If $N^2$ is a closed, orientable surface of genus at least $2$, and if $\phi$ is an (orientation-preserving) pseudo-Anosov mapping on $N$, then one can form the closed orientable 3-manifold $M^3$ by gluing the boundary components of $N^2\times [0,1]$ via $\phi$. In other words, $M^3$ is fibered over $S^1$, with fiber $N^2$. $M^3$ can also be given a hyperbolic structure (much like $N^2$ can). So my question is:

Does there exist a closed orientable hyperbolic manifold $M^k$, fibered over $S^1$ with a fiber $N^{k-1}$, when $k\neq 3$?

No such $M$ exists for $k=2$. It is also known (via Mostow rigidity) that for such an $M^k$ to exist for $k>3$, we cannot have the fiber $N^{k-1}$ hyperbolizable as well. But I don't know anything further than that.

  • 7
    $\begingroup$ It is impossible for even $k$ and wide-open for odd. An example would also give an example if non hyperbolic group of finite type without Baumslag-Solitar subgroups, existence if which is unknown. $\endgroup$
    – Misha
    Sep 19, 2013 at 17:36
  • 2
    $\begingroup$ For even $k$ this is just the Euler characteristic obstruction: It is nonzero for hyperbolic manifolds and zero for manifolds fibering over $S^1$. For odd $k$ my gut feeling is that such manifolds do not exist, but we have no tools for proving this. $\endgroup$
    – Misha
    Sep 19, 2013 at 18:52
  • 3
    $\begingroup$ The Euler characteristic of such a fibration is $0,$ so if $k$ is even, the volume of a hyperbolic such $M^k$ is also $0$ by Gauss-Bonnet... $\endgroup$
    – Igor Rivin
    Sep 19, 2013 at 18:54
  • 3
    $\begingroup$ One more thing that we know is that if such a fibration exists, then the fundamental group of the fiber cannot be word-hyperbolic: This is a corollary of the Rips theory (as the fiber group has an outer automorphism of infinite order). $\endgroup$
    – Misha
    Sep 19, 2013 at 19:53
  • 1
    $\begingroup$ There is an example in dimension 5 by Italiano, Martelli, and Migliorini arxiv.org/abs/2105.14795 $\endgroup$
    – user493370
    Oct 21, 2022 at 18:15

1 Answer 1


I make a remark here that is well-known to experts, that $\pi_1(M)$ is an extension of $\pi_1(N)$ by an infinite-order outer automorphism, explaining partly Misha's comment.

Consider the sequence $\pi_1(N^{k-1}) \to \pi_1(M^k) \overset{\varphi}{\to} \mathbb{Z} $. Choose $\alpha\in \pi_1(M)$ such that $\varphi(\alpha)=1\in \mathbb{Z}$. Then conjugation $g\mapsto \alpha^k g\alpha^{-k}, g\in \pi_1(N)$ gives an automorphism of $\pi_1(N)$. If this automorphism is conjugation by an element of $\pi_1(N)$ for some $k$, then there exists $h\in \pi_1(N)$ such that $\alpha^k g \alpha^{-k}=hgh^{-1}$ for all $g\in \pi_1(N)$. Let $\alpha'=h^{-1}\alpha^k$, then $\alpha' g= g \alpha'$ for all $g\in \pi_1(N)$. Consider the maximal cyclic subgroup $C<\pi_1(M)$ containing $\alpha'$ (there is a unique such group since $\pi_1(M)$ is the fundamental group of a closed hyperbolic manifold, so $C$ is the stabilizer of the axis of $\alpha'$ by the proof of Preissman's theorem). Also by Preissman's theorem, the subgroup generated by $\langle \alpha', g \rangle$ is abelian and therefore cyclic, and therefore $g\in C$. But this implies that $\pi_1(N)\leq C$, so $\pi_1(M)\leq C$, a contradiction. Thus, the conjugation by $\alpha$ maps to a infinite order element of $Out(\pi_1(N))$.

Then if $\pi_1(N)$ is hyperbolic, one can deduce that $\pi_1(N)$ splits over $\mathbb{Z}$ by Rips' theory. However, an aspherical closed $k-1$-manifold group cannot split over $\mathbb{Z}$ for $k>2$.

  • $\begingroup$ Thank you for the informative post. Let me just remark that there is a (slightly) different proof about $N$ not being hyperbolic, which I suggested in my post: using Mostow rigidity we can homotope $\alpha$ to be an isometry, and then the fact that the isometry group of $N$ is finite (indeed, isomorphic to $Out(\pi_1(N))$), we get $M$ finitely covered by $N\times S^1$. $\endgroup$
    – Steve D
    Sep 23, 2013 at 16:50
  • 2
    $\begingroup$ @SteveD: by $\pi_1(N)$ hyperbolic, I mean $\delta$-hyperbolic in the sense of Gromov. So, for example, this includes the case that $N$ admits a negatively curved metric, but might not be constant negative curvature (although it is more general than that). $\endgroup$
    – Ian Agol
    Sep 23, 2013 at 17:35

Your Answer

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

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