Unformatted text preview: d di ( t )
m
=,
dt
t
since each new node at each time spreads its m new links randomly over the
t existing nodes at time t .
This diﬀerential equation has a solution
di (t ) = m + m log �t �
i . From this solution, we derive an approximation to the degree distribution.
12 Networks: Lecture 6 “Expected” Degree Distribution
We ﬁrst note that the expected degrees of nodes are increasing over time.
If we ask how many nodes have degree ≤ 100 and we know that a
node born at time τ has degree = 100 at time t , then we are
equivalently asking how many nodes were born on or after time τ .
This implies that at time t , the fraction of nodes having degree less
than or equal to 100 would be t −τ .
t
For any d and any time t , let i (d ) be a node such that di (d ) (t ) = d . The
i (d )
resulting cumulative distribution function then is Ft (d ) = 1 − t . Applying this technique to the uniform attachment model, we solve for i (d )
such that
�t�
d −m
i (d )
d = m + m log
,
which yields
= e− m ,
i (d )
t
and therefore the distribution function Ft (d ) = 1 − e − d −m
m . This is an exponential distribution with support from m to inﬁnity and a
mean degree of 2m.
13 Networks: Lecture 6 Preferential Attachment Model
Nodes are born over time and indexed by their date of birth.
Assume that the system starts with a group of m nodes all connected to one
another.
Each node upon birth forms m (undirected) edges with preexisting nodes.
Instead of selecting m nodes uniformly at random, it attaches to nodes with
probabilities proportional to their degrees.
For example, if an existing node has 3 times as many links as some
other existing node, then it is 3 times as likely to be linked to by the
newborn node.
Thus, the probability that an existing node i receives a new link to the
newborn node at time t is m times i ’s degree relative to the overall degree
of all existing nodes at time t , or
d (t )
m t i
.
∑j =1 dj (t )
14 Networks: Lecture 6 Preferential Attachment Model
Since there are tm total links at time t in the system, it follows that
∑t=1 dj (t ) = 2tm. Therefore, the probability that node i gets a new link in
j
d (t ) time t is i2t .
Hence, we can write down the evolution of expected degrees in continuous
time as
d di ( t )
d (t )
=i ,
dt
2t
with initial condition di (i ) = m (assuming degree is a continuous variable).
This equation has a solution:
di (t ) = m � t �1/2 .
i
As before, expected degrees of nodes are increasing over time.
Hence to ﬁnd the fraction of nodes with...
View
Full Document
 Fall '09
 Acemoglu
 Probability theory, Pareto distribution, power law, Pareto principle, Scalefree network, Complex Networks

Click to edit the document details