Scribd
Upload
11 views3 pages

4.1 Configuration Model Introduction

The document discusses the configuration model, a generalized random graph model that allows for any degree distribution in random graphs. It explains how to match stubs to form edges while considering the probabilities of multi-edges and self-edges, which become negligible in large sparse networks. Additionally, it introduces the concept of excess degree distribution, highlighting that neighbors of a node tend to have a higher average degree than the node itself.

Uploaded by

Riley Collins
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
11 views3 pages

4.1 Configuration Model Introduction

The document discusses the configuration model, a generalized random graph model that allows for any degree distribution in random graphs. It explains how to match stubs to form edges while considering the probabilities of multi-edges and self-edges, which become negligible in large sparse networks. Additionally, it introduces the concept of excess degree distribution, highlighting that neighbors of a node tend to have a higher average degree than the node itself.

Uploaded by

Riley Collins
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/ 3

4.

1 The configuration model

In the previous section we considered a random graph model in which the


node degree has a Poisson distribution. Here we consider how to extend this to
random graphs with any degree distribution pk where pk is the probability that
a node selected uniformly at random has degree k. To do this, we develop the
configuration model which is a generalised random graph model.
For each node i ∈ {1, 2, ..., n}, specify the degree ki . Node i has
Pki ‘stubs’
(meaning the end of an edge where it connects to the node) giving i ki = 2m
stubs in total. The configuration model matches these stubs to each other at
random to form edges and a network with the required degree structure. Note
there must be an even number of stubs to avoid any left over at the end of this
matching process. In this process, to make the analysis tractable, we also allow
the possibility of multi-edges and self-edges being formed but we shall see that
these become negligible in large sparse networks.
Let us consider the probability pij that a stub of node i is connected to a
stub of node j. There are ki kj ways in which node i can be connected to node
j. Also there are 2m2 = m(2m − 1) ≈ 2m2 possible edges in the network. So
the probability that a particular edge connects i to j is ki kj /2m2 . Since there
are m edges, the total probability of a connection between i and j is
ki kj
pij = .
2m

1 Probability of a multiedge
The probability of one edge is as above. Given this, the probability of a second
edge forming between nodes i and j must account for one of the stubs being
used on each node and is (ki − 1)(kj − 1)/(2m) so the probability of at least 2
edges is:
ki kj (ki − 1)(kj − 1)
P (at least 2 edges) =
4m2
The expected number of multiedges is then:
1 X 1 X X
k i k j (k i − 1)(k j − 1) = k i (k i − 1) kj (kj − 1)
8m2 ij 2hk 2 in2 i j
2
1 hk 2 i − hki

=
2 hki

1
where the factor of 12 prevents double-counting of edges and we used 2m = nhki.
This remains constant as n gets large and so the probability of a multiedge tends
to zero as 1/n.

2 Probability of a self-edge
For a node i, the number of ways it can be connected to itself is
 
ki 1
= ki (ki − 1)
2 2
so the probability of any given edge going from node i to itself is
ki (ki − 1)
4m2
when m is large. Here we have used the fact that there are 2m

2 = m(2m − 1)
possible places for an edge to fall. Since there are m edges, the probability of a
self-edge from node i to itself is then
ki (ki − 1)
Pii =
4m
The expected number of self-edges in the network is:
X X ki (ki − 1) nhk 2 i − nhki
Pii = =
i i
4m 4m
hk 2 i − hki
=
2hki
Again, this remains constant as n → ∞ so the density of self-edges tends to
zero as 1/n.

3 Excess degree distribution


Let us consider a node and follow one of its edges to a neighbour. What is the
degree distribution of this neighbour?
There are 2m − 1 stubs in total excluding the one we are following. The
probability of following the edge to a particular node of degree k is k/(2m − 1),
or k/2m when m is large. The number of nodes of degree k is npk so the
probability that the neighbour has degree k is:
k kpk
npk · = . (1)
2m hki
We can use this to calculate the average degree of a neighbour:
X kpk hk 2 i
k· =
hki hki
k

2
so the average degree of a neighbour is not the average degree hki. In fact it is
bigger:

hk 2 i 1
hk 2 i − hki2

− hki =
hki hki
σ2
= >0
hki

where σ 2 is the variance of the degree distribution. On a social network, this


means that your friends have more friends than you do - on average.
A distribution that we will often need is the ‘excess degree’ distribution. This
is the distribution of a neighbour excluding the stub from the original node -
i.e. one less than the neighbour degree distribution. We can obtain it from the
degree distribution by noting that the probability of an excess distribution of
k is equal to the neighbour degree distribution of k + 1. So, we can obtain the
excess degree distribution from the neighbour degree distribution (Eq 1):

(k + 1)pk+1
qk = .
hki

You might also like