Graduate Texts in Mathematics

63
Editorial Board

F. W Gehring
P. R. Halmos
Managing Editor

c.e. Moore
 Bela Bollobas

Graph Theory
An Introductory Course

Springer-Verlag
New York Heidelberg Berlin
Bela Bollobas
Department of Pure Mathematics
and Mathematical Statistics
University of Cambridge
16 Mill Lane
Cambridge CB2 ISB
ENGLAND

Editorial Board

P. R. Halmos F. W. Gehring c. C. Moore

Managing Editor University of Michigan University of California
Indiana University Department of Mathematics at Berkeley
Department of Mathematics Ann Arbor, Michigan 48104 Department of Mathematics
Bloomington, Indiana 47401 USA Berkeley, California 94720
USA USA

AMS Subject Classification: 05Cxx

With 80 Figures

Library of Congress Cataloging in Publication Data

Bollobas, Bela.
Graph theory.

(Graduate texts in mathematics: 63)

Includes index.
I. Graph theory. I. Title. II. Series.
QA166.B662 511'.5 79-10720

All rights reserved.

No part of this book may be translated or reproduced in any form

without written permission from Springer-Verlag.

1979 by Springer-Verlag New York Inc.

Softcover reprint of the hardcover 1st edition 1979

9 876 54 32 1

ISBN-13: 978-1-4612-9969-1 e-ISBN-13: 978-1-4612-9967-7

001: 10.10077978-1-4612-9967-7
To Gabriella
There is no permanent place in the
world for ugly mathematics.
G. H. Hardy
A Mathematician's Apology
Preface

This book is intended for the young student who is interested in graph
theory and wishes to study it as part of his mathematical education. Ex-
perience at Cambridge shows that none of the currently available texts meet
this need. Either they are too specialized for their audience or they lack the
depth and development needed to reveal the nature of the subject.
We start from the premise that graph theory is one of several courses
which compete for the student's attention and should contribute to his
appreciation of mathematics as a whole. Therefore, the book does not
consist merely of a catalogue of results but also contains extensive descriptive
passages designed to convey the flavour of the subject and to arouse the
student's interest. Those theorems which are vital to the development are
stated clearly, together with full and detailed proofs. The book thereby
offers a leisurely introduction to graph theory which culminates in a thorough
grounding in most aspects of the subject.
Each chapter contains three or four sections, exercises and bibliographical
notes. Elementary exercises are marked with a - sign, while the difficult
ones, marked by + signs, are often accompanied by detailed hints. In the
opening sections the reader is led gently through the material: the results
are rather simple and their easy proofs are presented in detail. The later
sections are for those whose interest in the topic has been excited: the theorems
tend to be deeper and their proofs, which may not be simple, are described
more rapidly. Throughout this book the reader will discover connections
with various other branches of mathematics, including optimization theory,
linear algebra, group theory, projective geometry, representation theory,
probability theory, analysis, knot theory and ring theory. Although most
of these connections are nQt essential for an understanding of the book, the
reader would benefit greatly from a modest acquaintance with these SUbjects.

vii
viii Preface

The bibliographical notes are not intended to be exhaustive but rather to

guide the reader to additional material.
I am grateful to Andrew Thomason for reading the manuscript carefully
and making many useful suggestions. John Conway has also taught the
graph theory course at Cambridge and I am particularly indebted to him for
detailed advice and assistance with Chapters II and VIII. I would like to
thank Springer-Verlag and especially Joyce Schanbacher for their efficiency
and great skill in producing this book.

Cambridge Bela Bollobas

April 1979
Contents

Chapter I
Fundamentals
1. Definitions 1
2. Paths, Cycles and Trees 6
3. Hamilton Cycles and Euler Circuits 11
4. Planar Graphs 16
5. An Application of Euler Trails to Algebra 19
Exercises 22
Notes 25

Chapter II
Electrical Networks 26
1. Graphs and Electrical Networks 26
2. Squaring the Square 33
3. Vector Spaces and Matrices Associated with Graphs 35
Exercises 41
Notes 43

Chapter III
Flows, Connectivity and Matching 44

1. Flows in Directed Graphs 45

2. Connectivity and Menger's Theorem 50
3. Matching 53
4. Tutte's I-Factor Theorem 58
Exercises 61
Notes 66

IX
X Contents

Chapter IV
Extremal Problems 67
1. Paths and Cycles 68
2. Complete Subgraphs 71
3. Hamilton Paths and Cycles 75
4. The Structure of Graphs 80
Exercises 84
Notes 87

Chapter V
Colouring 88
I. Vertex Colouring 89
2. Edge Colouring 93
3. Graphs on Surfaces 95
fure~ ~
Notes 102

Chapter VI
Ramsey Theory 103
I. The Fundamental Ramsey Theorems 103
2. Monochromatic Subgraphs 107
3. Ramsey Theorems in Algebra and Geometry 110
4. Subsequences liS
Exercises 119
Notes 121

Chapter VII
Random Graphs 123
I. Complete Subgraphs and Ramsey Numbers-The Use of the Expectation 124
2. Girth and Chromatic Number-Altering a Random Graph 127
3. Simple Properties of Almost All Graphs-The Basic Use of Probability 130
4. Almost Determined Variables-The Use of the Variance 133
5. Hamilton Cycles-The Use of Graph Theoretic Tools 139
Exercises 142
Notes 144

Chapter VIIl
Graphs and Groups 146
I. Cayley and Schreier Diagrams 146
2. Applications of the Adjacency Matrix 155
3. Enumeration and P6lya's Theorem 162
Exercises 169
Notes 173

Subject Index 175

Index of Symbols 179

CHAPTER I
Fundamentals

The purpose of this introduction is to familiarise the reader with the basic
concepts and results of graph theory. The chapter inevitably contains a
large number of definitions and in order to prevent the reader growing
weary we prove simple results as soon as possible. The reader is not expected
to have complete mastery of Chapter I before sampling the rest of the
book, indeed, he is encouraged to skip ahead since most of the terminology
is self-explanatory. We should add at this stage that the terminology of
graph theory is far from being standard, though that used in this book is
well accepted.

1 Definitions
A graph G is an ordered pair of disjoint sets (V, E) such that E is a subset
of the set of unordered pairs of V. Unless it is explicitly stated otherwise, we
consider only finite graphs, that is V and E are always finite. The set V is
the set of vertices and E is the set of edges. If G is a graph then V = V(G)
is the vertex set of G and E = E(G) is the edge set. An edge {x, y} is said to
join the vertices x and y and is denoted by xy. Thus xy and yx mean exactly
the same edge; the vertices x and yare the end vertices of this edge. If x y E E( G)
then x and yare adjacent or neighbouring vertices of G and the vertices x
and yare incident with the edge xy. Two edges are adjacent if'they have
exactly one common end vertex.
As the terminology suggests, we do not usually think of a graph as an
ordered pair, but as a collection of vertices some of which are joined by
2 I Fundamentals

2 6

4
Figure 1.1. A graph.

edges. It is then a natural step to draw a picture of the graph. In fact, some-
times the easiest way to describe a graph is to draw it; the graph G =
({I, 2, 3, 4,5, 6}, {12, 14, 16,25,34,36,45, 56}) is immediately comprehended
by looking at Figure 1.1.
We say that G' = (V', E') is a subgraph of G = (V, E) if V' c V and
E' c E. In this case we write G' c G. If G' contains all edges of G that join
two vertices in V' then G' is said to be the subgraph induced or spanned by
V' and is denoted by G[V']. A subgraph G' of G is an induced subgraph if
G' = G[V(G')]. These concepts are illustrated in Figure 1.2.
We shall often construct new graphs from old ones by deleting or adding
some vertices and edges. If We V(G) then G - W = G[V\W] is the sub-
graph of G obtained by deleting the vertices in Wand all edges incident with
them. Similarly if E' c E(G) then G - E' = (V(G), E(G)\E'). If W = {w}
and E' = {xy} then this notation is simplified to G - wand G - xy.
Similarly, if x and yare non-adjacent vertices of G then G + xy is obtained
from G by joining x to y.
If x is a vertex of a graph G then instead of x E V(G) we usually write
x E G. The order of G is the number of vertices; it is denoted by IG I. The
same notation is used for the number of elements (cardinality) of a set: IXI
denotes the number of elements of the set X. Thus IGI = IV(G)I. The size
of G is the number of edges; it is denoted by e(G). We write Gn for an
arbitrary graph of order n. Similarly G(n, m) denotes an arbitrary graph of
order n and size m.

2 6

3
3~
4 4
Figure 1.2. A subgraph and an induced subgraph of the graph in Figure 1.1.
 I Definitions 3

Figure 1.3. Graphs of order at most 4 and size 3.

Two graphs are isomorphic if there is a correspondence between their

vertex sets that preserves adjacency. Thus G = (V, E) is isomorphic to
Gf = (V', E') if there is a bijection : V -> V' such that xy E E iff (x)(y)EE'.
Clearly isomorphic graphs have the same order and size. Usually we do
not distinguish between isomorphic graphs, unless we consider graphs with
a distinguished or labelled set of vertices (for example, subgraphs of a given
graph). In accordance with this convention, if G and H are isomorphic
graphs then we write either G ~ H or simply G = H. In Figure 1.3 we show
all graphs (within isomorphism) that have order at most 4 and size 3.
The size of a graph of order n is at least 0 and at most (2). Clearly for
every m, 0 S m S (2), there is a graph G(n, m). A graph of order n and size
(2) is called a complete n-graph and is denoted by Kn; an empty n-graph En
has order n and no edges. In K" every two vertices are adjacent, while in En
no two vertices are adjacent. The graph Kl = El is said to be trivial.
The set of vertices adjacent to a vertex x EGis denoted by r(x). The
degree of x is d(x) = 1r(x) I. If we want to emphasize that the underlying
graph is G then we write r G(x) and dG(x); a similar convention will be adopted
for other functions depending on an underlying graph. Thus if x E H = G[ W]
then

The minimum degree of the vertices of a graph G is denoted by <5(G) and

the maximum degree by ~(G). A vertex of degree 0 is said to be an isolated
vertex. If <5( G) = ~(G) = k, that is every vertex of G has degree k then G
is said to be k-regular or regular of degree k. A graph is regular if it is k-
regular for some k. A 3-regular graph is said to be cubic.
If V(G) = {Xl' X2, ... , x n} then (d(xJ)~ is a degree sequence of G. Usually
we order the vertices in such a way that the degree sequence obtained in
this way is monotone increasing or monotone decreasing, for example
<5(G) = d(xl) S ... S d(x n) = ~(G). Since each edge has two endvertices,
the sum of the degrees is exactly twice the number of edges:
n
I d(x;) = 2e(G). (1)
1

In particular, the sum of degrees is even:

n
L d(x;) == 0 (mod 2). (2)
4 I Fundamentals

This last observation is sometimes called the handshaking lemma, since it

expresses the fact that in any party the total number of hands shaken is
even. Equivalently, (2) states that the number of vertices of odd degree is
even. We see also from (1) that b( G) ~ L2e( G)/n J and ~(G) 2 f2e( G)/n1-
r
Here LxJ denotes the greatest integer not greater than x and x1= - L- xJ.
A path is a graph P of the form
V(P) = {Xo, Xl"'" Xl}, E(P) = {XOXl, XlX2,"" XI-lXI}'

This path P is usually denoted by xox 1 ... Xl' The vertices Xo and Xl are the
endvertices of P and I = e(P) is the length of P. We say that P is a path
from Xo to Xl or an XO-Xl path. Of course, P is also a path from Xl to Xo or an
XZ-Xo path. Sometimes we wish to emphasize that P is considered to go
from Xo to Xl and then call Xo the initial and Xl the terminal vertex of P. A
path with initial vertex X is an x-path.
The term independent will be used in connection with vertices, edges and
paths of a graph. A set of vertices (edges) is independent if no two elements
of it are adjacent; a set of paths is independent if for any two paths each
vertex belonging to both paths is an endvertex of both. Thus PI, P 2, ... , P k
are independent x-y paths iff V(P;) n V(P) = {x, y} whenever 1 =F j. Also,
W c V(G) consists of independent vertices iff G[W] is an empty graph.
Most paths we consider are subgraphs of a given graph G. A walk Win G
is an alternating sequence of vertices and edges, say Xo, (Xl' X2, (X2, ... , (Xz, Xo
where (Xi = Xi-IX;, 0 ~ i < I. In accordance with the terminology above,
W is an XO-Xl walk and is denoted by XOXI ... Xl; the length of W is l. This
walk W is called a trail if all its edges are distinct. Note that a path is a walk
with distinct vertices. A trail whose endvertices coincide (a closed trail) is
called a circuit. If a walk W = XOXI .. Xl is such that 12 3, Xo = Xl and
the vertices Xi' 0 < i < I, are distinct from each other and Xo then W is said
to be a cycle. For simplicity this cycle is denoted by XIX2 ... Xl' Note that
the notation differs from that of a path since X 1Xl is also an edge of this
cycle. Furthermore, XlX2'" Xl' XIXI- l '" Xl> X2X3'" XIX l , XiXi-l'"
XlXIXI- l . Xi+ 1 all denote the same cycle.
The symbol pi denotes an arbitrary path of length I and C l denotes a
cycle of length I. We call C 3 a triangle, C4 a quadrilateral, C5 a pentagon, etc.
(See Figure 1.4). A cycle is even (odd) if its length is even (odd).
Given vertices x, y, their distance d(x, y) is the minimum length of an
x-y path. If there is no x-y path then d(x, y) = 00.
A graph is connected if for every pair {x, y} of distinct vertices there is a
path from X to y. Note that a connected graph of order at least 2 cannot
contain an isolated vertex. A maximal connected subgraph is a component

DO
I Definitions 5

t Figure I.5. A forest.

lY
of the graph. A cutvertex is a vertex whose deletion increases the number of
components. Similarly an edge is a bridge if its deletion increases the number
of components. Thus an edge of a connected graph is a bridge if its deletion
disconnects the graph. A graph without any cycles is a forest or an acyclic
graph; a tree is a connected forest. (See Figure 1.5.) The relation of a tree to a
forest sounds less absurd if we note that a forest is a disjoint union of trees;
in other words, a forest is a graph whose every component is a tree.
A graph G is a bipartite graph with vertex classes VI and V2 if V(G) =
VI U V2, VI n V2 = 0 and each edge joins a vertex of VI to a vertex of V2.
Similarly G is r-partite with vertex classes VI' V2, ... , V. if V(G) = VI U V2 U
... U v., V; n J.j = 0 whenever 1 ::; i < j ::; r, and no edge joins two
vertices in the same class. The graphs in Figure 1.1 and Figure 1.5 are bi-
partite. The symbol K(n!, ... , nr ) denotes a complete r-partite graph: it
has ni vertices in the ith class and contains all edges joining vertices in
distinct classes. For simplicity we often write Kp,q instead of K(p, q) and
K,(t) instead of K(t, ... , t).
We shall write G u H = (V(G) u V(H), E(G) u E(H and kG for the
union of k disjoint copies of G. We obtain the join G + H from G u H
by adding all edges between G and H. Thus, for example, K 2,3 = E2 + E3
and Kr(t) = Et + ... + Et.
There are several notions closely related to that of a graph. A hypergraph
is a pair (V, E) such that V n E = 0 and E is a subset of .?l(V), the power
set of V, that is the set of all subsets of V. In fact, there is a simple 1-1 cor-
respondence between the class of hypergraphs and the class of certain
bipartite graphs. Indeed, given a hypergraph (V, E), construct a bipartite
graph with vertex classes V and E by joining a vertex x E V to a hyperedge
SEE iff XES.
By definition a graph does not contain a loop, an "edge" joining a vertex
to itself; neither does it contain multiple edges, that is several "edges"
joining the same two vertices. In a multigraph both multiple edges and
multiple loops are allowed; a loop is a special edge.
If the edges are ordered pairs of vertices then we get the notions of a
directed graph and directed multigraph. An ordered pair (a, b) is said to be
an edge directed from a to b. or an edge beginning at a and ending at b, and
is denoted by a6 or simply abo The notions defined for graphs are easily
carried over to multigraphs, directed graphs and directed multigraphs,
mutatis mutandis. Thus a (directed) trail in a directed multigraph is an
6 I Fundamentals

alternating sequence of vertices and edges: Xo, el' Xl' e2' ... , ez' Xl' such that
eibegins at Xi-l and ends at Xi.
An oriented graph is a directed graph obtained by orienting the edges,
that is by giving the edge ab a direction ab or liii. Thus an oriented graph is
a directed graph in which at most one of ab and bQ occurs.

2 Paths, Cycles and Trees

With the concepts defined so far we can start proving some results about
graphs. Though these results are hardly more than simple observations,
and our main aim in presenting them is to familiarize the reader with the
concepts, in keeping with the style of the other chapters we shall call them
theorems.

Theorem 1. Let X be a vertex of a graph G and let W be the vertex set of a

component containing x. Then the following assertions hold.
1. W = {y E G: G contains an x-y path}.
ii. W = {y E G: G contains an x-y trail}.
iii. W = {y E G: d(x, y) < co}.
IV. For u, v E V = V(G) put uRv iff uv E E(G), and let R be the smallest
equivalence relation on V containing R. Then W is the equivalence class
~x. 0
This little result implies that every graph is the vertex disjoint union of its
components (equivalently, every vertex is contained in a unique component),
and an edge is a bridge iff it is not contained in a cycle.

Theorem 2. A graph is bipartite iff it does not contain an odd cycle.

PROOF. Suppose G is bipartite with vertex classes VI and V2 . Let X I X2 ... Xl
be a cycle in G. We may assume that Xl E VI. Then X2 E V2 , X3 E Vb and
so on: Xi E VI iff i is odd. Since Xl E V2 , we find that I is even.
Suppose now that G does not contain an odd cycle. Since a graph is
bipartite iff each component of it is, we may assume that G is connected.
Pick a vertex X E V(G) and put VI = {y: d(x, y) is odd}. There is no edge
joining two vertices of the same class since otherwise G would contain an
odd cycle. Hence G is bipartite. 0

Theorem 3. A graph is a fores! iff for every pair {x, y} of distinct vertices it
contains at most one x-y path.

PROOF. If XIX2 ... Xl is a cycle in a graph G then XIX2 ... Xl and XIXI are
two Xl-Xl paths in G.
2 Paths, Cycles and Trees 7

Conversely, let PI = XOXI ... X, and P 2 = XOYlY2 ... YkX, be two

distinct XO-X, paths in a graph G. Let i + 1 be the minimal index for which
Xi+l #- Yi+l, and letj be the minimal index for whichj ~ i and Yj+l is a
vertex of PI' say Yj+l = Xh' Then XiXi+1 ... XhYjYj-1 .. , Yi+l is a cycle in G.
o
Theorem 4. The following assertions are equivalent for a graph G.

i. G is a tree.
n. G is a minimal connected graph, that is G is connected and if xY E E(G)
then G - xY is disconnected. [In other words, G is connected and every
edge is a bridge.]
iii. G is a maximal acyclic graph, that is G is acyclic and if X and yare non-
adjacent vertices of G then G + xY contains a cycle.

PROOF. Suppose G is a tree. Let xy E E(G). The graph G - xy cannot contain

an X-Y path XZ IZ2 ... ZkY since otherwise G contains the cycle XZtZ2'" ZkY'
Hence G - xy is disconnected and so G is a minimal connected graph.
Similarly if x and yare non-adjacent vertices of the tree G then G contains a
path XZIZ2 ... ZkY and so G + xy contains the cycle XZIZ2 ... ZkY' Hence
G + xy contains a cycle and so G is a maximal acyclic graph.
Suppose next that G is a minimal connected graph. If G contains a cycle
XZIZ2 ... ZkY then G - xy is still connected since in any u - v walk in G
the edge xy can be replaced by the path XZIZ2 ... XkY' As this contradicts
the minimality of G, we conclude that G is acyclic so it is a tree.
Suppose finally that G is a maximal acyclic graph. Is G connected? Yes,
since if x and Y belong to different components, the addition of xy to G
cannot create a cycle XZIZ2 ... ZkY since otherwise the path XZIZ2 ... ZkY is
in G. Thus G is a tree. 0

Corollary 5. Every connected graph contains a spanning tree, that is a tree

containing every vertex of the graph.

PROOF. Take a minimal connected spanning subgraph. There are several

simple constructions of a spanning tree of a graph G; we present two of
them. Pick a vertex x and put V; = {y E G: d(x, y) = i}, i = 0,1, .... Note
that if YiE V;, i > 0, and XZIZ2'" Zi-IYi is an x - Yi path (whose existence

is guaranteed by the definition of V;) then d(x, z) = j for every j, < j < i.

In particular, l'j #- 0 for ::s; j ::s; i and for every Y E V;, i > 0, there is a
vertex Y' E V;-t joined to y. (Of course, this vertex Y' is usually not unique but
for each Y #- x we pick only one y'.) Let T be the subgraph of G with vertex
set V and edge set E(T) = {yy': Y #- x}. Then T is connected since every
yE V - {x} is joined to x by a path yy'y" ... x. Furthermore, T is acyclic
since if W is any subset of V and w is a vertex in W furthest from x then w is
joined to at most one vertex in W. Thus T is a spanning tree. 0
8 I Fundamentals

The argument above shows that with k = max y d(x, y) we have Vi # 0

for 0 :::; i :::; k and V = V( G) = U~ Vi. At this point it is difficult to resist
the remark that diam G = maxx,y d(x, y) is called the diameter of G and
rad G = minx maxy d(x, y) is the radius of G.
If we choose x E G with k = max y d(x, y) = rad G then the spanning
tree T also has radius k.
A slight variant of the above construction of T goes as follows. Pick
x E G and let Tl be the subgraph of G with this single vertex x. Then TI is a
tree. Suppose we have constructed trees Tl C Tz C . . . C 7;, C G, where
T; has order i. If k < n = IG I then by the connectedness of G there is a
vertex y E V(G) - V(7;,) that is adjacent (in G) to a vertex z E 7;,. Let 7;,+ I
be obtained from 7;, by adding to it the vertex y and the edge yz. Then 7;,+ 1
is connected and as yz cannot be an edge of a cycle in 7;,+ b it is also acyclic.
Thus 7;,+ I is also a tree, so the sequence To C Tl C ... can be continued
to T". This tree T" is then a spanning tree of G.
The spanning trees constructed by either of the methods above have
order n (of course!) and size n - 1. In the first construction there is a 1-1
correspondence between V - {x} and E(T), given by y -> yy', and in the
second construction e(I;C> = k - 1 for each k since e(T1 ) = 0 and 7;,+ I
has one more edge than 7;,. Since by Theorem 4 every tree has a unique
spanning tree, namely itself, we have arrived at the following result.

Corollary 6. A tree of order n has size n - 1; a forest of order n with k com-

ponents has size n - k. 0

The first part of this corollary can be incorporated into several other
characterizations of trees. In particular, a graph of order n is a tree iff it is
connected and has size n - 1. The reader is invited to prove these character-
izations (Exercise 9).

Corollary 7. A tree of order at least 2 contains at least 2 vertices of degree 1.

PROOF. Let d l :::; d z :::; ... :::; dn be the degree sequence of a tree T of order
n 2:: 2. Since T is connected, (j(T) == d 1 2:: 1. Hence if T had at most one
vertex of degree 1, by (1) and Corollary 5 we would have
n
2e(T) = 2n - 2 = L d; 2:: 1 + 2(n -
1
1). o
A well known problem in optimization theory asks for a relatively easy
way of finding a spanning subgraph with a special property. Given a graph
G = (V, E) and a positive valued cost function f defined on the edges,
f: E -> IR +, find a connected spanning subgraph T = (V, E') of G for which
f(T) = L f(xy)
xyeE'
2 Paths, Cycles and Trees 9

is minimal. We call such a spanning subgraph T an economical spanning

subgraph. One does not need much imagination to translate this into a
"real life" problem. Suppose certain villages in an area are to be joined to a
water supply situated in one of the villages. The system of pipes is to consist
of pipelines connecting the water towers oftwo villages. For any two villages
we know how much it would cost to build a pipeline connecting them,
provided such a pipeline can be built at all. How can we find an economical
system of pipes?
In order to reduce the second problem to the above problem about
graphs, let G be the graph whose vertex set is the set of villages and in which
xy is an edge iff it is possible to build a pipeline joining x to y; denote the
cost of such a pipeline by f(xy) (see Figure 1.6). Then a system of pipes
corresponds to a connected spanning subgraph T of G. Since the system has
to be economical, T is a minimal connected spanning subgraph of G, that is a
spanning tree of G.
The connected spanning subgraph T we look for has to be a minimal
connected subgraph since otherwise we could find an edge a whose deletion
would leave T connected and then T - IX would be a more economical
spanning subgraph. Thus T is a spanning tree of G. Corresponding to the
various characterizations and constructions of a spanning tree, we have
several easy ways of finding an economical spanning tree; we shall describe
four of these methods.
(1) Given G and f: E(G) ---+ ~, we choose one of the cheapest edges of G,
that is an edge a for which f(a) is minimal. Each subsequent edge will be
chosen from among the cheapest remaining edges of G with the only re-
striction that we must not select all edges of any cycle, that is the subgraph of
G formed by the selected edges is acyclic.
The process terminates when no edge can be added to the set E' of edges
selected so far without creating a cycle. Then Tl = (V(G), E') is a maximal
acyclic subgraph of G so, by Theorem 4(iii) it is a spanning tree of G.
(2) This method is based on the fact that it is foolish to use a costly edge
unless it is needed to ensure the connectedness of the subgraph. Thus let us
delete one by one those costliest edges whose deletion does not disconnect
the graph. By Theorem 4(ii) the process ends in a spanning tree T2

4
Figure 1.6. A graph with a function f: E --+ IR+ ; the number next to an edge xy is the
cost f(xy) of the edge. .
10 I Fundamentals

(3) Pick a vertex x 1 of G and select one of the least costly edges incident
with Xl' say X 1 X2' Then choose one of the least costly edges of the form Xi."t.
where 1:::; i :::; 2 and X {Xl, X2}' Having found vertices Xl> X2," X k
and an edge XiX j ' i < j, for each vertex X j with j :::; k, select one of the least
costly edges of the form XiX, say XiXk+ 1 where 1:::; i :::; k and
Xk+l {X 1X 2 , . , xd The process terminates after we have selected
n - 1 edges. Denote by T3 the spanning tree given by these edges. (See
Figure 1.7.)

Figure I.7. Three of the six economical spanning trees of the graph shown in Figure 1.6.

(4) This method is applicable only if no two pipelines cost the same. The
advantage of the method is that every village can make its own decision and
start building a pipeline without bothering to find out what the other villages
are going to do. Of course, each village will start building the cheapest
pipeline ending in the village. It may happen that both village X and village y
will build the pipeline xy; in this case they meet in the middle and end up
with a single pipeline from X to y. Thus at the end of this stage some villages
will be joined by pipelines but the whole system of pipes need not be con-
nected. At the next stage each group of villages joined to each other by pipe-
lines finds the cheapest pipeline going to a village not in the group and begins
to build that single pipeline. The same procedure is repeated until a connected
system is obtained. Clearly the villages will never build all the pipes of a
cycle so the final system of pipes will be a spanning tree (See Figure 1.8.)

Theorem 8. Each of the four methods described above produces an economical

spanning tree. If no two edges have the same cost then there is a unique econo-
mical spanning tree.

1+

4+
Figure I.8. The graph of Figure 1.6 with a slightly altered cost function (0 < e < 1)
and its unique economical spanning tree.
3 Hamilton Cycles and Euler Circuits 11

PROOF. Choose an economical spanning tree T of G that has as many edges

in common with TI as possible. (T1 is the spanning tree constructed by the
first method.)
Suppose that E(T1 ) =f. E(T). The edges of TI have been selected one by
one; let x y be the first edge of TI that is not an edge of T. Then T contains a
unique x-y path, say P. This path P has at least one edge, say uv, that does
not belong to T1 , since otherwise TI would contain a cycle. When xy was
selected as an edge of T1 , the edge uv was also a candidate. As xy was chosen
and not uv, the edge xy can not be costlier than uv, that isf(xy) ~ f(uv). Then
T' = T - uv + xy is a spanning tree and since f(T') = f(T) - f(uv) +
f(xy) ~ f(T), T' is an economical spanning tree of G. (Of course, this
inequality implies that f(T') = f(T) and f(xy) = f(uv).) This tree T' has
more edges in common with TI than T, contradicting the choice of T. Hence
T = TI so TI is indeed an economical spanning tree.
Slight variants of the proof above show that the spanning trees T2 and
T3 , constructed by the second and third methods, are also economical. We
invite the reader to furnish the details (Exercise 19).
Suppose now that no two edges have the same cost, that it f(xy) =f. f(uv)
whenever xy =f. uv. Let T4 be the spanning tree constructed by the fourth
method and let T be an economical spanning tree. Suppose that T =f. T4 ,
and let xy be the first edge not in T that we select for T4 . The edge xy was
selected since it is the least costly edge of G joining a vertex of a subtree
F of T4 to a vertex outside F. The x-y path in T has an edge uv joining a
vertex of F to a vertex outside F so f(xy) < f(uv). However, this is impos-
sible since T' = T - lIl' + xy is a spanning tree of G and f(T') < f(T).
Hence T = T4 . This shows that T4 is indeed an economical spanning tree.
Furthermore, since the spanning tree constructed by the fourth method is
unique, the economical spanning tree is unique if no two edges have the
same cost. D

3 Hamilton Cycles and Euler Circuits

The so-called travelling salesman problem greatly resembles the economical

spanning tree problem discussed in the preceding section, but the similarity
is only superficial. A salesman is to make a tour of n cities, at the end of which
he has to return to the head office he starts from. The cost of the journey
between any two cities is known. The problem asks for an efficient algorithm
for finding a least expensive tour. (As we shall not deal with algorithmic
problems, we leave the term "efficient" undefined; loosely speaking an
algorithm is efficient if the computing time is bounded by a polynomial in
the number of vertices.) Though a considerable amount of work has been
done on this problem, since its solution would have important practical
12 I Fundamentals

Figure 1.9. A Hamiltonian cycle in the graph of the dodecahedron.

applications, it is not known whether or not there is an efficient algorithm

for finding a least expensive route.
In another version of the travelling salesman problem the route is required
to be a cycle, that is the salesman is not allowed to visit the same city twice
(except the city of the head office). A cycle containing all the vertices of a
graph is said to be a Hamilton cycle of the graph. The origin of this term is a
game invented in 1857 by Sir William Rowan Hamilton based on the
construction of cycles containing all the vertices in the graph of the dode-
cahedron (see Figure 1.9). A Hamilton path of a graph is a path containing
all the vertices of the graph. A graph containing a Hamilton cycle is said to
be Hamiltonian.
In fact, Hamilton cycles and paths in special graphs had been studied well
before Hamilton proposed his game. In particular, the puzzle of the knight's
tour on a chessboard, thoroughly analysed by Euler in 1759, asks for a
Hamilton cycle in the graph whose vertices are the 64 squares of a chessboard
and in which two vertices are adjacent if a knight can jump from one square
to the other. Figure 1.10. shows two solutions of this puzzle.
If in the second, more restrictive version ofthe travelling salesman problem
there are only two travel costs, 1 and 00 (expressing the impossibility of the
journey), then the question is whether or not the graph formed by the edges
with travel cost 1 contains a Hamilton cycle. Even this special case of the
travelling salesman problem is unsolved: no efficient algorithm is known

Figure 1.10. Two tours of a knight on a chessboard.

3 Hamilton Cycles and Euler Circuits 13

Figure 1.11. Three edge-disjoint Hamilton paths in K6.

for constructing a Hamilton cycle, though neither is it known that there is

no such algorithm.
If the travel cost between any two cities is the same, then our salesman has
no difficulty in finding a least expensive tour: any permutation of the n - 1
cities (the nth city is that of the head office) will do. Revelling in his new found
freedom, our salesman decides to connect duty and pleasure and promises
not to take the same road xy again whilst there is a road uv he hasn't seen yet.
Can he keep his promise? In order to plan a required sequence of journeys
for our salesman we have to decompose K n into the union of some edge
disjoint Hamilton cycles. For which values of n is this possible? Since K n is
(n - I)-regular and a Hamilton cycle is 2-regular, a necessary condition is
that n - 1 is even, that is n is odd. This necessary condition also follows
from the fact that e(Kn) = !n(n - 1) and a Hamilton cycle contains n edges,
so K n has to be the union of !(n - 1) Hamilton cycles.
Let us assume now that n is odd, n ~ 3. Deleting a vertex of K n we see that
if K n is the union of !(n - 1) Hamilton cycles K n- 1 is the union of !(n - 1)
Hamilton paths. (In fact n - 1 has to be even if K n - 1 is the union of some
Hamilton paths since e(K n- 1 ) = !(n - 1)(n - 2) and a Hamilton path in
K n - 1 has n - 2 edges.) With the hint shown in Figure 1.11 the reader can
show that for odd values of n the graph K n - 1 is indeed the union of !(n - 1)
Hamilton paths. In this decomposition of K n - 1 into !(n - 1) Hamilton
paths each vertex is the end vertex of exactly one Hamilton path. (In fact,
this holds for every decomposition of K n - 1 into !(n - 1) edge-disjoint
Hamilton paths since each vertex x of K n - 1 has odd degree so at least one
Hamilton path has to end in x.) Consequently if we add a new vertex to
K n- 1 and complete each Hamilton path in K n- 1 to a Hamilton cycle in K n
then we obtain a decomposition of K n into !(n - 1) edge disjoint Hamilton
cycles. Thus we have proved the following result.

Theorem 9. For n ~ 3 the complete graph K n is decomposable into edge

disjoint Hamilton cycles iff n is odd. For n ~ 2 the complete graph K n is decom-
posable into edge disjoint Hamilton paths iffn is even. D

The result above shows that if n ~ 3 is odd, then we can string together
!(n - 1) edge disjoint cycles in K n to obtain a circuit containing all the edges
of Kn. In general a circuit in a graph G containing all the edges is said to be
and Euler circuit of G. Similarly a trail containing all edges is an Euler trail.
14 I Fundamentals

Figure I.12. The seven bridges on the Pregel in Konigsberg.

A graph is Eulerian if it has an Euler circuit. Euler circuits and trails are
named after Leonhard Euler, who in 1736 characterized those graphs which
contain them. At the time Euler was a professor of mathematics in St.
Petersburg, and was led to the problem by the puzzle of the seven bridges
on the Pregel (see Figure 1.12) in the ancient Prussian city of Konigsberg
(birthplace of Kant; since 1944 it has belonged to the USSR and is called.
Kaliningrad). Could anyone plan a walk in such a way that he would cross
each bridge once and only once?
It is clear that such a walk is possible iff the graph in Figure 1.13 has an
Euler trail.

Theorem 10. A non-trivial connected graph has an Euler circuit iff each vertex
has even degree.
A connected graph has a Euler trail from a vertex x to a vertex y =1= x iff
x and yare the only vertices of odd degree.
PROOF. The conditions are clearly necessary. For example, if XIX2 ... Xm is
an Euler circuit in G and x occurs k times in the sequence Xl> X2'' X m ,
then d(x) = 2k.
We prove the sufficiency of the first condition by induction on the number
of edges. If there are no edges, there is nothing to prove so we proceed to the
induction step.
Let G be a non-trivial connected graph in which each vertex has even
degree. Since e( G) 2': 1, we find that b( G) 2': 2 so, by Corollary 7, G contains
a cycle. Let C be a circuit in G with the maximal number of edges. Suppose
C

A 4E-- -- - - -.... D

B
Figure f.13. A graph of the Konigsberg bridges.
3 Hamilton Cycles and Euler Circuits 15

Figure 1.14. The circuits C and D.

C is not Eulerian. As G is connected, C contains a vertex x which is in a

non-trivial component H of G - E( C). Every vertex of H has even degree
in H so by the induction hypothesis H contains an Euler circuit D. The
circuits C and D (see Figure 1.14) are edge-disjoint and have a vertex in
common, so they can be strung together to form a circuit with more edges
than C. As this contradicts the maximality of e(C), the circuit C is Eulerian.
Suppose now that G is connected and x and yare the only vertices of
odd degree. Let G* be obtained from G by adding to it a vertex u together
with the edges ux and uy. Then by the first part G* has an Euler circuit C*.
Clearly C* - u is an Euler trail from x to y. 0

Theorem 10 implies that there is no walk satisfying the conditions of the

Konigsberg bridge puzzle, since the associated graph in Figure 1.13 has
four vertices of odd degree. The plan of the corridors of an exhibition is also
easily turned into a graph: an edge corresponds to a corridor and a vertex
to the conjunction of several corridors. If the entrance and exit are the same,
a visitor can walk along every corridor exactly once iff the corresponding
graph has an Eulerian circuit. In general a visitor must have a plan in order
to achieve this: he cannot just walk through any new corridor he happens
to come to. However, in a well planned (!) exhibition a visitor would be
certain to see all the exhibits provided he avoided going along the same
corridor twice and continued his walk until there were no new exhibits
ahead of him. The graph of such an exhibition is said to be randomly Eulerian
from the vertex corresponding to the entrance (which is also the exit). See
Figure 1.15 for two examples. Randomly Eulerian graphs are also easily
characterized (Exercises 24-26).
_ _ _,"-:H

u v

x
Figj.lfe 1.15. G is randomly Eulerian from x; H is randomly Eulerian from both u and u.
16 I Fundamentals

4 Planar Graphs
The graph of the corridors of an exhibition is a planar graph: it can be drawn
in the plane in such a way that no two edges intersect. Putting it a bit more
rigorously, it is possible to represent it in the plane in such a way that the
vertices correspond to distinct points and the edges to simple Jordan curves
connecting the points of its end vertices in such a way that every two curves
are either disjoint or meet only at a common endpoint. The above representa-
tion of a graph is said to be a plane graph.
There is a simple way of associating a topological space with a graph,
which leads to another definition of planarity, trivially equivalent to the one
given above. Let PI' P2, ... be distinct points in ~3, the 3-dimensional
Euclidean space, such that every plane in ~3 contains at most 3 of these points.
Write (p;, p) for the straight line segment with endpoints Pi and Pj (open
or closed, as you like). Given a graph G = (V, E), V = (Xl, X2, ... , xn), the
topological space
n
R(G) = U{(Pi, P):XiXjEE} u UPi C ~3
1

is said to be a realization of G. A graph G is a planar if R( G) is homeomorphic

to a subset of ~2, the plane.
Let us make some more remarks in connection with R(G). A graph H
is said to be a subdivision of a graph G or a topological G graph if H is obtained
from G by subdividing some of the edges, that is by replacing the edges by
paths having at most their end vertices in common. We shall write TG for a
topological G graph. Thus TG denotes any member of a rather large family
of graphs; for example TK3 is an arbitrary cycle and TC 8 is an arbitrary
cycle oflength at least 8. It is clear that for any graph G the spaces R( G) and
R(TG) are homeomorphic. We shall say that a graph G is homeomorphic to a
graph H if R(G) is homeomorphic to R(H) or, equivalently, G and H have
isomorphic subdivisions.
At the first sight one may think that in the study of planar graphs one may
run into topological difficulties. This is certainly not the case. It is easily
seen that the Jordan curves corresponding to the edges can be assumed to
be polygons. More precisely, every plane graph is homotopic to a plane
graph representing the same graph, in which the Jordan curves are piecewise
linear. Indeed, given a plane graph, let (j > 0 be less than half the minimal
distance between two vertices. For each vertex a place a closed disc Da of
radius (j about a. Denote by J ~ the curve corresponding to an edge (X = ab
and let a~ be the last point of J~ in Da when going from a to b. Denote by
J~ the part of J~ from aa. to b~. Let 6 > 0 be such that if (X # f3 then J~ and
J'p are at a distance greater than 36. By the uniform continuity of a Jordan
curve each J~ can be approximated within 6 by a polygon J; from a~ to b~.
To get the required piecewise linear representation of the original graph
4 Planar Graphs 17

J:---
Figure 1.16. Constructing a piecewise linear n:presentation.

simply replace each J(J. by the poygon obtained from J~ by extending it in

both directions by the segments aa(J. and b(J.b (see Figure IJ6).
A less pedestrian argument shows that every planar graph has a straight
line representation: it can be drawn in the plane in such a way that the edges
are actually straight line segments (Exercise 28).
If we omit the vertices and edges of a plane graph G from the plane, the
remainder falls into connected components, called faces. Clearly each
plane graph has exactly one unbounded face. The boundary of a face is the
set of edges in its closure. Since a cycle (that is a simple closed polygon)
separates the points of the plane into two components, each edge of a cycle
is in the boundary of two faces. A plane graph together with the set of faces
it determines is called a plane map. The faces of a plane map are usually
called countries. Two countries are neighbouring if their boundaries have an
edge in common.
If we draw the graph of a convex polyhedron in the plane then the faces
of the polyhedron clearly correspond to the faces of the plane graph. This
leads us to another contribution of Leonhard Euler to graph theory, namely
Euler's polyhedron theorem or simply Euler's formula.

Theorem 11. If a connected plane graph G has n vertices, m edges and f faces,
then
n- m + f= 2.

PROOF. Let us apply induction on the number of faces. If f = 1 then G does

not contain a cycle so it is a tree and the result holds.
Suppose now that f > 1 and the result holds for smaller values of f.
Let ab be an edge in a cycle of G. Since a cycle separates the plane, the edge
ab is in the boundary of two faces, say Sand T. Omitting ab, in the new plane
graph G' the faces Sand T join up to form a new face, while all other faces of
G remain unchanged. Thus if n', m' and f' are the parameters of G' then
n' = n, m' = m - 1 and f' = f - 1. Hence n - m + f = n' - m' - f = 2.
o
18 I Fundamentals

Let G be a connected plane graph with n vertices, m edges and f faces;

furthermore, denote by ii the number of faces having exactly i edges in their
boundaries. Of course
Iii = f,
i
(3)

and if G has no bridge then

I ifi = 2m, (4)
i

since every edge is in the boundary oftwo faces. Relations (3), (4) and Euler's
formula give an upper bound for the number of edges of a planar graph of
order n. This bound can be improved if the girth of the graph, that is the
number of edges in a shortest cycle, is large. (The girth of an acyclic graph is
defined to be 00.)

Theorem 12. A planar graph of order n ~ 3 has at most 3n - 6 edges. Further-

more, a planar graph of order n and girth at least g, 3 ::;; g < 00, has size at
most

max {g ~ 2 (n - 2), n - I}.

PROOF. The first assertion is the case g = 3 of the second, so it suffices to

prove the second assertion. Let G be a planar graph of order n, size m and
girth at least g. If n ::;; g - 1 then G is acyclic so m ::;; n - 1. Assume now
that n ~ g and the assertion holds for smaller values of n. We may assume
without loss of generality that G is connected. If ab is a bridge then G - ab is
the union of two vertex disjoint subgraphs, say G 1 and G 2 Putting ni =
IGi I, mi = e( GJ, i = 1, 2, by induction we find

m = m1 + m2 + 1 ::;; max{g ~ 2 (nl - 2), nl - I}

+ max{g ~ 2(n 2 - 2),n2 - I} +1

::;; max{g ~ 2(n - 2), n - I}.

On the other hand, if Gis bridgeless, (3) and (4) imply

2m = Iifi = I ifi ~ IM = gf.

i i~g i

Hence, by Euler's formula,

2
m+2=n+f::;;n+-m,
g
5 An Application of Euler Trails to Algebra 19

and so
m ~ -g-(n - 2).
g-2
o
Theorem 12 can often be used to show that certain graphs are non-planar.
Thus K 5, the complete graph of order 5, is non-planar since e(K5) = 10 >
3(5 - 2), and K 3 3 the complete 3 by 3 bipartite graph is non-planar since
its girth is 4 and e(K 3 3 ) = 9 > (4/(4 - 2(6 - 2). The non-planarity of
K 3 3 implies that it is impossible to join each of 3 hours to each of 3 wells by
non-crossing paths, as demanded by a well-known puzzle (see Figure 1.17).

Figure I.17. Three houses and three wells.

If a graph G is non-planar then so is every topological G graph, and

every graph containing a topological G graph. Thus the graphs in Figure 1.18
are non-planar since they contain TK 5 and TK 3 3 , respectively.
It is somewhat surprising that the converse of the trivial remarks above
is also true. This beautiful result was proved by Kuratowski in 1930; as the
proof is rather long though elementary, we shall not give it here.

Figure I.18. G contains a TK 5 and H contains a TK 3 3 .

Theorem 13. A graph is planar iff it does not contain a subdivision of K 5 or

K 3 3 0

5 An Application of Euler Trails to Algebra

To conclude this chapter we shall show that even simple notions like the
ones presented so far may be of use in proving important results. The result
we are going to prove is the fundamental theorem of Amitsur and Levitzki
20 I Fundamentals

on polynomial identities. The commutator of two elements a and b of a

ring S is [a, bJ = ab - ba. Similarly, if ai E S, 1 sis k, we write
[ai' a2"'" akJ = L (-I)"a ulau2 ... auk'
u

where the summation is over all permutations (J of the integers 1, 2, ... , k.

If [at> a2, ... , akJ = 0 for all ai E S, 1 sis k, then S is said to satisfy the
k-th polynomial identity. The theorem of Amitsur and Levitzki states that the
ring Mk(R) of k by k matrices with entries in a commutative ring R satisfies
the 2k-th polynomial identity.

Theorem 14. Let R be a commutative ring and let Ai' A 2 , ... , A2k E Mk(R).
Then [Ai' A 2,, A 2k J = O.
PROOF. We shall deduce the result from a lemma about Euler trails in directed
multigraphs. Let Gbe a directed multigraph of order n with edges e 1, e2, ... ,
em' Thus each ei is an ordered pair of not necessarily distinct vertices. Every
(directed) Euler trail P is readily identified with a permutation of
{I, 2, ... , m}; define B(P) to be the sign of this permutation. Given not neces-
sarily distinct vertices x, y of G, put B(G; x, y) = LpB(P), where the sum-
mation is over all Euler trails from x to y.

Lemma 15. If m ~ 2n then B(G; x, y) = O.

Before proving this lemma, let us see how it implies Theorem 14. Write
Eij E Mn(R) for the matrix whose only non-zero entry is a 1 in the ith row and
jth column. Since [A 1, A 2 , ... ,A 2n J is R-linear in each variable and
{Eij: 1 S i,j S n} is a basis of MiR) as .an R-module, it suffices to prove
Theorem 14 when Ak = Ei0k for each k. Assuming this is the case, let Gbe the
directed multigraph with vertex set {I, 2, ... ,n}, whose set of directed edges is
{iljl' i 2j2"'" i2n j2n}' By the definition of matrix multiplication a product
A u1 A u2 '" A u2n is Eij if the corresponding sequence of edges is a (directed)
Euler trial from ito j and otherwise the product is O. Hence [Ai' A 2, .. ,A 2n J
= Li,j B(G; i,j)Eij' By Lemma 15 each summand is 0 so the sum is also O.

PROOF OF LEMMA 15. We may clearly assume that Ghas no isolated vertices.
Let G' be obtained from G by adding to it a vertex x', a path of length
m + 1 - 2n from x' to x and an edge from y to x' (see Figure 1.19). Then

Figure 1.19, The construction of G'.

5 An Application of Euler Trails to Algebra 21

G' has order n + (m + 1 - 2n) = m + 1 - n and size m + m + 1 - 2n + 1

= 2(m + 1 - n). Furthermore, it is easily checked that Ic(G; x, y)1 =
Ic(G'; x', x')I. Hence it suffices to prove the theorem when m = 2n and
x = y.
Given a vertex z, let d+(z) be the number of edges starting at z and let
r(z) be the number of edges ending at z. Call d(z) = d+(z) + d-(z) the
degree of z and f(z) = d+(z) - r(z) the flux at z. We may assume that
Gcontains an Euler circuit (an Euler trail from x to x), otherwise there is
nothing to prove. In this case each vertex has 0 flux and even degree at least 2.
Furthermore, we may assume that there is no double edge (and so no double
loop) for otherwise the assertion is trivial.
In order to prove the theorem in the case m = 2n and x = y we apply
induction on n. The case n = 1 being trivial, we turn to the induction step.
We shall distinguish three cases.
(i) There is a vertex b =F x of degree 2; say em = ab ends at band em-l = bc
starts at b. If a = c the assertion follows by applying the induction hypo-
thesis to G - b. If a =F c then without loss of generality x =F c. Let el = CCl,
e2 = CC2, ... , e t = CC t be the edges starting at c. For each i, 1 ::;; i ::;; t,
construct a graph G from G - b by omitting e; and adding e; = ac; (see
Figure 1.20). Then c(G; x, x) = 2::=1 c(G;; x, x) = o.

em e".-l

a. b c

Figure 1.20. The construction of G1.

(ii) There is a Loop at a vertex b =F x of degree 4. Let em be the loop at b

and let em- 2 = ab and em-l = bc be the other edges at b. Let Go be obtained
from G - b by adding to it an edge e~-2 = ac. Then c(6; x, x) = c(G o ; x, x)
= o.
(iii) The cases (i) and (ii) do not appLy. Since m = 2n - ! 2:1 d; and each
vertex distinct from x has degree at least 4, either each vertex has degree 4
or else d(x) = 2, there is a vertex of degree 6 and all other vertices have degree
4. It is easily checked (Exercise 34) that there are two adjacent vertices of
degree 4, say a and b, since otherwise (ii) would hold. Now we shall apply our
fourth and final graph transformation. This is more complicated than the
previous ones since we shall construct two pairs of essentially different
graphs from G: the graphs Gl , G2, Ii6 and Ii7 shown in Figure 1.21. Each
Euler trail from x to x in Gis transformed to an Euler trail in exactly one of
G1 and G2 However, the graphs G2 contain some spurious Euler trails:
Euler trails that do not come from Euler trails in G. As these spurious Euler
22 I Fundamentals

IX b

e,

e~

e'

e~ e,

H6 H,
,.~es e. e~

'rr.::
e.
e3 IX e3
b IX

e2 e2 e,
e"
Figure I.21. The graphs G., G2, ii6 and ii,.

trails are Euler trails in exactly one of H6 and H7 and they exhaust all the
Euler trails of H6 and H7, we find that
2 2
B(G; x, x) = L B(G;; x, x) - L B(H;; x, x).

1 1

The first two terms are because of (i) and the second two terms are
because of (ii) so B( G; x, x) = 0, completing the proof of Lemma 15. D

EXERCISES

1. (i) Show that every graph contains two vertices of equal degree.
(ii) Determine all graphs with one pair of vertices of equal degree.
2.- Prove that the complement of a disconnected graph is connected.
r Show that in a graph G there exists a set of cycles such that each edge of G belongs
to exactly one of these cycles iff every vertex has even degree.
4. Show that in an infinite graph G with countably many edges there exists a set
of cycles such that each edge of G belongs to exactly one of these cycles iff for
every X c V(G) the set of edges joining X to V(G) - X is even or infinite.
5. Show that d. ::;; d 2 ::;; ... ::;; dn is the degree sequence of a tree iff d 1 ::::: 1 and
D d; = 2n - 2.
Exercises 23

6. Show that every integer sequence d I ::::; d 2 ::::; .. ::::; dn with d l ;::: I and I~ di =
2n - 2k, k ;::: I, is the degree sequence of a forest with k components.
7. Characterize the degree sequence of forests!
8. Prove that a regular bipartite graph of degree at least 2 does not have a bridge.
9. Let G be a graph of order n. Prove the equivalence of the following assertions.
(i) G is a tree.
(ii) G is connected and has n - 1 edges.
(iii) G is acyclic and has n - 1 edges.
(iv) G = K n for n = I, 2, and if n ;::: 3 then G "# Kn and the addition of an
edge to G produces exactly one new cycle.
10. Let.91 = {A I, A 2' ... , An} be a family of n (;::: 1) distinct subsets of a set X with n
elements. Define a graph G with vertex set .91 in which Ai A j is an edge iff there
exists an x E X such that Ai f::::. Aj = {x}. Label the edge AiAj with x. For He G
let Lab(H) be the set of labels used for edges of H. Prove that there is a forest
Fe G such that Lab(F) = Lab(G).

11. (10 continued) Deduce from the result in the previous exercise that there is an
element x E X such that the sets AI - {x}, A2 - {x}, ... , {An} - {x} are all
distinct. Show that this need not hold for any n if 1.911 = n + 1.
12. A tournament is a complete oriented graph, that is a directed graph in which for
any two distinct vertices x and y either there is an edge from x to y or there is an
edge from y to x, but not both. Prove that every tournament contains a (directed)
Hamilton path.
13: Let G be a connected graph of order n and let 1 ::::; k ::::; n. Show that G contains
a connected subgraph of order k.
14~ Prove that the radius and diameter of a graph satisfy the inequalities
rad G ::::; diam G ::::; 2 rad G,
and both inequalities are best possible.
15:- Given d ;::: I, determine
max min{diam T: T is a spanning tree of G}.
diamG=d

16. Denote by fJo(G) the maximum number of independent vertices in G. Prove that
ifG does not contain a triangle then .1(G) ::::; /3o(G) and deduce that e(G) ::::; tn/3o(G),
where n = IGI.
IT Show that if for every vertex z of a directed graph there is an edge starting at z
(that is d+ (z) > 0) then the graph contains a (directed) cycle.

18. A grading of a directed graph G = V, if) is a partitioning of V into sets Vb V2 , . . , ~

such that if xy E if then x E J.-; and y E V;+ I for some i.
Given a directed graph G and a (non-directed) path P = XOXI ... x" denote
by v(xo, x,; P) the number of edges XiXi+ I minus the number of edges Xj+ IXj.
Prove that G has a grading iff v(xo, X,; P) is independent of P for every pair of
vertices X o , X,.
24 I Fundamentals

19. Complete the proof of Theorem 8 by showing in detail that both the second and
third methods construct an economical spanning tree.
20. Show how the fourth method in Theorem 8 can be applied to find an economical
spanning tree even if several edges have the same cost (cf. Figure 1.4).
21. Show that every economical spanning tree can be constructed by each of the first
three methods.
22. Deduce from Theorem 10 that a graph contains an Euler circuit iff all but at
most one of its components are isolated vertices and each vertex has even degree.
State and prove an analogous statement about the existence of an Eulerian trail
from x to y.
23. The following algorithm for finding an Euler circuit X t X 2 . . . Xn in a graph G
is due to Fleury. Pick x arbitrarily. Having chosen Xl' X2, . , Xk, put Gk =
G - {XIX2, X2 X3' .. , Xk-lXk}. If Xk is isolated in Gk> terminate the algorithm.
Otherwise let xk+ 1 be a neighbour of X k in Gk such that XkXk+ 1 is not a bridge,
. unless every edge of Gk incident with X k is a bridge.
Prove that if G has an Euler circuit then the trail X 1X 2 .. Xl constructed
by the algorithm is an Euler circuit.
24. A graph G is randomly Eulerian from a vertex x if any maximal trail starting at x
is an Euler circuit. (If T = xx 1 Xl then T is a maximal trail starting at x iff
Xl is an isolated vertex in G - E( T).) Prove that a non-empty graph G is randomly
Eulerian from X iff G has an Euler circuit and x is contained in each cycle of G.

25. Let F be a forest. Add a vertex x to F and join x to each vertex of odd degree
in F. Prove that the graph obtained in this way is randomly Eulerian from x.

26. Prove that a graph G is randomly Eulerian from each of two vertices x and y
iff G is the union of an even number of x-y paths, any two of which have only x
and y in common.

27. How would you define the number of sides of a face so that formula (4) continues
to hold for graphs with bridges? Rewrite the proof of Theorem 12 accordingly.

28.+ Prove that every planar graph has a drawing in the plane in which every edge is a
straight line segment. [Hint. Apply induction on the order of maximal planar
graphs by omitting a suitable vertex.]

29. A plane drawing of an infinite graph is defined as that of a finite graph with the
additional condition that each point has a neighbourhood containing at most one
vertex and meeting only edges incident with that vertex.
Show that Kuratowski's theorem does not hold for infinite graphs, that is
construct an infinite non-planar graph without TK 5 and T K 3.3.
Is there an infinite non-planar graph without a T K4?

30. Let d 1 :::; d 2 :::; . :::; dn be the degree sequence of a planar graph.
(i) By making use of an upper bound for 2:1 d i , show that if d 1 2': 4 then

2: df < 2(n + W- 62.

Notes 25

(ii) Prove by induction on n that if n ~ 4 then

I df ::; 2(n +W- 62.

I

Show that equality can hold for every n ~ 4.

31.+ Determine the maximum of II df where d l ::; d2 ::; .. ::; dn is the degree
sequence of a planar graph with girth at least 4 (that is without triangles). What is
the maximum if the girth is at least 9 > 4?
32. Fill in the small gap in the proof of Lemma 15: show that if cases (i) and (ii) do
not apply then there are two adjacent vertices of degree 4.

Notes
Theorem 14 is in K. Kuratowski, Sur Ie probIeme des courbes gauches en
topologie, Fund. Math. 15 (1930) 271-283; for a simpler proof see G. A.
Dirac and S. Schuster, A theorem of Kuratowski, Indag. Math. 16 (1954)
343-348.
The theorem ofS. A. Amitsur and J. Levitzki (Theorem 14) is in Minimal
identities for algebras, Proc. Amer. Math. Soc. 1 (1950) 449-463; the proof
given in the text is based on R. G. Swan, An application of graph theory to
algebra, Proc. Amer. Math. Soc. 14 (1963) 367-373 and Correction to " An
application of graph theory to algebra," Proc. Amer. Math. Soc. 21 (1969)
379-380.
CHAPTER II
Electrical Networks

This chapter is something of a diversion from the main line of the book, so
some readers may wish to skip it. Only the concepts introduced in the
first half of 3 will be used later, in 2 of Chapter VIII.
It does not take long to discover that an electrical network may be viewed
as a graph, so the simplest problems about currents in networks are exactly
questions about graphs. Does our brief acquaintance with graphs help us
to tackle the problems? As it will transpire in the first section, the answer is
yes; for after a short review of the basic ideas of electricity we make use of
spanning trees to obtain solutions. Some of these results can be reformulated
in terms of tilings of rectangles and squares, as we shall show in 2. The last
section introduces elementary algebraic graph theory which is then applied
to electrical networks.
It should be emphasised that in the problems we consider we use hardly
more than the terminology of graph theory; virtually the only concept to
be used is that of a spanning tree.

1 Graphs and Electrical Networks

A simple electrical network can be regarded as a graph in which each edge
ei has been assigned a real number ri called its resistance. If there is a potential
difference Pi between the endvertices of ej, say a and b, then an electrical
current Wi will flow in the edge ei from a to b according to Ohm's law:

26
l Graphs and Electrical Networks 27

Though to start with we could restrict our attention to electrical networks

corresponding to graphs, in the simplifications that follow it will be essential
to allow multiple edges, that is to consider multigraphs instead of graphs.
Furthermore, we orient each edge arbitrarily from one end vertex to the other
so that we may use Pi to denote the potential difference in the edge ei' meaning
the difference between the potentials of the initial vertex and the end vertex.
Similarly Wi is the current in the edge e;, meaning the current in ei in the
direction of the edge. (Note that we regard a negative current -Wi as a
positive current Wi in the other direction.) Thus, throughout the section we
consider directed multigraphs, that is directed graphs which may contain
several edges directed from a to b. However, in this section there is no danger
of confusion if we use ab to denote an edge from a to b; in the next section
we shall be more pedantic. Thus

In many practical problems electrical currents are made to enter the

network at some points and leave it at others, and we are interested in the
consequent currents and potential differences in the edges. These are
governed by the famous laws of Kirchhoff (another renowned citizen of
Konigsberg).
Kirchhoff's potential (or voltage) law states that the potential differences
round any cycle XIX2 ... Xk sum to 0:

Kirchhoff's current law postulates that the total current outflow from any
point is 0:
Wab + Wac + ... + Wau + Waco = O.
Here ab, ac, ... ,au are the edges incident with a, and Waco denotes the
amount of current that leaves the network at a. (In keeping with our con-
vention, Wcoa = -Waco is the amount of current entering the network at a).
For vertices not connected to external points we have
Wab + Wac + ... + Wau = O.
Note that if we know the resistances, then the potential law can be re-
written as a restriction on the currents in the edges. Thus we may consider
that the currents are governed by the Kirchhoff laws only; the physical
characteristics of the network (the resistances) affect only the parameters in
these laws.
It is also easily seen that the potential law is equivalent to saying that one
can assign absolute potentials Va, Yt" ... to the vertices a, b, ... so that the
potential difference between a and b is Va - Yt, = Pab. If the network is
connected and the potential differences Pab are given for the edges, then we
are free to choose arbitrarily the potential of one of the vertices, say Va, but
then all the other potentials are determined. In this section we shall work
28 II Electrical Networks

with the potentials, usually choosing the potential of one of the vertices to
be 0, but we must keep in mind that this is the same as the application of the
voltage law.
In the most fundamental problems current is only allowed to enter the
network at a single point s, the source, and only leave it at another point t,
the sink. (We shall indicate later that the general problem can be reduced
to these fundamental problems.) If the size of the current from s to t is w
and the potential difference between sand t is p, then by Ohm's law r = p/w
is the total resistance of the network between sand t. As an example of the
use of the Kirchhoff laws we shall evaluate the total resistance between sand
t of the simple network shown in Figure Ill.
s

.......- -... b .......- -... vb = 2(\ - e)

Figure 11.1. The resistances, the currents and the potentials.

This network has 5 resistors, of values 1,2,3,4 and 5 ohms, as shown in

the first picture. If we suppose that a unit current flows into the system at s
and leaves it at t, then the consequent edge currents must be as in the second
picture, for suitable values of e andf. Finally, the potentials Vr = 0, Va, V/" V.
assigned to the vertices must satisfy Ohm's law, so Va = 1 . e = e, V/, =
2(1 - e) and V. = Va + 5(e + f) = 6e + 5f. Ohm's law has to be satisfied
in two more edges, ab and bs, giving us
v,. = e = V/, + 3f = 2(1 - e) + 3f
and
v. = 6e + 5f = Vb + 4(1 - e - f) = 2(1 - e) + 4(1 - e - f).
Hence
e = 2 - 2e + 3f
and
+ 5f = 6 - 6e - 4{,
6e
giving e = 4/7, f = -2/21 and v" = 6e + 5f = 62/21. In particular, the
total resistance from s to t is (v" - Vr)/w = 62/21.
The calculations are often simplified if we note that Kirchhoff's equations
are linear and homogeneous in all currents and potential differences. This
implies the so-called principle of superposition: any combination of solutions
is again a solution. As an application of the principle of superposition one
l Graphs and Electrical Networks 29

can show that any current resulting from multiple sources and sinks can be
obtained by superposing flows belonging to one source and one sink, that is
solutions of the fundamental problems mentioned above can be used to
solve the general problem. Furthermore, the principle of superposition implies
immediately that there is at most one solution, no matter how the sources
and sinks are distributed. Indeed, the difference of two distinct solutions is a
flow in which no current enters or leaves the network at any point. If in
this flow there is a positive current in some edge from a to b, then by the
current law a positive current must go from b to c, then from c to d, etc.,
giving a trail abed . ... Since the network is finite, this trail has to return
to a point previously visited. Thus we obtain a circuit in whose edges positive
currents flow in one direction. But this is impossible since it implies that the
potential of each vertex is strictly greater than that of the next one round
the circuit.
Before proving the existence of a solution (which is obvious if we believe
in the physical interpretation), we shall calculate the total resistance of two
networks. Unless the networks are very small, the calculations can get very
heavy, and electrical engineers have a number of standard tricks to make them
easier.
The very simple networks of Figure 11.2 show two resistors r1 and r2
connected first in series and then in parallel. Let us put a current of size 1

Figure II.2. Resistors connected in series and in parallel.

through the networks, from s to t. What are the total resistances? In the first
case
v., = r1 and V. = v., + r2 = r1 + r2
so the total resistance is
r = r1 + r2'
In the second case, when they are connected in parallel, if a current of size e
goes through the first resistor and so a current of size 1 - e through the
second, then
30 II Electrical Networks

and the total resistance is given by

1 1 1
or - = - +-.
r r1 r2
This indicates that reciprocals of resistances, or conductances, are just as
natural as the resistances themselves, and indeed are slightly more convenient
in our presentation, especially in 3. (The conductance of an edge of resistance
1 ohm is 1 mho). What we have shown now is that for series connection the
resistances add and for parallel connection the conductances add.
The use of conductances is particularly convenient when considering
certain limiting cases of Ohm's law. If the resistance of an edge ab is 0, then
we necessarily have v" = JIb, and from an electrical point of view the vertices
can be regarded as identical. In the usual slang, a has been" shorted" (short-
circuited) to b. Of course, a may be shorted to b if there is some other reason
why v" = JIb. At the other extreme, we can introduce edges of 0 conductance
without effecting the currents and potentials. Conversely, we make an edge
have 0 conductance by "cutting" it. Of course, an edge of 0 resistance is said
to have 00 conductance and an edge of 0 conductance is said to have 00
resistance.
Let us see now how the acquaintance with resistors in series and in parallel
and the possibility of shorting vertices can help us to determine the total
resistance. As an example, let us take the network formed by the edges of a
cube, in which each edge has 1 ohm resistance. What is the total resistance
across an edge st? Using the notation of the first picture in Figure 11.3,
we see that by symmetry v" = .v" and ~ = VI' so c can be shorted to a and f
to d, giving us the second picture. From now on we can simplify resistors
connected parallel and in series, until we find that the total resistance is 172'
Knowing this, it is easy to recover the entire current flow.
Another important device in practical calculations is the so-called
star-delta transformation. If a vertex v is joined to just three vertices, say

I
b s tat b

e ] I! ! d ! e
f

Figure II.3.
r] t t d
Calculating the total resistance of a cube.
l Graphs and Electrical Networks 31

.{c
b~C
Figure 11.4. The star-delta transformation; S = AB + BC + CA.

a, band e, by edges of resistances A, Band C, then we call v the center of a

star, as in the first picture of Figure II.4. If no current is allowed to enter or
leave at v, then we are allowed to replace this star by the delta-configuration
shown in the second picture of Figure II.3, because, as the reader should
check, these two networks have the same total resistances across all three
pairs ab, be and ca. Of course we may apply the transformation in reverse,
replacing A', B', C' by A = B'C'/T, B = C'A'/T and C = A'B'/T, where
T = A' + B' + C'. Incidentally, the formulae become symmetrical if we
use resistances in the first transformation and conductances in the second:
A' = B + C + BCjA, and a = {3' + y' + {3'y'/a', where a, {3, ... are the
conductances.
As an application of the star-delta transformation, let us calculate the
total resistance of a tetrahedron across an edge, in which the resistances are
as in Figure II.5. The pictures speak for themselves.

J.
2
1
2
1 J.
4
1 5 !
'-4 l '-4

Figure II.5. Applications of the star-delta transformation.

We shall conclude this section on a slightly more theoretical note: we

shall prove the existence of a solution. More precisely, we shall show that if a
current of size 1 is put through a network then the current in an edge can be
expressed in terms of the numbers of certain spanning trees. For simplicity
we assume that the graph G of the network is connected, each edge has unit
resistance, and a current of size 1 enters at a vertex s and leaves at t.
32 II Electrical Networks

Theorem 1. Given an edge ab, denote by N(s, a, b, t) the number of spanning

trees of G in which the (unique) path from s to t contains a and b, in this order.
Define N(s, b, a, t) analogously and write N for the total number of spanning
trees. Finally, let Wab = {N(s, a, b, t) - N(s, b, a, t)}/N.
Distribute currents in the edges of G by sending a current of size wabfrom
a to b for every edge abo Then there is a total current size 1from s to t satisfying
the Kirchhoff laws.
PROOF. For each spanning tree T there is exactly one neighbour XT of s
that is on the Sot path P T contained in T. Hence Lber(s) N(s, s, b, t) = N,
where r(s) is the set of neighbours of s. Since N(s, b, s, t) = 0 for every vertex
bEr(S), we find that Lber(s)Wsb = 1. By symmetry Laer(t) Wat = 1; so
Kirchhoff's current law is satisfied at sand t provided a current of size 1
enters the network at s and leaves it at t.
To prove the theorem we have to show that ifno current is allowed to enter
or leave the network at any other point then the current and potential laws
are satisfied.
Let us take the current law first; to simplify the situation multiply all
currents by N. What is the contribution of a spanning tree T to the current
entering and leaving a vertex y distinct from sand t? If y is not on the Sot
path P T then T contributes nothing to N(s, x, y, t) or N(s, y, x, t). Now if y is
on the Sot path P T , say P T = s ... xyz ... t, then T contributes 1 to
N(s, y, z, t) and it also contributes 1 to N(s, x, y, t). In other words T con-
tributes a current of size 1 from y and a current of size 1 into y. Thus the
current law is satisfied.
As all edges have the same resistance, the potential law claims that the
total current in a cycle with some orientation is zero. To show this we proceed
as earlier, but first reformulate slightly the definition of N(s, a, b, t). Call a
forest F a thicket if it has exactly two components, say Fs and Fp such that
s is in Fs and t is in Ft. Then N(s, a, b, t) is the number ofthickets F = Fs u F t
for which a E Fs and bE F t and N(s, b, a, t) is defined analogously. What is
then the contribution of a thicket F = Fs u F t to the total current in a
cycle? It is the number of cycle edges from Fs to F t minus the number of
cycle edges from F t to Fs; so it is zero. 0

The proof above can be rewritten word for word to give a solution in
the case when the edges have arbitrary conductances. For a spanning tree T
define the weight w(T) of T as the product of the conductances of its edges.
Let N* be the sum of the weights of all the spanning trees, let N*(s, a, b, t)
be the sum of the weights of all the spanning trees in which b follows a on
the (unique) Sot path in the tree, and let N*(s, b, a, t) = N*(t, a, b, s).

Theorem 2. There is a distribution of currents satisfying Kirchhoff's laws in

which a current of size 1 enters at s and leaves at t. The value of the current in
and edge ab is given by {N*(s, a, b, t) - N*(s, b, a, t)}/N.* 0
2 Squaring the Square 33

Corollary 3. If the conductances of the edges are rational and a current of size
1 goes through the network then the current in each edge has rational value.
D

2 Squaring the Square

This is a diversion within a diversion; we feel bound to draw attention to a
famous problem arising from recreational mathematics which is related to
the theory of electrical networks. Is there a perfect squared square? In other
words, is it possible to subdivide a closed square into finitely many (but
at least two) square regions of distinct sizes that intersect only at their
boundaries?
Let us consider a squared rectangle, say the one shown in Figure II.6;
the number in a square is the length of its side. Let us cut this rectangle out
ofa sheet of nichrome (or any other material with low conductivity) and let us
put rods made of silver (or some other material of high conductivity) at the
top and bottom.

3 3

Figure II.6. A (non-perfect) squaring of a 6 x 7 rectangle.

What happens if we ensure that the silver rod at the top is at 7 volts while
the rod at the bottom is kept at O? Of course a uniform current will flow
from top to bottom. In fact the potential at a point of the rectangle will
depend only on the height of the point: the potential at height x will be x
volts. Furthermore, there will be no current across the rectangle, only from
top to bottom. Thus the current will not change at all if (i) we place silver
rods on the horizontal sides of the squares and (ii) cut narrow slits along
the vertical sides, as shown in the first picture of Figure II.7.
Now since silver is a very good conductor, the points of each silver rod
have been shortened, so can be identified. Thus as an electric conductor the
whole rectangle behaves like. the plane network shown in the second picture
of Figure II.7, in which the conductance of an edge is equal to the conductance
34 II Electrical Networks

-ys s
V,=7

a
a V. = 5

~t Y,=o
Figure II.7. The electrical network associated with our rectangle.

of the corresponding square from top to bottom. Clearly the conductance of

a rectangle from top to bottom is proportional to the length of a horizontal
side and the resistance is proportional to a vertical side. Consequently all
squares have the same resistance, say unit resistance, so all edges in Figure
II.7 have unit resistance. What is the potential drop in an edge? It is the side
length of the corresponding square. What is the resistance of the whole
system? The ratio of the horizontal side of the original big rectangle to the
vertical side, that is t.
Since the process above is reversible, that is every squared rectangle can
be obtained from some network we have an effective tool to help us in our
search for squared squares. Take a connected planar graph G and turn it
into an electrical network by giving each edge resistance 1. Calculate the
total resistance from a vertex s to a vertex t. If this is also 1, the network may
correspond to a suitably squared square. If the potential differences in the
edges are all distinct, all squares have different sizes, so we have a perfect
squared square.
Of course, at this stage our problem is far from being solved; we do not
even know that there must exist a squared square. However, we have a chance
to search systematically for a solution. Many squared squares have been

27
35
50
8\ 19
15 1217 111
~
Y..ll 24
25 18
29 16
4

37 42
33

Figure II.8. A perfect squared square: a tiling of a square with 21 incongruent squares.
3 Vector Spaces and Matrices Associated with Graphs 35

found with the help of computers, but the first examples were found without
computers by Sprague in 1939 and by Brooks, Smith, Stone and Tutte in
1940. The smallest number of squares that can tile a square is 21; Figure 11.8
shows this tiling, due to Duijvestijn. In fact, this is the only tiling of order 21.
The connection between squaring a rectangle and electrical networks
gives us immediately a beautiful result first proved by Dehn in 1903. Corollary
3 tells us that if each edge has resistance 1 and a current of size 1 flows
through the system then in each edge the value of the current is rational.
This translates to the following result about squared rectangles.

Theorem 4. If a rectangle can be tiled with squares then the ratio of two
neighbouring sides of the rectangle is rational. 0
Equivalently: a rectangle can be tiled with squares iff it can be tiled with
congruent squares.
It is easily seen that electrical networks can be used to obtain tilings of
rectangles of prescribed shapes: an edge e of resistance r corresponds to a
rectangle in which the height is r times the base. The reader is encouraged to
solve the exercises based on this simple idea (Exercises 9-12); some other
exercises concern related packing problems (Exercises 13-15).

3 Vector Spaces and Matrices Associated with

Graphs
The vertex space Co(G) of a graph G is the complex vector space of all
functions from V( G) into C. Similarly the edge space C I (G) is the complex
vector space of all functions from E(G) into C. In these definitions it is
sometimes convenient to replace the complex field by F 2, the field of order 2,
or by other fields. We shall take V(G) = {Vb V2, ... ' vn } and E(G) =
reb e2'' em} so that dim Co(G) = nand CI(G) = m. The elements of
Co(G) are usually written in the form x = Ii'=1 XiVi or x = (Xi)'i. The sum
Ii'=1 Xi Vi is a formal sum of the vertices but if we think of Vi as the function
V(G) ~ C which is 0 everywhere, except at the vertex Vi' where it is 1, then
VI' ... ' Vn is a basis of Co(G) and the sum above simply expresses an element
in terms of the basis elements. Similarly an element of C I (G) may be written
as y = I~ I Yiei or Y = (Yi)7 We call (VI' ... , vrn) the standard basis of the
vertex space Co(G) and (e 1 , . . . , em) the standard basis of the edge space.
We shall endow these spaces with the inner product in which the standard
bases are orthonormal: <x, y) = Ii XiYi.
In this section we shall be concerned mostly with the edge space CI(G);
to start with we define two subspaces which will turn out to be orthogonal
complements of each other. Let L be a cycle in G with a given cyclic orienta-
tion L = U I U 2 .. u/. If ei ~ UjU j + I and ei is oriented from uj to uj + I then
36 II Electrical Networks

Figure 11.9. If the thick cycle L is oriented anti-clockwise, its vector in C 1(L) is
ZL = ( - 1, 1, 1, - 1, 1, 0, ... , 0).

we say that ei is oriented as L. This oriented cycle L can be identified with

an element ZL of C I (G):
1 if ei E E(L) and ei is oriented as L,
zL(e;) = { -1 if ei E E(L) and ei is not oriented as L,
a if ei ~ E(L).
A simple example is shown in Figure II.9. Denote by Z(G) the subspace
of CI(G) spanned by the vectors ZL as L runs over the set of cycles; Z(G) is
the cycle space of G.
Let now P be a partition V = VI U V2 of the vertex set of G. Consider the
set (VI' V2 ) of edges from VI to V2 ; such a set of edges is called a cut. There
is a vector up in CI(G) called a cut vector, naturally associated with this
partition P:
1 if ei goes from VI to V2
up(ei) = { -1 if ei goes from V2 to VI
a if e ~ E(VI , V2 ).
We write U(G) for the subspace of the edge space CI(G) spanned by all the
cut vectors up; U(G) is the cut (or cocycle) space of G.

Theorem 5. The inner product space C I (G) is the orthogonal direct sum of
the cycle space Z(G) and the cut space U(G). If G has n vertices, m edges and
k components then
dim Z(G) = m - n +k and dim U(G) = n - k.
PROOF. Let us see first that Z(G) and U(G) are orthogonal. Let L be a cycle
and P a partition V = VI U V2 What is the product (ZL' up)? It is simply
the number of edges of L going from VI to V2 in the orientation of L, minus
the number of edges of L from V2 to VI' Thus (ZL, up) = a for every cycle L
and partition P, so Z(G) and U(G) are indeed orthogonal.
Since the dimension of CI(G) is the number of edges, m, both assertions
will be proved if we show that dim Z(G) ;;::: m - n + 1 and dim U(G) ;;:::
n - 1. We shall first prove this under the assumption that G is connected;
the general case will follow easily.
3 Vector Spaces and Matrices Associated with Graphs 37

Thus let us assume that G is connected, that is k = 1. Let T be a spanning

tree of G. We shall make use of T to exhibit m - n + 1 independent vectors
in Z(G) and n - 1 independent vectors in U(G). We may choose the indices
of the edges in such a way that el' e2"'" en-l are the tree edges and en'
en+ 1, ... , em are the remaining edges, the chords of T.
We know that for every chord ei (i 2=: n) there is a (unique) oriented cycle
C such that zc.(e;) = 1 and zc.(e) = 0 for every other chord ej' that is
whenever j 2=: nand j #- i. (Shortly: zc.(e) = ()ij if j 2=: n, where ()ij is the
Kronecker delta.) We call Ci the fundamental cycle belonging to ei (with
respect to T); zc; is a fundamental cycle vector (see Figure II.10). Similarly by

Figure 11.10. The fundamental cycle vector belonging to e9 is zc; = e9 - e 2 + e1 +

e4 - e6 , the fundamental cut vector belonging to e4 is up, = e4 - e lO - e9 .

deleting an edge ei of T the remainder of the spanning tree falls into two
components. Let Vii be the vertex set of the component containing the
initial vertex of ei and let V~ be the vertex set of the component containing
the terminal vertex of ei' If Pi is the partition V = V~ u V~ then clearly
uple) = ()ij for 1 ~ j ~ n - 1. The cut E(VL V~) is the fundamental cut
belonging to ei (with respect to T) and up; is the fundamental cut vector.
It is easily seen that {zc;: n ~ i ~ m} is an independent set of cycle vectors.
Indeed, if z = Li"=n AiZC; = 0 then for every j 2=: n we have 0 = z(e) =
Li"=n Ai()ij = Yj and so every coefficient A. j is O. Similarly the fundamental
cut vectors up;, 1 ~ i ~ n - 1, are also independent. Hence dim Z(G) 2=:
m - n + 1 and dim U(G) 2=: n - 1, as required.
Finally, the general case k 2=: 1 follows immediately from the case k = 1.
For if G has components G1 , G2 , ... , Gk then C 1 (G) is the orthogonal direct
sum of the subspaces C 1(G;), i = 1,2, ... , k; furthermore Z(G i ) =
Z(G) n C 1 (G i ) and U(G;) = U(G) n C 1(G;), 0

The proof above shows that dim Z(G), called the cyclomatic number of
G, and dim U(G) are independent of the field over which the edge space is
defined. The use of a spanning tree in the proof is not compulsory; in some
cases, for instance in the case of a planar graph, there are other natural cycle
and cut bases (cf. Exercise 16).
38 II Electrical Networks

There are several matrices naturally associated with a graph and its
vector spaces discussed above. The adjacency matrix A = A(G) = (ai) of a
graph G is the n x n matrix given by
a ..
I)
= {I ifvivjEE(G),
0 otherwise.
In order to define the incidence matrix of a graph we again consider an
orientation of the edges, as in the definition of the cycle and cut spaces. The
incidence matrix B = B( G) = (b i) of G is the n x m matrix defined by
I if Vi is the initial vertex of the edge ej'
bij = { -1 if Vi is the terminal vertex of the edge ej'
o otherwise.
There is a simple connection between the two matrices A and B.

Theorem 6. BBt = D - A, where Bt denotes the transpose of Band D is the

n x n diagonal matrix in which (D)ii is d(v;), the degree of the vertex Vi in G.
PROOF. What is (BBt)i/ It is Li= 1 bi/bjz, which is d(vi) if i = j, -1 if ViVj is
an edge (if el = ViVj is directed from Vi to Vj' then bi/bjl = 1( -1) = -1
and all other products are 0), and 0 if Vi Vj is not an edge and i # j. 0
We may and will identify the matrices A and B with the linear maps
A: Co( G) -+ C o( G) and B: C 1 (G) -+ C o( G) they define in the standard bases:
(AX)i = LJ= 1 aijXj and (BY)i = Lj= 1 bijYj If we wanted to be pedantic,
we would write the vectors in the vertex and edge spaces as column vectors,
or we would put Axt and Byt, where t stands for transposition; we shall not
do this since there is no danger of confusion. If C is a cycle then clearly
Bzc = 0 E C o( G); in fact, it is easily shown (cf. Exercise 17) that the cycle
space is exactly the kernel of B. Thus the rank of B is reB) = m - (m - n + k)
= n - k. Furthermore, the transpose of B maps Co(G) into C1(G), and the
image of Bt is exactly the cut space (cf. Exercise 18).

In Chapter VIII we shall discuss in some detail the eigenvalues and eigen-
vectors of the adjacency matrix; in this section we shall use the matrices to
solve the electrical network problem discussed in the first section. In fact,
it was Kirchhoff who first realized the applicability of matrix algebra to
graph theory, exactly in connection with the electrical network problem.
How can we formulate the Kirchhoff laws in terms of matrices and vectors
in the edge space? Let us assume that G' is the graph of our electrical network;
V(G') = {Vb V2,"" Vn- d, E(G') = {el, e2,' em'}' the network is con-
nected and we have a voltage generator ensuring that the potential difference
between Vi and Vj is gi - gj volts for 1 :::; i < j :::; k. In order to express
Kirchhoff's laws in a neat form, we add a vertex Vn to G', and join it to
Vb V2"'" Vk; the new graph is G. Let m = m' + k and em'+l = VnVi, i =
1,2, ... , k, so that V(G) = {Vl, C2"'" vn} and E(G) = {el, e2"'" em}.
3 Vector Spaces and Matrices Associated with Graphs 39

Give the edges of G' an arbitrary orientation and let Wi be the amount of
current flowing in the edge ei in the direction of ei; thus Wi = -1 means a
current of 1 ampere in the opposite direction. Direct each new edge em' + i
from Vn to Vi and let W m , + i be the total current entering the network at Vi.
Once again, W m' + i = - 1 means that a current of 1 ampere leaves the network
at Vi. The vector w = (W b W2' ' .. , Wm ) E C 1(G) is the current vector. In this
notation Kirchhoff's current law takes the form
Bw= O. (1)
We have no greater difficulty in formulating Kirchhoff's potential law
in matrix form. Let Pi be the potential difference in the edge ei and let P =

(PI, P2' . , . , Pm) E C I (G) be the potential vector. The potential law states that
<z, p) = for every cycle Z E C I (G). Instead of postulating this about every
cycle, we collect all necessary information into a single matrix. As before, we
choose a spanning tree Tin G and label the edges so that el' e2' "., en-I
are the tree edges and en, en+ b .. " em are the chords, Let C be the m x
(m - n + 1) matrix whose ith column is the fundamental cycle vector
ZC n _l+; belonging to the edge en -1+i, i = 1,2, ... , m - n + 1. Since the
fundamental cycle vectors form a basis of the cycle space, the potential law
takes the form
Cp = 0, (2)
where C denotes the transpose of C.
Now in order to find the current through the edges of G' we need one
more equation, namely the equation relating the potential to the current,
the conductance and the voltage generator. For i ::s; m' let di be the con-
ductance of the edge ei and postulate that each new edge ej (j ~ m' + 1)

has conductance dj = 0. We may assume that di > for every i ::s; m' since
otherwise the edge ei could have been cut. Let D be the m x m diagonal
matrix with (D)ii = d i Finally, let g = (0, ... ,0, gl' g2' , .. , gk) E C1(G) be
the vector of the voltage generator. Then clearly
p = Dw + g. (3)
This equation contains all the information we have about the electric current
in addition to the Kirchhoff laws,
In order to solve (1), (2) and (3) forw and p, we shall split C 1 (G) as ET + EN,
where ET is the subspace spanned by the tree edges and EN is spanned by the
chords, the edges not belonging to T. Let w = (WT' wN ) and p = (PT, PN)
be the corresponding splittings; furthermore, writing 1J for the matrix
obtained from B by omitting the last row, we have

C = (~:) and 1J = (BTB N ).

As the columns of C are the fundamental cycles, C N is the (m - n + 1) x

(m - n + 1) identity matrix 1m _n + I. Since the kernel of B contains the cycles,
40 II Electrical Networks

BC
= and so BC = 0, giving BTCT = -BN. Now BT is invertible, as the
reader should check (cf. Exercise 19), so
CT = -Bi1BN
After this preparation we can easily solve our equations.

Theorem 7. The electric current W satisfying p = Dw + g is given by w =

- C(CDC)-lCg.

PROOF. Equation (1) implies that BTw T + BNw N = 0, so WT = -Bi1BNw N

= CTW N. Hence w = Cw N. Combining (2) and (3) we find that CDw +
Ctg = 0 and so (CDC)w N = - Cg. As CDC is easily shown to be invertible,
the result follows. 0

Clearly Theorem 7 is valid in a somewhat more general situation, not

only when G and g are defined as above. In fact, the following conditions are

sufficient (and more or less necessary) for the existence of a unique current
w: gA = for every i and the edges ej with dj > form a connected subgraph.
Furthermore, the results hold for multigraphs, that is "graphs" in which
several edges may join the same pair of vertices. The reader may find it
necessary to check that all the concepts (incidence matrix, cycle and cut
spaces, fundamental cycles and cuts) can be defined as before and the proofs
of the results remain unchanged.
By considering multigraphs one can set up Theorem 7 in a slightly simpler
form without adding a new vertex to the graph G' of the network. Thus if the

current enters G' at a vertex and leaves it at a vertex b, then we join a to b by a
new edge e of conductance (even if a and b had been joined before), and
postulate (by choosing g = (0, 0, ... , 0, 1), where e is the last edge) that the
potential difference in e is 1. Using this set-up one can check that the ratio
of the current in ei to the total current (that is the current in e) is indeed
given by Theorem 1 of 1, though this checking is rather tedious and involved.
On the other hand, it is very easy to express the total number of spanning
trees in a graph in terms of the incidence matrix.
To do this we make use of two facts. The first is an extension ofthe observa-
tion that BT is invertible: the modulus of the determinant of an (n - 1)
x (n - 1) submatrix of B is 1 if the edges corresponding to the columns

form a tree and otherwise (cf. Exercise 19). The second is the Cauchy-Binet
formula oflinear algebra, stating that if K is a p x q matrix (p :::; q) and L is a
q x p matrix then det KL = Lpdet KpLp = Lpdet Kp det Lp. Here the
summation is over all p-subsets P of {l, 2, ... , q}, and Kp is the p x p sub-
matrix of K formed by the columns of K indexed with elements of P and Lp is
the submatrix of L formed by the corresponding rows of L. Putting the two
together we arrive at the following formula.

Theorem 8. The number of spanning trees of G is det BBt. o

Exercises 41

EXERCISES

1.- Verify the formulae in the star-delta transformation.

In Exercises 2-6 all edges are assumed to have unit resistance.
2.- Calculate the resistance of the network shown in Figure Ll measured between the
vertices 2 and 3.
3. For each different pair of vertices of a cube calculate the resistance between them.
4. What is the resistance between two adjacent vertices of (a) an octahedron, (b) a
dodecahedron and (c) an icosahedron?
5. Suppose each edge of a connected network is in the same number of spanning
trees. Prove that the total resistance between two adjacent vertices is (n - 1)/e,
where n is the order and e is the size of the network. Verify your answers to
Exercise 4.
6. By applying suitable star-delta transformations, calculate the resistance of a
dodecahedron between the midpoints of two adjacent edges.
7. Give a detailed proof of Theorem 2.
8. Find the squared rectangle indicated in Figure ILl I.

Figure 1I.11. A squaring of the 69 x 61 rectangle.

9. Which squared rectangle corresponds to the network in Figure 1I.12? Rotate the
rectangle through 90 and draw the network for this rectangle.

Figure 1I.12. A plane network.

10.- How many essentially different squared rectangles correspond to the network
of the cube in Exercise 3?
42 II Electrical Networks

11. Find the tiling associated with Figure 11.13.

t 2

Figure 11.13. A network with edges of differing resistance.

12. Find the simple perfect squared square given by the network in Figure 11.14.
(This example was found in 1964 by J. Wilson.)

Figure 11.14. A network giving a perfect square.

13. Show that an equilateral triangle cannot be dissected into finitely many in-
congruent equilateral triangles.
14.- Prove that if a rectangular parallelepiped can be decomposed into cubes then the
ratios of its sides are rational.
15. Show that a cube cannot be dissected into finitely many incongruent cubes.
16. Show that in a plane graph the boundaries of bounded faces form a. cycle basis.
17. Show that the cycle space is the kernel of the map C I (G) -+ Co( G) defined by the
incidence matrix B.
18. Let B' be the transpose of B. Show that the cut space is the image of the map
Co(G) -+ C1(G) defined by B'.

19. Let F be a set of n - 1 edges of a graph of order n with incidence matrix B. Let
Bf" be an (n - 1) x (n - 1) submatrix of B whose columns correspond to the
edges of F. Prove that BF is invertible iff F is the edge set of a tree.
20. Deduce from Theorem 8 that there are n" - 2 trees on n distinguishable vertices.
Notes 43

Notes
The first squared square was published by R. Sprague, Beispiel einer Zer-
legung des Quadrats in lauter verschiedene Quadrate, J. Reine Angew. Math.
182 (1940) 60-64, closely followed by R. L. Brooks, C. A. B. Smith, A. H.
Stone and W. T. Tutte, The dissection of rectangles into squares, Duke Math.
J. 7 (1940) 312-340. The square shown in Fig. 8 was published in A. J. W.
Duijvestijn, Simple perfect square of lowest order, J. Combinational Theory
Ser. B 25 (1978) 240-243.
The original proof of Theorem 4 is due to Max Dehn, Uber die Zer-
legung von Rechtecken in Rechtecke, Math. Ann. 57 (1903) 314-322. Two
survey articles the reader may wish to look at are W. T. Tutte, The quest of
the perfect square, Amer. Math. Monthly 72 (1965),29-35 and N. D. Kaza-
rinoff and R. Weitzenkamp, Squaring rectangles and squares, Amer. Math.
Monthly 80 (1973) 877-888.
CHAPTER III
Flows, Connectivity and Matching

The results of this chapter have many applications in diverse branches in

mathematics; in particular, Hall's marriage theorem is a useful tool in
algebra and analysis. Every result we are going to present is a necessary and
sufficient condition for the existence of certain objects: in each case the
beauty of the theorem is that a condition whose necessity is obvious is shown
to be also sufficient. In the natural formulation of our results we shall have
two functions, say J and g, clearly satisfyingJ :s; g, and we shall show that
max J = min g. The results of the chapter are closely interrelated so the
order they are proved in is a matter of taste; to emphasize this, some results
will be given several proofs.
In the first chapter we looked at the simplest properties of connectivity.
Now we shall go a step further and in 2 we shall prove the basic theorem
about connectivity, namely Menger's theorem, first proved in 1927. This
theorem turns out to be one of the many consequences of a fundamental
theorem about flows in directed graphs, the max-flow min-cut theorem.
Though this book is devoted almost exclusively to undirected graphs, we
introduce in the first paragraph the concept of a flow in a directed graph
and go on to prove the max-flow min-cut theorem, partly because it pro-
vides one of the simplest proofs of Menger's theorem and partly because
it has a number of other important consequences concerning undirected
graphs.
A set of independent edges in a graph is called a matching. In a bipartite
graph a matching can be identified with a flow and the max-flow min-cut
theorem takes the form of another widely applicable combinatorial result,
Hall's marriage theorem. We discuss various forms of this result in 3.
Though we deduce the theorems of both Menger and Hall from the
max-flow min-cut theorem, the results are in fact very closely related, and

44
l Flows in Directed Graphs 45

because of their fundamental importance we give also independent proofs

of each.
Hall's theorem tells us, in particular, when a bipartite graph has got a
ijactor, a subgraph whose vertex set is that of the original graph and in
which every vertex has degree 1. The question of the existence of a i-factor
in an arbitrary graph is considerably harder. It is answered by a theorem of
Tutte that we present in 4.

1 Flows in Directed Graphs

Let G be a (finite) directed graph with vertex set V and edge set if. We shall
study (static) flows in G from a vertex s (the source) to a vertex t (the sink).
A flow Jis a non-negative function defined on the edges; the value J(xy)
is the flow or current in the edge xy. For notational simplicity we shall write
J(x, y) instead ofJ(xy) and a similar convention will be used for other func-
tions. The only condition a flow from s to t has to satisfy is Kirchhoff's
current law: the total current flowing into each intermediate vertex (that is
vertex different from sand t) is equal to the total current leaving the vertex.
Thus if for x E V we put

r+(x) = {y E V: xy E if},
r-(x) = {y E V: yx E if},

then a flow Jrom s to t satisfies the following condition:

L
YEf+(x)
J(x, y) = L
ZEf-(x)
J(z, x) for each x E V - {s, t}.

Since

o= L {L
XEV-{S,t} YEf+(x)
J(x, y) - L
ZEf-(x)
J(z, X)}

uItJ tJ~(u/(Z' u) - YEh(u/(u, Z)}

we find that

L
YEf+(s)
J(s, y) - L
YEf-(s)
J(y, s) = L
YEf-(t)
J(Y, t) - L
YEf+(t)
J(t, y).

The common value, denoted by v(f), is called the value ojJ or the amlJunt oj
flow from s to t.
We wish to determine the maximal flow value from s to t provided the
flow satisfies certain constraints. First we shall deal with the case when the
46 III Flows, Connectivity and Matching

so called capacity of an edge restricts the current through the edge. It will
turn out that several other seemingly more complicated restrictions can be
reduced to this case.
Let us fix our directed graph G = (V, ) and two vertices in it, say sand
t. With each edge xy of G we associate a non-negative number c(x, y), called
the capacity of the edge. We shall assume that the current flowing through
the edge xy cannot be more than the capacity c(x, y).
Given two subsets X, Y of V we write (X, Y) for the set of directed
X-Yedges:
(X, Y) = {XY:XEX,yE Y}.

Whenever g: -+ IR is a function, we put

g(X, Y) = I g(x, y),
where the summation is over (X, Y). If S is a subset of V containing s but
nott then (S, S) is called a cut separating s from t. (Here S = V - S.) If
we delete the edges of a cut then no positive-valued flow from s to t can be
defined on the remainder. Conversely, it is easily seen that if F is a set of
edges after whose deletion there is no flow from s to t (that is v(f) = 0 for
every flow from s to t) then F contains a cut (Exercise 1). The capacity of a
cut (S, S) is c(S, S) (see Figure III.l). It is easily seen (Exercise 2) that the
capacity of a cut is at least as large as the value of any flow, so the minimum
of all cut capacities is at least as large as the maximum of all flow values.
The celebrated max-flow min-cut theorem of Ford and Fulkerson states
that this trivial inequality is, in fact, an equality. Before stating this theorem
and getting down to the proof, let us justify the above use of the words
minimum and maximum. Since there are only finitely many cuts, there is a
cut whose capacity is minimal. The existence of a flow with maximal value
is only slightly less trivial. Indeed,
v(f) ~ I c(x, y)
xyEif

for every flow I, so v = sup v(f) < 00. Let II> 12, ... be a sequence of flows
with limn v(fn) = v. Then, by passing to a subsequence, we may assume that
for each xy E the sequence (fix, y is convergent, say to I(x, y). The

"' ... -- ...... ,, ,,

,
/
/
I
I
I
5 I
\
\
I \ I
I
I
3 \ I
\
I \
I 4 1 7 9
I 1
IS r
\
\
,
r

\
I
I
\
2 \
,, \ I
,, /
I
6
\
,, t I
I
/
-'
Figure IIU. A cut with capacity 12. (The numbers next to the edges indicate their
capacity.)
l Flows in Directed Graphs 47

functionfis a flow with value v, that is a flow with maximal value. In a similar
way one can show that even if some of the edges have infinite capacity, there
is a flow with maximal value which can be either finite or infinite (Exercise 3).

Theorem 1. (Max-flow min-cut theorem.) The maximal flow value from s to t

is equal to the minimum of the capacities of cuts separating s from t.
PROOF. We have remarked already that there is a flow fwith maximal value,
say v, and the capacity of every cut is at least v. Thus, in order to prove the
theorem we have to show that there is a cut with capacity v. We shall, in
fact, do considerably more than this: we shall give a very simple procedure
for constructing such a cut from a flow fwith maximal value.
Define a subset S c V recursively as follows. Let s E S. If XES and
c(x, y) > f(x, y)
or
f(y, x) > 0
then let yES.
We claim that (S, S) is a cut separating s from t with capacity v = v(f).
Let us see first why t cannot belong to S. If t belongs to S, we can find vertices
Xo = s, Xl' ... , Xl = t such that

Ci = max{c(xi, Xi+1 ) - f(Xi' Xi+l),f(Xi+l, x;)} > 0

for every i, 0 sis I - 1. Put C = mini Ci' Then f can be augmented to a
flow f* in the following way: if Ci > f(Xi+ 1> x;) then increase the flow in
XiXi+ 1 by c, otherwise decrease the flow in Xi+ lXi by c. Clearly f* is a flow
and its value is v(f*) = v(f) + c, contradicting the maximality of f. This
shows that t tt s so (S, S) is a cut separating s from t.
Now v(f) is equal to the value of the flow from S to S defined in the ob-
vious way:
L f(x, y) - L f(x, y).
XES.YES XES. YES

By the definition of S the first sum is exactly

L c(x, y) = c(S, S)
XES, YES

and each summand in the second sum is zero. Hence c(S, S) = v(f), as
required. D

The max-flow min-cut theorem is the cornerstone of the theory to be

presented in this chapter. Note that the theorem remains valid (with ex-
actly the same proof) if some of the edges have infinite capacity but the
maximal flow value is finite.
The proof of the theorem also provides a surprisingly efficient algorithm
for finding a flow with maximal value if the capacity function is integral,
48 III Flows, Connectivity and Matching

that is if c(x, y) is an integer for every edge xy. We start with the identically
zero flow: fo(x, y) = 0 for every xy E E. We shall construct an increasing
sequence of flows fo, f1' f2' ... that has to terminate in a maximal flow.
Suppose we have constructed!;. As in the proof above, we find the set S
belonging to j;. Now if t f. S then!; is a maximal flow (and E(S, S) is a minimal
cut) so we terminate the sequence. If, on the other hand, t E S, then!; can be
augmented to a flow!; + 1 by increasing the flow along a path from s to t.
Since each v(!;) is an integer, we have v(!;) ~ V(!;-1) + 1, and the sequence
must end in at most Lx,y c(x, y) steps.
Moreover, if c is integral the algorithm constructs a maximal flow which
is also integral, that is a flow whose value is an integer in every edge. Indeed,
fo is integral and if!;_1 is integral then so is!;, since it is obtained from!;_1
by increasing the flow in a path by a value which is the minimum of positive
integers. This result is often called the integrality theorem.

Theorem 2. If the capacity function is integral then there is a maximal flow

which is also integral. 0

We shall rely on this simple result when we use the max-flow min-cut
theorem to find various paths in graphs. It is important to note that the
results do not claim uniqueness: the algorithm finds one of the maximal
flows (usually there are many) and Theorem 2 claims that one ofthe maximal
flows is integral.
The existence of the algorithm proves some other intuitively obvious
results as well. For instance, there is a maximal acyclic flow, that is one which
does not contain a flow around a cycle X 1 X 2 ' .. , x k :

It may seem somewhat surprising that if instead of one source and one
sink we take several of each, the situation does not become any more com-
plicated. The only occasion we have to be a little more careful is the defini-
tion of a cut. If S 1, ... , Sk are the sources and t 1, ... ,t l are the sinks then
E(S, S) is a cut if Si E Sand tj E S for every i, j, 1 :5: i :5: k, 1 :5: j ::; I.
In order to be able to apply the max-flow min-cut theorem, let us add a
new source S and a new sink t to G, together with all the edges SSi and tjt
of infinite capacity. Let Ii be the graph obtained in this way. Consider those
flows from S 1, ... , Sk to t 1, ... , tl in G in which the total current entering
(leaving) a source (sink) is not greater than the total current leaving (en-
tering) it. These flows can easily be extended to a flow from S to t in Ii and
this extension establishes a 1-1 correspondence between the two sets of
flows. Furthermore, a cut separating S from t in Ii that has finite capacity
cannot contain an edge _of the form SSi and tjt, so it corresponds to a cut
of the same capacity in G, separating S 1, ... , Sk from t 1, ... , t l . Thus Theo-
rem 1 has the following extension.
l Flows in Directed Graphs 49

Theorem 3. The maximum of the flow value from a set of sources to a set of
sinks is equal to the minimum of the capacity of cuts separating the sources
from the sinks. 0

Let us assume now that we have capacity restrictions on the vertices,

except the source and the sink. Thus we are given a function c: V - {s, t} - t
IR + and every flow f from s to t has to satisfy the following inequality:
I f(x, y) = L f(z, x) ~ c(x) for every x E V - {s, t}.
YE r+ (xl ZE r- (xl

How should we define a cut in this situation? A cut is a subset S of V - {s, t}

such that no positive-valued flow from s to t can be defined on G-S. (In
order to distinguish the two kinds of cuts we sometimes call this a vertex-cut
and the other one an edge-cut. However, it is almost always clear which cut
is in question.) Can we carryover the max-flow min-cut theorem to this
case? Yes, very easily, if we notice that a flow can be interpreted to flow in a
vetex as well, namely from the part where all the currents enter it to the part
where all the currents leave it. More precisely we can turn each vertex of G
into an edge (without changing the nature of the directed graph) in such a
way that any current entering (and leaving) the vertex will be forced through
the edge. To do this, replace each vertex x E V - {s, t} by two vertices, say
x _ and x + , send each incoming edge to x _ and send each outgoing edge to
x+. Finally, add edges from x_ to x+ with capacity c(x_, x+) = c(x) (see
Figure 111.2). There is a simple 1-1 correspondence between the flows from

'm,
x

.~
Figure III.2. Replacing a graph Gwith restrictions on the capacity of the vertices by a
graph H with restrictions on the capacity of the edges.

s to t in G satisfying the capacity restrictions on the vertices and the flows

in the new graph Ii satisfying the capacity restrictions on (some of) the
edges. Since in H only the edges Lx:;. have finite capacities, an edge-cut
of finite capacity in H consists entirely of edges of the form Lx:;. , so it
corresponds to a vertex-cut in G of the same capacity. Thus we have the
following form of Theorem 1.

Theorem 4. Let G be a directed graph with capacity bounds on the vertices

other than the source s and the sink t. Then the minimum of the capacity of a
vertex-cut is equal to the maximum of the flow value from s to t.
50 III Flows, Connectivity and Matching

Theorems 1, 3 and 4 can easily be combined into a single theorem. We

leave this to the reader (Exercise 6).

2 Connectivity and Menger's Theorem

RecalI that if any two vertices of a graph can be joined by a path then the
graph is said to be connected, otherwise it is disconnected. A maximal con-
nected subgraph of a graph G is a component of G.
If G is connected and G- W is disconnected, where W is a set of vertices
or a set of edges, then we say that W separates G. If in G- W two vertices s
and t belong to different components then W separates s from t. We say
that a graph G is k-connected (k ~ 2) if either G is aK k + 1 or else it has at
least k + 2 vertices and no set of k - 1 vertices separates it. Similarly G is
k-edge-connected (k ~ 2) if it has at least two vertices and no set of at most
k - 1 edges separates it. A connected graph is also said to be I-connected
and l-edge-connected. The maximal value of k for which a connected graph
G is k-connected is the connectivity of G, denoted by K(G). (If G is discon-
nected, we put K(G) = 0.) The edge-connectivity A(G) is defined analogously.
Clearly a graph is 2-connected iff it is connected, has at least 3 vertices
and contains no cut vertex. Similarly, a graph is 2-edge-connected iff it is
connected, has at least 2 vertices and contains no bridge. It is often easy to
determine the connectivity of a given graph. Thus if 1 ~ I ~ n then K(pl) =
A(pl) = 1, K(e") = A(e") = 2, K(Kn) = A(Kn) = n - 1 and K(KI,n) = A(KI,n)
= l. In order to correct the false impression that the vertex-connectivity is
equal to the edge-connectivity, note that if G is obtained from the disjoint
union of two Kl by adding a new vertex x and joining x to every old vertex,
then K(G) = 1, since x is a cut vertex, but A(G) = I (see also Exercise 11).
This last example shows that A(G - x) may be 0 even when A(G) is large.
However, it is clear from the definitions that for every vertex x and edge xy
we have

K(G) - 1 ~ K(G - x) and A(G) - 1 ~ A(G - xy) ~ A(G).

If G is non-trivial (that is has at least two vertices), then the parameters

b( G), ,1,( G) and K( G) satisfy the folIowing inequality:
K(G) ~ A(G) ~ b(G).

Indeed, if we delete alI the edges incident with a vertex, the graph becomes
disconnected, so the second inequality holds. To see the other inequality,
note first that if G is complete then K( G) = ,1,( G) = IG I - 1, and if K( G) ~ 1
then A(G) = K(G). Suppose now that G is not complete, A(G) = k ~ 2 and
{X 1 Yl,X2Yz, . ,Xk Yk} is a set of edges disconnecting G. If G-
{Xl' X2"'" Xk} is disconnected then K(G) ~ k. Otherwise each vertex Xi
2 Connectivity and Menger's Theorem 51

Figure 111.3. A 4-edge-connected graph G such that G - Xl' XZ, X3, X 4 is connected.

has degree at most k (and so exactly k), as shown in Figure 111.3. Deleting
the neighbours of Xl' we disconnect G. Hence K = A(G).
Another property immediate from the definition of vertex-connectivity
is that if G l and Gz are k-connected (k ~ 1) subgraphs of a graph G having
at least k common vertices, then Gl U G z is also k-connected. Indeed if
W c V(G l ) U V(G z ) has at most k - 1 vertices, then there is a vertex X
in V(G l ) n V(Gz)\w. Therefore the connected subgraphs Gl - Wand
G z - W of G have at least one vertex, namely x, in common, so G l U G z -
W = (G l - W) u (G z - W) is connected.
Having seen in Chapter I how useful it is to partition a graph into its
components, that is into its maximal connected subgraphs, let us attempt a
similar decomposition using all maximal 2-connected subgraphs. A sub-
graph B of a graph G is a block ofG if either B is a bridge (and its endvertices)
or else it is a maximal 2-connected subgraph of G. The remarks above show
that any two blocks have at most one vertex in common and if x, yare
distinct vertices of a block B then G - E(B) contains no x-y path. Therefore
every vertex belonging to two blocks is a cutvertex of G, and, conversely,
every cut vertex belongs to at least two blocks. Recalling that a cycle is 2-
connected and an edge is a bridge iff no cycle contains it, we find that G
decomposes into its blocks B l , B z , ... , Bp in the following sense:
p

E(G) = UE(B i ) and E(Bi) n E(B) =0 if i =f. j.

1

Suppose now that G is a non-trivial connected graph. Let bc(G) be the

graph whose vertices are the blocks and cutvertices of G and whose edges
join cutvertices to blocks: each cut vertex is joined to the blocks containing
it. Then bc(G), called the block-cut vertex graph of G, is a tree. Each end-
vertex of bc(G) is a block of G, called an endblock of G. If Gis 2-connected
or it is a K Z (an "edge") then it contains only one block, itself; otherwise
there are at least two end blocks, and a block is an endblock iff it contains
exactly one cutvertex (Figure 111.4).
The basic result in the theory of connectivity was proved by Menger in
1927. It is the analogue of the max-flow min-cut theorem for (non-directed)
graphs. Recall that two s-t paths are independent if they have only the vertices
sand t in common.
52 III Flows, Connectivity and Matching

Figure I1I.4. The construction of the block-cut vertex tree bc(G). Bl is an endblock.

Theorem 5.
i. Let sand t be distinct non-adjacent vertices of a graph G. Then the minimal
number of vertices separating s from t is equal to the maximal number of
independent s-t paths.
11. Let sand t be distinct vertices of G. Then the minimal number of edges
separating s from t is equal to the maximal number of edge-disjoint s-t
paths.

PROOF. (i) Replace each edge xy of G by two directed edges, .xy and yx, and
give each vertex other than sand t capacity 1. Then by Theorem 4 the
maximal flow value from s to t is equal to the minimum of the capacity of a
cut separating s from t. By the integrality theorem (Theorem 2) there is a
maximal flow with current 1 or 0 in each edge. Therefore the maximum flow
value from s to t is equal to the maximal number of independent s-t paths.
The minimum of the cut capacity is clearly the minimal number of vertices
separating s from t.
(ii) Proceed as in (i), except instead of restricting the capacity of the
vertices, give each directed edge capacity 1. 0

The two parts of the above theorem are called the vertex form of Menger's
theorem and the edge form of Menger's theorem. One can easily deduce
the edge form from the vertex form (Exercise 14), but the other implication
is not so easily proved. Since, as we have mentioned already, the max-flow
min-cut theorem can also be deduced from Menger's theorem, we shall give
another proof of the vertex form of Menger's theorem from first principles.

SECOND PROOF OF THE VERTEX FORM OF MENGER'S THEOREM. Denote by k

the minimal number of vertices separating sand t. Then clearly there are
at most k independent s-I paths and for k ~ I there are k independent s-I
paths.
Suppose the theorem fails. Take the minimal k 2:: 2 for which there is a
counterexample to the theorem and let G be a counterexample (for the
minimal k) with the minimal number of edges. Then there are at most k - 1
independent s-t paths and no vertex x is joined to both sand t, otherwise
G - x would be a counterexample for k - 1.
3 Matching 53

Let W be a set of k vertices separating s from t. Suppose neither s nor t

is adjacent to every vertex in W. Let Gs be obtained from G by replacing
the component of G - W containing s by a single vertex S' and joining S'
to each vertex in W. In Gs we still need k vertices to separate S' from t and
since the component we collapsed had at least two vertices, Gs has fewer
edges than G. Since G is a counterexample of minimal size, in Gs there are k
independent S' -t paths. The segments of these k paths from t to Ware such
that any two of them have exactly the vertex t in common. In particular, for
every WE W one of these paths is a t-w path. If we carry out the analogous
procedure for t instead of s then we get k paths from s to W. These two sets
of paths can be put together to give k independent s-t paths, contradicting
our assumption. Hence for any set W of k vertices separating s from t either
s or t is adjacent to each vertex of W.
Let SX 1 X2 ... x/t be a shortest s-t path. Then I? 2 and, by the minimality
of G, in the graph G - x 1 X 2 we can find a set Wo of k - 1 vertices separating
s from t. Then both W 1 = {xd u Wo and W 2 = {X2} u Wo are k-sets
separating s from t. Since t is not joined to Xl' S is joined to every vertex in
W 1 Similarly, s is not joined to X2' so t is joined to every vertex in W 2 . This
implies the contradiction that sand t have at least one common neighbour:
every vertex in Wo is a common neighbour of sand t and I Wo I = k - 1 ? 1.
D

Corollary 6. A graph is k-connected (k ? 2) iff it has at least two vertices and

any two vertices can be joined by k independent paths. A graph is k-edge-
connected (k ? 2) iff it has at least two vertices and any two vertices can be
joined by k edge-disjoint paths.

Another characterization of k-connectivity is given in Exercise 12.

Corresponding to the max-flow min-cut theorem for multiple sources
and sinks, one has the following version of Menger's theorem. If Sand T
are arbitrary subsets of vertices of G then the maximal number of vertex-
disjoint (induding endvertices!) S- T paths is min { I WI: We V( G), G - W
has no S- T path}. To see this add two new vertices to G, say sand t, join s
to every vertex in Sand t to every vertex in T, and apply Menger's theorem
to the vertices sand t in the new graph.

3 Matching
Given a finite group G and a subgroup H of index m, can you find m elements
of G, say gl' g2"'" gm, such that {glH, g2H, ... , gmH} is the set of left
co sets of Hand {Hgl> Hg 2 , ... , Hg m} is the set of right cosets? A reformula-
tion of this problem turns out to be a special case of the following problem
arising frequently in diverse branches of mathematics. Given a family
54 III Flows, Connectivity and Matching

d = {A 1> A 2, ... , Am} of subsets of a set X, can we find m distinct elements

of X, one from each Ai? A set {x I, X2' ... ,xm } with these properties (i.e.
Xi E Ai> Xi i= Xj if i i= j) is called a set of distinct representatives of the family
d. The set system d is naturally identifiable with a bipartite graph with
vertex classes VI = d and V2 = X in which Ai E d is joined to every X E X
contained in Ai' A system of distinct representatives is then a set of m in-
dependent edges (thus each vertex in VI is incident with one of these edges).
We also say that there is a complete matching from VI to V2
It is customary to formulate this problem in terms of marriage arrange-
ments. Given m girls and n boys, under what conditions can we marry off
all the girls provided we do not want to carry matchmaking as far as to marry
a girl to a boy she does not even know?
It is clear that both the max-flow min-cut theorem and Menger's theorem
imply a necessary and sufficient condition for the existence of a complete
matching. In fact, because of the special features of a bipartite graph, there
is a particularly simple and pleasant necessary and sufficient condition.
If there are k girls who altogether know at most k - 1 boys then we cannot
find suitable marriages for these girls. Equivalently, if there is a complete
matching from VI to V2 then for every S C VI there are at least I S I vertices
of V2 adjacent to a vertex in S, that is
Ir(S)1 ~ lSI.
The result that this necessary condition is also sufficient is usually called
Hairs theorem. (P. Hall proved it in 1935, and an equivalent form of it was
proved by Konig and Egervary in 1931, but both versions follow immediately
from Menger's theorem, proved in 1927.) We shall give three proofs. The
first is based on Menger's theorem or the max-flow min-cut theorem, the
other two prove the result from first principles.

Theorem 7. A bipartite graph G with vertex sets VI and V2 contains a complete

matching from VI to V2 iff
Ir(S) I ~ lSI for every S C VI'
We have already seen that the condition is necessary so we have to prove only
the sufficiency.
1ST PROOF. Both Menger's theorem (applied to the sets VI and V2 , as at the
end of 2) and the max-flow min-cut theorem (applied to the directed graph
obtained from G by sending each edge from VI to V2 , in which each vertex
has capacity 1) imply the following. If G does not contain a complete match-
ing from VI to V2 then there are TI C VI and T2 C V2 such that I TI I +
I T21 < I VI I and there is no edge from VI - TI to V2 - T2 . Then
r(VI - TI ) C T2 so
Ir(VI - TI)I:s; IT21 < IVII-ITII = IVI - Til.
This shows the sufficiency of the condition. D
3 Matching 55

2ND PROOF. In this proof, due to Halmos and Vaughan, we shall use the
matchmaking terminology. We shall apply induction on m, the number of
girls. For m = 1 the condition is clearly sufficient so we assume that m ~ 2
and the condition is sufficient for smaller values of m.
Suppose first that any k girls (1 ::;;; k < m) know at least k + 1 boys.
Then we arrange one marriage arbitrarily. The remaining sets of girls and
boys still satisfy the condition, so the other m - 1 girls can be married off
by induction.
Suppose now that for some k, there are k girls who altogether know ex-
actly k boys. These girls can clearly be married off by induction. What
about the other girls? We can marry them off (again by induction) if they
also satisfy the condition provided we do not count the boys who are already
married. But the condition is satisfied since if some I girls to be married
know fewer than I remaining boys then these girls together with the first k
girls would know fewer than k + I boys. 0

3RD PROOF. This proof is due to Rado. Let G be a minimal graph satisfying
the condition. It suffices to show that G consists of I VI I independent edges.
If this is not so then G contains two edges of the form al x, a2 x, where
aI' a2 E VI and x E V2 . Since the deletion of either of these edges invalidates
the condition, there are sets AI, A2 C VI such that I r(Ai) I = lAd and ai
is the only vertex of A;(i = 1, 2) adjacent to x. Then

Ir(A I ) nr(A 2)1 = Ir(AI - {al}) nr(A2 - {a2}) I + 1

~ jr(AI nA 2)1 + 1 ~ IAI nA21 + 1,

which implies the following contradiction:

Ir(A I u A 2 )1 = Ir(A I ) u r(A 2) I = jr(A I ) I + jr(A 2) I

- Ir(A I ) nr(A 2)1 ::;;;IAII + IA21-IAI nA21-l.

A regular bipartite graph satisfies the conditions of Hall's theorem, so

it has a complete matching. In turn this implies that we can indeed find group
elements gl' g2, ... , gm as required at the beginning of the section.
Let us reformulate the marriage theorem in terms of sets of distinct
representatives.

Theorem 7'. A family d = {AI' A 2 , ... , Am} of sets has a set of distinct
representatives iff

U Ai I ~
I ,eF IFI for every Fe {l, 2, ... , m}. o
There are two natural extensions of the marriage theorem. Suppose the
marriage condition is not satisfied. How near can we come to marrying off
56 III Flows, Connectivity and Matching

all the girls? When can we marry off all the girls but d of them? Clearly
only if any k of them know at least k - d boys. This obvious necessary con-
dition is again sufficient.

Corollary 8. Suppose the bipartite graph G = Gim, n) with vertex sets Vi' V2
satisfies the following condition:
!r(S) I ;::: lSI - d for every S C Vi'
Then G contains m - d independent edges.
PROOF. Add d vertices to V2 and join them to each vertex in Vi' The new
graph G* satisfies the condition for a complete matching. At least m - d
of the edges in this matching belong to G. D

The next extension concerns matchmaking for boys in a polygynous

country, where the ith boy intends to marry di girls.

Corollary 9. Let G be a bipartite graph with vertex classes Vi = {Xl' ... ,xm}
and V2 = {Yl, ... ,Yn}' Then G contains a subgraph H such that dH(x;) = di
and 0 ~ dH(Yi) ~ 1 iff
Ir(S)I;::: L di for every S C Vi'
XiES

PROOF. Replace each vertex Xi by di vertices joined to each vertex in rex;).

Then there is a subgraph H iff the new graph has a matching from the new
first vertex class to V2 The result follows from Theorem 7. D

Of course, Corollary 9 also has a defect form which the reader is en-
couraged to state and deduce from this.
The reader is probably aware of the fact that these corollaries are still
special cases of the max-flow min-cut theorem. In fact, the bipartite graph
version of the max-flow min-cut theorem is considerably more general
than the corollaries above.

Theorem 10. Let G = Gim, n) be a bipartite graph with vertex classes

Vi = {Xl,., xm} and V2 = {Yl"'" Yn}. For S C Vi and 1 ~j ~ n denote
by Sj the number of edges from Yj to S. Let db"" dm and e l , ... , en be natura'l
numbers and let d ;::: O. Then there exists a subgraph H of G with
n
e(H) ;::: Ld i - d,
1

1~ i ~ m
and
1 ~ j ~ n,
3 Matching 57

ifffor every S C VI we have

n

Ld
X,ES
i ::; L min{Sj' ej} + d.
j=l

PROOF. Turn G into a directed graph G by 'sending each edge from VI to V2

Give each edge capacity 1, a vertex Xi capacity di and a vertex Yj capacity ej.
Then there is a subgraph H with required properties iff in Gthere is a flow
from VI to V2 with value at least L~ d; - d and, by the max-flow min-cut
theorem, this happens iff every cut has capacity at least L~ d; - d. Now
minimal cut-sets are of the form T u U U E(VI - T, V2 - U), where
T C VI and U C V2 Given a set T, the capacity of such a cut will be minimal
if a vertex Yj belongs to U iff its capacity is smaller than the number of edges
from S = VI - T to Yj. With this choice of U the capacity of the cut is
exactly
n

X,E
L d; + Lmin{Sj'
T 1
e).

The condition that this is at least L~ d; - d is clearly the condition in the

theorem. 0

The reader is invited to check that the second proof of Theorem 7 can be
rewritten word for word to give a proof of the exact form of this result (that
is with d = 0) and the defect form (the case d ~ 0) can be deduced from it
as Corollary 8 was deduced from Theorem 7 (Exercise 26).
To conclude this section we prove another extension of the marriage
theorem. This is Dilworth's theorem concerning partially ordered sets. A
partial order < on a set is a transitive and irreflexive relation defined on
some ordered pairs of elements. Thus if X < Y and Y < x then x < Z, but
x < Y and Y < x cannot both hold. A set with a partial order on it is a
partially ordered set. The relation x ::; Y expresses the fact that either x = Y
or else x < y. A subset C of a partially ordered set P is a chain (or tower) if
for x, Y E C either x ::; Y or Y < x. A set A c P is an antichain if x < Y
implies that {x, y} A. See Figure Ill.5 for an example.

Figure III.5. A partially ordered set and a maximal antichain. (An edge indicates
that its upper endvertex is greater than its lower endvertex.)
58 III Flows, Connectivity and Matching

What is the smallest number of chains into which we can decompose a

partially ordered set? Since no two elements of an antichain can belong to
the same chain, we need at least as many chains as the minimal size of an
antichain. Once again, the trivial necessary condition is, in fact, sufficient.

Theorem 11. If every antichain in a (finite) partially ordered set P has at most
m elements then P is the union of m chains.
PROOF. Let us apply induction on 1P I. If P = 0 there is nothing to prove
so we suppose that 1PI> 0 and the theorem holds for sets with fewer ele-
ments.
Let C be a maximal chain in P. (Thus if x C then C u {x} is no longer a
chain.) If no antichain of P - C has m elements then we are home by in-
duction. Therefore we may assume that P - C contains an antichain
A = {aI, a2"" ,am}
Define the lower shadow of A as
s- = {x E P: x ::; a; for some i},
and define the upper shadow S + of A analogously. Then P is the union of the
two shadows since otherwise A could be extended to an antichain with
m + i elements. Furthermore, neither shadow is the whole of P since the
maximal element of C does not belong to S - and the minimal element of C
does not belong to S+. By the induction hypothesis both shadows can be
decomposed into m chains, say
m
and S+ = Uct.
I

Since different a; belong to different chains, we may assume that a; E C;-

and a;ECt.
The proof will be completed if we show that a; is the maximal element of
C;- and the minimal element of ct. For in that case the chains C;- and ct
can be strung together to give a single chain C; and then P = C;.U'r
Suppose then that, say, a; is not the maximal element of C;-: a; < x for
some XEC;-. Since x is in the lower shadow of A, there is an ajEA with
x ::; aj. However, this implies the contradiction a; < aj. 0

4 Tutte's I-Factor Theorem

A factor of a graph is a subgraph whose vertex set is that of the whole graph.
If every vertex of a factor has degree r then we call it an r1actor. How can
we characterize graphs with a i-factor? If G has a i-factor H and we delete
a set S of vertices of G, then in a component C of G - S and even number
of vertices are on edges of H contained in C and the other vertices of Care
4 Tutte's I-Factor Theorem 59

Figure III.6. A graph G with a I-factor: lSI = 4 and G - S has 2 odd components.

on edges of H joining a vertex of C to a vertex of S. In particular, for every

odd component C of G - S (that is a component with an odd number of
vertices) there is an edge of H joining a vertex of C to a vertex of S. Now the
edges of H are independent, so this implies that the graph G - S has at
most IS I odd components, one for each vertex in S (see Figure 111.6).
The necessity of the condition we have just found is rather trivial, but it
is not clear at all that the condition is also sufficient. This surprising and
deep result was first proved by Tutte in 1947. It will be convenient to denote
by q(H) the number of odd components of a graph H, that is the number of
components of odd order.

Theorem 12. A graph G has got a l1actor iff

q(G - S) ~ lSI for every S c V(G). (*)

PROOF. We know that the condition is necessary. We shall prove the suffi-
ciency by induction on the order of G. For IG I = 0 there is nothing to prove.
Now let G be a graph of order at least one satisfying (*) and suppose that
the theorem holds for graphs of smaller order.
Let So c V(G) be a non-empty set for which equality holds in (*). Denote
by C 10 C 2 , .. , Cm, m = I So I :2: 1, the odd components of G - So and let
D 1 , D2 , , Dk be the even components of G - So. If the theorem is true
and G does contain a I-factor F then for each Ci there is at least one edge
of F that joins a vertex of Ci to a vertex in So. Since m = I So I, for each Ci
there is exactly one such edge, say Ci Si' Ci E Ci , Si E So. Each Ci - Ci contains
a I-factor, (a subgraph of F), and each D j contains a I-factor (a subgraph of
F). Finally, the edges S1C1' S2C2,"" SmCm form a complete matching from
So into the set {C 1 , C 2 , . , Cm}.
The proof is based on the fact that one can find an So which has all the
properties described above. How shall we find such a set So? Let So be a
maximal non-empty subset of V(G) for which equality holds in (*). Of
course, a priori it is not even clear that there is such a set So. With S = 0
the condition (*) implies that G has an even order. If S is any vertex of G
then G - {s} has even ordet so it has at least one odd component. Since (*)
60 III Flows, Connectivity and Matching

holds, G - {s} has exactly one odd component. Hence for every S = {s}
we have equality in (*). This establishes the existence of So.
As before, C I , C 2 , , Cm, m = ISol, are the odd components of G - So
and D l , D2 , , Dk are the even components.
(i) Each Dj has a l1actor. Indeed, if S c V(D) then
q(G - So uS) = q(G - So) + q(D j - So) ~ ISo u SI = ISol + lSI,
so
q(Dj - S) ~ lSI.

Hence by the induction hypothesis D j has a I-factor.

(ii) If c E C j then C; - c has a I-factor. Assume that this is false. Then by
the induction hypothesis there is a set S c V(C;) - {c} such that
q(C; - {c} uS) > lSI.
Since
q(C; - {c} uS) + IS u {c}1 == IC;! == 1 (mod 2),
this implies that
q(C; - {c} uS) ~ lSI + 2.
Consequently
ISo u {c} u SI = ISol + 1 + lSI ~ q(G - So u {c} uS)
= q(G - So) - 1 + q(C; - {c} u S) ~ ISol + 1 + lSI,
so in (*) we have equality for the set So u {c} uS as well. This contradicts
the maximality of So.
(iii) G contains m independent edges of the form SiC;, s; E So and C; E C;,
i = 1, 2, ... , m. To show this let us consider the bipartite graph H = G2 (n, m)
with vertex classes {C l , C 2 , , Cm} and V2 = So, in which C; is joined to
a vertex s E So if and only if G contains an edge from s to C;. The assertion
above is true iff H has a I-factor, that is a matching from VI to V2 Fortu-
nately we have got the weapon to check this: Hall's theorem. Given A c VI'
put B = r H(A) c V2 (see Figure III.7). Then (*) implies that
IAI ~ q(G - B) ~ IBI.

Hence the graph H satisfies Hall's condition, so it has I-factor.

We are almost home. To complete the proof we just put together the
information from (i), (ii) and (iii). We start with the m independent edges
SiC;, s; E SO, C; E C;. Adding to this set of edges a I-factor of each C; - c;,
1 ~ i ~ m, and a I-factor of each D j , 1 ~ j ~ k, we arrive at a I-factor
~a 0
It is once again very easy to obtain a defect form of the above result.
Exercises 61

Figure 111.7. The construction of H from G. The set A = {C z , C 3 } determines Be So

by the rule B = ru(A).

Corollary 13. A graph G contains a set of independent edges covering all but
at most d of the vertices iff
q(G - S) ::;; lSI + d for every S c: V(G).
PROOF. Since the number of vertices not covered by a set of independent
edges is congruent to IG I modulo 2, we may assume that
d == IGI (mod 2).
Put H = G + K d , that is let H be obtained from G by adding to it a set W
of d vertices and joining every new vertex to every other vertex, old and new.
Then G contains a set of independent edges covering all but d of the vertices
itT H has got a I-factor. When does (*) hold for H? If W - S' "# 0 then
H - S' is connected so q(H - S') ::;; 1 and then (*) does hold; if W c: S'
then with S = S' - W we have q(H - S') = q(G - {S'\W}) = q(G - S),
so (*) is equivalent to
q( G - S) ::;; IS'I = ISI + d. o
EXERCISES

1.- Suppose F is a set of edges after whose deletion there is no flow from s to t with
positive value. Prove that F contains a cut separating s from t.
2.- By summing an appropriate set of equations show that the capacity of a cut is at
least as large as the maximum of the flow value.
3.- Let G = (V, E) be a directed graph and let c be an extended-real-valued capacity
function on E. (Thus c(x, y) is a non-negative real or + 00.) Let sand t be two
vertices. Prove that either there is a flow from s to t with infinite value or else there
is a flow with maximal finite value.
4. By successively reducing the number of circular flows in G, prove that there is a
maximal flow without circular flows in which no current enters the source and
no current leaves the sink.
5. Use the method of Exercise 4 to show that if the capacity function is integral then
there is a maximal flow that is also integral.
62 III Flows, Connectivity and Matching

6. Formulate and prove the max-flow min-cut theorem for multiple sources and
sinks with bounds on the capacity of the edges and vertices.
7. (Circulation theorem). A circulation in a directed graph G is a flow without a
source and a sink. Given a lower capacity I(x, y) and an upper capacity c(x, y)
for each edge xy with 0 S; I(x, y) s; c(x, y) we call a circulation 9 feasible if
I(x, y) s; g(x, y) s; c(x, y)
for every edge xy. Prove that there exists a feasible circulation iff
I(S, S) s; c(S, S) for every S c V.
[Note that the necessity of the condition is trivial since in a feasible circulation
the function I forces at least I(S, S) current from S to S and the function c allows
at most c(S, S) current into S from S. To prove the sufficiency, adjoin a sink sand
a source t to G, send an edge from s to each vertex of G and send an edge from
each vertex of Gto i. Define a capacity function c* on the edges of the new graph
G* by putting c*(x, y) = c(x, y) - I(x, y), c*(s, x) = I(V, x) and c*(x, t) = I(x, V) .
. Then the relation
f(x, y) = g(x, y) - I(x, y)
sets up a 1-1 correspondence between the feasible circulations 9 in G and flows
fin G* from s to t with value I( V, V). Rewrite the condition given in the max-flow
min-cut theorem in the form required in this result.]
8.+ Let H be a bipartite multigraph without loops, with vertex classes VI and V2 (Thus
H may contain mUltiple edges, that is two vertices belonging to different classes
may be joined by several edges, which are said to be parallel.) As usual, given a
vertex x we denote by rex) the set of edges incident with x and by d(x) = Irex) I
the degree of x. Prove that, given any natural number k, the set E of edges can be
partitioned into sets E I , E 2 , , Ek such that for every vertex x and every set E j
we have

lk
d(x)J
s; lr(x)nEjl S;
rd(X)l
k '
r
where, as in the rest of the book, z1 is the least integer not less than z and LzJ =
-r -zl
Thus if we think ofthe partition U~ E j as a colouring of the edges with k colours,
then the colouring is equitable in the sense that in each vertex the distribution of
colours is as equal as possible. [Hint. Construct a directed graph if = (VI U V2 , E)
from H by sending an edge from x to y iff x E VI' Y E V2 and H contains at least one

Figure III.8. The graphs H, fj and G.

Exercises 63

xyedge. Let G be obtained from Ii by adding a vertex u and all the edges yu, ux,
for x E VI and y E V2 , as shown in Figure III.8. Define an appropriate upper and
lower capacity for each edge of Gand prove that there is a feasible integral circula-
tion. Use this circulation to define one of the colour classes.]
9. (Exercise 8+ continued) Show that we may require that, in addition to the
property above, the colour classes are as equal as possible, say lEI I ::;; IE21
::;; ... ::;; IEk I ::;; lEI I + 1, and in each set of parallel edges the distribution of
colours is as equal as possible.
10. Let d l ::;; d2 ::;; ::;; dn be the degree sequence of a graph G. Suppose that

dj c. j + k - 1 for j = 1,2, .. '.' n - 1 - dn-k+ I'

11. Prove that G is k-connected.
Let k and I be integers 1 ::;; k ::;; I. Construct graphs G I , G2 and G3 such that
(i) K(G I ) = k and A(G I ) = I,
(ii) K(G 2 ) = k and K(G 2 - x) = I for some vertex x,
(iii) K(G 3 - x) = k and A(G 3 - xy) = I for some edge xy.
12. Given V c V(G) and a vertex x E V(G) - V, an x-V fan is a set of IV I paths from
x to V, any two of which have exactly the vertex x in common. Prove that a graph
G is k-connected iff IG I c. k + 1 and for any k-set V c V(G) and vertex x not
in V, there is an x-V fan in G. [Hint. Given a pair (x, V), add a vertex u to G
and join it to each vertex in V. Check that the new graph is k-connected if G
is. Apply Menger for x and u.]

13. Prove that if G is k-connected (k c. 2) and {x I, x 2 , .. , Xk} c V( G) then there

is a cycle in G that contains all Xi' 1 ::;; i ::;; k.
14. The line graph L(G) of a graph G = (V, E) has vertex set E and two vertices
f3 E E are adjacent iff they have exactly one vertex of G in common. By applying
0:,
the vertex form of Menger's theorem to the line graph L(G), prove that the vertex
form of Menger's theorem implies the edge form.
15. Show that if A(G) = k c. 2 then the deletion of k edges from G results in at most
2 components. Is there a similar result for vertex-connectivity?
16. Let G be a connected graph with minimum degree c5(G) = k c. 1. Prove that G
contains a path XIX2 ... X k such that G - {XI' x 2 , . , x k } is also connected.
[Hint. Let XIX2'" Xl be a longest path. Note that I c. k + 1. Suppose that
G - {XI> X2, ... , Xk} is disconnected and let YoYI ... Ym be a longest path in a
component C not containing Xk+ IXk+2 ... Xl' Then dc(Yo) ::;; m but Yo cannot be
joined to k - m of the vertices X I> .. , Xk']
17. Let G = G 2 (m, n) be a bipartite graph with vertex classes VI and V2 containing
a matching from VI to V2
(i) Prove that there is a vertex x E VI such that for every edge xy there is a
matching from VI to V2 that contains xy.
(ii) Deduce that if d(x) = d for every x E VI then G contains at least d! matchings
if d ::;; m and at least d(d - 1) (d - m + 1) matchings if d > m.
18. Let A = (aij)~ be an n x 11 stochastic matrix, that is let aij c. 0 and I?= I aij =
Ii= I aij = 1 for all i, j. Show that A is in the convex hull of n x n permutation
64 III Flows, Connectivity and Matching

matrices, i.e. there are Ai ~ 0, Ii Ai = 1, and permutation matrices P" P2' ... , Pm
such that A = Ii AiPi' [Let aij = raijl, A* = (aij)~, and let G = G2 (n, n) be
the bipartite graph naturally associated with A*. Show that G has a complete
matching and deduce that there are a permutation matrix P and a real A,O < il ~ 1,
such that A - ilP = B = (bij)'j is such that bij ~ 0, Ii=,
bij = D=,
bij = 1 - il
for alli,j, and B has at least one more 0 entry than A.]

19. An r x s Latin rectangle based on 1, 2, ... , n is an r x s matrix A = (aij) such

that each entry is one of the integers 1,2, ... , n and each integer occurs in each
row and column at most once. Prove that every r x n Latin rectangle A can be
extended to an n x n Latin square. [Hint. Assume that r < n and extend A to an
(r + 1) x n Latin rectangle. Let Aj be the set of possible values of ar + '.j, that
is let Aj = {k: 1 ::; k ::; n, k =1= aij}' Check that {A j : 1 ::; j ::; n} has a set of
distinct representatives.]
Prove that there are at least n! (n - I)! ... (n - r + 1)! distinct r + n Latin
rectangles based on 1,2, ... , n. [Hint. Apply Exercise I7(ii).]

20. . Let A be an r x s Latin rectangle and denote by A(i) the number of times the
symbol i occurs in A. Show that A can be extended to an n x n Latin square
iff A(i) ~ r + s - n for every i = 1,2, ... , n.

21. Prove the following form of the Schroder-Bernstein theorem. Let G be a bi-
partite graph with vertex classes X and Y having arbitrary cardinalities. Let
A c X and BeY. Suppose there are complete matchings from A into Y and
from B into X. Prove that G contains a set of independent edges covering all
the vertices of A v B. [Hint. Consider the components of the union of the
matchings.]
22. Deduce from Exercise 20 that every bipartite graph contains a set of independent
edges such that each vertex of maximum degree (that is degree ~(G is incident
with one of the edges. Deduce that a non-empty regular bipartite graph has a
i-factor.
23. We say that G is an (r, r - k)-regular graph if r - k ::; beG) ::; ~(G) ::; r. Prove
that for 1 ::; k ::; s ::; revery (r, r - k)-regular graph contains an (s, s - k)-regular
factor. [Hint. Assume s = r - 1. Take a minimal (r, r - k)-regular factor. Note
that in this factor no two vertices of degree r are adjacent. Remove a set of in-
dependent edges covering the vertices of degree r.]

24;- Recall Tychonov's theorem from general topology: the product of an arbitrary
family of compact spaces is compact. Deduce from this the following extension
of Hall's theorem.
Let G be an infinite bipartite graph with vertex classes X and Y, such that each
vertex in X is incident with finitely many edges. Then there is a complete matching
from X into Y iff 1rcA) 1 ~ 1A 1 for every finite subset A of X.
Show that the finiteness condition cannot be omitted.
25. Prove that a 2-edge-connected cubic graph has a I-factor. [This result is called
Petersen's theorem. In order to prove it check that the condition of Tutte's
theorem is satisfied. If 0 =1= S c V(G) and C is an odd component of G - S
then there are at lea~t two S - C edges, since G is 2-edge-connected. Furthermore,
since G is cubic, there are at least three S - C edges. Deduce that q( G - S) ::; 1S I.]
Exercises 65

26. Imitate the second proof of Theorem 7 to give a direct proof of the case d =0
of Theorem 10 and then deduce from it the general case d :;:: O.
27. Let G be a graph of order n with at most r :;:: 2 independent vertices. Prove that
if G is any orientation of G which does not contain a directed cycle (acyclic
orientation) then G contains a directed path of length at least rn/r1 - 1. [Hint.
Apply Dilworth's theorem, Theorem I1.J
28. Deduce from Exercise 27 the following result. Given a set of rk + 1 distinct
natural numbers, either there exists a set of r + 1 numbers, none of which divides
any of the other r numbers, or else there exists a sequence ao < al < ... < ak
such that if 0 ::;; i < j ::;; k then ai divides aj.
29. Describe all maximal graphs of order n = 21 which do not contain a I-factor.
[Hint. Read it out of Tutte's theorem (Theorem 12).J
30. (cf. Exercise 29) Describe all maximal graphs of order n which contain at most k
independent edges. [Hint. Apply Corollary 13.J
31. Make use of Exercise 30 and the convexity of the binomial coefficient CD, x :;:: 2,
to prove that if n :;:: k + 1 then the maximal size of a graph of order n with at
most k independent edges is

Show also that the extremal graphs (that is the graphs for which equality holds)
are one or both of the graphs K2k+ I U En - 2k - 1 and Kk + E"-k. (See Figure 111.9.)

K3 + En

0
K7 U E2

Figure 111.9. For k = 3, n = 9 there are two extremal graphs: K7 u E2 and

K3 + E6.

32. Call a sequence d l , d 2 , . , dn of integers graphic if there is a graph G with vertex

set V(G) = {XI' X 2 , ... , xn} such that d(x) = di' 1 ::;; i ::;; n. (The graph G is said to
realize (d)1.) Show that d I :;:: d 2 :;:: :;:: dn is graphic iff so is the sequence

33. Use the algorithm given in the previous exercise to decide which of the follow-
ing sequences are graphic: 5,4,3,2,2,2; 5,4,4,2,2,1; 4,4,3,3,2,2,2 and
5, 5, 5, 4, 2, 1, 1, 1. Draw tbe graphs realizing the appropriate sequences con-
structed by the algorithm.
66 III Flows, Connectivity and Matching

Notes
The basic book on flows is L. R. Ford, Jnr. and D. R. Fulkerson, Flows in
Networks, Princeton University Press, Princeton, 1962. It not only contains
all the results mentioned in the chapter concerning flows and circulations,
but also a number of applications to standard optimization problems.
The fundamental theorems of Menger, Hall and Tutte are in K. Menger,
Zur allgemeinen Kurventheorie, Fund. Math. 10 (1927) 96-115, P. Hall, On
representatives of subsets, J. London Math. Soc. 10 (1935) 26-30), and W. T.
Tutte, A factorization of linear graphs, J. London Math. Soc. 22 (1947)
107-111. The proof of Tutte's theorem we gave is due to T. Gallai, Neuer
Beweis eines Tutte'schen Satzes, Magyar Tud. Akad. Kozl. 8 (1963) 135-139,
and was rediscovered independently by I. Anderson, Perfect matchings in a
graph, J. Combinatorial Theory Ser. B 10 (1971) 183-186 and W. Mader,
Grad und lokaler Zusammenhang in endlichen Graphen, Math. Ann. 205
(1973) 9-11. The result in Exen::ises 8 and 9 is due to D. de Werra, Multi-
graphs with quasiweak odd cycles, J. Combinatorial Theory Ser. B 23 (1977)
75-82.
A slightly simpler form of the result in Exercise 23 is due to W. T. Tutte;
the proof indicated in the hint was found by C. Thomassen. An extensive
survey of results concerning connectivity and matching can be found in
Chapters I and II of B. BolloMs, Extremal Graph Theory, Academic Press,
London and New York, 1978.
CHAPTER IV
Extremal Problems

At least how many edges are in a graph of order n if it is forced to contain a

path of length I? A cycle of length at least I? A cycle of length at most I?
A complete graph K r ? These questions are special cases of the so called
forbidden subgraph problem: given a graph F, determine ex(n; F), the maximal
number of edges in a graph of order n not containing F. The forbidden sub-
graph problem is a prime example of the rather large family of extremal
problems in graph theory. In this chapter we present the fundamental result
concerning the forbidden subgraph problem and discuss some closely re-
lated extremal questions. We have to emphasize that we do not intend to
provide an overall view of extremal graph theory.
Before going into the details, it is appropriate to say a few words about
the terminology. If for a given class of graphs a certain graph parameter,
say the number of edges or the minimum degree, is at most some number f,
then the graphs for which equality holds are the extremal graphs of the
inequality. As a trivial example, note that an acyclic graph of order n has at
most n - 1 edges, and the extremal graphs are the trees of order n. In the
forbidden subgraph problem a graph is extremal if it does not contain F and
has ex(n; F) edges. In the previous chapter Tutte's factor theorem enabled
us to solve a beautiful extremal problem: how many edges guarantee k + 1
independent edges? In this case F consists of k + 1 independent edges, that
is F = (k + 1)K2. Exercise 31 of Chapter III claims that for n ~ 2k + 1
the extremal graphs of ex(n; F) are K2k+ 1 U En - 2k-l and/or Kk + E n - k.
The material in this chapter falls conveniently into two parts: the odd
sections concern paths and cycles while the even ones are about complete
subgraphs. We have chosen to alternate the topics in order to have the
simpler results first, as in most other chapters.
The first section is about paths and cycles (short and long) in graphs of

67
68 IV Extremal Problems

large size. Among others we shall give a good bound on ex(n; pi), the maximal
number of edges in a graph of order n without a path of length l. We shall
also present some fundamental results about Hamilton cycles.
One of the best known theorems in graph theory, Tunin's theorem,
determines the function ex(n; Kr). The second section is devoted to this
theorem together with some related results.
When discussing ex(n; pi) and ex(n; Kr), we most1y care about the case
when n is large compared to I and r. We get rather different problems if F
and G have the same order. A prime example of these problems will be
discussed in the third section, the problem of Hamilton cycles. Over the
years considerable effort has gone into the solution of this problem and in a
certain rather narrow sense the present answers are satisfactory.
The fourth section is devoted to a deep and surprising theorem of Erdos
and Stone. The theorem concerns ex(n; F), where F is a complete r-partite
graph with t vertices in each class, but as an immediate corollary of this
result one can determine lim n _ oo ex(n; F)/n 2 for every graph F.

1 Paths and Cycles

In its natural formulation our first theorem gives a lower bound on the order
of a graph in terms of the minimum degree and the girth, the length of a
shortest cycle. Equivalently, the result gives an upper bound on the girth
in terms of the order and minimum degree.

Theorem 1. For g ;;:: 3 and b ;;:: 3 put

I + _b_ {(b _ 1)<g-1)/2 - I} if g is odd,

{ b-2
no(g, b) =
_2_ {(b _ 1)g/2 - I} if g is even.
b-2
Then a graph G with minimum degree b and girth g has at least no(g, b) vertices.

PROOF. Suppose first that g is odd, say g = 2d + 1, d ;;:: 1. Pick a vertex x.

There is no vertex z for which G contains two distinct z-x paths of length at
most d, since otherwise G has a cycle of length at most 2d. Consequently
there are at least b vertices at distance 1 from x, at least b(b - 1) vertices at
distance 2, and so on, at least b(b - 1)d-1 vertices at distance d from x
(Figure IY.1). Thus

n ;;:: 1 + b + b(b - 1) + ... + b(b - l)d - \

as claimed.
l Paths and Cycles 69

*Figure IV. I. The cases i5 = 9 = 5 and i5 = 4, 9 = 6.

Suppose now that g is even, say g = 2d. Pick two adjacent vertices, say
x and y. Then there are 2(0 - 1) vertices at distance 1 from {x, y}, 2(0 - 1)2
vertices at distance 2, and so on, 2(0 - l)d-1 vertices at distance d - 1 from
{x, y}, implying the required inequality. D

Let Go be an extremal graph of Theorem 3, that is a graph with parameters

oand g, for which equality holds. The proof above implies that Go is regular
of degree 0; if g = 2d + 1 then Go has diameter d and if g = 2d then every
vertex is within distance d - 1 of each pair of adjacent vertices. It is easily
seen that no(g, 0) is also the maximal number for which there is a graph with
maximum degree 0 having the latter property (Exercise 1). We call Go a
Moore graph of degree 0 and girth g, or, if g = 2d + 1, a Moore graph of
degree 0 and diameter d. In Chapter VIII we shall use algebraic methods to
investigate Moore graphs.
Let us see now what we can say about long cycles and paths in a graph.
If a graph of order n is Hamiltonian then its circumference, that is the length
of a longest cycle, is n, while the length of a longest path is n - 1. However,
every non-Hamiltonian connected graph contains at least as long paths as
the circumference of the graph. Indeed, if C = x I X2 .. Xl is a longest cycle
and I < n then there is a vertex y not on C adjacent to a vertex of C, say x I,
and then YXIX2'" Xl is a path of length I.

Theorem 2. Let G be a connected graph of order n ;;::: 3 such that for any two
non-adjacent vertices x and y we have

d(x) + dey) ;;::: k.

If k = n then G is Hamiltonian and if k < n then G contains a path of length

k. and a cycle of length at least (k + 2)/2.

PROOF. Assume that G is not Hamiltonian and let P = XIX2 ... Xl be a

longest path in G. The maximality of P implies that the neighbours of XI
and Xl are vertices of P. As G does ~ot contain a cycle oflength I, XI is not
adjacent to Xl' Furthermore the path P cannot contain vertices Xi and Xi+ I
such that XI is adjacent to Xi+ I and Xl is adjacent to Xi since otherwise
X I X 2 ... XiXnXn-l ... X i + 1 is a cycle of length I (Figure IV.2). Consequently
the sets
70 IV Extremal Problems

2----.~
---------
X"'-;----~-X
.. . i~-~
X~
-..--X
..... ~
XI- 1 Xl

----
Figure IV.2. The construction of a cycle of length I.

are disjoint subsets of {X2, X3' ... , Xl} and so

k s d(xI) + d(xl) s 1- 1 s n - 1.
The first two assertions of the theorem follow from this inequality. If k = n,
this is impossible, so G is Hamiltonian. If k < n then P has length I - 1 ~ k.
Finally, the assertion about cycles is even simpler. Assume that d(xI) ~
d(xl) so d(xI) ~ rk/21, where rzl denotes the least integer not less than z.
Put t = max{i: XIXi E E(G)}. Then t ~ d(xI) + 1 ~ rk/21 + 1 and G con-
tains the cycle XIX2'" XI of length t. D

In 3 we shall make use of the proof of this theorem to obtain detailed

information about graphs without long cycles and paths. For the moment
we confine ourselves to noting two consequences of Theorem 2.

Theorem 3. Let G be a graph of order n without a path of length k( ~ 1). Then

k - 1
e(G) s -2-n.

A graph is an extremal graph (that is equality holds for it) iff all its components
are complete graphs of order k.

PROOF. We fix k and apply induction on n. The assertion is clearly true if

n s k. Assume now that n > k and the assertion holds for smaller values
ofn.
If G is disconnected, the induction hypothesis implies the result. Now if
G is connected, it contains no Kk and, by Theorem 1, it has a vertex X of
degree at most (k - 1)/2. Since G - X is not an extremal graph
k-l k-l k-l
e(G) s d(x) + e(G - x) < -2- + -2-(n - 1) = -2-n. D

Theorem 4. Let G be a graph of order n in which every cycle has length at most
k (k ~ 2). Then
k
e(G) s l(n - 1).

A graph is an extremal graph iff it is connected and all its blocks are complete
graphs of order k. D
2 Complete Subgraphs 71

The proof of this result is somewhat more involved than that of Theorem
3. Since a convenient way of presenting it uses 'simple transforms' to be
introduced in 3, the proof is left as an exercise (Exercise 25) with a detailed
hint.

2 Complete SUbgraphs
What is ex(n; K r ), the maximal number of edges in a graph of order n not
containing a K r , a complete graph of order r? If G is (r - I)-partite then it
does not contain a Kr since every vertex class of G contains at most one
vertex of a complete subgraph. Thus ex(n; Kr) is at least as large as the
maximal size of an (r - 1)-partite graph of order n. In fact, there is a unique
(r - I)-partite graph of order n that has maximal size. This graph is the
complete (r - I)-partite graph 1',.-l(n) that has L(n + k - I)/(r - I)J
vertices in the kth class (Figure IV.3). In other words, 1',.-1 (n) is the complete
(r - I)-partite graph of order n whose classes are as equal as possible: say
there are nk vertices in the kth class and n1 ::;; n2 ::;; ... ::;; nr - 1 ::;; n1 + 1.
Indeed, let G be an (r - 1)-partite graph of order n and maximal size.

Figure IV.3. The Turan graph T3(7).

Clearly G is a complete (r - I)-partite graph. Suppose the classes are not

as equal as possible, say there are m1 vertices in the first class and m2 ;:;::
m1 + 2 in the second. This is impossible since by transferring one vertex
from the second class to the first we would increase the number of edges by
(m1 + I)(m2 - 1) - m1m2 = m2 - m1 - 1 ;:;:: 1.
The number of edges in 1',.-l(n) is usually denoted by t r - 1(n); thus tin) =
Ln 2 /4J. A fundamental theorem of Tunin states that the trivial inequality
ex(n; Kr);:;:: t r - 1(n) is, in fact, an equality for every nand r. We shall show
first that the degree sequence of a graph without a K r is dominated by the
degree sequence of an (r - I)-partite graph. In view of the remarks above
this will imply Tunin's theorem.

Theorem 5. Let G be a graph with vertex set V that does not contain K r , a
complete graph of order r. Then there is an (r - I)-partite graph H with
vertex set V such that for every vertex Z E V we have
72 IV Extremal Problems

If G is not a complete (r - I)-partite graph then there is at least one vertex Z

for which the inequality above is strict.
PROOF. We shall apply induction on r. For r = 2 there is nothing to prove
since G is the empty graph En, which is I-partite. Assume now that r ~ 3
and the assertion holds for smaller values of r.
Pick a vertex x E V for which dG(x) is maximal and denote by W the set
of vertices of G that are joined to x. Then Go = G[W] does not contain a
K,-l otherwise with x it would form a K'. By the induction hypothesis we
can replace Go by an (r - 2)-partite graph Ho with vertex set W in such a
way that dGo(Y) ~ dHo(y) for every yEW. Add to Ho the vertices in V - W
and join each vertex in V - W to each vertex in W. To complete the proof
let us check that the graph H obtained in this way has the required properties.
If Z E V - W then dH(z) = dH(x) = dG(x) ~ dG(z) and if z E W then
dH(z) = dHo(z) + n - I WI ~ dGo(z) + n - I WI ~ dG(z). Thus dG(z) ~ dH(z)
holds for every z E V.
Can H contain a complete subgraph of order r? Clearly not, since a
complete subgraph of H has at most one vertex in V - Wand most r - 2
vertices in W.
What can we say about G is e(G) = e(H)? Then e(G o) = eCHo), so by
the induction hypothesis Go is a complete (r - 2)-partite graph and
eG(W, W) = eH(W, W) = IWIIWI.
Hence
e(G[W]) = e(H[W]) = 0,

implying that G is a complete (r - I)-partite graph. o

We now give Turon's theorem in its original form.

Theorem 6. ex(n; K') = t,-l(n) and T,.-l(n) is the unique graph of order nand
size t,_1 (n) that does not contain a complete graph of order r.

1sT PROOF. Since T,.-I(n) is the unique (r - I)-partite graph of order nand
maximum size, both assertions follow from Theorem 5. D
In order to emphasize the significance of Theorem 6, we prove it from
first principles also. This proof is based on the fact that T,.- 1(n) is almost
regular: .5(T,.- 1 (n = n - rnj(r - 1)1, ,1.(T,.- 1(n = n - Lnj(r - l)J, and if
x is a vertex of minimum degree in T,.-I(n) then T,.-I(n) - x is exactly
T,.-l(n - 1).

2ND PROOF. Fix r ~ 3 and apply induction on n. For n ~ r the assertion is

trivial so suppose that n > r and the theorem holds for smaller values of n.
Suppose G has n vertices, t,_l(n) edges and it contains no K'. As T,.-I(n)
is a maximal graph without a K' (that is no edge can be added to it without
2 Complete Subgraphs 73

creating a K') the induction step will follow if we show that G is exactly a
1',. - 1 (n). Since the degrees in 1',. _ 1 (n) differ by at most 1, we have

J(G) ~ J(1',.-I(n ~ ~(1',.-I(n ~ ~(G).

Let x be a vertex of G with degree d(x) = J(G) ~ J(1',.-I(n. Then

e(G - x) = e(G) - d(x) ~ e(1',.-I(n - 1,
so by the induction hypothesis Gx = G - x is exactly a 1',.-I(n - 1).
A smallest vertex class of Gx contains L(n - 1)/(r - l)J vertices and the
vertex x is joined to all but

n- 1- (n - rr ~ 11) = l: =~J
vertices of G x . Since x cannot be joined to a vertex in each class of Gx , it
has to be joined to all vertices of G x save the vertices in a smallest vertex
class. This shows that G is exactly a 1',.-1 (n) graph. 0

The proof above can easily be adapted to give a number of related results
(cf. Exercises 7, 8).
Now let us turn to the probLem of Zarankiewicz, which is the analog of
Tunin's problem in bipartite graphs. Write G 2 (m, n) for a bipartite graph
with m vertices in the first class and n in the second. What is the maximum
size of a G2 (m, n) if it does not contain a complete bipartite graph with s
vertices in the first class and t in the second? This maximum is usually
denoted by z(m, n; s, t). The following simple lemma seems to imply a very
good upper bound for the function z(m, n; s, t).

Lemma 7. Let m, n, s, t, r, k be integers, 2 ~ s ~ m, 2 ~ t ~ n, 0 ~ k, 0 ~

r < m, and Let G = G 2 (m, n) be a graph of size z = my = km + r without a
K(s, t) subgraph having s vertices in the first class and t in the second. Then

(1)

PROOF. Denote by VI and V2 the vertex classes of G. We shall say that at-set
(i.e., a set with t elements) T of V2 beLongs to a vertex x E VI if x is joined to
every vertex in T. The number of t-sets belonging to a vertex x E VI is (d}X).
Since the assumption on G i-s exactly that each t-set in V2 belongs to at most
s - 1 vertices of VI, we find that

L (d(X) ~ (s _ 1) (n). (2)

XEV, t t
As LXEV, d(x) = z = my =km + r, 0 ~; < m andf(u) = (~) is a convex
function of u for u ~ t, inequality (2) implies (1). 0
74 IV Extremal Problems

Theorem 8. z(m, n; s, t) :5: (s - l)l/t(n - t + 1)m 1 -(I/t) + (t - l)m.

PROOF. Let G = G2 (m, n) be an extremal graph for the function z(m, n; s, t),
that is let G be a bipartite graph of size z(m, n; s, t) = my without a K(s, t)
subgraph. As y :5: n, inequality (1) implies
(y - (t - l)Y :5: (s - l)(n - (t - 1)Ym- 1 . o
As a consequence of Theorem 8 we see that if t :c: 2 and c > (t - l)l/t
are fixed then for every sufficiently large n we have
z(n, n; t, t) < cn 2 -(1/t). (3)
The method of Lemma 7 gives also an upper bound for ex(n; Kz(t)), the
maximum number of edges in a graph of order n without a complete t by t
bipartite subgraph.

Theorem 9. ex(n; Kit)) :5: t(t - l)l/t(n - t + 1)n 1 -(I/t) + t{t - l)n
< t{t - l)l/t n Z-(I/t) + ttn.

PROOF. Let G be an extremal graph. As in Lemma 7, let us say that at-set

of the vertices belongs to a vertex x if x is joined to every vertex of the t-set.
Since G does not contain a Kit), every t-set belongs to at most t - 1 vertices.
Therefore if G has degree sequence (d;)~ then 2::1 d i = 2 ex(n; K 2 (t)) and

implying the result. o

It is very likely that inequality (3) gives the correct order of magnitude of
z(n, n; t, t) for every t :c: 2 as n ~ 00 but this has been proved only for t = 2
and 3. In fact, it is rather hard to find non-trivial lower bounds for
z(m, n; s, t). In Chapter VII we shall use the probabilistic method to obtain
a lower bound. To conclude the section we prove an elegant result for t = 2
that indicates the connection between the problem of Zarankiewicz and
designs, in particular, projective spaces.

Theorem 10. z(n, n; 2, 2) :5: tn{l + (4n - 3)I/Z} and equality holds for in-
finitely many values ofn.

PROOF. Note that with z = tn{l + (4n - 3)1/2} we have

(z - n)z = n 2 (n - 1).
Suppose that there is a bipartite graph G = Gz(n, n) of size greater than z
that does not contain a quadrilateral (that is a K(2, 2)). Denote by
d 1, d z , ... , dn the degrees of the vertices of G in the first vertex class, say VI'
3 Hamilton Paths and Cycles 75

Then L~ d; = e = e(G) > z and, as in the proof of Lemma 7, we have

(n)2 ~ t: (d.)2' = 2:1 t: dt - 2:1 t: d

n n n
i

~ 2 _ .: z(z - n) _ (n)
~ 2n e 2> 2n - 2'
This shows that the inequality does hold.
In fact, the proof so far tells us a considerable amount about the graphs
G for which equality can be attained. We must have d i = d 2 = ... = dn = d
and any two vertices in the second vertex class V2 of G have exactly one
common neighbour. By symmetry each vertex of V2 has degree d and any
two vertices in Vi have exactly one common neighbour.
Call the vertices in Vi points and the sets r(x), x E V2, lines. By the remarks
above there are n points and n lines, each point is on d lines and each line
containsd points, there is exactly one line through any two points and any
two lines meet in exactly one point. Thus we have arrived at the projective
plane of order d - 1. Since the steps are easy to trace back, we see that equa-
lity holds for every n for which there is a projective plane with n points. In
particular, equality holds for every n = q2 + q + 1 where q is a prime power.
In conclusion, let us see the actual construction of G for the above values
of n. Let q be a prime power and let PG(2, q) be the projective plane over
the field of order q. Let Vi be the set of points and V2 the set of lines. Then
I Vi I = I V2 1 = q2 +q+1= n.
Let G be the bipartite graph Gin, n) with vertex classes Vi and V2 in which
PI (P E VI' IE V2 ) is an edge iff the point P is on the line l. (See Figure IV.4.)
Then G has n(q + 1) = !n{l + (4n - 3)i/2} edges and it does not contain a
quadrilateral. D

n -> n + 0, 1,3

Figure IVA. PG(2, 2) and the corresponding bipartite graph.

3 Hamilton Paths and Cycles

A class of graphs is said to be monotone if whenever a graph L belongs to the
class and M is obtained from L by adding to it an edge (but no vertex) then
M also belongs to the class. Most theorems in graph theory can be expressed
by saying that a monotone class vIt is contained in a monotone class r!I.
76 IV Extremal Problems

Of course, these classes are usually described in terms of graph invariants

or subgraphs contained by them. For example, the simplest case of TUflin's
theorem, discussed in the previous section, states that the class uIt =
{G(n, m): m > n2 j4} is contained in [1 = {G: G contains a K3}. It is worth
noting that a class [1 of graphs is said to be a property of graphs if L E [1 and
L ~ M imply M E [1.
How should we go about deciding whether or not uIt is contained in &?
In some cases there is a simple and beautiful way of tackling this problem.
Suppose we have a class!Y of triples (G, x, y), where G is a graph and x and
yare non-adjacent vertices of G, such that if (G, x, y) E!Y and G E uIt then
G belongs to [1 iff G+ = G + xy does. This holds, for example, if [1 is the
property of containing a K r and !Y = {(G, x, y): /r(x) n r(y)/ < r - 2}.
In this case G can be replaced by G+. If G+ also contains two non-adjacent
vertices, say u and v, such that (G + , u, v) E !Y then we can repeat the opera-
tion: we can replace G+ by G+ + = G+ + uv. Continuing in this way we
arrive at a graph G* :::> G which belongs to [1 iff G does, and which is a
closure of G with respect to !Y, that is it has the additional property that for
no vertices a, bE G* does (G*, a, b) E!Y hold. Thus it is sufficient to decide
about these graphs G* E uIt whether or not they belong to [1.
Of course the method above is feasible only if (i) the class !Y is simple
enough, (ii) it is easy to show that G belongs to [1 iff G+ does and (iii) if we
start with a graph G E uIt then a graph G* E uIt is easily shown to belong to
[1. In this section we give two examples that satisfy all these requirements:
we shall give sufficient conditions for a graph to contain a Hamilton cycle
or a Hamilton path. Because of the special features of these examples it will
be convenient to use slightly different notation and terminology.
Let nand k be natural numbers and let [1 be a class of graphs of order n.
We say that [1 is k-stable ifwhenever G is an arbitrary graph of order n, x
and yare non-adjacent vertices of G and d(x) + d(y) ~ k then G has property
[1 iff G + = G + xy has it also. It is easily seen that for every graph G of
order n there is unique minimal graph G* = Ck(G) containing G such that
dG*(x) + dG*(y) s k - 1 for xy E(G*).
In the notation of the previous paragraph, we shall take
!Y = {(G, x, y): /G / = n, xy E(G), d(x) + dey) ~ k},
which is certainly simple enough, so (i) will be satisfied. It is also encouraging
that G* = Ck(G) is unique. Almost by definition we have the following
principle of stability: if [1 is a k-stable property of graphs of order n then G
has property [1 ijfCk(G) has it also. We call C k( G) the k-closure of G.
Requirement (ii) is also satisfied, since the gist of the proof of Theorem 2
is that the property of containing a Hamilton cycle is n-stable and the
property of containing a Hamilton path is (n - I)-stable. Indeed if d(x) +
dey) ~ n - 1 whenever x and yare non-adjacent distinct vertices, then the
graph is connected so the proof of Theorem 2 can be applied. (In fact, this
3 Hamilton Paths and Cycles 77

is exactly what inspired the notion of a k-closure.) By the stability principle

we obtain the following reformulation of Theorem 2 in the case k = n or
n-l.

Lemma 11. G is Hamiltonian iff Cn(G) is and G has a Hamilton cycle iff so
does CiG). 0

Depending on the amount of work we are able and willing to put in at

this stage (cf. requirement (iii, we obtain various sufficient conditions for
a graph to be Hamiltonian. Of course, the case k = n of Theorem 2 is ob-
tained without any work, and so is the case k = n - 1, since the conditions
imply immediately that Cn(G) = K n in the first case and Cn-I(G) = K n in
the second, and K n is Hamiltonian if n ~ 3. In order to make better use of
Lemma 11, we shall prove the following ungainly technical lemma.

Lemma 12. Let k, nand t be natural numbers, t < n, and let G be a graph with
vertex set V(G) = {Xl, X2,"" x n}, whose k-closure Ck(G) contains at most
t - 1 vertices of degree n - 1. Then there are indices i,j, 1 :::; i < j :::; n, such
that Xi X j E( G) and each of the following four inequalities holds:
j ~ max {2n - k - i, n - t + I}
d(x;):::; i +k - n, d(x) :::; j + k - n - 1, (4)
d(Xi) + d(x) :::; k - 1.

Remark. Note that it is not assumed that the degree sequence d(XI),
d(X2), ... , d(xn) of G is ordered in any way.

PROOF. The graph H = Ck(G) is not complete so we can define two indices
i and j as follows:
j = max{l: dH(XL) =F n - I},
i = max{l: x/Xj E(H)}.
Then XiXj E(H) so, by (3),
dH(xJ + dH(xj) :::; k - 1,
which implies the fourth inequality in (4). Each of the vertices

has degree n - 1 in H so
n-j:::;t-l
and
78 IV Extremal Problems

The vertex x j is joined to the n - j vertices following it and to the j - i - I

vertices preceding it, so
dH(x j ) ;;:: n- j +j - i-I =n- i- 1.
These inequalities enable us to show that the indices i, j, 1 ~ i <j ~ n,
satisfy the remaining three inequalities in (4). Indeed,
dG(x;) ~ dH(x;) ~ k - 1 - dH(x) ~ k - 1 - (n - i-I) = i +k - n,
dG(x;) ~ dH(x) ~ k - 1 - dH(x;) ~ k - 1 - (n - j) = j +k - n - 1,
i +j ;;:: (n - dH(x j) - 1) + (n - dH(x;)) ;;:: 2n - 1 - (k - 1) = 2n - k.
o
Combining Lemma 11 and Lemma 12 (with t = n - 1 and k = n or
n - 1) we obtain rather complicated but useful conditions for the existence
of a Hamilton path or cycle.

Theorem 13. Let G be a graph with vertex set V(G) = {Xl' X2' ... , Xn},
n ;;:: 3. Let e = 0 or 1 and suppose there are no indices i, 1, 1 ~ i < j ~ n,
such that X;Xj E(G) and
j;;::n-i+e,
d(x;) ~ i - e, d(x) ~ j - 1 - e,
d(x;) + d(xj) ~ n - 1 - e.
Ife = 0 then G has a Hamilton cycle and ife = 1 then G has a Hamilton path.
o
This theorem has the following beautiful consequence.

Corollary 14. Let G be a graph with degree sequence d l ~ d2 ~ ... ~ dn,

n ;;:: 3, and let e = 0 or 1. Suppose
dn- k - E ;;:: n- k whenever dk ~ k - e < !(n - e).
Ife = 0 then G has a Hamilton cycle and ife = 1 then G has a Hamilton path.
o
We draw the attention of the reader to Exercises 21 and 22 which show
that the assertions in the corollary above are in some sense best possible.
In particular, if d 1 ~ d 2 ~ ... ~ dn is a graphic sequence such that
n
dn - k < n - k and dk ~ k < 2
then there is a graph G with vertex set {Xl> X2, ... , xn} such that d(x;) ;;:: d;,
1 ~ i ~ n, and G does not have a Hamilton cycle.
3 Hamilton Paths and Cycles 79

x o---oi( x iIE---o(

Figure IV.S. An x-path and a simple transform of it.

There is another customary way of showing that a graph has a Hamilton

cycle or path. Let S be a longest xo-path in G, that is a longest path beginning
at xo: S = XOXl ... Xk Then r(Xk) C {Xo, Xl"'" Xk- d since otherwise S
could be continues to a longer path. If Xk is adjacent to Xj, 1 :5: j < k - 1,
then S' = Xo Xl' .. Xj Xk Xk _ 1 Xj + 1 is another longest xo-path. We call
S' a simple transform of S. It is obtained from S by erasing the edge XjXj+ 1
and adding to it the edge XkXj' Note that if S' is a simple transform of S
then S is a simple transform of S' and S has exactly d(xk) - 1 simple trans-
forms. The result of a sequence of simple transforms is called a transform
(See Figure IV.5).
Let L be the set of end vertices (different from xo) of transforms of Sand
put N = {XjES:Xj _ l E L or x j + l EL} and R = V - NuL. Thus L is the
collection of the last vertices of the transforms, N is the collection of their
neighbours on Sand R is the rest of the vertices.

Theorem 15. The graph G has no L-~ edges.

PROOF. Recall that there is no L - (V(G) - V(S)) edge, since S is a longest
xo-path, so in particular V(S) = V(P) for every transform P of S.
Suppose XiXjEE(G), where XiEL and XjER. Let Si be a transform of S
ending in Xi' Since at least one neighbour of Xj on Si is the end vertex of a
simple transform of Si' Xj cannot have the same neighbours on Sand Si'
since otherwise Xj would belong to N. However, when the edge xl Xl
j' = j - 1 or j + 1, is erased during a sequence S ~ S' ~ S" ~ ... ~ Si of
simple transformations, one of the vertices xl' Xj is put into L and the other
into N. Thus Xj E L u N = V(G) - R, contradicting our assumption. D
We will make use of Theorem 15 later in Chapter VII. The last two
results of this section are also obtained with the use of simple transforms.

Theorem 16. Let W be the set of vertices of even degree in a graph G and let
Xo be a vertex of G. Then there are an even number of longest xo-paths ending
in W.
PROOF. Let H be the graph whose vertex set is the set L of longest xo-paths
in G, in which P l E L is joined to P 2 E L iff P 2 is a simple transform of Pl'
Since the degree of P = XOXl ... Xk E Lin H is d(xk) - 1, the set of longest
paths ending in W is exactly the set of vertices of odd degree in H. The
80 IV Extremal Problems

number of vertices of odd degree is even in any graph, so the proof is com-
plete. []

Theorem 17. Let G be a graph in which every vertex has odd degree. Then
every edge ofG is contained in an even number of Hamilton cycles.
PROOF. Let XoY E E(G). Then in G' = G - xoy only Xo and y have even
degree, so in G' there are an even number of longest xo-paths that end in y.
Thus either G has no Hamilton cycle that contains XoY or it has a positive
even number of them. []

4 The Structure of Graphs

As we saw in 2 on the example of a Turim graph 1',.-1 (n), a graph of order n
with !r - 2)/(r - 1) + o(1n edges does not necessarily contain a K', a
complete graph of order r. (Here and in what follows we use Landau's nota-
tion 0(1) to denote a function tending to 0 as n --+ co.) The main aim of this
section is to prove the deep result of Erdos and Stone, proved in 1946, that
if e > 0 then en 2 more edges ensure not only a K', but a K,(t), a complete
r-partite graph with t vertices in each class, where t --+ co as n --+ co.
A sharper version of the theorem is due to Bollobas and Erdos; this is
the result we shall present here. The simple proof is based on two lemmas.
The first lemma enables us to replace a condition on the size by a condition
on the minimum degree; the second is a technical lemma in the vein of
Lemma 7.

Lemma 18. Let 0 < a < a +e<

1 and put I] = (e/2) 1/2. Then every graph
G of order n ~ 2a/e and size at least !(a + e)n 2 contains a subgraph H with
IH I = h ~ I]n and minimum degree at least ah.
PROOF. Define a sequence of graphs Go = G ~ G1 ~ G2 ::> .. with IGkl =
n - k as follows. If Gk has a vertex Xk of degree less than a(n - k) then put
Gk+ 1 = Gk - x k, if c5(Gk) ~ a(n - k) then terminate the sequence. The
assertion of the theorem follows if we show that the sequence has to terminate
in a graph Gk with h = n - k ~ I]n.
Suppose this is not so and we arrive at a graph Gk with h = n - k < I]n.
By construction
e(G k ) ~ !(a + e)n 2 - a{n + (n - 1) + ... + (h + I)}.
On the other hand
4 The Structure of Graphs 81

so
en 2 - an + ah + h ~ (1 - a)h 2,

contradicting our assumptions. o

Lemma 19. Let r ~ 2, 1 ~ t ~ q and N = n - (r - 1)q ~ 1. Let G be a
graph of order n that contains a Kr-I(q), say K, but does not contain a Kr(t).
Then G has at most
e = r - 2)q + t)N + 2qN 1 -(I/t)
edges of the form xy, where x E K and y E G - K.
PROOF. Let SI' S2' ... , S,_I be the vertex classes of K and put S = Url-I Si'
ThuslSd = qandlSI = (r -1)q.LetxI,x2, ... ,x N betheverticesofG - K
and put Ai = r(Xi) r. S, i = 1,2, ... ,N. With this notation the lemma
states that
N

I 1
lAd ~ e.

Call a set We S a centipede if it has exactly t vertices in each set Si'

Since G does not contain a KrCt), every centipede is contained in at most
t - 1 of the Ai' We may assume that IAil ~ (r - 2)q + t for i ~ M and
IAil < (r - 2)q + tfor i > M. Fori ~ Mandj ~ r - 1 putaij = IAi r. Sjl.
Then aij ~ t and Ai contains

rff (aij)
j= 1 t
centipedes. Since altogether there are m
r - 1 centipedes and each one of
them is contained in at most t - 1 sets Ai' we have

I rfl (aij)
i=lj=1 t
~ (t _ 1) (q)r-I.
t
The required estimate follows from this inequality and the convexity of
the binomial coefficients. Indeed putting ai = Ii:
t aij - (r - 2)q, one has
~nl (aij) ~ (q)r-2(ai),
J=I t t t
so

.f (a
,= 1 t
i) ~ (t - 1) (q).
t

As G) is a convex function of x, on putting a = I~ aJM the last inequality

implies
82 IV Extremal Problems

Thus

and
M
I IAil < 2qM- 1/t + M((r - 2)q + t).
1

Since
N
I lAd :-s.; (N - M)((r - 2)q + t),
M+1

the required estimate does hold. o

Theorem 20. Let r ~ 2 be an integer and let 0 < c < !(r - 1). Then there
exists ad = d(c, r) > 0 such that ifn is sufficiently large and
m ~ H(r - 2)/(r - 1) + c}n Z
then every graph of order n and size m contains a KrCt) with t ~ Ld log nJ.

PROOF. We shall apply induction on r. For r = 2 Theorem 9 implies that

any d < -1/log(2c) will do for d(c, 2). Indeed, if there is a graph G(n, m)
without a Kz(t) where m ~ en z and t = Ld log nJ, then by Theorem 9 we
have

that is
2en l/t :-s.; (t - 1)l/t + tn - 1+ (lft).
However, this inequality cannot hold if n is sufficiently large.
Now assume that r > 2 and the result holds for smaller values of r. For
the sake of convenience we write Kix) for KiLxJ) and we denote by Gin)
a graph of order n and size at least {!(r - 2)/(r - 1) + c}n z .
Put Cr = H(r - 2)/(r - 1) - (r - 3)j(r - 2)} = {2(r - l)(r - 2)} -1 > 0
and er = d(c" r - 1) > O. Then er depends only on r, so by the induction
hypothesis if n is large then every Gin) contains a K r- 1(e r log n) = Kr-1(q).
We shall show that d = d(c, r) = er(r - 1)c/4 will do for the theorem, that is
every Gin) contains a KrCd log n) = KrCt) ifn is large enough.
Suppose that, contrary to this assertion, there are arbitrarily large values
ofnfor which some Gin) does not contain a KrCd log n) = Kr(t). Given any
ho ~ l/c, let G = Gin), n ~ ho/YJ, contain a K r- 1(q), say K, but no KrCt).
Hence by Theorem 9 there are at most !(t - l)qZ-(l/t) + !qt edges in a
subgraph spanned by the q vertices in a class of K, so there are at most

A = ( r -2. 1)qZ + (r - l){!{t - 1)qZ-1/t + !qt}

4 The Structure of Graphs 83

edges in the subgraph spanned by the whole of K. Furthermore, by Lemma

19, at most B = r - 2)q + t)(n - (r - l)q) + 2qn 1- 0 /t) edges join K to
G - K. Finally, as <5(G) 2': {(r - 2)/(r - 1) + 6}n, we must have
2A +B 2': (r - l)q<5(G) 2': {(r - 2) + 6(r - l)}qn,
and so
2qn 1- O /t ) + (r - 1)q2 + tn :::;; 6(r - l)qn.
We have arrived at a contradiction, since this inequality does not hold for
larger values of n. D

Remark. In a certain sense Theorem 20 is best possible: for every 6 and r

there is a constant d:
tending to 0 with 6 such that the graph described in
the theorem need not contain a K,(t) with t = Ld: log nJ. In fact, we shall
see in Chapter VII Theorem 3 that if 0 < 6 < ! and d! > - 2jlog(26), then
for every sufficiently large n there is a graph G(n, m) not containing a Kit),
where m= Lcn 2J and t = Ld! log nJ. This result will imply immediately
(cf. Exercise 6 of Chapter VII) that if r > 2 and 0 < 6 < !(r - 1) - 2 then
any value greater than -2/log(2(r - 1)26) will do for,d:.
Since d log n --+ 00 as n --+ 00, Theorem 20 has the following immediate
corollaries.

Corollary 21. Let F = Kit), where r 2': 2 and t 2': 1. Then the maximal size
of a graph of order n without a Kr(t) is

ex(n; F) = 2:1{r-2
r _ 1 + 0(1) } n 2. D

Corollary 22. Let F 1, F 2, .. , F/ be non-empty graphs. Denote by r the mini-

mum of the chromatic numbers of the F;, that is let r be the minimum number
for which at least one of the Fi is contained in an F = Kr(t) for some t. Then
the maximal size of a graph of order n not containing any of the Fi is

ex(n; F 1 , F 2, ... , F/) = ~ {~ =~ + 0(1)}n 2.

PROOF. The Tunin graph 1',.-1 (n) does not contain any of the F;, so

ex(n; Fb F 2,, F/) 2': e(1',.-l(n)) = tr- 1(n) = ~ t =~ + 0(1)}n 2.

Conversely, since, say F j C F = K,(t) for some j and t,

ex(n; Fb F 2,, F/) :::;; ex(n; F) :::;; ex(n; F) = ~ t =~ + 0(1)}n 2. D

Theorem 20 is the basis of a rather detailed study of the structure of

extremal graphs, giving us considerably more accurate results than Corollary
22. This theory is, however, outside the scope of our book.
84 IV Extremal Problems

The density of a graph G of order n is defined to be e(G)/G). The upper

density of an infinite graph G is the supremum of the densities of arbitrarily
large finite subgraphs of G. It is surprising and fascinating that not every
value between 0 and 1 is the upper density of some infinite graph; in fact,
the range of the upper density is a countable set.

Corollary 23. The upper density of an infinite graph G is 1, t, ~, i, ... , or o.

Each of these values is the upper density of some infinite graph.

PROOF. Let Gr be the complete r-partite graph with infinitely many vertices
in each class. Since the density of Kit) tends to 1 - (1/r) as t tends to 00,
the upper density of Gr is 1 - (l/r), proving the second assertion.
Now let IY. be the upper density of G and suppose

1
IY. > 1 - -- (r ~ 2).
r - 1

Then there is an e > 0 such that G contain graphs H k of order nk with nk -+ 00

satisfying e(H k) ~ t(1 - (1/(r - 1)) + e)nr By Theorem 19 each Hk con-
tains a subgraph Kr(t k) with tk -+ 00; the subgraphs Kr(t k) show that IY. ~
1 - (l/r). 0

EXCERCISES

1. (i) Let G be a graph of order n, maximum degree ~ :2: 3 and diameter d. Let
no(g, b) be as in Theorem 1. Prove that n ::;; no(2d + 1, ~) with equality iff G
is ~-regular and has girth 2d + 1.
(ii) Let G be a graph of order n, maximum degree ~ :2: 3, and suppose every
vertex is within distance d - 1 of each pair of adjacent vertices. Prove that
n ::;; no(2d, ~), with equality iff G is ~-regular and has girth 2d.
2. Prove Theorem 4 for k = 3 and 4.
3. Prove that the maximal number of edges in a graph of order n without an even
cycle is Lt(n - 1)J. Compare this with the maximum size of a graph without
an odd cycle.
4. Show that a tree with 2k endvertices contains k edge-disjoint paths joining distinct
end vertices.
5. Suppose x is not a cutvertex and has degree 2k. Prove that there are k edge-
disjoint cycles containing x. [Cf. Exercise 4.]
6: Show that a graph with n vertices and minimum degree L(r - 2)n/(r - l)J +1
contains a K'.
7. Let G have n :2: r + 1 vertices and t,-1 (n) + 1 edges.
(i) Show that G contains two K' subgraphs with r - 1 vertices in common.
(ii) Prove that for every p, r ::;; p ::;; n, G has a subgraph with p vertices and at least
t'-l(P) + 1 edges.
Exercises 85

8. Prove that for n 2: 5 every graph of order n with Ln 2/4J + 2 edges contains two
triangles with exactly one vertex in common.
9. Prove that if a graph with n vertices and Ln 2/4J - I edges contains a triangle
then it contains at least Ln/2J - 1- 1 triangles. [Hint. Let XIX2X3 be a triangle
and denote by m the number of edges joining {XI' X2, X3} to V(G) - {XI' X 2 , X3}
Estimate the number of triangles in G - {Xl> X2' X3} and the number of triangles
sharing a side with XIX2X3.J
10; (i) Show that the edges of a graph of order n can be covered with not more than
Ln2/4J edges and triangles.
(ii) Let G be a graph with vertices XI' X2, .. , Xn, n 2: 4. Prove that there is a
set S, IS I ::; Ln 2/4J, containing non-empty subsets X I, X 2, ... , X n such that
XiXj is an edge of G if Xi (\ Xj oft 0.

11~ Let 1 ::; k < n. Show that every graph of order n and size (k - l)n - +1 m
contains a subgraph with minimum degree k, but there is a graph of order nand
size (k - l)n - m in which every subgraph has minimum degree at most k - 1.
[Hint. Imitate the proof of Lemma 18.J
IT Show that a graph of order n and size (k - l)n - m+ 1 contains every tree
of order k + 1.
13. Let G be a graph of order n which does not contain a cycle and one of its diagonals.
Prove that if n 2: 4 then G has at most 2n - 4 edges.
Show that if n 2: 6 and G has 2n - 4 edges then G is the complete bipartite
graph K(2, n - 2). [Hint. Consider a longest path in G.J
14~ Let k 2: 1 and let G be a graph of order n without an odd cycle of length less than
2k + 1 2: 5. Prove that b(G)::; Ln/2J and Tin) is the only extremal graph,
unless n = 2k + 1 = 5, in which case there is another extremal graph, C 5
15 -; Let G be a graph of order n without an odd cycle of length less than 2k + 1 2: 5.
r
Prove that if G does not contain n/21 independent vertices then (j(G)::;
2n/(2k + 1). Show that equality holds only for n = (2k + l)t and the extremal
graphs are obtained from a cycle C 2 k+ I by replacing each vertex by t vertices,
as in Figure IV.6.

Figure IV.6. The graph CSC2).

16. Let Xl> X2, ... , Xn be vectors of norm at least 1 in a Euclidean space. Prove that
there are at most Ln 2/4J unordered pairs i,j such that IXi + Xjl < 1. [Hint. Show
that if IXII = IX21 = IX31 = 1 then IXi + Xjl 2: 1 for some i,j, 1 ::; i < j ::; 3.J
1r Let X and Y be independent identically distributed random variables taking
values in a Euclidean space. Prove that P( IX + YI 2: x) 2: !P( IX I 2: X)2 for
every X 2: o.
86 IV Extremal Problems

18. Let x[, X2, ... , X3pE [R2 be such that IXi - xjl:$; 1. Prove that at most 3p2 of
the distances IXi - Xjl are greater than J2/2. [Hint. Show that among any four
of the points there are two within distance J2/2 of each other.]
19. The clique number cl( G) of a graph G is the maximum order of a complete subgraph
of G. Show that if a regular graph of order n contains a complete subgraph of
order Lnl2J + 1 then it is a complete graph, but for every p, I :$; p :$; n12, there
is a regular graph of order n with cl(G) = p.
20:- We say that a set We V(G) covers the edges of a graph G if every edge of Gis
incident with at least one vertex in W Denote by cxo(G) the minimum number of
vertices covering the edges of G. Prove that if G has n vertices and m edges then
cxo(G) :$; 2mnj(2m + n), with equality iff G = pK' for some p and r, that is iff each
component of G is a K' for some r. [Hint. Note that cxo(G) = n - cl(G), and if
cl(G) = p then by Tunin's theorem e(G) :$; tp(n), so m = (2) - e(G) ~ m- tin).]
21. (Cf. Corollary 14.) Let d 1 :$; d 2 :$; ... :$; d n be a graphic sequence such that for
some k
n
dk :$; k < 2 and dn - k :$; n - k - 1.

Show that there is a non-Hamiltonian graph G with vertex set {Xl' X2, ... , xn}
such that d(Xi) ~ d;, 1 :$; i :$; n (cf. Figure IV.7).

Figure IV.7. The graph (K 3 u E 3) + K3 has no Hamilton cycle and (K 2 u E 3)

+ K2 has no Hamilton path.
22. (Cf. Corollary 14) Let d l :$; d 2 :$; ... :$; dn be a graphic sequence such that for
some k
dk :$; k - 1 < ten - 1) and dn + l-k 2': n- k.
Prove that there is a graph G with vertex set {x 1> X2, ... , xn} such that d(Xi) ~ di,
1 :$; i :$; n, and G does not contain a Hamilton path (cf. Figure IV.7).
23. Prove that a non-Hamiltonian graph of order n 2': 3 has at most (2) - (n - 2)
edges and there is a unique extremal graph.
Prove that a graph of order n ~ 2 without a Hamilton path has at most
m- (n - 3) edges and K n - l u El is the unique extremal graph.

24. Given b < n12, determine the maximal number of edges in a graph G of order n
without a Hamilton cycle (path), provided beG) = b.
25. Prove Theorem 4 by making use of simple transforms of a longest xo-path P =
XOXI .. . x,. [Hint. Apply induction on n. If beG) :$; k12, the result follows by
induction, otherwise consider the set L of end vertices of simple transforms of P.
Put I = IL I, r = max xEL d(x) and note that I 2': r and the neighbours of each
X E L are contained in {x" X,_ [, ... , X'-k+ d. Deduce that e(G) - e(G - L)
:$; I(k - I) + I(r + I - k) :$; kl/2 and complete the proof by applying the induction
hypothesis to G - L.]
Notes 87

26. Let 1 < al < a2 < ... < ak S x be natural numbers. Suppose no ai divides
the product of any two others. Prove that k s n(x) + x 2i3 , where, as usual, n(x)
denotes the number of primes not exceeding x. [Hint. Put VI = {I, 2, ... , LX 213 J}
and V2 = {x: x2i3 S b s x and b is a prime}. Show first that ai = biCi' where
bi' ci E V = VI U V2 . Let G be the graph (with loops) with vertex set V whose
edges (loops) are biCi' Note that G does not contain a path of length 3.J
27.+ (Cf. Exercise 26). Let 1 < a l < a2 < ... < ak S x be natural numbers. Suppose
aiaj # akal unless {i,j} = {h, I}. Prove that k s n(x) + CX 314 for some constant
c> O. [Hint. The graph G of Exercise 26 contains no quadrilaterals; apply
Theorem 8 to the bipartite subgraph of G with vertex classes VI and V2 . Recall the
prime number theorem, namely that (n(x)log x)/x ---> 1 as x ---> oo.J
28. Denote by Dk(n) the maximum number of occurrences of the same positive
distance among n points in IRk. Prove that lim n _ oo Dk(n)/n 2 = t - (1/2Lk/2J) if
k ;::: 2. [Hint. (i) Note that if x E {z E IRk: ZI + d = 1 and Zi = 0 if i > 2} and
Y E {ZE IRk:Z~ + Z~ = 1 and Zi = 0 ifi # 3 or4} then Ix - yl = j2. (ii) Deduce
from Theorem 20 that Dk(n) is at least as large as claimed.J

Notes
The results concerning Hamilton cycles and paths presented in the chapter
all originate in a paper of G. A. Dirac: Some theorems on abstract graphs,
Proc. London Math. Soc. 2 (1952) 69-81. The notion of the k-closure dis-
cussed in 3 was introduced in 1. A. Bondy and V. Chvatal, A method in graph
theory, Discrete Math. 15 (1976) 111-135; Corollary 14, characterizing
minimal forcibly Hamiltonian degree sequences, is in V. Chvatal, On
Hamilton's ideals, J. Combinatorial Theory Ser. B 12 (1972) 163-168. The
special case of Theorem 17 concerning cubic (that is 3-regular) graphs was
proved by C. A. B. Smith, and was first mentioned in W. T. Tutte, On Hamil-
tonian circuits, J. London Math. Soc. 21 (1946) 98-101; the theorem itself is due
to A. G. Thomason, Hamiltonian cycles and uniquely edge colourable graphs,
Annals of Discrete Math. 3 (1978) 259-268.
The fundamental theorem of Turim concerning complete subgraphs is in
P. Turan, On an extremal problem in graph theory (in Hungarian), Mat. es
Fiz. Lapok 48 (1941) 436-452; its extension, Theorem 6, giving more in-
formation about the degree sequence of a graph without a f('" is in P. Erdos,
On the graph theorem of Tunin (in Hungarian), Mat. Lapok 21 (1970)
249-251. The Erdos-Stone theorem is in P. Erdos and A. H. Stone, On the
structure of linear graphs, Bull. Amer. Math. Soc. 52 (1946) 1087-1091, its
extension, Theorem 20, is in B. Bollobas and P. Erdos, On the structure of
edge graphs, Bull. London Math. Soc. 5 (1973) 317-321. A rather detailed
exposition of the results concerning the structure of extremal graphs,
referred to at the end of 4, can be found in Chapter VI of B. Bollobas,
Extremal Graph Theory, Academic Press, London-New York-San
Francisco, 1978; Chapter III of the same book concerns cycles and contains
many more results than we could present in this chapter.
CHAPTER V
Colouring

We wish to arrange the talks in a congress in such a way that no participant

will be forced to miss a talk he would like to hear, that is that there are no
undesirable clashes. Assuming a good supply of lecture rooms enabling us to
hold as many parallel talks as we like, how long will the program have to
last? Let us reformulate this question in terms of graphs. Let G be the graph
whose vertices are the talks and in which two talks are joined iff there is a
participant wishing to attend both. What is the minimum value of k for which
V(G) can be partitioned into k classes, say VI, V2 , . , l-k, such that no edge
joins two vertices of the same class? As in 4 Chapter IV, we denote this
minimum by X(G) and call it the (vertex) chromatic number of G. The termin-
ology originates in the usual definition of X( G): a colouring of the vertices
of G is an assignment of colours to the vertices in such a way that adjacent
vertices have distinct colours; X(G) is then the minimal number of colours in
a (vertex) colouring of G.
Let us remark here that we shall use real colours (red, blue, ... ) only if
there are few colours, otherwise the natural numbers will be our "colours".
Thus a k-colouring of the vertices of G is a function c: V(G) -> {l, 2, ... , k}
such that each set c-l(j) is independent.
Another scheduling problem goes as follows. Each of n businessmen
wishes to hold confidential meetings with some of the others. Assuming
that each meeting lasts a day and at each meeting exactly two businessmen
are present, in how many days can the meetings be over? In this case one
considers the graph H whose vertices correspond to the n businessmen and
two vertices are adjacent iff the two businessmen wish to hold a meeting.
Then the problem above asks for the minimal number of colours in an edge-
colouring of H, that is in a colouring of the edges of H in such a way that no

88
l Vertex Colouring 89

two adjacent edges have the same colour. This number, denoted by X'(H),
is the edge chromatic number of H.
In this chapter we shall present some of the basic properties of the vertex
and edge chromatic numbers. The final section (3) is devoted to colouring
planar graphs and to a very brief outline of the proof of the best known
result in graph theory, the so called four colour theorem.

1 Vertex Colouring
In Chapter I. 2 we noted the trivial fact that a graph is bipartite iff it does
not contain an odd cycle. Thus X(G) ;;::: 2 iff G contains an edge and X(G) ;;::: 3
iff G contains an odd cycle. For k ;;::: 4 we do not have a similar character-
ization of graphs with chromatic number at least k, though there are some
complicated characterizations (cf. Exercises 32-35). Of course, if G => Kk
then X( G) ;;::: k and if G does not contain h + 1 independent vertices then
X( G) ;;::: IG I/h. However, it is not easy to see that there are triangle-free
graphs of large chromatic number (cf. Exercise 11), though in Chapter VII
we shall show that there exist graphs with arbitrarily large chromatic
number and arbitrarily large girth. The difficulty we encounter in finding
such graphs shows that it would be unreasonable to expect a simple character-
ization of graphs with large chromatic number. Thus we shall concentrate
on finding efficient ways of colouring a graph.
How should one try to colour the vertices of a graph with colours 1,2, ... ,
using as few colours as possible? A simple way is as follows. Order the
vertices, say Xl' X 2 , , Xn and then colour them one by one: colour xlI,
then colour X 2 1 if X 1X 2 rtE(G) and 2 otherwise, and so on, colour each vertex
the smallest colour it can have at that stage. This so-called greedy algorithm
does produce a colouring, but this colouring may and usually does use many
more colours than necessary. Figure V.I shows a bipartite (i.e., 2-colourable)
graph for which the greedy algorithm wastes four colours. However, it is
easily seen (Exercise 3') that for every graph the vertices .can be ordered in
such a way that the greedy algorithm uses as few colours as possible. Thus it is
not surprising that it pays to investigate the number of colours needed by
the greedy algorithm in various orders of the vertices.

Figure V.l. In the order Xl>X2,"" Xs the greedy algorithm needs four colours.
90 V Colouring

Let us start with an immediate consequence of the algorithm.

Theorem 1. Let k = maxH c5(H), where the maximum is taken over all spanned
subgraphs of G. Then x( G) ::; k + 1.

PROOF. Write Xn for a vertex of degree at most k, and put H n - l = G - {x n }.

By assumption H n - l has a vertex of degree at most k. Let Xn-l be one of
them and put H n - 2 = H n - l - {x n - d = G - {x n , Xn- d. Continuing in
this way we enumerate all the vertices.
Now the sequence Xl' X2, .. , Xn is such that each Xj is joined to at most
k vertices preceding it. Hence the algorithm will never need colour k + 2
to colour a vertex. 0

It is, of course, very easy to improve the efficiency of the greedy algorithm.
If we already know a subgraph H 0 which can be coloured with few colours,
in particular, if we know a colouring of H 0 with X(H 0) colours, then we may
start our sequence with the vertices of H 0, colour H 0 in an efficient way
and apply only then the algorithm to colour the remaining vertices. This
gives us the following extension of Theorem 1.

Theorem 2. Let Ho be a spanned subgraph of G and suppose every subgraph

H satisfying Ho c H c G, V(Ho) =I V(H), contains a vertex x E V(H)
- V(Ho) with dH(x) ::; k. Then

x(G) ::; max{k + 1, X(H)}. o

In some cases the problem of colouring a graph can be reduced to the
problem of colouring certain subgraphs of it. This happens if the graph is
disconnected or has a cutvertex or, slightly more generally, contains a
complete subgraph whose vertex set disconnects the graph. Then we may
colour each part separately since, at worst by a change of notation, we can
fit these colourings together to produce a colouring of the original graph,
as shown in Figure V.2.

1~1 lJ;l'
G

, 3 1

3~ G3

3 2
Figure V.2. The vertex set of the thick triangle disconnects G;
l Vertex Colouring 91

As a rather crude consequence of Theorem 1 we see that x( G) ::; d + 1,

where d = d(G) is the maximum degree of G, since maxHcG (j(H) ::; d(G).
Furthermore, if G is connected and not d-regular then clearly maxHcG (j(H)
::; d - 1 so x( G) ::; A The following result, due to Brooks, takes care of the
regular case.

Theorem 3. Let G be a connected graph with maximal degree d. Suppose G is

neither a complete graph nor an odd cycle. Then X(G) ::; d.
PROOF. We know already that we may assume without loss of generality that
G is 2-connected and d-regular. Furthermore, we may assume that d ~ 3,
since a 2-regular 3-chromatic graph is an odd cycle.
If G is 3-connected, let Xn be any vertex of G and let XI' X2 be two non-
adjacent vertices in rexn). If G is not 3-connected, let Xn be a vertex for which
G - Xn is separable, so has at least two blocks. Since G is 2-connected, each
endblock.of G - Xn has a vertex adjacent to X n Let X I and X 2 be such vertices
belonging to different end blocks.
In either case we have found vertices XI' X2 and Xn such that G - {Xl> X2}
is connected, X I X2 f/: E(G) but XIX n E E(G) and X2Xn E E(G). Let Xn-I E V -
{x I, X 2, xn} be a neighbour of X n , let Xn - 2 be a neighbour of Xn or Xn _ I, etc.
Then the order XI' X 2 , X3""'Xn is such that each vertex other than Xn is
adjacent to at least one vertex following it. Thus the greedy algorithm will
use at most d colours, since XI and X2 get the same colour and Xn is adjacent
to both. 0

Another colouring algorithm can be obtained by reducing the problem to

colouring two other graphs derived from G. This reduction also enables us to
obtain some information about the number of colourings of a graph with a
given set of colours.
Let a and b be non-adjacent vertices of a graph G. Let G' be obtained from
G by joining a to b, and let G" be obtained from G by identifying a and b.
Thus in G" there is a new vertex ab instead of a and b, which is joined to the
vertices adjacent to at least one of a and b (Figure V.3). The colourings of G
in which a and b get distinct colours are in 1-1 correspondence with the
colourings of G'. Indeed c: V(G) -+ {I, 2, ... , k} is a colouring of G with
c(a) =1= c(b) iff c is a colouring of G'. Similarly the colourings of G in which

G G' GU

Figure V.3. The graphs G' and G U

92 V Colouring

a and b get the same colour are in a 1-1 correspondence with the colourings
of Gil. In particular, if for a natural number x and a graph H we write PH(X)
for the number of colourings of a graph H with colours 1, 2, ... , x, then
PG(X) = pdx) + PG"(x), (1)
By definition X(G) is the least natural number k for which PG(k) ~ 1. Thus
both the remarks above and relation (1) imply that
x( G) = min {xC G'), X( Gil)}. (2)
The basic properties of PH(X) are given in our next result.

Theorem 4. Let H be a graph with n ~ 1 vertices, m edges and k components.

Then
n-k
PH (X) = L (-IY ai xn - i,
i=O

whereao = l,a 1 = mandai> Oforeveryi,O.::; i.::; n - k.

PROOF. We apply induction on n + m. For n + m = 1 the assertions are
trivial so we pass to the induction step. If m = 0, we are again home since
in this case k = n and as every map f: V(H) ~ {I, 2, ... , x} is a colouring
of H, we have PH(X) = xn. If m > 0 we pick two adjacent vertices of H, say
a and b. Putting G = H - ab we find that G' = H. Since e( G) = m - 1 and
IGill + e( Gil) .::; n - 1 + m, by the induction hypothesis the assertions of the
theorem hold for PG(x) and PG''(x). Note now that Gil has k components and
G has at least k components. Therefore
n-k
PG(x) = xn - (m - l)x n - 1 + L (_l)ibix"-i
i=2

where bi ~ 0 for each i, and

n-k
PG"(x) = x n- 1 - L (-IYc xn - i i,
i=2

where Ci > 0 for each i. Hence, by (1),

n-k
PH(X) = pdx) = PG(x) - PG''(x) = xn - mx n - 1 + L (-IY(b i + cJx n - i
i=2

n-k
= xm - mx n - 1 + L (-l)iaixn-i,
i=2

where ai > 0 for each i. o

Because of Theorem 4 we are justified in calling PH(X) the chromatic
polynomial of H. The chromatic polynomial is closely connected to various
enumeration problems, some of which will be discussed in Ch. VIII. In par-
2 Edge Colouring 93

ticular, the modulus of each coefficient of PH(X) is the number of certain

subgraphs of H (cf. Exercise 12+).
Let us describe now the algorithm based on the reduction G --+ {G', Gil}.
Given a graph G, construct a sequence of graphs Go, G 1 , . . . as follows. Put
Go = G. Having constructed Gi , if Gi is complete, terminate the sequence,
otherwise let Gi + 1 be Gi or Gi'. The sequence has to end in a complete graph
Gn say of order IGt I = k. A k-colouring of Gt can easily be lifted to a k-
colouring of the original graph G, so X(G) :::; k. In fact, equality (2) shows
that X( G) is the minimum order of a complete graph in which a sequence
Go, G1, . can terminate.
There are other problems that can be tackled by the reduction G--+
{G', Gil}; a beautiful example is Exercise 16+.

2 Edge Colouring
In a colouring of the edges of a graph G the edges incident with a vertex
get distinct colours, so x'(G), the edge chromatic number, is at least as
large as the maximum degree, .1.(G) = max x d(x),
x'(G) ~ .1.(G). (3)
At first sight it is somewhat surprising that this trivial inequality is in fact,
an equality for large classes of graphs, including the class of bipartite graphs.
Indeed, Exercise 22 of Chapter III, which is a consequence of Hall's theorem,
asserts that the edge set E(G) of a bipartite graph G can be partitioned into
.1.(G) classes of independent edges, that is x'(G) = .1.(G).
Another trivial lower bound on x'(G) follows from the fact that if G does
not contain /3 + 1 independent edges, then each colour class has at most /3
edges so we need at least re( G)//31 colour classes to take care of all the edges:
x'(G) ~ re(G)//31. (4)
Similarly to Theorem 9 of Chapter 1, it is easily shown that if G is a complete
graph of order 2 then equality holds in (4), that is X'(Kn) = n - 1 if n is
even and x'(K n) = n if n ~ 3 is odd (Exercise 28).
How can one obtain an upper bound for x'(G)? Since each edge is adjacent
to at most 2(.1.(G) - 1) edges, Theorem 1 implies that
X'(G) :::; 2.1.(G) - 1.
Furthermore, if .1.(G) ~ 3, the theorem of Brooks gives
X'(G) :::; 2.1.(G) - 2.
At first sight this inequality seems reasonably good. However, the following
fundamental theorem of Vizing shows that this is not the case, for the edge-
chromatic number is almost determined by the maximum degree.
94 V Colouring

Theorem 5. ~(G) ::; X'(G) ::; ~(G) + 1.

PROOF. Put ~ = ~(G) and assume that we have used 1,2, ... , ~ + 1 to
colour all but one of the edges. We are home if we can show that this un-
coloured edge can also be coloured.
We say that a colour is missing at a vertex z if no edge incident with z
gets that colour. If z is incident with d'(z) ::; d(z) ::; ~ edges that have been
coloured, then ~ + 1 - d'(z) colours are missing at z. In particular, at each
vertex at least one colour is missing. Our aim is to move around the colours
and the uncoloured edge in such a way that a colour will be missing at both
end vertices of the uncoloured edge, enabling us to complete the colouring.
Let XY1 be the uncoloured edge; let s be a colour missing at x and let t1
be a colour missing at Y l' We shall construct a sequence of edges xY 1, XY2, ... ,
and a sequence of colours t 1, t 2 , ... such that ti is missing at Yi and XYi+1
has colour t i . Suppose we have constructed XY1"'" XYi and t 1,, t i
There is at most one edge xy of colour t i . If Y {Y1" .. ,y;}, we put Yi+ 1 = Y
and pick a colour ti+ 1 missing at Yi+ 1, otherwise we stop the sequence.
These sequences have to terminate at most ~(G) terms; let XY1' ... , XYh and
t b ... , th be the complete sequences. Let us examine the two reasons that
may have forced us to terminate these sequences.
(a) No edge xy has colour tho Then recolour the edges XYi' i < h, giving
XYi the colour t i . In the colouring we obtain every edge is coloured, except
XYh' However, since th occurs neither at x nor at Yh, we may complete the
colouring by assigning th to XYh'
(b) For some j < h the edge xYj has colour tho To start with, recolour the
edges XYi' i < j, giving XYi the colour t i . In this colouring the uncoloured edge
is xYJ' Let H(s, t h ) be the subgraph of G formed by the edges of colour s
and tho Each vertex of H(s, t h) is incident with at most 2 edges in H(s, t h) (one of
colour s and the other of colour t h ) so the components of H(s, t h ) are paths
and cycles. Each of the vertices x, Yj and Yh has degree at most 1 in H(s, th),
so they cannot beiong to the same component of H(s, th)' Thus at least one
of the following two cases has to hold.
(bl) The vertices x and Yj belong to distinct components of H(s, th)' Inter-
change the colours sand th in the component containing Yj. Then the colour
s is missing at both x and Yj' so we may complete the colouring by giving xYj
the colour S.
(b2) The vertices x and Yh belong to distinct components of H(s, th)'
Continue the recolouring ofthe edges incident with x by giving XYi the colour
ti for each i < h, thereby making XYh the uncoloured edge. This change does
not involve edges of colour sand t h , so H(s, t h ) has not been altered. Now
switch around the colours in the component containing Yh' This switch
makes sure that the colour s is missing at both x and Yh, so we can use s to
colour the so far uncoloured edge XYh' D

Note that the proof above gives an algorithm for colouring the edges
with at most ~ + 1 colours.
3 Graphs on Surfaces 95

3 Graphs on Surfaces
The most famous problem in graph theory is undoubtedly the four colour
problem: prove that every plane graph is 4-colourable. The weaker assertion,
namely that every plane graph is 5-colourable, is an almost immediate
consequence of Euler's formula.

Theorem 6. Every plane graph is 5-colourable.

PROOF. Suppose the assertion is false and let G be a 6-chromatic plane graph
with minimal number of vertices. By Theorem 12 of Chapter I G has a vertex
x of degree at most 5. Put H = G - x. Then H is 5-colourable, say with
colours 1,2, ... , 5. Each of these colours must be used to colour at least one
neighbourhood of x, otherwise the missing colour could be used to colour x.
Hence we may assume that x has 5 neighbours, say Xl> X2, ... , X5 in some
cyc1ic order about x, and the colour of Xi is i, i = 1,2, ... ,5. Denote by
H(i, j) the subgraph of H spanned by the vertices of colour i and j.
Suppose first that Xl and X3 belong to distinct components of H(1, 3).
Interchanging the colours 1 and 3 in the component of Xl> we obtain another
5-colouring of H. However, in this 5-colouring both Xl and X3 get colour 3,
so 1 is not used to colour any of the vertices Xl' ... , X5. This is impossible
since then X can be coloured l.
Since Xl and X3 belong to the same component of H(1, 3), there is an XI-X3
path P 13 in H whose vertices are coloured 1 and 3. Analogously H contains
an X2-X4 path P 24 whose vertices are coloured 2 and 4. However, this is
impossible, since the cycle XXIP13X3 of G separates X2 from X 4 but P 24
cannot meet this cycle (Figure V.4). D

Clearly not every plane graph is 3-colourable. Indeed, K4 is planar and

it does need 4 colours. Another 4-chromatic planar graph is obtained by join-
ing all five vertices of a C 5 to a sixth vertex. Thus if we want to determine
XO = max{x(G): G is planar} then we see immediately that XO 2:: 4 and
XO ~ 5 and the problem is to prove Xo ~ 4.
Instead of a plane graph we may wish to consider a graph drawn on an
orientable surface of genus y > 0 and ask for XY ' the maximum ofthe chromatic

4
-~--

Figure V.4. The paths P\3 and P 24

96 V Colouring

number of such a graph. We shall see in a moment that in this problem we

encounter a rather different difficulty. For our limited purpose it will suffice
to remark that an orientable surface Sy of genus I' ::::: 1 is obtained from a
sphere by attaching to it I' "handles", and a graph of order n drawn on Sy
has at most 3n + 6(1' - 1) edges. Thus S 1 is the torus, and every toroidal
graph (that is, graph that can be drawn on a torus) of order n has at most 3n
edges. The following easy upper bound on the chromatic number of a graph
drawn on an orientable surface of genus I' ::::: 1 was obtained by Heawood
in 1890.

Theorem 7. The number of colours needed to colour a graph drawn on an

orientable surface of genus I' ::::: 1 is at most
H(y) = Lt{7 + J1 + 48y}J.
PROOF. Let G be the graph in question and let H be any subgraph of G. By
Theorem 1 we are home if we show that (j(H) ::::;; H(y) - 1. If H has n vertices,
it has at most 3n + 6(1' - 1) edges so

(j(H) ::::;; min{n - 1,6+ l12(Y n- 1)J}.

Hence the result follows if

6+l12(Y - 1)J < H( I') - 1.

H(y) + 1 -
With a little work one can check that this inequality indeed holds. 0

By applying a stronger colouring result than the trivial Theorem 1, one

can show that H(y) - 1 colours suffice to colour a graph on Sy unless the
graph contains a complete graph of order H(y). Thus Xy = H(y) iff KH(y)
can be drawn on Sy- Heawood gave a false proof of the assertion Xy = H(y);
the first correct proof, due to Ringel and Youngs, was found only over 75
years later. Note that the difficulty in proving this deep result lies in finding
a drawing of a fixed and single graph, KH(Y), on a surface of genus I' > O. On
the other hand, in order to solve the four colour problem one has to show that
every plane graph can be coloured with four colours. Thus the problem of
determining Xo has almost nothing to do with the problem of determining
Xy forI' ::::: 1.
Before saying a few words about the solution of the four colour problem,
let us point out that H(l) = 7 is easily proved; in fact, this was indeed proved
by Heawood. All one needs is that K7 can be drawn on the torus. Recalling
that a torus can be obtained from a rectangle by identifying opposite sides,
Figure V.5 shows a required drawing.
We saw in Chapter I. 4 that a plane graph G determines a map M = M(G)
consisting of the plane graph G and the countries determined by the plane
graph. A colouring of a map is a colouring of the countries such that no two
3 Graphs on Surfaces 97

2 3

5 5

4 4

2 3
Figure V.S. A drawing of K7 on the torus.

countries sharing an edge in their boundaries get the same colour. Show
that every plane map can be coloured with four colours. This is the original
form of the four colour problem, as posed by Francis Guthrie in 1852. His
teacher, de Morgan, circulated it amongst his colleagues, but the problem was
first made popular in 1878 by Cayley, who mentioned it before the Royal
Society. Almost at once "proofs" appeared, by Kempe in 1879 and by Tait in
1880. Heawood's refutation of Kempe's proof was published in 1890, though
he modified the proof to obtain the five colour theorem. Tait's paper also
contained false assumptions which prompted Petersen to observe in 1891
that the four colour theorem is equivalent to the conjecture that every cubic
planar graph has edge chromatic number three (Exercise 27). Contributions
to the solution since the turn of the century include Birkhoff's introduction
of the chromatic polynomial and works by various authors giving lower
bounds on the order of a possible counterexample. In 1943 Hadwiger made
a deep conjecture containing the four colour problem as a special case:
if X(G) = k then G is contractible to Kk (see Exercises 18-20).
In hindsight the most important advance was made by Heesch, who
developed the method of discharging to find unavoidable sets of reducible
configurations. The problem was at last solved by Appel and Haken in 1976,
making use of a refinement of Heesch's method and fast electronic com-
puters. The interested reader is referred to some papers of Appel and Haken,
and to the book of Saaty and Kainen for a detailed explanation of the under-
lying ideas of the proof. All we have room for is a few superficial remarks.
What makes the five colour theorem true? The following two facts: (i) a
minimal 6-colourable plane graph cannot contain a vertex of degree at
most 5, and (ii) a plane graph has to contain a vertex of degree at most 5.
We can go a step further and ask why (i) and (ii) hold. A look at the proof
shows that (i) is proved by making a good use of the paths P ii , called Kempe
chains after Kempe, who used them in his false proof of 1879, and (ii) follows
immediately from Euler's formula n - e + f = 2.
The attack on the four colour problem goes along similar lines. A con-
figuration is a connected cluster of vertices of a plane graph together with
the degrees of the vertices. A configuration is reducible if no minimal 5-
chromatic plane graph can contain it, and a set of configurations is un-
avoidable if every plane graph contains at least configuration belonging to
98 V Colouring

the set. In order to prove that every plane graph is 4-colourable, one sets out
to find an unavoidable set of reducible configurations. How should one show
that a configuration is reducible? Replace the cluster of vertices by a smaller
cluster, 4-colour the obtained smaller graph and use Kempe chains to show
that the 4-colouring can be "pulled back" to the original graph. How
should one show that a set of configurations is unavoidable? Make extensive
use of Euler's formula. Of course, one may always assume that the graph is a
maximal plane graph. Assigning a charge of 6-k to a vertex of degree k,
Euler's formula guarantees that the total charge is 12. Push charges around
the vertices, that is transfer some charge from a vertex to some of its neigh-
bours, until it transpires that the plane graph has to contain one of the con-
figurations.
Looking again at the five colour theorem, we see that the proof was based
on the fact that set of configurations of single vertices of degree at most 5 is
an unavoidable set of configurations (for the five colour theorem).
The simplistic sketch above does not indicate the difficulty of the actual
proof. In order to rectify this a little, we mention that Appel and Haken
had to find over 1900 reducible configurations to complete the proof.
Furthermore, we invite the reader to prove the following two simple
assertions.
1. The configurations in Figure V.6 are reducible.

+
Figure V.6. Three reducible configurations; in the second examples the outer vertices
may have arbitrary degrees.

2. Let G be a maximal planar graph of order at most 25 and minimum

degree 5. Call a vertex a major vertex if its degree is at least 7, otherwise
call it a minor. Then G contains one of the following:
(a) a minor vertex with 3 consecutive neighbours of degree 5,
(b) a vertex of degree 5 with minor neighbours only,
(c) a major vertex with at most one neighbour of degree at least 6.

EXERCISES

1:- Show that a graph G has at least e~G)) edges.

2. For each k :2: 3 find a bipartite graph with vertices Xl, X2, ... , Xn for which the
greedy algorithm uses k colours. Can it be done with n = 2k -' 2? Show that it
cannot be done with n = 2k - 3.
Exercises 99

r Given a graph G, order its vertices in such a way that the greedy algorithm uses
only k = X(G) colours.
4.- Let d I 2':: d z 2':: .. 2':: dn be the degree sequence of G. Show that in an order
XI'Xz, ... , Xn' d(x;) = dj, the greedy algorithm uses at most max min{d i + 1, i}
colours, and so if k is the maximal natural number for which k ~ dk + 1 then
X(G) ~ k.
5. Deduce from Exercise 4 that if G has n vertices then
x(G) + X(G) ~ n + 1.
6.- Show that X(G) + X(G) 2':: 2J~.

7. Let d, d l and d z be non-negative integers with d l + d z = d - 1. Prove that if

~(G) = d then the vertex set V(G) of G can be partitioned into two classes, say
V(G) = VI U Vz , such that the graphs Gi = G[V;] satisfy ~(Gi) ~ d;, i = 1, 2.
[Hint. Consider a partition V(G) = VI U Vz for which dle(G z ) + dze(G I ) is
minimal.]
8. (Exercise 7 continued) Let now d, d l , d z , ... , dr be non-negative integers with
2:; (d i + 1) = d + 1. Prove that if ~(G) = d then there is a partition V(G) =
U; V; such that the graphs Gi = G[V;] satisfy ~(Gi) ~ d;, i = 1,2, ... , r.
9. Given natural numbers rand t,.2r ~ t, the Kneser graph Kjr) is constructed as
follows. Its vertex set is T(r), the set of r-element subsets of T = {I, 2, ... , t}, and
two vertices are joined iff they are disjoint subsets of T. Figure V.7 shows K~Z), the
so called Petersen graph. Prove that X(Kjr) ~ t - 2r + 2, x(Kf) = 3 and
X(K~Z) = 4.

Figure V.7. The Petersen graph and the Grotzsch graph.

10. Check that the Grotzsch graph (Figure V.7) has girth 4 and chromatic number 4.
Show that there is no graph of order 10 with girth at least 4 and chromatic number
4.

11 + Try to construct a triangle free graph of chromatic number 1526 without looking
at Chapters VI or VII.
12.+ Let H be a graph of order n with edges eh ez, ... , em and chromatic polynomial
PH(X) = 2:7=0 (-I)icixn- i. Call a set of edges a broken cycle if it is obtained from
the edge set of a cycle by omitting the edge of highest inClex. Show that cj is the
number of .i-element subsets of the edges containing no broken cycle [Hint.
Imitate the proof of Theorem 4, choosing e l for abo Put G = H - e l so that
H = G' and PII(X) = PG(x) - PG ..(x). If an edge of G" comes from two edges of G,
say from xa = ei and xb := eh' label it with eb where k = max{i, h}; otherwise
keep the label it had in G (and H). Let F be a set of edges of H containing no
100 V Colouring

broken cycle of H. Note that if e l rt F then F is a set of edges of G containing

no broken cycle, and if e l E F then F - {e l } is a set of edges of G" containing no
broken cycle.]

13. Let PG(x) = Li'=o (-IYaixn-i be the chromatic polynomial of G. Prove that a2 =
(2) - k3(G), where m = e(G) and k 3(G) is the number of triangles in G.

14. Show that there is a unique graph Go of order n and size m = Ln 2 j4J such that if G
is also of order n and size m then

whenever x is sufficiently large.

15.- Find graphs G and H of order n and same size such that X(G) < X(H) but PG(x) <
PH(X) if x is sufficiently large.

16:+ Given a connected graph G containing at least one cycle, define a graph H on the
.set S of all spanning trees of G by joining TI to T2 iff IE(T1 )\E(T2) I = 1. (Cf.
simple transform of an x-path, p. 79). Imitate the proof of the fact that PH(X) is a
polynomial (Theorem 4) and the proof suggested in Exercise 12 +, to show that
H is not only Hamiltonian, but every edge of it is contained in a Hamiltonian
cycle.
17.+ We say that a graph G has a subgraph contractible to a graph H with vertex
set {Y!, ... , Yk} if G contains vertex disjoint connected subgraphs G" ... , Gk
such that, for i #- j, YiYj E E(H) iff G has a GcG j edge.
Prove that for every natural number P there is a minimal integer c(p) such that
every graph with chromatic number at least c(p) has a subgraph contractible
to KP. Show that c(l) = I, c(2) = 2 and c(n + 1) ~ 2c(n) - I for n:2: 2.
[Hint. Pick Xo E G and let Sk be the "sphere" of centre Xo and radius k; Sk =
{x E G: d(xo, x) = k}. Show that X(G) ~ X(Sk) + X(Sk+ I) for some k.]
18.+ Hadwiger's conjecture states that c(P) = P for every p. Prove this for P ~ 4.

19. Deduce from Kuratowski's theorem, Theorem 13 of Chapter I, that a graph is

planar iff it has no subgraph contractible to K 5 or K 3 3.

20. Show that the truth of Hadwiger's conjecture for P = 5 implies the four colour
theorem.
21.- Show that a map M = M(G) can be 2-coloured (see p. 96) iff every vertex of G
has even degree.
22:+ Suppose G is a cubic plane graph. Prove that the map M(G) is 3-colourable iff
each country has an even number of sides.

23:- Let M = M(G) be a triangular map, that is a map in which every country has
three sides. Show that M is 3-colourable unless G = K4.
24:- Prove that a map M = M(G) is 4-colourable if G has a Hamilton cycle.

25:- For each plane graph G construct a cubic plane graph H such that if M(H) is
is 4-colourable then so is M(G).
Exercises 101

26~ According to Tait's conjecture every 3-connected cubic plane graph has a Hamilton
cycle. (i) Show that Tait's conjecture implies the four colour theorem. (ii) By
examining the graph in Figure V.8 disprove Tait's conjecture.

Figure V.8. Tutte's counterexample to Tait's conjecture

27: Let G be a cubic plane graph. Show that G is 3-edge-colourable iff M(G) is 4-
colourable. [Hint. Let 1, a, band c be the elements of the Klein fourgroup C 2 x C 2 ,
so that a 2 = b 2 = c 2 = 1. Colour the edges with a, band c, and the ountries
with 1, a, b and c.]
28.- Find the edge chromatic number of K".
29. The cubic graph G has exactly one edge colouring with x'(G) colours. Show that
X'(G) = 3 and that G has exactly 3 Hamilton cycles.
30. Let n = 2P Show that K n + 1 is not the union of p bipartite graphs but K" is.
Deduce that if there are 2P + 1 points in the plane then some three of them deter-
mine an angle of size at least n(l - (lIp)).
31. Let X(G) = k. What is the minimal number of r-chromatic graphs whose union
is G?
32. Show that a k-chromatic graph can be oriented in such a way that a longest
directed path has k vertices.
33. Show that if a graph G can be oriented in such a way that no directed path contains
more than k vertices than X(G) :s; k. [Hint. Omit a minimal set of edges to destroy
all directed cycles. For a vertex x let c(x) be the maximal number of vertices on a
directed path in the new graph starting at x. Check that c is a proper colouring.]
34. Denote by cf/k the set of (isomorphism classes of) graphs of chromatic number
at least k. Let G 1 and G2 be vertex disjoint graphs in (f/k and let aibi be an edge
of Gi , i = 1,2. Let G = G 1 V G2 be obtained from G 1 U G2 by omitting the edges

Figure V.9. The Haj6s operation.

102 V Colouring

alb l , azb z , adding the edge blb z and identifying the vertices al and az. (The opera-
tion v is said to be the Hajos operation, see Figure V.9.) Prove that G E (fjk'
35.+ Let Yf'k be the smallest collection of (isomorphism classes of) graphs such that
(I) KkEYf'k' (2) if HE,;ifk and G::J H then GE.;ifb (3) if HE,;ifk and Gis
obtained from H by identifying two non-adjacent vertices, then G E Yf'k> (4) if
G I , G z E Yf'k and G = GI V G z then G E Yf'k' Prove that Yf'k = Ck' [Hint.
Exercise 33 implies that Yf'k C Ck' Assume that the converse inclusion is false and
let G E Ck\Yf'k be a counterexample of minimal order and maximal size. G cannot
be a complete q-partite graph so it contains vertices a, b l and b z such that
blb z E E(G) but ab l , ab z rt E(G). Let G I = G + ab l and G z = G + ab z . Then
G I and G z are not counterexamples so belong to .Yf k . Find out how G can be
obtained from copies of G I and G z by the allowed operations.]
36. Let k be a natural number. Prove that an infinite graph is k-colourable iff every
finite subgraph of it is. [Hint. Apply Tychonov's theorem as in Exercise 24 of
Chapter IlL]

Notes
Theorem 3 is in R. L. Brooks, On colouring the nodes of a network, Proc.
Cambridge Phil. Soc. 37 (1941) 194-197, and Vizing's theorem, Theorem 5, is
in V. G. Vizing, On an estimate ofthe chromatic class of a p-graph (in Russian)
Diskret. Analiz 3 (1964) 23-30. A detailed account of results concerning
colouring graphs on surfaces, culminating in Ringel and Youngs' proof of
Heawood's conjecture, can be found in Map Color Theorem, Grundlehren
der math. Wiss. 209, Springer-Verlag, Berlin, 1974. The use of the dis-
charging procedure in attacking the four colour problem is described in H.
Heesch, Untersuchungen zum Vierfarbenproblem, B-I-Hochschulskripten
810/81 Oa/81 Ob, Bibliographisches Institut, Mannheim, Vienna, Zurich,
1969. The proof of the four colour theorem appears in K. Appel and W.
Haken, Every planar map is four colourable, Part I: discharging, Illinois J.
of Math. 21 (1977) 429-490 and K. Appel, W. Haken and J. Koch, Every
planar map is four colourable, Part II: reducibility, Illinois J. of Math. 21
(1977) 491-567. A history of the four colour problem and a digest of its
proof are provided byT. L. Saaty and P. C. Kainen, The Four Color'problem,
Assaults and Conquest, McGraw-Hill, New York, 1977. The partition results
of Exercises 7 and 8 are in L. Lovasz, On Decomposition of Graphs, Studia Sci.
Math., Hungar.l (1966) 237-238.
The result in Exercise 33 is due to T. Gallai, On directed paths and circuits,
in Theory of Graphs (P. Erdos and G. Katona, eds.), Academic Press, New
York, 1968, 115-118, and that of Exercise 35 was proved by G. Haj6s,
Uber eine Konstruktion nicht n-farbbarer Graphen, Wiss. Zeitschr. Martin
Luther Univ. Halle-Wittenberg, Math.-Natur. Reihe 10 (1961) 116-117.
Colouring is a naturally appealing part of graph theory and the subject
has a vast literature. Many of the fundamental results are due to G. A. Dirac;
for these and other results see Chapter V. of B. Bollobas, Extremal Graph
Theory, Academic Press, London and New York, 1978.
CHAPTER VI
Ramsey Theory

Show that in a party of six people there is always a group of three who
either all know each other or are all strangers to each other. This well known
puzzle is a special case of a theorem proved by Ramsey in 1928. The theorem
has many deep extensions which are important not only in graph theory and
combinatorics but in set theory (logic) and analysis as well. In this chapter
we prove the original theorems of Ramsey, indicate some variations and
present some applications of the "results.

1 The Fundamental Ramsey Theorems

We shall consider partitions of the edges of graphs and hypergraphs. For

the sake of convenience a partition will be called a colouring, but one should
bear in mind that a colouring in this sense has nothing to do with the edge
colourings considered in Chapter V. Adjacent edges may have the same
colour and, indeed, our aim is to show that there are large subgraphs all of
whose edges have the same colour. In a 2-colouring we shall always choose
red and blue as colours; a subgraph is red (blue) if all its edges are red (blue).
Given a natural number s, there is an n(s) such that if n 2: n(s) then every
colouring of the edges of K n with red and blu~ contains either a red K S or a
blue K S In order to show this and to give a bound on n(s), we introduce the
following notation: R(s, t) is the minimum of n for which every red-blue
colouring of Kn yields a red K S or a blue K'. (We assume that s, t 2: 2, for we
adopt the reasonable convention that every Kl is both red and blue since
it has no edges.) A priori ifis not clear that R(s, t) is finite for every sand t.
103
104 VI Ramsey Theory

However, it is obvious that

R(s, t) = R(t, s) for every s, t ::?: 2

and
R(s, s) = R(2, s) = s,
since in a red-blue colouring of K S either there is a blue edge or else every
edge is red. The following result shows that R(s, t) is finite for every sand t,
and at the same time it gives a bound on R(s, t).

Theorem 1. If s > 2 and t > 2 then

R(s, t) ::; R(s - 1, t) + R(s, t - 1) (1)
and

s+t -
R(s, t) ::; ( s _ 1
2) . (2)

PROOF. (i) When proving (1) we may assume that R(s - 1, t) and R(s, t - 1)
are finite. Let n = R(s - 1, t) + R(s, t - 1) and consider a colouring of the
edges of K n with red and blue. Wd have to show that this colouring contains
either a red K S or a blue Kt. Let x be a vertex of Kn. Since d(x) = n - 1 =
R(s - 1, t) + R(s, t - 1) - 1, either there are at least nl = R(s - 1, t) red
edges incident with x or there are at least n2 = R(s, t - 1) blue edges incident
with x. By symmetry we may assume that the first case holds. Consider a
subgraph Knl of K n spanned by nl vertices joined to x by a red edge. If Knl
has a blue K t, we are home. Otherwise Knl contains a red K S - 1 which forms
a red K S with x.
(ii) Inequality (2) holds if s = 2 ot t = 2 (in fact, we have equality since
R(s,2) = R(2, s) = s). Assume now that s > 2, t > 2 and (2) holds for every
pair (s', t') with 2 ::; s', t' and s' + t' < s + t. Then by (1) we have

R(s, t) ::; R(s - 1, t) + R(s, t - 1)

::; (s +t-
s-2
3) + (s +s - l 3) = (s + 2).
t - t -
s-1
D

The result easily extends to co lou rings with arbitrarily (but finitely)
many colours: given k and SI' S2' ... ' Sk' if n is sufficiently large then every
colour of K n with k colours is such that for some i, 1 ::; i ::; k, it contains a
KSi coloured with the i-th colour. (The minimal value of n for which this
holds is usually denoted by R k(SI, ... , Sk).) Indeed, if we know this for k - 1
colours, then in a k-colouring of K n we replace the first two colours by a new
colour. Ifn is sufficiently large (depending on SI' S2, . , Sk) then either there
is a KSi coloured with the i-th colour for some i, 3 ::; i ::; k, or else there is a
1 The Fundamental Ramsey Theorems 105

Km, m = R(sl> S2) coloured with the new colour, that is coloured with the
first two (original) colours. In the first case we are home and in the second, for
i = 1 or 2 we can find a K S ' in K m coloured with the i-th colour.
In fact, Theorem 1 also extends to hypergraphs, that is to colourings of the
set x(r) ofall r-tuples of a finite set X with k colours. This is one of the theorems
proved by Ramsey.
Denote by R(r)(s, t) the minimum of n for which every red-blue colouring
of x(r) yields a red s-set or a blue t-set, provided IX I = n. Of course, a set
Y c X is called red (blue) if every element of y(r) is red (blue). Note that
R(s, t) = R(2)(S, t). As in the case of Theorem 1, the next result not only
guarantees that R(r)(s, t) is finite for all values of the parameters (which
is not at all obvious a priori), but also gives an upper bound on R(r)(s, t). The
proof is an almost exact replica of the proof of Theorem 1. Note that if r >
min{s, t} then R(r)(s, t) = min{s, t} and ifr = s.::; t then R(r)(s, t) = t.

Theorem 2. Let 1 < r < min{s, t}. Then R(r)(s, t) is finite and
R(r)(s, t) .::; R(r-1)(R(r)(s - 1, t), R(r)(s, t - 1 + 1.
PROOF. Both assertions follow if we prove the inequality under the assump-
tion that R(r-l)(u, v) is finite for all u, v, and R(r)(s - 1, t), R(r)(s, t - 1) are
also finite.
Let X be a set with R(r-l)(R(r)(s - 1, t), R(r)(s, t - 1 + 1 elements.
Given any red-blue colouring of x(r), pick an x E X and define a red-blue
colouring of the (r - I)-sets of Y = X - {x} by colouring (J E y(r-l) the
colour of {x} U (J E x(r). By the definition of the function R(r- 1)(u, v) we may
assume that Y has a red subset Z with R(r)(s - 1, t) elements.
Now let us look at the colouring of z(r). If it has a blue t-set, we are home,
since z(r) c x(r) so a blue t-set of Z is also a blue t-set of X. On the other
hand if there is no blue t-set of Z then there is a red (s - I)-set, and its union
with {x} is then a red s-set of X. 0

Very few of the non-trivial Ramsey numbers are known, even in the case
r = 2. It is easily seen that R(3, 3) = 6 and with some work one can show
that R(3, 4) = 9, R(3, 5) = 14, R(3,6) = 18, R(3, 7) = 23 and R(4,4)
= 18. Because of (1) any upper bound on an R(s, t) helps to give an upper
bound on every R(s', t'), s' ~ s, t' ~ t. Lower bounds for R(s, t) are not easy
to come by either. In Chapter VII we shall apply the method of random
graphs to obtain lower bounds on the Ramsey numbers R(s, t).
As a consequence of Theorem 2 we see that in every red-blue colouring of
the r-tuples of the natural numbers contains arbitrarily large monochromatic
subsets; a subset if monochromatic if its r-tuples have the same colour.
Ramsey proved that, in fact, we can find an infinite monochromatic set.

Theorem 3. Let c: A (r) -> {l, . ~ . , k} be a k-colouring ofthe r-tuples (1 .::; r < (0)
of an infinite set A. Then A contains a monochromatic infinite set.
106 VI Ramsey Theory

PROOF. We apply induction on r. Note that the result is trivial for r = 1 so

assume that r > 1 and the theorem holds for smaller values of r.
Put Ao = A and pick an element Xl E Ao. As in the proof of Theorem 2,
define a colouring c 1 : B~ - 1) -> {I, ... , k} of the (r - 1)-tuples of B 1 =
Ao - {xd by putting C1C,) = c(, u {xd), ,E B~-1). By the induction
hypothesis Bl contains an infinite set Ai all whose (r - I)-tuples have the
same colour, say d l (d l E {I, ... , k}). Let now X2 E Ai, Bl = Ai - {X2}
and define a k-colouring c2:B~-1)->{I, ... ,k} by putting C2(')=
c(, u {X2}), ,E B~-1). Then B2 has an infinite set A2 all of whose (r - 1)-
tuples have the same colour, say d 2 Continuing in this way we obtain an
infinite sequence of elements: Xb X2' ... , an infinite sequence of colours:
d l , d 2, ... , and an infinite nested sequence of sets: Ao ::J Ai ::J A2 ::J "',
such that Xi E A i - l and all r-tuples whose only element outside Au is Xi have
the same colour d i The infinite sequence (dn)~ must take at least one of the
k values 1, 2, ... , k infinitely often, say ni -> 00 and dni = 1 for every i. Then
by the construction each r-tuple of the infinite set {x n " x n2 ' } has colour 1.
o
In some cases it is more convenient to apply the following version of
Theorem 3. (N is the set of natural numbers.)

Theorem 3'. Let kr EN, r EN, and colour the set N(r) of r-tuples of N with kr
colours. Then there is an infinite set MeN such that for every r any two r-
tuples of M have the same colour, provided their minimal elements are at least r.

PROOF. Put Mo = N. Having chosen infinite sets Mo::J ... ::J M r - l , let
Mr be an infinite subset of M r - l such that all the r-tuples of Mr have the same
colour. This way we obtain an infinite nested sequence of infinite sets:
Mo ::J Ml ::J '" .Pickal E Mba2 E M2 - {I, ... , ad,a3 E M3 - {l, ... ,a2},
etc. Clearly M = {ab a2,"'} has the required properties. 0

Either Theorem 2 and the colour-grouping argument described after

Theorem 1, or Theorem 3 and a compactness argument imply the following.
Given rand Sl' S2,"" Sk' then for large enough IXI every colouring of x(r)
with k colours is such that for some i, 1 :::;; i :::;; k, there is a set Si c X,
ISil = Si' all of whose r-sets have colour i. The smallest value of IXI for
which this is true is denoted by Rt)(Sl' S2,"" Sk); thus R(r)(s, t) = R~)(s, t)
and Risl,S2, ... ,Sk) = Rf)(SbS2, ... ,Sk)' The upper bound for
Rt)(Sl' S2"'" Sk) implied by Theorem 2 is not very good. Imitating the
proof of Theorem 2 one arrives at a better upper bound (cf. Exercise 8):

Rt)(Sl"'" Sk) :::;; Rt-l)(Rt)(Sl - 1, S2' ... , Sk),,

Rt)(Sl' ... , Sk-l' Sk - 1 + 1.

2 Monochromatic Subgraphs 107

2 Monochromatic Subgraphs
Let H 1 and H 2 be arbitrary graphs. Given n, is it true that every red-blue
colouring of the edges of K n contains a red H 1 or a blue H 2? Since Hi is a
subgraph of KSi, where Si = IHd, the answer is clearly "yes" ifn ?: R(Sl, S2).
Denote by r(H 1, H 2) the smallest value of n that will ensure an affirmative
answer. Note that this notation is similar to the one introduced earlier:
R(s l ' S2) = r(K S" KS2). Clearly r(H 1, H 2) - 1 is the maximal value of n for
which there is a graph G of order n such that H 1 G and H 2 G.
The numbers r(H 1, H 2), sometimes called generalized Ramsey numbers,
have been investigated fairly extensively in recent years. We shall determine
r(H1' H 2) for some simple pairs (H1' H2).

Theorem 4. Let T be a tree of order t. Then r(K S , T) = (s - l)(t - 1) + 1.

PROOF. The graph (s - 1)K t - 1 does not contain T, its complement,
K s - 1(t - 1), does not contain K S , so r(K S , T)?: (s - l)(t - 1) + 1.
Let now G be a graph of order (s - l)(t - 1) + 1 whose complement does
not contain K S Then x( G) ?: rn/(s - 1)1 = t so it contains a critical sub-
graph H of minimal degree at least t - 1 (see Theorem 1 of Chapter V). It
is easily seen that H contains (a copy of) T. Indeed, we may assume that
T1 c H, where T1 = T - x and x is an end vertex of T, adjacent to a vertex
y of T1 (and of H). Since y has at least t - 1 neighbours in H, at least one of its
neighbours, say z, does not belong to T1 Then the subgraph of H spanned
by T1 and z clearly contains (a copy of) T.

As we know very little about r(KS, K t ), it is only to be expected that

r(G 1, G2) has been calculated mostly in the cases when both G1 and G2 are
sparse (have few edges compared to their orders), e.g., when G1 = sH 1
and G2 = tH 2. The following simple lemma shows that for fixed H 1 and
H 2 the function r(sH 1, tH 2) is at most s IH 1I + t IH 21 + c, where c depends
only on H1 and H 2 , and not on sand t,

Lemma 5. r(G,H 1 u H 2)::; max{r(G, H 1) + IH21, r(G, H2)}. In particular,

r(sH1' H 2) ::; r(Hi> H 2) + (s - 1)IH11.

PROOF. Let n be greater than the right hand side and suppose there is a red-
blue colouring of K n without a red G. Then n ?: r(G, H 2) implies that there
is a blue H 2 Remove it. Since n - IH21 ?: r(G, H 1 ), the remainder contains
a blue H 1. Hence K n contains a blue H 1 U H 2. 0

Theorem 6. If s ?: t ?: 1 then
r(sK2, tK2) = 2s +t - 1.
108 VI Ramsey Theory

PROOF. The graph G = K 2 s- 1 U EI - 1 does not contain s independent edges

and G = E 2s - 1 + K I - 1 does not contain t independent edges. Hence
r(sK2, tK2) ~ 2s + t - 1.
Trivially r(sK2, K2) = 2s. We shall show that

rs + 1)K 2, (t + 1)K2) ~ r(sK2, tK 2) + 3.

This is sufficient to complete the proof since then

rs + 1)K2, (t + 1)K2) ~ r(sK2, tK2) + 3 ~ rs - l)K2, (t - 1)K2) + 6

~ rs - t + 1)K 2, K2) + 3t = 2(s - t + 1) + 3t
= 2s + t + 2.
Let G be a graph of order n = r(sK2, tK 2) + 3 ~ 2s + t + 1. If G = K n then
G J (s + 1)K2 and if G = En then G J (t + 1)K2. Otherwise there are
three vertices, say x, y and z such that xy E G, xz ~ G. Now either G - {x, y, z}
contains s independent edges and then xy can be added to them to form
s + 1 independent edges of G or else G - {x, y, z} contains t independent
edges and then xz can be added to them to form t + 1 independent edges
~~ 0

Theorem 7. If s ~ t ~ 1 and s ~ 2 then r(sK 3, tK 3) = 3s + 2t.

PROOF. Let G = K 3s - 1 U (Kl + E 2r - 1 ). Then G does not contain s in-
dependent triangles and G = E 3s - 1 + (K 1 U K 21 - 1 ) does not contain t
independent triangles. Hence r(sK 3, tK 3) is at least as large as claimed.
It is easily shown that r(2K 3, K 3) = 8 and r(2K 3, 2K 3) = 10 (Exercise
12). Hence repeated applications of Lemma 5 give

r(sK 3, K 3) ~ 3s,

and to complete the proof it suffices to show that for s ~ 1, t ~ 1 we have

To see this let n = r(sK 3, tK 3) + 5 and consider a red-blue colouring of

Kn. Select a monochromatic (say red) triangle Kr in Kn. If K n - Kr contains
a red sK 3 then we are home. Otherwise K n - Kr contains a blue triangle Kb
(it even contains a blue tK 3 ). We may assume that at least five of the nine
Kr - Kb edges are red. At least two of these edges are incident with a vertex
of K b, and, together with an edge of Kr they form a red triangle K: meeting
K b. Since K n - K: - Kb has r(sK 3, tK 3) vertices, it contains either a red
sK 3 or a blue tK 3. These are disjoint from both K: and K b , so K n contains
either a red (s + 1)K 3 or a blue (t + 1)K3. 0

By elaborating the idea used in the proofs of the previous two theorems
we can obtain good bounds on r(sKP, tKq), provided max(s, t) is large
2 Monochromatic Subgraphs 109

compared to max(p, q). Let p, q ~ 2 be fixed and choose to such that

to min{p, q} ~ 2r(KP, Kq).

Theorem 8. If s ~ t ~ 1 then
ps + (q - 1)t - 1 ~ r(sKP, tKq) ~ ps + (q - 1)t + C.
PROOF. The graph KPS- 1 U E(q- 1)1- 1 shows the first inequality. As in the
proofs of the previous theorems, we fix s - t and apply induction on t. By
Lemma 5 we have
r(sKP, tKq) ~ (s - t)p + r(tKP, tKq) ~ ps + C,
provided t ~ to. Assume now that t ~ to, and the second inequality holds
for s, t.
Let G be a graph of order n = p(s + 1) + (q - 1)(t + 1) + C + 1 such
that G t> (s + 1)KP and G t> (t + 1)Kq. Suppose no KP of G and Kq of G
ha ve a vertex in common. Denote by ~ the set of vertices of G that are in KP
subgraphs and put Vq = V - Vp, np = I Vp I, nq = I Vq I. No vertex x E Vq is
joined to more than r(KP-I, Kq) vertices of Vp since otherwise either there is
a KP of G containing x or else there is a Kq of G containing vertices of Vp.
Similarly every vertex Y E Vq is joined to all but at most r(KP, Kq-I) vertices
of Vp. Hence
nqr(KP-l, Kq) + npr(KP, Kq-I) ~ npnq.
However, this is impossible, since np ~ sp, nq ~ tq so np > 2r(KP-I, Kq)
and nq > 2r(KP, Kq- I). Therefore we can find a KP of G and a Kq of G with
a vertex in common.
When we omit the p + q - 1 vertices of these subgraphs, we find that the
remainder H is such that H t> sKP and H t> tKq. However, IHI = ps
+ (q - 1)t + C + 1, so this is impossible. 0

One is inclined to imagine that the various Ramsey theorems hold for
finite graphs because the graph whose edges we colour is K n and not some
sparse graph with few edges.
Let us consider now another generalization of the original Ramsey
problem for finite graphs. Let H, GI , G z , ... , Gk be graphs. Denote by
H -+ (G I , Gz ,., Gk )
the following statement: for every colouring of the edges of H with colours
C I, C 2, ... , C k there is an index i such that H contains a subgraph isomorphic
to G; whose edges are of colour C;. If G; = G for each i then instead of
H -+ (G I , Gz , ... , Gk) we usually write H -+ (G)k. Note that
r(G I , G z ) = min{n: K n + I -+ (G I , Gz )} = max{n: K n + (G I , G z )}.
110 VI Ramsey Theory

Figure VI. 1. The graph C 3 + C 5

Clearly it is harder to find a monochromatic subgraph in a sparse graph than

in a complete graph and for given G1 and G2 it may not be easy to find a
sparse graph H (sparse in some given sense) such that H ..... (G 1, G2 ). Thus
one may not find immediately a graph H without a K6 for which H .....
(K 3 , K 3 ). (Such a graph is shown in Figure VI.1). Going even further, one
may ask the following difficult question: is there a graph H with cl(H) = 3
(so without a K4) such that H ..... (K 3, K3)? Note that Theorem 1 does not
guarantee the existence of such an H. To conclude the section we only state
a deep theorem of N esetfil and R odl showing that the answers to the questions
of this type are in the affirmative.

Theorem 9. Given graphs G b G2 , ... , G k there is a graph H such that cl(H)

= max1:$i:$k cl(GJ and H ..... (G b G2 , ... , G k ). 0

3 Ramsey Theorems in Algebra and Geometry

Given an algebraic or geometric object and a collection f!J of finite subsets
of the elements, is it true that whenever we colour the elements with k
colours (i.e., partition them into k classes), at least one of the colour classes
contains at least one member of f!J? In this section we discuss some questions
of this type for Euclidean spaces, finite vector spaces, various semi groups
and objects belonging to rather general categories. We prove only one or
two simple results, since the proofs of the deep results are well beyond the
scope of these notes, though the statements themselves are very easily under-
stood.
Let f!J be a collection of finite subsets of a set M. In accordance with the
notation used in the previous section, M ..... (f!J)k means that in every k-
colouring of M there is a monochromatic set PeJJ, i.e., if M = M 1 u U Mk
then P c Mi for some i E [1, k] = {l, ... , k} and P E f!J. When discussing
whether M ..... (f!J)k holds or not, the following compactness theorem enables
us to replace M by a finite subset of it.

Theorem 10. M ..... (f!J)k iff there is a finite set X c M such that X ..... (f!J)k.
3 Ramsey Theorems in Algebra and Geometry 111

PROOF. If M f+ (.9')k then clearly X f+ (gk for every finite set X.

A k-colouring of M is a point of the space [1, k]M; for C E [1, k]M and
x E M the projection of c into the x component, 1!x(c), is exactly c(x) E [1, k]
= {I, 2, ... , k}, the colour assigned by c to x. As almost always, we give
[1, k] the discrete topology and [1, k]M the product topology. By the familiar
Tychonov theorem from general topology, the space [1, k]M is compact
since it is a product of compact spaces.
Given a finite set X c M, let N(X) be the set of co lou rings in which X
does not contain a monochromatic set PEg>. If eFt N(X) and dE [1, k]M
agrees with c on X (that is d IX = c IX) then d Ft N(X). Therefore
[1, k]M - N(X) is open and so N(X) is closed. (In fact, N(X) is trivially
clopen.)
Let us turn now to the proof of the second implication. Assume that
X f+ (gk for every finite set X c M. This means exactly that N(X) #- 0
for every finite set X. Since N(X) n N(Y) ::J N(X u Y), the system
{N(X): Xc M is finite} of closed sets has the finite intersection property.
The compactness of [1, k]M implies that nx N(X) #- 0. Every c E nx N(X)
is a k-colouring of M without a monochromatic set PEg>. 0

Let L be a finite set of points in [Rm, the m-dimensional Euclidean space.

As another variant of the notation used so far, [Rn -+ (L)k means that in
every k-colouring of [Rn there is a monochromatic L' congruent to L. We say
that L is Ramsey if for every k there is an n such that [Rn -+ (L)k.

Theorem 11. Let P be a pair of points at distance 1 apart. Then

[R2 -+ (Ph but [R2 f+ (Ph.

PROOF. Figure VI.2 shows the first assertion. For suppose that in a red-blue-
yellow colouring of the seven points there is no monochromatic adjacent

y,

Figure VI.2. Adjacent points are at distance 1.

pair. We may assume that x is red. Then Yl, Zl are blue and yellow so Xl is
red. Similarly X2 is red, but Xl and X2 are adjacent.
Figure VI.3 shows that [Rf -+ (Ph is false. 0
112 VI Ramsey Theory

Figure VI.3. The hexagons have side a, 1 < a < (4)5 - 5)/10.

It is interesting that we do not know the maximal value of k for which

IR z --+ (P)k'

Theorem 12. Let QZ be the (vertex set of a) unit square. Then 1R6 --+ (Qzh.

PROOF. Consider a red-blue colouring of 1R6. Then, in particular, we have a

red-blue colouring of the following fifteen points of 1R 6 :xij = (Xi}"", X~),
1 :=; i < j :=; 6, X~j = 0 unless k = i or j and xL = xlj = 1/)2. Consider a
K6 with vertex set {VI"'" V6}, and colour edge viv/i < j) with the colour
of xij' Since r(C 4 , C 4 ) = 6 (the easy proof of this is left to the reader as
Exercise 13), by symmetry we may assume that the edges VI V z , V z V3' V3 V 4
are all red. It is easily checked that the points X12, XZ3, X34 and X l4 form a
unit square and this square is, of course, red. 0

Given LI C IRm 1 and L z C IRm>, we define LI x L z C IRm 1 +m 2 by LI X

L z = {(Xl> ... , Xm, + m ,): (Xl,, Xm1 )EL I , (X m1 +l>"" Xm, + m')EL Z}'

Theorem 13. If LI and L z are Ramsey then so is LI X Lz

PROOF. Given k, there is an m such that IRm --+ (Llh. Hence by the com-
pactness theorem there is aftnite subset X c IRm such that X --+ (LI)k' Put
I = klxl. Since L z is Ramsey, there is an n such that IRn --+ (LZ)I' Every k-
colouring of IRm +n = IRm x IRn restricts to a k-colouring c: X x IRn-+
[1, kJ = {l, 2, ... , k}. This gives an 1= k1xl-colouring of IRn with colours
[1, kY: simply colour Yo E IR" with the functionf E [1, kY that tells us how
the points of X x {Yo} are coloured: f(yo) = c(x, Yo). In this I-colouring
there is a monochromatic L z c IR", i.e., an L z c IRn such that for X E X,
and YI' Y2 E L z the points (x, YI), (x, Yz) have the same colour (in the original
k-colouring). Assigning this common colour to x, we obtain a k-colouring
of X. It has a monochromatic LI C X, providing us with a monochromatic
LI x L z . D
3 Ramsey Theorems in Algebra and Geometry 113

A brick in IRn is a set congruent to

where ai > 0. A unit simplex in IRk shows that IRk ---+ ({O, 1})k so every 2-point
set is Ramsey. Hence Theorem 13 has the following consequence.

Theorem 14. Every subset of a brick is Ramsey. o

Call a set L c IRm spherical if it is imbeddable in some sphere (of arbitrary
dimension and radius). It is somewhat more complicated to show that if a
set is not spherical then it is not Ramsey (Exercise 20). It is rather curious
that though there are many sets (and many very simple sets, e.g., an obtuse
triangle) that are spherical but not subsets of bricks, it is not known about
any of them whether they are Ramsey or not.

The following very significant extension of Ramsey's theorem was proved

by Hales and Jewett.

Theorem 15. Let S be a commutative infinite semigroup. Let L be afinite subset

of S and let [7/ = {a + nL: a E S, n EN}, where nL = {nx: x E L}. Then
S ---+ ([7/)k for every k. o
The proof is too difficult to present in these notes. However, we state this
theorem partly because it is the cornerstone of very general results proved
by Graham, Leeb and Rothschild, asserting that certain categories are
"Ramsey", and partly because we shall prove a special case of this theorem.
The next theorem was conjectured by Rota; it is one of the most striking
consequences of the results about Ramsey categories.

Theorem 16. Given a finite field F and natural numbers nand k, there is a
natural number N such that if F N, the N-dimensional vector space over F, is
coloured with k colours then FN contains a monochromatic n-dimensional
affine subspace.
Furthermore, given a finite field F and natural numbers n, m and k, there is
an M such that if the set of k-dimensional subspaces of FM is coloured with k
colours, then there is an n-dimensional subspace in which all k-dimensional
subspaces have the same colour. 0

The classical theorem of van der Waerden (Theorem 17 below) is a special

case of Theorem 15. Here we present it with a proof due to Graham and
Rothschild, since the proof gives back the flavour of the arguments used in
the deepest results of the theory. As customary, we write N for the set of
natural numbers and em, n] for {m, m + 1, ... , n}.
114 VI Ramsey Theory

Theorem17. Given p, kE N, there exists an integer W(p, k) such that in any

k-colouring of [1, W(p, k)] there is a monochromatic arithmetic progression
consisting of p terms.

PROOF. Let I and m be natural numbers. Two m-tuples (Xl> ... , x m ),

... , Ym) E [0, l](m) are said to be I-equivalent if they agree up to and in-
(y 1,
cluding the last occurrences of I, with the convention that any two sequences
not containing I are equivalent. We write S(l, m) for the following state-
ment:
Given kEN, there exists W(l, m, k) such that for every function

c: [1, W(/, m, k)] -> [1, k]

there exist a, d 1, .... , dm E N such that

a+l~dj::S;W(/,m,k) and c(a+~Xjdj)

is constant on each I-equivalence class of [0, I](m).
Note that van der Waerden's theorem is exactly the assertion S(p, 1),
but the theorem holding for every p is equivalent to the truth of S(I, m) for
every I and m. We shall prove S(l, m) by induction, starting with the trivial
assertion S(l, 1) and proving two different induction steps.
(i) If S(l, m) holds for some m ~ 1 then so does S(l, m + 1). To see this,
fixkandput W = Hr{1,m,k), W' = W(l, 1,kW)and Jtj = [(j - l)W + l,jW],
1 ::s; j ::s; W'. Let c: [1, WW'] -> [1, k] be any k-colouring. This induces a kW_
colouring c' of [1, W] that colours jE[l, W] with the functionf E[l, kJll. Wj
describing the colouring of Jtj: f(w) = cj - l)W + w), 1 ::s; w ::s; W (cf.
the proof of Theorem 3). By the choice of W', there exist a' and d', a' + Id' ::s;
W' such that c'(a') = c'(a' + d') = c'(a' + (I - l)d'). Let us apply S(l, m) to
the interval [(a' - l)W + 1, a'W], k-coloured with c. Then there exist
a, d n ... , dm E N such that
m
(a' - l)W + 1 ::s; a ::s; a + I L d j ::s; a'W
1

and c(a + LT xjd;) is constant on I-equivalence classes. Put d m + 1 = d'W.

Then a, d 1, . . . , dm + 1 are such that
m+l
a + I L dj::s; WW'
1

and

C
( a + t
m+ 1
xjd j
)

is constant on each I-equivalence class of [0, IJ<m +). (See Figure VI.4.)
4 Subsequences 115

ex a +
1"'1
L x,d. a + d'W
I 1
a + 2d'W
1 I

.
a' a'
.
+ d'
. .
a' + 2d' W'
Figure VIA. Illustration to part (i). The colour ofj E [1, W'] tells us how It} is coloured.

(ii) IIS(l, m) hoidsior every m then S(l + 1,1) also holds. This is an almost
immediate consequence of the pigeonhole principle. For let c:[W(l, k, k)] ~
[1, k] be given. Then there exist a, d 1 , ,dk EN such that a + I L1 d i :::;;
W(l, k, k) and c(a + L1 XidJ is constant on I-equivalence classes. The.
colours c(a + L1ldJ, s = 0, 1, ... , k cannot be all different so there exist
: :;
s, t, s < t :::;; k, such that

Then

is constant for j E [O,j].

Since S(l, 1) is trivial, by induction S(l, m) is true for every I and m. 0

Van der Waerden's theorem has many interesting and deep extensions.
Among others, Rado determined when a system
n

L
j= 1
aijxj = 0, i = 1, ... , m,

of linear equations with integer coefficients has a monochromatic solution

in every colouring of N with finitely many colours. As a particular case of
Rado's theorem and the compactness theorem one obtains the following
result. Given integers k and n, there exists N = N(k, n) such that if [1, N]
is k-coloured then there is a set A of n natural numbers such that a :::;; N LaEA
and all the sums LbEB b, 0 "# B c A have the same color. At the end of the
next section we discuss a beautiful extension of this to infinite sets.

4 Subsequences
Let Un) be a sequence of functions on a space T. Then we can find an in-
finite subsequence Un') such that one of the following two alternatives holds:
a. if Un) is any subsequence of (!",) then sup IL~ In; I ::::: l/Nforevery N ::::: 1,
b. ifUn)isanysubsequenceofUn,)thensupIL~ In;1 < l/N for some N ::::: L
116 VI Ramsey Theory

This rather difficult assertion about sequences of functions is, in fact, an

immediate consequence of a Ramsey type result about infinite sets.
As usual, 2M denotes the set of subsets of M, M(r) is the set of r-tuples of
M and M(W) denotes the set of countably infinite subsets of M. N is the set
of natural numbers and [1, n] = {I, 2, ... ,n}. A family fF c 2~ is called
Ramsey iff there exists an ME N(W) such that either M(W) c fF or M(W) c
2~ - fF.
Of course, 2~ can be identified with the cartesian product fliE
~ T", where
T" = {O, I}. We give T" the discrete topology and the product 2N the product
topology. In this topology 2~ is a compact Hausdorff space. A weak form
of a theorem due to Galvin and Prikry states that open subsets of 2~ are
Ramsey. (This is easily seen to imply the above assertion about sequences
of functions.) To prove this result it is convenient to use the notation and
terminology introduced by Galvin and Prikry. M, N, A and B are infinite
subsets of N. X, Yare finite subsets of N. X < a means that x < a for every
x EX; X < M means that X < m for every m EM. An M -extension of X is
a set of the form X u N where X < Nand N c M. Let us fix now a family
fF c 2W. We say that M accepts X if every M-extension of X belongs to fF;
M rejects X if no N c M accepts X.

Lemma 18. If N rejects 0 then there exists an ME N(W) that rejects every
XcM.
PROOF. Note first that there is an M o such that every X c M o is either
accepted or rejected by Mo. Indeed, put No = N, a o = 1. Suppose that we
have defined No=> N I=> .. => Nk and ai E Ni - N i+ IO :s; i :s; k - 1. Pick
akEN k If Nk - {ak} rejects {a o , ,a k} then put Nk+1 = Nk - {ak},
otherwise let N k+ I be an infinite subset of Nk - {a k} that accepts {ao, ... , ak}'
Then M o = {ao, a l , ... } will do.

By assumption M rejects 0. Suppose now that we have chosen
bo , bI> ... , b k - I such that M o rejects every Xc {b o , b l , , bk - I }. Then
M o can not accept infinitely many sets of the form Xu {c),j = 1,2, ... ,
since otherwise {CI, C2, . '} accepts X. Hence M o rejects all but finitely
many sets of the form Xu {c}. As there are only 2k choices for X, there

exists a b k such that M rejects every X c {b o , b I, ... , b k }. By construction
the set M = {b o , b l , ... } has required property. 0

Theorem 19. Every open subset of2 N is Ramsey.

PROOF. Let fF c 2~ be open and assume that A(w) fF for every A E N(W),
i.e., N rejects 0. Let M be the set whose existence is guaranteed by Lemma
18. If M(W) 2W - fF, let A E M(W) n fF. Since fF is open, it contains a
neighbourhood of A so there is an integer a E A such that if B n {l, 2, ... , a}
= A n {I, 2, ... ,a} then BE fF. But this implies that M accepts An
{I, 2, ... ,a}, contrary to the choice of M. Hence M(W) c 2 w - fF, proving
that fF is Ramsey. 0
4 Subsequences 117

Denote by XI<W) the family of finite subsets of X. A family fi'o c NI<w)

is dense if fi' 0 n M' < w) i= 0 for every ME N'W) and it is thin if no member
of fi' 0 is an initial segment of another member (that is if X < Y implies
X ~ fi' 0 or X u Y ~ fi' 0).

Corollary 20. Let fi' 0 c NI<W) be dense. Then there is an ME N'W) such that
every A c M has an initial segment belonging to fi' 0 .

PROOF. Let fi' = {F eN: F has an initial segment belonging to fi' o}. Then
fi' is open so there is an M E Nlw) such that either MIW) c fi', in which case
we are home, or else Mlw) c 2N - fi'. The second alternative cannot hold
since it would imply M'<W) n fi' 0 = 0. D

This corollary enables us to deduce an extension of the original Ramsey

theorem for infinite sets (Theorem 3).

Corollary 21. Let fi' 0 c N'<W) be a thin family. Then for any k-colouring of
fi' 0 there is an irifjnite set A c N such that all members of fi' 0 contained in A
have the same colour.

PROOF. It suffices to prove the result for k = 2. Consider a red and blue
colouring of fi' 0: fi' 0 = fi'red U fi' blue. If fi'red is dense then let M be the
set guaranteed by Corollary 20. For every FE fi' 0 n 2M there is an infinite
set N c M with initial segment F. Since fi' 0 is thin, F is the unique initial
segment of N that belongs to fi' o. Hence FE fi'red, so every member of fi' 0
contained in M is red.
On the other hand if fi'red is not dense then 2M n fi'red = 0 for some
infinite set M. Hence 2M n fi' 0 c fi' blue. 0

Let us now turn to the result concerning monochromatic sums we pro-

mised at the end of the previous section. This beautiful result, conjectured
by Graham and Rothschild and first proved by Hindman, is not very near
to the results given in this section, but the striking proof given by Glazer
illustrates the rich methods that can be applied in infinite Ramsey theory.

Theorem 22. For any k-colouring of N there is an infinite set A c N such

that all sums LXEX x, 0 i= X c A, have the same colour. D

We shall not give a detailed proof but only sketch one for those who are
(at least rather vaguely) familiar with ultrafilters on N and know that the
set f3N of all ultrafilters is a compact topological space (the Stone-Cech
compactification of N with, of course, the discrete topology). The proof
which is due to Glazer, is at least as beautiful as the theorem and it is con-
siderably more surprising.
118 VI Ramsey Theory

Let us recall that a filter :F on N is a collection of subsets of N such that

(i) if A, BE:F, then An BE:F, (ii) if A E:F and A c B then BE:F and
(iii) :F -# 2N , that is 0 :F. Zorn's lemma implies that every filter is con-
tained in a maximal filter, called ultrafilter. IfOlt is an ultrafilter then for every
A c N either A EOlt or else N - A EOlt. This implies that every ultrafilter
Olt defines a finitely additive 0-1 measure m on 2 N :

meA) = {Io if A EOlt,

if N - A EOlt.
Conversely, clearly every finitely additive 0-1 measure on 2N defines an
ultrafilter. If there is a finite set of measure 1 then one of the elements, say a,
of that set has also measure 1, and so Olt = {A eN: a E A}. These ultra-
filters are called principal. Not every ultrafilter is principal: the ultrafilters
containing the filter:F = {A c N: N - A is finite} are not principal.
That ultrafilters can be useful in proofs of Ramsey theorems can be seen
from the following very simple proof of the case r = 2 of Theorem 3. Let Olt
be a non-principal ultrafilter. Let N(2) = P t U P 2 U U Pk. For n EN let
Aln) = {m: (n, m) E PJ. Then exactly one of the sets Aln), A~n), ... ,Aln) be-
longs to Olt, say the set A~(~). Now with Bi = {n: c(n) = i} we have N =
B t u ... U Bk so again exactly one of these sets, say B j , belongs toOlt. Finally,
pick at E B j , a2 E B j n AjUd, a3 E B j n Aja d n AjU21, etc. With A = {at, a2, ... }
we have A(2) c Pj'
Let us turn at last to the sketch of Glazer's proof of Theorem 22. Let us
define an addition on f3N by
Olt + 'Y = {A eN: {n EN: A - n EOlt} E 'Y}

where Olt, 'Y E f3N and A - n = {a - n: a E A, a > n}.

With some effort one can check that Olt + 'Y is indeed an ultrafilter and
that with this addition f3N becomes a semigroup. Furthermore, the semi-
group operation is right-continuous, i.e. for a fixed flJ E f3N the map f3N .....
f3N, given by Olt ..... flJ + Olt, is continuous. By applying a short and standard
topological argument we see that the properties above imply that f3N has
an idempotent element, that is an element flJ with flJ + flJ = flJ. This flJ is
nonprincipal since if {p} E flJ then {2p} E flJ + flJ so {p} .OJ> + flJ.
Let now A E flJ. Then by the definition of the addition
A* = {nEN:A - nEP}EflJ.
Thus if a E An A* then B = (A - a) n (A - {a}) E ;J/. (We could replace
A by A - {a} since:-Y is not principal) Hence for every A E :J} there exist
a E A and B c A - {a} such that BE ?J and a + B c A.
Of course, this ultrafilter P has nothing to do with any colouring of N.
However, just as any non-principal ultrafilter enabled us to find a mono-
chromatic infinite set in a direct way, this idempotent ;J}i enables us to find
an appropriate infinite set. Let N = C t u ... U Ck be the decomposition
Exercises 119

of N into colour classes. Exactly one of these colour classes, say Ci , belongs
to !I. Put Al = Ci . Select al E Al and A2 E g>, A2 C Al - {ad such that
a l + A z c AI' Then select a2 E A2 and A3 E g>, A3 c A2 - {a2} such that
az + A3 c A 2 , etc. The set A = {aI, a2,"'} clearly has the required
property: every infinite sum LXE x X, X c A, has colour i. 0

Finally, it should be emphasised that the infinite Ramsey results presented

in this section form only the tip of an iceberg: the Ramsey theory of infinite
sets, called partition calculus, is an essential and very cultivated branch of
set theory, and it has a huge literature.
The foundations of partition calculus were laid down over twenty years
ago by Erdos and Rado, who also introduced the arrow notation, used
throughout the chapter, to formulate assertions about large cardinals.

EXERCISES

1. Show the equivalence of Theorem 3 and Rt)(sl"'" Sk) < 00,

(i) directly,
(ii) using Theorem 10.
2. Prove that R(3, 4) = 9.

Figure VI.5. A graph showing R(3, 4) Z 9.

3. Extending the construction in Figure VI.5, find for each t z 2 a t-regular graph
which shows that R(3, t + 1) z 3t.
4. By considering the graph with vertex set ;Z 17 (the integers modulo 17) in which i
is joined to j iff i - j = 1, 2, 4 or 8, show that R(4, 4) = 18.
5. Prove that R(3, 5) = 14.
6. Let e be an edge of K4. Show that r(K 4 - e, K4) = 11 (cf. Exercise 4).
7. By considering the 3-colouring of KI6 with vertex set GF(16), the field of order 16,
in which the colour of an edge ij depends on the coset of the cubic residues to which
i - j belongs, show that R3(3, 3, 3) = 17. (You have to check that the graph is
well defined.)
8. Establish the upper bound for Rt)(sJ, ... , Sk) given at the end of 1.
120 VI Ramsey Theory

9:- Given 2 :::;; k :::;; n, denote by ck(n) the maximal integer which is such that in every
k-colouring of the edges of K n we can find a connected monochromatic subgraph
of order ck(n). Show that cin) = n. (See Exercise 13 of Chapter 1.)
10:- Prove that Cn -I (n) = 2 if n ;?: 2 is even and Cn - I (n) = 3 if n ;?: 3 is odd. [Hint.
Use Theorem 9 of Chapter 1.J
11:- Prove that
if n == 2 (mod 4),

otherwise

12. Check that r(2K 3 , K 3 ) = 8 and r(2K\ 2K 3 ) = 10.

13. Show that r(C 4 , C 4 ) = 6.
14. By using the proof of Theorem 9 of Chapter IV prove that rk(C 4 , C 4 , ... , C4 ) :::;;

e + k + 2.
15.+ Using two-dimensional vector spaces over finite fields (cf. Theorem 10 of Chapter
IV), show that
rk(C 4 , C 4 , .. , C4 ) = k2 + O(k).
16. By considering HI = p5 and H 2 = K 1,3, show that

r(H I , H 2 );?: min r(Hi' Ha

i = 1. 2
need not hold.
17.+ Let H P' H q be graphs of order p and q, respectively and let cl(B p) = i, cl(Bq) = j.
Then there is a constant C depending only on p and q such that
ps + qt - min{si, tj} - 2 :::;; r(sH p, tHq) :::;; ps + qt - min{si, tj} + C.
[Hint. Find a red Ki(p-i), say R, a blue Ki(q- ii, say B, and a set N of ij other vertices
such that the RN edges are red and the BN edges are blue. Cf. the proof of Theorem
8.J
18. Prove that a set {xo, x I' ... ,xtl c [Rn is non-spherical iff for some Cj
{ {

I dXi - Xo) = 0 and I cllxd 2 - IXoI2) = b =I O.

I I

19. Let b, Clo' .. , CI E [R, b =I O. Show that there exists an integer k and some k-colouring
of [RI such that the equation
I

Idxj - xo) = b
I

has no solution with X o , XI' ... , XI all the same colour.

20. Prove that every Ramsey set is spherical. [Hint. Given a non-spherical set
{xo, XI" ... , XI} C [Rn, find (Cj) and b as in Exercise 18. Choose a k and a k-colouring
of [RI as in Exercise 19. Use this k-colouring to define a k-colouring of [Rn by
colouring X E [Rn with the colour of 1x 12.J
Notes 121

21. Let fen) be the minimal integer N such that whenever X is a set of N points
in a plane, no three of which are collinear, X contains n points forming a convex
n-gon. Show that fen) ~ R(4)(5, n) for every n ~ 4. Can you give a better bound
for f(5)?
22. Let S be an infinite set of points in the plane. Show that there is an infinite set A
such that either A is contained in a line or no three points of A are collinear.
23. Show that Rk(3, 3, ... , 3) ~ Lek!J + 1.
24. Deduce from Exercise 23 the following theorem of Schur: If we partition the
natural numbers not exceeding ek! into k classes then the equation x + y = z
is solvable in at least one of the classes.
25. Show that there is an infinite set of natural numbers such that the sum of any two
elements has an even number of prime factors.
26:' Show that there is a sequence nl < n2 < ... of natural numbers such that if
r ~ i l < i2 < ... < ir then 2:i= 1 nij has an even number of prime factors iff r
has an odd number of prime factors.
27. Define a graph with vertex set [1, N](2) by joining a < b to b < c. Show that this
graph does not contain a triangle and its chromatic number tends to infinity
with N. (See Exercise 11 of Chapter V.)
28. Let gl(X), g2(X), ... , gn(x) be bounded real functions and let f(x) be another
real function. Let e and 5 be positive constants. Suppose maXi (gi(X) - g;(y)) > 5
whenever f(x) - fey) > B. Prove that f is bounded.

Notes
The fundamental Ramsey theorems are in F. P. Ramsey, On a problem of
formal logic, Proc. Lond. Math. Soc. (2) 30 (1930) 264-286. Theorems 4,7
and 8 are due to V. Chvatal, P. Erdos and J. Spencer and can be found in
S. A. Burr, Generalized Ramsey theory for graphs-a survey, in Graphs
and Combinatorics (R. Bari and F. Harary, eds.) Springer-Verlag, 1974,
pp. 52-75. For some of the deep results of Nesetfil and Rodl, one of which is
Theorem 9, see J. Nesetril and V. Rodl, The Ramsey property for graphs
with forbidden subgraphs, J. Combinatorial Theory Ser. B, 20 (1976) 243-
249, and Partitions of finite relational and set systems, J. Combinatorial
Theory Ser. A 22 (1977) 289-312.
The geometric Ramsey results of 3 can be found in P. Erdos, R. L.
Graham, B. L. Rothschild, J. Spencer and E. G. Straus, Euclidean Ramsey
theorems, I, J. Combinatorial Theory Ser. A, 14 (1973) 344-363. Theorem 15
is in A. Hales and R. 1. Jewett, Regularity and positional games, Trans.
Amer. Math. Soc. 106 (1963) 222-229. Rota's conjecture is proved in R. L.
Graham and B. L. Rothschild, Ramsey's theorem for n-parameter sets,
Trans. Amer. Math. Soc. 159 (1971) 257-292. The results concerning Ramsey
122 VI Ramsey Theory

categories are given in R. L. Graham, K. Leeb and B. L. Rothschild, Ramsey's

theorem for a class of categories, Adv. in Math. 8 (1972) 417-433 (Errata,
ibid 10 (1973) 326-327). Van der Waerden's theorem is in B. L. van der
Waerden, Beweis einer Baudetschen Vermutung, Nieuw Archie! va or
Wiskunde 15 (1927) 212-216. The simple proof we give is from R. L. Graham
and B. L. Rothschild, A short proof of van der Waerden's theorem on
arithmetic progressions, Proc. Amer. Math. Soc. 42 (1974) 385-386.
Rado's theorem, mentioned at the end of 3 is from R. Rado, Studien zur
Kombinatorik, Math. ZeitschriJt 36 (1933) 424-480.
The first results of 4 are in F. Galvin and K. Prikry, Borel sets and Ram-
sey's theorem, J. Symbolic Logic 38 (1973) 193-198; Glazer's proof of
Hindman' theorem is in W. W. Comfort, Ultrafilters: some old and some
new results, Bull. Amer. Math. Soc. 83 (1977) 417-455.
Exercise 5 is in R. E. Greenwood and A. M. Gleason, Combinatorial
relations and chromatic graphs, Canad. J. Math. 7 (1955) 1-7, Exercise 9
in a remark of F. Galvin and Schur's theorem (Exercise 24) is in I. Schur,
-ober die Kongruenz x'" + ym == zm (mod p), Jahr. Deutsch. Math. Ver. 25
(1916) 114-116.
CHAPTER VII
Random Graphs

"Give a 'good' lower bound on the Ramsey number R(s, s), that is show
that there exists a graph of large order such that neither the graph nor its
complement contains a KS". "Show that for every natural number k there
is a k-chromatic graph which does not contain a triangle". It does not take
long to realize that the constructions that seem to be demanded by questions
like these are not easily come by. Later we show that for every k there is a
graph with the latter property, but even for k = 4 our graph has at least
232 vertices. This book does not contain a picture of such a graph. Indeed,
the aim of this chapter is to show that in order to solve these problems we
can use probabilistic methods to demonstrate the existence of the graphs
without actually constructing them. (It should be noted that we never use
more than the convenient language of probability theory, since all the
probabilistic arguments we need can be replaced by counting the number of
objects in various sets.) This phenomenon is not confined to graph theory
and combinatorics; in the last decade or two probabilistic methods have
been used with striking success in Fourier analysis, in the theory of function
spaces, in number theory, in the geometry of Banach spaces, etc. However,
there is no area where probabilistic (or counting) methods are more natural
than in combinatorics.
In most cases we use one of two closely related probabilistic models. We
start with a fixed set of n distinguished (labelled) vertices. Then either we
choose every edge with some probability p, 0 < p < 1, independently of
the choices of other edges or else we take all graphs of order n and size M,
and consider them as points of a probability space, having equal probability.
In the first case we shall write C(n, P(edge) = p) for the probability space,
in the second case we write c(n, M). Of course in a more complicated model
123
124 VII Random Graphs

we would choose the edges with given probabilities from several disjoint
subsets of edges.
In the first two sections we shall show that a random graph can be a
valuable weapon in attacking straightforward questions concerning graphs,
including the ones mentioned above. However, random graphs are clearly
of interest in themselves. A random graph can be viewed as an organism
that develops by acquiring more and more edges as p or M increases. It is
fascinating and rather surprising that a given property may appear rather
suddenly at a certain stage of the development. In the last two sections we
present some examples of this phenomenon.

1 Complete Subgraphs and Ramsey Numbers-the

Use of the Expectation
Let n, Nand M be natural numbers, 0::;; M ::;; N = (2). We shall consider
the set n = '(n, M) of all graphs of size M with vertex set V = {I, 2, ... , n}.
Clearly n has (z,) elements; in particular, if M =0 then n = {En} and if
M = N = (2) then n = {Kn}. (We use M for the number of edges in order
to emphasise that M tends to be of larger order than n. We cannot use e
since the base of the natural logarithm occurs in many of our formulae.)
For the sake of convenience we view n as a probability space, in which the
points (i.e., graphs) have equal probability, 1/1 n I. Then all graphic in-
variants occur as random variables on n, so we may talk of their expected
value, standard deviation, etc. In our calculations we shall often make use
of various estimates of binomial coefficients. For convenience we list them
now. First recall Stirling's formula

n! = (~rjhmea/(12n), (1)

where IX depends on n but is between 0 and 1. The following estimates, in

which 0 ::;; x ::;; x + y ::;; a and x ::;; b ::;; a, are obtained by expanding the

: ; e:r
binomial coefficient and occasionally applying Stirling's formula (1).

:;
2:ra (~r 2~ C~ br-b(~r (~) (a ~ br-b(~r:; :;
(2)

a- b- y)Y(~)x
(a-x-y a-x : ; (a -b-x
x- y)(a)-l
:; e:
b
br(~r :; e-(b/a)y-(l-(b/ax. (3)
I Complete Subgraphs and Ramsey Numbers-the Use of the Expectation 125

As our first illustration of the use of the language of probability in graph

theory we give a lower bound for the Ramsey number R(s, t). This lower
bound is based on the expected number of complete subgraphs of a given
order. For G E 0 denote by Xs = X.(G) the number of K S subgraphs in G.
Then X s is an integer valued random variable on the probability space O.

Theorem 1. The expected number of KS subgraphs contained in a graph

G E 0 = f(n, m) is

E(X.) = (n)s (NM -- (~))

(2) M
(N) -1.

PROOF. As we noted earlier, 101 = (ft). In order to calculate the expected

number of K S subgraphs in G E 0, first we compute the number of graphs
G E 0 that contain a fixed complete subgraph Ko of order s. If Ko c G then
m ofthe.edges of G are determined (i.e., are in K 0) and the remaining M - m
edges have to be chosen from a set of N - = (2) - m edges. Thus m
(N
M-m
-m)
of the graphs G EO contain Ko. As there are G) choices for K o , the expected
number of KS subgraphs contained in G E 0 is as claimed. 0

Theorem 2. If s, t :2: 3 then

R(s, t) > exp {

(S - 1)(t -
2(s + t)
1)}.
In particular
R(s, s) > e(I/4)(s-I)2/ s

PROOF. Note first that the graph G = (t - 1)Ks- 1 does not contain a K S
and its complement, K s - 1(t - 1),
does not contain a K 1, so R(s, t) :2:
(s - l)(t - 1) + 1. Therefore we may assume that

n = r{
exp
(S - 1)(t -
2(s + t)
1)}1 :2: (s - l)(t - 1) + 1,

so, in particular 8 ::;; s ::;; t.

Let

M= l_s
s+
NJ, t
M
I

=N-M<--N+l
+ s
t
t '

let Es be the expected number of K S subgraphs in a graph of order n and size

M, and, similarly, let EI be the expected number of KI subgraphs in the
126 VII Random Graphs

complement of such graphs (i.e., in the graphs G(n, M'. Let us estimate the
expression for E., given in Theorem 1, by using inequalities (2) and (3).
We obtain
t s(s - 1)
log Es ::; s(log n +1- log s) - - - 2
s+t
s(s - 1)
::; - +1- s(log s - 1) < - 3.
2(s + t)
Similarly
s t(t - 1) t(t - 1)
log Et ::; t(log n + 1 - log t) - - -
s+t
2 + nn-
( 1)

t(t - 1) t(t - 1)
= - +1- t(log t - 1) + < - 3.
2(s + t) n(n - 1)

Hence Es + Et < 1 so there is a graph G E Q which does not contain a K S

and whose complement does not contain a Kt. Hence R(s, t) ~ n, as claimed.
o
Clearly the bound given in Theorem 2 is very bad for s = 3 and large t.
In 2 we shall sketch a more refined version of the proof above, that gives a
much better bound for R(3, t). However, first we apply a straightforward
argument to obtain lower bounds in the problem of Zarankiewicz (cf.
Theorems 8 and 10 in Chapter IV).

Theorem 3. Let 2::; s ::; nl, 2::; t ::; n 2 , a = (s - l)/(st - 1) and f3 =

(t - 1)/(st - 1). Then there is a bipartite graph G 2(n!, nz) of size

l( l - ___
1 )nl-anl-pJ
s!t! ! z

that does not contain a K(s, t) (with s vertices in the first class and t vertices
in the second class).

PROOF. Let
n = nl + n2'
Vi = {l, 2, ... , nd,
Vz = {nl + 1, n l + 2, ... , n! + n2},
E = {ij: i E VI, j E Vz },
M = Lnl-an~-PJ.

We shall consider the probability space Q = ':(n, M; E) consisting of the

(It I) graphs with vertex set V = VI U Vz having exactly M edges from E
and none outside E. (Note that this is not the probability space considered
2 Girth and Chromatic Number-Altering a Random Graph 127

in the previous theorems.) The expected number of K(s, t) subgraphs con-

tained in a graph G E Q is

since the first factor is the number of ways the first class of K(s, t) can be
chosen, the second factor is the number of ways the second class can be
chosen and since K(s, t) has st edges. By (3) we have

Es,1 < 1 S I ( M )SI I

- - , - , n 1 n2 - - < -,-, n 1I-a n 2I-P
S.t. n 1n2 S.t.

Thus there is a graph Go E Q that contains fewer than (l/s!t !)n~ -an~ - p
complete bipartite graphs K(s, t). Omit one edge from each K(s, t) in Go.
The obtained graph G = Gin 1 , n 2 ) has at least

Ln 1- an 1- PJ - _1_nl-an1-P >
1 2 s!t! 1 2 -
l(1 - _1_)n 1 - an 1 -
s!t! 1 2
fJ J
edges and contains no K(s, t). D

By similar methods one can construct a graph of order n and size

l! (1 - _1_)n
2 s!t!
2 -(S+I-2)/(SI-l)J

that does not contain a K(s, t) (see Exercise 4).

It is a fact of life that the answers produced with the aid of random graphs
usually differ from the best possible ones by a factor of 2. For example,
random graphs give R(s, s) > 2s / 2 , whereas one would expect the actual
value of R(s, s) to be around 2S Similarly it is thought that there are graphs
of order n with cl n2 -(1/1) edges containing no K(t, t), where C I is a positive
constant depending only on t; random graphs show that CI n2 - (2/1) edges
are possible. However, in both these cases random graphs produce the best
known lower bounds.

2 Girth and Chromatic Number-Altering

a Random Graph
When looking for graphs with a certain property, we sometimes turn a set
of graphs into a probability space and hope that most graphs in that space
have the required property. If this does not happen, all is not lost, for we can
try to alter the graph slightly to make it have that property. We shall il-
lustrate this point with a particularly simple example. Throughout the
128 VII Random Graphs

section, we work with the probability space 0 = rg(n, M) of all graphs of

size M with a fixed set of n distinguished vertices.

Theorem 4. Given natural numbers b z 3 and g z 4 there is a graph of order

at most (2b)g whose minimum degree is at least b and whose girth is at least g.

ZIZI
PROOF. Let 0 = rg(n, M), where n = (2b)g and M = bn. Denote by ZlG)
the number of cycles of length I in a graph G; thus is a random variable
on the space o. What is E(ZI), the expected value of ? A cycle Co of length
I determines I edges of a graph G containing it; the remaining M - I edges
of G have to be chosen from a set of CD - I = N - I edges. Since there are
(';) choices for the vertex set of Co and, given the vertex set, there are 1<1 - I)!
ways of choosing Co,

I
E(ZI)=1(I-l)! I - I) (N)-I
(n) (NM-I M 1 I(M)I
~2zn N .

In the inequality above we applied (3) with a = N, b = M, x = I and y = o.

Summing over I we find that
g-I g-Il {b}1 g-I
1~3 E(ZI) ~ 1~3 21 nl: < 1~3 (2by < n.

Hence there is a graph G E G(n, bn) that contains at most n - 1 cycles of

length at most g - 1. Therefore, by omitting one edge from each cycle of
length at most 9 - 1, we obtain a graph H of order n and size M' > (b - 1)n.
We know from the simple Exercise 11 of Chapter IV that H contains a graph
with minimum degree at least b. D

The next result solves the second problem posed in the introduction of
this chapter.

Theorem 5. Given natural numbers k z4 and g z 4, there is a k-chromatic

e
graph of girth 9 (and order g + g).

PROOF. Let 0 = rg(n, M), where n = k 4g and M = k 3 n. Then, as in the pre-

vious proof, we have
g-I g-I 1
L E(ZI) < L (2k 3Y < (2k 3 )g < - n7/8.
1=3 1=3 9
Putf = and denote by 0 1 the set of graphs in 0 that contain at most
(2k 3 )g
3f cycles of length less than g. Then trivially (or by Chebyshev's inequality,
see 4), we have
P(OI) zl
Now we put p = (njk) + 1 = k 4g - I + 1 and estimate P(02)' the measure
of the set O 2 of graphs G that do not contain a p-set with at most 3f edges.
2 Girth and Chromatic Number-Altering a Random Graph 129

Clearly the graphs in O 2 are not (k - l)-colourable. Furthermore, even if

we omit any 3Joftheir edges, we still cannot colour them with k - 1 colours.
In order to estimate P(02) we calculate the expectation of 1;1 = 1;I(G),
the number of p-sets of vertices in which G has exactly I edges. There are (;)
ways of choosing a set of p vertices; there are

ways of choosing I edges joining vertices of such a set; and the remaining
M - I edges can be chosen in

(N-m)
M -I
ways. Thus

Applying (3) and summing over I we find that

L E(l P
3f +1 ):::; - -p2 )1exp { - M (p) (n)
L (en)p(e
3f -1}
1=0 1=0 P 2 2 2
< kPeP+3fp6fexp{ -kn}.
The logarithm of the right hand side is at most

n
n + - + n7/8 + !n 7/8 log n - kn < -tn.
3
Consequently
3f
L E(l;I) < e- n/ 2 < t
1=0

and so

Therefore
P(01 n O 2 ) ~ 1-
Every graph HE 0 1 n O 2 can be altered in a very simple way to provide
a graph with the required properties. Indeed, since H E 0 1 , it contains at
most 3J cycles of length less than g. Let F be obtained from H by omitting
3J edges of H, at least one from each cycle of length less than g. Then F has
girth g(F) ~ g. Since HE O 2 , this graph F does not contain p = (n/k) + 1
independent vertices and so X(F) ~ k. Let G be the disjoint union of a cycle
Cg and a k-chromatic subgraph of F. 0
l30 VII Random Graphs

In 1 we promised to sketch a proof of a lower bound for the Ramsey

number R(3, t), given by Erdos. The sketch will be rather rough since the
details of the proof are somewhat technical and cumbersome.

Theorem 6. There is a constant c > 0 such that

R(3, t) ~ cCo~ tY
for every t ~ 2. o
It suffices to show that there is a (large) positive constant A such that for
every n there is a graph Gn without triangles and without LAnl/2 log nJ in-
dependent vertices. Indeed, it is easily checked that if there is such a constant
then c = tA - 2 will do in the theorem.
In order to show the existence of A, we choose a constant B > 0 and con-
sider the probability space Q = '!len, M), where M = LB- 1 / 2n 3 / 2J Given
G E Q, let E* be a minimal set of edges whose removal destroys every triangle
in G. If H = G - E* contains a set W of p = LBnl/2 log nJ independent
vertices, then every edge of G[W] has to be contained in a triangle of G whose
third vertex is outside W One can show that if Band n are sufficiently large,
say B ~ Bo and n ~ no, then most graphs G E Q do not contain such a set W
Clearly A = Bo n6 has the required properties.

3 Simple Properties of Almost All Graphs- the

Basic Use of Probability
In this section we shall introduce and examine the model <g(n, P(edge) = p).
The graphs in this model are such that the edges are chosen independently
and with the same probability p,O < P < 1. Thus <g(n, P(edge) = p) con-
sists of all graphs with a fixed set of n distinguishable vertices, and the
probability of a graph with m edges is pmqN-m, where q = 1 - p and, as
before, N = (D. Thus q is the probability that a given pair of vertices are
not joined by an edge, and N is the maximum possible number of edges. We
saw in the preceding section how useful it is to know that most graphs in a
model have a certain property. Now we shall go a step further, namely we
shall discuss properties shared by almost all graphs. Given a property Q,
we shall say that almost every (a.e.) graph (in a probability space Q n con-
sisting of graphs of order n) has property Q if peG E Q n : G has Q) ~ 1 as
n ~ 00. In this section we shall always take Q n = '!len, P(edge) = p), where
o < p < 1 may depend on n.
Let us assume first that 0 < p < 1 is fixed, that is p is independent of n.
There are many simple properties holding for a.e. graph in <g(n, P(edge) =
p). For instance, if H is an arbitrary fixed graph, then a.e. G E <g(n, P(edge) = p)
contains H as a spanned subgraph. Indeed, if IH I = h, then the probability
3 Simple Properties of Almost all Graphs-the Basic Use of Probability 131

that the subgraph of G spanned by a given set of h vertices is isomorphic

to H is positive, say r > O. Since V(G) contains Ln/hJ disjoint subsets of h
vertices each, the probability that no spanned subgraph of G is isomorphic
to H is at most (1 - ryn/hJ, which tends to 0 as n --+ 00. The following result
is a strengthened version of this observation.

Theorem 7. Let 1 S h s k be fixed natural numbers and let 0 < p < 1 be

fixed also. Then in C(n, P(edge) = p) a.e. graph G is such that for every
sequence ofk vertices Xl' X 2 , .. , X k there exists a vertex x such that XXi E E(G)
if 1 sis hand xxd: E( G) if h < i s k.

PROOF. Let XI> x 2 , , Xk be an arbitrary sequence. The probability that a

vertex X E W = V(G) - {Xl' ... ,xd has the required properties is phqk-h.
Since for x, YEW, X =1= y, the edges XXi are chosen independently of the
edges yXi' the probability that no suitable vertex X can be found for this
particular sequence is (1 - phqk-h)"-k. Since there are n(n - 1) ... (n - k + 1)
choices for the sequence X I> X 2 , ,Xb the probability that there is a
sequence Xl, X 2 , , X k for which no suitable X can be found is at most
c = nk(1 _ phqk-h)"-k.

Clearly c --+ 0 as n --+ 00. D

By a result of Gaifman concerning first order sentences, Theorem 7 im-

plies that for a fixed 0 < p < 1 every first order sentence about graphs is
either true for a.e. graph in C(n, P( edge) = p) or is false for a.e. graph. Though
this result looks rather sophisticated, it is in fact weaker than the shallow
Theorem 7, for given any first order sentence, Theorem 7 enables us to
deduce immediately whether the sentence holds for a.e. graph or it is false
for a.e. graph. In particular, each of the following statements concerning
the model C(n, P(edge) = p) for a fixed p,O < P < 1, is an immediate con-
sequence of Theorem 7.
1. For a fixed integer k, a.e. graph is k-connected.
2. A.e. graph has diameter 2.
3. Given a graph H, a.e. graph G is such that whenever Foe G is isomorphic
to a subgraph F of H, there exists an Ho isomorphic to H satisfying
Fo c Ho c G.
Rather naturally, most statements we are interested in are not first order
sentences, since they concern large subsets of vertices. "For a given e > 0,
a.e. graph has at least -!(p - c)n 2 edges and at most -!(p + c)n 2 edges".
"Almost no graph can be coloured with n l / 2 colours". "A.e. graph contains
a complete graph of order (log n)/(log(l/p))". "Given c > 0, a.e. graph is
!(p - c)n-connected". These statements are all true for a fixed p and are
easily proved (see Exercises 8-11); however, none of them is a first order
sentence.
132 VII Random Graphs

Now we shall examine the model n = C(n, P(edge) = p) under the as-
sumption that 0 < p < 1 depends on n but pn 2 ~ 00 and (1 - p)n 2 ~ 00 as
n ~ 00. As before, we put N = mand for M = 0, 1, ... , N we denote by
n M the set of graphs in C(n, M). Clearly n = Uz.;o
nM' and the elements
ofn M have equal probability both in C(n, M) and C(n, P(edge) = p).
We shall show that the models n = C(n, P(edge) = p) and C(n, M) are
very close to each other, provided M is about pN, the expected number of
edges of a graph in n.
Writing P for the probability in n, we see that

Hence
p(n M ) M +1q
(4)
p(n M + 1) N - M p.

This shows that p(nN)/p(nM+ 1) increases with M, and p(nM) is maximal

for some M satisfying pN - 1 ~ M ~ pN + 1. Furthermore, if 0 < e < 1
and n is sufficiently large then, since pn 2 ~ 00 as n ~ 00,
p(n M )
p(nM + 1 ) < 1 - e

provided M < (1 - e)pN, and

p(nM + 1 ) I + )-1
p(n M) e

when M > (1 + e)pN. Putting Ne = L(l + e)pN J and N -e = r(1 - e)pNl,

we see from these inequalities that a.e. graph G in n satisfies N -e ~ e(G) ~
N" that is

pC~-& nM) ~ 1 as n ~ 00. (5)

Another consequence of (4) is that there is an YJ > 0 (in fact, any 0 < YJ < t
would do) such that

(6)

if n is sufficiently large. Now (5) and (6) imply that if n* c n is such that
p(n*) ~ 1 as n -+ 00, then for any e > 0 there are M 1 and M 2, such that
(1 - e)pN ~ M 1 ~ pN ~ M 2 ~ (1 + e)pN and
InM ( l n*1
, ~1 asn~oo (i = 1, 2,). (7)
InMil
4 Almost Determined Variables-the Use of the Variance 133

We call a set Q* c Q convex if G E Q* whenever G 1 c G C G2 and

G 1, G 2 E Q*; a convex property of graphs is defined analogously. It is easily
seen that for a convex set Q* relation (7) implies that

(7')

whenever M 1 ~ M ~ M 2, in particular, if M = LPNJ. Let us restate the

assertions above as a theorem about the connection between the models
~(n, P(edge) = p) and ~(n, M).

Theorem 8. Let 0 < P = p(n) < 1 be such that pn 2 -+ 00 and (1 - p)n 2 -+ 00

as n -+ 00, and let Q be a property of graphs.
(i) Suppose e > 0 is fixed and, if (1 - e)pN < M < (1 + e)pN, then a.e.
graph in ~(n, M) has Q. Then a.e. graph in ~(n, P(edge) = p) has Q.
(ii) IfQ is a convex property and a.e. graph in~(n, P(edge) = p) has Q, then
a.e. graph in ~(n, LPNJ) has Q. D

4 Almost Determined Variables-the Use of the

Variance

If X = X(G) is a non-negative variable on Q = ~(n, M) or Q =

~(n, P(edge) = p), and the expectation of X is at most a, then

t - 1
P(X ~ ta) ~ - - for t > 1.
t

Thus if the expectation of X is very small then X is small for most graphs.
This simple fact was used over and over again in the first two sections.
However, if we want to show that X is large or non-zero for almost every
graph in Q then the expected value itself very rarely can help us, so we have
to try a slightly more sophisticated attack. We usually turn to the variance
for help. Recall that if J1. = E(X) is the expectation of X then

is the variance of X and u = u(X) > 0 is the standard deviation. Chebyshev's

inequality, which is immediate from first principles, states that if t > 0 then
134 VII Random Graphs

In particular, if 0 < t < p then

a2
P(X = 0) ~ P(IX - pi;;:: t) ~ 2 '
t

so
a2
P(X = 0) ~ 2. (8)
J1.

In the examples we consider X = X(G) is always the number of spanned

subgraphs of G contained in a family !F = {F 1, F 2, ... }. Here the family
!F may depend on n and every graph FE !F has the same labelled vertex set
as G. Then clearly

E(X2) = L peG contains F' and F"), (9)

(F'.F")

where the summation is over all ordered pairs (F', F"), F', F" E!F.
In this section we shall consider the space Q = <;(n, P(edge) = p), where
o < p < 1 is fixed; we know that this space is near to <;(n, M), where M =
LPn2j2J. As before, we write q = 1 - p for the probability that two given
vertices are not adjacent. Furthermore, we use Landau's notation O(J(n))
for a term that, when divided by fen), remains bounded as n -+ 00. Similarly,
o(J(n)) denotes a term that, when divided by fen), tends to 0 as n -+ 00.
Thus 0(1) is a bounded term and 0(1) is a term tending to o.
Our aim is to make use of the variance to show that certain graph in-
variants are almost determined in our model <;(n, P(edge) = p). The first
theorem concerns the maximal degree. In its proof we shall make use of a
special case of the classical De Moivre-Laplace theorem concerning the
approximation of the binomial distribution by the normal distribution.
Let 0 < c = c(n) = 0(1) and put d(c) = LPn + c(pqn log n)1/2 J Suppose
x(c) = c(log n)1/2 -+ 00. Then

Ln (n) pkqn-k = (1 + 0(1)) -1 foo e- u2 / 2 du

k=d(e) k J2n x(e)

Furthermore

k

=dIe)
{(n)pkqn_k}2
k
= (1 + 0(1))(2npqn)-1/2 foo e-
x(e)
u2 du

= (1 + 0(1))(8npqc 2n log n)-1/2n -e 2

= o(n -e2- 1/2). (11)

4 Almost Determined Variables-the Use of the Variance 135

Theorem 9. IfO < P < 1 is fixed then in Q = C(n, P(edge) = p) the maximal
degree of almost every graph is

pn + (2pqn log n)l/Z + o(n log n)l/Z.

PROOF. Let c > 0 and denote by Xc = Xi G) the number of vertices of degree
at least d(c) = Lpn + c(pqn log n)l/Z J. Then, by (10),

and

+ n(n - 1) L:
n-l L:
n-l ( n- 2)( 2) n- pk,+k 2 -Z q Zn-Z-k,-k 2
k, =d(c) k2 =d(c) k, - 1 kz - 1
= I1c + (1 + 0(111;
In the estimation of E(X;) the term

is the probability that, given vertices a, bEG, a =I b, the vertex a is joined

to k 1 - 1 vertices in V (G) - {a, b} and b is joined to k z - 1 vertices in
V(G) - {a, b}.
Now if c = J2 - c, where 0 < c < 1, then I1c ;::: (1 + o(In E, so E(X;) =
(1 + 0(111; and (JZ(Xc) = 0(11;). Therefore, by (8),

P(Xc = 0) = 0(1).

Thus G almost surely has a vertex of degree at least

pn + (J2 - c)(pqn log n)l/Z.

On the other hand, if c = J2 + c, c > 0, then I1c = o(n- E) so

o
The result above has an interesting consequence.

Corollary 10. IfO < P < 1 and c > J(3/2) then in C(n, P(edge) = p) almost
no graph has two vertices of equal degree whose degrees are at least

d(c) == LIm + c(pqn log n)l/Z J.

136 VII Random Graphs

PROOF. The probability ofthe existence of two such vertices is at most

and by (11) this is 0(1). o

Combining Theorem 9 with Corollary 10 we see that a.e. graph has a
unique vertex with maximum degree.
In the next theorem we again consider the case when p varies with n.
We say that a function t(n) is a threshold/unction for a property Q of graphs
if

1. P
1m n p
(Q) = {O if p/t(n) ~ 0,
. (
n-oo' 1 If p/t n) ~ 00.

Here Pn.p(Q) denotes the probability that in ~(n, P(edge) = p) a graph has
property Q. The existence of a threshold function means that in the evolu-
tion of a random graph, i.e. in the process a graph acquires more and more
edges, the property emerges rather abruptly. In this section we prove a basic
result about threshold functions. The next section will be devoted to the
beautiful result that the property of being Hamiltonian has a sharp threshold
function.

Theorem 11. Let k ~ 2, k - 1 :::; I :::; m

and let F = G(k, l) be a graph with
average degree (21/k) at least as large as that 0/ any 0/ its subgraphs. Then
n- k/1is a threshold/unction/or F, that is ijpnk/l ~ 0 then in ~(n, P(edge) = p)
almost no graph contains F and ijpnk/l ~ 00 then almost every graph contains F.

PROOF. Let p = yn- k/1, 0 < y < nk/l, and denote by X = X(G) the number
of copies of F contained in G E ~(n, P(edge) = p). Denote by kF the number
of graphs with a fixed set of k labelled vertices that are isomorphic to F.
Clearly kF :::; k!. Then

fl = fly = E(X) = (~)kFP1(1 - py~)-l :::; nk(y1n- k) = yl,

so E(X) ~ 0 as y ~ 0, showing the first assertion.

Now let us estimate the variance of X when y is large. Note that there is a
constant c 1 > 0 such that
(12)

According to (9), we have to estimate the probability that G contains two

fixed copies of F, say F' and F". Put
As = I P(G contains F' and F"),
4 Almost Determined Variables-the Use of the Variance 137

where Ls
means that the summation is over all pairs (P', F") with s vertices
in common. Clearly

Furthermore, in a set of s vertices F' has t ~ (l/k)s edges. Hence, counting

first the choices for P' and then for F" with s 2 1 common vertices with P',
we find that for some constants C 2 and c 3 ,

As ~ L (k) (n - k)k!pl-tq(~)-l+t
/1 t -5, ls/k S k- s
~ L c2nk-S(yn-k/I)I-t
t -5, ls/k

Here in the last step we separated the term with t = 0 from the rest. Con-
sequently, making use of (12), we find that

for some constant C4' Therefore, by (8),

so P(X = 0) --+ 0 as y --+ 00. D

The most striking example of a graphic invariant being almost determined

in a random graph is that of the clique number, the maximal order of a com-
plete subgraph. It turns out that for a fixed p, 0 < p < 1, the clique number
of almost every graph in ~(n, P(edge) = p) is one of two possible values. In
fact, for most values of n (in a well defined sense) the clique number of almost
every graph is just a function of p and n. We shall confine ourselves to proving
a simple result in this direction. As in Theorem 1, denote by X = X(G) the
number of K' subgraphs in G E ~(n, P(edge) = p). Then

since we have G) choices for the vertex set of a K' and, having chosen the
vertex set, we have no choice for the m
edges. Let d = d(n, p) be the positive
real number for which
138 VII Random Graphs

For simplicity we put b = lip. It is easily checked with the aid of Stirling's
formula that

and

2 log n
d = log b + O(log log n). (6)

Theorem 12. Let 0 < p < 1 be fixed. Then the clique number of almost every
graph G E ~(n, P(edge) = p) is LdJ or rdl.

PROOF. The assertion is equivalent to the following:

P(X, > 0) --+ 0 if r z d + 1,

P(X, > 0) --+ 1 if r :0::; d - 1.

Now if r z d + 1, then by (6),

E(X,) = (n)p(2) :0::; ~ p'( n )p(r 21):O::; ~ p' --+ 0,

r r r-1 r

so the first assertion is clear.

Let now r :0::; d - 1. Let us use (9) to calculate the second moment of X"
summing separately over pairs of K' subgraphs with exactly I vertices in
common:

Since

Jl; = E(X,>2 = (n)r (n)I (nr --

1=0
r)p2(2)
I
= (n)2 p2<2J
r

with u, = u(X,) we have

,
:0::; I r2I n- 1b(l-I)I/2.
1=2
5 Hamilton Cycles-the Use of Graph Theoretic Tools 139

Recall now that bd / 2 < n so bO - 1 )1/2 < nl - 1 Hence

proving the second assertion. o

5 Hamilton Cycles-the Use of Graph Theoretic

Tools
In the proofs so far we always adopted a more or less head-on attack. We
hardly needed more from graph theory than the definitions of the concepts
involved, the emphasis was on the use of probability theory. This section is
devoted to a beautiful result of P6sa, the proof of which is based on a non-
trivial result in graph theory. Of course, in the ideal use of probabilistic
methods in graph theory we would have a mixture of all the ideas presented
in the four sections. Thus we would prepare the ground by using non-trivial
graph theoretic results and would apply probability theory to get informa-
tion about graphs in a probability space tailor made for the problem. We
could then select an appropriate graph which we would afterwards alter
with the aid of powerful graph theoretic tools.
It is easily seen that if c; > 0 then almost all graphs of order n and size
M = L(t + c;)n log nJ are connected and one can also show that almost no
graph of order n and size M = L(t - c;)n log nJ is connected (Exercise 13).
In particular, almost no graph of order n and size M = L(t - c;)n log nJ
contains a Hamilton cycle. It is fascinating that roughly the same number of
edges as are needed to ensure connectivity already ensure the existence of a
Hamilton cycle. Our next aim is to prove this beautiful theorem of P6sa.
The basis of the proof of this result is Theorem 15 of Chapter IV. Let S
be a longest xo-path in a graph H and write L for the set of end vertices of the
transforms of S. Denote by N the set of neighbours of vertices of L on Sand
put flA = V(H) - L uN. Then Theorem 15 of Chapter IV states that H has
no L - flA edge. All we shall need from this is that if IL I = 1 s IH 1/3 then
there are disjoint sets of size 1and IH I - 31 + 1, that are joined by no edge
of H.
For the sake of convenience we shall work with the space

clog n
~(n, P(edge) = p), p = ~-.
n

We start with a simple lemma in the vein of Theorem 4. Denote by Dt the

number of pairs (X, Y) of disjoint subsets of V such that IX I = t, I Y I =
n - 1 - 3t, and G has no X - Y edge.
140 VII Randon Graphs

Lemma 13. Let c > 3 and 0 < y < !- be constants and let p = (c log n)/n.
Then in ~(n, P(edge) = p) we have
P(D t > 0 for some t, 1 S t S yn) = O(n 3- C).

PROOF. Put f3 = (c - 3)/4c. Clearly

L: E(Dt) = L:
[ynl [ynl (
n) (
n - t ) (1 _ py(n-3t)
t; 1 t~ 1 t n - 3t

+ L:
[ynl
22n(1 - py(n-3t).
[f/nl+ 1

Now since (1 - p)n < n-C, we have

n 3(1 _ p)n-3 < (1 _ p)3 n3-C,
n3t (1 - py(n-3t) < nt(3-(c(n-3t)/n) S n 3- c if2 S t s f3n

and
22n(1 - py(n-3t) < n(2n/(logn))-n-3t)t/n) = O(n-f/(l-3 y)n) if f3n S t S yn.

Consequently
[ynl
L: E(D t) = O(n 3 - C ),

t; 1

implying the assertion of the lemma. o

Theorem 14. Let p = (c log n)/n and consider the space ~(n, P(edge) = p).
If c > 3 and x and yare arbitrary vertices, then almost every graph contains a
Hamilton pathfrom x to Y.1f c > 9 then almost every graph is Hamiltonian con-
nected: every pair of distinct vertices is joined by a Hamilton path.

PROOF. Choose y < !- in such a way that cy > 3 if c > 9 and cy > 1 if c > 3.
Let us introduce the following notation for certain events in
~(n, P(edge) = p), whose general element is denoted by G.

D = {D t = 0 for every t, 1 S t s LynJ},

E(W, x) = {G[W] has a path of maximal length, whose endvertex is
joined to x},
E(W, xlw) = {G[W] has a w-path of maximal length among the w-paths,
whose endvertex is joined to x},
F(x) = {every path of maximal length contains x},
H(W) = {G[W] has a Hamilton path},
H(x, y) = {G has a Hamilton x-y path},
He = {G is Hamiltonian connected}.
5 Hamilton Cycles-the Use of Graph Theoretic Tools 141

The complement of an event A is A.

Note that by Lemma 13 we have

P(D) = 1 - P(D) = O(nC- 3 ).

Let I WI = n - 2 or n - 1 and let us estimate P(D n E(W, x, where

x W. Let G E D n E(W, x) and consider a path S = XOX1 ... Xk of maximal
length in G[W]. (By introducing an ordering in W, we can easily achieve
that S is determined by G[W].) Let L = L(G[W]) be the set of endvertices
of the transforms of the xo-path S and let M be as in Theorem 15 of Chapter
IV (applied to G[W]). Recall that IM I ~ I WI + 1 - 31 L I and there is no
L - M edge, so no L - M u {x} edge either. Since G ED and 1M u {x} I ~
n - 31 L I, we find that IL I ~ yn. As L was determined independently of the
edges incident with x, we have

P(D n E(W, x :s:; peG E D and x is not joined to L(G[W]

:s:; (1 - p)Yn < n- cy .

Exactly the same proof implies that

P(D n E(W, xlw < n- CY

provided I WI = n - 2 or n - 1, WE Wand x W.
Note now that F(x) c E(V - {x}, x), so

P(H(V = p( UF(X) :s:; P(D

xeV
n UF(X) + P(D)
xeV

:s:; L P(D n F(x + P(D) :s:; nP(D n E(V - {x}, x + P(D)

:s:; n 1 -cy + O(n 3 -C).

This proves that if c > 3 then almost every graph has a Hamilton path.
Now let x and y be distinct vertices and put W = V - {x, y}. By the
first part
P(H(W:s:; 2n 1 - cy + O(n 3 - c).
Since
H(x, y) c H(W) n E(W, y) n E(W, xly),
we have

P(H(x, y :s:; P(H(W + P(D n E(W, y + P(D n E(W, xly) + P(D)

:s:; 2n 1-cy + 2n -cy + O(NC- 3).

Therefore, if c > 3 then alm<;>st every graph contains a Hamilton path from
x to y.
142 VII Random Graphs

Finally, as there are G) choices for an unordered pair (x, y), x "# y,
P(HC) ~ L P(H(x, y)) ~ n 3 - cy + O(n 2 - Cy ).
x=y

Thus if c > 9 then almost every graph is Hamiltonian connected. D

Since every Hamiltonian connected graph is Hamiltonian (contains a

Hamilton cycle), by Theorem 8 we have, in particular, that if c > 9 then
almost every graph in "(n, L(c log n)/nJ) is Hamiltonian. In fact, one can see
that the essential part of the proof gives this result with c > 6 as well. Inde-
pendently of P6sa, Korshunov proved a sharper and essentially best possible
form of Theorem 14: the assertion holds for every c > 1.

EXERCISES

1. Let G = (V, E) be a directed graph with m edges and without loops. Prove that V
can be partitioned into sets VI and V2 such that Gcontains more than m/4 edges
from VI to V2 .
2. Given an integer k :2: 2 and a graph G, denote by p(k)(G) the minimal number of
edges of G whose omission results in a k-partite graph. Prove that if G = G(n, m)
then

where n = kr + 1,0 :0;; 1 < k. Show furthermore that

k- 1
m - kp(k)(G) :2: - - m.
n- 1
[Hint. Consider all graphs of the form H = IK'+ I U (k - l)K' with vertex
set V(G). Note that the expected number of edges common to G and H is
e(G)e(H)/e(Kn).J
3. Prove that there is a tournament of order n (see Exercise 12 of Chapter I) that
contains at least n! 2 - n + I directed Hamiltonian paths.
4. Show that there is a graph of order n and size Lt(l - (l/s! t !n 2 -S+I- 2)(sl- IJ
that does not contain a K(s, t).
5. Given 2 :0;; s :0;; n, let d be the maximal integer for which there is a G 3 (n, n, n)
without a K3(S, s, s), in which every vertex is joined by at least d edges to each of
the other two classes. Prove a lower bound for d.
6. Use Theorem 3 to prove that if r > 2,0 < e < tcr - 1) - 2 and
2
d* > - :---:-c---,-:--;;-c-
, log(2(r - 1)2(;)
then for every sufficiently large n there is a graph G(n, m) not containing a K,(t),
where m :2: {(r - 2)/2(r - 1) + e}n 2 and t = Ld: log nJ. (Note that this shows
that Theorem 20 of Chapter IV is essentially the best possible.)
Exercises 143

7. Show that a given vertex has degree 1 in about 2/e 2 of the graphs in <'(n, n).
In Exercises 8-13 the model <'(n, P(edge) = p) is used and 0 < p < 1 is assumed
to be fixed. The same model is used in Exercises 14-19 but p varies with n as
indicated.
8 ~ Show that for I: > 0 a.e. graph has at least t(p - l:)n 2 edges and at most t(p + l:)n 2
edges.

9 ~ Give a lower bound on the chromatic number of a.e. graph.

10. Show that a.e. graph contains a complete graph of order at least log n/log(l/p).

11. Prove that a.e. graph G satisfies.

o(G) = A(G) = K(G) = pn - (2pqn log n)1/2 + o(n log n)1/2,

where q = 1 - p.

12. Estimate the maximal value of t for which a.e. graph contains a spanned K'" =
K(t, t). Estimate the corresponding value for Kr(t) = K(t, ... , t).

13. Let 0 < C < I. Prove that a.e. graph has the property that for every set Wof
k = Lc log2 nJ vertices there is a vertex X z for each subset Z of W such that Xz is
joined to each vertex in Z and to none in W - Z. Note that for c = 1 it is
impossible to find even a set of 2k vertices disjoint from W. [Hint. Refine the proof
of Theorem 7.]

14. Prove that (log n)/n is a sharp threshold function for connectivity, in the sense
that if e > 0 and p = 1 - e)log n)/n then almost no graph is connected while
if e > 0 and p = 1 + I:)log n)/n then a.e. graph is connected.

15:'" Sharpen the previous result as follows. If p = ((log n)/n) + (2x/n) then the prob-
ability that G is connected is e - e - 2 x. [Show first that a.e. G consists of a component
and isolated vertices.]

16. Let p = ((log n)/n) + (C(n)/n), where C(n) -> 00 arbitrarily slowly. Prove that
a.e. G contains a I-factor. [Use Tutte's theorem, Theorem 12 of Chapter III,
ignoring the parity of the components.]

17. Show that l/n is a threshold function for F 1 in Figure VII.1, that is if pn -> 0 then
almost no graph contains F 1 and if pn -> 00 then a.e. graph does.

18. What is the threshold function for F 2 in Figure VII. 1 ?

A
Figure Vn.1. The graphs F l' F 2 and F 3 .
144 VII Random Graphs

19. Let e > O. Prove that if p = n-(1/2)-e then almost no graph contains F 3 in Figure
VII.l but if p = 11-(1/2)+e then a.e. graph does. [Find a suitable graph Fj that has
average degree 2 + e.J
20:- Consider random directed graphs in which all edges are chosen independently
and with the same probability p. Prove that there is a constant c such that if
p = c((log n)ln)1/2 then a.e. directed graph contains a directed Hamilton cycle.
[Hint. What is the probability that a graph contains both edges ab and bQ?
Apply Theorem 14 to the random graph formed by the double edges.J
21:- Note that the suggested solution of Exercise 20 gives two directed Hamilton
cycles with the same underlying (non-directed) edge set. Show that with p =
(1 - e)((log n)ln)1/2 almost no directed graph contains such a pair of Hamilton
cycles.
22. Show that there are at least (2N In!) + O(2NIII !) non-isomorphic graphs of order n.
[Hint. Show that a.e. graph in '09'(11, P(edge) = 1) has trivial automorphism group;
for the automorphism group see p. 157.J

Notes
Perhaps the first combinatorial result proved by probabilistic methods is
the assertion of Exercise 3, proved by T. Szele in Combinatorial investi-
gations concerning directed complete graphs (in Hungarian), Mat. Fiz.
Lapok 50 (1943) 223-256; for a German translation see Kombinatorische
Untersuchungen tiber gerichtete vollstandige Graphen, Publ. Math. Debre-
cen 13 (1966) 145-168. However, the first papers that gave real impetus to
the use of random graphs are two papers of P. Erdos: Graph theory and
probability, Canad. J. Math. 11 (1959) 34-38, contains Theorem 6 and Graph
theory and probability II, Can ad. J. Math. 13 (1961) 346-352, contains the
lower bound on R(3, t) given in Theorem 6.
The result about first order sentences which we mentioned after Theorem
7 is due to R. Fagin, Probabilities on finite models, J. Symb. Logic 41 (1976)
50-58.
The fundamental paper on the growth of random graphs is P. Erdos and
A. Renyi, On the evolution of random graphs, Publ. Math. Inst. Hungar.
A cad. Sci. 5 (1960) 17-61. This paper contains a detailed discussion of
sparse random graphs, covering amongst other things the distribution oftheir
components and the occurrence of small subgraphs (Theorem 11).
The sharpest results in the direction of Theorem 12 are in B. Bollobas
and P. Erdos, Cliques in random graphs, Math. Proc. Cambridge Phil. Soc.
80 (1976) 419-427. P6sa's theorem (Theorem 14) is in L. P6sa, Discrete Math.
14 (1976) 359-364, its sharper form is in A. D. Korshunov, Solution of a
problem of Erdos and Renyi, on Hamilton cycles in non oriented graphs,
Soviet Mat. Doklady 17 (1976) 760-764.
For standard results of probability theory, including Chebyshev's in-
equality and the approximation of the binomial distribution by the Poisson
Notes 145

distribution the reader is referred to W. Feller, An Introduction to Proba-

bility Theory and Its Applications, Vol. 1 3rd ed., Wiley, New York, 1974,
and A. Renyi, Foundations of Probability, Holden-Day, San Francisco, 1970,
especially to the sections on the De Moivre-Laplace formula.
Random graphs and other combinatorial structures are investigated
in depth in the book by P. Erdos and J. Spencer, Probabilistic Methods in
Combinatorics, Academic Press, New York and London, 1974.
CHAPTER VIII
Graphs and Groups

In this chapter we examine some interactions between graphs and groups.

Some problems concerning groups are best attacked by using graphs.
Although in this context a graph is hardly more than a visual or computa-
tional aid, its use does make the presentation clearer and the problems more
manageable. The methods are useful both in theory and in practice: they help
us to prove general results about groups and particular results about in-
dividual groups. The first section, about Cayley and Schreier diagrams,
illustrates well both these aspects. It also contains an informal account of
group presentations.
If each member of a class of graphs is known to have particularly pleasant
symmetry properties, in other words if it has a large automorphism group,
then we may often make use of this to obtain additional information about
the class. In some cases this extra information enables us to determine the
class almost completely. The second section shows how we can do this
using the properties of the adjacency matrix.
Some classes of labelled graphs are easily enumerated, as we shall see in
3. Other enumeration problems, such as counting isomorphism classes
of graphs, lead us to the study of orbits of permutation groups. The high-
light of the section is the fundamental theorem enumerating such orbits,
namely Polya's theorem.

1 Cayley and Schreier Diagrams

Let A be a group generated by a, b, .... The graph of A, also called its Cayley
diagram, with respect to these generators is a directed multigraph whose
edges are coloured with the generators: there is an edge from x to y coloured

146
l Cayley and Schreier Diagrams 147

Figure VIII.1. The Cayley diagrams of (i) the cyclic group C 4 generated by a, (ii) the
Klein four-group with generators a, b and (iii) the symmetric group S 3 with generators
a = (123) and b = (12).

with a generator 9 iff xg = y. To illustrate this concept in Figure VII 1.1 ,

we show the Cayley diagrams of three small groups.
A Cayley diagram is regular, and so is its colouring, in the following
sense: for each vertex x and each generator (colour) 9 there is exactly one
edge of colour 9 starting at x and exactly one edge of colour 9 ending at x.
Furthermore, at most one edge goes from x to another vertex y. If we know
the Cayley diagram of a group then we can easily answer questions in terms
of the generators. What is the element aba 2 b in S 3 ? It is the end of the directed
walk starting at 1 whose first edge has colour a, the second has colour b,
the third a, the fourth a and, finally, the fifth b. By following this walk in
the third picture in Figure 1, we find that aba 2 b = a 2 In general, two ele-
ments expressed as products of some generators are equal iff the correspond-
ing walks starting at 1 end at the same vertex.
The Schreier diagram is a slight extension of the Cayley diagram. This
time we have a group A, a set S of elements of A and a subgroup B of A.
The Schreier diagram of A mod B describes the effect of the elements of S on
the right cosets of B: it is a directed multigraph whose vertices are the right
co sets of B, in which an edge of colour s E S goes from a coset H to a coset K
iff Hs = K. (Thus a Cayley diagram is a Schreier diagram mod B = {I}.)
In most cases S is chosen to be a set of generators or a set which together
with B generates A. Instead of giving a separate illustration, note that if A
is the symmetric group on p, 2, 3, 4}, B is the subgroup of elements fixing
1 and S = {a} with a = (1234) then the Schreier diagram is exactly the first
picture in Figure VIII.1, that is the Cayley diagram of C 4 . Once again we
note that for each vertex H and each colour 9 E S exactly one edge coloured
9 starts at H and exactly one edge coloured 9 ends at H. However, some of
the edges may be loops, that is they may start amd end at the same coset.
Furthermore, there may be many edges of different colours joining two
vertices.
Group diagrams do not tell us anything about groups that cannot be
expressed algebraically. However, the disadvantage of the algebraic ap-
proach is that many lines of print are needed to express what is conveyed
almost instantaneously by a 'single diagram. These diagrams are especially
148 VIII Graphs and Groups

helpful when we have a concrete problem to be solved by hand, and more-

over the purely mechanical techniques involved are ideal for direct use on a
computer. Since the advent of fast electronic computers many otherwise
hopeless problems have been solved in this way.
Group diagrams are particularly useful when attacking problems con-
cerning groups given by means of their presentations. For the convenience
of the reader we recall the basic facts about group presentations. We aim
throughout for an intuitive description, rather than a rigorous treatment;
the interested reader may fill in the details himself or turn to some specialist
books on the subject. A word W in the symbols a, b, c, ... is a finite sequence
such as ba-lccaa- 1 b- 1a; the empty sequence is denoted by 1. We call two
words equivalent if one can be obtained from the other by repeatedly re-
placing xx - 1 or X-I X by 1 (the empty word) or vice versa. Thus abb - 1 a-I c - 1
and CC-1C-1dd- 1 are both equivalent to c- 1 In fact, we shall use the same
notation for a word and its equivalence class and so we write simply
abb-Ia-Ic- 1 = cc-Ic-1dd- 1 = c- I . Furthermore, for simplicity
abbc-Ic-Ic- I = ab 2 c- 3 , etc. The (equivalence classes of) words form a
group if multiplication is defined as juxtaposition: (ab-Ic)(c-Iba) =
ab-Icc-Iba = a 2 Clearly a-I is the inverse of a and (a-Ib-Ic)-I = c-Iba.
This group is the free group generated by a, b, c, ... and it is denoted by
<a, b, c, .. . ).
Let RI" Rv> ... be words in the symbols a, b, c, ... , let F = <a, b, c, ... )
and let K be the normal subgroup of F generated by RI" R v , . Then the
quotient group A = F/K is said to be the group generated by a, b, c, ... and
the relators RI" R v , " ' ; in notation A = <a, b, C, .. IRI" R v ," .).
Once again we use a word to denote its equivalence class and write
equality to express equivalence. More often than not, a group presentation
is written with defining relations instead of the more pedantic relators. Thus
<a, bla 2 = b 3 ) denotes the group <a, bla 2 b- 3 ). A group isfinitely presented
if in its presentation there are finitely many generators and relations. It is
easily seen that two words WI and W2 are equivalent in A iff W2 can be ob-
tained from WI by repeated insertions or deletions of aa - I, a - I a, bb - I, ... ,
the relators RI" R., ... and their inverses R;: 1, R; 1, . . . . As an example,
note that in A = <a, bla 3 b, b3, a4 ) we have a = aa 3 b = a 4 b = b. Hence
1 = a 3 b(b 3 )-1 = a = b and so A is the trivial group of order 1.
Even the trivial example above illustrates our difficulties when faced with
a group given in terms of defining relations. However, groups defined by
their presentations arise naturally in diverse areas of mathematics, es-
pecially in knot theory, topology and geometry, so we have to try to over-
come the difficulties. The fundamental problems concerning group presenta-
tions were formulated by Max Dehn in 1911. These problems ask for general
and effective methods for deciding in a finite number of steps (i) whether two
given words represent the same group element or not (word problem), (ii)
whether they represent conjugate elements (conjugacy problem) and (iii)
whether two finitely presented groups are isomorphic or not (isomorphism
I Cayley and Schreier Diagrams 149

problem). All these problems have turned out to be problems in logic, and
cannot be solved in general. Explicit solutions of these problems are always
based on specific presentations and often make use of group diagrams.
(Dehn himself was a particularly enthusiastic advocate of group diagrams.)
Let A = <a, b, ... IR Jl , R" .. .). We shall attempt to construct the Cayley
diagram of A with respect to the generators a, b, . ... Having got the Cayley
diagram, we clearly have a solution to the word problem for this presentation.
The Cayley diagram has the following two properties. (a) The (directed)
edges have a regular colouring with a, b, ... , that is for each vertex x and
generator g there is exactly one edge coloured g starting at x and exactly
one edge coloured g ending at x. (b) Every relation is satisfied at every
vertex, that is if x is a vertex and Rv is a relator then the walk starting at x
corresponding to Rv ends at x. How shall we go about finding the Cayley
diagram? We try to satisfy (a) and (b) without ever identifying two vertices
unless we are forced to do so. Thus at each stage we are free to take a new
vertex and an edge into it (or from it). We identify two vertices when either
(b) or (a) force us to do so. If the process stops, we have arrived at the Cayley
diagram. Note that until the end we do not know that distinct vertices rep-
resent distinct group elements.
As an example, let us see how we can find the Cayley diagram of A =
<a, bla 3 = b2 = (ab)2 = I). We replace each double edge corresponding to
b by a single undirected edge; this makes the Cayley diagram into a graph.
We start with the identity and with a triangle 123 corresponding to a 3 = 1;
for simplicity we use numbers 1,2, ... to denote the vertices, reserving 1 for
the identity element. An edge coloured b must start at each of the vertices
1, 2 and 3, giving vertices 4, 5 and 6. Now a 3 = 1 must be satisfied at 6,
giving another triangle, say 678, whose edges are coloured a, as in Figure
VIII.2. At this stage we may care to bring in the relation (ab)2 = abab = 1.
Checking itat 8, say, we see that the walk 86314 must end at 8, so the vertices
so far denoted by 8 and 4 have to coincide. Next we check abab = 1 at 7:
the walk 7(8 == 4)/125 must end at 7 so 5 and 7 are identical. All that remains
to check is that the diagram we obtained satisfies (a) and (b), so it is the
Cayley diagram of the group in question. In fact, the diagram is exactly the
third picture in Figure VIII.1, so the group is S3.
For p ;::: q ;::: r;::: 2 denote by (p, q, r) the group <a, b, claP = b q = cr =
abc = I). Given specific values of p, q and r, with a little effort the reader

I
--a
: -----b
I
I
I

~------2~------~,
Figure VIII.2 .. Construction of a Cayley diagram.
150 VIII Graphs and Groups

--a
---- b
........ c

a4 = b = c J = abc = I
Figure VIII.3. Some Cayley diagrams. The shaded regions correspond to abc = 1.

can find the Cayley diagram of the group (p, q, r) with respect to the genera-
tors a, band c. Figure VIII.3 shows some of these diagrams.
The diagrams above indicate some connection with tessellations. The
beauty of the use of Cayley diagrams is that one can make good use of this
connection. Indeed, the reader who is slightly familiar with tessellations of
the sphere, the Euclidean plane and the hyperbolic plane, can easily prove
the following result.

Theorem 1. If (l/p) + (l/q) + (l/r) < 1 then the group (p, q, r) is finite and
has order 2s where l/s = (l/p) + (l/q) + (l/r) - 1. The Cayley diagram is
a tessellation of the sphere (as in the first two pictures in Figure VIII.3).
If (l/p) + (l/q) + (1/r) ::;; 1 then the group (p, q, r) is infinite. If equality
holds, the Cayley diagram is a tessellation of the Euclidean plane, otherwise
it is a tessellation of the hyperbolic plane (as in the last two pictures in Figure
VIIl.3).

As we remarked earlier, groups given by means of their presentations

arise frequently in knot theory. In particular, Dehn showed how a presenta-
tion of the group of a (tame) knot (that is the fundamental group of [R3
after the removal of the knot) can be read off from a projection of the knot
l Cayley and Schreier Diagrams 151

Figure VIII.4. The trefoil and the figure of eight.

into a plane. The projection of the knot forms .the boundary of certain
bounded domains of the plane. To each of these domains there corresponds
a generator (the identity corresponds to the unbounded domain) and to
each cross-over there corresponds a relator. The general form of these
relators is easily deduced from the two examples shown in Figure VIll.4.
(Indeed, readers familiar with the fundamentals of algebraic topology can
easily prove the correctness of this presentation.) The group of the trefoil
(or clover leaf) knot is <a, b, e, d Iad - 1b, cd - I a, ebr I ) and the group of the
figure of eight knot is <a, b, e, d, e Iab -Ie, ad-Ieb -1, ed- 1 eb -1, aed- l ).
Of course, before embarking on an investigation of the group, it is sensible
to attempt to simplify the presentation. For example, ebd- 1 = 1 means that
d = eb or br Ie = 1. Thus the group of the trefoil knot is
<a,b,e,dlad- 1 b,bd- Ie,ed- 1 a)
or, equivalently,
<a, b, eleb = ba = ae).
We invite the reader to check that the Cayley diagram of this group is made
up of replicas of the ladder shown in Figure VIll.5. (Exercise 4). At each
edge three ladders are glued together in such a way that when looking at
these ladders from above, we see an infinite cubic tree (Figure VIIl.5).
a

b d
d
c

d
a

c d
d

Figure VIII.5. The ingn::dients of the Cayley diagram of the trefoil knot.
152 VIII Graphs and Groups

Having obtained the Cayley diagram, we can read off the properties we are
interested in. In the case of this group the method does not happen to be
too economical, but this is the way Dehn proved in 1910 that the group of
the trefoil knot is not the group of a circle, which is infinite cyclic.
Schreier diagrams can be constructed analogously to Cayley diagrams.
In fact, in order to determine the structure of a largish group given in a
presentation it is often more advantageous to determine first the Schreier
diagram of a subgroup. In order to show this, we work through another
example, once again due to Dehn. What is the group
A = <a, bla 2 = b S = (ba)3 = I)?
Let us construct the Schreier diagram of the cosets of the subgroup B gene-
rated by b. As before, we take a vertex 1 for B and try to construct as big a
diagram as conditions (a) and (b) allow us. (Recall that (a) requires the
colouring to be regular and (b) requires that each defining relation is satisfied
at each vertex.) However, in this case there is one more condition: the edge
coloured b starting at 1 must end in 1 (so it is a loop) since Bb = B. Thus
after two steps we have the diagram shown in Figure VIII.6. (Once again the
edges coloured a will not be directed since a 2 = 1.)

--a

,i ---- b
/-~

r
\
\ I
\ I
\,1

Figure VIII.6. The initial part of the Schreier diagram of A mod B.

Now let us check the condition bababa = 1 at vertex 6. The walk babab
takes us from 6 to 3, so there must be an edge coloured a from 3 to 6. In
order to have edges coloured a starting at 4 and 5, we take up new vertices
7 and 8, together with edges 47 and 58 coloured a. Next we check the con-
dition ab-1ab-1ab- 1 = 1, which is equivalent to (ba)3 = 1 at 7, and find
that there is an edge from 7 to 8 coloured b. To satisfy b S = 1 at 7 we take
three new vertices, 9, 10 and 11. Checking (ba)3 = 1 at 11 we find that there
is an edge from 9 to 11 coloured a. At this stage we are almost home, but no
edge coloured a begins at 10, so we take a new vertex 12 joined to 10 by an
edge coloured a. What does the condition ab - 1 ab - 1 ab - 1 = 1 tell us at
vertex 12? The walk ab-1ab-1a starting at 12 ends at 12, so there must be
an edge coloured b starting at 12 and ending at 12, giving us Figure VIII.7.
This is the Schreier diagram we have been looking for, since the colouring
is clearly regular and it is a simple matter to check that each defining relation
l Cayley and Schreier Diagrams 153

I
I
/ -- ,
\

f
I
/
3
/ -...- -,_ 4 7
,.-
_JlJ..-
11
\
\
(

\
,
~
/
I

.
\ \ I
I I I \
,

\
\ \ I
J / I I /

Ito
\j:
f f
1 2\ I I
12

-5
I
~\
\
\ _4:- L
8 -~-
- '* --a
6 9 ---b
Figure VIII.7. The complete Schreier diagram.

is satisfied at each vertex. In fact, a 2 = 1 and b 5 = 1 are obviously satisfied;

since (ba)3 = 1 holds at 6 by construction, it also holds at each vertex of
the walk 621236, etc.
This detailed and cumbersome description fails to do justice to the
method which, when performed on a piece of paper or on a blackboard, is
quick and efficient. The reader is encouraged to find this out for himself.
What the Schreier diagram certainly tells us is the index of B: it is simply
the number of vertices. Indeed, Schreier diagrams are often constructed on
computers just to determine the index of a subgroup. In some cases, as in
the example above, it tells us considerably more. The Schreier diagram is
essentially a shorthand for the representation of A as a group of permuta-
tions of the co sets of B. In this case a -+ (12)(36)(47)(58)(9 11)(10 12) and
b -+ (1)(23456)(78910 11)(12). Since the permutation corresponding to b
has order 5, which is exactly the order of b in A, we see that A is a group of
order 5.12 = 60 and is, in fact, isomorphic to A 5 , the alternating group on
5 symbols.
If we want our Schreier diagram to carry more information, then we fix
certain representatives of the cosets and keep track of the effect of the genera-
tors on these representatives. We decorate each coset by its representative:
if H is decorated by [h] then H = Bh. Now if there is an edge coloured a
from H = Bh to K = Bk, then we decorate this edge by [IX] if ha = ak. Since
K = Bk = H a = Bha, we see that IX E B, so the edges are decorated with
elements of B. Furthermore, if H, K and L are decorated with h, k and I,
and there are edges coloured a, b, c and decorated [a], [P] and [y] joining
them, as in Figure VIII.8, then habc = akbc = IXPlc = apyh. In particular,

K[k]
\
\
\
\
\

a[lX] ~ b[P]
\
\
\
\
\
\
H[h] ..................................... :. L[I]
cry]
Figure VIII.S. A cycle in a decorated Schreier diagram.
154 VIII Graphs and Groups

b[b] 3
/-~

{ \

'. f
\

B= 1~-+---<

a[l]
4
Figure VIII.9. The decorations induced by a spanning tree.

if h = 1 we have abc = af3y. Of course, an analogous assertion holds for

arbitrary walks starting and ending at B: the product of the colours equals
the product of the decorations.
One of the simplest ways of decorating the vertices and edges makes use
of spanning trees. Select a spanning tree of the Schreier diagram. Decorate
B, the subgroup itself, by 1 (the identity) and decorate the edges of the
spanning tree also by 1. This determines the decoration of every vertex (that
is coset) and every edge. Indeed, for each vertex H the spanning tree contains
a unique path from B to H; clearly H has to be decorated with the product
abc . .. corresponding to this path. These coset representatives are said to
form a Schreier system for B mod A. What is the decoration of a chord HK,
an edge not in the tree? By the remark above it is the product of the colours
on the B-H path, the edge HK and the K-B path, as in Figure VIII.9.
Since each element of B is the product of the colours on a closed walk
from B to B in the Schreier diagram, the decorations of the chords generate
B. Thus from a Schreier diagram we can read off a set of generators of B,
independently of the structure of A.

Theorem 2. The subgroup B of A is generated by the decorations of the chords.

o
In particular, the subgroup B in Figure VIII.9 is generated by b, ab 3 a-1,
ab - 1 ab - 1 a-I and ababa - 1.
It is equally simple to find a presentation of B, provided we have a presen-
tation of A. This is obtained by the Reidemeister-Schreier rewriting process;
we give a quick and loose description of it. The generators of this presenta-
tion are the chords of the spanning tree; to distinguish chords of the same
colour we write Ci for the edge coloured c starting at vertex i. For each vertex
i and each relator RJl denote by R~ the (word of the) walk starting at i given
by RJl expressed as a product of the cj ' say R~ = bic j . The reader can
easily fill in the missing details in the proof of the following beautiful result,
due to Reidemeister and Schreier.

Theorem 3. The subgroup B has a presentation

D
2 Applications of the Adjacency Matrix 155

Now if we wish to preserve the connection between this presentation of

B and the original presentation of A, we simply equate C i with the decoration
of the edge coloured c starting at vertex i.
If A is a free group, its presentation contains no relations. Hence the
above presentation of B contains no relations either, so B is also a free group.
This is a fundamental result of Nielsen and Schreier.

Theorem 4. A subgroup of afree group isfree. Furthermore, if A is afree group

of rank k (that is it has k free generators) and B is a subgroup of index n then
B has rank (k - l)n + 1.
PROOF. The presentation of B given in Theorem 3 is a free presentation on the
set of chords of the Schreier diagram. Altogether there are kn edges of which
n - 1 are tree edges; hence there are (k - l)n + 1 chords.

It is amusing to note that Theorem 3 implies that the rank of a free group
is well defined. Indeed, suppose that A is freely generated by a, b, .... Then
every directed multigraph with a distinguished vertex (corresponding to
the subgroup) and a regular colouring (by a, b, ...) is the Schreier diagram
of some subgroup B of A, for threre are no relations to be satisfied. Hence
subgroups of index n are in 1-1 correspondence with regularly coloured
multigraphs of order n. In particular, if A has k 2 2 generators, it has 2k - 1
subgroups of index 2, for there are 2k multigraphs of order 2 regularly
coloured with k colours, but one of those is disconnected. The number of
subgroups of index 2 is clearly independent of the presentation, so k is
determined by A.

2 Applications of the Adjacency Matrix

Recall from 3, Chapter II that the vertex space Co(G) of a graph G is the
complex vector space of all functions from V(G) into C. Once again we take
V(G) = {VI' V2"'" vn } so that dim Co(G) = n and write the elements of
Co(G) in the form x = I?= 1 XiV i or x = (xJ~; here Xi is the value of x at Vi'
also called the weight at Vi' The space Co(G) is given the natural inner product
associated with the basis (Vi)~: (x, y) = L?= 1 XiYi'
We shall concentrate on the adjacency matrix A = (aij) of G, the 0-1
matrix where aij = 1 iff Vi Vj is an edge. As usual, A is identified with a linear
endomorphism of Co(G). To start with, we recollect some simple facts from
linear algebra. The matrix A is real and symmetric, so it is hermitian, that is
(Ax, y) = (x, Ay). Hence its numerical range
V(A) = {(Ax, x): (x, x) = I}
is a closed interval of the real line. The eigenvalues of A are real, say Al >
..12 > ... > At> and V(A) isexactiy the interval [At> All (For simplicity an
156 VIII Graphs and Groups

eigenvalue of A is said to be an eigenvalue of G). We shall write Amax(G) for

the maximal eigenvalue Al and AmiJG) for the minimal eigenvalue At. The
inner product space Co(G) has got an orthonormal basis consisting of
eigenvectors of A. In particular, if m(A) denotes the (geometric or algebraic)
multiplicity of an eigenvalue A then L:l=
I m(Ai) = n. These remarks enable
us to deduce some basic properties of the eigenvalues.

Theorem 5. Let G be a connected graph of order n with adjacency matrix A.

(i) Every eigenvalue A of G satisfies IAI s ~ = ~(G).
(ii) ~is an eigenvalue of G iff G is regular; if ~ is an eigenvalue then m(M = 1.
(iii) If - ~ is an eigenvalue of G then G is regular and bipartite.
(iv) If G is bipartite and A is an eigenvalue of G then so is - A, and m(A) =
m( -A).
(v) b(G) s Amax(G) s ~(G).
(vi) If H is a spanned subgraph of G then Amin(G) S Amin(H) S Amax(H) S
AmaiG).
PROOF. (i) Let x = (xJ be a non-zero eigenvector with eigenvalue A. Let xp
be a weight with maximum modulus: Ixp I :?: IXi I for every i; we may assume
without loss of generality that xp = 1. Then

IAI = IAXpl = iltlaplxli s Jlapllxz! S IXpld(vp) s IXpl~ =~, (*)

showing IAI s ~.
(ii) If A = ~ is an eigenvalue and x, xp are chosen as in (i), then (*) im-
plies that d(v p) = ~ and XI = xp = 1 whenever VI is adjacent to vp. In turn
this implies that d(vI) = ~ and Xk = XI = 1 whenever Vk is adjacent to VI'
and so on; as G is connected d(vJ = ~ and Xi = 1 for every i. Hence G is
~-regular and x is j, the vector all of whose entries are 1.
Conversely, if G is ~-regular then (Aj)i = I7= I ail = ~ so Aj = ~j.
(iii) If A = - ~ is an eigenvalue then, as in (ii) inequality (*) implies that
d(v p) = ~ and XI = -xp = -1 whenever VI is adjacent to vp. As in (ii),
this implies that G is ~-regular. Furthermore, at each vertex Vk adjacent to
VI the weight is 1, at each neighbour of Vk it is -1, and so on. The weight is 1
at the vertices at an even distance from vp and it is - 1 at the other vertices
and every edge joins vertices of different weights. Thus G is bipartite, say
V = VI U V2 , where xp E VI.
(iv) Suppose G is bipartite with vertex classes VI and V2 Let b be the
function (vector) that is 1 on VI and - 1 on V2 Then x ~ bx = (b i Xi)~ is
an automorphism of the vector space Co(G). Now if Ax = AX and Vi E VI'
say, then
n

(A(bxi = L: aijbjxj = I aijXj - I aijXj = - L: aijXj

j= 1 VjEV 1 VjEV2 VjEV2

n
- L: aijX j - L: aijXj = - I aijXj = - AX i = -A(bx);.
VjEV t VjEV2 j= 1
2 Applications of the Adjacency Matrix 157

Hence b gives an isomorphism between ker(A - A) and ker(A + A), showing

m(A) = m( - A).
(v) We know already that Amax(G) ::::; ~(G). Note that for j = (1, 1, ... , 1)
we have <j, j) = n so

Hence Amax(G) = max V(A) ~ J(G).

(vi) It suffices to prove the result for a spanned subgraph H of order
n - 1, say V(H) = {Vi> V 2 , . , vn-d.
Let A' be the adjacency matrix of H. Then there is a vector y E Co(H)
such that <y, y) = 1 and <A'y, y) = Amax(H). Let x = (Yl, Y2' ... , Yn-l' 0).
Then x E Co(G), <x, x) = 1 and <Ax, x) = <A'y, y) = AmaiH) E V(A). Con-
sequently Amax(G) ~ Amax(H). The other inequality is proved analogously.
o
It is reasonable to expect that a graph with many automorphisms will
have particularly pleasing properties and that these will be reflected in the
adjacency matrix. The automorphism group of a graph G is the group, Aut G,
of permutations of the vertices preserving adjacency. Every abstract group
can be represented as the automorphism group of some graph. For instance,
if F is any finite group, consider its Cayley diagram with respect to some
set of generators. The automorphism group of this coloured and directed
mu[tigraph is exactly F. It only remains to replace each edge of this diagram
by a suitable subgraph which bears the information previously given by the
direction and colour. This produces a graph G with automorphism group
isomorphic to F. An example is shown in Figure VUllO.
Each n E Aut G induces an endomorphism of Co(G) which is described
by a permutation matrix P. In fact, an arbitrary permutation matrix Q
corresponds to an automorphism of G precisely when it commutes with the

Figure VULlO. A graph with automorphism group S3' constructed from the Cayley
diagram in Figure VIII. I.
158 VIII Graphs and Groups

adjacency matrix A, that is AQ = QA. The group of these matrices therefore

faithfully represents Aut G. Regarding A as an endomorphism of Co(G),
we find that the eigenspaces of A are invariant under P. In particular, if an
eigenvalue of A is simple (that is it has multiplicity 1) then P must map an
eigenvector to a multiple of itself. Thus if all eigenvalues are simple, we must
have p 2 = I. This is a strong restriction on those permutations which might
correspond to automorphisms of G. For example, if G has at least 3 vertices
and every eigenvalue is simple, then Aut G cannot be vertex-transitive, so
not every pair of vertices can be interchanged by an automorphism.
these remarks indicate how the methods of representation theory may
be used to deduce restrictions on the adjacency matrix of graphs which have
extensive automorphism groups. The lack of space prevents us from ex-
ploring this further. Instead, we shall use algebraic methods to study graphs
which are highly regular although this regularity is not expressed in terms
of the automorphism group.
Algebraic methods are particularly useful if we want to prove that certain
regularity conditions cannot be satisfied except perhaps for a small set of
parameters. A problem of this type arose in Chapter IV: for which values
of k is there a k-regular graph of order n = k 2 + 1 and girth 5? We shall
show later that if there is such a graph then k is 2, 3, 7 or 57. Group theory is
particularly rich in problems of this type, especially because of the search
for sporadic simple groups. Later on we shall mention some examples.
Regularity, that is the condition that all vertices have the same degree k,
is not sufficiently restrictive. There are too many k-regular graphs of order
n (provided k < nand kn is even) so we have to go considerably further to
find interesting classes. Call a connected graph G highly regular with col-
lapsed adjacency matrix C = (Ci) if for every vertex x E V = V(G) there is a
partition of V into sets V1 = {x}, V2 , . , Vp such that each vertex yE J.'j is
adjacent to exactly cij vertices in ~ (see Figure VIII.11). It is immediate

2
from the definition that G is regular, say every vertex has degree k. In this
case each column sum in the collapsed matrix is k. The collapsed matrix C

:: 2*2
1

3 3
3 3
3 3 3

(02 1 0)I (03 1 0)

1 1
(~ ~ ~ ~) 022
1

Figure VIlLI I. The pentagon, the cube and the Petersen graph together with collapsed
adjacency matrices.
2 Applications of the Adjacency Matrix 159

can be obtained from the adjacency matrix A as follows:

cij = Last, where Vt E lj.
VsEVi

The point is exactly that the above sum is independent of the representative
Vt of Jt;. We are especially interested in the collapsed adjacency matrix C
if it is of a much smaller size than A.
Consider a partition VI = {v d, Vz , ... ,~ belonging to the vertex
x = VI' Let P be the p-dimensional complex vector space of formal sums
Lf= I Zi Jt;, Zi E C, and let n: Co(G) -+ P be the linear map given by

Of course, C can be considered to map P into itself.

Theorem 6.
(i) nA = Cn, that is n(Ax) = C(nx) for every x E Co(G).
(ii) The adjacency matrix A and the collapsed adjacency matrix C have the
same minimal polynomial. In particular, A is an eigenvalue of A iff it is a
root of the characteristic polynomial of C.
PROOF. (i) Let us show that n(Av t) = C(nv t), where Vt is the basis vector
corresponding to an arbitrary vertex Vt E lj. To do this it suffices to check
that the ith coordinates of the two sides are equal. Clearly nV t = lj so

(n(Avt))i = Last
VsEVi

and

and these are equal by definition.

(ii) Let q be the minimal polynomial of C. In order to prove q(A) = 0,
by symmetry it suffices to check that (q(A)x) I = 0 for every x. This is indeed
so:
(q(A)x) I = (n(q(A)x))1 = (q(C)(nx))1 = (0)1 = o.
Conversely, the minimal polynomial of A annihilates C since nCo(G) = P.
D

This result enables us to restrict rather severely the matrices C that may
arise as collapsed adjacency matrices.

Theorem 7. Let G be a connected non-bipartite highly regular graph of order

n with collapsed adjacency matrix C. Let AI> ..1. 2 , .. , A, be the roots of the
160 VIII Graphs and Groups

characteristic polynomial of C different from k, the degree of the vertices of

G. Then there are natural numbers m 1 , m z , ... , mr such that
r

Lmi=n-l
i= 1
and

I: m;A'i = - k.
i= 1

PROOF. We know from Theorem 5 that ..1. 1 , A. z , ... ,A.r are the eigenvalues of
A in addition to k, which has multiplicity 1. Thus if m(A.;) is the multiplicity
of A.i then
r
1 + I: m(A.;) = n,
i= 1

since Co(G) has an orthonormal basis consisting of eigenvectors of A.

Furthermore, since the trace of A is 0 and a change of basis does not alter
the trace,
r
tr A = k + I: m(A.i)A.i = O. o
i= 1

The condition expressed in Theorem 7 is not easily satisfied, especially

if ..1. 1 , A. z , ... , A.r are not rational numbers, so it does rule out the possibility
of constructing many highly regular graphs with seemingly feasible param-
eters. For the so-called strongly regular graphs we shall rewrite the condition
in a more attractive form. A graph G is said to be strongly regular with
parameters (k, a, b) if it is k-regular, any two adjacent vertices have exactly
a common neighbours and any two non-adjacent vertices have b ~ 1
common neighbours. In other words, G is highly regular with collapsed
adjacency matrix
0
(
C = k a ob ) .
o k-a-l k-b

Theorem 8. If there is a strongly regular graph of order n with parameters

(k, a, b) then
1{ (n - l)(b - a) - 2k }
mj, mz ="2 n- 1 {(a _ b)Z + 4(k _ b)}l/Z

are natural numbers.

PROOF. The characteristic polynomial of the collapsed adjacency matrix
Cis
x3 + (b - a - k)x Z + a - b)k +b - k)x + k(k - b).
2 Applications of the Adjacency Matrix 161

On dividing by x - k, we find that the roots different from k are

A1, A2 = Ha - b {(a - b)2 + 4(k - b)}1/2}.
By theorem 7 there are natural numbers m 1 and m 2 satisfying
m1 + m2 = n - 1
and
m1A 1 + m2 A2 = -k.
Solving these for m 1 and m 2 we arrive at the assertion of the theorem. 0

Theorem 8 is sometimes called the rationality condition for strongly

regular graphs. It is also easily proved without invoking Theorem 7. Indeed,
if A is the adjacency matrix of a strongly regular graph with parameter
(n, k, a, b), then
kif i = j,
(A2);j = { a ifv;vjEE(G),
b otherwise
Hence
A 2 = kI + aA + b(J - I - A),
where J is the matrix all whose entries are 1. Therefore J is a quadratic
polynomial in A, so J and A are simultaneously diagonalizable. Noting
that J has only two eigenvalues: n, with multiplicity 1, and 0, with multi-
plicity n - 1, one can easily find that Al and A2 are as above (cf. Exercise 28).
Now we shall use the rationality condition to fulfil our promise con-
cerning Moore graphs of diameter 2 (or girth 5).

Theorem 9. Suppose there is a k-regular graph G of order n = k 2 + 1 and

diameter 2. Then k = 2, 3, 7 or 57.
PROOF. We know from Theorem 1 of Chapter IV that G is strongly regular
with parameters (k,O, 1). By the rationality condition at least one of the
following two conditions has to hold:
(i) (n - 1) - 2k = k 2 - 2k = 0 and n - 1 = k 2 is even,
(ii) 1 + 4(k - 1) = 4k - 3 is a square, say 4k - 3 = S2.
Now if (i) holds then k = 2.
If (ii) holds then k = S2 + 3); on putting this into the expression for
the mUltiplicity m1 we find that
_ ~ {~( 2 3)2 [(S2 + 3)2/16J - [(S2 + 3)/2J}
m1 - 2 16 s + + s '

that is
162 VIII Graphs and Groups

Hence s divides 15 so s is one of the values 1, 3, 5 and 15, giving k = 1, 3, 7

or 57. The case k = 1 is clearly unrealizable. 0

It is worth noting that for k = 2, 3 and 7 there are unique k-regular

graphs of order k 2 + 1 and diameter 2 (and so girth 5). In particular for
k = 2 it is a pentagon and for k = 3 it is the Petersen graph. However, it is
not known whether or not k = 57 can be realized.

Sporadic simple groups are those simple groups which do not belong to
one of the infinite sequences consisting of cyclic groups of prime order,
alternating groups of degree at least 5 and simple groups of Lie type. As
we remarked earlier, sporadic simple groups are often connected to strongly
regular graphs. For example, there is a strongly regular graph with param-
eters (162, 105, 81) and the Maclaughlin group of order 898 128000 is a
subgroup of index 2 of the automorphism group of this graph. Similarly,
there is a strongly regular graph with parameters (416, 100,96) and the
Suzuki group, which is a simple group of order 448 345 497 600, is a sub-
group of index 2 of the automorphism group of this graph.

3 Enumeration and P61ya's Theorem

We cannot end the book without considering perhaps the most basic ques-
tion about graphs, namely, how many of them are there? We may want to
count graphs with a given set of vertices or we may be interested in the
number of isomorphism classes of certain graphs. As we saw in Chapter VII,
counting labelled graphs is relatively easy, for instance there are 2m = 2N
labelled graphs on n vertices, of which (~) have m edges. Furthermore, by
applying Theorem 8 of Chapter II to the complete graph, one can easily
show that there are nn-2 labelled trees of order n. This result was first ob-
tained by Cayley; we present it here with a proof due to PrOfer, which is
independent of Theorem 8 of Chapter II.

Theorem 10. There are nn - 2 trees on n labelled vertices.

PROOF. As in Chapter VII, let V = {l, 2, ... , n} be the set of vertices. Given
a tree T, associate a code with T as follows. Remove the end vertex with the
smallest label and write down the label of the adjacent vertex. Repeat the
process until only two vertices remain. The code obtained is a sequence of
length n - 2 consisting of some numbers from 1, 2, ... , n; of course any
number may occur several times in the code (see Figure VIII.12). As the
reader should check, each of the nn- 2 possible codes corresponds to a unique
tree. 0
3 Enumeration and P6lya's Theorem 163

6
7

9
5

10 4
Figure VIII.l2. The Priifer code of this tree is (3, 8, 11,8, 5, 8, 3, 5, 3).

It is easily seen that the label of a vertex of degree d occurs exactly d - 1

times in the Priifer code of the tree. Thus the proof has the following con-
sequence.

Corollary 11. Let d 1 S d z S ... S dn be the degree sequence of a tree: d 1 ~ 1

and I?= 1 d; = 2n - 2. Then the number of labelled trees of order n with
degree sequence (d;)~ is given by the multinomial coefficient

(
d1 -
n- 2
1, d z - 1, ... , dn -
)
1 .
o
The difficulties we encounter change entirely if we wish to count certain
graphs within isomorphism. Given graphs G 1 and Gz with a common
vertex set V, when are they isomorphic? They are isomorphic if there is a
permutation 1[ of V which maps G 1 onto G z . Of course, strictly speaking 1[
does not act on graphs, it only induces a permutation a of X = VIZ), the
pairs of vertices, and it is a that maps an edge of G 1 into an edge of G z and a
non-edge of G 1 into a non-edge of Gz . Now G; is naturally identified with a
subset of X or, equivalently, with a function/;: X ~ {O, I}. Therefore G 1 is
isomorphic to Gz iff there is a permutation a of X = VIZ) (coming from a
permutation 1[ of V) such that a*fl = fz, where a* is the permutation of the
set of functions {O, I}X induced by a. Thus counting graphs within iso-
morphism is a special case of the following problem. Given sets X and Y,
and a group r acting on X, let r act on the set offunctions yX in the natural
way. How many orbits are there in yX? The main aim of this section is to
present a beautiful theorem of P61ya that answers this question.
Let r be a group of permutations acting on a (finite) set X. For x, y E X
put X ~ Y iff y = ax for some a E r. Then '" is an equivalence relation on
X; if x '" y we say that x is equivalent to y under r. The equivalence class of
x is called the r-orbit (or simply orbit) of x, and is denoted by [x]. For
x, yE X put
r(x, y) = {a E r: ax = y}.
Of course, r(x, y) is non-empty iff [x] = [y], that is x and y belong to the
same orbit. r(x) = r(x, x) is the stabiliser of x; it is a subgroup of r. Note
that if y = f3x then
r(x, y) = {a: ax = y} = {a: ax = f3x} = {IX: p-l a E r(x)} = f3r(x),
164 VIII Graphs and Groups

so r(x, y) is a coset of r(x). We see that Ir(x)1 depends on the equivalence

class of x, so we may put s([x]) = jr(x)l. Clearly
r = U r(x, y),
YE[X]
and this gives us
Irj = I[x] I jr(x) I = I[x] Is([x]).
The theorem mentioned in the title of this section is based on a version
of a lemma due to Burnside concerning the sum of the" weights" of orbits.
Let 0 1 , ,0, be the r-orbits, let A be an arbitrary Abelian group (written
additively) and let w: X ~ A be a function that is constant on orbits. We
call w a weight function and define the weight of 0i by w(O;) = w(x), X E 0i.
For a permutation a E r we denote by F(a) the set of elements fixed by a,
that is F(a) = {x E X: ax = x}. Thus x E F(a) iff a E r(x).

Lemma 12. Irj Ll= 1 W(Oi) = LaEr LXEF(a) W(X).

PROOF.
,
L xEF(a)
L W(X) = L L W(X) = L L L W(X)
XEDi
, , ,
aEr i= 1 aEr(x)

= L W(0 i) L L 1= L W( 0 i) I0;1 s( 0;) = jr I L W( 0;).

i= 1 XEDi aEr(x) i= 1 i= 1

The original form of Burnside's lemma is obtained on choosing A = 7L

and W == 1:
1 1
N(r) = -Irl
aEr
L xEF(a)
L 1 = -Irl L IF(a)l, aEr

where N(r) = I is the number of orbits.

We shall illustrate by two very simple examples, that even Burnside's
lemma can be used to calculate the number of equivalence classes of certain
objects.

EXAMPLE 1. Let X = {l, 2,3, 4} and r = {l, (12), (34), (l2)(34)}. What is
N(r)? Clearly F(1) = {I, 2, 3, 4}, F(12 = {3, 4}, F34 = {I, 2} and
F(12)(34 = 0. Thus N(r) = i{4 + 2 + 2 + O} = 2.

EXAMPLE 2. Consider all bracelets made up of 5 beads. The beads can be red,
blue and green, and two bracelets are considered to be identical if one can
be obtained from the other by rotation. (Reflections are not allowed!) How
many distinct bracelets are there?
3 Enumeration and P61ya's Theorem 165

In this case we choose X to be the set of all 35 = 243 bracelets and let r
be C s , the cyclic group of order 5, acting on X. Then the question is: how
many orbits does r have? For the identity 1 E r clearly F(1) = X. For
every non-trivial rotation rx E r only the 3 monochromatic bracelets are
invariant under rx, so N(r) = !{243 + 3 + 3 + 3 + 3} = 51.
The second example resembles a little the problem we really want to
tackle. Let r be a group of permutations of a (finite) set D. Let R be another
(finite) set and let us consider the set RD of all functions from D into R. Each
rx E r can be made to act on RD; namely define rx*: RD --+ RD by
(rx*f)(d) = f(rxd),
Then
r* = {rx*: rxE r}
is a group of permutations of RD; as an abstract group, it is isomorphic to
r and we distinguish it from r only to emphasize that it acts on RD while
r acts on D.
As customary in connection with P6lya's theorem, we adopt an intuitive
terminology. D is called the domain and its elements are places, R is the
range and its elements are figures, the functions in RD are called configura-
tions, finally a pattern is an equivalence class of configurations under r*,
that is a r*-orbit. Our main aim is to calculate the number of distinct pat-
terns.
The origin of this terminology is that a functionf E RD is an arrangement
of some figures into the places in such a way that for each place there is
exactly one figure in that place, but each figure can be put into as many
places as we like. Two configurations mapped into each other by an element
of r* have the same pattern and are not distinguished. Thus in the second
example the places are, say, 1, 2, 3, 4 and 5, the figures are r, band g (for red,
blue and green) and a configuration is a sequence of the type g, b, b, r, b, that
is a bracelet. The group r is generated by (12345) and distinct patterns
correspond to distinguishable bracelets.
In addition to counting the number of distinct patterns, we may wish to
count the number of patterns of a certain type. It turns out that all these
problems can be solved at once, provided we learn enough about the cycle
structure of permutations in r acting on D, and are willing to store much
information about the patterns.
Each element rx E r is an essentially unique product of disjoint cycles
(cyclic permutations) acting on D. If rx = ~1 ~2 .. ~m is such a product, we
say that ~ 1, ... , ~m are the cycles of rx. In the product we include cycles of
length 1 as well so that every rx E D appears in exactly one cycle; if I~ I denotes
the number of elements in ~ then Lk'=1 I~k I = d, where d = ID I is the number
of elements in D. Denote by Nrx) the number of cycles of rx having length k;
by the previous equality Lk'= 1 kNrx) = d. Note that IF(rx) I, appearing in
Burnside's lemma, is exactly jl(rx), the number of elements of D fixed by rx.
166 VIII Graphs and Groups

We define the cycle sum of r to be

d

Z(r; al"'" ad) = L TI alk(a\

aErk=1

where, as before d = ID I. The reader should bear in mind that Z depends on

the action of r on D, not only on the abstract group r. Note also that the
cycle sum is a polynomial in a l , a2' ... , ad with integer coefficients; it tells
us the distribution of cycles in the elements of r. When writing down a cycle
sum, it is useful to remember that L~= 1 kM(1.) = d for every (1.. The customary
cycle index of r is Z(r; a l , ... , ad) = (1/WDz(r; a l , ... , ad)' As we shall
consider general rings instead of the more usual polynomial ring with ra-
tional coefficients, we have to use the cycle sum since we. cannot divide
by WI.
Let A be an arbitrary commutative ring and let w: R -+ A be a function.
We call w(r) the weight of the figure r, and for k = 1,2, ... define the kth
figure sum as

Furthermore, the weight of a configurationf E RD is

w(f) = TI w(f(a)).
aED

Clearly any two configurations equivalent under r* have the same weight,
so we may define the weight of a pattern 0i by
w(0;) = w(f),
Our aim is to learn about the pattern sum
1
S= L w(O;),
i= I

where 0 1 , O 2 , ,0 1 are the r*-orbits, that is the distinct patterns.

Note that w(r), w(f), Sk and S are all elements of our commutative ring
A. If we have a way of determining the pattern sum S, it is up to us to choose
A and the weight function w: R -+ A in such a way that S can be "decoded"
to tell us all we want to know about various sets of patterns. In practice one
always chooses A to be a polynomial ring (Z[x], Q[x, y], etc.), and usually
w(r) is a monic polynomial; the information we look for is then given by
certain coefficients of the polynomial S. We shall give several examples
after the proof of our main result, P6lya's enumeration theorem.

Theorem 13. (P6Iya's enumeration theorem)

InS = z(r; SI' S2' '" ,Sd)'

3 Enumeration and P61ya's Theorem 167

PROOF. By Lemma 12
I

IriS = WI I w(OJ = L L w(f).

;= 1 aEl JEF(a*)

Now clearly
F(rx*) = {fERD:f is constant on cycles of rx}, so if ~1' ~2"'" ~m are
the cycles of rx and a E ~; means that a is an element of the cycle ~; then

F(rx*)={fERD:r;ER and f(a)=r; if aE~;,i=1,2, ... ,m}.

Hence
)jk(a)
I w(f) = I TI w(rJI~;I = TI I
m d (
w(r)k = TI si da ),
d

JEF(a*) (ri)cR ;= 1 k= 1 rER k= 1

giving
d
InS = L TI slk(a) = z(r; SI' S2"'" Sd)' o
aE 1 k = 1

If Ir I has an inverse in the ring A, say if A is a polynomial ring over the

rationals, then Theorem 13 can also be written in its more usual form:

Let us illustrate now how the theorem can be applied.

EXAMPLE 3. Let us consider again the bracelets made up of five beads, which
can be red, blue and green. Then D = {1, 2, 3, 4, 5} is the set of places of the
beads, R = {r, b, g} is the set of colours (figures) and r is C s , the cyclic
group of order 5 generated by the permutation (12345). The cycle sum is
Z = a~ + 4a s . On choosing A = 7L and w(r) = w(b) = w(g) = 1, we find
that Sk = 3 for every k, so 5S = 3 s + 4.3. Since each pattern (bracelet) has
weight 1, there are !{3s + 12} = 51 distinct patterns (bracelets).
On choosing A = 7L[x, y] and w(r) = 1, w(b) = x, w(g) = y, we find that
S = !{(l + x + y)S + 4(1 + X S + yS)}. Now it is easy to extract much
information from this form of S. For example, a bracelet has weight xy2 iff
it has 2 red, 1 blue and 2 green beads. Thus the number of such bracelets is
the coefficient of xy2 in the polynomial S, that is 0/5)(5 !/2!2!) = 6.

EXAMPLE 4. What happens if in the previous example we allow reflections?

Then r is the dihedral group Ds' the group of symmetries of the regular
pentagon, whose cycle sum is a~ + 4a s + 5a l a~. Thus if we take, as before,
A = 7L[x, y], w(r) = 1, w(b) = x and w(g) = y, we find that the number of
bracelets containing 2 red, 1 blue and 2 green beads is the coefficient of xy2 in
lo{(l + x + y)S + 40 + X S + yS) + 5(1 + x + y)(1 + x 2 + y2)2}, that
is,3 + 1 = 4. .
168 VIII Graphs and Groups

EXAMPLE 5. This is the example P6lya used to illustrate his theorem. Place
3 red, 2 blue and 1 yellow balls in the 6 vertices of an octahedron. In how
many distinct ways can this be done? The group of symmetries of the octa-
hedron has cycle sum a~ + 6ai a4 + 3ai a~ + 6a~ + 8a~: a~ comes from
the identity, 6ai a4 from rotations through rc about axes through the vertices,
6a~ from rotations through rc about axes through midpoints of edges, and,
finally, a~ is the summand corresponding to a rotation through 2rc/3 about
an axis going through the centre of a face. On taking A = Z[x, y], w(r) = 1,
w(b) = x and w(y) = y, we see that the required number is the coefficient of
x 2 y in
l4{(1+ x + y)6 + 6(1 + x + y)2(1 + X4 + y4)
+ 3(1 + x + y)2(1 + x 2 + y2)2 + 6(1 + x 2 + y2)3 + 8(1 + x 3 + y3)2},
that is, 3.
It should be clear by now that the theorem loses nothing from its general-
ity if instead of a general commutative ring A we take Z[x,: r E R], the poly-
nomial ring over the integers in variables with the elements of R, and we
define the weight function as w(r) = x,. Then the pattern sum S contains all
the information the theorem can ever give us. In particular, if w: R ~ A is
an arbitrary weight function then the corresponding pattern sum is obtained
by replacing x, by w(r) in S. However, if R is large, the calculations may get
out of hand if we do not choose a "smaller" ring than Z[x,: r E R], which is
tailor-made for the problem at hand. The choice of a smaller ring is, of
course, equivalent to a substitution into S.

EXAMPLE 6. Place red, blue, green and yellow balls into the vertices of an
octahedron. Denote by Pi the set of patterns in which the total number of
red and blue balls is congruent to i modulo 4. What is 1Pol - 1P 21 ?
The cycle sum of the rotation group of the octahedron was calculated in
Exercise 5 and was found to be a~ + 6ai a4 + 3ai a~ + 6a~ + 8a~.
Let A = C, the field of complex numbers, and put w(r) = w(b) = i,
w(g) = w(y) = 1. Then for a pattern/we have Re w(f)= 1 iffEP o , Re w(f)=
- 1 if / E P 2 and Re w(f) = 0 if / E PI U P 3 Thus IF 01 - jP 21 is exactly
the real part of the pattern sum. As s 1 = 2(1 + i), s 2 = 0, S 3 = 2(1 - i) and
S4 = 4, we see immediately that after substitution the real part of each term
is 0, so IF0 1= 1P 2 I

We were first led to our study of the orbits of a permutation group by

our desire to count the number of graphs within isomorphism. We realized
that this amounted to counting the orbits of the group r:
acting on {O, 1}x,
where X = V(2) and rn is the permutation group acting on X which is in-
duced by the symmetric group acting on V. So according to P6lya's theorem
our problem is solved when we know the cycle sum of the permutation group
rn. It is now a routine matter to write down an explicit expression for this
Exercises 169

cycle sum, though we don't display it here since its form is not very in-
spiring. Furthermore, except for small values of n it is too unwieldy for
practical calculations and it is much easier to use asymptotic formulae
derived by random graph techniques (see Exercises 22 and 23 of Chapter
VII).
We remark finally that an extension of P6lya's theorem covers the case
when there is also a group acting on the range of the functions. For instance,
if we let S2 act on {a, I} in the example above, we do not distinguish between
a graph and its complement, and may thereby compute the number of graphs
which are isomorphic to their complements.

EXERCISES

1. Draw the Cayley diagram of the quaternion group <a, b la 2 = b 2 = (ab)2).

2. Findtheordersof<a,bla 3 = b 3 = 1,ab = ba),<a,bla4 = b 3 = 1,ab = ba)and

<a, bla 3 = b4 = (aW = 1).
3. Use Euler's formula and information about the Cayley diagram to deduce that
in the group <a, b, cla 5 = b 3 = c2 = (abc)-') we have a 610 = 1.
4. Verify that the Cayley diagram associated with the trefoil knot is the diagram
described in Figure VIII.5.
5. Write down presentations of the knots shown in Figure VIII.13. Show that the
group of the quinquefoil is isomorphic to <J, 9 I f5 = g2) and the group of the
tweeny is <a, b Iababa-'b-'a-' babab-'a-'b -').

Figure VIII.13. The quinquefoil and the tweeny.

6. Prove that no two of the groups of the knots shown in Figures VIllA and VIII. 13
are isomorphic.
7. A closed orientable surface of genus 2 is obtained by identifying pairs of non-
adjacent sides of an octagon, say as in Figure VIII.14. The fundamental group
has a presentation <a" a2, a3, a41 a,az'a4ai'a3a2a4'a3'). Show that the Cayley
diagram is a tessellation of the hyperbolic plane. Deduce that any non-empty
reduced word W equal to 1 must contain a subword of length at least 5 which is
part of the cyclically written relator or its inverse. (A reduced word is one in which
no generator occurs next to its inverse.) [Hint. Consider the part of the walk W
furthest from 1.]
170 VIII Graphs and Groups

Figure VIII.14. An orientable surface of genus 2.

8. Show that the dihedral group Dn, the group of symmetries of a regular n-gon,
has a presentation of the form (a, bla n = b 2 = (ab)2 = 1). What is its Cayley
diagram?
9. Give a group whose Cayley diagram is the truncated cube having 8 triangular and
(i octagonal faces.

10. Draw the Cayley diagram of (i) (a, blab = ba) in the Euclidean plane, (ii)
(a, b Ibn, ab = ba) on an infinite cylinder, (iii) (a, b Iam = bn = 1, ab = ba) on a
torus, (iv) (a, b, c Ia4 = b 4 = c4 = abc = 1) in the hyperbolic plane.
11. Check the examples of Cayley diagrams illustrated in Figure VIII.3 and prove
Theorem 1.
12:- Let A = (a, bla 3 = b 4 = (ba)2 = 1) and B = (b). What is the Schreier diagram
of A mod B?
13.- A group A is generated by a, band c, the Schreier diagram of A modulo a subgroup
B is shown in Figure VIII.15. Read off a set of generators of B.

~/// ............... ~

--a
---b
......... c

Figure VIII. 15. The Schreier diagram of A mod B.

14. Let A be the free group on a, band c, and let B be the subgroup consisting of all
squares. What is the Schreier diagram of A mod B? Find a set of free generators
for B.
15. How many subgroups of index 2 are there in a free group of rank k?
16. How many subgroups of index n are there in a free group on 2 generators?
Exercises 171

17. Show that a subgroup B of a finitely generated free group F is of finite index in F
iff B is finitely generated and there is a natural number n for which W" E B for
every word W
18. What is the automorphism group of the Petersen graph (shown in Figure VIII.ll)?
Find the automorphism group of the Kneser graph Krl, where s ;::: 2r + 1 (see
p. 99. Deduce that the automorphism group of K~;+ I is 3-arc-transitive, that
is any path of length 3 can be mapped into any other path of length 3 by an auto-
morphism.
19. The Tutte 8-cage has vertices n l , n3, n5 for n = 1,2, ... , 10, with edges joining
n5 to nl and n3, and ni to mi iff In - ml == i (mod 10). Show that the auto-
morphism group of the Tutte 8-cage is 5-arc-transitive.
20. Find the automorphism group of the Grotzsch graph (see p. 99).
21. Does every tree have an Abelian automorphism group?
22. Construct a non-trivial tree with trivial automorphism group.
23. Let fen) be the minimal size of a graph of order with trivial automorphism group.
Prove that fen) < n for every n ;::: 7 and f(n)/n -> 1 as n -> 00.
24. Show that if 11 E Aut G has k odd cycles and 1 even cycles then G has at most
k + 21 simple eigenvalues.
25. Show that a connected graph G of odd order whose automorphism group is
vertex transitive has exactly one simple eigenvalue.
26.- Let A be the adjacency matrix of a graph G. Show that (A1)ij is the number of
Vi-Vj walks of length I.

27. Given k ;::: 2, Po (x), PI (x), ... be polynomials defined by Po(x) = 1, PI (x) = x,
pz(x) = X Z - k and

I ;::: 3.

Show that if A is the adjacency matrix of a k-regular graph then (Pl(Aij is the
number of Vi-Vj paths of length I.
28. Complete the details of the second proof of Theorem 8, as suggested on p. 161.
29. Prove that the matrix J (all of whose entries are 1) is a polynomial in the adjacency
matrix A iff G is regular and connected.

30. Combine Theorem 5(vi) and Theorem 1 of Chapter V to deduce that X(G) ::;
Am.x( G) + 1.
31. Let fG(x) = D~o CkX"-k be the characteristic polynomial of (the adjacency
matrix of) a graph G. Show that Co = 1, CI = 0, C2 = -e(G) and -C 3 is twice
the number of triangles in G.
32. Let k(x) be the characteristic polynomial of G. Show that if e = xy is a bridge
of G then
172 VIII Graphs and Groups

Now let F be a forest with 2n vertices and denote by dk the number of k-element
sets of independent edges. [Thus d; is the number of I-factors.] Prove that

What are the possible values for dn ?

33. Let G be a connected k-regular graph containing an odd cycle. At time 0 put a
counter on a vertex. For each counter that is on a vertex x at time t, place a counter
on every vertex adjacent to x at time t + 1 and remove the counters from t. Show
that n,(x)lk' tends to a limit as t -> 00, where nlx) is the number of counters on
x at time t. What is the corresponding assertion if the counters are not removed
from the vertices? [Hint. There is an orthonormal basis consisting of the eigen-
vectors of the adjacency matrix.]
34. A k-regular graph G of order n is such that any two non-adjacent vertices can be
mapped into any other such two by some automorphism of G. Show that G is
strongly regular. What are the eigenvalues of G? What is the condition that k
and n have to satisfy?
35. Let C be the collapsed adjacency matrix of a highly regular graph. Show that
(c l ) 11 is the number of walks of length I from a vertex to itself. Interpret the other
entries of C l
36. Why must the collapsed adjacency matrix of the Petersen graph, as shown in
Figure 11, have rational eigenvalues?
37. In the graph G every two adjacent vertices have exactly one common neighbour
and every two non-adjacent vertices have exactly two common neighbours.
Show that G is regular of degree k and has order 2k2 + 1, for some k in
{I, 2, 7,11,56, 497}.
38. Let G be a strongly regular graph of order 100 with parameters (k, 0, b). What are
the possible values of k? Find the eigenvalues when k = 22.
39. Make use of the result of Exercise 19 to calculate the eigenvalues of the Tutte
8-cage.
40. Give detailed proofs of Theorem 10 and Corollary 11.
41. Prove that the number of trees with n - 1 ~ 2 labelled edges is nn - 3.
42. Show that a given vertex has degree 1 in about lie of all labelled trees, where
e = 2.71828 .... (Cf. Exercise 7 in Chapter VII.)
43. Denote by R(n, k) the number of trees with n distinguishable vertices of which
exactly k have degree 1. Prove that

k
- R(n, k) = (n - k)R(n - 1, k - 1) + kR(n - 1, k).
n

44. Prove that there are

Notes 173

acyclic graphs on n distinguishable vertices have n - m edges. Deduce that there

are ten - l)(n + 6)nn ~ 4 forests with two components.
45. Show that the cycle sum of the symmetric group Sn acting on the usual n letters is

where the summation is over all partitionsjl + 2jz + ... + njn = n.

46. What is the cycle sum of S5 acting on the unordered pairs of 5 elements? How
many non-isomorphic graphs are there on 5 vertices? How many of them have
5 edges?
47. Let Z(r, ai' az, ... , ad) be the cycle index of a permutation group r acting on a set
D. Consider the action of r on the set of all k-subsets of D. How many orbits are
there?
48. Determine the cycle index of the rotations of the cube acting on (i) the vertices,
(ii) the edges, (iii) the faces, (iv) the faces and vertices. In how many distinct ways
can you colour the vertices using some of n colours? The edges? The faces? The
vertices and faces?
49. Show that the cycle sum of Cn, the cyclic group of order n, is Z(C n; ai' ... , an) =
Lk/n (k)ak n/\ where (k) is the Euler function.
50. Prove that the cycle index of the dihedral group Dn (cf. Exercise 8) is

. _I. {tala~~I)/z ifn is odd,

Z(Dn' aJ,"" az n) - zZ(Cn, al"'" an) + I( n/2
4 a2
2 (n~2)/Z)
+ ala2 'f .
1 n IS even.

How many bracelets are there with 20 beads coloured red, blue and green?
51. How many distinct ways are there of colouring the faces of a dodecahedron
with red, blue and green, using each colour at least once?

Notes
There is a vast literature concerned with group presentations, including
the use of Cayley and Schreier diagrams. The basic book is perhaps W.
Magnus, A. Karrass and D. Solitar, Combinatorial Group Theory, Second
Edition, Dover, New York, 1976; the connections with geometry are
emphasised in H. S. M. Coxeter, Regular Complex Polytopes, Cambridge
University Press, New York, 1974. Numerous articles deal with the com-
putational aspect, in particular J. A. Todd and H. S. M. Coxeter, A practical
method for enumerating cosets of a finite abstract group, Proc. Edinburgh
Math. Soc. (2) 5 (1936) 26~34, which was the first paper in this line and J.
Leech, Computer proof of relations in groups, in Topics in Group Theory
and Computation (M. P. J. Curran, ed.), Academic Press, New York, 1977,
in which some more recent developments are described.
174 VIII Graphs and Groups

Max Dehn posed the word problem in Dber unendliche diskontinuier-

liche Gruppen, Math. Ann. 71 (1911) 116-144, and gave the above discussed
presentation of the group of the trefoil in Dber die Topologie des dreidimen-
sionalen Raumes, Math. Ann. 69 (1910) 137-168. The word problem was
shown to be intrisically connected to logic by P. S. Novikov, On the algo-
rithmic unsolvability of the word problem, Amer. Math. Soc. Transl. (2) 9
(1958) 1-22 and G. Higman, Subgroups of finitely presented groups,
Proc. Royal Soc. A, 262 (1961) 455-475. For the properties of knots and
their groups the reader is referred to R. H. Crowell and R. H. Fox, Intro-
duction to Knot Theory, Graduate Texts in Mathematics, Vol. 57, Springer-
Verlag, New York, 1977.
An exposition of matrix methods in graph theory can be found in N. Biggs,
Algebraic Graph Theory, Cambridge University Press, New York, 1974.
The first striking result obtained in this way, Theorem 9, is due to A. J.
Hoffman and R. R. Singleton, On Moore graphs with diameters 2 and 3,
IBM J. Res. Dev. 4 (1960) 497-504. The connection between graphs and
sporadic simple groups is discussed in detail in R. Brauer and C. H. Sah
(editors), Theory of Finite Groups: a Symposium, Benjamin, Menlo Park,
1969. Since then however several more sporadic groups have been found.
The fundamental enumeration theorem of G, P61ya appeared in Kom-
binatorische Anzahlbestimmungen fUr Gruppen und chemische Verbin-
dungen, Acta Math. 68 (1937) 145-254. Many enumeration techniques were
anticipated by J. H. Redfield, The theory of group-reduced distributions,
Amer. J. Math. 49 (1927) 433-455. The standard reference book for P61ya-
type enumeration is F. Harary and E. M. Palmer, Graphical Enumeration,
Academic Press, New York, 1973.
Subject Index

Acyclic graph 5 Complete matching 54

Adjacency matrix 38, 155 Complete r-partite graph 5
Adjacent 1 Component 4, 50
Almost every 130 Conductance 29
Amount of flow 45 Configuration 97,165
Antichain 57 Conjugacy problem 148
Automorphism group 157 Connected graph 4, 50
Connection in parallel 29
Connection in series 29
Bipartite graph 5 Connectivity 50
Block 51 Convex property 133
Block-cutvertex graph 51 Country 17
Bridge 5 Cover 86
Broken cycle 99 Cubic graph 3
Burnside's lemma 164 Current 27, 45
Cut 36
space 36
Capacity 46 vector 36
Cayley diagram 146 Cut vertex 5
Chain 57 Cycle 4
Charge 98 space 36
Chord 37, 154 sum 166
Chromatic number 88 Cycles of a permutation 165
Chromatic polynomial 92 Cyclomatic number 37
Circulation 62
Clique number 86, 137
Closed trail 4 Defining relation 148
Cocycle space 36 Degree sequence 3
Collapsed adjacency matrix 158 Dehn's problems 148
Colouring 88 Density 84
Complete graph 3 Diameter 8

175
176 Subject Index

Dihedral group 167 Highly regular graph 158

Directed graph 5 Hypergraph 5
Directed multigraph 5
Discharging 97 Incidence matrix 38
Distance 4
Incident 1
Domain 165
Independent paths 51
Independent vertices, edges, paths 4
Economical spanning tree 9 Induced subgraph 2
Edge 1 Initial vertex 4
chromatic number 89 Integrality theorem 48
space 35 Isolated vertex 3
Eigenvalue 156 Isomorphic graphs 3
Electrical current 26 Isomorphism problem 148
Empty graph 2
Endblock 51 Join 5
Endvertex 1,4
Euler circuit 13
Eulerian graph 14 k-connected graph 50
Euler's formula 17 k-edge-connected graph 50
Euler trail 13 Kempe chain 97
Even cycle 4 Kirchhoff's laws 27
Extremal graph 67 Kneser graph 99
Knot 150

Face 17
Latin rectangle 64
Factor 45, 58
Fan 63 Length 4
Figure 165 Line graph 63
Loop 5
sum 166
Filter 118
Flow 45 Map 96
Forbidden subgraph problem 67 Max-flow min-cut theorem 47
Forest 5 Maximum degree 3
Four colour problem 97 Method of Heesch 97
Free group 148 Minimum degree 3
Fundamental cut 37 Monochromatic set 105
vector 37 Monotone class of graphs 75
Fundamental cycle 37 Moore graph 69, 161
vector 37 Multigraph 5
Multiple edge 5
Girth 18,68
Graph 1 Neighbouring 1
Graphic sequence 65 countries 17
Greedy algorithm 89
Grotzsch graph 99 Odd component 59
Odd cycle 4
Hadwiger's conjecture 97, 100 Ohm's law 26
Haj6s operation 102 Orbit 163
Hamilton cycle 12 Order of a graph 2
Hamiltonian graph 12 Oriented graph 6
Hamilton path 12
Handshaking lemma 4 Partially ordered set 57
Heawood's conjecture 96 Partition calculus 119
Subject Index 177

Path, xo-path 4, 79 Dilworth 57

Pattern 165 Dirac 87
sum 166 Erdos 87
Perfect squared square 33 Erdos and Stone 68, 80
Petersen graph 99 Gallai 102
Planar graph 16 Galvin and Prikry 116
Plane graph 16 Graham, Leeb and Rothschild 113
Polynomial identity 20 Haj6s 102
Potential difference 26 Hales and Jewett 113
Presentation of groups 148 Hall 54
Principle of superposition 28 Hindman 117
Problem of Zarankiewicz 73 Korshunov 144
Projective plane 75 Kuratowski 19
Lovasz 102
Radius 8 Menger 52
Ramsey set 111 Nielsen and Schreier 155
Randomly Eulerian 15, 24 Petersen 64
Range 165 P61ya 166
Rationality condition 161 P6sa 139
Realization of a graph 16 Rado 115
Reducible configuration 97 Ramsey 105
Regular graph 3 Ringel and Youngs 96
Reidemeister-Schreier rewriting process Smith 87
154 Schroder and Bernstein 64
Relator 148 Thomason 87
Resistance 26 Turin 72
r-partite graph 5 Tutte 58
Tychonov 64
van der Waerden 113
Schreier diagram 147 Topological graph 16
Schreier system 154 Total resistance 28
Set of distinct representatives 54 Tournament 23
Shadow 58 Trail 4
Simple transform 79 Transform 79
Sink 28,45 Travelling salesman problem 11
Size of a graph 2 Tree 5
Source 28, 45 Trefoil knot 151
Spanned subgraph 2 Trivial graph 3
Standard basis 35
Star-delta transformation 30
Stochastic matrix 63 Ultrafilter 118
Strongly regular graph 160 Unavoidable configuration 97
Subdivision 16 Upper density 84
Subgraph 2
Value of flow 45
Tait's conjecture 101 Vertex 1
Terminal vertex 4 space 35, 155
Theorem of
Amitsur and Levitzki 20 Walk 4
Appel and Haken 97 Weight 155
Bollobis and Erdos 80, 82 function 164
Bondy and Chvital 87 of a configuration 166
Brooks 91 Word 148
Dehn 35 problem 148
Index of Symbols

ab directed edge 5 GuH union 5

A = A(G) adjacency matrix 38 A(rx) cycles of length k
Aut G automorphism group 165
157 Kn complete graph 3
be (G) block-cut vertex graph K(p, q) = Kp,q complete bipartite
51 graph 5
B = B(G) incidence matrix 38 complete r-partite
e(x, y) capacity 46 graph 5
Co(G) vertex space 35, 155 K,(t) complete r-partite
C1(G) edge space 35 graph 5
C' cycle of length I 4 [m,n] the set m, m + 1, ... , n
diamG diameter 8 113
d(x) = dG(x) degree 3 countably infinite
d(x, y) distance 4 subsets 116
e = e(G) size of a graph 2 MW) finite subsets 117
E = E(G) edge set 2 N number of orbits 164
En empty graph 3 o(x),O(x) Landau's notation
ex(n; F) extremal size 67, 71 134
F(rx) fixed elements 164 [1, k] the set 1, 2, ... , k
G graph 1 110
IGI order of a graph 2 chromatic polynomial
G - E' deletion of edges 2 92
G[V'] spanned slibgraph 2 pi path of length I 4
G- W deletion of vertices 2 .9(V) power set 5
G + xy joining x to y 2 q(H) number of odd
Gn graph of order n 2 components 59
G(n, m) graph of order nand rad G radius ofG 8
size m 2 R(G) realization of G 16
bipartite graph 73 t,_l(n) size of a Tunin graph
isomorphic graphs 3 71
join 5 Tunin graph 71

179
180 Index of Symbols

TG topological graph 16 f30(G) independence number

2M power set 116 23
U(G) cocycle space 36 rex) = rG(x) vertices adjacent to x
V = V(G) vertex set 1 3
v(f) value of a flow 45 beG) minimum degree 3
w(f), w(r), w(o;) weights 166 ~(G) maximum degree 3
X Cr ) set of r-tuples 105 K(G) connectivity 50
LzJ greatest integer :-0;; z A(G) edge-connectivity 50
62 x(G) chromatic number
fzl least integer 2': z 62 88
z(m, n; s, t) Zarankiewicz number leG) edge-chromatic
73 number 88
Z(G) cycle space 36 chromatic number of
Z(r; a j , , ad) cycle index 166 surface 95
z(r; a j , , ad) cycle sum 166
Graduate Texts in Mathematics
Soft and hard cover editions are available for each volume up to vol. 14, hard cover
only from Vol. 15

1 TAKEun/ZARING. Introduction to Axiomatic Set Theory.

2 OXTOBY. Measure and Category.
3 SCHAEFFER. Topological Vector Spaces.
4 HILTON/ STAMMBACH. A Course in Homological Algebra.
5 MACLANE. Categories for the Working Mathematician.
6 HUGHES/ PIPER. Projective Planes.
7 SERRE. A course in Arithmetic.
8 TAKEun/ZARING. Axiomatic Set Theory.
9 HUMPHREYS. Introduction to Lie Algebras and Representation Theory.
2nd printing, revised.
10 COHEN. A Course in Simple Homotopy Theory.
11 CONWAY. Functions of One Complex Variable. 2nd ed.
12 BEALS. Advanced Mathematical Analysis, xi, 230 pages. 1973.
13 ANDERSON/ FULLER. Rings and Categories of Modules.
14 GOLUBITSKy/GUILLEMIN. Stable Mappings and Their Singularities.
15 BERBERIAN. Lectures in Functional Analysis and Operator Theory.
16 WINTER. The Structure of Fields.
17 ROSENBLATT. Random Processes. 2nd ed.
18 HALMOS. Measure Theory.
19 HALMOS. A Hilbert Space Problem Book.
20 HUSEMOLLER. Fibre Bundles. 2nd ed.
21 HUMPHREYS. Linear Algebraic Groups.
22 BARNES/ MACK. An Algebraic Introduction to Mathematical Logic.
23 GREUB. Linear Algebra. 4th ed.
24 HOLMES. Geometric Functional Analysis and Its Applications.
25 HEWITT / STROMBERG. Real and Abstract Analysis. 4th printing.
26 MANES. Algebraic Theories.
27 KELLEY. General Topology.
28 ZARISKI/ SAMUEL. Commutative Algebra I.
29 ZARISKI/ SAMUEL. Commutative Algebra II.
30 JACOBSON. Lectures in Abstract Algebra I: Basic Concepts.
31 JACOBSON. Lectures in Abstract Algebra II: Linear Algebra.
32 JACOBSON. Lectures in Abstract Algebra III: Theory of Fields and
Galois Theory.
33 HIRSCH. Differential Topology.
34 SPITZER. Principles of Random Walk. 2nd ed.
35 WERMER. Banach Algebras and Several Complex Variables. 2nd ed.
36 KELLEY/NAMIOKA. Linear Topological Spaces.
37 MONK. Mathematical Logic.
38 GRAUERT/FRITZSCHE. Several Complex Variables.
39 ARVESON. An Invitation to C*-Algebras.
40 KEMENY / SNELL/ KNAPP. Denumerable Markov Chains. 2nd ed.
41 ApOSTOL. Modular Functions and Dirichlet Series in Number Theory.
42 SEERE. Linear Representations of Finite Groups.
43 GILLMAN/ JERISON. Rings of Continuous Functions.
44 KENDIG. Elementary Algebraic Geometry.
45 LOEVE. Probability Theory. 4th ed. Vol. 1.
46 LOEVE. Probability Theory. 4th ed. Vol. 2.
47 MorSE. Geometric Topology in Dimensions 2 and 3.
48 SACHS/WU. General Relativity for Mathematicians.
49 GRUENBERG/WEIR. Linear Geometry. 2nd ed.
50 EDWARDS. Fermat's Last Theorem.
51 KLINGENBERG. A Course in Differential Geometry.
52 HARTSHORNE. Algebraic Geometry.
53 MANIN. A Course in Mathematical Logic.
54 GRAVER/WATKINS. Combinatorics with Emphasis on the Theory of
Graphs.
55 BROWN/PEARCY. Introduction to Operator Theory. Vol. 1: Elements
of Functional Analysis.
56 MASSEY. Algebraic Topology: An Introduction.
57 CROWELL/ Fox. Introduction to Knot Theory.
58 KOBLITz.p-adic Numbers, p-adic Analysis, and Zeta-Functions.
59 LANG. Cyclotomic Fields.
60 ARNOLD. Mathematical Methods in Classical Mechanics.
61 WHITEHEAD. Elements of Homotopy Theory.
62 KARGAPOLOV / MERZUAKOV. Fundamentals of the Theory of Groups.
63 BOLLOBAS. Graph Theory-An Introductory Course.
64 EDWARDS. Fourier Series, Volume 1. second edition.
65 WELLS. Differential Analysis on Complex Manifolds.
66 WATERHOUSE. Introduction to Affine Group Schemes.
67 SERRE. Local Fields.
68 WEIDMANN. Linear Operators in Hilbert Spaces.