On the rate of convergence to the semi-circular law (Q6827396)

Consider a Hermitian random matrix \(X=(X_{j,k})_{j,k=1,\ldots,n}\) with independent entries for \(1\leq j\leq k\leq n\) which are not necessarily identically distributed. Assume that \(E(X_{j,k})=0\), \(E(X_{j,k}^2)=1\), and that the \(X_{j,k}\) have a uniform exponential decay in the sense that for all \(j,k\) and all \(t\geq1\), \N\[\NP(| X_{j,k}|>t)\leq \kappa^{-1} e^{-t^\kappa}\N\]\N with some \(\kappa>0\). The authors prove in this case that the Kolmogorov distance between the empirical spectral distributions of the Wigner matrices \(X/\sqrt n\) and the Wigner semicircle law has order \(O(n^{-1}\cdot \log^bn)\) for some \(b>0\). The proof of this nontrivial result uses some recursion argument, the Stieltjes transform as well as methods from large deviations.\N\NFor the entire collection see [Zbl 1264.60004].