Edge universality for non-Hermitian random matrices (Q2660384)

The paper under review aims to proving edge universality for random matrix \(X_{n\times n}\) with centred i.i.d. entries in the edge regime, typically removing the four moment matching condition from \textit{T. Tao} and \textit{V. Vu} [Ann. Probab. 43, No. 2, 782--874 (2015; Zbl 1316.15042)]. This result is the non-Hermitian analogue of the Tracy-Widom edge universality in the Hermitian case with two essential issues. The first one is that non-normal spectral analysis of the non-Hermitian case is difficult since the resolvent \((X-z)^{-1}\) of a non-normal matrix is not effective for studied eigenvalues near \(z\); the second one is the lack of a good analogue of the Dyson Brownian motion. The authors here present two key inputs to prove the edge universality by proposing the first main key on the optimal local law for the resolvent of \(H^z\) (log-determinant of the symmetric matrix in (1)) and the second key on a lower tail estimate for the lowest singular value of \(X-z\) when \(|z| \sim 1\), to control the i.i.d. entries of X from the Ginibre ensemble with a Green function comparison argument using the optimal local law for \(H^z\). The main result (Theorem 1) and the assumptions are given in section 2 with the outline of its proof. Theorem 1 (edge universality) follows from combining the local law for \(H^z\) in Proposition 1, Girko's Hermitization identity, the bound on small singular value in Proposition 2 and a long-time Green function comparison argument. Section 1 lays out the background and introduction of the problem, with the model and notations; the main results are presented in Section 2. Section 3 is devoted to estimate the lower tail of the smallest singular value of \(X-z\), and starts the real or complex Ginibre matrix \(\tilde{X}\) and \(X_t\) to formulate \(H_t^z\) by replacing the non-Hermitian matrix \(X\) with \(X_t\). Proposition 3 estimates the difference between the normalized trace of resolvent \(G_t = G_t^z = (H_t-w)^{-1}\) for \(w\in \mathbf{H}\) by Itô's lemma and Assumption A on moments; then Proposition 2 on the bound of small singular value is given. Section 4 proves the main result on edge universality with the proof of Proposition 4 as technical support, several lemmas and propositions are proved to estimate the probability that there exists an eigenvalue of \(H^z\) very close to zero. The estimate splits into four analogous Ginibre ensembles by replacing \(H^z\) by \(\tilde{H}^z, G^z, \tilde{G}^z\). Appendix A gives an extension of the local law and use a standard grid argument to conclude the local law simultaneously in all \(z\). It is interesting to understand Assumption A and Assumption B in the edge universality without moment matching conditions, also it is quite interesting to know those local law and spectral singular analysis for the family of the resolvent \(H_t^z\).