Suppose that we have a finitely generated residually finite group $G = \langle g_1,\ldots,g_r \rangle$ and $H$ is a subgroup of $G$ with finite index $m$. Let $\phi$ be an automorphism on $G$.

Question: What is the bound for the smallest $n \in \mathbb{N} \setminus \{0\}$, such that $\phi ^n(H) = H$?

My thought so far: We know that $G$ has at most $(m!)^r$ number of subgroups with index $m$, so $n \leq (m!)^r$. I was wondering if the bound can be improved, it seems unlikely for the automorphism to go through all the subgroups of this index. Is it possible to show that $n$ is bounded by a polynomial (or even linear function) in $m$, i.e. $ n = \mathcal{O}_r(m^d)$ for some $d$?

  • $\begingroup$ Lots of typos and omissions. Is $r$ fixed? Is $N=H$? When write $n = \mathcal{O}(n^d)$ you presumably mean $n = O(m^d)$ but it's unclear what dependence on $r$ is allowed. $\endgroup$ Apr 3 at 18:26
  • $\begingroup$ Thank you for pointing them out, $r$ is fixed and any dependency on $r$ is allowed. $\endgroup$
    – ghc1997
    Apr 3 at 20:08

1 Answer 1


Let $G = \langle x,y,z \rangle = F_3$ be the free group on $3$ generators. If $w = w(x,y)$ is any word in $x$ and $y$, then $x$, $y$, and $zw$ generate $F_3$ because $x$ and $y$ generate $w$ and thus from $zw$ and $x$ and $y$ one can recover $x$, $y$, and $z$. Thus the map

$$\phi: x,y,z \rightarrow x,y,zw$$

is an automorphism of $F_3$. Now let $\psi: G\rightarrow S_m$ be the map such that:

$$\psi(x) = (1,2), \psi(y) = (1,2,3,\ldots,n), \psi(z) = e.$$

This is surjective because the first two elements generate $S_m$. Let $\sigma$ be any element of $S_m$. Then since $x$ and $y$ generate $S_m$, there is a word $w = w(x,y)$ such that $\psi(w) = \sigma$. Let $\phi$ denote the automorphism of $G = F_3$ which fixes $x$ and $y$ and sends $z$ to $zw$ for this $w$. Let $H$ denote the stabilizer of $1$ under the action of $G$ given by $\psi$, and let $K$ denote the normal closure of $H$, which is the kernel of $\psi$. If $\phi^n(H) = H$, then certainly $\phi^n(K) = K$ as well. But $\phi(z) = e$ so $z \in K$, whereas

$$\phi(\psi^n(z)) = \phi(zw^n) = \psi(w^n) = \sigma^n.$$

Hence $\phi^n(z)$ lies in $K$ only if $\sigma^n = e$. Thus any bound on $n$ (for $r = 3$) in terms of $m$ also gives a bound for the order of any element $\sigma \in S_m$. But this order is not bounded polynomially in $m$ (https://en.wikipedia.org/wiki/Landau%27s_function).

  • $\begingroup$ Free groups admit a huge number of subgroups of index $m$. If you take $F_3$, then if you take any of the $(m!)^3$ triples of elements in $S_m$ then with probability close to $1$ they generate $S_m$ or $A_m$, and so you get all of these subgroups. I expect, however, that the automorphism group of $F_3$ might be close to transitive on the subgroups whose normal closure has quotient $S_m$, so you get a transitive action on $n!^3$ points or so. It's not surprising then that this action has elements of large order. $\endgroup$ Apr 13 at 14:10
  • $\begingroup$ Thank you so much for your detailed answer and explanation!!👍 So we have to assume that the group $G$ has additional property for $n$ to be bounded by a polynomial. (Maybe the group itself needs to have polynomial growth) $\endgroup$
    – ghc1997
    Apr 13 at 15:45
  • $\begingroup$ @David If I answered your question you can accept the answer. $\endgroup$ Apr 19 at 18:48

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.