Water does not want to freeze.
That's the wrong word — want — but the right shape. At 0°C, ice and liquid water have the same free energy. The phase boundary is a line of indifference. Below zero, ice is more stable, thermodynamically favored. But the liquid doesn't transform. Not because it resists. Because there is nothing to transform onto.
A crystal needs a surface. Not a metaphor — a physical requirement. Ice formation requires a seed: a lattice fragment, a mineral grain, a scratch on the container wall. Something with geometry close enough to hexagonal ice that water molecules can align against it and begin to stack. Without a seed, the liquid holds.
−5°C. The molecules are slower now. Their thermal motion has dropped below what the liquid state requires for its particular kind of disorder. Hydrogen bonds flex and almost lock — a hundred million times per second, somewhere in the volume, six molecules arrange themselves into a hexagonal ring and immediately dissolve. The cluster needs to reach a critical radius before the energy gained by forming crystal exceeds the cost of creating a new surface. Below that size, the embryo costs more than it's worth.
The liquid holds.
−20°C. The critical radius has shrunk. The viscosity has increased — molecules stay in accidental arrangements longer, and the probability of the right arrangement persisting long enough to recruit its neighbors rises by orders of magnitude. Every thermodynamic gradient points toward ice. Every energetic calculation says the solid should exist. Still liquid.
−38°C. The homogeneous nucleation limit. Here, even without a seed, the probability of a spontaneous cluster reaching critical size becomes certainty. In laboratory droplets — ten micrometers across, purified of every nucleation site — ice appears from nothing but water. But most water never reaches −38°C. Most water meets a boundary first.
A speck of kaolinite drifts through the liquid. Clay mineral, lattice geometry close enough to hexagonal ice that water molecules find a template. They align against the surface. The first layer forms in nanoseconds. The second layer in microseconds. The crystal extends outward from the seed into liquid that has been ready for this since the first degree below zero.
The ice front moves at ten centimeters per second. Dendrites — branches along the crystal's preferred growth axis, each spawning side-branches at sixty degrees, because the hexagonal lattice permits no other angle. The geometry is not chosen. It is the only geometry the lattice allows.
Every molecule that joins the crystal releases heat. 334 joules per gram — the latent heat of fusion, running in reverse. Freezing warms the surrounding liquid. At twenty degrees of supercooling, roughly a quarter of the water crystallizes before the released heat raises the rest back to zero. Then the dendrites stop. The system has paid its energy debt — exactly, no more — and the remaining water freezes slowly, from the outside in, the way water always freezes when it has time.
What remains is ice laced with liquid. The dendrites are thin, elaborate, structurally intricate. Given hours, they will thicken, merge, simplify into the featureless block that ice becomes when there is nothing left to balance against. The branching was temporary — the shape of a transformation happening faster than equilibrium could follow.
No one chose to freeze. No one waited. The metastable state was not patience. It was the absence of a path. When the path appeared, the system took it — exactly as far as the energy allowed — and stopped.