# Cohomological dimension of a group acting on a cellular complex

Let $G$ be a group acting on $X$, $X$ a cellular complex and $cd(G)$ the cohomological dimension of $G$.

2 things:

(1) I'm looking for a reference (or proof!) of this:

Suppose $X$ is acyclic. Then $cd(G) \leq max_{\sigma} \space cd(Stab(\sigma) + dim \space \sigma$, where $\sigma$ runs over the cells of $X$.

(2) If $X$ isn't acyclic, can anything be said about $cd(G)$?

-
Presumably you want the action to be faithful? –  Qiaochu Yuan Sep 15 '10 at 16:40
@Qiaochu: Not necessarily; e.g. when X is a point the whole group is the stabilizer. –  Tyler Lawson Sep 15 '10 at 17:43

If you don't require $X$ to be acyclic, then you can't say anything. Indeed, every group acts freely on its Cayley graph, which is a 1-dimensional cell complex.

You're not even saved by assuming the $X$ is highly connected. By attaching cells to the Cayley graph in an equivariant manner, you can obtain a $k$-connected $(k+1)$-dimensional complex on which the group acts freely.

-
could we just use EG? –  Sean Tilson Sep 16 '10 at 2:25
Well, EG is in general infinite-dimensional, but you could take its k-skeleton. –  Nikita Sep 16 '10 at 16:10

Reference: Serre, J-P.(1971) "Cohomologie des groupes discrets," Ann. Math. Studies 70, 77-169 (Proposition 11, page 93). Serre credits this to Quillen, and I've never succeeded in locating this in any of Quillen's papers. Does anyone know where it is?

Proof: Use the equivariant cohomology spectral sequence.

-
The proof that I have in mind involves filtering the space $X\times_GEG$ in two ways -- by skeleta of $BG$ and by skeleta of $X$ -- and comparing the information that these two filtrations yield. Each of the two filtrations leads to a spectral sequence, I suppose (although on each side you can do the job without that tool). I don't know which of these two is what Bruce means by "the equivariant cohomology spectral sequence"! –  Tom Goodwillie Sep 15 '10 at 18:26
Its when you filter by skeleta of $X$; the other is the Serre spectral sequence. –  Oscar Randal-Williams Sep 15 '10 at 19:27
Tom, does looking at $X \times _G EG→X \times _{Stab(\sigma)} EG$ end up bounding $cd(G)$ by dim$\Sigma ^n BStab(\sigma) = n+dim BStab(\sigma)$? (n = dim $\sigma$) –  Romeo Sep 15 '10 at 20:08

If $X$ is a free contractible $G$-complex, this is the Eilenberg-Ganea Theorem

-
Your link seems to go through your university proxy, so I can't use it. Can you fix that? –  Nikita Sep 15 '10 at 19:01
Oops, sorry. I've linked instead to Math Reviews, hope that works. –  Mark Grant Sep 15 '10 at 21:14
If I've parsed the statement of the Eilenberg--Ganea theorem correctly, in their set up the action of $G$ on $X$ is free. –  HJRW Nov 1 '10 at 23:13
Yes. Then (1) says $cd(G)\leq \dim(X)$, and (assuming $G$ is discrete) this says $cd(G)$ is bounded above by its geometric dimension, ie the smallest dimension of a $BG$...Ah, I guess I meant to say "this is (the trivial part of) the Eilenberg-Ganea Theorem"... –  Mark Grant Nov 2 '10 at 10:04