Field notes on things that run themselves
A Slope Held Up by Its Own Collapse
Drop dry sand onto a table, one grain at a time, and for a while nothing interesting happens. The pile grows, roughly conical, its slope climbing along with it. Then, at some angle, the slope stops climbing. Grains keep arriving at the same steady rate as always, but the pile holds its shape — not because gravity or friction has changed, but because the pile has started shedding avalanches. Most move a single grain a few centimetres. Once in a while one runs the length of the slope. Nobody set that angle on purpose. The pile found it.
In 1987 Per Bak, Chao Tang and Kurt Wiesenfeld gave that behaviour a name and a machine. Their model sandpile was a grid of cells holding stacks of grains; add one grain, and once a cell’s height crosses a threshold it topples, handing a grain to each of its four neighbours, which may then topple in turn. Grains leave at the edges roughly as fast as new ones arrive, and the grid settles into a state needing no further adjustment. They called it self-organized criticality — critical because the system sits exactly at the boundary between stable and runaway, self-organized because nothing outside it tunes it there. An ordinary magnet must be heated to one precise temperature before it shows critical fluctuations. This pile finds its own critical point, and stays on it.
The signature they were after was statistical, not visual. Run the model long enough, measure how many cells topple in each cascade, and the sizes fall on a power law: no bump at some typical size, no cutoff except at the size of the grid itself. A cascade that moves one cell and a cascade that moves ten thousand are the same kind of event, produced by the same rule, differing only in how far the chain reaction happened to run before it stopped finding a taker.
That is a new shape for this series, worth saying plainly. Every steady state described here so far has been one quantity holding its value — a flame’s outline, a wind’s period. This is the first one steady only as a distribution: no single avalanche is stable or foreseeable, and the pile never settles on one slope so much as hovers near one. What holds still is the shape of the scatter, not any one event inside it — which separates it cleanly from a phantom traffic jam’s one travelling threshold, and from a reversing wind whose period comes from a two-term balance with an actual value, even a wandering one. A sandpile has no representative avalanche and no representative period. It has a distribution, and that is what is conserved.
Actual sand, dropped onto an actual table, does not reliably cooperate. In 1996 a team at the University of Oslo ran the experiment on rice instead, pouring grains one at a time and timing every avalanche automatically. Long, needle-shaped grains produced avalanches with no characteristic size, matching the model closely. Short, rounder grains produced avalanches clustered around a typical size, rarely growing large no matter how long the pile was fed. Same protocol, same slope, same one-grain-at-a-time rule; the difference was how a grain sheds energy against its neighbours as it slides — exactly the detail the clean cellular-automaton math is built to ignore. Self-organized criticality is not the universal, detail-blind law it was first sold as. It is a property some physical systems earn and others do not, depending on friction the elegant version of the theory never had to specify.
A clearer real-world confirmation came from an unlikelier place. In 2003 John Beggs and Dietmar Plenz recorded spontaneous activity in living rat cortex — cultured tissue and slices alike — with a sixty-channel electrode array, watching bursts ripple across it between quiet stretches. The bursts, which they named neuronal avalanches, followed a power law with an exponent close to −3/2, matching what theory predicts for a critical branching process, and the tissue’s own branching ratio sat close to the value at which a disturbance neither dies out nor runs away: one. It is the sandpile’s critical slope, measured in spikes instead of grains. Whether that means cortex genuinely sits at a self-organized critical point, or some other dynamic produces the same statistics, is still argued over two decades on — the evidence is suggestive, not settled. Earthquakes and forest fires get invoked in the same breath; both show power-law size distributions, and both are more contested as genuine self-organized criticality than the cortex is.
What the sandpile actually teaches, then, is narrower than the slogan built on it, and better for being narrower. Somewhere between one grain and one cascade, certain systems arrange themselves to fail at every scale at once instead of reliably at one scale — and where that happens, the failing is what keeps the whole thing standing. Not despite the collapses. Because of exactly how many there are, how small most of them stay, and how rarely one gets enormous.
One loop I’m watching
Next: a waterfall is not a place. It is a step in a river’s profile that retreats upstream a little at a time, undercutting its own lip until the overhang collapses forward into itself — the same drop, the same plunge pool, relocated a few metres, made now of whichever rock happened to be there when it arrived.
Tip: the ← and → arrow keys move between issues.
Prefer feeds? Follow by RSS or JSON Feed.
Want something playable between issues? Thought Toys has its own email list and RSS feed.
New to The Standing Wave? Start here →