Uniform exponent bounds for integral group homologyThanks: ORCID: https://orcid.org/0000-0001-8594-0699.Thanks: The author acknowledges the financial support from the Slovenian Research and Innovation Agency (ARIS), research core funding No. P1-0222, and project No. J1-50001.

Abstract.

We prove that, in each fixed degree, the exponent of the integral homology of a finite group is bounded solely in terms of the degree and the exponent of the group. The proof combines the solution of the restricted Burnside problem with a representability property of the bar construction and may be viewed as a torsion analogue of the method of acyclic models. We also use the Lyndon–Hochschild–Serre spectral sequence to obtain explicit bounds for finite solvable and nilpotent groups in terms of their derived length and nilpotency class.

1. Introduction

Let $G$ be a finite group and $H_{n}(G,\mathbb{Z})$ its $n$-th integral homology group. It is well known [1, Chapter III, Corollary 10.2] that $|G|\cdot H_{n}(G,\mathbb{Z})=0$. It is proved in [5] using Zel’manov’s solution of the restricted Burnside problem [8, 9] that given $e\geq 1$, there exists a constant $c=c(e)$ such that $c\cdot H_{2}(G,\mathbb{Z})=0$ for all (locally) finite groups $G$ of exponent $e$. Subsequently, considerable work has been devoted to obtaining good upper bounds for $\exp H_{2}(G,\mathbb{Z})$. A recent survey and further references are given in [6].

The purpose of this paper is to generalize the above-mentioned result of [5] to higher homology groups:

Theorem 1.1.

For every pair of positive integers $n$ and $e$ there exists an integer $N_{n}(e)$, depending only on $n$ and $e$, such that

$$N_{n}(e)H_{i}(G,\mathbb{Z})=0$$

for every finite group $G$ of exponent dividing $e$ and every $1\leq i\leq n$. In particular, $\exp H_{n}(G,\mathbb{Z})$ is bounded in terms of $n$ and $\exp(G)$ only.

One may take

$$N_{n}(e)=\prod_{q=1}^{n}|R(q,e)|,$$

where $R(q,e)$ stands for the $q$-generator restricted Burnside group of exponent $e$.

The argument uses the solution of the restricted Burnside problem together with a degreewise representability property of the bar complex. The proof is reminiscent of the method of acyclic models [4, Chapter VIII], but the models $R(q,e)$ are not acyclic. Instead, at the $q$-th stage one uses the fact that $H_{q}(R(q,e),\mathbb{Z})$ has a controlled annihilator. It may therefore be viewed as a torsion version of the acyclic-model argument.

The resulting bound stated in Theorem 1.1 is highly non-effective, since the orders of the restricted Burnside groups are in general enormous, see Vaughan-Lee and Zel’manov [7]. Better bounds are obtained when other invariants of the groups in question are allowed to be included in the estimates. We briefly present a method based on the Lyndon–Hochschild–Serre spectral sequence. We provide bounds for $\exp H_{n}(G,\mathbb{Z})$ in the cases when $G$ is solvable or nilpotent, in terms of the derived length or nilpotency class, respectively. The filtration argument used in the Lyndon–Hochschild–Serre bound follows an approach already taken by Hilton [3], and is also closely related to Browder’s work [2].

2. Uniform bounds for higher homology

In this section we show that, in each fixed degree, the exponent of the integral homology of a finite group is bounded in terms of the exponent of the group alone.

Fix a positive integer $e$, and denote by $\mathcal{C}_{e}$ the category of finite groups of exponent dividing $e$.

For $q\geq 1$, let $R(q,e)$ denote the restricted Burnside group of rank $q$ and exponent $e$, that is

$$R(q,e)=F_{q}/\bigcap_{N}N,$$

where $F_{q}$ is a free group of rank $q$, and the intersection is taken over all normal subgroups $N\trianglelefteq F_{q}$ such that $F_{q}/N$ is finite and has exponent dividing $e$. The group $R(q,e)$ is finite, by the solution of the restricted Burnside problem [8, 9], and its order depends on $e$ and $q$ only. Every finite $q$-generator group of exponent dividing $e$ is an image of $R(q,e)$.

Throughout the rest of this section, the exponent $e$ is fixed, and we use the shorthand notation $R_{q}=R(q,e)$. Fix a generator set $x_{1},\ldots,x_{q}$ of $R_{q}$, and denote $m_{q}=|R_{q}|$.

We use the standard inhomogeneous bar complex; see [1]. Given a set $X$, let $\mathbb{Z}[X]$ be the free abelian group on $X$. For a group $G$, denote by $C_{\bullet}(G)$ its inhomogeneous bar complex:

$$C_{0}(G) =\mathbb{Z},$$

$$C_{q}(G) =\mathbb{Z}[G^{q}],$$

where $q>0$. The basis element corresponding to the tuple $(g_{1},\ldots,g_{q})\in G^{q}$ in $C_{q}(G)$ is written in the bar notation as $[g_{1}\mid\cdots\mid g_{q}]$. The differential $\partial_{q}:C_{q}(G)\to C_{q-1}(G)$ is given by

(2.0.1) $$\begin{split}\partial_{q}[g_{1}\mid\cdots\mid g_{q}]={}&[g_{2}\mid\cdots\mid g_{q}]\\ &+\sum_{i=1}^{q-1}(-1)^{i}[g_{1}\mid\cdots\mid g_{i}g_{i+1}\mid\cdots\mid g_{q}]\\ &+(-1)^{q}[g_{1}\mid\cdots\mid g_{q-1}].\end{split}$$

In particular, $\partial_{1}=0$. The homology of this complex is $H_{\bullet}(G,\mathbb{Z})$. If $f:G\to H$ is a homomorphism, then the induced map $C_{\bullet}(G)\to C_{\bullet}(H)$ will be denoted by $f_{\#}$.

Lemma 2.1.

For every $G\in\mathcal{C}_{e}$ and every $q\geq 1$, the map

$$\Hom(R_{q},G)\longrightarrow G^{q},\qquad\varphi\longmapsto\bigl(\varphi(x_{1}),\ldots,\varphi(x_{q})\bigr),$$

is a bijection. Moreover, these bijections are natural in $G$. Consequently,

$$C_{q}(-)\cong\mathbb{Z}\bigl[\Hom(R_{q},-)\bigr]$$

as functors $\mathcal{C}_{e}\to\mathbf{Ab}$.

Proof.

A homomorphism $R_{q}\to G$ is determined by the images of the generators $x_{1},\ldots,x_{q}$, so the displayed map is injective. Conversely, a tuple $\mathbf{g}=(g_{1},\ldots,g_{q})\in G^{q}$ determines a unique homomorphism $\phi_{\mathbf{g}}:R_{q}\to G$ sending $x_{i}$ to $g_{i}$ for all $i$. This proves the bijection.

Let $f:G\to H$ be an arbitrary homomorphism in $\mathcal{C}_{e}$. The above bijections commute with the maps induced by $f$, that is, the diagram

is commutative. The bijections in question are therefore natural in $G$. Applying the free abelian group functor to the natural bijection $\Hom(R_{q},G)\cong G^{q}$ gives

$$\mathbb{Z}\bigl[\Hom(R_{q},G)\bigr]\cong\mathbb{Z}[G^{q}]=C_{q}(G).$$

The result is proved. ∎

Let $u_{q}=[x_{1}\mid\cdots\mid x_{q}]\in C_{q}(R_{q})$. Let $G$ be a finite group in $\mathcal{C}_{e}$. If $\mathbf{g}=(g_{1},\ldots,g_{q})\in G^{q}$, then $(\phi_{\mathbf{g}})_{\#}(u_{q})=[g_{1}\mid\cdots\mid g_{q}]$. Given $c\in C_{r}(R_{q})$, the rule $[g_{1}\mid\cdots\mid g_{q}]\mapsto(\phi_{\mathbf{g}})_{\#}(c)$ extends linearly to a natural transformation $C_{q}(-)\rightarrow C_{r}(-)$.

Proof of Theorem 1.1.

Recall that $m_{q}=|R_{q}|$ and define

$$N_{0}=1,\qquad N_{q}=m_{q}N_{q-1}.$$

Thus,

$$N_{q}=\prod_{j=1}^{q}|R(j,e)|.$$

We prove inductively that, for every $q\geq 1$, there exist natural homomorphisms $h_{i}^{(q)}:C_{i}(-)\rightarrow C_{i+1}(-)$, $1\leq i\leq q$, such that, with $h_{0}^{(q)}=0$,

(2.1.1) $$\partial_{i+1}h_{i}^{(q)}+h_{i-1}^{(q)}\partial_{i}=N_{q}\id_{C_{i}}$$

for every $1\leq i\leq q$. In other words, multiplication by $N_{q}$ is naturally null-homotopic in positive degrees through degree $q$.

We first treat the base case $q=1$. Here $R_{1}=\langle x_{1}\rangle\cong C_{e}$ and $m_{1}=e$. In the unnormalized bar complex of $R_{1}$, put

$$c_{1}=\sum_{j=0}^{e-1}[x_{1}^{j}\mid x_{1}]\in C_{2}(R_{1}).$$

Using $\partial_{2}[a\mid b]=[b]-[ab]+[a]$, we obtain

$$\partial_{2}c_{1}=\sum_{j=0}^{e-1}\bigl([x_{1}]-[x_{1}^{j+1}]+[x_{1}^{j}]\bigr)\\ =e[x_{1}],$$

because $x_{1}^{e}=1$. For $G\in\mathcal{C}_{e}$ and $g\in G$, let $\phi_{g}:R_{1}\to G$ be the homomorphism satisfying $\phi_{g}(x_{1})=g$, and define

$$h_{1}^{(1)}(G)[g]=(\phi_{g})_{\#}(c_{1}).$$

This construction is natural in $G$, and $\partial_{2}h_{1}^{(1)}(G)[g]=e[g]=N_{1}[g].$ This proves (2.1.1) for $q=1$.

Suppose the assertion is proved through degree $q-1$. To simplify notation, write $h_{i}=h_{i}^{(q-1)}$, where $i<q$. Consider the natural endomorphism $\alpha_{q}:C_{q}(-)\rightarrow C_{q}(-)$ defined by

(2.1.2) $$\alpha_{q}=N_{q-1}\id_{C_{q}}-h_{q-1}\partial_{q}.$$

We claim that $\alpha_{q}$ takes values in cycles. For $q=1$ this follows from $\partial_{1}=0$. If $q\geq 2$, the induction hypothesis in degree $q-1$ gives

(2.1.3) $$\partial_{q}h_{q-1}+h_{q-2}\partial_{q-1}=N_{q-1}\id_{C_{q-1}}.$$

Composing (2.1.3) on the right with $\partial_{q}$ and using $\partial_{q-1}\partial_{q}=0$ gives

$$\partial_{q}h_{q-1}\partial_{q}=N_{q-1}\partial_{q}.$$

Hence,

$$\partial_{q}\alpha_{q}=N_{q-1}\partial_{q}-\partial_{q}h_{q-1}\partial_{q}=0,$$

which proves the assertion, that is, $\alpha_{q}(G)(C_{q}(G))\subset Z_{q}(G)$ for every group in $\mathcal{C}_{e}$. In particular, $z_{q}=\alpha_{q}(R_{q})(u_{q})$ belongs to $Z_{q}(R_{q})$. By [1, Chapter III, Corollary 10.2] we have that $m_{q}[z_{q}]=0$ in $H_{q}(R_{q},\mathbb{Z})$. Therefore, there exists $c_{q}\in C_{q+1}(R_{q})$ such that

$$\partial_{q+1}c_{q}=m_{q}z_{q}=m_{q}\alpha_{q}(R_{q})(u_{q}).$$

Define $t_{q}(G):C_{q}(G)\to C_{q+1}(G)$ via

$$t_{q}(G)[g_{1}\mid\cdots\mid g_{q}]=(\phi_{\mathbf{g}})_{\#}(c_{q}).$$

Because this construction comes from the fixed $c_{q}\in C_{q+1}(R_{q})$, the maps $t_{q}(G)$ are natural in $G$. We have that

$$\partial_{q+1}t_{q}(G)[g_{1}\mid\cdots\mid g_{q}] =\partial_{q+1}(\phi_{\mathbf{g}})_{\#}(c_{q})$$

$$=(\phi_{\mathbf{g}})_{\#}(\partial_{q+1}c_{q})$$

$$=m_{q}\cdot(\phi_{\mathbf{g}})_{\#}(z_{q})$$

$$=m_{q}\cdot(\phi_{\mathbf{g}})_{\#}(\alpha_{q}(R_{q})(u_{q})).$$

As $\alpha_{q}$ is a natural transformation, we have that

$$(\phi_{\mathbf{g}})_{\#}(\alpha_{q}(R_{q})(u_{q}))=\alpha_{q}(G)(\phi_{\mathbf{g}})_{\#}(u_{q})=\alpha_{q}(G)[g_{1}\mid\cdots\mid g_{q}].$$

This, together with the calculation above, shows that

(2.1.4) $$\partial_{q+1}t_{q}(G)=m_{q}\alpha_{q}(G).$$

Now we define

$$h_{q}^{(q)}=t_{q}$$

and

$$h_{i}^{(q)}=m_{q}h_{i}$$

for $1\leq i<q$. For $i<q$, the induction hypothesis gives

$$\partial_{i+1}h_{i}^{(q)}+h_{i-1}^{(q)}\partial_{i} =m_{q}\left(\partial_{i+1}h_{i}+h_{i-1}\partial_{i}\right)$$

$$=m_{q}N_{q-1}\id_{C_{i}}$$

$$=N_{q}\id_{C_{i}}.$$

In degree $q$, the equation (2.1.4) gives

$$\partial_{q+1}h_{q}^{(q)}+h_{q-1}^{(q)}\partial_{q} =\partial_{q+1}t_{q}+m_{q}h_{q-1}\partial_{q}$$

$$=m_{q}\alpha_{q}+m_{q}h_{q-1}\partial_{q}$$

$$=m_{q}\left(N_{q-1}\id_{C_{q}}-h_{q-1}\partial_{q}+h_{q-1}\partial_{q}\right)$$

$$=N_{q}\id_{C_{q}}.$$

This completes the induction proof of (2.1.1).

Now fix $n$ and let $1\leq i\leq n$. If $z\in Z_{i}(G)$, then $\partial_{i}z=0$, and (2.1.1), with $q=n$, yields

$$N_{n}z=\partial_{i+1}h_{i}^{(n)}(z)+h_{i-1}^{(n)}(\partial_{i}z)=\partial_{i+1}h_{i}^{(n)}(z).$$

Thus, $N_{n}z$ is a boundary, and hence $N_{n}[z]=0$ in $H_{i}(G,\mathbb{Z})$. Therefore, $N_{n}H_{i}(G,\mathbb{Z})=0$ for every $1\leq i\leq n$, as required. ∎

We note that the choice $m_{q}=|R(q,e)|$ in the above proof is merely convenient. The only property required of $m_{q}$ is $m_{q}H_{q}(R(q,e),\mathbb{Z})=0$. Consequently, one may instead take $m_{q}=\exp H_{q}(R(q,e),\mathbb{Z})$, which gives the formally sharper bound

$$\exp H_{n}(G,\mathbb{Z})\mid\prod_{q=1}^{n}\exp H_{q}(R(q,e),\mathbb{Z}).$$

The same argument applies, without change, to locally finite groups of exponent dividing $e$.

3. Exponent bounds from the Lyndon–Hochschild–Serre spectral sequence

We record a general consequence of the Lyndon–Hochschild–Serre spectral sequence which gives useful exponent bounds for the homology of group extensions. Throughout this section all homology groups are taken with integral coefficients.

Let

$$1\longrightarrow N\longrightarrow G\longrightarrow Q\longrightarrow 1$$

be a group extension. The homological Lyndon–Hochschild–Serre spectral sequence is

$$E^{2}_{p,q}=H_{p}\bigl(Q,H_{q}(N,\mathbb{Z})\bigr)\Longrightarrow H_{p+q}(G,\mathbb{Z}),$$

see, for example, [1, Chapter VII, Section 6]. Here the action of $Q$ on $H_{q}(N,\mathbb{Z})$ is the one induced by conjugation in $G$.

The following standard filtration argument is closely related to [3]:

Proposition 3.1.

Let

$$1\longrightarrow N\longrightarrow G\longrightarrow Q\longrightarrow 1$$

be an extension of finite groups, and let $n\geq 1$. Then $\exp H_{n}(G,\mathbb{Z})$ divides

$$\exp H_{n}(Q,\mathbb{Z})\prod_{j=1}^{n}\exp H_{j}(N,\mathbb{Z}).$$

Proof.

Consider the Lyndon–Hochschild–Serre spectral sequence $E^{2}_{p,q}$ as above. Denote $a_{j}=\exp H_{j}(N,\mathbb{Z})$ and $b_{n}=\exp H_{n}(Q,\mathbb{Z})$. For $q>0$, the underlying abelian group $H_{q}(N,\mathbb{Z})$ is annihilated by $a_{q}$. Hence, $a_{q}E^{2}_{p,q}=0$ for $p\geq 0$ and $q\geq 1$. For $q=0$ we have $H_{0}(N,\mathbb{Z})\cong\mathbb{Z}$ with trivial $Q$-action, and therefore $E^{2}_{p,0}=H_{p}(Q,\mathbb{Z})$. In particular, $b_{n}E^{2}_{n,0}=0$. Every term on every subsequent page is a subquotient of the corresponding term on the preceding page. Consequently, an annihilator of $E^{2}_{p,q}$ also annihilates $E^{\infty}_{p,q}$. It follows that $a_{q}E^{\infty}_{n-q,q}=0$ for $1\leq q\leq n$, and $b_{n}E^{\infty}_{n,0}=0$.

Convergence of the spectral sequence gives a finite filtration

$$0=F_{-1}\subseteq F_{0}\subseteq\cdots\subseteq F_{n}=H_{n}(G,\mathbb{Z})$$

whose associated graded groups satisfy

$$F_{p}/F_{p-1}\cong E^{\infty}_{p,n-p}\qquad(0\leq p\leq n).$$

The quotient $F_{p}/F_{p-1}$ is annihilated by $a_{n-p}$ whenever $0\leq p<n$, while the last quotient $F_{n}/F_{n-1}$ is annihilated by $b_{n}$. We immediately deduce that $b_{n}a_{1}a_{2}\cdots a_{n}\,H_{n}(G,\mathbb{Z})=0$. ∎

Proposition 3.1 therefore becomes particularly simple when the kernel is abelian.

Corollary 3.2.

Let

$$1\longrightarrow A\longrightarrow G\longrightarrow Q\longrightarrow 1$$

be an extension of finite groups with $A$ abelian. Then, for every $n\geq 1$, $\exp H_{n}(G,\mathbb{Z})$ divides $(\exp A)^{n}\exp H_{n}(Q,\mathbb{Z})$.

As a first structural application, one obtains an explicit bound for metabelian groups.

Corollary 3.3.

Let $G$ be a finite metabelian group of exponent dividing $e$. Then for every $n\geq 1$, the exponent of $H_{n}(G,\mathbb{Z})$ divides $e^{n+1}$.

Proof.

Apply Corollary 3.2 to

$$1\longrightarrow G^{\prime}\longrightarrow G\longrightarrow G/G^{\prime}\longrightarrow 1.$$

Both $G^{\prime}$ and $G/G^{\prime}$ are abelian and have exponent dividing $e$, hence the result follows. ∎

The same argument can be iterated along the derived series.

Corollary 3.4.

Let $G$ be a finite solvable group of exponent dividing $e$ and derived length at most $d$. Then, for every $n\geq 1$, $\exp H_{n}(G,\mathbb{Z})$ divides $e^{\,1+n(d-1)}$.

Proof.

We argue by induction on $d$. The case $d=1$ is straightforward. If $d>1$, then $A=G^{(d-1)}$ is abelian and $G/A$ has derived length at most $d-1$. Corollary 3.2 and the induction hypothesis give that $\exp H_{n}(G,\mathbb{Z})$ divides $e^{n}\exp H_{n}(G/A,\mathbb{Z})$, which divides $e^{n}e^{1+n(d-2)}=e^{1+n(d-1)}$. ∎

An analogous argument gives a bound in terms of nilpotency class.

Corollary 3.5 (See [3], Lemma A.11).

Let $G$ be a finite nilpotent group of exponent dividing $e$ and nilpotency class at most $c$. Then, for every $n\geq 1$, $\exp H_{n}(G,\mathbb{Z})$ divides $e^{\,1+n(c-1)}$.

Proof.

We argue by induction on $c$. The case $c=1$ is the abelian case. If $c>1$, then $A=\gamma_{c}(G)$ is abelian and $G/A$ has nilpotency class at most $c-1$. Applying Corollary 3.2 and then the induction hypothesis gives $\exp H_{n}(G,\mathbb{Z})\mid e^{n}e^{1+n(c-2)}=e^{1+n(c-1)}$. ∎