Bulk eigenvalue fluctuations of sparse random matrices (Q2240477)

Let \(A\) be the adjacency matrix of an Erdős-Rényi random graph \(G(N,p)\). The authors are interested in the ``sparse'' regime when \(N\to\infty\) and \(p= p(N) \in [N^{-1+\epsilon}, N^{-\epsilon}]\) for a fixed \(\epsilon>0\). They identify the joint limiting distributions of the eigenvalues away from \(0\) and the spectral edges. More precisely, let \(\lambda_1\leq \dots \leq \lambda_N\) be the eigenvalues of \(A\) and let \(X_1,\dots, X_N\) be their centered and suitably normalized versions. If \(\mathcal I_N\) denotes the set \([\tau N, N/2 - N^{1-\delta}]\cup [N/2 + N^{1-\delta}, (1-\tau) N]\) with a fixed \(\tau \in (0, 1/2)\) and a suitable \(\delta>0\). Let \(k\in \mathbb N\) be fixed and \(i_1= i_1(N),\dots, i_k= i_k(N)\) be sequences of indices in \(\mathcal I_N\). The authors prove a joint central limit theorem for the random vector \((X_{i_1},\dots, X_{i_k})\). The covariance matrix of the limit normal distribution consists of \(\pm 1\)'s which means that the eigenvalues fluctuate simultaneously: the correlation of two eigenvalues of the same, respectively different, sign is asymptotically \(1\), respectively \(-1\). The authors also prove CLTs for the eigenvalue counting function and the trace of the resolvent at mesoscopic scales.