# Octonion orders with odd conductor: local zeros at 3, none on Re s = 1/2

Research, figure and page section, 11 November 2026, ~1:00–2:15 PM PST. Workspace `/workspace/math`, scripts in `oddcond1111/`. No messages sent. No RH claim: everything here is about finite local factors at the prime 3.

## Plain explanation for Grant

Before today, every octonion order we had looked at was "perfect" (maximal) at odd primes, so the odd primes produced nothing: the only ideals there were scaled copies, and the local count was 1/(1 − p^(−8s)), with no zeros at all.

Today we deliberately damaged the order at 3. Take the best integer octonions O (the E₈ order). Near 3 they look like the split octonions, which can be written as 2×2 "Zorn" arrays [[a, u], [v, b]] with a, b numbers and u, v 3-vectors. Keep only those octonions whose remainder mod 3 lies in a chosen smaller algebra (a "subalgebra" of O/3O). That gives a new order Λ, non-maximal at 3 and unchanged at every other prime.

Then count one-sided ideals at 3 exactly as before. The result is the following.
- **Zeros now appear at 3.** The odd prime finally produces local zeros and an "odd-prime signal" in the explicit formula.
- **None of them lie on Re s = ½.** We checked six natural orders, and the test is exact, not numerical.
- **There is no mirror symmetry s ↦ c − s.** The quaternion analogues (Eichler orders) put all their local zeros on Re s = ½, and the integer-coordinate octonions at 2 put half of them there. These odd-conductor octonion orders do neither.
- **The simplest choice, Z + 3O, is degenerate.** Almost every lattice counts as an ideal, so its zeros only reflect "count all subspaces" combinatorics.

## 1. The orders  [project; definitions]

Model: Zorn(Z) with x = (a, u₁, u₂, u₃, v₁, v₂, v₃, b), N(x) = ab − u·v, and

(a,u,v,b)(a′,u′,v′,b′) = (aa′ + u·v′, au′ + b′u − v×v′, a′v + bv′ + u×u′, bb′ + v·u′).

This is the split octonion algebra. It is a maximal order at 3, and at 3 it is isomorphic to E₈ ⊗ Z₃ ([known]: E₈ ⊗ Z₃ is a unimodular composition algebra over Z₃, hence split). The code checks the norm and alternative laws (`zorn.py`).

For a subalgebra S of O/3O, set Λ_S = {x ∈ O : x mod 3 ∈ S}. Then 3O ⊆ Λ_S ⊆ O and [O : Λ_S] = 3^(8 − dim S).

A computer search (in `explore.py`/notes) finds the coordinate subsets that are closed under multiplication. Twenty-eight of them contain both diagonal entries a and b. Under coordinate symmetries they fall into six shapes, plus the scalar case, and we use one representative of each:

| name | S (coordinates kept mod 3) | dim S | [O:Λ] | what S is |
|---|---|---|---|---|
| Z | scalars only | 1 | 3⁷ | Λ = Z₃ + 3O, the order asked about first |
| D | a, b | 2 | 3⁶ | split étale F₃ × F₃ (diagonal) |
| U | a, b, u₁ | 3 | 3⁵ | "Borel-type": upper triangular inside a quaternion M₂(F₃) |
| Q | a, b, u₁, v₁ | 4 | 3⁴ | a split quaternion subalgebra M₂(F₃) |
| N4 | a, b, u₁, v₂ | 4 | 3⁴ | 4-dim with a nilpotent part |
| P5 | a, b, u₁, u₂, v₃ | 5 | 3³ | 5-dim |
| P6 | a, b, u₁, u₂, v₁, v₃ | 6 | 3² | 6-dim, parabolic-type |

The orders are Z⁸-type: in a Zorn frame each is spanned by coordinate vectors e_i (i ∈ S) and 3e_j (j ∉ S). The E₈-relative "integer-coordinate" Gravesian-type orders have index 16, a power of 2, so they are maximal at 3. Coordinate sub-orders of 3-power index in a split frame are therefore the natural Z⁸-type objects at 3.

Symmetries used:
- [project, checked] (a,u,v,b) ↦ (b, −v, −u, a) is an automorphism, and signed permutations of the three coordinates act too.
- [known] Conjugation is an anti-automorphism that preserves every Λ_S, so **left and right ideal counts are equal**. We also computed Q both ways.

## 2. What is counted  [definition, same as earlier notes]

A right ideal of Λ = Λ_S ⊗ Z₃ is a finite-index Z₃-lattice L ⊆ Λ with L·Λ ⊆ L. This is the same notion as in `octonion-ideal-count.md` and `octonion-odd-primes-20261111.md`: compositions of right multiplications are matrix products, not octonion products. Then a_k = #{L : [Λ:L] = 3^k}, and Z_{Λ,3}(s) = Σ a_k x^k with x = 3^(−s).

**Reduction [project, proved].**
- Write L = 3^a L₀, with L₀ ⊆ O not inside 3O.
- If a ≥ 1, then 3^a L₀ ⊆ 3O ⊆ Λ automatically. So with P_O(x) = Σ x^([O:L₀]) over primitive stable L₀ ⊆ O, and P_Λ(x) = the same sum over those L₀ lying inside Λ (index taken in Λ), we get

  Z_{Λ,3}(x) = P_Λ(x) + x^(dim S) · P_O(x) / (1 − x⁸).

- The multiplication algebra of each Λ contains 3^c M₈(Z₃), with c = 2 for Q and P6 and c = 3 for the others (an exact HNF computation). Hence every primitive stable lattice contains 3^c O, there are finitely many, and both P's are polynomials.

## 3. Computations (two independent codes, plus a structural count)

**Method A** (`method_A.py`, Zorn model, right ideals):
- breadth-first search over maximal stable sublattices;
- maximal submodules of L/3L are found as annihilators of minimal submodules of the dual, using all 3,280 projective functionals and the word basis of the action algebra;
- lattices are compared by exact integer HNF.

**Method B** (`method_B.py`, written separately, left ideals):
- recursive Cayley–Dickson integer octonions, maximal at 3, with no Zorn code imported;
- an explicit Zorn frame (e, u₁,u₂,u₃, v₁,v₂,v₃, f) is found inside them mod 3⁶ from a w with N(w) ≡ −1, and all 64 structure constants are verified;
- each Λ_S is transported into this frame;
- the BFS uses **all** stable M with 3L ⊆ M ⊊ L, listed as sums of cyclic submodules, and a separately written Howell-type canonical form over Z/3⁶.

| order | primitive lattices | P_O (coefficients by index) | P_Λ | A = B? |
|---|---|---|---|---|
| P6 | 10 | 1,0,1,0,4,0,1,0,3 | 1,0,4,0,1,0,3 | yes |
| Q | 59 | 1,0,8,0,18,0,8,0,24 | 1,0,4,0,12 | yes (also left = right in A) |
| P5 | 502 | 1,1,13,13,14,40,130,40,40,39,117,27,27 | 1,2,40,130,40,40,39,117,27,27 | yes |
| N4 | 1018 | 1,2,25,26,36,80,259,80,89,78,234,54,54 | 1,2,25,26,35,78,234,54,54 | yes |
| U | 4874 | 1,8,90,176,411,392,1026,392,626,384,936,216,216 | 1,2,43,136,192,468,108,108 | yes (method B finished at 2:16 PM PST Nov 11, just after first publication; P_O and P_Λ identical) |

**Z = Z₃ + 3O (degenerate; structural count) [project, proved + computer-assisted].**
- L·Λ ⊆ L means exactly 3·span(L·O) ⊆ L.
- Every primitive L ⊆ Λ is Λ itself. Every lattice L with 3^(a+1)O ⊆ L ⊆ 3^a O (a ≥ 1) is an ideal, so all subspaces of F₃⁸ appear. **That is the degeneracy.**
- For primitive M ⊆ O: if M mod 3 contains an element of unit norm, then M ⊇ 3O. Otherwise W = M mod 3 is totally singular. Write M ∩ 3O = 3K. Then M ↔ (K, a lift of W), K is again stable, and span(W·O) ⊆ K mod 3.
- This gives P_O = C₁ + C₂, where:
  - C₁ sums [8 − t, k − t]₃ · 3^(w(8−k)) x^(16−k−w) over all 84,560 totally singular W. The counts by dimension are 1,120 / 36,400 / 44,800 / 2,240, matching the known O⁺(8,3) numbers, with t = dim span(W·O) computed for every W. Non-singular W add Gaussian binomials.
  - C₂ = 1,120 × 81 x¹² (the "deep" lattices over isotropic lines). These were enumerated and verified for each of the 1,120 lines.
- **Independent check:** a generic BFS (method A's routine, with no use of the recursion), restricted to the lattices M with M + 3O = W̃ + 3O, reproduces the per-W counts for each type of W. That covers a plane (1, 36, 81), a solid (1, 27), both kinds of maximal W ((1, 81) and (1)), and isotropic lines of both trace classes (1, 120, 1170, 1080, 81, 81 at indices 7–12; this includes the 81 deep lattices).
- P_O = 1, 3280, 896260, 25095280, 75913222, 25186000, 2105860, 1313680, 3082800, 1310400, 1209600, 90720, 90720 (indices 0–12).

**D (diagonal F₃ × F₃): not finished [honest gap].** The ideal theory is degenerate in the same way: R_e is a projection, and the top layer alone has 212² = 44,944 ideals. I did not complete a structural count, so no generating function is claimed.

## 4. Generating functions and zeros  [project; exact counts, then exact gcd tests; zero positions in double precision with 40-digit mpmath roots]

After cancelling common factors with 1 − x⁸:

| order | Z_{Λ,3}(x) = N(x)/den | zeros per period | on Re s = ½ | Re s range | mirror / palindrome |
|---|---|---|---|---|---|
| P6 | (1 + 4x²)/(1 − x⁴) | 2 | 0 | all at log 2/log 3 ≈ 0.6309 (exact) | trivial (one line, not ½) |
| Q | (1 + 4x² + 13x⁴ + 8x⁶ + 17x⁸ + 4x¹⁰ + 12x¹²)/(1 − x⁸) | 12 | 0 | −0.021, 0.048, 0.539 | none |
| P5 | (1 + 2x + 40x² + 130x³ + 39x⁴ + 39x⁵)/(1 − x⁴) | 5 | 0 | −0.515 … 1.705 | none |
| N4 | (1 + 2x + 25x² + 26x³ + 35x⁴ + 78x⁵ + 234x⁶ + 54x⁷ + 54x⁸)/(1 − x⁴) | 8 | 0 | −0.631, **0.5055, 0.5131**, 1.428 | none |
| U | (1 + 2x + 43x² + 137x³ + 199x⁴ + 556x⁵ + 241x⁶ + 382x⁷ + 192x⁸ + 468x⁹ + 108x¹⁰ + 108x¹¹)/(1 − x⁴) | 11 | 0 | −0.574 … 1.695 | none |
| Z | N_Z(x)/(1 − x⁸), degree 13, coefficients up to 75,913,222 | 13 | 0 | −1.092 … 4.239 | none |

- **On the half-line [project, exact].** A zero with Re s = ½ means |x|² = 1/3. Such a root would also be a root of the reflected polynomial x^d N(1/(3x)). For every order, gcd(N(x), x^d N(1/(3x))) = 1 over Q, so **no order has any zero on Re s = ½**. N4 comes closest: four zeros at Re s ≈ 0.5055 and 0.5131, near the line but provably not on it.
- **Mirror/palindrome [project].** Apart from P6, whose zeros all share one real part, the real parts are not symmetric about any vertical line. No coefficient palindrome c_{d−k} = λ^(d/2−k) c_k holds, so there is no local functional equation s ↦ β − s.
- **Comparison [known/project].**
  - Quaternion Eichler orders of level p^r at odd p: every local root is on Re s = ½ (`order-eichler-unitcircle-proof.md`, [project, proved]).
  - The Solomon closed form: also on the circle (conditional on the closed form).
  - Gravesian G at 2: 4 of 8 on ½, with an exact s ↦ 1 − s palindrome.
  - S₁ and R at 2: 0 on ½.
  - Odd-conductor octonions at 3 behave like S₁ and R, not like Eichler orders: **0 on ½, and no symmetry.**

## 5. Global zeta and explicit formula  [project, proved from the local results]

Λ_S is maximal at every p ≠ 3, and there the factor is 1/(1 − p^(−8s)) (Petersson 2021, Thm 4.1, [known]). Hence

  Z_Λ(s) = ζ(8s) · (1 − 3^(−8s)) · Z_{Λ,3}(s),

which equals:
- ζ(8s) · (1 + 3^(−4s)) · N(3^(−s)) for P6, P5, N4 and U;
- ζ(8s) · N(3^(−s)) for Q and Z.

Its zeros are:
- the inherited zeros of ζ(8s), on Re s = 1/16 exactly if RH holds (no claim);
- the trivial zeros;
- the zeros of 1 + 3^(−4s), on Re s = 0;
- the finite local zeros above.

**Explicit-formula signature at powers of 3.**
- −Z′/Z = 8(−ζ′/ζ)(8s) + (log 3) Σ_k e_k 3^(−ks), where e_k are exact integers (`explicit-weights.json`). For the maximal order e_k = 0 for all k.
- Q: e₂, e₄, e₆, e₈, e₁₀ = 8, 20, −136, 356, 88 (odd k: 0).
- P6: 8, −28, 128, −516, 2048 at k = 2, 4, …, i.e. e_k = −2(−4)^(k/2) plus the 1 − x⁴ correction.
- P5 / N4 / U: e₁ = 2, with rapidly growing signed weights.
- Z: e₁ = 1, e₂ = 6559, …

**So an odd-prime signal now appears [project].** The signed diagnostic B(x) = 8ψ₀(x^(1/8)) + (log 3)Σ*_{3^k ≤ x} e_k has jumps at powers of 3 that are not multiples of 3⁸. Their size grows like 3^(k·max Re ρ_local): 3^(0.539k) for Q, 2^k for P6, 3^(4.24k) for Z. It is a finite, exactly periodic-in-log-x local term, like the prime-2 jumps of G (b₄ = −39), and it carries no information about ζ's zeros. The ideal count itself, A_Λ(x) = Σ_j c_j ⌊(x/3^j)^(1/8)⌋ with C = (1 − x⁸)Z_{Λ,3}, still has bounded error and no zero waves, exactly as in `octonion-explicit-formula-20261111.md`.

## 6. What this settles / what stays open
- **Settled for these six shapes and this ideal notion (all six double-computed):** odd conductor at 3 produces local zeros, an odd-prime explicit-formula signal and no half-line or mirror structure. It is not an Eichler-like circle theorem.
- **Degenerate:** Z + 3O, where the counts are essentially Gaussian binomials, and D (diagonal), which was not completed.
- **Open:**
  - other ideal notions (two-sided ideals only; "invertible"/locally principal ideals; ideals relative to a bigger order), which might restore a functional equation;
  - p = 5, 7, where the same shapes should give 5- and 7-adic analogues (not computed);
  - finishing D;
  - an outside referee.

- **Follow-up (Nov 11, ~5 PM PST):** the same shapes at p = 5, 7 (and N4, P5 at 11) are in `octonion-odd-conductor-5-7-20261111.md`. The verdict is the same, and the coefficients follow polynomials in p [project conjecture].

## Reproduce

    cd oddcond1111
    python3 zorn.py
    python3 method_A.py Q,P6,P5,N4,U R
    python3 method_B.py Q,P5,N4,U
    python3 ts.py && python3 zstruct.py full && python3 zcheck.py all
    python3 analyze.py && python3 explicit.py && python3 figure.py
    python3 test_oddcond.py

Figures: `figures/octonion-odd-conductor.svg` (1240×1180) and `-mobile.svg` (400×2600). PNG: `/workspace/math/octonion-odd-conductor-20261111.png`.
