Abstract:We consider large non-Hermitian real or complex random matrices $X$ with independent, identically distributed centred entries. We prove that their local eigenvalue statistics near the spectral edge, the unit circle, coincide with those of the Ginibre ensemble, i.e. when the matrix elements of $X$ are Gaussian. This result is the non-Hermitian counterpart of the universality of the Tracy-Widom distribution at the spectral edges of the Wigner ensemble.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is the edge universality of non - Hermitian random matrices at the spectral boundary. Specifically, the author studied large non - Hermitian real or complex random matrices \(X\) with independent and identically distributed centered elements, and proved that the local eigenvalue statistics of these matrices near the spectral boundary are the same as those of the Ginibre ensemble, that is, the eigenvalue statistics when the matrix elements of \(X\) are Gaussian distributed.
### Background and Motivation
Most of the universality results on the local eigenvalue statistics of large random matrices focus on Hermitian matrices. For example, the well - known Wigner - Dyson statistics describe the eigenvalue statistics in the spectral bulk of Hermitian random matrices, and the Tracy - Widom statistics describe the eigenvalue statistics at the spectral boundary. However, little is known about the spectral universality of non - Hermitian models.
### Main Contributions of the Paper
1. **Proof of Edge Universality**: The author proved the edge universality at the spectral boundary of any \(n\times n\) random matrix \(X\) (whose elements are centered and independently and identically distributed). In particular, they removed the fourth - moment matching condition required in previous results.
2. **Application of the Local Circular Law**: By normalizing \(E|X_{ab}|^2=\frac{1}{n}\), the spectrum of the matrix \(X\) converges to the unit disk, and the spectral density conforms to the circular law. The typical distance between the nearest - neighbor eigenvalues is \(O(n^{- 1/2})\).
3. **Convergence of Eigenvalue Correlation Functions**: Select a reference point \(z\) on the boundary of the limit spectrum (i.e., \(|z| = 1\)), and scale the correlation function by a factor of \(n^{-1/2}\) to detect the correlations of individual eigenvalues. The author proved that these scaled correlation functions converge to the correlation functions of the Ginibre ensemble as \(n\rightarrow\infty\).
### Methods and Techniques
1. **Optimal Local Law**: The author uses the optimal local law to control the Green's function of \(H_z\), where \(H_z\) is the symmetrized matrix \(H_z=\begin{pmatrix}0&X - z\\X^* - z&0\end{pmatrix}\). This method allows them to compare the joint distribution of multiple Green's functions with their Gaussian counterparts under the natural Ornstein - Uhlenbeck flow.
2. **Lower - Tail Estimate of the Minimum Singular Value**: The author also provides a lower - tail estimate of the minimum singular value of \(X - z\), especially when \(|z|\approx1\). This estimate ensures that there are no too - small singular values in certain intervals, thus ensuring the optimality of the local law.
3. **Green's Function Comparison Argument**: Through the Green's function comparison argument, the author extends the control of the Ginibre ensemble, thereby proving the edge universality of non - Hermitian matrices.
### Conclusion
This paper successfully proved the edge universality of non - Hermitian random matrices at the spectral boundary, filled the gap in this field, and provided new tools and methods for future research. This result is an important progress in the theory of non - Hermitian random matrices, similar to the Tracy - Widom edge universality in Hermitian matrices.