[Paper Review] Growing Small-World Networks Generated by Attaching to Edges

Zhongzhi Zhang, Lili Rong|arXiv (Cornell University)|Mar 28, 2005

Complex Network Analysis Techniques3 citations

TL;DR

This paper proposes a minimal growing small-world network model where new vertices attach to the endpoints of randomly selected edges, forming a sparse, planar graph. The model exhibits exponential degree distribution, a high clustering coefficient of approximately 0.6479, and an average path length growing logarithmically or slower with network size, confirming small-world properties through analytical derivation and numerical validation.

ABSTRACT

We introduce a minimal model of small-world growing network generated by attaching to edges. The produced network is a plane graph which exists in real-life world. We obtain the analytic results of degree distribution decaying exponentially with degree and average clustering coefficient $C={3/2}ln3-1\approx 0.6479$, which are in good agreement with the numerical simulations. We also prove that the increasing tendency of average path length of the considered network is a little slower than the logarithm of the network order $N$.

Motivation & Objective

• To develop a simple, minimal model for generating growing small-world networks using edge attachment.
• To analyze the geometric and topological properties of such networks, particularly degree distribution, clustering coefficient, and average path length.
• To demonstrate that the model produces realistic small-world characteristics despite its simplicity and planar structure.
• To provide analytical solutions for key network metrics and validate them with numerical simulations.

Proposed method

• The network grows by adding one vertex at each time step, connecting it to both endpoints of a randomly selected, previously unused edge.
• The model maintains sparsity, with average degree approaching 4 as network size increases.
• Master equation (1) governs the time evolution of vertex degree distribution, leading to an exponential decay with degree.
• The clustering coefficient is derived analytically using the sum over degree-dependent contributions, yielding $ C = \frac{3}{2}\ln 3 - 1 \approx 0.6479 $.
• Average path length is analyzed via recursive total distance equations, leading to $ \sigma(N) \sim N^2 \ln N $, implying $ L(N) \sim \ln N $.
• The analysis assumes that shortest paths do not pass through vertices born after the maximum birth time of the pair, enabling recursive distance computation.

Experimental results

Research questions

• RQ1 Does a growing network model based solely on edge attachment produce small-world properties such as high clustering and short average path length?
• RQ2 What is the analytical form of the degree distribution in such a network model?
• RQ3 How does the clustering coefficient behave as a function of network size, and can it be computed exactly?
• RQ4 How does the average path length scale with network order, and does it grow logarithmically or slower?
• RQ5 Can the model generate a planar graph that reflects real-world network constraints while preserving small-world characteristics?

Key findings

• The degree distribution decays exponentially with degree, following $ P(k) \propto \left(\frac{2}{3}\right)^k $.
• The clustering coefficient converges to $ \frac{3}{2}\ln 3 - 1 \approx 0.6479 $, which is higher than that of regular lattices with the same average degree.
• The average path length grows at most logarithmically with network size $ N $, consistent with small-world behavior.
• The total distance function $ \sigma(N) $ scales as $ N^2 \ln N $, confirming the logarithmic scaling of average path length.
• The model produces a sparse, planar graph that is realizable in physical systems such as printed circuits and biological vein networks.
• Analytical results for degree distribution, clustering, and path length are in strong agreement with numerical simulations.