# Every order between the Gravesian octonions and E8: ideal counts, zeros, and a Gorenstein test

Sprint run, Wednesday 11 November 2026, 7:24–7:55 AM PST (box clock). Scripts, JSON and run logs are in `inter1111/` beside this note. No messages sent.

## Summary
- **[project, computed exactly]** There are exactly **16 orders containing G = Z^8** (basis 1, e1..e7, Fano lines {i, i+1, i+3} mod 7, the same table as every earlier script): G itself, **one** order S1 of index 2, **seven** orders of index 4 (R and its six images), and **seven** maximal (Coxeter/E8-type) orders of index 16. None have index 8. Up to the 1344 signed-permutation automorphisms of the octonion algebra that fix Z^8, there are four classes: {G}, {S1}, {the 7 index-4 orders}, {the 7 maximal orders}.
- **[project]** One correction to the Block 4 note: R is **not** a direct step up from G. R = Z⟨G, (1+e1+e2+e4)/2⟩ already contains ω = (1+e1+…+e7)/2, so the chain is G ⊂ S1 ⊂ R ⊂ O (indices 2, 2, 4). The orders strictly between G and any maximal O ⊇ R are S1 plus three index-4 orders, all conjugate to R under the stabiliser of O (order 192). R lies in exactly 3 of the 7 maximal orders (#12, #13, #15). So the only genuinely new intermediate order is **S1**.
- **[project, proved — computer-assisted, double-computed]** S1's 2-local left-ideal count (right gives the same) is
  **P_S1(x) = (1 + 2x + 15x² + 35x³ + 14x⁴ + 14x⁵)/(1 − x⁴)**, with x = 2^{-s}. The numerator Q_S1 is irreducible over Q, and Q_S1(1) = 81.
- **[project, exact]** S1 has **no functional equation** of any shape Q(x) = C·x⁵·Q(a/x). gcd(Q_S1, Q_S1*) = 1 with Q*(x) = (2x)⁵Q(1/(2x)), so **no zero lies on Re s = ½** (exact), and no zeros pair up as mirror images across ½. Its five zeros per period sit at Re s ≈ −0.5670 (a conjugate pair), 1.0551 (one zero at Im s = π/log 2) and 1.9431 (a pair) [project, numerical: PARI 60 digits and numpy agree to 1e-15].
- **[project, proved — computer-assisted]** Re-derived with the new code: G gives Q_G/(1−x⁴), with FE and 4 zeros on ½; R and a conjugate of it give (1+3x²)/(1−x⁴); the maximal order gives 1/(1−x⁸) (only 2^m O, consistent with [known] Petersson 2018).
- **[speculative adaptation, computed exactly]** An octonion version of Voight's Gorenstein test (codifferent locally principal, Def. 24.2.1 / Lemma 24.2.5): **G yes, S1 no, R yes, maximal yes**. In this adapted sense only R and the maximal orders are "Bass". So **R is Gorenstein-type yet has no ½-symmetric functional equation**, which refutes the Block 4 guess that R would fail Gorenstein. And S1 contains a copy of Z_2 × Z_2 but is still not Gorenstein-type. That breaks the quaternion implication "basic ⇒ Bass ⇒ Gorenstein" (Voight Prop. 24.5.8) once it is carried over to octonions.
- No RH content: all of these are finite Euler factors at the single prime 2.

## 1. Enumeration of the orders  [project, computed exactly]
- Any order S ⊇ G lies in ½G, because trace and norm integrality of x·e_i force 2x ∈ Z^8. So S = G + V/2 for an F_2-subspace V ⊆ F_2^8. The norm-form Gram determinant is 2^{8−2·dim V}, so dim V ≤ 4.
- `inter1111/enum_orders_full.py` tests all 309,193 subspaces of dimension ≤ 4 for ring closure, checking products G·V, V·G and V·V exactly. Run log: `enum-orders-full-run.txt`. Result: 16 orders, with dim V = 0, 1, 2, 4 occurring 1, 1, 7, 7 times.
- `inter1111/structure.py` (log `structure-run.txt`, data `structure.json`) gives the containments. Every index-4 order contains S1 and lies in 3 maximal orders, and every maximal order contains 3 index-4 orders, like the points and lines of a Fano plane. Aut(G) here is the 1344 signed permutations (= 2³·168) that preserve the table. Its orbits on the 16 orders are {G}, {S1}, {7 index-4}, {7 maximal}.
- Generators: S1 = G + Zω, ω = (1+e1+…+e7)/2, N(ω) = 2, t(ω) = 1, ω² − ω + 2 = 0 (checked, `quadratic-subrings-run.txt`). R = S1 + Z·h, h = (1+e1+e2+e4)/2, h² − h + 1 = 0.

## 2. Ideal counts, two independent methods  [project, double-computed]
Notion of ideal: exactly as in `octonion-ideal-count.md` and `block4/`. A **left ideal of 2-power index is an additive sublattice L ⊆ S with x·L ⊆ L for every x ∈ S**, i.e. stable under all left multiplications L_x in S's own Z-basis. Right ideals use R_x. a_k = #{L : [S:L] = 2^k}.

- **Method 1** (`method1_pari.py`, PARI HNF). The children of each stable L are all stable N with 2L ⊆ N ⊊ L, found from the full submodule lattice of L/2L as sums of cyclic submodules. Complete by Nakayama. This is the Block 4 "full" algorithm, rewritten to take any order.
- **Method 2** (`method2_dual.py`, pure Python with its own HNF and no PARI). The children are kernels of quotient modules of dimension ≤ DMAX, found as invariant subspaces of the dual (L/2L)*. It is complete because a maximal stable sublattice N ⊂ L has L/N simple, hence killed by 2 and of dimension equal to a composition factor of S/2S (Brauer–Nesbitt: every lattice has the same factors). Method 1 prints those factors: G and S1 have eight copies of F_2 (DMAX = 1); R has four 2-dimensional factors (DMAX = 2); the maximal order has a single 8-dimensional simple factor (method 2 not run there).
- **Method 3** (spot check, `bruteforce_small.py`). Every one of the 255 index-2 and 43,435 index-4 sublattices of S1 was tested directly: 2 and 15 are stable, left and right, which agrees with methods 1 and 2.

| order | side | method 1 | method 2 | a_k, k = 0, 1, 2, … |
|---|---|---|---|---|
| G | left | k ≤ 26 | k ≤ 26 | 1, 1, 7, 7, 15, 15, 35, 15, 31, then period 4 (15, 35, 15, 31) — same as Block 4 |
| S1 | left | k ≤ 24 | k ≤ 24 | **1, 2, 15, 35, 15, 16, then period 4: 15, 35, 15, 16** |
| S1 | right | k ≤ 24 | k ≤ 24 | identical to left |
| R | left / right | k ≤ 16 (right) | k ≤ 16 (left and right) | 1, 0, 3, 0, 1, 0, 3, 0, … — same as Block 4 |
| R′ (order #2, conjugate) | left | — | k ≤ 16 | identical to R |
| O (#12, maximal) | left | k ≤ 16 | — | 1 at k ≡ 0 mod 8, else 0 |

Logs: `m1-*-run.txt`, `m2-*-run.txt`, JSON `m1-*.json`, `m2-*.json`.

**Why the counts determine the factor exactly** [project, proved — computer-assisted]. `conductor.py` (log `conductor-run.txt`) closes the span of words in the L_x to an algebra A_S and finds [M_8(Z):A_S] and the least c with 2^c·M_8(Z) ⊆ A_S. Results: G 2^96, c = 3 (as in Block 4); **S1 2^68, c = 2**; R 2^40, c = 1; O index 1, c = 0. Left and right agree. Then, exactly as in `gravesian-periodicity-proof.md` Step 3: a stable L ⊄ 2S contains 2^c·S, so primitive stable lattices have index ≤ 2^{8c}, and P(x) = Π(x)/(1−x⁸) with Π of degree ≤ 8c. For S1, 8c = 16 and the counts reach k = 24. The primitive counts are p_k = 1, 2, 15, 35, 15, 16, 15, 35, 14, 14, then 0 for 10 ≤ k ≤ 24, which gives Π = Q_S1·(1+x⁴) and P = Q_S1/(1−x⁴) (`analyse.py`, `analyse-run.txt`, `analysis.json`).

**Period 4 has the same cause as for G** [project, computed exactly, `normaliser_S1.py`]. For each k = 1..7, g = L_{1+e_k} lies in A_{S1}, normalises it (gA g⁻¹ = A, by equal 64-dimensional HNFs), and det g = 16. Alternativity gives g² = 2·L_{e_k}, so stable lattices form chains L ⊃ gL ⊃ 2L with index step 2⁴. Q_S1(1) = 81 counts the chains.

## 3. Functional equations and zeros  [project, exact + numerical]
x = 2^{-s}, and Re s = ½ ⇔ |x| = 2^{-1/2}. A zero on Re s = ½ is a common root of Q and Q*(x) = (2x)^{deg}Q(1/(2x)), so gcd(Q, Q*) = 1 rules out on-line zeros exactly.

| order | local factor (2-part) | FE of G type, Q* ∝ Q (s ↦ 1−s) | zeros on Re s = ½ (per period) | other zeros (Re s) |
|---|---|---|---|---|
| G | Q_G/(1−x⁴), deg 8 | **yes** | 4 (Block 3 Sturm-exact) | 2 at 1.1834, 2 at −0.1834 (mirror pairs) |
| S1 | Q_S1/(1−x⁴), deg 5 | **no** (gcd = 1; no a, C give Q = C x⁵ Q(a/x)) | **0** (exact) | 2 at −0.5670, 1 at 1.0551, 2 at 1.9431 |
| R (×7) | (1+3x²)/(1−x⁴) | no s ↦ 1−s FE (gcd = 1) | **0** (exact) | 2 at log 3/log 4 ≈ 0.7925 |
| O (×7) | 1/(1−x⁸) | trivially | none | none (the ζ(8s) zeros are inherited) |

Fine print [project]: 1+3x² satisfies Q(x) = 3x²·Q(1/(3x)), a reflection about Re s = log_4 3 instead of ½. This is automatic for a polynomial in x² of degree 1, whose roots all share one modulus, so it does not count as a G-type equation. For Q_S1 no reflection of any centre works: matching coefficients gives a = ½ from the x¹ term but a² = 1/6 from the x² term. Root positions: `roots_check.py`, `roots-check-run.txt`.

## 4. Gorenstein test
**[known] The associative definition.** Voight, *Quaternion Algebras* (GTM 288, 2021; local copy `literature/sources/voight-quaternion-algebras.pdf`), §24.2: codiff(O) = O^♯ = {α ∈ B : trd(αO) ⊆ R}. **Def. 24.2.1:** O is Gorenstein if codiff(O) is invertible. **Lemma 24.2.5:** nrd(codiff O)·discrd(O) ⊆ R, with equality iff O is Gorenstein, i.e. iff O^♯ is locally principal. **Def. 24.5.1:** O is Bass if every superorder is Gorenstein. **Def. 24.5.6 / Prop. 24.5.8** (local R; Chari–Smertnig–Voight 2019; Brzezinski 1990 when 2 is a unit): O is basic, meaning it contains the maximal order of a maximal commutative subalgebra, iff O is Bass. Brzezinski's papers were not read directly. Voight is the primary source used here.

**[speculative] The octonion adaptation.** "Invertible" has no standard meaning for lattices in a nonassociative algebra: (I·J)·K ≠ I·(J·K) in general, and there is no group of fractional ideals. So I used the "locally principal" form of the criterion, which needs only one-sided multiplication by a single element:
- codiff(S) = S^♯ = {x : t(xs) ∈ Z for all s ∈ S}, with t = 2·Re. Because t((ab)c) = t(a(bc)) in an octonion algebra, S^♯ is a left and right S-module (proved in one line; also checked on all basis triples).
- **Adapted Gorenstein (test P):** S^♯ ⊗ Z_2 = S_2·α for some α. For octonions vol(S·α) = N(α)⁴·vol(S), the right-multiplication determinant, where the quaternion case has nrd(α)². So the test is: some α ∈ S^♯ has N(α)⁴·[S^♯:S] a 2-adic unit. It is exact by enumerating α mod 2S^♯, because 2^{d/4}·N is integral on S^♯ and its parity depends only on α mod 2S^♯.
- **Test N** (the Lemma 24.2.5 analogue): n(S^♯)⁴·[S^♯:S] = 1, where n is the Z-module generated by the norms. On every order tested, P and N gave the same answer.
- Calibration on quaternions, using the same code with exponent 2: Lipschitz Z⟨i,j⟩ yes, Hurwitz yes, Z + 2·Hurwitz no, Z + 2·Lipschitz no. This matches Lemma 24.2.5 and Voight Prop. 24.1.4 (Z + f·Gor(O) with f > 1 is not Gorenstein).

**Results** [speculative adaptation; computations exact; `gorenstein.py`, `gorenstein-run.txt`, `gorenstein-witness-run.txt`]:

| order | [S^♯:S] | n(S^♯) | Gorenstein-type? | generator |
|---|---|---|---|---|
| G | 2^8 | 2^{-2} | yes | G^♯ = ½·G |
| S1 | 2^6 | 2^{-1} | **no**: 2^6 is not a 4th power, so no α can work, even 2-adically | — |
| R and its 6 conjugates | 2^4 | 2^{-1} | **yes** | R^♯ = R·α = α·R, α = (e1+e2)/2, N(α) = ½ |
| maximal (×7) | 1 | 1 | yes | α a unit |
| Z + 2·O (control) | 2^14 | 2^{-2} | no | — |

Bass (adapted): the maximal orders yes; **R yes**, since its superorders R and the 3 maximal orders above it all pass; S1 no; G no, because its superorder S1 fails.

## 5. Where non-associativity changes the outcome (the "tiny ways")  [project, labels as stated]
1. **The volume exponent is 4, not 2.** A principal codifferent forces [S^♯:S] to be a 4th power. For S1 it is 2^6, a square but not a 4th power, so S1 fails for that reason alone. A quaternion order's discriminant is always a square, so this kind of obstruction cannot occur there. [project, exact]
2. **Basic does not imply Bass.** Z_2[ω] ≅ Z_2 × Z_2 (x² − x + 2 splits mod 2), so S1 contains the maximal order of a split quadratic subalgebra. In the quaternion case the relevant commutative subalgebras are quadratic, and here I take quadratic étale subalgebras as the octonion analogue of "maximal commutative subalgebra" [speculative]. Over Q_2 the octonion algebra is split and has larger commutative subalgebras containing nilpotents, so this choice is part of the adaptation. In quaternions, Prop. 24.5.8 would then make the order Bass, hence Gorenstein; S1 is not Gorenstein-type. G also contains Z_2[e1] = Z_2[i], the maximal order of Q_2(i), yet G is not Bass-type. Compare the Lipschitz quaternions, the quaternion analogue of G, which are basic, Bass and Gorenstein. [speculative, because it depends on the adaptation]
3. **Gorenstein-type does not give a ½-line equation.** R passes the adapted test, but its factor has no s ↦ 1−s symmetry and no zeros on ½. In the quaternion Eichler family the project had Gorenstein/Bass structure together with a palindromic factor and all local zeros on ½. For octonion orders that link is broken in both directions: G has the FE and passes the test, R passes and lacks the FE, S1 fails and lacks it. [project + speculative adaptation] Even in the associative case, Gorenstein ⇒ local FE is not a theorem for Solomon-type counts (see `order-eichler-r4.md`), so this does not refute anything known.
4. Unchanged from the quaternion picture: the period-4 chain is driven by a norm-2 element 1+e_k whose left multiplication squares to 2·unit (alternativity), for both G and S1. [project, exact]

## Status / not done
- Exact, computer-assisted and double-computed: the enumeration, all four factors, the zero/no-zero statements on ½, and the Gorenstein-type computations. Not done: an independent re-check by a second agent; a conceptual (non-computational) proof that A_{S1} has c = 2; Sturm-exact counts for the off-line S1 zeros (positions are numerical only, though gcd = 1 makes "none on ½" exact).
- The Gorenstein adaptation is not from the literature. I found no source defining Gorenstein orders in octonion algebras. Petersson's notes (`octonion-sources/`) treat only maximal orders.
- The Fano-plane incidence of the 7 index-4 orders against the 7 maximal orders was observed computationally and not related to a known source. Classically there are 7 maximal orders containing the Gravesian octonions [known, Coxeter 1946 / Conway–Smith ch. 9, recalled, not re-read]. That count matches.

## Reproducing
Run from `inter1111/` with `/workspace/tools/pari-venv/bin/python` (it has cypari2 and numpy), in this order: `enum_orders_full.py`, `structure.py`, `conductor.py`, `method1_pari.py <order#> <K> [left|right]`, `method2_dual.py <order#> <K> <DMAX> [side]`, `bruteforce_small.py 1`, `analyse.py`, `roots_check.py`, `normaliser_S1.py`, `gorenstein.py`, `gorenstein_witness.py`. The figure is `fig_octonion_orders_3.py`, run with system `python3` (matplotlib) from `hyper-critical/`; the previous figure is kept in `inter1111/figure-backup/`. Order numbers: 0 = G, 1 = S1, 8 = R, 12 = a maximal order containing R. Total compute was about 8 minutes.

Deployed 11 Nov 2026, 7:45 AM PST, as version 7b2321eb-42e7-4ef1-8e84-693cb05ea9cd, anchor `/critical-line-hypercomplex#octonion-intermediate-orders`.
