eduKateSG · How Science Works
Choose your reading route
Water can remain liquid below its equilibrium freezing temperature because making the first stable ice crystal takes more than cooling. A small crystal must overcome a nucleation barrier. Until a sufficiently stable nucleus forms, the liquid can persist in a metastable state called supercooling.
Full contents · Check your understanding · How Science Works master guide
A thermometer reads −3°C, but the water still flows. Then ice begins to spread and the temperature rises. Neither observation requires the thermometer to be broken. Together, they reveal an important distinction: the conditions that favour a change are not always enough to make that change happen immediately. This guide follows that distinction from the first crystal to the measurements that could test the explanation.
Part of the How Science Works collection. Parent guide: Statistical Mechanics.
Contents
Understand the puzzle · Chapters 1–3
Follow the evidence · Chapters 4–6
4. Read the cooling curve as a sequence
Check and connect · Chapters 7–8
For pure water at approximately ordinary atmospheric pressure, 0°C is the familiar equilibrium melting/freezing temperature. At that condition, ice and liquid water can coexist. Below it, ordinary ice is the thermodynamically favoured phase. But an equilibrium statement is not a stopwatch: it does not specify the exact instant when a particular sample must begin crystallising.
There are two separate questions. Which phase is favoured under these conditions? That is a thermodynamics question. How does the sample reach that phase, and how quickly? That is a kinetics question. Supercooling appears when those answers do not line up on the timescale of our observation.
This is not the same as adding salt and lowering the equilibrium freezing temperature. A salt solution can remain liquid below 0°C without being supercooled relative to its own freezing point. To identify supercooling, compare a sample with the appropriate equilibrium boundary for its composition and pressure, not with a memorised number alone.
For the wider heating-and-cooling picture, read Phase Changes, Melting and Boiling. Here our narrower question is why the first ice crystal can be delayed.
CHAPTER 2 OF 8 · Understand the puzzle
2. The first crystal has an obstacle to overcome
Liquid water is not motionless or structureless. Its molecules continually rearrange. Small, short-lived arrangements resembling a piece of ice can form and disappear. However, making an ice-like cluster also creates a boundary between solid and liquid. That interface has an energetic cost.
For a very small cluster, the interface cost can outweigh the benefit of converting its interior to the favoured solid phase. The cluster is then likely to disappear. A sufficiently large, stable nucleus can instead grow. This competition is the central idea behind the nucleation barrier.
A helpful analogy is a ball in a shallow hollow beside a deeper hollow. The deeper hollow represents the more stable state, but the ball must first cross a ridge. Cooling changes the landscape; it does not mean that every ball crosses the ridge at precisely the same moment. The analogy describes a barrier, not literal balls inside water.
In the simplest spherical-nucleus model, surface cost scales with radius squared while the bulk contribution scales with radius cubed. This explains why “a tiny amount of ice-like order” and “a crystal that can keep growing” are different things. Real water requires more detailed models of molecular structure and the solid–liquid interface.
CHAPTER 3 OF 8 · Understand the puzzle
3. Metastable does not mean permanently stable
A supercooled liquid can survive small disturbances for a while even though another phase is more favourable. That is what metastable means here. It is neither at ordinary stable equilibrium nor guaranteed to change immediately.
Nucleation can occur within the liquid itself, called homogeneous nucleation, or be helped by a suitable surface or particle, called heterogeneous nucleation. A pre-existing ice crystal offers a particularly direct route for further ice growth. Not every speck of dust is equally effective, and cleanliness alone does not guarantee an indefinitely liquid sample.
Container surfaces, sample volume, cooling history and observation time can all matter. So can the way a temperature probe touches the sample. Repeating an apparently identical procedure may produce a range of freezing-onset temperatures rather than a single exact value.
The scientific lesson is useful well beyond water: a delayed transition need not disprove the equilibrium model. It may reveal that the pathway and the timescale need to be included. The IAPWS guideline on supercooled water explicitly treats a metastable liquid region; it does not redefine that region as ordinary stable liquid equilibrium.
Imagine a measured temperature–time graph. First, liquid water cools toward 0°C. Next, it passes below that temperature while remaining liquid. When crystallisation begins, the graph can turn upward toward the melting/freezing region even though the surroundings are still cold.
The upward turn is called recalescence. Freezing releases latent energy, warming the remaining liquid and the newly formed ice. This does not create energy from nothing. The phase change redistributes energy while the apparatus continues exchanging energy with its surroundings.
It also does not mean that all the water instantly becomes solid. Some ice may form quickly, releasing enough energy to reduce the supercooling. Further freezing then depends on continued energy removal. Real traces may not reach exactly 0°C because of heat transfer, composition, sensor position and response time.
A graph should therefore be read alongside observations of the sample. A sudden temperature rise by itself is not proof of freezing: a probe being moved into warmer liquid could also change its reading. The explanation becomes stronger when the temperature event coincides with visible crystal growth and appears consistently under controlled conditions.
CHAPTER 5 OF 8 · Follow the evidence
5. An energy estimate: how much initially freezes?
Illustrative calculation, not measured data. Suppose 100 g of pure water starts at −4°C. Estimate the fraction that must freeze to warm the whole sample to approximately 0°C, neglecting exchange with the surroundings and using a constant specific heat capacity of 4.2 J g⁻¹ K⁻¹.
The warming requirement is approximately 100 × 4.2 × 4 = 1,680 J. If freezing releases approximately 334 J per gram, the estimated frozen mass is 1,680 ÷ 334 ≈ 5.0 g. That is about 5% of the original mass.
The estimate is deliberately simplified. It neglects the different heat capacities of ice and water during the changing phase fraction, as well as the container and thermometer. Its purpose is not a precision prediction. It shows why a rapid freezing front can leave a water–ice mixture instead of converting every gram immediately.
Notice how the calculation disciplines the visual story. “The bottle turned cloudy” is an observation. “Every molecule froze instantly” is a much stronger claim that the observation does not establish. Always ask which quantities a video actually measures.
A useful investigation needs separate records of sample temperature, surroundings and the onset of crystallisation. A teacher-provided video with synchronised data is a safe starting point. Professional experiments can compare carefully characterised samples under specified cooling conditions.
If the question concerns nucleation sites, change that factor while holding composition, sample volume, cooling programme and measurement method as constant as practicable. If several of those change together, an earlier freezing event cannot be attributed confidently to one cause.
Repeat the comparison. Record the full spread of onset temperatures, not only the most impressive run. A probe has uncertainty and a finite response time; the water near it may differ from water at the container wall. A credible report names those limitations rather than hiding them behind extra decimal places.
Also specify what would challenge the interpretation. If apparent “freezing” occurs without any detected ice, if warming precedes the event, or if the sample contains enough dissolved material to shift its equilibrium boundary, the original explanation needs revision. That willingness to revise is part of the investigation, not a failure of it.
Safety: use supplied datasets or supervised demonstrations. Do not improvise with sealed freezing containers, glass, dry ice or laboratory chemicals. The historical Harvard demonstration record is evidence of the phenomenon, not a home procedure; its preparation includes hazardous specialist materials.
Supercooled droplets occur in clouds. They are relevant to the growth of ice particles and to icing when droplets encounter suitable surfaces. NOAA’s explanation of hail formation connects supercooled water with the growth of hailstones. The existence of subzero liquid droplets is therefore more than a laboratory curiosity.
However, cloud droplets, a laboratory flask and a household bottle are different systems. Their sizes, surfaces, impurities and thermal histories differ. A freezing temperature observed in one should not be advertised as a universal minimum for all liquid water.
Do not confuse supercooling with ice below 0°C. Ice can cool further after it forms, but that is a solid losing thermal energy, not a liquid delaying crystallisation. For that separate question, see Can Ice Be Colder Than 0°C?
Nor is supercooling the same as water having “no energy”. At these temperatures, molecules remain active. The relevant issue is the competition between phases and the probability of forming a viable nucleus, not the complete cessation of molecular motion.
Question 1: A sample remains liquid at −2°C. Does that prove supercooling? Answer: No. First establish composition, pressure, temperature accuracy and the appropriate equilibrium freezing condition. A solution may have a lower freezing point.
Question 2: Why can the measured temperature rise when freezing starts? Answer: Formation of ice releases latent energy. Some of that energy warms the sample even while energy continues to leave it.
Question 3: Why would one freezing run be insufficient to establish an exact onset temperature? Answer: Nucleation is sensitive to conditions and can vary between runs. A distribution of results, with measurement uncertainty, is more informative than one dramatic observation.
The key chain is: favourable solid phase → barrier to the first stable nucleus → delayed crystallisation → latent-energy release → measured evidence. If you can explain each link without saying that water has “forgotten” to freeze, you understand the central mechanism.
For deeper context, continue to Statistical Mechanics. The source links above provide an institutional thermodynamic guideline, a university demonstration record and a meteorological application. The worked calculation here is an original teaching estimate, not a reported experiment.
Contents · Previous chapter · Continue to Statistical Mechanics