If $N$ is a normal subgroup of a group $G$ such that $G/N= \mathbb{Z}$. Suppose that the classifying space of $G$ is a finite CW-complex of dimension $n$. Does it follow that the classifying space of $N$ is a finite CW-complex of dimension $n-1$ ?
-
5$\begingroup$ The classifying space is not unique up to homeomorphism, so it should probably be phrased with "admits as classifying space a CW-complex of dimension $d$". $\endgroup$– YCorCommented Oct 9, 2020 at 19:23
-
17$\begingroup$ This fails on the easiest possible example: let G be the free group on 2 generators and let N be the kernel of the map to the integers taking both generators to 1. $\endgroup$– Andy PutmanCommented Oct 9, 2020 at 19:40
-
3$\begingroup$ Andy P: I wrote an answer which turns out to include your comment above before I had seen your comment - apologies for appearing to copy. $\endgroup$– IJLCommented Oct 12, 2020 at 14:53
3 Answers
No. Baumslag-Solitar groups of type $(1, n)$, which are semidirect products $\Bbb Z \ltimes \Bbb Z[1/n]$ have finite two-dimensional classifying spaces, but $\Bbb Z[1/n]$ clearly cannot have one-dimensional classifying space.
-
4$\begingroup$ A variant in which $N$ is finitely generated: write the Baumslag-Solitar group as $G=\langle t,x\mid txt^{-1}=x^n\rangle$, with $n\ge 2$. Then in $G^2$, the subgroup $N$ generated by $\{(t,t^{-1}),(x,1),(1,x)\}$ is finitely generated, and normal. Then $G^2$ admits a 4-dimensional finite classifying space. But $N$ is not finitely presented so doesn't admit any finite classifying space. (Nevertheless I'm not claiming that $N$ has no 3-dimensional classifying space.) $\endgroup$– YCorCommented Oct 9, 2020 at 19:29
Famous examples come from the so-called Bestvina-Brady groups. Given a simplicial graph $\Gamma$, define the right-angled Artin group $A(\Gamma)$ as $$\langle \text{$u$ vertex of $\Gamma$} \mid [u,v]=1 \text{ if $\{u,v\}$ is an edge of $\Gamma$} \rangle.$$ Sending each generator to $1$ yields a morphism $A(\Gamma) \to \mathbb{Z}$, and its kernel $BB_\Gamma$ is referred to as a Bestvina-Brady group.
If $\Gamma$ is finite, then $A(\Gamma)$ admits a finite classifying space of dimension $\mathrm{clique}(\Gamma)$ (:= maximal cardinality of a complete subgraph in $\Gamma$), but $BB_\Gamma$ may not have good finiteness properties.
Theorem. $BB_\Gamma$ is of type $F_n$ (i.e. admits a classifying space with finite $n$-skeleton) if and only if the flag completion of $\Gamma$ is $(n-1)$-connected.
If $\Gamma$ is a pair of isolated vertices, then we recover the example mentioned by Andy Putman in the comments. Here, $BB_\Gamma$ is not even finitely generated (because $\Gamma$ is not connected). If $\Gamma$ is a $4$-cycle, then $BB_\Gamma$ is finitely generated but not finitely presented, so it does not admit a classifying space with a finite $2$-skeleton. (This example was initially due to Stallings and Bieri.) You can find examples which are of type $F_n$ but not $F_{n-1}$ for every $n \geq 2$ by considering high dimensional spheres.
However, the initial question becomes more interesting if we assume that $N$ already admits a finite classifying space:
Question: Let $G$ be a group and $N \lhd G$ a subgroup such that $G/N \simeq \mathbb{Z}$. Assume that $G$ admits a finite classifying space of dimension $d$ and that $N$ admits a finite classifying space. Is it true that $N$ admits a finite classifying space of dimension $d-1$?
No, not even in the case when $G$ has a 1-dimensional classifying space: if $G$ is free of rank at least two, then for any $N$ with $G/N\cong \mathbb{Z}$, $N$ will be free of infinite rank and so $N$ will need a 1-dimensional classifying space too.
What you get easily is that if $G$ has cohomological dimension $n$ (modulo Eilenberg-Ganea issues for $n=2$ this is the same as the minimal dimension of a classifying space for $G$) then $N$ with $G/N\cong \mathbb{Z}$ cannot have a classifying space whose dimension is lower than $n-1$. Also easily $N$ does have a classifying space of dimension $n$ provided that $G$ does. But both of these possibilities can and do occur.
Concerning the question of whether $N$ can have a finite classifying space, the Euler characteristic gives a simple obstruction. If the Euler characteristic of the classifying space for $G$, $\chi(G)$ is non-zero, then $N$ cannot have a finite classifying space, because Euler characteristics multiply for group extensions and $\chi(\mathbb{Z})=0$, giving $\chi(G)=\chi(N)\chi(\mathbb{Z})=0$ whenever $\chi(N)$ is well-defined.