Supercooling: When Water Refuses to Freeze

The Video That Broke My Brain

There’s a video — you’ve probably seen some version of it — where someone takes a bottle of water out of a freezer. The water is clearly liquid. They tap the bottle once on the counter, and in about two seconds the entire thing turns to solid ice, spreading from the point of impact like a crystal wave. It looks like a magic trick. It looks fake.

It’s not fake. It’s supercooling, and the physics behind it is one of those things that seems straightforward until you actually think about it carefully. Then it becomes a rabbit hole that leads to nucleation theory, metastable thermodynamic states, and one of water’s many genuinely bizarre properties.

Why Water Doesn’t Always Freeze at 0 °C

Here’s the part they don’t emphasise enough in school: 0 °C is the temperature at which ice and liquid water are in equilibrium. It’s the temperature at which ice can exist stably alongside liquid water at atmospheric pressure. It is not necessarily the temperature at which water freezes.

Freezing requires more than just being cold enough. It requires nucleation — the formation of a tiny initial crystal structure (a nucleus) that the rest of the liquid can crystallise around. And forming that nucleus is energetically costly.

Think about it this way. When a few water molecules arrange themselves into the hexagonal structure of ice, they release energy — the latent heat of crystallization. Good, that drives the process forward. But they also create a new surface — the boundary between the ice nucleus and the surrounding liquid — and forming that surface costs energy (surface energy). For a tiny nucleus, the surface-to-volume ratio is enormous, so the energy cost of creating the surface exceeds the energy gained from crystallisation. The nucleus is thermodynamically unfavourable. It dissolves back into the liquid.

Only when the nucleus exceeds a critical size — typically a few nanometres — does the volume energy gain start to outweigh the surface energy cost. Above the critical size, the nucleus grows spontaneously. Below it, it melts back. This is classical nucleation theory, and it explains why phase transitions don’t happen instantly at the equilibrium temperature.

At 0 °C, the energy gain from crystallisation is so small that the critical nucleus size is enormous — far too large to form by random molecular fluctuations. The liquid is “supposed” to freeze, thermodynamically, but the kinetic barrier prevents it. As you cool further below 0 °C, the energy gain per molecule increases, the critical nucleus size shrinks, and nucleation becomes increasingly likely. But in very pure water with no foreign particles to serve as templates, the liquid can persist well below 0 °C — metastable, thermodynamically unstable but kinetically trapped.

Metastability: Stuck in the Wrong State

Supercooled water is a metastable state. It’s not the lowest-energy configuration (ice is), but it’s sitting in a local energy minimum with a barrier between it and the global minimum. It’s like a ball resting in a shallow dip on a hillside — technically, it should roll all the way to the bottom, but the dip is holding it in place. Jiggle the ball (tap the bottle, introduce a nucleation site) and it rolls down.

Metastability is everywhere in physics and chemistry. Diamond is metastable — graphite is the thermodynamically stable form of carbon at room temperature and pressure, but the conversion barrier is so high that diamonds last effectively forever. Supersaturated solutions are metastable — dissolved sugar beyond the saturation limit stays in solution until a crystal seed triggers rapid crystallisation (this is how rock candy is made). Some metallic glasses — metals cooled so fast that they solidify without crystallising — are metastable and will eventually crystallise if heated, sometimes explosively.

Supercooled water is just a particularly dramatic and accessible example. The metastable state looks perfectly normal — clear, liquid, unremarkable — until the barrier is overcome and the transition happens in seconds.

Nucleation: Homogeneous vs. Heterogeneous

There are two ways for ice to nucleate in supercooled water, and the distinction matters a lot.

Homogeneous nucleation is pure — no foreign surfaces, no impurities, just water molecules randomly fluctuating into an ice-like arrangement. This requires deep supercooling, typically below −35 °C for microscopic droplets. The probability of a critical nucleus forming by random fluctuation increases exponentially with the degree of supercooling, but at modest supercooling (say, −5 °C) it’s essentially zero. You can wait forever and homogeneous nucleation won’t happen.

Heterogeneous nucleation is the real-world version. A foreign surface — a dust particle, a scratch on the bottle wall, a dissolved mineral crystal — provides a template that lowers the activation energy for nucleus formation. The foreign surface effectively provides part of the crystal structure for free, reducing the critical nucleus size. This is why tap water freezes reliably at 0 °C (it’s full of dissolved particles and container defects) while very pure water in a smooth container can supercool to −10 °C or below.

The fact that ice formation depends on impurities is, when you think about it, remarkable. The phase transition itself — the rearrangement of H₂O molecules from disordered liquid to ordered crystal — is driven entirely by thermodynamics. But whether and when it happens depends on the presence of foreign particles that have nothing to do with the thermodynamics. Kinetics and thermodynamics are telling you different things, and in the supercooled regime, kinetics wins.

Supercooled Clouds and Rain

Up in the atmosphere, supercooled water is not a party trick — it’s a major driver of weather.

Clouds form when moist air rises, cools, and water vapour condenses onto tiny aerosol particles (cloud condensation nuclei). In a typical cloud at −10 °C, most water droplets are still liquid. They’re supercooled. Pure water droplets at this temperature are nowhere near cold enough for homogeneous nucleation, and not all aerosol particles are effective ice nuclei.

The Bergeron process, which produces most of the rain in mid-latitude regions, depends on this coexistence of supercooled liquid and ice. When an ice crystal does form (via heterogeneous nucleation on a suitable dust particle), it grows rapidly at the expense of surrounding supercooled droplets. Why? Because the saturation vapour pressure over ice is lower than over liquid water at the same temperature. Water molecules evaporate from the liquid droplets and deposit onto the ice crystal. The crystal grows while the droplets shrink and eventually disappear.

The ice crystal grows until it’s heavy enough to fall. If it passes through warm air on the way down, it melts and arrives as rain. If it stays frozen, it arrives as snow, sleet, or hail. Either way, the process started with supercooled liquid water that refused to freeze on its own.

Without supercooling, the physics of precipitation would be completely different. Clouds would freeze immediately upon reaching 0 °C, and the delicate interplay between ice and supercooled liquid that drives efficient precipitation wouldn’t exist. The water cycle, and thus weather and climate as we know them, depends on water’s stubborn refusal to freeze when thermodynamics says it should.

The Deep Puzzle: Water’s Second Critical Point

At the frontier of supercooling research, there’s something genuinely strange going on.

Water’s thermodynamic properties — heat capacity, compressibility, density — all behave anomalously as you cool it. They don’t just change smoothly; they diverge, getting more extreme as temperature drops. Liquid water at −20 °C is measurably denser than ice (obviously — ice floats) but also measurably more compressible than water at 20 °C, which is unusual for a liquid approaching its freezing point.

One hypothesis, proposed by Eugene Stanley and collaborators in 1992, suggests that water has a second critical point — a temperature and pressure at which two distinct liquid phases of water (high-density liquid and low-density liquid) become indistinguishable. This hypothesised critical point lies deep in the supercooled regime, around −45 °C and 1,000 atmospheres, in a region that’s extremely difficult to access experimentally because water nucleates ice so readily at those conditions.

If the second critical point exists, it would explain water’s anomalies as fluctuations between two liquid forms — similar to how near the familiar liquid-gas critical point (374 °C for water), density fluctuates wildly and the fluid becomes opalescent. The anomalous properties of supercooled water would be a distant echo of this hidden critical point, like hearing the rumble of a waterfall you can’t see.

This is still debated. Experimental evidence is tantalising but not definitive. Ultrafast X-ray scattering experiments on microdroplets at free-electron lasers have probed the deep supercooled regime in recent years, and the results are broadly consistent with the two-liquid hypothesis. But a smoking-gun confirmation remains elusive.

Tap the Bottle

Next time you see one of those supercooling videos, you’ll know what’s happening. The water was cooled below 0 °C in a smooth, clean container with minimal nucleation sites. It’s metastable — thermodynamically unstable but kinetically stuck. The tap provides a mechanical shock that either creates a nucleation site (a tiny crack in the surface, a bubble) or jostles existing micro-crystals into the critical size range. Once one critical nucleus forms, the crystallisation front propagates through the supercooled liquid at a few centimetres per second, releasing latent heat as it goes.

The temperature of the resulting ice? About 0 °C. All that latent heat released during freezing warms the ice-water mixture back up to the equilibrium temperature. You started below zero and ended at zero, because the system finally found its way to the ground state — the configuration it was always supposed to be in, if only it could get over the barrier.

That’s the thing about metastable states. They’re not wrong, exactly. They’re just waiting.