Central limit theorems for the real eigenvalues of large Gaussian random matrices

Abstract

Let $G$ be an $N \times N$ real matrix whose entries are independent identically distributed standard normal random variables $G_{ij} \sim \mathcal{N}(0,1)$. The eigenvalues of such matrices are known to form a two-component system consisting of purely real and complex conjugated points. The purpose of this note is to show that by appropriately adapting the methods of \cite{KPTTZ15}, we can prove a central limit theorem of the following form: if $\lambda_{1},\ldots,\lambda_{N_{\mathbb{R}}}$ are the real eigenvalues of $G$, then for any even polynomial function $P(x)$ and even $N=2n$, we have the convergence in distribution to a normal random variable \begin{equation} \frac{1}{\sqrt{\mathbb{E}(N_{\mathbb{R}})}}\left(\sum_{j=1}^{N_{\mathbb{R}}}P(\lambda_{j})-\mathbb{E}\sum_{j=1}^{N_{\mathbb{R}}}P(\lambda_{j})\right) \to \mathcal{N}(0,\sigma^{2}(P)) \end{equation} as $n \to \infty$, where $\sigma^{2}(P) = \frac{2-\sqrt{2}}{2}\int_{-1}^{1}P(x)^{2}\,dx$.