arXiv:physics/0512262v1 [physics.hist-ph] 29 Dec 2005 ohv enfrotnoc oemdr conceptions modern more it which once heat, forgotten of been have discussed an to repeatedly of was Ba- support effect Francis The in Bacon, Descartes. Roger and Aristotle, con, discussed others, was effect among The by, phenomenon. this on periments under- knowledge be any physics. without can advanced students important that of by of discussed method, wealth and a scientific stood up the bring consider- about will careful issues real, tasks is these effect experi- of the home ation not a or in Whether to that effect insists ment. Mpemba second, who the And non-scientist observed have a impossible. to they respond is pre- to as it how a explain, why decide to for try possible, pause to as First, should cisely things. thermodynamics two do of to laws moment possible. the fact by in den is it experi- that numerous see will in we observed and been sight.ments, first has at it impossible first. Nevertheless, theoretically freezes appears water effect hotter subzero The initially identical the to tem- and exposed higher are surroundings, a other, at the is than one that perature except way, every in identical eino neprmn.W e htteMeb effect Mpemba the that see We the experiment. in an theory of for sci- design need a the formulate and to carefully, need hypothesis the entific of illustration good a provides experi- of and interpretation can and the ments. acceptance understanding about our theoretical bias questions our should raises which to and observa-degree non-scientists, the of dismissing against tions Mpemba’s cautions Mpemba. story Erasto dramatic student, by community school of scientific secondary reintroduction modern a the the to describe phenomenon we that obser- this III, claimed early section incorrectly these In Kuhn replicate vations. that cannot see experiments will modern we fact, In nscinI,w ecieeryosrain n ex- and observations early describe we II, section In forbid- is effect the that certain quite are who Readers water, of bodies two when occurs effect Mpemba The nscinI,w e htta h pmaeetalso effect Mpemba the that that see we IV, section In hsc eatet o 64 otenIlni Universit Illinois Southern 1654, Box Department, Physics hl h ffc per mosbea rtsgt thsbeen has and it Descartes, sight, and Bacon, first Francis at Aristotle, by impossible on appears reported effect the While tde fteosrainta o ae ie r oelike more ex are modern pipes of water falsifiabil results hot described. the of that also observation and nature the effect, the of Mpemba and Studies the hypothesis, observati for scientific and mechanisms a experiment metho on of scientific theory ment of the influence about the culmi issues inquiry, descr which important to history, simple numerous reintroduced while fascinating trates who phenomenon, and The Mpemba, rich community. Erasto scientific a student, has school It secondary world. the erve h pmaeet hr nilyhtwtrfreezes water hot intially where effect, Mpemba the review We .INTRODUCTION I. appears incorrect ocnrdc,wr developed. were contradict, to hoyo et n appears and heat, of theory o ae a reefse hncold?!? than faster freeze can water Hot owe Jeng Monwhea r ol atog o yti aeutlfil re- fairly until name wrote Aristotle this B.C., by 350 Around not cently). (although world ern osblte o uueeprmnsta ol edn by done be discuss we could students. that VIII experiments section future in Finally, for of possibilities unclear. importance is possible role concluding its the ultimately that at effect, Mpemba look the to we in happened VII, supercooling has section what In “remember” it. can situation water clear the a provide where to they related because as closely effect, are supercooling Mpemba pipes the on target experiments and The effect, than cause. conclusive the Mpemba which more the been phenomenon, on have those VI, this section on in cold discuss adjacent Experiments we than freezing pipes. from burst water to likely more are for but confusing, modern are of which results reasons. effect, interesting the the discuss We on experiments important. may of be environment, also one surrounding all convection, the is of and cooling effects gasses, the Evaporative dissolved but been explanations, have strongest effect. impos- the explanations the Multiple is for effect flawed. proposed Mpemba fact the in that is sible, proof explain common We a section. and why this history to the straight mechanisms in jump uninterested can theoretical background readers possible and effect, discuss sec- the until we for not that is It V occurs. tion it and whether occurs, of effect regardless Mpemba indeed, the why of regardless hensible falsifiable. be must that thesis hypothesis of Popper’s scientific Analysis up a brings effect. naturally Mpemba problems experiments the common the on some experiments discuss lay and ap- with glance, might first it than at experimentally pear study to harder much is I XEIET EOETESCIENTIFIC THE BEFORE EXPERIMENTS II. fwtrhsbe rvosyhae,ti contributes this heated, previously been has water If h pmaeethsln enkoni h West- the in known been long has effect Mpemba The thssmtmsbe eotdta o ae pipes water hot that reported been sometimes has It compre- be should II-IV sections in discussions The ∗ dadvle dadvle L 62025 IL, Edwardsville, Edwardsville, y a enwl-nw sfllr around folklore as well-known been has n h edfrpeiini h state- the in precision for need the on, :terl fsetcs nscientific in skepticism of role the d: t.W uvypooe theoretical proposed survey We ity. b,i eetvl ope,adillus- and complex, deceptively is ibe, yt us hncl ae ie are pipes water cold than burst to ly eni ueoseprmns was experiments, numerous in seen ae ntedaai tr fthe of story dramatic the in nates h ffc otetetehcentury twentieth the to effect the atrta ntal odwater. cold initially than faster eiet ntephenomenon. the on periments REVOLUTION 1 2

to the rapidity with which it freezes: for it cools more middle of this debate, around 1461, Giovanni Marliani quickly. (Thus so many people when they want to reported on experiments, described here by Clagett5: cool water quickly first stand it in the sun: and the . . . To support his contention that heated water inhabitants of Pontus when they encamp on the ice freezes more rapidly [than cold], Marliani first points to fish. . . pour hot water on their rods because it to a passage in Aristotle’s Meteorologica affirming it. freezes quicker, using the ice like solder to fix their However, [Marliani] does not depend on Aristotle’s rods.) And water that condenses in the air in warm statement alone. He claims that not only has he of- districts and seasons gets hot quickly. ten tested its truth during a very cold winter night, Aristotle used this observation in support of antiperista- but that anyone may do so. You take four ounces sis, which is the “sudden increase in the intensity of a of boiling water and four ounces of non-heated wa- quality as a result of being surrounded by its contrary ter and place them in similar containers. Then the quality, for instance, the sudden heating of a warm body containers are exposed to the air on a cold winter’s when surrounded by a cold.”2,3 morning at the same time. The result is that the While the idea of antiperistasis today sounds ridicu- boiling water will freeze the faster. lous, with the hindsight of our modern understanding of The belief in the Mpemba effect continued strong into heat, energy, and temperature, it should be remembered the 17th century. Francis Bacon and Descartes both that Aristotle was working without these paradigms, and wrote extensive works on the scientific method, and ex- indeed, without a thermometer. The fact that ice re- periments, and both wrote about the Mpemba effect. In quires cold, yet hail comes in the summer, rather than the Novum Organum, Francis Bacon wrote6 the winter, requires an explanation—Aristotle’s explana- . . . water a little warmed is more easily frozen than tion was antiperistasis. Later, a number of medieval sci- that which is quite cold. . . entists used antiperistasis to explain the (apparent) facts that bodies of water are colder in the summer, and that And in 1637, Descartes wrote about this phenomenon in 2 Les Meteores, a work which was published attached to his human bodies are hotter in the winter . While we can 7 now explain these observations with our modern theory more famous Discourse on Method . He emphasized the of heat, the explanations are not obvious. The concept of importance of experiment, described how to analyze the temperature, and the zeroth law of thermodynamics, are density-dependence of water, and stated results about quite counterintuitive to anyone who has touched metal the freezing times: and wood, outside on a cold day. We can see this by experiment, if we fill a beaker— In the 13th century, well before the Scientific Revolu- or some other such container having a long, straight tion, Roger Bacon wrote Opus Majus, in which he argued neck—with hot water, and expose it to freezing cold repeatedly for the importance of experiments in science. air; for the water level will go down visibly, little He wrote4 by little, until the water reaches a certain level of Moreover, it is generally believed that hot water coldness, after which it will gradually swell and rise, freezes more quickly than cold water in vessels, and until it is completely frozen. Thus the same cold the argument in support of this is advanced that con- which will have condensed or shrunk it in the begin- trary is excited by contrary, just like enemies meet- ning will rarefy it afterwards. And we can also see ing each other. But it is certain that cold water by experiment that water which has been kept hot freezes more quickly for any one who makes the ex- for a long time freezes faster than any other sort, periment. People attribute this to Aristotle in the because those of its parts which can least cease to second book of Meteorologics; but he certainly does bend evaporate while it is being heated. (Emphasis not make this statement, but he does make one like added.) it, by which they have been deceived, namely, that if A modern writer on Descartes has commented on the cold water and hot water are poured on a cold place, italicized statement: “This statement, which the sim- as upon ice, the hot water freezes more quickly, and plest of experiments could have refuted, was repeated this is true. But if hot water and cold are placed in with elaborate details in a letter to Mersenne, and it em- two vessels, the cold will freeze more quickly. There- phasizes Descartes’ readiness to rely ona ` priori conclu- fore all things must be verified by experience. sions”8. But this modern writer’s position is contradicted What is particularly interesting about this is that by the letter to Mersenne, in which Descartes makes clear Roger Bacon agrees that hot water can, under some cir- that he has, in fact, done this experiment. In this 1638 9,10 cumstances, freeze faster than cold water, but argues that letter, Descrates wrote : specification of the precise experimental conditions is im- I appreciate once again what you have written me portant. We will see that this is a crucial observation, that my reputation is at stake in my response to Mr. equally important in discussions about modern experi- Fermat, in which I assure you that there is not one ments on the Mpemba effect. single word that I would like to have changed. . . In the Middle Ages, debates raged over whether ob- I dare to assure you that there is nothing incorrect, jects could only be cooled by extrinsic sources, or whether because I did these experiments myself, and particu- some objects might be able to cool themselves. In the larly the one which you commented on of the hot wa- 3

ter that freezes more quickly than cold; where I said to think about. not hot and cold, but that water that one has held The Mpemba effect illustrates the points raised by for a long time over the fire freezes more quickly than Kuhn. In modern times, because the Mpemba effect ap- the other; because in order to correctly do this ex- pears to contradict modern theories of heat, scientists are periment, one must first have boiled the water, then much more skeptical, or even dismissive, of experiments let it cool off, until it has the same degree of cool- that observe the Mpemba effect. Furthermore, like the ness as that in a fountain, and having tested it with eighteenth century experiments with pan balances, ex- a thermometer, then draw water from that fountain, periments on the Mpemba effect, for reasons we will dis- and put the two waters in the same quantity in same cuss, yield “neither consistent nor simple results.” Thus, vases. But there are few people who are capable of the Mpemba effect, while interesting, is a factual curios- correctly doing these experiments, and often, in do- ity, and not fundamental to our modern understanding ing them poorly, one finds the complete opposite of of heat. what one should find. (Emphasis in original) In The Structure of Scientific Revolutions, Kuhn As with Roger Bacon’s earlier experiment, we again see briefly mentions these experiments by Marliani and Ba- 14 that the details of the experiment are crucial. Descartes con : is not measuring the time for the hot water to freeze, but . . . the natural histories often juxtapose [correct] is saying that when water has been heated, it is some- descriptions. . . with others, say, heating by an- how changed so that it cools more easily, even after being tiperistasis (or by cooling), that we are now quite brought back to room temperature. While the observa- unable to confirm. tion described is different than our modern statement Kuhn does not cite any experimental evidence that we of the Mpemba effect, it is similar in that some sort of are unable to confirm these older results, and one sus- history-dependence (i.e. memory) of the water is implied pects that he is simply assuming this point15. In fact, as by the results. Descartes’s letter also indicates that both we will see, at the time that Kuhn wrote this, there were he, and others, have done this experiment, and that the multiple papers confirming the existence of the Mpemba results are contradictory, a problem that we will also see effect. Kuhn thus unintentionally, and ironically, demon- in more modern experiments. strates how our theoretical expectations can color our With the advent of the modern theory of heat, these experimental beliefs. earlier observations were forgotten. Since these experi- ments appear to contradict what we know about heat, it is natural to dismiss them as mistakes. III. ERASTO B. MPEMBA AND 20TH Presentations in textbooks typically show the progress CENTURY KNOWLEDGE of science as a simple, straight-line progression, with ex- periments pointing in a straightforward and unambigu- This strange phenomenon was reintroduced to the ous manner to new scientific theories. But, as Kuhn has modern scientific community by Erasto Mpemba, a sec- pointed out, the development of scientific theories is not 11 ondary school student in Tanzania, in 1963. Mpemba so simple . Most of the time, scientists are engaged in told his story in Physics Education16. Mpemba and his what Kuhn calls “normal science,” during which research fellow students were making ice cream, which used a mix- “. . . is a strenuous and devoted attempt to force na- ture that included boiled milk. Because excessively hot ture into the conceptual boxes supplied by professional 12 objects could damage the refrigerator, they were sup- education. . . .” When scientists are working under a posed to let their mixture cool before putting it in the re- certain paradigm, results that cannot be forced into the frigerator. However, space in the refrigerator was scarce, existing paradigm may be ignored, as attention is focused and when another student put his mixture in without on experiments that appear more promising for advanc- 13 boiling the milk, Mpemba decided to put his hot mix- ing knowledge. To give but one of Kuhn’s examples : ture in, without waiting for it to cool. Later, Mpemba In the eighteenth century, for example, little atten- found that his hot mixture froze first. tion was paid to the experiments that measured elec- Mpemba asked his teacher for an explanation, but trical attention with devices like the pan balance. his teacher said Mpemba must have been confused. Because they yielded neither consistent nor simple When the teacher later covered Newton’s law of cool- results, they could not be used to articulate the ing, Mpemba persisted in his questioning. The teacher paradigm from which they derived. Therefore, they responded that “Well, all I can say is that is Mpemba remained mere facts, unrelated and unrelatable to physics and not the universal physics,” and from then the continuing progress of electrical research. Only on the teacher and the class would mock his mistakes in retrospect, posessed of a subsequent paradigm, by saying “That is Mpemba’s mathematics,” or “That is can we see what characteristics of electrical phenom- Mpemba’s physics.” ena they display. Mpemba ran more systematic experiments, both with Whatever one’s opinion on Kuhn’s more controversial water and milk, and continued to get similar results. theories, it is clear that scientists’ theoretical views in- When Dr. Osborne, a professor at a nearby university, fluence what experiments they choose to do, to trust and visited Mpemba’s school, Mpemba asked him why water 4 at 100OC froze faster than water at 35OC. Upon return- delight, the hotter one was indeed the first to form ing to his university, Dr. Osborne asked a technician to a layer of ice on the surface. test Mpemba’s question. When Dr. Osborne later asked He was, to say the least, no longer quite so im- for the results, “The technician reported that the wa- pressed by my capabilities as a Science Teacher. ter that started hot did indeed freeze first and added in The skepticism with which scientists react to the a moment of unscientific enthusiasm ‘But we’ll keep on Mpemba effect is quite common. In this author’s ex- repeating the experiment until we get the right result’ ” 17 perience, scientists are much more likely to react with (emphasis added) . But future experiments gave sim- disbelief than laypersons; this is not surprising, since sci- ilar results, and Mpemba and Osborne later published 16 entists know enough to “know that this is impossible.” their results . In the same year, Dr. Kell of Canada Mpemba’s story provides a dramatic parable against independently reported the phenomenon, along with a 18 dismissing the observations of non-scientists. But his theoretical explanation that we will consider later . story is particularly interesting because it is more than Subsequent discussions in journals showed that the ef- just a story of bad, close-minded, scientists. There is, of fect was already known and believed by non-scientists course, no excuse for the Tanzanian teacher’s mockery of in diverse regions of the world. Kell stated that it was Mpemba. But is it “unscientific” for scientists to, upon widely believed in Canada, and that hearing of the Mpemba effect, immediately suspect er- Some will say that a car should not be washed with rors in the experiment? Kuhn emphasizes that with the hot water because the water will freeze on it more operation of “normal science,” scientists interpret exper- quickly than cold water will, or that a skating rink iments in light of the reigning paradigm—i.e. with their 11 should be flooded with hot water because it will preexisting theoretical understanding . freeze more quickly. What is interesting about the Mpemba effect is that unlike the examples commonly given in science text- Mpemba reported that after his initial experience, he books, where theory and experiment march hand-in- found that ice cream makers in Tanga City, Tanzania, hand, always leading to further progress, here theory used hot liquids to make the ice cream faster. British (rightly or wrongly) prevents acceptance of experiment. letters to the New Scientist reveal a wealth of lay ob- We are certainly not arguing that the reaction of the sci- servations. One writer stated that it was well-known entific community to surprising experimental results is that in the winter, hot water pipes were more likely to 19 arbitrary, or necessarily hostile. Our point is simply that freeze than cold water pipes (an issue we consider in the reaction to an experiment depends significantly on section VI). Another stated that the phenomenon was 20 how well that experiment matches accepted theoretical well-known in the food-freezing industry . A number preconceptions. Because experimental claims can be in of letters reported that their (non-scientist) friends and 21,22,23 error, scientists do not just accept all published claims. family had known of this effect even in the 1920’s . While few scientists would find this claim controversial, One writer was a science teacher, whose story closely par- it is quite different than the impression one gets from sci- allels Mpemba’s. The teacher describes his experiences ence textbooks, and from what appears in certain pos- with a student who asked him why hot water froze faster 24 itivistic views of science. The Mpemba effect provides than cold a lovely case for considering these issues, because while . . . I told him that it was most unlikely, but he it provokes skepticism, it has been observed in multiple replied that he had seen it happen when his mother experiments; yet, in support of the skeptical position, threw out her washing water onto the path. I ex- we will see that the experimental results are not entirely plained to him that the particles in the hot liquid consistent, and that the theoretical situation is still un- would be escaping more readily to the atmosphere settled. due to evaporation and that this would leave a thin- ner layer of liquid to freeze than in the colder one where the particles would be escaping more slowly. IV. WHAT IS THE QUESTION, AND IS IT He was not, however, convinced, and a few days later SCIENTIFIC? he reported that he had placed two cans, one con- taining hot water, and the other cold water, outside To analyze the Mpemba effect, we first need to pre- and that the hot one was still the first to freeze. cisely state what we are trying to test. At first sight, the Finally, he challenged me to explain that one, if I question is quite simple: “Does hot water freeze faster could. Still in no doubt I criticized his experiment than cold?” However, a little thought shows that this and suggested he attempt a more accurate one under formulation is not adequate. Clearly, a small drop of hot laboratory conditions. water can freeze faster than a cold ocean. Hot water in a He obtained two identical specimen jars and freezer will freeze faster than cold water on a warm day placed hot water (approx 50O C) in one and colder (as the latter will not freeze at all). These examples are water (approx 20O C) in the other; both tops were silly, but illustrate the need to state the question clearly. left off and they were placed in the freezing com- A better second attempt at stating the problem might partment of a refrigerator. To my disbelief, and his be “Given two bodies of water, which are identical in all 5 parameters (mass, shape, surroundings, etc. . .), except the process with my brand new Swatch watch, whose that one is initially at a higher uniform temperature than precision and smart styling have made it the number the other, the hotter water will freeze first.” But further one choice of scientists the world over. thought shows that this cannot be correct. If the initially I subsequently did the same with two trays of cold hotter water is at 99.9999OC, and the initially colder water, which had been chilled down to a starting water is at 0.0001OC, then the initially cold water is just temperature of 38 degrees. seconds away from freezing, and it is clear that the hot The results? The cold water froze about 10 or water cannot possibly overtake it. Furthermore, there is 15 minutes faster than the hot water, and there was clearly no reason to expect the Mpemba effect to occur no detectable difference between the boiled water for all possible initial parameters. and the other kind. Another old wives’ tale thus So a better phrasing might be “There exists a set of emphatically bites the dust. Science marches on. initial parameters, and a pair of temperatures, such that These discussions fail to appreciate that a single test given two bodies of water identical in these parameters, cannot show that the effect never occurs for any param- and differing only in their initial uniform temperatures, eters and initial temperatures. the hot one will freeze sooner.” This is a much better Further consideration of this point brings up another statement of the question, although we will see later that issue. Logically, our statement about the Mpemba effect deficiencies remain. can never be proven false. Regardless of how many exper- Once the Mpemba effect is properly stated, it is clear iments Cecil Adams runs, a true believer in the Mpemba that we are only looking for some set of parameters, effect can always claim that the effect occurs for other such that if we plot freezing time versus initial tempera- sets of initial parameters, that differ slightly from the ture, there is some range of the graph that is downwards ones that Cecil Adams used. Popper has famously ar- sloping—not that there exist initial parameters such that gued that the hallmark of a scientific hypothesis is that the graph is downwards-sloping all the way from 0OC to it be falsifiable27. That is, not that it can be proven 100OC, or that the graph has downwards-sloping parts true, but that it can be proven false. Is our most recent for any set of initial parameters. This is logically nec- statement of the Mpemba effect thus unscientific? essary for the problem to be at all reasonable, but this It is not unusual that a scientific phenomenon, strictly point is not always appreciated in popular discussions. speaking, cannot be proven impossible, because the pa- Consider the discussion of the Mpemba effect in Ann rameter space in which it might occur is, in principle, Landers’ column, as described in Myth-Informed25 infinite. However, if we search a “representative sam- [Does] warm water [freeze] faster than cold water. . ple” of the parameter space over which the phenomenon . The redoubtable Ann Landers got into an ongo- might be thought to occur, and the phenomenon is not ing row with readers in 1983 when she addressed seen, that would be fairly convincing evidence against it. this question, as well as the related cosmic issue We thus need a list of experimental parameters that of whether cold water boils faster than hot water. we might vary when studying the Mpemba effect. The She “went to the top” and consulted Dr. Jermore list is rather long (infinite). On a first page, we might Weisner, chancellor of the Massacusetts Institute of list the mass of the water, the shape of the container, the Technology, who kicked the problem over to the MIT surrounding environment of the refrigerator, and the gas dean of science, Dr. John W. Deutch. Landers never content of the water. Note that several items on this list recorded what Deutch thought of being given such a are not single parameters—characterizing the shape of problem by an advice column, but the eminent scien- the container requires multiple parameters. We may also tist reported “Neither statement was true.” Where- want to include boolean parameters, such as whether the upon “Self-Reliant in Riverdale”. . . upbraided her container has a lid. On a second page, we might list the for using “argument by authority” rather than doing color of the container, the electrical conductivity of the her own experiment. “Self-reliant” said she reached walls of the refrigerator, the gender of the experimenter, the same conclusion as Deutch by using a pan of hot etc. . . If we simply list all the parameters we might water, a thermometer, a stove, a refrigerator, and a think of, the list is infinite, and we are at a complete watch with a second hand. loss as to how to proceed with an experiment. Without In his popular column, The Straight Dope, Cecil Adams a theoretical framework in which to design and conduct also discussed the Mpemba effect26: the experiment, we are reduced to randomly collecting . . . I carefully measured a whole passel of water facts, such as the color of the container, and the situation into the Straight Dope tea kettle and boiled it for is impossible. However, we have strong theoretical rea- about five minutes. This was so I could compare the sons for ignoring all the parameters on the second page, freezing rate of boiled H2O with that of regular hot and the experimenter can safely ignore the color of the water from the tap. (Somehow I had the idea that container, neither varying, nor recording, its color. water that had been boiled would freeze faster.) Unfortunately, as we will see in next section, all the Finally I put equal quantities of each type into items of the first page can plausibly be considered im- trays in the freezer, checked the temp (125 degrees portant for the Mpemba effect. Furthermore, their ef- Fahrenheit all around), and sat back to wait, timing fects are not independent of one another. But an exper- 6 imenter can hardly be expected to establish a vast mul- The problem with this proof is that it implicitly as- tidimensional array of containers of different dimensions sumes that the water is completely characterized by a and shapes, independently varying masses, gas contents, single parameter—the temperature. We need to think of and refrigeration methods. a parameter that might change during the course of the O O We are, of course, not claiming that scientific investi- experiment; then, the 70 C water cooled to 30 C will O gation of the Mpemba effect is impossible. But a com- not be the same as the water initially at 30 C. mon response, upon first hearing of the Mpemba effect, One possible parameter is the mass of the water. Both is that it should be a straightforward matter to study bodies of water initially have the same mass. But if the experimentally. But for even this (deceptively) simple initially hotter water loses mass to evaporation, then the problem, productive experimental design requires at least 70OC water cooled to 30OC will, having less mass, be some theoretical understanding of why the effect might easier to freeze (i.e. less energy will need to be removed occur—otherwise we will not know whether we should be to cool and freeze it). This is one of the strongest theo- paying attention to, for example, the gas content of the retical explanations for the Mpemba effect. Kell numer- water. We will see in the next section that because the ically integrated the heat loss equations, assuming that time to freezing is sensitive to all the “first page” param- the cooling was by evaporation alone, and that the mass eters, the experimental results become very confusing. lost to evaporation never recondensed—-he found that Our statement of the Mpemba effect now reads “There with these assumptions, there were initial temperatures exists a set of initial parameters (water mass, gas con- where the hot water would freeze faster than the cold wa- tent of water, container shape and type, and refrigeration ter18. But this does not prove that evaporative cooling method), and a pair of temperatures, such that given two is the only factor behind the Mpemba effect. A number bodies of water identical in these parameters, and differ- of experimenters have claimed that the amount of mass ing only in their temperatures, the hot one will freeze they lost to evaporation was insufficient to explain their sooner.” One final difficulty that must be considered is results28,29,30. And Wojciechowski et. al. observed the how to define the time of freezing—are we considering Mpemba efect in a closed container, which further sug- it frozen when ice crystals first appear, or only when gests that evaporative cooling is not the sole cause of the the entire body of water is frozen? Or, to simplify the effect31. experiment, we might just measure the time until some A more complex “parameter” is the temperature dis- specified part of the water reaches 0O C. This might seem tribution of the water. As the water cools it will develop like a minor issue, but we will see in section VII, on su- convection currents, and the temperature will become percooling, that it is potentially crucial. non-uniform. This also defeats the impossibility proof above, since the water is no longer characterized by a single number. Analysis of the situation is now quite V. HOW COULD THE MPEMBA EFFECT complex, since we are no longer considering a single pa- OCCUR? rameter, but a scalar function, and computational fluid dynamics is notoriously difficult. Nevertheless, at least one general point can be made. For temperatures above We have deliberately avoided discussing the theoreti- O 4 C, hot water is less dense than cold water, and will cal explanations for the Mpemba effect until now. We thus rise to the top. So we can generally expect that when have done this both because the historical reaction to O the 70 C water has cooled to an average temperature of the Mpemba effect is only comprehensible in light of the O O 30 C, the top of the water will be hotter than 30 C, effect’s apparent inconsistency with modern conceptions O and the bottom of the water will be below 30 C. If the of heat, and to emphasize the need for an experimen- water primarily cools at its surface, the non-uniform dis- tal framework when designing experiments on the ef- O tribution with an average temperature of 30 C will thus fect. Here, we discuss some proposed mechanisms for O lose heat faster than uniformly 30 C water. Consistent the Mpemba effect, but will not attempt to analyze their with this, Deeson found that gentle stirring substantially relative plausibility in depth. raised to time to freezing32. Convection could work in To see how the effect might occur, it is useful to care- concert with other factors, such as evaporative cooling. fully think about why the effect appears impossible. The Convection currents are sensitive to the shape and di- careful reader may already have come up with a proof of mensions of the container, so this explanation may play impossibility that goes something like this: very different roles for different container shapes. Suppose that the initial temperatures for the hot This brings up the question of what “parameters” and cold water are 70OC and 30OC. Then the 70OC should be associated with the surrounding air. Modeling must first cool to 30OC, after which it must do ev- the cooling process should take into account the convec- erything the 30OC water must do. Since the 70OC tion currents of the air, which will depend on the shape water has to do everything the 30OC water must do, of the refrigerator. Firth’s studies of the Mpemba effect plus a little more, it must take longer to freeze. found that this was an important factor33: A good way of systematically analyzing the Mpemba ef- What these experiments have shown, however, is fect is to think about why this proof is wrong. that it is not just the beaker or the initial tempera- 7

ture of its contents that are important, but that its vestigations is guaranteed different results from all environment probably influences the cooling rates to others. a greater extent than any aspect of the beaker itself. In figure 1 we see experimental results by Walker, On a related note, one letter writer to The New Scien- showing the dependence of the time of freezing on the tist suggested that the Mpemba effect could be explained initial temperature, under various initial conditions. We if the containers of water were sitting on layers of frost. see that some graphs show a strong Mpemba effect, some He argued that the frost conducts heat poorly, and the only show a weak one, and some show no Mpemba effect hot water causes the layer of frost to melt, establishing at all. This indicates that the cooling is indeed sensi- better thermal conduct with the refrigerator floor34. This tive to a number of parameters. Furthermore, Walker may explain some everyday observations of the effect, but reported that while his results were mostly repeatable, the experiments published in refereed journals generally he “still obtained strange large deviations on some of the used containers on thermal insulators. results.”35. Another possibility is that the hot and cold water, while they appear identical to the naked eye, differ in their gas content. Hot water can hold less dissolved gas VI. SUPERCOOLING AND BURSTING WATER than cold water, and the gas content affects the proper- PIPES ties of the water. Mpemba and Osborne’s original exper- iments were done with recently boiled water, to remove It has often been said that hot water pipes burst 16 35 dissolved air , as were the experiments by Walker . from freezing more often than adjacent cold water These experiments suggest that dissolved gasses are not pipes19,36,37,38. This is different from the Mpemba ef- necessary to the Mpemba effect. (Although under typi- fect, but it is similar to it, in requiring the water to have cal conditions, the degassed water will slowly regain dis- a “memory.” The mechanism behind the bursting water solved gasses from the atmosphere, confusing matters.) pipes is better understood than that behind the Mpemba On the other hand, Freeman only observed the Mpemba effect. 30 effect when carbon dioxide was dissolved in the water . The water pipe claim was first investigated by F. C. Similarly, Wojciechowski et. al. only saw the effect for Brown, in 191636. He confirmed the claim by taking 100 31 non-degassed water . A number of explanations of have glass test tubes, and filling 50 with tap water, and 50 been proposed for how the amount of dissolved gas could with tap water that had been boiled. After allowing all affect the properties of the water, and thus cause the water to first reach room temperature, he placed them Mpemba effect, although they remain largely specula- outside in subzero temperatures. He found that 44 of tive. One of the few quantitative findings was by Woj- the tubes with the boiled water burst, while only 4 of ciechowski et. al., who reported that for water saturated the tubes with non-boiled water first. Since all water with carbon dioxide, the enthalpy of freezing was smaller was at the same temperature when placed outside (as for the initially warmer water (but that preheating was with Descartes’s experiment), and he was looking at the irrelevant to the enthalphy if dissolved gasses were ab- occurence of bursting, rather than the time until freezing, 31 sent) . We will return to the effects of dissolved gases, these are not tests of the Mpemba effect. and other impurities, when we discuss supercooling. Freezing water will generally supercool. Supercooling All the factors discussed are at least potentially im- to −4OC to −6OC is common, and much greater super- portant in explaining the Mpemba effect. What makes cooling can occur for small samples37,39. Once freezing the situation particularly difficult to analyze is that the starts, the ice-water mixture must go to 0OC. So when factors are not independent of each other—for example, freezing begins, a finite fraction of the water must lose rates of evaporative cooling will depend on the shape of energy, and turn to ice, transferring heat to the remain- the container. The experimental results described here ing subzero water, whose temperature will rise to 0OC. do not all point to a single, clear, conclusion, and the Thus, the more supercooling occurs (i.e. the lower the reader who further inspects the original papers will find temperature at which freezing begins), the larger the vol- more facts, but will not find the overall picture much ume fraction of water that will freeze initially (i.e. the 16,30,31,32,33,35,41 clearer . larger the fraction of H2O molecules that will be in ice Because there are so many factors that can be varied, structures). Also, a certain volume fraction of ice will not and the results of the experiments can depend sensitively always correspond to the same volume of region spanned on any of these factors, experimental results are varied by the ice. The ice will form a dendritic structure, inter- and difficult to organize into a consistent picture. (Recall spersed with liquid. If more supercooling has occurred, Kuhn’s statement about 18th century pan balance exper- more of the water will have reached subzero tempera- iments, quoted in section II.) It is thus unclear which of tures, and the dendritic structure will span a larger re- these explanations is the explanation, or indeed, whether gion, for a fixed volume fraction of ice. it is appropriate to look for a single explanatory factor, Brown observed that the water that had been boiled 33 isolated from the other factors. As Firth wrote , first, later supercooled, while the nonboiled water did There is a wealth of experimental variation in the not, and argued that this difference was responsible for problem so that any laboratory undertaking such in- the difference in bursting behavior36. He argued as fol- 8

FIG. 1: Dependence of time of freezing on initial temperature, for various experimental conditions: (a) 50 ml in small beaker (b) 50 ml in large beaker (c) 50 ml in large beaker in frost-free freezer (d) 100 ml in large beaker, thermocouple near bottom (e) 100 ml in large beaker, covered with plastic wrap, thermocouple near bottom (f) 100 ml in large beaker, thermocouple near top. Graph produced from data obtained by Walker35. lows: If water in a pipe freezes near 0OC, only a small found that the more supercooling, the greater the pres- amount of ice will be formed initially. This ice will be sure gradient needed, for the same amount of total ice localized to the coldest regions, at the sides of the pipes, formed (figure 2). He concluded, as Brown had, that leaving a hole in the center. With further cooling, the with more supercooling, the ice formed would be more hole will shrink, but until all the water has frozen, liquid likely to form a dendritic structure, spanning the pipe, water will still be able to flow through the hole. Further- and causing blockage. He also confirmed this picture with more, the water flow can break away the ice on the sides, photographs of the cross-section of the pipe during the relieving pressure. On the other hand, if the water su- freezing process. percools significantly before freezing, a larger structure of These experiments convincingly demonstrated that dendritic ice will form throughout the pipe—while only greater supercooling would lead to burst pipes, leading a finite fraction of the water will be turned to ice, this to the question of why initially hot water would super- dendritic ice will span more of the pipe, possibly leaving cool more than initially cold water. Brown argued it no hole, thus blocking the flow of water, and resulting was because boiled water had less dissolved gas, and in a burst pipe. Consistent with this explanation, Brown that the dissolved gas prevented supercooling36. How- found that the ice rose higher from expansion in the tubes ever, Dorsey, who carried out an extensive series of ex- with nonboiled water, indicating the existence of a cen- periments on the factors affecting supercooling, over the tral region in which the water was mobile, and the ice course of ten years, disputed this37,40. Dorsey found that able to push up. dissolved gasses were not a significant barrier to super- In 1977, Gilpin performed quantitative experiments cooling; he also pointed out that, unlike boiled water, that confirmed Brown’s qualitative explanation38. Gilpin the water in hot water pipes does contain significant exposed pipes to subzero temperatures, and measured amounts of dissolved gas. Dorsey agreed that heated wa- the pressure gradient necessary to induce the flow of ter supercooled more, and that this would result in burst water, at various times after the ice had formed. He pipes, but argued that the greater supercooling occurred 9

Flows in pipes (from Gilpin) complicated by the randomness of supercooling, and mul- tiple trials will be needed to get precise average times to 120 freezing. For simplicity, many experiments have studied the time for some specified location in the water to reach O 30,33,35 100 0 C . Supercooling is irrelevant for these exper- iments, if the specified location is at or near the place that first reaches 0O C. 80 Auerbach considered the relevance of supercooling to the effect41. He found that initially hot water would su- 60 percool less (measuring the time to the first appearance of ice crystals) than initially cold water. Auerbach did

40 not determine why this happened, but pointed out that

Pressure gradient (kPa/m) gradient Pressure the initially hotter water should have greater tempera- ture shear, and that this shear is known to trigger crys- 20 tallization42. However, Auerbach’s finding that heated water supercools less than nonheated water is directly

0 opposite to the findings of Brown and Dorsey, described 0 0.05 0.1 0.15 0.2 in the previous section. Furthermore, Auerbach had a Volume fraction of ice relatively small number of trials, so the significance of <2.5 °C 2.6-2.7 °C his results is unclear. 2.9-3.3 °C 4.0-5.0 °C While Auerbach’s result is in the correct direction to explain the Mpemba effect (since if the hot water su- FIG. 2: Effect of nucleation temperature on pressure gradient percools less, this will help it freeze sooner), he actually required to start flow through pipe. Each of the four sets of found that the initially hot water actually took longer to data points are for different amounts of supercooling. Graph freeze on average (due to the greater time the hot water O produced from data obtained by Gilpin38. Curves are power took to reach 0 C)41. law fits, and are guides to the eye only (no theoretical justi- Given this, it is unclear whether Auerbach’s ex- fication.) periments can be described as actually observing the Mpemba effect. Due to the random fluctuations in the times till freezing, Auerbach found that the hot water because heating served to deactivate or expel “motes” might freeze first by chance. He found that when the am- 37 O O (what we would today call nucleation sites) . Both ex- bient temperature was −5 C>Ta > −8 C, the proba- planations agree that some sorts of nucleation sites are bility for a randomly chosen container of initially hotter deactivated by heating the water. Gilpin not only con- water to freeze before a randomly chosen container of ini- firmed that hot tap water supercooled more than cold tap tially colder water was 53%41. For ambient temperatures O O water, but that tap water left in an open container super- of −8 C >Ta > −11 C, the probability was 24%. If cooled least of all—this can be explained by the fact that the hot water, on average, takes longer to freeze, but only water in an open container will absorb impurities from happens to freezes first in some specific samples due to the air, and these impurities can then act as nucleation random fluctuations, it is not clear that this should be sites38. called a Mpemba effect. (Given Auerbach’s small num- Again, the phenomenon here is not the same as the ber of samples, 53% is not signficantly greater than 50%.) Mpemba effect, but is related, in that it explains how While the experiments on pipes show that supercooling the water can remember its history. However, note that can induce significant memory effects, the role of super- the results here are in the wrong direction to explain the cooling in the Mpemba effect remains uncertain. Mpemba effect. If initially hot water supercools more, then to freeze it has to reach an even lower temperature than initially cold water, which will lengthen the time VIII. FUTURE PROSPECTS that it takes to freeze. It is clear that evaporative cooling can play an impor- tant role in the Mpemba effect, and that the history of the VII. SUPERCOOLING AND THE MPEMBA water can affect the amount of supercooling. But beyond EFFECT that, the experiments together paint a very muddled pic- ture. More experiments are needed to solve this 2000+ Consideration of supercooling greatly complicates the year-old puzzle. Despite the theoretical complexity of the Mpemba effect, and it is not clear how, or whether, it Mpemba effect, the experiments needed to probe it can helps to explain it. We first need to decide precisely how be done at the K-12 and undergraduate level. Indeed, we measure the “time to freezing.” If we wait until the simple experiments on the Mpemba effect are a common first appearance of ice, then the experimental situation is science fair project. 10

A fair amount of thought needs to go into the exper- know of only one published paper studying the Mpemba imental design before even the first data point is taken. effect in a closed container31, and a single experiment Walker discusses the basic set-up, and while you can can always be in error. If you run a series of experi- come up with your own design, reading Walker’s article ments in closed containers, and regularly fail to find a is a good way to appreciate some of the hidden complexi- Mpemba effect, that would provide good support for the ties in the experiment35. For example, Walker points out claims that evaporation is the primary cause of the effect. that the container should be heated along with the water, On the other hand, observations of the Mpemba effect in for if hot water is poured into a cold container, the sud- closed containers would show that it can occur without den change in the water’s temperature causes problems. evaporation. To make sure that all samples of water have the same Rather than looking at freezing time versus initial mass, masses need to be measured after heating, rather temperature, you could investigate supercooling. Sim- than before, as a fair amount of mass is lost during the ply reproducing earlier experiments would be valuable heating process. The precise environment surrounding in resolving the discrepancy between the recent re- the container is important, and it can make a difference sults of Auerbach, and the older results of Brown and whether the water is in a middle of an empty freezer, Dorsey36,37,41. The number of runs in Auerbach’s exper- or jammed between a frozen pizza and a frost-covered iment was too limited for firm conclusions to be drawn, bucket of ice cream. The temperature can be read with so a repetition of his experiment with a greater number of a common mercury thermometer, but a device that can runs would be useful. A modern repetition of Descartes’ more quickly and accurately register changes in temper- experiment would also be interesting. ature is better, if available. The reader is directed to More systematic studies of how the history of the wa- Walker’s article for further discussion of potential prob- ter affects its properties would be helpful. For example, 35 lems and solutions . if you find that pre-boiled tap water has different prop- A series of trials will produce a graph of freezing time erties than water straight out of the tap, you could in- vs. initial temperature, of the sort shown in figure 1. vestigate why it differs (dissolved oxygen? impurities?), A single curve may or may not show a Mpemba effect, by looking at how long it takes pre-boiled water’s prop- but is not, on its own, particularly useful for probing the erties to return to those of tap water’s, and under what cause(s) of this phenomenon. To see how the Mpemba conditions. effect depends on the various parameters of interest (ini- You should try rerunning the experiments several times tial mass, gas content, container shape and type, etc. . .) with exactly the same parameters, to get an idea of how several curves need to be produced, under different pa- big the error bars are. The size of the error bars is crucial, rameter settings. If you had infinite resources and time, since if a graph of freezing time vs. initial temperature you could vary all the initial parameters, producing a gi- shows only a weak local maximum, it will be unclear if ant multidimensional matrix of freezing times. In prac- this is a true Mpemba effect, or just the result of fluctu- tice, you will have to decide which factors you are most ations. And, as with any experiment, you want to make interested in, and design your experiment accordingly. sure that your results are repeatable. This greatly in- You can decide on your own what to vary to produce creases the time required for the experiment, and limits your series of curves. I give some suggestions below, but the amount you can vary the parameters, but it’s bet- the number of reasonable experimental designs is essen- ter to have a small amount of reliable data than a large tially unlimited. amount of unreliable data! Kell claimed that because the Mpemba effect relies What will constitute experimental success? You do on surface cooling, it is more likely to be observed in not need to observe the Mpemba effect for your experi- wooden pails than in metal ones, since in metal pails a ment to be a success. Finding that the Mpemba effect fair amount of heat is lost through the sides. This claim does not occur under certain conditions is still a good can be tested by producing a series of curves of freezing experimental result. On the other hand, it is certainly time vs. initial temperature, for containers with differing more dramatic and psychologically satisfying if you can degrees of thermal insulation on the sides. This would find conditions under which the effect occurs. And if demonstrate how the Mpemba effect is affected as the you find conditions under which it occurs, you can then relative importance of evaporation to the cooling process study what changes destroy the effect, which provides a is varied. Alternatively, varying the height of the wa- potentially valuable probe of the phenomenon. You may ter, while keeping the base fixed, provides another way want to do some preliminary testing to find parameters of varying the importance of evaporation. where the Mpemba effect occurs, and then decide how to The importance of evaporation can be largely elim- proceed. inated by putting the water in a closed container, or If you complete experiments on the Mpemba effect, by putting a layer of oil over the surface of the water. I’d be interested in hearing the results—if you have a Such experiments would be extremely useful in assessing chance, send me an e-mail telling me what you found. If claims that evaporation is the cause of the Mpemba effect. enough experiments are done, perhaps this 2000+ year- As already discussed, several authors have claimed that old problem can be solved! evaporation is insufficient to explain their results. But I Finally, those who like the counterintuitive nature of 11 the Mpemba effect might be interested in a similar phe- a preliminary translation of the letter from Descartes to nomenon: water drops will last longer on a skillet well Mersenne, and Leigh Anne Eubanks for providing the above 100OC, than on one only a little above 100OC. final translation. We would like to thank Nancy Ruff for This is easier to explain than the Mpemba effect43. confirming the correctness of Burke’s translation of Roger Bacon, over another, incorrect, published translation. A previous version of this article appeared on the sci.physics Acknowledgments FAQ web page44.

We would like to thank Debra Waxman for providing

∗ Electronic address: [email protected] 23 F. L. C. Blackwall, “Mpemba’s ice cream,” New Sci. 43, 1 Aristotle, Meteorologica, trans. H. D. P. Lee (Harvard Uni- 88-89 (1969). versity Press, London, 1962), Book I, Chapter XII, p.85-7. 24 S. M. Arkless, “Mpemba’s ice cream,” New Sci. 42, 655- 2 M. Clagett, Giovanni Marliani and the late medieval 656 (1969). physics, (AMS Press Inc., New York, 1967), p.79. 25 P. Dickson and J. C. Goulden, Myth-Informed: Legends, 3 Antiperistasis, in a different form, was also central to Aris- Credos, and Wrong-headed ‘Facts’ We All Believe, (Peri- totle’s explanation of how an arrow was able to continue in gree Books, New York, 1993), p. 127. motion when the initial force of the bow was removed. A. 26 “Straight Dope” archives, http://www.straightdope.com/ Franklin, “Principle of inertia in the Middle Ages,” Am. J. classics/a2 098b.html. Phys. 44 (6), 529-545 (1976). 27 K. Popper, The Logic of Scientific Discovery (Harper and 4 R. Bacon, The Opus Majus of Roger Bacon, trans. Robert Row, New York, 1968). Belle Burke, vol. II (Russell and Russell Inc., New York, 28 D. G. Osborne, “Mind on Ice,” Phys. Educ. 14, 414-417 1962), Part 6, p.584. (1979). 5 M. Clagett, Ibid., p.71-2. 29 D. G. Osborne, “Freezing of Water,” New. Sci. 44, 125-126 6 F. Bacon, Novum Organum, in The Physical and Meta- (1970). physical Works of Lord Francis Bacon, ed. J. Devey (G. 30 M. Freeman, “Cooler Still—An Answer?,” Phys. Educ. 14, Bell and Sons, Ltd., London, 1911), Book II, Chapter L, 417-421 (1979). p.559. 31 B. Wojciechowski, I. Owczarek, and G. Bednarz, “Freez- 7 R. Descartes, Discourse on Method, Optics, Geometry, and ing of Aqueous Solutions Containing Gases,” Cryst. Res. Meteorology, trans. P. J. Olscamp (Bobb-Merrill Company, Technol. 23 (7), 843-848 (1988). Indianapolis, 1965), Meteorology, Ch. 1, p.268. 32 E. Deeson, “Cooler-lower down,” Phys. Educ. 6, 42-44 8 J. F. Scott, The Scientific Work of Rene´ Descartes (1596- (1971). 1650) (Taylor and Francis, London), p.67, citation omitted 33 I. Firth, “Cooler?,” Phys. Educ. 6, 32-41 (1971). from quotation. 34 A. Osborne, “Faster freezing,” New Sci. 43, 662 (1969). 9 R. Descartes, Œuvres Lettres de Descartes (Librarie Gal- 35 J. Walker, “The Amateur Scientist: Hot water freezes limeru, 1953), p.998. faster than cold water. Why does it do so?,” Sci. Am. 237 10 Translation courtesy of Leigh Anne Eubanks. Preliminary (3), 246-257 (1977). translation provided by Debra Waxman. 36 F. C. Brown, “The frequent bursting of hot water pipes 11 T. S. Kuhn, The Structure of Scientific Revolutions (The in household plumbing systems,” Phys. Rev. 8, 500-503 University of Chicago Press, Chicago, 2nd ed., 1970). (1916). 12 T. S. Kuhn, Ibid., p.5. 37 N. E. Dorsey, “The freezing of supercooled water,” Trans. 13 T. S. Kuhn, Ibid., p.35. Am. Philos. Soc. 38, 246-328 (1948). 14 T. S. Kuhn, Ibid., p.16. 38 R. R. Gilpin, “The effects of dendritic ice formation in 15 The explanation of “heating by antiperistasis” is, of course, water pipes,” Int. J. Heat Mass Transfer 20 (6), 693-699 no longer valid. But Kuhn is criticizing not just the expla- (1977). nation, but the experimental results claimed by Marliani 39 C. A. Knight, The Freezing of Supercooled Liquids, D. Van and Bacon. Nostrand, Princeton, New Jersey (1967). 16 E. B. Mpemba and D. G. Osborne, “Cool?,” Phys. Educ. 40 C. A. Knight, “The Mpemba effect: the freezing times of 4, 172-175 (1969). hot and cold water,” Am. J. Phys. 64 (5), 524 (1996). 17 E. B. Mpemba, ibid., p.174. 41 D. Auerbach, “Supercooling and the Mpemba effect: 18 G. S. Kell, “The freezing of hot and cold water,” Am. J. When hot water freezes quicker than cold,” Am. J. Phys. Phys. 37 (5), 564-565 (1969). 63 (10), 882-885 (1995). 19 R. M. Robson, “Mpemba’s ice cream,” New Sci. 43, 89 42 J. van der Elsken, J. Dings, and J. C. F. Michielsen, “The (1969). freezing of supercooled water,” J. Mol. Structure 250, 245- 20 M. B. F. Ranken, “Mpemba explained,” New Sci. 45, 225- 251 (1991). 226 (1970). 43 J. Walker, “Boiling and the Leidenfrost Effect,” in Funda- 21 R. Jephson, “Mpemba’s ice cream,” New Sci. 42, 656 mentals of Physics, by D. Halliday and R. Resnick (Wiley, (1969). New York, 1988, 3rd ed.), p. E10-1-5. 22 J. C. Dixon, “Mpemba’s ice cream,” New Sci. 42, 656 44 Usenet Physics FAQ, http://math.ucr.edu/home/baez/ (1969). physics/General/hot water.html, 1998.