Graph Theory

Mathematical Olympiad Series

ISSN: 1793-8570

Series Editors: Lee Peng Yee (Nanyang Technological University. Singapore)

Xiong Bin (East China Normal University. China)

Published
VoL 1 A First Step to Mathematical Olympiad Problems
by Derek Holton (University of Otago. New Zealand)

VoL 2 Problems of Number Theory in Mathematical Competitions

by Yu Hong-Bing (Suzhou University, China)
translated by Lin Lei (East China Normal University. China)

VoL 3 Graph Theory

by Xiong Bin (East China Normal University, China) &
Zheng Zhongyi (High School Attached to Fudan University. China)
translated by Liu Ruifang. Zhai Mingqing & Lin Yuanqing
(East China Normal University, China)
 Xiong Bin
East China Normal University, China

Zheng Zhongyi
High School Attached to Fudan University, China

Vol. 31 Mather:natical
Olympiad
Series

Graph Theory

translated by

Liu Ruifang
Zhai Mingqing
Lin Yuanqing
East China Normal University, China

~ East China Normal

~ University Press ,It World Scientific
Published by
East China Normal University Press
3663 North Zhongshan Road
Shanghai 200062
China

and

World Scientific Publishing Co. Pte. Ltd.

5 Toh Tuck Link, Singapore 596224
USA office: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK office: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-Publication Data

A catalogue record for this book is available from the British Library.

GRAPH THEORY
Mathematical Olympiad Series - Vol. 3
Copyright © 2010 by East China Normal University Press and
World Scientific Publishing Co. Pte. Ltd.

All rights reserved. This book, or parts thereof, may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permission from the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to
photocopy is not required from the publisher.

ISBN-13 978-981-4271-12-7 (pbk)

ISBN-IO 981-4271-12-8 (pbk)

Printed in Singapore by B & Jo Enterprise Pte Ltd

Editors
XIONG Bin East China Normal University, China
Lee Peng Yee Nanyang Technological University, Singapore

Original Authors
XIONG Bin East China Normal University, China
ZHENG Zhongyi High School Affiliated to Fudan University, China

English Translators
LIU Ruifang East China Normal University, China
ZHAI Mingqing East China Normal University, China
LIN Yuanqing East China Normal University, China

Copy Editors
NI Ming East China Normal University Press, China
ZHANG Ji World Scientific Publishing Co . , Singapore
WONG Fook Sung Temasek Polytechnic, Singapore
KONG Lingzhi East China Normal University Press, China

v
This page intentionally left blank
 Introduction

In 1736, Euler founded Graph Theory by solving the Konigsberg

seven-bridge problem. It has been more than two hundred years till
now. Graph Theory is the core content of Discrete Mathematics, and
Discrete Mathematics is the theoretical basis of Computer Science and
Network Information Science. This book vulgarly introduces in an
elementary way some basic knowledge and the primary methods in
Graph Theory. Through some interesting mathematic problems and
games the authors expand the knowledge of Middle School Students
and improve their skills in analyzing problems and solving problems.

vii
This page intentionally left blank
 Contents

Introduction Vii
Chapter 1 Definition of Graph 1
Chapter 2 Degree of a Vertex 13
Chapter 3 Turin's Theorem 24
Chapter 4 Tree 40
Chapter 5 Euler's Problem 51
Chapter 6 Hamilton's Problem 63
Chapter 7 Planar Graph 75
Chapter 8 Ramsey's Problem 84
Chapter 9 Tournament 101
Solutions 110
Index 145

ix
This page intentionally left blank
Chapter 1 Definition of Graph ~
~ - .::.~-->

Graph theory is a branch of mathematics on the study of graphs.

The graph we consider here consists of a set of points together with
lines joining certain pairs of these points. The graph represents a set
that has binary relationship.
In recent years, graph theory has experienced an explosive growth
and has generated extensive applications in many fields .
We often encounter the following phenomena or problems:
In a group of people, some of them know each other, but others
do not.
There are some cities. Some pairs of them are connected by air-
lines and others are not.
There is a set of points in the plane. The distance between some
of them is one and others are not one.
All the above phenomena or problems contain two aspects: one IS
object, such as people, football teams, cities, points and so on; and
the other is a certain relationship between these objects, such as
"knowing each other", "having a contest", "the distance between"
and so on. In order to represent these objects and the relationships,
we could use a point as an object, which is called a vertex. If any two
objects have a relationship, then there is a line joining them, which is
called an edge. Then we have constructed a graph.
We call the figure a graph a) • For instance, the three graphs G 1 ,

a) The general definition of graphs: a graph is a triplet (V, E, .p), where V and E are
two disjoint sets, V is nonempty and.p is a mapping from V x V to E. The sets V, E , .p are
vertex set, edge set and incidence function, respectively .
2 Graph Theory

G 2 , G 3 in Fig. 1. 1 are isomorphic, which contain some vertices and

edges joining them, representing some objects and the relationships
between them.
Fig. 1. 1 shows three graphs G 1 , G 2 , G 3 , where vertices are
represented by small circles.

~~l A
v,

G,
v, v;

G,
v]

G,
Fig. 1. 1

We can see that in the definition of graphs there are no

requirement on the location of the vertices, the length and the
curvature of the edges, and the fact whether the vertices and the edges
are in the same plane or not. However, we do not allow an edge
passing through the third vertex and also not let an edge intersect
itself. In graph theory, if there is a bijection from the vertices of G to
the vertices of G' such that the number of edges joining v i and v j
equals the number of edges joining v/ and v/' then two graphs G and G'
are isomorphic and considered as the same graph.
A graphG' = (V', E') is called a subgraph of a graphG = (V, E)
if V' c V, E' c E, that is, all the vertices of G' are the vertices of G
and the edges of G' are the edges of G.
For instance, the graphs G 1 , G 2 in Fig. 1. 2 are the subgraphs
of G.

G G, G,
Fig. 1. 2
 Definition of Graph 3

If there is an edge joining v i and v j in graph G, then v i and v j

are adjacent. Otherwise, they are nonadjacent. If the vertex v is an
end of the edge e, then v is incident to e. In Fig. 1.3, Vl and V2 are
adjacent, but V2 and V4 are not. The
e,
vertex V3 is incident to the edge e 4 •
We called the edge a loop if there is e, v,

an edge joining the vertex and itself.

v, e,
For instance , the edge e6 in Fig. 1.3 is a
Fig. 1. 3
loop.
Two or more edges with the same pair of ends are called parallel
edges. For instance, the edges in Fig. 1. 3 are parallel edges.
e1 , e2
A graph is simple if it has no loops or parallel edges. The graphs
G1 , G 2 , G 3 in Fig. 1. 1 are simple, whereas the graph in Fig . 1.3 is
not. In a simple graph, the edge joining v i and v j is denoted by
(v i ' V j) . Certainly, (v i' V j) and (v j ' Vi) are considered as the
same edge.
A complete graph is a simple graph in which any two vertices are
adjacent. We denote the complete graph with n vertices by K n. The
graphs K 3 , K 4' K 5 in Fig. 1.4 are all complete graphs. The number of

edges of the complete graph K" is G) = ~ n (n - 1).

K,

Fig. 1. 4

A graph is finite if both the number of the vertices IV I ( IV I is

also said to be the order of G) and the number of edges IE I are finite.
A graph is infinite if IV I or IE I is infinite.
In this chapter, unless specified, all graphs under discussion
should be taken to be finite simple graphs.
4 Graph Theory

These fundamental concepts mentioned above help us to consider

and solve some questions .
Example 1 There are 605 people in a party. Suppose that each of
them shakes hands with at least one person. Prove that there must be
someone who shakes hands with at least two persons.
Proof We denote the 605 people by 605 vertices VI' V2' ... ,

V605 ' If any two of them shake hands, then there is an edge joining the
corresponding vertices.
In this example we are going to prove that there must be someone
who shakes hands with at least two persons. Otherwise, each of them
shakes hands with at most one person. Moreover, according to the
hypothesis each of them shakes hands with at least one person . Thus
we have each of them just shakes hands with one person. It implies
that the graph G consists of several figures that every two vertices are
joined by only one edge .
Suppose that G have r edges . So G has
2r (even) vertices. It contradicts the fact that
the number of vertices of G is 605 (odd).
We complete the proof. Fig. 1. 5
Example 2 Is it possible to change the state in Fig. 1. 6 to the
state in Fig. 1. 7 by moving the knights several times? (In the figures ,
W stands for white knight, and B stands for black knight . knight
should be moved by following the international chess regulation)
Solution As Fig. 1. 8 shows, the nine squares are numbered and
each of them is represented by a vertex in the plane. If the knight can
be moved from one square to anther square , then there is an edge
joining the two corresponding vertices, as Fig . 1. 9 shows.

~~
I 4 7

2 5 8

~~ Fig. 1. 6 Fig. 1. 7
3 6

Fig. 1. 8
9
 Definition of Graph 5

'0 0
8
3

Fig. 1. 9
4
9
I W
8
5.
B
3 4

Fig. 1. 10
.
B 9
0
I W

8
5.

B
3

Fig. 1.11
4
2
W 9

Thus the beginning state in Fig. 1. 6 and the state in Fig. 1. 7 are
represented by the two graphs as in Fig. 1. 10, Fig . 1. 11, respectively.
Obviously, the order of the knight on the circle cannot be
changed from the state that two white knight are followed by two
white knight into the state that white knight and black knight are
interlaced. So it is impossible to change the states as required.
Example 3 There are n people A I , A 2 , . • . , A n taking part in a
mathematics contest, where some people know each other and any two
people who do not know each other would have common
acquaintance . Suppose that Al and A 2 know each other, but do not
have common acquaintance. Prove that the acquaintances of Al are as
many as those of A 2 •
Proof Denote the n people AI, A 2 , . .. , An by n vertices VI'
V2' . . . , V n. If two people know each other, then there is an edge

joining the two corresponding vertices. Then we get a simple graph G.

The vertices of G satisfy that any two nonadjacent vertices have a
common neighbor. We shall prove two adjacent vertices VI and V2 have
the same number of neighbors.
The set of neighbors of the vertex VI is denoted by N (VI) and the
set of neighbors of the vertex V2 is denoted by , ", ,
N ( V2 ). If there is a vertex
V i#- V 2 '

Al
then V i is not in N (V 2 ). Otherwise
and A2 have the common acquaintance A i .
V i in N ( VI) and
\V
'O'- :
" N(7J ,)
''''- -,"
" " N( v ,)
.... ---"
Vj ':
"

Thus V2 and V i have a common neighbor V j and

v j #- v 1. So N (V2) contains v j' as Fig. 1. 12 VI v~

shows. For V i ' V k in N (V I )' which are Fig. 1.12

6 Graph Theory

distinct from V2' both of them cannot be adjacent to a vertex v j in

N (V2)' which is distinct from VI . Otherwise, two nonadjacent
vertices V I ' V j have three common neighbors V2' V i ' V k . Therefore
v k in N (VI)' which is distinct from v k ' must have a neighbor v I in
N (V2)' which is distinct from v j . So the number of vertices in N (VI)
is not greater than that of N (V 2 ) . Similarly the number of vertices in
N (V2) is not greater than that of N (VI). Thus the edges incident to VI
are as many as those incident with V2.
Example 4 Nine mathematicians meet at an international
mathematics conference. For any three persons , at least two of them
can have a talk in the same language . If each mathematician can speak
at most three languages, prove that at least three mathematicians can
have a talk in the same language. (USAMO 1978)
Proof Denote the 9 mathematicians by 9 vertices VI' V2' .. . ,
V9. If two of them can have a talk in the ith language, then there is an
edge joining the corresponding vertices and color them with the i th
color. Then we get a simple graph with 9 vertices and edges colored.
Every three vertices have at least one edge joining them and the edges
incident to a vertex are colored in at most three different colors . Prove
that there are three vertices in graph G, any two of which are adjacent
to the three edges colored with the same color. (This triangle is called
monochromatic triangle. )
If the edges (v i ' V j ) , have the i th color, then the
(v i ' V k )

vertices v j , V k are adjacent and edge (v j , V k) has the i th color. Thus

for vertex VI , there are two cases :
(1) The vertex V I is adjacent to V 2 ' . . . , V 9. By the pigeonhole
principle, at least two edges, without loss of generality, denoted by
(VI' V2), have the same color. Thus triangle
(VI' V 3 ) ' D V I V2V 3 is a
monochromatic triangle .
(2) The vertex VI is nonadjacent to at least one of V 2' . . . , V 9.
Without loss of generality, we suppose that VI is nonadjacent to V 2 .
For every three vertices there is at least one edge joining them, so
there are at least seven edges from vertices V3 ' V4' . .. , V9 to the
 Definition of Graph 7

'Vertex VI or V2. From that we know at least four

vertices of V3' V4' . . . , V9 are adjacent with
vertex VI or V2' Without loss of generality, we
suppose that V3, V4' vs, V6 are adjacent to VI' as
it is shown in Fig. 1. 13. Thus there must be two
v,o
edges of (VI' V3)' (VI' V4)' (VI' vs), (VI' V6)
Fig. I. 13
which have the same color. Suppose (VI' V3),
(VI' V4) have the same color, then DVIV3V4 is a monochromatic
triangle .
Remark If the number 9 in the question is replaced by 8, then
the proposition is not true. Fig. 1. 14 gives a counterexample. Denote
the 8 vertices by VI' V2' . . . , V8 and 12 colors by 1, 2, ... , 12, and
there is no monochromatic triangle in the graph.

~.
I~~I
v, 2 v,
7~:~:~~J
v, 8 v,
Fig. 1. 14

The following example is the third question of national semor

middle school mathematics contest in 2000.
Example 5 There are n people, any two of whom have a talk by
telephone at most once. Any n - 2 of them have a talk by telephone 3 m
times, where m is a natural number. Determine the value of n. (China
Mathematical Competition)
Solution Obviously n ;;? 5. Denote the n persons by the
vertices AI, A 2 , ••• , An. If A i ' Aj have a talk by telephone,
then there is an edge (A i , A j ). Thus there is an edge joining two
of the n vertices. Without loss of generality, we suppose that it is
(A I ,A 2 ) .
Suppose there is no edge joining A I and A 3 • Consider n - 2
vertices AI, A 4 , A s , ... , An; A 2 , A 4 , As, . .. , An and A 3 , A 4,
8 Graph Theory

As, ... , A n. We know the number of edges joining any of A I ' A 2 ,

A3 to all of A 4 , As, ... , An is equal and we denote it by k .

Add A2 to the set AI, A 4 , As, ... , An' then there are S = 3m +
k + 1 edges joining the n - 1 vertices. Take away any vertex from n -1
vertices, the number of edges joining the remaining n - 2 vertices is
always 3 m • So there are k + 1 edges joining every vertex and the
remaining n - 2 vertices. Therefore,

S = ; (n - 1)(k +1).

Similarly, add A 3 to the set AI, A 4 , As, ... , An. We get n -1

vertices and the number of edges is t = 3 m +k = 21 (n - 1 ) k.

For S = t + 1, we have

that is n = 3. A contradiction. Thus there is an edge joining AI, A 3.

Similarly, there is also an edge joining A2 and A 3. Moreover,
there must be edges joining AI, A2 and allAi(i = 3,4, ... , n).
For Ai, Aj (i "#- j), there is an edge joining A i and AI' So there is
an edge joining Ai and A j • Thus it is a complete graph. Therefore,

Hence we have n = 5.
Example 6 There are n (n > 3) persons. Some of them know
each other and others do not. At least one of them does not know the
others. What is the largest value of the number of persons who know
the others?
SolutionConstruct the graph G: denote the n persons by n
vertices and two vertices are adjacent if and only if the two
corresponding persons know each other.
For at least one of them does not know the others, in graph G
there are at least two vertices which are not adjacent. Suppose that
 Definition of Graph 9

there is no edge e = (v 1 , V 2) joining VI' V2. Thus G must be K n - e if

it has the most edges. That is the graph taken away an edge e from the
complete graph K n. The largest number of vertices which is adjacent
with the remaining vertices is n - 2. So the largest number of people
who know the others is n - 2.
The following example is from the 29th International Mathematical
Olympiad (1988).
Example 7 Suppose that n is a positive integer and A I , A 2 , ••• ,

A 2n + 1 is a subset of a set B.
Suppose that
(1) each Ai has exactly 2n elements;
(2) each A i n Aj (1 ~ i <j ~ 2n + 1) has exactly one element;
(3) each element of B belongs to at least two Ai'S.
For which values of n can one assign to every element of B one of
the number 0 and 1 in such a way that Ai has 0 assigned to exactly n of
its elements?
Solution At first, the words "at least" in (3) can be replaced by
"exactly". If there is an elemental E Al n A 2n n A 2n +1 , then each of
the remaining 2n - 2 subsets A 2 , A 3 , . . . , A 2n - 1 has at most one
element of AI. Thus there is at least one element in Al but not inA2 U
A3 U ... U A2n-1 U A 2n U A 2n +l •

It contradicts (3).
Construct the complete graph K 2n+1' where every vertex Vi

represents a subset A i and every edge (Vi' v j) = eij (1 ~i, j ~ 2n + 1,

i eft j) represents the common element of A i ' A j • So the question can
be changed into: what property does n satisfy such that by as signing
the edges of K 2n+1 to 0 or 1, exactly n edges of the 2n edges incident to
any vertex v i are assigned to O?
K 2n+1 has n (2n + 1) edges. If the required method of assigning can
be met, then there are ~ n (2n + 1) edges which are assigned to o. So n

must be even.
Conversely, if n = 2m is even, we assign the edges (v i ' V i-m) ,
10 Graph Theory

( v; , V ;-m+ 1 ) , ••• , (v ; , V ;- 1) ' (V ; , V ;+1)' ... , (V ; , V ;+m ) ' i = 1,

2, ... , 2n + 1, to 0, otherwise to 1 in K 2n+ 1. Then the method can
meet the requirement. (Note that v ( 2n+ ])+ ; = v ; ).
Therefore , the condition of the question is satisfied if and only if
n IS even .
The following problem is from the IMO preseleced questions
in 1995.
Example 8 There are 12k persons attending a conference . Each
of them shakes hands with 3k +6 persons, where any two of them
shake hands with the same number of people. How many persons are
there in the conference?
Solution Suppose that for any two persons, they shake hands
with n people . For one person a, the set of all the persons shaking
hands with a is denoted by A and the set, the other persons by B. We
know from the problem that I A I= 3k +6, IB I= 9k - 7. For b E A,
n persons shaking hands with a, b are all in A . Therefore, b shakes
hands with n persons in A and 3k +5 - n persons in B. For c E B, n
persons shaking hands with a, c are all in A. Thus the number of
persons in A who have shaken hands with someone in B is
Ok +6)Ok + 5 - n) = (9k - 7)n ,
Ok + 6) Ok + 5)
n = 12k - 1

So 16n
(i2k - 1 + 25) (i2k - 1 + 21)
(i2k - 1)
Obviously, 0 , 12k - 1) = 1. So (i2k - 1) I 25 X 7. For 12k - 1
divided by 4 leaves 3, 12k - 1 = 7, 5 X 7, 52 X 7. By calculating 12k -
1 = 5 X 7 has the only integer solution k = 3, n = 6.
Next we construct a figure consists of 36 points . Each point is
incident to 15 edges and for any two points there are 6 points adjacent
to them .
Naturally, we can use 6 complete graphs K 6 . Divide the 36 points
into 6 teams and label the points in the same team. We get a 6 X 6
square matrix
 Definition of Graph 11

1 234 5 6
612345
561234
456123
345612
2 3 456 1

For any point in the square matrix, it only connects with 15 points in
the same row, in the same column, or having the same label. It is
obvious that for any two persons there are 6 persons who have shaken
hands with them.

Exercise 1

1 Consider the graph G = (V, E), where V = {Vl' V2' ••• ,

V and E = {(v l'

5 }, V 2)' (v 2' V 4)' (v 3' V 4), (v 4' V 5), (V 1, V 3) }.
Draw the graph G.
2 Let G be a simple graph, where I V I = n, IE I = e. Prove that
,;::: n(n - 1)
e ~ 2 .

3 Show the following two graphs are isomorphic.

(1) (2)

Fig. 1. 15

4 There are n medicine boxes. Any two medicine boxes have the
same kind of medicine inside and every kind of medicine is contained
in just two medicine boxes. How many kinds of medicine are there?
5 There are n professors A 1 , A 2 , ••• , An in a conference.
Prove that these n professors can be divided into two teams such that
12 Graph Theory

for every Ai' the number d i of the people whom he has acquaintance
with in another team is not less than d/ in his team, i = 1, 2, ... , n.
6 There are 18 teams in a match. In every round, if one team
competes with another team then it does not compete with the same
team in another round. Now there have been 8 rounds. Prove that
there must be three teams that have never competed with each other in
the former 8 rounds.
7 n representatives attend a conference. For any four represent-
atives, there is one person who has shaked hands with the other three.
Prove that for any four representatives, there must be one person who
shakes hands with the rest of the n - 1 representatives.
8 There are three middle schools, each of which has n students.
Every student has acquaintance with n + 1 students in the other two
schools. Prove that we can choose one student from each school such
that the three students know each other.
9 There are 2n red squares on the a big chess board. For any two
red squares, we can go from one of them to the other by moving
horizontally or vertically to the adjacent red square in one step. Prove
that all the red squares can be divided into n rectangules.
10 There are 2000 people in a tour group. For any four people,
there is one person having acquaintance with the other three. What is
the least number of people having acquaintance with all the other
people in the tour group?
11 In a carriage, for anym(m ~ 3) travelers, they have only one
common friend. (If A is a friend of B, then B is a friend of A.
Anyone is not a friend of himself. ) How many people are there in the
carriage?
12 There are five points A, B, C, D, E in the plane, where
any three points are not on the same line. Suppose that we join some
points with segments, called edges, to form a figure. If there are no
above five points in the figure of which any three points are the
vertices of a triangle in the figure, then there cannot be seven or more
than seven edges.
Chapter2 Degree of a Vertex

The degree of a vertex v in a graph G, denoted by de (v), is the

number of edges of G incident to v, where each loop is counted as two
edges. Moreover, when there is no scope for ambiguity, we omit the
letter G from graph-theoretic symbols and write, for example, d (v)
instead of d e (v). We denote by 8 (G) and t::. (G) the minimum and
maximum degrees of the vertices of G, or 8 and t::. for brevity.
In Fig. 2. 1 , d ( V 1) = 1 , d ( V 2) = 3, d ( V 3) = d ( V 4) = 2, 8 = 1,
t::. = 3.
A vertex is odd if its degree is odd, otherwise, it is even . In
Fig. 2.1, VI and V2 are odd vertices, and V3 and V4 are even.

V,

V3 0-----~V ,

Fig. 2. 1 Fig. 2. 2

A graph G = (V, E) is said to be k - regular, if d ( v) = k for all

v E V. The complete graph on n vertices is (n - i)-regular. Fig. 2. 2
shows a 3-regular graph.
The connection between the sum of the degrees of the vertices of a
graph and the number of its edges is given as follows.
Theorem 1 For any graph G on n vertices, the sum of the degrees
of all of the vertices is twice as large as the number of the edges. In
symbols, if G with n edges has vertices Vl' V2' ... , v,,, then
14 Graph Theory

Proof The sum of the degrees of all of the vertices d ( VI) +

d ( V 2) + ... + d ( v ,.) represents the whole number of the edges one of
whose ends is VI' V2' ... , or v n' Since each edge has two ends, every
edge of G is counted twice in the sum d ( v j) + d ( V 2) + ... + d ( v" ). So
the sum of the degrees of all of the vertices is twice as large as the
number of the edges.
For instance, in Fig. 2.1, e = 4,

Theorem 1 is often called the Hand-Shaking Lemma. A famous

conclusion is given by Euler about two hundred years ago, that is to
say, if many people shake hands when they meet, then the number of
times of shaking hands is even. Then we can have the conclusion that
there is an even number people who shake hands an odd number of
times. The corollary is the following Theorem 2.
Theorem 2 In any graph G, the number of vertices with odd
degree is even.
Proof Suppose that G has vertices VI' V2' ... , V n'
where
VI' ... , V 1 are odd vertices and v 1+1 , .,. , V n are even. According to
Theorem 1,
d (v + ... + d (v I) + d (v ,+1) + ... + d (v,,) = 2e,
1)

deVl) + ... +dev,) = 2e - deV +l) - ••• - d(v,,).

1

Since de v 1+1) + ... + d ev,,) are all even, the right sideof the
equality is even. However de VI), ... , d (v I) are all odd, then t must
be even so that d(vj) + ... +dev l ) is even. That is, the number of
vertices with odd degrees is even.
Example 1 Among n (n > 2) people, there are at least 2 persons,
where the number of their friends are the same.
SolutionWe denote the n people by the vertices VI , V2' ... , V n •
If two persons are friends, we join the corresponding vertices. Then
we get a graph. The assertion follows if we can find at least 2 vertices
 Degree of a Vertex 15

with the same degree in G.

A vertex is at most adjacent to other n - 1 vertices in a simple
graph on n vertices, so d ( v) <n - 1, for all v E V. Hence the degree
of a vertex in G can take only the following values:

0,1,2, ... ,n-1.

However, not all of them are feasible. Note that a vertex with
degree zero could not be adjacent to any other vertex and that the
vertex with degree n - 1 must be adjacent to any other n - 1 vertices.
So in G, only the following degrees are possible:

0,1, 2, ... , n -2,

or

1,2,3, .. . ,n-1.

According to the pigeonhole principle, there are at least 2 vertices

with the same degree.
Example 2 There are 24 pairs of contestants taking part in the
International Table Tennis Mixed Doubles Contest. Some athletes
shake hands before the game, and the two in one pair do not shake
hands with each other. After the game, one male athlete asks
everybody the number of hand-shaking, and all the answers are
different. How many people does the male contestant's female partner
shake hands with?
Solution The 48 vertices v, v o, Vj, •.. , V46 represent the 48
contestants where the male contestant is represented by v, with edges
joining two people who had shaken hands before, then we can get a
graph G. In graph G, d (v i) < 46, i = 0, 1, 2, ... , 46, and d (v i) *"
d (v j ), if i =1= j. So except v, the degree of the other vertices are

0, 1, 2, ... , 45, 46.

Without loss of generality, we suppose that d (v i) = i, for i = 0, 1,

2, ... , 46. Vertex V46 is adjacent to every vertex except Va, so Vo and
V46 are partners. Deleting V46' Vo and the edges which are incident to
16 Graph Theory

them, we can get graph G l . The degree of every vertex in graph G l

except v is still different and has decreased by 1. Likewise, V45 and VI
are partners. So are V44 and V2' .• • , V24 and V 22' That is to say, the
partner of v is V23' so the male contestant's female partner shakes
hands for 23 times.
Remark Changing 24 to 34 in Example 2, "male and female
partners" to "group leader and deputy group leader", this is one of
pre-selected problems In the 26th International Mathematics
Olympiad. Changing 24 to 16, "male and female partners" to "2
football teams A and B", that is the third problem of the extra test of
China Mathematical Competition in 1985 .
Example 3 Every city in one country has 100 roads connecting to
other cities, and any city can be reached from any other city. Now,
one road is closed for repair. Prove that any city can still be reached
from any other city.
Proof Assume that the road closed is AB. We need to prove that
B can still be reached from A. Otherwise, except A, the degrees of all
the vertices in the connected subgraph containing vertex A are even.
By Theorem 2, it is a contradiction.
Remark The key point is to study the connected subgraph. The
concept of connecting is very important, and we will encounter it
later .
Example 4 Some cities are located on both sides of a river, and
there are no less than 3 cities. Some routes connect the cities, and a
pair of cities locating on both sides is connected by one route.
Furthermore, each city is connected only with k cities on the other
side. People could go to one city from any other city. Prove that
people could still go to any city from any other city if one route is
canceled .
Proof We may consider the two sides as the north side and the
south side. The n cities on the north side are represented by Xl'
X2' . . . , X" and the whole set is denoted by X = {Xl' X2' ••• , Xn}.

The m cities on the south side are represented by Yl , Y2' ... , Ym and
 Degree of a Vertex 17

Y = {Yl' Y2' ... , Ym }. If there is a route between the city Xi on the

north side and the city Yi on the south side, we connect them to form
an edge (x i ' Y i)' and the set formed by all the edges is denoted by E.
Then we can get a graph formed by the vertex sets X, Y and the edge
sets E, which is called a bipartite graph, or an even graph, denoted
by G = (X , Y; E). The last two conditions are that there are only k
edges incident to any vertex, and the graph G is connected in the sense
that there exist some paths between any two vertices. The conclusion
is that deleting any edge e from E, the graph is still connected.
Each vertex is incident to k edges, so

IXlk=IEI=IYlk,
where 1X I , 1E I , 1Y 1 denote the numbers of the elements in sets X,
E, Y respectively. So 1X I = 1Y I , andn =m. Since 1X 1+1 Y 1~ 3,
then 1X 1= 1Y 1~ 2.
Deleting one edge from G, we can get a graph G'. If G' is not
connected, G' is composed of two connected components G l and G 2 •
Let

X = Xl U Xn Xl nX 2 = 1>,
Y = Y l UY2 , Y l nY 2 = 1>,

The deleted edge connects the vertices in Xl and Y 2 , then

1Xli k - 1 = 1El l = 1Y l 1k,
1X 2 1k -1 = 1E 2 1= 1Y 2 1k - 1.

Then (I Xli -I Y l I) k = 1, and hence k = 1.

G is connected, so 1X 1= 1, which is contradictory to 1X 1~ 2.
So G' is connected, and the conclusion holds.
Example 5 There are 99 members in a club, and every member
claims that he would play bridge only with someone he knows. Each
member knows at least other 67 members. Prove that there must be 4
members who can play bridge together. (Polish Mathematics Contest in
18 Graph Theory

1996)
Proof 1 Construct a graph G: denote the 99 members by 99
vertices and join every two vertices whose corresponding members
know each other. The condition is that d (v) ~ 67, for all v E V. We
should prove that there is a complete graph K 4 in G. For any vertex u
in G, d (u) ~ 67, so there exists a vertex v such that there are at most
(99 - 1 - 67 = )31 vertices which are neighbors of v, but not of u.
Similarly , there are at most 31 vertices which are adjacent to u, not v .
That is to say there are at least (99 - 31 - 31 - 2 = )35 vertices adjacent
to both u and v. As shown in Fig . 2. 3, assume that vertex x IS

neighboring with both u and v. Vertex

x is at least adjacent to one vertex y
which is a neighbor of both u and v,
since d (x) ~ 67, and there are at most
(31 + 31 + 2 = )64 vertices which are ," ,
not adjacent to u and v at the same \, ~ 31': "
time. So u, v, x and yare neighbors, Fig. 2.3
which proves the proposition .
Proof 2 The members are represented with the vertices, with
edges joining two strangers, then we can get a graph G'. Since each
person knows at least 67 people, d (v) ~ 99 - 1 - 67 = 31, for all v.

We need to prove that there exist 4 vertices which are not mutually
adjacent in G'. For a vertex u, choose a vertex v which is not adjacent
to u. In the remaining 97 vertices, the number of vertices that are
neighbors of either u or v could not exceed

d(u) + d(v) ~ 31 + 13 = 62.

So there exists a vertex x which is not adjacent to both u and v, and

the number of vertices adjacent to u, v or x cannot exceed

d(u) + d(v) + d(x) ~ 31 X 3 = 93.

Thus in the remaining 96 vertices, there must be a vertex y which is

not adjacent to u, v and x, then the 4 people corresponding to u, v,
 Degree of a Vertex 19

x and know each other, and they can play bridge together.
y
Remark 1 Changing to 66, we might not be able to find out 4
people who know each other. The counter-example is shown in Fig. 2.
4. Separate the vertex set V into three subset {V I ' V2' . . . , V33}, {V34'

V35' and { V67 ' V68' . . . , V99 } ' Any two vertices in the same
. . . , V 66 }

set are not adjacent, and two vertices in

different sets are always adjacent.
Obviously, the degree of every vertex is
66, and there are at least 2 vertices
belonging to the same subset in any 4
vertices. It means that they are not
adjacent. That is to say, there does not
exist 4 vertices which are adjacent Fig. 2. 4
mutually.
Remark 2 We can generalize it as follows: there are n en ~ 4)

members in a club, and each person at least knows [2; ] + 1 people,

then there must exist 4 persons who know each other.
Remark 3 If G is a simple graph on n vertices, then we can get a
graph by deleting all the edges belonging to G from the complete graph
K n' which is called the complementary graph of G, denoted by G. The
graph G in Proof 1 and the graph G' in Proof 2 are complementary.
Remark 4 We can make use of the pigeonhole principle to get
another good solution. Please try it yourself.
Example 6 In a region, 20 members of a tennis club hold 14
single matches, and everybody plays at least once. Prove that there
must be 6 matches in which 12 participants are all different. (USAMO
1989)
ProofThe 20 members are represented by 20 vertices VI' V2' .. . ,
V20' with edges joining 2 participants who have played a match. Then

we get a graph G. There are 14 edges in G, and denote the degree of

the vertex by d i , i = 1, 2, ... , 20, d i ~ 1. According to Theorem 1,
d l +d 2 + ... +d 20 = 2 X 14 = 28.
20 Graph Theory

Delete d i - 1 edges incident to Vi. Since one edge might be deleted

twice, the number of deleted edges does not exceed

Cd t - 1) +Cd 2 - 1) + " '+ Cd 20 - 1) = 28 - 20 = 8.

So the graph G after deleting the edges, has at least 14 - 8 = 6

edges, and the degree of each vertex in G' is at most 1. Therefore the
12 vertices that are incident to the 6 edges are different. That is to
say, there must be 6 matches in which 12 participants are all different.
Example 7 Let 5 = {Xt, X2' • •• , xn} be a point set on the
plane, and the distance between any 2 vertices is at least 1. Prove that
there are at most 3n pairs of vertices, and the distance between 2
vertices in a pair is exactly 1.
Proof The n points are represented with n vertices, with edges
joining 2 points whose distance is 1, then we get a graph G . The
number of edges in G is denoted bye . Obviously, the vertex adjacent
to vertex x i in graph G is on the circle whose center is Xi' and whose
radius is 1. Since the distance between two points in set 5 is no less
than 1, there are at most 6 points of 5 on the circle. So d Cx i ) ~ 6.
Using Theorem 1 to graph G, we have

6n ~ 2e,

1. e. e ~ 3n. That is to say, the number of edges in graph G cannot

exceed 3n. So there are at most 3n pairs in the n vertices, whose
distance of every pair is exactly 1.
Example 8 There are n points on the plane. Prove that the

number of pairs of points whose distance is 1 would not exceed ~ +

J2
- n 1- 2
2 .
Proof Denote the n points by n vertices in graph G. Let V =

{ V t ' V 2' ••• , v ,, } is the vertex set in graph G. Join 2 vertices whose
distance is 1. According to Theorem 1,
 Degree of a Vertex 21

C i denotes a circle whose center is v i and radius is 1. The total

number of intersection points between any two of n circles does not

exceed2G) = nen - 1). On the other hand, if Vi is adjacent to both

v j and v k ' then viE Ck n C j. Therefore, v i regarded as an

intersection point of circles C l ' C 2 , ••• , C n is counted (deVi))

2

times, so

CD

According to the Cauchy Inequality, we have

(deVl)) + (deV2)) + ... + (deVn)) 2

~ -e
2
-e.
2 2 2 n

According to CD and (2), we have

~e 2 - e :::;;; n (n - 1) ,
n

1. e.

Then

e <-n J2 ~
+2- n . 2
~ 4

Exercise 2

1 Let G = ev, E) be a graph, where I V I = n and I E I = e.

2e
Prove that 0 :::;;; - :::;;; 6.
n
22 Graph Theory

2 G is a graph with n vertices and n + 1 edges. Prove that there

exists at least one vertex whose degree is no less than 3.
3 Does there exists a polyhedron such that the number of its
faces is odd and each face has an odd number of edges?
4 There are 15 telephones in a town. Can we connect them so
that each telephone is connected with the other 5 telephones?
5 There are 123 people attending an academic symposium, and
each person has discussed with at least 5 other participants. Prove that
there is at least one person who has discussed with more than 5
participan ts .
6 In a conference, each councilor does not know at most 3
person. Prove that all the councilors can be divided into 2 groups such
that each councilor in a group does not know at most 1 person.
7 2n people are getting together, and each one knows at least n
people. Prove that there must exist 4 persons such that they know the
persons sitting besides them when sitting around a round table (n :)0 2).
8 There are 9 people Vl' V 2 ' . . . , V9. V l has shaken hands with
2 people. V2 and V3 have shaken hands with 4 people respectively. V4'
VS, V6 and V7 have shaken hands with 5 people respectively. V 8 and V9

have shaken hands with 6 people respectively. Prove that there must
exist 3 people who have shaken hands with each other among
themselves.
9 There are 14 members in a tour group. When they are resting
on a hill, they would like to play bridge, and each of them has all
cooperated with 5 of them before. There is a rule that 4 people can
play together only when any 2 of them have never cooperated with
each other. If so, they cannot go on after 3 rounds. At this time,
there comes another tourist, and he has never cooperated with the
members in the tour group. Prove that there must be another round if
the new comer joins the bridge.
10 For a vertex set P consists of any n vertices on the plane, the
vertex set is stable if the distance between any 2 of them is fixed.
Prove that the vertex set P consists of n (n :)0 4) is stable if any 3 of
 Degree of a Vertex 23

them are not on the same line, and there are ~ n (n - 3) +4 pairs

whose distance are determined.

11 A cube with edge of length n is cut into n 3 unit cubes by
planes parallel to its sides. How many pairs of unit cubes whose
common vertices are no more than 2 are there?
12 There are 21 routes connecting the capital and other cities.
There is only one route connecting city A and one other city, and all
cities other than A are connected with some other places. Prove that
city A can be reached from the capital.
ChapterS Turan's Theorem

In 1941, a Hungarian mathematician Tunln brought forward his

famous theory so as to answer the question that if a graph with n
vertices does not contain a complete graph K m with m vertices as its
subgraph, how many edges can the graph contain at most? Then a new
branch of graph theory called "extremal graph theory" appeared. The
extremal graph theory is one of the most active branch of graph
theory. In 1978, a Hungarian mathematician B. Bollobas wrote a
book called "extremal graph theory" which is the authoritative book of
this branch.
Then we begin with the definition of a k-partite graph.
If the set of vertices V of a graph G can be decomposed into a
union set of k subsets which do not intersect each other. Namely,
k
V = UVi' V i
i= l
n Vj = 0, i =I=- j ,

and there is no edge whose two vertices are in the same subset. We call
such graph a k-partite graph denoted by G = eV 1 , V 2 , ••• , Vk ; E).
Fig. 3. 1 shows us a 2-partite graph which is also called bigraph.
Fig. 3.2 shows us a 3-partite graph.

' " ,,,

,
....... -- - - - ........ ,
,

AA
, '

--- ---- ---

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

Fig.3.1 Fig. 3. 2
 Turtin's Theorem 25

Clearly, any graph with n vertices is n-partite graph.

Suppose that there is a k-partite graph G = (VI' V 2 , ••• , Vk ; E)
with I Vi I = mi. A graph G is said to be a complete k-partite graph if
any two vertices of G satisfy u E Vi' V E V j , i -=I=- j , where i , j = 1,
2, ... , k, u and v are adjacent. We denote G by Kmj.m2 • . . • mk.

Fig. 3. 1 shows us a complete bigraph K 2 . 3' There are m 2 and m (m + 1)

edges in complete bigraphs K m. m and K m. m+1 , so the number of edges

of the graph is [~2 J, where n is the number of vertices in G. (Here

[x] denotes the largest integer no more than x.) Complete bigraphs
K m • m andKm,m+1 contain no triangle, In Theorem 1, we can see that
these two kinds of graphs contain the most number of edges among the
graphs without triangles.
Theorem 1 If a graph G with n vertices contain no triangle, the

largest number of edges of G is [n42 J.

Proof Assume that VI is the vertex with the maximum degree in
G, d (v I) = d and we denote d vertices adjacent to VI by

V n , V ll -1' ••• , V n -d+l.

Since G contains no triangle, any two of v n , V,, - I ' ••• , Vn - d+1 are
not adjacent. So the number of edges of G satisfies

e < d ( V I) + d ( V 2) + ... + d ( v" -d )

« n - d).d « n-~+df
n2
4'

Since e must be an integer, e < [~2 J.

The upper bound can be met only when G = Km. m if n = 2m and
G =Km,m + 1 ifn = 2m + 1.
We can also use induction to prove the theorem, and we leave it as
an exercise,
Example 1 Suppose that there are 20 vertices and 101 edges in a
26 Graph Theory

graph G. Prove that there must exist two triangles which have a
common edge in G.
Proof In general, we can replace 20 by more general number 2n
(n ~ 2). Now we use induction, if there are 2n (n ~ 2) vertices and
n 2 + 1 edges in a graph G, there must be two triangles which have a
common edge in G .
When n = 2, there are 4 vertices and 5 edges in the graph G .

Consider the complete graph K 4' there are (~) edges in K 4. It is not

difficult to prove that whatever edge we remove from K 4' there must
still be two triangles which have a common edge in G. So the theorem
holds when n = 2.
Suppose that the theorem holds for n = k (k = 2). Let G be a
graph with 2(k + 1) vertices which are denoted by VI' V2' . . . , V2k+2
and (k + 1)2 + 1 = k 2 + 2k + 2 edges. Note that

[ (2k : 2) 2 ] = [k 2 + 2k + 1] < P + 2k + 2.
According to Theorem 1, there must be a triangle in G. Without loss
of generality, we denote this triangle by L"> VI V2 V3 and d (v I) ~
d(V2) ~ d(V3).

If one of the vertices V4' VS' . . . , V 2k+2 is adjacent to two of VI ,

V2' V3' then we get two triangles with a common edge.

If each of V4' vs, .. . , V2k+2 is adjacent to at most one of V I , V2'

V3' then the number of edges which join the vertex set {V4' VS, . • . ,
to {VI' V2' V3} is less than (2k + 2) - 3 = 2k - 1. Then the
V2k+2 }

number of edges which join the vertex set {VI' V2} to {V4' VS' . . • ,
V 2k +2 } is no more than ; (2k - 1). We remove the vertices VI , V2 and

the edges adjacent to them from G to get G'. The number of vertices
in G' is 2k and the number of edges is

e' ~ k 2 + 2k + 2 - 3 - ; (2k - 1)

= k2 +; k - ~ ~ k 2 + 1, since k ~ 2.
 Turan's Theorem 27

By induction, there are two triangles with a common edge in G',

which are also in G. Then we complete the proof.
Example 2 We denote integer pair by (a , b) C1 < a, b < n, a 7":-
b) by S, where (a, b) is the same as (b, a). Prove that there are at

least ~: ( m - ~2 ) triples (a, b, c) satisfying (a, b), (a, c) and (b,

c) E S. (Asian-Pacific regional Mathematical Olympiad in 1989)

Solution We construct a graph G and denote the number i by v ; ,
wherei = 1, 2, ... , n,andCi,j) ES. Thevertexv;isadjacenttovj
if and only if Ci, j) E S. Then there are n vertices and m edges in G.

What we will prove is that there are at least ~: (m - n42

) triangles

in G .
Let the degree of vertex v; be d ; and denote the edge set of G by
E. If (v ; , v j ) E E, then there are d; + d j - 2 edges joining v;, v j
with all other n - 2 vertices. So there are at least d ; + d j - n pairs of
edges joining v;, v j with the same vertex . These edges together with
the edge (v ; , v j) form triangles. So there are at least d ; +d j - n
triangles containing (v;, v j) in G. Since we have counted every
triangle containing the edge (v;, v j) in G three times, there are

k =1- ~ (d ; +d j - n)
3 (Vi' 'lJj>EE

triangles in G. Since the degree d; of vertex v; has been counted 3

times in the above equation and the number of edges is m. So
1 n
K = - ( ~d7 - mn)
3 ;= 1

n
Since ~ d; = 2m, apply the Cauchy Inequality to CD and get
i= l

k ~ ; [ ~ (~d;) 2 - mn ]

= ; (4:2 - mn)
28 Graph Theory

= 4m(m _ ~)
3n 4'

Remark The problem derives from a question in graph theory: if

there are m edges and n vertices in a graph G, there must be no less

than ~ (m - ~2 ) triangles in the graph G .

Suppose that n = mk + r (k ~ 1, 0 ,,;;; r <

m). We denote the complete m-partite graph
K 'I, . "2 ' . ... "m by T m (n ), where n I = n 2 = .. . =
nr = k + 1, n r+ 1 = .. . = n m = k. We also denote
the number of edges of T m (n) by em (n ).
Fig. 3. 3 shows us T 3 (5), where e3 (5) = 8. The Fig. 3. 3

formula to calculate em (n) is :

em (n) = (n-k)
2 + (m - l) (k 2+l) , k = [nm J.
We leave it as an exercise.
Let G = (VI' V 2 , ••• , Vm ; E) be an m-partite graph G with n
vertices, and Pi = I v i I (~ :I Pi = n ) . We can show that the number
of edges in G is less than em (n) and the equality holds if and only if
G ~ T men). (We leave the proof as an exercise.) In other words,
T m(n) is the only m-partite graph with n vertices, which has the most
number of edges.
Clearly, any m-partite graph contains no K m + l • Furthermore,
Turan prove that T m(n) is the only m-partite graph with n vertices,
which contains the most number of edges and no K m+ I.
Theorem2 SupposeG contains noK m+ l , then e (G) ";;;em (n). The
equality holds if and only if G ~ T m (n).
This is Turan's theorem and we omit the proof. If you are
interested in it, you can read Graph Theory and Its Applications
written by J . A. Bondy and U. S. A. Murty.
Example 3 Let A I , A 2 , A 3 , A 4 , A s, A 6 be six points on a plane
and there are no three points on a line .
 Turrin's Theorem 29

(1) If we join the points randomly and get 13 line segments.

Prove that there must exist four points so that each of them is adjacent
to any other three points.
(2) If there are only 12 line segments joining these points. Draw a
graph to show that the conclusion of (i) is not true.
(3) Can the conclusion of (i) be modified so that there must exist
four copies of K 4? Give a counterexample or prove it.
Solution
(1)We can phrase the problem in the language of graph theory:
There are 6 vertices and 13 edges in a graph G, prove that the G
contains K 4.
It is easy to calculate e 4 (6) = 12 < 13. According to Theorem 2,
we know that G must contain K 4.
(2) Consider the complete 3-partite
graph K 2 • 2. 2 . According to Fig. 3. 4, we
choose any 4 vertices from K 2. 2. 2 and
there must be 2 vertices belonging to one :, A ,~:"'--I-.,4----lH""':"'~Ji 6,:
,
part. These two vertices are not adjacent. \ "

So the 4 vertices we choose arbitrarily

Fig. 3. 4
cannot form a K 4 •
Remark (1) Of course, we can use Theorem 2 to prove it and
there are many other ways. Here we list two different methods.
(1) Since the sum of the degrees of 6 vertices is 2 X 13 = 26, there
are at least 2 vertices whose degrees are 5 among the 6 vertices.
Otherwise, the sum of degrees is 5 + 5 X 4 = 25 < 26. Without loss of
generality, suppose that d (AI) = d (A 2 ) = 5, there are 9 edges
incident to A l or A 2 • According to Fig. 3. 5,
there are 13 - 9 = 4 edges joining A 3 , A 4 , A 5 ,
A 6 . Two ends of any of the four edges A6

together with Al , A2 can form a K 4 •

(2) Since there are 15 edges in a complete
graph with 6 vertices, we delete two edges.
We discuss the problem in two different cases. Fig. 3. 5
30 Graph Theory

CD If there are common vertices in two deleted edges as Fig. 3. 6

shows us, then A 2 , A 4 , A s, A 6 form a K 4.
(2) If there is no common vertex in two deleted edges as Fig. 3. 7
shows us, then AI , A 3 , A s , A 6 form a K 4.

A,
A, A~A'
AJ A,

Fig. 3. 6 Fig. 3. 7

(3) According to the above two cases, let us begin our discussion:
In case CD, there are six K 4 : (AI, A 4 , As, A 6 ), (A 2 , A 4 , As, A 6 ) ,
(A 3 , A 4 , A s , A 6 ) , CA 2 , A 3 , A 4 , A s), CAl' A 3 , A 4 , A s), (AI,
A3 , As, A 6)'
In case (2) , there are four K 4 : (AI, A 3 , A s , A 6 ) , (AI' A 4 , As,
A 6), (AI, A 3 , A 4 , As), (AI, A 2 , A4 , As). So there must exist four
copies of K 4 •

Example 4 In a simple graph with eight vertices, can you find the
maximum number of edges of a graph which contains no quadrangle?
(The quadrangle consists of four vertices A, B, C, D and four edges
AB, BC, CD, DA . ) (China Mathematical Olympiad in 1992)
Solution The maximum number of edges is 11.
First, Fig. 3. 8 shows us a graph with 8
vertices and 11 edges, which contains no
quadrangle.
Next, we will prove the fact that if any
simple graph contains 12 edges, the graph must
contain a quadrangle. Fig. 3. 8

First, we must point out two obvious facts .

(a) Let A, B be two vertices (A =F B). If A is adjacent to vertices
C I , C z , ... , C k and B is adjacent to at least two of vertices {C, . .. ,
 Tunin's Theorem 31

Ck }. the graph must contain a quadrangle.

(b) If there are 5 edges joining 4 vertices. the graph must contain
a quadrangle.
Suppose that a graph contains 8 vertices and 12 edges. which
contains a quadrangle. A is one of the vertex which is incident to the
most vertices.
(1) Suppose that A is incident to s ~ 5 edges and we denote the
edge set incident to A by 5 and the vertex set other than A and 5 by
T. According to (a) and Cb), we know that the number of the edges

joining the vertices in 5 is no more than [ ~ J. The number of the

edges joining the vertices in T is at most ( I: I ). The number of edges

joining the vertices in 5 and T is at most IT I. SO the sum of edges is

at most

When s ~ 5, the number of edges is less than 12. A contradiction.

(2) There are four edges from A, namely: AAj(j = 1,2,3,4).
Let another three vertices be B l , B 2 , B 3 • There are at most two edges
joining {A 1, A 2 , A3 , A4} and at most three edges joining {B 1, B 2 ,
B3 }. There are at most three edges joining these two vertex sets. Since
there are 12 edges in the graph G, the number of edges in each of the
three groups is 2, 3, 3 respectively. Without loss of generality, let the
three edges in the third group be A j B j Cj = 1, 2, 3). Since there are
two edges without a common vertex in the first group, there is one
edge joining A 1, A 2 , A 3 , denoted by A 1A 2. So A1A2B2Bl is a
quadrangle. A contradiction.
(3) There are three edges with an end point at A and every vertex
is incident to three edges. Suppose that A, B are not adjacent. We
denote the three edges with an endpoint at A or B by AA j . BB j (j =

1, 2, 3). respectively. So according to (a), there are at most one

32 Graph Theory

common vertex between {A 1 , A 2 , A 3 } and { B 1 , B 2 , B 3} '

If there is no common vertex between them, there is at most one
edge joining two of the three vertices . There are at most three edges
joining the two three-vertex sets, the number of edges is at most 11. It
is a contradiction.
If there are only one common vertex between the two three-vertex
set, we consider the eighth vertex. According to the Pigeonhole
Principle, among the three edges having the common vertex as an end
there must be two edges which have ends in one three-vertex set. Then
we get a quadrangle. It is a contradiction.
In conclusion, we have proved that the graph containing 8 vertices
and 12 edges must contain a quadrangle. So the maximum number of
edges is 11.
Example 5 G is a simple graph with n vertices. If G contains no
quadrangle, then the number of edges is

e < ! n C1 + J 4n - 3 ).

Solution We denote the vertex set of G by V = {v l' V 2 ' ••• ,

V n }. For any vertex vi E V, the number of vertex pairs Cx , y)

adjacent to Vi is (dCv2
i ))
. Because G contains no quadrangle, when V i

is changing in the V, all the vertices pairs ( x , y) are distinct.

Otherwise , vertices pairs Cx, y) are counted in both (dC; ;)) and

(d C; j ) ), respectively . Then V i ' x , V j , y construct a quadrangle. So

In virtue of the Cauchy inequality,

 Turan's Theorem 33

So

2
--;;e
2
- e~ (n)
2 '

e 2 - ~e - ~n2en -1) ~ O
2 4 "'" .

We solve the above inequality and get

e ~ ~ (1 + /4n - 3 ).

Remark This problem tells us an upper bound of the number of

edges ·of graphs which contain n vertices and no quadrangle . But it is
not the maximum number . For general n, we can do a further research
into its maximum number . Example 4 has showed when n = 8, the
maximum number is 11.
Example 6 There are n vertices and l edges in a graph. Then n =

q2 + q + 1, l ~ ~qeq + 1)2 + 1, q ~ 2, q EN .

We know that any four points in the graph do not lie on one plane
and every point must lie on at least one line . So there exists a point
that lies on at least q + 2 lines. Prove that the graph must contain a
quadrangle in the space, consisting of four points A, B, C, D and
four lines AB, BC, CD, DA. eChina Mathematical Competition in
2003)
Solution The condition that any four points cannot lie on a plane
is to ensure that there are no three points on a line. So in terms of
graph theory, we only need to prove that the graph contains a
quadrangle. To solve this problem, we need to use the idea of Example
5, but we cannot use it directly. Consider the removal of the de VI )
vertices which are adjacent to VI CdCV1) ~ q + 2). There will be
( n - deVl»)
2 pairs of vertices left.
34 Graph Theory

As in Example 5. when the graph contains no quadrangle .

Similarly.

(de v ) - 1)
2::
n

i~2
'2 = 2l - n +1 - d ev ) .

Then use the Cauchy equality.

[n - devI)J [ n - de v I) - 1J
2

~ 21 { ~
n
[n - de V i ) J2 - ~ [ n - de V i ) J
n }

~ ~ {n ~ 1 [2l - n +1 - d ev I) J2 - [ 2l - n +1 - d eVi )] }.

That is to say.

en - D[n - devI) J [n - de v I) - 1J
~ [2l - n + 1 - de v I) J [2l - 2n + 2 - de vI )J
~ [q3 + q2 - d ev I) + 2J[q 3 - q + 2 - de v I ) J
= [ nq - q + 2 - de v I)J [ nq - q - n + 3 - de v I) ].

It contradicts the fact that

and

q [ n - d ( V I) - 1J ~ nq - 1 - n +3 - d ev ).

So the graph must contain a quadrangle .

As an application of Tunin's Theorem. we give an example in
geometry.
In a point set 5 on the plane. let the maximum distance of any
two points be d . If d is a finite number. we call d the diameter of the
vertex set 5 .
Let 5 = {x l . X2 ' •••• xn} be a point set which consists of n
 TUriin's Theorem 35

points and whose diameter is 1. The n points determine the distance of

two points in the G) point pairs. For a number d between 0 and 1,

we can ask the following question: in a vertex set S = {x 1 , X 2 , ••• ,

x"} whose diameter is 1, how many points pairs are there such that the
distance between two points in the point pairs is more than d? Here,

we discuss only the special case when d = 1.

First, when n = 6, then S = {Xl' X2' X3' X 4' X S' X 6}. We put
them on the vertices of a regular hexagon so that the distance of two
points in the point pairs (X l ' X4), (X2' Xs), (X 3 ' X6) is 1. Fig. 3. 9
shows us that the diameter of S is 1. It is not difficult to find that the
distance of two points in the point pairs ( X l ' X 3 )' (X2' X4)' ( X3 '

X s ), (X4' X6)' (X S , Xl)' (X6' X2) is1. So there are 9 point pairs in

the point set S whose diameter is 1. The distance of these 9 points

. .IS more t h an 12
paIrs :2.

X2Q---~-----bX,

Fig. 3. 9 Fig. 3.10

But 9 is not the best answer for 6 points. Suppose we arrange the 6
points as Fig. 3. 10 shows us, namely, the vertices Xl' X 3 ' X s form a
regular triangle with edges of length 1. The vertices X2' X4' X6
construct a regular triangle the length of whose edges of length o. 8.
The center of the new triangle coincide with the center of D X l X 3 X S
and the edges of the new triangle are paralleled to those of D X l X 3 X S'
then the distance between two points in the point pairs other than (Xl'
36 Graph Theory

X2)' (X 3 ' X4)' ( X s' X6) is more than 1. So we have (~) - 3 = 12

point pairs and the distance between two points in the point pairs is

more than 1. In fact, it is the best answer we can get. For the

general case, Theorem 3 gives the a solution to the question.

Theorem 3 Let 5 = {X l ' X2' ••• , xn} be a point set on the plane
whose diameter is 1, then the maximum possible number of the point

pairs the distance of which is more than 1 is [~ ] . For every n,

there exists a point set {Xl' X2' .. • , Xn} whose diameter is 1, and

there are exactly [;3] point pairs such that the distance of two points
.
In each · ·IS more t h an
paIr J2
2.
Proof Draw a graph G: we denote n points by n vertices. Two
vertices are adjacent if and only if the distance of two vertices is more

than 1. First, we prove G contain no K 4 •

For any four points on the plane, their convex hull could have
only three case: a line segment, a triangle, or a quadrangle, as
Fig. 3.11 shows us. Clearly in every case there is an angle L X iXj Xk no
more than 90 For the three vertices X i ' X j , X k' it is impossible that
0
•

the distance of any two vertices of the three points is all greater than
J2
2 and less than or equal to 1. Here, we denote the distance between

o......~'--c~--co>----o
x, ~I Xk x,

Fig. 3.11
 Turan 's Theorem 37

x and y by d (x, y), d ( X j' Xk ) is more than 1 for all j, k and

L X iXjXk ~ 90°, then

d(Xi' Xk ) ~Vd2(X i ' X j ) + d 2 ( Xj ' Xk) > 1.

Since the diameter of the vertex set S is 1, among any four

vertices in G there is at least one pair whose vertices are not adjacent.
It means that G contains no K 4'

According to Theorem 2 the number of edges of G is no more than

e 3 (n ) = [n3
3
J.
We can construct a vertex set {x l ' X 2' ... , X n} which contains
3
[ n3 ] vertex pairs so that the distance of two vertices in each pair is

1.
more than The construction is as follows. Choose r so that 0 <

r < ! -1). (1 Then draw three circles whose radii are all 1 and the

distance of any two of their centers is all 1 - 2r. As Fig. 3.12 shows us,
weputxI' X2 ' ... , X[ -n inacircle, x [t J+ I' ... , x [~ J in another
circle and x [~ ] +I' . .. , x" in the third circle so
that the distance of XI and x" is 1. Obviously,
the diameter of this set is 1. Furthermore, d ( x i ,

Xj ) > 1 if and only if Xi and Xj belong to

3
different circles. So there exist exactly [n3 ] x,

vertex pairs (Xi' Xj ) such that d (Xi' Xj ) > 1. Fig. 3. 12

Exercise 3

1 Prove that if a bigraph G (X, Y; E) is 8-regular, then

38 Graph Theory

1X 1= 1Y I.
2 Prove by induction Theorem 1.
3 Draw a simple graph which contains 20 vertices, 100 edges and
no triangle .
4 Prove that if there are 2n + 1 vertices and n 2 + n + 1 edges in a
simple graph G, then G must contain a triangle.
S (1) The number of edges in a complete m-partite graph with n

verticese m(n) = C~k) + (m _ 1)(k ;1), wherek = [~ J.

(2) Let G be a complete m-partite graph with n vertices . Then
e(G) <em (n).
6 There are n students from each of the two countries X and Y.
Every student from X has danced with some (but not all) students
from Y and every student from Y has danced with at least one student
from X. Prove that it is possible to find two students x, X l from X
and two students y, yl from Y so that x has danced with y and has not
danced with yl while X l has danced with yl and has not danced
with y .
7 Use Tunin's Theorem to prove Problem 9 in Exercise 2 .
8 If a graph G contains n (n > 5) vertices, then G and its

complement G contain at least 214n(n - l)(n - 5) triangles in all.

9 X is a set of n elements and set its m k-subsets A l , A 2 , • • • ,

A m red k-subset. Prove that if m

. > (k - 1)( n
k2
- k) + k •
(k n )
_ 1 '

then there must exist a (k + 1) -subset of X so that all k -subsets are red
k-subset.
10 Let K 3 • 3 be a graph. Prove that a graph with 10 vertices and
40 edges must contain a K 3.3.
11 In a circular city whose radius is 6 kilometers, there are 18
police cars making the rounds. They use wireless equipments to
communicate with each other. If the wireless is effective within 9
kilometers . Prove that at any time , there must exist two cars, each of
 Turan's Theorem 39

them can communicate with five other cars.

12 There are 2n points in the space and any four of them are not
on one plane. If there are n 2 + 1 line segments joining them, then
these line segments can form at least n distinct triangles.
Chap ter 4 " -- -
Tree

Among all kinds of graphs , there is a simple but important graph ,

which is called " tree" . Tree is very important because it is widely used
not only in many other fields but also in graph theory itself. In graph
theory , tree is a very simple graph, so when we discuss some general
graph theory conjectures, we may study tree first.
First , we introduce several concepts.
Suppose, in graph G , a sequence consists of different edges:

If edge e i = (v i- 1 , V i ) , i = 1, 2 , . .. , m , then we call the sequence a

chain from Va to v m ' The number m is called the length of the chain.
Va and Vm are its ends. We denote the chain by VaV \ . . . v m •

A path is a chain where the v i are all distinct

If the ends Va and v m of a chain are the same , then the chain is
called a cycle CD .
In Fig. 4.1 , e1' e2' e3' e4 ' es form a chain , whereas e1' e2' e3
constitute a cycle . If for any two vertices u and v of graph G , there is
a chain from u to v, then G is called a connected graph. Otherwise , G
is called a disconnected graph.
The graph in Fig. 4.1 is a connected graph. The graph in Fig. 4 . 2
is a disconnected graph.
Now , let us give the definition of a tree.

CD In this book , "cycle" is actually a " closed chain" . It is different from the " cycle" in
general graph theory. T he "cycle" in graph theory is a "closed chain"vovI . . . v.,,(VO =v m ),

in which vertices V I ' V2 , ... , v." are distinct .

 Tree 41

v,
Fig. 4. 1
l~
v, v,

Fig. 4. 2
v,

A connected graph which contains no cycle is called a tree. We

usually denote a tree by T.
According to the definition of tree, tree is
obviously a simple graph. A tree with eight
vertices is shown in Fig. 4.3. Clearly, a graph
without cycles must be composed of one or
several trees whose vertices are disjoint. We
Fig. 4. 3
call such graph a forest.
The graph in Fig. 4. 4 is a forest, which 1S composed of three
trees. The vertex with degree 1 is called a pendant vertex (or leaf).

Fig.4.4

Theorem 1 If a tree T has no less than 2 vertices, then T

contains at least two pendant vertices.
Proof 1 Suppose we start from some vertex u, along the edges of
T, every edge can only be passed by once. Since a tree has no cycle, it
cannot return to the vertices which have been passed. It means that
each vertex can be passed at most once. If the vertex we pass is not a
pendant vertex, because its degree is more than 1, we can continue.
But the number of vertices of T is finite, so it is impossible to continue
forever. If we cannot continue further at v, then v is a pendant
42 Graph Theory

vertex.
We start from a pendant vertex v and we can end up at another
pendant vertex v'. So tree T has at least two pendant vertices.
Proof 2 Assume that f1. = uv 1 V 2 ••• V k V is the longest chain of T,
we can show that d (u) = d (v) = 1, i. e. u, v are pendant vertices.
In fact, if d (u) ~ 2, then there exists vertex w (w =I=- VI) adjacent
to u. If w is one of V2' ..• , V k' v, then a cycle occurs. It
contradicts the definition of tree. If w is different from V2' . . . , V k ,
v, then wuv 1 ••• V k V is a chain longer than f1.' aeon tradiction. So
d (u) = 1. Similarly, d ( v) = 1. Hence tree T has at least two
pendant vertices.
Remark Proof 1 is "by construction" and Proof 2 is by the
method of "the longest chain". They are two very important methods.
Theorem 2 Let n be the number of vertices of T, then the
number of edges e = n - 1.
Proof We prove by induction on n .
When n = 1, e = 0, the conclusion holds. Suppose the conclusion
holds for n = k. Let T be a tree on k + 1 vertices (k ~ 1) . By Theorem
1, T has at least two pendant vertices. Let v be one of them, then
deleting v and its adjacent edges. We can get tree T' on k vertices.
According to the induction hypothesis, T' has k - 1 edges. So the
number of edges of T is k, hence the conclusion holds for any natural
number n.
Theorem 3 Let T be a graph with n vertices and e edges. Then
the following three propositions are equivalent:
(1) T is a tree;
(2) T has no cycle, and e = n - 1;
(3) T is connected, and e = n - 1.
Proof (1)~(2):

Suppose T is a tree. According to the definition of T, T has no

cycle. By Theorem 2, e = n - 1. Hence (2) holds.
(2)~C3) :
It suffices to show that T IS connected and T is disconnected.
 Tree 43

Suppose T has k connected branches. Since each connected branch has

no cycle, hence each connected branch is a tree. If the i th branch has
P i vertices, according to Theorem 2, the ith branch has P i - 1 edges.
So
e = (Pl - 1) + ... + (Pi -1) = n - k :;:;;; n - 2.

This contradicts e = n - 1. Hence T is connected .

(3)-(1) :
It suffices to show that T has no cycle, then T is a tree. When
n = 1, the conclusion holds. Suppose n ~ 2, then T must have a
pendant vertex. Otherwise, since T is connected and n ~ 2, the degree
of each vertex of T is no less than 2. Hence

This contradicts e = n - 1.
Now by induction on n, we will show that T has no cycle.
When n = 2, e = 1, T is has no cycle.
Suppose the proposition holds for n = k. T is a graph on k +1
vertices, v is a pendant vertex. Deleting v and its adjacent edges, we
obtain graph T'. By induction and T' has no cycle. Add v and its
adjacent edges and we can obtain graph T again. Hence the
proposition is correct.
Theorem 3 shows that any two conditions of "connected",
"having no cycle" , "e = n - 1" imply that T is tree. Hence they all can
be the definition of a tree.
Example 1 If T is a tree, then
( 1) T is connected, but after deleting any edge of T, the
obtained graph G is disconnected;
(2) T has no cycle, but after adding any edge, the obtained graph
G contains only one unique cycle.
Conversely, if T satisfies (1) and (2), then T is tree.
Proof (1) If graph G is connected, then G is still a tree, so G has
n - 1 edges, equal to the number of edges of T, a contradiction.
44 Graph Theory

(2) If G has no cycle, then G is still a tree. So G has n - 1 edges,

equal to the number of edges of T . It is impossible, so G contains a
cycle. Clearly, G contains only one cycle.
If (1) and (2) hold, then T is tree. We leave it as an exercise.
This exercise characterizes a feature of tree. Among all graphs
with given vertices, tree is a connected graph with the least number of
edges. Tree is also graph without cycle and with the most number of
edges. From this, in any graph G, if e < n - 1, then G is
disconnected; if e > n - 1, then G must have a cycle.
Another feature of tree is also very useful, as shown below.
Example 2 If T is tree, then there is only one chain between any
two vertices of T. Conversely, if there is only one chain between any
two vertices of T, then T must be a tree.
Proof If T is a tree, then T is connected, and there is at least
one chain between any two vertices. Since T has no cycle, there is
only one chain between any two vertices of T.
Conversely, if there is only one chain between any two vertices of
T, then T is clearly connected and T has no cycle. Otherwise, there
would be at least two chains between any two vertices on a cycle, this
introduces a contradiction.
Example 3 There are n cities, each city can call another city
through some intermediate cities. Prove that there are at least n - 1
direct lines, each of which connects two cities. (Hungary Mathe-
matical Olympiad)
Proof Draw a graph G and denote n cities by n vertices. If there
is a direct line between two cities, then join two corresponding
vertices. By hypothesis, G is a connected graph. So the number of
edges of G is no less than n - 1 , then there are at least n - 1 lines
connecting every two cities.
This problem can be considered III the following way. If the
connected graph G obtained has a cycle, we delete an edge on the
cycle and obtain a graph G 1 • The number of edges of G 1 is less than
that of G by one, but G 1 is still connected . If graph G 1 still has a
 Tree 45

cycle, we delete an edge on the cycle and obtain a graph G 2 , and so on

until the graph obtained has no cycle. The graph is certainly a tree. It
has n - 1 edges, so graph G has at least n -1 edges.
The above tree obtained is called the generating tree of graph G.
Adding several edges to the generating tree, we can get the original
graph.
Example 4 In a certain region, twenty players of a tennis club
have played fourteen singles. Every person plays at least once. Prove
that there are six pairs singles, in which twelve players are distinct.
Proof This question has occurred in Chapter 2. Here we prove it
from the viewpoint of tree.
Denote twenty players by twenty vertices. If two persons have
played a game, add an edge between them. There are fourteen edges
in all, each vertex is incident to at least one edge. Our conclusion is
equivalent to: it is possible to find six edges so that any two of them
are not adjacent.
Suppose the graph has n connected branches, among which the
ith branch has Vi vertices, ei edges. Clearly, ei ~ Vi -1, so
n
" n
.z=e i ~
i= l
.z= (Vi -
i =l
1) .z=Vi - no
i=l

n "
But .z= e
i= l
i = 14, .z= Vi
i= l
= 20, so 14 ~ 20 - n, n ~ 20 -14 = 6. Since

every vertex is incident to at least one edge, it is impossible that there

exists a connected branch, which contains only one isolated vertex.
Hence choose an edge from every connected branch, which promises
that they are not adjacent and the number of edges is at least six. We
finish the proof.
From Fig. 4. 5, the number of vertices is twenty, the number of
edges is fourteen. Choose arbitrarily seven edges, then there must be
two edges that are in the same connected branch and adjacent. Hence
six is the best possible.
46 Graph Theory

Fig.4.S

Example 5 Given 2n vertices on a plane, cover them by some

circles. Prove that if each circle covers at least n + 1 vertices, then any
two vertices can be joined by a zigzag line on a plane, and this line is
fully covered by the cycles.
Proof We denote 2n points by 2n vertices. If there is a circle,
which covers two points, then adding an edge between these two
vertices, we obtain a graph G. By hypothesis, the degree of each
vertex is no less than n . Each edge in the graph can be fully covered by
a circle. So we only need to show that G is connected.
If G is disconnected, then there exists a connected branch G I ,
which has at most n vertices. We have d (v) :0::;; n - 1 for every vertex v
of G I . A contradiction. So G is connected.
Example 6 >
3) table tennis players play several single
n (n
games . The opponents of any two players are different. Prove that we
can remove one player so that the opponents of any two players in the
remaining players are still different. (China Mathematical
Competition in 1987)
Denote n players by n vertices VI' V2' . . . , v". If the
Proof
proposition is not true, i. e. any player cannot be removed. For player
v k (1 :0::;; k :0::;; n), since he cannot be removed, after removing v k' we
can always find a pair of players v j and v j ' whose opponents are the
same (if there are several pairs, choose anyone pair) . This indicates
that the opponents of v and v j are only different due to v k. Without
j

loss of generality, we suppose that v i has played a game with v k' but
v j has not played a game with v Add an edge between v j and v j ,
k.

label k . We obtain a graph with n vertices and n edges, and there are
n edges with n different numbers.
The graph with n vertices and n edges must have a cycle. Suppose
 Tree 47

Vi , ' Vi 2 , is a cycle. Along the cycle, going through an edge

••• , V ik

means adding or subtracting a player from the players and the added
player and the subtracted player are different. Going through a cycle,
we can return to Vi } , i. e. after adding or subtracting players from the
players who have played with Vi, ' the result is the same as the original
opponents of V i, • A contradiction.
Hence, there is at least one player who can be removed in n players.
Example 7 In a lecture, there are five mathematicians. Each of
them took a nap twice and every two of them took naps at the same
time. Prove that there must be three persons who took naps at the
same time. (USA MO 1986)
Proof Denote ten naps of five mathematicians by vertices VI , V2'
. . . , VlO' add an edge between V i and V j if and only if i th and j th
took naps at a common time. We obtain a graph.
By hypothesis, every two mathematicians took naps at a common

time, so graph G has at least G) = 10 edges. But graph G has ten

vertices , so G must have a cycle.

Let the cycle be V i } Vi 2 ••. V i k V i , ' suppose V i , waked up first. Then
as soon as Vi , waked up, where Vi 2 and Vi k were still taking a nap. This
proves that there must be three persons who took naps at a common
time.
Example 8 There are 1990 residents in a district. Every day each
of them tells the news they heard yesterday to all his acquaintances and
any news can gradually be known to all. Prove that we can select 180
residents so that they are informed of some news at the same time and
after at most ten days, this news will be known by all residents .
Proof Denote these residents by vertices, two adjacent vertices
means that the corresponding residents are familiar. We obtain a
graph G with 1990 vertices.
By hypothesis, G is connected . Let this graph be a tree T 1990 •
Otherwise replace it by its generating tree. In the tree T 1990 , choose
the longest chain, and denote it by
48 Graph Theory

el)
V 11 v(ll
"
Choose v~l) as a representative of the residents, and delete edge
(v\P, vW). Then T 1990 is divided into two trees. In the first tree, the
distance of every vertex v and v~p is no more than 10. Otherwise, in
tree T 1990 ' the chain from v to v~1) is longer than the chain from v~1) to
v~1). Hence the news known to the representative v\P will be known to
the persons in the first tree within ten days. The latter tree also have a
longest chain, note that
V (2) V (2)
11 m ,

where m ~ 1990 - 11 = 1979. Similarly, choose v\f) as a representative

of the residents, and delete edge (v\f) , vg». The tree is divided into
two trees again.

Fig. 4. 6

Continue in this way. After v~;) (i ~ 179) has been chosen, the
remaining tree has at most 11 vertices. At the same time, the number
of representatives is i + 1 ~ 180. The proposition holds. Otherwise we
can find the representatives
(1 79)
... , V11

Each representative can tell a news to the residents in the district

in ten days .
At the end, there is only one tree, which has at most
1990 - 11 x 179 = 21

vertices. Letv1v2 ... Vk be its longest chain. Ifk ~ 11, then choose V11
as the 180th representative v~1 80) . If k < 11, then choose V 1 as the
180th representative v~1 80) In this way, the 180 representatives
are obtained
 Tree 49

... ,
as required.

Exercise 4

1 If the number of vertices of a connected graph G is no less

than 2, then there exists at least two vertices in graph G. After
removing the two vertices and their adjacent edges, the graph is still
connected. (A graph without vertices is also considered as a connected
graph)
2 On a coordinate plane, eleven vertical lines and eleven
horizontal lines constitute a graph. The vertices of the graph are the
points of intersection of the vertical and horizontal lines (lattice
points), the edges are the vertical and horizontal segments between
two lattice points. How many edges at least should be removed so that
the degree of each vertex is less than four? How many edges at most
should be deleted so that the graph keeps connected?
3 If graph G has n vertices and n - 1 edges, then graph G is a
tree. Is this proposition correct? Why?
4 A tree T has three vertices of degree 3, one vertex of degree 2
and other vertices are all pendant vertices. (1) How many pendant
vertices are there in T? (2) Draw two trees which satisfy the above
requirement of degrees.
5 A tree has n i vertices whose degrees are i, i = 1, 2, ... , k. If
the numbers nz, ... , n k are all known, what is nl? If n r (3 ~ r ~ k)

is not known, and n j (j =F r) is known, what is n r?

n

6 Let d 1 , d 2 , ••• , d n be n positive integers, n ~ 2, and ~ d i

i=1

2n - 2. Prove that there exists a tree where the degrees of its vertices
are d 1 , d 2' ••• , d n •
7 There are n (n ~ 3) segments on the plane where any three of
50 Graph Theory

them have a common end, then these n segments have a common end.
8 Given a number table with n rows and n columns, every two
rows are different. Prove that there must exist a column, after
deleting it, every two rows are still different.
9 Let V be the set of all the vertices of G, E the set of all the
edges of G. Prove that if I E I ~ I V I + 4, then there must be two cycles
which have no common edges. (Posa Theorem)
10 In an evening, 21 persons make phone calls, someone finds
that these 21 persons call 102 times and every two persons call at most
once . He also finds there exist m persons, the first one calls the
second, the second calls the third, ... , the (m - 1) th calls the m th ,
and the m th calls the first. He does not tell the value of m, only says
m is odd. Prove that there exist 3 persons among 21 persons, each of
the three persons calls each other.
11 A country has a number of cities. There are roads connecting
the cities and each city has 3 roads connected to it. Prove that there
exist roads forming a cycle whose length is not divisible by 3.
Chap ter 5 Euler's Problem

Euler's problem arised from the famous seven-bridge problem.

Konigsberg is located in Europe and Pregel River with a beautiful view
runs across the city. In the river, there are two islands A and D. On
the river, there are seven bridges joining two islands and the riversides
Band C (Fig. 5. 1) .

Fig. 5. 1

Question: Can a traveler go through every bridge once and once

only? This is the famous Konigsberg seven-bridge problem.
Clever Euler, using his insight , found that it was an interesting
geometry problem. In 1936, he published a thesis named "seven
bridges of Konigsberg" and solved the seven-bridge problem. It was
generally considered as the first thesis in graph theory. Euler changed
Fig. 5. 1 into a graph G as shown in Fig. 5. 2. Islands A, D and
riversides B, C are four vertices and seven edges c
in graph G denote seven bridges. Hence the
seven-bridge game becomes the following
A c:-E'------¢D
problem: Can the graph be drawn without lifting
one's pen and traversing every edge only once?
Drawing a graph with a continuous line does not B
necessarily have to come back to the original Fig. 5. 2
52 Graph Theory

starting point, that is to say, the problem of drawing a graph with a

continuous line (or in the end, coming back to the original starting
point) is equivalent to the problem whether the graph is a chain (or a
cycle) .
If G is a chain from Vl to V n+ l , then each vertex V i (i = 2, 3, ... ,
n) different from Vl and V n +l are even vertices. Because to vertex Vi'

if there is an edge entering V i ' then there is also an edge leaving Vi.

The entering and . leaving edges cannot be repeated. Hence there are
always double edges adjacent to V i . Therefore, graph G has at most
two vertices with odd degrees, i. e. Vl and V n +l' If G is a cycle, by the
above reasoning, Vl and V n +l are also even vertices. Hence , if G is a
chain (cycle), then the number of vertices with odd degrees in G is
equal to 2(0). This is the necessary condition that G is a chain (cycle) .
In other words, if the number of vertices with odd degrees in G is
more than 2, then G is not a chain and it cannot be drawn without
lifting one's pen .
In Fig. 5. 2, A, B, C, D are all odd vertices, hence this graph is
not a chain and it cannot be drawn without lifting one's pen. That is to
say, it is impossible for a traveler to go through each bridge once and
only once.
The following is "the solution to Euler's problem".
Theorem 1 A finite graph G is a chain or a cycle if and only if G
is connected and the number of vertices with odd degree in G is equal
to 2 or O. When and only when the number of vertices with odd
degrees in G is equal to 0, the connected graph G is a cycle.
Proof The necessity has been proved above, and next we prove
the sufficiency.
First, we prove: If G is connected and the number of odd vertices
is 0, then G must be a cycle .
Start from any vertex Vo of G and go to Vl through the adjacent
edge el . Since d (V l ) is even, go to V2 through the adjacent edge e2
from Vl , and continue in this way. Each edge is chosen only once and
we must return to Vl after several steps, so we obtain a cycle f1. 1 :
 Euler's Problem 53

VOV1 ••• VI).

If is G itself, then the proposition holds. Otherwise we can

(.11

find a subgraph G I by deleting /11 in G, then every vertex in G I is also

an odd vertex. Since G is connected, then there must exist a common
vertex u in /11 and in G I and a cycle /12 from u to u in G I . So /11 and /12
still constitute a cycle. Repeat the above process. Since G has only a
finite number of edges in all, the cycle finally obtained is graph G
itself .
Now we prove the second case. Suppose G is connected, and the
number of odd vertices is 2. Let u, V be the two odd vertices, add an
edge e between u and v, we get graph G ' . Hence the number of odd
vertices is 0 in G' , so G' is a cycle. Therefore, after deleting e, G is a
chain.
We call a cycle in a graph an Euler tour if it traverses every edge
of the graph exactly once. A graph is Eulerian if it admits an Euler
tour.
Furthermore, there is the following question: If the number of
odd vertices in a connected graph G is not 0 or 2, then by lifting one's
pen how many times can G be drawn? We know that the number of
odd vertices is even, so we have the following conclusion.
Theorem 2 If G is connected and has 2k odd vertices, then graph
G can be drawn by lifting one's pen k times and at least k times.
Proof Divide these 2k odd vertices into k pairs: VI' V'I; V2'
I
V2;••• ; V k' V Ik' a dd an e d ge e i b etween v i an dv' d a b tam
i an . G' .
Graph G' has no odd vertex, so G' is a cycle. Delete these k added
edges, then this cycle is divided into at most k parts, i. e. k chains.
This indicates that G can be drawn by lifting one's pen k times.
Suppose G is divided into h chains, each chain has at most two
odd vertices. Hence 2h ;?: 2k, i. e. h ;?: k. Graph G can be drawn by
lifting one's pen at least k times.
Example 1 Fig. 5.3 is a plane graph of a building and there is a
living room. After entering the front door into the living room, there
are four other rooms . If you enter from the front door, can you enter
54 Graph Theory

all the rooms (including the living room) through all the doors and go
through each door only once?

front door

Fig. 5. 3 Fig. 5. 4

Proof The answer is no.

Consider five rooms, the area outside the front door and that
outside the back door can be denoted by vertices, add an edge between
the corresponding vertices if these places are connected by doors and
we get a graph G (see Fig. 5. 4). In graph G, the number of odd
vertices is 4, so G is not a chain. So the answer is no.
Example 2 Fig. 5. 5 can be drawn by lifting one's pen 5 times and
Fig. 5. 6 can be drawn by lifting one's pen 4 times .

Fig. 5. 5 Fig. 5. 6

Example 3 On an 8 x 8 chessboard with 64 black and white

squares we move a knight. In whatever direction the knight moves, is
it possible to make the knight move to all the squares and each square
only once? (A knight moves from a corner square of a rectangle with
2 X 3 black and white squares to another corner square diagonal to the
starting position. )
Proof Fig. 5.8 gives us an answer of the question.
 Euler's Problem 55

56 41 58 35 50 39 60 33
47 44 55 40 59 34 51 38
42 57 46 49 36 53 32 61
45 48 43 54 31 62 37 52
20 5 30 63 22 11 16 13
29 64 21 4 17 14 25 10
6 19 2 27 8 23 12 15
1 28 7 18 3 26 9 24

Fig. 5.7 Fig.S.8

Solving this question, we often try the following four methods.

1. Each time we place the knight in a position where the number
of squares where the knight can move to (though it has not moved) is
the least, i. e. first move to the squares where the knight has fewer
places to move to, last move to the squares where the knight has move
places to move to.
2. Divide the chessboard into several parts and find a Hamiltonian
chain in each part (see Chapter 6) and join them together.
3. Find several cycles on the chessboard and join them.
4. Add edges to a smaller chessboard and get a Hamiltonian chain
of the whole chessboard.
Example 4 In Fig. 5. 9, two ants lie in A

positions A, B. Ant A tells ant B: "Let us

compete and see who can first creep nine edges B E

of this graph and arrive at E first." Ant B

agrees . Suppose the two ants have the same
speed and start at the same time. Which ant G
Fig. 5. 9
will arrive at E first?
Proof Consider A, B, C, D, E as vertices, the original nine
edges as nine edges, and we obtain graph G . Then G is connected and
the number of odd vertices is 2. According to Theorem 1, it is a chain.
Since B is an odd vertex, E is also an odd vertex, there exists a
chain from B to E, for example

BCDACEABDE.
56 Graph Theory

For ant B, it can start from B and arrive at E along this chain.
But vertex A is an even vertex, it is impossible to go through all
edges of C from A to E without repeating. Ant A must repeat at least
an edge. So ant B may choose a suitable route to reach E earlier than
ant A .
Example 5 As shown in Fig. 5.10, the three vertices of the big
triangle are dyed in colors A, B, C. Choose some points in the
interior of the big triangle, divide the big triangle into several small
triangles. Each of the two small triangles has either a common vertex,
or a common edge, or no common vertex at all. Color the vertices of
each small triangle by one of A, B, C. Prove that whatever way you
color, there must be a small triangle whose three vertices are all
different.
A A

G G'
Fig. 5.10 Fig. 5.11

Proof Choose a point outside the big triangle and a point in the
interior of small triangle as vertices, respectively. When two faces
have a common edge AB, add an edge between the corresponding two
vertices, and obtain graph C', as shown in Fig. 5. 11.
A small triangle with its vertices colored A , B, C corresponds to
the any vertex C' with degree 1. Other small triangles correspond to
°
the vertices of C' with degree or 2. Since the degree of a vertex
outside the big triangle is 1 and the number of odd vertices is even,
Therefore C' has at least an odd vertex v except u . That is to say, in
Fig. 5. 11, there is at least a small triangle whose three vertices have
three colors A, B, C, respectively.
 Euler's Problem 57

Note By the conclusion of this example, the famous Brouwer

fixed point theorem follows.
Example 6 A graph which consists of a convex n-polygon and
n - 3 disjoint diagonal lines in the polygon is called a subdivision
graph.
Prove that there exists a subdivision graph which is a cycle drawn
without lifting one's pen (i. e. start from a vertex, go through each
segment only once and return the starting point) if and only if 3 In.
(The 5th China Mathematical Competition)
Proof First prove by induction that the condition 3 I n is
sufficient.
When n = 3, clearly the proposition holds.
Suppose for any convex 3k-polygon, there
A,
exists a subdivision graph which is a cycle drawn
without lifting one's pen. For a convex3(k +1) =
3k + 3 -polygon A A 2A 3. .. A 3k + 3 , join A 4A 3k + 3.
j

Since A4AS . .. A 3k + 3 is a convex 3k-polygon, by

A, A JI.- - J

induction, A4AS ... A 3k + 3 contains a subdivision

Fig. 5. 12
graph which is a cycle drawn without lifting
one's pen. Construct this subdivision graph and join A 2A 4 , A 2 A 3k + 3 ,
so we obtain a subdivision graph of a convex 3(k + 1) -polygon
A j A 2A 3. .. A 3k + 3. Since the subdivision graph of A4As . .. A 3k + 3 is a
cycle, we start from A 3k + 3 , go through each edge of the subdivision
graph only once and return to A 3k + 3. Then go through A 3k + 3A A A2 ,
j , j

A 2A 3 , A 3A4' A4A2' A 2A 3k + 3 , and return to A 3k + 3 , again. This proves

that for any convex Ok + 3) -polygon, there also exists a subdivision
graph, which is a cycle drawn without lifting one's pen. So the
sufficiency has been proved.
Next prove the necessity. Assume that a convex n-polygon has a
subdivision graph, which is a cycle drawn without lifting one's pen.
Then each vertex of the graph is an even vertex. Clearly a convex
quadrangle and a convex pentagon do not have a subdivision graph
such that each vertex is an even vertex. Thus when 3 < n < 6, if a
58 Graph Theory

convex n -polygon has a subdivision graph, such that each vertex is an

even vertex, then n = 3. When 3 ~ n < 3k (k > 2), if a convex
n-polygon has a subdivision graph, such that each vertex is an even
vertex, then 3 1n. Now consider 3k ~ n < 3 (k + 1). Suppose that a
convex n-polygon A I A 2 • •• An has a subdivision graph, such that each
vertex is an even vertex. It is easy to see that any subdivision of a
convex n (n > 3) -polygon can divide the convex n-polygon into n - 2
small triangles, which have no common interior, and at least two of
these triangles contain two adjacent edges of the convex n-polygon as
two edges. Hence without loss of generality, let AlA3 be a diagonal
line of a subdivision graph of the convex n-polygon A I A 2 • •• A n (as
shown in Fig . 5.13). So AlA3 is still an edge of another L~AIA 3 A i in
the subdivision graph. By hypothesis that A I A 2 ••• An has a
subdivision graph such that each vertex is an even vertex, hence i 4. "*
"*
Otherwise, A 3 is an odd vertex . Equally i n; otherwise, Al is an odd
vertex. Therefore 4 < i < n. The subdivision
graph of A I A 2 • •• An gives rise to subdivision
graphs of a convex (i - 2) -polygon A 3A4' " Ai
and a convex (n - i +2) -polygon A I A 2 ••• An'
respectively. Each vertex of the convex
polygons corresponding to these two subdivision
Fig. 5. 13
graphs is even. Hence by induction,

3 I i - 2,3 I n - i + 2,

so 3 1n. Hence the necessity has been proved.

The necessity also can be proved by the coloring method. For a
subdivision graph of a convex n-polygon, we can color the divided
triangles using two colors, such that two triangles with a common edge
have different colors. Do as follows: draw diagonal lines in sequence
so that each diagonal line divides the interior of polygon into two
parts, in one part keep the original color, in another part change color.
Finally, we draw all the diagonal lines and obtain the needed color.
Since convex polygon has a subdivision graph, which is a cycle
 Euler's Problem 59

drawn without lifting one's pen, each vertex is an even vertex. So the
number of triangles at each vertex is odd. In the above coloring
method, all the edges of the polygon belong to the triangles with the
same color. Let it be black (see Fig . 5.14). Denote
the number of edges of white triangles by m.
Clearly 3 1m, each edge of the white triangles is
also that of the black triangles . However, all the
edges of the polygon are those of the black
triangles, so the number of edges of the black Fig. 5.14
triangles is m + n, so 3 1n .
Example 7 Suppose n > 3 , consider the set E of 2n - 1 distinct
points on a circle . Color some points of E black, and other vertices no
color. If there exists at least a pair of two black points such that
between two arcs with the two black points as their endpoints we can
find one of them whose interior (not including endpoints) contains
exactly n points, then we call the coloring "good" . If each coloring
with k points of E colored black is good, find the minimum value of
k. (31 th International Mathematical Olympiad)
Proof Denote the points of E by V1' V2' • •• , V 2n -1 according to
the anti-clockwise direction and add an edge between v i and v i + ( n - J) ,

i = 1, 2, ... , 2n - 1. We assume that V j +(2n -J)k = Vj' for k = 1, 2,

3, .... Then we get a graph G . The degree of each vertex in G is 2
(i.e . every vertex is adjacent to two other vertices) and V i andv i+3 are
adjacent to a common vertex . Since each vertex of G is an even
vertex, G consists of one or several cycles.
(i) When 3 1 (2n - 1) , graph G consists of three cycles, the vertex
set of each cycle is

{vi i i = 3k, k = 1, 2, . . . , 2n 3- 1 } ,

{vii i = 3k + 1, k = 0, 1, ... , 2n 3- 4 } ,

{vii i = 3k + 2, k = 0, 1, .,. , 2n 3- 4 }.
60 Graph Theory

Since the number of vertices in each cycle is 2n 3- 1 • it is possible

1 (2n
to choose at most:2 -- - -1 - 1 )
3 =
n - 2 vertices
-3- . and every two a f

them are not adjacent (note that 2n 3- 1 is odd). So we may choose

n - 2 vertices which are all not adjacent pairwise. By the pigeonhole

principle. we must color at least n - 1 vertices black to assure that
there is at least a pair of adjacent black vertices.
(ii) When 3 '} (2n - 1). each vertex of Vl' V2' . . . • V2n - l can be
denoted in the form of V3k' So graph C is a cycle with length 2n - 1.
We can choose n - 1 nonadjacent vertices on this cycle and at most
n - 1 nonadjacent vertices. Hence color at least n - 1 vertices black so
that there is at least a pair of adjacent black vertices.
In other words. when 3'} (2n -1). the minimum value of k IS n.
and when 3 I (2n - 1). the minimum value of k is n - 1.

Exercise 5

1 What is the value n when the complete graph K n is a cycle?

What is the value n when the complete graph K" is a chain? What are
the values m. n. when the complete bipartite graph Kin." is a cycle?
2 Suppose graph C can be drawn by lifting one's pen at least k
times. C' is obtained by deleting an edge. How many times at least can
C' be drawn by lifting one's pen?
3 Determine whether each of the Fig. 5. 15 can be drawn without
lifting one's pen.

r 1

Fig. 5,15
 Euler's Problem 61

4 Choose arbitrarily n (n > 2) vertices, and join each vertex to

all other vertices. Can you draw these segments without lifting one's
pen, so that they join end to end and finally return to the starting
point?
5 If at a conference, each person exchanges views with at least
() ~ 2 persons. Prove that it is definitely possible to find k persons VI ,

V 2' such that V I changes opinion with V2' V2 exchanges views

. .. , V k ,

with V3 ' . • . , V k - I changes opinion with v k ' and v k exchanges views

with VI , where k is an integer greater than O.
6 As shown in Fig . 5.16, graph G has 4 vertices, and 6 edges .
They are all on a common plane. This plane is divided into 4 regions
I, IT , ill, N, and we call them regions faces. Suppose there are two
points QI' Q 2 on these faces. Prove that there is no line f-t joining Q I
and Q 2 which satisfies : (1) f-t cuts across each edge only once; (2) f-t
does not go through any vertex v j (j = 1, 2, 3, 4) .
V,

V)

Fig. 5. 16

7 Arrange n vertices V I ' V 2' . . . , V n in order on a line. Each

vertex is colored in red or blue. If the ends of a segment v i V i +1 are
colored differently, we call it a standard segment. Suppose the colors
of V I and v n are different. Prove that the number of the standard
segments is odd.
8 Choose some points on the edges and in the interior of L:,ABC.
Divide L:,ABC into various small triangles. Each two small triangles
has either a common vertex, or a common edge, or no common vertex
at all . Use A, B or C to label those vertices in the interior of L:,ABC.
Use A or B to label the vertices on the edge AB of the big triangle,
62 Graph Theory

label B or C to the vertices on the edge BC of the big triangle, and

label C or A to the vertices on the edge CA of the big triangle. Prove
that there must be a small triangle, whose three vertices are A, B, C.
9 In the following figure with 25 small squares, try to design a
walk starting from point A, going through the edges of all the small
squares and finally returning to A, such that the path is the shortest.

Fig. 5. 17
Chapter6 Hamilton's Problem

In 1856, the famous British mathematician Willian Rewan

Hamilton brought forward a game whose name was" go around the
world". He denoted twenty big cities by twenty vertices of a regular
dodecahedron. You should go along the edges, pass through every city
once and at last return to the starting point.
The game was welcomed all around the world. In this game, we
see a chain that it passes through every vertex only once. We call this
chain (cycle) a Hamiltonian chain (cycle). If a graph contains a
Hamiltonian cycle, we call it a Hamiltonian graph.
On the surface, Hamilton's problem is similar to Euler's problem.
But, in fact, they are different in nature. Hamilton's problem is one
difficult problem in graph theory that has not been solved. Until now
we can not find a necessary and sufficient condition to characterize it.
So we have different methods for different problems. We shall use
some examples to illustrate.
Example 1 Does Fig. 6.1 contain a 2

Hamiltonian chain or a Hamiltonian cycle?

Solution As Fig. 6. 1 shows us, according 3

to numbers shown we can find a Hamiltonian

cycle.
Here we use the" direct search" method to 7
solve the problem of "go around the world". Fig. 6. 1
That is, go from a vertex and search one by one
in order to find the Hamiltonian chain (cycle). If we find one chain,
we have found one solution. If not, there does not exist a solution .
64 Graph Theory

This method can always be used on simpler graphs and often to those
graphs which contain a Hamiltonian chain (cycle).
Example 2 In an international mathematics conference, there are
seven mathematicians come from different countries. The language
they can speak is
A: English
B: English and Chinese
C: English, Italian and Spanish
D: Chinese and Japanese
E : German and Italian
F: French, Japanese and Spanish
G: French and German
How can we arrange these seven mathematicians round a table so
that everyone can talk with the person beside him?
Solution We denote the seven mathematicians by seven vertices
A, B, C, D, E, F, G. If two persons can speak a common language ,
then we join the vertices representing them and we get a graph G. As
Fig. 6 . 2 shows us, the problem of arranging seats becomes a problem
of finding a Hamiltonian cycle. Arrange the seats in the order of the
cycle, so that everyone can talk with the person beside him.

A (En) A

B(En, Ch)

C (En, 1:k-~--........j..-..,,oF(Fr, .la, Sp)

D(Ch, .Ia) E (Ge, It) E

Fig. 6. 2 Fig. 6. 3

Note Ch = Chinese, En = English, Fr = French, Ge

German, It = Italian, Ja = Japanese, Sp = Spanish.
In Fig. 6. 2, we draw a cycle in a bold line and then we get our
solution, which also means if we arrange the seats in the order A, B,
D, F, G, E, C, everyone can talk to the persons beside him. The
 Hamilton's Problem 65

common language is labelled on each corresponding edge in Fig. 6. 3.

Example 3 Determine whether the graph G in Fig . 6. 4 contains a
Hamiltonian chain or cycle?

B
Fig. 6. 4 Fig. 6. 5

Solution We mark one vertex in graph G as A. For example, we

mark the vertex a as A and all the vertices adjacent to the vertex a as
B all the vertices adjacent to B as A . Then we mark the vertices
adjacent to the vertex which is marked B as A and the vertices
adjacent to the vertex which is marked A as B until we mark all the
vertices. As Fig . 6. 5 shows us, if G contains a Hamiltonian cycle, the
cycle must go through A and B in turn. So the difference between the
numbers of A and B is no more than 1. But in Fig. 6. 5, there are nine
A vertices and seven B vertices. The difference is 2, so there are no
Hamiltonian chain.
Generally, to a higraph G = (VI' V 2 , E), there is a simple
method to see whether the graph contains a Hamiltonian chain or a
Hamiltonian cycle.
Theorem 1 In a bigraph G = (VI' V 2 , E), if I VI 10:/=1 V 2 I , G
must contain no Hamiltonian cycle. If the difference between I VI I
and IV 2 I is more than 1, G must contain no Hamiltonian chain.
We can use the same method as Example 3 to prove it.
Example 4 Fig. 6. 6 shows us half of
a chessboard. A knight is at the bottom
right corner. Can the knight move along
every square continually once only? What
happens if we delete the black panes at Fig. 6. 6
66 Graph Theory

the two corners of the half chessboard?

Solution We consider the following graph. We denote the
squares in the half of a chessboard by the vertices of a graph. If a
knight can move along from one square to another square in one step,
we join the two vertices representing the panes. So the problem
becomes a new problem of determining whether the graph contains a
Hamiltonian chain. In the graph, whether the two vertices are
adjacent is determined by the . rule of how a knight moves. Two
vertices are adjacent if they are at the two ends of the shape of letter
"L" on the chessboard. Color a vertex by the color of the square
representing the vertex in the chessboard, the colors of the two
adjacent vertices are always different. So between the two adjacent
vertices in the graph, one vertex is black
15 18 7 22 11 28 5 24
and the other is white . The number of
8 21 16 27 6 23 2 29
the black vertices is the same as the 17 14 19 10 31 12 25 4
number of white vertices, so there exists 20 9 32 13 26 3 30 1
a Hamiltonian chain. We can use the trail
Fig. 6. 7
and error to find a chain.
Now let us consider the second part of the problem . Again we use
the above method to convert the problem to determining whether the
graph contains a Hamiltonian chain. The number of the black squares
is 14 and the number of white squares is 16. According to Theorem 1,
the graph contains no Hamiltonian chain. It means that the knight
cannot move along every square continually once only when we delete
the black squares.
Now we do not know the necessary and sufficient condition of the
problem, namely, whether a connected graph contains a Hamiltonian
chain (cycle). However many first-class mathematicians have done
some hard work for more than one century, they have found some
sufficient conditions and some necessary conditions. In what follows
we give a sufficient condition for the problem whether a simple graph
contains a Hamiltonian chain.
Theorem 2 G is a simple graph with n (n ~ 3) vertices. For every
 Hamilton's Problem 67

pair of vertices v, VI,

d(v) +d(v') ~ n -1,

then G contains a Hamiltonian chain.

Proof First, we prove that G is a connected graph. Suppose that
G contains two or more connected components. Suppose one of them
has n 1 vertices, and another has n 2 vertices. We take one vertex each,
VI and V2' from the two components. Thend(vl) <nl -1 andd(v2) <
n2-1. So

It contradicts the hypothesis. So G is a connected graph.

Now we prove that there exists a Hamiltonian chain. The method
of proof in fact gives us a method to construct a Hamiltonian cycle.
Suppose G contains a chain from VI to V p: VI V2 . •. V p. If VI or
V p is adjacent to one vertex which does not lie in the chain, we can
extend this chain so that the vertex lies in the chain. Otherwise, V I

and V p are only adjacent to the vertices of the chain. There must exist
a cycle with vertices Vj' V2' ... , v p. Suppose the vertex set adjacent
to Vj is {Vjl' Vj2' ... , V jk }. Here Vjl' VJ2' •.. , V jk are the vertices
on the chain and p < n.
If Vj is adjacent to v p' then there exists a cycle Vj V2 . .. V pVl .

If Vj is not adjacent to v p' then there exists a vertex v [(2 <l <
p) which is adjacent to Vj and v [-1 is adjacent to v p as Fig. 6.8 shows
us. If not, v p is adjacent to p - k -1 vertices at most, which excluding
V j l - l ' Vj2-j, ••• , Vjk-1 and Vp' means that
d (v 1) + d (v p) < k + (p - k - 1) = p - 1 < n - 1.

~ v,
'".~l _-----
__ -]--_VP
"""""~"------~~_----- V,-l

- - - - - - - - - - - - - - - - VI V I .. \

Fig. 6. 8
68 Graph Theory

It contradicts the hypothesis. So there exists a cycle containing VI ,

Vl ' • . . , Vp: V I V/V / +I ••• VpV/ - I V/-l ••• V1VI.

If P = n , already there exists a Hamiltonian cycle. If p < n , since

G is connected, G must contain a vertex v' which does not belong to
the chain and is adjacent to one vertex v k of VI Vl . . . v p. As Fig. 6. 9
shows us, we can get a cycle containing VI ' V2 ' ... , V p ' v' :
V 'Vk Vk+l _ •• Vt-1V p Vp - l ••• VIVI V 2 ••• VkV'. Repeat above process until
there exists a chain with n - 1 edges.

~: _
_L _ v:;jv,
"''''''' , VI. V " <l ......... ..- V,.--2

"'--, -- ---- - --~;:,-

v;

Fig. 6. 9 Fig. 6. 10

It is easy to see that the condition of Theorem 2 about the

existence of a Hamiltonian chain is sufficient, but not necessary.
Suppose G is a polygon of n sides, as Fig. 6.10 shows us. Here n = 6.
Though the sum of the degrees of any two vertices is 4 < 6 - 1, G
contains a Hamiltonian chain.
In 1960, Ore gave us a sufficient condition for a Hamiltonian
graph in American Mathematical Monthly.
Theorem 3 G is a simple graph with n vertices. For every two
nonadjacent vertices v, v'

dC v ) + dC v' ) ?:; n,

then G is Hamiltonian cycle.

Proof When n = 3, by the given condition we know that G must
be a complete graph K 3. The conclusion is true.
Suppose n ?:; 4, we prove by contradiction. Suppose G with n
vertices satisfies the given condition of degree and contains no
Hamiltonian cycle.
 Hamilton's Problem 69

Without loss of generality, G is the graph which satisfies the given

condition and contains the most number of edges. In other words,
after adding an edge to G, G contains a Hamiltonian cycle.
Otherwise, G can be added several edges until we can not add edges.
After adding edges, the degrees of vertices satisfy the condition of
degree. Then we get a Hamiltonian chain contains every vertex of G.
We denote the chain by VI V2 . .. v,,, then VI is not adjacent to v". So

Then among V2' V3' ... , V,, - I , there must be a vertex v i so that
VI is adjacent to Vi and v" is adjacent to Vi-I as Fig. 6.11 shows us.
Otherwise, there are d(vI) = k vertices Vi, ' Vi 2 , ••• , Vi k (2 :s;;; i l :s;;;
i2 :s;;; ••• :S;;;i k :s;;; n -1) adjacent to VI' and Vn is not adjacent to Vi,-I'
V i 2 -1' ••• , V ik - l ' So

then
dev I ) + d(v,.):S;;;k + n - l - k =n - 1 <n,

which contradicts the condition. So G contains a Hamiltonian cycle

VIV2 ... Vi - IV"VI/-"" V iVj, which also contradicts the hypothesis.
We complete the proof .
V,~_ _~ v,

V, _ l

Fig. 6.11

For the complete graph K 1/ (n ~ 3), clearly there is Hamiltonian

cycle .
Example 5 n persons take part in a conference. During the
conference time, everyday they must sit at a round table to have
dinner. Every evening, every person must sit beside different persons.
How many times at most will there be such dinners?
70 Graph Theory

Solution We denote n persons by n vertices. Draw a complete

graph K", then the Hamiltonian cycle in K n is a way of sitting round
the table. The most number of times is equal to the number of
Hamiltonian cycle with no common edges in K" .

K" contains ~ n (n - 1) edges and every Hamiltonian cycle

contains n edges. There are at most [n ~ 1] Hamiltonian cycle with

no common edges. When n = 2k + 1, we arrange the vertices 0, 1,
2, ... , 2k as Fig . 6. 12. First, we take a
Hamiltonian cycle (0, 1, 2, 2k, 3, 2k - 1,
4, ... , k + 3, k, k + 2, k + 1, 0), then
rr 2rr
rotate k' k' ... , (k
- 1) krr clockwise

around the 0 and get k [n ~ 1 ]

k+ 1
Hamiltonian cycle with no common edges. If
Fig. 6. 12
n = 2k + 2, add a vertex v in the center and
also get k Hamiltonian cycle .
From Theorem 3 we can induce next Theorem 4, which is given
by the mathematician Dirac in 1952 .
Theorem 4 G is a simple graph with n vertices. If the degree of

every vertex v is no less than ; , then G must contain a Hamiltonian

cycle.
Example 6 Arrange 7 examinations in 7 days so that the two
courses which are taught by one teacher cannot be arranged in two
consecutive days. If every teacher can teach at most 4 courses which
have examination. Prove that it is possible to arrange the examinations
in the above way.
Solution Suppose the graph G contains 7 vertices. Every vertex
represents an examination. If two courses are not taught by one
teacher, then we join the two vertices representing the two courses.
Since the number of courses taught by one teacher is no more than 4,
 Hamilton's Problem 71

the degree of every vertex is at least 3. The sum of degrees of any two
vertices is at least 6. By Theorem 2, G contains a Hamiltonian chain.
Since in this chain every two vertices which are adjacent represent two
courses which are not taught by one teacher, we can arrange 7
examinations in the order of vertices in this chain.
Example 7 A factory produces two-color cloth using 6 distinct
colored yarns. Among six kinds of yarns, every color must be matched
in groups with three other colors. Prove that it is possible to choose
three kinds of two-color cloth so that all 6 colors are present.
Proof We denote six colored yarns by six vertices. If two colors
are in the same group, then we join the vertices of the two colors, and
we get a graph G. What we know is that every color can be matched in
groups with three other colors. For any vertex v j , d (v j) ;? 3, what we
will prove is that graph G contains three edges, any two edges of
which contain no common end.
For any vertex v in the graph G, by Theorem 4, G contains a
Hamiltonian cycle which we denote by VIVZV 3 V4VSV6VI. Then edges
(VI' V 2 ) ' (V3' V4)' (v s ' V6) are the three edges which contain no

common vertex.
We always use the fact that the graph satisfies the sufficient
condition to determine that the graph is a Hamiltonian graph and the
fact that the graph does not satisfy the necessary condition to
determine that the graph is not a Hamiltonian graph. Next we give a
necessary condition.
Theorem 5 If G contains a Hamiltonian graph, remove several
vertices VI' VZ, . . . , V k and their adjacent edges from the G to get a
new graph G', then the number of connected components in G' is no
more than k.
Proof Suppose c is a Hamiltonian cycle in G. After removing
several vertices VI' vz, ... , V k and their adjacent edges from G, the
cycle c can be divided into at most k parts. So the number of
connected components in G' is at most k.
Example 8 Prove that there is no Hamiltonian graph in
72 Graph Theory

V,
Fig. 6 . 13.
Proof Remove the vertices VI' V 2 and
their adjacent edges from Fig. 6. 13. We
obtain G' with three connected components.
It does not satisfy the necessary condition of
Theorem 5. So there is no Hamiltonian graph
in Fig. 6.13. V,

Finally, we give another example to end Fig. 6. 13

this chapter.
Example 9 If A oAIA2 ... A 2n- 1 is a regular polygon of 2n sides.
Join all its diagonals to get a graph G. Prove that every Hamiltonian
cycle of G must contain two edges which are paralleled in the graph.
Proof Suppose A;A j is parallels to A kA l • Since the number of the
vertices between A i and A l is equal to the number of the vertices
between A j and A k , i - l = k - j. The sufficient and necessary
condition for A iA j parallel to A kA I is :

i +j - k +l (mod 2n ) .

Suppose A io A i, .. . A i2n _, is a Hamiltonian cycle. 1 0 ' 11 ' ... , i 2n - 1 is a

rearrangement of 0, 1, ... , 2n . Among them any two edges are not
paralleled. So among the 2n numbersi o + i " i , + i 2 , i2 + i 3 , •• • , i 2n - 1 +
i n any two numbers are not congruent module 2n . That is, the above
2n numbers is in a surplus system of module 2n. Then

U o + i , ) + U I +i 2 ) + U 2 +i 3 ) + .. . + U 2n - 1 + io)
= 0 + 1 + 2 + ... + 2n - 1 = 2n 2 - n
- n(mod 2n) .

On the other hand,

(ill + i , ) +U I + i 2 ) + (i 2 + i 3 ) + ... + (i2n-1 +io )

= 2(i o + i l + i 2 + ... +i 2n- l )

= 2 (0 + 1 + 2 + ... + 2n - 1) = 2n 2 - n
= 2n(2n - 1)
= O(mod 2n).
 Hamilton's Problem 73

We get two results which contradict each other. So the proof is

complete.

Exercise 6

1 What is the value n so that the complete graph K n is a

Hamiltonian graph? What are the values m, n so that the complete
bigraph Km, n is a Hamiltonian graph?
2 The graph representing the regular tetrahedron, hexahedron,
octahedron or icosahedron is a Hamiltonian graph.
3 We use paper to construct an octahedron. Can we cut it into
two parts so that every face is also cut into two parts and the cutting
lines do not go through the vertices of the octahedron?
4 A mouse eats the cheese whose size is the same as 3 X 3 X 3
cube . The way to eat it is to get through all the 27 of the 1 X 1 X 1
subcube. If the mouse begins from one corner, then goes to the next
subcube which has not been eaten . Can the mouse be at the center
when he has eaten the cheese .
S We divide 6 persons into 3 groups to finish 3 missions. There
are 2 persons in every group. Everyone can cooperate with at least 3
persons among the other 5 persons. (1) Can the two persons of every
group cooperate with each other? (2) How many distinct grouping 6
persons into 3 groups can you give?
6 A king has 2n ministers, among whom there are several
ministers hate each other. But the number of persons every minister
hate is no more than n - 1. Can they sit in a round table so that no two
adjacent ministers hate each other?
7 Among 9 children, every child knows at least four children.
Can these children be arranged in a line so that every child know the
child beside him?
8 A chef uses eight materials to do the cooking. He should use
two materials for each dish. Every material should be used in at least
74 Graph Theory

four dishes. Can this chef cook four dishes so that he uses 8 distinct
materials?
9 All the subsets of a finite set can be arranged side by side in a
way so that every two adjacent subsets differ only in one element.
10 In a plane, a graph with n vertices and several edges is not a
Hamiltonian graph. But if we remove one vertex and its adjacent edges
from the graph, the graph will become a Hamiltonian graph. Find the
minimum n.
11 Around a table there sit at least five persons and it is possible
to rearrange their seats so that beside everyone there are two new
neighbours.
Chap ter 7 Planar Graph
~ -=--:::..--

A graph is called a planar graph, if it can be drawn in the plane

so that its edges intersect only at their ends .
Some graphs seem to have edges intersecting, but it is not clear
that they are not planar graphs. See Fig. 7. 1 (1), it is isomorphic to
Fig. 7.1(2). So it is easy to see that Fig. 7.1(1) is a planar graph.

( I) (2)

Fig. 7. 1

When we talk about a planar graph G, we always assume that the

planar graph is drawn according to such requirement. The vertices and
edges of a graph partition the plane into
several separated regions. Each area is
called a face of G. The only unbounded
V ,<r------''<l: .O ------'>P V ,
face is called the outer face and the others
are the inner faces. In Fig. 7. 2, F 1 , F 2 ,
F 3 ' F 4 are inner faces, F 5 is the outer Fig. 7. 2
face.
We know Euler's Formula of a convex polyhedron in middle
schools. That is, if a convex polyhedron has v vertices, e edges and f
faces, then v - e + f = 2. We can generalize it to planar graphs.
Theorem 1 (Euler's Formula) For a connected planar graph G
wi th v vertices, e edges and f faces, then
76 Graph Theory

v -e +1 = 2.

Proof We prove by induction on e.

If G has only one vertex, then v = 1, e = 0, 1 = 1 , the assertion
v - e +1 = 2 holds.
If G has one edge, then v = 2, e = 1, 1 = 1, then the assertion
v -e +1 = 2 holds.
Suppose that it is true for all connected planar graphs with k
edges, i.e. Vk -ek + Ik = 2. Now we discuss the case when G hask +
1 edges .
If we add a new edge to a connected graph G with k edges such
that G is still a connected graph, there must be two cases.
(i) Add a new vertex v' which is adjacent to a vertex v of G as
Fig. 7.3(1) shows us . Then both v k and ek increase by 1, Ik does not
change. So

(ii) Add a new edge to join two vertices of G as Fig. 7. 3 (2) shows
us. Then both Ik' ek are added by 1, V k does not change. So

By the induction hypothesis, the theorem holds for every positive

integer e.

v'

(I) (2)
Fig. 7. 3

We can use Euler's Formula to determine the maximum number of

edges in a simple planar graph. Since a face has at least 3 edges, the
boundaries of 1 faces have at least 31 edges. In addition, one edge
 Planar Graph 77

2
belongs to the boundaries of at most 2 faces. So 2e ?:?- 31, and 1 ~ 3e.

Use Euler's Formula

2
2 = v - e +1 ~ v - e + 3e ,

that is

e ~ 3v - 6.

This proves the following theorem.

Theorem 2 For a connected simple planar graph G with v (v ?:?- 3)
vertices and e edges, then

e ~ 3v - 6.

In fact, Theorem 2 also holds for disconnected simple planar

graphs. Theorem 2 can be used to determine whether a graph is a
planar graph or not.
Example 1 Prove that the complete graph K 5 is not a planar
graph.
Proof Since v = 5, e = 10 do not satisfy e ~ 3v - 6. So K 5 is not
a planar graph.
Example 2 Prove that K 3 . 3 is not a planar graph.
Proof Suppose that K 3 • 3 is a planar graph. Since we choose 3
vertices randomly in K 3 . 3 ' there must be 2 vertices which are not
adjacent to each other. Therefore, each face has at least 4 edges as its
boundary. By

41 ~ 2e, 1 ~ ~.
Use Euler's Formula

e
2 = v-e + l ~ v-e+2'

that is
e ~ 2v - 4.
78 Graph Theory

In K 3.3 ' V = 6, e = 9, so 9 > 2 X 6 - 4, this leads to a

contradiction. So K 3 • 3 is not a planar graph.
Although Euler's Formula and the inequality derived from it can
be used to prove that a graph is not a planar graph, but it can do
nothing to prove that a graph is a planar graph. The Polish
mathematician Kuratowski showed us a brief result in 1930. Every
nonplanar graph has either K 5 or K 3.3 as its subgraph. To depict the
result clearly, we give the definition of homeomorphism first.
Two graphs G 1 , G 2 are called homeomophic, if G 1 can be obtained
by inserting some new vertices on G 2 's edges . The two graphs in
Fig. 7.4 are homeomorphic. By the definition of homeomorphism, we
know that inserting or deleting a number of 2-degree vertices does not
change the planarity. Kuratowski's theorem is as follows.

MM 0)
Fig. 7. 4
(2)

Theorem 3 A graph is planar if and only if it contains no

subgraph homeomorphic to K 5 or K 3 . 3 •
Although the theorem is basic, but the proof is too long to show.
Example 3 Are Fig. 7.5, Fig. 7.6 planar graphs?
Proof Fig. 7. 5 contains a K s as its subgraph, and Fig. 7. 6
includes a K 3 • 3 as its subgraph. By Kuratowski's Theorem, it is easy to
see that neither of them is a planar graph.

Fig. 7. 5 Fig. 7.6

 Planar Graph 79

Example 4 How many kinds of regular polyhedrons are there?

Can you give their numbers of edges, vetices and faces? How many
edges is a vertex incident to?
Proof As each vertex in a regular polyhedron lies on at least 3
faces, when every internal angle at a vertex in a regular polyhedron is
no less than 120 it cannot be a vertex of a regular polyhedron.
0
,

Therefore, we can only consider polyhedrons with regular pentagons,

squares and regular triangles as faces.
(1) Polyhedron constructed from regular pentagons.

Since the internal angle of a regular pentagon is ; TC, and ; TC X

4 > 2TC, each vertex is 3-degree. Hence 3v = 2e, ~f = e, and by

Euler's Formula

It is easy to see that

e = 30, v = 20, f = 12.

So, there is only one regular polyhedron constructed from regular
pentagons. It is a regular dodecahedron with 20 vertices, 30 edges,
and each vertex incident to 3 edges.
(2) You may prove polyhedrons constructed from squares and
regular triangles by yourselves. There are 4 kinds as follows shown in
Fig. 7. 7. Therefore, there are 5 kinds of regular polyhedrons.

regular dodecahedron cube regu.lar tetrahedron regular octahedron regular isosahedron

Fig. 7. 7
80 Graph Theory

Example 5 We partition a square into n convex polygons, where

n is fixed . What is the largest number of edges that a convex polygon
could have?
Proof By Euler's Formula, we know that when a convex polygon
is partitioned into n polygons, then v -e + n = 1 (because f = n + 1 ).
As we partition a square into n convex polygons, for each vertex
of these polygons, if it does not belong to the square, then it must be
the vertex of at least 3 convex polygons. We use A, B, C, D to
denote the vertices of the square, and v is an arbitrary vertex except
A, B, C, D, then

d(v) ~3(d(v) -2).

Calculating the sum of all the vertices except A, B, C, D,

2e - (d(A) + d(B) +d(C) + d(D»
~ 3(2e - (d(A) +d(B) + d(C) +d(D» - 6(v -4».

So,

4e ~ 2(d(A) + d(B) +d(C) + d(D» + 6(v -4).

Sinced(A) ~2, deB) ~2, d(C) ~2, d(D) ~2, hence

2e ~ 8 + 3 (v - 4).

Using

v-e + n = l,

we can obtain the following result

3(e + 1) = 3v + 3n ~ 2e + 4 + 3n,
that is, e ~ 3n + 1.
Draw n -1 lines crossing an edge of a square such that all the lines
are parallelled to their adjacent edge and divide the square into n
rectangles. The number of edges is

4+3(n - 1) = 3n+1.

In summary, the largest edge number is 3n + 1.

 Planar Graph 81

In 1968, two Soviet mathematicians Kozyrev and Grinberg gave a

necessary condition to planar graphs with Hamiltonian cycles.
Theorem 4 If a planar graph has a Hamiltonian cycle c, let f: be
the number of i-polygons inside c, and j':be the number of i-polygons
outside c, then
(1) 1 • f~ + 2 • f: + 3 • f~ + ... = n - 2;
( 2) 1 • j'~ + 2 • j'~ + 3 • j'~ + ... = n - 2;
(3) 1 • Cf~- j'~) +2· Cf: - j'~) + 3· Cf~- j'~) + ... = 0, where
n is the number of the vertices of G, and also the length of c.
Proof Suppose there are d edges inside c. As G is a planar
graph, its edges do not intersect and one edge divides the face into two
parts. Try to think of these edges putting side by side in the graph. We
get one more face each time we put an edge in the graph. So d edges
divide the interior of c into d + 1 faces . The total number of the faces
inside c is
(1)

We can mark each i-polygon inside c with number i. The sum of

all the numbers marked on the faces is the number of the edges that
form the faces. Each edge inside c is counted twice, but the edges on c
is counted once. So

2· f~ + 3. f ~+ 4. f:+5. f~+··· = 2d + n. (2)

Subtracting equation (1) twice from equation (2), we can get

1 • f~ + 2 • f: + 3 • f ~ + ... = n - 2. (3)

Similarly, we can get the following result

1 • j'~ + 2 • j'~ + 3 • j'~ + ... = n - 2. (4)

Equation (4) subtracted from equation (3), you can get

1 • (f ~ - j'~ ) + 2 • Cf: - j'~) + 3 • (f~ - j'~) + ... = o.

Example 6 Prove that the Grinberg graph (see Fig. 7. 8) has no
Hamiltonian cycle .
82 Graph Theory

Proof Suppose that the Grinberg graph

includes a Hamiltonian cycle. Since there are
only pentagon, octagon and enneagon, from
Theorem 4,

3 • (j~ - 1'~) +6 • (j~ - 1'~) +7 • (j~ - 1';)

= o.
That is
Fig. 7.8
7 (j~ - 1'~) - 0 (mod 3).

It contradicts I~ - 1'~ = 1. Therefore, the Grinberg graph has no

Hamiltonian cycle.

Exercise 7

1 Let G be a simple planar graph, then it must have a vertex

whose degree is no more than 5.
2 Prove that a simple planar graph with edges less then 30 must
have a vertex whose degree is no more than 4.
3 Prove that in a simple planar graph with 6 vertices and 12
edges, each face is surrounded by 3 edges.
4 G is a graph with 11 or more vertices. G is the complement of
G, that is, G has the same vertices as G and all the possible edges but
not in G . Prove that G or G is not a planar graph .
S 1 parts, and every two parts are
Divide the plane into
adjacent. What is the largest value of I?
6 Suppose that each vertex in a convex polyhedron is adjacent to
all the other vertices. Prove that except tetrahedron such convex
polyhedron does not exist.
7 In a bus network there are n stations. Each station IS

connected to at least 6 roads. Prove that there must be two roads

intersecting on the plane.
 Planar Graph 83

8 If a polyhedron containing n edges exists, what is the value of

n?
9 A convex polyhedron has iOn faces. Prove that there are n
faces which have the same number of edges.
10 Prove that the graph in Fig. 7.9 has no Hamiltonian cycle.

Fig. 7. 9 Fig. 7.10

11 The graph in Fig. 7. 10 contains a Hamiltonian cycle. Prove

that for any Hamiltonian cycle, if it contains edge e, then it must not
contain edge e'.
12 Let S = {Xl' X2' ••• , Xn} en ~ 3) be the set of vertices on
the plane. The distance between two arbitrary vertices is no less than
1. Prove that there are at most 3n - 6 pairs of vertices whose distances
are 1.
Chapter8 Ramsey's Problem

Usually, Ramsey's problem refers to problems involving coloring,

Ramsey number (Ramsey, an English logician) and the Pigeonhole
Principle.
First we begin with a mathematical Olympiad problem. The
problem appeared in a Hungarian mathematics competition in 1947.
Example 1 Prove that among every six persons, you can always
find three persons who know each other or do not know each other.
We denote six persons by six vertices. If there are two persons
who know each other, we join the corresponding vertices and color it
red. If there are two persons who do not know each other, we join the
corresponding vertices and color it blue. What we are to prove is that
there must exist a monochromatic triangle. That is, there is a triangle
whose edges are either all red or all blue.
The above is not the only one such question. A similar problem
appeared in Putnam Mathematics Competition: In the space, there are
six points among which any three points are not on a line and any four
points are not on a plane. We join six points in pairs to get 15 line
segments. We use red and green to color these line segments. (A line
segment can only be colored in one color. ) Prove that in whatever way
the line segments are colored, there must be a monochromatic
triangle.
Next let us begin our proof to the Putnam competition problem
which can be regarded as a standard pattern.
Proof Let A I , A 2 , ••• , A 6 be the six vertices given. Consider 5
line segments AIA 2' AIA 3' ... , AIA 6 adjacent to AI. Since there are
only two colors coloring the 5 line segments, there must be 3 line
 Ramsey's Problem 85

segments in one color . Without loss of generality, let the three line
segments be AIA 2' A IA 3 , AlA4 which are colored red. (We denote
red by a solid line and blue by a dotted line.) If DA2A3A4 are blue
(see Fig. 8.1), then it is a monochromatic triangle. If in DA2A 3A4'
there are at least one edge, for example, A2A 3 is red (as Fig. 8. 2
shows us), then D AlA2A3 is a monochromatic triangle. In other
words, in either case there is a monochromatic triangle.

AI AI

A,
A2
A2
A, A,
Fig.8.l Fig. 8.2

From this example, we can easily know that when n ~ 6, we use

two colors to color all the edges of K n which is called a two-color
complete graph K n for short. Then there must
exist a monochromatic triangle.
Fig. 8.3 shows us a complete graph K 5 in two
colors, which contains no monochromatic
triangle b) •
In summary, we can get the next conclusion. Fig. 8. 3
Theorem 1 If a complete graph K in two colors contains a
11

monochromatic triangle, then the minimum n is equal to 6.

Example 2 Prove that it is impossible to color the K 10 10 four
colors so that any subgraph K 4 of K 10 contains all four colors.
Proof We prove by contradiction. Suppose that we can find a
coloring satisfying the requirement.

b) Vice versa, it is easy to prove that the two-color complete graph K 5 without
monochromatic triangle consists of two pentagon with different colors. In other words, in the
two-color complete graph K 5 , if there are no blue triangle and pentagon, then there must be a
red triangle.
86 Graph Theory

If a vertex is adjacent to 4 edges with the same color, say blue, we

color AB, AC, AD, AE all in blue. Among the edges joining B, C,
D, E there must be an edge in blue. Let it be BC. Then there are 4
blue edges joining A, B, C, D, and the remaining two edges should
be colored in three colors. A contradiction. So A must be adjacent to
at most 3 monochromatic edges and there must be one color coloring
three edges. Without loss of generality, suppose that AB, AC, AD
are all blue. There are 6 edges joining A, B, C, D, so the remaining
three edges are colored in different colors. There are no blue edges in
BC, BD, CD.
Consider the remaining six vertices. By Theorem 1, the graph
must contain a blue triangle or a triangle without blue edges.
If there are three vertices E, F, G with no blue edges joining
them, then there is no blue edges joining A, E, F, G. It is a
contradiction. So without loss of generality, let DEFG be a blue
triangle. Since there are no blue edges joining B, C, D, E, there
must be one blue edge among BE, CE, DE . Suppose that BE is blue,
then there must exist four blue edges joining B, E, F, G. It is also a
contradiction. In summary, the proposition is true.
Using the above conclusion, we can solve the following problem
in the 33th International Mathematical Olympiad.
Example 3 Given nine points in the space, among them there are
no four points on a face. Join every pair of points and try to find the
minimum n so that you can color everyone of any n line segments by
red or blue arbitrarily . In the set of the n line segments, there must be
a monochromatic triangle.
Solution By assumption there are no four points in a face, which
assures that among nine points there are no three points on a line. So
this problem is still a planar graph problem . The problem can be stated
as follows: There are nine points on a plane and among them there are
no three points on a line and there are 36 lines . How many lines should
we take so that there must be a monochromatic triangle when we color
the graph in two colors randomly?
 Ramsey's Problem 87

We construct a graph G with 9 vertices, 32 edges and in two

colors. We color the edges joining the vertex VI and V2' V3 ' V S , V 9 by
red (a solid line) while we color the edges joining the vertex VI and V4'
vs, V6' V 7 by blue (a dotted line). We divide the vertices other than
VI into four groups: I: II: (V4' v s ); ill: (V6' V 7 ); N:
(V2' V 3 );

(v s ' V 9 ). We call I and II, II and ill, ill and N adjacent groups.
Except VI , two vertices belonging to one group are not adjacent, two
vertices belonging to two different adjacent groups are joined by a
solid line (red) and two vertices belonging to two different groups
which are not adjacent are joined by a
dotted line (blue). Fig. 8. 4 shows
v,

that there are G) - 4 = 32 edges in v,

the graph G which contains 16 red

edges (solid lines) and 16 blue edges
(dotted lines). It is not difficult to
Fig. 8. 4
know that G contains no
monochromatic triangle. So n ;? 33.
Next let us prove n ~ 33. Suppose 33 edges connecting have been
colored. While there are 3 edges which have not been colored.
Without loss of generality, we denote the 3 edges by el' e 2' e3'

Choose one end V I ' V2 ' V 3 from e l ' e2' e3' respectively. Then we
delete the three vertices from K 9 and the remaining 6 vertices form a
graph K 6 ' So if we color the graph by red and blue, the graph must
contain a monochromatic triangle.
So n = 33.
To generalize Theorem 1, we first need to increase the number of
colors.
We use k colors CI' C2' •.. , Ck to color the complete graph K".
We call the complete graph K" k-color complete graph K " if every
edge is colored in one color. We can imagine if n is large enough,
k-color complete graph K" must contain monochromatic triangle. We
denote the least n by r k' In Theorem 1, r2 = 6. It is clear that rl = 3.
88 Graph Theory

The existence of rk was firstly proved by British mathematician

and logician Ramsey. We call rk the Ramsey number. Concerning rk

we have the next conclusion.

Theorem 2 (1) For every positive integer k, the Ramsey number
rk exists. When k ~ 2,
rk ~ k (r k-j - 1) + 2.
(2) For any natural number k ,

k' k'
r k ~1 + 1 +k +k(k - 1) + "'+2! + 1f + k!.

Proof (1) Apply induction on k. We know that rl' r2 exist and

rl =3, r2 = 6~2(rl - 1)+2.

Suppose that rk exists and rk ~ k (rk-l - 1) + 2 holds. Take n =

(k + l)(r k -1) + 2 and let K n be a (k + 1) - color complete graph

whose vertices are AI, A 2 , ••• , An. Take a vertex A 1 from K n
randomly, which is adjacent to n - 1 = (k + 1)(rk - 1) + 1 edges.
There are k + 1 colors in these edges . By the Pigeonhole Principle, in
these edges there are at least rk monochromatic edges . Suppose that
these r k edges are A IA2' AIA 3' ... , A IArk + 1' which are all colored
in color C I. Consider r k -subset K rk consisting of the vertices A 2 ,
A 3 , ... , A rk+ l • If K rk contains an edge with colored CI such as A 2 A 3 ,
then L AIA 2A 3 is a monochromatic triangle. If K rk contains no edges
with color C I ' then there are k colors coloring the K r k ' which means
that K rk is k -color complete graph. By the induction hypothesis, K rk
contains a monochromatic triangle .
In summary, K n contains a monochromatic triangle . We know
thatM = {m I any (k + 1)-color complete graph with m vertices that
contains a monochromatic triangle} is a nonempty subset of the natural
number set N, then there is a minimum rk +1 and

r k+ 1 ~ n = (k + 1) (r k -1) + 2.
(2) Apply induction. Whenk = 1, rl =3 ~ 1 + 1 +1. Suppose that
the property is true for k, then by (1) and the induction hypothesis,
we get
 Ramsey's Problem 89

rk +1 :s;; (k + 1 )(rk - 1) +2
:s;; (k + 1) [ 1 + k + k(k - 1) i i
+ ... + ; + ~ + k ! ] + 2
= (k + 1) + (k + 1)k + (k + 1)k (k - 1) + ...
+ (k + 1)! + (k + 1)! + (k + 1)' + 2
2! 1! .
= 1 + 1 + (k + 1) + (k + l)k + ...
+ (k + 1)! + (k + 1)! + (k + 1)'
2! 1! .'
So the property is also true for k + 1.
If we make use of the expanding formula of base number of the
natural logarithm in advanced mathematics

e = 1 + -.1 + -.1 + ... + ---.L + ...

1! 2! n! '
we can simplify (2) and get rk :s;; [k !eJ + 1. Here we denote the largest
integer no more than x by [ x].
Though Theorem 2 proves the existence of rk and gives an upper
bound of r k , we only know three exact values of rk' Other than rl =
3, rz = 6 mentioned above, we also know r 3 = 17. In face, according
to (1) in Theorem 2, we know
r3 :s;; 3 (rz - 1) +2 = 3 X 5 +2 = 17.
We follow the conclusion of Theorem 2 and obtain a result which was set as
a problem in the 6th International Mathematical Olympiad in 1964:
There are 17 scientists among which
o
everyone communicates with any other 15

person. They only discuss three problems

when they are communicating and every
two scientists can only discuss one problem. 4

Prove that there are at least three scientists

discussing one problem . On the other hand,
we can color the complete graph K 16 to
using three colors so that the graph contains
no monochromatic triangle. As Fig. 8. 5
90 Graph Theory

shows us, a solid line represents a red edge, a dotted line represents a
blue edge and no line represents a yellow edge. This means that r 3 ~

17. So r 3 = 17.
We can have another generalization of Theorem 1.
Suppose every edge of the complete graph K n is colored in red or
blue, which means that K" is a two-color (red, blue) complete graph
K". For two constant natural numbers p, q, when n is large enough,
the two-color Cred, blue) complete graph K" must contain a red K p or
a blue K q . We denote the least n satisfying the above proposition by
rCp, q). We also call rCp, q) the Ramsey number.
Using the concept of subgraph and complementary graph, we call
rCp, q) the minimum n so that any n-subgraph G of the complete
graph K n contains a complete subgraph K p or its complementary graph
G contains a complete subgraph K q •
By definition and Theorem 1, we know that r (3, 3) = r2 = 6.
Furthermore, it is easy to find thatrCl, q) = rCp, 1) = 1.
In order to understand the next general conclusion of r Cp, q), we
prove an example: r (3, 4) = 9. First prove r (3, 4) ~ 9 and we can
also give another example with the same form as Example 1.
Example 4 Prove that among any nine persons, you can find
three persons knowing each other or four persons not knowing each
other.
ProofWe denote nine persons by nine vertices AI' A 2' . . . ,
A ~. We join every two vertices and if A; knows A j' then color
A;A j red. Otherwise, we color it blue. What we need prove is that
in the two-color complete graph K 9 there must exist a red K 3 or a
blue K 4.
If a vertex is adjacent to no less than four red edges, we denote
them by A t A 2' A t A 3 ' A t A 4' A t A 5 . If there is an edge joining
A 2 , A 3, A 4 , A5 by a red line, for example A 2A 3 . Then 6 A tA2A 3
is a red triangle. If there is no edge join A 2' A 3 ' A 4' A 5 by a red
line, then A 2, A 3 , A4 , A5 form a blue triangle. We complete the
proof.
 Ramsey's Problem 91

If the number of red edges adjacent to every vertices is less than

4, then the number of blue edges adjacent to every vertex is no less
than 5. Consider the graph consists of AI, A 2 , • • • , A 9 and all the
blue edges. Since the number of odd edges is even, there must be an
even vertex such as VI ' Since the number of blue edges adjacent to AI
is even, the number of blue edges adjacent to Al is no less than 6. Let
AIA 2' A IA 3' .. . , AlA7 be blue edges . Consider six vertices A 2 ,
A 3 , •• • , A 7 , every two of them are incident
s
to be red or a blue edge.
By Theorem 1, the graph contains a red triangle or a blue triangle . In
the first case, the proposition is true, and we complete the proof . In
the second case, let ~A2A 3 A4 be a blue triangle. Then the complete
graph K4 with vertices AI ' A 2 , A 3 , A 4 is blue. The proposition is also
true.
Considering a complete two-color complete
gra ph K 8 as Fig. 8. 6 shows us, we denote red edges
by solid lines and blue edges by dotted lines. There
exists a color of graph K 8 such that K 8 contains no
red K 3 and no blue K 4 • It means that r(3, 4) > 8.
In summary, r(3, 4) = 9. Fig. 8. 6

Concerning rCp, q), we have the next conclusion.

Theorem 3 C1) r C2, q) = q, r Cp, 2) = p.
(2) rCp, q) = rCq, p).
(3) When p ~ 2, q ~ 2,

rCp , q) ~ rCp - 1, q) + rCp, q - 1).

The inequality holds when rCp -1, q) and rCp, q - 1) are all even.
We prove the theorem using subgraphs and their complementary
graphs . The proof of (3) is very difficult. We may refer to the proof
of Example 4 .
Proof C1) Let G be a graph with q vertices . If two vertices in G
are adjacent, G contains K 2. Otherwise G contains a K q ' So r C2,
q) ~ q. Since q - 1 vertices which are not adjacent to each other form
a graph G. Obviously G contains no K 2 and its complementary graph
92 Graph Theory

GcontainsnoK q . SorC2, q) ~ q.
In summary, rC2, q) = q. Similarly, we can prove rCp, 2) = p.
(2) If G contains rCp, q) vertices, then G also contains r Cp, q)
vertices. So G contains K p or G contains K q. In other words, G
contains Kq or G contains K p. So rCp, q) ~ rCq, p). Similarly, rCq,
p) ~ rCp, q). SorCp, q) = rCq, p).
(3) SupposeG containsrCp, q - 1) +rCp - 1, q) vertices, and VI
is a vertex of G.
IfdCvI) ~ rCp - 1, q), let () = rCp - 1, q) vertices be Vu
V3' ... , 1.S, 1.S+1 which are adjacent to VI. Then we remove other
vertices and their edges from G to get G I . According to the definition
of () = r Cp - 1, q), G I contains K P- I or G I contains K q. If G I contains
K p- I' then in G, K p-I and VI form a complete graph K p. If G I
contains K q' then G also contains this K q •
If the number of the vertices adjacent to VI is less than rCp - 1,
q), then V I is not adjacent to at least r Cp, q - 1) vertices. We denote
the vertices not adjacent to VI by VZ, V3' ... , V s ' V,+I' where
£ = rCp, q - 1). We remove the vertices except vz, V3' ... , V,+I and
their adjacent edges to get G z . According to the definition of £ = rCp,
q - 1) , Gzcontains Kp or G z contains K q- I . If G z contains K p , G also

contains this Kp . If Gz contains K q- I , then in G, VI and K q- I form

a Kq•
In summary, we get

rCp, q) ~ rCp, q - 1) + rCp - 1, q).

If r Cp, q - 1) and r Cp - 1, q) are even, we choose a graph G with

rCp, q - 1) + rCp - 1, q) - 1 vertices. Since the number of odd
vertices is even and rCp, q - 1) + r(p - 1, q) - 1 is odd, G contains an
even vertex VI. For VI' either d CV I) ~ rCp - 1, q) - 1, or VI is not
adjacent to at least rCp, q - 1) vertices. Since the number of vertices
adjacent to VI is even, in the first case, d CV I) ~ rCp - 1, q).
According to the same method we get
 Ramsey's Problem 93

rep, q) ~ r(p, q -1) +r(p - 1, q) - 1

< rep, q - 1) + r(p -1, q).

We complete the proof of Theorem 3.

By Theorem 3, we get an upper bound of some Ramsey numbers
rep, q). For example,

r(3, 3) ~r(3, 2) +r(2, 3) = 3 + 3 = 6,

r(3, 4) ~r(3, 3) +r(2, 4) -1 ~6 + 4 - 1 =9,
r(3, 5) ~r(3, 4) +r(2, 5) ~ 9 +5 = 14,
r(4, 4) ~ r(4, 3) + r(3 , 4) = 9 +9 = 18.

We have provedr(3, 3) =6, r(3, 4) =9, similarly, we can prove

r(3, 5) = 14, r(4, 4) = 18. By a known inequality, of course we need
only to prove r(3, 5) > 13, r(4, 4) > 17. Here, we only prove the
first inequality and leave out the second proof.
Consider the graph which Fig. 8. 7 shows
us, it contains no K 3 and its complementary
graph G contains no Ks. So r(3, 5) > 13.
II 4
Applying Theorem 3, we can get a simple upper
bound of r (p, q), as shown in Theorem 4.
Theorem 4 When p ~ 2, and q ~ 2,
8 7
Fig. 8. 7
r(p,q) ~ ( P + q-2) .
p -1

Proof Note that l = P + q. We apply induction on l.

When l = 4, P = q = 2. The left side is r(2, 2) = 2 and the right

side is (4-1 2) = 2.

Suppose the theorem is true when l = k (k ~ 4). When l = k + 1,

consider the case when p = k - 1, q = 2, or P = 2, q = k - 1 ,

r(k - 1, 2) = r(2, k -1) = k -1 = (kk -_ 21) = (k -1 1) .

The theorem is true . Whenp ~3, and q ~3, we havep +q = k + 1,
94 Graph Theory

apply (3) in Theorem 3 and by induction, we have

rCp, q) ~ rCp - 1, q) + rCp, q - 1)

~ (P + q - 3)+ (P + q - 3)
p - 2 p - 1

= (P + q - 2).
P- 1

The theorem is also true. By induction, Theorem 4 is proved.

In spite of the result, it is difficult to find the exact value of rCp,
q) . Except forrO, q) = rep, 1) = 1, rCp, 2) = p , rC2, q) = q, we
know only a few values of rCp, q). They are listed in the next table in
which the numerator represents the low bound and the denominator
represents the upper bound.

~ 3 4 5 6 7 8

28
3 6 9 14 18 23
29
25 34
4 9 18
28 36
25 42 51
5 14
28 55 94
34 38 102
6 18
36 94 178

If we combine the above two generalizations , we obtain the next

generalization.
Color the complete graph K n in [ colors C I ' C 2' ••. , c,. Every
edge can be colored in only one color and we get an [-color complete
graph. When n is large enough, [-color complete graph K n contains a
c I -color complete subgraph K PI or a c 2 -color complete subgraph
K P2' . • . or a c ,-color complete subgraph K PI . We denote the
minimum n satisfying the above propositions by r CPI' Pn . .. , PI).
We also call it a Ramsey number .
If we introduce the definition of hypergraph , we can make
 Ramsey's Problem 95

further generalization. We do not elaborate here.

Example 5 We divide 1, 2, 3, 4, 5 into two groups A, B
randomly. Prove that it is possible to find two numbers in a group and
the difference of the two numbers is the same as one number of this
group.
Proof We divide 1, 2, 3, 4, 5 into two groups A, B randomly.
Choose six vertices and label them as 1, 2, 3, 4, 5, 6. For any two
vertices i > j , it is always true that 1 < i - j < 5. For the two vertices
i > j, if i - j is in group A, we color the edge ij red; if i - j is in
group B, we color the edge ij blue. So we get a 2-color complete
graph K 6. By Example 1, this K 6 contains a monochromatic triangle
which is D, ij k (i > j > k). This means that a = i - k, b = i - j ,
c = J - k. The three numbers are in one group, and

a -b = Ci -k) -Ci - j ) =j - k =c.

We have completed the proof.

Remark In this example, it is possible that b = c, then a = 2b.
The problem can be rewritten as follows. We divide 1, 2, 3, 4, 5 into
two groups A, B randomly. Prove that it is possible to find a number
in a group so that it is twice one number in the group or the sum of two
numbers in the same group.
Question 8 in Exercise 8 of this chapter is an IMO problem in 1978
which is an extension or generalization of this example. Generalize
Question 8 further, we get the famous Schur Theorem (Question 7).
A variate of monochromatic triangle is heterochromous triangle
whose three edges are colored in three distinct colors. The following is
a question from Hungarian Mathematical Olympiad.
Example 6 There are 3n + 1 persons in a club. Any two persons
can play one of the three games: Chinese chess, the game of go,
Chinese checkers. It is known that everyone must play Chinese chess
with n persons, the game of go with n persons and Chinese checkers
with n persons. Prove that among the 3n + 1 persons, there must be
three persons so that there are Chinese chess player, go player, and
96 Graph Theory

Chinese checkers player among the three.

Proof We denote 3n + 1 persons by 3n + 1 vertices. If two persons
play Chinese chess, the corresponding edge of them is colored red. If
two persons play the game of go, the corresponding edge of them is
colored blue. If two persons play Chinese checkers, the corresponding
edge of them is colored black. Then we get a 3-color complete graph
K 3n+1. What we must prove is that in this 3-color complete graph
K 3n + 1 , there must be a heterochromous triangle.
If two edges adjacent to one vertex are not monochromatic, we
call the angle of the two edges heterochromous angle. A triangle is
heterochromous if and only if its three angles are heterochromous
angles. Every vertex is adjacent to 3n edges during which there are n
red edges, n blue edges, n black edges, respectively. Therefore, the

number of the heterochromous angles induced by one vertex is (~) n 2

3n 2. The 3-color complete graph K 3n+1 contains 3n 2 On + 1)
heterochromous angles at all. On the other hand, complete graph
3n +
K 3n+1 contain ( 3
1) 21 On + 1) On -1) triangles. We can regard
= n

these triangles as holes and heterochromous angles as pigeons. Since 3n 2 On +

1) >n On + 1) On - 1), the number of heterochromous angles in the
3-color complete graph K 3,,+1 is twice more than the number of
triangles. By the Pigeonhole Principle, there must be a triangle with three
heterochromous angles. We call the triangle a heterochromous triangle.
We often find the similar problems about the Ramsey problem in
mathematics contest. We give some more examples to finish this
chapter.
Example 7 There are 100 guests in a hall. Everyone of them
knows at least 67 persons. Prove that among these guests you can find
4 persons any two of them know each other. (Polish Mathematical
Competition in 1966)
We denote the guests by 100 vertices A1 , A 2 , ••• , Awo. Join
Proof
every two vertices and color it in red or blue. The edge joining A; and Aj is
 Ramsey's Problem 97

colored in red if and only if A. and Aj know each other. We use the
language of graph theory to re-phrase this problem: In a red-blue two color
complete graph K tOO' if the number of red edges going out of every vertex is
at least 67, then K IOU contains a red complete subgraph K 4 •
Take one vertex AI. The number of red edges induced by it is no
less than 67, so there must exist a red edge A I A 2 • Since the number of
red edges induced by A2 is no less than 67, the number of blue edges
induced by Al and A2 is at most 32 X 2 = 64. They involve 66 vertices
and there must exist one vertex, for example A3 so that AIA3 and
A2A 3 are red edges. The number of blue edges induced by AI, A 2 , A3
is at most 32 X 3 = 96 and these blue edges involve 99 vertices. There
must exist one vertex A4 so that AIA4' A 2A 4 , A3A4 are red edges. So
the complete subgraph K4 with vertices AI, A 2 , A 3 , A4 is red.
Example 8 We use pentagons AlA2A3A4AS and BlB2B3B4BS as
the top and bottom faces of a prism. Every edge and every line
segment AiBj' where i, j = 1, 2, ... , 5, are colored in red or blue.
Every triangle which uses a vertex of the prism as its vertex and a line
segment which has been colored as its edge is not a monochromatic
triangle. Prove that the ten edges in the top and bottom faces are
colored the same color. (The 21th IMO)
Proof First we prove that the five edges on the top face are
colored the same color. Otherwise, there are at least two edges in the
pentagon which are not monochromatic. So there are two adjacent
edges such as A I A 2 , AlAs which are not monochromatic. Without loss
of generality, we suppose that AlA2 is red and AlAs is blue. Among
the edges joining AI and B l , B 2 , B 3 , B 4 , Bs there are at least 3
monochromatic edges. Let AIB i . AIBj' AIBk be red edges (i, j, k
are distinct). Since 6A I BiB j is not monochromatic, BiB j is a blue
edge. Similarly, A2Bi is also a blue edge. We can also know that A2B j
is a red edge. Then 6AlA2Bj is a red triangle. It is a contradiction.
Similarly, we can prove that the five edges on the bottom face are
also monochromatic.
If the edges in the top and bottom face are not monochromatic.
98 Graph Theory

We suppose that AtA2A 3A 4A s is red and BtB2B 3B4B s is blue.

Without loss of generality, let AtB I be a blue edge. By the assumption
that every triangle is not monochromatic, we can know that Al Bs and
AIB2 are all red edges. So AzB 2 and AsB s are blue edges. Similarly,
A sB I' A sB4' AzBI , A 2 B 3, . .. are all red edges and A 3B 3' A4B4 are
blue edges. So A 4BI and A4BZ are blue edges and we can get a blue
triangle D A4Bt B 2 • It is a contradiction.
So the ten edges in the top and bottom faces are monochromatic.
Example 9 There are two international airlines X and Y serving
10 districts. For any two districts, there is only one company
providing a direct flight (to and fro). Prove that there must be a
company which can provide two tour routes so that the two routes do
not pass through the same districts and each route passes through an
odd number of districts.
Proof We denote the 10 districts by 10 vertices Ut , U 2 ' • . . , U IO.
If the flight between U i and U j is provided by X, then we join U i and
U j by a red edge (a solid line): If the flight between U i and U j is

provided by Y, then we join U i and U j by a blue edge (a dotted line).

Then we can get a 2-color complete graph K 10. In order to prove the
conclusion, it suffices to prove that there must be two monochromatic
triangles or polygons having no common edge and an odd number of
edges in K 10 •
The 2-color complete graph K 10 contains a monochromatic
triangle. Let D USU9U1I1 be a monochromatic triangle. By Example 1,
we can know that the triangles constructed by the vertices Ut,
U2' • . • , U 7 must contain a monochromatic triangle . Let D U S U 6 U 7 be a

monochromatic triangle. If the color of DU s U 6 U 7 is the same as that of

D USU9U tu, the conclusion holds. Then let D U SU 6 U 7 be red and
D USU9UlO be blue.
The number of edges joining the vertex sets {u s , U 6' U 7} and {u s ,
U 9' U IO} is 3 x 3 = 9. By the Pigeonhole Principle, there must be five

monochromatic edges . Let them be red edges . The five edges are
induced by {us, U9 ' U lO} , so there must exist a vertex which is
 Ramsey's Problem 99

adjacent to two red edges which are U8U6, U g U7' As Fig. 8. 8 shows,
there must also be another red triangle DU6U7US.

Consider the 2-color complete graph Ks whose vertices are Ul'

U2' U3' U4' us. If the Ks contains a monochromatic triangle,

whatever color the triangle is, together with the red triangle D U6 U 7 Us

or the blue triangle D U s U9 U 10' K 10 contains two

".~"
monochromatic triangles with common edge and
the same color. Otherwise, the 2-color complete
graph K s contains no monochromatic triangle. It
IS easy to know that Ks contains two U9 .~~ - - - - - - - -.. :~.UIO

monochromatic pentagons which are one red and Fig. 8. 8

one blue. We complete the proof.

Remark If we replace 10 districts by 9 districts, the conclusion is
false. An example is given as follows. We divide 9 districts into 3
groups, i.e. {Ul' Uu U 3 ' Up us} = A, {U6' U7' us} =B, {Uy} =
C. The flights among the five districts in A are provided by X. The
flights among the three districts in B are provided by Y. The flights
between A and B, U 9 and A are provided by Y: The flights between Ug

and A are provided by X .

Exercise 8

1 In the space, there are six points. Join every two of them and
color the lines in red or blue. Prove that there must be two
monochromatic triangles.
2 In the space, there are eight points. Join every two of them
and color the lines in two colors. Prove that there must exist three
monochromatic lines which contain no common point.
3 In the space, there are six points. Any three points are not the
vertices of an equilateral triangle. Prove that among these triangles,
there is one triangle whose shortest side is also the longest line of
another triangle.
100 Graph Theory

4 Join nine distinct points on a circle to get 36 lines and color

them in red or blue. Suppose any triangle with three vertices coming
from the nine points contains a red line. Prove that there are four
points and any edge joining two of them is red.
S Prove that among any 19 persons, there must be 3 persons who
know each other or 6 persons who do not know each other.
6 Prove that among any 18 persons, there must be 4 persons who
know each other or do not know each other.
7 We divide the natural numbers 1, 2, . . . , N into n groups.
When N is large enough, there must be a group which contains x, y
and their difference 1x - y I. (Schur Theorem)
8 There are 1978 members in an international corporation. They
come from 6 countries. We label them as 1, 2, . .. , 1978. Prove that
there must be at least one member whose number is twice the number
of his one fellow-country or the sum of two fellow-countrymen.
9 Prove that in a 2-color complete graph K 7 ' there must be two
monochromatic triangles with no common edges.
10 In the space, there are six lines . Among them every three
lines do not lie on a plane. Prove that there must be three lines
satisfying one of the following three conditions: (i) Any two of them
do not lie on a plane . (ii) Any two of them is parallel to each other.
(iii) They meet at one point.
11 Find the minimum positive integer n so that any given n
irrational numbers always contain three irrational numbers among
which the sum of any two is also an irrational number.
12 Find the minimum positive integer n so that when the K is TI

colored by two colors arbitrarily, there must be two monochromatic

triangles which are colored by one color but contain no common edges.
13 In a football league, there are 20 football teams. In the first
round , they are divided into 10 matches. In the second round, they
are also divided into 10 matches. (Notice that the opponent of every
team in different rounds can be the same. ) Prove that before the third
round, you can find 10 teams which have not played with each other.
Chap ter 9 Tournament

In Chapter 1, we have said that the graph is a tool to describe the

special relationship of some objects. The graph in the above chapter is
an undirected graph. The relationship they describe is a symmetrical
relationship. In daily life, many relationships are not symmetrical
such as the relationship of knowing each other. When X knows y, it
does not mean that Y knows X. So is the relationship of winning or
losing in a match. So we can have a new definition of directed graph.
We call a graph directed graph if we assign to every edge of the
graph a direction. We call the edge of a directed graph an arc. If
there is an arc joining the vertices v i and v j and the arrow of the arc
points from v i to V j' we denote it by ( v i ' V j ) and call v i the starting
point and call v j the end point. Generally, we denote the directed
graph by D = (V , U). Here we denote the vertex set of D by V and
the arc set of D by U. Fig. 9.1 shows us a directed graph. The vertex
set is

and the arc set is

U = {(v ,' V2)' (V 2' V3) ' ( vs , V2) ' (vo V2), ( V4' V6),
( VS , V6 )' ( VS , V 4), ( V3 ' v s ), ( V4 ' vs ) } .

The directed graph in this chapter

is also a simple directed graph , which
is a graph without loops (an arc which
starts and ends at the same point) and
without multi-arcs (there are more Fig. 9. 1
102 Graph Theory

than one arc joining v i and v j ) •

We say that v i and v j are adjacent if there is an arc (v i ' V j) or

(v)' Vi) in the arc set of the directed graph G. Otherwise, we say
that v i and v j are not adjacent. We call the number of the arcs whose
starting points are v i an outdegree of Vi' which is denoted by d +(v i).
We call the number of the arcs whose end points are v i an indegree of
v" which is denoted by d- (v i).
We call a directed graph tournament graph if the graph contains n
vertices and there is only an arc joining every two vertices. We denote
the directed graph by K" .
Theorem 1 Let V1' V2' ... , V n be the vertices of tournament
Kn. Then
d +(V1) +d+(V2) + ... + d+(v n )

=d - (V1) +d - (V 2 ) +"'+d-(v n )

1
= 2n(n -1).

Proof Since every arc of K" induces one in degree and one
outdegree and there is only one arc joining every two vertices, the sum
of indegree of every vertices in K n is the same as the sum of outdegree
of every vertices.
d +(V1) +d+(V2) + ... +d +( v n )

=d - (V1) +d-(V2) + "'+d-(v,,)

1
= 2n(n -1).

Example 1 n players P 1 , P 2 , ••• , P" (n > 1) take part in a

round robin. Every player plays only one game with any of other n -1
players. Suppose that there is no tie in the result and we denote the
number of win and lose of P r by w r and l TO respectively. Prove that

wf +w~ + ···+w ;, =n +l~ + ··· + l~.

(The 26th American Putnam Mathematical Competition)
Solution Draw a tournament K". We denote the person P r by
 Tournament 103

the vertex v r ' If Pi defeats P j ' we join v i and v j to get an arc (v i ,

Vj) . So W rand lr are the indegree and outdegree of v r respectively.
By Theorem 1,

Wl+W2 + ···+Wn =ll+ l 2+···+ l n.

Note that Wi +li = n -1 C1 ~ i ~ n),

WT + w~ + ... +w~ - (iT +l~ + ... + l~)

= (WT - in +(W~ - lD +··· + (w~ - l~)
= (WI + ll)(wl -ll) + (W 2 +l2)(W 2 - l2) +··· + (w n + In)(w n - In)
=(n - 1)[(wl +W2+ ···+Wn) - (l1 +l2+··· + ln)J =0.

So
wf +W ~ + ... +W~ = If + n + ... + l~.

In a directed graph D = (v, u), there exists a sequence of distinct

arcs U I ' U 2' . • . , Un . If the starting point of U i is v i and the end
point of Ui is Vi+1 Ci = 1, 2, ... , n). We call n the length of the
directed path. VI is the starting point of the path and v n + I is the end
point. If v I = V n+ 1 , we call the path a circuit.
Example 2 The MO space city consists of 99 space stations. Any
two stations are connected by a channel. Among these channels there
are 99 two-way channels and others are one-way channels. If four
space stations can be arrived at from one to another, we call the set of
four space stations a connected four-station group.
Design a scheme for the space city so that we get the maximum
number of connected four-station groups. (Find the exact number and
prove your conclusion.) (The 14th China Mathematical Olympiad)
Solution We call an unconnecte@ four-station group a bad four-
station group. A bad four-station group has three possible situations:
(1) Station A has three channels AB , AC, AD which all leave A.
(2) Station A has three channels which all arrive at A.
(3) Stations A and B, stations C and D have two-way channels
but the channels AC, AD all leave A, and BC, BD all leave B.
We denote all the bad four-station groups in (1) by S and others
104 Graph Theory

by T. Let us calculate IS I.

Since the space city contains C;) - 99 = 99 x 48 one-way

channels . We denote the number of channels leaving the i - th station

by S i ' So
99

~Si = 99 X 48.
i= 1

Now the number of bad four-station group in (1) which contains

three channels induced by A is (~i). So

IS I= ~
99
(
S.)~ 99 x (48) .
I

1~ 1 3 3

The above inequality holds because (~) = ~ x(x - D(x - 2) for

x ~ 3 is a convex function . Since the number of all four-station groups

is (9:), so the number of connected four-station groups is no

more than

Then we give an example so that the number of the connected

four-station groups is (9:) - 99(~8) . Let the number of the channels

from and to every station Ai be both 48. Every station has two two-
way channels and there are only type S groups of bad four-stations and
no type T groups of bad four-stations.
We put 99 stations on the vertices of an inscribed polygon with 99
sides and assume the longest diagonals of the regular polygon with 99
sides all two-way channels. So every vertex is adjacent to two two-way
channels. For station A i , there are one-way channels leaving Ai and
joining the next 48 stations in the clockwise direction, and one-way
 Tournament 105

channels arriving at Ai and joining the next 48 stations in the counter-

clockwise stations. Then we will prove there are only type S groups of
bad four-station groups and no T groups of bad four-station groups.
Suppose {A, B, C, D} is a four-station group.
(i) If there are two two-way channels in the four-station group,
clearly they are connected.
Cii) If there are only one two-way channel AC in the four-station
group, each of Band D forms a cycle with A, C. Of course, they are
connected.
(iii) So if the bad four-station group contains no two-way
channel, it can only be one of (1) and (2). If it is (2), without loss of
generality, we suppose that the 3 channels of A all arrive at A and B,
C, D all come from the next 48 stations of A in the clockwise
direction . Let D be the farthest station from A. So AD, BD, CD all
leave from D, which means that all the bad four-station groups are of
the type S group.

In summary, there are at most C:) - 99(~8) connected four-

station groups.
Theorem 2 There exists a vertex in a tournament so that there is a
path from it to any other vertices. The maximum length of the paths
is 2.
Proof Suppose that the vertex with the maximum outdegree of a
tournament Kn is v\. We denote the end point set of the arcs whose
starting point is V1 by N+ (v 1). If the conclusion is false, there must be
a vertex V 2 (v 2 eft v \ ), where v 2 tl: N + (v \ ). For every vertex u E
N+ (v \) , there is one arc (V2' u) from V2 to u together with the arc
(V2' v\) . So d+ (v 2) "?:- d+ (v \) + 1, which contradicts the fact that

degree of v\ is maximum. The proof is complete.

Example 3 Every athlete who takes part in the single round robin
must play one game with any other athlete and there is no tie. Prove
that among these players, you can find such athletes that the persons
who were defeated by him and the persons who were defeated by the
106 Graph Theory

person he defeats can contain all other athletes. (Hungarian

mathematic contest)
Use tournament on the round robin, which is Theorem 2. We
omit the proof.
Example 4 n (n ~ 3) athletes take part in a single round robin
and use the result to find good athletes. The requirement that A is
selected to be a good athlete is that for any other athlete B, either A
defeats B or there exists C so that C defeats B and A defeats B. If
only one athlete meets the above requirement, show that he defeats
any other athletes.
Solution We denote n athletes by n vertices. If v i defeats v j , we
draw an arc from v i to v j and get a tournament K n. Without loss of
generality, we suppose that the outdegree of VI is maximum in the K" ,
according to Theorem 2, VI is a good athlete. What we will prove is
that VI can arrive at any other vertex by a path whose length is 1,
which means that the indegree of VI is d- (v I) = O.
Suppose that the proposition is false. We denote the set of arcs
with starting point VI by N- (v I) = {v iI' V i2' ••• , Vir}' r ~ 1.
Consider the K r consisting of ViI' V i 2 ' ••• , Vir. We suppose that the
outdegree of v i l is maximum in K r. By Theorem 2, the length of path
from v il to each of v i2' •.• , V ir is no more than 2. Since VI can arrive
at other vertices except {v i 1 , Vi'
2
... , Vi' Vi}'
r 1
then Vi
1
can arrive
at vertices except ViI' V i 2 ' . . . , V ir through the paths whose length is
no more than 2. Therefore, in the tournament K r , Vi
I
can arrive at
any other vertex through a path whose length is no more than 2.
Hence ViI is also a good athlete, which contradicts the fact that VI is
the only good athlete. So N- (V I) = 0 or d- (v I) = O. We have
completed the proof.
Remark This problem gives a property of the tournament IL: If
the vertex of Kn with the maximum outdegree is unique, then the
outdegree of this vertex is n - 1.
Theorem 3 Tournament K n contains a Hamiltonian path whose
 Tournament 107

length is n - 1.
Proof Apply induction on the number of vertices n. When n = 2,
clearly the proposition is true.
Suppose the proposition is true for n ~ k. When n = k + 1, from
k + 1 vertices we take a vertex v . Remove v and the arcs adjacent to
v from R u 1 . By induction, R k+ 1 - v contains a Hamiltonian path VI'

If there is an arc (v k' v), then VI' V2' .•. , V k' V IS a

Hamiltonian path. If there is an arc (v , VI)' then v , VI' V2' ... , V k

is also a Hamiltonian path.

V
Otherwise there exist arc
(v, and (VI' v). Then there
V k)

must be an i C1 ~ i ~ k - 1) so
that the arcs (v;, v), (v, Vi + I) VI V2 v)

both exist. Now VI' . . . , v;, v, Fig.9.2

v; + l ' . . . , V k is a Hamiltonian
path as Fig . 9. 2 shows us.
Example 5 In a match of Chinese chess, every two players should
playa game. Prove that we can label the players so that every player is
not defeated by the player whose number follows immediately
after his .
Solution Suppose there are n players. We denote n players by n
vertices VI' V2' ... , v". When v i is not defeated by v j' we draw an
arc from v i to V j to get (v i' V j ). Then we get a tournament K". By
Theorem 3 , K" contains a Hamiltonian path, so we can label them
according to the order of the path.
Theorem 4 The tournament K" (n ;?-3) contains a circuit which is
a triangle, if and only if there are two vertices v and v' satisfying

Proof Let vertices v and v' satisfy d+ (v) d+ (v'). We will

prove that K" contains a circuit which is a triangle.
Without loss of generality, we assume that there is an arc ( v , v')
108 Graph Theory

and draw arcs from V'to everyone of VI' V2' ... , V k where (k =
d+ (v)). Then there must be V j C1 ~ j ~ k) so that there is an arc from
Vj to v. Otherwise, d+ (v) ~ k +1 > d+ (v') and v, v', V j form a
triangle . We have proved the sufficient condition .
If the outdegree of every vertex of K n is different, we can prove
by induction that K n contains no triangular circuit. When n = 3, it is
easy to see that the outdegree of a triangle is 0, 1, 2 and the triangle
cannot form a circuit.
Suppose the proposition is true for n = k. Consider the tournament
K n +I ' If the outdegree of every vertex is different, they are 0, 1,
2, "', k in order. Supposed+(v ' ) = k. Remove the vertex v' and its
adjacent arcs. By induction hypothesis, Kk - v' contains no triangular
circuit. Clearly K k+1 contains no triangular circuit. We have proved
the necessary condition.

Exercise 9

1 Among n (n > 4) cities, every two cities have a path to join

them. Prove that we can change these paths to one-way path so that it
is possible to go from one city to another city through at most one
other city.
2 If a tournament K n contains a circuit, then K n contains a
trianglular circuit.
3 In a country, N cities are connected by air routes. For any
route, the airplanes can fly only along one direction. An air route
satisfies the condition f: Any plane which starts from one city cannot
return to the same city. Prove that it is possible to design an airline
system so that every two cities are connected by an air route and the
system also satisfies condition f.
4 In a volleyball round robin, if team A defeats team B or team
A defeats team C and team C defeats team B, we say that A is
 Tournament 109

superior to B and we also call the team superior to any other team the
champion. According to this regulation, can two teams both win the
championship?
5 n players take part in a match in which everyone plays with
several other players. Suppose that there is no tie in a game . If the
result that VI defeats V2' V2 defeats V 3 ' . . . , V k defeats VI does not
appear. Prove that there must be a player who wins all games and
another player who loses all games.
6 If among n persons v I ' V2' . . • , V n every two persons v i and
v j have one ancestry v k. Everyone can be an ancestry of oneself.
Prove that the n persons have a common ancestry.
7 A, B, C, D play table tennis and every two persons should
play against each other. At last, A defeats D and A, B , C win the
same number of games. How many games does C win?
8 n (n:;?o 3) players take part in a round robin. Every pair should
playa game and there is no tie . There is no player who defeats all
other players. Prove that among them there must be three persons A ,
B, C so that A defeats B, B defeats C and C defeats A.
9 There are 100 species of insects. Among every two of them
there is one species who can eliminate another species. (But A
eliminates B, B eliminates C, which does not mean that A eliminates
C . ) Prove that the 100 species of insects can be arranged in an order so
that any species can perish another species next to it.
 Solutions <1
--.::.--=::::.:::.,--.:.:,

Exercise 1

1 The graph G is shown in Fig. 1.

2 In a simple graph, every edge is
adjacent to two distinct vertices and among
every two vertices there is at most one edge
joining them. Now there are n vertices, so the

number of edges is at most G) = n en2- 1) . Fig. 1

Since not all simple graphs are complete graphs, then e ~ n en; 1) .

3 As shown in Fig. 2, we establish the relationship of the

vertices: V I +--+ U l ' V2 +--+ U 2 ' V3 +--+ U 3 ' V 4 +--+ U 4' Vs +--+ U s and the
relationship of the edges: e l+--+e~, e2 +--+ e; , e3 +--+ e~, ... , e8 +--+ e~ . The
number of the vertices and that of the edges of the two graphs are
equal. So we can establish a corresponding relationship between the
vertices and edges of the two graphs . So the two graphs are
isomorphic.

e,

( I)

Fig. 2
 Solutions 111

4 Construct a graph as follows. We denote a medicine-chest by a

vertex. We denote every two medicine-chests v i and v j which contain
one common medicine by the edges (v i ' V j ). By the hypothesis, the
graph is a complete graph K" and the number of the kinds of the
medicine is equal to the number of the edges ; n (n - 1).

5 We denote n professors by n vertices Vt , V2' . . . , v" and join

by an edge if the two corresponding persons know each other. We
divide the n persons into two groups randomly. There are finite ways
to divide them. Conside the number of edges joining the vertices of the
two groups. There must be a division so that S is the largest. Now
d i '?d; = 1, 2, ... , n). Otherwise, ifd t < d~, we transfer VI from
(i
a group to the other group. The number of S increases by d; -d l > 0,
which contradicts the fact that S is the largest.
6 Team A had played 8 matches with 8 teams and did not play
with the other 9 teams. Suppose the 9 teams had played with each
other in 8 rounds. Since every team had played 8 games, the 9 teams
had not played with other teams. But the 9 teams could only play 4
games, so there must be one team which had played with other teams.
A contradiction. So among the 9 teams there must be two teams Band
C which had not played with each other. Then A, B, C had not
played with each other.
7 We denote n delegates by n vertices. If two delegates have
shaken their hands, we join the corresponding vertices and get the
graph G. If among any four vertices Vt, V2' V3' V4 in G, every vertex
has its adjacent vertex, we denote them by V/t, v~, v~, v:. By the
known condition, among VI' V2' V3' V4 there is a vertex VI which is
not adjacent to the other three vertices SO V'I -=F v 2' V 3 ,
V2' V3' V4.
I ' f ·
V4. If V2 -=F Vj, among our vertices V2'
h .
V3' Vt, V2' t ere IS no vertex
I I

which is adjacent to the other three vertices. SO V '2 = V/t. Similarly,

v; = v~. Among four vertices Vt , V2' V3' v; , there is no vertex which
is adjacent to any other vertex. So among any 4 vertices there must be
one vertex which is not adjacent to the other n - 1 vertices.
112 Graph Theory

8 We denote these students by 3n vertices . We divide the

students from three schools into three vertex set which is denoted by
X, Y and Z. If u and v are from different schools and they know each
other, then join the vertices representing them to get a graph G .
Suppose x EX, we denote the number of vertices which are adjacent
to x and lie in Y and Z by k and l respectively . Then k +l = n + 1. We
denote the maximum number between k and l by m (x ). Let x go
through every in X and the maximum value of m (x) is denoted by
m x. Similarly, we definite my and m z . We denote the maximum of
m x, my and m z by m. Without loss of generality, suppose that m =

m x and Xo E X so that I Y1 I= m. ( IY 1 I is the number of the vertices

in Y adjacent to Xo. ) The number of vertices in Z adjacent to Xo is n +
1- m ~ 1. Suppose Zo E Z is adjacent to xo. If there is Yo E Y 1
adjacent to zo, then 6xoyozo is a triangle in G. If every vertex y in
Y1 is not adjacent to zo, the number of vertices in Y adjacent to Zo is
no more than n - m. So the number of vertices in X adjacent to Zo is
no less than n +1 - (n - m) = m + 1, which contradicts the fact that
m is maximum. Then we prove G must contain 6xoyozo . For a
diagram, see Fig.3.
9 When n = 1, there are two red squares which are adjacent.
Clearly, it is a rectangle. Suppose it is true when n ~ k , which means
we can divide 2k connected squares into k rectangles. When n = k + 1,
(i) For 2k + 2 squares, if we remove a pair of adjacent red squares
from the graph, the graph is also connected. By induction, the
conclusion is true.
(ii) If we remove a pair of adjacent red squares from the graph,
the graph becomes several connected
subgraphs. The number of red squares in
every subgraph is even. By induction, the
conclusion is true .
(iii) If we remove a pair of adjacent red
squares from the graph, there is a connected
subgraph in which there are odd red Fig. 3
 Solutions 113

squares:
When n = 2, there is a graph in the shape of the letter T with two
rectangles 1 x 3 and 1 x 1 satisfying the requirement.
When n =F 2, consider the T -shaped graph at the top-left corner of
all the squares. After removing these two squares, we get at most two
connected graphs .
If the number of the red squares of the two connected subgraphs is
odd after removing the two squares at the top-left corner, then remove
the two rectangles 1 x 3 and 1 Xl. It is easy to know they are also
connected and the number of red squares is even. In summary, when
n = k + 1, the conclusion is true. So the conclusion is true.
10 If 2000 members know each other, then the number of all the
members in the delegation is 2000. Therefore, we suppose that two
members u and v do not know each other. We prove in three steps.
( i) Any two members other than u and v know each other.
Suppose that a and b are another the two members. By the hypothesis,
among the four members a, b, u, v, there must be a person who
knows another three persons. The person must be a or b, which means
a and b know each other.
(ii) If u and v know everyone of the remaining 1998 persons,
then in the delegation there are 1998 persons who know all the other
persons. Suppose that a is any member other than u, v. By the
hypothesis, a knows u, v. Suppose that b is another member. By the
above proof, a and b know each other. By the fact that b is arbitrary,
a knows every other member. Also by the fact that a is arbitrary, the
1998 members in delegation other than u and v know all other
members.
(iii) If one of u, v does not know all other 1998 members, then
there are 1997 members knowing all the members. Suppose, other
than v, u does not know another member w. Let a be anyone of 1997
member other than u, v, w. By the hypothesis, among a, u, v, w
there is only one person a who knows the other three. It means that
each of u, v, w knows the other 1997 persons of the delegation.
114 Graph Theory

In summary, the number of persons who know all members of the

delegation is at least 1997.
11 By assumption, everyone has friends. Suppose that k (k ~ m)
persons are friends . Since they have a common friend, k + 1 persons
are friends. Similarly, we can deduce that m + 1 persons A l , A 2 , ••• ,

A m+l are friends. Next we will prove that there are no other persons in
the carriage except Al , A 2 , ••• , A m+1 • If B is a person other than the
m +1 persons and he has made friends with at least two of A 1 ,
A2, A m+1 • Suppose Band A 1 , A2 are friends. Then the m
••• ,

persons B, A 3 , A 4 , ••• , A m+l have two common friends A 1 , A 2 ,

which contradicts the given condition. So the person B other than A 1 ,
A 2, ... , A m+ l can make friends with at most one of A 1 , A 2 , ••• ,
A m+1 • Without loss of generality , we suppose that A 2 , A 3 , • •• , Am+l
except A 1 cannot be the friends of B. Of course, the common friend C
of m passengers B, A 1 , A 2 , •• • , A m+ l is neither A 2 , A 3 , • • • , A,n-l
nor A 1 • Since m ~ 3 , C makes friends with m - 1 ~ 2 persons among
A 1 , A 2 , ••• , A m+1 • It contradicts the fact that C makes friends with
at most one of Al , A 2 , •• • , A m+1 • So in the carriage, there are only
m + 1 passengers Al , A 2 , • • • , A m+ l each of whom has m friends.

12 Ks contains C) = 10 edges and C) = 10 triangles. The

number of triangles which is related to every edge is 3. If the number

of the edges of the graph is no less than 7, we remove at most 3 edges
from K s. Hence we can remove at most 3 X 3 = 9 triangles, so there is
still a triangle. It contradicts the condition, so the graph cannot
contain 7 or more than 7 edges.

Exercise 2
n

1 Since.z= d ( v;) = 2e, n (; ~ 2e ~ n D. , then (; ~ 2e ~ D..

; ~1 n
n

2 .z= d ( v ;) = 2 (n + 1), the average degree of the vertices IS

i= !
 Solutions 115

2(n + 1) > 2,
=-:..'-'---'-----"'-'-- hence there is at least one vertex whose degree is no less
n
than 3.
3 We denote the faces of a polyhedron by vertices. We join two
vertices if and only if the faces which are represented by the vertices
have a common edge. We obtain a graph G. By assumption, the
number of the odd-degree vertices is odd. So such polyhedron does not
exist.
4 No. Use Theorem 2.
5 Construct a graph G as follows. We denote 123 persons by 123
vertices VI , V2' . . . , V123. If two persons have discussed the problem,
we join the corresponding vertices representing them. So the degree of
every vertex in the graph is no less than 5. If the graph contains no
vertex whose degree is more than 5, the degree of every vertex in the
graph is 5. It means that the number of odd vertices in the graph is
odd, which is possible. Then the graph contains at least one vertex
whose degree is more than 5.
6 Construct a graph G as follows. We denote n congressmen by
n vertices. If two persons do not know each other, we join them in
the graph G. For every vertex v" deVi) ~ 3. Now divide G into
two components G I and G 2 • For two vertices in one component, if
there is an edge joining them in the original graph, we keep the
edge. For two vertices in the different components, if there is an
edge joining them in the original graph, the edge will not exist. The
removed edges form a set E. In the two components, suppose there
is a vertex whose degree is more than 1. Without loss of generality,
assume that G I contains a vertex VI , de V I) ~ 2. Then we move the
vertex to G 2 • Then G I will lose two edges. Since de v I) ~ 3, G 2
increases by at most one edge and then E increases by at least one
edge. Repeat this process. The number of E is increasing, but the
total number of edges is finite. So at one point, in the two
components, there is no vertex whose degree is more than 1. We
complete the proof.
116 Graph Theory

7 The problem can be rephrased as follows. In a graph G with

2n vertices, the degree of every vertex is no Less than n. Prove that G
contains a rectangle. If G = K 2n , the conclusion is true. If G oft K 2n ,

there exist vertices VI and '02 which are not adjacent. Since d Cv I) +
d C'02) ~ 2n, by the Pigeonhole Principle, among the remaining 2n - 2
vertices there must be two vertices '03 ' '04 which are adjacent to both
VI' '02. Then the 4 vertices form one rectangle.
8 Construct a graph G as follows. We denote 9 persons by 9
vertices. The two vertices are adjacent if and only if they have shaken
their hands. Since d C'09) = 6, there exists v k oft V I , '02' '0 3 which is
adjacent to '09. Clearly d C'Ok) ~ 5. Among the other 5 vertices adjacent
to '09 there is a vertex Vh adjacent to 'Ok. COtherwise, d C'Ok) ~ 9- 5- 1 =

3. ) Then V 9 ' Vh' Vk are what we need.

9 We denote the 14 persons by 14 vertices { VI' V2' .. • , VI4}.

Two vertices Vi' Vj are adjacent if and only if they have not
cooperated, then we get graph G . The degree of every vertex in G is
8 . After playing three sets, we will remove 6 edges. So there exist at
least two vertices whose degrees are also 8. Let one of them be VI'

among the 8 vertices adjacent to VI there must be one vertex whose

degree is no less than 7. We know V2 is adjacent to one vertex V3 which
is another vertex adjacent to VI. Then VI' V2' V 3 and a new vertex V

form K 4 •
10 Every two vertices are adjacent if the distance between them

is given. We prove this problem by induction. When n = 4, ~ n(n - 1) +

4 = 6. Among the 4 vertices, there are 6 distances which are all given.
The conclusion is true. Suppose n = k Ck ?:: 4) , the conclusion is true.

When n = k + 1, there are ~ Ck + 1) Ck - 2) +4 edges in the vertex set.

Let A k + 1 be the vertex with the smallest degree in the vertex set. Its
degree is
 Solutions 117

2[ ~(k + 1)(k -2) +4J

d (A k + l ) ~ k +1
8
= k-2 + k + 1

8
~ k-2+5 < k.

So d (A k+l ) ~ k - 1. Then among the remaining k vertices AI' A 2 ,

A 3 , ... , Ak there are at least

~ (k + 1) (k - 2) +4 - (k + 1) = ~ k (k - 3) +4
edges. By induction, the set of k vertices is stable. Also,

so A k+1 is adjacent to at least three vertices among Al , A 2 , ••• , A k .

Suppose that A k+1 is adjacent to vertices AI, A 2 , A 3 and thatAk+IA I =
x, A k+IA 2 = y, A k+I A 3 = z. It is easy to prove A k+1 can be uniquely
determined. If not, let A~+ I be another vertex. Also A'k+IA I = x,
A:+ I A 2 = y, A' k+IA 3 = z, then AI, A 2 , A 3 are all on the
perpendicular bisector of Ak+IA:+ I . It contradicts the hypothesis that
there are no three vertices on a common line. Then A k + 1 A 4 , ••• ,
Ak+IA k can be determined. The set { A I , A 2 , ••• , A k+l } is stable.
The conclusion is true when n = k + 1. In summary, the conclusion is
true.
11 We use the unit cube as the vertex. We join the two
corresponding vertices if and only if there is a common face between
the two unit cubes and we get a graph G. The number of edges of its
complementary graph G is what we need. It is easy to know the
number of edges of G is 3n 2 ( n - 1), the number of edges of K ,,3 is
1 -
2,n 3 (n 3 -1) and the number of edges of G is
118 Graph Theory

The number of the pairs of unit cubes which have no more than 2
common vertices is ; n 6 - ~ n 3 + 3n 2 at all.
12 We consider the connected part of the routes which contain
the capital. It is a connected graph. What we must prove is that this
connected graph contains city A. If not, one vertex (capital) of the
graph is adjacent to 21 edges, so every other vertex is all adjacent to 20
vertices. It means that the graph contains only one odd vertex, which
is impossible.

Exercise 3

1 Since C; 1 X 1 = C; 1 y 1 is equal to the total number of edges,

then 1 X 1 = 1 Y I .
2 We will prove the cases when n is odd and when n is even.
Here we only prove the latter. Suppose the theorem is true for all even
numbers n ~ 2k. Let G be a graph which contains 2k + 2 vertices and
no triangle. Remove two adjacent vertices v and v' from G to get G'.

By induction, G' contains at most [4!2 ] = k 2 edges. Since G contains

no triangle and any vertex v" cannot be adjacent to v, v' at the same
time, G contains at most

edges, where l is the number of the vertices adjacent to v in G'.

3 Construct a complete bigraph K 10, 10.
4 Use Theorem 1.
5 (1) Suppose that n = m k +r (0 ~ r < m). Then by the

definition of T m. n , em(n) = G) - r(k :l)_(m - r)(~), replace r

by n - mk and simplify it to get the solution.
(2) Let the number of the vertices of m parts of a complete
m-partite graph G be nl' nz, . .. , nm respectively. If G is not
 Solutions 119

isomorphic to T m Cn ), then there exists n i - n j > 1. Consider a

complete m-partite graph G' , the number of vertices of its m parts is nl ,
n 2' . .. , n i - 1, . .. , n j + 1 , . .. , n m • Since

e CG / )

1
+-
2 (n - n J - 1)(n J + 1)
= eCG) + (n i- n j ) - 1 > eCG) .

If G' is isomorphic to T m (n ), we complete the proof. Otherwise,

repeat the above step until we find a graph which is isomorphic
to T m . n '
6 Construct a bigraph G = CX, Y ; E) as follows. We denote
every student from country X by a vertex in X and every student from
country Y by a vertex in Y. If a student from X has danced with a
student from y, then join the vertices corresponding to them. Suppose
that the degree of x is the largest in X. Since d ( x ) < n, in y, there is
a vertex y' which is not adjacent to x. Suppose that x ' in X is adjacent
to y ' . Since except y', there are d ( x ') - 1 vertices adjacent to x' and
de x /) - 1 ~ de x ) - 1 < de x ), so there must be a vertex y which is
adjacent to x but not adjacent to x ' . Then we get four vertices x , x ' ,
y, y' corresponding to four persons who satisfy the requirement.
7 Construct a graph G as follows. We denote 14 persons by 14
vertices . Two vertices are adjacent if and only if these two
corresponding persons have not cooperated . By assumption, there are

C24) = 91 pairs among 14 persons . Everyone has teamed up with the

14 X 5
other 5 persons, so there are - 2- = 35 pairs. Now they play 3 sets

and have 6 new pairs . The number of edges of G is 91 - 35 - 6 = 50 ,

but e2 (14) = 49. By Tunln's theorem, G contains K 3 and the travelers
120 Graph Theory

corresponding to three vertices can playa set with the new traveler.
8 SetG = (V, E). There ared(x){n -1 - d(x)} triple group
{x, y, z}. They do not form a triangle in G or G, and x E V is the
end of the only edge in G. Every triple group {x, y, z} which does
not form a triangle in G or G contains one or two edges of G. Suppose
that (x, y) is one edge of G and (x, z), (y, z) are two edges of G.
In the sum ~d(x){n -1 -d(x)}, the triple group {x, y, z} has
rEV
been counted twice: one is about x, the other is about y. If (x, y),
(y, z) are the edge of G, (x, z) is the edge of G, in above sum, the
tuple group {x, y, z} has also been counted twice: one involves x,
the other involves z. The sum of triangles in G and G is

G)- ~ ~rEvd(x)(n -1 - d(x»

1
= 24n(n -D(n - 5).

9 Suppose there is no (k + 1) -element set which we need. We

will prove

m < -'-(k'-'-----~1)'_(~n'_-----.:..:k'_'_)_+_'_____'_=k ( n )
k2 k - 1 .

We denote all the red k -subsets by S and all the (k - 1) -subset by

f3 . For any (k - 1) -subset B, we denote the number of red k -element
subsets which contain B by a (B). For any A E S, A contains k (k -
1)-element subsets. For any element x E X\A, x together with at
most k - 1 of k (k - 1) -element subsets form a red k -element subset.
( Otherwise, there exists (k + 1) -element subset and all its k -element
subsets are red k -element subsets. ) So
~ a(B)«n - k)(k-1)+k.
BcA, IBI= k-l
So
 Solutions 121

m [ Cn - k)Ck - 1) + k] ~ ~ ~ aCB)
A E SB C A. IB I ~ k - l

= ~CaCB»2
B E {3

~ m((;{3aCB») 2
~
1
- --Ckm)2.
(k ~ 1)
So

[Cn - k)(k - n + k]( n )

~ _ _ __ _ ~_ _~k~-_1~
m ----, k2

Since

[Cn - k)Ck - n + k]( n )

m > _ ____~~--_k---1-
k2
there must exist a Ck + 1) -element subset of X such that all k-element
subsets are red k-element subsets.

10 Since C20 ) = 45, then a complete graph with 10 vertices

contains 45 edges . The figure in the problem is obtained by removing 5

edges from the complete graph with 10 vertices. We call the 5 edges
"Removed Edges" and denote 10 vertices by A l ' A 2' . . . , A 10 .
Without loss of generality, let A 1 A2 be a "Removed Edge", then
remove Al and its incident edges. The deduced graph with 9 vertices
contains at most 4 "Removed Edges". Without loss of generality, let
A2A 3 be a "Removed Edge". Then remove A2 and its incident edges.
CIf there is no "Removed Edge", remove any vertex . The same in the
later step.) The deduced graph with 8 vertices contains at most 3
"Removed Edges". Without loss of generality, let A3A4 be a
"Removed Edge". Then remove A3 and its incident edges. The
deduced graph with 7 vertices contains at most 2 "Removed Edges".
122 Graph Theory

Without loss of generality, let A4AS be a "Removed Edge". Then

remove A4 and its incident edges. The deduced graph with 6 vertices
contains at most 1"Removed Edge". The deduced graph is a complete
graph with 6 vertices or the graph which is obtained by removing one
edge from the complete graph with 6 vertices. Anyway, the graph
must contain a complete bipartite graph K 3 , 3 . We can generalize this
problem: If a graph with n vertices and m edges contains no K n, n' We
can prove that m < C • n 2--;' , where C depends on r.
11 We denote the positions of 18 police cars by 18 vertices Xl'

X2' ... , X lS . Suppose

By Theorem 3, 2 - [182J
I E I? ( 18) 3 = 45. It means there are at least

45 pairs of cars which can communicate with each other. If the above
condition of the graph does not hold, then there does not exist two
vertices whose degrees are more than 5 and

I E I~ ~ (1 X 17 + 4 X 17) < 43,

a contradiction.
12 When n = 2, n 2 + 1 = 5. There are 5 line segments among 4
vertices, which form two triangles . Suppose that proposition is true
for n = k. When n = k + 1, let us prove that there exists at least one
triangle . Suppose that AB is a given line segment and denote the
number of line segments from A and B to other 2k points by a and b.
( 1) If a +b
? 2k + 1 , there exists a vertex C other than A and B
so that AC and BC exist. Then there exists a triangle L:,.ABC.
(2) If a + b ~2k, if we remove A and B, among the remaining 2k
vertices, there exist at least k 2 + 1 line segments. By induction, there
exists a triangle.
Suppose L:,.ABC is a triangle formed by these line segments. We
 Solutions 123

denote the number of line segments from A, Band C to other 2k - 1

points by a, (3 and y.
(3) If a + (3 + y ~ 3k - 1, the total number of the triangles
including one of AB, BC, CA as an edge is at least k . These k
triangles together with L"ABC give k + 1 triangles.
(4) If a + (3 + Y ~ 3k - 2, there is at least one number which is no
more than 2k - 2 among the three numbers a +(3, (3 + y, y + a. Without
loss of generality, a +(3 ~ 2k - 2. When we remove two vertices A, B,
among the remaining 2k vertices there exist at least k 2 + 1 line
segments. By induction, there exist at least k triangles. These k
triangles together with L"ABC give k + 1 triangles. The proposition is
true when n = k + 1. We complete the proof by induction.

Exercise 4

1 The spanning tree of graph G contains two pendant vertices.

Remove these two vertices and the graph is still connected.
2 There are 9 X 9 = 81 vertices whose degrees are 4, so we should

remove at least [821 ] + 1 = 41 edges so that the degree of every vertex is

less than 4 . We can remove at most 2 X 11 X 10 - 120 = 100 edges so

that the graph is still connected.
3 The proposition is not true. Take K 3 and an isolated vertex
(the vertex which is not adjacent to any other vertex) to construct a
graph G. Then G contains 4 vertices and 3 edges, so it is not connected
and clearly not a tree.
4 (1) Suppose that T has x pendant vertices. The number of
vertices of the tree T is n = 3 + 1 + x, the edge number is e = n - 1 =
11

X + 3. 2: d (v,)
i= l
= 3 X3 +2 Xl +1Xx = 11 + x, so 11 + x = 2(x +
3), x = 5.
(2) Fig. 4 shows us two trees satisfying the requirement but they
are not isomorphic.
124 Graph Theory

Fig. 4

5 Suppose T contains n vertices and e edges. Then n

e = n - 1,
k k

~ in i = 2e = 2n - 2 = 2~ n i - 2.
i= 1 i= 1

So
k

nl = ~(i - 2)n i +2.

i= 2

For r ~ 3, by the above equality, we can obtain

6 Among d l ' d 2' ... , d n' there must be at least two which is

equal to 1. (Otherwise, t
i =l
di ~ 2n - 1). We apply induction on the

number of vertices n . When n = 2, the proposition is true. Suppose

that the conclusion is true when n = k . When n = k + 1, there exists a
number 1 among d 1 , d 2 , ••• , d k , d k + 1 • Without loss of generality, let
d k + l = 1. It is easy to know among the k + 1 numbers there exists a
number which is no less than 2, denoted by d k. Consider the k
numbersd 1 , d 2 , ••• ,dk-l , (d k - 1),

d 1 + .. ·+dk-l + (d k - 1) = 2(k + 1) - 2 - 1 - 1 = 2k - 2.

By induction, there exists tree T' whose vertices are VI' ••. , V k ,

~d(Vi) = d 1 + ... +d k - 1 + (d k - 1) = 2k - 2.
i=1

In T', there is an edge which is from Vk to Vk + l ' We obtain a tree

 Solutions 125

T, then
k+1
~d(v;) = 2k -2 +1 + 1 = 2(k + 1) - 2.
i= l

T is what we need.
7 Construct a graph G, we use the ends of n line segments as the
vertices of G and the line segments as the edges. By the condition, G is
connected and contains no loop. So G is a tree where the length of its
longest chain is 2. So G contains only one vertex which is not a
pendant vertex. The vertex is the common vertex of n line segments.
8 Refer to Example 6 in this chapter.
9 Suppose the conclusion is not true, then there must exist a
counterexample. Consider the counterexample in which 1E 1+1 V 1is
the smallest. In this counterexample, 1E 1= 1V 1+ 4. (Otherwise, we
can remove more edges and still get a counterexample, where 1 E 1 +
1V 1is smaller. A contradiction!) Then 1E 1> 1V I. The graph must
contain a cycle. The length of the shortest cycle is at least 5.
(Otherwise, the length of the shortest cycle is no more than 4, then
we remove this cycle. We still have 1E 1;;: 1V I. There still exists a
cycle. The cycle and the above cycle contain a common edge. A
contradiction! )
Furthermore, the degree of every vertex is at least 3. Otherwise,
if the degree of a vertex is 2, remove this vertex and change the two
edges adjacent to this vertex to one edge. We still have 1E 1= 1V 1 +
4 and 1E 1+ 1V 1gets smaller. A contradiction! If the degree of one
vertex is 1, remove this vertex and its adjacent edges, we still have
1E 1= 1V 1+4 and 1E 1+1 V 1is smaller. A contradiction! If there
exists an isolated vertex, remove it and 1E I > 1V 1+ 4 and 1E 1+
1 V 1 is smaller. A contradiction! Take a cycle Co, the length of which
is at least 5. The cycle contains at least 5 vertices. For every vertex on
the Co, it is adjacent to at least one edge which is not on the cycle and
the adjacent vertices are distinct. (Otherwise, there exists a cycle
whose length is less than 5. ) Then it is easy to see 1 V I ;;: 2 X 5 = 10.
126 Graph Theory

On the other hand, 2 1 E 1 = ~d (V) > ~3 = 3 1 V I. 1 E 1 = 1V 1 +

v EV v EV

4, so 2 1 V 1 + 8 ;;:, 3 1 VI, 1 V 1 ~ 8. A contradiction! Such

counterexample does not exist. We have proven the proposition.
Remark The method of proof in this problem is the proof by
contradiction. To prove that the proposition is true, we assume that
the proposition is false. Consider a variable V E
N. From the smallest counterexample of V we can
deduce a contradiction and the proof becomes
easier using this condition. The conclusion of this
problem is the best. When 1 E 1 = 1 V 1 + 3, we can
give a counterexample as Fig. 5 shows us. Fig. 5
10 We denote 21 persons by 21 vertices. There is an edge joining
two vertices if and only if the two delegates which are represented by
the two vertices have called each other. By assumption, there exists an
odd cycle whose length is m. (We call cycle whose length is odd
cycle. )
Let C be the smallest odd cycle in the graph, the length of which
is 2k + 1.
If k1 , let C be a triangle. It means the three persons have made
=

a phone call to each other.

If k > 1, set C Vl V2 • .• and there is no edge joining Vi and
V2k+l Vl

Vj. (1 ~ i, j ~2k +1, i - j *±1 (mod2k +1). ) Otherwise, suppose

that Vi' Vj are adjacent and the sum of the length of cycle Vl V2 • ••

V2k+l Vl and that of the cycle ViVi+l • . . VjVi is 2k + 3. So among

V iV j •••

them there must be an odd cycle whose length is less than 2k + 1. It

contradicts the fact that C is the shortest.
Suppose there is no triangle among the 21 - (2k + 1) = 20 - 2k
vertices other than Vl' V2' ••• , V2k+l. By Turan's theorem, there are
at least (10 - k)2 edges joining them. Any vertex among them is not
adjacent to two vertices which are adjacent to C, so it is adjacent to at
most k vertices. So the sum of the edges is:
 Solutions 127

2k + 1 + H20 - 2k) + 00 - k)2

= 100 + 2k + 1 - k 2

= 102 - (k - 1) 2

~ 102 - (2 _ 1)2 = 101.

A contradiction!
So the graph must contain a triangle, which means there exist
three persons who have called each other.
11 Suppose there exists such a graph that the degree of every
vertex is more than 2. But the length of any cycle of the graph is
divisible by 3. We consider the graph G which has this property and
the least number of vertices. Clearly , the graph contains the shortest
circle Z. The non-adjacent vertices on this cycle are not joined by an
edge. Since the degree of every vertex is more than 2, every vertex on
the cycle Z is adjacent to one vertex not on the cycle . Let Z pass the
vertex A 1 , A 2 ' ••. , A 3k .
Suppose that there exists a path 5 which joins the vertices A m and
A " and which does not include edges in Z. We consider the cycle Z 1
and Z z consisting of the two halves of 5 and Z. Since the length of
each of the two cycle is divisible by 3, it is not difficult to see the
length of path 5 is divisible by 3. Especially, for the given graph , we
can know that any vertex X which is not on the circle Z cannot have
edges incident to two distinct vertices of Z. It means that the edges
which are induced by the vertices on Z but not on the cycle should be
incident to distinct vertices, respectively.
Let us construct a graph G 1 • Collapse all vertices A 1 , A z , ... ,
A 3k on the cycle Z of G into one vertex A and keep all the vertices
which are not on the cycle and their incident edges. Join the A and the
vertices on the Z one by one. It is easy to know the degree of A is no
less than 3k. The number of vertices in G 1 is less than that of G and
the degree of every vertex is still more than 2. According to above
conclusion , the length of any cycle in G is divisible by 3. We arrive at
a contradiction . In view of the above proof, we can know G is the
128 Graph Theory

graph satisfying these properties and with the least vertices.

Then in a graph the degrees of its vertices are all more than 2,
there must exist a cycle whose length can be divided by 3. Then we
only need to apply this assertion to our problem. We denote the city by
a vertex and the path by the edge.

Exercise 5

1 When n en ~ 2) is odd, K" is a cycle. When n = 2, K 2 is a

chain. When m, n are both even, K m ." is a cycle.
2 Suppose G contains at least 2k odd vertices. Delete one edge to
get G'. There are three cases: e1) The number of odd vertices of G'
decreases by 2, then G' need at least k - 1 strokes to draw. (2) The
number of odd vertices increases by 2, G' need at least k + 1 strokes to
draw. (3) The number of odd vertices does not change, G' need at
least k strokes to draw.
3 These two graphs are all unicursal, that is, they can be drawn
in one strock, and start and end at the same vertex.
4 When n is odd, the graph is unicursal; when n IS even, the
graph is not unicursal.
5 Draw G as follows. We denote the persons by vertices. If two
person have exchanged views, then join the corresponding vertices.
Take the longest chain 1-'-' Let Vl be an end of 1-'-' then () vertices
V2' ••• , VS+l adjacent to Vl are all on the chain 1-'-' Otherwise, I-'- can

still be extended. Go along the I-'- through vertices V2' V3' •. • , VS+l'
and then return to v 1 . This is a cycle whose length is more than ().
6 Take a point v;
(j = 1, 2, 3, 4) on every v;

face. If the two faces have common edges, join the

two points. Then we get a new graph G * which we
call the dual graph of G. In the graph G, going from Vj

a face to another face through the edge e i is

equivalent to going from one vertex to another
vertex along an edge in the G *. Therefore, if G Fig. 6
 Solutions 129

contains a broken line f1 satisfying conditions (1) and (2), then G ' is
a chain (Q1 and Q2 are not on a face) or a cycle (Q1 and Q 2 are on a
face), i. e. the graph G ' is unicursal or it can be drawn in one strock.
But if the four vertices of G * are all odd, the graph G' needs two
strocks to draw.
7 Suppose that there are k lines and that one vertex V i

corresponds to one number ai in the following way. If Vi is red, then

ai = 1; Vi is blue, thenai =-1, i = 1,2, ... , n. Then

hence k is odd.
8 Use the conclusion of Exercise 7 A ___ 1!" __ <;:0" __l}1 __~! __ _
,, ,
,
and refer to Example 5 in Chapter 5. __ L
'- -- :B
I I I 3

9 The given graph contains 16 odd ,

I
"
I:C
r- - -
vertices B i ' C i (i = 1, 2, ... , 8). If we ,, -
'
'
- - -I -
,,
- - I }

C ' ,i=-=-=4::"::"::4i=-'=-=4.::.=+, -:...j. : B4 =--'-

want to make it a cycle, we should add at 7:: :: :::

least 8 edges so that the graph becomes a
B7 ~: ___ ~ ____ I ~ ___ :C
: __ ..
4

,
I I I
,
I

cycle . Fig. 7 shows the cycle after adding __ __

~_~~~:-:-~~-_-~~~~~
~ _~_,r._~_~_ '
8 edges BiC i (i = 1,2, ... ,8) so that the
Fig. 7
walk is the shortest.

Exercise 6

1 When n ;?: 3, K n is a Hamiltonian graph. When m = n ;?: 2, the

complete bigraph K m." is a Hamiltonian graph.
2 The reader can find it on the graph.
3 A regular icosahedron consists of 20 congruent equilateral
triangles. At the center of every triangle we mark a vertex. Only if
two triangles have a common edge, we join the corresponding vertices
and construct a regular dodecahedron which consists of 12 regular
pentagons. From the study of Hamiltonian cycles, we can know that
on the regular dodecahedron we can find a Hamiltonian cycle. Use
scissors to cut the dodecahedron along the Hamiltonian cycle. It means
130 Graph Theory

that cut the regular dodecahedron into two halves, and also cut each
regular triangle into two halves. The trace does not go through the
vertices of the regular dodecahedron.

Fig. 8

4 Draw a graph G as follows. We denote each 1 x 1 x 1 cube by a

vertex. We join the corresponding vertices if and only if two cubes
contain a common face. It is easy to know that G is a bigraph. Let
G = (X, Y; E). If the corresponding vertex corresponding to the
small cube at one corner belongs to X, the vertex which represents the
small cube at the center belongs to Y. Since I X 1= 14, I Y I = 13, G
contains no Hamiltonian chain.
S (1) We denote six persons by six vertices V I ' V 2 ' . . . , V 6 . If
v, can cooperate with Vj' v, is adjacent to Vj. By condition, d (v i) ~
3, i = 1, 2, ... , 6. According to Theorem 4, G contains a
Hamiltonian cycle C = v i l V i 2 • •• V i 3 V iI . In the cycle, two adjacent
vertices represent two persons who can cooperate with each other.
(2) Put ViI' Vi 2 into one group, Vi 3 ' Vi. into one group, Vis ' Vi6
into one group. Alternatively put Vi6 ' ViI into one group, Vi 2 ' Vi3 into
one group, Vi. , V is into one group . These are two different ways of
grouping .
6 We denote ministers by 2n vertices. If two persons are not
2n
enemies, join the corresponding vertices to get a graph G. In the
graph G, for every vertex v, d( v ) ~ (2n - 1) - (n - 1) = n.
According to the Theorem 4, G contains a Hamiltonian cycle and we
can arrange vertices according to the order in the cycle.
7Draw a graph G as follows. We denote 9 children by 9 vertices
and if two children know each other we join the corresponding
 Solutions 131

vertices . In graph G, for any two vertices v and VI, d (v) + d (VI) ~ 8.
According to Theorem 2, G contains a Hamiltonian chain and we can
arrange the children into one line by the order in the chain.
8 Draw a graph G as follows. We denote the materials by
vertices and every dish by an edge. In graph G, the degree of every
vertex is no less than 4. According to Theorem 4, G contains a
Hamiltonian cycle.
9 Suppose that set A contains n elements and we give every
element a number. Let A = {1, 2, 3, ... , n}. We use a sequence
whose length is n and consisting of 0 and 1 to denote a subset. The rule
is that if the element i of A is in this subset, the i-th position of this
sequence is 1, otherwise the i-th position is O. For example, the empty
set 0 = 0, 0 , 0, . . . , 0; {1} = 1, 0, 0, . . . , 0; {n} = 0, 0, ... , 1;
{2, 3} = 0, 1, 1, 0, ... , O. Then there are 2" subsets in A. We denote
the sequences corresponding to the 2" subsets by vertices and join two
vertices if and only if the sequences have only one different number at
the same position. Then we get a graph G. For example, when n = 1,
G is a line segment as Fig. 9 shows us: When n = 2, G is a square as
Fig. 10 shows us . Fig. 10 can be drawn using two Fig. 9. That is, add 0
before one pair of 0, 1 to get 00, 01 and add 1 before another pair of
0, 1 to get 10, 11. Then put one Fig. 9 on top of another one and join
them by two vertical edges to get a square. Copy two Fig. 10, put one
on top of another, add 0 before the number of every vertex of the top
square and add 1 before the number of every vertex of the bottom
square. Join the corresponding vertices of the two squares by four
vertical edges to get a graph G. When n = 3, G is a cube. If n = k ,
assume that the graph G has been drawn. Put G on top of another
copy. Then add 0 before the number of every vertex of the upper
graph and add 1 before the number of every vertex of below graph.
Join the corresponding vertices of the two graphs by vertical edges to
get a new graph G of n = k + 1. Fig. 11 is the case when n = 3 and Fig .
12 is the case of n = 4. We call the graph when n = k a cube graph with
dimensions n. It is easy to prove by induction: the cube graph with
132 Graph Theory

dimensions n contains a Hamiltonian cycle en ;?: 2). For n = 1, clearly

it is true. Since the cube graph with dimension 1 is K 2 which is a
Hamiltonian chain. For n = 2, it is a square which is a Hamiltonian
cycle. For n = k, if it is a Hamiltonian graph, consider n = k + 1.
Delete one corresponding edge from the upper and lower Hamiltonian
cycle of n = k in G, respectively and then combine the edges joining
the ends of deleted edges and the upper and lower Hamiltonian cycle
to get a Hamiltonian cycle when n = k + 1 as the bold lines of Fig. 12
show us . Put the vertices of G on a cycle according to the order of
Hamiltonian differs only cycle. Start from any vertex and order all
subsets clockwise (or counter clockwise) so that every adjacent subset
differs only in one element.

o0 > - - - - - - -.....01
Fig. 9
°D
10
Fig. 10
11

o1Of-~----"=-r

"
110 1111
Fig. 11 Fig. 12

10 First, the degree of every vertex is at least 3. Otherwise,

there exists one vertex A which is incident to at most two edges.
 Solutions 133

Remove one edge, then the remaining vertex A cannot lie on cycle
which contradicts the condition. So n ~ 3. It is easy to prove that n *
4, 5, 6.
If n =7, remove the vertex whose degree is the largest (clearly,
the degree is at least 3) to get a cycle whose length is 6. Since the
vertices adjacent to this vertex may be non-adjacent on the cycle.
(Otherwise, there will be a cycle whose length is 7.) The removed
vertex is adjacent to at most three non-adjacent vertices on the cycle.
So the degree of this vertex is at most 3. 3 X 7 = 21 is odd. In fact, the
sum of all the degrees is even. A contradiction.
If n = 8, after removing the vertex whose degree is the largest, we
get a cycle whose length is 7. The removed vertex is adjacent to at
most three non-adjacent vertices on the cycle. So the degree of this
vertex is at most 3. The degree of every vertex is 3. As Fig. 13 ( 1 )
shows us, the degrees of A, C, F, 0 are 3. They cannot be incident
to any edge. Every vertex of B, D, E, G is incident to one edge,
respectively. If B is adjacent to G, D is adjacent to E (there are two
edges). It is impossible. If B is adjacent to D, E is adjacent to G and
the graph contains a cycle whose length is 8. A contradiction. If B is
adjacent to E, D is adjacent to G and the graph contains a cycle whose
length is also 8. A contradiction.
If n = 9, since 3 X 9 = 27 is not even, it is impossible that the
degree of every vertex is 3. There exists one vertex whose degree is at
least 4. We remove the vertex whose degree is the largest to get a cycle
whose length is 8. So the removed vertex is adjacent to at most four
non-adjacent vertices on the cycle. The largest degree is 4 and the
smallest is 3. As Fig. 13 (2) shows us, B is at least incident to one
edge. Clearly we cannot join more edges between B and A, C. If B is
adjacent to D, the graph contains a cycle whose length is 9. A
contradiction. Similarly, B cannot be adjacent to H. If B is adjacent
to F, the graph contains a cycle whose length is also 9. A
contradiction. So B can only be adjacent to E or G. By symmetry, let
B be adjacent to E. Similar to the above argument, we can know H is
134 Graph Theory

adjacent to C. (If H is adjacent to E, the degree of E is 5. A

contradiction. ) F is adjacent to A, and D is adjacent to G. We cannot
join any two vertices by more edges and after removing A, the graph
must contain a cycle whose length is 8. In fact, if the graph contains a
cycle whose length is 8, BE, BC, HG, HC, FE, FG must lie on the
cycle. The six edges form a cycle whose length is 6. A contradiction.
By the above discussion, the n satisfying the condition is at least 10.
The example when n = 10 is Fig. 13(3), we call it the Peterson graph.

A A

A, A,
G{---,,+---4C
F C

E D E
(I) (2) (3)

Fig. 13

11 If there are 5 persons, let the original order of seats be

ABCDEA. Change it to ADBECA. If there are more than five
persons, join the two vertices only when they are not seated together.
Get a graph G. Since the degree of every vertex is 1V(G) 1- 3. The
sum of the degree of any two vertices is 2n - 6 where n is the number
of vertices. Also n > 5, 2n - 6 ~ n. By Theorem 3, G contains a
Hamiltonian cycle and we can arrange the seats by the order of the
cycle.

Exercise 7

1 Suppose that G is connected, otherwise consider one connected

component. If the degree of every vertex is no less than 6, 611 ~ 2e. It
means that 11 ~ ;. Since f ~ 2; , then

2 = 11 - e +f ~ 3"e - e
2e
+ 3" = o.
 Solutions 135

A contradiction.
2 Suppose the degree of every vertex is more than 4, then 2e

:8dC'O;)
[=1
~ 5'O, i.e. '0 < ;e. Sincee <3'0 -6, thene < ~e -6. It

means that e ~ 30. A contradiction.

3 By Euler's Formula, f = 2 + e -v = 8. Since there are ~ = 3

edges on every face in average. Since there are at least three edges on
every face, there are 3 edges on every face.
4 Suppose G and G are both planar graphs. The number of
vertices in G and G is v and the number of edges is e and e',

respectively. Thene +e' = 1 'OC'o -1). Adding the inequalities e <

3'0 -6, ande' <3'0 - 6, we get ~ 'OC'o - 1) = e + e' <6'0 - 12, '0 2 -

13'0 + 24 < 0, v < 11. It contradicts the hypothesis.

5 Consider the dual graph. CSee the sixth problem of Exercise
5. ) Since K 5 is not a planar graph, then f < 4.
6 There are G) edges in a convex polyhedron with n vertices

and every face contains at least 3 edges. So the number of the faces of
(n)
the polyhedron is no more than 32 2 . By Euler s Formula, ,

Simplify it to get n 2 - 7n + 12 < 0, where n can take only 3 or 4. We

complete the proof.
7 See Question 1 of Exercise 7.
8 We denote the vertices of a polyhedron by the vertices of a
graph and the edges by the edges of the graph. Then we get a
connected planar graph. Then v ~ 4, f ~ 4. By Euler's Formula e =

v +f - 2 ~ 6. It means that there is no polyhedron whose edge

number is less than 6. If there is a graph with e = 7, then 3f < 2 X 7,
136 Graph Theory

f = 4. But a polyhedron with four faces can only contain 6 edges. So

there is no polyhedron with 7 edges. Consider k ;;;: 4. The pyramid
whose base is a polygon with k edges is a polyhedron with 2k edges.
Cut a corner from the pyramid whose base is a polygon with k - 1 edges
to get a polyhedron with 2k + 1 edges. In conclusion, n ;;;: 6, n =F 7,
there is a polyhedron with n edges.
9 Suppose a convex polyhedron contains x vertices and the lOn
faces contain C 1 , C z , ••• , C lOn vertices and a1' az, ••• , a10n edges,
respectively. The number of the edges of the convex polyhedron is
1 10n
"2 ~a;. By Euler's Theorem,
1=1

1 10n
10n +x = "2 ~a; +2.
i= l

1 10n
Since x ~ :3 ~ a;, then
i=1

1 10n 1 10n

"2 ~ a; + 2 - 1On ~:3 ~ a;.

or
1011

~a; ~ 60n -12.

i=1

If among lOn faces there are no n faces whose edge numbers are
equal,
lOn

~a; ;;;: C3 + 4+···+12)(n-1) + 13 XlO

i=l

= 75n +55
> 60n -12,

a contradiction. So there are at least n faces whose edge numbers are

equal.
10 In the graph there are only two-side polygons and hexagons
and the number of each kind is 3. If the graph contains a Hamiltonian
 Solutions 137

cycle, by Theorem 4, 4(j~ - f~) = O. That is f~ = f~, but f~ + f~ = 3.

It is impossible.
11 By Theorem 4, 2 (f~ - f'~) + 3 (j~ - f'~) = 0, so f~ - f'~ is a
multiple of 3. It means among the 5 quadrilaterals there are 4
quadrilaterals outside the cycle, and another inside, or 4 quadrilaterals
inside. If this Hamiltonian cycle goes through both e and e', the two
quadrilaterals on the two sides of e are inside the Hamiltonian cycle
and outside the cycle, respectively. So are the two quadrilaterals on
the two sides of e'. There are at least two quadrilaterals inside the
cycle and two others outside. A contradiction.
12 Draw a graph G = (V, E), where V
{Xt, X2' ••• , XII}' In the graph G, two vertices
Xi' Xj are adjacent if and only if dCx" Xj) = 1.
Suppose that G contains two distinct edges AB,
°
CD which intersect at vertex as Fig. 14 shows us.
Fig. 14
Since d (A, B) = 1, d CC, D) = 1, without loss of

generality, we suppose that dCO, A) < ~ , dCO, C) < ~ and the

angle between AB and CD is e, d CA, C) = {d 2 CO, A) + d 2 (0, C) -
2d(0, A) xdCO, C)cose}-±. By the above condition only whene =

1[, dCO, A) = ~, d(O, C) = ~, then dCA, C) = 1. But now A

coincides with D and B coincides with C, it means that AB and DC

are the same edge. It contradicts the hypothesis that they are two
distinct edges. Other than this case, we have d CA, C) < 1. It
contradicts the hypothesis that the distance of any two vertices is no
less than 1. In summary, G is a planar graph and the edge number e of
G is no more than 3n - 6. CThe reader can compare it with the
conclusion of Example 7 in Chapter 2 . )

Exercise 8

1 By Example 1, there must exist a monochromatic triangle .

138 Graph Theory

Without loss of generality, let L AIA2A 3 be red. For the three edges
of L A4A SA6' there are two cases: (1) The three edges of L A4ASA6
are all red. We have completed the proof. (2) There is one blue edge
in LA4A SA6' say, A4A S. For AIA4' A 2A 4 , A 3A4' if there are two
red edges among them, there must exist a red triangle. If there are
two blue edges, say, AIA4 and A 2A 4 . Now if there is one blue edge
among AlAs and A 2A s ' there must exist a blue triangle. If AlAs and
A2AS are all red, LAIA2AS is a red triangle. We complete the proof.
2 If there are no three line segments such
~6_____~ 5
that they are monochromatic and contain no /,' :\
'1\
/:
"

I \ I I

common vertices. As Fig. 15 shows us, without loss ,;"..----;-...:.-'+'. ;. '----:71 A 4

of generality, let AIA2 be red. By hypothesis,
three line segments A 3A4' AsA s , A 6A 7 cannot be A21..::::..---T~-'-';-~A 3
all blue. Without loss of generality, let A 3A4 be
A, A,
red. Since AIA2 and A3A4 are red, the lines
Fig. IS
joining every two of the four points As, A 6 , A7 ,

A s are all blue. Similarly, the lines joining every two of the four
points A I ' A2 , A 3 , A4 are all red. Without loss of generality, let
AIA6 be blue, then A 3AS must be red. Whatever color A4As is, it
contradicts the hypothesis.
3 We color the shortest edge of every triangle red and others
blue. Since r2 = 6, there must be a monochromatic triangle which is
red and whose longest edge is the shortest edge of another triangle.
4 Take one vertex A of the two-color complete graph K 9 . If A
is incident to four blue edges AA I , AA 2 , AA 3 , AA 4 , the complete
subgraph K4 whose vertices are AI, A 2 , A 3 , A4 in K9 contains no blue
edge. If A is incident to six red edges AA I , AA 2 , ••• , AA 6 , the
complete subgraph whose vertices are A I , A2, • •• , A6 in K 9 contains
a monochromatic triangle LAiAjA k C1 ~ i, j, k ~ 6). Since K9
contains no blue triangle, LAiAjAk is a red triangle and the complete
subgraph K4 whose vertices are A, A i ' Aj , Ak is red. If every vertex
of K 9 is incident to 5 red edges, then the number of red edges in K 9 is
 Solutions 139

5 X9
2 It is impossible .

5 By Theorem 3,

rC3, 6) ~ rC3, 5) +r(2, 6 ) - 1 = 14 + 6 - 1 = 19.

6 By Theorem 3, r ( 4, 4) ~ r ( 4, 3) + r C3, 4) = 9 +9 = 18.

7 Consider the n -color complete graph K r ,, ' The method of
coloring is that color ( x , y) in the i -th color if and only if I x - y I is
in the i-th group. By Theorem 2, K r" must contain a monochromatic
triangle. Suppose that the three edges of the triangle is all colored in
thej-th color, then among 1,2, . .. , r", there are three natural
numbers a > b > c such that x = a - c, y = a - b, z = x - yare all
in the j -th group.
8 By Theorem 2, r6 ~ [6! eJ +1 = 1958 < 1978. By Schur's
Theorem, the proposition is true.
9 In Al , A 2, . . . , A 7 , among the triangles constructed from the
first six vertices there must be two monochromatic triangles which
have no common edge . If there is a common edge , let the two triangle
be DA1A2A 3 and D A1A 2A4' Now remove Al and add A 7 , then there
exist two monochromatic triangles. Between the two triangles, there
must be a tria ngle different from A 2A3 A4' There is no common edge
between this triangle and D A1A2A 3 or D A1A 2A4'
10 We denote the six lines by six vertices. If two lines lie on two
different faces, color the edge joining the corresponding vertices red.
If the lines lie on one face, color the edge joining the corresponding
vertices blue. We obtain a two-color complete graph K 6 ' So there
must exist a monochromatic triangle. If it is a red triangle, any two of
the lines which the vertices correspond to lie on different faces. If it is
a blue triangle, the three lines which the vertices correspond to lie on
one face . Since any three lines cannot lie on one face, the three faces
which the three lines lie in intersect each other in the three lines. Then
we know the three lines either are parallel to each other or intersect at
one point.
140 Graph Theory

11 Take a set consisting of four irrational numbers {j2, - j2 ,

J3, - J3}. Take any three numbers from the set and there must be:
either j2 +(- j2) = 0 is a rational number when j2, - j2 are chosen;
orJ3 + (- J3) = 0 is a rational number when J3, - J3 are chosen.
When n = 4, the conclusion is not true. So if the n satisfies the
requirement of the problem, n ~ 5. Next, we prove that among any
five given numbers, we may find three such that the sum of any two
irrational numbers is still an irrational number. Suppose that {x, y,
z, u, v} is a set of any given five irrational numbers. We denote the
five irrational numbers by five vertices. If the sum of two numbers is
an irrational number, join the corresponding vertices by a red edge. If
the sum of two numbers is a rational number, join the corresponding
vertices by a blue edge. Then we get a two-color complete graph K 5 .
First we prove that the two-color complete graph K 5 contains no blue
triangle . Otherwise, suppose that there is a blue triangle DXYZ which
means that x + y, Y + z, Z +x are all rational numbers. Then
1
x = 2[ (x + Y) + (z + x) - (Y + z)]
is also a rational number, which contradicts the fact that x is an
irrational number. Next, prove that the two-color complete graph K 5
contains no blue pentagon. Otherwise, suppose that there is a blue
pentagon xyzuv which means that x + y, y + z, Z + u, u + v, v + x
are rational numbers. Then

1
x = 2[(x + y) + (z + u) + (v +x) - (y + z) - (u + v) ]

is still a rational number, which contradicts the fact that x is an

irrational number. The two-color complete graph K 5 contains neither
a blue triangle nor a blue pentagon . By the vice versa after Fig. 8. 3,
there must exist a red triangle. Suppose that DXYz is a red triangle,
then x + y, y + z, Z + x are all irrational numbers.
12 As Fig. 16(1) shows us, color the K 7 in two colors and denote
 Solutions 141

a solid line by a red edge and a dotted line by a blue edge. There are 4
red triangles: ,6,AJA4A6' ,6,A2A4A6' ,6,A3A4A6' ,6,A7A4A6 and 4
blue triangles ,6,A JA2A 3' ,6,A2A 3A 7' ,6,AJA3A 7' ,6,AJA2A7. It is
easy to see that any two monochromatic triangles with the same color
contain a common edge. So n ~ 8.
Next, we prove when n = 8, the proposition is true. We prove by
contradiction.
First we prove a lemma: If the proposition is not true, there must
be a red triangle and a blue triangle which contain one common
vertex. First, K s which is colored in 2 colors must contain a
monochromatic triangle. Without loss of generality, let it be a blue
triangle ,6,A JA2A 3. Now among A3A4A sA6A7A S' there must exist a
monochromatic triangle which is not blue. If this red triangle contains
A 3 , the lemma is true. Otherwise, suppose that ,6,A4A sA 6 is a red
triangle. There are 9 edges joining ,6,AJA2A3 and ,6,A4A sA6 among
which there are at least five monochromatic edges. Without loss of
generality, let it be red. Then A J , Az , A3 are incident to at least 5 red
edges. Among them there must be one vertex which is incident to at
least two red edges. This triangle consisting of this vertex and
,6,AJA 2A 3 contains one common vertex. So the lemma must be true.
Next, we prove the proposition: Suppose the proposition is not true.
By the lemma, let ,6,AJA 2A3 be a blue triangle and ,6,A 3A4A s be a red
triangle. Consider the edges joining AJA4A6A7AS. Among them there
is no monochromatic triangle . So K 5 consists of a blue cycle of 5
vertices and a red cycle of 5 vertices (by the remark after Fig. 8.3). In
Fig. 16(2), represent red edges by solid lines and blue edges by dotted
lines. Without loss of generality, let AJA4A 6A7As be a blue cycle of 5
vertices and AJA7A4A SA 6 be a red cycle of 5 vertices. Next, we
discuss the color of A3A7 . If A 3A7 is blue, then A 3AS' A 3A6 must be
red. (Otherwise, ,6,A 3A7 As or ,6,A 3A 6A 7 is blue, which contradicts
,6,AJAzA3.) Now ,6,A3A6As is red, which contradicts ,6,A 3A4A S! If
A 3A 7 is red, we discuss the color of A 3A S.
142 Graph Theory

A,

(I) (2)
Fig. 16

If A 3A S is blue, A2A4 must be red, AlA s must be blue and A 2A7

must be red. So, L:"A 2A4A7 is a blue triangle, which contradicts
L:"A3A4AS ' If A3A S is red, AsA7 must be blue and AsAs must be blue.
So L:"AsA 7A8 is blue, which contradicts L:"A 1 A 2A3! In summary,
when n = 8, the proposition is true. So the smallest natural number
is 8.
13 We denote 20 teams by 20 vertices. Join the vertices of the
teams which have played matches in the first round by red edges and
the vertices of the teams which have played matches in the second
round by blue edges. Then every vertex is incident to a red edge and a
blue edges . The graph must consist of several even cycles. In every
even cycle, we can choose half of the vertices among which any two
vertices are not adjacent. Then we choose 10 teams which have not
played with each other. We complete the proof .

Exercise 9

1 (1) When n = 5, Fig. 17 is what we need . When n = 6, Fig. 18

is what we need.

Fig. 17 Fig. 18
 Solutions 143

(2) Suppose when n = k, there exists a directed graph satisfying

the requirement. When n = k + 2, first using the vertices VI' V 2 , •• • ,
V k , we draw a directed graph with k vertices satisfying the
requirement. For another two vertices V k + l , V k + 2 , suppose that VI ,
V 2 , . . . , V k all point to V k+ 1 and that V k+ 2 points to VI , V 2 , . .. , V k .
Suppose that V k+ 1 points to V k+ 2 . Then V k+ 1 gets to VI , V 2 , ... through
V k + 2• (Clearly, V k+ 2 can get to VI' V 2 , ... , V k .) VI' V 2 , ... , V k can
get to V k+ 2 through V k + 1 • (Clearly , VI , V 2, • • • , V k can get to V k+ 1 .)

So this graph with k + 2 vertices still satisfy the requirement.

By (1) and (2), we know for any 4 < n EN, there exists a scheme
of changing the path among n cities satisfying the requirement.
2 Suppose G contains a circuit (VI' V 2' .. . , V k ). In V 2'
V3 ' •• • , V k- I , take the first vertex V i so that the arc ( V i+ 1 , V I ) exists .

Then there exists an arc ( V I ' Vi )' so (V I ' Vi' V i+ l ) is a triangular

circuit.
3 We will prove that if an air route satisfies the condition f and
there is no flight between two cities A and B, then we can use airline
A--B or B --A so that the air route still satisfies the condition. If not,
the new route does not satisfy the condition f. Then after opening the
routeA--B, there exists a closed path B--CI-- .. ·--C,,--A--B.
Similarly, after opening the route B -- A, there exists a closed path
A--DI-- .. ·--Dm--B--A . But before opening route between A and
B, there exists a route A -- DI - - .. ·--D m --B--C I - - .. ·--C " --A.
(Maybe there are some vertices C i and D j which are overlapped . It
means that the former air route does not satisfy the condition f,
because it is possible to fly from A and return to A. A contradiction. )
4 Refer to Example 4.
5 We denote n players by n vertices . If Vi defeats V j ' we can
draw an arc from Vi to Vj to get a directed graph D. If there is no
circuit in D, there must exist a vertex v whose indegree is O. The
vertex represents the person who wins all the games. Similarly, we can
prove there is a person who loses all the game.
6 Suppose among v I ' V 2' •• • , v n , the vertex v p has the most
144 Graph Theory

number of offsprings, then v p is the common ancestor of the n

persons . Otherwise, we assume that V p is not the ancestor of V q ' and
then the common ancestor v r of v p and v q is not v p' and the offsprings
of v r are more than those of v p by one . A contradiction.
7 B has won two games .
8 One round rohin corresponds to one tournament. By assumption,
there is no vertex whose outdegree is n - 1. By the Pigeonhole
Principle, there exists at least two vertices whose outdegrees are the
same. By Theorem 4, we have completed the proof.
9 Use Theorem 3.
Index

Adjacent, 3 Finite graph, 3

Are, 101 Forest, 41

Bigraph , 24 Graph , 1
Brouwer, 57 Generating tree, 45

Chain, 40 Hamiltonian cycle , 63

Circui t , 103 Hamiltonian graph, 63
Coloring, 84 Hamiltonian path, 63, 106
Complete graph, 3 Hamiltonian, 63
Completet k-partite graph , 24 Hand - Shaking lemma , 14
Connected graph, 40 Homeomophic, 78
Cycle , 40 Hypergrah , 94

Degree , 13 Incident, 3
Dirac , 70 Indegree, 102
Directed graph, 101 Infinite graph, 3
Direct search, 63 Isomorphic graph, 2
Endpoint, 101
Euler's Formula , 75 Konigsberg, 51
Euler tour , 53
Eulerian, 53 Leaf vertex, 41
Even , 13 Length , 40 , 103
Extremal graph, 24 Loop, 3

Face , 75 Maximum degree, 13

146 Graph Theory

Minimum degree, 13 Regular graph, 13

Odd, 13 Schur Theorem, 95

Ore, 68 Simple graph, 3
Outdegree, 102 Starting point, 101
Subgraph, 2
Parallel edges, 3
Path, 40 Tournament, 102
Pendant vertex , 41 Tree, 41
Peterson graph, 132 Triangle, 25
Planar graph, 75 Turan's Theorem, 28

Ramsey number, 90 Vertex, 1

This page intentionally left blank