Contents

I RANDOM WALKS
1.1 The plot of the mean-squared displacement < R 2 > versus Time, for an
one dimensional system.
1.2 The plot of the mean value of the displacement < R > versus Time, for
an one dimensional system.
1.3 The plot of the distribution of R for N= 500 steps and N=1000 steps.
1.4 Trapping on random walks in two dimensional lattice.
1.5 The distribution of trapping times.
1.6 The survival probability.
1.7 Simulations for calculating S N .
1.8 The plot of S N versus Time for 1-D.
1.9 The plot of S N versus Time for 2-D.
1.10 The plot of S N versus Time for 3-D.
1.11 The plot of the distribution of S N for 1-D.
1.12 The plot of the distribution of S N for 2-D.
1.13 The plot of the distribution of S N for 3-D.
1.14 The plot of the relation of < S N > versus Time for 1-D.
1.15 The plot of the relation of < S N > versus Time for 2-D.
1.16 The plot of the relation of < S N > versus Time for 3-D.

II NETWORKS
2.1 What are networks?
2.1.1 Basic notions
2.1.2 Adjacency matrix
2.1.3 The adjacency matrix for a random network
2.1.4 Degree distribution
2.2 Random networks, the Erdős- Rényi model
2.2.1 The plot of the distribution of random networks
2.3 Scale-free networks
2.3.1 Characteristics of scale-free networks for g = 2.0, 2.5,3.0
2.3.2 Plot of the distribution of scale-free networks for g = 2.0
2.3.3 Plot of the distribution of scale-free networks for g = 2.5
2.3.4 Plot of the distribution of scale-free networks for g = 3.0
2.3.5 The algorithm of constructing the scale-free network
2.3.6 The Breadth-First search for finding the greatest cluster for the scale-
free networks
2.4 The Barabasi-Albert model
2.4.1 The plot of the distribution of B-A model
2.5 Real networks

1
 2.5.1 Internet robustness
2.5.2 The plot of distribution of the Internet
2.6 Resilience of scale-free networks and internet as a special case, to
random breakdowns
2.6.1 The theory of resilience to random failures
2.6.2 The plot of k versus P of scale-free networks under random attack
2.6.3 The plot of k versus P of scale-free networks and internet under
random attack
2.6.4 The plot of k versus Pc of scale-free networks under random attack
2.7 Breakdown of scale-free networks and internet as a special case, under
intentional attack
2.7.1 The plot of k versus P of scale-free networks under intentional attack
2.7.2 The plot of k versus Pc of scale-free networks under intentional
attack
2.7.3 The plot of k versus Pc of scale-free networks under random attack
and intentional attack

III REFERENCES

2
 Random Walks

In mathematics, computer science, and physics, a random walk is a formalization of

the intuitive idea of taking successive steps, each in a random direction.

A one-dimensional random walk can also be looked at as a Markov chain whose space
is given by the integers i = 0, �1, �2,L , for some probability p , 0 < p < 1 , Pi ,i +1 = p ,
Pi.i -1 = 1 - p . We can call it a random walk because we may think of it as being a
model for an individual walking on a straight line who at each point of time either
takes one step to the right with probability p or one step to the left with probability
1- p .

A random walk is a discrete domain (e.g. time) stochastic process. It can be viewed as
a domain summation of a Bernoulli process. The simplest random walk is a path
constructed according to the following rules:

 There is a starting point.

 The distance from one point in the path to the next is a constant.
 The direction from one point in the path to the next is chosen at random, and
no direction is more probable than the other.

Suppose we draw a line some distance from the origin of the walk. How many times
will the random walk cross the line? The following, perhaps surprising, theorem is the
answer: for any random walk in one dimension, every point in the domain will be
crossed an infinite number of times almost surely. This problem has many names: the
level-crossing problem, the recurrence problem or the gambler’s ruin problem. The
source of the last name is as follows: if you are a gambler with a finite amount of
money playing a fair against a bank with an infinite amount of money, you will surely
lose. The amount of money you have will perform a random walk but it will, almost
surely, reach at some time 0, and the game would be over.

Higher dimensions

Imagine now a drunkard walking around in the city. The city is infinite and arranged
in a square grid, and at every corner he chooses one of the four possible routes
(including the one he has come from) with equal probability. Formally, this is a
random walk on the set of all points in the plane with integer coordinates. Will the
drunkard ever get back to his home from the bar? It turns out that he will (almost
surely). This is the high dimensional equivalent of the level crossing problem

3
discussed above. However, the similarity stops here. In three dimensions and above,
this no longer holds. In other words, a drunk bird might forever wander around, never
finding its nest. The formal terms to describe this phenomenon is that random walks
in dimensions 1 and 2 is recurrent while in dimension 3 and above it is transient. This
was proved by Pólya in 1921.

1.1 The plot of the mean-squared displacement ( < R 2 > ) versus Time

One of the most basic quantities in the random walk theory is the mean-squared
displacement < R 2 (n) > of a particle diffusing in a given space, which is a measure of
the distance R covered by a random walker after performing n steps. In most cases,
this quantity is described by an expression of the form < R 2 (n) >: n a . The value of the
parameter a classifies the type of diffusion into normal linear diffusion (a = 1) ,
subdiffusion (a < 1) , or superlinear diffusion (a > 1) .

The description of the algorithm for constructing the plot of < R 2 > versus Time:

 We perform a random walk in a one dimensional system.

 At each step we generate a random number x between 0 and 1; if x �1/ 2 then
the walker will move to the right, if 0 < x < 1/ 2 the walker will move to the
left. So, in general if we want to perform a random walk in n-dimensional
system, the walker will move in one of the possible 2n directions, to each one
of them with probability 1/ 2n .
 After a fixed number of steps we see where the walker will be and measure R.
 Note: If a walk is composed by N steps then R at the end of a walk, will take
an integer value from [ - N , N ] . If the walk is performed in a two dimensional
lattice, we save in two arrays the displacement respectively for x-axe and y-
axe, then R 2 = x 2 + y 2 , here x and y are the displacement in x and y axes,
respectively.
 We perform a large number of walks (runs) and find the average of the result
taken from them.

4
 The plot of <R^2> vs Time, for a one dimentional system
1000

Linear Regressin: y = A + B*x

800
A = 1.78747
B = 0.99288

600
<R^2>

400

200
B
Linear Fit of Data1_B

0
0 200 400 600 800 1000
Time

1.2 The plot of the mean value of the displacement ( < R > ) versus Time

0.0

-0.5
<R>

-1.0

-1.5

0 100 200 300 400 500 600 700 800 900 1000
Time

5
1.3 The plot of the distribution of R for 500 steps and 1000 steps

Distribution of R
4000
for 1000 steps B
for 500 steps C
3500

3000

2500

2000
P(R)

1500

1000

500

-500
-150 -100 -50 0 50 100 150
R

1.4 Trapping of Random Walks in Two Dimensional Lattice

A random walk, sometimes called a “drunkard’s walk”, is a formalization of the

intuitive idea of taking successive steps, each in a random direction.

For the random walk in two dimensions, imagine a drunkard walking around in the
city. The city is infinite and arranged in a square grid, and at every corner he chooses
one of the four possible routes with equal probability. Formally, this is a random walk
on the set of all points in the plane with integer coordinates.

An important process related to random walk theory is trapping. Trapping reactions

have been widely studied in the frame of physical chemistry, as part of the general
reaction-diffusion scheme. The general idea includes two different species A and B,
which diffuse freely in a given space and upon proximity they react according to
A + B � B . Many different variations describe a plethora of physical phenomena. We
will consider the case where B particles are immobile. The simplest mean-field
analytical treatment predicts a simple exponential decay in the density of A’s.

We will perform a random walk in a two dimensional lattice, where we have placed at
random positions a number of trap molecules with concentration c. We place one
particle at a random position on the lattice and let it perform a random walk. The walk
will stop when the particle falls in a trap. The time needed for his to happen is the

6
trapping time. When the particle reaches the borders of the lattice it does not escape,
but it is placed in the opposite site of the lattice. We will perform a large number of
walks (runs) and we will save the trapping times and make the distribution of these
times.

1.5 The distribution of the trapping times

To find the distribution of the trapping times, we have performed a numerical

simulation for 100000 runs and we have saved the trapping times for these runs. We
find the frequency of these times, by counting the number of the trapping _ times and
normalize the results by dividing the array of frequency by the number of runs ( in
our case by 100000 ).
Below is shown the graph of the distribution of the trapping time in a two dimensional
lattice LAT [700] [700] and the concentration of the traps is c=1/100.
We argued with simulation techniques that the distribution of trapping time has an
exponential decay.

A
0.08

0.07

0.06
P(Trapping-ime)

0.05

0.04

0.03

0.02

0.01

0.00

-0.01
0 20 40 60 80 100
Trapping_time

1.6 The survival probability

The main property monitored during the process introduced in the previous section is
the survival probability, which denotes the probability that a particle A survives after
performing n steps in a space which includes traps B with a concentration c.

7
In the following lines we will describe the way of calculating the survival probability
of random walks that move in the presence of randomly distributed traps. We will
calculate the survival probability from the distribution of the trapping times.

If the probability that a particle will get trapped in the time t is the numerical value of
t

the integral �
p ( x)dx , than the survival probability that the particle will survive in
0
t

time t is 1 - �
p ( x )dx , where p(x) is the distribution of the trapping time. The
0

following plot shows the survival probability of the problem described in the previous
paragraph for the one we have find the distribution of the trapping time.

C
1.0

0.8
Survival Probability

0.6

0.4

0.2

0.0

0 20 40 60 80 100
Time

In the graph above are presented in the same plot the distribution of the trapping time
and the survival probability.

8
 1.0

Trapping time distribution

0.8 Survival probability

0.6

0.4

0.2

0.0

0 20 40 60 80 100

1.7 Simulations for calculating S N

The behavior of a random walk is also characterized by the coverage of the space, as
expressed by the average number of distinct sides visited < S N > after N steps.
S N is the number of distinct sites visited by a single random walker. We will see the
behavior of S N on one, two and three dimensional lattices, using Monte Carlo
simulation techniques to model a single random walk on spatially anisotropic lattice
structures.
The analytic expression for S N in the asymptotic limit of N � �, where N is
the number of steps, has been given by Montroll and Weiss for all three
dimensionalities. In 1-D, S N follows a t 1/ 2 power law, in 3-D it is linear with t, and in
2-D it is “almost” linear, with an additional logarithmic term:

1/ 2
�8 N �
SN : � � N � � 1-D (1)
�p �

pN
SN : N �� 2-D (2)
log N

SN : N N �� 3-D (3)

9
These are the asymptotic equations. Correction terms, which add accuracy to the early
time behavior, have also been derived. We use the analytical expressions 2 and 3, with
their associated correction terms.

1/ 2
�8 N � � 1 3 �
1-D: SN = � � � 1+ - +L � (4)
p
��� 4 N 64 N 2

AN � -d b
j
� �1 � �
2-D: SN = �
ln( BN ) j =0 ln BN b =2
1+ O � �
�
� �N �
�
�
(5)

3-D: S N = 0.65946267 N + 0.573921N 1/ 2 +

0.449530 + 0.40732N -1/ 2 +L (6)

where A and B in the equation 5 are constants.

The number of distinct sites visited, S N , as a function of time, for a single
random walker on one, two and three dimensional lattices is plotted in the following
figures.

The analytical solution, which in the plot corresponds with the continuous line is
found using equations (4), (5) and (6).

1.8 The plot of S N versus Time for 1-D

The description of the algorithm for finding S N :

 We perform a random walk in a n-dimensional lattice.

 Set the values of the n-dimensional lattice to 0.
 Find randomly one position in the lattice, where the walk will start, and set the
value of this position in the lattice with 1.
 Generate a random number, which will define the direction in which the
particle will move.
 The particle will move in the direction found in the previous step, if the
position where it has to go is 0, otherwise if its value is 1 we find randomly
another direction, till the end of the walk.
 Each time we find a 0 value the particle moves to this new position, which is
set to 0 and S N increases by 1.
 For finding the distribution of S N we save the frequency of different sites
visited from a particle after many walks (runs).

10
 exercice results
55 For a one dimentional lattice analitic solution
50

45

40

35

30
SN

25

20

15

10

0 200 400 600 800 1000

TIME

1.9 The plot of S N versus Time for 2-D

Simulation results
400 For a two dimentional lattice Analitic solution

350

300

250

200
Sn

150

100

50

-50
0 200 400 600 800 1000
Time

11
1.10 The plot of S N versus Time for 3-D

exercise results
For the 3 - D lattice analytic soluion
700

600

500

400
SN

300

200

100

-100
0 200 400 600 800 1000
Time

1.11 The plot of the distribution of S N for 1-D

350 Distribution of Sn in a one dimentional lattice

LAT[490000], NRUN=10000, NTIME=1000
300 Mean Value=49.7821

250

200
Frequency

150

100

50

20 40 60 80 100 120 140

Sn

12
1.12 The plot of the distribution of S N for 2-

Distribution of Sn in a two dimenional lattice

1000 LAT[700][700], NRUN=100000, NTIME=1000
Mean Value=361.91969

800

600
Frequency

400

200

100 200 300 400 500 600

Sn

1.13 The plot of the distribution of S N for 2-D

Distribution of Sn for a three dimentional lattice

200
LAT[50][50][50], NRUN=10000, NTIME=1000
Mean Value=716.6919

150
Frequency

100

50

550 600 650 700 750 800

Sn

1.14 The plot of the relation of < S N > versus Time for 1-D

13
 B
For a one dimentional lattice
50 LAT[100000], NRUN=100000

45

40

35
<Sn>

30

25

20

15

0 200 400 600 800 1000

Time

1.15 The plot of the relation of < S N > versus Time for 2-D

D
400
For 2-D lattice
350

300

250
<Sn>

200

150

100

50

0
0 200 400 600 800 1000
Time

14
1.16 The plot of the relation of < S N > versus Time for 3-D

B
3.6 For a three dimentional lattice
LAT[50][50][50],NRUN=10000
3.4

3.2

3.0

2.8
<Sn>

2.6

2.4

2.2

2.0

1.8
2 4 6 8 10 12
Time

15
 II NETWORKS

2.1 What are Networks?

2.1.1 Basic notions

In mathematical terms a network is represented by a graph. A graph is a pair of sets

G= {P, E}, where P is the set of N nodes (or vertices) P1 , P2 ,L , PN and E is a set of
edges (or links) that connect two elements of P.

Networks with undirected edges are called undirected networks, networks with
directed edges are directed networks.
The case of my study will be the undirected networks.
The total number of connection of a vertex is called its degree k. In general
terms, random networks are networks with a disordered arrangement of edges. Note
that, usually, in graph theory the meaning of a random graph is much more narrow. A
random network means an esemble of nets. In principle, random networks may
contain vertices of fixed degree. Usually, however, the degree of vertices are
statistically distributed.

2.1.2 Adjacency matrix

The adjacency matrix of a network provides its complete image. It indicates which of
the vertices are connected (adjacent). This is a square N �N matrix, where N is the
total number of vertices in the network. Its element aij is equal to 1 if there is an edge
that connects the vertex i with vertex j. The adjacency matrix of an undirected
network is symmetrical, and diagonal elements are equal to zero: aii = 0 .

2.1.3 The adjacency matrix for a random network

16
This is the adjacent matrix of a random network with 20 nodes, and two nodes are
linked with probability p = 1/ 6 (in this case).

00001000001110000000 The description of the algorithm:

00010000001000000000 (1) Consider the network as a two
00000000000100000011 dimensional lattice N �N ,
01000000001000000000 where N is the number of nodes the
10000000010000000000 network contains.
00000001000000000010 (2) We move randomly in the lattice and
00000000010000001000 pick up with probability p one
00000100100000100000 position in the lattice, and set its
00000001001011000011 value aij = 1 . This means that node i
00001010000100010000 is linked with node j.
11010000100011010000 (3) For the distribution of a random
10100000010000000001 network, we find the frequency
10000000101000110100 of the connections nodes have.
00000000101000010010
00000001000010000001
00000000011011001000
00000010000000010010
00000000000010000000
00100100100001001001
00100000100100100010

2.1.4 Degree distribution

Degree distribution is the simplest statistical characteristics of a network. This is the

distribution of an “one-vertex” quantity. So, the degree distribution characterizes only
the local properties of a network but, often, even this poor information about the
network structure is sufficient to determine its basic properties. Most of the empirical
data are degree distributions for various networks and are easy to measure.
When the vertices of a graph are statistically independent (equivalent), that is
connections are random, the degree distribution completely determines the statistical
properties of a network.
The following examples demonstrate typical degree distributions for networks.

(1) The Poisson distribution

e-k k k
P (k ) =
k!

here, k is the average degree, k = �k = 0 kP (k ) . A classical random graph

�

asymptotically has just this degree distribution, if its number of vertices approaches
infinity under the constraint that the mean degree is fixed.

17
 k
(2) Exponential distribution P (k ) = a e - k . For instance, this is the distribution of
the trapping times.

(3) Power-law distribution P (k ) = a k -g , k �0 . Here, g is the exponent of the

power-law distribution.
The power law distribution contrasts with Poisson and exponential
distributions. It has no natural scale, and hence may be called scale-free. Networks
with such distribution are called scale-free.

List of main characteristics

Here we list the main structural characteristics of networks. Some of them have been
introduced, the others will be necessary in what follows.

 The total number of vertices: N. Sometimes we call it the ‘size’ of the

network.
 The total number of edges: L.
 Degree of a vertex: k
 The total degree of a network: K. This is the sum of degrees of vertices.
 The mean degree of a network: < k > . This is the same as the average number
of the nearest neighbors of a vertex in a network.
 The mean of square degree of a network: < k 2 > .
 Degree distribution: P(k).
 g exponent of a degree distribution: P (k ) : k -g .

2.2 Random networks, Erdős- Rényi model

A particularly rich source of ideas has been the study of random graphs, graphs in
which the edges are distributed randomly. The theory of random graphs was
introduced by Paul Erdős and Alféd Rényi.

According to the Erdős- Rényi model, we start with N nodes (a fixed number) and
connect every pair of nodes randomly selected with probability p, creating a graph
with approximately pN ( N - 1) / 2 edges distributed randomly. The distribution below
is obtained for a network with 100000 nodes where two nodes are connected with
probability p = 1/ 6 .

2.2.1 The plot of the distribution of the random network

As we can see from the plot, the random network has a Poisson distribution. This
result suits with what we said in paragraph 2.1.4.

18
 Distribution of P(k)
400
Mean Value = 16666.69062
350 N_NODE = 100000 Count

300

250

200
Frequency

150

100

50

-50
16200 16400 16600 16800 17000 17200 17400
K

2.3 Scale – Free Network

The term “scale-free” refers to any functional form f(x) that remains unchanged to
within a multiplicative factor under a rescaling of the independent variable x. In effect
this means power-law forms, since these are the only solutions to f(ax)=bf(x), and
hence “power-law” and “scale-free” are, for our purposes, synonymous.
Here we find the formula for calculating the degree of a vertex k, for a power-law
distribution:

Pk (k ) = Ck -g where k �[kmin , kmax ] ,

�dx,0 < x < 1

Px ( x) = �
0, otherwise
�
+�

As Px ( x) is density function we can write �

P ( x)dx = 1 .
-�
x

dx
Px ( x) = , from this follows dx = Pk (k )dk
dk
k
x=
kmin
�P (k ')dk '
k

-g
Replacing Pk (k ) = Ck , we obtain
Pk (k ') = C ( k ') -g
k
-g
x= �(k ')
kmin
dk '

19
 (k ') -g +1 k C
x=C |kmin = [k 1-g - kmin
1-g
] , for g �1 .
-g + 1 1- g
C
So, x= [ k 1-g - kmin1-g
] for g �1 .
1- g
1- g
= k 1-g - kmin 1-g

C
(1 - g )x 1-g
k = 1-g + kmin (*)
C

For x = 1 we receive a formula for kmax :

1- g 1-g
kmax = 1-g + k min
C
Taking the power of ( 1 - g ) we receive:

1- g 1-g 1-g 1- g 1-g 1-g

+ kmin = kmax � = kmax - kmin
C C

Replacing this last equation in the equation (*) we receive:

1-g 1-g 1-g

k = 1-g (kmax - kmin ) x + kmin (**)

This equation will be used widely for finding the number of connections for each
node of a scale-free network.

2.3.1 Characteristics of scale-free networks

In the following table are shown the values of the parameters < K > , < K 2 > , where
K stands for the number of connections. To provide a heuristic characterization of the
level of heterogeneity of networks we define the parameter k =< k 2 > / < k > .
Indeed, fluctuations are denoted by the normalized variance that can be expressed as
k / < k > -1 , and scale-free networks are characterized by k � �, whereas
homogeneous networks have k �< k > . For this reason, we will generally refer all
networks with heterogeneity parameter k ? < k > as scale-free networks. We shall see
in the following paragraphs that k is a key parameter for all properties and physical
processes in networks which are affected by the degree fluctuations.

Gama = 2.0 Gama = 2.5 Gama = 3.0

<K> 8.4598 2.5258 1.6494
< K2 > 4165.82 54.9312 7.9352
k 492.426 21.748 4.81096

20
2.3.2 Plot of the distribution of scale-free network for g = 2.0

We receive the following distribution plot (with logarithmic axes). Where K is the
number of links and Frequency is the number of times each link is encountered.

10000

1000
log(Frequency)

100

10

1 10 100 1000 10000

log(K)

2.3.3 Plot of the distribution of scale-free network for g = 2.5

10000

1000
lo(Frequency)

100

10

1 10 100 1000
log(K)

2.3.4 Plot of the distribution of scale-free network for g = 3.0

21
 Count

10000

1000
log(Frequency)

100

10

1 10 100
log(K)

From this plots we see that the network for g = 2 is very dense, the network for
g = 2.5 is less dense and the network for g = 3 is less dense than the previous one.
Below we represent the distribution of scale-free networks for g = 2.0, 2.5,3.0 in the
same plot.

Distribution for Gama=2.0

Distribution for Gama=2.5
Distribution for Gama=3.0
log(Frequency)

1000

100

10
1 10 100
log(K)

22
2.3.5 The algorithm that constructs the scale – free network for
g = 2, 2.5,3 with N fixed nodes

The algorithm will pass in the following steps.

1. Find the number of connections for each node using the equation
1-g
k = 1-g (kmax 1-g
- kmin 1-g
) x + kmin , in our case kmin = 1 .
N -1
2. Find the total number of legs: T _ legs = �legs (i ) , where legs(i) are the
i=0
connections node i has.
3. If the total number of the legs is odd, choose randomly one node and add a leg
to it, so T _ legs will increase by one.
4. Find the number of links: N _ links = T _ legs / 2 .
5. Randomly choose a link and than two nodes. Using the rejection method we
give more priority to choose two nodes with biggest ‘weights’, or nodes that
have more connections. If we choose the nodes randomly then we will be
faced with the problem that nodes with many connections will have most of
them unconnected.
6. Construct the network.

After the network construction we ignore the links which cannot connect.

Below we present the scale-free network for g = 2.5 in a linked list format.

N Link Nodes linked

0 1 5
1 1 49
2 2 44 29
3 1 43
4 1 24
5 1 0
6 1 40
7 15 29 9 42 47 40 15 36 26 37 11
45 35 21 33 28
8 1 40
9 6 29 44 42 7 21 19
10 1 21
11 4 7 32 29 44
12 1 26
13 1 18
14 2 26 47
15 2 7 42
16 1 49
17 3 29 21 39
18 1 13
19 1 9
20 2 29 21
21 9 48 29 44 47 9 7 17 10 20

23
22 2 47 44
23 3 29 44 46
24 2 44 4
25 1 40
26 4 29 14 7 12
27 1 42
28 1 7
29 12 9 7 45 26 21 2 41 23 35 17
11 20
30 1 36
31 1 47
32 1 11
33 2 7 38
34 2 47 40
35 2 29 7
36 2 7 30
37 1 7
38 1 33
39 1 17
40 5 8 34 7 6 25
41 1 29
42 4 9 7 27 15
43 2 3 44
44 9 9 2 21 47 22 24 43 23 11
45 2 29 7
46 1 23
47 7 34 7 31 44 21 22 14
48 1 21
49 2 16 1

2.3.6 The Greatest _ Cluster for Scale-Free Network

Breadth – First search is an algorithm for searching a graph. Given a graph G= (V, E)
and a distinguished source vertex s, breadth – first search systematically explores the
edges of G to “discover” every vertex that is reachable from s. It computes the
distance (fewer number of edges) from s to all such reachable vertices. It also
produces a “breadth – search tree” with root s that contains all such reachable
vertices. From any vertex v reachable from s, the path in the breadth – first tree from
s to v corresponds to a “shortest path” from s to v in G, that is, a path containing the
fewest number of edges.

Breadth – first search is so named because it expands the frontier between discovered
and undiscovered vertices uniformly across the breadth to the frontier. That is, the
algorithm discovers all vertices a distance k from s before discovering any vertices at
distance k+1.

To keep track of progress, breadth – first search colors each vertex white, gray, or
black. All vertices start out white and may become gray and then black. A vertex is
discovered the first time it is encountered during the search, at which time it becomes

24
nonwhite. Gray and black vertices, therefore, have been discovered, but breadth – first
search distinguishes between them to ensure that the search proceeds in a breadth –
first manner. If (u , v ) �E and vertex u is black, then vertex v is either gray or black;
that is, all vertices adjacent to black vertices have been discovered. Gray vertices may
have some adjacent white vertices; they represent the frontier between discovered and
undiscovered vertices.

Breadth- first search constructs a breadth – first tree, initially containing only its root,
which is the source vertex s. Whenever a white vertex v is discovered in the course of
scanning the adjacency list of an already discovered vertex u, the vertex v and the
edge (u, v) are added to the tree. We say that u is the predecessor or parent of v in the
breadth – first tree. Since a vertex is discovered at most once, it has at most one
parent. Ancestor and descendant relationships in the breadth – first tree are defined
relative to the root s as usual: if u is on a path in the tree from the root s to vertex v,
then u is an ancestor of v and v is a descendant of u.

The breadth–first–search procedure BFS below assumes that the input graph G=(V,E)
is represented using adjacency lists. It maintains several additional data structures
with each vertex in the graph. The color of search vertex u �V is stored in the
variable color[u ] , and the predecessor of u is stored in the variable P[u ] . If u has no
predecessor (for example, if u=s or u has not been discovered), then P[u ] = NIL . The
distance from the source s to vertex u computed by the algorithm is stored in d [u ] .
The algorithm also uses a first-in, first-out queue Q to manage the set of gray vertices.

The pseudo-code of the algorithm BFS (G,s):

1. for each vertex u �V [G ] - {s}

2. do color[u] ← WHITE
3. d[u] ← ∞
4. P[u ] ← NIL
5. color[s] ← GRAY
6. d[s] ← 0
7. P[ s ] ← NIL
8. Q ← {s}
9. while Q �0
10. do u ← head[Q]
11. for each v �Adj[u ]
12. do if color[v] ← WHITE
13. then color[v] ← GRAY
14. d[v] ← d[u] + 1
15. P[v ] ← u
16. ENQUEUE (Q,v)
17. DEQUEUE (Q)
18. color[u] ← BLACK

25
The procedure BFS works as follows. Lines 1-4 paint every vertex white, set d[u] to
be infinity for every vertex u, and set the parent of every vertex to be NIL. Line 5
paint the source vertex s gray, since it is considered to be discovered when the
procedure begins. Line 6 initializes d[s] to 0, and line 7 sets the predecessor of the
source to be NIL. Line 8 initializes Q to the queue containing just the vertex s;
thereafter, Q always contains the set of gray vertices.

The main loop of the program is contained in lines 9-18. The loop iterates as long as
there remain gray vertices, which are discovered vertices that have not yet had their
adjacency lists fully examined. Line 10 determines the gray vertex u at the head of the
queue Q. The for loop of lines 11-16 considers each vertex v in the adjacency list of u.
If v is white, then it has not yet been discovered, and the algorithm discovers it by
executing lines 13-16. It is first grayed, and its distance d[v] is set to d[u] + 1. Then,
u is recorded as its parent. Finally, it is placed at the tail of the queue Q. When all the
vertices on u’s adjacency list have been examined, u is removed from Q and
blackened in lines 17-18.

The following figure illustrates the progress of BFS on a sample graph.

Figure: The operation of BFS on an undirected graph. Tree edges are shown shaded as
they are produced by BFS. Within each vertex u is shown d[u]. The queue Q is shown

26
at the beginning of each iteration of the while loop of lines 9-18. Vertex distances are
shown next to vertices in the queue.

After finding the distribution of scale-free network using the formula

1-g 1-g 1-g
k = 1-g (k max -k min )x + k min and constructing the network using the algorithm
introduced in paragraph 2.3.5, we find he greatest cluster for K min = 1 and then we
find also the greatest cluster for K min = 2 . As is seen from the results, for K min = 2 the
greatest cluster is the whole network.

Using the Breadth-First-Search we have these results:

Gama N _ Node Greatest _ Cluster Greatest _ Cluster

Kmin = 1 Kmin = 2
2.0 10000 9977 10000
2.5 10000 7844 10000
3.0 10000 4750 10000

2.4 The Barabasi – Albert model

The Poisson distribution is a rapidly decreasing function, and we desire fat-tailed

ones. The next step was made by Barabasi and Albert (1999) who combined the
growth and preferential linking (or rich-get-richer). This model is a citation graph
where new vertices become attacked to old ones with a probability to their degrees.

Networks growing under the mechanism of preferential linking are growing networks
where new edges become attacked to vertices preferentially. By the definition, this
means that more connected vertices have a better chance to get new connections.
The linear type of preferential attachment produces fat-tailed degree distributions. On
the other hand, if any preference of linking is absent, and new connections become
distributed at random, the degree distribution decreases rapidly. These networks are
strongly correlated.

We start with a fixed number N of vertices that are then randomly connected or
rewired, without modifying N. In contrast, most realworld networks describe open
systems that grow by the continuous addition of new nodes. Starting from a small
nucleus of nodes, the number of nodes increases throughout the lifetime of the
network by subsequent addition of new nodes.
Network models discussed so far assume that the probability that two nodes are
connected is independent of the nodes’ degree, i.e., new nodes are placed randomly.
Most real networks, however, exhibit preferential attachment, such that the likelihood
of connecting to a node depends on the nodes’ degree.

27
The algorithm of the Barabasi – Albert model is the following:

1. Growth: Starting with a small number ( m0 ) of nodes, at every time step, we

add a new node with m ( �m0 ) edges that link the new node to m different
nodes already present in the system.
2. Preferential attachment: When choosing the nodes to which the new node
connects, we assume that the probability P that a new node will be connected
to node i depends on the degree ki of node i, such that

ki
P ( ki ) =
�k j
j

After k time steps this procedure results in a network with nodes and tm edges.
Numerical simulations indicated that this network evolves into a scale – invariant
state with the probability that a node has k edges following a power law with an
exponent g BA �3 .

In the following plot is represented the distribution of the BA model. The distribution
is received for a network that contains 10000 nodes, and is taken the average result of
1000 runs, starting from 2 nodes linked together and adding each time a new node, for
the first plot, and the second one is received for a network that contains 100000
nodes, for the same number of runs.

The description of the algorithm step by step:

 We start with 2 nodes, which are connected with each other.

 At each step we add a new node, which is connected with m ( m �m0 ) nodes
chosen randomly, from m0 nodes that the network has in the current stage.
 We increase the number of legs in the network with 2, after two nodes are
connected.
 After reaching the desired size of the network, we find the number of legs
each node has.
 Construct the array of frequency, which contains the number of times a certain
connection is encountered.
 We perform a large number of runs and take the average result.

The following plots are constructed using this algorithm.

2.4.1 The plot of the distribution of B-A model

28
 B
10000 Linear Fit of Data1_B

1000
Slope = - 2.91344
Node = 10000
100
Run = 1000

10
log(P(K))

0.1

0.01

1E-3

1E-4
1 10 100
log(K)

B
100000 Linear Fit of Data1_B

10000 Slope = -2.9778

Node = 100000
1000 Run = 1000

100

10
log(P(K))

0.1

0.01

1E-3

1E-4
1 10 100 1000
log(K)

29
2.5 Real Networks

Only recently have we realized that we reside in a world of networks. The Internet
and World Wide Web (WWW) are changing our lives. Our physical existence is based
on various biological networks. The extent of the development of communication
networks is a good indicator of the level of development in a country. ‘Networks’ turn
to be a central notion in our time, and the explosion of interest in networks is already
a social and cultural phenomenon. The Internet will be the case of this study.

2.5.1 Internet robustness

The Internet is composed by thousands of different elements – both at the hardware

and software level – which are naturally susceptible to errors, malfunctioning, or other
kind or failures, such as power outages, hardware problems, or software errors
(Paxson, 1997; Laboviz, Ahuja, and Jahanian, 1999). Needless to say, the Internet is
also subject to malicious attacks. The most common of those are the denial – of –
service attacks, that encompass a broad set of attacks aimed at a diversity of the
Internet services, such as the consumption of limited resources or the physical
destruction of network components ( C.E.R. team, 2001 ). Given so many open
chances for the errors and failures, it might sometimes be surprising that the Internet
functions at all.
The design of a computer network resilient to local failures ( either random
malfunctions or intentional attacks ) was indeed one of the main motivations for the
original study of distributed networks by Paul Baran ( 1964 ). Considering the worst
possible scenario of an enemy attack directed towards the nodes of a nationwide
computer network, Baran analyzed the “survivability” ( defined as the average
fraction of surviving nodes capable of communication with any other surviving node )
of the several network designs available at that time. His conclusion was that the
optimal network, from the survivability point of view, was a mesh - like graph with a
sufficient amount of redundancy in the paths between vertices. Even in the case of a
severe enemy strike, depleting a large number of components, such network topology,
would ensure the connectivity among the surviving computers, diverting
communications along the ensemble of alternative paths. In addition, this sort of
resilience against attacks would be economically beneficial, since it would allow the
use of low – cost, unreliable connections between components, which would not
substantially hinder communications in case of a random failure.
Baran’s proposal for a designed distribution network was finally not taken into
account in the development of the primitive ARPANET, nor in the subsequent
evolution of the Internet. This network has undergone instead a self – organized
growth, following no pre – established plan, and its present scale – free topological
structure is far from that envisioned by Baran. In this respect, the possible fragility of

30
the Internet to random failures and intentional attacks is a major issue, with several
practical implications.
The study of the resilience of the Internet to failures is no an easy task. After
any router or connection fails, the Internet responds very quickly by updating the
routing tables of the routers in the neighborhood of the failure point. Therefore, the
error tolerance of this network is a dynamical process, which should take into account
the time response of the routers to different damage configurations. Though, a first
approach to the analysis of the Internet’s robustness can be achieved at the
topological level by studying the behavior of the AS and IR level maps under the
removal of vertices or edges. These studies have shown that the Internet presents two
faces in front of the component failures: it is extremely robust to the loss of a large
number of randomly selected vertices, but extremely fragile in response of a targeted
attack.
The Internet can be viewed as a special case of a random, scale-free network,
where its distribution follows a power-law: P (k ) : k -g ( g �2.1 , for the Internet).
The following plot shows the distribution of the Internet for results taken from
www.netdimes.org . From this web site I have taken the edges of the network and then
I construct the network (in a linked list format) and from the adjacency list of the
network I find the distribution of the Internet, which is shown in the following graph.
The data are taken from a registration of NETDIMES done in October 2004.

2.5.2 The plot of the distribution of the Internet

1000
Slope = -2.10297

100
log(frequency)

10

10
log(k)

31
 In this work I provide a review of results on the topological resilience of the
Internet damage. I will present numerical experiments which show that the Internet
can withstand a considerable amount of random damage and still maintain overall
connectivity in the surviving network. In particular the Internet’s tolerance to massive
random damage is much higher than for meshes or random homogeneous networks,
suggesting that the cause for this robustness resides in its power law degree
distribution. The very nature of Internet degree distribution, on the other hand, implies
the presence of heavily connected hubs. A targeted attack, aimed at knocking down
those hubs, has dramatic consequences for Internet connectivity. In this case we shall
see that the deletion of a very small fraction of hubs is enough to break the network
down into small, isolated components, hugely reducing its communication
capabilities.

2.4 Resilience of scale-free networks and internet as a special case,

under random breakdowns

A first empirical assessment of the robustness of the Internet in front of random

failures can be obtained by studying the topological response of Internet maps to the
removal of edges or vertices. Focusing on the effect of vertex removal, and assuming
that all vertices are equally likely to experience a failure, this theoretical experiment
can be performed on a connected map of size N by looking at the effect achieved by
removing a fraction p of randomly selected vertices. The deletion of a vertex implies
that all the edges adjacent to it are also removed. In order to monitor the response of
the network to the damage, one can control several topological metrics related to
network connectivity ( Albert, Jeong, and Barabasi, 2000 ). A quantity presented by
R.P.Satorras and A.Vespignani in their book Evolution and Structure of the Internet, is
the size of S p of the largest component in the network after damage with respect to
the size S0 of the undamaged network. In particular the ratio S p / S0 > 0 indicates that
a macroscopic fraction of vertices is still capable of communication, i.e. a giant
component still exists. On the contrary, S p / S0 ; 0 signals that the whole network has
been fragmented into small disconnected components, each one of them no containing
an appreciable fraction of the vertices.
The parameter I have measured is k =< k 2 > / < k > , where <k> is the average
degree of the network.
A natural question to ask in this context concerns the maximum amount of
damage that the network can take, i.e. the threshold value of the removal probability,
pc , above which k < 2 and the network can be considered destroyed.
The conclusion from the analysis presented in this section is that the Internet
appears as very robust with respect to large densities of random failures. The network
is able to sustain a considerable number of disabled components and still presents an
appreciable fraction of vertices in the largest component.
The main topological feature of the Internet resides in its power – law degree
distribution, it is natural to think of this property as the one responsible for the
Internet’s high resilience. Indeed, it is possible to understand this point qualitatively

32
by recalling that the power law form of this distribution implies that the vast majority
of vertices have a very small degree, while a few hubs collect a very large number of
edges, providing the necessary connectivity to the whole network. When removing
vertices at random, chances are that the largest fraction of deleted elements will have
a very small degree. Their deletion will imply, in turn, that only a limited number of
adjacent edges are eliminated. Therefore, the overall damage exerted on the network’s
global connectivity properties will be limited, even for very large values of p. This
intuition will be confirmed in the analytical study performed in the following
paragraphs, where it is shown that scale – free networks with degree exponent g �3
have a infinite tolerance to random damage, in the sense that it is necessary to delete a
fraction p � 1 in order to induce the complete breakdown of the largest component.
This fact has led on some occasions to the erroneous statement that scale – free
networks have a topology that is designed or optimized to resist random failures.

2.6.1 The theory of resilience to random failures

Recently there has been increasing interest in the formation of random networks and
in the formation of random networks and in the connectivity of these network,
especially in the context of the Internet. When such networks are subject to random
breakdowns – a fraction p of the nodes and their connections are removed randomly –
their integrity might be compromised: when p exceeds a certain threshold, p < pc , the
network disintegrates into smaller, disconnected parts. Below that critical threshold,
there still exists a connected cluster that spans the entire system.
In this paragraph we consider random breakdowns in the Internet and introduce an
analytical approach to finding the critical point. The site connectivity of the physical
structure of the Internet, where each communication node is considered as a site, is
power law, to a good approximation. We introduce a new general criterion for the
percolation critical threshold of randomly connected networks. Using this criterion,
we show analytically that the Internet undergoes no transition under random
breakdowns of its nodes. In other words, a connected cluster of sites that spans the
Internet survives even for arbitrarily large fractions of crashed sites.
We consider networks whose nodes are connected randomly to each other, so
that the probability for any two nodes to be connected depends solely on their
respective connectivity. We argue that, for randomly connected networks with
connectivity distribution P(k), the critical breakdown threshold may be found by the
following criterion: if loops of connected nodes may be neglected, the percolation
transition takes place when a node (i), connected to a node (j) in the spanning cluster,
is also connected to at least one other node – otherwise the spanning cluster is
fragmented. This may be written as

ki | i � j = �ki P (ki | i � j ) = 2 (1)

ki

33
where the angular brackets denote an ensemble average, ki is the connectivity of node
i, and P (ki | i � j ) is the conditional probability hat node i has connectivity ki , given
that it is connected to node j. But, by Bayes rule for conditional probabilities

P(ki | i � j ) = P(ki | i � j ) / P(i � j ) = P(i � j | ki ) P(ki ) / P(i � j )

where P (ki , i � j ) is the joint probability that node i has connectivity ki and that it is
connected to node j. For randomly connected networks P (i � j ) = k /( N - 1) and
P (i � j | ki ) = ki /( N - 1) , where N is the total number of nodes in the network. It
follows that the criterion (1) is equivalent to

k2
k� =2 (2)
k

at criticality.
The strategy thus consists in finding at which damage density p, the surviving
network fulfills the percolation condition k p / k p = 2 , where k p
2 2
and k p
refers to the degree distribution moments of the damaged graph.
Following the intuitive approach proposed by Cohen, Erez, ben Avraham, and
Havlin (2000), lets consider a sparse uncorrelated generalized random graph with
2
degree distribution P0 ( k ) and first moments k 0 and k . 0

Removing a fraction p of all vertices in the graph is equivalent to removing a fraction

p of the neighbors of any surviving vertex. Therefore, in the damaged network any
surviving vertex with original degree k ' will have a degree k with probability

�k '�
�� (1 - p ) k p k '- k , k ' �k ,
�k �

which essentially corresponds to the random deletion of k '- k neighboring vertices,

preserving k of them. The degree distribution of the network defined by the surviving
vertices will then be given by

�
�k '�
Pp (k ) = �P0 (k ') � � (1 - p) k p k '- k .
k '= k k
��

From this equation we can compute the first and second moments of the degree
distribution in the damaged network, obtaining

k p
= �kPp (k ) = (1 - p ) k 0 ,
k

k2 = �k 2 Pp (k ) = (1 - p ) 2 k 2 + p(1 - p) k 0 .
p 0
k

34
These expressions can be plugged in Equation (2) that gives the condition for the
presence of a giant component in the surviving network, yielding the precise value of
the threshold pc as the one that satisfies the equation

k2 (1 - pc ) k 2
pc
= 0
+ pc = 2 .
k pc
k 0

By using the heterogeneity parameter k = k

2
/ k 0 , the explicit solution for the
0

removal threshold is obtained as

1
pc = 1 - .
k -1

That is, the critical threshold for the destruction of the graph’s connectivity differs
from unity by a term that is inversely proportional to the degree fluctuations of the
undamaged network. This readily implies that the topological robustness to damage is
related to the graph’s degree heterogeneity. In homogeneous networks, in which the
heterogeneity parameter is k ; k , the threshold is finite and will simply depend on
the average degree. In highly heterogeneous networks, in which k ? k , the
2
threshold approaches larger values, being dominated by the magnitude of k . In
particular, all scale – free graphs with diverging k2
0
have k � �, therefore
exhibiting an infinite tolerance to random failures; i.e. pc � 1 . Scale – free graphs
thus define a class of networks characterized by a distinctive resistance to high levels
of damage that is clearly welcome in many situations.

2.6.2 The plot of k versus P for scale-free networks under random attack

The following plot presents the behavior of k for scale-free networks (

g = 2.0, 2.5,3.0 ), under a progressive density of damage p. In the next plot the result
of the same damage applied to Internet network is compared with the behavior
obtained for the scale-free networks for g = 2.0, 2.5 .

The algorithm step by step:

 Find the number of connections for each node using the equation
1-g
k = 1-g (kmax 1-g
- kmin 1-g
) x + kmin , in our case kmin = 1 and kmax = N .
 Construct the network, using a linked list format.
 Choose randomly 10% of the nodes and remove their legs and also the legs of
the nodes adjacent to them; then we choose randomly 20% and so on, till the
threshold will be reached. The parameter we use to test this is k = < k > < k >
2

35
 ( k - is the number of connections for each node), and the threshold is reached
when k =2.

g =2.0
g=2.5
Random Attack
g =3.0
300

250

200
Kapa

150

100

50

0.0 0.2 0.4 0.6 0.8 1.0

P

2.6.3 The plot of k versus P for scale-free networks & Internet under
random attack

36
 5/28/2007 15:34:14

300 g = 2.0
Internet
g = 2.5
250

200
Kapa

150

100

50

0.0 0.2 0.4 0.6 0.8 1.0

P

2.6.4 The plot of k versus Pc for scale-free networks under random

attack

37
 Random Attack

1.0

0.9

0.8
Pc

0.7

0.6

0.5

0.4
1.8 2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6
Gama

The results shown in the previous section point a scale – free graphs as very robust
networks that in principle can be broken apart only by damaging all their vertices.
From the plot in paragraph 2.6.3 we conclude that even the Internet network gives a
stunning robustness to random failures.
However, real – world networks, the Internet studied in this chase, necessarily show
finite size effects due to resource or size constraints.

2.7 Breakdown of scale-free networks and internet as a special case of

them, under intentional attack

The question of stability of scale – free networks to removal of a fraction of their

sites, especially in the context of the Internet, has recently been of interest. The
Internet can be viewed as a special case of a random, scale - free network, where the
probability of a site to be connected to k other sides follows a power law: P ( k ) : k -g (
g �2.1 , for the Internet). It is now well established that if a fraction p of the sites is
removed randomly, then for g > 3 there exist a threshold, pc , such that for p > pc the
network disintegrates; networks with g �3 are more resilient and do not undergo this

38
transition, although finite networks (such as the Internet) may eventually be disrupted
when nearly all of their sites are removed.
Albert et al. (2000) have introduced a model for intentional attack, or sabotage
of random networks: the removal of sites is not random, but rather sites with the
highest connectivity are targeted first. Their numerical simulations suggest that scale
– free networks are highly sensitive to this kind of attack. It is proved both
analytically and numerically, that scale – free networks are highly sensitive to
sabotage of a small fraction of the sides, for all values of g .

With simulations I have studied the problem of intentional attack in scale – free
networks. The study focuses on the exact value of the critical fraction needed for
disruption.

The algorithm step by step:

 Find the number of connections for each node using the equation
1-g
k = 1-g (kmax 1-g
- kmin 1-g
) x + kmin , in our case kmin = 1 and kmax = N .
 Sort the array of connections using Selection Sort.
 Construct the network for this sorted array of connections.
 Choose randomly 10% of the nodes and remove their legs and also the legs of
the nodes adjacent to them; then we choose randomly 20% and so on, till the
threshold will be reached. As in the case of random attach we test
k = < k > < k > ( k - is the number of connections for each node), and the
2

threshold is reached when k =2.

The first plot shows the relation of k versus P, where P is the percentage of removed
sites, for the scale-free networks for g �2.0, 2.5,3.0 .
The second plot represents the relation of k versus Pc , for different g . Pc is the
threshold, such that for values of P above that, the network disintegrates.

2.7.1 The plot of k versus P for scale-free networks under intentional

attack

39
 Gama = 2.0
Intentional Attack Gama = 2.5
18
Gama = 3.0
16

14

12

10
Kapa

0
0.01 0.02 0.03 0.04 0.05 0.06 0.07
P

2.7.2 The plot of k versus Pc for scale-free networks under intentional

attack

0.08
Intentional Attack

0.07

0.06

0.05

0.04
Pc

0.03

0.02

0.01

0.00
2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6
Gama

40
2.7.3 The plot of k versus Pc for scale-free networks under random and
intentional attack

Intentional attack
Random attack

1.0

0.8

0.6
Pc

0.4

0.2

0.0
2.0 2.2 2.4 2.6 2.8 3.0 3.2 3.4 3.6
Gama

The scale-free nature of the Internet protects it from random failures, since the hubs,
that hold the network together with their many links, are difficult to hit in a random
selection of vertices. Since hubs are the key elements ensuring the connectivity of the
network, however, it is easy to imagine that a targeted attack, aimed at the destruction
of the most connected vertices, should have a very disruptive effect. In practice this is
made by removing the vertices following a decreasing degree ordered list. The first
vertex to be removed is therefore the one with the highest degree. Then the second
highest degree vertex is removed, and so on until the total number of removed
vertices represents a fraction p of the total number of vertices forming the network. As
for the random removal, the behavior of k can be studied for increasing damages p.

The scale-free nature of the Internet graph makes the long tail of large degree vertices,
the hubs, extremely important for keeping the graph connected. Their removal leads
right away to networks collapse.

41
 10 Gama = 2.0
Intentional Attack
Internet
9
Gama = 2.5
8

6
Kapa

0
0.00 0.05 0.10 0.15 0.20 0.25 0.30
P

The targeted attack scenario is studied in the case of the Internet network. The
heterogeneous parameter k reaches the value 2 for Pc �0.11 .
This result does not suits with my expectations, because as the Internet shows a scale-
free behavior, with g �2.1 , I expect that 0.048 < Pc < 0.072 , where Pc = 0.048 for
g = 2.0 , and Pc = 0.072 for g = 2.5 .

The reasons I think why this happens are the following:

 These are the results taken just from one registration of the NETDIMES.
 If you see carefully the plot of the distribution of the Internet, it’s a lot of noise
in the tail of it.

Further work

 To use more registration of the NETDIMES, to see their behavior and take the
average of them.
 To try the same attacks even with the registrations of Oregon Route-Views
project and compare the results.

42
 REFERENCES

[1] S.N.Dorogovtsev, J.F.F. Mendes (2004), Evolution of Networks

[2] Romualdo Pastor-Satorras, Alessandro Vespignani (2004), Evolution and
Structure of the Internet
[3] Reka Albert, Albert-Lazlo Barabasi (2002), Statistical mechanics of complex
networks, Reviews of Modern Physics, 74.
[4] S.N.Dorogavtsev, J.F.F. Mendes (2002), Evolution of Networks, Advances in
Physics 51, 4,1079-1187.
[5] M.E.J. Newman (2003), The Structure of Complex Networks, SIAM Review 45,
2, 167-256.
[6] Lazaros K. Gallos (2004), Random walk and trapping processes on scale-free
networks, Physical Review E 70, 046116
[7] Reuven Cohen, Kerez Erez, Daniel ben-Avraham, Shlomo Havlin (2000),
Resilience of the Internet to Random Breakdowns, Physical Review Letters 85, 21.
[8] Reuven Cohen, Kerez Erez, Daniel ben-Avraham, Shlomo Havlin (2001),
Breakdowns of the Internet under Intentional Attack, Physical Review Letters, 86, 16.
[9] Lazaros K. Gallos, Panos Argyrakis, Armin Bunde, Reuven Cohen, Shlomo
Havlin (2004), Tolerance of scale-free networks: from friendly to intentional attack
strategies, Physica A 344, 504-509.
[10] Anna L. Lin, Raoul Kopelman, Panos Argyrakis (1997), Diffusion-Controlled
Elementary Reactions in Tubular Confinement: Extreme Nonclassicality, Segregation,
and anomalous Scaling Laws for Dimensional Crossovers, Physical Chemistri A 101,
5, 802-808.
[11] Brian H. Kaye (1989), A Random Walk Through Fractal Dimensions
[12] Joseph Rudnick, George Gaspari (2004), Elements of the Random Walk
[13] Thomas H. Comen, Charles E. Leiserson, Ronald L. Rivers, Introductions to
Algorithms
[14] Thomas Fuhrmann, On the Topology of Overlay-Networks
[15] Martin Rosvall (2003), Complex Networks and Dynamics of an Information
Network Model.
[16] X.Yao (2006), Random Graphs and Complex Networks, WIAS.
[17] Ulrike Muckstein (2002), Statistical Mechanics of Complex Networks, Institute
for Theoretical Chemistry and Sructural Biology, University Vienna.
[18] W.W.Powell, D.R.White, K.W.Koput, J.O.Smith (2002), Networks Dynamics and
Field Evolution: The Growth of Interorganizational Collaboration in the Life Sciences
[19] Alain Barrat, Alexander Hall, Maus Mihal’ak (2006), Network Discovery on
Snapshots of the Internet Graph, Project Number 001907 DELIS, Dynamically
Evolving, Large-scale Information Systems.
[20] Michael Molloy, Bruce Reed, Combinatorics, Probability and Computing (1998)
7, 295-305.
[21] Elliott W. Montroll, George H. Weiss, Random Walks on Lattices. II,
Mathematical Physics 6, 2, (1965).

43
[22] Frank S. Henyey, V. Seshadri, On the number of distinct sites visited in 2D
lattices, Mathematical Physics 6, 2, (1965).
[23] Harvey Gould, Jan Tobochnik (1996), An Introduction to Computer Simulation
Methods.

44