Finiteness of Cannon--Thurston fibers

Abstract: Let $Y\to X$ be a proper map between proper hyperbolic metric spaces. A Cannon--Thurston map is a continuous extension $\partial Y \to \partial X$. We prove that in most known settings in which a Cannon--Thurston map exists it is uniformly finite-to-one. This answers a question due to Swarup from Bestvina's problem list and generalizes previous results of Cannon--Thurston, Kapovich--Lustig, Dowdall--Kapovich--Taylor and Ghosh.

• The paper proves that Cannon–Thurston maps are uniformly finite-to-one in three major settings: hyperbolic group extensions, trees of hyperbolic spaces, and metric graph bundles.
• The authors replace lamination and index arguments with barycenter flows, contracting ladders, and a combinatorial counting lemma to obtain bounds depending only on geometric parameters.
• The results settle Swarup’s finiteness question and yield applications to commensurated subgroups, local cut points, slitherings, and rigidity, while leaving relatively hyperbolic analogues open.

Overview and main result

Let $Y \to X$ be a proper injective map between proper hyperbolic metric spaces. A Cannon–Thurston map is a continuous extension $\partial i : \partial Y \to \partial X$ of the inclusion to the Gromov boundaries. The paper by Bhattacharyya, Halder, Lazarovich, and Mj establishes that, in essentially all standard settings where such a map is known to exist, the boundary map is uniformly finite-to-one: there is a constant $N$, depending only on the uniform parameters of the setup (hyperbolicity constants, quasi-isometry constants, valence bounds, properness functions), such that every point of $\partial X$ has at most $N$ preimages in $\partial Y$.

The main theorem covers three settings:

1. $X$ the Cayley graph of a hyperbolic group and $Y$ that of a hyperbolic normal subgroup;
2. $X$ a hyperbolic tree of uniformly hyperbolic spaces with edge-to-vertex inclusions uniform qi embeddings (the Bestvina–Feighn combination theorem framework), with $Y$ a vertex space;
3. $\partial i : \partial Y \to \partial X$0 a hyperbolic metric graph bundle over a hyperbolic base with uniformly hyperbolic fibers and coarsely surjective barycenter maps (Mj–Sardar), with $\partial i : \partial Y \to \partial X$1 a fiber.

Case (2) answers affirmatively a question of Swarup from Bestvina's problem list (2603.22428), which asked whether the Cannon–Thurston map $\partial i : \partial Y \to \partial X$2 for a vertex space of a hyperbolic tree of spaces is finite-to-one. Case (1) generalizes prior finiteness results of Kapovich–Lustig, Dowdall–Kapovich–Taylor, and Ghosh, which were obtained for free normal subgroups via lamination descriptions and index theory of free group automorphisms.

Method: barycenter flows instead of laminations

The central methodological point is that the proof bypasses Cannon–Thurston laminations entirely. Previous approaches identified point-preimages via ending laminations and then used delicate index-theoretic arguments; here uniform finiteness is obtained directly from coarse geometry.

The key mechanism, sketched for a tree of spaces over a ray $\partial i : \partial Y \to \partial X$3, is as follows. Suppose $\partial i : \partial Y \to \partial X$4 lie in the same fiber of the Cannon–Thurston map $\partial i : \partial Y \to \partial X$5. The boundary flow maps $\partial i : \partial Y \to \partial X$6 transport these points to each fiber, and the coarse barycenter of each triple defines a ray

$\partial i : \partial Y \to \partial X$7

which is a quasigeodesic ray in $\partial i : \partial Y \to \partial X$8 with endpoint $\partial i : \partial Y \to \partial X$9. Consequently, barycenter rays arising from any two triples within a single fiber are asymptotic. Lemma 2.4-type arguments then force all these barycenters to remain within a uniformly bounded ball in some fixed fiber $N$0. A combinatorial counting lemma — Proposition 2.6 — completes the argument: if $N$1 is a set whose triples all have barycenters meeting a fixed ball $N$2, then the Gromov inner products $N$3 are uniformly bounded, so geodesic rays from $N$4 to points of $N$5 separate on a sphere of radius $N$6, giving

$N$7

where $N$8 bounds the valence. This yields an explicit, purely geometric bound on fiber cardinality.

Trees of hyperbolic spaces

For trees of spaces satisfying the qi-embedded condition, the argument requires the boundary flow technology developed by Kapovich–Sardar. Boundary flows along edges propagate boundary points of vertex spaces through the tree; Kapovich–Sardar's result that nontrivial fibers of the Cannon–Thurston map are witnessed by contracting ladders over a unique geodesic ray in the base tree reduces the problem to the ray case. Two structural lemmas do the work:

• Barycenter sections: if three distinct boundary points admit boundary flow along a ray, their successive barycenters form a uniform quasigeodesic section.
• Reduction to rays (Lemma 3.14): if $N$9 share a value under $\partial X$0, there is a unique $\partial X$1 such that the full fiber equals the fiber of the restricted map $\partial X$2.

Combining these, Theorem 3.15 proves part (I) for bundles over a ray and part (II) for arbitrary trees, with the bound depending only on $\partial X$3. The immediate corollary is that for a hyperbolic group splitting as a finite graph of hyperbolic groups with qi-embedded edge groups, every vertex-group inclusion has uniformly finite-to-one Cannon–Thurston map.

Metric graph bundles

The analogous theorem for metric graph bundles (Mj–Sardar's framework) proceeds by observing that the preimage of a geodesic ray in the base is quasi-isometric to a tree of spaces whose adjacent vertex spaces are uniformly quasi-isometric (via the natural nearest-neighbor maps). The ray case therefore follows from part (I) above; the general case again reduces to rays using Krishna–Sardar's uniqueness of the base direction witnessing multiple values. Since the Cannon–Thurston map for such bundles is surjective by recent work of Halder, one obtains the sharp statement $\partial X$4.

A notable corollary: if $\partial X$5 is a nonelementary hyperbolic normal or commensurated subgroup of infinite index in a hyperbolic group $\partial X$6, then $\partial X$7 is uniformly finite-to-one. The commensurated case goes beyond normality and connects to the coset-graph models of Margolis and Lazarovich–Margolis–Mj.

Applications

Two applications illustrate the force of the finiteness theorem:

• Commensurated subgroups: the authors give a short new proof of the main technical theorem of Lazarovich–Margolis–Mj: if a metric graph bundle over $\partial X$8 has fiber quasi-isometric to a one-ended hyperbolic group $\partial X$9, then $N$0 admits a local cut point, hence $N$1 virtually splits over a two-ended subgroup. The proof uses Bowditch's dendrite structure of $N$2: cut points of $N$3 have finite preimages, which must disconnect $N$4.
• Slitherings: translating Thurston's slithering manifolds into metric graph bundles (fibers uniformly quasi-isometric to $N$5 by Candel's theorem), the authors recover Thurston's theorem that the Cannon–Thurston map from a leaf boundary to $N$6 exists and is finite-to-one. They also deduce a rigidity statement: no codimension-one uniform foliation of a closed pinched-negatively-curved manifold of dimension $N$7 can have leaves of pinched negative curvature, since $N$8 ($N$9) has no local cut points.

Further extensions cover subtrees of spaces (using an induction showing any finite subset of a fiber admits boundary flow into a single vertex space) and the "subtrees of subspaces" setting of Halder–Sardar, where the bound multiplies the vertex-level bound $\partial Y$0 by the tree-of-spaces bound.

Limitations and open questions

The results are quantitative but not effective in a practical sense: the bounds depend exponentially on hyperbolicity and valence parameters through the counting lemma. The paper also concedes that the lamination-based description of point-preimages (Swarup's first question) remains a separate problem; the present methods bound fiber size without identifying the fibers. Most significantly, the relatively hyperbolic analogue remains open, as does the following general question posed at the end of the paper: if a hyperbolic group $\partial Y$1 acts on a proper hyperbolic space $\partial Y$2 without parabolics and the Cannon–Thurston map $\partial Y$3 exists, must it be finite-to-one? The type-preserving relatively hyperbolic version, with finiteness away from parabolic points, is likewise left open.

Conclusion

This paper settles Swarup's finiteness question in the affirmative across the principal known contexts for Cannon–Thurston maps — trees of hyperbolic spaces, metric graph bundles, and normal or commensurated hyperbolic subgroups — with uniform, parameter-only bounds. Its main technical contribution is a lamination-free strategy built on barycenter flows and contracting ladders, which both simplifies and substantially generalizes earlier index-theoretic arguments, while leaving the relatively hyperbolic setting as the natural next target.