Introduction to Network

Theory:
Modern Concepts, Algorithms
and Applications
Ernesto Estrada
Department of Mathematics, Department of Physics
Institute of Complex Systems at Strathclyde
University of Strathclyde
www.estradalab.org
Types of graphs
 Weighted graphs
 Multigraphs
 Pseudographs
 Digraphs
 Simple graphs
 Weighted graph

is a graph for which each edge has an associated

weight, usually given by a weight function
w: E  R, generally positive
Adjacency Matrix of Weighted
graphs
A B C D E
A 0 1.5 0 0 0
B 1.5 0 3.4 0 0
C 0 3.4 0 2.1 0.5
D 0 0 2.1 0 0.7
E 0 2.1 0.5 0.7 0
 Degree of Weighted graphs
The sum of the weights associated to every edge
incident to the corresponding node
The sum of the corresponding row or column of
the adjacency matrix
A B C D E Degree
A 0 1.5 0 0 0 1.5
B 1.5 0 3.4 0 0 4.9
C 0 3.4 0 2.1 0.5 6
D 0 0 2.1 0 0.7 2.8
E 0 2.1 0.5 0.7 0 3.3
 Multigraph or pseudograph

is a graph which is permitted to have multiple

edges. Is an ordered pair G:=(V,E) with
V a set of nodes
E a multiset of unordered pairs of vertices.
Adjacency Matrix of Multigraphs

A B C D E
A 2 1 0 0 0
B 1 0 3 0 4
C 0 3 0 1 1
D 0 0 1 0 2
E 0 4 1 2 0
Directed Graph (digraph)
 Edges have directions
 The adjacency matrix is not symmetric

A B C D E
A 0 1 0 0 0
B 0 1 0 0 2
C 0 1 0 0 1
D 0 0 1 0 1
E 0 0 0 1 0
 Simple Graphs
Simple graphs are graphs without
multiple edges or self-loops. They are
weighted graphs with all edge weights
equal to one. A

B C

D E
 Local metrics

 Local metrics provide a measurement of a

structural property of a single node
 Designed to characterise
 Functional role – what part does this node
play in system dynamics?
 Structural importance – how important is this
node to the structural characteristics of the
system?
Degree Centrality
degree
A A B C D E
A 0 1 1 0 0 1
B C
B 1 0 1 1 1 4
C 1 1 0 0 0 3
D 0 1 0 0 0 1
D E E 0 1 0 0 0 1
 Betweenness centrality

 The number of shortest paths in the graph

that pass through the node divided by the
total number of shortest paths.

  i, k , j 
BC  k    , i jk
i j   i, j 
 Betweenness centrality
 Shortest paths are:
 AB, AC, ABD, ABE, BC, BD,
A BE, CBD, CBE, DBE
  A, B, D   1;   A, D   1
B C   A, B, E   1;   A, E   1
  C , B, D   1;   B, D   1
  C , B, E   1;   C , E   1
D E
  D, B, E   1;   D, E   1
 B has a BC of 5
 Betweenness centrality

 Nodes with a high betweenness centrality

are interesting because they
 control information flow in a network
 may be required to carry more information

 And therefore, such nodes

 may be the subject of targeted attack
 Closeness centrality

 The normalised inverse of the sum of

topological distances in the graph.

N 1
CC  i  
 d  i, j 
j
Closeness centrality
n

A A B C D E
 d  i, j 
j 1

A 0 1 1 2 2 6
B C B 1 0 1 1 1 4
C 1 1 0 2 2 6
D 2 1 2 0 2 7
D E E 2 1 2 2 0 7
Closeness centrality

A Closeness
0.67
B C 1.00
0.67
0.57
D E
0.57
 Closeness centrality

Node B is the most central one in spreading

information from it to the other nodes in the
network.
Local metrics

A
Node B is the most central one
according to the degree,
B C
betweenness and closeness
centralities.

D E
and the winner is…

A is the most central

according to the
A
degree

B B is the most central

according to closeness
and betweenness

Which is the most central node?

Degree: Difficulties
 Extending the
Concept of Degree
Make xi proportional to the average of the centralities
of its i’s network neighbors
1 n
xi   Aij x j
 j 1

where  is a constant. In matrix-vector notation we

can write 1
x Ax

In order to make the centralities non-negative we select
the eigenvector corresponding to the principal eigenvalue
(Perron-Frobenius theorem).
 Eigenvalues and Eigenvectors
 The value λ is an eigenvalue of matrix A if
there exists a non-zero vector x, such that
Ax=λx. Vector x is an eigenvector of
matrix A
 The largest eigenvalue is called the principal
eigenvalue
 The corresponding eigenvector is the principal
eigenvector
 Corresponds to the direction of maximum
change
 Eigenvector Centrality

 The corresponding entry of the principal

eigenvector of the adjacency matrix of the
network.

 It assigns relative scores to all nodes in the

network based on the principle that
connections to high-scoring nodes
contribute more.
Eigenvector Centrality
Node EC

1 0.500
2 0.238
3 0.238
4 0.575
5 0.354
6 0.354
7 0.168
8 0.168
Eigenvector Centrality:
Difficulties

In regular graphs all the

nodes have exactly the
same value of the
eigenvector centrality,
which is equal to 1 n
 Subgraph Centrality
A closed walk of length
k in a graph is a succession
of k (not necessarily
different) edges starting
and ending at the same
node, e.g.
1,2,8,1 (length 3)
4,5,6,7,4 (length 4)
2,8,7,6,3,2 (length 5)
 Subgraph Centrality
 0 1 
The number of closed  
walk of length k starting A   1 0 
  
at the same node i is given  
by the ii-entry of the kth
  2 1 
power of the adjacency  
A 
2
 2  2 
matrix 
  

k  i    A k
 ii
  3 1

A 
3
 3  2



  

 Subgraph Centrality

We are interested in giving weights in decreasing

order of the length of the closed walks. Then,
visiting the closest neighbors receive more weight
that visiting very distant ones.
The subgraph centrality is then defined as the
following weighted sum
 
EE  i   c0 A0 ii  c1  A  ii  c  
2 A 2
ii  c3 
A 3
ii  c  
4 A 4
ii 

  cl l  i 
l 0
 Subgraph Centrality

By selecting cl=1/l! we obtain

l  i 
 

EE  i    EE  i   e A ii
l 0 l!

where eA is the exponential of the adjacency

matrix.
For simple graphs we have
 
n
EE  i    x j  i  e
2 j

j 1
Subgraph Centrality

Nodes EE(i)
1,2,8 3.902
4,6 3.705
3,5,7 3.638
Subgraph Centrality:
Comparsions
DC (i )  DC ( j ), i  V , j  V
CC (i )  CC ( j ), i  V , j  V
EC (i )  EC ( j ), i  V , j  V

Nodes BC(i) Nodes EE(i)

1,2,8 9.528 1,2,8 3.902
4,6 7.143 4,6 3.705
3,5,7 11.111 3,5,7 3.638
Subgraph Centrality:
Comparisons
DC (i )  DC ( j ), i  V , j  V
CC (i )  CC ( j ), i  V , j  V
BC (i )  BC ( j ), i  V , j  V
EC (i )  EC ( j ), i  V , j  V

Nodes EE(i)
45.696

45.651
 Communicability

Path of length 6 Walk of length 8

Shortest path
 Communicability
s
P
Let pq be the number of shortest paths of length s
between p and q.
k
Let W pq be the number of walks of length k>s
between p and q.

DEFINITION (Communicability):

G pq  bs P   c k W
s
pq
k
pq
k s
bs and ck must be selected such as the communicability converges.
 Communicability

By selecting bl=1/l! and cl=1/l! we obtain

G pq  
 A 
k
pq
 
 e A
pq
k 0 k!
where eA is the exponential of the adjacency
matrix.
For simple graphs we have
n
G pq   x j  p  x j  q  e
j

j 1
Communicability

j  1 1  p   0 p  V
 Communicability

j2

q
q
p
p

sgn  j  p   sgn  j  q  sgn  j  p   sgn  j  q 

 Communicability
intracluster
              
 
G pq  1  p 1  q  e    j  p  j  q  e    j  p  j  q  e
1 j j

j 2 j 2
 
   j  p  j  q  e    j  p  j  q  e
j j

j  2       j  2     
intercluster

intra  cluster inter  cluster

   p   q  e    p  q  e
j j
G pq  j j  j j
j 2 j 2
 Communicability &
Communities
A community is a group of nodes for wich the
intra-cluster communicability is larger than the
inter-cluster one
intra  cluster inter  cluster

   p   q  e    p   q  e
j j
j j  j j
j 2 j 2

These nodes communicates better among them

than with the rest of extra-community nodes.
 Communicability Graph

1 if x  0
Let x   
0 if x  0

The communicability graph (G)is the graph

whose adjacency matrix is given by (G)) results
from the elementwise application of the function
(G)to the matrix (G).
 Communicability Graph

 G 
Δ G    Δ  G   

ΔG pq  0,1
communicability
graph
 Communicability Graph

A community is defined as a clique

in the communicability graph.

Identifying communities is reduced

to the “all cliques problem” in the
communicability graph.
Communities: Example

Social (Friendship) Network

 Communities: Example

The Network

Its Communicability
Graph
Communities

Social Networks Metabolic Networks

 References
Aldous & Wilson, Graphs and Applications. An
Introductory Approach, Springer, 2000.
Wasserman & Faust, Social Network Analysis,
Cambridge University Press, 2008.
Estrada & Rodríguez-Velázquez, Phys. Rev. E
2005, 71, 056103.
Estrada & Hatano, Phys. Rev. E. 2008, 77,
036111.
 Exercise 1
Identify the most central node according to the following
criteria:
(a) the largest chance of receiving information from closest
neighbors;
(b) spreading information to the rest of nodes in the
network;
(c) passing information from some nodes to others.
 Exercise 2
T.M.Y. Chan collaborates with 9 scientists in
computational geometry. S.L. Abrams also collaborates with
other 9 (different) scientists in the same network. However,
Chan has a subgraph centrality of 109, while Abrams has 103.
The eigenvector centrality also shows the same trend,
EC(Chan) = 10-2; EC(Abrams) = 10-8.

(a) Which scientist has more chances of being informed about

the new trends in computational geometry?
(b) What are the possible causes of the observed differences
in the subgraph centrality and eigenvector centrality?
Exercise 2. Illustration.