A note on higher coherence of graphs of groups

Abstract.

For $n\in\mathbb{N}$, a group is called $n$-coherent if every subgroup of type $\mathsf{F}_{n}$ is of type $\mathsf{F}_{n+1}$. For $n\geq 1$, we observe that graphs of groups with $n$-coherent vertex groups and virtually poly-cyclic edge groups are $n$-coherent. We deduce the $n$-coherence of certain right-angled Artin groups.

1. Introduction

A group is said to be coherent if every finitely generated subgroup is finitely presented. This notion has many connections, e.g., to low-dimensional topology, geometric group theory, graph theory, and ring theory [Wis20]. The subject of this note is a higher version of coherence.

For $n\in\mathbb{N}$, a group $G$ is of type $\mathsf{F}_{n}$ (resp. of type $\mathsf{F}_{\infty}$) if there exists a CW-model for the classifying space $BG$ with finite $n$-skeleton (resp. of finite type). Every group is of type $\mathsf{F}_{0}$. For a group being of type $\mathsf{F}_{1}$ (resp. of type $\mathsf{F}_{2}$) is equivalent to being finitely generated (resp. finitely presented).

Definition 1.1.

Let $n,m\in\mathbb{N}\cup\{\infty\}$ with $n<m$. We say that a group $G$ is $(n,m)$-coherent if every subgroup of $G$ that is of type $\mathsf{F}_{n}$ is of type $\mathsf{F}_{m}$.

Definition 1.1 generalises the usual notions: $(0,1)$-coherence is being Noetherian (also called slender), $(1,2)$-coherence is coherence, and $(n,n+1)$-coherence is recently called $n$-coherence [KV23, Vid, KV25]. Clearly, $(n,m)$-coherence implies $(n^{\prime},m^{\prime})$-coherence for $n\leq n^{\prime}<m^{\prime}\leq m$ and is preserved under taking subgroups. Moreover, $(n,m)$-coherence passes to finite index overgroups and hence is invariant under commensurability. Virtually poly-cyclic groups are $(0,\infty)$-coherent. Groups of geometric dimension $\leq n$ are $(n,\infty)$-coherent. The direct product of free groups $(F_{2})^{k}$ is $(n,n+1)$-coherent if and only if $n\geq k$.

The following is a classical combination theorem for coherence of graphs of groups.

Theorem 1.2 (Karrass–Solitar [KS70]).

Let $G$ be the fundamental group of a graph of groups whose vertex groups are $(1,2)$-coherent and whose edge groups are $(0,1)$-coherent. Then $G$ is $(1,2)$-coherent.

We prove a combination theorem for higher coherence of graphs of groups. We use the convention that $\infty-1=\infty$.

Theorem 1.3.

Let $n,m\in\mathbb{N}\cup\{\infty\}$ with $1\leq n<m$. Let $G$ be the fundamental group of a graph of groups whose vertex groups are $(n,m)$-coherent and whose edge groups are

$$\begin{cases}(0,m-1)\text{-coherent}&\text{if }n=1;\\ (1,m-1)\text{-coherent}&\text{if }n\geq 2.\end{cases}$$

Then $G$ is $(n,m)$-coherent.

Theorem 1.2 was crucial to characterise the coherent right-angled Artin groups. A simplicial complex is flag if every complete subgraph in its 1-skeleton spans a simplex. Associated to a finite flag simplicial complex $L$ is the right-angled Artin group $A_{L}$ that has as generating set the vertices of $L$, subject only to the relations that two generators commute if the corresponding vertices are adjacent in $L$ [Kob22]. A graph is chordal if it does not contain a cycle of length $\geq 4$ as a full subgraph. It is a classical result of Dirac [Dir61] that in chordal graphs, minimal separating sets are cliques. It follows that a flag simplicial complex with chordal $1$-skeleton is a “tree of simplices” in the following sense.

Definition 1.4.

We define $\mathcal{T}_{1}$ to be the smallest class of finite flag simplicial complexes that satisfies the following:

1. All simplices (of all dimensions) lie in $\mathcal{T}_{1}$;

2. If $L_{1},L_{2}\in\mathcal{T}_{1}$ and $L_{0}$ is a full subcomplex of both $L_{1}$ and $L_{2}$ that is a simplex or is empty, then the gluing $L_{1}\cup_{L_{0}}L_{2}$ lies in $\mathcal{T}_{1}$.

Theorem 1.5 (Droms [Dro87]).

Let $L$ be a finite flag simplicial complex. Then the following are equivalent:

1. The group $A_{L}$ is $(1,\infty)$-coherent;

2. The group $A_{L}$ is $(1,2)$-coherent;

3. The $1$-skeleton of $L$ is a chordal graph;

4. The simplicial complex $L$ lies in the class $\mathcal{T}_{1}$.

We consider similar classes of simplicial complexes that allow for more vertex spaces and for edge spaces in $\mathcal{T}_{1}$.

Definition 1.6.

Let $n\in\mathbb{N}$ with $n\geq 2$. We define $\mathcal{T}_{n}$ to be the smallest class of finite flag simplicial complexes that satisfies the following:

1. All simplices (of all dimensions) lie in $\mathcal{T}_{n}$;

2. All finite flag simplicial complexes of dimension $\leq n-1$ lie in $\mathcal{T}_{n}$;

3. If $L_{1},L_{2}\in\mathcal{T}_{n}$ and $L_{0}$ is a full subcomplex of both $L_{1}$ and $L_{2}$ that lies in $\mathcal{T}_{1}$ or is empty, then the gluing $L_{1}\cup_{L_{0}}L_{2}$ lies in $\mathcal{T}_{n}$;

4. If $L\in\mathcal{T}_{n}$, then the cone $L\ast\Delta^{0}$ lies in $\mathcal{T}_{n}$.

As an application of Theorem 1.3, we deduce the higher coherence of certain right-angled Artin groups.

Theorem 1.7.

Let $n\in\mathbb{N}$ with $n\geq 2$ and let $L$ be a finite flag simplicial complex. If $L\in\mathcal{T}_{n}$, then the right-angled Artin group $A_{L}$ is $(n,\infty)$-coherent.

We do not expect the converse of Theorem 1.7 to hold. This leaves open a characterisation of higher coherence for right-angled Artin groups.

2. Inheritance of higher coherence

We recall some well-known inheritance results for finiteness properties [Bie81, Bro82, Geo08, HW21].

Theorem 2.1.

Let $G$ be the fundamental group of a finite graph of groups and let $n\in\mathbb{N}\cup\{\infty\}$. The following hold:

1. If all vertex groups are of type $\mathsf{F}_{n}$ and all edge groups are of type $\mathsf{F}_{n-1}$, then $G$ is of type $\mathsf{F}_{n}$;

2. If $G$ is of type $\mathsf{F}_{n}$ and all edge groups are of type $\mathsf{F}_{n}$, then all vertex groups are of type $\mathsf{F}_{n}$.

By Bass–Serre theory, there is a correspondence between graphs of groups and groups acting on trees (without inversions), under which vertex and edge groups correspond to vertex and edge stabilisers, respectively, of orbit representatives.

Theorem 2.2 ([GL17, Corollary 2.19]).

Let $G$ be a group and let $T$ be a cocompact $G$-tree. If $G$ is finitely presented, then there exists a cocompact $G$-tree $S$ with finitely generated edge stabilisers together with a $G$-map $S\to T$ that sends vertices to vertices and edges to non-trivial edge paths. In particular, the vertex and edge stabilisers of $S$ are subgroups of vertex and edge stabilisers of $T$, respectively.

Theorem 2.3.

Let $1\to N\to G\to Q\to 1$ be a group extension and let $n\in\mathbb{N}\cup\{\infty\}$. The following hold:

1. If $Q$ is of type $\mathsf{F}_{n}$ and $N$ is of type $\mathsf{F}_{n}$, then $G$ is of type $\mathsf{F}_{n}$;

2. If $G$ is of type $\mathsf{F}_{n}$ and $N$ is of type $\mathsf{F}_{n-1}$, then $Q$ is of type $\mathsf{F}_{n}$.

These inheritance results for finiteness properties give rise to corresponding inheritance results for higher coherence.

Proof of Theorem 1.3.

Let $H$ be a subgroup of $G$ that is of type $\mathsf{F}_{n}$. We show that $H$ is of type $\mathsf{F}_{m}$. Consider the Bass–Serre tree $T$ associated to $G$ as an $H$-tree by restriction. Since the stabilisers of the $H$-tree $T$ are subgroups of the stabilisers of the $G$-tree $T$, they inherit the corresponding coherence properties. Since $H$ is finitely generated, there exists a cocompact $H$-subtree $T^{\prime}$ of $T$. In particular, the vertex and edge stabilisers of the $H$-tree $T^{\prime}$ are subgroups of vertex and edge stabilisers of the $H$-tree $T$, respectively.

We treat the two cases $n=1$ and $n\geq 2$ separately. First, let $n=1$. Since the edge stabilisers of $T$ are $(0,m-1)$-coherent, the edge stabilisers of $T^{\prime}$ are of type $\mathsf{F}_{m-1}$ and, in particular, of type $\mathsf{F}_{1}$. By Theorem 2.1 () ‣ 2.1, the vertex stabilisers of $T^{\prime}$ are of type $\mathsf{F}_{1}$. Since the vertex stabilisers of $T$ are $(1,m)$-coherent, the vertex stabilisers of $T^{\prime}$ are of type $\mathsf{F}_{m}$. By Theorem 2.1 () ‣ 2.1, it follows that $H$ is of type $\mathsf{F}_{m}$.

Second, let $n\geq 2$. By Theorem 2.2, there exists a cocompact $H$-tree $S$ whose edge stabilisers are of type $\mathsf{F}_{1}$ and whose vertex and edge stabilisers are subgroups of vertex and edge stabilisers of $T^{\prime}$, respectively. Since the edge stabilisers of $T$ are $(1,m-1)$-coherent, the finitely generated edge stabilisers of $S$ are of type $\mathsf{F}_{m-1}$ and, in particular, of type $\mathsf{F}_{n}$. By Theorem 2.1 () ‣ 2.1, the vertex stabilisers of $S$ are of type $\mathsf{F}_{n}$. Since the vertex stabilisers of $T$ are $(n,m)$-coherent, the vertex stabilisers of $S$ are of type $\mathsf{F}_{m}$. By Theorem 2.1 () ‣ 2.1, it follows that $H$ is of type $\mathsf{F}_{m}$. ∎

Lemma 2.4.

Let $1\to N\to G\to Q\to 1$ be a group extension and let $n,m\in\mathbb{N}\cup\{\infty\}$ with $n<m$. The following hold:

1. If $Q$ is $(n,m)$-coherent and $N$ is $(0,m)$-coherent, then $G$ is $(n,m)$-coherent;

2. If $G$ is $(n,m)$-coherent and $N$ is of type $\mathsf{F}_{m-1}$, then $Q$ is $(n,m)$-coherent.

Proof.

Let $p\colon G\to Q$ denote the quotient map. For part (i), let $H$ be a subgroup of $G$ that is of type $\mathsf{F}_{n}$. Consider the group extension

$$1\to N\cap H\to H\to p(H)\to 1.$$

Since $N$ is $(0,m)$-coherent, the group $N\cap H$ is of type $\mathsf{F}_{m}$ and, in particular, of type $\mathsf{F}_{n-1}$. By Theorem 2.3 () ‣ 2.3, the group $p(H)$ is of type $\mathsf{F}_{n}$. Since $Q$ is $(n,m)$-coherent, the group $p(H)$ is of type $\mathsf{F}_{m}$. By Theorem 2.3 () ‣ 2.3, it follows that $H$ is of type $\mathsf{F}_{m}$.

For part (ii), let $K$ be a subgroup of $Q$ that is of type $\mathsf{F}_{n}$. Similarly, consider the group extension $1\to N\to p^{-1}(K)\to K\to 1$ and apply Theorem 2.3 () ‣ 2.3 and () ‣ 2.3. ∎

Remark 2.5.

In Theorem 1.3 for $n=1$, instead of assuming that the edge groups are $(0,m-1)$-coherent, it suffices that the intersection of any subgroup of $G$ that is of type $\mathsf{F}_{n}$ with any edge group is of type $\mathsf{F}_{m-1}$. Similarly, in Lemma 2.4 () ‣ 2.4, instead of assuming that $N$ is $(0,m)$-coherent, it suffices that the intersection of any subgroup of $G$ that is of type $\mathsf{F}_{n}$ with $N$ is of type $\mathsf{F}_{m}$.

3. Higher coherence of some right-angled Artin groups

We apply the inheritance results above to deduce the higher coherence of certain right-angled Artin groups.

Proof of Theorem 1.7.

We show that $(n,\infty)$-coherence is satisfied by the right-angled Artin groups associated to the flag simplicial complexes in Definition 1.6 () ‣ 1.6 and () ‣ 1.6 and is preserved under the operations in Definition 1.6 () ‣ 1.6 and () ‣ 1.6.

(i) If $L$ is a simplex, then the group $A_{L}$ is finitely generated free abelian and hence $(0,\infty)$-coherent and, in particular, $(n,\infty)$-coherent.

(ii) If $L$ is of dimension $\leq n-1$, then the group $A_{L}$ is of geometric dimension $\leq n$ and hence $(n,\infty)$-coherent [Geo08, Proposition 7.2.13].

(iii) For a gluing $L\coloneqq L_{1}\cup_{L_{0}}L_{2}$ with $L_{1},L_{2}\in\mathcal{T}_{n}$ and $L_{0}\in\mathcal{T}_{1}$ or $L=\emptyset$, the group $A_{L}$ splits as an amalgamated product $A_{L_{1}}\ast_{A_{L_{0}}}A_{L_{2}}$. (Here $A_{\emptyset}$ is the trivial group.) The group $A_{L_{0}}$ is $(1,\infty)$-coherent by Theorem 1.5. By induction on the number of vertices, we may assume that the groups $A_{L_{1}}$ and $A_{L_{2}}$ are $(n,\infty)$-coherent. Then the group $A_{L}$ is $(n,\infty)$-coherent by Theorem 1.3.

(iv) For a cone $CL\coloneqq L\ast\Delta^{0}$ with $L\in\mathcal{T}_{n}$, the group $A_{CL}$ splits as a direct product $A_{L}\times\mathbb{Z}$. By induction on the number of vertices, we may assume that the group $A_{L}$ is $(n,\infty)$-coherent. Then the group $A_{CL}$ is $(n,\infty)$-coherent by Lemma 2.4 () ‣ 2.4. ∎

We note that if a right-angled Artin group $A_{L}$ splits as an amalgamated product over a non-trivial $(1,2)$-coherent group, then there exists a separating full subcomplex of $L$ that lies in $\mathcal{T}_{1}$ [Hul21].

Remark 3.1.

Right-angled Artin groups are so-called graph products of $\mathbb{Z}$. Theorem 1.7 holds more generally for graph products of $(0,\infty)$-coherent groups when Definition 1.6 () ‣ 1.6 is modified with respect to their geometric dimension. The $(1,2)$-coherence of graph products was studied in [Var19].

Right-angled Artin groups can have subgroups with exotic finiteness properties, which are obvious obstructions to higher coherence. Let $L$ be a finite flag simplicial complex. The Bestvina–Brady group $H_{L}$ is defined as the kernel of the group homomorphism $A_{L}\to\mathbb{Z}$ that sends every standard generator of $A_{L}$ to $1\in\mathbb{Z}$. The finiteness properties of the group $H_{L}$ are determined by the higher connectivity of $L$ as follows.

Theorem 3.2 (Bestvina–Brady [BB97]).

Let $L$ be a finite flag simplicial complex and let $n\in\mathbb{N}$ with $n\geq 1$. Then the group $H_{L}$ is of type $\mathsf{F}_{n}$ if and only if $L$ is $(n-1)$-connected.

Corollary 3.3.

Let $L$ be a finite flag simplicial complex and let $n\in\mathbb{N}$ with $n\geq 1$. If the right-angled Artin group $A_{L}$ is $(n,n+1)$-coherent, then every full subcomplex of $L$ that is $(n-1)$-connected is $n$-connected.

For $n=1$, the converse of Corollary 3.3 holds by Theorem 1.5. For $n\geq 2$, we do not know if the converse of Corollary 3.3 holds.

Example 3.4.

Let $L$ be the flag simplicial complex spanned by the graph in Figure 1. The right-angled Artin group $A_{L}$ is isomorphic to $F_{2}\times F_{2}\times(\mathbb{Z}^{2}\ast\mathbb{Z})$. The group $A_{L}$ is not $(1,2)$-coherent by Theorem 1.5 because $L$ contains a cycle of length $4$ as a full subcomplex. The group $A_{L}$ is also not $(2,3)$-coherent by Corollary 3.3 because $L$ is $1$-connected and not $2$-connected. On the other hand, the group $A_{L}$ is $(4,\infty)$-coherent by Theorem 1.7 because $L$ is $3$-dimensional and hence lies in the class $\mathcal{T}_{4}$. We do not know if the group $A_{L}$ is $(3,\infty)$-coherent, although we suspect it to be. Note that $L$ does not lie in the class $\mathcal{T}_{3}$, while every full subcomplex of $L$ that is $2$-connected is $3$-connected.

Figure 1.

Remark 3.5.

There is an analogous notion of higher homological coherence [KV23, Fis]. Let $R$ be a ring and let $n\in\mathbb{N}$. A group $G$ is of type $\mathsf{FP}_{n}(R)$ (resp. of type $\mathsf{FP}_{\infty}(R)$) if the trivial $RG$-module $R$ admits a projective resolution that is finitely generated in degrees $\leq n$ (resp. in all degrees).

Let $n,m\in\mathbb{N}\cup\{\infty\}$ with $n<m$. We say that a group $G$ is homologically $(n,m)$-coherent over $R$ if every subgroup of $G$ that is of type $\mathsf{FP}_{n}(R)$ is of type $\mathsf{FP}_{m}(R)$. For $n\geq 2$, homological $(n,m)$-coherence over $\mathbb{Z}$ implies $(n,m)$-coherence because for $n\geq 2$, a group is of type $\mathsf{F}_{n}$ if and only if it is of type $\mathsf{F}_{2}$ and of type $\mathsf{FP}_{n}(\mathbb{Z})$. Inheritance results similar to the ones proved above hold also for higher homological coherence.

Groups of cohomological dimension $n$ over $R$ are homologically $(n,\infty)$-coherent over $R$ [Bro82, Proposition VIII.6.1]. Over the field $\mathbb{Q}$ (and suitably also over other fields), this can be improved if the top-dimensional $\ell^{2}$-Betti number vanishes [Fis, Proposition 2.3]: If $G$ is a locally indicable group with $\operatorname{cd}_{\mathbb{Q}}(G)=n$ and $b^{(2)}_{n}(G)=0$, then $G$ is homologically $(n-1,\infty)$-coherent over $\mathbb{Q}$.

Let $L$ be the flag simplicial complex spanned by the graph in Figure 1. The right-angled Artin group $A_{L}$ satisfies $\operatorname{cd}_{\mathbb{Q}}(A_{L})=4$ and by [DL03]

$$b^{(2)}_{4}(A_{L})=\dim_{\mathbb{Q}}H_{3}(L;\mathbb{Q})=0.$$

Hence $A_{L}$ is homologically $(3,\infty)$-coherent over $\mathbb{Q}$ (and, in fact, over every field [FHL24, Proposition 3.15][Fis, Proposition 2.3]).

Acknowledgements.

We thank Sam Fisher and Marco Linton for helpful comments. The first author was partially supported by the SFB 1085 Higher Invariants (Universität Regensburg, funded by the DFG).