The underlying question is: how closely related are isomorphisms and automorphisms? More precisely, if $G_1$ and $G_2$ are given in terms of generating sets, when can we factor an isomorphism as an automorphism and an "obvious" map.

Let $H=\langle S\rangle$ be a fixed group, and let $G_1=H\ast_{A_1^{t_1}=B_1}$ and $G_2=H\ast_{A_2^{t_2}=B_2}$. Note that $G_1=\langle S, t_1\rangle$ and $G_2=\langle S, t_2\rangle$. If $\alpha:G_1\rightarrow G_2$ is an isomorphism then does there exists an automorphism $\beta\in\operatorname{Aut}(G_2)$ such that $\alpha$ factors as $\beta\circ \operatorname{id}$, where $\operatorname{id}: G_1\rightarrow G_2$ is the map with $\operatorname{id}(s)=s$ for all $s\in S$ and $\operatorname{id}(t_1)=t_2$.

In my precise setting, $H$ has Serre's property FA. Note that there are some subtleties here with respect to generating sets. For example, if $H=\mathbb{Z}$ then $BS(2, 3)=\langle a, t; t^{-1}a^2t=a^3\rangle$ is generated both by the pair $(a, t)$ and by $(a^2, t)$, but there is no automorphism of $BS(2, 3)$ which takes one pair to the other (this is the classical Baumslag-Solitar map, with non-trivial kernel). This observation is a subtlety rather than an outright obstruction though: $(a^2, t)$ is not given as an HNN-extension unlike in the question.

  • 5
    $\begingroup$ If $\alpha: G_1 \rightarrow G_2$ is an isomorphism, then every isomorphism from $G_1$ to $G_2$ is of the form $\beta \circ \alpha$ where $\beta \in \mathrm{Aut}(G_2)$. So the only question is whether mapping one generating set to the other extends to an isomorphism, which sounds like a hard problem in general (but maybe feasible for some special kinds of presentation). $\endgroup$
    – Colin Reid
    Aug 23, 2017 at 12:17
  • 1
    $\begingroup$ @Colin In my specific case I want to show non-isomorphism. I know that mapping one generating set to the other is not an isomorphism, and I want to know if I can stop there. My question is whether is it feasible for HNN-extensions; whether they are special enough. $\endgroup$
    – user113727
    Aug 25, 2017 at 6:47
  • $\begingroup$ It took me a while to notice your restriction that $H$ has Serre's property FA. Before I noticed that, I was ready to write out the following: For the case where $H$ is the fundamental group of a closed, oriented surface of genus $g \ge 1$, there are many counterexamples arising from the Thurston norm (see this very closely related answer mathoverflow.net/questions/104873/…). But I suppose you knew that. $\endgroup$
    – Lee Mosher
    Sep 25, 2017 at 21:09

1 Answer 1


I think you can mostly get what you want if in addition, you assume $\mathrm{Out}(H)$ is trivial and one of the HNN extensions is not ascending (let's say $G_1$).

Suppose $\alpha: G_1 \rightarrow G_2$ is an isomorphism. Then via $\alpha$, $G_1$ acts on the Bass–Serre tree for $G_2$ and vice versa. By Serre's property (FA), $\alpha(H) \le gHg^{-1}$ for some $g \in G_2$ and $\alpha^{-1}(gHg^{-1}) \le hHh^{-1}$ for some $h \in G_1$. So we end up with $H \le \alpha^{-1}(gHg^{-1}) \le hHh^{-1}$. Since it's a nonascending HNN extension, $H$ is not conjugate in $G_1$ to any proper overgroup of itself, so $H = \alpha^{-1}(gHg^{-1}) = hHh^{-1}$; after conjugating in $G_2$ we can then assume $\alpha(H) = H$. Since $\mathrm{Out}(H)$ is trivial, actually $\alpha$ is then inner on $H$, and by conjugating again in $G_2$, we get an isomomorphism from $G_1$ to $G_2$ that restricts to the identity on $H$.

On the other hand, if $\mathrm{Out}(H)$ is nontrivial, you could have $G_1 = H \ast_{A^{t_1}_1 = B_1}$, $G_2 = H \ast_{(\beta(A_1))^{t_2} = \beta(B_2)}$, where $t_2 =\beta t_1 \beta^{-1}$ for some noninner automorphism $\beta$ of $H$. In that case $G_1$ and $G_2$ are certainly isomorphic, but probably not in a way that restricts to the identity on $S$.

I'm not sure what happens in the case when $G_1$ and $G_2$ are both ascending.

  • $\begingroup$ I do not understand why $\alpha(H)=H$ implies that $\alpha$ is inner. For example, in $BS(2, 4)=\langle a, t; t^{-1}a^2t=a^4\rangle$ the automorphism $\psi: a\mapsto a, t\mapsto t^{-1}a^2ta^{-2}t$ acts trivially on the base group $\langle a\rangle$ but is not inner. (I know that this isn't the precise setting you wrote, but I am not convinced that $\alpha$ is inner in your setting, and this example illustrates what I think the issue is.) $\endgroup$
    – user113727
    Aug 25, 2017 at 16:01
  • $\begingroup$ I mean it's inner on $H$ (edited). I don't think you can control where the stable letter will go: given a group $G = H \ast_t$, then typically $G$ will also split as $G = H \ast_s$ for many other elements $s$. $\endgroup$
    – Colin Reid
    Aug 26, 2017 at 17:38

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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