# Think hard: which higher-dimensional facts really reach ordinary zeta?

Research note, 10 November 2026, PST. No deployment. No claim of RH progress.

**Labels.** [known] means an established result, with primary references below;
[project] means a deduction/check organized here, not a new theorem of number
theory; [speculative] means a missing theorem or research proposal. A conditional
implication can itself be [known]: this does not mean its premise is proved.

## Revision notice and changelog

**[project; 10 November 2026, PST, “continue to think hard” rigor pass]** See
[`think-hard-heilbronn.md`](think-hard-heilbronn.md) for the full derivation,
source audit, and fresh interval certificates. This revision:

* distinguishes a valid multiplicity-extraction mechanism from an RH strategy:
  **the norm inequality is redundant in our already-holomorphic Q8 example**;
* proves that the compatible assignment `Theta_G=m*1_G` survives all the bare
  bookkeeping constraints, across arbitrarily many fields, even for a hypothetical
  multiple or off-line ordinary zero;
* adds Murty–Raghuram's **solvable-group** refinements with RHS `(M-m)^2`, not
  `M^2-m^2`, and the all-abelian truncation;
* supplies fresh 320-bit interval local field/subfield counts at the first three
  actual ordinary zeros and at 1/2, plus an Euler/theta certificate of root sign
  −1. **Field evaluations are factored, not independent downstairs evidence**;
* clarifies fixed-region family averages versus conductor-rescaled one-level
  density, and confirms the all-index Dedekind Li criterion from the original
  Bombieri–Lagarias text;
* marks Heilbronn's unretrieved 1972 contribution separately from the retrieved
  1973 journal paper, and distinguishes zero-multiset differences from pairwise
  noncoincidence. Wrong-title source downloads were quarantined, not cited.

The earlier Aramata–Brauer direction and all-index Li implication were correct.
The chief correction is the earlier impression of extra Q8/RH leverage, not a
new zero-location theorem. No live page or deployment was changed.

## 0. The answer in ordinary language

**[project; synthesis of the known results below]** The best bridge is not
“there are more directions.” It is **“the larger function contains ordinary zeta
as a factor, and nothing can cancel that factor's zeros.”** A theorem that the
larger function has no zeros in a region then rules out ordinary-zeta zeros there.
A bound on the larger function's number of zeros also bounds the smaller one's.
But a zero *found* upstairs need not be downstairs: it may belong to another factor.

**[known; H1–H5]** There is a precise multiplicity bridge beyond the first ledger
summary: **Heilbronn characters**, an exact bookkeeping system for zero
multiplicities across a Galois extension and its subfields. They constrain all the
Artin factors even without the general Artin holomorphy conjecture. In particular,
a simple field-zeta zero can only belong to a one-dimensional Artin factor. If it
also belongs to ordinary zeta, it is a simple ordinary-zeta zero and all the other
irreducible factors are nonzero there. This is real higher-to-lower information,
not a visual analogy. It still needs a theorem or certificate *about the larger
function* as input. In our monomial Q8 example, known factor holomorphy already
makes the norm bound automatic; it is not additional analytic leverage on RH.

**[project; interpretation]** Quaternionic and octonionic zero spheres are valuable
ways to see ordinary zeros, but for the intrinsic slice extension they are a
lossless re-expression of the complex picture. Their roundness is not an
additional equation that fixes the real coordinate. A genuinely new geometric
inequality might transfer through that re-expression; roundness by itself does not.

**[known; P1–P5]** The strongest other bridges are exact positivity/approximation
criteria. They change the language of RH to positivity, Hilbert-space approximation,
or a real-zero flow. They do not make the universal positivity/limit assertion
already true. The finite-field story shows that moving to a product surface really
*can* prove a curve's RH; it does not yet supply the corresponding surface,
operator, trace identity and positivity for ordinary zeta over the integers.

## 1. First separate six meanings of “higher-dimensional”

**[project; terminology]** We have used (1) H/O-valued inputs, (2) a field K/Q of
higher degree, (3) a representation of larger matrix dimension, (4) a variety of
larger geometric dimension, (5) an infinite-dimensional Hilbert space and (6) a
family with extra parameters. These are not one ladder. The arrows below name the
particular map; “up/down” only refers to that map.

**[known; elementary order arithmetic]** Write ord_a F for the signed order at a:
positive for a zero, negative for a pole, zero for a finite nonzero value. If

    F(s) = zeta(a*s+b)^r G(s),   a != 0, r >= 1,

and G is holomorphic near s0, then

    ord_s0 F = r ord_(a*s0+b) zeta + ord_s0 G.

In a nontrivial-zero strip, the last order is nonnegative. This simple equation is
the engine. An Euler-product identity only in Re(s)>1 must first be continued;
a meromorphic quotient with unexcluded poles is not enough.

### Ranked verdict / direction matrix

| Rank | Candidate and exact bridge | Direction of information | Status and missing input |
|---|---|---|---|
| **1** | Galois Dedekind factorization + Aramata–Brauer; Heilbronn orders and subfield projections | Zeros go **up**; zero exclusion/count upper bounds/nonvanishing and constrained multiplicities go **down**. Full factor data reconstructs both ways. | **[known; A1,H1–H5]** Fully load-bearing. No improved global upstairs zero-location theorem has been obtained here. |
| **2** | Li/Weil positivity; Nyman–Beurling/Báez–Duarte; de Bruijn–Newman | **Both**, as exact equivalences in their stated domains/normalizations | **[known; P1–P5]** Fully load-bearing logical reformulations. The missing universal assertion is RH-sized, not a completed experiment. |
| **3** | Quaternion-order/E8 shifted-zeta products; modular-surface scattering poles (§7.5) | Zeros up, exclusions down; the specified full zero-location assertions are **equivalent** to RH | **[known; Q1,Q2,R5]** Genuine but currently a repackaging, not an easier proved upper theorem. |
| **4** | Tensor/Rankin–Selberg factor containing 1, with a holomorphic adjoint quotient | Same one-way exclusion/count bridge as rank 1; **not** from zeros of the original standard L-function to tensor zeros | **[known in stated cases; R1–R3]** Q8 is unconditional. General-rank adjoint holomorphy is a gate; known small-rank cases are specified below. |
| **5** | Hurwitz: zero-free holomorphic approximants, locally uniform nonzero limit | **Down to the limit**, not a reverse assertion for arbitrary approximants | **[known]** Exact analytic lever. An appropriate all-stage sequence/convergence theorem is missing. |
| **6** | Weil/Deligne geometric positivity and Frobenius | Genuine within finite-field geometry; **no established map to ordinary zeta** | **[known finite-field theorem; speculative arithmetic bridge; G1–G3]** Missing global spectral/trace/positivity package, not merely extra coordinates. |
| **7** | Jacquet–Langlands, Selberg degree/orthogonality, family symmetry | JL preserves a specified GL2 L-function both ways on its image; other notions give no automatic zero-by-zero descent | **[known/conditional as below; R4,S1,S2,F1]** Useful structure, not ordinary RH by themselves. |
| **8** | Intrinsic slice zero spheres, tube degrees, holes at infinity; random-matrix resemblance | Slice extension/restriction is **both ways but information-equivalent**; listed coarse invariants do not retain the needed coordinate | **[known/project; X1,X2]** No additional ordinary-zeta constraint from these facts alone. This is not a theorem that all future higher-dimensional methods fail. |

## 2. Field zetas: zero exclusion and density really do descend

### 2.1 The theorem, including the no-cancellation hypothesis

**[known; A1,H1]** For finite Galois K/Q, the quotient

    Q_K(s) = zeta_K(s)/zeta(s)

is entire (Aramata–Brauer). This does **not** require holomorphy of every
irreducible Artin L-function. Hence, for a nontrivial ordinary zero rho,

    m_K(rho) >= m_zeta(rho).

A theorem proving GRH for this one field zeta would prove ordinary RH. Ordinary
RH alone does not prove GRH for the quotient. For arbitrary non-Galois extensions,
write down the holomorphy theorem actually available: the general Dedekind
quotient conjecture is not a corollary of merely seeing a trivial constituent in
a permutation representation. Solvable-normal-closure cases are known beyond the
Galois case; they do not license an all-fields statement. [H4]

**[known; consequence]** For any region D in the nontrivial strip:

    zeta_K zero-free on D  ==>  zeta zero-free on D.

For counts WITH multiplicity, with the same height and boundary conventions,

    N_zeta(sigma,T) <= N_K(sigma,T).

This includes off-line counts. No density estimate is improved automatically by
dividing the field count by [K:Q]: every inherited ordinary zero need occur only
once. A family average of N_K over extensions also bounds the same N_zeta, since
that fixed contribution occurs in every extension; an average tending to zero
for a fixed forbidden region would suffice. Establishing it without assuming the
answer is the missing part. Almost-all-field GRH, if genuinely proved for a
nonempty Galois family, would likewise imply ordinary RH, not merely a statistical
analogy.

**[known; logical qualification]** “Not a converse” means ordinary zero-freeness
places no factorization-based exclusion on zeros of Q_K. It is not an assertion
that the conjectural GRH of a particular field has been disproved.

**[known; important scaling distinction]** This fixed-region average is not a
standard low-lying one-level density assertion. A fixed ordinary ordinate gamma
has rescaled location gamma*log(C)/(2*pi), tending to infinity with conductor C.
It leaves any fixed low-lying window. No descent from a rescaled family statistic
to a fixed ordinary zero is justified without a separate argument.

### 2.2 Is this quantitatively better than the direct zeta theory?

**[project; assessment, not a new optimality theorem]** Usually no. A typical
Dedekind region is written with log D_K and degree-dependent height factors,
schematically

    Re(s) >= 1 - c/[log D_K + [K:Q] log(|t|+3)],

with a possible real exceptional zero and exact constants depending on the
particular theorem. Pulling it down is valid, but increases the complexity of the
conductor rather than removing it. Existing direct ordinary-zeta zero-free and
zero-density results are not improved by our finite Q8 data or by this generic
inequality. We have **not** made a comprehensive best-constant comparison across
all Dedekind results, and do not claim none could ever help. Any proposed gain
must state the identical target region, constants, height range, discriminant
cost, exceptional-zero clause and multiplicity convention, then beat a direct
ordinary-zeta theorem on those terms.

**[known; H2,H3]** Exceptional *real* zeros have special descent theorems (below).
For ordinary zeta there is already no real zero in (0,1): its alternating-series
representation has positive numerator and negative denominator there. Such
exceptional-zero localization is important for quadratic L-functions and
Chebotarev, but is not a new exclusion of ordinary nonreal zeros near 1/2.

## 3. Heilbronn characters: the overlooked rigorous bridge

### 3.1 The exact local object

**[known; H1–H4]** Fix a finite Galois K/k, group G, and s0 != 1. Brauer gives
meromorphic Artin L-functions. Put

    n_chi = ord_s0 L(s,chi),
    Theta = sum_(chi irreducible) n_chi chi,
    M = ord_s0 zeta_K = Theta(1) = sum_chi chi(1) n_chi.

Theta is a *virtual* character: n_chi may a priori be negative. Its trivial
coefficient is exactly ord_s0 zeta_k. This is not a new L-function and its
“dimension” is a zero order, not the number of spatial directions.

For H <= G and E=K^H, Artin induction and Frobenius reciprocity give

    ord_s0 zeta_E = <Theta|_H,1_H>
                  = sum_chi n_chi dim(V_chi^H).

More generally Theta|_H is the Heilbronn character for K/E. These equalities turn
orders in intermediate fields into linear measurements of the same order vector.
Their compatibility is a real extra constraint.

### 3.2 An elementary proof of the norm bound and Stark's lemma

**[known; norm bound reported in H4 Prop.3.2, H3; proof reconstructed here]** For
each cyclic subgroup C of G, all its Artin characters are one-dimensional Hecke
characters. At s0 != 1 their L-functions are holomorphic, including the trivial
character's field zeta. Thus Theta|_C is a genuine character, so

    |Theta(g)| <= Theta(1) = M   for every g in G.

Character orthogonality now gives the unconditional bound

    sum_chi n_chi^2 = (1/|G|) sum_g |Theta(g)|^2 <= M^2.             (H)

This proof also shows M>=0. It neither assumes GRH nor silently replaces a
virtual character by a genuine one on all of G.

**[known; Stark, H2 Theorem 3; H4 Prop.3.3]** If M=0, (H) forces every n_chi=0.
If M=1, the integer vector has precisely one nonzero coefficient, equal to +1,
on a degree-one character: sum chi(1)n_chi=1 excludes the minus sign and larger
degree. Consequently:

* A simple field-zeta zero makes **all** Artin L-functions holomorphic there.
* It is a simple zero of exactly one **one-dimensional** irreducible Artin factor;
  all other irreducible factors are nonzero there.
* If it is also an ordinary-zeta zero, the responsible character is trivial:
  ordinary zeta is simple there and every nontrivial factor is nonzero.
* A pole of an irreducible Artin factor, if one exists away from 1, can occur only
  at a field-zeta zero with M>=2.

The second bullet follows from the norm bound as well as holomorphy; do not read
“one-dimensional” as necessarily the trivial representation. Nonreal zeros of a
Dirichlet factor are legitimate simple upstairs zeros that are NOT zeta zeros.

### 3.3 Stronger order-budget statements and precise direction

**[known consequence of H]** Let k=Q and m=ord_s0 zeta>=0. Then

    sum_(chi != 1) n_chi^2 <= M^2-m^2.

If independent information says M=m, all other n_chi vanish, even without a
global Artin-holomorphy theorem. This equality condition is stronger than the
bare subset-of-zeros observation. It does not tell us m until we have some input
that identifies it; an integer optimization can bound rather than determine m.

**[known; H3,H4]** Foote–Murty developed this character framework and higher-order
local holomorphy statements. For solvable G whose order has at least two distinct
prime divisors p1<p2<..., ord zeta_K <= p2-2 is sufficient for local Artin
holomorphy. Foote–Wales prove the order-at-most-two result for solvable G.
Wong's 2019 Theorem 1.4 strengthens the solvable case to

    ord_s0 (zeta_K/zeta_k) <= 2  ==> every Artin factor holomorphic at s0.

His Theorem 1.3 gives the corresponding p2-2 quotient version. Do not apply a
second-smallest-prime notation to Q8, a prime-power group, without treating that
case; our Q8 factors are already monomial, so this holomorphy is known directly.
Also, Wong's abstract omits a solvability qualifier found in the actual theorem
statements. This note follows the theorem statements, not the abstract's broader
wording.

**[known; H1 Theorem 1 and H2]** For a **real simple** field-zeta zero, complex
conjugation pairs non-real-valued degree-one characters, so the unique responsible
character must be real, hence trivial or quadratic. Heilbronn's original theorem
puts that zero in the compositum of quadratic subextensions. Stark's exceptional
zero analysis makes this particularly effective close to 1. Over Q, ordinary
zeta is nonzero on (0,1), so a real simple zero there must belong to a nontrivial
quadratic factor. This is an excellent example of descent that actually narrows
the source of a zero. It does not address ordinary nonreal RH zeros directly.

**[project; revised verdict]** This is our most auditable *multiplicity-extraction
mechanism*, not presently our most promising RH attack. General Heilbronn/Stark
arguments can rule out unknown Artin poles from an independently small field
order. But for Q8 all factors are already holomorphic: `sum n_chi^2 <= M^2` is
then elementary from nonnegative orders and adds no extra analytic constraint.
Its practical promise here is certified separation/extraction, not new evidence
that an ordinary off-line zero or multiple zero is impossible.

### 3.4 Stronger solvable budgets and the family obstruction

**[known; H6 Theorems 4.1, 4.6, 5.4]** For solvable G, any degree-one chi0 satisfies

    sum_(chi != chi0) n_chi^2 <= (M-n_chi0)^2.

Deleting all degree-one characters S gives

    sum_(chi not in S) n_chi^2 <= (M-ord zeta_(K^ab))^2.

These use quotient-holomorphy theorems to preserve genuine cyclic restrictions;
they are not obtained by merely subtracting a summand in (H). In Q8 they too
are automatic from known holomorphy. The nonabelian quotient has no simple
zeros/poles (H6 Corollary 5.6); this is not a simplicity theorem for ordinary zeta.

**[project; exact obstruction]** At a hypothetical noncentral ordinary zero of
ANY multiplicity m, assign `Theta_G=m*1_G` in every field of a family or tower.
Every subfield zeta then has order m; all nontrivial Artin factors have order
zero. Restrictions, projectors, no-cancellation, the norm inequality (with
equality), and the sharpened trivial-component budget (0<=0) all hold. The same
assignment respects conjugate/reflected points and does not depend on Re(s0).
Thus infinitely many compatible bookkeeping constraints alone cannot disprove
an off-line zero or a multiple ordinary zero. A new independent zero-location,
small-multiplicity, or forced-coincidence theorem would have to break that model.
Even independently proving M=1 at every ordinary zero would imply simplicity,
not RH. Details and precise limits: `think-hard-heilbronn.md` §§1–3.

## 4. Apply it to our Q8 fields, without attributing Artin zeros to zeta

**[known identity; project field identification]** For the degree-eight field
already audited (discriminant 205^6), put A_D=L(s,chi_D), B=L(s,rho). Then

    zeta_K = zeta A_5 A_41 A_205 B^2,
    zeta_M = zeta A_5 A_41 A_205,
    M = Q(sqrt(5),sqrt(41)).

Every nontrivial factor is entire: the quadratic factors by Dirichlet/Hecke,
and B because rho is induced from an order-four character of an index-two cyclic
subgroup. Thus the stronger order formula has only nonnegative terms in the
nontrivial strip:

    m_K = m_0+m_5+m_41+m_205+2m_rho.                            (Q8)

**[known consequence; project application]** The sign -1 forces an odd-order
central zero of B. Our earlier interval derivative work identifies order one for
the specified examples. It therefore forces an upstairs field zero of order at
least two, not an ordinary-zeta zero. The earlier ordinary/quadratic central
nonvanishing certificates make the field order exactly two in the base example.
The central value zeta(1/2) is about -1.46035, not zero.

At an ordinary zero of multiplicity m, (Q8) forces m_K>=m. It does **not** force
m_rho>0. Under known holomorphy a shared Artin zero would force m_K>=m+2. Hence a
field multiplicity bound can forbid that coincidence. Conversely, knowing a
simple B zero only bounds the *added* contribution; it gives no zero or slope
of zeta at that point. A derivative certificate for B at 1/2 is not a simplicity
certificate for any ordinary-zeta zero. Proving all nontrivial zeros of zeta_K
simple is not a viable goal: its squared entire Artin factor gives infinitely
many multiple zeros, even away from the forced center. More generally H5 proves
infinitely many multiple zeros for any nonabelian Galois extension. The useful
claim would instead isolate the ordinary component or particular zero boxes.

**[known consequence]** Nonvanishing of the full zeta_K at a regular point forces
nonvanishing of zeta there. Nonvanishing of B alone does not. Even nonvanishing
of every nontrivial factor alone leaves the possibility that the trivial factor
zeta vanishes; it only makes the field and ordinary orders equal. Special-value
identities need this same factor-by-factor audit, not an inference from the
central sign or a nonzero Artin derivative.

### 4.1 An exact downward projector using intermediate fields

**[known character-theoretic consequence; project exact integer check]** Let
F_5,F_41,F_205 be the quadratic subfields. Their orders satisfy

    2 m_0 = m_(F_5)+m_(F_41)+m_(F_205)-m_M,
    2 m_rho = m_K-m_M.                                        (P)

This holds at each s0 != 1, using signed orders where needed. It follows from
zeta_(F_D)=zeta A_D and the biquadratic factorization; equivalently,

    1_G = (1/2)[sum_(three H_i of index 2) Ind_(H_i)^G 1
                         - Ind_(center)^G 1].

`think_hard_checks.py` checks this identity in the exact Q8 character table.
Thus independently known subfield orders **do** recover an ordinary-zeta order.
The minus sign is crucial: nonnegative subfield orders by themselves do not give
a useful lower bound for m_0. This is an extraction identity, not a new positivity
argument. A contour version gives the same identity for zero-minus-pole counts,
provided every boundary is zero/pole-free. Our existing certificates largely
travel **upward**, because they start with separate factor evaluations; calling
that an independent downward proof would be circular.

### 4.2 Twisting does not give eighteen independent tensor bridges

**[known consequence of quadratic twisting; project character check]** For each
admissible quadratic twist rho_D=rho tensor chi_D,

    End(rho_D) = rho_D tensor rho_D* = End(rho),
    L(s,End(rho_D)) = zeta_M(s).

The common biquadratic subfield remains fixed. Consequently

    zeta_(K_D) = zeta_M B_D^2.

The central-slope experiments distinguish B_D but do not vary this tensor
bridge to zeta. Repeating its identical End identity does not create independent
constraints. Even tensor powers likewise repeat the same four abelian characters;
odd powers repeat rho (with the appropriate twist). Higher symmetric powers can
contain zeta factors, but their zeros are not obtained by multiplying old zero
heights. See the prior exact tensor/symmetric-power audit in `ledger.md` A2.

**[speculative, exact target]** A uniform family theorem that separately bounded
the common zeta_M contribution, or recovered it with a positive projector on
suitable tests, could help. Averaging away B_D does not automatically leave a
nonnegative zeta component: one must prove the identity and error/positivity
estimate, not just invoke variation in D.

### 4.3 Fresh local certificates: orders, not decimal evaluations

**[project; interval-certified]** `scripts/heilbronn_local_certificate.py` and
`results/heilbronn-local/certificate.json` certify the following for the field
above. “At an ordinary zero” means the actual unique zero in a rigorous sign
bracket, not a rounded decimal seed.

| target | ordinary zeta | each quadratic subfield zeta | biquadratic zeta | degree-eight field zeta | two-dimensional B |
|---|---:|---:|---:|---:|---:|
| 1/2 | 0 | 0 | 0 | 2 | 1 |
| first ordinary zero, height about 14.13472514 | 1 | 1 | 1 | 1 | 0 |
| second ordinary zero, height about 21.02203964 | 1 | 1 | 1 | 1 | 0 |
| third ordinary zero, height about 25.01085758 | 1 | 1 | 1 | 1 | 0 |

Entries are **zero orders**. The three nontrivial quadratic Artin factors are
nonzero at all four targets. At each noncentral target every nontrivial factor
is nonzero on the whole rational square of half-width 10^-6. Fresh Arb contour
counts have explicit coefficient/integration/quadrature/Taylor errors. Root sign
−1 is additionally selected by a rigorous Euler-product/theta comparison at s=2.
Stark and (P) recover the claimed ordinary and Artin orders.

**[project; independence correction]** These field/subfield contours use the
factorization, not an independent degree-eight analytic evaluator. The previous
height-30 certificate already implied the local result. Therefore this is a
fresh, explicit finite witness and bookkeeping check, **not independent new
higher-to-lower evidence for ordinary simplicity or RH**. Full analytic bounds,
trust scope, and primary-source audit are in `think-hard-heilbronn.md` §4.
Figure: `figures/heilbronn-local-orders.png`.

## 5. Li coefficients and positivity: the subtraction trap, and the all-index exception

### 5.1 Completed products, not uncompleted Euler factors

**[known; P1,P2; additivity follows by differentiating logs]** For an entire
completed factor X nonzero near s=1, define

    lambda_n(X) = 1/(n-1)! [d^n/ds^n {s^(n-1) log X(s)}]_(s=1).

Constants multiplying X do not matter. All poles/endpoint zeros must be removed
in a convention consistent across factors. Here

    xi_K = const * xi_zeta Lambda_5 Lambda_41 Lambda_205 Lambda_rho^2,
    xi_zeta = const * s(s-1) pi^(-s/2) Gamma(s/2) zeta(s).

The field and ordinary zeta require **one** s(s-1) pole-removal factor, not one
for every Artin constituent. The nontrivial Artin completions here are entire
and nonzero at 1. Therefore exactly

    lambda_n(K) = lambda_n(zeta)+lambda_n(5)+lambda_n(41)
                               +lambda_n(205)+2 lambda_n(rho).       (L)

At the zero-sum level the same identity follows from the union of zero multisets,
with consistent symmetric limiting conventions. A list of low zeros alone does
NOT compute the full coefficient without an error bound for the omitted tail.

**[known; P1,P2,P6]** Li's criterion applies to completed Dedekind zeta: all lambda_n(K)
nonnegative iff its nontrivial zeros have real part 1/2. The appropriate real-part
criterion holds for the principal automorphic L-functions in Lagarias's theorem;
our self-dual monomial factors fit the usual real self-dual setting.

### 5.2 What does and does not follow from positivity upstairs?

**[known consequence]** At a **fixed** index, lambda_n(K)>=0 does not isolate the
sign of lambda_n(zeta). Nor does knowing all the other summands are positive make
subtraction sign-preserving. A sufficient quantitative estimate is

    lower_bound(lambda_n(K)) >= upper_bound(sum other summands).

For an entire Weil form the analogous usable statement is domination of the
quotient form, not just positivity of the total form.

**[known; important quantifier correction]** However, if **every** field Li
coefficient is nonnegative, it DOES imply every ordinary-zeta Li coefficient is
nonnegative: apply the full field Li criterion, then zero inclusion, then ordinary
Li's criterion. In our fully holomorphic Q8 product, all-index field positivity
actually implies GRH for all its factors. Thus “positivity of a sum does not imply
positivity of its pieces” must be qualified as a finite-index/algebraic inference;
it must not be used to deny the all-index theorem-level implication. A hypothetical
off-line ordinary zero is an upstairs zero and cannot disappear from that full
criterion.

### 5.3 New computation / figure: the small piece hidden inside a large positive total

**[project]** We reran the existing first-20 coefficient calculation at 110-digit
PARI precision, without changing its mathematical method. It evaluates completed
Taylor series at s=1, NOT truncated zero sums. Comparison to the old 90-digit
run agrees to better than 1.15e-83 on all stored factor/field values. Printed-value
additivity residuals are below 2.16e-103. These are floating-point consistency
checks, not interval enclosures and not independent analytic evaluations.

| n | ordinary zeta | three quadratic contributions | twice rho | field total |
|---:|---:|---:|---:|---:|
| 1 | 0.0230957089661 | 1.24775718230 | 6.80639127660 | 8.07724416786 |
| 2 | 0.0923457352280 | 4.75905473098 | 10.5059965102 | 15.3573969764 |
| 3 | 0.207638920554 | 9.91796761130 | 25.2768933188 | 35.4024998507 |

**[known calculation; project visualization]** A simple central zero rho=1/2
contributes

    1-(1-1/rho)^n = 1-(-1)^n

in the Li zero sum. Thus its **squared field factor** contributes 4 for odd n and
0 for even n. We separated that exact term in the plot. At n=1 ordinary zeta is
only 0.286% of the total, while the forced central field term alone is about
49.5%. A pleasing large positive field coefficient can mostly measure other
factors, including a forced zero that ordinary zeta does not possess.

Artifacts:

* `figures/think-hard-li-decomposition.png` and `.svg`.
* `scripts/li_transfer_think_hard.py` — new 110-digit run; same method as the prior script.
* `results/li-transfer-think-hard.json` — all 20 rows and direct product checks.
* `scripts/think_hard_checks.py`, `results/think-hard-checks.json` — precision comparison,
  exact character projector/twist checks, and figure generation.

This repeats/strengthens a numerical bookkeeping check already present in the
ledger, rather than pretending Li additivity was discovered today.

## 6. Rankin–Selberg, functoriality and Jacquet–Langlands: what is actually preserved?

**[known; R1–R3]** For a unitary cuspidal pi on GL(n)/Q, at the level of the
appropriate full local parameters,

    L(s,pi x pi*) = zeta(s) L(s,Ad pi).

The tensor parameter contains the scalar endomorphisms. To use zero inclusion,
require the adjoint quotient to be holomorphic at the target point. This is
proved for GL2 by Gelbart–Jacquet (Ad is the symmetric square with determinant
removed), and for GL3/GL4 by Yang's theorem and finite-part corollary. General
rank is not obtained merely from the Rankin–Selberg pole at 1. Yang's result is
published, not an assumption of arbitrary-rank adjoint functoriality.

A zero-free or zero-density result for the full Rankin–Selberg L-function, in a
region where the quotient is holomorphic, descends to zeta just as in §2. GRH
of the original L(s,pi) does NOT itself prove GRH of L(s,pi x pi*). Tensoring acts
on Frobenius/Satake parameters prime by prime, not on the zero ordinates of the
original function. Also pi x pi* is NOT the holomorphic function |L(s,pi)|^2.

**[known; R4]** Jacquet–Langlands is a genuine arithmetic bridge between suitable
automorphic representations on a quaternion algebra B^x and GL2. In the
non-character cuspidal setting it preserves the matching standard local
L/epsilon factors, hence the standard global L-function. This lets us transfer
a theorem about *that same* automorphic L-function across realizations. It is
not the slice lift q -> zeta(q), and does not by itself put ordinary zeta inside
a cuspidal degree-two standard L-function.

**[known consequence]** A further Rankin–Selberg/adjoint construction can then
supply the zeta factor. That is a two-step bridge:

    quaternionic automorphic form <-> GL2 cusp form
       -> Rankin–Selberg product = zeta * entire adjoint quotient.

The missing payload is a sufficiently strong theorem for the **product**, not
just JL matching. A quaternion ideal-counting zeta (Voight's §26 formula) is yet
another object: its direct zeta factors already give a simpler bridge without
invoking JL. Eichler/Shimizu/JL correspondences concern automorphic forms/Hecke
modules; the name “quaternion” does not identify all their zeta functions.

## 7. Exact equivalents and limits: stronger than analogy, not free proofs

### 7.1 Weil positivity and restriction from a larger space

**[known; P2,P3]** In one admissible convention, take smooth g compactly supported
in (0,infinity), Mellin transform G(s), and G(0)=G(1)=0. The Weil criterion is

    Q_zeta(g) = sum_rho G(rho) conjugate(G(1-conjugate(rho))) >= 0

for **every** such test. The explicit formula identifies this with a precisely
normalized prime/archimedean expression. On RH the summands become |G(rho)|^2.
Before RH, replacing the reflected argument by rho assumes the desired answer.

**[known; functional analysis]** If we could prove a positive form Q_big on a
larger space and an embedding J satisfying

    Q_big(Jg,Jh) = Q_zeta(g,h)

for the full admissible test class, positivity would descend by restriction.
This is a genuine bridge. The hard requirements are the exact identity,
positivity independent of RH, and domains/continuity or density control. A
positive norm of a quaternion or octonion supplies none of those identities.
For block decomposition the trivial-representation block must be identifiable:
positivity of a scalar *trace over all blocks* is not positivity of each block.

**[known; P2]** Lagarias relates the Li test functions to the Weil Gram form;
for ordinary self-dual zeta the finite Gram entries are

    W_nm = lambda_n + lambda_m - lambda_|n-m|,   lambda_0=0.

The prior interval-positive 32-dimensional compression checks its stated span,
not all tests. Ordinary norm positivity must not be confused with positivity of
this arithmetic form. Infinite-dimensional structure is useful only if it proves
the missing global statement.

### 7.2 Nyman–Beurling / Báez–Duarte

**[known; P4]** One exact formulation is

    1_(0,1] belongs to closure span{ {1/(a x)} : a=1,2,... }
       in L2((0,infinity), dx)   <==> RH.

The finite-dimensional spaces may grow, but the theorem asks that the distance
go to zero, not merely that it decrease. The Mellin transform exposes the
obstruction: evaluation at an off-line zero gives a continuous functional that
annihilates the approximating span but not the target (in the appropriate
half-plane). Explicit feasible approximation coefficients plus a rigorous error
bound tending to zero would transfer down. A fitted distance at finitely many N,
or an extrapolated rate, does not. Enlarging the span arbitrarily may remove the
obstruction while losing equivalence: specify the exact allowed subspace.

### 7.3 de Bruijn–Newman

**[known; P5]** In the Rodgers–Tao normalization H_0(z)=xi(1/2+iz/2)/8, the heat
deformation has all zeros real exactly for t>=Lambda. RH iff Lambda<=0;
Rodgers–Tao prove Lambda>=0, so RH iff Lambda=0. A proven sequence t_j decreasing
to 0 with every H_(t_j) all-real-rooted would suffice, using locally uniform
convergence and Hurwitz on the upper/lower half-planes. All-real zeros at one
positive time do not flow backward automatically; the deformation can repair
nonreal zeros before that time. None of the project computations proves Lambda<=0.

### 7.4 Hurwitz and which approximation theorem is missing

**[known; classical Hurwitz theorem]** On a connected open domain D, holomorphic
nonvanishing f_j converging locally uniformly to f imply either f is identically
zero or f has no zeros. Apply this to xi on each off-critical half-plane (or an
exhaustion of them) with f nonzero somewhere. This would prove RH if the
approximants were proved zero-free there. Pointwise convergence, convergence only
on the real axis, a finite sequence, or zero-freeness on domains that never cover
a fixed compact neighborhood of an alleged off-line zero is insufficient.

**[project; design constraint]** Approximants deliberately built from already
known on-line zeros are particularly vulnerable: proving they converge to xi
*as whole holomorphic functions* with no omitted zero factor is where RH could
have been inserted. State the convergence proof before reading the plot.

### 7.5 A genuine spectral-geometric bridge that self-adjointness does not finish

**[known; R5]** There is a particularly instructive larger-space realization on
the modular surface PSL2(Z)\H. Its scalar scattering function is exactly

    phi(s) = Lambda(2s-1)/Lambda(2s)
           = sqrt(pi) Gamma(s-1/2)/Gamma(s) * zeta(2s-1)/zeta(2s),
    Lambda(w)=pi^(-w/2) Gamma(w/2) zeta(w).

For a nonreal ordinary zero rho, s=rho/2 is a scattering pole of the same
multiplicity: the numerator Lambda(rho-1)=Lambda(2-rho) is finite and nonzero
by the functional equation and the Euler product in Re(2-rho)>1. Thus ordinary
RH is equivalent to all nonreal scattering poles in 0<Re(s)<1/2 lying on
Re(s)=1/4. This is a **both-way exact bridge**, stronger than a loose comparison
to Selberg spectra.

The Laplacian of that surface is self-adjoint on L2, but these scattering states
are not L2 eigenvectors. Poles of the continued resolvent/scattering function
are resonances, not the real spectrum of the self-adjoint L2 operator. Its
self-adjointness therefore does not prove the required resonance line. A new
geometric resonance-location theorem would transfer immediately; the existing
spectral theorem is not that theorem. This is another precise way geometry could
help without pretending it already solves RH.

## 8. The finite-field precedent: moving to a surface really can help

### 8.1 What is proved there

**[known; G1,G2]** For a smooth projective geometrically connected curve C/F_q,

    Z_C(T) = det(1-T F | H^1)/( (1-T)(1-qT) ),
    #C(F_(q^r)) = 1+q^r - Tr(F^r | H^1).

Frobenius is an actual endomorphism. Its H^1 eigenvalues alpha have |alpha|=sqrt(q),
so T=q^(-s) puts the zeros on Re(s)=1/2. The eigenvalue magnitude theorem is NOT
just the functional equation's pairing alpha <-> q/alpha.

**[known; G1, reconstructed surface argument]** Let S=C x C, Delta its diagonal,
and Gamma the graph of q-Frobenius. Put N=#C(F_q). Intersection theory gives

    Delta.Gamma=N,  Delta^2=2-2g,  Gamma^2=q(2-2g).

For D=a Gamma+b Delta, the Hodge-index inequality on the component perpendicular
to the two fiber directions gives

    D^2 <= 2 (D.fiber_1)(D.fiber_2),
    0 <= 2gq a^2 + 2(q+1-N)ab + 2g b^2.

The discriminant bound is |N-(q+1)|<=2g sqrt(q). Applying this over every finite
extension gives RH for C. This is a real success story of a two-dimensional
surface constraining a one-dimensional curve's zeta. Weil's original
correspondence positivity sigma(eta eta')>=0 gives the same trace-power bound.
The inspected 1948 primary text states it in §IV no.21, Theorem 13 Corollary 3,
and derives RH in no.22.

**[known; G2]** In cohomological language, the Künneth formula identifies pieces of
H^2(C x C) with H^1(C) tensor H^1(C), so correspondences become endomorphisms and
intersection/trace pairings can be compared. Künneth alone is a decomposition,
NOT positivity. Duality, algebraic correspondences, an appropriate polarization
and positivity or the corresponding trace bound do the work.

**[known; G2]** For smooth projective higher-dimensional X/F_q, Deligne proves
weights: eigenvalues on H^i have absolute value q^(i/2). The zeta function is an
alternating product of determinant factors with zeros **and poles** at their
weight lines. Do not assign one universal central line to its entire divisor,
and do not say Deligne proved the full standard conjectures of Hodge type.
His general proof is not simply “ordinary real positive definite matrices on
all l-adic cohomology”; l-adic coefficients do not carry that naive order.

### 8.2 What fails to be supplied over Q

**[known facts / speculative program; G3,P3]** We do have l-adic representations
and local Frobenius elements at unramified primes over number fields. The missing
object is NOT “any Frobenius over Q.” It is a global realization with the required
spectrum/determinant for completed ordinary zeta, compatible trace/explicit
formula, archimedean contribution, and a provably positive pairing that imposes
the weight or self-adjointness condition. Spec Z has no finite-field absolute
Frobenius t -> t^q at a fixed q; its primes have different residue characteristics.
Existing local purity does not settle the zeros of the global Euler product.

A global analogue of C x C with the right correspondences, intersection pairing,
Künneth behavior and Hodge-index inequality could carry such a proof. But one
cannot just take Spec Z x_(Spec Z) Spec Z, which is Spec Z again, and declare an
arithmetic surface with all these properties. Arakelov geometry and known
arithmetic intersection theorems are genuine mathematics; their existence alone
does not identify the particular Weil form with an independently positive one.

**[known component results / speculative bridge; G3,P3]** Deninger's cohomology,
flow and Hodge-star/height-pairing package explicitly lists conjectural objects
and compatibility. In its simplest target, an operator Theta with
Theta+Theta*=1 on a positive Hilbert space and determinant equal to xi would
force Re(rho)=1/2. Constructing the required global operator, domains and identity
is not accomplished by writing this equation.

Connes's spectral/trace-formula approach and Connes–Consani's arithmetic/scaling
site provide different frameworks and proved components. Their existence is not
a completed all-test positivity proof. The inspected Connes–Consani archimedean
paper proves positivity under specified support/vanishing restrictions; it does
not supply unrestricted all-prime/all-support Weil positivity. A spectral
realization whose positivity/absorption interpretation is equivalent to RH is
not an independent positive self-adjoint realization already proving RH.

**[project; verdict]** The transferable *lesson* is very precise: geometry can
help if its positive pairing equals the arithmetic pairing. The transferable
*theorem* from finite-field RH to ordinary RH is absent. This is a serious
program, not an implication we may claim now.

## 9. Hypercomplex extensions: an equivalence, not a second supply of zero data

### 9.1 The intrinsic slice lift contains exactly the old zero information

**[known; X1,X2]** Write the ordinary intrinsic function as
zeta(x+iy)=u(x,y)+iv(x,y), with u,v real. Away from its pole its H/O slice lift is

    zeta_A(x+Iy) = u(x,y)+I v(x,y),    I^2=-1,
    |zeta_A(x+Iy)|^2 = u(x,y)^2+v(x,y)^2.

Therefore it vanishes iff zeta(x+iy)=0. For y!=0 the entire sphere x+y S^(d-2)
vanishes, d=4 or 8. Restricting to one complex slice recovers the exact ordinary
function. In this intrinsic class the operations are inverse. At real zeros the
sphere collapses to a point.

**[known consequence]** Any full zero-location theorem for the lift descends;
any such theorem downstairs lifts. The two zero-location problems are
information-equivalent. A new *proof technique* in the larger geometry is not
logically excluded: an equivalent reformulation can be easier. But it must
supply a coordinate-sensitive inequality that was not simply roundness,
multiplicity, conjugation symmetry, or a restatement of RH.

**[project; inference boundary]** The spherical tube degree measures multiplicity
(2m for a nonreal spherical zero in our conventions), not the real coordinate x.
The coarse complement/end pattern explored in `hyper-critical/infinity.md` can
coexist with horizontal displacement of spheres. We do not have a theorem that
all possible topology is irrelevant; we do have a clear reason the tested
invariants alone cannot select x=1/2. Nonassociative polynomial examples that
move a point on an unchanged sphere supply no zeta factor or Weil-form identity.

### 9.2 What changes the problem, and what new bridge it would need

**[known distinctions; speculative project targets]**

1. **Non-real coefficients.** General slice functions can have isolated,
   direction-dependent zeros, and one slice contains multiple complex components.
   They are different functions, not extra values of the same scalar zeta.
   Keeping exact scalar zeta values on a real interval in the usual slice-regular
   identity-principle setting does not allow arbitrary new directional behavior:
   the extension is uniquely fixed. To return information from a genuinely
   changed function, produce a factor, determinant, projection or form identity.

2. **Fueter / Cauchy–Riemann–Fueter / Clifford monogenic notions.** These satisfy
   different PDEs from slice regularity. The quaternionic Fueter mapping uses
   the Euclidean Laplacian on a suitable slice lift to obtain a monogenic
   function. This map is not automatically zero-preserving or injective:
   Delta(1)=Delta(q)=0 although 1 has no zeros and q has a zero at 0.
   A zero-free theorem for the image cannot generally be pulled back to zeta.
   An inverse/reconstruction with kernel, boundary conditions and a proved
   implication for zeros would be a new gate. Do not quietly substitute one
   regularity notion for another. [X3]

3. **Quaternionic modular/automorphic forms.** They can encode different genuine
   arithmetic, with Hecke operators and JL matching. Their usable bridge is
   the specified automorphic L-function identity in §6, not the dimensionality
   of their values.

4. **Quaternion-algebra order zetas and lattices.** Here exact product formulas
   really do transfer. For the maximal definite rational order convention in
   our ledger,

       Z_D(s)=zeta(2s)zeta(2s-1) product_(p|D)(1-p^(1-2s)).

   Its complete nonreal zero-location assertion on the lines 1/4,1/2,3/4 is
   equivalent to ordinary RH. The finite factors give the middle line; in the
   two open inherited strips the other zeta factor cannot cancel a nonreal
   inherited zero. For E8,

       Z_E8(s)=240*2^(-s) zeta(s)zeta(s-3).

   The corresponding complete assertion on 1/2 and 7/2 is equivalent to RH.
   Its mirror center 2 is NOT the inherited critical line and is actually
   zero-free, as already proved in the ledger. The functional-equation center
   n/4 alone is normalization, not a zero-location theorem. [Q1,Q2]

## 10. Families, symmetry, Selberg degree and “orthogonality”

### 10.1 Q8 twists: two different symmetry vocabularies

**[known; F1 Theorems 1.4,8.6]** Our Q8 representation has Frobenius–Schur
indicator -1: it is symplectic/quaternionic as a representation. For the specific
coprime fundamental-discriminant quadratic-twist family in Shankar–Södergren–
Templier, the root number is constant, and the low-lying-zero symmetry is
**SO(even)** when the sign is +1 and **SO(odd)** when it is -1. Thus “symplectic
representation” and “orthogonal low-lying-zero family” are compatible, not a
contradiction. Neither is a claim of GSE spacing for ordinary zeta.

The inspected one-level theorem has a test Fourier-support restriction
alpha<4/11; on that support it gives the corresponding orthogonal statistic.
The paper also explicitly notes Rubinstein's stronger support alpha<1/2 result
for this family, and a GRH-conditional enlargement to alpha<1; the displayed
4/11 theorem is not the strongest result cited there. None is all-support.
SO(even) and SO(odd) cannot be fully distinguished by the small-support one-level
formula alone. The constant root sign and forced central zero supply the parity
information; the paper's theorem must not be upgraded to all-support density or
all central zeros simple. In versions not assuming GRH, zero parameters may be
complex and tests are interpreted accordingly, not as already-proved real
ordinates. [F1 §§2.6–2.8,8]

**[known consequence]** A limiting average about the B_D factors does not prove
anything zero-by-zero about the fixed ordinary zeta factor absent a separate
factor/projection mechanism. An average can ignore rare exceptions, and a
one-level statistic with limited tests is weaker still. This does not contradict
§2's genuine family-average descent: that descent requires zero counts of the
**full field zetas containing zeta**, a specified forbidden region and a bound
strong enough to exclude the common fixed contribution. Those are not the
one-level-density facts we have proved for B_D.

### 10.2 Selberg axioms and degree

**[known; S1,S2]** The Selberg class imposes Dirichlet series, Ramanujan-type
coefficients, Euler product, analytic continuation and functional equation.
Degree d=2 sum_j alpha_j comes from the gamma factors and is additive in a
product. It controls the leading total zero-count term, not the horizontal
position of individual zeros. Class membership does not prove RH; Selberg's RH
for the class is an additional conjecture. A holomorphic irreducible Artin
L-function has the expected class membership, but membership must not be
assumed in a proof of its own holomorphy.

### 10.3 Orthogonality is about prime coefficients, not disjoint zero sets

**[known as a conjecture with proved consequences; S1,S2]** Selberg orthogonality
for distinct primitive F,G is the assertion

    sum_(p<=x) a_F(p) conjugate(a_G(p))/p = O(1),
    sum_(p<=x) |a_F(p)|^2/p = log log x + O(1).

Conrey–Ghosh obtain uniqueness of primitive factorization under the relevant
Selberg conjectures. Murty proves that Selberg's Conjecture B implies Artin
holomorphy. These are genuine conditional structural bridges, not unconditional
proofs from the axioms alone and not a proof of ordinary RH even after Artin
holomorphy is granted.

**[known consequence of Chebotarev and character orthogonality]** For the finite
Artin characters of one fixed extension the prime correlations themselves have
the character-inner-product main term (with finitely many ramified primes
irrelevant). Yet sharing some zeros is not excluded by this prime-coefficient
orthogonality. Finite common zeros and central forced zeros are not the same
question as leading prime correlations. Therefore do not infer simplicity,
zero disjointness, or ordinates of zeta from the word “orthogonality.” General
Selberg-class primitive factorization is a different assertion from this
finite-group calculation.

**[known consequence; prime-side bridge]** Chebotarev estimates for all conjugacy
classes can be summed to count rational primes (up to finitely many ramified
ones), and the trivial character projects out the ordinary prime-counting term.
This is a genuine map back to primes. An estimate only for a traceless/nontrivial
Artin character does not control the missing trivial component. For Q8, the
2-dimensional character even vanishes on the order-four conjugacy classes;
its alone cannot recover an arbitrary constant-class contribution.

## 11. What would reach the primes, and how much?

**[known; P3 and the standard logarithmic-derivative explicit formula]** The route
is not “extra dimensions -> regular primes.” It is

    proved upstairs exclusion/order/count statement
       -> ordinary-zeta zero statement by a specified bridge
       -> explicit formula or contour shift for -zeta'/zeta
       -> a stated error for ordinary prime counting.

In particular RH is equivalent to

    psi(x) = x + O(sqrt(x) log^2 x),
    psi(x)=sum_(p^k<=x) log p,

and RH implies pi(x)=Li(x)+O(sqrt(x)log x). The historical von Koch source and
our logarithmic-derivative converse audit are in `ledger.md` after S24.
Finite-height certification gives finite-height/truncated-explicit-formula
control with explicit remainders, not an all-x RH error. A zero-density result
alone usually yields a different quantitative statement; it must not be sold
as excluding every exceptional zero.

**[project; caution]** A higher-dimensional theorem must improve something
*after* paying for conductor, multiplicity, quotient and remainder terms. A
beautiful larger total whose ordinary component is unknown is not yet an
improved bound on the primes.

## 12. Four concrete investigations, with explicit pass/fail tests

These are project-scale tasks, not promises of global RH progress.

### A. Independent Heilbronn/subfield order-budget certificate — first priority

**[project proposal]** Use the existing Q8 field and intermediate fields. Select
a small set of disjoint contour boxes, including the first ordinary zero, a
rho zero, the forced center, and at least one close cross-factor pair. Evaluate
zeta_K and subfield zetas through an analytic implementation not just assembling
the previously certified factor values; document any internal factorization.
A direct ideal/theta evaluator is the intended independence target, not PARI's
internally factorized Dedekind evaluator.

**Pass test:** interval-contour counts with zero-free boundaries satisfy (P),
recover the known ordinary count and rho multiplicity, and agree with the
independent factor certificates. An integer feasibility table using the
Heilbronn constraints must show exactly what each subset of field measurements
determines or leaves ambiguous. **Failure/limit:** if the evaluator shares the
same factor oracle, label it a consistency check, not an independent downward
certificate. This tests the mechanism's implementation; it does not improve the
current certified height or global RH unless separately demonstrated.

### B. Certified Li subtraction, with the central-zero contribution isolated

**[project proposal]** Upgrade the first 20 entries to interval-completed Taylor
series, including the monomial Hecke/Artin evaluator, not merely more floating
precision. Prove interval upper/lower estimates in (L), and compute the
ordinary block and field block of the Weil Gram form under identical conventions.

**Pass test:** the interval difference “field minus other factors” encloses the
independently computed ordinary coefficients and Gram entries. Show the exact
4/0 central parity term. Report widths and cancellation cost. **Failure/limit:**
positivity of the field total alone must not be counted as a proof of each
ordinary coefficient; a finite positive block still is not RH. Much of the base
ordinary/quadratic interval machinery already exists, so the new difficulty is
certified Artin Taylor data and an independently checked subtraction.

### C. A return-map gate for any new hypercomplex/positivity proposal

**[project proposal]** Before another large zero scan, require one explicit
identity of the form Q_big(Jg,Jh)=Q_zeta(g,h), or a holomorphic factor identity
with proved no cancellation, or an injective reconstruction with a zero-location
implication. Test a candidate on rational/polynomial stems and a deliberately
off-line conjugate/reflection-symmetric toy divisor first.

**Pass test:** it both recovers the original complex arithmetic form/function and
retains enough information to distinguish an off-line toy zero from an on-line
one. If it claims unconditional positivity, prove that positivity without
replacing reflected arguments by absolute squares. **Failure/limit:** norms,
round spheres or a Fueter transform that loses affine kernel data fail this
return-map gate. Passing a finite toy test only keeps a candidate alive; it is
not proof of the required zeta identity or global positivity.

### D. Quantitative “does it actually improve zeta?” comparison

**[project proposal]** Choose one precise published field zero-free/density
estimate and a direct ordinary-zeta estimate; implement both in exactly the
same target height/region convention using D_K=205^6. Then examine whether
any family averaging has a proved common-factor count bound rather than just
a low-lying B_D statistic.

**Pass test:** a rigorously stated transferred excluded region or count bound
strictly beats the selected direct bound on a stated range, with all constants
and exceptions accounted for. **Failure is useful:** an explicit dominated
comparison documents why the added degree currently costs more than it earns.
Do not describe an improvement over one old comparison theorem as the best
known bound without a separate literature audit.

## 13. Primary sources and source audit

Citations below name the source supplying the theorem, not just a popular
explanation. Source IDs from the earlier ledger are not reused here.

### Artin factorization and zero orders

* **A1.** H. Aramata, *Über die Teilbarkeit der Dedekindschen Zetafunktionen*,
  Proc. Imperial Academy 9 (1933), 31–34.
  <https://doi.org/10.3792/pia/1195580866>. Original archived earlier as
  `sources/aramata-jstage.pdf/.txt`; opening Galois theorem inspected in the earlier audit.
* **H1.** H. Heilbronn, *On Real Zeros of Dedekind zeta-Functions*, Canadian J.
  Math. 25 (1973), 870–873. Theorem 1 and its Artin/virtual-character proof.
  <https://doi.org/10.4153/CJM-1973-090-3>.
  Original downloaded/read in full here: `sources/think-hard/heilbronn-1973.pdf/.txt`.
* **H2.** H. M. Stark, *Some effective cases of the Brauer–Siegel theorem*, Invent.
  Math. 23 (1974), 135–152, Theorem 3.
  <https://doi.org/10.1007/BF01405166>;
  <https://eudml.org/doc/142257>. Publisher original PDF not retrieved here;
  theorem cross-checked in H4 Prop.3.3 and H5 Lemma 3.1, and the needed simple-zero
  consequence is proved explicitly above rather than accepted from a search summary.
* **H3.** R. Foote and V. K. Murty, *Zeros and poles of Artin L-series*, Math.
  Proc. Cambridge Philos. Soc. 105 (1989), 5–11.
  <https://doi.org/10.1017/S0305004100001316>.
  R. Foote and D. Wales, *Zeros of order 2 of Dedekind zeta-functions and Artin's
  conjecture*, J. Algebra 131 (1990), 226–257.
  <https://doi.org/10.1016/0021-8693(90)90173-L>.
  Original theorem attributions cross-checked in the following primary refinement;
  these two complete original proofs were not retrieved/read here.
* **H4.** P.-J. Wong, *Zeros of Dedekind zeta functions and holomorphy of Artin
  L-functions*, J. Ramanujan Math. Soc. 34 (2019), 253–261,
  Theorems 1.1–1.4, Propositions 3.1–3.3, Theorem 3.5 and Corollary 3.6.
  <https://www-math.nsysu.edu.tw/~pjwong/stuff/LowOrderZero.pdf>.
  Author original archived/read, including hypotheses and proof of quotient refinement.
* **H6.** M. Ram Murty and A. Raghuram, *Some Variations on the Dedekind
  Conjecture*, J. Ramanujan Math. Soc. 15 no.4 (2000), 225–245.
  <https://mrammurty-ai.github.io/murty-raghuram.pdf>.
  Original retrieved; Theorems 4.1, 4.4–4.6, 5.1, 5.4 and Corollary 5.6
  inspected on rendered scan pages. Fuller proof and source-status audit in
  `think-hard-heilbronn.md`. Heilbronn's Boulder 1972 contribution remains
  unretrieved; it must not be confused with H1's inspected 1973 journal proof.
* **H5.** D. Hu, I. Kaneko, S. Martin and C. Schildkraut,
  *Order of Zeros of Dedekind Zeta Functions*, arXiv:2107.03269.
  <https://arxiv.org/abs/2107.03269>. Original §§2–3 inspected; its new higher-order
  existence results also warn against conjecturing simplicity of all field-zeta zeros.

The Foote–Ginsberg–Murty 2015 *Heilbronn characters* survey was located but its AMS
PDF returned 403. We did not treat this as absence of the paper and did not use
its uninspected proof as the sole source. The original Heilbronn and Wong papers
supply the facts used here. Search-result prose incorrectly said the Q8 family
was “not orthogonal” while giving SO symmetry; the original F1 resolves that
vocabulary error. Another search summary overstated the inverse theorem in the
original JL book; the inspected text is more precise, as noted below.

### Positivity, approximation and deformation

* **P6.** E. Bombieri and J. C. Lagarias, *Complements to Li's Criterion for
  the Riemann Hypothesis*, J. Number Theory 77 (1999), 274–287.
  <https://doi.org/10.1006/jnth.1999.2392>;
  <https://websites.umich.edu/~lagarias/doc/bombieri.pdf>.
  Original introduction, Theorem 1, Corollary 1 and relevant proof inspected
  on rendered pages; explicitly confirms the Dedekind extension and all-index
  quantifier. Archived under `sources/heilbronn-deep/`.
* **P1.** X.-J. Li, *The Positivity of a Sequence of Numbers and the Riemann
  Hypothesis*, J. Number Theory 65 (1997), 325–333.
  <https://doi.org/10.1006/jnth.1997.2137>. Original criterion includes Dedekind
  zetas; original publisher retrieval was unsuccessful in the prior audit.
* **P2.** J. C. Lagarias, *Li coefficients for automorphic L-functions*, Ann.
  Inst. Fourier 57 (2007), 1689–1740, Introduction, Theorems 2.2 and 3.1.
  <https://doi.org/10.5802/aif.2311/>, <https://arxiv.org/abs/math/0404394>.
  Original archived earlier, relevant sections inspected. Product additivity here
  is a direct differentiation of the exact completed identity, not a separate
  theorem falsely attributed to this paper.
* **P3.** A. Weil, *Sur les “formules explicites” de la théorie des nombres
  premiers*, Lund seminar volume (1952), 252–265; A. Connes and C. Consani,
  *Weil positivity and Trace formula: the Archimedean place*, Selecta Math. 27
  (2021), article 77, <https://arxiv.org/abs/2006.13771>.
  The latter original is archived in `sources/connes-consani-positivity.*`;
  introduction equations (1)–(2) and support-restricted theorem audited earlier.
* **P4.** L. Báez-Duarte, *A strengthening of the Nyman–Beurling criterion for
  the Riemann hypothesis*, Rend. Mat. Acc. Lincei 14 (2003), 5–11, Theorem 1.1.
  <https://arxiv.org/abs/math/0202141>. Original archived earlier.
* **P5.** B. Rodgers and T. Tao, *The de Bruijn–Newman constant is non-negative*,
  Forum Math. Pi 8 (2020), e6, equations (1)–(4), Theorem 1.
  <https://arxiv.org/abs/1801.05914>. Original archived earlier; normalization
  and direction checked. Hurwitz's theorem is standard complex analysis;
  its precise hypotheses, not a numerical limit, are explicitly stated above.

### Automorphic transfers and class structure

* **R1.** S. Gelbart and H. Jacquet, *A relation between automorphic representations
  of GL(2) and GL(3)*, Ann. Sci. ENS 11 (1978), 471–542.
  <https://www.numdam.org/item/ASENS_1978_4_11_4_471_0/>.
  Original introduction archived/read in prior source audit.
* **R2.** L. Yang, *Holomorphy of adjoint L-functions for GL(n): n<=4*, Math.
  Ann. 381 (2021), 1745–1805.
  <https://doi.org/10.1007/s00208-021-02189-4>,
  <https://arxiv.org/abs/1903.09881>. Archived author original Theorem A and
  Corollary 3 inspected here. No arbitrary-rank assertion is inferred.
* **R3.** H. Jacquet, I. Piatetski-Shapiro and J. Shalika, *Rankin–Selberg
  convolutions*, Amer. J. Math. 105 (1983), 367–464.
  <https://doi.org/10.2307/2374264>. Standard tensor/contragredient factorization
  used with the explicit holomorphy gates R1/R2, not an assumed adjoint lift.
* **R4.** H. Jacquet and R. P. Langlands, *Automorphic Forms on GL(2)*,
  Lecture Notes in Math. 114 (1970), §14 local factor equalities and Theorem 14.4.
  <https://publications.ias.edu/sites/default/files/automorphic-forms-on-gl2_rpl_5.pdf>.
  Original archived; relevant §14 and §16 statements inspected. Historical caution:
  that original text calls its reverse-direction Theorem 16.1 a formal sketch
  still conjectural at that writing. We use the established quaternion-to-GL2
  direction and exact matching factors here, not that historical sketch as a
  complete proof of all modern converse statements.
* **R5.** Y. N. Petridis, N. Raulf and M. S. Risager, *Quantum Limits of
  Eisenstein Series and Scattering states*, <https://arxiv.org/abs/1111.6615>,
  introduction and equations (2.2)–(2.3). Primary research paper inspected online;
  the scattering identity and non-L2 nature of the states are explicit there.
  Its notation xi omits s(s-1)/2, so we call that function Lambda above.
* **S1.** J. B. Conrey and A. Ghosh, *On the Selberg class of Dirichlet series:
  small degrees*, Duke Math. J. 72 (1993), 673–693.
  <https://doi.org/10.1215/S0012-7094-93-07225-0>.
* **S2.** M. R. Murty, *Selberg's conjectures and Artin L-functions*, Bull. AMS
  31 (1994), 1–14, Proposition 2.4 and Theorem 3.1.
  <https://doi.org/10.1090/S0273-0979-1994-00479-3>,
  <https://mrammurty-ai.github.io/murty-ams.pdf>.
  Original author copy archived; conjectures and stated consequences inspected.

### Geometry, families and hypercomplex functions

* **G1.** A. Weil, *Sur les courbes algébriques et les variétés qui s'en déduisent*,
  Hermann (1948), correspondence trace positivity, §IV nos.21–22.
  <https://gallica.bnf.fr/ark:/12148/bpt6k3372566b>;
  inspected transcription <https://denisevellachemla.eu/Weil-courbes-varietes.pdf>.
  Archived here. The short surface calculation above is separately reconstructed
  from the intersection identities; it does not repeat typographical errors in
  an online lecture-note calculation.
* **G2.** P. Deligne, *La conjecture de Weil I*, Publ. IHÉS 43 (1974), 273–307,
  (1.5.4), Theorem (1.6).
  <https://www.numdam.org/item/PMIHES_1974__43__273_0/>. Original archived earlier.
* **G3.** C. Deninger, *The Hilbert–Polya strategy and height pairings*,
  <https://arxiv.org/abs/1001.1621>, Conjectures 1–2 and Hodge-star discussion;
  A. Connes, *Trace formula in noncommutative geometry and the zeros of the
  Riemann zeta function*, Selecta Math. 5 (1999), 29–106,
  <https://arxiv.org/abs/math/9811068>; A. Connes and C. Consani,
  *Geometry of the Arithmetic Site*, <https://arxiv.org/abs/1502.05580>.
  Deninger and arithmetic-site originals archived earlier; conjectural-vs-proved
  status is integral to the claim, not an implicit proof.
* **F1.** A. Shankar, A. Södergren and N. Templier, *Sato–Tate equidistribution of
  certain families of Artin L-functions*, Forum Math. Sigma 7 (2019), e23,
  Theorems 1.4 and 8.6, §8.
  <https://doi.org/10.1017/fms.2019.18>;
  <https://research.chalmers.se/publication/519413/file/519413_Fulltext.pdf>.
  Original archived/read relevant definitions and theorem, not just the abstract.
  The fraction in Theorem 8.6 was verified on the rendered PDF page (alpha<4/11);
  flattened PDF text can incorrectly suggest 1/14.
* **Q1.** J. Voight, *Quaternion Algebras*, §26, Theorem 26.4.9 and §26.8,
  <https://quatalg.org/>. Original book archived earlier; ideal-index versus
  reduced-norm conventions matter.
* **Q2.** M. Viazovska, *The sphere packing problem in dimension 8*, Ann. Math.
  185 (2017), 991–1015, <https://arxiv.org/abs/1603.04246>, the E8 theta/Eisenstein
  identity; its Dirichlet-series factorization gives the displayed product.
* **X1.** R. Ghiloni and A. Perotti, *Slice regular functions on real alternative
  algebras*, Adv. Math. 226 (2011), 1662–1691,
  <https://doi.org/10.1016/j.aim.2010.08.015>.
* **X2.** G. Han, *Quaternionic slice regular functions and quaternionic Laplace
  transforms*, <https://arxiv.org/abs/2006.07511>, §2.4 equations (2.4)–(2.6),
  §3 identity principle and intrinsic decomposition. Original archived/read here.
  For O use X1's alternative-algebra framework rather than assuming associativity.
* **X3.** R. Fueter, *Die Funktionentheorie der Differentialgleichungen Delta u=0
  und Delta Delta u=0 mit vier reellen Variablen*, Comment. Math. Helv. 7
  (1934/35), 307–330. The affine-kernel obstruction above needs only the direct
  elementary computation Delta(1)=Delta(q)=0; it does not assume an inverse theorem.

New downloaded PDFs/text and fetch manifests are under `sources/think-hard/`.
Unavailable original full texts and sources only checked via a later primary
paper have been marked rather than represented as fully rederived proofs.

## 14. Bottom line

**[project; synthesis]** The best next move is to make **the ordinary component
extractable**, not just make the ambient space larger. Our strongest established
return maps are (i) no-cancellation factorization with Heilbronn order constraints,
(ii) exact subfield/representation projectors, and (iii) positivity/approximation
identities whose universal hypotheses are explicit. The missing payload remains
a new independent inequality, exclusion theorem, or convergent construction.
Nothing here establishes it. The point of this note is to distinguish a usable
bridge from an attractive place on the other side of the river.
