# Octonion orders with odd conductor at p = 5 and 7 (and some p = 11): same picture, and a polynomial-in-p pattern

Research, figure and page section, 11 November 2026, ~4:00–5:15 PM PST. Follows `octonion-odd-conductor-20261111.md` (p = 3). Scripts and results are in `oddcond1111/`. No messages sent. No RH claim: everything here is about finite local factors at one odd prime p.

## Plain explanation for Grant

At 3 we damaged the octonion order on purpose and found local zeros, but none on Re s = ½. Now we repeat the same five shapes at p = 5 and p = 7, and for two of them at p = 11.
- **Same verdict at every prime [project, exact test].** No zero on Re s = ½, for every shape and every p computed. There is no mirror symmetry either, except the trivial P6 case, where all the zeros sit on a single line.
- **The counts follow a pattern [project conjecture].** Each coefficient of the local polynomial looks like a fixed polynomial in p. For N4 and P5 we fitted that polynomial from p = 3, 5, 7, then predicted p = 11 and computed it: every coefficient matched. For U the fit is from three primes only and is untested.
- **The zeros creep toward ½ but never land on it [project observation].** For P6 the real part is exactly log(p+1)/(2 log p): 0.631, 0.557 and 0.534 at p = 3, 5, 7. It moves toward ½ but is never ½. For N4 there are zeros on both sides of ½ (0.4986 and 0.5035 at p = 7) that close in as p grows. The exact test still says none is on the line.
- **What was not finished, and what came in later.** At first publication Q at p = 7 was still running, and P6 at p = 7 had only one program. Both have since come in (update 11 Nov, ~7:30 PM PST): Q at p = 7 by method A gives 211 lattices and numerator 1, 8, 57, 16, 65, 8, 56, exactly what the general-p proof predicts; P6 at p = 7 by method B gives 18 lattices, agreeing with method A. Method B for Q at p = 7 is still running. Z_p + pO at p = 5, 7 was not computed; only its general structure is stated (§5).

## 1. Shapes  [definitions as at p = 3]

For odd p, E₈ ⊗ Z_p ≅ Zorn(Z_p) (split octonions). Λ_S is the preimage of the coordinate subalgebra S of Zorn(F_p). The shapes, with coordinates (a, u₁,u₂,u₃, v₁,v₂,v₃, b) indexed 0–7:
U = (0,1,7), Q = (0,1,4,7), N4 = (0,1,5,7), P5 = (0,1,2,6,7), P6 = (0,1,2,4,6,7). Index of Λ_S in O: p^(8 − dim S). We count one-sided ideals (lattices L ⊆ Λ with L·Λ ⊆ L) by additive index, exactly as at 3: Z_{Λ,p}(x) = P_Λ(x) + x^(dim S) P_O(x)/(1 − x⁸), x = p^(−s).

## 2. Two independent programs  [project]

- **Method A, `genp.py`:** Zorn model, HNF over Z/p⁷, BFS over primitive stable lattices. Mode "eig" (common eigen-functionals) is used only when a certificate holds. The certificate: the commutator ideal of the multiplication algebra mod p is nilpotent and the algebra is split, so all simple modules are 1-dimensional and maximal submodules have codimension 1. It held for U, N4, P5 at every p. Q and P6 failed it (c = 2), so they used mode "brute" (all projective functionals, word basis).
- **Method B, `method_Bp.py`:** Cayley–Dickson model, frame built from w with N(w) ≡ −1 mod p⁶, all 64 structure constants checked, Howell canonical forms over Z/p⁶, and the left action instead of the right (by conjugation, left count = right count). Mode "hyp" (intersections of eigenspaces) is used for certified shapes and "cyc" (cyclic closure of every functional) otherwise. A dual-action convention bug in "cyc" (Xᵀ instead of X) was found by the P6 p = 3 regression and fixed; it then reproduced P6 and Q at p = 3 exactly.

Agreement (`compare-p.json`). P_O and P_Λ are identical, as is the number of primitive lattices:

| shape | p = 3 | p = 5 | p = 7 | p = 11 |
|---|---|---|---|---|
| U | 4874, A = B (B finished 2:16 PM PST) | 36010, A = B | 148490, A = B | — |
| N4 | 1018, A = B | 5146, A = B | 16298, A = B | 82714, A only (tests the pattern) |
| P5 | 502, A = B | 2550, A = B | 8102, A = B | 41238, A only (tests the pattern) |
| Q | 59, A = B | 123, A = B | 211, A only (B still running); matches the proved formula | — |
| P6 | 10, A = B | 14, A = B | 18, A = B (B finished ~7:02 PM PST) | — |

All six p = 3 orders are now double-computed: U's method B result matches method A exactly.

## 3. Local factors and zeros  [project; exact counts, exact gcd tests; zeros by 40-digit mpmath]

Here x = p^(−s), and the denominator is 1 − x⁴ (Q: 1 − x⁸).

| shape | p | numerator N(x) coefficients (x⁰ upward) | zeros on ½ | Re range |
|---|---|---|---|---|
| U | 3 | 1, 2, 43, 137, 199, 556, 241, 382, 192, 468, 108, 108 | 0 of 11 | -0.5740 … 1.6953 |
| U | 5 | 1, 2, 161, 817, 1091, 4986, 1261, 2456, 1080, 4650, 750, 750 | 0 of 11 | -0.5398 … 1.5911 |
| U | 7 | 1, 2, 407, 2865, 3599, 23192, 4033, 8730, 3584, 22344, 2744, 2744 | 0 of 11 | -0.5265 … 1.5575 |
| Q | 3 | 1, 0, 4, 0, 13, 0, 8, 0, 17, 0, 4, 0, 12 | 0 of 12 | -0.0210 … 0.5386 |
| Q | 5 | 1, 0, 6, 0, 31, 0, 12, 0, 37, 0, 6, 0, 30 | 0 of 12 | -0.0112 … 0.5220 |
| Q | 7 | 1, 0, 8, 0, 57, 0, 16, 0, 65, 0, 8, 0, 56 | 0 of 12 | -0.0073 … 0.5145 |
| N4 | 3 | 1, 2, 25, 26, 35, 78, 234, 54, 54 | 0 of 8 | -0.6310 … 1.4279 |
| N4 | 5 | 1, 2, 61, 62, 87, 310, 1550, 250, 250 | 0 of 8 | -0.5565 … 1.2670 |
| N4 | 7 | 1, 2, 113, 114, 163, 798, 5586, 686, 686 | 0 of 8 | -0.5342 … 1.2102 |
| N4 | 11 | 1, 2, 265, 266, 387, 2926, 32186, 2662, 2662 | 0 of 8 | -0.5181 … 1.1619 |
| P5 | 3 | 1, 2, 40, 130, 39, 39 | 0 of 5 | -0.5148 … 1.7048 |
| P5 | 5 | 1, 2, 156, 806, 155, 155 | 0 of 5 | -0.5024 … 1.5927 |
| P5 | 7 | 1, 2, 400, 2850, 399, 399 | 0 of 5 | -0.5007 … 1.5580 |
| P5 | 11 | 1, 2, 1464, 16226, 1463, 1463 | 0 of 5 | -0.5002 … 1.5320 |
| P6 | 3 | 1, 0, 4 | 0 of 2 | 0.6309 … 0.6309 |
| P6 | 5 | 1, 0, 6 | 0 of 2 | 0.5566 … 0.5566 |
| P6 | 7 | 1, 0, 8 | 0 of 2 | 0.5343 … 0.5343 |

**On-line test (exact).** A zero on Re s = ½ means |x|² = 1/p, which would make it a common root of N(x) and x^d N(1/(px)). The gcd over Q is 1 in every case, so **no zero lies on ½** [project, exact]. **Mirror test:** the real parts are not symmetric about any vertical line, except P6, whose zeros share one real part log(p+1)/(2 log p). For P6 there is a trivial functional equation N(x) ↔ x²N(1/((p+1)x)) about that line, never about ½ [project, exact].

## 4. Polynomial-in-p pattern  [project conjecture unless stated]

Fitted from p = 3, 5, 7, choosing the smallest integer coefficients. The base-p digits agree at p = 5 and 7, which is a non-trivial check. See `pattern_p.py` and `pattern-p.json`.
- **P6:** N = 1 + (p+1)x². Also P_O = 1 + x² + (p+1)x⁴ + x⁶ + p x⁸. [conjecture; data p = 3, 5, 7]
- **P5:** N = 1 + 2x + (p³+p²+p+1)x² + (p⁴+p³+2p²+p+1)x³ + (p³+p²+p)(x⁴+x⁵). **Predicted p = 11 and computed: all coefficients match.** [conjecture, confirmed at a fourth prime] **Update ~8:25 PM PST (pt212u): now [project, proved] for every odd prime; see `octonion-n4-p5-proof-20261111.md`.**
- **N4:** N = 1 + 2x + (2p²+2p+1)x² + (2p²+2p+2)x³ + (3p²+2p+2)x⁴ + (2p³+2p²+2p)x⁵ + (2p⁴+2p³+2p²)x⁶ + 2p³(x⁷+x⁸). **Confirmed at p = 11.** [conjecture, confirmed at a fourth prime] **Update ~8:25 PM PST (pt212u): now [project, proved] for every odd prime; see `octonion-n4-p5-proof-20261111.md`.**
- **U:** N = 1 + 2x + (p³+p²+2p+1)x² + (p⁴+p³+2p²+3p+2)x³ + (p⁴+3p³+3p²+3p+1)x⁴ + (p⁵+2p⁴+4p³+4p²+2p+1)x⁵ + (p⁴+4p³+5p²+2p+1)x⁶ + (3p⁴+4p³+3p²+p+1)x⁷ + (p⁴+3p³+3p²+p)x⁸ + (p⁵+2p⁴+2p³+p²)x⁹ + (p⁴+p³)(x¹⁰+x¹¹). [conjecture from three primes; untested at 11, where we estimate about 9×10⁵ lattices] **Update ~9:05 PM PST (pt216u): now [project, proved] for every odd p — see `octonion-u-proof-20261111.md` (52 two-step cells via the F_p[t]/t² look-alike count and a carry check).**
- **Q (p = 3, 5 only):** the even coefficients fit 1, p+1, p²+p+1, 2(p+1), p²+2p+2, p+1, p(p+1). [speculative: two primes only]

Observations [project]: the coefficient of x is 2 for U, N4 and P5 at every p. As p grows, the P5 extreme zeros tend toward Re s = −½ (−0.5148, −0.5024, −0.5007, −0.5002), and the P6 and N4 near-½ zeros approach ½ without reaching it. If the P6 pattern holds, log(p+1)/(2 log p) > ½ for every p, so P6 never touches ½. That last step is proved: p + 1 > p.

## 5. Z_p + pO  [project, proved structure; not computed at 5, 7]

The p = 3 argument works for any odd p:
- L·Λ ⊆ L ⇔ p·span(L·O) ⊆ L.
- If M mod p contains an element x with N(x) ≢ 0, then x is invertible in O_p, x·O = O (Moufang), and so pO ⊆ M.
- Otherwise W = M mod p is totally singular, and M ↔ (K, lift of W) with M ∩ pO = pK.

So P_O = C₁ + C₂, with C₁ = Σ_W [8−t, k−t]_p p^(w(8−k)) x^(16−k−w) over totally singular W (t = dim span(W·O)), plus Gaussian binomials for non-singular W, and C₂ the deep lattices over isotropic lines. Isotropic lines ℓ = ⟨x⟩ always have t = 4 (x·O is 4-dimensional for null x). The full t-distribution over all totally singular W, and the deep-lattice count (81 = 3⁴ at p = 3; p⁴ in general is a guess, [speculative]), were computed only at p = 3. So we give no closed formula for p = 5, 7.

## 6. Explicit-formula jumps at powers of p  [project, exact from the local polynomials]

−Z′/Z has p-part log p · Σ_k w_k p^(−ks). The extra weights e_k = w_k − 8·[8 | k] relative to the maximal order are, for k = 1..8:

| shape | p | e₁ … e₈ |
|---|---|---|
| U | 3 | 2, 82, 161, -3322, -9123, 123223, 526066, -4547506 |
| U | 5 | 2, 318, 1493, -51450, -374523, 7464015, 84208350, -1015842882 |
| U | 7 | 2, 810, 6161, -333322, -4052763, 123180567, 2317186482, -42104198898 |
| Q | 3 | 0, 8, 0, 20, 0, -136, 0, 356 |
| Q | 5 | 0, 12, 0, 52, 0, -612, 0, 2212 |
| Q | 7 | 0, 16, 0, 100, 0, -1616, 0, 7492 |
| N4 | 3 | 2, 46, -64, -930, 2592, 19468, -80890, -365074 |
| N4 | 5 | 2, 118, -172, -6626, 17812, 377884, -1454206, -20798130 |
| N4 | 7 | 2, 222, -328, -24002, 63432, 2618940, -9799074, -280634578 |
| N4 | 11 | 2, 526, -784, -136802, 355312, 35735212, -130519114, -9264981906 |
| P5 | 3 | 2, 76, 158, -3456, -9163, 133144, 540906, -5087608 |
| P5 | 5 | 2, 308, 1490, -52016, -376183, 7618496, 85096986, -1047643704 |
| P5 | 7 | 2, 796, 6158, -334816, -4060963, 124181224, 2327485106, -42622579128 |
| P5 | 11 | 2, 2924, 39902, -4387136, -97082683, 5965207736, 201321711642, -7554083925240 |
| P6 | 3 | 0, 8, 0, -28, 0, 128, 0, -516 |
| P6 | 5 | 0, 12, 0, -68, 0, 432, 0, -2596 |
| P6 | 7 | 0, 16, 0, -124, 0, 1024, 0, -8196 |

The P6 weights follow from N = 1 + (p+1)x²: e_{2j} = 2(−1)^(j+1)(p+1)^j + 4·[4 | 2j] − 8·[8 | 2j], and e_odd = 0 (proved wherever the P6 formula holds). The jump sizes grow like p^(k·max Re ρ). They come from the local polynomial, not from zeta zeros.

## 7. What this settles / what stays open

- **Settled at p = 3, 5, 7 for these shapes and this ideal notion:**
  - no local zero on ½;
  - no mirror (P6 trivial, about a line ≠ ½);
  - the odd-prime explicit-formula signal persists.
- **Open:**
  - prove the polynomial-in-p formulas, perhaps by a uniform cell decomposition of the lattice poset;
  - B-check Q at p = 7 (running);
  - Z_p + pO at p ≥ 5;
  - the diagonal order D;
  - other ideal notions.

- **Follow-up (Nov 11, ~5:10 PM PST):** P6 and Q are now proved for every odd prime, and U, N4, P5 have a partial proof; see `octonion-odd-conductor-general-p-20261111.md`.

## Reproduce

    cd oddcond1111
    python3 genp.py SHAPE p            # method A (auto mode from certificate)
    python3 method_Bp.py SHAPE p hyp   # method B, certified shapes (U, N4, P5)
    python3 method_Bp.py SHAPE p cyc   # method B, Q and P6
    python3 compare_p.py && python3 analyze_p.py && python3 pattern_p.py && python3 figure_p.py
