Two CLTs for Sparse Random Matrices

Abstract: Let

G = G (n, p_{n})

be a homogeneous Erdos-R'enyi graph,

A

its adjacency matrix with eigenvalues

l a m b d a_{1} (A) g e q l a m b d a_{2} (A) g e q ... g e q l a m b d a_{n} (A) .

In such cases, it has been shown using local laws that

l a m b d a_{2} (A)

can exhibit different behaviors: real-valued Tracy-Widom for

p_{n} g g n^{- 2 / 3},

normal for

n^{- 7 / 9} l l p_{n} l l n^{- 2 / 3},

or a mix of the two for

p_{n} = c n^{- 2 / 3} .

Additionally, this technique renders the largest eigenvalue

l a m b d a_{1} (A),

separated from the rest of the spectrum for

p_{n} g g n^{- 1},

has Gaussian fluctuations when

p_{n} g g n^{- 1 / 3} .

This paper extends the range of the last convergence to

n^{e p s i l o n - 1} l e q p_{n} l e q f r a c 12, e p s i l o n i n (0, 1) :

the tool behind this is a CLT for the eigenvalue statistics of

A,

which is justified by the method of moments.

This page was built for publication: Two CLTs for Sparse Random Matrices