Scribd
Upload
29 views

Lec 26

Uploaded by

Tushar
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
29 views

Lec 26

Uploaded by

Tushar
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 11

Configuration Model

Networks with Power-law degree distribution

MA 653: Network Science


Instructor: Ashok Singh Sairam
[email protected]
Networks with power-law degree distributions
• Look at the properties of configuration model networks with power
law distributions
• The power law
0 fo𝑟 𝑘 = 0
𝑝𝑘 = ቊ −𝛼
𝑘 /𝜁 𝛼 for 𝑘 ≥ 1
• Whether a conf. model that follows power law degree distribution will
have a giant component?

MA 653: Network Science 2


Power-law network: Giant component
• Will there be a giant component in a network with power-law
distribution?
𝑐2
• A network will have g.c. if 𝑐1
> 1 ⇒ 𝑘2 − 𝑘 > 𝑘 = 𝑘2 − 2 𝑘 > 0
• For power-law networks
𝑘 −𝛼+1 𝜍(𝛼−1)
• 𝑘 = σ𝑘 𝑘𝑝𝑘 = σ𝑘 =
𝜍(𝛼) 𝜍(𝛼)
2 2 𝑘 −𝛼+2 𝜍(𝛼−2)
• 𝑘 = σ𝑘 𝑘 𝑝𝑘 = σ𝑘 =
𝜍(𝛼) 𝜍(𝛼)
• Thus there is a giant component
𝜍 𝛼 − 2 > 2𝜍(𝛼 − 1)

MA 653: Network Science 3


Graphical solution
• A numerical solution of the
equation
𝜍 𝛼 − 2 = 2𝜍(𝛼 − 1)
gives a value
𝛼 = 3.4788

MA 653: Network Science 4


Power law: Giant component
1
• 𝑔0 𝑢 = σ𝑘 𝑝𝑘 𝑢𝑘 = σ∞
𝑘=1 𝑘 −𝛼 𝑘
𝑢
𝜁(𝛼)
1 σ∞ 𝑘 −(𝛼−1) 𝑢𝑘−1 1
• 𝑔1 𝑢 = σ𝑘 𝑘𝑝𝑘 𝑢 𝑘−1
= 𝑘=1
= σ∞
𝑘=1 𝑘 −𝛼+1 𝑘−1
𝑢
𝑘 σ∞
𝑘=1 𝑘 −(𝛼−1) 𝜁(𝛼−1)
1
•𝑢= σ∞
𝑘=0(𝑘 + 1) −𝛼+1 𝑘
𝑢
𝜁(𝛼−1)
• There is no closed loop solution
• If 𝜁(𝛼 − 1) diverges, then we will get a solution 𝑢 = 0, and indeed it diverges at
𝛼≤2
𝑆 = 1 − 𝑔𝑜 0 = 1 − 𝑝0

MA 653: Network Science 5


• For our particular choice of
degree distribution, there are no
nodes with degree 0
• Means 𝑝0 = 0 and 𝑆 = 1
• This does not mean there are no
small components
• Probability that a randomly chosen
node belongs to a small component
is 0

MA 653: Network Science 6


Diameter of a network with power-law
• Consider two nodes i and j
• Consider two sets of nodes, those at distance s from i and at distance t from j

MA 653: Network Science 7


𝑠 + 𝑡 + 1, if there is an edge between the surfaces
𝑑𝑖𝑗 = ቊ
> 𝑠 + 𝑡 + 1, otherwise
• Let 𝑙 = 𝑠 + 𝑡 + 1
• The diameter of the network is the smallest value of l such that 𝑃 𝑑𝑖𝑗 > 𝑙 is zero
• Prob. of an edge b/w pair of nodes u and v on the surface 𝑘𝑢 𝑘𝑣 /2𝑚
• Since nodes on the surface are reached by following a sequence of edges,
𝑘𝑢 and 𝑘𝑣 are the excess degrees
𝑐2
• Average excess degree = 𝑐
1
1 𝑐2 2
• Hence avg. prob. of an edge = 2𝑚 𝑐1

MA 653: Network Science 8


• Avg. # neighbours at distance 1 = 𝑐1
𝑐2 𝑠−1
• Avg. # neighbours at distance s = 𝑐1
𝑐1
𝑐2 𝑡−1
• Avg. # neighbours at distance t = 𝑐1
𝑐
1
𝑐2 𝑠−1 𝑐2 𝑡−1 𝑐2 𝑠+𝑡−2 2
• Total #pairs at the surfaces = 𝑐1 × 𝑐1 = 𝑐1
𝑐1 𝑐1 𝑐1
𝑐2 𝑠+𝑡−2 2
𝑐1
𝑐2 2 𝑐1
𝑐1
• 𝑃 𝑑𝑖𝑗 > 𝑠 + 𝑡 + 1 = 1 − 2𝑚

MA 653: Network Science 9


Exercise
• Ex 12.16: Consider a conf. model with degree distribution 𝑝𝑘 = 𝐶𝑘𝑎𝑘 ,
where a and C are positive constants and a < 1
a) Calculate the value of C as a function of a
b) Calculate the mean degree of the network
c) Calculate the mean-square degree of the network
d) Hence or otherwise, find the value of a that marks the phase
transition between the region in which the network has a giant
component and the region in which it does not. Does the giant
component exists for larger or smaller values than this?

MA 653: Network Science 10


Exercise
Ex 12.5: Show that in a configuration model network with nodes of
degree 2 and greater, but no nodes of degree 0 or 1, there are no small
components. (or more properly, the fraction of nodes belonging to
such components tend to 0 as 𝑛 → ∞)

MA 653: Network Science 11

You might also like