How Many Ping-Pong Balls Fill a Jumbo, and Why Thirty Million Is the One Impossible Answer

Dividing the cabin by the ball gives 29.8 million, which is exact arithmetic on the assumption that spheres tile space. They do not, and the correction is pinned on both sides by constants: a plain cubic grid anyone can build holds exactly 15,625,000 balls, and no arrangement whatever beats pi over root eighteen, which caps the count at 22.1 million. The answer is that interval, with a settled pour at 19.1 million sitting inside it.

A table tennis ball is 40 mm across. Take the cabin of a jumbo to be 1000 cubic metres. How many balls fit inside?

Almost everybody produces the same first number, and it is the one answer the geometry forbids. The interesting part of this problem is not the arithmetic. It is that the honest output is a range whose two endpoints are exact constants, with an empirical figure sitting between them, so the estimate can be defended without knowing anything about the aircraft.

Dividing the box by the ball

A regulation ball has radius $r = 2$ cm, so

v \;=\; \tfrac{4}{3}\pi r^3 \;=\; \tfrac{4}{3}\pi \cdot 2^3 \;=\; \frac{32\pi}{3} \;=\; 33.5103\ \text{cm}^3

(1)

A metre is a hundred centimetres and the conversion cubes, so 1000 cubic metres is exactly $10^9$ cubic centimetres. Divide:

\frac{10^9}{32\pi/3} \;=\; 29.84 \times 10^6

(2)

Just under thirty million. The division is exact and the answer is wrong, which is the combination worth noticing. Nothing was miscalculated. What was assumed is that spheres tile space, and they do not. Air is left over between them no matter how they are arranged, and the leftover is not a rounding error.

Fig. 1 — The gaps are the whole problem. On the right, the cell one ball owns in a plain cubic arrangement: a 4 cm cube of which the ball occupies π/6.

The quantity the naive division sets to one

Packing density — the fraction of space actually filled

For an arrangement of non-overlapping equal spheres, the packing density $\eta$ is the fraction of the container's volume lying inside a sphere. The count is then $N = \eta V / v$ , and the naive division is the special case $\eta = 1$ , which no arrangement of spheres in three dimensions attains.

So the real formula carries one extra factor, and every honest version of this estimate is a statement about that factor:

N \;=\; \eta\,\frac{V}{v} \;=\; \eta \cdot \frac{10^9}{32\pi/3}

(3)

Three values of $\eta$ matter here, and two of them are exact.

A count with no packing fraction in it at all

Start with the arrangement anyone can build. Lay the balls out on a plain cubic grid, so each one sits at the centre of a 4 cm cube and touches its six neighbours. Then the count needs no density argument whatsoever: each ball owns 64 cubic centimetres, and

N_{\text{cubic}} \;=\; \frac{10^9}{4^3} \;=\; \frac{10^9}{64} \;=\; 15{,}625{,}000

(4)

Feeding the simple cubic density $\eta = \pi/6$ into equation (3) returns the same integer, since the $\pi$ cancels against the one in the ball volume. Two routes, one from counting cells and one from a published density, agree exactly. That is a real check rather than a restatement, because the first route never mentions $\pi$ .

This number is a floor, and the distinction matters. It is not a rival estimate to be averaged with the others. It is a packing you could physically construct, so the true answer for any sensible pour cannot be below it. Building the arrangement in a one-metre cube gives $25^3 = 15{,}625$ balls, which scales to the cabin figure exactly.

The ceiling is a theorem

At the other end, the densest packing of equal spheres in space has density

\eta_{\max} \;=\; \frac{\pi}{\sqrt{18}} \;=\; \frac{\pi}{3\sqrt{2}} \;=\; 0.74048\ldots

(5)

Gauss showed in 1831 that no lattice arrangement beats it. The much harder statement, that no arrangement of any kind beats it, is the Kepler conjecture, and Thomas Hales proved it in work announced in 1998 and published in 2005. So the ceiling here is not a conservative habit of mind, it is a bound with a proof behind it: no arrangement of balls puts more than 22.1 million of them in 1000 cubic metres.

The bound is actually generous, because $\pi/\sqrt{18}$ is a density for infinite space. Any finite container loses volume near its walls, where the lattice cannot register perfectly against the boundary, so a real box stays strictly under the bulk value. The wall loss is not monotone in the box size either: a 160 cm box and a 240 cm box register identically against the same lattice, and their counts differ by exactly $1.5^3$ .

Where the answer actually lives

Between the two exact constants sits the empirical one. Pour equal spheres into a container and shake or tap until they stop settling, and they reach a reproducible density near 0.64, the random close packing figure. Applying it gives 19.1 million. Leave the pile completely alone, which is what happens with light hollow balls, and it sits nearer random loose packing at about 0.60, giving 17.9 million.

The two exact fractions and the two empirical ones are strictly ordered:

\underbrace{\tfrac{\pi}{6}}_{0.5236} \;<\; 0.60 \;<\; 0.64 \;<\; \underbrace{\tfrac{\pi}{\sqrt{18}}}_{0.7405} \;<\; 1 \qquad\Longrightarrow\qquad 15.6\times 10^6 \;\le\; N \;\le\; 22.1\times 10^6

(6)

Fig. 2 — The dashed bar is the impossible one. Everything below it is achievable, and the top achievable bar is a proved bound.

The bracket is the claim. The 19.1 million is a point inside it, chosen because 0.64 is the best-established figure for a settled pour, and moving to 0.60 changes nothing structural because 17.9 million is inside the same interval. Reporting a single digit here would be reporting the softest input as if it were the hardest.

One tempting objection turns out to help. Balls at the bottom of a thousand cubic metres of pile would be squashed by the weight above them, so the arrangement is not one of rigid spheres. Deformation can only raise the count, so it cannot break a ceiling derived for hard spheres, and equation (6) survives.

The cabin figure is a guess, and the count is linear in it

Every number above inherits the 1000 cubic metres, and that was the one quantity nobody measured. Equation (3) is exactly proportional to $V$ , so a reader with a better figure rescales by the same factor and no part of the packing argument changes. A real jumbo's main and upper decks come to something nearer 850 cubic metres, which moves the settled-pour answer to about 16 million.

That sensitivity is where the method stops working, and it is worth saying plainly. The count goes as the cube of a length, so guessing the cabin's linear dimensions 20 percent too large inflates the answer by more than 70 percent. Volume-to-volume ratios amplify small errors in a way that surprises people who are used to estimating lengths. The packing fraction, meanwhile, is confined to a window of width 0.22 by mathematics. In this problem the geometry is the part you can trust and the container is the part you cannot, which is the opposite of most people's intuition when they start.

Sources and further reading

Thomas C. Hales, “A proof of the Kepler conjecture”, Annals of Mathematics162 (2005), 1065–1185, cited for its own sake: the statement that $\pi/\sqrt{18}$ cannot be beaten.
Thomas Hales et al., “A formal proof of the Kepler conjecture”, Forum of Mathematics, Pi 5 (2017), the machine-checked version of the same result.
Background on the problem and on Gauss's 1831 lattice case: Kepler conjecture and Sphere packing
The empirical fraction and why it is reproducible: Random close packing
The genre, and why the deliverable is an interval: Fermi problem

The exact figures here were recomputed symbolically, and the lattice count was checked by placing sphere centres in a one-metre cube, verifying pairwise disjointness and counting what remained inside the walls rather than by quoting a density. No claim is made about the true number of balls that fits in any particular aircraft.