[Paper Review] Gaussian fluctuation for the number of particles in Airy, Bessel, sine and other determinantal random point fields

Alexander Soshnikov|ArXiv.org|Jul 15, 1999

Random Matrices and Applications35 references18 citations

TL;DR

This paper establishes the Central Limit Theorem (CLT) for the number of particles in determinantal random point fields associated with Airy, Bessel, and sine kernels—key models in random matrix theory. Using a general theorem by Costin and Lebowitz, it proves Gaussian fluctuation of particle counts in local bulk and edge regimes, with explicit logarithmic variance asymptotics for eigenvalue counts in shrinking intervals near the spectrum edge or bulk.

ABSTRACT

We prove the Central Limit Theorem for the number of eigenvalues near the spectrum edge for hermitian ensembles of random matrices. To derive our results, we use a general theorem, essentially due to Costin and Lebowitz, concerning the Gaussian fluctuation of the number of particles in random point fields with determinantal correlation functions. As another corollary of Costin-Lebowitz Theorem we prove CLT for the empirical distribution function of the eigenvalues of random matrices from classical compact groups.

Motivation & Objective

• To establish the Central Limit Theorem (CLT) for the number of eigenvalues in local bulk and edge regimes of random matrix ensembles.
• To analyze the Gaussian fluctuation of particle counts in determinantal point processes governed by Airy, Bessel, and sine kernels.
• To derive precise asymptotic expressions for the variance of particle counts in shrinking intervals near the spectrum edge or bulk.
• To extend the CLT to eigenvalue statistics of classical compact groups (U(n), SO(n), Sp(n)) under appropriate scaling.
• To provide a unified framework using the Costin-Lebowitz theorem for proving CLT in determinantal processes with general kernels.

Proposed method

• Applies a general CLT for determinantal point processes due to Costin and Lebowitz, which links particle count fluctuations to trace-class properties of the correlation kernel.
• Uses the asymptotic behavior of orthogonal polynomials (Hermite, Bessel, etc.) to derive limiting kernels such as the Airy kernel and sine kernel.
• Employs rescaling of eigenvalues near the edge (e.g., λ_i = x + y_i / ρ_n,1(x)) to obtain local correlation functions in the bulk and edge regimes.
• Derives the limiting correlation kernel K(x,y) via Plancherel-Rotach asymptotics for Hermite polynomials, leading to the Airy kernel K(x,y) = [Ai(x)Ai'(y) - Ai'(x)Ai(y)] / (x - y).
• Computes the variance of particle counts ν_n in an interval [θ, θ + δ_n] by evaluating the trace of K_n · χ_I · (I - K_n) · χ_I.
• Establishes convergence in distribution of the normalized particle count (ν_n - Eν_n)/√Var(ν_n) to a standard normal variable N(0,1) using the Costin-Lebowitz framework.

Experimental results

Research questions

• RQ1 Does the number of eigenvalues in a shrinking interval near the edge of the spectrum for GUE follow a central limit theorem?
• RQ2 What is the asymptotic variance of the number of particles in a local interval for determinantal point processes with Airy, Bessel, or sine kernels?
• RQ3 How does the CLT for particle counts differ between the bulk and edge regimes of classical random matrix ensembles?
• RQ4 Can the CLT be extended to eigenvalue statistics of classical compact groups (U(n), SO(n), Sp(n)) under appropriate scaling?
• RQ5 What is the joint limiting distribution of particle counts in multiple disjoint intervals for these ensembles?

Key findings

• For the Airy kernel (edge regime), the normalized number of particles in [θ, θ + δ_n] converges in distribution to N(0,1), with Var(ν_n) ~ (1/π²) log(nδ_n) for θ > 0 and (1/(2π²)) log(nδ_n) for θ = 0.
• For the sine kernel in the bulk (U(n), SO(n), Sp(n)), the variance of particle count in an interval of length δ_n is asymptotically (1/π²) log(nδ_n) when nδ_n → ∞.
• The joint distribution of normalized particle counts in multiple disjoint intervals converges to a centered Gaussian sequence with covariance Eξ_kξ_l = δ_{k,l} - (1/2)(δ_{k,l+1} + δ_{k,l-1}).
• For U(n), the variance of ν_n in [θ, θ + δ_n] is (1/π²) log(nδ_n) + o(1), and the normalized count converges to N(0,1).
• The CLT holds uniformly for intervals satisfying 0 < δ_n < 2π - θ - ε and nδ_n → ∞, ensuring the local bulk or edge scaling is valid.
• The results are extended to Bessel-type kernels via asymptotic analysis of Bessel functions, confirming the same CLT structure in the hard edge regime.