Supercooled water

Why can water sit below freezing as a liquid for hours and then turn solid the instant it is knocked?

A bottle of water sits below zero as a liquid, indefinitely. Knock it and ice races through it in a second.

The companion account of supercooling in this collection derives why that is possible at all: a new ice embryo pays a surface cost before it earns a volume reward, so there is a critical radius and an energy barrier, and freezing is therefore a stochastic event with a rate rather than a temperature-triggered switch. Take all of that as given.

What that leaves unexplained is a plain observational fact. If the barrier were the one that theory computes for pure water forming ice out of nothing but itself, water would routinely sit liquid to around −38 or −40 °C. It does not. Ordinary water in ordinary containers freezes somewhere between about −2 and −20 °C, sample by sample, with a scatter far too wide to be a property of water. Something is lowering the barrier, and it is not the water.

Start by asking what the barrier is actually made of. It is a surface cost: forming an ice embryo means creating an ice-water interface, and interfaces cost energy. The critical radius is where the volume reward finally overtakes that cost.

Now ask whether the embryo has to create all of that interface itself.

Suppose the embryo forms not in open liquid but against a solid surface — a speck of dust, a scratch in the glass, a mineral grain. Part of what would have been ice-water interface is now ice-solid interface instead. If the ice wets that solid well, the substitution is favourable: the embryo is buying some of its boundary at a discount.

Follow the geometry. An embryo sitting on a surface as a spherical cap needs a much smaller volume to reach the same critical radius than a full sphere does, because the surface has supplied part of its shape for free. The critical radius itself is unchanged — that is set by temperature — but the number of molecules that must assemble by chance to reach it can be smaller by orders of magnitude. And since the rate depends on how often a fluctuation of that size occurs, the rate goes up enormously.

That is the whole mechanism. Homogeneous nucleation — ice forming unaided in the bulk liquid — is what the textbook barrier describes, and it is almost never what happens. What happens is heterogeneous nucleation on whatever surface in the sample is the best catalyst, and the observed freezing temperature is a measurement of that surface, not of the water.

Several puzzling observations now fall out at once.

The scatter. Identical-looking samples freeze at different temperatures because they contain different impurities. The distribution of freezing temperatures across many samples is really a distribution of catalyst quality.

Volume dependence. A large sample freezes warmer than a small one from the same source, because a larger volume is more likely to contain a rare good nucleator. Divide water into many tiny droplets and the droplets that happen to contain no catalyst can be cooled much further — which is precisely how the homogeneous limit is measured experimentally.

The knock. Mechanical disturbance does not supply energy in any thermodynamically meaningful amount — the barrier is far too large for a tap to push a system over it directly. What agitation plausibly does is create transient cavities and surfaces, disturb the boundary layer at the container wall, or fracture and disperse a marginal embryo already present so that several fragments can grow. I should be honest that the details here are less settled than the rest of the account.

The runaway. Once one embryo passes critical size, growth is downhill, and the latent heat released warms the surrounding liquid toward 0 °C. So the whole vessel does not freeze solid: a slush forms fast and the rest freezes slowly as heat leaves. The dramatic second is the release of the supercooling, not the whole phase change.

Not all surfaces are catalysts, which is what makes the field interesting. What matters is not dirtiness but whether the solid's surface structure resembles ice closely enough for water to order against it. Certain mineral dusts and bacterial nucleators are strikingly good, effective a degree or two below zero; other materials are inert. So a filthy sample can supercool further than a clean one holding a single good nucleator — the strongest evidence that this is structural match, not contamination.

The homogeneous limit is real and is where the theory is tested. Around −38 to −40 °C, ice forms in pure water without help at rates so high that no sample survives. That figure is recalled rather than derived here. Everything warmer than it is a statement about catalysts; the limit itself is the property of water the theory actually predicts.

This matters well beyond the kitchen. Whether a cloud droplet freezes governs whether a cloud gives rain, snow or neither, and droplets are small and often clean, so they routinely supercool far below zero. Atmospheric ice nucleation is a live research area, and one place this laboratory curiosity has real consequences.

What is contested. The catalytic efficiency of specific materials, the contribution of airborne biological material, and the precise mechanism by which agitation triggers freezing are all argued over. The framework is well established; the parameters within it are not, and I quote almost none.

The falsification test. If observed freezing temperature is set by the best catalyst present rather than by water itself, then dividing one sample into many small droplets should push the distribution of freezing temperatures sharply colder, and adding a known good nucleator should collapse it to a warm, narrow spike — with the underlying liquid unchanged in both cases. If droplet size made no difference and every sample froze at the same temperature, nucleation would be homogeneous after all and this account would be wrong.