From the factorial to the primes
The first two guides of this rung built the gamma function — the smooth complex interpolation of the factorial — and tamed its growth with the reflection formula and Stirling's asymptotics. The gamma function was a single function dressed up in many forms. The Riemann zeta function we now meet is a different and deeper kind of object: it starts life as an innocent-looking infinite sum, yet it turns out to braid together two worlds that seem to have nothing to do with each other — the additive world of counting (1, 2, 3, 4, ...) and the multiplicative world of the primes (2, 3, 5, 7, 11, ...). The bridge between them is a single equation, and it is the subject of this guide.
Here is the definition. For a complex number s = sigma + i t (number theorists write the variable as s, with real part sigma and imaginary part t — a tradition we will keep), the Riemann zeta function is the sum zeta(s) = sum over n >= 1 of 1/n^s = 1 + 1/2^s + 1/3^s + 1/4^s + .... Each term 1/n^s is the complex power n^(-s) = e^(-s log n), where log n is the ordinary real logarithm, so there is no branch ambiguity at all — n is a positive integer and log n is one real number. The only question is when this infinite sum converges, and to that we now turn.
Where the series lives: the half-plane Re(s) > 1
Convergence is decided entirely by the REAL part of s. The size of a term is |1/n^s| = |n^(-s)| = |e^(-s log n)| = e^(-sigma log n) = n^(-sigma), where sigma = Re(s). The imaginary part i t only rotates each term by the phase e^(-i t log n); it spins the term around the circle of radius n^(-sigma) but never changes its length. So the sum of absolute values is exactly sum of 1/n^sigma — the familiar real p-series — which converges precisely when sigma > 1. That single inequality is the whole story of where the raw series makes sense.
On the half-plane Re(s) > 1 the function is not just defined but genuinely holomorphic. The series converges absolutely there, and on any compact piece staying to the right of, say, sigma = 1 + delta the convergence is uniform — every term is bounded by n^(-(1+delta)), a fixed convergent series independent of t. By the theorem that a locally uniform limit of holomorphic functions is again holomorphic, zeta(s) is holomorphic on the open half-plane Re(s) > 1. It is the cleanest member of a whole family: a Dirichlet series.
Dirichlet series and the abscissa of convergence
A Dirichlet series is any sum of the shape sum over n >= 1 of a_n / n^s, where a_n is a sequence of complex coefficients. Zeta is the special case where every a_n = 1. These play the same organizing role for arithmetic that ordinary power series sum a_n (z - z_0)^n play for analysis: just as a power series converges inside a disk whose radius is its radius of convergence, a Dirichlet series converges in a HALF-PLANE Re(s) > sigma_c, to the right of a vertical line. That cutoff sigma_c is called the abscissa of convergence.
The geometry is worth pausing on, because it is a recurring surprise. For power series the boundary of convergence is a CIRCLE, because |z^n| depends only on |z|; for Dirichlet series the boundary is a vertical LINE, because, as we just saw, |n^(-s)| depends only on Re(s). One subtlety carries over from the real theory and is sharper here: the abscissa of ABSOLUTE convergence can sit strictly to the right of the abscissa of plain convergence, so a Dirichlet series can converge conditionally in a thin strip where it does not converge absolutely. For zeta both abscissae equal 1, so we need not fuss over the distinction yet — but it becomes the heart of the matter for the alternating cousin of zeta later on.
Euler's product over the primes
Now comes the discovery, due to Euler in 1737, that makes zeta extraordinary rather than merely convergent. For Re(s) > 1, the sum over ALL the integers equals a product over ONLY the primes: zeta(s) = product over primes p of 1/(1 - p^(-s)). On the left, every whole number; on the right, only 2, 3, 5, 7, 11, and so on. The equals sign between an additive sum and a multiplicative product is the Euler product, and it is the precise analytic statement that every integer factors into primes in exactly one way.
Where does it come from? Expand each factor as a geometric series, which is legitimate because |p^(-s)| = p^(-sigma) < 1 when sigma > 1. The factor for the prime p becomes 1/(1 - p^(-s)) = 1 + p^(-s) + p^(-2s) + p^(-3s) + ..., a sum over all powers of that one prime. Now multiply these geometric series together over every prime and ask what a single term of the expanded product looks like.
1/(1 - 2^-s) = 1 + 2^-s + 4^-s + 8^-s + ... 1/(1 - 3^-s) = 1 + 3^-s + 9^-s + ... 1/(1 - 5^-s) = 1 + 5^-s + ... Pick one term from each row and multiply: 2^-s * 3^-s * 1 * 1 * ... = (2*3)^-s = 6^-s 4^-s * 1 * 5^-s * ... = (4*5)^-s = 20^-s Each choice 2^a * 3^b * 5^c * ... = n^-s hits exactly one integer n, ONCE (unique prime factorization). Summing over all choices rebuilds 1 + 2^-s + 3^-s + ... = zeta(s).
To pick one term from the expanded product you choose a power p^(-a*s) from each prime's geometric series — a 2-power, times a 3-power, times a 5-power, and so on — and multiply them. The product is (2^a * 3^b * 5^c * ...)^(-s) = n^(-s) for the integer n = 2^a * 3^b * 5^c * .... The Fundamental Theorem of Arithmetic says every n arises from exactly one such choice of exponents, so as we range over all choices we hit each 1/n^s once and only once. Summing them all back up reconstructs 1 + 1/2^s + 1/3^s + ... = zeta(s). The unique factorization of the integers IS the Euler product.
Why the product matters: zeta knows the primes
The Euler product is not a curiosity; it is the doorway to analytic number theory, and it earns that status by a single, immediate consequence. The product is an infinite product of nonzero factors, each 1/(1 - p^(-s)), and a convergent infinite product of nonzero terms cannot equal zero. Therefore zeta(s) is NEVER ZERO on the half-plane Re(s) > 1. A whole open region is swept clean of zeros for free, purely because every integer factors into primes — a fact about arithmetic delivering a fact about a holomorphic function.
Run the logic the other way and you can hear the primes inside zeta. Take logarithms of the Euler product to turn the product into a sum: log zeta(s) = sum over primes p of -log(1 - p^(-s)) = sum over primes p of (p^(-s) + (1/2) p^(-2s) + ...). To leading order log zeta(s) is roughly sum over primes of p^(-s), a sum running over the primes ALONE. So the analytic behaviour of zeta near s = 1 — where, recall, the harmonic series diverges and zeta blows up — is controlled by how the primes are distributed. This is exactly the thread that, pulled hard enough, yields the prime number theorem: the count of primes up to x is asymptotic to x / log x.