Graph Theory

Michlmas 2003

ii

Contents
1

Introduction

Graphs

Flows, connectivity and matching

Connectivity and the theorems of Menger

11

Matchings

13

Extremal graph theory

6.1 Complete subgraphs and Turans theorem . . . . . . . . . . . . . . .
6.2 The problem of Zarankiewicz . . . . . . . . . . . . . . . . . . . . . .

15
16
18

Graph colouring
7.1 Vertex colouring and Brooks theorem . . . . . . . . . . . . . . . . .

19
19

Edge colouring and Vizings theorem

21

Colouring graphs on surfaces

9.1 Plane graphs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

23
23

iii

iv

CONTENTS

Chapter 1

Introduction
The notes were typed up by me, John Fremlin [email protected].
These notes are based on the part IIA mathematics course Graph theory given
by Dr Fisher in Cambridge in Michlmas 2003. These notes are not connected to Dr
Fisher in any way. If there are any mistakes in them, it is more than very likely that
they are my fault.
I added a few clumsy elucidations, to the arguments that I initially did not understand, which will no doubt ensure that there are at least some errors, because I could
not find any in Dr Fishers lectures.
Furthermore these notes are very definitely no substitute for actually going to lectures, because they do not include all of the material and especially examples covered,
or any of the asides.
Finally, I would like to thank Dr Fisher for his supervisions, where he taught me a
lot about mathematics.

CHAPTER 1. INTRODUCTION

Chapter 2

Graphs
Definition 2.1 (Graph). A graph is a pair (V, E) with E V (2) .
Loops and multiple edges are forbidden.
Definition 2.2 (Vertices). V = V (G) is the set of vertices of G.
Definition 2.3 (Edges). E = E(G) is the set of edges of G.
Example 2.4 (Complete graph). Kn is the complete graph on n vertices (all
edges).

n
2

Example 2.5 (Cycle). Cn ,n 3 is a cycle of length n; it has vertices v1 , v2 , , vn

and edges v1 v2 , v2 v3 , .
Definition 2.6 (Isomorphic graphs). Two graphs G1 , G2 are isomorphic (symbol
=) if
there is a bijection : V (G1 ) V (G2 ) : xy E(G1 ) (x)(y) E(G2 ).
= (V, V (2) \ E).
Definition 2.7 (Complement). The complement of G = (V, E) is G
Example 2.8 (C 5 isomorphic to C5 ).
Definition 2.9 (Subgraph). A subgraph of a graph G is a graph H with V (H) V (G)
and E(H) E(G).
T
Definition 2.10 (Induced subgraph). H is an induced subgraph if E(H) = E(G) V (2) (H).
The induced subgraph with vertex set W V (G) is written G[W ].
Definition 2.11 (Spanning subgraph). H is a spanning subgraph if V (H) = V (G).
Definition 2.12 (Walk). A walk in a graph is a sequence of vertices v1 , v2 , , vk
E(G)1 i < k. Its length is k 1. It is closed if v1 = vk .
Definition 2.13 (Trail). A trail is a walk with no repeated edges.
Definition 2.14 (Euler trail/circuit). An Euler trail/circuit is a trail or circuit which
passes through all the edges.
Definition 2.15 (Eulerian). A graph is Eulerian if it has an Euler circuit.
Definition 2.16 (Path). A path is a walk (a trail) with no repeated vertices.
3

CHAPTER 2. GRAPHS

Definition 2.17 (Cycle). A cycle is a closed walk of length 3 with distinct vertices
except that the frist and last are equal (isomorphic to Cn ).
Definition 2.18 (Hamilton trail/circuit). A Hamilton path/cycle is a path/cycle which
goes through every vertex of the graph.
Definition 2.19 (Hamiltonian). A graph is Hamiltonian if it has a Hamilton cycle.
Definition 2.20 (Neighbours of a vertex). Let x be a vertex of G, its neighbours or
adjacent vertices are (x) = {y V (G) : xy E(G)}.
Definition 2.21 (Degree of a vertex). The degree of a vertex x is d(x) = k(x)k.
Definition 2.22 (Degree sequence of a graph). The degree sequence of a graph is a
graph with vertices v1 , , vn is d(v1 ), , d(vn ).
Definition 2.23 (Maximum, minimum degree). The minimum/maximum degree of a
vertex of G is denoted (G) or (G) respectively.
Definition 2.24 (Regular). A graph G is regular if (G) = (G).
Definition 2.25 (Size of a graph). e(G) = kE(G)k is the number of edges of G, its
size.
Definition 2.26 (Order of a graph). kGk = kV (G)k is the number of edges of G, its
order.
P
Lemma 2.27 (Handshaking lemma).
vV (G) d(v) = 2e(G)
Proof. Double count {(x, xy) : x V (G), xy E(g)}.

Observation 2.28. To solve problems try induction (on vertices, edges, and other
things), try double counting (counting something in more than one way), consider the
pigeon hole principle and consider extreme cases.
Definition 2.29 (Connected). A graph is connected if every pair of vertices is joined
by a path.
The relationship on V (G) where x y where is a path between x and y is an
equivalence relation. The equivalence classes are called components of G.
Definition 2.30 (Acyclic). A graph is acyclic if it has no cycles.
Definition 2.31 (Tree). A tree is a connected acyclic graph.
Definition 2.32 (Forest). A forest is an acycle graph.
Theorem 2.33 (Properties of a tree). The following are equivalent
1. G is a tree (connected, acyclic)
2. G is minimal connected
3. G is maximal acyclic

5
Proof. i = ii. Let G be connected and acyclic. Let uv E(G). If G uv were
connected then any u v path in G uv would give a cycle in G.
ii = i. If C were a cycle in G and uv E(G) then any x y path in G passing
via uv could be patched up to C uv to give a path in G uv. Contradiction.
i = iii. Let uv
/ E(G). Since G connected u v path in G so there is a cycle
in G + uv. Contradiction.
iii = i. Suppose G is not connected. Let u, v be vertices of G belonging to
different components. G + uv contains no cycles. Contradiction.
Corollary 2.34. A graph is connected iff it has a spanning tree (i.e. G has a subgroup
T , T a tree and V (T ) = V (G)).
Proof. Assume G has a spanning tree T . Then there is a path in T and so in G between
all vertices.
Let T be minimal connected spanning subgraph of G. By previous theorem it is a
tree.
Definition 2.35 (Leaf). A leaf is a vertex v with d(v) = 1.
Lemma 2.36 (Trees have leaves). Every tree T of order n 2 has at least two leaves.
Proof. Let P = x1 xk be a path of maximal length in T . Then (x1 ) P , since P
is maximal. T is acyclic so (x1 ) P = {x2 }. Similarly for xk .
Theorem 2.37 (Size of a tree). A tree of order n has size n 1. e(T ) = kT k 1. (In
fact a connected graph G with e(T ) = kT k 1 is a tree, exercise.)
Proof. By induction on n. Clearly true for n 2. Let T be a tree of order n and let
v V (T ) be a leaf. Let x, y be vertices of T v. There is an x y path in T . So
T connected implies T v connected so T v is a tree. By induction e(T v) =
n 2 = e(T ) = e(T v) + 1 = n 1.
Observation 2.38 (Number of possible graphs). If vertices are labelled 1, , n there


n
are n2 possible edges and each subset gives a graph so there are 2( 2 ) possible graphs.
If the vertices are unlabelled then each graph can be labelled in at most n! possible
(n2 )
ways. So there are at least 2 n! graphs.
Theorem 2.39 (Cayley). There are nn2 labelled trees of order n.
Proof. (Due to Prufer.) Given a labelled tree we constract a sequence of n 2 numbers
in the range 1, , n as follows.
Select the lowest labelled leaf. Write down its neighbour. Delete the leaf. Repeat
until just two vertices remain. Now have a function f from set of labelled trees of order
n to {1, n}n2 .
Claim: f is injective. Each vertex v appears d(v) 1 times in the sequence. Leaves
are vertices not appearing. Given a sequence (a1 , , an2 ) let v1 be the least vertex
not appearing, join it to a1 . Let v2 be the least vertex not appearing in the new sequence (v1 , a2 , , an2 ), join it to a2 . Repeat until there are only two nodes not in
(v1 , , vn2 ), join them together. The original graph is reconstructed.

CHAPTER 2. GRAPHS

Claim: f is surjective. Every sequence produces a connected acyclic graph G with

e(G) = kGk 1 which must be a tree (or else add more edges to make a tree and
produce a contradiction).
Deduce that f is a bijection.
Definition 2.40 (Bipartite). A graph is bipartite with vertex classes X, Y if X and Y
partition V (G) and no edge of G lies within X or Y (every edge goes between them).
We say that X and Y are independent sets. We sometimes think of colouring X red
and Y blue.
Theorem 2.41 (Konig). A graph is bipartite iff it contains no odd cycles.
Proof. Any cycle alternates between the two vertex classes, so has even length.
We may suppose that the graph G is connected, since a graph is bipartite if its
components are bipartite.
Let the distance d(u, v) between u, v V (G) be the length of the shortest u v
path. Let u V (G). Define Ui = {v V (G) : d(u, v) = i} for i = 0, 1, 2, .
Note that an edge of G can join vertices in Ui and Uj only if j = i or j = i + 1 or
j = i 1.
Claim. There is no edge between vertices in Ui . Proof. If yy E(G) : y, y Ui
then select paths u y and u y of length i. Let w be the last common vertex. Then
w y,w y and yy form a cycle of length 2(i d(u, w)) + 1 contradicting absence
of odd cycles. S S S
S S S
Colour U0 U2 U4 red and U1 U3 U5 blue to give a bipartition
of G.
Definition 2.42 (Complete bipartite graph Km,n ). Km,n is the complete bipartite
graph with vertex classes of order m and n with all mn edges.
Definition 2.43 (Planar graphs). A graph that can be drawn in the plane without crossings is planar. A graph drawn in the plane is a plane graph.
Definition 2.44 (Face or country). If we omit (cut out) the vertices and edges of a
plane graph from the plane the remainder falls into connected components called faces
or countries.
Theorem 2.45 (Euler). Let G be a connected plane graph with n vertices, e edges and
f faces. Then n e + f = 2.
Proof. By induction on e = e(G). If e = n 1 then G is a tree and f = 1. Done.
If e > n 1 there is a cycle C in G. Then any xy E(C) separates two different
faces. Apply induction to G xy gives n (e 1) + (f 1) = 2.
Observation 2.46. It is convenient to use stereographic projection to consider a plane
graph as drawn on the sphere. Then G is connected iff all the faces are simply connected (homeomorphic to the unit disc).
Definition 2.47 (Bridge). A bridge in a plane graph is an edge whose removal increases the number of components.

7
Definition 2.48 (Bridgeless). In a bridgeless plane graph every edge separates two
different faces.
P
2e(G) = i ifi where fi is the number of i-sided faces.
Definition 2.49 (Girth). The girth of a graph G is the length of the shortest cycle.
Theorem 2.50 (Bound for number of edges in a planar graph). Let G be a bridgeless
g
(n 2). In particular a
plane graph with n vertices and girth g. Then e(G) g2
plane graph has at most 3n 6 edges.
Proof. Add edges ifPnecessary P
to ensure that G is connected. It remains bridgeless and
planar. Then 2e = i ifi g i fi = gf . So f 2e
g .
g2
By Eulers theorem n 2 = e f . But e f e 2e
g =
g e = e
g
(n

2).
g2
Theorem 2.51 (Kuratowski (unproved)). Kuratowski showed that the only non-planar
graphs are those that contain a subdivision of K5 or K3,3 obtained by replacing edges
with paths.
Theorem 2.52 (Eulerian condition). A graph is Eulerian iff it is connected and every
vertex has even degree.
Proof. If the graph is Eulerian it must be connected. If a vertex has odd degree the path
must pass in a different number of times from that which it passes out.
By induction on the number of edges. True for the empty graph. Since (G) 2
there are no leaves so G is not a tree. Therefore G must contain a cycle C. Each
component of G \ E(C) has vertices of even degree, so by induction hypothesis each
has an Euler circuit. By traversing C take time out to traverse each of these circuits
when first encountered. We produce an Euler circuit for G as G is connected.
Corollary 2.53. A connected graph G has an Euler trail from a vertex x to a vertex
y 6= x iff x and y are the only vertices of odd degree.
Proof. If there is an Euler trail then obvious.
If x and y are the only vertices of odd degree then form G by adding a new vertex
u and joining it to x and y. By the theorem G has an Euler circuit. Deleting u gives
an Euler trail from x to y.

CHAPTER 2. GRAPHS

Chapter 3

Flows, connectivity and

matching
Definition 3.1 (Flow). A flow in a digraph D with source s and sink t is a function
f : E(D) 7 R0 such that f+ (x) = f (x) x PV (D) \ {s, t} where f+ (x) =
P
y:yxE(D) f (yx) (the flow into
y:xyE(D) f (xy) (the flow out of x) and f (x) =
x). Convenient to set f (xy) = 0 if xy
/ E(G).
Definition 3.2 (Directed edge). xy is the directed edge from x to y. Loops and multiple
edges are forbidden, so certainly f (xy) = 0 or f (yx) = 0.
Observation 3.3. We imagine f as representing a flow of water or electricity in a
network, being pumped in at s, and being pumped out at t. The condition f+ (x) =
f (x) expresses that flow is conserved at each vertex.
Definition 3.4 (Value of the flow). v(f ) = f+ (s) f (s) = f (t) f+ (t) is called
the value of the flow where s is the source and t is the sink.
Definition 3.5 (Net flow). Let S V (D)
/ S. Let S = V (D) \ S.
P with s S and t

The net flow from S to S is f (S, S) = xS,yS (f (xy) f (yx)).

P
P

P Since x,yS (f (xy)f (yx)) = 0 we get f (S, S) = xS,yV (f (xy)f (yx)) =

xS (f+ (x) f (x)) = f+ (s) f (s) = v(f ).
Definition 3.6 (Capacity function). Given a capacity function c : E(D) 7 R0 we
require that 0 f (xy) c(xy) for all xy E(D).
Definition 3.7 (Cut). Let S V (D) with s S and t
/ S. Let S = V (D) \ S. The
= {xy E(D) : x S, y S}
is called a cut with capacity
set of edges
E(S,
S)
=P
c(S, S)
c(xy).
xS,yS
c(S, S).

For any cut we have v(f ) = f (S, S)

Theorem 3.8 (Max-flow min-cut (MFMC)). The maximum value of a flow (source s
and sink t) in a network is equal to the minimum cut capacity.
Proof. The maximum flow in a network is less than or equal to the minimum cut capacity. Let f be a flow. We construct S V (D). Initially let S = {s}. Then put
y S if either x S : f (xy) < c(xy) or x S with f (yx) > 0. Repeat.
If t S then there is a sequence of vertices starting with s = y0 , y1 , , yk = t
where for each 1 i k either c(yi1 , yi ) f (yi1 , yi ) = i > 0 or f (yi , yi1 ) =
9

10

CHAPTER 3. FLOWS, CONNECTIVITY AND MATCHING

i > 0. Let = min1ik i . We construct a new flow f : E(D) 7 R0 equal to

f except on the s t path but with f (yi1 , yi ) = f (yi1 , yi ) + or f (yi , yi1 ) =
f (yi , yi1 ) as appropriate. f is a flow. v(f ) = v(f ) + so f was not maximal,
contradiction.
Algorithm increases. Take a subsequence monotonic on each edge. Take the limit.
This flow has value of limit so it is maximal.
is a cut. For every xy with x S and
If t
/ S. Let S = V (D) \ S. E(S, S)
y S we have f (xy)
=
c(xy).
For
every
edge
yx,Px S y S we have f (yx) = 0.
=P
v(f ) = f (S, S)
(f
(xy)

f
(yx))
=

c(xy). That is, the value

xS,yS
xS,yS
of the flow is the cut capacity of some cut.
Corollary 3.9 (Integrability theorem). In a network with integer valued edge capacities
there is a maximal flow whose value is equal to the minimum cut capacity, which has
integer values at each edge.
Proof. Follow MFMC starting with the zero flow. Then is always integer.
Observation 3.10. Note that there may be a maximal flow which is not integer valued
on each edge.
Definition 3.11 (Multiple sinks and sources). Consider multiple sources s1 , , sk
Pk
and multiple sinks t1 , , tk . The value of the flow is v(f ) =
i=1 (f+ (si )
Pl

f (si )) = j=1 (f (tj ) f+ (tj )) and a cut is E(S, S) where s1 , , sk S and

t1 , , tk S.
Corollary 3.12 (MFMC for multiple sources and sinks). In a multiple source network
the maximum value of a flow is equal to the minimum cut capacity.
Proof. We construct a new network by joining a supersource s to s1 , , sk and a
supersink t to t1 , , tl using edges of infinite capacity.
Apply MFMC. Note that any cut of finite capacity cannot touch edges ssi or tj t so
cut can be applied to original graph.
Definition 3.13 (Vertex capacities). We can consider vertex capacities instead of edge
capacities. The capacity function is now C : V (D) \ {s, t} 7 R0 and we want
0 f+ (v) = f (v) c(v) v V (D) \ {s, P
t}. A cut is now a set of vertices S such
that D S has no s t path. It has capacity vS c(v).
Corollary 3.14 (Vertex capacity version of MFMC). The maximum value of a flow in
a network with vertex capacities is equal to the minimum cut capacity.
Proof. Construct from D a new network D with edge capacities. Replace each vertex
v by two vertices v and v+ and an edge v v+ . For each edge xy E(D) add an
edge x+ y to D . We give v v+ capacity c(v) give x+ y infinite capacity. Apply
MFMC.

Chapter 4

Connectivity and the theorems

of Menger
Definition 4.1 (Notation for subgraphs). If G graph and S V (G) then G S is
the induced subgraph with edges in S deleted. Given F E(G) then G F is the
spanning subgraph obtained by deleting edges in F .
Definition 4.2 (k connected). A graph is k connected if |G| > k and GS is connected
for any S V (G) with |S| < k.
Observation 4.3. The only k connected graph with k+1 vertices is the complete graph
Kk+1 .
Definition 4.4 (Connectivity). The connectivity of G or (G) = max{k : Giskconnected}.
Definition 4.5 (Local connectivity). If a, b V (G) distinct and ab
/ E(G) the local
connectivity (a, b; G) = min{k : S V (G) \ {a, b} : |S| = k : G S has no a b
path }, the minimum number vertices separating a and b.
Observation 4.6. If G is not complete then (G) = mina,b {(a, b; G)}, which is the
same as mina,b:ab,baE(G)
{(a, b; G)}.
/
If G is complete then mina,b {(a, b; G)} does not have a value, but (Kn ) = n1.
Definition 4.7 (k-edge connected). A graph G is k-edge connected if G F is connected for all F E(G) with |F | < k.
Definition 4.8 (Edge connectivity (G)). The edge connectivity of G is (G) =
max{k : G is k edge connected}.
Definition 4.9. If a, b V (G) distinct then the local edge connectivity is (a, b; G) =
min{k : F E(G) : |F | = k, G F has no a b path }. Note that (G) =
mina,b (a, b; G).
Definition 4.10 (Independent paths). Two a b paths are independent if their only
common vertices are a, b.
Theorem 4.11 (Menger). Vertex form. Let a, b V (G) distinct. ab
/ E(G) then
(a, b; G) = maximum number of pairwise independent a b paths.
Edge form. a, b V (G) distinct. Then (a, b; G) = maximum number of edge
disjoint a b paths.
11

12

CHAPTER 4. CONNECTIVITY AND THE THEOREMS OF MENGER

Proof. Vertex form. Replace all edges uv E(G) with two directed edges and give
each vertex capacity 1. Apply vertex form of max-flow min-cut to get an integer flow
from a b, (a, b; G) since each vertex has capacity 1 or 0.
Edge form. Do the same thing but use the edge form of max-flow min-cut.
Definition 4.12 (Line graph). The line graph L(G) is the graph with a vertex ve for
each e E(G) and an edge ve vf whenever e and f have a common end vertex in G.
Observation 4.13. The edge form can be deduced from the vertex from, using the line
graph.
Observation 4.14. Given a set P of independent a b paths then certainly |P |
(a, b; G), but it is not necessarily possible to add paths to P to form (a, b; G) independent paths.
Observation 4.15. There is a multiple source - sink version of Mengers theorem.
Lemma 4.16. Assume ab E(G) and let G = G ab. Then (a, b; G ) (G) 1.
Proof. Let k = (a, b; G ). Choose S V (G) \ {a, b} with |S| = k and G S
disconnected. Let x and y be vertices in different components of G S, so that
|{x, y} {a, b}| is minimal. Then either
S
1. a 6= x, y so G (S a) is disconnected; or
S
2. b 6= x, y so G (S b) is disconnected; or
S
3. {x, y} = {a, b} so V (G) = S {a, b}, so |G| k + 2.
All cases imply that (G) k + 1.
Corollary 4.17. A graph is k-connected iff any two vertices are joined by at least k
independent paths.
Proof. If G is not a complete graph, (G) = min (a, b; G). Apply Mengers theorem.
Kn is k-connected iff n k + 1, so true for a complete graph.
Corollary 4.18. A graph is k-edge-connected iff any two vertices are joined by at least
k edge disjoint paths.

Chapter 5

Matchings
Definition 5.1 (k-factor). A k-factor of a graph G is a k-regular spanning subgraph,
that is, a subgraph H with (H) = (H) = k.
Definition 5.2 (Matching). Let G be a bipartite graph with vertex classes X and Y . A
matching in G is a set of |X| independent edges.
If |X| = |Y | then a matching is a 1-factor.
Theorem 5.3 (Halls marriage theorem). Let G be a bipartite graph with vertex classes
X and Y . Then G has a matching iff |(S)| |S| for every S X.
Using Mengers theorem. Join a new vertex a to all elements of X and a new vertex b
to all elements of Y to form G .
Suppose C is a set of vertices separating a from b. Then (X \ C) Y C. Now
|C| = |C X| + |C Y |. So |C| |C X| + |(X \ C)|. By Halls condition
|(X \ C)| |X \ C|. So |C| |C X| + |X \ C| = |X|. By the choice of C,
(a, b; G) |X|.
Using Mengers theorem there are |X| independent paths, giving a matching in G.
Direct. By induction on |X|.
The case |X| 1 is trivial.
Suppose that for every S X with S 6= , X we have |(S)| > |S|. Then take any
xy E(G). G x satisfies Halls condition. By the induction hypothesis G {x, y}
has a matching, so G has a matching.
Otherwise
S there exists a critical set T S X, T 6= , X with |(T )| = |T |. Let
G1 = G[T (T )] and G2 = G[(X \ T ) (Y \ (T ))].
S
G1 clearly satisfies Halls condition. Let S X \ T . Now G2 (S) = G (S T ) \
G (T ).
|G2 (S)| |G (S
|S

T )| |G (T )|

T | |T | = |S|

So by induction hypothesis there is a matching on G1 and G2 , giving a matching

in G.

13

14

CHAPTER 5. MATCHINGS

Corollary 5.4 (Defect form of Halls theorem). Let d 0 be an integer. If |(S)|

|S| d for all S X then there is a matching of all but d elements of X.
Proof. Add d extra vertices to Y connected to all vertices of X. Then by Halls theorem
there is a matching. Remove the additional vertices, to make a matching of all but d
elements of X.
Corollary 5.5 (Polygamous form). Let d 1 be an integer. If |(S)| d|S| for all
S X, then we can match each x X to d elemens of Y , the different d element sets
being disjoint.
Proof. Replace each x X with d nodes connected to (x). Halls theorem gives a
matching.

Chapter 6

Extremal graph theory

Definition 6.1 (Monotone). A property of a graph is monotone if the whole graph has
the property when a subgraph does.
Definition 6.2 (Non-trival property). A property of a graph is non-trivial if the empty
graph does not have the property.
Definition 6.3 (Extremal problem). The study of the minimum size of a graph with a
monotone, non-trivial property, or the maximum size of a graph without it.
Theorem 6.4 (Condition for a graph to be Hamiltonian). Let G be a connected graph of
order n 3. Suppose that for every pair of non-adjacent vertices a, b, d(a)+d(b) k.
If k < n then G has a path of length k. If k n then G is Hamiltonian.
Note that if k n then the degree condition implies that G is connected.
Proof. Suppose that G is not Hamiltonian. Let P = x1 xl be a path of maximal
length.
Claim. G has no cycle of length l. Proof of claim. Suppose there is a cycle.
Obviously if l = n then G is Hamiltonian, contradiction. If l < n then there is a vertex
w not in the cycle. G is connected, so we can find a path from the cycle to w, giving a
path longer than l, contradiction.
In particular, x1 xl
/ E(G). Now by hypothesis d(x1 ) + d(xl ) k.
Let S = {2 i l : x1 xi E(G)}, T = {2 i l : xi1 xl E(G)}. If
j S T then x1 xj xj+1 xl xj1 xj2 x1 is a cycle of length l. So S T = .
S
Certainly l
/ S, T and S T {1, , l}, so in fact k d(x1 ) + d(xl ) =
|S| + |T | l 1.
If k < n then we have a path of length l 1 k, so a path of length k.
If k = n then we have a path of length n, which is a contradiction because it must
involve n + 1 vertices. Therefore G is Hamiltonian.
Corollary 6.5 (G.A. Dirac). If (G)

n
2

then G is Hamiltonian.

Theorem 6.6. Let G be a graph with n vertices and no path of length k. Then e(G)
(k 1) n2 with equality iff G is a union of copies of Kk .
Proof. By induction on n.

If n k then e(G) n2 (k 1) n2 with equality iff G = Kk .
15

CHAPTER 6. EXTREMAL GRAPH THEORY

16

Consider n > k. If G is disconnected apply induction to each component. Otherwise G is connected and Kk 6 G, or there would be a vertex w that could be joined to
a path traversing the Kk to make a path of length k.
By theorem 6.4 at least one v V (G) has d(v) k1
2 . By the induction hypothek1
sis e(G v) < k1
(n

1).
Therefore
e(G)
<
n.
2
2

6.1 Complete subgraphs and Turans theorem

Weve seen the maximum size of a graph containing no path of a certain length. What
is the maximum size of a graph containing no Kr ?
Definition 6.7 (r-partite graph). An r-partite graph is a graph with a vertex partition
V1 , . . . , Vr so that each Vi is an independent set (i.e. G[Vi ] is empty of edges).
Certainly no (r 1)-partite graph contains Kr .
What is the maximum size of a (r 1)-partite graph? Clearly we should look at
a complete (r 1)-partite graph, where any pair of vertices in different classes are
joined.
Suppose two class orders differ by more than 1, |Vi | |Vj | + 2. Moving a vertex
from Vi to Vj increases the number of edges by |Vi | |Vj | 1 1. Therefore the
classes are as equal as possible.
Definition 6.8 (Turan graph, Tr and tr ). The Turan graph Tr (n) is the complete rpartite graph of order n with classes orders nr or nr .
We write tr (n) = e(Tr (n)).
Theorem 6.9 (Turans theorem; maximum sized graph not containing Kr ). Let G be
a Kr free graph of order n with e(G) tr1 (n) then G = Tr1 (n).
Proof. The vertices in Tr (n) with least degree belong to vertex class of greatest order,
by removing one of these we get Tr (n 1).
tr1 (n) (Tr1 (n)) = tr1 (n 1)

(6.1)

Furthermore the degrees are as equal as possible.

(Tr1 (n)) (Tr1 (n)) 1

(6.2)

By induction on n. True for n r 1, because Tr1 (n) is Kn for n r 1.

In general let G G be a spanning subgraph with tr1 edges. Certainly G is also
Kr free. Now apply the handshaking lemma to G
n(G ) 2e(G )
= 2tr1
(Tr1 (n)) + (n 1)(Tr1 (n))
Using 6.2
n(Tr1 (n)) + n 1
(G ) (Tr1 (n))


6.1. COMPLETE SUBGRAPHS AND TURANS
THEOREM

17

Pick v V (G ) a vertex of minimum degree. Then G v is again Kr -free.

e(G v) = e(G ) (G )
= tr1 (n 1)
by 6.1.
So by the induction hypothesis, G v is a complete (r 1)-partite graph. In G , v
cannot be joined to all vertex classes in G v, since otherwise G would contain Kr .
So G is (r 1)-partite. But e(G ) = tr1 (n) and Tr1 (n) is the only (r 1)-partite
graph of that size, so G = Tr1 (n).
Adding any new edge to Tr1 (n) creates a Kr , so G = G = Tr1 (n).
Theorem 6.10 (Turans theorem; alternative forumulation and proof). Let G be a Kr free graph with vertex set V . Then there exists a (r 1)-partite graph H with vertex
set V , and dH (v) dG (v) for all v V .
Proof. By induction on r.
Case r = 2 trivial.
In general, we pick x a vertex of G of maximum degree.
Let G = G[(x)]. Then G is Kr1 free. Let V = (x).
By induction, there is a (r 2)-partite graph H with vertex set V and dH (v)

dG (v) v V . Form H by joining every vertex in V \ V to H .

Now construct H by joining every vertex in V \V to H . Then H is (r 1)-partite.
Need to show dH (v) dG (v) for all v V .
If v V \ V , dH (v) = kV k = dG (x). But dG (x) dG (v) v V .
Otherwise, v V . dH (v) = kV \ V k + dH (v). But dH (v) dG (v). Now
dG (v) dG (v) + kV \ V k, so dH (v) dG (v). Done.
Lemma 6.11 (Alternative formulation of Turans theorem implies original). Let G be
a Kr free graph of order n with e(G) tr1 (n) then G = Tr1 (n).
Proof. Only remains to show e(G) tr1 (n), as certainly Tr1 (n) is the unique
(r 1)-partite graph of size tr1 (n). By previous theorem there exists a (r 1)-partite
graph H with vertex
set V (G), and
P
PdH (v) dG (v) for all v V (G).
e(G) = 12 vV (dG (v) 21 vV dH (v) = e(H). Now certainly H is (r 1)partite, and tr1 (n) is the maximum number of edges of an (r 1)-partite graph. So
e(G) = e(H) tr1 (n).
Lemma 6.12 (Variant on the alternative formulation of Turans theorem). Let G be a
graph of order n, and let x V (G) of degree d = (G). If e(G[(x)]) tr2 (d),
then e(G) tr1 (n).
Proof. Let V = (x). Let H be a copy of Tr2 (d) with vertex set V . Form an
(r 1)-partite graph H by joining every vertex in V \ V to H . Certainly e(H)
tr1 (n).
Now e(H) e(H ) e(G) e(G[V ]), so if e(G[V ]) e(H ) then e(G)
e(H) tr1 (n).

18

CHAPTER 6. EXTREMAL GRAPH THEORY

Corollary 6.13 (Variant on alternative formulation of Turans theorem). If G has order

n and size e(G) > tr1 (n), then G contains Kr as a subgraph.
Proof. By induction on r. Case r = 2 is trivial.
In general we take a vertex x V (G) with degree d = (G). By the last lemma
e(G[(x)]) > tr2 (d) as e(G) > tr1 (n). By induction, G[(x)] contains Kr1 , so
G contains Kr .
Observation 6.14 (Method for finding Kr in a graph). This corollary gives us an
algorithm for finding Kr in a graph that has more edges than tr1 (kGk). Take vertex
xr of degree (G). Let Gr = G[(xr )]. Given xn , Gn with n > 1, take Gn1 to be
Gn [Gn (xn )] and xn1 to be a vertex of highest degree in Gn .
By the lemma, Gn , n > 1 contains Kn1 . Therefore G[x1 , , xr ]
= Kr .

6.2 The problem of Zarankiewicz

Definition 6.15 (Zarankiewicz problem, z(n; t)). The bipartite analogue of Turans
problem is to find the maximum number of edges z(n; t) in an (n, n)-bipartite graph
not containing a subgraph Kt,t .
The value of z(n; t) is not known.




n
Theorem 6.16 (Zarankiewicz). Let y = n1 z(n; t). Then yt t1
t .
n

Chapter 7

Graph colouring
7.1 Vertex colouring and Brooks theorem
Definition 7.1 (Vertex colouring). A vertex k-colouring of a graph G is a function
c : V (G) 7 {1, 2, , k}, such that c(x) 6= c(y) xy E(G).
We say G is k-colourable if there is a k-colouring of G. Clearly G is k-colourable
iff it is k-partite.
Definition 7.2 (Chromatic number). The chromatic number (G) of a graph G is the
minimum k for which G is k-colourable.
Definition 7.3 (Greedy algorithm). Given an ordering v1 , , vn of V (G), the greedy
algorithm colours the vertices sequentially, giving vertex vi the smallest colour in
{1, 2, , n} that is not in c((vi ) {v1 , v2 , , vi1 }).
Clearly the number of colours used by the greedy algorithm depends on the order of
the vertices. Furthermore given a colouring that uses only (G) colours it is possible
to order the vertices so that the greedy algorithm will use no more than (G) colours.
Lemma 7.4 (Upper bound on (G)). Given a graph G, (G) 1 + (G)
Proof. Given a vertex x, then the greedy colouring algorithm will not give it a colour
value more than d(x) + 1.
Therefore the greedy colouring algorithm colours all graphs with no more than
(G) + 1 colours, so (G) 1 + (G).
Theorem 7.5 (Upper bound on (G)). (G) 1 + maxHG (H) where H ranges
over all induced subgraphs of G (including G).
Proof. We constructively describe a way of colouring G, by specifying a vertex order
for the greedy colouring algorithm.
Let n = kV k. Let Hn = G. Let xn be a vertex in Hn with degree (Hn ). Certainly
(G) < 1 + maxHG (H). Let Hn1 = Hn {xn }, n > 1.
The sequence x1 , x2 , x3 , , xn presents to the greedy colouring algorithm a series of vertices which have had at most maxHG (H) neighbours already coloured.
Therefore at most 1 + maxHG (H) colours can be used.
This bound is certainly exact for a complete graph.

19

20

CHAPTER 7. GRAPH COLOURING

Theorem 7.6 (Brooks). If G is connected then (G) (G), unless G is complete

or G is an odd cycle.
Proof. Certainly by the greedy algorithm (G) (G) + 1.
Let = (G).
If (G) = + 1 then by 7.5 there is a subgraph H of G with (H) = (H is
-regular). But G is connected and no vertex not in H can connect to H, so G = H
and G is -regular.
Now suppose Delta = 1. Then G = K2 . Suppose = 2. Then G = C3 .
Therefore we need only consider 3.
Sorry, but the rest of the proof will be omitted. The method is to take a vertex of
degree (the minimal degree) and as in the proof of Vizings theorem, consider the
components Hi j of vertices coloured either i or j and the relationship its neighbours.
By considering switching i,j in these components one can show that the neighbours
are pairwise joined.
Definition 7.7 (Chromatic polynomial). Let G be a graph. Let PG (x) be the number
of ways of colouring G using x colours.
Definition 7.8 (Notation for contracting edges G/e). Let G be a graph. Let e E(G).
Then G/e is the graph obtained by identifying the endpoints of e. That is if e joins
vertices x, y then for each edge yz join z to x if z is not already joined to x. Then
remove y.
Theorem 7.9 (Induction step on the chromatic polynomial). Let e E(G). Then
PG (x) = PGe (x) PG/e (x).
Proof. Let e = uv. The colourings c of G e that are not colourings of G are those
with c(u) = c(v), which are precisely the colourings of PG/e (x).
Corollary 7.10 (The chromatic polynomial is a polynomial). PG (x) is a polynomial
in x. Furthermore, PG (x) = xn a1 xn1 + a2 xn2 + + (1)n an with n = kGk,
a1 = e(G) and all ai 0.
Proof. By induction on e(G). True for the empty graph. In general let e E(G).
Now PGe (x) = xn b1 xn1 + b2 xn2 + + (1)n bn , PG/e (x) = xn1
n2
c1 x
+ (1)( n 1)cn1 , where n = kGk, b1 = e(G) 1 and all bi , ci 0.
Now by the theorem, PG (x) = PGe (x) PG/e (x), so PG (x) = xn (b1 +
1)xn1 + (c1 + b2 )xn2 + + (1)n (bn + cn1 ), a polynomial of the same form.

Chapter 8

Edge colouring and Vizings

theorem
Definition 8.1 (Edge colouring). A k-edge colouring of a graph G is a function c :
E(G) 7 {1, 2, , k} such that incident edges receive different colours.
Definition 8.2 (Edge chromatic number, chromatic index). Given a graph G, the edge
chromatic number or chromatic index (G) is the least k for which G is k-edgecolourable.
Certainly (G) (G).
Theorem 8.3 (Vizing). (G) (G) (G) + 1
Proof. The lower bound is trivial. For the upper bound we do induction on the number
of edges.
Suppose we have a colouring of all but one edge xy E(G) using colours {1, 2, , (G)+
1}. Then we wish to recolour so all the edges are coloured.
Note that one colour is unused (missing) at every vertex.
Let xy0 be the uncoloured edge. We construct a sequence of edges xy0 , xy1 ,
and a sequence of colours c0 , c1 , as follows.
Pick ci to be a colour missing at yi . Let xyi+1 be an edge with colour ci . We stop
with k = i when either ck is a colour unused at x, or ck is already used on xyj for
j < k.
If ck was a colour unused at x then we recolour xyi with ci for 0 i k. This
finishes the easy case where we can recolour the edges touching x to give a a colouring
for G.
Otherwise we recolour xyi with ci for 0 i < j and uncolour xyj . Notice that ck
(red) is missing at both yj and yk . Let blue be a colour unused at x.
1. If red is missing at x, we colour xyj red.
2. If blue is missing at yj we colour xyj blue.
3. If blue is missing at yk we colour xyi with ci for j i < k and colour xyk blue.
(None of the xyi , j i < k are red or blue.)
If none of the above hold, then we consider the subgraph of red and blue edges.
The components of this subgraph are paths or cycles. The vertices x, yi , yk are the end
vertices of paths. Therefore they cannot all belong to the same component.
21

22

CHAPTER 8. EDGE COLOURING AND VIZINGS THEOREM

Select a component that contains exactly one of these vertices. Now swap over red
and blue in this component. Now one of the conditions above must apply.
Theorem 8.4 (Edge chromatic number of a bipartite graph). If G is bipartite, then
(G) = (G).
Proof. We embed G in a (G)-regular bipartite multi-graph as follows: We replace
G by two copies of G and for v , v , the copies of v V (G), we join v to v with
(G)d
S G (v) parallel
S edges. This creates a bipartite multi-graph H with vertex classes
(X Y ) and (Y X ) if X and Y were the original vertex classes in G.
Now we prove the theorem for -regular bipartite multi-graphs H by induction on
kHk + (H).
Clearly true for (H) = 1.
Let uv be an edge of H, Delete the vertices u and v. Because u and v were in
different vertex classes, it is possible to add fewer than new edges to make a new
-regular bipartite multi-graph H . Now we colour H by the induction hypothesis.
Certainly not all the colours were used to colour the new edges. Let red be one of these
colours. Certainly the red edges in H with uv form a 1-factor in G. Deleting this
1-factor gives a ( 1)-regular bipartite multigraph H .
Now colour H by the induction hypothesis, then add the 1-factor back, to obtain
a colouring of H.
Definition 8.5 (List colouring, L-choosable). Let L : V (G) 7 { finite subsets of N
}, so that each vertex v has a paint box L(v). We say that G is L-choosable if
c : V (G) 7 N such that c(v) L(v) v and c is a vertex colouring.
Definition 8.6 (k-choosable). We say that G is k-choosable if G is L-choosable whenever kL(v)k = k, v V (G).
Clearly if G is k-choosable it is k + 1-choosable. However it is not necessarily true
that if G is k-choosable it is k-colourable.
Definition 8.7 (List chromatic number l (G)). The list chromatic number l (G) of a
graph G is the least k such that G is k-choosable.

Chapter 9

Colouring graphs on surfaces

9.1 Plane graphs
Definition 9.1 (Dual graph G). Given a plane graph G we can form a new graph G
by drawing a vertex in the middle of each of Gs faces, and joining vertices if their
corresponding faces are adjacent.
Lemma 9.2 (Condition for duality of dual). G = G iff G connected and (G) > 2.
Proof. If G is always connected, so G is as well, so G must be.
Always e(G) e(G) as each edge in the original graph may be crossed at most
once by an edge in the dual. If (G) is 1 then G has a bridge, which certainly will not
be crossed by an edge in the dual, so e(G) < e(G). If (G) is 2 then there is an edge,
which if removed, would leave a bridge in G. so again e(G) < e(G). In both cases
e(G ) e(G) < e(G).
If (G) 3 then any pair of faces has at most a single common boundary edge so
G = G.
Definition 9.3 (Plane map). A plane map is a plane graph together with its set of faces.
Definition 9.4 (Face colouring). A (face) colouring of a plane map is an assignment
of colours to the faces of the map with the condition that faces sharing an edge have
different colours.
Example 9.5. Let G be a plane graph with every vertex of even degree. Then every
face in the dual has an even number of sides.
Therefore every cycle in the dual is even, so the dual is bipartite, so there is a face
colouring of G with just two colours.
Definition 9.6 (Four colour conjecture). The four colour conjecture (4CC) asserts that
every plane map can be 4-face coloured. Alternatively every plane graph has (G) 4
by considering the dual.
The 4CC was made popular by Cayley in 1878. Almost at once, proofs appeared:
Kempe 1879, Tait 1880.

23