# Octonion orders with odd conductor, every odd prime at once: proofs for P6 and Q, partial results for U, N4, P5

Research note, 11 November 2026, ~5:00–5:15 PM PST (Grant pt207u, "start"). Follows `octonion-odd-conductor-20261111.md` (p = 3) and `octonion-odd-conductor-5-7-20261111.md` (p = 5, 7, 11). Checks are in `oddcond1111/proof_blocks.py`, with output in `proof-blocks-check.json`. No messages sent.

**Update ~9:05 PM PST (Grant pt216u):** U is now proved for every odd prime as well (`octonion-u-proof-20261111.md`), so all five patterns P6, Q, N4, P5, U are proved and none has a zero on ½; the U limits in §5 (e.g. ½ + 1/(2p log p)) are now unconditional.

**Update ~8:25 PM PST (Grant pt212u):** N4 and P5 are now proved for every odd prime, and 750 of U's 802 cells are proved. See `octonion-n4-p5-proof-20261111.md`, which uses a hull/core/glue reduction plus an exact count. Their "no zero on ½" no longer needs the fitted formulas. N4's 1/(4√2 p^(3/2) log p) constant is now derived. What remains for U is stated there (§4). No RH claim: everything is about finite local factors at one prime p.

## Plain explanation for Grant (intuition first)

Until now we computed prime by prime and then guessed a pattern. Here we prove two of the patterns for every odd prime at once.

**Why a proof is possible.** Near p, an octonion is a 2×2 "Zorn" array [[a, u], [v, b]]: two numbers a and b, and two 3-vectors u and v. Multiplying on the right by the corner element e = [[1,0],[0,0]] keeps only (a, v). Multiplying by f = [[0,0],[0,1]] keeps only (u, b). Both corners lie in all our orders. So every ideal splits cleanly into an "(a, v) half" and a "(u, b) half".

For two of the orders, P6 and Q, we can split further. They contain both u₁ and v₁ unharmed, and multiplying by u₁ and then by v₁ copies a coordinate pair across and back. That shows:
- the ideal is made of four 2-dimensional pieces;
- the pieces come in matching pairs.

So the whole ideal is determined by **two lattices in the plane**, L₁ and L₂, plus a short list of rules linking them. Counting lattices in the plane is classical: there are exactly p + 1 lattices of index p. So the counts become explicit polynomials in p.

**What we get:**
- **P6 [project, proved]:** the local factor is (1 + (p+1)x²)/(1 − x⁴), with x = p^(−s), for every odd prime p.
- **Q [project, proved]:** the local factor is (1 + (p+1)y + (p²+p+1)y² + 2(p+1)y³ + (p²+2p+2)y⁴ + (p+1)y⁵ + p(p+1)y⁶)/(1 − y⁴), with y = x² = p^(−2s), for every odd prime p. The p = 7 run was not finished when this was proved, so this formula *predicted* it: 211 lattices, numerator 1, 8, 57, 16, 65, 8, 56. The exhaustive run (method A) has since finished and gives exactly that (update ~7:30 PM PST).
- **No zero on Re s = ½, for every prime, for P6 and Q [project, proved].** P6's zeros all sit at Re s = log(p+1)/(2 log p). That is always a little more than ½, since p+1 > p, and it tends to ½ like 1/(2p log p). Q has four zeros that approach ½ from the right, like 1/(4p log p), and never reach it.
- **U, N4, P5 [project, partial proof at the time of writing; N4 and P5 proved later the same evening, see the update above].** For these we prove:
  - the general set-up;
  - the coefficient of x, which is exactly 2 for every p, explained by a short theorem.

  The full classification is not done. For these orders, multiplication is "nilpotent mod p": it pushes coordinates along chains and never copies a pair back. So there is no splitting into small pieces. The gap is stated in §5.
- **If the conjectured U, N4, P5 formulas are right, then they have no zero on ½ for any prime [project, proved conditionally].** We also give their limits as p → ∞, where N4 has four zeros closing in on ½ from both sides extremely fast.

## 1. Set-up  [known + definitions]

- O_p = Zorn(Z_p), with x = (a, u₁,u₂,u₃, v₁,v₂,v₃, b), coordinates indexed 0–7, and multiplication (a,u,v,b)(a′,u′,v′,b′) = (aa′+u·v′, au′+b′u−v×v′, a′v+bv′+u×u′, bb′+v·u′). For odd p this is E₈ ⊗ Z_p [known]. Nothing in §§2–4 uses p ≠ 2, but our shapes are defined for odd p.
- Λ_S = {x ∈ O_p : x mod p ∈ S} for a coordinate subalgebra S. So Λ_S has Z_p-basis E_i (i ∈ S) and pE_j (j ∉ S).
  - P6: S = {a, u₁, u₂, v₁, v₃, b}.
  - Q: S = {a, u₁, v₁, b}.
  - U: S = {a, u₁, b}.
  - N4: S = {a, u₁, v₂, b}.
  - P5: S = {a, u₁, u₂, v₃, b}.
- A right ideal is a full lattice L ⊆ Λ with L·Λ ⊆ L, and Z_{Λ,p}(s) = Σ [Λ : L]^(−s).

**Lemma 1 (reduction) [project, proved].** Every right ideal L is p^k M, where M ⊆ O_p is a "primitive" Λ-stable lattice (M·Λ ⊆ M, M ⊄ pO_p).
- If k ≥ 1, then L ⊆ pO_p ⊆ Λ automatically.
- If k = 0, we need M ⊆ Λ.

So Z_{Λ,p} = P_Λ(x) + x^(dim S) P_O(x)/(1 − x⁸), where P_O = Σ x^[O:M] over primitive stable M, and P_Λ = Σ x^([O:M] − (8 − dim S)) over those inside Λ.

*Proof.* Take k maximal with L ⊆ p^k O. Stability is scale-invariant, and [Λ : p^k M] = 8k + [O : M] − (8 − dim S). ∎

**Lemma 2 (stability is checked on a basis) [project, proved].** M is Λ-stable if and only if M·y ⊆ M for each y in the Z_p-basis of Λ above. This holds because right multiplication is linear in y.

## 2. The block maps  [project, proved; checked against the multiplication table by `proof_blocks.py`]

Write A = (a, v₁, v₂, v₃) and B = (u₁, u₂, u₃, b). Right multiplication acts as follows:
- by e: (a, v) is kept and (u, b) ↦ 0;
- by f: the reverse.
- by E_{u_j}: A → B by a ↦ u_j, v_j ↦ b (call it φ_j); B → A by u ↦ u × e_j in the v-slot (ψ_j).
- by E_{v_j}: A → B by v ↦ −v × e_j in the u-slot (χ_j); B → A by u_j ↦ a, b ↦ v_j (ω_j).

Each is a signed single-coordinate map with no p's. In Λ_S, the maps for j ∉ S appear multiplied by p.

**Lemma 3 (Peirce split) [project, proved].** For every shape, e and f lie in Λ, so M = Me ⊕ Mf with Me ⊆ A-coordinates and Mf ⊆ B-coordinates.

## 3. P6 and Q: four blocks, two plane lattices  [project, proved]

Assume u₁ and v₁ are both in S (true for P6 and Q). Then:
- ω₁φ₁ is the projection of A onto (a, v₁);
- ψ₁χ₁ is the projection of A onto (v₂, v₃);
- φ₁ω₁ and χ₁ψ₁ are the projections of B onto (u₁, b) and (u₂, u₃).

These are compositions of right multiplications, so M is stable under them. Hence M = A′ ⊕ A″ ⊕ B′ ⊕ B″, with each piece in one coordinate pair. Moreover φ₁ maps A′ onto B′ (inverse ω₁), and χ₁ maps A″ onto B″ (inverse ψ₁). So

  **M ↔ (L₁, L₂)**: L₁ ⊆ Z_p² gives A′ in (a, v₁) and B′ in (u₁, b) by the same coordinates, and L₂ ⊆ Z_p² gives A″ in (v₂, v₃) and B″ in (u₃, −u₂).

Here [O : M] = 2([Z²:L₁] + [Z²:L₂]). The remaining eight maps φ₂, φ₃, χ₂, χ₃, ψ₂, ψ₃, ω₂, ω₃ each send one coordinate of L₁ to one coordinate of L₂, or back. Written out:
- X → Y: (x₁,x₂) ↦ (0,x₁) [φ₂, ψ₂], (0,x₂) [χ₃, ω₃], (x₁,0) [φ₃, ψ₃], (x₂,0) [χ₂, ω₂];
- Y → X: (y₁,y₂) ↦ (0,y₁) [φ₂, ψ₂], (y₁,0) [χ₃, ω₃], (0,y₂) [φ₃, ψ₃], (y₂,0) [χ₂, ω₂].

Signs do not matter, because each map moves a single coordinate. Notation: c(L) = p^m means the smallest valuation of an entry of L is m, and π₁, π₂ are the coordinate projections.

**P6** (u₂, v₃ in S, so φ₂, ψ₂, χ₃, ω₃ are unscaled; u₃, v₂ not in S, so φ₃, ψ₃, χ₂, ω₂ carry a factor p). M is stable if and only if:
- (i) p^(m+1)Z ⊕ p^m Z ⊆ L₂, where c(L₁) = p^m;
- (ii) p^k Z² ⊆ L₁, where k = min(r, s+1), π₁(L₂) = p^r and π₂(L₂) = p^s.

Primitivity means min(c(L₁), c(L₂)) = 1, and M ⊆ Λ if and only if r ≥ 1.

*Counting.*
- **m = 0.** Then s = 0 and r ≤ 1.
  - r = 0 forces L₁ = L₂ = Z². This is O, index 0.
  - r = 1 forces L₂ = pZ ⊕ Z, and L₁ ⊇ pZ² is primitive. With L₁ = Z² this is Λ itself (index 2); the p+1 index-p choices of L₁ give index 4.
- **m ≥ 1.** Then s = 0, m = 1 and L₁ = pZ².
  - r = 1: L₂ is pZ ⊕ Z (index 6), or one of the p−1 non-coordinate lines between p²Z ⊕ pZ and pZ ⊕ Z (index 8).
  - r = 2: L₂ = p²Z ⊕ Z (index 8).

So

  P_O = 1 + x² + (p+1)x⁴ + x⁶ + p x⁸,  P_Λ = 1 + (p+1)x² + x⁴ + p x⁶,

and Lemma 1 gives **Z_{P6,p} = (1 + (p+1)x² + x⁴ + (p+1)x⁶)/(1 − x⁸) = (1 + (p+1)x²)/(1 − x⁴).** ∎

**Q** (all eight maps carry a factor p). M is stable if and only if p·c(L₁)Z² ⊆ L₂ and p·c(L₂)Z² ⊆ L₁, and M ⊆ Λ if and only if L₂ ⊆ pZ².

*Counting.*
- c(L₁) = c(L₂) = 1: both lattices contain pZ² and are primitive, giving (1 + (p+1)x²)².
- One content is p: that lattice is pZ², and the other is a primitive lattice containing p²Z². The primitive lattices containing p²Z² number 1, p+1, p(p+1) at index 0, 1, 2. This case gives x⁴(1 + (p+1)x² + p(p+1)x⁴), twice.

Hence:
- P_O = 1 + 2(p+1)x² + ((p+1)²+2)x⁴ + 2(p+1)x⁶ + 2p(p+1)x⁸;
- P_Λ = 1 + (p+1)x² + p(p+1)x⁴;
- **Z_{Q,p} = N_Q(x)/(1 − x⁸)**, with N_Q = 1 + (p+1)x² + (p²+p+1)x⁴ + 2(p+1)x⁶ + (p²+2p+2)x⁸ + (p+1)x¹⁰ + p(p+1)x¹². ∎

**Checks (`proof_blocks.py`, `proof-blocks-check.json`):**
- the 64 block-map claims match the octonion multiplication table;
- the reduced (L₁, L₂) count, with a looser bound than the proof needs, reproduces the formulas at p = 3, 5, 7, 11, 13;
- every lattice built from a pair is confirmed to be a right ideal by direct octonion multiplication (10/14/18/26/30 for P6 and 59/123/211/459/619 for Q, at p = 3/5/7/11/13).

Agreement with the exhaustive programs:

| | p = 3 | p = 5 | p = 7 | p = 11 |
|---|---|---|---|---|
| P6 | both programs | both programs | both programs (method B finished ~7:02 PM PST) | not run to completion (a ~10-hour brute-force run was started; see §7) |
| Q | both programs | both programs (method B finished ~5:11 PM PST) | method A: 211 lattices, as the proof predicts (method B still running) | — |

For p = 11 the evidence is the proof itself plus the direct right-ideal check of every predicted lattice. No independent exhaustive run exists at 11 for these two shapes, so the "checked against computed data at 3, 5, 7, 11" standard is met only at 3, 5, 7.

## 4. Consequences for P6 and Q  [project, proved]

- **No zero on ½, any prime.**
  - P6: the zeros are x = ±i(p+1)^(−1/2), so Re s = log(p+1)/(2 log p) = ½ + log(1 + 1/p)/(2 log p) > ½.
  - Q: the resultant in x of N_Q(x) and p^12 x^12 N_Q(1/(px)) is p²⁶ (p+1)⁸ (p⁴+2)² (3p⁵+2p⁴+4p³+4p²+4p+2)² F(p)⁴, where F is a degree-22 integer polynomial with leading and constant coefficients 1. A zero on ½ would need this to vanish at p, but no factor has a prime root: an integer root must divide the constant term, which is ±1 or 2, and p = 2 fails too. So no zero lies on ½ for every prime p.
- **Mirror.** P6 has the trivial functional equation N(x) = (p+1)x² N(1/((p+1)x)) about Re s = log(p+1)/(2 log p). Q has no mirror (checked numerically at p = 3, 5).
- **Asymptotics.**
  - P6: Re s − ½ = log(1+1/p)/(2 log p) ~ 1/(2p log p).
  - Q, with y = w/p: N_Q = (1 + w + w²) + (w + w²)/p + O(p⁻²). So four zeros have w → a primitive cube root of unity ω, with |w| = 1 − 1/(2p) + O(p⁻²). Their real parts are **½ + 1/(4p log p) + …**, approaching ½ from the right. Numerically, (Re s − ½)·p log p = 0.2163, 0.2463, 0.25006 at p = 11, 101, 10⁶+3. The other eight zeros tend to Re s = 0.
- **Explicit-formula jumps.**
  - P6: e_{2j} = 2(−1)^(j+1)(p+1)^j + 4[4|2j] − 8[8|2j], and e_odd = 0, for every p.
  - Q: the weights follow from N_Q by the same Newton recursion (`analyze_p.py`). Their size grows like p^(k·max Re ρ), with max Re ρ → ½.

## 5. U, N4, P5  [project, partial proof here; superseded for N4 and P5 by `octonion-n4-p5-proof-20261111.md`]

**Proved for all p.**
- Lemmas 1–3: the reduction, the Peirce split M = Me ⊕ Mf, and the block maps.
- **Theorem 4 (the coefficient of x).** [x¹]Z_{Λ,p} equals the number of ring homomorphisms Λ → F_p. This is 2 when no pair u_i, v_i both lies in S (U, N4, P5), and 0 otherwise (P6, Q). This matches every computed case (p = 3, 5, 7, and 11 where computed).

  *Proof.* An index-p right ideal L contains pΛ. On the 1-dimensional quotient Λ/L, right multiplication by y acts as a scalar χ(y). So z·y − χ(y)z ∈ L for all z. Taking z = 1 gives L = ker χ. Applying χ again gives χ(zy) = χ(z)χ(y), so χ is a ring homomorphism. Conversely, the kernel of any homomorphism to F_p is such an ideal.

  χ(e) ∈ {0, 1} and χ(f) = 1 − χ(e). Every u- or v-element w is off-diagonal (ew = w = wf for u, and the mirror for v), which forces χ(w) = 0. So χ(a,u,v,b) is a mod p or b mod p. This is multiplicative exactly when u·v′ ≡ v·u′ ≡ 0 mod p on Λ, which holds if and only if no index i has both u_i and v_i in S. If such a pair exists, u_i v_i = e and v_i u_i = f would force χ(e) = χ(f), which is impossible. ∎

**The gap.** For U, N4, P5 no unscaled pair u_i, v_i exists, so there is no projection splitting A and B into 2-dimensional blocks. All available maps are nilpotent chains mod p; this matches the machine certificate, which found a nilpotent commutator ideal. So the classification is a problem about lattices that are stable under a nilpotent family of single-coordinate maps, some scaled by p, on two rank-4 lattices. We have not found a uniform classification. There are 4874/36010/148490 lattices for U at p = 3/5/7, versus 2p+4 for P6. **Exactly what is missing:** a uniform-in-p description, for instance as a cell decomposition into pieces counted by powers of p, of the primitive stable pairs (Me, Mf) for each of the three shapes. With that, the fitted numerators would follow from Lemma 1.

**Conditional results [project, proved given the fitted formulas, which remain project conjectures].**
- **No zero on ½ for any prime.** The resultants of N and p^d x^d N(1/(px)) factor as:
  - P5: p¹² (p⁸ + p⁷ + 3p⁶ + 4p⁵ + 5p⁴ + 11p³ + 16p² + 12p + 4)(4p⁹ + … − 2)²;
  - N4: 16p⁴⁰ (4p³ − p − 2)² (16p⁶ + … + 4)(2p⁹ − 19p⁸ + … − 16)²;
  - U: p⁵⁸ (p − 1)⁴ (degree-12 factor, constant 1)(degree-38 factor, constant −1)².

  An integer root of a factor divides its constant term (±1, 2, 4 or 16), so no odd prime is a root. Hence no zero lies on ½, for every odd prime, *if* the formulas hold.
- **Limits as p → ∞ (Newton polygon in p).**
  - P5: 2 zeros → Re s = 3/2, 1 → 1, 2 → −½.
  - N4: 2 → 1, **4 → ½**, 2 → −½. The four middle zeros straddle ½ in pairs: Re s = ½ ± 1/(4√2 p^(3/2) log p) + …. The constant was identified numerically: (Re s − ½) p^(3/2) log p → ±0.1768 = ±1/(4√2) for p up to 10⁸.
  - U: 2 → 3/2, 1 → 1, **2 → ½ from the right like 1/(2p log p)**, 4 → 0, 2 → −½.

## 6. What this settles / what stays open

**Settled for every odd prime [project, proved]:**
- the P6 and Q local factors;
- no zero on ½ for either;
- P6's line log(p+1)/(2 log p) > ½;
- the asymptotics above for P6 and Q;
- the coefficient of x for all five shapes.

**Open:**
- ~~the uniform classification for U~~ (N4 and P5 done at ~8:25 PM PST; U done at ~9:05 PM PST, pt216u);
- the diagonal order D;
- Z_p + pO for p ≥ 5;
- an independent exhaustive check of P6 and Q at p = 11;
- ~~deriving N4's ½ ± 1/(4√2 p^(3/2) log p) constant analytically~~ (done, `n4proof/pinch.py`).

## 7. Reproduce

    cd oddcond1111
    python3 proof_blocks.py      # block maps vs multiplication table; reduced counts p=3..13; direct ideal checks; comparison with exhaustive runs

The exhaustive P6 run at p = 11 (`genp-P6-11.txt`) was started at ~5:05 PM PST in the background. At about 23 minutes per lattice it may take ~10 hours. When it and the Q p = 7 runs finish, re-run `proof_blocks.py`, which compares them automatically. Update ~7:30 PM PST: Q p = 7 method A and P6 p = 7 method B have finished and `proof_blocks.py` reports both match; P6 p = 11 and Q p = 7 method B are still running.
