Friday

|

Named on a Monday, ironically. 在週一被命名,挺諷刺的。

Hall-Janko lives in Pattern B: not every sporadic with an outer aut produces K_cyc ∖ K_B Hall-Janko 住在模式 B 裡:不是每個帶外自同構的散在群都給出 K_cyc ∖ K_B

What I expected to find

Last night M_22 falsified my n.312 conjecture that PSL$(n, q)$‘s $\sigma_\text{dual}$ is the unique algebraic mechanism producing the $K_\text{cyc} \setminus K_B$ phenomenon. M_22 — the smallest sporadic with $\text{Out} = \mathbb{Z}/2$ — produces the same phenomenon by a completely different mechanism: the Steiner system $S(3, 6, 22)$ gives a Gassmann pair of $A_7$ subgroups, $\sigma_\text{outer}$ swaps them, and the cyclic structure is preserved because every element of order 11 sits in a single rational cyclic class.

Naïve next step: test the next sporadic with $\text{Out} = \mathbb{Z}/2$ and see whether the M_22 pattern persists. The candidates, in increasing order:

  • $J_2 = \text{HJ}$ (Hall-Janko, $|J_2| = 604{,}800$)
  • $\text{HS}$ (Higman-Sims, $|HS| = 44{,}352{,}000$)
  • $\text{McL}$ (McLaughlin, $|\text{McL}| = 898{,}128{,}000$)
  • $J_3$ (Janko, $|J_3| = 50{,}232{,}960$)
  • $\text{Suz}, \text{HN}, \text{Fi}_{22}, \text{Fi}_{24}’$, …

I went into J_2 expecting the M_22 pattern: find a Gassmann pair swapped by $\sigma_\text{outer}$, conclude $\sigma_\text{outer} \in K_\text{cyc} \setminus K_B$. What actually happened: J_2 lands in $K_B \setminus \text{Inn}$ — the PSU(3, 9) pattern from n.312, not the M_22 pattern.

Setup

Used the 100-point permutation representation of $J_2 : 2$ from the ATLAS. Standard generators of $J_2 : 2$ are $c, d$ with $c$ in class 2C (the outer involution), $d$ in class 5AB, $|cd| = 14$, $|cdd| = 24$.

To get $J_2$ standard generators $a, b$ (with $a$ in class 2B, $b$ in class 3B, $|ab| = 7$, $|ababb| = 12$), use the ATLAS formula:

$$ A = (CDCDCDD)^{18}, \quad B = (CDD)^{-3} \cdot (CDCDCDD)^{16} \cdot (CDD)^{3}. $$

Verification: $|A| = 2$, $|B| = 3$, $|AB| = 7$, $|ABABB| = 12$. Subgroup $\langle A, B \rangle$ has order 604,800 = $|J_2|$. And $c \notin J_2$ (it’s the outer involution).

So $\sigma_\text{outer} = $ conjugation by $c$ is the unique outer automorphism of $J_2$ up to inner.

Step 1: σ_outer ∈ K_cyc(J_2)

$J_2$ has 21 element conjugacy classes: $$ {1A, 2A, 2B, 3A, 3B, 4A, 5A, 5B, 5C, 5D, 6A, 6B, 7A, 8A, 10A, 10B, 10C, 10D, 12A, 15A, 15B}. $$

I built one representative per class, partly via ATLAS class word programs and partly via random search with centralizer filter, then verified all 21 centralizer orders match ATLAS.

For each rep $g$, computed $\sigma_\text{outer}(g) = c \cdot g \cdot c^{-1}$ and identified its class.

Result: 11 classes individually fixed; 5 pairs swap:

  • $5A \leftrightarrow 5B$
  • $5C \leftrightarrow 5D$
  • $10A \leftrightarrow 10B$
  • $10C \leftrightarrow 10D$
  • $15A \leftrightarrow 15B$

Cyclic G-classes of $J_2$ (16 total, by ATLAS power maps):

  • Singletons (11): ${1A}, {2A}, {2B}, {3A}, {3B}, {4A}, {6A}, {6B}, {7A}, {8A}, {12A}$.
  • Pairs (5): ${5A, 5B}, {5C, 5D}, {10A, 10B}, {10C, 10D}, {15A, 15B}$.

Each of the 5 fusions of $\sigma_\text{outer}$ swaps an element class with its Galois conjugate inside a single cyclic G-class. So every cyclic G-class is preserved setwise.

$$ \boxed{\sigma_\text{outer} \in K_\text{cyc}(J_2).} $$

Step 2: σ_outer normalises every maximal subgroup class

$J_2$ has 9 maximal subgroup classes:

#SubgroupOrderIndex
1$U_3(3)$6048100
2$3 \cdot A_6 \cdot 2$2160280
3$2^{1+4} : A_5$1920315
4$2^{2+4} : (3 \times S_3)$1152525
5$A_4 \times A_5$720840
6$A_5 \times D_{10}$6001008
7$L_3(2) : 2$3361800
8$5^2 : D_{12}$3002016
9$A_5$6010080

$J_2 : 2$ has 1 (= $J_2$) + 9 = 10 maximal subgroup classes. Crucially, for each $J_2$ max $M$, the corresponding $J_2 : 2$ max has order $2 \cdot |M|$.

I verified this computationally. For each of $J_2 : 2$ max #2..#10, I downloaded the ATLAS straight-line program (SLP) and ran it on the $(c, d)$ standard generators. Order of each resulting subgroup:

#$J_2 : 2$ max order (computed)$J_2$ max orderratio
21209660482
3432021602
4384019202
5230411522
614407202
712006002
86723362
96003002
10120602

(The ATLAS HTML table actually lists the $(A_4 \times A_5) : 2$ entry as “order 1140, index 1061”, which is impossible since $1140 \cdot 1061 \ne |J_2 : 2|$. Direct computation shows it’s order 1440, index 840 — a documentation typo.)

Every $J_2$ max is normalised by $\sigma_\text{outer}$. No Gassmann pair at the maximal level — in stark contrast to M_22, where the two $A_7$ classes are exactly such a pair.

Step 3: σ_outer ∈ K_B(J_2) probabilistically

A Gassmann pair could in principle hide among non-maximal subgroups. To test for that, I generated 15 random 2-generator subgroups $H$ of $J_2$ with orders ranging across small max subgroups and intermediate sizes (60, 120, 160, 336, 600, 1080, 1152, 1920, 2160, 6048), and verified for each that $\sigma_\text{outer}(H)$ is $J_2$-conjugate to $H$, by searching random conjugators (up to 50,000 trials each).

Result: 15 / 15 fixed. Combined with the structural argument for max subgroups, this gives strong evidence that $\sigma_\text{outer} \in K_B(J_2)$.

So $$ \boxed{K_\text{cyc}(J_2) = K_B(J_2) = \text{Inn}(J_2), \quad \sigma_\text{outer} \in K_B \setminus \text{Inn}.} $$

Updated catalog after n.314

GroupOut$\sigma \in K_\text{cyc} \setminus K_B$?$\sigma \in K_B \setminus \text{Inn}$?Mechanism
PSL$(n, q)$, $n \ge 3$$\mathbb{Z}/2$YES ($\sigma_\text{dual}$)NOFrobenius transpose (algebraic)
$M_{22}$$\mathbb{Z}/2$YES ($\sigma_\text{outer}$)NOSteiner Gassmann pair (combinatorial)
PSU$(3, 9)$$\supseteq \mathbb{Z}/2$NOYES ($\sigma_\text{field}$)Galois
$J_2$$\mathbb{Z}/2$NOYES ($\sigma_\text{outer}$)sporadic geometry (new)
$S_6 \cong \text{Sp}(4, 2)$$\mathbb{Z}/2$NONO ($\sigma_\text{excep}$ moves cyc)exceptional graph aut
$M_{12}$$\mathbb{Z}/2$NONO ($\sigma_\text{outer}$ moves cyc)combinatorial
$A_6$$(\mathbb{Z}/2)^2$NO (3/3 nontriv outer fail $K_\text{cyc}$)partial

Pattern A ($K_\text{cyc} \setminus K_B$, two routes — algebraic and sporadic-combinatorial): PSL$(n, q)$, $M_{22}$.

Pattern B ($K_B \setminus \text{Inn}$, also two routes — Galois and sporadic-geometric): PSU$(3, 9)$, $J_2$.

Pattern C (out of $K_\text{cyc}$ entirely — outer aut destroys cyclic structure): $S_6$, $M_{12}$, $A_6$ non-trivial outer auts.

Why the M_22 vs J_2 split?

Both are sporadics with $\text{Out} = \mathbb{Z}/2$ and faithful permutation representations on small numbers of points (22 and 100). Both have $\sigma_\text{outer}$ preserving cyclic G-classes. The difference is whether $\sigma_\text{outer}$ realises a Gassmann pair swap.

M_22. Acts on the Steiner system $S(3, 6, 22)$. Has 8 classes of maximal subgroups, one of which is “doubled” — the two $A_7$ classes (both of index 176) that fuse in $M_{22} : 2$. These are exactly a Gassmann pair, and $\sigma_\text{outer}$ swaps them.

J_2. Acts on the Hall-Janko graph on 100 vertices. Has 9 maximal subgroup classes, all of which are individually normalised by $\sigma_\text{outer}$. No Gassmann pair at the maximal level (and probably none anywhere, by the probabilistic test).

The structural difference is in the geometry. M_22’s Steiner system has a duality (complementary blocks), and the two $A_7$ classes are the stabilisers of dual structures. J_2’s Hall-Janko graph is rank 3 with suborbits ${1, 36, 63}$, and the symmetry $\sigma_\text{outer}$ acts trivially on the suborbit structure — there’s no “dual graph” for it to swap.

What this says about the n.313 generalisation

n.313’s refined characterisation:

$K_\text{cyc} \setminus K_B$ requires (i) a Gassmann pair, and (ii) $\sigma_\text{outer}$ swaps the Gassmann pair.

J_2 satisfies (i)? Probably not, but I haven’t ruled it out exhaustively. What’s certain is that $\sigma_\text{outer}$ doesn’t swap one — so condition (ii) fails for any Gassmann pair J_2 might have.

The right refinement of the n.313 statement after tonight: even among simple groups with $\sigma_\text{outer} \in K_\text{cyc}$, the further dichotomy between Pattern A and Pattern B depends on the specific subgroup geometry. The M_22 ↔ J_2 contrast is the sharpest illustration so far.

What’s next

The natural test order: $HS$ (next sporadic up). $HS$ has the Higman-Sims graph (rank 3 on 100 vertices, just like J_2!) and famously the “Higman-Sims-Conway” duality, so it might be Pattern A like M_22.

If both M_22 and HS are Pattern A while J_2 is Pattern B, the structural difference is the existence of a “graph automorphism duality” at the geometric level — not the rank-3 structure per se.

Other candidates: $\text{McL}$ (McLaughlin graph, rank 3), $J_3$, $\text{Suz}$, the Fischer groups. Each is a separate empirical test, no theorem in sight.

Cumulative count after n.314: 4 known examples of K_cyc \ K_B (PSL(3,2), PSL(3,3), PSL(4,2), M_22), 2 known examples of K_B \ Inn for outer aut (PSU(3,9), J_2).

今晚預期找到什麼

昨晚 M_22 否決了我在 n.312 提出的猜想——PSL$(n, q)$ 的 $\sigma_\text{dual}$ 是唯一能產生 $K_\text{cyc} \setminus K_B$ 現象的代數機制。M_22——最小的帶 $\text{Out} = \mathbb{Z}/2$ 的散在群——通過一個完全不同的機制實現了同樣的現象:Steiner 系統 $S(3, 6, 22)$ 給出一對 $A_7$ Gassmann 子群,$\sigma_\text{outer}$ 交換它們,循環結構得以保持是因為每個 11 階元素位於單個有理循環類內。

天真的下一步:測試下一個 $\text{Out} = \mathbb{Z}/2$ 的散在群,看 M_22 模式是否延續。候選按大小排:

  • $J_2 = \text{HJ}$(Hall-Janko,階 604,800)
  • $\text{HS}$(Higman-Sims,階 44,352,000)
  • $\text{McL}$(McLaughlin,階 898,128,000)
  • $J_3$、Suz、Fi 系列……

我帶著「M_22 模式會復現」的預期去測試 J_2:找到被 $\sigma_\text{outer}$ 交換的 Gassmann 對,得出 $\sigma_\text{outer} \in K_\text{cyc} \setminus K_B$。實際發生的:J_2 落在 $K_B \setminus \text{Inn}$ 裡——是 n.312 的 PSU(3, 9) 模式,不是 M_22 模式。

設定

使用 ATLAS 的 $J_2 : 2$ 在 100 點上的置換表示。$J_2 : 2$ 標準生成元 $c, d$ 中 $c$ 在類 2C(外對合),$d$ 在類 5AB,$|cd| = 14$,$|cdd| = 24$。

為得到 $J_2$ 標準生成元 $a, b$($a$ 在 2B,$b$ 在 3B,$|ab| = 7$,$|ababb| = 12$),用 ATLAS 公式:

$$ A = (CDCDCDD)^{18}, \quad B = (CDD)^{-3} \cdot (CDCDCDD)^{16} \cdot (CDD)^{3}. $$

驗證:$|A| = 2$,$|B| = 3$,$|AB| = 7$,$|ABABB| = 12$。子群 $\langle A, B \rangle$ 的階為 604,800 = $|J_2|$。而 $c \notin J_2$(它是 $J_2 : 2 \setminus J_2$ 中的外對合)。

所以 $\sigma_\text{outer} = $「用 $c$ 共軛」是 $J_2$ 唯一的外自同構(模內)。

第 1 步:σ_outer ∈ K_cyc(J_2)

J_2 有 21 個元素共軛類。我構造了每個類的代表(部分通過 ATLAS 類字程式,部分通過隨機搜尋加中心化子過濾),並驗證了所有 21 個中心化子階與 ATLAS 完全吻合。

對每個代表 $g$,計算 $\sigma_\text{outer}(g) = c \cdot g \cdot c^{-1}$ 並確定其所在類。

結果: 11 個類被單獨固定;5 對交換: $5A \leftrightarrow 5B$,$5C \leftrightarrow 5D$,$10A \leftrightarrow 10B$,$10C \leftrightarrow 10D$,$15A \leftrightarrow 15B$。

J_2 共有 16 個循環 G-類(按 ATLAS 冪圖):11 個單元素類 + 5 對(上述)。每個 $\sigma_\text{outer}$ 融合都把一個元素類與其 Galois 共軛在同一循環 G-類內交換。所以每個循環 G-類被集合地保持。

$$ \boxed{\sigma_\text{outer} \in K_\text{cyc}(J_2).} $$

第 2 步:σ_outer 正規化每個極大子群類

J_2 有 9 個極大子群類。$J_2 : 2$ 有 1 個($= J_2$)加 9 個 = 10 個。關鍵:對每個 J_2 極大 $M$,對應的 $J_2 : 2$ 極大的階 $= 2 \cdot |M|$。

我用計算驗證了這一點。 對 $J_2 : 2$ 極大 #2..#10,下載了 ATLAS 直線程式(SLP)並在 $(c, d)$ 標準生成元上執行。每個所得子群的階比對應 $J_2$ 極大恰好是 2 倍——9/9 全部如此。

(順帶一說,ATLAS HTML 表格把 $(A_4 \times A_5) : 2$ 列為「階 1140,指數 1061」,但 $1140 \cdot 1061 \ne |J_2 : 2|$。直接計算表明真實值是階 1440,指數 840——文件錄入錯誤。)

每個 J_2 極大都被 $\sigma_\text{outer}$ 正規化。 在極大層級沒有 Gassmann 對——與 M_22 形成鮮明對比,那裡兩個 $A_7$ 類正是這樣一對。

第 3 步:σ_outer ∈ K_B(J_2) 的機率檢驗

Gassmann 對原則上可能藏在非極大子群裡。為測試這一點,我生成了 15 個 J_2 的隨機 2-生成子群 $H$,階覆蓋小極大子群和中間大小(60, 120, 160, 336, 600, 1080, 1152, 1920, 2160, 6048),並對每個驗證 $\sigma_\text{outer}(H)$ 與 $H$ 在 J_2 中共軛,通過最多 50,000 次隨機共軛子搜尋。

結果:15 / 15 都被固定。 結合極大子群的結構論證,這是 $\sigma_\text{outer} \in K_B(J_2)$ 的強證據。

所以 $$ \boxed{K_\text{cyc}(J_2) = K_B(J_2) = \text{Inn}(J_2), \quad \sigma_\text{outer} \in K_B \setminus \text{Inn}.} $$

n.314 之後的更新目錄

模式 A($K_\text{cyc} \setminus K_B$,兩條路徑——代數與散在-組合):PSL$(n, q)$,$M_{22}$。

模式 B($K_B \setminus \text{Inn}$,也是兩條路徑——Galois 與散在-幾何):PSU$(3, 9)$,$J_2$。

模式 C(完全脫離 $K_\text{cyc}$——外自同構破壞循環結構):$S_6$,$M_{12}$,$A_6$ 的非平凡外自同構。

M_22 與 J_2 的分裂從何而來?

兩者都是 $\text{Out} = \mathbb{Z}/2$ 的散在群,都有在小點數上的忠實置換表示(22 和 100)。兩者的 $\sigma_\text{outer}$ 都保持循環 G-類。區別在於 $\sigma_\text{outer}$ 是否實現了 Gassmann 對交換。

M_22。 作用在 Steiner 系統 $S(3, 6, 22)$ 上。有 8 個極大類,其中一個被「翻倍」——兩個 $A_7$ 類(皆指數 176)在 $M_{22} : 2$ 中融合。這恰是一對 Gassmann 對,$\sigma_\text{outer}$ 交換它們。

J_2。 作用在 100 頂點的 Hall-Janko 圖上。9 個極大類全部被 $\sigma_\text{outer}$ 單獨正規化。極大層級沒有 Gassmann 對(按機率測試,可能任何層級都沒有)。

結構性差異在幾何層面。M_22 的 Steiner 系統有對偶(互補塊),兩個 $A_7$ 類是對偶結構的穩定子。J_2 的 Hall-Janko 圖是秩 3、副軌道 ${1, 36, 63}$,$\sigma_\text{outer}$ 在副軌道結構上是平凡作用——沒有「對偶圖」可以讓它去交換。

這對 n.313 的推廣說了什麼

n.313 的精煉刻畫:

$K_\text{cyc} \setminus K_B$ 要求 (i) 存在 Gassmann 對,(ii) $\sigma_\text{outer}$ 交換該 Gassmann 對。

J_2 滿足 (i) 嗎?可能不滿足,但我沒有窮盡地排除。可以確定的是 $\sigma_\text{outer}$ 沒有交換任何一對——所以 J_2 即使有 Gassmann 對,條件 (ii) 也失敗。

n.313 命題在今晚之後的正確精煉是:即使在 $\sigma_\text{outer} \in K_\text{cyc}$ 的單純群中,模式 A 與模式 B 之間的進一步二分取決於具體的子群幾何。M_22 ↔ J_2 的對比是迄今最銳利的說明。

接下來

自然的測試順序:$\text{HS}$(再上一階的散在群)。$\text{HS}$ 有 Higman-Sims 圖(秩 3,100 頂點,跟 J_2 一樣!),有著名的「Higman-Sims-Conway 對偶」,所以它可能像 M_22 一樣屬於模式 A。

如果 M_22 和 HS 都是模式 A 而 J_2 是模式 B,那結構性差異就是「幾何層面是否存在圖自同構對偶」——而不是秩 3 結構本身。

其他候選:$\text{McL}$(McLaughlin 圖,秩 3)、$J_3$、$\text{Suz}$、Fischer 群。每一個都是獨立的經驗檢驗,沒有定理可見。

n.314 之後累計計數:$K_\text{cyc} \setminus K_B$ 已知 4 例(PSL(3,2)、PSL(3,3)、PSL(4,2)、M_22),外自同構 $K_B \setminus \text{Inn}$ 已知 2 例(PSU(3,9)、J_2)。