The Riemann zeta function \zeta(s) = \sum_{n\ge1} n^{-s} only converges when \Re(s) > 1. Yet Riemann, in his nine-page bombshell of 1859, showed it can be continued to almost the whole complex plane — and, more astonishingly, that the continued function possesses a hidden mirror symmetry. Its value at s is rigidly locked to its value at 1-s, the point reflected across the vertical line \Re(s) = \tfrac12.

This is the functional equation. It is the beating heart of analytic number theory: it manufactures the trivial zeros out of thin air, it explains why the critical line sits at \Re(s) = \tfrac12 and nowhere else, and it turns a series that only makes sense on the right half-plane into a single object that knows about the left half-plane too. Let's build it, prove it the way Riemann did, and read off its consequences.

The symmetric form — a function that equals its own reflection

The cleanest statement uses the completed zeta function (also called the Riemann \xi-function in this normalisation). You take \zeta(s) and dress it with a Gamma factor and a power of \pi:

Read the symmetry out loud: whatever \xi does at a point, it does the identical thing at the mirror-image point on the other side of \Re(s) = \tfrac12. The awkward-looking factors \pi^{-s/2}\Gamma(s/2) are not decoration — they are precisely the extra machinery needed to make the raw \zeta symmetric. Where do they come from? Riemann pulled them out of a single, gorgeous identity about a sum of Gaussians.

Picturing the reflection

Before the proof, fix the geometry in your mind. In the complex plane, write s = \sigma + it. The map s \mapsto 1-s sends \sigma + it \mapsto (1-\sigma) - it: it reflects the real part across \sigma = \tfrac12 and flips the sign of the imaginary part. The two points s and 1-s straddle the line \Re(s) = \tfrac12, with their midpoint sitting exactly on it. Step through:

A point on the line — where \sigma = \tfrac12 — is its own mirror image up to the sign of t. That is why the line \Re(s) = \tfrac12 is called the critical line: it is the fixed axis of the whole symmetry, the natural place for the non-trivial zeros to congregate.

Riemann's engine: the Jacobi theta function

Define the theta function, a sum of Gaussians sampled at every integer:

This object has a miraculous modular transformation — a Poisson summation identity that relates \theta at t to \theta at 1/t:

This single identity is where the whole symmetry s \leftrightarrow 1-s is born: inverting t \mapsto 1/t inside the integral is exactly what will become inverting s \mapsto 1-s on the outside. It is worth pausing on how strange it is — a sum that decays slowly for small t and a sum that decays fast for large t turn out to be the same sum, rescaled.

The Mellin transform: from theta to zeta

Now the bridge. Take a single term of the completed sum and integrate it against t^{s/2-1}. Starting from the Gamma integral \Gamma(s/2) = \int_0^\infty t^{s/2-1}e^{-t}\,dt and substituting t \to \pi n^2 t gives, for each n \ge 1,

Sum over n = 1, 2, 3, \dots. The left side assembles into \pi^{-s/2}\Gamma(s/2)\zeta(s) = \xi(s); the right side gathers the Gaussians into the "one-sided" theta piece. Writing \psi(t) = \sum_{n\ge1} e^{-\pi n^2 t} = \tfrac{\theta(t)-1}{2} (the n=0 term is 1, and the sum is symmetric in \pm n):

This is a Mellin transform — the multiplicative cousin of the Fourier transform — and it is the exact object we can now feed Jacobi's identity into.

Split at t = 1 and symmetrise

The integral runs from 0 to \infty. Split it at t = 1 into a "small-t" piece and a "large-t" piece:

In the first integral, substitute t \mapsto 1/t and apply Jacobi's transformation \theta(1/t) = \sqrt{t}\,\theta(t). After the dust settles, the small-t divergences are tamed and the whole thing collapses into a beautifully symmetric shape:

Look at the right-hand side: swapping s \leftrightarrow 1-s leaves the integrand's bracket t^{s/2-1} + t^{(1-s)/2-1} unchanged, and it leaves -\tfrac{1}{s(1-s)} unchanged too (since s(1-s) = (1-s)s). The expression is manifestly symmetric — so \xi(s) = \xi(1-s). That is the functional equation, and Riemann's proof, in one card. As a bonus, the -\tfrac{1}{s(1-s)} term openly displays the two simple poles of \xi at s=0 and s=1; the integral converges for all s, so it hands us the analytic continuation for free.

The asymmetric form — what you actually compute with

Unpack \xi(s) = \xi(1-s) by writing out both Gamma factors and using the reflection and duplication formulae for \Gamma. Solving for \zeta(s) alone gives the form most often quoted:

This is the same symmetry in disguise, but now expressed as an explicit formula for \zeta(s) in terms of \zeta(1-s). Notice the newcomer: a factor of \sin(\pi s/2). That sine is about to do something wonderful — it will hand us an entire infinite family of zeros for free.

Reading off the trivial zeros

Look at the factor \sin(\pi s/2) in the asymmetric equation. It vanishes whenever s/2 is an integer, i.e. at every even integer s = \dots, -4, -2, 0, 2, 4, \dots. For each negative even integer s = -2, -4, -6, \dots, the other factors on the right (2^s, \pi^{s-1}, \Gamma(1-s), \zeta(1-s)) are all finite and non-zero, so the whole product is zero:

These are the trivial zeros — trivial precisely because the functional equation tells you exactly where they are, no mystery required. (The positive even integers s = 2, 4, \dots are not zeros: there the vanishing sine is exactly cancelled by a pole of \Gamma(1-s) on the right — more on that danger in the "Watch out!" below.)

Worked example — \zeta(-2) = 0. Put s = -2 into the asymmetric equation:

Every factor is finite, and \sin(-\pi) = 0, so \zeta(-2) = 0. The same collapse happens at s = -4 (\sin(-2\pi)=0), at s = -6, and so on down the negative even integers forever.

Worked example — tying ζ(2) to ζ(−1)

The functional equation lets you leap from a known value on one side of the critical line to a value on the other. Everyone knows the Basel result \zeta(2) = \tfrac{\pi^2}{6}. Let's use it to pin down \zeta(-1). Set s = -1 in the asymmetric form, so 1-s = 2:

Now evaluate each factor: 2^{-1} = \tfrac12, \sin(-\pi/2) = -1, \Gamma(2) = 1! = 1, and \zeta(2) = \tfrac{\pi^2}{6}. Substituting:

So \zeta(-1) = -\tfrac{1}{12} — the notorious value that, read carelessly, gets written as the "sum 1+2+3+\cdots = -\tfrac{1}{12}." It is nothing of the sort: it is the value of the continued \zeta at s=-1, delivered here directly by the functional equation from the honest, convergent Basel sum \zeta(2) = \tfrac{\pi^2}{6} on the other side of the line.

The whole picture in one breath

Riemann's chain of ideas: the theta function \theta(t) = \sum e^{-\pi n^2 t} obeys the modular law \theta(1/t) = \sqrt t\,\theta(t); feeding it through the Mellin transform turns \pi^{-s/2}\Gamma(s/2)\zeta(s) into an integral; splitting that integral at t=1 and using the modular law makes the result visibly symmetric under s \leftrightarrow 1-s. Out drops \xi(s) = \xi(1-s), and with it the analytic continuation of \zeta, the trivial zeros, the value \zeta(-1) = -\tfrac1{12}, and the special status of the critical line — all from one identity about a sum of Gaussians. That is why this equation, more than any single formula in the subject, deserves to be called deep.