(

GRAPHS &
DIGR~PHS
THIRD EDITION

G.CHARTRAND I
Western Mishigan Unlversi~
,

~nd
I

L. LESNIAK
DrewjUniversi~

8r1!
Facultat de
Matematiques i Estadistica
BIBLIOTECA

CHAPMAN & HALUCRC

Boca Raton London New York Washington, D.C.
 Dedicated with appreciation to our friends
and distinguished graph theory colleagues

Mike Henning
University of Natal, Pietermaritzburg

and

Library of Congress Cataloging-in-Pnblication Data Ed Palmer

Michigan State University
Catalog record is available from the Library of Congress,

This book contains information obtained from authentic and highiy regarded sources, Reprinted
material is quoted with pennission, and sources are indicated. A wide variety of references are listed.
Reasonable efforts have been made to publish reliable data and infonnation, but the author and the
publisher cannot assume responsibility for the validity ()f all materials or forthe consequences of their use.
Apart from any fair dealing for the purposes of research or private study, or criticism or review, as
pennitted under the UK Copyright Designs and PatentsAct. 1988, this publication may not be reproduced,
stored or'transmitted, in any fomi or by any means, electronic or mechanical, including photocopying,
microfilming, and recording, or by any information Stomge or retrieval system, without the prior permis-
sion in writing of the publishers, or in the case of reprographic reproduction only in accordance with
the tenns of the licenses issued by the Copyright Licensing Agency in the UK, or in accordance with
the tenns of the license issued by the appropriate Reproduction Rights Organization outside the UK.
The consent of CRC Press LLC does not extend to copying for geneml distribution, for promotion,
for creating new works, or for resale. Specific pennission must be obtained in writing from CRC Press
LLC for such copying.
. Direct all inquiries to CRC Press LLC, 2000 N.W. Corpomte Blvd., Boca Raton, Florida 33431.

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

First edition 1979

t, Second edition 1986
Third edition 1996

I
First CRC Press reprint 2000
Originally published by Chapman & Hall

© 1979, 1986 by Wadsworth, Inc., Belmont, California 94002

I © 1996 by Chapman & HalVCRC

No claim to original U.S. Government works

•f International Standard Book Number 0-412-98721-X

Library of Congress Card Number 96-83483
Printed in the United States of America 1 2 3 4 5 6 7 8 9 0
Printed on acid-free paper
 Contents

Preface to the third edition ix

f':' Introduction to graphs 1

1.1 Graphs 1
1.2 Degree sequences 11
1.3 Distance in graphs 16
1.4 Digraphs and multigraphs 25
2 Structure and symmetry of graphs 33
2.1 Cut-vertices, bridges and blocks 33
2.2 The automorphism group of a graph 40
2.3 Cayley color graphs 45
2.4 The Reconstruction Problem 50

:? Trees and connectivity 57

3.1 Elementary properties of trees 57
3.2 Arboricity and verlex-arboricity 68
3.3 Conne~vity arid edge-connectivity 74
3.4 Menger's theorem 79
3.5 Vulnerability of graphs 86
4 Eulerian and hamiltonian graphs and digraphs 92
4.1 Eulerian graphs and digraphs 92
4.2 Hamiltonian graphs and digraphs 103
4.3 Line graphs and powers of graphs 122
5 Directed graphs 131
5.1 Connectedness of digraphs 131
5.2 T04rnaments 138
5.3 Hamiltonian tournaments 149
6 Planar graphs 154
6,1 Euler's Formula 154
6.2 Characterizations of planar graphs 161
6.3 Hamiltonian planar graphs 169
6,4 Crossing number and thickness 174
viii Contents
7 Graph embeddings 183
7.1 The genus of a graph 183
7.2 2-cell embeddings of graphs 191
7.3 The maximum genus of a graph 207
7.4 Voltage graphs 214 Preface to the third edition
8 Graph colorings 220
8.1 Vertex colorings 220
8.2 Edge colorings 236
8.3 Map colorings and flows 246
Graph theory is a major area of combinatorics, and during recent decades,
9 Matchings, factors and decompositions 259 graph theory has developed into a major area of mathematics. In addition
9.1 Matchings and independence in graphs 259 to its growing interest and importance as a mathematical subject, it has
9.2 Factorizations and decompositions 272 applications to many fields, including computer science and chemistry.
9.3 Labelings of graphs 288 As in the first edition of Graphs & Digraphs (M. Behzad, G. Chartrand,
L. Lesniak) and the second edition, our major, indeed our sole, objective
10 Domination in graphs 302 is to introduce and to treat graph theory in the way we have always found
10.1 The domination number of a graph 302 it, namely, as the beautiful area of mathematics it is. We have strived
10.2 The independent domination number of a graph 312 to write a reader-friendly, carefully written book that emphasizes the
10.3 Other domination parameters 317 mathematical theory of graphs and digraphs.
New to the third edition are expanded treatments of hamiltonian graph
11 Extremal graph theory 329 theory, graph decompositions, and extremal graph theory, a study of
11.1 Turan's theorem 329 graph vulnerability and domination~in graphs; and introductions to
11.2 Extremal results on graphs 334 . voltage graphs, graph labelings, and the probabilistic method in graph
11.3 Cages 343 theory. Numerous original exercises have been added. A comprehensive
bibliography has been included together with an extensive list of graph
12 Ramsey theory 351 theory books so that avid graph theory readers have many avenues to
12.1 Classical Ramsey numbers " 351 pursue their interests.
12.2 Generalized Ramsey theory 357 This text is intended for an introductory sequence in graph theory at
12.3 Other Ramsey numbers 363 the advanced undergraduate or beginning graduate level. A one-semester
course can easily be designed by selecting those topics of major impor-
13 The probabilistic method in graph theory 370 tance and interest to the instructor and students. Indeed, mathematical
13.1 The probabilistic method 370 maturity is the only prerequisite for an understanding and an apprecia-
13.2 Random graphs 376 tion of the material presented.
It is with great pleasure and sincere appreciation that we thank a
Glossary of symbols 382 number of mathematicians who gave of their time to read portions of
Graph theory books 1936-1996 385 earlier versions of this edition, to offer suggestions for additions and
References 392 improvements, or both. Consequently, we thank
Index 415
Ghidewon Abay Asmeron Frank Boesch
Brian Alspach Bob Brigham
Dan Archdeacon Paul Catlin
Jay Bagga Guantao Chen
Marge Bayer David Craft
Lowell Beineke Roger Entringer
x Preface to the third edition
Ralph Faudree Christina Mynhardt CHAPTER 1
Herbert Fleischner Ladislav NebeskY
Joe Gallian Ortrud Oellermann
Heather Gavlas
John Gimbel
Ed Palmer
Rich Ringeisen Introduction to graphs
Wayne Goddard Ed Schmeichel
Ron Gould Kelly Schultz
Frank Harary Michelle Schultz
Johan Hattingh Allen Schwenk
Teresa Haynes Paul Seymour
Steve Hedetniemi Bill Staton We begin our study of graphs by introducing many ofthe basic concepts
Mike Henning Henda Swart that we shall encounter throughout our investigations. The related topic
Hector Hevia Bjarne Toft of digraphs is introduced as well.
Mike Jacobson Henk Jan Veldman
Hong-Jian Lai Lutz Volkmann
Terri J"indquester Doug West 1.1 GRAPHS
Terry McKee Art White
John Mitchem A graph C is a finite nonempty set of objects called vertices (the singular is
vertex) together with a (possibly empty) set of unordered pairs of distinct
A special acknowledgement is due to Julie Yates for her impeccable vertices of C called edges. The vertex set of C is denoted by V(C), while the
typing and to Marci Israel for her significant assistance in the preparation edge set is denoted by E(C).
of the manuscript. We cannot thank both of you enough. The edge e = {u, v} is said to Join the vertices u and v. If e = {u, v} is
Finally, we thank all those editors with whom we have worked, namely an edgeofa graph C, then uand v are adjacent vertices, while u and e
John Kimmel, Achi Dosanjh, Emma Broomby and Stephanie Harding, are incident, as are v and e. Furthermore, if e} and e2 are distinct edges
and the staff of Chapman & Hall for their interest in and assistance of G incident with a common vertex, then el and e2 are adjacent edges. It
with the third edition. is convenient to henceforth denote an edge by uv or vu rather than by
{u,v}.
C.c. &L L The cardinality of the vertex set of a graph C is called the order of C
and is commonly denoted by n(C), or more simply by n when the
graph under consideration is clear; while the cardinality of its edge set
is the size of C and is often denoted by m(C) or m. An (n,m) graph has
order n and size m.
It is customary to define or describe a graph by means of a diagram in
. which each vertex is represented by a point (which we draw as a small
circle) and each edge e = uv is represented by a line segment or curve
joining the points corresponding to u and v.
A graph C with vertex set V(C) = {v}, V2,"" vn } and edge set
E(C) = {e},e2"" ,em} can also be described by means of matrices. One
such matrix is the n x n adjacency matrix A( C) = [aij]' where

aij = {Io If~£VjVj

VjVd
E E(C)
E(C).

Thus, the adjacency matrix of a graph G is a symmetric (0,1) matrix

having zero entries along the main. diagonaL·· Another matrix is the
2 Introduction to graphs Graphs 3
twice. We state this as our first theorem, which, not so coincidentally, is

G, t-/f o1
101
1 0
1
11000]
10 1 10
sometimes called The First Theorem of Graph Theory.

1L-J" ·=1 Theorem 1.1 .

110 1 o1 10 1
o1 1 0 o0 0 1 1 Let G be an (n, m) graph where V(G) = {VI, V2,'" 1 Vn }. Then
11
V3 eS v4
L deg vi=2m.
Figure 1.1 A graph and its adjacency and incidence matrices. i=1

This result has an interesting consequence.

n x m incidence matrix B(G) = [bij]/where
b.. = {I if Vj and ej are incident Corollary 1.2
'I 0 otherwise.
In any graph, there is an even number of odd vertices.
For example, a graph G is defined by the sets
V(G) = {VllV2,V3,V4} and E(G) = {el,e2,e3,e4,eS}, Proof
where el = VlV2/ e2 = VIV3, e3 = V2V3, e4 = V2 V4 and es = V3V4' A diagram Let G be a graph of size m. Also, let W be the set of odd vertices of G and
of this graph and its adjacency and incidence matrices is shown in let U be the set of even vertices of G. By Theorem 1.1,
Figure 1.1.
With the exception of the order and the size, the parameter that one L degv = L degv + L degv = 2m.
encounters most frequently in the study of graphs is the degree of a VEV(C) VEW VEU

vertex. The degree of a vertex v in a graph G is the number of edges of G Certainly, EVEU degv is even; hence EVEW degv is event implying that
incident with v/ which is denoted by degcv or simply by deg v if G is clear IWI is even and thereby proving the corollary. 0
from the context. A vertex is called even or odd according to whether its
degree is even or odd. A vertex of degree 0 in G is called an isolated vertex Two graphs often have the same structure, differing only in the way
and a vertex of degree 1 is an end-vertex of G. The minimum degree of G is their vertices and edges are labeled or in the' way they are drawn. To
the minimum degree among the vertices of G and is denoted by 8(G). The make this idea more precise, we introduce the concept of isomorphism.
maximum degree is defined similarly and is denoted by A (G). In Figure 1.2/ A graph Gl ,is isomorphic to a graph Gz if there exists a one.;to-one mapping
a graph G is shown together with the degrees of its vertices. In this case, ¢, called an isomorphism, from V(Gl) onto V(Cz) such that ¢ preserves
8(G) = 1 and A(G) = 5. adjacency; that is, uv EE(G1 ) if and only if ¢u¢V EE(Gz). It is easy to
For the graph G of Figure 1.2/ n = 9 and m = 11/ while the sum of the see that 'is isomorphic to' is an equivalence' relation on graphs; hence,
degrees of its nine vertices is 22. That this last number equals 2m this relation divides the collection of all graphs into eqUivalence classes,
illustrates a basic relationship involVing the size of a graph and the two graphs being nonisomorphic if they belong to different equivalence
degrees of its vertices. Every edge is incident with two vertices; hence, classes.
when the degrees of the vertices are summed, each edge is counted If G1 is isomorphic to G2, then we say Gl and G2 are isomorphic or equal
and write Gl = G2 • If G1 = G2, then, by definition, there exists art iso-
2 3 4 1 morphism ¢: V(G 1) -+ V(G z). Since ¢ is a one-to-one mapping, G1 and

1\
Gz have the same order. Since adjacent vertices in G1 are mapped into
G: adjacent vertices in G2 and nonadjacent vertices of G1 are mapped into
nonadjacent vertices in G2, the graphs G1 and G2 have the same size.
Since each vertex v in G1 and its image vertex ¢v in G2 must have the
same degree in their respective graphs, the degrees of the vertices of G1
1 3 5 2 are exactly the degrees of the vertices of G2 (counting multiplicities).
Figure 1.2 The degree of the vertices of a graph. Although these conditions are necessary for G1 and G2 to be isomorphic,
4 Introduction to graphs Graphs 5
o o o
o
o o o 0--0

o o 0--0 0--0

o o o o 0---0

Figure 1.4 All graphs of order 4 or less.

H: G:

Figure 1.3 Isomorphic and nonisomorphic graphs.

they are not sufficient. For example, consider the graphs Gi , i = 1,2,3, of Figure 1.5 Subgraphs.
Figure 1.3. Each is a (6,9) graph and the degree of every vertex of each
graph is 3. Here, G1 = G2 . For example, the mapping 0/: V(Gd - 7 V(G 2 ) set V(G) and edge set E(G) - {e}. The deletionofa set of vertices orset of
defined by edges is defined analogously. These concepts are illustrated in Figure 1.6.
If u and v are nonadjacent vertices of a graph G, then G + f, where
o/VI = VI, o/V2 = V3, o/V3 = VS, o/V4 = V2, O/VS = V4, o/V6 = v6
f = uv,denotes the graph with vertex set V(G) and edge set E(G) U {f},
is an isomorphism, although there are many other isomorphisms. On the Clearly, G ~ G +f.
other hand, G3 contains three pairwise adjacent vertices whereas GI does We have seen that C - e has the same vertex set as G and that G has the
not; so there is no isomorphism from GI to G3 and therefore GI oJ~. Of same vertex set as G + f. Whenever a subgraph H of a graph G has the
course, G2 =I G3 • same order as G, then H is called a spanning subgraph of G.
If G is an (n, m) graph, then n ~ 1 and 0 ::;; m ::;; (~) = n(n -1)/2. There Among the most important subgraphs we shall encounter are the
is only one (1,0) graph (up to isomorpJ:.li.sm), and this is referred to as 'induced subgraphs'. If U is a nonempty subset of the vertex set V(G)
the trivial graph. A nontrivial graph then has n ~ 2. The distinct (non- of a graph G, then the subgraph (U) of G induced by U is the graph
isomorphic) graphs of order 4 or less are shown in Figure 1.4. having vertex set U and whose edge set consists of those edges of G
Frequently, a graph under study is contained within some larger graph incident with two elements of U. A subgraph H of G is called vertex-
also being investigated. We consider several instances of this now. A induced or simply induced if H = (U) for Some subset U of V(G). Similarly,
graph H is a subgraph of a graph G if V(H) ~ V(G) and E(H) ~ E(G); in
such a case, we also say that G is a supergraph of H. If H is isomorphic to a
subgraph of G, we also say that H isa subgraph of G. In Figure 1.5, H is a
subgraph of G but H is not a subgraph of F.1f H is a subgraph of G, then
we write H ~ G.
The simplest type of subgraph of a graph G is that obtained by deleting
a vertex or edge. If v E V(G) and IV(G)!:;;:: 2, then G - v denotes the
subgraph with vertex set V( G) - {v} and whose edges are all those of G
not incident with v; if e E E(G), then G - e is the subgraph having vertex Figure 1.6 The deletion of a vertex or edge of a graph.
6 Introduction to graphs Graphs 7

V4

Figure 1.7 Induced and edge-induced subgraphs. Figure 1.9 The Petersen graph.

if X is a nonempty subset of E(G), then the subgraph (X) induced by X is Theorem 1.3
the graph whose vertex set consists of those vertices of G incident with at
least one edge of X and whose edge set is X. A subgraph H of Gis edge- For every graph G and every integer r ~ ~(G), there exists an r-regular graph H
induced if H = (X) for some subset X of E(G). It is a simple consequence of containing G as an induced subgraph.
the definitions that every induced subgraph of a graph G can be obtained
by removing vertices from G while every subgraph of G can be obtained Proof
by deleting vertices and edges. These concepts are illustrated in Figure 1.7
for the graph G, where HG is r-regular, then we may take H = G. Otherwise, let G' be another
copy of G and join corresponding vertices whose degrees are less than r,
V(G) = {Vl,V2,V3,V4,VS,V6}, U={VllV2,VS} and X={VIV4,V2VS}.
calling the resulting graph G1 . If G1 is r-regular, then let H = G1 . If not, we
There are certain classes of graphs that occur so often that they deserve continue this procedure until arriving at an r-regular graph Gk where
special mention and in some cases, special notation. We describe the most k = r - 8(G). (The proof of Theorem· 1.3 is illustrated for the graph G of
prominent of these now. Figure 1.10 where r = ~(G) = 3.) 0
A graph G is regular of degree r if degv = r for each vertex v of G. Such
graphs are called r-regular. A graph is complete if every two of its vertices
are adjacent. A complete (n, m) graph is therefore a regular graph of G:
degree n -1 having m = n(n -1)/2; we .denote this graph by K n • In
Figure 1.8 are shown all (nonisomorphid regular graphs with n = 4,
including the complete graph G3 = K4.
A 3-regular graph is also called a cubic graph. The graphs of Figure 1.3
are cubic as is the complete graph K4 • However, the best known cubic
graph is probably the Petersen graph, shown in Figure 1.9. We will have
many occasions to encounter this graph. .
In 1936 Denes Konig [KlO] wrote the first book on graph theory. In it he
proved that if G is a graph with ~(G) = d, there exists a d-regular graph H
containing G as an induced subgraph. Konig's result actually first appeared
in 1916 (see [K8]). His technique proves a somewhat stronger result.

o G 1: 0----0

o o 0---0
Figure 1.8 The regular graphs of order 4. Figure 1.10 A 3-regular graph H containing G as an induced subgraph.
8 Introduction to graphs Graphs 9
Of course, there is no pretense that the graph H constructed in the proof o
of Theorem 1.3 is one of the smallest order with the desired property.
Indeed, for the graph G of Figure 1.10, the graph H has order 16, while
the minimum order of a3-regular graph containing G as an induced
subgraph is actually 6. In fact, in 1963 Erdos and Kelly [EKl] produced
a method for determining the minimum order of an r-regular graph H
o I I I II\
Figure 1.12 The union of graphs.
containing a given graph G as an induced subgraph. The graph KI,s is called a star. A graph is a complete multipartite graph if it is
The complement G of a graph G is that graph with vertex set V(G) such a complete k-partite graph for some k ~ 2.
that two vertices are adjacent in G if and only if t~ese vertices are not There are many ways of combining graphs to produce new graphs. We
adjacent in G. Hence, if G is an (n, m) graph, then G is an (n, m) graph, next describe some binary operations defined on graphs. This discussion
where m + m= G). In Figure 1.8, the graphs Go and G3 are complemen- introduces notation that will prove useful in giving examples. In the
tary, as are GI and Gz. The complement Kn of the complete graph K n has n follOWing definitions, we assume that G) and Gz are two graphs with
vertices and no edges and is referred to as the empty graph of order n. A disjoint vertex sets.
graph G is self-complementary if G = G. Certainly, if G is a self-complemen- The union G= G I U Gz has V(G) = V(G I ) U V(G z ) and E(G) =
tary graph of order n, then its size is m = n(n -1)/4. Since only one of n E(G I ) U E(G z ). If a graph G consists of k (2':2) disjoint copies of a graph
and n - 1 is even, either 4 I nor 4 I n - 1; that is, if G is a self-complemen- H, then we write G = ill. The graph 2KI U 3Kz U KI,3 is shown in
tary graph of order n, then either n := 0 (mod 4) or n := 1 (mod 4). Figure 1.12.
A graph G is k-partite, k ~ I, if it is possible to partition V(G) into k The join G = G1 + Gz has V(G) = V(G 1 ) U V(G z ) and
subsets Vb V z ,.'" Vk (called partite sets) such that every element of
E(G) joins a vertex of Vi to a vertex of Vi' i:l' j. If G is a I-partite graph E(G) = E(G1) U E(G z) U {uvlu E V(G 1 ) and v E V(G z )}.
of order n, then G = Kn. For k = 2, such graphs are called bipartite graphs; Using the join operation, we see that Kr,s = Kr + K.. Another illustration is
this class of graphs is particularly. important and will be encountered given in Figure 1.13.
many times. In Figure 1.l1(a), a bipartite graph G is given. Then Gis The cartesian product G = G1 X Gz has V(G) = V(G 1 ) x V(G z), and two
redrawn in Figure 1.11(b) to emphasize its bipartite character. If G is vertices (U1' uz) and (Vb Vz) of G are adjacent if and only if either
an r-regular bipartite graph, r;" I, with partite sets VI and V 2I then
IVII = IVz\· This follows since its size m = rlVII = rlVz\· UI = VI and UZVz E E(Gz )
A complete k-partite graph G is a k-partite graph with partite sets or
VI' Vz ,.··, Vk having the added property that if u E Vi and v E Vi' i:l' j, Uz = Vz and U1VI E E(G I ).
then uv E E(G). lilVd = nj, then this graph is denoted by K(nI' nz, ... ,nk)
A convenient way of drawing G1 x Gz is first to place a copy of Gz at each
or Kn\>nz,...,nk' (The order in which the numbers nI, nz, ... ,nk are written is
vertex of G1 (Figure 1.14(b» and then to join corresponding vertices of Gz
not important) Note that a complete k-partite graph is complete if and
in those copies of Gz placed at adjacent vertices of G1 (Figure 1.14(c».
only if nj = 1 for all i, in which case it is ~. If nj = t for all i, then the
Equivalently, G1 x Gz can be constructed by placing a copy of G1 at
complete k-partite graph is regular and is also denoted by Kk(t). Thus,
each vertex of Gz and adding the appropriate edges. As expected,
Kk(1) = Kk. A complete bipartite graph with partite sets V 1 and V z, where
G1 x Gz = Gz X G1 for all graphs G1 and Gz.
IVII = rand IVz\ = s, is then denoted by K(r,s) or more commonly Kr,s' An important class of graphs is defined in terms of cartesian prodlJocts.
The n-cube Qn is the graph Kz if n = I, while for n ;" 2, Qn is defined

G:

(a) (b)
I
Figure 1.11 A bipartite graph. Figure 1.13 The join of two graphs.
10 Introduction to graphs Degree sequences 11
G\: 0--0---0 1.4 For each integer k ;;" 2, give an example of k nonisomotphic regular
graphs, all of the same order and same size.
1.5 A nontrivial graph G is called irregular if no two vertices of G have
the same degrees. Prove that no graph is irregular.
1.6 Let G be a graph of order n containing vertices of degree r, where r is
(a) (b) (c) a positive integer, and exactly one vertex of each of the degrees
r -1, r - 2, ... , r - j, where 1 < j < r. By Konig's proof of Theorem
Figure 1.14 The cartesian product of two graphs. 1.3, there is an r-regular graph of order 21n containing G as an
induced subgraph. Show, in fact, that there exists an r-regular
(0, 1,0) (1,1,0) graph of order 2n containing G as an induced subgraph.
o 1 1.7 If H is an induced subgraph of G, does it follow that R is an induced
Q\: 0----------10 (0,1,1) subgraph of G?
(I, 1, 1)
1.8 Prove that there exists a self-complementary graph of order n for
(0,1) (1,1)
every positive integer n with n == 0 (mod 4) or n == 1 (mod 4).
1.9 Determine all self-complementary graphs of order 5 or less.
(0,0,1) 1.10 Let G be a self-eomplementary graph of order n, where
n == 1 (mod 4). Prove that G contains at least one vertex of degree
(n -1)/2. (Hint: Prove the stronger result that G contains an odd
I!umber of vertices of degree (n - 1)/2.)
(0,0) (1,0) (0,0,0) 0, O. 0)
1.11 Let G' be a nonempty graph with the property that whenever
Figure 1.15 Cubes. uv rj, E(G) and vw rj,E(G), then uw rj, E(G). Prove that G has this
property if and only if G is a complete k-partite graph for some k ;;" 2.
recursively as Qn-l x K2. The n-cube Qn can also be considered as that
graph whose vertices are labeled, by the binary n-tuples (aba2"" ,an)
(that is, aiis 0 or 1 for 1 ~ i ~ n) and such that two vertices are adjacent
if and only if their corresponding n-tuples differ at precisely one coordi- 1.2 DEGREE SEQUENCES
nate. It is easily observed that Qn is an n-regular graph of order 2n. The n-
cubes, n = 1,2 and 3, are shown in Figure 1.15 with appropriate labelings. In this section, we investigate the concept of degree in. more detail. A
The graphs Qn are often called hypercubes. sequence dt , d2 , ••• ,dn of nonnegative integers is called a degree sequence
of a graph G if the vertices of G can be labeled Vt, V2, ..• ,Vn so that
degvi = d j for all i. For example, a degree sequence of the graph of
Figure 1.16 is 4,3, 2,2, 1 (or 1, 2, 2, 3, 4, or 2, 1,4,2,3, etc.).
EXERCISES 1.1

1.1 . Determine all nonisomorphic graphs of order 5.

1.2 Let n be a given positive integer, and let rand s be nonnegative
integers sitch that r + s = nand s is even. Show that there exists a
graph G of order n having r even vertices and s odd vertices.
1.3 Figure 1.3 shows two regular nonisomorphic (6,9) graphs. Give 4,3,2,2,1
another example of two nonisomorphic regular graphs of the same
order and same size. Figure 1.16 A degree sequence of a graph.
12 Introduction to graphs Degree sequences 13
Given a graph G, a degree sequence of G can be easily determined, of Theorem 1.4 actually provides an algorithm for determining whether a
course. On the other hand, if a sequence s: d1 , dz, ... ,dn of nonnegative given finite sequence of nonnegative integers is graphical. If, upon
integers is given, then under what conditions is s a degree sequence of repeated application of Theorem 1.4, we arrive at a sequence every
saine graph? If such a graph exists, then s is called a graphical sequence. term of which is 0, then the original sequence is graphical. On the other
Certainly the conditions di ~ n - 1 for all i and 2:::7=1 di is even are hand, if we arrive at a sequence containing a negative integer, then the
necessary for a sequence to be graphical and should be checked first, given sequence is not graphical.
but these conditions are not sufficient. The sequence 3,3,3,1 is not We now illustrate Theorem 1.4 with the sequence
graphical, for example. A necessary and sufficient condition for a s:5,3,3,3,3,2,2,2,1,1,1.
sequence to be graphical was found by Havel [H81 and later rediscovered
by Hakimi [H3]. After one application of Theorem 1.4, we get
8;: 2,2,2,2,1,2,2,1,1,1.
Theorem 1.4 Reordering this sequence, we obtain
A sequence s: d1 , dz, ... ,dn of nonnegative integers with d1 ?: dz ?: .... ?: dnf 81: 2,2,2,2,2,2,1,1,1,1.
n ?: 2, d1 ?: 1, is graphical if and only if the sequence S1: dz - 1, d3 - 1, ... , ContinUing, we have
dd 1+1 -1, dd ,+Z,"" dn is graphical.
s~: 1,1,2,2,2,1,1,1,1

Proof 8Z: 2,2,2,1,1,1,1,1,1

Assume that S1 is a graphical sequence. Then there exists a graph Gl of 5~ = 53: 1,1,1,1,1,1,1,1
order n - 1 such that sl is a degree sequence of Gt . Thus, the vertices of 5~: 0,1,1,1,1,1,1
Gl can be labeled as Vz, V3,"" V n so that
54: 1,1,1,1,1,1,0
d; - 1 2 ~ i ~ dl +. 1
degvi = { . s~:O,l,l,l,l,O
di dl + 2 ~ l ~ n.
A new graph G can now be constructed by adding a new vertex VI and the s5: 1,1,1,1,0,0
dl edges vlv;,2 ~ i ~ dl + 1. Then in G, degvi = d; for 1 ~ i ~ n, and so 8~: 0,1,1,0,0
s: dl , dz,... , dn is graphical.
Conversely, let s be a graphical sequence. Hence there exist graphs 86: 1,1,0,0,0
of order n with degree sequence s. Among all such graphs let G be s; = S7: 0,0, 0,0.
one such that V(G) = {VI' Vz,. '," V n }, degv; = di for i = 1, 2, ... , n, and
the sum of the degrees of the vertices adjacent with VI is maximum. Therefore, 8 is graphical. Of course, if we observe that some sequence
We show first that VI is adjacent with vertices having degrees prior to S7 is graphical, then we can conclude by Theorem 1.4 that 8 is
dz,d3 , .. • ,dd1 +1' graphical. For example, the sequence 83 is easily seen to be graphical
Suppose, to the contrary, that VI is not adjacent with vertices having since it is the degree sequence of the graph G3 of Figure 1.17. By Theorem
degrees dz, d:3, ... , dd , + l' Then there exist vertices V r and V s with dr > ds 1.4, each of the sequences 5Z,5It and 5 is in turn graphical. To construct
such that VI is adjacent to V s but not to V r • Since the degree of. V r exceeds a graph with degree sequence 5z, we proceed in reverse from 8~ to 82'
that of vs , there exists a vertex Vt such that v t is adjacent to V r but not to observing that a vertex should be added to G3 so that it is adjacent to
vs ' Removing the edges VIVs and VrVt and adding the edges VIVr and VsVt two vertices of degree 1. We thus obtain a graph Gz with degree sequence
results in a graph G' having the same degree sequence as G. However, in 8z (or 8~). Proceeding from 8~ to 8It we again add a new vertex joining it to
G' the sum of the degrees of the vertices adjacent with VI is larger than that two vertices of degree 1 in G2• This gives a graph Gl with degree sequence
in G, contradicting the choice of G. 81 (or 5;). Finally, we obtain a graph G with degree sequence 8 by con-
Thus, VI is adjacent with vertices haVing degrees dz, d3 , ••• ,dd, +It and sidering 5}; that is, a new vertex is added to Glf joining it to vertices of
the graph G - VI has degree sequence SIt so SI is graphical. 0 degrees 2, 2,2,2, 1.
14 Introduction to graphs Degree sequences 15
most (k) = k(k -1)/2. Every vertex Vi of Vz (so k + 1 ~ i ~ n) is clearly
joined ~o at most dj vertices of VI' Also, every vertex V; of V z is joined
to at most k vertices of VI' Hence each vertex Vi of V z is adjacent to at
most rnin{k, di } vertiCes of VI' The desired result now follows. 0

A proof of the sufficiency of Theorem 1.5 was given in a paper by

Sierksma and Hoogeveen [SHll, where several other characterizations
of graphical sequences are also presented.
G: When considering degree sequences, we are interested not only in
degrees but also in their frequencies. We now delete this last requirement.
Denote the degree set of a graph G (that is, the set of degrees of the vertices
of G) by V(G). For example, if G = K1,z,4, then V(G) = {3, 5, 6}. We now
Figure 1.17 Construction of a graph G with given degree sequence. investigate the question of which sets of positive integers are the degree
sets of graphs. This question is complet~ly answered by a result of
It should be pointed out that the graph G in Figure 1.17 is not the only Kapoor, Polimeni and Wall {KPWl].
graph with degree sequence s. Indeed, there are graphs that cannot be
produced by the method used to construct graph G of Figure 1.17. For
example, the graph H of Figure 1.18 is such a graph. Theorem 1.6
Another result that determines which sequences are graphical is due to For every set 5 = {al,aZ"" ,ak},k ~ 1, of positive integers, with
Erdos and Gallai [EG3]. We give a proof of the necessity only since the al < az < ... < ak, there exists a graph G such that V(G) = 5. Furthermore,
proof of the sufficiency is lengthy and not useful for our purposes. the minimum order p,(5) = p,(al, az, ... ,ak) of such a graph G is p,(5) = ak + 1.

I
Theorem 1.5
Proof
A sequence dl , dz ,... ,dn (n ~ 2) of nonnegative integers with dl ~
If G is a graph such that V(G) = 5, then G has order at least ak + 1. Thus
dz ~ ... ~ dn is graphical if and only if L:f=1 di is even and for each integer
k, 1 ~ k ~ n - 1, . we must show that such a graph G having order ak + 1 exists. We proceed
by induction on k. For k= 1, we observe that every vertex of the complete
k n
graph Kat +1 has d~gree alJ SO p,(al) = at + 1. For k = 2, the vertices of the
I: d ~ k(k -1) + I:
i rnin{k,d;}. graph F = Ka l + (Ka2 - al +1) have degrees al and az, and since F has order
i=1 i=k+l
az + I, we conclude that p,(al,aZ) = az + 1.
Let k ~ 2. Assume, for every set 5 containing i positive integers, where
Proof (of the necessity) 1 ~ i ~ k, that p,(5) = aj + 1, where ai is the largest element of S. Let
Let G be a graph of order n with V(G) = {vt>vz, ... ,vn} such that 5 t = {b 1,bz , .. ·,bk+1} be a set of k+l positive integers such that
degv; = d; for 1 ~ i ~ n. By Theorem 1.1, L:f=1 di is even. For b1 < bz < ... < bk+1' By the inductive hypothesis,
l~k~n-l, let VI = {VbVZ,,,,,Vk} and Vz =V(G)-:'V1 . The sum p,(bz - b1,b3 - b1, ... , bk - b1) = (bk - b1) + 1.
E~=1 di counts every edge in (VI) twice and every edge joining a
vertex of V t and a vertex of Vz once. The number of edges in (VI) is at Hence, there exists a graph H of order (bk - b1 ) + 1 such that
V(H) = {b2 - b1,b3 - b11 ... ,bk --' bl}'
0 0 0
The graph
H:
0 0 0 G = Kbl + (Kbk+t-bk UH)
Figure 1.18 A graph that cannot be constructed by the method following Theorem has order bk+1 + 1, and V(G) = {b 1 ,b2 , ... , bk +1 }; hence Jt(b1 , b2 , ••• ,bk+1) =
1.4. . bk+ 1 + 1, which completes the proof. 0
 Distance in graphs 17
16 Introduction to graphs

EXERCISES 1.2

1.12 Determine whether the following sequences are graphical. If so, G:

construct a graph with the appropriate degree sequence.
(a) 4,4,3,2, 1
(b) 3, 3,2,2,2, 2, 1, 1
(c) 7, 7, 6, 5, 4, 4, 3, 2
(d) 7, 6, 6, 5, 4, 3, 2, 1
(e) 7, 4, 3, 3,2, 2, 2, 1, 1, 1
Figure 1.19 ,Walks, trails and paths.
1.13 Show that the sequence dl , d2l , dn is graphical if and only if the
sequence n - dl - 1, n - d2 1, , n - dn - 1 is graphical. only if k = t and Ui = Vi for 0 :::::; i :::::; k; otherwise, they are differen,t. Observe
1.14 (a) Using Theorem 1.4, show that s: 7, 6, 5, 4, 4, 3, 2, 1 is graphicaL that the edges of two different u-v walks of G may very well mduce the
(b) Prove that there exists exactly one graph with degree sequence s. same subgraph of G.
A u-v walk is closed or open depending on whether u = V or u =1= v. A u-v
1.15 Show that for every finite set S of positive integers, there exists a trail is a u-v walk in which no edge is repeated, while a u-v path is a u-v
positive integer k such that the sequence obtained by listing each walk in which no vertex is repeated. A vertex u forms the trivial u-u
element of S a total k times is graphicaL Find the minimum such k path. Every path is therefore a trail. In the graph G of Figure 1.19,
for S = {2, 6, 7}. WI: VI' V2, v3, V2,VS, v3, V4 is a Vl-v4 walk that is not a trail,W2 : VI, V2;VS,
1.16 Two finite sequences SI and S2 of nonnegative integers are called VI, V3, V4 is a VI-V4 trail that is not a path and W3 : VI' V3, V4 is aVl-v4 pat~. '

bigraphical if there exists a bipartite graph G with partite sets VI By definition, every path is a walk. Although the c~nverse of thiS
and V2 such that SI and S2 are the degrees in G of the vertices in statement is not true in general, we do have the followmg theorem. A
VI and V 2, respectively. Prove that the sequences 51: aI,a2,'" ,ar walk W is said to contain a walk W' if Wi is a subsequence of W.
and 52: bl ,b2,··., bt of nonnegative integers with r ~ 2,
al ~ a2 ~ ... ~ art bl ~ b2 ~ ... ~ bt, 0 < al :::::; t, and 0 < bl : : :; rare Theorem 1.7
bigraphical if and only if the sequences s~:a2,a3, ... ,ar and
s~: hI "":'1, b2 -1, ... , ba, -1, bal +1>"" bt are bigraphical. Every u-v walk in ~ graph contains a u-'-v path.
1.17 Find a graph G of order 8 having V( G) = {3, 4, 5, 7}.
Proof
Let W be a u-v walk in a graph G. If W is dosed, the resultis trivial. Let
1.3 DISTANCE IN GRAPHS W: u = Uo, Ul, u2,"" Uk = V be an open u-v walk of a graph G. (A vertex
may have received more than one label.) If no vertex of G occurs in W
Let U and v be (not necessarily distinct) vertices of a graph G. A u-v walk more than once, then W isa u-v path. Otherwise, there are vertices of G
of G is a finite, alternating sequence that occur in W twice or more. Let i and jbe distinct positive integers, with
i < j say, such that Ui = Uj' If the terms Uil Ui+i,"" Uj-l are deleted fro~
U = uO,ell Ulle2,'" ,'Uk-1>ekl Uk= v W, a u-v walk WI is obtained having fewer terms than that of W. If there IS
of vertices and edges, beginning with vertexu and ending with vertex v, nb repetitibn of vertices in WI' then WI is a u-v path. If this is not the case,
such that ei = Ui-IUi for i = 1,2, ... ,k. The number k (the number of we continue the above procedure until finally arriving at a u-v walk that is
occurrences of edges) is called the length of the walk. A trivial walk a u-v path. 0
cont.ains no edges, ~hat is, k = O. We note that there may be repetition of
vertIces and edges m a walk. Often only the vertices of a walk are indi- As the next theorem indicates, the kth power of the adjacency matrix of
cated since the edges present are then evident. Two u-v walks u = uo, a graph can be used to compute the number of walks of various lengths in
UI,"" Uk = v and U = VA' VI"", Vt = V are considered to be equal if and the graph.
 Distance in graphs 19
18 Introduction to graphs
o
Theorem 1.8
If A is the adjacency matrix of a graph G with V(G) = {VI, Vz,···, vn }, then
the (i, j) entry of Ak, k;?; I, is the number of different Vi-Vj walks of length k
in G.
GN01\I
Figure 1.20 A graph with six components.
o

Proof
Theorem 1.9
The proof is by induction on k. The result is obvious for k = 1 since there A nontrivial graph is bipartite if and only if it contains no odd cycles.
exists a Vj-Vj walk of length 1 if and only if VjVj E E(G). Let A k- I = [a~r-I)]
and assume that a~-I) is the number of different Vi-Vj walks of length
k - 1 in G; furthermore, let A k = fait)]. Since A k = A k- I . A, we have Proof
n Let G be a bipartite graph with partite sets VI and Vz· S~ppose that
a~k) =L a~~-I)atj' (1.1) C: VI' VZ, ... , Vb VI is a cycle of G. Without loss of generalIty, we may
t=I
assume that VI E VI' Bowever, then Vz E V z ,V3 E VI> V4 E V z, and so on.
Every V,Vj walk of length k in G consists of a V,-Vt walk of length k - I, This implies k = 2s for some positive integer s; bence, C h.a~ even length.
where Vr is adjacent to Vj, followed by the edge VtVj and the vertex Vj' Thus For the converse, it suffices to prove that every nontnvial connected
by the inductive hypothesis and (1.1), we have the desired result. D graph G without odd cycles is bipartite, since a nont~ivi~l gr.aph is
bipartite if and only if each of its nontrivial components IS bIpartIte. Let
A nontrivial closed trail of a graph G is referred to as a circuit of G, and a V E V(G) and denote by VI the subset of V(G) consisting of V and all
circuit vI, Vz, ... ,Vn , VI (n;?; 3) whose n vertices Vi are distinct is called vertices u of G with the property that any shortest u-v path of G has
a cycle. An acyclic graph has no cycles. The subgraph of a graph G induced even length. Let Vz = V(G) - VI. We now prove that th~ Rarti~on
by the edges of a trail, path, circuit or cycle is also referred to as a trail, V U V of V(G) has the appropriate properties to show that G IS bIpartite.
path, circuit or cycle of G. A cycle is even if its length is even; otherwise it 1Let:and wbe elements of VI, and suppose that uw E E(G). Necessarily,
is odd. A cycle of length n is an n-cycle; a 3-cycle is also called a triangle. then, neither u nor w is the vertex v. Let V = uo, UI, ••• , U2s = U, S ;?; I, and
A graph of order n that is a path or a cycle is denoted by P n or C n, V = Wo, WI' ... , WZt ;::= W, t ;?;1, be a shortest v-u path and a shortest v-w?f
respectively. G, respectively, and suppose that w' is a vertex that the two paths have ill
We now consider a very basic concept in graph theory, namely COn- common such that the w'''::'u subpathand. w'-w subpath have pnly w' in
nected and disconnected graphs. A vertex u is said to be connected to a common. (Note that w' may be v.) The two v-w' subpaths so determined
vertex V in a graph G if there exists a u-v path in G. A graph G is are then shortest v-w' paths. Thus, there exists an i such that uf = Ui = Wi'
connected if every two of its vertices are connected. A graph that is not However, Ui,Ui+!> ... ,U2s,WZhWZt-l, ... ,Wi=Ui is an odd cycle of G,
connected is disconnected. The relation 'is. connected to' is an equivalence 'which is a contradiction to our hypothesis. Similarly, no two vertices of
relation on the vertex set of every graph G. Each subgraph induced by the V z are adjacent. D
vertices ina resulting equivalence class is called a connected component or
simply a component of G. EqUivalently, a component of a graph G is a For a connected graph G, we define the distance d(u, v) between two
connected subgraph of G not properly contained in any other connected vertices u and v as the minimum of the lengths of the u-v paths of G.
subgraph of Gi that is, a component of G is a subgraph that is maximal Under this distance function, the set V(G) is a metric space, that is, the
with respect to the property of being connected. Bence, a connected sub- following p'roperties hold:
graph F of a graph G is a component of G if for each connected graph H
with F ~ H ~ G where V(F)~ V(H) and E(F) ~ E(H), it follows that 1. d(u, v) ;?; 0 for all pairs u, v of vertices of G, and d(u, v) = 0 if and only
F = H. The number of components of G is denoted by k( G)i of course, if u = V;
k(G) = 1 if and only if G is connected. For the graph G of Figure 1.20, 2. (symmetric property)
k(G) = 6. d(u, v) = d(v, u) for all pairs u, v of vertices of G;
We are now prepared to present a useful characterization of bipartite 3. (triangle inequality)
graphs. d(u, v) + d(v, w);?; d(u, w) for all triples u, v, w of vertices of G.
20 Introduction to graphs Distance in graphs 21
r For the graph G of Figure 1.22, the vertices x and y are peripheral vertices
r s t
and Per(G) = 2K1 •
The radius and diCl.meter are related by the following inequalities.

1J)('\,
v
G:
Theorem 1.10
w
v w x y z For every connected graph G,
radG ~ diamG ~ 2radG.
z
Figure 1.21 The distance levels from the vertex r. Proof
The inequality rad G ~ diam G is a direct consequence of the definitions.
Each vertex of the graph G of Figure 1.21 is labeled with its distance
In order to verify the second inequality, select vertices u and v in G such
from r. The graph G is then redrawn to illustrate these distances better.
that d(u, v) = diam G. Furthermore, let w be a central vertex of G. Since d
The vertices of G are thus partitioned into levels, according to their
is a metric on V(G),
distance from r. There are a number of instances when it is useful to
draw: a graph in this manner. d(u, v) ~ d(u, w) + dew, v) ~ 2e(w) = 2radG. 0
If G is a disconnected graph, then we can define distance as above
between vertices in the same component of G. If u and v are vertices in Theorem 1.10 gives a lower bound (namely, rad G) for the diameter of
distinct components of G, then d(u, v) is undefined (or we could define a connected graph G as well as an upper bound (namely, 2 rad G). This
d(u, v) = (0). is the first of many results we shall encounter for which a question of
The eccentricity e(v) of a vertex v of a connected graph G is the number 'sharpness' is involved. In other words, just how good is this result?
maxUEV(G) d(u, v). That is, e(v) is the. distance between v and a vertex Ordinarily, there are many interpretations of such a question. We
furthest from v. The radius rad G of G is the minimum eccentricity shall consider some possible interpretations in the case of the upper
among the vertices of G, while the diameter diam G of G is the maximum bound.
eccentricity. COili';equently, diam Gis the greatest distance between any Certainly, the upper bound in Theorem 1.10 would not be considered
two vertices of G. Also, a graph G has radius 1 if and only if G contains sharp if diam G < 2 rad G for every graph G; however, it would be
a vertex adjacent to all other vertices of G. A vertex v is a central vertex if considered sharp indeed if diamG =2radG for every graph G. In the
e(v) = rad G and the center Cen(G) is the subgraph of G induced by its latter case, we would have a formula, not just a bound. Actually, there
central vertices. are graphs G for which diam G < 2 rad G and graphs H for which
For the graph G of Figure 1.22, rad G = 3 and diam G == 5. Here, diam H = 2 rad H. This alone may ,be a· satisfactory definition of
Cen(G) = K3 . A vertex v is a peripheral vertex if e(v) = diam G, while the 'sharpness'. A more likely interpretation is the existence of an infinite
periphery PerCG) is the subgraph of G iJ:\duced by its peripheral vertices. class 1i of graphs H such that diam H = 2 rad H for each H E 1i. Such a
class exists; for example, let 1i consist of the graphs of the type Kt + 1(2'
One disadvantage of this example is that for each H E1i, diamH == 2
u

1>.
and rad H = 1. Perhaps a more satisfactory class (which fills a more
u satisfactory requirement for sharpness) is the class of paths P2k+v k ~ 1.
Cen(G): In this case, diamP2k +1 = 2k and radP2k+l = k; that is, for each positive
G: integer k, there exists a graph G such that diam G = 2 rad G = 2k (Exercise
1.30).
v
In the graph G of Figure 1.22, we saw thafCen(G) = K3 • It is not difficult
to see that Cen(P2k +1 ) = K1 and Cen(P2k ) = K2 for all k;;;, 1. Also,
v Cen(CIl ) = CIl for all n ~ 3. Hence there are many graphs that are centers
Per(G): 0 0
x y of graphs. Hedetniemi (see Buckley, Miller and Slater [BMS1]) showed
Figure 1.22 A graph with radius 3 and diameter 5. that there is no restriction on which graphs are centers.
22 Introduction to graphs Distance in graphs 23
There is also no restriction on which graphs can be medians. Slater [S71
showed that every graph is the median of some connected graph. Indeed,
"2 Hendry [HI I] showed that for every two graphs GI and Gz, there exists a
H: connected graph H such that Cen(H) = GI and Med(H) = Gz.
Not every graph is the periphery of some graph, as Bielak and Syslo
[BS2] verified.

Figure 1.23 A graph with given center. Theorem 1.12

A graph G is the periphery of some connected graph if and only if every vertex of
Theorem 1.11 G has eccentricity 1 or no vertex of G has eccentricity 1.
Every graph is the center of some connected graph.
Proof
Proof
Suppose first that every vertex of G has eccentricity 1. Then Per(G) = G
Let G be a given graph. We construct a graph H from G by adding four and G is the periphery of itself. Next, suppose that no vertex oiG has
new vertices UI, VI, U2, V2 and for i = 1, 2, every vertex of G is joined to Vi' eccentricity 1. Define H to be the graph obtained from G by adding a
and Ui is joined to Vi' (This construction is illustrated in Figure 1.23.) Since new vertex v to G and joining v to every vertex of G. Then e(v) = 1. Let
e(ui) = 4 and e(vi) = 3 for i = 1, 2, while eH(x) = 2 for every vertex x of G, x E V(G). Since ec(x) f 1, there is some vertex y in G such that xy rf:. E(G).
it follows that Cen(H) = G. 0 However, dH(x, y) = 2 since x, v,y is a path of length 2 in H. Thus
eH(x) = 2. Indeed, then, every vertex of G has eccentricity 2 in H. There-
The center of a connected graph G was introduced as one means of fore Per(H) = G.
describing the 'middle' of a graph. Since this concept is vague and is For the converse, assume that G is a graph for which some but not
open to interpretation, other attempts have been made to identify the all vertices have eccentricity 1, and suppose, to the contrary, that G
middle of a graph. We describe another of these. The total distance td(u) is the periphery of a connected graph H. Certainly G is a proper
of a vertex U in a connected graph Gis defined by subgraph of H. Therefore, for each vertex x of G, it follows that
eH(x) = diamH ~ 2. Let u be a vertex of G having eccentricity 1 in G.
td(u) = Ed(u, v).
VEV(C)
Thus, u is adjacent to all other vertices of G. Let vbe a vertex of H
such that dH(u, v) = eH(u) = diamH ~ 2. Therefore eH(v) = diamHand
A vertex v in G is called a median vertex if v has the minimum total distance v is a peripheral vertex of H. On the other hand, since dH(u, v) ~ 2,
among the vertices of G. Equivalently, v is a median vertex if v has the the vertex v is not adjacent to u and so v is not in G, which produces a
minimum average distance to all vertices of G. The median Med(G) of G is contradiction. 0
then thesubgraph of G induced by its median vertices. In the graph G of
Figure 1.24, each vertex is labeled with its total distance. The median of G
is also shown.
EXERCISES 1.3
38
29 1.18 Let uand v be arbitrary vertices of a connected graph G. Show that
26 42 S2 there exists a u-v walk containing all vertices of G.
G: Med(G): 0--0
" v 1.19 Prove that 'is connected to' is an equivalence relation on the vertex
" 29
set of a graph.

38 1.20 (a) Let G be a graph of order nsuch that degv ~(n -1)/2 for every
v EV(G). Prove that G is connected. .
Figure 1.24 The median of a graph. (b) Examine the sharpness of the bound in (a).
24 Introduction to graphs Digraphs and multigraphs 25
1.21 Let n ~ 2 be an integer. Determine the minimum positive integer m Let w be a vertex with e(w) = k, and let u be a vertex with
such that every graph of order n and size m is connected. d(w, u) = e(w) = k. For a central vertex v of G, let P be a v-u
path of length d(v, u). Show that e(v) < k ~ e(u). Then show
1.22 Prove that a graph G is connected if and only if for every partition that there is a vertex x (distinct from w) on P such that e(x) = k.)
V(G) = VI U V 2 , there exists an edge of G joining a vertex of VI and a
vertex of V 2 • 1.33 Show that for every pair r, s of positive integers, there exists a
positive integer n such that for every connected graph G of order
1.23 Prove that if G is a graph with 8(G) ~ 2, then G contains a cycle. n, either ~(G) ~ r or diamG ~ s.
1.24 Show that if G is a graph of order n and size n2 /4, theri either G 1.34 Let F and H be two subgraphs in a connected graph G. Define the
contains an odd cycle or G = Kn/ 2,n/2' distance d(F, H) between Fand H as
1.25 (a) Show that there are exactly two 4-regular graphs G of order 7. d(F, H) = min{d(u, v) I U E V(F), v E V(H)}.
(Hint: Consider G.)
Show that for every positive integer k, there exists a connected graph
(b) How many 6-regular graphs of order 9 are there?
G such that d(Cen(G),Med(G)) = k.
1.26 Characterize those graphsG having the property that every induced 1.35 Every complete graph is the periphery of itself. Can a complete
subgraph.of G is a connected subgraph of G. graph be the periphery of a connected graph G with diam G ~ 2?
1.27 .Define a connected graph G to be degree linear if G contains a path P
with the property that for each dE V(G) (the degree set of G), there
exists a vertex of degree d on P.
(a) Let G be a connected graph with V(G) = {dt>dz},d I < dz . Prove 1.4 DIGRAPHS AND MULTIGRAPHS
that G is degree linear by proving that G contains a path of length
1 containing vertices of degrees d1 and d2 • There are occasions when the standard definition of graph does not serve
(b) Determine the maximum value of k such that every connected our purposes. This remark leads us ,to our next topics. When the
graph having a k-element degree set is degree linear. symmetric nature of graphs does not satisfy our requirements, we are
led to directed graphs. A directed graph or digraph D is a finite nonempty
1.28 Let G be a nontrivial connected subgraph that is not bipartite. Show set of objects called vertices together with a (possibly empty) set of ordered
that G contains adjacent vertices u and v such that deg u + deg v is pairs of distinct vertices of D called arcs or directed edges. As with graphs,
even. the vertex set of D is denoted by V(D) and the are set is denoted by E(D).
1.29 Prove that if G is a disconnected graph, then G is connected and, in A digraph D with V(D) = {u, v; w} and E(D) ={(u, w), (w, u),(u, v)} is
fact, diam G ~ 2. illustrated in Figure 1.25. Observe that when a digraph is described by
means of a diagram, the 'direction' of each arc is indicated by an arrow-
1.30 Let a and b be positive integers with a ~ b ~ 2a. Show that there head.
exists a graph G with rad G = a and diam G = b. The terminology used in discussing digraphs is quite similar to that
1.31 Define the central appendage number of a graph G to be the minimum used for graphs. The cardinality of the vertex set of a digraph D is
number of vertices that must be added to G to produce a connected
graph H with Cen(H) = G. Show that the central appendage number w
of a graph (a) can be 0, (b) can never be 1, and (c) is at most 4.
1.32 Let G be a connected graph.
(a) If u and '. v are adjacent vertices of G, then show that D:
le(u) - e(v)1 ~ 1.
(b) If k is an integer such that rad G ~ k ~ diam G, then show that
there is a vertex w such that e(w) = k.
(c) If k is an integer such that rad G < k ~ diam G, then show that u 1I

there are at least two, vertices of G with eccentricity k. (Hint: Figure 1.25 A digraph.
26 Introduction to graphs Digraphs and multigraphs 27
called the order of 0 and is denoted by nCO), or simply n. The size m(O) (or
m) of 0 is the cardinality of its arc set. An (n, m) digraph is a digraph of
order n and size m.
If a = (u, v) is an arc of a digraph 0, thena is said t? joi~ u. ~d v. We
further say that a is incident from u and incident to v, w~l1e UIS incIdent to.a
and v is incident from a. Moreover, u is said to be adJacent. to v .and v is
adjacent from u. In the digraph D of Figure .1.25, vertex u is adJacent to
vertex v, but v is not adjacent to u. Two vertices u and v of a digraph 0
are nonadjacent if u is neither adjacent to nor adj~cent from v in D. .
The outdegree od v of a vertex v of a digraph 0 is the number of vertices
of 0 that are adjacent from v. The indegree id v of v is th~ nu~ber of
vertices of 0 adjacent to v. The degree deg v of a vertex v of 0 is defined by
degv = odv +id v.
In the digraph 0 of Figure 1.25, od u = 2, id u = id v = id w = od w = 1, Figure 1.26 The (3,3) digraphs.
while od v = O. For the same digraph, deg u = 3, deg w = 2 and deg v = 1.
We now present The First Theorem of Digraph Theory. say that 0 1 is a subdigraph of D. We write D1 ~ 0 to indicate that D1 is a
subdigraph of O. A subdigraph 0 1 of 0 is a spanning subdigraph if 0 1
Theorem 1.13 has the same order as O. Vertex-deleted, arc-deleted, induced and arc-
induced sUbdigraphs are defined in the expected manner. These last
lt~ is a digraph of order nand sizem with V(O) = {Vl'VZ"" ,vn }, then two concepts are illustrated for the digraph 0 of Figure 1.27, where
n n
L od Vi = L id Vi = m. V(O) = {Vr, V2,V3,V4}, U={Vl,VZ,V3}, and
i=1 i=1 F = {(vr, vz), (vz, V4)}'
We now consider certain types of digraphs that occur regularly in our
Proof discussions. A digraph 0 is called symmetric if, whenever (u, v) is an arc of
When the outdegrees of the vertices are summed, each arc is counted 0, then (v, u) is also. There is a natural one-to-one correspondence
once, since every arc is incident from exactly one vertex. Similarly, whe.n between the set of symmetric digraphs and the set of graphs. A digraph
the indegrees are summed, an arc is counted just once since every arc IS o is called an asymmetric digraph or an oriented graph if whenever (u, v) is
incident toa single vertex. D an arc of 0, then (v, u) is not an arc of O. Thus, an oriented graph 0 can be
obtained from a graph G by assigning a direction to (or by 'orienting')
A digraph 0 1 is isomorphic to a digraph Oz if there exists a one-to-one
mapping ifJ, called an isomorphism, from V(D 1 ) onto V(Oz) such that
(u, v) E E(OI) if and only if (ifJu, ifJv) E E(Dz). The relation 'is isomorphic
to' is an equivalence relation on digraphs. Thus, this relation partitions
the set of all digraphs into equivalence classes; two digraphs are
nonisomorphic if they belong to different equivalence classes. If D1 is iso-
morphic to Oz, then we say D1 and Dz are isomorphic and write D1 = Dz.
There is only one 0,0) digraph (up to isomorphism); this is the trivial
digraph. Also, there is only one (2,0), (2,1) and (2,2) digraph (up !O
isomorphism). There are four (3,3) digraphs, and they are shown m
Figure 1.26.
A digraph D1 is a subdigraph of a digraph D if V(D 1 ) ~ V(D) and D (u) (F)
E(D1 ) ~ E(D). If D 1 is isomorphic to a subdigraph of D, then we also Figure 1.27 Induced and arc-induced subdigraphs.
28 Introduction to graphs Digraphs and multigraphs 29

Figure 1.30 Regular digraphs.

Figure 1.28 Symmetric and asymmetric digraphs.
important difference being that the dir~ctions of the arcs must be followed
each edge of C, thereby transforming each edge of C iilto an arc and trans- in each of these walks. In particular, when referring to digraphs, the terms
forming C itself into an asymmetric digraph; 0 is also called an orientation directed path and directed cycle are synonymous with path and cycle. More
of C. The digraph 0 1 of Figure 1.28 is symmetric while O2 is asymmetric; formally, for vertices u and v in a digraph 0, a u-v walk in 0 is a finite,
the digraph 0 3 has neither property. alternating sequence
A digraph 0 is called complete if for every two distinct vertices u and v of
0, at least one of the arcs (u, v) and (v, u) is present in O. The complete
symmetric digraph of order n has both arcs (u, v) and (v, u) for every two of vertices and arcs, beginning with u and ending with v, such that
distinct vertices u and v and is denoted by K~. Indeed, if C is a graph, then aj = (Ui-l, Ui) for i = 1,2, ... ,k.The number k is the length ofthe walk.
C* denotes the symmetric digraph obtained by replacing each edge of G A term that is unique to digraph theory is that of a setniwalk. A u'-v semi-
a
by symmetric pair of arcs. The digraph K~ has size n(n ~ 1) and walk in 0 is a finite, alternating sequence
od v = id v = n - 1 for every vertex v of K~. The digraphs Kj', Ki, K3 and
K: are shown in Figure 1.29. The underlying graph of a digraph D is that
graph obtained by replacing each arc (u, v) or symmetric pairs (u, v), of vertices and arcs, beginning with u and ending with v, such that either
(v, u) of arcs by the edge of uv. Certainly the underlying graph of C* is C. aj = (Uj_l, Uj) or aj = (Uj, Uj-l) for i = 1, 2, ... ,k. If the vertices
A complete asymmetric digraph is called a tournament and will be uo, Ul, ... ,uk are distinct, then the u-v semiwalk is a u'--v semipath. If
studied in some detail in Chapter 5. Uo = Uk! where k ~ 3, and the vertices Ul, U2, •.• ,Uk are distinct, then the
A digraph D is called regular of degree r or r-regular if od v = id v = r for semiwalk is called a semicycle.
every vertex v of O. The digraph K~ is (n - 1)-regular. A I-regular digraph A digraph D is connected (or weakly connected) if for every pair u, vof
0 1 and 2-regular digraph D2 are shown in Figure 1.30. The digraph D2 is a vertices, 0 contains a u-v semipath. (Of course, if D contains a u-v semi-
tournament. path, then 0 contains a v-u sernipath.) Equivalently, D is connected if the
Unlike the Situation for graphs, there are several types of connectedness underlying graph of D is connected. A digraph 0 is strong (or strongly
for digraphs. The terms walk, open and closed walk, trail, path, circuit connected) if for every pair u, v of vertices, D contains both a u-v path
and cycle for graphs have natural counterparts in digraph theory, the and -a v-u path.
A digraph 0 is unilateral if for every pair u, v of vertices, 0 contains
either a u-v path or a v-u path; while D is anticonnected if for every pair

o
u, v of vertices, D contains a u-v semipath that contains no subpath of
length 2. Such a sernipath is referred to as an antidirected semipath orl1n
o antipath. Certainly, if 0 contains a u-v antipath, then 0 contains a v-u
antipath. In' Figure 1.31, the qigraph 0 1 is strong, O2 is unilateral, D3 is
anticonnected and 0 4 is connected but has none of the other connected-
ness properties.
Distance can be defined in digraphs as well. For vertices u and v in a
K*1 K*2 K*3 digraph 0 containing a u-v path, the (directed) distance d(u, v) from u to v is
K*4 the length of a shortest u-v path in O. Thus the distances d(u, v) and
Figure 1.29 Complete symmetric digraphs. d(v, u) are defined for all pairs u, v of vertices in astrong digraph. This
30 Introduction to graphs Digraphs and 1;/1.ultigraphs 31

(a) (b)

Figure 1.31 Connectedness properties of digraphs.

distance is not a metric, in general. Although the distance satisfies the

triangle inequality, it is not symmetric - unless D is symmetric, in (d)
(c)
which case D is, in actuality, a graph. Eccentricity can be defined as
before, as well as radius and diameter. The eccentricity e(u) of a vertex u Figure 1.33 Multigraphs, multidigraphs, pseudographs and pseudodigraphs.
in D is the distance from u to a vertex furthest from u. The minimum
eccentricity of the vertices of D is the radius rad D of D, while the allow parallel edg~s and loops are called pseudographs, while digraphs
diameter diamD is the greatest eccentricity. that permit parallel arcs and loops are called pseudodigraphs. Although
The vertices of the strong digraph D of Figure 1.32 are labeled with their these concepts are useful at times, they will be encountered rarely in
eccentricities. Observe that rad D = 2 and diam D = 5, so it is not true, in our study. Figure 1.33 (a)-(d) shows a multigraph, multidigraph, pseudo-
general, that diamD ::::; 2rad D, as is the case with graphs. graph and pseudodigraph, respectively.
In the definition of a graph G, either one edge or no edge joins a pair of
distinct vertices of G. For a digraph D, two (directed) edges can join dis-
tinct vertices of D - if they are directed oppositely. There are occasions
when we will want topermit more than one edge to join distinct vertices EXERCISES 1.4
(and in the same direction in the case of digraphs).
If one allows more than one edge (but yet a finite number) between the 1.36 Determine all (pairwise nonisomorphic) digraphs of order 4 and
same pair of vertices in a graph, the resulting structure is a multigraph. size 4.
Such edges are called parallel edges. If more than one arc in the same 1.37 Prove or disprove: For every integer n ~ 2, there exists a digraph D
direction is permitted to join two vertices in a digraph, a multidigraph of order n such that for every two distinct vertices u and v of D,
results. A loop is an edge (or arc) that joins a vertex to itself. Graphs that od u =1= od v and id u =1= id v.
1.38 Prove or disprove: No digraph contains an odd number of vertices of
5 odd outdegree or an odd number of vertices of odd indegree.
1.39 Prove or disprove: If D1 and D2 are two digraphs with
V(D 1) = {UI, Uz,"" un} and V(Dz ) = {VI, Vz,···, vn} such that
4 4 idOl U"i = id 02 Vi and od o1 Ui = od o2 Vi for i = 1, 2, ... , n, then
D 1 =D2 •
1.40 Prove that if every proper induced subdigraph of a digraph D of
order n ~ 4 is regular, then E(D) = 0 or D = K~.
3 2 1.41 Prove that there exist regular tournaments of every odd order but
Figure 1.32 Eccentricities in a strong digraph. there are no regular tournaments of even order.
32 Introduction to graphs
1.42 The adjacency ma~rix A(D) ~f a digraph D with V (D) = {VI' VZ, ... , VI!} CHAPTER 2
IS the n. x n matrIX [aij] defined byaij = 1 if (Vi, vo) E E(D) and a·· = 0
otherwIse. J IJ

(a) W~at information do the row sums and column sums of the
adjacency matrix of a digraph provide? Structure and symmetry
(b) Characterize matrices that are adjacency matrices of digraphs.
1.43 Prove that if D is a digraph with od V ;::, 1 imd id v ;::, 1 for every of graphs
vertex v of D, then D contains a cycle.
1.44 Pr?ve that for every two positive integers a and b with a ~ b, there
eXIsts a strong digraph D with rad D = a and diam D = b.
1.45 The center Cen(D) of a strong digraph D is the subdigraph induced
by t~ose. vertices V with e(v) = rad D. Prove that for every asym- Although being connected is the most basic structural property that a
metnc dIgraph D v there exists a strong asymmetric digraph D graph may possess, more information about its structure is provided by
such that Cen(D) = DI. special vertices, edges and subgraphs it contains and the symmetry it
. possesses. This is the theme of the current chapter.

2.1 CUT-VERTICES, BRIDGES AND BLOCKS

Some graphs are connected so slightly that they can be disconnected by

the removal of a single vertex or single edge. Such vertices and edges play
a special role in graph theory, and we discuss. these next.
A vertex v of a graph G is called a cut-vertex ofG if k(G - v) > k(G).
Thus, a vertex of a connected graph is a cut-vertex if its removal produces
a disconnected graph. In general, a vertex v of a graph G is a cut-vertex of
G if its removal disconnects a component of G. The following theorem
characterizes cut-vertices.

Theorem 2.1
A vertex v of a connected graph G is a cut-vertex of G if and only if there exist
vertices u and w (u, w =1= v) such that v is on every u-w path of G.

Proof
Let v be a cut-vertex of G; so the graph G - v is disconnected. If u and w
are vertices in different components of G - v, then there are no u-w
paths in G.- v. However, since G is connected, there are u-w paths in G.
Therefore, every u-w path of G contains v.
Conversely, assume that there exist vertices u and win G such that the
vertex v lies on every u-w path of G. Then there are no u-w paths in G - v,
implying that G - v is disconnected and that v is a cut-vertex of G. 0

The complete graphs have no cut~vertices while, at the other extreme,

each nontrivial path contains only two vertices that are not cut-vertices.
 Cut-vertices, bridges and blocks 35
34 Structure and symmetry of graphs
and Wz be arbitrary distinct vertices of G. If e does not lie on aWrW2 path P
In order to see that this is the other extreme, we prove the following
of G, then P is also a W1-W2 path of G - e. If, however, e lies on a W1-W 2
theorem.
path Q of G, then replacing e by the u-v path (or v-u path) on C ~ot
containing e produces a w1-WZ walk in G - e. By Theorem 1.7, there IS a
Theorem 2.2 w1-W 2 path in G - e. Thus, W1 and Wz are connected in G - e and so e is not

Every nontrivial connected graph contains at least two vertices that are not cut- a bridge.
Conversely, suppose that e is not a bridge of G. Thus G - e is connected.
vertices.
Hence there exists a. u-v path P in G - e; however, P together with e
produce a cycle in G containing e. 0
Proof
Assume that the theorem is false. Then there exists a nontrivial connected A cycle edge is an edge that lies on a cycle. From Theorem 2.4, a cycle
graph G containing at most one vertex that is not a cut-vertex; that is, edge of a graph G is an edge that is not a bridge of G. A bridge incident
every vertex of G, with at most one exception, is a cut-vertex. Let u and with an end-vertex is called a pendant edge.
v be vertices of G such that d( u, v) = diam G. Many of the graphs we encounter do not contain cut-vertices; we
At least one of u and v is a cut-vertex, say v. Let w ~e a vertex belonging discuss these next. A nontrivial connected graph with no cut-vertices is
to a component of G - v not containing u. Since every u-w· path in G called a nonseparable graph. Nontrivial graphs with cut-vertices contain
contains v, we conclude that special subgraphs in which we are also interested. A block of a graph G
is a maximal nonseparable subgraph of G. A block is necessarily an
d(u,w) >d(u,v) =diamG, induced subgraph, 'and, moreover, the blocks of a graph partition its
which is impossible. The desired result now follows. 0 edge set. If a connected graph G contains a single block, {hen G is non-
separable. For this reason, a nonseparable graph is also referred to as a
Analogous to the cut-vertex is the concept of a bridge. A bridge of block itself. Every two blocks have at most one vertex in common, namely
a graph G is an edge e such that k(G - e) > k(G). If e'is a bridge of G, a cut-vertex. The graph of Figure 2.1 has five blocks Bj, 1 ~ i ~ 5, as
then it is immediately evident that k(G - e) = k(G) + 1. Furthermore, if indicated. The vertices V3, Vs and va are cut-vertices, while V3 V S and V4 V S
e = uv, then u is a cut-vertex of G if and only if degu > 1. Indeed, the are bridges; moreover, V4VS is a pendant edge..
complete graph K2 is the only connected graph containing a bridge but Two useful criteria for a graph to be nonseparable are now presented.
no cut-vertices. Bridges are characterized in a manner similar to that
of cut-vertices; the proof too is similar to that of Theorem 2.1 and is
v2 VI V2
omitted. Vi

Theorem 2.3
An edge e ofa connected graph G is a bridge of G if and only if there exist vertices
u and w such that e is on every u-w path of G.
B 1:

V v3

v4
For bridges, there is another useful characterization.
B3: V40----0 VS
V7
Theorem 2.4
An edge e of a graph G is abridge of G if and only if e lies on no cycle of G.

Proof
Assume, without loss of generality, that G is connected. Let e = uv be an
V9 VlO
'Bs:

&
V9 v lO

edge of G, and suppose that e lies on a cycle C of G. Furthermore, let W1 Figure 2.1 A graph and its five blocks.
36 Structure and symmetry of graphs Cut-vertices, bridges and blocks 37
Theorem 2.5 Theorem 2.5 suggests the following definitions: A block of order at least
3 is called a cyclic block, while the block Kz is called the acyclic block.
A graph G of order at least 3 is nonseparable if and only if every two vertices of
We now state a theorem of which Theorem 2.2 is a corollary.
G lie on a common cycle of G.

Proof Theorem 2.7

Let G be a graph such that each two of its vertices lie on a: cycle. Thus G is Let G be a connected graph with one or more cut-vertices. Then among the blocks
connected. Suppose that G is not nonseparable. Hence G contilins a cut- of G, there are at least two which contain exactly one cut-vertex of G.
vertex v. By Theorem 2.1, there exist vertices U and w such that v is on
every u-w path in G. Let C be a cycle of G containing u and w. The In view of Theorem 2.7, we define an end-block of a graph G as a block
cycle C determines two distinctu-w paths, one of which does not contain containing exactly one cut-vertex of G. Henceevery connected graph with
v, contradicting the fact that every u-w path contains v. Therefore, G is at least one cut-vertex contains at least two end-blocks. In this context,
nonseparable. another result that is often useful is presented. Its proof will become
Conversely, let Gbe a nonseparable graph with n > 3 vertices. We show evident in the next chapter.
that every two vertices of G lie on a common cycle of G. Let u be an arbi-
trary vertex of G, and denote by U the set of all vertices that lie on a cycle
containing u. We now show that U = V = V(G). Assume that U::f= V; Theorem 2.8
so there exists a vertex v E V - U. Since G is nonseparable, it contains Let G be a graph with at least one cut-vertex. Then G contains a cut-vertex v
no cut-vertices, and furthermore, since n > 3, the graph G contains no with the property that, with at most one exception, all blocks of G containing v
bridge. By Theorem 2.4, every edge of G lies on a cycle of G; hence, are end-blocks.
every vertex adjacent to u is an element of U. Since G is connected,
there exists a u-v path W: u = Uo, UI, UZ, ... ,Uk = v in G. Let i be the
Another interesting property of blocks of graphs was pointed out by
smallest integer, 2 ~ i ~ k, such that Ui ¢ U; thus Ui-I E U. Let C be a
Hararyand Norman [HN2].
cycle containing u and ui-l' Because ui-l is not a cut-vertex of G, there
exists a Ui-U path P: Ui = vo, VI' Vz, ... , VI = U not containing Ui-l' If the
0u: y verte:: common to P and C is u, then a cycle containing U and Ui Theorem 2.9
eXIsts, which produces a contradiction. Hence P and C have a vertex in
common different from u. Let j be the smallest integer, 1 :::s; j :::s; t, such that The center of every connected graph G lies in a single block of G.
Vj belongs to both P and C. A cycle containing u and u;can now be con-
structed by beginning with the· Ui-Vj subpath of P, proceeding along C Proof
from Vj to U and then to Ui-lt and fmally taking the edge Ui-IUi back to
Ui' Thus, a contradiction arises again, implying that the vertex v does not Suppose that G is a connected graph whose center Cen(G) does not lie
exist and that every two vertices lie on a cycle. 0 within a single block of G. Then G has a cut-vertex v such that G - v con-
tains components G} andGz, each of which contains vertices of Cen(G).
An internal vertex of a u-v path P is any vertex of P different from U or v. Let ube a vertex such that d(u, v) = e(v), and let PI be a v-upath of G
A collection {PlJ Pz ,... , Pk} of paths is called internally disjoint if each having length e(v). At least one of G} and Gz, say Gz, contains no vertices
internal vertex of Pi (i = 1,2, ... , k) lies on no path Pj U::f= i). In particular, of Pl' Let w be a vertex QfCen(G) belonging to Gz, and let P z be a w-v path
two u-v paths are internally disjoint if they have no vertices in common, of minimum length. The paths PI and P z together form a u-w path P3 ,
other than U and v. Edge-disjoint u-v paths have no edges in common. A which is necessarily a u~w path of length d(u, w). However, then
second characterization of nonseparable graphs is now apparent. e(w) >e(v), which contradicts the fact that w is a central vertex. Thus
Cen(G) lies in a single block of G. 0
Corollary 2.6
A graph G is a critical block if G is a block and for every vertex v, the
A graph G of order at least 3 is nonseparable. if and only if there exist two graph G - v is not a block. Hence Cl block G is noncritical if and only if
internally disjoint u-v paths for every two distinct vertices uand v of G. there exists a vertex v of G such that G- v is also a block. There is an
38 Structure and symmetry of graphs Cut-vertices, bridges and blocks 39
Proof
Suppose that G is a minimal block of order at least 4, but that G contains
no vertices of degree 2. By Theorem 2.10, G is not a critical block. Thus,
G contains a vertex W such that G - W is a block. Let e be an edge of G
incident with w. Since G is a minimal block, G - e is not a block, and there-
fore G - e contains a cut-vertex u (iow). Hence G - e - u (= G - u - e) is
Figure 2.2 Minimal and critical blocks. disconnected, so e is a bridge of G - u. On the other hand, since G - u - W
(= G - W - u) is connected, e is a pendant edge of G - u and W is an end-
analogous concept concerning edges. A graph G is a minimal block if G is a vertex of G - u. Therefore, W has degree 1 in G - u and degree 2 in G. This
block and for every edge e, the graph G - e is not a block. is a contradiction. 0
The block GI of Figure 2.2 is minimal and noncritical, while the block Gz
is critical but nonminimal.
In each of the graphs of Figure 2.2, there are vertices of degree 2. All
minimal and critical blocks have this property, as we shall see. EXERCISES 2.1

2.1 Prove that if ~s a cut-vertex of a connected graph G, then v is not a

Theorem 2.10 cut-vertex of G.
If G is a critical block of order at least 4, then G contains a vertex of degree 2. 2.2 Prove Theorem 2.3.
2.3 Prove that every graph containing only even vertices is bridgeless.
Proof 2.4 Prove Corollary 2.6.
For each vertex x of G, there exists a vertex y of G - x such that G - x - y
2.5 Let u and v be distinct vertices of a nonseparable graph G of order
is disconnected. Among all such pairs x, y of vertices of G, let u, v be a
n ~ 3. If P is a given u-v path of G, does there always exist jl u-v path
pair such that G - U - v is disconnected and contains a component GI
Q such that P and Q are internally disjoint u-v paths?
of minimum order k. If k = I, then the vertex of G 1 has degree 2 in G.
Thus we may assume that k ~ 2. Let Gz denote the union of the com- 2.6 Let G and H be graphs with V(G) = {VI, V2,"" vn } and V(H) =
ponents of G - u - v that are different from G1 . Further, let H = {uJ, U z,"" un}, n ~ 3..
(V(G I ) U {u, v}). (a) Vertices Ui and Uj are adjacent in H if and onlyif Vi and v· belong
Let WI E V(G I ). There exists a vertex W2 in G - WI such that to a common cycle in G. Characterize those graphs G fdr which
G - WI - W2 is disconnected. We now consider two cases. H is complete.
(b) Vertices Ui and Uj are adjacent in H if and only if degG Vi + degG v·
Case 1. Assume that Wz E V(H). Since both (V(G 2) u {u}) and is odd. Prove that H is bipartite. ]
(V(G2 ) U {v}) are connected, some component of G - WI - Wz has order
less than k, producing a contradiction. 2.7 An element of a graph G is a vertex or an edge of G. Prove that a graph
G of order at least 3 is nonseparable if and only if every pair of
Case 2. Assume that Wz E V(G z ). Since G - WI - Wz is disconnected, elements of G lie on a common cycle of G.
H -: WI must contain exactly two components, namely a component
H u containing u and a component H v containing v. If either H u or H v 2.8 Let G be a graph of order n'~ 3 with the property that
is trivial, then G has a vertex (namely u or v) of degree 2; so we may degu + degv ~ n for every pair u, v of nonadjacent vertices of G.
assume that H u and H v are nontrivial. However, G - WI - U is then Show t-hat G is nonseparable.
disconnected and has,a component of order less than k, again producing 2.9 Does there exist a noncritical blocJ< G containing an edge e = uv such
a contradiction. 0 that G - e is a block, but neither G - u nor G - v is a block?
2.10 Does there exist a graph other than K2 and the n-cycles, n ~ 4, that is
Corollary 2.11 a critical block as well as a minimal block?
If G is aminimal block of order at least 4, then G contains avertex of degree 2.
40 Structure and symmetry of graphs The automorphism group of a graph 41
2.2 THE AUTOMORPHISM GROUP OF A GRAPH Theorem 2.12
For every graph G, Aut(G) ~ Aut(G).
We have already described one way of studying the structur~ of graphs,
namely, by determining the number and location of specIal vertices,
edges and subgraphs. Another natural way of studying the structur~ of We mentioned previously that Aut(KIl ) ~ Sn for every positive integer
graphs is by investigating their symmetries. A common method of domg n. Certainly, if G is a graph of order n containing adjacent vertices as well
this is by means of groups. . . as nonadjacent vertices, then Aut(G) is isomorphic to a proper subgroup
An automorphism of a graph G is an isomorphism between G and Itself. of the symmetric group 5 n• Combining this observation with Theorem
Thus an automorphism of G is a permutation of V(G) that preserves 2.12 and Lagrange's Theorem on the order of a subgroup of a finite
adjacency (and nonadjacency). Of course, the identity .function. on V( G) group, we arrive at the follOWing.
is an automorphism of G. The inverse of an automorphIsm of G IS also an
automorphism of G, as is the composition of two automorphisms ~f G. Theorem 2.13
These observations lead us to the fact that the Set of all automorphisms
The order IAut(G) Iofthe automorphism group ofagraph G oforder n is a divisor
of a graph G form a group (under the operation of composition), called the
ofn! and equals n! if and only if G = Kn or G = Kn .
automorphism group or simply the group of G and denoted by Aut(G).
The. automorphism group of the graph G1 of Figure 2.3 is cyclic of order
Two labelings of a graph G of order n from the same set of n labels are
2, which we write as Aut(G1 ) ~ Z2' In addition to the identity permuta-
considered distinct if they do not produce the same edge set. With the aid
tion E on V(G I ), the group Aut(G1 ) contains the 'reflection' a = (~y)(v x),
of the automorphism group of a graph G of order n, it is possible to
where a is expressed as 'permutation cycles'. The graph G2 of FIgure 2.3
determine the number of distinct labelings of G.
has only the identity automorphism, so Aut(G2 ) ~ Zl'
Every permutation of the vertex set of Kn is an automorphism and so
Aut(Kn) is the symmetric group Sn of order n! On the other hand, the Theorem 2.14
automorphism group of Cn , n ~ 3, is the dihedral group D n of order 2n, The number of distinct labelings of a graph G of order n from a set of n labels is
consisting of n rotations and n reflections. The 4-cycle C4 and its auto- n!/IAut(G)I·
morphism group are illustrated in Figure 2.4.
Next we present a few basic facts concerning automorphism groups of
graphs. We have already noted that every automorphism of a graph Proof
preserves both adjacency and nonadjacency. This observation leads to Let 5 be a set of n labels. Certainly, there exist n! labelings of G using the
the following result. elements of 5 without regard to which labelings are distinct. For a given
w
labeling of G, each automorphism of G gives rise to an identical labeling
w
of G; that is, each labeling of G from S determines IAut(G)I identical

~
labelings of G. Hence there are n!/IAut(G)1 distinct labelings of G. 0

u v x y u v x y z As an illustration of Theorem 2.14, consider the graph G = P3 of Figure

2.5 and the set 5 = {I, 2, 3}. Since Aut(G) ~ Z2, the number of distinct
Figure 2.3 Graphs with cyclic automorphism groups of orders 2 and 1.
labelings of G is 3!/2 = 3. The three distinct labelings of Theorem 2.14
are shown in Figure 2.5.
Automorphisms: £, (Xl = (u v w x ),
G: '0--0--0 S ={I, 2, 3}
= (u w ) (v x ), (X3 = (u x w v ),
(X2
q»1 = (u w), q»2 = (v X),
q»3 = (u v) (w x ). q»4 = (u x) (v w )
12 3 1 3 2 2 1 3
x w
0--0--0 0--0--0 0--0--0
Figure 2.4 The 4-cycle and its automorphism group. Figure 2.5 Distinct labelings of graphs.
42 Structure and symmetry of graphs The automorphism group of a graph 43

orbits:
G: 5 8 {l},(2,3,4}
(5 7, 9), (6,8, lO}

10 9
Figure 2.6 The orbit of a graph.

If a relation R is defined on the vertex set of a graph G by uRv if ¢u = v

for some automorphism ¢ of G, then R is an equivalence relation. The
equivalence classes of R are referred to as orb~ts, and two vertices b:lo~g­
ingto the same orbit are called similar vertIces. Of course, two sImIlar
Figure 2.8 Vertex-transitive graphs and regular graphs that are not vertex-
vertices have the same degree. The automorphism group of the graph G transitive.
of Figure 2.6 is cyclic of order 3 and G has four orbits. . .
In Chapter 1 we defined the total distance td(u) of a vertex u ~ a con- Two edges el = UlVI and e2 = U2V2 of a graph G are similar if there exists
nected graph G as the sum of the distances between u and all vertices of G. an automorphism ¢ of G such that ¢(el) = e2, that is, either 4>(Ul) = U2 and
For the graph G of Figure 1.24 we computed the total distance of each
4>(Vl) = v2J or 4>(Ul) = v2 and 4>(Vl) = uz. A graph G is edge-transitive if
vertex. This graph is shown again in Figure 2.7. The orbits of this graph every two edges of G are similar. The graph HI = K3 X K2 of Figure 2.9
are {rl, r2}, {s}, {t 1 , t z}, {u}, {v}, {Wl, wz}, {x}, {y} and {z}. S~nce rl and rz is vertex-transitive but not edge-transitive, while Hz = P3 is edge-
are similar vertices, they necessarily }lave the same total distances. The transitive but not vertex-transitive.
same can be said of tl 'and t2 , as well as of Wl and Wz. Of course, it is possible The following result is due to E. Dauber (see Harary [H7, p. 172]).
for vertices that are not similar to have the same total distances, as is the
case with u and v. For the graph G of Figure 2.7, Aut(G) 9;! Zz x Zz X Zz.
A graph that contains a single' orbit is called vertex-transitive. Thus a Theorem 2.15
graph G is vertex-transitive if and only if for every two vertices u an? v Every edge-transitive graph without isolated vertices is either vertex-transitive or
of G, there exists an automorphism ¢ of G such that ¢u = v. Necessarily, bipartite.
every vertex-transitive graph is regular. The graphs K n (n ~ I), Cn (n ~ 3)
and Kr,r (r ~ 1) are vertex-transitive. The Petersen graph (Figure 1.9) is
vertex-transitive. Also the regular graphs G1 = Cs X K2 and Gz = K2,z,z Proof
of Figure 2.8 are vertex-transitive, while the regular graphs G3 and G4
Let G be an edge-transitive graph without isolated vertices such that
are not vertex-transitive.
E(G) = {el, ez, ... ,em }. Suppose that el=vlv2, For each integer i

G: 0---0---0
y z

t2
Figure 2.9 A vertex-transitive graph that is not edge-transitive and an edge-
Figure 2.7 Similar vertices in a graph. transitive graph that is not vertex-transitive.
44 Structure and symmetry of graphs Cayley color graphs 45
2.12 Describe the elements of Aut(Cs ).

o
Figure 2.10 Digraphs with cyclic automorphism groups.

(1 ::;; i ::;; m), there exists an automorphism cPi of G such that. cPi(el) = ei'
2.13 Determine the number of distinct labelings of Kr,r'
2.14 For which pairs k, n of positive integers with k ::;; n does there exist a
graph G of order n having k orbits?
2.15 For which pairs k, n of positive integers does there exist a graph G
of order n and a vertex v of G such that there are exactly k vertices
Let VI = {cPi(Vl)ll ::;; i::;; n} and V2 = {cPi(V2) 11 ::;; i::;; m}. Since G has no similar to v?
isolated vertices, VI U V 2 = V(G). We now consider two cases. 2.16 Describe the automorphism groups of the digraphs below.
Case 1. Assume that VI and V 2 are disjoint. We show that G is a bipartite
graph with partite sets VI and V 2. Suppose, to the contrary, that one of u u v u v

[I) X
VI and V2 contains adjacent vertices; say VI contains adjacent vertices x
and y, with xy = ej' Thus cPj(el) = ej' By definition, one of x and y is in V2,
contrary to the hypothesis.
Case 2. Assume that VI and V 2 are not disjoint. We show that G is vertex- v w w x x y
transitive. Let u and w be any two vertices of G. Suppose, first, that both
vertices belong to VI or both belong to V 2, say the former. Then there exist
1
automorphisms cPk and cPt such that cPkVl= u and cPtVI = w. Then cPtcPi: is
an automorphism and <PtcP-,;l(u) = w; so u and ware similar. Suppose,
next, that U E VI and w E V 2. Let v E VI n V 2. From what we have just 2.3 CAYLEY COLOR GRAPHS
shown, u and v are'similar, as are v and w. Thus, u and ware similar. 0
We have seen that we can associate a group with every graph or digraph.
Every digraph also has an automorphism group. An automorphism of a We now consider the reverse question of associating a graph and a
digraph D is an isomorphism of D with itself, that is; an q\ltomorphism of digraph with a given group. We consider only finite groups in this con-
D is a permutation O! on V(D) such that (u, v) is an arc of D if and only if text. A nontrivial groupr is said to be generated by nonidentity elements
(O!u, O!v) is an arc of D. The set of all automorphisms under composition hI, h2,··· ,hk (and these elements are called generators) of r if every
forms a group, called the automorphism group of D and denoted by Aut(D). element of r can be expressed as a (finite) product of generators. Every
While we have seen (the graph G of Figure 2.6) that it is not necessarily nontrivial finite group has a finite generating set (often several such
easy to find a graph G with Aut(G) ~ Z3' this is actually quite easy for sets) since the set of all nonidentity elements of the group is always a
digraphs. For digraphs D1 and D2 of Figure 2.10, Aut(D1) e:! Z3 and generating set for r. .
Aut(Dz) e:! Zs. Let r be a given nontrivial finite group with .6. = {hI, hZ1 ••• , hk} a
generating set for r. We associate a digraph with rand .6. called the
Cayley color graph of r with respect to .6. and denoted by D6,(r). The
vertex set of D6, (r) is the set of group elements of r; therefore, D6, (r)
EXERCISES 2.2
has order Ifi. Each generator hi is now regarded as a color. For gv
gz E r, there exists an arc (gIl gz) colored hi in D6,(r) if and only if
2.11 For the graphs GI and Gz below, describe the automorphisms of G1
g2 = glh;. If hi is a group element of order 2 (and is therefore self-inverse)
and of G2 in tenns of permutation cycles.
and g2 = glh i, then necessarily gl = g2hi' When a Cayley color graph
t y 1 6 D6,(r) contains each of the arcs (gl,~8z) and (82, gl), both colored hi,

~ ~
then it is customary to represent this symmetric pair of arcs by the
single edge g182'
We now illustrate the concepts just introduced. Let r denote the sym-
u z 2 7 metric group 53 of all permutations on the set {1, 2, 3}, and let.6. = {a, b},
46 Structure and symmetry of graphs Cayley color graphs 47
b Proof
Let f={gj,g2, ... ,gll}' For i=1,2, ... ,n, define ai:V(DA(r))->
V(DD.(r) by aigs = gigs for 1 ~ s ~ n. Since f is a group, the mapping
ai is one-to-one and onto. Let h E 1:1. Then for each i, 1 ~ i ~ n, and for
each s, 1 ~ s ~ n,
ai(gsh) = gi(gsh) = (gigs)h = (aigs)h.
Hence, by Theorem 2.16, ai is a color-preserving automorphism of DD. (f).
Next, we verify that the mapping ¢, defined by ¢gi = ai, is an iso-
morphism from f to the group of color-preserving automorphisms of
ba ~:::::""_------"",,--"";;:l()
ba 2
a DA(r)· The mapping </> is clearly one-to-one since ai =f. aj for i =f. j.
To show that </> is operation-preserving, let gi, gj E r be given, and sup-
Figure 2.11 A Cayley color graph. pose that gigj = gk· Then ¢(gigj) = ¢gk = ak and (</>gi)(¢gj) = aiaj' Now,
for each s, 1 ~ s ~ n, akgs = gkgs' Furthermore, gkgs = (gigj)gs = gi(gjgs) =
where a = (123) and b = (12). The Cayley color graph DD. (r)in this case is ai(gjgs) = aj(ajgs) = (aiaj)gS' Hence, for each s, 1 ~ s ~ n,akgs = (ajaj)gs
shown in Figure 2.11. so that ak = aiaji that is, </>(gjgj) = (¢gj)(¢gj)'
If the generating set 1:1 of a given nontrivial finite group r with n Finally, we show that the mapping </> is onto. Let a be a color-preserving
elements is chosen to be the set of all nonidentity group elements, then automorphism of DA(r). We show that a = ai for some i (1 ~ i ~ n).
for every two vertices gl, g2 of DD.(f), both (gl, g2) and (g2, gl) are arcs Suppose that agI = g" where gj is the identity of f. Let gs E f. Then gs
(although not necessarily of the same color) and DD.(r) is the complete can be expressed as a product of generators, say
symmetric digraph K~ in this case. gs = hI h2 •• .h r,
Let f be a nontrivial finite group with generating set 1:1. An element
a E Aut(DD. (f» is said to be color-preserving if for every arc (gl, g2) of where hj E 1:1, 1 ~ j ~ t. Hence,
DD.(r), the arcs (gl, g2) and (agI, ag2) have the same color. For a given ags = a(glgS) = a(gl hIh2" .ht )·
nontrivial finite group f with generating set 1:1, it is a routine exercise to
prove that the set of all color-preserving automorphisms of DD. (r) forms a By successive applications of Theorem 2.16, it follows that
subgroup of Aut(DD.(f». A useful characterization of color-preserving ags = (agj)h Ih2.·· ht = (agI)gs = grgs = args'
automorphisms is given in the next result.
Therefore, a = a r and the proof is complete. 0

Theorem 2.16 For more information on Cayley color graphs, see White [W5].
In 1936 the first book on graph theory was pUblished. In this book the
Let f be a nontrivial finite group with generating set 1:1 and let a be a permutation
author Konig [KlO, p. 5] proposed the problem of determining all finite
ofV(DA(f». Then a is a color-preserving automorphism OfDD.(f) if and only if
groups r for which there exists a graph G such that Aut(G) ~ f. The
a(gh) = (ag)h problem was solved in 1938 by Frucht [FIll who proved that every
finite group has this property. We are now in a position to present a
for every g E rand h E 1:1. proof of this result.
If r is the trivial group, then Aut(G) ~ f for G = KI . Therefore, let
The major significance of the group of color-preserving automorphisms r = {gl> g2, .... ,gn}, n ~ 2, be a given finite group, and let 1:1 =
of a Cayley color graph is contained in the following theorem. {hI, h2 ,· .. ,ht }, 1 ~ t < n, be a generating set for r. We first construct
the Cayley color graph DA(f) of r with respect to 1:1; the Cayley color
graph is actually a digraph, of course. By Theorem 2.17, the group of
Theorem 2.17
color-preserving automorphisms of DD.(f) is isomorphic to f. We now
Let f be a nontrivial group with generating set 1:1. Then the group of color- transform the digraph DD.(r) into a graph G by the follOWing technique.
preserving automorphisms of DD.(r) is isomorphic to r. Let (gi, gj) be an arc of DA(r) colored hk• Delete this arc and replace it by
48 Structure and symmetry of graphs Cayley color graphs 49
underlying graph of a Cayley color graph DA(f) is called a Cayley graph
and is denoted by GE,(r). Thus a graph G is a Cayley graph if and only if
there exists a finite group f and a generating set II for f such that
G = GA(f), that is, the vertices of G are the elements off and two vertices
gl and g2 of G are adjacent if and only if either gl = g2h or gz = glh for
some hEll.
As observed earlier, K~ is a Cayley color graph; consequently, every
complete graph is a Cayley graph. Since K2 x K3 is the underlying graph
of the Cayley color graph of Figure 2.11, K2 x K3 is a Cayley graph. Every
Cayley graph is regular. In fact, if Il' denotes those elements of II of order
at least 3 whose inverse also belongs to ll, then degv = Illl-!lll'l for
every vertex v in GA(f). Indeed every Cayley graph is vertex-transitive.
The converse is not true, however. For example, the Petersen graph
(Figure 1.9) is vertex-transitive but it is not a Cayley graph.

EXERCISES 2.3
h3
Of--.-..
..~....,O 2.17 Construct the Cayley color graph of the cyclic group of order 4 when
D~(r)
the generating set II has (a) one element and (b) three elements.
G
2.18 Prove Theorem 2.16.
Figure 2.12 Constructing a graph G from a given group f.
2.19 Determine the group of color-preserving automorphisms for the
the 'graphical' path gi, Uij' U;j' gj' At the vertex uij we construct a new path Cayley color graph DA(f) below.
P ij of length 2k - 1 and at the vertex U;j a path I1j of length '2Jc. This
a
construction is now performed with every arc of DE, (f), and is illustrated
in Figure 2.12 for k = 1, 2 and 3.
The addition of the paths Pij and I1j in the formation of G is, in a sense,
equivalent to the direction and the color of the arcs in the construction
of DA(f). a
a
It now remains to observe that every color-preserving automorphism of
DA(r) induces an automorphism of G, and conversely. Thus we have a
proof of Frucht's theorem.

Theorem 2.18 a
2.20 Determine the smallest integer n > 1 such that there exists a con-
For every finite group f, there exists a graph G such that Aut(G) ~ f.
nected ~raph G of order n such that IAut(G)1 = 1.
The condition of having a given group prescribed is not a particularly 2.21 Find a nonseparable graph G whose automorphism group is iso-
stringent one for graphs. For example, Izbicki [12] showed thaUor every morphic to the cyclic group of order 4.
finite group f and integer r ~ 3, there exists an r-regular graph G with
Aut(G) ~ f. 2.22 For a given finite group f, determine an infinite number of mutually
We have now seen that for every finite group f and generating set ~ nonisomorphic graphs whose groups are isomorphic to f.
there is an associated digraph, namely the Cayley color graph DA(f). The 2.23 Show that every n-cycie is a Cayley graph.
50 Structure and symmetry of graphs The Reconstruction Problem 51
2.24 Show that the cube Q3 is a Cayley graph. The Reconstruction Problem is thus to determine the truth or falsity of
the Reconstruction Conjecture.
2.25 Let f be a finite group generated by ~. The condition on the order in the Reconstruction Conjecture is needed
(a) Show that the group of color-preserving automorphisms of since if G1 = Kz, then G1 is not reconstructible. If G2 = 2K1, then the sub-
D~(f) is a subgroup of Aut(G~(r). graphs G1 - v, where v E V(G 1 ), and the subgraphs G2 - v, for v E V(G z),
(b) If the group of color-preserving automorphisms of D~(f) is are precisely the same. Thus G1 is not uniquely determined by its sub-
isomorphic to Aut(G~(r) and G~(f) = K", then find f. graphs G1 - v, v E V(G1 ). By the same reasoning, G2 r:::,,; 2K1 is also non-
(c) Prove or disprove: If the group of color-preserving automorph- reconstructible. The Reconstruction Conjecture claims that K z and 2K1
isms of D~, (f 1) is isomorphic to the group of color-preserving are the only nonreconstructible graphs.
automorphisms of D~2(fz), then f 1 r:::,,; fz. Before proceeding further, we note that there is a related problem.that
(d) Prove or disprove: If Aut(G~l (f1 » r:::,,; Aut(G~2(f2»' then we shall not consider. Given graphs G1 , Gz, ... , Gn , does there eXIst a
f 1 r:::,,; f 2. graph G with V(G) = {Vb vz, ... ,Vn } such that Gi = G ~ Vi for i = I,
2, ... , n? The answer to this question is not known in general. Although
there is a similarity between this question and the Reconstruction
Problem, the question is quite distinct from the problem in which we
2.4 THE RECONSTRUCTION PROBLEM are interested.
If there is a counterexample to the Reconstruction Conjecture, then it
If ¢ is an automorphism of a nontrivial graph G and u is a vertex of G, then must have order at least 10, for, with the aid of computers, McKay [M4]
G - u = G - ¢u, that is, if u and v are similar vertices, then G - u = G - v. and Nijenhuis [NS] have shown that all graphs of order less than 10 (and
The converse of this statement is not true, however. Indeed, for the greater than 2) are req:mstructible. The graph G of Figure 2.14 is therefore
vertices u and v of the graph G of Figure 2.13, G - u = G - v, but u and reconstructible since its order is less than 10. Hence the graphs G -'- Vi
v are not similar vertices of G. (1 ,,;; i ,,;; 6) uniquely determine G. However, there exists a graph H with
These observations suggest the problem of determining how much V(H)={Vl,VZ"",V6} such that G-vi=H-vj for l,,;;i,,;;S, but
structure of a graph G is discernible from its vertex-deleted subgraphs. G - V6 =1= H - V6' Therefore, the graphs G - Vi (1 ,,;; i,,;; 5) do not uniquely
This, in fact, brings us to a famous problem in graph theory. determine G. On the other hand, the graphs G - Vi (4 ,,;; i ,,;; 6) do uniquely
Probably the foremost unsolved problem in graph theory is the Recon- determine G.
struction Problem. This problem is due to P. J. Kelly and S. M. Ulam and its Digraphs are not reconstructible, however. The vertex-deleted sub-
origin dates back to 1941. We discuss it briefly in this section. digraphs of the digraphs D1 and Dz of Figure 2.15 are the same; yet
A graph G with V(G) = {V1, Vz, ... , v,,}, n ~ 2, is said to be reconstruc-
tible if for every graph H having V (H) = {U1, Uz, ... , u,,}, G - Vi = H - Ui
for i = 1,2, ... , n implies that G = H. Hence, if G is a reconstructible
graph, then the subgraphs G - v, v E V(G), determine G uniquely.
We now state a conjecture of Kelly and Ulam, the following formulation G:
of which is due to F. Harary.

The Reconstruction Conjecture

Every graph of order at least 3 is reconstructible.

G:

u v
Figure 2.13 A graph with nonsimilar vertices whose vertex-deleted subgraphs are
isomorphic. Figure 2.14 A reconstructible graph.
52 Structure and symmetry of graphs The Reconstruction Problem 53

Figure 2.15 Two nonreconstructible digraphs.

G 1:

~ G2:

~
0 0
D1 f= D2 · Indeed, for digraphs, there are infinite pairs of counterexamples
(5tockmeyer [59]). . G 3: G 4:
There are several properties of a graph G that can be found by consid-
ering the subgraphs G - v, v E V(G). We begin with the most elementary
of these.

Theorem 2.19
If G is an (n, m) graph with n ;;, 3, then nand m as well as the degrees of the
vertices of G are determined from the n subgraphs G - v, v E V(G).

Proof
It is trivial to determine the number n, which is necessarily one greater
than the order of any subgraph G - v. Also, n is equal to the number of
G s:

Z
Figure 2.16 The subgraphs G - v of a graph G.
G 6:

Z
graph regularity is recognizable; indeed, the degree of regularity is a
subgraphsG - v. recognizable parameter. For regular graphs, much more can be said.
Todeterminem label these subgraphsby Gi, i = 1,2, ... , n,and suppose
thatG i = C - Vi' w);ereVi E V(G). Letrni denote the size ofG i. Consider an
Theorem 2.20
arbitrary edge e of G, saye = VjVk' Then e belongs to n - 2 of the subgraphs
Git namely all except Gj and Gk • Hence, 2:7=1 mi counts each edge n - 2 Every regular graph of order at least 3 is reconstructible.
times; that is, 2:7=1 mi = (n - 2)m. Therefore;
2:7=1 mi Proof
m= n-2 .
As we have already mentioned, regularity and the degree of regularity are
The degrees of the vertices of G can be qetermined by simply noting that recognizable. Thus, without loss of generality, we may assume that G is an
deg Vi = m - mi, i = 1, 2, ... , n. 0 r-regular graph with V(G) = {v" V2,"" Vn }, n ;;, 3. It remains to show
that G is uniquely determined by its subgraphs G - Vi, i = 1, 2, ... , n.
We illustrate Theorem 2.19 with the six subgraphs G - v shown in Consider G - Vv say. Add vertex Vl to G - VI together with all those
Figure 2.16 of some unspecified graph G. From these subgraphs we deter- edges V1V where degG-vI V = r - 1. This produces G. 0
mine n, m and degvi for i = 1, 2, ... ,6. Clearly, n = 6. By calculating the
integers mi, i = 1, 2, ... ,6, we find that m = 9. Thus, deg VI = deg V2 = 2, If G has order n ;;, 3, then it is discernible whether G is connected from
degv3 = degv4 = 3, and degvs = degv6 =j:. the n subgraphs G - v, v E V(G).
We say that a graphical parameter or graphical property is recognizable
if, for each graph G of order at least 3, it is possible to determine the value
Theorem 2.21
of the parameter for G or whether G has the property. from the subgraphs
G -' v, v E V(G). Theorem 2.19 thus states that for a graph of order at least For graphs of order at least 3, connectedness is a recognizable property. In parti-
3, the order, the size, and the degrees of its vertices are recognizable cular, ifG is agraph with V(G) == {Vl, v2, ... ,Vn }, n ~ 3, then G is connected if
parameters. From Theorem 2.19, it also follows that the property of and only if at least two of the subgraphs G - Vi are connected.
 The Reconstruction Problem 55
54 Structure and symmetry of graphs
denote the set of those graphs Gi in S' having a maximum number of
Proof components isomorphic to F'.
Let G be a connected graph. By Theorem 2.2, G contains at least two Assume that Gl , G2 , •.. , G1 (t ~ 1) are the elements of S". Each graph Gi
vertices that are not cut-vertices, implying the result. in S" has k(G) components. Since each graph Gi (1 ::s:; i ::s:; t) has a minimum
Conversely, assume that there exist vertices V1, V2 E V( G) such that both number of components isomorphic to F, each vertex Vi (1 ::s:; i ::s:; t) belongs
G - Vl and G - V2 are connected. Thus, in G - V1 and also in G, vertex v2 is to a component Fi of G isomorphic to F, where the components Fj of G
connected to Vj, i ~ 3. Moreover, in G - V2 (and thus in G), v1 is connected (1 ::s:; i :::; t) are not necessarily distinct. Further, since each graph Gi
to each Vi, i ~ 3. Hence every pair of vertices of G are connected and so G is (1 ::s:; i::s:; t) has a maximum number of components isomorphic to F', it
connected. 0 follows that F j - Vi = p' for each i = 1, 2, ... , t. Hence, every two of the
graphs G1 , G2 , ... ,G1 are isomorphic, and G can be produced from G"
Since connectedness is a recognizable property, it is possible to deter- say, by replacing a component of G1 isomorphic to F' by a component
mine from the subgraphs G - v, v E V(G), whether a graph G of order at isomorphic to F. 0
least 3 is disconnected. We now show that disconnected graphs are
reconstructible. There have been several proofs of this fact. The proof It can be shown that (connected) graphs of order at least 3 whose com-
given here is due to Manvel [MIl. plements are disconnected are reconstructible (Exercise 2.30). However,
it remains to be shown that all connected graphs of order at least 3 are
reconstructible.
Theorem 2.22
Disconnected graphs of order at least 3 are reconstructible.
EXERCISES 2.4

Proof 2.26 Reconstruct the graph G whose subgraphs G - V, v E V( G), are given
We have already noted that disconnectedness in graphs of order at least 3 in Figure 2.13.
is a recognizable property. Thus, we assume without loss of generality 2.27 Reconstruct the graph G whose subgraphs G - V, V E V(G), are given
that G is a disconnected graph with V(G) = {Vb V2,"·' V n }, n ~ 3. in the accompanying figure.
Further, let Gi = G - Vi fori = 1, 2, ... ,no From Theorem 2.19, the degrees
of the vertices Vi, i = 1, 2, ... , n, can be determined from the graphs G - Vi'
Hence, if G contains an isolated vertex, then G is reconstructible. Assume
then that G has no isolated vertices.
Since every component of G is nontrivial, it follows that k(Gi) ~ k(G) for
i = 1, 2, ... , n and that k(Gj ) = k(G) for some integer j satisfying 1 ::s:; j ::s:; n.
Hence the number of components of G is rnin{k(Gi)li = 1, 2, ... , n}.
Suppose that F is a component of G of maximum order. Necessarily, F 2.28 Let G be a graph with V(G) = {Vl' V2,"" V7} such that G - Vi = K2,4
is a component of maximum order among the components of the for i = 1, 2, 3 and G - Vi =·K33 for i = 4,5,6,7. Show that G is
graphs G;; that is, F is recognizable. Delete a vertex that is not a cut- reconstructible. '
vertex from F, obtaining F'. 2.29 Show that the digraphs of Figure 2.15 are not isomorphic.
Assume that there are r (~1) components of G isomorphic to F. The
number r is recognizable, as we shall see. Let 2.30 (a) Prove that if G is reconstructible, then G is reconstructible.
(b) Prove'that every graph of order n(~3) whose complement is
5 = {G;ik(G i ) = k(G)} , disconnected is reconstructible.
and let 5' be the subset of 5 consisting of all those graphs Gi having a 2.31 Prove that bipartiteness is a recognizable property.
minimum number t of components isomorphic toP. (Observe that if
r = I, then there exist graphs Gi in S containing no components iso- 2.32 Reconstruct the graph G whose subgraphs G - V, V E V(G), are given
morphic to F; that is, t = 0.) In general, then, r =t + 1. Next let 5" in the accompanying figure.
56 Structure and symmetry of graphs

ii G'=GI~<> <> I
CHAPTER 3

G,(i~1'2"8)<> T'rees and connectivity

IG
1 <><>:
Connectedness of graphs is explored more fully in this chapter. Among
2.33 Show that no graph of order at least 3 can be reconstructed from
the connected graphs, the simplest yet most important are the trees.
exactly two of the sUbgraphs G - v, v E V(G).
Several results and concepts involving trees are presented. Connectivity
and edge-connectivity are described and related theorems by Menger are
stated. Other measures of connectedness are also given.

3.1 ELEMENTARY PROPERTIES OF TREES

A tree is an acyclic connected graph, while a forest is an acyclic graph. Thus

every component of a forest is a tree. There are several observations that
can be made regarding trees. First, by Theorem 2.4, it follows that every
edge of a tree T is a bridge; that is, every block of T is acyclic. Conversely,
if every edge of a connected graph G is a bridge, then G is a tree.
There is one tree of each of the orders I, 2 and 3; while there are two
trees of order 4, three trees of order 5, and six trees of order 6. Figure 3.1
shows all trees of order 6.
If u and v are any two nonadjacent vertices of a tree T, then T + uv
contains precisely one cycle C. If, in turn, e is any edge of C in T + uv,·
then the graph T + uv - e is once again a tree.
In a nontrivial tree T, it is immediate that the number of blocks to which
a vertex v of T belongs equals degv. So T - v is a forest with degv com-
ponents. If deg v = I, then T - v is a tree. Thus, every vertex of T that is
not an end-vertex belongs to at least two blocks and is necessarily a cut-
vertex. The next result is a basic property of trees and very useful when
using mathematical induction for proving theorems dealing with trees.

Theorem 3.1
Every nontriVial tree has at least two end-vertices.

Proof
Let T be a nontrivial tree and let P be a maximal path (a path not properly
contained in any other path of T). Suppose that P is a u-v path. Since P is a
58 Trees and connectivity Elementary properties of trees 59
G2 , • •• , Gk the components of G, where k~ 1. Furthermore, let Gi be an
(n;, mi) graph. Since each graph G; is a tree, ni = mi + 1. Hence,
k k
n- 1= m= L mi = L (ni - 1) = n - k

*
i=l i~l

so that k = 1 and G is connected. 0

The proof of Theorem 3.2 provides us with another result.

Corollary 3.3
A forest F of order n has n - keF) edges.
A spanning tree of a graph G is a spanning subgraph of G that is a tree.
Every connected graph G contains a spanning tree. If G is itself a tree,
then this observation is trivial. If G is not a tree, then a spanning tree T
of G can be obtained by removing cycle edges from G one at a time until
only bridges remain. If G has order n and size m, then since T has size
n -1, it is necessary to delete a total of m - (n - 1) = m - n + 1 edges to
Figure 3.1 The trees of order 6. produce T. This, of course, implies that m ~ n - 1, that is, every connected
graph of order n has at least n - 1 edges.
maximal path, neither u nor v is adjacent to any vertex not on P. Certainly,
Another characterization of trees is presented next.
u is adjacent to the vertex following it on P, and v is adjacent to the vertex
preceding it on P; however, neither u nor v is adjacent to any other vertex
of P since T contains no cycles. Therefore, deg u = deg v = 1. 0 Theorem 3.4
An (n, m) graph G is a tree if and only if G is connected and n = m + 1.
There are a number of ways to characterize trees (for example, see Berge
[B7, p. 152] and Harary [H7, p. 32]). Three of these are particularly useful.
Ptoof
Theorem 3.2 Let T be an (n, m) tree. By definition, T is connected and by Theorem 3.2,
n = m + 1. For the converse, we assumethat G is a connected (n, m) graph
An (n, m) graph G is a tree if and only if G is acyclic and n = m + 1. with n = m + 1. It suffices to show that G is acyclic. If G contains a cycle C
and e is an edge orc, then G - e is a connected graph of order n having
Proof n - 2 edges, which is impossible as we have observed. Therefore, G is
acyclic and is a tree. 0
If G is a tree, then it is acyclic by definition. To verify the equality
n = m + 1, we employ induction on n. For n = 1, the result (and graph) Hence, if G is an (n, m) graph, then any two of the properties (i) G is
is trivial. Assume, then, that the equality n = m + 1 holds for all (n, m) connected, (ii) G is acyclic and (iii) n = m + 1 characterize G as a tree.
trees with n(~l) vertices, and let T be a tree with n + 1 vertices. Let v There is yet another interesting characterization of trees that deserves
be an end-vertex of T. The graph T' = T - v is a tree of order n, and so mention.
T' has m = n - 1 edges by the inductive hypothesis. Since T has one
more edge than T', it follows that T has m + 1 = n edges. Since
n + 1 = (m + 1) + 1, the desired result follows. Theorem 3.5
Conversely, let G be an acyclic (n, m) graph with n = m + 1. To show A graph G is a tree ifand only ifevery two distinct vertices of G are connected by a
that G is a tree, we need only verify that G is conneCted. Denote by Gv unique path of G.
60 Trees and connectivity Elementary properties of trees 61
Proof 0---0----0--0--
If G is a tree, then certainly every two vertices u and v are connected by at

X>+
least one path. If u and v are connected by two different paths, then a cycle
of G is determined, producing a contradiction.
On the other hand, suppose that G is a graph for which every two
distinct vertices are connected by a unique path. This implies that G is
connected. If G has a cycle C containing vertices u and v, then u and v
are connected by at least two paths. This contradicts our hypothesis. Figure 3.2 Paths, stars, double stars and caterpillars.
Thus, G is acyclic and so G is a tree. 0
Proof
We now discuss some other properties of trees, particularly related to
Suppose that T has order n and size m. Then I:~=l ni = nand
the degrees of its vertices. It is very easy to determine whether a sequence
of positive integers is the degree sequence of a tree. k k
Lin; = 2m = 2n - 2 = 2 L ni - 2
Theorem 3.6 i=l ;=1
or
A sequence d1 ,dz ,... , dn ofn ;;. 2 positive integers is the degree sequence ofa tree k
;'=
of order n if and only if I: 1 d; = 2n - 2. L (i - 2)ni + 2 = O. (3.1)
i=l
Proof
Solving (3.1) for nl gives the desired result. 0
First, let T be a tree of order n and size m with degree sequence dv
dz,··· ,dn' ThenI:7=1 di = 2m = 2(n -1) = 2n - 2. We verify the converse There are some classes of trees with which we should be familiar. Of
by induction. For n = 2, the only sequence of two positive integers with course, the paths Pn and stars K 1 s are trees. A tree T is a double star if it
sum equal to2is 1, I, and this is the degree sequence ofthetreeKz . Assume contains exactly two vertices that are not end-vertices; necessarily, these
now that whenever a sequence of n - 1 ;;. 2 positive integers has the sum vertices are adjacent. A caterpillar is a tree T with the property that the
2(n - 2) = 2n - 4, then it is the degree sequence of a tree of order n -1. removal of the end-vertices of T results in a path. This path is referred
Let d1 , dz , ... ,dn be a sequence of n positive integers with to as the spine of the caterpillar. If the spine is trivial, the caterpillar is a
I: 7= 1di = 2n - 2. We show thatthis is the degree sequence of a tree. Sup- star; if the spine is Kz, then the caterpillar is a double star. Examples of a
pose that d1 ;;. dz ;;. •.. ;;. dn' Since each term d; is a positive integer and path, star, double star and caterpillar are shown in Figure 32.
I: 7= 1di = 2n - 2, it follows that 2::;;;; d1 ::;;;; n -1 and dn - 1 = dn = 1. Hence Knowledge of the properties of trees is often useful when attempting to
d1 - 1, dz, d3 , ••• , dn -1 is a sequence of n - 1 positive integers whose sum prove certain results about graphs in general. Because of the simplicity of
is 2n - 4. By the inductive hypothesis, then, there exists a tree T' of order the structure of trees, every graph ordinarily contains a number of trees as
n -1 with V(T') = {VI' VZ, . .. , vn~ I} such that deg Vl= d1 - 1 and subgraphs. Of course, every tree of order n or less is a subgraph of K n .
degvi = d; for 2 ::;;;; i ::;;;; n 1. Let T be the tree obtained from T' by A more general result is given next.
adding a new vertex V n and joining it to VI' The tree T then has the
degree sequence dltd z , ... ,dn' 0 Theorem 3.8
There is a simple connection between the number of end-vertices in a Let T be a tree of order k, and let G be a graph with 6(G) ;;. k -1. Then T is a
tree and the number of vertices of the various degrees exceeding 2. subgraph ofG. .

Theorem 3.7
Proof
Let T be a nontrivial tree with .6.(T) = k, and let ni be the number ofvertices of
degree i for i = 1,2, ... ,k. Then The proof is by induction on k. The result is obvious for k = 1 sjnce Xl is a
subgraph of every graph and for k =2 since K2 is a subgraph of every
nl = n3 + 2n4 +3ns +... + (k - 2)nk + 2. nonempty graph.
62 Trees and connectivity Elementary properties of trees 63
Assume for each tree T' of order k - 1, k ~ 3, and every graph H with
r5(H) ~ k:- 2 that T' is a subgraph of H. Let T be a tree of order k and let G
,A, ,L, ,~,
l' I, /2
be a graph with r5(G) ~ k - 1. We show that T ~ G.
Let v be an end-vertex of T and let u be the vertex of T adjacent with v.
The graph T - v is necessarily a tree of order k - 1. The graph G has
r5(G) ~ k - 1 > k - 2; thus by the inductive hypothesis, T - v is a sub-
graph of G. Let u' denote the vertex of G that corresponds to u. Since 1~2 1~2 ~2
3~ ~4 !430-~~~t
degG u' ~ k - 1 and T - v has order k - 1, the vertex u' is adjacent to a
vertex w that corresponds to no vertex of T - v. Therefore, T ~ G. 0

Theorem 3.8 can be restated to read that if G is a graph with r5( G) ~ k,

then G contains every tree of size k as a subgraph. This was extended to
forests by Brandt [B14].
2 2
lU lC In 1:J' 2 3/ 3h4

122IV12 IS2 IN2

3 43 43 43 4
Theorem 3.9
Let nand k be positive integers with n ~ 2k. Then every graph G of order n
with r5(G) ~ k contains every forest of size k without isolated vertices as a
subgraph.
3 43 43 43 4

IV2 11\/'2 1"72 1"v12

Although no convenient closed formula is known for the number of
nonisomorphic trees of order n, a formula does exist for the number
of distinct labeled trees (whose vertices are labeled from a fixed set of
cardinality n). For n = 3 and n = 4, the answer is sufficiently simple
that we can actually draw all three distinct trees of order 3 whose vertices
are labeled with elements of the set {I, 2, 3} and all 16 distinct trees of
order 4 whose vertices are labeled with elements of the set {1,2,3,4}.
3A4 3A4 3A4 3~4
These are shown in Figure 3.3. Figure 3.3 The labeled trees of orders 3 and 4.
In general, the number of distinct trees of order n whose vertices are
labeled with the same set of n labels is nn-Z. This result is due to
Cayley [C4]. There have been a number of proofs of Cayley's theorem.
The one that we describe here is d1,1e to Priifer [P6]. The proof consists

T"~ T"~
6 3
of showing the existence.of a one-to-one correspondence between the
trees of order n whose vertices are labeled with elements of the set
{1,2, ... ,n} and the sequences (called Priiter sequences) of length n - 2 TO=T'~
whose entries are from the set {I, 2, ... , n}. Since the number of such 7 ~. 2 5 8 4 72584 7258
sequences is nn-z, once the one-to-one correspondence has been estab-
lished, the proof is complete.
Before stating Cayley's theorem formally, we illustrate the technique
with an example. Consider the tree T of Figure 3.4 of order n = 8 whose
vertices are labeled with elements of {I, 2, ... ,8}. The end-vertex of To = T T 4: 0---<>--0-<:> T 5: 0-0-0
having the smallest label is found, its neighbor is the first term of the 7258 258
Priifer sequence for T, and this end-vertex is deleted, producing a new
tree T 1• The neighbor of the end-vertex of T1 having the smallest label is Priifer sequence for T: (1,8, 1,5,2,5)
the second term of the Priifer sequence for Ti this end-vertex is deleted, Figure 3.4 Determining the Priifer sequence of a tree.
64 Trees and connectivity Elementary properties of trees 65
producing the tree Tz. We continue this until we arrive at T,,_z = Kz. The 6 3 6 3
0
resulting sequence of length n - 2 is the Priifer sequence for T.
In the example just described, observe that every vertex v of T appears
(1.8.1,5,2,5)
1,2,3,4.5,6,7,8
0
df
(8.1.5.2.5)
1, 2. 4, 5, 6, 7, 8 df
in its Priifer sequence deg v-I times. This is true in general. Therefore, 0 0 0 0-0
0 0 0 0 0
no end-vertex of T appears in the Priifer sequence for T. So, if T is a tree of 7 2 5 8 4 7 2 5 8 4
order n and size m, then the number of terms in its Priifer sequence is
L (degv -1) = 2m - n = 2(n -1) - n = n - 2. 6 3 6 3
VEV(T) (5,2.5)

We now consider the converse question, that is, if (al,aZ,'" ,a,,-z) is a

sequence of length n - 2 such that each ai E {I, 2, ... , n}, then we construct
(1,5,2,5)
1,2.5,6,7,8 '\! 1,2,5,7,8

0);0-<0
0 0 0 0-0 0
a labeled tree T of order n such that the given sequence is the Profer 7 2 5 8 4 7 2 5 8 4
sequence for T. Suppose that we are given the sequence (1,8,1,5,2,5).
We determine the smallest element of the set {I, 2, ... ,8} not appearing
6 3 6 3
in this sequence. This element is 3. In T, we join 3 to 1 (the first element

Jo-<o
(2,5) (5)
of the sequence). The first term is deleted and· the reduced sequence 2,5,8
2,5,7,8
(8,1,5,2,5) is now considered. Also, the element 3 is deleted from the
set {1, 2, ... , 8}, and the smallest element of this set not appearing in
(8,1,5,2,5) is found, which is 4, and is joined to 8. This procedure is
0-<0);0-<0
72584 7 2 5 8 4

continued until two elements of the set remain. These two vertices are
joined and T is constructed. This is illustrated in Figure 3.5.
6 3
Since a step in the second procedure is simply the reverse of a step in

~
the first procedure, we have the desired one-to-one correspondence. We 5.8
have now described a proof of Cayley'S theorem.
7 2 584
Theorem 3.10
Figure 3.5 Constructing a tree with a given Priifer sequence.
2
There are nrl - distinct labeled trees a/order n.
As an illustration, we have

G-~ ~][~ -~)~[-: :],

Theorem 3.10 might be considered as a formula for determining the
number of distinct spanning trees in the labeled graph KIl" We now
consider the same question for graphs in general.
The next result, namely Theorem 3.11, is due to Kirchhoff [K4] and is
often referred to as the Matrix-Tree Theorem. The proof given here is based which has a determinant of -36. Writing IAI = det A, we see that

I~ -~I'I~ -~I+I~ ~I'I~ -~I+I-~ ~I'I~ ~1=-36.

on that given in Harary [H6].
This proof will employ a useful result of matrix theory. Let M and M' be
r x s and s x r matrices, respectively, with r ~ s. An r x r submatrix M i of
M is said to correspond to the r x r submatrix Mi of M' if the column Let Gbe a graph with V(G)={Vl,V2, ... ,V"J The degree matrix
numbers of M determining M i are the same as the row numbers of M' V(G) = [d;j] is the n x n matrix with dii = degvi and dij = 0 for i =f:. j. We
determining M~. Then now state the Matrix-Tree Theorem.

det(M·M') = L(detMi)(detMD, Theorem 3.11

where the sum is taken over all r x r submatrices M i of M, and where Mi is If G is a nontrivial labeled graph with adjacency matrix A and degree matrix V,
the r x r submatrix of M' corresponding to Mi' The numbers detM; and then the number of distinct spanning trees of G is the value of any cofactor of the
detMi are called the major determinants of M and M' , respectively. matrix V - A.
 Trees and connectivity Elementary properties of trees 67
66
corresponding major determinants of Ci and C;. However, corresponding
Proof
major determinants have the same value and their product is 1 if the
We note first that the sum of the entries of row i (column i) of A is deg Vi, defining columns correspond to a spanning tree of G and is a otherwise.
so every row (column) sum of 0 - A is O. It is a result of matrix theory that This completes the proof. 0
all cofactors of 0 ~ A have the same value.
Assume first that G is a disconnected graph of order n, and that GI
is a component of G with V( GI ) = {VI, V2, ... , v r }. Let 0' be the
(n - 1) x (n - 1) submatrix obtained by deleting from 0 - A the last EXERCISES 3.1
row and last column. Since the sum of the first r rows of 0' is the zero ,
vector with n - 1 entries, the rows of 0' are linearly dependent, implying 3.1 Draw all forests of order 6.
that det 0' = O. Hence one cofactor of 0 - A has value O. This is, of course, 3.2 Prove that a graph G is a forest if and only if every induced subgraph
the number of spanning trees of G. of G contains a vertex of degree at most 1.
We henceforth assume that G is a connected (n, m) graph; so, m ;;. n -1.
Let B denote the incidence matrix of G and in each column of B, replace 3.3 Characterize those graphs with the property that every connected
one of the two nonzero entries by -1. Denote the resulting matrix by subgraph is an induced subgraph.
C = [cij]' We now show that the product of C and its transpose C I is 3.4 A tree is called central if its center is KI and bicentral if its center is K2•
0- A. The (i,j) entry of CC I is Show that every tree is central or bicentral.
3.5 Let T be a tree of order 3 or more, and let T' be the tree obtained from
T by deleting its end-vertices.
(a) Show that diam T = diam T' + 2, rad T = rad T' + 1, and
which has the value deg Vi if i = j, the value -1 if ViVj E E( G), and a other- Cen(T) = Cen(T').
wise. Therefore, CCI = 0 - A. (b) Show that a tree T is central orbicentral (see Exercise 3.4) accord-
Consider a spanning subgraph H of G containing n - 1 edges. Let C' be ing to whether diam T = 2rad Tor diam T = 2rad T -1.
the (n -1) x (n -1) submatrixof C determined by the columns associated
with the edges of H and by all rows of C with one exception, say row k. 3.6 Let G be a nontrivial connected graph. Define the block-cut-vertex
We now determine IdetC'I. If H is not connected, then H has a compo- graph BC(G) of G as that graph whose vertices are the blocks and
nent HI not containing vk' The sum of the row vectors of C' corresponding cut-vertices of G and such that two vertices are adjacent if and
to the vertices of HI is the zero vector with n - 1 entries; hence det C' = O. only if one vertex is a block and the other isa cut-vertex belonging
Assume now that H is connected so that H is (by Theorem 3.4) a span- to the block.
ning tree of G. Let UI (:;fvk) be an end-vertex of H, and eI the edge incident (a) Show that BC(G) is a tree for every graph G.
with it. Next, let UZ(:;fVk) be an end-vertex of the tree H - UI and ez the (b) Show that BC(G) is a central tree.
edge of H - UI incident with Uz. We continue this procedure until finally (c) Prove Theorem 2.7.
only Vk remains. A matrix C" = [cij) can now be obtained by a permutation 3.7 (a) Show that every tree of order at least 3 contains a cut-vertex V
of the rows and columns of C' such that Ic~jl = 1 if and only if Ui and e· are such that every vertex adjacent to v, with at most one exception,
incident. From the manner in which C" was defined, any vertex ~i is is an end-vertex.
incident only with edges ej, where j ~ i. This, however, implies that C" is (b) Prove Theorem 2.8. (Hint: Consider the block-cut-vertex graph
lower triangular, and since leijl = 1 for all i, we conclude that Idet C"I = 1. (Exercise 3.6).)
However, the permutation of rows and columns of a matrix affects only
the sign of its determinant, implying that Idet C'I = Idet C"I = 1. 3.8 Let T be a tree of order n such that T:;f KI,n_l.Prove that T ~ f.
Since every cofactor of 0 - A has the same value, we evaluate only the 3.9 Determine the Priifer sequences of the trees in Figure 3.3.
ith principal cofactor; that is, the determinant of the matrix obtained by
deleting from D - A both row i and column i. Denote by Ci the matrix 3.10 (a) Which trees have constant Priifer sequences?
obtained from C by removing row i; so the aforementioned cofactor (b) Which trees have Priifer sequences each term of which is one of
equals det(CiG), which, by the remark preceding the statement of this two nl,lmbers?
theorem, implies that this number is the sum of the products of the (c) Which trees have Profer sequences with distinct terms?
68 Trees and connectivity Arboricity and vertex-arboricity 69.
3.11 Determine the labeled tree having Prufer sequence (4,5,7,2,1, a(C Il ) = 2. If n is even, a(KIl ) = n12; while if n is odd, a(KIl ) = (n + 1)/2. So,
1,6,6,7). in general, a(KIl ) = rn/21 Also, a(K"s) = 1 if r = 1 or s = 1, and a(Kr,s) = 2
3.12 Let G be the labeled graph below. otherwise. No formula is known in general, however, for the vertex-
arboricity of a graph although some bounds for this number exist. First,

'I~/r
it is clear that for every graph G of order n,
G:
~ r~l (3.2)

"~'3
a(G)

The bound (3.2) is not a particularly good one in general. In order to

(a) Use the Matrix-Tree Theorem to compute the number of distinct present a better bound, a new concept is introduced at this point.
labeled spanning trees of G. A graph G is called critical with respect to vertex-arboricity if
(b) Draw all the distinct labeled spanning trees of G. a(G - v) < a(G) for all vertices v of G. This is the first of several occasions
when a graph will be defined as critical with respect to a certain
3.13 (a) Let G = K4 with V(G) = {Vj, VZ, V3, V4}' Draw all spanning trees parameter. In order to avoid cumbersome phrases, we will simply use
of G in which V4 is an end-vertex. the term 'critical' when the parameter involved is clear by context. In
(b) Let v be a fixed vertex of G= KII • Determine the number of particular, a graph G that is critical with respect to vertex-arbericity
spanning trees of G in which v is an end-vertex. will be referred to in this section as a critical graph and, further, as a
3.14 Prove Theorem 3.10 as a corollary to Theorem 3.11. k-critical graph if a(G) = k. A k-critical graph necessarily has k ~ 2. The
complete graph K2k - J is k-critical while each cycle is 2-critical. It is
3.15 A graph G of order n with degree sequence dj , dz, ... ,dll , where not difficult to give examples of critical graphs; indeed, every graph G
d1 ~ dz ~ ... ~ dlt , is defined to be degree dominated if di ",;;; with a( G) = k ~ 2 contains an induced k-critical subgraph. In fact,
r(n - 1)/il for each integer i (1 ",;;; i ",;;; n). Prove that every tree is every induced subgraph G' of G of minimum order with a(G') = k is
degree dominated. k-critical.
Before presenting the aforementioned bound for a(G), we give another
result.

3.2 ARBORICITY AND VERTEX-ARBORICITY

Theorem 3.12
One of the most common problems in graph theory. deals with
If G is a graph having a(G) = k ~ 2 that is critical with respect to vertex-
decomposition of a graph into various subgraphs possessing some arboricity, then 6(G) ~. 2k - 2.
prescribed property. There are ordinarily two prOblems of this type,
one dealing with a decomposition of the vertex set and the other with
a decompoSition of the edge set. One such property that has been the Proof
subject of investigation is that of being acyclic, which we now
cohsider. Let G be a k-critical graph, k ~ 2, and suppose that G contains a vertex v of
For a graph G, it is always possible to partition V(G) into subsets Vi, degree 2k - 3 or less. Since G is k-critical, a(G - v) = k -1 and there is a
1 .:;; i",;;; k, such that each induced subgraph (Vi) is acyclic, that is, is a partition VI Vz, ... ,Vk - 1 of the vertex set of G - v such that each sub-
forest. One way to accomplish this is by selecting each subset Vi so that graph (Vi) is acyclic. Because degv ~ 2k 3, at least one of these subsets,
lVil ",; ; 2; however, the major problem is to partition V(c;) so that as few say Vi' contains at most one vertex adjacent with v in G. The subgraph
subsets as possible are involved. This suggests our next concept. The (Vi U {v}} is necessarily acyclic. Hence Vb V2 , .•• , Vi U {v}, ... , Vk-1 is a
vertex-arboricity a(G) of a graph G is the minimum number of subsets partition of the vertex set of G into k - 1 subsets, each of which induces an
into which V(G) can be partitioned so that each subset induces an acyclic acyclic subgraph. This contradicts the fact that a(G) = k. 0
subgraph. It is obvious that a(G) = 1 if and only if G is acyclic. For a few
classes of graphs, the vertex-arboricity is easily determined; For example, We are now in a position to present the desired upper bound [CKl].
70 Trees and connectivity Arboricity and vertex-arboricity 71
Theorem 3.13 Proof
For each graph G, LetG t be an induced subgraph of G haVing order nt and size mt. Thus Gt
a(G) ~ 1+ lmax;(G') j, can be decomposed into at (G) or fewer acyclic subgraphs, each of which
has size at most nt - 1. Then

where the maximum is taken over all induced 8ubgraphs G' of G. "(G)
u t ~
2mt
-,,;;.
2(nt - 1 )at (G) < 2at (G) .
nt nt
Proof Hence max8(G') < 2a t (G), where the maximum is taken over all induced
subgraphs G' of G. Therefore, 2a t (G) ~ 1 + max 8(G') which yields the
The result is obvious for acyclic graphs; thus, let G be a graph with desired result. 0
a(G) = k ~ 2. Now let H be an induced k-critical subgraph of G. Since H
itself is an induced subgraph of G, We now have a result that relates a(G) and at (G), also due to Burr [B161.
o(H) ~ maxo(G'), (3.3)
where the maximum is taken over all induced subgraphs G' of G. By Corollary 3.16
Theorem 3.12, o(H) ~ 2k - 2, so by (3.3), For every graph G, a(G) ~ at (G).
max 8(G') ~ 2k - 2 = 2a(G) - 2.
Proof
This inequality now produces the desired result. 0
By Theorems 3.13 and 3.15, we have
Since o(G') ~ ~(G) for each induced subgraph G' of G, we note the
following consequence of the preceding result. 2a(G) -2 ~ maxo(G') ~ 2a t (G) -1,
where the maximum is taken over all induced subgraphs G' of G. So
Corollary 3.14 a(G) ,,;;at(G) +!. Since a(G) and at (G) are integers, the result follows. 0
For every graph G, Unlike vertex-arboricity there is a formula for the arboricity of a graph,
a(G) l
~ 1 + ~~G) j. which was discovered by Nash-Williams [Nl]. The proof we give is due to
Hal.<i.mi, Mitchem and Schmeichel [HMSl].

We now turn to the second decomposition problem. The edge-arboricity, Theorem 3.17
or simply the arboricity at (G) of a nonempty graph G is the minimum
number of subsets into which E(G) can be partitioned so that each For every nonempty graph G,
subset induces an acyclic subgraph. As with vertex-arboricity, a non-
empty graph has arboricity 1 if and only if it is a forest. The following at (G)
IE(H)I
= max rIV(H)I- 1
1 '
lower bound for the arboricity of a graph was established by Burr [B161.
where the maximum is taken over all nontrivial induced subgraphs H of G.
Theorem 3.15
For every graph G, Proof
the statement of the theorem is equivalent to the following, and it is in
1+max8(G')1
at (G) ~
r 2 ' this form that we prove the result: The arboricity of a graph G is at most k
if and only if
where the maximum is taken over all induced subgraphs G' of G. IEW)I ~ k(IV(H)I-l)
72 Trees and connectivity Arboricity and vertex-arboricity 73
for all induced subgraphs H of G. Assume first that al (G) ~ k. Then there Eidx(e,_l) and so on, until, finally, in the last and (r + l)st step, eo is
is a partition of E(G) into subsets El , Ez,"" Et , where t ~ k and such that relocated to Eidx (e,), After the completion of these r + 1 steps, we now
(E i) is acyclic for all i (1 ~ i ~ t). have a partition El , Ez, ... ,Ek of E(G).
Let H be an induced subgraph of G, and define Xi = Ei n E(H) for Initially, each of the subsets Ei (1 ~ i ~ k) induces an acyclic subgraph,
i = 1,2, ... ,t. Hence E(H) is partitioned into the nonempty subsets namely the forest F;. We show now that after each of the r + 1steps, each
among Xl' Xz, ... ,Xt · Since each such subset Xi induces an acyclic sub- of the resulting sets Ei induces an acyclic graph. We verify this by finite
graph of H, it follows that IE((Xi))1 ~ 1V((Xi})1-1. Hence induction. The desired result is certainly true after the first step. Assume,
after the completion of the jth step (1 ~ j ~ r), that each of the resulting
IE(H)I ~ t(IV(H)I-1) ~ k(IV(H)I-1).
subsets E1 , Ez,' .. , Ek induces an acyclic subgraph. In the (j + l)st step, the
We now verify the sufficiency. Suppose, to the contrary, that there exist edgeer_j is added to the set Eidx (e'_i+l)' By the minimality of r, the original
graphs G' for which (a) IE(H') I ~ k(IV(H')1 -1) for all induced subgraphs (unique) path Pr - j + 1 in Fidx (e,_i+l) that connects the two incident vertices
H' of G' and (b) al(G') > k. Let G be such a graph of minimum size. Let of er_j and contains the edge er_j+l does not contain any edge ej with
eo be an edge of G. Because of the defining property of G, there exists a i > r - j + 1. Therefore, this path is not affected by the relocation of the
partition El , Ez, ... ,Ek of E(G - eo) such that Fi = (E j ) is a forest for edges ener-b'" ,er-j+z, However, when er-j+l is relocated, there is no
i = 1,2, ... ,k. longer a path in Fidx(e, _i+1) connecting the two incident vertices of er _) and
For each edge e of G -eo, we define the index idx(e) of e as the integer i after the U+ l)st step, Eidx(e'_i+') induces an acyclic subgraph, as does
(1 ~ i ~ k) such that e E Ei. In this context, an edge e of G is said to be each set Ei (1 ~ i ~ k). This completes the proof of the inductive step
reachable (from eo) if either (a) e = eo or (b) there exists a sequence eo, and, consequently, after the (r + l)st step, each of the sets El , Ez, ~ .. ,Ek
el,' .. ,er = e of edges of G such that for each i (1 ~ i ~ r), the two vertices in the partition of E(G) induces an acyclic subgraph. However, then,
incident with ei -I are connected by a path containing ej all of whose edges al (G) ~ k, contrary to assumption. Thus, as claimed, every reachable
belong to Fidx(ei)' An edge e of G is forest-connected (with respect to the edge of G is forest-connected.
partition El , Ez, ... ,Ek ) if the two vertices incident with e are connected Let J be the subgraph of G induced by the set of reachable edges of G.
by a path in Fi for every i (1 ~ i ~ k). Equivalently, e is forest-connected Necessarily, J is connected. For i = 1,2, ... ,k, define F; as that forest with
if e lies on a cycle in (E i U {e}} for every i (1 ~ i ~ k) with i =I idx(e). vertex set V(J) and edge set E(Fi) n E(J). Ne.xt we show that every forest
First, we claim that every reachable edge of G is forest-connected. By Fi (1 ~ i ~ k) is, in fact, a tree. Let j (1 ~ j ~ k) be a fixed integer, and let v
definition, eo is reachable. The edge eo is certainly forest-connected; for r
and w be distinct vertices of Fi. Since is connected, there exists a v-w path
otherwise. there is some j (1 ~ j ~ k) such that (Ej U {eo}) has no cycle P: v = Vo,V!>.'" Vt = win J. Let e; = Vi-lVi for i = 1,2, ... ,to Thus each
containing eo, which implies that al (G) ~ k, contrary to assumption. edge e; (1 ~ i ~ t) is reachable and so is forest-connected as well. Since
Suppose, to the contrary, that there exists a reachable edge e of G - eo each edge e; (1 ~ i ~ t) is forest-connected, the two incident vertices of
that is not forest-connected. Since e is reachable, there exists a sequence e; e;
are connected by a path Pi in Fj • Since each edge (1 ~ i ~ t) is reach-
eO,el,'" ,er = e of minimum length r(~l) such that the two vertices able, so is every edge of h Thus, Pi is a path of Fi. The walk consisting
incident with ei-l are connected by il path Pi in Fidx(ei) that contains ei' ofthe path PI followed by P Z ,P3 , ••• ,Ptis a v-wwalk. Thus, Fi contains a
Since e is not forest-connected, there exists an integer s (1.~. s,~ k) with v-w path and so Fj is connected. Therefore, Fi is a spanning tree of J; so
s =I idx(e) such that (Es U {e}) is acyclic. IE(Fi)1 = 1V(J)1-1 for each j (1 ~ j ~ k). Let H be the subgraph of G
From the given partition El , Ez, ... , Ek of E(G - eo), we now recursively induced by V(J). Then
construct a partition of E(G) into k subsets, which we will also refer to as
k
E1 , Ez,... ,Ek • For an edge f of G - eo, the integer idx( f) refers to the index
(subscript) of the set to which f originally belongs, regardless of what IE(H)I~IE(J)I = 1 + L IE(Fi) 1 = 1 +k(lV.(H)I-1),
j=l
subsets f may belong as the sets El , Ez, ... ,Ek are redefined. As the first
of r + 1 sequential steps, we relocate the edge e = er to Es' denoting the which produces a contradiction. 0
resulting set by Es . Of course, the set Eidx(e,) no longer contains ell but the
resulting set is still denoted by Eidx(e,). All other sets Ei (l~ i ~ k) are As a consequence of Theorem 3.17, it follows that
unchanged and are denoted as before.
For the second step, the edge er -I is relocated to the set Eidx(e,), and the and at (Kr s) = r5 1·
rr+5-1
sets E1 , Ez, ... , Ek are redefined. For the third step, we relocate er-z to ,
74 Trees and connectivity Connectivity and edge-connectivity 75
It is interesting to note that when n is even, Kn can be decomposed into n/2 positive integer n. Hence K( G) is the minimum number of vertices whose
spanning paths, as shown by Beineke [B2], and when n is odd, Kn can be removal results in a disconnected or trivial graph. It is an immediate con-
decomposed into (n + 1)/2 subgraphs, (n - 1) /2 of which are isomorphic sequence of the definition thata nontrivial graph has connectivity 0 if and
to Pn -] UK] and the other isomorphic to K t ,n-l' Decomposing a graph only if G is disconnected. Furthermore, a graph G has connectivity 1 if
into pairwise edge-disjoint acyclic subgraphs is a special case of the and only if G = Kz or G is a connected graph with cut-vertices; K( G) ~ 2
more general subject of· decomposition, which will be considered in if and only if G isnonseparab!e of order 3 or more.
Chapter 9. Connectivity has an edge analogue. An edge-cut in a graph G is a set X of
edges of G such that G - X is disconnected. If X is a minimal edge-cut of a
connected graph G, then, necessarily, G - X contains exactly two compo-
nents. Every nontrivial graph has an edge-cut. The edge-connectivity K] (G)
EXERCISES 3.2 of a graph G is the minimum cardinality of an edge-cut of G if G is non-
trivial, and ,.,.,] (K1 ) = O. So ""'1 (G) is the minimum number of edges whose
3.16 Are there graphs of order n other than Kn with a(Kn ) = n/2l? r removal from G results in a disconnected or trivial graph. Thus K1 (G) = 0
r
3.17 For all pairs k, n of positive integers with k :::;; n /21, give an example if and only if G is disconnected or trivial; while ,.,.,] (G) = 1 if and only if G
of a graph G of order n with a(G) = k. is connected and contains a bridge.
We now describe a basic relationship between vertex-cuts and edge-
3.18 What upper bounds for a(K1 ,s) are given by Theorem 3.13 and
cuts in graphs. The following result is due to Brualdi and Csima [BC4].
Corollary 3.14?
3.19 Let G be a k-critical graph with respect to vertex-arboricity (k ~ 3). Theorem 3.18
Prove that for each vertex v of G, the graph G - v is not (k - I)-critical
with respect to vertex-arboricity. Let G be a connected graph of order n ~ 3 that is not complete. For each edge-cut
X of G, there is a vertex-cut Uof G such that lUI:::;; IXI·
3.20 Show that the formula given for al (G) for a nonempty graph G in
Theorem 3.17 is, in fact, equivalent to the statement: The arboricity Proof
of G is at mosH if and only if IE(H)I :::;; k(\V(H)I-l) for all induced
subgraphs H of G. Assume, without loss of generality, that Xis a minimal edge-cut of G.
Then G - X is a disconnected graph containing exactly two components
3.21 Give an example of a graph G that has a nonempty induced sub- ~ and Gz. We consider two cases.
graph H such that
Case 1. Every vertex of G1 is adjacent to every vertex of Gz. Then IXI ~ n -1.

r
jE(G)1
\V(G)I-l <
1r IE(H) I
\V(H)I-l '
1 Since G is not complete, either G1 or Gz contains two nonadjacent vertices.
Say G] has nonadjacent vertices u and v. Then U = V(G) - {u, v}, which
thereby proving that, in gen~ral, al(G) =1= f1E(G)j/(\V(G)I-l)l has cardinality n - 2, has the desired property.
Determine a] (G) for this graph. Case 2. There are vertices u in G] and v in Gz that are not adjacent in G. For
each edge e in X, we select a vertex for U in the following way. If u is
incident with e, then choose the other vertex (in Gz) incident with e for
U; otherwise, select for U the vertex that is incident with e and belongs
3.3 CONNECTIVITY AND EDGE-CONNECTIVITY to G]. Now lUI:::;; IXI. Furthermore, u, v E V(G - U), but G - U contains
no u-v path, so U is a vertex-cut. 0
A vertex-cut in a graph G is a set U of vertices of G such that G - U is
disconnected. Every graph that is not complete has a vertex-cut. We are now in a position to present a result due to Whitney [W6].
Indeed, the set of all vertices distinct from two nonadjacent vertices is a
vertex-cut. Of course, the removal of any proper subset of vertices from a Theorem 3.19
complete graph leaves another complete graph. The vertex-connectivity or
simply the connectivity 1\:( G) of a graph G is the minimum cardinality of a For every graph G,
vertex-cut of G if G is not complete, and K(G) = n - 1 if G = Kn for some
76 Trees and connectivity Connectivity and edge-connectivity 77
C is not a complete graph. Hence there exists a vertex-cut U of C such that
lUI = t ,;,;; k - 1. The graph G - U is therefore disconnected of order n - t.
G: Let G 1 be a component of C - U of smallest order, say n1' Thus
n1 ,;,;; l(n - t)/2j. Let v be a vertex of Cj • Necessarily, v is adjacent in C
only to vertices of U or other vertices of C 1 - Hence
Figure 3.6 A graph G with /'i;(G) = 2, /'i;1(G) = 3 and 8(G) = 4. degv,;,;; t + (n1 -1) ,;,;; t + l(n - t)/2j -1

Proof = l(n+t-2)/2j';';; l(n+k-3)/2j,

Let v be a vertex of G such that degv = 6(G). The removal of the 6(G) contrary to the hypothesis. 0
edges of G incident with v results in a graph G' in which v is isolated,
so C/ is either disconnected or trivial. Therefore, Kl (G) ,;,;;. 6( G). A graph G is' k-edge-connected, k ~ 1, if I\;j (G) ~ k. Equivalently, G is k-
We now verify the other inequality. If G = K n for some positive integer edge-connected if the removal of fewer than k edges from C results in
n, then/'\,(Kn ) = /'\,1 (Kn ) = n - 1. Suppose next that G is not complete, and neither a disconnected graph nor a trivial graph. The class of k-edge-
let Xbe an edge-cut such that IXI = /'\,1 (G). By Theorem 3.18, there exists a connected graphs is characterized in the following simple but useful
vertex-cut U such that lUI,;,;; IXI. Thus theorem.
/'\,(G) ,;,;; lUI,;,;; IXI = 1\;1 (G). D'
Theorem 3.21
Figure 3.6 shows a graph G with I\;(G) = 2, /'\,1(G) = 3, and 6(G) = 4. It
can be shown (Exercise 3.31) that if a, b and c are positive integers with A nontrivial graph C is k-edge-connected if and only if there exists no nonempty
a ~ b ~ c, then there is a graph G with 1\;(G) = a, 1>:.1 (C) = band 6(C) = c. proper subset W of V(C) such that the number of edges joining Wand
A graph G is said to be k-connected, k ~ 1, if 1\;(C) ~ k. Thus C is V(G) - W is less than k.
I-connected if and only if G is nontrivial and connected; while C is 2-
connected if and only if G is nonseparable and has order at least 3. In
general, C isk-connected if and only if the removal of fewer than k vertices Proof
results in neither a disconnected nor trivial graph. First, assume that there exists no nonempty proper subset W of V(G) for
It is often the case that knowing that a graph is k-connected for some which the number of edges joining Wand V(C) - Wis less thankbutthat
specified positive integer k is as valuable as knowing the actual connec- G is not k-edge-connected. Since G is nontrivial, this implies that there
tivity of the graph. As would be expected, the higher the degrees of the exist t edges, 0 ,;,;; t ~ k, such that their deletion from G results in a dis-
vertices of a graph, the more likely it is that the graph has large connec- connected graph H. Let H j be a component of H. Since the number of
tivity. There are several sufficient conditions of this type. We present one edges joining V(H1 ) and V(C) - V(H1)is at most I, where 1< k, this is
of the simplest of these, originally presented in [CHI]. a contradiction.
Conversely, suppose that G is a k-edge-connected graph. If there should
Theorem 3.20 exist a subset W of V(C) such that j edges, j < k, join Wand V(G) - W,
then the deletion of these j edges produces a disconnected graph - again a
Let G be agraph oforder n ~ 2, and let k be an integer such that 1 ~ k ,;,;; n - 1. If
contradiction. The characterization now follows. 0
deg v ~ rn + ~ - 21 According' to Theorem 3.19, 1\;1(G)';';; 6(G) for every graph G. The
for every vertex v of C, then G is k-connected. following theorem of Plesnik [P5] gives a sufficient condition for equality
to hold.
Proof
'l}1eorem 3.22
Suppose that the theorem is false. Then there is a graph G satisfying the
hypothesis of the theorem such that G is noek~connected. Certainly, then, If G is a graph of diameter 2, then /'\,1 (G) = 6(G).
78 Trees and connectivity Menger's theorem 79
Proof 3.24 Let H = G + KIt where G is k-connected. Prove that H is (k + 1)-
connected.
Let 5 be a set of ~1 (G) edges of G whose removal disconnects G, and let HI
and Hz denote the components of G - 5, with orders nl and nz, respec- 3.25 Let G be a graph with degree sequence dl , dZ, ' •. , dnt where
tively. Without loss of generality, assume that nl ::;;; nz· d1 ~ dz ~ ... ~ dll • Define H = G1 + K1. Determine ~1(H).
Suppose that some vertex U of HI is adjacent in G to no vertex of Hz· 3.26 Show that every k-connected graph contains every tree of order k + 1
Then dc(u ,v) = 2 for each vertex v of Hz, and each vertex v of Hz is adja- as a s'ubgraph.
cent to some vertex of HI' Thus, either each vertex of HI is adjacent to
some vertex of Hz or each vertex of Hz is adjacent to some vertex of HI- 3.27 Let G be a noncomplete graph of order n and connectivity k such that
In either case, degv ~ (n + 2k - 2)/3 for every vertex v of G. Show that if 5 is a
vertex-cut of cardinality "'( G), then G - 5 has exactly two components.
~1 (G) = 151 ~ min{nll nz} = nl'
3.28 For a graph G of order n ~ 2, define the k-connectivity Kk(G) of G,
For each vertex U E V(H1 ), letdi(u) denote the number of vertices of
2 ::;;; k::;;; n, as the minimum number of vertices whose removal
Hi (i = 1,2) adjacent to U in G. Then
from G results in a graph with at least k components or a graph of
8(G) ::;;; degu = d1 (u) +dz(u) ::;;; nl -1 +dz(u) order less than k. (Note that KZ(G) = ~(G).) A graph G is defined to
be (t, k)-connected if K.k(G) ~ t. Let G be a graph of order n contain-
::;;; ~1 (G) -1 + dz(u) ::;;; 8(G) 1 + dz(u). (3.4)
ing a set of at least k pairwise nonadjacent vertices. Show that if
Since 8(G) ~ "'I (G), it follows from (3.4) that dz(u) ~ 1 for each vertex U of
ri+(k-l)(t-2)1
HI' Let V(H1 ) = {Ul, Uz , ••• , Uk}' where k = nl' Then
k k-l
d ege v ~
r k

"'I (G) = 151 = L

dZ(Ui) =
i=1
L
+dZ(Uk) ~ (k -1) + dZ(Uk)
i=1
for every v E V (G), then G is (t 1 k)-connected.
3.29 Prove Corollary 3.23.
= nl -1 + dZ(Uk). (3.5) 3.30 Let G be a graph of diameter 2. Show that if S is a set of "'I (G) edges
Again since 8(G) ~ "'I (G), it follows from (3.4) and (3.5) that whose removal disconnects G, then at least one of the components of
G - S is isomorphic to K1 or K 8(G)'
ni -1 + dZ(Uk) ~ 8(G) ~ "'I (G) ~ nl - 1 + dZ(Uk)'
3.31 Let a, band c be positive integers with a ::;;; b ::;;; c. Prove that there
Thus, "'I (G) = 8(G). 0 exists a graph G with K.(G) = a, K.l(G) = band 8(G) = c.
3.32 Verify that Theorem 3.20 is best possible by showing that for each
Corollary 3.23
positive integer k, there exists a graph G of order n( ~ k + 1) such that
If G is a graph of order n ~ 2 such that for all distinct nonadjacent vertices u 8(G) = r(n + k - 3)/21 and K.(G) < k.
and v,
3.33 Verify that Theorem 3.22 is best possible by finding an infinite class
deg U + deg v ~n - 1, of graphs G of diameter 3 fOfwhich ~1(G) =1= 8(G).
then "'1 (G) = 8(G). 3.34 The connection number c(G) of a connected graph G of order n ~ 2 is
the smallest integer k with 2 ::;;; k ~ n such that every induced sub-
graph of order kinG is connected. State and prove a theorem that
EXERCISES 3.3 gives a relation between K.(G) and c( G) for a graph G of order n.

3.22 Determine th~ connectivity and edge-connectivity of each complete

k-partite graph.
3.4 MENGER'S THEOREM
3.23 Let VI 1 vz, ... 1 Vk be k distinct vertices of a k-connected graph G. Let H
be the graph formed from G by adding a new vertex of degree k that A nontrivial graph G is connected (or, equivalently, I-connected) if
is adjacent to each of VI, Vz, .•• , Vb Show that ~(H) = k. between every two distinct vertices of G there exists at least one path.
80 Trees and connectivity Menger's theorem 81
1. For every two adjacent vertices VI and V2 of H, where neither VI nor V2 is U
or v, there exists a set U of t - 1 vertices of H such that U U {Vi}' i = 1,2,
separates Uand v.
To see this, let e = vIV2 and observe that St(u, v) is false for H - e.
G: u ~~------:::;,oc v However, we claim that 5t - 1 (u, v) is true for H - e. If not, there exists
a set W of ver.tices that separates u and v in H - e, where IWI ,:;; t - 2.
Then W separates u and v in both H - VI and H - V2;SO W U {vJ,
i = 1,2, separates u and v in H, which is impossible. Thus, as claimed,
W3 5 t - 1 (u, v) is true in H - e. So there exists a set U that separates u
Figure 3.7 A graph illustrating Menger's theorem. and v in H - e, where lUI = t -1. However, then Uu {Vi}' i = 1,2,
separates u and v in H.
2. For each vertex w(fu, v) in H, not both uw and vw are edges of H.
This fact can be generalized in many ways, most of which involve, either
Suppose that this is not true. Then 5/- 1 (u, v) is true for H - w.
directly or indirectly, a theorem due to Menger [M6]. In this section, we
However, then, H - W contains t - 1 internally disjoint u-v
discuss the major ones of these, beginning with Dirac's proof [08] of
paths. So H contains t internally disjoint u-v paths, which is a
Menger's theorem itself.
contradiction.
A set 5 of vertices (or edges) of a graph G is said to separate two vertices
3. If W = {WI, W2, ... ,w/} is a set of vertices that separates u and v in H,
u and v of G if the removal of the elements of 5 from G produces a dis-
then either UWi E E(H) for all i (1 ,:;; i :::;; t) or vWi E E(H) for all i
connected graph in which u and v lie in different components. Certainly,
(1':;; i:::;;t).
then,S is a vertex-cut (edge-cut) of G.
Define H u as the subgraph induced by the edges on all U-Wi paths
In the graph G of Figure 3.7, there is a set 5 = {WI, W2, W3} of vertices of
in H that contain only one vertex of W. Define Hv similarly. Observe
G that separate the vertices u and v. No set with fewer than three vertices
that V(Hu ) n V(Hv ) = W. Suppose that the above statement is not
separates u and v. As is guaranteed by Menger's theorem [M6], stated
true. Then both H u and H v have order at least t + 2. Define H~ to
next, there are three internally disjoint u-v paths inG.
consist of H u , a new vertex v· together with all edges V·Wi. Also,
define II;; to consist of Hv , a new vertex u· together with all edges
Theorem 3.24 U'Wi' Observe that II;; and H; have smaller order than H. So
St(u,v') is true in. II;; and 5/(u',V) is true in II;;. Therefore,
Let ~ and v be nonadjacent vertices in a graph G. Then the minimum number of there exist t internally disjoint u-v' paths in II;; and t internally
vertIces that separate u and v is equal to the maximum number o/internally disjoint u'-v paths in H;. These 2t paths produce t internally disjoint
disjoint u-v paths in G. u-v paths in H, a contradiction.
Let P be a u-v path of length d(u, v). Then d(u, v)·~ 3 by (2). Thus we
Proof may write P: u, Ul, U2, ... , v where ul, u2 f V. By (1), there exists a set U of
First, if u and v lie in different components of G, then the result is true; so t - 1 vertices such that both U U {ud and U U {U2} separate u and v. In
we may assume that the graphs under consideration are connected. If the particular, U U {Ul} separates u and v. So every vertex of U is adjacent to
~imum number of vertices that separate u and v is k( ~ 1), then the max-
u. Consider U U {U2}' No vertex of U is adjacent to v, and so u is adjacent
Imum number of internally. disjoint u-v paths in G is at most k. Thus, if to U2' which is impossible. 0
k = 1, the result is true (since we are assuming that G is connected).
Denote by Sk(U, v) the statement that the minimum number of vertices With the aid of Menger's theorem, it is now possible to present
that separate u and v is k. Whitney's characterization [W6] of k-connected graphs.
. Suppose that the theorem is false. Then there exists a smallest positive
mteger t(~2) such that 5/(u, v) is true in some graph G but the maximum
number of internally disjoint u-v paths is less than t. Among all such Theorem 3.25
graphs G of smallest order, let H be one of minimum size. A nontrivial graph G is k-connected if and only if for each pair u, v of distinct
We now establish three properties of the graph H. vertices there are at least k internally disjoint u-v paths in G.
82 Trees and connectivity Menger' 5 theorem 83
Proof Proof
Assume that G is a k-connected graph. Suppose, to the contrary, that there For k = 2, the result follows from Theorem 2.5; hence, we assume that
are two vertices u and v such that the maximum number of internally k ): 3. Let W be a set of k vertices of G. Among all cycles of G, let C be a
disjoint u-v paths in G is t, where t < k. If uv (j. E(G) then, by Theorem cycle containing a maximum number, say t, of vertices of W. We observe
3.24, K,(G) ~ t < k, which is contrary to hypothesis. If uv E E(G), then that t): 2. We wish to show that t = k. Assume, to the contrary, that
the maximum number of internally disjoint u~v paths in G - uv is t < k. Let W be a vertex of W such that W does not lie on C.
t - 1 < k -1; hence K,(G - uv) < k-l. Therefore, there exists a set U of Necessarily, C contains at least t + 1 vertices; for if this were not the
fewer than k -1 vertices such that G - uv - U is a disconnected graph. case, then the vertices of C could be labeled so that C: W1, Wz, ... , W(, W1,
Therefore, at least one of G-(UU{u}) and G-(UU{v}) is discon- where Wi E W for 1 ~ i ~ t. By Theorem 3.26, there exist internally
nected, implying that K(G) < k. This also produces a contradiction. disjoint W-Wi paths Qi, 1 ~ i ~ t. Replacing the edge W1 Wz on C by the
Conversely, suppose that G is a nontrivial graph that is not k-connected wl-wZ path determined by Q1 and Qz, we obtain a cycle containing at
but in which every pair of distinct vertices are connected by at least k least t + 1 vertices of W, which is a contradiction. Therefore, C contains
internally disjoint paths. Certainly, G is not complete. at least t + 1 vertices.
Since G is not k-connected, K,(G) < k. Let W be a set of K,(G) vertices of G Thus we may assume that C contains vertices W1, W2, ... ,W(, W( + 11
such that G - W is disconnected, and let u and v be in different com- Wi E W for 1 ~ i ~ t and W(+l (j. W. Since k): t + I, we may apply
ponents of G -'- W. The vertices u and v are necessarily nonadjacent; how- Theorem 3.26 again to conclude that there exist t + 1 internally disjoint
ever, by hypothesis, there are at least k internally disjoint u-v paths. By W-Wi paths Pi (1 ~ i ~ t + 1). For,each i = 1,2, ... , t + I, let Vi be the first
Theorem 3.24, u and v cannot be separated by fewer than k vertices, so a P:
vertex on Pi that belongs to C (possibly Vi = Wi) and let denote the w-vi
contradiction arises. 0 subpath of Pi' Since C contains exactly t vertices of W, there are distinct
integers 5 and t, 1 ~ 5, t ~ t + I, such that one of the two Vs-VI paths, say
With the aid of Theorem 3.25, the following result can now be estab- P, determined by C contains no interior vertex belonging to W. Replacing
lished rather easily. P by the vs-v! path determined by Ps and Pi, we obtain a cycle of G
containing at least t + 1 vertices of W. This contradiction gives the
desired result that t = k. 0
Theorem 3.26
If G is a k-connected graph and V, V1 , Vz, •.. , Vk are k + 1 distinct vertices of G, Both Theorems 3.24 and 3.25 have 'edge' analogues; the analogue to
then there exist internally disjoint V-Vi paths (1 ~ i ~ k). Theorem 3.24 was proved in Elias, Feinstein and Shannon [EFS1] and
Ford and Fulkerson [FFll. It is not surprising that the edge analogue of
Proof Theorem 3.24 can be proved in a manner that bears a striking similarity to
the proof of Theorem 3.24.
Construct a new graph H from G by adding a new vertex u to G together
with the edges UVi, i = 1,2, ... , k. Since Gis k-connected, H is k-connected
(Exercise 3.23). By Theorem 3.25, there exist k internally disjoint u-v paths Theorem 3.28
in H. The restriction of these paths to G yields the desired internally dis-
If u and v are distinct vertices of a graph G, then the maximum number of edge-
joint V-Vi paths. 0
disjoint u-v paths in G equals the minimum number ofedges of G that separate u
and v.
One of the interesting properties of 2-connected graphs is that every
two 'vertices of such graphs lie on a common cycle. (This is a direct
consequence of Theorem 2.5.) There is a generalization of this fact to k- Proof
connected graphs by Dirac [D5J.
We actually prove a stronger result here by allowing G to be a multi-
graph.
Theorem 3.27 If u and V are vertices in different components of a multigraph G, then
Let G be ak-connected graph, k ~ 2. Then every k vertice5 of G lie on a common the theorem is true. Thus 1 without loss of generality, we may assume that
cycle of G. the mu1tigraphs under consideration are connected. If the minimum
84 Trees and connectivity Menger's theorem 85
number of edges that separate u and v is k, where k> 1, then the
maximum number of edge-disjoint u-v paths is at most k. Thus, the EXERCISES 3.4
result is true if k = 1.
For vertices u and v of a multigraph G, let 5k (u, v) denote the statement 3.35 Prove that a graph G of order n > k + 1 > 3 is k-connected if and only
that the minimum number of edges that separate u and v is k. if for each .set 5 of k distinct vertices of G and for each two-vertex
If the theorem is not true, then there exists a positive integer 1(>2) for subset T of 5, there is a cycle of G that contains the vertices of T
which there are multigraphs G containing vertices u and v such that and avoids the vertices of 5 - T.
5Au, v) is true, but there is no set of I edge-disjoint u-v paths. Among
3.36 Prove that a graph G of order n > 2k is k-connected if and only if for
all such multigraphs G, let F denote one of minimum size.
every two disjoint sets VI and V 2 of k vertices each, there exist k
If every u-v path of F has length 1 or 2, then since the minimum
disjoint paths connecting VI and V 2 .
number of edges of F that separate u and v is I, it follows that there
are I edge-disjoint u-v paths in F, producing a contradiction. Thus F 3.37 LetG be a k-connected graph and let v be a vertex of G. For a positive
contains at least one u-v path P of length 3 or more. Let el be an integer t, define Gt to be the graph obtained from G by adding t new
edge of P incident with neither u nor v. Then for F - ev the statement vertices Ul, U2,' .. , Ut and all edges of the form UiW, where 1 ~ i ~ t
5t (u, v) is false but 5t - l (u, v) is true. This implies that el belongs to a set and for which vw E E(G). Show that Gt is k-connected.
cif I edges of F that separate u and v, say {el,e2"" ,eJ. We now sub-
3.38 Show that if G is a k-connected graph with nonempty disjoint subsets
divide each cif the edges ei, 1 ~ i ~ I, that is, let ei = UiVi, replace each ei
51 and 52 of V(G), then there exist k internally disjoint paths
by a new vertex Wi, and add the 21 edges UiWi and WiVj. The vertices Wi
Pll P 2 , . .. , P k such that Pi is a Ui-Vi path, where Ui E 51 and Vi E 52,
are now identified, producing a new vertex wand a new multigraph H.
The vertex W in H is a cut-vertex, and every u-v path of H contains w. for i = 1,2, ... , k, and 151 n V(P i ) I = 152 n V(Pi)1 = 1.
Denote by H u the submultigraph of H determined by all u-w paths of Hi 3.39 Let G be a k-connected graph, k> 3, and let v., VI' V2, ... ,Vk-l be k
the submultigraph H v is defined similarly. Each of the m:ultigraphs H u vertices of G. Show that G has a cycle C containing all of
and Hv has fewer edges than does F (since el was chosen to be an edge Vll V2, ... ,Vk-l but not v and k - 1 internally disjoint v - Vi paths Pi
of a u-v path in F incident with neither u nor v). Also, the minimum (1 ~ i ~ k - 1) such that for each i, the vertex Vi is the only vertex of
number of edges separating u and W in H u is I, and the minimum Pi on C.
number of edges separating v and W in H v is I. Thus, the multigraph H u
3.40 Prove Theorem 3.29.
satisfies 5t (u, wyand the multigraphHv satisfies 5t (w, v). This implies that
Hu contains a set of I edge-disjoint u-w paths and Hv contains a set of I 3.41 Prove or disprove: If G is a k-edge-connected graph and
edge-disjoint w-v paths. For each i = 1,2, ... ,I, a u-w path and w-v v, VI' V2, , Vk are k + 1 distinct vertices of G, then for
path can be paired off to produce a u-v path in H containing the two i = 1,2, ,k, there exist V-Vi paths Pi such that each path Pi contaiJ.'s
edges UiW and WVi' These I u-v paths of H are edge-disjoint. The process exactly one vertex of {VI, V2, •.• , Vk}, namely Vi, and for i =I j, Pi and Pj
of subdividing the edges ei = UiVi of F and identifying the vertices Wi to are edge-disjoint.
obtain W can now be reversed to produce I edge-disjoint u-v paths in F.
This, however, produces a contradiction. 3.42 Prove or disprove: If G is a k-edge-connected graph with nonempty
Since the theorem has been proved for multigraphs G, its validity disjoint subsets 51 and 52 of V(G), then there exist kedge-disjoint
follows in the case where G is a graph. 0 . paths Pl,P2 , ... ,Pk such that Pi is a Ui-Vi path, where Ui E 51 and
Vi E52' for i = 1,2, ... , k, and 151 n V(Pi)1 = 152 n V(Pi)1 = 1.
With the aid of Theorem 3.28, it is now possible to present an edge 3.43 Show that I\;(Qn) = I\;I(Qn) "= n for all positive integers n.
analogue of Theorem 3.25.
3.44 Assume that G is a graph in the proof of Theorem 3.28. Does the proof
go through? If not, where does it fail?
Theorem 3.29
3.45 Let G be a graph of order n with I\;(G) > 1. Prove that
A nontrivial graph G is k-edge-connected if and only if for every two distinct
vertices u and v of G, there exist at least kedge-disjoint u-v paths in G. n> I\;(G)[diamG -lJ +2.
86 Trees and connectivity Vulnerability of graphs 87
3.5 VULNERABILITY OF GRAPHS The independence number is related to toughness in the sense that
among all the vertex-cuts 5 of a noncomplete graph G, the maximum
The connectivity of a graph is one measure of how strongly connected the value of keG - 5) is tJ(G), so for every vertex-cut S of G, we have that
graph is, that is, the smaller the connectivity the more vulnerable a graph ;;;(G) ~ ISland keG - 5) ~ tJ(G). This leads us to bounds for the toughness
is. There are other measures of vulnerability. We will discuss some of of a graph, a result due to Chvatal [C7].
these in the current section.
If G is a noncomplete graph and t is a nonnegative real number such
that t ~ ISI/k(G - S) for every vertex-cut S of G, then G is defined to be Theorem 3.30
t-taugh. If G is a t-tough graph and s is a nonnegative real number such For every noncomplete graph G,
that s < t, then G is also s-tough. The maximum real number t for which a
graph G is t-tough is called the toughness of G and is denoted by t( G). Since ;;;(G) s:. t(G) s:. ;;;( G)
complete graphs do not contain vertex-cuts, this definition does not apply tJ(G) ~ ~ 2 .
to such graphs. Consequently, we define t(Kn ) = +00 for every positive
integer n. Certainly, the toughness of a noncomplete graph is a rational Proof
number. Also t(G) = 0 if and-only if G is disconnected. Indeed, it follows
that if G is a noncomplete graph, then According to (3.6),
t(G) = minISI/k(G - 5), (3.6) . 151 ;;;(G)
t(G) = mm keG _ 5) ~ tJ(G)'
where the minimum is taken over all vertex-cuts 5 of G.
For the graph G of Figure 3.8, S1 = {u, V, w}, S2 = {w}, and S3 = {u, v} Let 5' be a vertex-cut with 15'1 = ;;;(G). Thus keG - 5') ~ 2, so
are three (of many) vertex-cuts. Observe that IS11/k(G - 51) = i, . 151 15'1 ;;;(G)
15 2 1/k(G - 52) = ~ and 153 1/k(G - 53) =~. There is no vertex-cut 5 of G
t(G) = mm keG _ 5) ~ keG _ 5') ~ -z-· 0
with 151/k(G - 5) < ~; thus t(G) = ~.
The toughness of a graph G is then a measure of how tightly the sub-
graphs of G are held together. Thus the smaller the toughness the more Let F be a graph. A graph G is F-free if G contains no induced subgraph
vulnerable the graph is. A I-tough graph, for example, has the property isomorphic to F. Thus a K2-free graph is empty. In this context, a graph of
that breaking the graph into k components (if this is possible) requires the particular interest is K1•3 • A K1•3 -free graph is also referred to as a claw-free
removal of at least k vertices; while breaking. a Z-tough graph into k graph. The following result by Matthews and Sumner [MSI] provides a
components requires the removal of at least Zk vertices. cla:ss of graphs for which the upper bound given in Theorem 3.30 becomes
A parameter that plays an important role in the study of toughness is an equality.
the independence number. Two vertices that are not adjacent in a graph G
are said to be independent. A set 5 of vertices is independent if every two
vertices of 5 are independent. The vertex independence number or simply the Theorem 3.31
independence number tJ(G) of a graph G is the maximum cardinality among If G is a noncamplete claw-free graph, then t(G) = !;;;(G).
the independent sets of vertices of G. For example, tJ(Kr•s ) = max{r,s},
tJ(C n ) = Ln/ZJ and tJ(Kn ) = 1.
Proof

a If G is disconnected, then t(G) = ;;;(G) = 0 and the result follows. So we

assume that· ;;;(G) = r ~ 1. Let 5 be a vertex-cut such that t(G) =
G: b ISI/k(G - 5). Suppose that keG - 5) = k and that G}, G2, .. ·, Gk are the
~l----Oy components of G - S.
c Let Uj E V(G j ) and Uj E V(Gj ), where i =I- j. Since G is r-connected, it
follows by Theorem 3.25 that G contains at least r internally disjoint
d
Ui-Uj paths. Each of these paths contains a vertex of S. Consequently,
Figure 3.8 A graph of toughness ~ . there are at least r edges joining the vertices of Sand tfie vertices of Gj
88 Trees and connectivity Vulnerability of graphs 89
for each i (1 ~ i ~ k) such that no two of these edges are incident with the In Theorem 3.30 we saw a relationship between the toughness of a
same vertex of 5. graph, the connectivity and the independence number. Theorem 3.32
Hence there is a set Xcontaining at least kr edges between 5 and G - 5 and Corollary 3.33 relate integrity to various graphical parameters. The
such that any two edges incident with a vertex of 5 are incident with following result is due to Goddard and Swart [GSl].
vertices in distinct components of G - S. However, since G is claw-free,
no vertex of 5 is joined to vertices in three components of G - 5. Theorem 3.32
Therefore,
Let G be a graph of order n ~ 2 with degree sequence d 1 , dz , .. . , d ll , where
kr = IXI ~ 2151 = 2kt(G), d1 ~ dz ~ .... ~ dl1' Then
so kr ~ 2kt(G). Thus t(G) ~ r/2 = !I\;(G). By Theorem 3.30, t(G) = (a) I(G) ~ min {max{t,dt+ I}};
!I\;(G). 0 1~ t~ n

In defining the toughness of a graph we were in some sense fine-tuning

the idea of connectivity. For example, if a graph G is 2-connected, then the
(b) I(G) ~ fn~(~\G)l+I\;(G);
removal of one vertex from G does not result in a disconnected graph. The
removal of two vertices, however, may not only result in a disconnected (c) I(G) ~ f2Jnt(G) - t(G)l, if G =J K".
graph but in fact may result in a graph with many components. If,
however, we know that G is I-tough, then not only is G 2-connected but Proof
also the removal of any two vertices of G can result in a graph with at most
two components. Another measure of vulnerability that reflects overall (a) Let 5 be a proper subset of V (G) with 151 = s. Then N(G - 5) ~ I and
vulnerability rather than local weaknesses in a graph is the integrity of N(G- 5) ~ .0.(G- 5) + 1
a graph.
The integrity I(G) of a nontrivial graph G is defined as ~ max
VEV(G)-S
degG v - S +1
I(G) = min
SCV(G)
{lSI + N(G - S)},
~ dS + 1 - S + 1.
where N(G - 5) is the maximum order of a component of G - 5. Therefore,
Each of G1 = K1,n-l and Gz =K1 + (K1 UKn _ z) has order n and con- 151 + N(G - 5) ~ max{s + 1, dS + 1 + I}.
nectivity Ii however, I(GI ) = 2 while I(G z ) = n - 1. The high integrity Thus
of Gz reflects the fact that although the removal of a single vertex may
disconnect Gz, in doing so all but one of the remaining vertices lie in I(G) = min {151 + N(G - 5)}
ScV(G)
the same component.
Figure 3.9 gives the values of I\;(G), t(G) and I(G) for three classes of ~min {max{t,d t +l}}.
1", t '" n
graphs.
(b) If G is complete, then I(G) = n = f(n -1\;(G))/,B(G)l + I\;(G).
Assume, then, that G is not complete, and let 5 be a proper subset of
G I\;(G) t(G) I(G) V(G) for which I(G) = 151 +N(G - 5). Then 5 is a vertex-cut and so
151 ~ I\;(G) (Exercise 3.56).
K Il n-1 00 n
Now, keG - 5) ~ ,B(G - 5) ~ ,B(G), so
k
~,Il-k k k+1 n - 151 . n - 151
n-k N(G S) ~ keG _ S) ~ f3(G) .
(k ~ n/2)
K] + (Kk u KIl-k-d 1 1
2: n-k Therefore,
(k ~ (n - 1)/2)
151 + N(G - S) ~ f(n - I\;(G))/f3(G)l + I\;(G) , (3.7)
Figure 3.9 Connectivity, toughness and integrity. and the desired result follows.
90 Trees and connectivity Vulnerability of graphs 91
(c) If t(G) = 0, then clearly I(G) ;;:. r2jnt(G) - t(G)l Assume, then, that 3.52 Show that for every nonnegative rational number r, there exists a
t(G) > 0, that is, G is connected. Let 5 be a proper subset of V(G) for which graph G with t(G) = r.
s = 151 and I(G) = 151 + N(G - 5). Then 5 is a vertex-cut of G and so
3.53 Let G be a noncomplete graph of order n. Show that if 5 is a vertex-
151 cut of cardinality I\",(G), then G - 5 contains a component of order at
t(G) ~ k(G - 5)" least r(n - K(G))/(3(G)l
Therefore, 3.54 We have seen thatthe vertex-arboricity a(G) ~ r n/21 for every graph
G of order n. Show that a(G) ~ r(n - (3(G) + 1)/21 for every graph G
n -151 n- s (n)
N(G - 5) ;;:. k(G _ S) ;;:'s/t(G) = t(G) 1 . s- of order n.
3.55 Characterize the K1,z-free graphs.
Hence
3.56 Show that if G is a noncomplete graph of order nand 5 is a proper
I(G) ;;:. min
l';s,;n-l
{s + t(G) (~-1)}.
S
subsetofV(G) for which I(G) = lSI +N(G - S), thenSis a vertex-cut
ofG.
Setting f(5) = 5 + t(G)(n/5 -I), we observe that f is minimized at s = 3.57 Justify inequality (3.7).
jnt(G), and the result follows. 0
3.58 Prove Corollary 3.33.
Corollary 3.33 3.59 Show that for positive integers rand 5, I(Kr,s) = 1 + rnin{r, 5}.
If G is a graph, then I(G) ~ 1 + 6(G). 3.60 (a) Show that if r vertices are removed from the path P n' one of the
resulting components contains 1
at least (n - r)/(r + 1) vertices.
Other measures of graph vulnerability include the binding number and (b) Show that I(P n ) = r2(n + lYQ - 2.
the edge integrity of a graph. Excellent sources for more information in
this area are found in the surveys by Barefoot, Entringer and Swart
[BES1], Bagga, Beineke, Goddard, Lipman and Pippert [BBGLPl, and
Bagga, Beineke, Lipman and Pippert [BBLP1].

EXERCISES 3.5

3.46 Determine the toughness and the integrity of the complete tripartite
graph Kr,r,r (r ;;:. 2).
3.47 Show that if H is a spanning subgraph of a noncomplete graph G,
then t(H) ~ t(G) and I(H) ~ I(G).
3.48 Show that if G is a noncomplete graph of order n, then
t(G) ~ (n - (3(G))/(3(G).
3.49 Show that the order of every noncomplete connected graph G is at
least (3(G)(l + t(G)).
3.50 Show that for positive integers rand 5 with r + s ;;:. 3, t(Kr,s) =
min{r, 5} /max{r, s}.
3.51 Show that every I-tough graph is 2-connected.
 Eulerian graphs and digraphs 93
CHAPTER 4

Eulerian and hamiltonian

graphs and digraphs
Figure 4.1 The bridges of Konigsberg.

Also, since five bridges lead to region A, Euler saw that the letter A must
appear in the sequence a total of three times - twice to ind~cate an
In this chapter we investigate graphs and digraphs with special circuits entrance to and exit from land area A, and once to denote either an
and cycles. In particular, we will be concerned with circuits containing entrance to A or exit from A. Similarly, each of the letters B, C and D
every edge of a graph and cycles containing every vertex. must appear in the sequence twice. However, this implies that nine
terms are needed in the sequence, an impossibility; hence the desired
route through Konigsberg is also impossible.
4.1 EULERIAN GRAPHS AND DIGRAPHS The Konigsberg Bridge Problem has graphical overtones in many ways;
indeed, even Euler's representation of a route through Konigsberg is
In this section we discuss those trails and circuits in graphs and digraphs essentially that of a walk in a graph. If each land region of Konigsberg
which are historically the most famous. is represented by a vertex and two vertices are joined by a number of
It is difficult to say just when and where graphs originated, but there is edges equal to the number of bridges joining corresponding land areas,
justification to the belief that graphs and graph theory may have begun in then the resulting structure (Figure4.2) is a multigraph.
Switzerland in the early 18th century. In any case, it is evident that the The Konigsberg Bridge Problem is then equivalent to the problem of
great Swiss mathematician Leonhard Euler [E6] was thinking in graphical determining whether the multigraph of Figure 4.2 has a trail containing
terms when he considered the problem of the seven Konigsberg bridges. all of its edges.
Figure 4.1 shows a map of Konigsberg as it appeared in the 18th The Konigsberg Bridge Problem suggests the follOWing two concepts.
century. The river Pregel was crossed by seven bridges, which connected An eulerian trail of a graph G is an open trail of G containing all of the
two islands in the river with each other and with the opposite banks. We edges and vertices of G, while an eulerian circuit of G is a circuit containing
denote the land regions by the letters A, B, C and D (as Euler himself did). all of the edges and vertices of G. A graph possessing an eulerian circuit
It is said that the townsfolk of Konigsberg amused themselves by trying is called an eulerian graph. Necessarily, then, graphs containing eulerian
to devise a route that crossed each bridge just once. (For a more detailed trails and eulerian circuits are nontrivial connected graphs. The graph G1
account of the Konigsberg Bridge Problem, see Biggs, Lloyd and Wilson of Figure 4.3 contains an eulerian trail while Gz is an eulerian graph.
[BLWl, p.l].)
Euler proved that such a route over the bridges of Konigsberg is impos-
sible - a fact of which many of the people of Konigsberg had already c
convinced themselves. However, it is probable that Euler's approach to
the problem was considerably more sophisticated.
Euler observed that if such a route were possible it could be represented
by a sequence of eight letters, each chosen from A, B, C and D. A term of A ~------~---~;oD
the sequence would indicate the particular land area to which the route
had progressed while two consecutive terms would denote a bridge
traversed while proceeding from one land area to another. Since each
bridge was to be crossed only once, the letters A and B would necessarily B
appear in the sequence as consecutive terms twice, as would A and C. Figure 4.2 The multigraph of Konigsberg.
94 Eulerian and hamiltonian graphs and digraphs Eulerian graphs and digraphs 95
component of G' is a nontrivial graph with fewer than m edges and is
eulerian by hypothesis. Since G is connected, every component of G'
has a vertex that also belongs to W. Hence an eulerian circuit of G can
be constructed by inserting an eulerian circuit of each component H' of
G' at a vertex of H' also belonging to W. 0

Figure 4.3 Graphs with eulerian trails and eulerian circuits. A graph G is defined to be an even graph (odd graph) if all of its vertices
have even (odd) degree. Thus, by Theorem 4.1, the nontrivial connected
Simple but useful characterizations of both eulerian graphs and graphs even graphs are precisely the eulerian graphs. A characterization of
with eulerian trails exist; in fact, in each case the characterization was graphs containing eulerian trails can now be presented.
known to Euler [E6]. Complete proofs of these results were not given
until 1873, however, in a paper by Hierholzer [fII3].
Theorem 4.2

Theorem 4.1 Let G be a nontrivial connected graph. Then G contains an eulerian trail if and
only iiG has exactly two odd vertices. Furthermore, the trail begins at one of these
Let G be a nontrivial connected graph. Then G is eulerian if and only if every odd vertices and terminates at the other.
vertex of G is even.

Proof
Proof
If G contains an eulerian u-v trail, then, as in the proof of Theorem 4.1,
Let G be an eulerian graph with eulerian circuit C, and let v be an arbitrary
every vertex of G different from u and v is even. It is likewise immediate
vertex of G. Ifv is not the initial vertex of C (and therefore not the final
that each of u and v is odd.
vertex either), then each time v is encountered on C, it is entered and left
Conversely, let G be a connected graph having exactly two odd vertices
by means of distinct edges. Thus each occurrence of v in C represents a
u and v. If G does not contain the edge e = uv, then the graph G + e is
contribution of 2 to the degree of v so that v has even degree. If v is the
eulerian. If the edge e is deleted from an eulerian circuit of G + e, then
initial vertex of C,then C begins and ends with v, each term representing a
an eulerian trail of G results. In any case, however, a new vertex w can
contribution of 1 to its degree while every other occurrence of v indicates
be added to G together with the edges uw and vw, producing a connected
an addition of 2 to its degree. This gives an even degree to v. In either case,
graph H in which every vertex is even. Therefore, H is eulerian and
vis even.
contains an eulerian circuit C. The circuit C necessarily contains uw and
Conversely, let G be a nontrivial connected graph in which every vertex
vw as consecutive edges so that the deletion of w from C yields an eulerian
is even. We employ induction on the number m of edges of G. For m = 3,
trail of G. Moreover, this trail begins at u or v and terminates at the
the smallest possible value, there is only one such graph, namely K 3 , and
this graph is eulerian. Assume then that all nontrivial connected graphs other. 0
having only even vertices and with fewer than m edges, m ~ 4, are
eulerian, and let G be such a graph with m edges. If G is a connected graph with 2k odd vertices (k ~ 1), then the edge set
Select some vertex u in G, and let W be a u-u circuit of G. Such a circuit of G can be partitioned into k subsets, each of which induces a trail
exists in G since if Wi is any u-v trail of G, where u =1= v, then necessarily an connecting odd vertices (Exercise4.2). However, even more can be said.
odd number of edges of G incident with v are present in W', implying that This result was extended in [CPSl]. (See Exercise 4.3 for a special case of
W' can be extended to a trail W" containing more edges than that of W'. the next theorem.)
Hence Wi can be extended to a u-u circuit W otG.
If the circuit W contains every edge of G, then W is an eulerian circuit of
Theorem 4.3
G and G is eulerian. Otherwise, there are edges of G that are not in W.
Remove from G all those edges that are in W together with any resulting If G is a connected graph with 2k odd vertices (k ~ 1), then E(G) can be parti-
isolated vertices, obtaining the graph G'. Since each vertex of W is incident tioned into subsets E1, E2 , • •• , Ek so that for each i, (E j ) is a trail connecting odd
with an even number of edges of W, every vertex of G' is even. Every vertices and such that at most one of thes'e trails has odd length.
96 Eulerian and hamiltonian graphs and digraphs Eulerian graphs and digraphs 97
We note that analogues to Theorems 4.1 and 4.2 exist for multigraphs. only once, namely as the terminal vertex. There is an odd number of edges
It therefore follows that the multigraph of Figure 4.2 contains neither an possible for the initial edge of such a trail. Once the initial edge has b~en
eulerian trail nor an eulerian circuit. Eulerian graphs have several useful chosen and the trail has then proceeded to the next vertex, say w, then
characterizations. The following result, due to Veblen [V2], characterizes again there is an odd number of choices for edges that are incident with
eulerian graphs in terms oftheir cycle structure. w but different from uw. Wecontinue this process until we arrive at vertex
v. At eachvertex different from v in such a trail, there is an odd number of
edges available for a continuation of the trail. Hence there is an odd
Theorem 4.4
number of these trails.
A nontrivial connected graph G is eulerian if and only if E(G) can be partitioned Suppose that TI is a u-v trail that is not a u-v path and TI contains v only
into subsets Ej , 1 ~ i ~ k, where each subgraph (Ej) is a cycle. once. Then som~ vertex VI (=rfv) occurs at least twice on Tv implying that
TI contains a VI-VI circuit, say C: VI' V2,"" Vb VI' Hence, there exists a u-v
trail T2 identical to TI except that C is replaced by the 'reverse' circuit
Proof
C': Vb Vb Vk-b' .• , V2, VI' This implies that the u-v trails that are not u-v
Let G be an eulerian graph. We employ induction on the number m of paths occur in pairs. Therefore, there is an even number of such u-v trails
edges of G. If m = 3, then G = K3 and G has the desired property. that are not u-v paths and, consequently, there is an odd number of u-v
Assume, then, that the edge set of every eulerian graph with fewer than paths in G - e. This, in tum, implies that there is an odd number of cycles
m edges, m ~ 4, can be partitioned into subsets each of which induces a containing e.
cycle, and let G be an eulerian graph with m edges. Since G is eulerian, G is For the converse, suppose that G is a nontrivial connected graph that is
an even graph and G has at least one cycle C. If E(G) = E(C), then we not eulerian. Then G. contains a vertex V of odd degree. For each edge e
have the desired (triyial) partition of E(G). Otherwise, there are edges incident with v, denote by c(e) the number of cycles of G containing e.
of G not in C. Remove the edges of C to obtain the graph G'. As in the Since each such cycle contains two edges incident with v, it follows that
proof of Theorem 4.1, every nontrivial component of G' is a nontrivial E c(e) equals twice the number of cycles containing v. Because there is an
connected even graph and so, by :rheorem 4.1, is an eulerian graph odd number of terms in this sum, some c(e) is even. 0
with fewer than m edges. Thus, by the inductive hypothesis, the edge
set of each nontrivial component of G' can be partitioned into subsets, We now briefly consider the directed analogue of eulerian graphs. An
each inducing a cycle. These subsets, together with E(C), give the desired eulerian trail of a digraph D is an open trail of D containing all of the arcs
partition of E(G). and vertices of D, and an eulerian circuit is a circuit containing every arc
For the converse, suppose that the edge set of a nontrivial connected and vertex of D. A digraph that contains an eulerian circuit is; called an
graph G can be partitioned into subsets Ei , 1 ~ i ~ k, where each subgraph eulerian digraph. The digraph Di of Figure 4.4 is eulerian while D2 has an
(Ei) is a cycle. This implies that G is a nontrivial connected even graph and eulerian trail.
so, by Theorem 4.1, G is eulerian. 0 We now present a characterization of eulerian digraphs whose state-
ment and proof are very similar to Theorem 4.1. Recall that a digraph D
We next present a characterization of eulerian graphs involving parity is connected if D contains a u-v semipath for every pair u, V of vertices
and cycle structure. The necessity is due to Toida [T7] and the sufficiency ofD.
to McKee [M5].

Theorem 4.5
A nontrivial connected graph Gis eulerian ifand only if every edge of G lies on an
odd number of cycles. .

Proof
First, let G bean eulerian graph and let e = uv be an edge of G. Then G -e
is connected. Consider the set of all u-v trails in G - e for which v appears Figure 4.4 Digraphs with eulerian circuits' and eulerian trails.
98 Eulerian and hamiltonian graphs and digraphs Eulerian graphs and digraphs 99
Theorem 4.6 i = 1,2, ... , k. The set P is called a pairing of S in G. A minimum pairing of S
in G is one for which
Let D be a nontrivial connected digraph. Then D is eulerian if and only if
ad v = id v for every vertex v of D.

With the aid of Theorem 4.6, it is easy to give a characterization of

digraphs containing eulerian trails. is minimum. Certainly if G is connected, then G contains a pairing of S.
Indeed, G contains a pairing of 5 if and only if every component of G
contains an even number of vertices of S.
Theorem 4.7 Another useful observation is that the paths in a minimum pairing P
Let D be a nontrivial connected digraph. Then D has an eulerian trail ifand only if are necessarily pairwise edge-disjoint; for, otherwise, suppose that
D contains vertices u and v such that P: xI,xZ,'" ,Xt and Q: WI' wz,'''' wr are paths in P with XjXj+1 = WjWj+1
for some i and j. Then P and Q can be replaced by the paths
ad u = id u + 1 and id v = ad v + 1
and od W = id Wfor all other vertices Wof D. Furthermore, the trail begins at u
P: xI,xZ,'" ,Xj = Wj,Wj_I,··· ,WI
and ends at v. and

We now return to graphs. Neither of the graphs GI and Gz of Figure 4.5

is eulerian. However, we can obtain an eulerian graph fromG I by adding contradicting the minimality of P.
the edges ad, bb' and ec'. On the other hand, as we shall see, there are The next theorem is due to Jaeger [J3].
no edges that can be added to Gz to produce an eulerian graph. These
observations lead to our next topic. Theorem 4.8
A graph G is called subeulerian if it is possible to add edges to G to obtain
an eulerian graph, that is, if G is a spanning subgraph of an eulerian Let G be a connected graph with X <;;; E (G). Then G contains an even subgraph H
graph. Thus the graph GI of Figure 4.5 is subeulerian while the graph with X <;;; E(H) if and only if X contains no minimal edge-cut of G having odd
Gz of Figure 4.5 is not subeulerian. So graphs of even order mayor may cardinality.
not be subeulerian. However, every graph of odd order at least 3 is sub-
eulerian since it is a spanning subgraph of a complete graph of odd order, Proof
which is necessarily eulerian.
Theorem 4.8 will provide us with a tool to characterize connected sub- Suppose first that H is a subgraph of G containing a minimal edge-cut Y of
eulerian graphs. Before presenting this result, we introduce some useful G haVing odd cardinality. We show that H is not an even graph, which
terminology. will in turn show that every subgraph of G that contains a set X of edges of
~et Sbe a .set of 2k vertices of a graph·G. We say there is a pairing of Son G with Y <;;; Xis not even. Since Y is a minimal edge-cut of G, the set Y is
G if the vertices of S can be labeled as UI, uz, ... , Uk> VI , Vz, ••• , Vk so that G certainly an edge-cut of H as well <although not necessarily a minimal
contains a set P = {Pt>Pz, ... ,Pk } of k paths, where P j is a Uj-Vj path for edge-cut). Consequently, there is a partition of V(H) into subsets "VI
and V z such that the edges of H between VI and V z are precisely those
b c of Y. Thus,
L degHv = 2I E((VI))1 + IYI·
VEV1

a O---<:>-----(')...-1h a' Since \YI is odd, H contains a vertex of odd degree and so H is not even.
For the converse, assume that X is a set of edges of G that contains no
minimal edge-cut of G of odd cardinality. We show that there is an even
subgraph H of G containing X. If (X) itself is an even graph, then H = (X)
b' c' is the desired subgraph of G. Supposethen that (X) contains odd vertices,
Figure 4.5 Subeulerian graphs. and let 0 be the set of odd vertices of'(X). We show that there is a pairing
100 Eulerian and hamiltonian graphs and digraphs Eulerian graphs and digraphs 101
of the set 0 in the graph G - X. It suffices to show that every component a b
of G - X contains an even number of vertices of O. This is certainly true if
G - X is connected.
Suppose, then, that G - X is disconnected and that Go is a component of
G - X. We show that Go contains an even number of vertices of O. Let c' ()----¥----Oc
Gll Gz, ,Gdk ~ 1) be the components of G - V(Go) and for
i = 1,2, , k, let Xi denote the edges between Go and Gi . Hence each set
Xi (1 ~ i ~ k) is a minimal edge-cut of G and Xi ~ X. By hypothesis then, b'
IXii isevenforeachi (1:::; i ~ k). Let X' = Uf=l Xi. Thus IX'I is even and X'
consists of all those edges of X that are incident with exactly one vertex Figure 4.6 A supereulerian graph.
of Go.
We now construct a graph G' consisting of the vertices and edges of (X) cardinality are precisely the sets of edges inducing a spanning odd
that belong to Go plus IX'I new vertices, one for each edge of X' and joined complete bipartite graph, the result follows. 0
to the vertex of Go that is incident with the corresponding edge of
X'. Hence each newly added vertex has degree 1 in G', and The graph G3 of Figure 4.6 is, of course, not eulerian. It is not sub-
degG' v = deg(X} v for each vertex v in Go. The number of odd vertices of eulerian either. Since G3 = K3 ,3, this follows from Corollary 4.9.
G'is Ion V(Go)1 + IX'I. Since G' has an even number of odd vertices and We now turn to a concept that is, in a certain sense, opposite to sub-
IX'I is even, the number of elements of 0 in Go is even. eulerian graphs. We have just observed that we cannot produce an
Thus, as claimed, there is a pairing of 0 in G - X; so there is a minimum eulerian graph by adding edges to G3 • However, the removal of the
pairing ofP = {P1 ,PZ"" ,Pt } of 0 in G - X. Therefore, the paths in Pare edges aa', bb' and cc' results in an eulerian graph.
pairwise edge-disjoint. Hence the graph H with A graphG is supereulerian if it is possible to remove edges from G to
obtain an eulerian graph, i.e. if G has a spanning eulerian subgraph.
V(H) = V({X)) u (~ V(Pi )) Therefore, the graph G3 of Figure 4.6 is supereulerian. The graph
G4 = KZ,3 is not supereulerian.
and No characterization is known for supereulerian graphs although there
are many sufficient conditions for a graph to be supereulerian (see Catlin
u (~
[C2]). Corollary 4.10, due to Jaeger U3), gives one such condition.
E(H) = X E(Pi ))

contains X and is even. 0 Corollary 4.10

If a graph G contains two edge-disjoint spanning trees, then G is supereulerian.
Corollary 4.9, first discovered by Boesch, Suffel and Tindell [BSTl],
characterizes connected subeulerian graphs.
Proof

Corollary 4.9 Let T1 and Tz be edge-disjoint spanning trees of G. Since Tz ~ G - E(T1 ),

the graph G - E(T1 ) is connected and so E(T1 ) contains no minimal edge-
A connected graph G is subeulerian if and only if G contains no spanning odd cut (of odd cardinality). Thus there is an even subgraph H of G with
complete bipartite graph. E(T1 ) ~ E(H). Since T1 ~ H, it follows that H is a spanning connected
even subgraph of G; so G is supereulerian. 0
Proof
According to Theorem 4.4, the edge set of an eulerian graph G can be
Let G be a connected graph of order n. Now G is subeulerian if and only if partitioned into subsets E1 ,Ez, ... , Ek, where Gi {E i ) is a cycle for
there is an even subgraph H of Kn with E(G) ~ E(H). By Theorem 4.8, i=1,2, ... ,k. Thus the collection C={Gl,GllGZ,GZ, ... ,GbGk} of
such a subgraph H exists if and only if E(G) contains no minimal edge- cycles of G(where multiplicities are allowed) has the property that
cut of Kn of odd cardinality. Since the minimal edge-cuts of Kn of odd every edge of G is in exactly two cycles of C. Such a collection of cycles
102 Eulerian and hamiltonian graphs and digraphs Hamiltonian graphs and digraphs 103
is called a cycle double cover of G. Many graphs that are not eulerian also 4.7 Show that if 0 is a connected digraph such that
have cycle double covers. For example, the Petersen graph has a cycle LVEV(D) lod v- id vi = 2t, where t ~ 1, then E(O) can be partitioned
double cover, consisting of five cycles. Szekeres [S11] in 1973 and into subsets Ei , 1 ~ i ;;:; t, so that (E i ) is an open trail for each i.
Seymour [S3] in 1979 independently made the following conjecture. 4.8 For each integer k ~ 2, give an example of a connected graph Gk of
order 2k that is neither subeulerian nor supereulerian.
The Cycle Double Cover Conjecture
4.9 Characterize disconnected subeulerian graphs.
Every 2-edge-connected graph has a cycle double cover.
4.10 Show that the Petersen graph has a cycle double cover consisting of
A related conjecture, due to Bondy [B13], bounds the number of cycles five cycles.
in a cycle double cover.

The Small Cycle Double Cover Conjecture

4.2 HAMILTONIAN GRAPHS AND DIGRAPHS
Every 2-edge-connected graph of order n ~ 3 has a cycle double cover consisting
of fewer than n cycles.
A graph G is defined to be hamiltonian if it has a cycle containing all the
We close this section with suggestions for further reading. Survey vertices of G. The name 'hamiltonian' is derived from Sir William Rowan
articles on eulerian graphs and related topics can be found in Fleischner Hamilton, the well-known Irish mathematician. Surprisingly, though,
[F6] and in [L01], and a book by Fleischner [F7] is another useful resource. Hamilton's relationship with the graphs bearing his name is not strictly
Readers interested in more information on supereulerian graphs should mathematical (see Biggs, Lloyd and Wilson [BLWl, p.3l]). In 1857,
consult the survey by Catlin [C2], which also includes information on the Hamilton introduced a game consisting of a solid regular dodecahedron
cycle double cover conjecture. made of wood, twenty pegs (one inserted at each corner of the dodeca-
hedron), and a supply of string. Each corner was marked with an
important city of the time. The aim of the game was to find a route
along the edges of the dodecahedron that passed through each city
EXERCISES 4.1 exactly once and that ended at the city where the route began. In order
for the player to recall which cities in a route had already been visited, the
4.1 In present-day Konigsberg (Kaliningrad), there are two additional string was used to connect the appropriate pegs in the appropriate order.
bridges, one between regions Band C, and one between regions There is no indication that the game was ever successful.
Band O. Is it now possible to devise a route over all bridges of The object of Hamilton's game may be described in graphical terms,
Konigsberg without recrossing any of them? namely, to determine whether the graph of the dodecahedron has a
4.2 Let G be a connected graph with 2k odd vertices, k ~ 1. Show that cycle containing each of its vertices (Figure4.7). It is from this that we
E(G) can be partitioned into subsets Ei , 1 ~ i ~ k, so that (E i ) is an get the term 'hamiltonian'.
open trail for each i. Then show that for t < k, E(G) cannot be parti- It is interesting to note that in 1855 (two years before Hamilton intro-
tioned into subsets Ei , 1 ~ i ~ t, sothat (E i ) is an open trail for each i. duced his game) the English mathematician Thomas P. Kirkman posed
the follOWing question in a paper submitted to the Royal Society: Given
4.3 Show that every nontrivial connected graph G of even size having the graph of a polyhedron, can one always find a circuit that passes
exactly four odd vertices contains two trai.lsT1 and T z of even length through each vertex once and only once? Thus, Kirkman apparently intro-
suchthat {E(T1),E(Tz)} is a partition of E(G). duced the general study of 'hamiltonian graphs' although Hamilton's
4.4 Prove Theorem 4.6. game generated interest in the problem.
A cycle of a graph G containing every vertex of G is called a hamiltonian
4.5 Prove Theorem 4.7. cycle of G; thus, a hamiltonian graph is one that possesses a hamiltonian
4.6 Prove that a nontrivial connected digraph 0 is eulerian if and only if cycle. Because of the similarity in the definitions of eulerian graphs and
E(D) can be partitioned into subsets Ei , 1 ~i ~ k, where (E i ) is a hamiltonian graphs, and because a particularly useful characterization of
cycle for each i. eulerian graphs exists, one might well expect an analogous criterion for
104 Eulerian and hamiltonian graphs and digraphs Hamiltonian graphs and digraphs 105
hamiltonian. For example, let k ~ 2 be a positive integer and· consider
G = Kk,k+ It which is k-connected. Let S denote the partite set of cardinality
k. Then keG - 5) = k+ 1> 151, which implies by Theorem 4.11 that G is
not hamiltonian.
We have also seen in Corollary 4.12 that every hamiltonian graph is
I-tough. Altl:1ough the converse is not true here either, it has been conjec-
tured by Chvatal [C7] that there is a constant k such that every k-tough
graph is hamiltonian. Indeed, Chvatal has conjectured that every 2-tough
graph is hamiltonian.
There have been several sufficient conditions established for a graph to
be hamiltonian. We consider some of these in this section. The following
Figure 4.7 The graph of the dodecahedron. result is due to are [01].
hamiltonian graphs. However, such is not the casei indeed it must be
considered one of the major unsolved problems of graph theory to Theorem 4.13
develop an applicable characterization of hamiltonian graphs.
If G is a graph of order n ~3 such that for all distinct nonadjacent vertices u
If G is a hamiltonian graph, then certainly G is connected, G contains a
and v,
hamiltonian cycle and so G has no cut-vertices, and, of course, G has order
at least 3; so G is 2-connected.Therefore, a necessary condition for G to be degu + degv ~ n,
hamiltonian is that G be 2-connected; that is, if G is not 2-connected, then then G is hamiltonian.
G is not hamiltonian. A less obvious necessary condition is presented
next.
Proof
Theorem 4.11 Assume that the theorem is not true. Hence there exists a maximal non-
hamiltonian graph G of order n ~ 3 that satisfies the hypothesis of the
If G is a hamiltonian graph, then for every proper nonempty set 5 of vertices of G,
theoremi that is, G is nonhamiltonian and for every two nonadjacent
keG - 5) :::;; 151. vertices WI and Wz of G, the graph G + WIWZ is hamiltonian. Since G is
nonhamiltonian, G is not complete.
Proof Let u and v be two nonadjacent vertices of G. Thus, G + uv is hamilto-
nian, and, furthermore, every hamiltonian cycle ofG + uv contains the
Let 5 be a proper nonempty subset of V(G), and suppose that edge uv. Thus there is a u-v path P: U = UI, Uz, ••• ,Un = v in G containing
keG - 5) = k ~ 1, where GI , Gz, ... , Gk are the components of G -,- S. Let every vertex of G.
C be a hamiltonian cycle of G. When Cleaves Gj (1 ::;; j ::;; k); the next If UIUj E E(G),2::;; i::;; n, then Uj_IU n ¢ E(G)i for otherwise,
vertex of C belongs to 5. Thus keG - 5) = k::;; 151. 0

Consequently, for every. vertex-cut 5 of a hamiltonian graph G, it fol- is a hamiltonian cycle of G. Hence for each vertex of {U2, U3,'" ,un} adja-
lows that ISI/k(G - 5) ~ 1. cent to U1 there is a vertex of {U1' U2, •.• , Un -I} not adjacent with Un' Thus,
degu n :::;; (n -1) - degul so that
Corollary 4.12 deg u + deg v :::;; n - 1.
Every hamiltonian graph is I-tough. This presents a contradictioni so G is hamiltonian. 0

We now turn our attention to sufficient conditions for a graph to be If a graph G is hamiltonian, then certainly so is the graph G + uv, where
hamiltonian. Although every hamiltonian graph is 2-connected, the U and v are distinct nonadjacent vertices of G. Conversely, suppose that G
converse is not true. Indeed; no connectivity guarantees that a graph is is a graph of order n with nonadjacent vertices U and v such that G + uv is
106 Eulerian and hamiltonian graphs and digraphs Hamiltonian graphs and digraphs 107
Our next theorem is a simple consequence of the definition of closure
and Theorem 4.14.

Theorem 4.16
A graph is hamiltonian if and only if its closure is hamiltonian.
C(G)
G
Since each complete graph with at least three vertices is hamiltonian,
Figure 4.8 The closure function. we obtain Bondy and Chvatal's [BC3] sufficient condition for a graph to
be hamiltonian.
hamiltonian; furthermore, suppose that degG U + degG v ~ n. If G is not
hamiltonian, then, as in the proof of Theorem 4.13, we arrive at the contra-
diction that degG u + degG v ~ n - 1. Hence we have the following result, Theorem 4.17
which was first observed by Bondy and Chvatal [BC3].
Let G be a graph with at least three vertices. If C(G) is complete, then G is
hamiltonian.
Theorem 4.14
Let u and v be distinct nonadjacent vertices of a graph G of order n such that If a graph G satisfies the conditions of Theorem 4.13, then C(G) is
deg u + deg v ~ n. Then G + uv is hamiltonian if and only if G is hamiltonian. complete and so, by Theorem 4.17, G is hamiltonian. Thus, Ore's theorem
is an immediate corollary of Theorem 4.17 (although chronologically it
Theorem 4.14 motivates our next definition. The closure of a graph G of preceded the theorem of Bondy and Chvatal by several years). Perhaps
order n, denoted by C(G),.is the graph obtained from G by recursively surprisingly, many well-known sufficient conditions for a graph to be
joining pairs of nonadjacent vertices whose degree sum is at least n (in hamiltonian based on vertex degrees can be deduced from Theorem
the resulting graph at each stage) until no such pair remains. Figure 4.8 4.17. Theorem 4.18, due to Chvatal [C6t is an example of one of the
illustrates the closure function. That C(G) is well-defined is established strongest of these.
next.
Theorem 4.18
Theorem 4.15 Let G be a graph of order n ~ 3, the degrees di of whose vertices satisfy
If G1 and Gz are two graphs obtained from a graph G of order n ,by recursively d1 ~ dz ~ ... ~ dn . If there is no value of k < n/2 for which dk ~ k and
joining pairs of nonadjacent vertices whose degree sum is at least n, then dn - k ~ n - k -1, then G is hamiltonian.
G1 = Gz·
Proof
Proof
We show that C(G) is complete which, by Theorem 4.17, implies that G is
Let el,eZ,'" ,ej and 11,12, ... ,fk be the sequences of edges added to.G t? hamiltonian. Assume, to the contrary, that C(G) is not complete. Let u and
obtain G1 and Gz, respectively. It suffices to show that each ei (1 ~ l ~J) w be nonadjacent vertices of C(G) for which degC(G) u + degc:(G) W is as
is an edge of Gz and that each fi (1 ~ i ~ k) is an edge of G1 • Assume, to the large as possible. Since u and ware nonadjacent vertices of C(G), it follows
contrary, that this is not the case. Thus we may assume, without loss of that degC(G) U + degC(G) w ~ n - 1. Assume, without loss of generality,
generality, that for some t satisfying 0 ~ t ~ j - 1, the edge et+ 1 = uv does that degC(G) u ~ degC(G) w. Thus if k = degC(G) u, we have that k ~
not belong to Gz; furthermore, ei E E(G z) for i ~ t. Let G3 be the graph (n -1)/2 < n/2 and degC(G) w ~ n -1-- k. Let W denote the vertices
obtained from G by adding the edges el' ez, ... ,et . It follows from the other than w that are not adjacent to w in C(G). Then IWI =
definition of G1 that degG3 u + deBe3 v ~ n. This is a contradiction, how- n -1 - degC(G) w ~ k. Also, by the choice of u and w, every vertex v E W
ever, since u and v are nonadjacent vertices of G2• Thus each edge ej satisfies degG v ~ degC(G) v ~ degC(G) u = k. Thus, G has at least k vertices
belongs to Gz and each edge fi belongs to G1; that is, G1 = G2·D of degree at most k and so dk ~ k. Similarly, let U denote the vertices other
II
108 Eulerian and hamiltonian graphs and digraphs Hamiltonian graphs and digraphs 109
than u that are not adjacent to u in C(G). Then lUI = n -1 - degC(G) U = results for r-regular graphs, the condition that G be 2-connected is
n - k - 1. Every vertex v E U satisfies degG v < degC(G) v < degC(G) w< included. Certainly, this is a necessary condition for G to be hamiltonian.
n -1- k, implying that d,,-k-l < n - k -1. However, degG U < However, unlike the situation in Theorems 4.13, 4.17 and 4.18, 2-connect-
degC(G) u < degC(G) W < n -1- k, so d l1 - k < n - k -1. This, however, edness is not guaranteed by the degree condition. For example, there are
contradicts the hypothesis of the theorem. Thus, C(G) is complete. 0 (n/3)-regular graphs of order n that are not 2-connected and hence not
hamiltonian. The neighborhood N(v) of a vertex v in a graph G is the set
Perhaps the simplest sufficient condition for a graph to be hamiltonian of all vertices of G that are adjacent to v. The following result of Faudree,
is due to Dirac [D4]. It is a simple consequence of each of Theorems 4.13, Gould, Jacobson and Schelp [FGJSl] gives a sufficient condition for
4.17 and 4.18. hamiltonicity based on the cardinality of the union of the neighborhoods
of nonadjacent vertices. The proof given here is due to Fraisse [FlO].
Corollary 4.19
Theorem·4.20
If G is a graph of order n ;:, 3 such that deg v ;:, n/2 for every vertex v ofG, then
G is hamiltonian. If G is a 2-connected graph of order n such that for all distinct nonadjacent
vertices U and v,
Each of the sufficient conditions presented so far requires that the graph 2n -1
under consideration contains some vertices of degree at least n/2. In the IN(u) UN(v)!;:' - 3 - '
case of regular graphs, however, this situation can be improved. Jackson
£II] has shown that every 2-connected r-regular graph of order at most 3r then G is hamiltonian.
is hamiltonian. The Petersen graph, for exarhple~ shows that 3r cannot be
replaced by 3r +1. Define the class:F to be the set of all 2-connected Proof
graphs that are spanning subgraphs of one of the graphs GI , Gz and G3
Since Gis 2-connected, G contains at least one cycle. Among all cycles of
of Figure 4.9. Since none of these three graphs is hamiltonian, no graph G, let C be one of maximum length. We show that C is a hamiltonian cycle.
in :F is hamiltonian. In particular, the regular graphs in :F are non- Assume, to the contrary, that there is a vertex Wo of G that is not on C. Thus
hamiltonian. However, in [BJVV1] the following generalization of
we may apply Theorem 3.26 to obtain two paths PI and Pz having initial
Jackson's result was conjectured. Let G be a 2-connected r-regular
vertex Wo that are pairwise disjoint, except for wo, and that share with C
graph of order at most 4r. Then for r ;:, 4, G is hamiltonian or G is a
only their terminal vertices Xl and Xz. For i= 1, 2, let Wi be the vertex
member of the class :F. It is known that this conjecture is true for
following Xi in some fixed cyclic orientation of C.
graphs with at most (7r -14)/2 vertices.
No vertex Wi (i = 1,2) is adjacent to Wo; for otherwise we could replace
Other approaches have been used to determine sufficient conditions for
the edge XiWi in C by the Xi-Wi path determined by the path Pi and the
hamiltonicity that are satisfied by graphs G with minimum degree less
edge WOWi to obtain a cycle having length at least IV(C)! + 1, which is
than n/2. For these results, as in the case of the previously mentioned
impossible. Similarly, WIWZ 'I E(G); otherwise we could replace the
edges Xl WI and XzWz in C by the paths PI and P z together with the edge
WI Wz and obtain a cycle longer than C. Thus {Wo, Wll wz} is an
independent set. Finally, no vertex not on C is adjacent to two of
WO,WI.WZ or again a cycle longer that C would be produced. We show
that IN(wk) UN(wt)! < (2n -1)/3 for some integers k and t with
o< k #l <,2, producing a contradiction and completing the proof,
Let R denote the vertices of G adjacent to none of Wo, WI' Wz; let S denote
the vertices of G adjacent to exactly one of Wo, WI. Wz; and let T denote the
vertices of G adjacent to at least tWo of Wo, WI> Wz. Furthermore, let
P.q,s~ 1 p, q, s ~ 2 p,q,s~3 Rc = R n V(C) and let Rc = R - V(C). Then {wo. wI> wz} ~ R and, in
p+q+s=n-2 p+q+s=n-l p+q+s"'n particular, Wo E RC' so that
Figure 4.9 Members of the class :F. IRel ~ 1. (4.1)
110 Eulerian and hamiltonian graphs and digraphs Hamiltonian graphs and digraphs 111
Also, as observed, From (4.1) we know that IR c,I;31. Thus IN(Wk)UN(w()I:s;
(V(G) - V(C)) n T = 0. (4.2) (2n - 2)/3 < (2n -1)/3, and the proof is complete. D
We next show that there are at most two elements of T between con-
In [FGJL2]' another neighborhood result shows that if G is a 2-con-
secutive elements (with respect to the fixed cyclic ordering of C) of Ron C.
nected graph of sufficiently large order n such that IN(u) UN(v)1 ;3 n/2
In the remainder of the proof we assume that the discussion is always
for all pairs u,v of vertices of G, then G is hamiltonian. Clearly, this
with respect to this ordering of C. Let a1 and az be consecutive vertices
theorem generalizes Dirac's result for graphs of sufficiently large order.
of Ron C. Since {wo, W1, wz} ~ R, it follows that a1 and az belong either
~e?rem 4.21, presented without proof because of the length and
to the W1-wZ subpath or WZ-WI subpath of C. Without loss of generality,
dIffIculty of the proof, has the preceding result, Dirac's theorem, Ore's
assume that a1 and az, in this order, belong to the W1-WZ subpath. Let P be
theorem, and Theorem 4.18 as corollaries. This result is due to Broersma,
the a1-aZ subpath of C. Then every internal vertex of P is adjacent to at
van den Heuvel and Veldman [BVV1]. Recall that F is the class of all 2-
least one of wo, W1, Wz. Suppose that v and Vi denote a vertex and its
connected spanning subgraphsof one of the graphs of Figure 4.9.
successor on P with the property that v E N(w;) and Vi E N(wj) for some
integers i and j with 0 :s; i i- j :s; 2. First, i i- 0; for otherwise, j = 1 or j = 2,
and in either case we obtain a cycle longer than C. Thus i = 1 or i = 2. If Theorem 4.21
i = 2, then j = 0; otherwise j = 1 and we obtain a cycle longer than C. IfG is a 2-connected graph of order n such that for all distinct nonadjacent
Finally, if i = 1, then j = 0 or j = 2 since i i- j. It follows that if vertices u and v,
No = yep) n N(wo), N 1 = yep) n N(w1) and N z = yep) n N(wz), then
n
(i) each Ni consists of a (possibly empty) sequence of consecutive IN(u) U N(v) I ;3 2'
internal vertices of P;
then G is hamiltonian, G is the Petersen graph, or G E F.
(ii) each element of N I precedes or equals each element of N z and each
element of N z precedes each element of No; and
The ~ufficient conditions for hamiltonicity that we have presented so
(iii) IN1 n Nzi :s; 1, INz n Nol :s; 1 and INI n Nol is 0 or at most 1 depending
far all Involve the number of adjacencies of the vertices of the graph.
on whether N z i- 0 or N z = 0·
Our next result, however, involves the cardinality of independent sets
Since each element of T between a1 and az must lie in the intersection of of vertices and the connectivity of the graph. The next result is due to
two of the sets Ni (0 :s; i :s; 2), it follows that there are at most two Chvatal and Erdos [CEl]. The proof technique has become a standard
elements of T between a1 and az· tool (see, for example, the proof of Theorem 4.20).
To complete the proof, let j (0 :s; j :s; 2) be chosen so that Wj has the .
maximum number s of neighbors in 5. Then s ;3 151/3. Since, by (4.2), no
verh~x not on C is adjacent to two of WO, W1, Wz, the number of vertices of 5
Theorem 4.22
not on C is n -IV(c)I-IRe'I. From the preceding discussion we see that Let G be a graph with at least three vertices. If ~(G) ;3 j3(G) then G is
the number of vertices of 5 on C is at least 1V(c)I- 31Rcl. Thus, hamiltonian. '
151 = IS - V(C)I + IS n V(C)I ;3 (n -1V(c)I-I Re'1) + (IV(C)I- 31Rc l)
= n -IRe'1 ~ 31Rc l, Proof
and so If j3( G) = 1, then the result follows since G is complete. Hence we assume
151 n IRe' I that j3(G) ;3 2. Let ~(G) = k. Since k;3 2, G contains at least one cycle.
s;3 3 ;3 3" - 3 - IReI·
Among all cycles of G, let Cbe one of maximum length. By Theorem
However, then, Wk and W( (O:S; t i- k:s; 2; t,k i- j) are nonadjacent 3.27, there ar-e at least k vertices on C. We show thatC i$a hamiltonian
vertices of G such that cycle of G. Assume, to the contrary, that there is a vertex W of G that does
IN(wk) UN(w()1 =n -IRI- s not lie on C. Since IV(C)! ;3 k, we may-apply Theorem 3.26 to conclude
t~a~ ~here are k paths PI' Pz,... ,Pk having initial vertex W that are pairwise
= n - IRe' I - IRc I- s
dlsJomt, except for w, and that share with C only their terminal vertices
2n 21 RC' I VI' vz,' .. , vk, respectively. For each i = 1,2, ... , k, let Uj be the vertex fol-
::::::----
'" 3 3' lowing Vi in some fixed cyclic ordering of C. No vertex Uj is adjacent to W
112 Eulerian and hamiltonian graphs and digraphs Hamiltonian graphs and digraphs 113
in G; for otherwise we could replace the edge ViUi in C by the Vi-Uj path Case 3. Assume that R' = 0 and T' = 0. Let u E R and vET. Then, in a
determined by the path Pj and the edge UjW to obtain a cycle having length manner similar to Case 2, we have
at least IV(C)I + I, which is impossible. Let 5 = {W,Ul'UZ'" "Uk}' Since
151 = k + 1 > f3( G) and WUj ¢ E(G) for i = 1,2, ... , k, there are integers j n ::;:; deg U + deg v ::;:; r + s - 1 + (t - t') + (5 - s')
and t such that UjUt E E(G). Thus by deleting the edges VjUj and VtUt = (r + 5 + t) - (s' + t') + (s - 1)
from C and adding the edge UjUt together with the paths Pj and Pt , we
obtain a cycle of G that is longer than C. This produces a contradiction, = n -;- f3(G) + ",(G) -1;
so that C is a hamiltonian cycle of G. 0 so {3(G) ::;:; K(G) -1 < K(G).
Our next result, due to Bondy [Bl1], shows, perhaps somewhat sur- Case 4. Assume that R' i= 0 and T i= 0. Let u E R' and vET'. Then
prisingly, that Ore's theorem (Theorem4.13) follows as a corollary to
Theorem 4.22. n ::;:; deg U+ deg v ::;:; (r - r' ) + (5 - 5') + (t - t' ) + (5 - 5')
= (r + s + t) - (r' + 5' + t') -+ (5 - 5')
Theorem 4.23
=n - f3(G) + K(G) - 5'.
Let G be a graph of order n ~ 3 such that for all distinct nonadjacent vertices U

and v, Thus f3(G) ::;:; K(G) - s' ::;:; II:(G). 0

degu + degv ~ n.
Theorem 4.13 has been extended in other ways. The next result, due to
Then I'i:(G) ~ f3(G). Bondy [BI2], illustrates this.

Proof
Theorem 4.24
If G is complete, then clearly I'i:(G) ~ f3(G). Thus we may assume that Gis
not complete. Let 5 be a set of vertices of G such that G - 5 is disconnected If G is a 2-connected graph of order n such that
and 151 = I'i:(G). Let R be the vertex set of one of the components of G - 5
3n
and define T = V(G) - (5 u R). Let U be an independent set of vertices of degu+degv+degw ~ 2"
G with lUI = f3(G), and let
g=Rn~ ~=5n~ T=Tn~ for every set {u, v, w} of three independent vertices of G, then G is hamiltonian.
Finally, set
As we have already remarked, obtaining an applicable characterization
r= IRI, s= 151, t= ITI, r' = IR'I, 5' = 15'1', t' = IT'I· of hamiltonian graphs remains an unsolved problem in graph theory. In
We proceed by cases. view of the lack of success in developing such a characterization, it is not
surprising that special subclasses of hamiltonian graphs have been
Case 1. Assume that R' = T' = 0. Then U= 5' s:;; 5; so lUI::;:; lSI, that is, singled out for investigation as well as certain classes of nonharniltonian-
f3(G) ::;:; II:(G). graphs. We now discuss several types of 'highly hamiltonian' graphs and
Case 2. Assume that R' = 0 and T' = 0. Let U E R' and vET. Then then briefly consider graphs that are 'nearly hamiltonian'.
uv ¢ E(G); so degu + degv ~n. Furthermore, N(u) s:;; (R - R') u (5 - 5') A path in a graph G containing every vertex of G is called a hamiltonian
ahd N(v)S:;;SU(T-{v}); so degu::;:;(r-r')+(s-s') and degv::;:; path. A graph G is hamiltonian-connected if for every pair u, v of distinct
5 +t - 1. Therefore vertices of'G, there exists a hamiltonian u-v path. It is immediate that a
hamiltonian-connected graph with at least three vertices is hamiltonian.
n ::;:; degu + degv::;:; (r - r') + (5 - 5') + s + t - l
We define the (n + I)-closure Cn + 1 (G) of a graph G of order n to be the
= (r + s + t) - (r' + 5') + (s -1) graph obtained from G by recursively joining pairs of nonadjacent
vertices whose degree sum is at least n + 1 until no such pair remains.
=n - f3( G) + 11:(G) - 1.
We then have the following analogue to Theorem 4.17, also due to
It follows that f3(G) ::;:; I'i:(G) -1 < II:(G). Bondy and Chvatal [BC3l.
114 Eulerian and hamiltonian graphs and digraphs Hamiltonian graphs and digraphs 115
Theorem 4.25 A connected graph G of order n is said to be panconnected if for each pair
u, v of distinct vertices of G, there exists a u-v path of length t for each
Let G be a graph of order n. If CIl+I(G) is complete, then C is hamiltonian- t satisfying d(u, v) ~ t ~ n -1. If a graph is panconnected, then it is
connected. hamiltonian-connected; the next example indicates that these concepts
are not equivalent.
Proof For k~ 3, let Ck be that graph such that V(G k ) = {VI, Vz, ... , V2k} and

If n = 1, then the result is obvious; so we assume that n ~ 2. Let C be a E(C) = {ViVi+Ili = 1,2, ... ,2k} U {vivi+3Ii = 2,4, ... ,2k - 4}
graph of order n whose (n + I)-closure is complete, and let u and v be any U {VIV3, V2k-ZVZk},
two vertices of C. We show that G contains a hamiltonian u-v path, which
will then give us the desired result. where all subscripts are expressed modulo 2k. Although for each pair u, v
Define the graph H to consist of C together with a new vertex wand the of distinct vertices and for each integer t satisfying k ~ t ~ 2k - 1, the
edges uw and vw. Then H has order n + 1. Since CIl+I(G) is complete, graph Gk contains a u-v path of length t, there is no VI-v2k path of
(V(C))C(H) = KIl • Thus degC(H) x ~ n -1 for x E V(G); so length t if 1 < t < k. Since d( VI' V2k) = 1, it follows that Ck is not pan-
connected.
degC(H) x + degC(H) w ~ n + l. A sufficient condition for a graph G to be panconnected, due to
Therefore, C(H) = K Il + I and by Theorem 4.17, H is hamiltonian. Any Williamson [W7], can be given in terms of the minimum degree of C,.
hamiltonian cycle of H necessarily contains the edges uw and vw, imply-
ing that G has a hamiltonian u-v path. 0 Theorem 4.29

Two immediate corollaries now follow, the first of which is due to If C is a graph of order n ~ 4 such that degv ~ (n + 2)/2 for every vertex v of
Ore [03]. C, then G is panconnected.

Proof
Corollary 4.26
If n = 4, then C = K4 and the statement is true.
If C is a graph of order n such that for all distinct nonadjacent vertices u and v, Suppose that the theorem is not true. Thus there exists a graph G of
degu+degv ~ n +1, order n ~ 5 with 8( G)~ (n + 2) /2 that is not panconnected; that is,
there are vertices u and v of C that are connected by no path of length
then G is hamiltonian-connected. t for some t satisfying d(u, v) < t < n -1. Let C* :=: C - {u, v}. Then
C* has order n* = n - 2 ~ 3 and 8(C*) ~ (n + 2)/2 - 2 = n· /2. Therefore
Corollary 4.27 by Corollary 4.19, the graph C* contains a hamiltonian cycle
C: VI' Vz,.··, V n·, VI'
If C is agraph oforder n such that deg v ~ (n + 1)/2 for every vertex v of G, then If UVi E E(G), 1 ~ i ~ n*, then VVi+(-Z f/- E(C), where the subscripts are
C is hamiltonian-connected. expressed modulo n*; for otherwise,

A number of other sufficient conditions for a graph to be harniltonian-

connected can be deduced from Theorem 4.25.· One of these is the is a u-v path of length t in C. Thus for each vertex of C tha,t is adjacent
analogue to Theorem 4.18 (see [B8, p.218]). with u in C; there is a vertex of C that is not adjacent with v in G. Since
degG u ~ (n + 2) /2, we conclude that u is adjacent with at least n/2
Corollary 4.28 vertices of C, so

Let G be a graph of order n ~ 3, the degrees di of whose vertices satisfy d egGv~l+n * --=--l.
n n
2 2
d1 ~ d2 ~ •.. ~ dn• If there is no value of k ~ n/2 for which dk ~ k and
dn _ k ~ n - k, then G is hamiltonian-connected. This, however, produces a contradiction. 0
116 Eulerian and hamiltonian graphs and digraphs Hamiltonian graphs and digraphs 117
The result presented in Theorem 4.29 cannot be improved in general. Suppose that n is odd. Then by (4.3) there is some vertex, say VII' such
Let n = 2k + 1 ;?: 7, and consider the graph Kk,k+ 1 with partite sets V 1 and that deg VII ~ (n - 1)/2. But then
V z , where 1V11 = k on IVzl = k+ 1. The graph G is obtained from K k ,k+1 ,,-1
by constructing a path Pk-1 and k -1 vertices of V 2• Join the remaining
two vertices x and y of V 2 by an edge. Then degv ;?: (n + 1)/2 for every
2m = L degvi + degv"
j=1
vertex v but G is not panconnected since G contains no x-y path of
length 3. n(n-l) (n-l) nZ
~ 2 +--2-<2'
A graph G of order n ;?: 3 is called pancyclic if G contains a cycle of
length t for each t satisfying 3 < { < n. We say that G is vertex-pancyclic which contradicts the fact that m;?: n Z/4. Thus n is even and 2m =
if for each vertex v of G and for every integer { satisfying 3 < { < n, 2:;'=1 degvj ~ nZ/2, so that m ~ n2 /4. Since m ~ n2 /4, we have that
there is a cycle of G of length {that contains v. Certainly every pancyclic m = n Z /4. This implies that equality is attained in (4.3) for each j. Therefore,
graph is hamiltonian, as is every vertex-pancyclic graph,. although the
VjVk E E(G) if and only if Vj+1Vk+2 r¢ E(G), k f: j -1,j. (4.4)
converse is not true. The next theorem, due to Bondy [BID], gives a
Z
sufficient condition for a hamiltonian graph to be pancyclic. In order Suppose that G f: K,,/z,n/z. Since m = n /4, by Exercise 1.24, G has an
to present a proof due to C. Thomassen, a preliminary definition is odd cycle. This implies that G contains an outer cycle of 9dd length. Let
useful. Vj, Vj+ 1,' .. , Vj+t, Vj be a shortest outer cycle of G of odd length {+ 1
Let G be a hamiltonian graph and C: V1, Vz,"" v"' V1 a hamiltonian where, then, { is even and 4 < { ~ n - 4 since G contains no (n -1)-
cycle of G. With respect to this cycle, every edge of G either lies on C or cycle. Since VjVj+t E E(G), by (4.4), Vj-1VjH-2 r¢ E(G). Then, again by
joins two nonconsecutive vertices of C and is referred to as a chord. Any (4.4), Vj-ZVjH-4 EE(G). Therefore Vj_2, Vj_1,"" Vj+(-4, Vj_Z is an outer
cycle of G containing precisely one chord is an outer cycle of G (with cycle of (odd) length { - 1, which is a contradiction. Thus G = Kn/ z,n/2'
respect to the fixed hamiltonian cycle C). We now show by induction on n that if G is a hamiltonian (n, m) graph,
where m ~ n2 / 4, then either G is pancyclic or n is even and G = K,,/2,n/2' If
n = 3, then G = C3 and G is pancyclic. Assume for all hamiltonian graphs
Theorem 4.30 H of order n -1(~3) with at least (n -1)2/4 edges that either H is pan-
If G is a hamiltonian (n, m) graph where m ;?: n2 / 4, then either G is pancyclic or n cyclic or n -1 is even and H = K(n-1)/z,(n-1)/2' Let G be a hamiltonian
2
is even and G = K Il / 2,1l/2' (n, m) graph with m;?: n /4. Assume that either (0 n is even and
G f: K n / 2,n/2 or (ii) n is odd. We show that G is pancyclic. Under these
assumptions, it follows from the first part of the proof that G contains
Proof an (n - I)-cycle C*: W1, W2,"" W n "':1> WI' Let W be the single vertex of G
not on Co.
We first show that if G is a hamiltonian (n, m) graph, where n;?: 4
If deg W ;?: n /2, then for each integer { satisfying 3 ~ { ~ n, the vertex W
and m ~ nZ/4, and G contains no (n - 1)-cycle, then n is even and
lies on an {-cycle of G; otherwise, whenever WWj E E(G), 1 ~ i ~ n - 1, it
G =K Il / 2,1l/2'
follows that WWt r¢ E(G), where t == i + t ~ 2 (mod n -1). This, however,
Let C: Vb vz, ... , VII' V1 be a hamiltonian cycle of G and let Vj and Vj+ 1 be
any two consecutive vertices of C (where all subscripts are expressed implies that degw < (n -1)/2, which is a contradiction. Thus G is
modulo n). If 1 ~ k ~ n but k f: j - 1 and k f: j, then at most one of VjVk pancyclic if degw ~ n/2.
If deg W < n/2, then G - W is a hamiltonian graph of order n - 1 with at
and Vj+1Vk+2 is an edge of G; otherwise,
least nZ/4 - (n -1)/2 edges. Since n2 /4 - (n -1)/2 > (n -1)2/ 4, it fol-
lows that G - W f: K(n -1)/2,(n -1)/2' Applying the inductive hypothesis,
we conclude that G - W is pancyclic. Thus G is pancyclic and the proof
is an (n - I)-cycle of G. Thus for each of the degvj - 1 vertices in is complete. 0
V (G) - {Vj _ b Vj} that is adjacent to Vj' there is a vertex in
V(G) - {Vj+1,Vj+Z} that is not adjacent to Vj+1' Thus degvj+1 ~ If the sum of the degrees of each pair of nonadjacent vertices of a
(n - 2) - (degvj - 1) + 1, so graphG is atleast n, where n = W(G)I ~ 3, then by Theorem 4.13, G is
hamiltonian. Our next result shows that the condition of Theorem 4.13
degvj + degVj+1 ~ n. (4.3) actually implies much more about the cycle structure of G.
118 Eulerian and hamiltonian graphs and digraphs Hamiltonian graphs and digraphs 119
Corollary 4.31 cycle is called a hamiltonian cycle. As with hamiltonian graphs, no
characterization of hamiltonian digraphs exists. Indeed, if anything, the
Let G be a graph of order n ;;;: 3 such that for all distinct nonadjacent vertices u situation for hamiltonian digraphs is even more complex than it is for
and v, hamiltonian graphs. While there are sufficient conditions for a digraph
deg u + deg v ;;;: n. to be hamiltonian, they are analogues of the simpler sufficient conditions
Then either G is pancyclic or n is even and G = Kil / 2,il/2' for hamiltonian graphs.
Recall that a digraph D is strong if for every two distinct vertices u and v
of D, there is both a u-v (directed) path and a v-u path. Clearly, every
Proof hamiltonian digraph is strong (though not conversely).
Let G have size m. We need only show that m;;;: n2 /4, since G is hamilto- Because of the difficulty of the proof, we state without proof the
nian by Theorem 4.13. Let k be the minimum degree among the vertices of following theorem of Meyniel [M7] that gives a sufficient condition for
G.1f k;;;: n/Z, then clearly m ;;;: n2 /4. Thus we may assume that k < n/Z. a digraph to be hamiltonian. It should remind the reader of Ore's theorem
Let t denote the number of vertices of G of degree k. These t vertices (Theorem 4.13).
induce a subgraph H that is complete; for if any two vertices of H were
not adjacent, then there would exist two nonadjacent vertices the sum of
Theorem 4.32
whose degrees would be less than n. This implies that t :::;; k + 1. How-
ever, t i= k + 1; for otherwise, each vertex of H is adjacent only to vertices If D is a strong nontrivial digraph of order n such that for every pair u, v of
of H, which is impossible since G is connected. distinct nonadjacent vertices,
Let u be a vertex of degree k. Since t :::;; k, one of the k vertices adjacent to
u has degree at least k + 1, while each of the other k - 1 vertices adjacent
degu + degv;;;: 2n -1,
to u has degree at least k.lfw i= uis one of the n - k - 1 vertices of G that is then D is hamiltonian.
not adjacent to u, then deg w + deg u ;;;: n, so that deg w ;;;: n - k. Hence,
1 1 Theorem 4.32 has a large number of consequences. We consider these
m=Z L degv;;;:z[(n-k-1)(n-k)+~+k+1J now, beginning with a result originally discovered by Woodall [WI0].
VEV(G)

1 ~,2
2
. 2 n +1
=Z[LJ\+(Z-Zn)k+(n -n+l)J~-4-' Corollary 4.33

the last inequality holding since for k:::;; (n -1)/2, the expression
If D is a nontrivial digraph of order n such that whenever u and v are distinct
H2~ + (2 - 2n)k + (n 2 - n + 1)J takes on its minimum value when vertices and (u, v) f/. E(D),
k = (n - 1)/2. 0 odu +idv;;;: n, (4.5)

It is interesting to note that many other conditions that imply that a then D is hamiltonian.
graph is hamiltonian have been shown to imply that either the graph is
pancyclic or else belongs to a simple family of exceptional graphs.
Proof
We next briefly consider graphs that are, in certain senses, 'nearly
hamiltonian'. Of course, ifG is hamiltonian, then G has a hamiltonian First we show that condition (4.5) implies that D is strong. Let u and v be
path. Sufficient· conditions for a graph to possess a hamiltonian path any two vertices of D. We show that there is a u-v path in D. If
can be obtained from the sufficient conditions for a graph to be hamilto- (u, v) E E(D), then this is obvious. If (u, v) fj. E(D), then by (4.5), there
nian. For example, suppose that G is a graph of order n ;;;: 2 such that for must exist a vertex w in D, with w i= u, v, such that (u, w), (w, v) E E(D).
all distinct nonadjacent vertices u and v, we have deg u + degv ;;;: n - 1. However, then, u, to, v is the desired u-v ·path.
Then the graph G + K1 satisfies the hypothesis of Theorem 4,13 and so is To complete the proof we apply Meyniel's theorem. Let u and v be any
hamiltonian. This, of course, implies that G contains a hamiltonian path. two nonadjacent vertices of D. Then by (4.5), od u + id v ~ nand
We close this section with a brief discussion of hamiltonian digraphs. A od v + id u ~ n so that deg u + deg v ~ 2n. Thus; by Theorem 4.32, D is
digraph D is called hamiltonian if it contains a spanning cycle; such a hamiltonian. 0
120 Eulerian and hamiltonian graphs and digraphs Hamiltonian graphs and digraphs 121
The following well-known theorem is due to Ghouila-Houri [G3]. The 4.17 Show that Theorem 4.13 is sharp, that is, show that for infinitely
proof is an immediate consequence of Theorem 4.32. many n ;;, 3 there are nonhamiltonian graphs G of ordern such
that deg u + deg v ;;, n - 1 for all distinct nonadjacent vertices u
Corollary 4.34 and v.

If D is a strong digraph such that degv ~ n for every vertex v of D, then D is 4.18 Show that Theorem 4.20 is sharp, that is, show that for infinitely
hamiltonian. many n;;' 3 there are 2-connected nonhamiltonian graphs G of
ordern such that IN(u) UN(v)1 ;;, [(2n -1)/3] -1 for all distinct non-
This result also has a rather immediate corollary. adjacent vertices u and v.
4.19 Show that Theorem 4.22 is sharp, that is, show that for infinitely
Corollary 4.35 many n ;;, 3 there are nonhamiltonian graphs G of order n such
that K(G) ;;, f3(G) -1.
If D is a digraph such that
od v ~ n/2 and id v;;, n/2 4.20 Let G be an (n, m) graph, where n ;;, 3 and m ;;, ("2 1) + 2. Prove that
G is hamiltonian.
for every vertex v of D, then D is hamiltonian.
4.21 Let G be a bipartite graph with partite sets U and W such that
A spanning path in a digraph D is called a hamiltonian path of D. As in lUI = IWI = k;;, 2. Prove that if degv > k/2 for every vertex v of G,
the case of graphs, sufficient conditions for hamiltonian cycles in digraphs then G is hamiltonian.
often have easily obtainable analogues for hamiltonian paths (Exercise 4.22 Let G be a graph of order n ~ 2, the degrees di of whose vertices
4.33). Readers interested in more information on hamiltonian graphs satisfy d1 ~ dz ~ ... ~ dn • Show that if there is no value of
and digraphs should consult the surveys in Faudree [F2], Gould [G7] k< (n+l)/2 for which dk~k-l and dn - k+1 ~n-k-l, then G
and Lesniak [L2]. has a hamiltonian path.
4.23 Show that if G is a graph with at least two vertices for which
K(G) ;;, f3(G) - 1, then G has a hamiltonian path.
EXERCISES 4.2
4.24 Show that if G is a connected graph of order n ;;, 2 such that for all
4.11 Show that the graph of the dodecahedron is hamiltonian. distinct nonadjacent vertices u and v
4.12 (a) Show that if G is a 2-connected graph containing a vertex that (2n -2)
IN(u) U N(v)1 ;;, 3 '
is adjacent to at least three vertices of degree 2, then G is not
hamiltonian. then G has a hamiltonian path.
(b) The subdivision graph S(G) of a graph G is that graph obtained
from G by replacing each edge uv of G by a vertex wand 4.25 Show that if G is a k-connected graph (k ;;, 2) of order n ;;, 3 such that
edges uw and vw. Determine, with proof, all graphs G for' for every independent set {U1' uz, ... , Uk} of k vertices,
which S(G) is hamiltonian. k(n -1)
IN(U1) U N(uz) U· .. U N(uk)1 > (k + 1) ,
4.13 Give an example of a I-tough graph that is not hamiltonian.
4.14 (a) Prove that Kr ,Zr,3r is hamiltonian for every positive integer r. then G is hamiltonian.
(b) Prove that Kr ,zr,3r+1 is hamiltonian for no positive integer r.
4.26 Show that there exists a function f with f(n) < 3n/2 such that the
4.15 (a) Prove that if G and H are hamiltonian graphs, then G x H is following result is true: If G is a connected graph of order n such
hamiltonian. that degu+degv+degw~f(n)'forevery set {u,v,w} of three
(b) Prove that the n-cube Qn, n ;;, 2, is hamiltonian. independent vertices of G, then G has a hamiltonian path.
4.16 Give a direct proof of;Corollary 4.19 (without using Theorems 4.13 4.27 Show that if G is an (n,m) graph, where n ~ 4 and m ~ (n 21) + 3,
or 4.18). then G is hamiltonian~connected.
122 Eulerian and hamiltonian graphs and digraphs Line graphs and powers of graphs 123
4.28 Prove that every hamiltonian-connected graph of order 4 or more is C3 is hamiltonian. It is true, in fact, that the cube of every connected graph
3-connected. of order at least 3 is hamiltonian. Indeed, a stronger result exists, discov-
4.29 Give an example of a graph C that is pancyclic but not panconnected. ered independently by Karaganis [Kl] and Sekanina [51].

4.30 Prove or disprove: If C is any graph of order n ~ 4 such that for all
distinct nonadjacent vertices u and v, Theorem 4.36
deg u + deg v ~ n + 2, If C is a connected graph, then C3 is hamiltonian-connected.
then C is panconnected.
Proof
4.31 Let C be a connected graph of order nand let k be an integer such 3
that 2 :0:;; k :0:;; n - 1. Show that if deg u + deg v ~ k forevery pair u, v If H is a spanning tree of C and H 3 is hamiltonian-connected, thenC is
of nonadjacent vertices of C, then C contains a path of length k. hamiltonian-connected. Hence it is sufficient to prove that the cube of
every tree is hamiltonian-connected. To show this we proceed by induc-
4.32 Prove Corollaries 4.33 and 4.34. tion on n, the order of the tree. For small values of n the result is obvious.
4.33 State and prove an analogue to Theorem 4.32 which gives a sufficient Assume for all trees H of order less than n that H 3 is hamiltonian-
condition for a digraph to have a hamiltonian path. Show that the connected, and let T be a tree of order n. Let u and v be any two vertices
bound in this result is sharp. of T. We consider two cases.
Case 1. Suppose that u and v are adjacent in T. Let e = uv, and consider the
forest T - e. This forest has two components, one tree Tu containing u and
the other tree Tv containing v. By hypothesis, T~ and T~ are hamiltonian-
4.3 LINE GRAPHS AND POWERS OF GRAPHS
connected. Let Ul be any vertex of Tu adjacent to u, and let VI be any vertex
of Tv adjacent to v. If T u or Tv is trivial, we define Ul = U or VI = v,
The kth power C of a graph C, where k ~ I, is that· graph with
k
k respectively. Note that Ul and VI are adjacent in y3 since dr (Ul' VI) :0:;; 3.
V(C ) = V(C) for which uv E E(d) if and only if 1 :0:;; dc(u, v) :0:;; k. The
Let Pu be a hamiltonian U-Ul path (which may be trivial) of T~, and let
graphs C2 and C3 are also referred to as the square and cube, respectively,
Pv be a hamiltonian VI-V path of T~. The path formed by begi.nriing
of C. A graph with its square and cube are shown in Figure 4.10.
with P u and then follOWing with the edge Ul VI and the path Pv is a
Since the kth power Ck (k ~ 2) of a graph C contains C as a subgraph (as
hamiltonian u-v path of T 3 .
a proper subgraph if C is not complete), it follows that d is hamiltonian if
C is hamiltonian. Whether or not C is hamiltonian, for a connected graph Case 2. Suppose that u and v are not adjacent in T. Since T is a tree, there exists
C of order at least 3 and for a sufficiently large integer k, the graph c" is a unique path between every two of its vertices. Let P be the unique u-v
hamiltonian since Cd is complete if G has diameter d. It is therefore natural path of T, and let f = uw be the edge of P incident with u. The graph
to ask for the minimum k for which d is hamiltonian. Certainly, for T - f consists of two trees, one tree Tu containing u and the other tree Tw
connected graphs in general, k = 2 will not suffice since if G is the containing w. By hypothesis, there exists a hamiltonian w-v path Pw in T~.
graph of Figure 4.10, then CZ is not hamiltonian. We see, however, that Let Ul be a vertex of Tu adjacent to u, orlet Ul = uif Tu is trivial, and let Pu
be a hamiltonian U-Ul path in T~. Because dr (ull w) :0:;; 2, the edge ulw is
present in T 3 • Hence the path formed by starting with Pu and then follow-
ing with Ul wand Pw is a hamiltonian u-v path of T 3 . 0

It is, of course, an immediate corollary that for any connected graph C

of order at least 3, C3 is hamiltonian.
Although it is not true that the squares of all connected graphs of order
at least 3 are hamiltonian, it was conjectured independently by C. Nash-
Williams and M. D. Plummer that for 2-connected graphs this is the case.
In 1974, Fleischner [FSl proved the conjecture to be correct. Its lengthy
Figure 4.10 A graph whose square is not hamiltonian. proof is not included.
124 Eulerian and hamiltonian graphs and digraphs Line graphs and powers of graphs 125
Theorem 4.37 Proof
Z
If G is a 2-connected graph, then G is hamiltonian. Since (G)z f::. K n , either G is disconnected or G is connected with
diam G;;;, 3. If G is disconnected, then G is connected and diam G :::; 2
A variety of results strengthening (but employing) Fleischner's work (Exercise 1.29). Thus GZ is complete and therefore hamiltonian-connected.
have since been obtained; for example, it has been verified [CHIKNl] Thus we may assume that G is connected with diam G ;;;, 3. This implies
that the square of a 2-connected graph is hamiltonian-connected. that n;;;, 4 and that there are vertices Uv Uz E V(G) such that
dC(Ul, uz) = 3. __
For i = 1,2, define Vi = {v E V(G)IUiV E E(G)}. Since dc(Ul, uz) = 3, it
Theorem 4.38 follows that Vi =1= 0 (i = 1,2) and VI n V z = 0· Furthermore, UI ¢ Vz
and Uz ¢ VI' We consider two cases. C

If G is a 2-connected graph, then GZ is hamiltonian-connected.

Case 1. Assume that VI U V2 = V(G) - {UI'UZ}' If every vertex of VI is
adjacent in Gto every vertex of Vz, then (el = Kn - e, which is a contra-
Proof diction. Thus there are vertices VI E VI and Vz E V z such that VIVZ ¢ E(G).
Since G is 2-eonnected, G has order at least 3. Let U and v be any two Let FI denote the graph with V(F I ) = V(G) and
vertices .of G. Let Gll Gz, ... , Gs be five distinct copies of G and let Ui E(FI ) = {UIUZ, VIVZ} U {UIWZ!WZ E V z } U {uzwIlwI E VJl.
and Vi (I = 1,2, ... ,5) be the vertices in Gi corresponding to U and v in
Then FI ~ G and (F I )z is hamiltonian-connected. Since V(F I ) = V(G), it
G. Form a new graph F by adding to GI U Gz U ••. U Gs two new vertices
follows that GZ is hamiltonian-connected.
WI and Wz and ten new edges WIUi and WZVi (i = 1,2, ... ,5). Clearly,
neither WI nor Wz is a cut-vertex of F. Furthermore, since each graph Gi Case 2. Assume that VI U V z =1= V(G) - {Ull uz}. Select an arbitrary vertex
is 2-connected and contains two vertices adjacent to vertices in
Vo E V(G) - {u}, uz} - VI - V z·
V(F) - V(G i), no vertex of Gi is a cut-vertex of F. Hence F is a 2-eonnected
graph with at least three vertices and so, by the aforementioned result of Let Fz denote the graph with V(F z ) = V(G) and
Fleischner, FZ has a hamiltonian cycle C.
E(Fz ) = {VOUI, UIUZ, uzvo} U {UZWI!WI E VI}
Since each of WI and Wz has degree 2 in C, one of the subgraphs Gi of F,
say Gb contains no vertex adjacent to either WI or Wz in C. Suppose that U{UIWZ!WZ E V(G) - {UI,UZ} - VI}'
WE V(G k) - {Uk, Vk} and w' E V(F) - V(G k). If WW' E E(FZ ), then necessa-
rily w' = WI or uf = Wz. Therefore, since no vertex of Gk is adjacent to Then Fz ~ G and (Fd is hamiltonian-connected. Thus, since V(F z) =
either WI or Wz in C, it must be the case that Uk and Vk are the only vertices V(G), we see that G is also hamiltonian-connected. 0
of Gk that are adjacent in C to vertices in V(F) - V(G k). This implies that
one of the Uk-vk paths P determined by C contains exactly the vertices of Corollary 4.40
Gk· The proof will be complete once we have shown that E(P) ~ E(G~).
Let x,y E V(G k ) such that xy E E(P). Hence dF(x,y) :::; 2. From the way Let G be a graph with G f::. P4 • Then either GZ or (G)z is hamiltonian-connected.
in which F was constructed, we conclude that dck(x,y) :::; 2. Therefore
xy E E(G~) and P is a hamiltonian Uk - Vk path of Gt. 0 Corollary 4.41
Let G be a graph of order at least 3. Then either GZ or (Gl is hamiltonian.
Our next result, due to Nebesky [N4], gives another condition under
which.Gz is hamiltonian-connected. Recall thatKn - e denotes the graph . There is an interesting relationship between Fleischner's result
obtained by deleting an edge from the complete graph of order n. (Theorem 4.37) and Chvatal's conjecture that every 2-tough graph is
hamiltonian. It is not difficult to show (Exercise 4.39) that t(GZ) ;;;, K(G)
Theorem 4.39 for every graph G. Thus a proof of Chvatal's conjecture would imply
Fleischner's result.
Let G be. a graph of order n. If (G)2 f::. Kn and (G)2 f::. Kn - e, then CZ is We have seen that for every graph C and positive integer k we can
hamiltonian-connected. determine a new graph, the kth power Ck of C. There are other associated
 Line graphs and powers of graphs 127
126 Eulerian and hamiltonian graphs and digraphs

A~~
"lo-----oUZ

G: L(G):

Figure 4.11 A graph and its line graph.

graphs of interest. The line graph L(G) of a graph G is that graph whose
vertices can be put in one-to-one correspondence with the edges of G
~~+
~t?1*
in such a way that two vertices of L(G) are adjacent if and only if the
corresponding edges of G are adjacent. A graph and its line graph are
shown in Figure 4.11.
It is relatively easy to determine the number of vertices and the number
of edges of the line graph L( G) of a graph G in terms of easily computed
quantities in G. Indeed, if G is an (n, m) graph with degree sequence
dll dz, ... ,dn and L(G) is an (n', m') graph, then n' = m and Figure 4.12 The induced subgraphs not contained in any line graph.

m' = t (d
i=1 2
i
), Theorem 4.43
Let G be a graph without isolated vertices. Then L(G) is hamiltonian if and only if
since each edge of L(G) corresponds to a pair of adjacent edges of G. G = K Il , for some t ~ 3, or G contains a dominating circuit.
A graph H is called a line graph if there exists a graph G such that
H = L(G). A natural question to ask is whether a given graph H is a
line graph. Several characterizations of line graphs have been obtained, Proof
perhaps the best known of which is a 'forbidden subgraph' characteri- If G = KI,t for some t ~ 3, then L(G) is hamiltonian since L(G) = K{.
zation due to Beineke [B4]. We present this result without its lengthy
Suppose, then, that G contains a dominating circuit
proof.

Theorem 4.42 It suffices to show that there exists an ordering S: el, e2, ... ,em of the m
A graph H is a line graph if and only if none of the graphs of Figure 4.12 is an edges of G such that ei and ei+l are adjacent edges of G, for
induced subgraph of H. 1 ~ i ~ m - I, as are el and em, since such an ordering S corresponds to a
hamiltonian cycle of L(G). Begin the ordering S by selecting, in any order,
We tum to the problem of determining the relationships between a all edges of G incident with VI that are not edges oK, followed by the edge
graph and hamiltonian properties of its line graph. Theorem 4.43, due VIVZ' At each 'Successive vertex Vi, 2 ~ i ~ t - I, select, in any order, all
to Harary and Nash-Williams [HNl], provides a characterization of edges of G incident with Vi that are neither edges of C nor previously
those graphs having hamiltonian line graphs. A set X of edges in a selected edges, followed by the edge. ViVi + 1- This process terminates
graph is called a dominating set if every edge of G either belongs to X or with the edge Vt-IVt. The ordering S is completed by adding the edge
is adjacent to an edge of X. If (X) is a circuit C, then C is called a dominating VtVI' Since C is a dominating circuit of G, every edge of G appears exactly
circuit of G. Equivalently, a circuit ein a graph G is a dominating circuit if once in S. Furthermore, consecutive edges of S as well as the first and last
every edge of G is incident with a vertex of C. edges of 5 are adjacent in G.
128 Eulerian and hamiltonian graphs and digraphs Line graphs and powers of graphs 129
Z
Conversely, suppose that G is not a star but L(G) is hamiltonian. We Earlier in this section we saw that for any graph G =1= P4, either G or
show that G contains a dominating circuit. Since L( G) is hamiltonian, (el is hamiltonian-connected (Corollary 4.40). Consequently, for every
there is an ordering S: el, ez,' .. ,em of the m edges of G such that ei and graph G of order at least 5 either GZ or (G)z is hamiltonian (Corollary
ei+l are adjacent edges of G, for 1 ~ i ~ m -1, as are el and em' For 4.41). It is perhaps surprising that there is a line graph analogue of
1 ~ i ~ m-l, let Vi be the vertex of G incident with both ei and ei+l' Corollary 4.40 but no such analogue of Corollary 4.41. In particular,
(Note that 1 ~ k -I q ~ m - 1 does not necessarily imply that Vk -I v q.) Nebesky [N2] showed that if G is a graph of order at least 5, then either
Since G is not a star, there is a smallest integer h exceeding 1 such that L(G) or L(G) is hamiltonian. Furthermore, Nebesky [N3] showed that for
vj, =1= VI' The vertex Vj, -1 is incident with ej,' the vertex vj, is incident with every positive integer n there is a graph G of order n such that neither L( G)
ej,' and vj, -I = VI' Thus, ej, = VI vj, . Next, let h (if it exists) be the smallest nor L(G) is hamiltonian-connected.
integer exceeding h such that viz ¥ vj, . The vertex viz -1 is incident with eiz' , In [T6] Thomassen conjectured that every 4-connected line graph is
the vertex viz is incident with eh' and viz -1 = vj,' Thus, eh = vj, viz. Contin- hamiltonian. This conjecture is related to Chvatal's conjecture that
uing in this fashion, we finally arrive at a vertex vJ'1 such that eJ•1 = vJ'(1-1) v'It , every 2-tough graph is hamiltonian as follows. As we saw in Chapter 3,
where Vj, = Vm_I' Since every edge of G appears exactly once in Sand for every graph G we have t(G) ~. K( G) /2 and equality holds if G is claw-
since 1 < h < h < .. , < h ~ m -1, this construction yields a trail free. Since a line graph does not contain K1 ,3 as an induced subgraph, it
follows that every 4-connected line graph is 2-tough. Thus the truth of
Chvatal's results would also imply the conjecture made independently
in G with the properties that (i) every edge of G is incident with a vertex of by Matthews and Sumner [MS1J that every 4-connected claw-free graph
T, and (li) neither el nor em is an edge of T. is hamiltonian.
Let W be the vertex of G incident with both el and em' We consider four
possible cases.
Case 1. Suppose that . w = VI = v m-l' Then T itself is a dominating circuit EXERCISES 4.3
ofG.
4.34 Show that the graph GZ of Figure 4.10 is not hamiltonian.
Case 2. Suppose that W = VI and W =1= vm-l' Since em is incident with both
wand vm-J, it follows that em = Vm_lW = Vm_lVI' Thus C: T,e m,VI is a 4.35 Prove that if v is any vertex of a connected graph G of order at least 4,
dominating .circuit of G. then G3 - v is hamiltonian.

Case 3. Suppose that W = Vm-1 and W =1= VI' Since el is incident with both w 4.36 Give an infinite faniily 9 of graphs such that for each G E g, neither G
and VI, we have that el = WVI = Vm_1 VI' Thus C: T, el, VI is a dominating nor G is 2-connected. (Thus, Corollaries 4.40 and 4.41 do not follow
circuit of G. from Theorems 4.37 and 4.38.)

Case 4. Suppose that w =1= Vm-l and W =I=. VI' Since em is incident with both 4.37 Prove Corollary 4.40.
wand Vm_ J, it follows that em = Vm-1 w. Since el is incident with both W
4.38 Prove that if G is a self-complementary graph of order at least 5, then
and VJ, we have that el = WVl' Thus VI =1= Vm-l, and C: T, em, w, el, VI is a GZ is hamiltonian-connected.
dominating circuit of G. D
4.39 (a) Let G be a graph, and let S be a vertex-cut of GZ• Further, let
It follows from Theorem 4.43 that if G is either eulerian or hamiltonian VI' V 2 , ••• , Vk be the vertex sets of the components of G2 - S.
then L(G) is hamiltonian. More generally,we have the following corollary: For each i = 1,2, ... , k, let
. Sj = {u E Siu is adjacent to a vertex of Vj in G}.
Corollary 4.44 Show that IS;! ~ K(G) for each i = 1,2, ... , k and that Si n Sj =0
IfG issupereulerian, then L(G) is hamiltonian. for 1 ~ i =1= j ~ k.
(b) Use (a) to show that t(G2 ) ~ /l,(G) for every graph G.
The strong interest in supereulerian graphs can, at least in part, be 4.40 Determine a formula for the number of triangles in the line graph
attributed to Corollary 4.44. L(G) in terms of quantities in G.
130 Eulerian and hamiltonian graphs and digraphs
4.41 Prove that L(G) is eulerian if G is eulerian. CHAPTER 5
4.42 Find a necessary and sufficient condition for a graph G to have the
property G = L(G).
4.43 (a) Show that if G is a connected graph with 8(G) ~ 3, then L(G) is Directed graphs
supereulerian. (Hint: Each vertex v in G corresponds to a com-
plete subgraph K(v) in L(G). Select a hamiltonian cycle from each
K(v) and use. this t~ build a spanning eulerian subgraph of L(G).)
(b) Show that If G IS a connected graph with 8(G) ~ 3, then
e(G) = L(L(G)) is hamiltonian.
(c) ~or ~tegers k ~ k2, the kth iterated line graph Lk(G) of a graph G We return to digraphs, first considering graphs for which some or all
~s defined as L(L -1(G)), where L1(G) denotes L(G) and Lk-I(G) orientations have a certain connectedness type. The main emphasis
IS ass~med to ~e nonempty. Show that for a connected graph G, here, however, is the study of tournaments.
some Iterated hne graph of G is hamiltonian if and only if G is not
a path.
4.44 For each of the following, prove or disprove. 5.1 CONNECTEDNESS OF DIGRAPHS
(a) If G is hamiltonian, then C2 is hamiltonian-connected.
(b) If G is supereulerian, then G2 is supereulerian. In Chapter 1 we described various types of connectedness that a digraph
(c) If G is connected and L(G) is eulerian, then G is eulerian. may possess. In this section we explore these in more detail. Recall that a
(d) If G is hamiltonian, then L(G) is hamiltonian-connected. digraph 0 is strong if for every pair u, v of vertices of 0, there is both a u-v
(e) If G has a dominating circuit, then L(G) has a dominating circuit. path and a v-u path. Strong digraphs are characterized in the following
theorem.

Theorem 5.1
A digraph 0 is strong if and only if 0 contains a closed spanning walk.

Proof
AssuII\e that W: UI, U2, ... , Uk> UI is a closed spanning walk in O. Let u,
v E V(O). Then u = Ui and v = Uj for some i,j with 1 ~ i, j ~ kand i:::/= j.
Without loss of generality, we assume that i <j. Then WI: Ui, Ui+l,"" Uj
is a Ui-Uj walk in 0 and W2 : Uj,Uj+I'"'' Uk, Ul,"" ui is a UrUi walk in O.
Consequently, 0 contains both a Ui-Uj path and a UJ-Ui path in O.
Conversely, assume that 0 is a nontrivial strong digraph. We show that
o contains a closed spanning walk. Suppose thjlt this is not the case, and let
W be a closed walk containing a maximurtfnumber of vertices of O. Let x
be a vertex of 0 that is not on W, and let v be a vertex on W. Since 0 is
strong, 0 contains a v-x path P1 and an x-v path P 2. When v is encountered
on W, we insert P 1 followed by P2 . This results in a closed walk W' contain-
ing more vertices of 0 than W, which produces a contradiction.Thus 0 is
strong. 0
Recall that a digraph 0 is unilateral if for every pair u, v of vertices of 0,
there is either a u-v path or a v-u path. Unilateral digraphs can be
characterized in much the same way as strong digraphs.
 132 Directed graphs Connectedness of digraphs 133
Theorem 5.2 Assign directions to the edges of H so that the resulting digraph D is
A digraph D is unilateral if and only if D contains a spanning walk. strong, but assign no directions to the edges of G - E(H). Let U E V(H)
and let v E V(G) - V(H). Since G is 2-edge-connected, there exist two
edge-disjoint (graphical) u-v paths in G. Let P be one of these u-v paths
Proof
and let Q be the v-u path that results from the other u-v path. Further, let
Suppose first that D contains a spanning walk W: V1, VZ, ... ,Vk' Let u and v U1 be the last vertex of P that belongs to H, and let V1 be the first vertex of Q
be distinct vertices of D. Then u = Vi and v = v· for some integers i and belonging to H. Next, let P1 be the UI-V subpath of P and let Q1 be the V-V1
j with 1 ,,; i,j"; k. Assume, without loss of ge~erality, that i < j. Then subpath of Q. Direct the edges of PI from Ul toward v, producing the
W1: U = Vi, vi+1,"" Vj = v is a u-v walk. Thus D contains a u-v path directed path P~, and direct the edges of Ql from v toward VI' producing
and so D is unilateraL the directed path Q~.
For the converse, assume that D is a unilateral digraph and suppose, to Define the digraph D' by V(D') = V(D) U V(~) U V(Q~) and E(D') =
the contrary, that D does not have a spanning walk. Let W: U1, UZ, ... 1 Ut E(D) U E(P~) U E(QD. Since D is strong, so is D', which contradicts the
be a walk containing a maximum number of vertices of D. Let x be a choice of H. 0
vertex of.D that is not on W. If D contains an X-U1 path P or a UrX path
Q, then el~her P followed by W or W followed by Q is a walk containing Not every connected graph has a unilateral orientation. For example,
more vertices than W, which produces a contradiction. Hence we may the only trees that have unilateral orientations are paths.
assume that D does not contain these paths. Consequently, D contains a
UI-X path and an X-Ut path. Hence there exists an integer i (1 ,,; i < t) Theorem 5.4
such that D contains a Ui-X path PI and an x-ui+1 path Pz. Denote the
UI-Ui subwalk of W by WI and the Ui+1-Ut subwalk of Wby Wz. Then A tree T ha5 a unilateral orientation if and only if T is a path.
W1 followed by P1 , Pz and Wz produces a walk containing more vertices
than W, which isa contradiction. 0
Proof
Rec~ll t~at an ~sy~metric digraph D can be obtained from a graph~ Clearly every path has a unilateral orientation. Next, assume that T is
by assIgnmg a dIrection to each edge of G and that D is also called an a tree that is not a path. Then T has a vertex v with deg V ;;:. 3. Now
or~entat~on of G. We are now interested in those graphs having a strong suppose that we assign directions to the edges of T producing a digraph
onentatlOn. Certainly, if G has a strong orientation, then G must be con- D. Then there are at least two edges of T directed toward or away from v
ne.cted. ~lso, if G has a bridge, then it is impossible to produce a strong in D; say e1 = VUl and ez == vUz are edges of T incident with v such that
onentation of G. On the other hand, if G is a bridgeless connected graph, ei = (v, Ui) in D for i = 1,2 or ej = (Ui, v) in D for i = 1,2. In either. case,
then G always has a strong orientation. This observation was first made by D contains neither a UI-UZ path nor a UZ-ul path. Hence D is not
Robbins [R9]. unilateraL 0

Theorem 5.3 Indeed, only those graphs G with the property that all of their
bridges lie on a single path of G have a unilateral orientation. In the
A nontrivial graph G has a strong orientation ifand only if Gis 2-edge-connected. following proof, we use graphs that are obtained by identifying sets of
vertices in a graph G. Let G be a graph and let 51,52 , ... , Sk be non-
Proof empty, pairwise disjoint subsets of V(G). The graph G' is obtained by
identifying the vertices of 5i for each i if V(G') = V(G - U7=1 5 i ) U
W~ have already observed that if a graph G has a strong orientation, then {51' 5Z, ... , 5k}' and the edges of G' consist of the edges of G - U7=1 Sj
G IS 2-edge-connected. Suppose that the converse is false. Then there together with the edges of the type XSi (1 ,,; i,,; k) whenever
exists a 2-edge-connected graph G that has no strong orientation. x E V(G - U7=1 Sj) is adjacent to some vertex of Sj and edges of the
Among the subgraphs of G, let H be one of maximum order that has a type 5i5j (1 ,,; i,j ,,; kj i =1= j) whenever some vertex of Si is adjacent to
strong orientation; such a subgraph exists since for each v E V(G), the some vertex of Sj' This is illustrated in Figure 5.1, where the vertices
s~bgraph ({V})tr~viany has a strong orientation. Thus IV(H)j < jV(G)I, of 51 == {VI, V2, V4, Vs} are identified, as are the vertices of 52 ==
Since, by assumptIon, G has no strong orientation. {V6, V7,VS, Vg, VlO}.
134 Directed graphs Connectedness of digraphs 135
r r r

G: G':

V12
Figure 5.1 Identification of vertices.
Figure 5.2 An anticonnected orientation of a graph.
Theorem 5.5
by reversing the directions of all those arcs directed toward a vertex v for
A connected graph G has a unilateral orientation if and only if all of the bridges of which the (directed) distance d(r, v) is even (see Figure 5.2 for an illustra-
G lie on a common path. tion). Then any orientation of G containing To is antiConnected.
Although there exists an anticonnected orientation of every connected
graph, it is certainly not the case that every orientation of every graph is
Proof
anticonnected. The follOWing result due to L. Eroh (see [CGSW1]) provides
By Theorem 5.4, the result holds if G is a tree. So let G be a connected a necessary condition for a graph to have this property.
graph that is not a tree. Denote the bridges (if any) of G by el, e2,' .. , er
and the maximal 2-edge-connected subgraphs of G by G1 ,G2 , ••• ,Gk •
Theorem 5.6
Let G' be the graph obtained from G by identifying the vertices of Gi for
each i = 1,2, ... , k. Then E(G') = {ell e2," ., erl, the set of bridges of G. By If every orientation ofa graph G of order at least 4 is anticonnected, then G is 2-
this construction, G' is a tree, and if G has no bridges, then G' == K1 • It connected and 6(G) ~ 3.
follows that G' is a path if and only if every bridgeei of G lies on a
common path P of G. Thus, by Theorem 5.4, it remains to show that G
Proof
has a unilateral orientation if and only if G' does.
Clearly, a unilateral orientation of G induces a unilateral orientation of Suppose thatG contains a cut-vertex w, and let B1 and B2 be distinct blocks
G'. For·the converse, suppose that G' has a unilateral orientation D. By containing w. Furthermore, let u(#w) be a vertex of B1 and v(#w) be a
Theorem 5.4, G' is a path and D is a directed path. Assign the same vertex of B2 • An orientation of G in which all edges of B1 incident with
directions to the edges ell e2,'''' er in G as they have in D. Further, by ware directed toward wand all edges of B2 incident with ware directed
Theorem 5.3, we may orient the edges of each sUbgraph Gi (1 ,,;;; i ,,;;; k) away from w contains no u~vantipath.Consequently, such an orientation
so that a strong subdigraph is obtained. This gives a unilateral orientation is not anticonnected. Thus Gis 2-connected.
ofG. 0 Suppose that there is a vertex v of G such that deg v = 2, with v adjacent
to u and w. Orientthe edges vu and vw as (v, u) and (w, v). For each vertex
Recall that for vertices u and v in a digraph D, a u~v antipath is a u~v x(#v) that is adjacentto u, or~ent ux as (u, x). Also, for every vertex y(# v)
semipath that contains no subpath of length 2, and that D is anticonnected that is adjacent to w, orient yw as (y, w). All other edges are oriented arbi-
if D contains a u~v antipath for every pair u, v of vertices of D. Every con- trarily. Let z be a vertex such that d(v, z) = 2. Then, there is no v-z antipath
nected graph has an anticonnectedorientation. In order to see this, let G in this orientation of G, contrary to assumption. 0
be a connected graph and let T' be a spanning tree of G and r a vertex of T'.
Let T be that directed tree obtained by orienting the edges of T' such that Since every orientation of a complete graph is anticonnected, it follows
all arcs of T are directed away from r. Now let To be that digraph obtained that if the degrees of the vertices of a graph G are sufficiently large, then
 136 Directed graphs Connectedness· of digraphs 137
every orientation of G is anticonnected. The precise bound on the degrees
is described next [CGSWl].

Theorem 5.7
If G is a graph of order n :?: 3 such that deg v :?: (3n - 1)/4 for every vertex v of
G, then every orientation of G is anticonnected.

Proof
Suppose, tn the contrary, that there exists a graph G satisfying the hypoth-
esis of the theorem but having an orientation V that is not anticonnected.
Then V contains two nonadjacent vertices x and y for which there is no x-y
antipath. Since each of x and y is adjacent to at least (3n -1)/4 vertices,
they are mutually adjacent to at least (n + 3)/2 vertices. Of these, x is
directed toward or away from at least (n + 3)/4 vertices, say the former.
Let 5 denote this set of vertices. Since V contains no x-y antipath, every Figure 5.3 An orientation of a graph that is notanticonnected.
vertex of S is directed toward y. Let w E S. Since degw :?: (3n - 1)/4, w is
In each case, the minimum degree of the graph is r(3n -1)/41 -1 and
adjacent to at least one vertex z of S. Assume, without loss of generality,
there exists an orientation that is not anticonnected.
that (w,z) is an arc of V. Then X,Z, w,y is an antipath, producing a
contradiction. 0
EXERCISES 5.1
Theorem 5.7 is best possible in the sense that for every integer n :?: 3,
. there exists a graph G of order n such that degv :?: r(3n -1)/41 -1 for 5.1 Let e = uv be an edge of a graph G. Show that if G has a strong orien-
every vertex v of G and there exists an orientation of G that is not
tation, then G has a strong orientation in which u is adjacent to v and
anticonnected. The vertex set of such a graph G consists of mutually
a strong orientation in which v is adjacent to u.
disjoint sets {x,y},A, B,C and V, where {x,y}, A and B are independent
and the subgraphs of G induced by both C and V are complete. The 5.2 Determine all graphs for which every orientation is unilateral.
vert~x x is joined to every vertex of A, Band C; while y is joined to 5.3 Let G be a connected graph with cut-vertices. Show that an orienta-
each vertex in A, Band D. Furthermore, each vertex of A is adjacent tion D of G is strong if and only if the subdigraph of Dinduced by the
to every vertex of B, C and V; while every vertex of B is adjacent to vertices of each block of G is strong.
every vertex of C and D. The orientation of G is described in Figure
5.4 Prove that a graph G has an eulerian orientation if and only if G is
5.3. For example, x is directed toward every vertex of A and away
eulerian.
from every vertex of B, while x is adjacent to every vertex of C with
the directions of the arcs between x and the vertices of C chosen arbitra- 5.5 Prove or disprove: If D is an anticonnected digraph and v is a vertex
rily. Also, every vertex of C is directed toward every vertex of A and that is added to V together with some edges joining v and some
away from every vertex of B. Since there is no x-y antipath, this orienta- vertices of D, then these edges may be oriented in such a way that
tion is not anticonnected. the resulting digraph is anticonnected.
Now, for a giveninteger n :?: 3, the cardinalities of A, B, C and Dare 5.6 Prove or disprove: If every orientation of a graph G of order at least 5
selecte~ as follows: .
is anticonnected, then 8(G) :?: 4.
1. n == 0 (mod 4): IAI = IBI = n/4,ICI =101= (n ,- 4)/4; 5.7 Show that if G is a graph of order n :?: 3 such that degu + degv:?:
2. n == 1 (mod 4): IAI = (n + 3)/4, IBI = (n -1)/4, ICI = IDI = (n - 5)/4; (3n - 1)/2 for every pair u, v of nonadjacent vertices of G, then
3. n == 2 (mod 4): IAI = IBI = (n + 2)/4, ICI = IDI = (n - 6)/4; every orientation oiG is anticonnected. .'
4. n == 3 (mod 4): IAI = (n + 1)/4, IBI = ICI =101 = (n - 3)/4.
138 Directed graphs Tournaments 139
5.2 TOURNAMENTS teams. In a round robin tournament, each team plays every other team
exactly once and ties are not permitted. Let V], Vz, ... ,Vn represent the
The class of oriented graphs that has received the greatest attention are teams as well as the vertices of the corresponding tournament T. If, in
the tournaments; that is, those digraphs obtained by orienting the edges the competition between Vi and vi' i 'I j, team Vi defeats team vi' then
of complete graphs. (Vi'Vi) is an arc of T. The number of victories by team Vi is the outdegree
The number of nonisomorphic tournaments increases sharply with of Vi' For this reason, the outdegree of the vertex Vi in a tournament is also
order. For example, there is only one tournament of order 1 and one of referred to as the score of Vi'
order 2. There are two tournaments of order 3, namely the tournaments T] A sequence 5],52, .•. ,Sn of nonnegative integers is called a score sequence
and T z shown in Figure 5.4. There are four tournaments of order 4, 12 of (of a tournament) if there eXIsts a tournament T of order n whose vertices
order 5, 56 of order 6, and over 154 billion of order 12. can be labeled V], V2, ... ,Vn such that ad Vi = Si for i = 1,2, ... ,n. The
If T is a tournament of order n, then it follows since T is complete that its following result describes precisely which sequences are score sequences
size is (~) and that, consequently, of transitive tournaments.

L
vEV(T)
od v = L
VEV(T)
id v = G)' Theorem 5.9
A tournament T is transitive if whenever (u, v) and (v, w) are arcs of T, then A nondecreasing sequence S of n(;::: 1) nonnegative integers is a score sequence
(u, w) is also an arc of T. The tournament Tz of Figure 5.4 is transitive of a transitive tournament of order n if and only if S is the sequence
while T1 is not. The following result gives an elementary property of 0,1, ... , n - 1.
transitive tournaments.
Proof
Theorem 5.8
First we show that S: 0,1, ... , n - 1 is a score sequence of a transitive
A tournament is transitive if and only if it is acyclic. tournament. Let T be the transitive tournament defined by V(T) =
{Vl> Vz, , v n} and E(T) = {(Vi, Vi)ll ::; j < i::; n}. Then od Vi = i-I for
Proof i = 1,2, , n; so S is a score sequence of a transitive tournament.
Conversely, assume that T is a transitive tournament of order n. We
Let T be an acyclic tournament and suppose that (u, v) and (v, w) are arcs show that S: 0,1, ... ,n -1 is a score sequence of T. It suffices to show
of T. Since T is acyclic, (w, u) 1. E(T). Therefore, (u, w) E E(T) and T is that no two vertices of T have the same score (outdegree). Let
transitive. u, V E V(T) and assume, without loss of generality, that (u, v) E E(T). If
Conversely, suppose that T is a transitive tournament and assume that W denotes the set of vertices of T adjacent from v, then od V = IWI.
T contains a cycle, say C: vI, vz,'''' vb vl (where k;::: 3 since T is Since (v, w) E E(T) for each wE Wand (u, v) E E(T), it follows that
asymmetric). Since (v], vz) and (vz, V3) are arcs of the transitive tourna- (u,w) E E(T) for each WE W, since T is transitive. Thus; odu;:::
ment T, (V],V3) is an arc of T. Similarly, (V]'V4)'(v],vs),· .. , (V],Vk) are 1 + IWI = 1 + ad v. 0
arcs of T. However, this contradicts the fact that (Vb v]) is an arc of T.
Thus, T is acyclic. 0
The proof of Theorem 5.9 shows that the structure of a transitive tour-
nament is uniquely determined.
Every tournament of order n can be thought of as representing or
modeling a round robin tournament involving competition among n
Coronary 5.10
For every positive integer n, there is exactly one transitive tournament of
order n.

Combining this corollary with Theorem 5.8, we arrive at yet another

Figure 5.4 The tournaments of order 3. corollary.
140 Directed graphs Tournaments 141
Corollary 5.11 and if Vi E Vi and Vi E Vi' where i > j, then (Vi, Vi) E E(T). Thisdecompo-
sition is often useful when studying the properties of tournaments that
For every positive integer n, there is exactly one acyclic tournament of order n.
are not strong.
We already noted that there are four tournaments of order 4. Of course,
Although there is only one transitive tournament of each order n, in a
one of these is transitive, which consists of four trivial strong components
certain sense that we now explore, every tournament has the structure of a
51 ,52 ,53 ,54, where the vertex of 5j is adjacent to the vertex of 5i if and only
transitive tournament. Let T be a tournament. We define a relation on
if j > i. There are two tournaments of order 4 containing two strong
V(T) by u is related to v if there is both a u-v path and .a v-u ~ath in ~.
components 51 and 521 depending on whether 51 or 52 is the strong com-
This relation is an equivalence relation and, as such, thIS relation parti-
ponent of order 3. (No strong component has order 2.) Since there are four
tions V(T) into equivalence classes VI, V 2 , ••. , V k (k;;:' 1). Let 5i = (V j )
tournaments of order 4, there is eactly one strong tournament of order 4.
for i = 1,2, ... ,k. Then each 5j is a strong subdigraph and, indeed, is max-
These tournaments are depicted in Figure 5.6. The arcs not drawn in Tv T2
imal with respect to the property of being strong. The subdigraphs
and T3 are all directed downward, as indicated by the 'double arrow.
51 ,52 , ••. ,5k are called the strong components of T. So the vertex sets of
We also stated that there are 12 tournaments of order 5. There are six
the strong components of T produce a partition of V(T).
tournaments oforder 5 that are not strong, shown in Figure 5.7. Again all
Let T be a tournament with strong components 51' 52, ... ,5k, and let
arcs that are not drawn are directed downward. Thus there are six strong
T denote that digraph whose vertices Ul, U2, ... , Uk are in one-to-one tournaments of order 5.
correspondence with the strong compone,!lts (u; corresponds to 5i ,
Theorem 5.9 characterizes score sequences of transitive tournaments.
i = 1,2, ... ,k) such that (Ui' Uj) is an arc of T, i =/= j, if and only if some
We next investigate score sequences in more generality. We begin with
vertex of 5 j is adjacent to at least one vertex of 5j • Since 5i and 5j are dis-
a theorem similar to Theorem 1.4.
tinct strong components o! T, it follows that every vertex of 5; is adjacent to
every vertex of 5·. Hence, T is obtained by identifying the -yertices of 5j for
i = 1,2, ... ,k.A tournament T and associated digraph T are shown in OS4
Figure 5.5. _ 0

~ VSZ
S2
OS3
Observe that for the tournament T of Figure 5.5, T is necessarily a
tournament and, in fact, a transitive tournament. That this always J!, OS2 J!,
occurs follows from Theorem 5.12 (Exercise 5.7). VS\
0 SI
OSI

Theorem 5.12 Tl T2

If T is a tournament with (exactly) k strong components, then T is the transitive Figure 5.6 The four tournaments of order 4.
tournament of order k.
Theorem 5.13
Since for every tournament T, the tournament T is transitive, it follows
that if T is a tournament that is not strong, then V (T) can be partitioned as
A nondecreasing sequence S: sl, S2,' •• , Sn (n ;;:, 2) of nonnegative integers
VI U V 2 U ... U V k (k;;:' 2) such that (Vi) is a strong tournament for each i,
is a score sequence if and only if the sequence SI: S11 S 2, ... ,S5n '
S5 n + 1 - 1, ... ,Sn -1 - 1 is a score sequence.

T: '1':
o
o
o
.~
o
o o
m
om
o

o o
T} T2 T3 T4 T
S
Figure 5.5 A tournament T and T. Figure 5.7 The six tournaments of order 5 that are not strong.
142 Directed graphs Tournaments 143
Proof 1
Assume that SI is a score sequence. Then there exists a tournament Tl of
order n - 1 such that SI is a score sequence of Tl . Hence the vertices of Tl
can be labeled as VI' V2, ... , V n -1 such that
~
1 1
Sj for 1 ,,;; i ,,;; Sn
odvj= {
sj-1 fori>sn'
2
We construct a tournament T by adding a vertex V n to Tl • Furthermore,
2 2
for 1 ,,;; i ,,;; n, V n is adjacent to Vj if 1 ,,;; i ,,;; Sn, and V n is adjacent from Vj
otherwise. The tournament T then has S as a score sequence.
For the converse, we assume that 5 is a score sequence. Hence there
T:
exist tournaments of order n whose score sequence is 5. Among all
such tournaments, let T be one such that V (T) = {VI' V2, .•. , vn },
od Vj = Sj for i = 1,2, ... , n, and the sum of the scores of the vertices adja-
cent from V n is minimum. We claim that V n is adjacent to vertices haVing 2 3
scores 51,52,"" S5.' Suppose to the contrary that V n is not adjacent to
vertices having scores 51, S2,' •• , S5,,' Necessarily, then, there exist vertices • 4
Vi and Vk with j < k and si < Sk such that V n is adjacent to Vb and V n is Figure 5.9 Construction of a tournament with a given score sequence.
adjacent from Vi' Since the score of Vk exceeds the score of Vi' there
exists a vertex VI such that Vk is adjacent to VI' and VI is adjacent to Vi
(Figure 5.8(a)). Thus, a 4-cycle C: V n , Vb VI> Vi' V n is produced. If we reverse Rearranging, we have
the directions of the arcs of C, a tournament T' is obtained also having 5 as 51: 1,2,2,2,3.
a score sequence (Figure 5.8(b». However, in T', the vertex V n is adjacent
to Vi rather than Vk' Hence the sum of the scores of the vertices adjacent Repeating this process twice more, we have
from V n is smaller in T' than in T, which is impossible. Thus, as claimed,
V n is adjacent to vertices having scores 5 1,52, •.. ,55., Then T - u is a
~: 1,2,2,1
tournament having score sequence 5. 0 8 2 : 1,1,2,2
As an illustration of Theorem 5.13, we consider the sequence 53: 1,1,1.
8: 1,2,2,3,3,4.
The sequence 53 is clearly a score sequence. We can use this information
In this case, 5 n (actually 56) has the value 4; thus, we delete the last term, to construct a tournament with score sequence 5. The sequence S3 is
repeat the first sn = 4 terms, and subtract 1 from the remaining terms, the score sequence of the tournament T3 of Figure 5.9: Proceeding
obtaining from 8 3 to 8 2 , we add a new vertex to T3 and join it to two vertices
5;,: 1,2,2,3,2. of T3 and from the other, producing a tournament T2 with score
sequence 8 2 , To proceed from 52 to SIt we add a new vertex to T2
and join it to vertices having scores 1,2 and 2, and from the remaining
vertex of T2 , producing a tournament TI with score sequence SI'
Continuing 'in the same fashion, we finally produce a desired tourna-
. ment T with score sequence 8 by adding a new vertex to TI and joining
it to vertices having scores 1, 2, 2 and 3, and joining it from the other
vertex.
The following theorem by Landau [Ll] gives a nonconstructive criter-
ion for a sequence to be a score sequence. There are many proofs of this
Figure 5.8 A step in the proof of Theorem 5.13. result; the one we give is due to Thomassen [T4].
144 Directed graphs Tournaments 145
Theorem 5.14 For each r satisfying 1 <r <n - k, we have
A nondecreasing sequence S: SI, S2, ... ,SIl of nonnegative integers is a score r r r r+k k

sequence if and only if for each k (1 < k < n), .?: ti

1=1
= 2: (Sk+i -
1=1
k) = L
1=1
5k+i - rk = .?: .?: Si - rk.
1=1
Si -
.=1

(5.1) Since

r+k (r+k)
with equality holding when k = n. L>i~ 2
i=1

and
Proof
Assume that S: SI ,S2, •.• ,SIl is a score sequence. Then 'there exists a
tournament T of order n with VeT) = {VI, v2, ... ,VII} such that
t
1=1
Si= (~),
odT Vi = 5i for i = 1,2, ... ,n. Let k be an integer with 1 < k < n. Then it follows that
T1 = ({VI> V2,"" vd) is a tournament of order k and size (~). Since
odT vi ~ od T1 Vi for 1 < i < k, it follows that

8 =8
k
Si
k
od T Vi ~ 8
k
od T! Vi =
(k)2 ' with equality holding for r = n - k. Thus, T satisfies (5.1)and so, by the
minimality of n, there exists a tournament T 2 of order n - k having score
with equality holding when k = n. sequence T.
We prove the converse by contradiction. Assume that S: SI, S2, ... , SII is Let T be a tournament with VeT) = V(Tl ) U V(T2 ) and
a counterexample to the theorem, chosen so that n is as small as possible E(T) = E(Tl ) U E(T2 ) U {(u, v)lu E V(Tz ), V E V(T1 )}.
and so that SI is as small as possible among all these counterexamples."
Suppose first that there exists an integer k with 1 < k < n - 1 such that Then S is a score sequence for T, contrary to assumption. Thus for
k = 1,2, ... , n - I,
tSi=
.=1
(~). (5.2)

Thus the sequence SI: SI, S2, •.• ,Sk satisfies (5.1) and so, by the minimality
of n, there exists a tournament T1 of order k having score sequence Sl' In particular, 51 > O.
Consider the sequence T:t 1 ,t2, ... ,tll _ k, where ti=Sk+i- k for Consider the sequence S': 51 - 1, 52, S3, ..• , SIl-l ,Sn + 1. Clearly S is a
i = 1,2, ... , n - k. Since nondecreasing sequence of nonnegative integers that satisfy (5.1). By

(k+ 1)
the minimality of Sit then, there exists a tournament T' ()f order n
k+l having score sequence Sf. Let x and y be vertices of T' such that
2: 5i ~ 2 '
i=l odT' x = sn + 1 and od T, y = SI -1. Since od T, x ~ odT' y + 2, there is a
vertex w =I x,y such that (x, w) E E(T') and (y, w) f/. E(T'). Thus,
it follows from (5.2) that
P: x, w,y is a path in T'. .
Let T be the tournament obtained from T' by reversing the directions of
L
5k+l = k+l Si- k
i=l
LSi ~ (k+l)
i=l 2
- (k)
2
=k. the arcs ofP. Then S is a score sequence for T, again producing a contra-
diction and completing the proof. 0 .
Thus, since S is a nondecreasing sequence,
With a slight alteration in the hypothesis of the preceding theorem,
ti = 5k+i - k ~ Sk+l - k ~ 0
we obtain a necessary and sufficient condition for a score sequence of a
f~r i.= 1,2, ... , n - k, and so T is a nondecreasing sequence of nonnega- strong tournament. This result is due to L. Moser (see Harary and Moser
tIve mtegers. We show that T satisfies (5.1). [HMl]).
146 Directed graphs Tournaments 147
Theorem 5.15 By Theorem 5.14, S is the score sequence of a tournament. Let T be a
tournament with V(T) = {VI, V2,"" vn } such that Si = od Vi (1 ~ i~ n).
A nondecreasing sequence S: SI, S2, ... , Sn of nonnegative integers is a score We show that T is strong.
sequence of a strong tournament if and only if If T is not strong, it follows from Theorem 5.12 (and the discussion pre-
ceding it) that V(T) can be partitioned as U U W such that (u, w) E E(T)
for every U E U and WE W. Suppose that IWI = k. Then W =
{VI' V2, ... ~ Vk}' Let T1 = (W). Then odr Vj = odT! Vj for 1 ~ i ~ k. Since T1
is a tournament of order k,we have
for 1 ~ k ~ n - 1 and

tSj=
.=1
(~). tSj=
i=l
L
WEW
odrw= L
WEW
odTjw= G),
Furthermore, if S is a score sequence of a strong tournament, then every tourna- contradicting the hypothesis. D
ment with S as a score sequence is strong.
We close this section with a brief'discussion involving distance in a
Proof toumament. Recall that if U and v are vertices of a digraph 0, and 0 con-
tains at least one (directed) u-v path, then the length of a shortest u-v path
Let T be a strong tournament with V(T) = {v], V2,"" v n } and suppose is called the directed distance from u to v and is denoted by d(u, v).
that the nondecreasing sequence S: s}, 52,"" 5n is a score sequence of T,
where Sj = od Vj for i = 1,2, ... ,n. Since T is a tournament of order n,
Theorem 5.16

t
•=1
Sj= (~). Let v be avertex ofmaximum score in a nontrivial tournament T. If u is a vertex of
T different from v, then d(v, u) ~ 2.
Let 1 ~ k ~ n -1 and define Tl = ({v], V2,"" Vk})' Since T1 is a tourna- "
ment of order k, Proof

t
j=1
odT!vj = (k).
2
Assume that od v = k. Necessarily, k;;:. 1. Let v1> V2,"" Vk denote the
vertices of T adjacent from v. Then d(v, Vj) = 1 for i = 1,2, ... ,k. If
V(T) = {v, v}, V2,"" Vk}, then the proof is complete.
Since T is a strong tournament, some vertex Vj in T1 (1 ~ j ~ k) must be Assume, then, that V(T) - {v, v}, V2' ••• ,vd is nonempty, and let
adjacent in T to a vertex not in T1 so that odT Vj > odT, Vj' Since u E V(T) - {v, VI' V2,"" vkl. If u is adjacent from some vertex Vj,
odT Vj ;;:. odT]Vj for all i (1 ~ i ~ k), we o\:>tain 1 ~ i ~ n, then d(v, u) = 2, producing the desired result. Suppose that

k k
?= 5j =?= odTvj > ?= odT,vj =
k (k)2 . this is not the case. Then u is adjacent to all of the vertices VI' V2, .•• , Vk!
as well as to v, so od u ;;:. 1 + k = 1 + od v. However, this contradicts the
,=1 ,=1 .=1 fact that v is a vertex of maximum score. D
For the converse, we assume that S: S},S2,'" ,sn is a nondecreasing Theorem 5.16 was first discovered by the sociologist Landau [Ll]
sequence of nonnegative integers such that
during a study of pecking orders and domination among chickens. In
the case of chickens, the theorem says that if chicken c pecks the largest
number of chickens, then for every other chicken d, either c pecks d, or c
pecks some chicken that pecks d. Thus c dominates every other chicken
forl~k~n-land either directly or indirectly in two steps. .
Let D be a strong digraph. Recall that the eccentricity e(v) of a vertex v of
tSj= (;). o is defined as e(v) =maxWEV(D) d(v,w). The radius of 0 is radO=
•=1 minVEV(D) e(v) and the center Cen(O) of 0 is defined as ({ vle(v) = rad O}) .
148 Directed graphs Hamiltonian tournaments 149
Theorem 5.16 provides an immediate result dealing with the radius of a 5.13 Which of the following sequences are score sequences? Which are
strong tournament. score sequences of strong tournaments? For each sequence that is
a score sequence, construct a tournament having the given sequence
as a score sequence.
Corollary 5.17 (a) 0, 1, 1,4,4
Every nontrivial strong tournament has radius 2. (b) 1, 1, 1,4,4,4
(c) 1, 3, 3, 3, 3, 3, 5
We conclude this section with a result on the center of a strong tourna- (d) 2,3,3,4,4,4,4,4
ment. 5.14 What can be said about a tournament T with score sequence
S1>S2,'" ,Sn such that equality holds in (5.1) for every k, 1 ~ k ~ n?

Theorem 5.18
5.15 Show that if S: 81, s2, ... , 8n is a score sequence of a tournament, then
The center of every nontrivial strong tournament contains at least three vertices. Sl:n-l-S1,n-l-sz, ... ,n-I-sn is a score sequence of a
tournament.
Proof 5.16 Give two different proofs that every regular tournament is strong.
Let T be a nontrivial strong tournament. By Corollary 5.17, rad T = 2. Let 5.17 Prove that every two vertices in a nontrivial regular tournament lie
w be a vertex having eccentricity 2. Since T is strong, there are vertices on a 3-cycle.
adjacent to w; let v be one of these having maximum score. Among the
vertices adjacent to v, let u be one of maximum score. We show that both u 5.18 Prove that if T is a nontrivial regular tournament, then diam T = 2.
and v have eccentricity 2, which will complete the proof.
5.19 Prove that every vertex of a nontrivial strong tournament lies on a
Assume, to the contrary, that one of the vertices u and v does not have
eccentricity 2. Suppose, then, that x E {u, v} and e(x) ~ 3. Hence, ther~ 3-cycle.
exists a vertex y in T such that d(x, y) ~ 3. Thus, Y is adjacent to x. More- 5.20 Prove Corollary 5.17.
over, y is adjacent to every vertex adjacent from x. These observations
imply that ody > odx. 5;21 (a) A vertex v of a tournament T is called a winner if d(v, u) ~ 2 for
Suppose that x = v. Since x is adjacent to w, it follows thaty is adjacent every u E V(T). Show that no tournament has exactly two
to w. However, ad y > ad v, which contradicts the defining property of v. winners.
Therefore, x = u. Here x is adjacent to v so that y is adjacent to v, but (b) Show that if n is a positive integer, n ~ 2,4, then there is a
ad y > ad u. Hence, x ~ u and the proof is complete. 0 tournament of order n in which every vertex is a winner.

EXERCISES 5.2
5.3 HAMILTONIAN TOURNAMENTS
5.8 Draw all four (nonisomorphic) tournaments of order 4.
The large number of arcs that a tournament has often produces a
5.9 Give an example of two nonisomorphic regular tournaments of the variety of paths and cycles. In this section we investigate these types
same order. of subdigraphs in tournaments. We begin with perhaps the most basic
5.10 Prove Theorem 5.12. result of this type, a property of tournaments first observed by Redei
[R21.
5.11 Determine those positive integers n for which there exist regular
tournaments of order n.
Theorem 5.19
5.12 Show that if two vertices u and v have the same score in a tourna-
ment T, then u and v belong to the same strong component of T. Every tournament contains a hamiltonian path.
 150 Directed graphs Hamiltonian tournaments 151
Proof that for some i (1:0::; i:o::; t), both (Vi'V) and (V,Vi+j) are arcs ofT (where all
subscripts are expressed modulo t). Thus, VI lies on the (( + I)-cycle
~et T be ~ tournamex:t of ?rder n, and let P: VI' Vz," ., Vkbe a longest path
m T. If P IS not a hamiltoman path ofT, then 1 :0::; k < n and there is a vertex
V of T not on P. Since P is a longest path, (v, Vj)1 (Vb V) rt E(T), and so Case 2. Suppose that no vertex v exists as in Case 1. Let A denote the set of
(Vj,v), (V,Vk) E E(T). This implies that there is an integer i (1:0::; i < k)
all vertices in V(T) - V(C) that are adjacent to every vertex of C, and let B
such that (Vi, V) E E(T) and (v, Vi+j) E E(T). But then
be the set of all vertices in V (T) - V (C) that are adjacent from every vertex
of C. Then Au B = V(T) - V(C). Since T is strong, neither A nor B is
empty. Furthermore, there is a vertex b in B and a vertex a in A such
is a path whose length exceeds that of P, producing a contradiction. 0 that (b,a) E E(T). Thus, VI lies on the (t + I)-cycle
A simple but useful consequence of Theorem 5.19 concerns transitive a,vj,vz, ... ,vt_I,b,a. 0
tournaments.
Corollary 5.22
Corollary 5.20 Every nontrivial strong tournament is pancyc1ic.
Every transitive tournament contains exactly one hamiltonian path.
We consider next a class'of oriented graphs that properly includes
The preceding corollary is a special case of a result by Szele [513], who tournaments, and see an extension of Theorem 5.21 within this class.
showed that every tournament contains an odd number of hamiltonian For k ~ 2, a k-partite tournament is a digraph obtained by orienting
paths. the edges of a complete k-partite graph. Thus a tournament of order n
While not every to~rnament is hamiltonian, such is the case for strong is an n-partite tournament with exactly n vertices. The partite sets of the
tou~aments, a fact dIscovered by Carnian [Cll. It is perhaps surprising underlying k-partite graphs are also referred to as the partite sets of the k-
that if a tourna~ent is .hamiltonian, then it must possess significantly partite tournament.
stronger properties. A dIgraph D of order n ~ 3 is pancyclic if it contains Bondy [B9] proved that every strong k-partite tournament (k ~ 3) con-
a cycle of length ( for each ( = 3,4, ... , n and is vertex-pancyclic if each tains an (-cycle for t = 3,4, ... ,k. This is a generalization of Corollary
vertex v of D lies on a cycle of length ( for each ( = 3,4, ... ,n. Harary 5.22. Our next result, due to Guo and Volkmann [GV2l, is the analogous
and Moser [HMl] showed that every nontrivial. strong tournament is generalization of Theorem 5.21.
pancyclic. The following result was discovered by Moon [M9]. The
proof here is due to C. Thomassen. Theorem 5.23
Let D be a strong k-partite (k ~ 3) tournament. Then every partite set of D has at
Theorem 5.21 least one vertex that lies on an {-cycle for t = 3,4, ... , k.
Every nontrivial strong tournament is vertex-pancyc1ic.
Proof
Proof Let Vj, V z , ... , V k be the partite sets of D. We show, without loss of
Let T be a strong tournament of order n ~3, and letvj be a vertex of T. generality, that VI has a vertex contained in an {-cycle for t = 3,4, ... , k.
"!'Ie sh~w that Vj lies on an t-cycle for each t = 3,4, ... , n. We proceed by We proceed by induction on t, first showing that VI has a vertex that
mductlOn on (. belongs to a 3-cycle. Let v E VI' Since D is strong, v is contained in at
Since T is strong, it follows from Exercise 5.19 that Vj lies on a 3-cycle. least one cycle. Let C: v = VI' Vz, ... , Vil VI be a shortest cycle containing
Assumethatvjlies onant-cyclevj, vz, ... ,Vt, VI, where 3 :0::; t:o::; n -1. We v = VI' If V3 rt VI' then, because of the minimality of C, it follows that.
prove that Vj lies on an (t + I)-cycle. (V3, VI) E E(D) and so Vj is on a 3-cycle. Assume then that V3 E VI; so
t ~ 4 and V4 rt VI' Since C is a smallest cycle containing VI' it follows
Case 1. Suppose that there is a vertex v not on C that is adjacent to at least that (V4' vd E E(D) and t = 4. If Vz and V4 belong to distinct partite sets,
one vertex of C and is adjacent from at least one vertex of C. This implies then either (a) (vz, V4) E E(D) and Vb Vz, V4, VI is a 3-cyc1e containing VI
152 Directed graphs Hamiltonian tournaments 153
or (b) (V4,V2) E E(D) and V2,V3,V4,V2 is a 3-cycle containing V3' Thus we is of length j + q + t - r. Thus, Ul lies on an (-cycle for t = 3,
may assume thatv2, V4 E V 2. Ifthere is a vertex X E Vi fori ~ 3 such that X is 4, ... , t + q -1.
adjacent to at least one vertex of C and also adjacent from at least one It remains to consider the case that V(P) n 52 =1= 0. Since every vertex of
vertex of C, then at least one of VI and V3 is on a 3-cycle. 52 is adjacent to every vertex of C, the vertex Yq-l E 52' If Ul and Yq-2 are
Therefore, we can assume that V(D) - VI - V 2 can be partitioned into in distinct partite sets, then (UbYq-2) E E(D) and so Ul,Yq-2,Yq-b
two sets 51 and 52 such that every vertex of 51 is adjacent from every U3, U4, ... , Uh Ul is a (t + l)-cycle. If Yq-2 is in V v then (U2, Yq-2) E E(D)
vertex of C and every vertex of 52 is adjacent to every vertex of C. Since and then U1' U2, Yq-2' Yq_ b U4, Us,· .. 'Urn' Ul or (if t = 3) Ul , U2, Yq-2,
k ~ 3, at least one of 51 and 52 is nonempty; say 51 =1= 0· Since D is strong, Yq-ll U1 is a (t + l)-cycle and the proof is complete. 0
there is a path from every vertex in 51 to every vertex of C. Let
P: Xl, X2,' .. , xq be a shortest such path, where necessarily q;:' 3.. If
V(P) n 52 = 0, then one of X2 and X3 belongs to V 1 and the other belongs
EXERCISES 5.3
to V 2' But then Xl' X2, X3, Xl is a 3-cycle. Therefore, suppose that
V(P) n 52 =1= 0. Since P is a shortest path from the vertices in 51 to those
5.22 Prove that if T is a tournament that is not transitive, then T has at
in C and since every vertex of 52 is adjacent to every vertex of C, it follows
least three hamiltonian paths.
that Xq_l E 52' If q = 3, then VI' Xl' X2, VI is a 3-cycle. So, assume that.q ;:. 4.
Then Xq-2 E VI or Xq-2 E V 2. In the first case, Xq_2, Xq_l, V2, Xq_2 1S a 3- 5.23 Use Corollary 5.20 to give an alternative proof of Theorem 5.9.
cycle; in the second case, Vl,X q_2,Xq_l,Vl is a 3-cycle. Thus VI has at
5.24 Prove or disprove: Every arc of a nontrivial strong tournament T lies
least one vertex that lies on a 3-cycIe. on a hamiltonian cycle of T.
Suppose now that U is a vertex of VI that lies on an t-cycle for
( = 3,4, ... , t, where t < k. We show that either u is on a (t + l)-cycle or 5.25 Prove or disprove: Every vertex-pancyclic tournament is hamiltonian-
VI contains another vertex that lies on an t-cycle for t = 3,4, ... , t, t + 1. connected.
Let C: Ul, U2, ... , Ut , Ul be a t-cycle with U = Ul and let 5 be the set of
5.26 Show that if a tournament T has an t -cycle, then T has an s-cycle for
vertices that belong to partite sets not represented on C. If there is a
s = 3,4, ... ,t.
vertex x of 5 adjacent to and from vertices of C,then x can be inserted
in C to form a (t + l)-cycle containing u. Otherwise, 5 can be decomposed 5.27 A digraph D has a hamiltonian antipath if D has an antipath contain-
into two sets 51 and 52 such that every vertex of Sl is adjacent from every ing all vertices of D. There are only three tournaments that do not
vertex of C and every vertex of 52 is adjacent to every vertex of C. Without contain hamiltonian antipaths, one of order 3, one of order 5, and one
loss of generality, assume that 51 =1= 0. Since D is strong, there is a path of order 7. Find the two smallest ofthese.
from 51 to C. Let P: Yl, Y2," ., Yq be a shortest such path, where necessarily
5.28 A hamilton~an anticycle in a digraph D is a spanning semicycle of
q ;:. 3. D containing no subpath of length 2. A digraph D is anti1u;tmiltonian
Suppose first that V(P) n 52 = 0. Then (y;, Yl) E E(D) for i ;:. 3 since P
if Dcontains a hamiltonian anticycle. Show that a graph G has an
is a shortest path from 51 to C. If P includes at least one vertex of V v then
antihamiltonian orientation if and only if G is a hamiltonian graph
choose the minimum value oUforwhichYt E V(P) n VI' We claim thatYt
of even order.
lies on a j-cycle for j = 3,4, ... , t + q - 1 ;:. t + 2. If t = 2 or t = 3, then it is
straightforward to verify this claim. If t ;:. 4, then by the choice of Yt, we 5.29 A digraph D is antihamiltonian-connected if for every pair u, v of
see that (Yt, yJ E E(D) for i ~ t - 2. But then clearly Yt is contained in vertices, D contains a hamiltonian u-v antipath. Prove that no
cycles of lengths 3,4, ... , t + q -1. If, on the other hand, V(P) n VI = 0, cubic graph has an antihamiltonian-connected orientation.
then (UllYi) E E(D) for i ~ q -1. Since Yq E V(C) and Yq =1= Ul, we have
that Yq = Ur for some r ;:. 2. Then for every i with 1 ~ i ~. q - 1, the cycle

is of length i + t - r + 2. Furthermore, for every j with 1·~ j ~ r -1, the

cycle
 Euler's Formula 155
CHAPTER 6

Planar graphs
Figure 6.1 Planar, plane and nonplanar graphs.

of A neither corresponds to a vertex of G nor lies on any curve correspond-

We now consider graphs that can be drawn in the plane without their ing to an edge of G. Intuitively, the regions of G are connected portions of
edges crossing. A formula developed by Euler plays a central role in the plane remaining after all curves and points corresponding, respec-
the study of these 'planar' graphs. We describe two characterizations of tively, to edges and vertices of G have been deleted. For a plane graph
planar graphs. A necessary condition for a planar graph to be hamiltonian G, the boundary of a region R consists of all those points x corresponding
is discussed. Two parameters associated with nonplanar graphs are then to vertices and edges of G having the property that x can be joined to a
considered. point of R by a curve, allof whosepoints·tlifferent from x belong to R.
Every plane graph G contains an unbounded region called the exterior
region of G. If G is embedded on the sphere, then no region of G can be
6.1 EULER'S FORMULA regarded as being exterior. On the other hand, it is equally clear that a
plane graph G can always be embedded in the plane so that a given
An (n, m) graph G is said to be realizable or embeddable on a surface 5 if it is region of G becomes the exterior region. Hence a plane graph G can
possible to distinguish a collection of n distinct points of 5 that correspond always be realized in the plane so that any vertex or edge lies on the
to the vertices of G and a collection of m curves, pairwise disjoint except boundary of its exterior region. The plane graph Gz of Figure 6.1 has
possibly for endpoints, on 5 that correspond to the edges of G such that if three regions, and the boundary of each is a 4-cycle.
a curve A corresponds to the edge e = uv, then only the endpoints of A The order, size and number of regions of any connected plane graph are
correspond to vertices of G, namely u and v. Intuitively, G is embeddable related by a well-known formula discovered by Euler [E7].
on S if G can be drawn on S so that edges (more precisely, the curves
corresponding to edges) intersect only at a vertex (that is, a point corre-
sponding to a vertex) mutually incident with them. In this chapter we are Theorem 6.1 (Euler's Formula)
concerned exclusively with the case in which S is a plane or sphere. If G is a connected plane graph with n vertices, m edges and r regions, then
A graph is planar if it can be embedded in the plane. Embedding a graph
in the plane is equivalent to embedding it on the sphere. In order to see n -m+r=2.
this, we perform a stereographic projection. Let 5 be a sphere tangent to a
plane 1r, where P is the point of S diametrically opposite to the point of Proof
tangency. If a graph G is embedded on S in such a way that no vertex of G We employ induction on m, the result being obvious for m = 0 since in this
is P and no edge of G passes through P, then G may be projected onto 1r to case n = 1 and r = 1. Assume that the result is true for all connected plane
produce an embedding of G on1r. The inverse of this projection shows that graphs with fewer than m edges, where m ~ 1, and suppose that G has m
any graph that can be embedded in the plane can also be embedded on edges. If G is a tree, then n = m + 1 and r = 1 so the desired formula
the sphere. follows. On the other hand, if G'is not a tree, let e be a cycle edge of G
If a planar graph is embedded in the plane, then it is called a plane graph. and consider G - e. The connected plane graph G --, e has n vertices,
The graph G1 = K2,3 of Figure 6.1 is planar, although, as drawn, it is not m- 1 edges, and r - 1 regions so that by the inductive hypothesis,
plane; however, Gz = K 2 ,3 is both planar and plane. The graph G3 = K ,3 is n - (m -1) + (r-1) = 2, which implies that n - m+r = 2. 0
nonplanar. This last statement will be proved presently. 1
Given a plane graph G, a region of G is a maximal portion of the plane
From the preceding theorem, it follows that every two embeddings of a
for which any two points may be joined by a curve A such that each point connected planar graph in the plane result in plane graphs having the
 Planar graphs Euler's Formula 157
156
same number of regions; thus one can speak of the number of regions of Corollary 6.5
a connected planar graph. For planar graphs in general, we have the Every planar graph contains a vertex of degree at most 5.
following result.
Proof
Corollary 6.2 Let G be a planar (n, m) graph with V(G) = {Vb vz, .. " v n }. If n ~ 6, then
If G is a plane graph with n vertices, m edges and r regions, then the result is obvious. Otherwise, m ~ 3n - 6 implies that
n-m+r=l+k(G). n
L degvi = 2m ~ 6n -12.
A planar graph G is called maximal pl~nar if, for every p~ir u, v of non- ;=1
adjacent vertices of G, the graph G + uv IS nonplanar. Thus In any embed- Not all vertices of·G have degree 6 or more, for then 2m ~ 6n. Thus G
ding of a maximal planar graphG having order n ~ 3, the boundary of contains a vertex of degree 5 or less. .0
every region of G is a triangle. For this reason, maximal planar graphs are
also referred toas triangulated planar graphs; triangulated plane graphs are We next consider another corollary involving degrees. In it we make
often called simply triangulations. . . ., use of the fact that the minimum degree is at least 3 in a maximal
On a given number n of vertices, a planar graph IS qUlte hmlted as to planar graph of order at least 4.
how large its size m can be. A bound on m follows from our next result.
Corollary 6.6
Theorem 6.3
Let G be a maximal planar graph oforder n ~ 4, and let ni denote the number of
If G is a maximal planar (n, m) graph with n ~ 3, then vertices of degree i in G, for i = 3,4, ... ,k = 6.(G). Then
m =3n-6. 3n3 + 2n4 + ns = n7 + 2ng + ... + (k - 6)nk + 12.

Proof Proof
Denote by r the number of regions of G. In G the boundary of every region Let G have size m. Then,by Theorem 6.3, m = 3n - 6. Since
is a triangle, and each edge is on the boundary of two regions. Therefore, k k
if the number of edges on the boundary of a region is summed over all n = Ln; and 2m = Lin;,
regions, the result is 3r. On the other hand, such a sum counts each edge ;=3 ;=3
twice, so 3r = 2m. Applying Theorem 6.1, we obtain m = 3n - 6. 0
it follows that
k k
Corollary 6.4 Lin; = 6 L n; - 12
i=3 ;=3
IfG is a planar (n,m) graph with n ~ 3, then
and, consequently,
m ~3n-6.
3n3 + 2n4 + ns = n7 + 2ng + ... + (k - 6)nk + 12. 0
Proof An interesting feature of planar graphs is that they can be embedded in
Add to G sufficiently m\l.ny edges so that the resulting (n', m') g;aph c;! is the plane so that every edge is a straight line segment. This result was
maximal planar. Clearly, n = n' and m ~ m'. By Theorem 6.3, m = 3n - 6 proved independently by F<iry [Fl] and Wagner [WI].
and so m ~ 3n - 6. 0 The theory of planar graphs is very closely allied with the study of poly-
hedra; in fact, with every polyhedron P is associated a connected planar
An immediate but important consequence of Corollary 6.4 is given graph G(P) whose vertices and edges are the vertices and edges of P.
next. Necessarily, then, every vertex of G(P) has degree at least 3. Moreover,
158 Planar graphs Euler's Formula 159
Hence E ;;0 3F. Also,

2£ = LkVk;;o L3Vk =3L Vk = 3V.

k;;>3 k;;>3 k;;>3

By Theorem 6.3, V - £ + F = 2; therefore, E ~ ~E +lE - 2 = E- 2. This

is a contradiction. 0

A regular polyhedron is a polyhedron whose faces are bounded by

congruent regular polygons and whose polyhedral angles are congruent.
Figure 6.2 A polyhedron and its associated graph. In particular, for a regular polyhedron, F = Fs for some s and V = VI
for some t. For example, a cube is a regular polyhedron with V = V 3
if G(P) is a plane graph, then the faces of P are the regions of G(P) and and F = F4 • There are only four other regular polyhedra. These five
every edge of G(P) is on the boundary of two. regions. A polyhedron and regular polyhedra are also called platonic solids. The Greeks were
its associated plane graph are shown in Figure 6.2. aware, over two thousand years ago, that there are only five such
It is customary to denote the number of vertices, edges and faces of a polyhedra.
polyhedron P by V, E and F, respectively. However, these are the number
of vertices, number of edges, and number of regions of a connected planar
Theorem 6.9
graph, namely G(P). According to Theorem 6.1, V, E and F are related. In
this form, the statement of this result is known as the Euler Polyhedron There are exactly five regular polyhedra.
Formula.
Proof
Theorem 6.7 (Euler Polyhedron Formula)
Let P be a regular polyhedron and let G(P) be an associated planar graph.
If V, E and F are the number of vertices, edges and faces of a polyhedron, then Then V - E + F = 2, where V, E and F denote the number of vertices,
V-E+F=2. edges and faces of P and G(P). Therefore,
-8 = 4E - 4V - 4F
When dealing with a polyhedron P (as well as the graph G(P), it is
customary to represent the number of vertices of degree k by V k and =2E+2E-4V-4F
number of faces (regions) bounded by a k-cycle by Fk • It follows then that
= L kFk + L kVk - 4L V k - 4 L Fk
2E = L kVk = k;;>3
k;;>
3
L kFk' (6.1) k;;.3 k;;>3 k;;>3 k;;>3

= L(k- 4)Fk + L(k - 4)Vk·

By Corollary 6.5, every polyhedron has at least one vertex of degree 3, 4 k;;.3 k;;>3
or 5. As an analogue to this result, we have the following.
Since P is regular, there exist integers s(~3) and t(;;o3) such that F = Fs
and V = VI' Hence -8 = (s - 4)Fs + (t - 4)Vt • Moreover, we note that
Theorem 6.8 3 ~ s ~ 5, 3 ~ t ~ 5, and sFs = 2E = tVI • This gives us nine cases to
At least one face of every polyhedron is bounded by a k-cyde for some k = 3,4,5. consider.
Case 1. Assume that s = 3 and t = 3. Here we have
Proof
-8 = -F3 - V3 and. 3F3 = 3V3 ,
Assume that F3 = F4 = F5 = 0, so by equation (6.1),
so F3 = V 3 = 4. Thus P is the tetrahedron. (That the tetrahedron is the
2£ = I)F k ~ L 6Fk = 6 L h = 6F. only regular polyhedron with V 3 = F3 = 4 follows from geometric con-
k;;>6 k;;>6 k;;>6 siderations.) .
160 Planar graphs Characterizations of planar graphs 161
Case 2. Assume that s = 3 and t = 4. Therefore Case 8. Assume that s = 5 and t = 4. Here -8 = Fs, which is impossible.
-8 = -F3 and 3F3 = 4V4 . Case 9. Assume that s = 5 and t = 5. This, too, is impossible since.
Hence F3 = 8 and V4 = 6, implying that P is the octahedron. -8 =f. Fs + V s· This completes the proof. 0

Case 3. Assume that s = 3 and t = 5. In this case, The graphs of the five regular polyhedra are shown in Figure 6.3.
-8 = -F3 + V s and 3F3 = 5Vs ,
so F3 = 20, V s = 12 and P is the icosahedron.
EXERCISES 6.1
Case 4. Assume that s = 4 and t = 3. We find here that
-8 = - V 3 and 4F4 = 3V3 • 6.1 Give an example of a planar graph that contains no vertex of degree
Thus V3 = 8; F4 = 6 and P is the cube. less than 5.

Case 5. Assume that s = 4 and t = 4. This is impossible since -8 =f. O. 6.2 Show that every planar graph of order n ;;:,: 4 has at least four vertices
of degree less than or equal to 5.
Case 6. Assume that s = 4 and t = 5. This case, too, cannot occur, for
6.3 Prove Corollary 6.2.
otherwise -8 = V s.
6.4 Prove that a planar (n, m) graph with n ;;:,: 3 is maximal planar if and
Case 7. Assume that s = 5 and t = 3. For these values, only if m = 3n - 6.
-8 = Fs - V 3 and 5Fs = 3V3 •
6.5 Prove that there exists only one 4-regular maximal planar graph.
Solving for Fs and V 3 , we find that Fs = 12 and V 3 = 20, so P is the 6.6 Let k;;:,: 3 be an integer, and let G be an (n, m) plane graph of order
dodecahedron. n(;;:':k).
(a) If the length of every cycle is at least k, then determine an upper
bound B for m in terms of nand k.
(b) Show that the bound B obtained in (a) is sharp by determining,
for arbitrary k ;;:,: 3, an (n, B) plane graph G, every cycle of which
has length at least k.
6.7 Show that every 2-edge-connected planar graph has a cycle double
cover.
tetrahedron cube octahedron

6.2 CHARACTERIZATIONS OF PLANAR GRAPHS

There are two graphs, namely Ks and K 3,3 (Figure 6.4), that play an
important role in the study of planar graphs.

dodecahedron icosahedron
Figure 6.3 The graphs of the regular polyhedra. Figure 6.4 The nonplanar graphs Ks and' K3 ,3'
162 Planar graphs Characterizations of planar graphs 163
Theorem 6.10 is undoubtedly one of the best known theorems in the theory of graphs.
Before presenting a proof of this result, first discovered by Kuratowski
The graphs Ks and K3,3 are nonplanar. [KIll, we need one additional fact about planar graphs.

Proof
Theorem 6.12
Suppose, to the contrary, that Ks is a planar graph. Since Ks has n = 5
A graph is planar if and only if each of its blocks is planar.
vertices and m = 10 edges,
10 = m > 3n - 6 = 9,
Proof
which contradicts Corollary 6.4. Thus Ks is nonplanar.
Suppose next that K 3 ,3 is a planar graph, and consider any plane embed- Certainly, a graph G is planar if and only if each of its components is
ding of it. Since K3 ,3 is bipartite, it has no triangles; thus each of its regions planar, so we may assume G to be connected. It is equally dear that if
is bounded by at least four edges. Let the number of edges bounding a G is planar, then each block of G is planar. For the converse, we employ
region be summed over all r regions of K 3 ,3, denoting the result by N. induction on the number of blocks of G. If G has only one block and this
Thus, N ~ 4r. Since the. sum N counts each edge twice and K 3 ,3 contains block is planar, then, of course, G is planar. Assume that every graph with
m = 9 edges, N = 18 so that r::::; ~. However, by Theorem 6.1, r = 5, and fewer than k ~ 2 blocks, each of which is planar, is a planar graph, and
this is a contradiction. Hence K3 ,3 is nonplanar. 0 suppose that G has k blocks, all of which are planar. Let B be an end-block
of G, and denote by v the cut-vertex of G common to B. Delete from Gall
For the purpose of presenting two useful, interesting criteria for graphs vertices of B different from v, calling the resulting graph G'. By the'
to be planar, we describe two relations on graphs in this section. inductive hypothesis, G' is a planar graph. Since the block B is planar, it
An elementary subdivision of a nonempty graph G is a graph obtained may be embedded in the plane so that v lies on the exterior region. In any
from G by removing some edge e = uv and adding a new vertex wand region of a plane embedding of G' containing v, the plane block B may
edges uw and vw. A subdivision of G is a graph obtained from G by a now be suitably placed so that the two vertices of G' and B labeled v
succession of elementary subdivisions (including the possibility of are 'identified'. The result is a plane graph of G; hence G is planar. 0
none). In Figure 6.5 the graphs G1 andGz are subdivisions of G3 .
It should be clear that any subdivision of a graph G is planar or We can now give a characterization of planar graphs. The proof of the
nonplanar according to whether G is planar or nonplanar. Also it is an following result, known as Kuratowski's theorem [Kll], is based on a
elementary observation that if a graph G contains a nonplanar subgraph, proof by Dirac and Schuster [DSl].
then Gis nonplanar. Combining these facts with our preceding results, we
obtain the following. Theorem 6.13
A graph is planar if and only if it contains no subgraph that is a subdivision of
Theorem 6.11 either Ks or K3,3'
If agraph G contains asubgraph that is asubdivision ofeither Ks or K3 3' then G is
nonplanar. . ,
Proof
The remarkable property of Theorem 6.11 is that its converse is also The necessity is precisely the statement of Theorem 6.11; thus we need
true. These two results provide a characterization of planar graphs that only consider the sufficiency. In view of Theorem 6.12, it is sufficient to
show that if a block contains no subgraph that is a subdivision of Ksor
K3 ,3, then it is planar. Assume, to the contrary, that suchis not the case.
Hence among all nonplanar blocks containing no subgraphs that are
subdivisions of either Ks or K3 ,3, let G be one of minimum size.
First we verify that 6(G) ~ 3. Since G is a block, it contains no end-
vertices. Assume, then, that G contains a vertex v with deg v = 2, such
Figure 6.5 Subdivision. that v is adjacent with uand w. We Consider two possibilities. Suppose
164 Planar graphs Characterizations of planar graphs 165
that uw E E(G). Then G - v is also a block. Since G - v is a subgraph of G,
it follows that G - v also contains no subgraph that is a subdivision of Ks
or K3,3; however, G is a nonplanar block of minimum size having this
property, so G - v is planar. However, in any plane embedding of
G - v, the vertex v and edges uv and vw may be inserted so that the
H:
resulting graph G is plane, which contradicts the fact that G is nonplanar.
Next, suppose that uw ~ E(G). The graph G' = G - v + uw is a block
having smaller size than G. Furthermore, G' contains no subgraph that
is a subdivision of either Ks or K3,3; for suppose it contained such a sub-
graph F. If F failed to contain the edge uw, then G would also contain F, V,
which is impossible; thus F contains uw. !fto F - uw we add the vertex v
and edges uv and WV, the resulting graph F' is a subdivision of F. Figure 6.6 Structure of the graph H of Theorem 6.13.
However, F' is a subgraph of G, which is impossible. Thus G' is a block
having size less than G that contains no subgraph that is a subdivision of connecting a vertex of P with a vertex of C must contain at least one of
either Ks or K3,3, so G' is planar. However, since G is a subdivision of G', Vs and VI'
this implies that G too is planar, which is a contradiction. Thus, G cannot Let H] be the component of H - {v r I 0 :;;; r < k, r =J s, t} containing P.
contain a vertex of degree 2; so 8(G) ;;:, 3, as claimed. By the choice of C, the subgraph H t cannot be inserted in the interior of
By Corollary 2.11, G is not a minimal block, so there exists an edge C in a plane manner. This, together with the assumption that G is non-
e = uv such that H = G - e is also a block. Since H has no subgraph planar, implies that the interior subgraph of H must contain one of the
s
that is a subdivision of either K Or K3 ,3 and H has fewer edges than following:
does G, the graph H is planar. Since H is a cyclic block, it follows by
1. A Va-Vb path Q, 0 < a < 5, i < b < t (or, equivalently, s < a < i and
Theorem 2.5 that H possesses cycles containing both u andv. We hence-
t < b < k), none of whose vertices different from Va and Vb belong
forth assume H to be a plane graph having a cycle, say C, containing u and
to C.
v. such that the number of regions interior to C is maximum. Assume that
2. A vertex w not on C that is connected to C by three internally disjoint
C is given by
paths such that the end-vertex of one suchpathP' is one of vo, vs , Vj and
u= Vo, V], ... , Vi = V, ... ,Vk = U, Vt· If P' ends at vo, the end-vertices of the other paths are Va and Vb,
where 5 :;;; a < i and i < b :;;; t but not both a = 5 and b = t hold. If P'
where 1 < i < k - 1.
ends at any of vs, Vi or Vt, there are three analogous cases.
Several observations regarding the plane graph H can now be made. In
3. A vertex w not on C that is connected to C by three internally disjoint
order to do this, it is convenient to define two special subgraphs of H. By
the exterior sUbgraph (interior subgraph) of H, we mean the subgraph 9f G paths PI! P2I P3 such that the end-vertices of the path (different from w)
induced by those edges lying exterior (interior) to the cycle C. First, since are three of the four vertices vo, vs' Vi, Vt, say Vo, Vi, vs, respectively,
the graph G is nonplanar, both the exterior and interior subgraphs exist, together with a Vc-V t path P4 (vc =J vo, Vi'W) where Vc is on P t o. Pz,
and P4 is disjoint from PlI P z and C except for V c and Vt. The remaining
for otherwise, the edge e could be added to H (either exterior to C or
choices for PlI Pz and P3 produce three analogous cases.
interior to C) so that the resulting graph, namely G, is planar.
4. A vertex w not on C that is connected to the vertices Vo, v s, Vi, Vt by four
We note further that no two distinct vertices of the set {vo, V], . .. , v;} are
internally disjoint paths.
connected by a path in the exterior subgraph of H, for otherwise this would
contradict the choice of C as being a cycle containing u and v having the These four cases exhaust the possibilities. In each of the first three cases,
maximum number of regions interior to it. A similar statement can be the graph G has a subgraph that isa subdivision of K3,3 while in the fourth
made regarding the set {Vi, Vi+], ... ,Vk}' These remarks in connection case, Ghas a subgraph that is a subdivision of Ks. However, in any case,
with the fact that H + e is nonplanar imply the existence of a Vs-VI path this is contrary to assumption. Thus no such graph G exists, and the proof
P, 0 < s < i < t < k, in the exterior subgraph of H such that no vertex of is complete. 0
P different from V o and V r belongs to C. This structure is illustrated in
Figure 6.6. We further note that no vertex of P different from V o and VI Thus the Petersen graph (Figure 6.7(a» is nonplanar since it contains
is adjacent to a vertex of C other than V s or VI' and, moreover, any path the subgraph of Figure 6.7(b) that i!> a subdivision of K3 ,3' Despite its
166 Planar graphs Characterizations of planar graphs 167

H: G:

v4o------i::>v 3

Figure 6.8 Contraction.

(a) (b)
Figure 6.7 The Petersen graph and a subgraph that is a subdivision of K3 ,3' Theorem 6.14
resemblance to the complete graph K s, the Petersen graph does not If a graph H is a subdivision of a graph G, then G is a contraction of H.
contain a subgraph that is a subdivision of Ks.
For graphs GI and Gz, a mapping ¢ from V (GI ) onto V (G z) is called an
Proof
elementary contraction if there exist adjacent vertices u and v of GI such that
If G = H, then clearly G is a contraction of H. Hence we may assume
(i) ¢u = ¢v, and {Ull VI} =1= {u, V} implies that ¢Ul =1= tPVl;
that H is obtained from G by a sequence of elementary subdivisions.
(ii) {Ul,vl}n{U,v}=0 implies that UIVI EE(GI ) if and only if
Suppose that G' is an elementary subdivision of G; then C/ is obtained
¢UI¢VI E E(G z); and
from G by removing some edge uv and adding a vertex W together with
(iii) for WE V(G I ), W =1= u, v, then uw E E(G1 ) or vw E E(G 1 ) if and only if
the edges uw and VW. However, then, G' is contractible to G by an
¢u¢w E E(G z).
elementary contraction ¢, which fixes every element of V(G) and
We say here that Gz is obtained from GI by the identification of the adjacent ¢w = ¢u. Hence G can be obtained from H by a mapping that is a com-
vertices U and v. A contraction is then a mapping from V(G I ) onto V(G z) position of finitely many elementary contractions so that G is a contraction
that is either an isomorphism or a composition of finitely many elemen- ofH. 0 .
tary contractions.
If there exists a contraction from V(G I ) onto V(G z), then Gzis a contrac- Corollary 6.15 will actually prove to be of more use than the theorem
0/
tion Glf and GI contracts to or is contractible to G z. A subcontraction of a itself.
graph G is a contraction of a sUbgraph of G.
There is an alternative and more intuitive manner in which to define
'contraction'. A graph Gz may be defined as a contraction of a graph G1 if Corollary 6.15
there exists a one-to-one correspondence between V (G z) and the elements If a graph H contains a sUbgraph that is a subdivision of a connected nontrivial
of a partition of V(G I ) such that each element of the partition induces a
graph G, then G is a subcontraction of H.
connected subgraph of GI , and two vertices of Gz are adjacent if and only
if the sUbgraph induced by the union of the corresponding subsets is
connected. We can now present our second characterization of planar graphs
In Figure 6.8, the graph G is a contraction of· H, obtained by the (Halin [H4], .Wagner [W2] and Harary and Tutte [HTl]), which is often
identification of Vi and vs. It might also be considered as the contraction referred to as Wagner's theorem.
resulting from the partition
V(H) = {vd U {vz,vs} U {V3} U {V4}' Theorem 6.16
A relationship between contraction and subdivision is given in the A graph G is planar if and only if neither Ks nor K3,3 is a subcontraction
following theorem. of G.
168 Planar graphs Hamiltonian planar graphs 169
Proof lllay coincide with V2lt V24 or V25) connected by internally disjoint
(possibly trivial) paths in G2 to the distirtct elements of {V2l, V24, V2S}' We
Let G be a nonplanar graph. By Theorem 6.14, G contains a subgraph that then let E2 denote the edge sets of these paths. In an analogous manner,
is a subdivision of K s or K 3 ,3' Thus by Corollary 6.15, K s or K 3,3 is a we define accordingly the sets E3 , E4 and Es with the aid of the sets
subcontraction of G. {V3l, V34, V3S}, {V4b V42, V43} and {VSb VS2, vs3 }, respectively..The subgraph
In order to verify the converse, we first suppose that G is a graph such induced by the union of the sets Ei and the edges VijVji contams a subgraph
that H = K3,3 is a subcontraction of G. We show, in this case, that G F that is a subdivision of K3 3 such that the vertices of degree 3 of Fare Wl I

contains a subgraph that is a subdivision of K3 ,3, implying that G is non- w; and the vertices Wi, i = 2,3,4,5. In either case, G is nonplanar. 0
planar. Denote the vertices of H by Ui and ui, 1 ::s;; i::s;; 3, such that every~
edge of H is of the type UiUj. Taking the alternate definition of contraction, As an application of this theorem, we again note the nonplanarity of
we let Gi , 1 ::s;; i ::s;; 3, be the connected subgraph of G corresponding to Ui the Petersen graph of Figure 6.7(a). The Petersen graph contains Ks as a
and let Gi correspond to ui. Since UiUj E E(H) for 1 ::s;; i ::s;; 3,1 ::s;; j ::s;; 3, in the subcontraction, which follows by considering the partition V lt V2 , V3 , V4 ,
graph G there exists a vertex Vij of Gi adjacent with a vertex vij of Gj. V s of its vertex set, where Vi = {Vi, Vi+S}'
Among the vertices Vilt Vi2, Vi3 of Gi, two or possibly all three may actually
represent the same vertex. If Vil = Vi2 = Vi3' we set each Vij = Vj; otherwise,
we define Vi to be a vertex of Gi connected to the distinct elements of
{Vil, Vj2, vd with internally disjoint paths in Gi . (It is possible that
EXERCISES 6.2
Vj = Vij for some j.) We now proceed as above with the subgraphs Gi,

thereby obtaining vertices vi. The subgraph of G induced by the nine 6.8 Show that the converse of Theorem 6.14 is not, in general, true.
edges vijVij together with the edge sets of any necessary aforementioned ; 6.9 Show that the Petersen graph of Figure 6.7(a) is nonplanar by
paths from a vertex Vi or vi is a subdivision of K 3 ,3' (a) showing that it has K3 ,3 as a subcontraction, and
Assume now that H = Ks is a subcontraction of G. Let V(H) = (b) using Exercise 6.6(a).
{Ui I ::s;; i::s;; 5}, and suppose that Gi is the connected subgraph of G that
corresponds to Ui' As before, there exists a vertex Vij of Gj adjacent with 6.10 Let T be a tree of order at least 4, and let el, e21 e3 E E(T). Prove that
a vertex Vji of Gj , i =I j, 1 ::s;; i, j ::s;; 5. For a fixed i, 1 ::s;; i ::s;; 5, we consider the T + el + e2 + e3 is planar.
vertices Vij'} =I i. If the vertices Vjj represent the same vertex, we denote 6,n A graph G is auterplanar if it can be embedded in the plane so that
this vertex by Vi' If the vertices Vij are distinct and there exists a vertex every vertex of G lies on the boundary of the exterior region. Prove
(possibly some Vij) from which there are internally disjoint paths (one of the following:
which may be trivial) to the Vij' then denote this vertex by Vi' If three of the (a) A graph G is outerplanar if and orilyif G + Kl is planar.
vertices Vij are the same vertex, call this vertex Vi' If two vertices vij are the (b) A graph is outerplanar if and orily if it contains no subgraph that
same while the other two are distinct, then denote the two coinciding ver- is a subdivision of either .Kt or K2,3'
tices by Vi if there exist internally disjoint paths to the other two vertices. (c) If G is an (n, m) outerplanar graph with n ~ 2, then m ::s;; 2h - 3.
Hence in several instances we have defined a vertex Vi, for 1 ::s;; i ::s;; 5.
Should Vi exist for each i = 1,2, ... ,5, then G contains a sUbgraph that is
a subdivision of Ks.
Otherwise, for some i, there exist distinct vertices Wi and wi of Gi, each of 6.3 HAMILTONIAN PLANAR GRAPHS
which is connected to two of the Vij by internally disjoint (possibly trivial)
paths of Gi while Wi and wi are conneCted by a path of Gi , none of whose We have encountered many sufficient conditions for a graph to be
internal vertices are the vertices Vij' If two vertices Vij coincide, then this hamiltonian but only two necessary conditions. In this section we reverse
vertex is Wi' If the other two vertices Vij should also coincide, then this our point of view and consider a necessary condition for a planar graph to
vertex is wi· Without loss of generality, we assume that i = 1 and that be hamiltonian.
Wl is connected to V12 and Vl3' while w; is. connected to Vl4 and VlS as Let G be a hamiltonian plane graph of order n and let C be a fixed
described above. hamiltonian cycle in G. With respect to this cycle, a chord is, as before,
Denote the edge set of these five paths of G1 by E1• We riowturn to G2. If an edge of G that does not lie on C. Let ri (i = 3,4, ... , n) denote the
V21 = V24 = V25, we set E2 = 0; otherwise, there is a vertex W2 of G2 (which number of regions of G in the interior of C whose boundary contains
170 Planar graphs Hamiltonian planar graphs 171
However, N counts each interior chord twice and each edge of C once,
so that N = 2d + n. Thus,
11
G:
Lir; = 2d+ n. (6.3)
;=3

Substituting (6.2) into (6.3), we obtain

11 n
L ir; = 2 L r; - 2 + n,
;=3 ;=3

so
n
Figure 6.9 A hamiltonian plane graph. L (i - 2)r; = n- 2. (6.4)
;=3

r;
exactly i edges. Similarly, let denote the number of regions of G in By considering the exterior of C, we conclude in a similar fashion that
the exterior of C whose boundary contains i edges. To illustrate these n

definitions, let G be the plane graph of Figure 6.9 with hamiltonian L (i - 2)r; =n- 2. (6.5)
cycle C: Vb V6, V7, VB, V9, Vl0, VS, V4, V3, V2' Vl' Then r; = 0 if i =1= 4 and i=3
r4 = 4. Also, r; = 0 if i =1= 4, 5 while r~ = 1 and r~ = 2. It follows from (6.4) and (6.5) that
Using the notation of the previous paragraph, we have the following' n
necessary condition, due to Grinberg [G8J, for a plane graph to be
hamiltonian.
L (i - 2)(r; - rD = O. 0
;=3

The following observations often prove quite useful in applying

Theorem 6.17
Theorem 6.17. Let G be a plane graph with hamiltonian cycle C Further-
Let G be a plane graph of order n with hamiltonian cycle C Then with respect to more, suppose that the edge e of G is on the boundary of two regions R1
this cycle C, and R 2 of G. If e is an edge of C, then one of R 1 and R 2 is in the interior of C
n and the other is in the exterior of CIf, on the other hand, e is not an edge of
L (i - 2)(r; - ri) = o. C, then R 1 and R2 are either both in the interior of C or both in the exterior
;=3 of C.
In 1880, the English mathematician P. G. Tait conjectured that every
Proof 3-connected cubic planar graph is hamiltonian. This conjecture was dis-
proved in 1946 by Tutte [TIl], who produced the graph GinFigure6.10 as
We first consider the interior of CIf d denotes the number of chords of G a counterexample. In addition to disproving Tait's conjecture, Tutte [T14]
in the interior of C, then exactly d + 1 regions of G lie inside C Therefore, proved that every 4-connected planar graph is hamiltonian. This result
n was later extended by Thomassen [T5J.
Lr;=d+l,
;=3
Theorem 6.18.
implying that
Every 4-connected planar graph is hamiltonian-connected.
, d=(t r
1=3
;)-l. (6.2)
As an illustration of Grinberg's theorem, we now verify that the Tutte
graph (Figure 6.10) is not hamiltonian. Assume, to the contrary, that the
Let the number of edges bounding a region interior to C be summed Tutte graphG, which has order 46, contains a hamiltonian cycle C.
over all d + 1 such regions, denoting the results by N. Hence N = E~=3 ir;. Observe that C must contain exactly two of the edges e, hand h.
172 Planar graphs Hamiltonian planar graphs 173
For many years, Tutte's graph was the only known example of a 3-
connected cubic planar graph that was not hamiltonian. Much later, how-
ever, other such graphs have been found; for example, Grinberg himself
provided the graph in Exercise 6.12 as another counterexample to Tait's
conjecture.

G:
EXERCISES 6.3

6.12 Show, by applying Theorem 6.17, that the Grinberg graph (below) is
non-hamiltonian. .,

Figure 6.10 The Tutte graph.

Consider the regions Rv Rz and R3 of G. Suppose that two of them, say
RI and Rz, lie in the exterior of C. Then the edges h and Iz do not belong to
C since the unbounded region of G also lies in the exterior of C. This,
however, is impossible; thus at most one of the regions Rv R z and R3
lies in the exterior of C. We conclude that at least two of these regions,
say RI and R z, lie in the interior of C. This, of course, implies that their
common boundary edge e does not belong to C. Therefore, It and fz are
edges of C. Now let GI denote the component of G - {e, h, fz} containing
w. Then the cycle C contains a Vl-VZ subpath P that is a hamiltonian path
of G1 . Consider the graph Gz = GI + VI Vz. Then Gz has a hamiltonian cycle
Cz consisting of P together with the edge Vl Vz.
An application of Theorem 6.17 to Gz and Cz yields
1(r3 - r~) + 2(r4 - t{) +,3(r5 - r;) + 6(r8 - Is) = O. (6.6)
The Grinberg graph.
Since Vl Vz is an edge of Cz and since the unbounded region of Gz lies in the
exterior of C2I we have that 6.13 Show, by applying Theorem 6.17, that the Herschel graph (below) is
non-hamiltonian.
r3 - r~ = 1- 0 = 1 and r8 - r~ = 0 -1 =-1.
Therefore, from (6.6) we obtain
2(r4 - r~) + 3(r5 - r;) = 5.
Since degG2 W = 2, both WWl and WWz are edges of Cz . This implies that
r4 ~ I, so

~-t{=I-I=O m ~-t{=2-0=2
If r4 - r~ = 0, then 3(r5 - r~) =:' 5, which is impossible. If, on the other ,,<j'
hand, r4 - r4 = 2, then 3(r5 - r5) = I, again iinpossible. We conclude
that Tutte's graph is not hamiltonian. The Herschel graph.
174 Planar graphs Crossing number and thickness 175
6.14 Show, by applying Theorem 6.17, that no hamiltonian cycle in the'
graph of Figure 6.9 contains both the edges e and f.

6.4 CROSSING NUMBER AND THICKNESS

There are several ways of measuring how nonplanar a graph is. In this
section, we discuss two of these measures.
Nonplanar graphs cannot, of course, be embedded in the plane. Hence,
whenever a nonplanar graph is 'drawn' in the plane, some of its edges
must cross. This rather simple observation suggests our next concept.
The crossing number v( G) of a graph G is the minimum number of Figure 6.11 A drawing of Ks with one crossing.
crossings (of its edges) among the drawings of G in the plane. Before
proceeding further, we comment on the assumptions we are making a drawing (Figure 6.11) of Ks in the plane with one crossing so that
regarding the idea, of 'drawings'. In all drawings under consideration,
v(Ks) = 1.
we assume that '
The inequality v(K6 ) :::;;; 3 follows from Figure 6.12, where a drawing of
• adjacent edges never cross ~ with three crossings is shown. We now verify that II(K6) ~ 3, complet-
• two nonadjacent edges cross at most once ing the proof that v(K6 ) = 3. Let there be given a drawing of ~ in the
• no edge crosses itself plane with c = v(K6 ) crossings, where, of course, c ~ 1. At each crossing
• no more than two edges cross at a point of the plane and we introduce a new vertex, producing a connected plane graph G of order
• the (open) arc in the plane corresponding to an edge of the graph 6 + c and size 15 + 2c. By Corollary 6.4,
contains no vertex of the graph. 15 + 2c :::;;; 3(6 + c) - 6,
A few observations will prove useful. Clearly a graph G is planar if so that c ~ 3 and, consequently, v(K6 ) ~ 3.
and only if v(G) = O. Further, if G ~ H, then v(G) :::;;; v(H), while if H is Considerably more specialized techniques are required to verify
a subdivision of G, then II(G) = II(H). For very few classes of graphs is Theorem 6.19 for 7:::;;; n:::;;; 10.
the crossing number known. It has been shown by Blazek and Koman lt was mentioned in section 6.1 that every planar graph can be
[BK1] and Guy [G10], among others, that for complete graphs, embedded in the plane so that each. edge is a straight line segment.

v(K n ) :::;;; ~ l~J In; JIn; Jl

1 2 n; 3 j. (6.7)
Thus, if a graph G has crossing number 0, this fact can be realized by

and Guy has conjectured that equality holds in (6.7) for all n. As far
as exact results are concerned, the best obtained is the following (Guy
[Gll]).

Theorem 6.19
For 1:::;;; n:::;;; 10,

(6.8)

Since Kn is planar for 1 :::;;; n :::;;; 4, Theorem 6.19 is obvious for 1 :::;;; n :::;;; 4.
Further, Ks is nonplanar; thus, v(Ks ) ~ 1. On the other hand, there exists Figure 6.12 A drawing of K6 with three crossings.
176 Planar graphs Crossing number and thickness 177
considering only drawings in the plane in which the edges are straight ' Zarankiewicz [Z2] thus thought that he had proved
line segments. One may very well ask if, in general, it is sufficient to con-
sider only drawings of graphs inwhich edges are straight line segments in (6.10)
determining crossing numbers. With this question in mind, we introduce
a variation of the crossing number. but, in actuality, he only verified that the right hand expression of (6.10)
The rectilinear crossing number v(G) of a graph G is the minimum is an upper bound for v(Ks,t). As it turned out, both P. C. Kainen and
number of crossings among all those drawings of G in the plane in G. Ringel found flaws in Zarankiewicz's proof. Hence, (6.10) remains
which each edge is a straight line segment. Since the crossing number .only a conjecture. It is further conjectured that v(Ks,t) = v(Ks,t)· The best
v(G) considers all drawings of G in the plane (not just those for which general result on crossing number of complete bipartite graphs is the
edges are straight line segments), we have the obvious inequality following, due to the combined work of Kleitman [K71 and Woodall
[W11].
v(G) <s; v(G). (6.9)
As previously stated, v(G) = v(G) for every planar graph G. It has also Theorem 6.20
been verified that v(KI1 ) = v(KI1 ) for 1 <s; n <s; 7 and n = 9; however,
If sand t are integers (s <S; t) and either s <S; 6 or s = 7 and t <S; 10, then
v(Ks) = 18 and v(Ks ) = 19
(Guy [GIl]), so strict inequality in (6.9) is indeed a possibility.
We return to our chief interest, namely the crossing number, and
J
v(Ks,t) = l~J lS; 1J l~j It; j.1

consider the complete bipartite graphs. The problem of determining It follows, therefore, from Theorem 6.20 that
v(Ks,t) has a rather curious history. It is sometimes referred to as Turrin's
Brick-Factory Problem (named for Paul Tunin). We quote from Tunin [TID]: v(K3,t) = UJ It;l j, v(K4 ,t) = 2l~ Jlt; j,
1

6l~J It; 1 j
We worked near Budapest, in a brick factory. There were some kilns
where the bricks were made and some open storage yards where the v(Ks,t) = 4l~j It; 1 J and v(K6,t) =
bricks were stored. All the kilns were connected by rail with all
the storage yards. The bricks were carried on small wheeled trucks for all t. For example, v(K3,3) = I, V(E(4,4) = 4, v(Ks,s) = 16, v(K6,6) = 36
to the storage yards. All we had to do was to put the bricks on the and v(K7,7) = 81. A drawing of K4,4 with four crossings is shown in
trucks at the kilns, push the trucks to the storage yards, and unload Figure 6.13. .
them there. We had a reasonable piece rate for the trucks, and the As would be expected, the situation regarding crossing numbers of
work itself was not difficult; the trouble was only at the crossings. complete k-partite graphs, k ~ 3, is even more complicated. For the
The trucks generally jumped the rails there, and the bricks fell out most part, only bounds and highly specific results have been obtained
of them; in short this caused a lot of trouble and loss of time which in these cases. On the other hand, some of the proof techniques employed
was precious to all of us. We were all sweating and cursing at such
occasions, I too; but nolens volens the idea occurred to me that this loss
of time could have been minimized if the number of crossings of the
rails had been minimized. But what is the minimum number of cross-
ings? I realized after several days that the actual s~tuation could have
been improved, but the exact solution of the general problem with s
kilns and t storage yards seemed to be very difficult ... the problem
occurred to me again ... at my first visit to Poland where I met
Zarankiewicz. I mentioned to· him my 'brick-faCtory'-problem ...
and Zarankiewicz thought to have solved (it). But Ringel found a
gap in his published proof, which nobody has been able to fill so
far - in spite of much effort. This problem has also become a
notoriously difficult unsolved problem.... Figure 6.13 A drawing of ~,4 with four crossings.
 Planar graphs Crossing number and thickness 179
178
showed that v(C 3 x C3 ) = 3. Their proof consisted of the following three
have been enlightening. As an example, we establish the crossing number
of K Z,Z,3 (see White [W5], p. 77). steps:
Step 1. Exhibiting a drawing of C3 x C3 with three crossings so that
Theorem 6.21 v(C 3 x C3 ) ~ 3.
Step 2. Showing that C3 x C3 - e is nonplanar for every edge e of C3 x C3
The crossing number of KZ,Z,3 is v(KZ,Z,3) = 2. so that v(C 3 x C3 ) ): 2.
Step 3. Showing, by case exhaustion, that it is impossible to have a draw-
ing of C3 x C3 with exactly two crossings so that v(C3 x C3) ): 3
Proof
(Exercise 6.19).
Let V(K Z,2,3) = c. Since K 3 ,3 is nonplanar and K3,3 ~ KZ,Z,3, it follows that
Ringeisen and Beineke [RBI] then extended this result significantly by
K Z,Z,3 is nonplanar so that c ): 1. Let there be given a drawing of K Z,Z,3 in
the plane with c crossings. At each crossing we introduce a new vertex, determining v( C3 x Ct ) for all integers t ): 3.
producing a connected plane graph G of order n = 7 + c and size
m = 16 + 2c. By Corollary 6.4, m ~ 3n - 6. Theorem 6.22
Let UjUZ and VjVZ be two (nonadjacent) edges of KZ,Z,3 that cross in the
given drawing, giving rise to a new vertex. If G is a triangulation, then For all t ): 3,
C: Uj, VI' Uz, vz, Uv is a cycle of G, implying that the induced subgraph
({U1>UZ,Vl'VZ}) in K Z,Z,3 is isomorphic to K 4· However, KZ,Z,3 contains no
such subgraph; thus, G is not a triangulation so that m < 3n - 6. We have
Proof
16 + 2c < 3(7 + c) - 6,
We label the vertices of C3 x Ct by the 3t ordered pairs (O,j), (1,j) and
from which it follows that c ): 2. The inequality c ~ 2 follows from the fact (2,j), where j = 0,1, ... , t -I, and, for convenience, we let
that there exists a drawing of KZ,Z,3 with two crossings (Figure 6.14). 0 Uj = (O,j), Vj = (1,j) and Wj = (2,j).
Other graphs whose crossing numbers have been investigated with First, we note that v(C 3 X C t ) ~ t. This observation follows from the fact
little success are the n-cubes Qn' Since Q" is planar for n = 1,2,3, of that there exists a dtawing of C3 x Ct with t crossings. A drawing of
°
course, v(Q,,) = for such n. Eggleton and Guy [EGl] have shown that
V(Q4) = 8 but v(Qn) is unknown for n ): 5. One might observe that
C3 X C4 with four crossings is shown in Figure 6,15. Drawings of
C3 x Ct with t crossings for other values of t can be given similarly.
Q4 = Kz X K z X K z X Kz = C4 X C4 ,
To complete the proof, we show that v(C3 x Ct ) ): t. We verify this by
induction on t ): 3. For t = 3, we recall the preViously mentioned result
so that v(C4 x C4) = 8. This raises the problem of determining v(Cs x Ct ) v(C3 x C3 ) = 3.
for s, t ): 3. For the case s = t = 3, Harary, Kainen and Schwenk [HI<SI] Assume that v(C3 x Ck) ): k, where k): 3, and consider the graph
C3 x Ck + l' We show that v( C3 x Ck + 1) ): k + 1. Let there be given a
drawing of C3 x Ck+l with v(C3 x Ck+l) crossings. We consider two
cases.
Case 1. Suppose that no edge of any triangle Tj = ({Uj, Vj, Wj}), j = 0,1, ... ,k,
is crossed. For j = 0,1, ... , k, define
Hj = ({Uj, Vj, Wj, Uj+l> Vj+j, Wj+l}),
where the subscripts are expressed modulo k + 1. We show that for each
j = 0,1, ... ,k, the number of times edges of H j are crossed totals at least
two. Since, by assumption; no triangle Tj has an edge crossed and since
every edge not in any Tj belongs to exactly one subgraph Hjf it will follow
Figure 6.14 A drawing of KZ,2,3 with two crossings. that there are at least k + 1 crossings in the drawing because then every
180 Planar graphs Crossing numberand thickness '. 181
The only other result giving the crossing number of graphs Cs x Ct is
the following formula by Beineke and Ringeisen [BRn This theorem and
the succeeding theorems in this chapter are stated only to illustrate the
types of results obtained in this area.

Theorem 6.23
For all t ;::. 4,
v(C4 x Ct ) = 2t.
Beineke and Ringeisen [BRl] have also found a formula for v(K4 x C t )·

Theorem 6.24
For all t ;::. 3,
W3

Figure 6.15 A drawing of C3 x C4 with four crossings. In addition to the crossing number, another parameter that is interesting
for nonplanar graphs only is the thickness. The edge-thickness or simply the
crossing of an edge in H j involves either two edges of H j or an edge of H j thickness 81 (G) of a nonempty graph G is the minimum number of pairwise
and an edge of Hi for some i =1= j. edge-disjoint planar spanning subgraphs of G whose edge sets is a
If two of the edges UjUj+l, VjVj+l and WjWj+l cross each other, then two partition of E(G). This provides another measure of the nonplanarity of
edges of H j are crossed. Assume then that no two edges of H j cross each a graph. Once again, it is the complete graphs, complete bipartite graphs
other. Thus, Hj is a plane subgraph in the drawing of C3 x Ck+ 1 (Figure and n-cubes that have received the most attention.
6.16). The triangle Tj + 2 must lie within some region of H j • If Tj + 2 lies in A formula for the thickness of the complete graphs was established
a region of Hj bounded by a triangle, say Tj , then at least one edge of the primarily due tci the efforts of Beineke [B3], Beineke and Harary [BH2],
cycle uo, ul, ... , Uk> Uo, for example, must cross an edge of Tj , contradicting Vasak [VI] and Alekseev and Gonchakov [AGl].
our assumption. Thus, Tj +2 must lie in a region of Hj bounded by a 4-cycle,
sayuj' Uj+v Wj+v Wj' Uj, withoutloss of generality. However, then edges of
Theorem 6.25
the cycle Vo, VI!"" Vk, Vo must cross edges of the cycle Uj, Uj+l, Wj+l, Wj' Uj
at least twice and hence edges of Hj at least twice, as asserted. The thickness of Kn is given by
Case 2. Assume that some triangle, say To, has at least one of its edges crossed.
Suppose that v(C3 x Ck + 1 ) < k+ 1. Then the graph C3 x Ck + 1 - E(To),
which is a subdivision of C3 x Ck, is drawn with fewer than k crossings,
.,(K") ~ {P7 j n #9,10
n = 9,10.
contradicting the inductive hypothesis. 0
Although only partial results exist for the thickness of complete bipar-
"j ( } - - - - - - - - - - - - ( ) "j+l
tite graphs (Beineke, Harary and Moon [BHMl]), a formula is known for
the thickness of the n-cubes, due to Kleinert [K6].
~:
Theorem 6.26
The thickness of Qn is given by
~ ~+l
Figure 6.16 The subgraph Hj in the proof of Theorem 4.15.
r 11-
81 (QI!) = n ;
182 Planar graphs
CHAPTER 7
EXERCISES 6.4

6.15 Draw K7 in the plane with nine crossings.

6.16 Determine v(K3,3) without using Theorem 6.20.
Graph embeddings
6.17 Show that v(K7,7) ~ 81.
6.18 Determine v(K2,2,2)'
6.19 Determine v(K1,2,3)'
6.20 Show that 2 ~ v(C 3 x C3 ) ~ 3.
In Chapter 6 the emphasis was on embeddings of graphs in the plane.
Here this notion is extended to embeddings of graphs on other surfaces.
6.21 Prove that v(C3 x Ct ) = t for t ~ 3.
6.22 (a) It is known that v(W4 x K2 ) = 2, where W4 is the wheel C4 + K1
7.1 THE GENUS OF A GRAPH
of order 5. Draw W4 x K2 in the plane with two crossings.
(b) Prove or disprove: If G is a nonplanar graph containing an edge e
such that G - e is planar, then v(G) = 1. We now introduce the best known parameter involving nonplanar
graphs. A compact orientable 2-manifold is a surface that may be thought
6.23 Prove that 81 (Kn ) ~ L(n + 7)/6J for all positive integers n. of as a sphere on which has been placed a number of 'handles' or,
6.24 Verifythat81(KIl )= L(n+7)/6J forn=4,5,6,7,8. equivalently, a sphere in which has been inserted a number of 'holes'.
The number of handles (or holes) is referred to as the genus of the
surface. By the genus gen(G) of a graph G is meant the smallest genus of
all surfaces (compact or.ientable 2-manifolds) on which G can be
embedded. Every graph has a genus; in fact, it is a relatively simple
observation that a graph of size m can be embedded on a surface of
genus m.
Since the embedding of graphs on spheres and planes is equivalent, the
graphs of genus 0 are precisely the planar graphs. The graphs with genus
1 are therefore the nonplanar graphs that are embeddable on the torus.
The (nonplanar) graphs K s and K 3,3 have genus 1. Embeddings of K3 ,3 on
the torus and on the surface of genus 2 are shown in Figure 7.1(a) and (b).
Not only is Ks embeddable on the torus, but so are.l<t> and K7 • (The
graph Ks is not embeddable on the torus.) Figure 7.2 gives an embedding
of K 7 on the torus. The torus is obtained by identifying opposite sides of
the square. The vertices of K7 are labeled vo, VI, ... , V6' Thus note that the
'vertices' located at the corners of the square actually represent the same
vertex of K7 , namely the one labeled vs.
For graphs embedded on surfaces of positive genus, the regions and the
boundaries of the regions are defined in entirely the same manner as for
embeddings in the plane. Thus, if G is embedded on a surface 5, then the
components of 5 - G are the regions of the embedding. In Figure 7.1(a)
there are three regions, in Figure 7.1 (b) there are two regions and in Figure
7.2 there are 14 regions.
A region is called a 2-cell if any simple dosed curve in that region can
be continuously deformed or contracted in that region to a single point.
184 Graph embeddings The genus of a graph 185
An embedding of a graph Gon a surface 5 is called a 2-cell embedding
of G on 5 if all the regions so determined are 2-cells. The embeddings in
Figure 7.1 (a) and Figure 7.2 are both 2-cell embeddings.

Theorem 7.1
Let G be a connected (n, m) pseudograph with a 2-cell embedding on the surface of
genus g and having r regions. Then
n-m+r=2-2g. (7.1)

Proof
The proof is by induction on g. For g = 0, the formula holds for connected
graphs by Theorem 6.1. If G is a connected (n, m) pseUdograph (which is
(a)
not a graph) embedded in the plane and having r regions, then a plane
graph H is obtained by deleting from G all loops and all but one edge in
Figure 7.1 Embeddings of K3,3 on surfaces of genus 1 and 2. any set of multiple edges joining the same two vertices. If H has order nv
size ml and r1 regions, then nl - ml +rl = 2 by Theorem 6.1. We now add
back the deleted edges to form the originally embedded pseudograph G.
Equivalently, a region is a 2-cell if it is topologically homeomorphic to 2-
Note that the addition of each such edge increases the number of regions
dimensional Euclidean space. Although every region of a connected
graph embedded on the sphere is necessarily a 2-cell, this need not be by 1. IfG has k more edges than does H, then n = nv m = ml + k and
the case for connected graphs embedded on surfaces of positive genus. r = rt+k so that
Of the two regions determined by the embedding of K3,3 on the 'double n - m + r = nl - (ml + k) + (r1 + k)= nl - m1 +'1 = 2,
torus' in Figure 7.1 (b), one is a 2-cell and the other is not. The boundary of producing the desired result for g = O.
the 2-cell is a 4-cycle while the boundary of the other region consists of all Assume the theorem to be true for all connected pseudographs that are
vertices and edges of K3 ,3' Indeed, in the closed walk bounding the second 2-cell embedded on the surface of genus g - 1, where g > 0, and let G be a
region, five of the edges are encountered twice. connected (n, m) pseudograph that is 2-cell embedded on the surface 5 of
genus g and haVing r regions. We verify that (7.1) holds.
Since the surface 5 has genus g and g > 0, 5 has handles. Draw a
curve C around a handle of 5 such that C contains no vertices of G.
Necessarily, C will cross edges of G; for otherwise C lies in a region of
G and cannot be contracted in that region to a single point, contradicting
the fact that the embedding on 5 is a 2-cell embedding. By re-embedding
G on 5, if necessary, we may assume that the total number of intersections
of C with edges of G is finite, say k, where k > O. If e1, ez, ... ,et are
the edges of G that are crossed by C, then 1:::::; t:::::; k (Figure 7.3).
Moreover,
t
if edge ei, 1 :::::; i ::::; t, is .crossed by C a total of Ii funes, then
2:i=1 Ii = k:
At each of the k intersections of C with the edges of G we add a new
vertex; further, each subset of C lying between consecutive new vertices
is specified as a new edge. Moreover, each edge of Gthatis crossed by C,
v5 G ---l'--_J:.._..L-_-t_.....;:o v5 say a total of I times, is subdivided into t + 1 new edges.
Let the new pseudograph so formed be denoted by G'; further, suppose
l
Figure 7.2 An embedding of K7 on the torus. G' has order n , size m' and rl regions. Since k new vertices have been
186 Graph embeddings The genus of a graph 187
n" - m" + r" = 2 - 2(g - 1) or
(n + 2k) - (m + 3k) + (r + k + 2) = 2 - 2(g - 1);
thus,
n - m +r = 2 - 2g,
giving the desired result. 0

Restating Theorem 7.1 for graphs, we have the following.

Figure 7.3 A curve C drawn on a handle of the surface S. Corollary 7.2

Let G be a connected (n, m) graph with a 2-cell embedding on the surface ofgenus
introduced in forming G', it follows that n' = n + k. The curve C has
g and having r regions. Then
resulted in an increase of k in the number of edges. Also, each edg'~ ei,
1 ~ i ~ t, has given rise to an increase of t i edges and since 2:::=1 t j =k, n - m +r = 2 - 2g.
the total increase in siZe from G to G' is 2k; that is, m' = m + 2k.
Each portion of C that became an edge of G' is in a region of G. Thus, the In connection with Corollary 7.2 is the following result. Proofs of this
addition of such an edge divides that region into two regions. Since there theorem (e.g. Youngs [Yl}) are strictly topological in nature; we present
exist k such edges, r' = r + k. Because every region of G is a 2-cell, it no proof.
follows that every region of G' is a 2-cell.
We now make a 'cut' in the handle along C, separating the handle into Theorem 7.3
two pieces (as shown in Figure 7.4). The two resulting holes are 'patched'
or 'capped', producing a new (2-cell) region in each case. (This is called a If G is a connected graph embedded on the surface of genus gen(G), then every
'capping' operation.) region of G is a 2-cell.
In the process of performing this capping operation, several changes
have occurred. First, the surface S has been transformed into a new sur- Corollary 7.2 and Theorem 7.3 now immediately imply the following.
face S". The two capped pieces of the handle of S are now part of the
sphere of 5". Hence 5" has one less handle than 5 so that 5" has genus Theorem 7.4
g - 1. Furthermore, the pseudograph G' itself has been altered. The
vertices and edges resulting from the curve C have been divided into If G is a connected (n, m) graph embedded on the surface of genus gen(G) and
two copies, one copy on each of the two pieces of the capped handle. If haVing r regions, then
Gil denotes this new pseudograph, then G" has order n" = n + k = n + 2k
l
n - m +r =2 -2 gen(G).
and size m" = m' + k = m + 3k. Also, the number r" of regions satisfies
r" = r' + 2 = r + k + 2. Since each of these r" regions in the connected An important conclusion, which can be reached with the aid of
pseUdograph G" is a 2-celt the inductive hypothesis applies so that Theorem 7.4, is that every two embeddings of a connected graph G on
the surface of genus gen(G) result in the same number of regions. With
the theorems obtained thus far, we can now proVide a lower bound for the
genus of a connected graph in terms of its order and size.

Theorem 7.5
If G is a connected (n, m) graph (n ~ 3), then
m n
Figure 7.4 Capping a cut handle, gen( G) ~ 6' - 2: + 1.
188 Graph embeddings The genus of a graph 189
Proof The following corollary is a consequence of.the preceding result.
The result is immediate for n = 3, so we assume that n ~ 4. Let G be
embedded on the surface of genus gen(G). By Theorem 7.4, n - m + r = Corollary 7.9
2 - 2 gen(G), where r is the number of regions of G. (Necessarily, each of If G is a graph with components Gj , G2 , ••• , Gk , then
these regions is a 2-cell by Theorem 7.3.) Since the boundary of every k
region contains at least three edges and every edge is on the boundary
of at most two regions, 3r ~ 2m. Thus,
gen(G) = L gen(G;).
i=1
2m
2 - 2 gen(G) = n - m + r ~ n- m+ T' As is often the case, when no general formula exists for the value of a
parameter for an arbitrary graph, formulas (or partial formulas) are
and the desired result follows. 0 established for certa.in families of graphs. Ordinarily the first classes to
be considered are the complete graphs, the complete bipartite graphs,
The lower bound for gen(G) presented in Theorem 7.5 can be improved and the n-cubes. The genus offers no exception to this rule.
when more information on cycle lengths in G is available. The proof of the In 1968, Ringel and Youngs [RY1] completed a proof of a result that has
next theorem is entirely analogous to that of the preceding one. a remarkable history. They solved a problem that became known as the
Heawood Map Coloring Problem; this problem will be discussed in Chapter
Theorem 7.6 8. The solution involved the verification of a conjectured formula for the
genus of a complete graph; the proof can be found in (and, in fact, is) the
If G is a connected (n, m) graph with smallest cycle of length k, then book by Ringel [R8].
gen(G)~ ~(1-~)-~+1. Theorem 7.10

A special case of Theorem 7.6 that includes bipartite graphs is of special The genus of the complete graph is given by
interest. Recall that a graph is called triangle-free if it contains no triangles.
gen (Kn)
= r(n - 3)(n
12
- 4) 1
, n ~ 3.
Corollary 7.7
A formula for the genus of the complete bipartite graph was discovered
If G is a connected, triangle-free (n, m) graph (n ~ 3), then by Ringel [R7]. We shall also omit the proof of this result.
m n
gen(G) ~ '4-2+1.
Theorem 7.11
As one might have deduced by now, no general formula for the genus The genus of the complete bipartite graph is given by
of an arbitrary graph is known. Indeed, it is unlikely that such a formula
will ever be developed in terms of quantities that are easily calculable. On gen(Kr,s) = r(r - 2~S - 2) 1, r, s ~ 2.
the other hand, the following result by Battle, Harary, Kodama and
Youngs [BHKY1] i~plies that, as far as genus formulas are concerned,
one need only investigate blocks. We omit the proof. A formula for the genus of the n-cube was found by Ringel [R5] and by
Beineke and ~rary [BH1]. We prove this result to illustrate some of the
techniques involved. We omit the obvious equality gen(Qt) = O.
Theorem 7.8
If G is a graph having blocks B1 , B2 , .•• , Bk , then Theorem 7.12
k For n ~ 2, the genus of the n-cube is given by
gen(G) = :L gen(B j ).
gen(Qn) = (n - 4) .2n - 3 + 1.
;=1
 190 Graph embeddings 2-cell embeddings of graphs 191
Proof 7.3 Prove Theorem 7.6.
l
The n-cube is a triangle-free (2/l, n . 2/l- ) graph; thus, by Corollary 7.7, 7.4 Use Theorem 7.8 to prove Corollary 7.9.
gen(Qll) ~ (n - 4). 2/l- 3 + 1. 7.5 Show that

gen(K/l) ~
(n - 3)(n - 4) 1for n ~ 3.
To verify the inequality in the other direction, we employ induction on
n. For n ~ 2, define the statement A(n) as follows: The graph Qn can be r 12

embedded on the surface of genus (n - 4) ·2n - 3 + 1 such that the 7.6 Show that
boundary of every region is a 4-cycle and such that there exist 2,,-2
gen(Kr,s) ~
(r -2)(s -
4
2)1 for r, S ~ 2.
regions with pairwise disjoint boundaries. That the statements A(2) and
A(3) are true is trivial. Assume A(k - 1) to be true, k ~ 4, and, accordingly, r
let 5 be the surface of genus (k - 5) . 2k - 4 + 1 on which Qk-l is embedded 7.7 (a) Find a lower bound for gen(K3,3 + K/l)'
such
k 3
that the boundary of each region is a 4-cycle and such that there exist (b) Determine gen(K3 ,3 + Kit) exactly for n = 1,2 and 3.
2 - regions with pairwise disjoint boundaries. We note that since Qk-l
k 7.8 Determine gen(K2 x C4 x C6 ).
has order 2 -1, each vertex of Qk -1 belongs to the boundary of precisely
one of the aforementioned 2k - 3 regions. Now let Qk-l be embedded on 7.9 Prove, for every positive integer g, that there exists a connected
another copy 5' of the surface of genus (k - 5) .2k - 4 + 1 such that the graph G of genus g.
embedding of Qk-1 on 5' is a 'mirror image' of the embedding of Qk-l 7.10 Prove, for each positive integer k, that there exists a planar graph G
on 5 (that is, if VI' VZ, V3, V4 are the vertices of the boundary of a region of such that gen(G x Kz) ~ k.
Qk-l on 5, where the vertices are listed clockwise about the 4-cycle, then
there is a region on 5', with the vertices VI, Vz, V3, V4 on its boundary listed
counterclockwise). We now consider the 2k - 3 distinguished regions of 5
together with the corresponding regions of 5', and join each pair of 7.2 2-eELL EMBEDDINGS OF GRAPHS
associated regions by a handle. The addition of the first handle produces
the surface
k 3
of genus Z[(k - 5) . Zk-4 + 1] while the addition of each of the In the preceding section we saw that every graph G has a genus; that is,
other 2 - - 1 handles results in an increase of one to the genus. Thus, the there exists a surface (a compact orientable 2-manifold) of minimum
surface just constructed has genus (k - 4) . 2k - 3 + 1. Now each set of four genus on which G can be embedded. Indeed, by Theorem 7.3 ifG is a
vertices on the boundary of a distinguished region can be joined to the connected graph that is embedded on the surface of genus gen(G), then
corresponding four vertices on the boundary of the associated region so the embedding is necessarily a 2-cell embedding. On the other hand, if G
that the four edges are embedded on the handle joining the regions. It is is disconnected, then no embedding of G is a 2-cell embedding.
now immediate that the resulting graph is isomorphic to Qk and that Our primary interest lies with embeddings of (connected) graphs that
every region is bounded by a 4-cycle. Furthermore, each added handle are 2-cell embeddings. In this section, we investigate graphs and the
gives rise to four regions, 'opposite' ones of which have disjoint bound- surfaces on which they can be 2-cell embedded. It is convenient to
aries, so there exist 2k - Z regions of Qk that are pairwise disjoint. denote the surface of genus k by 5k • Thus, So represents the sphere (or
Thus, A(n) is true for all n ~ 2, proving the result. 0 plane), 51 represents the torus, and 52 represents the double torus (or
sphere with two handles). .
We have already mentioned that the torus can be represented as a
EXERCISES 7~1 square with opposite sides identified. More generally, the surface
5k (k> 0) carr be represented as a regular 4k-gon whose 4k sides can be
7.1 Determine g = gen(K4,4) without using Theorem 7.11 and label the listed in clockwise order as
regions in a 2-eell embedding of K4 ,4 on the surface of genus g. a1 b1al-lb-l . b -lb- l ' b -lb- l
1 a2 za2 z ... ak kak k, (7.2)
7.2 (a) Show that gen(G) ::;; v(G) for every graph G. where, for example, a1 is a side directed clockwise and all is a side also
(b) Prove that for every positive integer k, there exists a graph G labeled al but directed counterclockwise. These two sides are then
such that gen(G) = 1 and 1/( G) = k. identified in a manner consistent with their directions. Thus, the double
192 Graph embeddings 2-cell embeddings of graphs 193

G:

(a)

y a2 y
Figure 7.5 A representation of the double torus.

torus can be represented as shown in Figure 7.5. The 'two' points labeled
X are actually the same point on 52 while the 'eight' points labeled Yare,
in fact,a single point.
Although it is probably obvious that there exist a variety of graphs that (b) (c)
can be embedded on the surface 5k for a given nonnegative integer k,
.....
it may not be entirely obvious that there always exist graphs for which
a 2-cell embedding on 5k exists. ,. )
,.
Theorem 7.13
.....
..., -
For every nonnegative integer k, there exists a connected graph that has a 2-cell ""
embedding on Sk'
()

Proof
For k = 0, every connected planar graph has the desired property; thus, (d) (e)
we assume that k > O.
We represent 5k as a regular 4k-gon whose 4k sides are described and Figure 7.6 A graph 2-cell embedded on the torus.
identified as in (7.2). First, we define a pseUdograph H as follows. At each
vertex of the 4k-gon, let there be a vertex of H. Actually, the identification Figure 7.6 illustrates the construction given in the proof of Theorem
process associated with the 4k-gon implies that there is only one vertex of 7.13 in the case of the torus 51' The graph G so constructed is shown in
H. Let each side of the 4k-gon represent an edge of H. The identification Figure 7.6(a). In Figures 7.6(bHe) we see a variety of ways of visualizing
produces 2k distinct edges, each of which is a loop. This completes the the embedding. In Figure 7.6(b), a 3-dimensional embedding is
construction of H. Hence, the pseUdograph H has order 1 and size 2k. described. Ih Figures 7.6(c) and (d), the torus is represented as a
Furthermore, there is only one region, namely the interior of the polygon; rectangle with opposite sides identified. (Figure 7.6(c) is the actual
this region is clearly a 2-c'ell. Therefore, there exists a 2-cell embedding drawing described in the proof of the theorem.) In Figure 7.6(e), a
H on Sk' portion of G is drawn in the plane, then two circular holes are made
To convert the pseUdograph H into a graph, we subdivide each loop in the plane and a tube (or handle) is placed over the plane joining
twice, producing a graph G having order 4k + 1, size 6k, and again a the two holes. The edge uv is then drawn over the handle, completing
single 2-cell region. 0 the 2-cell embedding.
194 Graph embeddings 2-cell embeddings of graphs 195

)
Figure 7.7 A 2-cell embedding of Ks on the torus. V2

The graphs G constructed in the proof of Theorem 7.13 are planar. Figure 7.8 Tracing out a region.
Hence, for every nonnegative integer k, there exist planar graphs that
In the 2-cell embedding of Ks on 51 shown in Figure 7.7, there are
can be 2-c.e~1 e~bedded on 5k • It is also true that for every planar graph
five regions, labeled R1 ,Rz, ... ,Rs. Each region Ri (1 ~ i ~ 5) is, of
G and posztzve mteger k, there exists an embedding of G on Sk that is not a
course, a 2-cell. The boundary of the region R l consists of the vertices
2-~ell embedding: In general, for a given graph G and positive integer k
VI, Vz and Vs and the edges VIVt, VZVs and VSVI' If we trace out the edges
WIth k > gen(G), there always exists an embedding of G on 5k that is not a
of the boundary in a clockwise direction, that is, keeping the boundary at
2-cell embedding, which can be obtained from an embedding of G on
our left and the region to our right (Figure 7.8), beginning with the edge
Sgen(G) by adding k - gen(G) handles to the interior of some region of G.
VI Vz, we have VI Vz, followed by Vz Vs, and finally vSvl' This information can
If k = gen(G) and G is connected, then by Theorem 7.3 every embedding
also be obtained from the cyclic permutations 11"1, 1I"z, •.• ,1I"s; indeed, the
of G on 5k is a 2-cell embedding while, of course, if k < gen(G), there is no
edge following VI Vz = VZVl as we trace the boundary edges of R1 in a
embedding whatsoever of G on Sk'
Thus far, ';henever we have described a 2-cell embedding (or, in fact, clockwise direction is precisely the edge incident with Vz that follows
VZVl if one proceeds counterclockwise about Vz; that is, the edge following
any embeddmg) of a graph G on a surface 5k, we have resorted to a
VIVZ in the boundary of Rl is VZV1f2 (1) = vzvs. Similarly, the edge following
geometric description, such as the ones shown in Figure 7.6. There is a
vzvs = vsvz as we trace out the edges of the boundary of R1 in a clockwise
far more useful method, algebraic in nature, which we shall now discuss.
Consider the 2-cell embedding of Ks on 51 shown in Figure 7.7, with the direction is VSv1fs (Z) = VSVI ~ Hence with the aid of the cyclic permutations
11"1, 1I"z, .•. ,1I"s, we can trace out the edges of the boun~ary of R l • In a like
vertice~ o~ Ks lab~led as indicated. Observe that in this embedding the
manner, the boundary of every region of the embedding can be so
edges InCIdent WIth VI are arranged cyclically counterclockwise about
VI in the order VIVZ, VIV3, VIV4, VIVS (or, equivalently, VIV3, VIV4, VIVS,
described.
Since the direction (namely, clockwise) in which the edges of the
VIVZ' and s? on). ~his ind~ces a cyclic permutation 11"1 of the subscripts
boundary of a region are traced in the above description is of utmost
of the vertIces adjacent WIth VI' namely 11"1 = (2 3 4 5), expressed as a
importance, it is convenient to regard each edge of Ks as a symmetric
~ermutation cycle. Similarly, this embedding induces a cyclic permuta-
tIon 1I"z of the subscripts of the vertices adjacent with Vz; in particular, pair of ares and, thus, to interpret Ks itself as a digraph D. With this
1I"z = (1543). In fact, for each vertex Vi (1 ~ i ~ 5), one can associate a interpretation, the boundary of the region Rl and thus R1 itself can be
cyclic permutation 1I"i with Vi' In this case, we have described, starting at Vv as
(VI, Vz), (vz, V"2(1))' (Vs, V"S(Z))
11"1 = (2345),
or
1I"z = (1543),
(VI' Vz), (Vz, Vs),(Vs, VI)' (7.3)
11"3 = (1254), We now define a mapping 11": E(D) ~ E(D) as follows. Let a E E(D),
11"4 = (1 325), where a = (Vir Vj)' Then
1I"s = (1432). 1I"(a) = 1I"((Vi, Vj)) = 1I"(Vir Vj) = (Vj, V1fj (i))'
196 Graph embeddings 2-cell embeddings of graphs 197
The mapping 1r is one-to-one and so is a permutation of E(O). Thus, 1r can where for i = 1,2, ... , n, 1rj: V(i) ~ V(i) is a cyclic permutation that describes
be expressed as a product of disjoint permutation cycles. In this context, the subscripts of the vertices adjacent to Vi in counterclockwise order about Vi'
each permutation cycle of 1r is referred to as an 'orbit' of 1r. Hence (7.3) Conversely, for each such n-tuple (7l"j, 1rz, ... , 1r,,), there exists a 2-cell embedding
corresponds to an orbit of 1r and is often denoted more compactly as of G on some surface such that for i = 1,2, ... , n, the subscripts of the vertices
Vj - Vz - Vs - Vj' (Although this orbit corresponds to a cycle in the adjacent to Vi and in counterclockwise order about Vi are given by 1rj.
graph, this is not always the case for an arbitrary orbit in a graph that is
2-cell embedded.) For the embedding of Ks on S1 shown in Figure 7.7, the Proof
list of all five orbits (one for each region) is given below:
Let there be given a 2-cell embedding of G on some surface. For each
Rj : Vj - Vz - Vs - Vj, vertex Vj of G, define 1r( V(i) ~ V(i) as follows: If VjVj E E(G) and ViVI
Rz: Vj - V3 - Vz - V1, (possibly t = j) is the next edge encountered after VjVj as we proceed
counterclockwise about Vi, then we define 1riU) = t. Each 7l"i so defined
R3:Vj - V4 - V3 - V1, is a cyclic permutation.
R4:Vj - Vs - V4 - V1, Conversely, assume that we are given an n-tuple (1r1, 7l"z, ... ,1r,,) such
that for each i (1 ::<; i::<; n), 7l"i: V(i) ~ V(i) is a cyclic permutation. We
Rs: v2 - V3 - Vs - Vz - v4 - Vs - V3 - V4 - Vz· show that this determines a 2-cell embedding of G on some surface. (By
The orbits of 1r form a partition of E(O) and, as such, each arc of 0 necessity, this proof requires the use of properties of compact orientable
appears in exactly one orbit of 1r. Since 0 is the digraph obtained by 2-manifolds.)
replacing each edge of K s by a symmetric pair of arcs, each edge of K s Let 0 denote the digraph obtained from G by replacing each edge of G
appears twice among the orbits of 1r, once for each of the two possible by a symmetric pair of arcs. We define a mapping 1r: E(O) ~ E(O) by
directions that are assigned to the edge. 1r((vj, Vj)) = 1r(Vj, Vj) = (Vj, V"i(i))'
The 2-cell embedding of Ks on S1 shown in Figure 7.7 uniquely deter-
mines the collection {1r1' 1rz, ... , 1rs} of permutations of the subscripts of The mapping 1r is one-to-one and, thus, is a permutation of E(O). Hence, 1r
the vertices adjacent to the vertices of K s. This set of permutations, in turn, can be expressed as a product of disjoint permutation cycles. Each of these
completely describes the embedding of Ks on S1 shown in Figure 7.7. permutation cycles is called an orbit of 1r. Thus, the orbits partition the set
This method of describing an embedding is referred to as the Rotational E(O). Assume that
Embedding Scheme. Such a scheme was observed and used by Dyck [DI0] R: ((vj, Vj)(Vj' VI) ... (VilVi))
in 1888 and by Heffter [HI0]in 1891. It was formalized by Edmonds [El]
in 1960 and discussed in more detail by Youngs [Yl] in 1963. is an orbit of 1r, which we also write as
We now describe the Rotational Embedding Scheme in a more general R: Vi - Vj - VI - ... - VI - Vj'
setting. Let Gbe a nontrivial connected graph with V(G) = {V1' vz, .. . , v n }.
Let Hence, this implies that in the desired embedding, if we begin at,vi and
proceed along (Vi, Vj) to Vj' then the next arc we must encounter after
V(i) = {jlvj E N(Vi)}' (Vj, Vj) in a counterclockwise direction about Vj is (Vj, v,,'(i)) = (Vj,vI)'
For each i (1 ::<; i ::<; n), let 1ri: V(i) ~ V(i) be a cyclic permutation (or Continuing in this manner, we must finally arrive at the a~c (Vil VI) and
rotation) of V(i). Thus, each permutation 1ri can be represented by a return to Vi, in the process describing the boundary of a (2-cell) region
(permutation) cycle of length lV(i)1 = IN(vj)1 = degvi' The Rotational (considered as a subset of the plane) corresponding to the orbit R.
Embedding Scheme states that there is a one-to-one correspoI1denc~,. Therefore, each orbit of 1r gives rise to a 2-cell region in the desired
between the 2-cell embeddings of G (on all possible surfaces) and the n- embedding,
tuples (1r1' 1rz, ... ,1r n) of cyclic permutations. To obtain the surface S on which G is 2-cellembedded, pairs of regions,
with their boundaries, are 'pasted' along certain arcs; in particular, if
(Vi' Vj) is an arc on the boundary of R1 and (Vj, Vi) is an arc on the boundary
Theorem 7.14 (The Rotational Embedding Scheme)
of Rz, then (Vi, Vj) is id~ntified with (Vj, Vi) as shown in Figure 7.9. The
Let G be a nontrivial connected graph with V (G) = {V1' V2, ... , vn }. For each 2- properties of compact orientable 2-manifolds imply that S is indeed an
cell embedding of G on a surface, there exists a unique n-tuple (1r1' 1r2, ... ,1rn ), appropriate surface.
 198 Graph embeddirzgs 2-cell embeddings of graphs 199

Figure 7.9 A step in the proof of Theorem 7.14.

In order to determine the genus of 5, one needs only to observe that the
number r of regions equals the number of orbits. Thus, if G has order n
and size m, then by Corollary 7.2, 5 = 5k where k is the nonnegative
integer satisfying the equation n - m + r = 2 - 2k. 0

As an illustration of the Rotational Embedding Scheme, we once again

consider the complete graph Ks , with V(Ks ) = {Vl, V2, V3' V4, vs}. Let there
be given the 5-tuple (1Tl, 1T2, 1T3, 1T4, 1Ts), where (a)
1Tl = (2345),
1T2 = (1 345),
1T3 = (1245),
1T4 = (1 235),
1Ts = (1 234).
Thus, by Theorem 7.14, this 5-tuple describes a 2-cell embedding of Ks on
some surface 5k• To evaluate k, we consider the digraph D obtained by
replacing each edge of K s by a symmetric pair of arcs and determine
the orbits of the permutation 1T: E(D) -> E(D) defined in the proof of
Theorem 7.14. The orbits are
R 1: Vl -V2 -V3 -V4 -vs -Vl,

R2 : vl - v3 - v2 - v4 - v3 - Vs - V4 - Vl - Vs - V2 - Vb

R3 : Vl - V4 - V2 - Vs - V3 - vl;

and each orbit corresponds to a 2-eell region. Thus, the number of regions
in the embedding is r = 3. Since Ks has order n = 5 and size m = 10,
and since n - m + r = -2 = 2 -'- 2k, it follows that k = 2, so that the (b)
. given 5-tuple describes an embedding of Ks on 52'
Given an n-tuple of cyclic permutations as we have described, it is not Figure 7.10 A 2-cell embedding of Ks on the double torus.
necessarily an easy problem to present a geometric description of the
embedding, particularly on surfaces of high genus.· For the example just edge e2 = v3vS is drawn along H2 while e3 = Vl v3 is drawn along both H 1
presented, however, we give two geometric descriptions in Figure 7.10. and H2 • The three 2-cell regions produced are denoted by RlI R2 and R3 •
Figure 7.IO(a), a portion of Ks is drawn in the plane. Two handles are then In Figure 7.IO(b), this 2-cell embedding of K s on 52 is shown on the
inserted over the plane, as indicated, and the remainder of Ks is drawn regular octagon. The labeling of the eight sides (as in (7.2» indicates the
along these handles. The edge el = V2VS is drawn along the handle H 1, the identification used in producing 52'
200 Graph embeddings 2-cell embeddings of graphs 201
As a more general illustration of Theorem 7.14 we determine the genus so that h = (a -1)(b -1). Hence there is a 2-cell embedding of K 2a ,2b on
of the complete bipartite graph K2a ,2b' According to Theorem 7.11, S(a-l)(b-l), as we wished to show.
gen(K2a ,2b) = (a -l)(b -1). That (a -l)(b -1) is a lower bound for As a theoretical application of Theorem 7.14, we present a result that is
gen(K2a ,2b) follows from Exercise 7.6. We use Theorem 7.14 to show referred to as the Ringeisen-White Edge-Adding Lemma (see Ringeisen
that K2a ,2b is 2-cell embeddable on S(a-l)(b-l)' thereby proving that [R4]).
gen(K2a ,2b) ::::: (a - l)(b -1) and completing the argument.
Denote the partite sets of K2a ,2b by U and W, where lUI = 2a and Theorem 7.15
IWI = 2b. Further, label the vertices so that
Let G be a connected graph with V(G) = {VI> V2,' .. , vn } such that Vi and Vj are
U = {VI' V3, VS, ... , v4a-d and W = {V2, V4, V6,"" V4b}'
distinct nonadjacent vertices. Suppose that there exists a 2-cell embedding of G on
Let there be given the (2a + 2b)-tuple (assuming that a::::: b) some surface Sh with r regions such that Vi is on the boundary of region Rj and Vj
(1I"J, 11"2, ••• ,11"4a -1, 1I"4a, 11"40+2, 11"4a+4, ..•. , 1I"4b), is on the boundary of region Rj.· Let H = G + ViVj' Then
where (1) if Rj =f Rj,there exists a 2-cell embedding of Han Sh+ 1 with r - 1 regions in
11"1 = 1I"s = = 1I"4a-3 = (246 ... 4b), which Vi and Vj are on the boundary of the same region;
Oi) if Rj = Rj, there exists a 2-cell embedding of H on Sh with r + 1 regions in
11"3 = 11"7 = = 1I"4a-l = (4b ... 642), which each of Vi and Vj belongs to the boundaries of (the same) two distinct
11"2 = 11"6 = = 1I"4b-2 = (1 35 4a - 1), regions.
11"4 = 11"8 = = 1I"4b = (4a -1 531).
Proof
By Theorem 7.14, then, this (2a + 2b )-tuple describes a 2-cell embedding
of K2a 2b on some surface Sh' In order to evaluate h, we let D denote the By hypothesis, there exists a 2-cell embedding of the (n, m) graph G on Sh
digraph obtained by replacing each edge of K2a ,2b by a symmetric pair of with r regions such that Vi is on the boundary of region Rj and Vi is on the
arcs and determine the orbits of the permutation 11": E(D) --> E(D) defmed boundary of region R j . By TheOrem 7.14, an n-tuple (1I"J,1I"2, ... ,1I"n) of
in the proof of Theorem 7.14. cyclic permutations corresponds to this embedding, namely for
Every orbit of 11" contains an arc of the type (vs , VI)' where Vs E U and t = 1,2, ... , n, 11"1: Vet) ....... Vet) is a cyclic permutation of the subscripts of
VI E W. If 5 == 1 (mod4) and t == 2 (mod 4), then the resulting orbit R the vertices of N(vl) in counterclockwise order about VI'
containing (VS,v l) is Let D denote the symmetric digraph obtained from G by replacing each
R: V s - VI - v s +2 - VI-2 - v s , edge by a symmetric pair of arcs and let 11": E(D) --> E(D) be defined by
1I"(v", v(3) = (v(3' v ..p(a)' Since the given embedding has r regions, 11" has r
with 5 + 2 expressed modulo4a and t - 2 expressed modul04b. Note that
orbits. Denote each region and its corresponding orbit by the same
R also contains the arc (Vs+2' VI_2), where, then,s + 2 == 3 (mod4) and
symbol; in particular, R i and Rj are orbits of 11".
t - 2 == 0 (mod4). If s== 1 (mod4) and t == 0 (mod 4), then the orbit R'
Suppose that R i =f R i • We can therefore represent orbits Rj and R j as
containing (vs , VI) is
R': Vs -. VI - V s -2 - VI-2 - V S ,
Rj : Vi - Vk - ••• - Vk' - Vi
where, again, 5 - 2 is expressed modulo 4a and t - 2 is expressed and
modulo 4b. The orbit R' also contains the arc (Vs -2, VI_2), where Rj:Vj - V( - ••. - V(' - Vi'
5 - 2 == 3 (mod 4) and t - 2 == 2 (mod 4). Thus every orbit of 11" is either.
It therefore follows that
of the type R (where s== 1 (mod4) and t == 2 (mod 4» or the type R"
(where s== 1 (mod4) and t == 0 (mod4». Since there are a choices for s 1I"i(l() = k and 1I"/t') = t.
and b choices for tin eac1:l case, the total number of orbits is 2ab; therefore,
the number of regions in this embedding is r = 2ab. We now consider the graph H = G + ViVj and define
Since K2a ,2b has order n = 2a +-2b and size m = 4ab and because V'(t) = {r I VrVI E E(H)}
n - m + r = 2 - 2h, we have
for t=1,2, ... ,n. Thus V'(t)=V(t) for t#i,j, and V'(i) = V(i)U{j}
(2a + 2b) - 4ab + 2ab = 2 - 2h, while V'(j) = V(j) U {i}. For the graph H, we define an n-tuple
202 Graph embeddings 2-cell embeddings of graphs 203
(7r;, 1/z, ,7r~) of cyclic' permutations, where 7r;: V' (t) -+ V'(t) for Again we denote by D' the digraph obtained from H by replacing each
t = 1,2, ,n such that n; = 7rt for t f:. i,j. Furthermore, edge of H by a symmetric pair of arcs and define the permutation
7ri(a) ifai=k' 7r': E(D') -+ E(D') by 7r'(v a,v/3) = (v/3,v"/i(aj)' The orbits of 7r' consist of
all orbits of 7r different from Ri together with the orbits
7I;(a) ={ ~ ifa=k'
if a = j R': Vi - Vj - v( - ... - vk' - Vi

and and
7rj(a) if a f:. t' R" : Vj - Vi - Vk - ... - Vt' - Vj'
7rj(a)= ~ ifa=t' Therefore, 7r' has r + 1 orbits and the resulting 2-cell embedding of H
{
~ ifa=i. has r + 1 regions. Furthermore, each of Vi and Vj belongs to the boundaries
of both R' and R". Here n - m + r = 2 - 2himplies that n - (m + 1) +
Let D' be the digraph obtained from H by replacing each edge of H by (r + 1) = 2 - 2h, and H is 2-cell embedded on Sh, which verifies (ii). 0
a symmetric pair of arcs. Define the permutation 7r': E(D') -+ E(D') by
7r' (von v,B) = (v,B, v"/i{a)' The orbits of -rr' then consist of all orbits of 7r
A consequence of Theorem 7.15 will prove to be useful.
different from R i and Rj together with the orbit
R: Vi - Vj - v( - ... - VI' - Vj - Vi - Vk - ... - vk' - Vi'
Corollary 7.16
Thus, -rr' has r - 1 orbits and the corresponding 2-cell embedding of H has
Let eand fbe adjacent edges ofa connected graph G.lf there exists a 2-cell embed-
r - 1 regions. Moreover, Vi and Vj lie on the boundary of R. Since
ding of G' = G - e - f with one region, then there exists a 2-cell embedding of G
n - m + r = 2 - 2h, it follows that n (m + 1) + (r - 1) = 2 - 2(h + 1)
with one region.
and H is 2-cell embedded on Sh+l' This completes the proof of (i).
Suppose that R i = Rj • We can represent the orbit Ri(=Rj ) as
R i : Vi - Vk - ... - VI' - Vj - V( - ... - Vk' - Vi' Proof

(Note that Vi and Vj cannot be consecutive in Ri since ViVj ~ E(G).) It

Let e = uv and f = vw, where then u f:. w. Let there be given a 2-cell
follows that embedding of G' with one region R. Thus all vertices of G' belong to the
boundary of R, including u and v. By Theorem 7.15(ii), there exists a 2-cell
7ri(k') = k and 7rj(t') = t. embedding of G' + e with two regions where u and V lie on the boundary
We again consider the graph H = G + ViVj and once more define of both regions. Therefore, V is on the boundary of one region and w is on
the boundary of the other region in the 2-cell embedding of G' + e. Apply-
V' (t) = {r I VrVt E E(H)} ing Theorem 7.15(i), we conclude that there exists a 2-cell embedding of
for t = 1,2, ... , n. We define an n-tuple (n;, 7r~; ... , -rr'n) of cyclic permuta- G' + e + f = G with one region. 0
tions, where 7r;: V' (t) -+ V' (t) for t = 1,2, ... , n such that 7r;::= 7rt for
t f:. i, j. Also, We now turn our attention for the remainder of the section to the
following question: Given a (connected) graph G, on which surfaces Sk
{ .,(aJ ifai=k'
do there exist 2-cell embeddings of G? As a major step towards answer-
7I;(a) = ~ ifa=k' ing this question, we present the following 'interpolation theorem' of
ifa=j puke [D9].
and

r<'l if a f:. t' Theorem 7.17

7rj(a) = ~ ifa=t' ·If there exist 2-cell embeddings ofa connected graph G on the surfaces Sp and Sq,
if a = i. where p ~ q, and k is any integer such that p ~ k ~ q, then there exists a 2-cell
embedding of G on the surface Sk'
204 Graph embeddings 2-cell embeddings of graphs 205
Proof Case 1. Assume that all arcs of EI belong to a single orbit R of 1f. Suppose, first,
that the orbit R has the form
Observe that there exist 2-cell embeddings of K I only on the sphere; thus,
we assume that G is nontrivial. R: VI - Vy - ... - Vb - VI - v~ - ... - Va - VI - Vb - .•. - V x - VI'
Assume that there exists a 2-cell embedding of G on some surface St. Let
"f{ere the orbits of'lr' are the orbits of 'lr except that the orbit R is replaced by
V(G) = {Vb Vz, ... , V n }, n ;;. 2. By Theorem 7.14, there exists an n-tuple the three orbits
(7I"I,7I"Z, ... ,7I"n) of cyclic permutations associated with thisembeddmg
such that for i = 1,2, ... , n, 7I"j: V(i) -7 V(i) is a cyclic permutation of the R;: VI - Vy - - Vb - VI'
subscripts of the vertices of N(vj) in counterclockwise order about Vj' R~: VI - Vc - - Va - Vb
Let 0 be the symmetric digraph obtained from G by replacing each
edge by a symmetric pair of arcs. Let 71": E(O) -7 E(O) be the. pe:mutation R~: VI - Vb - ••. - V x - VI'
defined by 7I"(Vi, vi) = (Vi' v".(i))' Denote the number of orbIts In 71" by r;
{Ience, 'lr' describes a 2-cell embedding of G with r + 2 regions on a surface
that is, assume that there a;e r 2-cell regions in the given embedding of
5'. Necessarily, then, S' = Sf-I'
G on St.
The other possible form that the orbit R may take is
Assume there exists some vertex of G, say Vv such that deg VI ;;. 3. Then
1fI = (a b c ...), where a, band c are distinct. Let V x be any vertex adjacent R: VI - v y - ••• - Va - VI - vb- •.. - Vb - VI - Vc - ... - V x - VI'
with VI other than Va and Vb, and suppose that 1fI (x) = y. Thus
Illthis situation, the orbits of n' are the orbits of 1f, except for R, which is
7l'I = (a be ... x y ...), replaced by the orbit

where, possibly, x = c or y = a. Let EI be the subset of E(D) consisting of

R': VI - v y - ... - Va - VI - Vc - •.• - V x - VI - Vb - ... - Vb - VI'

the three pairs of arcs lienee, 'lr' has r orbits.

(va, VI), (VI, Vb); (Vb, VI), (VI' V c ); (v x , VI), (VI' Vy )' (7.4) Case 2. Assume that 7l' has two orbits, say R1 and Rz, with R1 containing two
of the pairs of arcs in E1 , and Rz containing the remaining pair of arcs. In this
Note that the six arcs listed in (7.4) are all distinct. By the definition of the case, the orbits of 'lr' are those of 'lr, except for R1 and Rz, which are
permutation 7l', we have replaced by two orbits R;
and R~, where one of R; and R~ contains two
7l'(Va , VI) = (VbVb), 7l'(Vb, VI) = (VbVc) and 7I"(Vx ,Vl) = (Vl'V y )' arcs of E1 and the other contains the remaining four arcs of E1• In this case,
i' has r orbits.
This implies that the arc (va, VI) is followed by the arc (VI, Vb) in some orbit
~ase 3. Assume that 1f has three orbits R1 , Rz and R3 such that (va, VI) is
of 71", and that the edge VaVl of G is followed by the edge VI Vbas we proceed
[allowed by (VI' Vb) in R 1, (Vb, VI) is followed by (VbVc) in Rz, and (V x ,Vl) is
clockwise around the boundary of the corresponding region in the given
followed by (Vb V y ) in R3 . In this case, the orbits of 'lrJ are the orbits of'lr,
embedding of G inS,. Also, (Vb, VI) is followed by (VI, V c) in some orbit of
except for R v Rz <:Ind R 3 , which are replaced by a single orbit R' of the form
'lr and (v x , VI) is followed by (VI' V y) in some orbit.
We now define a new permutation 7l": E(D) -7 E(D) with the aid of the R': VI -Vy - ... - V x - VI - Vb - ... - Va - VI - V c - ... - Vb - VI'
n-tuple (n;, ni, ... ,n'n), where for i = 1,2, ... , n, 1f:: V(i) -7 V(i) is a cyclic
In this case, 'lr' has r - 2 orbits so that 7l" describes a 2-cell embedding of G
permutation defined by
on S'+l'
, {(ac ... xb Y . .,) ifi=l Thus, we can now conclude that the shifting of a single term in 1fl
'lrj = 1f; if 2 ::::; i ::::; n. (producing n;) changes the genus of the resulting surface on which G is
f7Fell embedded by at most 1. Having made this observation, we can now
We then define n'(vj, Vj) = (Vi' v,r,(j)). By Theorem 7.14, the n-tuple complete the proof.
(1f;, 'lr~, ... , 'lr~) determines a 2-cellJ embedding of G on some surface, Let (P,1' P,z, ... , P,n) be the n-tuple of cyclic permutations associated with
where for i = 1,2, ... , n, 1f: is a cyclic permutation of the subscripts of a 2"cell embedding of G on 5p and let (VI ,Vz, ... , v n ) be the n-tuple of cyclic
vertices adjacent to Vj in counterclockwise order about Vi' permutations associated with a 2-cell embedding ofG on Sq. If degvi is 1
Three cases are now considered, depending on the possible distribution or 2 for each i, 1 ~ i ~ n, then !J.j= Vj so thatp = q and the desired result
of the pairs (7.4) of arcs in E1 among the orbits of 7l'. ~follows. Hence, we may assume that for some i, 1 ::::; i ::::; n, deg Vj ;;. 3. For
206 Graph embeddings The maximum genus of a graph 207
each such i, J..Li can be transformed into Vi by a finite number of single 7.3 THE MAXIMUM GENUS OF A GRAPH
term shifts, as described above. Each such single term shift describes art
embedding of G on a surface whose genus differs by at most 1 from If G is a connected graph with gen(G) = p, and q is the largest positive
the genus of the surface on which G is embedded prior to the shift. integer such that G is 2-cell embeddable on 5q, then it follows from
Therefore, by performing sequences of single term shifts on those fLi Theorem 7.17 that G can be 2-cell embedded on 5k if and only if
for which deg Vi ~ 3, the n-tuple (J..L1, J..L2, ... ,J..Ln) can be transformed into p ~ k ~ q.- This suggests the following concept.
(v], V2,"" vn ). Since p ~ k ~ q, there must be at least one term Let G be a connected graph. The maximum genus genM(G) of G is the
(1T], 1T2,' .. , 1Tn ) in the aforementioned sequence beginning with maximum among the genera of all surfaces on which G can be 2-cell
(J..L1>J..L2, ... ,J..Ln) and ending with (V1,V2,''''Vn) that describes a 2-cell embedded. At the outset, it may not even be clear that every graph has
embedding of G on 5k • 0 a maximum genus since, perhaps, some graphs may be 2-cell embeddable
on infinitely many surfaces. However, there are no graphs that can be 2-
cell embedded on infinitely many surfaces, for suppose that G is a non-
EXERCISES 7.2 trivial connected graph with V(G) = {V1, V2,"" v n}. By Theorem 7.14,
there exists a one-to-one correspondence between the set of all 2-cell
7.11 (a) For the 2-cell embedding of K3 ,3 shown in Figure 7.l(a), deter- embeddings of G and the n-tuples (1T1>1T2, ... ,1Tn), where for
mine the 6-tuple of cyclic permutations 1Ti associated with i = 1,2, ... , n,1Ti: V(i) -+ V(i) is a cyclic permutation. Since the number
this embedding. Determine the orbits of the resulting permu- of such n-tuples is finite, and in fact is equal to
n
tation 1T.
(b) For the 2-cell embedding of K7 on 51 shown in Figure 7.2, II (degv; - I)!,
i=]
determine the 7-tuple of cyclic permutations 1Ti associated with
this embedding. Determine the orbits of the resulting permu- it follows that there are only finitely many 2-cell embeddings of G and
tation 1T. therefore that there exists a surface of maximum genus on which G can
be 2-cell embedded. We can now state an immediate consequence of
7.12 Let G = Kt X K2 • Theorem 7.17.
(a) Show that G is nonplanar.
(b) Show, in fact, that gen(G) = 1 by findingan 8-tuple of cyclic Corollary 7.18
permutations that describes a 2-cell embedding of G on 51'
Determine the orbits of the resulting permutation 1T. A connected graph G has a 2-cell embedding on the surface 5k if and only if
7.13 Let G be a graph with V(G) = {V1,V2,V3,V4,VS,V6} and let gen(G) ~ k ~ genM(G).
(1T1, 11"2, 11"3, 1T4, 1I"s, 11"6) describe a 2-cell embedding of G on the surface We now present an upper bound for the maximum genus of any con-
5kt where
nected graph. This bound employs a new but very useful concept.
1Tt = (2563), 1T2 = (36 1 4), 1T3 = (41 25), The Betti number B(G) of an (n, m) graph G having k components is
defined as
1T4 = (5236), 11"5 = (6341), 1T6 = (1 452).
B( G) = m - n + k.
(a) What is this familiar graph G?
Thus, if G is connected, then
(b) What is k?
(c) Is k = gen(G)? B( G) = m - n + 1.
7.14 How many of the 2-cell embeddings of Kt are embeddings in the The following result is due to Nordhaus, Stewart and White [NSW1j.
plane? On the torus? On the double toms?
Theorem 7.19
7.15 (a) Describe an embedding of K3,3 on 52 by means of a 6-tuple of
cyclic permutations. IfG is a connected graph, then
(b) Show that there exists no 2-cell embedding of K3,3 on 53'
genM(G) ~ lB~G) J
208 Graph embeddings The maximum genus of a graph 209
Furthermore, equality holds if and only if there exists a 2-cell embedding of G
on the surface of genus genM(G) with exactly one or two regions according to
whether B(G) is even or odd, respectively.

Proof
Let G be a connected (n, m) graph that is 2-cell embedded on the surface of G:
genus genM(G), producing r (2-cell) regions. By Theorem 7.1,
n - m+r =2 - 2genM(G).
Thus,
B(G) = m - n + 1 = 2genM(G) + r -1,
so that Figure 7.11 Splitting trees of graphs.

Theorem 7.21
producing the desired bound. Let T bea splitting tree ofan (n,m) graph G. Then every component ofG ~ E(T)
Moreover, we have has even size ifand only if B(G) is even.

G) - B(G) + 1- r -lB(G)J
genM ( - 2 - 2 ' Proof

if and only if r = 1 (which can only occur when E(G) is even) or r = 2 Suppose that every component of G - E(T) has even size. Then G - E(T)
(which is only possible when B(G) is odd). 0 has even size. Since every tree of order n has size n - I, the graph
G - E(T) has size m - (n -1) = m - n + 1. Therefore, B(G) = m:- n + 1
A (connected) graph G is called upper embeddable if the maximum genus is even.
of G attains the upper bound given in Theorem 7.19; that is, if Conversely, suppose that B(G) is even. The graph G - E(T) has size
genM(G) = LB(G)/2J. The graph G is said to be upper embeddable on a m - n + 1 = B(G). Since T is a splitting tree of G, at most one component
surface S if S = SgenM(G)' We can now state an immediate consequence of of G - E(T) has odd size. Since the sum of the sizes of the components of
Theorem 7.19. G - E(T) is even, it is impossible for exactly one such component to have
odd size, producing the desired result. 0
Corollary 7.20
We now state a characterization of upper embeddable graphs, which
Let G bea graph with even (odd) Betti number. Then G is upper embeddable on a was discovered independently by Jungerman [J4l and Xuong [Xl].
surface S if and only if there exists a 2-cell embedding of G on S with one (two)
region(s).
Theorem 7.22
In order to present a characterization of upper embeddable graphs, it is A graph G is upper embeddable if and only if G has a splitting tree.
necessary to introduce a new concept.
A spanning tree T of a connected graph G is a splitting tree of G if at most Returning to the graph G of Figure 7.11, we now see that G is
one component of G - E(T) has odd size. It follows therefore that if upper embeddable since G contains T1 as a splitting tree. On the other
G - E(T) is connected, then T is a splitting tree. For the graph G of hand, neither the graph G1nor the graph ~ of Figure 7.12 has a single
Figure 7.11, the tree T 1 is a splitting tree. On the other hand, T2 is not f'iplitting tree, so, by Theorem 7.22, neither of these graphs is upper
splitting tree of G. embeddable.
The following observation that relates splitting trees and Betti numbers We mentioned earlier that no formula is known for the genus of an
is elementary, but useful. arbitrary graph; However, such is not the case with maximum genus.
210 Graph embeddings The maximum genus of a graph 211
even. Hence, by Corollary 7.20, H can be 2-cell embedded on SgeuM(H) with
one region. Adding the edges ell e2," ., e~(G) to H produces the graph G.
By Theorem 7.15, there exists a 2-cell embedding of H + e] on some
~urface (namely on SgenM(H) in this case) with two regions. By e(G) appli-
cations of Theorem 7.15, it follows that there exists a 2-cell embedding of
G = H + e] + e2 + ... + e{(G) on some surface 5 with at most 1 + e(G)
regions. Therefore, if G is 2-cell embedded on 5 with s regions, then
necessarily s ~ 1 + t;(G). Since the minimum number of regions of any
2-cell embedding of G occurs when G is 2-cell embedded on SgenM(G)
and since such an embedding produces r regions, by assumption, we
conclude that r ~ s so that r ~ 1 + e(G).
Figure 7.12 Graphs that are not upper embeddable. To verify that r ~ 1 + t;(G), we again assume that G is 2-cell embedded
on the surface of genus genM(G) with r(~2) regions. Let h be an edge
With the aid of Theorem 7.22, Xuong [Xl] developed a formula for the belonging to the boundary of two regions of G. (Necessarily, It is not a
maximum genus of any connected graph. bridge of G.) Then G - It is 2-cell embeddable on the surface of genus
For a graph H we denote by eo(H) the number of components of H genM(G) with. r -1 regions. Furthermore, if r> 2, then for
of odd size. For a connected graph G, we define the number {(G) as k = 2,3, ... , r - 1, let A be an edge belonging to the boundary of two
follows: regions of G - {It ,12, ... ,A-d. then for k = 1,2, ... , r -I, the graph
G - {1t,f2"" ,M is 2-cell embeddable on the surface of genus genM(G)
{(G) = min {o(G - E(T)), with r - k regions; in particular, the graph G' = G - {h,f2"" ,fr-d is
where the minimum is taken over all spanning trees T of G. 2-cell embeddable on the surface on genus genM(G) with one region.
1:'herefore, B(G') is even and, by Corollary 7.18, the graph G' is upper
embeddable on the surface of genus genM(G). By Theorem 7.22, G' con-
Theorem 7.23
tains a splitting tree T', and all components of G' - E(T') have even size.'
The maXimum genus of a connected graph G is given by Thus, eo(G - E(T')) ~ r -1. Consequently, e(G) ~ {o(G - E(T')) ~ r-1
so that r ~ 1 + e(G). Therefore, r = 1 + t;(G).
genM(G) = !CB(G) ~ {(G)). By Theorem 7.1,
n - m +r = 2 -2genM(G).
Proof
Assume that G has order n and size m. Let G be 2-cell embedded on the Since r = 1 + e(G), it follows that
surface of genus genM(G) such that r regions are produced. First we show 2genM(G) = m - n + 1 - {(G)
that r = 1 + {(G). Note that if {(G) = 0, then G contains a splitting tree and
B( G) is even. Thus, G is upper embeddable on SgenM(G) with one region; or
that is, r = 1 + {(G)if t;(G) = O. Similarly, if r = 1, then B(G) is even and G genM(G) =!(B(G) -{(G)). 0
is upper embeddable. Thus G has a splitting tree andt;(G) = 0, so
r = 1 + e(G) if r = 1. Therefore, we assume that e(G) > 0 and r ~ 2. Returning to the graph G] of Figure 7.12, we see that B(G]) = 6 and
Let T] be a spanning tree of G such that that co(G] - E(T)) = 6 for every spanning tree T. Therefore, €(G]) = 6
eo(G - E(T1 )) = {(G). so that
Let Gi , i = 1,2, ... ,t;(G)~ be the components of odd size in G - E(T]). For genM(G]) =! (B(G]) - t;(G1 )) =0
i = 1,2, ... , t;(G), let ej be a pendant edge of Gj if Gj is a tree and let ej be a
and G] is 2-cell embeddable only on the sphere.
cyc.le edge of. Gj if Gj is not a tree. Defin~ H = G - {eb e2, .•• , e~(G)}' Since
T] IS a spannmg tree of H, the graph H IS connected. Since every compo~ With the aid of Theorem 7.23 (or Theorem 7.22), it is possible to show
nent of H - E(T1 ) has even size,I1 is a splitting tree of H. Therefore, by that a wide variety of graphs are upper embeddable. The following result
Theorem 7.22, H is upper embeddable. Also, by Theorem7.21, B(H) is is due to Kronk, Ringeisen and White [KRWl].
212 Graph embeddings The maximum genus of a graph 213
Corollary 7.24
Every complete k-partite graph, k ~ 2, is upper embeddable.

From Corollary 7.24, it follows at once that every complete graph is

upper embeddable, a result due to Nordhaus, Stewart and White
[NSW1] .. We present a proof using Theorem 7.22.

Corollary 7.25
The maximum genus of Kn is given by

genM (Kn ) -l(n

-
-l)(n
4 - 2) .
J

Proof
If T is a spanning path of 1(", then Kn - E(T) contains at most one nontri-
vial component. Therefore, T is a splitting tree of K n and, by Theorem 7.22,
Kn is upper embeddable. Since B(Kn ) = (n -l)(n - 2)(2, the result fol-
Figure 7.13 An embedding of Ks on the Mobius strip.
lows. 0

A formula for the maximum genus of complete bipartite graphs was EXERCISES 7.3
discovered by Ringeisen [R4].
7.16 Describe an embedding of Ks on SgenM(Ks) by means of a 5-tuple of
Corollary 7.26 cyclic permutations.

The maximum genus of Ks,1 is given by 7.17 Determine the maximum genus of the graph Gz of Figure 7.12.
7.18 Determine the maximum genus of the Petersen graph.
genM (Ks,1 ) -l(S-1)(t-1)J
- 2 . 7.19 (a) Let G be a connected graph with blocks B1 , Bz, ... ,Bk • Prove that
k
Zaks [21] discovered a formula for the maximum genus of the n-cube. genM(G) ~ L: genM(B;).
i=1

Corollary 7.27 (b) Show that the inequality in (a) may be strict.
7.20 Prove Theorem 7.22 as a corollary t6Theorem 7.23.
The maximum genus of Qn, n ~ 2, is given by
7.21 Prove Corollary 7.24.
genM(Qn) = (n - 2)2n - z.
7.22 Prove Corollary 7.26.
Finally, we also note that it is possible to speak of embedding graphs on 7.22 Prove Corollary 7.27.
nonorientable surfaces such as the Mobius strip, projective plane and
Klein bottle. As might be expected, every planar graph (as well as ~o111e. 7.24 Prove or disprove: For every positive integer k, there exists a con-
nonplanar graphs) can be embedded on such surfaces. Figure 7.13 shows nected graph Gk such that LB(G)(2J - genM(Gk ) = k.
Ks embedded on the Mobius strip. These topics shall not be the subject of 7.25 Prove or disprove: If H is a connected spanning subgraph of an
further discussion, however. upper embeddable graph G, then H is upper embeddable.
214 Graph embeddings Voltage graphs 215
7.26 For G = Cs x Ct (5, t ~ 3), determine gen(G) and genM(G).
7.27 Prove that if each vertex of a connected graph G lies on at most one
cycle, then G is only 2-cell embeddable on the sphere. H:
7.28 Prove, for positive integers p and q with p ~ q, that there exists a
graph G of genus p that can be 2-cell embedded on 5q•
7.29 Prove that if G is upper embeddable, then G x K2 is upper embed-
dable.
(a) (b)
Figure 7.15 A voltage graph (with an embedding of it on 51)'
7.4 VOLTAGE GRAPHS
embedding of H on 51 in which we are interested is shown in Figure
We have already seen that describing an embedding of a given graph on a 7.15(b). Notice that when we embed a pseudodigraph on 51 we do not
given surface is potentially complicated. However, if the graph is suffi- show the usual identification arrows for the torus; it is understood that
ciently symmetric, then there may be another approach available to us. this embedding is on 51' We will show that this embedding of H contains
In Figure 7.2 we described an embedding of K7 on the torus 51' Since the necessary information to reproduce the original embedding of K7 on
gen(K7 ) = I, this embedding is, of course, a 2-cell embedding. By 5 t shown in Figure 7.2. To see how this is accomplished, we begin by
Theorem 7.1, it follows that n - m + r = 0 for this embedding. Indeed, finding the cyclic permutation associated with the arcs incident with
in this case, n = 7, m = 21 and r = 14, that is, 7 - 21 + 14 = O. Notice vertex 0 in the embedding of H given in Figure 7.15(b).
that 7 is a factor of each of these numbers; so we can write this expression Starting with the arc labeled I, where it is directed away from vertex 0,
as 7(1 - 3 + 2) = 0, which might suggest considering a 2-cell embedding we proceed counterclockwise about the vertex 0 and record either the
of a 'graph' of order 1 and size 3 on 51 that results in two regions. Of element of ~ on each arc encountered or its inverse depending on
course, there is no graph of order 1 and size 3, but there is a pseUdograph whether the arc is directed away from or toward 0, respectively. Thus
with these properties, namelythe pseUdograph G shown in Figure 7.14(a), 11'0 begins with the element 1 and is followed by 5 (the inverse of 2 since
where the single vertex of G is labeled O. A 2-cell embedding of G on 51 the arc labeled 2 is encountered next and is directed toward 0). Continu-
that has two regions (one of which is shaded) is shown in Figure 7.14(b). ing in this manner, we obtain 11'0 = (1 54623).
We now return to K7 • The graph K7 is a Cayley graph. Indeed, The cyclic permutation .1ro just calculated now determines all seven
K7 = G.6.(~), where ~ denotes the group of integers modulo 7 (under cyclic permutations that comprise the rotational embedding scheme
addition) and fl consists of the three generators 1,2,3 of Z7' We now that produces the embedding of K7 on $1 shown in Figure 7.2. Here we
assign a direction to each edge (loop) of the pseUdograph G of Figure use the fact that K7 is the Cayley graph G.6. (~), where fl = {I, 2, 3}. Hence
7.14(a) and label each arc so produced with an element of fl. The resulting the vertex set of K7 is ~ = {O, 1,2, ... ,6}. The rotational embedding
pseudodigraph H is referred to as a voltage graph (the formal definition of scheme contains 11'0 = (1 546 2 3). The remaining six cyclic permutations
which will be given shortly) and is shown in Figure 7.15(a). The particular 11'; (1 ~ i ~ 6) are obtained from 11'0 by adding i (addition modulo 7) to
every entry of 11'0' Thus, the rotational embedding scheme for K7 is the
7-tuple (11'0, 1I't, .•. ,71'6), where
G:
71'0 = (1 54 6 2 3),
11'1 = (265034),
71'2 = (306 1 4 5),
11'3 = (410256),
71'4 = (521 3 6 0),
(~ (b) 11'5=(632401),
Figure 7.14 A 2-cell embedding of a pseudograph on St. 1l'6 = (043 5 1 2).
216 Graph embeddings Voltage graphs 217
This is precisely the rotational embedding scheme for the embedding of lifts (or corresponds) to Irjls regions in the covering graph embedding, where
K7 shown in Figure 7.2. If we had not recognized this as the embedding of each such region is bounded by an orbit of length ks. Moreover, two distinct
Figure 7.2, then, of course, we could use the methods described in section regions of H correspond to distinct sets of regions in the covering graph.
7.2 to compute the number of regions of this embedding and show that it
does, in fact, describe an embedding of K7 on 51 with 14 triangular Theorem 7.28 gives us a method of computing the number of regions in
regions. Doing so, however, would not be using the power of the voltage a covering graph embedding directly from the voltage graph. To illustrate
graph H. this, consider once again the voltage graph H and its embedding in Figure
More generally, let r be a finite group with identity element e, and let 6 7.15(b). Since the order of the boundary element of each of the two regions
be a generating set for r, where e rt 6, such that if h E 6., then h- l rt 6., is s = I, it follows that each region lifts to Irjls = 711 = 7 regions in the
unless, of course, h2 = e. Then a pseudodigraph H of order 1 is called a embedding of G!J,(r) = K7 . Hence the voltage graph H lifts to an embed-
voltage graph of index 1 corresponding to rand 6 if H has size 161 and the ding of K7 that has 14 regions. Also, since each region of H is bounded by
arcs of H are labeled with the distinct elements of 6. The general theory of an orbit of length 3, all 14 regions of the embedding of K7 are triangular
voltage graphs allows for voltage graphs of order greater than 1; however, regions. Thus, we have shown that the embedding of H in Figure 7.15(b)
it is advantageous here to consider only those of order 1. lifts to the embedding of K7 on 51 shown in Figure 7.2.
Let H be a voltage graph of index 1 corresponding to a finite group r In the previous example, the boundary element of every region of the
and a generating set 6 for r, and suppose that H is 2-cell embedded on voltage graph embedding is the identity element of the group. When this
some surface. Then this embedding of H determines a 2-cellembedding of occurs, we say that the embedding satisfies the Kirchhoff Voltage Law (or,
the Cayley graph G!J, (r) on some surface. We say that H lifts to the covering more simply, the KVL). It is also possible to have a voltage graph embed-
graph G!J,(r). Also, the voltage graph is referred to as being below the ding in which some but not all regions have the identity for its boundary
covering graph and the covering graph as being above the voltage element. In this case, we say that the particular region satisfies the KVLi
graph. Let R be a region of the embedding of H, and let hI, h2 , ... ,hk that is, if R is a region whose boundary element is the identity, then R
denote the sequence of arcs (elements of 6.) encountered as the boundary satisfies the KVL. For example, an embedding of the voltage graph H of
of R is traversed in the clockwise direction. Then we say that R is bounded Figure 7.15(a) corresponding to Z7 al,ld 6 = {1,2,3} on 50 is shown in
by an orbit of length k, and the boundary element of R is defined as the Figure 7.16. Here, only the exterior region R satisfies the KVL.
product (or sum, if the group operation is addition) Next, we state an important special case of Theorem 7.28.
k

II hmj
i=1
I , 2

where mi = 1 if hi is oriented in the same direction as the boundary of R is

traversed and mj == -1 if hi is oriented in the opposite direction. For exam-
ple, the boundary element of each region of the voltage graph embedding
shown in Figure 7.15(b) is 0 since 1 + 2 + 4 = 0 and 3 + 6 + 5 = 0 in ~.
Since a voltage graph is a pseudodigraph, a region of an embedding of a
voltage graph may be founded by an orbit of length 1 or 2. Figure 7.16 A region oian embedding of a voltage graph that satisfies the KVL.
The next result, which is a special case of a theorem of Gross and Alpert
[GAl], describes the relationship between regions of a voltage graph Corollary 7.29
embedding and regions of its covering graph embedding.
For a finite group r with generating set 6, let H be a voltage graph of index 1
corresponding to rand 6. that is2-cell embedded on a surface. If R is a region of
Theorem 7.28 this embedding that is bounded by an orbit oflength k( ~ 1) and has the identity of
r as its boundary element, then R lifts to Irl regions, each of which is bounded by
For a finite group r with generating set 6., let H be a voltage graph of indexl' an orbit of length k.
corresponding to rand 6 that is 2-cell embedded on some surface. Furthermore,
let R be a region of the embedding of H that is bounded by an orbit of length Let r = £';5 and Li = {I,Z}. A voltage graph, denoted by H', correspond-
k (k ~ I), and let s denote the order of the boundary element of R in r. Then R ing to rand Li is shown in Figure 7.17(a) embedded on 51' and resulting in
218 Graph embeddings Voltage graphs 219
2 n----r--_.--1)

(2.1) (0.1) 0, 1)

H': CD.' (0.1)

(2.0) (0.0) (1.0)

~ 0,0)
(2.2) (0.2) (1.2)
2O:='=------:L.--....-o2

(a) (b)
Figure 7.17 A voltage graph and its covering graph. (a) (b)

Figure 7.19 A voltage graph and its covering graph.

one region R. Since 1 + 3 + 4 + 2 = 0 in 2 5 , it ~ollows th~t R satis~ies the
KVL and is bounded by an orbit of length 4. This embeddmg of H hfts to a
2-cell embedding of Ge,.(r) = Ks that has five regions, each of which is
EXERCISES 7.4
bounded by an orbit of length 4. Thus K s is embedded on 51 (Figure
7.l7(b».
7.30 Consider the embedding of the voltage graph shown below for the
Next, consider the voltage graphH" corresponding to 2 s and.6. = {I, 2}
given group C where .6. = {hJ,h z}. Determine the covering graph
shown in Figure 7.18. Here, HI! is embedded on So. The boundary ele~ent
embedding; that is, find the covering graph, the surface on WhICh
of each of the regions R 1 and R3 is 4 and the boundary element of R z IS 2.
it is embedded, and describe the regions.
All of these elements have order 5 in 2 s so that each of the regions R1 and
(a) r = Zs; h1 = t hz = 2,
Rzlifts to a region bounded by an orbit of length 5, and R 3 (being bounded
(b) r = 2 3 x 2 3 ; hI = (1,0), hz = (0,1).
by an orbit of length 2) lifts to a region b?unded b~ ~n orbit of length .10.
Thus the embedding above has three regIOns; so H' Mts to an embeddmg
of Ks on 5z.
As a fmal example, consider the voltage graph of Figure 7.19(a) corre-
:=
sponding to r = 2 3 x 2 3 and .6. = {(I, 0), (0, I)}. Then ~e,.(r) C3 x C3
and, by Corollary 7.29, the embedding of C3 x C3 has nme reglOns, each
of which is bounded by a 4-cyde. Figure 7.19(b) shows the covering graph
embedding. .
This section provides only a brief introduction to voltage graphs.
7.31 Find a voltage graph embedding that lifts to an embedding of
Indeed, we have restricted the discussion to voltage graphs of index 1.
C3 x C3 X C3 X C3 that contains 81 regions, each of which is bounded
For such voltage graphs, the covering graphs are always Cayley graphs.
by an 8-cycle. On what surface does this embedding take place?
However, the theory of voltage graphs is often used to describe embed-
(Hint: Recall the representation of the double torus shown in
dings of graphs that are not Cayley graphs. We refer the reader to Cross
Figure 7.5.)
and Tucker [CTl] and to White [W5].

1 2

HfI:~
R3
Figure 7.18 Another voltage graph embedded on So;
 Vertex colorings 221
232 3
CHAPTER 8
G:

Graph colorings
Figure 8.1 A 3-chromatic graph.

that X(G) ~ 3. These two inequalities imply that X(G) = 3, that is, G is
3-chromatic.
The graph-theoretic patameter that has received the most attention over
If G is a k-partite graph, then X(G) ~ k since the partite sets of G
the yea~s is the chromatic number. Its prominence in graph theory is
determine the color classes in a k-coloring of G. Conversely, every
undoubtedly due to its involvement with the Four Color Problem,
graph G with X(G) ::::;k is necessarily k-partite. Similarly, X(G) = kif and
:-vhich is di~cussed in this chapter. The main goal ofthis chapter, however,
only if Gis k-partite but G is not t-partite for t < k. Consequently, the 1-
IS to descnbe the many ways in which a graph can be colored and to
chromatic graphs are precisely the empty graphs and the 2-chromatic
present results on these topics.
graphs are the nonempty bipartite graphs. However, for no value of k
greater than 2 is such an applicable characterization known.
We need only be concerned with determining the chromatic numbers of
8.1 VERTEX COLORINGS
nonseparable graphs since the chromatic number of a disconnected graph
is the maximum of the chromatic numbers of its components and the
~ coloring of a graph G is an assignment of colors (which ate actually con-
chromatic number of a connected graph with cut-vertices is the maximum
SIdered as elements of some set) to the vertices of G, one color to each
of the chromatic numbers of its blocks.
:ertex: so that adjacent ve~ices are a~signed different colors. A coloring
Although the chromatic number is one of the most studied parameters
m whIch k colors are used IS a k-colormg. A graph G is k-colorable if there
in graph theory, no formula exists for the chromatic number of an arbi-
exists an s-coloring of G for some s ::::; k. Itisobvious that if G has order n,
trary graph. Thus, for the most part, one must be content with supplying
then G can be n-colored, so that G is n-colorable.
bounds for the chromatic nllmber of graphs. In order to present such'
The minimum integer k for which a graph G is k-colorable is called the
bounds, we now discuss graphs that are critical or minimal with respect
vertex chromatic number, or simply the chromatic number of G, and is
to chromatic number.
denoted by x(G). If G is a graph for which X(G) = k, then G is k-chromatic.
For an integer k ;:. 2, we say that a graph G is critically k-chromatic if
Certainly, if H <;:: G, then X(H) ~ X(G).
~(G) = k and X(G - v) = k -1 for all v E V(G); G is minimally k-chromatic
I~ a given coloring of a graph G, a set consisting of all those vertices
aSSIgned the same color is referred to as a color class. The chromatic If X(G) = k and X(G - e) = k -1 for all e E E(G). There are several results
number of G may be defined alternatively as the minimum number of dealing with critically k-chromatic graphs and minimally k-chromatic
'~ndependent su~sets into which V(G) can be partitioned. Each such
graphs, many of which are due to Dirac [D3]. We shall consider here
mdependent set IS then a color class inthe X(G)-coloring ofG so defined. only one of the more elementary, yet very useful, of these.
For some graphs, the chromatic number is quite easy to determine. For Every critically k-chromatic graph is nonseparable, while every mini-
example, mally k-chromatic graph without isolated vertices is nonseparable. Further-
more, every minimally k-chromatic graph (without isolated vertices) is
critically k-chromatic. The converse is not true in general, howeveri for
and, in general, example, t~e graph of Figure 8.2 is critically 4-chromatic but not minimally
4-~~omatIc. For k =.2 and k = 3, the converse is true. In fact, K2 is the only
x(K(nl' n2,' .. ,nk)) = k. cntIcally 2·chromatic graph as well as the only minimally 2-ehromatic
The graph G of Figure 8.1 is 3-colorablei a 3-coloring of G is indicated with graph wit~out isolated verticesi while the odd cycles are the only critically
the colors denoted by the integers 1, 2, 3. Therefore, G is k-colorable for ?-ehromahc graphs and the only minimally 3-ehromatic graphs having no
1r '> ~ "",.1 vrr.., < q C::'",.,p r. ,<:» <:nho-r»nh of G imd y(e.) = 3. it follows
Isolated vertices. For. k ;:. 4, neither the critically k-chrom'atic graphs nor
222 Graph colorings Vertex colorings 223
If for each til in Ull the vertex Ul is adjacent only with vertices assigned
colors different from that assigned to Ul, then the assignment of colors to
the vertices of G is not altered. On the other hand, if some vertex UI of U1 is
adjacent with a vertex of Gz that is assigned the same color as that of Ull
then in G1 we may permute the k - 1 colors so that in the new assignment.
of colors to the vertices of G, no vertex of U1 is adjacent to a vertex of Gz
having the same color. This is possible since the vertices of UI may be
assigned anyone of at least k -1 - kl colors and k -1 - kl > O.
If, in this new assignment of colors to the vertices of G, each vertex Uz of
Figure 8.2 A critically 4-ehromatic graph that is not minimally 4-chromatic. Uz is adjacent only with vertices assigned colors different from that
assigned to uz, then no (additional) permutation of colors of G1 occur~.
the minimally k-chromatic graphs have been characterized. Although it is However, if saine vertex Uz of Uz is adjacent with a vertex of Gz that IS
quite difficult, in general, to determine whether a given k-chromatic graph assigned the same color as that of Uz, then in G1 we may permute the k - 1
G is critical or minimal, G contains both critically k-chromatic subgraphs colors, leaving thecolor assigned to U1 fixed, so that no vertex of UI U Uz
and minimally k-chromatic subgraphs. A k-chromatic subgraph of G of is adjacent to a vertex of Gz having the same color. This can be done since
minimum order is critically k-chromatic, while a k-chromatic subgraph of the vertices of Uz can be assigned any of (k - 1) - (k z + 1) colors, and
G of minimum size is minimally k-chromatic. (k -1) - (k z + 1) ~ (k -1) - (kl + kz) > O. Continuing this process, we
The first theorem of this chapter concerns the structure of critically (and arrive at a (k - 1)-coloring of G, producing the desired contradiction. 0
minimally) k-chromatic graphs.
Since' every connected, minimally k-chromatic graph is critically k-
Theorem 8.1 chromatic, the preceding result has an immediate consequence.
Every critically k-chromatic graph, k ~. 2, is (k -I)-edge-connected.
Corollary 8.2
Proof If G is a connected, minimally k-chromatic graph, k ~ 2, then G is (k - 1 )-edge-
Let G be critically k-chromatic, k ~ 2. If k = 2 or k = 3, then G = Kz or G is connected.
an odd cycle, respectively; therefore, G is I-edge-connected or 2-edge-
Theorem 8.1 and Corollary 8.2 imply that 11:1 (G) ~ k -1 for every
connected.
critically k-chromatic graph G or connected, minimally k-chromatic
Assume that k ~ 4 and suppose, to the contrary, that G is not (k -1)-
edge-connected. Hence by Theorem 3.20, there exists a partition of V(G) graph G. The next corollary now follows directly from Theorem 3.18.
into subsets VI and Vz such that the set E' of edges VI and Vz contains
fewer than k - 1 elements. Since G is critically k-chromatic, the subgraphs Corollary 8.3
GI = (VI) and Gz = (Vz ) are (k -I)-colorable. Let each qf GI and Gz be If G is critically k-chromatic or connected and minimally k-chromatic, then
colored with at most k - 1 colors, using the same set of k - 1 colors. If each
6(G)~k-l.
edge in E' is incident with vertices of different colors, then G is (k - 1)-
colorable. This contradicts the fact that x( G) = k. Hence we may assume We are now prepared to present bounds for the chromatic number of a
that there are edges of E' incident with vertices assigned the same color.
graph. We give here several upper bounds, beginning with the best
We show that the colors assigned to the elements of VI may be permuted
known and. most applicable. The theorem is due to Brooks [BI5] but the
so that each edge in E' joins vertices assigned different colors. Again this proof here is due to Lovasz [L5].
will imply that x( G) ~ k - I, produce a contradiction, and complete the
proof.
In the coloring of GlI let UI , Uz, ... , Ut be those color classes of GI such Theorem 8.4
that for each i, 1 ~ i ~ t ~ k - 2, there is at least one edge joining Uj aru::t If G is a connected graph that is neither an odd cycle nor a complete graph; then
G2• For i = 1,2, ... , t, assume that there are kj edges joining Uj and G2•
Hence, for each i, 1 ~ i ~ t, it follows that ki > a and 2:::=1 kj ~ k - 2. X(G) ~·~(G).
224 Graph colorings Vertex colorings 225
Proof If K(H - u) = 1, then we consider two end-blocks Bl and Bz containing
cut-vertices WI and W2, respectively, of H - u. Since H is 2-connected,
Let G be a connected graph that is neither an odd cycle nor a complete
there exist vertices UI in B1 - WI and U2 in B2 - W2 that are adjacent to
graph, and suppose that X(G) = k, where, necessarily, k;:: 2. Let H be a
u. Let x = UI a~d y = Uz, and proceed as in Case 1.
critically k-chromatic subgraph of G. Then H is nonseparable and
This completes the proof. D
t:J.(H) ~ t:J.(G).
Suppose that H = Kk or that H is an odd cycle. Then G 1= H. Since G is
connected, t:J.(G) > t:J.(H). If H = Kk' then t:J.(H) = k -1 and t:J.(G) ;:: k; so The bound for the chromatic number given in Theorem 8.4 is not
particularly good for certain classes of graphs. For example, the bound
X(G) = k ~ t:J.(G). provided for the star graph Kl,n -1 of order ndiffers from its chromatic
If H is an odd cycle, then number by n - 3. We shall see in section 8.3 that 4 serves as an upper
bound for the chromatic number of all planar graphs; however, Theorem
t:J.(G) ;:: 3 = k = X(G). 8.4 gives no bound for the entire class. Thus, there are several important
Hence, we may assume that H is critically k-chromatic and is neither an classes of graphs for which the bound t:J.(G) for X( G) is poor indeed. From
odd cycle nor a complete graph; this implies that k ;:: 4. the next result, due to Szekeres and Wilf [SWn a number of other bounds
Let H have order n. Since x(H) = k ;:: 4 and H is not complete, it follows for X(G) follow as corollaries.
that n ;:: 5. We now consider two cases, depending on the connectivity
ofH.
Theorem 8.5
Case 1. Suppose that H is 3-connected. Let x and y be vertices of H such Let f be a real-valued function defined on the class of all graphs that satisfies the
that dH(x,y) = 2, and suppose that x, w,y is a path in H. The graph
following two properties:
H - x - Y is connected. Let Xl = W,Xz, •.. ,Xn-z be the vertices of
H - X - y, listed so that each vertex Xi (2 ~ i ~ n - 2) is adjacent to at (i) If H is an induced subgraph of G, then f(H) ~ f(G).
least one vertex preceding it. By letting Xn-I = X and X n = y, we have (ii) f(G) ;:: 6(G) for every graph G.
the sequence
Then X(G) ~1 + f(G).
Assign the color 1 to the vertices Xn-l and Xn- We successively color
xn _Z,xn _3,,,,,x2 with one of the colors 1,2, ... ,t:J.(H) that was not Proof
used in coloring adjacent vertices following it in the sequence. Such a Let G be a k-chromatic graph, where k;:: 2, and let H be an induced k-
color is available since each Xi (2 ~ i ~ n - 2) is adjacent to at most critical subgraph of G. By Corollary 8.3, 6(H) ;:: k -1. Therefore,
t:J.(H) - 1 vertices following it in the sequence. Since Xl = W is adjacent
to two vertices colored 1 (namely, Xn -1 and xn ), a color is available for f(G) ;:: f(H) ;;;. 6(H) ;::k -1,
Xl' Therefore,
so that
x(G) = X(H) ~ t:J.(H) ~ t:J.(G).
x(G) = k ~ 1 + f(G). 0
Case 2. Suppose that K(H) = 2. We begin with an observation in this case;
namely, H does not contain only vertices of degrees 2 and n-1. Since As corollaries of Theorem 8.5 we have the results of Brooks [B15L
X(H) ;:: 4, H cannot contain only vertices of degree 2. Since H is not c0!p... ,. Welsh and Powell [WPIt Gallai [G2L and Szekeres and Wilt [SWn
plete, H cannot contain only vertices of degree n - 1. If H contains vertices"c' which are stated in chronological order.
of both degrees (and no others), then H must contain two vertices of
degree n -1 and n - 2 vertices of degree 2; that is, H = KI,I,n-Z. However,
then, X(H) = 3, which is impossible. Corollary 8.6
Let u E V(H) such that 2 < degH u < n -1. If H - u is 2-connected, For every graph G,
then let v be a vertex with dH(u, v) = 2. We may let X = u and y = v,
and proceed as in Case 1. x(G) ~ 1 + t:J.(G).
226 Graph colorings Vertex colorings 227
Corollary 8.7
Let G be a graph with V(G) = {VI, V2,"" v n }. Then 51 = 0,2, 3)
5 2 =(1,2, 3)
X(G) ~
1
max {min{i, deg
~ l~n
Vi + I}}. 5 3= (2, 3,4,5, 6}
5 4 = (4, 5}
Corollary 8.8
For every graph G, Figure 8.3 An intersection graph.

x(G) ~ 1 + t(G),
degree at most 5 (by Corollary 6.5) and since every subgraph of a planar
where t( G) denotes the length of a longest path in G.
graph is planar, a bound of 6 is provided for the chromatic number of
planar graphs by Corollary 8.9. In each of these cases a marked improve-
Corollary 8.9 ment is shown over the result offered by Theorem 8.4. If G is a regular
graph of degree r, then both Theorem 8.4 and Corollary 8:9 give r + 1
For every graph G,
as an upper bound for X(G); however, this bound is poor fqr many r-
x(G) ~ 1 + maxb(H), regular graphs, such as Kr,r'
where the maximum is taken over all induced subgraphs H of G. We now direct our attention briefly to lower bounds for the chromatic
number. The clique. number w(G) of a graph G is the maximum order
The proof of each of Corollaries 8.6, 8.7, 8.8 and 8.9 consists of deter- among the complete subgraphs of G. Clearly, w(G) = (3(G) for every
mining an appropriate function f and then applying Theorem 8.5. For graph G. If Kk ~ G for some k, then X(G) ~ X(Kk ) = k. It follows that
example, in Corollary 8.9 we define the function f by f(G) = max b(H), X(G) ~ w(G). Although this lower bound for X(G) is not particularly
where the maximum is taken over all induced subgraphs H of G. Then good in general, X( G) actually equals w( G) for some special but important
clearly f satisfies the properties required to apply Theorem 8.5. For classes of graphs. For example, if G is bipartite, then either G is empty and
proofs of the other corollaries, see Exercise 8.8. x(G) = 1 = w(G), or else X(G) = 2 = w(G). In order to give another
It is not difficult to show that among the corollaries stated above, example where equality holds, we introduce the notions of intersection
Corollary 8.9 is the strongest application of Theorem 8.5. graphs and interval graphs.
Let F be a finite family of not necessarily distinct nonempty sets. The
intersectiongraph of F is obtained by representing each set in F by a vertex
Theorem 8.10 and then adding an edge between two vertices whose corresponding sets
If a r~~l-valued function f defined on the class of all graphs satisfies properties (i) have a nonempty intersection. A graph G is called an intersection graph if
and (ll) of Theorem 8.5, then for each graph G, it is the intersection graph of some family F. A family F = {51' 52, 53, 54}
and its intersection graph is shown in Figure 8.3. Here, vertex Vi
f(G) ~ max8(H), corresponds to set 5i •
where the maximum is taken over all induced subgraphs H of G. When F is allowed to be an arbitrary family of sets, the class of graphs
obtained as intersection graphs is Simply all graphs (see Marczewski
[M2]). By restricting the sets in F, many interesting classes of graphs
Proof
are obtained. For example, the intersection graph of a family of closed
Let G be a graph, and let k = max 8(H), where the maximum is taken over intervals of real numbers is called an interval graph. The graph G of
all induced subgraphs Hof G. If f(G) < k, then G contains an induced Figure 8.3 is seen to be an interval graph by considering the intervals
su~graph HI with f(G) <: k = 8(HI ). By (i), then, f(H I ) ~ f(G) < 6(H1 ). h = [1,3], 12 = [1,3], 13 = [2,6] and 14 =[4,5]. However, not all graphs
ThiS, however, contradicts (ii). 0 are interval graphs. For example, C4 is not an interval graph.
IT G is an interval graph, then the vertices of G correspond to closed
Corollary 8.9 gives an upper bound of 2 for the chromatic number of intervals; say Vi corresponds to Ii = [ti, r;]. We show thatG has an w(G)-
Kl ,n -I' which, of course, is exact. Since every planar graph has minimum coloring. Assume that the vertices of'C have been labeled so that t; ::::; t j
 Graph colorings Vertex colorings 229
228
Theorem 8.11
1 2
A graph G is a chordal graph if and only if either G is complete or G can be

IZ obtained from two chordal graphs G1 and G2 (having orde~s less than that of
G) by identifying two complete sUbgraphs of the same order In G1 and G2·

3 1 Proof
Figure 8.4 A greedy 3-coloring. If G can be obtained as described, then clearly G is chordal. Conversely,
since every complete graph is chordal, let G be a noncomplete chordal
if i ~ j. Color VI' Vl,' .. ,Vn in order, assigning to Vi the smallest color graph and let S be any minimal vertex-cut of G. Let A be the vertex set
(positive integer) j that has not been assigned to a neighbor of Vi' (Such of one component of G - S and let B = V (G) - 5 - A. Define the (chordal)
a coloring is called a greedy coloring.) For example, using this procedure subgraphs GI and G2 of Gby G1 = (A U 5) and G1 = (B U 5). Then G can be
on the graph G of Figure 8.3, we obtain the 3-coloring of G given in obtained from G1 and G1 by identifying the vertices of 5. We show that (5)
Figure 8.4. Note that w(G) = 3. is complete. This is certainly true if 151 = l,so we may assume that 151 ;<: 2.
In general, suppose that we have given a greedy coloring to an interval Since 5 is minimal, each x E 5 is adjacent to some vertex of each com-
graph G and that vertex Vi has been assigned color k. This implies that Vi is ponent of G - 5. Therefore, for each pair x,y E 5, there exist paths x,
adjacent to vi" Vi2 , ' •• , Vik_l (i1 < i1 < ... < ik - 1 < i) colored 1,2, ... , k -I, a1, a2,"" a" Y and x,b1,b1).., , bt , Y where each ai E A and bi E B, suc.h
I
respectively. We show that these vertices, together with Vi, induce that these paths are chosen to be of minimum length. Thus,
a complete graph. The left endpOints of the corresponding intervals satisfy C: x,aba2,'" ,anY, bi, bt - 1 , ... , bbX is a cycle of G of length at least 4,
t i, ~ t i2 ~ ... ~ t ik _l ~ ti' implying that C has a chord. However, aibj (j. E(G) since 5 is a vertex-
cut and aiaj (j. E(G) and bibj (j. E(G) by the minimality of rand t. Thus
If til = t iz = ... = t ik _, = t i, then certainly ({Vi" Viz, ... ,Vik_, ' Vi}) is com- xy E E(G). 0
plete. Otherwise there is an integer t with 1 ,;;; t ,;;; k - 1 for which t ij < t i
for 1,;;; j ,;;; t and ti. = t i for each j with t + 1 ,;;; j ~ k - 1. Clearly,
Vit+1 ,Vi
. t+2
, .... ,Vik-1 ,Vi J induce a complete graph. Now, since t iJ < t i for Corollary 8.12
1 ~ j ,;;; t and Vij and Vi are adjacent, it follows that rij ;<: Ii' Thus If G is a chordal graph, then X( G) = w( G).
Vi" Viz" .. , Vik_" Vi induce a complete graph. We conclude that the
greedy coloring so produced is an w( G)-coloring.
As we have seen, if G is a bipartite graph or an interval graph, then Proof
X(G) = w(G). Furthermore, since every induced sUbgraph of a bipartite We proceed by induction on the order n of G. If n = I, then G = K 1 and
(interval) graph is also a bipartite (interval) graph, we see that X( G) = w( G) = 1. Assume that the clique. and chromatic~umbers are
X(H) = w(H) for every induced subgraph H of a bipartite or interval equal for chordal graphs of order less than n and let G be a chordal
graph. A graph G is called perfeCt if X(H) = w(H) for each induced sub- graph of order n ;<: 2.
graph H of G. Thus bipartite graphs and interval graphs are examples of If G is complete, then X(G) = w(G). If G is not complete, then G can be
perfect graphs. obtained from two chordal graphs G1 and G1 of order less than that of G
Interval graphs form a subset of a larger class of perfect graphs called by identifying two complete subgraphs of the same order ~ G1 and G2•
chordal graphs. A graph G is called chordal if every cycle of G of length Since there are no edges between V(G 1 ) - Sand V(G 2 ) - 5, It follows that
greater than 3 has a chord, that is, an edge joining two nonconsecutive
vertices of the cycle. In the literature, chordal graphs have also been w(G) = max{w(G1),W(G1 )}.
called triangulated, rigid circuit and perfect elimination graphs. Clearly, X(G) ;<: max{x(G1 ), X(G 2 )}. However, since a X(Gi)-coloring of
Since every induced sUbgraph of a chordal graph is also chordal, to Gi assigns distinct colors to the vertices of 5 (i == 1,2), it follows that ~e
show that chordal graphs are perfect we need only show that can make the colorings agree on 5 to obtain a max{x(G 1 ), X(G2 )}-colormg
X(G) = w(G) for every chordal graph G. This follows from the following of G. Thus,
characterization of chordal graphs due to Hajnal and Suranyi [HSl] and
Dirac [06].
230 Graph colorings Vertex colorings 231
By the inductive hypothesis, X(G I ) = w(G I ) and X(G z) = w(G z ). Thus,
X(G) = w(G). 0

Corollary 8.13
Every chordal graph is perfect.

Perfect graphs were introduced by Berge [B61, who conjectured that a

graph G is perfect if and only if G is perfect. This conjecture (sometimes
referred to as the Perfect Graph Conjecture) was verified by Lovasz [U]. Its
rather lengthy proof is omitted.

Figure 8.5 The Grotzsch graph: a 4-chromatic triangle-free graph.

Theorem 8.14
A graph G is perfect if and only ifG is perfect. Proof
Since X(C 2k + I ) = 3 i= w(C2k + I ), k)o 2, it follows that if an induced sub- The proof is by induction on k. If k is 1, 2 or 3, then the graphs KI , Kz and
graph of a graph G is an odd cycle of length at least 5, then G is not perfect. Cs, respectively, have the required properties. Assume that H is a triangle-
Similarly, if G contains an induced odd cycle of length at least 5, then G free graph with X(H) = k, where k)o 3. We show that there exists a
and, by Theorem 8.14, G are not perfect. Berge conjectured ([B8, p. 361]) triangle-free graph with chromatic number k+ 1. Let V(H) =
that every graph that is not perfect contains either an induced odd cycle of {VI> V2,' .. , vn }. We construct a graph G from H by adding n + 1 new
length at least 5 or its complement contains such a cycle. vertices u, UI, U2,'''' Un' The vertex u is joined to each vertex Ui and, in
addition, Ui is joined to each neighbor of Vj.
To s.ee that G is triangle-free, first observe that U belongs to no triangle.
Strong Perfect Graph Conjecture Since no two vertices Uj are adjacent, any triangle would consist of a
A graph G is perfect if and only if no induced subgraph of G or G is an odd cycle vertex Ui and vertices Vj and Vt, i i= j, t, but by the construction, this
of length at least 5. would imply that ({ Vj, Vj' Vt }) is a triangle in H, which is j,mpossible.
Let a k-coloring of H be given. Now assign to Uj the same color assigned
This conjecture remains open, although it has been verified for several to Vi and assign a (k + 1)st color to u. This produces a (k + 1)-coloring of
classes of graphs including planar graphs (Tucker [T8]) and claw-free G. Hence X(G) ~ k+ 1. Suppose that X(G) ~ k, and let there be given a
graphs (Parthasarathy and Ravindra [PRl]). k-coloring of G, with colors 1,2, ... ,k, say. Necessarily the vertex U is
As we have noted, X(G) )0 w(G) for every graph G. Hence if G contains colored differently from each Uj. Suppose that U is assigned color k.
triangles, then X( G) )0 3. However, there exist graphs G that are triangle- Since X(H) = k, the color k is assigned to some vertices of H. Recolor
free such thatX(G) )0 3. For example, the odd cycles C2k + I , with k)o 2, each Vj that is colored k with the color assigned to Uj. This produces a
have chromatic number 3 and are, of course, triangle-free. The graph of (k - 1)-coloring .of H and a contradiction. Thus, X( G) = k + 1, and the
Figure 8.5, called the Grotzsch graph, is 4-chromatic and triangle-free and proof is complete. 0
is, in fi;lct, the smallest such graph (in terms of order).
It may be surprising that there exist triangle-free graphs with arbitrarily This result has been extended significantly by Erdos [E3] and Lovasz
large chromatic number. This fact has been established by a number [L3]. We postpone the proof of Theorem 8.16 until Chapter 13 (Theorem
of mathematicians, including Descartes [D21 Kelly and Kelly [KKl] and 13.5). The girth of a graph is the length of its shortest cycle.
Zykov [Z3]. The following construction is due to Mycielski [MIl], however.
Theorem 8.16
Theorem 8.15
For every two integers k )0 2 and t ;:;: 3 there exists a k-chromatic graph whose
For every positive integer k, there exists a k-chromatic triangle-free graph. girth exceeds t.
 232 Graph colorings Vertex colorings 233
According to Theorem 8.15, a k-chromatic graph may contain no
triangles and, therefore, no large complete subgraphs. In particular, a
k-chromatic graph need not contain Kk • There are two well-known vn ~ VX(G). x(G)~ x(G); x(G).
conjectures related to this observation.
For k ~ 3, it is trivial to see that every k-chromatic graph contains a This verifies the lower bound of (i).
subdivision of Kk • Dirac [03] showed that this is also true for k = 4 and To verify the upper bound in nu, we make use of Corollary 8.9. Suppose
Hajos [H2] conjectured that for each positive integer k, every k-chromatic that k = max8(H), where the maximum is taken over all induced sub-
graph contains a subdivision of Kk . This conjecture, however, was shown graphs H of G. Hence every induced subgraph of G has minimum
to be false for k ~ 7. A weaker conjecture was proposed by Hadwiger degree at most k and, by Corollary 8.9, it fol~ws that X(G)';:;; 1 +k.
[Hl]. . Next we show that every induced subgraph of G has minimum degree
at most n - k - 1. Assume, to the contrary, that there is an induced sub-
graph H of G so that 8(Ft) ~ n - k. Thus the vertices of H have degree at
Hadwiger's Conjecture
most k - 1 in G.
If G is a k-chromatic graph, where k is a positive integer, then Kk is a subcon- Let.F be an induced subgraph of G with o(F) = k: Thus no vertex of F
traction of G. belongs to H. Since the order of F is at least k + I, the order of H is at most
n -k - I, contradicting the fact· that· 8(H) ~ n - k. We may therefore
This conjecture was verified for k = 4 by Dirac [031. The proofs for k = 5 conclude that every induced subgraph of G has minimum degree at
and k = 6 depend on results of Wagner [W2] and of Robertson, Seymour most n - k -1 and, by Corollary 8.9, that X(G) ~ 1 + (n - k -1),= n - k.
and Thomas [RSTl] and on the proof of the famous Four Color Theorem Thus,
which we will encounter in section 8.3. For k ~, 7, this conjecture remains x(G) + X(G) ,;:;; (1 +k) + (n - k) = n + 1,
open.
Our next result, due to Nordhaus and Gaddum [NGl], is the best completing the proof of the upper bound in (i). The upper bound in (ii)
known result on chromatic numbers and complementary graphs. The now follows by consideration of geometric and arithmetic me<\ns. 0
proof is based on one by H. V. Kronk.
We close this section with some examples of variations of graph color-
ing. We observed that the chromatic number of a graph G can be defined
Theorem 8.17 . as the minimum number of independent sets into which V(G) can be
If G is a graph of order n, then partitioned. More generally, let P denote a family of graphs. The P
chromatic number Xp(G) is the minimum number of sets into which V(G)
(i) 2n / ~ X(G) + X(G) ~ n + I,
12
can be partitioned so that each set induces a graph that belongs to P.
(ii) n ~ X(G)· X(G) <::;; ((n + 1)/2)2. This parameter is well-defined whenever K1 E P. If P is the family of
empty, graphs, then Xp(G) = X(G). If P is the family of forests, then
Xp(G) = a(G), the vertex-arboricity of G. IfP is the family of empty
Proof graphs and complete graphs, then Xp(G) is the cochromatic number of G,
Let a X(G)-coloring for G and a X(G)-coloring for Gbe given. Using these first introduced in [LS1]. Alternatively, the cochromatic number of G is
colorings, we obtain a coloring of Kn . Assign a vertex v of Kn the color the minimum number of subsets in a partition of'V(G) so that each
(C1' c~), ~h~re c1 is the color assigned to v inG and c2 is the color assigned subset is independent in G or in G. Certainly for each family P described
to v In G. Smce every two vertices of K n are adjacent either in G or in G, above; Xp(G) ,;:;; x(G) for every graph G inP.
they are assigned different colors in that subgraph of K". Thus, this is a We now desGribe another coloring number. Suppose thatfor each vertex
coloring of KIl using' at most X( G) . x(G) colors, so v of a graph G there is associated a list L(v) of allowable colors for v. The
list chromatic number xtC G) is the smallest positive integer k such that for
n = X(K n ) ~ X(G)· X(G). each assignment of a list L(v) of cardinality at least k to every vertex v of
G, it is possible to color G so that every vertex is assigned a color from its
This establishes the lower bound in (ii). Since the arithmetic mean of two list. Thus X(G) ,;:;; Xt(G) for every graph G. That this inequality may be
positive numbers is always at least as large as their geometric mean, strict is illustrated by the bipartite graph G of Figure 8.6 for which
234 Graph colorings Vertex colorings 235
8.9 Let G1 , G2 , l Gk be pairwise disjoint graphs, and define G =
{I, 3}
G1 + G2 + + Gk • Prove that
k k

{I,2} {2,3} X(G) = LX(Gil and w(G) = LW(G;).

;=1 ;=1

8.10 (a) Show that every graph is an intersection graph.

G:
(b) Let G be a nonempty graph. Show that a set:F can be associated
with G so that the intersection graph of :F is the line graph of G.
{t,2} {I,3}
8.11 Show that every induced subgraph of an interval graph is an interval
graph.
{2,3} 8.12 Show that every interval graph is a chordal graph.
Figure 8.6 A 2-chromatic graph with list-chromatic number 3. 8.13 (a) Show that if G is a chordal graph, then for every proper complete
subgraphH of G there is a vertex v E V(G) - V(H) for which the
xAG) = 3. A 2-coloring of G cannot be given from the indicated lists. neighpors of v in G induce a complete sUbgraph in G.
A graph G is k-choosable if xA G) ~ k. (b) Show, without using Theorem 8.14, that the complement of a
The book by Jensen and Toft [JT1] on graph coloring problems is an chordal graph is perfect.
excellent source of additional information. More information on perfect 8.14 For each integer n ~ 7, give an example of a graph Gll of order n such
graphs can be found in the book by Golumbic [G6]. that no induced subgraph of Gn is an odd cycle of length at least 5 but
Gll is not perfect.
8.15 Determine G if, in the proof of Theorem 8.15,
EXERCISES 8.1 (a) H = K 2 ; (b) H = Cs.

8.1 Prove, for every graph G of order n, that n//3(G) ~X(C) ~

8.16 Prove that for every two integers k ~ 3 and t ~ 3, with k ~ t, there
n+ 1- j3(G). exists a graph G such that X(G) = k and w(G) = t.
8.2 Determine and prove a result analogous to Exercise 8.1 for vertex~
8.17 Show that the conjectures of Hajos and Hadwiger are true for k ~ 3.
arboricity. 8.18 Show that all the bounds given in Theorem 8.17 are sharp.
8.3 Prove a result analogous to Theorem 8.4 for disconnected graphs. 8.19 Determine and prove a theorem analogous to Theorem 8.17 for
8.4 What bound is given for X(G) by Corollary 8.9 in the case that G is vertex-arboricity.
(a) a tree? (b) an outerplanar graph (Exercise 6.11)? 8.20 Define a graph G to be k-degenerate, k ~ 0, if for every induced sub-
8.5 Let G be a k-chromatic graph, where k ~ 2, and let r be apositive graph H of G, 8(H) ~ k. Then the O-degenerate graphs are the empty
integer such that r ~ ~(G). Prove that there exists an r-regular k- graphs, and by Exercise 3.2, the I-degenerate graphs are precisely
chromatic graph H such that G is an induced subgraph of H. the forests. By Corollary 6.5, every planar graph is 5-degenerate. A
k-degenerate graph is maximal k-degenerate if, for every two non-
8.6 Determine (and prove) a necessary and sufficient condition for a adjacent vertices u and v of G, the graph G + uv is not k-degenerate.
graph to have a 2-colorable line graph. For k ~ 0, .let P k denote the family of k-degenerate graphs. Then
8.7 Let G be a corinectep., cubic graph of order n > 4 having girth 3. XPk(G) is the minimum number of subsets into which V(G) can be
Determine X(G). partitioned so that each subset induces a k-degenerate subgraph of
G. A graph is said to be t-critical with respect to XPk' t ~ 2, if
8.8 (a) Prove Corollary 8.6 without using Theorem 8.4. XPk(G) = t and X'Pk(G - v) = t -1 for every v E V(G).
(b) Prove Corollary 8.7. (a) Prove that if Gis a maximal k-degenerate graph of order n, where
(c) Prove Corollary 8.8. n ;;:: k + 1, then 6(G)':::t: k.
236 Graph colorings Edge colorings 237
(b) Determine XPk (K Il )· Set of vertices of degree d in G is empty or independent. We show that Gis
(c) Prove that if G is a graph that is t-critical with respect to XPk' then d-edge colorable.
beG) ~ (k + 1)(t -1). Let u be a vertex of G such that degu = ~(G). Furthermore, let Yo be a
vertex adjacent to u and let eo = uYo. Since d ~ ~(G) ~ ~(G - eo) and the
8.21 (a) Give an example of agraph H for which the cochromatic number
set of vertices of G - eo of degree d is empty or independent, it follows by
is 3. the inductive hypothesis that G - eo is dc.edge colorable.
(b) Give an example of a graph H that is not the union of (disjoint) Let there be given a d-edge coloring of G - eo; that is, every edge of G
complete graphs for which the cochromatic number equals X(H). except eo is assigned one of d colors so that adjacent edges are colored
(c) Give an example of a graph H that is not a union of complete differently. For each edge e = uY of G incident with u, we define its dual
graphs for which the cochromatic number is less than X(H). color as anyone of the d colors that is not used to color edges incident with
8.22 Show that K3,3 has list-chromatic number 3. y. Since y is adjacent to u, it follows that deg y < d regardless of whether
d = ~(G) ord > ~(G), and so there is at least one color available for a dual
color. Let eo have dual color 01' (The color 01 is not the color of any edge of
G incident with Yo.) We may assume that some edge el = UYI incident
8.2 EDGE COLORINGS withu has been assigned the coldr hI; for otherwise the edge eo could
be colored OJ, thereby producing a d-e<;l.ge coloring of Gandcompleting
We now switch our attention from coloring the vertices of a graph to the proof.
coloring its edges. An assignment of colors to the edges of a nonempty Let 02 be the dual color of el' If 0z of 01 and there is an edge e2 = uyz
graph G so that adjacent edges are colored differently is an edge coloring of incident with u that has been assigned the color 0z, then let 03 be the dual
G (a k-edge coloring if k colors are used). The graph G is k-edge colorable if color of ez. If 03 of 01, Oz and there is an edge e3 = UY3 incident with u that
there exists an t -edge coloring of G for some t :,,-; k. The minimum k for has been assigned the color 03' then let 04 be the dual color of e3' We
which a graph G is k-edge colorable is its edge chromatic number (or continue in this inanner until we have constructed a sequence eo = uYo,
chromatic index) and is denoted by Xl (G). eI = uYv ez = uYz,··· ,ek-l = UYk-I of distinct edges with dual colors
The determination of Xl (G) can be transformed into a problem dealing 01, 02, ... ,Ok such that OJ of OJ for i < j < k and either (i) Ok = OJ for
with chromatic numbers; namely, from the definitions it is immediate that some j < k or (iil there is no edge incident with u that has been assigned
the color Ok' If (ii) holds then we may assign each of the edges
Xl (G) = X(L(G)),
eo, eI, ... ,ek-l its dual color and obtain a d-edge coloring of G. Hence
where L(G) is the line graph of G. This observation appears to be of little we may assume that Ok = OJ for some j < k.
value in computing edge chromatic numbers, however, since chromatic °
Let be a color that has not been assigned to any edge incident with u.
numbers are extremely difficult to evaluate in general. Such a color exists since degG-eo u < ~(G) :,,-; d. If no edge incident with
It is obvious that ~(G) is lower bound for Xl (G). In what must be Yk-1 is colored 0, then we redefine the dual color of ek-I to be 0. Then (ii)
considered the fundamental result on edge colorings, Vizing [V3] holds and we can obtain a d-edge coloring of G as before. Hence we may
proved that Xl (G) equals ~(G) or 1 + ~(G). We prove a more general assume that some edge incident with Yk-l is colored 0.
result due to Berge and Fournier [BF1]. Define G' to be that spanning sUbgraph of G whose edge set consists of
°
those edges of G - eo colored or Ok' Since some edge incident with Yk-I
Theorem 8.18
°
is colored but no edge incident with Ykc..l is colored Ok, the component of
G' containing Yk-l is aYk-l - zpathP for some vertexi.If u lies on P, then
Let G be a nonempty graph and let d be a positive integer such that d ~ ~(G). z = u since no edge incident with u is colored 0. Furthermore, since
If the set of vertices of G of degree d is empty or independent, then Xl (G) :,,-; d. Ok = OJ, it follows that if z = u, then the final edge of P is ej = uYj' Also,
if u isonP, thenYj_l is not onP since no edge incident withYj_1 is colored
Proof
OJ = Ok' Therefore, at most one of u and Yj-1 is on P, and so at most one of
these vertices is z.
We proceed by induction on the size m 6f G. When m = 1, the result We now obtain a new d-edge coloring of G- eo by interchanging the
obvious. Assume, then, that the result holds for graphs of size m - 1, colors a and Ok for the edges of P. We consider three cases according to
where m ~ 2. Let G be a graph of size m for which d ~ ~(G) and the the possibilities for the vertex z.
238 Graph colorings Edge colorings 239
Case 1. Suppose that z = u. Then Yj -I is not on P and no edge incident Although it is probably not obvious, there are considerably more class
with u or y. -1 has been assigned the color Ctk = Ctj' Thus we may assign one graphs than class two graphs, relatively speaking. Indeed, Erdos and
each of the ~dges eo, el, ... ,ej-l its dual color and obtain a d-edge coloring Wilson [EWl] have proved that the probability that a graph of order n is
ofG. of class one approaches 1 as n approaches infinity. However, the problem
Case 2. Suppose that z = Yj_I' Then u is not on P and no edge i~cident of determining which graphs belong to which class is unsolved.
with u or Yj -1 has been assigned the color Ct. Thus we may aSSIgn the Corollary 8.20 gives a sufficient condition for a graph to be of class one.
edge ej_l the color Ct and each of the edges eo, el, ... ,ej-2 its dual color The following result, due to Beineke and Wilson [BWl], gives a sufficient
and obtain a d-edge coloring of G. .
condition for' a graph to be of class two. An independent set of edges in a
graph G is a set of edges, each two of which are independent (non-
Case 3. Suppose that z ¥= u and z ¥= Yj-l' Then neither u nor Yj-l is on P. adjacent). The edge independence number (31 (G) of G is the maximum
Thus, in particular, no edge incident with u or Yk-l is assigned the cardinality among the independent sets of edges of G.
color Ct. Thus we may assign the edge ek-l the color·a:: and each of the
edges eo, el," . ,ek-2 its dual color to obtain a d-edge coloring of G. 0
Theorem 8.21
If we take d = ~(G) + 1 in Theorem 8.18, then we obtain the afore- Let G be a graph of size m. If
mentioned result of Vizing [V3].
m> ~(G). (31 (G),

then G is of class two.

Corollary 8.19
IfG is a nonempty graph, then Xl(G) ~ 1 + ~(G).
Proof
By setting d = ~(G) in Theorem 8.18, we obtain a result due to Fournier Assume that G is of class one. Then Xl (G) = l:.(G). Let a Xl (G)-edge
[F9]. coloring of G be given. Each edge color class of G has at most (31 (G)
edges. Therefore, m ~ fl(G)· (31 (G). 0
Corollary 8.20
Since (31 (G) ~ L! nJ for every graph G of order n, we have an immediate
If G is a nonempty graph in which the vertices of maximum degree are indepen- consequence of the preceding result. An (n, m) graph G is called overfull if
dent, then Xl (G) = ~(G). m> fl(G). L!nJ.
With the aid of Corollary 8.19, the set of all nonempty graphs can be
divided naturally into two classes. A nonempty graph G is said to be of Corollary 8.22
class one if Xl (G) = ~(G) and of class two if Xl(G) = 1 + ~(G). The ~ain Every overfull graph is of class two.
problem, theri, is to determine whether a given graph is of class one or of
class two. It should be emphasized that Theorem 8.21 and· its corollary provide
The set of all edges of a graph G receiving the same color in an strictly sufficient conditions for a graph to be of class two. There exist
edge coloring of G is called an edge color class. By Vizing's theorem, the graphs with relatively few edges that are of class two. Of course, the
edge chromatic number of an r-regular graph G is either r or r + 1. For odd cycles are of class two, but, then, they are regular of odd order.
example, C n (n ~ 3) is of class one if n is even and of class two if n ir' The Petersen graph is of class two. A cubic graph of class two whose
odd. If Xl (G) = r for an r-regular graph, then necessarily each colcir girth is at least 5 is called a snark. Thus the Petersen graph is a snark.
class in a Xl (G)-edge coloring of G induces a spanning, I-regular sub- Isaacs [Il] has shown that there exist infinitely many snarks. For example,
graph of G. Thus, as we shall see in Chapter 9, Kn is of class one if n is the graph of Figure 8.7 is called the double-star snark.
even and of class two if n is odd. More generally, every regular graph Hilton [H14] and Chetwynd and Hilton [CH3] conjectured that a graph
of odd order is of class two. It .is not true, however, that every regular G of order n with ~(G) > Inis of class two if and only if G contains an
graph of even order is of class one; the Petersen graph, for example, is overfull subgraph H with ~(G) = l:.(H). Certainly if G contains an over-
of class two. full sUbgraph H with ~(G) = D.-(H), then G is of class two. The converse
240 Graph colorings Edge colorings 241
Theorem 8.23
Let G be a connected graph of class two that is minimal with respect to edge
chromatic number. Then every vertex of G is adjacent to at least two vertices of
degree .6.(G). In particular, G contains at least three vertices of degree .6.(G).

Theorem 8.24
Let G be a connected graph of class two that is minimal with respect to edge
chromatic number. If u and v are adjacent vertices with degu = k, then v is
adjacent to at least .6.(G) - k + 1 vertices of degree .6.(G).

We next examine to which class a graph belongs if it is minimal with

respect to edge chromatic number.

Theorem 8.25
Let G be a connected graph with .6.(G) = d ~ 2. Then G is minimal with respect
to edge chromatic number if and only if either:
(i) G is of class one and G = Kl,d or
(ii) G is of class two and G - e is of class one for every edge e of G.

Figure 8.7 The double-star snark.

Proof

has been established for several classes of graphs; For example, if G is Assume first that G=KI,d' Then Xl(G)=.6.(G)~2 while Xl(G-e) =
a complete s-partite graph for some s, then G is of class two only if G .6.(G) -1 for every edge e of G. Next, sqppose that G is of class two and
contains an overfull subgraph H with .6.(G) = .6.(H). that G - e is of class one for every edge e of G. Then, for an arbitrary edge
When discussing vertex colorings, we found it useful to consider e of G, we have
graphs that are critical with respect to chromatic number. Now that we Xl (G - e) = .6.(G - e) < 1 + .6.(G) = Xl (G).
are investigating edge colorings, it proves valuable to consider certain
minimal graphs. Conversely, assume that Xl (G - e) < Xl (G) for every edge e of G. If G is
A graph G with at least two edges is minimal with respect to edge of class one, then
chromatic number (or simply minimal if the parameter is clear from .6.(G):::;; .6.(G - e) + 1:::;; Xl(G - e) +1 = Xl(G) = .6.(G).
context) if Xl (G - e) = Xl (G) - 1 for every edge e of G. Since isolated
vertices have no' effect on edge colorings, it is natural to rule out isolated Therefore, .6.(G - e) = .6.(G) -1 for every edge e of G, which implies that
vertices when considering such minimal graphs. Also, since the edge G = KId'
chromatic number of a disconnected graph G having only nontrivial If G is of class two, then
components is the maximum of the edge . chromatic numbers of the
components of G, every minimal graph without isolated vertices is Xl(G - e) + 1 = Xl (G) = .6.(G) + 1
connected. Therefore, the added hypothesis that a. minimal graph G so that Xl (G - e) = .6.(G) for every edge e of G. Suppose that G contains an
is connected is equivalent to the assumption that G has no isol<;lt~~l> edge el such that G - el is of class two. Then Xl(G - eI) = .6.(G - ed + 1.
vertices. Hence, .6.(G - eI) < .6.(G), implying that G has at most two vertices of
Two of the most useful results dealing with these minimal graphs are degree ~(G). This, however, contradicts Theorem 8.23 and completes
also results of Vizing [V4], which are presented without proof. the proof. 0
 Edge colorings 243
242 Graph colorings
A graph G with at least two edges is called class minimal if G is of class
two and G - e is of class one for every edge e of G. It follows that a class
minimal graph without isolated vertices is necessarily connected. On the
basis of Theorem 8.25, we conclude that except for star graphs, class
minimal graphs are connected graphs that are minimal with respect to
edge chromatic number, and conversely.
A lower bound on the size of class minimal graphs is given next in yet
Figure 8.8 Planar graphs of class two.
another result by Vizing [V4].
Theorem 8.27
Theorem 8.26 If G is a planargraph with i.l(G) ?o 10, then G is of class one.
If G is a class minimal graph of size m with i.l(G) = d, then
Proof
m ?o 1(3d2 + 6d -1).
Suppose that the theorem is not true. Then among the graphs for which the
theorem is false, let G be a connected graph of minimum size. Thus, G is
Proof
planar, i.l(G) = d ~ 10, and Xl (G) = d + 1. Furthermore, G is minimal with
Without loss of generality, we assume that G is connected. Suppose that respect to edge chromatic number. By Corollary 6.5, G contains vertices of
8( G) = k and that deg u = k. By Theorem 8.23, the vertex u is adjacent to at degree 5 or less. Let S denote the set of all such vertices. Define H = G - S.
least two vertices of degree d. Let v be such a vertex. By Theorem 8.24, v is Since His planar, H contains a vertex w such that degH w ~ 5. Because
adjacent to at least d - k + 1 vertices of degree d. Since the order of G is at degG w > 5, the vertex w is adjacent to vertices of S. Let v E S such that
least d + 1, we arrive at the following lower bound on the sum of the wv E E(G), and let degG v = k ~ 5. Then, by Theorem 8.24,w is adjacent
degrees of G: to at least d - k + 1 vertices of degree d, but d - k + 1 ?o 6 so that w is
adjacent to at least six vertices of degree d. Since d ~ 10, w is adjacent to
2m ?o [d(d - k + 2) + k(k - 1)] = [~ - (d + l)k + (d 2 + 2d)]. (8.1)
at least six vertices of H, contradicting the fact that degH W ~ 5. 0
However, expression (8.1) is minimized when k = (d + 1}/2 so that
Seymour [52] conjectured that a planar graph is of class two if and only
2m?o (d;,1)2_(d~1)2 +d2+2d if G contains an overfull subgraph H with i.l(G) = i.l(H). If true, this
conjecture would imply that every planar graph G with i.l(G) ?o 6 is of
or class one.
More on edge colorings can be found in Fiorini and Wilson [FW1],
which is devoted to that subject.
There is a coloring that assigns colors to both the vertices and the edges
In the next section we shall be discussing various colorings of plan~r
of a graph. A total coloring of a graph G is an assignment of colors to the
graphs, primarily vertex colorings and region colorings. We briefly con-
elements (vertices and edges) of G so that adjacent elements and incident
sider edge colorings of planar graphs here. In this context, our chief
elements of G are colored differently. A k-total coloring is a total coloring
problem remains to determine which planar graphs are of class one and
that uses k colors. The minimum k for which a graph G admits a k-total
which are of class two. It is easy to find planar graphs G of class one for
coloring is called the total chromatic number of G and is denoted by X2(G).
which i.l(G) = d for each d ?o 2 since all star graphs are planar and of class
one. There exist planar graphs G of class two with i.l(G) = d for d = 2,3,4, Certainly, xdG) ?o 1 + i.l(G). The total chromatic number was introduced
by Behzad [Bl] who made the following conjecture.
5. For d = 2, the graphK3 has the desired properties. For d = 3, 4, 5, the
graphs of Figure 8.8 satisfy the required conditions.
It is not known whether there exist planar graphs of class two having Total Coloring Conjecture
maximum degree 6 or 7; however, Vizing [V4] proved that if G is plana!,'; For every graph G,
and i.l(G) ;;:: 8, then G must be of class one. We prove the follOWing, .
X2(G) ~ 2+ ~(G).
somewhat weaker, result.
 Graph colorings Edge colorings 245
244
Strong evidence for the truth of the Total Coloring Conjecture waspro~ and so there is a good permutation of the colors assigned to the edges of
vided by McDiarmid and Reed [MRl]. They showed that the probability G. Now permute the edge colors according to this good permutation. Let
that a graph G of order n satisfies Xz(G) ~ 2 + ~(G) approaches 1 as n G' be the supgraph of G induced by the edges of G that have the same
approaches infinity. Our next result, and its proof, indicate the types of color as one of their incident vertices. Then ~(G') ~ t and so, by Corollary
known bounds on the total chromatic number. A slightly stronger version 8.19, the edges of G' can be colored with t + 1 new colors. This gives a
of this result was obtained independently by Chetwynd and Hiiggkvist (Xl (G) + t + 2)-total coloring of G. 0
[CH2] and McDiarmid and Reed [MRl].
Since Xl (G) ~ 1 + ~(G), Theorem 8.28 has an immediate corollary.

Theorem 8.28 Corollary 8.29

If G is a nonempty graph of order nand t is an integer for which t! > n, then If G is a nonempty graph of order nand t is an integer satisfying tl >' n, then
Xz(G) ~XI(G)+t+2. Xz(G) ~ ~(G) + t + 3.

Proof
Let ~(G) = d. By Corollary 8.6, X(G) ~ d + 1. Since d + 1 ~ Xl (G) + I, it EXERCISES 8.2
follows that G is (Xl (G) + 1)-colorable and (Xl (G) + 1)-edge colorable. Let
8.23 Show that every nonempty regular graph of odd order is of class
a (Xl (G) + I)-coloring of G and a (Xl (G) + I)-edge coloring of G be given
using the same set of colors. We now consider permutations of the colors two.
assigned to the edges of G. Such a permutation is good if at each vertex vof 8.24 Let H be a nonempty regular graph of odd order, and let G be a
G, there are at most t -1 edges vu for whichvu and II are colored the graph obtained from H by deleting ! (~(H) -1) or fewer edges.
same. We first show that there is a good permutation of colors assigned Show that G is of class two.
to the edges of G (with the vertex colors remaining unchanged). For a
fixed vertex v and a set T of t incident edges, a permutation of the 8.25 Prove or disprove: If GI and Gz are class one graphs and H is a graph
colors assigned to the edges of G is called bad for T if every edge in T is with Gl ~ H ~ Gz, then H is of class one.
assigned the color of its incident vertex u I- v. Oearly, there are at most 8.26 Show that the Petersen graph is Qf class two.
(Xl (G) + 1 - t)! such permutations. Since at each vertex v there are at
most (~) sets of t edges incident with v, the total number of permutations 8.27 Prove that every hamiltonian cubic graph is of class one.
of the colors assigned to the edges of G that are bad for some t-element set
8.28 (a) Show that each graph in Figure 8.8 is of class two.
T is at most
(b) Show that the two graphs of order 5 in Figure 8.8 are class
minimal.
nG)(XI(G) + 1- t)!.
8.29 Determine the class of each of the five regular polyhedra.
If n(1)«XI(G)+ 1- t)! is less than (XI(G) + 1)1, the total number of 8.30 Determine the class of Kr,r-
permutations of the colors assigned to the edges of G, then the desired
good permutation exists. Since n < t!, it follows that 8.31 Show that a cubic graph with a bridge has edge chromatic number 4.
8.3~ Show that there are nO connected class rr4nimal graphs of order 4
n(~)(d - t)! < d!. or 6.
8.33 Let G be a graph of class two. Prove that G contains a class minimal
Thus, since Xl (G) + 1 > d, we conclude that subgraph H such that ~(H) = ~(G).

n(~)<Xl(G) + 1- t)! < (Xl (G) + I)!, 8.34 (a) Prove that X2(G) ;::: 1 + ~(G) for every graph G.
(b) Verify the Total Coloring Conjecture for graphs G with ~(G) ~ 2.
246 Graph colorings Map colorings and flows 247
(c) Determine X(G), Xl(G) and X2(G) for the (5,7) graph G in them (and who is now professor of mathematics at the South African
Figure 8.8. University, Cape Town), showed me the fact that the greatest
necessary number of colours to be used in colouring a map so as to
avoid identity of colour in lineally contiguous districts is four. I
should not be justified, after this lapse of time, in trying to give his
proof....
8.3 MAP COLORINGS AND FLOWS
With my brother's permission I submitted the theorem to Professor
It has been said that the mapmakers of many centuries past were aware of
De Morgan, who expressed himself very pleased with it; accepted it
the 'fact' that any map on the plane (or sphere) could be colored with four as new; and, as I am informed by those who subsequently attended
or fewer colors so that no two adjacent countries were colored alike.. 'fwo his classes, was in the habit of acknowledging whence he got his
countries are considered to be adjacent if they share a common boundary information.
line (not simply a single point). As was pointed out by May [M3], If I remember rightly, the proof which my brother gave did not
however, there has been no indication in ancient atlases, books on cartog- seem altogether satisfactory to himself; but I must refer to him
raphy, or books on the history of mapmaking that people were familiar those interested in the subject.
with this so-called fact. Indeed, it is probable that the Four Color Problem, . On the basis of this note, we seem to be justified in proclaiming that the
that is, the problem of determining whether the countries of any map on Four Color Problem was the creation of one Francis Guthrie.
the plane (or sphere) can be colored with four or fewer colors such that Returning to the letter of De Mor&an to Hamilton, we note the very
adjacent countries are colored differently, originated and grew in the prompt reply of disinterest by Hamilton to De Morgan on October 26
minds of mathematicians. 1852: '
What, then, is the origin of the Four Color Problem? The first written
reference to the problem appears to be in a letter, dated October 23,1852, I am not likely to attempt your 'quaternion of colours' very soon.
by Augustus De Morgan, mathematics professor at University College,
London, to Sir William Rowan Hamilton (after whom 'hamiltonian Before proceeding further with this brief historical encounter with the
graphs' are named) of Trinity College, Dublin. The letter by De Morgan Four Color Problem, we pause in order to give a more pn;!cise mathema-
reads in part: tical statement of the problem.
A plane graph G is said to be n-regioncolorable if the regions of G can be
A student of mine asked me today to give him a reason for a fact
colored with n or fewer colors so that adjacent regions are colored differ-
which I did not know was a fact - and do not yet. He says that if
entlr The Four Color Problem is thus the problem of settling the following
a figure be anyhow divided and the compartments differently
conjecture. .
coloured so that figures with any portion of common boundary
line are differently coloured - four colours may be wanted, but no
more.... Query cannot a necessity for five or more be invented.... The Four Color Conjecture
But it is tricky work ... what do you say? And has it, if true, been
noticed? My pupil says he guessed it in colouring a map of England. Every map (plane graph) is 4-region colorable.
The more I think of it, the more evident it seems. If you retort with
some very simple case which makes me out a stupid animal, I think I In dealing with the Four Color Conjecture, one need not consider aJJ
must do as the Sphynx did.... plane graphs, as we shall now see. .,
The region chromatic number X* (G) of a plane graph G is the minimum n
The student referred to by De Morgan was Frederick Guthrie. By 1880
the problem had become quite well-known. During that year, Frederick for whic.h Gis n-region col?rable. Since X*(G) is the maximum regi,Qu
Guthrie published a note in which he stated that the originator of the chromatic number among Its blocks, the Four Color Problem. can' be
restated as determining whether every plane nonseparable graph is 4~
question asked of De Morgan was his brother, Francis Guthrie. We
region colorable. )
quote from Frederick Guthrie's note [G9]:
In graph theory the Four Color Problem is more often stated in terms of
Some thirty years ago, when I was attending Professor De Morgan's coloring the vertices of a graph; that is, coloring the graph. In this form,
class, my brother, Francis Guthrie, who had recently ceased to attend the Four Color Conjecture is stated as follows.
248 Graph colorings Map colorings and flows 249
The Four Color Conjecture Proof
Every planar graph is 4-colorable. Without loss of generality, we may assume that the graphs under consid-
eration are connected.
It is in terms of this second statement that the Four Color Problem will Suppose that every planar graph is 4-colorable. Let G be an arbitrary
be primarily considered. We now verify that these two formulations oIthe connected plane graph, and consider Gd , the underlying graph of its
Four Color Conjecture are indeed equivalent. Before doing this, however, dual Gd . T~o regions of G are adj~cent if and only if the corresponding
we require the concept of the dual of a plane graph. vertices of Gd are adjacent. Since Gd is planar, it follows, by hypothesis,
Recall that a pseudograph G is a multigraph in which loops are per- that Gd is 4-colorable; thus, G is 4-region colorable.
mitted. For a given connected plane graph G we construct a pseUdograph For the converse, assume that every plane graph is 4-region colorable,
Gd as follows. A vertex is placed in each region of G, and these vt;;rtices and let G be an arbitrary connected plane graph. As we have noted, the
constitute the vertex set of Gd • Two distinct vertices of Gd are then joined dual Gd oiG can be embedded in the plane so that each region of Gd con-
by an edge for eac:h edge common to the boundaries of the two corre- tains exactly one vertex of G. If Gd is not a graph, then it can be converted
sponding regions of G. In addition, a loop is added at a vertex v of Gd into a graph G' by inserting two vertices into each loop of Gd and by
for each bridge of G that belongs to the boundary of the corresponding placing a vertex in all but one edge in each set of parallel edges joining
region. Each edge of Gd is drawn so that it crosses its associated edge of the same two vertices. Two vertices of G are adjacent if and only if the
G but no other edge of G or Gd (which is always possible); hence, Gd is corresponding regions of G' are adjacent. Since G' is 4-region colorable,
planar. The pseUdograph Gd is referred to as the dual of G. In addition to G is 4-colorable. 0
being planar, Gd has the property that it has the same size as G and can be
drawn so that each region of Gd contains a single vertex of G; indeed;
With these concepts at hand, we now return to our historical account
(Gd)d = G. If each set of parallel edges of Gd joining the same two vertices
of the Four Color Problem. We indicated that this problem was
is replaced by a single edge and a~ loops are deleted, the result is a graph,
evidently invented in 1852 by Francis Guthrie. The growing awareness
referred to as the underlying graph Gd of Gd • These concepts are illustrated in
of the problem was quite probably aided by De Morgan, who often spoke
Figure 8.9, with the vertices of Gd represented by solid circles.
of it to other mathematicians. The first known published reference to the
Four Color Problem is attributed to De Morgan in an anonymous article
in the April 14, 1860 issue of the journal Athenaeum. By the 1860s
the problem was becoming rather Widely known. The Four Color
Problem received added attention when on June 13, 1878, Arthur
Cayley asked, during a meeting of the London Mathematical· Society,
whether the problem had been solved. Soon afterwards, Cayley [C3]
publisheda paper in which he presented his views on why the problem
appeared to be so difficult. From his discussion, onemight very well infer
the existence of planar graphs with an arbitrarily large chromatic
number.
One of the most important events related to the Four Color Problem
occurred on July 17, 1879, when the magazine Nature carried an
announcement that the Four Color Conjecture had been verified by
Alfred Bray :{<empe. His proof of the conjecture appeared in a paper
[K2] published in 1879 and was also described in a paper [K3] published
FigureB.9 The dual (and its underlying graph) of a plane graph. in 1880. For approximately ten years, the Four Color Conjecture was
considered to be settled. Then in 1890, Percy John Heawood [H9]
discovered an error in Kempe's proof. However, using Kempe's tech-
Theorem 8.30
nique, Heawood was able to prove that every planar graph is 5-
Every planar graph is 4-colorable if and only if every plane graph is 4-region colorable; This result was referred to, quite naturally, as the Five Color
colorable. Theorem.
250 Graph colorings
Map colorings and flows 251
Theorem 8.31
distinctions, and nearly 1200 hours of computer time were required to
Every planar graph is 5-colorable. perform extensive computations. A simpler, and more easily checked,
proof of the Four Color Theorem was obtained much later by Robertson,
Sanders, Seymour and Thomas [RSSTl]. Even their proof, however,
Proof required extensive computer calculations.
The proof is by induction on the order n of the graph. For n :::; 5,the result
is obvious. Theorem 8.32
Assume that all planar graphs with n -1 vertices, n > 5, are 5-color-
able, and let G be a plane graph of order n. By Corollary 6.5: G contains Every planar graph is 4-colorable.
a vertex v of degree 5 or less. By deleting v from G, we obtaill~he pl~ne
graph G - v. Since G - v has order n - 1, it is 5-colorable ?y the mductIve Although the Four Color Theorem has been established, other
hypothesis. Let there be given a 5-coloring of G ~ v, denotmg ~he col~rs by approaches have been suggested that might eliminate the heavy depen-
1, 2, 3, 4 and 5. If some color is not used in coloring the vertices adjacent dence on the computer. Here, the idea is to solve an equivalent problem.
with v, then V may be assigned that color, producing a 5-coloring ~f G In section 8.1, for example, we saw the conjecture of Hadwiger that Kk is a
itself. Otherwise, deg v = 5 and all five colors are used for the vertices subcontraction of every k-chromatic graph. For k = 5, this conjecture is
adjacent with v. . . equivalent to the Four Color Theorem. We sketch the proof below. The
Without loss of generality, we assume that Vb V2, V3, V4, Vs are the fi~e details can be found in Wagner [W2].
vertices adjacent with and arranged cyclically about v and t~at Vi IS
assigned the color i, 1 :::; i :::; 5. Now consider any two colors assIgned to
nonconsecutive vertices Vj, say 1 and 3, and let H be the subgraph of G - v Theorem 8.33
induced by all those vertices colored 1 or 3. If VI and v3.belong to d~fere~t Every planar graph is 4-colorable if and only if Ks is a subcontraction of every 5-
components of H, then by interchangingthe colors asslgne~ to vertices r,rl chromatic graph.
the component of H containing Vv for example, a 5-colormg of G - v IS
produced in which no vertex adjacent with v is assigned the color 1. Thus
if we color V with 1, a 5-eoloring of G results. Proof
Suppose then that VI and V3 belong to the same cox,nponent of H.
Consequently, there exists a vI-V3 path P, all of whose vertIces are col~red Suppose that Ks is a subcontraction of every 5-chrorilatic graph. Let G
be a planar graph. Then, by Theorem 6.16, neither Ks nor K3 ,3 is a sub-
lor 3. The path P, together with the path V3,V, VI, produces a cycle C ill G
contraction of G. It follows that x( G) :::; 4.
that encloses v2, or V4 and vs. Hence there exists no v2-v4 path in G, all of
whose vertices are colored 2 or 4. Denote by F the subgraph of G induced For the converse, assume that every planar graph is 4-colorable, but that
there are 5~chromatic graphs for which Ks is not a subcontraction. Let G
by all those vertices colored 2 or 4..Interchanging the colors. of the vertic~s
in the component of F containiilg V2, we arrive at a 5-colormg of G - V. ill be a counterexample of minimum order. It is straightforward to show that
which no vertex adjacent with v is assigned the color 2. If we color v WIth G is 4-eonnected. But then G is a 4-connected graph for which K s is not a
2, a 5-coloring of G results. 0 subcontraction, implying that G is planar <Wagner [W2]). Since X(G) = 5,
this produces a contradiction to the Four Color Theorem. 0
In the 86 years that followed the appearance of Heawood's paper,
Another approach, due to Tait [Tl], involves coloring the edges of
. numerous attempts were made to unlock the mystery of the Four Color
bridgeless cubic planar graphs. In establishing this result it is convenient
Problem. Then on June 21, 1976, Kenneth Appel and Wolfgang Haken
to make use of the group Z2 x Zz, denoting its elements by (0,0), (0, I),
announced that they, with the aid of John Koch, had verified the Four (1,0) ·and 0,1).
Color Conjecture. .. .
Appel and Haken's proof [AHKll was logically qUite slffiple;. m fact,
many of the essential ideas were the same as those used (unsuccessfully) Theorem 8.34
by Kempe and, then, by Heawood. However, their proof was combinafo~'
rially complicated by the extremely large number of necessary case Every planar graph is 4-colorable ifand only ifevery bridgeless cubic planar graph
is 3-edge colorable.
252 Graph colorings Map colorings and flows 253
Proof
H: H:
By Theorem 8.30, it suffices to show that every plane graph is 4-region
colorable if and only if every bridgeless cubic planar graph is 3-edg:~
colorable. ...
Assume that every plane graph is 4-region colorable and let G be a
bridgeless cubic plane graph. Let the regions of G be colored with the
elements of Zz x Zz. Since G contains no bridges, each edge of G belongs
to the boundary of two (adjacent) regions. Define the color of an edg~tobe
the sum of the colors of those two regions bounded, in part, by the edge.
Since every element of Zz x Z2 is self-inverse, no edge of G is assigned the
color (0,0). However, since Z2 x Zz is a group, it follows that the three
H':
edges incident with a vertex are assigned the colors (0,1), (1,0) and
0, 1). Hence G is 3-edge colorable.
Conversely, assume that every bridgeless cubic planar graph is 3-edge
colorable. We show that every plane graph is 4-region colorable.
Certainly, every plane graph is 4-region colorable if and only if every
(a) (b)
bridgeless plane graph is 4-region colorable. Furthermore, suppose that
every cubic bridgeless plane graph is 4-region colorable. Let H be a Figure 8.10 Constructing a cubic graph H' from a graph H.
bridgeless plane graph. We now construct a cubic plane block H' from
H as follows. If H contains a vertex v of degree 2 that is incident with as e is crossed. In order to show that the color of 5 is well-defined, we
edges e and f, we subdivide e and f by introducing vertices V1 and Vz verify that the color assigned to 5 is independent of the curve A;
into e and f, respectively, remove v, and then identify V1 and V2, respec- however, this will be accomplished once it has been shown that if C is
tively, with the vertices of degree 2 in a copy of the graph K1,1,Z (Figure any simple closed curve not passing through vertices of G, then the
8.10(a». If H contains a vertex U of degree t ~ 4, incident with the con- sum of the colors of the edges crossed by C is (0,0). Let C be such a
secutive edges el, ez, ... ,et, then we subdivide each ei by inserting a curve. If no vertex of G lies interior to C, then each edge crossed by C
vertex Ui in each ei, i = 1,2, ... , t,.removing the vertex u, and identifying is crossed an even number of times; and since each element of Zz x Zz
each Ui with the corresponding vertex of the t-cycle Ul, Uz, ••• , Ut> Ul is self-inverse, it follows' that the sum of the colors of the edges crossed
(Figure 8.10(b». By hypothesis, X*(H' ) ::::; 4, for the resulting cubic plane by C is (0,0). If C encloses vertices, then, Without loss of generality, we
block Hi hence there exists a k-region coloring, k::::; 4, of H'. However, by may assume that any edge crossed by C is crossed exactly once. We
identifying all vertices of the graph K1,1,Z for each vertex of degree 2 and proceed as follows. Let el' ez, ... ,es be those edges crossed by or lying
by identifying the vertices of the t~cycle for each vertex of degree t· ~ 4, the interior to C, and suppose that the first r of these edges are crossed by
graphH is reproduced and a k-region coloring of His induced. Hence H is C. Observe that the sum of the colors of the three edges incident with any
4-region colorable. vertex is (0,0); hence, if we were to total these sums for all vertices lying
Thus the proof will be complete once we have shown that if every interior to C we, of course, arrive at (0,0) also. However, this sum also
bridgeless cubic planar graph is 3-edge colorable, then every bridgeless .equals
cubic plane graph is 4-region colorable.
Let G be a bridgeless cubic plane graph. Then G is 3-edge colorable. Le,t; c(el) +c(ez) + ... + c(er) +2[c(er+1) + c(er+z) + ... + c(es )],
the edges of G be colored with the nonzero elements of Zz x Zz. Let R be where c(ei)' indicates the color of the edge ei' Therefore, c(el) +
some region ofG and assign the color (0,0) to it. Let S be some other c(ez) + ... + c(er) = (0,0), that is, the sum of the color of the edges crossed
region of G. We now assign a color (an element of Zz x Zz) according by C is (0,0).
to the following rule. Let A be a continuous curve joining a point of It now remains to show that this procedure yields a+region coloring
region R with a point of region S such that A passes through no vertex of G. However, if R1 and R z are two adjacent regions, sharing the edge e
of G. We now define the color of S to be the sum of the colors of those in their boundaries, then the colors assigned to R1 and Rz differ by
edges crossed by A, where the color of an edge e is counted as many times c(e) =f. (0,0). This completes the proof. 0
254 Graph colorings Map colorings and flows 255
It follows from Theorems 8.32 and 8.34 that every bridgeless cllbic
e ep(e)
planar graph is 3-edge colorable.
G: 3 (v2 , VI) 2
Corollary 8.35 v2 (V6 • VI) ~2
v6 V6
v2
(V2 • V:v -1
Every bridgeless cubic planar graph is 3-edge colorable.
(V6 • vi) 1
v v3
Several conjectured extensions of the Four Color Theorem remain open. Vs 3 Vs (v3 • V& 1
For example, Hadwiger's Conjecture, equivalent to the Four Color (vs• V:v 1
Theorem for k = 5, is unsettled for k ~ 7. In another direction, Corollary v4 v4 (V3 • v~ -1
8.3S states that everybridgeless cubic planar graph is 3-edge colorable.
(vs• v~ 1
The condition that the cubic graph is bridgeless is certainly necessary
(Vs• V& -2
here since no cubic graph with a bridge is 3-edge colorable (Exercise
8.31). Furthermore, since the Petersen graph is bridgeless and has edge- Figure 8.n· Flow construction.
chromatic number 4, we cannot drop the requirement of planarity in
Corollary 8.35. Tutte [T1S], however, made the following conjecture. Nowhere-zero flows in planar graphs are of particular interest because
of their relationship to region colorings. Let G be a bridgeless plane graph,
Tutte's First Conjecture with the edges of G oriented arbitrarily, and let c be a k-region coloring
If G is a bridgeless cubic graph with Xl (G) = 4, then the Petersen graph is a of G. Thus for each region R of G, the region R is colored c(R), where
subcontraction of G. c(R) E {1,2, ... ,k}. For each oriented edge e = (u,v) of G, define </>(e) to
be c(R 1 ) - c(Rz), where Rl is the region to the right of e = (u, v) as we
Since the Petersen graph is a subcontraction of no planar graph, Tutte's travel along e from u to v and R2 is the region to the left of e. An example
First Conjecture, if true, together with Theorem 8.34, implies the Four of this construction is given in Figure 8.11.
Color Theorem. It is straIghtforward to verify that the integer-valued function </> defined
Tutte's First Conjecture about the edge-chromatic number of bridgeless above on the bridgeless plane graph G is a nowhere-zero k-flow. Thus if a
cubic graphs is a special case of yet another conjecture of Tutte concerning bridgeless plane graph G is k-region colorable, then G has a nowhere-zero
nowhere-zero flows. k-flow. Conversely, Tutte [T12] showed that if a bridgeless plane graph G
Let D be an oriented graph and let k ~ 2 be an integer. A nowhere-zero has a nowhere-zero k-flow, then G is k-region colorable. Consequently,
k-fIow in D is a function </> defined on E(D) so that another equivalent form of the Four Color Theorem 'can be given.

(i) </>(e) E {±l, ±2, ... , ±(k -I)} for each arc e of D, and
Theorem 8.36
(ii) for each vertex v of D, .
L
(v,u) E E(D)
</>(v, u) = L
(u,v) EE(D)
</>(u, v).
Every planar graph is 4-colorable ifand only ifevery bridgeless planar graph has a
nowhere-zero 4-flow.

If an orientation D of a graph G has a nowhere-zero k-flow </> and D' Applying Theorems 8.32 and 8.36, we obtain Corollary 8.37, the 'flow
is the orientation of G obtained by replacing some arc (u, v) with the analogue' of Corollary 8.35.
arc (v, u), then D' has the nowhere-zero k-flow </>0' defined by
</>(e), if e I- (u, v) Corollary.8.37
</>0' ()
e = {
-</>(e) , if e = (u,v). Every bridgeless planar graph has a nowhere-zero 4-fIow.
Thus if some orientation of G has a nowhere-zero k-flow, then so does
every orientation of G. Consequently, when we say that a graph G has a As with Corollary 8.35, the condition 'bridgeless' is necessary in
nowhere-zero k-flow, then we mean, in fact, that every orientation of G Corollary 8.37 since no graph with a bridge has a nowhere-zero k-flow
has a nowhere-zero k-flow. for any k ~ 2 (Exercise 8.37). Perhaps not surprisingly, the Petersen
256 Graph colorings Map colorings and flows 257
graph has no nowhere-zero 4-flow and so the planarity condition in For cubic graphs, the two Tutte conjectures are equivalent. Tutte also
Corollary 8.37 cannot be deleted. conjectured that every bridgeless graph has a nowhere-zero S-flow.
Although this conjecture was made in 1954, it was not even known
unti11975 whether 5 could be replaced by some larger number k. Jaeger
Theorem 8.38 . • [J2] showed that every bridgeless graph has a nowhere-zero 8-flow; this
The Petersen graph has no nowhere-zero 4-flow. was improved by Seymour [54] who showed that every bridgeless graph
has a nowhere-zero 6-flow. Tutte's 5-flow conjecture remains open.
More information on the history of the Four Color Problem can be
Proof found in Biggs, Lloyd and Wilson [BLW1] and in Jensen and Toft [TTl].
Suppose, to the contrary, that the Petersen graph P has a nowhere-zero 4- Additional material on flows and their relationship to graph colorings can
flow cf> corresponding to some orientation pi of P. If cf>(u, v) < 0 for some be found in Seymour [55, 56].
arc (u, v), then we can reverse the direction of the arc(u, v) to produce The Four Color Theorem deals with the maximum chromatic number
(v, u) and assign the value cf>(v, u) = -cf>(u, v), obtaining another among all graphs that can be embedded in the plane. The chromatic number
nowhere-zero 4-flow in pl. By repeating this procedure, we obtain a ofasurface (where, as always, a surface is a compact orientable 2-manifold)
nowhere-zero 4-flow cP in pi with 0 < cP(e) < 4 for every arc e of p'. Sk of genus k, denoted X(Sk), is the maximum chromatic number among
Since P is cubic, it follows that for each vertex v of p' the three arcs incident all graphs that can be embedded on 5k • The surface 50 is the sphere and
with v have flow values 1, 1, 2 or 1, 2, 3. the Four Color Theorem states that X(5 0 ) = 4. Heawood [H9] showed that
Now, those arcs e with cf>(e) = 2 produce a 1-regular spanning sUbgraph X(5 1 ) = 7; that is, the chromatic number of the torus is 7. Moreover,
of P. Thus, the corresponding edges can be assigned a single color in an Heawood was under the impression that he had proved
edge coloring of P. Therefore, those arcs e for which cf>(e) = 1 or cP(e) = 3
produces a 2-regular spanning subgraph H of P. X(Sk) = l7+~J
A vertex v of H is said to be oftype I if there is an arc e of pi with cP(e) = 1
that is incident to v; while v is oftype II if there is an arc e of pI with cf>(e) = 1 for all k > O. However, Heffter [HI0] pointed out that Heawood had only
that is incident from v. Consequently, every vertex of H is of type I or of established the upper bound:
type II, but not both. Moreover, in any cycle of H the vertices alternate
between type I and type II. Thus every component. of H is an even
cycle. It follows that H is 2-edge. colorable, so P is 3-edge colorable,
X(Sk) ~ l7 + ~ J. (8.2)

The statement that X(Sk) = l(7 +';1 + 48k/2)J for all k> 0 eventually
which produces a contradiction. 0
became known as the Heawood Map Coloring Conjecture. In 1968, Ringel
and Youngs [RY1] completed a remarkable proof of the conjecture,
The technique used in the proof of Theorem 8.38 can be used to show
which has involved a number of people over a period of many decades.
that if a bridgeless cubic graph G has a nowhere-zero 4-flow, then G is 3-
This result is now known as the Heawood Map Coloring Theorem. The proof
edge colorable. In fact, the converse is also true.
we present assumes inequality (8.2).

Theorem 8.39 Theorem 8.40

Let G be a bridge/ess cubic graph. Then G has a nowhere-zero 4-flow ifand only if For every positive integer k,
G is 3-edge colorable.
X(Sk) = l7+~J.
In view of Corollary 8.37 and Theorem 8.38, Tutte [T1S] proposed a
second conjecture.
Proof
Tutte's Second Conjecture Because of inequality (8.2), it remains only to verify that

If G is a bridgeless graph with no nowhere-zero 4-flow, then the Petersen graph

is a subcontraction of G.
X(Sk) ~ l7+~J
258 Graph colorings
for all k > O. Define CHAPTER 9

n= l7+~l
so that n ~ (7 + VI + 48k/2). From this, it follows that k?;
Matchings, factors and
(n - 3)(n - 4)/12. Therefore,
decompositions
k?; l
(n - 3i~n - 4) j. (8.3)

Since the right-hand expression of· (8.3) equals the genus of K n (by
Theorem 7.10), gen(KIl ) ~ k so that
X(Sgen(Kuj) ~ X(Sk)' We now consider special subgraphs that a graph may contain or into
Clearly KIl is embeddable on Sgen(Kuj; consequently, X(Sgen(Kuj)?; n, which a graph may be decomposed. In particular, we emphasize iso-
implying that X(Sk) ?; n. D morphic decompositions. This leads us to a consideration of graph
labelings.
As a consequence of the Four Color Theorem, Theorem 8.40 also holds
for k = O. A thorough discussion of the Heawood Map Coloring Problem
can be found in Ringel [R8] and White [W4]. 9.1 MATCHINGS AND INDEPENDENCE IN GRAPHS

Recall that two edges in a graph G are independent if they are not adjacent
in G. A set of pairwise independent edges of G is called a matching in G,
EXERCISES 8.3
while a matching of maximum cardinality is a maximum matching in G.
8.35 Use a proof similar to that of Theorem 8.31 to show that a(G) ~ 3 for
Thus the number of edges in a maximum matching of G is the edge
independence number ,81 (G) of G. In the graph G of Figure 9.1, the set
every planar graph G.
M 1 = {el,e4} is a matching that is not a maximum matching, while
8.36 Show that if a bridgelessplane graph is k-region colorable, then G M 2 = {el,e3,eS} and M 3 = {el,e3,e6} are maximum matchings in G.
has a nowhere-zero k-flow. If M is a matching in a graph G with the property that every vertex of G
8.37 Show that if G has a nowhere-zero k-flow, k ?; 2, then Gis bridgeless.
is incident with an edge of M, then M is a perfect matching in G. Clearly, if G
has a perfect matching M, then G has even order and (M) is a I-regular
8.38 Give an example of a graph G for which gen(G) = 2 and X(G) = spanning subgraph of G. Thus, the graph G of Figure 9.1 cannot have a
X(S2)' Verify that your example has these properties. perfect matching.
If M is a specified matching in a graph G, then every vertex ofG is
8.39 Use the result given in Corollary 8.9 to establish an upper bound for
the chromatic numbers of graphs embeddable on the torus. Discuss incident with at most one edge of M. A vertex that is incident with no
the sharpness of your bound. edges of M is called an M-vertex. The following theorem will prove to
be useful.

Theorem 9.1
Let M1 and M 2 be matchings in a graph G..Then each component of the spanning
subgraph H of G with E(H) = (M1 - M2 ) U (M2 - M 1 ) is one of the following
types:
(i) an isolated vertex,
(ii) an even cycIewhose edges are alternately in M 1 and in M2,
260 Matchings, factors and decompositions Matchings and independence in graphs 261
(iii) a nontrivial path whose edges are alternately in M I and in M z and such that edges of P belonging to M, and let M" = E(P) - M'. Since IM"I =IM'I + 1,
each end-vertex of the path is either an Mrvertex or an Mz-vertex but not the set (M ...,. M') U M" is a matching having cardinality exceeding that of
M, producing a contradiction.
both.
Conversely, let M I be a matching in a graph G, and suppose that there
exists no Mraugmenting path in G. We verify that M I is a maximum
Proof matching. Let M z be a maximum matching in G. By the first part of the
First we note that L1(H) ~ 2, for if H contains a vertex v such that proof, there exists no Mz-augmenting path in G. Let H be the spanning
degH v ~ 3, then v is incident with at least two edges in the same match- subgraph of G with E(H) = (M I - M z) U (M 2 - M I ). Suppose that HI is a
ing. Since t::..(H) ~ 2, every component of H is a path (possibly trivial) or a component of H that is neither an isolated vertex nor an even cycle. Then
cycle. Since no two edges in a matching are adjacent, the edges of each it follows from Theorem 9.1 that HI is a path of even length whose edges
cycle and path in H are alternately in M I and in M z. Thus each cycle in H is are alternately in M I and in M 2 , for otherwise, there would exist a path in
even. G that is either M I-augmenting or M 2-augmenting, which is impossible. It
Suppose that e = uv is an edge of Hand u is the end-vertex of a path P now follows by Theorem 9.1 that IMI - Mzl = 1M2 - MIl, which, in turn,
that is a component of H. The proof will be complete once we have shown implies that IMII = IM2 1. Hence, Mlis a maximum matching. 0
that u is an MI-vertex or an Mz-vertex but not both. Sincee E E(H), it
follows that e E M I M z or e E M z - MI' If e E M I - Mz, then u is not According to Theorem 9.2, if a matching M is given, it i~ possible to
an MI-vertex. We show that u is an Mz-vertex. If this is not the case, decide whether M is a maximum matching by determining whether G
then there is an edge f in Mz(thus f =1= e) such that f is incident to u. has an M-augmenting path.
Since e and f are adjacent, f ¢ MI' Thus, f E Mz - MI ~ E(H). This, In applications, maximum matchings in bipartite graphs have proved
however, is impossible since u is the end-vertex of P. Therefore, u is an to be most useful. The next result, namely Theorem 9.3, attributed to
M 2-vertex; similarly, if e E M z - M I , then u is an Mrvertex. 0 Konig [K9] and Hall [H5], is of interest in its own right.
In a graph G, a nonempty subset UI of V(G) is said to be matched to a
subset Uz of V(G) disjoint from UI if there exists a matching M in G such
that each edge of M is incident with a vertex of U I and a vertex of U2 , and
every vertex of UI is incident with an edge of M, as is every vertex of Uz. If
G:
M ~ M*, where M* is also a matching in G, we also say that UI is matched
under M* to Uz.
Let U be a nonempty set of vertices of a graph G and let its neighborhood
Figure 9.1 Matchings and maximum matchings. N(U) denote the set of all vertices of G adjacent with at least one element
of U. Then the set U is said to be nondeficient if IN(S)I ~ 151 for every non-
In order to present a characterization of maximum matchings, we intro- empty subset 5 of U..
duce two new terms. Let M be a matching in a graph G. An M-alternating
path of G is a path whose edges are alternately in M and not in M. An M-
augmenting path is an M-alternating path both of whose end-vertices are Theorem 9.3
M-vertices. The following characterization of maximum matchings is due Let G be a bipartite graph with partite sets VI and V2 • The set VI can be matched
to Berge [B5]. to a subset of V2 if and only if VI is nondeficient.

Theorem 9.2 Proof

A matching M in a graph G is a maximum matching if and only if there exists no
Suppose that VI can be matched to a subset of V z under a matching M*.
M-augmenting path in G.
Then every nonempty subset 5 of VI can be matched under M* to some
subset of V z, implying that IN(S)I ~ 151; so VI is nondeficient.
Proof To verify the converse, let G be a bipartite graph for which VI is non-
Assume that M is a maximum matching in G and that there exists an deficient and suppose that VI cannot be matched to a subset of V2 • Let M
augmenting path P of G. Necessarily, P has odd length, Let M' denote the be a maximum matching in G. By assumption, there is a vertex v in VI that
262 Matchings, factors and decompositions Matchings and independence in graphs 263
is an M-vertex. Let 5 be the set of all vertices of G that are connected to v 'may be reformulated to produce what is often referred to as Hall's
by an M-alternating path. Since M is a maximum matching, an application Marriage Theorem: If there are k girls, then the Marriage Problem has a
of Theorerr:. 9.2 yields v as the only M-vertex in S. solution if and only if every subset of j girls (1::::; j ::::; k) collectively
Let W1 = 5 n V 1 and let Wz = 5 n V z. Using the definition of the set S, know at least j boys.
together with the fact that no vertex of 5 - {v} is an M-vertex, we We have already noted that if M is a perfect matching in a graph G, then
conclude that W1 - {v} is matched under M to Wz· Therefore, (M) is a I-regular spanning subgraph of G. Any spanning subgraph of a
IWzl = IW11-1 and Wz <::;; N(W1 ). Furthermore, for every wE N(W1), graph G is referred to as a factor of G. A k-regular factor is called a k-factor.
the graph G contains an M-alternating v-w path so that N(W1) <::;; Wz· Thus F is a I-factor of a graph G if and only if E(F) is a perfect matching in
Thus,N(W1 ) = Wz and , G. The determination of whether a given graph contains a I-factor is a
problem that has received much attention in the literature. Of course, if
IN(W1 )1 = IWzl= IW1 1-1 <IW11· a graph G has a I-factor, then G has even order. A characterization of
This, however, contradicts the fact that V 1 is nondeficient. 0 ,graphs that contain I-factors has been obtained by Tutte [T13]. The
following proof of TuUe's theorem is due to Anderson [Al]. An odd
We are now in a position to present a well-known theorem due to Hall pomponent of a graph is a component of odd order.
[HS]. A collection 51, Sz, ... , Sk, k ~ I, of finite nonempty sets is said to
have a system of distinct representatives or a transversal if there exists a set
{S1' SZ, ... ,Sk} of distinct elements such that Sj E Sj for 1 ::::; i ::::; k. (For a
thorough treatment of transversals, see Mirsky [M8].) .A nontrivial graph G has a I-factor if and only if for every proper subset S of
V(G), the number of odd components of G - 5 does not exceed 151.

Theorem 9.4
A collection 51, Sz, ... ,Ski k ~ I, of finite nonempty sets has a system of distinct Let F be a I-factor of G. Assume, to the contrary, that there exists a proper
representatives if and only if the union of any j of these sets contains at least j subset Waf V(G) such that the number of odd components of G.,..W
elements, for each j such that 1 ::::; j ::::; k. exceeds IWI.For each odd component H ofG - W, there is necessarily
an edge of F joining a vertex of H with a vertex of W. This implies,
Proof however, that at least one vertex of W is incident with at least two
edges ofF, which is impossible. This establishes the necessity.
From the collection 51, Sz, ... ,Sk, k ~ I, of finite, nonempty sets we con- Next we consider the sufficiency. For a subset S of V(G), denote the
struct a bipartite graph G with partite sets VI and Vz in the following number of odd components of G - 5 by ko(G - 5). Hence, the hypothesis
manner. Let VI be the set {VI, Vz," ., Vk} of distinct vertices, where Vj of G may now be restated as ko(G - 5) ::::; 151 for every proper subset S of
corresponds to the set Sj, and let V z be a set of vertices disjoint from V 1
V(G). In particular, ko(G - 0) : : ; 101 = 0, implying that G has only even
such that IVzl = IU;=1 Sjl, where there is a one-to-one correspondence components and therefore has even order n. Furthermore, we note that for
k
between the elements of V z and those of Uj=1 Sj. The construction of G each proper subset S of V(G), the numbers ko(G - 5) and lSI are of the
is completed by joining a vertex V of VI with a vertex w of Vz if and same parity, since n is even.
only if V corresponds to a set Sj and w corresponds to an element of 5 j.
We proceed by induction on even positive integers n. If G is a graph of
From the manner in which G is defined, it follows that VI is nondeficient if
order n = 2 such that ko(G - 5) ::::; 151 for every proper subset S of V(G),
and only if the union of any j of the sets Sj contains at least j elements.
then G = Kz and G has a I-factor.
Now obviously, the sets Sj have a system of distinct representatives if and
Assume for, all graphs H of even order less than n (where n ~ 4 is an
only if V 1 can be matched to a subset of V z. Theorem 9.3 now produces the
even integer) that if ko(H - W) ::::; IWI for every proper subset Waf V(H),
desired result. 0 then H has a I-factor. Let G be a graph of order n and assume that
ko(G - 5) ::::; 151 for each proper subset S of V(G). We consider two cases.
The preceding discussion is directly related to a well-known combina-
torial problem called the Marriage Problem: Given a set of boys and a set of Case 1. Suppose that ko(G - S) < lSI for all subsets 5 of V(G) with
girls where each girl knows some of the boys, under what conditions can 2::::; 151 < n. Since ko(G- 5) and 151 are of the same parity, ko(G - 5) : : ;
all girls get married, each to a boy she knows? In this context, Theorem 9.4 151- 2 for all subsets 5 of V'(G) with 2 ~ 151 < n.Let e = uv be an edge
264 Matchings, factors and decompositions Matchings and independence in graphs 265
of G and consider G - u - v. Let T be a proper subset of V(G - u - v). It similar proof technique to that used in the proof of Theorem 9.5, Berge
follows that ko(G - u - v - T) ~ In
for suppose, to the contrary, that [BS] obtained the following extension of Tutte's theorem.
ko(G - u - v - T) > ITj. Then
ko(G - u - v - T) > ITI = jT U {u, v}j - 2, Theorem 9.6
so that ko(G (TU{u,v})) ~ ITu{u,v}I, contradicting our supposition. Let G be a graph of order n. If k is the smallest nonnegative integer such that for
Thus, by the inductive hypothesis, G - u - v has a I-factor and, hence, each proper subset 5 of V(G), the number of odd components of G - 5 does not
so does G. exceed 151 + 2k, then
Case 2. Suppose that there exists a subset R ofV(G) such that ko(G - R) = IRI,
where 2 ~ IRj < n. Among all such sets R, let 5 be one of maximum
cardinality, where ko(G - S) = 151 = k. Further, let G1> G2 ,···, Gk denote
the odd components of G - S. These are the only components of Again, by Tutte's theorem, if G is a graph such that for every proper
G - S, for if Go were an even component of G - 5 and Uo E V(Go), subset 5 of V( G), the number of odd components of G - S does not exceed
then kn(G - (5 U {uo})) ~ k+ 1 = IS U {uo}l, implying necessarily that lSI, then G has a I-factor. Of course, if G is a graph of even order such that
kn(G - (5 U {uo})) = IS U {uo}!, which contradicts the maximum property for every vertex-cut 5 of V(G), the number of components (odd or even) of
of 5. G - 5 does not exceed 151, then G has a I-factor. Hence every I-tough
For i = 1,2, ... ,k, let Sj denote the set of those vertices of 5 adjacent to graph of even order contains a I-factor. Enomoto, Jackson, Katernis and
one or more vertices of G;. Each set 5; is nonempty; otherwise some G; Saito [EJKSl] extended this result.
would be an odd component of G. The union of any j of the sets
51 ,52 " .• , 5k contains at least j vertices for each j with 1 ~ j ~ k; for other-
wise, there exists j (1 ~ j ~ k) such that the union T of some j sets contains Theorem 9.7
less than j vertices. This would imply, however, that kn(G - T) > In If G is a k-tough graph oforder n ~ k + I, where kn is even, then G has a k-factor.
which is impossible. Thus, we may employ Theorem 9.4 to produce a
system of distinct representatives for 51 ,52 " " , 5k • This implies that 5 By definition, every I-regular graph contains a I-factor. A 2-regular
contains vertices v1' V2, ••. , Vk! and each G; contains a vertex U; graph Gcontains a I-factor if and only if every component of G is an
(1 ~ i ~ k) such thatuiv; E E(G) for i =1,2, ... ,k. even cycle. This brings us to the 3-regular or cubic graphs. Petersen
Let W be a proper subset of V(G; - u;),1 ~i ~ k. We show that [P3] proVided a sufficient condition for a cubic graph to contain a I-factor.
kn(G i - Ui - W) ~ IWI, for suppose that ko(G; - Uj - W» IWI. Since
G; - U; has even order, kn(G; - U; - W) and IWI are of the same parity
and so kn(G; - U; - W) ~ IWI + 2. Thus, Theorem 9.8

ko(G-(5UWU{u;})) = ko(Gj-uj-W)+ko(G 5) 1 Every bridgeless cubic graph contains a I-factor.

~ j51+IWI+l
Proof
= 15UWU{uj}l·
Let G be a bridgeless cubic graph and assume that V(G) has a proper
This, however, contradicts the maximum property of 5. Therefore, subset S such that the number of odd components of G - S exceeds 151.
ko(G; - U; - W) ~ IWI as claimed, implying by the inductive hypothesis Let j = 151 and let G1 , G2 , ••. , Gk (k > j) be the odd components of G - 5.
that, for i = 1,2, ... , k, the subgraph G; - ujhas a I-factor. This I-factor, There must be at least one edge joining a vertex of G; to a vertex of S, for
together with the existence of the edges UjV; (1 ~ i ~ k), produces a 1- each i = 1,2, ... , k; for otherwise, G; is a cubic graph of odd order. On the
factor in G. 0 other hand, since G contains no bridges/there cannot be exactly one such
edge; that is, there are at least two edges joining G; and 5, for each
By Tutte's theorem, it follows, of course, that if G is a graph of order" i = 1,2, ... ,k.
n ~ 2 such that for each proper subset S of V(G), the number of odd Suppose that for some i = 1,2, ... , k, there are exactly two edges joining
components of G - S does not exceed lSI, then ,81 (G) = n12. Using a G; and S. Then there are an odd number of odd vertices in the component
266 Matchings, factors and decompositions Matchings and independence in graphs 267
Theorem 9.10
Ifall the bridges ofa connected cubic graph G lie on asingle path of G, then G has a
I-factor.

The three bridges of the cubic graph G of Figure 9.2 do not lie on a single
path of G, of course, and G does not have a I-factor. It is a direct con-
sequence of Errera's theorem that if the bridges of a cubic graph G of
order n lie on a single path, then f31 (G) = nj2. If the bridges of a cubic
graph G do not lie on a single path, then it may very well occur that
f31 (G) < nj2. But more (see [CKORl]) can be said about this.

Theorem· 9.11
If G is a connected cubic graph oforder n all ofwhose bridges lie on r edge-disjoint
paths of G, then
Figure 9.2 A cubic graph containing no I-factors.

Gi of G - S, which cannot happen. Hence, for each i = 1,2, ... ,k, there are
f31 (G) -l
~ ~ ~ j.
at least three edges joining Gi and S. Therefore, the total number of edges
joining U~=l V(G;) and S is at least 3k. However, since each of the jvertices Recall that an independent set of vertices in a graph G is one whose
of S has degree 3, the number of edges joining U~=l V(G i ) and S is at most elements are pairwise independent (nonadjacent) and that the vertex
3j. Therefore, 3j ~ 3k, which is a contradiction since 3k > 3j. Hence, no independence number f3( G) of G is the maximum cardinality among the
such setS exists. By Theorem 9.5, then, we conclude that G has a 1- independent sets of vertices in G. For example,if s ~ t, then f3(Ks,d = t and
factor. 0 f31 (Ks,t) = S.
A vertex and an edge are said to cover each other in a graph G if they ate
Indeed, Petersen [P31 showed that if a cubic graph G contains at most incident in G. A vertex cover in G is a set of vertices that covers all the edges
two bridges, then G has a I-factor (Exercise 9.5). On the other hand, this of G. An edgecover in a graph G without isolated vertices is a set of edges
result cannot be extended further since the cubic graph of Figure 9.2 has that covers all vertices of G.
three bridges but contains no I-factor. In fact, it is the unique cubic graph The minimum cardinality of a vertex cover in a graph G is called the
of minimum order with this property. vertex covering number of G and is denoted by a(G). As expected, the edge
From Petersen's work, we know that if G is a cubic graph of order n covering number a1 (G) of a graph G (without isolated vertices) is the mini-
containing at most two bridges, then f31 (G) = nj2. The following result mum cardinality of an edge cover in G. For s ~ t, we have a(Ks,d = sand
[CKLSl] provides a sharp bound on the edge independence number of a1 (Ks,t) = t. As another illustration of these four parameters, we note that
a cubic graph in terms of the number of bridges it possesses. for n ~ 2, f3(Kn) = 1, f31 (Kn) = Lnj2J, a(Kn) = n - 1 and a1 (Kn) = nj2l r
Observe that for the two graphs G of order n considered above, namely
Ks,t, with n = s + t, and Knt we have
Theorem 9.9
a(G) + f3(G) = a1(G) + f31(G) = n.
If G is a connected cubic graph oforder n containing fewer than 3(k + 1) bridges,
then These two examples serve to illustrate the next theorem, due to Gallai
[Gl].
n-2k
f31(G) ~-2-'
Theorem 9.12
Errera [E5] showed, however, that if all the bridges of a cubic graph,
regardless of how many there might be, lie on a single path, then G has a If G is a graph of order n having no isolated vertices, then
I-factor. a(G) + f3(G) = n (9.1)
268 Matchings, factors and decompositions Matchings and independence in graphs 269
and As in the proof of Theorem 9.3, we have that WI - U is matched to W2
al (G) + (31 (G) = n. (9.2) and that N(WI) = W z. Since WI - U is matched to Wz, it follows that
IWI I-IW2 1 = lUI·
Observe that C = (VI - WI) U W2 is a vertex cover in G; for otherwise
Proof there is an edge vw in G such that v E WI and w <f. W2 . Furthermore, ,
We begin with (9.1). Let U be an independent set of vertices of G with ICI = IVd -IWII + IWzl = lVII-lUI = IMI.
lUI = (3(G). Clearly, the set V(G) - U is a vertex cover in G. Therefore,
a(G) ~ n - (3(G). If, however, W is a set of a(G) vertices that covers all Therefore, a( G) ~ ICI = IMI = (31 (G) and the proof is complete. 0
edges of G, then V(G) - W is independent; thus (3(G) > n - a(G). This
proves (9.1). Next we present upper and lower bounds for the edge independence
To verify (9.2), let EI be an independent set of edges of G with number, due to Weinstein [W3].
IEII = (31 (G). Obviously, EI covers 2(31 (G) vertices of G. For each vert~x
of G not covered by EI, select an incident edge and define Ez to be the
Theorem 9.14
union of this set of edges and EI . Necessarily, E2 is an edge cover in
G so that IE 2 1>al(G). Also we note that IE I I+IE2 1=n; hence Let G be a graph of order n without isolated vertices. Then
al (G) + (31 (G) ~ n. Now suppose that E' is an edge cover in G with
IE'I = al(G). The minimality of E' implies that each component of (E') is
a tree. Select from each component of (E') one edge, denoting the resulting
set of edges by E". We observe that IE"I ~ (31 (G) and thatlE'1 + IE"I = n.
f1 + ~(G) 1~ (31 (C) ~ l~j.
Furthermore, these bounds are sharp.
These two facts imply t~at al (G) + (31 (G) > n, completing the proof of
(9.2) and the theorem. 0
Proof
If C is a vertex cover in a graph G and E is an independent set of edges,
then for each edge e of E there is a vertex Ve in C that is incident with e. It suffices to prove the theorem for connected graphs. The upper bound
Furthermore, if e, fEE, then V e 1:' vf' Thus for any independent set E of for (31 (G) is immediate and clearly sharp.
edges and any vertex cover C in G, we have ICI > lEI. This, of course, In order to {rerify the lower bound, we employ induction on the size m
implies that a(G) > (31 (G). In general, equality does not hold here. If, of a connected graph. If m ::;= 1 or m = 2, then the lower bound follows.
however, G is bipartite, then we do have a( G) = (31 (G), as was shown ~ssume that t~e lower bound holds for all connected graphs of positive
by Konig [K9]. SIze not exceedmg k, where k > 2, and let G be a connected graph of order
n having size k + 1. If G has a cycle edge e, then .

Theorem 9.13 (31 (G) > (31 (G _ e) >- n >- n

r l+ll(C-e) r l+ll(G)'
If G is a bipartite graph, then
Oth~rwise, G is a tree. If G=KI,n_I' then (31(G)=nj(l+ll(G» =1
(WhICh also .shows the sharpness of the lower bound). If G 1:' KI,n_I,
then G contams an edge e such that G - e has two nontrivial components
Proof GI and G~. Let ni denote the order of Gi , i = 1,2. Applying the inductive
hypothesIs to GI. and Gz, we obtain
Since a(G) > (31 (G), it suffices to show that a(G) ~ (31 (G). Let VI and V 2
be the partite sets of G and let M be a maximum matching in G.
Then (31 (G) = IMI. Denote by U the set of all M-vertices in VI' at
(31(G) > (31 (GI) + (31 (G2 ) > 1 +i(G1) + 1 + i(G )
2
U = 0, then the proof is, of course, complete.) Observe that >- nl nz n
IMI = IVII - IUI· Let 5 be the set of all vertices of G that are connected r 1 +~(G) + 1 + ll(G) = 1 + ll(G)' 0
to some vertex in U by an M-altemating path. Define WI = S n VI and
W2 = Sn V 2 • Combining Theorems 9.12 and 9.14, we have our next result.
270 Matchings, factors and decompositions Matchings and independence in graphs 271
(
Corollary 9.15
Let G be a graph of order n without isolated vertices. Then EXERCISES 9.1

r"2nl ~ a:](G) ~ In.b.(G) J 9.1 Show that every tree has at most one perfect matching.
1 + b.(G) .
9.2 Determine the maximum size of a graph of order n having a
Furthermore, these bounds are sharp. maximum matching of k edges if (a) n = 2k, (b) n = 2k + 2.
It is easy to see for a graph G of order n without isolated vertices that 9.3 Use Menger's theorem to prove Theorem 9.3.
1 ~ jJ(G) ~ n -1 and that these bounds are sharp. This implies that
9.4 Let G be a bipartite graph with partite sets VI and V2, where
1 ~a:(G) ~ n -1 are sharp bounds for a:(G).
IVII ~ 1V21· The deficiency def(U) of a set U ~ VI is defined as
A set S of vertices or edges in a graph G is said to be maximal with respect
max{ISI-IN(S)I}, where the maximum is taken over all nonempty
to a property P if S has property P but no proper superset of S has property
subsets 5 of U. Show that (3I (G) = min{IVII, IVII - def(Vl)}'
P; while S is minimal with respect to P if S has property P but no proper
subset of S has property P. Although isolated instances of these concepts 9.5 Prove that every cubic graph with at most two bridges contains a 1-
have been discussed earlier, we will be encountering such ideas in a more factor.
systematic manner in this chapter and the next. In particular, for certain
properties P, we will be interested in the maximum and! or minimum 9.6 (a) Let G be an odd graph and let VI U V2 be a partition of V(G),
cardinality of a maximal or minimal set with property P. where E' is the set of edges joining VI and V2 • Prove that !VIi
For example, a maximal independent set of vertices of maximum and IE'I are of the same parity.
cardinality in a graph G is called a maximum independent set of vertices. (b) Prove that every (2k + I)-regular, 2k-edge-connected graph,
The number of vertices in a maximum independent set has been called k ~ I, contains a I-factor.
the independence number of G, which we have denoted by jJ(G). We 9.7 Prove that if Gis an r-regular, (r - 2)-edge-connected graph (r ~ 3)
define the lower independence number i( G) of G as the minimum cardinality of even order containing at most r - 1 distinct edge-cuts of cardin-
of a maximal independent set of vertices of G. Thus, for Ks,I, where s < t, ality r - 2, then G has a I-factor.
there are only two maximal independent sets of vertices, namely, the
partite sets of Ks,t. Hence, (3(Ks,t) = t, while i(Ks,t) == s.. . 9.8 Show that a. graph G is bipartite if and only if jJ(H) ~ ! !V(H) I for
Likewise, a maximal matching or a maximal independent set of edges every sUbgraph H of G.
of maximum cardinality in a graph G is a maximum matching. The 9.9 Let Gbe a graph and let U ~ V(G). Use Tutte'stheorem to prove that
number of edges in a maximum matching is the edge independence
G has a matching that covers U if and only if for every proper subset
number jJ](G) of G. The minimum cardinality of a maximal matching is
S of V(G), the number of odd components of G - S containing only
the lower edge independence number of G and is denoted by i] (G). For
vertices of U does not exceed lSI.
example, for the path P6, jJ](P6 ) = 3 and i](P6 ) = 2. Two results of
[IRSl] dealing with maximal independent sets of edges are stated next 9.10 Prove Theorem 9.16.
(Exercises 9.10 and 9.11).
9.11 Prove Theorem 9.17.

Theorem 9.16 9.12 Show that Theorems 9.16 and 9.17 have no analogues to maximal
independent sets of vertices.
For every nonempty graph G,
i I (G) ~jJl(G)~ 2i](G).
9.13 Characterize those nonempty graphs with the property that every
pair of distinct maximal independent sets of vertices is disjoint.

Theorem 9.17 9.14 The matching graph M(G) of a nonempty graph G has the maximum
matchings of G as its vertices, and two vertices M 1 and M 2 of M(G)
Let G be anonempty graph. If k is an integer such that i1(G) ~ k ~ {31 {G), then G are adjacent .if M1 and M2 differ in only one edge. Show that each
contains a maximal matching with kedges. cycle en,
n = 3,4,5,6, is the matching graph of some graph.
272 Matchings, factors and decompositions Factorizations and decompositions 273
9.15 Prove or disprove: A graph G without isolated vertices has a perfect
matching if and only if al (G) = .81 (G).
9.16 Show that if G is a bipartite graph without isolated vertices, then
a1 (G) = ;3(G).

9.2 FACTORIZATIONS AND DECOMPOSITIONS

A graph G is said to be factorable into the factors GI , Gz,... , GI if these Figure 9.3 The Petersen graph: a bridgeless cubic graph that is not I-factorable.
factors are pairwise edge-disjoint and U:=I E(Gi ) = E(G). If G is factored
into G1 , Gz, ... , GI , then we represent this by G = GI ffi Gz ffi ... ffi GI ,
which is called a factorization of G. is I-factorable, and let G be a regular bipartite graph of degree r, where VI
If there exists a factorization of a graph G such that each factor is a k- and V z are the partite sets of G.
factor (for a fixed k), then G is k-factorable. If G is a k-factorable graph, then We now show that VI is nondeficient. Let 5 be a nonempty subset of VI'
necessarily G is r-regular for some integer r that is a multiple of k. The number of edges of G incident with the vertices of 5 is r151. These
If a graph G is factorable into GI , Gz ,... ,GI , where Gi = H for some edges are, of course, also incident with the vertices of N(5). Since G is
graph H and each integer i (1 ~ i ~ t), then we say that G is H-factorable r-regular, the number of edges joining 5 and N(S) cannot exceed
and that G has an isomorphic factorization into the factor H. Certainly, if a rIN(5)1. Hence, rIN(S) I ~ rl51 so that IN(5)1 ~ 151. Therefore, VI is non-
graph G is H-factorable, then the size of H divides the size of G. A graph G deficient, implying by Theorem 9.3 that VI can be matched to a subset
of order n = 2k is I-factorable if and only if G is kKz-factorable. of V z. Since G is regular of positive degree, IVII = IVzl; thus, G has a
The problem in this area that has received the most attention is the I-factor F. The removal of the edges of F from Gresults in a bipartite
determination of which graphs are I-factorable. Of course, only regular graph G' that is regular of degree r - 1. By the inductive hypothesis, G'
graphs of even order can be I-factorable. Trivially, every I-regular graph is I-factorable, implying that G is I-factorable as well. 0
is I-factorable. Since a 2-regular graph contains a I-factor if and only if
every component is an even cycle, it is precisely these 2-regular graphs The following result is part of mathematical folklore.
that are I-factorable. The situation for r-regular graphs, r ~ 3, in general,
or even only 3-regular graphs in particular, is considerably more com- Theorem 9.19
plicated. By Petersen's theorem, every bridgeless cubic graph contains a
I-factor. Consequently, every bridgeless cubic graph can be factored into The complete graph K2k is I-factorable.
a I-factor and a 2-factor. Not every bridgeless cubic graph is I-factorable,
however. Indeed, as Petersen himself observed [P4], the Petersen graph Proof
(Figure 9.3) is not I-factorable; for otherwise, it would have edge
chromatic number 3, which is not the case (Exercise 8.26). The result is obvious for k = I, so we assume that k ~ 2. Denote the vertex
We now describe two classes of I-factorable graphs. The first of these is set of K2k by {Vo, VI, ... , VZk-l} and arrange the vertices V1> V2,"" V2k-1 in
due to Konig [K8]. a regular (2k -1)-gon, placing Vo in the center. Now join every two ver-
tices by a straight line segment, producing K2k . For i = 1,2, ... ,2k - I,
define the ~-factor Fi to consist of the edge VOVi together with all those
Theorem 9.18
edges perpendicular to VOVi' Then K2k = F I ffi Fz EB··· EB F2k - I , so K2k is
Every regular bipartite graph of degree r ~ 1 is l..,factorable. I-factorable. 0

Proof The construction described in the proof of Theorem 9.19 is illustrated in

Figure 9.4 for the graph KG'
We proceed by induction on r, the result being obvious for r = 1. We return briefly to the I-factorization of K2k described in the proof of
Assume, then, that every regular bipartite graph of degree r -I, r ~ 2, Theorem 9.19. Recall that the I-factor FI consists of the edge VOVt and all
 Factorizations and decompositions 275
274 Matchings, factors and decompositions
Proof
We have already noted that if G is a 2-factorable graph, then G is regular
of positive even degree. Conversely, suppose that G is 2k-regular for some
integer k ~ 1. Assume, without loss of generality, that G is connected.

\ Hence, G is eulerian and so contains an eulerian circuit C.

Let V{G) = {VI, Vz, .•. , v n }. We define a bipartite graph H with partite
sets U = {UI, Uz,"" un} and W = {WI, Wz,···, wnh where
E(H) = {UiWj IVj immediately follows Vi on C}.

The graph His k-regular and so, by Theorem 9.18, is I-factorable. Hence,
H = FI EEl Fz EEl ••• EEl Fk is a I-factorization of H.
Corresponding to each I-factor Ft (1 ~ t ~ k) ofH is a permutation at on

Figure 9.4 A I-factorization of ~.

J the set {I,2, ... , n},defined by at(i) = jif UiWj E E(Ft ). Let (}:t be expressed
as a product of disjoint permutation cycles. There is no permutation cycle
of length 1 in this product; for if (i) were a permutation cycle, then
this would imply that at(i) = i. However, this further implies that
UiWj E E(Ft) and that ViVj E E( C); which is impossible. Also there is no
permutation cycle of length 2 in this product; for if (i j) were a permutation
edges perpendicular to VOVI, namely, VZVZk-I, V3V2k-Z,' .• , vkvk+I' If the k cycle, then at(i) = j and (}:t(j) = i. This would indicate that UjWj'
edges of FI are rotated clockwise through an angle of 211"1 (2k - 1) radians,
UjWj E E(Ft) and that Vj both immediately follows and precedes Vj on C,
then the I-factor Fz is obtained. In general, if the edges of FI are rotated
contradicting the fact that no edge is repeated on a circuit. Thus, the
clockwise through an angle of 27rjI (2k - 1) radians, where 0 ~ j ~ 2k - 2,
length of every permutation cycle in (}:t is at least 3.
then the I-factor Fj + I is produced. A factorization of a graph obtained in
Each permutation cycle in (}:t therefore gives rise to a cycle in G, and
this manner is referred to as a cyclic factorization.
the product of disjoint permutation cycles in (}:t produces a collection of
Such a factorization can be viewed in another way. Let K2k be drawn as
mutually disjoint cycles in G containing all vertices of G; that is, a 2-
described in the proof of Theorem 9.19. We now label each edge of K 2k
factor in G. Since the I-factors Ft in H are mutually edge-disjoint,
with one of the integers 0, 1, ... ,k - 1. Indeed, we will assign 2k - 1
.the resulting 2-factors in· G are mutually edge-disjoint. Hence, G is 2-
edge~ of K 2k the label i for i = 0, 1, ... ,k - 1. Every edge of the type VOVi
factorable. D
(1 ~ 1 ~ 2k -1) is labeled O. Now let C denote the cycle
v:' Vz, ••• , V2k-I, VI' For 1 ~ s < t ~ 2k - I, the edge VsVt is assigned the
By Theorem 9.20, then, there exists a factorization of every regular
dIstance label dcCv s , Vt). Observe that 1 ~ dcCv s , Vt) ~ k - 1. Thus, the
graph G of positive even degree in which every factor is a union of
2k - 1 edges of C are labeled 1; in general, then, 2k - 1 edges of K2k are
cycles. We next consider the problem of whether there exists a
labeled the integer i for 0 ~ i ~ k - 1. Observe, further, that FI contains k
facto~za?on of G such that every factor is a single cycle. A hamiltonian
edges, one of which is labeled i for 0 ~ i ~ k - 1. Moreover, when an edge
of F1 labeled i (0 ~ i ~ k -1) is rotated clockwise through an angle of
factonzation of a graph G is a factorization of G such that every factor is
a hamiltonian cycle of G. Certainly, if a graph G has a hamiltonian factor-
27rjj(2k-I) radians, 0~j~2k-2, and edge of Fj + I also labeled i is
ization, then G is a 2-connected regular graph of positive even degree. The
obtained. Hence a I-factorization of K2k is produced.
converse of this statement is not true, however, as the graph of Figure 9.5
We now tum to 2-factorable graphs. Of course, for a graph to be 4-
shows. .
factorable, it is necessary that it be 2k-regular for some integer k ~ 1.
For complete graphs, 2-factorable and hamiltonian factorable are
Petersen [P3] showed that this obvious necessary condition is sufficient
equivalent concepts.
as well.

Theorem 9.21
Theorem 9.20
A graph G is 2-factorable if and only if G is 2k-regular for some integer k ~ 1. For every positive integer k, the graph K2k + 1 is hamiltonian factorable.
276 Matchings, factors and decompositions Factorizations and decompositions 277
The factorization described in the proof of Theorem 9.21 is again a
cyclic factorization. If we place the vertex Vo in the center of the regular
2k-gon and rotate the edges of the hamiltonian cycle FI clockwise through
an angle of 21f /2k = Jr /k radians, then the hamiltonian cycle F2 is pro-
duced. Indeed, if we rotate the edges of F1 clockwise through an angle
of 1fj/k radians for any integer j with 1 :( j :( k - 1, then the hamiltonian
cycle Fj + 1 is produced and the desired hamiltonian factorization of K2k + 1
is obtained (Exercise 9.21).
Another factorization result now follows readily from Theorem 9.21.

Corollary 9.22
The complete graph K 2k can be factored into k hamiltonian paths.

Theorems 9.19 and 9.21 have an interesting consequence of a different

Figure 9.5 A 2-factorable graph that is not hamiltonian factorable. nature. If G is an r-regular graph of order n, then, of course, 0:( r :( n - 1.
On the other hand, if rand, n are odd positive integers (with
Proof o:( r :( n - 1), then there can be no r-regular graph of order n. With
this lone exception, every other type of regular graph is possible.
Since the result is clear for k = 1, we may assume that k ~ 2. Let
V(K2k + l ) = {Vo, VI, •• • , V2k}' Arrange the vertices VI' v2,' .• , v2k ina regular
2k-gon and place Vo in some convenient position. Join every two vertices Corollary 9.23
by a straight line segment, thereby producing K2k + I · We define the edge
set of FI to consist of VOVlt VOVk+ It all edges parallel to VI V2 and all edges Let rand n be integers with 0 :( r :( n -1. Then there exists an r-regular graph
parallel to v2kv2 (see F I in Figure 9.6 for the case k = 3). In general, for of order n if and only if rand n are not both odd.
i = 1,2, ... , k, we define the edge set of the factor Fi to consist of VOVi,
vOvk+i, all edges parallel to ViVi+1 and all edges parallel to Vi_IVi+l,
Proof
where the subscripts are expressed modulo 2k. Then K2k + I =
F 1 El7 F 2 El7 ••. El7 Fk, where Pi is the hamiltonian cycle It suffices to show that there exists an r-regular graph of order n if at least
one of rand n is even (and 0 :( r :( n - 1). Suppose first that n is even.
Vo, Vi, Vi+l, Vi-I, Vi+2, Vi-2,"" Vk+i-l, Vk+i+lo Vk+;, Vo· 0
Then n = 2k for some positive integer k. By Theorem 9.19, K2k can be
This result is illustrated in Figure 9.6 for K7 · factored into I-factors F1 ,F2 , .•• ,F2k - 1 • The union of r of these I-factors
produces an r-regular graph of order n.
Next, suppose that n is odd. Then r is necessarily even. The graph KI is
VI VI VI a-regular of order 1, so we may assume that n = 2k + 1 ~ 3. By Theorem
9.21, K 2k + 1 can be factored into hamiltonian cycles F 1 , F 2 , •• • , Fk • The
union of r/2 of these hamiltonian cycles gives an r-regular graph of
V6 V2 V6 V2 V6 V2
order n. 0

Vs v3 Vs v3 Vs V3 Using the construction employed in the proof of Theorem9.2I, we can

obtain another factorization result.
V4 V4 v4

F2 F3 Theorem 9.24
PI
Figure 9.6 A hamiltonian factorization of K7 . The graph K2k can be factored into k - I hamiltonian cycles and a I-factor.
278 Matchings, factors and decompositions Factorizations and decompositions 279
Very similar to the concept of factorization is decomposition. A decom- The decomposition shown in Figure 9.7 is a cyclic decomposition. In
position of a graph G is a collection {Hi} of nonempty subgraphs such that general, a cyclic decomposition of a graph G into k copies of a subgraph H
Hi = (E i) for some (nonempty) subset Ei of E(G), where {Ei} is a partition is obtained by (a) drawing G in an appropriate manner, (b) selecting a
of E(G). Thus no subgraph Hi in a decomposition of G contains .isolated suitable subgraph HI of G that is isomorphic to H, and (c) rotating the
vertices. If {Hj } is a decomposition of G, then we wnte G = vertices and edges of HI through an appropriate angle k - 1 times to
HI EE:l Hz EE:l ••• EE:l H t , as we do with factorizations, and say G is decom- produce the k copies of H in the decomposition.
posed into the subgraphs H1>Hz, ... ,Ht , where then I{Hi}1 = t. Indeed, If G is an H-decomposable graph for some graph H, then certainly H is a
if G = HI EE:l Hz EE:l ••• EE:l H t is a decomposition of a graph G of order subgraph of Gand the size of H divides the size of G. Although this last
n and we define Fi = Hi U [n - JV(H i)IJK1 for 1 ~ i ~ t, then condition is necessary, it is not sufficient. For example, the graph K I 4 is a
F EE:l Fz EE:l ••• EE:l Ft is a factorization of G. On the other hand, every factor- ~ubgraph of the graph G of Figure 9.7 and the size 4 of Kl4 divides th~ size
1
ization of a nonempty graph G also gives rise to a decomposition of G. 12 of G, but G is not KIA-decomposable (Exercise 9.26). '
Suppose that G = F1 E9 Fz EE:l ••• EE:l Fs is a factorization of a nonempty The basic problem in this context is whether, for a graph G and a
graph G, so written that F1,F2, ... ,Ft are none~pty (t~s).. ~et subgraph H of G whose size divides that of G, the graph G is H-
Hi = (E(F j )} for i = 1,2, . ; . , t. Then HI EE:l Hz E9 ... EE:l H t IS a decomposItIon decomposable. First, we consider H-decomposable graphs for graphs H
of small size. Of course, every (nonempty) graph is Kz-decomposable.
ofG.
If {Hi} is a decomposition of a graph G such thatBi = H for some graph Every component of a P3~decomposable graph must have even size. In
H for each i, then G is said to be H-decomposable. If G is an H-decomposable fact, this condition is sufficient for a graph to be P3-decomposable (see
graph, then we also write HI G and say that H divides G. Also H is said [CPSl]).
to be a divisor of G, and G is a multiple of H. For G = K2,z,z (the graph of
the octahedron) and for the graph H shown in Figure 9.7, we have that Theorem 9.25
G is H-decomposable. An H-decomposition of G is also shown in
Figure 9.7. A nontrivial connected graph G is P3-decomposable if and only if G has even size.

Proof
uo::=--------~:Dv
H:
We have already noted that if G is P 3-decomposable, then G has even size.
G: For the converse, assume that G has even size. Suppose, first, that G is
eulerian, where the edges of G are encountered in the order
e1> ez, .. · ,em' Then each of the sets {el, ez}, {e3,e4},"" {em-bern} induce
a copy of P3 , so G is P 3-decomposable. Otherwise, G has 2k odd vertices
for some k ~ 1. By Theorem 4.3, E(G) can be partitioned into subsets
E1 , Ez, ..,' , Ekt where. for each i, (E i ) is an open trail Ti of even length
w

._
connectmg odd vertIces of G. Then, as with the eulerian circuit above,
~he edges of each trail Ti can be paired off so that each pair of edges
v mduces a copy of P3 • Thus G is P3-decomposable. 0
Ol-----~: ~
u
U l' The only other graph of size 2 without isolated vertices is 2Kz. The class
of 2Kz-~ecomposable graphs was discovered by Y. Caro (unpublished)
and RUlZ [R12]. Since the proof of the next result is quite lengthy, we
omit it.

w w Theorem 9.26
HI H2 A nontrivial connected graph is 2Krdecomposable if and only if G has even size
Figure 9.7 An H-decomposable graph. m, Ll(G) ~ ~m, and G i K3 U Kz. .
280 Matchings, factors and decompositions Factorizations and decompositions 281
Most of the interest in K3 -decompositions has involved complete 3 4
graphs. A K3-decomposable complete graph is called a Steiner triple

" ,.', ,D'

system. Kirkman [K5] characterized Steiner triple systems.

Theorem 9.27
The complete graph Kn is K3 -decomposable if and only if n is odd and 3 I G)·

For Kn to be Kp+ 1-decomposable, the conditions pi (n -1) and

n 2
Figure 9.8 Graceful graphs.
rsJ
152
4 1
43

(P +1) I C) are certainly necessary. These conditions are not sufficient in result also appears in Fink [F4], where specific H-decomposable (not
ge~eral,2 however. For n = p2 + P + 1, Ryser [RI3] showed that Kn is necessarily complete) graphs are described. We give a proof of this
K 1-decomposable if and only if there exists a projective plane of result, but prior to doing this, it is convenient to introduce some addi-
p+ . tional terminology, which will be explored further in section 9.3.
order Pi and in order for a projective plane of order p to eXIst, p must
satisfy the Bruck-Ryser conditions [BR2] that p == 0 (mod 4) or A graph G of size m is called graceful if it is possible to label the vertices
p == 1 (mod4), and p = ~ + y2 for some integers x and y. The smallest of G with distinct elements from the set {O, 1, ... m} in such a way that the
1

value of p for which the existence of a projective plane of order p is induced edge labeling, which prescribes the integer Ii - jl to the edge
joining vertices labeled i and j, assigns the labels 1,2, ... , m to the edges
unknown is p = 10.
Whenever Kn is Kp+rdecomposable, we have an example of a combi~ of G. Such a labeling is called a graceful labeling. Thus, a graceful graph is a
natorial structure referred to as a balanced incomplete block design. Thus graph that admits a graceful labeling.
graph decompositions may be viewed as generalized block designs. The graphs K3 , K4, ~ - e and C4 are graceful as is illustrated in Figure
There is another important interpretation of a special type of decompo- 9.8. Here the vertex labels are placed within the vertices and the induced
sition. In particular, the minimum number of I-regular subgraphs into edge labels are placed near the relevant edges.
which a nonempty graph G can be decomposed is the edge chromatic Not every graph is graceful, however. For example, the graphs K s, Cs
number Xl(G) of G. By Corollary 8.19, the edge chromatic number of and K 1 + 2K3 are not graceful.
an r-regular graph G (r ~ 1) is r or r + 1. If Xl(G) = r, then each edge The gracefulness grac(G) of a graph G with V(G) = {Vl,V2,""V n } and
color class in a Xl (G)-edge coloring of G induces a I-factor of G. Thus, without isolated vertices is the smallest positive integer k for which it is
an r-regular graph has edge chromatic number r if and only if it is 1- possible to label the vertices of G with distinct elements from the set
factorable.
{a, 1, ... , k} in such a way that distinct edges receive distinct labels. Such
The vast majority of factorization and decomposition results deal with vertex labelings always exist, one of which is to label Vi by i -1. Hence for
factoring or decomposing complete graphs into a specific graph or every (n,m) graph G without isolated vertices, m ~ grac(G) ~ 2n - 1 • If G
graphs. R. M. Wilson [W8] proved that for every graph H withoutisolated is a graph of size m with grac(G) = m, then G is graceful. Thus the
vertices, there exist infinitely many positive integers n such that Kn is H- gracefulness of a graph G is a measure of how close G is to being
decomposable. graceful. By definition, it is possible to label the vertices of a graph G
with distinct elements of the set {O, 1, ... ,grac(G)} so that the edges of
G receive distinct labels. Of course, some vertex of G must be labeled
Theorem 9.28 grac(G), but it is not known whether an edge of G must then be labeled
For every graph H without isolated vertices and having size m, there exists grac(G).
a positive integer N such that if (i) n ~ N, (ii) m I mand (iii) d I (n -1), All graphs of order at most 4 are graceful. There are exactly three non-
where graceful graphs of order 5. In each case the gracefulness is one more than
t~e size. These three graphs with an appropriate labeling are shown in
d = gcd{degv Iv E V(H)}, FIgure 9.9. ~

then Kn is H-decomposable. We. now return to our discussion of decompositions and give a con-
structive proof that for every graph H without isolated vertices there
As an immediate consequence of Theorem 9.28; there exist regular exists a regular H-decomposable graph. The proof is due to Ficlc and
decomposable graphs for every graph H without isolated vertices. This Ruiz [FRl] and was inspired by a proof technique of Rosa [RIO].
282 Matchings, factors and decompositions Factorizations and decompositions 283

~
4

2 3 4

..
Figure 9.9 The three graphs of order 5 that are not graceful.
I
I
0
.
...
d
Theorem 9.29 Vs v4

For every graph H without isolated vertices, there exists a regular H-decomposable Figure 9.10 Construction of a regular H-decomposable graph.
graph. G = KZm +1 and, consequently, KZm + 1 is H-decomposable. Indeed, then,
for any graceful graph H of size m, the complete graph Kzm + I is H-
Proof decomposable; in fact, there is a cyclic decomposition of KZm + I into H.
Let H be an (n, m) graph without isolated vertices and suppose that This observation is due to Rosa [RlOl. Because of its importance, we
grac(H) = k. Hence there exists a labeling ¢: V(H) -> {O, 1, ... ,k} of give a direct proof due to Rosa of this result.
vertices of H so that distinct edges of H are labeled differently and
max{¢(x) I x E V(H)} = k. Theorem 9.30
We now construct a regular H-decomposable graph G of order If His a graceful graph of size m, then KZm + I is H-decomposable. Indeed, KZm + I
p =2k+ 1. Let V(G) = {vo,v}, ... ,vp-d and arrange these vertices cycli- can be cyclically decomposed into copies of H.
cally in clockwise order about a regular p-gon.Next we define a graph HI
by
Proof
Since H is graceful, there is a graceful labeling of H, that is, the vertices of
and H can be labeled from a subset of {O, 1, ... ,m} so that the induced edge
E(HI ) = {v.p(x)v.p(y) I xy E E(H)}. labels are 1,2, ... ,m. Let V(KZm + I ) = {VO,VI, ... , vzm } where the vertices
of KZ~+I are arranged cyclically in a regular (2m + 1)-gon, denoting the
For i = 2,3, ... , p, define Hi by cyclically rotating HI through a clockwise resultmg (2m + 1)-cycle by C. A vertex labeled i (0 ,,;; i ,,;; m) in H is placed
angle of 21r(i - 1)!p radians. Therefore, for 1 ,,;; i ,,;; p, at Vi in K2m + I and this is done for each vertex of H. Every edge of His
V(Hi ) = {V,p(~)+i-I I x E V(H)} dr~wn as a straight line segment in KZm + 1t denoting the resulting copy of
HmKzm + I as HI' Hence V(HI ) S;;; {VQ,VI""'Vm },
and Each edge VsV t of KZm +I (0 ,,;; s, t ~ 2m) is labeled dc(v s,Vt), where then
E(Hi ) = {V,p(x)+i-IVq,(y)+i-I I xy E E(H)}. 1 ,,;; ddv;, Vt) ,,;; m.. Cons:quently, KZnH I contains exactly 2m + 1 edges
~abele~ I for each 1(1 ,,;; I ,,;; m) and HI contains exactly one edge labeled
The definition of G is completed by defining E(G) = Uf=1 E(Hi). (See
Figure 9.10 for a given graph H, a possible labeling of the vertices of H,
I(! , ; I ,,;; m). Whenever an edge of HI is rotated through an angle (clock-
WIse, .say) o~21rk/(2m + 1) radians, where 1";; k,,;; m, an edge of the same
the induced edge labels of H, the vertices of G, and the subgraphs HI and
label IS .obtamed. Denote the subgraph obtained by rotating HI through a
Hz of G, where the edges of Hz are drawn with dashed lines.)
clockWise angle of 21rk/(2m + 1) radiaI1s by Hk +I . Then Hk +I = H and a
The graph G is therefore decomposable into the graphs HI, Hz, ... , H p,
cyclic decomposition of KZm + 1 into 2m + 1 copies of H results. 0
each of which is isomorphic to H, and G is 2m-regular. 0
As an illustration of Theorem 9.30, we consider the graceful graph
If, in the proof of Theorem 9.29, the graph H is graceful, then
grac(H) = m and G is a 2m-regular graph of order 2m + 1, that is,
H = P3' A graceful labeling of H is shown in Figure 9.11 as well as the
resulting cyclic H-decomposition of Ks .
 Matchings, factors and decompositions Factorizations and decompositions 285
284
Furthermore, the following conjecture concerning decompositions of
Vo Vo complete graphs into trees has been made by Ringel [R6].
0

H: V40

0
~" ',0/1"V2 V3
2

v2
1
Ringel's Conjecture
For every tree T of size m, the complete graph KZm + 1 is T-decomposable.

Of course, if Kotzig's conjecture is true, so is Ringel's. Indeed, the truth

V3
HI H2 of Kotzig's conjecture implies the truth of the following conjecture, due
jointly to G. Ringel and A. Kotzig.

"V ,.:<
Vo Vo Vo

,.~
The Ringel-Kotzig Conjecture
For every tree T of size m, KZm +1 can be cyclically decomposed into T.
0'1 0 0'1
We close this section by presenting another result involving cyclic
o decompositions. A linear forest is a forest of which each component is a
V3 1 V2 v3 v2 v3 v2 path. Since every I-factor is a linear forest, the follOWing result, due to
H H4 HS Ruiz [RUl, is a generalization of Theorem 9.19.
3
Figure 9.11 A cyclic decomposition of Ks into the graceful graph P3'
Theorem 9.31
Although Kzm +1 has a cyclic decomposition into every graceful graph H
of size m, it is not necessary for H to be graceful in order for Kzm + 1 t~ have If F is a linear forest of size k hoving no isolated vertices, then K2k is F-
a cyclic H-decomposition. For example, we have seen that C5 . 1.S n~t decomposable.
graceful; yet Ku has cyclic C5-decomposition. Such a decomposItIon IS
depicted in Figure 9.12. . . Proof
It has been conjectured by Kotzig (see Rosa [RIO]) that every nontnvlal
Since the result is obvious for k = 1, we assume that k ~ 2. Let the vertex
tree is graceful. set of K 2k be denoted by {VO,Vl,VZ, ••• ,V2k-l}' Arrange the vertices
Vb vz,·.·, V2k-l cyclically in clockwise order about a regular (2k - 1)-
Kotzig's Conjecture gon, calling the resulting cycle C, and place Vo in the center of the
Every nontrivial tree is graceful. (2k- I)-gon. Join every two vertices by a straight line segment to obtain
the edges of K2k • We label each edge that joins Vo to a vertex of C by O.
There are 2k - 1 such edges. Every other edge of K2k joins two vertices
of C. If uv is an edge joining two vertices of C, then label uv by i if
dcCu, v) = i. Note that 1 .:;; i .:;; k -1 and that for each i = I,2, ... ,k -1,
the graph K2k contains 2k - 1 edges labeled i.
We now describe two paths P andQ of length kin K2k • If k is even, then
P: vo, VI' V2k-I, V2, V2k-2, V3, ... , Vk/Z, V3k/2
VgO
and

v,o Q: Vo, Vb Vk+ 1> Vk_I' Vk+Z, Vk_Z,·· ., V(k+Z)/Z, V3k/Z;

o while if k is odd, then

v6 Vs
Figure 9.12 A cyclic decomposition of Kll into the nongraceful graph Cs·
286 Matchings, factors and decompositions Factorizations and decompositions 287
and (b) Show that every bipartite graph G is of class one, that is,
X1(G) = Li(G).
Q: Vo, Vb Vk+l, Vk-l> Vk+Z, Vk-Z,"" V(3k-l)/Z, v(k+l)/Z'
9.18 Give an example of a connected graph G of composite size having
Observe that, in either case, for i = 1,2, ... ,k, the ith edge of P and the ith the property that whenever F is a factor of G and the size of F divides
edge of Q are labeled i. the size of G, then G is F-factorable.
Assume that the linear forest 9.19 (a) Prove that Q" is I-factorable for all n ~ 1.
F = Pk, + j U Pkz + j u· .. U Pk, + 1, (b) Prove that Qn is k-factorable if and only if kin.
where then 2::=1 kj = k. We define a sUbgraph H of K Zk as follows. The 9.20 Use the proof of Theorem 9.20 to give a 2-factorization of the graph
edge set E of H consists of the first k1 edges of P, edges kj + 1 through of the octahedron (namely Kz,z,z).
kj + kz of Q, edges k1 + kz + 1 through k1 + kz + k3 of P, and so on until 9.21 Use the proof of Theorem 9.21 to produce a hamiltonian factoriza-
finally the last kt edges of Q if t is even or the last kt edges of P if t. is tion of~.
odd. Define H = (E). Note that H = F and that H contains exactly one
9.22 Let k be a nonnegative even integer and n ~ 5 an odd integer with
edge labeled i for each i = 0, 1, ... ,k 1.
k :( n - 3. Prove that there exists a graph G of order n with degree
. Now for j = 1,2, ... ,2k - I, define Hj to be the subgraph of K2k obtained
set {k, k + 2}.
by revolving Habout the (2k - 1)-gon in a clockwise angle of
27r(j - 1)/ (2k - 1) radians. Observe that for each i = 0, 1, ... ,k - 1 and 9.23 Prove Corollary 9.22.
each j = 1,2, ... ,2k - I, the subgraph Hj contains exactly one edge labeled 9.24 Prove that K2k+ 1 cannot be factored into hamiltonian paths.
i. Since Hj = F for each j = 1,2, ... , 2k - 1 and K2k is decomposed into the
subgraphs H1 ,Hz, ... ,H2k - lt it follows that K 2k is F-decomposable. 0 9.25 Give a constructive proof of Theorem 9.24.
9.26 Show that the graph of the octahedron is not K1,4-decomposable.
9.27 (a) Use the fact that K3 is graceful to find a K3-decomposition of K7 •
(b) Find a noncomplete regular K3-decomposable graph.
9.28 Find an F-decomposition of K12 where F = 2Pz U 2P3 •
9.29 Find a P 6 -decomposition of KlO •
9.30 For each integer k ~ i, show that
(a) K2k + 1 is Kl,k-decomposable.
(b) K2k is K1,rdecomposable.
VsO ~V5 9.31 Use the drawing of the Petersen graph shown below to find cyclic
v7 v6 decompositions into Flt Fz and F3 .
Figure 9.13 A step in the construction of an F-decomposition of K12 for
F = P2 UP3 uP4 •

The preceding theorem and its proof are illustrated in Figure 9.13 for
2k = 12 and F = P z U P3 U P4 • The labeling of the vertices of K12 is shown

~:~
along with the subgraph H (or H j ).

EXERCISES 9.2

9.17 (a) Show that every bipartite graph G is a subgraph of a

regular bipartite graph.
~(G)­ F,:X
288 Matchings, factors and decompositions Labelings of graphs 289
9.32 Find all graphs F of size 3 that are subgraphs of the Petersen graph
P for which P is F-decomposable. (Hint: Use the drawing of the

~
Petersen graph shown in Exercise 9.31.)

9.3 LABELINGS OF GRAPHS o 1 4 6 014 7 9

In the previous section we discussed gracefullabelings of graphs for the m=6 m=9
purpose of describing cyclic H-decompositions of certain complete Figure 9.14 The Ruler Problem for 6 and 9 unit rulers.
graphs. In this section we discuss graceful labelings in more detail as
well as describe two other well-known labelings of graphs. A necessary condition for an eulerian graph to be graceful was discov-
Recall that a graceful labeling of a graph G of siz~ m is an assignment of ered by Rosa [RIO].
distinct elements of the set {O, 1, ... , m} to the vertices of G so that the
edge labeling, which prescribes Ii - jj to the edge joining vertices labeled Theorem 9.33
i and j, assigns the labels 1, 2, ... , m to the edges of G. A graph possessing a IfG is agraceful eulerian graph ofsize m, then m == 0 (mod 4) orm == 3 (mod 4).
graceful labeling is a graceful graph.
The topic of graceful labelings of graphs has a distinctive number
theoretic flavor to it. Indeed, in number theory, a restricted difference Proof
basis (with respect to a positive integer m) is a set {al,aZ, ... ,an } ~ Let C ·. v0, v1,·'·' V m- 1 , V m = 0 be an eulerian circuit of G, and let a graceful
{O, 1, ... , m} such that every integer k with 1 ~ k ~ m can be represented labeling of G be given that assigns the integer ai (0 ~ ai ~ m) to Vi for
in the form k = aj - ai' Hence a graph G of size m has a graceful labeling if o ~ i ~ m, where, of course, ai = aj if Vi = Vj' Thus the label of the edge
the vertices of G can be labeled with the elements of a restricted difference Vi-lVi is lai - ai-II. Observe that
basis in such a way that for each integer k with 1 ~ k ~ m, there is a unique
la; - ai-II == (a; - ai-I) (mod2)
pair of adjacent vertices, labeled ai andaj say, so that k = aj - ai' A related
problem is the Ruler Problem: For a given positive integer m, construct a for 1 ~ i ~ m. Thus the sum of the labels of the edges of G is
ruler m units in length so that all integral distances from 1 to m can be m m
measured and the ruler is marked at a minimum number of places. L ja; - ai-II == L (aj - ai-I) == 0 (mod 2),
Rulers for m = 6 and m = 9 are shown in Figure 9.14. i=1 i=1
Although there are no general sufficient conditions for a graph to be that is, the sum of the edge labels of G is even. However, the sum of the
graceful, there are necessary conditions. edge labels is 2:1=1 i = m(m + 1)/2; so m(m + 1)/2 is even. Consequently,
41 m(m + I), which implies that 41 m or 41 m + 1 so that m == 0 (mod 4) or
Theorem 9.32 m == 3 (mod 4). 0
If G is a graceful graph of size m, then there exists a partition of V(G) into We now determine which graphs in some well-known classes of graphs
two subsets Ve and Va such that the number of edges joining V e and Va is are graceful. Rosa [RIO] determined the graceful cycles.
rm/2l
Theorem 9.34 .
Proof
" The cycle Cn is graceful if and only if n == 0 (mod 4) or n == 3 (mod 4).
Let a graceful labeling of G be given. Denote the set of vertices labeled
with an even integer by V, a,nd the set of vertices labeled with an odd Proof
integer by Va' All edges labeled with an odd integer must then join a
vertex of Ve and a vertex of Va' Since there are rm/21 such edges, the Since C is an eulerian graph, it follows by Theorem 9.33 that if
result follows. 0 n == 1 (rr:od4) or n ~ 2 (mod 4), then en is not graceful; so it remains
290 Matchings, factors and decompositions Labelings of graphs 291

Figure 9.16 A graceful labeling of Cll .

Figure 9.15 A graceful labeling of C 12 ·

°
onlytoshowthatifn == (mod 4) orn == 3 (mod 4), then Cn is graceful. Let
Cn: VI' vz, ... , V m VI' Assume firstthat n == 0 (mod 4). We assign Vi the label
Theorem 9.35
The complete graph Kn (n ~ 2) is graceful if and only if n ~ 4.
ai, where

(i -1)/2 if i is odd
Proof
ai = n + 1 - il2 if i is even and i ~ nl2
{ We have already observed that Kn is graceful if 2 ~ n ~ 4. Assume then
n - i/2 if i is even and i> n12.
that n ~ 5 and suppose, to the contrary, that K n is graceful. Hence there
It remains to observe that this labeling is graceful. This labeling is exists a graceful labeling of the vertices of Kn from an n-element subset of
illustrated in Figure 9.15 for n = 12. {O, 1, ... , m}, where m = G). .
Next, assume that n == 3 (mod 4). In this case we assign Vi the label bi, W~ have already seen that every graceful labeling of a graph of size m
where requITes oand m to be vertex labels. Since some edge of K n must be labeled
m.-l, some vertex of ~nmust be labeled 1 or m -1. We may assume,
n + 1 - i/2 if i is even wIthout loss of generality, that a vertex of Kn is labeled 1; otherwise, we
bi = (i - 1) 12 if i is odd and i ~ (n - 1) 12 may use the complementary labeling.
{
(i + 1)/2 if i is odd and i> (n -1)/2. To produce an edge labeled m - 2, we must have adjacent vertices
labeled 0, m - 2 or 1, m - 1 or 2, m. If a vertex is labeled 2 or m - 1, then
This is a graceful labeling of Cn' An illustration is given in Figure 9.16 for we have two edges labeled 1, which is impossible. Thus, some vertex of Kn
n = 11. 0 must be labeled m - 2.
Since we now have vertices labeled 0, I, m - 2 and m, we have edges
If G is a graceful graph of order n and size m, then, of course, the labeled 1,2, m - 3, m - 2, m - 1 and m. To have an edge labeled m - 4, we
vertices of G can be labeled with the elements of a set {al' az, ... ,an} <;;; must have a vertex labeled 4 for all other choices result in two edges with
{O, 1, ... ,m} so that the induced edge labels are precisely 1,2, ... , m. the same label.
This means one vertex-in some pairs of adjacent vertices is labeled 0 Now we have vertices labeled 0, I, 4, m- 2 and m, which results in
and the other vertex in the pair is labeled m. Also, if we were to replace ~~ges l~beled 1,2,3,4, m - 6, m - 4, m - 3, m - 2, m - 1 and m. However,
each vertex label ai by m ~ ai, then we have a new graceful labeling, called It IS qUIckly se:n that there is no vertex label that will produce the edge
the complementary labeling. label m - 5 Without also producing a duplicate edge label. Hence no
We saw in Figure 9.8 that the complete graphs K3 and K4 are graceful. If graceful labeling of Kn exists. 0
is very easy to show that K2 is graceful. The folloWing result of Golomb
[G5) shows that there are no other graceful complete graphs. Theorem 9.35 adds credence to the'following conjecture [CHOn
292 Matchings, factors and decompositions Labelings of graphs 293

Ks, t:

Figure 9.17 A graceful labeling of Ks,t.

Conjecture
Graceful graphs with arbitrarily large chromatic numbers do not exist.
Figure 9.18 Gracefullabelings of paths.
Unlike the classes of graphs we have considered, every complete
bipartite graph is graceful. Other familiar classes of graceful trees include stars, double stars and
caterpillars (Exercise 9.35). Recall that a caterpillar is a tree the removal
of whose end-vertices produces a path. A lobster is a tree the removal of
Theorem 9.36 whose end-vertices produces a caterpillar. It is not known whether every
Every complete bipartite graph is graceful. lobster is graceful but of course, it is conjectured that this is the case.
We now consider a graph labeling that is similar in nature to graceful
labeling. A connected (n, m) graph G with m ~ n is harmonious if there
Proof exists a labeling ¢: V(G) -4 Zm of the vertices of G with distinct elements
Let Ks,t have partite sets V r and V z, where 1V11 = s and IV~I = t. Label the 0,1, ... ,m -1 of Zm such that each edge uv of G is labeled ¢(u) + ¢(v)
vertices of V 1 with 0,1, ... ,s -1 and label the vertices of V 2 by (addition in Zm) and the resulting edge labels are distinct. Such a labeling
S, 25, ... , (t -l)s, ts (see Figure 9.17). This is a graceful labeling. 0 is called a harmonious labeling. If G is a tree (so that m = n - 1) exactly two
verticesareJabeled the same; otherwise, the definition is the same. Four
There is no result on graceful graphs as well-known as Kotzig's conjec- harmonious graphs of order 5 (with harmonious labelings) are shown in
ture, which we recall. Figure 9.19.
Some examples of graphs that are not harmonious are shown in Figure
9.20.
Kotzig's Conjecture
We have now seen that C5 is harmonious but C4 is not. This serves as an
Every nontrivial tree is graceful. illustration of a theorem of Graham and Sloane [GS21.

Many classes of trees have been shown to be graceful. One of the most Theorem 9.38
familiar of these is the class of paths.
The cycle Cn is harmonious if and only if n is odd.
Theorem 9.37
Every nontrivial path is graceful.

Proof
Let P: vo, Vj, ••• ,Vm be a path of size m. For i even, assign Vj the label it'J.,i'~.
If i is odd, then Vi is labeled m - (i - 1) /2. It remains only to observe that
this labeling is graceful (see Figure 9.18 for m = 5 and m = 8). 0 Figure 9.19 Harmonious graphs.
294 Matchings, factors and decompositions Labelings of graphs 295
labeling of Ks,t. Suppose that this labeling assigns the integers aI, az,··· ,as

D<l>@
to the vertices of VI and bl,bz, ... ,bt to the vertices of V z. Thus,
A={al,aZ, ... ,as} and B={bl,bz, ... ,bt } are disjoint subsets of
{O,l, ... ,st -1} and
{ai + bj 11 ~ i ~ sand 1 ~ j ~ t} = {O, 1, ... , st - 1}.
Figure 9.20 Three graphs that are not harmonious.
Since for (i,j) =1= (k, t), we have ai + bj =1= ak + b(, it follows that
ai - b( i= ak - bj or, equivalently,
Proof
I{ai - bj 11 ~ i ~ sand 1 ~ j ~ t}1 = st.
Assume first that n is odd and let C: Vo, VI, ... ,Vn-1, Vo be a cycle of length
n. The labeling that assigns Vi (0 ~ i ~ n - 1) the label i is harmonious and Hence, for some i (1 ~ i ~ s) and j (1 ~ j ~ t), it follows that ai - bj = 0; so
hence Cn is harmonious if n is odd. (This labeling is illustrated for Cs in ai = bj , which contradicts the fact that A and B are disjoint. 0
Figure 9.19.)
Assume now that n = 2k ~ 4 is even and suppose, to the contrary, that We saw in Theorem 9.35 that the complete graph Kn (n ~ 2) is graceful
Cn ishatmonious. Let Cn: Vo, VI,"" Vn-l, Vo. Suppose thatthe labeling that if and only if n ~ 4. Harmonious complete graphs are characterized in
assigns Vi the label ai is harmonious. Consequently, the integersao, exactly the same way. Since the proof of this result, also due to Graham
all'" ,an-l are distinct and, in fact, {aO,al"" ,an-d = {O, 1, ... , n - 1}. and Sloane [GS2], requires considerable reliance on number theoretic
Therefore, the edge labels are ao + aI, al + az, , an -1 + ao and, further- results, we omit its proof.
more, {an + aI, al + az, ... , an-l + ao} = {O, 1, , n - l}. Let S = 2.:7';;;01 ai'
The sum ofthe edge labels of Cn is Theorem 9.40
(ao +ad + (al +az) + '" + (an_I +ao) The complete graph Kn (n ~ 2) is harmonious if and only if n ~ 4.
n-l
== L:>
i=O
== ao +al+'" +an-l (modn). We now turn our attention to trees. First, we show that every nontrivial
path is harmonious. In the proof it is convenient to label some vertices by
Thus, 2S == n(n -1)/2 == S (mod n); so S == 0 (mod n) and S == ken -1} == -n + a rather than a, where 0 ~ a ~ n - 1, which, of course, are equivalent
k (mod n). Hence k == 0 (mod 2k), which is impossible. 0 in Z~.

Although there is some similarity between the results for graceful Theorem 9.41
cycles and harmonious cycles, there is no such similarity for complete
bipartite graphs. The following result is due to Graham and Sloane [GS2]. Every nontrivial path is harmonious.

Theorem 9.39 Proof

The complete bipartite graph Ks,t is harmonious if and only if s = 1 or t = 1. Let Pn:VI, vz, ... ,Vn, where n ~ 2. If n is even, write n = 2k + 2; while if n
is odd, write n = 2k + 3. For an integer t, label the vertex Vi with ai' where
Proof a.= {-t+(i-1}/2 ifiisodd
The labeling that assigns the central vertex of KI,t the label 0 and assigns
I t + (i - 2}/2 if i is even.
the end-vertices the labels 0,1, ... , t -1 is harmonious. Consequently, Regardless of the value of t, the edges of Pn are labeled 0,1, ... , n -1.
every star is harmonious. Now, if k is even, then let t = k/2i- while if k is odd, then let
It remains to show that no other complete bipartite graph is harmo- t = (k + 1)/2. In either case, this is a harmonious labeling of P nt where if
nious. Suppose, to the contrary, that some complete bipartite graph Ks,t, k is even, then t is the repeated label of Pn; while if k is odd, then -t is the
where s, t ~ 2, is harmonious. Let the partite sets of Ks,f be VI and V 2, repeated label. Thus P n is harmonious. (Harmonious labelings of P,,,
where IVII=s and 1V21 = t. By assumption, there is a harmonious 5 ~ n ~ 8, are shown in Figure 9.21.) 0
296 Matchings, factors and decompositions Labelings of graphs 297

banj{P3): 1 2 2 ba¥K3): 2

130202
ban(P3} = 1 ban(K3) =2

Figure 9.21 Harmonious labelings of P/1' 5 ~ n ~ 8.

As with gracefullabelings, many classes of trees have been shown to be

harmonious, but whether all trees are harmonious is not known.

Graham-Sloane Conjecture
Every nontrivial tree is harmonious.
banj{C4):
042433 4
3
ban(C4) =2
3 2

Figure 9.23 The bandwidths of P3 , K 3 and C4 •

We have seen that the only graceful and harmonious complete graphs
have order at most 4. Thus, the maximum size of a graceful or harmoniolls
Therefore, for every graph G of order n, it follows that 1 :::;; ban(G) ::;;
graph of order n ~ 5 is less than G). For 5:::;; n :::;; 10, the maximum sizes of
n -1. A numberingf of a graph G for which banf(G) = ban(G) is called
a graceful graph of order n and a harmonious graph of order n are shown
a bandwidth labeling of G. All numberings of the graphs P3 , K3 and C4 are
in Figure 9.22. The similarities are obvious.
shown in Figure 9.23 together with their bandwidths. Consequently,
Another problem concerns labeling the vertices of a graph in terms of
ban(P3 ) = 1, ban(K3 ) = 2 and ban(C4 ) = 2.
its order rather than its size. A numbering f of a graph G of order n is a
These examples serve to illustrate parts of the following result.
labeling that assigns distinct elements of the set {I, 2, ... , n} to the verti.ces
of G, where each edge uv of G is labeled If(u) - f(v)J. The bandwldth
banf (G) of a numbering f: V( G) ..... {I, 2, ... ,n} of G is defined by Theorem 9.42
banf(G) = max{lf(u) - f(v)11 uv E E(G)}, (i) For n ~ 2, ban(Pn) = 1.
(iO For n ~ 2, ban(Kn ) = n - 1.
that is, banf (G) is the maximum edge label of G, and the bandwidth ban(G)
(iii) For n ~ 3, ban(Cn ) = 2.
of a graph G itself is
(iv) For 1 :::; s :::; t, ban(Ks,t) = ftj21 +s - 1.
ban(G) = min{banf(G) If is a numbering of G}.
Proof
n Maximum size of a graceful Maximum size of a
graph of order n harmonious graph of order n (i) Let P/1: Vb V2,'··' V n • If the labeling f assigns i to Vi for i = 1,2, ... , n,
then banf(P n ) = 1. Consequently, ban(Pn ) = 1.
5 9 9 (ii) There is only one numbering f of Kn and the label of the edge
6 13 13 joining the vertices labeled 1 and n is n - 1. Thus, banf(Kn ) =
7 17 17
ban(Kn ) = n - 1.
8 24 23
9 30 29 (iii) First, we show that ban(Cn ) ~ 2. Ifban(Cn ) = 1, then there exists a
10 36 36 numbering of Cn in which every edge label is 1. However, for this to occur,
every edge must join vertices labeled i and i + 1 for 1 :::; i :::; n - 1. But this
Figure 9.22 The maximum size of a graceful or harmonious graph of would imply that Cn has at most n - 1 edges, which is not the case. Hence,
order n (5 ::;; n ::;; 10). ban(Cn ) ~ 2, as claimed.
298 Matchings, factors and decompositions Labelings of graphs 299
VI There is a possibly surprising connection between the bandwidth of a
graph G of order n and powers of the path P Il .

Theorem 9.43
A graph G of order n has bandwidth k (1 ~ k ~ n - 1) if and only if k is the
smallest positive integer for which G is a subgraph of P~.

Proof
Suppose, first, that ban(G)=k. Let V(G) = {VI,VZ,''''v,,}, We may
Figure 9.24 Bandwidth labelings of C6 and C7 • assume, without loss of generality, that there exists a bandwidth labeling
of G that assigns the label i to Vi (1 ~ i ~ n). Since ban(G) = k, every two
To show that ban(C,,) = 2, it suffices to show the existence of a vertices Vi and Vj for which Ii - jl > k are not adjacent. Let P 11 denote the
numbering f of CIl for which banf(CIl) = 2. Suppose first that n = 2k is path VI, Vz,· .. 'VI1' Hence every edge of G joins vertices Vi and Vj' with
even. Then the labeling f such that Ii - jl ~ k, that is, G is a subgraph of P~.
2i-1 ifI~i~k
If G were a subgraph of p~-I, then every edge of G joins vertices Vi and
Vj with Ii - jl ~ k - 1. However, then, the labeling f that assigns i to Vi for
f(Vi) = { 2n + 2 - 2i if k + 1 ~ i ~ 2k = n
1 ~ i ~ n has banf(G) ~ k - I, which produces a contradiction.
has the property that banr(CIl) = 2 (see Figure 9.24 for n = 6). Next The converse is now immediate. 0
suppose that n = 2k + 1 is odd. Then the labeling f for which
Several bounds for the bandwidth of a graph have been found in terms
2i-I ifI~i~k+1 of other parameters defined on graphs. We present some of these now.
f(Vi)= { 2n+2-2i ifk+2~i~2k+1=n The first three of these and Theorem 9.48 are due to Dewdney (see
Chvatalovcl, Dewdney, Gibbs and Korfhage [CDGKl]).
has banf(CIl) = 2 (see Figure 9.24 for n = 7). Thus ban(CIl ) = 2.
(iv) Let VI and V z be the partite sets of Ks,I' where IVII = s ~ t = IVzl; Theorem 9.44
and let f be a bandwidth labeling of Ks,t. Certainly, the vertices labeled 1
and s + t must be in the same partite set since f minimizes the greatest For every graph G,
edge label. Also, vertices labeled 1 and 5 + t - I, as well as vertices labeled
2 and s + t, must be in the same partite set. Continuing in this manner, we
ban(G) ~ r~~G) 1.
see that the vertices of V z must be labeled with the Lt/2J smallest (or
largest) integers of the set S = {I, 2, ... , s + t} and the rt/21 largest 50r Proof
smallest, respectively) integers of S. Thus, we may assume that Vz consIsts
Let a bandwidth labeling of G be given and let v be a vertex of G for which
of vertices labeled 1,2, ... , Lt/2J and 5 + It/2J + 1, S + Lt/2J + 2, ... ,5 + t.
Consequently, ban(Ks,t) = bani (Ks,t) = rt/21 + S - 1. (A bandwidth label-
degv = ~(G) = k. Furthermore, let N(v) = {VI, vz, ... ,Vk}' Suppose that
the bandwidth labeling assigns ai to Vi for 1 ~ i ~ k and ao to v. We may
ing of KS,9 is given in Figure 9.25. Only VI and V z are shown.) 0
assume that al < az < ... < ak' Thus max{lal - aol, lak - aol} ~ k/2. So
ban(G) ~ r~(G)/2l 0

CD CD ® The next two lower bounds for the bandwidth of a graph follow from
Theorem 9.43.
KS,9:

Theorem 9.45
<D000®@@@@V2 For every graph G,
Figure 9.25 A bandwidth labeling of KS,9' ban(G) ~ X(G) - 1.
300 Matchings, factors and decompositions Labelings of graphs 301
Proof 9.34 Determine gracefullabelings of P6 , P7 , P9 and P IO .
Let G be a graph of order n with V(G) = {V1, V2,"" v,,} and X(G) = k and 9.35 Show that the following classes of trees are graceful:
suppose, to the contrary, that ban(G) :( k - 2. Let P denote the path (a) stars, (b) double stars, (c) caterpillars.
v1 v 2,' ,v". Hence G is a subgraph of pk-2. Assign the color i
9.36 Determine harmonious labelings of C7 and C9 •
00

(1 :( i :( k - 1) to Vj if j =- i (mod k - 1). Hence for each i (1 :( i :( k - 1),

the set Vi of all vertices colored i is independent in G, and thus this is 9.37 Use the proof of Theorem 9.37 to give harmonious labelings of P9
a proper coloring of G. Therefore, X(G) :( k - 1, which is a contradic- and PIO •
tion. 0 9.38 Show that every double star is harmonious.

Theorem 9.46 9.39 Use the proof of Theorem 9.38 to give bandwidth labelings of C8
and C9 •
For every graph G,
9.40 Give a bandwidth labeling of K4,11'
ban(G) ~ K,(G).
9.41 Show that if G is an (n, m) graph for which m > G) - (/1 2k), then
ban(G) > k.
Proof
Let G be a graph of order n with ban(G) = k (1 :( k:( n - 1), and let 9.42 Show that if G is a nonplanar graph, then ban(G) ~ 4.
P,,: V1' V2,' .. , v" denote a path of order n. By Theorem 9.43, G is a 9.43 Prove Theorem 9.47.
subgraph of ~. If k = n -1, then certainly K.(G) :( n - 1. Otherwise,
9.44 Prove Theorem 9.48.
{V2' V3, ooVk+ 1} is a vertex-cut of cardinality k in ~. Thus, K.(G) :(
00

K,(~) :( k. 0

The next result prOVides lower and upper bounds in terms of the
independence number. The lower bound is due to Chvatal [CS] and the
upper bound to Chvatalova (see Chvcitalovci, Dewdney, Gibbs and
Korfhage [CDGKl]).

Theorem 9.47
For every graph G of order n,
rn/}3(G)l -1 :( ban(G) :( n - L}3(G)/2J -1.

We conclude with an upper bound for the bandwidth of a graph in

terms of its diameter.

Theorem 9.48
For every connected graphG of order n,
ban(G) :( n - diamG.

EXERCISES 9.3

9.33 Determine gracefullabelings of C15 and CJ6 .

 The domination number of a graph 303

CHAPTER 10

Domination in graphs
Y1 Z1 Zz YZ
Figure 10.1 Dominating sets.
Next we turn our attention to sets of vertices in a graph G that are close to
all vertices of G, in a variety of ways, and study minimum such sets and
their cardinality.

10.1 THE DOMINATION NUMBER OF A GRAPH

A vertex v in a graph G is said to dominate itself and each of its neighbors,

that is, v dominates the vertices in its closed neighborhood N[v]. A set S
of vertices of G is a dominating set of G if every vertex of G is dominated
by at least one vertex of S. Equivalently, a set 5 of vertices of G is a
dominating set if every vertex in V(G) - 5 is adjacent to at least one
vertex in 5. The minimum cardinality among the dominating sets of G
is called the domination number of G and is denoted by I(G).
A dominating set of cardinality ,( G) is then referred to as a minimum
dominating set. Figure 10.2 The minimum number of queens that dominate the squares of a
The sets S1 = {Vt>V2,Y1,Y2} and 52 = {W1,W2,x} are both dominating chessboard.
sets in the graph G of Figure 10.1, indicated by solid circles. Since 52 is
a dominating set of minimum cardinality, I(G) = 3.
Dominating sets appear to have their origins in the game of chess,
where the goal is to cover or dominate various squares of a chessboard
by certain chess pieces. In 1862 de Jaenisch [D1] considered the problem
of determining the minimum number of queens (which can move either
horizontally, vertically or diagonally over any number of unoccupied
squares) that can be placed on a chessboard such that every square is
either occupied by a queen or can be occupied by one of the queens in
a single move. The minimum number of such queens is 5 and one possible
placement of five such queens is shown in Figure 10.2.
Two queens on a chessboard are attacking if the square occupied by one
of the queens can be reached by the other queen in a single move; other-
wise, they are nonattacking queens. Clearly, every pair of queens on the
chessboard of Figure 10.2 are attacking. The minimum number of non-
attacking queens such that every square of the chessboard can be reached
by one of the queens is seven. Apossible placement of seven nonattacking Figure 10.3 The minimum number of nonattacking queens that dominate the
queens is shown in Figure 10.3. squares of a chessboard.
304 Domination in graphs The domination number of a graph 305
The connection between the chessboard problem described above and Theorem 10.2
dominating sets in graphs is immediate. The 64 squares of ~ chess?oar~ If G is a graph without isolated vertices and 5 is a minimal dominating set of G,
are the vertices of a graph G and two vertices (squares) are adjacent m G If then V(G) - 5 is a dominating set of G.
each square can be reached by a queen on t~e other square by ~ ~ingle
move. The graph G is referred to as the queen s graph. Then the ml~umum Proof
number of queens that dominate all the squares of a chessb?ard IS ,,(G).
The minimum number of nonattacking queens that dommate all the Let v E 5. Then v has at least one of the two properties 00.1) and 00.2)
squares of a chessboard is the minimum cardinality of an independent described in the statement of Theorem 10.1. Suppose first that there exists
dominating set in G. . a vertex w in V(G) - 5 such that N(w) n S = {v}. Hence v is adjacent to
Domination as a theoretical area in graph theory was formalized by some vertex in V(G) - S. Suppose next that v is adjacent to no vertex in S.
Berge in 1958 [see B7, p. 40] and Ore [02, Chap. 13] in 1962. Sinc~ 19.77, Then v is an isolated vertex of the subgraph (5). Since v is not isolated in
when Cockayne and Hedetniemi [CH4] presented a survey of dommatIon G, the vertex v is adjacent to some vertex of V(G) - 5. Thus V(G) - 5 is a
results, domination theory has received considerable attention. . dominating set of G. 0 .
A minimal dominating set in a graph G is a dominating set that contams
no dominating set as a proper subset. A minimal dominating set of For graphs G without isolated vertices, we now have an upper bound
minimum cardinality is, of course, a minimum dominating set and con- for ,,(G) in terms of the order of G.
sists of ,,(G) vertices. For the graph G of Figure 10.1, the set 51 =
{Vb vz, Y1' yz} is a minimal dominating set that is not .a minimum Corollary 10.3
dominating set. Minimal dominating sets were charactenzed by Ore If G is a graph of order n without isolated vertices, then ,,( G) ~ n/2.
[02, p. 206].
Proof
Theorem 10.1 Let 5 be a minimal dominating set of G. By Theorem 10.2, V(G) - 5 is a
A dominating set 5 of a graph G is a minimal domin?ting set of G if.and only if dominating set of G. Thus
every vertex v in S satisfies at least one of the followmg two properties: ,,(G) ~ min{ISI, !V{G) - 51} ~ n/2. 0
CD there exists a vertex win V(G) - 5 such that N(w) n 5 = {v}; (10.1) Many graphs attaining the bound in Corollary 10.3 can be produced by
(ii) v is adjacent to no vertex of S. (10.2) the following operation. The corona cor (H) of a graph H is that graph
obtained from H by adding a pendant edge to each vertex of H. Let
G = cor(H), where G has order n. Then G has no isolated vertices and
Proof ,,(G) = nj2. Indeed, Payan and Xuong [PX1] showed that every compo-
First, observe that if each vertex v in 5 has at least one of the properties nent of a graph G of order n without isolated vertices haVing ,,(G) = nj2 is
(10.1) and (10.2), then S - {v} is not a dominating set ofG. Consequently, either C4 or the corona of some (connected) graph.
S is a minimal dominating set of G. Hence, if G is a graph of order n, then ,(G) ~ ni while, by Corollary
Conversely, assume that 5 is a minimal dominating set of G. Then cer- 10.3, if 6(G) ~ 1, then ')'(G) ~ n/2. McCuaig and Shepherd [MS2] showed
tainly for each v E 5, the set 5 - {v} is not a dominating set of G. Hence that if 6(G) ~ 2, and G is not one of seven exceptional graphs, then
there is a vertex w in V (G) - (5 - {v}) that is. adjacent to no vertex of 6(G) ~ 2n/5. Reed [R3] showed that if 6(G) ~ 3, then 6(G) ~ 3n/8. A
5- {v}. If w = v, then v is adjacent to no vertex of 5. Suppose then that more general result is due to Payan [P2]. We delay its proof until Chapter
w :f= v. Since 5 is a dominating set of G and w rf. 5, the vertex w is adjacent 13 (Theorem 13.4) when the probabilistic method of proof is described.
to at least one vertex of 5. However, w is adjacent to no vertex of S - {v}.
Consequently, N(w) n S = {v}. 0 Theorem 10.4
Let G be a graph of order n with 6 = 6(G) ~ 2. Then
The following result of Ore [02, p. 207] gives a property of the comple-
mentary set of a minimal dominating set in a graph without isolated G) ~ n(1 + In(6 + 1))
vertices. ')' ( "" 6+1 .
 The domination number of a graph 307
306 Domination in graphs
Since K:(G) ~ ~(G) for every graph G, we have the following conse-
Bollohls and Cockayne [BC2] showed that every graph without iso-
quence of Theorem 10.6, due to Walikar, Acharya and Sampathkumar
lated vertices contains a minimum dominating set in which every
[WASl].
vertex satisfies (l0.1).

Theorem 10.5 Corollary 10.7

Every graph G without isolated vertices contains a minimum dominating set 5 If G is a graph of order n, then
such that for every vertex v of 5, there exists a vertex w of G - 5 such that ')'(G) ~ n - K(G).
N(w) n 5 = {v}.
The domination number of a graph without isolated vertices is also
Proof bounded above by all of the covering and independence numbers.
Among all minimum dominating sets of G, let 5 be one such that (5) has
maximum size. Suppose, to the contrary, that 5 contains a vertex v that Theorem 10.8
does not have the desired property. Then by Theorem 10.1, v is an isolated
vertex in (5). Moreover, every vertex of V(G) - 5 that is adjacent to v is
If G is a graph without isolated vertices, then
adjacent to some other vertex of 5 as well. Since G contains no isolated oy(G) ~ min{0:(G),0:1(G),,6(G),,61(G)}.
vertices, v is adjacent to a vertex w in V(G) - 5. Consequently,
(5 - {v}) u {w} is a minimum dominating set of G whose induced sub-
graph contains at least one edge incident with wand hence has a greater Proof
size than (5). This produces a contradiction. 0 Since every vertex cover of a graph without isolated vertices is a dominat-
iIlg set, as is every maximal independent set of vertices, ')'(G) ~ o(G) and
Bounds for the domination number of a graph can be given in terms of ')'(G) ~ ,6(G). Let X be an edge cover of cardinality 0:1 (G). Then every
the order and the maximum degree of the graph. The lower bound in the vertex of G is incident with at least one edge in X.
following theorem is due to Walikar, Acharya and Sampathkumar Let 5 be a set of vertices, obtained by selecting an incident vertex with
[WASl], while the upper bound is due to Berge [B8]. each edge in X. Then 5 is a dominating set of vertices and oy(G) ~ 151 ~
IXI =01(G).
Theorem 10.6 Next, let M be a maximum matching in G. We construct a set 5 ofvertices
consisting of one vertex incident with an edge of M for each edge of M.
If G is a graph of order n, then Let uv E M. The vertices u and v cannot be adjacent to distinct M-vertices

fl + ~(G) 1~ I(G) ~ n - ~(G).

x and y, respectively; for otherwise, x, u, v, Y is an M-augmenting path
in G, contradicting Theorem 9.2. If u is adjacent to an M-vertex, place u
in 5; otherwise, place v in 5. This is done for each edge of M. Thus, 5 is a
dominating set of G, and oy(G) ~ 151 = IMI =,61 (G). 0
Proof
We begin with the lower bound. Let 5 be a minimum dominating set of G. Vizing [V5] obtained an upper bound for the size of a graph in terms of
Then its order and domination number. We omit the proof of this result.
V(G) - 5 ~ U N(v),
vES
Theorem 10.9
implying that W(G) - SI·~ 151· ~(G). Therefore, n -,(G) ~ oy(G) . ~(G)
and so oy(G) ~ rnj(l + ~(G»l. If G is an (n, m) graph for which I = oy(G) ~ 2, then
Next we establish the upper bound. Let v be a vertex of G with
(n-')')(n-,+2)
degv = ~(G). Then V(G) - N(v) is a dominating set of cardinality m~ 2 . (10.3)
n - ~(G); so oy(G) ~ n - ~(G). 0
308 Domination in graphs The domination number of a graph 309
With the aid of Theorem 10.9, we can now supply bounds for the domi- Next we verify the upper bound in (i). If G has an isolated vertex, then
nation number of a graph in terms of its order and size. The lower bound /(G) ~ nand /(G) = 1; while if G has an isolated vertex, then ,(G) ~ n
is due to Berge [BS]. and ,(G) = 1. So, in these cases, ,(G) + ,(G) ~ n + 1. If neither G nor G
has isolated vert~es, then /(G) ~ n/2 and /(G) ~ n/2 by Corollary 10.3
and so ,(G) + ,(G) ~ n. .
Theorem 10.10
It remains then only to verify the upper bound in (ii). The upper bound
If G is an (n, m) graph, then is immediate if ,(G) = 1, so we assume that ,(G) = k ~ 2. Let
n - m ~ /(G) ~ n + 1 - VI + 2m. 5 = {VI, VZ, ... ,vd be a minimum dominating set of G and partition
V(G) into /(G) = k subsets VI, Vz, ... , V k subject to the conditions that
Furthermore, /(G) = n - m if and only if each component of G is a star or an (a) Vi E Vi for 1 ~ i ~ k and all vertices in Vi are dominated by Vi and
isolated vertex. (b) the sum over all integers i (1 ~ i ~ k) of the number of vertices in Vi
adjacent to all other vertices in Vi is a maximum.
Proof _ We now show that each set Vi (1 ~ i ~ k) is a dominating set of
G. Suppose that this is not the case. Then there exists a vertex x E VI
Rewriting the inequality (10.3) given in Theorem 10.9, we have
that is adjacent in G to no vertex of V s for distinct integers sand t
(n _/(G))z + 2(n -/(G)) - 2m;:' O. (lOA) with 1 ~ s, t ~ k. Then x is adjacent in G to every vertex of V s' If x = VI'
then 5 - {v s } is a dominating set of G having cardinality less than
Solving the inequality (10.4) for n -/(G) and using the fact that
n -,(G) ;:. 0, we have that /(G), which is impossible. Consequently, x E VI - {VI}' If x is adjacent
in G to every other vertex of VI' then (5 - {v s' VI} ) u {x} is a dominating
n - /(G);:' -1 + VI + 2m, set of G having cardinality less than /(G), which is again impossible.
which establishes the desired upper bound. Therefore, x is adjacent in G to every vertex of Vs but not to every
Since /(G) ;:. I, the lower bound is established when m ;:. n -I, which vertex of VI'
includes all connected graphs. Assume then that m ~. n - 1. Then G is a Define V~=VI-{X} and V~=Vsu{x}. For r=ls,t, define V~=Vr'
graph with at least n - m components. The domination number of each Thus, we now have a partition of V (G) into subsets V~, V~, ... , V~ such
component of G is at least 1; so ,( G) ;:. n - m, with equality if and only if G that Vj E vi for 1 ~ i ~ k and all vertices in vi are dominated by Vi.
has exactly n - m components, each with domination number 1. This can However, the sum over all subsets vi (1 ~ i ~ k) of the number of vertices
occur only, however, if G is a forest with n - m components, each of which in vi adjacent to all other vertices of V; exceeds the corresponding sum for
is a star or an isolated vertex. 0 the partition Vb V z, . .. , Vb which is a contradiction.
Th~s, as claimed, each subset Vi (1 ~ i ~ k) is a dominating set in G;
The Nordhaus-Gaddum theorem (Theorem 8.17) provided sharp so ,( G) ~ lViI for each i. Hence
bounds on the sum and product of the chromatic numbers of a graph
k
and its complement. We now present the corresponding result for the
domination number. The following result is due to Jaeger and Payan n= L lViI ~ /(G) . /(G).
i=l
0
UPI], the proof of which is based on a proof by E. J. Cockayne.
The upper bound in (i) in Theorem 10.11 can be restated as: If Kn (n ~ 2)
Theorem 10.11 is factored into GI and Gz, then /( G1 ) +,(Gz) ~ n + 1. Goddard, Henning
If G is a graph of order n ;:. 2, then and Swart [GHSl] obtained the corresponding upper bound for three
factors.
(i) 3 ~ ,(G) + /(G) ~ n + I,
(ii) 2 ~ ,(G) ·/CG) ~ n.
Corollary 10.12
Proof
If Kn is factored into G1 , Gz and G3 , then
The lower bounds in .Q) and (ii) follo~ immediately from the observation
that if ,(G) = 1 or ,(G) = I, then ,(G) ~ 2 or ,(G) ~ 2, respectively.
310 Domination in graphs The domination number of a graph 311
Proof Theorem 10.14
Since G2 EEl G3 = GIt it follows from Theorem 10.11 that i(G1 )+ A graph G is 2-domination maximal if and only ifG is a galaxy.
i( G2 EEl G3 ) .:;; n + 1. Now 'let S be a dominating set for G2 EEl G3 . Thus
every vertex of G2 E9 G3 is dominated by a vertex of S. Consequently, Proof
for each vertex v of G2 EEl G3 , the vertex v is not dominated by a vertex
Assume first that G is a graph of order n ~ 2 such that G is a galaxy. Then
of S in at most one of G2 and G3 • Thus, in extending S to dominating
t.(G) = n - 2; so i(G) ~ 2. Let u and v be nonadjacent vertices of G. By
sets S2 and S3 for G2 and G3, respectively, each vertex of G2 EEl G3 need
hypothesis, uv is an edge of a star in G, where, say, u is an end-vertex of
be added at most once. So i(G 2 ) + I(G 3 ) .:;; i(G 2 EEl G3 ) + n. Therefore,
the star. Then in G + uv, the vertex u has degree n - 1. Thus,
i(G I ) + i(G 2 ) + I(G 3 ) .:;; i(GI ) + I(G2 EEl G3 ) + n ~ 2n + 1. 0 I(G + uv) = 1. Hence i(G) = 2, and Gis 2-domination maximal.
For the converse, assume that G is a 2-domination· maximal graph.
All of the bounds presented in Theorem 10.11 are sharp (Exercise 10.8); Then, for every two nonadjacent vertices u and v of G, it follows that
however, if neither G nor G has isolated vertices, then an improved upper I(G + uv) = 1. Hence either u or v, say u, has degree n -1 in G + uv,
bound for i( G) + I(G), due to Joseph and Arumugam GAl], can be given. which implies that u has degree 1 in G. This implies that uv and, in fact,
every edge of G is a pendant edge. Thus G is a galaxy. 0
Theorem 10.13
Although 3-domination maximal graphs have not been characterized,
If G is agraph oforder n ~ 2 such that neither G nor G has isolated vertices, then some properties of these graphs have been found. The following result is
- n+4 due to Sumner and Blitch [SBll.
i(G) +i(G)':;; -2-'
Theorem 10.15
Proof Every connected 3-domination maximal graph of even order has a I-factor.
Since neither G nor G has isolated vertices, it follows from Corollary 10.3 Wojcicka [W9] showed that 3-domination maximal graphs p0i>sess
that I(G) .:;; n/2 and I(G) .:;; n/2. Hence if either i(G) = 2 or I(G) = 2, some hamiltonian properties.
then the proof is complete. If i( G) ~ 4 and i(G) ~ 4, then by the upper
bound in (ii) in Theorem 10.11, we have that I(G) .:;; nh(G) .:;; n/4 and Theorem 10.16
I(G) .:;; nh(G) .:;; n/4; so I(G) + I(G) .:;; n/2. Hence we may assume that
I(G) = 3 or I(G) = 3, say the former. Thus 3 = I(G) .:;; n/2, so n ~ 6. By Every connected 3-domination maximal graph of order at least 7 contains a
Theorem 10.11, I(G) .:;; n/3. Therefore hamiltonian path.
- n n
I(G) + i(G) .:;;3 +3':;; 2+ , 0 It is not known whether every connected 3-dornination maximal graph
2 of order at least 7 is hamiltonian. We close this section with another
For the bound stated in Theorem 10.13 to be attained, either G or G must unsolved problem, namely, a conjecture due to V. G. Vizing.
have domination number n/2. In the discussion following Corollary 10.3,
graphs G of order n without isolated vertices and having 8(G) = n/2 were Vizing's Conjecture
described.
For every two graphs G and H,
If G is a graph containing nonadjacent vertices u and v, then either
I(G + uv) = I(G) or I(G + uv) == I(G) - 1. A graph G is called domination I(G x H) ~ i(G) "i(H).
maximal if I( G + uv) = I( G) - 1 for every two nonadjacent vertices u and v
of G. If G is a domination maximal graph with i(G) = k, then G is k-
domination maximal. The I-domination maximal graphs are (vacuously) EXERCISES 10.1
the complete graphs. The 2-domination maximal graphs were character-
ized by Sumner and Blitch [SBll. A galaxy is a forest, every component of 10.1 Determine the domination numbers of the 3-cube Q3 and the 4-
which is a (nontrivial) star. cube Q4'
312 Domination in graphs The independent domination number of a graph 313
10.2 (a) Determine (with proof) a formula for '}'(C n ). u UZ VI V
(b) Determine (with proof) a formula for '}'(P n )·
10.3 Obtain a sharp lower bound for the domination number of a con- G:
nected graph G of order n and diameter k.
10.4 State and prove a characterization of those graphs G with '}'(G) = 1.
10.5 Investigate the sharpness of the bounds given in Theorem 10.6.
10.6 (a) Does there exist a graph G such that I(G) = a(G) but I(G) is w
strictly less than each of the numbers a1 (G), (3( G) and. (31 (G)?
(b) The question in (a) suggests three other questions. State and
answer these questions.
10.7 Show that. equality is possible for the upper bound given in
Theorem 10.10.
10.8 Show that all bounds given in Theorem 10.11 are sharp.

S2= {x,y,z} 83= {u, V, w}

Figure 10.4 Dominating sets and maximal independent sets.
10.2 THE INDEPENDENT DOMINATION NUMBER OF A GRAPH

It is not difficult to see that every maximal independent set of vertices in

Proof
a graph G is a dominating set of G. Thus, I(G) ~ i(G), where, recall, i(G) We have already noted that every maximal independent set of vertices is a
is the lower independence number of G. Not every dominating set is dominating set. Conversely, suppose that 5 is an independent dominating
independent, however. Indeed, not every minimum dominating set is set. Then 5 is independent and every vertex not in S is adjacent to a vertex
independent. For example in the graph G of Figure 10.4, the set 51 = of 5, that is,S is maximal independent. D
{Ul, Uz, VI' Vz, WI' w2}is a maximal independent set (and consequently a
dominating set) of G; while 5 z = {x,y,z} is a minimum dominating set Another observation now tollows.
of G and certainly 52 is not independent. (These sets are indiCated in
Figure 10.4 by solid circles.) However, G does contain a minimum
dominating set of G that is independent, namely, 53 = {u, v, w}. Corollary 10.18
Our attention now shifts in this section to dominating sets that are also Every maximal independent set ofvertices in agraph is a minimal dominating set.
maximal independent sets. A set 5 of vertices in a graph G is called an
independent dominating set of G if 5 is both an independent and a dominat-
ing set of G. Thus the sets 51 and 53 in Figure 10.4 are independent domi- Proof
nating sets while 52 and 54 = {Ul' VI, wd are not. The independent Let 5 be a maximal independent set of vertices in a graph G. By Theorem
domination number i(G) of G is the minimum cardinality among all 10.17,5 is a dominating set. Since 5 is independent, certainly every vertex
independent dominating sets of G. That this is precisely the notation of 5 is adjacent to no vertex of S. Thus, every vertex of 5 satisfies property
used for the lower independence number of a graph is justified by the (ii) of Theorem 10.1. So, by Theorem 10.1, 5 is a minimal dominating
following observation of Berge [BBl. set. D

Theorem 10.17 We have noted that I(G) ~ i(G) for every graph G. That this inequality
can be strict is illustrated by the queen's graph G for which I(G) = 5 and
A set S of vertices in a graph is an independent dominating set if and only if S is =
i(C) 7. Also, equality can hold since 'Y(K1,t) = i(K1,t) = 1 for every
maximal independent. positive integer t. For 1 ~ s < t, let H be the graph obtained from Ks,t by
314 Domination in graphs The independent domination number of a graph 315
adding a pendant edge to each vertex of the partite set of cardinality s. The converse of Corollary 10.20 is certainly not true, though, since
Then ,(H) = i(H) = s. That the difference between i(G) and ,(G) can be ,(K1,3) = i(K1,3) = 1.
arbitrarily large can be seen in the double star T containing two vertices of Since every line graph is claw-free (by Theorem 4.42), we have a
degree k ~ 2, where i(T) = k and ,(T) = 2. consequence of Corollary 10.20.
For some special classes of graphs, Bollobas and Cockayne [BCl]
determined an upper bound for i(G) in terms of ,(G).
Corollary 10.21

Theorem 10.19 For every graph G,

If G is a K1,k+ I-free graph, where k ~ 2, then

,(L(G)) = i(L(G)).
i(G) ~ (k - l),(G) - (k - 2). No forbidden subgraph characterization of graphs G for which
,(G) = i(G) is possible; for suppose that H is a given graph and we
define G = K 1 +H. Then ,(G) = i(G) = 1.
Proof In Chapter 8, a graph G is defined to be perfect if X(H) = w(H) for every
Let 5 be a minimum dominating set of vertices of G and let 5' be a max: i- induced subgraph H of G. A graph G is domination perfect if ,(H) = i(H)
mal independent set of vertices of 5 in G. Thus, 151 = ,(G) and 15'1 ;::: 1. for every induced sUbgraph H of G. A class of domination perfect graphs
Now, let T denote the set of all vertices in V(G) - 5 that are adjacent in G is provided by Corollary 10.20.
to no vertex of 5', and let T' be a maximal independent set of vertices in T.
Certainly·
. I
then, 5' u T' is an independent set of verticesI of G. Since every
vertex of V(G) - (5' U T) is adjacent to some vertex of 5 and every vertex Corollary 10.22
of T - T' is adjacent to some vertex of T', it follows that 5' U T' is a max- Every claw-free graph is domination perfect.
imal independent set of vertices. Thus, by Theorem 10.17, 5' u T' is an
independent dominating set. Sumner and Moore [SM1] stated that it is not necessary to consider
. Observe that every vertex of 5 - 5' is adjacent to at most k - 1 vertices every induced subgraph of a graph G in order to show that G is domina-
of T'; for if this were not the case, then some vertex v of 5 - 5' is adjacent to tion perfect.
at least k vertices of T' and also at least one vertex of 5', which contradicts
the hypothesis that G contains no induced sUbgraph isomorphic to K1,k+ l'
Also, ol:~serve that every vertex of T' is adjacent to some vertex of 5 - 5'. Theorem 10.23
Therefore,
A graph G is domination perfect if and only if ,(H) = i(H) for every induced
IT'I ~ (k -1)15 - 5'1 = (k -1)(151-15'1) = (k l)(,(G) -15'1)· subgraph H of G with ,(H) = 2.
Consequently, Fulman [F12] obtained a characterization of domination perfect
i(G) ~ IS' u T'I = 15'1 + IT'I graphs in terms of eight forbidden subgraphs; while Zverovich and
Zverovieh [ZZl] discovered such a characterization in terms of seventeen
~ 15'1 + (k -l)(r(G) -lS'I)
forbidden induced subgraphs. A survey of domination perfect graphs
= (k -l)r(G) - (k - 2)15'1 is given in Sumner [SI0]. Allan, Laskar and Hedetniemi [ALH1]
presented an upperbound for ,(G) + i(G) for a graph G without isolated
~ (k-l)r(G) - (k- 2). 0 vertices. .
The special case of Theorem 10.19 where k = 2 is of particular interest.
Theorem 10.24
Corollary 10.20 If G is a graph of order n without isolated vertices, then
If G is a claw-free graph, then ')'(G) = i(G). ,(G) + i(G) ::::; n.
316 Domination in graphs Other domination parameters 317
Proof 10.3 OTHER DOMINATION PARAMETERS
By Theorem 10.8,,(C;) :s;; a(G) and by definition, i(G) :s;; {3(G). By Gallai's
In the previous section we introduced a variant of the classical domina-
theorem (Theorem 9.12), {3(G) + a(G) = n; so ,(G) + i(G) :s;; n. 0
tion number, namely, the independent domination number. In this
The bound presented in Theorem 10.24 is sharp in the sense that there are section, we describe several other domination parameters that have
infinitely many graphs G for which "((G) + i(G) = IV(G)\ (Exercisel0.12). been the object of study.
Gimbel and Vestergaard [GV1] discovered an upper bound for the For a set A of vertices in a graph G, the closed neighborhood N[A] of A is
independent domination number of an arbitrary graph. defined by N[A] = UVEA N[v]. Equivalently, N[A] = N(A) U A. A set 5 of
vertices in G is called an irredundant set of G if for every vertex v E 5, there
exists a vertex WE N[v] such that w r¢ N[5 - {v}]. EqUivalently, 5 is an
Theorem 10.25 irredundant set of vertices of G if N[S - {v}] =1= N[5] for every vertex
If G is a connected graph of order n ~ 2, then vE 5. Every vertex v withthis property is an irredundant vertex. Therefore,
every vertex in an irredundant set is an irredundant vertex. A set 5 of
i(G) ~ n +2 - 2y'ii. vertices that is not irredundant is called redundant. Consequently, a set
The bound presented in Theorem 10.25 is sharp as well (Exercise 10.13). 5 of vertices in a graph G is redundant if and only if. there exists a
For bipartite graphs, however, a simple improved upper bound for the vertex v in 5 for which N[5 - {v}] = N[S]. Such a vertex v is called a
redundant vertex (with respect to 5).
domination number exists.
For the graph G of Figure 10.5, let 5 = {w,y,s}. Then 5 is anirredundant
set of G. For example, u E N[S] but u r¢ N[S - {w}]. Similarly
Theorem 10.26 x r¢ N[5 - {y}] and s r¢ N[S - {s}].
If G is a connected bipartite graph of order n, then i(G) ~ n/2. A characterization of irredundant sets is presented next.

Proof Theorem 10.27

Denote the partite sets of G by VI and V2, where IVII ~ 1V21· Since VI A s.et 5 of vertices in a graph G is irredundant if and only if every vertex v in 5
is an independent dominating set and IVII:s;; n/2, it follows that sattsfies at least one of the following two properties:
i(G) ~ n/2. 0 (i) there exists a vertex w in V(G) - S such that N(w) n 5 = {v}. (10.5)
(ii) v is adjacent to no vertex of S. (10.6)

EXERCISES 10.2
Proof
10.9 Show that a graph need not have any minimum dominating set that First, let 5 be a set of vertices of G such that for every vertex v E 5, at least
is independent. one of the properties (10.5) and (10.6) is satisfied. If (10.5) is satisfied, then
10.10 Prove or disprove: If a graph G contains an independent minimum
dominating set of vertices, then ,(G) = i(G).
10.11 For each integer k ~ 3, show that there exists a graph G such that
i(G) = k and "((G) = 3.
10.12 Show that there ~re infinitely many graphs G for which G: z s= {w,Y,sj
"((G) + i(G) = IV(G)I.
10.13 Show that the bound stated in Theorem 10.25 is sharp.
10.14 Show that the bound given in Theorem 10.26 is sharp. r s
Figure 10.5 Irredundant sets of vertices.
318 Domination in graphs Other domination parameters 319
there exists a vertex W E N[v] such that W tf. N[S - {v}]. If (10.6) is satisfied,
then v tf. N[S - {v}]. In either case, S is irredundant. w x
Conversely, let S be an irredundant set of vertices in G, and let v E S,
UI VI
Since 5 is irredundant, there exists W E N[v] such that w tf. N[S - {v}]. If H: T:
w =f v, then (10.5) is satisfied; while if w = v, then (10.6) is satisfied. 0
U2 V2
By Theorem 10.1, then, a minimal dominating set of vertices in a graph
Y z
is an irredundant set. Hence, every graph has an irredundant dominating U3 V3
set of vertices. Figure 10.6 Graphs whose domination numbers exceed their irredundance
If S is an irredundant set of vertices in a graph G, then for each v E S, the numbers.
set N[v] - N[S - {v}] is nonempty. Each vertex in N[v] - N[S - {v}] is
referred to as a private neighbor of v. The vertex v may, in fact, be a private Figure 10.6, we have '}'(H) = 3 and ir(H) = 2. The set {Uj, Vj} is a maximal
neighbor of itself. Consequently, a nonempty set 5 of vertices in a graph G irredundant set of minimum cardinality in H. In order to see that {Uj, Vj} is
is irredundant if and only if every vertex of S has a private neighbor. a maximal irredundant set in H, observe that (1) t has no private neighbor
Certainly every nonempty subset of an irredundant set of vertices in a in {t, uj, Vj}, (2) Uj has no private neighbor in {uj, Vj, U2} and {Uj, Vj, U3},
graph G is irredundant. Also, every independent set of vertices is an and (3) VI has no private neighbor in {Uj, Vj, V2} and {Uj, Vj, V3}' Moreover,
irredundant set. for the tree TofFigure 10.6, ,(T) = 5 and ir(T) = 4. The set {w,x,y,z} is a
The irredundance number ir( G) of a graph G is the minimum cardinality maximal irredundant set of minimum cardinality in T.
among the maximal irredundant sets of vertices of G. Since the set Cockayne and Mynhardt [CM1] provided a lower bound for the
5 = {r, z} is a maximal irredundant set of vertices of minimum cardinality irredundance number of a graph in terms of its order and maximum
for the graph G of Figure 10.5, it follows that for this graph, ir(G) = 2. To degree.
see that S is irredundant, observe that y is a private neighbor of r, and w is
a private neighbor of z. To see that 5 is a maximal irredundant set, note that
. (1) {5, r, t} is not irredundant since 5 would have no private neighbor, (2)
Theorem 10.29
{x, r, t} and {y, r, t} are not irredundant since r would have no private If G is a graph of order n and maximum degree 6.( G) ;:, 2, then
neighbor, and (3) {u, r,z}, {v, r, t} and {w, r, z} are not irredundant since
z would have no private neighbor. Hence a maximal irredundant set need ir( G) ;:, 3~~G) .
not be a dominating set and, strictly speaking, the irredundance number
is not a domination parameter.
An inequality relating the irredundance number of a graph and its
The next result summarizes how the parameters discussed thus far in
domination number was discovered by both Allan and Laskar [All]
this chapter are related.
and Bollobas and Cockayne [BC1].

Theorem 10.28
Theorem 10.30
For every graph G,
For every graph G,
ir(G) :::; ,(G) :::; i(G).
,(G) :::; 2 ir(G) - 1.
Proof A forbidderr subgraph characterization of graphs G for which
We have already observed that 'Y( G) :::; i( G). The inequality ir( G) :::; ,(G) is ir(G) = '}'(G) cannot exist. To see this, let G be a graph and define
a consequence of the fact that every minimal dominating set of vertices of H = Kj + G. Then H contains G as an induced subgraph and
G is an irredundant set. 0 ir(H) = '}'(G) = 1. A sufficient condition in terms of forbidden subgraphs
for a graph to have equal irredundance number and domination number
That the inequality ir(G) ~ ,(G) may be strict is illustrated in the graph was found by Laskar and Pfaff [LPl]. Recall that a graph G is chordal if
G of Figure 10.5, where ,(G) = 3 and ir(G) = 2. Also, for the graph H of every cycle of order 4 or more contains a chord.
320 Domination in graphs Other domination parameters 321
Theorem 1031 0. Favaron conjectured, however, that not all six graphs of Figure 10.7
are needed as forbidden subgraphs for an irredundance perfect graph.
If G is a chordal graph that contains neither of the graphs Hand T of Figure 10.6
as an induced subgraph, then ir(G) = 'Y(G).
Favaron's Conjecture
A graph G is then defined to be irredundance perfect if ir(H) = 'Y(H) If a graph G has no induced subgraph isomorphic to any of the graphs Gl , Gz and
for every induced subgraph H of G. We now have two immediate G3 of Figure 10.7, then Gis irredundance perfect.
corollaries.
In this connection, Henning [H12l obtained the following result.

Corollary 10.32
Theorem 10.36
A chordal graph G is irredundance perfect if and only if G contains neither of the
graphs Hand T of Figure 10.6 as an induced subgraph. If a graph G has no induced subgraph isomorphic to any of the graphs Gl , G2 and
G3 of Figure 10.7, then 'Y(H) = ir(H) for every induced sUbgraph H of G with
ir(H) ~ 4.
Corollary 10.33
From Theorem 10.23, a graph G is domination perfect if and only if
A tree G is irredundance perfect if and only if G does not contain the tree T of 'Y(H) = i(H) for every induced subgraph H of G with 'Y(H) = 2. Henning
Figure 10.6 as a subgraph. [H12l conjectured that such a result exists for irredundance perfect graphs
as well.
Favaron [F3l established a forbidden subgraph sufficient condition for a
graph G to have ir(G) = 'Y(G) = i(G).
Henning's Conjecture

Theorem 10.34 A graph G is irredundance perfect if and only if 'Y(H) = ir(H) for every induced
subgraph H of G with ir(H) ~ 4.
If a graph G is both claw-free and H-free, for the graph H of Figure 10.6, then
ire G) = 'Y( G) = i( G). It follows that if Henning's Conjecture is true, then so too is Favaron's.
We now introduce two other domination parameters - the so-called
Favaron [F3l also established a sufficient condition for a (not necessarily upper domination parameters. The upper domination number reG) of a
chordal) graph to be irredundance perfect in terms of six forbidden graph G is the maximum cardinality of a minimal dominating set of G;
subgraphs. while the upper irredundance number IR(G) of G is the maximum cardinal-
ity of an irredundant set of G. We now summarize in Figure 10.8 the para-
meters we have introduced thus far in this chapter together with the
Theorem 10.35
vertex independence parameters we introduced earlier.
If a graph G has no induced subgraph isomorphic to any of the six graphs Gi The six parameters described in Figure 10.8 make up a string of inequal-
(1 ~ i ~ 6) shown in Figure 10.7, then G is irredundance perfect. ities, which was first observed by Cockayne and Hedetniemi [CH4].

minimum cardinality maximum cardinality

maximal independent lower independence independence number
set of vertices in G number i(G) (3(G)
minimal dominating domination number upper domination
set of vertices of G 'Y(G) .' number reG)
() 0 maximal irredundant irredundance number upper irredundance
set of vertices of G ir(G) numberIR(G)
Ch Gs
Figure 10.7 Forbidden subgraphs for irredundance perfect graphs. Figure 10.8 Summary of definitions of parameters.
322 Domination in graphs Other domination parameters 323
Theorem 10.37 Ul U2 U3

For every graph G, G:

ir(G) ~ f(G) ~ i(G) ~ (3(G) ~ r(G) ~ IR(G).

VO VI V2 v3 V4 vs v6 1I7 vg vg VI0

Proof Figure 10.9 A graph with 2-domination number 3.

Since every minimal dominating set is an irredundant set, it follows

extended adjacency, while open domination is ordinary adjacency. These
that r(G) ~ IR(G). Moreover, every maximum independent set is a
two types of domination can be described in terms of distance. The vertex
dominating set; so (3(G) ~ r(G). Also, since an independent dominating
v dominates all vertices w with d(v, w) ~ 1 and openly dominates all
set is independent, i(G) ~ (3(G). The result now follows from Theorem
vertices w with d(v, w) = 1. For a positive integer k, the k-neighborhood
10.28. 0
Nk(v) is defined by
Cockayne, Favaron, Payan and Thomason [CFPTl] have shown that Nk(V) = {w E V(G) Id(v, w) = k}.
graphs exist having distinct values for all six parameters mentioned in
Theorem 10.37. For bipartite graphs, however, three of these parameters The vertex v k-dominates all vertices w withd(v, w) ~ k and k-step dominates
must have the same value. all vertices w with d(v, w) = k, that is, v k-step dominates all vertices in
its k-neighborhood. Consequently, I-domination and domination are
equivalent, as are I-step domination and open domination.
Theorem 10.38 For a graph G and a positive integer k, a set 5 is a k-dominating set if every
For every bipartite graph G, vertex of G is at distance at most k from some vertex of S. The k-domination
number fk(G) of G is the minimum cardinality among the k-dominating
(3(G) = f(G) = IR(G).
sets of G. Thus, fl(G) =f(G) and fi+I(G) ~fi(G) for every positive
integer i. For the graph G of Figure 10.9, the set {V2' Vs, V8} is a minimum
Proof 2-dorninating set; so f2(G) = 3.
Let G be a bipartite graph with partite sets U and W. Let S be a maximum Minimal k-dominating sets were characterized by Henning, Oeller-
irredundant set of vertices of G, and let T be the set of isolated vertices of mann and Swart [HOSI] in a theorem that generalizes Theorem 10.1
(5). Furthermore, let (Exercise 10.20).
UI = Tn U, Uz = (S n U) - T, WI = Tn W, Wz = (S n W) - T,
where one or more of these sets may be empty. Each vertex w E Wz is Theorem 10.39
irredundant in S. Since w is not isolated in (S),the vertex w is not its A k-dominating set S (k ~ 1) of a graph G is a minimal k-dominating set of G if
own private neighbor. However, since 5 is an irredundant set, w is a pri- and onlyif every vertex v in 5 satisfies at least one of the following two properties:
vate neighbor of some vertex of V( G) - 5. Hence for w E Wz, there exists a
vertex w' E V(G) - 5 such that N(w' ) n 5 = {w}. Moreover, since WE W, en there exists a vertex win V(G) - 5 such that v is the unique vertex of S
it follows that w' E U. whose distance from w is at most k; (l0.7)
Let A = {w' IWE W z}. Then IAI ~ IWzl and A cU. Furthermore, no (ii) for all x E 5, x i= v, d(v, x) ~ k + 1. (l0.8)
vertex. of A is adjacent to a vertex of WI' Consequently, UI U Uz U
The following result relates k-dominating sets with the kth power of a
WI U A is independent in G. Hence
graph (Exercise Hl.2I).
(3(G) ~ lUll + IUzl + IWtI + IAI ~ 151 = IR(G).
The result now follows from Theorem 10.37. 0 Theorem 10.40
While a vertex v in a graph G dominates the vertices in its closed neigh- For every connected graph G and positive integer k,
borhood N[v], the vertex v is said to openly dominate the vertices in its
neighborhood N(v). Thus ordinary domination can be considered as
324 Domination in graphs Other domination parameters 325
The following result is due to Henning, Oellermann and Swart (HOS2]. Theorem 10.43
Let G be a connected graph. The k-step domination number of G is well-defined if
Theorem 10.41 and only if rad G ~ k.
If G is a connected graph of order at least k + 1 ~ 2, then G has a minimum k- Let G be a connected graph with radius r and diameter d. While the
dominating set 5 with the property that for each v E 5, there exists a vertex sequence 'Y(G) = 1'1 (G), I'2(G), ... , 'Yd(G) = 1 is nonincreasing, such need
WE V(G) S with d(v, w) = k such that v is the unique vertex of S whose not be the case for P1(G),P2(G), ... ,Pr(G). For positive integers i and j
distance from w is at most k. with i < j, there exist graphs G and R such that Pi(G) - Pj(G) and
Pj(G) - Pi(G) are arbitrarily large «(CHSl]). Indeed, it may be more effi-
As a consequence of Theorem 10.41, we have an upper bound for the k- cient to step dominate the vertices of G using distinct distances.
domination number of a graph in terms of its order. This result is also due Let G be an (n,m) graph with V(G) = {Vl,V2'''''V n}. A sequence
to Henning, Oellermann and Swart (HOSl]. s: t 1, t 2, ... , In of positive integers is called a universal dominating sequence
for G if every vertex of G is ti-step dominated by Vi for some i
Corollary 10.42 (1 ~ i ~ n). The integer Ii is referred to as the step of Vi (1 ~ i~ n).
Every nontrivial connected graph has a universal dominating sequence;
If G is a connected graph of order n ~ k+1 ~ 2, then indeed, the constant sequence 1,1, ... ,1 of length n is such a sequence.
n The value val(s) of a universal dominating sequence s: t l ,t2, ... ,tn is the
'Yk(G) ~ k+ l' number of vertices of G that are ti-step dominated for some i (1 ~ i ~ n)
counting multiplicities, that is,
Proof n

By Theorem 10.41, G contains a k-dominating set S of cardinality 'Yk(G)

val(s) = 2: IN(.(v;)l.
i=1
with the property that for each v E 5, there exists a vertex WE V(G) - 5
with d(v, w) = k such that v is the unique vertex of S whose distance from For the constant sequence s = {I} of length n, val(s) = 2m. Thus the First
W is at most k. Let 'Yk(G) = I and S = {VI,V2""'V(}. For 1 ~ i ~ I, let Theorem of Graph Theory is a special instance of the value of a universal
Wi E V(G) - S such, that d(Vi' Wi) = k such that Vi is the unique vertex of dominating sequence.
S whose distance from Wi is at most k. Hence G contains pairwise disjoint A graph G is a constant universal graph if there exists a positive integer N
Vi-Wi paths Pi (1 ~ i ~ t) of length k. Consequently, n ~ I(k + 1), which such that val(s) = N for every universal dominating sequence s of G. Every
produces the desired bound. 0 complete graph and every odd cycle is a constant uniyersal graph. Hayes,
Schultz and Yates [HSYl] characterized constant universal graphs.
For a graph G and a positive integer k, a set S is a k-step dominating set if
every vertex of G is at distance exactly kfrom some vertex of 5. The k-step Theorem 10.44
domination number Pk(G) is the minimum cardinality of a k-step dominat-
ing set of G. For the graph G of Figure 10.9 (shown again in Figure 10.10), A nontrivial connected graph G is a constant universal graph if and only if
pz(G) = 7. A minimum 2-step dominating set of G is indicated in Figure INj(v) I = INk(v)1 for every vertex v of G and every pair j, k of integers with
10.10 by solid circles. l~j,k~e(v).
Hayes, Schultz and Yates (HSY1] determined those graphs G for which
the k-step domination number of G is well-defined (Exercise 10.22). Proof

"1 "2 "3 U4 If INj(v)1 = INk(v)1 for every vertex v of G and every pair j, k of integers
with 1 ~ j, k ~ e(v), then certainly G is a constant universal graph.
G:
a-l • • I • I • • L
vQ VI v2 v3 v4 vs v6 VB '17 V9 vlO
Fgr the converse, assume that G is a constant universal graph with
V(G) = {VIlV2""'Vn}, and suppose, to the contrary, that some vertex,
say Vv has the property that INj(Vl) I i-INk (Vl)! for some integers j and k
Figure 10.10 A graph with 2-step domination number 7. with l:::;;j < k:::;; e(vl)'
326 Domination in graphs Other domination parameters 327
Suppose first that V1 is not adjacent to an end-vertex of G. Then the V2' Since all other vertices of G are I-step dominated by some vertex of G, it
sequences {h}:'=1 and {kj }:'=1 defined by follows that {k i } is a universal sequence for G. However, then,
. {j
]i=
ifi=1
:L;'=1INj;(Vi)! i= :L7=1INk;(Vi)l, which is a contradiction. 0
1 if2~i~n
and
k ifi=1 G:
i
k = { 1 if2~i~n
are universal dominating sequences for G. However,
n IJ

L: INj,(Vi)1 i= L: !Nk;(Vi)l, (a) pl(G) =3 (b) Pt(G) = 4 (c) Pz(G) =4

i=1 i=1 Figure 10.11 A graph with planetary domination number 3.
which is a contradiction.
Corresponding to the notion of universal dominating sequences is
Next, suppose that V1 is adjacent to an end-vertex V2 and j i= 1. Then
another domination parameter. Let G be a graph of order n. A sequence
INj + 1(V2)1 = !Nj (V1)! i= INk (v1)1 = INk + 1(V2)1· s: t b t 2 , ... , t k (k ~ n) of positive integers is called a planetary domination
Thus the sequences {h}i'=1 and {kJ:'=1 defined by sequence for G if G contains distinct vertices Vb V2' .... ,Vk such that every
vertex of G is {i-step dominated by Vi for some i (1 ~ i ~ k). A planetary
. {j+l ifi=2 dominating sequence s for G is minimal if no proper subsequence of s is a
]i = 1 if i i= 2 (1 ~ i ~ n) planetary dominating sequence for G. The length of a minimum planetary
and dominating sequence for G is called the planetary domination number pl(G)
of G. It is a direct consequence of the definitions that pl(G) ~ Pi(G)
k_{k+l ifi=2 for every nontrivial connected graph G and every integer i with
,- 1 ifii=2 (l~i~n) 1 ~ j ~ rad G. For the graph G of radius 2 shown in Figure 10.11,
are universal sequences for which 2:7=1 INj,(Vj) II 2:7=1 INk;(Vi)!, pro- pl(G) = 3 and P1 (G) = P2 (G) = 4. A minimum planetary domination
ducing a contradiction. sequence and minimum I-step and 2-step dominating sets are indicated
Finally, suppose that V1 is adja'cent to an end-vertex V2 and j = 1. Let V3 in Figure 10.11(a), (b) and (c) respectively.
be a vertex of G such that d(v1' V3) = e(v1) = t ~ 2. Thus, V3 is not adjacent Thus for the graph G of Figure 10.11, pl(G) < Pi(G) for 1 ~ i ~ radG.
to an end-vertex of G. Moreover, d( V3, V2) = t + 1. Consider the sequences Indeed, it was shown in [CHS1] that for every integer k ~ 2, there exists
{j;}?=1 and {kili'=1 defined by a graph G with rad G ~ k such that pI(G) < Pi( G) for every integer i
(1 ~ i ~ radG).
2 ifi=2 ' We note, in closing, that the domination number may be defined for
j;= t+l ifi=3 digraphs as well. Ina digraph 0, a vertex v dominates itself and all vertices
{ adjacent from v. The domination number -y(O) of 0 is the minimum cardin-
1 if i ;i: 2,3 (1 ~ i ~ n)
ality of a set 5 of vertices of 0 such that every vertex of 0 is dominated by
and some vertex of S.
k Hi = 1 For a thorough study of domination in graphs, see Haynes, Hedetniemi
and Slater [HHSl].
k 2 ifi =2
i=
{ ~+1 if i = 3
if4 ~ i ~n.
EXERCISES 10.3
The sequence {j;} is certainly a universal sequence for G. We show that
{ki } is also a universal sequence for G. The vertex V2 2-step dominates all 10.15 (a) Prove that a set S of two or more vertices of G is irredundant if
neighbors of V1 (except V2 itselt of course); while V3 (t + 1)-step dominates and only if it contains no redundant vertex.
328 Domination in graphs
(b) Characterize those graphs G of order n ~ 2 such that every s~t 5 CHAPTER 11
of vertices of G with 151 ~ 2 contains a redundant verte~.
10.16 Verify that ,,(T) = 5 and ir(T) = 4 for the tree T of Figure 10.6.
10.17 Let G be a graph for which ir(G) = ,,(G). Prove that if S is a minimal Extremal graph theory
dominating set of G, then S is a maximal irredundant set.
10.18 Let G be a graph for which ir(G) = ,,(G). If S is a maximal irredun-
dant set of vertices of G of minimum cardinality, is 5 a dominating
set of G?
10.19 Give an example of an infinite class of graphs G for which We have seen results which state that if a graph G of a fixed order n has at
ir(G) < ,,(G). least f(n) edges, then G contains a particular subgraph or G has some
specified property. If the bound f(n) on the number of edges is sharp,
10.20 Prove Theorem 10.39.
then there exists a graph of order n and size f(n) -1 that doesn't contain
10.21 Prove Theorem 10040. the subgraph or doesn't possess the property involved. Such a graph is
called an extremal graph: The problems of determining such sharp
10.22 Prove Theorem 10.43.
bounds f(n) and resulting extremal graphs constitute a major part of an
10.23 Prove that if G is a constant universal graph with rad G = I, then area of graph theory called extremal graph theory. Several problems of
G = pKl UH for some regular graph H and positive integer p. this type are considered in this chapter. There are also extremal problems
10.24 .For the tree T shown below, determine pl(T) and Pi(T) for every that deal with determining the minimum order of a graph possessing
some specified properties. Here this problem is discussed when a
integer i with 1 ~ i ~ rad T.
degree of regularity and girth are prescribed.

T:
11.1 TURAN'S THEOREM
We have seen that if a graph G of order n ~ 3 has at least n edges, then G
has a cycle. Indeed, if G has order n ~ 3 and at least (n; 1) + 2 edges, then
G has a hamiltonian cycle (Exercise 4.20). Both bounds are sharp since
every tree of order n has size n - 1 and certainly contains no cycles;
10.25 For a graph G, define while the graph of order n obtained. by adding a pendant edge to Kn - 1
domG = min{ ,,(0) I[j is an orientation of G} has (n;l) + 1 edges but is not hamiltonian. Furthermore, if a graph G of
order n ~ 2 has at least (n; 1) + 1 edges, then G is connected; indeed, G has
and a hamiltonian path. Moreover, if n is even, then G contains a I-factor. The
DOM G = max{,,(D) I D is an orientation of G}. graph Kn - l U K1 shows that all of these bounds are sharp. These observa-
tions lead us to the main topic of this section and the next.
(a) Determine dom K3 and DOM K3 • For a graph F of order k and an integer n with n ~ k, the extremal number
(b) Show that domG = ,,(G) for every graph G. ex(nj F) of F is the maximum number of edges in a graph of order n that
(c) Show that if k is an integer such that dom G ~ k ~ DOM G, then
does not contaIn F as a subgraph. Consequently, every graph of order n
there exists an orientation 0' of G such that ,,(0') = k. and size ex(n; F) + 1 contains F as a subgraph. The graphs of order nand
size ex(n; F) not containing F as a suhgraph are the extremal graphs.
From the discussion above, ex(n; en) = (nil) + 1 for n ): 3. We now deter-
mine ex(n; F) for some 'small' graphs F. If F = K2 , then ex(n; F) = 0 for
n ~ 2; while if F = P3, then ex(n; F) = In/2J for n ): 3. Furthermore, if
F = 2K2, then ex(n; F) = n - 1 for n ): 4 (Exercises 11.1 and 11.2).
330 Extremal graph theory Turan's theorem 331
Of all the extremal numbers ex(n; F) that have been investigated, the Proof
best known ones have been when F is complete. We begin with the case
F = K3 • The followIng result is due to Tunin [T91. First observe that the result is true if r = 2. Hence it suffices to assume that
n ~ r ~ 3. We now proceed by induction on r. The result follows for r = 3
and all integers n ~ 3 by Theorem 11.1. Assume, for an integer r - 1 ~ 3
Theorem 11.1 and all integers n ~ r - 1, .that every graph of order n and size at least
Every graph of order n ~ 3 and size at least Ln2 / 4J + 1 contains a triangle.
Gr~34)n2 +1
Proof contains Kr - 1 as a subgraph.
It remains to show that every graph G of order n and size m, where
We proceed by induction on n. For n = 3, the only graph of order n and n ~ rand
size at least Ln 2 /4J + 1 is K3 , which, of course, is a triangle. For n = 4,
the only graphs with the given conditions are K4 - e and ~, both of m~
r-2)
- n2 +1
(-
2r-2
which contain triangles. Thus the result is true for n = 3 and n = 4.
Assume that every graph of order k and size at least L0/4J + 1 contains contains Kr as a subgraph. We verify this by induction on n. For n = r,we
a triangle for every integer k with 3 ~ k < n, where n ~ 5. Now let G be a have
graph of order n containing at least Ln2 /4J + 1 edges. Let u and v be
adjacent vertices of G and define H = G - u - v. If u and v are mutually
adjacent to a vertex of H, then G contains a triangle.· Otherwise, each
m ~ ( ; ~22 ) n2 + 1 ~ G)'
Thus,G = Kn = Kr and the result follows.
vertex of H is adjacent to at most one of u and v, and the size of H is at
least Assume now that every graph H of order k, where r ~ k < n, and size at
least
l:J+l-(n-l)= ln2-~n+4J+l = l(n~2fJ+1. (;r~~)0+1
By the inductive hypothesis, H contains a triangle and, consequently, so contains Kr as a sUbgraph. Let G be an (n, m) graph, where
does G. D
m ~ (;r~~)n2+1.
That the bound presented in Theorem 11.1 is best possible follows from
• . 2 We show that G contains Kr as a sUbgraph. Since
the fact that for n ~ 3, the graph Kl n/ 2J,[n/21 has order n, has size Ln /4J
and, of course, is triangle-free. This verifies that ex(nj K3 ) = Ln /4J for
2

n ~ 3. We shall soon see that Kln/ 2J ,[n/21 is, in fact, the unique extremal (;r~22)n2 ~ Gr~~}p ~ m,
graph.
By Theorem 11.1, it follows, of course, that if G is a graph of order n ~ 3 it follows hom the inductive hypothesis that G contains Kr- 1 as a sub-
and size at least (n 2 /4) + 1, then G contains K 3 as a subgraph. This result is graph. Let U be the vertex set of a subgraph of G that is isomorphic to
now extended to complete graphs of any order r ~ 2. Kr _ 1 , and define H = G - U. If some vertex of H is adjacent to all vertices
of U, then G contains Kr as a sUbgraph. Otherwise, every vertex of H is
adjacent to at most r - 2 vertices of U. Thus, the size of G is at most
Theorem 11.2
Let rand n be positive integers, where n ~ r ~ 2. Then every graph oforder nand (1'; 1) +(n _ r +l)(r _ 2) + (n -; +1).
~~~ -
If n - r + 1 < r, then n ~ 2(r - 1). However, the inequalities
r-2) 2 r ~ n ~ 2(r -1) are equivalent to the inequality
( 2r-2 n +1
contains Kr as a subgraph. (r;1)+(n_r+l)(r_2)+(n-;+I) ~ (;r-=-22)n2,
332 Extremal graph theory Turin's theorem' 333
which contradicts the fact that the size of G is at least Theorem 11~3

Let nand p be integers with 2 ::::; n ::::; p. Every graph of order p and size at least
(~)n2+L
2r -2 m(p, n -1) + 1 contains Kn as a subgraph. Furthermore, the onlyKn-free graph
of order p and size m(p, n -1) is R(p, n - 1).
Thus n - r + 1 ~ r. Since H has order n - r + 1 ~ r and size at least

r-_2) nz+ 1 - (r - 1).-(n - r + l)(r - 2)

(_
Proof
2r-2 2 We proceed by induction on n(~2). For n = 2 and p ~ 2, the graph
R(p, n - 1) = R(p, 1) = Kp; so m(p, 1) = O. Consequently, every graph of
= (r-2 )(n-r+l)2+ 1, orderp and size at least m(p, 1) + 1 contains K z as a subgraph. Certainly,
2r-2
Kp is the unique graph of order p containing no edges. Therefore, the
it follows by the inductive hypothesis that H contains Kr as a subgraph. result is true for n = 2.
Therefore, G contains Kr as a subgraph. D Assume, for n ~ 3, that every graph of order s( ~ n - 1) and size at least
m(s, n - 2) + 1 contains Kn - 1 as a subgraph and that R(s, n - 2) is the only
By Theorem 11.2, it follows that for n ~ r, graph of order 5 and size m(s, n 2) that does not contain Kn- 1 as a
subgraph. For p ~ n, let G be a Kn-free graph of maximum size having
ex(n; Kr )::::; Gr~22)nz. order p.
Let v be a vertex of G such that degG v = .0.( G) = .0.. Since G does not
Also, by Theorem 11.2, for r = 4 and n = 10, every graph of order 10 and contain K n as a subgraph, the subgraph (N(v)) induced by the neighbors
size at least 35 contains K4 as a subgraph. However, every graph of order of v does not contain Kn - l'
10 and size 34 also contains K4 as a subgraph. Hence the bound presented Next we show that .0. ~ n - 1; for suppose, to the contrary, that
in Theorem 11.2 is not sharp for r ~ 4. The exact value of ex(n; Kr) for all .0. ::::; n - 2. Since G has order p and p ~ n, it follows that there is a
integers nand r with n ~ r ~ 2 is due to Tunln [T9]. vertex u(# v) in G such that u is not adjacent to v. Since G is a Kn-free
Prior to presenting this more general result, we introduce some graph of maximum size having order p, the graph G + uv contains a
terminology and notation that will be useful in its proof. A near regular subgraph F isomorphic to Kn. With the possible exception of u and v,
complete multipartite graph is a complete multipartite graph, the all of the vertices of G + uv have degree at most n - 2. However, F is
cardinalities of whose partite sets differ by at most 1. Thus the degree (n -I)-regular and has order at least 3. This produces a contradiction;
set of a near regu.lar complete multipartite graph has at most two so .0. ~ n - I, as claimed.
elements. A near regular complete k-partite graph of order n is unique Since (N(v)) does not contain Kn- 1 as a subgraph, the size of (N(v)) does
and we denote this graph by R(n, k). If q = Ln/kJ is the quotient not exceed the size m(.0., n - 2) of the graph R(.0., n - 2).
obtained when n is divided by k, then, by the division algorithm, Let U = {Ul> uz, ... , Ut} denote the vertex set of the graph G - N[v].
n = qk + r, where 0 ::::; r < k. Necessarily, then, r of the partite sets in Since each vertex Uj (1 ::::; i ::::; t) has. degree at most .0. in G, it follows
R(n, k) containq + 1 vertices, while the remaining k - r partite sets contain that
q vertices. Thus the size of R(n,k) is r(q!l) + (k - r)(i) and the size of IE(G)I ::::; (t + 1).0. + m(.0., n - 2).
R(n,k) is
If, in fact, IE(G)I = (t+l).0.+m(.0.,n-2), then (N(v))=R(.0.,n-2).
Define
G' = R(.0.,n -2) +Kt + 1 •
We denote the size of R(n,k) by m(n,k). Consequently,
Thus, G' is a complete (n -I)-partite graph of order p and size
(t + 1).0. + m(.0., n - 2). Since G'is (n -I)-partite, it does not contain Kn
as a subgraph. Therefore,

The following proof of Turan's theorem is based on one due to A. J. (t + 1).0. + m(l1, n - 2) = !E(G')! ~ IE(G)I ~ (t + 1)11 + m(l1, n - 2).
Schwenk. Consequently, G has size (t + 1)61 + m(.0., n - 2).
334 Extremal graph theory Extremal results on graphs 335
Next we show that G = G', that is, G' is the unique Kn~free graph of We have already seen that the minimum size which guarantees that
order p and size (t + 1).6. + m(.6., n - 2). The degree in G of every vertex every graph of order n contains a cycle is n. Although barely a teenager
of U is .6.; for otherwise IE(G)I < IE(G')I. Moreover, U is independent; at the time, Posa (see Erdos [E4]) determined the minimum size of a graph
otherwise, IE(G)I < IE(G')I. Therefore, U U {v} is independent and G of order n ~ 6 which guarantees that G contains two disjoint cycles.
G = G', as claimed.
Since G = R(.6., n - 2) + Kt + v it follows that G = K t + l ,Pl,P2,...,Pn_z' where
Theorem 11.5
we may assume that PI ~ P2 ~ ... ~ Pn _ 2' It remains only to show that G
is near regular. By the induction hypothesis, R(.6., n - 2) = Kp1 ,Pz"",PH-2 is Every graph of order n ~ 6 and size at least 3n - 5 contains two disjoint cycles.
near regular, so pn-2 ~ P1 + 1. Since v is a vertex of maximum degree in G,
it follows that t + 1 ~ Pl' Proof
Hence, it remains to show that pn-Z ~ t + 2. Suppose, to the contrary,
that Pn-Z ~ t + 3. LetH = Kt + Z,Pl,P2,...,P,_3,Pn_z -1' Thus It suffices to show that every (n, 3n - 5) graph contains two disjoint cycles
for n ;,. 6. We employ induction on n. There are only two (6,13) graphs,
IE(H)I-IE(G)I = (Pn-2 -1) - (t+ 1) ~ 1, one obtained by removing two nonadjacent edges from K6 and the other
which contradicts the defining property of G. Therefore, Pn -z ~ t + 2 and obtained by removing two adjacent edges from K6 . In both cases, the
G = R(p,n -1). 0 graph has two disjoint triangles. Thus, the result is true. for n = 6.
Assume for all k with 6 ~ k < n that every graph of order k and size
3k ~ 5 contains two disjoint cycles. Let G be an (n, 3n - 5) graph. Since

EXERCISES 11.1 L:: degv = 6n -10,

VEV(G)

11.1 Show that every graph of order n ~ 3 and size ln12] + 1 contains P3 there exists a vertex Vo of G such that degvo ~ 5. Assume first that
as a subgraph. Describe the extremal graphs. degvo = 5, and N(vo) = {Vb Vz, ... , vs}. If (N[voJ) contains 13 or more
11.2 Show that every graph of order n ;,. 4 and size n contains 2Kz as a edges, then we have already noted that (N[voD has two disjoint cycles,
subgraph. Describe the extremal graphs. implying that G has two disjoint cycles. If, on the other hand, (N[vo]) con-
tains 12 or fewer edges, then, since deg Vo = 5, some neighbor of vo, say Vt,
11.3 For n ;,. 4, determine ex(n; Kl ,3) and all extremal graphs. is not adjacent with two other neighbors of vo, say Vz and V3' Add to G the
edges V1 V2 and VI v3 and delete the vertex vo, obtaining the graph G'; that
is, G' = G + V1vZ + V1V3 - vo, The graph G' is an (n - 1, 3n - 8) graph and,
by the inductive hypothesis, contains two disjoint cycles Ct and Cz. At
11.2 EXTREMAL RESULTS ON GRAPHS least one of these cycles, say Cl , does not contain the vertex V1 and thus
contains neither the edge VI v2 nor the edge V1 V3' Hence C1 is a cycle of G. If
In this section, we consider a variety of other extremal results in graph Cz contains neither V1VZ nor V1V3, then C1 and Cz are disjoint cycles of G.
theory. By Turan's theorem, ex(n; K 4 ) = m(n,3). Dirac [D7] obtained a If C2 contains V1 V2 but not V1V3, then by removing VIVz and adding Vo, VOV1
related result. and VOV2, we produce a cycle of G that is disjoint from Cl . The procedure is
similar if C2 contains VrV3 but not VtV2' If Cz contains both VtV2 and V1V3,
then by removing v1 from Cz and adding vo, VoVz and VOV3, a cycle of G
Theorem 11.4
disjoint from Ct is produced.
Every graph oforder n ~ 4 and size at least 2n - 2 contains a subdivision of K4 as Suppose next that degvo = 4, whereN(vo) = {Vl,V2,V3,V4}' If (N[voJ) is
a subgraph. not complete, then some two vertices of N(vo) are not adjacent, say V1 and
vz. By addingvlv2 to G and deletingvo, we obtain an (n -1,3n - 8) graph
It has been conjectured by G. A. Dirac that every graph of order n ;,. 5 G', which by the inductive hypothesis contains two disjoint cycles. We
and size at least 3n - 5 contains a subdivision of Ks as a subgraph; may proceed as before to show now that G has two disjoint cycles.
however, it has only been verified, by Thomassen [T3], that every graph Assume then that (N[voD is a complete graph of order 5. If some vertex
of order n ~ 5 and size at least 4n - 10 contains such a subgraph. of V(G) - N[vol is adjacent with two or more neighbors of Vo, then G
336 Extremal graph theory Extremal results on graphs 337
contains two disjoint cycles. Hence we may assume that no vertex of That is, the inequality (11.3) states that the sum of the integers a;b i
V(G) - N[val is adjacent to more than one vertex of N(va). Remove the (1:::;; i:::;; n) is at least n times the product of the averages of alla2,'" ,an
vertices va, VI' V2 from G, and note that the resulting graph Gil has order and of b1 , b2 ,· •• , bn .
n - 3 and contains at least (3n - 5) - (n - 5) - 9 = 2n - 9 edges. How- Suppose then that G is an (n,m) graph with vertex set V(G) =
ever, n ~ 6 implies that 2n - 9 ~ n - 3; so Gil contains at least one cycle {VI' V2,"" v n } such that degv; = di (1 :::;; i :::;; n) and d1 ,d2 , ... ,dn is a non-
C. The cycles C and the cycle Va, VI' V2, Va are disjoint and belong to G. increasing sequence. Then, of course, L~=I d; = 2m. From inequality
Finally, we assume that degva :::;; 3. The graph G - Va is an (n -1, m) (11.3), it follows that
graph, where m ~ 3n - 8. Hence by the inductive hypothesis, G - Va
(and therefore G) contains two disjoint cycles. 0 t
;=1
(d;)=!
2 2
t
;=1
di(d; -1) ~ :!. (2m) (2m - n).
2 n n
(11.4)

To see that the bound 3n - 5 presented in Theorem 11.5 is sharp, Weare now prepared to present the aforementioned result dealing with
observe that the complete 4-partite graph Kl,l,I,n -3 has order n and size 4-cycles (see Lovasz [L6, p. 69]).
3n - 6, and that every cycle contains at least two of the three vertices
having degree n -1. Thus no two cycles of K1,I,I,n-3 are disjoint. Theorem 11.7
For a graph of order n to contain two edge-disjoint cycles, only n + 4
edges are required. This result is also due to Posa (see Erdos [E4]). If G is an (n, m) graph with n ~ 4 and
n+n~
Theorem 11.6
m~ ,4 + 1,
Every graph of order n ~ 5 and size at least n + 4 contains two edge-disjoint then G contains a 4-cycIe.
cycles.
Proof
Theorem 11.1 could very well be interpreted as a result concerning
cycles rather than a result concerning complete graphs. From this point Suppose that G is an (n, m) graph with n ~ 4 that contains no 4-cycle. For a
of view, we know that if G is a graph of ordern ~ 3 and size m, where vertex V of G, the number of distinct pairs of vertices that are mutually
m ~ (n 2 /4) + I, then G contains a 3-cycle. We now turn to 4-cycles. adjacent to V is (dey). However, since G contains no 4-cycles, each pair of
In order to present a proof of the next result, it is convenient to be vertices that are mutually adjacent to another vertex is counted exactly
acquainted with an inequality involving nonincreasing sequences of once in the sum LVEV(G)(dey). Hence "
integers. Let all a2, ... ,an and b1 , b2 , ••• , bn be nonincreasing sequences
of integers. Then L
VEV(G)
(degV):::;;
2
(n).
2
(11.5)

L (a; - a;)(b; - b;) ~ O. (11.1)

Denote the degrees of the vertices of G by d1 , d2l .. . , dn . Then
1", i<j '" n
By rearranging the terms in (11.1), we arrive at
n
L (deg v) = t (d;) ~ :!. (2m) (2m - n),
2 ;=1 2 2 n n
(n -1) La;b; ~ L a;bj • (11.2)
VEV(G)

;=1 1 ",;i'j",n where the inequality follows from (11.4). Combining (11.4) and (11.5), we
Adding L7=1 aib; to both sides of (11.2), we obtain have
m(2m -
n) n(n -1)
~.~ 4 :::;; 2 (11.6)
n ta;b; a;b; = ( t a;) (tb;)
1=1 1~1J/~n . ,1=1 1=1 Solving inequality (11.6) for m gives us
or, equivalently,
n+n~
m:::;; 4 '
(11.3)
which completes the proof. 0
338 Extremal graph theory Extremal results on graphs 339
Somewhat fewer edges guarantee that a graph contains either a 3-cycle Theorem 11.9
or a 4-cycle. Let k and n be integers with 1 ~ k < n. Every graph of order n and size at
least
Theorem 11.8
If G is an (n, m) graph with n ~ 4 and
(k-l)n- G)+1
nJn=l contains a subgraph with minimum degree k.
m~ 2 + 1,
then G contains a 3-cycle or a 4-cycle. Proof
We proceed by induction on n ~k + 1. First, assume that n = k + 1. Let G
Proof be a graph of order n and size at least

Suppose that G is an (n, m) graph with n ~ 4 that contains no 3-cycle or

4-cycle. We proceed as in the proof of Theorem 11.7. SinceG contains no
4-cycles, each pair of vertices that are mutually adjacent to another
vertex is counted exactly once in the sum I:vEV(G)(der). Since G contains Then G = K n = K k + 1 and so G itself is a graph with minimum degree k.
no 3-cycles, each pair of vertices that are mutually adjacent to another Assume that every graph of order n - 1 ~ k + 1 and size at least
vertex are themselves not adjacent. Hence

L: (degV) ~ (n)_m.
(k - 1) (n - 1) - G) + 1
(11.7)
VEV(G) 2 2 contains a subgraph with lninimum degree k. Let G be a graph of order n
and size m, where
Denote the degrees of the vertices of G by d1 ,d2 , ... ,dn . Applying the
inequality (11.4), we obtain
m ~(k - l)n - G) + 1.
L:
nEV(G)
(degV)=
2
t(d2;=1
i
) ~ !!-
2
(2m)
n
(2mn-n). (11.8) We show that G contains a subgraph with minimum degree k. If G itself is
not such a graph, then G contains a vertex v with deg v ~ k - 1. Then the
Combining (11.7) and (11.8), we have order of G - v is n - 1 and its size is at least

2
m ~
n2 (it-l)
4 '
m - degv ~ (k -1)n - G) + 1 - (k -1) = (k -1)(n -1) - (~) + 1.
which yields the desired result. D By the induction hypothesis, G - v, and therefore G as well, contains a
subgraph with minimum degree k. D
Letting n = 5 in Theorem 11.8, we find that if the size of a graph G of
order 5 is at least 6, then G contains a 3-cyde or a 4-cycle. This cannot be The bound given in Theorem 11.7 cannot be improved. If k = 1, then
improved because of the 5-cycle. For n = 10 it follows that if a graph G of the graph K:;" contains no subgraph with lninimum degree 1. More
order 10 has at least 16 edges, then G has a 3-cyde or a 4-cycle. This too generally, for 2 ~ k < n, the graph Kn - k + 1 + Kk - 1 has order n and size
cannot be improved since the Petersen graph has order 10, size 15, and (k -1)n - (~) but contains no subgraph with minimum degree k. For
contains no 3-cydes or 4-cycles. n = 9 and k = 4, the graph Kn -k+1 + Kk - 1 is shown in Figure 11.1.
We now turn to the problem of determining the number of edges that a . ByTheorem 3.8, if G is a graph such that 6(G) ~ k for some positive
graph G of order n must have to guarantee that G contains a subgraph integer k, then Gcontains every tree of size k as a subgraph. Combining
with a specified minimum degree. this result with Theorem 11.9 gives us the following corollary.
340 Extremal graph theory Extremal results on graphs 341
vertex of U that is adjacent to u or v (or both). Let a = IAI and b = IBli so
a + b = k-1.
Now observe that if X1Yl and xzYz are edges of A, where each of Yl and
Yz is adjacent to at least two vertices of U, then X1XZ If. E(G)i for otherwise
G contains an M-augmenting path. Note that for every two edges el and ez
of A, at least one vertex incident with el is not adjacent to a vertex incident
with ez; so the size of (A) is at most (~) - (~). For each edge e of A, there is
a vertex incident with e that is adjacent to no vertex of U. Thus the number
of edges joining a vertex of (A) and a vertex of U is at most a(n - 2k + 2).
Figure 11.1 A (9,21) graph containing no subgraph with minimum degree 4.
Therefore, the number of edges that are incident with two vertices oiA or
Corollary 11.10 with a vertex of A and a vertex of U is at most

Let k and n be integers with 1 ~ k < n. If G is a graph of order n and size at (~) -(;)+a(n - 2k+2).
least

(k-l)n- G)+l, Next observe that if xlYl and xzYz are edges of A, where Yl and yz are
adjacent to two or more vertices of U, then Xl and Xz are not adjacent to
distinct incident vertices of an edge uv of B; for otherwise G contains an
then G contains every tree of size k as a subgraph. M-augmenting path. Moreover, if u is adjacent to Xl' say, then v is not
adjacent to any vertex of U, for once again an M-augmenting path results.
In 1959 Erdos and Gallai [EG2] determined the maximum number of This implies that the number of edges of G incident with at least one
edges in a graph of order n ~ 2k that contains no matching of size k; that vertex of B is at most
is, for n ~ 2k, they determined ex(n;kKz). In 1968 Moon [MIa] obtained a
more general result. It is Moon's proof of the formula for ex( n; kKz) that (~)+ (2b)a +ba +2b.
we present here.
Since every edge of G is one of the types described above, it follows that'
TheoJ'em 11.11
m ~ (~)-(;)+(~)+a(n - 2k+2) + 3ab+ 2b
For positive integers nand k with n ~ 2k,
= (2k - 1 ) a(2n - 5k + 2) _ ab
ex(n; kKz ) = max { (k - l)n - G), (2k; 1 ) }. 2 + 2 2

~ Ck ;l)+ a(2n-;k+2) (11.9)

Proof
Let G be a graph of order n and size m containing a maximum matching =(k-l)n- G)_b(2n-;k+2). (11.10)
M of size k - 1, and let U denote the set of vertices of G that are incident
with no edge of M. Since M is a maximum matching, U is an independent e
1f2n - 5k + 2 = a, then, of course, k;l) = (k -1) - (~). If2n - 5k+ 2> 0,
set of vertices. Moreover, U does not contain two vertices that are adjacent then (11.9) and (11.10) attain their maximum value when b = Oi while if
to distinct vertices of V (G) - U that are joined by an edge of M; otherwise, 2n - 5k + 2 < 0, then (11.9) and (11.10) attain their maximum values when
G contains an M-augmenting path, which is impossible since M is a a = O. Thus.
maximum matching (Theorem 9.2).
We now partition the edges of M into two subsets A and B. The set A
consists of those edges xy of M such that one of x andy, say y, is adjacent and so
to at least two vertices of U. The vertex xof each such edge xy of A is
therefore not adjacent to any vertex of U. The set B then consists of the
remaining edges of M. Hence if uv is an edge of B, then there is at most one
ex(n;kK2) ~ max{ (k -l)n - G), C'k; 1)}.
342 Extremal graph theory Cages 343
The graph K2k - 1 UK n -2k+1 has order n, size (2k;11 and a maximum 11.3 CAGES
matching of size k -1. Moreover, the graph Kk - 1 + Kn - k + 1 has order n,
size (k -1)n - (~), and a maximum matching of size k -1. Therefore, We close this chapter with a different type of extremal topic. Recall that the
length of a smallest cycle in a graph G that contains cycles is called the girth
ex(n;kK2 ) ~ max { (k -l)n - G), (2k; 1)}, of G which we denote by g(G). Therefore, g(Kn) = 3 for n ~ 3, g(Ks,l) = 4
for 5, t ~ 2, andg(C n ) = n for n ~ 3. We are interested in the smallest order
of an r-regular graph of girth g for given integers rand g. Thus, for
which yields the desired result. 0 positive integers r ~ 2 and g ~ 3, we define f(r,g) as the smallest positive
integer n for which there exists an r-regular graph of girth g having order n.
An equivalent statement of Theorem 11.11 is that for a positive integer
The r-regular graphs of order f(r,g) with girth g have been the object of
k, every graph of order n ~ 2k and size at least many investigations; such graphs are called (r,g)-cages. The (3,g)-cages

max{l+(k-l)n- G),I+ Ck;I)} are commonly referred to simply as g-cages. We introduce the notation
[r,g]-graph to indicate an r-regular graph haVing girth g. Thus, an (r,g)-
cage is an [r, g]-graph; indeed, it is one of minimum order.
contains a matching of size k. From Corollary 11.10, we know that if Gis It is clear that f(r,g) ~ max{r + l,g}. Thus, f(2,g) = g since Cg is a
a graph of order n > k ~ 1 and size at least (~-I)n - (~) +~, t~en G 2-regular graph with girth g. Likewise, fer, 3) = r + 1 since Kr+1 is an r-
contains every tree of size k. Brandt [BI4] obtamed a generalIzation of regular graph haVing girth 3. In fact, the complete graph K4 is the
both Corollary 11.10 and Theorem 11.11. unique 3-cage. A lower bound for the order of any [r,g]-graph is
presented next (see Holton and Sheehan [HS3, p. 184]). For r,g ~ 3, we
define
Theorem 11.12
r[(r-l)(g-1)/2_1]
Let n and k be positive integers with n ~ 2k. Every graph of order n and size at if g is odd
1+ r-2
least fo(r,g) = { 2[(r _ 1)g/2 -1]
maX{I+(k-l)n- G),l+ C k
;I)} r-2
if g is even.

contains every forest of size k without isolated vertices as a subgraph. Theorem 11.13
If G is an [r,g]-graph of order n, then n ~fo(r,g).

EXERCISES 11.2
Proof
11.4 Prove Theorem 11.4. First, suppose that g is odd. Then g = 2k + 1 for some positive integer k.
Let v E V(G). For 1 ~ i ~ k, the number of vertices at distance i from v is
11.5 For n ~ 9 determine the smallest positive integer m such that r(r - l)i-l. Hence
every graph of order n and size m contains three pairwise disjoint
cycles. n ~ 1 + r + r(r -1) + r(r _1)2 + ... + r(r -ll- 1
11.6 Let nand k be positive integers such that n ~ (5k - 2) /2. Prove that r[(r-l)k-l]
if G is a graph of order n and size at least (k -1)n - (~) + 1, then G - +-':":"_-'---"
-.1 r-2'
contains every forest of size k and without isolated vertices as a
Next, suppose thatg is even. Theng = 2(, where (~2. Lete = uv E E(G).
subgraph.
For 1 ~ i ~ ( - 1, the number of vertices at distance i from u or v is
11.7 Let G be a graph containing a sUbgraph H of order at least 2k 2(r _1)i. Thus
such that 8(H) ~ k. Prove that G contains every forest of size k
and without isolated vertices as a subgraph. n~2+2(r-l)+2(r'-1)2+... +2(r-l)'''-1 =2[(r-1/
. r~2
-1]. 0
344 Extremal graph theory Cages 345
Consequently, if for integers r, g ~ 3, there exists an (r,g)-ca.ge, then Denote by W the set of all those vertices W of G such that d( u, w) :::; g - 2
f(r,g) ~ fo(r,g). We now show that for every pair r, g.~ 3 of mt.egers, or d(v, w) :::; g -1. From our earlier remark, it follows that u, v E W. The
there is at least one (r,g)-cage. The proof of the followmg result IS due number of vertices different from u at a distance at most g - 2 from u
cannot exceed
to Erdos and Sachs [ES1].

Theorem 11.14
g-Z
.E(r _1)i =
(1)
r = ',[(r _1)g-Z - 1];
i=l r 2
For every pair r, g of integers at least 3, the number f (r, g) exists and, i}1fact,
while the number of vertices different from v and at a distance at most
f(r,g) :::; G=~)l(r _1)g-1 + (r - W- z
+ (r - 4)]. g - 1 from v cannot exceed

Proof
g-l . (
2:)r _1)' =
i=l
r -1
)
-=-
r
1
[(r -1)g- - 1].
2 .
Since Hence the number of elements in W is at most
-1)
?=(r-1/= ( ~-2 [(r-1)g-1_1]
g-I r-1) - z · (r-1) -1
( -r-2 [(r -1)g -1] + -r-2 [(r -1)g -1];
,=1

and however,
g-Z . (r - 1)
?=(r-1)'= r-2 [(r-1)
g-z]
-1, r -1) [(r _1)g-1 + (r _1)g-z - 2] = n- r + 1 < n.
( r-2
,=1

it follows that Therefore, there is a vertex WI E V(G) - W, so d(u, WI) ~ g -1 and

d(v, WI) ~ g.
r
( r-2
-1)
[(r _l)g-l + (r _1)g-Z + (r - 4)] Since d(u, WI) > d(G) and u E M(G), it follows that WI rt M(G) and
degwl = r ~ 3. Therefore, there exists an edge e incident with WI whose
is an integer. Denote this integer by n, and let S be the set of all graphs H removal from G results in a graph having girth g. Suppose that e = WI wz.
of order n such that g(H) = g and ~(H) :::;r. Note that n ~ g. T~e set S Clearly, d(v, wz) ~ g -1, so Wz ~ M(G) and degwz = r.
is nonempty since the graph consisting of a g-cyc1e and n - g Isolated We now add the edge UWl to Gand delete the edge WIWZ' producing
vertices belongs to S. For each H E S, define the graph G1 . The graph G1 also belongs to S and, in fact, belongs to Sl'
M(H) = {v E V(H) I degv < r}. The set M(G 1) contains all the members of M(G) except possibly u and,
in addition, contains Wz. From the manner in which G was chosen,
If for some H E S, M(H) = 0, then we have the CJ.esi~ed result; thus we IM(G1 )1 :::; IM(G)I; so u.~ M(G1 ) and IM(G1 )1 = IM(G)I. Therefore,
assume for all H ES,M(H) =f:. 0. For H E S, we define d(H) to be ~he degu = r in GIt implying that, in G, degu = r -1. Furthermore, G1
maximum distance between two vertices of M(H). (We define belongs to Sz.
d(Ul' uz) = +00 if Ul and Uz are not .c~nnected.) . We now show that u is not the only vertex of M(G), for suppose that it
Let Sl be those graphs in S contammg the maXlmum num?er of edge~, is. Since there is an even number of odd vertices, we must have r arid n
and denote by Sz the set of all those graphs H of SI for which IM(H) I ~s odd; however, this cannot occur since n is even when r. is odd. We con-
maximum. Now among the graphs of Sz, let G be chosen so that d(G) IS clude that u and v are distinct vertices ofM(G).
maximum. ' . The vertices v and Wz are distinct vertices of M(Gl)' If there exists no
Let u, v E M(G) such thatd(u, v)= d(G). Suppose t~at d(G) ~ g -1 ~ 2. v-wz path in G1, then d(G1 ) = +00, and this is contrary to the fact that
By adding the edge uv to G, we obtain a graph G of order n havmg d( G1 ) :::; d( G). Thus v and Wz are connected in G1 . Let P be a shortest V-Wz
g(G') = g and ~(G'):::; r. Hence G' E S; however, G' has more edges path in G1 . If P is also in G, then P has length at least dc(v,wz)in G, but
than G, and this produces a contradiction. There~or:, d(G) .~ g - 2 and
dc(v, wz) ~ g - 1 > d(G),
d(u, v):::; g 2. (The vertices u and v may notbedlStmct.)
346 Extremal graph theory Cages 347
VI Proof
As mentioned earlier, it is not difficult to show that the Petersen graph is a
5-cage. In order to see that it is unique, assume that G is a [3, 5]-graph of
V5 order 10. We show that G is isomorphic to the Petersen graph.
Let Vj E V(G), and suppose V2' V3 and V4 are the vertices adjacent to VI'
Since g( G) = 5, each vertex Vj, i = 2,3,4, is adjacent to two new vertices of
G. Let Vs and V6 be adjacent with Vz; V7 and Vs with V3; and V9 and vlO with
V4' Hence V(G) = {Vi I i = 1,2, ... , 10}. The fact that the girth oiG is 5 and
that every vertex of G has degree 3 implies that Vs is adjacent with one of
V4 V3 V7 and Vs and one of V9 and VlO' Without loss of generality, we assume Vs to
Figure 11.2 The Petersen graph: the unique 5-cage. be adjacent to V7 and V9' We must now have V6 adjacent to Vs andvlO'
Therefore, the edges V7VlO and VSV9 are also present and no others.
which is impossible. If P is not in G, then P contains the edge UWI and a Thus, G is isomorphic to the Petersen graph. 0
u-v path of length dc(u, v) as a subpath. Hence P has length exceeding
dc(u, v) = d(G), again a contradiction. Since fo(r, 5) = rZ + I, a Moore graph of girth 5 has order r2 + 1. We
It follows for some H in S that M(H) = 0, that is, H is an r-regular have seen that the Petersen graph is the only cubic Moore graph of
graph of order n having girth g. 0 pirth 5. However, the Petersen graph is one of only two (or possibly
three) Moore graphs of girth 5. This fact was established by Hoffman
We now determine the value of the number f(r, 4). and Singleton [HS2]. We omit this proof.

Theorem 11.15 Theorem 11.17

For r ~ 2, f(r, 4) = 2r. Furthermore, there is only one (r,4)-cage; namely, Kr,r' If G is an r-regular Moore graph (r ~ 3) ofgirth 5, then r = 3, r = 7 or, possibly,
r= 57.

Proof A graph referred to as the Hoffman-Singleton graph is the 7-regular

By Theorem 11.13, f(r,4) ~ 2r. Obviously, the graph Kr,r is r-regular, has Moore graph of girth 5 (and of order 50). Its construction is described
girth 4, and has order 2r, thus implying that f(r, 4) = 2r. in Holton and Sheehan [HS3, p. 202].
To show that Kr,r is the only (r, 4)-cage, let G bean [r, 4]-graph of order
2r, and let UI E V(G). Denote by VI' Vz, ... , V r the vertices oiG adjacent
with UI' Since g(G) = 4, VI is adjacent to none of the vertices Vj, 2 :::;; i:::;; r;
hence G contains r - 1 additional vertices uz, U3,' .• ,ur• Since every vertex
has degree rand G contains no triangle, each vertex Uj (1 :::;; i :::;; r) is
adjacent to every vertex Vj (1 :::;; j :::;; r); therefore, G = Kr,r' 0

An [r,g)-graph of order n is called a Moore graph if n = f(r,g) = fo(r,g).

Hence, a Moore graph is an (r,g)-cage of order fo(r,g). Consequently,
the graphs Kr,T for r ~ 3 are Moore graphs. The best known Moore
graph is the Petersen graph (Figure 11.2). It is not difficult to verify
that the Petersen graph is a 5-cage. That it is the only 5-cage is verified next.

Theorem 11.16
The Petersen graph is the unique 5-cage. Figure 11.3 The 4-regular Moore graph of girth 6.
348 Extremal graph theory Cages 349
We now summarize the information concerning Moore graphs; namely,
r-regular Moore graphs of odd girth g exist when
• g = 3, r ~ 3, and Kr +1 is the unique Moore graph;
• g = 5, r= 3, and the Petersen graph is the unique Moore graph;
• g = 5, r= 7, and the Hoffman-Singleton graph is the unique Moore
graph; and
• g = 5 and r = 57 is undecided.
Furthermore, r-regular Moore graphs of even girth g exist when
• g = 4, r ~ 4, and KY,r is the unique Moore graph;
• g = 6 and for all r for which there exists a projective plane oforder r - 1;
• g = 8 and for all r for which there exists a certain projective geometry;
and
• g = 12 and for all r for which there exists a certain projective geometry.
As an additional example of Ii Moore graph referred to above, there is a
unique 4-regular Moore graph of girth 6 and order fo(4, 6) = 26. This
graph is shown in Figure 11.3.
Additional information on Moore graphs can be found in Holton and
Sheehan [HS3, Chap. 6]. We now return to cages that are not necessarily
Moore graphs. The (4,5)-cage and (5,5)-cage are shown in Figure 11.4.
The (6,5)-cage has order 40 while the (7,5):cage, as mentioned earlier,
is known to have order 50.
There is only one 6-cage, referred to as the Heawood graph, and this is
shown in Figure 11.5.
There are only a few known g-cages, g ~ 7. The 7-cage (known as the
McGee graph) and the 8-cage(the so-called Tutte-Coxeter graph) are shown
in Figure 11.6. The 12-cage has order 126.
Some other known values of f(r,g) are shown in Figure 11.7.

Figure 11.4 The (4,5)-cage and (5,5)-cage.

Moore graphs can be considered from a different point of view.

Consider integers r, g ~,3, where.g = 2k + 1 is odd. If G is an r-regul~r
graph of order n and diameter k, then by the proof of ~heorem 11.13, it
follows that fo(r,g) ~ n: Hence a Moore grap~ of odd gIrth ~as the II\ax-
imum order consistent with its degree and diameter constramts and the
minimum order consistent with its degree and girth constraints. Such a
statement also applies to Moore graphs with even girth. Figure 11.5 The Heawood graph: the unique 6-cage.
350 Extremal graph theory
CHAPTER 12

Ramsey theory

Probably the best known and most studied area within extremal graph
theory is Ramsey theory. We begin this study with the classical Ramsey
numbers.
Figure 11.6 The 7-cage\and 8-cage.

g=3 g=4 g=5 g=6 g=7 g=8 g=9 g= 10 12.1 CLASSICAL RAMSEY NUMBERS
r=3 4 6 10 14 24 30 58 70
r=4 5 8 19 26 ? 80 ? ?
For positive integers sand t, the Ramsey number r(s, t) is the least
r=5 6 10 30 42 ? 170 ? ? positive integer n such that for every graph G of order n, either G
r=6 7 12 40 62 ? 312 ? ? contains Ks as a subgraph or G contains Kt as a subgraph; that is, G
r=7 8 14 50 90 ? ? ? ? contains either s mutually adjacent vertices or an independent set of t
vertices. The Ramsey number is named for Frank Ramsey [R1l, who
Figure 11.7 Some known values of f(r,g). studied this concept in a set theoretic fram~ork and essentially verified
the existence of Ramsey numbers. Since (G) = G for every graph G, it
For those values of rand g for which f(r,g) is unknown, the known follows that the Ramsey number r(s, t) is symmetric in sand t and
bounds are not especially close. The closest known bounds for an r(s, t) = r(t, s).
unknown value of f(r,g) is 53 ~ f(4, 7) ~ 728. At the opposite extreme, It is rather straightforward to show that r(s, t) exists if at least one of s
3110 ~ f(7, 10)~ 68 • _ and t does not exceed 2 and that
r(l, t) = 1 and r(2, t) = t.
The degree of difficulty in determining the values of other Ramsey num-
EXERCISES 11.3
bers increases sharply as sand t increase, and no general values like the
above are known.
11.8 Let G be a connected graph with cycles. Show that
It is sometimes convenient to investigate Ramsey numbers from .an
g(G) ~ 2diam(G) + 1.
'edge coloring' point of view. For every graph G of order n, the edge
11.9 (a) Prove that f(3, 6) = 14. sets of G and G partition the edges of Kn • Thus, r(s, t) cap be thought of
(b) Prove that the Heawood graph is the only 6-cage. as the least positive integer n such that if every edge of K n is arbitrarily
colored red or blue (where, of course, adjacent edges may receive· the
11.10 Let G be an [r,g]-graph (r ~ 2,g ~ 3) of order f(r,g); that is, let G be
same color), then there exists either a complete subgraph ()f order s, all
an (r,g)-cage. Prqve that if H = G X K2 is an [s,g]-graph, then H
of whose edges are colored red, or a completesubgraph of order t, all
cannot be an (s,g)~cage.
of whose edges are colored blue. In the first case, we say that there is a
red Ks ; in the second case, a blue Kt • We call the coloring a red-blue coloring
of Kn • For example, for t ~ 2, r(2, t) > t -1 since if all C;l) edges of Kt - 1
are colored blue, then Kt _I contains neither a red K2 nor a blue Kt ; How-
ever, r(2, t) ~ t since in art arbitrary red-blue coloring of Kt , either all the
352 Ramsey theory Classical Ramsey numbers 353
edges are blue and we have a blue K or at least one edge is red and we and that
have a red K2 . Thus, r(2, t) = t. "
Theorem 12.1 gives the value of the first nontrivial Ramsey number r(s-1,t)=:;;; ( S+t-3)
s-2 and r(s,t-1)=:;;; (S+t-3)
s-l .
r(3,3). Since
S+t-3)+(S+t-3)= (S+t-2),
Theorem 12.1 ( s-2 s-1 s-1
The Ramsey number r(3, 3) = 6. we have that

r(s -1,t) + r(s,t -1) =:;;;

S+t-2) . (12.1 )
( s-1
Proof
Since neither Cs nor C\ contains K3 as a subgraph, r(3, 3).;;' 6. Consider Let G be a graph of order r(s -1, t) + r(s, t - n
We show that either G
any red-blue coloring of K 6 and let v be a vertex of K 6 • Clearly, v is contains Ks as a sUbgraph orG contains Kt as a subgraph. Let v E V(G).
incident with at least three edges of the same color. Without loss of We consider two cases.
generality, we assume that vv1f VV2 and VV3 are red edges. If any of V1V2f Case 1. Assume that degG v ;;. r(s ~ 1, t). Thus if 5 is the set of vertices adja-
V1 V3 and V2V3 is a red edge, then there is a red K3 ; otherwise, these three cent to v in G, then either (SIG contains Ks - 1 as a subgraph or (SIG = (5)c
edges ani blue and we have a blue K3 . Thus, r(3,3) =:;;; 6. Combining the .<:9ntains K t as a sUbgraph. If (SIc contains K t as a subgraph, then so does
two inequalities, we have r(3,3) = 6. 0 G. If (SIG contains Ks - 1 as a subgraph, then G contains Ks as a subgraph
since in G, the vertex v is adjacent to each vertex in S. Hence in this case,
Before proceeding further, we show that all Ramsey numbers exist and, Ks ~ G or Kt ~ G.
at the same time, establish an upper bound for r(s, t), which was discov-
ered originally by Erdos and Szekeres [ES2]. In the proof of Theorem 12.2 Case 2. Assume that degG v < r(s - 1, t). Then iegc v ;;. r(s, t - 1). Thus if
we use the definition of the Ramsey number directly, rather than the T denotes the set of vertices adjacent to v in G, then ITI ;;. r(s, t - 1) and
equivalent edge coloring point of view. either (T)G contains Ks as a subgraph or (T)c; contaiE.s Ks -1 as a subgraph.
It follows, as in Case 1, that either K s ~ G or K 1 ~ G.

Theorem 12.2 Since G was an arbitrary graph of order r(s -1, t) + r(s, t -1), we
conclude that r(5, t) exists and that
For every two positive integers sand t, the Ramsey number r(s, t) exists;
r(s, t) =:;;; r(5 - 1, t) + r(s, t -1). (12.2)
moreover,
Combining (12.1) and (12.2), we obtain the desired result. 0
S+t-2)
r(s,t)=:;;; ( s-1 .
The proof of Theorem 12.2 gives a potentially improved upper bound
for r(5, t). This is stated next, together with another interesting fact.
Proof
We proceed by induction on k = s + t. Note that we have equality for s = 1 Corollary 12.3
or s = 2, and arbitrary t; and for t = 1 or t = 2, and arbitrary s. Hence the
For integers s ;;. 2 and t ;;. 2,
result is true for k =:;;; 5. Furthermore, we may assume that s ;;. 3 and t ;;. 3.
Assume that r(s', t') exists for all positive integers s' and t' with r(s, t) =:;;; r(s - 1, t) + r(s, t -1). (12.3)
s' + t' < k, where k ;;. 6, and that Moreover, if r(s -1, t) and r(5, t '-1) are both even, then 5trictinequality holds
in (12.3).
"
r(s,t)=:;;; (S'+t'-2)
s'-l .
Proof
Let sand t be positive integers such that s + t = k, s ~ 3 andt ~ 3. By
the inductive hypothesis, it follows that r(s -1, t) and r(s, t - 1) exist, The inequality in (12.3) follows from the proof of Theorem 12.2,
354 Ramsey theory Classical Ramsey numbers 355
In order to complete the proof of the corollary, assume that res - 1, t)
and res, t -1) are both even, and let G be any graph of order
res - 1, t) + res, t -1) -1. We show that either G contains Ks as a sub-
graph or G contains Kt as a subgraph.
Since G has odd order, some vertex v of G has even. degree. If G:
degG v;;, res - 1, t), then, as in Case 1 of Theorem 12.2, either G contains
Ks as a subgraph Or G contains K1 as a subgraph. If, on the other hand,
degG < res -1, t), then degG v ~ res - 1, t) - 2 since degG v and res ~ 1, t)
are both even. But then degev ;;. res, t -1), and we may proceed as i.p.
Case 2 of Theorem 12.2. 0

As we have already noted, the bound given in Theorem 12.2 for res, t) Figure 12.1 An extremal graph showing r(3,S) "" 14.
is exact if one of sand t is 1 or 2. The bound is also exact for s = t = 3.
By Theorem 12.2, According to Theorem 12.4, r(3,4) ~ 9 and r(3,5) ~ 14. Actually,
P+t equality holds in both of these cases. The equality r(3,5) = 14 follows
r(3, t) ~ . since there exists a graph G of order 13 containing neither a triangle nor
an independent set of five vertices; that is, K 3 ~ G and Ks ~ G. The graph
An improved bound for r(3, t) is now presented. G is shown in Figure 12.1.
Theorem 12.2 gives an upper bound for the 'diagonal' Ramsey number
Theorem 12.4 r(s,s), namely res,s) ~ e:.:::-12 ). There are three ways in which lower
bounds for res, s) have been obtained: the constructive method, a counting
For every integer t ;;. 3,
method and the probabilistic method. In the constructive method, a lower
t 2 +3 bound Jor r(s,s) is established by explicitly constructing a graph G of an
r(3,t) ~ - 2 - ' (12.4) appropriate order such that neither G nor G contains K as a subgraph.
Better lower bounds, however, have been obtained using a counting
Proof method, which we describe here briefly: (The probabilistic method will
be. discussed in Chapter 13.) Suppose that we wish to prove the
We proceed byinductionont. Fort = 3, r(3, t) = 6 while (t2 + 3)/2 = 6,so eXIstence of a graph G of order n having some given property P. If we
that (12.4) holds if t == 3. Assume that r(3, t -1) ~ ((t -If
+ 3)/2, for can estimate the number of graphs of order n that do not have property
some t ;;. 4, and consider r(3, t). By Corollary 12.3, P and we can show that this number is strictly less than the total number
r(3,t) ~ t +r(3, t -1). (12.5) of graphs of order n, then there must exist a graph G of order n having
property P. Of course, this procedure offers no method for constructing G.
Moreover, strict inequality holds if t andr(3, t -1) are both even. In 1947, in one of the first applications of a counting method, Erdos [E2]
Combining (12.5) and the inductive hypothesis, we have established the follOWing bound.
(t - 1)2 + 3 t2 + 4
r (3 ,t) ~t+ 2 =-2-' (11.6)
Theorem 12.5
To complete the proof, it suffices to show that the inequality given in For every int~ger t ;;. 3,
(12.6) is strict.
If t is odd, then r(3, t) < (t2 + 4)/2 since t2 + 4 is odd. Thus we may
assume that t is even. If r(3, t - 1) < ((t _1)2 + 3)/2, then clearly the
inequality in (12.6) is strict. If, on the other hand, r(3, t -1) =
Proof
((t-1)2+ 3 )/2=t2/2-t+2, then r(3,t-1) is even since t is even.
/2
Therefore the inequality in (12.5) is strict, which implies the desired Let n = li j. We ~monstrate the existence of a graph G of order n such
result. 0 that neither G nor G contains Kt as a subgraph.
356 Ramsey theory Generalized Ramsey theory 357
There are 2(;) distinct labeled graphs of order n with the same vertex set Thus, r(4,4) > 17. Show that r(4,4) =18.
V. For each subset 5of V with 151 = t, the number of these graphs in which
5 induces a complete graph is 2 Gl - (~). Thus, if M denotes the number of
graphs with vertex set V that contain a subgraph isomorphic to K t , then
t
M ~ (n)2G)-(D < n 2Gl-(~). (12.7)
"" t t!
2
By hypothesis, n ~ Thus, nt ~ i
t2 2
i/2. 2
/ • Since t;?; 3, we haye 2 / <
mt!2(~), and so

nt < G)t! 2(~). (12.8)

Combining (12.7) and (12.8), we conclude that

M < (~)2(D. 12.6 The value of the Ramsey number r(5, 5) is unknown. Establish
If we list the M graphs with vertex set V that contain a subgraph iso- upper and lower bounds (with explanations) for this number.
morphic to Kt, together with their complemen!s, then there are at most
2M < 2Gl graphs in the list. Since there are 2(2) graphs with verte~ set
V, we conclude that there is a graph G with vertex set V such that nelth~
12.2 GENERALIZED RAMSEY THEORY
G nor G appears in the aforementioned list, that is, neither G nor G
contains a subgraph isomorphic to Kt · 0
For positive integers S1 and S2, the classical Ramsey number r(s1, S2), dis-
cussed in section 12.1, may be defined as the least positive integer n such
By Theorems 12.2 and 12.5, we have 4 < r(4, 4) ~ 20. Actually
that for any factorization Kn = F1 EtJ Fz (therefore, Fz = 1\), either KS1 s;;: F1
r(4,4) = 18 (Exercise 12.5); in fact, the only known Ramsey numbers
or Ks2 s;;: F z· Defining the Ramsey number in this manner suggests a variety
r(s, t) for 3 ~ s ~ tare
of interesting generalizations. In this section we consider a sample of the
r(3,3) = 6 r(3,6) = 18 r(3,9) = 36 many directions of investigation in the field of Ramsey theory.
r(3,4) = 9 r(3,7) = 23 r(4,4) = 18 Let Gll Gz,· .. , Gk (k;?; 2) be graphs. The (generalized) Ramsey number
r(3,5) = 14 r(3, 8) = 28 r(4,5) = 25. r(G ll Gz,···, Gk ) is the least positive integer n such that for any factoriza-
tion
Kn = F1 EBFz EB··· EtJFk>
EXERCISES 12.1 the graph Gi is a subgraph of Fi for at least one i = 1,2, ... ,k. Hence,
r(KSl ,1<",) = r(s1' sz). Furthermore, we denote r(KS1 ' KSz , ••• ,KSk ) by
12.1 Show that r(s, t) = r(t, s) for all positive integers sand t. r(s1'sZ" .. , Sk)' The existence of such Ramsey numbers is guaranteed by
the existence of the classical Ramsey numbers, as we now see.
12.2 Show that if G is a graph of order r(s, t) - 1, then
(a) Ks - 1 s;;: GorKt S;;:G, Theorem 12.6
(b) Ks s;;: G or K t -1 s;;: G.
Let the graphs G1', Gz, ... ,Gk (k;;': 2) be given. Then the Ramsey number
12.3 If 2 ~ Sf ~ sand 2 ~ t f ~ t, then prove that r(sf, t f) ~ r(s, t). Further- r( G1 , Gz, ... , Gk ) "exists.
more, prove that equality holds if and only if Sf = sand tf = t.
12.4 Show that r(3, 4) = 9. Proof
12.5 The accompanying graph has order 17 and contains neither four It suffices to show that if 51,52"", 5k are positive integers, then
mutually adjacent vertices nor an independent set of four vertices. r(s1, 52,' .. , 5k) exists; for suppose tliat G1 , Gz, ... ,Gk have orders s1l
358 Ramsey theory Generalized Ramsey theory 359
S2,' .. , Ski respectively, and that r(SI' S2,' .. ,Sk) exists. If Fl EB F2 EB . ". EB Fk is Let F be any graph of order 1 + (s - l)(t - 1). We show that Ts ~ F or
any factorization of the complete graph of order r(sl' S2,' .. , Sk), then Kt ~ F, implying that r(Ts , Kt ) ~ 1 + (s - 1) (t - 1) and completing the
Ks; ~ F; for some i, 1 ~ i ~ k. Since Gi ~ KSi , it follows that G; ~ h Thus proof. If Kt is not a subgraph of "P, then fJ(F) ~ t - 1. Therefore, since F
r(G l ,G2, ... ,Gk) exists and r(G l ,G2,··· ,Gk) ~. r(sl,s2,'" ,Sk)' has order 1 + (s -1)(t -1) and f3(F) ~ t - I, it follows that X(F) ~ 5
We proceed by induction on k, where r(s1> S2) exists for all positive (Exercise 8.1). Let H be a subgraph of F that is critically s-chromatic. By
integers SI and 52 by Theorem 12.2. Assume that r(sl,52,'" ,Sk-l) exists Corollary 8.3, 6(H) ~ S - 1. Now applying Theorem 3.8, we have that
(k ~ 3) for any k - 1 positive integers SI, S2, ... , Sk-lI and let SI, S2, ... ,sk Ts ~ H, so that Ts ~ F. 0
be k positive integers. We show that r(sl, S2,' .. , Sk) exists. By the inductive
hypothesis, r(S1>S2"" ,Sk-'l) exists; say r(sl,s2,'" ,Sk-l) = no. Let For k ~ 3, the determination of Ramsey numbers r(G1> G2 , ... , Gk) has
r(no, Sk) = n. We now verify that r(s1> 52," . , 5k) ~ n; thereby establishing proved to be quite difficult, for the most part. For only a very few classes
the required existence. of graphs has any real progress been made. One such example, however,
Let Kn = Fl EB F2 EB ... EB Fk bean· arbitrary factorization of Kn into k is where each Gi , 1 ~ i ~ k, is a star graph. The following result is by Burr
factors. We show that Ks; ~ F; for.at least one i, 1 ~ i ~ k. Let and Roberts [BR3].
H = Fl EB F2 EB ... EB Fk- l ; hence Kn = H EB Fk. Since r(nO,sk) = n, it follows
that Kno ~ H or KSk ~ Fk. Theorem 12.8
Suppose that K no ~ H. Let V o be a set of no mutually adjacent vertices
of H, and define F; = (Voh; for i = 1,2, ... ,k -1. Since H = Let SI, 52, ... ,sk (k ;;:. 2) be positive integers, t of which are even. Then
Fl EB F2 EB ... EB Fk - lI it follows that K no = F~ EB F~ EB ... EB FL l' However, k
r(sl,s2, ... ,sk_l)=nO; so Ks;~F; for some i, 1~i~k-1. Because r(Kl,Sl,Kl,S2"" ,Kl,sk) = 2:)s; -1) + O/>
F; ~ F; for all i, 1 ~ i ~ k - 1, the graph Ks; is a subgraph of F j for at ;=1
least one i, 1 ~ i ~ k - 1. where 0t = 1 if t is positive and even and 0t == 2 otherwise.
Hence, we may conclude that K s; ~ F i for some i, 1 ~ i ~ k. D .

While it is known that r(3, 3, 3) = 17, no other nontrivial numbers of the Proof
type r(51, S2, ... ,Sk), k ~ 2, have been evaluated exceptthose mentioned in Let r(Kl ,Sl' K1,sz"'" K1,Sk) = n, and let E~=lsi = N. First, we show that
the preceding section. It may be surprising that there has been consider- n ~ N - k + 0/. Since each vertex of KN -k+2 has degree N - k + i =
ably more success in evaluating the numbers r( G1 , G2 , ••. , Gk ) when not E;=I(Si -1) + 1, any factorization
all the graphs Gi are complete.· One of the most interesting results in this
direction is due to Chvatal [C8], who determined the Ramsey number K N __ k+2 = FI EB F 2 EB ... Efj Fk
r(Ts' Kt ), where Ts is an arbitrary tree of order s. This very general necessarily has K1,s; ~ Fi for at least one i, 1 ~ i ~ k. Thus, n ~ N - k + 2.
result has a remarkably simple proof. To complete the proof of the inequality n ~ N - k + 0/, it remams to show
that n ~ N - k + 1 if t is positive and even. Observe that, in this ease,
Theorem 12.7 N - k + 1 is odd. Suppose, to the contrary, that there exists a factorization
Let Ts be any tree of order s ~ 1 and let t be a positive integer. Then KN - k + 1 = F1 EBF2 EB··· EBFk
r(Ts,Kt ) = 1 + (s -1)(t -1). such that K1,s; is not a subgraph of Fi for each i = 1,2, ... ,k.Since each
vertex of KN -k+ 1has degree N - k = 2:;=1 (S; ~ 1), this implies that Fi is
an (Si - 1)-factor of KN -k+ 1 for each i = 1,2, ... , k. However, N - k + 1 is
Proof
odd and nj - 1· is odd for some j (1 ~ j ~ k); thus, Fj contains an odd
For s = lor t = 1, r(Ts, Kt ) = 1 = 1 + (s - l)(t - 1). Thus, we may assume number of odd vertices, which is impossible.
that s ;;:. 2 and t ;;:. 2. Next we show that n ~ N - k + 0t. If t = 0, then each integer Si is odd as
The graphF = (t -l)Ks _ l does not contain Ts as a subgraphsince each is N - k + 1. By Theorem 9.21, the complete graph KN -k+ 1 can be factored
component of F has order s -1. The complete (t -I)-partite graph iI1to (N - k)j2 hamiltonian cycles. For each i = 1,2, ... , k, let F; be the
F = K(s -1,s -1, ... ,s -1) does not contain K t as a subgraph. Therefore, union of (Si -1)/2 of these cycles, so F; is an (s; -I)-factor of KN-kH
r(Ts,K t ) ~ 1 + (s -1)(t -1). Hence there-exists a factorization KN - k+ 1 = F1 EBF2 EB··· EBFk such that
 Ramsey theory Generalized Ramsey theory 361
360
Kl ,s; is not a subgraph of h for each i (1 ;;;; i ;;;; k). This implies that and Lovasz [BELl] have shown that a similar result holds in the case of
chromatic numbers.
n ~ N - k + 2 ift = O.
Assume that tis odd. ThenN - k + 1 is even. By Theorem 9.19, KN -k+l
is I-factorable and therefore factors into N - k I-factors. For i = 1,2, ... , k,
Theorem 12.10
let Fi be the union of Si - 1 of these I-factors, so each F; is an (5; - 1)-factor
of KN- k+ l . Thus, there exists a factorization KN-k+I = FI ffiFz ffi .. · ffiF k For all positive integers 5 and t, if G --> (Ks,K t ), then X(G) ~ X(K r ), where
such that K l ,s; is not a subgraph of Fi for each i. Thus n ~ N - k + 2 if t is r = r(s, t).
odd.
Finally, assume that t is even and positive; and suppose that 51' say, is
even. Then there is an odd number of even integers among 51 -1, Proof
Sz, .•• ,Sk, which implies by the previous remark that The result holds if 5 = 1 or if t = 1. Thus we may assume that s ~ 2 and
n ~ r(KI,Sl-I,KI ,S2'''' ,KI,Sk) ~ N - k + 1. t ;?:. 2, so that r(s, t) :?: 2.
Suppose that X(G) ;;;; r - 1. Since the order of G is at least r, there is an
Hence, in all cases n ~ N - k+ 8t , so that n = N - k+ 0/. 0 (r - I)-coloring of G with resulting color classes UI , Uz,···, Ur - I ·
By definition of r = r(s, t), there is a factorization Kr _ I = FI ffi Fz such
For k = 2 in Theorem 12.8, we have the following. that Ks g FI and K1 g Fz. Label the vertices of Kr -1 as vI, Vz, ... ,V,-I'
We construct a factorization of G as follows. Let V(G I ) = V(G z) = V(G).
Corollary 12.9 Each. edge e of G i~ of the form e = UjUk where Uj E Uj and Uk E Uk
(1 ;;;;J < k;;;; r -1). Smce VjVk is an edge of K r- I, either VjVk E E(FI ) or
Let 5 and t be positive integers. Then VjVk E E(Fz ) .in the factorization Kr - I = FI ffi Fz. Let UjUk E E(G i ) if
s+t-l if 5 and t are both even VjVk E E(F;), I = 1,2. Then G = GI ffi Gz.
rK K -
( I,S,I,I)- { s+t otherwise. Suppose that Ks ~ GI. Thus G1 contains s mutually adjacent vertices,
say WI, Wz, ... ,Ws ' an? there are distinct color classes U i " Uiz , ••• , U;,
For graphs GI , Gz, ... , Gb where k ~ 2, we know (as a result of such that Wj E U;. for J = 1,2, ... ,5. From the way in which GI was con-
Ramsey's theorem) that if G is a complete graph of sufficiently large structed, it follo~s that ({Vi" Vi2 , . . . , v;Jh, = K" which is impossible.
order, then for every factorization G = FI ffi Fz $ ... ffi Fb the graph Gi is Therefore, Ks g; GI . Similarly, Kt g; Gz, so G f> (Ks , Kt ). This is a contra-
a subgraph of F; for at least one i, 1 ;;;; i ;;;; k. This suggests the following. diction. Thus, X(G) ;?:. r, and the proof is complete. 0 .
For graphs G,GI,GZ " " ,Gk (k;?:. 2), we say G arrows GloGZ' ' ' ' ,Gk,
written G --> (G I , Gz, ., . , Gk ), if it is the case that for every factorization
G = F I ffi Fz ffi··· ffi Fk , we have G; ~ F; for at least onei, 1 ;;;; i ;;;; k. The Corollary 12.11
natural problem, then, is to determine those graphs G for which
G --> (G I , Gz,... , Gk) for given graphs GI , Gz , ... , Gk· For all positive integers 5 and t, if G --> (Ks , Kt ), then the size of G is at least (~),
In a few special cases of pairs of graphs GI and Gz, the aforementioned where r = r(s, t).
problem has been solved. In general, however, the problem is extremely
difficult. Therefore most attention has been centered on the case k = 2, For arbitrary graphs GI and Gz, ifG -4 (G I , Gz), then the order of G is at
and, for given graphs GI and Gz, on the properties a graph G can possess least r, where r = r( GI , Gz). However, it is not true ip general that
if G --> (G I , Gz). For example, if G -->(Ks' Kt ) where 5, t ~ 2, then clearly X(G) :?: X(K r ) or that the size of G is at least G), thesj.ze of Kr •
w(G) ~ max(s, t). Folkman [FB] has shown that this is a sharp bound on We note in closing that Graham, Rothschild and Spencer [GRSll have
w(G); specifically, given integers s, t ~ 2, there is a graph G' with clique written a book On Ramsey theory. .
number max(s, t) foJ;' which G' --> (K s , Kt ). Nesetfil and Rodl [NRl) have
extended this result by showing that for any graph Hand integer k ;?:. 2,
there exists a graph G with. clique number w(H) for which
G --> (Hl ,Hz, ... ,Hk), where Hi = H for i = 1,2, ... ,k. EXERCISES 12.2
If G -4 (Ks,K t ), then it is easily seen that the,order of G is at least r(s, t)i
that is, if G -4 (Ks , Kt ), then the n(G) ~ n(Kr ), wherer= r(s, t),Burr, Erdos 12.7 Show that r(3, 3, 3) ;;;; 17.
362 Ramsey theory Other Ramsey numbers 363
12.8 Show for graphs G1, GZ, .•• ,Gk (k;;::' 2) that 12.3 OTHER RAMSEY NUMBERS
r(G 1, Gz,··· , GklKz ) = r(GbG Z" ' " Gk )·
12.9 Show for positive integers t1, t2 , • •• ,tk (k ;;::, 2) that In this section we consider three examples of Ramsey-like numbers.
For positive integers sandt the classical Ramsey number res, t) can be
r(Ktt ,Kt2 , · · · ,Ktk , Ts ) = 1 + (r -l)(s -1), defined as the least positive integer n such that for every graph G of order
where Ts is any tree of order s ;;::, 1 and r = r(tj, tz, ... , tk ). n, either [3(G) ;;::, s or [3(G) ;;::, t, that is, either G contains Ks as a subgraph or
G contains Kt as a sUbgraph. Since every independent set is also an irre-
12.10 Let sand t be integers with s ;;::, 3 and t ;;::, 1. Show that dundant set, this alternative formulation suggests a concept analogous to
_ { 2t + 1 if s is odd and s < 2t + 1 Ramsey numbers for irredundance.
r(CS ,K1,t)- s ifs;;::'2t. The irredundant Ramsey number ires, t) is the least positive integer n such
that for any graph G of order n, either IR(G) ;;::, s or IR(G) ;;::,t.
(Note that this does not cover the case where s is even and s < 2t.) Since IR(G) ;;::, !3(G) for every graph G, it follows that ires, t) exists for all
12.11 Let Gl be a graph whose largest component has order s, and let Gz positive integers sand t and, in fact, ires, t) < res, t). The irredundant
be a graph with x(Gz) =t. Prove that r(G l , Gz) ;;::, 1 + (s -l)(t ~1). Ramsey number is symmetric in sand t, as is the classical Ramsey
number, and, as noted by Brewster, Cockayne and Mynhardt [BCM1],
12.12 Show for positive integers t and t, that r(K{ + Kti Ts ) < t(s -1) + t, recurrence relations analogous to (12.2) hold for irredundant Ramsey
where Ts is any tree of order s ;;::, 1. numbers. The proof of this last fact is left as an exercise.
12.13 Let sand t be positive integers, and recall that a(G) denotes the
vertex-arboricity of a graph G. Determine a formula for a(s, t), Theorem 12.12
where a(s, t) is the least positive integer n such that for any factor-
ization Kn = F1 EB Fz, either a(Fl ) ;;::, s or a(Fz ) ;;::, t. For every two positive integers s ;;::, 2 and t ;;::, 2, the irredundant Ramsey number
ires, t) exists; moreover,
12.14 Show that if G, Gl and Gz are graphs such that G -+ (G1 , Gz), then
the order of G is at least r, where r = r( G], Gz).
ires, t) < ires -1, t) + ires, t -1).
12.15 Prove Corollary 12.11. Certainly, ir(l, t) = 1 and ir(2, t) =t (Exercise 12.21). Our next result
establishes the value of ir(3, 3).
12.16 (a) Let 5 and t be positive integers. Show that if G ....... (K1,s, Kl ,t),
then the size of G is at least s + t - 1.
(b) Give an example ofagraph G for which G -+ (Kl ,s,K1,t) and the Theorem 12.13
size of G is s + t - 1. The irredundant Ramsey number ir(3, 3) = 6.
12.17 (a) Give an example of graphs G, G v Gz such that G -+ (G l , Gz) but
X(G) < X(Kr(ch{;z))· • Proof
(b) Give an example of graphs G, G1, Gz such that G ....... (G l , Gz) but
the size of Gis less than the size of Kr(c t ,c2 ), Since ir(3,3) < r(3, 3) = 6, it suffices to exhibit a graph G of order 5 for
which IR(G) < 3 and IR(G) < 3. Consider G = G = Cs . Assume, to the
12.18 (a) Prove that there exists no triangle-free graph G of order 4k + 3 contrary, that IR(Cs ) ~3. Since [3(Cs ) = 2, it follows that Cs contains an
(k;;::' 1) for which 8(G) ;;::, 2k + 2. irredundant set 5 of cardinality 3 that is not independent. Thus,· (S) == P3
(b) Let Gl , Gz, ... ,Gk + 1 be k + 1(;;::, 2) graphs such that G1 = K3 and or (5) = Pl U Pz. In either case, however, this implies that 5 is redundant,
Gi =K13 for 2<i<k+1. Prove that r(G 1,GZ ,···,Gk + 1) = and produces a contradiction. Thus IR(Cs ) = 2 and, consequently,
4k+3. ' ir(3, 3) ~ 6. D
12.19 (a) Let S11SZ,S3(;;::,2) be integers. Prove that r(sl,s2,S3) <
r(sl' S2,S3 -1) + r(sj,sz -1, S3) + r(sl -1,s2,53) -1. Recall that for a graph G, the independent domination number i(G) is
(b) Generalize the result in part (a). the minimum cardinCllity among all independent dominating sets of G.
Equivalently, i(G) is the minimum cardinality among all maximal inde-
pendent sets of G. So i(G) and [3(G) represent the smallest and largest
364 Ramsey theory Other Ramsey numbers 365
cardinalities of maximal independent sets in G. The clique number w(G) contains a maximal independent set of cardinality at most t, contradicting
of a graph G is the largest cardinality among the complete subgraphs i(G) :;, t + 1. We conclude that no such Gexists and Ir(s, t) :;, s + t + 1. 0
(equivalently, among maximal complete subgraphs) of G and we let
u(G) denote the corresponding smallest cardinality among the maximal Very few exact values oflower Ramsey numbers are known. However,
complete subgraphs of G. The parameter u(G) is called the lower clique Mynhardt [MI3] ob.tained an improved upper bound for t'r(s, t), and this
number. We have already referred to i(G) as the lower independence bound was shown to be exact for s = 1 and all t in [FGJL1]. We establish
number. The classical Ramsey number res, t) can be defined as the least the case when t is a perfect square.
positive integer n such that for every graph G of order n, either w(G) :;, s or
,6(G) :;, t. Analogously, we define the lower Ramsey number (res, t) as the Theorem 12.15
greatest positive integer n such that for every graph G of order n, either
u(G) ~ s or i(G) ~ t. For every positive integer k, the lower Ramsey number lr(I,~) = ~ + 2k.
Clearly, lower Ramsey numbers are symmetric in sand t. Mynhardt
[MI2] showed that if every graph G of order n has u(G) ~ s or i(G) ~ t, Proof
then so has every graph of order less than n. Moreover, it was shown that
the lower Ramsey number Ir(s, t) exists, and bounds on (res, t) were To show that t'r(I,~) < ~ + 2k + 1, let G be the graph of order (k + l)z
obtained.. obtained from k + 1 disjoint copies of the star KI,k of order k + 1 by
adding all possible edges between the vertices of degree k in the stars.
Then the maximal independent sets of G have order 0 + 1 or 0 + k,
Theorem 12.14 and so i(G) = 0 + 1. Furthermore, the maximal complete sUbgraphs
For every two positive integers sand t, the lower Ramsey number Ir(s, t) exists; have order 2 ork+l so that u(G) =2. Thus t'r(I,~)~~+2k since
moreover, there is a graph of order (k + Il with u(G) = 2 and i(C) = 0 + 1.
To show that t'r(I,~) :;, 0 + 2k, assume, to the contrary, that there is a
s + t + 1 ~ t'r(s, t) ~ 2(s + t) - 1. graph G of order ~ + 2k for which u(G) :;, 2 and i(G) :;, 0 + 1. Since
i( G) :;, 0 + 1, G has an independent set of cardinality at least ~ + l.
Partition V(G) into sets A and B, where A is an independent set of
Proof o + 1 vertices. The following observations will be useful. Since
Let M be the set of all positive integers n for which every graph G of order i(G) :;, 0 + 1, any independent set of fewer than ~ + 1 vertices lies
n has u(G) ~ s or i(G) ~ t. Certainly 1 EM and so M f= 0. To show that within an independent set of ~ + 1 vertices. In particular, if x E B is
(res, t) ~ 2(s + t) - 1, we exhibit a graph H of order 2(s + t) in which adjacent to t' vertices of A, then B contains an independent set of I
u(G) :;, s + 1 and i(G) :;, t + 1. Let H have vertex set VI U V z, where vertices. Also since u(G) :;, 2, there are no isolated vertices in G.
VI n V z = 0 and Wll = W21 = s + t. Furthermore, let (VI) = K5 + 1 and Let B' ~ B be a maximum independent set in (B) of cardinality k - a,
(Vz) = K5 + 1• Finally, add s(s + t) edges between the vertices of VI and where 1 - k ~ a ~ k - 1. We show that for every such a, a contradiction
the vertices of V z so that every vertex of VI U V z is adjacent to s of these arises, which completes th~ proof.
edges. Let T be a maximal independent set. If Tn VI = 0, then T = V z As observed, we must be able to extend If to an independent set
and so ITI = s + t. Suppose, then, that Tn VI f= 0. Then IT n VII = 1, say of ~ + 1 vertices. By our choice of B', then, there is a set A' of
VI E Tn VI' Then, since VI is adjacent to precisely s vertices of V z, it ~ + 1 - k + a vertices in A such that A' U B' is an independent set in G.
follows that IT n Vzl = t and so ITI = t + 1. Thus i(H) = t + 1. Similarly, Each vertex of A' has degree at least 1, and consequently there are at least
u(H) = s + 1, and so Ir(s, t) ~ 2(s + t) - 1. o + 1 - k + a edges between the vertices in A' and the k + a-I vertices
To show that Ir(s, t) :;, s + t + 1, suppose, to the contrary, that there is a in B ~ B'. Thus some vertex x in B - B' is adjacent to at least
graph G of order s + t + 1 for which u(G) :;, s + 1 and i(G) :;, t + 1. Then, (~ + 1 - k + a) / (k + a-I) vertices of A. But this implies that Bcontains
in particular, G contains sets 5 and Tsuch that (5) = KS+ I and (T) = K1+1 • an independent set of at least (~ + 1 - k+ a) / (k + a-I) vertices and, by
Since W(G)I = s + t + I, it follows thatu(G) = s + 1 and i(G) = t + 1; also, the choice of B', that .
IS n TI = 1, say 5 n T = {v}. Letu E T {v}. If degu = 0, then ({u}) is a
maximal complete subgraph, so that u(G)= 1 < s + 1. Thus u is adjacent k _ a > (f + 1 - k + a). (12.9)
to some vertex w where, necessarily, wE S. But then T - {u, v} U {w} (k+ a-I)
366 Ramsey theory Other Ramsey numbers 367
However, the inequality in (12.9) implies that ei + 1 ;:;;:; 0, which is Let sand t be positive integers such that s + t = k, 5 ~ 2 and t ~ 2. By the
impossible. 0 inductive hypothesis, it follows that br(s -1, t) and br(s, t -1) exist, and
that
As a final example of another Ramsey-like number we turn our
attention to bipartite graphs. The classical Ramsey number r(s, t) can be br(s-I,t):::; ( S-I+t)
s-1 -1 and br(s,t-l):::; (S+t-l)
5 -1.
defined as the least positive integer n such that any factorization of KIl into
GI and G2 has the property that either GI contains Ks as a subgraph or G2 Since
contains Kt as a subgraph. By replacing the complete graphs involved in
this definition with complete bipartite graphs, we obtain bipartite Ramsey
numbers. Specifically, for positive integers sand t the bipartite Ramsey
number. br(s, t) is the least positive integer n such that each factorization it follows that
of KIl,1l into GI and G2 has the property that either GI contains Ks,s as a
subgraph or G2 contains Kt,t. Equivalently, the bipartite Ramsey br(s-l,t)+br(s,t~I)+I;:;;:;e;t)-1. (12.10)
number br(s, t) is the least positive integer n such that if every edge of
KIl,1l is colored either red or blue, then there is a red Ks,s (that is, a copy Let n = br(s - 1, t) + br(s, t -1) -1 and consider any red-blue coloring
of Ks,s in KIl,Il, all of whose edges are red) or a blue Kt,t. of Kn,n, where VI and V 2 denote the partite sets of Kn,n'
Given a red-blue coloring of the edges of K"," and a vertex v in Kn,Il, let The following observation will be useful. Suppose that v is a vertex
R v denote the set of vertices in KIl,n that are joined to v by red edges and let of Kn,n with degR v ~ br(s - 1, t) + 1 and that w E Rv with degR w ~
Bv be the set of vertices joined to v by blue edges. The red degree of v (with br(5 -1, t) + 1. Thus the subgraph F = (Rv U Rw - {v, w}) contains Kbb ,
respect to the given red-blue coloring) is degR v = IRvl and the blue degree where b = br(s -1, t), so that F contains either a red Ks-I,s-I or a bl~e
of v is degBv = IBv I· Finally, DR (Kn,n) denotes the minimum red degree of a KI,I' If F contains a blue Kt,I, then so does Kn,n' If F contains a red
vertex in Kn,n and DB(Kn,n) denotes the minimum blue degree. Ks-I,s-I, then (Rv U Rw ) contains a red Ks,s, as does Kn,n' Thus if v is a
Obviously, the bipartite Ramsey numbers are symmetric in 5 and t, and vertex of Kn,n with degR v ~ br(s - 1, t) + I, and wE Rv with
for every positive integer t; br(l, t) = t (Exercise 12.23). The existence of degR w ~ br(s - 1, t) + I, then Kn,n contains a red Ks,s or a blue Kt,t. Simi-
the bipartite Ramsey number br(s, t) follows from the work of Erdos and larly, if degB v ~ br(s, t -1) + I, and WE Bv withdegB W ~ br(s, t - 1) + I,
Rado IERl]; while Hattingh and Henning [HHl] presented an upper then Kn,n contains a red Ks,s or a blue Kt,t. In particular, if 5R =
bound for br(s, t). 5R(Kn,n) ~ br(s -, 1, t) + lor 6B = DB(Kn,n) ~ br(s, t -1) + I, then Kn,n con-
tains a red Ks,s or a blue KI,I' Assume, then, that 5R ;:;;:; br(s - 1, i) and
Theorem 12.16 5B ;:;;:; br(s,t -1).
If degR v ~ br(s -1, t) + 1 for all vertices v in V v then the number of red
For every two positive integers 5 and t the bipartite Ramsey number br(s, t) exists; edges between VI and V 2 is at least
moreover,
[br(s -1, t) + br(5, t - 1) + 1]. [br(s 1, t) + IJ.
br(s, t) ;:;;:;.( s+t)
s -1.
This implies that degR W ~ br(s - 1, t) + 1 for some W E V 2• Therefore for
some v in VI we have degR v ~ br(s - 1, t) + I, and W E Rv with
Proof degR W ~ br(s -1, t) + 1, implying that G contains a red Ks,s or a blue
KI,t· Thus we may assume that at least one vertex VI of VI (and, similarly,
We proceed by induction on k, where k = 5 + t. Note that we have equality one vertex V2 of V2 ) has red degree at most br(s - 1, i). Lety be a vertex of
for 5 = 1 and arbitrary i, and for t = 1 and arbitrary s. Hence the result is minimum blue degree in Kn,n- Without loss of generality, assume y E VI'
true for k = 2 and k =·3. Furthermore, we may assume that s ~ 2 and t ~ 2. Then VI and y are vertices of VI for which degR y = n - degB y =
Assume that br(s', t') exists for all positive integers s' and t' with n - 6B ~ br(s -I,t) + 1 and degB VI = n -degR VI ~ br(s,t -1) + 1.
5' + t' < k, where k ~ 4, and that Since 1V21 = br(s - 1, t) + br(5,t - 1) + I, .it follows that IRy n Bv,l ::f= 0.

br(s',t') ~. (
5' +
5'
t') -1. Let x E Ry n Bv1 ' If degR x ~ br(s ~ 1, t) + I, then Kn,n contains a red Ks,s
or a blue Kt,I' If, on the other hand, degR x ~ br(s -1, t), then degB x ~
br(s, t - 1) + 1 and again Kn,n contains ared Ks,s or a blue Kt,t.
368 Ramsey theory Other Ramsey numbers 369
Thus The values br(2, 2) = 5 and br(3, 3) = 17 had earlier been determined by
br(s, t) ~ br(s -1, t) + br(s, t -1) + 1 (12.11) Beineke and Schwenk [BSl]. Also, an improved upper bound for br(s, t)
when s = t was discovered by Thomason [T2].
and the desired result follows from (12.10). 0
Theorem 12.18 '--~

For every positive integer t,

br(t,t) ~ zt(t-l) + 1.

EXERCISES 12.3

12.20 Prove Theorem 12.12.

12.21 Show that ir(2, t) = t, for all t ~ 2.
Figure 12.2 An extremal graph showing that br(2, 3) ;?: 9.
12.22 Show that Ir(l, 5) = 10.
Inequality (12.11) gives a potentially improved bound for br(s, t). We
state this as a corollary.
12.23 Show that br(l, t) = t, for all t ~ 1.
12.24 Show that br(2, 2) = 5.
Corollary 12.17
For integers s ;?: 2 and t ;?: 2,
br(s, t) ~ br(s -1, t) + br(s, t - 1) + 1.

The bound given in Theorem 12.16 for br(s, t) is exact if s = 1 or t = 1.

The bound is also exact for s = 2 and t = 2, 3 or 4. Figures 12.2 and
12.3 indicate the extremal colorings showing that br(2, 3) ~ 9 and
br(2,4) ~ 14. Here the edges shown are the 'red' edges of Ks,s and K13,13'

Figure 12.3 An extremal graph showing that br(2, 4);?: 14.

 CHAPTER 13

•
The probabilistic method In
graph theory

In this chapter we investigate a powerful, nonconstructive proof tech-

nique known as the probabilistic method and then study properties of
random graphs.

13.1 THE PROBABILISTIC METHOD

In Chapter 12 we showed that for every integer t ;?: 3, the Ramsey number
ret, t) > L2 t / 2 J. We did so by proving the existence of a graph G of order
n = l2 t / 2 J such that neither G nor G contains K t as a subgraph. More
specifically, we counted the number of different labeled graphs of order
nthat contain a subgraph isomorphic to Kt , together with their comple-
ments, and showed that there were fewer than 2G) of these graphs. Here
we revisit this proof from a probabilistic point of view. Recall that the The proof of Theorem 13.1 illustrates the basic technique of the
assignment of the colors red or blue to the edges of a graph G is called probabilistic method. An appropriate probability space is defined on a
a red-blue coloring of G. set of objects (in our case, red-blue colorings of K n ). An event A is
then defined representing the desired structure. In the proof of Theorem
Theorem 13.1 13.1, this event A = Ms. We then show that A has positive probability
so that an object with the desired characteristics or structure must
For every integer t ;?: 3, exist.
r(t,t) > Li/2 J. Before presenting another example of the probabilistic method We
introduce some standard terminology. In the proof of Theorem 13.1 we
Proof defined a probability space whose objects consisted of all red-blue color-
ings of Kn in which each such coloring was equally likely. In such a case
Let n = Li/ J. We show that there exists a red-blue coloring of Kn that
2
we refer to a random red-blue coloring of Kn-
contains no monochromatic Kt , that is, neither a red Kt nor a blue Kt• Our second-example of the probabilistic method involves tournaments.
Consider the probability space whose elements are red-blue colorings A tournament Tof order n ;?:. 2 has property Sk (1 ~ k ~ n ~ 1) if for every
of K n , where V(Kn ) = {VI, V2," ., V n }. The probabilities are defined by set S of k vertices of T there is a vertex-w¢:S such that (w,V)E E(T) for
setting every v in S, that is, there is a vertex w¢: 5 that is adjacent to every vertex of
P[VjVj is red] = P[VjVj is blue] = ! S. Using the probabilistic method, we show that for every such integer k
there is a tournament T of order nhaving property Skfor all sufficiently
for each pair Vi, Vj of distinct vertices of Kn (where prE] denotes the prob-
large n.
ability of event E), and letting these events be mutually independent.
372 The probabilistic method in graph theory The probabilistic method 373
Theorem 13.2 Theorem 13.3
For every positive integer k and sufficiently large integer n, there is a tournament For each positive integer n there is a tournament of order n with at least nl 2n - 1

T of order n with property Sk' hamiltonian paths.

Proof Proof
For a fixed integer n, consider a random tournament T on n vertices. More Consider a random tournament T of order n, and let X be the number of
specifically, consider the probability space whose elements are the 2 G) hamiltonian paths in T. For each of the nl permutations (j of VeT), let X(J be
different labeled tournaments T with vertex set {VI, V2 , ... , V ,l }. The the indicator random variable for (j giving a hamiltonian path, that is; X(J is 1
probabilities are defined by setting . or 0 depending on whether (j does or does not describe a hamiltonian path
P[(v;, Vj) E E(T)] =P[(Vj, Vi) E E(T)] =!' in T. Then P[X(J = 11 = (!)n -1 (since each of the n - 1 arcs in the potential
hamiltonian path must be correct) and so E[X(J] = (!t -1. Let X = L: X(J'
aIJd then letting these events be mutually independent. Thus each of these where the Summation is taken over all permutations (j of VeT). Then X
2(2)different labeled tournaments is equally likely. gives the number of hamiltonian paths in T and
Fora fixed k-element set S s:;; {VI, V2,'" ,vn }, let As denote the event that
there is no wE VeT) - 5 that is adjacent to every vertex of S. Each vertex E[X] = E[L: X(J] = nlT(n-l).
w E VeT) - S has probability (!)k of being adjacent to every vertex of 5, Hence there is a point in the probability space, namely a specific tourna-
and there are n - k such vertices w, all of whose chances are mutually ment T, for which X exceeds or equals its expectation. This T has at least
independent. Thus, nlT(n-1) hamiltonian paths. 0

In our fourth example of the probabilistic method in graph theory, the

and so
objects in the sample space unde)) consideration are the vertex subsets of a
P[VAsl:E; LP[As1 = G)(1- Tk t- k . fixed graph G. Here we obtam an upper bound on the domination
number·1'(G) of G, due to Payan [P2], in terms of the minimum degree
Thus, if we choose n so that m(l - 2-k )n-k < 1, then P[VAs] < 1. For of G (Theorem 10.4).
such an integer n, it follows that P[VAs1 > 0 = P[Ms] > O. Thus there is
a point in the probability space for which /\A s is true, that is, there exists Theorem 13.4
a tournament T with property Sk' 0
Let G be a graph of order n with 8 = 8(G) f.. Then
~

Observe that again we have defined an appropriate probability space G) ~ n(t + !n(8 + 1)) .
(
and event A. This is done so that A has positive probability and, conse- "I '" 8+1
quently, the desired object (in this case, a tournament of order n with
property Sk) exists. Proof
.For a probability space S, a random variable X on S is a real-valued
function on S. The expected value E[X] of X is the weighted average Set p = (In(8 + 1))/(8 + 1) and consider a random set 5 ~ V(G) whose
vertices are chosen independently with probability p. That is, consider
E[X] = LkP[X=k], the probability space whose elements are the 2n subsets of V(G) =
{VI> V2,"" v n }. The probabilities are assigned by setting P[v; E 51 = p,
where the sum is takenover all possible values k of X. It is easy to see that
and letting' these events be mutually independent. Thus a subset
expectation is linear, that is, if Xv X2 and X are random variables on a
S ~ V(G) OCcurs with probability plsl(t ..... p)n-1sl.For a random set S, let
probability space S and X = Xl + X2 , thenE[X] = E[X1l + E[X21.· Further-
more, if E[X] = t, then X(Sl) ~ t and X(S2) ~ t for some elements 81 and S2
y = Ys be the set of vertices not in 5 haVingno neighborS inS. Then 5 l0 Y s
of S. This second observation will prove to be very powerful, as indicated is a dominatin~ set oiG. We show that
in the proof of Theorem 13.3. This result of 5zele [512] is often considered
the first use of the probabilistic method.
E[jSI + IYlr~ n(l +6;~ + 1)) .
 The probabilistic method 375
independent. Thus each o~, the different labeled graphs of size m occurs
with probability pm(1- p)(2)-m.
Let X be the random variable that gives the number of cycles of length
at most t. For a fixed i, 3':;; i':;; t, there are G) i-element subsets of
{VI'VZ'" .,vn }· For each such set S there are i!/(2i) = (i-l)!/2 different
cyclic orderings of the vertices in S. Thus there are
n) (i -I)! = n(n - 1) ... (n - i + 1)
(i 2 2i
potential cycles' of length i, and so X is the sum of
E1=3 (n(n -1) .. ; (n - i + 1))/2£ indicator v,:!-riables. Furthermore, since a
cycle of length i' occurs with probabilipr p', the linearity of expectation
gives

E[X] = ti=3 n(n -1) ... .(n - i + 1) pi.

21
(13.1)

Since p = nO-I, it follows from {13.1).that

8 ;i'
( OJ \..

E[X]':;; (13.2)
One important new idea is involved in the previous proof. The random
choice did not give the required dominating set immediately; it gave us By the choice of fJ, it follows that Ot,= 1 - c for some real number c with
the set S which then needed to be altered (in this case, by adding Y s) to 0< c < 1. Thus
obtain the desired dominating set. The proof of Theorem 13.5 employs the
E[X] (nO i (nO( K
same technique of alteration. This proof also uses Markov's inequality,
which states that for a random variable X and positive number t,
-.:;;
n/2
L: --.-.:;; L: -. =--;;:,
i=3 m i=3 m n
E[X] where K = E1=3 Iii, and so limn -+ oo E[x]/(n/2) = O.
P[X~t].:;; --.
t By Markov's inequality,
Theorem 13.5 was first stated as Theorem 8.16 without proof. E[X]
PIX ~ n/2]':;; n/2'
Theorem 13.5
Thus, for n sufficiently large, P[X ~ n12] < 0.5. .
For every two integers k ~ 2 and t ~ 3 there exists a k-chromatic graph whose Let t = f3(ln n )/pl The probability t~at a given t-element subset of
girth exceeds t. {VI, V2, ••• , v n } is independent is (1 - p) (2). Since there are G) such sets,
it follows that
Proof
For k = 2, any even cycle of length greater than t has the desired proper-
P[,B(G) ~ t] .:;; G) (1- p)(i).

ties. Assume, then, that k ~ 3. Let 0 < fJ < l/t and, for a fixed positive However, 1 - p < e- P, and so
integer n, let p = nO-I. Consider a random graph G of order n whose
edges are chosen independently with probability p; that is, consider the
probability space whose elements are the 2(~) different labeled graphs G
P[,B(G) ~ tj < G) e-p(i) < (ne-p(t-I)/Z)t.

with vertex set {VI' V2, •.. , v n}. The probabilities are defined by setting Since ne- p(t-I)/2 <1 for n sufficiently large, it follows that we can
P[ViVj is an edge of Gj = p, and then letting these events be mutually choose n so that P[,B(G) ~ t] < 0.5 and p[X ~ n/2] < 0.5. For such an n,
376 The probabilistic method in graph theory Random graphsr 377
then, P[,8(G) < t and X < n/Z) > O. Thus there is a specific graph G of For a positive integer n and positive real number p less than I, the
order n with fewer than n/Z cycles of length at most I with ,8(G) < ,t. random graph G(n,p) denotes the probability space whose elements are
Since t = f3(ln n)/p1 = f3(In n)n 1 - 01, we may assume that n is sufficiently the 2(2) different labeled graphs with vertex set {VI, V2,"" vn }. The prob-
large to ensure that t < n/Zk. abilities are determined by setting P[ViVj E E(G)) = p, with these events
Remove one vertex from each cycle of G of length at most I, denoting mutually independent, s~ that the probability of any specific graph
the resulting graph by G*. Then G* has girth greater than I and ,8(0.) ~ with m edges is pm(1 - p)(2)-m. Although we refer to the 'random graph
,8(G) < n/2k. Furthermore, G(n,p)', it is important to remember that we are, in fact, referring to an
* jV(G*)1 n/2 element selected from the probability space G(n, pl.
X(G ) ~ ,8(G*) ~ n/2k = k. In this section we discuss properties shared by almost all graphs.
Specifically, given a graph theoretic property Q, we say that almost all
Finally, we remove vertices froni. G*, if necessary, to produce a graph G** graphs (in G(n, p)) have property Q if lirnn-> 00 P[G E G(n, p) has property
with girth greater than t and X(G**) = k. 0 Q] = 1. A useful technique to establish that almost all graphs have prop-
erty Qis to CIefine a nonnegative integer-valued random variable X on
G(n,p) so that Ghas property Q if X=O.Then P[X = 0);';;; P[G E G(n,p)
EXERCISES 13.1 has property Q); so that if limn->oo P[X = 0) = I, then we also know
that lim n-+ oo P[G E G(n,p)) = 1. Since X is an integer-valued function,
13.1 (a) Show that if G)Zl-(~) < I, then r(t,t) > n. limn-+ oo P[X = 0) = 1 if and only if limn~oo P[X ~ 1] = O. Using Markov's
(b) Stirling's formula states that limt~oo t!/(t/e)' == v'27it. Use this inequality, we see that since P[X ~ 1) ;,;;; E[X] it follows that if
fact to prove that limn~oo E[X] = 0, then limn->oo P[X ~ 1] = 0 and so almost all graphs
t1' l2 have property Q.
r(t, t) > M' Our first result shows that for constant real number p (0 < p <I),
ey2
almost all graphs are connected with diameter 2. This strengthens the
13.2 (a) Show, without probabilistic techniques, that every (n, m) graph original result of Gilbert-[G4] that almost all graphs are connected.
contains a bipartite subgraph with at least m/2 edges.
(b) Give a probabilistic proof that every (n, m) graph G contains a
bipartite subgraph with at least m/2 edges. (Hint: COJ.1.$ider the Theorem 13.6
probability space whose elements are the Zn subsets of V(G) = For any fixed positive real number p < I, almost all graphs are connected with
{Vl' v2,"" v n }. For a random set 5 ~ V(G), the probabilities are diameter 2. .
assigned by setting P[Vi E 5) = 0.5 an<,i Jetting these events be
mutually independent. Let X be the random variable defined
so that X(5) is the number of edges incident with exactly one Proof
vertex of 5, and consider the expected value of X.) For each graph G in G(n,p), let the random variableX(G) be the number
13.3 Give a probabilistic proof that there is a red-blue coloring of Kn with of (unordered) pairs of distinct vertices of G with no common adjacency.
at most G) .21 -<:) monochromatic copies of K•. Certainly, if X(G) =0 then Gis connected with diameter 2 (or G is the
13.4 Show that if G) . Z-G) + (7) . Z-(~) < I, then r(s, t) > n. single exception K n ). Thus (by Markov's inequality), it suffices to show
that lirn n-+ oo E[X] = O.
List the (~) 'pairs of vertices of G. Then X can be written as the sum of G)
indicator variables Xi, 1 ;,;;; i ~ (~), where Xi = 1 if the ,ith pair has no
13.2 RANDOM GRAPHS common adjacency and 0, otherwise. Thel), X = Xl + Xz + ... + XCi) and,
by the linearity of expectation, E[X] = L~~l E[X;). If the ith pair is u, v,
In section 13.1 we found it useful to define appropriate probability spaces then P[Xi = 1) is the probability that no other vertex is adjacent to u and
in order to prove the existence of graphs with desired properties. In this v. Fora fixed vertexz (# u, v), the probability thatz is not adja~enttoboth
section we give a formal model for a random graph and answer que~tions u and v is 1 - p2. This probability is independent of the probability that
about the probability that a random graph has certain properties such as any other vertex is not adjacent to u and v. Thus the probability that none
nonplanarity or k-connectedness.. of the n - 2 verticesz (# u, v) is adjacent to both u and v is (1 - pr-
2 Z
and
378 The probabilistic method in graph theory Random graphs 379
so E[X; = 1] = (1 - pZt-Z. It follows that If Ql and Qz are graphical properties such that almost all graphs (in
E[X] = G) (1 - pZt-
Z G(n,p)) have property Ql and almost all graphs have property Q2, then
almost all graphs have both properties Ql and Q2 (Exercise 13.5).

and, c1early,lim n..... oo E[X] = O. 0

Corollary 13.9
The basic idea used to define the random variable X in t~e proof of For fixed nonnegative integers k and { and a positive real number p < ·1, let Qbe a
Theorem 13.6 was generalized by Blass and Harary [BH3] m order to graphical property deducible from finitely many applications of Corollary 13.8.
study other properties of almost all graphs. Then almost all graphs have property Q.

Theorem 13.7 As an example of the use of Corollary 13.9 to show that almost all
graphs have property Q, we prove that for each graph H and fixed real
For fixed nonnegative integers k and { and a posi!ip~real number p < 1, almost number p (0 < p < 1), almost all graphs (in G(n,p)) contain Has an
all graphs have the property that if 5 and T are dlsJom~ k-el~ment and (-element induced subgraph.
subsets of vertices, then there is a vertex z¢. 5 U T that IS adjacent to every vertex
of 5 and to no vertex of T.
Theorem 13.10
Proof For each graph H and fixed positive real number p < 1, almost all graphs contain
Define a pair 5, T of disjoint k-element and t -element subsets of V (G) to be
Has im induced subgraph.
bad if no vertex z 1:- 5 U T is adjacent to every vertex of 5 and to no vertex of
T. For each G in G(n, p), let X(G) be the number of such bad pairs 5, T. We Proof
wish to show that almost all graphs have no bad pairs of sets and, as in
Let k = IV(H)I. We proceed by induction on k. For all k = 1, all graphs
the proof of Theorem 13.6, we need only show .th~t limn-too .E[X] = O.
The variable X can be written as the sum of mdlcator varIables Xi, contain H as an induced subgraph since H= K 1 . Assume that fbr every
graph H'· of order k - 1 > 1, almost all graphs contain H' as an induced
where Xi = 1 if the ith pair 5, T is bad, and Xi = 0, othe~se; The~
P[Xi ::::; 1] = (1- ,1'(1 - p)<')l n - k -(. Since .the number of paIrs S, T IS subo-raph, and consider a graph H of order k. Select a vertex v of H and let
N = (~Wtk) = n!/(k!(n - k)!(n - k- t)!), It follows that
H' :;: H - v. Then, by the inductive hypothesis, almost all graphs contain
H' as an induced subgraph. Furthermore, if v is adjacent to eXcktly s
~ n! (1 _ k(1 _ )()n-k-(. vertices of H' in H, then since almost all graphs have property Qk,k, it
E[X] = £:1. E[X;] = k!(n - k)!(n - k - f)! P P follows from Corollary 13.9· that almost all graphs contain Hasan
induced subgraph. 0
As n tends to infinity, the first factor in the expression for E[X] tends to
infinity (as a polynomial in n) and· the second factor tends to 0 exponen- If Q is a property like planarity that implies certain graphs (such as K s
tially. Thus, limn ....."" E[X] = 0, and the proof is complete. 0 and K 33 ) do not exist as induced subgraphs, then Theorem 13.10 imme-
diately'implies that for p fixed, almost no graph in G( n, p) has property Q.
In the case k = 2 and { =0, Theorem 13.7 reduces to Theorem 13.6. Here, of course, we mean that limn-too P[GE G(n,p) has property QI = o.
For fixed nonnegative integers k and t, let Qk,( denote the property that
if 5 arid T are disjoint sets of vertices of a graph with 151 ~ k and ITI ~ t,
then there is a vertex z¢. S uT that is adjacent to every vertex of S and to Corollary 13.11-
no vertex of T. . For any fixed positive real number p < 1, almost no graphs are planar.

Corollary 13.8
Corollary 13.12
For fixed nonnegative integerskand t and a positive real number p < 1; almost
For any positive integer kand fixed positive real numberp < 1, almost no graphs
allgraphs have property Qk,(' are k-colorable.
380 The probabilistic method in graph theory Random graphs 381
Corollary 13.13
Property Threshold
For any positive integer k and fixed positive real number p < 1, almost no graphs
Contains a path of length k r(n) = n-(k+l}/k
have genus k. Is not planar r(n) = lin
Contains a hamiltonian path r(li) = (In n)ln
Other results can be obtained in a manner similar to that used in the Is connected r(n) = (Inn)ln
proof of Theorem 13.10. Contains a copy of Kk r(n) = n-2 /(k-l}

Figure 13.1 Threshold functions.

Theorem 13.14
For any fixed positive integer k and positive real number p < 1, almost all graphs no graph G has clique number w(G) ?:: 4 while if limn...; 00 p(n)/n- Z/ 3 = 00,
are k-connected. then almost every graph G has w( G) ?:: 4.
Gene~a~ly, let Q be a graph theoretic propetty that is not destroyed by
It should be noted that for a fixedreal number p (0 < p < 1), there are the addItion of edges to a graph. A function r(n) is called a threshold
interesting properties of almost all graphs that cannot be proved by apply- function for Q if limn -+ oo p(n)/r(n) = 0 implies that almost no graph
ing Corollary 13.9. For example, Blass and Harary [BH3] showed that ~as property Q, that is, limn...;oo P[G E G(n,p(n)) has Q] = 0 and
almost all graphs are hamiltonian; however, Corollary 13.9 cannot be lun n -+ oo p(n)/r(n) = 00 implies that almost every graph has property Q,
used to establish this result. that is, Iim n-+ oo P[G E G(n, p(n)) has Q] = 1.
If p is fixed and X is the random variable defined on G(n, p) by Figure 13.1 indicates some of the properties Q for which a threshold
X(G) = IE( G) I, then the expected value of X is p('2), and consequently we function r(n) exists and is known (see [58, p. 17]). .
are dealing with dense graphs. So, in some sense, the preceding results of The books by Palmer [PH Alon and Spencer [ASl] and Spencer [58]
this section are not surprising. We next briefly consider G(n,p(n)), that is, are excellent sources of additional material on random graphs and the
G(n,p) where p is not fixed and p = p(n) is a function of n. We begin with probabilistic method.
an example involving complete subgraphs.
Let Q be the property that a graph G has clique number w( G) .< 4, and
let p(n) be a function of n. For each graph G in G(n,p(n)),let the random EXERCISES 13.2
variable X(G) denote the number of copies of K4 in G. If X(G) = 0, then
G has property Q. Thus, by MarkQv's inequality, if limn...;oo E[X] = 0,
13.5 (a) Show that P[A and B] ?:: 1 - (P[A] + P[B"]).
then almost every graph in G(n,p) has clique number less than 4. For
(b) Show that if almost all graphs (in G(n, p)) have property QI and
each 4-element subset S of V(G), let Xs be the indicator variable with
almost all graphs have property Qz, then almost all graphs have
X s = 1 if (5) is complete and X s = 0, otherwise. Then X = L X s, where
both properties Ql and Qz. .
the sum is taken over all 4-element subsets of V(G). Furthermore,
E[XsJ =:= P[X s] = (p(n))6. By the linearity of expectation, then, 13.6 Prove Corollary 13.13.

E[X] = L:E[Xs] = G) (p(n))6 < n\p(~))6.

13.7 Prove Theorem 13.14.
13.8 Without using the results given in Figure 13.1, show that if
If lim n-+ oo E[X] = 0, then almost all graphs in G(n,p(n)) have clique Ii~ll-+oo p(n)/n- Z/(k-l) = 0, then almost no graph in G(n, p(n)) con-
number less than 4. Consequently, if lim n-+ oo p(n)/n- Z/ 3 = 0, then tams a copy of Kk •
almost all graphs have clique number less than 4. We can think of this 13.9 F~f ~ fixea, let T(n,!) denote the probability space consisting of the
result as saying that if p(n) is 'significantly smaller' than n-Z/ 3 , then 2 different labeled tournaments T of order n with vertex set
almost all graphs in G( n, p) have clique number less than 4. The surprising {VI' Vz, ... ,VII}' where the probabilities are defined by setting
fact is that, using the second moment method of probability theory, it !:an =
P[(V,., Vj) E E(T)J P[(Vj,Vi) E E{T)] =!.Show that for a fixed posi-
be shown that iflim n -+ oo p(nl/n-:z/3 = 00, then almost no graph has clique tive integer k, almost all tournaments have property 5k (Theorem
numberless than 4. Thusn- 13 can be thought of as a threshold for clique 13.2).
number less than 4. Equivalently, if limn -+ oo p(n)/n- Z/ 3 = 0, then almost
 Glossary of symbols 383
I(G) integrity 88
i(G) lower independence number 270
i(G) independent domination number 312
i1 (G) lower edge independence number 270
Glossary of symbols IR(G) upper irredundance number 321
ir(G) irredundance number 318
ir(s,t) irredundantRamsey number 363
Kn complete graph of order n 6
K(r,s) or Kr,s complete bipartite graph 8
Symbol Meaning Page K(nl, n2, ... ,nk) complete k-partite graph 8
Kn1 ,n2,... ,nk complete k-partite graph 8
A(G) adjacency matrix 1 Kk(t) regular complete k-partite graph 8
Aut(G) automorphism group 40 k(G) number of components 18
a(G) vertex-arboricity 68 ko(G) number of odd components 263
edge-arboricity 70 (r(s, t) lower Ramsey.number 364
al (G)
B(G) incidence matrix 2 M(G) matching graph 271
ban(G) bandwidth 296 m or m(G) size 1
bandwidth of a numbering 296 Med(G) median 22
banf(G)
BC(G) block-cut-vertex graph 67 nor n(G) order 1
br(s, t) bipartite Ramsey number 366 (n,m) order n and size m 1
Cn cycle of order n 18 N(v), N(U) neighborhood 109,261
C(G) closure 106 N(G) maximum order of a component 88
Cn + 1 (G) (n + I)-closure 113 Nk(v) k-neighborhood 323
Cen(G) center 20 N[v],N[U] closed neighborhood 302,317
D digraph 25 odv outdegree 26
D(G) degree matrix 65 Pn path of order n 18
Dt.(r) Cayley color graph 45 Per(G) periphery 20
V(G) degree set 15 pi (G) planetary domination number 327
d(u,v) distance 19 PIX] probability 370
.~;
degc v or deg v degree 2 Qn n-cube 9
20 r number of regions 155
Ii
~-'
diamG
E(G)
diameter
edge set of graph 1 radG radius 20
E(D) edge set of digraph 25 r(n) threshold function 381
)l e(v) eccentricity 20 R(n, k) near regular complete multipartite graph 332
E[X] expected value 372 r(8, t) Ramsey number 351
~,
ex(n;F) extremal number 329 r(G 1 , G2,··. ,Gk) (generalized) Ramsey number 357
f(r,g) . smallest order of an r-regular graph with (r,g)-cage smallest [r,g]-graph 343
girth g 343 [r,g]-graph r-regular graph with girth g 343
G graph 1 Sk surface of genus k 191
g(G) girth 231 t(G) toughness 86
gen(G) genus 183 td(v) total distance 22
I, genM(G) maximum genus 207 u(G)
V(G)
lower clique number 364
G(n,p) random graph 377 vertex set of graph 1
grac(G) gracefulness 281 V(D) vertex set of digraph 20
~: '::
id v indegree 26 a(G) vertex covering number 267
384 Glossary of symbols
Q](G) edge covering number 267
(3(G) independence number 86
(31 (G) edge independence number 239
qG) upper domination number 321
i(G) domination number 302 Graph theory books
ik(G) k-domination number 323
.6.(G)
5(G)
maximum degree
minimum degree
2
2
1936-1996
()1(G) edge-thickness or thickness 181
K(G) vertex-connectivity 74
K1 (G) edge~connectivity 75
V(G) crossing number 174
V(G) rectilinear crossing number 176 1936
eO(G) number of components of odd size 210
e(G) minimum eo(G) 210 Konig, D. (1936) Theorie der endlichen und unendlichen Graphen, Teubner, Leipzig.
Pk(G) k-step domination number 324
X(G) chromatic number 220
edge chromatic number or chromatic index 236 1958
Xt(G)
X2(G) total chromatic number 243 Berge, C. (1958) Theorie des Graphes et Ses Applications, Dunod, Paris.
Xr(G) list chromatic number 233
Xp(G) P chromatic number 233
X(Sk) chromatic number of a surface 257 1959
W(G) clique number 227
Ringel, G. (1959) Farbungsprobleme auf Fliichen and Graphen, Veb Deutscher Verlag
G complement 8 cler Wissenschaften, Berlin.
d kth power 122
G* symmetric digraph 27
Gd dual 248 1961
G1 = G2 isomorphic 3
G1 UG2 union Seshu, S. and Reed, M. B. (1961) Linear Graphs and Electrical Networks, Addison-
9 Wesley, Reading, MA.
G1 +G2 join 9
G1 X G2 cartesian product 9
G1 EEl G2 factorization, decomposition 272,278 1962
H~G subgraph 4
(U) subgraph induced by U 5 Berge, C. (1962) The Theory of Graphs and its Applications, Methuen, London.
(X) Ford, L. R. and Fulkerson, D. R. (1962) Flows in Networks, Princeton University
subgraph induced by X 6 Press, Princeton.
G-v deletion of a vertex 4 Ore, O. (1962) Theory of Graphs, 38 Amer. Math. Soc. Colloq. Pub., Providence, RI.
G-e deletion of an edge 4
G+f addition of an edge 5
G --+ (Gil G2 ,···, Gk ) arrows 360 1963
t associated tournament 140 are, o. (1963) Graphs and Their Uses, Random.House, New York.

1964

Grossman, 1. and Magnus, W. (1964) Groups and Their Graphs, RandotIl House,
New York. .
386 Graph theory books 1936-1996 Graph theory books 1936-1996 387
1972
1965

Busacker, R. G. and Saaty, T. L. (1965) Finite Graphs and Networks: An Introduction Johnson, D. E. and Johnson, J. R. (1972) Graph Theory With Engineering Applications,
With Applications, McGraw-HilL New York Ronald Press, New York.
Harary, F., Norman,R. Z, and Cartwright, D. (1965) Structural Models: An Intro- Mayeda, W. (1972) Graph Theory, Wiley-Interscience, New York
duction to the Theory of Directed Graphs, John Wiley & Sons, New York Read, R. C. (ed.) (1972) Graph Theory and Computing, Academic Press, New York.
Sa~hs, H. (1972) Einfahrung i~ die Theorie der endlichen Graphen II, Teubner, Leipzig.
WIlson, R. J. (1972) IntroductIOn to Graph Theory, Academic Press, New York.
1966

Tutte, W. T. (1966) Connectivity in Graphs, University of Toronto Press, London. 1973

Berge, C. (1973) Graphs and Hypergraphs, North-Holland, London.

1967 DorfIer, yv. and Miihlbacher, J. (1973) Graphentheorie far Informatiker, de Gruyter,
Berlin.
Ore, O. (1967) The Four-Color Problem, Academic Press, New York. Harary, F. and Palmer, E. M. (1973) Graphical Enumeration, Academic Press,
New York. .
White, A. T. (1973) Graphs, Groups and Surfaces, North-Holland, Amsterdam.
1968

Moon, J. W. (1968) Topics on Tournaments, Holt, Rinehart and Winston, New York 1974

Biggs, ~. (1974) Algebraic Graph Theory, Cambridge University Press, Cambridge.

1969 MalkeVltch, J. and Meyer, N. (1974) Graphs, Models, and Finite Mathematics
Prentice-Hall, Englewood Cliffs, NJ. '
Harary, F. (1969) Graph Theory, Addison-Wesley, Reading, MA. Ringel, G. (1974) Map Color Theorem, Springer-Verlag, Berlin.
Knodel, W. (1969) Graphentheoretische Methoden und ihre Anwendungen, Springer- Sache, A. (1974) La Theorie des Graphes, Presses Universitaire de France, Paris.
. Verlag, Berlin.
Nebesky, L. (1969) Algebraic Properties of Trees, Universita Karlova, Praha.
Roy, B. and liorps, M. (1969) Algebre Moderne et Theorie des Graphes Orientees vers les 1975
Sciences Economique et Sociales, Dunod, Paris.
Zykov, A. A. (1969) Teoriia Konechnykh Grafov, Nauka, Novosibirsk Christofides, N. (1975) Graph Theory, An Algorithmic Approach, Academic Press,
London. .
Fulkerson, D. R. (ed.) (1975) Studies in Graph Theory, Parts 1 and 2, Mathematical
1970 Association of America, New York.

Elmaghraby, S. E. (1970) Some Network Models in Management Science, Springer-

Verlag, Berlin. 1976
Harris, B. (ed.) (1970) Graph Theory and Its Applications, Academic Press, New
York. Biggs,!'!. L., Lloyd, E. K. and Wilson, R. J. (1976) Graph Theory 1736~1936, Oxford
Moon, J. W. (1970) Counting Labelled Trees, Canadian Mathematical Congress. Umversity Press, London.
Wagner, K. (1970) Graphentheorie, Bibliographisches Inst., Mannheim. Bondy, J. A. and Murty, U. S. R. (1976) Graph Theory with Applications, North-
Holland, New York
Noltemeier, H. (1976) Graphentheorie mit Algorithmen und Anwendunuen de Gruyter
Berli~ u· , ,
1971
Teh, H ..H. and .Shee, S. ~. (1976) Algebraic Theory of Graphs, Lee Kong Chien
Behzad, M. and Chartrand, G. (1971) Introduction to the Theory of Graphs, Allyn and Institute of Mathematics and Computer Science, Singapore.
Bacon, Boston. Trudeau, R J. (1976) Dots and Lines, Kent State University Press, Kent, OH.
Marshall, C. W. (1971) Applied Graph Theory, Wiley-Interscience, New York
Maxwell, L. M. and Reed, M. B. (1971) The Theory of Graphs: A Basis for Network
Theory, Pergamon Press, New York 1977
Nakanishi, N. (1971) Graph Theory and Feynman Integrals, Gordon and Breach,
New York. Chartrand, G. (1977) Graphs as Mathematical Models, Prindle, Weber & Schmidt,
Price, W. L. (1971) Graphs and Networks: An Introduction, Auerbach, New York. Boston.
388 Graph theory books 1936-1996 Graph theory books 1936-1996 389
Fiorini, S. and Wilson, R. J. (1977) Edge-Colourings of Graphs, Pitman, London. 1983
Giblin, P. J. (1977) Graphs, Surfaces and Homology, Chapman & Hall, London.
Graver, J. E. and Watkins, M. E. (1977) Combinatorics With Emphasis on the Theory of Barnette, D. (1983) Map Colorings, Polyhedra, and the Four-Color Problem, Mathe-
Graphs, Springer, New York. matical Association of America, New York.
Saaty, T. 1. and Kainen, P. C. (1977) The Four-Color Problem: Assaults and Conquests, Beineke, 1. W. and Wilson, R. J. (eds) (1983) Selected Topics in Graph Theory 2,
McGraw-Hill, New York. Academic Press, London.

1978 1984

Beineke, 1. W. and Wilson, R. J. (eds) (1978) Selected Topics in Graph Theory, Aigner, M. (1984) Graphentheorie: Eine Entwicklung aus der 4-Farben Problem,
Academic Press, London. Teubner, Stuttgart.
Bollobas, B. (1978) Extremal Graph Theory, Academic Press, New York. Gondran, M. and Minoux, M. (1984) Graphs and Algorithms,Wiley-Interscience,
Capobianco, M. and Molluzzo; J. C. (1978) Examples and Counterexamples in Graph New York.
Theory, North-Holland, New York. Mehlhorn, K. (1984) Graph Algorithms and NP-Completeness, Springer-Verlag,Berlin.
Minieka, E. (1978) Optimization Algorithms for Networks and Graphs, Marcel Dekker, Mirkin, B. G. and Rodin, S. N. (trans. H. 1. Beus) (1984) Graphs and Genes, Springer-
New York. Verlag, Berlin
Roberts, F. S. (1978) Graph Theory and Its Applications to Problems of Society, SIAM, Tutte, W. T. (1984) Graph Theory, Addison-Wesley, Reading, MA.
Philadelphia. Walther, D. (1984) Ten Applications of Graph Theory, Kluwer, Hingham, MA.
White, A. T. (1984) Graphs, Groups and Surfaces, Revised edn, North-Holland,
Amsterdam.
1979

Behzad, M., Chartrand, G. and Lesniak-Foster, 1. (1979) Graphs & Digraphs, 1985
Prindle, Weber & Schmidt, Boston.
Bollobas, B. (1979) Graph Theory: An Introductory Course, Springer, New York. Berge, C. (1985) Graphs, Second revised edn, North-Holland, New York.
Carrie, B. (1979) Graphs and Networks, Clarendon Press, Oxford. . Bollobas, B. (1985) Random Graphs, Academic Press, New York.
Chachra, V., Ghare, P. M. and Moore, J. M. (1979) Applications of Graph Theory Chartrand,G. (1985) Introductory Graph Theory, Dover Publications, New York.
Algorithms, North-Holland, New York. Gibbons, A. (1985) Algorithmic Graph Theory, Cambridge University Press.
Even, S. (1979) Graph Algorithms, Computer Science Press, Potomac, MD. Palmer, E. M. (1985) Graphical Evolution, John Wiley & Sons, New York.
Wilson, R. J. and Beineke, 1. W. (eds) (1979) Applications ofGraph Theory, Academic
Press, London.
1986

1980 Chartrand, G. and Lesniak, 1. (1986) Graphs & Digraphs, 2nd edn, Wadsworth &
Brooks/Cole, Menlo Park, CA. .
Cameron, P. J. and van Lint, J. H. (1980) Graphs, Codes, and Designs, Cambridge Lovasz, 1. and Plummer, M. D. (1986) Matching Theory, AkademiaiKiado, Budapest.
University Press, Cambridge. Yap, H.P. (1986) Some Topics in Graph Theory, London Mathematical Society
Golumbic, M. C. (1980) Algorithmic Graph Theory and Perfect Graphs, Academic Lecture Notes 108.
"Press, New York.
Graham, R. 1., Rothschild, B. 1. and Spencer, J. H. (1980) Ramsey Theory, Wiley-
Interscience, New York. 1987
Haggard, G. (1980) Excursions in Graph Theory, University of Maine, Orono.
Aigner, M. (1. Boron, C. Christenson, and B. Smith, translators) Graph Theory:
A Development from the 4-Color Problem, BCS Associates, Moscow, ID.
1981 Gross, J. 1. and Tucker, T. W. (1987) Topological Graph Theory, Wiley-Interscience,
New York..
Temperley, H. N. V. (1981) Graph Theory and Applications, Halstead Press, New Zykov, A. A. (1987) Osnovy Teorii Grafov, Nauka, Moscow.
York.

1988
1982
Beineke, 1. W. and Wilson, R. J. (oos) (1988) Selected Topics in Graph Theory 3,
Boffey, T. B. (1982) Graph Theory in Operations Research, MacMillan Press, London. Academic Press, London.
390 Graph theory books 1936-1996 Graph theory books 1936-1996 391
Gould, R. (1988) Graph Theory, Benjamin-Cummings, Menlo Park, CA. 1994
Lau, H. T. (1988) Algorithms on Graphs, Tab Books Inc., Blue Ridge Summit, PA.
Hartsfield, N. and Ringel, G. (1994) Pearls in Graph Theory: A Comprehensive
Introduction, Revised and augmented, Academic Press, Boston.
1989

Brower, A. E., Cohen, A. M. and Neumaier, A. (1989) Distance-Regular Graphs, 1995

Springer-Verlag, Berlin.
Bonnington, C. P. and Little, C. H. C. (1995) The Foundations of Topological Graph
Theory, Springer, New York.
1990 Jensen, T. R. and Toft, B. (1995) Graph Coloring Problems, John Wiley & Sons,
New York.
Bosak, J. (1990) Decompositions of Graphs, Kluwer Academic, Boston. Lovasz, L., Graham, R. L. and Grotschel, M. (eds) (1995) Handbook of Combinatorics,
Buckley, F. and Harary, F. Distance in Graphs, Addison-Wesley, Redwood City, Elsevier Science, Amsterdam.
CA.
Fleischner, H. (1990) Eulerian Graphs and Related Topics, Part 1, Vol. 1, Annals
of Discrete Mathematics 45, North-Holland, Amsterdam. 1996 (AND BEYOND)
Graham, R. L., Rothschild, B. L. and Spencer, J. H. (1990) Ramsey Theory, 2ndedn,
Wiley-Interscience, New York. Beineke, L. W. and Wilson, R. J. (eds) (1996) Graph Connections, Oxford University
Hartsfield, N. and Ringel, G. (1990) Pearls in Graph Theory: A Comprehensive Press.
Introduction, Academic Press, Boston. Chartrand, G. and Lesniak, L. (1996) Graphs and Digraphs, 3rd edn, Chapman &
Konig, D. (R. McCoart, translator) (1990) Theory of Finite and Infinite Graphs, Hall, London.
Birkhiiuser, Boston. Chartrand, G. and Jacobson, M. S. (eds) (1996) Surveys in Graph Theory, Congressus
McHugh, J. A. (1990) Algorithmic Graph Theory, Prentice-Hall, Englewood Cliffs, NJ. Numerantium, Winnipeg.
Ore, O. and Wilson, R. J. (1990) Graphs and their Uses (updated), Mathematical Haynes, T., Hedetniemi, S. T. and Slater, P. J. Domination in Graphs, Marcel Dekker,
Association of America, New York. New York. To appear.
Steinbach, P. (}990) Field Guide to Simple Graphs, Design Lab, Albuquerque. Haynes, T., Hedetniemi, S. T. and Slater, P. J. (eds) Domination in Graphs: Selected
Wilson, R. J. and Watkins, J. J. (1990) Graphs: An Introductory Approach: A First Topics, Marcel Dekker, New York. To appear.
Course in Discrete Mathematics, John Wiley & Sons, New York. McKee, T. A. and McMorris, F. R. Topics in Intersection Graph Theory. To appear.
Prisner, E. (1996) Graph Dynamics, Pitman, London.
West, D. B. (1996) Introduction to Graph Theory, Prentice-Hall, Upper Saddle
1991 River, NJ.
Zhang, C.-Q. Integer Flows and Cycle Covers of Graphs, Marcel Dekker, New York.
Clark, J. and Holton, D. A. (1991) A First Look at Graph Theory, World Scientific, To appear.
New Jersey.
Fleischner, H. (1991) Eulerian Graphs and Related Topics, Part 1, Vol. 2, Annals of
Discrete Mathematics 50, North-Holland, Amsterdam.
Voss, H.-J. (1991) Cycles and Bridges in Graphs, Kluwer Academic, Norweel, MA.

1992

Foulds, L. R. (1992) Graph Theory Applications, Springer, New York.

1993

Ahuja, R. K, Magnanti,T. L. and Orlin, J. (1993) Network Flows, Prentice-Hall,

Englewood Cliffs, NJ.
Biggs, N. (1993) Algebraic Graph Theory, 2nd edn, Cambridge University Press.
Chartrand, G. and Oellermann, O. R. (1993) Applied and Algorithmic Graph Theory,
McGraW-Hill, New York.
Holton, D. A. and Sheehan, J. (1993) The Petersen Graph, Cambridge University
Press.
 References 393
Battle, J.
[BHKYl] (with F. Harary, Y. Kodama and J. W. T. Youngs) Additivity of the
genus of a graph. Bull. Amer. Math. Soc. 68 (1962) 565-8. [188]

References Behzad,M.
[Bl] Graphs and Their Chromatic Numbers. Doctoral thesis, Michigan State
University (1965). [243]
Beineke,L. W.
[B2] Decompositions of complete graphs in forests. Magyar Tud. Adad. Mat.
Kutato Int. Kozl. 9 (1964) 589-94. [74]
[B3] The decomposition of complete graphs into planar subgraphs, in
Graph Theory and Theoretical Physics. Academic Press, New York
Each item in the bibliography is followed by one or more numbers in square (1967) 139-54. [181]
brackets indicating the pages on which the item is referenced. The starred entries [B4] Derived graphs and digraphs, in Beitriige zur Graphentheorie. Teubner,
denote books. Leipzig (968) 17-33. [126]
[BHl] (with F. Harary) The genus of the n-cube. Canad. J. Math. 17 (1965)
Acharya, B. D. 494-6. [189]
See [WASl]. [BH2] (with F. Harary) The thickness of the complete graph. Canad. J. Math.
Alekseev, V. B. 17 (1965) 850-9. [181]
[AGl] (with V. S. Gonchakov) Thickness of arbitrary complete graphs. Mlt. [BHMl] (with F. Harary and J. W. Moon) On the thickness of the complete
Sbornik 101 (1976) 212-30. [181] bipartite graph. Proc. Cambridge Phi/os. Soc. 60 (1964) 1-5. [181]
[BRl] (with R. D. Ringeisen) On the crossing number of products of
Allan, R. B. cycles and graphs of order four. J. Graph Theory 4 (1980) 145-55.
[ALl] (with R. Laskar) On domination and independent domination [181]
numbers of a graph. Discrete Math. 23 (1978) 73-6. [319] [BS1] (with A. J. Schwenk) On a bipartite form of the Ramsey problem.
[ALH1] (with R. Laskar and S. T. Hedetniemi) A note' on total domination. Congress. Numer. 15 (1975) 17-22. [369]
Discrete Math. 49 (1984) 7-13. [315] [BWl] (withR. J. Wilson) On the edge-chromatic number of a graph. Discrete
Alan, N. Math. 5 (1973) 15-20. [239]
*[ASl] (with J. H. Spencer) The Probabilistic Method. John Wiley & Sons, New See also [BBGLPll, [BBLP1] and [RBI].
York (1992). [38U Berge, C.
Alpert, S. R. [B5] Two theorems on graph theory. Proc. Nat. Acad. Sci. USA 43 (1957)
See [GAl]. 842-4. [260, 265]
[B6] Farbung von Graphen, deren samtliche bzw. deren ungeraden Kreise
Anderson, I. starr sind. Wissenschaftliche Zeitung, Martin Luther Univ., Halle Witten-
[Al] Perfect matchings of a graph. J. Combin. Theory lOB (1971) 18~. berg l14 (1961). [230]
[263] *[B7] The Theory of Graphs and its Applications. Methuen, London (1962).
Appel, K. . [58,304]
*[B8] Graphs and Hypergraphs. North-Holland, London (1973).
[AHKl] (with W. Haken and J. Koch) Every planar map is four-colorable.
Illinois J. Mlth. 21 (1977) 429-567. [250] [114, 230, 306, 308, 312]
See also [250]. [BF1] (with J. C. Fournier) A short proof for a generalization of Vizing's
theorem. J. Graph Theory 15 (1991) 333-6. [236]
Arumugam, S.
See [JAl]. Bielak, H.
[BS2] (with M. M. Syslo) Peripheral vertices in graphs. Studia Sci. Math.
Bagga, K. S. Huugar. 18 (1983)269-75. [23]
[BBGLPl] (with 1. W. Beineke, W. D. Goddard, M. J. Lipman and R. E. Pippert)
A survey of integrity. Discrete Appl. Math. 37/38 (1992) 13-28. [90] Biggs, N. 1.
[BBLPl] (with 1. W. Beineke, M. J. Lipman and R. E. Pippert) Edge-integrity: *[BLWl] (with E. K. Lloyd and R. J. Wilson) Graph Theory 1736-1936. Oxford
a survey. Discrete Appl. Math. 124 (1994) 3-12. [90] University Press, London (1976). [92, 103, 257]
Barefoot, C. A. Blass, A.
[BES1] (with R. Entringer and H. C. Swart) Integrity of trees and powers of [BH3] (with F. Harary) Properties of almost all graphs and complexes.
cycles. Congress Numer. 58 (1987) 103-14. [90] J. Graph Theory 3 (1979) 225-10. [378,380]
 j :

394 References References 395

Blazek, J. Buckley, F.
[BKl] (with M. Koman) A minimal problem concerning complete plane [BMSl] (with Z. Miller and P. J. Slater) On graphs containing a given graph as
graphs, in Theory of Graphs and its Applications. Academic Press, center. J. Graph Theory 5 (1981) 427-34. [21]
New York (1964) 113-7. [174]
Burr, S. A.
Blitch, P. [B16] An inequality involving the vertex arboricity and edge arboricity of a
See [SBl]. graph. J. Graph Theory 10 (1986) 403-4. [70,71]
[BEll] (with P. Erdos and L. Lovasz)On graphs of Ramsey type. Ars Combin.
Boesch, F. T. 1 (1976) 167-90. [361]
[BSTl] (With C. Suffel and R. Tindell) The spanning subgraphs of eulerian [BR3] (with J. A. Roberts) On Ramsey numbers for stars. Utilitas Math. 4
graphs. J. Graph Theory 1 (1977) 79--85. [100] (1973) 217-20. [359]
Bollobas, B. Camion,P.
[BCl] (with E. J. Cockayne) Graph-theoretic parameters concerning domina- [Cl] Chemins et circuits hamiltoniens des graphes complets. C. R. Acad.
tion, independence, and irredundance. J. Graph Theory 3 (1979) 241-9. Sci. Paris 249 (1959) 2151-2. [150]
[314,319]
[BC2] (with E. J. Cockayne) The irredundance number and maximum Caro, Y.
degree of a graph. Discrete Math. 49 (1984) 197-9. [306] See [279].
Bondy,J. A. Catlin, P. A.
[B9] Disconnected orientation and a conjecture of Las Vergnas. J. London [C2] Supereulerian graphs: a survey. J. Graph Theory 16 (1992) 177-96.
Math. Soc. 14 (1976) 277--82. [151] [101, 102]
[BI0] Pancyclic graphs. J. Combin. Theory lIB (1977) 80-4. .[116] Cayley, A.
[Bll] A remark on two sufficient conditions for Hamilton cycles. Dzscrete [C3] On the colouring of maps. Proc. Royal Geog. Soc. 1 (1879) 259--61.
Math. 22 (1978) 191-4. [112]
[249]
[B12] Longest paths and cycles in graphs of high degree. Research Report [C4] A theorem on trees. Quart. J. Math. 23 (1889) 376--8. Collected papers,
CORR 80-16, University of Waterloo (1980). [113]
Cambridge, 13 (1892) 26--8. [62]
[B13] Small cycle double covers of graphs. Cycles and Rllys. Kluwer See also [249].
Academic Publishers, Dordrecht(1990) 21-40. [102]
[BC3] (with V. Chvatal) A method in graph theory. Discrete Math. 15 (1976) Chartrand, G.
111-36. [106,107,113] [CGSWl] (with H. Gavlas, M. Schultz and C. E. Wall) Anticonnected digraphs.
Preprint. [135, 136]
Brandt, S. [CHI] (with F. Harary) Graphs with prescribed connectivities, in Theory of
[B14] Subtrees and subforests of graphs. J. Cambin. Theory 61B (1994) 63-70. Graphs: Proceedings ofthe Colloquium Held at Tihany, Hungary. Budapest
[62,342] (1968) 61-3. [76]
Brewster, R. C. [CHSl] (with M. A. Henning and K. Schultz) On planetary domination
[BCMl] (with E. J. Cockayne and C. M. Mynhardt) Irredundant Ramsey numbers of graphs. Preprint. [325, 321]
numbers for graphs. J. Graph Theory 13 (1989) 283-90. [363] [CHOl] (with H. Hevia and O. R. Oellermann) The chromatic number of a
factorization of a graph. Preprint. [291]
Broersma, H. J. [CHJKNl] (with A. M. Hobbs, H. A. Jung, S. F. Kapoor and C. St. J. A. Nash-
[BJVVl] (with B. Jackson, J. van den Heuvel and H. J. Veldman) Hamiltonicity Williams) The square of a block is hamiltonian-connected. J. Combin.
of regular 2-connected graphs. Preprint. [100] Theory 16B (1974) 290-2. [124]
[BWl] (with J. van den Heuvel and H. J. Veldman) A generalization of [CKLS1] (with S. F. Kapoor, L. Lesniak and S. Schuster) Near I-factors in
Ore's theorem involving neighborhood unions. Discrete Math. 122 graphs. Congress. Numer. 41 (1984) 131-47. [266]
(1995) 37-49. [111] [CKORl] (with S. F. Kapoor, O. R. Oellermann and S.. Ruiz) On maximum
Brooks, R. L. mathings in cubic graphs with a bounded number of bridge-covering
[B15] On coloring the nodes of a network. Proc. Cambridge Philos. Soc. 37 paths. Bull. Austral. Math. Soc. 36 (l986) 441-7. [267]
. (1941) 194-7. [223,225] [CKl] (with H: V. Kronk)The point arboricity of planar graphs. J. London
Math. Soc. 44 (1969) 612-6. [69]
Brualdi, R. A. [CPSl] (with A. D. Polimeri and M. J. Stewart) The existence of I-factors in
[BC4] (with J. Csima) A note on vertex- and edge-connectivity. Bull. Insf. line graphs, squares, and total graphs. Indag. Math. 35 (1973) 228-32.
Comb. Appl. 2 (1991) 67-70. [75] [95,279]
Bruck, R. H. Chetwynd, A. G.
[BR2] (with H. Ryser) The nonexistence of certain finite projective planes. [CHZ] (with R. Hiiggkvist) Some upper bounds on the total and list chromatic
Canad. J. Math. 1 (1949) 88-93. [280] numbers of multigraphs. J. Grf!ph Theory 16 (1992) 505-16. [244]
396 References References 397
[CH3] (with A. J. Hilton) Star multigraphs with three vertices of maximum [06] On rigid circuit graphs. Abh.Math. Sem. Univ. Hamburg 25 (961) 71-6.
degree. Math. Proc. Cambridge Philos. Soc. 100 (986) 303-17. [239] [228)
[D7] Homomorphism theorems for graphs. Math. Ann. 153 (1964) 69-80.
Chvatal, V. [334)
[C5] A remark on a problem of Harary. Czech. Math. J. 20 (970) 109-11. [D8] Short proof of Menger's graph theorem. Mathematika 13 (1966) 42-4.
[300] [80)
[C6] On Hamilton's ideals. J. Combin. Theory 12B (972) 163-8. [107] [DS1] (with S. Schuster) A theorem of Kuratowski. Nederl. Akad. Wetensch.
[C7] Tough graphs and Hamiltonian circuits. Discrete Math. 5 (1973) Proc. Ser. A 57 (954) 343-8. [163]
215-28. [87, 105] See also [334).
[C8] Tree-complete graph ramsey numbers. J. Graph Theory 1 (977) 93.
[358] Duke, R. A.
[CEll (with P. Erdos) A note on hamiltonian circuits. Discrete Math. 2 (1972) [D9] The genus, regional number, and Betti number of a graph. Canad. J.
111-3. [Ill] Math. 1.8 (1966) 817-22. [203)
See also [BC3J, and [125, 129). Dyck, W.
Chvatalova, J. [DIO] Beitrage zur Analysis Situs. Math. Ann. 32 (1888)457-512. (196)
[CDGKl] (with A. K. Dewdney, N. E. Gibbs and R. R. Korfhage) The bandwidth Edmonds, J.
problem for graphs: a collection of recent results. Research Report [Ell A combinatorial representation for polyhedral surfaces. Notices Amer.
#24, Department of Computer Science, UWO, London, Ontario (1975). Math. Soc. 7 (1960) 646. [196]
[299,300]
Eggleton, R. B.
Cockayne, E. J. [EGl] (with R. K. Guy) The crossing number of the n-cube. Notices Amer.
[CH4] (with S. T. Hedetniemi) Towards a theory of domination. Networks 7 Math. Soc. 17 (1970) 757. (178]
(1977) 247-61. [304,321]
[CFPTl) (with O. Favaron, C. Payan and A. Thomason) Contributions to the Elias, P.
theory of domination, independence and irredundance in graphs. [EFS1] (with A. Feinstein and C. E. Shannon) A note on the maximum flow
Discrete Math. 33 (981) 249-58. [322) through a network. IRE Trans. Inform. Theory IT-2 (956) 117-9. [83]
[CMl] (with C. M. Mynhardt) Irredundance and maximum degree in Enomoto,H.
graphs. Preprint. [319] [EJKSl] (with B. Jackson, P. Katernis and A. Saito) Toughness and the existence
See also [BCI], [BC2], [BCMl] and [308]. of k-factors. J. Graph Theory 9 (1985) 87-95. [265]
Csima, J. Entringer, R.
See [BC4]. See [BES1].
Dauber, E. Erdos, P.
See [48]. [E2] Some remarks on the theory of graphs. Bull. Amer. Math. Soc. 53 (1947)
de Jaenisch, c. F. 292-4. [359]
[E3] Graph theory and probability II. Canad. J. Math. 13 (1961) 346-52.
*[Dl] Applications de I'Analyse Mathematique an Jenudes Echecs. Petrograd
(1862). [302) [231]
[E4] Extremal problems in graph theory, in A Seminar in Graph Theory.
De Morgan, A. Holt, Rinehart & Winston, New York 0%7) 54-9. [335,336]
See [246] and [249]. [EG2] (with T. Gallai) On maximal paths and circuits of graphs. Acta Math.
Acad. Sci. Hungar. 10 (1959) 337-56. [340]
Descartes, B. [EG3] (with T. Gallai) Graphs with prescribed degrees of vertices
[D2] A three colour problem. Eureka 9 (947) 21. [230] (Hungarian). Mat. Lapok 11 (960) 264-74. [14]
[EK1] (withP. J. Kelly) The minimal regular graph containing a given graph,
Dewdney, A. K. in A Seminar on Graph Theory (F. Harary, ed.). Holt, Rinehart &
See [CDGKl]. Winston, New York (967) 65-9. [8]
Dirac, G. A. [ER1] (with R. Rado) A partition calculus in set theory. Bull. Amer. Math. Soc.
[D3] A property of 4-chromatic graphs and some remarks on critical 53 (1947) 292-4. [366]
graphs. J. London Math. Soc. 27 (952) 85-92. [221,232] [ES1) (with H. Sachs) Regulare Graphen gegebener Taillenweite mit mirii-
[D4] Some theorems on abstract graphs. Proc. London Math. Soc. 2 (1952) maIer Knotenzahl. Wiss. Z. Univ. Halle, Math-Nat. 12 (1963) 251-8.
69-81. [108] [344]
[05] In abstrakten Graphen vorhande vollstandige 4~Graphen und ihre [ES2] (with G. Szekeres) A combinatorial problem in geometry. Compositio
unterteilungen. Math. Nachr. 22 (960) 61-85. [82] Math. 2 (1935) 463-70. [352]
398 References References 399
IEWl] (with R. J. Wilson) On the chromatic index of almost all graphs. Folkman, J. H.
J. Combin. Theory 23B (1977) 255-7. [239] [F8] Graphs with monochromatic complete subgraphs in every edge
See also [BEll] and [CEl]. coloring. SIAM J. Appl. Math. 18 (1970) 19-24. [360]
Eroh, L. Ford, L. R., Jr.
See [135]. [FF1] (with D. R. Fulkerson) Maximal flow through a network. Canad. J.
Errera, A. Math. 8 (1956) 399-404. [83]
[E5] Du colorage des cartes. Mathesis 36 (1922) 56-60. [266] Fournier, J. C.
Euler, L. [F9] Methode et theoreme general de coloration des aretes d'um multi-
[E6] Solutio problematis ad geometriam situs pertinentis. Comment. graphe. J. Math. Pures Appl. 56 (1977) 437-53. [238]
Academiae Sci. 1. Petropolitanae 8 (1736) 128-40. [92,94] See also [BFl].
[E7] Demonstratio nonnullarum insignium proprietatum quibus solida Fraisse, P.
hedris planis inclusa sunt praedita. Novi Comm. Acad. Sci. Imp. Petro- [FlO] A new sufficient condition for hamiltonian graphs. J. Graph Theory 10
pol. 4 (1758) 140--60. [155] (1986) 405--9. [109]
See also [158].
Frucht, R.
Fary, I. [Fll] Herstellung von Graphen mit vorgegebener abstrakten Gruppe.
[Fl] On the straight line representations of planar graphs. Acta Sci. Math. Compositio Math. 6 (1938) 239-50.
11 (1948) 229-33. [157] [47]
Faudree, R. Fulkerson, D. R.
[F2] Forbidden subgraphs and hamiltonian properties, in Surveys in Graph See [FFl].
Theory. Preprint. [120] Fulman,J.
[FGJU] (with R. J. Gould, M. S. Jacobson and L. Lesniak) Lower bounds for [F12] A note on the characterization of domination perfect graphs. J. Graph
lower Ramsey numbers. J. Graph Theory 14 (1990) 723-30. [365] Theory 17 (1993) 47-51. [315]
[FGJL2] (with R. J. Gould, M. S. Jacobson and L. Lesniak) Neighborhood
unions and a generalization of Dirac's theorem. Discrete !0ath. 105 Gaddum, J. W.
(1992) 61-71. [111] See [NGl].
[FqSl] (with R. J. Gould, M. S. Jacobson and R. H. Schelp) Neighborhood Gallai, T.
unions and Hamiltonian properties in graphs. J. Combin. Theory 46B [Gl] Uber extreme Punkt- und Kantenmengen. Ann. Univ. Sci. Budapest,
(1989) 1-20. [109] E6tv6s Sect. Math. 2 (1959) 133-8. [267]
Favaron, O. [G2] On directed paths and circuits, in Theory of Graphs. Academic Press,
[F3] Two relations between the parameters of independence and New York (1968) 115--8. [225]
irredundance. Discrete Math. 70 (1988) 17-20. [320] See also [EG2] and [EG3].
See also [CFPTl] and [321]. Gavlas, H.
Feinstein, A. See [CGSW1].
See [EFSl]. Ghouila-Houri, A.
Fink, J. F. [G3] Une condition suffisante d'existence d'un circuit Hamiltonien. C. R.
[F4] Every graph is an induced isopart of a connected regular graph. Acad. Sci. Paris 156 (1960) 495--7. [120]
Preprint. [281] Gibbs, N. E.
[FR1] (with S. Ruiz) Every graph is an induced isopart of a circulant. Czech. See [CDGKl].
Math. J. 36 (1986) 172-6. . [281]
Fiorini, S. Gilbert, E. N.
*[FWl] (with R. J. Wilson) Edge-Colourings of Graphs. Pitman, London (1977). [G4] Random graphs. Ann. Math. Stat. 30 (1959) 1141-4. [377]
[243] Gimbel,J.
Fleischner, H. [GVl] (with P. D. Vestergaard) Inequalities for total matchings of graphs.
[F5] The square of every two-connected graph is hamiltonian. J. Combin. Ars Combin. To appear. [316]
Theory 16B (1974) 29-34. [123] Goddard, W.
[F6] Eulerian graphs, in Selected Topics in Graph Theory. Academic Press, [GHSl] (with M. A. Henning and H. C. Swart) Some Nordhaus-Gaddum-
New York (983) 17-53. [102] type results. J. Graph Theory 16 (1992) 221-31. [309]
"[F7] Eulerian Graphs and Related Topics, Part I, Vol. 1, Annals of Discrete [GSl] (with H. C. Swart) Integrity in graphs: bounds and basics. J. Comb.
Mathematics 45, North-Holland, Amsterdam (1990). [102] Math. Cambin. Camput. 7 (1990) 139.,.51. [89]
See also [125]. See also [BBGLPl].
400 References References 401
Golomb, S. W. Hajos, G.
[G5] How to number a graph, in Graph Theory and Computing. Academic [H2] Ober e,inen Satz von K. Wagner zum Vierfarbenproblem. Math. Ann.
Press, New York (1972) 23-37. [290] 153 (1964) 47-62. [232]
Golumbic, M. C. Haken, W,
'[G6] Algorithmic Graph Theory and Perfect Graphs. Academic Press, New See [AHK1] and [250].
York (1980). [234]
Hakimi, S. L.
Gonchakov, V. S. [H3] On the realizability of a set of integers as degrees of the vertices of a
See [AG1], graph. J. SIAM App!. Math. 10 (1962) 496-506. [12]
Gould, R. J. [HMSl] (with J. Mitchem and E. F. Schmeichel) Short proofs of theorems of
[G7] Updating the hamiltonian problems-a survey. J. Graph Theory i5 Tutte and Nash-Williams. Preprint. [71]
(1991) 121-58. See also [FGJLI], [FGJL2] and [FGJS1]. [120] Halin, R.
Graham, R. L. [H4] Bemerkungen tiber ebene Graphen. Math. Ann. 53 (1964) 38-46.
•[GRSl] (with B. L. Rothschild and J. H. Spencer) Ramsey Theory, 2nd edn. [167]
Wiley-Interscience, New York (1990). [361] Hall, P.
[GS2] (with N. J. A. Sloane) On additive bases and harmonious graphs. [H5] On representation of subsets. J. London Math. Soc. 10 (1935) 26-30.
SIAM]. Alg, Disc. Meth. 1 (1980) 382-404. [293,294,295] [261,262]
Grinberg, E. J.
[G8] Plane homogeneous graphs of degree three without hamiltonian Hamilton, W. R.
circuits. Latvian Math. Yearbook 4 (1968) 51-8. [170] See [103] and [246].
See also [173]. Harary, F.
Gross, J. L. [H6] Graphs and matrices. SIAM Review 9 (1967) 83-90. [64]
[GAl] (with S. R. Alpert) Branched coverings of graph imbeddings. Bull. •[H7] Graph Theory. Addison-Wesley, Reading, MA (1969). [43,58]
Amer. Math. Soc. 79 (1973) 942-5. [216] [HKS1] (with P. C. Kainen and A. J. Schwenk) Toroidal graphs with arbitrarily
'[GTl] (with T. W. Tucker) Topological Graph Theory. Wiley-Interscience, New high crossing numbers. Nanta Math. 6 (1973) 58-67. [178]
York (1987). [218] [HMl] (with L. Moser) The theory of round robin tournaments. Amer. Math.
Guo, Y. Monthly 73 (1966) 231-46. [145, 150]
[GV2]
[HNl] (with C. St. J. A. Nash-Williams) On eulerian and hamiltonian graphs
(with L. Volkmann) Cycles in multipartite tournaments. Preprint. and line graphs. Canad. Math, Bull. 8 (1965) 701-10. [126]
[151] [HN2] (with R. Z. Norman) The dissimilarity characteristic of Husimi trees.
Guthrie, Francis Ann. of Math. 58 (1953) 134-41. [37]
See [246] and [249]. [HTl] (with W. T. Tutte) A dual form of Kuratowski's Theorem. Canad. Moth.
Guthrie, Frederick Bull. 8 (1965) 17-20, 373. [167]
[G9] Note on the colouring of maps. Proc. Royal Soc. Edinburgh 10 (1880) See also [BHKY1], [BHl], [BH2], [BHMl], [BH3], [CHI] and [50].
727-8. [246] Hattingh, J. H.
See also [246]. [HHl] (with M. A. Henning) Bipartite ramsey numbers. Preprint. [366]
Guy, R. K.
[GI0] Havel, V.
A combinatorial problem. Bull. Malayan Math. Soc. 7 (1960) 68-72. [H8] A remark on the existence of finite graphs (Czech.) Casopis pest. Mat.
[174] 80 (1955) 477-80. [12]
[Gll] Crossing number of graphs, in Graph Theory and Applications.
Springer-Verlag, New York (1972) 111-4. [174,176] Hayes, L.
See [EG1]. [HSYl] (with K. Schultz and J. Yates) Universal dominating sequences in
Hadwig~r, R: graphs. Preprint. [324, 325]
[HI] Uber eine Klassifikation der Streckenkomplexe. Vierteljschr. Natur- Haynes, T.
forsch Ges. Zurich 88 (1943) 133-42. [232] '[HHS1] (with S. T. Hedetniemi and P. J. Slater) Domination in Graphs. Marcel
Haggkvist, R. Dekker, New York. To appear. [327]
See [CH2].
Hajnal, A. Heawood, P. J.
[H9] Map-colour Theorem. Quart. ]. Math. 24 (1890) 332-9. [249, 257]
[HSl] (with J. Suranyi) Uber die Auflosung von Graphen in vollstiindige
Teilgraphen. Ann. Univ. Sci. Budapest, Eotvos Sect. Math. 1 (1958) Hedetniemi, S. T,
113-21. [228] See [ALHlJ, [CH4] and [HHSl].
402 References References 403
Heffter, L. Jaeger, F.
[HI0] Uber das Problem der Nachbargebiete. Math. Ann.. 38 (1891) 477-508. [J2] On nowhere-zero flows in multigraphs. Congress. Numer. 15, Winnipeg
[196,257] (1976) 373-8. [257]
[J3] A note on sub-eulerian graphs. J. Graph Theory 3 (1979) 91-3.
Hendry, G. R. T. [99,101]
[Hll] On graphs with prescribed median I. J. Graph Theory 9 (1985) 477-81. [JPl] (with C. Payan) Relations du type Nordhaus-Gaddum pour Ie
[23] nombre d'absorption d'un graphe simple. C. R. Acad. Sci. Ser. A. 274
(1972) 728-30. [308]
Henning, M. A.
[H12] Irredundance perfect graphs. Discrete Math. 142 (1995) 107-20. [321] Jensen, T. R.
[HOSl] (with O. R. Oellermann and H. C. Swart) Bounds on distance domina- ['JTl] (with B. Toft) Graph Coloring Problems. John Wiley & Sons, New York
tion parameters. J. Combin. Inf. Sys. Sci. 16 (1991) 11-18. [323,3241 (995). [234,257]
[HOS2] (with O. R. Oellermann and H. C. Swart) Relating pairs of distance
domination parameters. J. Combin. Math. Combin. Comput. 18 (995) Jones, D. M.
233-44. [324] [JRSl] (with D. J. Roehm and M. Schultz) On matchings in graphs. Preprint.
See also [CHSlJ, [GHSl] and [HH1]. [270]

Hevia, H. Joseph, J. P.
See [CHOn [JAIl (with S. Arumugam) A note on domination in graphs. Preprint.
[310]
Hierholzer, C.
[H13] Uber die M6glichkeit, einen Linienzug ohne Wiederholung und ohne Jung, H. A.
Unterbrechnung zu umfahren. Math. Ann. 6 (1873) 30-2. [94] See [CHJKNl].

Hilton, A. J. Jungerman, M.
[H14] Recent progress in edge-colouring graphs. Discrete Math. 64 (1987) []4] A characterization of upper embeddable graphs. Trans. Amer. Math.
303-7. [239] Soc. 241 (1978) 401-6. [209]
See also [CH3]. Kainen, P. C.
Hobbs, A. M. See [HKSl] and [177].
See [CHJKNl]. Kapoor, S. F.
Hoffman, A. J. [KPWl] (with A. D. Polimeni and C. E. Wall) Degree sets for graphs. Fund.
[HS2] (with R. R. Singleton) On Moore graphs with diameters two and three. Math. 95 (977) 189-94. [15]
IBM J. Res. Develop. 4 (1960) 497-504. [347] See also [CHJKNl], [CKLS1] and [CKORl].
Holton, D. A. Karaganis, J. J.
'[HS3] (with J. Sheehan) The Petersen Graph. Cambridge University Press [K1] On the cube of a graph. Canad. Math. Bull. 11 (1968) 295-6. [123]
(1993). [343,347,349] Katemis, P.
Hoogeveen, H. See [EJKSl].
See [SHl]. Kelly, J. B.
Isaacs, R. [KK1] (with L.M. Kelly) Paths and circuits in critical graphs. Amer. J. Math.
[Ill Infinite families of nontrivial trivalent graphs which are not Tait 76 (1954) 786-92. [230]
colorable. Amer. Math. Monthly 82 (1975) 221-39. [239] Kelly, L. M.
Izbicki, H. See [KKl].
II2] Reguliire Graphen beliebigen Grades mit Vorgegebenen Eigen- Kelly, P. J.
schaften. Monatsh. Math. 64 (1960) 15-21. [48] See [EK1] and [50].
Jackson, B. Kempe, A. B.
[Jl] Hamilton cycles in regular 2-connected graphs. J. Combin. Theory 29B [K21 On the geographical problem of the four colors. Amer. J. Math. 2 (1879)
(1980) 27--46. [l08] 193-200. [249]
See also [BJVVl] and [EJKSl]. [K3] How to color a map with four colors. Nature 21 (1880) 399--400.
Jacobson, M. S. [249J
See [FGJU], [FGJL2] and [FGJSl]. See also [249].
404 References References 405
Lesniak, L.
Kirchhoff, G.
[K4] Uber die Auflosung der Gleichungen, auf welche man bei der Unter- [L2] Hamiltonicity in some special classes of graphs, in Surveys in Graph
suchung der linearen Verteilung galvanischer Strome gefiihrt wird. Theory. Preprint. [120]
Ann. Phys. Chem. 72 (1847) 497-508. Gesammelte Abhandlungen, [L01] (with O. R. Oellermann) An eulerian exposition. J. Graph Theory 10
Leipzig (1882) 22-3. [64] (1986) 277-97. [102]
[LSI] (with H.]. Straight) The cochromatic number of a graph. Ars Combin. 3
Kirkman, T. P. (1977) 39-46. [233]
[K5] On a problem in combinations. Cambridge and Dublin Math. J. 2 (1847) See [CKLS1], [FGJLl] and [FG]L2].
191-204. [280]
Lipman, M. J.
See also [103].
See [BBGLP1] and [BBLP1].
Kleinert, M. Lloyd, E. K.
[K6] Die Dicke des n-dimensionalen Wiirfel-Graphen. J. Combin. Theory 3 See [BLW1].
(1967) 10-15. [181]
Lovasz, L.
Kleitman, D.]. [13] On chromatic number of finite set-systems. Acta Math. Acad. Sci.
[K7] The crossing number of KS,/f' J. Combin. Theory 9 (1970) 315-23. Hungar. 79 (1967) 59-67. [231]
[177]
[14] Normal hypergraphs and the perfect graph conjecture. Discrete Math.
Koch,]. 2 (1972) 253-67. [230]
See [AHK1] and [250]. [15] Three short proofs in graph theory. J. Combin. Theory 19B (1975)
269-71. [223]
Kodama, Y. *[L6] Combinatorial Problems and Exercises. North-Holland, Amsterdam
See [BHKY1]. (1979). [337]
Koman,M. See also [BELl].
See [BK1]. Manvel, B.
Konig, D. [M1] On Reconstruction of Graphs. Doctoral thesis, University of Michigan
[K8] Uber Graphen und ihre Anwendung auf Determinantheorie und (1m~ ~
Mengenlehre. Math. Ann. 77 (1916) 453-65. [6,292] Marczewski, E.
[K9] Graphen und Matrizen. Math. Riz. Lapok. 38 (1931) 116-19. [M2] Sur deux proprietes des classes d'ensembles. Fund. Math. 33 (1945)
[261,268]
303-7. [227]
'[K10] Theorie der endlichen und unendlichen Graphen, Teubner, Leipzig.
Reprinted Chelsea, New York (1950). [6,47] Matthews, M. M.
[MS1] (with D. P. Sumner) Hamiltonian results in K1s-free graphs. J. Graph
Korfhage, R. R. Theory 8 (1984) 139-46. ' . [87, 129]
See [CDGK1].
May,K. O.
Kotzig, D. [M3] The origin of the four-color conjecture. Isis 56 (1965) 346-8. [246]
See [284].
McCuaig, W.
Kronk, H. V. [MS2] (with B. Shepherd) Domination in graphs with minimum degree two.
[KRW1] (with R. D. Ringeisen and A. T. White) On 2-cell embeddings of J. Graph Theory 13 (1989) 749-62. [305]
complete n-partite graphs. Colloq. Math. 36 (1976) 295-304. [211]
See also [CK1] and [232]. McDiarmid, C. J. H.
Kuratowski, K.
[MR1] (with B. Reed) On total colourings of graphs. J. Combin. Theory B.
To appear. [244]
[Kll] Sur Ie probleme des courbes gauches en topologie. Fund. Math, 15
(1930) 271-83, [163] McKay, B. D.
Landau, H. G.
[M4] Computer reconstruction of small graphs. J. Graph Theory 1 (1977)
281-3. . [51]
[LlI On dominance relations and the structure of animal societies. III. The
condition for a score structure. Bull. Math. Biophys. 15 (1953) 143-8. McKee, T. A.
[143,147] [M5] Recharacterizing Eulerian: intimations of new duality. Discrete Math.
51 (1984) 237-42. [96]
Laskar, R.
[LP1] (with J. Pfaff) Domination and irredundance in graphs. Tech. Rept. Menger, K.
434, Dept. Mathematical Sciences, Clemson University (1983). [3191 [M6] Zur allgeminen Kurventheorie. Fund. Math. 16 (1927) 95-115. [80]
See also [ALl] and [ALH1]. See also [57].
406 References References 407
Meyniel, M. [NSW1] (with B. M. Stewart and A. T. White) On the maximum genus of a
[M7] Une condition suffisante d'existence d'un circuit Hamiltonien dans graph. J. Combin. Theory HB (1971) 258--67. [207, 212]
un graph oriente. J. Combin. Theory 14B (1973) 137--47. [119]
Norman, R. Z.
Miller, Z. See [HN2].
See [BMS1].
Oellermann, O. R.
Mirsky, L. See [CHOlJ, [CKORl], [HOS1), [HOS2] and [LOl].
'[M8] Transversal Theory. Academic Press, New York (1971). [262]
Ore,O.
Mitchem, J. [01] Note on Hamilton circuits. Amer. Math. Monthly 67 (1960) 55. [105]
See [HMS1]. '[02] Theory of Graphs. Amer. Math. Soc. Colloq. Pub., Providence, Rl 38
(1962). [304]
Moon,J. W. [03] Hamilton connected graphs. J. Math. Pures App/. 42 (1963) 21-7.
[M9] On subtournaments of a tournament. Canad. Math. Bull. 9 (1966) [114)
297-301. [150]
[M10] On independent complete subgraphs in a graph. Canad. J. Math. 20 Palmer, E. M.
(1968) 95-102. [340] '[PI] Graphical Evolution. John Wiley & Sons, New York (1985). [381]
See also [BHM1]. Parthasarathy, K. R.
Moore,]. L. [PRl] (with G. Ravindra) The strong perfect graph conjecture is true for
See [SM1]. K1,3-free graphs. J. Combin. Theory 22B (1976) 212-23. [230]

Moser, L. Payan, C.
See [HM1] and [145]. [P2] Sur Ie nombre d'absorption d'un graphe simple. Cahiers Centre Etudes
Recherche Oper. 17 (1975) 307-17. [305,373]
Mycielski, J. [PXll (with N. H. Xuong) Domination-balanced graphs. J. Graph Theory 6
[M11] Sur Ie coloriage des graphes. Colloq. Math. 3 (1955) 161-2. [230] (1982) 23-32. [305]
See also [CFPTl] and UP1].
Mynhardt, C. M.
[M12] Generalised maximal independence and clique numbers of a graph. Petersen, J.
Quaestiones Math. 11 (1988) 383-98. [364] [P3] Die Theorie der regularen Graphen. Acta Math. 15 (1891) 193-220.
[M13] Lower Ramsey numbers for graphs. Discrete Math. 91 (1991) 69-75. [265,266,274]
[365] [P4] Sur Ie theoreme de Tait. L'Intermediaire des Mathematiciens 5 (1898)
See also [BCM1] and [CMIl. 225-7. [272]
Nash-Williams, C. St.]. A. pfatt, J.
[N1] Decomposition of finite graphs into forests. J. London Math. Soc. 39 See [LP1].
(1964) 12. [71]
Pippert, R. E.
See also [CHJKNl], [HN1] and [123]. See [BBGLP1] and [BBLP1].
Nebesky, L. Plesnik, J.
[N2] A theorem on hamiltonian line graphs. Comment. Math. Univ. Carolinae [P5] Critical graphs of given diameter. Acta Fac. Rerum Natur. Univ.
14 (1973) 107-11. [129] Comenian. Math. 30 (1975) 71-93. [77]
[N3] On pancyclic line graphs. Czech. Math. J. 28 (1978) 650-5. [129]
!N4] On squares of complementary graphs. Math. Slovaca 30 (1980) 247-9. Plummer, M. D.
[124] See [4.34.12].
Nesetfil, J. Polimeni, A. D.
[NR1] (with V. Rod!) The Ramsey property for graphs with forbidden com- See [CPS1] and [KPWIl.
plete subgraphs. J. Combin. Theory 20B (1976) 243-9. [360] Posa, L.
Nijenhuis, A. See [335) and [336].
[N5] ,Not~ on the unique determination of graphs by proper subgraphs. Powell, M. B.
Notices Amer. Math. Soc. 24 (1977) A-290. [51] See [WP1].
Nordhaus, E. A. Priifer, H.
[NG1) (with J. W. Gaddum) On complementary graphs. Amer. Math. Monthly [P6] Neuer Beweis eines Satzes iiber Permutationen. Arch. Math. Phys. 27
63 (1956) 175-7. [232) (1918) 142--4. [62]
408 References References 409
Rado, R. Rothschild, B. L.
See [ER1). See [GRS1].
Ramsey, F. Ruiz,S.
[Rl] On a problem of formal logic. Proc. London Math. Soc. 30 (1930) 264-86. [Rlll Isomorphic decompositions of complete graphs into linear forests.
[251] J. Graph Theory 9 (1985) 189-91. . [285]
Ravindra, G. [R12] Randomly decomposable graphs. Discrete Math. 57 (1985) 123-8.
See [PRl). [279]
See also [CKOR1] and [FRl).
Redei, L.
[RZ] Ein kombinatorischer Satz. Acta Litt. Szeged 7 (1934) 39--43. [149] Ryser, H.
Reed, B. '[R13] Combinatorial Mathematics. Wiley, New York (1963). [280]
[R3] Paths, stars, and the number three. Preprint. See also [BRZ].
[305]
See also [MR1]. Sachs, H.
Ringeisen, R. D. See [ESl).
[R4] Determining all compact orientable 2-manifolds upon which Km,n has Saito, A.
2-cell imbeddings. J. Combin. Theory 12B (1972) 101--4. [201,212] See [EJKS1].
[RB1] (with L. W. Beineke) The crossing number of C3 x Cn- J. Combin.
Theory 24B (1978) 134-6. [179] Sampathkumar, E.
See also [BR1] and [KRW1]. See [WASl].
Ringel, G. Sanders, D.
[R5] Uber drei kombinatorische Probleme am n-dimensionalen Wiirfel See [RSSTl].
und Wiirfelgitter. Abh. Math. Sem. Univ. Hamburg 20 (1955) 10-19. Schelp, R. H.
[189] See [FGJS1].
[R6] Problem 25. Theory of graphs and its applications. Proc. Int. Symp.
Smolenice 1963, Prague (1964) 162. [285] Schmeichel, E. F.
[R7] Das Geschlecht des vollstandigen paaren Graphen. Abh, Math. Bem. See [HMS1].
Univ.Hamburg 28 (1965) 139-50. [189] Schultz, K.
'[R8] Map Color Theorem. Springer-Verlag, Berlin (1974). [189,258] See [CHS1] and [HSY1].
[RY1] (with J. W. r.Youngs) Solution of the Heawood map-coloring
Schultz, M.
problem. Proc. Nat. Acad. Sci. USA 60 (1968) 438--45. [189,257]
See also [177]. See [CGSW1] and [JRS1].
Robbins, H. E. Schuster, S.
[R9] A theorem on graphs, with an application to a problem in traffic See [CKLS1] and [DS1].
control. Amer. Math. Monthly 46 (1939) 281-3. [132] Schwenk, A. J.
Roberts, J. A. See [BS1], [HKS1] and [332].
See [BR3]. Sekanina, M.
Robertson, N. [Sl] On an ordering of the set of vertices of a connected graph. Publ. Fac.
[RSSTl] (with D. Sanders, P. D. Seymour and R. Thomas) The four-colour Sci. Univ. Bmo. 412 (1960) 137--42. [123]
theorem. Preprint. [251] Seymour, P. D.
[RSTl] (with P. D. Seymour and R. Thomas) Hadwiger's conjecture for IY,- [S2] On multicolourings of cubic graphs, and conjectures of Fulkerson and
free graphs. Combinatorica 13 (1993) 279-361. [232] Tutte. Proc. London Math. Soc. 38 (1979) 423-60. [243]
Rodl, V. [$3] Sums of circuits, in Graph Theory and Related Topics. Academic Press,
See [NRl]. New York (1979) 341-55. [102]
Roehm, D.]. [54] Nowhere-zero 6-flows. J. Combin. Theory 30B (1981) 130-5. [257]
See [JRS1]. [55) On Tutte's extension of the four-colour problem. J. Combinatorial
Theory 31B (1981) 82-94. [257]
Rosa, A. [56] Progress on the four-colour theorem. Preprint. [257]
[RI0] On certain valuations of the vertices of a graph, in Theory of Graphs, 5ee also [RSSTl] and [R5Tl].
Proc. Internat. Sympos. Rome 1966. Gordon and Breach, New York
Shannon, C. E.
(1967) 349-55. [281; 283,284,289]
See [EFSl].
410 References References 411
Sheehan, J. [SWl] (with H. 5. Wil£) An inequality for the chromatic number of a graph.
See [HS3]. J. Combin. Theory 4 (1968) 1-3. [225]
Shepherd, B. See also [E52].
See [MS2].
Sze1e, T.
Sierksma, G. [S12] Kombinatorikai vizsg<ilatok az ininyitott teljes graftel kapcsolatban.
[SHIl (with H. Hoogeveen) Seven criteria for integer sequences to be Mat. Fiz. Lapok 50 (1943) 223-56. [372]
graphic. J. Graph Theory 15 (1991) 223-31. [15] [513] Kombinatorische Untersuchungen tiber gerichtete vollstandige
Graphen. Pub/. Math. Debrecen. 13 (1966) 145-68. [150]
Singleton, R. R.
See [HS2]. Tait, P. G.
Slater, P. J. [Tl] Remarks on the colouring of maps. Froc. Royal Soc. London 10 (1880)
[S7] Medians of arbitrary graphs. ]. Graph Theory 4 (1980) 389-92. [23] 729. [251]
See also [BMSIl and [HHSl]. See also [17l}.

Sloane, N. J. A. Thomas, R.
See [GS2]. See [RSSTl] and [RSTl].
Spencer, J. H. Thomason, A.
[58] Ten Lectures on the Probabilistic Method. Capital Press, Montpelier [T2] On finite Ramsey numbers. Europ.]. Combin. 3 (1982) 263-72. [369]
(1994). [38Il See also [CFPTl].
See also [ASIl and [GR51].
Stewart, B. M. Thomassen, C.
See [NSWl]. [T3] Some homeomorphism properties of graphs. Math. Nachr. 64 (1974)
119-33. [334]
Stewart, M. J. [T4] Landau's characterization of tournament score sequences, in The
See [CP5l}. Theory and Applications of Graphs. Wiley, New York (1981) 589-91.
Stockmeyer, P. J. [143]
[S9] The falsity of the reconstruction conjecture for tournaments. J. Graph [T5] A theorem on paths in planar graphs. J. Graph Theory 7 (1983) 169-76.
Theory 1 (1977) 19-25. [52] [171]
[T6] Reflections on graph theory. J. Graph Theory 10 (1986) 309-24.
Straight, H. J. [129]
See [LSI]. See also [116] all,d [150].
Suffel, C. Tindell, R.
See [B5Tl]. See [BSTI].
Sumner, D. P.
[SlO] Critical concepts in domination. Discrete Math. 86 (1990) 22-46. Toft, B.
. [315] See UTI].
[SB1] (with P. Blitch) Domination critical graphs. J. Combin. Theory 34B Toida, S.
(1983) 65-76. [310,311] [T7] Properties of an Euler graph. J. Franklin Inst. 295 (1973) 343-5. [96]
[SMl] (with J. 1. Moore) Domination perfect graphs. Notices Amer. Math. Soc.
26 (1979) A-569. [315] Tucker, A,
See also [MSl}. [T8] The strong perfect graph conjecture for planar graphs. Canad. J. Math.
25 (1973) 103-14. . [230]
Suranyi, J.
See [HSl]. Tucker, T. W.
Swart, H. C. See [GTl].
See [BESl],[GHSl], [GSl], [HOSl] and [HOS2].
Turan, P.
Syslo,M. M. [T9J Eine Extremalaufgabe aus der Graphentheorie. Mat. Fiz. Lapok 48
See [BS2]. (1941) 436-52. [330,332]
Szekeres, G. [TIO] A note of welcome. J. Graph Theory 1 (1977) 7-9. [176]
[511] Polyhedral decomposition of cubic graphs. Bull. Austral. Math. Soc. 8
(1973) 367-87. [102] Tutte, W. T.
[TIl] On hamiltonian circuits. J. London Math. Soc. 2 (1946) 98-101. [171]
412 References References 413
[TI2] A contribution to the theory of chromatic polynomials. Canad. J. Math. Welsh, D. J. A.
6 (1954) 80-91. [255] [WPll (with M. B. Powell) An upper bound for the chromatic number of a
[TI3] A short proof of the factor theorem for finite graphs. Canad. J. Math. 6 graph and its application to timetabling problems. Computer J. 10
(1954) 347-52. [263] (1967) 85-7. [225]
[TI4] A theorem on planar graphs. Trans. Amer. Math. Soc. 82 (1956) 99-116.
White, A. T.
[171]
[W4] The proof of the Heawood Conjecture, in Selected Topics in Graph
[TIS] On the algebraic theory of graph colorings. J. Combin. Theory 1 (1966)
Theory. Academic Press, New York (1978) 51-81. [258]
15-50. [254, 256]
*[W5] Graphs, Groups and Surfaces, Revised edn. North-Holland, Amsterdam
See also [HTI].
(1984). [47,178,218]
Ulam,S. M. See also [KRWl] and [NSWl].
See [50]. Whitney,H.
van den Heuvel, J. [W6] Congruent graphs and the connectivity of graphs. Amer. J. Math. 54
See [BJVVl] and [BVVl]. (1932) 150-68. [75,81]
Vasak, J. Wilf, H. S.
[VI] See [SWl].
The thickness of the complete graph having 6m+4 points. Preprint.
[181] Williamson, J. E.
[W7] Panconnected graphs II. Period. Math. Hungar. 8 (1977) 105-16.
Veblen,O. [115]
[V2] An application of modular equations in analysis situs. Math. Ann. 14
(1912-1913) 86-94. [96] Wilson, R. J.
See [BWlj, [BLWlj, lEWl] and [FWl].
Veldman, H. J.
See [BJVVl] and [BVVl]. Wilson, R. M.
[W8] Decompositions of complete graphs into subgraphs isomorphic to a
Vestergaard, P. D. given graph. Proceedings of the Fifth British Combinatorial Conference
See [GVl]. (1975) 647-59. [280]
Vizing, V. G. Wojcicka, E.
[V3] On an estimate of the chromatic class of a p-graph. Diskret. Analiz. 3 [W9] Hamiltonian properties of domination-critical graphs. J. Graph Theory
(1964) 25-30. [236,238] 15 (1991) 109-14. [311]
[V4] Critical graphs with a given chromatic class. Diskret. Analiz. 5 (1965) Woodall, D. R.
9-17. [240, 242] [W1O] Sufficient conditions for circuits in graphs. Proc. London Math. Soc. 24
[V5] A bound on the external stability number of a graph. Doklady A. N.164 (1972) 739-55. [119]
(1965) 729-31. [307] [W11] Cyclic-order graphs and Zarankiewicz's crossing-number conjecture.
See also [311]. Preprint. [177]
Volkmann, L. Xuong,N.H.
See [GV2]. [Xl] How to determine the maximum genus of a graph. J. Combin. Theory
Wagner, K. 26B (1979) 217-25. [209,210]
[WI] Bemerkungen zum Vierfarbenproblem. Jber. Deutsch. Math. Verein. 46 See also [PXl].
(1936) 21-2. [157] Yates, J.
[W2] Uber eine Eigenschaft der ebene Komplexe. Math. Ann. 114 (1937) See [HSYl].
570-90. [167, 232, 251]
Youngs, J. W. T.
Walikar, H. B. [Yl] Minimal imbeddings and the genus of a graph. J. Math. Mech. 12
[WASl] (with B. D. Acharya and E. Sampathkumar) Recent developments (1963) 303-15. [187, 196]
in the theory of domination in graphs. Mehta Research Institute See also [BHKYl] and [RYl].
Allahabad, MRI Lecture Notes in Math. 1 (1979). [306,307]
Zaks, J.
Wall, C. E. [Zl] The maximum genus of cartesian products of graphs. Canad. J. Math.
See [CGSWl] and [KPWl]. 26 (1974) 1025-35. [212]
Weinstein, J. M. Zarankiewicz, K.
[W3] On the number of disjoint edges in a graph. Canad. f. Math. 15 (1963) [22] On a problem of P. Turan concerning graphs. Fund. Math. 41 (1954)
106-11. [269] 137-45. [177]
I
414 References
Zverovich, L R .. f
[221] (with V. E. ZveroV'ich) An induced subgraph charactenzatlOn 0
domination perfect graphs. J. Graph Theory 20 (1995) 375-95.
1-
[315)
Zverovich, V. E.
See [ZZI].
Index
Zykov, A. A. " . Ma Sb 'k
[Z3] On some properties of linear complexes (RussIan). t. arm 24
(1949) 163-88. Amer. Math. Soc. Translations No. 79 (1952). [230]

Acyclic block 37 complete 8

Acyclic graph 18 Bipartite Ramsey number 366
Adjacency matrix Block 35
of a digraph 32 acyclic 37
of a graph 1 critical 37
Adjacent edges 1 cyclic 37
Adjacent from (vertices in a digraph) 26 end- 37
Adjacent to (vertices in a digraph) 26 Ininimal38
Adjacent vertices 1 Block-cut-vertex graph 67
Almost all graphs 377 Blue degree 366
M-Altemating path 260 Boundary element 216
Anticonnected digraph 29 Boundary of a region 155
Antidirected semipath 29 Bridge 34
Antihamiltonian digraph 153
Antihamiltonian-eonnected digraph 153 Cage
Antipath 29 g- 343
Arboricity 70 (r,g)- 343
edge-70 Capping opetation 186
vertex- 68 Cartesian product of two graphs 9
Arc 25 Caterpillar 61
Arc-deleted subdigraph 27 Cayley color graph 45
Arc-induced subdigraph 27 Cayley graph 49
Arrows 360 Cayley'S theorem 64
Asymmetric digraph 27 2-Cell embedding 185
Attacking queens 302 2-Cell region 183
M-Augmenting path 260 Center
Automorphism of a (strong) digraph 32
color-preseJ:ving 46 ofa graph 20
of a digraph 44 Central appendage number of a graph
of a graph 40 24
Automorphism group Central tree 67
of a digraph 44 Central vertex 20
of a graph 40 k-Choosable 234
,,!,';
Chord of a cycle 116
Balanced incomplete block design 280 Chordal graph 228

I
;:";~";>:" Bandwidth labeling 297 k-Chromatic 220
Bandwidth of a graph 296 . critically 221
, Bandwidth of a numbering 296
, Ininimally 221
Betti number 207 Chromatic index 236
Bicentral tree 67 Chromatic number 220
Bigraphical sequences 16 edge 236
Bipartite graph 8 list- 233
416 Index Index 417
of a surface 257 Connectivity 74 Degree matrix of a graph 65 Eccentricity of a vertex
region 247 k-connectivity of a graph 79 Degree sequence 11 in a digraph 30
total 243 Constant universal graph 325 of a graph 11 in a graph 20
vertex 220 Containing a walk within a walk 17 Degree set 15 Edge 1
P chromatic number 233 Contractible to 166 Diameter addition of 5
Circuit Contraction 166 of a digraph 30 deletion of 4
in a digraph 28 Contracts to 166 of a graph 10 directed 25
in a graph 18 Corona of a graph 305 Different walks 17 Edge chromatic number 236
Class minimal 242 Cover 267 Digraph 25 Edge color class 238
Class one 238 edge 267 Directed cycle 29 Edge cover 267
Class two 238 vertex 267 Directed distance 29 Edge covering number 267
Claw-free graph 87 Covering graph 216 Directed edge 25 Edge independence number 239
Clique number 227 Covering number Directed graph 25 lower 270
lower 364 edge 267 Directed path 29 Edge set 1
Closed neighborhood vertex 267 Disconnected graph 18 Edge-arboricity 70
of a set of vertices 317 Critical block 37 Distance k-Edge-colorable 236
of a vertex 302 Critical (with respect to vertex- in a digraph 29 Edge-coloring 236
Closed walk arboricity) 69 in a graph 19 k-edge-coloring 236
in a digraph 28 k-critical (with respect to vertex- total (of a vertex) 22 k-Edge-connected graph 77
in a graph 17 arboricity) 69 Distinct labelings of a graph 41 Edge-connectivity 75
Closure of a graph 106 Critically k-chromatic 221 Divides 278 Edge-cut 75
(n + I)-Closure of a graph 113 Crossing number 174 Divisor 278 Edge-deleted subgraph 4
Cochromatic number 233 rectilinear 176 Dodecahedron 160 Edge-disjoint paths 36
Color class 220 Cube 160 Dominate Edge-induced subgraph 6
edge 238 n-cube 9 in a digraph 327 Edge-thickness 181
Color-preserving automorphism 46 Cube of a graph 122 in a graph 302 Edge-transitive graph 43
Colorable Cubic graph 6 k-dominate 323 Element of a graph 39
k-colorable 220 Cut-vertex 33 Dominating circuit 126 Elementary contraction 166
k-edge 236 Cycle Dominating set 302 Elementary subdivision 162
n-region 247 n- 18 independent 312 Embeddable
Coloring 220 even 18 k-dominating set 323 on a surface 154
edge 236 in a digraph 28 k-step 324. upper 208
k-coloring 220 in a graph 18 minimal 304 Empty graph 8
total 243 odd 18 minimuin 304 End-block 37
Complement 8 Cycle double cover 102 Dominating set of edges 126 End-vertex 2
of a graph 8 Cycle Double Cover Conjecture 102 Domination maximal 310 Equal graphs 3
Complementary labeling 290 Small 102 k-domination maximal 310 Equal walks 16
Complete bipartite gr<,\ph 8 Cycle edge 35 Domination number Euler Polyhedron FOrinula 158
Complete digraph 28 Cyclic block 37 independent 312 Euler's Formula 155
Complete graph 6 Cyclic decomposition 279 k-domination number 323 Eulerian circuit
Complete multipartite graph 9 k-step 324 in a digraph 97
Complete symmetric digraph 28 H-Decomposable 278 of a digraph 327 in a graph 93
Component Decomposition 278 of a graph 302 Eulerian digraph 97
connected 18 cyclic 279 planetary 327 Eulerian graph 93
of a graph 18 Deficiency 271 upper 321 Eulerian trail
Connected component 18 k-Degenerate 235 Domination perfect graph 315 in a digraph 97
Connected digraph 29 maximal 235 Double cover in a graph 93
strongly 29 Degree cycle 102 Even cycle 18
weakly 29 of a regular digraph 28 Double Cover Conjecture Even graph 95
Connected graph 18 of a regular graph 6 Cycle 102 Even vertex 2
k-connected graph 76 of a vertex in a digraph 26 Small Cycle 102 Expected value of a random
(t,k)-connected graph 79 of a vertex in a graph 2 Double star 61 variable 372
Connected vertices 18 Degree dominated graph 68 Double-star snark 239 Exterior region of a plane graph 155
Connection number 79 Degree linear graph 24 Dual of a graph 248 Exterior subgraph 164
418 Index Index 419
Extremal graphs 329 Hamiltonian graph 103 Irredundant set of vertices 317 Matrix (associated with a graph)
Extremal number 329 Hamiltonian path Irredundant vertex 317 adjacency 1
in a digraph 120 Irregular graph 11 incidence 2
Factor 263 in a graph 113 Isolated vertex 2 Matrix-Tree Theorem 64
I-factor 263 Hamiltonian-connected graph 113 Isomorphic digraphs 26 Maximal k-degenerate 235
k-factor 263 Harmonious Isomorphic factorization 272 Maximal planar graph 156
Factorable 272 graph 293 Isomorphic graphs 3 Maximal with respect to a property P
I-factorable 272 labeling 293 Isomorphism 270
H-factorable 272 . Heawood graph 349 of digraphs 26 Maximum degree 2
k-factorable 272 Heawood Map Coloring Conjecture of graphs 3 Maximum genus 207
Factorization 272 257 kth Iterated line graph of a graph 130 Maximum independent set of vertices
I-factorization 272 Heawood Map Coloring Problem 189 270
hamiltonian 275 Heawood Map Coloring Theorem 257 Join of two graphs 9 Maximum matching 259
isomorphic 272 Henning's Conjecture 321 Joining of two vertices McGee graph 349
Favaron's Conjecture 321 Herschel graph 193 by an arc in a digraph 26 Median of a graph 22
First Theorem of Digraph Theory 26 Hoffman-Singleton graph 347 by an edge of a graph 1 Median vertex 22
First Theorem of Graph Theory 3 Hypercube 10 Menger's theorem 80
Forest 57 Kirchhoff Voltage Law Minimal
linear 285 Icosahedron 160 for a region of a voltage graph 217 class 242
Forest-connected 72 Identification of adjacent vertices 166 for an embedding of a voltage graph Minimal block 38
Four Color Conjecture 247 Incidence matrix 2 217 Minimal dominating set 304
Four Color Problem 246 Incident from (in digraphs) 26 Konigsberg Bridge Problem 92 Minimal with respect to a property P
Four Color Theorem 251 Incident to (in digraphs) 26 Kotzig's Conjecture 284 270
F-Free graph 89 Incident vertex and edge 1 Kuratowski's theorem 163 Minimal with respect to edge
Frucht's theorem 48 Indegree 26 chromatic number 240
Independence number 86 Labelings of a graph 41 Minimally k-chromatic 221
Galaxy 310 edge 239 Length of a walk Minimum degree 2
Generated by elements (group) 45 lower 270 in a digraph 29 Minimum dominating set 302
Generator(s) of a group 45 lower edge 270 in a graph 16 Minimum pairing 99
Genus vertex 86 Lift of a voltage graph 216 Moore graph 346
maximum 207 Independent dominating set 312 Line graph 126 Multidigraph 30
of a graph 183 Independent domination number 312 Line graph of a graph 126 Multigraph 30
of a surface 183 Independent set of edges 239 kth iterated 130 Multipartite graph
Girth 231 Independent set of vertices 86 Linear expectation 372 complete 9
Graceful graph 281 maximum 270 Linear forest 285 Multiple 278
Graceful labeling 281 Independent vertices 86 List-chromatic number 233
Gracefulness 281 Index (of an edge) 72 Lobster 293 (n,m) digraph 26
Graham-Sloane Conjecture 296 Indicator random variable 373 Loop 30 (n,m) graph 1
Graph 1 Induced subdigraph 27 Lower clique number 364 Nearregular complete multipartite
Graphical sequence 12 arc- 27 Lower edge independence number 270 graph 332
Greedy coloring 228 Induced subgraph 5 Lower independence number 270 Neighborhood of a set of vertices 261
Grinberg graph 173 edge- 6 Lower Ramsey number 364 closed 317
Grotzsch graph 230 vertex- 5 Neighborhood of a vertex 109
Group of a graph 40 Integrity 88 Major determinants of matrices 64 closed 302
Interior subgraph 164 Markov's inequality 374 k-neighborhood 323
Hadwiger's Conjecture 232 Internal vertex of a path 36 Marriage Problem 262 Nonadjacent vertices in a digraph 26
Hall's Marriage Theorem 26;3 Internally disjoint paths 36 Marriage Theorem Nonattacking queens 302
Hamiltonian.anticycle 153 Intersection graph 227 Hall's 263 Nondeficient 261
Hamiltonian antipath 153 of a family of graphs 227 Matched to 261 Nonisomorphic digraphs 26
Hamiltonian cycle Interval graph 227 Matched under 261 Nonisomorphic graphs 3
in a digraph 119 Irredundance number 318 Matching 259 Nonseparable graph 35
in a graph 103 upper 321 maximum 259 Nontrivial graph 4
Hamiltonian digraph 118 Irredundance perfect graph 320 perfect 259 Nowhere-zero k-flow 254
Hamiltonian factorization 275 Irredundant Ramsey number 363 Matching graph 271 Numbering of a graph 296
420 Index Index 421
Octahedron 160 Planetary domination sequence 327 Rotational Embedding Scheme 196 Thickness 181
Odd component 263 minimal 327 Ruler Problem 288 edge-181
Odd cycle 18 Platonic solid 159 Threshold function for a graph
Odd graph 95 kth Power of a graph 122 Score of a vertex (in a tournament) theoretic property 381
Odd vertex 2 Private neighbor 318 139 Total chromatic number 243
Open walk Product of two graphs 9 Score sequence of a tournament 139 Total coloring 243
in a digraph 28 cartesian 9 Self-complementary graph 8 k-total coloring 243
in a graph 17 Priifer sequence 62 Semicycle 29 Total Coloring Conjecture 243
Openly dominate 322 Pseudodigraph 31 Semipath 29 Total distance of a vertex 22
Orbit Pseudograph 31 antidirected 29 t-Tough graph 86
of a permutation 196 Semiwalk29 Toughness 86
of a region in a voltage graph Queen's graph 304 Separation of two vertices 80 Tournament 28
216 Similar edges 43 transitive 138
of similar vertices. in a graph 42 [r, q] graph 343 Similar vertices 43 Trail
Order Radius Size in a digraph 28
of a digraph 26 of a digraph 30 of a digraph 26 in a graph 17
of a graph 1 of a graph 20 of a graph 1 Transitive tournament 138
Orientation of a graph 28 Ramsey number 351 Small Cycle Double Cover Conjecture Transversal 262
Oriented graph 27 bipartite 366 102 Tree 57
Outdegree 26 generalized 357 Snark 239 Triangle 18
Outer cycle 116 irredundant 363 double-star 239 Triangle inequality involving distance
Outerplanar graph 169 lower 364 Spanning subdigraph 27 19
Overfull graph 239 Random graph 377 Spanning subgraph 5 Triangl~free graph 188
Random red-blue coloring of Kn 371 Spanning tree 59 Triangulated graph 228
Pairing 98 Random variable 372 Spine of a caterpillar 61 Triangulated planar graph 156
minimum 99 indicator 373 Splitting tree 208 Triangulation 156
Panconnected graph 115 Reachable edge 72 Square of a graph 122 Trivial digraph 26
Pancyclic digraph 150 Realizable on a surface 154 Star 9 Trivial graph 4
Pancyclic graph 116 Recognizable double 61 Trivial path 17
vertex-116 parameter (of a graph) 52 Steiner triple system 280 Trivial walk 16
Parallel edges 30 property (of a graph) 52 k-Step dominate 323 Turan's Brick-Factory Problem 176
k-Partite graph 8 Reconstructible graph 50 k-Step dominating set 324 Turan's theorem 333
complete 8 Reconstruction Conjecture 50 k-Step domination number 324 Tutte graph 172
Partite set 8 Reconstruction Problem 50 Step of a vertex 325 Tutte-Coxeter graph 349
k-Partite tournament 151 Rectilinear crossing number 176 Stereographic projection 154 Tutte's conjecture 254, 256
Path Red degree 366 Strong component 140 Tutte's I-factor theorem 263
in a digraph 28 Red-blue coloring (of Kn ) 351 Strong digraph 29 Type I vertex 256
in a graph 17 random 371 Strong Perfect Graph Conjecture 230 Type II vertex 256
trivial 17 Redundant set of vertices 317 Strongly connected digraph 29
Pendant edge 35 Redundant vertex 317 Subcontraction 166 Underlying graph
Perfect elimination graph 228 Region chromatic number 247 Subdigraph 26 of a digraph 28
Perfect graph 228 n-Region colorable 247 Subdivision 162 of a dual 248
Perfect Graph Conjecture 230 Region of a plane graph 154 elementary 162 Unilateral digraph 29
strong 230 Regular digraph 28 Subdivision graph 120 Union of two graphs 9
Perfect matching 259 r- 28 Subeulerian graph 98 Universal dominating sequence 325
Peripheral vertex 20 Regular graph 6 Subgraph 4 Upper domination number 321
Periphery r- 6 Supereulerian graph 101 Upper embeddable 208
of a graph 20 Regular polyhedron 159 Supergraph 4 on a surface 208
Petersen graph 6 Restricted difference basis 288 Symmetric digraph 27 Upper irredundance number 321
Planar graph 154 Rigid circuit graph 228 complete 28
maximal 156 Ringeisen-White Edg~Adding Lemma Symmetric property of distance 19 Value of a universal dominating
triangulated 156 201 System of distinct representatives 262 sequence 325
Plane graph 154 Ringel's Conjecture 285 Vertex 1
Planetary domination number 327 Ringel-Kotzig Conjecture 285 Tetrahedron 159 deletion of 4
422 Index
M-Vertex 259 Vizing's conjecture 311
Vertex chromatic number 220 Voltage graph of index 1 216
Vertex cover 267
Vertex covering number 267 Walk(s) 16
Vertex independence number 86 closed 17
Vertex set 1 contained in a walk 17
Vertex-arboricity 68 different 17
Vertex-connectivity 74 equal 16
Vertex-cut 74 in a digraph 29
Vertex-deleted subdigraph 27 in a graph 16
Vertex-deleted subgraph 4 length of 16
Vertex-induced subgraph 5 open 17
Vertex-pancyclic trivial 16
digraph 150 Weakly connected digraph 29
graph 116 Winner in a tournament 149
Vertex-transitive graph 42