Extended proof of the Statement: Convergence rate of the expected spectral functions of symmetric random matrices Ξ n is equal to O (n—1/2) and the method of critical steepest descent (Q4780942)
From MaRDI portal
!
This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:
scientific article; zbMATH DE number 1833687
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | Extended proof of the Statement: Convergence rate of the expected spectral functions of symmetric random matrices Ξ n is equal to O (n—1/2) and the method of critical steepest descent |
scientific article; zbMATH DE number 1833687 |
Statements
Extended proof of the Statement: Convergence rate of the expected spectral functions of symmetric random matrices Ξ n is equal to O (n—1/2) and the method of critical steepest descent (English)
0 references
21 November 2002
0 references
resolvents
0 references
spectral functions
0 references
Hermitian Gaussian matrices
0 references
Christoffel-Darboux equality
0 references
Hermitian polynomials
0 references
Plancherel-Rotach asymptotics
0 references
convergence rates
0 references
spectral distributions
0 references
0.9117085337638856
0 references
0.9117077589035034
0 references
The author deals with the convergence of normalized spectral functions (n.s.f.) \(\mu_n(x,\Sigma_n)\) of symmetric matrices \(\Sigma_n\) with independent random entries on the main diagonal. In his papers [see, e.g., Random Oper. Stoch. Equ. 6, No. 4, 359-408 (1998; Zbl 0912.60004); `An introduction to statistical analysis of random arrays.' (1998; Zbl 0934.62055] he proved that under some conditions the n.s.f. \(\mu_n(x,\Sigma_n)\) converge to the corresponding n.s.f. \(\mu_n(x,H_n)\) of the Hermitian Gaussian random matrices \(H_n\): \[ \sup_{x}|{\mathbf E}\mu_n(x,\Sigma_n)-{\mathbf E}\mu_n(x,H_n)|\leq cn^{-1/2}. \] The proof is based on the developed asymptotic techniques for Hermitian Gaussian matrices. In this paper the author clarifies these techniques and corrects some misprints from previous papers. He considers asymptotics of Hermitian polynomials on the real line and generalizes the Plancherel-Rotach type asymptotics. The starting point for the analysis of densities of expected normalized spectral matrices of Hermitian Gaussian matrices is a formulation of the corresponding orthogonal polynomial problem as a problem for the Christoffel-Darboux equality for Hermitian polynomials. For this purpose the ``critical steepest descent method'' is used. A new proof is given for estimates of the convergence rate for the normalized spectral function of Hermitian Gaussian matrices \(H_n\): \[ \sup_{x}|{\mathbf E}\Gamma_n(x)-W_n(x)|\leq cn^{-1/2}, \] where \(W_n(x)\) is a Wigner semi-circle law d.f. The first result on the convergence rate of the n.s.f. outside of the support of the semi-circle density for Gaussian symmetric matrices with independent entries on the main diagonal and above was established by \textit{B.V. Bronk} [J. Math. Phys. 5, 1661-1663 (1964; Zbl 0151.34201)]. The next result was presented by \textit{V.L. Girko} in Ukr. Math. J. 41, No. 1, 20-24 (1989; Zbl 0677.60037), where a new method (called REFORM method) was proposed and a basic equation for the resolvent of random matrices was described. Some remarks of the author in the introduction will help to get an impression of the article: 1. By the REFORM method the following inequality it was established: \[ \sup_{x\geq 2+\delta}|1-{\mathbf E}\mu_n(x)|+ \sup_{x\leq -2-\delta}{\mathbf E}\mu_n(x) \leq cn^{-4\delta_1/5}, \] where \(\delta>0\) is a certain number. I never stated that I have established the convergence rate \(N^{-1/14}\) inside of the support of the semi-circle law, but some authors continue to cite my pseudo-result. I consider their publications as some kind of mockery. 2. \textit{Z.D. Bai} [Ann. Probab. 21, No. 2, 625-648 (1993; Zbl 0779.60024); ibid., 649-672 (1993; Zbl 0779.60025)] obtained the convergence rate \(O(n^{-1/4})\) of the n.s.f. of symmetric random matrices under the conditions that \(\xi_{ij}^{(n)}=n^{-1/2}\xi_{ij}\), the random entries \(\xi_{ij},\;i\geq j,\;ij=1,2,\dots\), are independent, \({\mathbf E}\xi_{ij}=0\), \({\mathbf E}\xi_{ij}^2=1\), \({\mathbf E}\xi_{ij}^4<\infty\) under a certain discussed definition of the root from complex numbers. Usually such a definition is delicate in the theory of random matrices, hence subject to multiple interpretations and controversies. This definition was clarified in many papers [see, e.g., the above cited author's introduction to the statistical analysis of random arrays]. 3. In the first part of this paper we discuss systematically ideas underlying the so-called REFORM method approach to the problem of finding the convergence rate of n.s.f. We give a completely new method for finding the convergence rate of expected spectral distributions of large random matrices. I have established [see `Random matrices.' (Russian) (1975; Zbl 0344.60001)] an equality which is the basis of my proof and which has been used by some authors without any reference. Therefore, to brake off speculation, we give the original name for this equality: `The equation \(M_1\) of the First Law equation for eigenvalues and eigenvectors of symmetric random matrices.' 4. The main idea consists in using the invariance principle for random matrices, in using certain expressions for distributions of eigenvalues of Gaussian Hermitian random matrices and the `imaginary logarithmic' inequality. This method allows us to find a more accurate convergence rate of normalized spectral functions of random matrices to the n.s.f. of Hermitian Gaussian matrices and, in particular, to the limit semi-circle law and to generalize Bronk's result. The theorems of this paper form the basis of the theory of the convergence rate of normalized spectral functions of large symmetric random matrices, which has wide applications in applied research.
0 references