Gaussian fluctuations in complex sample covariance matrices (Q2461954)

Let \(X= (X_{i,j})\) be a complex \(m \times n\), \(m \geq n\) random matrix, the entries of which are independent complex Gaussian random variables with mean zero and variance \(\frac{1}{n}\). Let \(S= X^* X\) be a sample covariance matrix. In the paper, the limiting distribution of eigenvalues firstly analyzed by \textit{V. A. Marchenko} and \textit{L. A. Pastur} [Math. Sb., N. Ser. 72(114), 507--536 (1967; Zbl 0152.16101)] when \(\frac{m}{n} \rightarrow \gamma \geq 1\) as \(n \rightarrow \infty\) is studied. It is shown that the limiting distribution of the \(k\)-largest eigenvalues is described by the central limit theorem under certain condition on \(k\) when \(n\) and \(k\) tend to infinity. Proofs of this behaviour are based on the Costin-Lebowitz-Soshnikov theorem [\textit{O. Costin} and \textit{J. Lebowitz}, Phys. Rev. Lett. 75, 69--72 (1995); \textit{A. B. Soshnikov}, J.~Stat. Phys. 100, No. 3--4, 491--522 (2000; Zbl 1041.82001)]. An exact formula and asymptotics for the mean density of eigenvalues \(\lambda_1, \dots, \lambda_n\) uniformly valid on the entire real axis based on the standard technique of asymptotic analysis of RH problems is given. As a direct consequence the convergence speed of the mean density of eigenvalues to the Marchenko-Pastur distribution density is obtained when \(| \frac{m}{n} - \gamma| = O(\frac{1}{n})\).