Optimal local law and central limit theorem for \(\beta\)-ensembles (Q2113501)

The \(\beta\)-ensemble is a collection of real random variables (eigenvalues) \(\lambda_1< \dots <\lambda_N\) whose joint density is proportional to \[ e^{-\frac {\beta N}{2}\sum_{k=1}^N V(\lambda_k)} \prod_{1\leq j < k \leq N} |\lambda_k-\lambda_j|^\beta, \] where \(V(\lambda)\) is a function satisfying some hypotheses. The authors study the fluctuations of \(\lambda_k\) around their expected values as \(N,k\to\infty\). They prove a rigidity result on the eigenvalues which is an estimate on the maximum of \(|\lambda_k - \gamma_k|\) taken over all \(k\in [a_N, N-a_N]\) with \(a_N/\log N\to\infty\), where \(\gamma_k\) is the \(k/N\)-th quantile of the equilibrium measure associated with \(V\). Another result is a multivariate central limit theorem for the fluctuations of the \(\lambda_k\)'s in the bulk of the spectrum wih covariance function corresponding to a logarithmically correlated field. In particular, the authors show that each eigenvalue in the bulk fluctuates on scale \(\sqrt {\log N}/N\). They also establish a similar CLT (again, with logarithmic correlations), for the logarithmic potential generated by the eigenvalues. An important ingredient in the proofs of these theorems is an upper bound on \(\mathbb E |s_N(z) - m_V(z)|^q\), where \(s_N(z) = \frac 1N \sum_{k=1}^N \frac{1}{\lambda_k-z}\) is the Stieltjes transform of the empirical eigenvalue distribution and \(m_V(z)\) is the Stieltjes transform of the equilibrium measure associated with the external potential \(V\). Contrary to much progress on random matrix universality, the proofs do not proceed by comparison with the Gaussian case. In fact, the results are new even for the Gaussian \(\beta\)-ensembles (corresponding to quadratic external potential \(V(x) = x^2/2\)).