Dendrite Growth: The Physics Behind Snowflake Symmetry

A flat front is an unstable front

Picture a perfectly flat solid-liquid interface advancing into an undercooled or supersaturated liquid. In principle it could stay flat forever. In practice it almost never does, because any tiny random bump that pokes slightly ahead of the front finds itself in liquid that is more undercooled than the liquid right at the flat interface — undercooling that increases with distance ahead of the front as the diffusing latent heat or rejected solute spreads out. That extra undercooling drives the bump to grow faster than the flat surface around it, so the bump gets bigger, ahead further, and grows even faster still. Left unchecked, this positive feedback loop turns a flat front into a jagged one almost immediately.

What stops the front from turning to chaos at every length scale is surface tension. A very sharp, tightly curved bump costs a lot of surface energy per unit volume compared to a broad, gentle one, so surface tension actively suppresses short-wavelength bumps even while the diffusion field encourages long-wavelength ones. Balancing these two competing effects — William Mullins and Robert Sekerka worked out the balance explicitly in 1963 — picks out a single fastest-growing wavelength, and it is bumps near that wavelength that survive and grow into the primary arms of a dendrite.

The Mullins-Sekerka instability, sketched

The formal analysis perturbs a flat front with a small sinusoidal ripple of wavenumber k and asks whether the amplitude of that ripple grows or decays over time. The diffusion field ahead of the front is destabilising and its contribution to the growth rate scales roughly with k; surface tension is stabilising and its contribution scales with k^3, because curvature (and the energy cost of curvature) grows faster than amplitude for short wavelengths. The two effects trade off at a critical wavenumber, and everywhere below that critical wavelength perturbations decay while everywhere above it they amplify — with one particular wavelength growing fastest of all:

growth rate of a ripple of wavenumber k, roughly:

  omega(k)  ~   k * G_diffusion   -   k^3 * gamma * (curvature stiffness)

  omega(k) > 0   ->   ripple amplifies (unstable)
  omega(k) < 0   ->   ripple decays    (stable, surface tension wins)

  a single wavenumber k_max maximises omega(k)
  ->   sets the characteristic spacing between primary dendrite arms

Once a primary arm has grown out into the liquid, the same instability repeats on its sides at a smaller scale, producing secondary side branches, and then tertiary branches off those — a self-similar cascade that produces the characteristic tree-like (dendritic, from the Greek for tree) shape seen in everything from frost on a window to castings solidifying inside a mould.

Anisotropy: why the branches line up

The linear stability analysis alone would let bumps form in every direction around a growing crystal — it does not by itself explain why real dendrites branch along specific, repeatable directions. That comes from crystalline anisotropy: surface energy and interface kinetics are not the same in every direction relative to the crystal lattice, and a small amount of directional preference is enough to select which directions grow fastest and suppress everything in between. For ice, whose molecules pack into a hexagonal lattice, the preferred growth directions are set six degrees apart, which is exactly why every snowflake — no matter how different its detailed branching pattern is from any other, because that detail is set by the exact, chaotic history of humidity and temperature it experienced while falling — shares the same underlying six-fold symmetry.

From snowflakes to castings and welds

The exact same instability governs metal solidification, where it is usually driven by constitutional undercooling rather than thermal undercooling: as an alloy solidifies, the solid phase typically rejects one alloying element into the remaining liquid, building up a solute-rich layer just ahead of the front whose local freezing point is depressed below the actual temperature there. Metals crystallise into cubic lattices rather than hexagonal ones, so metallic dendrites branch in four-fold or six-fold patterns set by the cubic symmetry rather than ice's hexagonal one, but the underlying Mullins-Sekerka mechanism is identical. Controlling this dendritic structure — its arm spacing, its orientation, whether it forms at all — is central to alloy casting, welding, and additive manufacturing, because the spacing between dendrite arms directly controls how far solute has to diffuse during subsequent heat treatment, and therefore the final mechanical properties of the part.

Frequently asked questions

Try it live

Everything above runs in your browser — open Dendrite Growth and change the parameters while it is running. Nothing is installed, nothing is uploaded, the whole model lives in one tab.