Download as pdf or txt
Download as pdf or txt
You are on page 1of 17

Configuration Model Properties

Giant Component

MA 653: Network Science


Instructor: Ashok Singh Sairam
ashok@iitg.ac.in
Giant component
• u: prob. that the node does not belong to the giant component
• Prob. of a node i with degree k ∉ giant component = 𝑢𝑘
• It is not connected to the giant component via any of its k neighbours
• Implicitly assumes there is no direct connection between the neighbours
• locally tree-like neighbours
• Avg. prob., over entire network, that a node ∉ giant component
• 𝑔0 𝑢 = σ𝑘 𝑝𝑘 𝑢𝑘 , where 𝑝𝑘 is the usual degree distribution
• 𝑔0 𝑢 is called the probability generating function for the prob. distribution 𝑝𝑘
• Prob. that node belongs to giant component, 𝑆 = 1 − 𝑔0 (𝑢)

MA 653: Network Science 2


• We still need to know the value of u
• Prob. that a node is not connected through any of its k neighbours
𝑔1 𝑢 = ෍ 𝑞𝑘 𝑢𝑘
𝑘
here we talk about neighbours, so k is now distributed according to the excess
degree distribution.
𝑔1 𝑢 is the prob. generating function of qk .
𝑘
1 𝑘
1 𝑘−1
1 ′
𝑔1 𝑢 = ෍ 𝑞𝑘 𝑢 = ෍ 𝑘 + 1𝑝𝑘+1 𝑢 = ෍ 𝑘𝑝𝑘 𝑢 = 𝑔0 (𝑢)
𝑘 𝑘 𝑘
𝑘 𝑘 𝑘

𝑔0 denotes the first derivative of g0

MA 653: Network Science 3


1 ′ 1
𝑔0 𝑢 = ෍ 𝑘𝑝𝑘 𝑢𝑘−1
𝑘 𝑘
𝑘
𝑔0′ 1 = ෍ 𝑘𝑝𝑘 = 𝑘
′𝑘
𝑔0 (𝑢) 𝑔0′ (𝑢)
𝑔1 𝑢 = = ′
𝑘 𝑔0 1

• Using this formula, we can compute 𝑔1 𝑢 directly from 𝑔0 𝑢

MA 653: Network Science 4


Example: Giant component
• Consider a network that has nodes only of degree 0, 1, 2 and 3
• If we use the configuration model, does the network have a giant
component? Compute the value of S.

MA 653: Network Science 5


Example: Giant component
• Consider a network that has nodes only of degree 0, 1, 2 and 3
𝑔0 𝑢 = 𝑝0 + 𝑝1 𝑢 + 𝑝2 𝑢2 + 𝑝3 𝑢3
𝑔′0 (𝑢) 𝑝1 + 2𝑝2 𝑢 + 3𝑝3 𝑢2
𝑢 = 𝑔1 𝑢 = ′ = = 𝑞0 + 𝑞1 𝑢 + 𝑞2 𝑢2
𝑔 0 (1) 𝑝1 + 2𝑝2 + 3𝑝3
𝑞0 𝑞0 𝑝1
Solve for u; 𝑢 = 1 𝑜𝑟 ; For g.c to exist 𝑢 = =
𝑞2 𝑞2 3𝑝3
𝑝1 1
That is g.c exists only if < 1 𝑜𝑟 𝑝3 > 𝑝1
3𝑝3 3
𝑝12 𝑝12 𝑝2 𝑝13
𝑆 = 1 − 𝑔0 𝑢 = 1 − 𝑝0 − − 2 −
3𝑝3 9𝑝3 27𝑝32
MA 653: Network Science 6
General solution for the size of giant component
• Given that 𝑔1 1 = σ𝑘 𝑞𝑘 = 1, since 𝑞𝑘 is a properly normalized probability
distribution
• The equation 𝑢 = 𝑔1 (𝑢) will always have a trivial solution u = 1
• What about the non-trivial solution that does give a giant component?
• 𝑔1 (𝑢) is a power series of the probabilities 𝑞𝑘
• The derivatives of 𝑔1 (𝑢) are also non-negative for 𝑢 ≥ 0
• It is in general positive, an increasing function of its argument and upward concave

MA 653: Network Science 7


• Solution of the equation
𝑢 = 𝑔1 (𝑢) is given
by the point at which the
curve y = g1(u) intercepts the
line y = u.
• Condition: slope of the
curve 𝑔1′ 1 > 1 at u = 1 is
greater than slope of the
dotted line

MA 653: Network Science 8


Review
Ex: 12.4 Consider a configuration model in which every node has the
same degree k.
a) What is the degree distribution 𝑝𝑘 ? What are the generating
functions 𝑔0 and 𝑔1 for the degree distribution and the excess
degree distribution?
b) Show that the giant component fills the whole network for all 𝑘 ≥
3.
c) What happens when k=1?

MA 653: Network Science 9


Configuration Model Properties
Small Component

MA 653: Network Science


Instructor: Ashok Singh Sairam
ashok@iitg.ac.in
Size of a small component
• Consider any vertex i of degree k in a small
component
• The component is the “connection” of k
sub-graphs in the same component
• Avg. size of a node of degree k
• 𝑠 = 1 + σ𝑘𝑚=1 𝑡𝑚 = 1 + 𝑘 𝑠𝑚𝑎𝑙𝑙 𝑡
where 𝑡 is the average size of the subnetwork

MA 653: Network Science 11


Small component
• Remove node i from the figure
• The set of nodes becomes small components
in their own right
• Means their average size is 𝑡
• Suppose one of these nodes have degree
k, then
𝑡 =1+k 𝑡
1
𝑡 =
1 − 𝑘 𝑛𝑒𝑖𝑔ℎ𝑏𝑜𝑢𝑟
where 𝑘 𝑛𝑒𝑖𝑔ℎ𝑏𝑜𝑢𝑟 is degree of k neighbours
of nodes in the small component
MA 653: Network Science 12
Average size of small component
𝑢𝑔0′′ (𝑢) 1 1
• 𝑠 = 1 + k 𝑠𝑚𝑎𝑙𝑙 𝑡 = 1 + 𝑢𝑔′′
𝑠𝑖𝑛𝑐𝑒 𝑡 =
𝑔0′ (𝑢) (𝑢) 1− 𝑘 𝑠𝑚𝑎𝑙𝑙
1− ′0
𝑔0 (𝑢)
𝑢𝑔0′′ (𝑢) 1
• 𝑠 =1 + 𝑔′ (𝑢) 𝑢𝑔′′ (𝑢)
0 1− ′0
𝑔0 (𝑢)
𝑔′0 (𝑢)
• We know 𝑔1 𝑢 = or 𝑔′0 𝑢 = 𝑔1 𝑢 𝑔′0 (1) and 𝑔′′0 𝑢 = 𝑔1′ 𝑢 𝑔′0 (1)
𝑔′0 (1)
𝑢2 𝑔0′ (1)
𝑠 =1+
𝑔0 𝑢 [1 − 𝑔1′ 1 ]
• When there is no giant component u=1
𝑢2 𝑔0′ (1) 𝑔0′ (1)
• 𝑠 =1 + = 𝑠 =1 +
𝑔0 𝑢 [1−𝑔1′ 1 ] 1−𝑔1′ (1)

MA 653: Network Science 13


Review
𝑘2 − 𝑘 𝑐2
• Average excess degree = =
𝑘 𝑐1
• c3: sum of excess degree of second neighbours
𝑐2 𝑐2 3−1
𝑐3 = 𝑐2 𝑐 = 𝑐1
1 𝑐1
• On similar lines, average number of neighbours at distance d
𝑑−1
𝑐2
𝑐𝑑 = 𝑐1
𝑐1
𝑐2
• Size of the network grows or falls depending on the value of 𝑐1
𝑐2
• Conf. model has a giant component if >1
𝑐1

MA 653: Network Science 14


Exercise
Ex 12.15: Consider a configuration model network that has nodes only of degree 0,
1, 2 and 3.
(a) In the regime in which there is no giant component, show that the average size
of the component to which a randomly chosen node belongs is
𝑝1 + 2𝑝2 + 3𝑝3 2
𝑠 =1+
𝑝1 − 3𝑝3
(b) In the same regime find the probability that a randomly chosen node belongs
to components of size 1, 2 and 3.

MA 653: Network Science 15


Example
• Consider a network with 100 nodes. The degree distribution of the
network is 𝑝𝑘 = 𝐶𝑘 −2.5 . Find the degrees of the first 10 nodes.

MA 653: Network Science 16


Example
• Consider a network with 100 nodes. The degree distribution of the
network is 𝑝𝑘 = 𝐶𝑘 −2.5 . Find the degrees of the first 10 nodes.

1
• 𝜁 2.5 = 1.3415 ⇒ 𝐶 = = 0.745
1.3425
• We can generate the degree by drawing samples from 𝑘𝑖 ~𝑝𝑘 = 0.745𝑘 −2.5
• k={1, 1, 1, 2, 2, 2, 2, 2, 2, 3 …}

MA 653: Network Science 17

You might also like