The Verification Venue · a thing everyone repeats, checked
The Effect That Melts the Moment You Measure It Carefully
Hot water freezing faster than cold, the Mpemba effect, is served everywhere as a proven fact with a tidy cause. Put it under careful control and it dissolves. The answer is not "yes," and it is not "debunked": it is contingent, and it turns on a choice nobody tells you they are making, what the word frozen means.
Here is the number that should end the confident version. In the tightest controlled study, Burridge and Linden (2016, Scientific Reports 6:37665) measured the time for water to reach 0°C with thermometers matched in height. The Mpemba effect reproduced 0 times. Their killer finding: a thermometer misplaced by even about 1 cm in height was enough to manufacture false Mpemba evidence, because the temperature at the instant you compare depends on where in the beaker you probe. The one historical dataset too large to blame on that error is Mpemba and Osborne's own 1969 experiment, a single result never robustly replicated. So: for a glass of water the popular "established fact" is not established. Below you can operate the physics and watch exactly where the confident answers come apart.
Under your current definition of “frozen”
–
–
Hot reaches 0°C at
–
–
Cold reaches 0°C at
–
–
Curve gap never changes sign
–
zero-crossings of hot minus cold
The load-bearing control: what does “frozen” mean?
Mpemba and Osborne's 1969 samples ran hottest near 90°C.
Kept below the hot start. Their effect looked largest starting from 25°C.
Newton's law: warmer than ambient loses heat in proportion to the gap.
A home freezer sits near -18°C. Must be below 0 for anything to freeze.
Only bites under definition (b): boiling-hot water loses more mass, so it has less ice to make. At 1.0 the hot sample loses about half its mass cooling to 0°C.
The Burridge and Linden artifact: nudge the hot sample's probe up in a vertical gradient and it reads colder than the water is.
The 1 cm lie: how a misplaced probe fakes the effect
Burridge and Linden's central mechanism for prior "sightings" was not physics but measurement. Cooling water is not the same temperature top to bottom, and the moment you compare two samples you read whatever your probe is sitting in. Assume a representative near-surface gradient of G = 2 °C/cm. A probe placed Δh too high reads G·Δh degrees too cold, so the hot sample can cross the reported 0°C line while its water is still genuinely warmer than the cold sample's.
Drag Δh above, or hit Probe regime. The artifact is powerless when the two starts are far apart (the true gap at the crossing is tens of degrees) and devastating when they are close, which is exactly the regime real "Mpemba" experiments run in. That is the whole point of the 2016 paper.
The answer depends on a word: "frozen"
Twist 1: which finish line you draw decides the winner
Under Newton's law of cooling, the hotter sample can never reach 0°C first, and this is provable. Both samples obey T(t) = Tenv + (T0 − Tenv)·e−kt, so their difference is D(t) = (Th − Tc)·e−kt, which is positive for every finite time. The hot curve sits strictly above the cold curve forever: it must pass through every temperature the cold sample already holds, so it arrives at 0°C later, guaranteed. That is why the top readout says the curves cross zero times.
But redefine "frozen" and the guarantee evaporates, literally. Call it fully solid and add strong evaporation: the boiling-hot sample can shed enough mass that it has less ice to make, so even after taking longer to reach 0°C it can finish the latent-heat plateau first. Or call it onset of freezing via nucleation: real water supercools past 0°C to a random temperature before ice appears, and if the cold sample draws a deep-supercool ticket while the hot sample nucleates early, the hot sample can win the lottery. Flip the definition on the buttons above and watch the verdict line change with nothing else moving. That definitional fragility is precisely why two centuries of casual experiments, from Aristotle and Bacon and Descartes to every kid with an ice tray, disagree with each other.
Time to 0°C, well-mixed, no mass loss
The clean limiting case. Hot cannot win: zero crossings, a theorem, not a measurement. This is the baseline the simulator starts from.
Fully solid, or nucleation onset
Now evaporation (less mass to freeze) or the supercooling lottery can occasionally flip the winner. The page reports how often across repeated seeded runs, so you see the odds, not a slogan.
Twist 2: the effect is genuinely real, just not in your ice tray
Firewall these two regimes and never let them blur. Water in a freezer: contested, not robust, not settled. The general Mpemba effect as a non-equilibrium relaxation phenomenon: real, and actively researched. It concerns cooling rates and relaxation paths, not energy conservation, so the old line that it "violates the first law of thermodynamics" is simply false; modern theory shows a hotter state relaxing faster is thermodynamically allowed. Lu and Raz (PNAS 2017) predicted the effect and an inverse in Markovian systems, and Kumar and Bechhoefer (Nature 2020) demonstrated a "strong Mpemba effect" in a single colloidal particle, where a hot state relaxed exponentially faster than a merely warm one. None of that means your kettle water freezes first. It means the lazy "yes, proven fact" and the lazy "no, debunked" are both wrong.
Same water, three finish lines, three verdicts
This is the whole confusion in one table. Nothing here is hidden; the answer is contingent on a free choice the muddled search results never surface. Numbers are for the current sliders.
| definition of “frozen” | can hot win? | why / how often |
|---|
The check: every number recomputed in front of you
Nothing on this page is a stored figure. For your current sliders the page recomputes the cooling times, the no-crossing proof, the evaporation flip point, the seeded supercooling win-rate, and the probe artifact, live:
Free choices and uncertainty, named. The physics that hot cannot reach 0°C first under lumped Newton cooling is exact (a theorem about e−kt), and it is an idealization: well-mixed samples, identical environment, no mass loss. Everything that can flip the winner lives in modelling choices that are here to be seen and moved: the evaporation coefficient and the latent-plateau constant Λ set how much mass loss it takes to win definition (b); the nucleation temperatures are drawn Uniform[-15°C, -1°C] and the gradient is fixed at 2°C/cm. Change any of them and the win-rates change, honestly. The offline gate recomputes all of it, twice where it can: node research/does-hot-water-freeze-faster-than-cold/verify-does-hot-water-freeze-faster-than-cold.mjs.
What is exactly true here, and what is a stated model
Exactly true (the water verdict and the proof). In the tightest controlled study, Burridge and Linden (2016) found no statistically robust Mpemba effect for water once time-to-0°C was measured with height-matched thermometers, and showed that a probe misplaced by about 1 cm could fake it. Under lumped Newton cooling with identical environments and no mass loss, the initially-hotter sample provably cannot reach any lower temperature first: the curve difference (Th−Tc)e−kt never changes sign. The Mpemba and Osborne 1969 numbers are a single historical dataset, cited here as the effect's origin, never as confirmation.
Not "debunked," not "impossible." The careful reading is "fails to reproduce in the tightest controlled water study," not "proven false." The water case remains genuinely unsettled and actively studied. The Mpemba effect does not violate the first law of thermodynamics; it is about rates and relaxation paths, and non-equilibrium theory shows it is allowed.
A model, not a measurement (everything that can flip the winner). The latent-heat plateau, the evaporative mass-loss law massLost = E·T0/k (the integral of Newton cooling down to 0°C), the plateau constant Λ = 200 min per unit mass, the nucleation distribution Uniform[-15°C, -1°C], and the 2°C/cm gradient are a transparent, representative model built to show the shape of the answer: which finish line flips it and how often. The exact minutes and percentages are illustrative; the orderings (never under (a); sometimes under (b) or (c); artifact only when starts are close) are the robust content.
The honest firewall. "Real in colloids" (Kumar and Bechhoefer 2020) must never be read as "so your hot water does freeze first." Engineered single-particle systems and a beaker in a freezer are different regimes. This page keeps them apart on purpose.