You are on page 1of 35

MAST30011 Graph Theory

Practical Class 6

PS4-2
The following network N has source x and sink y with arc capacities as shown.

( '

#
) $
%

& !
#
#
"

(a) Use the labelling algorithm to find a maximum flow f in N . Show at least three stages of the
algorithm, and state the value of the maximum flow.
(b) Find a minimum cut in N .
Answer: See the working (without explanation) on a separate page. The numbers in brackets are the
labels of the corresponding vertices, the numbers in blue colour show the flow values of the corresponding
arcs, and the numbers in circles indicate the order of vertices labelled.
The value of a maximum flow is 11, and S = {x} produces a minimum cut @ + (S) which is the set of
two outgoing arcs from x.
Exercise: Give details by yourself.

1
PS4-4

Let N be the network shown below, where each arc has unlimited capacity. Note that the underlying
digraph of N is weakly connected (i.e. there exists a path between any two vertices) and that every
vertex, except possibly the source and sink, has positive indegree and positive outdegree. Show that
there is no flow in N such that the flow along every arc is positive.

N:
s t

Hint: Apply the Conservative Condition to each vertex. Consider the unique arc incident to s. Let the
flow on this arc be a > 0. Apply the Conservative Condition to the head of this arc, and so on.
Exercise: Give a complete proof by yourself including all details.

PS5-3

(Mantel’s Theorem 1907) A graph is triangle-free if it contains no K3 -subgraph. What is the maximum
number of edges in a triangle-free graph with n vertices?
Answer the same question for graphs with no Kt -subgraph for any integer t 3 (Turán’s Theorem).
Solution: The maximum number of edges in a triangle-free graph with n vertices is equal to bn2 /4c.
And the only triangle-free graph with bn2 /4c edges is the complete bipartite graph Kbn/2c,dn/2e . (Note
that bn/2cdn/2e = bn2 /4c.)
Proof: Let G be a triangle-free graph with n vertices. Let be the maximum degree of G. Let v be a
vertex of G with degree . Let S = V (G) \ N (v). Then S is the set of vertices not adjacent to v. Note
that v 2 S and |S| = n . Since G is triangle-free, no two vertices in S are adjacent, for otherwise
two adjacent vertices in S together with v would induce a copy of K3 . Hence every edge of G is incident
to at least one vertex in S (that is, every edge of G is incident to one vertex or two vertices in S). Since
G has maximum degree , we obtain that the size m(G) of G satisfies
X
m(G)  deg(x)  |S| · = (n ) .
x2S

Moreover, m(G) = (n ) holds if and only if every vertex in S has degree and every edge
of G is incident to exactly one vertex in S. In other words, m(G) = (n ) holds if and only if no
two vertices in S are adjacent and every vertex in S is adjacent to exactly vertices in N (v). Since
|N (v)| = , we obtain further that m(G) = (n ) holds if and only if G is a bipartite graph with
bipartition {N (v), S} such that every vertex in S is adjacent to all vertices in N (v).
Note that 0   n 1 and for in this range the function (n ) achieves maximum when
= bn/2c (prove this by yourself). Therefore,

m(G)  (n )  (n bn/2c)bn/2c = dn/2ebn/2c = bn2 /4c.

Moreover, the discussion above implies that m(G) = bn2 /4c holds if and only if G = Kbn/2c,dn/2e .
Turán’s Theorem: Let G be a simple graph of order n which contains no Kt , where t 2. Then
|E(G)|  |E(Tt 1,n )|, with equality if and only if G ⇠
= Tt 1,n , where Tt 1,n is the complete (t 1)-partite
graph on n vertices in which all parts are as equal in size as possible. (Tt 1,n is called a Turán graph.)
A proof of Turán’s Theorem can be found in most textbooks on graph theory, e.g. Theorem 19.2
(pp. 504-505) in [CLZ] and Theorem 5.2.3 in [D. West, Introduction to Graph Theory, 2nd ed., Prentice
Hall, 2001].

2
PS5-4

Let G be a connected graph with at least three vertices. Let G0 be the graph obtained from G by adding
an edge joining x, y whenever dG (x, y) = 2. Prove that G0 is 2-connected.
Proof: By the definition of G0 , G is a spanning subgraph of G0 . So pG0 (u, v) pG (u, v) for any pair of
distinct vertices u, v of G. By the definition of the connectivity of a graph, we then have

(G0 ) = min{pG0 (u, v) : u, v 2 V (G), u 6= v}


min{pG (u, v) : u, v 2 V (G), u 6= v}
= (G).

Thus, if G is 2-connected, then (G0 ) (G) 2 and hence G0 is 2-connected as well.


Assume that G is connected but not 2-connected. As G has order at least three, it follows that
G contains at least one cut-vertex. Consider an arbitrary cut-vertex v of G. Let G1 , . . . , Gt be the
connected components of G v, where t 2 as v is a cut-vertex. Then, for any two distinct i, j between
1 and t and any neighbours ui , uj of v in Gi , Gj respectively, ui and uj are at distance 2 in G and hence
must be adjacent in G0 . This implies that V (Gi ) and V (Gj ) are in the same component of G0 v.
(Exercise: Give a detailed proof of this statement by yourself.) Since this is true for any i 6= j,
it follows that G0 v is connected. In other words, v is not a cut-vertex of G0 . On the other hand, since
G is a spanning subgraph of G0 , if a vertex is not a cut-vertex of G, then it cannot be a cut-vertex of G0 .
So G0 contains no cut-vertex. Therefore, G0 is 2-connected.
Second proof: Consider an arbitrary pair of distinct vertices v, w of G. Since G is connected, there
exists a path between v and w, say,
v 1 , v2 , v3 , . . . , v k ,
where v = v1 , vk = w. If k is odd, then

v 1 , v3 , v 5 , . . . , v k 2 , vk

and
v 1 , v2 , v4 , v6 , . . . , v k 1 , vk

are internally disjoint paths between v and w in G0 .


If k is even, then
v 1 , v 3 , v5 , . . . , v k 3 , vk 1 , vk

and
v 1 , v2 , v4 , v6 , . . . , v k 2 , vk

are internally disjoint paths between v and w in G0 .


So we have proved that there are two internally disjoint paths between each pair of distinct vertices.
Therefore, G0 is 2-connected.

PS5-6

For each integer ` 1, give an example of a graph G` with connectivity (G` ) = 1 and edge-connectivity
(G` ) = `.
Solution: There are many examples of such graphs. We give one example here.
Let G be the graph with vertex set { `, ` + 1, . . . , 1, 0, 1, . . . , ` 1, `} such that { `, ` +
1, . . . , 1, 0} and {0, 1, . . . , ` 1, `} induce K`+1 and there is no edge between any vertex in { `, ` +
1, . . . , 1} and any vertex in {1, . . . , ` 1, `}. We claim that (G) = 1 and (G) = `.
Since the path `, ` + 1, . . . , 1, 0, 1, . . . , ` 1, ` is a spanning path, there is a path between any two
vertices in G. Hence G is connected. On the other hand, by the construction of G, we see that vertex 0
is a cut-vertex. Hence (G) = 1.
We now prove (G) = `.

3
First, @({1, . . . , `}) = {0i : 1  i  `} is an edge cut of G with ` edges. Hence (G)  ` by the edge
version of Menger’s Theorem.
To prove that (G) `, it suffices to construct ` edge-disjoint paths between each pair of distinct
vertices. In fact, for 0  i < j  `, the paths i, k, j, where k 2 {0, 1, . . . , `} \ {i, j}, plus the edge ij
form ` edge-disjoint paths between i and j. Similarly, we can show that there are ` edge-disjoint paths
between i and j for `  i < j  0.
If `  i < 0 < j  `, then, as shown above, there are ` edge-disjoint paths between i and 0, say,
P1 , . . . , P` , and there are ` edge-disjoint paths between 0 and j, say, Q1 , . . . , Q` . The concatenation of
Pt and Qt (that is, the path from i to 0 along Pt and then from 0 to j along Qt ) is a path between i and
j, for 1  t  `, and these ` paths are edge-disjoint.
In summary, we have proved that there are ` edge-disjoint paths between each pair of distinct vertices.
Hence (G) `. This together with (G)  ` implies (G) = `.

4
MAST30011 Graph Theory
Practical Class 1

PS1-1

Draw the multigraph that represents the Hampton Court Maze shown below. Have one vertex for the
start, one vertex for the destination, one vertex for each location where there are three di↵erent places
to walk, and one vertex for each dead-end. How many vertices are there. Have one edge for each passage
that joins two locations. How many edges are there? Write down the degree of each vertex. What is the
degree sequence of the graph? How many paths are there from the start to the destination?

Answer: The graph has 15 vertices and 15 edges. The degree sequence is

3, 3, 3, 3, 3, 3, 3, 2, 1, 1, 1, 1, 1, 1, 1.

There are two paths from the start to the destination.

PS1-2

How many edges are there in each of the following graphs: (a) the complete graph K10 (b) the complete
bipartite graph K4,9 (c) the hypercube Q4 (d) the wheel W8 (e) the Petersen graph?
10
Answer: (a) 2 = 45 (b) 4 · 9 = 36 (c) 12 (24 · 4) = 32 (d) 16 (e) 15

PS1-4

Let G be a graph which contains m vertices of degree m and n vertices of degree n. Prove that if G
contains an odd vertex, then every vertex of G is odd.
Proof: By the assumption and the Handshaking Theorem, we have
X
2|E(G)| = deg(v) = m2 + n2 .
v2V (G)

Suppose that G contains an odd vertex. Then at least one of m and n is odd. Since 2|E(G)| is even, the
equation above implies that both m and n must be odd. Hence every vertex of G is odd.

PS1-5

Prove that every set of six people contains (at least) three mutual acquaintances or three mutual strangers.
Is it true that every set of five people has this property? How can this result be generalised?
Proof: Construct a graph G such that each of the six people corresponds to a vertex and two vertices
are adjacent if and only if the two persons are acquaintances. Obviously, this graph has order 6.

1
Now the statement in the question can be restated as follows: In G either there are three vertices
which are mutually adjacent (that is, a cycle of length 3) or there are three vertices which are mutually
nonadjacent.
Fix a vertex v of G. Denote
V1 = N (v) (set of acquaintances of v)
V2 = V (G) (N (v) [ {v}) (set of strangers of v).
Then V1 [ V2 = V (G) \ {v} and V1 \ V2 = ;.
Since |V1 | + |V2 | = 5, either |V1 | 3 or |V2 | 3.
Case 1. Assume |V1 | 3 first.
If there exists a pair of adjacent vertices in V1 , say, u, w, then {v, u, w} is a triangle in G.
If any two vertices of V1 are nonadjacent, then since |V1 | 3 there are three vertices in V1 which are
mutually nonadjacent.
Case 2. Assume |V2 | 3 next.
If there exists a pair of nonadjacent vertices in V2 , say, u, w, then v, u, w are mutually nonadjacent.
If such a pair does not exist, then any two vertices of V2 are adjacent. Since |V2 | 3, this implies
that there are three vertices in V2 which are mutually adjacent.
In each case above, we have either a triangle or a set of three mutually nonadjacent vertices in G.
It is not true that every set of five people contains (at least) three mutual acquaintances or three
mutual strangers. That is, there exists at least one edge-colouring of K5 using two colours such that
there is no monochromatic triangle. For example, one may assign red colour to all edges on a 5-cycle in
K5 and assign blue colour to the remaining edges. Then there is no red or blue triangle in K5 .
Generalization of PS1-5: Ramsey theory
The result of PS1-5 can be restated as follows: If we colour the edges of a complete graph of order 6 by
two colours, say, red and blue, then there is either a red triangle or a blue triangle.
In general, Ramsey (1930) proved the following result: Let s, t 3 be fixed integers. If n is sufficiently
large, then when colouring the edges of the complete graph Kn of order n by red and blue, there always
exists a red Ks (i.e. with all edges coloured red) or a blue Kt (i.e. with all edges coloured blue).
The smallest integer n such that this holds is called the Ramsey number and is denoted by R(s, t).
It is notoriously difficult to determine R(s, t) for given integers s and t. Currently, this number is know
only for some small integers s and t. See: https://en.wikipedia.org/wiki/Ramsey\%27s_theorem
“Erdös asks us to imagine an alien force, vastly more powerful than us, landing on Earth and de-
manding the value of R(5, 5) or they will destroy our planet. In that case, he claims, we should marshal
all our computers and all our mathematicians and attempt to find the value. But suppose, instead, that
they ask for R(6, 6). In that case, he believes, we should attempt to destroy the aliens.” (Joel Spencer)
PS1-18
Let G be a graph. Let S(G) be the subdivision of G obtained by replacing each edge e = uv of G by a
new vertex ve , and joining ve to u and v. Show that if G is any nontrivial graph, then S(G) is bipartite.
Proof 1: Recall that a graph is bipartite if and only if it has no cycle of odd length (i.e. odd cycle).
For each edge e = uv of G, let e0 = uve and e00 = vve denote the two edges of S(G) obtained by
subdividing e. Since ve has degree 2 in S(G), if a cycle contains one of e0 and e00 , it must contain both
of them. This implies that the edges of a cycle of S(G) come in pairs. Hence any cycle of S(G) has an
even length. In other words, S(G) does not contain any odd cycle. Hence S(G) is bipartite.
Proof 2: Let X be the set of vertices in the original graph G, and Y the set of new vertices which
subdivide edges of G. That is,
X = V (G), Y = {ve : e 2 E(G)}.
Then X [ Y = V (S(G)) and X \ Y = ;. In other words, {X, Y } is a partition of V (S(G)).
By the definition of S(G), any two vertices in X are not adjacent in S(G) and any two vertices in Y
are not adjacent in S(G). Hence S(G) is a bipartite graph with bipartition {X, Y }.

2
MAST30011 Graph Theory
Practical Class 2

PS1-20

Let G be a graph with the property that every edge joins an even vertex and an odd vertex. Prove that
G is bipartite.
Proof: Let X denote the set of even vertices of G and Y the set of odd vertices of G. Since every vertex
is either even or odd but not both, {X, Y } is a partition of V (G) (that is, X [ Y = V (G) and X \ Y = ;).
By the assumption every edge of G joins a vertex in X and a vertex in Y . Thus, by the definition of
a bipartite graph, G is bipartite with bipartition {X, Y }.

PS1-24

Show that the following two graphs are not isomorphic.


g
a f e f
g
b e d
h c
d h
c a b

Proof: There are several ways to prove that the two graphs are non-isomorphic.
Proof 1: The graph on the left-hand side contains three 3-cycles (namely, (a, g, h, a), (a, b, h, a) and
(c, b, h, c)) such that one of them has exactly one common edge with each of the other two 3-cycles, while
the graph on the right-hand side does not have three 3-cycles with this property. So the two graphs
cannot be isomorphic to each other.
Proof 2: Denote by G and H the graphs on the left and on the right, respectively. Note that h is the
only maximum degree vertex of G and e is the only maximum degree vertex of H. So if there is an
isomorphism from G to H, it must map h to e and therefore must map the neighbourhood NG (h) of h
in G to the neighbourhood NH (e) of e in H. This implies that the set of degrees of the neighbours of h
in G is equal to the set of degrees of the neighbours of e in H. However, the former is {3, 3, 3, 3, 4} while
the latter is {2, 3, 3, 3, 3}, so the two sets are not equal. This contradiction shows that the two graphs
cannot be isomorphic to each other.

PS1-26

Draw all the nonisomorphic 3-regular graphs of order 6 (hint: Consider their complements).
Solution: Note that two graphs are isomorphic if and only if their complements are isomorphic. So,
if we know all the nonisomorphic complements of 3-regular graphs of order 6, then we know all the
nonisomorphic 3-regular graphs of order 6.
The complement of any 3-regular graph of order 6 is a 2-regular graph of order 6. Since every 2-regular
graph is a vertex-disjoint union of cycles, there are exactly two 2-regular graphs of order 6, namely C6
and C3 [ C3 . Taking complements of them, we obtain exactly two 3-regular graphs of order 6, which are
K3,3 and K3 ⇤K2 .

1
PS1-29

How many distinct graphs are there with vertex set {1, 2, . . . , n}?
How many distinct graphs are there with vertex set {1, 2, . . . , n} and m edges?
For these questions, what happens if “with vertex set {1, 2, . . . , n}” is replaced by “with n vertices”?
Solution: For each pair of distinct vertices v and w, either the edge vw is present or not present. So
there are two choices for each pair of distinct vertices, and for di↵erent pairs of distinct vertices the
n
choices are made independently. Each set of choices gives a distinct graph. So there are 2( 2 ) distinct
graphs with vertex set {1, 2, . . . , n}.
n
Similarly, there are (m2) distinct graphs with vertex set {1, 2, . . . , n} and m edges.
The first two questions are about enumeration of labelled graphs, while the third question is about
enumeration of unlabelled graphs. The di↵erence is that in the labelled case isomorphic graphs whose
vertices are labelled di↵erently are counted as di↵erent graphs, while in the unlabelled case all graphs
which are isomorphic to each other are counted once. The latter is not elementary at all.
Pólya observed that the number of unlabelled graphs with order n and the number of unlabelled
n n
graphs with order n and size m are asymptotically 2( 2 ) /n! and (m
2 ) /n! as n ! 1, respectively.

PS1-30

Show for each integer n 2 that, up to isomorphism, there is exactly one bipartite graph of order n
having size bn2 /4c. Show further that this graph achieves the maximum size among all bipartite graphs
of order n.
Proof: Let G be an arbitrary bipartite graph of order n. Denote its bipartition by {X, Y } and set
x = |X|. Then 1  x  n 1 and |Y | = n x. The size m(G) of G satisfies

m(G)  x(n x)

and equality holds if and only if G ⇠


= Kx,n x (that is, every vertex in X is adjacent every vertex in Y ).
Using calculus, we can show (exercise!) that in the interval [1, n 1] the quadratic function x(n x)
has a unique maximum point, namely x = n/2. So

n2
m(G)  x(n x)  .
4
(This can also be proved as follows: Using the inequality 2ab  a2 + b2 , we obtain m(G)  x(n x) =
p p ⇣ p 2 p 2
⌘2 2
( x n x)2  ( x) +(2 n x) = n4 .)
Since m(G) is an integer and bn2 /4c is the largest integer no more than n2 /4, from m(G)  x(n x) 
n2
4 it follows that
m(G)  bn2 /4c.

If n is even, then m(G)  n2 /4 = bn2 /4c = m(Kn/2,n/2 ). Hence Kn/2,n/2 is the unique bipartite
graph of order n with maximum possible size.
Assume that n is odd. Since the function x(n x) is increasing in [1, n/2] and decreasing in [n/2, n 1],
it has two integer maximum points in [1, n 1], namely x = (n 1)/2 = bn/2c and x = (n+1)/2 = dn/2e,
and the values of the function at both points are equal to (n 1)(n + 1)/4 = (n2 1)/4 = bn2 /4c. Thus,
for any integer x in [1, n 1],

x(n x)  bn2 /4c = m(K(n 1)/2,(n+1)/2 ),

and K(n 1)/2,(n+1)/2 is the unique bipartite graph of order n with maximum possible size.
In summary, Kbn/2c,dn/2e is the unique bipartite graph of order n with maximum size, and its size is
equal to bn2 /4c. The uniqueness implies that, up to isomorphism, Kbn/2c,dn/2e is the unique bipartite
graph of order n having size bn2 /4c.

2
MAST30011 Graph Theory
Practical Class 3

PS1-32
Prove that the spectrum of Kn is ((n 1)1 , ( 1)n 1
).
Proof: We have A = A(Kn ) = Jn In , where In and Jn are the identity matrix and all-1 matrix of
order n, respectively. Set
n (x) = Kn (x) = det(xIn A).
Then
x 1 1 ··· 1 1
1 x 1 ··· 1 1
1 1 x ··· 1 1
n (x) = .
···
1 1 1 ··· x 1
1 1 1 ··· 1 x

Subtracting the first row from each of the other n 1 rows and expanding along the last row, and
then expanding the first determinant along the last column, we obtain
x 1 1 ··· 1 1
x 1 x+1 0 ··· 0 0
x 1 0 x+1 ··· 0 0
n (x) =
···
x 1 0 0 ··· x+1 0
x 1 0 0 ··· 0 x+1
1 1 ··· 1 1
x+1 0 ··· 0 0
= ( 1)n+1 ( x 1) 0 x + 1 ··· 0 0
···
0 0 ··· x + 1 0
+(x + 1) n 1 (x)
= ( 1)n+1 ( x 1)( 1)n ( 1)(x + 1)n 2
+ (x + 1) n 1 (x)
= (x + 1)n 1 + (x + 1) n 1 (x).

Applying this relation recursively, we obtain

n (x) = (x + 1)n 1 + (x + 1) n 1 (x)


= (x + 1)n 1 + (x + 1){ (x + 1)n 2
+ (x + 1) n 2 (x)}
= ···
= (x + 1)n 1 (x n + 1).
Setting n (x) = 0, we obtain that the spectrum of Kn is
((n 1)1 , ( 1)n 1
).

Second proof: This proof is based on the following known result from linear algebra: If is an eigenvalue
of a square matric M with multiplicity m and f (x) is polynomial, then f ( ) is an eigenvalue of f (M )
with multiplicity m.

1
We have A = A(Kn ) = Jn In , where In and Jn are the identity matrix and all-1 matrix of order n,
respectively. So A = f (Jn ) is a polynomial of Jn , where f (x) = x 1. Thus the eigenvalues of A can be
obtained from that of Jn .
A non-zero column vector x = (x1 , x2 , . . . , xn )T is an eigenvector of Jn with eigenvalue if and only
if Jn x = x, which is true if and only if

x1 + x2 + . . . + xn = xi , for i = 1, 2, . . . , n.

Setting = 0, this system of linear equations is satisfied by all vectors x satisfying x1 +x2 +. . .+xn = 0.
Since the solution space of the equation x1 + x2 + . . . + xn = 0 has dimension n 1, it follows that 0 is
an eigenvalue of Jn with multiplicity n 1.
On the other hand, setting = n, the above system of linear equations is satisfied by all vectors x
satisfying x1 = x2 = . . . = xn . Since the solution space of this system has dimension 1, it follows that n
is an eigenvalue of Jn with multiplicity 1.
Since A = Jn In , we obtain that 0 1 = 1 is an eigenvalue of A with multiplicity n 1, and n 1
is an eigenvalue of A with multiplicity 1. In other words, the spectrum of Kn is

((n 1)1 , ( 1)n 1


).

PS1-33

Let P3 be the path of length 2. Compute the characteristic polynomial P3 (x) and determine the
spectrum of P3 .
p p
Solution: P3 (x) = x3 2x. Spectrum: ( 2, 0, 2).
(Exercise: Give details.)

PS2-1

Prove that if |V (G)| 2 and G is disconnected then G is connected.


Proof: Let H be a connected component of G. Denote

X := V (H), Y := V (G) V (H).

Since G is disconnected, we have X 6= ; and Y 6= ;. Since H is a component of G, there is no edge of G


between X and Y . Hence any vertex of X is adjacent to any vertex of Y in G.
For any x, x0 2 X, the sequence x, y, x0 is an (x, x0 )-path in G, where y is any vertex in Y . Similarly,
for any y, y 0 2 Y , the sequence y, x, y 0 is a (y, y 0 )-path in G, where x is any vertex in X.
We have proved that there exists a path of G between any two vertices. Hence G is connected.

PS2-2

Recall that (G) is the minimum degree of a vertex of G.

(a) Prove that every graph G with order n and (G) (n 1)/2 is connected.
(b) Prove that this result is best possible by finding, for every even integer n 2, a disconnected graph
G of order n with (G) = (n 2)/2.
First proof of (a): Suppose to the contrary that G is disconnected. Let v 2 V (G), and let S be the
set of vertices in the component of G containing v. Then v cannot be adjacent to any vertex outside S.
Also, v cannot be adjacent to itself. So
d(v)  |S| 1.

2
On the other hand, choose a vertex u not in S. Then u cannot be adjacent to any vertex in S. And
u cannot be adjacent to itself. Hence
d(u)  n |S| 1.

Combining the two inequalities above, we obtain

d(u) + d(v)  n 2.

Thus min{d(u), d(v)}  (n 2)/2 < (n 1)/2, but this contradicts the hypothesis that (G) (n 1)/2.
Hence G is connected.
Second proof of (a): Suppose to the contrary that G is disconnected. Take G1 to be a connected
component of G with smallest order. Since G has order n, by the choice of G1 , the order of G1 is at
most bn/2c. That is, ( n
2 , if n is even
|V (G1 )|  n 1
2 , if n is odd.
Each vertex in V (G1 ) has degree  |V (G1 )| 1 in G. So the degrees of the vertices of G1 are at most
( n
2 1 = n 2 2 , if n is even
n 1 n 3
2 1= 2 , if n is odd.

This contradicts the assumption (G) (n 1)/2. Therefore, G must be connected.


Third proof of (a): By the definition of a connected graph, it suffices to prove that there exists at
least one path in G between any two vertices of G.
Let u and v be any two distinct vertices of G. If u and v are adjacent, then they are joined by a path
of G, namely the edge uv.
Assume that u and v are non-adjacent. Then u 62 N (v) and v 62 N (u). We have

|N (u)| = d(u) (G) (n 1)/2

|N (v)| = d(v) (G) (n 1)/2.

If N (u) \ N (v) = ;, then N (u), N (v) and {u, v} are pairwise disjoint subsets of V (G). Hence

n = |V (G)| 2 + |N (u)| + |N (v)| 2 + (n 1)/2 + (n 1)/2 = n + 1.

This contradiction shows that N (u) \ N (v) 6= ;. So we can take a vertex w 2 N (u) \ N (v). Thus the
sequence u, w, v is a path between u and v.
We have proved that there is at least one path in G between any two vertices. Therefore, G is
connected.
(b) Let r = n/2 and G = Kr [ Kr (the union of two disjoint copies of Kr ). Then G is disconnected
with order n and (G) = (n 2)/2.

PS2-4

Let G be the following weighted graph.

a b c
1 4

13 6
8 5 7
d 6 8
e f
11
3 5
g

3
Apply Dijkstra’s algorithm to find the distance from a to all other vertices of G. State what the labels of
vertices are when each new vertex is chosen to add to the set S. Finally, give a tree determining shortest
paths from a to all other vertices of G.
Solution: By following Dijkstra’s algorithm, one can find a shortest path tree T with root a and edge
set
E(T ) = {ab, bc, be, bf, ad, dg}.
See the table below, where the first coordinate in each pair indicates the label and the second coordinate
indicates the parent of the corresponding vertex, and the last column lists vertices added to S in that
order. When a pair is highlighted, it means that the corresponding vertex has received a permanent
label and so its label will not be corrected in subsequent iterations.
Initialization: l(a) = 0 and l(b) = l(c) = l(d) = l(e) = l(f ) = l(g) = 1 and S := ;

a b c d e f g Added to S
(0, ) (1, ) (1, ) (1, ) (1, ) (1, ) (1, ) a
(1, a) (1, ) (8, a) (13, a) (1, ) (1, ) b
(5, b) (8, a) (6, b) (7, b) (1, ) c
(8, a) (6, b) (7, b) (1, ) e
(8, a) (7, b) (17, e) f
(8, a) (12, f ) d
(11, d) g

A shortest path from a to b: a, b, with length 1


A shortest path from a to c: a, b, c, with length 5
A shortest path from a to d: a, d, with length 8
A shortest path from a to e: a, b, e, with length 6
A shortest path from a to f : a, b, f , with length 7
A shortest path from a to g: a, d, g, with length 11

4
MAST30011 Graph Theory
Practical Class 4

PS3-2

Find four spanning trees in the graph on the right.

Answer: Since this graph has order 7 and size 9, any spanning tree of it has size 6 and can be obtained
by removing 3 edges while maintaining the connectedness. For example, we can remove 35, 46, 37 to
obtain the spanning tree T with E(T ) = {12, 16, 56, 36, 67, 74}. There are several other spanning trees
of this graph. (Find at least three other spanning trees by yourself!)

PS3-3

Prove that every graph of order n with m edges contains at least m n + 1 cycles.
Proof: We first assume that G is connected. Then it has at least one spanning tree.
Let T be a spanning tree of G. Then T contains exactly n 1 edges, and the remaining m (n 1)
edges of G are not in T . Each of these m (n 1) edges, say, e, gives rise to a unique cycle of G which
consists of e and the unique path in T between the two end-vertices of e. Note that e is the only edge
not in T on this cycle. Hence, for di↵erent edges e, e0 of G not in T , the cycles obtained this way from e
and e0 are di↵erent. Therefore, G contains at least m (n 1) = m n + 1 cycles.
So the result is true for connected graphs.
In general, suppose that G has k components G1 , . . . , Gk with n1 , . . . , nk vertices and m1 , . . . , mk
edges, respectively. By what we proved above, each Gi contains at least mi ni + 1 cycles. Of course
no two components have a common cycle. Therefore, G contains at least
k k
! k
!
X X X
(mi ni + 1) = mi ni + k = m n+k m n+1
i=1 i=1 i=1

cycles.
Remark: What we proved is a stronger result: Any graph with n vertices, m edges and k components
contains at least m n + k cycles.

PS3-4

Let T be a tree with maximum vertex degree 3. What is the relationship between the number of degree-1
vertices and the number of degree-3 vertices in T ? First, draw small examples to find the relationship.
Then prove your answer using the handshaking theorem. Then try to find an inductive proof.

1
Proof: Let ni be the number of degree-i vertices of T , for i = 1, 2, 3. Since T is a tree with maximum
degree 3, it is nontrivial and has no isolated vertices. So the order of T is given by n = n1 + n2 + n3 .
On the other hand, since T is a tree, its size is equal to n 1. So by the handshaking theorem we have
X
n1 + 2n2 + 3n3 = deg(v)
v2V (T )

= 2|E(T )|
= 2(n 1)
= 2n1 + 2n2 + 2n3 2.

Thus n3 = n1 2. That is, the number of degree-3 vertices is two less than the number of degree-1
vertices.
Proof by induction: Any tree T with maximum degree 3 has order at least 4, with each vertex having
degree 1, 2 or 3. Let ni (T ) be the number of degree-i vertices of T , for i = 1, 2, 3. Then T has order
n(T ) = n1 (T ) + n2 (T ) + n3 (T ).
We prove n1 (T ) n3 (T ) = 2 by induction on the order n(T ) of T .
If T is a tree with maximum degree 3 and order 4, then T is K1,3 and hence n1 (T ) n3 (T ) = 3 1 = 2,
as required.
Now suppose that the result is true for any tree with maximum degree 3 and order less than n, for
some n 4. Consider a tree T with maximum degree 3 and order n. Since every tree with order at
least two contains at least two leaves, we may take a leaf v of T . Then T v is a tree. Let u be the
unique neighbour of v in T . Since n 4, u cannot be a leaf for otherwise T would be disconnected, a
contradiction. Hence u has degree 2 or 3 in T .
Case 1: u has degree 2.
In this case u is a leaf of T v and T v has maximum degree 3. Hence n1 (T v) = n1 (T ),
n2 (T v) = n2 (T ) 1 and n3 (T v) = n3 (T ). By our hypothesis,

n1 (T v) n3 (T v) = 2.

Therefore,
n1 (T ) n3 (T ) = n1 (T v) n3 (T v) = 2.

Case 2: u has degree 3.


If u is the only degree-3 vertex of T , then n3 (T ) = 1 and every vertex of T v has degree 1 or
2. So T v is a path and hence has exactly two vertices of degree 2. Hence n1 (T ) = 3. So we have
n1 (T ) n3 (T ) = 2, as desired.
Now assume that T contains at least one degree-3 vertex other than u. Then T v is has maximum
degree 3 and hence n1 (T v) n3 (T v) = 2 by our hypothesis. Since u has degree 2 in T v,
we have n1 (T v) = n1 (T ) 1, n2 (T v) = n2 (T ) + 1 and n3 (T v) = n3 (T ) 1. Therefore,
n1 (T ) n3 (T ) = (n1 (T v) + 1) (n3 (T v) + 1) = n1 (T v) n3 (T v) = 2.
By mathematical induction, the proof is complete.
Remark: The counting argument above generalises for an arbitrary nontrivial tree T . Let be the
maximum degree of T . Let ni be the number of vertices with degree i, for i = 1, . . . , . By the
handshaking lemma,

X X ⇣ X ⌘
ini = deg(v) = 2|E(T )| = 2(n 1) = 2 ni 1 .
i=1 v2V (T ) i=1

Thus
X
(i 2)ni = 2.
i=1

2
Hence
X
(i 2)ni = n1 2.
i=3

In PS3-8 and PS3-9, G is the following weighted graph.

a 2 c 5 e

2 4
3 3
3
2 4
b d f
PS3-8

Use Kruskal’s algorithm to find a minimum spanning tree T of G. What is the weight of T ?
Solution:
Order Edge added to S Weight
1 ac 2
2 af 2
3 bd 2
4 ef 3
5 cd 3
The tree T with edges ac, af, bd, ef and cd is a minimum spanning tree, and its weight is 12. Another
minimum spanning tree is the one with edges ac, af, bd, ef and ab, and its weight is 12.
Second version of Kruskal’s algorithm: Order the edges of G according to a non-decreasing order
of their weights (breaking ties arbitrarily), and scan them one by one in this order. Maintain a spanning
forest beginning with the one without any edge. Accept an edge if adding it to the current forest does
not create cycles, and reject it otherwise. Stop when all edges have been scanned.

Edges ac af bd ab cd ef cf df ce
Weights 2 2 2 3 3 3 4 4 5
Accept (Yes or No)? Y Y Y Y N Y N N N
In this way we obtain the minimum spanning tree with edges ac, af, bd, ef and ab, and with weight
12.

PS3-9

Use the Prim-Jarnik algorithm to grow a minimum spanning tree Ta of G from the vertex a, and another,
Tb , from b.
Solution: Minimum spanning tree Ta from vertex a:

Vertices added to Ta Edges added to Ta Weight


a
c ac 2
f af 2
d cd 3
b db 2
e fe 3

3
The weight of Ta is 12.
Minimum spanning tree Ta from vertex b:

Vertices added to Tb Edges added to Tb Weight


b
d bd 2
a ab 3
c ac 2
f af 2
e fe 3

The weight of Tb is 12.

4
MAST30011 Graph Theory
Practical Class 5

PS3-1

Let
Pd1 , . . . , dn be positive integers (n 2). Prove that there is a tree with degrees d1 , . . . , dn if and only
if i di = 2n 2.
P
Proof: ()) If T is a tree with vertex degrees d1 , . . . , dn , then T has exactly n 1 edges and i di =
2|E(T )| by the handshaking theorem. Hence
X
di = 2|E(T )| = 2(n 1) = 2n 2.
i

(() We need to prove the following statement for n 2:


P
(*) If d1 , . . . , dn are n positive integers with i di = 2n 2, then there exists a tree with order n
and vertex degrees d1 , . . . , dn .
We prove this statement by induction on n 2.
If n = 2, then d1 = d2 = 1 and K2 is our desired tree.
Now assume that n > 2 and the statement (*) holds for values less than n.
P
Let d1 , . . . , dn be positive integers with i di = 2nP 2. Without loss of generality we may assume
d1 ··· dn . The average of d1 , . . . , dn is equal to ( i di )/n = 2 (2/n), which is strictly between
1 and 2 as n > 2. Thus dn = 1 and d1 > 1. Hence d1 1, d2 , . . . , dn 1 are n 1 positive integers.
Moreover, !
X n
(d1 1) + d2 + . . . + dn 1 = di 2 = 2(n 1) 2.
i=1

So by our induction hypothesis there is a tree T with order n 1 and degrees d1 1, d2 , . . . , dn 1 .


Let v be a vertex of T with degree d1 1. Construct a graph T 0 from T by adding a new vertex, say,
x, and a new edge joining x and v. Since T is a tree, T 0 is connected with no cycles. So T 0 is a tree.
Moreover, T 0 has order n and vertex degrees d1 , d2 , . . . , dn . By mathematical induction, this completes
the proof of (*) for all n 2.

PS3-10

Let G be a connected weighted graph. Suppose that G has two di↵erent minimum spanning trees. Prove
that G has at least two edges of the same weight.
Proof: Suppose that G has two distinct spanning trees, say, T1 and T2 . Then E(T1 ) 6= E(T2 ). Suppose
to the contrary that no two edges of G have the same weight. Then we can order the edges e1 , e2 , . . . , en 1
(where n = |V (G)|) of T1 so that

w(e1 ) < w(e2 ) < · · · < w(en 1 ).

Let i be the smallest subscript such that ei is not an edge of T2 . That is, e1 , . . . , ei 1 2 E(T1 ) \ E(T2 )
but ei 2 E(T1 ) E(T2 ). Let C be the unique cycle of T2 + ei created by adding ei to T2 .
There exists an edge f on C such that f 62 E(T1 ), for otherwise the whole cycle C would be in
T1 , which contradicts the assumption that T1 is a tree. Note that f 2 E(T2 ). Hence f 6= ei and so
w(f ) 6= w(ei ) by our assumption.

1
If w(ei ) < w(f ), then T2 + ei f is a spanning tree of G with weight w(T2 ) + w(ei ) w(f ) < w(T2 ),
which is impossible since T2 is a minimum spanning tree of G. Hence w(f ) < w(ei ).
Since f 62 E(T1 ), T1 + f contains a unique cycle, say, D. All edges of D other than f are edges of T1 .
There exists an edge ei⇤ on D such that ei⇤ 2 E(T1 ) E(T2 ) for otherwise the cycle D is contained in
T2 , which is a contradiction. Since T1 is a minimum spanning tree and T1 + f ei⇤ is a spanning tree,
we have w(T1 )  w(T1 ) + w(f ) w(ei⇤ ), which implies w(ei⇤ )  w(f ).
Combining this and w(f ) < w(ei ), we obtain w(ei⇤ ) < w(ei ). By the order of the edges of T1 , this
implies i⇤ < i. Thus ei⇤ is an edge of T1 not in T2 which has a smaller subscript than ei . However, this
contradicts the choice of i. Therefore, G must have two edges with the same weight.

PS3-11

Using the definition or a known characterization of a tree, prove the following statements:

(a) Suppose that G is not a complete graph. Prove that G is a tree if and only if adding any edge with
end-vertices in V (G) creates exactly once cycle.
(b) A graph G is a tree if and only if it is loopless (i.e. without any edge from a vertex to itself) and
has exactly one spanning tree.

Proof: (a) ()) Suppose that G is a tree. Then G has no cycles. Let u, v be distinct vertices of G
such that uv 62 E(G). Then there is a unique path in G joining u and v. Call this path Puv . So the
graph G + uv obtained by adding uv to G contains a cycle, namely the cycle consisting of uv and Puv .
Moreover, this is the unique cycle in G + uv because G has no cycle and so any cycle in G + uv must
contain uv and Puv . Hence adding any edge with end-vertices in V (G) creates exactly one cycle.
(() Suppose that G is not a complete graph and adding any edge with end-vertices in V (G) creates
exactly one cycle.
We claim first that G is connected. Suppose otherwise. Let G1 and G2 be two components of G.
Let x 2 V (G1 ) and y 2 V (G2 ). Then x, y 2 V (G), xy 62 E(G), and adding xy to G does not create any
cycle, which is a contradiction. Hence G must be connected.
Next we claim that G has no cycle. Suppose otherwise. Let C : v1 , v2 , . . . , vk , v1 be a cycle of G. For
any distinct vertices vi , vj on C, we should have vi vj 2 E(G) for otherwise adding vi vj to G creates two
cycles, which is a contradiction. (There are exactly two paths between vi and vj on C. Each of them
together with vi vj gives rise to a cycle.) In other words, any two vertices on C must be adjacent in G.
Since G is not a complete graph, there exists at least one vertex u outside C which is not adjacent
to at least one vertex on C. Without loss of generality we may assume that u 2 V (G) V (C) is not
adjacent to v1 2 V (C). Since G is connected, there exists a path between u and a vertex on C. Let P
be such a path with shortest length and assume that it connects u and vi . Then all internal vertices of
P are outside C by the minimality of the length of P .
Case 1: P has length 1. Then uvi 2 E(G) and so vi 6= v1 . Hence adding uv1 to G creates two cycles,
of which the first one is the concatenation of the path u, v1 , v2 , . . . , vi and P , and the second one is the
the concatenation of the path u, v1 , vk , vk 1 , . . . , vi and P .
Case 2: P has length at least 2. Then by the choice of P , u is not adjacent to any vertex on C. In
particular, u is not adjacent to vi 1 . So adding uvi 1 creates two cycles, namely the concatenation of
u, vi 1 , vi and P , and the concatenation of u, vi 1 , vi 2 , . . . , vi+1 , vi and P .
In each case we obtain a contradiction. This shows that G has no cycles.
In summary, we have proved that G is connected without cycles. Hence G is a tree.
(b) If G is a tree, then it is loopless and G itself is the unique spanning tree of G.
Conversely, suppose that G is loopless and has exactly one spanning tree T . If there exists an edge
in G but not in T , say, e, then adding e to T results in a subgraph of G which contains a cycle, namely
the fundamental cycle C(T, e). Deleting an edge on C(T, e) other than e results in a spanning tree of G
which is di↵erent from T , but this contradicts our assumption. Hence there exists no edge of G not in
T . In other words, G is identical to T and hence is a tree.

2
PS3-12

Prove that every tree other than a path with at least 4 vertices has at least 3 leaves (vertices of degree
1).
Proof 1: Let T be a tree with order n 4 which is not a path. Suppose to the contrary that T has at
most 2 leaves. Then it has at least n 2 vertices of degree 2. Hence, by the handshaking theorem,
X
2|E(T )| = deg(v) 2(n 2) + 1 + 1 = 2n 2.
v2V (T )

On the other hand, 2|E(T )| = 2(n 1) since T is a tree. So the inequality above is tight, which implies
that T has exactly two leaves, denoted by u and v, and all other vertices have degree exactly 2.
Let P denote the unique (u, v)-path in T . Since T is not a path, we have T 6= P , so there is at least
one vertex, say w, outside P . Since T is connected, there is a path from w to some vertex x on P . Let y
be the second last vertex on this path (when traversing from w to x). Then x and y are adjacent, and y
is outside P . If x is u or v, then u or v has degree 2, a contradiction. If x is neither u nor v, then it has
degree 3, which contradicts what we proved above (that is, every vertex other than u and v has degree
2).
Therefore, T has at least three leaves.
Proof 2: Let
P : u, x, . . . , y, v
be a longest path in T (i.e. P has the maximum length among all paths of T ), where u and v are the
first and last vertices of P , respectively. Since T has at least four vertices, P has length at least 1 and
so u 6= v.
We claim that x is the only vertex adjacent to u in T . Suppose otherwise. Let z 6= x be adjacent
to u. Then z is not a vertex on P , for otherwise T would contain a cycle, which is impossible since T
has no cycle at all. Thus adding vertex z and edge zu to P yields a path which is longer than P . This
contradicts the choice of P . Hence x is the only vertex adjacent to u in T and consequently deg(u) = 1.
Similarly, one can show that deg(v) = 1.
Since T is not a path, we have T 6= P . So there is at least one vertex outside P . Since T is connected,
any vertex outside P is joined to some vertex on P by a unique path in T . Choose a vertex w outside
P such that the unique path between w and a vertex on P has maximum possible length. Denote this
path by
Q : w, x, . . . , p,
where p 2 V (P ).
We claim that w is a leaf. Suppose otherwise. Then there exists a vertex z 6= x adjacent to w. For
this vertex z, the path
z, w, x, . . . , p
connects z and a vertex on P and it is longer than Q. This contradicts the choice of w. Hence w is a leaf.
Since w is not on P , it is clear that w is distinct from u and v. Hence T has at least three leaves.

3
MAST30011 Graph Theory
Practical Class 7

PS5-8
Show that if G is a connected graph of order at least 2 that contains a bridge, then either G has order
2 or G contains cut-vertices.
Proof: It suffices to prove the following equivalent statement: If a connected graph G has order at least
3 and contains a bridge, then G contains cut-vertices.
Let G be a connected graph with order at least 3. Suppose that G contains a bridge, say, e = uv.
Then G e is disconnected with exactly two components. Denote the component of G e containing u
by Hu and the other component of G e by Hv . Since |V (G)| = |V (Hu )| + |V (Hv )| 3, at least one of
Hu and Hv has order at least 2. Without loss of generality we may assume |V (Hu )| 2. Then G u
has at least two components, one of which is Hv and the others are components of Hu u. Hence u is
a cut-vertex of G.
PS5-13
Let v be a vertex of a connected graph G. Prove that v has a neighbour in every connected component
of G v. Conclude that no graph has a cut-vertex of degree 1.
Proof: If G is the trivial graph, then nothing needs to be proved as G v is the null graph and hence
has no connected component.
Assume G is non-trivial. Since G is a non-trivial connected graph, it has no isolated vertices. So
k := deg(v) 1. Let the neighbours of v in G be u1 , . . . , uk . Let H be any component of G v. Since G
is connected, there exists a path from v to some vertex w 2 V (H). This path must use some neighbour
of v, say u1 , and hence is of the form v, u1 , . . . , w. Hence its segment u1 , . . . , w is a path in G v, and
it is a path in H since w 2 V (H) and H is a component of G v. Therefore, H contains the neighbour
u1 of v. This shows that every connected component of G v contains a neighbour of v.
Let K be any (not necessarily connected) graph and let v be a vertex of K with degree 1. Consider
the component G of K containing v. Applying the result above to G, we know that every component of
G v must contain the unique neighbour of v. Hence G v must have exactly one component. Therefore,
(K v) = (K) and so v is not a cut-vertex of K.
PS6-1
Find a cubic graph with no perfect matching.
Answer: There are many examples. A well-known example is as follows (https://en.wikipedia.org/
wiki/Petersen\%27s_theorem). Convince yourself that indeed this graph has no perfect matching.

1
PS6-2

Take two disjoint paths P4 and P5 of lengths 3 and 4 respectively, and add an edge from each vertex of
one to each vertex of the other. Does the resulting graph have a perfect matching?
Answer: No, because it has an odd number of vertices.

PS6-3

Let G be the following graph:

a e

b f

c g

d h

(a) Consider the matching M = {gd, ah} in G. Find an augmenting path for M in G. Use this path
to find a matching of cardinality 3 in G. What is the cardinality of a maximum matching in G?
Give a reason for your answer.
(b) Apply the algorithm given in lectures, from the start, to find a maximum matching in G.

Answer: (a) An augmenting path: c, g, d, h, a, e (there are others).


Augmenting along this path, we obtain the matching M 0 = {cg, dh, ae}.
Note that the given graph is bipartite with bipartition {{a, b, c, d}, {e, f, g, h}}. For S = {b, c} ⇢
{a, b, c, d}, we have N (S) = {g} and so |N (S)| < |S|. Thus, by the König-Hall Theorem, there is no
matching that matches all vertices in {a, b, c, d}. Hence the cardinality of a maximum matching is at
most three. But M 0 is a matching of cardinality three. So the cardinality of a maximum matching is
equal to three, and M 0 is a maximum matching.
(b) Omitted. Work out details by yourself.

2
MAST30011 Graph Theory
Practical Class 8

PS6-5

A vertex cover of a graph is a set Q ✓ V (G) such that each edge of G is incident with at least one vertex
in Q (that is, vertices in Q cover E(G)). Prove the following:
König-Egerváry Theorem: If G is a bipartite graph, then the maximum size of a matching in G is
equal to the minimum size of a vertex cover of G.
Proof: We prove this by using the König-Hall Theorem.
Let G be a bipartite graph with bipartition {X, Y }. Let M be a maximum matching in G, and let
Q be a vertex cover of G with minimum size. It suffices to prove that |M | = |Q|.
Since M is a matching, di↵erent edges in M must be covered by di↵erent vertices in Q. Hence
|Q| |M |. We will construct a matching M ⇤ with size |M ⇤ | = |Q|. Once this is achieved, we then have
|M | |M ⇤ | = |Q|.
Set R = Q \ X and T = Q \ Y . Then Q = R [ T and {R, T } is a partition of Q. Let H and K be
the subgraphs of G induced by R [ (Y T ) and T [ (X R), respectively. Since Q = R [ T is a vertex
cover of G, there is no edge of G between X R and Y T .
Thus, for any S ✓ R, NH (S) is contained in Y T . Since NH (S) covers all edges incident to S
that are not covered by (R S) [ T , we see that (R S) [ NH (S) [ T is a vertex cover of G. If
|NH (S)| < |S|, then this vertex cover would be smaller than Q, which contradicts the minimality of
Q. Hence |NH (S)| |S| for every S ✓ R. So by the König-Hall Theorem, H has a matching M1 that
matches every vertex in R.
The same argument applied to K yields a matching M2 in K that matches every vertex in T .
Since H and K are vertex-disjoint subgraphs of G,

M ⇤ := M1 [ M2

is a matching in G. Moreover, we have M1 \ M2 = ;. Hence

|M ⇤ | = |M1 | + |M2 | = |R| + |T | = |Q|.

Since M is a maximum matching, we have |M | |M ⇤ | = |Q|. However, we also have |Q| |M | as


proved earlier. Therefore, |M | = |Q| and the proof is complete.

1
R T

S
K
H

NH(S)
No edges between
X-R and Y-T in G

X⎯R Y⎯T

X Y
Q is the union of R and T
|Q| = |R| + |T|

PS6-6

Let G be the following graph:


a h

b i

c j

d k

e m

f n

g p

Assume the matching algorithm given in lectures finds the matching

M = {ah, bi, cj, dk, em}

in G after 5 steps. Complete the algorithm and hence find a maximum matching.
Answer: There are many ways to complete the algorithm. The first alternating tree which is grown
from f can include edges such as f m, me, ej, jc, cp. Since p is unmatched we can stop now and use the
alternating path f, m, e, j, c, p. Switching the edges in this path in and out of M gives the new matching

M = {ah, bi, dk, f m, ej, cp}.

Starting at g, grow an alternating tree with respect to the new matching. For example, we may add
edges gm, gi, gh, f m, ib, ha, an to the tree and reach the unmatched vertex n. So take the path in this
tree from g to n: g, h, a, n. Augment along this path to obtain the matching

M = {bi, dk, f m, ej, cp, gh, an}.

Since the last vertex g on the left has been treated (or since the current matching is perfect already),
this must be a maximum matching by the maximum bipartite matching algorithm.

2
PS6-6 (from scratch)

Let G be the graph above. Using the matching algorithm, find a maximum matching from scratch (i.e.
start with the empty matching).
Answer: See below for an illustration (without detailed explanation). In each step, if there are multiple
augmenting paths, we can choose any one of them to augment. Di↵erent maximum matchings may be
obtained if di↵erent augmenting paths are used.

PS6-7

The complete graph K2n+1 has no perfect matching since it has an odd number of vertices. Convince
yourself that the complete graph K2n has perfect matchings. How many perfect matchings are there in
K2n ?
Solution: Let fn denote the number of perfect matchings in K2n . Then fn is the number of ways to
pair up 2n vertices.

3
Denote the vertices of K2n by v1 , v2 , . . . , v2n . There are 2n 1 choices for the partner of v2n , and for
each such choice there are fn 1 ways to complete the matching. Hence, for any n 2,

fn = (2n 1)fn 1.

Note that f1 = 1. Applying the equation above recursively, we obtain fn = (2n 1)fn 1 = (2n
1)(2n 3)fn 2 = · · · = (2n 1)(2n 3) · · · 3 f1 = (2n 1) · (2n 3) · · · 3 · 1.

PS6-8

Prove that every tree has at most one perfect matching.


Proof: Let T be a tree. If the order of T is odd, then T has no perfect matching. Assume in the sequel
that T has order 2n for some n 1.
Proof 1: Suppose to the contrary that T has two distinct perfect matchings, say, M1 and M2 . Let H
be the spanning subgraph of T whose edge set is the symmetric di↵erence between M1 and M2 (that is,
(M1 \ M2 ) [ (M2 \ M1 )). Since T has no cycles, each component of H is a path (possibly a trivial path).
Moreover, since both M1 and M2 are perfect and hence maximum matchings, there is no augmenting
path with respect to each of them. Hence each component of H must be a path of even length whose
edges belong to M1 and M2 (or M2 and M1 ) alternatively. Furthermore, H has at least one nontrivial
component as M1 6= M2 .
Let
P : v0 , v1 , . . . , v2k 1 , v2k
be a path that is a nontrivial component of H. Without loss of generality we may assume that
v0 v1 , v2 v3 , . . . , v2k 2 v2k 1 2 M1 and v1 v2 , v3 v4 , . . . , v2k 1 v2k 2 M2 . Since M2 is a perfect matching
of T , v0 must be matched to some vertex u by M2 (that is, v0 u 2 M2 ), and moreover u 62 V (P ) for oth-
erwise T would contain a cycle, a contradiction. If v0 u 2 M1 , then v0 is matched to both u and v1 (6= u)
by M1 , a contradiction. If v0 u 62 M1 , then v0 u 2 E(H) by the definition of H, but this contradicts the
assumption that P is a component of H. Therefore, T cannot have two distinct perfect matchings.
Proof 2: Consider a leaf of the tree. The corresponding pendant edge must be a matching edge in any
perfect matching. Use this fact recursively to prove the desired result.
Exercise: Complete this proof by yourself ! (The best way to proceed is by making induction on
the (even) order of the tree.)

4
MAST30011 Graph Theory
Practical Class 9

PS7-1

For which values of n is Kn eulerian? Why?


Answer: All odd positive integers n. This follows immediately from Euler’s Theorem and the fact that
Kn is (n 1)-regular.

PS7-2

What is the minimum number of edges which must be added to a 5-regular graph of order 40, in order
to make a graph with an eulerian trail?
Answer: 19.
Proof: Let G be a 5-regular graph with order 40. Then G has no eulerian trails as it has more than two
odd vertices. To turn G into a graph with an eulerian trail, we need to add a set of edges to G in such
a way that the new graph has exactly two odd vertices. So the degrees of 38 vertices must be increased
by some odd positive integers (and the degree of each of the remaining two vertices is unchanged or
increased by an even positive integer). So the sum of the degrees must be increased by at least 38. That
is, at least 19 edges must be added to G.
We now show that 19 can be achieved. In fact, the complement G of G is a 34-regular graph with
order 40. Since 34 40/2, by Ore’s Theorem, G contains a Hamilton cycle C. Since C has length 40, it
contains a matching M of size 19. Of course, M is a matching in G. By adding all 19 edges of M to G,
we obtain a graph which has exactly 38 vertices of degree 6 and two vertices of degree 5. So this graph
contains an eulerian trail.
In summary, the minimum number of edges which must be added to G in order to obtain a graph
with an eulerian trail is 19.

PS7-3

Prove that Cm ⇤Cn is eulerian for all m, n 3.


Proof: Label the vertices of Cm by u1 , . . . , um in such a way that

E(Cm ) = {u1 u2 , u2 u3 , . . . , um 1 um , um u1 }.

Label the vertices of Cn by v1 , . . . , vn in such a way that

E(Cn ) = {v1 v2 , v2 v3 , . . . , vn 1 vn , vn v1 }.

Then V (Cm ⇤Cn ) = {(ui , vj ) : 1  i  m, 1  j  n} and each (ui , vj ) is adjacent to (ui , vj 1 ), (ui , vj+1 ),
(ui 1 , vj ) and (ui+1 , vj ) in Cm ⇤Cn only, where the subscripts in the first coordinate are taken modulo m
and the subscripts in the second coordinate taken modulo n. Hence Cm ⇤Cn is 4-regular and so contains
no odd vertices. By Euler’s Theorem, Cm ⇤Cn is eulerian.
It is well known that Cm ⇤Cn is a grid graph on a torus. Draw a small example e.g. C3 ⇤C4 on a
torus by yourself.

PS7-5

1
Show that a connected multigraph G is eulerian if and only if the edge set of G can be partitioned into
subsets, each of which induces a cycle (where 2-cycles are permitted).
Proof: If G is trivial, then there is nothing to prove. Henceforth we assume that G is a nontrivial
connected multigraph.
(() We first prove that if E(G) can be partitioned into subsets each of which induces a cycle, then
G is eulerian.
Suppose that E(G) can be partitioned into E1 , . . . , Er such that each G[Ei ] is a cycle, i = 1, . . . , r.
Let u be an arbitrary vertex of G. If a cycle G[Ei ] passes through u, then exactly two edges of G[Ei ]
are incident with u. Since {E1 , . . . , Er } is a partition of E(G), this implies that u is incident with an
even number of edges of G; that is, deg(u) is even. By the arbitrariness of u, all vertices of G are even.
Hence G is eulerian by Euler’s Theorem.
()) We need to prove that if G is eulerian then E(G) can be partitioned into subsets each inducing
a cycle.
Suppose that G is eulerian. Then by Euler’s Theorem all vertices of G are even. Note that G contains
cycles, for otherwise it is a nontrivial tree and so contains at least one vertex of degree 1 (leaf), which is
a contradiction.
We prove, by induction on the number of cycles, that:
(*) if a nontrivial connected multigraph is eulerian, then its edge set can be partitioned into subsets
each inducing a cycle.
First, this is true if a nontrivial connected multigraph contains exactly one cycle, because in this case
the graph itself is a cycle.
Suppose that (*) holds for any nontrivial connected eulerian multigraph containing less than m cycles.
Let G be a nontrivial connected eulerian multigraph containing m cycles, where m 2. Since G is
eulerian, by Euler’s Theorem all vertices of G are even. As proved above, G contains at least one cycle.
Let C be a cycle of G (which may be a 2-cycle). Then G E(C) is a spanning subgraph of G whose
vertices are even. Equivalently, all vertices in each component of G E(C) are even.
Let G1 , . . . , Gk denote the components of G E(C) with at least two vertices. Since all vertices of
Gi are even, by Euler’s Theorem, Gi is eulerian for i = 1, . . . , k. Moreover, for each i,

# cycles in Gi < # cycles in G = m.

By the hypothesis, E(Gi ) can be partitioned into subsets, say, Ei,1 , . . . , Ei,ni (for some integer ni 1),
each of which induces a cycle of Gi . Since Gi is a subgraph of G, each Ei,j induces a cycle of G,
j = 1, . . . , ni . Therefore, E(G) is partitioned into

E(C), E1,1 , . . . , E1,n1 , . . . , Ek,1 , . . . , Ek,nk

each of which induces a cycle of G.

PS7-7

Prove that if (G) 2 then G contains at least one cycle.


First proof: Let P be a longest path in G. Let u0 be the first vertex of P , and u1 the second. Since
(G) 2, we have deg(v) 2 for all v 2 V (G), and so u0 must be adjacent to at least one vertex other
than u1 . Let w 6= u1 be such a vertex adjacent to u0 . If w is not on P , then P can be made into a
longer path by adding the edge wu0 , which contradicts the assumption that P is a longest path in G.
So w must be on P . Then the segment of P from u0 to w together with the edge wu0 forms a cycle in
G. Therefore, G contains at least one cycle.
First proof (with a slightly di↵erent presentation): Take a longest path

P : v 0 , v1 , . . . , v k 1 , vk

2
in G. The length k of P must be at least 2, for otherwise the component of G containing P would be
isomorphic K1 or K2 and so the minimum degree of G is at most 1, contradicting our assumption.
Since G has minimum degree at least 2, each of its vertices has degree at least 2. So v0 is adjacent
to at least one vertex, say u, other than v1 . If u 62 V (P ), then u, v0 , v1 , . . . , vk 1 , vk is a path of G
longer than P , which contradicts our choice of P . So we must have u 2 V (P ), say, u = vi , where we
have 2  i  k as u 6= v1 . So we get a cycle v0 , v1 , . . . , vi 1 , vi , v0 in G. Hence G contains at least one
cycle.
Second proof: Since G has minimum degree at least 2, no connected component of it can be isomorphic
K1 or K2 . That is, every component of G has order at least 3.
Suppose to the contrary that G contains no cycles. Then G is a forest with at least three vertices in
each of its components. That is, every component of G is a tree with at least three vertices. So every
component of G contains at least two leaves, and these leaves are also leaves of G. (It is known that any
tree with at least two vertices contains at least two leaves.) But this implies that the minimum degree
of G is at most 1, a contradiction. Hence G must contain cycles.

3
MAST30011 Graph Theory
Practical Class 10

PS7-8
Show that if G is a graph containing a vertex that is adjacent to at least three vertices of degree 2, then
G is not Hamiltonian.
Proof 1: Let u be a vertex of G which is adjacent to at least three vertices of degree 2. Then we
can take three distinct neighbours v1 , v2 , v3 of u with degree 2. So for i = 1, 2, 3, vi has a unique
neighbour other than u, which we denote by wi . Note that w1 , w2 and w3 are not necessarily distinct,
and {w1 , w2 , w3 } \ {v1 , v2 , v3 } may or may not be empty.
Suppose to the contrary that G is Hamiltonian. Let C be a Hamilton cycle of G. Since u and wi
are the only neighbours of vi , C must use both uvi and vi wi for i = 1, 2, 3. However, this implies that
uv1 , uv2 and uv3 are all in C, which cannot happen as C is a cycle but all these three edges are incident
to u. This contradiction shows that G is not Hamiltonian.
Proof 2: Suppose, for a contradiction, that G is Hamiltonian. Let
C : v 1 , v2 , . . . , v n , v1
be a Hamilton cycle of G, where n 3 is the order of G. By our assumption, G contains a vertex which
is adjacent to at least three vertices of degree 2. Without loss of generality we may assume that v1 is
adjacent to at least three vertices of degree 2.
Note that v2 and vn are neighbours of v1 (but their degrees may or may not be equal to 2). Since
v1 has at least three neighbours with degree 2, there exists a vertex vi other than v2 , vn such that
deg(vi ) = 2 and vi is adjacent to v1 in G. So we have 3  i  n 1, but this implies that v1 , vi 1 , vi+1
are distinct neighbours of vi . Hence deg(vi ) 3, but this contradicts the assumption that deg(vi ) = 2.
This contradiction shows that G is not Hamiltonian.
PS7-10
Show that the bound in the Chvátal-Erdös Theorem is sharp by proving that the graph G = Kk,k+1
(where k 2) satisfies (G) = ↵(G) 1 but is not Hamiltonian.
Proof: Note that G = Kk,k+1 is a non-complete graph. Let {X, Y } be the bipartition of G, where
|X| = k and |Y | = k + 1.
Observe that both X and Y are vertex cuts of G. On the other hand, for any S ✓ V (G) = X [ Y
such that X S 6= ; and Y S 6= ;, we have s1 := |S \ X| < k and s2 := |S \ Y | < k + 1 and so
G S⇠ = Kk s1 ,k+1 s2 is a complete bipartite graph. Hence G S is connected. Thus any vertex cut of
G must contain either X or Y as a subset. Since |X| < |Y |, it follows that X is the unique vertex cut of
G with minimum cardinality. Hence (G) = |X| = k.
On the other hand, both X and Y are independent sets of G. Since every vertex in X is adjacent
every vertex in Y , any S ✓ V (G) such that X \ S 6= ; and Y \ S 6= ; cannot be an independent set of
G. That is, any independent set of G must be a subset of X or Y . Since |Y | > |X|, it follows that Y is
the unique independent set of G with maximum cardinality. So ↵(G) = |Y | = k + 1.
Therefore, G satisfies (G) = ↵(G) 1. In particular, G does not satisfy the condition (G) ↵(G)
in the Chvátal-Erdös Theorem.
Since G is bipartite, it contains no odd cycles. In particular, it does not contain any cycle of length
2k + 1 = |V (G)|. That is, G does not contain any Hamilton cycle and hence is not Hamiltonian.

1
PS7-12

Prove that Kr,2r,3r+1 is non-Hamiltonian for any positive integer r.


Proof: Recall that Kr,2r,3r+1 is the complete 3-partite graph with partite sets of sizes r, 2r, 3r + 1,
respectively. Let {X, Y, Z} be the tri-partition of Kr,2r,3r+1 , where |X| = r, |Y | = 2r and |Z| = 3r + 1.
Take S = X [ Y . Then |S| = 3r and k(G S) = 3r + 1 as each vertex in Z is an isolated vertex (and
hence a connected component) of G S. Since k(G S) > |S|, it follows that G is not tough. Hence G
is non-Hamiltonian.

PS7-15

The m ⇥ n rectangular grid Gm,n (where m, n 1 are integers) can be defined as follows:

V (Gm,n ) = {(i, j) : i, j are integers, 1  i  m, 1  j  n}

E(Gm,n ) = {uv : u = (i, j) 2 V (Gm,n ), v = (i0 , j 0 ) 2 V (Gm,n ),


|i i0 | + |j j 0 | = 1}.

Prove that Gm,n is Hamiltonian if and only if either m or n is even.


Proof: Sufficiency: Suppose that either m or n is even. Try to find a Hamilton cycle in Gm,n by yourself.
(Start with a concrete example, say G3,6 , and find a Hamilton cycle in it. Then generalise your idea to
the general case.)
Necessity: Suppose that Gm,n is Hamiltonian. We aim to prove that either m or n is even. We prove
the contrapositive: If both m and n are odd, then Gm,n is not Hamiltonian.
Suppose that both m and n are odd. Let

S = {(i, j) 2 V (Gm,n ) : i + j is odd}.

Then |S| = (mn 1)/2. Also, k(Gm,n S) = (mn + 1)/2 as every vertex in Gm,n S is an isolated
vertex. Hence k(Gm,n S) > |S|. So Gm,n is not tough. Therefore, Gm,n is not Hamiltonian.
Necessity (second proof): Let

S = {(i, j) 2 V (Gm,n ) : i + j is odd}

and
T = {(i, j) 2 V (Gm,n ) : i + j is even}.
Then {S, T } is a partition of V (Gm,n ). By the definition of Gm,n , no two vertices in S are adjacent,
and no two vertices in T are adjacent. Hence Gm,n is a bipartite graph with bipartition {S, T }. So it
contains no odd cycles. In particular, if both m and n are odd, then |V (Gm,n )| = mn is odd and so
Gm,n contains no cycle of length mn. In other words, if both m and n are odd, then Gm,n contains no
Hamilton cycle and hence is not Hamiltonian.

PS8-1

(a) Find a simple planar graph with degree sequence (4, 4, 4, 4, 3, 3).
(b) Find a simple non-planar graph with the same degree sequence.

(This exercise shows that we cannot tell whether a graph is planar or not by simply looking at its
degree sequence.)
Solution: (a) The octahedron graph is planar (draw it in the plane without edge-crossing by yourself)
and has degree sequence (4, 4, 4, 4, 4, 4). Removing one edge from the octahedron graph yields a graph
with degree sequence (4, 4, 4, 4, 3, 3). This simple graph is planar as it is a subgraph of a planar graph.

2
(b) The complete bipartite graph K3,3 is non-planar with degree sequence (3, 3, 3, 3, 3, 3). Adding
two edges with no common end-vertex to K3,3 results in a graph with degree sequence (4, 4, 4, 4, 3, 3).
This simple graph is non-planar as it contains the non-planar graph K3,3 as a subgraph.

3
MAST30011 Graph Theory
Practical Class 11

PS8-2

Is the following graph planar? Is it maximal planar? Remove or add edges to it to make it maximal
planar.

Answer: This graph is planar as it can be embedded in the plane without edge-crossing (draw this graph
in the plane without edge-crossing by yourself).
Recall that a maximal planar graph is a planar graph that is maximal with respect the property of
being planar, that is, adding any edge (but not vertex) results in a non-planar graph. It is known that a
planar graph with n vertices is maximal if and only if it has exactly 3n 6 edges. Since this graph has
12 vertices and 26 (< 3 · 12 6) edges, it is not maximal planar.
Moreover, we need to add (3 · 12 6) 26 = 4 edges to obtain a maximal planar graph. There are
several ways to achieve this. See below for an example, where dashed edges are new edges.

PS8-3

Draw a 5-regular simple planar graph. What is the minimum number of vertices in such a graph? What
is the maximum number of vertices in such a graph?
Answer: The icosahedron graph is the standard example of a 5-regular simple planar graph of order 12.

1
By Euler’s formula, every 5-regular simple planar graph G of order n has at most 3n 6 edges. On
the other hand, G has exactly 5n 5n
2 edges as it is 5-regular. Hence, 2  3n 6, which implies n 12.
Thus every 5-regular planar graph has at least 12 vertices, and 12 can be achieved by the icosahedron
graph. Therefore, the minimum number of vertices in a 5-regular simple planar graph is 12. (In fact,
the icosahedron graph is the only 5-regular planar graph with 12 vertices.)
There is no upper bound on the number of vertices in a 5-regular simple planar graph. Why? Here
is one construction of a 5-regular simple planar graph with arbitrarily many vertices. Start with a cycle
(w1 , w2 , . . . , wk ) drawn in the plane. Next, draw a new cycle (x1 , y1 , x2 , y2 , . . . , xk , yk ) around the outside,
and make wi adjacent to yi 1 , xi , yi (where y0 means yk ). Finally, draw another cycle (z1 , z2 , . . . , zk )
around the outside, and make zi adjacent to xi , yi , xi+1 (where xk+1 means x1 ).
You should check that we obtain a 5-regular simple planar graph with 4k vertices. Since k can be
arbitrarily large, there is no maximum number of vertices in a 5-regular simple planar graph.
z2

y2
x3 x2
z3 z1
y1
y3
w3 w2

x4 w4 w1 x1

w5 w6
y6
y4
z4 z6
x5 x6
y5

z5

PS8-4
Prove that the Petersen graph is nonplanar by using the fact that every cycle of it has length 5 or more.
Proof: Suppose to the contrary that the Petersen graph is planar. Let r be the number of faces in
a planar embedding of the Petersen graph. Since the graph has 10 vertices and 15 edges, we have
10 15 + r = 2 by Euler’s formula, and so r = 7.
Since the Petersen graph has no loops or cut vertices (show this by yourself), the boundary of each
face is a cycle. Since every cycle of the Petersen graph has length at least 5, the total length of the 7
cycles which are boundaries of the 7 faces is at least 7 · 5 = 35. On the other hand, each edge belongs to
exactly two of these cycles. So the total number of edges is at least 35/2 > 17, which is a contradiction.
Therefore, the Petersen graph is nonplanar.
PS8-5
(a) Let G be a simple planar graph with n vertices and m edges. Show that if a shortest cycle in G has
length k, then m  k(n 2)/(k 2).
(Definition: If G contains cycles, then its girth is defined as the length of a shortest cycle in G; if G
has no cycles (i.e. G is a forest), then its girth is defined to be 1.)

2
(b) Deduce that a simple planar bipartite graph of order n 1 has at most 2n 4 edges.

Proof: (a) Let G be a simple planar graph with n vertices, m edges, and f faces. Let k be the length
of a shortest cycle in G. An underlying assumption in this question is that k is finite, so G contains at
least one cycle and thus n 3. By abuse notation, we also use G to denote a planar embedding of G.
Assume first that G is connected with no cut vertices. Then G is 2-connected and the boundary of each
face of G is a cycle. So each face of G has at least k edges on its boundary. Since each edge of G is in two
faces, it follows that 2m kf . Combining this with Euler’s formula, we have n m+ 2m k n m+f = 2.
That is,
k
m (n 2),
k 2
as required.
Now assume that G is connected, and G contains cut vertices but not bridges. Suppose G has t 1
cut vertices, where t 2. Then it has exactly t blocks, say, G1 , . . . , Gt . Then each Gi contains no cut
vertices, and moreover Gi has order at least three as G has no bridges. Let ni , mi , ki be the order, size
and girth of Gi , respectively, for 1  i  t. Then k = min{k1 , . . . , kt } and hence kiki 2  k k 2 for each i.
Applying what we have proved above to Gi , we obtain
ki k
mi  (ni 2)  (ni 2).
ki 2 k 2
Hence ! !
t
X t
X
k
mi  ni 2t .
i=1
k 2 i=1
Pt Pt
Note that m = i=1 mi and n = ( i=1 ni ) (t 1). So
k k k
m (n + (t 1) 2t) = (n t 1) < (n 2)
k 2 k 2 k 2
as desired.
Next assume that G is connected and contains bridges. Let u1 v1 , . . . , ub vb be the bridges of G, where
b 1. Let G0 be the graph obtained from G by contracting all these edges. Then G0 contains at least
b cut vertices but no bridges. (Each bridge of G gives rise to a cut vertex of G0 , and each cut vertex of
G not in {u1 , v1 , . . . , ub , vb } is also a cut vertex of G0 .) Let n0 , m0 , k 0 be the order, size and girth of G0 .
Then n0 = n b, m0 = m b and k 0 = k. Since G0 is connected but contains no bridges, by what we
0
have proved above, we obtain m0  k0k 2 (n0 2), that is, m b  k k 2 (n b 2). Hence
k 2 k
m (n 2) b< (n 2)
k 2 k 2 k 2
as desired.
So far we have proved the desired inequality when G is connected. It remains to consider the case when
G is disconnected, say, with connected components H1 , . . . , Hc , where c 2. Let ni , mi , ki be the order,
size and girth of Hi , respectively, for 1  i  c. Then k = min{k1 , . . . , kt } and hence kiki 2  k k 2 for each
i. Since we have proved the result for simple planar graphs, we have mi  kiki 2 (ni 2)  k k 2 (ni 2)
for each i. Hence !
Xc Xc
k
mi  ( ni ) 2c .
i=1
k 2 i=1
Pc Pc
Note that m = i=1 mi and n = i=1 ni . So
k k
m (n 2c) < (n 2)
k 2 k 2
as desired. This completes the proof.
(b) If G is a simple planar bipartite graph, then every cycle in G has at least four edges. Hence k 4.
By part (a), we obtain m  k(n 2)/(k 2)  4(n 2)/(4 2) = 2n 4. In other words, G has at most
2n 4 edges.

3
PS9-10

An edge-colouring of a graph is an assignment of colours to its edges such that no two edges with a
common end-vertex receive the same colour. The least number of colours required is called the edge-
chromatic number of the graph, denoted by 0 .
0
Show that (T ) = (T ) for any tree T .
Proof: Since the (T ) edges incident with a maximum degree vertices require distinct colours, we have
0
(T ) (T ).
0
We prove (T )  (T ) by induction on the number of vertices.
A tree with order 1 is isomorphic to K1 . Since 0 (K1 ) = (K1 ) = 0, the result is true for trees of
order 1. Similarly, since a tree of order 2 is isomorphic to K2 and 0 (K2 ) = (K2 ) = 1, the result is
true for trees of order 2.
Suppose 0 (T1 )  (T1 ) for any tree T1 of order n 1 2.
Let T be a tree of order n. Since n 3, T contains at least one leaf (vertex of degree 1). Let v be
a leaf of T and e = uv the unique edge incident with v. Denote k = (T v). Since T v is a tree of
order n 1, by our hypothesis, 0 (T v)  (T v) = k and so T v can be edge-coloured by at most
k colours, say, 1, 2, . . . , k.
Case 1: dT v (u) = k.
Then (T ) = k + 1. Assign colour k + 1 to e. In this way we obtain an edge-colouring of T by at
most k + 1 colours. Hence 0 (T )  k + 1 = (T ).
Case 2: dT v (u) < k.
Then dT (u) = dT v (u) + 1  k and so (T ) = k. In this case at least one of the k colours used for
T v is not used by any edge incident to u. So we can assign this colour to e to obtain an edge-colouring
of T by at most k colours. Hence 0 (T )  k = (T ).
0
So we have proved (T )  (T ). By mathematical induction, the proof is complete.

You might also like