Abstract:Spectral clustering has been widely used for community detection in network sciences. While its empirical successes are well-documented, a clear theoretical understanding, particularly for sparse networks where degrees are much smaller than $\log n$, remains unclear. In this paper, we address this significant gap by demonstrating that spectral clustering offers exponentially small error rates when applied to sparse networks under Stochastic Block Models. Our analysis provides sharp characterizations of its performance, backed by matching upper and lower bounds possessing an identical exponent with the same leading constant. The key to our results is a novel truncated $\ell_2$ perturbation analysis for eigenvectors, coupled with a new analysis idea of eigenvectors truncation.

What problem does this paper attempt to address?

This paper attempts to address the issue of unclear theoretical performance analysis when using spectral clustering for community detection in sparse networks. Specifically, although spectral clustering performs well in practical applications, its performance has not been fully understood theoretically, especially in sparse networks where the degree is much smaller than $\log n$. The paper fills this gap by proving that spectral clustering can achieve an exponentially small error rate under the Stochastic Block Model (SBM). ### Main Issues 1. **Insufficient Theoretical Performance Analysis**: Existing research mainly focuses on dense networks, and there is a significant gap in the theoretical performance analysis of spectral clustering for sparse networks (i.e., networks with degrees much smaller than $\log n$). 2. **Exponential Error Bound**: Although spectral clustering performs excellently in practical applications, there is a lack of rigorous theoretical proof, particularly regarding whether it can achieve an exponentially small error rate. ### Main Contributions of the Paper 1. **Proof of Exponential Error Bound**: The paper proves that in sparse networks, spectral clustering can achieve an exponentially small error rate and provides matching upper and lower bounds with the same asymptotic exponent and leading constant. 2. **Precise Performance Characterization**: Through a new truncated $\ell_2$ perturbation analysis method and eigenvector truncation technique, the paper provides a precise characterization of the performance of spectral clustering, including the leading constant. 3. **New Analytical Tools**: The paper introduces a new truncated $\ell_2$ perturbation analysis method, which may also be valuable for fine spectral perturbation analysis of other binary random matrices. ### Methods and Techniques 1. **Truncated $\ell_2$ Perturbation Analysis**: By truncating the coordinates of the eigenvectors, the $\ell_\infty$ norm of the eigenvectors is controlled within a selected threshold $t_0$, thus overcoming the $\log n$ factor issue in classical concentration inequalities. 2. **Eigenvector Truncation Technique**: Through the eigenvector truncation technique, the paper effectively addresses the tail probability problem in high-dimensional data. 3. **Leave-One-Out Technique**: To decompose the tail probability of $(\tilde{A}_i - E A_i)(I - U^* U^{*T}) U$, the paper uses an improved leave-one-out technique to decouple the dependency between $\tilde{A}_i$ and $U$. ### Conclusion Through rigorous mathematical analysis, the paper proves that spectral clustering can achieve an exponentially small error rate in sparse networks and provides matching upper and lower bounds. These results not only fill the theoretical gap but also provide a solid theoretical foundation for the application of spectral clustering in sparse networks.

Fundamental Limits of Spectral Clustering in Stochastic Block Models

Randomized Spectral Clustering in Large-Scale Stochastic Block Models

Strong Consistency, Graph Laplacians, and the Stochastic Block Model

Strong Consistency of Spectral Clustering for Stochastic Block Models.

Stochastic Block Model and Community Detection in the Sparse Graphs: A spectral algorithm with optimal rate of recovery

Spectral Properties and Weak Detection in Stochastic Block Models

Spectral clustering on spherical coordinates under the degree-corrected stochastic blockmodel

Joint Spectral Clustering in Multilayer Degree-Corrected Stochastic Blockmodels

Strong Consistency of Spectral Clustering for the Sparse Degree-Corrected Hypergraph Stochastic Block Model

Analysis of spectral clustering algorithms for community detection: the general bipartite setting

Percolated stochastic block model via EM algorithm and belief propagation with non-backtracking spectra

Strong Consistency Guarantees for Clustering High-Dimensional Bipartite Graphs with the Spectral Method

Robust spectral clustering with rank statistics

Simultaneous Dimensionality and Complexity Model Selection for Spectral Graph Clustering

Partial Recovery Bounds for the Sparse Stochastic Block Model

Exact Recovery and Bregman Hard Clustering of Node-Attributed Stochastic Block Model

A spectral based goodness-of-fit test for stochastic block models

Computational and Statistical Thresholds in Multi-layer Stochastic Block Models

Robustness for Spectral Clustering of General Graphs under Local Differential Privacy

A Tighter Analysis of Spectral Clustering, and Beyond

Weak recovery, hypothesis testing, and mutual information in stochastic block models and planted factor graphs