loading...
running...
The science
What this models. A snow crystal grows when water vapour diffuses to its surface and freezes there. The shape is the result of a competition: tips stick out into more vapour and grow faster (diffusion instability), while surface physics decides where molecules can attach (facets). Here the process is simulated with the Reiter / Gravner-Griffeath hexagonal cellular automaton, the standard minimal model for snow-crystal morphology.
1. The lattice. Cells are hexagons (axial coordinates q, r with six neighbours), so growth is naturally six-fold. Each cell holds diffusive vapour d, a boundary mass b (quasi-liquid layer on the ice surface), crystal mass c, and a flag "attached" (part of the crystal). A cell is receptive if it is attached or touches an attached cell.
2. Diffusion. Only non-receptive cells hand out vapour. With u = d for non-receptive cells and 0 for receptive ones, and ū the mean of u over the six neighbours,
d(x) ← d(x)[receptive] + u(x) + (α/2) (ū(x) - u(x))
On a hexagonal lattice with unit spacing ū - u = (1/4) ∇2u, so this is the diffusion equation ∂d/∂t = D ∇2d with D = α/8 cell2 per step. A point pulse therefore spreads as <r2> = 4 D t = α t / 2 (check 2 below). The far boundary of the hexagonal domain is held at the vapour density ρ.
3. Freezing. On each receptive, not yet attached cell the vapour turns into boundary mass and a bit of crystal: b ← b + (1 - κ) d, c ← c + κ d, d ← 0.
4. Attachment. A boundary cell joins the crystal when the surface is ready for it. With n attached neighbours: n ≤ 2 and b ≥ β (a tip needs a lot of vapour, which makes arms), or n = 3 and (b ≥ 1, or the vapour around it is below θ and b ≥ α), or n ≥ 4 (a hole fills in at once). Then c ← c + b and b ← 0. The larger β is, the more the crystal prefers to fill in as a plate instead of branching.
5. Melting and noise. Boundary cells that did not attach return a fraction μ of b and γ of c to the vapour (re-evaporation smooths the surface). Optional noise multiplies the vapour by 1 ± σ. With σ = 0 and a one-cell seed the whole system is exactly symmetric under 60 degree rotations: every arm is a copy of every other, which is why real flakes look symmetric (they share the same history) but are never identical when the noise differs.
Checks (run in the page). (1) total mass d + b + c is conserved to machine precision on a closed torus with no melting or noise; (2) the mean-square spread of a point pulse after 120 steps equals α t / 2; (3) the grown crystal maps onto itself exactly under a 60 degree rotation (cells with no partner = 0). The live readout also shows the 6-fold mismatch count for the crystal you are growing.
Simplifications. This is a phenomenological automaton with dimensionless parameters, not a quantitative model: it has no temperature or supersaturation scale, no surface-energy anisotropy beyond the lattice, no latent-heat diffusion, and one cell state per hexagon. The parameter sets in the presets were chosen to show the main morphologies (fern dendrites, stellar arms, sectored stars, plates); mapping them to a particular temperature on the Nakaya diagram would be an overreach. In the 3D page the thickness of a cell grows with its age as an artistic cue; the physics is two-dimensional.