[Paper Review] A steady state network model with a 1/k scale-free degree distribution

Simon Laird, Henrik Jeldtoft Jensen|arXiv (Cornell University)|Mar 8, 2006

Complex Network Analysis Techniques6 citations

TL;DR

This paper proposes a steady-state network model that achieves a 1/k scale-free degree distribution through node duplication and deletion without network growth. By balancing preferential attachment with preferential detachment, the model produces a power-law degree distribution with exponent γ ≈ 1 in the limit of low connectance, verified via mean-field analysis and simulations, demonstrating that scale-free structure can emerge without growth or traditional preferential attachment.

ABSTRACT

Using a steady state process of node duplication and deletion we produce networks with 1/k scale-free degree distributions in the limit of vanishing connectance. This occurs even though there is no growth involved and inherent preferential attachment is counterbalanced by preferential detachment. The mean field evolution is considered and the 1/k law is verified under certain approximations. An ansatz for the degree distribution is proposed on the basis of symmetry considerations and is shown to coincide well with the simulation data. Distributional forms other than power law are also shown to arise when the condition of perfect duplication is relaxed.

Motivation & Objective

• To develop a network model that produces scale-free degree distributions without network growth.
• To investigate whether preferential attachment and detachment can coexist to yield a 1/k power-law distribution.
• To explore how duplication fidelity affects degree distribution forms, including exponential and power-law behavior.
• To verify the emergence of γ = 1 exponent via mean-field approximation and simulation.
• To relate the model’s findings to real-world systems like ecological and biological networks with low-exponent power laws.

Proposed method

• The model uses a fixed network size N with stochastic node deletion and duplication of a randomly selected parent node.
• After deleting a node, a daughter node is created by copying all edges from the parent with probability Pe, and connecting to non-neighbors with probability Pn, and to the parent with probability Pp.
• Connectance C₀ is maintained as a control parameter by setting Pp = C₀, allowing Pe to vary as a fidelity parameter from random (Pe = C₀) to perfect (Pe = 1, Pn = 0).
• An ansatz for the degree distribution is derived from symmetry considerations and tested against simulation data.
• Mean-field equations are derived for degree evolution, with transitions modeled via hypergeometric-like probabilities and edge-configuration sampling.
• Edge-degree correlations are estimated using an urn-model approximation to compute the probability of selecting edges connected to nodes of a given degree.

Experimental results

Research questions

• RQ1 Can a 1/k scale-free degree distribution emerge in a network model without node growth?
• RQ2 What role does the balance between preferential attachment and preferential detachment play in shaping the degree distribution?
• RQ3 How does duplication fidelity affect the functional form of the degree distribution?
• RQ4 Does the mean-field solution support the observed exponent γ ≈ 1 in the low-connectance limit?
• RQ5 Can this model explain the emergence of exponential or power-law distributions in ecological and biological networks?

Key findings

• The model produces a 1/k power-law degree distribution in the limit of vanishing connectance (C₀ → 0), with the exponent γ ≈ 1 confirmed by simulation and mean-field analysis.
• The degree distribution transitions from binomial (for random duplication) to exponential (for partial fidelity) to power-law (for perfect duplication) as duplication fidelity increases.
• The mean-field solution of the degree evolution equation supports the 1/k law under the approximations of low connectance and high fidelity.
• The proposed ansatz based on symmetry considerations fits the simulation data well, especially in the low-connectance regime.
• Preferential detachment counterbalances preferential attachment, leading to a system where 'more begets less' in terms of degree evolution.
• The model suggests that ecological and biological networks may naturally exhibit 1/k-like distributions due to steady-state duplication and extinction dynamics, not growth-driven mechanisms.