Singularity degree of structured random matrices

From MaRDI portal





For random Hermitian matrices with a \(K\times K\) block variance structure, i.e., the variance of the entries depends only on which block they lie in, the authors determine the behavior of the self-consistent density of states at the origin. They prove that the density at the origin is determined by the zero structure of the variance matrix \(S=(s_{ij})_{i,j=1}^{K}\) and classify all possible singularities.\N\NThis classification is based on a novel normal form for \(S\) introduced by the authors and a relation this normal form induces on the indices of its block structure. Their main result states that if \(\rho\) is the Lebesgue density of the self-consistent density of states at the origin, then\N\[\N\lim\limits_{\tau\rightarrow 0}|\tau|^{\sigma}\rho(\tau)>0,\N\]\Nexists and is a finite positive number for some \(\sigma=\frac{k}{k+2}\), \(k=0,1,2,\dots\). The proofs follow by considering the power law scaling of the unique solution to the vector Dyson equation associated to the variance matrix \(S\), and the stability of a rescaled version of these equations based on this power law behavior.\N\NThe paper includes several examples and a conjecture for the condition number of such structured random matrices based on the singularity degree classified in the main result.











This page was built for publication: Singularity degree of structured random matrices

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6948495)