On the spectrum of random anti-symmetric and tournament matrices (Q2822700)

An analysis of the small-scale properties of the spectrum of large random tournament matrices is presented. Let notice that a tournament of size \(N\) is an \(N\times N\) matrix \(D= (D_{i,j})_{1\leq i,j\leq N}\) with entries in \(\{0,1\}\) such that the diagonal entries are zero and \(D_{i,j}= 1- D_{j,i}\) for \(i\neq j.\) The spectrum of \(D\) is complex but \(D\) is associated with a (non-Hermitian) rank-one perturbation of a Hermitian matrix \(M\) such that \(M= 2iD - i (|\mathbf{1}\rangle\langle\mathbf{1}| -I_{N})\), where \( \langle\mathbf{1}| = (1, 1, \dots, 1)\) and \(|\mathbf{1}\rangle= \langle \mathbf{1}| ^{T}\). The matrix \(M\) is said to be anti-symmetric Hermitian. On the other hand, when \(N\) is odd, the spectrum of \(M\) consists of \(0\) and \(N-1\) real eigenvalues symmetrically distributed in pairs about \(0\). If \(N\) is even, then the spectrum is also symmetric but \(0\) is not necessarily an eigenvalue of \(M\). According to Wigner's semicircle law the spectrum is concentrated on \([-2 \sqrt{N}, 2\sqrt{N}]\).NEWLINENEWLINEA first result (Theorem 1) of this contribution deals with sinc kernel universality in the bulk for matrices \(M\). Indeed, if \(W\) is a matrix in the ensemble of antisymmetric \(N\times N\) matrices, chosen uniformly at random, with \(\pm1\) entries and \(M=iW\), for any \(E\in(0,2)\) and any \(b>0\), such that \(I_{E}= [ E-b, E+b] \subset(0,2)\), the \(n\)-point correlation functions of \(M\), with a proper scaling and averaged over \(I_{E}\), converge to those of the sinc kernel process.NEWLINENEWLINEThe second main result describes the spectrum of \(D\). In Theorem 2 it is proved that if \(N\) is odd, the matrix \(D\) has a real eigenvalue \(\lambda_{0}(D)\), at distance of order \(N\) from the imaginary axis such that \(\frac{\lambda_{0}(D)}{(N-1)/2} \rightarrow 1\) almost surely when \(N\rightarrow \infty\) through the positive odd integer numbers. On the other hand, for odd and even \(N\), a probabilistic interlacing property between the eigenvalues of \(M\) and the imaginary part of the eigenvalues of \(D\) is proved as well as an estimate of the real part of the eigenvalues of \(D\) for \(N\) large enough.NEWLINENEWLINEThe proofs of the above results rely on analogs for anti-symmetric Hermitian matrices of those for symmetric and Hermitian matrices by \textit{P. Bourgade} and \textit{H.-T. Yau} [``The eigenvector moment flow and local quantum unique ergodicity'', Preprint, \url{arXiv:1312.1301}] and \textit{L. Erdős} et al. [Ann. Inst. Henri Poincaré, Probab. Stat. 48, No. 1, 1--46 (2012; Zbl 1285.82029)].NEWLINENEWLINEThe asymptotic correlations functions of anti-Hermitian finite-rank deformations of Gaussian unitary ensembles have been deduced in [\textit{Y. V. Fydorov} and \textit{B. A. Khoruzhenko}, ``Systematic analytical approach to correlation functions of resonances on quantum chaotic scattering'', Phys. Rev. Lett. 83, No. 1, 65--68 (1999; \url{doi:10.1103/PhysRevLett.83.65})]. Results concerning the location of bulk eigenvalues for low rank deformations of Wigner and covariance matrices appear in the work of \textit{A. Knowles} and \textit{J. Yin} [Commun. Pure Appl. Math. 66, No. 11, 1663--1749 (2013; Zbl 1290.60004)].