Universality in complex Wishart ensembles for general covariance matrices with 2 distinct eigenvalues (Q962215)

Let \(X\) be an \(M \times N\) (\(M \geq N\)) matrix with independent and identically distributed Gaussian entries whose real and imaginary parts have variance \(\frac{1}{2}\) and zero mean. Let \(\Sigma_N\) be an \(N \times N\) positive definite Hermitian matrix with eigenvalues \(a_1 \geq \dots \geq a_N\) (not necessarily distinct). Then the case where \(\Sigma_N\) has only 2 distinct eigenvalues, 1 and \(a\), such that \(N_1\) of its eigenvalues are \(a\) and \(N - N_1\) of them are 1 is considered. The asymptotic limit of the Wishart distribution with \(\frac{N}{M} \rightarrow c\) and \(\frac{N_1}{N} \rightarrow \beta\) as \(M,N,N_1 \rightarrow \infty\) and \(0 < c, \beta = 1\) is studied. The empirical distribution function \(F_N\) of the eigenvalues will converge weakly to a nonrandom positive definite function, which is supported on either 1 or 2 intervals in \({\mathbb R}_+\). It is shown, using a Riemann-Hilbert approach, that the local eigenvalue statistics at the phase transition between the cut 1 and 2 case near the critical point can be described by a Pearcey kernel.