Supercooling
A Socratic walk-through of supercooling — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why can very pure water sit well below freezing as a liquid, and then turn to ice the instant you tap the bottle?
We are taught that water freezes at 0 °C, and the phrasing invites us to hear it as a command. Yet a clean bottle of still water, cooled slowly, will sit at -8 °C as a liquid indefinitely — until you knock it against the table, at which point ice races through it in a second.
Two things need explaining, and they pull in opposite directions. Why does freezing not happen when thermodynamics says ice is the stable phase? And why, once it starts, does it finish so fast?
Reasoning it through
REASONING #Take the second question first, because it constrains the answer to the first. If freezing is fast once begun, growing ice is not what is difficult. The difficulty must be in starting — in making the first piece of ice out of nothing but liquid.
What would that first piece cost? Imagine a tiny sphere of ice appearing inside the cold water. Its interior is the stable phase, so its volume pays a free-energy dividend scaling as the cube of the radius. But it also has an interface with the water, and interfaces cost energy — a cost scaling as the square of the radius. For a small sphere the square term dominates, so the embryo is uphill: it costs energy to make, even though the phase it is made of is the favoured one.
Follow the two terms as the sphere grows. The cost rises, peaks, then falls as the volume term overtakes the surface term. There is a critical radius: below it an embryo shrinks even though ice is stable, because shrinking lowers the total energy; above it, growth is downhill and runaway. The barrier is real, and it is geometric rather than chemical.
Then what gets over it? Thermal fluctuation. Water molecules constantly assemble into transient ordered clusters and fall apart again; almost all are sub-critical, but occasionally one exceeds the critical size by luck and grows. That makes freezing a stochastic event with a rate, not a temperature-triggered switch — which is the shape of the observation, since supercooled samples do not all freeze at the same temperature.
The barrier shrinks rapidly as you go colder, because the bulk dividend grows while the surface cost does not, so there should be a temperature at which chance succeeds almost immediately even in perfectly clean water. There is: pure water in small droplets can be carried to roughly -38 °C, and near there liquid cannot persist. Household supercooling never gets close, which tells us ordinary water at -5 °C is not freezing by luck alone.
It is freezing on something. A foreign surface with a lattice resembling ice lets an embryo form as a cap against that surface rather than as a full sphere, so far less new interface has to be made for the same volume, and the barrier collapses. This is heterogeneous nucleation, and it is why real water freezes near 0 °C: dust, container scratches, and specialist nucleators do the work — silver iodide, whose lattice spacing is close to ice's, well enough to be used in cloud seeding. So the question's premise is only half right. Purity matters, but so do the container and stillness, and the bottle is doing at least as much work as its contents.
Now the tap. Mechanical disturbance nucleates ice, and how is genuinely still argued: the leading account has a sharp impact producing cavitation, with the pressure transients around a collapsing bubble favouring an embryo, though dislodging an already-present nucleus is also plausible. Neither is settled.
One last prediction, checkable and slightly surprising. Freezing releases latent heat, and a supercooled sample cannot shed it quickly, so the liquid warms itself as it freezes and stops at 0 °C. How much ice can it make? Water's latent heat of fusion is about 334 joules per gram and liquid water's specific heat about 4.18 joules per gram per kelvin, so 6 °C of supercooling converts only about 4.18 x 6 / 334 — roughly 7.5 percent — of the sample. That is why the bottle turns to slush rather than a block, the rest freezing slowly afterwards as heat leaves through the walls.
The analogy
THE ANALOGY #Think of a boulder resting in a shallow dip near the top of a long hillside. Downhill is where it belongs, and the dip is barely holding it — but it is holding it, and it will hold it forever unless something jostles the boulder over the lip. Tilt the whole hill steeper and the dip becomes shallower, until at some angle no jostle is needed at all.
one boulder either goes or does not, whereas a supercooled bottle contains a vast number of independent attempts happening constantly, so the correct description is a probability per second, not a yes or no.
Clarifying the model
THE MODEL #Supercooled water is not unstable, and it is not "trying" to freeze. It is metastable: genuinely at rest in a local minimum, with a barrier between it and the deeper one, and left alone it sits there indefinitely.
Nothing here contradicts the freezing point. 0 °C remains where ice and water are in equilibrium; supercooling is a statement about kinetics. That is worth separating from the freezing-point-depression explanation in this collection, which shares the phase diagram but moves a different quantity: dissolved salt shifts where equilibrium lies. Supercooling leaves equilibrium exactly where it was and asks why the system fails to reach it — and salted water can supercool below its own depressed freezing point. The barrier itself is general, governing limescale precipitating from solution and bubbles in a carbonated drink alike.
And the stochastic character has one counterintuitive consequence: a larger volume of the same water freezes at a higher temperature than a small droplet, because it offers more chances per second for a critical embryo to appear.
A picture of it
THE PICTURE #How to readThis is a worked plot of the standard nucleation expression, not measured data — written in units of the critical radius and the barrier height, the curve is exactly 3x squared minus 2x cubed, with no material constants in it. Trace left to right as an embryo grows: everything left of 1.0 is uphill, so an embryo that size shrinks back even though ice is stable. At 1.0 the curve turns over, and beyond it growth releases energy and runs away. Cooling does not change the shape; it moves the peak leftward and downward until thermal jostling clears it easily.
What became clearer
WHAT CLEARED #Freezing is two separate problems, and only one is thermodynamics. Below 0 °C ice is favoured, but that says nothing about how the first crystal appears — and the first crystal must pay a surface cost before it can collect a volume dividend, so small embryos are penalised and dissolve. Ordinary water escapes the barrier by cheating, nucleating on dust and scratches; clean, still water in a smooth bottle has nothing to cheat with, and waits. The tap supplies what was missing, and after that the only limit is the latent heat the water must shed — hence slush rather than a block of ice.
Where to go next
ONWARD #- How insects and polar fish use antifreeze proteins to block ice growth rather than lower the freezing point.
- Why cooling fast enough to skip nucleation entirely produces glassy water, and what that means for cryopreservation.
Key terms
TERMS #| Term | What it means |
|---|---|
| Metastable | genuinely at rest in a local energy minimum, with a barrier separating it from a lower one. |
| Critical radius | the embryo size above which further growth lowers total free energy and so becomes self-sustaining. |
| Homogeneous nucleation | formation of a critical embryo from the liquid alone; in water it needs roughly -38 °C. |
| Heterogeneous nucleation | formation of an embryo against a foreign surface, which lowers the barrier and is how nearly all real freezing begins. |
Every term the collection defines is gathered in the glossary.