π is in the bell curve, and there is no circle anywhere in it

By the 42Math team · 26 July 20264 min read

A tool on this site finds π seven different ways. Only two of them involve drawing a circle.

Everyone learns π as a fact about circles: the circumference divided by the diameter, a number a little over three. That definition is true and it is also the least interesting thing about it.

The π secrets tool finds the same number seven ways: polygons, a rolling wheel, onion rings, colliding blocks, dropped needles, random points, and an infinite sum. Two of those are about circles. The rest are not, and the fact that they agree is the thing worth explaining.

Needles on a floor

Buffon's needle, from 1777, is the cleanest case. Rule parallel lines on the floor a distance d apart. Drop a needle of length ℓ ≤ d at random. The probability that it crosses a line is

P = 2ℓ / (πd)

So drop a few thousand needles, count the crossings, and you have measured π with a matchstick and a floorboard. No circle was drawn, no radius measured, nothing curved was involved at any point.

π got in through the angle. Whether a needle crosses depends on its orientation, and averaging over all orientations means integrating over a full turn, and a full turn is 2π radians by definition. Every appearance of π in probability that involves "in a random direction" comes from that same integral.

Ramaley's 1969 note makes the point unimprovably. Bend the needle into any shape you like — a noodle — and the expected number of crossings depends only on its total length, not its shape. The result survives because it never depended on straightness in the first place, only on the average over directions. That is what it looks like when a constant belongs to rotation rather than to geometry.

The bell curve, which has no angle at all

Now the case with no rotation in sight. The normal distribution the normal tool draws is

f(x) = (1/√(2π σ²)) · e^{−(x−μ)²/(2σ²)}

A π, sitting in the front of the most-used formula in statistics, describing exam marks and measurement errors and the heights of conscripts. Nothing here is circular.

It arrives through a trick worth knowing, because it explains rather than asserts. To make the curve integrate to 1 you need ∫e^−x²dx, which has no elementary antiderivative. The standard escape is to square it — turning one integral over a line into a double integral over the plane — whereupon the exponent becomes e^−(x²+y²). And x² + y² is a radius. The moment the sum of two squares appears, the natural move is polar coordinates, the angle integrates to 2π, and the answer is √π.

So π entered the bell curve because the square of the function depends on x² + y², and a quantity that depends only on x² + y² is a quantity that does not care about direction. Circularity was hiding inside the algebra rather than in the picture.

What π is actually about

That is the pattern, and once you have it the other appearances stop being surprising.

π shows up wherever there is rotation: an average over all directions, a full turn, an oscillation coming back to where it started. And it shows up wherever there is a sum of two squares, because that is the algebra of distance, and distance is what does not change when you turn. Circles are simply the most familiar object with both properties, which is why they got the naming rights.

The colliding-blocks demonstration on that page is the strangest member of the family: two blocks with a mass ratio of 100ⁿ and a wall produce exactly the first n + 1 digits of π in the number of collisions. It looks like a coincidence and it is not. Elastic collisions conserve kinetic energy, which is a sum of squares of velocities, so the state of the system lives on a circle in velocity space, and each collision is a fixed rotation around it. The number of collisions is the number of steps before you have gone all the way round. It is Buffon's argument again, wearing a different hat.

One practical warning, since two of these methods are simulations. Random-point estimates converge at the pace all Monte Carlo does: the error falls as 1/√n, so each extra correct digit costs a hundred times the work. As a way to compute π they are hopeless, and the polygon method that Archimedes ran by hand beats them badly. As a demonstration that π lives outside geometry, they are the best thing on the page.

Published 26 July 2026 · corrections welcome via the corrections page.