[Paper Review] The cavity method for counting spanning subgraphs subject to local constraints

Justin Salez|arXiv (Cornell University)|Mar 16, 2011

Markov Chains and Monte Carlo Methods29 references5 citations

TL;DR

This paper rigorously validates the cavity method for counting spanning subgraphs under local constraints in asymptotically tree-like graphs by leveraging negative association and random weak limits. It establishes convergence of the free entropy density to a limit expressible via the unique solution of a limiting cavity equation, with explicit formulas derived for b-matchings in Erdős-Rényi random graphs with fixed average degree.

ABSTRACT

Using the theory of negative association for measures and the notion of random weak limits of sparse graphs, we establish the validity of the cavity method for counting spanning subgraphs subject to local constraints in asymptotically tree-like graphs. Specifically, the corresponding free entropy density is shown to converge along any sequence of graphs whose random weak limit is a tree, and the limit is directly expressed in terms of the unique solution to a limiting cavity equation. On a Galton-Watson tree, the latter simplifies into a recursive distributional equation which can be solved explicitely. As an illustration, we provide an explicit-limit formula for the $b-$matching number of an Erdős-Rényi random graph with fixed average degree and diverging size, for any $b\in\mathbb N$.

Motivation & Objective

• To establish the validity of the cavity method for counting spanning subgraphs subject to local constraints in asymptotically tree-like graphs.
• To resolve the convergence and correctness questions of the cavity method under general conditions using negative association and random weak limits.
• To derive an explicit limit formula for the b-matching number in Erdős-Rényi random graphs with fixed average degree as the graph size diverges.
• To show that the free entropy density converges along any sequence of graphs whose random weak limit is a tree, with the limit determined by the unique solution to a limiting cavity equation.

Proposed method

• Uses the theory of negative association for measures, particularly the cavity-monotone property, to ensure convergence of the cavity equation.
• Applies the framework of local weak convergence and unimodularity to analyze the infinite-volume limit of graph sequences.
• Reduces the cavity equation to a recursive distributional equation on Galton-Watson trees, enabling explicit solution via distributional self-similarity.
• Employs the Gibbs-Boltzmann measure and generating polynomial to relate the free energy and energy to the cavity method's predictions.
• Uses recursive equations involving random variables and expectations to characterize the limiting behavior of the b-matching number.
• Applies fixed-point analysis and monotonicity arguments to prove convergence of iterative cavity updates and to identify historical minima of the entropy function.

Experimental results

Research questions

• RQ1 Does the cavity method converge to a unique fixed point for counting spanning subgraphs under local constraints in sparse, locally tree-like graphs?
• RQ2 Is the solution to the cavity equation asymptotically correct, i.e., does it accurately describe the free entropy density in the infinite-volume limit?
• RQ3 Can the limiting free entropy density be explicitly computed for specific models such as b-matchings in Erdős-Rényi random graphs?
• RQ4 What conditions ensure the validity of the cavity method beyond heuristic applications, particularly in the presence of cycles?
• RQ5 How does the b-matching number behave in the infinite-volume limit for random graphs with fixed average degree?

Key findings

• The free entropy density converges along any sequence of graphs whose random weak limit is a tree, with the limit given by the unique solution to the limiting cavity equation.
• For Galton-Watson trees, the cavity equation reduces to a recursive distributional equation that can be solved explicitly, enabling analytical computation of the limiting free energy.
• An explicit formula is derived for the b-matching number in Erdős-Rényi random graphs with fixed average degree c, showing that the normalized matching number converges almost surely to a closed-form expression involving the solution to a transcendental equation.
• The limiting b-matching number is given by the formula: 1 - (tc + e^{-c tc} + c tc e^{-c tc}) / 2, where tc is the smallest root of t = e^{-c e^{-c t}}.
• The method establishes the validity of the cavity method under the cavity-monotone property, which generalizes ultra-log-concavity and ensures convergence and correctness.
• The analysis confirms that the cavity method correctly predicts the asymptotic behavior of the energy and partition function in the infinite-volume limit for unimodular Galton-Watson trees.