On a problem of E. Meckes for the unitary eigenvalue process on an arc

Abstract

We study the problem originally communicated by E. Meckes on the asymptotics for the eigenvalues of the kernel of the unitary eigenvalue process of a random $n\times n$ matrix. The eigenvalues $p_j$ of the kernel are, in turn, associated with the discrete prolate spheroidal wave functions. We consider the eigenvalue counting function $|G(x,n)|:=\#\{j:p_j>C{e}^{-xn}\}$, ($C>0$ here is a fixed constant) and establish the asymptotic behavior of its average over the interval $x\in (\lambda -\epsilon ,\lambda +\epsilon )$ by relating the function $|G(x,n)|$ to the solution $J(y)$ of the following energy problem on the unit circle $S^1$, which is of independent interest. Namely, for given $\theta$, $0<\theta <2\pi$, and given $q$, $0<q<1$, we determine the function $J(q)=\inf \{I(\mu ):\mu \in \mathcal{P}(S^1),\mu (A_{\theta })=q\}$, where $I(\mu ):=\iint \log \frac{1}{|z-\zeta |}d\mu (z)d\mu (\zeta )$ is the logarithmic energy of a probability measure $\mu$ supported on the unit circle and $A_{\theta }$ is the arc from $e^{-i\theta /2}$ to $e^{i\theta /2}$.

15 pages, 1 figure