The Routing of Complex Contagion in Kleinberg's Small-World Networks

TL;DR

This paper investigates the difficulty of routing complex contagion in Kleinberg's small-world networks, showing that decentralized routing times grow polynomially with network size, unlike the more efficient simple contagion.

Contribution

It introduces the concept of complex routing in small-world networks and analyzes its routing time, revealing fundamental differences from simple contagion and complex diffusion.

Findings

Routing time is polynomial in network size for all lpha, both in expectation and with high probability.

Complex routing is significantly harder than complex diffusion, with exponential differences in routing times.

For lpha=2, simple contagion has polylogarithmic routing time, contrasting with complex contagion.

Abstract

In Kleinberg's small-world network model, strong ties are modeled as deterministic edges in the underlying base grid and weak ties are modeled as random edges connecting remote nodes. The probability of connecting a node $u$ with node $v$ through a weak tie is proportional to $1/|uv|^\alpha$, where $|uv|$ is the grid distance between $u$ and $v$ and $\alpha\ge 0$ is the parameter of the model. Complex contagion refers to the propagation mechanism in a network where each node is activated only after $k \ge 2$ neighbors of the node are activated. In this paper, we propose the concept of routing of complex contagion (or complex routing), where we can activate one node at one time step with the goal of activating the targeted node in the end. We consider decentralized routing scheme where only the weak ties from the activated nodes are revealed. We study the routing time of complex contagion and compare the result with simple routing and complex diffusion (the diffusion of complex contagion, where all nodes that could be activated are activated immediately in the same step with the goal of activating all nodes in the end). We show that for decentralized complex routing, the routing time is lower bounded by a polynomial in $n$ (the number of nodes in the network) for all range of $\alpha$ both in expectation and with high probability (in particular, $\Omega(n^{\frac{1}{\alpha+2}})$ for $\alpha \le 2$ and $\Omega(n^{\frac{\alpha}{2(\alpha+2)}})$ for $\alpha > 2$ in expectation), while the routing time of simple contagion has polylogarithmic upper bound when $\alpha = 2$. Our results indicate that complex routing is harder than complex diffusion and the routing time of complex contagion differs exponentially compared to simple contagion at sweetspot.