# Octonion right-ideal counts at odd primes: no local zeros

Research and figure, 2026-11-11 10:23 AM PST. Workspace `/workspace/math`. No messages sent.

## Plain explanation for Grant

**[known] 2 really is special in this family.** At 3, 5 and 7, all sixteen orders give exactly the same answer. The only right ideals are copies of the entire order shrunk by p, p², p³, and so on. Shrinking eight coordinates by p gives index p⁸. Thus the count is

    1 + p^(-8s) + p^(-16s) + ... = 1/(1 - p^(-8s)).

Its numerator is 1, so it has **no zeros anywhere**, including on Re s = ½. This is not a failed numerical search: it is an exact known theorem, independently checked by two calculations here. The interesting local zero pattern of G is confined to 2, not repeated at odd primes. This says nothing new about RH.

## 1. What is counted

For S one of G, S₁, the seven R-type orders or the seven maximal E₈-type orders, put Oₚ = S ⊗ Zₚ. A right ideal is a finite-index Zₚ-sublattice L ⊆ Oₚ such that **L·Oₚ ⊆ L**. Equivalently, every right-multiplication linear map R_y: x ↦ xy preserves L. Compositions of these maps are ordinary associative compositions of linear operators; we do **not** replace R_yR_z by a single octonion product. Nonassociativity is precisely why those maps can generate a much larger matrix algebra. This is not a count of arbitrary sets xOₚ, whose right stability must not be assumed, nor a count of subalgebras or element norms.

Define aₖ = #{L : [Oₚ:L] = pᵏ}, and Zₛ,ₚ(s) = Σ aₖ p^(-ks). This fixes the **additive index** convention; an element-norm exponent would be different.

All [S:G] are powers of 2. Consequently G⊗Zₚ = S⊗Zₚ for every odd p, under the actual inclusions, not just up to an abstract isomorphism. The norm of G is Σxᵢ², with polar Gram matrix 2I₈. At odd p its determinant 2⁸ is a unit, so this is an octonion **composition algebra over Zₚ**. At 2 the polar form for G is degenerate, so the theorem below cannot be applied to G₂, S₁,₂ or R₂ just because each is called an order. Maximal E₈ octonions do have the nonsingular composition-algebra structure at 2. The odd local algebra is the split octonion algebra; our calculation and theorem do not require selecting a particular split isomorphism.

## 2. Literature: stronger, published source located and read

**[known]** H. P. Petersson, *Norm equivalences and ideals of composition algebras*, **Münster Journal of Mathematics 14 (2021), 283–293**, DOI [10.17879/06089651315](https://doi.org/10.17879/06089651315). Primary PDF downloaded and section 4, pp. 289–292, read directly this turn:

https://www.fernuni-hagen.de/mi/fakultaet/emeriti/docs/petersson/norm-equiv.-ideals-comp.-alg.pdf

Local PDF/text: `octonion-sources/petersson-norm-equivalences-ideals.{pdf,txt}`.

**Theorem 4.1(i), p. 289:** for a rank-8 composition algebra C over **any commutative ring k**, a ↦ aC bijects the ideals of k with the one-sided ideals of C; inverse I ↦ I∩k. The proof first shows right ideals are two-sided (Cayley–Dickson), then that two-sided ideals extend from k. Section 4.4(b) emphasizes that the analogous one-sided statement is false for split quaternions. This supersedes the narrower Dedekind-domain formulation in the already inspected 2018 Malaga notes, Theorem 1(b), without changing its consequence here. No mathematical novelty claim.

Apply the published theorem to k=Zₚ. Every nonzero base ideal is pᵃZₚ; hence L=pᵃOₚ for a≥0, and [Oₚ:L]=p^(8a). It follows exactly, for **every odd prime**, that

    aₖ = 1 if 8 divides k, and 0 otherwise;
    Zₛ,ₚ(s) = 1/(1-p^(-8s)).

This is [known], not a fitted [project conjecture]. It holds for left and right ideals alike. No assumption of RH is used.

## 3. Independent calculations: what was actually computed

**[project, proved, computer-assisted]** Two fresh programs, no imports from the earlier counting programs and no shared multiplication code, give exact certificates. Both write counts through k=32 for p=3,5,7. They obtain the counts by **classifying stable lattices from their action algebra**, not by brute-force sweeping all Hermite forms of index p³². This is stronger than a finite fit, but it must not be described as two exhaustive HNF enumerations.

### A. Integer Fano words and explicit matrix units

`odd1111/method_A.py`, `run-A.txt`, `results-A.json`, `certificate-A.npz`.

Use the project's Fano lines (i,i+1,i+3) mod 7 to construct the seven 8×8 signed-permutation right-multiplication matrices. Take all products with increasing distinct indices and lengths 0,1,2,3. There are 1+7+21+35=64. Their vectorized rows W have exact Gram matrix **WWᵀ=8I₆₄**. The code also checks, separately for every a,b, the identity

    8E_ab = Σ_w w_ab · w.

Thus 8M₈(Z) is in the operator algebra A; its 64-word span has index 2⁹⁶ (determinant squared 8⁶⁴). Since 8 is invertible at every odd prime, A⊗Zₚ=M₈(Zₚ). Every stable lattice is a scalar ideal times Zₚ⁸: elementary matrices isolate each coordinate and transfer it to all positions. Hence there is exactly one ideal at k=8a and none otherwise, for all k. The five scalar candidates through k=32 are directly checked stable.

This is also a short uniform explanation of why only the prime 2 can be exceptional; no enormous p=7 enumeration is needed.

### B. Independent Cayley–Dickson multiplication and modular saturation

`odd1111/method_B.py`, `run-B.txt`, `results-B.json`.

Recursively double the real, complex and Hamilton bases, with

    (a,b)(c,d) = (ac - conjugate(d)b, da + b conjugate(c)).

Construct right actions afresh (different basis labelling, no Fano lookup). Over each Fₚ, close their associative matrix algebra with independent row elimination. For p=3,5,7 the dimensions by closure round are **8,29,64,64**. Thus the reduction acts as all M₈(Fₚ), and has no proper nonzero invariant subspace.

Why this determines the count at every p-power level: given finite-index stable L⊆Oₚ, divide by its largest common coordinate power pᵃ to get a primitive stable lattice L₀. Its reduction in Oₚ/pOₚ is nonzero, hence full. Therefore Oₚ/L₀ = p(Oₚ/L₀); this finite p-primary module is zero (Nakayama, or iterate multiplication by p). Thus L₀=Oₚ and L=pᵃOₚ. The code constructs the scalar chain and its determinant indices through p³². The standard Cayley–Dickson model has the same localized octonion algebra; the classification also follows directly in that model without relying on A's certificate.

### Count table (each prime, each kind, right ideals)

| p | G | S₁ | each R-type | each E₈-type |
|---|---|---|---|---|
| 3 | 1/(1−3⁻⁸ˢ) | same | same | same |
| 5 | 1/(1−5⁻⁸ˢ) | same | same | same |
| 7 | 1/(1−7⁻⁸ˢ) | same | same | same |

For all twelve prime/kind combinations, aₖ for 0≤k≤32 is 1 exactly at **0,8,16,24,32**, and zero at the other 28 exponents. A and B agree exactly. The proofs extend to every index; the uniform A certificate extends to every odd prime. p=7 is not an infeasibility caveat: the structural method avoids a huge lattice search.

## 4. Zeros, poles and the comparison figure

**[known]** With x=p⁻ˢ, the odd factor is 1/(1−x⁸), whose numerator is 1. It has **no finite complex zeros**. It has simple poles at s=2πin/(8 log p), n∈Z, all on Re s=0. In particular s=0 is a pole, not a zero. The empty panels depict the absence of zeros, not a failed root finder.

The prime-2 panels retain the established exact project factors and numerical root positions:

| kind | numerator (x=2⁻ˢ) | denominator | zeros on ½ / total per 2π/log2 period |
|---|---|---|---|
| G | 1+x+7x²+7x³+14x⁴+14x⁵+28x⁶+8x⁷+16x⁸ | 1−x⁴ | 4 / 8 |
| S₁ | 1+2x+15x²+35x³+14x⁴+14x⁵ | 1−x⁴ | 0 / 5 |
| R | 1+3x² | 1−x⁴ | 0 / 2 |
| E₈ | 1 | 1−x⁸ | 0 / 0 |

`odd1111/figure.py` recomputes these polynomial roots and checks residuals/root counts; the exact on-line counts are from the earlier Sturm/gcd proofs, not upgraded from these floats. In the diagram the vertical coordinate is **u = Im(s) log(p)/(2π)**, one normalized period −½<u≤½ for each prime; the horizontal coordinate remains Re s. Thus panels compare periods, not equal raw heights. Dashed gold marks Re s=½; green dots are the four on-line G zeros and rust dots are off-line zeros. Poles are not plotted.

Figures: `figures/octonion-odd-primes.svg` (1240×1030), `figures/octonion-odd-primes-mobile.svg` (400×2230). Site copies have `hyper-` prepended. PNG still requested at `/workspace/math/octonion-odd-primes-20261111.png`.

The local functions here have a **complex parameter s**. The figure concerns arithmetic inside octonion orders; it is not an eight-dimensional rendering of the intrinsic slice-lift zero spheres. An empty local panel does not mean the global ideal zeta has no zeros: even ζ(8s) has zero-free Euler factors in its convergence domain while its continuation inherits the ordinary zeta zeros. Infinite products and continuation must not be confused with a finite product of entire functions. No RH claim.

## 5. What remains open / reproducibility

This closes the odd-prime test for **these sixteen orders** and this ideal notion at every odd prime, not just 3,5,7. Do not queue repeated counts at 11 or 13 expecting different behavior. To find other odd-prime local phenomena, one would have to change the order so its conductor has an odd-prime factor (outside today's computation), or change the arithmetic object with explicit definitions. No such new count was performed here. The conceptual prime-2 normalization/duality questions in the earlier notes remain open; this does not settle them. No outside referee for the project's 2-local factors.

Reproduce from `odd1111/` with system Python and numpy:

    python3 method_A.py
    python3 method_B.py
    python3 figure.py
    python3 test_acceptance.py

The acceptance test is green only after the site build publishes the new section and assets. The mathematical certificates use exact integer/modular arithmetic; figure positions use ordinary numpy complex roots, not rigorous intervals.

## Publication

Published 11 November 2026, 10:25 AM PST, version `b6dcb774-673f-4033-8de3-c0144d312053`.
Live: https://zeta-zeros.grant-golland.workers.dev/critical-line-hypercomplex#octonion-odd-primes
Build/check: NO ERRORS (13 pages, 1885 local references); two acceptance tests pass; Wrangler dry-run passed. Live page and 11 new assets/notes return HTTP 200 and match the built bytes; all 316 previous live page IDs survive. Receipt: `odd1111/live-receipt.json`; `DEPLOY-LOG.md` updated. Desktop/mobile PNG renders inspected; unsupported superscript glyphs replaced with unambiguous ASCII formulas. The publication receipt was appended to this local research copy after deploying; the published note has the same mathematics without this final receipt.

The existing build automatically copied two pre-existing Fox–Artin SVG assets from the shared figures directory. No Fox–Artin page section or new research claim was added.

Authorized figure: Grant t190u, “go ahead and make a new figure”. No outgoing messages.
