What you're seeing
A grid of cells, each holding a small integer number of grains. We drive the system by dropping one grain at a time. Whenever a cell's pile reaches four, it is too steep to hold: it topples, handing one grain to each of its four orthogonal neighbors. Those neighbors may now be over the limit too, so the toppling spreads — an avalanche. Grains that topple off the edge of the grid simply fall away and are lost.
What makes this mesmerizing rather than trivial is the size of the avalanches. After a brief warm-up, the pile settles into a state where most dropped grains do almost nothing, but every so often a single grain unleashes a cascade that crosses the entire grid. There is no typical avalanche size; they come in all sizes, following a power law. The system reaches this knife-edge on its own, without anyone tuning it — that is self-organized criticality.
The rule
For every cell with height h:
- If h ≥ 4, the cell topples: subtract 4 from it and add 1 to each of its four orthogonal neighbors (up, down, left, right).
- Open boundary: grains sent past the edge of the grid are removed from the system. This loss is essential — it is the relief valve that lets the pile reach a steady critical state instead of filling up forever.
- Drive then relax: add one grain, then topple every unstable cell repeatedly until none remain (the pile is “stable,” all heights < 4). The number of topplings in that relaxation is the avalanche size.
The model's defining surprise is the abelian property, proved by Deepak Dhar (1990): when several cells are unstable, the final stable configuration — and the total number of topplings — does not depend on the order in which you topple them. You can relax cell A before cell B or the reverse and reach exactly the same answer. That is why the simulation is well-defined no matter how the work queue happens to schedule the cascade, and it is the reason these objects are called abelian sandpiles. (Our simulation lets a cell topple as many times as it can in a single visit and uses a work queue of only the currently-unstable cells, but by the abelian property the outcome is identical to toppling one grain at a time.)
Why it matters
A century of physics taught that a system reaches a critical point — where fluctuations occur on every scale and correlations stretch across the whole system — only when an external parameter (temperature, pressure) is tuned to a precise value. Water becomes critical at exactly one temperature-and-pressure; nudge it and the criticality is gone. Bak, Tang, and Wiesenfeld's 1987 insight was that some systems are different: driven slowly and dissipating at their boundaries, they drift to the critical point and stay there, with no fine tuning. The sandpile is the canonical example.
The signature of this critical state is scale invariance. Plot how often avalanches of each size occur and you get a straight line on log–log axes — a power law, P(s) ∝ s−τ, meaning there is no characteristic avalanche size. Double the size and the frequency drops by a fixed factor, forever; small avalanches are common, enormous ones are rare but never impossible. Watch the size-distribution bins in the readout fill in to see this happen. The same idea explains 1/f noise (also called “flicker noise”): a signal whose power is spread across all timescales, with slow fluctuations as important as fast ones. Bak, Tang, and Wiesenfeld titled their paper precisely as an explanation of 1/f noise — they proposed that many natural 1/f signals are the time-series fingerprint of an underlying self-organized critical state. Power laws and 1/f noise are two faces of the same scale-free coin: one in size, one in time.
In the wild
The sandpile is a toy model: real sand, poured in a real pile, does not behave like this cellular automaton (real grains have inertia, friction, and humidity, and experiments on actual sand and rice piles give mixed results — long-grained rice shows cleaner power-law avalanches than sand). The model's value is conceptual: it shows that scale-free behavior can arise for free from slow driving plus dissipation. With that caveat stated plainly, candidate self-organized-critical systems include:
- Earthquakes. The Gutenberg–Richter law — that earthquake frequency falls as a power law of magnitude (energy) — is one of the most robust empirical scaling laws in geophysics, and SOC is a leading candidate explanation. Whether the crust is genuinely self-organized-critical, or whether the scaling has another origin, remains actively debated.
- Landslides and avalanches. Landslide-size distributions are often approximately power-law, consistent with (but not proof of) SOC dynamics.
- Forest fires. The forest-fire cellular automaton is a sibling SOC model; some real fire-size records show power-law tails, though land management and weather complicate the picture.
- Neuronal avalanches. Cascades of activity in cortical tissue (Beggs & Plenz, 2003) follow power-law size distributions strikingly close to the critical exponent, and the “critical brain hypothesis” proposes the brain operates near a critical point. This is suggestive and influential, but contested — alternative non-critical mechanisms can produce similar statistics.
- Financial markets. Price-change distributions have famously fat (power-law) tails, and SOC has been proposed as an analogy for market crashes. This is the most speculative application; most economists regard it as a loose metaphor rather than an established mechanism.
The honest summary: the sandpile definitively establishes that self-organized criticality is possible and what it looks like. Whether any specific natural system is “really” SOC is, in most cases, an open and contested empirical question — power laws are necessary evidence but not sufficient proof.
Try this
- Start in Self-organized criticality mode and let it run. Watch the max avalanche in the readout keep jumping to new records — there is no ceiling, only diminishing odds. The bins show small avalanches vastly outnumbering large ones: that is the power law.
- Press +1000 at center repeatedly in either mode to inject a big load and watch a single giant relaxation ripple outward.
- Switch to Single pile, lower the speed, and watch the deterministic fractal grow — concentric boundaries and self-similar patches of teal, amber, and orchid. This is a genuine mathematical fractal, the same whether grown fast or slow (abelian!).
- Drag to paint a tall wall of grains by hand, then let it relax and collapse.
References
- Bak, P., Tang, C., Wiesenfeld, K. (1987). “Self-Organized Criticality: An Explanation of 1/f Noise.” Physical Review Letters 59(4), 381–384.
- Bak, P., Tang, C., Wiesenfeld, K. (1988). “Self-organized criticality.” Physical Review A 38(1), 364–374.
- Dhar, D. (1990). “Self-organized critical state of sandpile automaton models.” Physical Review Letters 64(14), 1613–1616. (Proof of the abelian property.)
- Bak, P. (1996). How Nature Works: The Science of Self-Organized Criticality. Copernicus / Springer.
- Gutenberg, B., Richter, C. F. (1944). “Frequency of earthquakes in California.” Bulletin of the Seismological Society of America 34(4), 185–188.
- Beggs, J. M., Plenz, D. (2003). “Neuronal avalanches in neocortical circuits.” Journal of Neuroscience 23(35), 11167–11177.