Fluctuation around the circular law for random matrices with real entries (Q2042871)

Let \((X_n)\) be a sequence of \(n\times n\) matrices with i.i.d. real valued entries of zero mean and variance \(1/n\) possessing moments of arbitrary order and let \((\sigma_i)_{i=1,\ldots,n}\) denote the eigenvalues of \(X_n\). The circular law states that for bounded continuous functions \(f\) the arithmetic mean of \(f(\sigma_i)_{i=1,\ldots,n}\) converges almost surely to \(\mathbb E[f(U)]\), where \(U\) is uniformly distributed on the unit disc. The aim of this paper is to provide a corresponding CLT for centered linear statistics \[L_n(f)=\sum_{i=1}^n f(\sigma_i)-\mathbb E \sum_{i=1}^n f(\sigma_i),\] which can be seen as a follow up to a study of the corresponding model with complex valued entries by the same authors, and it is nicely described, why the case of real valued matrices is much more involved than its complex valued counterpart. The authors' main result proves that for complex valued functions \(f\) with differentiability of at least order \(2+\delta\) there is convergence in distribution of \(L_n(f)\) to a centered complex Gaussian random variable with variance \(V_f\). An explicit formula for \(V_f\) is given, which depends on the fourth cumulant of the entry distribution, but in contrast to the complex valued case a special symmetry on the real axis is also incorporated. As a consequence, the limiting random field \(L(f)\) is identified as a projection of the Gaussian free field symmetrized with respect to the real axis, reflecting the fact that for complex eigenvalues also their complex conjugates are eigenvalues of the real matrices. As by-products, the authors obtain universality of the small singular values of \(X_n-z\) and asymptotic independence of \(X_n-z_1\), \(X_n-z_2\) for generic \(z_1,z_2\in\mathbb C\setminus\mathbb R\) and compactly supported test functions in terms of the \(k\)-point correlation functions.