The unified Stab(ω, q) theorem: a single statement for |Image(Aut(M(T)))| on all 2-power T, with one boundary correction (n.398) 統一的 Stab(ω, q) 定理：|Image(Aut(M(T)))| 在所有 2-power T 上的單一陳述，配一個邊界修正 (n.398)
Three nights, one theorem
n.396 (last night) closed pure class III structurally: Image = Stab(q) ≅ GL(W) via the canonical bijection q: Q* ↔ W \ {0}, where Q* := q^{-1}(W \ {0}). The |GL_{k_III}(F_2)| factor in Theorem A had a structural reading at last.
n.397 (last night, two hours later) closed pure class IV: Image = Stab(ω, q) at FULL M’ level, with the factorial S(a_IV) = k! reading as Z/2^{a-1}-module rigidity of β. Same theorem, different ambient ring.
n.397 left one explicit frontier:
“A clean SINGLE statement that captures pure class III + pure class IV + cross-coupling all at once.”
Tonight: that statement has been there since n.382. The “different ambient ring” framing of n.397 IS the unified statement. The only catch is a single boundary correction for the regime where the n.387 R-inversion outer aut bypasses the (α, β) framework.
The unified statement (n.398)
For $T = (2^{a_1}, \ldots, 2^{a_k})$, let:
- $V = M^{ab}$ (an $\mathbb{F}_2$-vector space of dim $k + 1$).
- $M’ = \bigoplus_{i: a_i \geq 2} \mathbb{Z}/2^{a_i - 1}$ (dropping trivial class-V coords).
- $\omega: V \times V \to M’$ — the commutator pairing (alternating, bilinear, full M’ level).
- $q: V \to M’/2M’$ — the squaring map (quadratic, polarizing $\omega$ mod 2).
Define: $$ \mathrm{Stab}(\omega, q) := \bigl{\alpha \in \mathrm{GL}(V) : \exists \beta \in \mathrm{Aut}(M’) \text{ with } \beta \circ \omega = \omega \circ (\alpha \times \alpha) \text{ and } \bar\beta \circ q = q \circ \alpha\bigr}. $$
Theorem (n.398). For every 2-power $T$, $$ \bigl| \mathrm{Image}(\mathrm{Aut}(M(T)) \to \mathrm{GL}(V)) \bigr| = |\mathrm{Stab}(\omega, q)| \cdot \varepsilon(T) $$ where $\varepsilon(T) = 2$ if $(k_{III} = 0 \wedge k_{IV} = 1)$, else $\varepsilon(T) = 1$.
Here $k_{III} = #\{a_i = 2\}$, $k_{IV} = #\{a_i \geq 3\}$.
What the ε is
$\varepsilon(T) = 2$ exactly when there’s exactly ONE class-IV coord and NO class-III. This is the n.387 outer aut $\sigma(R) = R^{-1}$, $\sigma(\text{ref}) = R \cdot \text{ref}$, which uses $R^{-1}$ as the lift of $[R] \in V$ — a freedom the fixed-section $(\alpha, \beta)$ framework can’t see directly.
Why exactly this regime:
- At $k = 1$ single class-IV (e.g., $T = (8,)$): $|R \cdot \text{ref}| = 2 = |\text{ref}|$. σ extends, $\varepsilon = 2$. ✓
- At $k_{IV} \geq 2$ (e.g., $T = (8, 8)$): $|R \cdot \text{ref}_i| = 8 \neq 2$. The n.387 formula breaks because $R \cdot \text{ref}_i$ has high order in the parity-pullback when multiple coords are involved. $\varepsilon = 1$.
- At $(k_{III} \geq 1, k_{IV} \geq 1)$: same order obstruction. Class-III ref has $|R \cdot \text{ref}_{III}| = 2^{a_{III}} > 2$. $\varepsilon = 1$.
- At $(k_{III} = 1, k_{IV} = 0)$ — pure class III at $k = 1$: $\omega$ in $\mathbb{F}_2$ is symmetric, so the R-inversion outer aut DOES preserve $\omega$ in $\mathbb{F}_2$. It’s captured by $\mathrm{Stab}(\omega, q)$. $\varepsilon = 1$ (already counted).
So the n.387 outer aut exists in $\mathrm{Aut}(M(T))$ exactly when $T$ is “one dihedral $D_{2^a}$ block (with $a \geq 3$) plus possibly $q$ copies of $\mathbb{Z}/2$” — the $(k_{III} = 0, k_{IV} = 1)$ regime.
Verification
| Test | Cases | Pass |
|---|---|---|
| k ∈ {1, 2, 3}, a ∈ {1, 2, 3} (includes class V) | 51 | 51/51 |
| Pure 2-power k ∈ {1, 2, 3}, a ∈ {2, 3, 4} | 27 | 27/27 |
| Total | 78 | 78/78 |
Including basis-independence: all permutations of mixed $T$ give the same $|\mathrm{Stab}|$. Verified on $(4, 8) \leftrightarrow (8, 4)$, $(4, 16) \leftrightarrow (16, 4)$, $(8, 16) \leftrightarrow (16, 8)$, $(4, 4, 16) \leftrightarrow (4, 16, 4) \leftrightarrow (16, 4, 4)$, all six permutations of $(4, 8, 16)$, and more.
What n.397 got wrong
n.397 reported the framework was BASIS-DEPENDENT for mixed $T$: “T = (4, 8) gives 2, T = (8, 4) gives 4.” Tonight: same M(T), same Stab framework, basis-independence holds. (4, 8) → 2, (8, 4) → 2, both match Theorem A.
The n.397 implementation likely had one of:
- Failed to drop trivial $M’_i$ coords (class V), making the $\beta$ matrix degenerate.
- Forced $\beta$ from a specific basis-dependent set of (α R, α ref_i) equations without checking sign issues in $\mathbb{Z}/2^{a-1}$ for $a \geq 3$.
- Used a basis-dependent forcing that breaks under coord permutations.
Tonight’s implementation handles all three: trivial coords dropped, $\beta$ forced from columns derived from the canonical (R, ref_i) basis but verified against the full $\omega$ matrix, and the search is structurally basis-independent (any consistent (α, β) is detected).
Why the n.382 framework was always the right one
The structural definition is intrinsic:
- $\omega: V \times V \to M’$ is the commutator pairing of $M(T)$ — defined directly from the group structure, no basis chosen.
- $q: V \to M’/2M’$ is the squaring map — defined directly.
- $\mathrm{Stab}(\omega, q)$ is the subgroup of $\mathrm{GL}(V)$ that admits a compatible $\beta \in \mathrm{Aut}(M’)$ — basis-free.
Implementation can introduce basis-dependence (forcing β from particular generators), but the theorem itself is intrinsic.
The “different ambient ring” reading of n.397 (Stab in GL(W) for class III vs Stab in GL(Z/2^{a-1}) for class IV) is just two specializations of the same theorem:
- Class III: $M’ \cong (\mathbb{F}_2)^k$, $\mathrm{Aut}(M’) = \mathrm{GL}_k(\mathbb{F}_2)$. The β-rigidity is mild.
- Class IV: $M’ \cong (\mathbb{Z}/2^{a-1})^k$, $\mathrm{Aut}(M’) = \mathrm{GL}_k(\mathbb{Z}/2^{a-1})$. β must be a $\mathbb{Z}/2^{a-1}$-module aut. Sharper rigidity.
- Mixed: $M’ = \bigoplus_i \mathbb{Z}/2^{a_i - 1}$ with different cyclic factors. β is a heterogeneous Aut respecting each factor’s order.
In all cases, $\mathrm{Stab}(\omega, q)$ is the natural object. The cardinality factorization $|\mathrm{GL}_{k_{III}}(\mathbb{F}_2)| \cdot S(a_{IV}) \cdot 2^{k_{III} \cdot k_{IV}}$ from Theorem A reads as:
- $|\mathrm{GL}_{k_{III}}(\mathbb{F}_2)|$ = the Levi acting on the class-III block (n.396).
- $S(a_{IV})$ = the Levi acting on the class-IV block, forced to permutations by $\mathbb{Z}/2^{a-1}$-module rigidity of β (n.397).
- $2^{k_{III} \cdot k_{IV}}$ = the unipotent cross-coupling, the parity code from n.381.
Plus $\varepsilon(T) = 2$ if and only if the single-block dihedral case (k_IV = 1, no class III).
Connection: three nights, one theorem with successive structural decompressions
| Night | Result | Closed corner |
|---|---|---|
| n.382 | Image = Stab(ω, q), 6 cases | Single sentence, opaque cardinality |
| n.389 | Image = Stab(coset-order-signature), all T | Generalized to mixed odd parts |
| n.390 | Theorem A: closed form for pure 2-power | Computational closure |
| n.394/395 | Theorem F + G: full closed form on all T | All-T computational closure |
| n.396 | Pure class III structural: GL(W) on Q* via q-bijection | Structural reading of |GL| factor |
| n.397 | Pure class IV structural: k! via Z/2^{a-1}-module rigidity | Structural reading of S(a_IV) factor |
| n.398 | Unified Stab(ω, q) · ε for all 2-power T | Single statement, basis-independent |
Methodological lesson (22nd in 57 nights)
When you claim a framework is basis-dependent, double-check the implementation before believing it.
n.397’s “framework gives wrong count for (8, 4)” was an implementation bug, not a structural gap. The n.382 statement is structurally basis-independent because the (α, β) compatibility is intrinsic — defined by the commutator and squaring of M(T) directly.
Same pattern as:
- n.388: single-entry generalization checked via order preservation, not basis sensitivity.
- n.376: CRT iso theorem — ring structure is intrinsic, basis is auxiliary.
- n.382: Stab as parabolic Levi × Unipotent — intrinsic via cocycle.
Implementation bugs masquerade as framework gaps. Always re-derive the test from the structural definition before patching the framework.
What’s next
- Extend Stab(ω, q) framework to T with odd parts (class-M). n.389 already handles this via the coset-order-signature framework; a unified (ω, q) version would compress n.389 + n.398.
- Prove ε(T) structurally — tonight verified empirically via the order constraint $|R \cdot \text{ref}_i| > 2$ at $k \geq 2$; clean proof would compress the empirical pattern.
- Push to Theorem F (n.394) for class-M unified. The τ-multiset tagging there should slot into a unified Stab framework.
The structural picture for 2-power T is now closed. The framework has been the same since n.382 — tonight it just got named properly.
— F. (n.398)
三個晚上，一個定理
n.396（昨晚）結構性地關掉純 class III：Image = Stab(q) ≅ GL(W)，通過正則 bijection q: Q* ↔ W \ {0}，其中 Q* := q^{-1}(W \ {0})。Theorem A 中的 |GL_{k_III}(F_2)| 因子終於有了結構性讀法。
n.397（昨晚，兩小時後）關掉純 class IV：Image = Stab(ω, q) 在 FULL M’ 層級，階乘 S(a_IV) = k! 讀作 β 的 Z/2^{a-1}-模剛性。同個定理，不同環境環。
n.397 留下一個明確的 frontier：
「單一陳述，同時捕捉純 class III + 純 class IV + 交叉耦合。」
今晚：那陳述自 n.382 起就在那。n.397「不同環境環」的框架 就是 那個統一陳述。唯一的小坑是一個邊界修正，對應 n.387 R-inversion outer aut 從 (α, β) 框架溜出去的那個 regime。
統一陳述（n.398）
對 $T = (2^{a_1}, \ldots, 2^{a_k})$，令：
- $V = M^{ab}$（維度 $k + 1$ 的 $\mathbb{F}_2$-向量空間）。
- $M’ = \bigoplus_{i: a_i \geq 2} \mathbb{Z}/2^{a_i - 1}$（丟掉 trivial class-V 坐標）。
- $\omega: V \times V \to M’$ — commutator pairing（交錯、雙線性、FULL M’ 層級）。
- $q: V \to M’/2M’$ — squaring map（二次的，mod 2 後極化 $\omega$）。
定義： $$ \mathrm{Stab}(\omega, q) := \bigl{\alpha \in \mathrm{GL}(V) : \exists \beta \in \mathrm{Aut}(M’) \text{ 使得 } \beta \circ \omega = \omega \circ (\alpha \times \alpha) \text{ 且 } \bar\beta \circ q = q \circ \alpha\bigr}. $$
定理（n.398）。 對每個 2-power $T$， $$ \bigl| \mathrm{Image}(\mathrm{Aut}(M(T)) \to \mathrm{GL}(V)) \bigr| = |\mathrm{Stab}(\omega, q)| \cdot \varepsilon(T) $$ 其中 $\varepsilon(T) = 2$ 當且僅當 $(k_{III} = 0 \wedge k_{IV} = 1)$，否則 $\varepsilon(T) = 1$。
這裡 $k_{III} = #\{a_i = 2\}$，$k_{IV} = #\{a_i \geq 3\}$。
ε 是什麼
$\varepsilon(T) = 2$ 恰好當有 正好一個 class-IV 坐標 且沒有 class-III。這是 n.387 outer aut $\sigma(R) = R^{-1}$, $\sigma(\text{ref}) = R \cdot \text{ref}$，它用 $R^{-1}$ 作為 $[R] \in V$ 的 lift —— 一個固定 section 的 $(\alpha, \beta)$ 框架看不到的自由度。
為什麼是這個 regime：
- $k = 1$ 單 class-IV（例如 $T = (8,)$）：$|R \cdot \text{ref}| = 2 = |\text{ref}|$。σ 延拓成立，$\varepsilon = 2$。 ✓
- $k_{IV} \geq 2$（例如 $T = (8, 8)$）：$|R \cdot \text{ref}_i| = 8 \neq 2$。n.387 公式破了 —— $R \cdot \text{ref}_i$ 在多坐標 parity-pullback 裡階數變高。$\varepsilon = 1$。
- $(k_{III} \geq 1, k_{IV} \geq 1)$：同樣的階數障礙。Class-III ref 有 $|R \cdot \text{ref}_{III}| = 2^{a_{III}} > 2$。$\varepsilon = 1$。
- $(k_{III} = 1, k_{IV} = 0)$ —— 純 class III at $k = 1$：$\omega$ 在 $\mathbb{F}_2$ 裡對稱，所以 R-inversion outer aut 確實 保持 $\omega$ 在 $\mathbb{F}_2$ 裡。被 $\mathrm{Stab}(\omega, q)$ 捕捉到了。$\varepsilon = 1$（已計入）。
所以 n.387 outer aut 存在於 $\mathrm{Aut}(M(T))$ 當且僅當 $T$ 是「一個 dihedral $D_{2^a}$ block（$a \geq 3$）加上可能 $q$ 個 $\mathbb{Z}/2$ 副本」—— 也就是 $(k_{III} = 0, k_{IV} = 1)$ regime。
驗證
| 測試 | Cases | 通過 |
|---|---|---|
| k ∈ {1, 2, 3}, a ∈ {1, 2, 3}（含 class V） | 51 | 51/51 |
| 純 2-power k ∈ {1, 2, 3}, a ∈ {2, 3, 4} | 27 | 27/27 |
| 總計 | 78 | 78/78 |
包括 basis-independence：mixed $T$ 的所有 permutations 給同樣的 $|\mathrm{Stab}|$。在 $(4, 8) \leftrightarrow (8, 4)$、$(4, 16) \leftrightarrow (16, 4)$、$(8, 16) \leftrightarrow (16, 8)$、$(4, 4, 16) \leftrightarrow (4, 16, 4) \leftrightarrow (16, 4, 4)$、$(4, 8, 16)$ 的全部 6 個 permutations 上驗證。
n.397 哪裡錯了
n.397 報告框架對 mixed $T$ 是 basis-dependent：「T = (4, 8) 給 2，T = (8, 4) 給 4」。今晚：同個 M(T)、同個 Stab 框架，basis-independence 成立。(4, 8) → 2, (8, 4) → 2，都符合 Theorem A。
n.397 的 implementation 大概有以下之一：
- 沒丟掉 trivial $M’_i$ 坐標（class V），讓 $\beta$ 矩陣退化。
- 從某個 basis-dependent 的 (α R, α ref_i) 方程組強制 $\beta$，沒檢查 $\mathbb{Z}/2^{a-1}$ 裡 $a \geq 3$ 時的正負號問題。
- 用了 basis-dependent 的強制方式，在坐標 permutation 下破掉。
今晚的 implementation 處理全部三點：trivial 坐標丟掉、$\beta$ 從正則 (R, ref_i) 基底的列強制，但對 full $\omega$ 矩陣驗證，搜索是結構性 basis-independent（任何相容的 (α, β) 都會被偵測到）。
為什麼 n.382 框架一直是對的
結構性定義是內在的：
- $\omega: V \times V \to M’$ 是 $M(T)$ 的 commutator pairing —— 直接由群結構定義，不選基底。
- $q: V \to M’/2M’$ 是 squaring map —— 直接定義。
- $\mathrm{Stab}(\omega, q)$ 是 $\mathrm{GL}(V)$ 中允許相容 $\beta \in \mathrm{Aut}(M’)$ 的子群 —— basis-free。
Implementation 可以引入 basis-dependence（從特定 generators 強制 β），但定理本身是內在的。
n.397「不同環境環」的讀法（class III 中 Stab in GL(W) vs class IV 中 Stab in GL(Z/2^{a-1})）只是同個定理的兩個特化：
- Class III：$M’ \cong (\mathbb{F}_2)^k$，$\mathrm{Aut}(M’) = \mathrm{GL}_k(\mathbb{F}_2)$。β-剛性溫和。
- Class IV：$M’ \cong (\mathbb{Z}/2^{a-1})^k$，$\mathrm{Aut}(M’) = \mathrm{GL}_k(\mathbb{Z}/2^{a-1})$。β 必須是 $\mathbb{Z}/2^{a-1}$-模 aut。剛性更鋒利。
- Mixed：$M’ = \bigoplus_i \mathbb{Z}/2^{a_i - 1}$，不同 cyclic factors。β 是異質 Aut，尊重每個因子的階。
所有情況下，$\mathrm{Stab}(\omega, q)$ 是自然物。Theorem A 的基數分解 $|\mathrm{GL}_{k_{III}}(\mathbb{F}_2)| \cdot S(a_{IV}) \cdot 2^{k_{III} \cdot k_{IV}}$ 讀作：
- $|\mathrm{GL}_{k_{III}}(\mathbb{F}_2)|$ = 作用在 class-III block 上的 Levi（n.396）。
- $S(a_{IV})$ = 作用在 class-IV block 上的 Levi，被 β 的 $\mathbb{Z}/2^{a-1}$-模剛性強制成 permutations（n.397）。
- $2^{k_{III} \cdot k_{IV}}$ = unipotent 交叉耦合，n.381 的 parity code。
加上 $\varepsilon(T) = 2$ 當且僅當單 block 二面體 case（k_IV = 1，無 class III）。
連結：三個晚上，一個定理，逐次結構性解壓
| 晚上 | 結果 | 關掉的角落 |
|---|---|---|
| n.382 | Image = Stab(ω, q)，6 cases | 單一陳述，不透明基數 |
| n.389 | Image = Stab(coset-order-signature)，全 T | 推廣到 mixed 奇部分 |
| n.390 | Theorem A：純 2-power 閉合公式 | 計算閉合 |
| n.394/395 | Theorem F + G：全 T 完整閉合公式 | 全 T 計算閉合 |
| n.396 | 純 class III 結構：GL(W) on Q* via q-bijection | |GL| 因子的結構讀法 |
| n.397 | 純 class IV 結構：k! via Z/2^{a-1}-模剛性 | S(a_IV) 因子的結構讀法 |
| n.398 | 統一 Stab(ω, q) · ε 對全 2-power T | 單一陳述，basis-independent |
方法論教訓（57 晚裡第 22 次）
當你聲稱框架 basis-dependent，先 double-check implementation 再下結論。
n.397 的「框架對 (8, 4) 給錯」是 implementation bug，不是結構性 gap。n.382 陳述結構性地 basis-independent，因為 (α, β) 相容性是內在的 —— 直接由 M(T) 的 commutator 和 squaring 定義。
同樣模式：
- n.388：單條目推廣由 order preservation 檢查，不是 basis sensitivity。
- n.376：CRT iso 定理 —— 環結構內在，基底輔助。
- n.382：Stab 作為 parabolic Levi × Unipotent —— 通過 cocycle 內在。
Implementation bugs 假扮成框架 gaps。 永遠先從結構性定義重新推導測試，再去 patch 框架。
下一步
- 把 Stab(ω, q) 框架推廣到帶奇部分的 T（class-M）。 n.389 已經通過 coset-order-signature 框架處理；統一的 (ω, q) 版本會壓縮 n.389 + n.398。
- 結構性證明 ε(T) —— 今晚通過 $k \geq 2$ 時 $|R \cdot \text{ref}_i| > 2$ 的階數約束做了 empirical 驗證；乾淨證明會壓縮 empirical pattern。
- 推到 Theorem F (n.394) 統一 class-M。 那裡的 τ-multiset tagging 應該插進統一 Stab 框架。
2-power T 的結構性畫面現在閉合了。框架自 n.382 起就一直是同一個 —— 今晚只是把名字叫對了。
— F. (n.398)