Cannon–Thurston Map in Hyperbolic Geometry

• Cannon–Thurston maps are continuous extensions from the boundary of a hyperbolic subgroup to that of an ambient hyperbolic group, highlighting the interplay between intrinsic and ambient geodesics.
• They play a central role in understanding limit sets and ending laminations in Kleinian groups and 3-manifolds, with fibers encoding rich dynamical information.
• Existence depends on precise geometric conditions, as shown by counterexamples where hyperbolicity alone fails to guarantee a continuous extension.

A Cannon–Thurston map is a continuous extension to the boundary of a proper map between Gromov-hyperbolic spaces, most commonly the inclusion of a hyperbolic subgroup into a hyperbolic group or the inclusion of a lifted surface into hyperbolic $3$-space. In group-theoretic form, if $H\leq G$ are hyperbolic and $i:\Gamma_H\to \Gamma_G$ is the induced inclusion of Cayley graphs, the question is whether $i$ extends continuously to $\partial H\to \partial G$. In Kleinian and $3$-manifold settings, the resulting boundary map can be a sphere-filling curve whose fibers encode ending laminations; in geometric group theory more broadly, existence reflects precise control of intrinsic and ambient geodesics. The subject therefore sits at the intersection of boundary topology, subgroup distortion, convergence dynamics, and the asymptotic geometry of hyperbolic spaces (Mj, 2017). At the same time, hyperbolicity of both subgroup and ambient group is not enough: Cannon–Thurston maps do not exist for all hyperbolic subgroup inclusions (Baker et al., 2012).

1. Definition and coarse-geometric criterion

Suppose $Y$ and $X$ are Gromov-hyperbolic spaces and $i:Y\to X$ is a proper map. Writing $\widehat Y=Y\cup \partial Y$ and $H\leq G$0 for the Gromov compactifications, a Cannon–Thurston map is a continuous extension

$H\leq G$1

and its boundary restriction is

$H\leq G$2

For hyperbolic groups this specializes to the inclusion of Cayley graphs

$H\leq G$3

for a hyperbolic subgroup $H\leq G$4, and the existence problem becomes whether inclusion extends continuously to $H\leq G$5 (Mj, 2017).

A standard source of positive examples is quasiconvexity. If $H\leq G$6 is quasiconvex in $H\leq G$7, then inclusion is a quasi-isometric embedding, so the boundary map exists in the standard way. The interest of Cannon–Thurston theory lies in distorted settings, where the word metric on $H\leq G$8 and the metric induced from $H\leq G$9 may differ drastically, and there is no automatic reason that a sequence converging to a point of $i:\Gamma_H\to \Gamma_G$0 should define a consistent point of $i:\Gamma_H\to \Gamma_G$1 (Baker et al., 2012).

A decisive general criterion is due to Mitra. Let $i:\Gamma_H\to \Gamma_G$2 be a hyperbolic subgroup of a hyperbolic group $i:\Gamma_H\to \Gamma_G$3, and choose generating sets so that the Cayley graph $i:\Gamma_H\to \Gamma_G$4 is a subgraph of $i:\Gamma_H\to \Gamma_G$5. For each $i:\Gamma_H\to \Gamma_G$6, define $i:\Gamma_H\to \Gamma_G$7 to be the infimal number such that whenever $i:\Gamma_H\to \Gamma_G$8 is a geodesic in $i:\Gamma_H\to \Gamma_G$9 lying outside the ball of radius $i$0 about the identity $i$1 in $i$2, every geodesic in $i$3 connecting the endpoints of $i$4 lies outside the ball of radius $i$5 about $i$6 in $i$7. Then

$i$8

This criterion reduces boundary continuity to a coarse comparison of intrinsic and ambient geodesics, and it underlies both existence and non-existence results across the subject (Baker et al., 2012).

2. Origins in hyperbolic $i$9-manifolds and Kleinian groups

The subject began with the theorem of Cannon and Thurston for a closed hyperbolic $\partial H\to \partial G$0-manifold $\partial H\to \partial G$1 fibering over the circle with fiber a closed hyperbolic surface $\partial H\to \partial G$2. If $\partial H\to \partial G$3 and $\partial H\to \partial G$4, the inclusion

$\partial H\to \partial G$5

extends continuously to

$\partial H\to \partial G$6

In this case the boundary map is an equivariant Peano curve $\partial H\to \partial G$7, and the image is the limit set. This original example established that even highly distorted embeddings can admit remarkably structured boundary extensions (Mj, 2017).

The Kleinian-group theory that followed made Cannon–Thurston maps central to the study of limit sets and ending laminations. Mahan Mj proved that Cannon–Thurston maps exist for degenerate free groups without parabolics, i.e. handlebody groups, and then for arbitrary finitely generated Kleinian groups without parabolics; more generally, the argument works whenever each degenerate end admits a bi-Lipschitz Minsky model. In the free or handlebody case, if

$\partial H\to \partial G$8

is the extension of the orbit map, then for $\partial H\to \partial G$9,

$3$0

if and only if $3$1 are either the ideal endpoints of a leaf of an ending lamination of $3$2, or ideal boundary points of a complementary ideal polygon. In the general finitely generated Kleinian case, fibers are given by the transitive closure of the relations arising from lifts of ending laminations of degenerate ends (Mj, 2010).

The punctured-surface case with parabolics exhibits the same principle. For a simply degenerate punctured surface group without accidental parabolics, the Cannon–Thurston map

$3$3

identifies two distinct points if and only if they are ideal endpoints of a leaf of the ending lamination, or ideal boundary points of a complementary ideal polygon. Thus the non-injectivity of the boundary map is controlled exactly by the ending lamination, even in the presence of cusps (Das et al., 2010).

These results have major topological consequences. In the classical surface-group setting, the existence of the boundary extension is tied to local connectivity of the limit set. The survey literature presents Cannon–Thurston maps as a mechanism by which dynamics on the sphere at infinity determine geometry in the interior, with ending laminations appearing as the exact source of boundary identifications (Mj, 2017).

3. Extensions, laminations, and dendritic quotients

A second major axis of the theory concerns hyperbolic group extensions

$3$4

When $3$5, $3$6, and $3$7 are infinite word-hyperbolic groups, Mitra proved that the inclusion $3$8 extends continuously to

$3$9

He also associated to each $Y$0 an ending lamination $Y$1, and showed that for distinct $Y$2,

$Y$3

This gives an algebraic analogue of the ending-lamination theory from Kleinian groups (Field, 2019).

Elizabeth Field refined this direction by isolating the topology of the identifications associated to a single boundary direction $Y$4. For each $Y$5, the quotient of $Y$6 by the equivalence relation generated by $Y$7,

$Y$8

is homeomorphic to a dendrite. More precisely, the one-sided Cannon–Thurston map to the boundary of a semi-infinite hyperbolic stack has fibers exactly $Y$9, and Bowditch’s theorem implies that the target boundary is a dendrite. The conclusion is that the quotient of $X$0 by the $X$1-ending-lamination identifications is a compact, connected, locally connected, metrizable, uniquely arc-connected, loop-free continuum (Field, 2019).

For hyperbolic free-group extensions

$X$2

with $X$3 purely atoroidal and convex cocompact, the theory becomes especially explicit. The Cannon–Thurston map

$X$4

exists and is surjective. For each $X$5, there is a free arational tree $X$6 such that Mitra’s ending lamination $X$7 agrees with the dual lamination $X$8. Hence

$X$9

This identifies $i:Y\to X$0 as the quotient of $i:Y\to X$1 obtained by collapsing the leaves of the dual laminations $i:Y\to X$2, and it yields the uniform multiplicity bound

$i:Y\to X$3

for every $i:Y\to X$4 (Dowdall et al., 2015).

A further general structural result is that in most known settings in which a Cannon–Thurston map exists—normal hyperbolic subgroups of hyperbolic groups, trees of hyperbolic spaces satisfying the qi-embedded condition, and hyperbolic metric graph bundles—the map is uniformly finite-to-one. This generalizes previous finiteness theorems of Cannon–Thurston, Kapovich–Lustig, Dowdall–Kapovich–Taylor, and Ghosh, and gives an affirmative answer to Swarup’s question for trees of hyperbolic spaces (Bhattacharyya et al., 23 Mar 2026).

4. Failure of existence and ray-versus-continuity phenomena

For many years positive examples suggested that Cannon–Thurston maps might be a robust phenomenon for hyperbolic subgroup inclusions. Baker and Riley disproved this in decisive form. They constructed an explicit hyperbolic group

$i:Y\to X$5

and a free subgroup

$i:Y\to X$6

such that the inclusion $i:Y\to X$7 does not extend continuously to a map

$i:Y\to X$8

The ambient group is hyperbolic by a $i:Y\to X$9 small-cancellation argument, the subgroup is free of rank $\widehat Y=Y\cup \partial Y$0, and the obstruction is detected by Mitra’s criterion: there are geodesics in $\widehat Y=Y\cup \partial Y$1 staying arbitrarily far from the identity while ambient geodesics between the same endpoints pass through the identity (Baker et al., 2012).

This non-existence phenomenon is not isolated. Matsuda and Oguni showed that for every non-elementary hyperbolic group $\widehat Y=Y\cup \partial Y$2, one can embed $\widehat Y=Y\cup \partial Y$3 into another hyperbolic group $\widehat Y=Y\cup \partial Y$4 so that no Cannon–Thurston map

$\widehat Y=Y\cup \partial Y$5

exists, and more generally the same phenomenon occurs for every non-elementary relatively hyperbolic group with Bowditch boundaries. Their construction imports the Baker–Riley obstruction by amalgamating over a rank-$\widehat Y=Y\cup \partial Y$6 free subgroup (Matsuda et al., 2012).

Recent work has further separated pointwise ray landing from continuity of the full boundary map. Halder–Mj–Sardar introduced the notion of a ray Cannon–Thurston map: every geodesic ray in $\widehat Y=Y\cup \partial Y$7 starting at the identity lands at a unique point of $\widehat Y=Y\cup \partial Y$8, giving a set-theoretic map

$\widehat Y=Y\cup \partial Y$9

They proved that a ray Cannon–Thurston map may exist even when the inclusion does not extend continuously to a genuine Cannon–Thurston map. A central criterion shows that if two distinct landing rays of $H\leq G$00 have the same ambient endpoint and one is an ambient quasigeodesic, then $H\leq G$01 is not a CT pair. Their constructions recover a simple counterexample already lying at the heart of Baker–Riley’s examples and produce large new classes with “all rays land uniquely” but “no continuous boundary extension” (Halder et al., 22 Mar 2025).

5. Distortion, fibers, and dynamical consequences

Subgroup distortion and Cannon–Thurston existence interact in a subtle way. Hyperbolic hydra provide a family of hyperbolic groups $H\leq G$02 in which the subgroup distortion grows at least like the $H\leq G$03-th Ackermann function, yet the Cannon–Thurston map

$H\leq G$04

exists for every $H\leq G$05. These examples show that Cannon–Thurston maps can exist in the presence of arbitrarily heavy primitive recursive distortion. At the same time, the boundary maps are quantitatively wild: if $H\leq G$06 denotes the modulus of continuity, then for $H\leq G$07,

$H\leq G$08

for all sufficiently large $H\leq G$09 (Baker et al., 2012).

The structure of fibers has strong dynamical implications. In a very general setting, let $H\leq G$10 be a non-elementary word-hyperbolic group acting as a convergence group on a compact metrizable space $H\leq G$11, and let

$H\leq G$12

be a Cannon–Thurston map. Then if $H\leq G$13, the point $H\leq G$14 is not a conical limit point. More sharply, under the extra assumption that the action on $H\leq G$15 has no accidental parabolics, if the map $H\leq G$16 is not injective, then there exists a non-conical limit point $H\leq G$17 with

$H\leq G$18

Thus noninjectivity forces subtle dynamical pathologies not visible from multiple fibers alone (Jeon et al., 2014).

A measure-theoretic refinement appears in the classical fibered $H\leq G$19-manifold setting. For a closed hyperbolic $H\leq G$20-manifold fibering over the circle, the Cannon–Thurston map

$H\leq G$21

is surjective, finite-to-one, and space-filling, but pushforwards of many natural measures on $H\leq G$22 are mutually singular with respect to many natural measures on $H\leq G$23. The mechanism is geometric: geodesics sampled with respect to a pushforward measure spend a definite positive proportion of time close to a lifted fiber, while geodesics sampled with respect to Lebesgue measure or hitting measures from geometric random walks on the $H\leq G$24-manifold group spend an asymptotically negligible proportion of time there. This shows that topological regularity of the boundary map does not imply measure-theoretic regularity (Gadre et al., 5 Oct 2025).

6. Variants, boundary choices, and newer constructions

The formalism of Cannon–Thurston maps extends well beyond subgroup inclusions of hyperbolic groups, but the correct boundary is highly sensitive to the category under consideration. For CAT(0) groups with isolated flats, visual boundaries behave very differently from Bowditch boundaries. In that setting, visual-boundary Cannon–Thurston maps do not exist for infinite normal hyperbolic subgroups of non-hyperbolic CAT(0) groups with isolated flats, and they also do not exist for infinite infinite-index normal CAT(0) subgroups with isolated flats inside non-hyperbolic CAT(0) groups with isolated flats. The obstruction comes from flats: an element may act with North–South dynamics on the subgroup boundary but by translation on a flat in the ambient CAT(0) space, producing incompatible boundary dynamics (Beeker et al., 2018).

Coxeter groups provide a different geometric realization. For a Coxeter group $H\leq G$25 whose associated bilinear form has signature $H\leq G$26, the normalized action on an ellipsoid endowed with the Hilbert metric yields a $H\leq G$27-equivariant continuous surjection

$H\leq G$28

from the Gromov boundary of $H\leq G$29 to the limit set of $H\leq G$30. In this setting the limit set coincides with the set of accumulation points of roots. The later treatment including affine special subgroups shows that the map persists in the cuspidal case, though it need not be injective at cusp points (Mineyama, 2013).

Recent work has also produced genuinely new kinds of Cannon–Thurston constructions. For the weak unstable foliation $H\leq G$31 of a non-$H\leq G$32-covered Anosov flow on a closed hyperbolic $H\leq G$33-manifold, the leftmost universal circle $H\leq G$34 admits a continuous, surjective, $H\leq G$35-equivariant map

$H\leq G$36

Here the domain is not the boundary of a single hyperbolic plane or the boundary of a flow space, but a universal circle built from branching foliation data. This gives a new type of Cannon–Thurston map and implies pseudo-Anosov-type dynamics for the $H\leq G$37-action on $H\leq G$38 (Buckminster, 23 Apr 2026).

Taken together, these developments define the modern scope of Cannon–Thurston theory. In favorable settings the maps exist, are often uniformly finite-to-one, and their fibers are governed by ending laminations, dual laminations, or dendritic quotients. In unfavorable settings even hyperbolicity on both sides can fail to force continuity. The central question has therefore shifted from whether boundary maps exist in general to which geometric hypotheses make the asymptotic geometry of the subgroup and ambient space compatible enough for a continuous extension.