Unitary correlations and the Fejér kernel (Q1611950)

The authors study eigenvalues of unitary random matrices. For a function \(f\) in \(L^2\) of the unit circle \(T\) one defines an additive function \(X_f(M)\) on the eigenvalues by adding \(f(t_j)\) for all eigenvalues of the matrix \(M\). Convolutions are defined by \(\int_T f(t)g(xt^{-1}) dt.\) Theorem 1 states for such \(f,g\) \[ \text{Cov}(X_f,X_g)=n(f*\tilde{g})(1)-n(f*\tilde{g}*K_n(1)), \] where \(K_n\) is the Fejér kernel. This is derived from a formula of \textit{F. J. Dyson} [J. Math. Phys. 3, 166-175 (1962; Zbl 0105.41604)], and also given a second proof and extends various known results for special choices of \(f\), such as studies of the numbers of eigenvalues in intervals. Theorem 2 gives a formula for the \(m\)-level correlation function. Theorem 4 generalizes this formula to general representations. Theorems 10, 11 give asymptotic versions of Theorems 1,4.