André Weil

Zeta over a finite field

Take a smooth projective curve $C$ over the finite field $\mathbb{F}_q$ and count its points over every extension $\mathbb{F}_{q^{n}}$. Package the counts into a generating function:

$Z_C(T) \;=\; \exp\!\left(\sum_{n \ge 1} \frac{\#C(\mathbb{F}_{q^{n}})}{n}\, T^{n}\right)$

Three things are then true, and all three are theorems rather than hopes. $Z_C(T)$ is a rational function, $P(T)/\big((1-T)(1-qT)\big)$, with $P$ of degree $2g$ and integer coefficients. It satisfies a functional equation relating $T$ to $1/(qT)$ — the exact analogue of $s \mapsto 1-s$. And every reciprocal root of $P$ has absolute value $\sqrt{q}$, which is the Riemann hypothesis for $C$: written in the variable $T = q^{-s}$, it says every zero sits on $\operatorname{Re}(s) = \tfrac12$.

book7/ch-weil-verify.py takes the genus-1 case down to arithmetic that leaves nothing to trust. Five curves over fourteen primes, every one of the $p^{2}$ pairs $(x,y)$ tested — no sampling — giving 66 exhaustive point counts, with four singular triples excluded rather than counted. Every $|a_p| \le 2\sqrt{p}$, the tightest case reaching $0.9615$ of the bound. All 132 reciprocal roots come out at $|\alpha| = \sqrt{p}$ to a maximum error of $8.88 \times 10^{-16}$ — and, more to the point, the algebra behind that is exhibited rather than measured: $\alpha\beta = p$ and $a_p^{2} < 4p$ force the pair to be complex conjugate, so $|\alpha|^{2} = \alpha\bar{\alpha} = \alpha\beta = p$ exactly, and the floating-point check only confirms that no arithmetic slipped. The functional equation holds to $2.66 \times 10^{-15}$, and in twelve cases the count over $\mathbb{F}_{p^{2}}$ is predicted from the roots and then confirmed by direct exhaustion over all $p^{4}$ pairs — the moment at which the zeta function stops being bookkeeping and starts making a claim about a field it was not built from.

1949: a conjecture about the shape of an absent theory

In the 1949 Bulletin paper Weil stated for varieties of any dimension what he had proved for curves, and added the observation that made it famous. The statements have the shape they would have if there were a cohomology theory for varieties over finite fields with a Lefschetz fixed-point formula: the points of $X(\mathbb{F}_{q^{n}})$ are exactly the fixed points of the $n$-th power of Frobenius, the zeta function is then an alternating product of characteristic polynomials, rationality is automatic, the functional equation is Poincaré duality, the degrees of the factors are Betti numbers, and the Riemann hypothesis is a statement about the eigenvalues of one operator. The conjecture was, in effect, a specification for a machine nobody had built.

Three columns, written from a prison cell

In March 1940, in a military prison in Rouen, Weil wrote to his sister Simone the letter that is now the standard statement of the analogy: three parallel columns — number fields, function fields over a finite field, and Riemann surfaces — with the middle column as the bridge, because it is algebraic enough to look like the first and geometric enough to look like the third. Translating a statement across is how you find out what it means. That letter is the reason anyone believes the function-field case is evidence for the number-field one.

It is also where the honest reading begins. Why does the middle column close? Because over $\mathbb{F}_q$ the Frobenius map is an actual endomorphism of an actual variety, and the curve $C$ gives you a surface $C \times C$ on which correspondences can be intersected. Weil’s proof runs on the Castelnuovo–Severi inequality — an intersection-theoretic positivity on that surface — and Deligne’s runs on a different positivity in $\ell$-adic cohomology. In both cases a geometric object supplies an inequality, and the inequality is the proof.

Over $\mathbb{Q}$ there is no such surface. There is no Frobenius, no ambient variety to intersect in, and consequently no source for the inequality. Weil himself supplied the sharpest form of the gap: his 1952 explicit formula turns the Riemann hypothesis into the positivity of a distribution. That is an exact equivalence and it is not a proof; it relocates the difficulty into a single sentence and leaves it there.

Place in the Series

The register for a Weil citation in this corpus is architectural, never evidential. The function-field case tells you what shape a statement about zeros ought to have — a symmetry with a fixed locus, an operator whose spectrum is the zero set, a positivity that closes the argument. It does not tell you that any of those exist over $\mathbb{Q}$, and citing it as though it did is the single most common way an RH-adjacent manuscript loses its reader in the first page.

The mathematics lives in Book 4, not here. Ch 11 · The Arithmetic Seed builds the contact form $\alpha_{\mathrm{arith}}$ and states plainly, in its Honest Inventory, that the three-dimensional lift is exposition rather than contribution — the argument principle made visible — with Suzuki’s screw line cited as the nearest named object in the literature. Ch 12 · The Critical Contact reads the functional equation as a contactomorphism whose fixed locus is the critical line, and carries the supersession record for the Mathlib merge of September 2026 together with the one result that survives it: a domain widened at the negative odd and positive even integers, by routing through the archimedean factor rather than through $\zeta$ directly. § 12.8 is where the function-field comparison is written out as mathematics, with the two positivity statements set side by side.

Alain Connes belongs beside this chapter for the obvious reason: the spectral-triple programme was aimed at exactly this target, is the most developed attempt anyone has made, and arrives at the same positivity. Gilbert Strang belongs beside it for a less obvious one — every one of these programmes ends by asking whether some operator is positive, and that is a sentence about a spectrum, which is the one piece of machinery this corpus can actually put under the kernel today.

Outside the mathematics, this history is also a convergence record, and the series reads it that way in Omega Point · In the Air: Dwork reaching rationality in 1960 with no cohomology while Grothendieck was building the cohomology that would reach it again; Deligne closing the hardest statement in 1974 by a route its architect did not intend; and the intended route still unwalked fifty-seven years after it was named. A destination arrived at twice, with a third road visible, mapped, and untravelled — which is the cleanest example in mathematics of the distinction that chapter is built on, between the route being contingent and the destination being constrained.

References