Skip to content
Mintnib™

Mintnib › Oxbridge interview questions › Coffee and milk

Coffee and milk: when should you add it?

Print this page: a free PDF

A classic interview question, answered in full. In an ordinary mug, adding the milk immediately leaves the coffee about 2.5 °C hotter after ten minutes than adding the same milk at the ten-minute mark. This page explains why, level by level as an interviewer's follow-ups add complications, how each variable increases or decreases that difference, which kind of mug makes adding the milk at the end the better choice, and how to present the answer in an interview.

Schools and organisations we have worked with

Shown with their permission. Not endorsements. See all

The question, as an interviewer asks it

“You have just made a cup of coffee, but you will not drink it for ten minutes. You take milk. When should you add the milk, if you want the coffee as hot as possible when you drink it?”

That is the whole question, and it is deliberately incomplete. Nobody has said how much milk, how cold it is, how warm the room is, what shape the mug is, or whether anyone blows on it. Part of what the interviewer is watching is whether you notice.

Before you calculate: ask, or assume out loud

A strong answer starts with the missing information, either as questions or as stated assumptions. Each one changes the answer by an amount this page computes with its full model (level 4), for fridge milk in a straight mug unless the line says otherwise:

Then commit to a base case and say it: 250 ml of coffee at 90 °C in a straight ceramic mug 8 cm across, 30 ml of milk from the fridge at 5 °C, a room at 20 °C with still air, and ten minutes to wait. Students who raise air motion and mug shape without being asked are doing exactly what the question is for.

The short answer, and what it depends on

Add the milk immediately. In the base case, on the page's full model (level 4), the coffee is at 65.5 °C at minute ten with the milk added immediately, and at 63.0 °C with the same milk added at minute ten: 2.5 °C hotter. With milk at room temperature the difference is 2.0 °C.

On this page, Δ always means the temperature at minute ten with the milk added immediately minus the temperature with the milk added at minute ten. A positive Δ means adding the milk immediately leaves the coffee hotter.

Δ depends on the mug (with 30 ml of milk, in a cup that flares out just above the coffee, Δ is negative), on the amount of milk (in a straight mug, more milk gives a larger Δ), and on whether the milk is kept hot while you wait (above about 69 °C, Δ is negative). The ladder below builds the answer one follow-up question at a time, from the 30-second answer to the most complete model.

The ladder: from the 30-second answer to the full model

Each level opens with the follow-up question an interviewer asks next, adds one complication, and says what changed. Stop at the level you need.

  1. Level 0: the 30-second answer: So, when do you add the milk?
  2. Level 1: the coffee cools while you wait: What if the coffee cools while you wait?
  3. Level 2: the milk is not at room temperature: What if the milk comes from the fridge, or is kept hot?
  4. Level 3: where the heat actually goes: Is convection the only way the coffee loses heat?
  5. Level 4: the leak per degree is not constant: Does the coffee really lose the same fraction of its excess every minute?
  6. Level 5: the shape of the mug: Does the shape of the mug matter?
  7. Level 6: the other variables, one at a time: What else changes the answer, and by how much?
  8. Level 7: the cup left for hours: What happens if nobody drinks it?

After the ladder: what an interviewer is testing at each level; what students often get wrong; where students get stuck, and the step that unsticks it; more follow-up questions an interviewer may ask; check it against your kitchen; where this puzzle comes from.

Level 0: the 30-second answer

The follow-up. “So, when do you add the milk?”

The intuition. A hot drink loses heat faster than a warm one, because the rate of heat loss increases with the difference between the drink and the room. Adding cold milk immediately lowers that difference for the whole wait, so the drink loses less heat in total. That reasoning, said in two sentences, is a strong first answer.

The simplest model: mixing is a weighted average, and no heat is lost.

Mixing is a weighted average

No heat leaves during the instant of mixing, so the heat the coffee gives up is the heat the milk takes in. With C=1300 J K⁻¹ for coffee and mug, c=121 J K⁻¹ for the milk and Tm the milk's temperature:

C(T−Tmix)=c(Tmix−Tm)⇒Tmix=CT+cTmC+c

Fridge milk into coffee at 90 °C gives 82.8 °C. Celsius and kelvin give the same answer here: add 273.15 to both temperatures and the result increases by exactly 273.15, because the two weights add up to one. A weighted average works in any temperature scale. Kelvin is needed wherever a formula uses the temperature itself rather than a temperature difference or a weighted average whose weights add up to one: radiation's T4 in level 3, and the film temperature Tfilm in the Rayleigh number (the assumptions of level 3).

The result at this level. If no heat is lost while you wait, both orders end at the same temperature, 82.8 °C: the coffee is still at 90 °C when the milk goes in, whenever that is. So the order can matter only through the heat lost during the wait, which is what every level below adds.

Assumptions added at this level, each with why it holds:

What changed from the level before. This is the starting point: with no heat loss, Δ is exactly zero.

Level 1: the coffee cools while you wait

The follow-up. “What if the coffee cools while you wait?”

Now heat leaks out, at a rate proportional to the difference between the drink and the room. This is Newton's law of cooling, derived in full below. To isolate the effect of the wait, this level uses milk at room temperature; level 2 adds the milk's own temperature.

Newton's law of cooling, derived in full

Step 1, energy. T(t) is the drink's temperature at time t, starting at T0=90 °C, in a room at Tr=20 °C. Let C be the heat capacity of coffee and mug together: 1045 J K⁻¹ for the coffee plus 255 J K⁻¹ for the ceramic, C=1300 J K⁻¹. The heat stored above a reference temperature Tref is C(T−Tref), so its rate of change is CdT/dt. If heat leaves at Q watts, energy conservation says the two are equal and opposite:

CdTdt=−Q(T)

This is a differential equation: it gives the rate of change of T, not T itself. At 90 °C, Q≈69 W (level 3 shows where each watt goes), so the coffee starts by losing 3.2 °C a minute. The minus sign says the temperature decreases while heat leaves.

Step 2, the simplest leak. “Loss proportional to the gap” becomes Q=K(T−Tr), where Tr is the room temperature and K, in W K⁻¹, measures the leakiness. For convection alone K=hA, with h the convection coefficient and A the area. In our mug at 90 °C, the four routes of level 3 together give K=Q/(T−Tr)≈69/70≈0.99 W K⁻¹. This level treats K as one measured number; level 3 shows what it is made of. This is Newton's law of cooling: exact for a fixed coefficient, and a fair first approximation for any smooth loss over a small range of temperature.

Step 3, a substitution: measure from the room. Let u=T−Tr, the excess over the room. The room is fixed, so du/dt=dT/dt, and with τ=C/K:

dudt=−uτ

Why substitute: in T the right side is −K(T−Tr)/C, which carries the constant Tr; in u the rate is simply proportional to the unknown itself, and the next two steps take one line each. Another way: keep T and integrate dT/(T−Tr) directly, which gives the same logarithm; or multiply by the integrating factor et/τ. The substitution is a convenience, not a necessity.

Step 4, why the answer must be exponential. The equation says that in each small interval the gap loses the same fraction of itself: 1/τ per unit time, whatever the gap is. In our mug that is about 4.5 % of the gap in the first minute. A quantity that loses a fixed fraction per unit time is an exponential, like radioactive decay or compound interest run backwards. Steps 5 and 6 prove it.

Step 5, separate the variables. The equation mixes u and t. Divide both sides by u, which is allowed because u>0 throughout: in this model the coffee approaches the room but never reaches it.

1ududt=−1τ

By the chain rule the left side is the derivative of lnu with respect to t. Both sides are now derivatives of something we know: ddtlnu=ddt(−tτ). Why this step: integrating undoes a derivative, but only when each side is a derivative of a known function; separating the variables arranges exactly that. Writing du/u=−dt/τ with differentials is shorthand for this chain-rule step.

Step 6, integrate from the start to time t. Integrate both sides from 0 to t; on the left the variable runs from the starting gap u0=70 K to u:

∫u0udu′u′=−1τ∫0tdt′⇒lnu−lnu0=−tτ

Definite limits fix the constant of integration for us. With indefinite integrals you would write lnu=−t/τ+k and find k=lnu0 by putting t=0: the same thing. Take the exponential of both sides:

u=u0e−t/τ,τ=CK

The constant of integration has become a multiplying factor, u0, not an added constant. Another way: guess u=aeλt, substitute it into Step 3 to get λ=−1/τ, and use the start to get a=u0. Guessing works because Step 4 told us what to guess.

Step 7, read the time constant. τ is stored heat per kelvin over leak per kelvin: J K⁻¹ divided by W K⁻¹ is J W⁻¹, which is seconds. It is the time to lose all the excess heat at the starting rate; in that time the gap falls to 1/e≈37 % of its starting value, after 2τ to 14 %, and it halves every τln2. Ours is about 22 minutes at 90 °C, so the gap would halve in about 15 minutes.

Comparing the two orders: the general condition

Let K be the leak of the black coffee and K′ the leak after the milk raises the level. Measure from the room, so the milk's excess is um (−15 K for fridge milk; u0=70 K for the coffee). After a wait t, with the milk added immediately (first line) or at the end of the wait (second line):

uimm=Cu0+cumC+ce−K′t/(C+c)uend=Cu0e−Kt/C+cumC+c

With milk at room temperature (um=0) each is the same constant, Cu0/(C+c), times one exponential. Why divide one by the other: the shared constant cancels and the comparison becomes a single exponent whose sign decides it (subtracting works too, but leaves a difference of exponentials to compare). The ratio is

uimmuend=exp[t(KC−K′C+c)]

It is greater than 1 exactly when its exponent is positive, K/C>K′/(C+c). Multiply both sides by C/K, which is positive: adding the milk immediately leaves the coffee hotter exactly when

K′K<C+cC

If one coefficient h applies to every surface, substitute K=hA: h cancels and the condition becomes pure geometry, Aafter/Abefore<Cafter/Cbefore: the milk must add relatively more heat capacity than surface. Ignoring the mug and the small difference in specific heat, that is Aafter/Abefore<mafter/mbefore. In our mug the milk adds 9.3 % of heat capacity and only 2.3 % of leak. With one h for every surface, our mug would only just pass: its total area increases by 8.8 % against 9.3 %. It passes by a wide margin because the wetted wall loses far less heat per square centimetre than the open top (the loss budget of level 3).

The result at this level. With milk at room temperature and K fixed at its 90 °C value, the coffee is at 61.8 °C at minute ten with the milk added immediately and 60.6 °C with it added at the end: Δ = 1.2 °C.

Assumptions added at this level, each with why it holds:

What changed from the level before. Heat now leaks while you wait, so the order matters. With milk at room temperature and Newton's law, Δ is positive exactly when the milk adds a larger fraction of heat capacity than of leak; in our mug Δ = 1.2 °C, against zero at level 0.

Level 2: the milk is not at room temperature

The follow-up. “What if the milk comes from the fridge, or is kept hot?”

For milk at any temperature, subtract the two formulas of level 1. To keep it short, write E=e−Kt/C and E′=e−K′t/(C+c), the two decay factors, and multiply both sides by C+c:

(C+c)(uimm−uend)=Cu0(E′−E)+cum(E′−1)

The first term is the room-temperature result of level 1. The second has E′−1<0, so it is positive when um<0 and negative when um>0:

That is also why a cup must flare further before adding the milk at the end is better with fridge milk (50 degrees, against 41 degrees with milk at room temperature; level 5).

The result at this level. With milk from the fridge at 5 °C and the same fixed K, the coffee is at 61.0 °C at minute ten with the milk added immediately and 59.3 °C with it added at the end: Δ = 1.7 °C, against 1.2 °C with milk at room temperature.

The results at this level that surprise people, each explained in full:

Assumptions added at this level, each with why it holds:

What changed from the level before. The milk's own temperature now enters. On the same fixed-K model, fridge milk increases Δ from 1.2 to 1.7 °C; milk kept hotter than about 69 °C makes Δ negative (full model).

Level 3: where the heat actually goes

The follow-up. “Is convection the only way the coffee loses heat?”

The loss budget

Heat leaves a hot drink four ways. Here is each, estimated for our mug, with its working.

1. Convection from the top. Air next to the hot surface warms, rises and is replaced. For a warm surface facing up, the standard correlation (see the assumptions at the end of this level) gives a coefficient of about h≈10.4 W m⁻² K⁻¹ at 90 °C (typical range 5 to 10 W m⁻² K⁻¹ for still air; ours is at the top of the range because the gap is 70 °C). With Atop=50.3 cm² and a 70 °C difference, hAtopΔT≈3.7 W. The sides lose about 8 W the same way, with h≈9.1 W m⁻² K⁻¹.

2. Radiation. Every surface radiates, and the room radiates back; the net loss is εσ(T4−Tr4) per square metre, with temperatures in kelvin: this is the formula that is not linear in temperature, so here, unlike in the mixing of level 0, Celsius would give a wrong answer. Here σ=5.67×10−8 W m⁻² K⁻⁴ is the Stefan and Boltzmann constant, and the emissivity ε is the fraction of a perfect emitter's output that the surface gives off: 0.95 for water and 0.9 for glazed ceramic in the infrared, whatever colour they look. To compare it with convection, factor the difference of fourth powers:

T4−Tr4=(T−Tr)(T+Tr)(T2+Tr2)

With each of the last two factors taken at the mean temperature, T+Tr=2Tmean exactly and T2+Tr2≈2Tmean2, so

T4−Tr4≈4Tmean3(T−Tr)

where Tmean is the mean of coffee and room, 55 °C or 328 K. Replacing the curve by a straight line like this is called linearising; at 90 °C it is 1.1 % low. Radiation then acts like a second coefficient, hr=4εσTmean3≈7.2 W m⁻² K⁻¹ (with ε=0.9): nearly as large as natural convection. Total radiation is about 9 W at 90 °C.

3. Evaporation. Each gram that leaves takes about 2.3 MJ kg⁻¹ × 0.001 kg ≈ 2.3 kJ, enough to cool the whole mug and coffee by 1.8 °C. The vapour pressure is the part of the air's pressure due to water vapour. Air touching the water holds as much vapour as it can at the water's temperature (the saturation vapour pressure), and vapour moves from there towards the drier room. At 90 °C the vapour pressure at the surface is 70 kPa against 1.2 kPa in the room (half of the 2.3 kPa that saturated air at 20 °C would hold), so the coffee loses about 1.3 g a minute and evaporation carries about 48 W. Vapour pressure doubles about every 16 °C near 70 °C, so this route decreases quickly as the coffee cools.

4. Conduction through the wall and into the table. A ceramic wall 4 mm thick (conductivity about 1.5 W m⁻¹ K⁻¹) conducts far better than the air film outside it, so the outside of the mug is nearly as hot as the coffee and the wall itself adds little resistance. A double-walled vacuum mug stops the side losses almost completely. A paper cup has thin walls and almost no heat capacity of its own. Through the base: a mug on wood loses roughly 1 W (a steady estimate: heat spreading from a disc of radius a into a solid of conductivity k meets a resistance 1/(4ka); the contact disc is 7 cm across), small enough to ignore; on a stone worktop the same estimate gives about 27 W, which is not small.

RouteHeat loss at 90 °C [W]Share [%]Heat loss at 60 °C [W]Share [%]
Evaporation47.76912.754
Convection12.1176.126
Radiation9.4144.720
Base (on wood)about 1left out
Total69.210023.5100
Heat loss by route at 90 and 60 degreesAt 90 degrees: evaporation 48 W, convection 12 W, radiation 9 W. At 60 degrees: 13, 6 and 5 W.90 °C48 + 12 + 9 = 69 W60 °C13 + 6 + 5 = 23 Wevaporationconvectionradiation
Where does the heat go when the coffee is hot, and when it is warm? Our mug. Evaporation is the biggest share at both temperatures but shrinks fastest; the three segments are evaporation, convection and radiation from left to right.
Heat lost by each route against temperatureEvaporation rises steeply with temperature; convection and radiation rise roughly in proportion to the temperature above the room.3040506070809001020304050coffee temperature [°C]heat loss rate [W]evaporationconvectionradiation
How does each route change as the coffee cools? Evaporation (solid) dominates when hot and falls away fastest; convection (dashed) and radiation (dotted) are close to each other all the way down.

What each route does to the time constant (heat capacity over leak per degree, at 90 °C, as in Step 7 of level 1):

So the textbook simplification, convection alone, makes the time constant about 126 minutes instead of 22. Interviews often use it anyway, because it makes the equation linear and the comparison clean; that is fine as a teaching model, and the honest thing is to say that radiation and evaporation are usually NOT negligible for a hot drink: including them shortens τ from about 126 to 22 minutes, and level 4 shows what that does to Δ.

The calculation, in order

The sums of levels 1 and 2 in the order you would do them on paper, now with the leak taken from this budget and fixed at its 90 °C value. Milk from the fridge, a wait of t=10 minutes.

  1. The leak, from the loss budget. Convection coefficients from the correlations (h≈10.4 W m⁻² K⁻¹ on top, 9.1 on the side), the radiation coefficient hr≈7.2 W m⁻² K⁻¹, and evaporation from the vapour pressures: Q≈69 W in all at 90 °C.
  2. Heat capacities. C=1045+255=1300 J K⁻¹ for coffee and mug; c=121 J K⁻¹ for the milk.
  3. Leaks. K≈0.99 W K⁻¹ from the loss budget. With the milk in, the level rises and wets more wall: K′≈1.01 W K⁻¹ (2.3 % more).
  4. Time constants. τ=C/K≈22 minutes for black coffee; τ′=(C+c)/K′≈23 minutes with the milk in.
  5. Milk added at the end. Cool the black coffee for t: Tr+u0e−t/τ≈64.4 °C. Then mix: (CT+cTm)/(C+c)≈59.3 °C.
  6. Milk added immediately. Mix first: 82.8 °C. Then cool the mixture for t with τ′: 61.0 °C.
  7. Compare. Adding the milk immediately leaves the coffee 1.7 °C hotter. Check the sign against the condition of level 1: the milk added 9.3 % of heat capacity and 2.3 % of leak, so Δ must be positive.

The results at this level that surprise people, each explained in full:

Assumptions added at this level, each with why it holds:

What changed from the level before. The leak is no longer one number: level 1's K≈0.99 W K⁻¹ is now four routes, with evaporation 69 % of it at 90 °C. Convection alone would give τ ≈ 126 minutes instead of 22. With the same K the hand calculation still gives Δ = 1.7 °C; what this level adds is the reason level 4 is needed: evaporation, the largest route, decreases fastest as the coffee cools.

Level 4: the leak per degree is not constant

The follow-up. “Does the coffee really lose the same fraction of its excess every minute?”

Step 8, what changes when the loss is not proportional to the gap. Steps 5 and 6 of level 1 work for any loss that depends only on the temperature: separating the variables gives the time to cool from T0 to T as

t=C∫TT0dT′Q(T′)

What changes is whether the integral can be done by hand.

So the full model does not integrate by hand. It steps the equation forward in time: from the temperature now, estimate the slope −Q/C four times across a 2 second step: at the start, twice at the middle (each time using the previous estimate to get there) and at the end. Move on by their weighted average, with weights 1, 2, 2 and 1 over 6, and repeat. This is the classical fourth order method of Runge and Kutta. The local time constant τ(T)=Cu/Q(T) still describes how fast it is cooling at each moment.

Step 9, our τ is not constant. At 60 °C the leak per degree is lower and τ≈37 minutes, against 22 minutes at 90 °C. Hot coffee loses more heat per degree than warm coffee because all three routes increase faster than the gap: evaporation steeply, radiation as T4, and even convection, whose coefficient increases as u1/4. Newton's law is the right first model, not the last word.

The temperature gap on a log scaleNewton's law with a fixed time constant is a straight line on this scale. The real coffee bends away from it, cooling more slowly as it cools, because evaporation fades.0102030405060510204070time [min]coffee minus room [°C], log scalereal coffeeNewton, fixed τ
Is the cooling really exponential? Black coffee in our mug for an hour, on a logarithmic vertical scale, where an exponential is a straight line. A pure exponential with the starting time constant (22 minutes) is the dashed line; the real curve flattens as the time constant increases.

The race in a real mug. The full model solves the energy balance every 2 seconds with every route recomputed at the current temperature and level, and at the chosen minute mixes in the milk with the weighted average of level 0.

Temperature over ten minutes, milk added immediately against milk added at minute tenWith the milk added immediately the coffee drops at once to 82.8 degrees and ends at 65.5. With the milk added at minute ten the black coffee stays hotter, reaching 68.3, then drops to 63.0.02468105565758595time after pouring [min]temperature [°C]added immediatelyadded at minute 1065.563.0
Which order leaves the coffee hotter when you drink it? Our mug with fridge milk. Milk added immediately (solid): the temperature drops by 7.2 °C when the milk goes in, then decreases more slowly than the black coffee. Milk added at minute ten (dashed): the black coffee stays hotter and loses heat faster, then drops by 5.4 °C when the milk goes in at minute ten.
Temperature at minute ten against the minute the milk is addedFrom 65.5 degrees with the milk at once to 63.0 with the milk at the end, falling steadily in between.024681061626364656667minute the milk goes in [min]temperature when you drink [°C]
Does it matter exactly when the milk goes in? Our mug, fridge milk, drunk at minute ten. Each minute of delay lowers the final temperature by about 0.3 °C, so the earliest moment is the best one.

Three causes add up to Δ = 2.5 °C. With milk at room temperature only the first two act, and give 2.0 °C; fridge milk adds the third, 0.5 °C:

  1. More mass without more surface. The milk adds 9.3 % of heat capacity behind almost the same surface.
  2. Evaporation cut early. Adding the milk immediately lowers the coffee at once, by 5.9 °C with milk at room temperature, which lowers the vapour pressure at the surface, and evaporation with it, for all ten minutes.
  3. The extra effect of colder milk (level 2): its deficit decays with the drink, and the colder mixture evaporates even less. Fridge milk lowers the coffee by 7 °C at once, and the vapour pressure at the surface by 24 %.
Heat loss rate over time for both ordersBlack coffee starts losing 69 watts; with the milk added immediately the loss rate starts lower and stays lower all the way to minute ten.0246810020406080time after pouring [min]heat loss rate [W]added at the end (black)added immediately
How fast is heat leaving, minute by minute? The area under each curve is the heat lost. With the milk added immediately (solid) the coffee loses heat more slowly at every moment, which is the whole answer in one picture.

Adding the milk at minute 0, 1, 2 and so on, the final temperature decreases steadily from 65.5 °C (minute 0) to 63.0 °C (minute 10): the earliest moment is the best one.

What changed from the level before. The leak per degree now decreases as the coffee cools (τ = 22 minutes at 90 °C, 37 at 60 °C). With fridge milk the full model gives 65.5 and 63.0 °C, Δ = 2.5 °C, against 1.7 °C on the fixed-K model of levels 2 and 3; with milk at room temperature, 2.0 °C against level 1's 1.2 °C. The fixed-K model has the right sign and a size about 32 % low, because it keeps the hottest, leakiest K for the whole wait.

Level 5: the shape of the mug

The follow-up. “Does the shape of the mug matter?”

Our seven mugs: a straight mug 8 cm across; a tall thin glass 5.5 cm and a wide cup 10.5 cm across; a cone widening at α=15° from a base radius r0=3 cm, and one narrowing at 15° from r0=5 cm; a round bowl of radius R=6.5 cm; and a straight mug 8 cm across that flares out at 55° from the coffee's surface upwards. All seven use the same convection correlations; for a mouth that narrows or flares that is an estimate, and level 3's robustness test shows every mug keeps the sign of its Δ when every air-side coefficient is halved or doubled.

The milk raises the level, so what matters is how the surface increases with volume. With coffee volume V, milk volume Vm and liquid depth z:

Cylinder of radius R: A=πR2 whatever the level. The milk raises the level by Vm/(πR2) (6 mm in our mug) and adds only a band of side wall.

Cone widening upwards, base radius r0, walls at angle α to the vertical: r(z)=r0+ztanα, so

A(z)=π(r0+ztanα)2,V(z)=π3tanα[(r0+ztanα)3−r03]

Eliminate z, because the milk is given as a volume, so the surface is wanted as a function of volume: from V(z), (r0+ztanα)3=r03+3Vtanα/π, and A is π times that to the power 2/3. (Another way: solve V(z) for the new height numerically, which is what the model does for every mug.)

A(V)=π(r03+3Vtanαπ)2/3

Write s(V)=r03+3Vtanα/π, the cube of the surface radius. Adding milk multiplies the surface by [s(V+Vm)/s(V)]2/3, always less than (1+Vm/V)2/3, the pure cone's limit: a 12 % increase in volume widens a cone's surface by at most 7.8 %, less than the 9.3 % the milk adds in heat capacity. So in our mug a smooth cone never makes Δ negative on its own (on its top surface; the wetted wall adds a little, but it loses far less heat per square centimetre).

Narrowing upwards: the same formulas with α<0. The surface shrinks as the level rises, so Δ is larger than in a straight mug.

Round bowl of radius R, filled to depth z: A=π(2Rz−z2) and V=πz2(R−z/3), so dA/dV=2(R−z)/(2Rz−z2): the surface increases quickly when the bowl is shallow and stops increasing at the equator.

Tall thin glass against wide cup, same volume: the wide cup has the larger surface, so a shorter time constant. A shorter time constant makes the ten minutes a larger part of the cooling, so lowering the temperature early saves more heat and Δ is larger.

A cup that flares out exactly at the coffee's surface: all of the milk's volume goes into the widening part, and the surface increases fast. This is the shape that makes Δ negative: on the full model the sign changes at a flare of about 41 degrees from the vertical with room temperature milk and 50 degrees with fridge milk.

Each row is 250 ml of coffee with 30 ml of fridge milk added.

MugDepth [cm]Top surface [cm²]Area added [%]Leak added [%]τ [min]Δ [°C]Better to add the milk
1. a straight mug5.050.3 → 50.38.82.3222.5immediately
2. a tall thin glass10.523.8 → 23.810.84.5311.8immediately
3. a wide cup2.986.6 → 86.66.61.3153.2immediately
4. a cup widening upwards5.663.5 → 67.18.65.6182.3immediately
5. a cup narrowing upwards4.048.7 → 44.57.9−3.4233.1immediately
6. a round bowl3.9111.7 → 115.85.33.5123.0immediately
7. a cup flaring out at the coffee's surface5.050.3 → 69.924.330.122−0.5at minute ten

Heat capacity added is 9.3 % for every mug (the same 30 ml). Every mug is given the same ceramic, 255 J K⁻¹ and a 4 mm wall, so only the shape differs. “Area added” counts the top and the wetted wall; “leak added” weights each by how much heat it actually loses, which is why the top, where evaporation happens, contributes most.

The seven mugs in cross-section, to scale1. straight mug: coffee 5.0 cm deep, 5.6 cm with the milk; 2. tall thin glass: coffee 10.5 cm deep, 11.8 cm with the milk; 3. wide cup: coffee 2.9 cm deep, 3.2 cm with the milk; 4. widening cup: coffee 5.6 cm deep, 6.0 cm with the milk; 5. narrowing cup: coffee 4.0 cm deep, 4.6 cm with the milk; 6. round bowl: coffee 3.9 cm deep, 4.2 cm with the milk; 7. flared at the surface: coffee 5.0 cm deep, 5.5 cm with the milk1234567
What do the seven mugs look like? Cross-sections to one scale, numbered as in the table: 1. straight mug; 2. tall thin glass; 3. wide cup; 4. widening cup; 5. narrowing cup; 6. round bowl; 7. flared at the surface. Solid line: the coffee's surface; dashed: the surface after the milk.
Leak added by the milk in each mug, against the heat capacity it addsThe dashed line is the heat capacity added, 9.3 percent, the same in every mug. A dot left of it favours adding the milk immediately on Newton cooling. 1. straight mug: 2.3 percent; 2. tall thin glass: 4.5 percent; 3. wide cup: 1.3 percent; 4. widening cup: 5.6 percent; 5. narrowing cup: -3.4 percent; 6. round bowl: 3.5 percent; 7. flared at the surface: 30.1 percent-100102030401. straight mug2. tall thin glass3. wide cup4. widening cup5. narrowing cup6. round bowl7. flared at the surfaceleak added by the milk [%]capacity +9.3 %
In which mugs does adding the milk immediately keep the coffee hotter, on Newton cooling alone? The milk adds 9.3 % to the heat capacity in every mug (dashed line). A mug whose leak grows by less, a dot to the left, favours adding the milk immediately; only the cup flaring at the surface lies to the right.
Mug 1, a straight mugMilk added immediately ends at 65.5, milk added at minute ten at 63.0.1. straight mug: Δ = +2.5 °C0246810455565758595time [min]temperature [°C]Mug 2, a tall thin glassMilk added immediately ends at 68.8, milk added at minute ten at 67.0.2. tall thin glass: Δ = +1.8 °C0246810455565758595time [min]temperature [°C]Mug 3, a wide cupMilk added immediately ends at 61.2, milk added at minute ten at 58.0.3. wide cup: Δ = +3.2 °C0246810455565758595time [min]temperature [°C]Mug 4, a cup widening upwardsMilk added immediately ends at 62.8, milk added at minute ten at 60.5.4. widening cup: Δ = +2.3 °C0246810455565758595time [min]temperature [°C]Mug 5, a cup narrowing upwardsMilk added immediately ends at 66.6, milk added at minute ten at 63.5.5. narrowing cup: Δ = +3.1 °C0246810455565758595time [min]temperature [°C]Mug 6, a round bowlMilk added immediately ends at 56.6, milk added at minute ten at 53.6.6. round bowl: Δ = +3.0 °C0246810455565758595time [min]temperature [°C]Mug 7, a cup flaring out at the coffee's surfaceMilk added immediately ends at 62.5, milk added at minute ten at 63.0.7. flared at the surface: Δ = −0.5 °C0246810455565758595time [min]temperature [°C]Mug 7, the last minute, zoomed inMilk added immediately ends at 62.5, milk added at minute ten at 63.0: adding it at minute ten leaves the coffee hotter.7, zoomed: the last minute99.51060626466687072time [min]temperature [°C]62.563.0
How does the mug's shape change the race? Temperature against time in each mug: milk added immediately solid, milk added at minute ten dashed, with 30 ml of fridge milk and 250 ml of coffee at 90 °C. Each title gives Δ, the final temperature with the milk added immediately minus that with the milk added at minute ten.
The difference Δ against the amount of milk, for each mugThe advantage of adding the milk immediately increases with the amount of milk in every mug except the flared cup, where it is below zero for moderate amounts.020406080-1012345678milk added [ml]difference Δ [°C]7241563
How much milk does it take for the order to matter? Each line is one mug, numbered as in the table. Δ is the final temperature with the milk added immediately minus that with the milk added at minute ten; positive means adding it immediately leaves the coffee hotter. With 5 ml of milk, Δ is at most 0.6 °C in any mug.

The results at this level that surprise people, each explained in full:

Assumptions added at this level, each with why it holds:

What changed from the level before. The surface can now change when the milk goes in. Against Δ = 2.5 °C in the straight mug (full model, fridge milk), Δ is between 1.8 and 3.2 °C in the five other mugs that do not flare, and −0.5 °C in the cup that flares at the surface.

Level 6: the other variables, one at a time

The follow-up. “What else changes the answer, and by how much?”

This level is a sensitivity sweep: the one complication it adds is that the base case's fixed values now move, each on its own, while the rest stay at the base case. All numbers are from the full model with fridge milk.

The difference Δ against milk added [ml]From 0.5 to 6.1 degrees across 5 to 120.2040608010012001234567milk added [ml]difference Δ [°C]The difference Δ against milk temperature [°C]From 2.6 to 1.7 degrees across 3 to 25.51015202501234567milk temperature [°C]difference Δ [°C]The difference Δ against room temperature [°C]From 2.5 to 2.6 degrees across 10 to 30.101520253001234567room temperature [°C]difference Δ [°C]The difference Δ against air motion [× still air]From 2.5 to 3.3 degrees across 1 to 4.11.522.533.5401234567air motion [× still air]difference Δ [°C]The difference Δ against starting temperature [°C]From 1.4 to 2.8 degrees across 70 to 95.70758085909501234567starting temperature [°C]difference Δ [°C]The difference Δ against waiting time [min]From 0.9 to 3.2 degrees across 2 to 30.5101520253001234567waiting time [min]difference Δ [°C]The difference Δ against mug width [cm]From 1.8 to 3.3 degrees across 5.5 to 11.6789101101234567mug width [cm]difference Δ [°C]
How much hotter does adding the milk immediately leave the coffee, as each variable changes? Δ is the final temperature with the milk added immediately minus that with the milk added at minute ten. One variable at a time from our base case (250 ml at 90 °C, 30 ml of fridge milk, room 20 °C, still air, ten minutes). The vertical scale is the same in every panel. The horizontal scales are in different units, so compare panels with the tornado chart below, which ranks the variables over realistic ranges.
Final temperature against milk added [ml]Milk added immediately: from 67.9 to 57.3 degrees across 5 to 120.204060801001204050607080milk added [ml]temperature at the end [°C]Final temperature against milk temperature [°C]Milk added immediately: from 65.4 to 66.4 degrees across 3 to 25.5101520254050607080milk temperature [°C]temperature at the end [°C]Final temperature against room temperature [°C]Milk added immediately: from 64.1 to 67.1 degrees across 10 to 30.10152025304050607080room temperature [°C]temperature at the end [°C]Final temperature against air motion [× still air]Milk added immediately: from 65.5 to 46.3 degrees across 1 to 4.11.522.533.544050607080air motion [× still air]temperature at the end [°C]Final temperature against starting temperature [°C]Milk added immediately: from 54.5 to 67.9 degrees across 70 to 95.7075808590954050607080starting temperature [°C]temperature at the end [°C]Final temperature against waiting time [min]Milk added immediately: from 78.4 to 48.2 degrees across 2 to 30.510152025304050607080waiting time [min]temperature at the end [°C]Final temperature against mug width [cm]Milk added immediately: from 68.8 to 60.3 degrees across 5.5 to 11.678910114050607080mug width [cm]temperature at the end [°C]
How hot is the coffee when you drink it, as each thing changes? Milk added immediately solid, milk added at minute ten dashed, same base case. Air motion and waiting time change the final temperature most (ranked in the chart below); the vertical distance between the two lines is Δ, plotted in the figure above.

How each graph changes:

Tornado chartair motion: +0.0 to -14.6 degrees; waiting time: +7.3 to -10.3 degrees; starting temperature: -5.2 to +2.4 degrees; milk amount: +1.9 to -4.7 degrees; mug width: +2.8 to -4.3 degrees; room temperature: -0.7 to +0.8 degrees; milk temperature: -0.1 to +0.7 degreesair motion-14.6waiting time+7.3-10.3starting temperature-5.2+2.4milk amount+1.9-4.7mug width+2.8-4.3room temperature-0.7+0.8milk temperature-0.1+0.7change from the base case [°C]
Which variable matters most for how hot the coffee is? Change in the final temperature (milk added immediately) when each variable moves across its range, all else fixed. Ranges: air motion still air to three times the coefficient; waiting time 5 to 20 minutes; starting temperature 80 to 95 °C; milk amount 10 to 80 ml; mug width 6 to 10.5 cm across; room temperature 15 to 25 °C; milk temperature 3 to 20 °C.
Tornado chartmilk amount: -1.6 to +2.5 degrees; air motion: +0.0 to +0.8 degrees; waiting time: -0.8 to +0.5 degrees; mug width: -0.6 to +0.6 degrees; starting temperature: -0.6 to +0.3 degrees; milk temperature: +0.1 to -0.6 degrees; room temperature: -0.0 to +0.0 degreesmilk amount-1.6+2.5air motion+0.8waiting time-0.8+0.5mug width-0.6+0.6starting temperature-0.6+0.3milk temperature+0.1-0.6room temperaturechange from the base case [°C]
Which variable matters most for whether the timing matters? Change in Δ when each variable moves across its range, all else fixed. The amount of milk changes Δ the most. Ranges: milk amount 10 to 80 ml; air motion still air to three times the coefficient; waiting time 5 to 20 minutes; mug width 6 to 10.5 cm across; starting temperature 80 to 95 °C; milk temperature 3 to 20 °C; room temperature 15 to 25 °C.

The mug's material changes the final temperature too: the same comparison in a paper cup ends at 62.8 °C, in a double-walled vacuum mug at 67.3 °C (milk added immediately in both, against 65.5 °C in ceramic). In the model the paper cup has a heat capacity of 17 J K⁻¹, against the ceramic's 255; the vacuum mug has 60 J K⁻¹ and loses nothing through its sides.

Milk added immediately, in three kinds of cupceramic mug: 65.5 degrees at minute ten; paper cup: 62.8 degrees at minute ten; double-walled vacuum mug: 67.3 degrees at minute ten02468105565758595time after pouring [min]temperature [°C]vacuum, ends 67.3 °Cceramic, ends 65.5 °Cpaper, ends 62.8 °C
Does the cup's material matter? Milk added immediately in each. The vacuum mug (solid) loses almost nothing through its sides; the paper cup (dotted) holds little heat itself; the ceramic mug (dashed) is our base case.

The results at this level that surprise people, each explained in full:

Assumptions added at this level, each with why it holds:

What changed from the level before. Level 5's base case had every variable fixed (Δ = 2.5 °C). Moved one at a time over realistic ranges, the amount of milk changes Δ the most (0.5 to 6.1 °C), and air motion and the waiting time change the final temperature the most (down to 46.3 °C and 48.2 °C).

Level 7: the cup left for hours

The follow-up. “What happens if nobody drinks it?”

On Newton's law the coffee approaches the room and never goes below it. With evaporation it does go below: this is the most complete case on the page, because every route of level 3 and the humidity of the room decide where it ends.

The result at this level that surprises people, explained in full:

Assumptions added at this level, each with why it holds:

What changed from the level before. Levels 1 to 6 stop at minute ten, with the coffee at 65.5 °C. Run for hours, Newton's law says the coffee approaches the room temperature and never goes below it; with evaporation, an open cup settles about 1 °C below the room, after several hours.

What an interviewer is testing at each level

Thinking aloud (more in how to approach an interview question). Say assumptions out loud before calculating, solve the simplest model fully, check limiting cases (no milk; milk at the coffee's temperature; a mug with no leak) and units, then say what would change the answer, and in which direction. Students often worry that thinking aloud sounds unclear: it does not, and a short silence while you think is fine too.

What students often get wrong

Where students get stuck, and the step that unsticks it

More follow-up questions an interviewer may ask

Check it against your kitchen

The model predicts about 3.2 °C a minute of cooling at first, black coffee at 68.3 °C after ten minutes and 36 °C after an hour, and about 8 g of water gone as vapour in the first ten minutes. Pour two identical mugs from one pot, add milk to one immediately and to the other after ten minutes, and read both with a kitchen thermometer; weigh a third mug of black coffee before and after. The least certain part of the model is evaporation (good to a factor of about two); none of the uncertainties changes the sign of Δ in a straight mug.

Where this puzzle comes from

This is a classic, argued over in kitchens and staff rooms for generations. Jearl Walker's Amateur Scientist column in Scientific American took the physics of a cup of coffee or tea as its subject in November 1977, and W. G. Rees and C. Viney studied the cooling of tea and coffee in theory and with household equipment in the American Journal of Physics (volume 56, 1988), concluding, as we do, that adding the milk at once usually keeps the drink hotter, though not in every case. The explanation, model, numbers and figures on this page are our own.

Try Mintnib Pro free for the ESAT

Keep going