Does there exist a finitely generated non-finitely presentable group $$ G=\langle S \mid R\rangle $$ with $S$ finite, where we can enumerate $R=\{r_1,r_2,\ldots\}$ such that for every finite subset $I\subset \mathbb{N}$, the group $$ G_I=\langle S \mid \{r_i\mid i\in I\}\rangle $$ is virtually free?

One potential candidate is the lamplighter $\mathbb{Z}_2\wr\mathbb{Z}$ with the presentation $$ \langle a,t \mid a^2, [a,t^kat^{-k}] \textrm{ for all } k\in\mathbb{Z}\rangle. $$

(edit) Let me ask a more specific question related to lamplighters. Is it true that if $$ G_I=\langle a,t \mid a^2, [a,t^kat^{-k}] \textrm{ for all } k\in I\rangle $$ is virtually free for some finite set $I\subset\mathbb N$, then for every large enough $k$, $G_{I\cup\{k\}}$ is also virtually free?

  • $\begingroup$ If you assume $S$ to be finite, you should say it. $\endgroup$
    – YCor
    Feb 20, 2019 at 19:18
  • 2
    $\begingroup$ @YCor I have to confess that I have been guilty of writing "Let $G = \langle X \rangle$ be a finitely generated group'' without explicitly saying that $X$ is assumed to be finite. $\endgroup$
    – Derek Holt
    Feb 20, 2019 at 19:21
  • 4
    $\begingroup$ The truncated presentations $\Gamma_J=\langle a,t\mid a^2,[a,t^kat^{-k}]:k\in J\rangle$ are HNN extensions of finite groups, hence virtually free, when $J$ is a segment (i.e. equal to $\{1,\dots,n\}$ -- we can ignore those relators for $k\le 0$ since they are redundant). $\endgroup$
    – YCor
    Feb 20, 2019 at 19:24
  • 3
    $\begingroup$ @YCor: I think the group $G_{\{1,3\}}$ is not virtually free. The kernel of the map to $\mathbb{Z}$ given by $a\mapsto 0, t\mapsto 1$ is a right-angled Coxeter group with vertices labelled by $a_k, k\in\mathbb{Z}$ ($a_k=t^kat^{-k}$ in $G_I$). Then I think $\langle a_0, a_1, a_2, a_3\rangle \cong (\mathbb{Z}_2 \ast \mathbb{Z}_2) \times (\mathbb{Z}_2\ast \mathbb{Z}_2)$, so is not virtually free. $\endgroup$
    – Ian Agol
    Feb 20, 2019 at 23:58
  • 3
    $\begingroup$ @YCor, A RA Coxeter group is virtually free iff the graph is chordal. This is not the case for a 4-gon or higher. $\endgroup$ Feb 21, 2019 at 12:11

1 Answer 1


Here are some observations on your question. Firstly, consider the case that we take the presentation $\langle S | r_i\rangle$ for some $i$. This is a 1-relator group, and it will be virtually free iff $r_i$ is a power of a primitive element of $S$ (see Theorem 3 of this paper in the torsion case; and otherwise this answer and induction in the torsion-free case).

As a special case, suppose there is a relator $r_i$ which is not a proper power. Then it must be a primitive element in $S$, so $\langle S | r_i\rangle$ must be a free group on 1 fewer generator. Thus by induction on the rank, we may assume that all of the $r_i$ are proper powers of primitive elements.

Maybe there is a generalization of the theory of 1-relator groups (Freiheitzatz and Magnus-Moldovanskii hierarchy) to 1-relator quotients of virtually free groups? If so, the above arguments might carry over to this context. See results of Howie for some progress on this.

  • 1
    $\begingroup$ But the easiest argument is that $\langle a,t \mid [a,a^t] \rangle \cong \langle a,b,t \mid [a,b] = 1, b = a^t \rangle$ is an HNN extension of ${\mathbb Z}^2$ and hence has ${\mathbb Z^2}$ as a subgroup and cannot be virtually free. Of course that still requires the theory of HNN extensions. $\endgroup$
    – Derek Holt
    Feb 21, 2019 at 7:53
  • $\begingroup$ @IanAgol Thanks, this seems like a very interesting approach in general, I just wondered if the community knew of an example before diving into this. $\endgroup$
    – DavidHume
    Feb 21, 2019 at 8:53
  • $\begingroup$ Perhaps it's worth pointing out that it's not terribly difficult to ensure that all the one-relator subpresentations are free. Just add an extra generator $t$ to $S$, add the relation $t$, and change every $r_i$ to $r_it$. Of course, it's much harder to do something similar for larger subsets of relators, but then one gets out of the theory of one-relator groups. $\endgroup$
    – HJRW
    Feb 21, 2019 at 9:37
  • $\begingroup$ @DerekHolt no need of HNN theory to see that it contains $\mathbf{Z}^2$ (generated by $\{a,a^t\}$): instead just map onto $\mathbf{Z}\wr\mathbf{Z}$. $\endgroup$
    – YCor
    Feb 21, 2019 at 11:29
  • $\begingroup$ I suspect now that such presentations cannot exist. I think at least one should be able to use small cancellation theory to rule out presentations where the power is high, maybe at least 6. $\endgroup$
    – Ian Agol
    Feb 21, 2019 at 19:58

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.