Expander Graphs

1 Preliminaries

This blueprint follows Fan Chung: Spectral Graph Theory (first published: AMS, 1992).
Unless otherwise specified, all matrices are \(n \times n\) where \(n\) is the number of vertices in \(G\). We omit natural language proofs for theorems that only involve matrix multiplication and basic operations, although we formally prove them in Lean.

1.1 Weighted Graph

Definition 1 Weighted Graph
#

We extend the definition of a general graph to that of a weighted graph by defining a weight function over its edges.

\[ \omega : \beta \to \mathbf{R}_{+} \]

with the condition that

\[ e \in edgeSet \iff 0 {\lt} w(e) \]

We also add an arbitray orientation for undirected edges.

Definition 2
#

In a graph G, let \(d_v\) denote the degree (number of incident edges) of vertex \(v\).

Definition 3
#

For \(k \in \mathbb {N}\), a graph is said to be \(k\)-regular if every vertex \(v\) has degree \(d_v = k\).

Definition 4
#

The volume of a graph is defined as the sum of the degrees of its vertices.

Definition 5
#

A vertex v is said to be isolated if \(d_v = 0\).

Definition 6
#

A graph is said to be nontrivial if it contains at least one edge.

1.2 Matrices and Operators

Definition 7
#

The adjacency matrix of a graph \(G\), denoted \(A\) is defined as follows,

\[ A(u,v)= \begin{cases} 1 & \text{if } u \text{ and } v \text{ are adjacent}, \\ 0 & \text{otherwise}. \end{cases} \]
Definition 8
#

Consider the matrix L, defined as follows

\[ L(u,v)= \begin{cases} d_v & \text{if}\ u=v, \\ -1 & \text{if}\ u \text{ and }v \text{ are ajacent},\\ 0 & \text{otherwise}. \end{cases} \]
Definition 9 Laplacian Matrix
#

Consider the matrix \(\mathcal{L}\), defined as follows

\[ \mathcal{L}(u,v) = \left\{ \begin{array}{cl} 1 & \text{if } u = v \text{ and } d_v \neq 0, \\[1.5ex] -\dfrac {1}{\sqrt{d_u d_v}} & \text{if } u \text{ and } v \text{ are adjacent}, \\[1.5ex] 0 & \text{otherwise}. \end{array} \right. \]
Definition 10
#

Let \(T\) denote the diagonal matrix with the \((u,v)\)-th entry having value \(d_v\).

Definition 11
#

We also define \(T^{-1/2}\) with the convention that \(T^{-1/2} = 0\) when \(d_v = 0\).

Lemma 12 Laplacian symmetric normalization
\[ \mathcal{L} = T^{-1/2} \text{ } L \text{ } T^{-1/2} \]
Definition 13
#

The Laplacian of a graph can be viewed as an operator on the space of functions \(g : V(G) \rightarrow \mathbb {R}\).

Definition 14
#

The Laplacian Operator satisfies

\[ \mathcal{L}g(u) = \frac{1}{\sqrt{d_u}} \sum _{\substack {v \\ u \sim v}} \left( \frac{g(u)}{\sqrt{d_u}} - \frac{g(v)}{\sqrt{d_v}} \right). \]
Lemma 15

When \(G\) is \(k\)-regular,

\[ \mathcal{L} = I - \frac{1}{k} A, \]

where \(k {\gt} 0\), \(A\) is the adjacency matrix of \(G\), and \(I\) is the identity matrix.

Lemma 16
#

For a general graph without isolated vertices, we have

\[ \begin{aligned} \mathcal{L} = I - T^{-1/2} A T^{-1/2}. \end{aligned} \]
Definition 17 Boundary Operator
#

Let \(S\) denote the matrix whose rows are indexed by the vertices and whose columns are indexed by the edges of G. Each column corresponding to an edge \(e = \{ u, v\} \) has an entry \(1/\sqrt{d_u}\) in the row corresponding to \(u\), an entry \(−1/\sqrt{d_v}\) in the row corresponding to \(v\), and has zero entries elsewhere.

Lemma 18
#

We note that \(\mathcal{L}\) can be written as

\[ \mathcal{L} = SS^*. \]
Lemma 19
#

The Matrix \(\mathcal{L}\) is hermitian.

Proof

By spectral theorem.

Definition 20
#

Following 19, the eigenvalues of \(\mathcal{L}\) are all real and non-negative.

Definition 21
#

The Dirichlet sum of a graph \(G\) for a function \(f : \alpha \to \mathbb {R}\) is defined as

\[ \sum _{u \sim v}{(f(u)-f(v))^2}. \]
Definition 22
#

Let \(\tau \) denote the constant function which assigns the value \(1\) on each vertex.

Theorem 23

\(T^{1/2} * \tau \) is an eigenfunction of \(\mathcal{L}\) with eigenvalue \(0\).

1.3 Basic facts about the spectrum of a graph

Lemma 24
#

The sum of the eigenvalues of the graph is at most its number of vertices, that is

\[ \sum _{i}{\lambda _i} \le n. \]
Proof

Follows from considering the trace of \(\mathcal{L}\).

Lemma 25

In the previous lemma, the equality holds if and only if \(G\) has no isolated vertices.

Lemma 26
#

When \(n{\gt}1\), we obtain the following upper bound on the second eigenvalue:

\[ \lambda _1 \le \frac{n}{n-1}. \]