teal-sea / zeta-labstate of record · compiled 14 Aug 2026 · revision 9ebdea0 · source

Library · docs/09-new-ontologies.md

New Ontologies: What "RH Needs New Mathematics" Actually Means

4,138 words · 397 lines · source

The sequel to docs/08-why-it-is-hard.md. That document catalogued why the existing tools provably stall. This one describes the professional consensus about what a real attack would have to look like, and the one time in history the strategy worked.

The short version

The working belief of the field is not "we need a cleverer inequality" but "we need new objects." The evidence is a precedent: the Riemann Hypothesis over finite fields resisted direct attack for roughly twenty-five years after Weil conjectured it in 1949, and it fell in 1974 — not because someone found a smarter estimate, but because Grothendieck and his school spent the 1960s rebuilding the ontology of geometry itself (schemes, topoi, étale cohomology) until Frobenius became an honest linear operator on an honest cohomology group, at which point Deligne could finish. The proof is literally unstatable in the pre-1960 language. "RH needs new mathematics" means: do that again, for the integers. There are serious ongoing attempts — the field with one element, Connes' noncommutative trace-formula program, Deninger's dynamical-systems program — and every one of them is a PROGRAM, not a result. Below: the precedent in detail, the Selberg-class rulebook that any new ontology must respect, the round-two attempts, and four concrete falsification gates that let you evaluate a proposed ontology in an afternoon.

1. The one victory: the Weil conjectures

1.1 A Riemann Hypothesis you can check by counting

Take a polynomial equation and count its solutions over the finite field F_p. For the curve E : y² = x³ − x, write N_p for the number of solutions mod p, plus one point at infinity.

p = 1009
N = 1 + sum(1 for x in range(p) for y in range(p)
            if (y*y - (x**3 - x)) % p == 0)
# N = 1040;  a_p = p + 1 - N = -30;  2*sqrt(1009) = 63.53...

THEOREM (Hasse, 1930s, for elliptic curves; Weil, 1940s, for all curves).

    | N_p − (p + 1) |  ≤  2g · sqrt(p)          (g = genus; g = 1 here)

Run the snippet: a_p = −30, comfortably inside ±63.5. Why is this a Riemann Hypothesis? Package the counts over all extensions F_{p^n} into a zeta function; for a curve it is a rational function whose numerator has roots of absolute value exactly sqrt(p) — i.e. "all zeros on the critical line" for the local analogue. The sqrt in the point-count bound is the 1/2 in Re s = 1/2.

THEOREM (Weil, published 1948). RH holds for all curves over finite fields. Weil then conjectured (1949) the analogue for varieties of every dimension: rationality of the zeta function, a functional equation, and the RH statement that the zeros and poles have the exact absolute values q^{i/2}.

1.2 Twenty-five years, and what actually worked

Direct and partial attacks ran for a generation. Dwork proved the rationality part in 1960 by p-adic analysis — brilliant, but a technique, not a home; it did not touch the RH part. What worked was Grothendieck's rebuild of the foundations, roughly 1958–1970:

    N_{p^n}  =  Σ_i  (−1)^i  Tr( Frob^n | H^i_et )

Rationality and the functional equation fall out of this formula almost immediately. The RH part says: the eigenvalues of Frobenius on H^i have absolute value exactly p^{i/2}. That is a positivity-flavoured statement about an operator — precisely the shape the Hilbert–Pólya dream (docs/06-hilbert-polya-and-gue.md) wants for zeta.

1.3 The positivity input

The eigenvalue bound does not come for free from the formalism. In Weil's own proof for curves, the input is a genuine positivity theorem: the Castelnuovo–Severi inequality (a Hodge-index-type statement) on the surface C × C, applied to the graph of Frobenius. Note the move: to prove RH for the curve C, you work on the square C × C. Grothendieck hoped to finish the general case the same way, via his "standard conjectures" — positivity statements for algebraic cycles that remain CONJECTURES to this day. Deligne's actual 1974 proof (La conjecture de Weil I) took a different, more analytic route — a tensor-power amplification argument with monodromy, borrowing a squaring trick from the classical theory of modular forms — so the general case was won without the standard conjectures. But in every version, symmetry (the functional equation, supplied by duality) plus an input of positivity is what pins eigenvalues to the critical absolute value. Symmetry alone never suffices; docs/08-why-it-is-hard.md makes the same point on the analytic side.

The professional meaning of "RH needs new mathematics" is exactly this template. Find the category in which Spec Z is a curve over something, in which "zeta of Spec Z" is the Riemann zeta function, in which some Frobenius-like flow acts on a cohomology that nobody has constructed, and in which a positivity theorem is available. Then Deligne's blueprint applies. Nobody has found the category.

2. Category theory: the working ontology, in one paragraph

The language all of this is written in deserves one honest paragraph. In category theory an object has no insides: you are forbidden from asking what it is made of, and permitted only to ask how it maps to and from other objects, and how those maps compose. The (informal) content of the Yoneda lemma is that this costs nothing: an object is completely determined by the totality of maps into it. This sounds like philosophy but is an engineering decision, and it is the one that made Section 1 possible — "a space with no classical points but real cohomology" is not a paradox once space means "a category of sheaves obeying axioms." Every round-two program below is an attempt to write down a category in which the primes are forced to be the points of something.

3. The Selberg class: an actual rulebook

If new objects are coming, we should at least axiomatise the old ones. Selberg (1989) proposed a definition of "everything that deserves to be called an L-function": the Selberg class S. A Dirichlet series F(s) = Σ a_n n^{−s} belongs to S if it satisfies:

    (1) Convergence:   the series converges absolutely for Re s > 1  (a_1 = 1).
    (2) Continuation:  (s−1)^m F(s) extends to an entire function of finite order,
                       for some integer m ≥ 0.
    (3) Symmetry:      a functional equation  Φ(s) = ω · conj(Φ(1 − conj(s))), |ω| = 1,
                       where Φ(s) = Q^s · Π_j Γ(λ_j s + μ_j) · F(s),  λ_j > 0, Re μ_j ≥ 0.
    (4) Euler product: log F(s) = Σ b_n n^{−s} with b_n = 0 unless n is a prime power,
                       and b_n ≪ n^θ for some θ < 1/2.
    (5) Ramanujan:     a_n ≪ n^ε for every ε > 0.

CONJECTURE (Grand Riemann Hypothesis). Every F in S has all its non-trivial zeros on Re s = 1/2. Note the design: axioms (3) and (4) are both required, which is exactly the lesson of the counterexamples in docs/08-why-it-is-hard.md — drop either one and the conjecture is false.

The class carries an invariant, the degree d = 2 Σ_j λ_j (zeta has d = 1; a modular-form L-function has d = 2). The rulebook is slowly being proven to contain exactly the expected inhabitants. THEOREM (Conrey–Ghosh, commonly cited from their 1993 paper): the only element of degree 0 is the constant function 1, and there are no elements with 0 < d < 1. THEOREM (Kaczorowski–Perelli, in their "structure of the Selberg class" series — degree 1 around 1999, and the non-existence of degrees strictly between 1 and 2 in a later installment, commonly cited as 2011): the degree-1 elements are precisely ζ(s) and shifted Dirichlet L-functions of primitive characters. (I am confident in the statements; verify the exact dates and paper numbers before citing them formally.) Degree 2 — where modular forms live — is not yet classified. It is CONJECTURED that S coincides with the class of automorphic L-functions; that identification is the Langlands program viewed from the analytic side.

4. Round two: attempts at the missing geometry of Spec Z

All three items below are labelled PROGRAM. None contains a proof of RH, and none is close by its own authors' account. They are serious because each one produces genuine theorems inside its own formalism and each one targets the specific gap Section 1 identified.

4.1 The field with one element, F1 — PROGRAM

The oldest hint predates schemes: Tits observed in the 1950s that formulas for algebraic groups over F_q degenerate, at q = 1, into true statements about Weyl groups — as if a "field with one element" existed. The number-theoretic dream is precise in shape: if Spec Z were a curve over F1, then Spec Z ×_{F1} Spec Z would be a surface, and Weil's positivity proof for curves (Section 1.3, which runs on C × C) would have a home. The obstruction in ordinary algebraic geometry is a theorem-level fact: Z has no coefficient field, and the product collapses (Z ⊗_Z Z = Z). So F1-geometry tries to rebuild the rulebook beneath rings: Soulé gave a candidate definition of varieties over F1 (2004); Connes–Consani have developed several successive versions; Borger's approach identifies "descent to F1" with a Λ-ring structure — roughly, a ring equipped with commuting lifts of all Frobenius maps, making "Frobenius over the integers" the definition rather than the mystery. These formalisms genuinely differ and none has yet produced the surface with a workable intersection theory. That is the state of play.

4.2 Connes' noncommutative trace formula — PROGRAM

THEOREM (Weil's positivity criterion, 1952). RH is equivalent to the positivity of a certain quadratic form built from the explicit formula of docs/04-explicit-formula.md: RH holds if and only if the explicit-formula functional is ≥ 0 on all test functions of the form g * g̃. The equivalence is a theorem; the positivity itself is the open problem, restated.

Connes (1999) constructed a noncommutative space — the adele class space, a quotient so badly non-Hausdorff that only the topos/operator-algebra rulebook can see it — on which a natural scaling flow acts, and proved a trace formula for it in which the explicit formula's shape appears with the primes on the geometric side. In this frame RH becomes exactly a Weil-positivity statement about the flow. Connes–Consani have continued the program (the "scaling site," and papers around 2020–2021 proving positivity in restricted settings); I have not verified the precise current scope of those results, and neither author claims RH is proven. The honest summary: the program has succeeded in restating RH as positivity on a genuinely existing (noncommutative) space, and the positivity remains open.

4.3 Deninger's dynamical program — PROGRAM

Deninger (from the early 1990s; see his ICM address, commonly cited as 1998) argues from the shape of known formulas that there should exist an infinite-dimensional cohomology theory and a flow (a one-parameter dynamical system on some foliated space) such that: zeta is a regularized determinant det(s − Θ) over the conjectural H^1; the closed orbits of the flow are the primes, with period log p; and the explicit formula becomes a Lefschetz-type trace formula for the flow — compare the Guinand–Weil form in docs/04-explicit-formula.md, where each prime p really does enter through the sequence log p, 2 log p, 3 log p, ..., exactly like a closed orbit and its iterates. Deninger has built fragments (e.g. determinant formulas that reproduce the Gamma factor), but the space carrying the flow has not been found. Note how precisely this program and 4.1–4.2 triangulate the same missing object from different sides: primes as closed orbits (Deninger), primes as points of a curve (F1), primes on the geometric side of a trace on a noncommutative space (Connes).

5. Four design gates (how to falsify a proposed ontology quickly)

Any proposed new ontology for RH can be stress-tested against four gates. The packaging into "four gates" is this document's own synthesis, not a standard framework you will find named in the literature — but each individual requirement below is standard, and the sources for each are given with it. These are not aesthetics; each is a concrete test with known failure modes.

Gate 1 — Recover the classics. Inside the new language, re-derive the functional equation and the Prime Number Theorem. If the formalism cannot even see docs/03-functional-equation.md and the zero-free region story of docs/08, it is not about zeta. (This is the gate most F1 formalisms are still working toward: producing any classical analytic statement as output.)

Gate 2 — A home for positivity. The ontology must contain a slot where a positivity input (Hodge-index-like, Weil-criterion-like) can live. Symmetry alone is provably insufficient — the functional equation is satisfied by functions with off-line zeros (Gate 3), so a framework whose only mechanism is symmetry cannot possibly output RH.

Gate 3 — The counterexample gate (the sharpest one). Whatever structure the ontology grants to ζ must be ungrantable to the Davenport–Heilbronn function and to generic Epstein zeta functions (THEOREM, 1936 and onward: these satisfy functional equations, have no Euler product, and have zeros off the critical line — see docs/08-why-it-is-hard.md). An ontology roomy enough to accommodate them "proves too much" and is dead on arrival. Concretely: when someone presents a framework, ask "where exactly does Davenport–Heilbronn fail to embed?" If there is no crisp answer, stop reading.

Gate 4 — The Euler product must be structural. In the new ontology the primes must be the points (or orbits, or spectrum) of the hidden object, so that the Euler product is the statement "the zeta function is a product over points" — a tautology of the geometry, as it is over finite fields — rather than an identity bolted on afterward. All three programs of Section 4 pass this gate by design; it is Gates 1 and 2 where they are stuck.

5.1 The strengthened gates: factorization, not positivity

The four gates above are eliminative: they kill ontologies that cannot possibly work. This subsection states the positive target at the same resolution — what a candidate that survived all four would still have to deliver. The organizing observation: since Weil positivity over the full admissible test class is already equivalent to RH (Section 4.2), "prove the quadratic form is positive" is not a strategy — it is RH restated. Trying to estimate the prime-power sum directly is merely another formulation of the problem. The demand must instead be factorization:

Construct, functorially from the prime-power and archimedean data, an arithmetic cohomology or spectral representation in which the Weil quadratic form is the norm square of a naturally defined operator or vector — schematically −W(f ∗ f̃) = ‖Φ(f)‖² in a genuine Hilbert, Hodge, intersection or C*-module structure — so that its sign becomes formal.

Arithmetic supplies the quadratic form; geometry must explain its sign. Once the norm identity holds inside a structure that is positive for structural reasons, ⟨v, v⟩ ≥ 0 is linear algebra, and RH follows by the already-known equivalence. Three requirements, in causal order:

Requirement A — arithmetic provenance. Every ingredient of the construction is generated from the prime-power data {(p, m, log p, p^{−m/2})} together with the archimedean local factor. No zeros, no ξ-phases, no zero-counting functions may enter the definition. (Given any real sequence one can manufacture a self-adjoint diagonal operator with that spectrum; a construction that imports the zeros explains nothing.)

Requirement B — exact trace realization. An identity Str π(f) = W(f) on the entire admissible test algebra, with the prime powers arising as primitive orbit repetitions, local fixed-point terms, or an equivalent intrinsic mechanism — not inserted afterward as coefficients. This is where the analytic difficulty relocates, not where it disappears: in the function-field case, Castelnuovo's inequality is formal given the surface, but building the surface and proving the graph–diagonal intersection computes the right thing was the hard part. Expect a real attempt to bleed here, on the full-class quantifier.

Requirement C — structural positivity. A canonical positive pairing such that −W(f ∗ f̃) = ‖Φ(f)‖² for test functions satisfying the pole-removal conditions, with the signature of the pairing forced independently of the zeros. The conceptual flow is then

    prime arithmetic ⟶ object ⟶ pairing ⟶ norm identity ⟶ Weil positivity ⟶ RH

and the load-bearing arrow is not the last one (that arrow is Weil's theorem); it is prime arithmetic ⟶ positive pairing.

The pseudo-solution taxonomy. Each of the following evasions has absorbed a real research program, which is why they deserve names:

  1. Direct estimation — proves positivity only for restricted support, special test functions, or finite numerical ranges. (The fate of the truncated-Weil numerical literature: positive-definite matrices Q_N for every computed N, worth nothing by Littlewood's rule, docs/08.)
  2. Tautological completion — defines a Hilbert space by completing test functions under the Weil form itself, assuming the very positivity at issue. (The standing critique of the de Branges route: Hilbert spaces of entire functions are a genuine positive category, but the required positivity conditions on the structure function are not prime-derived facts.)
  3. Zero-importing — defines the pairing spectrally via Σ_ρ |ĥ(ρ)|². (Every Hilbert–Pólya toy operator since 1999; fails Requirement A.)
  4. Formal C\-positivity* — observes that a*a ≥ 0 somewhere, but never proves that the arithmetic Weil distribution is the resulting positive functional. (The gap between "a trace formula exists on the adele class space" and "RH", Section 4.2.)
  5. Finite approximants — obtains positive matrices Q_N without proving the limit is the full Weil form on the whole admissible domain. (Variant of 1, with the gap hidden in the limit.)

Sharpening Gate 3/4: linear combination is the destroyer. The precise statement of what the counterexamples prove: functional equation, gamma factors, Dirichlet-series structure, and even being assembled from legitimate Euler products do not suffice; the primitive multiplicative structure must survive in the global object, and linear combination destroys it. The Davenport–Heilbronn function is a self-dual linear combination of the two Dirichlet L-functions of the quartic characters mod 5; Epstein zeta functions of class number greater than one are linear combinations of the Hecke L-functions of the class group — and both have zeros off the critical line. So the operational form of the gate is:

The construction must detect primitive multiplicative local data, and must not apply unchanged to linear combinations of completed L-functions.

zeta.epstein.battery runs exactly this: its default rivals are Davenport–Heilbronn plus both discriminant −23 forms (class number 3), i.e. two independent linear combinations of genuine Euler products. One caution keeps the gate honest in the other direction: no known example satisfies the full Selberg-class package (Section 3, axioms 1–5) and violates RH — the counterexamples conclusively reject weaker packages only. Believing the full package suffices is essentially believing GRH; the gate is eliminative, never probative (prime-blind ⟹ not an explanation, but not prime-sensitive ⟹ proof).

What this laboratory can and cannot test. The three requirements split cleanly. Requirement A is mechanically checkable (provenance scans of a construction's definition — the same technology as the discovery/ seam tests). Requirement B is exactly what this repo is built to test: an identity on the test algebra is a measured defect function in the house style — compute both sides independently over test functions from unrelated families, the protocol that validated the zeta/weil.py convention; a candidate passing at 1e-30 across families is not proof, but a candidate failing is dead. Requirement C's norm identity is numerically checkable as an identity; its naturality — "canonical", "signature forced independently of the zeros" — is a judgment no computation can render, and marks the exact boundary where the laboratory's writ ends and human mathematics begins. (Convention landmine for any implementation: the sign of W and the direction of the inequality vary across Weil/Bombieri/Connes normalizations — calibrate numerically against a test function of known sign, per the house rule; never trust a source's sign.)

6. Cost honesty, and the Langlands ledger

The one time this strategy worked, the bill was: roughly twenty years (late 1950s to 1974), a generation of exceptionally strong mathematicians working in concert, thousands of pages of foundations (EGA, SGA) — and, crucially, a working blueprint the whole time: Weil had already proven the curves case, so everyone knew the target theorems and could test the machinery against them. Round two has the same blueprint (the finite-field case is now the known model) and, counting from Tits' remark, more than fifty years of searching without finding the objects. That is not an argument it will fail; it is an argument about the scale of the ask.

One more piece of honesty about the Langlands program, the master rulebook conjecture: it predicts that the inhabitants of the Selberg class are exactly the automorphic L-functions, all tied together by functoriality. If the full structure existed, RH would be one line-item in the ledger — the natural positivity statement about the whole family at once. But state it carefully: no known formulation of Langlands functoriality is known to imply RH. Langlands organises the objects; it does not, as currently formulated, supply the positivity of Gate 2. The two quests are complementary, not identical.

7. Back to this repo: shadows of the missing ontology

Two things you can compute in this laboratory are, on this view, shadows cast by the object nobody has found.

The computations are real; the interpretation — that they are low-dimensional projections of one missing geometry — is the field's best-supported HEURISTIC.

Where to go next