Extremal eigenvalues and eigenvectors of deformed Wigner matrices (Q2634899)

The analysis of some spectral properties of deformed Wigner matrices constitutes the aim of this contribution. Given a real symmetric or complex Hermitian Wigner matrix of size \(N\times N,\) whose centered entries have variance \(N^{-1}\) and subexponential decay, and \(V\) a real diagonal random matrix of size \(N\times N,\) whose entries are bounded i.i.d. random variables which are independent of \(W\), let us consider the set of the matrices \(H= W + \lambda V\), where \(\lambda\) is a positive real number. If \(W\) belongs to the Gaussian unitary ensemble (GUE, in short), the above model is said to be the deformed GUE. The edge eigenvalues of the deformed GUE are governed by the Tracy-Widom distribution for \(\lambda\ll N^{-1/6}\). For \(\lambda\sim N^{-1/6}\), the fluctuations of the edge eigenvalues change from the Tracy-Widom to Gaussian distribution. Indeed, in [Probab. Theory Relat. Fields 138, No. 1--2, 75--112 (2007; Zbl 1116.60020)], \textit{K. Johansson} has proved that the limiting distribution of the edge eigenvalues for \(\lambda= \alpha N^{-1/6}\) is given by the convolution of the Tracy-Widom and the centered Gaussian distribution, with variance depending on \(\alpha\). These have not been established for the Gaussian orthogonal ensemble or for general Wigner matrices. In the paper under review, the authors deal with the edge behavior of the deformed model in the regime \(\lambda\sim 1\) when \(W\) is a real symmetric or complex Hermitian matrix. For certain \(V\), it is shown that there is another transition for the limit behavior of the largest eigenvalues of \(H\) as \(\lambda\) varies. It is assumed that the distribution of the entries of \(V\) is centered and given by the density \(\mu(v)= Z^{-1} (1+v)^{a} (1-v)^{b} d(v) \mathbf{1}_{[-1,1]}(v)\), \(a,b > -1\), where \(d\) is a strictly positive \(C^{1}\)-function and \(Z\) is a normalization constant. Two limiting distributions appear for the largest eigenvalues of \(H\). The first one is the Weinbull distribution with parameter \(b+1\), assuming that \(b>1\) and \(\lambda > \lambda_{+}\), where \(\lambda_{+}= ( \int_{-1}^{1} \frac {\mu(v) dv}{(1-v)^{2}})^{1/2},\) while the second one is the cumulative distribution function of the centered Gaussian distribution assuming that \(b>1\) and \(\lambda < \lambda_{+}\). When \(\lambda < \lambda_{+}\), the location of the largest eigenvalue depends on the collective behavior of the random variables in the diagonal entries of \(V\) while, when \(\lambda > \lambda_{+}\), the location of the largest eigenvalue is determined by the largest diagonal entry of \(V\). Similar results hold for the lowest eigenvalues. Furthermore, for deformed Wigner matrices, the eigenvectors of the largest eigenvalue are partially localized in the regime where \(\lambda > \lambda_{+}\) and completely delocalized in the case \(\lambda < \lambda_{+}\). To understand the transition from partial localization to delocalization, further efforts are required. Let notice that some results are known on bulk universality for the deformed model (see [\textit{T. Shcherbina}, ``On universality of bulk local regime of the deformed Gaussian unitary ensemble'', \url{arXiv:0804.2116})]). The proofs of the main results are based on a deep analysis of the empirical eigenvalue distribution of the deformed model that converges to a deterministic distribution (the so-called deformed semicircle law), when \(N\) tends to \(\infty\) under some weak assumption on \(\lambda V\). The deformed semicircle law can be defined in terms of a functional equation for the Stieltjes transforms of the limiting eigenvalue distributions of \(W\) and \(\lambda V\) (see [\textit{L. Pastur}, ``On the spectrum of random matrices'', Theor. Math. Phys. 10, 67--74 (1972)]). In a first step, a local law is deduced for the empirical eigenvalue density. In a second step, using the self-averaging property of the Wigner matrix \(W\), it is proved that the imaginary part of the function \(m(z)= N^{-1} Tr (H-z)^{-1}\) can be controlled on scales much smaller than \(N^{-1/2}\). The technical input is the ``fluctuation averaging lemma'' (see [\textit{L. Erdős} et al., Adv. Math. 229, No. 3, 1435--1515 (2012; Zbl 1238.15017)]). Finally, the main steps in the proof of the partial localization of eigenvectors are described. Since the typical eigenvalue spacing at the edge is much larger than \(N^{-1/2}\), the Helffer-Sjöstrand formula allows to translate information on the diagonal entries of the Green function \((H-z)^{-1}\) into information on the eigenvectors at the edge and thus the partial localization of the eigenvectors at the edge is proved.