A physical description of coffee cooling in a pot

Jan Sedláček∗
Czech University of Agriculture in Prague, Technical Faculty,
Department of Physics, Kamýcká 129, 16521 Prague 6, Czech Republic

Jiřı́ Dolejšı́†
Charles University in Prague, Faculty of Mathematics and Physics,
Inst. of Particle and Nuclear Physics, V Holešovičkách 2, 180 00 Prague 8, Czech Republic

One everyday phenomenon – coffee getting cold in a pot – is studied as an example of a typical
method used in physics. We meet this topic in many articles in which authors usually limit them-
selves on using of Newtons law of cooling (see references1,2,3,4,5,6,7,8,9 ). Our effort is aimed at the
non-professional audience, we suppose only elementary knowledge of the subject and our approach
should be obvious in the common sense.
The obvious beginning is to focus attention to the phenomenon and to collect some experience.
The second step is the experiment in controlled conditions and with different variants of studied
objects. The third step is an attempt to understand the results of the experiment and the nature of
the phenomenon. To reach that in the case of cooling coffee, we try to follow the pathway of heat
in the studied system within a sequence of models realized as spreadsheets in Excel.
We can guess that some heat is used for evaporation, some heat is radiated to surroundings and
other amount of heat accepted from surroundings. We also must not forget the transport of heat
between liquid and pot, through the pot and finally between pot and surroundings. We do not
measure the individual heat flows, but we put them into the model and from the comparison of the
model prediction for the time-dependent temperature with the data we deduce the role of different
mechanisms and the parameters which enter the model.

PACS numbers: 01.40.Ej, 01.50.My, 01.50.Pa, 01.50.Qb

Keywords: Newton’s law of cooling

Introduction minutes, either with a usual mercury thermometer (our

case) or with some more sophisticated equipment. The
estimated error of our measurement is about 0.3◦C. We
A standard situation from the everyday life is a pot
would like to concentrate to the detailed understanding
of coffee slowly getting cold when we forget about it (see
of the measured dependence and to identifying the rele-
fig. 1). Does it matter what kind of the pot we use? What
vant physics phenomena contributing to the process. We
are the physics processes which play some role in this or-
will try to achieve this goal by constructing a suitable
dinary occurrence? Which are more important, which
simulation model and by refining it to fit the data.
less? Can we influence the evolution in the direction we
need, like cool faster to drink in a hurry or cool slower
during our busy activity? Let us turn to experiments. At I. THE FIRST MODEL
first we measure the time-dependent temperature of the
cooling coffee in a china pot on a table. The measurement
should be done in reasonable intervals, e.g. in every five We will start with a simple assumption that the heat
transferred from the hot coffee to the colder surround-
ings per unit time is dependent only on the temperature
difference and that this dependence is linear:
temperature (°C)
90
∆Q/∆τ = k(tc − ts ),
80

70 too hot to drink

where k is the coefficient of the heat exchange between
60 coffee and surroundings, tc is the changing temperature of
50 nice the coffee, ts is the constant temperature of surroundings.
40
We assume that the temperature of the coffee is uniform.
Than we can write the equation relating the heat loss
30 cold
with the change of the coffee temperature with time τ in
20
0 20 40 60 80 100 120 140 160 the form
time (min)
∆tc ∆Q
mc = = k(tc − ts )
FIG. 1: The measured cooling of coffee. The personal views ∆τ ∆τ
of the optimal drinking temperature could be different, of where m is the mass of coffee and c is the specific heat
course. capacity of it. This relates to Newtons law of cooling.
 2

This equation we approximately and naively solve in the in the heat transfer? We may expect heat losses by evap-
simplest possible manner doing small steps in time with oration and related loss of coffee mass. Heat could be
the help of Excel (ignoring all the knowledge on solution radiated to surroundings and accepted from it. The pot
of differential equations), we only check that the time is heated first and then it cools, it may have the tempera-
step is sufficiently small not to influence the results. We ture different from coffee. Heat is transferred through all
fitted the coefficient k to meet the data (using Solver the surfaces. Maybe there are other effects still missing
from Excel). As we may conclude from fig. 2, the first in our list
temperature (°C)
90 experiment
III. THE THIRD MODEL
80 model 1

70
We will try to introduce all the abovementioned effects:
60

50 • transfer of heat from coffee to the pot:

40 ∆Q1 /∆τ = k1 (tc − tp );
30
• coffee evaporation to surroundings:
20
0 20 40 60 80 100 120 140 160
∆Q2 /∆τ = vlv tc P ,
time (min) where v is the speed of evaporation, lv is specific
heat of vaporization, P open surface;
FIG. 2: The time-dependent temperature from the experi-
ment confronted with the results of the first model (fitted • radiation from coffee to surroundings:
k = 0.54 J.s−1 .K−1 ). ∆Q3 /∆τ = ασTc4 P ,
where σ is the Stefan-Boltzmann constant, α is the
model systematically underestimates the heat loss at the absorptivity of coffee;
early moments and overestimates it later.
• radiation from surroundings to coffee:
∆Q4 /∆τ = ασTs4 P
II. THE SECOND MODEL
• transfer of heat from pot to surroundings:
∆Q5 /∆τ = k2 (tp − ts );
To improve the agreement of the model and data we
will try to use the ”brute force method” - we intro- • radiation from pot to surroundings:
duce the arbitrary exponent x into the heat transfer: ∆Q6 /∆τ = βσTp4 S,
∆Q/∆τ = k(tc − ts )x . The introduction of this expo- where S is the emitting surface of the cup, β is the
nent is clearly a substitute for better and more detailed absorptivity of its material;
description of the process. Both parameters k and x are
• radiation from surroundings to pot:
fitted with Excel. The fig. 3 shows better accuracy of
∆Q7 /∆τ = βσTs4 S;
the second model. The quality of the fit we judge from
the value of residual sum of squares divided by the num-
ber of degrees of freedom - ”χ2 /N DF ”, see any textbook temperature (°C)
90
on statistical methods, e.g.10 . Values much larger than 1 experiment
80 model 3
signalize the poor fit.
70
temperature (°C) 60
90 experiment
50
80 model 1
40
70 model 2
30
60
20
50 0 20 40 60 80 100 120 140 160
time (min)
40

30
FIG. 4: The measured temperature and the third model.
20 The fitted constants are k1 = 31.43 J.s−1 .K−1 , k2 =
0 20 40 60 80 100 120 140 160
time (min)
0.31 J.s−1 .K−1

FIG. 3: The measured temperature and the both models. Our model 3 (see fig. 4) is not yet ideal - the pot is
Fitted k = 0.17 J.s−1 .K−x , x = 1, 34. χ2 /NDF changed from still thin (i. e. it has only one temperature). But that
32.1 for the first model to 2.4 for the second model. model is more ”physical” than the former - at least coffee
evaporates. We are lacking any insight into the details
What are the effects which are mimicked by the power of evaporation, so we employed the linear dependence on
 3

tc . All the parameters α, β , v, k1 , k2 are fitted within

reasonable bounds. We are not much successful with this
model - the residual sum 2/NDF = 81.3 compared to
2/NDF = 32.1 for the first model. Before turning to the
next model we include one more experimental datum -
the change of coffee weight = the weight of evaporated
liquid. With this datum included we have χ2 /NDF =
81.6.
The lack of faster cooling at the beginning persists in
the model. Maybe the evaporation speed linear in tem-
perature is unrealistic?

IV. THE FOURTH MODEL

Although we do not have any detailed knowledge on

the dependence of evaporation speed on temperature and
other conditions, we can try to put a free power z in the
”vaporising” term:
∆Q2 /∆τ = vlv tzc P.
FIG. 6: The measurement of the evaporation speed: The
The fit results in z = 3.73! Then the agreement of the immersion heater controlled by the simple switching circuit
model and data is remarkable (see fig. 5) - χ2 /NDF = act as thermostat, the pot with water is regularly weighted.
1.0.
temperature (°C)
90
where tpi , tpe are the internal and the external tempera-
experiment
80 model 4
ture of the pot. We must also introduce another term in
our model - heat transfer through the pot:
70

60
∆Q8 /∆τ = k3 (tpi − tpe )
50

40 .
30

20
0 20 40 60 80 100 120 140 160
time (min)

FIG. 5: The measured temperature and the fourth model.

The fitted constants are k1 = 29.87 J.s−1 .K−1 , k2 =
0.18 J.s−1 .K−1

But the natural question appears: Is the funny expo-

nent 3.73 reasonable? We have made with our colleagues
a simple dedicated experiment for measuring of the evap-
oration speed, see fig. 6. The result of the experiment,
power z = 3.7, really agrees with the above value!

V. THE FIFTH MODEL

FIG. 7: The three various experimental pots.
In the next model we suppose the pot with nonzero
thickness to treat different pots. It means we assume
there are an external and an internal temperature of the
pot. Then we need to adapt the two already used terms VI. THE COMPARISON OF THE INFLUENCE
in the following way: OF DIFFERENT PARAMETERS

∆Q1 /∆τ = k1 (tc − tpi ),

By introduction of the parameters in the sequence of
our models and by fitting them to meet best the data
∆Q5 /∆τ = k2 (tpe − ts ), we get sets of parameters, which describe the experiment
 4

comparatively well. Clearly we need further experiments By determination and matching of all coefficients we
with exactly directed setup for specification of different have justified following facts (look at values at table I):
parameters. To check the importance of the evaporation
we put some fat on the liquid surface. We use various • coffee surface (dark) radiates and absorbs heat bet-
pots with various specific heat capacities and the heat ter than water surface
conductivities, see fig. 7. Comparison of dark coffee and
clear water shows us the role of the absorptivity of the
liquid surface. We performed a set of experiments with • shining stainless pot radiates and absorbs heat
different conditions: At first we discovered the results do worse than the dim one
not depend on the surface tension what is important for
the comparison of coffee and water. The next step was • fat on the liquid surface prevents evaporation
measurement for:
• three pots (china, ordinary metal sheet pot and • heat transfers through the china pot a little worse
vacuum-pot, it means the pot with double wall) than through the thin metal pot and much better
than through the vacuum-pot
• three liquids (coffee, water and water with fat on
its surface)

VII. CONCLUSION
TABLE I: The fitted parameters of the fifth model.
11 This contribution shows the beginning of a story orig-
α β
v.10
kg.◦C
  kJ1   kJ2   kJ3  inating from one phenomenon of our everyday life. We
Liquid Pot s.m2 s.K s.K s.K formulated few models to describe the cooling down of a
coffee china 0.990 0.192 15.8 2.90 0.360 2.04 cup of coffee with rather good final result. The sequence
water china 0.005 0.192 15.8 2.90 0.360 2.04 of approximations shows how we are pushed towards the
w.w.fat china 0.005 0.192 1.94 2.90 0.360 2.04 more detailed description and towards the study which
coffee metal 0.990 0.178 15.8 2.90 0.360 2.42 processes are important and which not. We were forced
water metal 0.005 0.178 15.8 2.90 0.360 2.42 to introduce some nonlinearity; we tried to do it by in-
w.w.fat metal 0.005 0.178 1.94 2.90 0.360 2.42 troducing power into evaporation speed. We verified its
power dependence on the temperature in an independent
coffee vac.pot 0.990 0.050 15.8 2.90 0.360 0.187
experiment. The whole story shows by simple means the
water vac.pot 0.005 0.050 15.8 2.90 0.360 0.187
path from observation to the description on some level
w.w.fat vac. pot 0.005 0.050 1.94 2.90 0.360 0.187 and may continue in further studies.

∗
[email protected] http://www.aw-bc.com/ide/Media/JavaTools/nlhcdata.
†
[email protected] html.
1 7
M. A. Karls and J. E. Scherschel, Am. J. of Physics 71, J. Martino, Newton’s law of cooling, http://www.math.jhu.
678 (2003). edu/courses/107/archives/notes/f01coursenotes/node9.
2
S. J. Greenwald and W. C. Bauldry, Coffee cooling on html.
8
a ti-cbl unit and in maple, http://www.mathsci.appstate. E. Yanik, Coffee, tea or not?, http://archives.math.utk.
edu/˜sjg/papers/CoffeeCooling.pdf. edu/ICTCM/EP-7/SA14/pdf/paper.pdf.
3 9
Cooling a cup of coffee, http://hyperphysics.phy-astr.gsu. My hot chocolate is too hot!, http://mathforum.org/ cal-
edu/hbase/thermo/coocof2.html. cpow/solutions/solution.ehtml?puzzle=48 (1999).
4 10
The coffee problem, http://www.rogerfrost.com/cof.htm. S. Brandt, Statistical and computational methods in data
5
Newton’s law of cooling, http://www.aw-bc.com/ide/ Me- analysis (North Holland/American Elsevier Publishing
dia/pdf/Documents/Part1/Labs/Lab01.pdf. Company, 1970).
6
How to use the newton’s law of cooling: Curve fitting tool,