The Riemann Hypothesis says that every non-trivial zero of \zeta(s) sits exactly on the critical line \Re(s) = \tfrac12. For fifty years after Riemann's 1859 paper, nobody could prove that even one zero lands there. Numerics found zeros on the line, but a computed value near \tfrac12 is not a proof that the real part is exactly \tfrac12 — it might be 0.5000001.

In 1914, G. H. Hardy broke the deadlock. He proved that \zeta has infinitely many non-trivial zeros lying precisely on the critical line. It was the first genuine progress toward the Riemann Hypothesis, and — as we'll be careful to stress — it is dramatically weaker than the Hypothesis itself. But the device Hardy invented to do it, a way of turning \zeta on the line into an honest real-valued function, is still the engine behind every large-scale zero computation today.

The obstacle: a complex value can't "cross" zero

Here is the difficulty. On the critical line we substitute s = \tfrac12 + it and study \zeta(\tfrac12 + it) as t runs up the line. This is a complex number for each t. A complex-valued function of a real variable traces a curve wandering around the plane; it can circle the origin, dodge it, brush past it — but "it changed sign, so it must have passed through zero" is an argument available only to a real function of a real variable. Complex numbers aren't ordered. There is no Intermediate Value Theorem to lean on.

Hardy's idea was to rotate away the phase. At each height t, \zeta(\tfrac12 + it) points in some direction in the complex plane; if we can compute that direction and multiply by a unit complex number that cancels it exactly, we are left with a real number of the same magnitude — which is zero iff \zeta(\tfrac12 + it) was zero. That rotation is the Riemann–Siegel theta function.

Hardy's Z-function

Define the Riemann–Siegel theta function

a smooth, real-valued function of t (its large-t shape is \theta(t) \approx \tfrac{t}{2}\ln\tfrac{t}{2\pi} - \tfrac{t}{2} - \tfrac{\pi}{8}). Now define Hardy's Z-function (also called Hardy's function):

Why is Z(t) real? The functional equation makes \zeta(\tfrac12+it) and its complex conjugate \zeta(\tfrac12-it) = \overline{\zeta(\tfrac12+it)} related by exactly the phase factor e^{2i\theta(t)}. The theta rotation is precisely tuned to undo half of that phase, landing the product on the real axis. The factor e^{i\theta(t)} is a pure "un-twist": it never vanishes and never changes the size of anything, it only spins \zeta(\tfrac12+it) back onto the real line.

The key equivalence: zeros become sign changes

Because e^{i\theta(t)} is never zero, the two functions vanish at exactly the same places:

This is the whole trick in one sentence: a sign change of the real function Z is a certificate that \zeta has a zero sitting precisely on \Re(s) = \tfrac12 — no rounding, no "close to \tfrac12", but exactly on it. Counting on-line zeros up to height T becomes counting sign changes of one real graph.

Seeing it: Z(t) crossing zero

Below is Hardy's Z(t) for real t, computed by the Riemann–Siegel formula (the same tool used in the record-breaking searches). It is a real, wiggling curve. Every time it crosses the horizontal axis, that crossing is a non-trivial zero of \zeta sitting exactly on the critical line.

Worked example — pinning the first zero. Evaluate the real number Z(t) at two heights straddling t = 14. A short computation gives Z(13) \approx -0.80 < 0 and Z(15) \approx +0.72 > 0. Since Z is continuous and has swapped from negative to positive, the Intermediate Value Theorem forces a t^* \in (13,15) with Z(t^*) = 0 — and therefore \zeta(\tfrac12 + it^*) = 0, a zero exactly on the line. Refining the bracket lands on t^* = 14.1347\ldots, the famous first non-trivial zero. Notice what we did not need: we never had to compute \zeta off the line, and we never had to trust that a number was "close enough" to \tfrac12. Two real evaluations and a sign flip did it.

Hardy's theorem, and the shape of the proof

The elegance is in how Hardy shows Z keeps changing sign without ever locating a single zero exactly. Suppose, for contradiction, that Z(t) were eventually one sign — say Z(t) \ge 0 for all large t. Then averages (integrals) of Z against smooth weights would settle into a predictable one-sided behaviour. Hardy studied exactly such moments — integrals of the form

for cleverly chosen weights \varphi. Using the functional equation he pinned down their size and — crucially — showed they take both positive and negative values for arbitrarily large T. A function that stayed one sign could never make its running averages swing both ways. The contradiction means Z must go positive and negative infinitely often, i.e. it has infinitely many sign changes — infinitely many on-line zeros. It is a pure "you can't have it both ways" argument: the existence of zeros is forced, even though none is ever named.

After Hardy: from "infinitely many" toward "all"

Hardy's theorem opened a century-long campaign to enlarge the fraction of zeros proven to be on the line. Write N(T) for the number of non-trivial zeros up to height T, and N_0(T) for how many of those are on the critical line. The milestones:

Notice the qualitative jump at Selberg. Hardy–Littlewood's N_0(T) \ge cT sounds strong, but since the total count N(T) grows like T\ln T, a bound of order T is still a vanishing fraction. Selberg was the first to prove a fixed percentage of zeros lie on the line, and Levinson and Conrey pushed that percentage up toward a half. Even so, Conrey's 2/5 is a long way from 100\% — the Riemann Hypothesis remains unproven, and the ceiling on the proportion has crept up only slowly since.