The Mpemba Effect: Why Hot Water Sometimes Beats Cold to Freeze

Why does hot water freeze faster than cold water?

Physicists have been fighting about that one sentence for sixty years, and a lot of them think it's the wrong question. In a properly controlled freezer, hot water usually doesn't freeze faster. Except sometimes it does. A study published this year finally has a convincing explanation for why the effect shows up in some experiments and refuses to show up in others, and the answer has nothing to do with the water being hot or cold at all.

The ice cream tub that started a physics fight

Start with the name. In 1963, a Tanzanian secondary school student named Erasto Mpemba was making ice cream with his classmates and noticed his mixture, which he'd put in the freezer while it was still hot, froze before a classmate's mixture that started cold. His teacher told him he was wrong. Mpemba didn't let it go, and in 1969 he and physicist Denis Osborne published a paper, titled simply Cool?, laying out the effect for the scientific record. It wasn't actually new. Aristotle noted something similar more than two thousand years earlier. But Mpemba's name stuck to it.

Since then, researchers have chased the effect into water, ice cream, polymers, magnetic alloys, colloids, and even quantum systems, with wildly inconsistent results. Some labs reproduce it reliably. Others, running what looks like the identical setup, find hot water never beats cold water to the freezing point. A widely cited 2016 study out of Imperial College London went as far as arguing the whole thing was probably a measurement artifact.

The 2025 study that narrowed it down

A team at Russia's Institute of Protein Research, led by Andrei Klimov and Alexei Finkelstein, ran a tightly controlled version of Mpemba's original test this year and found the effect is real, but only inside a narrow temperature window, and only because of randomness, not physics that favors hot water. They cooled batches of water starting at 20°C and 70°C in a chest freezer, tracking each sample with its own temperature sensor, and varied how cold the freezer itself was set.

Above a freezer setting of about −7°C, the timing of when each sample actually started to freeze (not just cooled down, but formed its first ice crystal) became wildly unpredictable, varying by 500 to 1,000 seconds or more between otherwise identical samples. Below about −8°C, that variability nearly vanished. The reason, the researchers argue, is that ice doesn't form the instant water hits 0°C; it needs a nucleation event, a single molecule-scale trigger, and how long that takes to happen by chance grows exponentially as the freezer temperature creeps closer to the freezing point itself.

"We do not want to create the impression that our finding of the stochasticity of the Mpemba effect in freezing disproves the very existence of such effects."

Andrei Klimov and Alexei Finkelstein, Institute of Protein Research, Russian Academy of Sciences

Put plainly: when the freezer is set close to freezing, the wait for that first ice crystal is so unpredictable that it can swallow up the head start the cold sample should have. Sometimes the hot water, by pure chance, gets its first ice crystal first. Set the freezer colder, and that randomness shrinks to nothing. The cold sample wins every time, exactly as intuition says it should.

Why this matters beyond a kitchen trivia question

This is also why home tests of the Mpemba effect are so unreliable: a freezer door opening and closing, a slightly different ice tray, a stray impurity in the water can all nudge the result. The paper even flags one specific contaminant, a bacterium called P. syringae that's a potent natural ice-nucleator, as something that could tip an ordinary tap-water test toward or away from the effect depending on how much of it happens to be in the sample.

The finding fits into a bigger pattern in physics right now: Mpemba-style effects, where a system further from equilibrium reaches it faster than one that started closer, have turned up in trapped-ion qubits and nuclear spin systems over the past two years, and researchers have built a new mathematical framework, called thermomajorization, specifically to quantify when and why. The freezer version, oddly, may end up being the least clean example of the phenomenon, precisely because ordinary water freezing is so sensitive to chance nucleation events that the "true" underlying effect gets buried in noise.

None of this makes the original ice cream story wrong. Mpemba really did watch his mixture freeze first. He just picked, without knowing it, exactly the freezer conditions (close to the ice-nucleation threshold) where chance could make it happen. Try to repeat the experiment in a deep freezer set well below zero, and the effect that made him famous mostly disappears.