Spectral radii of sparse non-Hermitian random matrices

Hyungwon Han
2024-12-02
Abstract:We provide estimates for the spectral radii of an $n\times n$ sparse non-Hermitian random matrix $Z$ with general entries in the regime $p=d/n$ where $0<d<1$ is fixed. Utilizing the structural results of (Łuczak, '90), we show that the spectral radius $\rho (Z)$ is $0$ with probability converging to some nonzero value, and satisfies the inequality $(\phi (n))^{-1}\leq \rho (Z)\leq \phi (n)$ in the asymptotic sense for any function $\phi$ satisfying $\lim_{n\to\infty}\phi (n)=\infty$ with the remaining probability.
Probability,Combinatorics
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to estimate the spectral radius \( \rho(Z) \) of a sparse non - Hermitian random matrix \( Z \), especially in the case where the average degree is a constant \( d/n \), where \( 0 < d < 1 \) is fixed. Specifically, the paper focuses on the behavior of the spectral radius of sparse non - Hermitian random matrices when the connection probability \( p = d/n \). ### Main problems and goals 1. **Estimation of spectral radius**: - The paper provides an estimate of the spectral radius \( \rho(Z) \) of an \( n\times n \) sparse non - Hermitian random matrix \( Z \) in the case of \( p = d/n \). - Using the structural results of Erdős - Rényi directed graphs, it is proved that the spectral radius \( \rho(Z) \) is 0 with certain probabilities, and satisfies the inequality \( (\varphi(n))^{-1}\leq\rho(Z)\leq\varphi(n) \) with other probabilities, where \( \varphi(n) \) is an arbitrarily slowly growing function, that is, \( \lim_{n\rightarrow\infty}\varphi(n)=\infty \). 2. **Phase - transition phenomenon**: - The paper explores the phase - transition characteristics of Erdős - Rényi directed graphs \( G_d(n, d/n) \) when \( d < 1 \), that is, all non - trivial strongly connected components are directed cycles. - This phase - transition characteristic plays a crucial role in the estimation of the spectral radius, especially in the case of \( d < 1 \), where the structure of the graph is relatively simple and the spectral radius can be estimated more precisely. 3. **Combination of upper and lower bounds**: - By combining the results of the upper and lower bounds, the paper obtains the main conclusion about the spectral radius: when \( p = d/n \), the probability that \( \rho(Z) = 0 \) converges to \( (1 - d)e^d + d^2/2 \), and in the remaining probability, \( \rho(Z)=\Theta_p(1) \). ### Formula summary - **Definition of spectral radius**: \[ \rho(A)=\lim_{k\rightarrow\infty}\|A^k\|^{1/k} \] where \( \|\cdot\| \) represents the matrix norm. - **Asymptotic notation**: - \( X_n = O_p(f(n)) \) means that for any function \( \varphi \) satisfying \( \lim_{n\rightarrow\infty}\varphi(n)=\infty \), the inequality \( X_n\leq\varphi(n)f(n) \) holds with high probability. - \( X_n=\Omega_p(f(n)) \) means that for any function \( \varphi \) satisfying \( \lim_{n\rightarrow\infty}\varphi(n)=\infty \), the inequality \( X_n\geq f(n)/\varphi(n) \) holds with high probability. - \( X_n = \Theta_p(f(n)) \) means that both \( X_n = O_p(f(n)) \) and \( X_n=\Omega_p(f(n)) \) hold. - **Main theorems**: - Theorem 1.2: Assume that \( p = d/n \), where \( 0 < d < 1 \) is fixed, and \( Z \) satisfies the above conditions, then \( \rho(Z)=O_p(1) \). - Theorem 1.3: Assume that \( p = d/n \), where \( 0 < d < 1 \) is fixed, and \( G_d(n, p) \) contains cycles, then \( \rho(Z) \)