3D Sandpile Criticality

Bak-Tang-Wiesenfeld sandpile on a lab platform: drop grains, watch avalanches, and measure the power-law avalanche statistics.

Switch to 2D

Presets

Playback

Grid actions

Dropping

Rules

Camera

Layers

Quality

Model check

The science

What is modelled

The Bak-Tang-Wiesenfeld sandpile on a square grid. Every site holds an integer number of grains z. A grain is dropped on a site; any site with z ≥ zc (default 4) topples: it sends one grain to each of its four neighbours (z → z − 4). Grains pushed over the edge of the platform fall into the catch pan. Toppling goes on in parallel sweeps until all sites are stable; that burst is one avalanche. The 3D view shows the grain count of each site as a column (not a true surface height: in the model z is a grain count, not a slope).

Equations

  • Toppling rule: zi → zi − 4, zj → zj + 1 for the four neighbours j of i.
  • Exact bookkeeping (abelian property): if ti is the number of times site i toppled and ni the grains added at i, then zifinal = ziinitial + ni − (4 ti − ∑j~i tj), with the sum over neighbours inside the grid. The final state and every ti do not depend on the order of toppling.
  • Avalanche size s = total topplings; duration T = number of parallel sweeps; area = distinct sites that toppled. At the critical state P(s) ∝ s−τ up to a cut-off set by the system size; published estimates of τ for the two-dimensional model scatter around 1.0 to 1.3, so the fitted value here is only an estimate.
  • Mean grain count of the self-organised state on a large grid: ⟨z⟩ ≈ 2.125 for the threshold 4 (finite grids and open edges give somewhat lower values).
  • Dissipation: each transferred grain is lost with probability f; then the avalanche size has a finite cut-off that shrinks as f grows.

Checks run in the page

  • Independent re-implementation (depth-first stack, one grain at a time) against the live parallel-sweep code on the same random start: identical final grid and identical toppling count at every site.
  • Conservation: grains added equal grains on the grid plus grains lost, and all sites end below the threshold.
  • The exact toppling identity above holds at every site after 1500 grains at the centre, and the resulting pile has the full eight-fold symmetry of the square.

Honest simplifications

  • The model is a cellular automaton, not granular physics: real sand has friction, inertia and a tilt-dependent avalanche angle, and real piles are not perfectly critical. The grains falling from the nozzle and into the pan are drawn only for illustration; the height of the pan heap is not to scale. The drop site moves instantly in the model while the gantry glides to it.

What to try

  • Start from "Empty grid, random drops" and watch the mean grain count climb to about 2.1 before big avalanches become common.
  • Use "Fill to maximum stable" and drop one grain: the whole platform can topple.
  • Raise the dissipation and watch the large avalanches vanish from the log-log plot.
  • Use "Centre pile" to grow the symmetric fractal pattern; switch on the toppling heat map.
Loading model...
Loading Blender model...
Mean grain count vs grains dropped
Avalanche size distribution (log-log)
Recent avalanche sizes
Made with 🟠 Blender · Three.js