Popularity-Driven Networking

We investigate the growth of connectivity in a network. In our model, starting with a set of disjoint nodes, links are added sequentially. Each link connects two nodes, and the connection rate governing this random process is proportional to the degrees of the two nodes. Interestingly, this network exhibits two abrupt transitions, both occurring at finite times. The first is a percolation transition in which a giant component, containing a finite fraction of all nodes, is born. The second is a condensation transition in which the entire system condenses into a single, fully connected, component. We derive the size distribution of connected components as well as the degree distribution, which is purely exponential throughout the evolution. Furthermore, we present a criterion for the emergence of sudden condensation for general homogeneous connection rates.

💡 Research Summary

The paper studies the evolution of a fixed‑size random graph in which links are added one by one, and the rate at which two nodes of degrees i and j become connected is proportional to (i + 1)(j + 1). This “popularity‑driven” rule implements a rich‑get‑richer mechanism directly on the existing nodes, rather than on newly arriving nodes as in the classic preferential‑attachment model.

First, the authors derive the degree dynamics. Let n_j(t) be the fraction of nodes of degree j at time t and h(t)=∑_j j n_j(t) the average degree. The total connection rate of degree‑j nodes is ν_j=∑i C{i,j} n_i, which for the chosen rate reduces to ν_j=(j+1)(1−t)^{−1}−1. The average degree obeys dh/dt=(h+1)^2/2, giving h(t)=t/(1−t) for t<1. Hence the average degree diverges at a finite time t_c=1, signalling a continuous condensation of degrees.

The degree distribution satisfies (1−t) dn_j/dt=j n_{j−1}−(j+1) n_j and, with the initial condition n_j(0)=δ_{j,0}, yields the exact solution n_j(t)=(1−t) t^j. Thus the degree distribution remains a pure exponential (geometric) throughout the evolution and collapses to zero at t_c.

Next, the authors turn to the size distribution of connected components (clusters). A cluster of size k contains k−1 links (almost always a tree in the infinite‑size limit). The aggregation rate for two finite clusters of sizes ℓ and m follows from the degree‑based rule and is K_{ℓ,m}=(3ℓ−2)(3m−2). The master equation for the cluster density c_k(t) is a binary aggregation equation with a gain term proportional to K_{ℓ,m}c_ℓc_m and a loss term proportional to c_k∑m K{k,m}c_m.

Moments M_n=∑_k k^n c_k are introduced. The second moment obeys dM_2/dt=(3M_2−2)^2, giving M_2(t)=1−2t/(1−3t) for t<1/3. Its divergence at t_g=1/3 marks the percolation transition: a macroscopic (giant) component appears and the finite‑size clusters no longer contain the whole mass.

The mass of finite clusters is M_1(t)=∑_k k c_k. For t<t_g, M_1=1; for t_g<t<t_c, M_1 decreases according to a duality relation derived from the cubic equation τ(1−τ)^2=t(1−t)^2, where τ<t_g is a non‑trivial root. The giant component’s mass is g(t)=1−M_1(t). Near the percolation point, g(t) grows linearly, g≈3(t−t_g). As t→t_c^−, the giant component absorbs the remaining finite clusters, and at t_c=1 the whole network condenses into a single fully connected component. At this condensation time both the degree distribution n_j and the cluster densities c_k vanish simultaneously.

The exact cluster‑size distribution is obtained in closed form: c_k(t)=A_k t^{k−1}(1−t)^{2k−1} with coefficients A_k=(3k−3)!/(k!(2k−1)!). The generating function B(x)=∑_k (3k−2)A_k x^k satisfies B(1−B)^2=x, leading to the asymptotic tail c_k∼k^{−5/2} e^{−k/k*}. At the percolation point the tail is a pure power law c_k∼k^{−5/2}.

The paper also generalizes the connection rule to C_{i,j}∝(i j)^α. Using scaling arguments, the average degree obeys dh/dt∼h^{2α}. For α<½, h grows algebraically, the percolation time is finite, but condensation occurs only at infinite time. At α=½, h grows exponentially. For ½<α≤1 both percolation and condensation happen at finite times, with h diverging as t→t_c. For α>1 the scaling breaks down and condensation becomes instantaneous (t_c=t_g=0), analogous to instantaneous gelation in aggregation theory.

Comparing with the classic Erdős–Rényi model (C_{i,j}=1), which has a finite percolation time (t_g=1/9) but an infinite condensation time, the popularity‑driven model exhibits two finite‑time abrupt transitions. The rich‑get‑richer mechanism accelerates the absorption of finite clusters by the giant component, leading to a rapid, system‑wide condensation.

In conclusion, the authors provide a complete analytical description of a network whose link formation is governed by node popularity. They derive exact expressions for degree and component‑size distributions, identify the percolation and condensation thresholds (t_g=1/3, t_c=1), and establish a general criterion for sudden condensation in homogeneous connection‑rate models. The results illuminate how popularity‑based attachment can produce abrupt structural transitions in social, communication, and other complex networks.