Self-similarity in Fractal and Non-fractal Networks

TL;DR

This paper investigates the conditions under which scale-free networks exhibit self-similarity after coarse graining, revealing that fractality and self-similarity are distinct properties influenced by specific exponents.

Contribution

It introduces a framework linking degree distribution self-similarity to exponents governing box mass and renormalized degree, clarifying their roles in fractal and non-fractal networks.

Findings

Self-similarity occurs when γ ≤ η or under a specific relation between exponents.

Fractality and self-similarity are shown to be unrelated properties.

The study provides conditions for self-similarity independent of fractal structure.

Abstract

We study the origin of scale invariance (SI) of the degree distribution in scale-free (SF) networks with a degree exponent $\gamma$ under coarse graining. A varying number of vertices belonging to a community or a box in a fractal analysis is grouped into a supernode, where the box mass $M$ follows a power-law distribution, $P_m(M)\sim M^{-\eta}$. The renormalized degree $k^{\prime}$ of a supernode scales with its box mass $M$ as $k^{\prime} \sim M^{\theta}$. The two exponents $\eta$ and $\theta$ can be nontrivial as $\eta \ne \gamma$ and $\theta <1$. They act as relevant parameters in determining the self-similarity, i.e., the SI of the degree distribution, as follows: The self-similarity appears either when $\gamma \le \eta$ or under the condition $\theta=(\eta-1)/(\gamma-1)$ when $\gamma> \eta$, irrespective of whether the original SF network is fractal or non-fractal. Thus, fractality and self-similarity are disparate notions in SF networks.