Sandpile — self-organized criticality

The hypothesis

Take a grid. On each cell, stack grains of sand. When a cell reaches four, it topples: it hands one grain to each of its four neighbours, and any grain that falls off the edge of the board is lost. One toppling can trigger others — that is an avalanche. Add the grains one at a time, let the pile relax between each, and watch what happens.

What we set out to show: without ever touching a parameter, the pile finds a particular state on its own — the critical state — where avalanches have no characteristic scale. There is no “typical” avalanche: one grain may do nothing, or set off a cascade that sweeps across the whole board.

What the run shows

On a 128×128 grid, 60,000 grains dropped at random (seed 42), the critical state is reached and holds:

• Mean height: 2.12 grains per cell. The pile neither blows up nor dies out — it settles around the critical value (~2.1 for this model). No one tuned it: dissipation at the edges and the toppling rule are enough.
• Avalanches range from 1 to 82,083 topplings — nearly five orders of magnitude, across more than fifteen thousand avalanches. The median is 13. In other words: most are tiny, a few are enormous, and there are avalanches at every size in between.
• The size distribution follows a power law, with an estimated exponent τ ≈ 1.31. On a log-log plot the sizes fall on a straight line — the signature of a scale-free system. Hence the critical verdict.

The contrast with the start is stark: while the grid is nearly empty, dropping a grain triggers nothing. The scale of avalanches does not exist at first — it emerges as the pile fills up and self-organizes. That is what rules out an artefact: the power law is not hidden in the local rule, it is the product of self-organization.

Why it matters

This is self-organized criticality, described by Per Bak, Chao Tang and Kurt Wiesenfeld in 1987. Their idea: many natural systems tune themselves to just the edge of chaos, with no fine-tuning, and then produce power-law events — with no typical size. The same principle is invoked for earthquakes (the Gutenberg-Richter law), forest fires, snow avalanches, species extinctions, sometimes market crashes: cascades in which a small trigger can have consequences of any magnitude.

In the LifeLabs catalogue, the sandpile adds a kind of emergence that was missing. Lenia and reaction-diffusion grow a pattern; the agents and the avatar, a behaviour; Physarum, a network; the boids, collective motion. The sandpile grows a statistic: a power law, the mark of a system living at the edge of a transition.

A note on vocabulary

“Critical” here is nothing dramatic: it is the physicists’ word for a system at the boundary between two regimes, where correlations reach across every scale. The pile “decides” nothing, “seeks” no balance — it applies a toppling rule, over and over, and that simple rule is enough to place it at the critical point. That is the whole point: the complexity we observe (the absence of scale) comes not from tuning, but from repeating a trivial mechanism.

The demo is interactive: drop grains with a click, adjust the rain, and watch the log-log histogram fill in until the straight line of the power law appears, in real time.