(Kluwer Texts in The Mathematical Sciences 19) O. Melnikov, V. Sarvanov, R. Tyshkevich, V. Yemelichev, I. Zverovich (Auth.) - Exercises in Graph Theory-Springer Netherlands (1998) PDF

Exercises in Graph Theory
Kluwer Texts in the Mathematical Sciences

VOLUME 19
A Graduate-Level Book Series
The titles published in this series are listed at the end of this volume.
Exercises
in Graph Theory
by
o. Melnikov
Department of Mathematics,
Belarus State University,
Minsk, Belarus
v. Sarvanov
Institute ofMathematics,
Belarus Academy of Sciences,
Minsk, Belarus
R. Tysbkevich
V. Yemelichev
and
I. Zverovich
Department of Mathematics,
Belarus State University,
Minsk, Belarus
SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-90-481-4979-7 ISBN 978-94-017-1514-0 (eBook)

DOI 10.l007/978-94-017-1514-0
Printed on acid-free paper
AII Rights Reserved

@1998 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 1998
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner
Contents
Introduction . . . . . . . 1
1 ABC of Graph Theory 3

1.1 Graphs: Basic Notions . . . . . . . . . 3
1.2 Walks, Paths, Components . . . . . . 13
1.3 Subgraphs and Hereditary Properties
of Graphs. Reconstructibility . . 20
1.4 Operations on Graphs . . . . . . 27
1.5 Matrices Associated with Graphs 31
1.6 Automorphism Group of Graph 35
2 Trees 41
2.1 Trees: Basic Notions . . . . . 41
2.2 Skeletons and Spanning Trees 48
3 Independence and Coverings 55

3.1 Independent Vertex Sets and Cliques 55
3.2 Coverings . . . . 62
3.3 Dominating Sets . . . . . . . . 64
3.4 Matchings............ 66
3.5 Matchings in Bipartite Graphs 68
4 Connectivity 71
4.1 Biconnected Graphs and Biconnected Components 71
4.2 k-connectivity .. 75
4.3 Cycles and Cuts 77
5 Matroids 81
5.1 Independence Systems 81
5.2 Matroids.. . . . 83
5.3 Binary Matroids 90
6 Planarity 93
6.1 Embeddings of Graphs. Euler Formula. 93
6.2 Plane Triangulation 98
6.3 Planarity Criteria . . . . . . . 99
v
VI CONTENTS
604 Duality and Planarity .. 101

6.5 Measures of Displanarity . 107
7 Graph Traversals 111

7.1 Eulerian Graphs 111
7.2 Hamiltonian Graphs 112
8 Degree Sequences 117

8.1 Graphical Sequences 117
8.2 P-graphical Sequences . . . 126
8.3 Split and Threshold Graphs 129
804 Degree Sets and Arity Partitions 132
9 Graph Colorings 135

9.1 Vertex Coloring. 135
9.2 Chromatic Polynomial 141
9.3 Edge Coloring . . . . . 142
904 Colorings of Planar Graphs 144
9.5 Perfect Graphs . . . . . . . 147
10 Directed Graphs 15-1

10.1 Directed Graphs: Basic Notions. 151
10.2 Reachability and Components .. 154
10.3 Matrices Associated with Digraph 160
lOA Tours and Paths 163
10.5 Tournaments .. 166
10.6 Base and Kernel 167
11 Hypergraphs 173
11.1 Hypergraphs: Basic Notions. 173
11.2 Hypergraph Realizations . . . 179
Answers to Chapter 1: ABC of Graph Theory 183

1.1 Graphs: Basic Notions . . . . . . . . . 183
1.2 Walks, Paths, Components . . . . . . 187
1.3 Sub graphs and Hereditary Properties
of Graphs. Reconstructibility . . 192
104 Operations on Graphs . . . . . . 196
1.5 Matrices Associated with Graphs 197
1.6 Automorphism Group of Graph . 201
Answers to Chapter 2: Trees 207

2.1 Trees: Basic Notions . . . 207
2.2 Skeletons and Spanning Trees 218
CONTENTS Vll
Answers to Chapter 3: Independence and Coverings 223

3.1 Independent Vertex Sets and Cliques. 223
3.2 Coverings . . . . 231
3.3 Dominating Sets . . . . . . . . 233
3.4 Matchings . . . . . . . . . . . . 234
3.5 Matchings in Bipartite Graphs 237
Answers to Chapter 4: Connectivity 241

4.1 Biconnected Graphs and Biconnected Components 241
4.2 k-connectivity .. 245
4.3 Cycles and Cuts 248
Answers to Chapter 5: Matroids 251

5.1 Independence Systems 251
5.2 Matroids . . . . . 252
5.3 Binary Matroids 255
Answers to Chapter 6: Planarity 257

6.1 Embeddings of Graphs. Euler Formula. 257
6.2 Plane Triangulation . 262
6.3 Planarity Criteria . . . . . 263
6.4 Duality and Planarity .. 267
6.5 Measures of Displanarity . 270
Answers to Chapter 7: Graph Traversals 275

7.1 Eulerian Graphs .. 275
7.2 Hamiltonian Graphs . . . . . . . . . . . 276
Answers to Chapter 8: Degree Sequences 281

8.1 Graphical Sequences . . . . 281
8.2 P-graphical Sequences . . . . . . 286
8.3 Split and Threshold Graphs .. . 289
8.4 Degree Sets and Arity Partitions 293
Answers to Chapter 9: Graph Colorings 297

9.1 Vertex Coloring. 297
9.2 Chromatic Polynomial .. . 304
9.3 Edge Coloring . . . . . . . . 305
9.4 Colorings of Planar Graphs 307
9.5 Perfect Graphs . . . . . . . 309
Answers to Chapter 10: Directed Graphs 313

10.1 Directed Graphs: Basic Notions .. 313
10.2 Reachability and Components . . . 314
10.3 Matrices Associated with Digraph 318
10.4 Tours and Paths 320
10.5 Tournaments . . . . . . . . . . . . 322
viii CONTENTS
10.6 Base and Kernel . 326
Answers to Chapter 11: Hypergraphs 333

11.1 Hypergraphs: Basic Notions. 333
11.2 Hypergraph Realizations . 336
Bibliography 341
Index . . . 342
Notations .. 351
Introduction
This book supplements the textbook of the authors" Lectures on Graph The-
ory" [6] by more than thousand exercises of varying complexity. The books match
each other in their contents, notations, and terminology. The authors hope that
both students and lecturers will find this book helpful for mastering and verifying
the understanding of the peculiarities of graphs.
The exercises are grouped into eleven chapters and numerous sections accord-
ing to the topics of graph theory: paths, cycles, components, subgraphs, re-
constructibility, operations on graphs, graphs and matrices, trees, independence,
matchings, coverings, connectivity, matroids, planarity, Eulerian and Hamiltonian
graphs, degree sequences, colorings, digraphs, hypergraphs.
Each section starts with main definitions and brief theoretical discussions.
They constitute a minimal background, just a reminder, for solving the exercises.
Proofs of the presented facts and a more extended exposition may be found in
the mentioned textbook of the authors, as well as in many other books in graph
theory.
Most exercises are supplied with answers and hints. In many cases complete
solutions are given.
At the end of the book you may find the index of terms and the glossary of
notations.
The "Bibliography" list refers only to the books used by the authors during the
preparation of the exercisebook. Clearly, it mentions only a fraction of available
books in graph theory. The invention of the authors was also driven by numerous
journal articles, which are impossible to list here.
The authors are greatly indebted to Professors Eberhard Girlich, Michael
Hazewinkel, Uno Kaljulaid, and Alexander Mikhalev, who have stimulated and
promoted publishing this book in English.
The authors wish to give their credit to Dr. Nikolai Korneenko for improving
the text during the translation from Russian.
O. Melnikov et al., Exercises in Graph Theory

© Springer Science+Business Media Dordrecht 1998
Chapter 1
ABC of Graph Theory
1.1 Graphs: Basic Notions

Let V(2) be the set of all two-elements subsets of a nonempty set V. An ordered
pair (V, E), where E ~ V(2), is called a graph. The elements of V are called the
vertices and the elements of E are called the edges. The sets of vertices and edges
of a graph G will be denoted by VG and EG respectively.
A graph is called finite if the set of its vertices is finite. This book deals with
finite graphs only; therefore in the sequel we shall omit the qualifier "finite".
The number JVGI of vertices of a graph G is called its order and is denoted by
IGI. A graph of order n is called an n-vertex graph.
--- -------
P2 P3
• • •
P4
• • • • • •
P5
L C3
D 0
C4 C5
a) b)
•
Kl -- L I2?J
K2
c)
K3 K4
Figure 1.1.1: Examples of graphs

* d)
Graphs are conveniently depicted by drawings that consist of points and con-
tinuous lines connecting some pairs of the points. The points correspond to the
vertices of a graph and the connecting lines correspond to the edges. Fig. 1.1.1
shows the following graphs: a) the simple paths P n , n = 2, ... ,5, b) the simple
cycles Cn, n =
3, ... ,5, c) the complete graphs K n , n =
1, ... ,4, d) the Pe-
3
4 1. ABC of Graph Theory
Tetrahedron
Cube
Octahedron
Dodecahedron
Icosahedron
Figure 1.1.2: Platoni c solids and their graphs

1.1. Graphs: Basic Notions 5
tersen graph. Fig. 1.1.2 shows five Platonic solids (tetrahedron, cube, octahedron,
dodecahedron, icosahedron) and the graphs of these solids, or Platonic graphs.
Two vertices u, v of a graph are said to be adjacent if the set {u, v} is an edge
of the graph. An edge {u,v} is often denoted by uv (or, equivalently, vu.) The
vertices u and v are called the endpoints, or ends of the edge uv. It is also said
that the edge uv connects the vertices u and v.
Two edges are called adjacent if they share an end.
A vertex v and an edge e are called incident if v is an end of e.
The set of vertices of a graph incident to a vertex v is called the neighborhood
of v and denoted by N(v). Similarly, for a subset X ~ VG, the neighborhood of X
is defined as follows:
N(X) = (U N(x))\X.
xEX
Let G and H be graphs and let cp : VG -+ V H be a bijection. If the images
cp(u), cp(v) of any two vertices u, v of G are adjacent in H if and only if u, v are
adjacent in G, then cp is called isomorphism of the graph G onto the graph H. If
such bijection exists, we say that G and H are isomorphic and write G ~ H (and
hence, H ~ G).
A graph is called complete if every two its vertices are adjacent. A graph with
no edges is called empty. The complete and the empty graphs of order n are
denoted by Kn (Fig. 1.1.1c) and On respectively.
A single-vertex graph is called trivial.
The number of edges incident to a vertex v is called the degree of v and is
denoted by deg v. The minimal and maximal degrees of the vertices of a graph G
are denoted by 6( G) and Do. ( G) respectively. The list of vertex degrees of a graph
is called its degree sequence. The order of elements in this sequence is indifferent.
Proposition 1.1.1 (Handshake Lemma) The sum of degrees of the vertices of

a graph is twice the number of the edges.
A vertex of degree 0 is called isolated; a vertex of degree 1 is called pendant.

The edge incident to a pendant vertex is also called pendant. A vertex incident
to all other vertices is called dominating.
Figure 1.1.3: Bipartite graphs
A graph is called bipartite if there is a partition if its vertex set into two subsets
sets (called parts) such that the ends of each edge are in different parts, see Fig.
1.1.3. Such edge partition is called bipartition. A bipartite graph will be denoted
by G = (A, B; E), where A, B are the parts of G and E is the edge set. (The
trivial graph is by definition bipartite, one of its parts being the empty set.) If a
bipartite graph G = (A, B; E) with IAI = p ~ 1, IBI = q ~ 1 is such that every

vertex from the part A is adjacent to all vertices of the part B then G is called
the complete bipartite graph and is denoted by Kp,q. In addition, the graph Kl,q
is called the star graph, or star.
Figure 1.1.4: Graph and its line graph
A graph is called regular if all its vertices are of the same degree. For example,
the complete graphs K n , Fig. 1.1.1c, the Petersen graph, Fig. 1.1.1d, the Platonic
graphs, Fig. 1.1.2,are regular. The degree deg G 0/ a regular graph G is the degree
of its vertices. Regular graphs of degree r are called r-regular. Graphs of degree
three are called cubic graphs.
For a graph G, the complementary graph G is defined as follows:
1. VG= VG;
2. Two vertices are adjacent in G if and only if they are not adjacent in G.
A graph G is called sel/complementary if G ~ G. For example, the path graph

P4 and the cycle graph C5 are selfcomplementary.
Figure 1.1.5: Line graph of Petersen graph
For a graph G the line graph L( G) is defined as follows:
1. The vertices of L(G) bijectively correspond to the edges of G.
2. Two vertices of L( G) are adjacent if and only if the corresponding edges of

G are adjacent.
Fig. 1.1.4 shows graphs G and L( G) overimposed: the vertices of G are denoted
by dots and the edges of G by thin lines; the vertices of L( G) are denoted by circles
and the edges of L( G) by bold lines. Fig. 1.1.5 shows the line graph of the Petersen
graph.
A multigraph is a pair (V, E), where V is a nonempty set (vertices) and E is
a collection of subsets of the set V(2) (edges.) We use the term" collection" to
mean that E may contain several copies of elements of V(2); in such cases equal
elements of E are considered to be different edges and are called multiple edges.
A further generalization is to allow loops, i.e., edges connecting a vertex with
itself. A pseudograph is a pair (V, E), where E is a collection of unordered pairs
of vertices (called edges), not necessarily different. Fig. 1.1.6 shows a multigraph
and a pseudograph.
Figure 1.1.6: Multigraph and pseudograph
For graphs of all these types it is natural to define isomorphism to be a bi-

jection between the sets of vertices that preserves the adjacency relation and the
multiplicities of edges and loops.
The Handshake Lemma holds also for pseudographs, provided that each loop
contributes the count two into the degree of the corresponding vertex.
Graphs in the sense of our initial definition will be called simple, or ordinary,
in order to distinguish them from the above generalizations.
Figure 1.1.7: Labelled graph
A graph of order n is called labelled if its vertices are provided with pairwise
different labels, e.g., the indices 1,2, ... , n. We may identify each vertex of a
graph with its index (and hence we identify the vertex set with the set of indices
{I, 2, ... , n}). Then we may define the equality oflabelled graphs of the same order:
G = H if and only if EG = EH. Fig. 1.1.7 shows a labelled (5,6)-graph G with
EG = {12, 14,23,24,34, 45}.
* * *
1.1.1. Draw:
a) a simple graph;
b) a multigraph that is not a simple graph;
c) a pseudograph that is not a multigraph.
1.1.2. Find the neighborhoods N(u) and N(v) of the vertices u and v for each
graph depicted at Fig. 1.1.8.
u
u v
* Gs
u
Figure 1.1.8: To Exercise 1.1.2
1.1.3. Describe a graph satisfying one of the following conditions:

a) every its vertex is isolated;
b) every its vertex is dominating;
c) the neighborhoods of any two vertices are equal.
1.1.4. Describe the structure of a graph G that satisfies the following condition:
N ( u) U N ( v) = V G for every pair of different vertices u, v.
1.1.5. How many edges has Kn?
1.1.6. a) Does there exist a (4, 7)-graph?
b) Prove that if there exists an (n, m )-graph then
o~ m ~ C~. (1.1.1)
c) Prove that for any two positive integers n, m satisfying the inequalities
(1.1.1) there exists an (n, m)-graph.
1.1. 7. Does there exist a nontrivial graph that has both isolated and dominating
vertices?
1.1.8. Find the number of edges in a graph with the degree sequence
a) (3,3,3,3,3,3);
b) (3,3,2,2,2,2);
c) (3,3,3,2,2);
Can you draw such graphs?
1.1.9. Prove that every nontrivial graph has two vertices of equal degrees.
1.1.10. Does there exist a bipartite graph G such that 6(G) + ~(G) > IGI?
1.1.11. a) Prove that for a nonempty regular bipartite graph the numbers of
vertices in both parts are equal.
b) In a group of people, each girl knows exactly k boys (k > 0) and each
boy knows exactly k girls. Is it true that there is as many boys as girls
in the group?
c) Prove that if G is a bipartite r-regular graph then r ~ IGI/2.
1.1.12. a) Prove that an n-vertex graph with more than n 2 /4 edges is not bi-
partite.
b) Find all n-vertex bipartite graphs with n 2 /4 edges.
1.1.13. Characterize the bipartite graphs with the following property: for every
pair nonadjacent vertices there exists a vertex adjacent to both of them.
1.1.14. Draw the complementary graph for each of the following graphs: On, Kn,
Kn,m, 2K2 (Fig. 1.1.9), C 4 , C s , P4 . Which of the above graphs are selfcom-
plementary?
I I
1.1.15. Prove that if graphs G and H are isomorphic then the complementary
graphs G and H are isomorphic.
1.1.16. Give an example (if possible) of

a) a bipartite r-regular graph of order n;
b) a cubic graph of order 9;
c) an (n, m)-graph with m = C~_l.

1.1.17. Find the number of pairwise nonisomorphic (n - 2)-regular graphs of order
n.
1.1.18. Find the number of pairwise distinct labelled
a) graphs of order n;
b) (n, m)-graphs.
1.1.19. Prove that the isomorphism relation for graphs is an equivalence relation,
i.e.,
a) every graph is isomorphic to itself;
b) if G ~ H then H ~ G;
c) if G ~ Hand H ~ I then G ~ I.
1.1.20. Prove that isomorphic graphs have
a) equal orders;
b) equal numbers of edges;
c) coinciding degree sequences (up to the orders of elements in them).
Are the conditions a)-c) sufficient for graphs to be isomorphic?
1.1.21. Prove that each isomorphism of graphs G and H naturally defines an
isomorphism of line graphs L(G) and L(H).
Figure 1.1.10: To Exercise 1.1.22a
1.1.22. a) Prove that the two graphs from Fig. 1.1.10 are isomorphic and those
from Fig. 1.1.11 are nonisomorphic;
b) Which pairs of graphs at Fig. 1.1.8 are isomorphic?
1.1.23. Prove that if a graph G is bipartite and G ~ H then H is also bipartite.
1.1.24. Draw all pairwise nonisomorphic graphs of orders 3,4,5.
1.1.25. Draw all pairwise nonisomorphic cubic graphs of orders 4,6,8.
1.1.26. Circulant graph: Let n > 1 be an integer and let Zn be an additive group
of modulo n residuals. For a set U C Zn that satisfies the following two
conditions:
1.1. Gra.phs: Basic Notions 11
(1) 0 ¢ U,
(2) x E U implies -x E U
= =
we may define the graph G Gn,u by setting V G Zn and defining a pair
of vertices x, y to be adjacent if and only if x - y E U. Such graph is called
a circulant graph.
a) Prove that
1. the above conditions do specify a graph;
2. a circulant graph Gn,u is a regular graph of degree lUI;
3. a graph complementary to a circulant graph is also circulant;
4. for any integers n, r, 0 ~ r ~ n - 1, such that at least one of them
is even there exists an r-regular circulant graph of order n.
b) Draw a cubic circulant graph of order 6. Determine the number of pair-
wise nonisomorphic cubic circulant graphs of order 6.
c) Do there exist regular graphs that are not circulant?
d) Let Gn,u;, i = 1,2, be two circulant graphs such that
for some positive integer k coprime to n. Prove that the mapping
is an isomorphism of graphs Gn,u, and G n ,u2'
1.1.27. n-Permuiaiion graph: A graph of order n! whose vertices bijectively corre-

spond to the permutations ofthe numbers 1,2, ... , n, and whose two vertices
are adjacent if and only if the corresponding permutations are transformed
into e~th other by a single transposition is called the n-permuiaiion graph.
a) Draw the 3-permutation graph;

b) What is the degree of the n-permutation graph? How many edges does
it have? Is it regular?
c) Prove that the n-permutation graph is bipartite.
1.1.28. Common divisor graph: Let V be a nonempty finite subset of the set of
positive integers. A graph G with the vertex set V is called a common divisor
graph if two vertices u, v E V are adjacent in G if and only if the numbers
u an v are not coprime. Prove that every graph is isomorphic to a common
divisor graph.
1.1.29. Propose the definitions of isomorphism for multigraphs and pseudographs

similar to that for ordinary graphs. Give an example of non isomorphic multi-
graphs with the same numbers of vertices and with the same numbers of
edges of the same multiplicity.
1.1.30. a) How many edges has a graph complementary to an (n, m)-graph?
How many edges has a selfcomplementary graph of order n?
b) Prove that if n is the order of a selfcomplementary graph then
either n == 0 (mod 4) or n == 1 (mod 4). (1.1.2)
c) Let G be a graph. Let us construct two new graphs HG and FG as

follows. Take G I ~ G, G 2 ~ G, G3 ~ G, G 4 ~ G with disjoint vertex
sets. HG is produced from G I , ... , G4 by connecting all vertices from
VG;, with all vertices from VGi+I, i = 1,2,3, by edges. FG is produced
from H G by adding a new vertex and connecting it with all vertices of
VG I and VG 4 . Prove that HG and FG are selfcomplementary graphs
for all G.
d) Prove that for any positive integer n satisfying the condition (1.1.2)
there exists a selfcomplementary graph of order n.
e) Prove that the problem of recognition of isomorphism for arbitrary
graphs is reducible to that for selfcomplementary graphs, i.e., for any
graphs G I , G 2 one may construct selfcomplementary graphs SG 1 and
SG 2 such that G I ~ G 2 if and only if SG 1 ~ SG 2 •
1.1.31. Find all selfcomplementary graphs of orders: (a) 4; (b) 5; (c) 6.

1.1.32. Prove that if (d l , d2 , ... , dn ) is the degree sequence of a selfcomplementary
graph and dI 2:: d2 2:: ... 2:: dn , then
d; = n - 1 - dn +l - i, i = 1,2, ... , In/2J.
1.1.33. Prove that if there exists an r-regular selfcomplementary graph of order n

then n == 1 (mod 4) and r = (n - 1)/2.
1.1.34. Is it possible for a selfcomplementary graph of order 100 to have exactly
one vertex of order 50?
1.1.35. Find the numbers of vertices and edges for the line graph L( G) of a graph
G with the degree sequence
1.2. Walks, Paths, Components 13
1.1.36. Prove that the line graph L( G) cannot be isomorphic to K I,q, q ~ 3, for
any graph G.
1.1.37. Give examples of
a) a graph with L( G) ~ G;
b) a Platonic graph isomorphic to the line graph of another Platonic graph.
1.1.38. Is it true that the line graph
a) of a regular graph is regular;

b) of a bipartite graph is bipartite?
1.1.39. Prove that the graph L(Kn) is a (2n-4)-regular graph of order C~. State
and prove the similar statement for the graph Kp,q.
1.1.40. Prove that the graph L(K5) is isomorphic to the Petersen graph.
1.2 Walks, Paths, Components

An alternating sequence
(1.2.1)
of vertices and edges of a graph such that ei = ViVi+I, i = 1,2, ... , k, is called a
walk connecting the vertices VI and Vk+1 (or simply an (VI, Vk+I)-walk). In (1.2.1)
it is possible that Vk+1 = VI, i.e., a walk may connect a vertex with itself. The
vertices VI and Vk+1 are called the terminal vertices (or endvertices) of the walk.
A walk is called a path if all its edges are different. A path is called a simple path
if all its vertices are different, with the possible exception of Vk+1 = VI. A walk is
called cyclic if its endvertices coincide and non cyclic otherwise. A cyclic path is
called a cycle, a cyclic simple path is called a simple cycle.
The number k of edges in a walk is called its length. A simple cycle of length
k is called a k-cycle and is denoted by Ck. The 3-cycle is often called triangle.
A simple non cyclic path with k vertices (i.e., of length k -1) is denoted by Pk.
A graph without cycles is called acyclic.
In the case of ordinary graphs the walk (1.2.1) may be specified in a simpler
way by writing down its vertices (or edges) in the order of their occurrence III
(1.2.1):
Any vertex of a cycle may be taken as a "starting" vertex. Moreover, the

sequences
specify the same cycle. For example, the sequences (1,2,3,4,1), (3,4,1,2,3), and
(1,4,3,2,1) specify the same cycle. The cycles (1,2,3,4,1) and (1,2,4,3,1) are
different: while the sets of their vertices coincide, the orders of their traversal are
different.
The length of the shortest cycle in a graph is called the girth of the graph. The
girth of an acyclic graph is defined to be infinite.
There exists a biparticity criterion for graphs in terms of lengths of cycles:
Theorem 1.2.1 (Konig's Theorem) A graph is bipartite if and only if it has

no cycles of odd length.
A graph H is called a subgraph of a graph G if V H ~ VG and EH ~ EG.

A graph in which any two vertices are connected by a walk is called connected.
A maximal with respect to inclusion connected subgraph of a graph G is called a
connected component, or simply component, of G. The vertex set of a connected
component is called a connected domain of a graph.
The length of the shortest (u, v)-walk in a connected graph G is called the
distance between the vertices u and v and is denoted by dG (u, v), or even simply
d(u, v).
The value
e(u) = max d(u, v)
vEVG
is called the excentricity of the vertex u. The maximum among the excentricities
of the vertices of a graph G is called the diameter of the graph and is denoted by
d( G). If e( v) = d( G) then the vertex v is called peripherical. A simple path of
length d( G) with distance d( G) between its endpoints is called a diametral path.
The minimum among the excentricities of the vertices of a graph G is called
the radius of the graph and is denoted by r( G). If e( v) = r( G) then the vertex v is
called central. The set ZG of all central vertices of a graph G is called the center
of the graph G. The subgraph G(ZG) with the vertex set ZG such that two its
vertices are adjacent if and only if they are adjacent in G will be called the central
subgraph, and sometimes simply the center.
Let us describe a simple procedure of analysis of a graph called breadth-first
search. It may be used, e.g., to determine whether a graph is connected or bipar-
tite, to find the distance between two vertices, vertex excentricities, diameter and
radius of a graph, to determine the connected domains of a graph.
Breadth-First Search. Given a graph G, we start from an arbitrary vertex
v and assign label 0 to it. Each vertex from the neighborhood of v obtains label
1. Each unlabelled vertex from the neighborhoods of the vertices with label 1
obtains label 2. Labels 3 are assigned in a similar way. This process continues
until possible. If G is connected then eventually the breadth-first search will label
all its vertices, see Fig. 1.2.1.
* * *
1.2.1. Find all simple paths and simple cycles in the graph shown at Fig. 1.2.2.
1.2.2. How many labelled simple cycles are there in the graphs a) Kn and b) Kp,q?
1.2.3. Find cycles of lengths 5,6,8,9 in the Petersen graph.
1.2.4. Find the girths of a) Kn, b) Kp,q, c) Petersen graph.
Figure 1.2.1: Breadth-first search labelling of graph
4 3
Figure 1.2.2: Example graph
1.2.5. a) Prove that every noncyclic (u, v)-walk contains a simple (u, v)-path. Is
this statement true for cyclic walks?
b) Is it true that every (u, v)-walk passing through a vertex w distinct
from u and v contains a simple (u, v)-path passing through w?
c) Prove that every cyclic walk of odd length contains a simple cycle. Is
this statement true for cyclic walks of even length?
d) Prove that every cycle contains a simple cycle.
1.2.6. Prove that in any graph G the distance d(u, v), u, v E VG, satisfies the
following axioms of metric:
1) d(u, v) ~ 0;
2) d(u, v) = 0 if and only if u = v;
3) d(u,v) = d(v,u);
4) d( u, v) + d( v, w) ~ d( u, w) (triangle inequality)~
1.2.7. Prove that a graph G is connected if and only if
a) given a fixed vertex u, there exist (u, v)-paths in G to all other vertices
v E VG; .
b) for every partition of the vertex set V G = Vi UV2, Vi nV2 = 0, there
exists an edge ab E EG such that a E Vi, bE V2 •
1.2.8. Prove that each vertex of a graph G lies in exactly one of its components.
Prove the similar statement for edges.
1.2.9. a) Prove that in every connected graph any two simple paths of maximal
length share a vertex.
b) With the help of the graph from Fig. 1.2.3 verify that the following
statement is false: in every connected graph all simple paths of maximal
length share a vertex.
1.2.10. Prove that if a graph has exactly two vertices of odd degrees then they are
connected by a path.
1.2.11. Prove that if the minimal degree 8(G) of the vertices of a graph G is greater
than 1 then G has a cycle.
1.2.12. Define the union G U H of a pair of graphs as follows:
V(GUH) = VGUVH, E(GUH) = EGuEH.
Prove that:
a) if P, Q are different simple (u, v)-paths then the graph P U Q contains
a simple cycle;
b) if C, D are different simple cycles with a common edge e then the graph
(CUD)-e obtained by the deletion ofthe edge e from the graph CuD
contains a simple cycle.
1.2.13. a) Prove that a graph is bipartite if and only if it has no simple cycles
of odd length.
b) Prove that the bipartition of a connected bipartite graph is unique. Is
this true for disconnected bipartite graphs?
1.2.14. Find all bipartite graphs whose complementary graphs are also bipartite.
Indicate the selfcomplementary ones among them.
1.2.15. Find all bipartite graphs of diameter 2.
1.2.16. Prove that for any integer number r > 1 there exists an r-regular graph of
order 2r and diameter 2.
1.2.17. Draw all pairwise nonisomorphic connected graphs with exactly four edges.
1.2.18. Describe the structure of a connected graph G that satisfies one of the
following conditions:
a) the length of any maximal with respect to inclusion simple path is 2;

b) for any vertex v E V G all vertices from its neighborhood are pairwise
adjacent;
c) every vertex is pendant;
d) every edge is pendant;
e) all edges are pairwise adjacent;
f) the number of edges is equal to the number of pendant vertices;
g) 6(G) = 1, ~(G) = 2;
h) G ~ LCG).
1.2.19. Describe the structure of a connected r-regular graph for a) r = 0, b)
=
r 1, c) r 2. =
1.2.20. Prove that the deletion of any edge that belongs to a cycle of a connected
graph produces a connected graph. What will occur if we delete an edge
which does not belong to any cycle?
1.2.21. What is the minimal possible number of edges in a connected graph of

order n?
1.2.22. Prove that for any two edges el, e2 of a connected graph there exists a
simple path such that el is its first edge and e2 is its last edge.
1.2.23. Prove the following statements.
a) If an (n, m)-graph has k components then
(n-k)(n-k+l)
n- k :::;m:::; 2 . (1.2.2)
Describe the cases for which these bounds are attained.

b) If in addition the graph is bipartite then
(n-k)(n-k+2) if n - k is even
{
n - k :::; m :::; (n _ k +21)2 ' (1.2.3)
if n - k is odd.
2 '
c) Prove that for any positive integers satisfying the inequalities (1.2.2)
there exists an (n, m)-graph with k components, and if (1.2.3) holds
then the required graph exists among the bipartite ones.
1.2.24. a) Prove that a graph G of order n is connected if 6(G) ~ (n - 1)/2.
Will the statement remain valid if we replace (n - 1)/2 by l( n - 1)/2J?
b) Prove that a graph G of order n > 2 is connected if the number of edges
is greater than (n -1)/2. Can we replace here "greater" by "not less"?
c) Prove that an (n, m )-graph is connected if it has no cycles of odd length
and m > «n - 1)/2)2. Will this statement remain true if we replace
">" by "~"?
1.2.25. a) Prove that the n-permutation graph (see Exercise 1.1.27) is connected
for any n.
b) Find the diameter of this graph.
c) Find all central and peripherical vertices of this graph.
1.2.26. Prove that
a) the diameter of a graph is at most twice its radius;
b) a diametral path of a graph contains a central vertex.
1.2.27. For any n ~ 4 give an example of an n-vertex graph other than the com-
plete graph such that
a) its radius and diameter are equal;
b) each vertex is peripherical;
c) each vertex is central.
1.2.28. Prove that
a) for any graph G either G or G is connected;
b) if G is not connected then d( G) ~ 2;
c) if G and G are connected and d(G) ~ 3 then d(G) ~ 3;
d) 1 ~ d(G) ~ 3 for any nontrivial selfcomplementary graph G.
1.2.29. The metrical dimension p,(G) of a connected graph G is the minimal car-
dinality of the subsets A ~ VG with the following property: for any two
vertices u,v E VG there is a vertex a E A such that d(u,a) =F d(v,a). Find
the metrical dimensions of the following graphs: a) K n , b) Kp,q, c) Cn , d) .
Pn .
1.2.30. Prove that p,(G) ~ IGI-d(G) (see Exercise 1.2.29) for any connected graph
G.
1.2.31. Give an example of a cubic graph of order 2n, n ~ 3, with girth at least 4.
1.2.32. Prove that IGI ~ L\(G) + 6(G), if the girth of Gis 4.
1.2.33. Prove that the order of a k-regular graph with girth 4 is at least 2k.
1.2.34. Give an example of a graph G whose center is not the set of all vertices
and consists of a) one vertex, b) two vertices, c) three vertices.
1.2.35. Prove that for any graph G there exists a graph H such that G:=:: Z(H),
where Z(H) is the central graph of H.
1.2.36. The Konig representation of a graph G is the bipartite graph K( G) such
that the vertices of one part bijectively correspond to the vertices of G, the
vertices of the other part bijectively correspond to the edges of G, and two
vertices of K (G) are adjacent if and only if the corresponding vertex and
edge of G are incident to each other. Prove the following statements.
a) K(G) may be obtained from G by subdivisions of all its edges, i.e., for
each edge e = uv E EG we add a new vertex We and the edge e is
replaced by the two edges UWe and WeV.
b) Two graphs are isomorphic if and only if its Konig representations are
isomorphic.
1.2.37. Determine whether the graphs from Fig. 1.2.4 are bipartite, using breadth-
first search. Partition the vertex sets of the bipartite graphs into parts.
f g
Figure 1.2.4: To Exercises 1.2.37,1.2.39
1.2.38. Using breadth-first search, find the diameter, radius, central and periph-
erical vertices of the following graphs: a) Pn ; b) en; c)Kn; d) K1,n, n 2: 2;
e) Km,n, n 2': 2, m 2': 2.
20 1. ABC of Gra.ph Theory
1.2.39. Using breadth-first search, find the radius, diameter, central and periph-
erical vertices of the graphs from Fig. 1.2.4.
1.2.40. Let f : VG -+ V H be an isomorphism of graphs G and H. Prove the
following statements.
a) If VG = VI U V2 U ... U Vk is the partition of VG into the connected

domains of G and
f(Vi} = {f(v): v E Vi} = Wi, i = 1,2, .. . ,k,

then V H = WI UW 2 U ... U Wk is the partition of V H into the connected
domains of H.
b) For any two vertices u, v of G,
d(u, v) = d(f(u), f(v)); e(u) = e(f(u)).

c) d(G) = d(H), r(G) = r(H).
d) If ZG denotes the center of G then f(ZG) = ZH.
1.2.41. Prove that the number of components of a circulant graph Gn,u (see Exer-
cise 1.1.26) is equal to the greatest common divisor GCD(UI' U2, ... , Uk, n),
where {Ul' U2,"" Uk} = U. In particular, Gn,u is connected if and only if
GCD(Ul, U2, ... , Uk, n) = 1.

of Graphs. Reconstructibility
A graph H is called a subgraph of a graph G if V H ~ VG and EH ~ EG. For
X ~ VG, we define the induced subgraph G(X) to be the subgraph of G such
that VG(X) = X and u, v E VG(X) are adjacent in G(X) if and only if they
are adjacent in G. It is said that G(X) is induced by X. A subgraph is called a
spanning subgraph if VG = V H. A spanning induced subgraph of a graph Gis
clearly the graph G itself.
In the cases when no confusion is possible the term "subgraph" will be used to
denote the" graph isomorphic to a subgraph" of the graph involved.
A property of graphs will be called graph-theoreticaliffor any pair of isomorphic
graphs both of them either possess the property or not.
Formally, a property of graphs may be treated as a class of graphs. Therefore
we shall use the notation G E P in order to say "G has a property P" .
A graph-theoretical property P is called hereditary if the fact G E P implies
H E P for every induced subgraph H of G. It is said that the set T of graphs char-
acterizes a graph-theoretical property P in terms of forbidden induced subgraphs
(FIS) if G E P if and only if no induced sub graph of G belongs to T.
Theorem 1.3.1 The hereditary properties are exactly the ones which may be char-
acterized in terms of forbidden induced subgraphs.
1.3. Subgraphs and Hereditary Properties of Graphs. Reconstructibility 21
In an investigation of a hereditary property, a natural intention is to find a

minimal under inclusion set of forbidden induced subgraphs (minimal FIS set)
characterizing it.
Theorem 1.3.2 For a given hereditary property P a minimal FIS set is a set of
graphs G satisfying the following conditions:
1. G(j.P;
2. every induced subgraph of G other than G itself has the property P.
Let v be a vertex of G. The graph G v = G - v is o~tained from G by the

deletion of v together with the edges incident to it; it is called a vertex-deleted
subgraph.
The graphs G v are related with the K elly- Ulam reconstruction conjecture,
stated as follows. For a graph G the collection of graphs P( G) = (G v : v E V G)
is called the deck of the graph and the graphs G v are called the cards of the deck.
The equality of decks is defined up to permutations of cards in decks. It is im-
portant to emphasize that there are no correspondences among the vertex sets of
cards, i.e., cards are considered to be unlabelled graphs. Alternatively, one may
consider the labelled deck PI(G) of a labelled graph G.
For example, if G = P3 then P(G) =(K2' K 2, O 2 ). An example of a labelled
deck is shown at Fig. 1.3.1.
2
~
G
Figure 1.3.1: Labelled deck
A graph G is called a reconstruction of a graph H if P(G) = P(H). For

example, P(I{2) = P(02) = (01,0t), therefore K2 and O 2 are reconstructions
of each other. A graph G is called reconstructible if it is isomorphic to every its
reconstruction.
Reconstruction Conjecture (P.Kelly, S.Ulam, 1941). The graphs of order
n > 2 are reconstructible.
The Kelly-Ulam conjecture is often called the vertex reconstruction conjecture.
In addition, the edge reconstruction conjecture due to F.Harary is known. It is
stated in a similar way in terms of edge-deleted subgraphs: the subgraph G e = G-e
is obtained from G by the deletion of the edge e (the ends of e are not deleted, i.e.,
G e is a spanning subgraph). Accordingly, the collection P'(G) = (G e : e E EG)
is called the edge deck of the graph G. A graph G is called an edge reconstruction
of a graph H if P( G) = P(H). A graph G is called edge reconstructible if it is
isomorphic to every its edge reconstruction.
Edge Reconstruction Conjecture (F.Harary, 1964). All graphs with more
than three edges are edge reconstructible.
A nonempty subset of vertices of a graph is called a clique if the subgraph

induced by it is a complete graph. In particular, all single-element vertex subsets
are cliques. A clique is called maximal if it is not contained in a larger clique of the
graph. For example, the graph shown at Fig. 1.3.2 has three maximal cliques.
Figure 1.3.2: Example graph
Let G be a graph. The clique graph Q( G) is defined as follows. The vertices

of Q( G) bijectively correspond to the maximal cliques of G; two vertices of Q( G)
are adjacent if and only if the corresponding cliques have a nonempty intersection.
For example, the clique graph for the graph from Fig. 1.3.2 is P3.
Theorem 1.3.3 For every graph H there exists a graph G such that H is isomor-
phic to an induced subgraph of the clique graph Q( G).
* * *
1.3.1. Prove that
a) a subgraph of a subgraph of a graph G is a subgraph of G;

b) a spanning subgraph of a spanning subgraph of a graph G is a spanning
subgraph of G;
c) an induced subgraph of an induced subgraph of a graph G is an induced
subgraph of G.
1.3.2. Prove that

a) any subgraph of a graph G may be obtained by deletion of some vertices
and/or edges of G;
b) any spanning subgraph of a graph G may be obtained by deletion of
some edges of G;
c) any induced subgraph of a graph G may be obtained by deletion of
some vertices of G.
1.3.3. (a) For a graph from Fig. 1.3.3 determine whether the cycles C5 , C6 , C7 , Cs
are among its
(1) subgraphs;
(2) spanning subgraphs;
(3) induced subgraphs.
1.3. Subgraphs and Hereditary Properties of Graphs. ReconstructibjJity 23
4 3
(b) Is it true that C(X) = P5 for X = {1, 2, 4, 6, 8} for a graph G shown at

Fig. 1.3.3?
1.3.4. Let C n be a graph with the vertex set {Vl,"" vn } such that Vi, Vj are
adjacent if and only if i, j are coprime. Draw the graphs G4 and C 8 . Prove
that if m < n then C m is an induced subgraph of C n .
1.3.5. Let F and H be subgraphs of a labelled (n, m)-graph C. We shall say that
F is equal to H if V F =
V Hand EF =
EH. Determine the numbers of
different
(a) induced subgraphs of G;

(b) spanning subgraphs of C;
(c) subgraphs of C for G = Kn.
1.3.6. Suppose that VC = V H for graphs C, Hand degG v ~ degH v for every
vertex v. Is it true that H is a subgraph of C?
1.3.7. Is it true that o( C) ~ o(H) if H is

(a) a subgraph of C;
(b) a spanning subgraph of C?
1.3.8. Find all graphs of order n ~ 4 for which all induced subgraphs of order 3
are isomorphic to (a) 0 3 ; (b) P3 ; (c) P3 ; Cd) K 3 .
1.3.9. What is the structure of a graph in which every simple path is an induced
subgraph?
1.3.10. Prove that for every graph G there exists a selfcomplementary graph that
has an induced subgraph isomorphic to C.
1.3.11. Is it true that the vertex set of any graph C may be partitioned into parts
V1 and V2 such that C(Vt) and C(V2 ) are regular graphs?
1.3.12. Prove that there are no graphs of order n ~ 5 such that every four-vertex
induced subgraph is the cycle graph C4 .
1.3.13. Prove that every graph G is an induced subgraph of a A(G)-regular graph.

1.3.14. Characterize connected bipartite graphs that have no induced subgraphs
P5 and C4 •
1.3.15. Prove the theorems 1.3.1 and 1.3.2.
1.3.16. Characterize all graphs in which the sets N(v) induce complete graphs for
all vertices v.
1.3.17. Find a graph of order 10 in which the neighborhoods of all vertices induce
the graph P 3 .
1.3.18. Prove that there are no regular graphs in which the neighborhoods of all
vertices induce the star KI,n, n ~ 2.
1.3.19. Find all connected graphs in which the sets N(v) for all vertices v induce
the graph
1.3.20. Let {HI, H 2 , . •• } be the set offorbidden induced subgraphs for some fixed
hereditary property P. Find a condition under which G E P implies G E P.
1.3.21. Determine which of the following graph-theoretical properties are heredi-
tary. For every hereditary property find a minimal FIS set.
(a) to be an empty graph;
(b) to be a complete graph;
(c) to be a connected graph;
(d) to be a disconnected graph;
(e) to be a complete bipartite graph;
(f) every component of a graph in question is a complete graph;
(g) every component of a graph in question is a path graph;
(h) every component of a graph in question is either a star or a trivial
graph;
(i) to have a vertex of even degree;
(j) 6(G) :::; 2;
(k) A(G):::; 2;
(1) to have an isolated vertex;
(m) to be a selfcomplementary graph;
(n) to be a bipartite graph;
( 0) to have no cycles;
(p) to be regular;
1.3. Subgraphs and Hereditary Properties of Graphs. Reconstructibility 25
(q) to have an induced subgraph isomorphic to a given graph H of K 1 .

1.3.22. Let P be a hereditary graph-theoretical property with at least one graph
G E P. Are the following properties hereditary?
(a) to have not the property P;

(b) to have no subgraphs with the property P.
1.3.23. Prove that a graph is complete bipartite or empty if and only if it has no
induced graphs K3 and P3.
1.3.24. Let P be a hereditary property. Define 15 to be the property such that

G E 15 if and only if G E P. Is 15 a hereditary property?
1.3.25. Is it true that for hereditary properties P1, P2
(a) the disjunction
is a hereditary property?
(b) the conjunction
is a hereditary property?
1.3.26. (a) Let K be a class of graphs that contains a graph K 1. Prove that it has
a unique maximal under inclusion hereditary subclass [K] ~ K.
(b) Let K be a class of graphs that contains all disconnected graphs, their
complements, and the graph K 1 . Characterize the subclass [K] defined above,
in terms of forbidden induced subgraphs.
1.3.27. Prove that a labelled graph of order at least 3 is reconstructible from its
labelled deck.
1.3.28. Are labelled graphs of order at least 3 reconstructible from

(a) two,
(b) three
cards of the labelled deck?
1.3.29. Determine whether the following collection of graphs constitute the deck
of some graph:
(a) Kl U P3• Kl U P3, Kl U P3, Kl U P3, P4 ;

(b) Kl U P3• C4 , C4 , K2 U K 2, K2 U K 2.
1.3.30. Prove that all regular graphs of order at least 3 are reconstructible.
1.3.31. Determine the degrees of all vertices of a graph of order at least 3 in terms
of its deck.
1.3.32. Prove that given a deck of a graph of order at least 3 without isolated
vertices, one may determine the numbers of edges and components of the
graph.
1.3.33. Prove the Kelly lemma: if the order of a graph F is less than the order
of a graph G then the number S(F, G) of subgraphs of G isomorphic to F
satisfies the relation:
" S(F,Gv )
S(F, G) = L..J IGI- IFI·
vEVG
1.3.34. Find all graphs whose deck contains only simple cycles.
1.3.35. Is the property "to be reconstructible" hereditary?
1.3.36. Does there exist a graph of order at least 3 whose cards in the deck are
pairwise nonisomorphic?
1.3.37. Find a clique graph for the graph shown at Fig. 1.3.4.
1.3.38. Find the graphs Q(G) and L(G) for (a) G = Pn , (b) G = Cn .
1.3.39. Prove that if a graph G has no isolated vertices and triangles then Q( G) ~
L(G).
1.3.40. Describe the structure of Q(L(K5» (the clique graph of the line graph of
K5).
1.3.41. Prove that
(a) every graph G with minimal degree 8(G) greater than 2 is isomorphic to
an induced subgraph of Q(L(G)).
(b) G ~ Q(L(G» if 8(G) > 1 and G is triangle-free.
1.3.42. If a graph G is isomorphic to the line graph of some graph then G is also
called a line graph. Prove the following statements.
(a) The property "to be a line graph" is hereditary.
(b) The property "to be a line graph of a triangle-free graph" is hereditary.
1.4. Operations on Graphs 27
(c) A graph G is the line graph of a triangle-free graph if and only if G has
no induced subgraphs K 1 ,3 and K4 - e.
1.3.43. Let B r be the set of (0, I)-sequences of length r. The Hamming distance
between the sequences
is the number
r
dH(a, (3) = L la; - (3;1·
;=1
The Hamming graph Hr,k is defined as follows. V Hr,k = B r , and two vertices
a, (3 are adjacent if and only if dH(a, (3) :::; k. Prove that every (n, m)-graph
is isomorphic to an induced subgraph of the Hamming graph Hr,k with r =
n 2 - m, k = 2n - 2.
1.4 Operations on Graphs

We have already defined the operations of vertex deletion and edge deletion in
Sect. 1.3. The deletion of any set X of elements of a graph G is defined as the
successive deletion of all vertices and edges contained in X from G. The resulting
graph is denoted by G - X.
Let two vertices u, v of G be nonadjacent. Then one may construct the graph
G I = G + e, where e = UV, such that VGI = VG and EG I = EG U {e}. This
operation is called the addition of the edge e.
Sometimes the notations G - v, G - e, G + e are used without specifying the
elements v, e. In such cases it is assumed that these elements are arbitrary, and
either the result is independent of the choice, see Fig. 1.4.1, or the resulting graph
is indifferent for us.
Ou(Jv C5 -v C5 - e c5 + e
Figure 1.4.1: Operations on graph
The union of graphs 0, H is the graph 0 U H with the vertex set V 0 U V H

and the edge set EO U EH. If VO n V H = 0 then the union is called disjoint. If
VO and/or V H are not specified then the union is assumed to be disjoint. Fig.
1.4.2 shows the graph K2 U K2 U K4 U K4 U K 4.
For the sake of brevity, one may use the notation nO for the disjoint union of
n graphs isomorphic to 0 (n copies of 0). Thus, the graph from Fig. 1.4.2 may
be denoted as 2K2 U 3K4 .
Figure 1.4.2: Disjoint union of graphs
The intersection G n H = (VG n V H, EG n EH) is defined for any pair of

graphs G, H with common vertices.
The join G + H is obtained from the disjoint union G U H by the addition of
the set of all edges of type uv : U E VG, v E V H.
The merging of vertices u, v of a graph G is the addition of a new vertex w to
the graph G - u - v, together with all edges connecting w with all vertices from
NG(u) UNG(v)\{u,v}.
The contraction of an edge uv is the merging of the adjacent vertices u, v. For
example, the contraction of an edge of C 5 produces C4 .
Consider the unary operation of raising a graph G to the k-th power. The
k-th power G k is defined as follows. VG k = VG; two vertices u, v E VG k are
adjacent if and only if the distance d(u, v) in G is at most k. Clearly, G l = G.
For the following four binary operations of (Cartesian) product G l x G 2 , com-
position GI[G 2 ], conjunction G l A G 2 , and modular product G l 0 G 2 the vertex set
of the resulting graph is the Cartesian product VG l x VG 2 . The adjacencies of
= =
vertices u (Ul, U2) and v (Vl, V2) are defined according to the following rules.
For G l x G 2 :
- either Ul = VI and U2V2 E EG2,
- or U2 = V2 and UlVl E EG l .
For G l A G 2 :
- Ul VI E EG I and U2V2 E EG 2 .
For G l oG 2 :
Ul i= Vl and U2 i= V2 and
- either Ul VI E EG I and U2V2 E EG 2 ,
- or UlVl tI. EG I and U2 V 2 tI. EG 2 .
Using the product operation, an important class of graphs called n-dimensional
cubes, or n-hypercubes Qn may be defined in a recurrent way: Ql = J{2, and
Qn = J{2 X Qn-l, n> 1, see Fig. 1.4.3.
* * *
1.4. Operations on Graphs 29
Figure 1.4.3: Cube graphs Q3 and Q4
1.4.1. Demonstrate that the operation G-X is defined correctly, i.e., the resulting
graph does not depend on the order of deletions of the elements of X.
1.4.2. Prove that the union of a connected graph G and any graph H is connected
if and only if every component of H contains a vertex of G.
1.4.3. Express the operation of join of graphs in terms of the operations of union
and complement.
1.4.4. Is it true that the intersection G n H is an induced subgraph of either G or

H?
1.4.5. Let P be a hereditary property. Is it true that if G E P and H E P then

GnH E P?
1.4.6. Let G' be a graph obtained from a bipartite graph G by the contraction
of an edge e. Find necessary and sufficient conditions under which G' is
bipartite.
1.4.7. Let u, v be two nonadjacent vertices from different parts of a connected

bipartite graph. Prove that the graph obtained by merging u with v is not
bipartite.
1.4.8. Find the diameter of P;.

1.4.9. What is the minimal number k such that the k-th power of a connected
graph is the complete graph?
1.4.10. Is it true that (a) (cm)n = c m+n , (b) (cm)n = c mn ?

1.4.11. How many edges has C + H?
1.4.12. Prove that if a graph C has n :?: 2 vertices and m :?:
(n
:2 1)2 edges then
C = HI + H2 for some graphs HI, H 2 .
1.4.13. Construct the graphs (a) C4 x P3 ; (b) C 4 [P2 ]; (c) C4 /\ P3 ; (d) C 4 <> P3 .
1.4.14. Let C i be (ni' mi)-graphs, i = 1,2. Find the numbers of edges of the graphs
(a) C I x C 2 ; (b) CdC 2 ]; (c) C I /\ C 2 ; (d) G I <> C 2 .
1.4.15. Which of the operations listed in the Exercise 1.4.14 are commutative?
1.4.16. Let db" .dn be the degree sequence of a graph G and let di, ... d~ be the
degree sequence of a graph G'. What are the degrees of the vertices of the
graph G x G'?
1.4.17. Prove that G x H is connected if and only if both G and H are connected.
1.4.18. Is it true that G x H is bipartite if and only if both G and H are bipartite.
1.4.19. Prove that

(a) G 1 /\ G2 is a spanning sub graph of G 1 <> G 2 .
(b) G 1 x G 2 is a spanning subgraph of GdG 2].
1.4.20. Prove that

(a) C 2m +1 /\ C 2m +1 is isomorphic to C 2m +1 X C 2m +1 ;
(b) for connected graphs G 1 • G 2 • the conjunction G 1 /\ G 2 is connected if and
only if one of these graphs has a simple cycle of odd length.
1.4.21. The major modular product GoG' is the graph specified as follows.
(1) V(GoG') = VG x VG';
(2) two vertices (u, u'). (v, v') of GoG' are adjacent if and only if u :f. v,
u' :f. v' and either uv E EG, u'v' E EG' or uv f/:. EG.
a. Construct the major modular product of pairs of graphs {P4, P3} and
{2K 2 , C 3 }.
b. Prove that G <> G' is a sub graph of GoG'.
c. What is the degree of a vertex (u, v) of GoG'?
d. Find graphs G, G' such that GoG' and G'oG are nonisomorphic.
1.4.22. Let us define the operation of extension of a graph G by a graph H denoted

by G +- H. For every vertex v E VG we take a copy Hv of the graph H; all
copies are pairwise disjoint. We set V(G +- H) = UVEVG V Hv. Two vertices
x E Hv, y E Hw are adjacent in G +- H if either v = wand x, yare adjacent
in Hv or v :f. wand v, ware adjacent in G.
Prove the following statements.
a. For every graph G, graphs Hl and H2 are isomorphic if and only if

G +- Hl and G +- H2 are isomorphic.
b. IE(G +- H)I = IEHIIVGI + IVHI2IEGI·
1.4.23. Using the operation of graph extension defined in the Exercise 1.4.22,
(a) construct a selfcomplementary graph of order 5n for every n 2': 1;
(b) prove that for any selfcomplementary graphs G, H the graph G +- His
also selfcomplementary.
1.5. Matrices Associated with Graphs 31
1.4.24. Prove that the n-cube Qn may be defined as follows. The vertices are
taken to be the (0, 1)-vectors of length n. Two vertices are adjacent if and
only if the corresponding vectors differ exactly by one component.
1.4.25. Draw the 5-cube.
1.4.26. Find the numbers of vertices and edges of the n-cube Qn.
1.4.27. Find the length of the longest induced simple cycle in Qn for (a) n = 4,
(b) n = 5.
1.4.28. Prove that Qn is a regular bipartite graph.
1.5 Matrices Associated with Graphs

In what follows, the element of a matrix M at position (i, j) is denoted by Mij.
A matrix in which every element is either 0 or 1 is called a binary matrix or
(0, I)-matrix.
Let G be a graph of order n whose vertices are labelled by the numbers 1, ... , n,
i.e., V G = {VI, ... , v n }. Let us define a binary n x n-matrix A = A( G) by setting
A;j = 1 if Vi, Vj are adjacent, and Aij = 0 otherwise. The matrix A(G) is called
the adjacency matrix of the graph G.
The adjacency matrices of multi- and pseudographs are defined in a similar
way: Aij is equal to the number of edges connecting the vertices Vi, Vj assuming
that a loop is counted twice.
Proposition 1.5.1 Two graphs (multigraphs, pseudographs) are isomorphic if

and only if the adjacency matrix of the first one is obtained from the adjacency
matrix of the second one by the same permutations of rows and columns.
Corollary 1.5.2 The characteristic polynomials of adjacency matrices of isomor-

phic graphs are equal.
Therefore we may define the characteristic polynomial of a graph to be the

characteristic polynomial of its adjacency matrix.
Graphs with equal characteristic polynomials are called cospectral. Notice that
there exist nonisomorphic cospectral graphs. This means that a graph is not
defined by its characteristic polynomial. For example, Fig. 1.5.1 shows two pairs
of nonisomorphic cospectral graphs.
Theorem 1.5.3 The degree of a regular graph is a maximal-magnitude root of its

characteristic polynomial.
The Kirchhoff matrix K(G) of a graph G is defined as follows:
-I if Vi and Vj are adjacent,

Kij = { 0: if i "# j and Vi and Vj are nonadjacent,
deg Vi, if i = j.
DA
Figure 1.5.1: Nonisomorphic cospectral graphs
Proposi tion 1.5.1 remains valid if we take the Kirchhoff matrices instead of the
adjacency ones.
Assume now that the edges of G are indexed, i.e., EG = {el, ... , em}.
Define the binary matrix J( G) called the incidence matrix as follows:
I .. _ {I, if the vertex Vi and the edge ej are incident,

ZJ - 0, otherwise.
Proposition 1.5.4 Graphs are isomorphic if and only if their incidence matrices
may be obtained from each other by permutations of rows and columns.
* * *
1.5.1. Prove that the sum of the elements of a row of the adjacency matrix of a
graph is equal to the degree of the corresponding vertex.
1.5.2. Prove that the sum of the elements of a row or of a column of the Kirchhoff
matrix is zero.
1.5.3. The trace of a square matrix is the sum of its diagonal elements. Prove that
the trace of the adjacency matrix of a pseudograph is twice the number of
its loops, and the trace of the Kirchhoff matrix of an ordinary graph is twice
the number of its edges.
1.5.4. Prove the following statements:
a) If graphs G, H are isomorphic then the following conditions hold:
1. A( G) is obtained from A( H) by the same permutations of rows and
columns;
2. K(G) is obtained from K(H) by the same permutations of rows
and columns;
3. J(G) is obtained from J(H) by permutations ofrows and columns;
1.5. Matrices Associated with Graphs 33
b) If one of the above conditions 1-3 holds then G ~ H.
1.5.5. Prove that by the same permutations ofrows and columns the matrix A( G)
may be reduced to
a) the block-diagonal form:
where Ai, i = 1, 2, ... , k, are square matrices, if and only if G is not

connected.
b) the block form
[~ tm]'
where Op is the zero (p x p)-matrix, if and only if G is a nontrivial
bipartite graph.
1.5.6. Let s be a permutation acting on the set {I, 2, ... , n}. Define the permuta-
tion matrix Ms to be the binary (n x n )-matrix such that
if s(j) = i,
otherwise.
Prove the following statements:
a) Each row and each column of Ms contains a single one and det Ms = ±1.
b) (MsA)ij =
AS-l(i)j, i =
1,2, ... ,n, j = 1,2, ... ,m, for any permuta-
tion s and any (n x m)-matrix A.
c) (BMs )ij = =
Bis(j), i 1,2, ... , m, j = 1,2, ... , n, for any permutation
s and any (m x n)-matrix B.
d) Graphs G,H are isomorphic if and only if A(H) = MsA(G)(Ms)-l for
some permutation s.
e) Graphs G, H are isomorphic if and only if I(H) = Ms I(G)(Mt )-l for
some permutations sand t.
1.5.7. Let G and H be graphs of order n with the same vertex set V. Prove that
the solutions of the equation
A(H)X = XA(G), (1.5.1)
where X is a permutation matrix of size (n x n), bijectively correspond to

isomorphisms of G and H, i.e., a permutation s : V -+ V is an isomorphism
of G and H if and only if X = Ms is a solution of (1.5.1).
1.5.8. Let f(x) = xn + C1Xn-1 + ... + Cn-1X + Cn be the characteristic polynomial
. of a graph G. Prove that
a) C1 = 0;
b) -C2 is the number of edges of G;

c) -C3 is twice the number of triangles in G.
1.5.9. Prove that the graphs GI, G2 and HI, H2 from Fig. 1.5.1 are indeed pairs
of nonisomorphic cospectral graphs.
1.5.10. Prove that the (i,j)-th element of the matrix (A(G))k is the number of
(vi,vj)-walksoflength k in G.
1.5.11. Prove that a graph G of order n is connected if and only if the matrix
(A(G) + En)n-l has no zeroes. Here En denotes the identity matrix.
1.5.12. Prove that after a renumbering of the vertices and edges of a graph the
rank of its incidence matrix remains unchanged.
1.5.13. Prove that the rank of the incidence matrix of a connected graph of order
n > 1 is n - 1, if the graph is bipartite; otherwise the rank is equal to n.
1.5.14. A matrix is called totally unimodular if all its minors have the values 0, ±1.
Prove that the incidence matrix of a bipartite graph is totally unimodular.
1.5.15. Let s = (1,2, ... , n) be a cyclic permutation; define
h ~ ~
M, [:: :: 1
Let further f(x) = ao + alX + ... + an_1X n - 1 be a polynomial with real
coefficients. We may define a matrix f(h) as follows:
ao an-l an-2 a2 al
al ao an-l a3 a2
a2 al ao a4 a3
f(h)=ao+a1h+ ... +an_1h n - 1 =
an-2 an-3 an-4 ao an-l

an-l an-2 an-3 al ao
A matrix of such type is called a circulant of order n.
Prove that a graph G is isomorphic to a circulant graph (see Exercise 1.1.26)
if and only if under an appropriate numbering of the vertices the adjacency
matrix of G is a circulant.
1.5.16. Let n be a prime number and let Gi =
Gn,u" i =
1,2, be two circulant
graphs (see Exercise 1.1.26). Prove that each of the conditions below is
necessary and sufficient for G 1 and G 2 to be isomorphic:
a) The spectra of G 1 and G 2 are the same.
b) U2 = kU1 for some positive integer k not divisible by n (see Exercise
1.1.26d).
1.6. Automorphism Group of Graph 35
1.6 Automorphism Group of Graph
First we present some notions from the theory of permutation groups.
A bijective mapping X ---+ X of a nonempty set X is called a permutation of X,
or a permutation acting on X. The product fg of two permutations f, g is specified
by the equality fg(x) = f(g(x)) for any x E X. This product operation defines
the structure of group over the set of all permutations acting on X. This group
is called the symmetric group and is denoted by S(X). The unit element of this
group is the identity transformation
e :X ---+ X, X I--> x, x EX.
If IXI = n, the group S(X) is denoted by Sn, and it is called the symmetric
group of degree n.
Any subgroup F of S(X) is called a permutation group over the set X.
If for any pair Xl, X2 E X the group F contains a permutation f such that
f(xt} = X2 then the group F is called transitive. Otherwise the group is called
intransitive.
We may specify an equivalence relation", over X by setting x '" y if and only
if the group F has a permutation f such that f(x) = y. This relation partitions
X into the classes of equivalent elements: any two elements from the same class
are mapped into each other by an appropriate permutation from F, while elements
from different classes cannot be mapped into each other by permutations from F.
These classes are called the orbits of the group F. The orbits of the group Fare
F -invariant subsets of X, i.e., if Y is an orbit, then F(Y) = {f(y) : y E Y, f E
F} = Y. Every F-invariant subset of X is the union of some orbits of F.
Let F be a permutation group over X and suppose that a E X. Denote Fa =
{f E F : f( a) = a}. The set Fa is a subgroup of F. It is called the stabilizer of the
element a.
Theorem 1.6.1 Let F be a permutation group over X. Let us take a E X, and let
Y be an orbit of F that contains a. For every element y E Y let us pick fy E F such
that fy(a) = y. Then {fy : y E Y} is a complete system of left coset representatives
of F modulo Fa, i.e.,
F= fyFa U
yEY
is the decomposition of F into left cosets modulo Fa.
Let F be a permutation group over X and let H be a permutation group over

Y. If there exists a bijection cp : X ---+ Y such that
(1.6.1)
then it is said that the permutation groups F and H are similar. The relation
(1.6.1) implies that the mapping
'lj;: F ---+ H: f I--> cpfcp-l, f E F

is an isomorphism of groups F and H. However similarity is a stronger relation

among groups than isomorphism: groups may be isomorphic but not similar.
Consider one of such examples. In what follows, (a, b, ... ,c} denotes the sub-
group of a group generated by the elements a, b, ... , c.
Let a = (1,2) and b = (3,4) be transpositions and let c = (1,2)(3,4) and
d = (1,3)(2,4) be products of two transpositions. Suppose further F = (a, b} and
H = (c, d}. Then F and H are isomorphic subgroups of S4 which are not similar.
The group H is called the Klein four-group.
Similar permutation groups are isomorphic, and what is more, they" act sim-
ilarly", i.e., a permutation f E F acts on x E X in the same way as cpjcp-l acts
on cp(x) = y, i.e., if f(x) = x' then cpfcp-1(cp(x)) = cp(x').
Let us define two operations on the set of permutation groups. The first one is
the direct product. If Xl, X 2 are nonempty disjoint sets and Fi is a permutation
group over Xi, i = 1,2, then the direct product F1 x F2 is the set
that acts on the union Xl U X 2 as follows:
Clearly, IFl x F21 = IFd1F21·

The second operation is the wreath product. Let F be a permutation group
over X and let H be a permutation group over Y. Let further II be the set of all
mappings Y --+ F. Define the operation" +" over II by setting
(1.6.2)
for any 7r1, 7r2 E II, Y E Y. Then the set II turns out to be a group isomorphic to
the direct product of k = IYI copies of F. The zero of the group II is the mapping
0: Y --+ F: y 1-+ ex, y E Y,
where ex is the unit element of F. For 7r E II, (-7r)(Y) = (7r(y))-l.

If 7r E II, h E H, then 7rh E II.
Let F Wr H denote the set of all pairs (h, 7r), 7r E II, h E H. Define the action
of a pair (h, 71") on the Cartesian product Z = X x Y as follows: if (x, y) E Z, then
(h, 7r)(x, y) = (7r(Y)(x), hey)). (1.6.3)

The set F Wr H of all pairs (h,7r) acting on Z according to (1.6.3) is called the
wreath product of permutation groups F and H.
Theorem 1.6.2 The wreath product F Wr H is a permutation group over the set
Z = X x Y such that
(1.6.4)
for any (hi, 7ri) E F Wr H, i = 1,2.

The equality (1.6.4) implies that the set of all pairs of form (ey, 11"), where ey
is the unit of H, constitutes a subgroup II' isomorphic to II, and the set of all pairs
of form (h, 0) constitutes a subgroup H' isomorphic to H. The group II' is normal
in F Wr H. In addition, the following equalities are true:
FWr H = H'II', IFWr HI = IHIIFIIYI.

Proposition 1.6.3 F Wr H is transitive if and only if both groups F and Hare
transitive.
A permutation <p that acts on the set VG of the vertices of a graph G and pre-
serves the adjacency relation (i.e., such that the images <p(u) and <p(v) of vertices
u, v are adjacent if and only if u, v are adjacent) is called an automorphism of the
graph G. In other words, an automorphism of a graph is an isomorphism of the
graph onto itself.
Theorem 1.6.4 The set of all automorphisms of a graph is a permutation group

over the set of its vertices.
It is called the automorphism group of the graph G and is denoted by Aut G.

Theorem 1.6.5 If a graph G is the disjoint union of graphs F, H such that no
component of F is isomorphic to a component of H then Aut G = Aut Fx Aut H.
Theorem 1.6.6 If all components of an n-component graph G are isomorphic to

each other then Aut G is similar to the wreath product (Aut F) Wr Sk, where F
is a component of G and Sk is the symmetric group of degree k.
The latter two theorems imply that finding the automorphism group of a graph
amounts to two problems for connected graphs: recognition of isomorphism of
connected graphs and finding the automorphism group of a connected graph.
* * *
1.6.1. Let F be a permutation group over X and let Y be an F-invariant subset
of X. Prove that Y is the union of some orbits of F.
1.6.2. Prove that isomorphic subgroups of the symmetric group may turn out to
be dissimilar.
1.6.3. Prove that IF Wr HI = IFlklHI if H is a subgroup of the symmetric group
Sk.
1.6.4. Prove a) the Theorem 1.6.2; b) the Proposition 1.6.3.
1.6.5. Prove the Theorem 1.6.4.
1.6.6. Prove that Aut G = Aut G for any graph G.
1.6.7. Prove that the automorphism group of the graph from Fig. 1.6.1 is the
identity group: Aut G = (e), where e is the identity permutation. Find
other graphs with such automorphism group.

7
5 8
1.6.8. a)Prove that if G is a graph of order n ~ 5 then Aut G:/= (e).

b) Find all six-vertex graphs with the identity automorphism group.
1.6.9. Prove that if graphs are isomorphic then their automorphism groups are
similar.
1.6.10. Find the automorphism groups for a) On; b) Kn; c) Pn ; d) C n ; e) the graph
from Fig. 1.6.2.
1.6.11. A permutation of the finite set is called even if it is a product of even

number of transpositions. Prove that
a) the set An of all even permutations is the subgroup of the symmetric

group Sn (it is called the alternate group).
b) there are no graphs of order n > 3 with Aut G = An.
1.6.12. Prove a) the Theorem 1.6.5; b) the Theorem 1.6.6.
1.6.13. Prove that the automorphism group of a disconnected graph is transitive
if and only if all components of the graph are isomorphic and their automor-
phism groups are transitive.
1.6.14. Find Aut G if G is a graph that has exactly
a) four components: C4, C4 , C4 , P3;

b) two components: Km, Kn, m :/= n.
c) two components: Kn, Kn.
1.6.15. Solve the Exercise 1.6.14b after the addition of an edge making the graph
connected.
1.6.16. Let G be a graph. For f E Aut G, let U ~ VG be a subset invariant with
respect to f and let f' be the restriction of the mapping f onto U, i.e.,
f' : U ~ U, f'(x) = f(x), x E U.
Prove that f' E Aut (G(U)), where G(U) is the subgraph induced by the
set U.
1.6.17. Let f be an automorphism of a graph G. Define a mapping 1 of the set of
maximal cliques of G by setting I(X) = {f(x) : x E X} for any maximal
clique X. Prove that 1 is an automorphism of the clique graph Q( G).
1.6.18. Let G be a nonempty graph, f E Aut G. Define a mapping fE : EG ~ EG
by setting fE(e) = f(a)f(b) for e = abo Further, define
AutEG = {IE: f E Aut G}.
It is called the edge group of graph G. Prove the following statements:
a) AutEG is a subgroup ofthe group Aut (L(G)).

b) AutEG ~ Aut G if and only if the following two conditions hold:
(1) G has at most one isolated vertex;
(2) no component of G is isomorphic to K 2 •
c) If one of the conditions
(1) G has no triangles and 5(G) > 1;
(2) 5(G) > 3
holds, then Aut (L(G)) = AutE(G) ~ Aut G.
1.6.19. Find Aut (L(G)) for a) G = K 4 ; b) G = [{5.
1.6.20. Find the automorphism group of the Petersen graph.

1.6.21. Find an edge-symmetric graph which is not vertex-symmetric. (This means
that AutEG is symmetric but Aut G is not.)
1.6.22. Find the edge group for the graph from Fig. 1.6.3.
1.6.23. Prove that the number of pairwise distinct labelled graphs isomorphic to
a graph G of order n is n!/I Aut GI.
1.6.24. Let G be a graph of order n. Demonstrate that
Aut G = {s E Sn : MsA(G) = A(G)Ms},

where Ms is the permutation matrix corresponding to s (see Exercise 1.5.6),
A(G) is the adjacency matrix of G.
1.6.25. Prove that a graph G of order n is isomorphic to a circulant graph (see
Exercise 1.1.26) if and only if Aut G contains a cyclic permutation of length
n.
1.6.26. Let G, F be connected graphs such that ~(G) = ~(H). Define a new graph
H = G U F + ab by adding an edge ab to the disjoint union G U F. Prove that
for any vertex a of maximal degree in G there exists a vertex b of maximal
degree in F such that the following statement is valid: G e:: F if and only if
there exists an automorphism <p of H such that <pea) f:. a.
Chapter 2
Trees
2.1 Trees: Basic Notions

A tree is a connected graph without cycles. An arbitrary graph without cycles is
called a forest (or acyclic graph). Therefore the components of a forest are trees.
Theorem 2.1.1 The following statements are equivalent for an (n, m )-graph G :
1. G is a tree.
2. G is a connected graph and m =n - 1.
3. G is a forest and m = n - 1.
4. Any two vertices of G are connected by a single simple path;
5. G is a forest with the following property: if we connect any pair of nonadja-

cent vertices by an edge, a single cycle will be created in the graph.
* * *
2.1.1. Prove that an acyclic graph with n vertices and k components has exactly
n - kedges.
2.l. 2. Are there isomorphic graphs among the trees shown at Fig. 2.1.1?
.11 ... I.I.....L.+

2.l.3. Draw all pairwise distinct labelled trees of order 4.
41
42 2. Trees
2.1.4. Draw all pairwise nonisomorphic trees of order n for a) n = 5; b) n = 6; c)

=
n 7; d) n 8. =
2.1.5. Let G be a connected (n, m)-graph. Prove that if G is not a tree then
m>n-1.
2.1.6. Prove that each tree is a bipartite graph. Which complete bipartite graphs
are trees?
2.1.7. Prove that every connected subgraph of a tree is the induced one.
2.1.8. What is the maximal possible degree ~(G) of the vertices of a tree G of
order n ~ 3?
2.1.9. Find all trees that are selfcomplementary graphs.
2.1.10. Which trees are regular graphs?
2.1.11. Prove that the following statements are equivalent for a connected graph
G:
- G has a single cycle;

- G is an (n, n)-graph for some n;
- G has an edge e such that G - e is a tree.
2.1.12. Let Tk be the set ofall trees without vertices of degree 2 and with k pendant
vertices (and with any number of all vertices). Prove that Tk is finite for any
k.
2.1.13. The mean degree (of vertices) of a graph of order n is the value dmean =
itL~=l deg v. Express the number of vertices of a tree in terms of the mean
degree.
2.1.14. A vertex i of a tree is called branching if di = deg i ~ 3. Express the
number of the pendant vertices of a tree in terms of the degrees di of the
branching ones.
2.1.15. Prove that any nontrivial tree

a) has at least two pendant vertices;
b) has exactly two pendant vertices if and only if the tree is a path.
2.1.16. Prove the following statements for a tree:
a) For any simple path P starting at a vertex v there exists a pendant
vertex w such that P is a subpath of the simple (v, w)-path.
b) If a vertex v is the most distant one from a vertex u then v is a pendant
vertex.
c) A vertex of maximal excentricity is pendant.
2.1. Trees: Basic Notions 43
2.1.17. Consider a tree T with exactly k pendant vertices. If we number these

vertices by L 2, ... , k, we may define the k x k-matrix D of distances between
the pendant vertices of T as follows: Dij = d( i, j).
a) How will D change if we change the numbering of the pendant vertices

ofT?
b) Prove that
Dij + Dj k - Dik > 0;
Dij + Djk - Dik == 0 (mod 2)

for any three different indices i, j, k.
c) Prove that a tree is specified by its matrix D up to isomorphism.
d) Can you draw the trees with the following matrices D?
0 2
[ 2 0 1
1 2
2] 2 2
[0 0333 ]
1 1 o 1 ,D2 = 3 3 0 2 .
2 2 1 0 3 320
e) Design an algorithm for reconstructing a tree from its matrix D.
2.1.18. Find a formula for the number of pendant vertices for a tree of order n
whose non-pendant vertices are of the same degree.
2.1.19. Prove that for any tree T of order n 2:: 3 the number of pendant vertices
is at most
n(A(T) - 2) + 2
A(T) -1
where A(T) is the maximal vertex degree in T.
2.1.20. Let a tree of order n 2:: 2 have k pendant vertices and no vertices of degree
2.
a) Prove that k 2:: (n + 2)/2.
b) Characterize the trees that attain this bound (i.e., the trees with n =
2k - 2.)

a) Every central vertex of a tree belongs to every its diametral path.
b) The center of a tree coincides with the center of any its diametral path.
c) The center of a tree consists of either one or two vertices (Jordan's
theorem).
d) The center of a tree T may be found by means of the following Jordan's
algorithm. Delete all pendant vertices of T to obtain a tree T'. Delete
all pendant vertices of T' to obtain Til . ... Repeat this operation until
only one or two vertices remain, which constitute the center ofthe initial
tree T.
44 2. Trees
e) In a tree of odd diameter any two diametral paths have a common edge.
2.1.22. Using Jordan's algorithm, find the center of the tree shown at Fig. 2.1.2.
2.1.23. Prove that a tree is of diameter 2 if and only if it is a star.
2.1.24. Draw all pairwise nonisomorphic trees of order n with one and with two
central vertices for n = 4,5,6,7,8 (cf. Exercise 2.1.4).
2.1.25. Characterize all connected graphs in which every maximal under inclusion
simple path is diametral.
2.1.26. Prove that the radius reT) and the diameter d(T) of a tree T are related
by
r
reT) = d(T)j21·
2.1.27. Find all trees with the diameter equal to the radius.
2.1.28. What values can the diameter assume for the graph complementary to a
nontrivial tree that is not a star?
2.1.29. Prove that a tree of order n with diameter at least 2k - 3 has at least n - k
different paths of length k.
2.1.30. Centroid of Tree. A branch of a tree T at vertex v is a maximal subtree

of T in which v is a pendant vertex. Thus the number of the branches at a
vertex v is deg v. Define weight of a vertex v E VT to be the number
w(v) = max{IEHI: HE Tv},
where Tv is the number of branches ofT at v. Clearly, the weight of a pendant

vertex is IETI. A vertex Vo is called centroidal for a tree T if
w(vo) = min{w(v) : v E VT}.
The number w( vo) is called the weight of T and is denoted by w(T). The
centroid of a tree is the set of its centroidal vertices.
a) Draw a tree with 1) one central and one centroidal vertices; 2) one
central and two centroidal vertices; 3) two central and one centroidal
vertices; 4) two central and two centroidal vertices;
b) Construct a tree that has disjoint center and centroid, both consisting
of two adjacent vertices.
c) Prove that for every tree the centroid consists either of one vertex or
two adjacent vertices.
d) Prove that a tree T has a single centroidal vertex if and only if
1
w(T) :::; "2IET1,
and it has two centroidal vertices if and only if
1
w(T) = "2(IETI + 1).
e) Prove that the centroid of a tree may consist of two vertices only if the
tree is of even order.
2.1.31. A caterpillar is a tree that turns into a simple path after the deletion of
all pendant vertices. Prove that:
a) a tree is a caterpillar if and only if it contains no subgraph isomorphic to
the graph from Fig. 2.1.3.
b) A tree T of order n is a caterpillar if and only if there exists a bijection

'P : ET ---> {I, 2, ... , n -1} such that the following condition holds: for every
vertex v E VT the set of edges {ei} incident to v may be indexed in such a
=
way that 'P(ei+d 'P(ed + 1, i = 1,2, ... , deg v - 1. A tree that satisfies the
latter condition is called an interval tree.
2.1.32. Find a tree of minimal order n 2 2 with the identi~y automorphism group.
2.1.33. Find all trees with transitive automorphism group.
2.1.34. Prove that every tree of order n is isomorphic to a subgraph of every graph
with minimal vertex degree at least n - 1.
2.1.35. Characterize the classes of trees specified by the following forbidden sub-
graphs, i.e., the classes of graphs that do not contain subgraphs isomorphic
to: a) K 1 ,3; b) P3 ; c) Pic, k 2 4; d) the graph from Fig. 2.1.4; e) the graph
from Fig. 2.1.5.
46 2. Trees
Figure 2.1.4: To Exercise 2.1.35d
Figure 2.1.5: To Exercise 2.1.35e
2.1.36. Let A be the incidence matrix of a tree of order n ~ 2. Prove that any
n - 1 rows of A are linearly independent over the field {O, I}.
2.1.37. Priifer Codes. These codes are intended for coding the trees. Consider a
tree T of order n with the vertex set {1,2, ... ,n}. The Priifer code ofT is
constructed as follows. Let bl be the smallest number which is a pendant
vertex of T. Let el = bl al denote the corresponding pendant edge. Deleting
el and bl from T we obtain another tree T I . We repeat this operation for
TI : we find the smallest number b2 which is a pendant vertex, and e2 = b2a2
denotes the corresponding pendant edge. This operation is repeated until
we are left with a single edge en-l = bn-Ian-l' The resulting sequence
(aI, ... , a n - 2) is called the Priifer code of the tree T.
Construct the Priifer codes for the trees from Fig. 2.1.6.
2.1.38. Let (aI, a2,.'" an -2) be the Priifer code of a tree T with the vertex set
{I, 2, ... , n}. Prove that
{1,2, ... ,n}\{al,a2, ... ,a n-2}
is the set of the pendant vertices of T.

2.1.39. a) Design an algorithm for reconstructing a tree from its Priifer code.
b) Reconstruct the trees from their Priifer codes: ~
(1) (3,3,5,5,6,6); (2) (1,5,1,5,9,8,2); (3) (1,5,2,2,1,5,5).
a) The Priifer codes of two different labelled trees are different.
b) Every sequence of length n - 2 of numbers from the set {I, 2, ... , n} is
the Priifer code of some labelled tree.
2.1.41. What tree has the Priifer code (i, i, ... , i)?
/tr7
a b c
d d
2.1.42. For what trees of order n ~ 4 all components of the Priifer code are
different?
2.1.43. Using Priifer codes, prove the Cayley Theorem: the number of labelled
trees of order n is nn-2.
2.1.44. Suppose that the collection of graphs P(G) = (G l , G 2 , ••• , Gn ) is the deck
of a graph of order n ~ 3 (see Sect. 1.3). The graphs Gi are unlabelled.
Prove that the collection P( G) allows to recognize whether G is a tree.
2.1.45. Let T 1 , T 2 , ... , Tk, k > 1, be subtrees of a tree T such that any two ofthem
intersect each other. Prove that
a) there exists a vertex of T common to all T;;
b) the subgraph induced by n~=l V'li is a tree.
2.1.46. Consider the following function defined over the vertex set of a tree T:
e(x) = vEVT
max d(x, y), x E VT
(the excentricity of x). Prove that this function has the convexity property:
2e(x) :::; e(y) + e(z),

where y and z are vertices adjacent to x.
2.1.47. Consider the following function defined over the vertex set of a tree T :
O"(x) =L d(x, v), x E VT.

vEVT
48 2. Trees
Prove that it has the strict convexity property:
2cr(x) < cr(y) + cr(z),

where y and z are vertices adjacent to x.
2.1.48. Barycenter of Tree. The vertex Va of a tree T is called barycentral if
cr(va) = min{cr(v) : V E VT},
where cr function is defined in Exercise 2.1.47.

The barycenter of a tree is the set of its barycentral vertices.
a) Prove that each tree has a barycenter consisting of one or two adjacent
vertices.
b) Give an example of a tree whose barycentral and central vertices are at
distance 100 apart.
2.1.49. Find trees of a given order for which :L(x,y) d(x, y), where the summation
is over all unordered pairs of vertices, is a) minimal; b) maximal.
2.1.50. Find a recurrence relation for the number ten, k) of labelled trees of order
n with k pendant vertices.
2.1.51. Prove that in an odd-order tree every automorphism has a stationary ver-
tex.
2.1.52. Prove that the excentricity of a vertex v of a tree T may be computed by

the formula
e(v) = d(v, Z(T)) + reT),
where Z(T) is the center of T and reT) is the radius.
2.1.53. The subtree intersection graph SJ(T) of a tree T is defined as follows. The
vertices of SJ(T) bijectively correspond to the subtrees of T. Two vertices
of SI(T) are adjacent if and only if the corresponding subtrees of T have a
common vertex.
Prove that for any tree T the graph SJ(T) has no induced subgraphs iso-
morphic to Ck, k ~ 4.
2.2 Skeletons and Spanning Trees

Let H be a spanning subgraph of a graph G. If H induces a tree on every connected
domain of G then H is called a skeleton of the graph G. Thus, a skeleton of a graph
is a forest (spanning forest). For a connected graph, any spanning subgraph which
is a tree (spanning tree) is a skeleton. Clearly, every graph has a skeleton: one can
obtain it successively destructing cycles by deletions of edges.
2.2. Skeletons and Spanning Trees 49
Theorem 2.2.1 Let G be an (n, m)-graph with k components. The number of edges
that must be deleted from G to obtain a skeleton is independent on the order of
their deletion and is equal to m - n + k.
The number m - n + k is called the cyclomatic number or cyclic rank of the

graph and is denoted by v( G). The number n - k of edges of its skeleton is called
cocyclic rank of the graph and is denoted by v*(G).
Theorem 2.2.2 Every acyclic subgraph of a graph G is contained in some skeleton

ofG.
Theorem 2.2.3 (Kirchhoff Theorem) The number of skeletons in a connected

labelled graph G of order n 2: 2 is equal to the cofactor of any element of the
J( irchhoff matrix f{ (G).
Theorem 2.2.4 (Cayley Theorem) The number of labelled trees of order n is

nn-2.
A graph G is called weighted if a weight function w: EG -+ R + is specified

that assigns a weight wee) to every edge e E EG. The weight w(H) of a subgraph
H of G is the sum of the weights of edges of H :
w(H) = L wee).
eEEH
A weighted graph G with the weight function w is denoted by (G, w).

Minimal Spanning Tree Problem: Find a spanning tree of minimal weight in
a weighted connected graph (G, w).
We describe the algorithms due to Kruskal and Prim for solving this problem.
Kruskal's Algorithm
1. Construct a graph Tl = On + el by adding an edge el E EG of minimal
weight to the empty graph with the vertex set VG.
2. Having T i , i < n -1, constructed, we construct Ti+l = Ti + ei+l, where ei+l

is an edge of G with minimal weight among the edges which are not in ~
and do not produce a cycle when added to ~.
Tn - 1 is the minimum weight spanning tree for (G, w).
Prim's Algorithm
1. Pick an edge el = ab of minimal weight and construct the tree Tl by setting
VT1 = {a,b}, ETl = {ed.
2. Having ~ of order i + 1, i < n -1, constructed, among the edges connecting

it with the vertices of G\T; we pick an edge ei+l of minimal weight. Ti+l is
obtained from T; by adding ei+l together with the endpoint which is not in
Ti .
50 2. Trees
Tn - l is the minimum weight spanning tree for (G, w).

The problem of finding a spanning tree of maximal weight may also be solved
by the algorithms of Kruskal and Prim; one must only pick edges of maximal
weight rather than of minimal.
* * *
2.2.1. Prove that the cyclic rank of a graph cannot be negative.
2.2.2. Consider the following operations with a graph:
=
1. if a vertex v is incident only to two edges el VUl and e2 VU2, then =
v is deleted (together with el and e2) and the edge UlU2 is added;
2. an edge Ul U2 is replaced by two edges VUl, VU2, v being a new vertex.
Prove that the cyclic rank of a graph is invariant with respect to these
operations.
2.2.3. Prove that a graph is a forest if and only if its cyclic rank is zero.
2.2.4. Find all skeletons of the following labelled graphs: a) [{4, b) P6, c) C7 , d)
[{l,3, e) [{2,3'
2.2.5. Find the numbers of nonisomorphic skeletons of the following graphs: a)

[{4, b) [{5, c) [{6, d) C7, e) a 6-vertex wheel graph W 5 , f) [{2,3, g) 1(3,3,
2.2.6. Using the Kirchhoff theorem, find the number of skeletons in graphs from
Fig. 2.2.1 (the vertices of graphs must be indexed first).
a b c d
2.2.7. In what cases a graph has a single skeleton?

2.2.8. Prove that any nontrivial connected graph has a skeleton such that if we
delete all its edges, the graph becomes disconnected.
2.2.9. Demonstrate how one can find a skeleton of a graph using the breadth-first
search.
2.2.10. Let H be a subgraph of a connected graph G of order n. Consider the

following conditions:
a) H has n vertices;
b) H has n -1 edges;
c) H is connected;
d) H is a forest.
Prove that:
1. Any three of these conditions taken together imply that H is a spanning

tree of G.
2. The conditions b), d) together imply this;
3. No other pair of conditions a)-d) imply this.
2.2.11. Prove that a subgraph H is a skeleton of a graph G if and only if the

following equalities hold:
IHI = IGI; m(H) = v*(G); k(H) = k(G).

2.2.12. Find spanning trees of maximal (minimal) weights for graphs from Fig.
2.2.2 using algorithms of Kruskal and Prim.
a c
2.2.13. Prove that every pendant vertex of a graph is contained in every skeleton
of the graph.
2.2.14. Prove that every edge of a graph belongs to some skeleton of the graph.
2.2.15. Prove that a subgraph of a graph G is its skeleton if and only if it is a
maximal subgraph of G without cycles.
52 2. Trees
2.2.16. Let G be obtained by "gluing together" two connected graphs G I , Gil at

a single vertex. Express the number of spanning trees of G in terms of the
numbers of spanning trees of G I and Gil.
2.2.17. Prove that if a graph has a simple cycle of length k then the graph has at
least k skeletons.
2.2.18. Does there exist a graph with exactly two skeletons?
2.2.19. Which graphs are such that every skeleton has exactly two pendant ver-
tices?
2.2.20. Draw a graph that has two skeletons with disjoint edge sets.
2.2.21. Prove the following two statements for connected graphs.
a) For any vertex v of a graph G there exists a spanning tree T of G such

that dG(v, w) = dT(v, w) for any vertex wE VG. Here dH(V, w) denotes
the distance between v and w in the graph H.
b) The center of a graph is contained in the union of the centers of its
skeletons.
2.2.22. Let T be a spanning tree of a connected graph G. Prove that d(G) ::; d(T),
where d(X) denotes the diameter of the graph X.
2.2.23. Is it true that in a connected graph of diameter d ~ 2

a) the diameter of any spanning tree is at least d?
b) a spanning tree of the same diameter exists?
2.2.24. Prove or disprove the statements:
a) If the diameter of a graph is 2 then the graph has a skeleton which is a

star.
b) If a graph has a skeleton which is a star then the diameter of G is 2.
2.2.25. How many skeletons are there in the labelled complete graph of order n.
2.2.26. Using the Kirchhoff theorem, find the number of skeletons in the labelled
complete bipartite graph J{m,n.
2.2.27. Prove that the number oflabelled trees which are labelled bipartite graphs
with sizes of parts m and n, is mn-1n m - 1 .
2.2.28. In a random labelled tree of order n a random vertex is picked. Assuming

that all outcomes are equiprobable, prove that
.
11m 1
Pn = -,
n--+oo e
where Pn denotes the probability of the event that the picked vertex is pen-
dant.
2.2.29. Let T 1 , T2 be skeletons of a graph C.
a) Prove that for any edge el E ETI there exists an edge e2 E ET2 such
that (Tl - eJ U e2 is also a skeleton of C. This replacement of el bye2
will be called switch.
b) Prove that any skeleton may be obtained from any other one by a
sequence of switches (producing a skeleton at each switch).
1 3 4
1 2
4 3
--------------~.
2.2.30. A graph of skeletons S( G) of a graph G is the graph whose vertex set is

the set T 1 , T 2 , ... ,Tr of skeletons of G and two its vertices T; and 'Fj are
adjacent if and only if the union of skeletons T; U Tj in G is a graph with a
single cycle.
a) Construct a graph of skeletons for the graph from Fig. 2.2.3.

b) Prove that the skeletons T; and Tj of G are adjacent in S( G) if and only
if there exist edges el E ET; and e2 E E'Fj such that (T; - el) U e2 = 'Fj.
c) Prove that S( G) is connected.
d) Prove that for skeletons T; and'Fj of a connected graph G the distance
between the corresponding vertices in S( G) is equal to IETi \ ETj I.
e) Prove that the diameter of S( G) is at most IGI - 1.
f) Characterize nontrivial graphs for which the diameter d( S( G)) of its
graph of skeletons is IGI - 1.
g) Prove that IGI- d(S(G)) may be arbitrarily large.
h) What is the maximal possible diameter of S(G) if keG) = k?
i) Characterize graphs G with the mentioned above diameter of S( G).
54 2. Trees
2.2.31. Construct all skeletons of the labelled graph G from Fig. 2.2.4 and draw
the S(G).
2.2.32. Prove that every minimal spanning tree of a connected graph may be
constructed by Kruskal's algorithm.
2.2.33. Let T be a minimal spanning tree of a connected weighted graph (G, w)

with weights Cl ::; C2 ::; ... ::; Cm of the edges of T. Let further T' be a
spanning tree of the same graph with weights d 1 ::; d 2 ::; ... ::; dm of the
edges of T'. Prove that Ci ::; di , i = 1, 2, ... , .
Chapter 3
Independence and
Coverings
3.1 Independent Vertex Sets and Cliques

A set of vertices of a graph is called independent if no two vertices in it are adjacent.
In other words, a set 5 ~ V G is independent in G if the induced subgraph 5( G)
is empty. Clearly, in this case any subset 5' ~ 5 is also independent.
An independent vertex set is called maximal if it is not a proper subset of
any other independent set. An independent set with maximal number of elements
is called a maximum cardinality independent set. The number of vertices in a
maximum cardinality independent set of a graph G is called the independence
number of G and is denoted by ao(G).
Below are some estimates for the independence number.
ao(G) 2 I: (l+degv)-l. (3.1.1)

vEVG
IGI
a o(G)2 1 + d , (3.1.2)
IGI
ao(G) 2 1 + b.(G) , (3.1.3)
where d = (L":vEVG deg v)/IGI·

The notion of the clique is antipodal to the notion of the independent set. A
nonempty subset of vertices of a graph is called a clique if it induces a complete
subgraph. A clique is called maximal if it is not a proper subset of any other clique.
A clique with maximal number of elements is called a maximum cardinality clique.
The number of vertices in the maximum cardinality clique of a graph G is called
the toughness of the graph G and is denoted by cp( G).
Clearly, a subset of vertices of a graph G is a clique if and only if it is an
independent set in the complementary graph G. Therefore cp( G) = ao (G), and the
inequalities (3.1.1)-(3.1.3) provide bounds for cp(G) as well.
55
56 3. Independence and Coverings
All cliques, as well as all maximal cliques, constitute a covering of the vertex
set of a graph. The minimal number of cliques of G covering VG is denoted by
c(G). Clearly, c(G) ~ ao(G) for any G.
The minimal number of complete graphs whose union is G is denoted by cc(G).
This number is also related to cliques, since cliques induce complete subgraphs.
The intersection graph for the set of maximal cliques of G is called the clique
graph and is denoted by Q(G). Thus the vertices of QCG) bijectively correspond
to the maximal cliques of G and two vertices of Q( G) are adjacent if and only if
the corresponding cliques intersect.
* * *
3.1.1. Define the independence numbers for the graphs from Fig. 1.2.4.
3.1.2. Define the independence numbers for the following graphs: a) Kn; b) K1,n;
c) Km,n; d) Pn ; e) en; f) Petersen graph; g) n-cube Qn.
3.1.3. Define the independence numbers for the graphs of Platonic solids (Fig.
1.1.2).
3.1.4. Characterize the graphs for which
a) every vertex subset is independent;

b) no independent set has more than one vertex.
3.1.5. Is it true that every maximal independent set of vertices of a graph is the
maximum cardinality independent set?
3.1.6. How may the independence number of a graph change
a) after the deletion of one vertex;

b) after the deletion of one edge;
c) after the addition of one edge.
3.1.7. Let GCA, B; E) be a bipartite graph. Is it true that ao(G) = max{IAI, IBI}?
3.1.8. Prove that G is bipartite if and only if aoCG) ~ IHI/2 for every subgraph
H ofG.
3.1.9. Prove that if G is a tree then ao(G) ~ IGI/2.

3.1.10. Prove that every connected graph other than K2 has a maximal indepen-
dent set of vertices which contains all pendant vertices of the graph.
3.1.11. Propose an algorithm for the construction of a maximal independent set

of vertices of a tree.
3.1.12. Prove that if the distance between any two pendant vertices of a tree is
even then the tree has a single maximum cardinality independent vertex set.
Is the converse true?
3.1. Independent Vertex Sets and Cliques 57
3.1.13. Let a~(G) denote the minimal cardinality of maximal independent sets of
vertices of a graph G. Find a~(G) for a) Kn; b) K1,n; c) Km.n; d) Pn ; e) en;
f) Petersen graph; g) n-cube Qn.
3.1.14. For any positive integer a, construct a graph G such that
ao(G) - a~(G) = a.
3.1.15. Prove that for graphs G, H
a) ao( G U H) = ao( G) + ao(H), if VG n V H = 0.

b) ao( G + H) = max( ao( G), ao(H»;
c) ao(G x H) ;::: ao(G)ao(H).
See Section 1.4 for the operation with graphs.
3.1.16. a) Prove the formula (3.1.1).

b) Prove that the bound (3.1.1) is attainable for a nontrivial graph.
c) Design an algorithm for finding an independent set of cardinality at
least I:vEVG(1 + deg V)-l in a graph G.
3.1.17. Prove the formula (3.1.2).
3.1.18. Let I be a maximal independent vertex set ofa graph G and let p(I) denote
the number of edges connecting the vertices of I with the vertices of VG\I.
Prove that p(I) + III;::: IGI·
3.1.19. Prove the formula (3.1.3) without making use of (3.1.2).
3.1.20. Let l( G) be the minimal number of paths of a graph whose union contains
all vertices of G. Prove that I(G) ::::; ao(G).
3.1.21. Prove that ao(G) ::::; IEGI/D(G) for all graphs G without isolated vertices.
3.1.22. Prove that ao(G) ::::; pO + (p+ + p-)/2 for all graphs G, where pO,p+,p-
are the numbers of zero, positive and negative eigenvalues of the adjacency
matrix A( G) respectively.
3.1.23. Let ei =
UVi, i =
1,2, ... , deg u, be the edges of a graph G incident to
a vertex u. Consider the graph G' obtained from G by the addition of a
new vertex u' and new edges uu' and U'Vi, see Fig. 3.1.1. (This operation
producing G' from G is called vertex splitting.) Prove that ao( G') = ao( G).
3.1.24. Thran graph: Let G be a graph satisfying the following conditions:
(1) IGI = n;
(2) ao( G) ::::; a;
58 3. Independence a.nd Coverings
(3) among all graphs satisfying the conditions (I), (2), G has the minimal
number of edges.
Such graph is called the Turan graph and is denoted by G~ or' where m is the
number of its edges. '
Prove the following properties of the Thran graph:
a) any two adjacent vertices are of the same degree;

b) the degrees of any two nonadjacent vertices differ by at most 1;
c) every component of G~or is a complete graph;
d) the components of G~or are q graphs KP+l and a - q graphs K p , where
p is the quotient and q is the remainder under the division of n by a
(n = ap + q, 0 ~ q ~ n - 1.)
3.1.25. Express the number m of edges of Thran's graph G~ or in terms of p, a, q

(see the previous exercise). '
3.1.26. Prove that if ao(G) = a for an (n, m)-graph G then

a ap
m>p(n- - - - )
- 2 2'
wherep= In/aJ.
3.1.27. Prove that
n2
ao(G) > 2(m + n)
for any (n, m)-graph G.
3.1.28. a) Prove that ao( G) = ao( GP), p ;::: 3, (see Sect. 1.4) if and only if every
component of G is a complete graph.
b) Construct a graph for which the above statement fails for p = 2.

3.1.29. A graph is called edge aD-critical if after the deletion of any edge the
independence number increases. Give examples of edge ao-critical graphs
and of graphs not of this class.
3.1.30. Prove that
a) every nonempty graph has a subgraph which is an edge ao-critical

graph;
b) every graph is an induced subgraph of an edge aD-critical graph.
3.1.31. Which Platonic solids (see Fig. 1.1.2) are edge aD-critical graphs?
3.1.32. Construct the clique graphs Q(G) for a) On; b) Kn; c) Kl,n; d) Pn ; e) C n ;
f) Km,n; g) Petersen graph.
3.1.33. Prove that if <p( G) ~ p then G has at least p vertices with degrees at least
p - 1.
3.1.34. Find a graph with maximal number of edges for a given number of vertices
n and toughness <po How many edges does it have?
3.1.35. Prove that a graph G is isomorphically embeddable into G' if and only if
<p(G <> G') ~ IGI (see Sect. 1.4).
3.1.36. Prove that a graph G has a subgraph isomorphic to G' if and only if
<p(GoG') ~ IGI (see Exercise 1.4.21).
3.1.37. Threshold Graph: Let G be a graph with vertex set VG = {Vl' V2, ... , v n }.
For a subset U ~ VG the characteristic vector Xu = (Xl, X2,' .. , xn) is
defined as follows:
X. _ { 1, if Vi E U;
, - 0, otherwise.
A graph G is called a threshold graph if there exist nonnegative reals
such that a subset U ~ VG is independent if and only if the vector Xu

satisfies the inequality
which is called a separating inequality for G. (The empty set is defined to be

independent. )
a) The mapping U 1-+ Xu is a bijection of the set of subsets of VG onto

the set of strings of length n consisting of zeroes and ones.
b) On is a threshold graph.
c) Kn is a threshold graph.
d) A graph obtained from a threshold graph by adding an isolated vertex
is a threshold one.
e) Every threshold graph has a separating inequality with positive coeffi-
cients.
f) A graph obtained from a threshold graph by adding a dominating vertex
is a threshold one.
g) the graph at Fig. 3.1.2 is a threshold graph. (Find a separating in-
equality for it.)
1 ."...-------::. 2
5 6
Figure 3.1.2: To Exercise 3.1.37g
h) Every induced subgraph of a threshold graph is a threshold one.

i) Graphs 2K2 , P4 , C4 are not threshold.
j) Every threshold graph has either an isolated vertex or a dominating
vertex.
k) A graph is a threshold graph if and only if it may be obtained from KI
by a sequence of additions of isolated and dominating vertices.
3.1.38. Draw all pairwise nonisomorphic four-vertex threshold graphs.
3.1.39. a) Find cc(G) for the graphs from Fig. 1.2.4.
b) Prove than c( G) :::; cc( G) for all graphs G.
c) Find a graph with c(G) = cc(G); find a graph with c(G) < cc(G).
3.1.40. Column Intersection Graph: Let A be a binary (m x n)-matrix without
zero columns. Let 1 denote the column of ones of height m and let 0 denote
the row of zeroes of height n. The set
peA) = {x : Ax:::; 1, x ~ O},
i.e., the set of all real-valued solutions x = (Xl, X2, ... , x n ) with nonnegative
components Xi of the system of inequalities Ax :::; 1 will be called a polytope.
We define the column intersection graph G A for the matrix A to be the graph
with VG A = {I, 2, ... , n} in which two vertices i, j are adjacent if and only
if at least one row of A contains ones at positions i and j.
a) A subset U of vertices of a graph G A is independent if and only if the

characteristic vector Xu (see Exercise 3.1.37) belongs to the polytope
P(A).
b) For every graph G there exists a matrix A such that G ~ GA.
3.1.41. Let G l , G 2 , ... , Gp be a covering of an n-vertex graph G by complete sub-

graphs. A binary matrix A is called the clique covering matrix of the graph
G if Aij = 1 if and only if the vertex Vj E V G belongs to the clique Gi.
Reconstruct graphs from the following clique covering matrices:
1 100)
o1
(1 1 0
1101 o 00 )
1 01 0
1 1 1 0 ,b) A = 0 1 1 0 0 0 1 0 .
0011 000110 o 1
3.1.42. Prove that a graph G is isomorphic to a column intersection graph G A if

and only if the matrix A is the clique covering matrix for G.
3.1.43. An integer point of the polytope P(A) (see Exercise 3.1.40) is a vector
X E P(A) all whose components are integer. Let A, B be binary matrices
without zero columns.
a) Find conditions necessary and sufficient for the coincidence of the sets
of integer points of polytopes P(A), P(B).
b) Does the coincidence of the polytopes P(A), P(B) follow from the co-
incidence of their sets of integer points?
3.1.44. Let G be a graph. Consider all binary matrices A such that G = G A (see
Exercises 3.1.40 and 3.1.42). Prove that the minimal number of rows for
such matrices is cc( G).
3.1.45. Prove that a graph G is isomorphic to the line graph L(H) of a graph H if
and only if there exists a covering {G 1 , G 2 , .•• , G p } of G by complete graphs
satisfying the conditions:
a) every vertex of G is in at most two Gi;

b) every edge of G is in exactly one G i .
3.2 Coverings
We shall say that a vertex and an edge cover each other if they are incident to
each other. Thus, an edge e = ab covers a and b and each of these vertices covers
e. A set V' ~ VG is called a covering (vertex covering) of a graph G if each
edge from EG is incident to at least one vertex from V'. A covering of a graph is
called minimal if it does not contain a covering with less vertices and it is called
minimum cardinality vertex covering if it has minimal number of elements among
the coverings of G. This minimal number is called the covering number, or the
vertex covering number of the graph and is denoted by Po(G). The independence
number and the covering number are related with each other as follows:
ao(G) + Po(G) = IGI· (3.2.1)
An edge covering of a graph G is a subset of edges E' ~ EG that covers all

vertices of a graph G, i.e., such that each vertex from VG is incident to at least one
edge from E'. An edge covering of a graph is called minimal if it does not contain
a covering with less edges, and it is called a minimum cardinality edge covering if
it has minimal number of elements among the edge coverings of G. This minimal
number is called the edge covering number of the graph and is denoted by /31(G).
* * *
3.2.1. Define the values of /30 (G) and /31 (G) for graphs from Fig. 1.2.4.
3.2.2. Define the values of /30 (G) and /31 (G) for the following graphs: a) Pn ; b)
Cn ; c) Kn; d) KI,n; e) Km,n; f) the Petersen graph.
3.2.3. Give an example of a bipartite graph with /30 < min(m, n), where m and n
are the cardinalities of the parts.
3.2.4. Does there exist a connected graph whose minimal covering contains all
vertices?
3.2.5. Characterize connected graphs for which a minimal edge covering contains
all edges.
3.2.6. Prove that there exists a minimum cardinality vertex covering of a tree other
than K2 that contains all vertices adjacent to pendant vertices.
3.2.7. Give an algorithm for the construction of a minimum cardinality vertex

covering of a tree.
3.2.8. Find a graph whose minimal vertex covering is not a minimum cardinality
one.
3.2.9. Find a graph whose minimal edge covering is not a minimum cardinality
one.
3.2. Coverings 63
3.2.10. Prove that for a graph G without isolated vertices the following inequalities
hold:
IVGI/2 ~ rh(G) ~ IVGI- 1.
For any positive integer n, find a graph of order n for which one of the
inequalities turns into the equality.
3.2.11. a) Prove that every edge covering of a graph contains all pendant edges.
b) Design an algorithm for the construction of a minimum cardinality edge
covering of a tree.
3.2.12. Prove that a set U of vertices of a graph G is a (minimal, minimum cardi-

nality) covering if and only if VG\U is a (maximal, maximum cardinality)
independent set.
3.2.13. Prove the equality (3.2.1).
3.2.14. Prove that for a bipartite graph G with q edges the relation ao(G),8o(G) 2:
q holds, and the equality takes place only for complete bipartite graphs.
3.2.15. Prove that ,8o(G) ~ beG) for any graph G. Find a graph for which the
equality holds.
3.2.16. For the complete n-partite graph G prove
a) ,80 (G) = IVGI- N, where N is the maximum of the cardinalities of the
parts.
b) ,8o(G) = b( G).
3.2.17. Prove that an edge covering of a graph is minimal if and only if every
component of a graph induced by the edges of the covering is a star.
3.2.18. A vertex v of a graph G is called critical if ,8o(G - v) < ,8o(G).
Prove that a vertex v is critical if and only if it belongs to some minimum
cardinality vertex covering.
3.2.19. An edge e of a graph G is called critical if ,8o(G - e) < ,8o(G).
Prove that if an edge is critical then its ends are critical vertices.
3.2.20. Find a graph with critical vertices and without critical edges.
3.2.21. A graph is called vertex-critical if every its vertex is critical. A graph is
called edge-critical if every its edge is critical.
a) Give examples of vertex-critical and edge-critical graphs.
b) Prove that every edge-critical graph without isolated vertices is vertex-
critical.
c) Give an example of a vertex-critical graph which is not edge-critical.
3.2.22. Prove or disprove the statement: an edge of a graph is critical if and only
if it belongs to a minimum cardinality edge covering of the graph.
3.3 Dominating Sets

A set D of vertices of a graph G is called a dominating set if every vertex from
VG\D is adjacent to a vertex from D, i.e., every vertex of G is at distance at most
1 from some vertex of a dominating set. A dominating set is called minimal if it
does not contain a dominating set with less vertices; and it is called a minimum
cardinality dominating set if it has minimal number of elements among the dom-
inating sets of G. This minimal number is called the dominating number and is
denoted by ,(G).
A subset of vertices of a graph that is both independent and dominating is
called a kernel of the graph.
* * *
3.3.1. Evaluate ,(G) for the graphs shown at Fig. 1.2.4.
3.3.2. Evaluate ,(G) for the graphs a) Pn ; b) Cn ; c) Kn; d) K1,n; e) Km,n; f) the
Petersen graph.
3.3.3. Find a graph for which a minimal dominating set is not a minimum cardi-
nality dominating set.
3.3.4. Find a graph for which a minimal dominating set is not independent.
=
3.3.5. Find a graph for which ooo(G) ,(G) =
3 and every maximum cardinality
independent set is a minimum cardinality dominating set.
3.3.6. Prove that
a) an independent set of vertices of a graph is maximal if and only if it is

dominating.
b) every maximal independent vert.ex set is a kernel.
3.3.7. Prove that ooo(G) 2: ""{(G) for every graph G.
3.3.8. Prove the inequalities WGI/(6.(G) + 1) :S ""{(G) < WGI - 6.(G) for all
graphs G.
Give examples of graphs G 1 , G 2 such that
3.3.9. Prove that a dominating set D is minimal if and only if for every v E Done
of the following conditions holds:
(1) the graph has no edges vu with u ED;
(2) if the graph has an edge vu with u E D then it has an edge uw with
w (j. D, where u is the unique vertex from D adjacent to w.
3.3.10. Prove that a graph G without isolated vertices has a dominating set D
such that VG\D is also dominating.
3.3. Dominating Sets 65
3.3.11. Prove that if a graph G has no isolated vertices then ,(G) ~ IGI/2.
For every even n, find a graph of order n with ,(G) = IGI/2.
3.3.12. Prove that if D is a minimal dominating set of a graph G without isolated

vertices then VG\D is also a dominating set.
3.3.13. Prove that for every minimal dominating set of a graph without isolated
vertices there exists another minimal dominating set disjoint from the former
one.
3.3.14. Prove that every graph has a minimal dominating set that contains all
vertices adjacent to pendant vertices.
3.3.15. Give an algorithm for the construction of a minimum cardinality dominat-

ing set of a tree.
3.3.16. Prove that for every graph G there exists a graph H containing a minimum
cardinality dominating set D such that H(D) ~ G.
3.3.17. A kernel D of a graph G is called a minmax kernel if \D\ = ll'o(G) = ,(G).

Find a graph with a minmax kernel.
3.3.1S. The domatic number d(G) is the maximal number of dominating sets the
vertex set of G may be partitioned into.
Prove the inequality d( G) ~ a( G) + 1.
3.3.19. Let ,/(G) denote the minimal number of vertices in a dominating set of
vertices of a graph G, which is independent.
a) Prove that ,(G) ~ ,/(G) ~ ll'o(G) for any graph G.

b) Give examples of graphs for which the last relation holds as a strict
inequality and as an equality.
c) Find a graph G of the minimal order such that ,(G) < ,I (G).
d) Prove that ,I (G) - ,( G) may be arbitrarily large.
3.3.20. Compute the numbers ,/(G) for graphs shown at Fig. 1.2.4.
3.3.21. Compute the numbers ,/(G) for a) Pn ; b) en; c) Kn; d) K1,n; e) Km,n; f)
Petersen graph.
3.3.22. Prove that ,(G) ~ f3o(G). Find a graph with ,(G) < f3o(G).
3.3.23. Find a graph with (a) ,/(G) > f3o(G). (b) ,/(G) < f3o(G).
3.4 Matchings
A set of pairwise nonadjacent edges of a graph is called a matching, or an in-
dependent set of edges. A matching is called a perfect matching, or a i-factor if
every vertex of a graph is incident to an edge of the matching. Clearly, a perfect
matching is a minimum cardinality edge covering of a graph.
A graph is called i-factorizable if it may be represented as a union of edge-
disjoint I-factors.
Sometimes, when the set of edges of a graph is the (only) matching, the graph
itself is called a matching. We have already met such kind of graph in the exercises,
namely, 2K2.
A matching in a graph G is called maximal if it is not contained in a larger
matching; it is called a maximum cardinality matching if it contains the maximal
number of edges among the matchings of the graph. This maximal number is
called the matching number of the graph and is denoted by O:'l(G).
The independent edge sets of a graph G bijectively correspond to the indepen-
dent vertex sets of the line graph L(G), hence, O:'l(G) = O:'o(L(G)).
For a graph without isolated vertices, the matching number is related with the
edge covering number as follows:
(3.4.1 )
The number of edges in any matching of a graph does not exceed the number
of vertices in any vertex cover; in particular,
(3.4.2)
Let M be a matching in a graph G. A path of G whose edges are alternately in

M and in EG\M is called an alternating path with respect to M. By definition,
a path of length 1 is alternating. The vertices of G incident to the edges of M
are called saturated (by M), the remaining ones are called unsaturated (by M).
A path alternating with respect to M that connects two unsaturated vertices is
called an augmenting path with respect to M.
Theorem 3.4.1 (Berge Theorem) A matching M in a graph is of maximal

cardinality if and only if the graph has no augmenting paths with respect to M.
* * *
3.4.1. Evaluate 0:'1 (G) for the graphs shown at Fig. 1.2.4. Which of these graphs
have a perfect matching?
3.4.2. Evaluate O:'l(G) for the graphs a) Pn ; b) Cn ; c) Kn; d) Kl,n; e) Km,n; f)

Petersen graph.
3.4.3. Give an example when a maximal matching is not of maximal cardinality.

3.4. Matchings 67
3.4.4. Find a graph such that

a) f3o(G) = al(G);
b) f3o(G) > al(G).
3.4.5. Prove that the graphs Qn and K 2n are 1-factorizable and the Petersen graph
is not.
3.4.6. Prove the relation (3.4.1).
3.4.7. Prove that f3l (G) = al( G) if and only ifthe graph G has a perfect matching.
3.4.8. Prove that if a connected graph of even order has no induced subgraphs
isomorphic to K l ,3 then it has a perfect matching.
3.4.9. Let M l , M2 be matchings in a graph G and let G' be a graph induced by
the set of edges
Prove that the components of G' may be only of two types:

(1) cycles of even length with edges alternately in Ml and M 2 ;
(2) paths whose edges are alternately in Ml and M2 and whose terminal
vertices are unsaturated in one of the matchings.
3.4.10. Prove the Berge Theorem.
3.4.11. Let M, N be disjoint matchings in a graph G, IMI > INI. Prove that there
exist disjoint matchings M', N' such that IM'I = IMI- 1, IN'I = INI + 1,
M'UN'=MUN.
3.4.12. Prove that a tree T has a perfect matching if and only ifT - v has a single
component of odd order for any v E VT.
3.4.13. Prove that every tree has at most one perfect matching.
3.4.14. Prove that al(G) = min{8(G), LlGI/2J} if G is a complete multipartite
graph.
3.4.15. Prove that the following inequalities hold for a maximal matching M of a
graph G:
a) L8(G)/2J ~ IMI;
b) al(G) ~ 21MI.
3.4.16. We shall say that an edge weakly covers itself and the adjacent edges. The
minimal numbers of edges and independent edges necessary to weakly cover
EG are denoted by f311(G) and f3~l(G) respectively.
Evaluate f311 (G) and f3~1 (G) for the graphs shown at Fig. 1.2.4.
3.4.17. Evaluate f311(G) and f3~l(G) for the graphs a) Pn; b) en; c) Kn; d) Kl,n;
e) Km,n; f) Petersen graph.
3.4.18. Prove that every component of a graph induced by a minimal number of

edges weakly covering the edges of a graph G is a star K I ,p, p ~ 1.
3.4.19. Prove that (311(G) = (3~l(G) for every graph G.
3.5 Matchings in Bipartite Graphs

We shall consider a bipartite graph G = (X, Y-; E) with fixed parts X and Y. Let
M be a matching in this graph. Clearly, M may be interpreted as an injective
mapping from some subset A ~ X into Y : if an edge xy belongs to M then the
vertex y is the image of x. In such situation M will be called a matching from A
into Y or simply a matching into Y. Similarly, we may speak about a matching
from A into B, B ~ Y.
For A ~ V G we denote
NG(A) = N(A) = UN(v)\A.

vEA
Theorem 3.5.1 A bipartite graph G(X, Y; E) has a matching from X into Y if

and only if
IAI:::; IN(A)I (3.5.1)
for every A ~ X.
Corollary 3.5.2 A regular bipartite graph of nonzero degree is l-factorizable.
Define
6(A) = {IAI-IN(A)!, A i= 0, .
0, otherWIse.
The value
60 (X, Y; E) = max6(A)
A~X
is called the deficiency of the bipartite graph G = (X, Y; E).
Theorem 3.5.3 For a bipartite graph G with nonempty part X, Y,
al(G) = IXI- 60 (X, Y; E).

Clearly, for a graph G the number of vertices in a covering is not less than the
number of edges in any matching; in particular, (3o(G) ~ al(G).
Theorem 3.5.4 For a bipartite graph G, (3o(G) = al(G).
Corollary 3.5.5 For a bipartite graph G,
* * *
3.5. Matchings in Bipartite Graphs 69
3.5.1. Let G = (X, Y; E) be a bipartite graph and let Mi, i =

1,2, be perfect
matchings from Xi into Y;, Xi ~ X, Y; ~ Y, respectively. Prove that there
exists a matching M' ~ M1 U M2 such that the vertices of Xl and Y2 are
saturated.
3.5.2. Prove that a nonempty bipartite graph has a matching such that all vertices
of maximal degree are saturated.
3.5.3. Prove that the edge set of a bipartite graph G may be partitioned into ~(G)
matchings.
3.5.4. Let G = (X,Y;E) be a bipartite graph and let t be a positive integer,

t ~ IXI. Prove that G has a matching of cardinality t if and only if
ING(A)I ~ IAI + t - IXI
for every A ~ X.
3.5.5. Prove that

IXI- oo(X, Y; E) = IYI- oo(Y, X; E)
for every bipartite graph G = (X, Y; E).
3.5.6. Prove that G is bipartite if and only if ao(H) = f3l(H) for every subgraph
H without isolated vertices.
3.5.7. A line of a matrix is either its row or its column. Two elements of a matrix
are called independent if they do not share a line.
Prove the Konig theorem: the maximal number of pairwise independent ones
of a binary matrix is equal to the minimal number of its lines containing all
ones of this matrix.
3.5.8. Prove that a binary matrix in which every line contains exactly k 'l's may
be represented as a sum of k binary matrices in which every line contains
exactly one ' 1' .
=
3.5.9. Let P {Pl, ... , Pn} be a finite non empty set and let 5 =
(51' ... ' 5 m ) be
an m-element collection of its subsets. A subset T ~ P is called a transver-
sal (or a system of different representatives) of 5 if there exists a bijective
mapping 'P: T -+ {1, 2, ... , m} such that t E 5",(t) for every t E T. This
means that we may index the elements of T in such a way that ti E 5i,
i = 1,2, ... , m.
Prove the following Hall theorem: 5 has a transversal if and only if for any
k E {I, ... , m} the union
of any k subsets 5i contains at least k different elements.

3.5.10. A subset T ~ P (see Exercise 3.5.9) is called a partial transversal of a

collection S if there exists an injective mapping 'P : T -+ {1, 2, ... , m} such
that t E S<p(t) for every t E T.
Prove that a collection S has a partial transversal of cardinality p ::; m if
and only if for every 1 ::; k ::; m the union of any k subsets Si contains at
least k + p - m different elements.
3.5.11. A collection (P1 , P2 , •.• , Pm) of subsets of P (see Exercise 3.5.9) is called
a
(k 1 , k 2 , .•. , km)-transversal of a collection S, where (k 1 , k 2 , ... , k m ) is a se-
quence of positive integers, if
(1) Pi ~ S;, i= 1,2, ... ,m;

(2) IPil = ki, i = 1,2, .. . ,m;
(3) Pi n Pj = 0, i f. j.
Prove that a collection S = (Sl,"" Sm) has a (k 1 , k2"'" km)-transversal if
and only if
IUS;! ~ Lk;
iEI iEI
for every I ~ {1, 2, ... , m}.

Chapter 4
Connectivity
4.1 Biconnected Graphs and Biconnected Com-

ponents
A vertex v of a graph G is called a cutpoint if G - v has more components than G.
In particular, if G is connected and v is a cutpoint then G-v is disconnected. Sim-
ilarly, an edge is called a bridge if its deletion increases the number of components
in the graph.
A connected graph is called biconnected (2-connected) if it is not K2 and has
no cutpoints.
Theorem 4.1.1 For a connected graph G, IGI > 2, the following statements are
equivalent:
1) G is biconnected;
2) any two vertices of G are on a common simple cycle;
3) any vertex and any edge of G are on a common simple cycle;
4) any two edges of G are on a common simple cycle;

5) for any two vertices a, b and any edge e, the graph G has a simple (a, b)-path
containing e;
6) for any three vertices a, b, c, the graph G has a simple (a, b)-path containing
c.
A maximal under inclusion connected subgraph without cutpoints is called

block. A biconnected graph G is called critically biconnected if G - e is not bicon-
nected for every edge e.
* * *
71
72 4. Connectivity
4.1.1. Let G be a biconnected graph. Prove the following statements without any
usage of the Theorem 4.1.1:
a) 6(G) ~ 2;
b) if P is a simple path with both ends belonging to G then G U P is also
biconnected;
c) for any three different vertices a, b, c, the graph G has an (a, b)-path not
containing c.
4.1.2. Prove that an edge of a graph is not a bridge if and only if it belongs to a
cycle.
4.1.3. Let cp(G) denote the number of cutpoints of a graph G. Prove the following
statements:
a) if e is an edge that belongs to a cycle of G then cp(G) ~ cp(G - e);

b) cp(G) ~ cp(T) for any skeleton T of G;
c) cp(G) ~ IGI- 2;
d) if cp(G) = IGI- 2 then G is a path graph.
4.1.4. Prove that if 6( G) ~ 3 and G =I K2 then G has a cutpoint if and only if G
has a bridge. Show that this statement fails for 6( G) ~ 4.
4.1.5. Prove that every biconnected graph G has at least n = IGI edges; moreover,
the unique biconnected graph with n edges is C n .
4.1.6. State necessary and sufficient conditions for a line graph L(G) to have a
cutpoint.
4.1. 7. Let G be a connected graph with 6( G) ~ 3. Prove that L( G) has no bridges.
4.1.8. Prove that d(G) ~ LlGI/2J for a biconnected graph G, where d(G) is the
diameter of G.
4.1.9. Construct biconnected n-vertex graphs G with d(G) = 2, IEGI = 2n - 5 for

all n ~ 5.
4.1.10. Let G be a biconnected graph. For VI ~ VG and v E VG\VI prove that

G has a simple (a, b)-path H such that V H n VI = {a, b} and H contains v.
4.1.11. Let B = {Bd and C = {Cj} be the block set and the cutpoint set of a
graph G respectively. Let bc( G) denote the graph with the vertex set B U C
and the edge set {BiCj : Bi E B, Cj E C, Cj E V Bd.
a) For a graph G shown at Fig. 4.1.1 construct the graph bc(G).
b) Prove that if G is connected then bc( G) is a tree.
4.1. Biconnected Graphs and Biconnected Components 73
4.1.12. Let G be a connected graph and let b( v) denote the number of blocks of
G containing a vertex v. Prove that the number of blocks of G is
1+ L (b(v) -1).
vEVG
4.1.13. Let G be a biconnected graph with 6(G) 2: 3. Prove that there exists a
vertex v E VG such that G - v is also biconnected.
4.1.14. Prove that if every two blocks of a graph have a common vertex then all
blocks have a common vertex.
4.1.15. Let x be vertex of graph G and let an edge e be incident to x. Prove that
if x is not a cutpoint of G then x is not a cutpoint of G - e.
4.1.16. Let G be a biconnected graph. Prove that if G - e has a cutpoint, then
the ends of e are in different blocks of G - e.
4.1.17. Prove that for every simple cycle of a graph there exists a block of the
graph that contains this cycle.
4.1.18. Prove that any two blocks of a graph have at most one common vertex.
4.1.19. Prove that every edge of a graph belongs to a single block.
4.1.20. Prove that if a block B and a simple cycle C of a graph are such that
WB n VCI2: 2 then VC ~ VB.
4.1.21. An edge ab is called a diagonal of a cycle C if a, b E VC and ab f/. EC.
Prove that an edge e of a biconnected graph G is a diagonal of a simple cycle
in G if and only if G - e is biconnected.
74 4. Connectivity
4.1.22. Prove that a critically biconnected graph with more than 3 vertices has no
triangles.
4.1.23. Let G be a critically biconnected graph. Prove that every block of G - e,

e E EG, not isomorphic to K2 is also a critically biconnected graph.
4.1.24. Prove that a critically biconnected graph has a vertex of degree 2.
4.1.25. Prove that IEGI :::; 21VGI-4 for a critically biconnected graph G, IGI ;::: 4.
4.1.26. Is it true that for a critically biconnected graph every vertex of degree
greater than 2 is adjacent to a vertex of degree 2?
4.1.27. Let v be a vertex of a connected graph G such that G(N(v» is connected.

Prove that G - v is connected.
4.1.28. Let v be a cutpoint of G. Prove that v is not a cutpoint of G.
4.1.29. A block is called pendant if it contains only one cutpoint of the graph.
Prove that every graph with cutpoints has at least two pendant blocks.
4.1.30. Let G be a biconnected graph such that G - a - b is disconnected for every

two its nonadjacent vertices a, b. Prove that G is a simple cycle.
4.1.31. Let {BI' B2, ... , Bd be the block set of a graph G. The block graph B(G)
for the graph G is the graph with the vertex set {VI, V2, ... , Vt} such that
ViVj E EB(G) if and only if Bi n Bj i= 0. Prove the following statements:
a) If G is connected and a block Bk is not pendant then the graph U:=l,i;Ck B;
is disconnected.
b) If B( G) is biconnected then it is a complete graph.

4.2. k-connectivity 75
4.2 k-connectivity
The vertex connectivity number K(G) of a graph G is the minimal number of
vertices whose removal makes G disconnected or single-vertex.
The edge connectivity number A( G) of a nontrivial graph G is the minimal
number of edges whose removal makes G disconnected. The edge connectivity of
a single-vertex graph is defined to be zero.
A graph is called k-connected if K( G) ~ k; it is called k-edge connected if
A(G) 2: k.
A maximal k-connected subgraph of G is called its k-component.
Theorem 4.2.1 Any two different k-components of a graph have at most k - 1

common vertices.
Theorem 4.2.2 For any graph G,
K(G) ~ A(G) ~ 6(G).
* * *
4.2.1. Find K(G), A(G) for the graph G shown at Fig. 4.1.2.
4.2.2. Find a 3-component in the graph G shown at Fig. 4.2.1.
4.2.3. How many edges must be added to the graph shown at Fig. 4.2.2 to make
it triconnected?
76 4. Connectivity
4.2.4. Prove that IECI ~ k1CI/2 for a k-edge connected graph.
4.2.5. Prove that if 6(C) > (ICI + k - 2)/2 then C is k-connected.
4.2.6. Find a graph C with /l,(C) < k and O(C) = L(ICI + k - 3)/2j.
4.2.7. Prove that if 6(C) ~ ICI- 2 then /l,(C) = 6(C).
4.2.8. For every n = 4,5, ... , construct an n-vertex graph with 6( C) = ICI - 3
and /l,(C) < 6(C).
4.2.9. Prove that if an edge e is a bridge of a connected graph C then C - e has

exactly two components.
4.2.10. Let S be a subset of edges of a connected graph C of size lSI = A(C).

Prove that if C - S is disconnected then it has exactly two components.
4.2.11. Prove that if O(C) ~ ICI/2 then A(C) = o(C).

4.2.12. Find an n-vertex connected graph C with A(C) < 6(C) = LlCI/2j - 1.
4.2.13. Prove that if C is not a complete graph then /l,(C) ~ 26(C) -ici + 2.
4.2.14. Prove that A(G) = /l,(G) for a cubic graph C.
4.2.15. Prove that IECI ~ /l,(C)ICI/2.

4.2.16. Prove that if C is an r-regular graph with a cutpoint then A(C) ::; Lr/2j.
4.2.17. Prove that if C is of diameter less than 3 then A(C) = 6(C).
4.2.18. Prove that /l,(C) = 1 + millvEVG /l,(C - v) for a connected graph C.
4.2.19. Prove that there are no triconnected graphs with 7 edges.
4.2.20. How many are there triconnected graphs with 8 edges?

4.3. Cycles a.nd Cuts 77
4.2.21. For which k there exists a k-connected graph with (k 2 + k + 2)/2 edges?
4.2.22. Is it true that in a triconnected graph any three edges not incident to the
same vertex belong to a common simple cycle?
4.2.23. Prove that in a triconnected graph any three vertices belong to a common
simple cycle.
4.2.24. Is it true that in a triconnected graph any four vertices belong to a common
simple cycle?
4.2.25. Let G be a bipartite graph. Prove that if 8( G) ~ IGI/4 + 1 then >.( G) =

8(G).
4.2.26. Prove that if G is an r-regular graph with IGI = 2r + 1 then >.( G) = r.
4.2.27. A connected graph is called rigid if IXI ~ k(G - X) for any X ~ VG,
X f. 0,where k(H) is the number of components of the graph H. Is it true
that in a rigid graph any three vertices belong to a common simple cycle?
4.2.28. For each n > 2 find a rigid graph (see Exercise 4.2.27) which is not tricon-
nected.
4.2.29. For each n > 1 find an n-connected graph which is not rigid (see Exercise
4.2.27).
4.2.30. A graph is called cyclically k-connected if there is no S ~ EG, lSI < k,

such that G - S is disconnected and every component of G - S contains a
cycle. Find a cyclically 4-connected graph which is not 4-connected.
4.2.31. Find a graph in which every 3 vertices are on a common simple cycle but
there are 4 vertices that are not on a common simple cycle.
4.3 Cycles and Cuts

The length of a shortest cycle of a graph G is called the girth of G and is denoted
by g(G).
A subset U ~ EG is called a separator set of G if the number of components of
G - U is greater than that of G. A minimal under inclusion separator set is called
a cut. The notion of cut is also applicable to pseudographs.
* * *
4.3.1. Find a graph G with IGI = 7, 6(G) ~ 3, g(G) = 4.
4.3.2. Prove that if 8(G) ~ 3 and IGI ~ 8 then g(G) :::; IGI/2.
4.3.3. Prove that if G is biconnected and has no cycles of even length then G is a
cycle of odd length.
78 4. Connectivity
4.3.4. Let h( G) denote the length of the longest simple cycle in G. Prove that if
8(G) 2: 2 then
a) h(G) 2: 8(G) + 1;
b) h(G) 2: 2IEGI/(IGI- 1).
4.3.5. Prove that if X is a cut of a connected graph G then G - X has exactly
two components.
4.3.6. Suppose that every vertex of a simple cycle C is colored in one of two colors.
Prove that the number of edges of the cycle with ends colored differently is
even.
4.3.7. Prove that any simple cycle and any cut in a graph have even number of
common edges.
4.3.8. Prove that for any cut X of a graph G there exists a partition of the vertex
set V G = VI U V2 such that X = {ab: a E VI, b E V2 , ab E EG}.
4.3.9. How many cuts has a graph shown at Fig. 4.3.1? Find them.
4.3.10. Prove that a cut of a graph induces a bipartite subgraph.

4.3.11. Prove that every cut of a graph contains at least one edge of every skeleton
of the graph.
4.3.12. For a vertex v of a graph G, let X be a set of edges incident to v belonging
to the same block of G. Prove that X is a cut of G.
4.3.13. Prove that every edge of a graph is in some cut of the graph.
4.3.14. Is it true that every pair of edges of a graph is in some cut of the graph?
4.3.15. Prove that every separator set of a graph has a subset which is a separator
set for some block of the graph.
4.3.16. Prove that every cut of a graph belongs to some block of the graph.
4.3.17. Let VG = VI U V2 be a partition of the vertex set of a connected graph G.

Prove that the set of edges (VI, V2 ) = {ab: a E VI, b E V2, ab E EG} is a
cut of G if and only if G - (VI, V2 ) has exactly two components.
4.3.18. Let C be a simple cycle of a graph G and let el, e2 be any edges of the
cycle. Prove that G has a cut S such that S n EC = {el' e2}.
4.3. Cycles and Cuts 79
G H
4.3.19. A pseudograph G* is called abstract dual for a pseudograph G if there

exists a bijection EG --+ EG* such that the subset of edges of EG consti-
tutes a simple cycle if and only if the corresponding subset of edges of EG*
constitutes a cut. Prove the following statements:
a) For graphs G and H shown at Fig. 4.3.2, H :::: G*.

b) Graphs J{5 and J{3,3 have no abstract duals.
Chapter 5
Matroids
5.1 Independence Systems

Let E be a finite nonempty set and let I be a set of its subsets satisfying the
following two conditions, called independence axzoms:
1.1. 0 E I.
1.2. If X E I and Y <;;; X then Y E I.
The pair 5 = (E, I) is called the independence system with support E and set of
independent sets I. The number lEI is called the order of the system.
Maximal under inclusion elements of I are called bases of the independence
system 5. A subset X <;;; E is called dependent if it is not independent, i.e., if
X ~ I. Minimal under inclusion dependent sets are called cycles of 5.
The set of independent sets of an independence system 5, the set of its bases
and the set of its cycles will be denoted 7(5), 8(5), and C(5), respectively. Each
of these sets together with the support E uniquely specifies the remaining two
ones and hence the system 5.
It is possible that C(5) = 0. In this case the system 5 is called free. Each
subset of a free independence system is independent. At the other extremity,
when 8(5) = {0}, the independence system is called trivial.
For any independence system 5 the set of bases 8(5) is an antichain with
respect to set inclusion, i.e., if B 1, B2 E 8(5) and B1 <;;; B2 then B1 = B 2 . The set
C(5) is also an antichain.
Let 5k = (Ek' h), k = 1,2, be two independence systems and let t.p : E1 -+ E2
be a bijection. Then if t.p(X) E h if and only if X E It for every X <;;; E 1 , then the
mapping t.p is called isomorphism of the independence systems 5 1 ,52 , and these
systems are called isomorphic (51 :::: 52).
If 5 k = (E, h), k = 1,2, are two independence systems with the same support
and I = II n h, then the pair (E, I) is also an independence system; it is called
the intersection of systems 5 1,52 and is denoted by 51 n 52.
* * *
81
82 5. Matroids
5.1.1. Let E be {1,2,3,4}. Check whether the pair S = (E,I) is an independence

system for
a) 1= {{1},{2},{3},{1,2},{1,2,3,4}};
b) 1= {{1},{2},{3},{1,2},{1,3}};
c) 1= {{1},{2},{3},{1,2},{1,3},0}.
If yes then find 8(S) and C(S).
5.1.2. Let E be a nonempty finite set and let 8 be a nonempty set of its subsets
that is an antichain. Prove that there exists an independence system S with
support E such that
a) 8(S) = 8;
b) C(S) = 8;
= =
5.1.3. Let Sk (Ek, h), k 1,2, be two independence systems and let r.p : El -+
E2 be their isomorphism. Prove that for any X ~ E 1 ,
a) r.p(X) E 8(S2) ¢=> X E 8(Sd;

b) r.p(X) E C(S2) ¢=> X E C(Sd.
= =
5.1.4. Let Sk (Ek, Ik), k 1,2, be two independence systems and let r.p : El -+
E2 be a bijection. Prove that if any of the claims of the previous exercise
holds then r.p is an isomorphism of the independence systems.
5.1.5. Prove that the intersection of independence systems is also an independence

system. How does the set of cycles of the intersection look like?
5.1.6. Let G be a graph. Let IG denote the set of independent subsets of vertices
of G plus the empty set.
a) Prove that the pair (VG,IG) is an independence system. It is called

graphical independence system.
b) Which subsets of vertices of G are bases of the graphical independence
system (VG,IG)?
c) Describe the cycles of the graphical independence system.
5.1.7. Prove that an independence system S is isomorphic to a graphical indepen-

dence system if and only if one of the following conditions holds:
a) S is a free independence system;

b) the number of elements in every cycle of S is 2.
5.1.8. Find the number of pairwise nonisomorphic independence systems of or-

der 3. How many of them are graphical independence systems? Draw the
corresponding graphs.
5.2. Matroids 83
5.1.9. Let f be a monotonic Boolean function in n variables, f ;j. 1. For E =

{1,2, ... ,n}, let I denote the set of subsets U
~ E whose characteristic
vectors (see Exercise 3.1.37) are zeroes of f.
a) Prove that S = (E, 1) is an independence system. We shall say that it

is specified by the function f.
b) Characterize the cycles of S.
c) Prove that every independence system is isomorphic to the indepen-
dence system specified by some monotonic Boolean function.
d) Which monotonic Boolean functions specify graphical independence
systems?
5.1.10. Find the cycles and the bases of the independence system specified by the
Boolean function:
a) f(Xl, X2, .. " xn) = Xl V X2 V ... V Xn;

b) f(X1, X2, X3, X4) = Xl V X2X3 V XIX2 V X4·
5.1.11. For E = {1, 2, ... , n}, let I denote the set of subsets U ~ E whose char-
acteristic vectors (see Exercise 3.1.37) satisfy the system of inequalities
allx1 + a12X2 + ... + alnXn < bl ,

a2l X l + a22 X 2 + ... + a2n X n < b2 ,
(5.1.1)
amlXl + a m 2 X 2 + ... + amnXn < bm ,
where all aij, bi are nonnegative reals. Prove the following statements:
a) The pair (E, 1) is an independence system. It is said that it is specified
by the system of inequalities (5.1.1).
b) Prove that every independence system is isomorphic to the indepen-
dence system specified by some system of inequalities.
c) Which system of inequalities specify graphical independence systems?
5.1.12. For a non empty graph G let IG denote the set of matchings of G plus the
empty set.
a) Prove that the pair (EG, IG) is an independence system.

b) Find the set of cycles of this system.
c) Characterize the bases of the system.
5.2 Matroids
A matroid with support E is a pair (E, I), where E is a nonempty finite set and
I is a set of subsets of E satisfying the following three conditions (independence
axioms of matroid):
84 5. Matroids
1.1. 0 E I.
1.2. If X E I and Y ~ X then Y E I.
1.3. If X, Y E I and IXI < IYI then there exists y E Y\X such that Xu {y} E I.
Thus, a matroid is an independence system satisfying the additional axiom 1.3.

Therefore bases and cycles of such system (matroid) have additional properties.
Theorem 5.2.1 The set B(M) of bases of a matroid M satisfies the following two
conditions:
B.1. No base is contained in another base.
B.2. If B l , B2 are bases then for any element b E Bl \B 2 there exists an element
c E B2 \B l such that (Bl \ {b}) U {c} E B(M).
Theorem 5.2.2 Let E be a non empty finite set and let B be a nonempty set of
its subsets satisfying both axioms of bases. Then if I denotes the set of all subsets
of elements of B (I = {X ~ B : B E B}), then the pair (E, I) is the matroid with
the support E, the set of independent sets I and the set of bases B.
All bases of a matroid are of the same cardinality; this number is called the
rank of the matroid and is denoted by p(M).
Theorem 5.2.3 The set C(M) of cycles of a matroid satisfies the following two
conditions:
C.l. No cycle is contained in another cycle.
C.2. If C l , C 2 are different cycles then for any element c E C l n C2 the set (Cl u
C 2 )\{c} contains a cycle.
The conditions of the above theorem are called the axioms of cycles.
A single-element cycle is called a loop.
Theorem 5.2.4 Let E be a nonempty finite set and let C be a set of its subsets
satisfying both axioms of cycles. Let I be the set of subsets X ~ E such that no
subset of X is in C. Then the pair (E,1) is the matroid with support E, set of
independent sets I, and set of cycles C.
Proposition 5.2.5 For any base B of a matroid and for any element e fI. B of
the matroid the set B U {e} contains a unique cycle of the matroid.
The rank function p of a matroid M
p(A) = max{IXI : X ~ A, X E I(M)}
assigns a number peA) to each subset A ~ E, which is called the rank of A.

In what follows, (E, B), (E, C), and (E, p) will denote, respectively, the matroid
with the set of bases B, with the set of cycles C and with the rank function p.
5.2. Matroids 85
Proposition 5.2.6 A matroid (E, p) has k pairwise nonintersecting bases if and
only if kp(A) + IAI ~ kp(E) for any A ~ E.
Let M = =
(E, B) be a matroid. For B E B, the set B E\B is called a cobase
of the matroid M. The set of all cobases of M will be denoted by B*(M).
Proposition 5.2.7 The pair (E, B*(M)) is a matroid with the set of bases B* (M).
The matroid B* (M) is called dual to M and is denoted by M* . An independent

(dependent) set of elements of M* is called coindependent (codependent) set of
M. A cycle of M* is called a cocycle of M. The rank function of M* is called the
corank function of M and is denoted by p* (M). For every statement about matroid
in terms of cycles, bases, etc., there exists the dual one about the dual matroid in
terms of cocycles, cobases, etc. If any of the two mutually dual statements is valid
for all matroids then the second one is also valid.
Proposition 5.2.8 For every nonempty independent set X of a matroid there

exists a cocycle Y such that IX n YI = 1.
Proposition 5.2.9 If matroids M 1 , M2 are isomorphic then Mi, M; are isomor-

phic too.
Consider some examples of matroids. Let L be a linear space over a field F. If

E ~ L is a finite nonempty set and I is the set of all linearly independent over F
sets of vectors in L plus the empty set, then (E, I) is a matroid. It is called the
vector matroid induced by the vector set E. A matroid isomorphic to the vector
one is also called vector matroid.
Let G be a nonempty graph. We may define the matroid M(G) with the
support E = EG whose bases are the sets of edges of the skeletons of G. This
matroid is called the cyclic matroid, or the matroid of cycles of the graph G. A
matroid isomorphic to cyclic matroid is called graphic matroid.
Notice a terminological whim: a graphic matroid is not necessarily a graphical
independence system.
The dual matroid (M(G))* = M*(G) is called the matroid of cuts of the graph
G. A matroid isomorphic to the matroid of cuts of a graph is called cographic.
Let E be a finite nonempty set and let S = {Ei : i = 1,2, ... , k, Ei =J 0} be
a covering of E, i.e., E = El U E2 U ... U Ek. Let I denote the set of the partial
transversals of the collection S (see Exercise 3.5.9) plus the empty set. Then the
pair (E, I) is a matroid. A matroid isomorphic to the one described above is called
a transversal matroid.
Let Mk = (Ek, h), k = 1,2, be two matroids specified by the sets of indepen-
dent sets h, h If E = El U E2 and I = {X U Y : X E h, Y E I 2} then the pair
(E, I) is a matroid with the system of independent sets I. It is called the union of
the matroids M 1 , M2 and is denoted by Ml U M 2.
Consider the following problem of discrete optimization. Let E be a nonempty
finite set and let w : E -+ R + be a function providing every e E E with a
nonnegative real weight w( e). For X ~ E, the weight of X is defined as the sum of
weights of all x E X. Let further I be a set of subsets of E. It is necessary to find
86 5. Matroids
an element in I of maximal weight. Let us describe a simple algorithm for solving

such problems.
Greedy Algorithm
Step 1: Find an element el E E such that
{ed E I, w(eI) = max{w(e): {e} E I}.
Step k : Find ek E E\{el, ... , ek-d such that {el, ... , ek-l, ed E I and
w(ek) = max{w(e) : e =I ei, i = 1, ... , k - 1, {el, ... , ek-l, e} E I}.

If there is no such ek then terminate.
The greedy algorithm always produces an element of I maximal under inclusion

but not necessarily of maximal weight. But if I is the set of independent sets
of a matroid with support E then the greedy algorithm always constructs an
independent set of maximal weight (this set turns out to be a base).
One may easily modify the greedy algorithm to make it construct a minimum
weight independent set of a matroid.
* * *
5.2.1. Prove that an independence system whose all cycles are single-element is a
matroid.
5.2.2. Prove that an independence system of order 2 is a matroid, but not all
independence systems of order 3 are matroids.
5.2.3. List all pairwise nonisomorphic matroids of order 3. For each of them,
find its rank, all bases, and all cycles. How many of them are graphical
independence systems? And graphic matroids?
5.2.4. Let M = (E, I) be a matroid, 0 =I A ~ E,
h = {X E I: X n A = 0}, h = {X n A: X E I}.
Prove that the pairs (E, h), k = 1,2, are the matroids with the sets of
independent subsets h.
5.2.5. Prove that for a matroid M of order n
a) the number of bases is at most G~(M);
b) the number of cycles is at most Gh(M)+!.
5.2.6. Let (E, G) be a matroid with the set of cycles C. For a positive integer k
let G' denote the set obtained from C by the deletion of all cycles with more
than k elements. Prove the following statements.
a) If k :::; 2 then (E, G') is the matroid with the cycle set C'.
b) If k = 3 then the statement from a) is not necessarily valid.
5.2. Matroids 87
5.2.7. a) Prove that the vector matroid is matroid indeed. What are its bases and
cycles? What is its rank?
b) Find all bases, cycles, cobases, cocycles, rank and corank of the vector
matroid induced by the columns of the matrix
1231]
A= [ 0 1 1 2 .
1 101
5.2.8. Let E be a finite nonempty set and let Bk be the set of k-element subsets
of E. Prove that (E, Bk) is the matroid with the set of bases Bk. It is called
the uniform matroid (of rank k).
5.2.9. Prove that the union of matroids is matroid.
5.2.10. Let E be a finite nonempty set and let
{Ei, i = 1, ... , k} (5.2.1)
be a covering of E, i.e., E = U~=l Ei. Let further m = (ml, ... , mk) be an

integer-valued vector with 0 :S mi :S !Eil, i = 1, ... , k. Set
I={X~E: !XnEi!:Smi, i=1, ... ,k}.
a) The pair 5 = (E, 1) is the independence system with the support E and
the set of independent sets I.
b) Every independence system is isomorphic to some independence system
5 defined in a).
c) If the covering (5.2.1) is a partition, i.e., Ei n Ej = 0 for all i,j, i 1= j,
then 5 = (E,1) is a matroid. It is called the partition matroid specified by
the vector m.
d) Every independence system of order 2 is a partition matroid.
e) Every matroid of order 3 is a partition matroid, but this is not so for
matroids of larger orders.
a) The transversal matroids and only such ones may be written as unions of
uniform matroids of rank 1 (see Exercise 5.2.8).
b) The uniform matroids of rank k and only such ones may be written as
the union of k copies of a uniform matroid of rank 1.
c) The partition matroids (see Exercise 5.2.10) and only such ones may be
written as unions uniform matroids with pairwise disjoint supports.
d) A uniform matroid is a partition matroid, but there are partition matroids
that are not uniform ones.
e) A partition matroid is a transversal matroid, but there are transversal
matroids that are not partition ones.
88 5. Matroids
5.2.12. Prove that the number of pairwise nonisomorphic

a) matroids of order n is at most 22n;
b) transversal matroids of order n is at most 2n 2.
5.2.13. Prove that every matroid dual to a uniform one is uniform, dual to a
partition one is a partition matroid, but there are transversal matroids whose
duals are not transversal.
5.2.14. A graph is called an M -graph if every its component is a complete graph.
Prove the equivalence of the following statements.
a) A graph G is an M-graph.
b) The graphical independence system (VG, IG) (see the Exercise 5.1.6) is
a matroid.
c) The path graph P3 is not an induced subgraph of G.
a) Every independence system is an intersection of matroids considered as
independence systems.
b) The minimal number of matroids whose intersection is the graphical in-
dependence system (VG, IG) (see the Exercise 5.1.6) is equal to the minimal
number of M-graphs (see the Exercise 5.2.14) whose union is the graph G.
5.2.16. Let (EG, IG) be the independence system specified in the Exercise 5.1.12.
Prove the equivalence of the following statements.
a) The independence system (EG, IG) is the matroid with the set of inde-
pendent sets IG.
b) Every nontrivial component of G is either the triangle K3 or the star
Kl,n, n ;::: 1.
c) P4 is not a subgraph of G.
5.2.17. Let G be a nonempty graph.
a) Prove that the cyclic matroid M(G) is matroid indeed. What are its bases
and cycles, cobases and cocycles? What are its rank and corank?
b) Find the bases, cycles, cobases and cocycles, rank and corank for M (G),
where G is shown at Fig. 5.2.1.
5.2.18. Does the isomorphism of the cyclic matroids M(G)and M(H) imply the
isomorphism of the graphs G and H?
5.2.19. Characterize the graphs whose cyclic matroids are uniform (the Exercise
5.2.8).
5.2.20. Prove that a non empty graph G has s skeletons with pairwise disjoint
edge sets if and only if every its spanning subgraph H satisfies the inequality
m(G)-m(H) ;::: s(k(H)-k(G», where k denotes the number of components,
m denotes the number of edges.
5.2. Matroids 89

a) If D1 , D2 are two different cuts of a graph G and e E Dl n D2 then the
set (Dl U D 2 ) - e is a separator set.
b) For every nonempty acyclic subgraph F of a graph G there exists a cut
in G with exactly one edge in common with F.
5.2.22. Prove that if a graph G* is an abstract dual for a graph G (the Exercise
4.3.19) then G is an abstract dual for G*.
5.2.23. Prove that the matroids of cycles M(Ks) and M(K3,3) are not cographic
and the matroids of cuts M*(Ks) and M*(K3,3) are not graphic.
5.2.24. For the matrix

1 0 2 0 3
0 3 1 0 1
2 2 3 3 6
1 2 0 2 1
0 0 1 1 0
1 2 1 3 1
find
a) a linearly independent set of rows with maximal sum of elements;
b) a maximal linearly independent set of rows with minimal sum of elements.
5.2.25. A set of tasks must be executed on a single computer. The jobs have their
deadlines. All tasks have the same execution time.
a) Prove that the set of all subsets of tasks that can be performed within
the deadlines constitute the set of independent sets of some matroid.
b) Assume that to every task that fails to meet its deadline a penalty is
assigned, all penalties being of the same value. Which order of the execution
of the tasks minimizes the total penalty?
90 5. Matroids
5.3 Binary Matroids

Let M be a matroid with support E = {el,"" em} and let F n be the linear
space of columns of height n over a field F. A mapping 1(/: E -+ F n is called
a representation of the matroid M over the field F if it satisfies the following
condition: for any subset
the system of vectors I(/(ei,), l(/(ei 2 ), ••• , I(/(ei k ) is linearly independent over F if
and only if X E I(M). If 1(/ is a representation then the matrix
whose columns 1(/( ei) are the images of the elements of the support E under the
representation 1(/ is also called a representation of the matroid M. A matroid is
called representable over the field F if there exists a representation of the matroid
over F.
Theorem 5.3.1 If a matroid M is representable over a field F then the dual

matroid M* is also representable over F.
In what follows, Z2 is the two-element field. A matroid representable over Z2

is called binary matroid.
Let M be a binary matroid and let L be the set whose elements are the empty
set, all cycles of M and the unions of pairwise disjoint cycles. Let us define the
operations of addition modulo 2 and multiplication by 0 and 1 on L :
X +Y = (X U Y)\(X n Y), 0 . X = 0, 1· X = X.
Let us select a base B of M. For any e E E\B, exactly one cycle of M is a subset
of B U {e }. Let us denote this cycle by C e .
Theorem 5.3.2 The set L supplied with the above operations is a linear space
over Z2 of dimension p*(M). The set of cycles
(5.3.1)
is its basis.
The space L is called the cycle space and the basis (5.3.1) is called the cycle
basis (with respect to base B) of the matroid M.
The cocycle space L* is the object dual to L. The set of cocycles {C; : e E B}
is its basis, where B is some fixed base of M and C; is the unique co cycle which
is a subset of (E\B) U {e}. This basis is called the cocycle basis (with respect to
base B) of the matroid M.
The cycle and cocycle bases with respect to the same base are easily expressed
in terms of each other; namely, the following theorem is valid.
5.3. Binary Matroids 91
Theorem 5.3.3 If fEB then
where {Cep C e2 , ••• , C ek } is the set of cycles from the basis (5.3.1) containing f.
The cyclic matroid M(G) for every non empty graph G is binary, since the
following theorem is valid.
Theorem 5.3.4 The incidence matrix I(G) is a representation of the cyclic ma-
troid M(G) over Z2.
Corollary 5.3.5 The cyclic matroid M(G) is isomorphic to the vector matroid
over Z2 induced by the columns of the incidence matrix I(G).
The cycle basis of M (G) is called the cycle basis of G and the co cycle basis
M (G) is called the cut basis of G.
* * *
5.3.1. Prove the following two statements.
a) A matroid is a vector matroid if and only if there exists a field over which
it is representable.
b) If a matroid has more than one loop then it cannot be a vector one.
5.3.2. a) Prove that the uniform matroid of order 4 and rank 2 is not graphic.
b) Is it a vector matroid?
5.3.3 . a) Prove that the matroid (E, B), where E = {I, 2, 3,4} and the set of bases
B is the set of all three-element subsets of E containing 1, is binary.
b) Find all cycles of this matroid.
c) Is this matroid graphic?
92 5. Matroids
5.3.4. Prove that

a) the dimension of the cycle space of a non empty graph G is equal to the
cyclic rank v(G);
a) the dimension of the cocycle space of a nonempty graph G is equal to the
cocyclic rank v* (G);
c) the number of cycles of a graph G is at most 2v(G) - 1, and the number
of cuts is at most 2v ·(G) - 1 (both these bounds are attainable).
5.3.5. Find a cycle basis, all simple cycles, a cut basis, and all cuts of the graph
shown at Fig. 5.3.1.
5.3.6. Find a cycle basis and a cut basis of the graph with the incidence matrix
1 0 1 0 0 0
1 1 0 1 0 0
I(G) = 0 1 0 0 1 0
0 0 0 1 1 1
0 0 1 0 0 1
Chapter 6
Planarity
6.1 Embeddings of Graphs. Euler Formula

Recall that a continuous rectifiable curve without self-intersections is called a Jor-
dan curve.
It is said that a graph G is embeddable into a space L if there exists a mapping
of its vertices and edges into different points and Jordan curves of L respectively
such that
~ the ends of the edges are mapped into the endpoints of the corresponding
curves;
- these curves intersect only at their common endpoints.
Such mapping (and its image) is called an embedding of the graph G into the space
L.
If a graph is embeddable onto the plane, it is called planar and the (image of
the) embedding is called a plane graph. In this case it is said that the graph has
a plane embedding.
Note. Rigorously speaking, a plane graph is not a graph; it is rather a represen-
tation of the graph which in addition to all graph-theoretical properties possesses
some specific ones.
It is assumed that a graph is isomorphic to all its embeddings and all em-
beddings of the same graph are isomorphic to each other. In particular, a planar
graph is isomorphic to the plane graphs that are its plane embeddings.
The graphs of Platonic solids, Fig. 1.1.2, are examples of planar graphs.
It is evident that
a) a subgraph of a planar graph is planar;
b) a graph is planar if and only if every its component is a planar graph.
Theorem 6.1.1 (Wagner Theorem) For every planar graph there exists an
isomorphic plane graph whose edges are straight line segments.
Theorem 6.1.2 Every graph is embeddable into the three-dimensional space E3.
Theorem 6.1.3 A graph is embeddable onto a sphere if and only if it is planar.
93
94 6. Planarity
A face of a plane graph G is a maximal under inclusion set of points of the

plane such that any two points of it may be connected by a Jordan curve that does
not intersect the edges of the graph, with a possible exception of the endpoints of
the curve.
The set of vertices and edges of a plane graph that belong to a face is called the
boundary of the face. A face of a plane graph bounded by a 3-cycle, or triangle,
will also be called triangle.
A plane graph has exactly one face of infinite area (face 3 at Fig. 6.1.1). Such
face is called external, or outer, the remaining ones are called internal (faces 1,
2 at Fig. 6.1.1).
2 3
Figure 6.1.1: Plane graph
The notions of plane embedding and face are naturally extended to pseudo-
and multigraphs.
A connected graph is called outerplanar if it has a plane embedding such that
all its vertices belong to the boundary of the same face (it is convenient for this
face to be the outer one). An outerplanar graph is called maximal outerplanar if
it ceases to be outerplanar after the addition of any new edge.
In what follows, the numbers n = neG), m = meG), f = f(G) will denote the
numbers of the vertices, edges and faces of a plane graph respectively.
Euler Formula. For a connected plane graph,
n-m+f = 2.
Corollary 6.1.4 The number of faces of every embedding of a planar connected

(n, m)-graph is m - n + 2.
Corollary 6.1.5 For a connected plane (n, m)-graph with n 2: 3,
m ~ 3n - 6.
Corollary 6.1.6 If the boundary of every face of a connected plane (n, m)-graph
is an r-cycle, r 2: 3, then mer - 2) 2: r(n - 2).
Theorem 6.1. 7 A plane graph is biconnected if and only if the boundaries of all
its faces are simple cycles.
6.1. Embeddings of Graphs. Euler Formula 95
Theorem 6.1.8 A connected graph is planar if and only if every its block is pla-
nar.
* * *
6.1.1. Prove that every internal face of a plane graph G may be transformed into
the external face of a plane graph isomorphic to G using the stereographic
projection.
6.1.2. Redraw the plane graph shown at Fig. 6.1.2 making the face a) 1, b) 2, c)
3 external.
6.1.3. Using the Corollary 6.1.5, prove that K5 is nonplanar.

6.1.4. Ihree Houses and Three Wells Problem: There are three houses and three
wells. Each family uses each of the wells. After a quarrel, they decided to
lay non crossing paths from every house to every well. Is this possible?
6.1.5. Construct plane embeddings of the graphs a) K5 - e, b) K 3 ,3 - e, c) K 2 ,n
such that the edges are presented by straight line segments.
6.1.6. Verify the Euler formula for a) the wheel W n , b) graphs of Platonic solids,
c) the graph defined by the crossings and edges on the m x n-chessboard, d)
K 2 ,n.
6.1.7. Prove that the Euler formula may be generalized for the case of a discon-
nected plane (n, m )-graph with k components: n - m + f = k + 1.
6.1.8. To how many faces of a plane graph maya point of an edge that is not a
vertex belong?
6.1.9. To how many faces of a plane graph maya vertex of degree d ~ 1 belong?
6.1.10. Let G be a plane biconnected graph. Prove that the set of all simple cycles
that are the boundaries of the internal faces of G is the cycle basis of G (see
Sect. 5.3).
6.1.11. Let the boundaries of all faces of a plane biconnected graph contain even
numbers of edges. Prove that the lengths of simple cycles of the graph are
even.
96 6. Planarity
6.1.12. Prove that m(h - 2) ~ hen - 2) for a connected planar (n, m)-graph with
n ? 3 of girth h ? 3. Using this relation, prove that the Petersen graph is
nonplanar.
6.1.13. Prove that m ::; 2n - 4 for a connected planar (n, m)-graph with n ~ 3
and without triangular faces.
6.1.14. Prove that m ~ 2(p+ q - 2) for a planar bipartite m--edge graph with the
sizes p, q of the parts at least 2.
6.1.15. Prove that IEHnl ::; 2n+1 - 4, where Hn is a planar spanning subgraph of
Qn, n? 2.
6.1.16. Prove that every planar graph has a vertex of degree at most 5.
6.1.17. Prove that there are no planar 6-connected graphs.
6.1.18. Prove that every planar 5-connected graph has at least 12 vertices. Con-
struct one of them.
6.1.19. For which r there exists a planar r-regular graph?
6.1.20. Prove that a planar connected cubic graph whose every face has at least 5
vertices is of order at least 20.
6.1.21. Let G be a planar connected cubic graph. Prove that
I)6 - i)I; = 12,

i;::3
where Ii is the number of faces of G each of which is bounded by i edges.

(Here and in the sequel, an edge that belongs to a single face is counted
twice in Ii') Prove that G has at least one face with at most 5 edges.
6.1.22. Let G be a plane graph with the number offaces less than 12. Prove that
if the degrees of the vertices of G are at least 3 then G has a face bounded
by at most 4 edges.
6.1.23. Prove that a plane graph of order n ? 4 without triangular faces has the
maximal number of edges 2n - 4 if and only if every its face (including the
external one) is bounded by a 4-cycle.
6.1.24. Prove that the maximal number of edges in a planar spanning subgraph
of Kp,q, p, q ? 2, is 2(p + q - 2).
6.1.25. Prove that a planar spanning subgraph of Kp,q, p, q ~ 2, has 2(p + q - 2)

edges if and only if every its face (including the external one) is bounded by
a 4-cycle.
6.1.26. Solve problem 6.1.15 using problem 6.1.23.

6.1.27. Prove that a planar spanning subgraph of On, n ~ 2, has 2n +1 - 4 edges

if and only if every its face (including the external one) is bounded by a
4-cycle.
6.1.28. Prove that
a) if the faces of a planar graph are bounded by pentagons and hexagons,
then the number of pentagonal faces is even;
b) if in addition this graph is cubic, then the number of pentagonal faces is
12 and it has no hexagonal faces.
6.1.29. For which n the square of the cycle Cn is planar?

6.1.30. Prove that the following statements are equivalent.
1) a graph G is outerplanar;
2) a graph G' obtained from G by adding a new vertex and connecting it
with all other vertices of G by new edges is planar;
3) every block of Gis outerplanar.
6.1.31. Prove that K4 and K 2 ,3 are not outerplanar.

6.1.32. Prove that a connected subgraph of an outerplanar graph is outerplanar.
6.1.33. Which graphs shown at Fig. 6.1.3 are outerplanar?
G3
6.1.34. For which n ~ 3 the wheel Wn is outerplanar?

6.1.35. Prove that (a) m ~ 2n-3 and (b) 3(1 -l)+n ~ 2m for every outerplanar
(n, m)-graph, n ~ 2.
6.1.36. For which n ~ 3 the square ofthe cycle Cn is outerplanar?

6.1.37. Let every connected subgraph F of G (F -=1= G) be outerplanar. Is G
necessarily outerplanar?
6.1.38. Prove that if G contains a subgraph isomorphic to K 1 ,3 then G 2 is not
outerplanar.
98 6. Planarity
6.1.39. Prove that the square of a connected graph G is outerplanar if and only if
G is either K3 or a path graph.
6.1.40. Prove that m ::; (3n - 4)/2 for an outerplanar (n, m)-graph without trian-
gles.
6.1.41. Let a maximal outerplanar graph G have n ~ 3 vertices. Prove that G has
a) a vertex of degree 2; b) a plane embedding with n - 2 internal faces.
6.1.42. Draw all pairwise nonisomorphic maximal outerplanar graphs of order 6.
6.1.43. Prove that every maximal outerplanar graph G of order n ~ 3 has a) 2n-3
edges; b) at least two vertices of degree 2; c) II:(G) = 2; d) at least 3 vertices
of degree at most 3.
6.1.44. Prove that every graph may be embedded into the three-dimensional space
in such a way that all its edges are represented by straight line segments.
6.2 Plane Triangulation

A connected plane graph is called a plane triangulation if all its faces are triangles.
A maximal planar graph is a graph of order n ~ 3 that ceases to be planar
after the addition of any edge. A maximal plane graph is a plane graph isomorphic
to a maximal planar graph.
Theorem 6.2.1 A graph is maximal plane graph if and only if it is a plane tri-
angulation.
Corollary 6.2.2 Every plane graph is a spanning subgraph of a plane triangula-

tion.
Corollary 6.2.3 For any maximal planar (n, m)-graph, m = 3n - 6.
Proposition 6.2.4 If the number of vertices of a plane triangulation is at least 4

then the degree of every vertex is at least 3.
* * *
6.2.1. Prove that a plane triangulation has no cutpoints.
6.2.2. How many faces has an n-vertex plane triangulation?
6.2.3. Prove that a planar graph of order n ~ 4 has at least four vertices of degree
at most 5.
6.2.4. Prove that the single plane triangulation which is a 5-regular graph is the
icosahedron graph, see Fig. 1.1.2.
6.2.5. What is the maximal number of edges in a planar graph of order n ~ 3?

6.3. Planarity Criteria 99
6.2.6. Draw a plane graph of order n > 5 that has exactly four vertices of degree
at most 5.
6.2.7. Does there exist a plane triangulation that has exactly two vertices of odd
degree and these vertices are adjacent?
6.2.8. The vertices of a plane triangulation are colored in three colors. A face is
said to be tricolored if its vertices are colored in different colors. Prove that
the number of tricolored faces is even.
6.3 Planarity Criteria

The edge subdivision operation for an edge e = ab is the deletion of e from the
graph and the addition of two edges av and vb, where v is a new vertex.
Two graphs are called homeomorphic if they may be obtained from the same
graph by edge subdivisions. By the definition, every graph is homeomorphic to
itself.
Kuratowski Criterion: A graph is planar if and only if it has no subgraphs
homeomorphic to K5 or K 3 ,3'
A graph G is said to be contractible to a graph H if H may be obtained from
G by edge contractions. By the definition, every graph is contractible to itself.
Wagner Criterion: A graph is planar if and only if it has no subgraphs
contractible to K5 or K 3 ,3.
* * *
6.3.1. Using the Kuratowski criterion, prove that the Petersen graph is nonplanar.
6.3.2. Prove that the n-transposition graph (see Exercise 1.1.27 for the definition)
is non planar for n 2:: 3.
6.3.3. For which n the graphs G n defined in the Exercise 1.3.4 are planar?
6.3.4. How many are there nonisomorphic nonplanar graphs of sixth order?
6.3.5. Which of graphs shown at Fig. 6.3.1 are nonplanar?
6.3.6. For which n the graphs of order 2n shown at Fig. 6.3.2 are planar?
6.3.7. Let Hn = (Vi, V2; E) be a bipartite graph defined as follows.
Xi = aibi, i = 1, .. . ,n; Yi = aibi+l, i = 1, .. . ,n -1; Yn = anb1;

Zi = aibi+2, i = 1, ... , n - 2; Zn-l = an-1b1; Zn = an b2.
For which n 2:: 3 the graphs Hn are planar?
100 6. Planarity
a b c d
a b
6.3.8. Let G be a graph of order n :::; 7. Prove that either G or G is planar.

6.3.9. What is the minimal number n such that there exists a nonplanar n-vertex
graph with nonplanar complement.
6.3.10. Find a graph G of order 8 such that
a) both G and G are planar;
b) both G and G are nonplanar;
c) one of G and G is planar and the other is not.
6.3.11. Is it true that for every 8-vertex triangulation the complementary graph is
planar?
6.3.12. Let G be a graph of order n ~ 11. Prove that G and G cannot be both
planar.
6.3.13. Prove that if an (nl' mI}-graph and an (n2' m2)-graph are homeomorphic
then
ni + m2 = n2 + mI·
6.3.14. Give examples of homeomorphic graphs with homeomorphic line graphs.
6.3.15. Is it true that for homeomorphic graphs their line graphs are homeomor-
phic?
6.3.16. Using the Wagner criterion, prove that the Petersen graph is nonplanar.
6.4. Duality and Planarity 101
6.3.17. Let G l be contractible to G 2 . Is it true that L(G l ) is always contractible

to L(G 2 )?
6.3.18. Is it true that if a graph is contractible to K5 then it has a subgraph

homeomorphic to K5?
6.3.19. Prove that a graph obtained by a contraction of an edge of a planar graph

is planar.
6.3.20. Prove that if a graph is contractible to a nonplanar graph then it is non-

planar.
6.3.2l. Prove or disprove: every connected nonplanar graph is contractible either

to f{5 or to f{3,3'
6.3.22. Verify that the line graphs of f{5 and f{3,3 are nonplanar.
6.3.23. Let a graph G l be homeomorphic to a graph G 2 that has no vertices of

degree two. Prove that L(G l ) is contractible to L(G 2 ).
6.3.24. Prove that if a graph is nonplanar then its line graph is also nonplanar.
6.3.25. Is it true that the line graph of a planar graph is planar?
6.3.26. For which r 2: 1 the line graph of an r-regular graph is planar?
6.3.27. Let the line graph of a graph G be planar. Prove that a) G is planar; b)
6.(G) ::; 4; c) every vertex of G of degree 4 is a cutpoint.
6.3.28. Prove that if the square of a graph G is planar then 6.( G) ::; 3.
6.3.29. Prove that a graph is outerplanar if and only if it has no subgraphs home-
omorphic to f{4 or f{2,3.
6.3.30. A connected nonplanar graph is called s-critical if the deletion of any its
edge produces a planar graph. Describe all s-critical graphs.
6.3.3l. A connected nonplanar graph is called t-critical if the contraction of any

its edge produces a planar graph. Describe all t-critical graphs.
6.4 Duality and Planarity

Note. In this section, the term" graph" will mean" pseudograph" .
For a plane graph G we may construct another plane graph G* called geometric
dual for G as follows. Pick a point vi inside every face Fi of G. These points
will be the vertices of G*. To every edge e E EG we assign a Jordan curve e*
that crosses only this edge and connects the points vi, vJ lying in the faces whose
boundaries contain e (there may be one or two such faces; in the former case the
curve connects a point vi with itself, i.e., it is a loop). The curves e* will be the
edges of G*. Clearly, one may choose the edges e* to be pairwise nonintersecting,
with the exception of their common endvertices. At Fig. 6.4.1, the graph G is
102 6. Planarity
drawn in straight lines and the geometric dual graph G* is drawn curved. As the
figure shows, this construction not always gives an ordinary graph, hence the note
above.
-----"-"--------------
Figure 6.4.1: A graph and its geometric dual
One may easily verify that a plane graph G* geometric dual to a plane graph
G is unique up to isomorphism.
Proposition 6.4.1 IfG is a connected plane (n, m)-graph with f faces and G* is
an (n*, m*)-
graph with f* faces geometric dual to G, then n* = f, m* = m, f* = n.
Since G* is a plane graph, one may construct a geometric dual to it, naturally
denoted by G** .
Theorem 6.4.2 If G is a plane connected graph then G** is isomorphic to G.
A graph geometrically dual to a planar graph is defined as follows: construct
any plane embedding of the graph and take its geometric dual.
Theorem 6.4.3 Let G be a planar graph with geometric dual G*. A subset of
edges in G constitutes a simple cycle if and only if the corresponding subset of
edges of G* constitutes a cut.
This theorem naturally leads to the following combinatorial definition of duality

for graphs that generalizes the geometric duality.
A graph G+ is called abstract dual to G if there exists a bijection between the
sets EG and EG+ such that a subset of edges in G constitutes a simple cycle if
and only if the corresponding subset of edges of G+ constitutes a cut.
Fig. 6.4.2 shows a graph and its abstract dual. Fig. 6.4.3 shows another ex-
ample of graph and its abstract dual. Notice that here G+ is not a geometric dual
to G, but G is a geometric dual to G+.
G G+
Figure 6.4.2: A graph and its abstract dual
Proposition 6.4.4 A graph geometrically dual for a plane graph is also an ab-
stract dual for it.
Proposition 6.4.5 A graph H is an abstract dual for a graph G if and only if the
matroid M( G) of cycles of G is isomorphic to the matroid M*(H) of cuts of H.
Theorem 6.4.6 If G+ is an abstract dual for a graph G then G is an abstract

dual for a graph G+.
a e
d h
G G+
Figure 6.4.3: Another graph and its abstract dual
Notice that in contrast to Theorem 6.4.2, G is not required to be connected.

In view of Theorem 6.4.6, graphs G and G+ may be called dual to each other.
Theorem 6.4.7 If a graph has an abstract dual then every its subgraph has an
abstract dual.
Theorem 6.4.8 If a graph G has an abstract dual then every graph homeomorphic
to G has an abstract dual.
104 6. Planarity
Whitney Criterion: A gmph is planar if and only if it has an abstmct dual.

The Whitney criterion implies that K5 and K 3 ,3 have no abstract duals.
* * *
6.4.1. Construct geometric duals for the graphs from Fig. 6.4.4.
6.4.2. Construct geometric duals for the graphs of Platonic solids, Fig. 1.1.2.
6.4.3. Prove that a graph geometrically dual to a planar graph is connected.
6.4.4. Prove that if a plane graph has a cutpoint then its geometric dual also has
a cutpoint.
6.4.5. Prove that a plane graph is biconnected if and only if its geometric dual is
biconnected.
6.4.6. Prove that a graph geometric dual to a plane triangulation of order n ~ 4

is a biconnected cubic plane graph.
6.4.7. Verify the isomorphism of graphs shown at Fig. 6.4.5. Are their geometric
duals isomorphic?
6.4.8. Let a plane graph G be disconnected. Prove that the iterated geometric
dual G** is not isomorphic to G.
6.4.9. Show that a geometric dual to a wheel is a wheel.

6.4.10. Under which conditions a graph geometric dual to an ordinary plane graph
is an ordinary graph?
6.4.11. A plane graph is called selfdual if it is isomorphic to its geometric dual.

Prove that m = 2n - 2 for a selfdual (n, m)-graph.
6.4.12. Show that a wheel Wn , n 2: 3, is selfdual. Give other examples of selfdual

graphs.
6.4.13. How many are there pairwise nonisomorphic selfdual graphs of 6th order?
6.4.14. Prove that there are no plane graphs with five faces such that any two
faces have a common edge.
6.4.15. Let G be a connected ordinary plane graph such that its geometric dual
G* is also an ordinary graph (cf. Exercise 6.4.10). Prove that G and G*
have the same number of spanning trees.
6.4.16. Prove that if the degrees of the vertices of a biconnected plane graph are
even then the geometric dual is bipartite.
6.4.17. Find two nonisomorphic graphs abstractly dual to the graph shown at
Fig. 6.4.6.
6.4.18. Do there exist nonisomorphic graphs abstractly dual to the graph from
Fig. 6.4.7?
106 6. Planarity
6.4.19. Prove that if G and G+ are dual to each other then n(G) - k(G) = v(G+)
and n(G+) - k(G+) = v(G), where n(G) is the number of vertices, k(G) is
the number of components, v(G) is the cyclic rank of a graph G.
6.4.20. Which pairs of graphs from Fig. 6.4.8 are dual to each other?
G H Q
6.4.21. Let G+ be an abstract dual to a planar graph G and let G++ be an abstract
dual to G+. Is it always G++ ~ G?
6.4.22. a) Prove that the following statements are equivalent:
1. G is a planar graph.
2. The cyclic matroid M(G) is cographic.
3. The matroid of cuts M*(G) is graphic.
b) Prove that K5 and K 3 ,3 are nonplanar using this criterion.
6.4.23. A graph G* is called a generalized dual to a graph G if there exists a bijec-
tion between the sets EG and EG* such that for every spanning subgraph
H ofG
v(H) + v*(H*) = v*(G*),
where H* is a subgraph of G* that corresponds to H, H* is the graph
obtained from G* by the deletion of the edges of H* , v( G) denotes the cyclic
rank, v*(G) denotes the co cyclic rank of a graph G.
6.5. Measures of Displanarity 107
Prove the following equalities.

a)v(G) = II*(G*);
b)v(G*) = v*(G);
c) v*(H) + v(H*) = v*(G).
6.4.24. Prove that if G* is a generalized dual to a graph G then G is a generalized

dual to G*.
6.4.25. Prove that a geometrical dual to a connected plane graph is a generalized

dual to this graph.
_...---_._--
Figure 6.4.9: Embedding of K 3 ,3 onto the torus
6.5 Measures of Displanarity

The genus of a graph G is defined to be the minimal number r = r(G) of handles
necessary to attach to the sphere so that G is embeddable onto the resulting
surface. Such surface is said to be of genus r. The graph of genus r is a graph
embeddable onto the surface of genus r but not onto the surface of genus r - 1.
Clearly, the planar graphs are the graphs of genus O.
A sphere with one handle (a genus 1 surface) is a torus. The graphs of genus
1, i.e., the ones that cannot be embeddable onto a plane (hence onto a sphere)
but that can be embedded onto the torus are called toroidal. Fig. 6.4.9 shows an
embedding of K3,3 onto the torus.
An embedding of a graph onto the torus is convenient to sketch using the rect-
angle with opposite sides pairwise identified as shown at Fig. 6.5.1. At Fig. 6.5.2,
an embedding of K 3 ,3 onto the torus is shown in such a way.
108 6. Planarity
a b a
c c
a b a
Figure 6.5.1: A representation of the torus
The following theorem generalizes the Euler formula (Sect.6.1). In this gener-
alization a face of a genu8-, graph is naturally defined in terms of Jordan curves
on the surface of genus ,.
Theorem 6.5.1 For a connected graph of genus 'Y with n vertices, m edges and
f faces, n - m + f = 2 - 2'Y.
1
.1
1
1
l ______ ~------.------
Figure 6.5.2: Embedding of K 3 ,3 onto the torus
The following formulae are known for the genus of some graphs:
,(Kn) = r(n - 3)(n - 4)/121, n 2:: 3. (6.5.1)
,(Kp,q) = r(p - 2)(q - 2)/41- (6.5.2)

,(Qn) = (n - 4)2 n- 3 + 1. (6.5.3)
The crossing number cr(G) of a graph G is the minimal number of pairwise

intersections of edges possible when drawing a graph on the plane. Clearly, cr( G) =
o if and only if G is a planar graph.
The thickness t( G) of a graph G is the minimal number of its planar subgraphs

whose union is the graph G. Clearly, the thickness of a planar graph is 1.
The skewness sk( G) of a graph G is the minimal number of edges whose deletion
produces a planar graph. Clearly, sk( G) = 0 if and only if G is a planar graph.
The arboricity ( G) of a graph G is the minimal number of edge-disjoint acyclic
spanning subgraphs whose union is G.
Theorem 6.5.2 (Nash-Williams Formula) If G is a nontrivial (n, m)-graph

and mp denotes the maximal number of edges in a subgraph of G with p vertices
then
(G) = max rmp/(p-1)1.
p=2, ... ,n
* * *
6.5.1. Prove that a) "Y(Ks) = 1; a) "Y(K7) = 1; a) "Y(K4,4) = 1.
6.5.2. Find the genus of the Petersen graph.
6.5.3. Embed Q4 onto the torus.
6.5.4. Give an example of a graph of genus 2.
6.5.5. Prove that there are no complete graphs of genus 7. What is the next integer
which is cannot be the value of the genus of a complete graph?
6.5.6. Using the Theorem 6.5.1 prove that

a) "Y(G) 2:: rem - 3n)/61 + 1 for a connected (n, m)-graph G;
b) "Y(G) 2:: rem - 2n)/41 + 1 for a connected triangle-free (n, m)-graph G;
c) "Y(Kn) 2:: r(n - 3)(n - 4)/121;
d) "Y(Qn) 2:: (n - 4)2 n- 3 + 1;
e) "Y(Kp,q) 2:: rep - 2)(q - 2)/41-
6.5.7. Find a lower bound for the genus of an (n, m)-graph of girth g. Deduce from
it that "Y(Kp,q) = rep - 2)(q - 2)/41·
6.5.8. What are the crossing numbers of K 3 ,3, Ks and the Petersen graph?
6.5.9. Prove that if graphs G l , G 2 are homeomorphic then cr(Gl) = cr(G 2 ).

6.5.10. Prove that if p, q are even integers then cr(Kp,q) :::; pq(p - 2)(q - 2)/16.
What is the upper bound for the crossing number of Kp,q when p, q are not
necessarily even?
6.5.11. Prove or disprove the statement: cr(G) = 1 for a nonplanar graph G if

and only if it has an edge e such that cr( G - e) = O.
110 6. Planarity
6.5.12. Prove that 7(G) :s cr(G). Give examples of graphs such that a) 7(G) <
cr(G);
b) 7(G) = cr(G).
6.5.13. Find the thickness for /{3,3, /{5, and the Petersen graph.
6.5.14. Prove that every nonplanar graph is homeomorphic to a graph of thickness
2.
6.5.15. Prove that t(G) 2:: fm/(3n - 6)l for a connected (n, m)-graph G.
6.5.16. Prove that t(I<n) 2:: len + 7)/6J.
6.5.17. Prove that t( G) 2:: fm/2(n - 2)l for a connected bipartite (n, m)-graph G,
n> 2.
6.5.18. Prove that

a) t(Kp,q) 2:: fpq/2(p + q - 2)l, p + q 2:: 2;
b) t(Qn) 2:: In/4J + 1, n 2:: 1.
6.5.19. What is the minimal number of edges to be deleted from Q4 to make it
planar?
6.5.20. Find the skewness for K 3 ,3, K 5 , and the Petersen graph.
6.5.21. Prove that sk(Kn) = C~ - 3n + 6, n 2:: 3.

6.5.22. Find the skewness for Kp,q and Qn.
6.5.23. Prove that the arboricity of a planar graph is at most 3.
6.5.24. Construct a planar graph of arboricity 3.
6.5.25. Prove that every graph is homeomorphic to a graph of arboricity 1 or 2

and hence of thickness 1 or 2.
Chapter 7
Graph Traversals
7.1 Eulerian Graphs

A cycle in a graph is called Eulerian if it contains all edges of the graph. (Recall
that by definition any edge may enter a cycle at most once.) A connected graph
with an Eulerian cycle is called Eulerian.
Theorem 7.1.1 A connected graph is Eulerian if and only if the degrees of all its
vertices are even.
We shall say that a set of edge-disjoint paths (not necessarily simple) covers a
graph if every edge of the graph belongs to some of these paths.
Theorem 7.1.2 If a connected graph contains exactly k vertices of odd degree

then the minimal number of edge-disjoint paths that cover the graph is k/2.
In particular, if k = 2 then the graph has a path connecting the odd-degree

vertices and containing all edges of the graph. Such path is called Eulerian.
* * *
7.1.1. Prove that a connected graph is Eulerian if and only if every its block is
Eulerian.
7.1.2. Prove that every graph G may be represented as the union of two edge-
disjoint graphs GIl G 2 such that G I is a forest and the degrees of the vertices
of G 2 are even.
7.1.3. Prove that an Eulerian graph is bridgeless.
7.1.4. Prove that a connected graph is the union of edge-disjoint simple cycles if
and only if it is Eulerian.
7.1.5. Let every edge e of an Eulerian graph G be provided with a real number
w(e). Prove that if w(G) = L:eEEG w(e) > 0 then G has a simple cycle C
such that w(C) = L:eEEC w(e) > O.
111
112 7. Graph Traversals
7.1.6. A cycle (ao, ... ,ap-l,ao) is called a subcycle of cycle (bo, ... ,bq-1,bo) if
ao = bk, al = h+l, ... ap-l = h+p-l, ao = bk +p for some k (here the
addition in the indices of bi is modulo q). Is it true that every simple cycle
of an Eulerian graph is a sub cycle of an Eulerian cycle?
7.1.7. Let G be a tree. Under which conditions G 2 is Eulerian?
7.1.8. Let G be a connected graph distinct from K 2 • Under which conditions L(G)
is Eulerian?
7.1.9. Find the minimal number of edge-disjoint paths covering the graph shown
at
Fig. 7.1.1.
Figure 7.1.1: To Exercise 7.1. 9
7.1.10. Prove that every connected graph contains a walk in which every edge of
the graph is repeated at most twice.
7.1.11. Prove that an Eulerian graph contains a single Eulerian cycle if and only
if the graph is a cycle graph.
7.1.12. Design an algorithm for the construction of an Eulerian cycle in an Eulerian

graph.
7.2 Hamiltonian Graphs

A graph is called Hamiltonian if it has a simple cycle containing all vertices of
the graph. Such cycle is called Hamiltonian. A simple path is called Hamiltonian
if it contains all vertices of the graph. There are quite a few theorems stating
sufficient conditions for the existence of a Hamiltonian cycle in a graph.
Theorem 7.2.1 (V.Chvatal) A graph with the degree sequence d1 ~ d2 ~ ... ~

dn , d1 ;::: 1, is Hamiltonian if either die > k or dn - Ie ;::: n - k for every k, 1 ~ k <
n/2.
7.2. Hamiltonian Graphs 113
A graph is called locally-connected if the neighborhood of every its vertex

induces a connected graph.
Theorem 7.2.2 Let G be a connected locally-connected graph of order n 2: 3. If

G has no induced subgraphs isomorphic to K 1 ,3 then G is Hamiltonian.
Theorem 7.2.3 Every 4-connected planar graph is Hamiltonian.
Theorem 7.2.4 If G is biconnected then G 2 is Hamiltonian.
The following theorem gives a necessary condition for the Hamiltonicity.
Theorem 7.2.5 Let fk be the number of faces of a plane graph G with exactly
k edges at the boundary. If G is Hamiltonian then for every k 2: 3 there exist
nonnegative integers f~, fr such that
L f~(k - 2) =L fr(k - 2).
* * *
7.2.1. Find a Hamiltonian cycle a) in the graph shown at Fig. 7.2.1; b) in the
dodecahedron graph.
7.2.2. Is the Petersen graph Hamiltonian?
7.2.3. A graph is called Hamiltonian-connected if every two its vertices are con-
nected by a Hamiltonian path. Is Q3 Hamiltonian-connected?
7.2.4. A non-Hamiltonian graph G is called hypohamiltonian if G - x is Hamilto-

nian for every vertex x. Prove that the Petersen graph is hypohamiltonian.
7.2.5. Find a connected graph G of order n ~ 3 such that G 2 is non-Hamiltonian.

Find a graph of minimal order with these properties.
7.2.6. Prove that if a cubic graph is Hamiltonian then it contains three edge-
disjoint perfect matchings. Is the converse statement true?
~ !>
7.2.7. An n-vertex graph is called pancyclic if it contains a simple cycle oflength

k for every k = 3, ... n. Is the graph shown at Fig. 7.2.2 pancyclic?
7.2.8. Prove that iffor a graph G there exists X ~ VG such that k(G - X) > IXI
then G is non-Hamiltonian. (Here k(H) denotes the number of components
of a graph H.)
7.2.9. Prove that if a graph G has an independent vertex subset X ~ VG such

that IXI > IN(X)I then G is non-Hamiltonian.
7.2.10. Prove that ao(G) :::; IGI/2 for a Hamiltonian graph G.
7.2.11. Let Hn,m be an n-vertex graph with the degree sequence
Pn m = (m, ... , m, n - m -1, ... , n - m -1, n -1, ... , n - 1).

, ~' v "v~--~·
m times n - 2m times m times
Prove that Hn,m is non-Hamiltonian for all m, 0 < m < n/2.

7.2.12. Let d 1 :::; d2 :::; ••• :::; dn be the degree sequence ofa graph that does not sat-
isfy the hypotheses of Theorem 7.2.1. Prove the existence of a number m, 0 <
m < n/2, such that di:::; Pi,m, i = 1, ... ,n, where (Pl,m,P2,m, ... ,Pn,m) =
Pn,m is the sequence defined in the previous exercise.
7.2.13. Prove that an n-vertex graph G is Hamiltonian if IEGI > (n - 2t + n.
7.2.14. Prove that every connected locally-connected line graph of order n > 2 is
Hamiltonian.
7.2.15. Prove that if a graph G is connected and every its edge belongs to a triangle
then L(G) is Hamiltonian.
7.2.16. Prove that if a graph G, IGI 2:: 3, is connected and locally-connected then
L( G) is Hamiltonian.
7.2.17. Prove that if a graph G, IGI2:: 3, is connected then L(G2) is Hamiltonian.
7.2.18. For positive integers m, n a graph G is called a grid graph if VG = {(x, y) :

= = =
x 1, ... , m, y 1, ... , n} and ab E EG if and only if a (x, y), b (z, t) =
and Ix - zl + Iy - tl = 1. For which m, n G is Hamiltonian?
7.2.19. Prove that a connected locally connected regular graph of degree 4 is

Hamiltonian.
7.2.20. Let an n-vertex graph G satisfy the hypotheses of Theorem 7.2.1. Prove
that IEGI;::: (n - 1)2/8.
7.2.21. Prove that if deg v;::: IGI/2 for any vertex of a graph G then G is Hamil-
tonian.
7.2.22. For each n ~ 4 construct a non-Hamiltonian (l n/2J-I)-connected n-vertex

graph.
7.2.23. For each theorem 7.2.1,7.2.2,7.2.3 find a graph satisfying the hypothesis of
the theorem and not satisfying the hypotheses of the remaining ones.
7.2.24. Find a cubic non-Hamiltonian graph in which every pair of nonadjacent

vertices is connected by a Hamiltonian path.
7.2.25. Prove that a Hamiltonian-connected graph of order n ~ 4 is 3-connected.
7.2.26. Prove that .6.(G) ::; (n - 1)/2 for a hypohamiltonian graph G.
7.2.27. Let t(X) denote the minimal number of vertex-disjoint paths covering the
subset X ~ VG of a graph G.Prove that if t(X) > IVG\XI for some
X E VG, then G is non-Hamiltonian.
7.2.28. Construct a biconnected cubic bipartite planar non-Hamiltonian graph.
7.2.29. Construct a triconnected non-Hamiltonian graph of radius 1.
7.2.30. A planar triconnected k-regular graph is called k-polytope. Let G be a non-

Hamiltonian bipartite 3-polytope with minimal number of vertices. Prove
that G has no vertex that belongs only to faces bounded by 4-cycles.
7.2.31. Prove that a graph shown at Fig. 7.2.3 has no Hamiltonian cycle.
7.2.32. A graph is called a caterpillar if the deletion of all its pendant vertices
produces a path graph. Prove than the square of a caterpillar is Hamiltonian.
7.2.33. Construct a connected locally connected non-Hamiltonian graph with more

than two vertices.
7.2.34. Let P = (Pl, ... ,Pn), Q = (ql, ... ,qn) be degree sequences such that
qi ;::: Pi, i = 1, ... , n. Is it true that if every graph with degree sequence P is
Hamiltonian then every graph with degree sequence Q is also Hamiltonian?
7.2.35. Prove that if deg u + deg v ;::: IGI for any two nonadjacent vertices u, v of a
graph G then G is Hamiltonian.
7.2.36. Prove that if deg u + deg v;::: IGI- 1 for any two nonadjacent vertices u, v
of a graph G then G contains a Hamiltonian path.
7.2.37. Suppose that deg u + deg v 2=: IGI for some two vertices u, v of a graph G.
Prove that if G contains a Hamiltonian (u, v)-path then G is Hamiltonian.
7.2.38. For every n 2=: 4 find an n-vertex non-Hamiltonian graph such that deg u + .
deg v 2=: n - 1 for any two vertices u, v of a graph G
7.2.39. A graph is called locally Hamiltonian if the induced graph G(N(v)) is
Hamiltonian for every vertex v. Construct a non-Hamiltonian locally Hamil-
tonian graph.
7.2.40. Let G be a complete n-partite graph, IGI2=: 3. Prove that:
a) G is Hamiltonian if and only if IGI2=: 2ao(G).
b) If G is non-Hamiltonian then it has a cycle of length 2f3o(G).
7.2.41. Let a graph G have a cycle that contains a vertex covering of the graph.
Prove that L( G) is Hamiltonian.
7.2.42. Let I be the incidence matrix of a graph G, considered over the field Z2.
Prove that G is Hamiltonian if and only if I has IGI-1linearly independent
columns such that their sum is also a column of I.
7.2.43. Let ab be an edge of a graph G such that G - a - b is disconnected. Prove
then G has no Hamiltonian cycles passing through abo
7.2.44. Prove that every biconnected outerplanar graph has a single Hamiltonian
cycle.
7.2.45. Is it true that for a graph G, IGI2=: 4, one of L(G), L(G) is Hamiltonian?
Chapter 8
Degree Sequences
8.1 Graphical Sequences

A sequence (d 1 , d 2 , ... d n ) of nonnegative integers will be called an n-sequence and
denoted by the single letter d. The sequence d is called graphical if there exists a
graph whose degree sequence is d. Such graph is called a realization of the sequence
d.
A nonincreasing n-sequence d is called proper if its sum is even and d 1 ~ n - 1.
Without loss of generality any graphical sequence may be ordered and considered
to be proper.
There is a convenient" shorthand" for graphical sequences: we write down only
different elements and their multiplicities, for example, (2 3 ,12) = (2,2,2,1,1).
As a rule a graphical sequence may have several realizations. For example,
(2 3 ,12) has two non isomorphic realizations shown at Fig. 8.1.1.
• •
G H
Figure 8.1.1: Two realizations of the same sequence
Let a, b, c, d be four different vertices of a graph G such that ab, cd E EG and

ac, bd tf. EG. In such case it is said that G admits the switching s = (ab, cd), which
is the deletion of the edges ab, cd followed by the addition of the edges ac, bd. The
resulting graph G - ab - cd + ac + bd is denoted by sG.
Theorem 8.1.1 (Switching Theorem) Any realization of a graphical sequence

may be obtained from any other one by a finite sequence of switchings.
117
118 8. Degree Sequences
If a sequence has a unique (up to isomorphism) realization G then such sequence

is called unigraphical and G is called a unigraph. The structure of unigraphs is
known, but its description is cumbersome. We give here only the theorem about
unigraphicity for multigraphs.
Theorem 8.1.2 (K.Senior, S.Hakimi) Let a proper n-sequence d be realized by

a multigraph. The realization is unique if and only if one of the following conditions
hold:
a) n ~ 3;
c) d 1 + 2 = d2 + ... + dn - 1, d2 = da = .. , = dn ;
d) n = 4, da > d4 = 1;
e) d2 = da = ... = dn = 1.
For a proper n-sequence, n > 1, the derived sequence di , 1 ~ i ~ n is defined

as follows. The element di is deleted from d and the first di remaining elements
are decreased by 1.
Theorem 8.1.3 (V.Havel, S.Hakimi) A proper n-sequence d -=f. (on) is graph-

ical if and only if every derived sequence di , 1 ~ i ~ n, is graphical.
Theorem 8.1.4 (P.Erdos, T.Gallai) A proper n-sequence, n > 1, is graphical

if and only if
k n
L d; ~ k(k - 1) + L min{k, di }
;=1
for all k = 1, 2, ... , n - 1.
Clearly, to verify the graphicity using any of the two latter theorems, n inequal-
ities must be verified in the worst case. However it is often possible to test the
graphicity of a sequence much faster, as the following sufficient condition suggests.
Theorem 8.1.5 A proper n-sequence d without zeroes such that
is graphical.
= =
A pair d (d 1 , ... , dn ), c (C1, •.. , cm ) of proper sequences is called graphical
if there exists a bipartite graph (called a realization of the pair) for which d, care
the lists of degrees of the vertices of its parts.
8.1. Graphical Sequences 119
Theorem 8.1.6 (D.Gale) A pair of proper sequences d and c with equal sums
of elements is graphical if and only if
k m
Ld i ::; Lmin{k,ci}.
i=l i=l
for all k = 1,2, ... , n.

Theorem 8.1. 7 A pair of proper sequences d and c with equal sums of elements
is graphical if and only if the sequence
C1 +m - 1, C2 +m - 1, ... , Cm +m - 1, d 1 , d2 , ... , dn
is graphical.
The following algorithm called the I-procedure may be used for testing a proper
n-sequence for graphicity and for constructing a possible realization.
Let V be an n-element set of vertices (of a graph to be constructed) such that
every element v E V is provided with a nonnegative integer label d( v) < n. Let
further S( v) ~ V be a subset of d( v) vertices distinct from v with maximal possible
labels (S( v) is not unique).
An iteration of the I-procedure consists of the following steps.
1. pick a leading vertex with nonzero label; denote it by v;
2. pick a subset S(v);
3. connect the leading vertex by edges with all vertices from S(v);
4. set d(v)=O, set d(u) = d(u) - 1 for all u E S(v).
If after an iteration some label becomes negative, we shall say that the iteration
fails.
The I-procedure is applied to a proper n-sequence d. Initially it is assumed that
d( vd = di and then the above iterations are successively applied. Two outcomes
are possible: either some iteration fails, or we obtain the zero sequence. In the
first case d is not graphical. In the second case we obtain a realization of d.
* * *
8.1.1. Find all pairwise non isomorphic realizations of the following sequences: a)
(42,3 4); b) (210); c) (5,43,34).
8.1.2. Find a sequence with minimal sum of elements that has more than one
realization.
8.1.3. a) Prove that if a graph of order n has exactly two vertices of degree k and
the degrees of the remaining vertices are different, then k = Ln/2J.
b) Describe the structure of a graph in which only two vertices are of the
same degree.
8.1.4. Show that a proper n-sequence with some coinciding elements is not neces-
sarily graphical.
8.1.5. Prove that the three sequences
are graphical or not simultaneously and have the same number of realizations.
8.1.6. Let d = (dl, ... ,dn ), C = (Cl,""Cm ) be two graphical sequences. Prove
that the following sequences are also graphical.
a) the union dU C = (d l , ... , dn , Cl, ... , cm);
b) the join d + C = (d l + m, ... , dn + m, Cl + n, ... , Cm + n).
8.1.7. Let graphical sequences d and C have Nand M realizations respectively. Is
it true that the number of realizations for the union dUe (see the previous
exercise) is N M?
8.1.8. A molecular graph is a multigraph whose vertices correspond to the atoms
of the molecule and the edges correspond to the bonds between the atoms.
In general, a molecular formula may correspond to several molecular graphs,
which are called structural isomeres.
a) Find all structural isomeres of the compound N 2 0 a (the nitrogen N has
valency 3 and the oxygen 0 is of valency 2.)
b) A saturated hydrocarbonate is a compound of the hydrogen H (valency 1)
and the carbon C (valency 4) whose molecular graph is a tree. Derive the
molecular formula Cn H 2n +2 for saturated hydrocarbonates.
c) Draw the graphs of saturated hydrocarbonates with 1,2,3 and 4 atoms of
carbon.
8.1.9. Let
d = (d l , d2 , . .. , dn ), d' = (d~, d~, . .. , d~)
be two graphical sequences with di 2: d~, i = 1,2, ... , n. Is it true that there
exists a graph G realizing d such that a spanning subgraph of G realizes d'?
8.1.10. Find a sequence of switchings that transforms the graph G shown at
Fig. 8.1.2 into 2K4 .
8.1.11. Suppose that a switching s is applied to a graph G and to its sub graph H.
Is it true that if H '1f. sH then G '1f. sG?
8.1.12. Let the vertices Vi,"" Vn of a graph G of order n have degrees d1 2:

... 2: dn respectively. Prove that for every vertex Vi there exists a sequence
S1, S2, ... , St of switchings such that the vertex Vi in the graph St . .. S2S1 G
is adjacent to the first di vertices distinct from Vi.
8.1.13. What are the possible values of the diameter of a graph that does not
admit switchings?
1 2
8,------~-+------:3
7-------+--~----~~
6 5
8.1.14. State and prove the Switching Theorem for a) multigraphs and b) pseudo-
graphs, modifying the definition of switching in a proper way.
8.1.15. How can one switching change the number of components of a graph?
8.1.16. Let d be a graphical sequence. The graph of realizations Gd is defined as
follows. Its vertices bijectively correspond to pairwise nonisomorphic realiza-
tions of d. Two vertices of Gd are defined to be adjacent if the corresponding
realizations may be obtained from each other by a switching.
a) Construct the graphs of realizations for the sequences d = (2\ 12) and
d' = (4,2\ 12 ).
b) Prove that the graphs of realizations of sequences from Exercise 8.1.5 are
isomorphic.
c) Prove that the graph of realizations of any graphical sequence is connected.
d) What is the degree of the vertex corresponding to the cycle Gn in the
=
graph Gd with d (2n), n 2: 3?
8.1.17. An alternating cycle in a graph G is a sequence G = (Vi, V2, ... , V2k, Vi)
of vertices that constitutes a cycle in the complete graph with the vertex
set VG such that every other edge of G is not in EG (antiedge) and the
remaining edges are in EG. A long switching s(G) is the deletion of the
edges of an alternating cycle G from G and the addition of the antiedges of
Gto G.
a) Prove that any realization of a graphical sequence may be obtained from
any other one by a finite sequence of long switchings.
b) Let s( G, G') denote the minimal number of long switchings necessary to
produce a realization G' of a graphical n-sequence d from another realization
G of the same sequence. Let us denote s(d) = maxs(G, G'), where the
maximum is over all pairs of realizations of d. Prove that s(d) $ Ln/4J.
8.1.18. A labelled realization of a graphical n-sequence is a graph G with the
vertex set V = = =
{Vi, V2, ... , vn } such that deg Vi di, i 1,2, ... , n. Labelled
realizations are considered to be different if EG "# EH.
a) Prove that the Switching Theorem remains valid for labelled realizations.
b) Construct all labelled realizations of the sequence (2 3 , 12 ).
c) Give the definition of the graph of labelled realizations G~, similar to the
graph Gd from Exercise 8.1.16. and construct it for the sequence (2 2 ,1 2 ).
d) In the graph of labelled realizations G~, two vertices are said to be equiva-
lent if the corresponding labelled realizations are isomorphic. Let V1 , ... , Vk
be the resulting equivalence classes. Construct the graph Gd from the graph
G~ and the partition V1 U ... U Vk = V.
e) Prove that d(D~) ?: d(Dd), where d(G) denotes the diameter of the graph
G.
f) Similarly to the Exercise 8.1.17, one may define the "labelled" parameter
s/(d). Let us define further s/(n) =
maxs/(d), where the maximum is over
all graphical n-sequences d. Prove that s/(n) = O(n 2 ).
8.1.19. A graph-theoretical property P is said to be s-complete iffor any graphical

sequence and any two its realizations G, G' E P there exists a sequence of
switchings transforming G into G' such that all intermediate graphs have
the property P.
a) Prove that if a property P is specified in terms of the graphical sequence
of a graph then it is s-complete. Give examples of such properties.
b) Restate the definition of the s-complete property in terms of graphs of
realizations.
c) Prove that the properties "to be bipartite graph" and "to be cactus" are
not s-complete. (A cactus is a connected graph in which no pair of simple
cycles have common edges.)
8.1.20. Let nonincreasing n-sequences d, c, b be such that di ~ Ci ~ bi,

1,2, ... , n, the sum of the elements of c is even, and d, b are graphical. Is it
true that c is also graphical?
8.1.21. Let d = =
(d 1 , ..• , dn ), C (C1, ••• , cm) be two graphical sequences. Is it
true that the (n x m)-sequence of the sums di + Cj, i, j = 1,2, ... , n, is also
graphical?
8.1.22. Prove the validity of the I-procedure.

8.1.23. Use the I-procedure to test the graphicity of the following sequences. a)
(9 4 ,6 2 ,83,5 2 ), b) (93,83,6,5,4,2,1,01000), c) (8 7 ,5 6 ,3 11 ).
8.1.24. Prove that if a proper graphical n-sequence d is realized by a graph G that

has a subgraph H isomorphic to f{p then there exists a realization G' of d
with a subgraph H' isomorphic to f{p such that V H' is the set of vertices
with first largest degrees d 1 , ... , dp .
8.1.25. Let a sequence d satisfy the Erdos-Gallai inequalities for all k = 1, ... , k( d),
where k( d) = max{ i : di ?: i}. Prove that d is graphical.
8.1.26. Use the Erdos-Gallai theorem to test the graphicity of the following se-
quences.
a) (7 6 ,6 4 ,5 6 );
b) (123,11,92,82,62,42,3);
8.1.27. Find all nonnegative integers m, n for which the following sequences are
graphical.
a) (mn); b) (m,l n ); c) (mn,nm); d) ((2m)2n,(2m_l)2n-l).
8.1.28. For which values of n the sequence nn, (n - 1)n-l, ... ,2 2 ,1 1 is graphical?
8.1.29. Let nl, n2, ... , nk be natural numbers with the sum equal to n. Prove that
t he sequence (d nl
l ' dn2
nk - n - n". '1 -- 1
2 , ... , d k ) , d·, - , 2
, ...k . grap h'lca.
, , IS I
8.1.30. Prove that a sequence of nonnegative integers is realizable by a pseudo-

graph if and only if the sum of its elements is even.
8.1.31. Prove that a proper n-sequence is realizable by a multigraph if and only if
d2 + d3 + ... + dn 2: d 1 .
8.1.32. Let d be a proper n-sequence with dj > 0 for some j. Prove that the
sequences d and
are realizable or not realizable by a multigraph simultaneously.

8.1.33. The Ferrer diagram of a proper n-sequence d is the (0, I)-matrix of size
n x n constructed as follows. For all i, we set the first di elements (excluding
the i-th element) of the i-th row to one. The remaining elements are set to
zero.
a) Construct the Ferrer diagram for the sequence (4,4,3,2,2,1).
b) Let di, i = 1,2, ... , n, denote the sum of elements of the i-th column of
the Ferrer diagra.!? The sequence d = d1 , d2 , ••• , dn is said to be conjugate
to d. Prove that d = d.
c) Prove the Fulkerson theorem: a proper n-sequence d is graphical if and
only if
k k
"d· < "d·
L.Jz_~z
;=1 i=1
for all k, = 1,2, ... , n.

8.1.34. Let d be a proper graphical n-sequence and let dn +1 be an even number
such that dn 2: dn+1 2: O. Prove that the sequence d U (dn+d (see Exercise
8.1.6) is graphical.
8.1.35. Prove that there exist infinitely many graphical n-sequences for which the
inequality from Theorem 8.1.5 turns into the equality.
8.1.36. Prove that if a proper n-sequence d contains no zero elements and d 1 <
2Vn - no - 2 then it is graphical.
8.1.37. Consider the class K of sequences whose elements are the numbers between
1000 and 1500 and whose sums of elements are even. Find the minimal
number N such that a sequence d E K is graphical of and only if it length
is at least N.
8.1.38. Which of the following pairs are graphical?

a) (4 4,2), (5,4 2 ,3,1 2 ); b) (5 4 ,24), (7 2 ,6,5,3).
8.1.39. Propose an algorithm for testing the graphicity of pairs of sequences and
constructing a realization (if it exists). Is it possible to apply the I-procedure?
8.1.40. Let d = (an) and c = (b m ) be two sequences. For which positive integers
a, band m, n ;::: 1 the pair (d, c) is graphical?
8.1.41. Prove that an n-sequence d is realizable by a multigraph if and only if the

pair (d, c) is graphical, where c = (2 m ), m = (d 1 + ... + dn )/2.
8.1.42. Let G be a bipartite graph. A switching (ab, cd) is called bipartite if a, dare
in one part and b, c are in the other part. Using this notion, state and prove
the counterpart of the Switching Theorem for graphical pairs of sequences.
8.1.43. For which n, m the graphs a) Kn, b) Pn , c) en, d) Km,n are unigraphs?
8.1.44. Is the sequence (16 3 ,14,12,11,8 5 ,6 2 ,4 2 ,3 2 ) unigraphical?
8.1.45. Is it true that a graph is a unigraph if and only if it admits no switchings?
8.1.46. Prove that for n == 2 (mod 4) or n == 3 (mod 4) the number ofunigraphs

of order n is even.
8.1.47. Find all regular unigraphs of de!1;ree two.
8.1.48. Find all selfcomplementary unigraphs of orders n :::; 7.
8.1.49. Is the property "to be a unigraph" preserved under the transition to a

subgraph?
8.1.50. A set X of vertices of a graph G is called d-complete if deg u :f. deg v for
any u E X, v E VG\X. Prove that for every d-complete subset X of vertices
of a unigraph G the subgraph G(X) is a unigraph.
8.1.51. Let k1 , ... , k/ be positive integers with sum s. Prove that the sequence
is unigraphical.
8.1.52. Prove that a graph is a unigraph if and only if its complement is a unigraph.
8.1.53. Is it true that the join G + H of unigraphs is a unigraph?
8.1.54. Is the conjunction of the properties "to be a bipartite graph" and "to be
a unigraph" a hereditary property?
8.1.55. Prove that every disconnected unigraph without isolated vertices has no
cycles. Describe the components of such graph.
8.1.56. Which of the following statements are valid?

a) Every component of a unigraph is a unigraph.
b) The disjoint union of unigraphs is a unigraph.
8.1.57. An n-sequence is called l-unigraphical if it has exactly one labelled real-

ization (called an l-unigraph).
a) Give examples of l-unigraphical n-sequences for all n 2: 1.
b) Prove that the property "to be l-unigraphical" is hereditary and find a
minimal FIS set.
8.1.58. Prove that if an n-sequence d satisfies the conditions
then it is realizable by a single multigraph.
8.1.59. The degree of an edge is the unordered pair of degrees of its endvertices.
The edge sequence of a graph is the list of degrees of its edges.
a) Does there exist a graph with edge sequence
b) Prove that if a graph has no isolated vertices then it is possible to recon-

struct the degree sequence of the graph from its edge sequence.
c) Prove that a graph is not specified up to isomorphism by its edge sequence.
d) Construct an infinite series of graphs which are specified up to isomor-
phism by their edge sequences but which are not unigraphs.
8.1.60. For a vertex v let deg1(v) denote the list of degrees of the vertices of its
neighborhood. If deg v = 0 then we set deg 1(v) = 0.
Prove that the sequence (deg 1(v) : v E V G) uniquely specifies the edge
sequence.
8.2 P-graphical Sequences

Let P be a graph-theoretical property and let d be a graphical sequence. Let
T( d, P) denote the set of realizations of d with the property P and let T( d) denote
the set of all realizations of d. IfI( d, P) ::f 0 then d is called potentially P-graphical.
If T(d, P) = T(d) then d is called forcibly P-graphical.
Characterizations of forcibly and potentially P-graphical sequences are known
for many properties P. We give here several theorems.
Theorem 8.2.1 (G.Edmonds) A proper graphical n-sequence d is realizable by

a k-edge-connected graph, k::::: 2, if and only if dn ::::: k.
Theorem 8.2.2 (D. Wang, D.Kleitman) A proper graphical n-sequence d zs

realizable by a k-connected graph, k ::::: 2, if and only if dn ::::: k and
n k-1
L d; - 2 L d; + (k - 1)(k - 2) ::::: 2(n - k).
;=1 i=l
Theorem 8.2.3 (V. Chungphaisan) If a proper n-sequence d has a realization

with a Hamiltonian path starting at a vertex of degree d; then this realization is
produced by the I-procedure in which at the first iteration the leading vertex is a
vertex of degree d; and at all remaining iterations the leading vertex is the one
with the minimal label in the set S( v), where v is the vertex leading at the previous
iteration.
Theorem 8.2.4 (V. Chungphaisan) A proper n-sequence d without zeroes is re-

alizable by a Hamiltonian graph if and only if there exists a realization of d with a
Hamiltonian path starting at a vertex of degree d l .
A graphical sequence is called P-unigraphical if II(d, P)I = 1. Notice that the

set of P-unigraphical sequences contains all unigraphical sequences whose realiza-
tions have the property P.
* * *
8.2.1. Prove that a graphical sequence is forcibly P-graphical if and only if it is
not potentially P-graphical, where P is the negation of P.
8.2.2. Let PI, P2 be two graph-theoretical properties. Interpret the following

statements in terms of potentially and forcibly Pl- and P2-graphical sequen-
ces.
a) The sequence is forcibly (P l &P2)-graphical.
b) The sequence is potentially (PI &P2 )-graphical.
c) The sequence is forcibly (PI V P2)-graphical.
d) The sequence is potentially (PI V P2)-graphical.
B.2. P-graphical Sequences 127
8.2.3. Let a class of graphs with property P be contained in class of graphs with
property P'. How are the following pairs of classes related to each other?
a) Classes of potentially P-graphical and potentially pI-graphical sequences.
b) Classes of forcibly P-graphical and forcibly PI-graphical sequences.
8.2.4. Let A, C, and T be the properties of " being acyclic" , "being connected", and
"being tree" respectively. Let Px and Fx, X = A, C, T, denote respectively
the sets of potentially and forcibly X-graphical sequences. Prove that
a) FT = FA n Fe; a) PT = PA n Pe; a) FA n Pe = PA n Fe = FT.
8.2.5. For which properties P the sets of potentially and forcibly P-graphical se-
quences coincide?
8.2.6. Prove that for every graph-theoretical property P the set of all graphical
sequences is partitioned into three subsets:
1) forcibly P-graphical sequences;
2) forcibly P-graphical sequences;
3) the sequences that are simultaneously P- and P-graphical.
8.2.7. Design algorithms for recognition of graphical sequences which are realizable
by the complete k-partite graph in the following cases:
a) the number k is known;
b) the sizes n1, ... , nk of parts are known;
c) the number k is unknown.
8.2.8. Prove that a proper graphical n-sequence without zeroes is potentially con-
nected if and only if 2:7=1 di ~ 2(n - 1).
8.2.9. a) Prove that a proper n-sequence without zeroes is realizable by a tree if
and only if 2:7=1 di = 2(n - 1).
b) Let a sequence d be realizable by a tree. Design an algorithm for the
construction of a tree realizing d.
c) Find necessary and sufficient conditions under which the sequence is re-
alizable by a unique tree.
8.2.10. Find necessary and sufficient conditions under which the sequence is real-
izable by a graph with at least one cycle.
8.2.11. Test whether the sequence (7,6,5 2 ,4,3,23 ) is realizable by a Hamiltonian
graph.
8.2.12. A graph is called unicyclic if it is connected and has exactly one cycle.
Prove that a proper n-sequence d without zeroes is potentially unicyclic if
and only if n ~ 3, d3 > 1, and 2:7=1 di = 2n.
8.2.13. Find all graphical n-sequence d with d 1 = d 2 = n-1 ~ 3 that are realizable
by line graphs.
8.2.14. Let d be a potentially acyclic n-sequence.

a) Prove that d~ + ... + d; ~ n 2 - n.
b) Give an infinite series of examples for which this inequality turns into the
equality.
8.2.15. Let d be a proper potentially planar n-sequence. Prove that d1 + d2 + d3 ~
2n-2.
8.2.16. Characterize the forcibly bipartite sequences.
8.2.17. Prove that a proper n-sequence d without zeroes is forcibly realizable by
a tree only in the following two cases:
a) n = 2 and d 1 = d = 1; 2
b) n ~ 3, d3 = 1, and d1 + d2 = n.
8.2.18. Characterize the forcibly unicyclic sequences.
8.2.19. Is the sequence (6 2 ,3 6 ) a) forcibly planar; a) potentially planar?
8.2.20. Characterize proper forcibly outerplanar n-sequences d such that d 1
n -1.
8.2.21. Let d, c be two graphical sequences. We shall say that c is a hereditary
subsequence of d (and we shall denote this relation by c ~ d) if there exist
realizations G and H of d and c respectively such that H is isomorphic to
an induced subgraph of G.
a) Which of the three sequences d = (3 6 ), c = (34 ), b = (2,12) are related
by" ~"?
b) Prove that the relation " ~ " defines a partial order on the set of graphical
sequences.
8.2.22. Let P be a hereditary graph-theoretical property and let Im(P) be the
corresponding minimal FIS set. Let dm (P) denote the set of degree sequences
of graphs from Im(P) and let d~(P) denote the set of minimal elements of
dm(P) with respect to the partial order " ~ " (see Exercise 8.2.21).
a) Find a property P such that d~(P) =f. dm(P).
b) Prove that a graphical sequence d is forcibly P-graphical if and only if
c ~ d (see Exercise 8.2.21) for no sequence c E d~(P).
8.2.23. The following Rao Conjecture is known: for every hereditary property P
the set d~(P) is finite. Verify its validity for a) acyclic graphs; b) bipartite
graphs.
8.2.24. Let Up denote the set of unigraphical sequences whose realizations are
with property P and let Pu denote the set of P-unigraphical sequences.
a) prove that Up ~ Pu.
b) Find examples showing that these two sets may coincide, as well as be
different.
8.3. Split and Threshold Graphs 129
8.2.25. Let T denote the property "being tree". Characterize the T-unigraphical
sequences.
8.2.26. Suppose that a graph-theoretical property P is specified by the degree

sequence of a graph. Prove that Pu = Up (see Exercise 8.2.24). Is the
converse true?
8.2.27. Suppose that every graph that has a property P also has a property pl. Is
it true that Pu ~ Ph (see Exercise 8.2.24)?
8.3 Split and Threshold Graphs

This section considers two properties that may be determined from the degree
sequence of a graph: split ness and thresholdness.
A graph G is called split if there exists a partition VG = AU B, where A is
a clique and B is an independent set. A is called the upper part of G, and B is
called the lower part of G. One of these parts may be empty.
Theorem 8.3.1 (P.Hammer, B.Simeone) Let d be a proper graphical n-seq-

uence. A realization G of d is a split graph if and only if
m n
Ld = L di+ m(m-l),
i
i::l i::m+l
where m = m(d) = max{i: di 2': i-I}.

An interesting subclass of split graphs is the threshold graphs. Let the vertices
ofa graph G be indexed by the numbers 1,2, ... , n, i.e., VG = {I, 2, ... , n}. Let IG
denote the set of all independent sets of vertices of G plus the empty set. If there
exist nonnegative numbers all a2, ... , an, f3 such that the set of (0, 1}-solutions of
the inequality
(8.3.1)
is the set of characteristic vectors of the elements of I G then G is called a threshold
graph and (8.3.1) is called the separating inequality (see Exercise 3.1.37).
Let K 0 G and 0 0 G denote the graphs obtained from G by the addition of a
dominating and an isolated vertex respectively.
Theorem 8.3.2 A graph G is a threshold graph if and only if it is of the form

G = G 1 0 (G2 0 ( .. • oGn ) .. . ), where G n is Kl and for i = 1,2, ... , n -1 the terms
Gi are either K or O.
* * *
8.3.1. Prove that for n ~ 3 all graphs are split.
8.3.2. Is the graph shown at Fig. 8.3.1 split?

8.3.3. Give examples of pairs of nonisomorphic split graphs with the same degree
sequences.
8.3.4. Which of the following sequences are realizable by split graphs?

a) (8,7,6 2 ,5,4,3,2,1 3 ); b)(7,6,5 2 ,4,3 2 ,1 2 ).
8.3.5. Prove that the graph G is split if and only if its complement G is split.
8.3.6. Let A, B be the upper and lower parts of a split graph G. Prove that deg u ~
deg v for any u E A, v E B.
8.3.7. How many partitions into parts maya split graph of order n have?
8.3.8. Prove that if one realization of a graphical sequence is split then all its
realizations are split.
8.3.9. Prove that the splitness property is hereditary and find a corresponding
minimal FIS set.
8.3.10. Construct an algorithm for

a) recognition of split graphs by the degree sequence;
b) construction of all partitions of a split graph into parts.
8.3.11. Reduce the problem of recognition of isomorphism of arbitrary graphs to

the isomorphism problem for split graphs.
8.3.12. Is it true that every split selfcomplementary graph has parts of equal size?
8.3.13. Find all split trees.
8.3.14. Which regular graphs are split?
8.3.15. Prove that every threshold graph is split, but not conversely.
8.3.16. Which graphs of order four are threshold? Find their separating inequali-
ties.
B.3. Split and Threshold Graphs 131
8.3.17. Give a geometric interpretation of the separating inequality and of the

thresholdness of a graph.
8.3.18. Prove that if a graph G is threshold then G+Kl and GUK1 are threshold.
8.3.19. Prove that the vertices of a threshold graph may be numbered in such
a way that the closed neighborhood N(v) U {v} of any vertex v contains
neighborhoods of all other vertices with smaller numbers.
8.3.20. Design algorithms for recognition of
a) threshold graphs;
b) graphical sequences that have a threshold realization.
8.3.21. Prove that the property "to be threshold graph" is hereditary and find a
corresponding minimal FIS set.
8.3.22. Design an algorithm for the construction of a separating inequality for a
threshold graph.
8.3.23. Prove that the number of threshold graphs of order n is 2n - 1•
8.3.24. Which graphs without triangles are threshold?

8.3.25. Prove that
a) threshold graphs are unigraphs;
b) the l-unigraphs are the threshold graphs.
8.3.26. Find all selfcomplementary threshold graphs.
8.3.27. Prove that a graph G is threshold if and only if for every subset S ~ VG
there exists a vertex u E S such that it is either adjacent to all vertices of
S - {u}, or it is adjacent to none of them.
8.3.28. Prove that a graph G is threshold if and only if there exists a labelling of
the vertices c: V G --+ N and an integer t such that
(uv E EG) ¢=> (u # v, c(u) + c(v) > t).
8.3.29. Prove that a split graph is threshold if and only if it has no induced sub-
graphs isomorphic to P4.
8.3.30. Prove that a split graph G is threshold if and only if the vertices of its
parts may be ordered A = {Ul, ... ,u/}, B = {VI, ... ,vm } in such a way that
(UiVj E EG) ¢=> (UilVjl E EG, if = i, ... , I, jf = j, ... , m).
8.3.31. Let a graph G be represented as the union G = U!=l Gi of threshold

graphs with the same vertex set. This representation is called threshold
decomposition. Prove that
a) every graph admits a threshold decomposition;
b) there exists a threshold decomposition of G in which t :::; \VGI.
8.3.32. The minimal number of components over the threshold decompositions of

a graph G is called the threshold numberofG and is denoted by th(G). Prove
that
a) th( G) is the minimal value of t such that there exists a system of t in-
equalities
aUxl + al2X2 ~::. + alnXn :::; bl ,
{
attXI + at2 X 2 + ... + atnXn :::; bt ,

such that I E IG if and only if the characteristic vector of the set I is a
(0, I)-solution of the system;
b) if H is an induced sub graph of a graph G then th(H) :::; th(G).
8.3.33. Determine a) theGn), n > 3; b) th(Km,n); c) th(Pn).

8.3.34. Prove the following statements. (ao, aI, f30 are introduced in Sect. 3.1.)
a) th(G) + ao(G) :::; IGI for every nonempty graph G.
b) If a graph G is triangle-free then th(G) = f3o(G).
c) th(G) = al(G) for bipartite graphs.
8.3.35. Using the graphs shown at Fig. 8.3.2 prove that the equality th(G) = th(G)
is not always true.
8.4 Degree Sets and Arity Partitions

Let us define two more objects basing on the degree sequence, namely the degree set
(the set of degrees of vertices) and the arity partition (the list of the multiplicities
of the degrees).
Suppose that a graphical sequence d realizable by a graph G is written in the
form d = (ki l , k~2, ... , k~n), where ki "# k j for i "# j and Pi is the multiplicity of
ki. Then the set S = {k l , ... , k n } is called the degree set of the graph G and the
list F = (PI, ... , Pn) is called the arity partition of G. Accordingly, the graph Gis
called a realization of the set S and the list F. Notice that the arity partition, as
well as the degree sequence, is the mu/tiset, i.e., it is considered up to permutations
of the elements, and the elements may repeat.
8.4. Degree Sets and Arity Partitions 133
* * *
8.4.1. Construct realizations of the sets {4}, {4, 2}, {4, 1, a} with minimal numbers
of vertices.
8.4.2. Prove that
a) any finite set 5 of positive integers is the degree set of some graph;
b) the minimal order of a realization of a set 5 is kl + 1, where kl is the
maximal element of 5.
8.4.3. Prove that the claim of the previous exercise remains valid
a) in the class of connected graphs;
b) in the class of threshold graphs.
8.4.4. Let S = {k 1 , ... , k n } be a set of positive integers such that kl > ... > k n .
a) Prove that if kn = 1 then 5 is realizable by a tree.
b) Find the minimal order of such tree.
c) Design an algorithm for the construction of a tree realization of minimal
order.
8.4.5. Prove that two threshold graphs without isolated vertices are isomorphic if
and only if their degree sets coincide.
8.4.6. Does there exist a graph which is uniquely specified by its degree set?
8.4.7. Let 5 = {k 1, ... , k n } be a set of integers such that kl > ... > k n ~ 3. Prove
that there exists a Hamiltonian graph of order k1 + 1 realizing 5.
8.4.8. Find necessary and sufficient conditions under which a degree set 5 is real-
ized by a planar graph.
8.4.9. Suppose that k1 > k2 and let J.Lp(k 1 , k 2) denote the minimal number of
vertices among planar realizations of the degree set {k 1 , k2 }. Prove that
a)J.Lp(k1, k 2) = k1 + 1 if 1:::; k2 :::; 3,
b) J.Lp(k1, k2) = k1 + 2 if k2 = 4,
c) J.Lp(k 1 ,k2):::; 2k1 + 2 if k2 = 5.
8.4.10. For which positive integers k, n there exists a graph of order n realizing
the degree set {I, 2k }?
8.4.11. Is it true that every set of nonnegative integers is realizable by a bipartite
graph?
8.4.12. Let 5 1 ,52 be two sets of positive integers. Prove that
a) there exists a bipartite graph such that 5 1,52 are the degree sets of its
parts;
b) if 51 = =
{I}, 52 {k 1 , ... , kn } then the minimal order of graphs from item
a is at most nl + 2:7=1 k i .
8.4.13. A graph with the degree set S = {k l , ... , kn } and girth g is called a
(k 1 , ... , k n ; g)-eell.
Construct a) a (3,4; 5)-cell of order 13; b) a (3,4; 6)-cell of order 18.
8.4.14. A graph G is called homotraceable if for any its vertex v there exists a
Hamiltonian path starting at v. Construct homotraceable non-Hamiltonian
graphs with degree sets a) {2, 3}; b) {2}; c) {3}.
8.4.15. For n ~ 3 find arity partitions of the graphs a) Pn ; b) en; c) Km,n'
8.4.16. Let d be a graphical sequence and let S, F be the corresponding degree set
and arity partition respectively. Is it true that the set of realizations of d is
the intersection of the sets of realizations of Sand F?
8.4.17. Prove that
a) a graph and its complement have equal arity partitions;
b) there are no graphs with arity partition (1,1, ... ,1).
8.4.18. Find necessary and sufficient conditions under which a list (PI, .. . ,Pn) of
positive integers is the arity partition of a connected graph.
8.4.19. Is it true that for any n ~ 2 there exists a graph of order n that is uniquely
specified by its arity partition?
8.4.20. Is it true that every list (PI, ... , Pn) f. (1 n) of positive integers is the arity
partition of a threshold graph?
Chapter 9
Graph Colorings
9.1 Vertex Coloring

A proper coloring of a graph Gin k colors, or k-coloring, is a function f: VG-+
{I, 2, ... , k} such that f( u) i- f( v) for any adjacent vertices u, v of G. The number
f( u) is called the color of the vertex u. A graph that has a proper k-coloring is
called k-colorable.
Notice that a k-coloring does not necessarily use all k colors. The set of vertices
of the same color is called color class.
The minimal k such that a graph G is k-colorable is called the chromatic
number of G and is denoted by X(G). If X(G) = k then G is called k-chromatic.
A proper X( G)-coloring of a graph G is called its minimal coloring.
Clearly, a graph is I-colorable if and only if it is empty, and it is 2-colorable
(bicolorable) if and only if it is bipartite.
Theorem 9.1.1 (A.Korshunov) The chromatic number of almost all graphs G
of order n satisfies the relation X( G) ,...- 21 n .
og2 n
Theorem 9.1.2 For every graph G, X(G) :::; 1 + max6(H), where the maximum
is over all induced subgraphs H of G.
Corollary 9.1.3 For every graph G, X(G) :::; 1 + Ll(G).

Theorem 9.1.4 (R.Brooks) If G is a non-complete connected graph with
Ll(G) ~ 3 then X(G) :::; Ll(G).
Theorem 9.1.5 (E.Nordhouse, J.Gaddum) For an n-vertex graph G and its
complement the following inequalities hold:
2vn:::;x+x:::;n+l, n:::; XX:::; (n+l)2j4,

where X = X(G), X = X(G).
A. Zykov proved that there exist triangle-free graphs with arbitrarily large
chromatic number. P.Erdos generalized this result as follows.
135
136 9. Graph Colorings
Theorem 9.1.6 For any positive integers g, X there exists a x-chromatic graph
with girth at least g.
Recall that the girth is the length of the shortest cycle in a graph.
A labelled graph G is called uniquely k-colorable if x( G) = k and all minimal
colorings induce the same partition of the vertex set into color classes up to the
numbering of colors.
Theorem 9.1. 7 Every uniquely k-colorable graph is (k - I)-connected.
Theorem 9.1.8 For every k 2 3 there exists a uniquely k-colorable graph without
complete subgraphs of order k.
A graph G is called critical if X( G - v) < X( G) for any its vertex v. If in addition

X(G) = k then G is called k-critical.
Characterizations of k-critical graphs for k 2 4 are unknown. Only some their
properties are studied.
The circumference of a graph is the length of the longest cycle of the graph.
Theorem 9.1.9 (H.Dirac) If G is a k-critical graph, k 2 3, then either G is

Hamiltonian or its circumference at least 2k - 2.
A graph G is called edge-critical if X(G - e) < X(G) for any its edge e. If in
addition X(G) = k then G is called k-edge-critical.
Theorem 9.1.10 If a k-edge-critical (n, m)-graph G is not complete, and k 2 4,

then 2m 2 (k - l)n + k - 3.
* * *
9.1.1. Consider the following algorithm of sequential coloring of the vertices of a
graph.
1. An arbitrary vertex Vl obtains color l.
2. If the vertices Vl, V2, .•. ,Vi are already colored then an arbitrary next
vertex Vi+l obtains the minimal color not used for the vertices of its
neighborhood.
a) Prove that the resulting coloring is proper.
b) Prove that a sequential coloring of a complete k-partite graph is minimal.
c) Give an example showing that the difference between the number of colors
in a sequential coloring and the chromatic number may be arbitrarily large.
9.1.2. Consider one more algorithm of coloring of the vertices of a graph.
1. Set G 1 = G.
2. If the graph G k is already constructed, select a maximal cardinality
independent vertex set Ik in it and color the vertices of Ik in color k.
9.1. Vertex Coloring 137
3. Set GA:+1 = GA: - lA:.

4. Repeat the two last steps until all vertices become colored.
a) Prove that this algorithm does not necessarily produce a minimal coloring.
b) Construct a sequence of bichromatic graphs H 1, H 2, ... such that the
described algorithm applied to Hi produces a coloring of the graph with at
least i colors.
9.1.3. Prove that in some minimal coloring of a graph G the neighborhood N (v)
of every vertex v is colored in the same color if and only if G is bipartite.
9.1.4. Find the chromatic numbers of graphs a) /{n; b) Kn,m; c) Pn ; d) en; e)

Petersen graph.
9.1.5. Prove that if every block of a graph is k-colorable then the graph is k-col-
orable.
9.1.6. Prove that the difference b.(G) - X(G) may be arbitrarily large.
9.1.7. Find all counterexamples to the proposition: every connected graph G con-
tains a vertex v such that deg v ~ X(G).
9.1.8. Prove that for every minimal coloring of the vertices of a connected graph
which is neither a complete graph nor C3 A: the graph has two vertices of the
same color at distance 2.
9.1. 9. Prove that for any two different colors i, j of any minimal coloring of the
vertices of a graph G there exists an edge uv E EG such that u has color i
and v has color j.
9.1.10. A chess piece graph is a graph whose vertices are the squares of the chess-
board and the edges are the pairs of squares linked by a move of the given
piece.
a) Find chromatic numbers for the castle, the knight, the bishop, and the
king graphs.
b) Prove that X(GQ) ~ 8 for the queen graph GQ.
c) Extend the problem a) to the (n x n)-chessboard case.
9.1.11. Prove that in every proper coloring of a line graph every vertex is adjacent
to at most two vertices of the same color.
9.1.12. Prove that the octahedron graph L(K4) cannot be properly colored in four
colors in such a way that every two-colored subgraph has no cycles.
9.1.13. Let a graph G have a proper coloring in which every color is used at least
twice. Prove that G has a coloring in X(G) colors with the same property.
2
9.1.14. Prove that X(G) ~ 2 n for an (n, m)-graph G.
n -2m
138 9. Gra.ph Colorings
9.1.15. Prove that X(G) 2:: n ~ r for an r-regular graph G of order n.
9.1.16. Prove that the number of edges of a graph G is at least X(G)(X(G) -1)/2.
9.1.17. Let (dl," .,dn ) be the degree sequence ofagraph G, dl ~ ... ~ dn . Prove
that
X(G) < max min{i, di + I}.
-19::;n
9.1.18. a) Prove that nlao ~ X(G) ~ n-ao+1 for every graph G, where n = IGI,
ao is the independence number of G.
b) Characterize the graphs with X( G) = n - aD + 1.
9.1.19. Is it true that for any positive integers n, aD, X that satisfy the inequalities
from the Exercise 9.1.18a there exists an n-vertex graph with independence
number aD and chromatic number X?
9.1.20. Prove that the chromatic number X(G) is equal to the minimal number m
such that ao(G x Km) = IG/, where aD denotes the independence number
and G I x G 2 denotes the Cartesian product of graphs defined in Sect. 1.4.
9.1.21. Is it true that if degv < X(G) - 1 for some vertex v of a graph G then
X(G) = X(G - v)?
9.1.22. Prove that X(G I +G 2 ) = X(Gt)+X(G 2 ) holds for all graphs G I , G 2 , where
G I + G 2 denotes the join of graphs defined in Sect. 1.4.
9.1.23. a) Prove that X(G I U G 2 ) ~ x(Gdx(G 2 ) even if G I , G 2 have common

vertices.
b) Give an example when X(G I U G2 ) = X(G I )x(G 2 ).
c) Express the chromatic number X(G I U G 2 ) for the disjoint union of G I
and G 2 in terms of X(Gt) and X(G 2 ).
9.1.24. Prove the Nordhouse-Gaddum Theorem 9.1.5.
9.1.25. Construct infinite series of graphs such that

a) X + X = n + 1 and XX = n;
b) X + X = 2vn and XX = n;
c) X + X = n + 1 and XX = (n + 1)2/4,
where X, X are the chromatic numbers of a graph and its complement.
9.1.26. Prove that there are no nontrivial graphs such that X +X= 2vn and
XX = (n + 1)2/4.
9.1.27. The merging of two nonadjacent vertices of a graph into one is called
an elementary homomorphism of the graph. Prove that X( G) ~ X( fG) ~
X(G) + 1 for any graph G and for any elementary homomorphism f of G.
9.1.28. The merging of two adjacent vertices of a graph into one is called an
elementary contraction of the graph.
a) Prove that Ix(G) - x(vG)1 ~ 1 for any graph G and for any elementary
contraction v of G.
b) Give examples of graphs for all three possible values of x( G) - x( vG) for
an elementary contraction v.

a) For every nonempty graph G there exists a partition VG = V1 U V2 such
that X(G(Vl» + X(G(V2» = X(G).
b) For every non-complete graph G there exists a partition V G = V1 U V2
such that X(G(V1 » + X(G(V2» > X(G).
9.1.30. Let a graph G admit a partition VG = V1 U ... U Vk such that for any i,j,
1 ~ i < j ~ k, there exist nonadjacent vertices x E Vi, y E Vj. Prove that
x( G) ~ IGI - k + 1.
9.1.31. Prove that X(G) ~ 1+ 1, where I is the length of the longest simple path
ofG.
9.1.32. Construct 4-regular 4-chromatic graphs with girth a) 4; b) 5.
9.1.33. Which graphs shown at Fig. 9.1.1 are uniquely 3-colorable?
9.1.34. Is it true that the union of any I color classes of a uniquely k-colorable
graph, 1 ~ I ~ k, induces a uniquely I-colorable graph?
9.1.35. Construct a uniquely 3-colorable graph without triangles.
9.1.36. Prove that if G is a uniquely k-colorable graph then 8( G) ~ k - 1.
9.1.37. Prove that for k-coloring of a uniquely k-colorable graph the subgraph
induced by the union of any two color classes is connected.
9.1.38. Prove that a uniquely k-colorable graph of order n has at least (k-l)n-Cf
edges.
9.1.39. Prove that if a graph G is connected and x( G) > k then G*Kk is a uniquely
k-colorable graph, where" *" denotes here the weak product of graphs defined
as follows: V(G * H) = VG x V H, E(G * H) = {((UI' U2), (VI, V2)) : UIVI E
EG, U2V2 E EH}.
9.1.40. The merging of G 1 and G 2 by a common subgraph K is the operation

producing a graph H = G~ UG~ such that G~ ~ G 1 , G~ ~ G 2 , and H(VGi n
VG~) ~ K. The set of k-trees is defined inductively.
1. Kk+1 is a k-tree.
2. If G is a k-tree then G' obtained by merging of G and Kk+1 by a
complete subgraph of order k is also a k-tree.
Is it true that the k-trees are uniquely (k + I)-colorable?

9.1.41. Prove that if X(G - u - v) = X(G) - 2 for every pair u, v of vertices of a
graph G, then G is a complete graph.
9.1.42. Prove that X(G - v) = X(G) - 1 for every vertex v of a critical graph G.
9.1.43. Find all 2-critical and 3-critical graphs.
9.1.44. Prove that critical graphs other than K2 are biconnected.
9.1.45. Prove that 8(G) ~ k -1 for a k-critical graph G.
9.1.46. Prove that X(G - I) = X(G) - 1 for any independent set I of vertices of a
critical graph G.
9.1.47. Is it true that every k-chromatic graph, k ~ 2, contains a k-critical sub-

graph?
9.1.48. Prove that the join G 1 + G 2 is a critical graph if and only if both G 1 and
G 2 are critical graphs.
9.1.49. Let u, v be two nonadjacent vertices of a non-complete k-critical graph G.

Prove that there exists a k-coloring of G such that the vertices u, v belong
to a) the same color class; b) different color classes.
9.1.50. Which of the following statements are valid?

a) Every edge-critical graph is critical.
b) Every critical graph is edge-critical.
9.1.51. Find all 3-edge-critical graphs.
9.1.52. Is it true that the join (defined in Sect. 1.4) of a k1- and a k2-edge-critical
graph is a (k1 + k2 )-edge-critical graph?
9.2. Chromatic Polynomial 141
9.1.53. Construct 6-edge-critical graphs of order 4n - 2, n 2: 3 with 8(G) > 2n.

9.1.54. Let G be an edge-critical graph. Let us construct a graph G' as follows.
For every vertex v E V G we introduce a new vertex v' and connect it with all
vertices from N(v). Then we introduce one more vertex wand connect it with
all vertices of type v'. Prove that G' is edge-critical and x( G') = x( G) + 1.
9.2 Chromatic Polynomial

Two proper t-colorings <p: VG -> {I, 2, ... , t} and e: VG -> {I, 2, ... , t} of a
labelled graph G are considered to be different if there exists a vertex v E V G
such that <p( v) =f. e( v). The number of different proper t-colorings of G is denoted
by f(G, t) and is called the chromatic function of graph G. Clearly, if t < X(G)
then f(G,t) = O.
The chromatic function of the complete graph Kn is easily computable:
f(Kn,t) ={ t(t-l) ... (t-n+lJ: if t 2: n;

otherwise.
Theorem 9.2.1 Let u, v be nonadjacent vertices of a graph G. If G1 = G + uv

and G 2 is produced by merging u and v then f(G, t) = f(G 1 , t) + f(G 2 , t).
Since IEGtI > IEGI and IVG 2 1 < IVGI, multiple application of Theorem 9.2.1
eventually reduces the computation of f( G, t) to the computation of chromatic
polynomials of complete graphs (the technique due to Zykov). Noticing that
f(I<n, t) is a polynomial in t, we have a corollary.
Corollary 9.2.2 The chromatic function f( G, t) of a graph G is a polynomial in

t.
Therefore f(G, t) is usually called the chromatic polynomial of G.
Theorem 9.2.3 IfG is an (n, m)-graph with k connected components Gl, ... , Gk
then
(i) f( G, t) is of degree n;
(ii) the coefficient at t n in f(G,t) is 1;
(iii) the coefficient at t n - 1 in f(G, t) is -m;
(iv) f( G, t) = I1~=1 f( Gi, t);
(v) aktk is the lowest term of f(G, t) with nonzero coefficient; in particular,
the constant term of f(G, t) is zero.
A graph is called chromatically unique if it is specified by its chromatic poly-

nomial up to isomorphism.
* * *
9.2.1. Find the chromatic polynomials of the graphs shown at Fig. 9.2.1.
9.2.2. Prove that

a) f(I{n,i) = t(t - 1) .. . (t - n + 1), ift ?: n;
b)f(On, t) = tn.
9.2.3. Find the chromatic polynomials of a) C n , n ?: 3; b) the wheel Wn = J{1 +Cn ,
n 2: 3.
9.2.4. Does there exist a graph with the chromatic polynomial a) t 4 - 3t 3 + 3t 2?
b) t 4 - 3t 3 + 2t2?
9.2.5. Prove that if G 1 , . .. , Gk are the connected components of a graph G then
f(G, t) = TI7=1 f(Gi, t).
9.2.6. Let a graph G be obtained from graphs G 1 , G 2 by merging by a common
subgraph J{n (as defined in the Exercise 9.1.40). Express leG, t) in terms of
f(G 1 ,t), f(G 2 ,t).
9.2.7. Prove the Theorem 9.2.1
9.2.8. Prove that a graph G of order n is a tree if and only if f(G, t) = t(t _1)n-l.
9.2.9. Prove that f(G, t) s; t(t _1)n-1 for any n-vertex graph G and any positive
integer t.
9.2.10. Prove that the maximal real root of a chromatic polynomial of order n is
at most n - 1.
9.2.11. Prove that if for a vertex v of a graph G the neighborhood N(v) induces
the complete subgraph then f( G, t) = (t - deg v)f( G - v, t).
9.3 Edge Coloring

An edge coloring of a graph Gin k colors is a function <p : EG -+ {I, 2, ... , k} such
that adjacent edges are colored differently. The number <pC e) is called the color of
the edge e. An edge k-coloring of a graph uses at most k colors. The chromatic
index X' = X' (G) of a graph G is the minimal k such that G has an edge k-coloring.
Other notions concerning edge colorings (edge color class, minimal edge coloring,
etc.) are defined similarly to the vertex-related case.
9.3. Edge Coloring 143
Theorem 9.3.1 (V.Vizing) For every graph G, 6.(G) :S X'(G) :S 6.(G) + 1.
* * *
9.3.1. Find all minimal edge colorings of graphs shown at Fig. 9.3.l. (Colorings
differ up to renaming the colors.)
9.3.2. Evaluate a) X'(Cn), n 2: 3; b) X'(Kn), n 2: 2.
9.3.3. Design an algorithm constructing a minimal edge coloring of Kn.
9.3.4. Fig. 9.3.2 shows two edge colorings of a graph G with X'(G) = 4 colors.
Is it possible to convert one coloring into another by means of recoloring of
maximal under inclusion bicolored paths?
9.3.5. Describe the graphs with X' :S 2.

9.3.6. Are the following equalities valid?

a) X(G) = X'(L(G». b) X'(G) = X(L(G».
9.3.7. Design an algorithm for finding a minimal edge coloring of a tree.
9.3.S. Prove that X'(G) = .6.(G) for a bipartite graph G.

9.3.9. Let G be a connected graph which is not a simple cycle of odd length. Prove
that X'(G) = .6.(G) if the lengths of all simple cycles of G are of the same
parity.
9.3.10. Let a cubic graph G have exactly one edge coloring with X'(G) colors.
Prove that G has three Hamiltonian cycles and X' (G) = 3.
9.3.11. Prove that X'(G) = .6.(G) + 1 for nonempty regular graphs of odd order.
9.3.12. A graph G is called isochromatic if X(G) =
X'(G). Find all isochromatic
graphs among a) complete graphs; b) bipartite graphs.
9.3.13. Find a cubic graph G of minimal order with X'(G) = 4.
9.3.14. A graph is called uniquely edge-colorable if all its minimal edge colorings in-
duce the same partition of the edge set. Construct a uniquely edge-colorable
graph with X' (G) = .6.( G) + 1.
9.3.15. We shall say that a graph G is of class 1 if X'(G) = .6.(G) and of class 2 if
X'(G) = .6.(G) + 1. A graph is called x'-critical if it is connected, belongs to
class 2 and the deletion of any edge decreases the chromatic index.
a) Construct a x'-critical graph G of fifth order with .6.(G) = 3.
b) For what values of n Kn is x'-critical?
c) Prove that there are no x'-critical graphs of orders 4 or 6 with .6.(G) = 3.
9.3.16. If in the definition of the x'-critical graph we replace edges by vertices, we
obtain the definition of the x'-vertex-critical graph.
a) Is it true that every x'-vertex-critical graph is critical?
b) For which n the graph Kn is x'--vertex-critical?
c) Is it true that if a graph G is x'-vertex-critical then X' (G - v) = X' (G) - 1
for any v E VG?
9.4 Colorings of Planar Graphs

A connected plane graph without bridges is called a planar map. Two faces of a
map are said to be adjacent if they have at least one edge in common. A k-coloring
of a map is a function f from the set of faces of the map into the set {I, 2, ... , k}
of colors such that the colors of adjacent faces are different. A map is called
k-colorable if it has a k-coloring.
The following conjecture has been known for more than 100 years.
9.4. Colorings of Planar Graphs 145
Four-Color Conjecture: Every planar map is four-colorable.

The following theorem allows to cast it in terms of graphs.
Theorem 9.4.1 A map G is k-colorable if and only if its geometric dual graph is
vertex-k-colorable.
In 1890 P.Heawood proved the" Five-Color Conjecture":
Theorem 9.4.2 Every planar graph is 5-colorable.
In 1976 K.Appel and W.Haken proved the Four-Color Conjecture with the help
of the computer.
Below is a generalization of the Four-Color Conjecture.
Hadwiger Conjecture, 1943: Every connected n-chromatic graph is con-
tractible to Kn.
The notions of a planar map and its coloring are naturally extendable to other
surfaces. The chromatic number x(S) of a surface S is the number maxX(G),
where the maximum is over all graphs embeddable onto S.
Theorem 9.4.3 (Ringel, Youngs) If Sp is the sphere with p > 0 handles then
x(Sp) = l(7 + Jl + 48p)/2J .
For p = 0 this formula amounts to the Four-Color Conjecture.
* * *
9.4.1. What is the minimal number of colors to color the map shown at Fig. 9A.1?
Figure 904:1: To Exercise 904.1
9.4.2. Construct a map with minimal number of faces that cannot be colored in
less than four colors.
9.4.3. Let A be an arrangement of nonoverlapping but possibly tangent disks of

equal radius in the plane. Let X(A) denote the minimal number of colors
to color the disks in such a way that tangent disks are colored differently.
Prove that X(A) ::; 4 for all A.
9.4.4. Prove that a map produced by the subdivision of the plane by a set of
arbitrary circles is colorable in two colors.
9.4.5. Does the statement of the previous exercise remain valid if we replace circles
by straight lines?
9.4.6. Prove that a planar map admits a coloring in two colors if and only if the
degree of every vertex is even.
9.4.7. Prove that every outerplanar graph is tricolorable.
9.4.8. Consider an arrangement of straight lines in the plane such that no three
lines pass through the same line. Assume that the intersection points are the
vertices of a graph whose edges are the line segments between the neighboring
intersection points. Prove that the resulting plane graph is tricolorable.
9.4.9. Prove that for a planar graph G with 6( G) < .6.( G) = 5 the chromatic
number is at most four.
9.4.10. Prove that a Hamiltonian planar map is 4-colorable.
9.4.11. Prove that the Four-Color Conjecture is true for planar cubic graphs.
9.4.12. Prove that it is sufficient to prove the Four-Color Conjecture for planar
bridgeless cubic maps.
9.4.13. Prove that a planar 3-regular 2-connected map is 4-colorable if and only if
its chromatic index is 3.
9.4.14. Prove that every uniquely four-colorable planar graph is a maximal planar
graph.
9.4.15. Verify the validity of the Hadwiger conjecture for n = 1,2,3.
9.4.16. Prove that the Hadwiger conjecture for n = 5 implies the Four-Color Con-
jecture.
Note: The converse is also true (the Wagner Theorem).
9.4.17. Does Theorem 9.4.1 remain valid for an arbitrary surface?
9.4.18. Prove that the chromatic number of the sphere is equal to the chromatic
number of the plane.
9.4.19. Prove that X(T) ~ 7 for the torus T.

9.5. Perfect Graphs 147
9.5 Perfect Graphs

A graph G is called x-perfect if X(H) = <p(H) for every induced subgraph H of
G, where <p denotes the toughness.
Similarly, G is called a-perfect if ao( G) = c(H) for every induced subgraph H
of G, where c denotes the clique covering number.
Theorem 9.5.1 (L.Lovasz, D.Fulkerson) The following statements are equiv-

alent for a graph G :
1. G is x-perfect;
2. G is a-perfect;
3. Every nonempty induced subgraph H of G has an independent set I such that
<p(H - 1) < <p(H).
By virtue of the above theorem we may simply speak about perfect graphs.
Corollary 9.5.2 If a graph is perfect then its complement is also perfect.
The most famous conjecture about perfect graphs is the following one.
Strong Berge Conjecture: A graph is perfect if and only if neither it nor
its complement contain induced subgraphs C 2 k+l, k 2:: 2.
A graph G is called triangulated if it has no induced subgraphs C k , k 2:: 4.
Theorem 9.5.3 Every triangulated graph is perfect.
Theorem 9.5.4 A connected graph is triangulated if and only if every its minimal
under inclusion separating set of vertices is a clique.
The following theorem is a tool for the analysis oftriangulated graphs. A vertex
of a graph is called simplicial if its neighborhood induces a complete graph.
Theorem 9.5.5 Every triangulated graph has at least two simplicial vertices.
Let the vertices of a graph G be ordered as follows: vo, VI, ... , V n - l . Let us
construct the sequence of sets
Yo=VG, Yi=Yi-l\{Vi-d, i=1,2, ... ,n-1.
A dismantling of a graph G is a sequence of graphs obtained form G by a

successive deletion of the vertices vo, VI,"" V n -l. A dismantling is called perfect
if every vertex Vi is simplicial in the subgraph G(Y;). A dismantling is called normal
if every vertex Vi is simplicial either in G(Y;) or in its complement.
It is known that a triangulated graph admits a perfect dismantling. A graph
admitting a normal dismantling is called quasitriangulated.
* * *
9.5.1. Is the graph shown at Fig. 9.5.1 perfect?
9.5.2. Verify directly that the graphs C 2k+l and C 2 k+1, n ~ 2, are not perfect.
9.5.3. Prove that the join (see Sect. 1.4) of perfect graphs is perfect.
9.5.4. Let G, H be two perfect graphs whose intersection is a complete graph.

Prove that G U H is perfect.
9.5.5. Prove that the following graphs are perfect: a) bipartite graphs; b) split
graphs.
9.5.6. Prove that X(G) = <p(G) for any bipartite graph G.
9.5.7. Characterize the set of triangle-free perfect graphs.
9.5.8. Let (P,~) be a finite partially ordered set. A comparability graph G is the
graph with VG = P such that x, y E VG are adjacent if and only if either
x ~ y or y ~ x.
a) Prove that X(G) = <p(G).
b) Is a comparability graph perfect?
9.5.9. The strong product G l · G 2 of graphs G l , G 2 is the graph with V(G l · G 2 ) =

VG l x VG 2 and (( Ul ,U2), (Vl, V2» E E( G l . G2 ) if and only if
- either Ul Vl E EG l and U2V2 E EG 2 , or

- Ul = Vl and U2V2 E EG 2 , or
- UlVl E EG l and U2 = V2.
a) Prove that if X(G i ) = <p(Gi), i = 1,2, then X(G l . G 2 ) = <p(G l . G 2 ).

b) Is it true that the strong product of perfect graphs is perfect?
9.5.10. Prove that a graph G is perfect if and only if every its induced subgraph
H has an independent vertex set I which intersects all maximum cardinality
cliques of H.
9.5.11. Let I = {h, ... , In} be a collection of closed segments of the real line.
The interval graph of the collection I is the intersection graph of I, i.e., its
vertices bijectively correspond to the segments from I and two vertices are
adjacent if and only if the corresponding segments intersect.
a) Prove that the property "to be an interval graph" is hereditary.
b) Prove that interval graphs are perfect.
9.5.12. Which of graphs shown at Fig. 9.5.2 are triangulated?
G H
9.5.13. Is it true that a graph G is triangulated if and only if every its induced
subgraph has a simplicial vertex?
9.5.14. a) Is it true that every plane triangulation is a triangulated graph?

b) Is it true that every plane triangulated graph is a plane triangulation?
9.5.15. Under what conditions a bipartite graph is triangulated?
9.5.16. Prove that every split graph is triangulated.
9.5.17. Is it true that a complement of a triangulated graph is triangulated?
9.5.1S. Prove that the join G 1 + G 2 is a triangulated graph if both G 1 and G 2 are
triangulated graphs.
9.5.20. Prove that the interval graphs are triangulated, but not conversely.
9.5.21. The Vizing- Wi If number w(G) of a graph G is the value w(G) = max8(H),
where the maximum is over all induced subgraphs H of G. Prove that a graph
G is triangulated if and only if w(H) = <.p(H) -1 for every induced subgraph
H ofG.
9.5.22. Let T 1 , ... , T k , k ?:: 2, be a family of subtrees of a tree T. Prove that

a) if every two subtrees of the family have a common vertex then all subtrees
of the family have a common vertex;
b) the intersection graph of the family of subtrees is a triangulated graph.
9.5.23. Prove that the comparability graph (see Exercise 9.5.8) has no induced
cycles C 2n + 1 , n?:: 2.
9.5.24. Prove that a graph is triangulated if and only if it admits a perfect dis-
mantling.
9.5.25. Prove that every simplicial vertex of a triangulated graph may be the first
vertex of a perfect dismantling.
9.5.26. Prove that every subgraph of a quasitriangulated graph is quasitriangu-

lated.
9.5.27. Find a graph which is not quasitriangulated but every its component is
quasitriangulated.
9.5.28. Prove that every quasitriangulated graph is perfect.

Chapter 10
Directed Graphs
10.1 Directed Graphs: Basic Notions

Let V be a finite nonempty set and let V 2 be its Cartesian square. A directed
graph, or digraph is a pair (V, A), where A ~ V2. The elements of V are called
the vertices of the digraph, and the elements of A are called its arcs. The sets
of vertices and arcs of a digraph G are denoted by VG and AG respectively. The
number WGI is called the order of the digraph G and is also denoted by IGI. If
= =
IGI nand IAGI m then G is said to be an (n, m)-digraph.
If x = (u, v) is an arc then the vertices u, v are called the endvertices, or ends
of the arc; u is the tail of the arc, v is the head of the arc. It is said that an arc is
incident to its ends. It is also said that an arc outgoes from its tail and ingoes into
its head. An arc with coinciding head and tail, i.e., of kind (u, u), is called a loop.
One may consider digraphs (multidigraphs) that have several arcs with common
head and common tail. Such arcs are called multiple, or parallel.
In drawings, an arc is depicted by an arrowed line pointing from the tail and
to the head. An example digraph is shown at Fig. 10.1.1. At this figure, X7, Xs
are parallel arcs and Xg is a loop.
Two vertices of a digraph are called adjacent if they are the ends of the same
arc. Arcs are called adjacent if they have a common endvertex.
The digraph (V, 0) is called empty. The digraph (V, V2) is called complete.
The digraph G obtained from a digraph G by reversing the directions of all
arcs is called the digraph opposite to G.
The set of heads of the arcs outgoing from a vertex v is denoted by rv. The
set of tails of the arcs in going into a vertex v is denoted by r-1v. For any subset
W ~ VG we denote
rw = U rw, r-1w = U r-1w.

wEW wEW
The outdegree d+ (v) of a vertex v is the number of arcs outgoing from it. If
there are no parallel arcs, d+(v) = Irvl.
151
152 10. Directed Graphs
:l ••
(i ;rlO 7
Figure 10.1.1: Example digraph
The indegree d- (v) of a vertex v is the number of arcs ingoing into it. If there
are no parallel arcs, d- (v) = Ir-lvl. The degree deg v of a vertex v of a digraph is
the number of arcs incident to it, i.e., deg v = d+ (v) + d- (v).
Theorem 10.1.1 The sum of in degrees of the vertices of a digraph G is equal to

the sum of outdegrees and is equal to the number of arcs of G :
L: d+(v) = L: d-(v) = IAGI·

vEVG vEVG
Many graph-theoretical notions, such as isomorphism, connectivity, may be

naturally extended to digraphs.
* * *
10.1.1. Find the maximal number of arcs in a digraph of order n without parallel
arcs.
10.1.2. Prove that for any integer numbers n, m, 0 < m < n 2 there exists an
(n, m)-digraph without parallel arcs.
10.1.3. Find the number of different labelled digraphs of order n without parallel
arcs.
10.1.4. For every positive integer n find a connected digraph of order n isomorphic
to the opposite digraph.
10.1.5. Prove that the automorphism groups of a digraph and of the opposite
digraph coincide.
10.1.6. Find the sets rVi and r-1Vi for the vertices Vl, V2, V3 for the digraphs
shown at Fig. 10.1.2. Find also the sets r(Vl, V2), r-l(Vl, V2), r(Vl' V2, V3),
r- 1 (Vl,V2,V3) for them.
10.1. Directed Graphs: Basic Notions 153
v]
ISJ
V2 'IJ:I VI V2
•
Vs E)
1'4 115 l1H 1'4 113
a b
10.1.7. Describe the structure of a digraph of order n without parallel arcs such
that every its vertex satisfies one of the following conditions: a) d+( v) = 0;
b) d-(v) = 0; c) d+(v) = n; d) d-(v) = n.
10.1.8. Prove that for isomorphic digraphs

a) the orders are equal;
b) the numbers of arcs are equal;
c) the collections of in degrees and out degrees coincide.
Are these conditions together sufficient for the isomorphism of digraphs?
10.1.9. Which of digraphs shown at Fig. 10.1.3 are isomorphic?
b c
10.1.10. Find all pairwise nonisomorphic digraphs of third order without loops
and parallel arcs.
10.1.11. The line digraph for a digraph G is the digraph L(G) such that its vertices
bijectively correspond to the arcs of G and (u, v) E AL( G) if and only if the
head of u is the tail of v in G.
a) Express the numbers of vertices and arcs of the line digraph L(G) in terms
of the parameters of G.
b) Prove that any isomorphism of G and H induces an isomorphism of their
line digraphs L(G) and L(H).
10.1.12. Prove that the conditions L:~=1 Ci = L:~=1 di , 0 ::; Ci ::; n, 0 ::; di ::; n,
i = 1,2, ... , n, are insufficient for the existence of a digraph without parallel
arcs with the in degrees (C1, ... , cn ) and out degrees (d 1 , ..• , dn ).
10.1.13. Construct all digraphs without loops and parallel arcs for which the list
(3,2,1,1) is the list both of the in degrees and the out degrees.
10.2 Reachability and Components

A walk, or a (u, v)-walk in a digraph G is an alternating sequence
(10.2.1)
of vertices Vi and arcs Xi of G such that Xi = (Vi-1, Vi), i = 1,2, ... , n. The
vertices Va, Vn are called the terminal vertices and the remaining vertices of the
walk (10.2.1) are called inner, or intermediate. The length of the walk is the
number of arcs contained in it. The length of the shortest (u, v)-walk is called the
distance from u to V and is denoted by p( u, v). A walk is called a chain if all its
arcs are different. A walk is called a path if all its vertices are different, except
that the terminal ones may coincide.
A walk may be specified by the sequence of its arcs; in a digraph with no
parallel arcs a walk may be also specified by the sequence of its vertices.
A walk (chain, path) is called cyclic if its terminal vertices coincide, otherwise
it is called noncyclic. A cyclic path is called a contour. A digraph without contours
is called contour/ess, or acyclic.
An alternating sequence (10.2.1) of vertices and arcs of a digraph G such that
either Xi = (Vi-l,Vi), or Xi = (Vi,Vi-l), i = 1,2, ... ,n, is its arc is called a
semiwalk. A semichain, semipath, and semicontour are defined in a similar way.
If a digraph has a (u, v)-walk then it is said that v is reachable from u. Every
vertex is assumed to be reachable from itself.
A digraph is said to be strongly connected, or strong if any two vertices are
mutually reachable. A digraph is said to be unilaterally connected, or unilateral
if for any two vertices at least one of them is reachable from the other one. A
digraph is said to be weakly connected, or simply weak if any two its vertices are
connected by a semiwalk.
Subgraphs, induced subgraphs, and spanning subgraphs are defined in the same
way as for the ordinary graphs.
A strong component of a digraph is a maximal under inclusion strong subgraph.
A weak component, or connected component is a maximal under inclusion weak
subgraph.
A condensation of a digraph G is the digraph G* with vertex set {Sl, ... , Sm}
of all strong components of G, and (Si, Sj) E EG* if and only if G has an arc with
10.2. Reachability and Components 155
the tail in Si and the head in Sj. Fig. 10.2.1 shows a digraph and its condensation.
8,
G G*
Figure 10.2.1: A digraph and its condensation
If we ignore the orientations of the arcs of a digraph G, we obtain an ordinary

(non-oriented) pseudograph which is called the substrate of G and is denoted by G s .
Accordingly, the digraph G is called an orientation of the pseudograph G s. Clearly
a digraph is connected if and only if its substrate is a connected pseudograph.
A digraph is called disconnected if its substrate is a disconnected pseudograph.
* * *
10.2.1. Prove that every (u,v)-walk contains a (u,v)-path, which is a contour if
u= v.
10.2.2. Prove that every digraph containing both a (u, v)-walk and a (v, u)-walk
contains a contour. Give an example showing that this digraph does not
necessarily contain a contour passing through both u and v.
10.2.3. Prove that every arc of a cyclic walk belongs to at least one contour con-
tained in this walk.
10.2.4. Prove that if d+(v) 2: 1 for every v E VG of a digraph G then G contains

at least one contour.
10.2.5. Prove that every contourless digraph has a vertex with zero indegree and
a vertex with zero outdegree.
10.2.6. Is it true that if d+(v) 2:: 1 and d-(v) 2:: 1 for all vertices v of a digraph G
then every vertex of G belongs to at least one contour?
10.2.7. Prove that if a vertex v is reachable from u and w is reachable from v then
w is reachable from u.
10.2.8. Prove the equivalence of the following statements about digraph:

a) The vertices u, v are mutually reachable.
b) The sets of vertices reachable from u and from v coincide.
c) If u, v are different vertices then they are connected by a path whose every
arc belongs to a contour.
d) There exists a sequence of contours C I , ... , Ck such that u E CI, v E Ck
and every two consecutive contours have a common vertex.
10.2.9. Prove that a digraph

a) is strong if and only if it has a spanning cyclic walk;
b) is unilateral if and only if it has a spanning walk;
b) is weak if and only if it has a spanning semiwalk.
10.2.10. A digraph is called irreducible if its vertex set cannot be partitioned into
disjoint sets VI, V2 in such a way that every arc of the digraph connecting
the vertices from different subsets is directed from VI to V2.
Prove that a digraph is strong if and only if it is irreducible.
10.2.11. Prove that a digraph G is strong if and only if one of the following con-
ditions hold:
a) there exists a cyclic walk that contains all arcs of G;
b) there is no nonempty set VI C VG such that rVI ~ VI;
c) G may be represented as the union of contours C I , ... , Ck such that every

two consecutive contours of this list have a common vertex.
10.2.12. An arc of a digraph is called acyclic if it does not belong to any contour
of the digraph. Prove that a digraph is strong if and only if it is connected
and has no acyclic arcs.
10.2.13. Prove that

a) a non-strong digraph has arcs that belong to no strong component;
b) all such arcs are acyclic.
10.2.14. Prove that a connected non-unilateral digraph has no vertices whose dele-
tion produces a strong digraph.
10.2.15. Let G be a (n, m)-digraph without loops and parallel arcs. Prove that
a) if G is a connected non-strong digraph then n - 1 :::; m :::; (n - 1)2;
b) if G is a strong digraph then n ~ m :::; n(n - I);
c) if G is a connected contourless digraph then n - 1 ~ m ~ n(n - 1)/2.
10.2.16. A vertex of a digraph is called source if every vertex of the digraph is

reachable from it. Prove that every unilateral digraph has a source.
10.2.17. Prove that every unilateral digraph may be converted into a strong one
by the addition of a single arc.
10.2.18. A topological sorting of the vertices of a digraph G is a numbering of its

vertices by numbers {I, 2, ... , IGIl in such a way that for every arc the num-
ber of the tail is less than the number of the head. Prove that a topological
sorting exists if and only if the digraph is contourless.
10.2.19. Construct the condensations of digraphs shown at Fig. 10.2.2.
a h
d
10.2.20. Prove that the following properties of a digraph G are equivalent:

a) G is a contourless digraph;
b) the digraph G and its condensation G* are isomorphic;
c) every walk in G is a path;
d) the vertices of G may be numbered in such a way that its adjacency matrix
(see Sect. 10.3) will be triangular.
10.2.21. Let V ~ VG be a subset of vertices of a connected digraph G such that

no arcs are ingoing into V. Then the set of arcs (v, u), v E V, u E V G\ V, is
called a directed cut of G. Prove that every arc of a digraph belongs either
to a directed cut or to a contour, but not to both simultaneously.
10.2.22. Prove that a digraph is contourless if and only if every its arc belongs to
a directed cut.
10.2.23. Find maximal possible number of arcs in an n-vertex digraph without

parallel arcs that has: a) k connected components; b) k strong components.
10.2.24. A graph G is called directable if al its edges may be directed in such a

way that the resulting digraph will be strong.
a) Prove that the Petersen graph and the graphs of Platonic solids are di-
rectable.
b) Prove that a connected graph is directable if and only if it is bridgeless.
10.2.25. Prove that every graph has an acyclic orientation.
10.2.26. Prove that the minimal number of strong components achievable when
orienting an ordinary graph is equal to the number of bridges of the graph
plus one.
10.2.27. A digraph is called minimally connected if it is strong and the deletion

of any its edge destroys this property. Construct a minimally connected
digraph of order n for every n ~ 2.
10.2.28. A digraph is called quasistrong if any two its vertices Vl, v2 are reach-
able from some vertex V3 (not necessarily distinct from Vl, V2)' For every n
construct a quasistrong digraph of order n that is not strong.
10.2.29. Prove that a digraph has a source (see Exercise 10.2.16) if and only if it
is quasistrong.
10.2.30. A rooted tree is an acyclic digraph G that has a vertex v with d- (v) = 0,
and d- (u) = 1 for every vertex u E V G\ v. The vertex v is called the root of
the tree. A vertex u with d+ (u) = 0 is called a leaf of the tree.
Prove that the following statements are equivalent for a digraph G :
a) G is a rooted tree;
b) G has a vertex that reaches every remaining vertex by a single walk;
c) there exists a vertex v E V G such that
1. d-(v) = 0,
2. for every other vertex u E V G there exists a (v, u)- walk and d- (u) = 1.
d) G is a quasistrong digraph which loses this property after the deletion of

any edge.
10.2.31. A binary tree is a rooted tree such that d+(v) ::; 2 for every vertex v.
The depth of a vertex is the length of the longest path starting from it. The
height of a tree is the depth of its root. Prove that a binary tree of height n
has at most 2n leaves and at most 2n +1 - 1 vertices.
10.2.32. A 2 - 3-tree is a rooted tree in which d+(v) = 2 or 3 for every non-root

vertex v and the lengths of all paths from the root to the leaves are equal.
Prove the following bounds for the number n of vertices and the number 1
of leaves of a 2-3-tree of height h :
10.2.33. A subgraph of a digraph G is called a rooted spanning tree if it is a

rooted tree and contains all vertices of G. Prove that a digraph has a rooted
spanning tree if and only if it is quasistrong.
10.2.34. A (di)graph with m edges (arcs) is called edge-labelled if its edges (arcs)
are provided with labels, e.g., numbers 1,2, ... , m. Two edge-labelled
(di )graphs G, H are said to be isomorphic if there exists a bijection f :
V G ---+ V H such that uv E EG and f( u )f( v) E E H have the same label
(respectively, (u,v) E AG and (J(u),f(v)) E AH have the same label) for
all edges (arcs).
Find:
a) the number of edge-labelled rooted trees of order n.
b) edge-labelled trees of order n.
10.2.35. The transitive closure of a digraph G is a digraph G such that VG =
V G and (u, v) E G if and only if the vertex v is reachable from u in G.
The transitive reduction of a digraph G is a subgraph of G minimal under
inclusion whose transitive closure coincides with that of G.
a) Find the transitive closure and transitive reduction of a digraph shown at
Fig. 10.2.3.
b) Find a digraph with minimal number of vertices that has nonisomorphic

transitive reductions.
c) Prove that a contourless digraph has a single transitive reduction.
10.2.36. A digraph G is called transitive if ((x, y) E AG, (y, z) E AG) implies
(x, z) E AG. Prove that a transitive digraph coincides with its transitive
closure.
10.2.37. Prove that the distance p(u, v) between vertices in a digraph satisfies
the first, second, and fourth axioms of metric from Exercise 1.2.6, but not
necessarily the third one.
10.2.38. Let every arc x of a digraph G without loops be provided with a positive
number q(x). Let us define a function over VG 2 :
pq(U, v) = m)n L q(x),

xEL(P)
where the minimum is over all (u, v)-paths P of G and L(P) is the set of
arcs of the path P. Prove that pq(u,v) satisfies the first, second, and fourth
metric axioms from Exercise 1.2.6, but not necessarily the third one.
10.2.39. Let I be a permutation over the set V = {I, 2, ... , n}. The permutation
digraph GJ is the one with VG J = V and (i, j) E AGJ if and only if I(i) = j.
a) Construct permutation digraphs for the permutations
!1 = (1,2,3,4,5,6); h = (2,3,4,5,6,1); Is = (4,5,2,1,3,6).

b) Describe the structure of the permutation digraph.
c) State a criterion of isomorphism of permutation digraphs in terms of the
underlying permutations.
10.3 Matrices Associated with Digraph

In this section it is convenient to consider digraphs with the vertex set (numbered
by) {I, 2, ... , n}.
The (n x n)-matrix A = A(G) such that Aij =1 if (i,j) E AG and Aij 0 =
otherwise is called the adjacency matrix of the digraph G.
Let G be a digraph without loops and with AG = {al,"" am}. The (n x
m)-matrix I = I(G) such that
if the vertex i is the tail of the arc aj;

if the vertex i is the head of the arc aj;
if the vertex i and the arc aj are not incident.
is called the incidence matrix of the digraph G

The (n x n)-matrix R = R(G) such that Rij = 1 if the vertex j is reachable
from i and Rij = 0 otherwise is called the reachability matrix of the digraph G.
* * *
10.3.1. Find the matrices A(G), I(G), R(G) for the digraph shown at Fig. 10.3.1.
10.3.2. Prove that

a) the number of ones in the i-th row (i-the column) of A(G) is equal to the
in degree (outdegree) of the digraph G;
b) the number of elements of the i-th row of I( G) equal to 1 (-1) is equal to
the indegree (outdegree) of the digraph G.
10.3. Matrices Associated with Digraph 161
1 2
•6
4
10.3.3. How are the adjacency, incidence and reachability matrices of the opposite
<-
digraph G related with those of G?
a) If the digraphs G and H are isomorphic then the following conditions are
valid:
(1) A(G) may be obtained from A(H) by equal permutations of rows and
columns;
(2) I( G) may be obtained from I(H) by permutations of rows and columns;
b) if one of the above conditions (1), (2) is valid then G-== H.
10.3.5. Demonstrate that R(G) = R(H) does not imply G -== H.
10.3.6. Prove that the matrix A( G) can be reduced to the block-diagonal form
diag[A 1 , ... , Ak], k> 1,
where Ai, i = 1, ... , k are square matrices, by equal permutations of rows
and columns if and only if G is not connected.
10.3.7. Let G 1 , ... , Gk be the strong components of a digraph G. Prove that
a) A( G) may be reduced to the following form
by equal permutations of rows and columns, where Ai is the adjacency matrix

of the digraph G i , i = 1, ... , k, and 0 is the zero matrix;
a) R( G) may be reduced to the following form
[:. o
.;,.]
by equal permutations of rows and columns, where Ii is the square matrix

of ones of order IGil, i = 1, ... , k.
10.3.8. Prove that for an acyclic digraph G the matrix A(G) may be reduced to
the triangular form by equal permutations 'of rows and columns.
10.3.9. Prove that the rank of the incidence matrix I(G) of a digraph G of order
n IS
a) n - 1, if G is connected;
b) n - p, if G has p connected components.
10.3.10. A matrix is called totally unimodular if every its minor is equal to 0, 1,

or -1. Prove that the incidence matrix of a digraph is totally unimodular.
10.3.11. Prove that the (i,j)-th element of the matrix (A(G))k is equal to the
number of (i, j)-walks of length k in the digraph G.
10.3.12. Let G be a digraph of order n. Let A(k)(G) denote the binary (n x

n)-matrix such that A~J) is equal to 1 if and only if G has an (i,j)-path
of length at most k.
a) Find the matrices A(1)(G), A(2)(G), A(3)(G), A(4)(G) for a digraph G
shown at Fig. 10.3.2.
b) prove that A(k)(G) = A(G), k = 1,2, ... ,n, for a transitive digraph G
(see Exercise 10.2.36) of order n.
c) Prove that a digraph G of order n is transitive if and only if A(k)(G) =
A(G) for some k ~ 2.
d) Prove that A(G) = A(n)(G), where G is the transitive closure of a digraph
G of order n.
10.3.13, Prove that for a digraph, its transitive closure, and its transitive reduction
(see Exercise 10.2.35) the strong components partition the vertex set into
subsets in the same way.
10.4. Tours and Paths 163
10.3.14. Prove that the (i, i)-th element of the matrix (R(G))2 is equal to the
number of vertices in the strong component of G containing the vertex i.
10.3.15. For a loopless digraph G we define an (n x n)-matrix M(G) with Mu =

d-(i), Mij = -k, i =1= j, where k is the number of parallel arcs (i,j). Prove
that G is a rooted tree with the root l' (see Exercise 10.2.30) if and only if
Mrr = 0, Mi; = 1 for i =1= l' and the complementary minor of the (r,r)-th
element of M (G) is equal to 1;
10.4 Tours and Paths

The definitions of Eulerness and Hamiltonicity for digraphs are similar to those
for ordinary graphs.
An Eulerian chain is a chain containing all arcs of a digraph. A connected
digraph is called Eulerian if it has a cyclic Eulerian chain.
Theorem 10.4.1 For a connected digraph G the following statements are equiv-
alent:
1. G is Eulerian;
2. d+(v) = d-(v) for every v E VG;

3. G is the union of contours such that no pair of them have common arcs.
A contour or path of a digraph is said to be Hamiltonian if it contains all

vertices of the digraph. A digraph is called Hamiltonian if it has a Hamiltonian
contour.
Theorem 10.4.2 Let G be a strong digraph of order n > 1 without loops and
parallel arcs. If
deg u + deg v ~ 2n - 1 (10.4.1)
for any pair of different nonadjacent vertices of G then G is Hamiltonian.
The independence number Qo(G), proper coloring, and the chromatic number
X(G) of a digraph G are defined in terms of its substrate graph G s : Qo(G) =
Qo(G s ), X(G) = X(G s ).
Let M = {Pi, P2 , ••• , PI} be a set of vertex-disjoint paths in a digraph G. If
VG = U;=l V Pi then M is called a path partition of digraph G. The minimal
number of paths that constitute a path partition of G is denoted by I(G).
Theorem 10.4.3 l( G) ~ Qo( G) for any digraph G.
Theorem 10.4.4 If k is the maximal length of a path in a digraph G then X( G) ~

k + 1.
* * *
10.4.1. For a digraph shown at Fig. 10.4.1 find contours of lengths 2, 3,4,5,6; a
cyclic Eulerian chain; a Hamiltonian contour.
10.4.2. Find a digraph that is not Eulerian but has a cyclic Eulerian chain.
10.4.3. Find a digraph that
a) is Eulerian but is not Hamiltonian;
b) is Hamiltonian but is not Eulerian;
c) is neither Eulerian, nor Hamiltonian.
10.4.5. Prove that a connected digraph contains a noncyclic Eulerian chain if and
only if it has two vertices Vl, V2 such that d- (v!) = d+ (v!) + 1, d- (V2) =
d+(V2) - 1, and d-(v) = d+(v) for all vertices v other than Vl, V2.
10.4.6. Let a digraph G have two vertices u, v such that d+(v) - d-(v) = k =
d-(u) - d+(u), and d-(w) = d+(w) for all vertices w other than u, v. Prove
that G may be partitioned into kedge-disjoint (v, u )-chains.
10.4.7. Prove that the edges of any graph may be directed in such a way that
Id+(v) - d-(v)1 :::; 1 for any vertex v in the resulting digraph.
10.4.S. Prove that the inequality (10.4.1) of Theorem 10.4.2 cannot be relaxed.
10.4.9. Prove that a strong digraph of order n without loops and parallel arcs has
a Hamiltonian contour if one of the following conditions holds for every its
vertex v :
(a) degv ~ n;
(b) d+(v) ~ n/2, d-(v) ~ n/2.
10.4.10. Prove that if deg u + deg v 2:: 2n - 3 for any two nonadjacent vertices u, v
of a digraph G of order n then G has a Hamiltonian path.
10.4.11. A set S of vertices of a digraph G is called incomparable if no vertex
from S is reachable from some other vertex from S. Prove that the maximal
cardinality of an incomparable set of a digraph G is equal to the independence
number of the transitive closure of G.
10.4.12. Prove that for any orientation of edges of the graph shown at Fig. 10.4.2
a) it may be partitioned into at most 5 paths;

b) there exists a path of length at least 3.
10.4.13. Prove that the minimal number of paths in a path partition of a transitive
digraph G is equal to ao(G) .
. 10.4.14. Let r(G) denote the minimal integer such that every orientation of the
graph G has a path of length r( G) and there exists an orientation of G that
has no path of length r( G) + 1. Prove that x( G) = r( G) + 1.
10.4.15. Prove that X(L(G)) 2:: log2 X(G), where L(G) is the line digraph of a
digraph G, see Exercise 10.1.11.
10.4.16. For a digraph G without loops and parallel arcs we denote
kl = vEVG
min d+(v), k2 = vEVG
min d-(v), k = max{k 1 , k 2}.
Prove that
a) for any vertex v of a digraph there exists at least kl! different paths of
length at least kl starting at v;
b) for any vertex v of a digraph there exists at least k2! different paths of
length at least k2 ending at v;
c) if k > 0 then the digraph has a contour of length at least k + 1.
166 10. Directed Gra.phs
10.4.17. Let G be a digraph. For A ~ AG define:
6(A) = { h~l- IfAI, if A f 0,

6o(G) = max 6(A).
otherwise 0~AS;VG
Prove that G has a spanning subgraph whose all connected components are
contours if and only if 6o(G) = O.
10.5 Tournaments
A digraph whose substrate is a complete graph is called a tournament.
Theorem 10.5.1 For every vertex of a strong tournament of order n and for
every integer k, 3 ~ k ~ n, there exists a contour of length k passing through this
vertex.
Corollary 10.5.2 Every strong tournament is Hamiltonian.
* * *
10.5.1. Draw all pairwise nonisomorphic tournaments of orders 2, 3, 4.
10.5.2. Prove that every tournament has a Hamiltonian path.
10.5.3. Prove that every tournament has a vertex such that all other vertices are
reachable from it.
10.5.4. Prove that every tournament either is strong or may be converted into
strong by reversing a single arc.
10.5.5. Prove that a transitive tournament has a single Hamiltonian path.
10.5.6. Prove that a tournament is strong if and only if it is Hamiltonian.
10.5.7. Prove that a Hamiltonian tournament T has a vertex v such that the
tournament T - v is Hamiltonian.
10.5.B. Prove that the automorphism group of a tournament is of odd order.
10.5.9. Prove that for every tournament

a) the distance from a vertex with maximal out degree to any other vertex is
at most two;
b) the distance from any vertex to a vertex with minimal in degree is at most
two.
10.6. Base and Kernel 167
10.5.10. Let (S1 ~ S2 ~ ... ~ sn) be the sequence of out degrees of vertices of a
tournament T. Prove that
a) 2:~=1 Si = n(n - 1)/2;
b) 2::=1 Si ~ k(k -1)/2, k = 1,2, ... , n -1, and these inequalities are strict
if and only if T is strong;
c) (k - 1)/2 ~ Sk ~ (n +k- 2)/2, k = 1,2, ... , n - 1;
d) T is transitive if and only if Sk =k - 1, k = 1,2, ... , n.
10.5.11. Prove that
vEVT vEVT
for a tournament T.
10.5.12. Let T be a maximal contourless subgraph of a tournament T. Prove that

IATI ~ LlVTI/2J l(IVTI + 1)/2J.
10.5.13. A digraph is called a bipartite tournament if its substrate is a complete
bipartite graph. Is it true that a strong bipartite tournament is Hamiltonian?
10.6 Base and Kernel

Let G be a digraph and let B be a subset of vertices such that every vertex from
VG\B is reachable from some vertex from B. If B is minimal under inclusion
among the subsets with this property then B is called a base of the digraph G.
D( v) will denote the set of vertices of G reachable from v E V G and D( v, A)
will denote the set of vertices of G reachable from v E V G via the arcs from
A~AG.
A minimal under inclusion set A of arcs of G is called an arc base of the digraph
G if D(v) = D(v, A) for every v E VG. The minimality property for A implies that
for every A' ~ A there exists a vertex u E VG such that D( u, A)\D( u, A') f. 0.
A set S of vertices of a digraph G is called dominating if for every wE VG\S
there exists v E S such that (w, v) E AG. The number of vertices in a dominating
set of minimal cardinality will be denoted by (3( G).
A set S of vertices of a digraph G is called independent if no two vertices from
S are adjacent.
A set of vertices of a digraph G which is both dominating and independent is
called a kernel of the digraph G.
Theorem 10.6.1 A digraph without contours of odd length has a kernel.
* * *
10.6.1. Prove that no two vertices of a base are connected by a walk.
10.6.2. Prove that every vertex with zero in degree belongs to a base.
10.6.3. A base component of a digraph is a strong component such that the corre-
sponding vertex in the condensation of the digraph is of zero indegree. Prove
that a subset B of vertices of a digraph is a base if and only if it consists of
vertices belonging to base components and every base component contains
exactly one vertex from B.
10.6.4. Prove that all bases of a digraph are of the same cardinality.
10.6.5. Which vertices constitute a base of a contourless digraph? Prove that such
digraph has a unique base.
10.6.6. Prove that a digraph has a unique base if and only if every base component
has exactly one vertex.
10.6.7. Prove that every base of a digraph has a vertex such that maximal number
of vertices of the digraph is reachable from it.
10.6.8. An antibase B of a digraph G is a minimal under inclusion set of vertices

of G such that for every vertex v E VG\B some vertex from B is reachable
from v.
Prove that an antibase of a digraph G coincides with some base of the op-
T-
posite digraph G .
10.6.9. Prove that for a digraph

a) all vertices of zero out degree belong to every antibase;
b) no two vertices of an antibase are connected by a walk;
c) all antibases are of the same cardinality.
10.6.10. Prove that a vertex v of a digraph G may belong both to a base Band
to an anti base B if and only if the strong component of G containing v
corresponds to an isolated vertex in the condensation G*.
10.6.11. Prove that a set A ~ AG is an arc base of a digraph G if and only if it

satisfies the following two conditions:
(1) for every arc (x, y) E AG\A there exists a path from x to y with all arcs
in A;
(2) for no arc (x, y) E AG\A there exists a path from x to y with all arcs in
A\{(x, y)}.
10.6.12. Prove that every digraph has at least one arc base.
10.6.13. Find a digraph with two arc bases of different cardinalities.
10.6.14. Prove that a contourless digraph without parallel arcs has a single arc
base.
10.6.15. Prove that in a connected digraph G a subgraph induced by the set of

arcs from an arc base is connected; it is strong if G is strong.
10.6.16. Prove that a connected (strong) digraph of order n has an arc base with
at least n - 1 (respectively, n) arcs.
10.6.17. Prove that an arc base of a strong digraph of order n has at most 2n - 2
arcs.
10.6.18. For every positive integer n construct a strong digraph of order n with
an arc base of size 2n - 2.
10.6.19. Prove that a digraph of order n has a Hamiltonian contour if and only if
it is strong and has an arc base of size n.
10.6.20. Prove that (3(G) ~ INI ~ ao(G) for any kernel N of a digraph G.
10.6.21. Find a loopless digraph of minimal order without kernel.
10.6.22. Prove that a set K of vertices of a digraph is a kernel if and only if K is
both maximal independent set and minimal dominating set.
10.6.23. Find a digraph in which
(a) some maximal independent vertex set is not a kernel;
(b) some minimal dominating vertex set is not a kernel.
10.6.24. A set N ~ VG for a digraph G is called an outkernel if r N = VG\N and
an inkernel if r- 1 N = VG\N.
Prove that an out kernel of a digraph is an inkernel of the opposite digraph.
10.6.25. Prove that a set N of vertices of a digraph is a kernel if and only if N is
an inkernel.
10.6.26. Find a digraph that has
(a) an out kernel but no inkernel;
(b) an inkernel but no outkernel;
(c) different out kernel and inkernel;
(d) coinciding outkernel and inkernel;
(e) two out kernels of different cardinalities.
10.6.27. For any integers k, n, 1 ~ k ~ n, find a digraph of order n with k-vertex
kernel.
10.6.28. Describe a digraph in which every vertex constitutes a kernel.
10.6.29. Prove that a subset N of vertices of a digraph G is an outkernel if and
only if the following three conditions are satisfied (where V+ is the set of
vertices with zero indegree):
(a) V+ ~ N;
(b) N n rv+ = 0;
(c) if V'= VG\(V+ U rv+) =1= 0 then the set N ' = N\V+ is an out kernel
of a subgraph G' ~ G induced by V'.
10.6.30. State and prove a criterion for the inkernel similar to the previous exercise.
10.6.31. Prove that a digraph with no semi contours of odd length and
(a) with no vertices of zero indegree has at least two outkernels;
(b) with no vertices of zero outdegree has at least two inkernels.
Show that these statements will fail if we discard the requirements about
vertices.
10.6.32. Prove that a contourless digraph cannot have
(a) more than one inkernel;
(b) more than one outkernel.
10.6.33. Prove that the set V+ of vertices of a digraph G of zero in degree is the
only out kernel of G if V+ n r-1v ::j: 0 for every v E VG\ V+.
State and prove a similar statement about inkernels.
10.6.34. Prove that a digraph whose substrate is a bipartite graph has both an
inkernel and outkernel.
10.6.35. Prove that every graph has an orientation with out kernel.
10.6.36. A function 9 with nonnegative integer values defined over the vertex set
of a (di )graph G such that for every vertex v g( v) is the minimal positive
integer outside the set {g( u) : u E rv} is called a Grundy function for the
(di)graph G.
(a) Find Grundy functions for the digraphs shown at Fig. 10.6.1.
a h c
(b) Prove that every graph admits a Grundy function.

(c) Find a loop less digraph of minimal order that has no Grundy functions.
(d) Find a digraph of minimal order that has several Grundy functions.
(e) Find a digraph that has a Grundy function but the opposite digraph has
no Grundy functions.
(f) Prove that a digraph with a Grundy function has a kernel.
(g) Prove that a digraph has a Grundy function if every its subgraph has a
kernel.
(h) Prove that a contourless digraph has a Grundy function.
(i) Find a digraph that has a Grundy function whose opposite digraph also
has a Grundy function.
(j) Prove that is a (di)graph G has a Grundy function 9 then X(G) ~
maXvEVG g(v) + 1.
(k) Prove that if a graph G is p-chromatic then it has a Grundy function 9
such that maxvEVG g(v) ~ p - 1.
Chapter 11
Hypergraphs
11.1 Hypergraphs: Basic Notions

A hypergraph is a generalization of a graph such that any nonempty subset of
vertices may be an edge.
Let V be a finite non empty set and let £ be a collection of any (not necessarily
different) subsets of V.
The pair (V, £) is called the hypergraph with the vertex set V and edge (or
hyperedge) collection £. Equal subsets in £ are called multiple edges of the hy-
pergraph. The set of vertices and the collection of edges of a hypergraph Hare
denoted by V Hand £H respectively. The number IV HI of vertices of H is called
=
the order of the hypergraph and is denoted by IHI. If IHI nand I£HI m then=
H is called an (n, m)-hypergraph.
If a vertex v E V H belongs to an edge e E £ H then we shall say that v and
e are incident (to each other). The collection of edges incident to a vertex v will
be denoted by £(v). The number 1£(v)1 is called the degree of the vertex v and
is denoted by deg v. The number lei is called the degree of the edge e. Since only
nonempty sets of vertices may be edges, the degrees of edges are at least 1. A
vertex of a hypergraph incident to no edge is called isolated.
Two vertices of a hypergraph are called adjacent if some edge contains both of
them. Two edges el and e2 of a hypergraph are called adjacent if el n e2 =1= 0.
In drawings it is convenient to depict an edge e of a hypergraph by an area
bounded by a closed curve encompassing the vertices of e if lei =1= 2. If lei = 2,
then e is depicted by a curve segment connecting its vertices, as in the case of an
ordinary graph.
Similarly to ordinary graphs, we may consider both labelled hypergraphs (the
vertices have different labels) and unlabelled ones.
If a hypergraph has no multiple edges and the degrees of all edges are equal to
k then the hyper graph is called k-uniform. Clearly, the pseudograph is a special
case of the hypergraph and the ordinary graphs are the 2-uniform hypergraphs.
A complete k-uniform hypergraph is a hypergraph with the edge set being the set
of all k-element subsets of vertices.
173
174 11. Hypergraphs
Let HI, H2 be hypergraphs and let r.p: V HI ---> V H2 and ~ : £H 1 ---> £H 2 be

bijections. If for any v E V HI, e E £ H I the images r.p( v) and ~(e) are incident
in H 2 if and only if v and e are incident in H 1 then H land H 2 are said to be
isomorphic hypergraphs, and this relation is denoted by HI ~ H2.
Every (n, m)-hypergraph H = (V, £) without isolated vertices may be related
with an (m,n)-hypergraph H* = (V*,£*) as follows.
If V = =
{VI, ... , vn } and £ {el, ... , em} then V* = =
{vi, ... , v;;"} and £*
{ei, ... , e~} and vi is incident to eJ if and only if ei and Vj are incident in H. The
hypergraph H* is called dual to H.
For a hypergraph H = (V, £), we define the incidence graph to be the bipartite
graph K(H) with the vertex set V U £ and the edge set
{(v, e) : (v,e) E V x £,v E e}.
The graph K(H) is also called the Konig representation of hypergraph H.

A path of length I > 0 in a hypergraph H, or a (Vl' VI+1 )-path is an alternating
sequence
of vertices and edges of a hypergraph H that satisfies the following conditions:

(1) el, ... ,el are different edges of H;
(2) Vi,Vi+1 Ee;, i= 1, ... ,1.
Notice different edges of a hypergraph H are considered to be different elements
of the collection £H (they may coincide as subsets of V H).
A path is called simple if all its vertices are different, with the possible excep-
tion that the first and the last vertices may coincide.
If I > 1 and Vl = VI+l then the path is called cycle and the simple path is
called simple cycle.
If a hypergraph H has a (u, v)-path for every two vertices u, v, then H is called
a connected hypergraph.
A hypergraph P is called a part of a hypergraph G if V P ~ V Hand £P ~ £H.
A maximal under inclusion connected part of a hypergraph is called a component
of the hypergraph.
A deletion of edge e of a hypergraph H = (V, £) is an operation producing the
hypergraph HI = = =
H - e (V, £/) with £1 £\{e}.
A deletion of vertex v of a hypergraph H = (V, £) is an operation producing
= =
the hypergraph H' H - v (V I,£') with VI = = =
V\{v}, £1 {e': e' e\{v} =F
0, e E £}.
A vertex v of a hypergraph H is called a cutpoint if the hypergraph H - v has
more components than H.
Let H be a connected hypergraph with IHI > 1. A maximal under inclusion
connected part of H without cut points is called a block. A connected hypergraph
without cutpoints has a single block, which is the hypergraph itself.
The vertex connectivity number K( H) of a connected hypergraph H is the min-
imal number of vertices whose deletion produces a disconnected or a single-vertex
hypergraph. A hypergraph H is called k-connected if K(H) ~ k.
11.1. Hypergraphs: Basic Notions 175
The edge connectivity number >.( H) of a connected hypergraph H is the minimal

number of edges whose deletion produces a disconnected hypergraph.
A matching of a hypergraph H is a subset £1 ~ £H such that el n e2 = 0
for any el, e2 E £', i.e., no two edges from £1 are adjacent. A matching is called
maximum cardinality matching if it has maximal possible number of edges among
the matchings of the hypergraph; this number is called matching number of the
hypergraph and denoted by al (H). A covering of a hyper graph H is a subset
U ~ V H such that Un e 1= 0 for every edge e E £H. A covering is called
minimum cardinality covering if it has minimal possible number of vertices among
the coverings of the hypergraph; this number is called the covering number of the
hypergraph and is denoted by f3o(H).
A set of vertices of a hypergraph is called independent if no its subset is an
edge of the hypergraph. The empty vertex set is defined to be independent. An
independent vertex set of maximal size is called maximum cardinality independent
vertex set, and its size is called the independence number and is denoted by ao(H).
A (vertex) coloring of a hypergraph is a partition of its vertex set into color
classes. A vertex coloring of a hypergraph H is called proper if every edge e E £ H
contains two vertices colored differently. Clearly a proper coloring is possible only
for hypergraphs in which every edge is of degree at least 2.
A hypergraph is called k-colorable if it admits a proper coloring in k colors.
The chromatic number X(H) of a hypergraph H is the minimal number of colors
necessary for proper coloring H.
The line graph of a hypergraph H = =
(V, £) is the graph L(H) (W, E) such
that there is a bijection f between the set Wand the collection £, and two vertices
Wl, W2 E Ware adjacent if and only if the edges f( wI) and f( W2) are adjacent.
In other words, L(H) is the graph of intersections of edges of H.
* * *
11.1.1. Find a 3-uniform hypergraph of order 10. How many are there labelled
hypergraphs of such type?
11.1.2. How many are there labelled (6, 3)-hypergraphs a) without multiple edges;
b) with possible multiple edges?
11.1.3. Generalize the Handshake Lemma for the case of a) hypergraphs;
b) k-uniform hypergraphs.
11.1.4. A matroid M with the support V may be related with the hypergraph
C(M) = (V, C), where C is the set of cycles of M.
a) Characterize the independent sets of M in terms of the hypergraph C(M).
b) What is the independence number of C(M)?
c) What can you say about the components of C(M) if M is the partition
matroid defined in the Exercise 5.2.10?
11.1.5. Draw all hypergraphs H with a) K(H) ~ K 5 ,5; b) K(H) ~ K2,3; c)
K(H) ~ K l ,5; d) K(H) ~ K l ,3 U K 2,2' Here U denotes the disjoint union of
graphs.
176 11. Hypergraphs
11.1.6. Which bipartite graph is the Konig representation for the cycle en?
11.1. 7. Is it true that for every bipartite graph G without isolated vertices there
exists a hypergraph H such that K(H) =:! G?
11.1.8. Find a hypergraph of tenth order that has a single cycle. Construct the
Konig representation of this hypergraph and verify that it has a single cycle
too.
11.1.9. Prove that the number of simple cycles of a hypergraph is equal to the
number of simple cycles of its Konig representation.
11.1.10. Prove that the Konig representations of hypergraphs Hand H* are iso-
morphic to each other.
11.1.11. A hypergraph H is called sel/dual if H =:! H*. Construct a selfdual
hypergraph of order n for every positive integer n.
11.1.12. Is it true that a hypergraph in which every vertex and every edge is at
least of degree 2 always has a cycle?
11.1.13. A hypergraph is called planar if its Konig representation is planar. Prove

that
a) a hypergraph H is planar if and only if its dual is planar;
b) for a planar k-uniform (n, m)-hypergraph n ~ (k - 3)m/3 + 2.
11.1.14. For hypergraphs shown at Fig. 11.1.1 find a) a maximum cardinality
matching; b) a maximum cardinality independent set.
11.1.15. Draw line graphs for hypergraphs shown at Fig. l1.1.I.

11.1.16. The following theorem due to Whitney holds for graphs: iffor connected
graphs of orders at least four the line graphs are isomorphic then the graphs
are isomorphic themselves. Does a similar theorem hold for hypergraphs?
11.1.17. Which graph may be a line graph of a) a hypergraph; b) a connected
hypergraph; c) a disconnected hypergraph?
11.1.18. Find a line graph for the hypergraph Hn = (V, {Et, ... , En}), where
V={{i,j}: i,j=1,2, ... ,n, i#j},E;={{i,j}: j=I,2, ... ,n, i#j}.
11.1.19. Construct a hypergraph dual to the graph [(n'
11.1.20. A hypergraph without isolated vertices is called linear if every pair of

different vertices belongs to at most one common edge. Is it true that a
hypergraph dual to a linear one is also linear?
11.1.21. Let H be a linear hypergraph with all vertices of degree two. Prove that
H* ~ L(H).
11.1.22. Prove that every graph is a line graph of some linear hypergraph.
11.1.23. Prove that for a graph H without isolated vertices the line graph L(H*)
of the dual hypergraph H* is isomorphic to H.
11.1.24. For a hypergraph H = (V, £) we may define a graph (Hh with the same
vertex set such that two vertices of (Hh are adjacent if and only if they are
adjacent in H. Prove that if H has no isolated vertices then (Hh :::: L(H*).
11.1.25. A hypergraph H is called conformal if the set of maximal cliques of (Hh

is the set of maximal edges of H. Prove the following Gilmor theorem: a
hypergraph is conformal if and only if for any triple e1,e2,e3 E £H there
exists eo E £H such that (e1 n e2) U (e1 n e3) U (e2 n e3) ~ eo.
11.1.26. Prove that for a hypergraph H without isolated vertices

a) a subset U ~ V H is a covering of H if and only if V H\U is independent;
b) a subset U ~ V H is a minimum cardinality covering of H if and only if
V H\U is a maximum cardinality independent set of H;
c) if IHI = n then O!o(H) + !3o(H) = n.
11.1.27. A vertex v of a hypergraph H = (V, £) is called hypopendant if there
exists another vertex Vi such that £(v) ~ £(v l ) (recall that £(x) denotes the
set of edges containing a vertex x). Prove that if v is a hypopendant vertex
of a hypergraph H 1- K2 without isolated vertices then
a) v belongs to some maximum cardinality independent set of H;
b) !3o(H) = !3o(H - v);
c) O!o(H) = O!o(H - v) + 1;
d) O!l(H) = O!l(H - v).
11.1.28. Prove the validity of the following statement: a graph L is a line graph
of some k-uniform hypergraph if and only if there exists a covering of L by
complete graphs L 1 , ... , Ls such that
a) every k + 1 different graphs from L1, ... L. have empty intersection;
b) every k different graphs from L 1 , ••• Ls have at most one common vertex.
11.1.29. Prove that there exists a graph with arbitrarily many vertices which is
not the line graph of a 3-uniform linear hypergraph but every its induced
subgraph is.
178 11. Hypergraphs
11.1.30. Prove that in a hypergraph

a) every two blocks have at most one common vertex;
b) every edge of size at least two belongs to exactly one block.
11.1.31. Prove that
a) if a vertex v of a hypergraph belongs to more than one block then it is a
cutpoint;
b) if a block of a hypergraph contains vertices u, v then it contains every
simple (u, v)-path of the hypergraph.
11.1.32. Is it true that a hypergraph is biconnected if and only if every two its
vertices belong to a common simple cycle?
11.1.33. Is it true that a connected hypergraph is biconnected if and only if every
two its edges of degrees at least two belong to a common simple cycle?
11.1.34. Are the hypergraphs shown at Fig. l1.1.23-colorable?
11.1.35. Prove that the following inequalities are valid for a connected hypergraph
H:
a) )"(H) ::; 6(H);
b) 6(H) ::; I£Hlr(H)/IHI.
11.1.36. Find the chromatic number for a) the complete k-uniform hypergraph H~
of order n, 2::; k::; n; b) hypergraphs Hi, H2 , H3 shown at Fig. 11.1.3.
11.1.37. How may the chromatic number ofa hypergraph change after the deletion
of a) an edge b) a vertex which is not incident to an edge of degree two?
Give examples.
11.2. Hypergrapb Realizations 179
11.1.38. Prove that a hypergraph is bicolorable if it has no cycles of odd length.
Is the converse statement true?
11.2 Hypergraph Realizations

A realization of a hypergraph H = (V, £) is a graph G that satisfies the following
conditions:
(1) VG = VH;
(2) every edge of G is contained in some edge of H;
(3) for every edge e E £ the induced subgraph G( e) is connected.
A realization of an edge e E £ of a hypergraph H is any connected graph
G e with VGe = e. Thus, every realization of a hypergraph is the union of some
realizations of all its edges.
If a realization of a hypergraph is a planar graph then we speak about a planar
realization. When a realization is a tree or a cycle, etc., we call it a realization of
the hypergraph by tree, by cycle, etc., and the hypergraph is said to be realizable
by tree, cycle, etc.
It is said that a hypergraph satisfies the Helly conditions if the intersection of
any subset of pairwise adjacent edges is nonempty.
Theorem 11.2.1 A connected hypergraph H is realizable by tree if and only if H

satisfies the Helly condition and the line graph L(H) is triangulated.
A realization G of a hypergraph H = (V, f), £ = {el, e2, ... , em}, is called

strict if there exists a partition of EG into subsets E l , ... , Em such that G( E i )
is a realization of ei, i = 1,2, ... , m.
* * *
11.2.1. Are the hypergraphs H l , H2 from Fig. 11.2.1 realizable by tree?
• • •
2 2
6 (i
• •
Figure 11. 2.1: To Exercise 11. 2.1
11.2.2. Do the hypergraphs shown at Fig. 11.2.2 and Fig. 11.2.3 have planar re-
alizations?
180 11. Hypergraphs
6 5
11.2.3. The contraction of an edge eo of a hypergraph H = (V, £) is the operation

that produces the hypergraph H' = =
(V',f') such that V' (V\eo) U {vol,
where Vo f/. V, and f' = f1 U f2' where
f1 = {e' : e' = (e\eo) U {vo}, e E f\{eo}, en eo i- 0},

f2 = {e E f : en eo = 0}.
The insertion of a vertex Vo f/. V H into an edge eo of a hypergraph H is an
=
operation producing the hypergraph H' (V', f') such that V' V U {vo}, =
f' = (f\{eo}) U e*, e* = eo U {vol.
a) If there exists a planar realization of a hypergraph H then there exists a
planar realization of a hypergraph iI obtained from H by (1) edge deletion;
(2) edge contraction; (3) insertion of vertex into edge; (4) deletion of vertex
of degree 1;
b) If there exists a planar realization of a hypergraph iI obtained from H by
the contraction of the edges of a subset £ ~ fH then there exists a planar
realization of a hypergraph obtained from H by the contraction of the edges
of any subset £", £' ~ £" ~ £H.
11.2. Hypergrapb Realizations 181
(.'"'
• •)
(. • •)
(. • •~
'-....J '---"
11.2.4. Find a hypergraph H such that it has a planar realization but no hyper-
graph H - v, deg v 2: 2, has a planar realization.
11.2.5. Prove that a planar realization of a hypergraph H exists if the Konig

representation of H is planar. Give an example when a planar realization of
a hypergraph exists but its Konig representation is nonplanar.
11.2.6. a) Prove that a connected hypergraph with at most three edges is realizable
either by tree or by simple cycle.
b) Under what conditions it is realizable by tree?
11.2.7. Prove that a hypergraph with at most four edges has a planar realization.
11.2.8. Prove that a hypergraph has a planar realization if and only if every its
block has a planar realization.
11.2.9. Is it true that if the Konig representation of a hypergraph is planar then

the dual hypergraph has a planar realization?
11.2.10. Is it true that if a hypergraph has a planar realization then the dual
hypergraph has a planar realization?
11.2.11. Prove that if the hypergraph obtained from a hypergraph H by the dele-
tion of two vertices is realizable by path graph then H has a planar realiza-
tion.
182 11. Hypergraphs
11.2.12. Suppose that a hypergraph H is realizable by a path graph P. Prove that

for any two adjacent edges e', e" of H the induced subgraph P( e' n e") is
connected.
11.2.13. Suppose that a hypergraph H' is obtained from a hypergraph H by the
deletion of two vertices which do not constitute an edge of degree two. Prove
that if H' is realizable by cycle graph then H has a planar realization.
11.2.14. Suppose that a hypergraph has a planar realization. Is it true that any
its minimal under inclusion realization is a planar graph?
11.2.15. Suppose that a cycle graph C is a realization of a hypergraph H = (V, £)
and there exist two adjacent edges e', e" E £ such that the induced subgraph
C( e' n ell) is not connected. Prove that e' U e" = V.
11.2.16. Prove that a hypergraph is k-colorable if and only if it is realizable by
k-colorable graph.
11.2.17. What is the chromatic number of a hypergraph realizable by tree?
11.2.18. Prove that the chromatic number of a hypergraph with planar realization
is at most four.
11.2.19. Do the following hypergraphs have strict realizations:
a) H=(V,£) with V={1,2,3,4},£={{1,2,3},{1,2,4},{1,3,4},{2,3,4}};
b) the hypergraph HI shown at Fig. 11.2.1?
11.2.20. Find a non-strict realization of the hypergraph HI shown at Fig. 11.2.l.
11.2.21. Prove that a strict realization of a connected hypergraph by tree exists if
and only if the hypergraph has no cycles.
11.2.22. Let v be a hyperpendant vertex of a hypergraph H (see Exercise 11.1.27)
and let t( H) denote the minimal number of edges in a realization of H. Prove
that a) t(H) = t(H - v) + 1; b) if H is realizable by tree then H - v is also
realizable by tree.
11.2.23. Prove that if a hypergraph is realizable by tree then it satisfies the Belly
condition.
11.2.24. A cycle of a hypergraph that contains an edge incident to three vertices of
this cycle is called balanced. A hypergraph is called balanced if every cycle
of odd length I ;::: 3 is balanced. Prove that every balanced hypergraph
satisfies the Belly condition.
11.2.25. A hypergraph is called strongly balanced if every cycle of length I ;::: 3 is
balanced. Prove that a hypergraph is strongly balanced if and only if every
its connected part is realizable by tree.
11.2.26. Prove that a hypergraph is conformal (see Exercise 11.1.25) if and only
if its dual satisfies the Belly conditions.
Answers to Chapter 1:
ABC of Graph Theory
1.1 Graphs: Basic Notions

1.1.3. c: The empty graph.
1.1.4. Kn.
1.1.5. C;.
1.1.6. a: No.
1.1.7. No.
1.1.8. a: 9; b: 7; c: There are no such graphs.
1.1.9. Hint: See Exercise 1.1.7 and proceed by contradiction.
1.1.10. No.
1.1.11. a: Hint: Count the number of edges in such graph.
b: Yes. Hint: Consider an appropriate bipartite graph.
c: Hint: See Exercise 1.1.11a.
1.1.12. a: Hint: Use the fact that if a, b are real numbers and a +b= n then
ab:::; n 2 j4.
b: Kp,p for n = 2p.
1.1.13. Complete bipartite graphs.
1.1.14. Selfcomplementary graphs: 0 1 , P4, C5 •
1.1.16. a: Bipartite r-regular graphs of order n exist only for n = 2m, 0:::; r :::; m.
One of them with parts A and B may be defined as follows. Let A Zm = =
{O, 1, ... , m - I} be the additive group of residuals modulo m and let B be
another copy of this group. If x E B then we set N (x) = {x + yEA: y =
0,1, ... , r - I}. Here x E B denotes the copy of x E A.
183
184 Answers, Hints, Solutions
b: There are no such graphs.

c: Hint: If 0 is the required graph then IEOI = n - 1.
1.1.17. For even n the graph is unique, for odd n there are no such graphs.
1.1.18.. a: 2c ;; b: C C2 ifm:S; C~, and 0 otherwise.

n
1.1.20. Hint: Two non isomorphic graphs satisfying conditions a)-c) are shown at
Fig. A1.1.1.
Figure A1.1.1: To Answer 1.1.20
1.1.21. Hint: If f : VO -+ V H is an isomorphism of graphs 0 and H then for

any edge e = uv E EO the set f(e) = {f(u), f(v)} is an edge of H. Prove
that the mapping e ...... f( e) specifies an isomorphism of line graphs L( 0) and
L(H).
1.1.22. b: 0 1 ~ O 2 ~ 0 3 , and there are no other pairs of isomorphic graphs at
Fig. 1.1.8.
1.1.24. Hint: The numbers of pairwise nonisomorphic graphs of order n are: 4 for
n = 3; 11 for n = 4; 34 for n = 5.
1.1.25. a: The complete graph J{4 is the unique cubic graph of order 4.
b: The number of pairwise nonisomorphic cubic graphs of order 6 is 2, see
Fig. A1.1.2. Hint: Consider the complementary graphs.
Figure A1.1.2: Cubic graphs of order 6
c: The number of pairwise non isomorphic cubic graphs of order 8 is 6, see

Fig. A1.1.3.
Hint: Let G be the required graph. First prove that G contains either
the triangle C3 , i.e., three pairwise adjacent vertices, or the cycle C4 , i.e.,
four vertices a, b, e, d such that {ab, be, cd, da} C EG. Further, consider the
following five possibilities:
1.1: Graphs: Basic Notions 185
Figure A1.1.3: Cubic graphs of order 8
(1) G contains K 4 , i.e., four pairwise adjacent vertices;

(2) G does not contain K 4 , but it contains K4 - e, i.e., four vertices a, b, e, d
such that {ab, be, cd, da, ae} ~ EG and bd (j. EG;
(3) G does not contain neither K 4 , nor K4 - e, but it contains two triangles;
(4) G contains exactly one triangle;
(5) G contains no triangles.
Prove that G ~ Gi (Fig. A1.1.3) in the case (i), i = 1, ... ,6. In the case (5)
G ~ G 5 or G ~ G 6 . G 5 contains the cycle C5 and G 6 does not.
1.1.26. b: The number of pairwise non isomorphic cubic circulant graphs of order
6 is 2.
c: Yes.
1.1.27. a: Hint: See Fig. 1.2.4a.

b: A regular graph of degree C;. The number of edges is C;~.
c: Solution: For n = 1 the statement is trivial. Assume that n ;::: 2. Since the
application of a single transposition changes the parity of the permutation,
no two permutations of the same parity are adjacent in the n-permutation
graph. Hence it is a bipartite graph, one part being all even permutations,
the other one being all odd permutations.
1.1.28. Solution: Consider the set of all prime numbers in the ascending or-
der: 2, 3, 5 .... Let Pi be the i-th prime number. Let further G be an
(n, m)-graph. Let us label its vertices by numbers PI, .. . Pn and let us label
its edges by the numbers Pn+1, ... , Pn+m. Afterwards we multiply the label
of each vertex by the labels of the edges incident to it. This gives a common
divisor graph isomorphic to G.
1.1.30. a: The number of edges of G is C~ - m. The number of edges of a selfcom-

plementary graph of order n is n(n - 1)/4.
b: Hint: See Exercise 1.1.30a.
1.1.31. a: P4; b: C 5 and the graph shown at Fig. ALIA; c: there are no such
graphs.
A
Figure ALIA: To Exercise 1.1.31
1.1.33. Hint: Use Exercise 1.1.32.
1.1.34. No. Hint: Such graph would contain a single vertex of degree 49. Let u be
a vertex of degree 50 and let v be of degree 49. Is there an edge uv in the
graph?
1.1.35. Solution: The number of vertices of the graph L(G) is m = IEGI =

~ E~=l di· Two edges of G are adjacent as vertices of L(G) if and only
if G has a vertex incident to both of them. Therefore all edges of G incident
to the same vertex are pairwise adjacent in L(G). Hence
1.1.37. a: I<3; b: The line graph of the tetrahedron is isomorphic to the octahe-
dron.
1.1.38. a: Yes; b: No.
1.1.40. Solution: Suppose that V = V I<5 = {I, 2, 3,4, 5}. Then

VL(I<5) = VL(I<5) = {xy: x,y E V, x =1= y}.
The vertices xy and uv are adjacent in L(I<5) ifand only if {x, y}n{u, v} = 0.
Hence G = L(I<5) is a 3-regular graph. The induced subgraphs (see Sect.
1.3) G( {12, 34, 15,23, 45}) and G( {13, 24,35,14, 25}) are isomorphic to C5 •
Further see Fig. A1.1.5.
12
23 15
Figure Al.1.5: To Exercise 1.1.40
1.2 Walks, Paths, Components

1.2.2. a: ~ I:~=3(k - 1)!C~ if n ~ 3, and 0 for n :S 2;
b: ~ I:;~n2(p,q) k!(k - l)!C;C~ if p ~ 2, q ~ 2, and 0 otherwise.
Hint: Let A, B be the parts of Kp,q, P :S q, The length of any simple cycle
of Kp,q is an even number 2k, 2 :S k :S p. In such cycle the vertices from
A and from B alternate. Finally, the number of all simple cycles with the
same vertex set V, IVI = 2k, is (k - 1)!k!/2.
1.2.4. a: 3 for n ~ 3, 00 for n = 1,2;

b: 4 for p, q ~ 2; 00 for p or q equal to 1;
c: 5.
1.2.5. a: Hint: Delete the "excessive" pieces of the walk. b: False; c: False.
1.2.10. Hint: Proof is by contradiction using the Handshake Lemma.
1.2.11. Hint: Starting from any vertex v, move along an incident edge e into the
adjacent vertex w. Similarly, move from w further along an edge other than
e. Continue this process until some vertex will be visited twice. Select a cycle
in the traversed walk.
1.2.12. a: Hint: Suppose that
are given paths, UI VI = U, Uk = Vm = =

V, U a and Va are the first
(counting from u) different vertices of these paths, u{3 and v1' are the first
coinciding vertices of these paths after U a and Va. Consider the union of the
(Ua-I, u{3)-subpath of P and (Va-I, v1')-subpath of Q.
1.2.13. a: Hint: Use Konig's theorem.

b: False.
1.2.14. See Fig. A1.2.1. The graphs G 1 = KI and G6 = P4 are selfcomplementary.

II AIINtx1
•
• •
Gs
Figure A1.2.1: To Exercise 1.2.14
1.2.15. Km,n, m, n ;::: 1, m + n ;::: 3. Hint: Demonstrate that after the deletion of
any edge e = uv from Km,n either d(u, v) = 3 in the resulting graph or it
becomes disconnected.
1.2.16. Hint: Use the previous exercise.
1.2.17. Hint: The number of such graphs is 5.
1.2.18. a: the star K1,n, n ;::: 2;

b: Kn;
c: K2 ;
d: the star K1,n;
e: the star K1,n or the triangle C3 ;
f: the star K1,n, n ;::: 2;
g: the path graph Pn , n ;::: 3;
h: the cycle graph Cn. Hint: Using the Handshake Lemma show that G has
no vertices of degrees greater than 2. Then notice that WGI = IEGI.
1.2.20. Hint: In any walk that contains the deleted edge one may replace this
edge by the remaining edges of the cycle. The deletion of an edge that does
not belong to any cycle of a connected graph produces a graph with two
connected components.
1.2.21. n - 1. Hint: Use the previous exercise and then the induction over the
number of edges of the graph.
1.2.22. Hint: Consider the shortest path connecting the ends of the edges el, e2.
1.2.23. a: The lower bound is attained for the graphs with no cycles. The upper
bound is attained only for the graph with one component isomorphic to
Kn-k+l and the remaining components trivial.
Hint: To prove the lower bound, use the Exercise 1.2.21. For the upper
bound, notice that in a graph with maximal number of edges all components
are complete graphs and then use the fact that if p ;::: q ;::: 1 then
b: If n - k f= 2 then there exists a unique n-vertex bipartite graph G with
maximal number of edges. It is obtained by the addition of k - 1 isolated
vertices to Kp,q, where p = = = =
(n - 1)/2, q p + 1 or p q (n - 1+1)/2,
depending on parity. For n - I = 2 there is one more such graph, i.e.,
2K2 + (1- 2)K 1 .
Hint: The reasoning is the same as in item a.
c: Hint: Using the item b, construct a graph of order n with k components
and maximal possible number of edges. Deleting the edges from it while
keeping the number of the components constant, we obtain the required
graphs.
1.2.24. a: The statement will become false. Hint: Suppose that the graph is
disconnected and consider the maximal possible degree of a vertex in the
component of the smallest order.
b: No, we cannot. Hint: Use the Exercise 1.2.23a.
c: It will become invalid. Hint: Use Konig's theorem and the Exercise
1.2.23a.
1.2.25. a: Hint: Use the theorem stating that any permutation of n numbers may
be produced from any other one via an appropriate sequence of transposi-
tions.
b: n -1. Hint: Use the theorem stating that the minimal number of transpo-
sitions whose product is equal to a given permutation s of the set {I, 2, ... , n}
is equal to its decrement n - c, where c is the number of independent cycles
of s.
c: Each vertex is both central and peripherical. Hint: For any permutation
ai, a2, .. . , an the decrement of the substitution
is n - 1.
1.2.26. a: Solution: Let u be a central vertex of G. Then
d(u, w) ~ max d(u, v) = e(u) = r(G)

vEVG
for every vertex w E YG. Consider any diametral (a, b)-path L. Since L is
the shortest (a, b)-path, its length does not exceed the length of the walk
composed of the shortest (a, u)-path and the shortest (u, b)-path.
b: Hint: Prove that a diametral (a, b)-path contains two adjacent vertices
= =
u, v such that e(u) d(u, a), e(v) d(v, b). Further, use the Exercise 1.2.26a.
1.2.28. a: Hint: Suppose that G is disconnected and consider the partition VG =

AU B, where A is a connected domain of G, B = VG\A.
c: Solution: Let L be a diametral (a, b)-path. Since its length is at least 3,

the vertices a and b are nonadjacent. By the same reason, every v E VG is
not adjacent either to a or to b. Hence G has the edge ab and each vertex of
G is adjacent either to a or to b. This implies that d(G) :S 3.
1.2.29. a: n - 1, if n > 1; b: p + q - 2, if p + q 2: 3; c: 2; d: 1.
1.2.30. Hint: Consider a diametral path (VI, V2, ... , Vd, Vd+l) of G and the set
1.2.31. Hint: Use Konig's theorem.
1.2.32. Solution: Let V be a vertex of degree .6.(G) and let N(v) be its neighbor-
hood. Consider u E N(v); clearly, IN(u)l2: 8(G). Since G has no triangles,
=
N(v) n N(u) 0. This implies the required inequality.
1.2.33. Hint: Use the Exercise 1.2.32.
1.2.35. Hint: See the graph at Fig. A1.2.2 in which u and v are adjacent to all
vertices of G.
BAg B A
1.2.36. a: Hint: Let the vertex it E V K( G) correspond to the edge x = uv of G.

From the definition of K(G) it follows that x is adjacent only to u and v
which correspond to the vertices uand v of G.
b: Hint: Use the Exercise 1.2.36a.
1.2.37. The graphs a, b, c, g are bipartite. At Fig. A1.2.3 the vertices of one part
are marked by A and the vertices of the other one are marked by B.
. _ . _{ (n-l)/2, ifnisodd,
1.2.38. a. d(Pn ) - n - 1, r(Pn ) - n /2 - 1, 0
th ·
erWlse.
The center of Pn consists of a single vertex if n is odd and of two vertices if
n is even. The endvertices are the peripherical vertices.
b: d(C ) = r(C ) = { (n -1)/2, ifn is.odd,

n n n/2, otherwlse.
Every vertex is both central and peripherical.
c: d(Kn) = r(I{n) = 1(n ~ 2); every vertex is both central and peripherical.
d: d(Kl,n) = 2, r(Kl,n) = 1(n ~ 2); every pendant vertex is peripherical.
The center consists of a single vertex.
e: d(Km,n) = r(Km,n) = 2(n ~ 2, m ~ 2); every vertex is both central and
peripherical.
1.2.39. a: r(G) = 2, d(G) = 2; very vertex is both central and peripherical.

b: r(G) = 3, d(G) = 4; the central and the peripherical vertices are shown
at Fig. A1.2.4a. (At Fig. A1.2.4, the central vertices are marked by "*" and
the peripherical ones are marked by "+" .)
* *
* *
a
+ +
c: r( G) = 2, d( G) = 2; each vertex is both central and peripherical.

d: r(G) = 3, d(G) = 3; each vertex is both central and peripherical.
e: r(G) = 2, d(G) = 2; each vertex is both central and peripherical.
f: r( G) = 2, d( G) = 3; the central and the peripherical vertices are shown
at Fig. A1.2.4b.
g: r( G) = 4, d( G) = 6; the central and the peripherical vertices are shown
at Fig. A1.2.4c.
1.2.41. Solution: Let G(n,U) be a circulant graph with U = {Ul, ... ,Uk}. Let d
denote the largest common divisor of Ul, ... , Uk, n. Vertices a, b are in the
same component of G if and only if G has an (a, b)-path. In the considered
situation this means that b is obtained from a by the addition of elements

from U. In other words, there are integers Xl, ... , Xk, Y such that
But the above equation is solvable in integers if and only if b - a is divisible

by d. Therefore the number of components of G is equal to the number of
residuals modulo d, i.e., it is equal to d.

of Graphs. Reconstructibility
1.3.3. a: (1) Yes; (2) only C s ; (3) only C5 .
b: No.
1.3.6. No.
1.3.7. (a) No. (b) Yes.
1.3.8. (a) On. (b) 2K2 • (c) C4 • (d) Kn.
1.3.9. Graphs without cycles.
1.3.10. Hint: Consider the graph HG described in Exercise 1.1.30c.
1.3.11. No.
1.3.12. Hint: Prove that such graph must be bipartite. Therefore it must have
three pairwise nonadjacent vertices.
1.3.13. Let G be a graph with d(G) = d. We extend it as follows. For every

vertex v E VG we add a new set Sv of d - deg v pendant vertices adjacent
to v (the sets Sv are pairwise disjoint). Further, we take a nonnegative
integer t such that I UVEVG SV 1+ 2t ~ d - 1 and add 2t more new vertices
Xl, ... ,Xt, Y1, ... , Yt together with edges X;Y;, i = 1,2, ... , t. The set of all
vertices added to G is denoted by Sl, and the resulting graph is denoted by
Gt. Let G 2 be a copy of G1 (with VG2 disjoint from VG1) with S2 C VG 2
being the corresponding copy of Sl. Finally, we construct a (d - I)-regular
bipartite graph G3 with parts Sl and S2 (see the answer to the Exercise
1.1.16a). Clearly, the graph G 1 U G2 U G 3 is the required graph.
1.3.14. K1 and the graphs shown at Fig. A1.3.1.
1.3.15. Hint: Consider the set of graphs which do not possess the given hereditary
property.
1.3.16. Graphs whose components are complete graphs.

1.3: Subgraphs and Hereditary Properties of Graphs. Reconstructibility 193
n~O m~O
1.3.17. A graph shown at Fig. A1.3.2.

1.3.18. Suppose that the required regular graph G exists. Let w denote a central
vertex of the star K1,n induced by the neighborhood of a vertex v E VG.
Since n ~ 2, there exists a vertex x E N(v)\w.
The vertex x is adjacent to no vertex from N(v)\w, since N(v) induces
the star K1,n. Moreover, it is adjacent to no vertex y f/: N(v) U v, since its
neighborhood N(x) must also induce the star K1,n. Therefore degx = 2 ~
deg v, which contradicts to the regularity of G.
1.3.19. a: K 4 . b: 3K2 .
1.3.20. {H 1 , H 2 , ... } = {H 1 , H 2 , .. . }.
1.3.21. (a) Yes; a minimal FIS is {K 2}. (b) Yes; a minimal FIS is {02}' (c) No.
(d) No. (e) No. (f) Yes; a minimal FIS is {P3}' (g) Yes; a minimal FIS is
{K 1 ,3,Cn ,n = 3,4, ...}. (h) Yes; a minimal FIS is {C3,C4,P4}, (i) No. (j)
No. (k) Yes; a minimal FIS is
(1) No. (m) No. (n) Yes; a minimal FIS is {C3 , C 5 , .•• , C 2k+l, ... }. (0) Yes;
=
a minimal FIS is {Cn, n 3,4, ...}. (p) No. (q) No.
1.3.22. a: No. b: Yes.
1.3.23. Hint: Sufficiency. Every simple cycle C 2k+l is either isomorphic to K3 or
contains the induced subgraph P3 . Therefore a graph without induced K3
and P3 is bipartite. Furthermore, every non-isolated vertex is adjacent to all
vertices from the opposite part; otherwise one could readily find the induced
P3'
1.3.24. Yes.
1.3.25. a: Yes. b: Yes.

1.3.26. a: Hint: Use the Exercise 1.3.25a.
b:Hint: Clearly, every graph from the class [It] has no induced subgraphs
P4, since P4 (j. [It]. Now we can demonstrate that every graph that has
no induced P4 is in· [K]. To this end one may apply the induction over the
number of vertices. Let us assume the induction hypothesis (i.e., every graph
of order n > 1 with no induced subgraphs P4 is in [K]) and consider a graph
G of order n + 1 with no induced subgraphs P4 • Consider G - v, v E VG,
and show that we may assume that this subgraph is disconnected. Using the
fact that G has no induced subgraphs P4 , prove that v is adjacent to every
vertex wE VG\{v}, and hence the graph is not connected.
1.3.27. Hint: First reconstruct the set of vertices of the graph together with their
labels. Then reconstruct the edge set following the rule: the vertices with
labels i, j are adjacent in the reconstruction if and only if they are adjacent
in at least one card of the labelled deck. Substantiate this rule.
1.3.28. a: No. b: Yes.
1.3.29. a: Yes. b: No.
1.3.31. deg v = LuevG mu/(IGI- 2) - m v, where mx = IEGxl.

1.3.32. The number of edges of Gis (LvevG m(Gv»/(IGI- 1), where m(Gv) is
the number of edges of the card Gv .
The number of components of Gis min{k(G v) : v E VG}, where k(G v ) is
the number of components of the card G v .
1.3.33. Hint: Use the fact that every subgraph F of G is contained in each of
IGI- IFI cards Gv such that v ¢ V F.
1.3.34. K 4 .
1.3.35. No.
1.3.36. Yes.
1.3.40. Solution: Assume that V = V(K5) = {I, 2, 3,4, 5} and V(L(K5» = {xy :
x, y E V, x f; y}. The set of maximal cliques of L(K5) is separable into two
classes. The first class consists of the four-element cliques:
Qv={vx: xEV, xf;v}, vEV.
The second class consists of the three-element cliques:
Quvw = {uv,uw,vw}, u,v,wE V,
where u, v, ware pairwise different. A clique Qv consists of the edges of K5

incident to the vertex v. A clique Quvw corresponds to the triangle in K5
with the vertices u, v, w.
1.3: Subgraphs and Hereditary Properties of Graphs. Reconstructibility 195
Thus, IQ(L(K5 ))1 = 5 + C~ = 15.

The neighborhoods of the vertices of Q(L(K5)) are the following.
N (Q v) = {Q x, Qvxy : x, y E V, x =I y, x =I v, y =I v};
N(Quvw) = {Qx, Qyzt : x = u, v, w, I{y, z, t} n {u, v, w}1 = 2};
deg Qv = 10; deg Quvw = 9.
1.3.41. Hint: Consider the correspondence v 1---+ Qv, where v E VG and Qv is the
clique of L( G) consisting of all edges of G incident to v.
1.3.42. c: Solution: Necessity. Since K 1 ,3 is not a line graph (see Exercise 1.1.36a),
then by item a) the graph G contains no induced subgraphs K 1 ,3. It is easily
verified that there exists a single connected graph (see Fig. A1.3.3) such that
L( H) = ]{4 - e. Since H contains a triangle, by item b) the graph G contains
no induced subgraphs ]{4 - e.
1.3.43. Hint: Assume that

VG={1,2, ... n}, EG={el,e2, ... em }
and construct the sequences
by setting a; = 1 if the vertex i is incident to the edge ej and a} = 0 other-

wise. To each ai, i = 1, ... , n, we append the sequence {3i = ({3f,·· .{3~2-2m)'
where J3j = 1 if
(i - l)n - (d 1 + ... + di- 1 ) + 15:. j 5:. in - (d 1 + ... + di )
and {3J = 0 otherwise. If i= 1 we assume that d1 + ... + di - 1 = O. Denote
, i = (i ai ai )
al,···a im ,1-'1,···l-'n'-2m·
Then for adjacent vertices i, j,
d( ,i ,
,i) = di - 1 + dj - 1 + n - di + n - dj = 2n - 2,
and for nonadjacent vertices i, j,
To conclude the proof, consider the Hamming graph H n 2 - m ,2n-2'

1.4 Operations on Graphs

1.4.3. G + H = G u H.
1.4.4. No.
1.4.5. No.
Hint: Consider the property specified by the forbidden induced subgraph
2K2 ·
1.4.6. The edge e belongs to no cycle of the graph G.
1.4.7. Hint: Use the fact that the distance d(u, v) is odd.
1.4.8. Ln/2J.
1.4.9. The number k is equal to the diameter of the graph.
1.4.10. a: No. b: Yes.
1.4.11. IEGI + IEHI + IVGIIVHI·

1.4.12. Hint: Prove that G has a dominating vertex.
1.4.13. See Fig. A1.4.1.
a b
c d
Figure A1.4.I: To Exercise 1.4.13
1.4.14. a: nlm2+n2ml. b: nlm2+n~ml' c: 2mlm2 d: 2(mlm2+m~m~), where

m~ =
ni(ni - 1)/2 - mi, i 1,2. =
1.4.15. Cartesian product, conjunction and modular product.
1.5: Matrices Associated with Graphs 197
1.4.16. d; + dj, i = 1. ... , n" j = 1, ... , n.
1.4.18. Yes.
1.4.20. a: Hint: Let the vertex set of a simple cycle C 2m +1 be V = {O, 1, ... , 2m},
with each vertex i E V being adjacent to i + 1 (addition is modulo 2m + 1).
Verify that the mapping t.p: V 2 ~ V 2 : t.p( i, j) =
(i + j, 2m + 1 - i + j) is
an isomorphism of the graphs C2m+1 X C 2m +1 and C 2m +1 1\ C 2m +1 for any
pair (i,j) E V 2 •
1.4.21. c: degudegv + (IGI- degu -1)(IG'I-1).

1.4.23. a:Solution: Set Hl = C 5 . Suppose that the graph Hn have already been
constructed. Set H n+1 = C 5 ; - Hn. Clearly, Hn is a selfcomplementary
graph of order 5n for all n > O.
1.4.26. IQnl = 2n, IEQnl = n2 n- 1 .

1.4.27. a: 8. b: 14.
1.5 Matrices Associated with Graphs

1.5.4. a,b: Hint: Without loss of generality assume that VG = V H = {1, 2, ... , n}.
G ~ H implies that there exists a permutation of the set VG such that
ij E EG ¢:::::} s(i)s(j) E EH. If A A(G) and B = A(H) then the above=
condition may be rewritten as
Bs(i)s(j) = Aij, i,j = 1,2, ... , n. (A1.5.1)
1.5.6. a: Solution: Since a permutation s is a bijective mapping, for each i E

X = {1,2, ... ,n} there exists exactly one j E X such that s(j) = i. Hence
exactly one element of the i-th row of Ms, say, the j-th element, is equal to
1. A k-th column of Ms also contains a single 1, say, at the position s(k). A
determinant which has a single nonzero element in each row and column is
equal to the product of its nonzero elements up to the sign.
b: Solution: For M = Ms,
n
(MA)ij = LMikAkj.
k=l
But Mik = 1 for i = s(k) and Mik = 0 otherwise. Hence
d: Solution: Graphs G, H of order n are isomorphic if and only if there exists

a permutation s of the set {l, 2, ... , n} satisfying the equalities (A1.5.1).
This system of equalities is equivalent to a single matrix equality
B = MsAM.-l. (A1.5.2)
In fact, (A1.5.2) is equivalent to BM. = M.A, i.e.,
(BMs)pq =(MsA)pq, p, q = 1,2, ... , n. (A1.5.3)

But
n
(BMs)pq = L Bpk(M.h q = Bpk, k = seq)·
k=l
In the same way,

(M.A)pq = Akq, k = S-l(p).
Hence the system (A1.5.3) assumes the form
Bp.(q) = A.-l(p)q, p, q = 1,2, ... , n.
Setting s-l(p) = i, q =j, we obtain (A1.5.1).

1.5.8. a: Hint: If the highest coefficient of a polynomial of degree n > 0 is 1 then
the coefficient at x n - 1 is the sum of the roots of the polynomial with the
opposite sign. On the other hand, the sum of the roots of the characteristic
polynomial of a matrix is equal to the trace of the matrix.
b, c: Hint: The coefficient of the characteristic polynomial of an n x n-matrix
at x n - 2 is equal to the sum of all second-order minors along the main diag-
onal, and the coefficient at x n - 3 is the sum of all third-order minors along
the main diagonal with the opposite sign.
1.5.9. x 3 (x - 2)(x + 2) is the characteristic polynomial for the first pair of graphs
and (x + l)2(x -1)(x 3 - x 2 - 5x + 1) is the characteristic polynomial for the
second pair of graphs.
1.5.10. Hint: Use induction with respect to k.
1.5.13. Solution: Let G be a connected graph of order n > 1. First, let us prove
that
rank I(G) ~ n - 1. (A1.5.4)
To this end, let us choose a spanning tree Tl in G (see Section 2.2). Assign
index 1 to a pendant vertex and to the edge incident to it. Delete these
vertex and edge obtaining a tree T2 with n - 1 vertices. Assign index 2 to
a pendant vertex of T2 and to the edge incident to it. Proceed in the same
way until we are left with a single vertex. Assign index n to it. Thus all
vertices of G and n - 1 its edges become indexed. The remaining edges (if
1.5: Matrices Associated with Graphs 199
any) are indexed by n, n + 1, ... in an arbitrary way. Under this indexing,

the incidence matrix of G assumes the form
1 0 0 0
1 0 0
*
* *
I(G) = n
0
* * ... 1 1
*
* * ... * 1
*
* * * *
and its upper left minor of order n-l is equal to 1. Hence (A1.5.4) is proved.
Let now G(A, B; E) be a bipartite graph, IAI = p. Let us rewrite the matrix
I( G) after indexing its vertices is such way that the indices of the vertices
from part A become less than those for part B. If we add the sum of rows
2,3, ... , P to row 1 and subtract the sum of the remaining rows from it,
then the first row will contain only zeroes. Hence, rank I(G) :::; n - 1, i.e.,
actually, rank I( G) = n - l.
Now let us consider the case when G = Cn is a simple cycle of odd length.
Let us index its vertices and edges according to their natural order:
Under this indexing,
0 0 0
:1,
1 0 0
1 1 1
I(C.) = [ :
0 0 1
detl(Cn ) = 2, rank I(Cn ) = n.

Let now G be any connected non-bipartite graph. Then G has a simple
cycle CP1 of odd length Pl. If V CP1 f. V G, then either the induced subgraph
G l = G(VG\VCp1 ) is bipartite or it contains a simple cycle CP2 of odd
length P2. In the latter case consider the induced subgraph
Iterating this process, we arrive at one of the following two partitions:
VG = VCP1 U VCP2 U ... U VCpk , (Al.5.5)

VG = VCP1 U VCP2 U ... U VCPk U W, (Al.5.6)
where G(W) is a bipartite graph.
Suppose that we arrive at A1.5.5. If Iq = I(ep.), then under an appropriate
indexing of the vertices and edges the matrix I( G) assumes the form
I(G) = [AB], A = diag[lt, h.··, h];

rank I( G) = PI + ... +Pk = n.
If otherwise we arrive at A1.5.6, then similarly I( G) assumes the form
I( G) = [AB], A = diag[II , h ... , Ik, 1],

where 1= I(G(W», with the additional feature that B has a column such
that one of its 1's is at the position with index at most the sum PI + ... + Pk
and another 1 has the index larger than the sum. Therefore,
rank I(G) = PI + ... + Pk + IWI-l = n.
1.5.14. Hint: Index the vertices of a bipartite graph in such a way that the indices
of the vertices of one part be less than those of the other part.
1.5.15. Solution: Consider the adjacency matrix A of the circulant graph G =
Gn,u, It is convenient to index the rows and columns of this matrix by
modulo n residuals: the row and the column that correspond to vertex i are
indexed by i E Zn. By the definition of circulant graph,
N(x) = {x + u : u E U} (A1.5.7)
for x E Zn, where N(x) is the neighborhood of x. If Aij = 1, i.e., ij E EG,
then j = i + u, u E U. But then j + 1 = (i + 1) + u, i.e., A i +1,j+l = 1. Thus
the elements of A satisfy the conditions
Ai+l,j+I=Aij , i,j=1,2, ... ,n, (A1.5.8)

i.e., A is a circulant.
Conversely, let the vertices of G be indexed by elements of the additive group
Zn of modulo n residuals and let A = A(G) be a circulant. Set
U = {x E Zn : Ax+I,1 = I}.
For the elements of the circulant A the equalities (A1.5.8) are valid. There-
fore (A1.5.7) is valid for each vertex x E Zn, i.e., G = G(n,U)'
1.5.16. a, b: Solution: Let G 1 = = G G(n,ul) be a nonempty circulant graph
and let n be prime. Then the adjacency matrix A = A(Gd is a circulant:
A = II(h) is a polynomial in h (see Exercise 1.5.15) of degree at most
n - 1. The characteristic polynomial of h is xn - 1, its roots are the complex
numbers 1, t, t 2 , ... , tn-I, where en = 1, t i= 1. Hence
is the spectrum of A = II(h).

1.6: Automorphism Group of Graph 201
If G = G(n,U 2 ) is another circulant graph with the incidence matrix B =

A(G 2 ), then similarly, B = h(h) and
12(1), h( f), h( (2), ... , h( fn-l)
is the spectrum of B.
Now, if the spectra of A and B are the same, then, in particular, h(f) =
h(flc); hence f is a root of the polynomial p(x) = h(x) - h(x lc ). But it
is known that for prime n the polynomial q(x) = 1 + x + x 2 + ... + xn - 1
with root f is irreducible over the field of rational numbers. Therefore p(x)
is divisible by q( x).
It is easy to understand that p( x) is also divisible by x-I. In fact, the
elements of the first column of the matrix h(h) are the coefficients of the
polynomial
h(x) = ao + alX + a2x2 + ... + an_Ix n- l .
Hence ao = 0 and h (1) is the sum of the elements of a column, i.e., it is the
degree of G I . The same is true for 12(1). But by Theorem 1.5.3 the degree of
a regular graph is uniquely specified by its characteristic polynomial, hence
we have
degG l = degG 2, h(l) = h{l),p(l) = 12(1) - h{1h) = O.
Thus we obtain
p{x) == 0 (mod (xn - 1»,
p(h) = h(h) - h(h lc ) = 0,
h(h) = h(h lc ).
The latter equality means that U2 = kUI (see Exercise 1.5.15). 0 rf; U2,
therefore k =F O. Further, see Exercise 1.1.26a.
1.6 Automorphism Group of Graph

1.6.2. Set F = ((1,2),(3,4)}, H = (1,2),{3,4),(1,3),(2,4)}. Then F ~ H, but
the groups F and H are not similar.
1.6.3. Let F be a subgroup of the symmetric group S{X) and let H be a subgroup
of S(Y), IYI = k. Suppose that (hi,7ri) E FWr H, i = 1,2 and (hl,7rl) =
(h 2 , 7r2)' Then for any x E X, y E Y,
i.e., by the definition,
I.e.,
Therefore IFWr HI = IIIIIHI. Since the group II is isomorphic to the direct

product of k copies of the group F, we have IIII = IFlk, hence IF Wr HI =
IFlklHI·
1.6.4. a: Solution: Suppose that (hi, 71'i) E F Wr H, i = 1,2. Consider the product
(hI, 71'1)(h 2, 71'2). By formula (1.6.3), for any (x, y) E X x Y,
(h l , 71'!)(h 2 , 71'2)(X, y) = (h l , 71'!)«h 2, 71'2)(X, y)) =
(h l , 71'1)( 71'2 (y)( X), h2(y)) = (7I'1h2(Y)7I'2(Y)(X), h1 h2(Y))'

But by formula (1.6.2),
71'1h2(Y)7I'2(Y) = (7I'1h2 + 71'2)(Y)·

Thus,
Hence we have proved that if we multiply the elements of the wreath product
as mappings of X x Y, then
(A1.6.1)
Further, let 0 denote the mapping Y --> F such that o(y) = ex for all y E Y.
This mapping is the zero of the group II. By (1.6.3),
(ey, o)(x, y) = (o(y)(x), ey(y)) = (ex (x), y) = (x, y)

for any (x, y) E X x Y, i.e., (ey, 0) = e is the unit of the symmetric group
SeX x Y).
For any pair (h, 71'), formula (A1.6.1) implies
Similarly,
This proves that the wreath product F Wr H is a subgroup of the symmetric

group SeX x Y).
b: Solution: Let F be a transitive subgroup of SeX) and let H be a transitive
subgroup of S(Y). Let us prove the transitivity of the wreath product as a
subgroup of the group S(XxY). Take a, b EX, c, dEY. F has a permutation
I such that I( a) = band H has a permutation h such that h( c) = d. Consider
71' : Y --> F such that 71'(y) = I for all y E Y. Then by (1.6.3),
(h, 71')(a, c) = (7I'(c)(a), h(c)) = (I(a), d) = (b, d),

i.e., F Wr H is indeed transitive. Proof of the converse is similar.
• • •
1.6.8. Graphs shown at Fig. A1.6.1 and their complements.

1.6.10. a, b: Aut G = Sn.
c: If 1,2, ... , n are the vertices of the path Pn in the natural order, then
Aut Pn = (a), where
a = { (1, n)(2, n - 1) ... (k, k + 1), n = 2k;

(1,n)(2,n-1) ... (k,k+2), n=2k+1;
d: Hint: Let the vertices of Cn be indexed by elements of the additive

group Zn of modulo n residuals in the natural order along the cycle: VCn =
{O, 1,2, ... , n -1}. For x E Zn the actions of permutations a, b will be defined
by the following equalities: a(x) = -x, b(x) = x + 1. If A = (a), then
Aut Cn = (a,b) = AUbAu ... Ubn-1A, IAut Cnl = 2n.

1.6.11. Hint: Prove first that for n > 3 An is twice transitive, i.e., for any two
pairs (i,j), (k,l) such that i i= j, k i= 1 there exists a permutation sEAn
satisfying the conditions s( i) = k, s(j) = l.
1.6.12. a: Hint: If t.p E Aut G and V is the connected domain of a graph G then
t.p(V) = {t.p(v): v E V} is also a connected domain.
b: Solution: Let {G b ... , Gk} be the set of components of G and let them
be isomorphic to each other. Set
G 1 = F, Vi = VGi, i = 1,2, . .. ,k,

and for each i = 2, ... , k we pick an isomorphism 'Pi : V1 -+ Vi of G 1 and G i .
We shall write the elements v E V1 in the form of pairs (v, 1) :
Similarly, we may assume that t.pi(V, 1) = (v,i), i.e., we may write
Vi = {(vl,i),(V2,i), ... ,(vn ,i)},i= 1,2, ... ,k,

where {Vb . .. , V n } = V F. Let us prove the inclusion
(Aut F) Wr Sk ~ Aut G. (A1.6.2)
Suppose that 7r : {I, 2, ... , k} -+ Aut F. Consider the element (e,7r) of the
wreath product, where e is the identity permutation of the set {I, 2, ... , k}.
By formula (1.6.3), (e, 1I")(v, i) = (1I"i)(V), i), hence (e,1I") E Aut G. Con-
sider the element (h,o), where h E Aut F and 0: {1,2, ... ,k} ---+ Aut F
is such that o(i) = e, where e is the unit of Aut F. By formula (1.6.3),
(h, o)(v, i) = (v, h(i». Therefore (h,o) E Aut G. But by formula (1.6.4)
each element (h,1I") E (Aut F) Wr Sk may be represented as a product
(h,1I") = (e, 1I")(h, 0). Therefore the inclusion (A1.6.2) is proved.
Let now s be any automorphism of G. Define s E Sk to be such that sCi) = j
if s(lI;) = Vi, i = 1,2, ... , k, and consider the mapping (s,o). We have
previously demonstrated that it is in Aut G. Set t = (s, o)-l s . Then t E
Aut G, t(lI;) = lI;, i = 1,2, ... ,k. Hence, t = (e,1I"), 11" : {1,2, ... ,k} ---+
Aut F.
Thus, s = (s,o)t E (Aut F) Wr Sk, and together with (A1.6.2) this implies
Aut G = (Aut F) Wr Sk.
1.6.14. a: Under an appropriate indexing of vertices, Aut G = (S2 Wr S3) x S2.
b: Aut G = Sm x Sn.
c: Aut G = Sm Wr S2.
1.6.15. Aut G = Sm-1 Wr S2.
1.6.18. b: Solution: Consider the mapping
cp: Aut G ---+ AutEG, f 1--+ fE, f E Aut G.
It is a surjective homomorphism of groups. If a and b are two isolated

vertices of G or if the edge ab together with a, b is a component of G then the
transposition (a, b) is an automorphism of G, and moreover if f = (a, b) then
IE is the identity mapping, i.e., f E Ker cpo Hence if one of the conditions
( 1), (2) is violated then
AutEG ~ Aut G/Ker cp 1- G.
Assume now that both these conditions hold, f E Ker cp and f(a) ::/= a for
some vertex a E VG. Suppose that x = ab E EG. Since x = fE(X) =
f(a)f(b), we have f(a) =
b, feb) =
a. There exists a third vertex c E VG
adjacent to either a or b; assume, e.g., that Xl ac E EG. Set d = f(c). =
Then fE(xd = bd =
Xl, b =
c. The resulting contradiction shows that
Ker cp = (e), and hence cp is the isomorphism of groups.
c: Solution: For any vertex v E G let Qv denote the clique of L( G) that
consists of the edges of G incident to v. If Q is the set of all such cliques
then the mapping
a:VG-Q, V 1--+ Qv, vEVG
is the isomorphism of G onto the subgraph F of the clique graph Q(L(G»

induced by the vertex set Q (see Exercise 1.3.41). If G has no triangles then
Q is the set of all maximal cliques of L(G). If G has triangles then, by the
assumption, it has no vertices of degree 3. Therefore the cardinalities of

the cliques from the set Q are not equal to 3 and the cardinalities of the
remaining maximal cliques of L( G) are equal to 3.
Suppose now that g E Aut L(G) and 9 is the corresponding automorphism
of Q(L(G)), see Exercise 1.6.17. From the above discussion it follows that Q
is invariant with respect to g. If g' is the restriction of 9 onto Q (see Exercise
1.6.16) then
Consider now the automorphism fE E AutEG. If
then
Similarly, feb) = v. Hence JE(e) = f(a)f(b) = uv. On the other hand,

{e} = Qa n Qb. Therefore
=
But QunQv {uv}, hence gee) uv = = JE(e). This proves that Aut L(G) ~
AutEG, hence Aut L(G) = AutEG.
Under the considered conditions AutEG ~ Aut G, see item b.
1.6.19. a: The group Aut L(K4) is similar to the wreath product S2 Wr S3,
=
and \ Aut L(K4)\ 48. Hint: L(K4) 3K2 • =
b: Solution: The degrees of all vertices of K5 are 4, hence (see Exercise
1.6.18c)
Aut L(K5) = AutEK5 ~ Aut K5 = S5.
The symmetric group S5 is generated by the cycle a = (1,2,3,4,5) and the

=
transposition b (1,2). Hence Aut L(I<5) (al, bl ), =
al = (12,23,34,45,51)(13,24,35,41), bl = (13,23)(14,24)(15,25).
1.6.20. a: Solution: Let G be the Petersen graph. Indexing its vertices as shown
at Fig. A1.1.4, we see that G = L(K5) (see Exercise 1.1.40). Hence
Aut G = Aut L(K5) = Aut L(K5).

Further, see Exercise 1.6.19b.
1.6.22. AutEG coincides with the Klein four-group.

1.6.23. Solution: Assume that V = {1,2, ... ,n} is the vertex set for all labelled
graphs isomorphic to G. If G 1 is one of these graphs, then reindexing its
vertices by means of a permutation s E S(V) we obtain the graph s(Gd.
The equality s(Gd = t(Gt), s, t E S(V), holds if and only ifrls(G!) = G l ,
i.e., t- 1 s E Aut G1.
In other words, the permutations s, t must belong to the same left coset of
S(V) modulo the subgroup Aut G 1 . Therefore the number of graphs iso-
morphic to G is equal to the number of these cosets, i.e., n!/I Aut Gd. But
since the groups Aut G1 and Aut G are similar, I Aut Gli = I Aut GI.
1.6.25. Let G be a circulant graph of order n. Under an appropriate indexing ofthe
vertices the adjacency matrix A( G) coincides with some polynomial in h (see
Exercise 1.5.15), where s = (1,2, ... , n) is the cyclic permutation of length
n. Clearly, hl(h) = I(h)h, therefore Ms E Aut G (see Exercise 1.6.24).
Conversely, let Aut G contain a cyclic permutation of length n. We may
index the vertices of G in such a way that s = {1,2, ... ,n}. Under this
indexing, the adjacency matrix A(G) will commute with Ms, i.e., MsA(G) =
A( G)Ms. But the characteristic and minimal polynomials of M. coincide,
therefore all matrices commuting with Ms are polynomials in M. of degree
at most n - 1. Thus, A(G) = I(Ms), i.e., it is a circulant, therefore G is a
circulant graph.
1.6.26. b: Hint: Let I : V F -+ VG be an isomorphism of graphs F and G and let
a be a vertex of maximal degree in F. Suppose further V F n V G ::J 0 and let
<p be the permutation of the set V F U VG specified by the conditions
( ) _ { I(x), if x E VF,
<p x - 1- 1 (x), if x E VG.
Then b = I(a) and <p satisfies the assumptions of the exercise.

Conversely, if <p E Aut Hand <pC a) ::J a then b = <pc a) is a vertex of maximal
degree in G and the mapping
I: VF -+ VG, x 1-+ <p(x),x E VF
is an isomorphism of graphs.
Trees
2.1 Trees: Basic Notions

2.1.2. No.
2.1.3. Hint: The number of such trees is 16.
2.1.4. Hint: The number of such trees is 3 for n = 5, 6 for n = 6, 11 for n = 7, 23
for n = 8.
2.1.6. Stars.
2.1.8.2,3, ... ,n-1.

2.1.9. ]{1 and P4·
Solution: Let T be a selfcomplementary tree with n vertices. Then IETI =
IETI, i.e.,
n - 1 = n(n - 1)/2 - (n - 1),
hence n = lorn = 4. An immediate check gives the answer.
2.1.10. ]{1 and K 2 .
2.1.11. Hint: Use the Theorem 2.1.1.
2.1.12. Solution: For any n-vertex tree the handshake Lemma and the Theorem
2.1.1 imply
(A2.1.1)
;=1
where d; is the degree of the i-th vertex. Let k ;::: 2 be a fixed integer. Since
d; i= 2 and the number of pendant vertices (d; = 1) is equal to k, it follows
that n is limited (otherwise the left side of (A2.1.1) could be arbitrarily
large). Therefore (h" is a finite set.
2.1.13. 2/(2 - dmean ).
207
2.1.14. 2 + Li(di - 2), where the summation is over all branching vertices.
Hint: Use the equation (A2.1.1).
2.1.15. a: Solution: Let a tree have n > 1 vertices. Then by Theorem 2.1.1 it has
n - 1 edges. Therefore by Handshake Lemma, I:~=1 d; = 2n - 2, where di
is the degree of the i-th vertex. Since d; > 0, the latter equality holds only
if at least two right-hand terms are equal to 1, i.e., the tree has at least two
pendant vertices.
b: Hint: Proof is by induction, based on the uniqueness of the path connect-
ing any pair of vertices in a tree.
2.1.17. c: Solution: If D = D(T) is the matrix of distances between the pendant

vertices of a tree T then T is said to be a realization of the matrix D. We
shall prove that all realizations of the matrix D are isomorphic using the
induction over the number of pendant vertices. This is evident if the order k
of D is 2, since every tree with two pendant vertices is a path (see Exercise
2.1.15b ).
Suppose that k > 2 and let all realizations of D(T') for any tree T' with k-l
pendant vertices be isomorphic, i.e., D(T') has a unique realization. Let
further T be a tree with pendant vertices 1,2, ... , k. Then T has branching
vertices. Let a be the one closest to the vertex k and let A denote the simple
(k, a)-path whose length is denoted by I. Deleting all vertices of A - a from
T, we obtain a tree T' whose matrix D(T') may be obtained from D(T) by
the deletion of the last row and the last column.
Clearly, the tree T is uniquely constructed from T' if we know the length I
and the branching vertex a E VT'. Therefore the problem will be solved if
we express I in terms of the elements of D(T) and indicate the vertex a in
T ' , since by the induction hypothesis, all realizations of the matrix D(T')
are isomorphic.
B
e-
io jo
Figure A2.1.1: To Exercise 2.1.17, c
Let band c be two different vertices of T' adjacent to a; B = (a, b, ... , io),
C = Ca, c, ... , jo) are simple paths in T', with io, jo being pendant vertices,
see Fig. A2.1.1. Then AU B is the simple (k, io)-path, AU C is the simple
2.1: Trees: Basic Notions 209
(k,jo)-path, B U C is the simple (io,jo)-path. Therefore if x and yare the

lengths of paths Band C respectively then
Dkio = X + I, Dkjo = Y + I, Diojo = X + y.

Therefore 1= (Dkio + D kjo - Di ojo)/2.
On the other hand, for arbitrary pendant vertices i, j, the simple (i, j)-,
(i, k)-, and (j, k)-paths have a single common vertex t, see Fig. A2.1.2, with
Figure A2.1.2: To Exercise 2.1.17, c
Therefore the length I of the simple path A may be determined as follows:
I = min {( Dki + Dkj - Dij )/2: 1~ i<j ~ k- I} .
Further, if io, io are the values of i, i which deliver the minimum, then the
vertex a belongs to the simple (io,io)-path of T', and its location at this
path is specified by the condition
d( io, a) = Diok - l.
d: Hint: Use of the solution for item c.
2.1.18. nd d~1+ 2, where d is the degree of a nonpendant vertex.

Solution: The Handshake Lemma for a tree of order n implies 2:7=1 di =
2(n - 1), where di is the degree of the i-th vertex. Since all nonpendant
vertices of T have the same degree d,
den - p) + p = 2{n - 1),
where p is the number of pendant vertices in T, hence the formula.
2.1.19. Hint: Modify the solution of the Exercise 2.1.18.

2.1.20. a: Hint: Use the answer for the Exercise 2.1.14.

b: The bound is attained only for trees with vertex degrees either 1 or 3.
2.1.21. a: Solution: Let P be a diametral (a, b)-path of a tree T. Then a, bare

pendant vertices of T (see Exercise 2.1.16). Consider any vertex c of T not
in P. T has a unique (a, c)-path Q and the latter is the union Q = 5 U R,
where S is the (a, d)-subpath of P and R is a (d, c)-path with d ::j:. a, see
Fig. A2.1.3. Let x, y, z be the lengths of the paths S, P\S, R respectively
a b
Figure A2.1.3: To Exercise 2.1.21a
and suppose that, e.g., x ::; y. Then we know the excentricities: e(d) =y,
e(c) = z + y, and hence e(d) < e(c), i.e., c is not central. This proves that
the center of T is in P.
b: Hint: This follows from the item a.
c: Hint: Bearing in mind the item a, it is sufficient to demonstrate the
Jordan theorem for the simple path Pn . In this case for odd n the center is
a single vertex and for even n the center is two adjacent vertices.
d: Hint: When passing from T to T' (as well as at all similar steps of
the algorithm), the excentricity of all remaining vertices is decreased by 1,
therefore the centers of T and T' coincide.
e: Hint: See Exercises 2.1.21a,b, as well as the hint to the Exercise 2.1.21c.
2.1.25. The graphs that may be obtained from a star by replacing each edge by a
path (all paths are of equal length).
Hint: First prove that the graph is a tree.
2.1.26. Hint: To do this, prove the inequality:
2r(T) - 1 ::; d(T) ::; 2r(T)
using the Exercises 2.1.21 and 1.2.26a.
2.1.28. 2 or 3.
2.1.29. Hint: Two paths VI,V2, ... ,Vk and UI,U2, ... ,Uk are said to be equal if
and only if either Ui = Vi, i = 1,2, ... , k or Ui = Vk-i+l, i = 1,2, ... , k.
Solution: Such tree has a simple path P = VI, V2, ... , V2k-2 of length 2k -
3. This path contains k - 2 subpaths of length k starting at the vertices
VI, V2, ... , Vk-2' In addition, there are n - 2k - 2 vertices not lying on the
path P. Each such vertex U is connected with P by a simple path Qu. If Qu
is longer than k - 1 then the required path of length k is taken to be the
segment of Qu of length k starting at u. Otherwise the required path consists
of Qu plus a part of P. Hence there additionally exist n - 2k + 2 simple paths
of length k. Hence the tree contains at least k - 2 + n - 2k + 2 = n - k simple
paths of length k.
2.1.30. a: Hint: See Table JQ.1.1.
Table JQ.1.1:
1 2
2
...<
--< •
• •
b: Hint: See Fig. A2.1.4, where VI and Ul are the central vertices and V2, U2
are the centroidal vertices .
• • • • • •
Figure A2.1.4: To Exercise 2.1.30b
c: Solution: Let v be a centroidal vertex of the tree T. The branches at this

vertex are denoted by Hi, i = 1,2, ... , k = deg v, i.e., Tv = {HI, H2, ... , Hd.
A branch Hi E Tv such that IV Hi I = w{ v) will be called a branch weighted
at v. Let HI be such branch, i.e., IV HII = w(v), and let el be the appending
edge of the branch, i.e., el =
(v, u). The branches at the vertex u are denoted
by HI, i = 1,2, ... , p = deg u, where el E EHf, see Fig. A2.1.5. Let us
Figure A2.1.5: To Exercise 2.l.30c
demonstrate that H~ is the branch weighted at u. From Fig. A2.l.5 we see

that
IEHll = L IEH:I + l.
i#
Therefore if the branch weighted at u were a branch Hj, j =I 1, then it would

be
IEHll ~ w(u) + 1
or w( v) > w( u). But this would contradict to the fact that v is a centroidal
vertex. Hence H~ is the single branch weighted at u.
Case 1: The vertex u is not centroidal, i.e., w(u) > w(v). Let us demonstrate
that in this case T has a single centroidal vertex (which is v). Indeed, Fig.
A2.l.5 shows that no vertex v' E Ui# V Hi, v' =I v, can be centroidal since
it has a branch that strictly contains H l , i.e., w(v' ) > w(v).
Similarly, every vertex v" E Ui# V HI has a branch that strictly contains
H~. Therefore w(v") > w(u). But in view of w(u) > w(v) this means that
v" cannot be centroidal.
Case 2: The vertex u is centroidal. Repeating the reasoning similar to the
previous case, we verify that no vertex other than u, v can be centroidal in
T.
d: Hint: See the solution for the Exercise 2.1.30c. Basing on Fig. A2.1.2,
the equality w(v) + w(u) = IETI + 1 is evident, which implies the required
statement.
e: Hint: See Exercise 2.1.30d.
2.1.31. Hint: a: The proofs of both the necessity and sufficiency are by contradic-
tion.
b: To prove the necessity, verify that every subtree of an interval tree is also
an interval tree, and the tree shown at Fig. 2.1.3 is not the interval one.
The sufficiency is easily proved by direct construction of the required bijec-
tion.
2.1.32. The tree of 7-th order shown at Fig. A2.1.6.

2.1.33. Kl and K 2 ·
Hint: If a graph has the transitive automorphism group, it is regular. Next,
see Exercise 2.1.10.
2.1.34. Let T be a tree of order n and let G be a graph with 8( G) 2:: n-1. We prove
that G has a subgraph isomorphic to T. For n = 2 the statement is evident.
Let us apply induction over n. Assume that n > 2 and the statement is true
for trees with less vertices. If T' is a tree of order n - 1, then by the induction
hypothesis we may assume that T' is a subtree of G. For any a E VT' the
following inequalities hold:
degT' a ~ n - 2, degG a 2:: n - 1.
Hence there exists a vertex bENG(a) \ VT'. Adding the vertex b and the
edge ab to T', we obtain a subgraph of G which is a tree of order n. Since
every tree of order n is obtained from some tree of order n - 1 by adding a
new vertex and a new edge, the required statement is proved.
2.1.35. a: Simple paths. b: Stars. c: Trees of diameter less than k - 1. d: Simple
paths or K1,m. e: Trees with maximal degree at most 3.
2.1.36. Hint: See Exercise 1.5.13.
2.1.37. a: (3,3,4,4,6,6).
b: (3,3,4,4,6,6).
c: (8,5,8,6,5,6).
d: (6,6,7,7,7,8,8,9,9,8,12,13,13).
e: (3,5,5,3,2,8,2,1,10,10,13,13,12,12,10).
2.1.38. Hint: Using induction over n 2:: 3 prove the equivalent statement that
is the set of all nonpendant vertices of the graph.

2.1.39. a: Hint: Use the previous exercise. b: See Fig. A2.1.7.
2.1.40. a: Hint: Proof may be carried out by contradiction using the algorithm of
the reconstruction of a tree by its Priifer code (see Exercise 2.1.39).
b: Hint: Use the algorithm of the reconstruction of a tree by its Priifer code.
1 2
7 WJV 3
2.1.41. The star with at least three rays.
2.1.42. Pn , n ~ 4.
2.1.44. Hint: According to the Exercise 1.3.32, the family peG) allows to deter-
mine the number of edges of G and whether G is connected.
2.1.45. a: Solution: We shall use induction over k. Two subtrees surely satisfy our
statement. Suppose that any k - 1 ~ 2 pairwise intersecting trees have a
common vertex. This, in particular, means that
VI = n
k-1
;=1
V'Ii # 0, V2 = nk
;=1
V'Ii # 0, V3 = VT1 n VT2 # 0.
It is sufficient to consider the case when there exist three different vertices
otherwise the statement would obviously be true. For the graph T, let PI
denote the (u, v)-path, let P2 denote the (v, w)-path, and let P3 denote the
(v, w)-path. Since T is a tree,
(otherwise T would contain a cycle). At the same time, every tree 'Ii, i =
1,2, ... , k, contains at least two vertices from {u, v, w}. Therefore every tree
'Ii, i = 1,2, ... , k, contains all vertices of one of the paths PI, P2, P3. There-
fore,
0# V4 ~ n
;=1
k
V'Ii.
b: Hint: It is sufficient to prove the connectivity of the induced subgraph.

2.1.46. Solution: Let P be a path oflength e(x) starting at x. If P does not contain
the vertices y and z then clearly
e(y) = e(x) + 1, e(z) = e(x) + 1,

and the statement is proved. If otherwise P contains one the vertices y, z,

say, y, then
e(y) ~ e(x) - 1, e(z) e(x) + 1,=
and the statement is proved just the same.
2.1.47. Solution: First of all let us determine the relation between u(x!) and U(X2)
when Xl, X2 are adjacent vertices ofT. Denote e = X1X2; T-e consists of two
components T l , T2. Suppose that ITll = nl, IT21 = n2, i.e., nl +n2 = ITI.
Fig. A2.1.8 shows us that
u(x!) = U(X2) - nl + n2. (A2.1.2)
Let now y and z be the vertices ofT adjacent to x. Denote el = xy, e2 = xz;
T- el - e2 consists of three components K l , K 2, K3,
Y E Kl , Z E K2, X E K3,
IKd = kl' IK21 = k2, IK 31 = k3,

see Fig. A2.1.9. Clearly, ki > 1, i = 1,2,3, kl + k2 + k3 = ITI. Bearing in
K3
mind the relation A2.1.2, we may see from Fig. A2.1.9 that
u(x) = u(y) - k2 - k3 + kl'
u(x) = u(z) - kl - k3 + k2,
hence
2u(x) = u(y) + u(z) - 2k3 < u(y) + u(z).
2.1.48. a: Solution: Suppose that T has two nonadjacent vertices u, v such that
the value IT( u) = IT( v) is minimal. Consider the (u, v)-path of T
Xl = U, X2, ... , Xp-l, xp = v, p> 2.

Since IT(u) ~ IT(v), strict convexity of the function IT(x) (see Exercise 2.1.47)
implies
IT(xd + IT(X3) > 2lT(X2) ~ IT(xd + IT(X2),
IT(X3) > IT(X2) ~ IT(xI),
IT(X2) + IT(X4) > 2lT(X3) ~ IT(xd + IT(X2),
IT(X4) > IT(xI},
and so on. Finally, we derive
IT(v) = IT(xp) > IT(l) = IT(u),

which contradicts to our assumption.
b: Solution: Consider a tree T of order m + n :
where Pm is the (VI, vm)-path (VI, V2, ... , vm ) and ei = ViUi, i = 1,2, ... , n.
If m is even then the center of T is the vertex V m /2. At the same time if
n = C!, then VI is the barycenter. Therefore the distance between the
center and the barycenter is m/2 - 1, i.e., it may be arbitrarily large.
2.1.49. a: A star graph. b: A path graph.
Solution: a: For any tree,. the expression L(x,y) d(x, y) has exactly n - 1
terms equal to 1. The remaining terms are at least 2 and they are equal to
2 only for the star.
b: Let a be a pendant vertex of T. Then
d(x,y) = IT(a) + d(x, y), (A2.1.3)

(x,y): x,yEVT (x,y): x,yEV(T-a)
where IT(a) is defined in Exercise 2.1.47.

If T is a path then IT(a) = 1 + 2 + ... + (n - 1). If T is an arbitrary tree
and e(a) = e (see Exercise 2.1.46) then for every i = 1,2, ... , e there exists
a vertex at distance i from a. Therefore
IT(a) = 1 + 2 + ... + e + el + e2 + ... + en-I-e,

where ek < e, k = 1,2, ... , n - 1 - e.
Clearly,
1 + 2 + ... + (n - 1) ~ 1 + 2 + ... + e + el + e2 + ... + en-I-e,

where the equality takes place only if e = n - 1, i.e., when T is a path.

Thus, the first term 0'(1) in (A2.1.3) is maximal if and only if T is a path.
The maximality of the second term on the right side of (A2.1.3) in the case
when T is a path is proved by induction over the number of vertices.
2.1.50.
n
t(n, k) = k(t(n - 1, k) + (n - k)t(n - 1, k - 1), n ~ 3, k ~ 1.
Solution: Assume that the vertex set of the tree is {1, 2, ... , n}. Clearly,
t(2, 2) = 1, t(n, 0) = t(n, 1) = t(n, n) = 0, n ~ 3.
Let T be the set of labelled trees of order n with k pendant vertices, 2 ~

k ~ n - 1, n ~ 3. Let 'Ii, be the subset of T of trees whose i-th vertex is
pendant. It is easily proved that
I'Ii, I = kt(n - 1, k) + (n - k)t(n - 1, k - 1).
But T = U7=1 1£ , and moreover, every tree from T belongs to exactly k sets
'Ii,. Therefore,
n
t(n, k) = ITI = 11k L I'Ii, I = nlk(t(n - 1, k) + (n - k)t(n - 1, k - 1)),
i=1
1 ~ k ~ n - 1, n ~ 3.
2.1.51. Solution: Let T be a tree, ITI = 2k + 1. Suppose that there exists an
automorphism f for which no vertex of T is stationary. Choose a maximal
subtree T' such that f(VT') n VT' = 0. Since ITI = 2k + 1, there exists a
vertex v E VT such that v fj. VT', v fj. f(VT'), but v is adjacent to a vertex
of the tree T'. Therefere
f(V(T' + v)) n V(T' + v) = 0,
contrary to the maximality of T' .
2.1.52. Hint: Let v be a vertex of T and let v' be a vertex maximally distant from
v, i.e., e(v) = d(v, v'). To prove the required formula, it is sufficient to prove
that all vertices of C(T) belong to the (v, v')-path. To this end, use the
Exercise 2.1.21.
2.1.53. Solution: Suppose the contrary, i.e., the graph SI(T) has an induced cycle
CIc = (vo, ... , VIc-1, vol, k ~ 4. Let 1£ denote the subtree of T that corre-
sponds to Vi, i = 1,2, . .. ,k. Consider Wi E Vl£ n Vl£+1, i = 0,1, . .. ,k-1
(here and below the summation of indices is modulo k). Since CIc is an in-
duced cycle and k ~ 4, all vertices Wo, ... , WIc-1 are pairwise different. Let
Qi denote the unique simple path connecting Wi and Wi+! , i = 0, 1, ... , k - 1.
Let further Xi+1 denote the common vertex of Qi and Qi+1 which is most
distant from Wi+1, i = 0,1, ... , k - 1. It is easily seen that the union of
the (Xi, xi+d-subpaths Qi constitutes an induced simple cycle in T, which
is impossible.
2.2 Skeletons and Spanning Trees

2.2.1. Solution: Let an (n, m)-graph 0 have k components, k > O. Clearly,
mi ~ ni - 1, i = 1,2, ... , k,
where mi is the number of edges of the i-th component, ni is the number of

vertices of O. Summing the inequalities, we have
k k
m = L mi ~ L(ni - 1) = n- k,
i=1 ;=1
i.e., v(O) = m - n + k ~ O.
2.2.3. Hint: The solution immediately follows from the definition of the cyclic rank
of graph.
2.2.4. Hint: The numbers of such skeletons are: a: 16; b: 1; c: 7; d: 1; e: 12.
2.2.5. a: 2; b: 3; c: 6; d: 1; e: 4; f: 2; g: 5.
2.2.6. a: 21; b: 15; c: 81; d: 64.
2.2.7. In the case when the graph is acyclic.
2.2.8. Solution: Choose an arbitrary vertex of the graph and complement the set
of edges incident to it to obtain a skeleton of the graph. Deleting the edges
of this skeleton from the graph, we obtain an isolated vertex.
2.2.9. Hint: Let 0 be a connected graph. Pick any vertex and label it by O. Let v be
the vertex with the smallest label i whose neighborhood contains unlabelled
vertices Ul, ... , Uk. We label all these vertices by i + 1 and select all edges
VU1, ... ,VUk. Repeat this until all vertices are labelled. The selected edges
constitute a skeleton.
2.2.10. 1: Hint: See Theorem 2.1.1.
3: Hint: See Theorem 2.1.1.
2.2.11. Hint: See Theorem 2.1.1.
2.2.12. Hint: The minimal(maximal) weight of the skeleton IS: a: 37(59); b:
28(60); c: 18(60).
2.2.15. The number of skeletons of 0 is the product of the numbers of skeletons
of 0' and 0".
2.2.16. Solution: For any edge e from a cycle one may construct a skeleton that
contains all edges of the cycle except of e.
2.2.17. No. Hint: See Exercise 2.2.17.
2.2: Skeletons and Spanning Trees 219
2.2.18. Pn , n 2: 2 and Cn, n 2: 3.

2.2.20. a: Hint: Use breadth-first search from v (see Exercise 2.2.9).
b: Hint: See Exercise 2.1.21. Apply Exercise 2.2.21a for every central vertex.
2.2.22. a: Yes. b: No.
2.2.23. Both statements are false. Hint: a: Consider the cycle C4 . b: Consider
K n , n2:3.
2.2.24. nn-2.
2.2.27. Solution: Since the number of labelled trees of order n is nn-2 (Cayley
Theorem), the total number of the outcomes is nn-l. Let us count the num-
ber of favorable outcomes, i.e., when the picked vertex is pendant. Its label
may be one of numbers 1,2, ... , n. It may be adjacent to any of the remain-
ing n - 1 vertices of the tree T - v. The number of labelled trees of order n-l
is (n - 1 )n-3. Thus the number of favorable outcomes is n( n - 1)( n - 1 )n-3.
Therefore
· Pn = l'1m (n-l)n_2
11m -- = 1/ e.
n--co n-+oo n
2.2.28. a: Hint: Without loss of generality assume that the graph G is connected.
If we delete an edge e from the skeleton T 1 , it becomes a two-component
graph with the vertex sets denoted by VI and V". The tree T2 has an edge
f = u l u" such that u l E VI, u" E V". It is easily seen that (Tl - e) + f is a
skeleton of G.
b: Hint: Use item a of the exercise.
2.2.29. a: See Fig. A2.2.1.

b-e: Hint: See Exercise 2.2.29.
f: Connected graphs that have two edge-disjoint skeletons.
g: For the graph of order n shown an Fig. A2.2.2 this difference is n - 3,
i.e., it may be arbitrarily large.
~~. n - 2 vertices
Figure A2.2.2: To Exercise 2.2.29g
h: v*(G), which is equal to n - k.

i: Graphs for which every component has two edge-disjoint skeletons.
2.2.30. See Fig. A2.2.3.
UC:JH71zk:Y1 3
Ta
4 3
T5
4 3
T6
4 3
Ts
4
~----------------~~T6
~~--------------~T5
S(G)
2.2.31. Let (G, w) be a connected weighted graph and let T be its minimal span-
ning tree which cannot be constructed by Kruskal's algorithm. Let us order
the list of weights of the edges of T by magnitude: Wl ::; W2 ::; ..• ::; Wm and
reindex the edges ofT correspondingly: w(ed Wi, i = =
1,2, ... , m, with the
2.2: Skeletons and Spanning Trees 221
additional requirement that among possible indexings we select the one for
which an application of Kruskal's algorithm produces the maximal number
ofthe first successive edges el e2, ... ek of this ordered edge list. By our sup-
position, k < m. Since G is connected, the (k + l)-st iteration of Kruskal's
algorithm is possible. Let e~+l be the edge selected at this iteration. Clearly,
Wk+l > w( e~+l)' and therefore e~+l ~ ET. The graph T + e~+l has a single
cycle C. It contains the edge e~+l and some edge ep with p 2: k+ 1. The graph
T' = T+e~+l -e p is a tree such that w(T') = w(T)+w(e~+l)-wP < w(T).
This contradiction proves the statement of the exercise.
2.2.32. Solution: Assume that
and
ET' = {e~, e~, ... , e~}, w(eD = di , i = 1,2, ... , m.
By Exercise 2.2.31, we may assume that T = Tm is a spanning tree con-
structed by some run of Kruskal's algorithm. Suppose that there exists a
number k such that w(ek) > w(eD. Let us choose the minimal possible k.
Clearly, k> 1. Let T k - 1 denote the spanning forest of G constructed at the
(k - l)-th iteration of Kruskal's algorithm. Then because of w(ek) > w(eD,
for every ej, j = 1,2, ... , k, either ej E ETk-l or Tk-l + ej contains a cycle,
otherwise ek would not have been selected at the k-th iteration. Therefore
both ends of ej belong to the same component ofTk _ 1 . Let S be a component
of Tk-l, lSI = p. Since the edges of the set E' = {e~, ... , eD produce no
cycles, E'nE(G(S))1 ::; p-1 = IESI (recall that G(S) denotes the subgraph
of G induced by S). Summing these inequalities over all components S, we
= =
obtain a contradiction: k - 1 IETk-ll LS IESI 2: IE'I k. =
Independence and
Coverings
3.1 Independent Vertex Sets and Cliques

3.1.1. a: 3; b: 6; c: 4; d: 4; e: 3; f: 5; g: 8.
3.1.2. a: 1; b: n; c: max{n, m}; d: n/2 for even n, (n + 1)/2 for odd n; e: n/2 for
even n, (n - 1)/2 for odd n; f: 4; g: 2n - 1 .
Hint: Look for a pattern for small m, n and then proceed by induction.
3.1.3. Tetrahedron: 1; cube: 4; octahedron: 2; dodecahedron: 8; icosahedron: 3.
Hint: The dodecahedron graph D contains an independent set of cardinality
8, see Fig. A3.1.1, i.e., O:o(D) 2 8. On the other hand, every face of the
dodecahedron contains at most two independent vertices. Since every vertex
223
belongs to 3 faces of the dodecahedron and there are 12 faces, we may write
3ao(D) ~ 2·12.
3.1.4. a: On; b: Kn.
3.1.5. No.
3.1.6.
a: ao(G) - 1 ~ ao(G - v) ~ ao(G).
b: ao(G) ~ ao(G - e) ~ ao(G) + 1.

c: ao(G) - 1 ~ ao(G + e) ~ ao(G).
Here v is the vertex being deleted, e is the edge being inserted or deleted.
3.1.7. No.
3.1.B. Hint: To proof the sufficiency, consider a cycle of odd length in the graph.
3.1.9. Hint: Take advantage of the fact that a tree is a bipartite graph.
3.1.10. Hint: Notice that either a pendant vertex or the one adjacent to it, but
not both, belongs to a maximum cardinality independent set.
3.1.12. No. See, e.g., Fig. A3.1.2
3.1.13. a: l;b: l;c: min{m,n};d,e: rn/31;f: 3;g: 2n - l .
3.1.14. Hint: Consider the stars Kl,n, n ~ 2.
3.1.15. Hint: c: If Xl, X 2 are independent sets of graphs G and H respectively

then Xl x X 2 is an independent set of the graph G x H.
3.1: Independent Vertex Sets and Cliques 225
3.1.16. Hint: a: A proof is by induction over the number of vertices of graph.

Clearly, (3.1.1) holds for n = 2. Suppose that !VGI = n > 2 and the inequal-
ity holds for graphs of order less than n. Let G' denote the graph obtained
from G by the deletion of a vertex x of minimal degree and of all vertices of
its neighborhood. If I' is a maximum cardinality independent set of vertices
of G', then the set I = I' U {x} is independent in G. To prove (3.1.1), it is
sufficient to prove that
S1 = L: (1 + degG v)-l ~ S2 = L: (1 + degG' v)-l + 1.

vEVG vEVG'
b: Consider the complete graphs.

c: Consider an algorithm such that at each iteration a vertex of minimal
degree is put into the required independent set X and then deleted from the
graph together with its neighborhood.
3.1.17. Hint: Use formula (3.1.1) and the Cauchy-Bunyakowsky inequality
where ai, bi ;::: 0, assuming ai = 1+deg Vi, bi = (l+deg Vi)-l for all Vi E VG.
3.1.18. Solution: Since the set I is maximal, every vertex of VG\I is adjacent to
at least one vertex from I, Le., !VG\II ~ p(I). Adding III to both sides, we
arrive at the required inequality.
3.1.19. Hint: Use the previous exercise and the fact that Ll(G)III;::: p(I).
3.1.20. Solution: Let
be a required partition of VG into the minimal number of paths such that

I( G) ;::: ao( G) + 1. Then among al, ... , a, a pair ai, aj of adjacent vertices
exists. Linking Pi,Pj into a path P = (bj, ... ,aj,ai, ... ,bi ), we obtain
the partition of the vertex set into less number of paths, contrary to the
assumption.
3.1.21. Solution: Every vertex of a maximum cardinality independent set of G is

incident to at least c5(G) edges. Hence IEGI;::: ao(G)c5(G).
3.1.22. Hint: Denoting the independence number of G by a, reindex the vertices
of G in such a way that its adjacency matrix assumes the form A( G) = A =
[~ g], where 00' denotes the zero a x a-matrix.
Prove that rank A ~ 2(n - a). Then notice that n = pO + p+ + p- and

rank A = p+ + p- .
3.1.23. Solution: From one hand, ao(G') ~ ao(G), since G is an induced subgraph
of G'. From the other hand, if an independent vertex set I of G' contains u',
it does not contain u, since u is adjacent to u'. Since G ~ G' - u, replacing
u' by u, we obtain an independent vertex set of G, i.e., ao(G') S ao(G).
Finally, if an independent vertex set I of G' does not contain u, it is also
independent in G.
3.1.24. Solution: a: Let u, v be adjacent in G~ a with deg v > deg u. Deleting v
from the graph and splitting u (see Exercise 3.1.23), we obtain a graph G'
of order n with less edges.
From the other hand,
a = ao(G~a) ~ ao(G~a - v) = ao(G'),

which is a contradiction.
b: Let u, v be nonadjacent in G~ a with deg v > deg u + 1. Deleting v from
the graph and splitting u (see Ex~rcise 3.1.23), we obtain a graph G' of order
n with less edges and with ao(G') S a, which is a contradiction.
c: Assume that a vertex u is adjacent to v and wand v is not adjacent to
= =
w. By item a of the exercise, deg v deg u deg w. Delete v, w from G~ a
denoting G' = G~ a - V - w. Splitting u in G' and then splitting it again
in the resulting gr~ph, we obtain a graph Gil of order n with less edges and
with ao(G") S a. The resulting contradiction means that v, ware adjacent
to each other. This implies that every component of G~ a is a complete
graph. '
d: Take G = G~ a; k is the number of its components. From the above it
follows that '
G = qKp+1 u (k - q)Kp, q S k = ao(G) S a. (A3.1.1)
Without loss of generality assume that
q < k, (A3.1.2)
for otherwise
G = kKp+1 = q' Kpl+1 U (k - q')Kpl, q' = 0, p' = p + 1,

and since n = kp' = ap' + 0, the required relation holds.
By (A3.1.1),
n = kp+ q. (A3.1.3)
(A3.1.2) and (A3.1.3) together imply that p is the quotient and q is the
remainder of the division of n by k. It remains to prove that k = a.
Suppose the contrary, i.e., k < a. Suppose that the division of n by k +1
gives the quotient PI and the remainder ql and
Clearly,
Pl ~ p, n = (k + 1)Pl + ql, ao(Gd = k + 1 ~ a. (A3.1.4)
Let us evaluate the difference t = IEGI-IEGd. We have

21EGI = q(p + 1)p + (k - q)p(p - 1) = p(2q + k(p - 1)).
Similarly, 21EG l i = Pl (2ql +(k+1)(Pl -1)). Taking into account the relations
(A3.1.3), (A3.1.4), we obtain
ql = k(p - pt} + q - Pl, 21EGd = Pl(k(2p - Pl -1) - Pl -1 + 2q).

Therefore,
2t = 2q(p - Pl) + k(p - pt}(p - Pl - 1) + P~ + Pl· (A3.1.5)
Since k + 1 ~ n, it follows that Pl > 0, and (A3.1.5) implies that t > o. The
latter conclusion contradicts to the definition of G, and this contradiction
proves that k = a.
3.1.25.
m = C;+lq + C;(a - q).
Hint: Use the previous Exercise.
3.1.26. Hint: Use the previous two Exercises.
3.1.27. Solution: Denote a = ao(G). Since n/a - 1 < P ~ n/a, the previous
exercise and the fact that n - a ~ 0 together imply that
m ~ p(n - a/2 - ap/2) > (n/a - 1)(n - a/2 - a(n/a)/2) =

n2 a n2
(n/a - 1)(n/2 - a/2) = - - n + - > - - n.
2a 2 2a
2 2
This implies n +m > ~, or ao(G) = a > 2(n ~ m)·
3.1.28. a: Solution: The sufficiency is evident. Let us prove the necessity. Assume
that G has a component K which is not a complete graph. Then ao(K) ~ 2.
Since ao(H) = ao(HP), on may choose a maximum cardinality independent
set N of the graph H which is a maximum cardinality independent set of
HP. Let us choose two vertices u, v from N such that the distance dH(U, v)
is minimal possible. Since u, v are not adjacent in HP, dH( U, v) ~ P + 1.
Consider a vertex w from a shortest (u, v)-path such that d H ( u, w) = 2, see
Fig. A3.1.3; w tI. N, since d(u,w) < p+ 1. In addition, w must be adjacent
to some vertex z from N, otherwise N U { w} would be a larger independent
set. But the (u, w, z)-path is shorter than the chosen (u, v)-path, contrary to
the choice of u, v. Therefore the assumption that K is not a complete graph
is false.
b: For example, P4.
z
u
,... .
v
3.1.30. a: Hint: Delete edges from the graph successively until the required graph
is obtained.
b: Solution: Assume that aa(G) = aa and EG = {e1,e2, ... ,em }. Let us
construct a graph iI as follows:
(l)V iI = VG U V1 U ... U Vm , IViI = aa - 1.
(2)EG ~ EH.
(3) Every vertex v E Vj, j = 1,2, ... , m, is adjacent to all vertices of iI with
the exception of the vertices from Vj and the ends of ej. Clearly, aa( iI) = aa.
Delete an edge ej = ab E EG from iI. Since the set Vj U a U b is independent,
aa(iI - ej) = aa + 1.
Therefore to obtain an aa-critical graph H, it is sufficient to delete "redun-
dant" edges from EiI\EG.
3.1.31. Tetrahedron, icosahedron.
3.1.32. a: On; b: K 1; c: Kn; d: Pn-1; e: K 1 , if n = 3; Cn , if n > 3; f: Km X Kn;

g: the line graph of the Petersen graph.
3.1.33. Hint: Consider a graph induced by the maximal clique.
3.1.34. G = =
G;r'.<p with m C;+1 r + C;(cp - r), where p is the quotient and r is
the remainder of the division of n by cp. IEGI = C~ - m.
Hint: See the Exercises 3.1.24,3.1.25.
3.1.35. Solution: Assume that VG =

{1,2, ... ,n}. Let G' have an induced sub-
graph H isomorphic to G and let
cp: VG --+ VH: i f--+ Vi, i = 1,2, .. . ,n,
be the corresponding isomorphism. By the definition of the modular product

it immediately follows that the vertices (1, V1), (2, V2), ... , (n, vn ) are pairwise
adjacent in GoG' and hence cp( GoG') 2: n.
Conversely, suppose that <p( GoG') ~ n and let C be a clique in GoG'

with n vertices. Suppose that C = {(I, Vl), (2, V2)"'" (n, vn )}. The vertices
=
Vi, i =
1,2, ... , n, are pairwise distinct. If we set H G'( Vl, V2, ... , vn ) then
the mapping i t-+ Vi will clearly be an isomorphism of G and H.
3.1.36. Hint: The solution is similar to that of the previous exercise.
3.1.37. b: Hint: Prove that

Xl + X2 + ... + Xn :s: n
is a separating inequality for the graph On.
c: Hint: Prove that
Xl + X2 + ... + Xn :s: 1
is a separating inequality for the graph Kn.
d: Hint: Let G be a threshold graph of order n, G f:. On, let
(A3.1.6)
be a separating inequality for G and let H be a graph obtained by the

addition of an isolated vertex Vn+l to G.
Then
alxl + a2 x 2 + ... + anXn + an+1Xn+l :s: (3,
where an+l = 0, is a separating inequality for H.
e: Hint: Let G be a threshold graph with a separating inequality (A3.1.6).
Bearing in mind item c, we assume that G f:. On. Let b be the minimum
among the sums alXl + ... + anx n , Xi E {O, I}, i = 1, ... , n, greater than (3.
Then
aiXl + a~x2 + ... + a~xn :s: b
where a: = ai + (b - (3)/n is also a separating inequality for G.

f: Hint: Let G be a threshold graph with a separating inequality (A3.1.6) in
which all coefficients are positive. If H = G+v, where v is a new dominating
vertex then
alXl + a2 X2 + ... + anXn + (3x n +1 :s: (3
is a separating inequality for H.

g: Hint: The separating inequality for the graph G shown at Fig. 3.1.2 is
convenient to look for in the form
It must be:
a,(3,1:S: 6; 2a,a+(3,a+l,2(3 > 6; (3+ 21:S: 6.

We may take a = 5, (3 = 3, 1= 1, 6 = 5.
h: Hint: Let G be a threshold graph with a separating inequality (A3.1.6)

and let H be a subgraph of G induced by the set of vertices {Vi l l · · · , Vik}'
Consider the inequality
(A3.1.7)
where {) is the maximum of the sums l¥i,Xi, + ... + l¥ikXik obtained by the
substitution of the characteristic vectors of the independent sets of vertices
of H into the left side of (A3.1.7).
i: Hint: Suppose that the graph G = 2K2 is a threshold one with EG =
{( Vi, V2), (vg, V4)} and with a threshold inequality l¥lXi + l¥2X2 + l¥gXg +
l¥4 X 4 :::; (3. Then
Summing the first two inequalities and then the second two ones, we obtain
a contradiction.
In a similar way one may prove that P4 and C4 are not threshold graphs.
j: Hint: Use items h), i) of this exercise.
k: Hint: Use items h), j) of this exercise.
3.1.38. Hint: There are 8 such graphs.
3.1.39. a: Hint: The graphs at Figs. 1.2.4a,b,c,d,g have no triangles, therefore

cc( G) for them is equal to the number of edges. For graph e the answer is
8; for graph f it is 6.
3.1.41. See Fig. A3.1.4.
3.1.43. a: GA = GB. b: Does not follow. Suppose for example that
1 1
A= [ 0 0 1 1
1 0] ' B=
3.2: Coverings 231
4
•
Since G A = GB (see Fig. A3.1.5), the sets of integer points of the polyhedra
P(A), P(B) coincide. But for x = (1/2,1/2,1/2,1/2) it is easily seen that
x E PB, and x tI. PA·
3.1.44. Hint: Use the Exercise 3.1.40
3.1.45. Solution: Sufficiency: Let Q = (Qi, i = 1, ... , p) be a covering of a graph
G by cliques satisfying the conditions 1 and 2. Without loss of generality one
may assume that every vertex v belongs to exactly two cliques. Otherwise
one may add a single-vertex clique {v} to the covering Q.
=
Consider the intersection graph Q Q(Q) of the collection Q, where VQ Q =
and QiQj E EQ if and only if Qi n Qj =F 0. Let us prove that L(Q) ~ G.
Recall that V L(Q) = EQ. Consider the mapping t.p: VG -+ V L(Q) such
that t.p(v) = Q;Qj, where v E Q;, v E Qj. The bijectivity of the mapping
follows from the conditions 1 and 2 of the exercise.
Let us prove that the mapping preserves the adjacency relation. Suppose
that vw E EG for v, w E VG. This means that v, w belong to the same
clique Qi of G. Therefore t.p(v) = QiQa, t.p(v) = QiQfj. The edges QiQa and
QiQfj have a common endpoint in Q, therefore the corresponding vertices of
L(Q) are adjacent. The converse implication about the adjacency may be
proved in a similar way.
Necessity: Suppose that G = L(H), where H is a graph. Let Q(v) denote
the set of edges of H incident to a vertex v E V H. Then the collection
(Qu : v E V H) is a covering of G by cliques satisfying the conditions 1 and
2.
3.2 Coverings
3.2.1. a: 3, 3; b: 6, 6; c: 4, 4; d: 6, 5; e: 3, 3; f: 4, 5; g: 8, 8.
3.2.2. a: f30 = Ln/2j; f31 = L(n + 1)/2j.
b: f30 = L(n+ 1)/2j; f31 = L(n+ 1)/2j.
c: f30 = n -1; f31 = L(n + 1)/2j.
d: f30 = 1; f31 = n.
e: f30 = min{m, n}; f31 = max{m, n}.
f: f30 = 6; f31 = 5.
3.2.4. No. Hint: Use the formula (3.2.1).

3.2.5. Stars K1,n.
3.2.6. Hint: If e = uv is a pendant edge of the tree and v is a pendant vertex then
the replacement of v by u does not increase the cardinality of the covering.
3.2.10. For example, paths with even number of vertices and stars.
3.2.11. b: Hint: Use item a) of the exercise.
3.2.12. Hint: Use the fact that the set U is independent if and only if the graph
has no edge with both ends in U, i.e., at least one end of every edge is in U.
3.2.14. Solution: Since O!o(I<m,n) = maxim, n} and /3o(Km,n) = minim, n}, it

follows that O!o(Km,n)/3o(Km,n) = mn = q.
Suppose that G = (X, Y; E) is a non-complete bipartite graph. If O!o( G) =
IXI then /3o(G) = =
IYI, and q < IXIIYI O!o(G)/3o(G).
Suppose now that X = AU B, Y = G U D, An B = 0, G n D = 0, IAI ::f 0,
IGI ::f 0, where AuG is a maximum cardinality independent set of G. Then
the Exercise 3.2.12 implies that BUD is a minimum cardinality covering of
G.
Since X and Yare independent sets of G, we have
IAI + IBI :::; IAI + IGI and IBI :::; IGI,

(A3.2.1)
IGI + IDI :::; IAI + IGI and IDI :::; IAI·
Since the vertices ofthe sets A and G do not adjacent to each other, we have
q :::; IBIICI + IAIIDI + IBIIDI·
Now we use the relations (A3.2.1): q:S IBllel + IAIIDI + IBIIDI :S IBIICI +
=
IAIIDI+IGIIDI < IBIICI+IAIIDI+IGIIDI+IAIIBI (IAI+IGI)(IBI+IDI) =
O!o(G)/3o(G).
3.2.15. Hint: Consider a vertex v of degree o. Prove that a covering of edges
incident to the vertex set v U N(v) contains at least 0 vertices.
3.2.17. Hint: Use the fact that if an edge covering H contains the edges ab, be, ed,
then H - be is also a covering.
3.2.18. Solution: If M is a minimum cardinality vertex covering of a graph G

containing v then the set M\ {v} covers G - v. Hence /30 (G - v) :S 1M 1- 1 =
/30 (G) - 1. Therefore v is a critical vertex.
Conversely, let v be a critical vertex of G. Consider a minimum cardinality
vertex covering M' of the graph G - v. Since the set M' U {v} is a vertex
covering of G and the number of vertices in it is only by 1 more than in M',
then it is of minimal cardinality.
3.3: Dominating Sets 233
3.2.19. Hint: Use the fact that the deletion of an end of an edge leads to the
deletion of the edge itself.
3.2.22. The statement is false. Hint: Consider, e.g., KI,n, n 2: 2.
3.3 Dominating Sets

3.3.1. a: 2; b: 3; c: 2; d: 3; e: 2; f: 2; g: 4.
3.3.2. a: fn/31;b: fn/31;c: l;d: l;e: 2;f: 3.
3.3.5. See Fig. A3.3.1.
3.3.6. a: Solution: Let I be a maximal independent set of vertices of a graph G.

Then every v E VG\I is adjacent to a vertex of I, for otherwise the set
I U {v} would have also been independent. Hence I is a dominating set.
The proof of the converse statement is similar.
b: Hint: Use item a) of the exercise.
3.3.7. Hint: Use the Exercise 3.3.6a.
3.3.8. Hint: To prove the left inequality, notice that
IDI(~(G) + 1) 2: IVGI
for a minimum cardinality dominating set D.
To prove the right inequality, notice that (VG\N(v)) is a dominating set,
where v is a vertex of maximal degree in G.
3.3.9. Hint: Proof is by contradiction.
3.3.10. Hint: We may assume that G is connected. Consider any its spanning tree
T and a vertex v E VT. If D is the set of vertices at even distances from v
in T, then VG\D is the set of vertices at odd distances from v in T, and D
and VG\D are dominating sets.

3.3.12. Hint: Prove that if for some vertex Va E D there are no edges adjacent to
it whose second end is in VG\D then D is not minimal.
3.3.14. Hint: Use the fact that either a pendant vertex or the vertex adjacent to
it, but not both, is in a minimum cardinality dominating set.
3.3.16. Hint: Construct a graph H satisfying the conditions
=
(1) VH VGu VI; VGn VI = 0; IVd = IVGI;
(2) EG <;; EH;
(3) Every vertex of VG is adjacent to exactly one vertex of VI and vice versa.
K ·
3.3.19. Hint: d: See Fig. A3.3.2.
. .
3.3.20. a: 3; b: 3; c: 4; d: 4; e: 2; f: 3; g: 4.
3.3.21. a: rn/31; b: rn/31; c: 1; d: 1; e: min{m,n}; f: 3.
3.3.23. a: See, e.g., Fig. A3.3.2. b: ]{3.
3.4 Matchings
3.4.1. a: 3; b: 6; c: 4; d: 5; e: 3; f: 4; g: 8.
Hint: Each of these graphs has a perfect matching, with the exception of the
graph shown at Fig. 1.2.4f.
3.4.2. a: Ln/2J; b: Ln/2J; c: Ln/2J; d: 1; e: min{m,n}; f: 5.
3.4.5. Hint: For Qn, use the induction.
For ]{2n, the required I-factors are, e.g., the sets of edges
Xi = {ViV2n} U {Vi+jVi_j : j = 1,2, .. . ,n -I}, i = 1,2, ... , 2n - 1,
where the summation of the indices of vertices is modulo 2n - 1.
For the Petersen graph, notice that every its I-factor contains either two
edges of the external cycle or no such edges.
3.4: Matcbings 235
3.4.6. Solution: Set 0:1 =
O:l(G), (31 = (31(G). In order to prove 0:1 + (31 = n, we
shall prove the inequalities
Let M be a maximum cardinality matching in G. Consider the subset V'

of all unsaturated vertices. Clearly, either Viis an independent set or it
is empty (otherwise M would not be of maximal cardinality). In addition,
=
V' \G\- 20:1. If V' -I 0, then for every v E V' we pick in G an edge
incident to it. The set of picked edges is denoted by E'. If V' 0, we set =
E' = 0.
Since G has no isolated vertices and V' is independent,
\E'\ = \V' \ = \G\- 20:1.

Clearly, E' U M is an edge covering, therefore
(31 ::; \E' U M\ = \E'\ + \M\ = (\G\- 20:t) + 0:1 = \G\- 0:1.
Let now P be a minimum cardinality edge covering of G. Consider a subgraph
G' = G(P) induced by the edges of the covering P. Clearly, G ' has no cycles.
Let t denote the number of the components of G and let k i be the number
of edges in the i-th component, i = 1,2, ... , t. Selecting an edge from each
component, we obtain a matching pI of cardinality t. Therefore t :s 0:1. Since
G has no isolated vertices,
t t
\G\ = I)ki + 1) = Lki + t = \P\ +t = (31 +t:S (31 + 0:1·
i=l ;=1
3.4.8. Solution: The proof is by induction over the number of vertices. For K2
the statement is true. Let it be true for graphs of orders less than p. Then
for our purposes it is sufficient to prove that if a graph G of order p has
no induced subgraphs isomorphic to K 1 ,3 then it has two adjacent vertices
whose deletion leaves the graph connected.
Let T be a spanning tree of G and let D = (a, b, c, ... ) be a diametral path
ofT.
If a vertex b is adjacent in T to the vertices a and c only, then G - a - b is
connected.
Suppose now that R = NT(b)\ {a, c} -10. Since D is a diametral path, every
7' E R is pendant in T. Therefore if a is adjacent in G to some 7' E R then
G - a - b is connected.
Let a be adjacent to no vertex from R. If \R\ > 1, then since G has no
induced subgraphs isomorphic to K 1 ,3, the vertices of R induce a complete
graph. Therefore any two vertices of R are the required ones.
If R = {7'}, then by the same reason 7' and care adj acent in G, so a, bare
the required ones.
3.4.9. Hint: Notice that every vertex is incident to at most one edge of every
matching.
3.4.10. Hint: To prove the sufficiency, suppose the existence of a matching Ml
such that IMll > IMI. Next, consider the graph induced by the edges of
Ml EEl M and use the previous exercise.
3.4.11. Use the Exercise 3.4.9.
3.4.12. Solution: Necessity: Let a tree T have a perfect matching M. Consider any
vertex v and the vertices u, Wl, W2, ... , Wk adjacent to it, where the edge vu
is in M. Let T, T 1 , T2"'" n denote the components of T - v that contain
the vertices
u, Wl, W2, ... , Wk respectively. The graphs T 1 , T 2, ... , Tk have perfect match-
ings, for otherwise T would also have no perfect matchings. Therefore the
orders of T 1 , T 2, ... , Tk are even. By the same reason, T U {u} has a perfect
matching. Hence T is of odd order.
Sufficiency: We shall proceed by induction. For K2 the statement is true.
Let it be true for trees with less than n vertices.
Let u be a pendant vertex and let v be adjacent to it. One of the components
of T - v is the vertex u, the remaining ones are denoted by T 1 , T 2, ... , Tk .
From the assumption of the exercise for T it follows that the orders of
T 1 , T 2, ... , Tk are even.
Let us prove that every T; satisfies the assumption of the exercise. Refer
to Fig. A3.4.1 to keep the track of the proof. If W E VT1 then one of
components of Tl - W is obtained from a component of T - W by the deletion
of u, v, VT2, ... , VTk. Therefore the orders of these components are of the
same parity. The remaining components of these graphs coincide.
Proofs for the remaining subtrees Ti are similar.
By the induction hypothesis every tree T 1 , T 2, ... , Tk has a perfect matching.
Their union plus the edge uv is a perfect matching in T.
3.5: Matchings in Bipartite Graphs 237
3.4.14. Hint: Notice first that o(G) = IGI-lVll, where Vl is the largest part, and
furthermore, if LlGI/2 J < 8( G) then IVII < LlGI/2 J, otherwise IVll ;::: LlGI/2 J.
3.4.15. a: Solution: If M is a perfect matching then the inequality is evident.
Suppose that a maximal matching M is not perfect and Vl is the set of ver-
tices not saturated by M. Clearly, Vl is an independent set. Therefore every
vertex v E Vl may be adjacent only to vertices saturated by the matching,
i.e., 6(G) :::; deg v:::; 21MI.
b: Solution: Suppose that there exists a maximal matching Mo such that
IMol < cxl(G)/2. For every edge e E Mo, consider a graph T(e) induced
by an edge e and edges adjacent to it. Since Mo is a maximal matching,
UeEMo T(e) = G. On the other hand, since every graph T(e) contains at
most two independent edges, CXl(G) = CXl(UeEMo T(e)) :::; 21Mol < CXl(G),
which is a contradiction.
3.4.16. a: 3,3; b: 4,4; c: 4,4; d: 4,4; e: 3,3; f: 2,2, g: 6,6.
3.4.17. a: f(n -1)/31, f(n -1)/31; b: fn/31, fn/31; c: Ln/2J, Ln/2J; d: 1,1; e:
min{m, n}, min{m, n}; f: 3,3.
3.4.18. Hint: Notice that if edges uv, vw, wz are in a covering M then M\vw is
also a covering.
3.4.19. Hint: The proof may be carried out by contradiction. Among all weak
coverings, pick the one with maximal number of independent edges and sup-
pose that this covering has adjacent edges. Such edges will induce stars
(see the previous Exercise). Analyze the properties of the covering in the
neighborhood of such star T.
Verify that every pendant vertex of T either belongs to a pendant edge of
the graph or at distance 2 from it there is a vertex incident to an edge e of
the covering, e rf. T. If either it belongs to a pendant edge or the mentioned
distance is 2, then it is easy to construct another covering with less number
of edges.
If the mentioned distance is 1, construct a sequence of edges alternately
belonging/not belonging to the covering. Since the number of vertices of
a graph is finite, this sequence terminates either at a pendant vertex or
at an already traversed vertex. Switching the edges of the covering in an
appropriate way, one may obtain another covering with less number of edges.
3.5 Matchings in Bipartite Graphs

3.5.1. Solution: Consider a bipartite graph
G' = (Xl U X 2 , Yl U Y2 ; Ml U M2)'

Since every vertex of G' is of degree 1 or 2, every its component is either a
path or cycle with edges alternately in Ml and M2, or an edge that belongs
to both Ml and M 2 •
The degree in aof every vertex y E Y2 \ YI is 1. Therefore every y E Y2 \ YI

belongs to a component which is a (y, x)-path Py , where x E X 2 \X I or
x E Y1 \Y2 .
In the first case, the last edge of Py belongs to M2 . Therefore Ml EB Py is a
matching from Xl U {x} into YI U {V} (recall that AEBB = (AUB)\(AnB».
In the second case, the last edge of Py belongs to MI. Therefore Ml EB Py is
a matching from Xl into (YI \{x}) U {V}.
In both cases the matching MI EBPy saturates YI nY2 . Therefore the matching
M{ = MI EB Py saturates all vertices of the set X I,· the vertex y E Y2 \ YI , and
the vertices of YI n Y2.
Performing this operation several times, we obtain the required matching.
3.5.2. Hint: Let Xl denote the subset of X of vertices of maximal degree. Consider
the graph induced by Xl U N(Xt). Prove that it satisfies the assumptions
of Theorem 3.5.1, Le., there exists a matching MI saturating the vertices of
Xl. Next, let YI denote the subset of Y of verti ces of maximal degree. Prove
that there exists a matching M2 saturating all vertices of YI . Now you may
use the previous exercise.
3.5.4. Solution: Let us construct a new bipartite graph G' = (X, Y', E') by
adding IXI - t new vertices to the part Y and connecting each of them
with every vertex of X. Clearly, the existence of a matching of cardinality
t in C is equivalent to the existence of a matching M from X into Y' in
C'. A necessary condition for the existence of such M is IAI ~ INGI(A)I.
The latter inequality is equivalent to the assumption of the theorem, since
INGI(A)I = ING(A)I + IXI- t.
3.5.6. Hint: The sufficiency may be proved by contradiction considering a cycle of

odd length in the graph.
3.5.7. Hint: A binary matrix may be regarded as a reduced adjacency matrix of

a bipartite graph; the ones of the matrix correspond to the edges of the
bipartite graph. The independence of two ones of the matrix means that the
corresponding two edges are not adjacent.
3.5.8. Hint: Use Corollary 3.5.2.
3.5.9. Hint: Consider a bipartite graph C = (X, Y, E), where the vertices of X
correspond to the elements of 5 and the vertices of Y correspond to the
elements of P, and XiYj E E if and only if Pj E 5". Now you may use the
Theorem 3.5.1.

3.5: Matchings in Bipartite Graphs 239
3.5.11. Hint: Use the Hall theorem replacing the collection 5 by the collection
S' = (511 , S12, Sal' S21, S22," ,S2k 2 , " " 5m1 , 5m2 ,., .5m k m ,
where
5ij = 5i , i = 1,2, . .. ,m, j = 1,2, ... ,kt , t = 1,2, .. . ,m.

Connectivity
4.1 Biconnected Graphs and Biconnected Com-

ponents
4.1.3. d: Hint: Exercises 4.1.3b,c and 2.1.15 imply that every skeleton ofthe graph
G is a simple path.
4.1.5. Hint: Use Exercise 4.1.1a and the Handshake Lemma.
4.1.6. A necessary and sufficient condition is that the graph G must have a bridge.
4.1.7. Hint: Use Exercise 4.1.2.
4.1.8. Hint: Use item 2) of Theorem 4.1.1.
4.1.9. See Fig. A4.1.1.
Figure A4.1.1: To Exercise 4.1. 9
4.1.10. Hint: By item 6) of Theorem 4.1.1, for any two vertices a, b the graph G
has a simple (a, b)-path Hi containing v. Prove that Hi contains a subpath
that is the path required for the exercise.
241
4.1.11. a: See Fig. A4.1.2.
• • • • • • •
b: Solution: Clearly, the graph be( G) is connected. Therefore it is sufficient

to prove that it is acyclic. Suppose that it contains a simple cycle
Each of blocks Bi;, i = 2, ... , k, contains a simple (ci;, cj;+J-path, and the
union of these paths is a simple (cil, ch)-path P of G. By Exercise 4.1.1b,
the graph Bh UP is biconnected, hence Bh is not a block of G, which is a
contradiction.
4.1.12. Solution: Let Band C be the sets of blocks and cutpoints of the graph
G respectively and let E be the set of edges of bc(G). Since bc(G) is a tree
(Exercise 4.1.11b), IBI = IEI-ICI + 1.
On the other hand,
1+ L (b(v) - 1) = 1 + L(b(v) - 1) = 1 + L b(v) -ici = 1 + IEI-ICI·

vEVG vEe
The two equalities imply IBI = 1 + LVEVG(b(v) - 1).
4.1.13. Solution: Proof is by contradiction. Suppose that the graph G v = G - v
has cutpoints for every v E VG. Let Bv denote a block of Gv with maximal
number of vertices; denote tv = 1Bv I. Choose a vertex v maximizing tv. Let
e be an edge of G with both ends in Bv. By Theorem 4.1.1, G has a simple
cycle passing through e and v. This implies that G has two simple paths
such that
Vh l n Vh 2 = v, Vh l n VBv = xp, Vh2 n VBv = Yq.
Set H = V hl U V h2 U V Bv. If we suppose that G has a vertex z ~ H then
this would imply that G z = G - z has a block B' such that H ~ VB',
i.e., tz > tv, contrary to the choice of the vertex v. Therefore VG = H. This
implies that max{p, q} > 1; otherwise it would be G - v = Bv. To be specific,
assume that p > 1.
Consider the vertex Xp-l of the path hl. By the assumptions of the exercise,
deg Xp-l ~ 3. Hence G has a vertex w adjacent to Xp-l such that WXp_l ~
Eh l • There are four possibilities for the location of W :
4.1. Biconnected Graphs and Biconnected Components 243
The first and second cases contradict to the choice of v; the third and fourth
cases contradict to the fact that Bv is a block of Gv .
4.1.14. Hint: Use the induction over the number of blocks of the graph.
4.1.15. Hint: This follows from the equality G - x = G - x-e.
4.1.16. Solution: Assume the contrary. Let B be a block of G - e containing a, b,

where ab = e, and let x be a vertex of (G - e)\B. Since G is biconnected, by
Theorem 4.1.1 it contains a simple cycle containing both e and x. Now the
Exercise 4.1.1b readily leads to the contradiction with the assumption that
B is a block of G - e.
4.1.18. Solution: Proof is by contradiction. Let blocks B 1 , B2 and vertices a, b
be such that a, b E V Bl n V B 2 . It is sufficient to consider the case when
B 1 , B2 are distinct from K 2. For an arbitrary vertex x E V B2 \ V B 1 , by
Theorem 4.1.1 B2 has a simple (a, b)-path containing x. It has a subpath
passing through x and having exactly two vertices in common with B 1 • Now
Exercise 4.1.1b contradicts to the fact that Bl is a block.
4.1.20. Hint: Use the Exercises 4.1.19, 4.1.17.
4.1.21. Solution: The sufficiency readily follows from Theorem 4.1.1, item 2. Let
us prove the necessity. Assume the contrary. Then by Exercise 4.1.17 the
cycle C entirely lies in a block of G - e. On the other hand, Exercise 4.1.16
implies that a, b belong to different blocks of G - e.
4.1.24. Solution: Proof is by contradiction. For IGI = 3 the statement is evident.
Consider a graph G of minimal order which is critically biconnected and has
no vertices of degree 2. Since 8(G) > 2, the graph G - e, e E EG, has a
block B which is not K 2 . By Exercise 4.1.23, B is a critically biconnected
graph, contrary to the minimality assumption for G.
4.1.25. Solution: Proof is by induction over IGI. For IGI = 4 the statement is
evident. Assume that IGI > 4. For e E EG let G 1 , G 2 , ... , Gm denote the
blocks of G - e distinct from K 2 and let R 1 , R 2 , ... , Rk denote the two-vertex
blocks of G - e. By Exercises 4.1.22,4.1.23, each of blocks Gi is a critically
biconnected graph and IGil ~ 4. Clearly, G - e has k + m - 1 cutpoints and
each cutpoint belongs to exactly two blocks of G - e. Therefore
m k
L IG;! + L IR;I = IGI + k + m - 1.
i=1 i=1
By the induction hypothesis, IEGil :S 21G;I - 4, i = 1,2, ... , m. In addition,

obviously, IE Rj 1 = 1 = 21Rj 1 - 3, j = 1,2, ... , k. These relations imply
m k
IEGI = L IEG;I + L IERj 1+ 1 ::;

i=1 j=1
m k
2 L IGil- 4m + 2 L IRjl- 3k + 1 = 21GI- 2m - k - 1 :S 21GI- 4.

j=l
The last inequality is always valid, if we observe that if m = 0 then k ~ 3,

since G is not the triangle, and if m = 1 then k ~ 1, since G - e has at least
two blocks.
4.1.26. No; see Fig. A4.1.3.

4.1.29. Hint: Use the Exercise 4.1.l1b.
4.1.30. Solution: Let G be a vertex of G. It is easily seen that degG G < IGI- 1.
Therefore there exists b E VG not adjacent to G. The hypothesis of the
exercise implies that b is a cutpoint of G - G. Let s, t be two vertices of
G - G that are not its cutpoints and belong to its different pendant blocks
(their existence follows from Exercise 4.1.29). Clearly, s, t are not adjacent.
Suppose that degG G > 2. Then G has a vertex v # s, t adjacent to G. This
means that G - s - t is connected, contrary to the assumptions of the exercise.
This contradiction implies that degG G = 2, i.e., G is a simple cycle.
4.1.31. a: Hint: Use the Exercise 4.1.20.
b: Solution: : The following proof is by contradiction. Suppose that H is a
maximal complete subgraph of B( G) and V B( G) # V H. The connectedness
of B( G) and the choice of the set V H together imply the existence of vertices
Vi,Vj,Vk such that Vi E VB(G)\VH, Vj,Vk E VH, ViVj E EB(G), ViVk r:J.
EB(G). This means that the block Bj contains two cutpoints of G, i.e., it is
not a pendant block. But then according to item a of the exercise B( G) - Vj
is disconnected, contrary to the assumption that B( G) is biconnected.
4.2 k-connectivity
4.2.1. ~(G) = 3, A(G) = 4.
4.2.2. A subgraph induced by the set of vertices 1,2, 3, 10.
4.2.3. One edge.

4.2.4. Hint: Use the Handshake Lemma and Theorem 4.2.2.
4.2.5. Solution: Proof is by contradiction. Let the subset 5 C VG be such that

:s
151 k - 1 and the graph G - 5 is disconnected. Let VI denote the vertex
set of a component of G - 5 of smallest size. If x E VI, then clearly deg x :s
:s
IV I I+ 151- 1. This together with the inequality IV I I (IGI- 151)/2 implies
degG x :s (IGI- 151- 2)/2 :s (IGI + k - 3)/2,
which contradicts to the assumption of the exercise.
4.2.6. A possible solution is the union of two copies of K n - 2 with n - 4 common

vertices.
4.2.7. Hint: Use the exercise 4.2.5.
4.2.8. A possible solution is the union of two copies of K 2 k-2 with k - 4 common
vertices.
4.2.10. Hint: Use the exercise 4.2.9.
4.2.11. Solution: Let us take 5 c EG with 151 = A(G) such that the graph G-5 is
disconnected. Then G - 5 has exactly two components G I , G 2 (see Exercise
4.2.10). Without loss of generality assume that IG 2 1 ~ IGd = t. By the
condition, 1 :s t :s 6. Let us estimate 151 :
151 ~ L: (deg v - t + 1) ~ t(6 + 1) - t 2 = (6 - t)(t - 1) + 6 ~ 6.

vEVG ,
4.2.12. Hint: Consider, e.g., the graph G = G I U G 2 U G 3 , where
Inl - n21 :s 1; EG = {ab}, a E VGI,

3 bE VG 2 ·
4.2.13. Solution: Let 5 C VG be such that 151 = ~(G) and G - 5 is disconnected.

Let G I denote a component of G - 5 of the smallest size. Clearly, IG/I :s
(IGI- 151)/2. If v E VGI then
6(G) :s degG v :s IG/I- 1 + 151 :s (IGI + 151)/2 - 1.
This implies that

26(G) - IGI + 2::; 151 = ~(G).
4.2.14. Solution: Without loss of generality assume that G is connected. In the

= = =
cases x;( G) 1 and x;( G) 3 the equality x;( G) A( G) follows from Exercise
4.1.4 and Theorem 4.2.2. Consider the case x;(G) = 2. By Theorem 4.2.2 it
is sufficient to prove that the graph has two edges e1, e2 such that G - e1 - e2
is disconnected. By the assumption of the exercise, G has two vertices a, b
such that G-a-b is disconnected, i.e., both G-a and G-b have a cutpoint.
Hence by Exercise 4.1.4 G - a has a bridge el and G - b has a bridge e2. It
is easily seen that G - el - e2 is disconnected.
4.2.15. Hint: This follows from the Theorem 4.2.2 and the Handshake Lemma.
4.2.16. Hint: Among the sets of edges incident to a cutpoint and belonging to the
same block, consider the one of the smallest size.
4.2.17. Solution: Suppose that S C EG is such that lSI = A(G) and G - Sis
disconnected. By Exercise 4.2.10, G - S has exactly two components Gt, G 2 •
Let VI ~ VGI and V2 ~ VG 2 denote the subsets of vertices incident to
edges from S. Clearly, !Vd ::; A(G) and !V21 ::; A(G). Moreover, the condition
d(G) = 2 implies either VI = VG I or V2 = VG 2 ; say, VI = VG I holds.
Consider v E VI and let TI ~ VI, T2 ~ V2 be the sets of vertices adjacent
to v. If v E VI, then let m( v) and m' (v) denote the sets of edges incident
to v that belong to S and not belong to S respectively. Clearly, degG v =
Im(v)I+lm'(v)1 and Im(v)l2: 1 for v E VI, and S = UVEV1 m(v). Let us select
a vertex Va E VI maximizing Im( v) I. Then the above relations produce:
i.e., A(G) 2: 8(G). This together with the Theorem 4.2.2 delivers the required
equality.
4.2.19. Hint: Use the Handshake Lemma and the Exercise 4.2.15.
4.2.20. One graph, the wheel W4 .
4.2.21. k = 1,2.
4.2.22. No; see Fig. A4.2.1

G~
Cl
Vl hl Vl
X
dl \ dl \
)
h2
)/
C2 C2
a b
4.2.23. Solution: Let Vl, V2, V3 be three arbitrary vertices of a triconnected graph
G. By Theorem 4.1.1, it has a simple cycle containing the vertices Vl, V2.
Let S denote the set of all vertices v of G such that G has a simple cycle
containing Vl, V2, v. Suppose that S # V. Since G is connected, it has an edge
e = xy such that XES, Y E V\S. By the choice of S, G has a simple cycle
C passing through Vl, V2, x. Let Cll C2 denote the two simple (Vl, v2)-paths
such that Cl UC2 = C and x E VC1. Since G-x is biconnected, by Exercise
4.1.10 it has a simple (y, dl)-path hl and a simple (y, d 2 )-path h2 such that
Consider two cases: (1) one of d l , d2 , say, d l is in VC1; (2) d l , d 2 E VC 2 .

It is easily seen that in both cases G contains a simple cycle passing through
Vl, V2, y. For the first case its structure is shown at Fig. A4.2.2a, for the
second case its structure is shown at Fig. A4.2.2h. This is the contradiction
with the choice of S, y. Therefore, C = V, i.e., V3 E S.
4.2.24. No; see the graph /{3,4.
4.2.25. Let G(A, B; E) be a bipartite graph with o(G) = O. It easy to show that
the assumptions of the exercise imply that G is connected. Suppose that
S C EG is such that lSI = ~(G) and G - S is disconnected. G - S has
two connected components, G l and G 2 (see Exercise 4.2.10). Without loss
of generality assume that IGll ::; IG 21. Set Xl = VG l n A, X 2 = VG l n B.
Clearly, IXl! + IX21 = IGll ::; IGI/2. It is easily seen that if Xl = 0 or X2 = 0
then IGll =1, and then, clearly, o(G) =~(G). Therefore we assume that
Xl # 0 and X 2 # 0, and hence IGd 2: 2. Let us hound the value lSI from
the below:
6(IX 1 1+ IX2 J) - 21 X III X zI ~ 6(IXti + IX2J) - (IXti + IX2J)2/2 =

6IG 1 1-IG l I2 /2 = 6 + (IGd - 1)(6 -IG l I2 /(2IG 1 1- 2)) ~
6 + (IGd - 1)(6 -IGI/4 - 1) ~ 6.

This and Theorem 4.2.1 imply ..\(G) = 6(G).
4.2.26. Hint: See the solution for the Exercise 4.2.11.
4.2.27. No. A counterexample is shown at Fig. A4.2.3.
4.2.28. Cn.
4.2.29. Kn,n+l.
4.2.30. K3,3.
4.2.31. K 3 ,4.
4.3 Cycles and Cuts

4.3.1. K 3 ,4.
4.3.2. Solution: Let C be a simple cycle of the graph G such that ICI = g(G).
The case g( G) ::; 4 is obvious. Assume that g( G) ~ 5 and consider the set T
of vertices from VG\ VC adjacent to the vertices of C. Since 6(G) ~ 3 and
cycle C has no diagonals, every vertex of C is adjacent to a vertex from T.
Furthermore, it is easily seen that every vertex of T is adjacent to a single
vertex of C, i.e., ICI ::; ITI, and hence 21CI ::; ICI + ITI ::; IGI·
4.3.3. Solution: Proof is by contradiction. Let C be a simple cycle of G, C f; G. If

ICI = IGI then C must have a diagonal, and in this case G has a cycle of even
length. Assume that ICI < IGI and consider v E VG\ VC. By Exercise 4.1.10,
G has a simple (a, b)-path h passing through v such that V h n V C = {a, b}.
It is easily seen that h U C has a cycle of even length.
4.3. Cycles and Cuts 249
4.3.4. a: Solution: Let h = (VI, V2, .•• , vp ) be a simple path of maximal length in
G and let Vk be a vertex of h incident to VI with the largest index k. The
maximality of the path h implies that VI is adjacent only to the vertices of
h. Therefore k ~ beG) + 1. Hence the simple cycle (VI, V2, ... , Vk, vd is of
length at least b( G) + 1.
4.3.5. Hint: See the Exercise 4.2.10.
4.3.6. Solution: Let El , E2 denote the sets of edges of a cycle C with equicolored
and differently colored ends respectively. Let us assign the weight w( v) = 0
to the vertices of the first color and the weight w( v) = 1 to the vertices of
=
the second color. An edge e uv obtains the weight w( e) =
w( u) + w( v).
Clearly, LeEE 2 wee) = IE21 and t = LeEE l wee) is even. On the other hand,
L w(uv) =L (w{u) + w(v)) =2 L w(x), L wee) =t + IE21·

uvEEC uvEEC xEVC eEEC
Therefore IE21 is even.

4.3.7. Hint: Use the Exercises 4.3.5, 4.3.6.
4.3.9. Ten cuts. Hint: To find them, use the Exercise 4.3.8.
4.3.14. No.
4.3.17. Solution: Suppose that keG - (VI, V2 )) = 2 and let X be a proper subset
of the separator set (VI, V2 ). The ends of an edge e E (VI, V2 )\X belong to
different components of G - (VI, V2), therefore adding e to G - (VI, V2) we
obtain a connected graph. Hence X is not separating, and therefore (Vl' V2)
is a cut of G. The converse is proved as in the Exercise 4.3.5.
4.3.18. Solution: Let C l , C 2 denote the vertex sets of paths produced from C by
the deletions of the edges el and e2 respectively. Among the separator sets
of form
X = (Vl' V2), C l ~ Vl, C2 ~ V2
we select a set XO = (Vt, V2') such that G - XO has the minimal num-
ber of components. Suppose that k( G - XO) ~ 3. Let Gl, G 2 denote the
components of G - XO containing C l , C 2 respectively. G has an edge con-
necting a vertex of some component Ga f=. G l , G 2 with a vertex of G l or
G2, say with that of G 1 . Therefore VGa C V2Q. Consider the separator set
X = (Vt U VGa, V2'\VG a). Clearly, C l ~ Vt U VG a, C 2 C V20\VG a and
keG - X) < keG - XO), which is a contradiction. Therefore keG - XO) = 2,
and by Exercise 4.3.17, XO is a cut.
Matroids
5.1 Independence Systems

5.1.1. a: No; b: No; c: Yes, 8(S) = {{I, 2}, {I, 3}}, C(S) = {{4}, {2, 3}}.
5.1.2. Hint: a: I(S) is the set of all subsets of elements B E 8.
b: A subset X ~ E is independent if and only if no element of 8 is a subset
of X.
5.1.5. If Sl, S2 are independence systems with the same support then C(Sl n S2)
is the set of all minimal under inclusion elements of the set C(SI) U C(S2).
5.1.6. b: The maximal independent sets of vertices of G.
c: The edges of G.
5.1.8. Nine; four graphical ones.
5.1.9. b: The characteristic vectors of lower ones of the function f.
c: Hint: Let (E,1) be an independence system of order n with E assumed
to be {I, 2, ... , n}. Let us define the Boolean function / in n variables by
setting
/(Xl, ... , xn) = 0 if and only if the subset U ~ E with the characteristic
vector Xu = (Xl, ... , xn) is independent.
d: / == 0 and the monotonic Boolean functions for which the number of
nonzero components of every lower one is equal to 2.
Hint: See the Exercise 5.1.7.
=
5.1.10. a: C(S) {{I}, {2}, ... , in}}, 8(S) {0}.=
b: C(S) = {{I}, {4}, {2, 3}}, 8(S) = {{2}, {3}}.
5.1.11. b: Solution: Let S = (E, I) be an independence system of order n, with E
assumed to be {I, 2, ... , n}. If S is a free system then it is specified by the
single inequality
Xl + X2 + ... + Xn ::; n.
251
If S is not free, take G E C(S) and let Xc = (ac" ... ,a be the charac-
Cn )
teristic vector of the cycle C, with kc denoting the number of its nonzero
components. Then the independence system S is specified by the system of
inequalities
c: The systems of inequalities of form (5.1.1) with aij E {D, I}, bi = 1.
5.1.12. b: The pairs of adjacent edges of G.

c: The maximal matchings of G.
5.2 Matroids
5.2.3. The support for all matroids is {1,2,3}. See the table below. There are
three graphical independence systems and two graphic matroids.
p Bases Cycles
D 0 {1},{2},{3}
1 {I} {2},{3}
1 {1},{2} {3},{I,2}
1 {1},{2},{3} {1,2},{1,3},{2,3}
2 {I,2} {3}
2 {1,2},{1,3} {2,3}
2 {1,2},{1,3},{2,3} {I,2,3}
3 {l,2,3} No
5.2.5. b:Hint: Use the Proposition 5.2.5.
5.2.6. a: Solution:
The set G' as an antichain, since it is a subset of the antichain G, i.e., Axiom
C.I holds for G'. Therefore the pair M' = (E, G') is an independence system
with the cycle set G' (Exercise 5.1.2b). By Theorem 5.2.4, M' is a matroid if
G' satisfies Axiom C.2. Take X, Y E G', X =P Y, e EX ny. Since X, Y E G,
the set Z = (X u Y)\ {e} contains a cycle U of the matroid M. In the
considered situation IXI = WI = IZI = 2, lUI :s; 2. Therefore U E G', i.e., G'
satisfies Axiom C.2.
b: A counterexample is the matroid with the set of cycles
{{1,2,3},{1,4,5},{2,3,4,5}}.
5.2.7. b: Solution: Recalling that an application of elementary operations to rows

of a matrix preserves linear relations among the columns, we transform the
matrix A as follows:
5.2. Matroids 253
1 2 3
A= [ 0 1 1
2
1
3
1 11 [1 2 n~ 3
2 -> 0 1 1
1 1 0 -1 -3 0 0 0 -2
[
1 2 3
o1 1
o
1
1
1 2-3] [1 0 oo -2]
-> 0 1 3 .
001 o 1 -1 0 0 1 -1
It is seen now that rank A = 3 and every three columns are linearly inde-
pendent. Thus, the three-element sets of columns of A are the bases of the
matroid M and the single-element sets of columns of A are the cobases of M.
The set of all columns is the single cycle. The two-element sets of columns
are the cocycles.
5.2.10. b: Hint: A possible covering is the set of cycles of the considered indepen-
dence system plus the set of all elements contained in no cycle.
d: Hint: For matroids of order three, use the table from the answer to
Exercise 5.2.3. When passing to matroids of higher orders you must take
into account that if M is the matroid of the partition E = El U ... U Ek
corresponding to the choice m = (ml, ... , mk), then the cycles of M are the
(mi + I)-element subsets of Ei, i = 1,2, ... , k. Let now M be the matroid
with the support {I, 2, 3, 4} and the cycle set C = {{I, 2, 3}, {I, 2,4}, {3,4}}.
If it were a partition matroid then all three cycles should belong to the same
part Ei of the partition and hence the number of elements in them should
be the same. Therefore M is not a partition matroid.
= =
5.2.11. a: Hint: If E is a finite non empty set, 5 {Ei : i 1, ... , k} is a covering
of E, M = (E, 8) is the transversal matroid whose bases are the transversals
of 5, Mi, i = 1, ... ,k are uniform matroids of rank 1 with the supports Ei
respectively, then M = Ml U ... U Mk.
e: The matroid with the support {I, 2, 3, 4} and with the set of bases
{{1,2},{1,3},{1,4},{2,3},{2,4}}
is transversal, but it is not a partition matroid.
5.2.12. b: Hint: Prove that if M is a transversal matroid of order n then to every
its independent set there corresponds a matching in the complete bipartite
graph Kn,n' The number of possible sets of matchings is at most 2n2.
Solution: Let M = (E,1) be a transversal matroid of order n. Without
loss of generality let us assume that M is the matroid of transversals of a
=
collection F (Ei : i = 1,2, ... , k) of nonempty subsets of E, where F is a
covering of E. Thus, the independent sets of M are the partial transversals
of F plus the empty set. Assume that k > n first.
=
Let us define a bipartite graph G F with parts A {I, 2, ... , k} and B = E
as follows: N(i) = Ei, i = 1,2, ... , k. Clearly, the mapping F 1---+ GF is a
bijection between the set of coverings of E of length k and the set of bipartite
graphs with parts A and B without isolated vertices. It is also clear that
a subset X ~ E is independent with respect to M if and only if G F has a
matching that covers X.
Let now B = {b l , ... , bp } be a base of M and C be a matching that covers it.
Without loss of generality assume that C = {lb l , 2b 2 ••• , pb p }. The graph
GF has no edges of type xy, x E {p + 1, ... , k}, y rt N(l) U ... U N(p),
otherwise C would not be a maximal matching and hence B would not be a
base. Therefore the subset F' = (Ei : i = 1, ... , p) is also a covering of E.
Let M' = (E, I') be the matroid of transversals of the family F'. Let us
prove that M = M', i.e., I =
1'. By the construction, I' C; I. Suppose
that X = {eO': a = 1, ... ,m} E I, D = {iae a : a = 1, ... ,m} is a
matching of G F that covers X such that ia :::; p if and only if a :::; q < m.
Set X = = =
Xl U X2, where Xl {eO' : a:::; q}, X2 X\XI . The matching D
is partitioned accordingly: D = DI U D 2 , iaea E DI if and only if eO' E Xl.
Clearly, every edge xy E D2 is incident to a vertex bi E B. Replacing every
edge xy of D2 by the corresponding edge ib i E C we obtain a matching of
GFt. Hence we have proved that x E I', I C; 1', I =
I', M =
M'.
Thus, we may assume that k :::; n in the original collection F. Therefore the
matroid M is specified by a bipartite graph GF with parts of sizes at most
n, i.e., by a subset of edges of /{n,n. The number of such subsets is 2n 2 •
5.2.13. Hint: Consider the dual matroid M* for the matroid M of transversals of
the collection of sets {{ I}, {2, 3}}.
5.2.14. Hint: Use the axioms of cycles for the matroid.
5.2.15. b: Use Exercises 5.2.14 and 5.1.6.
5.2.16. Hint: Consider the line graph L(G) and use the Exercise 5.2.14.
5.2.17. a: The only bases of the matroid M (G) are the sets of edges of skeletons
of the graph G. The sets of edges of the simple cycles of G are the cycles of
the matroid. The cuts of G are the cocycles; p(M(G)) = v*(G) (the co cyclic
rank of G); p*(M(G)) = v(G) (the cyclic rank of G).
b: Hint: See the Exercise 5.2.17a.
5.2.18. No.
5.2.19. The forests and the simple cycles.
5.2.20. Hint: Use the proposition 5.2.6.
5.2.21. Hint: The cuts of G are the cocycles of the matroid M(G).
5.2.22. Hint: Use the proposition 5.2.9.
5.2.23. Hint: See the Exercise 4.3.19.

5.3. Binary Matroids 255
5.2.24. a: The rows with numbers 1, 2, 3, 4, 6. The sum of its elements is 4l.
b: The rows with numbers 1, 2, 3, 4, 5. The sum of its elements is 35.
5.2.25. a: Solution: Let E = {e1, ... , en} be the set of all tasks with the same
execution time t and let ti be the deadline of the i-th task ei. Let I be the
set of sets of tasks that may be performed within their deadlines. Let us
prove that M = (E, 1) is a matroid.
The validity of Axioms 11 and 12 is evident. Let us prove the validity of
Axiom 13 . Take X, Y E I, IXI < IYI. Without loss of the generality assume
that Y = {e 1, ... , ek }, k :::; n and the tasks in the list Yare written in the
order of the possible execution. The execution time for Y is kt and for X it is
at most (k-1)t. Therefore if ek tI. X then XUek E X, since tk :::; kt. If ek EX
then consider the task ek-1. We have tk-1 :::; (k - l)t. Therefore if ek-1 tI. X
then X U ek-1 E I, since the task ek-1 may be executed within the time
interval [(k - 2)t, (k - l)t] and the task ek within the interval [(k - l)t, kt].
If further ek-1 is also in X, we consider ek-2 just in the same way, and so
on. Clearly, eventually we will find a task ei such that ei tI. X and ej E X,
j = i + 1, ... , k. Therefore Xu ei E I, since the task ei may be executed
within the time interval [( i-1)t, it], and each task ej, j = i + 1, ... , k within
the interval [(j - l)t,jt].
b: The tasks must be performed in the nondecreasing order oftheir deadlines.
5.3 Binary Matroids

5.3.2. a: Solution: This matroid is not binary. In fact, the two-dimensional linear
space over Z2 has only three nonzero vectors: a, b, a + b, where {a, b} is a
basis. Hence it has no four elements such that any two of them are linearly
independent.
b: Yes.
5.3.3. Hint: The matrix
is a representation of the considered matroid over Z2.

b: {2, 3, 4} is the single cycle.
c: Yes.
5.3.6. Solution: Let us transform the matrix I(G) by elementary transformations

of rows over Z2.
1 0 1 0 0 0 1 0 1 0 0 0 1
[~
0 0
r] ~
0
1 1 0 1 0 0 0 1 1 1 0 0 1 1 1 0
0 1 0 0 1 0 -> 0 1 0 0 1 0 -> 0 1 1 1
0 0 0 1 1 1 0 0 0 1 1 1 0 0 1 1
0 0 1 0 0 1 0 0 1 0 0 0 1 0 0
1 0 0 0
[~
0 1 0 1 0 0 0 1 0 1 0 0
1] ~
11 1 0 0 0 1 1 1 0 0 0 1 1 1 0
0 1 1 1 0 -> 0 0 1 1 1 0 -> 0 0 1 0 0
0 0 1 1 1 0 0 0 1 1 1 0 0 0 1 1
0 0 1 1 1 0 0 0 0 0 0 0 0 0 0 0
0 1 0
[~
0 1
i] ~
0 0 0 0
1 0 0 1 0 1 0 0 1
t]
0 1 0 0 0 0 1 0 0
0 0 1 1 0 0 0 1 1
0 0 0 0 0 0 0 0 0
=
Let ei be the i-th column of J(G), i 1, ... ,6. The final form of the matrix
implies that {el,e2,e3,e4} is a column basis,
e5 = e2 + e4, e6 = el + e3 + e4.
If now ei is the i-th edge of G then {el, e2, e3, e4} is a skeleton of G and
Ceo = {e2,e4,e5}, C e6 = {el,e3,e4,e6}

is the cycle basis with respect to this skeleton. To obtain the cut basis, we
apply the Theorem 5.3.2:
C;\ = {el,e6}, C;, = {e2,e5}, C;3 = {e3,e6}, C;. = {e4,e5,ed.

Planarity
6.1 Embeddings of Graphs. Euler Formula

6.1.1. Hint: Using the Theorem 6.1.3 we embed a plane graph G onto the sphere
in such a way that the north pole falls into the chosen internal face. Next,
consider the stereographic projection Gt of G onto the plane tangent to the
sphere at the south pole. Clearly, the chosen face of G will be mapped into
the external face of G t •
6.1.2. See Fig. A6.1.1.
a b c
Figure A6.1.l: To Exercise 6.1.2
6.1.4. No.
Hint: Prove by contradiction that /{3,3 is non planar using the Euler formula
and noticing that if the bipartite graph /{3,3 were planar, every its face would
have been bounded by at least 4 edges.
6.1.5. See Fig. A6.1.2.
257
K5 -e
6.1.7. Hint: Apply the Euler formula to each component counting the external
face only once.
6.1.8. To one or two faces.

Hint: If the edge is the bridge then the point belongs to one face, otherwise
it belongs to two faces.
6.1.9. It may belong to any number 1, ... , d offaces.
6.L12. Hint: Use the Euler formula and the inequality hI :::; 2m.
6.1.13. Hint: Use the Euler formula and the inequality 41 :::; 2m.
6.1.14. Hint: Notice that a bipartite graph is triangle-free (see Konig's theorem
on bipartite graphs). Next, use the previous exercise.
6.1.15. Hint: Use the Exercise 6.1.13 noticing that the n-cube Qn is triangle-free
(see Exercise 1.4.28 and Konig's theorem on bipartite graphs).
6.1.16. Solution: A possible proof is by contradiction. By Corollary 6.1.5, m :::;

3n - 6 for any planar (n, m)-graph G. Supposing that the degree of each
vertex of G is at least 6 we would have 6n :::; 2m, and hence the contradictive
inequality 3n :::; 3n - 6.
6.1.17. Solution: Let G be a 6-connected planar graph. Then 6(G) ~ K(G) ~ 6.

But G has vertices with degrees less than 5 (see Exercise 6.1.16).
6.1.18. Solution: Let G be a 5-connected planar (n, m)-graph. Then 6(G) 2:

K(G) ~ 5 and this implies 5:::; 2m/n :::; 2(3n - 6)/n = 6 - 12/n, i.e., n ~ 12.
An example of a 5-connected planar graph with minimal number of vertices
n = 12 is the icosahedron graph, Fig. 1.1.2.
6.1.19. For r = 0, 1, ... ,5.

Hint: Such graphs are f{l, f{2, f{3, f{4, the graph from Fig. A6.1.3 and
the icosahedron graph. For r 2: 6 there are no r-regular planar graphs, see
Exercise 6.1.16.
6.1.20. Hint: Use the Euler formula and the Handshake Lemma.
6.1.21. Hint: The Handshake Lemma, the Euler formula, and the equality
Li>3 iii = 2m, where m is the number of edges of the graph, together imply
Li~3(6 - i)li = 12. This equality is impossible if fa = 14 = 15 = O.
6.1.22. Hint: Similarly to the solution of Exercise 6.1.21, noticing that the degrees
of the vertices are at least 3 (3n 2: 2m), one may prove
12 :s: 3fa + 2/4 + 15 - h - 2/8 - 3/9 - ....

If fa = 14 = 0 then
12 :s: 15 - h - 2/8 - 319 - ... :s: 15,
which contradicts to the inequality I = Li?3 Ii < 12. Therefore fa + 14 2: 1.
6.1.23. Solution: Sufficiency: Since 41 = 2m, the Euler formula implies m =
2n - 4 (see Exercise 6.1.13).
Necessity: Since m = 2n - 4, the Euler formula implies 21 = m. Because of
the equalities
we deduce Ii = 0, i 2: 5 and 14 = m/2 = f.

6.1.24. Hint: Proof is by induction basing on the Exercises 6.1.14, 6.1.23, and on
the fact that bipartite graphs are triangle-free. During the inductive step,
a new vertex is placed inside any face and is connected with two opposite
vertices of the face.
6.1.25. Hint: Use the Exercises 6.1.24,6.1.23.
D
6.1.26. Hint: Proof is by induction basing on the Exercises 6.1.15,6.1.23, and on

the fact that Qn is triangle-free. See Fig. A6.1.4, where plane subgraphs of
Q2, Q3, Q4 with maximal number of edges are shown.
6.1.27. Hint: Use the Exercises 6.1.26,6.1.23.
6.1.28. a: Hint: It follows from the equality 5/5 + 6/6 = 2m (see Exercise 6.1.21
for notations).
b: Hint: Use the Exercise 6.1.21.
6.1.29. For n = 3 and for even n ~ 4.
6.1.31. Hint: Consider all faces as possible outer faces.
6.1.33. G 1 .
6.1.34. For no n ~ 3 the wheel Wn is outerplanar.

Hint: Use the Exercise 6.1.30.
6.1.35. Solution: a: It is sufficient to prove the inequality for a maximal outerpla-

nar
(n, m)-graph G. Let us construct a planar graph G' by adding a new vertex
adjacent to all vertices of G. Clearly, n' = n(G') = n + 1, m' =m(G') =
m + n. By Corollary 6.1.5, 3n' - 6 ~ m'. Therefore m ~ 2n - 3.
b: Adding the equality 3(1 - 1) = 3 + 3m - 3n (that follows from the Euler
formula) to the inequality m ~ 2n - 3, we obtain the required relation.
6.1.36. For n = 3.
Hint: The graph Cl = K4 is not outerplanar, see the Exercise 6.1.31. For any
n ~ 5 the graph G~ has m = 2n edges, which contradicts to the inequality
of the Exercise 6.1.35a.
6.1.37. No.
Hint: A counterexample is [{4.
6.1.38. Solution: The graph G2 has a subgraph K 4 , therefore (see Exercises 6.1.31,
6.1.32) G 2 is not outerplanar.
6.1.39. Solution: The sufficiency is evident. The necessity my be proved by con-

tradiction. Suppose that a connected graph G is neither the path graph nor
K 3 . Then two cases are possible.
Case 1: G is the simple cycle en, n ~ 4. Then G 2 is not outerplanar, see
Exercise 6.1.36.
Case 2: G has a subgraph [{1,3. Then G2 is not outerplanar by Exercise
6.1.38.
6.1.40. Hint: Since no internal face is a triangle, 4(1 - 1) + n :s 2m. Now apply
the Euler formula.
6.1.41. Solution: First of all notice that every internal face of a maximal outer-
planar graph is triangular.
a: Proof is by contradiction. Suppose that deg v ~ 3 for each vertex v of the
graph. Then the number of internal faces (triangles) is n. Therefore noticing
that the external face is bounded by an n-cycle (In = 1), we have
L iii = 3n + n = 4n = 2m,
i~3
where Ii is the number of faces bounded by i edges. This implies m = 2n,

1= n + 1, which contradicts to the Euler formula.
b: Proof is by induction over n ~ 3. For n = 3 the statement is evident.
Assume that it is true for n = k 2: 3 and consider a graph G with n = k + 1
vertices and q internal faces. Then G - v, where v is a vertex of degree 2
(see item a), has q -1 internal faces, i.e., q -1 = k - 2, or q = k -1 = n - 2.
6.1.42. See Fig. A6.1.5.
6.1.43. Hint: a: Use the Euler formula and the Exercise 6.1.41.
b: A possible proof is by induction over the number of vertices taking into
account that the vertices of degree 2 are not adjacent to each other if n ~ 4.
c: Use items a, b of the exercise and the Handshake Lemma.
d: Noticing that every maximal outerplanar graph has at least one vertex
of degree 2 (see Exercise 6.1.41a), a simple proof is by induction over the
number of vertices.
6.1.44. Hint: It is sufficient to place the vertices in the space in such a way that
no four lie in a common plane.
6.2 Plane Triangulation

6.2.1. Hint: Use Theorem 6.2.1.
6.2.2. 2n - 4.
Hint: Use the Euler formula and Corollary 6.2.3.
6.2.3. Solution: Let G be a plane graph satisfying the conditions of the theorem.
Adding the edges, convert G into a maximal plane graph G' with all vertices
of degrees at least 3 (see Proposition 6.2.4). Then by the Corollary 6.2.4 and
the Handshake Lemma we have
2(3n - 6) = 2m = 3ng + 4n4 + 4ns + ... ~ 3(ng + n4 + ns) + 6 L ni,

i~6
where m is the number of edges of G', n is the number of its vertices and
ni is the number of vertices of degree i. This inequality implies the required
ng + n4 + ns ~ 4, if we notice that Li~g ni = n.
6.2.4. Hint: For such (n, m)-graph the following equalities are valid:
5n = 2m, m = 3n - 6, n + f -= 2,
m
where f is the number of faces of the graph. Hence m = 30, n = 12, f = 20.
The uniqueness follows from the construction of a 5-regular triangulation.
6.2.5. 3n - 6.
Hint: See Corollaries 6.1.5 and 6.2.3.
6.2.6. See Fig. A6.2.1.
6.2.7. No.
Solution: Suppose that such triangulation exists. Deleting the edge connect-
ing the vertices of odd degree, we obtain the graph G' whose all faces are
triangular with the exception of one, bounded by four edges. All vertices of
G' are of even degree. Therefore the faces of G' may be colored red and blue
in such a way that any neighboring faces are colored differently (see Exercise
9.4.6). Assume that the unique 4-face is colored red. Let r, b denote the
numbers of faces G' colored red and blue respectively. Since every edge of
G' separates red and blue faces, 3b = 4 + 3(r - 1). But this equation has no
integer solutions; hence such triangulation cannot exist.
6.2.8. Solution: Let the colors be numbered 1,2,3. Each face may be characterized
by the triple (i,j, k) of colors of its vertices. Let us transform the triangu-
lation G into another triangulation G' as follows. We place a new vertex
inside each mono colored face, i.e., a face of type (i, i, i), i = 1,2,3, color
this vertex by a color j =F i, and connect it with all vertices of this face, see
Fig. A6.2.2. This operation preserves the number of tricolored faces (and
the total number of faces remains even, see Exercise 6.2.2). Clearly, G' has
no mono colored faces, i.e., it has only bi- and tricolored faces.
Every bicolored face of G' has a mono colored edge. At the same time, every
mono colored edge belongs to two bicolored faces, see Fig. A6.2.3, where
k =F i, j =F i. This implies that the number of bicolored faces of G' is even,
the same for the number of tricolored faces of G' and hence of G.
6.3 Planarity Criteria

6.3.1. The Petersen graph contains a subgraph homeomorphic to K 3 ,3 with the
....------..;.i
parts {1, 2, 3} and {4, 5, 6}, see Fig. A6.3.1.
6.3.2. Hint: Find a subgraph isomorphic to K 3 ,3 in the n-permutation graph.
6.3.3. n = 2,3, ... , 6.

Hint: G7 is nonplanar since it contains a subgraph homeomorphic to K 5 •
Further, if s < n then G s is the subgraph of G n .
6.3.4. 14. Hint: Construct all ten graphs of sixth order that contain the subgraph
K 3 ,3' Further, let G be a unique graph of sixth order homeomorphic to K5.
Construct all graphs containing G as a spanning subgraph and verify that
exactly two of them (having the vertex of degree two) were not constructed
previously. Finally, it is easy to find two six-vertex graphs containing the
subgraph K5 but neither K 3 ,3 nor G.
6.3.5. All graphs are nonplanar.

Hint: All of them contain subgraphs homeomorphic to K 3 ,3'
6.3.6. a: For every n ~ 2. b: For n = 2 only.

Hint: a: A plane embedding may be constructed inductively.
b: For n = 3 it contains a subgraph K 3 ,3' For n > 3 it contains a subgraph
homeomorphic to K 3 ,3'
6.3.7. The graph Hn is planar for every even n ~ 4. For any odd n ~ 3 Hn is
nonplanar: H 3 = J{3,3, and H 2/<+1, k > 1, contains a subgraph contractible
to J{3,3.
6.3.8. Hint: If n :::; 5, the statement is evident. Assume that n = 6 and the graph
G is nonplanar. Then three cases are possible: 1) G = J{3,3; 2) G has a
subgraph J{5; 3) G is obtained from J{5 by a subdivision of an edge. In all
these cases G is planar.
The case n = 7 may be analyzed in a similar way.
6.3.9. n = 8.
Hint: The graph J{5 U 0 3 has the property. See also the previous exercise.
7 5
1 •
6
4 •
2.---------- 8
3 2
G
6.3.10. a: See Fig. A6.3.2. b: The graph J{5 U 0 3 . c: J{8.
6.3.11. For anyone.

Hint: The graph G must have 10 edges. A possible proof is by contradiction:
enumerate all possible cases of nonplanar graphs G of 8th order with 10 edges
(J{3,3 U P2 , J{5 U 0 3 , (J{3,3 + e) U O 2 , etc.) and verify that for no case the
graph G = G may be a planar triangulation.
6.3.12. Solution: Clearly, IEGI + IEGI = n(n-l)/2. If both G, G were planar then
it would have been IEGI :::; 3n - 6, IEGI :::; 3n - 6, i.e., n(n -1)/2:::; 6n -12,
which is impossible for n ~ 11.
6.3.14. Pn and Pn+k, C n and Cn+k'
6.3.15. Not always. At Fig. A6.3.3 G 1 , G 2 are homeomorphic but L(G 1 ), L(G2)
are not homeomorphic.
6.3.17. Not always; see Fig. A6.3.4.

[01><1
6.3.18. No. Hint: A counterexample is the Petersen graph: it is contractible to

K5 but it has no subgraph homeomorphic to K5 since the vertex degrees of
the Petersen graph are all equal to 3.
6.3.19. Hint: A possible proof is by contradiction using the Wagner criterion.
6.3.21. A counterexample is the graph from Fig. A6.3.5.
6.3.22. Hint: All vertices of L(K5) are of degree six, which is impossible for a
planar graph, see Exercise 6.1.16. Verify that L(K3,3) contains a subgraph
homeomorphic to K 5 •
6.3.24. Solution: By the Kuratowski criterion, the graph G contains a subgraph

H homeomorphic either to K5 or to K3,3. Therefore by Exercise 6.3.23 L(H)
is contractible either to L(K5) or to L(K3,3), which are both nonplanar (see
Exercise 6.3.22). Therefore by Exercise 6.3.20 L(H) is nonplanar, hence
L(G) is nonplanar.
6.3.25. Not always.

Hint: If, for example, a planar graph G has a vertex of degree 5, then L( G)
contains a subgraph K 5 •
6.3.26. r = 1,2,3.
Hint: If a graph G is four-regular then L(G) is nonplanar, since it is six-
regular, cf. Exercise 6.1.19. Similarly for r = 5.
6.3.27. Hint:
1) The planarity of G follows from the Exercise 6.3.24.
2) Proof is by contradiction, see the hint for Exercise 6.3.25.
3) If some degree-4 vertex were not a cutpoint of G then L(G) would have
been nonplanar since it would contain a subgraph contractible to [{5.
6.3.28. Hint: Otherwise G 2 would contain a subgraph [{5.
6.3.29. Hint: Let G' be obtained from a graph G by the addition of a vertex
adjacent to all vertices of G. It is easily seen (see Exercise 6.1.30) that G is
outerplanar if and only if G' is planar. Further, use the Kuratowski criterion.
6.3.30. Graphs homeomorphic to K5 and [{3,3.
Hint: Use the Kuratowski criterion.

6.3.31. K5 and all 6-vertex graphs G that have the following two properties:
1) l{vEVG: degv=3}1~3;
2) G has a subgraph [{3,3.
Hint: Use the Wagner criterion.
6.4 Duality and Planarity

6.4.3. Hint: Use induction over the number of vertices of G*, i.e., over the number
of faces of G.
6.4.4. Hint: A cutpoint of G* is a vertex v* lying inside a face of G that contains

boundaries of at least two blocks of G.
6.4.5. Hint: Proof is by contradiction using the previous exercise and the Theorem
6.4.2.
6.4.6. Hint: See the previous exercise.
6.4.7. Their geometric duals are nonisomorphic.

6.4.8. Hint: The graphs G* and G** are connected.
6.4.10. When the graph is bridgeless and any two faces have at most one common
edge.
6.4.11. Hint: Use the Euler formula and the Proposition 6.4.1.
6.4.13. Three.
Solution: Let G be a plane selfdual (6, m)-graph with I faces. Then by the
Euler formula and the Proposition 6.4.1 we have
n = n* = I = 1* = 6, m = m* = 10.
Since the degree of each vertex of G is at least 3 (see Exercise 6.4.10) and at
most 5, we have
n3 + n4 + n5 = 6, 3n3 + 4n4 + 5n5 = 20,

where ni is the number of vertices of degree i in G. Therefore only the
following cases are possible.
= =
Case 1: n3 4, n4 2. In addition, due to the selfduality of G, /3 4,14 = =
2, where Ii is the number of faces of G bounded by i edges. Furthermore,
two vertices of G of degree 4 may be either adjacent or nonadjacent. In both
cases a selfdual graph exists, see Fig. A6.4.l.
Case 2: n3 = 5, n5 = 1, with /3 = 5, 15 = 1, i.e., G is the wheel W 5.

6.4.14. Hint: The geometric dual of such graph would contain a subgraph K 5 .
6.4.15. Hint: Let T be a spanning tree of G. Prove that the edges of G* which do
not cross the edges of T constitute a spanning tree of G*.
Solution: Let T* be the subgraph of G* constituted by the edges that do not
cross the edges of the spanning tree T. Clearly, T* is acyclic, otherwise the
vertices of T inside the cycle of T* could not be connected with the vertices
outside the cycle, i.e., T would have been disconnected. Assume further that
=
n IGI, n* = = =
IG*I, m m(G), m* m(G*), I is the number offaces of G.
Since m(T) = n - 1, the Proposition 6.4.1 and the Euler formula imply
m(T*) = m* - n + 1 = m - n + 1 = 1-1 = n* - l.
In addition, clearly, IT* I = n*. Hence T* is the spanning tree of G* (see
Exercise 2.2.10). Thus a bijection T --> T* between the spanning trees of G
and G* is established.
6.4.16. Solution: In the biconnected graph G* (see Exercise 6.4.5) the boundary
of each face has even number of edges. Therefore (see Exercise 6.1.11) the
length of any simple cycle in G* is even. By Konig's theorem about bipartite
graphs, G* is bipartite.
6.4.17. Hint: Consider two embeddings of the graph in the plane with nonisomor-
phic geometric duals.
6.4.18. Yes.
Hint: See Fig. A6.4.2, where Hand Q are abstract duals to G and H ~ Q.
The corresponding edges are labelled by the same letters.
c
a c
h h
G H Q
6.4.19. Hint: Since n( G) - k( G) is the number of edges of the skeleton of the graph
G (see Exercise 2.2.11) and v( G) is the number of edges to be deleted from
G in order to obtain a skeleton of G (see Theorem 2.2.1), the exercise follows
from the definition of duality and Theorem 6.4.6.
6.4.20. G, H are dual graphs, H, Q are dual graphs, see Fig. A6.4.3.
,~
~
G H Q
6.4.21. Not always.

Hint: See the previous exercise: H = G+, Q = H+ = G++, but Q ~ G.
6.4.22. a: Hint: This follows from the Whitney criterion, the Proposition 6.4.5,
and the Theorem 6.4.6.
b: See the Exercise 5.2.24.
6.4.23. Solution: a: The equality immediately follows from the definition of the
generalized dual graph, if we substitute H =
G and notice that v* (G*) O. =
b: This equality follows from the equality a) and the equality
v*{G*) + v{G*) = m{G*) = m{G) = v*{G) + v{G).

c:
v*(H) + v(fi*) = m(H) - v(H) + m(fi*) - v*(fi*) =
m(H) + m(H) - v*(G*) = m(G) - v(G) = II*(G),
since m(fi*) = m(H).
6.4.24. Hint: This follows from the equality c) of the previous exercise.
6.4.25. Hint: A possible proof is by induction over the number of edges of a

subgraph H considering two cases: 1) after the addition of an edge from G
to H the number II(H) increases by one; 2) v(H) remains unchanged after
such operation. In both cases it is easily seen that the equality specifying the
generalized duality does not fail. At the same time, this equality is obviously
true for H = 0IGI.
6.5 Measures of Displanarity

6.5.1. Solution: The graphs K 5 , K 7 , and K 4 ,4 are nonplanar but may be embedded
onto the torus, see Fig. A6.5.1.
4 1
4 2 2
5 3 3
3 1 2 3 6 4 1 6
K5
6.5.2. The genus of the Petersen graph is 1. Hint: It is nonplanar; at the same
time, it may be embedded onto the torus, see Fig. A6.5.2.
5 1 2 3 4 5
9e-______~6----~.8 9
10
5 1 2 3 4 5
6.5.3. See Fig. A6.5.3.

6.5.4. Solution: From the formula (6.5.1) for the genus of the complete graph it
follows that i(I<s) = 2. Therefore all graphs homeomorphic to Ks are of
genus 2.
6.5.5. Hint: Analyze the formula (6.5.1). The next missing value for the genus of
Kn is 9.
0000 0001 0101 0100 0000
0010 0011 0111 0110 0010
1010 1011 1111 1110 1010
1000 1001 1101 1100 1000
0000 0001 0101 0100 0000
6.5.6. Hint:
a: With the exception of trivial cases K 1 , K 2 , P3 , every face of a connected
(n, m)-graph is bounded by at least three edges, hence 3/ $ 2m. Substituting
this inequality into the generalized Euler formula (Theorem 6.5.1) and taking
into 'account that the genus is an integer number, we obtain the required
inequality.
b: For a triangle-free graph an additional condition is 4/ $ 2m.
c: This follows from item a.

d,e: Notice that Qn and f{p,q are triangle-free (see Exercise 1.4.28 and
Konig's theorem from Sect. 1.2).
6.5.7. -y(G) ~ rm(I/2 - l/g) - (n - 2)/21-

Hint: See the hint to the previous exercise.
6.5.8. cr(f{5) = cr(f{3,3) = 1, cr(P) = 2, where P denotes the Petersen graph.

Hint: Since P is nonplanar, cr(P) > O. If it were cr(P) = 1, then the
deletion of some edge from P would produce a planar graph. But one may
show that even the deletion of a vertex from the Petersen graph leaves the
graph contractible to f{3,3' On the other hand, it is easy to draw the Petersen
graph on the plane with two crossings.
6.5.10.
cr(f{p,q)~ l~J lp;IJ l~J lq;IJ
for any p, q ~ 1.
Hint: Take the vertices ai, i = 1, ... ,p, at the plane points with coordinates
{( -1 )ii, 0) and the vertices bj, j = 1, ... , q, at the plane points with coordi-
nates (0, (-l)j j). Connect every ai with all bj by line segments and count
the number of crossings in such construction (in each quadrant separately).
6.5.11. The statement is false as a whole.

Solution: Let us prove that the sufficiency part fails. Let G be a maximal
planar graph (cr( G) = 0) that has two vertices u, v at the distance at least
3. Then one may prove by contradiction that cr(G + e) > 1, e = uv. In fact,
suppose that cr( G + e) = 1. Then there exists a plane embedding of G such
that one may add the edge e to it that crosses only a single edge e' of G.
Clearly, G' = G - e' + e is also planar. Therefore G' has a triangular face
uvw, see Fig. A6.5.4. But this means that dG(u, v) ~ 2. The contradiction
proves that cr(G + e) > 1.
e'
6.5.13. i{f{3,3) = i{f{5) = t{P) = 2, where P denotes the Petersen graph.

6.5.14. Hint: Consider the graph G' obtained from a nonplanar graph G by sub-
dividing every its edge with two new vertices of degree 2. Its thickness is
2.
6.5.15. Hint: It follows from the Corollary 6.1.5.
6.5.16. Hint: Use the Exercise 6.5.15 and the evident equality ra/b1 = L(a + b-
1)/bJ for positive integers a,b.
6.5.17. Hint: Notice that 2n-4 2: m for a connected planar bipartite (n, m)-graph,
see Exercise 6.1.13.
6.5.18. Hint: a: Use the previous exercise.

b: For Ql, Q2 the statement is obvious. Further, Qn, n 2: 3, has no triangles
(see Exercise 1.4.28); it has 2n vertices and n2 n - 1 edges. Therefore the
previous exercise impliest(Qn) =
rs(n)l, n 2: 2, where sen) n2n-l/2(2n-=
2). It is easy to prove that n/4 < s(n):::; (n+ 1)/4, n 2: 3. From the evident
inequalities
r(n + 1)/41 2: (n + 1)/4; r(n + 1)/41- 1 :::; n/4, n 2: 3,

we obtain rs(n)l = r(n + 1)/41, n 2: 3. This implies t(Qn) 2: r(n + 1)/41 =
Ln/4J + 1, n 2: 3.
6.5.19. Four edges.

Hint: The 4-cube Q4 contains a maximal (by the number of edges) planar
graph without triangles (see Exercise 6.1.27 and Fig. A6.5.5). Since Q4 has
32 edges, it is necessary to delete 4 edges.

6.5.20. sk(K3,3) = sk(K5) = 1, sk(P) = 2, where P denotes the Petersen graph.

Hint: Deleting two nonadjacent edges from the Petersen graph we may obtain
a planar graph. Therefore sk(P) ~ 2. On the other hand, the deletion of
a vertex from P produces a graph contractible to /(3,3 (while proving the
latter fact, notice the vertex symmetry ofthe Petersen graph). Therefore the
deletion of any edge from the Petersen graph results in a non planar graph;
hence sk(P) = 2.
6.5.21. Hint: The formula is true because the maximal number of edges in a planar
graph of order n is 3n - 6 (see Corollary 6.2.3).
6.5.22. sk(/(p,q) = pq - 2(p + q) + 4, p, q? 2; sk(Qn) = n2n - 1 - 2n +1 + 4, n ? 2.

Hint: Use the Exercises 6.1.24,6.1.26.
6.5.23. Hint: Use the Nash-Williams formula (Theorem 6.5.2). Since mp ~ 3(p-2)
for a planar graph G, (G) ~ 3.
6.5.24. A planar triangulation of order 5.
6.5.25. Hint: See the hint to the Exercise 6.5.14.
Graph Traversals
7.1 Eulerian Graphs

7.1.1. Hint: To prove the necessity, show first that if x is a cutpoint of an Eulerian
graph that belongs to a pendant block B of the graph then degB x is even.
Then apply induction over the number of blocks.
7.1.2. Hint: Consider an algorithm that at every iteration deletes the edges of
some simple cycle from the graph G. Then take G l to be the graph induced
by the set of all deleted edges and take G2 = G - EGI.
7.1.4. Hint: Use the fact that the deletion of the edges of some simple cycle from
an Eulerian graph produces a graph whose all nontrivial components are
Eulerian graphs.
7.1.6. No.
7.1.7. For every vertex x E VG the number LVEN(x) degv must be even.
7.1.8. The degrees of all vertices are of the same parity.
7.1.10. Hint: Consider the multigraph obtained by the duplication of every edge
of the graph.
7.1.11. Hint: Prove that if an Eulerian cycle is not simple then changing the
direction of traversal of any of its sub cycle produces a new Eulerian cycle
(see Exercise 7.1. 6) .
7.1.12. Hint: Use, e.g., the Exercise 7.1.4.
275
7.2 Hamiltonian Graphs
7.2.2. No.
7.2.3. No.
7.2.4. Hint: Due to the symmetry of the Petersen graph G, it is sufficient to

construct a Hamiltonian cycle in G - x for an arbitrary vertex x.
7.2.5. The required graph may be obtained from K 1 ,3 by subdivision of every its
edge.
7.2.6. The converse statement is false. Hint: See, e.g., the graph from Fig. A7.2.2.
7.2.7. Yes.
7.2.8. Hint: Use the fact that the deletion of m vertices from en produces a graph.
with at most m components.
7.2.11. Hint: Prove that Hn,m = Km +(Om UK n- 2m ) ("+" denotes the operation
of join) and use the Exercise 7.2.8.
7.2.12. Hint: Take m to be the minimal index such that dm < m and dn - m <
n-m.
(n-2)2+ n
7.2.13. Hint: Prove that IEHn,ml::; 2 . Further, make use the fact
that every non-Hamiltonian graph G does not satisfy the conditions of the
Theorem 7.2.1 and by Exercise 7.2.12 it has at most IEHn,ml edges.
7.2.14. Hint: Prove that the line graph does not contain induced subgraphs iso-
morphic to K 1 ,3 and use the Theorem 7.2.2.
7.2.15. Hint: Prove that the line graph is locally connected and use the Theorem
7.2.2.
7.2.17. Hint: Prove that every edge of G2 belongs to a triangle and use the Exercise
7.2.15.
7.2.18. Solution: The graph G is Hamiltonian if m, n ~ 2 and one of n, m is even.

If both n, m are odd then G has an independent set with (mn + 1) /2 vertices.
By Exercise 7.2.10, G is non-Hamiltonian.
7.2.19. Hint: Prove that the graph does not contain induced subgraphs isomorphic
to K l ,3 and use the Theorem 7.2.3.
7.2.20. Hint: Let d 1 ~ d 2 ~ ... ~ d n be the degree sequence of a graph G and let
further t be the maximal index such that 1 ~ t < n/2 and d t ~ t (the case
when d t > t for all t, 1 ~ t < n/2, presents no difficulties). Consider the
= =
sequence Q Q(t) (ql, ... , qn) with the following elements:
qk = 2 for k = 1, ... , t;
qk = k + 1 for such k that t + 1 ~ k < n/2;
qk = n/2 for such k that n/2 ~ k ~ n - t - 1;
qk= n-tfork=n-t,n-t+l, ... ,n.
7.2.22. Solution: The graph 0r3-1+ 1 + KL3-J-1 is a (In/2j - I)-connected graph

and by Exercise 7.2.10 it is non-Hamiltonian.
7.2.23. See Fig. A7.2.1.
7.2.24. Petersen graph.
7.2.25. Hint: Suppose the contrary and consider a Hamiltonian path connecting
two vertices of the graph whose deletion disconnects the graph.
7.2.26. Hint: Let x be a vertex of maximal degree in G. Prove that if degG x >
(n - 1) /2 then using a Hamiltonian path in G - x and the edges of G incident
to x it is possible to construct a Hamiltonian cycle in G.
7.2.27. Hint: Use the fact that t(X) ~ IVG\X! for every subset X of vertices of
en·
7.2.28. See Fig. A7.2.2.
7.2.29. See Fig. A7.2.3.

7.2.30. Solution: A possible solution is by contradiction. Let G contain a ver-

tex v that belongs only to 4-faces, see Fig. A7 .2.4a. Deleting this vertex
and contracting the resulting cycle, Fig. A7.2.4b, we obtain the graph G'
which is also a bipartite 3-polytope. The minimality of G implies that G' is
Hamiltonian. But given a Hamiltonian cycle in G', one may easily produce
a Hamiltonian cycle in G, see Fig. A7.2.4c,d, hence the contradiction.
7.2.31. Use the Theorem 7.2.5.
y y
z x z x z x
...-.-- .. z
a) b) c) d)

7.2.32. Hint: Using the induction over the number of vertices prove that for any
pendant edge of a caterpillar G belonging to a diametral path of G there
exists a Hamiltonian cycle in G2 containing this edge.
7.2.33. See Fig. A7.2.5.
7.2.34. No. A counterexample is the pair of sequences
P = (3,3,2,2,2), Q = (2,2,2,2,2).
7.2.35. Solution: Let us prove that the graph G of order n satisfies a condition
of the Theorem 7.2.1, namely, d k > k, for every k, 1 :S k < n/2. Suppose
the contrary, i.e., dt :S t for some t, 1 :S t < n/2. Consider a subgraph of
G induced by the set of vertices {Vl, ... , Vt} such that deg Vi = di . It is a
complete graph since the conditions 1 < t and V/Vt fI. EG imply d/ + dt :S n,
contrary to the assumptions. This together with the inequalities di :S t,
i = 1, ... , t, imply that every vertex Vi, i = 1, ... , t, is adjacent to at most
one vertex Vj, j = t + 1, ... , n. But n - t > t (since t < n/2), hence there
exists a vertex Vj, t + 1 :S j :S n, adjacent to no vertex Vi, i = 1, ... , t.
Therefore dj :S n - t - 1 and dt + dj :S n, which is a contradiction.
7.2.36. Hint: Add a new vertex to the graph, connect it with all other vertices by
edges, and use the Exercise 7.2.35.
7.2.37. Solution: The statement is evident if the vertices are adjacent. Without
loss of generality assume that degu ~ degv. Let H = (U,Xl, ... ,Xn -2,V)
be a Hamiltonian (u, v)-path in G. The inequality deg u + deg v ~ n implies
that {vxk,uxk+d ~ EG for some k, 1 :S k :S n - 2. Therefore G contains
a Hamiltonian cycle C = H' U H" U {VXk, UXk+1}, where H', H" are the
(u, Xk)- and (Xk+l, v)-subpaths of H respectively, see Fig. A7.2.6.
7.2.38. G = G 1 U G 2 , where G l ~ /(2, G2 ~ /(n-l, IVG l n VG21 = l.
7.2.39. See Fig. A7.2.7.
7.2.40. a: Solution: The necessity follows from the Exercise 7.2.10. The suffi-
ciency will be proved by contradiction. Suppose that the graph G has no
Hamiltonian cycles. Let C denote a simple cycle of maximal length in G and
280 Answers, Hints. Solutions
. . . . . . . . . . ·"iii·········--··· --.-_. . . _...... "H-'--,,- ---....

..~.................
•... • • • • •••• • >" ....• • ... ••- - 1•. ____•. - -......
-, .
Xk Xk+l X n -2 V
Figure A7.2.B: To Exercise 7.2.37
denote V' = VG\ VC. It is easily seen that if x E V' and Vk is the part of G
containing x then the vertices of Vk cover all edges of C. Clearly, the graph
may have at most two parts with this property. Therefore only two cases are
possible.
Case 1: V' ~ VI for a part VI of G. Then IVGI = IVII + IVC\VII < 21VII ::;
2ao( G), contrary to the assumptions of the exercise.
Case 2: V' ~ Vi U Vi for some parts Vi, Vi of G. Since each of Vi, Vi must
cover all edges of C and ao( G) ::; IVG1/2, it follows that G is the complete
bipartite graph with parts of equal sizes. This contradicts to the assumption
that G is non-Hamiltonian.
b: Solution: Let VI denote the largest part of G. Since G is not Hamiltonian,
the Exercise 7.2.40a implies IVII = ao( G) > IG1/2. Let G' denote the graph
obtained from G by the deletion of any 2IVII - IGI vertices from the part
VI. Clearly, IG'I = 2ao(G') and hence the Exercise 7.2.40a implies that G'
is Hamiltonian, i.e., G contains a simple cycle of length
IG'I = IGI- (21Vd - IGI) = 2(IGI- ao(G)) = 2f3o(G).
7.2.42. Hint: Use the Corollary 5.3.5.

7.2.44. Use the Exercise 7.2.43.
7.2.45. No. See, e.g., G = P4'
Degree Sequences
8.1 Graphical Sequences

8.1.1. a: Three realizations. b: Five realizations. c: The sequence is not graphical.
8.1.3. Hint: Use the fact that the described graph has either a dominating or an
isolated vertex.
8.1.7. No.
8.1.8. a: See Fig. A8.l.l.
o 0
N===N
I
N
I
N
I I
~O/
o
o ==N--O - N== 0 O==N-<::o

b: Solution: Let a tree T be the graph of a hydrocarbonate. Let Q denote

the subtree of T induced by the vertices of degree four. If IQI = n then
LVEVQ degQv = 2(n - 1). Now it is clear that the number of the pendant
vertices (which correspond to the hydrogen atoms) must be 4n - 2(n - 1) =
2n+2.
8.1.9. No. For example, d = (3,1 3 ), d' = (1 4 ).
281
8.1.10. s = (15,62), s = (83,65), s = (12,74).

8.1.11. No. Hint: Consider the graph shown at Fig. A8.1.2.
G sG
8.1.12. Solution: Consider a realization G of an n-sequence d in which the vertex

Vi is adjacent to the maximal number of vertices of first maximal degrees.
Suppose that the claim of the Exercise is false. Then among the first di
vertices (excluding Vi) there exists a vertex Vj which is not adjacent to Vi.
Let Vk be a vertex of maximal degree from N(vd. Clearly, deg Vj 2: deg Vk.
Therefore there exists a vertex VI E N (Vj) not adjacent to vi' The graph sG
for the switching s = (ViVk,VjVI) contradicts to the choice ofG.
8.1.13. One or two.
8.1.14. Hint: For multigraphs the switching is admissible even in the case when
a, band/or c, d are adjacent pairs of vertices. For pseudographs, additionally,
any vertices from a, b, c, d may coincide.
8.1.15. May change at most by 1.
8.1.16. a: Gd = P3 (the graph of realizations Gd' and the realizations of the
sequence d' are shown at Fig. A8.1.3).
c: Hint: Use the Switching Theorem.
d: 0 for n ~ 5, In - 4/2J for n 2: 6. Hint: Use the Exercise 8.1.15.
8.1.17. a: Hint: Use the Switching Theorem.
b: Hint: Let realizations G, G' be constructed over the same vertex set
V, where degG V = degG' V for all V E V. Consider the auxiliary graph
H = (V, E'), where E' =
(EG\EG') U (EG'\EG), and prove that every
nontrivial component of H may be considered as an alternating cycle of G.
Clearly, a nontrivial component of H contains at least four vertices, therefore
sed) ~ lIdl/4J.
• •
G5
S.1.1S. b: Hint: The sequence (2 3 ,1 2 ) has 7 labelled realizations.

c: G~ ~ I<2.
d: Assume that the sets Vl,"" VA: are the vertices of Gd. The vertices
=1= j, are adjacent if and only if G~ has adjacent vertices
\Ii, Vj E VGd, i
VE\Ii,WEVj.
e: Hint: Use the previous exercise.
f: Hint: The proof of the Switching Theorem implies that the function
s/(n) grows not faster than O(n 2 ). On the other hand, consider the sequence
d = « 2k )4k+ 1 ) and two its disjoint realizations G, G'. To transform G into G' ,
at least k(4k+1)/2 switchings are necessary. Therefore s(n) ~ s(d) = O(n2),
where n = 4k + 1.
S.1.19. b: Let G d be the graph of realizations and let the set Vp ~ V G d consist of
the vertices that correspond to the realizations with the property P. If P is
s-complete then for any graphical sequence d either Vp = 0 or the induced
subgraph Gd(Vp) is connected, and vice versa.
c: Hint: Consider two cacti shown at Fig. AS.1.4.
Figure AS.1.4: To Exercise 8.1.19c

8.1.20. No.
8.1.21. Yes. Hint: Consider the Cartesian product of realizations of the sequences
d and c.
8.1.23. a: Yes. b: No. c: No.

8.1.25. Assume that k = k(d) and m > k. Then using the inequalities k + 1 2::
dk+1 2:: ... 2:: d n and 2mk - k 2 - k :::; m 2 - m we may compute
m k n n m
L d; = L d; + L d;:::; k( k - 1) + L min{ i, dol + L d; =
;=1 ;=1 i=k+1 i=k+1
n m n
k(k-l)+ L min{i, dol+2 L di :::; k(k-l)+2(m-k)+ L min{i, di } :::;
i=k+1
n n
2mk-k2-k+ L min{i,dol :::;m(m-l)+ Lmin{i,d;}.
i=k+1 i=1
8.1.26. a: Yes. b: No.
8.1.27. a: nm is even and m < n. b: m + n is even and 0 < m :::; n.

c: For m 2 + n 2 > O. d: For no m, n.
8.1.28. For n = 4k or n = 4k - 1, k 2:: 1.
8.1.29. Hint: Prove that this sequence is realized by the complete k-partite graph.
8.1.31. The necessity is evident. The sufficiency will be proved by induction over
u = 2:7=1 di . The base of the induction (u = 0) is evident. Assuming the
inductive hypothesis, consider a proper n-sequence d subject to the condition
u 2:: 2d 1 > 0 (the latter implies d 2 > 0). Let us construct the sequence
d' = (d l -1,d2 -1,d3 , ..• ,dn ) and let us order it: d' = (dl,d~, ... ,d~),
d1 > d~ > ... > d~.
Let us prove that u' = 2d'. If d~ = d 1 - 1 then u' = u - 22:: 2d 1 - 2 = 2d~.
Else if di = d 1 then d3 = d 1 and then for d 1 > 1 we have (J' > 3d 1. Therefore
u' = (J' - 2 2:: 3d l - 2 2:: 2d 1 = 2di. Finally, if di = d 1 = 1 then the evenness
of u and the equality di = d 1 imply d3 = 1. Therefore d4 = 1, and hence
u' = u - 2 2:: 4 - 2 = 2 = 2di.
Thus, by the induction hypothesis, the sequence d' is realizable by a multi-
graph G'. It remains to connect the vertices u, v of G' with degrees d1 - 1
and d2 - 1 by the edge to obtain a realization G of the sequence d.
5 4 2 3 2 0
4 0 1 1 1 1 0
4 1 0 1 1 1 0
8.1.33. a: 3 1 1 0 1 0 0
2 1 1 0 0 0 0
2 1 1 0 0 0 0
1
1 0 0 0 0 0
c: Hint: Prove that the Fulkerson theorem is equivalent to the Erdos-Gallai
one.
8.1.34. Hint: Verify that the sequence dU(dn1 ) satisfies the Erdos-Gallai inequal-
ities.
8.1.35. Hint: Consider the sequence ((8n - 3)4n, 28n2-4n-l), n ~ 1.

8.1.38. a: Yes. b: No.
8.1.39. Hint: Let d =
(d1, ... ,dn ), d' =
(d~, ... ,d~) be two integer sequences.
Prove that the pair (d, d') is graphical if and only if the (m + n)-sequence
(d 1 + n - 1, ... , dn + n - 1, d~ , ... , d~) is graphical.
8.1.40. a ~ m, b ~ n, na = mb.
8.1.41. Hint: Use the Konig representation of the multigraph, which is defined
similarly to the Konig representation of the graph, see Exercise 1.2.36.
8.1.42. Hint: The Switching Theorem for pairs is stated as follows. Any realization
of a graphical pair of sequences may be obtained from any other one by a
sequence of bipartite switchings.
8.1.43. a: n 2: 1. b: n ~ 4. c: n ~ 5. d: n ~ 1 or m ~ 1 or m = n = 2.
8.1.44. Yes.
8.1.45. No. Hint: Consider, for example, C4 .
8.1.46. Hint: Use the following statement: graphs G and G are unigraphs or not
simultaneously.
8.1.47. C3, C 4 , C 5 .
8.1.48. /{l, P4, C 5 .
8.1.49. No.
8.1.51. Hint: Prove that the only realization of the sequence is the graph in which
the vertices of degree s induce the complete graph and the remaining vertices
induce the empty graph, and the neighborhoods of the vertices of degrees
k 1 , ... ,k/ do not intersect.
8.1.53. Yes.
8.1.54. Yes.
8.1.55. The components are stars and graphs that may be produced from two
vertex-disjoint stars by connecting their centers by an edge.
8.1.56. a: Yes. b: No.
8.1.57. b: A minimal FIS set is {2K2' P4 , C4 }.
8.1.59. a: Yes. b: Hint: Notice that every vertex v is contained in degv edges of
the graph. d: The simple paths Pn , n ~ 5.
8.1.60. Hint: One may easily produce the degrees of all edges incident to a vertex
v from the list deg l v.
8.2 P-graphical Sequences

8.2.2. a: The sequence d is forcibly Pi-graphical for i = 1,2.
b,c: These properties cannot be expressed in terms of potentially and forcibly
Pi-graphical sequences, i = 1,2.
d: The sequence d is either potentially PI-graphical or potentially Pl-grap-
hical.
8.2.3. a,b: The first class is contained in the second one.
8.2.4. a: Hint: Use the Exercise 8.2.2a.
b: The inclusion PT ~ P A nPc is evident. Let us prove the reverse one. Let
d be a graphical sequence such that d EPA n Pc. Let m = I:~=1 dd2 denote
the number of edges in every realization of d; m ~ n - 1, since dE PA. On
the other hand, m ~ n - 1, since dE Pc. Thus every acyclic (or connected)
realization of d is a tree, i.e., dEPT.
8.2.5. For those and only for those properties which may be defined in terms of
degree sequences.
8.2.7. c: Hint: Let us select any element d; of the tested n-sequence d. If d is
realized by a complete multipartite graph then the size of the part containing
a vertex of degree d; is ni = n - di . Let Ii denote the arity (multiplicity)
of di in d. If ni > Ii, then d cannot be realized by a complete multipartite
graph. Otherwise we construct the sequence d' = d\{dfi) and apply the
above reasoning to it.
8.2.8. Hint: The sufficiency may be proved by induction over n. The inductive
step may be based on the Havel-Hakimi theorem.
8.2.9. Hint: a: The sufficiency may be proved by demonstrating that dn = 1 and
the sequence d' = (d l - 1, d 2 , .•• , dn-d also satisfies the assumptions of the
exerCIse.
8.2. P-graphical Sequences 287
8.2.10. n ~ 3 and d3 ~ 2.
8.2.12. Solution: The necessity is easily verified. Sufficiency: Let G be a labelled
realization of a sequence d = (dl,"" dn) with n ~ 3, d l , ~ d2 ~ d3 > 1,
2: di = 2n. It is sufficient to prove that d is realizable by a connected graph.
Suppose that G is not connected. Then at least one its component HI
contains a cycle C. Let us switch an edge of C with an edge of a component
H2 =1= H l . Clearly, this produces a realization of d with less components
than G has. Repeating such switchings, we ultimately arrive at a connected
realization.
8.2.13. «n - l)n) or «n - 1)2, (n - 1)n-3, 2). Hint: Prove that if the line graph
L(G) has exactly two dominating vertices then G = KI + (K2 U On-3), and
then L(G) = K2 + (K2 U K n - 3 ).
8.2.14. Solution: a: Proceed by induction over n. For n = 1 and for the empty
sequence the statement is evident. Suppose that the inequality holds for
any (n - 1)-sequence. Let d = (d l , ... , dn ) be some nonzero potentially
acyclic sequence and let G be one of its realizations. G has a pendant
vertex v. We delete it and obtain a graph G' with the degree sequence d' =
(d\(dj, 1)) U (dj - 1), where dj is the degree of the vertex adjacent to v. By
the induction hypothesis, 2:?;;/ d'~ :::; (n - 1)2 - (n - 1). Therefore
2:?=1 d; = 2:?;;/ d + 2dj + 2 :::;(n -
l; 1)2 - (n - 1) + 2(n - 2) + 2 = n 2 - n.
b: The stars.
8.2.15. Let G be a labelled realization of a potentially planar sequence d. Assume

that the vertices are indexed in such a way that deg Vi di , i =
1,2, ... , n. =
Denote Xi = N(Vi)\{Vl,V2,V3}, i = 1,2,3. Then
Further,
IX l n X 2 n Xgl = IX l n X21 + IXgl-I(Xl n X 2 ) U Xgl ~
«d l - 2) + (d 2 - 2) - (n - 3)) + (dg - 2) - (n - 3) = d l + d2 + d3 - 2n.

On the other hand, IXl nX2 nXgl :::; 2, since a planar graph does not contain
Kg,g.
8.2.16. Any graphical sequence d, d 1 ~ d2 ~ ••• ~ dn , with dg :::; 1 and the

sequences (24, on-4), n ~ 4.
Solution: Clearly, these sequences are forcibly bipartite. Let us prove that
there are no more forcibly bipartite sequences. Consider a forcibly bipar-
tite n-sequence d with d l ~ d2 ~ ••. ~ dn ~ 0 and d3 > 1. Using the
Havel-Hakimi theorem, it is easy to construct a labelled realization G of this
sequence in which the vertex Vl is adjacent to the vertices V2, V3. Since Gis
bipartite, Vl, V3 are not adjacent. Since we are given that deg V2 = d2 ~ 2
and deg V3 = d3 ~ 2, the graph G has vertices x and y adjacent to V2 and V3

respectively.
Suppose that x f. y first. The vertices x, yare not adjacent, otherwise the
bipartite graph G would contain an odd-length cycle (Vl,V2,x,y,Va). The
switching s = (V2X, V3Y) produces a bipartite graph sG. But this contradicts
to the fact that G has the triangle (VI, V2, V3)'
It remains to consider the case x = y. Clearly, the vertices VI, V2, V3, x induce
the cycle C 4 . Let us prove that in fact G is this cycle. Assume the contrary:
there exists a vertex z (/. VG\ VC4 . Since the considered sequence has no
zeroes, deg z 2:: 1. If z is adjacent to a vertex of the induced C 4, then G(VC4U
{z}) has the degree sequence either (3,2 3 , 1) or (3 2 ,23 ). Both these sequences
are not forcibly bipartite, which contradicts to the forcible bipartiteness of
d.
Therefore G has no vertex z E VG\VC4 adjacent to a vertex of the induced
C4 . But if G f. C4 then it has the induced subgraph C4 U K2 with the
degree sequence (24, 12), which has a nonbipartite realization C 3 U P3, which
is impossible.
8.2.18. An n-sequence d with d l ~ ... 2:: dn ~ 1 is forcibly unicyclic only in the

following cases:
1) d = (2 3 ), d = (24), or d = (2 5 ),
2) d = (3,2 4,1),
3) d = (n - 2,2 3 , 1n-4), n 2:: 5,
4) n 2:: 4, da 2:: 2, d4 = 1, d l + d2 + d3 = n + 3.
Hint: A unicyclic graph is a simple cycle Ck at every vertex V of which a
tree Tv is " planted" . Use the following scheme for the solution. Let G be a
realization of a forcibly unicyclic sequence d. Prove that
1) the length of the cycle Ck is at most 5;
2) every Tv may be assumed to be a star or KI;
3) For k = 5 there may be only a single star Tv = K 2 ;
4) For k = 4 there may be only a single star Tv = KI,t.
8.2.19. a: No. b: Yes.
8.2.20. A proper graphical n-sequence d is forcibly outerplanar if and only if the

sequence d' = (d 2 - 1, d3 - 1, ... , dn - 1) is forcibly realizable by a disjoint
union of simple paths. Therefore d2 - 1 ::; 2 and d4 - 1 ::; 1.
8.2.21. a: b ::; c.
8.2.22. a: For example, the property specified by two forbidden induced subgraphs
K2 U K3 and P6 •
8.3. P-graphical Sequences 289
8.2.23. a: d~(P) = {(2 3 ), (24), (2 5 )}.
b: d~(P) = {(2 3 ), (2 5 )}.
8.2.24. b: For example, Up = Pu holds for the property P "to be a unigraph"

and Us = Bu does not hold for the property B "to be a bipartite graph".
In fact, the sequence (2 6 ) is not unigraphical, therefore (2 6 ) ¢ Us. But this
sequence has a single bipartite realization C6, hence (2 6 ) E Bu.
8.2.25. Let a proper n-sequence d have a tree realization (see Exercise 8.2.9).
Then dE Tu only in the following cases: 1) n :::; 2; 2) n ~ 3 and d3 = 1; 3)
d1 = d2 = d3 ~ 3; 4) d1 = 2.
Hint: Let an n-sequence d with n ~ 3 be realizable by a tree T. Let W denote
the set of non pendant vertices of T and let us supply every vertex w E W
with a nonnegative integer weight c5(w) = degT w-degT(w) w. Let d'denote
the degree sequence of the subtree T(W). Two realizations T' and Til of d'
over the vertex set Ware called c5-isomorphic if there exists an isomorphism
cp: T' -+ Til that preserves labels, i.e., c5(cp(w» = c5(w) for all w E W. Prove
that the sequence d is T-unigraphical if and only if any two realizations of
d' are c5-isomorphic. Further, find all sequences d' that have single (up to
c5-isomorphism) realizations, as well as the corresponding labelling functions
for vertices. You will obtain the following: for d' = (0) or d' = (12) the
labelling function may be arbitrary. For d' = (2, 12) the labelling function is
c5 = const ~ 2. For d' = (2 n , 12) the labelling function is c5 = const = 2.
8.2.26. The converse statement fails. For example, consider the property "to be
isomorphic to one of graphs C8 or 2C4 ", which is not specified by a degree
sequence.
8.2.27. No. Let the property P be "to be a tree" and let the property P' be "to
be a bipartite graph." Clearly, P ~ P'. But (24,12) E Pu and (24,12) ¢ Ph.
8.3 Split and Threshold Graphs

8.3.2. Yes.
8.3.4. a: Yes. b: No.
8.3.5. n.
8.3.7. At most n + 1. The maximum is achieved for I<n.

8.3.8. Hint: Use Theorem 8.3.1.
8.3.9. A minimal FIS set is 2I<2, C4 , C5 •

Solution: It is immediately seen that the above graphs are not split. Further,
let G have no induced subgraphs 2I<2, C4 , C5 . Let us select a maXImum
cardinality clique I< in G such that G(VG\I<) is empty.
Suppose the contrary, i.e., I = VG\K has a pair of adjacent vertices u, v.

Since the clique K is of maximal cardinality, there are vertices u', v' E K
that are not adjacent to u and v respectively. We may consider that u' # v',
for if it were impossible to choose u' "# Vi, then K would be not of maximal
cardinality: (K\ {u'}) U {u, v} would be a larger clique.
Since G has no induced subgraphs 2K2 , C4, exactly one of ulv, vlu is in EG.
Assume, e.g., that u'v E EG and v'u f!. EG.
Let us prove further that v is adj acent to all vertices from K\ {Vi}. Sup-
pose the contrary, i.e., that v is not adjacent to w E K\{v'}. Clearly, w
is adjacent to u' and v'. Moreover, wand u are adjacent, for otherwise it
would be G({u,v,v',w}) = 2K2 • But then G({u,v,u',w}) = C 4 , which is
also impossible.
Therefore G contains a clique K' = (K\{ v'} )U{ v} of cardinality equal to IKI.
The definition of K implies that IEG(I) I $ IEG(I')I, where I' = VG\K'.
Since uv E EG(I) and uv f!. EG(I') , it follows that there exists a vertex
x E I adjacent to v' but not adjacent to v, see Fig. A8.3.l.
Figure A8.3.I: To Exercise 8.3.9
The vertices u, x are adjacent, for otherwise it would be G{ {u, v, v', x}) =
2K2. The vertices u', x are not adjacent, for otherwise it would be
= =
G( {u, v, u', x}) C4 . But then G( {u, v, u', v', x}) C5, which is also impos-
sible.
The resulting contradiction show that G(I) is empty.
8.3.10. a: Hint: Use the Theorem 8.3.1.

b: Hint: Consider the set of vertices with degrees di = i-I.
8.3.11. Hint: Use the following mapping of the set of graphs into the set of the
split graphs. For a given graph G we construct its Konig representation
K(G) (see Exercise 1.2.36). Then we construct the complete graph over the
vertices of K (G) that correspond to the set V G.
8.3.12. No. See the example at Fig. AB.3.2.

B.3. P-graphical Sequences 291
8.3.13. Trees with diameters at most three.

8.3.14. The complete and the empty graphs.
8.3.15. Hint: Prove that 2K 2 , C4 , C5 are not threshold and use the characterization
of split graphs in terms of forbidden induced subgraphs, see Exercise 8.3.9.
8.3.16. Let us assume that the vertices are numbered in the non decreasing order
of their degrees. Then the table below shows all graphs of order four and
the corresponding separating inequalities.
04 Xl + X2 + X3 + X4 ~ 4
K2 U O2 Xl + X2 ~ 2
P3 UK l 2Xl+X2+X3~2
K 3 UK l xl+x2+x3~1
K l ,3 3Xl + X2 + X3 + X4 ~ 3
P3 UKl 3Xl+2x2+2x3+X4~3
K4 - e2Xl + 2X2 + X3 + X4 ~ 2
K4 Xl + X2 + X3 + X4 ~ 1
The remaining four-vertex graphs are non-threshold.
Hint: See Exercise 3.1.37i.
8.3.17. The vertices of the n-dimensional cube correspond to the (0, I)-vectors
of length n. A subset of the vertices of this cube correspond to the char-
acteristic vectors of independent sets; let us denote it by I. To construct
a separating inequality means to construct a hyperplane in R n such that
I lie in the negative (i.e., containing the origin) closed halfspace and the
remaining vertices are in the positive open halfspace. An example of the
construction of the separating inequality 2Xl + X2 + X3 = 2 for G = Pa ,
V P3 = {I, 2, 3}, EP3 = {12, 13} is shown at Fig. AS.3.3:
8.3.18. Solution: Let alXl + ... anXn ~ b be a separating inequality for a threshold
graph G. Clearly, it will also be separating for G U K 1.
Further, by the previous exercise we may assume that the coefficients ai are
positive. Then the inequality alXl + ... anXn + bX n +l ~ b is separating for
G+K 1 .
8.3.19. Solution: Let us renumber the vertices of a threshold graph G in the

nondecreasing order of their degrees. Let u have a larger index than v and
z
N( v) ¢ N( u)U{ u}, i.e., there exists a vertex w E N( v) which is distinct from

v and not adjacent to it. Since deg u ;::: deg v, there exists a vertex x E N( u)
distinct from u and not adjacent to it. Clearly, G( {u, v, w, x}) is isomorphic
to one of graphs 2K2 , P4, C 4 . This is impossible, since these graphs are not
threshold graphs, and the thresholdness is a hereditary property.
8.3.21. A minimal FIS set is 2K2, P4 , C4 . Hint: Use the Theorem 8.3.2 and the
fact that if a graph G does not contain the induced subgraphs 2K2 , P4 , C 4 ,
then G U K 1 and G + K 1 does not contain such induced subgraphs.
8.3.22. Hint: See Exercise 3.1.37.
8.3.24. Om, K 1 ,/, K 1,/ U Om, I ~ 1, m ~ 1.
8.3.25. Hint: Prove that the threshold graphs and only the ones do not admit
switchings.
8.3.26. K 1 •
Hint: Use the Exercise 3.1.37j.
8.3.27. Hint: Use the Theorem 8.3.2 and the fact that the thresholdness property
is hereditary.
8.3.28. Hint: A possible labelling is the coefficients of the separating inequality

constructed for the solution of the Exercise 8.3.22; t = 8.
8.3.29. Hint: Compare the characterizations of split and threshold graphs in terms
of forbidden induced subgraphs (see Exercises 8.3.9,8.3.21).
8.3.30. Hint: Renumber the vertices of each part in the non decreasing order of
their degrees and use the previous exercise.
8.3.31. Hint: Use the fact that the stars are threshold graphs.
8.3.32. a:Hint: Use the following property: if G =

U7=1 Gi, V VG VGi,= =
i = 1,2, ... , n, then the set I ~ V is independent in G if and only if it is
independent in all G i , i = 1,2, ... , n.
8.3.33. a: Ln/2J. b: min{m, n}. c: Ln/2J.
8.3.34. a: Hint: Let I be a maximum cardinality independent set of a graph G

and VG\I = {I, 2, ... , t}. Construct the spanning subgraphs G I , G2 ,· •• , Gt
in such a way that Gi contains all edges incident to i, i = 1,2, ... , t.
b: Hint: : Consider the fact that in a triangle-free graph the edges of ev-
ery threshold subgraph without isolated vertices constitute a star. Use the
equality ao(G) = IGI- Po(G).
c: Hint: Use the equality al(G) = Po(G) for bipartite graphs.
8.3.35. The threshold numbers are 3 and 2 respectively.
8.4 Degree Sets and Arity Partitions

8.4.2. b: Hint: To construct a realization of order kl + 1, one may apply the
induction over even and odd values of lSI separately.
Consider the case of odd 151. Assume that 5 = {k l ,k 2 , •.• ,kn }, kl ~ .. , ~
k n . For n = 1 the required realization is the complete graph of order kl + 1.
Assume that n ~ 3. By the induction hypothesis the set 5' = {k2 - kn, kg -
k n , ... , k n - I - k n } is realizable by a graph G' of order n' = k2 - kn + 1.
Consider the complete split graph H = H(A, B) with the complete part A
of size kn and the empty part B of size kl - k n . Finally, construct a graph
realizing the set S by connecting every vertex of G' with every vertex of the
part A of H by an edge.
8.4.3. Hint: See the hint to Exercise 8.4.2b.
8.4.4. b: L~=I (ki - 1) + 2.

8.4.5. Hint: Use the Exercise 3.1.37i.
8.4.6. No.
8.4.7. Hint: Use the induction over n. The induction step for n ~ 3 is to be
performed as follows: H = «H')o U Ok 1 -k 2 ) U Ok n -2, where H' is a Hamil-
tonian realization ofthe set S' = {k2 - k n + 2, k3 - k n + 2, ... , k n - 1 - k n + 2},
which exists by the induction hypothesis, and (H'Y is the graph obtained
from H' by the deletion of the edges of some Hamiltonian cycle.
8.4.8. 1:S k n :S 5.
8.4.9. c:Hint: Use the graph shown at Fig. A8.4.I.
a b
8.4.10. For n = 2k + 1 and n ~ 2k + 3.

8.4.11. Yes.
8.4.12. a:Hint: First, reduce the problem to the case IS11 = IS21 = 1 and then
construct an appropriate complete bipartite graph.
b: Hint: Construct a bipartite graph realizing the pair of sequences
(kL k~, ... , k~) and (lk), where k = kl + k2 + ... + kn .
8.4.13. See Fig. A8.4.2.
8.4.14. a: A graph shown at Fig. A8.4.3. b: Such graph does not exist. c: Petersen
graph.
8.4.15. a: (n - 2,2). b: (n). c: (m, n).
8.4.16. No.
8.4.18. For n 2:: 2 and (Pl, P2, ... , Pn) :f:. (1,1, ... ,1).
Hint: Use the induction over n. The induction step is performed as follows:
add Pn isolated vertices to a realization of the partition (Pl,P2,'" ,Pn-l)
and take the complementary graph.
8.4.19. No. Hint: Consider the case n = 2.
8.4.20. Yes. Hint: Use the method of the solution of Exercise 8.4.18.
Graph Colorings
9.1 Vertex Coloring

9.1.1. c: Hint: Consider the bipartite graph Kn,n - M, where M is a perfect
matching and n 2:: 3.
9.1.2. b: Set HI = K I . If Hj-l is already constructed with V Hj-l = {Xl, ... , X n },

Hj is constructed as follows:
v
see Fig. A9.1.1.
•
H3
Clearly the single maximum cardinality independent set of vertices of Hj

is I = VHj\VHj _ l , where Hj - 1= Hj-I. Therefore the algorithm uses
exactly j colors. Since Hj is a tree, X(Hj) = 2.
297
9.1.3. Hint: To prove the necessity, show that if a graph G has a cycle of odd length
then its vertices must be properly colored in two colors, which is impossible.
9.1.4. a: n. b: 2. c: 2. d: 2 if n is even and 3 otherwise. e: 3.
9.1.5. Hint: Use the induction over the number of blocks.

9.1.6. Hint: Consider the graph /{I,n.
9.1.7. /{n and C2n+l, n ~ 1. Hint: Use the Brooks theorem.
9.1.8. Hint: For A ~ 3, consider the neighborhood of some vertex of maximal

degree and apply the Brooks theorem.
9.1.9. Hint: Otherwise it would be possible to recolor the vertices of color i in

color j.
9.1.10. a: 8,2,8,4. Hint: The castle and bishop graphs contain /{s as a subgraph.
Since the knight changes the color of the square at each move, its graph
is bipartite. For the case of the king, use induction over the size of the
chessboard.
c: n, 2, n, 4.
9.1.11. Hint: Use the fact that a line graph has no induced subgraphs /{1,3.
9.1.12. Hint: If two nonadjacent vertices of the octahedron graph are colored in
colors 1 and 2 respectively then the remaining four vertices (constituting a 4-
cycle) must be colored in two colors 3,4. Otherwise, if every two nonadjacent
vertices of the octahedron graph are colored in the same color, then only three
colors are used, and some 4-cycle is bicolored.
9.1.13. Solution: Consider the coloring <p in which every color is used at least
twice. Let ~ be a minimal coloring which has a maximal number of color
classes common with <p. Suppose the contrary, i.e., color 1 is used in ~ only
once, say, at vertex 2:. Let C denote the color class of <p containing 2:. Convert
~ into a coloring ~' by coloring the vertices of C in color 1. Clearly, e' is a
minimal coloring. All vertices of the set C\ {2:} have different colors in the
colorings <p and ~, therefore ~' has more color classes in common with <p,
contrary to the choice of ~.
9.1.14. Solution: Set X = X(G). Let the values of n, ~ be fixed. Find the maximal
possible value of m. Let k 1 , ... , kx denote the cardinalities of color classes of
some minimal coloring. Then
x x
2m ~ k 1 (n- kd+ ... +kx(n - k x ) = (kl + ... + kx)n - L k; = n2 - Lkl-
;=1 ;=1
Consider the optimization problem:

. . .
mInImIZe ",x k 2 b·
9 = 6;=1 i su Ject to 6i=1 i
",x k n. It is well known that its
solution is ki = nix, i = 1, ... , x. Therefore g 2:: n2 lx. Then

x
2m:::; n2 - Lkl:::; n2 - n2lx,
i=l
and hence X 2:: n 2 /(n 2 - 2m).

9.1.15. Hint: Use the previous Exercise.
9.1.16. Hint: In a minimal coloring, for any two colors i,j there exists an edge
whose one end has color i and the other one has color j.
9.1.17. Solution: Suppose that io = max{i: di + 1 2:: i}, i.e, dio + 1 2:: io, but
dio +1 + 1 < io + 1 or dio +1 + 1 :::; i o. Further,
. {.Z, di
mm + 1} = {i' + 1
di ,Z
i. <
- io;
.
> Zoo
Therefore
m!tX min{i, di
l$a$n
+ 1} = max{io, dio+1 + 1} = io.
Thus it is necessary to prove that x( G) :::; i o.
Let us color the vertices VI, ... , Vio with degrees dl , ... , di o in colors 1, ... , io
respectively. The neighborhood of every remaining vertex contains at most
dio+1 :::; io - 1 vertices. Therefore the consecutive coloring of the remaining
vertices will not use new colors.
9.1.18. a: Hint: Use the fact that the size of every color class does not exceed the
independence number.
b: K n - Olo +0 010 •
9.1.19. Yes.
9.1.20. Solution: Clearly, ao(G x Km) :::; IGI for any m ~ 1, since the vertices of
G x Km may be partitioned into cliques of size m.
Let us prove that m ~ X(G) if and only if ao(G x Km) = IGI.
Necessity: Suppose that X(G) = x.
V(G x Km) = {(V, i) : V E VG,i = 1, .. . ,m}.
Let C 1, ..• Cx denote the color classes of G. Consider the sets Ri = Ci X i,
i = 1, ... , m. Clearly, the sets Ri are independent, and moreover, their union
=
R is also independent. Since IRI IGI, the necessity is proved.
Sufficiency: Let I be a maximum cardinality independent set of vertices of
a graph G x Km. Then I contains at most one vertex of type (v, i) for every
v E VG. Since III = IGI, exactly one vertex of kind (v, i) belongs to I for
every v E V G. Let us define an m-coloring of G as follows. If (v, i) E I
then v obtains color i. It is easily seen that this coloring is proper, hence
X(G):::; m.
9.1.21. Yes.
9.1.23. a: Solution: Assume that VG l =VG 2 , for otherwise one could add isolated
vertices to G l or G 2 without changing the chromatic numbers. Further, every
vertex u of G 1 U G 2 is colored in color (0:1,0:2), where O:i is the color of the
vertex u in the minimal coloring of Gi, i = 1,2.
c: X(G 1 U G 2 ) = max(x(GI), X(G 2 )).
9.1.24. Let G l , ... , Gx be the color classes in a minimal coloring of G. The sets Gi
induce complete graphs in G. Therefore X 2: maXi IGd 2: nix, i.e., XX 2: n.
It is known that (a + b)/2 2: Vab for any positive a, b, hence X + X 2: 2"fii.
The inequality X + X ::; n + 1 is proved by induction over n. It is valid for
n = 1. Assume that X(G)+X(G) ::; n for any graph G of order n-1. Consider
a graph H of order n and a vertex v E V H. The graphs G = H - v and G =
H - v have n - 1 vertices. Clearly, X( H) ::; X( G) + 1 and X( H) ::; X( G) + 1. If
at least one if these inequalities is strict then X(H)+X(H) ::; X(G)+X(G)+I,
and we are done.
Suppose that X(H) =
X(G) + 1 and X(H) =
X(G) + 1. This means that
the deletion of the vertex v from G decreases its chromatic number. From
the Exercise 9.1.21 it follows that degG v 2: X(H). Similarly, degav = n-
degG v-I 2: X(H). But then X(H) + X(H) ::; n - 1.
Finally, the evident inequality XX ::; (X(H) + X(HW /4 implies XX ::; (n +
1?/4.
9.1.25. a: Kn, n > 1. b: Kq U ... U Kq (q terms) for n = q2. c: K t U Ot-l for
n = 2t - 1.
=
9.1.26. Solution: Suppose that there exists a graph G such that X + X 2"fii and
= = =
X+X (n+l)2/4. Then X 2fo-x implies XX X(2"fii-X) (n+l)2/4. =
This gives
Since this equation must have real roots, its discriminant is nonnegative:
4n - 4(n + 1)2/4 2: 0, i.e., -en - 1)2 2: O. Hence n = 1, contrary to the
assumption.
9.1.29. a: Hint: Take VI to be a color class of a minimal coloring of the graph G.

b: Solution: Consider a maximal complete subgraph Hl of G. Denote Vl =
= = =
VH 1 , V2 VG\VH 1 , H2 G(V2), IVtI k. We must prove that X(H 2) >
X(G) - k. Suppose the contrary, i.e., H2 admits a proper (X(G) - k)-coloring
in colors 1,2, ... , X(G) - k. Let us color Hl in colors X(G) - k + 1, ... , X(G).
Let T denote the set of all vertices of color 1 adjacent to a vertex u of color
X(G). Since Hl is a maximal complete subgraph, every vertex vET is
not adjacent to some Vi E HI. If we color every vET in the color of the
corresponding vertex Vi, the resulting coloring will be proper. Now we color
u in color 1 and obtain a proper coloring of G with X( G) - 1 colors, which
is a contradiction ..
9.1.30. Solution: For k = 1 the statement is trivial. Let us apply induction over
k. Assuming the induction hypothesis, consider the graph G\Vk • We have
Let us color G\ Vk in colors 1,2, ... , n -lVkl- k + 2 and the vertices of Vk in

colors n -lVkl- k + 3, ... , n - k + 2. Now let us recolor the graph to get rid
of one color. By the assumption, every set Vi, 1 :S i :S k - 1, has a vertex Yi
which is not adjacent to some vertex Xi E Vk. There exists a color, say, color
1, which is used only for the vertices of the set Y = {Yl, ... , Yk-d, since
IVG\YI = n - k + 1 is less than the number of colors. Let us recolor every
vertex Yi of color 1 in the color of the vertex Xi. This produces a proper
coloring with n - k + 1 colors. (Recall that Xi has a color distinct from colors
of all other vertices.)
9.1.31. a: Solution: Let us color G in colors 1,2, ... , X(G). Then we make the
following recoloring. Every vertex of color 1 which is not adjacent to the
vertices of color 2 we color in color 2. Every vertex of color 2 which is
not adjacent to the vertices of color 3 we color in color 3, and so on up to
color X - 1. This results in a minimal coloring such that every sequence of
vertices Vl, V2, •.. , Vx; of colors 1,2, ... , X is a simple path. Its length is X-I,
therefore l 2: X-I.
9.1.32. Hint: See Fig. A9.1.2.
9.1.33. G l and G3 .
9.1.34. Yes.
9.1.35. Hint: See Fig. A9.1.3.

9.1.36. If the graph G had a vertex v of degree degv :S k - 2 then for every
k-coloring of G it would have been possible to recolor the vertex v preserving
the properness of coloring, which is a contradiction.
9.1.37. Solution: Suppose that there exist color classes C,Cl such that G(CUCt)
has at least two components H1, H2. Swapping the colors of the vertices from
H l , we obtain a new k-coloring, contrary to its uniqueness for G.
9.1.38. Hint: Let C l , ... , Ck be the color classes of the k-coloring. Verify that
G(Gi U Gj) has at least IGil + IGjl- 1 edges (use the previous exercise).
Therefore the total number of edges in the graph is at least
E (IGil + IGjl- 1) = (k - l)n - Gi·

l~i<j~k
9.1.39. Hint: Let <p be a k-coloring of G * Kk (it exists, since G * Kk is a k-partite

graph). If for every vertex v E VG there exists a color which is used twice
among the vertices of kind (v, w), wE V Kk, then we may easily construct a
proper k-coloring of G by coloring v in this color, contrary to the assumption
that X(G) > k. Therefore there exists a vertex Va E VG such that all vertices
(va,w), w E VKk, are colored differently. If u E N(va) then <p(u,w) =1=
=
<p( Va, x) for w =1= x, since <p is a proper coloring. This implies that <p( u, x)
<p( Va, x). By the connectedness of G, the latter equality holds for every vertex
v. Therefore the considered k-coloring is induced by a single k-coloring of
Kk·
9.1.40. Yes.
9.1.41. Hint: A non complete graph with minimal coloring has a pair of vertices of
the same color.
9.1.42. Hint: Every k-coloring of the graph G - v is extendible to a (k+ i)-coloring

ofG.
9.1.43. The only 2-critical graph is J{2. The 3-critical graphs are the simple cycles
of odd length.
9.1.44. Hint: Use the fact that X(G) = maxx(B), where the maximum is over all
blocks B of G.
9.1.45. Hint: Use the fact that ifX(G) = k and degv < k-l then X(G) = X(G-v).
9.1.46. Solution: The inequality X(G - I) ::; X(G) - 1 holds, since G is critical.
The converse inequality is easily proved by contradiction.
9.1.47. Yes. Solution: If a k-chromatic graph G is not k-critical then there exists
=
a vertex Vl E VG such that X(G-vt) X(G). The graph G 1 G-Vl either =
is k-critical or has a vertex V2 such that X(G 1 - V2) = X(G 1 ) = k. Repeating
this reasoning, we ultimately find a k-critical subgraph in G.
9.1.48. Hint: Use the equality X(G 1 + G 2 ) = X(Gt) + X(G 2 ).

9.1.49. Hint: a: Delete the vertices u, v from G, color the graph G - u - v in colors
1, ... , k - 1, finally color u, v in color k.
b: Color the graph G - u in colors 1, ... , k - 1, finally color u in color k.
9.1.50. a: Yes. b: No; Fig. A9.1.4 shows a 4-critical graph with X(G - e) = 4.
e
9.1.51. The simple cycles of odd length.
9.1.52. No. Hint: Consider the graph (I{l U J{2) + (I{l U J{2).
9.1.53. C 2n - 1 + C 2n - 1 .
9.1.54. Hint: Prove first that if there exists a proper k-coloring t.p of the graph G',
then it induces a (k - 1)-coloring ~ of G as follows. Suppose that t.p( w) = 1.
Then ~(v) = t.p(v) if t.p(v) i- 1, otherwise ~(v) = t.p(v'). But this contradicts
to the assumption, therefore x( G') > k.
Further, prove that X(G' -e) S k for any edge e E EG'. To this end, consider
three cases: e E EG; e =
xy', x E VG; e wy'. =
9.2 Chromatic Polynomial

9.2.1. a: t 5 - 7t4 + 19t 3 - 23t 2 + lOt. b: t 5 - 7t4 + 18t 3 - 20t 2 + 8t.
9.2.2. a: A vertex vl may be colored in any of t colors, V2 may be colored in any
of the remaining t - 1 colors, etc., Vn may be colored in any of the remaining
t - n + 1 colors.
b: Every vertex may be independently colored in any of t colors.
9.2.3. a: (t - l)n + (-1)n(t - 1).

Solution: Clearly, f( Cn , t) is equal to the number of proper t-colorings of
the path graph Pn such that the terminal vertices are colored differently.
Let f'(Pn , t) denote the number of colorings of Pn such that the terminal
vertices are colored identically. It is easily seen that f'(Pn , t) = f(Cn- l , t).
Further,
f(C n , t) = f(Pn , t) - f'(Pn , t) = t(t - 1)n-l - f(C n- l , t).
Using this recurrence relation, the required formula is easily proved by in-
duction.
b: t(t - 2)n-l +t(-I)n-l(t - 2).
Hint: f(Wn,t)=tf(C n ,t-1).
9.2.4. a: No. Solution: By Theorem 9.2.3 this graph must have 4 vertices, 3
edges and 2 connected components. The only such graph is [{3 U [{ l, but its
chromatic polynomial is given in the item b.
b: Yes.
9.2.5. Hint: Use the fact that every component is colored independently of other
components.
9.2.6.
f(G,t) = (f(G l ,t){(G2 ,t) , t ~ n.

tt-1) ... t-n+l)
Hint: Use the Exercise 9.2.2a.
9.2.7. Solution: The number of proper t-colorings of G under which the colors of
the vertices u, v are different does not change if we add the edge uv to G.
Hence this number is f( G l , t). Similarly, the number of proper t-colorings
of G under which u, v are of the same color is f(G 2 , t). Therefore f(G, t) =
f(G l , t) + f(G 2 , t).
9.2.8. Solution: Necessity is proved by induction over n. For n = 1,2 the state-
ment is evident. Assume that the chromatic polynomial of a tree of order
n -1 is t{t _1)n-2. Consider a tree T of order n and any its pendant vertex
v. By the induction hypothesis, f(T - v, t) = t(t _1)n-2. The vertex v may
9.3. Edge Coloring 305
be colored in n - 1 ways, since it is adj acent to a single vertex of T. Therefore
f(T,t) = f(T - v,t)(n -1) = t(t - I t - 1 .
Sufficiency: Assume that f(T,t) = t(t - l)n-l. By Theorem 9.2.3, G is
connected (since the coefficient at t is nonzero). The coefficient at t n - 1 is
C~_l = n - 1. Therefore by Theorem 9.2.3 IEGI = n - 1. Thus G is a tree.
9.2.9. Hint: Use the following evident statement: every skeleton has no less color-
ings with t colors than the graph itself, and apply the previous exercise.
9.2.10. Solution: Let P be a partition of the vertex set of a graph G into p inde-
pendent sets; p ~ n = IGI. Under this partition, the number of colorings with
t colors is t(t -1) ... (t -p+ 1). Therefore f(G, t) = L:pt(t -1) ... (t- p+ 1),
where the summation is over all partitioning of VG into independent sets.
Since all summands are positive, it follows that for t > P - 1 we have
f(G, t) > 0, i.e., no t larger than p - 1 is a root of f(G, t). It remains
to notice that if t > n - 1 then t > p - 1.
9.3 Edge Coloring

9.3.1. 8 for the first graph, see Fig. A9.3.1; 2 for the second graph.
9.3.2. a: 2, if n is even and 3 otherwise.

b: n - 1, if n is even and n otherwise; n 2: 3.
Solution: For n = 2t, a possible partition Eo, ... , E n - 2 of the edges of [{n
into n - 1 color classes is as follows. If V Kn = {vo, ... , vn-d,
Ei = {vivn-d U {vi-ivi+i : j = 1,2, ... , t - I},

where the indices i - j and i + j are calculated modulo n.

The case of odd n is easily reducible to the above.
9.3.4. No.
9.3.5. Disjoint union of simple cycles of even lengths and simple paths.
9.3.6. a: No. b: Yes.
9.3.7. Hint: Start the coloring from the edges incident to a vertex of maximal
degree.
9.3.8. Hint: Use the fact that every bipartite graph has a matching that covers all
its vertices of maximal degree.
9.3.9. Hint: First analyze the structure of the graph G. If all its cycles are even or
there are no cycles then the graph is bipartite and the previous exercise is
applicable. If all cycles are odd then it is easily seen that no two cycles have
a common edge. Since G is connected and distinct from C2n+1, it follows
that ~(G) ~ 3. Using all the above, it is easy to construct an algorithm of
edge coloring of G in ~(G) colors.
9.3.10. Hint: If X'(G) -=f:. 3 then by the Vizing theorem, X'(G) = 4. Consider the
subgraph H of G induced by the edges of colors 1 and 2. Since G is a cubic
graph and X'(G) = 4, H is a spanning subgraph. If H were not connected,
swapping the colors 1, 2 in one of its components would produce a new
coloring of G, contrary to the uniqueness of the coloring of G. Therefore H
is connected, i.e., its edges constitute a Hamiltonian path or cycle in G. A
similar reasoning is valid for colors 3 and 4. Hence G is a cubic graph with
two edge-disjoint Hamiltonian paths. However such graph is unique (K4)
and it is 3-edge-colorable. This contradiction implies that X'( G) = 3, and
G has an edge coloring with colors 1,2,3. Further, prove that Hamiltonian
cycles are constituted only of the edges colored in only two colors.
9.3.11. Solution: If the statement is false then by the Vizing theorem X'(G) =
~(G). By the assumption, n = WGI is odd, and hence every color class
contains at most (n - 1)/2 edges. Therefore the total number of edges is at
most d(n - 1)/2. On the other hand, G is regular and by the Handshake
Lemma it has dn/2 edges, a contradiction.
9.3.12. a: The complete graphs of odd order n ~ 3.
b: The nonempty bipartite graphs with maximal vertex degree 2.
9.3.13. Petersen graph.
9.3.14. K 3 •
9.3.15. a: The graph shown at Fig. A9.3.2. b: Hint: Use the Exercise 9.3.2b.
9.3.16. a: No. Hint: Consider the graph K 5 • b: For odd n ~ 3.
c: No. Hint: Consider the graph K 5 •
9.4. Colorings of Planar Graphs 307
9.4 Colorings of Planar Graphs

9.4.1. 3.
9.4.2. A plane embedding of K 4 .
9.4.3. 4. Hint: Every such arrangement is naturally associated with a plane graph.
9.4.4. Hint: Use the induction over the number of circles. When a new circle is
overlaid onto a bicolored map, one must recolor the faces inside of this circle.
9.4.5. Yes.
9.4.6. Hint: Consider the geometric dual graph.
9.4.7. Hint: Use the fact that 8(G) ~ 2 for every outerplanar graph G and apply
the induction.
9.4.8. Solution: Let us choose the coordinate system in which the abscissae of
the intersection points (Xl, Yl), ... , (Xn, Yn) are pairwise different. One may
assume that Xl < ... < X n . We shall color the vertices in the indicated
order. Each next vertex (Xi,Yi) is adjacent to at most two previous vertices,
since exactly two lines pass through every intersection point. Therefore the
coloring requires at most three colors.
9.4.9. Solution: Use the induction over n = IGI. Consider a plane embedding of
O. For n ~ 5 there is nothing to prove. Assume that the statement is true
for graphs with less than n vertices.
If a has a vertex v with degree at most 3 then a proper 4-coloring of a may
be obtained from a proper 4-coloring of a-v. Therefore we may assume
that deg v 2:: 4 for all v E va. Assume that deg Vo = 4. The set N( vo) =
{ Vl, V2, V3, V4} has two nonadjacent vertices, say, Vl and V2, for otherwise G
would contain K5 and hence would be nonplanar. The graph 0 1 obtained
from 0 - Vo by merging the vertices Vl, V2 into ii is a plane graph, and by
the induction hypothesis it is 4-colorable. Now it is easy to produce a proper
coloring of a : Vl, V2 are colored in the color of v, all vertices distinct from Vo
inherit their colors from G I , finally, Vo is colored in a color unused in N(vo).
9.4.10. Hint: Prove that the faces inside (outside) the Hamiltonian cycle may be
colored in two colors, and use the colors 1,2 (respectively, 3,4) for the faces
inside (respectively, outside) the Hamiltonian cycle.
9.4.11. Hint: Use the inequality X(G) ::; .:l(G) + 1.
9.4.12. Hint: The contraction of a bridge does not influence on the adjacency of
faces. The vertices of degrees greater than 3 may be transformed as shown
at Fig. A9.4.1.
9.4.13. Hint: Let the faces be colored in colors 1, 2, 3, 4. Every edge borders on
two faces. An edge e is said to be of type ij if it borders on faces colored in
colors i and j. Prove that the following coloring a of the edges is proper:
aCe) = 1, if e is an edge of type 12 or 34;
aCe) = 2, if e is an edge of type 13 or 24;
aCe) = 3, if e is an edge of type 14 or 23.
Conversely, let there exist a proper coloring of the edges in colors 1, 2, 3.
Prove that the maps G 12 and G23 produced respectively by the edges of colors
1,2 and 2,3 are bicolored. Using their bicolorings, construct the 4-coloring
ofG.
9.4.14. Solution: Let C1, C 2 , C3 , C4 be the color classes of a 4-coloring of a planar
graph G, ICd = ni. By Exercise 9.1.37, all induced subgraphs G(Ci U Cj),
i :j:. j, are connected. Therefore the number of edges of G is at least
L (ni + nj - 1) = 3(n1 + n2 + n3 + n4) - 6 = 3n - 6.

19<j<4
By the corollary 6.1.5, G is a maximal planar graph.
9.4.15. Hint: Notice that a trichromatic graph always has a cycle.
9.4.16. Hint: For n = 5 the Hadwiger conjecture states that a 5-chromatic graph
is contractible to K 5 • But this means that the graph is nonplanar.
9.4.17. Yes.
9.4.18. Hint: Use the stereographic projection.
9.4.19. Hint: Fig. A9.4.2 shows a map on the torus that requires at least 7 colors.
t t
)
9.5 Perfect Graphs

9.5.1. No.
9.5.3. Hint: Use the equalities X(G + H) = X(G) + X(H) (Exercise 9.1.22) and
<peG + H) = <p(G) + <p(H).
9.5.4. Hint: Prove that X(G U H) = max{x(G), X(H)} and

<peG U H) = max{ <p(G), <p(H)}.
9.5.6. Hint: Use the Exercise 9.5.5a and the Corollary 9.5.2.
9.5.7. Bipartite graphs.
9.5.8. Solution: a: Let us prove that X(G) ::; ep(G) (and hence X(G) = ep(G)).
Let ~(x) denote the length of the longest path x = Xl ::; x2 ::; ... ::; xe
starting at a vertex x. Clearly, 1 ::; ~ Cx) ::; <pC G), because the vertices of
such path induce a clique in G. Let us prove that ~ is a proper coloring of
G. Let the vertices x 2:: y be adjacent in G. Then the longest path from x is
surely shorter than the longest path from y. Hence ~(y) "f:. ~(x). Therefore
xCG) ::; maxxEP xCx) ::; ep(G), and hence X(G) = <p(G).
b: Yes.
9.5.9. Solution: Let Hi denote a maximal complete subgraph of G i , i = 1,2. Then

V Hl x \l H2 induces a complete subgraph of G l · G 2 . Therefore <p(Gl . G 2 ) 2::
<pC Gt}<p( G2 ).
Further, let ~i be a minimal coloring ofGi, i = 1,2. Let us define the coloring
~ of G 1 . G 2 : ~(Xl' X2) = (~(xt), ~(X2». It is easily seen that ~ is a proper
coloring of G 1 . G 2 using X(Gt}X(G 2 ) colors. Therefore
The converse inequality X(G 1 . G 2 ) ~ <p(G 1 . G 2 ) is evident.

b: No. Hint: Consider the graph from Exercise 9.5.1.
9.5.10. Hint: Necessity: A possible required independent set I is a color class in

some <p( H)-coloring.
Sufficiency may be proved by induction over the number of vertices. It is
easily seen that
X(H) :::; X(H - 1) + 1, <p(H) ~ <p(H - 1) + 1.
By the induction hypothesis, <p(H - 1) = X(H - I). Therefore X(H) :::; <p(H),
and the converse inequality is evident.
9.5.11. b: Hint: Use the fact that every interval graph is triangulated; see Exercise
9.5.20.
9.5.12. H.
9.5.13. Yes.
9.5.14. a:No. b: No.
9.5.15. If it is a forest.
9.5.17. No.
9.5.19. Hint: Consider two nonadjacent vertices u, v and a minimal set S of ver-
tices separating them. Prove by induction over the number of vertices that
the components G u and Gv of G\S that contain u and v respectively have
simplicial vertices. Further, apply the Theorem 9.5.4.
9.5.20. Hint: Fig. A9.5.1 shows a triangulated graph which is not an interval
graph.
9.5.21. Solution: Necessity: Let H be an induced subgraph of a triangulated

graph G. Let us construct a perfect dismantling of H, i.e., a sequence of
graphs
where n' = IHI and every vertex Vi is simplicial in Hi, i = 1,2, ... , n' - 1.
Then s = maXi degH, Vi :::; <p(H) - 1. Let H' be an induced subgraph of H.
Then
where U E V H' is a vertex of maximal degree in H' and Vj is the first

one in the dismantling sequence VI, .•• Vn'-I that belongs to H'. We obtain
w(H) ::; s ::; 'P(H) - 1. On the other hand, the inequality w(H) ~ 'P(H) - 1
is evident.
Sufficiency: Suppose that G is not triangulated, i.e., it contains an induced
simple cycle Ck, k ~ 4. Then 8(Ck) = =
2 'P(Ck), contrary to the assump-
tions of the exercise.
9.5.22. a: Hint: Proof is by induction over k. For k = 2 the statement is trivial.
The inductive hypothesis is n~~ll VIi = Uk-1 i= 0. Suppose that VTk n
Uk-1 = 0. Let us prove that there exists a vertex v E VTk (and it is unique)
such that every path connecting any pair x E Uk-1, Y E VTk passes through
v. Let us take u E Uk -1. Then every tree Ii contains the vertex u and some
vertex of Tk, and hence the simple path P connecting these vertices. The
vertex v belongs to P, therefore v E VIi, i = 1,2, ... , k, which contradicts
to the supposition VTk n Uk-1 = 0, v rt. Uk-1.
9.5.23. Solution: Suppose that the comparability graph G is not triangulated,
i.e., G contains an induced simple cycle Ck = (V1,V2, ... ,Vk,V!), k ~ 4.
Consider three consecutive vertices Vi, Vi+1, Vi+2 of this cycle. If Vi ::; Vi+1
and Vi+l ::; Vi+2 then Vi ::; Vi+2, and hence G has an edge ViVi+2, which
is impossible, since Ck is an induced subgraph. The case Vi ~ Vi+1 and
Vi+! ~ Vi+2 is excluded in a similar way. Therefore either Vi ::; Vi+1 and
Vi+! ~ Vi+2 or Vi ~ Vi+1 and Vi+l ::; Vi+2, i.e. the relation "::;" on the vertex
set of Ck alternates along the cycle, hence the length of Ck must be even.
9.5.24. Hint: Use the Theorem 9.5.5 and the hereditarity of the property "to be a
triangulated graph" .
9.5.25. Hint: Use the hereditarity of the property "to be a triangulated graph" .
9.5.26. Hint: Use a normal dismantling of the graph.
9.5.27. Hint: Consider the graph 2C4 .
9.5.28. Hint: Perform the proof by contradiction considering a minimal non-
perfect quasitriangulated graph and applying the Corollary 9.5.2.
Directed Graphs
10.1 Directed Graphs: Basic Notions

10.1.1. n 2 .
10.1.3. 2n2.
10.1.7. a,b: Empty digraphs. c,d: Complete digraphs.

10.1.8. No. Hint: See Fig. A10.1.1
10.1.9. The isomorphic digraphs are shown at Fig. 10.1.3a,b.

10.1.10. Hint: There are 16 such digraphs.
10.1.11. a: IV L(G)I = IAGI, IAL(G)I = LVEVG d-(v)d+(v).
b: Hint: See the hint to the Exercise 1.1.21.
10.1.12. Hint: Consider, for example, the sequences (3,2,1) and (3,3,0).
313
Figure A1O.1.2: To Exercise 10.1.13
10.1.13. See Fig. AI0.1.2.
10.2 Reachability and Components

10.2.6. No. Hint: See Fig. AI0.2.1
Figure AI0.2.1: To Exercise 10.2.6
10.2.9. Hint: To prove the sufficiency, extract a walk (semiwalk) connecting the
required vertices from the existing spanning walk (semiwalk).
10.2.10. Solution: Necessity. Suppose that a digraph G is strong. Consider a

partition of the vertex set of the digraph into two disjoint sets Vi, V2 and
take Vi E Vi, V2 E V2 • Since the digraph is assumed to be strong, Vi, V2 are
reachable from each other. Hence there exists an arc outgoing from Vi and
ingoing into V2 and an arc outgoing from V2 and ingoing into Vi. Hence G
is irreducible.
Sufficiency. Suppose that a digraph G is irreducible. Consider a vertex u E
VG and the set V(u) of vertices reachable from u. Suppose that V(u) :f:. VG.
Then the irreducibility of G implies that there exists an arc (v, w) such that
v E V(u) and w ¢ V(u). This contradicts to the definition of V(u). Hence
V(u) = VG. This means that every vertex is reachable from every other
vertex, or G is strong.
10.2.11. a: Hint: Use the hint for the Exercise 10.2.9. b: Hint: Use the Exercise
10.2.10. c: Hint: Use the Exercise 10.2.8.
10.2.15. Hint: a: To prove the inequality m ~ (n -1)2, consider a digraph with

the vertex set {VI, ... , vn } and the arc set
((Vi,Vj), i,j=1,2, ... ,n-l, i#j}U{(Vi,Vn ), i=1,2, ... ,n-l}.
Prove that G has the maximal number of arcs among connected but not
strongly connected digraphs of order n.
c: To prove the inequality m ~ n(n-l)/2, consider a digraph with the vertex
set {VI, ... , vn } such that (Vi, Vj) E AG if and only if i < j. Prove that G
has the maximal number of arcs among connected contourless digraphs of
order n.
10.2.16. Hint: Consider a vertex from which one can reach a maximal number of
vertices of the digraph and prove that this vertex is a source.
10.2.17. Solution: The previous exercise implies that a unilateral digraph G has a
vertex u from which all other vertices are reachable. The opposite digraph
Gis also unilateral, and it has a vertex V with similar properties. In G, the
vertex V is reachable from all vertices. Therefore G + (v, u) is strong.
10.2.18. Hint: Consider the following numbering procedure. A vertex of the
digraph G with zero in degree obtains the number 1. If the vertices from
S ~ VG have already obtained the numbers 1,2, ... , k, a vertex of the di-
graph G - S with zero in degree obtains the number k + 1. Proceed in this
way until all vertices are numbered.
The existence of vertices with zero indegree in all considered digraphs follows
from the Exercise 10.2.4.
10.2.19. See Fig. A10.2.2 .
•
a b,c d
10.2.20. a::} d: Hint: Use the fact that every vertex of a contourless digraph con-
stitutes a strong component. Perform the topological sorting of the digraph
(Exercise 10.2.18).
10.2.23. a: (n - k + 1)2 + (k -1). Solution: Let us demonstrate that a digraph G

containing a maximal number of edges has k - 1 single-vertex components.
Suppose the contrary; let G have components GI and G 2 , IGII = nl, IG21 =
n2, n1 ~ n2 > 1. Clearly, GI and G2 are complete digraphs with IEG1 1 +
IEG 2 1= ni + n~. Consider a digraph G in which the components G I and
G2 are replaced by complete graphs G1 and G2 with IG1 1= ni + 1 and
IG2 1= n2 - 1. One may easily verify that this increases the number of edges,
contrary to the choice of the digraph G.
b: n(n - k + 1) + k(k - 1)/2. Hint: Argue similarly to the item a, in terms
of strong components.
10.2.24. b: Hint: To prove the sufficiency, consider a maximal under inclusion
direct able subgraph G of a bridgeless graph G. If VG =J va
then prove the
existence of a simple (u, v)-path L such that u, v E VG, and the remaining
vertices of L do not belong to G. Prove that the subgraph GUL is directable,
contrary to the choice of G.
10.2.25. Hint: Consider some numbering of the vertices of the graph and direct
every edge in such a way that the number of the tail of the resulting arc is
smaller than the number of the head.
10.2.26. Hint: Use the Exercise 10.2.24b.
10.2.29. Hint: Prove that a vertex U1 of a quasistrong digraph G from which a
maximal number k of vertices Vt, ... ,Vk is reachable is a source.
10.2.30. Solution: a) implies b): If G is a rooted tree then G is connected; oth-
erwise G has either more than one vertex with zero indegree or a contour.
Consider a semiwalk in G connecting the root with some other vertex. Since
d-(u) = 1 for all vertices distinct from the root, it follows that this semiwalk
is a walk, and moreover, this walk is unique.
b) implies c): Let v denote a vertex which is connected with every ver-
tex of the digraph by a single walk. Suppose that d- (v) > 0 and con-
sider (u,v) E AG. Then the existence of the walk L = (V,V1, ... ,VI,U)
would imply the existence of other walks connecting v and u, e.g., L1 =
(v, VI,"" v" U, v, VI,.'" VI, u). Therefore d-(v) = O. Similarly we may ob-
serve that d- (u) = 1 for all vertices u other than v.
c) implies a): It is necessary to prove that G has no contours. Suppose the
contrary, i.e., G has a contour. If a vertex v belongs to it, then d-(v) > O.
If v does not belong to the contour then the existence of walks connecting v
to the vertices of the contour implies that d- (u) > 1 for some vertex of the
contour. In both cases we have a contradiction.
a) and b) imply d): Clearly, a rooted tree is a quasistrong digraph. Sup-
pose that it remains quasistrong after the deletion of an arc (u, w). The
quasistrongness implies that there exists a vertex z from which u, ware
reachable. Hence the original digraph has two walks from z (and hence from
the root) to w, which contradicts to item b).
10.2. ReachabiJity and Components 317
d) implies a): Since G is quasistrong, by Exercise 10.2.29, it has a source s.

Therefore d-(v) > 0 for all vertices v 1= s. Suppose that there exists a vertex
v with d-(v) > 1, i.e., the digraph has two different arcs (u, v), (w, v). After
the deletion of one of them the' digraph remains quasistrong, contrary to the
assumption. Therefore d-(v) = 1. Similarly one may prove that d-(s) = O.
10.2.33. Hint: The sufficiency follows from Exercise 1O.2.30d.

10.2.34. a: nn-2. Solution: A bijection exists between the sets of rooted edge-
labelled trees of order n and labelled trees, and the required number follows
from the Kelly theorem, Sect. 2.2. A possible bijection is as follows. The
root obtains the label n and every remaining vertex obtains the label of the
arc on the shortest path from the root to it.
b: nn-3. Hint: Use the item a.
10.2.35. a: See Fig. A10.2.3. b: See Fig. AI0.2.4.
Figure AI0.2.3: To Exercise 10.2.35a
Figure AI0.2.4: To Exercise 10.2.35b
c: Hint: Suppose the contrary and consider two transitive reductions Rl, R2
of a contour less digraph G. Suppose that (u, v) EARl, (u, v) f/. AR2 • Since
(u, v) E AG, there exists a path L = (u, Wl, ... , Wk, v) in R2. By the defi-
nition of transitive reduction, (u, v) and all arcs of L cannot simultaneously
belong to R 1 . Assume that (WI, wl+d fj. R 1 . Since (WI, wl+d E AG, there
exists a path L = (WI, w~, ... , WI+l) in R 1 . Since G is contourless, L2 does
not pass through the vertices of L1 other than WI and WI+1.
Just the same, some arc (w~, W~+l) E L2 cannot belong to R 2, and its ends
may be connected by a (w~, w~+1)-path with all arcs from R2, and so on.
But since the digraph has a finite number of arcs, at some step it will become
impossible to select a new path.
10.2.39. a: See Fig. A10.2.5.
Q Q
b: The digraph Gf is a disjoint union of contours.

c: Digraphs G It, G h are isomorphic if and only if the permutations 1I,!2
have the same numbers of cycles of the same length (including the cycles of
length 1).
10.3 Matrices Associated with Digraph

0 1 0 0 0 0 1 1 1 1 1 0
0 0 0 1 1 0 1 1 1 1 1 0
1 0 0 0 0 0 1 1 1 1 1 0
10.3.1. A(G) = . R(G) =
1 0 1 0 1 0 1 1 1 1 1 0
0 0 0 0 0 0 0 0 0 0 1 0
0 0 0 0 0 0 0 0 0 0 0 1
1 0 -1 -1 0 0 0
-1 1 0 0 1 0 0
0 0 1 0 0 -1 0
I(G) =
0 0 0 1 -1 1 1
0 -1 0 0 0 0 -1
0 0 0 0 0 0 0
10.3. Matrices Associated with Digraph 319
10.3.3. The matrices A(G) and R(G) are the transposed A(G) and R(G) respec-
tively, and I( G) is obtained from I( G) by changing the signs of all entries.
10.3.4. Hint: See the hints to Exercise 1.5.4.
10.3.5. Hint: A graph and its transitive closure have the same reachability matrix.
10.3.7. Hint: Use the topological sorting (see Exercise 10.2.18) of the vertices of
the condensation of the digraph.
10.3.8. Hint: Use the topological sorting (see Exercise 10.2.18) of the vertices of
the digraph.
10.3.9. a: Hint: The sum of all rows of the matrix I(G) is equal to the zero row,
therefore rank I( G) :::; n - 1. Since the digraph G is connected, the deletion
of any k < n rows from I( G) results in a matrix which has a column with
exactly one nonzero element. The rows of such matrix cannot be linearly
dependent.
b:Hint: Use the item a.
10.3.10. Solution: Proof is by induction over the dimensions ofthe minor. For the
dimensions 1 x 1 the statement is evident. Suppose that it is valid for the
dimensions (n - 1) x (n - 1). Consider a submatrix A of dimensions n x n.
If A has a zero column then det A = O. If A has a column with exactly one
nonzero element then we decompose its determinant over the elements of
this column and obtain det A = ±A' , where A' is the cofactor of the nonzero
element of the column, which is either 0 or ±1, by the induction hypothesis.
If every column of A has exactly two nonzero elements then its rows are
linearly dependent and det A = O.
10.3.11. Hint: Use induction over the length of the walk.
[01001] [0 1 1 1
~]
00110 1 1 1 1
10.3.12. a: A(1) = o 1 0 0 0 ; A (2) = 0 1 1 1
1 000 1 1 1 0 0
o 000 1 0 0 0 0
[i 1 1]
1 1
1 1 1 1
A(3) = A(4) = 1 1 1 1 .
1 1 1 1
0 0 o 1
c: Solution: Consider the proof for the case k =2. Suppose that A( G) =
A (2)( G) for a digraph G but G is not transitive. This means that there exists
a path (UI, ... , up), P ~ 3 such that (UI' up) fj. AG. Let L = (UI' U2, U3, ... )
be the shortest among the (UI, up)-paths. Since A(G) = A2(G), it follows
that (UI, U3) E AG, and the path (UI, U3, ... ) obtained from L is shorter
than L, contrary to the selection of L.
For k > 2 the proof may be performed in a similar way.

10.3.13. Hint: Notice that the reachability matrices of all these graphs coincide.
10.3.14. Hint: Use the fact that a term of the sum R~;) = I:~=1 RiI.Rki is equal
to 1 if and only if the vertices i and k are in the same strong component.
10.3.15. Hint: Necessity: Perform the topological sorting of the vertices of the
rooted tree G (Exercise 10.2.18) and show that the matrix M1(G) under the
new numeration of G is obtained from the original matrix M (G) by the same
permutations of rows and columns. Next, prove that the determinant of the
matrix obtained by the deletion of the first row and the first column from
M1(G) is equal to 1.
Sufficiency: From the hypothesis of the exercise it follows that d-(r) = 0
and d-(i) = 1 for i # r. Suppose that the digraph has a contour. Determine
the structure of the connected component T of the digraph containing this
contour. Prove that det M(F) # 1 for the digraph F induced by the vertices
of the contour. Then using induction over the number of vertices prove that
detM(T) # 1.
10.4 Tours and Paths

10.4.2. Hint: This digraph cannot be connected.
10.4.4. Hint: To prove the implication 1 => 2, notice that a cyclic Eulerian chain
ingoes into every vertex and outgoes from it the same number of times.
To prove 2 => 3, select some contour L in the digraph, notice that claim 2
holds for the connected components of the digraph G\AL, and use induction
over the number of contours.
To prove 3 => 1, use the fact that the union of two Eulerian digraphs with a
single common vertex is an Eulerian digraph.
1004.5. Hint: Add a path L = (V1' W, V2) to the digraph G, where W is a new
vertex, and use the Theorem 1004.1.
1004.6. Hint: Add k paths L1 = (u, Wl, v), ... , Lk = (u, Wk, v) to the digraph G,
where Wi is new vertices, and use the Theorem 1004.1.
1004.7. Hint: Turn the graph into an Eulerian multigraph by augmenting it with
edges connecting pairs of odd-degree vertices. Further, direct the edges ac-
cording to the traversal of an Eulerian cycle of the multigraph, and then
delete the arcs corresponding to the added edges.
1004.8. Hint: Consider a digraph shown at Fig. AlOA.I.
1004.9. Hint: Use the Theorem 1004.2.

10.4.10. Hint: Add a new vertex v to the digraph and connect it by two arcs
(v, u) and (u, v) with every vertex u of the digraph. Prove that the resulting
digraph has a Hamiltonian contour.
10.4.12. Hint: a: Prove that no(G) = 5 and use the Theorem 10.4.3.
b: Prove that x( G) = 4 and use the Theorem 10.4.4.
10.4.13. Hint: To prove the inequality l(G) 2: na(G), notice that every path con-
tains at most one vertex from an independent set of vertices of a digraph.
The opposite inequality follows from the Theorem 10.4.3.
10.4.14. Solution: Consider a coloring g : VG -+ {I, 2, ... , X(G)} of the graph G.

Direct every edge in such a way that the tail of the resulting arc has color
of smaller number than the head. The resulting digraph has no paths longer
than X(G). Therefore r(G) ~ x(G) - 1. The opposite inequality r(G) + 12:
X( G) follows from Theorem 10.4.4.
10.4.15. Hint: Consider a k-coloring of the arcs of a digraph G corresponding

to a minimal k-coloring of vertices of the digraph L( G). Prove that the
substrate of the directed subgraph of G induced by the arcs of the same color
is a bipartite graph. and the substrate G s of G is the union of k spanning
bipartite graphs: G s =L~=l Gk. The vertices of every graph G; may be
colored in two colors 0, 1. Therefore every vertex v of G may be associated
with a sequence (al," .ak), where ai is the color of v in G;. Clearly, these
(0, I)-sequences produce a proper coloring of G; therefore X(G) ~ 2k, or
X(L(G» = k 2: log2X(G).
10.4.16. Solution: a: Suppose that kl = miIlveVG d+(v). Then at least kl arcs

outgoes from a vertex v. Taking any of them, say, (v, Vl)' we have a path of
=
length 1. Assume that we have already constructed a path L (v, Vl , ... , VI)
of length I < k l . Then there are at least kl - I arcs outgoing from VI that
are not incident to the vertices v, ... , VI-l. Anyone of them may be used to
augment the path L. This procedure allows to construct a path oflength k l .
Since at the first step we have at least kr arcs to select from, at the second
step we have kl - 1 possibilities, etc., there are at least kl! possible paths of
length k 1 .
b: The solution is similar to that for item a.
= =
c: Suppose that max{ k l , k 2 } kl k. Item a implies that G has a path
L = (V,VI, ... ,Vk) of length k. If G has an arc (vk,v), then we have the
required contour. Otherwise L may be augmented by an arc (Vk, Vk+d, where
Vk+1 ~ {VI, ... , Vk}' If G has an arc (Vk+I, v) or (Vk+l, VI) then we have the
required contour. Otherwise the path may be augmented still further, and
so on. Clearly, at some moment the path cannot be extended further, and
the addition of a new arc produces a contour of length at least k + l.
10.4.17. Solution: Sufficiency: Construct a bipartite graph C = (V, V'; E) for a
digraph G in the following way. V, V' are two disjoint copies of the vertex
set of the digraph G, and uv' E E if and only if uv E AG, see Fig. A10.4.2.
Clearly, bo(G) is equal to the deficiency b(V, V', E) of the bipartite graph
C. Theorem 3.5.4 implies the existence of a perfect matching in C. The

matching aoa~, aIa~, a2a~ ... is associated with the collection of contours
and loops (ao,al,a2, ... ,akpao), (akl+l,akl+2, ... ,ak2,akl+t), ... in G. It
is easily seen that this collection is the required spanning subgraph.
Necessity may be proved using the same construction.
10.5 Tournaments
10.5.1. See Fig. AlO.5.l.
10.5.5. Hint: To prove the uniqueness, notice that a transitive tournament has no
contours.
10.5.6. Hint: Use the Corollary 10.5.2.
10.5. Tournaments 323
10.5.7. Solution: Let L = (vo, VI, ... , Vn -l, vo) be a Hamiltonian contour of a
tournament T. If T has an arc ofform (Vi, Vi+2), 0 :::; i :::; n - 1, (the addition
of indices is modulo n here) then Vi is the required vertex. Suppose now that
T has the arcs ofform (Vi +2 , Vi)' 0 :::; i :::; n-1. If n is odd then the tournament
T-vo has the path L1 = (V n -l, Vn -3, ... , V3, vt) from Vn -l to VI' If n is even
then the tournament T - Vo has the path L2 = (V n -l,Vn -3, ... ,V2,V3,vd
from Vn -l to VI. Therefore T - Vo is strong, and by Corollary 10.5.2 it is
Hamiltonian.
10.5.S. Hint: If the automorphism group of a tournament were of even order
then it would have a subgroup of second order, which is impossible, since a
tournament cannot have both (u, v) and (v, u) arcs.
10.5.9. a: Solution: Let u be a vertex of a tournament T with maximaloutdegree
and let v be an arbitrary vertex of T. If (u, v) E AT then p( u, v) = 1. If
(v, u) E AT then let us consider the vertices from fu. Suppose that there
exists an arc (w,v) E AT with w E fu. In this case p(u, v) = 2. 1fT has
the arc (v, w) for every vertex w E fu, then d+(v) > d+(u), contrary to the
assumption of the exercise.
-
b: Hint: Consider the opposite digraph T and use the item a.
10.5.10. Solution: a: The equality follows from the Theorem 10.1.1.
b: Consider the tournament Tk induced by the vertices corresponding to
the list (81, ... , 8k). If (81, ... , 8k) is the list of outdegrees of Tk, then the
equality from item a) of the exercise implies that L~=1 8i = k(k - 1)/2,
k = 1, ... , n - 1, which in turn implies the required inequality.
If T is a strong tournament then there exists an arc outgoing from a vertex
of the set Tk. Therefore L~=1 8i > k(k -1)/2.
Suppose now that L~=l 8i > k(k - 1)/2 holds for all k = 1, ... , n - 1, and
let us prove that T is strong. Suppose the contrary, i.e., T is not strong.
Let T 1, ... Tk be its strong components. Assume that T1 contains a vertex

VI with minimal outdegree. Consider two components T1 and Ii, i 2:: 2. The
arcs connecting them must either outgo from every vertex of T1 and ingo into
every vertex of Ii or vice versa. Suppose that T has an arc (VI, vp), where
vp E Ii. Should there exist an arc (u, VI) with u E rvp then VI, vp would be
in the same strong component. Therefore (VI, u) E AT for all u E rv p , see
A10.5.2 But then Sl > sp, contrary to the choice ofTl .
Figure AlO.5.2: To Exercise 10.5.10
Thus it is proved that for every vertex V of T1 there are arcs outgoing from
all vertices of the remaining components and ingoing into v. Therefore if
St is the maximal outdegree for the vertices from T1 then all vertices with
outdegrees at most St belong to T 1 . But in this case L:~::1 Si = t(t - 1)/2,
contrary to the assumptions of the exercise.
c: To prove the left-hand inequalities, we shall use item h.
k
k(k-1)/2 ~ 2:Si ~ kSk, k = 1, ... ,n-l.
;::1
Hence Sk 2:: (k - 1)/2.

To prove the right-hand inequalities, consider the opposite tournament T
and its list of outdegrees (Sl, ... , sn) arranged in the non decreasing order.
Since the outdegree of a vertex of a digraph is equal to the indegree of the
vertex in the opposite digraph, we have Sk + Sn+1-k = n - 1. The fact
Sk 2:: (k -1)/2 implies that Sn+1-k 2:: ((n + 1- k) -1)/2 = (n - k)/2. Hence
Sk ~ (n + k - 2)/2.
10.5. Tournaments 325
d: Let T be a transitive tournament. Let us perform topological sorting of
the vertices of T, see Exercise 10.2.18. Let 1,2, ... n be the numbers assigned
to the vertices by this sorting. Then 81 = =
d+(n) 0, 82 = =
d+(n - 1) 1,
... , 8k = d+(n + 1 - k) = k - 1, k = 1, ... , n.
Suppose now that 8k = k -1 holds for all k = 1, ... , n. Since 8 n = n -1, the
arcs outgoing from the vertex corresponding to the number 8 n ingo into all
remaining vertices of the tournament. Since 8 n -1 = n - 2, all arcs outgoing
from the vertex corresponding to the number 8 n -l ingo into all remaining
vertices of the tournament except of V n . Arguing in a similar way for all
vertices, we conclude that the tournament is transitive.
10.5.11. Hint: Since d+(v) + d-(v) = n - 1 for every vertex v of a tournament, it

is sufficient to prove the equality
vEVT vEVT
which may be deduced from
L d+(v) = n(n - 1)/2.

vEVT
10.5.12. Solution: Proof is by induction over the number of vertices of a tourna-

ment. For a tournament with two vertices the statement is evident. Suppose
that it is valid for tournaments T of order n, i.e., IATI ~ In/2JL(n + 1)/2J.
Consider a tournament Tl of order n+ 1. Since EVEVT1 d+(v) = n(n+ 1)/2,
there exists a vertex Va with the out degree at least Ln/2J = L(n + 1)/2J for
even n or at least L(n+ 1)/2J for odd n. Let T2 denote a maximal contourless
subgraph of Tl - Va. If we add the arcs outgoing from Va into T2, we obtain
a contourless digraph. By the induction hypothesis, we deduce
IATd ~ IAT21+d+(va) ~ Ln/2J l(n+1)/2J+l(n+ 1)/2J = l(n+1)/2J L(n+2)/2J.
10.5.13. No. Hint: See Fig. AI0.5.3.

10.6 Base and Kernel

10.6.2. Hint: Notice that no one of these vertices is reachable from any other one.
10.6.3. Hint: Use the fact that all vertices of a strong component are mutually
reachable and every vertex of a non-base component is reachable from every
vertex of some base component.

10.6.5. All vertices with zero indegree. Hint: Use the Exercise 10.6.3.
10.6.7. Solution: Suppose that v is a vertex from which a maximal number of

vertices is reachable and v does not belong to a base B. Then there exists
a vertex u E B such that v is reachable from u. Since all vertices reachable
from v are also reachable from u, u is the required vertex for B.
Figure AIO.6.I: To Exercise 10.6.13

10.6.11. Solution: Let A be an arc base of a digraph G. If (a:, y) E AG then
y E D(a:), and by the definition y E D(a:, A), i.e., the condition (1) holds. If
(a:, y) E A then the existence of a path L that consists of arcs from the set
A' = A\{(a:,y)} would imply D(u,A) =D(u, A') for every vertex u of the
digraph, contrary to the definition.
Suppose now that A satisfies both conditions (1), (2). The first one im-
plies that for every (a:, y)-path we may replace its arcs which are not in
A by paths that consist of arcs from A, i.e., D( a:) = D( a:, A) for every
a: E VG. Suppose that A' c A, and (a:, y) E A\A'. Condition (2) implies
that D(a:, A)\D(a:, A') :j:. 0. Therefore A is an arc base of G.
10.6.12. Hint: Consequently delete arcs from a digraph G according to the follow-
ing rule: an arc (a:, y) is deleted if after its deletion y remains reachable from
a:. Using the Exercise 10.6.11 prove that this procedure produces an arc base
ofG.
10.6.13. For a digraph shown at Fig. A10.6.1 the sets {a, b, c, d} and {a, c, e} are
the bases.
10.6.14. Solution: Suppose the contrary, i.e., a digraph G has two arc bases
U1 =/:. U2, and suppose that an arc (x, y) E U1 \U2. Since U2 is an arc base,
there exists a walk L = (x, Xl, ••• , x p , y) with all arcs in U2 . Since U1 is a base,
there exist walks with all arcs in U1 connecting x with Xl, Xl with X2, ••• , xp
with y. Some of these walks contain the arc (x, y), for otherwise there would
exist an (x,y)-walk with all arcs in U1 \{(x,y)}, which is impossible, see Ex-
ercise 10.6.11. Suppose that a walk Li = (Xi-1, Vb V2, .. . , x, y, U1,.·., Us, Xi)
contains the arc (x, y). Consider two cyclic walks:
At least one of them contains two different vertices (otherwise G would

contain parallel arc connecting x and y). One may select a contour from this
walk (see Exercise 10.2.1), contrary to the assumption of the exercise.
10.6.17. Hint: Consider a subgraph D of a strong digraph induced by the arcs of

an arc base. By Exercise 10.6.15 D is strong. Further, use the induction
over the number of contours whose union is D (see Exercise 10.2.11).
10.6.18. See Fig. A10.6.2.
10.6.19. Hint: Consider a digraph induced by the arcs of an arc base.

10.6.21. See Fig. A10.6.3.
10.6.23. See, e.g., Fig. A10.6.4, where the required vertex sets are marked.
10.6.25. Hint: Prove that the set N ~ VG is dominating if and only if VG\N ~
r- 1 N, and it is independent if and only if r- 1 N ~ VG\N.
@r--_....
a
10.6.26. See Fig. A10.6.5. Here the required inkernels are marked by "-" and the
out kernels are marked by "+". At Fig. A10.6.5e, {2, 4} and {I, 3, 5} are
kernels.
10.6.28. A complete loopless digraph.
10.6.29. Solution: Suppose that N ~ VC satisfies (a), (b). Then f N ~ VC\N.

=
Suppose that V' VC\(V+ U fV+) # 0 and N' = N\V+ is the outkernel
of the subgraph C', i.e., fN' = V'\N'. Then
fN = fN' U fV+ = (V'\N') U fV+ =

((VC\(V+ U fV+))\N') U fV+ = (VC\(N U fV+)) U fV+.
Since N n fV+ = 0, it follows that fN = VC\N, i.e., N is an outkernel.
Suppose now that N is an outkernel. Its definition implies conditions (a)
and (b). If the set N' is nonempty then for the set f' N' of vertices of C'
adjacent to N',
f'N' = fN\fV+ = (VC\N)\fV+ = V'\N'.

10.6.30. Hint: Replace f by f- 1 and V+ by V-, where V- is the set of vertices
of the digraph with zero outdegree.
10.6.31. Hint: Use the fact that the substrate of the digraph is a bipartite graph
and prove that its parts are the required in- or outkernels.
To demonstrate that the statements cease to be valid, consider the digraph
shown at Fig. AlO.6.6.
10.6.32. Solution: a: Let N I , N2 be different inkernels of C. From Exercises

10.6.25, 10.6.22 in follows that neither Nl C N2 nor N2 C N I .
a b + c ± d
Let us pick a vertex UI E NI \N2. Since N2 is an inkernel, there exists a vertex

VI E N2 such that UI E r-IVI and VI ~ N I , since NI is independent. There-
fore VI E N2 \NI . By the same reason there exists a vertex U2 E N2 \N I such
that VI E r-IU2, and so on. Since G is finite, the vertices UI,Vl,U2,V2, ...
cannot be all different, hence G contains a contour, contrary to the statement
of the exercise.
b: The solut.ion is similar to that for item a.
10.6.34. Hint: Use the Theorem 10.6.1 and the Exercise 10.6.24.
10.6.35. Hint: Select a maximal independent set in the graph and direct the edges
in an appropriate way.
10.6.36. a: See Fig. A10.6.7.

b: Hint: Select a maximal independent set S I ~ V G and set g( v) = 0 for
all v E ,'it. Then select a maximal independent set S2 ~ VG\SI and set
g( v) = 1 for all v E S2, and so on. Prove that this process produces a
Grundy function.
c: See Fig. AIO.6.S. d: See Fig. AIO.6.9. e: See Fig. AIO.6.10.
f: Hint: Prove that the vertices of a digraph for which a Grundy function is
equal to zero constitute a kernel of the digraph.
oe-----i..--___e 0
oe<------4.._-_eo
2
b
Figure AIO.6.7: To Exercise 10.6.36a
Figure AIO.6.8: To Exercise IO.6.36c
010
~
Figure AIO.6.9: To Exercise 10.6.36d
...
Figure AIO.6.10: To Exercise 1O.6.36e

g: Hint: Let 51 be a kernel of a digraph a.

Set g( v) = 0 for all v E 51. Let
further 52 be a kernel of a-51 , Set g( v) = 1 for all v E 52, and so on.
Prove that this process produces a Grundy function.
h: Hint: Use the Theorem 10.6.1 and item g of the exercise.
i: Hint: Use, e.g., item h.
j: Hint: If such a Grundy function exists then we may assume its values to
be colors of the vertices. Prove that such coloring is proper.
a
k: If is a p-chromatic graph and va= LJf::-~ is a partition of its vertex set
into color classes then one may construct another partition as follows. Vti is
Va plus all vertices from Va\ Va adjacent to no vertices of Va. V{ is V1 \ Vti
plus all vertices from Va\(V1 U Vo) adjacent to no vertices of V1\ Vo, and so
on. Prove that the function g(v) = k, v E V; is a Grundy function.
Hypergraphs
11.1 Hypergraphs: Basic Notions
11.1.2. a: Cl3 .
b: Cf89.
Hint: a: The number of different nonempty subsets of a six-element set is
26 -1 = 63. Therefore the maximal possible number of edges in a hypergraph
of order six without multiple edges is 63. Therefore the number of labelled
(6,3)-hypergraphs without multiple edges is Cl
3.
b: Continuing the reasoning of item a, the maximal possible number of edges

in a hypergraph of order six with edge multiplicity at most 3 is 63 . 3 = 189,
hence the given answer.
11.1.3. a: L:vEvHdegv=L:eE£HIeJ.
b: L:vEVH deg v = kl£HI·
11.1.4. a: The independent sets of the matroid M are the independent sets of
vertices of the hypergraph C(M).
b: ao(C(M)) = p(M).
c: If M is the partition matroid for V = VI U ... U Vk specified by the vector
m = (ml, ... , mk), 1 S mi < lVii, i = 1,2, ... , k, then the hypergraph C(M)
has k components Hi =(Vi, £i), i= 1,2, ... , k, where Hi is the complete
(mi + I)-uniform hypergraph.
11.1.6. The cycle C2n.
11.1.7. Yes.
11.1.11. Hint: Consider the Konig representation of such hypergraph. What prop-
erty must it possess?
333
334 11. Hypergraphs
11.1.12. Yes.
11.1.13. b: Hint: Use the inequality m:::; 3n - 6 valid for planar (n, m)-graphs.
11.1.14. a: Possible examples are {{I, 2}} for hypergraph HI, {{I, 2}, {4, 5}} for
H 2•
b: Possible examples are {2,3} for hypergraph HI, {2,3,4} for H 2 .
11.1.16. No; see Fig. Al1.1.l.
Figure A 11.1.1: To Exercise 11.1.16
11.1.17. a: Any graph. b: Any connected graph. c: Any disconnected graph.

Hint: For a graph G without isolated vertices consider the line graph L(G*).
11.1.18. Kn.
11.1.19. The hypergraph Hn defined in Exercise 11.1.18.
11.1.20. Yes.
11.1.21. Hint: Prove that K(H*) ~ K(L(H)).
11.1.22. Hint: For a graph H without isolated vertices, the dual hypergraph H*
is a linear hypergraph. Further, see the hint to the Exercise 11.1.17.
11.1.23. Hint: Use the Exercise 11.1.2l.
11.1.25. Hint: It is sufficient to prove the following: every clique of (Hh is con-
tained in some edge of H if and only if for any triple el, e2, e3 E £H there
exists eo E £H such that (el n e2) U (el n e3) U (e2 n e3) ~ eo.
The necessity part of the latter statement is evident. The sufficiency may
be proved by induction over the number of vertices of a clique C using the
equality C = (ClnC2)U(ClnC3)U(C2nC3), where Ci = C\Vi and Vl,V2, V3
are three different vertices of C.
11.1.27. a: Hint: Suppose that £(v) ~ £(V'), X is a maximal independent vertex
set such that v ¢ X. Then v' E X and X\ {v'} U {v} is also an independent
set.
b: Clearly, H has a minimum cardinality covering without vertex v. There-
fore !3o(H) ~ !3o(H -v). On the other hand, the inequality !3o(H) :::; !3o(H -v)
is evidently valid for any vertex v E V.
c: Hint: Use the Exercises 11.1.27b and 11.1.26c.

d: Solution: For every maximum cardinality matching of a hypergraph H
there corresponds some matching of a hypergraph H -v. Therefore Q;l(H) :s
Q;l(H - v). On the other hand, no matching of H - v may contain two edges
from £(v), since v is a hyperpendant vertex of H. This implies that every
matching of H - v is a matching of H, i.e., Q;l(H) ~ Q;l(H - v).
11.1.28. Hint: Let G be a graph with a covering C = (L1' ... , Ls) described in the
statement of the exercise. Then v belongs to at most k complete graphs Li
from C. If it belongs to I such graphs then we add k - I copies of J( 1 with
the vertex set {v} to C. Doing the same for all vertices of G, we obtain a
covering C' =(L1' ... , L t ). Consider the hypergraph H =
(V, £) such that
= =
V VG and £ {VL 1,. '" VLd. Prove that L(H*) ~ G.
Conversely, let V H = {Vl,"" vn } be the vertex set of a hypergraph H.
Consider the covering of the line graph L(H) by the graphs L 1 , •.. , Ln ,
where Li , i = 1, 2, ... ,n, is a subgraph of L( H) induced by the set of edges
of H incident to Vi.
11.1.29. Using the previous exercise, prove that the graph shown at Fig. A11.1.2
is the required one for any t.
Vt-l Vt
11.1.31. See the similar statement for graphs in Chapter 4 "Connectivity".
11.1.32. Only the sufficiency part is true. See, e.g., Fig. A11.1.3.
11.1.33. Only the sufficiency part is true. See, e.g., Fig. A11.1.3.
11.1.34. Yes for H 1 , no for H 2 .

336 11. Hypergraphs
11.1.35. Solution: a: Let Va be a vertex of the hypergraph H of degree 8(H).

The deletion of all edges of E( va) from H clearly increases the number of
components, therefore )"(H) :::; 8(H).
b: The following relations are evident:
r(H)IE(H)I2: L lei = L IE(v)l2: IHI8(H).

eEeH vEVH
Therefore 8(H) :::; IEHlr(H)/IHI.

11.1.36. a: X(H~) = rn/(k - 1)1. b: x(HI) = 2; X(H 2 ) = 3; X(H3) = 2.
11.1.37. a, b: It either remains unchanged or decreases by one.
11.1.38. The converse statement is false.
Hint: The proof is similar to the proof of the sufficiency for the Konig theo-
rem (Exercise 1.2.15). An example of a bicolorable hypergraph with a cycle
of odd length is shown at Fig. A11.1.4.
11.2 Hypergraph Realizations

11.2.1. Yes for HI, no for H 2 •
11.2. Hypergraph Realizations 337
11.2.2. Yes for Hi, H2, H4 no for H3 . A planar realization for the hypergraph H2
is shown at Fig. A11.1.5.
11.2.3. Hint: a: A required realization of the hypergraph fI may be obtained from
a planar realization R = (V, E) of H, respectively, by
- deletion of all edges from E that belong to no edge of fI;
- contraction of all edges from E that belong to the contracted edge of H;
- addition of a new vertex and the edge connecting it with some vertex of
the given edge of the hypergraphj
- contraction of some edge from E incident to the deleted vertex.
b: Use the item a.
11.2.4. Solution: A planar realization for the hypergraph shown at Fig. A11.2.1
exists (see Exercise 11.2.2). But the deletion of the vertex v produces the
hypergraph which is the graph K 3 ,3'
11.2.5. Solution: Suppose that H = (V, f). Let E( v) denote the set of edges of
the graph K(H) = (VU£, E) incident to v E £. Selecting an edge e(v) from
every set E(v), v E £, and contracting it, we obtain a planar realization of
H.
An example when a planar realization of a hypergraph exists but its Konig
representation is nonplanar is H4 from Fig. 11.2.3.
11.2.6. a: Hint: Use the Theorem 11.2.1. b: A connected hypergraph H with at

most three edges is realizable by tree if either L(H) ?F C 3 or L(H) ~ C3 and
H has a vertex belonging to all three edges.
11.2.7. Solution: Without loss of generality we may consider only connected hy-
pergraphs, and taking the previous exercise into account, we may consider
only four-edge hypergraphs. Let H = (V, £) be a four-edge hypergraph and
let H' = (V', e') be a connected part of H with le'l = 3. Let us construct
a realization R' of H' by tree or cycle. Select a vertex in R' belonging to
an edge e E £\£' and connect it with all remaining vertices from e. The
resulting graph is a planar realization of H.
338 11. Hypergraphs
11.2.8. Hint: Use the similar statement for graphs.

11.2.9. Yes. Hint: The Konig representation of a hypergraph H is isomorphic to
that of H*. Now you may use the Exercise 11.2.5.
11.2.10. No. Solution: The hypergraph H4 from Fig. 11.2.3 has a planar realiza-
tion, but H: is K 3 ,3.
11.2.11. Hint: Let P be a path graph which is a realization of the hypergraph
obtained from H by the deletion of two vertices VI, V2. Then the required
planar realization of H may be obtained by the addition of VI, V2 to P and
connecting Vi, i = 1,2, with the vertices of P which are adjacent to Vi in H.
If VI, V2 are adjacent in H then we add the edge VI v2.
11.2.13. Solution: The required realization may be obtained by the addition of

the two vertices deleted from H to the cycle graph C which is a realization
of H' and connecting them by edges with all vertices of C with which they
are adjacent in H.
11.2.14. No. Hint: See the hypergraph H2 from Fig. 11.2.2.
11.2.16. Hint: Necessity: Let f: V H -+ {I, 2, ... , k} be a proper coloring of a
k-colorable hypergraph H. Consider the graph R with the same vertex set
V H whose edge set is all pairs of differently colored vertices that belong to
the same edge of H. Clearly, R is k-colorable and is a realization of H.
Sufficiency: Clearly, every proper coloring of a realization of a hypergraph
is a proper coloring of the hypergraph itself.
11.2.17. 2. Hint: Use the Exercise 11.2.16.
11.2.18. Hint: Use the Exercise 11.2.16 and the Four Color Conjecture.
11.2.19. a: No. b: Yes.
Solution: a: Should there exist a strict realization of the hypergraph H, it
would contain at least 2·4 = 8 edges, which is impossible for a four-vertex
graph.
b: See Fig. Al1.2.2.
3
Figure A11.2.2: To Exercise l1.2.19b
11.2.20. See Fig. A11.2.3.

11.2. Hypergraph Realizations 339
3
4 5
8 9
e---------~~---e
7
11.2.22. a: Hint: A realization of a hypergraph H with t(H) edges is obtained

from a realization of H - v with t(H - v) edges by the addition of a vertex
v and an edge vv / , where v'is such that t'(v) ~ t'(v' ).
b: Hint: Use the item a.
11.2.23. Hint: A possible proof is by induction over the cardinality of the set of
pairwise adjacent edges in the following way. Let {et, ... , ek} be a set of
pairwise adjacent edges. Consider three sets:
n n n
k-i k
Vi = ei -::j:. 0, V2 = ei -::j:. 0, V3 = ei ek -::j:. 0
and prove that Vi n V2 n V3 -::j:. 0.

11.2.25. Hint: To prove the necessity, use the Exercise 11.2.24 and the Theorem
11.2.1. The sufficiency may be proved by contradiction.
Bibliography
[1] Berge C. Graphs and Hypergraphs, North-Holland Pub!. Co., 1973.
[2] Christofides N. Graph Theory: An Algorithmic Approach, Academic Press,

1975.
[3] Evstigneev V .A., Melnikov 1.S. Problems and Exercises in Graph Theory and
Combinatorics, Novosibirsk University Pub!., Novosibirsk, 1981 (in Russian).
[4] Gavrilov G.P., Sapozhenko A.A. Collected Exercises in Discrete Mathematics,

Nauka, Moscow, 1977 (in Russian).
[5] Harary F. Graph Theory, Addison-Wesley Publ.Co., 1969.
[6] Melnikov 0., Tyshkevich R., Yemelichev V., and Sarvanov V. Lectures on
Graph Theory, BI-Wissenschaftsverlag, Mannheim, 1994.
[7] Ore O. Theory of Graphs, AMS, Providence, Rhode Island, 1962.

[8] Swamy M.N.S, Thulasiraman K. Graphs, Networks and Algorithms, Wiley,
1981.
[9] Tutte W.T. Graph Theory, Addison-Wesley Publ.Co., 1984.
[10] Wilson R.J. Introduction to Graph Theory, Oliver and Boyd, Edinburgh, 1972.
[11] Zykov A.A. Fundamentals of Graph Theory, BCS Associates, Moscow, Idaho,
USA, 1990. (Transl. from Russian original.)
341
*
Index
n-permutation graph, 11
r-regular graph, 6
s-complete property, 122
s-critical, 101
(0, I)-matrix, 31
i-critical, 101
(kl' k 2 , . .. , km)-transversal, 70 P-unigraphical sequence, 126
(n, m)-hypergraph, 173 I-procedure, 119
(u, v)-walk in digraph, 154 I-factor, 66
(VI, vk+d-walk, 13 1-factorizable graph, 66
(VI, vl+d-path in hypergraph, 174
2 - 3-tree, 158 abstract dual, 102
2-connected graph, 71 abstract dual pseudograph, 79
F-invariant subset, 35 acyclic graph, 13, 41
a-perfect graph, 147 addition of edge, 27
x-perfect graph, 147 adjacency matrix, 31
d-complete vertex subset, 124 adjacency matrix of digraph, 160
k-chromatic graph, 135 adjacent edges, 5, 173
k-colorable graph, 135 adjacent vertices, 5, 173
k-colorable hypergraph, 175 alternate group, 38
k-colorable map, 144 alternating cycle, 121
k-coloring, 135 alternating path, 66
k-coloring of map, 144 antibase of digraph, 168
k-component, 75 antichain, 81
k-connected graph, 75 arboricity of graph, 109
k-connected hypergraph, 174 arc base of digraph, 167
k-critical graph, 136 arc of digraph, 151
k-cycle, 13 arity partition of graph, 132
k-edge connected graph, 75 automorphism group of graph, 37
k-edge-critical graph, 136 automorphism of graph, 37
k-polytope, 115 axioms of cycles, 84
k-th power of graph, 28 axioms of independence, 81
k-tree, 140 axioms of metric, 15
k-uniform hypergraph, 173
l-unigraph, 125 balanced cycle, 182
l-unigraphical sequence, 125 balanced hypergraph, 182
n-dimensional cube, 28 barycenter of tree, 48
n-hypercube, 28 barycentral vertex, 48
342
Index 343
base component of digraph, 168 clique, 22, 55

base of digraph, 167 clique covering matrix of graph, 61
base of independence system, 81 clique graph, 22, 56
biconnected graph, 71 clique, maximal, 22, 55
binary matrix, 31 cobase of matroid, 85
binary matroid, 90 cocycle, 85
binary tree, 158 cocycle basis, 90
bipartite graph, 5 co cycle space, 90
bipartite switching, 124 cocyclic rank, 49
bipartite tournament, 167 codependent set, 85
bipartition, 5 cographic matroid, 85
block, 71 coindependent set, 85
block graph, 74 color class, 175
block of hypergraph, 174 color class of vertices, 135
block, pendant, 74 coloring of hypergraph, 175
boundary of face of plane graph, 94 column intersection graph, 61
branch of tree, 44 common divisor graph, 12
branching vertex, 42 comparability graph, 148
breadth-first search, 14 complementary graph, 6
bridge, 71 complete k-uniform hypergraph, 173
complete bipartite graph, 6
cactus, 122 complete digraph, 151
caterpillar, 45, 115 complete graph, 3, 5
Cayley Theorem, 47 component of graph, 14
center of graph, 14 component of hypergraph, 174
central subgraph, 14 component, connected, 154
central vertex, 14 component, strong, 154
centroid of tree, 44 component, weak, 154
centroidal vertex, 44 condensation of digraph, 154
chain in digraph, 154 conformal hypergraph, 177
chain, cyclic, 154 conjugate sequence, 123
chain, Eulerian, 163 conjunction of properties, 25
characteristic polynomial connected component, 154
of graph, 31 connected component of graph, 14
chess piece graph, 137 connected domain of graph, 14
chromatic function of graph, 141 connected graph, 14
chromatic index, 142 connected hypergraph, 174
chromatic number of graph, 135 contour, 154
chromatic number contour, Hamiltonian, 163
of hypergraph, 175 contractible graph, 99
chromatic number of surface, 145 contraction of edge, 28
chromatic polynomial of graph, 141 corank function, 85
chromatically unique graph, 141 cospectral graphs, 31
circulant graph, 10 covering number of hypergraph, 175
circulant matrix, 34 covering of hypergraph, 175
circumference of graph, 136 critical graph, 136
344 Index
critically biconnected graph, 71 digraph, Eulerian, 163

crossing number of graph, 108 digraph, Hamiltonian, 163
cubic graph, 6 digraph, irreducible, 156
cut, 77 digraph, minimally connected, 158
cut basis of graph, 91 digraph, permutation, 160
cutpoint, 71 digraph, quasistrong, 158
cutpoint of hypergraph, 174 digraph, strong, 154
cycle, 13 digraph, strongly connected, 154
cycle basis, 90 digraph, transitive, 159
cycle basis of graph, 91 digraph, unilateral, 154
cycle in hypergraph, 174 digraph, unilaterally connected, 154
cycle in hypergraph, simple, 174 digraph, weak, 154
cycle of independence system, 81 digraph, weakly connected, 154
cycle space, 90 direct product of groups, 36
cycle, balanced, 182 directable graph, 158
cycle, Eulerian, III directed cut, 157
cycle, Hamiltonian, 112 directed graph, 151
cycle, simple, 3, 13 disconnected digraph, 155
cyclic chain, 154 disjoint union, 27
cyclic matroid, 85 disjunction of properties, 25
cyclic path, 154 dismantling of graph, 147
cyclic rank, 49 distance in digraph, 154
cyclic walk, 13, 154 distance in graph, 14
cyclically k-connected graph, 77 domatic number of graph, 65
cyclomatic number, 49 dominating number, 64
dominating set, 64
deck, 21 dominating vertex, 5
deficiency of bipartite graph, 68 dominating vertex set in digraph, 167
degree of edge, 125 dual hypergraph, 174
degree of edge of hypergraph, 173 dual matroid, 85
degree of regular graph, 6
degree of vertex, 5 edge ao-critical graph, 59
degree of vertex of digraph, 152 edge k-coloring, 142
degree sequence, 5 edge addition, 27
degree set of graph, 132 edge color class, 142
deletion of elements of graph, 27 edge coloring, 142
dependent subset, 81 edge connectivity number, 75
depth of vertex, 158 edge connectivity number of hyper-
derived sequence, 118 graph, 175
diagonal of cycle, 73 edge contraction, 180
diameter of graph, 14 edge covering number, 62
diametral path, 14 edge covering of graph, 62
different edges of hyper graph , 174 edge deck, 21
digraph, 151 edge group of graph, 39
digraph, disconnected, 155 edge of graph, 3
digraph, edge-labelled, 159 edge of hypergraph, 173
Index 345
edge reconstructible graph, 21 generalized dual graph, 106

edge reconstruction conjecture, 21 genus, 107
edge sequence, 125 geometric dual graph, 101
edge subdivision, 19, 99 girth, 77
edge, pendant, 5 girth of graph, 14
edge-critical graph, 63, 136 graph,3
edge-deleted subgraph, 21 graph of genus ,,(, 107
edge-labelled digraph, 159 graph of realizations, 121
edge-labelled graph, 159 graph, 2-connected, 71
elementary contraction of graph, 139 graph, et-perfect, 147
elementary homomorphism of graph, graph, x-perfect, 147
138 graph, k-chromatic, 135
embeddable graph, 93 graph, k-colorable, 135
embedding of graph into space, 93 graph, k-connected, 75
empty digraph, 151 graph, k-critical, 136
empty graph, 5 graph, k-edge connected, 75
end of arc, 151 graph, k-edge-critical, 136
end of edge, 5 graph, r-regular, 6
endpoint of edge, 5 graph, 1-factorizable, 66
endvertex of arc, 151 graph, acyclic, 13, 41
endvertex of walk, 13 graph, biconnected, 71
equality of labelled graphs, 7 graph, bipartite, 5
Eulerian chain, 163 graph, chromatically unique, 141
Eulerian cycle, 111 graph, circulant, 10
Eulerian digraph, 163 graph, complementary, 6
Eulerian graph, 111 graph, complete, 3, 5
Eulerian path, 111 graph, complete bipartite, 6
even permutation, 38 graph, connected, 14
excentricity of vertex, 14 graph, critical, 136
extension operation, 30 graph, critically biconnected, 71
external face, 94 graph, cubic, 6
graph, cyclically k-connected, 77
face of plane graph, 94 graph, directable, 158
face, external, 94 graph, edge eto-critical, 59
face, internal, 94 graph, edge reconstructible, 21
face, outer, 94 graph, edge-critical, 63, 136
Ferrer diagram, 123 graph, edge-labelled, 159
finite graph, 3 graph, embeddable, 93
FIS, 20 graph, empty, 5
forbidden induced graph, 20 graph, Eulerian, 111
forbidden subgraph, 45 graph, finite, 3
forcibly P-graphical sequence, 126 graph, generalized dual, 106
forest, 41 graph, geometric dual, 101
Four-Color Conjecture, 145 graph, Hamiltonian, 112
free independence system, 81 graph, Hamiltonian-connected, 113
Fulkerson theorem, 123 graph, homotraceable, 134
346 Index
graph, hypohamiltonian, 113 Hamiltonian graph, 112

graph, isochromatic, 144 Hamiltonian path, 112, 163
graph, labelled, 7 Hamiltonian-connected graph, 113
graph, line, 6 Hamming distance, 27
graph, locally Hamiltonian, 116 Hamming graph, 27
graph, locally-connected, 113 head of are, 151
graph, maximal outerplanar, 94 height of tree, 158
graph, maximal planar, 98 Helly conditions, 179
graph, maximal plane, 98 hereditary property, 20
graph, ordinary, 7 hereditary subsequence, 128
graph, outerplanar, 94 homeomorphic, 99
graph, pancyclic, 114 homotraceable graph, 134
graph, perfect, 147 hyperedge, 173
graph, Platonic, 5 hypergraph, 173
graph, quasitriangulated, 147 hypergraph, k-colorable, 175
graph, reconstructible, 21 hypergraph, k-connected, 174
graph, regular, 6 hypergraph, k-uniform, 173
graph, rigid, 77 hypergraph, balanced, 182
graph, selfcomplementary, 6 hypergraph, complete k-uniform, 173
graph, selfdual, 105 hypergraph, conformal, 177
graph, simple, 7 hypergraph, connected, 174
graph, split, 129 hypergraph, linear, 177
graph, star, 6 hypergraph, planar, 176
graph, threshold, 59, 129 hypergraph, selfdual, 176
graph, toroidal, 107 hypergraph, strongly balanced, 182
graph, triangulated, 147 hypohamiltonian graph, 113
graph, trivial, 5 hypopendant vertex, 177
graph, unicyclic, 127
graph, uniquely k-colorable, 136 incidence graph, 174
graph, uniquely edge-colorable, 144 incidence matrix, 32
graph, vertex-critical, 63 incidence matrix of digraph, 160
graph, weighted, 49 incident, 5, 151
graph-theoretical property, 20 incomparable vertices, 165
graphic matroid, 85 indegree of vertex of digraph, 152
graphical independence system, 82 independence axioms, 81
graphical pair of sequences, 118 independence axioms of matroid, 83
graphical sequence, 117, 118 independence number of graph, 55
grid graph, 114 independence number of hyper graph,
group, intransitive, 35 175
group, transitive, 35 independence system, 81
Grundy function, 170 independent set of edges, 66
independent set of vertices, 55
Hadwiger Conjecture, 145 independent set of vertices of hyper-
Hamiltonian contour, 163 graph,175
Hamiltonian cycle, 112 independent set of vertices, maximal,
Hamiltonian digraph, 163 55
Index 347
independent vertex set loop of matroid, 84

in digraph, 167 lower part of split graph, 129
induced subgraph, 20, 154
major modular product, 30
inkernel, 169
map, 144
inner vertex, 154
map, k-colorable, 144
intermediate vertex, 154
internal face, 94 matching, 66
matching number, 66
intersection of graphs, 28
matching number of hypergraph, 175
interval graph, 149
matching of hypergraph, 175
interval tree, 45
matching, maximal, 66
intransitive group, 35
matching, maximum cardinality, 66
irreducible digraph, 156
matrix, totally unimodular, 34, 162
isochromatic graph, 144
isolated vertex, 5, 173 matroid,83
matroid of cuts, 85
isomorphic, 5
matroid of cycles, 85
isomorphic hypergraphs, 174
matroid, representable, 90
isomorphism, 5
maximal clique, 22, 55
join of graphs, 28 maximal independent set of vertices,
Jordan curve, 93 55
Jordan's algorithm, 43 maximal matching, 66
Jordan's theorem, 43 maximal outerplanar graph, 94
maximal planar graph, 98
Konig representation of graph, 19 maximal plane graph, 98
Konig representation of hypergraph, maximum cardinality clique, 55
174 maximum cardinality independent
Kelly-Ulam conjecture, 21 set, 55
kernel of digraph, 167 maximum cardinality independent
kernel of graph, 64 vertex set of hyper graph, 175
Kirchhoff matrix, 31 maximum cardinality matching, 66
Klein four-group, 36 maximum cardinality matching
Kruskal's Algorithm, 49 of hypergraph, 175
mean degree, 42
labelled deck, 21 merging by common subgraph, 140
labelled graph, 7 merging of vertices, 28
labelled realization, 121 metrical dimension of graph, 18
leaf of tree, 158 minimal coloring of graph, 135
length of path in hypergraph, 174 minimal dominating set, 64
length of walk, 13, 154 minimal edge coloring, 142
line digraph, 153 minimal edge covering, 62
line graph, 6 Minimal Spanning Tree Problem, 49
line graph of hypergraph, 175 minimal vertex covering, 62
linear hypergraph, 177 minimally connected digraph, 158
locally Hamiltonian graph, 116 minimum cardinality covering of hy-
locally-connected graph, 113 pergraph, 175
long switching, 121 minimum cardinality dominating set,
loop, 7, 151 64
348 Index
minimum cardinality edge covering, perfect matching, 66

62 peripherical vertex, 14
minimum cardinality vertex covering, permutation, 35
62 permutation digraph, 160
minmax kernel, 65 permutation group, 35
multidigraph, 151 permutation matrix, 33
multigraph, 7 permutation, even, 38
multiple arc, 151 Petersen graph, 5
multiple edge of hypergraph, 173 planar hypergraph, 176
multiple edges, 7 planar realization of hypergraph, 179
multiset, 132 plane graph, 93
plane triangulation, 98
neighborhood, 5 Platonic graph, 5
noncyclic walk, 13 Platonic solid, 5
polytope, 61
opposite digraph, 151 potentially P-graphical
orbit of group, 35 sequence, 126
order of digraph, 151 Priifer code of tree, 46
order of graph, 3 Prim's Algorithm, 49
order of hypergraph, 173 proper coloring, 135
ordinary graph, 7 proper sequence, 117
orientation of pseudograph, 155 proper vertex coloring of hyper graph,
outdegree of vertex of digraph, 151 175
outer face, 94 property, s-complete, 122
outerplanar graph, 94 pseudograph, 7
out kernel , 169
quasistrong digraph, 158
pancyclic graph, 114 quasitriangulated graph, 147
parallel arc, 151
part of bipartite graph, 5 radius of graph, 14
part of hypergraph, 174 raising a graph to k-th power, 28
partial transversal, 70 rank function of matroid, 84
partition matroid, 87 rank of matroid, 84
path, 13 rank of subset, 84
path in digraph, 154 Rao Conjecture, 128
path in hypergraph, 174 reach ability matrix of digraph, 160
path in hypergraph, simple, 174 reachable, 154
path partition of digraph, 163 realization of edge of hypergraph, 179
path, alternating, 66 realization of graphical sequence, 117
path, cyclic, 154 realization of hypergraph, 179
path, Eulerian, 111 realization, labelled, 121
path, Hamiltonian, 112, 163 realization, strict, 179
path, simple, 3, 13 reconstructible graph, 21
pendant block, 74 Reconstruction Conjecture, 21
pendant edge, 5 reconstruction conjecture, 21
pendant vertex, 5 regular graph, 6
perfect graph, 147 representable matroid, 90
Index 349
representation of matroid, 90 strongly balanced hypergraph, 182

restriction of mapping, 39 strongly connected digraph, 154
rigid graph, 77 subcycle, 112
root of tree, 158 subgraph, 14, 20, 154
rooted spanning tree, 159 subgraph, forbidden, 45
rooted tree, 158 subgraph, induced, 20, 154
subgraph, spanning, 20, 154
selfcomplementary graph, 6 subsequence, hereditary, 128
selfdual graph, 105 substrate of digraph, 155
selfdual hypergraph, 176 subtree intersection graph, 48
semichain, 154 support of independence system, 81
semicontour, 154 switching, 117
semipath, 154 switching, bipartite, 124
semiwalk, 154 switching, long, 121
separating inequality, 129 symmetric group, 35
separator set, 77 symmetric group of degree n, 35
sequence, l-unigraphical, 125 system of different
sequence, P-unigraphical, 126 representatives, 69
sequence, forcibly P-graphical, 126
sequence, graphical, 117, 118 tail of are, 151
sequence, potentially terminal vertex, 154
P-graphical, 126 terminal vertex of walk, 13
sequence, proper, 117 thickness of graph, 109
sequential coloring of graph, 136 threshold decomposition
set of independent sets, 81 of graph, 131
similar permutation groups, 35 threshold graph, 59, 129
simple cycle, 3, 13 threshold number of graph, 132
simple cycle in hypergraph, 174 topological sorting, 157
simple graph, 7 toroidal graph, 107
simple path, 3, 13 totally unimodular matrix, 34, 162
simple path in hypergraph, 174 toughness of graph, 55
simplicial vertex, 147 tournament, 166
skeleton of graph, 48 tournament, bipartite, 167
skewness of graph, 109 trace of square matrix, 32
source, 157 transitive closure, 159
spanning forest, 48 transitive digraph, 159
spanning subgraph, 20, 154 transitive group, 35
spanning tree, 48 transitive reduction, 159
split graph, 129 transversal, 69
stabilizer of element of group, 35 transversal matroid, 85
star graph, 6 tree, 41
strict realization, 179 tree, binary, 158
Strong Berge Conjecture, 147 tree, rooted, 158
strong component, 154 tree, rooted spanning, 159
strong digraph, 154 triangle, 13, 94
strong product of graphs, 148 triangle inequality, 15
350 Index
triangulated graph, 147 Wagner Theorem, 93

trivial graph, 5 walk,13
trivial independence system, 81 walk in digraph, 154
Thran graph, 58 walk, cyclic, 13, 154
walk, noncyclic, 13
unicyclic graph, 127 weak component, 154
uniform matroid, 87 weak digraph, 154
unigraph, 118 weak edge covering, 67
unilateral digraph, 154 weak product of graphs, 140
unilaterally connected digraph, 154 weakly connected digraph, 154
union of graphs, 27 weight, 49
union of matroids, 85 weight function, 49
uniquely k-colorable graph, 136 weight of vertex, 44
uniquely edge-colorable graph, 144 weighted graph, 49
upper part of split graph, 129 Whitney criterion, 104
wreath product of groups, 36
vector matroid, 85
vertex coloring of hypergraph, 175
vertex coloring of hypergraph,
proper, 175
vertex connectivity number, 75
vertex connectivity number of hyper-
graph, 174
vertex covering, 62
vertex covering number, 62
vertex deletion in hypergraph, 174
vertex merging, 28
vertex of digraph, 151
vertex of graph, 3
vertex reconstruction conjecture, 21
vertex splitting, 57
vertex, branching, 42
vertex, central, 14
vertex, centroidal, 44
vertex, dominating, 5
vertex, hypopendant, 177
vertex, inner, 154
vertex, intermediate, 154
vertex, isolated, 5, 173
vertex, pendant, 5
vertex, peripherical, 14
vertex, simplicial, 147
vertex, terminal, 154
vertex-critical graph, 63
vertex-deleted subgraph, 21
Vizing-Wilf number, 149
Notations
x - sign of Cartesian product (of graphs, sets)
~ - sign of isomorphism (of graphs, digraphs, etc.)
u - sign of union (of graphs, matroids, sets)

n - sign of intersection (of graphs, matroids, sets)
(A, B; E) - bipartite graph with parts A, B and edge set E
AG - set of arcs of digraph G
A( G) - adjacency matrix of graph G

Aut E(G) - edge group of graph G
Aut (G) - automorphism group of graph G
B( G) - block graph of graph G
c(G) - minimal number of cliques covering VG

cc( G) - minimal number of complete graphs whose union is G
8(S) - set of bases of independence system S
Cn - cycle graph of order n
C~ - number of two-element subsets of an n-element set
cr( G) - crossing number of graph G

C(S) - set of cycles of independence system S
deg G - degree of regular graph G

degG v - degree of vertex v in graph G
deg v - degree of vertex v in the considered graph
d( G) - diameter of graph G
352 Notations
dG(u, v) - distance between vertices u and v in graph G
d- (v) - in degree of vertex v
d+ (v) - outdegree of vertex v
EG - edge set of graph G

£ H - edge set of hypergraph H
£( v) - set of edges incident to vertex v of a hypergraph
e( v) - excentricity of vertex v
G - complementary graph for graph G
Gb - substrate of digraph G
G - e - subgraph of G produced by deletion of edge e from G
g( G) - girth of G
G[H] - composition of graphs G and G
G /\ H - conjunction of graphs G and G
Go H - modular product of graphs G and G
G + H - join of graphs G and G

GP - p-th degree of graph G
G - v - subgraph of G produced by deletion of vertex v and the edges incident
to it from G
G(X) - subgraph of G induced by subset X ~ VG
IG - set of independent subsets of vertices of graph G plus empty set

I( G) - incidence matrix of graph G
I(S) - set of independent sets of independence system S
K(G) - (1) Kirchhoff matrix of graph G (2) Konig representation of graph
or hypergraph G
k( G) - number of components of graph G
Kn - complete graph of order n
Kp,q - complete bipartite graph with parts of cardinalities p and q
L(H) - line graph of graph or hypergraph H
M(G) - matroid of cycles of graph G

Notations 353
M*(G) - matroid of cuts of graph G
m( G) - number of edges of graph G
Mij - entry of matrix M at position (i, j)
mJ{2 - graph of order 2m whose edge set is m pairwise nonadjacent edges
n( G) - number of vertices of graph G
N G ( v) - neighborhood of vertex v in graph G
On - empty graph of order n
P( G) - deck of graph G
Pn - path graph of order n
Q( G) - clique graph for graph G
R( G) - reachability matrix of graph G
r( G) - radius of graph G
S( G) - graph of skeletons of graph G
sk( G) - skewness of graph G
t( G) - thickness of graph G
the G) - threshold number of graph G

uv - the edge of a graph connecting the vertices u and v
V H - vertex set of graph or hypergraph H
Wn - wheel graph of order n +1

IXI - cardinality of set X, order of graph, hypergraph or matroid X
X U a - shorthand for X U {a}
X na- shorthand for X n {a}

Zn - additive group of integer residuals modulo n
Z2 - two-element field
O'o(H) - independence number of graph or hypergraph H

0'1 (H) - matching number of graph or hypergraph H
/30 (H) - vertex covering number of graph or hypergraph H
/31(H) - edge covering number of graph or hypergraph H

354 Notations
,( G) ~ genus of graph G
rv ~ the set of heads of arcs out coming from vertex v

r~lv ~ the set of tails of arcs incoming to vertex v
~(G) ~ maximal degree of vertices of graph G
8( G) ~ minimal degree of vertices of graph G
I;? ( G) ~ toughness of graph G
K(H) ~ vertex connectivity number of graph or hypergraph H
>.(H) ~ edge connectivity number of graph or hypergraph H
v( G) ~ cyclic rank of graph G
v* (G) ~ co cyclic rank of graph G
~(G) ~ arboricity of graph G
p(M) ~ rank of matroid M
X(H) ~ chromatic number of graph or hypergraph G
X'(H) ~ chromatic index of graph G

Kluwer Texts in the Mathematical Sciences
1. A.A. Harms and D.R. Wyman: Mathematics and Physics ofNeutron Radiography. 1986
ISBN 90-277-2191-2
2. H.A. Mavromatis: Exercises in Quantum Mechanics. A Collection of lllustrative Problems
and Their Solutions. 1987 ISBN 90-277-2288-9
3. V.1. Kukulin, V.M. Krasnopol'sky and J. Horacek: Theory of Resonances. Principles and
Applications. 1989 ISBN 90-277-2364-8
4. M. Anderson and Todd Feil: Lattice-Ordered Groups. An Introduction. 1988
ISBN 90-277-2643-4
5. J. Avery: Hyperspherical Harmonics. Applications in Quantum Theory. 1989
ISBN 0-7923-0165-X
6. H.A. Mavromatis: Exercises in Quantum Mechanics. A Collection of lllustrative Problems
and Their Solutions. Second Revised Edition. 1992 ISBN 0-7923-1557-X
7. G. Micula and P. Pavel: Differential and Integral Equations through Practical Problems and
Exercises. 1992 ISBN 0-7923-1890-0
8. W.S. Anglin: The Queen ofMathematics. An Introduction to Number Theory. 1995
ISBN 0-7923-3287-3
9. Y.G. Borisovich, N.M. Bliznyakov, T.N. Fomenko and Y.A. Izrailevich: Introduction to Dif-
ferential and Algebraic Topology. 1995 ISBN 0-7923-3499-X
10. J. Schmeelk, D. TakacQi and A. TakacQi: Elementary Analysis through Examples and Exercis-
es.1995 ISBN 0-7923-3597-X
11. J.S. Golan: Foundations ofLinear Algebra. 1995 ISBN 0-7923-3614-3
12. S.S. Kutateladze: Fundamentals of Functional Analysis. 1996 ISBN 0-7923-3898-7
13. R. Lavendhomme: Basic Concepts of Synthetic Differential Geometry. 1996
ISBN 0-7923-3941-X
14. G.P. Gavrilov and A.A. Sapozhenko: Problems and Exercises in Discrete Mathematics. 1996
ISBN 0-7923-4036-1
15. R. Singh and N. Singh Mangat: Elements of Survey Sampling. 1996 ISBN 0-7923-4045-0
16. C.D. Ahlbrandt and A.C. Peterson: Discrete Hamiltonian Systems. Difference Equations, Con-
tinued Fractions, and Riccati Equations. 1996 ISBN 0-7923-4277-1
17. J. Engelbrecht: Nonlinear Wave Dynamics. Complexity and Simplicity. 1997
ISBN 0-7923-4508-8
18. E. Pap, A. TakaCi and D. Takaci: Partial Differential Equations through Examples and Exer-
cises. 1997 ISBN 0-7923-4724-2
19. O. Melnikov, V. Sarvanov, R. Tyshkevich, V. Yemelichev and I. Zverovich: Exercises in Graph
Theory. 1998 ISBN 0-7923-4906-7
KLUWER ACADEMIC PUBLISHERS - DORDRECHT I BOSTON I LONDON

(Kluwer Texts in The Mathematical Sciences 19) O. Melnikov, V. Sarvanov, R. Tyshkevich, V. Yemelichev, I. Zverovich (Auth.) - Exercises in Graph Theory-Springer Netherlands (1998) PDF

Uploaded by

Copyright:

Available Formats

(Kluwer Texts in The Mathematical Sciences 19) O. Melnikov, V. Sarvanov, R. Tyshkevich, V. Yemelichev, I. Zverovich (Auth.) - Exercises in Graph Theory-Springer Netherlands (1998) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Kluwer Texts in The Mathematical Sciences 19) O. Melnikov, V. Sarvanov, R. Tyshkevich, V. Yemelichev, I. Zverovich (Auth.) - Exercises in Graph Theory-Springer Netherlands (1998) PDF

Uploaded by

Copyright:

Available Formats

What is the book about?

What is the book about?

What topics are covered in the book?

What topics are covered in the book?

Exercises in Graph Theory

Kluwer Texts in the Mathematical Sciences

A Graduate-Level Book Series

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

ISBN 978-90-481-4979-7 ISBN 978-94-017-1514-0 (eBook)

Printed on acid-free paper

AII Rights Reserved

1 ABC of Graph Theory 3

3 Independence and Coverings 55

604 Duality and Planarity .. 101

7 Graph Traversals 111

8 Degree Sequences 117

9 Graph Colorings 135

10 Directed Graphs 15-1

Answers to Chapter 1: ABC of Graph Theory 183

Answers to Chapter 2: Trees 207

Answers to Chapter 3: Independence and Coverings 223

Answers to Chapter 4: Connectivity 241

Answers to Chapter 5: Matroids 251

Answers to Chapter 6: Planarity 257

Answers to Chapter 7: Graph Traversals 275

Answers to Chapter 8: Degree Sequences 281

Answers to Chapter 9: Graph Colorings 297

Answers to Chapter 10: Directed Graphs 313

10.6 Base and Kernel . 326

Answers to Chapter 11: Hypergraphs 333

O. Melnikov et al., Exercises in Graph Theory

ABC of Graph Theory

1.1 Graphs: Basic Notions

Figure 1.1.1: Examples of graphs

Figure 1.1.2: Platoni c solids and their graphs

Proposition 1.1.1 (Handshake Lemma) The sum of degrees of the vertices of

A vertex of degree 0 is called isolated; a vertex of degree 1 is called pendant.

Figure 1.1.3: Bipartite graphs

bipartite graph G = (A, B; E) with IAI = p ~ 1, IBI = q ~ 1 is such that every

Figure 1.1.4: Graph and its line graph

A graph G is called sel/complementary if G ~ G. For example, the path graph

Figure 1.1.5: Line graph of Petersen graph

For a graph G the line graph L( G) is defined as follows:

1. The vertices of L(G) bijectively correspond to the edges of G.

2. Two vertices of L( G) are adjacent if and only if the corresponding edges of

Figure 1.1.6: Multigraph and pseudograph

For graphs of all these types it is natural to define isomorphism to be a bi-

Figure 1.1.7: Labelled graph

Figure 1.1.8: To Exercise 1.1.2

1.1.3. Describe a graph satisfying one of the following conditions:

1.1.16. Give an example (if possible) of

c) an (n, m)-graph with m = C~_l.

Figure 1.1.10: To Exercise 1.1.22a

Figure 1.1.11: To Exercise 1.1.22a

for some positive integer k coprime to n. Prove that the mapping

is an isomorphism of graphs Gn,u, and G n ,u2'

1.1.27. n-Permuiaiion graph: A graph of order n! whose vertices bijectively corre-

a) Draw the 3-permutation graph;

1.1.29. Propose the definitions of isomorphism for multigraphs and pseudographs