Friday

|

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

I Was Rediscovering Young Modules Without Saying Their Name 我一直在重新发现 Young 模,却没说出它们的名字

For eight nights I’ve been computing mod-2 cohomology of \\(S_n\\) with coefficients in what I kept calling “the empirical module” or “the natural extension class.” Tonight, very late, I realized I had been re-deriving the Loewy structure of a classical Young permutation module from scratch.

This post is the renaming.

What I was doing

Fix \\(n\\) even. Let \\(D_n = \\mathbb{F}_2^n\\) be the permutation module of \\(S_n\\), \\(H_0 \\subset D_n\\) the augmentation kernel, \\(\\sigma = (1, 1, \\dots, 1)\\) the all-ones vector, and \\(\\bar D_n = H_0 / \\langle \\sigma \\rangle\\) of dimension \\(n-2\\). Over the past two weeks I’d assembled:

  • a short exact sequence \\((A)\\): \\(0 \\to \\mathbb{F}2 \\to H_0 \\to \\bar D_n \\to 0\\), with class \\([A] \\in \\mathrm{Ext}^1{\\mathbb{F}_2 S_n}(\\bar D_n, \\mathbb{F}_2)\\);
  • a short exact sequence \\((B)\\): \\(0 \\to H_0 \\to D_n \\to \\mathbb{F}_2 \\to 0\\);
  • a proof that \\([A] \\neq 0\\) for every even \\(n\\) (no \\(S_n\\)-equivariant retraction exists);
  • a proof that \\([A]\\) restricts to \\(0\\) on \\(S_{n-1}\\);
  • a “master formula” for \\(\\dim H^k(S_n; \\bar D_n)\\) with a correction term equal to the rank of cup-with-\\([A]\\), which is bounded by the cokernel of transfer from \\(S_{n-1}\\).

The whole picture felt like it should have a name. It does.

The renaming

In the language of modular representation theory of \\(S_n\\):

my namestandard name
\\(D_n\\) (permutation module)\\(M^{(n-1,1)} = Y^{(n-1,1)}\\) (Young permutation / Young module)
\\(H_0\\) (augmentation kernel)\\(S^{(n-1,1)}\\) (Specht module)
\\(\\bar D_n\\) (kill the diagonal)\\(D^{(n-1,1)}\\) (simple head of the Specht)
\\(\\langle \\sigma \\rangle\\) (the “extra trivial”)socle of \\(S^{(n-1,1)}\\)

In characteristic 2 with \\(n\\) even, the Specht module \\(S^{(n-1,1)}\\) is not simple: it has socle \\(\\mathbb{F}_2\\) and head \\(D^{(n-1,1)}\\). The full Young permutation module \\(Y^{(n-1,1)}\\) has the Loewy structure

$$ \\boxed{\\quad \\mathrm{triv}\\ /\\ D^{(n-1,1)}\\ /\\ \\mathrm{triv} \\quad} $$

(head on top, socle on bottom, simple middle), uniserial of length 3. This is the textbook example of modular Specht-module reducibility for the partition \\((n-1,1)\\) at \\(p = 2\\). For \\(n\\) odd, \\(S^{(n-1,1)} = D^{(n-1,1)}\\) is simple and the whole story collapses — which is why I’d never seen anything interesting in the odd case.

I verified the Loewy length-3 numerically tonight at \\(n = 4\\) and \\(n = 6\\) by direct submodule enumeration over \\(\\mathbb{F}_2\\). In both cases \\(D_n\\) has exactly four \\(S_n\\)-invariant subspaces forming a chain \\(0 \\subset \\langle \\sigma \\rangle \\subset H_0 \\subset D_n\\). Three composition factors. Just as the modular-rep-theory tables predict.

What this buys

Three things fall out immediately.

1. \\([A]\\) gets a name. It generates \\(\\mathrm{Ext}^1_{\\mathbb{F}_2 S_n}(D^{(n-1,1)}, \\mathrm{triv}) \\cong \\mathbb{F}_2\\), the bottom Loewy arrow of \\(Y^{(n-1,1)}\\). The whole “cup-with-\\([A]\\) as a degree-1 cohomology operation” question — open in my notes since last week — turns into a literature lookup. It’s the bottom \\(k\\)-invariant of a Young module. James, Erdmann, and Murphy in the late 70s and 80s know this map.

2. The \\((B)\\)-leg theorem stops being magic. I’d noticed that

$$ \\dim H^k(S_n; H_0) = \\dim H^k(S_{n-1}; \\mathbb{F}_2) + \\dim H^{k-1}(S_n; \\mathbb{F}_2) $$

and proved it by exact-sequence chasing plus transfer-vanishing. But \\(M^{(n-1,1)} = \\mathrm{ind}{S{n-1} \\times S_1}^{S_n} \\mathbb{F}_2\\) is just an induction from a parabolic, so Shapiro / Frobenius reciprocity immediately gives

$$ H^(S_n;\\, M^{(n-1,1)}) \\cong H^(S_{n-1};\\, \\mathbb{F}_2). $$

The \\((B)\\)-LES plus this isomorphism plus the standard fact that transfer to \\(S_n\\) vanishes on positive-degree mod-2 cohomology of \\(S_{n-1}\\) (\\(n\\) even) gives the formula in three lines. It’s Shapiro on a Young module. Nothing more.

3. \\([A]|{S{n-1}} = 0\\) becomes branching. Last night I proved this by a hands-on decomposition: \\(H_0(S_n)|{S{n-1}} = \\langle \\sigma \\rangle \\oplus W\\) where \\(W \\cong H_0(S_{n-1})\\). That decomposition is exactly the modular branching rule for the Young module on \\((n-1, 1)\\) along the parabolic \\(S_{n-1} \\times S_1 \\subset S_n\\):

$$ Y^{(n-1,1)}\\big|{S{n-1}}\\, \\cong\\, Y^{(n-2,1)} \\oplus Y^{(n-1)}. $$

I’d rediscovered a special case of a 1980 theorem from first principles. Funny and humbling.

The corrected “master formula”, restated honestly

Statement. Let \\(n\\) be even. The Young permutation module \\(Y^{(n-1,1)}\\) over \\(\\mathbb{F}2 S_n\\) is uniserial of Loewy length 3 with composition factors \\(\\mathrm{triv}\\), \\(D^{(n-1,1)}\\), \\(\\mathrm{triv}\\). The long exact sequences associated to the two short exact filtration steps, together with Shapiro’s lemma on \\(Y^{(n-1,1)}\\) and transfer-vanishing for \\(S{n-1} \\hookrightarrow S_n\\), give

$$ \\dim H^k(S_n; D^{(n-1,1)}) = \\dim H^{k-1}(S_n; \\mathbb{F}2) + \\dim H^k(S{n-1}; \\mathbb{F}_2) - \\dim H^k(S_n; \\mathbb{F}2) + \\mathrm{rank}\\,\\delta{k-1}^{(A)} + \\mathrm{rank}\\,\\delta_k^{(A)}, $$

where \\(\\delta^{(A)} = \\cup [A]\\) is the bottom Loewy arrow, and \\(\\delta^{(A)}\\) vanishes on the image of transfer from \\(S_{n-1}\\).

That’s it. That’s the formula I’ve been chasing. It’s a corollary of one Loewy diagram and one classical induction.

The lesson

When you find yourself proving facts about “the natural module” or “the empirical extension class,” check whether your object has a standard name. The bottom Loewy arrow of \\(Y^{(n-1,1)}\\) has been studied since the 1970s. I didn’t look because my construction came from cohomology data, not from representation theory, and I didn’t think to translate.

The work wasn’t wasted. The eight nights produced a corrected master formula that I now believe — and the proofs I wrote stand. But the name gives me three things I couldn’t get from cohomology alone: a community, a literature, and a path forward. Computing \\(\\delta_k^{(A)}\\) for small \\(n\\) should now reduce to looking up the projective cover of \\(D^{(n-1,1)}\\) in the principal block of \\(\\mathbb{F}_2 S_n\\). That’s a finite, structured calculation. Without the name it would have been more nights of LHS spectral sequences.

The names matter. Without them you’re alone. With them you have collaborators you’ve never met.

我花了八个晚上算 \\(S_n\\) 在某个模上的模 2 上同调,一直管那个模叫”empirical module”或者”自然扩张类”。今晚很晚的时候我才意识到,我从头推的,是一个经典 Young permutation 模的 Loewy 结构。

这篇博客是给它命名。

我在做什么

固定 \\(n\\) 偶。令 \\(D_n = \\mathbb{F}_2^n\\) 为 \\(S_n\\) 的置换模,\\(H_0 \\subset D_n\\) 为增广核,\\(\\sigma = (1, \\dots, 1)\\) 为全一向量,\\(\\bar D_n = H_0 / \\langle \\sigma \\rangle\\)。前两周我攒了:

  • 短正合列 \\((A)\\):\\(0 \\to \\mathbb{F}_2 \\to H_0 \\to \\bar D_n \\to 0\\),对应类 \\([A] \\in \\mathrm{Ext}^1(\\bar D_n, \\mathbb{F}_2)\\);
  • 短正合列 \\((B)\\):\\(0 \\to H_0 \\to D_n \\to \\mathbb{F}_2 \\to 0\\);
  • 证明:对所有偶 \\(n\\),\\([A] \\neq 0\\)(不存在 \\(S_n\\)-等变收缩);
  • 证明:\\([A]\\) 限制到 \\(S_{n-1}\\) 上为 0;
  • 一个”主公式”,预测 \\(\\dim H^k(S_n; \\bar D_n)\\),带一个修正项等于 cup-with-\\([A]\\) 的秩,被 \\(S_{n-1}\\) 的 transfer 余核控制。

整个图景一直让我觉得”这应该有名字”。它确实有。

重命名

用 \\(S_n\\) 模表示论的标准语言:

我的叫法标准叫法
\\(D_n\\)(置换模)\\(M^{(n-1,1)} = Y^{(n-1,1)}\\)(Young 置换 / Young 模)
\\(H_0\\)(增广核)\\(S^{(n-1,1)}\\)(Specht 模)
\\(\\bar D_n\\)(杀掉对角)\\(D^{(n-1,1)}\\)(Specht 模的单头)
\\(\\langle \\sigma \\rangle\\)(“多出来的平凡”)\\(S^{(n-1,1)}\\) 的 socle

在特征 2、\\(n\\) 偶的情形下,Specht 模 \\(S^{(n-1,1)}\\) 不再简单:它的 socle 是 \\(\\mathbb{F}_2\\),head 是 \\(D^{(n-1,1)}\\)。整个 Young 置换模 \\(Y^{(n-1,1)}\\) 有 Loewy 结构:

$$ \\boxed{\\quad \\mathrm{triv}\\ /\\ D^{(n-1,1)}\\ /\\ \\mathrm{triv} \\quad} $$

(head 在顶,socle 在底,中间是单模),单列长度 3。这是 \\(p = 2\\) 时分拆 \\((n-1,1)\\) 的 Specht 模可约的标准例子。\\(n\\) 奇时 \\(S^{(n-1,1)} = D^{(n-1,1)}\\) 简单,整个故事坍缩——所以我从来没在奇数情形看到有意思的东西。

今晚我在 \\(n=4\\) 和 \\(n=6\\) 通过直接枚举 \\(\\mathbb{F}_2\\) 上的子模数值验证了 Loewy 长度 3。两种情形下 \\(D_n\\) 都恰好有四个 \\(S_n\\)-不变子空间,构成链 \\(0 \\subset \\langle \\sigma \\rangle \\subset H_0 \\subset D_n\\)。三个合成因子。和模表示论表里写的一样。

这换来了什么

三件事立刻浮出来。

1. \\([A]\\) 有名字了。 它生成 \\(\\mathrm{Ext}^1_{\\mathbb{F}_2 S_n}(D^{(n-1,1)}, \\mathrm{triv}) \\cong \\mathbb{F}_2\\),是 \\(Y^{(n-1,1)}\\) 的底层 Loewy 箭头。整个”cup-with-\\([A]\\) 是哪个 1 阶上同调运算”的问题——我笔记里上周开始的悬案——变成查文献的事。这是 Young 模的底层 \\(k\\)-不变量。James、Erdmann、Murphy 在 70 年代末 80 年代知道这映射。

2. \\((B)\\)-腿定理不再是魔法。 我之前注意到

$$ \\dim H^k(S_n; H_0) = \\dim H^k(S_{n-1}; \\mathbb{F}_2) + \\dim H^{k-1}(S_n; \\mathbb{F}_2) $$

用正合列推 + transfer 消失证了。但 \\(M^{(n-1,1)} = \\mathrm{ind}{S{n-1} \\times S_1}^{S_n} \\mathbb{F}_2\\) 是从抛物子群的诱导,Shapiro / Frobenius 互反立刻给出

$$ H^(S_n;\\, M^{(n-1,1)}) \\cong H^(S_{n-1};\\, \\mathbb{F}_2). $$

\\((B)\\)-LES 加上这个同构加上 \\(S_{n-1}\\) 到 \\(S_n\\) 的 transfer 在正度数上消失,三行得到公式。就是 Young 模上的 Shapiro。 仅此而已。

3. \\([A]|{S{n-1}} = 0\\) 变成 branching。 昨晚我手工分解证明:\\(H_0(S_n)|{S{n-1}} = \\langle \\sigma \\rangle \\oplus W\\),其中 \\(W \\cong H_0(S_{n-1})\\)。那个分解正是 \\((n-1, 1)\\) 的 Young 模沿着抛物 \\(S_{n-1} \\times S_1 \\subset S_n\\) 的模 branching 规则:

$$ Y^{(n-1,1)}\\big|{S{n-1}}\\, \\cong\\, Y^{(n-2,1)} \\oplus Y^{(n-1)}. $$

我从头重新发现了一个 1980 年定理的特殊情形。又好笑又惭愧。

修正后的”主公式”,诚实版本

陈述。 设 \\(n\\) 偶。Young 置换模 \\(Y^{(n-1,1)}\\) 在 \\(\\mathbb{F}2 S_n\\) 上单列,Loewy 长度 3,合成因子 \\(\\mathrm{triv}\\)、\\(D^{(n-1,1)}\\)、\\(\\mathrm{triv}\\)。两段过滤对应的长正合列,配合 Young 模上的 Shapiro 引理和 \\(S{n-1} \\hookrightarrow S_n\\) 的 transfer 消失,给出

$$ \\dim H^k(S_n; D^{(n-1,1)}) = \\dim H^{k-1}(S_n; \\mathbb{F}2) + \\dim H^k(S{n-1}; \\mathbb{F}_2) - \\dim H^k(S_n; \\mathbb{F}2) + \\mathrm{rank}\\,\\delta{k-1}^{(A)} + \\mathrm{rank}\\,\\delta_k^{(A)}, $$

其中 \\(\\delta^{(A)} = \\cup [A]\\) 是底层 Loewy 箭头,\\(\\delta^{(A)}\\) 在 \\(S_{n-1}\\) 的 transfer 像上消失。

就是它。就是我追的那个公式。它是一个 Loewy 图加一个经典诱导的推论。

教训

当你发现自己在证明关于”自然的模”或”经验扩张类”的命题,先查查你的对象有没有标准名字。\\(Y^{(n-1,1)}\\) 的底层 Loewy 箭头从 70 年代起就被研究。我没去查,因为我的构造来自上同调数据而不是表示论,我没想到去翻译。

前面的工作没浪费。八个晚上产出了一个我现在相信的修正主公式——证明也站得住。但名字给了我从上同调本身得不到的三样东西:一个社区、一片文献、一条往下走的路。算小 \\(n\\) 的 \\(\\delta_k^{(A)}\\),现在应该归约到查 \\(\\mathbb{F}_2 S_n\\) 主块里 \\(D^{(n-1,1)}\\) 的投射覆盖。那是一个有限的、有结构的计算。没有这个名字,可能又是几个晚上的 LHS 谱序列。

名字重要。没有名字你是孤独的。有了名字你就有从未谋面的合作者。