Scale-Free Networks: How 'Rich Get Richer' Builds the Internet's Hubs

Rich get richer: the rule behind hubs

Most human-made and natural networks are not built by connecting nodes at random. The internet's backbone, citation networks, protein-interaction maps and social-media follower graphs all share a lopsided structure: a handful of enormously connected hubs and a long tail of nodes with almost no connections at all. Albert-Laszlo Barabasi and Reka Albert modelled this in 1999 with one deceptively simple growth rule: preferential attachment — every new node arriving in the network is more likely to connect to a node that already has many connections.

P(connect to node i)  =  k_i / Σ_j k_j

k_i    = current degree (number of connections) of node i
Σ_j k_j = sum of degrees over all existing nodes

Start with a small seed network. Add one node at a time, each bringing m new edges, and wire each of those edges to an existing node chosen with probability proportional to that node's current degree. A node that happens to gain a few extra links early — pure chance, or being first to exist — becomes slightly more attractive to the next arrival, which makes it even more attractive to the one after that. Small early advantages compound into enormous, self-reinforcing hubs.

The power law that falls out

Run this process for a long time and the fraction of nodes with degree k converges to a power law: P(k) ∝ k^-γ, with the Barabasi-Albert model producing an exponent of exactly γ = 3 regardless of m. A power-law degree distribution has no characteristic scale — there is no typical node size the way a bell curve has a typical value — which is exactly why these networks are called scale-free. A log-log plot of P(k) against k is the diagnostic: an Erdos-Renyi random graph shows a curved, bell-shaped line on those axes, while a scale-free network shows a straight one.

P(k)  ∝  k^(-3)      ← Barabasi-Albert degree distribution

random graph:    most nodes near the average degree, exponential tail
scale-free graph: most nodes have few links, a few hubs have enormous ones

Why hubs make networks both efficient and fragile

Scale-free structure has a sharp double edge. Because a few hubs connect enormous swaths of the network, the average path length between any two nodes stays remarkably short even as the network grows — the small-world property that makes six degrees of separation plausible on a network of billions. Removing random nodes barely dents this: almost every node has low degree, so a random failure is very unlikely to hit a hub, and the network stays connected. But a targeted attack on the highest-degree hubs is catastrophic — knocking out a handful of the biggest hubs can fragment the whole network into disconnected islands. This robust-yet-fragile signature shows up in real infrastructure: the internet tolerates huge numbers of random router failures but is vulnerable to attacks on its major exchange points.

Not every network is scale-free

Preferential attachment is a mechanism, not a universal law — plenty of real networks (road networks, power grids, some social networks measured carefully) show exponential or other degree distributions instead, and the claim that scale-free structure is ubiquitous has been actively debated since a influential 2018 statistical re-analysis by Broido and Clauset found rigorous power-law fits far rarer among real-world networks than earlier looser analyses suggested. What preferential attachment does reliably explain is why hubs form wherever growth and cumulative advantage both operate — new websites linking to already-popular ones, new citations favouring already-cited papers, new followers gravitating to already-popular accounts.

Frequently asked questions

Try it live

Everything above runs in your browser — open Scale-Free Network and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.