THE CRITICAL SANDPILE

One falling grain may do nothing at all, or bring down a whole flank of the pile, and there is no telling which in advance. The rule of Bak, Tang and Wiesenfeld from 1987 is almost insultingly simple: any cell holding 4 grains topples, handing one grain to each of its 4 neighbours, and grains falling off the edge are gone for good. Keep dropping grains and the average load per cell settles near 2.1 with nobody tuning it, while avalanche sizes follow a power law: on log-log axes they line up almost straight, so no size counts as typical.

▸ Building the cell grid — each cell counts sand grains…
▸ Loading 1 single rule: cell ≥ 4 grains TOPPLES to 4 neighbours
▸ Grains fall, the system builds up gradually…
▸ One grain can ignite a chain AVALANCHE
▸ The system drives itself to the CRITICAL state (SOC)
▸ Ready — Online. ✅
0%
⌂ Complex systems

Simulation room The critical sandpile

Self-Organized Criticality · Bak–Tang–Wiesenfeld
Online
1 rule · topple threshold ≥ 4
Observation panel
🏖️ Đống cát
thumbnail of the whole grid
Mean grains / cell—
Total drops—
Last avalanche—
Largest avalanche—
Near-critical cells (≥3)—
State—
The idea
Just one rule: any cell stacking ≥ 4 grains topples, splitting 4 grains to 4 neighbours. Yet the system crawls to the critical point — where a small grain can cause an avalanche of any size. Most are tiny, rarely a giant one — exactly a power law.
—
Cell color by grain count 0–3 (dark→bright) · cell ≥4 topples · click the grid to drop grains · yellow region = toppling. Try "Drop at center" or "Sand rain".
Your browser has canvas disabled.
Avalanche size over time · log-log distribution sizedistribution
About Sandpile (SOC)

One falling grain may do nothing at all, or bring down a whole flank of the pile, and there is no telling which in advance. The rule of Bak, Tang and Wiesenfeld from 1987 is almost insultingly simple: any cell holding 4 grains topples, handing one grain to each of its 4 neighbours, and grains falling off the edge are gone for good. Keep dropping grains and the average load per cell settles near 2.1 with nobody tuning it, while avalanche sizes follow a power law: on log-log axes they line up almost straight, so no size counts as typical.

Related topics in Complexity & Chaos