(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
(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
(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
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
Introduction . . . . . . . 1
2 Trees 41
2.1 Trees: Basic Notions . . . . . 41
2.2 Skeletons and Spanning Trees 48
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
11 Hypergraphs 173
11.1 Hypergraphs: Basic Notions. 173
11.2 Hypergraph Realizations . . . 179
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.
--- -------
P2 P3
• • •
P4
• • • • • •
P5
L C3
D 0
C4 C5
a) b)
•
Kl -- L I2?J
K2
c)
K3 K4
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
O. Melnikov et al., Exercises in Graph Theory
© Springer Science+Business Media Dordrecht 1998
4 1. ABC of Graph Theory
Tetrahedron
Cube
Octahedron
Dodecahedron
Icosahedron
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
6 1. ABC of Graph Theory
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.
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.
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}.
* * *
8 1. ABC of Graph Theory
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
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. Graphs: Basic Notions 9
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
Figure 1.1.9: To Exercise 1.1.14
1.1.15. Prove that if graphs G and H are isomorphic then the complementary
graphs G and H are isomorphic.
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
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.
1.1.36. 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
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.
(1.2.1)
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):
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.
14 1. ABC of Graph Theory
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:
* * *
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.
1.2. Walks, Paths, Components 15
4 3
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)~
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.
16 1. ABC of Graph Theory
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:
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. Walks, Paths, Components 17
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:
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.
(n-k)(n-k+l)
n- k :::;m:::; 2 . (1.2.2)
(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 '
18 1. ABC of Graph Theory
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. Walks, Paths, Components 19
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
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.
2
~
G
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
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
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
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?
1.3.8. 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 .
24 1. ABC of Graph Theory
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
1.3.23. Prove that a graph is complete bipartite or empty if and only if it has no
induced graphs K3 and P3.
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.
1.3.29. Determine whether the following collection of graphs constitute the deck
of some graph:
1.3.30. Prove that all regular graphs of order at least 3 are reconstructible.
26 1. ABC of Graph Theory
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.
Ou(Jv C5 -v C5 - e c5 + e
Figure 1.4.1: Operations on graph
- 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
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.
1.4.6. 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.
1.4.14. 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?
30 1. ABC of Graph Theory
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.
1.4.21. 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.
1.4.26. 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.
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:
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
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:
[~ 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.
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) C1 = 0;
34 1. ABC of Graph Theory
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 =
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
(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
(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
(1.6.4)
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:
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.
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.
38 1. ABC of Graph Theory
5 8
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.
1.6.15. 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
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.
40 1. ABC of Graph Theory
Trees
Theorem 2.1.1 The following statements are equivalent for an (n, m )-graph G :
1. G is a tree.
3. G is a forest and m = n - 1.
* * *
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?
41
O. Melnikov et al., Exercises in Graph Theory
© Springer Science+Business Media Dordrecht 1998
42 2. Trees
2.1.11. Prove that the following statements are equivalent for a connected graph
G:
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.
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
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.)
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.
2.1.24. 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.
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.
2.1. Trees: Basic Notions 45
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.
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
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:
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.
2.1.53. 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.
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.
w(H) = L wee).
eEEH
* * *
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;
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'
2.2.6. 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
a) H has n vertices;
b) H has n -1 edges;
c) H is connected;
d) H is a forest.
Prove that:
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
2.2.17. Prove that if a graph has a simple cycle of length k then the graph has at
least k 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.
2.2.22. 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.
2.2.25. 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 .
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
--------------~.
Figure 2.2.4: To Exercise 2.2.31
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.
Independence and
Coverings
IGI
a o(G)2 1 + d , (3.1.2)
IGI
ao(G) 2 1 + b.(G) , (3.1.3)
55
O. Melnikov et al., Exercises in Graph Theory
© Springer Science+Business Media Dordrecht 1998
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
3.1.5. Is it true that every maximal independent set of vertices of a graph is the
maximum cardinality independent set?
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.
3.1.12. 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.
ao(G) - a~(G) = a.
3.1.15. Prove that for graphs G, H
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).
(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:
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.
3.1. Independent Vertex Sets and Cliques 59
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.
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
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.
3.1. Independent Vertex Sets and Cliques 61
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.
Prove the following statements:
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
3.1.43. 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:
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:
* * *
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.
3.2.8. 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.
* * *
3.3.1. 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.
3.3.8. 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.
3.3.13. 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.
3.3.16. Prove that for every graph G there exists a graph H containing a minimum
cardinality dominating set D such that H(D) ~ G.
3.3.1S. 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.
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).
66 3. Independence and Coverings
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)
* * *
3.4.1. Evaluate 0:'1 (G) for the graphs shown at Fig. 1.2.4. Which of these graphs
have a perfect matching?
3.4.8. 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
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.
68 3. Independence and Coverings
Define
6(A) = {IAI-IN(A)!, A i= 0, .
0, otherWIse.
The value
60 (X, Y; E) = max6(A)
A~X
* * *
3.5. Matchings in Bipartite Graphs 69
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.
for every A ~ X.
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
Connectivity
Theorem 4.1.1 For a connected graph G, IGI > 2, the following statements are
equivalent:
1) G is biconnected;
6) for any three vertices a, b, c, the graph G has a simple (a, b)-path containing
c.
* * *
71
O. Melnikov et al., Exercises in Graph Theory
© Springer Science+Business Media Dordrecht 1998
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:
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.
4.1.8. Prove that d(G) ~ LlGI/2J for a biconnected graph G, where d(G) is the
diameter of G.
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.
4.1.25. 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?
4.1.29. 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.
4.1.31. 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.
* * *
4.2.1. 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
4.2.6. 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).
4.2.23. 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?
4.2.27. 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).
* * *
4.3.1. 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.
G H
Matroids
* * *
81
O. Melnikov et al., Exercises in Graph Theory
© Springer Science+Business Media Dordrecht 1998
82 5. Matroids
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}.
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 ,
= =
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.
5.1.6. Let G be a graph. Let IG denote the set of independent subsets of vertices
of G plus the empty set.
5.1.10. Find the cycles and the bases of the independence system specified by the
Boolean function:
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.
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.
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.
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).
Step k : Find ek E E\{el, ... , ek-d such that {el, ... , ek-l, ed E I and
* * *
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)
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.
5.2.15. Prove the following statements.
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
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.
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
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.
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
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
5.3.5. 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
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.
93
O. Melnikov et al., Exercises in Graph Theory
© Springer Science+Business Media Dordrecht 1998
94 6. Planarity
2 3
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.
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.
6.1.12. 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.
6.1.18. Prove that every planar 5-connected graph has at least 12 vertices. Con-
struct one of them.
6.1.20. Prove that a planar connected cubic graph whose every face has at least 5
vertices is of order at least 20.
6.1.22. 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).
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.
G3
Figure 6.1.3: To Exercise 6.1.33
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.
Theorem 6.2.1 A graph is maximal plane graph if and only if it is a plane tri-
angulation.
* * *
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.
* * *
6.3.1. 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.
a b c d
a b
6.3.16. Using the Wagner criterion, prove that the Petersen graph is nonplanar.
6.4. Duality and Planarity 101
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.
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.
-----"-"--------------
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.
G G+
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.
a e
d h
G G+
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
* * *
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.
6.4.4. 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.
6.4.7. 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.
6.4.10. Under which conditions a graph geometric dual to an ordinary plane graph
is an ordinary graph?
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
_...---_._--
a b a
c c
a b a
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 ______ ~------.------
The following formulae are known for the genus of some graphs:
* * *
6.5.1. Prove that a) "Y(Ks) = 1; a) "Y(K7) = 1; a) "Y(K4,4) = 1.
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?
6.5.8. What are the crossing numbers of K 3 ,3, Ks and the Petersen graph?
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.
6.5.20. Find the skewness for K 3 ,3, K 5 , and the Petersen graph.
Graph Traversals
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.
* * *
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
O. Melnikov et al., Exercises in Graph Theory
© Springer Science+Business Media Dordrecht 1998
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?
7.1.8. 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.
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.
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
* * *
7.2.1. Find a Hamiltonian cycle a) in the graph shown at Fig. 7.2.1; b) in the
dodecahedron graph.
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?
7.2.6. Prove that if a cubic graph is Hamiltonian then it contains three edge-
disjoint perfect matchings. Is the converse statement true?
114 7. Graph Traversals
~ !>
Figure 7.2.2: To Exercise 7.2.7
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.)
7.2.14. 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.
7.2.20. 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.
7.2.23. 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.
7.2.27. 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.
7.2.31. 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.
7.2.34. 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.
116 7. Graph Traversals
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
• •
G H
117
O. Melnikov et al., Exercises in Graph Theory
© Springer Science+Business Media Dordrecht 1998
118 8. Degree Sequences
a) n ~ 3;
c) d 1 + 2 = d2 + ... + dn - 1, d2 = da = .. , = dn ;
d) n = 4, da > d4 = 1;
e) d2 = da = ... = dn = 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.
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
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.
120 8. Degree Sequences
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,------~-+------: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.
122 8. Degree Sequences
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 ).
8.1.21. 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?
8.1.25. 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. Graphical Sequences 123
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
8.1.35. Prove that there exist infinitely many graphical n-sequences for which the
inequality from Theorem 8.1.5 turns into the equality.
124 8. Degree Sequences
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.
8.1.39. 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?
8.1.42. 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?
8.1.50. 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. Graphical Sequences 125
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.
8.1.59. 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
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.
126 8. Degree Sequences
* * *
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.
8.2.3. 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.
128 8. Degree Sequences
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.
8.2.27. 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)?
* * *
8.3.1. Prove that for n ~ 3 all graphs are split.
8.3.3. Give examples of pairs of nonisomorphic split graphs with the same degree
sequences.
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.
8.3.12. Is it true that every split selfcomplementary graph has parts of equal size?
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
8.3.35. Using the graphs shown at Fig. 8.3.2 prove that the equality th(G) = th(G)
is not always true.
* * *
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 .
134 8. Degree Sequences
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
Theorem 9.1.2 For every graph G, X(G) :::; 1 + max6(H), where the maximum
is over all induced subgraphs H of G.
135
O. Melnikov et al., Exercises in Graph Theory
© Springer Science+Business Media Dordrecht 1998
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 9.1.8 For every k 2 3 there exists a uniquely k-colorable graph without
complete subgraphs of order k.
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.
* * *
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
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.
9.1.5. 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
9.1.16. 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.
9.1.26. 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. Vertex Coloring 139
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.
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.
9.1.34. 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?
9.1.37. Prove that for k-coloring of a uniquely k-colorable graph the subgraph
induced by the union of any two color classes is connected.
140 9. Graph Colorings
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}.
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.
9.1.42. Prove that X(G - v) = X(G) - 1 for every vertex v of a 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.
9.1.48. 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.
9.1.52. 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
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.
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.
* * *
9.2.1. Find the chromatic polynomials of the graphs shown at Fig. 9.2.1.
142 9. Graph Colorings
* * *
9.3.1. Find all minimal edge colorings of graphs shown at Fig. 9.3.l. (Colorings
differ up to renaming the colors.)
9.3.4. 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?
9.3.14. 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?
Theorem 9.4.1 A map G is k-colorable if and only if its geometric dual graph is
vertex-k-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
* * *
9.4.1. What is the minimal number of colors to color the map shown at Fig. 9A.1?
9.4.2. Construct a map with minimal number of faces that cannot be colored in
less than four colors.
146 9. Graph Colorings
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.
9.4.8. 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.
9.4.11. 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.
9.4.16. Prove that the Hadwiger conjecture for n = 5 implies the Four-Color Con-
jecture.
Note: The converse is also true (the Wagner Theorem).
9.4.18. Prove that the chromatic number of the sphere is equal to the chromatic
number of the plane.
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.
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
* * *
148 9. Graph Colorings
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.
9.5.5. Prove that the following graphs are perfect: a) bipartite graphs; b) split
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?
9.5.10. 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. Perfect Graphs 149
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.
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?
9.5.1S. 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.
150 9. Graph Colorings
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.
9.5.27. Find a graph which is not quasitriangulated but every its component is
quasitriangulated.
Directed Graphs
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
O. Melnikov et al., Exercises in Graph Theory
© Springer Science+Business Media Dordrecht 1998
152 10. Directed Graphs
:l ••
(i ;rlO 7
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).
* * *
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
Figure 10.1.2: To Exercise 10.1.6
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.
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.
154 10. Directed Graphs
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.
(10.2.1)
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*
* * *
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.
10.2.5. 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?
156 10. Directed Graphs
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.
10.2.10. 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;
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.
10.2.14. 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. Reachability and Components 157
10.2.17. Prove that every unilateral digraph may be converted into a strong one
by the addition of a single arc.
a h
d
Figure 10.2.2: To Exercise 10.2.19
10.2.22. Prove that a digraph is contourless if and only if every its arc belongs to
a directed cut.
158 10. Directed Graphs
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.
10.2.28. 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.
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.
Prove the following bounds for the number n of vertices and the number 1
of leaves of a 2-3-tree of height h :
10.2.34. 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.
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 :
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
* * *
10.3.1. Find the matrices A(G), I(G), R(G) for the digraph shown at Fig. 10.3.1.
•6
4
10.3.3. How are the adjacency, incidence and reachability matrices of the opposite
<-
digraph G related with those of G?
10.3.4. Prove the following statements:
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
[:. o
.;,.]
162 10. Directed Graphs
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.
10.3.11. 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.
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.
Theorem 10.4.1 For a connected digraph G the following statements are equiv-
alent:
1. G is Eulerian;
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).
* * *
164 10. Directed Graphs
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.
10.4.4. Prove the Theorem 10.4.1.
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. Tours and Paths 165
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
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
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.
* * *
10.5.1. Draw all pairwise nonisomorphic tournaments of orders 2, 3, 4.
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.
10.5.7. Prove that a Hamiltonian tournament T has a vertex v such that the
tournament T - v is Hamiltonian.
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.
* * *
10.6.1. 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.
168 10. Directed Graphs
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.
posite digraph G .
10.6.14. Prove that a contourless digraph without parallel arcs has a single arc
base.
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'.
170 10. Directed Graphs
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
(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
173
O. Melnikov et al., Exercises in Graph Theory
© Springer Science+Business Media Dordrecht 1998
174 11. Hypergraphs
* * *
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?
11.1.21. 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*).
11.1.29. 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
11.1.35. 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?
* * *
11.2.1. Are the hypergraphs H l , H2 from Fig. 11.2.1 realizable by tree?
• • •
2 2
6 (i
• •
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
(.'"'
• •)
(. • •)
(. • •~
'-....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.
11.2.6. 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.
11.2.10. 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
183
184 Answers, Hints, Solutions
1.1.17. For even n the graph is unique, for odd n there are no such graphs.
1.1.20. Hint: Two non isomorphic graphs satisfying conditions a)-c) are shown at
Fig. A1.1.1.
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.
1.1.26. b: The number of pairwise non isomorphic cubic circulant graphs of order
6 is 2.
c: Yes.
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
186 Answers, Hints, Solutions
(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.
1.1.31. 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
1.1.34. 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?
1.1.37. a: I<3; b: The line graph of the tetrahedron is isomorphic to the octahe-
dron.
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.
1.2: Graphs: Basic Notions 187
12
23 15
1.2.5. a: Hint: Delete the "excessive" pieces of the walk. b: False; c: False.
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.
II AIINtx1
•
• •
Gs
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.
1.2.20. 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
1.2: Graphs: Basic Notions 189
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.
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.
1.2.29. 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
1.2.32. 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
1.2.37. 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.
* *
* *
a
+ +
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
192 Answers, Hints, Solutions
1.3.6. No.
1.3.11. No.
1.3.12. Hint: Prove that such graph must be bipartite. Therefore it must have
three pairwise nonadjacent vertices.
1.3.15. Hint: Consider the set of graphs which do not possess the given hereditary
property.
n~O m~O
(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.
194 Answers, Hints, Solutions
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.
, 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,
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.
a b
c d
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 •
n
(MA)ij = LMikAkj.
k=l
B = MsAM.-l. (A1.5.2)
In fact, (A1.5.2) is equivalent to BM. = M.A, i.e.,
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.
1.5.13. Solution: Let G be a connected graph of order n > 1. First, let us prove
that
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
0 0 0
:1,
1 0 0
1 1 1
I(C.) = [ :
0 0 1
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
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 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
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.
I.e.,
202 Answers, Hints, Solutions
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),
Similarly,
• • •
Figure A1.6.1: To Exercise 1.6.8
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}.
204 Answers, Hints, Solutions
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
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
then
=
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.
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.
is an isomorphism of graphs.
Answers to Chapter 2:
Trees
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
208 Answers, Hints, Solutions
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.
B
e-
io jo
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
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.
a b
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.
2.1.28. 2 or 3.
2.1: Trees: Basic Notions 211
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 .
• • • • • •
by HI, i = 1,2, ... , p = deg u, where el E EHf, see Fig. A2.1.5. Let us
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.
2.1.33. 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:
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
1 2
7 WJV 3
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.
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,
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).
216 Answers, Hints, Solutions
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
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
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.
mi ~ ni - 1, i = 1,2, ... , k,
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).
2.2.14. Hint: See Theorem 2.1.1.
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
2.2.23. 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.
220 Answers, Hints, Solutions
~~. n - 2 vertices
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.
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. =
Answers to Chapter 3:
Independence and
Coverings
223
224 Answers, Hints, Solutions
belongs to 3 faces of the dodecahedron and there are 12 faces, we may write
3ao(D) ~ 2·12.
3.1.5. No.
3.1.6.
a: ao(G) - 1 ~ ao(G - v) ~ 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.
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
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,
q < k, (A3.1.2)
for otherwise
Similarly, 21EG l i = Pl (2ql +(k+1)(Pl -1)). Taking into account the relations
(A3.1.3), (A3.1.4), we obtain
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
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.
3.1.34. 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.
(A3.1.6)
It must be:
(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.
1 1
A= [ 0 0 1 1
1 0] ' B=
3.2: Coverings 231
4
•
Figure A3.1.5: To Exercise 3.1.43
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.
232 Answers, Hints, Solutions
3.2.19. Hint: Use the fact that the deletion of an end of an edge leads to the
deletion of the edge itself.
3.2.20. Hint: Use the Exercise 3.2.19.
3.2.22. The statement is false. Hint: Consider, e.g., KI,n, n 2: 2.
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.
234 Answers, Hints, Solutions
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
(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.
236 Answers, Hints, Solutions
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.
3.5.2. 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.
3.5.5. Hint: Use the Theorem 3.5.3.
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.
3.5.11. 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
Connectivity
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
242 Answers, Hints, Solutions
• • • • • • •
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,
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.
m k
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.
4.2.8. 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 :
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.
4.2.21. k = 1,2.
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
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:
248 Answers, Hints, Solutions
4.2.28. Cn.
4.2.29. Kn,n+l.
4.2.30. K3,3.
4.2.31. 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·
4.3.4. 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,
Matroids
251
252 Answers, Hints, Solutions
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
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
5.2.6. 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}}.
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
254 Answers, Hints, Solutions
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.
5.2.16. 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.
5.2.21. Hint: The cuts of G are the cocycles of the matroid M(G).
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.
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
Planarity
a b c
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.
257
258 Answers, Hints, Solutions
K5 -e
6.1.7. Hint: Apply the Euler formula to each component counting the external
face only once.
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).
6.1.20. 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
D
Figure A6.1.4: To Exercise 6.1.26
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.
6.1.33. G 1 .
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. Embeddings of Graphs. Euler Formula 261
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.
6.1.40. 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
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.
262 Answers, Hints, Solutions
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
6.3. Planarity Criteria 263
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.
....------..;.i
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.
6.3.7. 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
6.3.12. 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.
6.3.15. Not always. At Fig. A6.3.3 G 1 , G 2 are homeomorphic but L(G 1 ), L(G2)
are not homeomorphic.
[01><1
Figure A6.3.4: To Exercise 6.3.17
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 •
6.3.27. 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.
6.3.29. 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.
1) l{vEVG: degv=3}1~3;
2) G has a subgraph [{3,3.
6.4.5. Hint: Proof is by contradiction using the previous exercise and the Theorem
6.4.2.
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.
268 Answers, Hints, Solutions
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
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
Figure A6.4.2: To Exercise 6.4.18
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
6.4.23. 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
6.4.24. Hint: This follows from the equality c) of the previous exercise.
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.
6.5. Measures of Displanarity 271
5 1 2 3 4 5
9e-______~6----~.8 9
10
5 1 2 3 4 5
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
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.
272 Answers, Hints, Solutions
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).
e'
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.
6.5.16. 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.
6.5.23. 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.
Answers to Chapter 7:
Graph Traversals
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.
7.1.10. 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) .
275
276 Answers, Hints, Solutions
7.2 Hamiltonian Graphs
7.2.2. No.
7.2.3. No.
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. Hamiltonian Graphs 277
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.
7.2.19. 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.
7.2.25. 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.
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.
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
Xk Xk+l X n -2 V
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
Degree Sequences
I
N
I
N
I I
~O/
o
G sG
• •
G5
Figure A8.1.3: To Exercise 8.1.16
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.
8.1.20. No.
8.1.21. Yes. Hint: Consider the Cartesian product of realizations of the sequences
d and c.
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
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.
8.1.40. 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.
286 Answers, Hints, Solutions
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.
8.2.10. 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.
Further,
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 )}.
8.2.25. 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.
8.3.5. n.
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.
8.3.11. 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.
8.3.29. 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.
8.4.6. No.
294 Answers, Hints, Solutions
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.
a b
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 .
8.4.14. 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.
Answers to Chapter 9:
Graph Colorings
v
see Fig. A9.1.1.
•
H3
297
298 Answers, Hints, Solutions
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.
9.1.10. 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
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.
300 Answers, Hints, Solutions
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.
9.1.30. 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
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.
9.1.33. G l and G3 .
9.1.34. Yes.
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
9.1.40. Yes.
9.1.41. Hint: A non complete graph with minimal coloring has a pair of vertices of
the same color.
9.1.43. 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.
9.1.50. a: Yes. b: No; Fig. A9.1.4 shows a 4-critical graph with X(G - e) = 4.
e
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'. =
304 Answers, Hints, Solutions
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.
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.
9.4.3. 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.
9.4.7. 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).
308 Answers, Hints, Solutions
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.
9.4.12. 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
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.
9.5. Perfect Graphs 309
9.4.19. Hint: Fig. A9.4.2 shows a map on the torus that requires at least 7 colors.
t t
)
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).
9.5.6. Hint: Use the Exercise 9.5.5a and the Corollary 9.5.2.
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.
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
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.
9.5.15. 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.
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
9.5. Perfect Graphs 311
Directed Graphs
10.1.3. 2n2.
10.1.12. Hint: Consider, for example, the sequences (3,2,1) and (3,3,0).
313
314 Answers, Hints, Solutions
10.2.9. Hint: To prove the sufficiency, extract a walk (semiwalk) connecting the
required vertices from the existing spanning walk (semiwalk).
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.
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.
•
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).
316 Answers, Hints, Solutions
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
318 Answers, Hints, Solutions
(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
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.
10.3.5. 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.
320 Answers, Hints, Solutions
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.
10.4.4. 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.
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.
= =
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
10.5 Tournaments
10.5.1. See Fig. AlO.5.l.
10.5.2. Hint: Use the Theorem 10.4.3.
10.5.3. Hint: Use the Exercise 10.5.2.
10.5.4. Hint: Use the Exercise 10.5.2.
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.
324 Answers, Hints, Solutions
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
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.
vEVT vEVT
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.
10.6.12. 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. Base and Kernel 327
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:
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.
328 Answers, Hints, Solutions
@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.
10.6.31. 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.
a b + c ± d
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.
oe-----i..--___e 0
oe<------4.._-_eo
2
b
010
~
...
Hypergraphs
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.
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.
11.1.7. 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.
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
11.1.32. 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.
11.2.2. 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
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.
11.2.7. 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
11.2.10. 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.
4 5
8 9
e---------~~---e
7
Figure A11.2.3: To Exercise 11.2.20
n n n
k-i k
[3] Evstigneev V .A., Melnikov 1.S. Problems and Exercises in Graph Theory and
Combinatorics, Novosibirsk University Pub!., Novosibirsk, 1981 (in Russian).
[6] Melnikov 0., Tyshkevich R., Yemelichev V., and Sarvanov V. Lectures on
Graph Theory, BI-Wissenschaftsverlag, Mannheim, 1994.
[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
d( G) - diameter of graph G
352 Notations
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
P( G) - deck of graph G
r( G) - radius of graph G
t( G) - thickness of graph G
Z2 - two-element field
,( G) ~ genus of graph G