A BRIEF OVERVIEW OF GRAPH THEORY: ERDOS-RENYI

RANDOM GRAPH MODEL AND SMALL WORLD

PHENOMENON

JIATONG LI (LOGEN)

Abstract. This paper briefly introduces graph theory, Erdos-Renyi random

graph model, small world phenomenon, and its application in the engineer-
ing industry as supplemental material. We begin with basic definitions and
notations in graph theory, then move to graph networks’ properties. Then,
finally, we utilize the definitions and theorem we introduced to analyze the
Erdos-Renyi random graph model and the small world phenomenon. At the
end of the paper, we also point out some further applications of graph theory.

Contents
1. Introduction 1
2. Basic Definitions and Notations 2
3. Properties of Graph Networks 3
4. Random Graph Model 5
5. Small World Phenomenon 6
6. Application 8
6.1. Analyzing Degree Distribution of Random Graph Model 8
6.2. Analyzing Clustering Coefficient of Random Graph Model 8
6.3. Application of Graph Theory in Computer Science 8
7. Acknowledgements 9
References 9

1. Introduction
A graph is a structure which contains a set of objects - vertices and edges. A
basic example is when we connect two vertices with a line, we create a simple graph.
However, such a simple network can quickly become very complex to analyze as
the number of vertices and edges increases. To simulate what’s happening, we use
a random graph as a model in this paper. We will also explore some extensions of
random graph model such as small world phenomenon.
We begin with basic definitions and notations of graph theory. Section 3 will
introduce four important properties of graph networks: degree distribution, path
length, clustering coefficient, and connected component. Then we apply the knowl-
edge from Section 2 and Section 3 to analyze Erdos-Renyi random graph. Next, we
discuss the small world phenomenon - the principle that we are all connected by
our nearby neighbors. Finally, we zoom out and briefly mention the application of
graph theory in other subjects such as engineering.
1
 A BRIEF OVERVIEW OF GRAPH THEORY

2. Basic Definitions and Notations

Definition 2.1. Let V (G) and E(G) denote the vertex and the edge set of simple
graph G, respectively.
Definition 2.2. The order of a graph G is the cardinality of the vertex set, denoted
as |V(G)|. The size of G is defined as the cardinality of the edge set, denoted as
|E(G)|.
Example 2.3. In Figure 1, V (G) = {1, 2, 3, 4} and E(G) = {13, 23, 12, 24}. Based
on Definition 2.2, the graph has an order of 4 and a size of 4.

Figure 1

Definition 2.4. The degree of a vertex is the number of edges that are connected
to that vertex. When the degree of v is 5, we write as deg(v) = 5. The minimum
degree of G is symbolized by δ(G), and the maximum degree is symbolized by ∆(G).
Definition 2.5. The degree sequence of G is a list of degrees of all vertices. By
convention, we list them in a nonincreasing order.
Theorem 2.6. The sum of the degrees of all vertices is twice the size of the graph.
X
deg (vi ) = 2|E(G)|

Lemma 2.7. (Pigeonhole Principle)

If we put n pigeons in less than n pigeonhole, at least one pigeonhole is going to
contain more than one pigeons.
Definition 2.8. If an edge e joins two vertices v1 and v2 , we say the two vertices
are adjacent. If an edge e joins two vertices v1 and v2 , we say edge e is incident
upon v1 and v2 .
Definition 2.9. A simple graph is a graph that doesn’t have multiple edges and
loops. A loop is created when the starting vertex and ending vertex are the same.
Definition 2.10. A complete graph is a graph in which every pair of vertices is
connected by a unique edge.
Example 2.11. Figure 2 is a complete connected graph, whereas Figure 1 is a
simple graph.

Lemma2.12.
 The maximum number of edges in a connected graph H equals to
N
Emax = .
2
 A BRIEF OVERVIEW OF GRAPH THEORY

Figure 2

Proof. The proof is relatively intuitive. The maximum number equals to “N choose
2” (choose 2 vertices among N vertices). 
Definition 2.13. A walk is a finite number of sequences in the form of v0 v1 , v1 v2 ,
· · · , vm−1 vm . One can also write v0 → v1 → v2 → · · · . By definition, v0 is called
the initial vertex, and vm is called the final vertex. The number of edges in the
walk is called length.
Definition 2.14. In Definition 2.13, if the vertices are distinct, then the walk can
be called a path. Moreover, when v0 = vm , the walk is a cycle.
Definition 2.15. An undirected graph means the edge is bidirectional. By con-
trast, if the edge of a graph points to a specific direction, it is called directed graph.

Figure 3

Definition 2.16. The adjacency matrix A of a graph G is n×n, where n = |V (G)|,

and is defined as: 
1 vi ∼ v j
Ai,j =
0 otherwise
3. Properties of Graph Networks
Definition 3.1. We define the degree distribution P (k) of a graph as the proba-
bility of a randomly chosen node has degree k. If a graph has Nk vertices of degree
Nk
k, then P (k) = .
|V (G)|
Example 3.2. Figure 4 is a simple graph and we plot its degree distribution bar
1
plot in Figure 5. For example, the degree distribution of k = 1 equals to since
6
there is only one node with degree 1 among the six nodes.
 A BRIEF OVERVIEW OF GRAPH THEORY

Figure 5. Degree dis-

Figure 4. A simple graph
tribution bar plot of
Figure 4

Definition 3.3. The distance between a pair of nodes is defined as the shortest
path connecting the nodes. If two nodes are not connected, the distance is defined
as infinite.
Example 3.4. In Figure 4, h12 = 2 and h13 = 3.
2ei
Definition 3.5. We define the clustering coefficient of a graph as Ci =
ki (ki − 1)
where ki is a degree of node i and ei is the number of edges among the adjacent
vertices of node i.
ki (ki − 1)
Remark 3.6. In Definition 3.5, we should notice that is the maximum
2
number of edges among the adjacent vertices of node i. Note that Ci ∈ [0, 1].

Figure 6

Example 3.7. In Figure 6, the vertex set of the graph G(V, E) is V (G) = {0, 1, 2, 3, 4, 5, 6, 7, 8}.
The graph G contains two subgraphs G1 and G2 where V (G1 ) = {0, 1, 2, 3, 4, 5} and
V (G2 ) = {6, 7, 8}. Since G2 contains more elements, it is the largest component of
graph G.
 A BRIEF OVERVIEW OF GRAPH THEORY

Remark 3.8. In computer science, there is a special type of algorithm called Depth-
first search (DFS) which can calculate the number of connected components in a
given graph.
Algorithm 1: Finding the number of connected components using DFS
input : An undirected graph G(V, E)
output: An int value which indicates the number of connected components
count = 0;
for each vertex k ∈ V do
Visited[k] = false;
end
/* Initialize all nodes to status "unvisited" */
for each vertex k ∈ V do
Visited[k] = false;
if Visited[k]==false then
DF S(V, k);
count = count + 1;
end
end
Print count;
/* Recursively enumerate all nodes and count the number of connected component
*/
DF S(V, k) function: Visited[k] = true;
for each vertex q ∈ adjacent − neighbor − V [k] do
if Visited[q] == false then
DF S(V, q);
end
end

4. Random Graph Model

   
n
Suppose Tn,N has n possible vertices and N edges, then  2  possible
N
graphs can be formed. We define the following variable Nc to simplify the compu-
tation.
1
Nc = n log n + cn
2
We define a graph is type A if it has k isolated vertices and n − k effective vertices.
Everything that is not type A is type Ā.
Theorem 4.1. Let P0 (n, Nc ) denote the probability of Tn,N being connected. Then
−2c
lim P0 (n, Nc ) = e−e
n→+∞

Proof. Let N (n, Nc ) denote the number of connected graph, which is equal to the
number of graphs of type A without isolated points.
   
n   n−k
X n 
N (n, Nc ) = (−1)k 2 
k
k=0 Nc
 A BRIEF OVERVIEW OF GRAPH THEORY

Thus we obtain
∞
X (−1)k e−2kc
N (n, Nc ) −2c
lim    = = e−e
n→∞ n k!
k=0
 2 
Nc

Theorem 4.2. Define Pk (n, Nc ) as the probability that the greatest connected com-
ponent of Tn,N which contains exactly n − k effective points. Then we have

k −2c
e−2c e−e
lim Pk (n, Nc ) =
n→∞ k!
Proof. Suppose the graph has k isolated points and n − k effective points. It follows
that
 
n−k
 
n 2
Pk (n, Nc ) ∼     P0 (n − k, Nc ) .
k n
 2 
Nc

For a fixed value k, we have
Nc − 21 (n − k) log(n − k)
lim =c
n→∞ n−k
Combined with Theorem 1, the theorem is proved.
Theorem 4.3. Let Πk (n, Nc ) denote the probability of Tn,Nc consisting exactly
k + 1 connected components. Then

k 2c
e−2c e−e
lim Πk (n, Nc ) =
n→+∞ k!
Proof. The proof of the theorem is out of scope of this paper. Detailed proof of
Theorem 4.3 can be found in [1]. 

5. Small World Phenomenon

5.0.1. Six Degrees of Separation.
Before we dive into the generalized network model, I want to provide a little
background on the origin of small world phenomenon, also known as ”six degrees of
separation”. The idea is the distance between you and another person on the planet,
on average, is exactly six people away. Initially, this conclusion seems unrealistic,
but we can analyze the argument in this way: suppose you have 100 connections,
and each person you know also has 100 connections. After two steps, there are
100 × 100 = 10, 000 people. After five steps, there are 10 billion connections in
total. Mathematically, there is nothing wrong with our analysis. However, the
conclusion can barely help us to understand what a real social network is like. This
is why we need to introduce the concept of analyzing decentralized search.
 A BRIEF OVERVIEW OF GRAPH THEORY

Figure 7. In myopic search, the message carrier will choose the

contact that is closest to the target

5.0.2. Analysis of Decentralized Search - Myopic Search.

In Figure 7, we have a series of nodes arranged into an irregular ring shape. The
ring indicates that for a particular node, it is connected to its closest two neighbors.
For example, 3 is connected to both 2 and 4. The red arrow implies the path if
person 2 wants to find person 8.
The principle of myopic search is: when a specific node v has an information, it
passes to the contact that lies as close to the target node as possible. For example,
if we choose 2 as the initial node and 8 as the target node.
(1) Node 2 first send message to node 4 since compared to node 3, node 4 lies
closer to node 8.
(2) Node 4 passes to node 5 since there is no other way.
(3) Node 6 finally reaches node 8 which is our target node.
It is worth mentioning that myopic search is not the shortest path. Since we lack
the information of the overall structure of the diagram, it is possible to find other
more convenient paths. For example, if 2 is connected to 7, then the path will be
2-7-8 instead of 2-4-5-6-8.
Suppose the number of steps by myopic search is X. Define the expected value
of X is E[X]. When a message starts from the initial node and head to the target
node, we say the node is in phase j if its distance from the target node is between
2j and 2j+1 . Notice the maximum number of phase is log2 n. Therefore, we have:

Xf ullsearch = X1 + X2 + · · · + Xlog n

E[X] = E [X1 + X2 + · · · + Xlog n ] = E [X1 ] + E [X2 ] + · · · + E [Xlog n ]

5.0.3. Find the Normalizing Constant.

Let u be the initial node and v be the target node. We say the probability of u link
to v is proportional to d(u, v)−1 . The proof process is based on empirical experiment
and we will not discuss the detail in our paper. Like any other probability problem,
there is a missing constant 1/Z in our equation.
To find the value of Z we can think in this way: if n = 6 (6 nodes in total),
 A BRIEF OVERVIEW OF GRAPH THEORY

 
1 1
Z ≤2 1+ +
2 3
Therefore, assuming n is even,
  Z k !
1 1 1 1 1
Z ≤ 2 1 + + + + ··· + ≤2 1+ dx = 2 (1 + ln k)
2 3 4 n/2 1 x

We know since ln x ≤ log2 x,

Z ≤ 2 log2 n
Therefore, the probability equals to:
1 1
d(v, w)−1 ≥ d(v, w)−1
Z 2 log n

6. Application
6.1. Analyzing Degree Distribution of Random Graph Model.
We already introduced the definition of degree distribution in Definition 3.1.
Now we can use it to analyze the degree distribution of a random graph. Let P (k)
represents the fraction of nodes with degree k. Then
 
n−1
P (k) = pk (1 − p)n−1−k
k
Notice the degree distribution is binomial. We can come up with the mean and
variance of a binomial distribution easily as follows.

k̄ = p(n − 1)
σ 2 = p(1 − p)(n − 1)

6.2. Analyzing Clustering Coefficient of Random Graph Model.

For a specific random graph

ki (ki − 1)
E[ei ] = p
2
Plug in ei into the equation in Definition 3.5, we have

p · ki (ki − 1)
E[Ci ] = =p
ki (ki − 1)

6.3. Application of Graph Theory in Computer Science.

Since graph theory can perfectly visualize the concept of ”relationship” in a
complex network, computer scientists have integrated graph structure in a variety
of fields such as graph database management, data mining, and graph machine
learning. A famous product is the initial Google PageRank algorithm which treats
web pages as nodes and web links as edges. By computing the credibility of website,
Google assigns weight to each edges. Clearly, the application of graph theory have
drastically increased the accuracy of early day search engine.
 A BRIEF OVERVIEW OF GRAPH THEORY

7. Acknowledgements
It is my great pleasure to thank my advisor Sehyun Ji for all the help during the
REU program. Sehyun is extremely helpful in helping me understand the weekly
challenge problems and graph theory. I also want to thank Peter May for organizing
the REU program.

References
[1] P. Erdos, A. Renyi, On Random Graphs I, Publ. Math. Debrecen, 1959.
[2] David Easley, Jon Kleinberg, Networks, Crowds, and Markets: Reasoning about a Highly
Connected World, Cambridge University Press, 2010.
[3] Robin J. Wilson, Introduction to Graph Theory, Prentice Hall, 1996.
[4] https://www.baeldung.com/cs/graph-connected-components
[5] snap.stanford.edu/class/cs224w-2019/