[论文解读] Statistical Mechanics of Random Hyperbolic Graphs within the Fermionic Maximum-Entropy Framework

M. Ángeles Serrano|arXiv (Cornell University)|Mar 18, 2026

Complex Network Analysis Techniques被引用 0

一句话总结

该论文将双曲线随机图模型置于最大熵（MaxEnt）框架中进行上下文化与整合，揭示费米（排他）统计学如何支撑链接形成，并为在双曲空间中建立复杂网络的无偏见建模提供了一个 principled 的方法。

ABSTRACT

The intricate relations between elements in natural and human-made systems sustain the complex processes that shape our world, forming multiscale networks of interactions. These networks can be represented as graphs composed of nodes connected by links and, regardless of their domain, they share a set of fundamental structural properties. The family of network models in hyperbolic space constitutes one of the most advanced frameworks accounting for such properties, including sparsity, the small-world property, heterogeneity and hierarchical organization, high clustering, and scale invariance under network renormalization transformations. These geometric models also exhibit other intriguing phenomena, such as an anomalous, temperature-dependent phase transition between a geometric and a non-geometric phase. In simple graph representations, where network links are unweighted, the model can be derived within a statistical-mechanics framework by maximizing the Gibbs entropy of the graph ensemble subject to constraints imposed by observations, with links effectively behaving as fermionic particles. In this topical review, I revisit these derivations previously scattered across different sources and complement them, in order to properly contextualize and consolidate hyperbolic random graphs within the broad framework of the maximum-entropy principle in the statistical mechanics of complex networks. The approach presented here represents the least-biased prediction of the fundamental set of core network properties and establishes a principled framework for analyzing network structure, offering new perspectives and powerful analytical tools for both theoretical and empirical studies.

研究动机与目标

• Motivate the use of geometric (hyperbolic) representations to capture sparsity, small-world behavior, heterogeneity, clustering, and scale invariance in complex networks.
• Revisit and consolidate the derivation of hyperbolic random graph models within the maximum-entropy (MaxEnt) statistical mechanics framework.
• Demonstrate how simple, unweighted networks can be described with fermionic-like (exclusion) statistics leading to Fermi-Dirac-type probabilities for link formation.
• Provide a principled, least-biased analytical framework to analyze core network properties using MaxEnt principles in geometric spaces.

提出的方法

• Represent networks as ensembles constrained by observable graph properties within the Exponential Random Graph (ERG) framework.
• Derive the Hamiltonian H(G) as a linear combination of observables with Lagrange multipliers, leading to P(G) = e^{-H(G)}/Z.
• Show that for simple unweighted graphs, link formation probabilities exhibit fermionic-like occupancy and connect this to hyperbolic geometric embedding.
• Introduce homogeneous and heterogeneous node-attribute scenarios in hyperbolic geometry, with energy E(G) = sum_{i<j} ε_{ij} a_{ij} and ε_{ij} = f(x_{ij}).
• Relate the CM and SCM (soft configuration model) to a Fermi-Dirac-like occupancy, highlighting ensemble equivalence considerations and the role of hidden degrees.
• Discuss the hypersoft configuration model (HSCM) as a hypercanonical mixture, extending MaxEnt perspectives to ensembles with varying expected degrees.

实验结果

研究问题

• RQ1 How can hyperbolic geometry be used within a MaxEnt framework to describe sparse, small-world networks with heterogeneous degree distributions?
• RQ2 What is the role of fermionic (exclusion) statistics in the probability of link formation in simple unweighted graphs modeled in hyperbolic space?
• RQ3 How do various configuration models (ER, CM, SCM, HSCM) compare within the MaxEnt framework in terms of clustering, degree correlations, and ensemble equivalence?
• RQ4 Can a unified MaxEnt approach reproduce core network properties such as sparsity, clustering, and scale invariance observed in real networks when embedded in hyperbolic geometry?
• RQ5 What are the implications of geometric max-entropy constructions for analyzing real network structure and evolution?

主要发现

• Fermionic-like probabilities emerge for link formation in simple unweighted graphs when embedding in hyperbolic space under MaxEnt constraints.
• Maximum-entropy methods yield the ERG family, unifying classical random graphs, configuration models, and hyperbolic network models under a single principled framework.
• Soft and hypersoft configuration models relate to Fermi-Dirac-like occupancy and reveal potential ensemble non-equivalence in the thermodynamic limit, especially with heterogeneous degree sequences.
• Hyperbolic geometric ensembles naturally support sparsity, small-world behavior, and the possibility of scale invariance under renormalization, aligning with observed network properties.
• The framework provides a least-biased, information-theoretic approach to inferring and analyzing core network properties from observed data within a geometric setting.