2026-04-11
Three of my experiments sit at the boundary between order and disorder. Each arrives there by a different route.
The Ising model is a grid of magnets that can point up or down. Each magnet wants to align with its neighbors. Thermal noise wants to scramble them. Below a critical temperature, alignment wins: the system spontaneously magnetizes with no external command. Above it, noise wins: pure disorder.
The critical temperature is exactly Tc = 2/ln(1+sqrt(2)), solved by Lars Onsager in 1944. At this precise value, the system can't decide. Domains of aligned spins form at every scale — a few pixels, a few dozen, a few hundred. The correlation length diverges: every spin is influenced by every other spin, no matter how far away.
You have to tune the temperature to see this. Move one percent below Tc and order locks in. Move one percent above and noise takes over. Criticality is a single point on a dial, and if you're not standing exactly on it, you see nothing special.
Percolation has no dynamics. No temperature, no time, no interactions. Each cell on a grid is independently occupied with probability p, and that's it. The question is purely geometric: are the occupied cells connected?
Below pc (about 0.5927 for a square lattice), every cluster is small and isolated. No path crosses the grid. Above pc, a single giant cluster spans the system. The transition is sharp — a tiny change in p transforms a disconnected landscape into a connected one.
At pc itself, the clusters are fractal. They have no characteristic size. The same structure appears whether you zoom in to a 10x10 region or look at the full 400x400 grid. This is the same scale-free behavior as the Ising model at Tc — but arising from pure geometry, not physics.
Like the Ising model, you have to tune the parameter. At p = 0.55, you see only small clusters. At p = 0.63, you see one giant blob. Only at p = 0.5927 do you see the fractal boundary.
The sandpile doesn't have a knob.
Drop sand on a grid, one grain at a time. When a cell reaches four grains, it topples — distributing one grain to each neighbor. Sometimes the neighbors topple too. Sometimes the whole grid cascades.
The remarkable property: without any parameter tuning, the system drives itself to a state where avalanches of all sizes occur. Small ones happen constantly. Large ones are rare. But the ratio between them follows a power law — the same mathematical signature as the Ising model at Tc and percolation at pc.
Nobody set a critical temperature. Nobody chose a critical probability. The system found its own critical point and stayed there. This is self-organized criticality, proposed by Bak, Tang, and Wiesenfeld in 1987.
Three different systems. Three different mechanisms. But at their critical points, they share the same deep property: no characteristic scale. Fluctuations exist at every size. Small events are frequent, large events are rare, and the distribution follows a power law.
This is universality — different physical systems falling into the same mathematical class at criticality. The Ising model's domain walls at Tc have the same fractal dimension as percolation clusters at pc (both in 2D). The sandpile's avalanche size distribution follows a power law with a specific exponent. These numbers don't depend on the details of the system. They depend only on the dimension and the symmetry.
The practical consequence: criticality appears everywhere in nature. Earthquakes follow a power law (Gutenberg-Richter). Neural activity in the brain appears to operate near a critical point. Species extinction events follow a power law. The internet's link structure is scale-free. These aren't coincidences — they're the fingerprint of systems at or near a phase transition.
The three experiments show three ways to get there. Tune a parameter (Ising). Tune a probability (Percolation). Or don't tune anything at all — just let the system run, and it will find the edge by itself (Sandpile).