Robustness of a Network of Networks

Almost all network research has been focused on the properties of a single network that does not interact and depends on other networks. In reality, many real-world networks interact with other networks. Here we develop an analytical framework for studying interacting networks and present an exact percolation law for a network of $n$ interdependent networks. In particular, we find that for $n$ Erd\H{o}s-R'{e}nyi networks each of average degree $k$, the giant component, $P_{\infty}$, is given by $P_{\infty}=p[1-\exp(-kP_{\infty})]^n$ where $1-p$ is the initial fraction of removed nodes. Our general result coincides for $n=1$ with the known Erd\H{o}s-R'{e}nyi second-order phase transition for a single network. For any $n \geq 2$ cascading failures occur and the transition becomes a first-order percolation transition. The new law for $P_{\infty}$ shows that percolation theory that is extensively studied in physics and mathematics is a limiting case ($n=1$) of a more general general and different percolation law for interdependent networks.

💡 Research Summary

The paper “Robustness of a Network of Networks” tackles a fundamental limitation of most complex‑network research: the assumption that a single graph can be studied in isolation. In reality, many critical infrastructures—power grids, communication systems, transportation, financial markets—are tightly interwoven, and failures in one layer can cascade to others. To capture this, the authors introduce a formal “network of networks” (NON) framework in which n distinct networks are coupled through one‑to‑one dependency links. Each network is modeled as an Erdős–Rényi (ER) random graph with the same average degree k, and a fraction 1 − p of nodes is removed uniformly at the outset. After the initial removal, each network is pruned to its giant component; any node that does not belong to the giant component is considered failed, and because of the dependency links, its counterpart in every other network also fails. This process repeats iteratively, producing a cascade of failures that may either stop with a non‑zero surviving giant component or collapse entirely.

The core analytical contribution is a closed‑form self‑consistency equation for the size of the mutually connected giant component, denoted P∞. For a single ER network (n = 1) the classic percolation relation is P∞ = p