Extreme gaps between eigenvalues of random matrices (Q359681)

The paper studies extreme gaps between eigenvalues of random matrices from the circular unitary ensemble (CUE) and the Gaussian unitary emsemble (GUE). In the case of the CUE, the main results can be stated as follows. Let \(e^{i\theta_1},\dots, e^{i\theta_n}\), where \(0<\theta_1<\dots<\theta_n<2\pi\), be the eigenvalues of a random matrix distributed according to the Haar measure on the unitary group \(U(n)\). It is shown that the point process \[ \sum_{k=1}^n \delta(n^{4/3} (\theta_{k+1} - \theta_k), \theta_k) \] converges to the Poisson point process on \(\mathbb R_+\times [0,2\pi]\) with intensity \[ \frac {u^2 du}{24 \pi} \times \frac {dv}{2\pi}. \] From this result the authors deduce the limiting distribution for the \(k\)-th smallest order statistics \(\tau_{k}^{(n)}\) of the spacings \(\theta_{i+1}-\theta_i\), \(1\leq i\leq n\). Namely, it is shown that \(n^{4/3} \tau_k^{(n)}\) converges in distribution to a random variable \(\tau_k\) with the density \[ \frac{3}{(k-1)!} x^{3k-1} e^{-x^3}, \;\; x>0. \] This limiting distribution can be guessed by recalling that the density between two successive points in the determinantal point process with the sine kernel \(K(x,y) = \frac 1{\pi} \frac{\sin (x-y)}{x-y}\) satisfies \(p(s)\sim \frac{\pi^2}{3}s^2\), as \(s\to 0\), and treating the spacings as independent random variables. For the largest eigenvalue spacing \(T_1^{(n)}\) it is shown that \[ \frac{n}{\sqrt{32 \log n}} T_1^{(n)} \to 1 \] in \(L^p\), for every \(p>0\). Similar results are established in the setting of the Gaussian unitary ensemble. These results are compared with the extreme gaps between the zeros of the Riemann zeta function.