Sandbox:DavidKohler/Relativized Alon Conjecture/Trace method
Here, we review the Trace Method.
Basics of the Trace Method
Consider the random regular graph model Gn,d. The trace method allows us to determine information on the eigenvalues of a random regular graph by computing the expected value of the trace of a sufficiently high power of the adjacency matrix. Since the trace of a power of the adjacency matrix is the sum of the corresponding power of its eigenvalues, it passes through expectations and we have
And so
Since λ1 = d for d-regular graphs.
The trace of the kth power of the adjacency matrix of a graph can also be interpreted as the number of closed walks of length k in the graph. In the case of the Gn,d model, each edge is labeled with a letter in the alphabet
where each πi corresponds to a randomly selected permutation of n objects. Hence a walk of length k corresponds to a word w in this alphabet and hence corresponds to a permutation of n objects (using the usual composition of permutations). And the (i,j) entry of the the kth power of the adjacency matrix is the number of words w of length k taking i to j. Given a word w, the probability P(w) that the word takes i to itself is clearly independent of i and hence we have
Finding ways to estimate the right-hand side of this equation allows to estimate the expected value of the second largest eigenvalue.
In Broder & Shamir
In their paper, Broder and Shamir give an estimate of the right-hand side of the above equation.
To do | |
---|---|
![]() |
Describe the Broder Shamir approach and main result |
In Friedman
In his proof of the regular case of the Alon conjecture, Friedman shows that
ROUGH VERSION
First, P(w) has an expansion in 1/n:
Now set
Since P0(w) = 0 for irreducible words and k at least 1.
And so we can define the Irreducible trace as
and get
What makes the estimation work is to show that the gi functions are d-Ramanujan functions for all i less than √d-1/2.