# Downward transfer ledger

Version: downward-20261110-v1. Research date: 10 November 2026, PST.

**[project] Purpose.** Ask not merely whether a larger setting resembles ordinary
zeta, but what exact theorem carries information back to its nontrivial zeros.
This is a map of implications, not RH progress. No new common field, new joint
arithmetic object, or global RH result is claimed.

**[known] Terminology.** Ordinary RH says every nontrivial zero of ζ(s) has real
part 1/2. It is not a claim that the whole real number line consists of zeros.
Information about ζ zeros can constrain prime-counting errors through the explicit
formula, but that is a separate, quantitative implication. “Bigger” below means
several different things: a larger coefficient algebra H or O; a number-field
extension K/Q; a higher-dimensional variety; a larger representation; a family
parameter; or an infinite-dimensional Hilbert space. None is automatically a
higher rung of one common ordering.

## The implication test

**[known] Factor test.** If F(s)=ζ(as+b)G(s) on a region, a≠0, and G is holomorphic
there, every zero ρ of ζ whose preimage lies there gives a zero (ρ−b)/a of F.
Orders add: ord F = ord ζ(as+b)+ord G. If G has a pole, zero cancellation is
possible and this conclusion needs a separate check. Hence a zero-free region for
F pulls back to one for ζ, while a total zero count for F gives an upper bound
for the inherited ζ count; an exact ζ count requires subtracting the other
factors' counts. A good asymptotic count alone is not an RH proof.

**[known] Restriction test.** If an extended function restricts exactly to ζ on
every complex slice, a theorem about its entire zero set implies the corresponding
slice theorem. The extension does not provide the theorem for free.

**[known] Limit test.** A locally uniform limit of holomorphic zero-free functions
on a connected domain is zero-free unless it is identically zero (Hurwitz).
Thus proven zero-free approximants to ξ on both open half-planes beside the
critical line, with locally uniform convergence and a nonzero limit, would prove
RH. Pointwise convergence, finitely many approximants, and a resemblance of zero
plots do not meet these hypotheses.

## Ledger A — actual factor inheritance

### A1. Galois number fields: a genuine bridge

**[known; S1–S3]** For a finite Galois extension K/Q, Artin formalism gives

    ζK(s) = ∏_(χ irreducible) L(s,χ)^(χ(1)).

The trivial character contributes ζ(s). Aramata–Brauer proves that ζK/ζ is entire,
without assuming the Artin holomorphy conjecture for individual irreducible
factors. Therefore every nontrivial ζ zero is a ζK zero, with at least the same
multiplicity. RH for ζK implies RH for ζ. A zero-free region for ζK also excludes
ζ zeros there. This is a theorem-level transfer, not a numerical observation.

**[known; S1,S2] Important scope correction.** For an arbitrary non-Galois extension,
the permutation representation still contains a trivial constituent, but the
required holomorphy of the quotient cannot simply be assumed from that algebraic
identity. General Dedekind divisibility is a conjectural issue; there are important
proved cases beyond Galois extensions. This notebook relies only on the Galois
case. Do not write “ζK/ζ is entire for every number field” as if Aramata–Brauer
established that generality.

**[known; S3] Our explicit field.** The totally real Q8 extension underlying Artin
representation `2.42025.8t5.a.a` has degree 8 and discriminant 205^6 =
74220378765625. With χD the quadratic Kronecker character of fundamental
discriminant D,

    ζK(s) = ζ(s) L(s,χ5) L(s,χ41) L(s,χ205) L(s,ρ)^2.

The degree-two representation ρ is quaternionic, self-dual, determinant 1, and its
Artin conductor is 42025. Its root number is −1. The nontrivial factors here are
entire: the quadratic factors are Dirichlet L-functions; ρ is induced from a
nontrivial degree-one character of an index-two cyclic subgroup, so its L-function
is a Hecke L-function. There is no cancellation of ζ zeros by poles in this product.
A Q8 Galois field is not “ordinary quaternion inputs”; these are different uses
of quaternion structure.

**[project] Exact arithmetic audit.** `scripts/q8_transfer.py` reconstructs the
field and all five characters with PARI, verifies SmallGroup(8,4), and checks all
2000 Dedekind Dirichlet coefficients against the multiplicative convolution of
the factors, including the ramified primes. This finite coefficient test checks
implementation; Artin formalism, not 2000 samples, supplies the all-s theorem.

**[project] Numerical zero ledger, 0<t<30.** Non-interval PARI lists give:

| Factor | Distinct positive heights found | Multiplicity in ζK | Contribution |
|---|---:|---:|---:|
| ζ | 3 | 1 | 3 |
| L(χ5) | 11 | 1 | 11 |
| L(χ41) | 21 | 1 | 21 |
| L(χ205) | 28 | 1 | 28 |
| L(ρ) | 55 | 2 | 110 |
| **Total** | **118** | | **173** |

**[project]** The first ordinary height is 14.13472514173469…; the first ρ height
is 1.26470005850918…. Local sampled circle windings give 1 at the first ζ zero
and 2 at the first ρ zero. The central ρ zero contributes a double ζK zero, while
ζ(1/2)=−1.46035450880958… is not zero. Thus **a zero of the larger function need
not be an ordinary-zeta zero**. The implication goes from small zero to larger
zero, or from a theorem excluding large zeros to an exclusion for small ones.

**[project; numerical limitations]** For the entire completed field function
s(s−1)ΛK(s), two sampled rectangular contours give winding 348 = 2×173+2. These
are meshes of 992 and 1984 vertices, with maximum phase steps about 2.804 and
1.402 radians. Stable sampled winding is NOT a certified contour enclosure.
PARI's Dedekind evaluator internally factorizes: agreement between that evaluator
and our factor product is NOT an independent analytic proof. `lfunzeros` itself
assumes simple off-center zeros and may miss zeros; we deliberately do not run it
on ζK, whose ρ zeros are double. No new completeness or interval proof is claimed
for this combined ledger. Earlier project interval certificates for ρ are
separate evidence and are not retroactively certificates for the quadratic lists.

**[project; later independent finite-height audit]** We subsequently closed the
ordinary/quadratic completeness gap at height 30 with a different analytic
evaluator: 160-bit Arb finite Hurwitz-zeta character sums. Full contour-segment
images exclude zero and their argument increments enclose windings 6,22,42,56
for ζ, χ5, χ41, χ205. Disjoint critical-line brackets of radius 10^−12 saturate
each count, proving simplicity and completeness for these four factors in the
rectangle Re(s)∈[−1/2,3/2], |Im(s)|≤30. The central values are enclosed away
from zero. `quadratic-proof.md` states the exact argument-principle proof.

**[project; explicit dependency]** Combining these new certificates with the
**freshly rerun**, complete Artin certificate through height 50 yields the
factored finite-height field count 173 positive zeros, plus central multiplicity
2. The earlier Artin file has 101 positive brackets through 50, exactly 55 below
30; the aggregation records its SHA256 and dependency scope. All 118 distinct
positive factor brackets are disjoint. The smallest cross-factor gap is between
χ5 and χ205, exceeding .000347004409. This is a certificate by factorization,
not an independent direct contour evaluation of ζK. The initial sampled field
winding above remains a sample; it is not retroactively the certified contour.
See `results/combined-certificate.json` and the four new factor JSON files.

### A2. A tensor product really does bring ζ back

**[known consequence of character theory; S2,S3]** For this Q8 representation,

    ρ ⊗ ρ* = 1 ⊕ χ5 ⊕ χ41 ⊕ χ205,
    L(s,ρ ⊗ ρ*) = ζ(s)L(s,χ5)L(s,χ41)L(s,χ205)
                = ζ_{Q(√5,√41)}(s).

This tensor L-function has a genuine ζ factor and an entire quotient. RH for it
would imply ordinary RH. But knowledge of the zeros of L(s,ρ) alone does NOT give
the tensor L zeros: tensoring acts on local Frobenius eigenvalues, not by taking
tensor products of old zeros.

**[known consequence / project exact check]** The contrast is surprisingly clean:

    Sym²ρ = χ5 ⊕ χ41 ⊕ χ205,     exterior²ρ = 1.

So the symmetric square has no trivial ζ factor in this decomposition. For every
m≥1, character values (2,−2,0,0,0) give

    ρ^(⊗m) = 2^(m−1)ρ                                  (m odd),
    ρ^(⊗m) = 2^(m−2)(1⊕χ5⊕χ41⊕χ205)                    (m even).

`q8_tensor.py` checks symmetric/exterior squares and m=1,…,12 exactly. This is
standard finite-group representation theory, newly organized and checked here,
not new number theory. Even tensor powers carry ζ; odd powers merely repeat the
old ρ factor. More dimensions do not automatically mean more information.

**[known consequence / project exact audit] A symmetric-power qualification.**
“No automatic ζ factor” is not “never.” For this finite Q8 representation,

    Sym^m ρ = ((m+1)/2)ρ                                 (m odd),
    Sym^(2k)ρ = A·1 ⊕ B·(χ5⊕χ41⊕χ205),
    A=(2k+1+3(−1)^k)/4,   B=(2k+1−(−1)^k)/4.

The degree-two determinant-one character recurrence S_m=χρ S_(m−1)−S_(m−2)
proves the formula; the script checks all m=0,…,24 exactly. In particular

    Sym^4ρ = 2·1 ⊕ χ5 ⊕ χ41 ⊕ χ205,
    L(s,Sym^4ρ)=ζ(s)^2 L(s,χ5)L(s,χ41)L(s,χ205).

Thus the fourth symmetric power **does** carry ordinary ζ zeros (doubled), and
all even symmetric powers of degree m≥4 carry them. Odd powers repeat ρ; the
symmetric square alone has no ζ factor. At height 30 the fourth-power product
has 2×3+11+21+28=66 positive zeros with multiplicity, by the factor certificates.
Its ζ quotient is holomorphic in the nontrivial strip but can have a pole at
s=1 because another ζ remains. No inference about the symmetric-power zeros is
available just from the original scalar L(s,ρ) zero list. These are known
representation-theoretic consequences, not a new functoriality theorem.

### A3. Quaternion ideal zetas: the quarter-lines are an exact RH reformulation

**[known; S4]** For a maximal order in a definite rational quaternion algebra with
finite reduced discriminant D, using the absolute **ideal-index** convention,

    ZD(s)=ζ(2s)ζ(2s−1)∏_(p|D)(1−p^(1−2s)).

Its nonreal inherited zeros are ρ/2 and (ρ+1)/2. The finite Euler-factor zeros
are on Re(s)=1/2. In the first inherited strip 0<Re(s)<1/2, the shifted ζ factor
is finite and nonzero at nonreal inherited zeros; in the second strip 1/2<Re(s)<1,
the other ζ factor is finite and nonzero. Local factors cannot cancel these zeros.
Consequently, for any such fixed order,

    RH ⇔ every nonreal zero of ZD lies on Re(s)=1/4, 1/2, or 3/4.

**[project]** Forty inherited-zero checks across the quaternion and E8 formulas
have residuals below 10^−55 at 70 decimal digits. This illustrates the theorem;
it does not establish the global right-hand condition. The quarter-lines are a
bridge back, not independent evidence for ordinary RH. The ideal-index zeta is
NOT the four-dimensional lattice Epstein zeta: their summation objects and
functional-equation normalizations differ.

### A4. E8: its centre is empty of zeros, while its factors still transfer

**[known; S5,S4]** In the standard minimum-squared-norm-2 convention,

    ZE8(s)=∑_(v≠0)|v|^(−2s)=240·2^(−s)ζ(s)ζ(s−3),
    ΛE8(s)=π^(−s)Γ(s)ZE8(s)=ΛE8(4−s).

Nonreal inherited zero strips are 0<Re(s)<1 and 3<Re(s)<4. Hence

    RH ⇔ every nonreal zero of ZE8 has Re(s)=1/2 or 7/2.

The mirror centre 2 is **not** a prediction that zeros lie there. Using the
ordinary zeta functional equation gives the exact identity, for real t,

    ΛE8(2+it) = −30 π^(−4) cosh(πt/2)
                   ·|Γ(2+it)|² ·|ζ(2+it)|² < 0.

The strict inequality follows from the Euler product at real part 2 and the fact
that Γ has no zeros. Thus the **whole centre line is unconditionally zero-free**;
indeed 1<Re(s)<3 is zero-free for ZE8. This is an elementary consequence of known
factorizations, not a newly proved general theorem or a constraint on the still
unknown ζ zeros in 0<Re(s)<1.

**[project]** `lattice_transfer.py` checks the sign identity at nine heights
0,.1,.5,1,2,5,10,20,50. Units matter:

    raw ZE8(2) = −5π²/6 = −8.22467033424113…,
    completed ΛE8(2) = −5/6.

The older page correctly names −5/6 as the completed value; conversational
shorthand should retain that qualifier. Rescaling s=4u changes the centre to
1/2, but its inherited nontrivial-zero lines to 1/8 and 7/8 under RH. Merely
renaming the exponent cannot force them onto the centre.

### A5. Higher-dimensional arithmetic geometry that literally contains ζ

**[known; S6,S7]** Projective space over the integers gives an elementary example:

    ζ_{P^n_Z}(s)=ζ(s)ζ(s−1)…ζ(s−n).

Count points of P^n over each finite field to derive the local factors, then take
the Euler product. For nonreal zeros, its shifted critical strips are disjoint,
and there are no pole cancellations. Thus a complete theorem locating its
nonreal zeros on Re(s)=j+1/2 (j=0,…,n) would imply ordinary RH, and conversely.
For P^3_Z the mirror is centred at 2, but RH predicts four lines 1/2,3/2,5/2,7/2,
not a single line 2. The larger geometry contains repeated shifted ζ information.

**[known]** In other varieties the global zeta is an alternating product/quotient
of cohomological L-functions. One must examine zeros and poles, not just assert
“it contains H0, therefore every ζ zero survives.” A quotient can cancel zeros.
For example an elliptic-curve scheme zeta away from bad primes has the shape
ζ(s)ζ(s−1)/L(E,s). Its zero inheritance requires a cancellation audit.

## Ledger B — solved RH analogues are structural lessons, not factor transfers

### B1. Curves over finite fields: Weil's theorem

**[known; S7,S8]** For a smooth projective geometrically connected curve C/Fq of
genus g,

    Z(C,T)=P1(T)/((1−T)(1−qT)),
    P1(T)=∏_(j=1)^(2g)(1−αj T),     |αj|=√q.

So its zeros are at |T|=q^(−1/2); with T=q^(−s) these are periodic copies on
Re(s)=1/2. Hasse established the elliptic case; Weil the general curve case.
In a geometric proof, the intersection inequality on C×C (Hodge-index-type
negativity after removing fibre classes), applied to Frobenius correspondences,
provides the decisive estimate. The important ingredient is not extra spatial
directions alone: it is a real operator/correspondence and a proved inequality.
There is no automatic map carrying this finite-field theorem to ζ over Q.

**[project] Exact example.** For E: y²=x³+x+1 over F5, #E(F5)=9 and a5=−3,

    P1(T)=1+3T+5T²,
    T=(−3±i√11)/10,       |T|=1/√5=.447213595499958… .

Direct PARI extension-field point counts for degrees 1,…,8 agree with
#E(F_(5^r))=5^r+1−α^r−β^r. First counts: 9,27,108,675. These are exact finite
field counts, not interval approximations to global zeta zeros.

### B2. Deligne: all dimensions, but separated by cohomological weight

**[known; S8]** For smooth projective X/Fq,

    Z(X,T)=∏_i det(1−T Frobenius | H^i_et(Xbar,Q_l))^((-1)^(i+1)),
    |α|=q^(i/2) for Frobenius eigenvalues on H^i.

Deligne's 1974 theorem is a theorem in every dimension. It does **not** put all
zeros on one common critical circle: each cohomological degree has its own
weight. For general nonproper or singular varieties the statements involve
mixed weights; this notebook's clean equality assumes smooth and projective.

**[known; S8] Proof ingredient precision.** Grothendieck's cohomology and trace
formula identify the correct operator and finite-dimensional spaces. Deligne's
proof of purity uses geometric estimates, Lefschetz pencils and monodromy; it is
not justified to compress the entire higher-dimensional proof into “a positive
Hilbert-space inner product was constructed.” The general standard-conjecture
positivity story has its own unresolved aspects.

**[project] A surface checked, not merely drawn.** For X=E×E over F5,

    Z(X,T) = (1+3T+5T²)²(1+15T+125T²)² /
             [(1−T)(1+T+25T²)(1−5T)^4(1−25T)].

Its zero circles are |T|=5^(−1/2) and 5^(−3/2), each root with multiplicity 2.
The coefficient of T^r in T d/dT log Z equals #E(F_(5^r))² for r=1,…,8, checked
exactly. The exponent **4** on (1−5T) is necessary: H² has dimension 6. This
computation catches a tempting but incorrect exponent 2.

### B3. Selberg: an honest spectral model, with exceptions

**[known; S9]** For a compact hyperbolic surface, the spectral zeros satisfy
s(1−s)=λ for Laplace eigenvalues λ≥0. Self-adjointness forces λ real. Eigenvalues
λ≥1/4 produce Re(s)=1/2; 0<λ<1/4 produce **real exceptional zeros** in (0,1).
Such surfaces exist. There are also the known topological/trivial zeros and
end-point behaviour. Therefore “RH is a theorem for compact Selberg zeta, with
all nontrivial zeros on the line” is too strong without the real exception.

**[known / speculative distinction]** Selberg's trace formula and self-adjoint
Laplacian are actual objects with an actual spectral identity. For ordinary ζ,
a Hilbert–Pólya construction would need a self-adjoint operator whose determinant
or trace identity recovers **all** ξ zeros with correct multiplicities and local
prime terms. Choosing an operator with already-known real zero ordinates as its
spectrum is circular and says nothing about hypothetical off-line zeros.

### B4. Hasse–Weil over number fields is not already settled

**[known; S6,S10]** Local factors at good primes of a smooth projective variety
satisfy finite-field weight theorems. Their global Euler product over primes is
a different analytic object. Its analytic continuation/functional equation are
known in many cases but not universally, and its global RH is generally open.
Elliptic L-functions in arithmetic normalization have centre 1 (unitary
normalization shifts it to 1/2). Local eigenvalues of absolute value √p do not
by themselves prove the global zeros lie at Re(s)=1.

**[speculative programs, with known component results; S10–S12]** Deninger's
arithmetic cohomology/foliation program seeks a flow generator θ and a positive
Hodge-type pairing forcing θ−w/2 to be skew-adjoint in the relevant degree.
Connes and **Consani**, not Deninger, developed the arithmetic/scaling sites and
Frobenius correspondences; Connes's adèle-class-space trace framework and their
semi-local positivity work seek the needed explicit-formula pairing. The spaces,
correspondences and several local theorems are established; the full positivity
or spectral identification sufficient for ordinary RH is not supplied by these
papers. A proposed bridge is not a completed transfer.

## Ledger C — higher rank, products and statistics

**[known; S13,S14]** For a cuspidal unitary automorphic representation π of GL(n),
the standard L-function has its own nontrivial zeros in unitary normalization.
Its GRH is generally open. Larger n does not imply containment of the GL(1)
trivial L-function ζ. Functorial symmetric powers, when established, identify
Euler products; they do not automatically identify or include old zero sets.

**[known, with a gate; S13]** Locally the trivial representation inside End(π)
suggests

    L(s,π×π*) = ζ(s)L(s,Ad π).

The Rankin–Selberg product has a pole at 1. A formal quotient or local identity
alone does not guarantee ordinary ζ zeros survive globally: the adjoint quotient
must be holomorphic in the region of interest. One must use an actual holomorphy
theorem in the relevant class, not assert the general case. For GL(2), Gelbart–Jacquet establishes the needed adjoint holomorphy; Yang
establishes the complete adjoint result for GL(3) and GL(4), with finite-part
holomorphy in Corollary 3 of the cited text. Over Q these supply genuine
Rankin–Selberg-to-ζ transfers for the stated unitary cuspidal representations: a
zero-free theorem (or GRH) for the whole Rankin–Selberg function implies the
corresponding assertion for ζ. The general-rank statement still requires its
holomorphy gate. The Q8 tensor identity
in A2 is an explicit safe instance with an entire product of quadratic factors.
The notation π×π* means the contragredient representation, not |L(s,π)|² as a
holomorphic function of s. Absolute-square identities on a line are a separate
calculation.

**[known / conjectural distinction; S14,S15]** Montgomery proved a restricted
pair-correlation theorem (in the usual ordinate formulation assuming RH).
Rudnick–Sarnak established n-level correlation results for restricted Fourier
support; versions involving complex parameters avoid assuming RH, while their
usual real-ordinate statistical interpretations use RH/GRH and appropriate
coefficient hypotheses. This does not prove unrestricted GUE. Katz–Sarnak relate
family symmetry types U, symplectic and orthogonal to low-lying-zero statistics;
finite-field equidistribution results are theorems in their stated limits, while
many number-field family predictions remain conjectural or have support-limited
proved cases. Bulk correlations and low-lying family symmetry are different
statistics.

**[known; S15] Symmetry is not assigned by the word quaternion.** A quaternionic
representation type, quaternion-valued inputs and the Gaussian symplectic random
matrix ensemble are different notions. An ensemble prediction needs a specified
family and its arithmetic symmetry; it does not follow from the coefficient
algebra of one function. **[project]** For our Q8 character, the Frobenius–Schur
indicator is (2+2+6×(−2))/8=−1, certifying quaternionic representation type. That
calculation alone supplies no GSE spacing theorem or ordinary-ζ restriction.

**[project] The 500-zero illustration.** mpmath at 25 decimal digits gives heights
14.134725… through 811.184359…. Unfolding by the smooth Riemann–von Mangoldt
count gives mean adjacent gap .99969336 and minimum .23632537. We use all positive
pair differences below 4, exclude self-pairs, and apply a translation edge
correction. There are 1725 such pairs. The 20-bin RMS discrepancy from the GUE
bin averages is .0934906; this is descriptive, **not** a significance test or a
new theorem. First low zeros are not asymptotic data.

**[project] Completed 5000-zero comparison.** The extension took 1714.307 seconds
of computation and ends at height 5447.86199830130. There are 17492 positive
pairs below unfolded separation 4; mean adjacent gap 1.0000161940, minimum
.0462896374, maximum 2.4679062149. The same estimator gives 20-bin RMS discrepancy
.0380867018. Both curves are shown, with the same warning: a smaller discrepancy
is not a significance test, asymptotic theorem, or RH evidence about every zero.
Machine-readable output: `results/pair-correlation-5000.json`.

**[known] Why statistics do not prove RH.** A fixed finite exceptional zero set
contributes a vanishing fraction to normalized bulk statistics. Consequently a
limiting bulk law can be right while finitely many exceptions remain. More
subtle density-zero alterations can also evade suitable averages; no universal
claim about all such alterations is needed here. A universal zero-by-zero
statement requires something stronger than a limiting averaged histogram.

## Ledger D — audit of our six hypercomplex differences

| Thread | What is actually different | What reaches ordinary ζ? |
|---|---|---|
| **[known / project] Direction-dependent zeros** | Non-real coefficients select directions; a complex-character L zero can lift to an isolated point rather than a whole sphere. | Nothing automatic: it is a different L-function. A character-product/cyclotomic factorization can provide a separate bridge, not the directional effect itself. |
| **[known] Forced central zeros** | A self-dual completed Artin L-function with sign −1 vanishes at 1/2. | It forces a field-zeta zero through its squared factor, but does not force ζ(1/2)=0; ζ(1/2) is nonzero. |
| **[known / project] Bracket-moved octonion star zeros** | Nonassociative slice/star products can relocate a zero on a fixed real-part/radius sphere. Ordinary pointwise products obey multiplicativity of the norm. | No ζ factor identity was produced; no transfer. These examples are toy polynomials, not new Euler products or L-functions. |
| **[known] Quarter-lines** | Ideal-index zetas contain ζ(2s) and ζ(2s−1). | Yes: the nonreal zero-line statement is equivalent to ordinary RH, by the no-cancellation proof in A3. |
| **[project] Holes at infinity** | Unbounded round zero spheres create complement loops arbitrarily far out; this is not a locally horned zero sphere. | No real-part inequality follows. The same topological pattern accommodates bounded real-coordinate displacements; the required arithmetic position is missing. No general theorem that topology can never help is claimed. |
| **[known] Centre n/4** | Epstein zeta with |v|^(−2s) pairs s with n/2−s (self-dual lattice; otherwise it pairs with the dual lattice). | Normalization alone gives no zero location. E8 does transfer through its explicit ζ product, not through n/4; the centre line 2 is actually zero-free. |

**[known] Intrinsic ζ lifts.** For ordinary ζ's intrinsic quaternion/octonion
slice extension, a zero a+ib yields the full (d−2)-sphere a+bS^(d−2). Restriction
to a complex slice recovers ζ exactly. “Every nontrivial zero sphere is centred
at real coordinate 1/2” is thus equivalent to ordinary RH. The tube-degree count
2m for a spherical zero is a multiplicity count and does not encode a. A new
useful theorem about a would transfer; the known roundness alone does not supply
it. This is a boundary of the **present inference**, not a reason to close the
whole research program.

## Ledger E — exact reformulations in larger spaces

### E1. Li coefficients: all of them, not a finite attractive list

**[known; S16]** For ξ(s)=s(s−1)π^(−s/2)Γ(s/2)ζ(s)/2, define

    λn = (1/(n−1)!) [d^n/ds^n (s^(n−1)log ξ(s))]_(s=1).

RH is equivalent to λn≥0 for **every** n≥1. The equivalent zero sum is
Σρ[1−(1−1/ρ)^n], interpreted in the standard symmetric limiting order. A
finite zero truncation is not an exact Li coefficient unless its tail is controlled.

**[project]** From 90-digit PARI expansions at s=1, the first twenty generalized
Li coefficients satisfy exact-product additivity to numerical error below 10^−65:

    λn(K)=λn(ζ)+λn(χ5)+λn(χ41)+λn(χ205)+2λn(ρ).

At n=1, λ1(ζ)=.0230957089661210… and λ1(K)=8.07724416785970… . Total positivity
for one n does not let us subtract unknown positive contributions and conclude
the ζ term is positive. Positivity for **all** field coefficients can instead
imply field RH via the appropriate generalized criterion, then transfer by A1.
Our twenty numerical coefficients prove neither premise.

### E2. Weil's explicit-formula positivity

**[known; S11,S17]** The Riemann–Weil explicit formula equates a sum over all ζ
zeros to prime/archimedean terms for admissible test functions. With the proper
involution and normalizations, positivity on every convolution square is
**equivalent** to RH. Under RH the zero side takes an absolute-square form.
Without RH one must retain the conjugate-reflected arguments; replacing them by
absolute squares silently assumes the conclusion. Compact support means each
test invokes finitely many prime terms, but the criterion quantifies over all
tests/support scales. A finite positive matrix compression is useful evidence
and a diagnostic, not automatically positivity of the infinite form.

**[speculative project direction]** Seek a quaternionic/octonionic operator or
pairing whose restriction is exactly the Weil pairing and whose positivity is
proved independently. That would be a real downward bridge. A norm |f|²≥0
without the required explicit-formula identity is not that bridge. The proof
obligation is the equality of forms plus domain/density control, not a picture.

**[known; S11, exact convention]** With G(s)=∫_0^∞g(x)x^(s−1)dx,
g smooth compactly supported on (0,∞) and G(0)=G(1)=0, an equivalent
criterion is

    Σ_ρ G(ρ) overline(G(1−overline(ρ))) ≥ 0   for every such g.

Indeed the multiplicative convolution square g * overline(g)^sharp, with
h^sharp(x)=x^−1 h(1/x), has precisely this Mellin transform. Its pole terms
vanish by the two moment conditions. On RH the summands become |G(ρ)|².
The smooth compact-support class gives sufficiently rapid vertical decay for
the sum. Source S11 equation (2) states the equivalent inequality on the prime
side with the opposite sign. This convention avoids calling the unreflected
product an absolute square before RH has been proved.

**[known; S22] Li and Weil are not unrelated checklists.** Put
G_n(s)=1−(1−1/s)^n. Lagarias's Theorem 3.1 gives

    <G_n,G_m>_W = λn+λ_(−m)−λ_(n−m).

For ordinary self-dual ζ this is λn+λm−λ_|n−m|; in particular the diagonal is
2λn. This is an exact return mechanism between two reformulations, not a
statistical analogy. The rational G_n test class has endpoint poles and is
not literally the compact-support Mellin class above; the theorem justifies
its own enlarged admissible class.

**[project] A rigorous finite compression.** `li_weil_compression.py` computes
50 ordinary Li coefficients with 2048-bit Arb deflated-zeta Taylor series at
s=1. All 50 are strictly positive. It then forms the full 32×32 matrix above
and interval-certifies every LDL pivot as positive: the quadratic form is
positive on the span of these 32 test functions, not merely on 32 individual
diagonals. The last pivot is about **1.40803030073×10^−110**. A plain double-precision
LDL calculation instead reports its first nonpositive pivot at index 7
(about −5.51×10^−14): that is rounding/cancellation, not a certified negative
Weil direction. Results and full enclosures are in `li-weil-compression.json`.
A finite positive compression still does not give the full all-test theorem.
No improvement to known RH criteria or verified ranges is claimed.

### E3. Nyman–Beurling and Báez–Duarte

**[known; S18]** In H=L²(0,∞), let ρa(x)={1/(a x)} and χ=1_(0,1]. Báez–Duarte
proved RH iff χ is in the closed span of ρa with a a positive integer. Equivalently
one can use the classical tail-cancelling span in L²(0,1). This is an exact
transfer from an infinite-dimensional approximation problem to ordinary zeros.
Finite-dimensional numerical improvement alone is not density.

**[known consequence; derivation included]** In the tail-cancelling formulation,
b_a(x)={1/(a x)}−a^−1{1/x} for integer a≥2. For 0<Re(s)<1,

    ∫_0^1 b_a(x)x^(s−1) dx = −ζ(s)(a^−s−a^−1)/s.

At any hypothetical zero ρ=β+it with β>1/2, this Mellin functional annihilates
all finite combinations of b_a. Applying Cauchy–Schwarz to the residual r=1−Σc_a b_a
therefore gives the explicit obstruction

    ||r||²_L²(0,1) ≥ (2β−1)/|ρ|².

Thus rigorous approximants of norm at most ε constrain every off-line zero by
2β−1 ≤ ε²|ρ|². A sequence with ε→0 excludes every fixed off-line zero and, with
the functional equation, proves RH. One fixed ε cannot exclude arbitrarily high
zeros. The b_a subspace lies inside the classical Nyman–Beurling subspace;
we do not need to claim a converse for this particular integer tail-cancelled
span to justify the displayed one-way obstruction.

**[project] A finite interval experiment.** `nyman_transfer.py` computes the
infinite Gram integrals with 192-bit Arb enclosures, not by forgetting their
tails. Put m=floor(1/x). Then b_a(x)=(m mod a)/a on
(1/(m+1),1/m]. With L=lcm(a,b), periodicity gives

    <b_a,b_b> = (1/L) Σ_(r=1)^(L−1) [(r mod a)/a][(r mod b)/b]
                      ·[ψ((r+1)/L)−ψ(r/L)],
    <1,b_a> = log(a)/a,

where ψ is the digamma function. The first identity is obtained by summing
1/(r+kL)−1/(r+1+kL) over k≥0; Arb encloses ψ at positive rational arguments.
We solve the finite Gram system to choose decimal coefficients, then re-evaluate
the squared error for those **fixed exact rational-decimal coefficients** in
interval arithmetic. We need not claim those rounded coefficients are optimal.

| Largest generator a | Squared L² error (rounded) |
|---:|---:|
| 2 | .306852819440055 = 1−log 2 |
| 3 | .0957477839277540 |
| 5 | .0363190179381870 |
| 10 | .0238525533188830 |
| 20 | .0165382521093925 |
| 30 | .0145567335769970 |

**[project]** Full enclosures, rational upper bounds and chosen coefficients are
in `results/nyman-transfer.json`. The N=2 case independently reduces to
b_2=0 or 1/2 on alternating cells, optimal coefficient 2 and error 1−log 2;
unit tests check it. These modest finite errors yield **no improvement** to known
zero-free regions or RH verification. The experiment illustrates an actual
all-zero inequality, while leaving the convergence-to-zero requirement open.

### E4. Heat flow: a genuine one-parameter enlargement

**[known; S19]** In the Rodgers–Tao convention,

    H_t(z)=∫_0^∞ e^(t u²) Φ(u)cos(zu)du,
    H_0(z)=ξ(1/2+iz/2)/8,       ∂t H_t = −∂zz H_t.

All zeros are real exactly when t≥Λ, for a finite de Bruijn–Newman constant Λ.
Ordinary RH is equivalent to Λ≤0. Rodgers–Tao proved Λ≥0, so RH is now equivalent
to Λ=0. Real-rootedness at one positive t does not run backwards to t=0.
Proving real-rootedness for a sequence t_j↓0, together with locally uniform
convergence, would transfer by Hurwitz. A finite time sweep does not.

**[known]** Close zero pairs and zero dynamics already produced rigorous
information about this larger family. That makes it a real example of a bigger
setting returning information—not a proof of the remaining inequality.

### E5. Jensen polynomials: a model for a successful bridge and a quantifier trap

**[known; S20]** For the Taylor coefficient sequence γ(n) specified in
Griffin–Ono–Rolen–Zagier, define J^(d,n)(X)=Σ_(j=0)^d binom(d,j)γ(n+j)X^j.
RH is equivalent to hyperbolicity (all roots real) for every degree/shift; it
suffices to establish all degrees at shift 0. Their 2019 theorem establishes,
for each fixed degree d, hyperbolicity for sufficiently large shifts n. The
threshold depends on d. It is not the missing uniform all-degree theorem.
Their Hermite limits are rigorous, but interchanging “for every d eventually n”
with “all d,n” would be a logical error.

## Research gates kept open

**[project] Gate 1 — now closed at the stated finite height, with a dependency.**
New ordinary/quadratic interval contours and sign brackets, together with the
freshly rerun Artin certificate, give the factored count through height 30. The
Artin source and analytic bounds were also reviewed. A genuinely independent
implementation or direct field contour is a separate possible audit; repeating
the same algorithm does not constitute independence. Extending beyond the
certified heights is another finite task. Neither closes the global problem.

**[project] Gate 2 — an actual new position-sensitive pairing.** Specify a
quaternionic or octonionic construction and show its restriction equals a known
ζ explicit-formula, Li, or Nyman–Beurling object. Record where noncommutativity
or nonassociativity genuinely survives this restriction. If it disappears,
report that fact rather than conclude the whole field is exhausted.

**[project] Gate 3 — transfer without circular input.** Any new claimed spectrum
must capture all ξ zeros without inserting the observed ordinates as parameters.
Any proposed positive form needs all admissible tests, not only finite samples.
Any limiting construction needs locally uniform convergence and a nonzero limit.

**[project] Gate 4 — source and numerics hygiene.** Preserve exact conventions:
raw versus completed E8, lattice versus ideal-index quaternion zeta, local versus
global Hasse–Weil, and Galois versus arbitrary Dedekind quotients. Higher-dimensional
facts are worth gathering; these distinctions make them usable rather than less
interesting.

## Primary and authoritative sources

**[known-source records]** Original papers where accessible; manuals/books are
identified as such rather than passed off as original proofs. PDFs and extracted
text are archived in `sources/` when retrieval succeeded. A download failure is
not evidence a source does not exist.

- **S1.** H. Aramata, *Über die Teilbarkeit der Dedekindschen Zetafunktionen*,
  Proc. Imp. Acad. 9 (1933), 31–34. Original theorem:
  https://doi.org/10.3792/pia/1195580866 ; J-STAGE record:
  https://www.jstage.jst.go.jp/article/pjab1912/9/2/9_2_31/_article .
- **S2.** E. Artin, *Über eine neue Art von L-Reihen*, Abh. Math. Sem. Hamburg
  3 (1924), 89–108; R. Brauer, *On Artin's L-series with general group characters*,
  Ann. Math. 48 (1947), 502–514, https://doi.org/10.2307/1969161 . Standard
  induction/direct-sum formalism; the primary zero-inclusion theorem used is S1.
- **S3.** Magma handbook, example H139E4, *Built-in L-series* (authoritative
  reproducible example, not original theorem):
  https://magma.maths.usyd.edu.au/magma/handbook/text/1641 . Database parameters:
  https://www.lmfdb.org/ArtinRepresentation/2.42025.8t5.a.a . PARI manual
  cautions and Artin constructors:
  https://pari.math.u-bordeaux.fr/dochtml/html-stable/_L_minusfunctions.html .
- **S4.** J. Voight, *Quaternion Algebras*, §26, Theorem 26.4.9 and
  equations 26.8.12–14 (textbook): https://jvoight.github.io/quat-book.pdf .
- **S5.** M. Viazovska, *The sphere packing problem in dimension 8*,
  https://arxiv.org/abs/1603.04246 , for E8 and its E4 theta series. The displayed
  zeta product follows by Mellin transform and Σσ3(n)n^−s=ζ(s)ζ(s−3).
- **S6.** C. Deninger, *The Hilbert–Polya strategy and height pairings*,
  https://arxiv.org/abs/1001.1621 . Explicitly conjectural arithmetic cohomology;
  also distinguishes global scheme zetas from individual motivic L-functions.
- **S7.** A. Weil, *Numbers of solutions of equations in finite fields*,
  Bull. AMS 55 (1949), 497–508, https://doi.org/10.1090/S0002-9904-1949-09219-4 ;
  *Sur les courbes algébriques et les variétés qui s'en déduisent*, Hermann, 1948.
- **S8.** P. Deligne, *La conjecture de Weil I*, Publ. IHÉS 43 (1974),
  273–307, especially (1.5.4), Theorem (1.6):
  https://numdam.org/item/10.1007/BF02684373.pdf .
- **S9.** A. Selberg, *Harmonic analysis and discontinuous groups in weakly
  symmetric Riemannian spaces with applications to Dirichlet series*, J. Indian
  Math. Soc. 20 (1956), 47–87. For an original-paper statement including the real
  exception: B. Randol, *The Riemann hypothesis for Selberg's zeta-function and the
  asymptotic behavior of eigenvalues of the Laplace operator*, Trans. AMS 236
  (1978), https://doi.org/10.1090/S0002-9947-1978-0472728-1 .
- **S10.** C. Deninger, *Analysis on foliated spaces and arithmetic geometry*,
  in LMS Lecture Notes 354 (2008); the explicit operator/positivity argument is
  also in S6. This is not the Connes–Consani arithmetic site.
- **S11.** A. Connes and C. Consani, *Weil positivity and Trace formula: the
  archimedean place*, https://arxiv.org/abs/2006.13771 , Introduction (1)–(2),
  Theorem 1 and its restricted-support hypotheses. Local result, not full RH.
- **S12.** A. Connes and C. Consani, *Geometry of the Arithmetic Site*,
  https://alainconnes.org/wp-content/uploads/arithmeticsite.pdf , Theorem 1.1.
- **S13.** F. Shahidi, *On certain L-functions*, Amer. J. Math. 103 (1981),
  297–355; H. Jacquet, I. Piatetski-Shapiro and J. Shalika, *Rankin–Selberg
  convolutions*, Amer. J. Math. 105 (1983), 367–464. For a precise warning about
  adjoint holomorphy rather than treating it as automatic:
  L. Yang, *Holomorphy of adjoint L-functions for GL(n): n≤4*,
  https://doi.org/10.1007/s00208-021-02189-4 (published online 28 April 2021),
  and https://arxiv.org/abs/1903.09881 , Theorem A and Corollary 3. GL(2) primary:
  S. Gelbart and H. Jacquet, Ann. ENS 11 (1978), 471–542,
  https://doi.org/10.24033/asens.1355 . We do not assert general-rank holomorphy.
- **S14.** Z. Rudnick and P. Sarnak, *The n-level correlations of zeros of the
  zeta function*, C. R. Acad. Sci. Paris 319 (1994), 1027–1032,
  https://www.math.tau.ac.il/~rudnick/papers/RudnickSarnakCRAS1994.pdf ; full
  paper *Zeros of principal L-functions and random matrix theory*, Duke Math.
  J. 85 (1996), 269–322, https://doi.org/10.1215/S0012-7094-96-08115-6 .
- **S15.** N. Katz and P. Sarnak, *Zeroes of zeta functions and symmetry*,
  Bull. AMS 36 (1999), 1–26 (authors' survey):
  https://web.math.princeton.edu/~nmk/old/katzsarnakarticle.pdf . H. Montgomery,
  *The pair correlation of zeros of the zeta function*, Proc. Symp. Pure Math.
  24 (1973), 181–193 is the primary pair-correlation source.
- **S16.** 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 . Publisher full-text retrieval timed
  out this turn; the criterion is also explicitly cited in Conrey's survey
  https://www.ams.org/notices/200303/fea-conrey-web.pdf .
- **S17.** A. Weil, *Sur les “formules explicites” de la théorie des nombres
  premiers*, Comm. Sém. Math. Lund, suppl. (1952), 252–265. A modern precise
  formulation with involution and test-class conditions is S11.
- **S18.** L. Báez-Duarte, *A strengthening of the Nyman–Beurling criterion for
  the Riemann hypothesis*, 2003; preprint https://arxiv.org/abs/math/0202141 ,
  Theorem 1.1 and Mellin identity (2.6)–(2.7). Original Beurling theorem:
  *A closure problem related to the Riemann zeta-function*, PNAS 41 (1955),
  312–314.
- **S19.** B. Rodgers and T. Tao, *The de Bruijn–Newman constant is non-negative*,
  Forum Math. Pi 8 (2020), e6, https://arxiv.org/abs/1801.05914 , equations
  (1)–(4) and Theorem 1. This fixes the heat-flow normalization used here.
- **S20.** M. Griffin, K. Ono, L. Rolen and D. Zagier, *Jensen polynomials for
  the Riemann zeta function and other sequences*, PNAS 116 (2019), 11103–11110,
  https://arxiv.org/abs/1902.07321 , equations (1)–(2), Theorems 1–3.

- **S21.** R. Ghiloni and A. Perotti, *Zeros of regular functions of quaternionic
  and octonionic variable*, https://arxiv.org/abs/0904.2667 ; the earlier
  hypercomplex page contains the exact slice-product examples and their separate
  numerical audits. Its raw toys are not claimed to be arithmetic L-functions.

- **S22.** J. C. Lagarias, *Li coefficients for automorphic L-functions*,
  Ann. Inst. Fourier 57 (2007), 1689–1740,
  https://doi.org/10.5802/aif.2311 ; author text
  https://arxiv.org/abs/math/0404394 , Introduction and Theorems 2.2, 3.1.
  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 .

## Additional dependency audit

**[project]** All 5000 reused Artin coefficients were independently reconstructed
from the field in exact PARI arithmetic, and the full field-factor coefficient
convolution matches through 5000. The root number −1 and conductor 42025 were
also reconstructed. The earlier contour output/source/inputs are frozen under
`sources/artin-dependency/`, with hashes. `artin-proof-audit.md` records a
complete source read and the tail/quadrature/homotopy checks. The earlier Artin
algorithm was subsequently rerun completely in 221.824 seconds; every non-timing
field agrees with the pinned earlier output. This is reproduction of the same
algorithm, not an independent implementation. See `results/artin-rerun-audit.json`.

## A shifted family whose positivity already transfers: the line Re(s)=1

**[known; S23]** There is an instructive success story which is neither a new
spatial dimension nor a conjectural geometric construction. For σ>1 and real t,

    ζ(σ)^3 |ζ(σ+it)|^4 |ζ(σ+2it)| ≥ 1.

**[known; self-contained proof]** Absolute Euler-product convergence gives

    log(product) = Σ_p Σ_(k≥1) p^(−kσ)/k
                   · [3+4cos(kt log p)+cos(2kt log p)].

The bracket is exactly **2(1+cos θ)^2≥0**, so the whole sum is nonnegative.
The shifted family contains multiple samples of the *same* ζ, and a precise
positive identity ties them together. Positivity here is not an unrelated norm.

**[known; self-contained proof]** Suppose ζ had a zero of order m≥1 at 1+it,
t≠0. As σ decreases to 1 from the right, ζ(σ)^3 has pole order 3, the fourth
power of |ζ(σ+it)| vanishes to order 4m, and |ζ(σ+2it)| remains bounded because
1+2it is not the pole at 1. The product would tend to zero like
O((σ−1)^(4m−3)), contradicting its lower bound 1. Thus ζ has **no zeros on
Re(s)=1**. This is an actual, completed downward restriction on ordinary zeros.

**[known] Why not the critical line?** The positive logarithmic expansion was
justified in σ>1; moving σ toward a hypothetical β<1 leaves that convergence
region. The pole/zero order contradiction at the boundary does not magically
supply a corresponding inequality inside the strip. The exact missing extension
is visible rather than hidden behind the word 'positivity'.

**[project]** `scripts/shifted_positivity.py` checks 20 shifted products with
192-bit intervals, including σ=1.001. The proof is the identity and limiting
argument above, not those finite examples. `results/shifted-positivity.json`
records the enclosures. This classical mechanism is included as a useful design
pattern, not a new result.

- **S23.** J. Hadamard, *Sur la distribution des zéros de la fonction ζ(s) et ses
  conséquences arithmétiques*, Bull. Soc. Math. France 24 (1896), 199–220,
  https://www.numdam.org/item/BSMF_1896__24__199_1/ (original zero-free-boundary
  theorem). The elementary trigonometric presentation is associated with
  F. Mertens, *Über eine Eigenschaft der Riemann’schen ζ-Function*, Sitzungsber.
  Akad. Wiss. Wien, math.-naturw. Kl. 107 (1898), 1429–1434. Bibliography checked
  at https://dml.cz/handle/10338.dmlcz/136881 ; the Mertens original scan was not
  retrieved. The argument used here is fully written out above.

**[known; S23, Hadamard §5] Structural lesson.** The argument works more generally
for a function whose logarithm has a convergent Dirichlet series with nonnegative
coefficients, and which continues across the boundary except for one simple pole.
Dedekind ζ_K has that structure. Its proven boundary zero-freeness can therefore
transfer through the entire quotient in the Galois case. Positive Euler-log
coefficients suffice here for the *boundary*, not for the whole critical strip.
This is a different positivity from geometric intersection theory or the full
Weil test-function pairing; the connecting identity must still be named.

## The last return: from zero locations to the ordinary number line

**[known; S24]** Define ψ(x)=Σ_(p^k≤x) log p (the prime-power counting function,
not the digamma ψ used in the Gram formulas). Ordinary RH implies

    ψ(x) = x + O(sqrt(x) log² x),
    π(x) = Li(x) + O(sqrt(x) log x).

The constants are uniform as real x→∞. These are statements about the whole
ordinary positive number line, not merely the finitely many zeros in a plot.
For a Galois field, **RH_K ⇒ RH_ζ ⇒ these prime-error bounds** is a rigorous
chain. Proving the first premise independently remains the hard part.

**[known; self-contained converse for ψ]** Conversely, the stated ψ bound implies
RH. For Re(s)>1,

    −ζ'(s)/ζ(s) = s/(s−1) + s∫_1^∞ (ψ(x)−x)x^(−s−1) dx.

The error bound makes the integral holomorphic for Re(s)>1/2. Thus the logarithmic
derivative has no poles there apart from the known pole contribution at 1;
ζ has no zeros there. Functional-equation symmetry excludes nontrivial zeros on the other
side of the halfway line as well. This exhibits the exact analytic bridge from
a uniform number-line estimate back to *every* zero.

**[project] Scope of our finite evidence.** Certification through height 30 is
not the all-height premise in that implication. Turning finite zero information
into a finite-range prime estimate requires explicit-formula truncation and
rigorous tail/error bounds; we have not claimed or computed such an estimate
here. More certified data are genuine data, not a substitute for that analysis.

- **S24.** H. von Koch, *Sur la distribution des nombres premiers*, Acta Math.
  24 (1901), 159–182, https://doi.org/10.1007/BF02403071 ; original paper archived
  at https://zenodo.org/records/2347595 . The additional note on p.182 was read:
  it explicitly gives the RH-conditional √x log x prime-counting error. The
  ψ converse above is written out rather than attributed to that page.
