Nicholas Christoffersen,Kyle Luh,Hoi H. Nguyen,Jingheng Wang
Abstract:We show that, with very high probability, the random graph Laplacian has simple spectrum. Our method provides a quantitatively effective estimate of the spectral gaps. Along the way, we establish results on affine no-gaps delocalization, no-structure delocalization, overcrowding and small entries of the eigenvectors for the Laplacian model. These findings are of independent interest.
What problem does this paper attempt to address?
The main problem that this paper attempts to solve is the eigenvalue gap problem of the random graph Laplacian matrix. Specifically, the authors are concerned with whether the Laplacian matrix of a random graph has a simple spectrum (i.e., all eigenvalues are single - multiplicity), and they provide a quantitative estimate of the eigenvalue gap. These problems are of great significance in theoretical computer science, graph theory, and random matrix theory.
### Background and Motivation of the Paper
1. **Random Graph Model**:
- The research object is the Erdős - Rényi random graph \(G(n, p)\), where \(n\) is the number of vertices and \(p\) is the probability of an edge.
- The Laplacian matrix \(L_n = D_n - A_n\), where \(D_n\) is the degree matrix and \(A_n\) is the adjacency matrix.
2. **Existing Results**:
- Previous research has shown that for the adjacency matrix \(A_n\) of a random graph, its eigenvalue sequence \(\lambda_1(A_n)\leq\cdots\leq\lambda_n(A_n)\) is almost surely simple [TV17, NTV17].
- These results have been extended to sparse graphs [LV20, LL21] and Wigner matrices [CJMS24].
3. **Research Motivation**:
- From the perspective of random matrices, the size of the eigenvalue gap was one of the original motivations for introducing random matrices in physics and is a standard research topic in this field [Wig55, BS10, AGZ10, EYY12, Tao13, EKYY12].
- The minimum eigenvalue gap has also received extensive attention in Wigner matrices [Vin01, BAB13, FTW19, Bou22].
### Main Problems and Objectives
The main objective of the paper is to prove that the eigenvalues of the random graph Laplacian matrix are almost surely simple and to provide a quantitative estimate of the eigenvalue gap. Specifically:
- **Theorem 1.2**: Let \(0 < p < 1\) be fixed and \(A\) be an arbitrary constant. For the random Erdős - Rényi graph \(G_n = G(n, p)\) with parameter \(p\), the eigenvalues of its Laplacian matrix \(L_n\) satisfy:
\[
\max_{1\leq i\leq n - 1}P(|\lambda_{i + 1}(L_n)-\lambda_i(L_n)|\leq\delta n^{- 1/2})=O(n^{o(1)}\delta)
\]
where the implicit constant depends on \(p\) and \(A\).
- **Corollary**: This shows that the spectrum of the Laplacian matrix is almost surely simple and the minimum eigenvalue gap is at least \(n^{-3/2 - o(1)}\).
### Applications and Significance
These results provide a theoretical basis for many applications, including but not limited to:
- **Graph Partitioning and Clustering**: The eigenvalue gap affects the partitioning stability of the graph [AEGL21].
- **Quantum Random Walk**: The minimum eigenvalue gap affects the mixing time of certain quantum random walk models [AAKV01, Kem03, CLR20b].
- **Graph Neural Networks**: The eigenvalue gap affects the convergence speed of graph neural networks [WYZL22, ZL20, MWW23, GBR20].
Through these studies, the authors not only solve important theoretical problems but also provide a solid theoretical foundation for algorithm design in practical applications.