[Paper Review] A Linear Preferential Attachment Process Approaching the Rado Graph

Richard Elwes|arXiv (Cornell University)|Mar 29, 2016

Complex Network Analysis Techniques6 references3 citations

TL;DR

This paper studies a linear preferential attachment process where new vertices connect to existing ones with probability proportional to their current degree. Using martingale techniques, it proves that, unless the initial graph is edgeless or complete, the process almost surely converges to the Rado graph with finitely many isolated or universal vertices added.

ABSTRACT

We consider a preferential attachment network growth process under a simple linear dynamic: vertices are added to the network one at a time, and at stage $t$ the probability that the new $t+1$st vertex connects to vertex $u \leq t$ is exactly $\frac{d_u(t)}{t}$ where $d_u(t)$ is the degree of $u$ at time $t$. Building on some Martingale machinery introduced by Kleinberg and Kleinberg in [9], we show that, so long as the initial graph is neither edgeless nor complete, with probability 1 the infinite limit of the process will be a copy of the Rado graph augmented with a finite number of either isolated or universal vertices. Finally, we offer an informal argument that this result does not generalise to processes with attachment probability $\lambda \cdot \frac{d_u(t)}{t}$ where $\lambda <1$.

Motivation & Objective

• To analyze the almost sure limit of a linear preferential attachment process in random network growth.
• To determine under what conditions the process converges to the Rado graph, a universal random graph.
• To extend prior martingale-based analysis from Kleinberg and Kleinberg to establish convergence properties.
• To investigate whether the result holds under sublinear attachment probabilities (λ < 1).

Proposed method

• The process adds one vertex at a time, with connection probability to vertex u at time t equal to d_u(t)/t, where d_u(t) is u's degree at time t.
• Martingale theory is applied to analyze the almost sure convergence of vertex degrees and connectivity patterns.
• The analysis focuses on the limiting structure of the infinite random graph formed by the process.
• The method relies on structural properties of the Rado graph, particularly its universality and homogeneity.
• The proof establishes that the limit graph is isomorphic to the Rado graph, possibly with finitely many isolated or universal vertices.
• An informal argument is provided to suggest that the result fails when attachment probability is scaled by λ < 1.

Experimental results

Research questions

• RQ1 Under what conditions does a linear preferential attachment process converge to the Rado graph?
• RQ2 How does the initial graph structure affect the almost sure limit of the process?
• RQ3 Can martingale techniques from prior work be extended to prove convergence to the Rado graph?
• RQ4 Does the result generalize to sublinear attachment probabilities (λ < 1)?
• RQ5 What is the role of isolated or universal vertices in the limiting structure?

Key findings

• With probability 1, the infinite limit of the process is isomorphic to the Rado graph, provided the initial graph is neither edgeless nor complete.
• The limiting graph may include a finite number of isolated or universal vertices, but otherwise matches the Rado graph structure.
• The convergence result relies on martingale analysis of vertex degrees and connectivity patterns over time.
• The process does not converge to the Rado graph when the attachment probability is scaled by λ < 1, based on an informal argument.
• The result establishes a precise almost-sure convergence to a universal random graph with minimal structural assumptions on the initial graph.