9. Networks
Source: ebook_Network Theory.pdf; network-topology material also drawn from ebook_Complexity_Theory.pdf / Slide_Complexity_Theory_Guide.pdf (same series).
Chapter 1 introduced the relational paradigm: connections, not properties, are the primary explanatory factor. This chapter gives that idea a precise, structural vocabulary — the formal language for describing how things are connected, not just that they are.
Nodes, edges, and paths
A network models a system as nodes (entities) and edges (connections between two nodes, which may be directed or undirected, weighted or unweighted) — a direct graph-theoretic reformulation of “element” and “relation” (ch. 2, 4).
- Degree — the number of edges a node has; the simplest measure of a node’s local connectivity.
- Path — a sequence of edges connecting two nodes; the geodesic is the shortest such path between them.
- Centrality — several distinct ways a node can be structurally “important”: high degree centrality (many direct connections), high closeness centrality (short paths to everyone else), high betweenness centrality (lies on many other nodes’ shortest paths — a bridge or bottleneck), high eigenvector centrality (connected to other well-connected nodes).
Network “distance” is not physical distance — it’s a function of connectivity. Two cities equally far apart on a map can be very differently “close” in an air-traffic network depending on whether either is a hub; a well-connected hub can be network-closer to a distant city than a geographically nearer but poorly-connected one.
Topology: how a network is shaped
- Random — connections form with uniform, independent probability between any two nodes (Erdős–Rényi). No particular structure; degree is roughly uniform across nodes.
- Distributed / mesh — even degree across nodes, high local autonomy, no dominant hubs. Robust to random node loss (many redundant paths) but relatively inefficient (long average paths).
- Decentralized / small-world — local clusters of densely-connected nodes linked by a relatively small number of longer-range connections. Combines high local clustering with surprisingly short average path length — the “six degrees of separation” phenomenon.
- Centralized / scale-free — a small number of hubs account for a disproportionate share of all connections, following a power-law degree distribution generated by preferential attachment (“rich get richer”: new nodes preferentially connect to already well-connected ones). Efficient (short paths through hubs) but structurally fragile.
Clusters and weak ties
A cluster (or clique) is a subset of nodes far more densely interconnected with each other than with the rest of the network — often built on homophily, the tendency of similar nodes to connect (“birds of a feather”). This is a sharper, more structural cousin of “differentiation” (ch. 4): differentiation is about functional specialization, clustering is about the actual density and similarity of ties.
Weak ties (Granovetter) are the comparatively infrequent connections that bridge separate clusters. Individually minor, they are disproportionately important for a network’s overall diversity and integration — most new information, opportunities, and innovations cross a system via its weak ties, not its strong ones, because strong ties mostly just circulate what a cluster already knows.
Robustness and contagion
The same connectivity that integrates a system is also the pathway anything harmful propagates through — connectivity is a double-edged sword:
- Distributed networks tolerate random node failure well (many redundant paths) but are inefficient day-to-day.
- Scale-free/centralized networks are efficient and robust to random failure (most nodes aren’t hubs) but fragile to targeted attack on their hubs — losing one hub can fragment the whole network.
Diffusion/contagion across a network depends on the topology plus how “infectious” the thing spreading is and how resistant nodes are to it:
- Simple contagion — a single exposure is enough to spread (a rumor, a virus).
- Complex contagion — spreading requires reinforcement from multiple sources or crossing a
threshold (costly behavior change, adoption of an unproven innovation) — this connects back to
the phase transition idea in
04-relations-synergy-emergence.md: below a percolation threshold nothing spreads system-wide; above it, the same local rule produces a qualitatively different, system-wide outcome. The network effect (a network’s value grows disproportionately, often quadratically, with its number of connected nodes — Metcalfe’s law) is the same underlying dynamic viewed from the value side rather than the contagion side.