Eigenvalue gaps of random perturbations of large matrices
The paper studies eigenvalue gap statistics for random matrices of the form \(M_n = F_n + N_n\), where \(F_n\) is a deterministic symmetric matrix and \(N_n\) is a symmetric random matrix with independent, centred sub-Gaussian entries. The main focus is on obtaining tail bounds for the smallest eigenvalue gap \(\delta_{\min}\) in regimes where the operator norm of \(F_n\) is allowed to be very large.\N\NBuilding on earlier work of \textit{H. Nguyen} et al. [Probab. Theory Relat. Fields 167, No. 3--4, 777--816 (2017; Zbl 1391.15111)], which established polynomial tail bounds under the assumption \(\|F_n\| \le n^C\), the authors develop new techniques that allow \(\|F_n\|\) to grow as fast as \(\exp(n^{1/16})\). Their main result provides exponentially small upper bounds for the probability that \(\delta_{\min}\) is small. As a consequence, they show that \(M_n\) has simple spectrum with probability at least \(1 - \exp(-n^2/16)\), significantly improving previous polynomial bounds.\N\NThe proof combines the classical reduction of eigenvalue gap problems to anti-concentration estimates (via interlacing and eigenvector analysis) with recent combinatorial advances due to Jain. In particular, the authors employ lattice approximation techniques to control the arithmetic structure of approximate eigenvectors, thereby avoiding limitations of inverse Littlewood-Offord theory and least common denominator arguments, which typically yield only polynomial probability bounds or require restrictive norm conditions.\N\NA key intermediate step is a structural result showing that approximate null-vectors of \(M_n\) are ``unstructured in a suitable sense, quantified via a regularised Lévy concentration function. This is combined with a refined counting argument over integer lattice approximations and small-ball probability estimates to obtain the desired tail bounds.\N\NThe results contribute to the broader program of understanding eigenvalue repulsion and simplicity of spectrum in random matrix theory, particularly in settings with large deterministic perturbations. The methods are of independent interest and may have further applications to related problems in random matrices and high-dimensional probability.
- Approximate Spielman-Teng theorems for the least singular value of random combinatorial matrices
- Eigenvectors and controllability of non-Hermitian random matrices and directed graphs
- Exact Wigner surmise type evaluation of the spacing distribution in the bulk of the scaled random matrix ensembles
- Extreme gaps between eigenvalues of random matrices
- High-dimensional probability. An introduction with applications in data science
- scientific article; zbMATH DE number 3069216 (Why is no real title available?)
- scientific article; zbMATH DE number 3099315 (Why is no real title available?)
- Inverse Littlewood-Offord theorems and the condition number of random discrete matrices
- Invertibility of symmetric random matrices
- On the counting problem in inverse Littlewood-Offord theory
- Quantitative invertibility of random matrices: a combinatorial perspective
- Random matrices have simple spectrum
- Random matrices: tail bounds for gaps between eigenvalues
- RANDOM MATRICES: THE CIRCULAR LAW
- Random matrices: universality of local eigenvalue statistics
- Resilience of the rank of random matrices
- Singularity of random symmetric matrices -- a combinatorial approach to improved bounds
- Smooth analysis of the condition number and the least singular value
- Some new results in random matrices over finite fields
- The asymptotic distribution of a single eigenvalue gap of a Wigner matrix
- The generalised product moment distribution in samples from a normal multivariate population.
- The Littlewood-Offord problem and invertibility of random matrices
- The strong circular law: A combinatorial view
- Universality of random matrices and local relaxation flow
- Wegner Estimate and Level Repulsion for Wigner Random Matrices
This page was built for publication: Eigenvalue gaps of random perturbations of large matrices
Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q6894051)
