Percolation Theory: Clusters, the Critical Threshold & Phase Transitions

Connectivity from a coin flip per site

Percolation theory is the branch of probability and statistical physics that studies how connectivity emerges in disordered systems built from many randomly arranged components. Imagine a lattice in which each site is independently either open or blocked; percolation asks whether open sites link together into a cluster large enough to span the whole structure. This seemingly simple question conceals remarkably rich behaviour, including a sharp critical point at which the global character of the system changes abruptly. Because so many real systems are inherently disordered — porous rocks, composite materials, social contact networks, power grids — percolation provides a unifying language for describing when large-scale connectivity, conduction or contagion suddenly becomes possible.

The percolation threshold and the spanning cluster

The central object in percolation is the occupation probability p — the chance that any individual site is open. When p is small, open sites scatter into tiny isolated clusters; as p increases, clusters grow and merge. There exists a sharp critical value, the percolation threshold p_c, at which an infinite, system-spanning cluster appears for the first time. For an infinite lattice the probability that the origin belongs to an infinite cluster is exactly zero for p < p_c and strictly positive for p > p_c — a genuinely discontinuous change in global character, not a gradual one.

P(p) ~ (p - p_c)^beta        strength of the spanning cluster, p > p_c
xi   ~ |p - p_c|^(-nu)        correlation length, diverges at p_c

site percolation, square lattice:      p_c ≈ 0.5927
bond percolation, square lattice:      p_c =  0.5   (exact)
site percolation, triangular lattice:  p_c =  0.5   (exact)

Universality: one set of exponents, many lattices

Approaching p_c from either side, the system develops a characteristic length scale — the correlation length ξ — measuring the typical size of finite clusters; at the threshold itself there is no finite scale at all, and the spanning cluster becomes a fractal, self-similar across many length scales. The most profound result here is universality: although p_c itself is non-universal and depends on microscopic lattice details, the critical exponents β, ν and others depend only on the spatial dimension, not on whether one studies site or bond percolation or which particular lattice is used. In two dimensions these exponents take exact rational values derived through conformal field theory and Schramm–Loewner evolution; above the upper critical dimension of six, they settle to mean-field values. Universality is why a single model illuminates phenomena as diverse as gelation in polymers and the fragmentation of communication networks.

Where the threshold shows up

In porous media and oil recovery, the flow of water or hydrocarbons through rock depends on whether open pores connect into a spanning network. In conductive composites, mixing conducting particles into an insulating matrix produces electrical conduction only once the filler fraction exceeds a percolation threshold — the principle behind carbon-loaded conductive plastics. In epidemics and forest fires, spread through a contact network or a forest halts unless the transmission probability crosses a threshold. And in network robustness, randomly removing nodes from the internet or a power grid is equivalent to inverse percolation, with the threshold predicting how many failures a system can absorb before it fragments.

Common misconceptions

A frequent misunderstanding is that the percolation transition is gradual; for an infinite system the appearance of the spanning cluster is perfectly sharp, and the apparent smoothing seen in computer experiments is a finite-size effect, not a property of the underlying transition. Another error is assuming the threshold is a single universal number — it is not, it varies with lattice type and dimension, and only the critical exponents are universal. Site and bond percolation are also often conflated, yet the two generally have different thresholds even on the same lattice. Finally, percolation is not the same as diffusion: it describes whether a path exists at all, not how quickly something travels along it.

Frequently asked questions

Try it live

Everything above runs in your browser — open Percolation — Clusters & Critical Threshold, sweep the occupation probability p, and watch a spanning cluster snap into existence right around p_c ≈ 0.593. Nothing is installed, nothing is uploaded.