# Global ideal zetas of G, S₁, R and E₈

Research and new figure, 2026-11-11 10:38 AM PST. Grant's figure request t191u. No messages sent.

## Plain explanation for Grant

**[project, proved from the established local factors]** The whole ideal-counting function has two parts: the ordinary zeta function with its input multiplied by eight, and a finite correction recording the order's special behavior at 2. Multiplying the input by eight means a zero ρ becomes a zero at ρ/8: **both coordinates shrink by eight**. A checked ordinary zero ½+14.134725…i therefore becomes 1/16+1.766840…i. It is the same inherited zero, not a new discovery.

**Important qualification [known]:** all four functions have infinitely many inherited zeros on Re s=1/16, because ordinary zeta has infinitely many on ½. But saying **every** inherited nontrivial zero lies on 1/16 is equivalent to RH, not a result here. The 102 inherited zeros displayed through height 30 are taken from the existing finite project certificates and need no RH assumption.

**The multiplication changes the picture.** All apparent poles from the prime-2 denominators cancel. G, S₁ and R acquire an additional column of zeros on Re s=0 from 1+2^(-4s). Their prime-2 numerator zeros survive too. E₈ reduces to ζ(8s) exactly. Only G has extra zeros on Re s=½. In the ordinary zeta completion, E₈ has a mirror centered at 1/16; the other three functions do not share that mirror. G's local half-line symmetry does not become a symmetry of its whole global zeta.

## 1. Definition and Euler product

We retain the exact ideal notion and additive-index convention from `octonion-odd-primes-20261111.md`: a right ideal is a finite-index additive lattice L⊆S satisfying L·S⊆L; not an arbitrary set xS. Define

    Z_S(s) = Σ_L [S:L]^(-s).

An integral lattice is determined by its local lattices, almost all equal to S⊗Zₚ. The finite quotient and its stability under the finitely many integral right-multiplication matrices decompose prime by prime (Chinese remainder theorem). Conversely compatible local lattices glue to a stable global lattice. Indices multiply, so the Euler product is valid in its convergence region. This argument concerns linear stability and does not assume associative octonion multiplication.

All odd factors are 1/(1−p^(-8s)) [known, Petersson 2021, Thm. 4.1(i), and independently certified in the preceding note]. Therefore, initially for Re s>1/8,

    Z_S(s) = ζ(8s) (1−2^(-8s)) F_S,2(s).

The resulting identity provides meromorphic continuation to C. Let x=2^(-s), Q_G=1+x+7x²+7x³+14x⁴+14x⁵+28x⁶+8x⁷+16x⁸, and Q_S1=1+2x+15x²+35x³+14x⁴+14x⁵. The established F₂ factors give **exactly**:

| kind | F₂ | global function |
|---|---|---|
| G | Q_G/(1−x⁴) | **ζ(8s) Q_G(x)(1+x⁴)** |
| S₁ | Q_S1/(1−x⁴) | **ζ(8s) Q_S1(x)(1+x⁴)** |
| R (each of 7) | (1+3x²)/(1−x⁴) | **ζ(8s)(1+3x²)(1+x⁴)** |
| E₈ (each of 7) | 1/(1−x⁸) | **ζ(8s)** |

Label: [project, proved], using the previously computer-assisted 2-local factors; E₈ also follows directly from the known ideal theorem. This is algebraic assembly, not a new ideal enumeration.

Write the finite correction as D_S(x). Its coefficient lists (ascending powers) are:

- G: `[1,1,7,7,15,15,35,15,30,14,28,8,16]`;
- S₁: `[1,2,15,35,15,16,15,35,14,14]`;
- R: `[1,0,3,0,1,0,3]`;
- E₈: `[1]`.

These are also the primitive 2-homothety-class polynomials already implicit in the earlier conductor proof: their coefficient sums are 192,162,8,1. All coefficients are nonnegative, with constant term 1. Thus the Dirichlet series converge absolutely for Re s>1/8.

## 2. Complete zero and pole description

The factorization gives the full zero multiset, with multiplicities added if zeros coincide:

1. **[known] Inherited nontrivial zeros:** s=ρ/8 for every nontrivial ζ-zero ρ. Their strip is 0<Re s<1/8. The symmetry center of this inherited family is **1/16**. Every zero on the ordinary half-line maps to 1/16; an off-half-line ordinary zero, if any, would map off 1/16. No RH assumption is made.
2. **[known] Inherited trivial zeros:** s=−k/4, k=1,2,…, from ζ(−2k)=0. They stay simple: D_S(2^(k/4))>0, so the correction neither cancels nor adds a zero there. The displayed window includes −¼ and −½, on the real axis.
3. **[project, proved] Prime-2 numerator zeros:** if Q_S(z)=0, then all solutions 2^(-s)=z are zeros, with period 2πi/log2. G has four on Re s=½ per period and four off; S₁ has five off; R two off; E₈ none. The original project Sturm/gcd counts justify the half-line assertions. Plotted off-line coordinates remain [project, numerical], cross-checked with numpy and 60-digit mpmath this turn.
4. **[project, proved] Euler-replacement zeros:** for G, S₁ and R, the additional factor 1+x⁴ vanishes at

       s = i(2n+1)π/(4 log2), n∈Z.

   These lie on Re s=0. There are 13 positive-height examples through t=30. They are **not zeros of F₂'s numerator**, so a map made by merely adding the F₂ zeros to the inherited zeros would miss them. They arise when the missing odd-prime product is assembled with the 2-part. E₈ has no such column.
5. **[project, proved] Only one finite pole:** s=1/8, simple, with residue D_S(2^(-1/8))/8>0. Every apparent F₂ denominator pole has canceled in the global formula. In particular s=0 is **not** a pole or zero: Z_S(0)=−D_S(1)/2, namely −96,−81,−4,−½ for G,S₁,R,E₈. The plot is of Z_S, not a completed function.

All polynomial extra roots are simple: exact gcd(D,D′)=1. The bridge zeros cannot overlap ζ-zeros: ζ has no nontrivial zeros on Re w=0 and no trivial zeros at nonreal w. Exact algebraic gcds also rule out **any correction zero on Re s=1/16**: put α=2^(-1/8) and compute gcd(D(x),x^d D(α/x)) over Q(α); it is 1 for all nonconstant D. For real coefficients a root on |x|=2^(-1/16) would have to occur in this gcd. Hence zeros of these functions on 1/16 are exclusively inherited, not local.

Only G has global zeros on Re s=½: its four-per-period local ones. ζ(8s) has no zeros there, since Re(8s)=4>1, and the other corrections have none on ½ by the established exact tests.

### Finite plot data and status

- Window Re s∈[−0.6,2], 0≤Im s≤30, no vertical normalization this time.
- **102 inherited zeros in every panel.** Source: the first four chunks `order-height/zeros/0000001.json`, `0000033.json`, `0000065.json`, `0000097.json`, with exact rational brackets divided by 8. Completeness comes from the earlier Hardy-sign/Turing certificate through ordinary height 600 (`order-height/turing-600.json`), not a new interval proof this turn. A separate mpmath nzeros(240)=102 sanity check agrees; that check alone is not a certificate. Source hashes and transported brackets are saved in `global1111/results.json`.
- Positive-height numerator / replacement counts: G **27+13**, S₁ **17+13**, R **7+13**, E₈ **0+0**. Add 102 inherited zeros and the two real trivial zeros to get the shown raw-zero counts: 144,134,124,104 respectively. Each panel also has its one pole.
- Local polynomial positions: numpy roots and mpmath polyroots agree to <1e−10, with exact factor identities and gcd checks; not interval enclosures for the local root coordinates.

## 3. Completion and the meaning of a critical line

### E₈: inherited functional equation [known]

DLMF §25.4.3–4 gives ξ(w)=ξ(1−w), with ξ(w)=½w(w−1)Γ(w/2)π^(-w/2)ζ(w). Substituting w=8s yields the entire completion

    H_E8(s) = ξ(8s)
            = ½(8s)(8s−1) Γ(4s) π^(-4s) Z_E8(s),
    H_E8(s) = H_E8(1/8−s).

Its natural center is **1/16**, not ½. Gamma factors remove the trivial-zero/pole features in the usual way; the raw-function plot intentionally shows them. A change of variable u=8s would rename that center ½, without new mathematics.

### G, S₁ and R: the usual completion fails [project, proved]

Define the same natural entire completion

    H_S(s) = ½(8s)(8s−1) Γ(4s) π^(-4s) Z_S(s)
           = ξ(8s) D_S(2^(-s)).

It does **not** satisfy H_S(s)=ε C^s H_S(1/8−s) for any nonzero ε,C, for any of these three orders. Such nonvanishing exponential normalizations cannot repair an asymmetric zero set.

Here is an exact, common obstruction, rather than a numerical failure to match values. Let s₀=iπ/(4log2), a zero of 1+x⁴. The reflected point is s₁=1/8−s₀. Its x-coordinate z=2^(-1/8) exp(iπ/4) has **2z⁸=1**. The polynomial 2x⁸−1 is irreducible over Q (take its reciprocal and apply Eisenstein at 2 to x⁸−2). Exact gcd(Q_S,2x⁸−1)=1 for G,S₁,R shows Q_S(z)≠0; also 1+z⁴=1−1/√2≠0. Finally ζ(8s₁)=ζ(1−2πi/log2)≠0 by the known zero-free line Re w=1 (DLMF §25.10). Its gamma and xi prefactors are finite and nonzero. So H_S(s₀)=0 but H_S(s₁)≠0. The mirror fails exactly.

G's local reciprocal identity remains true:

    Q_G(x) = (2x²)^4 Q_G(1/(2x)).

Its local reflection has center Re s=½, not the inherited 1/16. The added 1+x⁴ polynomial has center 0. R's numerator 1+3x² reflects about log3/log4≈0.79248, not ½ or 1/16; S₁'s numerator has no reflection center of this type. Therefore none of these local properties supplies a global functional equation.

**An extra exact test [project, computer-assisted]:** not even D_G,D_S1,D_R alone has a single reflection center. If D(x)=a_d x^d D(A/x), then A^d=a_d^(-2) and coefficient matching forces A^k=a_(d−k)/(a_d a_k). For G and S₁ the k=1 relation forces A=½, contradicting the degree relation (d=12,a_d=16 and d=9,a_d=14). For R, k=2 forces A²=1/9, inconsistent with A⁶=1/9. The script checks these with rational arithmetic. This explains why retaining G's reciprocal Q while ignoring 1+x⁴ would give a false global mirror.

### Limits of the conclusion

We have ruled out the **usual rescaled-zeta gamma completion**, even with a nonvanishing exponential conductor factor. We have not proved that no imaginable completion or functional identity exists. An artificial symmetric entire function can always be made:

    ξ(8s) D(2^(-s)) D(2^(-(1/8−s))).

This deliberately adds the missing mirror zeros; it is not the original arithmetic count. Dividing out D simply returns ξ(8s), erasing the order-specific zeros. Neither construction is evidence of a new arithmetic functional equation.

Thus E₈ has a natural global symmetry center 1/16. For G,S₁,R, **1/16 remains the inherited zeta-family center, not an established symmetry line of the whole function**. G additionally has a genuine local half-line family; no single line contains all global nontrivial zeros. The term “critical line” must always name which factor and which variable is intended. No RH progress or new universal half-line theorem is claimed.

## 4. Sources, files, reproducibility

Opened this turn: [DLMF §25.4](https://dlmf.nist.gov/25.4) (xi completion), [§25.10](https://dlmf.nist.gov/25.10) (zero-free boundaries, trivial zeros, critical strip and RH). Prior primary source read directly in the preceding session: Petersson, *Norm equivalences and ideals of composition algebras*, Münster J. Math. 14 (2021), 283–293, Theorem 4.1(i), DOI [10.17879/06089651315](https://doi.org/10.17879/06089651315).

Project inputs: `octonion-intermediate-orders-20261111.md`, `gravesian-periodicity-proof.md`, `REFINEMENT-20261111.md`, `octonion-odd-primes-20261111.md`, and the earlier interval/Turing archive. The global theorem inherits the computer-assisted status of the nonmaximal 2-local factors; it is not an independently derived ideal-count formula.

Reproduce:

    /workspace/tools/pari-venv/bin/python global1111/compute.py
    python3 global1111/figure.py
    python3 global1111/test_acceptance.py

System Python lacks sympy here; the existing PARI venv contains sympy, numpy and mpmath. Figures: `figures/octonion-global-zetas.svg`, `figures/octonion-global-zetas-mobile.svg`; site copies use `hyper-` prefix. PNG: `/workspace/math/octonion-global-zetas-20261111.png`.

## 5. Research status

Closed: assembling the four global ideal zetas, exact cancellation of apparent local poles, complete factor-level zero descriptions, the usual rescaled-zeta completion test, and the finite t≤30 illustration. Still open: conceptual proofs of the prime-2 duality/normalizer behavior; any genuinely motivated alternative arithmetic completion with its hypotheses specified; odd-conductor orders outside this sixteen-order family. Do not describe failure of the usual completion as an unrestricted no-functional-equation theorem. No schedule change and no messages.

Publication receipt is recorded in `/workspace/math/DEPLOY-LOG.md` and `global1111/live-receipt.json` after verification.
