The two-star model: exact solution in the sparse regime and condensation transition

TL;DR

This paper provides an exact solution for the 2-star exponential random graph model in the sparse regime, revealing a condensation transition and limitations in generating highly heterogeneous degree distributions.

Contribution

It extends the understanding of the 2-star model to finite connectivity, identifying a phase transition and its implications for modeling real-world networks.

Findings

Identifies a condensation transition from fluid to condensate phase.

Shows the model's inability to produce graphs with degree heterogeneity beyond Erd"os-Rényi.

In the condensed phase, excess degree condenses on a single node with degree ~√N.

Abstract

The $2$-star model is the simplest exponential random graph model that displays complex behavior, such as degeneracy and phase transition. Despite its importance, this model has been solved only in the regime of dense connectivity. In this work we solve the model in the finite connectivity regime, far more prevalent in real world networks. We show that the model undergoes a condensation transition from a liquid to a condensate phase along the critical line corresponding, in the ensemble parameters space, to the Erd\"os-R\'enyi graphs. In the fluid phase the model can produce graphs with a narrow degree statistics, ranging from regular to Erd\"os-R\'enyi graphs, while in the condensed phase, the "excess" degree heterogeneity condenses on a single site with degree $\sim\sqrt{N}$. This shows the unsuitability of the two-star model, in its standard definition, to produce arbitrary finitely connected graphs with degree heterogeneity higher than Erd\"os-R\'enyi graphs and suggests that non-pathological variants of this model may be attained by softening the global constraint on the two-stars, while keeping the number of links hardly constrained.