Empirical spectral distributions of sparse random graphs (Q1983055)

This work studies the spectrum of a random multigraph \(G_n=([n],E_n)\) of \(n\) vertices with degree sequences \(D_i^{(n)}\) for \(i\in[n]\). Assume that \(\sum_{i=1}^nE_i^{(n)}=2|E_n|=n\omega_n(1+o(1))\) with \(\omega_m=o(n)\). The empirical spectral distribution for a symmetric matrix \(A\) is \(L^A=\frac1n\sum_{i=1}^n\delta_{\lambda_i}\), where \(\lambda_1\ge\cdots\ge\lambda_n\) are the eigenvalues. The normalized degrees \(D_{U_n}^{(n)}/\omega_n\) are uniformly integrable with the expectation of its square equal \(o(\sqrt{n/\omega_n})\), where \(U_n\) is uniformly selected from \([n]\). Let \(\Lambda_n\) be the diagonal matrix with diagonal elements \(D_i^{(n)}/\omega_n\) for \(i\in[n]\). If the empirical spectral distribution \(L^{\Lambda_n}\) converges weakly to a limit \(\nu\), it is shown that the empirical spectral distribution \(L^{A_n}\) converges weakly in probability to \(\nu\circ\sigma\), where \(A_n=A/\sqrt{\omega_n}\), \(A\) is the adjacency matrix of \(G_n\), \(\sigma\) is the standard semicircle law, and \(\circ\) means the free multiplicative convolution. For the entire collection see [Zbl 1462.60001].