Sandbox:DavidKohler/Alon try 1/Irreducible eigenvalues

From UBC Wiki

Given a directed graph one can define its irreducible eigenvalues as the eigenvalues of the graph describing irreducible walks in the graph.

Definition

Irreducible eigenvalues are eigenvalues of a corresponding graph in which we only keep track of irreducible (or non-backtracking) walks. We'll define this here for directed graphs, for undirected graphs and for variable-length graphs.

For a directed graph

Let G be a directed graph, we associate to this graph a new graph that we call its graph of irreducible walks and denote by GIrred. This graph is constructed as follow: its vertices are all the directed edges of the graph G and there is an edge from a directed edge e1 to another directed edge e2 if the directed path e1e2 is irreducible (or non-backtracking). Note that walks in this new graph do indeed correspond to irreducible walks in the original graph. We define the irreducible eigenvalues of the graph G to be the eigenvalues of the graph GIrred and will denote by λIrred(G) the largest irreducible eigenvalue of the graph G.

For an undirected graph

If G is an undirected graph, we define its irreducible eigenvalues as the irreducible eigenvalues of its corresponding directed graph (where for each undirected edge we create a pair of opposed directed edges).

For a VLG

Let G be a variable-length graph, then we can define GIrred as a variable-length graph as well by using the same construction as above and making the length of an edge in the graph GIrred be the length of the corresponding irreducible walk in the graph G.

Example

Consider the directed graph G on four vertices arranged as a basis (v1, v2, v3, v4) with adjacency matrix (in this basis):

Then the irreducible graph GIrred has eight vertices (the eight directed edges of the graph G) which we'll arrange in the basis (e12, e21, e23, e32, e24, e42, e34, e43) in which the adjacency matrix is: