Expected value for edge existence and independence, given a class of random graphs

191 Views Asked by At

Set up

Let $G$ be a graph with $n$ vertex: $\{ e_i \}_{i=1}^n = E$. Define, for all $i$ and $j$ from $1$ to $n$,

$e_{i,j}: E \times E \rightarrow \{0,1\}$,

the random variable which takes $1$ if the edge between $i$ and $j$ exists and $0$ otherwise.

I'm interested in calculating the expectation of $e_{i,j}$ ($\mathbb{E}(e_{i,j})$) and the independence of these variables ($Var(e_{i,j}, e_{k,l})$), knowing the "class" of the graph I'm dealing with.

For example for an Erdos-Renyi graph with probability $p$, it's trivial to show:

$\mathbb{E}(e_{i,j})=p$ and $Var(e_{i,j}, e_{k,l})=0$ for all $(i,j) \neq (k,l)$.

Questions

  • Is it possible to estimate these quantities for, let's say, a scale free graph with fixed degree distribution?
  • Do other classes of graphs exist, as the ER-graph, with "easy" computations?
  • Even if in general the variance will be strictly positive, I have the intuition that for a large class of graphs, it can be bounded. Is there any result about that?

Example: Scale free network case

Naively I thought that, if I want to know the probability of the existence of a certain vertex, let's say $e_{i,j}$, knowing that the edge $e_i$ as $k$ connections, I should just compute:

$\mathbb{P}($ existence of $e_{i,j}\,| \, \|e_i\|=k<n) = \tfrac{k}{n}$ so that

$\mathbb{E}(e_{i,j})= \mathbb{P}($ existence of $e_{i,j}) = \sum_{l=0}^n \mathbb{P}($ existence of $e_{i,j}\,| \, \|e_i\|=l) \mathbb{P}(\| e_i \| = l) = \sum_{l=0}^n \tfrac{l}{n} \tfrac{A}{l^{\gamma}} \approx \tfrac{A}{n} \sum_{l=1}^n \tfrac{1}{l^{\gamma-1}}$, with $A>0$ normalizing constant and $\gamma$ the fixed exponent for the degree distribution.

Is it correct? How can I estimate the variance know?

A general Example: configuration model

We can take a configuration model, which includes the previous case, chosing the degree sequence to follow a power law distribution. Reading here Probability that exists at least an edge in the configuration model, it gives me an hint, saying that the probability $p_{ij}$ that exists at least an edge between two vertices $i$ and $j$ is evaluated as $p_{ij} = \frac{k_i k_j}{2m}$, where $k_i$ and $k_j$ denote the degrees of two particular vertices in the network and $2m = \sum_i k_i $.

(in particular, following the same reference notes: Let $v$ be an array of length $2m$ and let us write the index of each vertex $i$ exactly $k_i$ times in the vector $v$. Each of these entries will represent a single edge stub attached to vertex $i$. Then, we take a random permutation of the entries of $v$ and read the contents of the array in order, in pairs. For each pair that we read, we add the corresponding edge to the network.)

  • Is the expectation made by computing this sum: $\mathbb{E}(e_{i,j})= \sum_{i,j}^n p_{i,j} \mathbb{P}(k_i) \mathbb{P}(k_j)$ ?

  • How can I compute the variance?