Pick any number you like and ask whether it's rational. Usually the definition tells you right away. But π is strange: everywhere we meet it, it looks like it's begging to be a ratio. It's the ratio of a circle's circumference to its diameter — literally a quotient. It's a limit of perfectly rational approximations (3, 22/7, 355/113, …) that get closer and closer without ever quite arriving. It shows up as an area, an arc length, a root of trigonometric equations built from power series with rational coefficients. Nothing in its birth certificate screams "irrational" the way, say, √2 does when you corner it with the Pythagorean theorem and a parity argument.
And yet π is irrational — not just irrational, but transcendental. The first fact was proved by Johann Lambert in 1761 using continued fractions, and it has a reputation for being hard. So it's a small shock to find, tucked into a "starred," skippable chapter of Michael Spivak's Calculus, a proof that needs nothing more exotic than derivatives, the Fundamental Theorem of Calculus, and two facts about sine you already know cold. Spivak calls the chapter a diversion from the main road of the book. It's also one of the cleverest ten pages in it.
Why This Isn't Supposed to Be Easy
The trouble with proving π irrational is that π doesn't hand you an equation to work with the way √2 does. To show √2 is irrational, you assume √2 = p/q in lowest terms, square both sides, and chase parity until you hit a contradiction. There's an algebraic identity sitting right there waiting to be exploited. π, by contrast, is defined analytically — as half the period of sine, or as the smallest positive root of sin x = 0, or via a limit of polygon areas. None of those descriptions is an algebraic equation you can just manipulate.
So any proof has to manufacture an algebraic contradiction out of analytic material: derivatives, integrals, growth rates. That mismatch is exactly why the proof feels like a magic trick even after you've checked every line. Spivak is admirably honest about this up front, warning that "the motivation for many steps in our proof cannot be supplied" — you can verify each step, but the proof was clearly reverse-engineered from its conclusion. This particular version is due to Ivan Niven (1947), and it's the cleanest one anybody has found in nearly 80 years.
The Star of the Show: A Polynomial Built to Order
Everything hinges on one strange little function:
fn(x) = xn(1 − x)n / n!
At first glance it looks arbitrary, but it's engineered to satisfy two properties simultaneously. First, on the interval (0, 1) it's squeezed tightly: 0 < fn(x) < 1/n!. Second — and this is the part that takes real work to see — if you multiply out xn(1 − x)n, every coefficient is an integer, and the powers of x run only from n up to 2n. Grinding through the derivatives at x = 0 and x = 1, Spivak shows that every derivative fn(k)(0) and fn(k)(1), for every order k, is a whole number. That's the entire payload of the function: it's a bridge between the world of smooth, continuous calculus and the world of integers, where contradictions are easy to spot (there's no integer strictly between 0 and 1).
Smuggling in the Assumption
Now suppose, for contradiction, that π2 = a/b for positive integers a and b (Spivak works with π2 rather than π itself, since a rational π would force π2 rational too — so it suffices to rule out the stronger statement). Using a, b, and fn, build a second function:
G(x) = bn[π2nfn(x) − π2n−2fn″(x) + π2n−4fn(4)(x) − ··· ± fn(2n)(x)]
It looks intimidating, but every factor of the form bnπ2n−2k equals bn(a/b)n−k = an−kbk, which is an integer precisely because we assumed π2 is rational. Combine that with the fact that the derivatives of fn are integers at 0 and 1, and you get that G(0) and G(1) are both integers. The assumption that π is rational has been converted into a concrete arithmetic fact about a function we control.
Where Sine Earns Its Keep
Here's the trick that makes the whole argument click. Define H(x) = G′(x) sin(πx) − πG(x) cos(πx). A short computation using G″ + π2G = π2anfn (which follows from how G was built) shows that H′(x) = π2anfn(x) sin(πx). By the Fundamental Theorem of Calculus, that means:
π ∫01 anfn(x) sin(πx) dx = H(1) − H(0) = π[G(1) + G(0)],
so the integral on the left is G(1) + G(0) — an integer, guaranteed by the work in the last section. Sine's only job here is to supply an antiderivative that closes the loop back to an integer; the proof uses nothing about sine beyond sin″ = −sin, sin(0) = 0, and cos(0) = 1.
The Squeeze That Ends It
Now compare that fact to the size of the integral. Since 0 < fn(x) < 1/n! on (0, 1) and sin(πx) is positive there too, the integral π∫01 anfn(x) sin(πx) dx is trapped strictly between 0 and πan/n!. And here is the second essential trick: no matter how big a is, the quantity an/n! shrinks to 0 as n grows, because factorial growth eventually crushes any fixed exponential. So for n large enough, the integral is squeezed into the open interval (0, 1) — while we already proved it must be an integer. There is no integer strictly between 0 and 1. Contradiction. The assumption that π2 is rational collapses, and π is irrational.
What's striking, as Spivak points out afterward, is how little of "π-ness" the argument actually uses. Re-examine every step and the only fact about π that gets used is sin(π) = 0. The proof is really showing that the smallest positive zero of any function satisfying sin″ + sin = 0, sin(0) = 0, sin′(0) = 1 must be irrational — it just happens that this smallest positive zero is π by definition. You can prove a famous constant irrational without ever writing down what makes it famous.
If you want to see every algebraic step spelled out in full — including how the derivative bookkeeping for fn actually gets nailed down, and the problems that follow with Viète's 1579 infinite product for π — Chapter 16 of Spivak's Calculus, 4th Edition is exactly nine pages, and it rewards reading with a pencil in hand.
Further reading: How to self-study Spivak's Calculus · Table of contents · Index · Calculus, 4th Edition · Answer Book
If you could only smuggle one extra fact about a mystery constant into a proof like this — the way this one smuggles in sin(π) = 0 — what would you want it to be, and what could you prove with it?
