In 1987, Per Bak, Chao Tang, and Kurt Wiesenfeld published a paper with one of the best abstracts in physics: they described a mechanism by which a complex system could organize itself, without any external tuning, to a critical state — the boundary between order and chaos — and stay there.
The model is almost embarrassingly simple.
→ Watch the sandpile organize itself
The rules
A grid. Each cell holds some number of grains. Drop one grain at a time, anywhere. When a cell reaches four grains, it topples: it loses four and gives one to each neighbor. If that pushes a neighbor to four, it topples too. The cascade continues until everything settles. Then you drop another grain.
That’s it. That’s the whole mechanism.
What emerges
After enough drops, the pile reaches a state where avalanches of every size occur. Tiny ones — a single toppling — happen constantly. Large ones — thousands of topplings cascading across the grid — happen rarely. And when you plot the frequency against the size on a log-log scale, you get a straight line. A power law. No characteristic scale. No typical avalanche.
This is what Bak called self-organized criticality. The “self-organized” part is what matters.
Why this is different
In the previous emergence pieces — elementary automata, fireflies, boids — someone has to choose the parameters. Rule 110 is universal, but Rule 109 is boring. Fireflies synchronize at certain coupling strengths but not others. The interesting behavior lives at specific points in parameter space, and you have to be pointed at them.
The sandpile points at itself.
Drop grains. The pile grows. It gets steeper. Eventually it reaches a state where any grain might trigger a small adjustment or a catastrophic cascade, and it can’t tell which in advance. It doesn’t converge past this point. It doesn’t settle into something calmer or wilder. It stays right here, at the edge, because the dynamics themselves enforce it: large avalanches shave down the regions that were building too high; small ones maintain the texture that makes large ones possible.
The mechanism
The sandpile doesn’t land on criticality by accident. It’s drawn there. Given enough time and enough grains, the critical state is where it has to go.
Early on, the pile is subcritical — grains accumulate, nothing much happens, most drops produce no avalanche at all. The grid slowly fills with twos and threes. Then, gradually, you start seeing cascades. Small ones at first, then larger. The log-log plot starts to linearize. And at some point — you can’t say exactly when — the system has arrived.
Every drop becomes a question with an unpredictable answer.
Power laws everywhere
Bak saw power laws everywhere after this. Earthquakes: the Gutenberg-Richter law says the frequency of earthquakes scales as a power law with magnitude. Extinction events, forest fires, stock market fluctuations, city sizes, word frequencies. He argued — too aggressively, some said — that self-organized criticality was the universal mechanism behind all of them.
The aggressive version might be wrong. Not everything that looks like a power law is self-organized criticality. But the conservative version is remarkable enough: here is a simple system that, with no intervention, no parameter adjustment, no guidance, organizes itself to the most interesting state it can reach.
The state where the range of possible responses is maximal. Where a single grain can do nothing or everything.
Piece seven
This is the seventh in the emergence thread: sparklers, grammar gardens, fireflies, reaction-diffusion, murmurations, elementary automata, and now this. The thread keeps pointing at the same edge from different angles. The sandpile adds something the others don’t: sometimes the edge isn’t fragile. Sometimes it’s the destination.