Approximate Spielman-Teng theorems for the least singular value of random combinatorial matrices

Vishesh Jain

DOI: https://doi.org/10.48550/arXiv.1904.10592

2019-04-24

Abstract:An approximate Spielman-Teng theorem for the least singular value $s_n(M_n)$ of a random $n\times n$ square matrix $M_n$ is a statement of the following form: there exist constants $C,c >0$ such that for all $\eta \geq 0$, $\Pr(s_n(M_n) \leq \eta) \lesssim n^{C}\eta + \exp(-n^{c})$. The goal of this paper is to develop a simple and novel framework for proving such results for discrete random matrices. As an application, we prove an approximate Spielman-Teng theorem for $\{0,1\}$-valued matrices, each of whose rows is an independent vector with exactly $n/2$ zero components. This improves on previous work of Nguyen and Vu, and is the first such result in a `truly combinatorial' setting.

Probability,Combinatorics

What problem does this paper attempt to address?

The problem that this paper attempts to solve is about the probability estimation of the minimum singular value of randomly composed matrices. Specifically, the author aims to develop a simple and novel framework for proving the approximate Spielman - Teng theorem for discrete random matrices. Such theorems are usually stated as follows: There exist constants \(C, c>0\), such that for all \(\eta \geq 0\),

\[ \Pr(s_n(M_n) \leq \eta) \lesssim n^C \eta+\exp(-n^c) \]

where \(s_n(M_n)\) is the minimum singular value of the \(n\times n\) random matrix \(M_n\).

### Background and Motivation

1. **Importance of Singular Values**
- The maximum and minimum singular values play a crucial role in multiple mathematical fields such as applied linear algebra, computer science, statistics, and asymptotic geometric analysis.
- They are also often key components in the proofs of other results in random matrix theory, such as the circular law (the counterpart of the classical semicircular law for non - Hermitian matrices) and the delocalization property of eigenvectors.

2. **Existing Research**
- The behavior of the maximum singular value of random matrices with independent entries is relatively well - understood.
- However, the study of the behavior of the minimum singular value is more difficult. For example, for random matrices with standard Gaussian entries, Edelman proved that:

\[ \Pr(s_n(M_n) \leq \epsilon n^{-1/2}) \sim \epsilon \]

- Spielman and Teng conjectured that the above result also holds for random Rademacher matrices (each entry independently takes ±1 with equal probability), but an additive error of \(cn\) needs to be added to account for the probability of the matrix being singular.

3. **Existing Results**
- Rudelson and Vershynin et al. have made significant progress and proved similar results.
- However, less is known about the behavior of the minimum singular value for random matrix models with non - trivial dependencies.

### Main Contributions of the Paper

1. **Proposing a New Framework**
- The author proposes a new framework, using the recent approach to counting problems in the inverse Littlewood - Offord theory, to simplify and uniformly prove the approximate Spielman - Teng theorem in the discrete setting.

2. **Specific Results**
- Proved an approximate Spielman - Teng theorem for {0, 1}-valued matrices, where each row is an independent vector with exactly \(n/2\) zero components.
- This is an improvement on the previous work of Nguyen and Vu and is also the first result obtained in a "truly combinatorial" setting.

### Conclusion

This paper simplifies the probability estimation of the minimum singular value of discrete random matrices by proposing a new framework and has made important progress in a specific combinatorial setting. This work not only promotes the development of random matrix theory but also provides new tools and methods for applications in related fields.