Flash Freezing: How Supercooled Water Crystallizes in a Blink

A liquid that has no business staying liquid

Below its melting point, ice is the thermodynamically stable form of water: the Gibbs free energy of the solid is lower than that of the liquid. The driving force is the free energy of fusion, ΔG = ΔH − TΔS. At the melting point Tₘ = 273.15 K the two phases are in balance and ΔG = 0, because ΔHᶠ = 6.01 kJ/mol is exactly offset by TₘΔSᶠ with ΔSᶠ = ΔHᶠ/Tₘ ≈ 22 J/(mol·K). Cool below Tₘ and the balance tips: freezing becomes downhill in free energy, and the deeper the undercooling ΔT, the larger that thermodynamic push.

So why doesn't the water just freeze? Because thermodynamics tells you where a system wants to go, not how fast it can get there. To become ice, water first has to assemble a minute cluster of molecules locked into the ice lattice — and building that first cluster costs energy it doesn't want to pay. Until a cluster appears, the supercooled liquid sits in a metastable valley: perfectly stable to small nudges, poised to avalanche given a big enough one. Pure, quiescent water can hold this state for hours at modest supercooling, though its metastable lifetime collapses as it approaches about −40 °C, where homogeneous nucleation becomes essentially instantaneous.

The nucleation barrier: why ice needs a seed

Classical nucleation theory pins down the cost. Imagine a spherical embryo of ice, radius r, forming inside the liquid. Two energies compete. Building bulk ice releases free energy proportional to the volume, −(4/3)πr³·|ΔGᵥ|, where ΔGᵥ is the (favorable) free-energy change per unit volume. But creating the new ice–water boundary costs energy proportional to the surface, +4πr²γ, where the interfacial energy γ ≈ 25–32 mJ/m². Add them:

• ΔG(r) = 4πr²γ − (4/3)πr³|ΔGᵥ|

For small embryos the surface term (r²) dominates, so ΔG rises as the cluster grows — every added molecule is uphill and the embryo tends to redissolve. Only once the cluster passes a critical radius r* = 2γ/|ΔGᵥ| does the volume term take over and growth becomes downhill. The peak of that curve is the nucleation barrier ΔG* = 16πγ³ / (3ΔGᵥ²).

The crucial fact is that |ΔGᵥ| grows with undercooling (roughly |ΔGᵥ| ≈ ΔHᶠ,ᵥₒₗ·ΔT/Tₘ), so both r* and ΔG* shrink as the water gets colder — ΔG* falls off as 1/ΔT². Near 0 °C the critical nucleus is enormous (tens of nanometers, ~10⁶–10⁷ molecules) and forms essentially never. By −40 °C it has shrunk to only ~1–2 nm and a few hundred water molecules — small enough that random thermal fluctuations finally throw one together. The nucleation rate follows J = A·exp(−ΔG*/kʙT), and because ΔG* sits in an exponential, J is almost zero for tens of degrees and then explodes over a window of a few degrees around −38 °C.

Homogeneous versus heterogeneous: the −40 °C ceiling

Everything above describes homogeneous nucleation — ice appearing spontaneously in the bulk liquid. It only wins near −38 to −40 °C, which is why that temperature is the practical floor for liquid water: below it, pure water freezes no matter what. Real water almost never gets that cold before freezing, because it cheats via heterogeneous nucleation. A foreign surface — a dust mote, a mineral grain, a scratch on the container, a gas bubble, even a specialized bacterial protein — acts as a template. The embryo grows against that wall instead of enclosing itself in liquid, so much of the expensive interface is replaced by a cheaper solid–ice contact.

Mathematically the barrier is simply scaled by a geometric factor set by the contact angle θ: ΔG*ₕₑₜ = ΔG*ₕₒₘₒ·f(θ), with 0 ≤ f(θ) ≤ 1. A well-matched surface makes f small, collapsing the barrier and letting ice form just a degree or two below 0 °C. This is the entire difference between tap water and the demo:

• Tap or mineral water is a suspension of nucleators — dissolved minerals, silt, microbubbles — so it nucleates heterogeneously at 0 to −5 °C and can't be deeply supercooled.
• Purified, filtered, degassed, undisturbed water has stripped out the good nucleators, so it must wait for the homogeneous route and can be chilled far below 0 °C.

Nature and industry exploit both sides. Silver iodide (AgI) has a crystal lattice almost identical to ice and is a superb ice nucleus — the basis of cloud seeding. The bacterium Pseudomonas syringae makes an ice-nucleating protein that templates ice at nearly −2 °C; freeze-dried, it is sold as Snomax to help ski resorts make snow at marginal temperatures.

The blink: a racing front and its self-limiting heat

Once one critical nucleus survives, the metastable state is over. Ice grows fastest not as smooth crystals but as branching dendrites — needle-like fingers that shoot forward because their sharp tips shed the released heat efficiently into the surrounding cold liquid. Their growth velocity climbs steeply with undercooling, roughly as ΔT², from about a centimeter per second at a few degrees of supercooling to on the order of ~1 m/s near the deep-supercooling limit. That is why a bottle or a poured stream appears to freeze all at once: the front crosses a 10–30 cm column in about a second.

Now the beautiful part. Freezing is exothermic — it releases the enthalpy of fusion, L = 334 J/g — and in the fast, near-adiabatic flash that heat has nowhere to go but back into the ice–water mush. This warms the slush and is called recalescence. The system self-limits: it can only freeze as much ice as the released latent heat is able to warm from the supercooled temperature back up to 0 °C, where solid and liquid coexist and further freezing halts. A simple energy balance gives the frozen fraction:

• f · L ≈ cₚ · ΔT  →  f ≈ cₚΔT / L, with cₚ ≈ 4.18 J/(g·K).

So supercooling of ΔT = 8 K freezes only about 10% of the water; ΔT = 40 K freezes roughly 50%. That is exactly why the flash-freeze demo yields a slush at 0 °C rather than a solid block — the water literally cannot pay the full latent-heat bill from the modest thermal deficit it carried. Only if you keep extracting heat afterward does the rest finish freezing conventionally. (And note the ice that forms is ordinary hexagonal ice Ih, density 0.917 g/cm³ — about 9% less dense than liquid water, which is why the crystal front can crack a full, capped bottle as it expands.)

Where supercooling actually matters

This is far more than a kitchen trick. Supercooled water is everywhere the demonstration is rare:

• Clouds and aviation. Cloud droplets are tiny and clean, so they routinely stay liquid down to −20 °C and below, freezing only near −38 °C (homogeneous freezing of droplets is a main route to cirrus ice). When such droplets strike a cold airframe they nucleate on contact — aircraft icing, a serious flight hazard that de-icing systems exist to fight.
• Freezing rain. Raindrops that fall through a shallow sub-freezing layer supercool without nucleating, then flash-freeze the instant they hit a road or power line, glazing everything in clear ice.
• Snowmaking. Snow guns atomize water into fine supercooled droplets; nucleators like Snomax or AgI ensure they crystallize before hitting the ground even when it's only a couple of degrees below freezing.
• Cryopreservation. Here nucleation is the enemy: ice crystals shred cell membranes. Vitrification sidesteps it entirely — cooling faster than ~10⁴ K/s, often with cryoprotectants that raise viscosity, outruns nucleation so the water sets into a glass (amorphous solid) with no crystals at all. It's the same physics as the demo, deliberately denied its nucleus.
• Food and biology. Industrial “flash freezing” of food aims for many tiny crystals (fast cooling, small grains, less cell damage) rather than the few large ones slow freezing makes; and cold-hardy insects, fish, and plants survive winter by producing antifreeze proteins that block or control ice nucleation and growth in their tissues.

Common misconceptions (and the safe way to see it)

“The water was frozen and the tap just shook it loose.” No — before the tap it is genuinely liquid, just metastable. The disturbance doesn't dislodge ice; it supplies (or exposes) a nucleation site — a shock-formed bubble, a bit of dust, or the ice you pour it onto — letting the very first crystal form.

“Colder means more will freeze instantly.” Partly. Colder water carries a bigger thermal deficit, so a larger fraction (f ≈ cₚΔT/L) flashes to ice — but because of recalescence it always stops at 0 °C as a partial slush, never a solid block, unless heat keeps leaving.

“It's the dissolved salt/minerals lowering the freezing point.” Colligative freezing-point depression is real but tiny for tap water (fractions of a degree). The dominant reason tap water doesn't supercool is the opposite — its impurities are excellent nucleators that make it freeze near 0 °C, not solutes that keep it liquid.

“Supercooled water can be any temperature.” There's a hard floor: around −38 to −40 °C homogeneous nucleation becomes unavoidable, so no ordinary liquid water persists much below that at 1 atm.

As a controlled demonstration the phenomenon is mild — the only real cautions are that a sealed bottle can crack as ice expands ~9%, and that the slush is at 0 °C. The scientific payoff is understanding why: a metastable liquid, a nucleation barrier that falls as 1/ΔT², a dendritic front, and latent heat that caps the result — the same chemistry that governs clouds, icy roads, and frozen food.

Why some water supercools to −38 °C and some freezes at 0 °C — it is all about the seed

Water sample	Typical freeze onset	Governing mechanism	What you see
Pure, still, degassed water	as low as −38 to −40 °C	must nucleate homogeneously — barrier huge	stays liquid, then flash-freezes when disturbed
Tap / mineral water	0 to −5 °C	heterogeneous nucleation on dust, ions, bubbles	freezes gradually, little supercooling
Seeded water (ice chip, AgI, dust)	at or just below 0 °C	template lowers the barrier to ~zero	freezes immediately on contact
Cloud droplet (few µm, clean)	down to −38 °C	few impurities → deep supercooling common	supercooled clouds; freezing rain; aircraft icing
Vitrified cryo-sample	never crystallizes	cooled >10⁴ K/s + cryoprotectant → glass	amorphous solid, no ice crystals form