Does Hot Water Freeze Faster Than Cold?

On its own, “does hot water freeze faster than cold water?” is too loose to test. Too much is left unstated. To test it you must fix the starting temperatures, what the water is made of, the shape and material of the container, and what surrounds it as it cools. You must also fix the slipperiest variable of the lot: what you are willing to call “frozen.”

“Frozen” might mean the first ice crystal forming, or the surface reaching 0°C, or a thermocouple in the middle reading 0°C, or the whole volume going solid. These are four separate moments. Score the race by one of them and the hot sample wins; score it by another and the cold one does.

In a 2016 paper reviewing the evidence, Henry Burridge and Paul Linden of the University of Cambridge found no universally accepted definition. They noted that Mpemba and Osborne, in their 1969 paper, timed how long water took to start freezing, while other studies included the freezing itself.

Stated carefully, the real question runs: under what reproducible conditions does water that starts hotter reach a fixed freezing endpoint before an otherwise identical sample that starts colder? Put that way, there is no clean yes or no. Sometimes it happens, sometimes it doesn’t, and even where it does, the physics behind it is only partly worked out.

Aristotle put it plainly in Meteorology (Book I, Part 12), from about the year 350 before the Common Era: “The fact that the water has previously been warmed contributes to its freezing quickly: for so it cools sooner. Hence many people, when they want to cool hot water quickly, begin by putting it in the sun.” The explanation he reached for, antiperistasis (a quality supposedly intensifying when hemmed in by its opposite), is wrong by any modern reading.

The note keeps recurring. Francis Bacon, in Novum Organum (1620), wrote that “water a little warmed is more easily frozen than that which is quite cold”. René Descartes took it up in Meteorology, an essay published with his Discourse on Method in 1637. Water kept hot for a long time, he wrote, freezes faster than any other sort, because “those of its parts which can least cease to bend evaporate while it is being heated”.

The modern thread begins in 1963. Erasto Mpemba was in Form 3, the third of four years of secondary school, at Magamba Secondary School in Tanzania. He was making ice cream, as boys at the school did, by boiling milk, mixing in sugar and freezing it.

The mixture was meant to cool first, but there was a rush for space in the refrigerator. Rather than lose the last ice tray, he put his milk in hot; another boy had put his in without boiling it. An hour and a half later his tray had frozen into ice cream while the other boy’s was still a thick liquid.

His physics teacher said, “You were confused, that cannot happen.” Later, at Mkwawa High School, a teacher called it “Mpemba’s physics and not the universal physics”, and his classmates took the phrase up as a joke.

The question stayed with him. The physicist Denis Osborne, of University College, Dar es Salaam, visited Mkwawa to give a talk. Mpemba asked him: “If you take two similar containers with equal volumes of water, one at 35°C and the other at 100°C, and put them into a refrigerator, the one that started at 100°C freezes first. Why?”

Osborne could have brushed it aside. He admitted instead that he did not know, and promised to check. Back at University College in Dar es Salaam, he asked a young technician to test it, and the water that started hot did freeze first.

Mpemba and Osborne’s joint paper ran in Physics Education in 1969, under the title “Cool?” They used 70 ml of water in 100 ml beakers, stood on polystyrene in the icebox of a domestic refrigerator. In their graph, water that started at about 90°C began to freeze after about 30 minutes. Water that started at about 20°C took about 100.

Evaporation, Mpemba and Osborne worked out, could account for no more than 30% of the cooling. They offered a tentative explanation instead: convection kept the top of the hot water warmer, so it went on losing heat quickly from its surface.

The interest never quite died down. In 2012 the Royal Society of Chemistry put up £1,000 for the best explanation. It drew 22,000 entries, whittled to eleven finalists by expert judges and a public vote. The prize went to Nikola Bregović of the University of Zagreb, announced on 10 January 2013. Awarding it settled who had explained the effect best that year, which is not the same as settling what causes it.

Newton’s law of cooling says a body sheds heat at a rate set by the gap between it and its surroundings: dQ/dt ∝ ΔT. So the hotter sample loses heat faster at the start, while that gap is widest. The catch is that it also has more heat to get rid of on the way down to 0°C. Run the figures as though nothing else about the sample changes, and the colder one reaches freezing first every time. It is plain bookkeeping.

By that logic, hot water beating cold water to ice is impossible. Monwhea Jeng’s 2006 review of the effect points to where the logic leaks: it assumes the water is completely described by one number, its temperature. Burridge and Linden make the same point: for the effect to happen, the samples must differ in something besides temperature.

Heating can drive off some of the water by evaporation and lower the amount of dissolved gas. It can also set up stronger convection currents, change the temperature at which ice starts to form, and melt the frost under a container. The hot sample put in to cool is not the cold sample with extra joules added. It has become a system of its own, and once two different systems are racing, the one the equation favours can still come second.

None of these rules out the others, and they are not independent: how much water evaporates, for example, depends on the shape of the container (Jeng 2006). That is what makes the results so touchy about conditions, and why no single mechanism has been accepted as the full answer.

The evidence for it starts with Mpemba and Osborne’s 1969 paper, which reported a series of measurements. But its authors called the experiments “relatively crude” and said more sophisticated ones were needed.

Brownridge (2011, American Journal of Physics) pushed it further. He cycled a pair of sealed vials through 28 freeze-thaw runs, and the hotter one froze first every time. The catch is that he had picked those vials because the temperatures at which they froze differed by about 5.5°C. So the two samples were not identical to begin with, and the supercooling gap is doing the work.

Burridge and Hallstadius (2020) got a repeatable effect by roughening the inside of the hot sample’s container with sandpaper, while leaving the cold one smooth. Roughening gives ice more places to start forming. Both results depend on something besides temperature differing between the samples: the temperatures at which Brownridge’s vials froze, and the surface of Burridge and Hallstadius’s container.

A 2023 paper in InfoMat, by Tang and colleagues, is titled “Direct observation of the Mpemba effect with water”. Its authors report that the effect appears only under specific conditions, which depend on the starting temperatures, the container’s shape and the volume of water.

Related speed-ups have been reported outside water: in tetrahydrofuran clathrate hydrate, but only at lower starting temperatures (2016), and in computer simulations of granular fluids (2017). Others have been reported in a colloidal system immersed in water (2020) and in a single trapped ion (2025). Kumar and Bechhoefer, who ran the colloid experiment, argue that the effect is not just a curiosity of water but “the prototype for a wide range of anomalous relaxation phenomena”. None of these systems is water freezing, so none of them shows that hot water freezes faster.

The case against it starts with Burridge and Linden (Cambridge, Scientific Reports, 2016). They ran carefully controlled experiments, defined the effect as reaching 0°C first, and could not observe it at all.

They showed that nearly all earlier positive results were marginal, with the hot sample only just ahead, and could have come from errors in where the temperature was measured. Misplace the thermometer vertically by as little as a centimetre and you can manufacture an apparent effect out of nothing. That is because the water column isn’t one temperature top to bottom.

The exception was Mpemba and Osborne’s 1969 data, which showed a far larger gap. No study in the 47 years since had reproduced it, and Burridge and Linden judged it erroneous unless it can be repeated.

Their wider point is the harder one to dodge. Pin freezing to 0°C, control the rest, and the textbook prediction, that the colder sample gets there first, holds up fine. They accepted that single runs had shown warm water cooling and freezing first. What had not been shown, they wrote, was water doing so consistently, by a repeatable and statistically significant margin, simply because it started hotter.

The defensible answer is a narrow one. Hot water freezing first is neither a dependable rule nor, as the thermodynamic argument has it, impossible.

Auerbach’s 1995 experiments gave different odds at different temperatures. As Jeng’s 2006 review reports them, a hot sample froze before a cold sample in 53% of comparisons with the surroundings at −5 to −8°C. With the surroundings at −8 to −11°C, it did so in 24%. With so few samples, Jeng notes, 53% is not significantly more than half.

No one mechanism covers every case. Jeng’s review found it clear only that evaporation can play an important role and that a sample’s history can change how far it supercools. Beyond that, it says, the experiments “paint a very muddled picture”.

Sixty years of modern work, on top of centuries of scattered observation, still haven’t produced one accepted theory that covers every result. What is missing is a model that can say in advance which conditions will produce the effect, and how strongly. Whether hydrogen-bond restructuring plays a part is unsettled. So is the exact link between a container’s shape, where ice tends to start forming, and how often the effect shows at all.

The episode is a small study in how a field absorbs a result that supposedly can’t happen. Mpemba’s teacher gave the quick response in 1963: that it could not happen. The slower response is to state the claim exactly, strip out the confounding variables, rerun it carefully, and keep whatever still stands. Osborne did the slow version. What stood up did so in some tests and fell over in others, and that is more or less where the matter rests today.

It also shows what reproducibility means for a result this sensitive. When two labs use different container materials, different water, different freezer temperatures and different definitions of freezing, they are not running the same experiment. Firth, writing in Physics Education in 1971 and quoted in Jeng’s review, put it this way: “There is a wealth of experimental variation in the problem so that any laboratory undertaking such investigations is guaranteed different results from all others.”

The definitional problem belongs at the centre of the story, not in a footnote. The truthful answer depends on it. Change the moment that counts as frozen, and the count of cases changes too, as it did in the 2026 study, from 46 cases to 58. A piece that never says what it means by “freezing” hasn’t engaged the question.

The practical claims are folk beliefs. Ice-cream sellers in Tanga told Mpemba they put their mixture in the refrigerator hot, and the chemist G. S. Kell reported in 1969, as Jeng’s review quotes him, a Canadian belief that a skating rink should be flooded with hot water because it freezes more quickly. The experiments do not show that starting with hot water reliably makes ice faster in a home freezer.

Sometimes, but not reliably. Hot water can beat cold water to ice in a single run. In Janni and colleagues’ 2026 study of 176 measurements, it did so 46 or 58 times, depending on the definition. The authors traced those cases to chance variations in supercooling and in the freezer’s air currents.

Burridge and Linden’s 2016 analysis found that water which started hotter generally took longer to cool to 0°C, and their own controlled experiments showed no effect. The set-ups that produce it every time change something else as well: Brownridge chose vials that froze at different temperatures, and Burridge and Hallstadius roughened the hot sample’s container.

Download the conclusion

The verdict above as one image, with the 2026 study’s counts. The bars show how many of 176 measurements the hot sample won: 46 counting which reached 0°C first, and 58 counting which started to freeze first. The image is 1,200 by 630 pixels, the size social networks use for link previews, and anyone may download it and post it.

Download the card (PNG, 1,200 × 630)