Cluster expansion of the log-likelihood ratio: Optimal detection of planted matchings

To understand how hidden information can be extracted from statistical networks, planted models in random graphs have been the focus of intensive study in recent years. In this work, we consider the detection of a planted matching, i.e., an independent edge set, hidden in an Erdős-Rényi random graph, which is formulated as a hypothesis testing problem. We identify the critical regime for this testing problem and prove that the log-likelihood ratio is asymptotically normal. Via analyses of computationally efficient edge or wedge count test statistics that attain the optimal limits of detection, our results also reveal the absence of a statistical-to-computational gap. Our main technical tool is the cluster expansion from statistical physics, which allows us to prove a precise, non-asymptotic characterization of the log-likelihood ratio. Our analyses rely on a careful reorganization and cancellation of terms that occur in the difference between monomer-dimer log partition functions on the complete and Erdős-Rényi graphs. This combinatorial and statistical physics approach represents a significant departure from the more established methods such as orthogonal decompositions, and positions the cluster expansion as a viable technique in the study of log-likelihood ratios for planted models in general.

💡 Research Summary

This paper addresses a fundamental problem in the field of random graph theory and information retrieval: the detection of a hidden “planted matching” within an Erdős-Rényi random graph. The researchers frame this as a hypothesis testing problem, comparing a null model $G(n,q)$ with a planted model $G(n,p;M)$, where $M$ is an independent edge set of size $\Theta(n)$. The primary objective is to identify the critical threshold for detection and to analyze the behavior of the log-likelihood ratio (LLR) in this regime.

The study’s most significant contribution lies in its methodological innovation. Moving away from traditional techniques like orthogonal decompositions, the authors employ “cluster expansion,” a sophisticated tool derived from statistical physics. By treating the problem through the lens of the monomer-dimer model’s partition function, they decompose the log-likelihood ratio into a sum of “clusters”—connected multi-graphs. Through a rigorous analysis of Ursell functions $\phi(H)$, which quantify the interactions within these clusters, the authors prove that the log-likelihood ratio is asymptotically normally distributed.

A crucial finding of this research is the identification of the critical regime at $p = \Theta(1/\sqrt{n})$. The authors demonstrate that within this regime, even when the edge density $q$ is significantly higher than the connectivity threshold $\log n/n$, the presence of the planted matching remains statistically distinguishable. Furthermore, the paper provides a profound insight into the relationship between statistical limits and computational complexity. By analyzing computationally efficient statistics, such as edge and wedge counts, the authors prove the absence of a “statistical-to-computational gap.” This implies that the theoretical limit of detection can be achieved using simple, polynomial-time algorithms, making the detection of hidden structures not just theoretically possible but practically feasible.

In summary, this work bridges the gap between statistical physics and combinatorial graph theory. The successful application of cluster expansion to characterize the LLR provides a new, powerful framework for studying a wide range of planted models, potentially paving the way for more advanced algorithms in detecting hidden information in large-scale complex networks.