Denote by $F_n$ the free group of rank $n$. We say that an automorphism $\phi\in Aut(F_n)$ is geometric if there exists a surface with boundary $M$ and a homeomorphism $h\colon M\to M$ such that $h$ induces $\phi$ on $\pi_1$.

Every automorphism of $F_2$ is geometric, but in higher rank geometric automorphisms are rare. I have seen several examples of nongeometric automorphisms, but all of which were of infinite order.

Question: Are there any examples of periodic (i.e. finite-order) nongeometric automorphism? Or can one show that periodic automorphisms are always geometric, possibly making use of the fact that every periodic automorphism can be realized as a simplicial graph automorphism? I wouldn't be surprised if we could nicely embed the graph into a suitable surface and extend the automorphism of the graph to a surface homeomorphism.

Observation: Every periodic automorphism $\phi$ fixes some nontrivial conjugacy class (maybe up to inversion), which is a necessary condition for $\phi$ to possibly be induced by a homeomorphism of a surface with connected boundary. But this observation really only plays a role when $n$ is even, as the fundamental group of a surface with connected boundary has even rank.

  • 1
    $\begingroup$ When you speak about automorphisms of free groups, do you really mean outer automorphisms? I ask because $h : M \to M$ does not induce a well-defined automorphism of $\pi_1 M$, at best it induces an outer automorphism. $\endgroup$
    – Lee Mosher
    May 6, 2013 at 21:22
  • $\begingroup$ The definition of geometric automorphisms also makes sense for actual (i.e. not outer) automorphisms, but basepoints would need to be preserved. If an outer auto $\mathcal{O}$ is geometrically realized by $h$ then any auto $\phi$ representing $\mathcal{O}$ is realized by the basepoint-preserving homeomorphism $p_{\phi}\circ h$, where $p_{\phi}$ is a suitable push map, a surface homeomorphism isotopic to the identity that pushes $h(\mathrm{basepoint})$ back to the basepoint. However, an arbitrary surface homeomorphism does not induce a well-defined actual free group auto, that's correct. $\endgroup$
    – Sebastian
    May 7, 2013 at 9:54

1 Answer 1


Assuming that you mean outer automorphisms of $F_n$ (as per my comment), the answer is that if $n \ge 3$ then there is a nongeometric finite order element of $Out(F_n)$, for example the outer automorphism class of $a_1 \mapsto a_1^{-1}$ and $a_i \mapsto a_i$ for $2 \le i \le n$.

For the proof, given $h : M \to M$ that induces a nonidentity, finite order, outer automorphism $\phi$ of $\pi_1 M$, I'll show that there does not exist a free factor of $\pi_1 M$ of rank $\ge 2$ whose conjugacy class is fixed by $\phi$ and whose elements all have conjugacy classes fixed by $\phi$, in contrast with the above example having such a free factor of rank $n-1$.

By Nielsen realization, after an isotopy of $h$ we may assume that $h$ is a finite order isometry with respect to some hyperbolic structure on $M$ with totally geodesic boundary. If there is a free factor $G < \pi_1 M$ of rank $\ge 2$ with conjugacy class fixed by $h$ then $h$ lifts to the $G$-cover $\tilde M$, and that lift restricts to an isometry of the convex hull $\tilde h : \mathcal{H}(\tilde M) \to \mathcal{H}(\tilde M)$. Since $rank(G) \ge 2$, the convex hull $\mathcal{H}(\tilde M)$ is itself a compact hyperbolic surface with totally geodesic boundary (if $rank(G)$ were equal to $1$ then the convex hull would just be a closed geodesic, and no contradiction would arise). Since $h$ fixes the conjugacy classes of all the elements of $G$, and since $G$ is a free factor (malnormality of $G$ suffices), it follows that $h$ induces the identity outer automorphism of $\pi_1(G)$. From this it follows in turn that $\tilde h$ is the identity isometry on $\mathcal{H}(\tilde M)$. It follows by projection that $h$ is locally the identity at some points of $M$, which by analytic continuation implies that $h$ is globally the identity. So $h$ induces the identity outer automorphism.


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.