Eigenvalue distribution of large dilute random matrices (Q4345174)
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 1036960
| Language | Label | Description | Also known as |
|---|---|---|---|
| default for all languages | No label defined |
||
| English | Eigenvalue distribution of large dilute random matrices |
scientific article; zbMATH DE number 1036960 |
Statements
Eigenvalue distribution of large dilute random matrices (English)
0 references
17 July 1997
0 references
eigenvalue distributions
0 references
diluted random Gram matrices
0 references
Wigner semicircle distribution law
0 references
spectral function
0 references
Lindeberg condition
0 references
Marchenko and Pastur distribution law
0 references
0.8860538601875305
0 references
0.8662224411964417
0 references
0.8564527034759521
0 references
Limit spectra of random \(N\)-dimensional matrices \(H\) are investigated whose entries \(H_{ij}\) are normed (by \(N^{-1})\) sums of products \(\xi_{\mu i} \xi_{\mu j} a_{ij}\) over \(\mu=1, \dots, m\) as \(N\to\infty\) and \(m\to\infty\) so that \(m/N\to c>0\). It is assumed that the matrices \(\{a_{ij}\}\) are symmetric with zero diagonal elements, and all values \(\xi_{\mu i}\) and \(a_{ij}\), \(i<j\), are jointly independent. All \(\xi_{\mu i}\) have zero expectation values and one and the same variance \(v^2\), while \(a_{ij}\) take one and the same decreasing value with a vanishing probability and, otherwise, have zero value (the matrices \(H\) are called ``diluted'' random Gram matrices). A spectral function \(\sigma (\lambda)\) is considered that is a portion of eigenvalues of the matrix \(H\) that do not exceed \(\lambda\). Its limit behavior is studied under a generalized version of the joint Lindeberg condition. The following two statements are proved. If each of \(a_{ij}\) is not zero and equals to the (common) value \(N^{-\nu}/ \sqrt a\) with the (common) probability \(aN^{-2(1-\nu)}\), where \(a>0\), \(1/2\leq \nu<1\), where either \(a\) is fixed or \(a<N\) and \(\nu=1/2\), then \(\sigma(\lambda)\) is weakly convergent to the Wigner ``semicircle distribution law'' depending on a single parameter \(u= \sqrt cv^2\) with the support \([-2u,2u]\). If each of \(a_{ij}\) is nor zero and equals to the (common) value \(N^{-\alpha}/ \sqrt a\) with the (common) probability \(aN^{2\alpha-1}/ \sqrt a\), \(a>0\), \(0\leq\alpha \leq 1/2\), where either \(a\) is fixed or \(a<N\) and \(\alpha=0\), then \(\sigma(\lambda)\) is weaky convergent to the Marchenko and Pastur distribution law depending on two parameters \(c\) and \(v\) with a bounded support. The proofs are based on analysis of moments and a special diagram technique.
0 references