---
title: 'Bowditch Boundary: A Relatively Hyperbolic Perspective'
url: https://www.emergentmind.com/topics/bowditch-boundary
type: topic
---

# Bowditch Boundary: A Relatively Hyperbolic Perspective

Searching arXiv for recent and foundational papers on the Bowditch boundary.
The **Bowditch boundary** is the canonical compact boundary attached to a relatively hyperbolic group pair \((G,\mathcal P)\). In the standard geometric models, it is the Gromov boundary of a cusped or combinatorial cusped space built from \(G\) by attaching horoball-like pieces to left cosets of the peripheral subgroups, and it is equivalently the boundary associated to the coned-off relative geometry [1806]. It is the natural analogue, for relatively hyperbolic groups, of the Gromov boundary of a hyperbolic group; it records both conical directions and peripheral, or parabolic, directions at infinity, and it supports a geometrically finite convergence action of \(G\) [1806].

## 1. Definition and equivalent geometric models

A common setup is a **group pair** \((G,\mathcal P)\) with \(G\) finitely generated and \(\mathcal P\) a finite collection of finitely generated proper subgroups. One model forms the **combinatorial cusped space** \(X_{CH}(G,\mathcal P,S)\) by taking a Cayley graph of \(G\) and gluing a combinatorial horoball to each left coset \(gP\) of each \(P\in\mathcal P\); the Bowditch boundary is then the Gromov boundary
\[
\partial(G,\mathcal P),
\]
the boundary of that hyperbolic cusped space [1806]. A closely related formulation uses the relative Cayley graph \(\hat{}\) with respect to \(X\cup \mathcal H\), or equivalently the coned-off Cayley graph obtained by adjoining cone vertices for left cosets \(gH_i\) and edges of length \(1/2\) from elements of \(gH_i\) to the corresponding cone point [2212].

In the coned-off formulation used for Farb–Bowditch relative hyperbolicity, the Bowditch boundary may be written as
\[
\partial_B(G,\mathcal P)=\partial\hat\Gamma\cup\{*_P:P\in\mathcal P\},
\]
where \(\partial\hat\Gamma\) is the Gromov boundary of the coned-off graph and the additional points \(*_P\) represent parabolic directions [1710]. This description makes explicit that the relative boundary consists of hyperbolic boundary points together with isolated peripheral points.

Several quotient descriptions are also fundamental. If \(G\) is already hyperbolic and \(\mathcal H=\{H_1,\dots,H_n\}\) is an almost malnormal collection of proper quasiconvex subgroups, then
\[
\partial(G,\mathcal H)\cong \partial G/\mathcal L,
\]
where \(\mathcal L\) is the collection of \(G\)-translates of the limit sets \(g\partial H_i\); thus the Bowditch boundary is obtained from the Gromov boundary by collapsing each peripheral limit set to a point [1504]. In a different but analogous setting, for a relatively hyperbolic hierarchically hyperbolic group, the Bowditch boundary is the quotient of the HHS boundary obtained by collapsing each peripheral limit set to a point [2305].

A useful comparison theorem between relative models states that if \(G\) is hyperbolic relative to \(\mathcal P\), then the geodesic boundary \(\partial\hat{}\) embeds \(G\)-equivariantly and homeomorphically into \(\partial(G,\mathcal P)\), and the complement is countable [2212]. This makes precise the relation between the purely hyperbolic part of the relative geometry and the full Bowditch boundary.

## 2. Dynamical characterization and boundary points

The Bowditch boundary is characterized not only geometrically but dynamically. Bowditch and Yaman showed that a geometrically finite convergence action on a perfect metrizable compactum \(M\) is equivalent to \(M\) being equivariantly homeomorphic to \(\partial(G,\mathcal P)\) for an appropriate relatively hyperbolic structure [1806]. In this sense, the Bowditch boundary is the compactum on which \(G\) acts as a **geometrically finite convergence group**.

In the cusp-uniform formulation, \(G\) acts properly and isometrically on a proper \(\delta\)-hyperbolic space \(X\) with a \(G\)-invariant collection of disjoint open horoballs based at parabolic points, and the Bowditch boundary is the Gromov boundary of \(X\) [1708]. Every boundary point is then either a **conical limit point** or a **bounded parabolic point**, and under tame peripheral hypotheses Bowditch proved that if \(\partial(G,\mathbb P)\) is connected then it is locally connected, with every global cut point parabolic [1708].

The parabolic points admit a concrete geometric interpretation in cusped-space models. For a peripheral subgroup or coset \(H\), the boundary of the corresponding horoball is a singleton \(\{a\}\subset \partial G\); such points are called **parabolic endpoints** or **parabolic points**, and conversely a point is parabolic if it is the boundary point of some horoball [2503]. This description is central in later work on quasiconformal boundary maps, where horoball shadows encode the peripheral geometry.

For non-elementary relatively hyperbolic groups, the Bowditch boundary is uncountable and perfect [2401]. This gives the boundary the same large-scale compactness profile that the Gromov boundary has in the absolute hyperbolic case, while still retaining the additional parabolic structure that distinguishes the relative setting.

## 3. Topology, cohomology, and manifold phenomena

A major structural theorem identifies the Čech cohomology of the Bowditch boundary with relative group cohomology. If \((G,\mathcal P)\) is relatively hyperbolic and of type \(F_\infty\), then for every \(k\) there is an isomorphism of \(AG\)-modules
\[
H^k(G,\mathcal P;AG)\cong \check{H}^{k-1}(\partial(G,\mathcal P);A).
\]
This is the relative analogue of the Bestvina–Mess theorem for hyperbolic groups and shows that the Bowditch boundary encodes relative cohomology “at infinity” [1806].

The same work shows that the boundary reflects relative Poincaré duality. If \((G,\mathcal P)\) is relatively hyperbolic and type \(F\), then the following are equivalent: \((G,\mathcal P)\) is a \(PD(n)\) pair, and \(\partial(G,\mathcal P)\) is a homology \((n-1)\)-manifold and an integral Čech cohomology \((n-1)\)-sphere [1806]. In particular, if \((G,\mathcal P)\) is a type \(F\) relatively hyperbolic \(PD(n)\) pair, then \(\partial(G,\mathcal P)\) is a homology \((n-1)\)-manifold, and in dimension three one has
\[
(G,\mathcal P)\text{ is }PD(3) \iff \partial(G,\mathcal P)\cong S^2.
\]
This recovers the Tshishiku–Walsh theorem in the relative setting [1806].

The boundary dimension can also be read cohomologically. If \((G,\mathcal P)\) is type \(F\), relatively hyperbolic, and satisfies
\[
\operatorname{cd}(G)<\operatorname{cd}(G,\mathcal P),
\]
then
\[
\dim(\partial(G,\mathcal P))=\operatorname{cd}(G,\mathcal P)-1.
\]
This is the relative counterpart of the Bestvina–Mess dimension theorem [1806].

Local connectedness has been established in full generality for relatively one-ended pairs. If \((\Gamma,\mathbb P)\) is relatively hyperbolic and relatively one ended, then the Bowditch boundary \(M(\Gamma,\mathbb P)\) is locally connected; this removes Bowditch’s earlier restrictions on cardinality and on the peripheral subgroups [2204]. The same work states that if the boundary is connected, then every cut point is parabolic, and such a cut point with stabilizer \(P\in\mathbb P\) exists if and only if \(\Gamma\) splits relative to \(\mathbb P\) over a subgroup of \(P\) [2204].

## 4. Splittings, cut points, and deformation under fillings

The Bowditch boundary is closely tied to the algebraic splitting theory of relatively hyperbolic groups. A central result is that if \((G,\mathbb P)\) has tame peripherals, \(\partial(G,\mathbb P)\) is connected, and \(\partial(G,\mathbb P)\) is not homeomorphic to a circle, then non-parabolic local cut points are equivalent to splittings over \(2\)-ended subgroups: if \(G\) does not split over a \(2\)-ended subgroup, then the boundary contains no non-parabolic local cut point, while a splitting over a non-parabolic \(2\)-ended subgroup relative to \(\mathbb P\) produces such a local cut point [1708].

This local-cut-point theory yields a classification of certain \(1\)-dimensional Bowditch boundaries. For a \(1\)-ended relatively hyperbolic group with tame peripherals, under the assumptions of no splitting over a virtually cyclic subgroup, no peripheral splitting, one-ended peripherals, and \(1\)-dimensional boundary, the Bowditch boundary is homeomorphic to exactly one of
\[
S^1,\qquad \text{the Sierpiński carpet},\qquad \text{the Menger curve}
\]
[1708].

Global cut points are controlled by peripheral splittings. A theorem of Dasgupta used in Dehn filling theory states that if \((G,\mathcal P)\) is relatively hyperbolic and \(\partial(G,\mathcal P)\) is connected, then \(\partial(G,\mathcal P)\) has a cut point if and only if \((G,\mathcal P)\) has a nontrivial peripheral splitting [2209]. This gives a direct bridge between boundary topology and relative JSJ-type structure.

The Bowditch boundary also behaves predictably under sufficiently long Dehn fillings. If \((G,\mathcal P)\) has connected Bowditch boundary with no cut point, then for all sufficiently long \(\mathcal M\)-finite fillings
\[
(G,\mathcal P)\twoheadrightarrow (\bar G,\bar{\mathcal P}),
\]
the resulting boundary
\[
\partial(\bar G,\bar{\mathcal P}^{\infty})
\]
is connected and has no cut points [2209]. The same work emphasizes that the older virtually polycyclic hypothesis on peripheral subgroups is not needed for this connectedness theorem [2209].

A related picture appears for groups with isolated flats. There, Tran’s theorem identifies the Bowditch boundary as the quotient of the CAT(0) visual boundary obtained by identifying all points lying in the boundary of the same flat, and Bowditch-type splitting theorems can then be transferred to the CAT(0) boundary [1704]. This suggests that the Bowditch boundary often serves as the quotient object in which peripheral Euclidean phenomena are compressed to parabolic points.

## 5. Comparison maps, quasi-isometries, and boundary constructions

Boundary comparison theorems are a recurrent feature of the theory. The embedding
\[
\partial\hat{}\hookrightarrow \partial(G,\mathcal P)
\]
with countable complement is one such result, and it has measurable consequences: hyperfiniteness of the orbit relation on \(\partial\hat{}\) passes to the Bowditch boundary because adding only countably many points preserves hyperfiniteness in that context [2212].

For quasi-isometric classification, recent work extends Paulin’s theorem from hyperbolic to relatively hyperbolic groups. If \(\varphi:G_1\to G_2\) is a \(C\)-coarsely cusp-preserving \((\lambda,K)\)-quasi-isometry between relatively hyperbolic groups, then the induced map on Bowditch boundaries is quasiconformal with respect to the ring structures based at boundary points, with distortion
\[
\eta(t)=e^{A\ln(t)+B}=e^B\,t^A,
\]
for constants \(A\ge 1\), \(B\ge 0\) depending on \(\delta,C,K,R,\lambda\) [2503]. Conversely, a quasiconformal homeomorphism between Bowditch boundaries that coarsely preserves the shadows of horoballs relative to every boundary point induces a coarsely cusp-preserving quasi-isometry between the groups [2503]. In this framework, parabolic points are exactly the boundary points corresponding to horoballs, and horoball shadows serve as the relative analogue of classical visual balls [2503].

For hierarchically hyperbolic groups that are relatively hyperbolic, the Bowditch boundary is recovered as a quotient of the HHS boundary. There is a quotient map
\[
\Psi \colon \partial(G,\mathfrak S)\to \partial \cusp(G,\mathcal P)
\]
such that for distinct \(p,q\in \partial(G,\mathfrak S)\),
\[
\Psi(p)=\Psi(q)\quad\Longleftrightarrow\quad \exists\, g\in G,\ H\in \mathcal P \text{ so that } p,q \text{ are both in the limit set of } gH,
\]
and the inclusion \(G\to \cusp(G,\mathcal P)\) extends continuously to \(\Psi\) [2305]. This makes the Bowditch boundary a coarse quotient of a finer boundary theory.

The same quotient theme appears in the quasi-redirecting boundary. If \(G\) is relatively hyperbolic and the Cayley graphs of the peripheral subgroups are mono-directional, then the quasi-redirecting boundary is homeomorphic to the Bowditch boundary [2406]. In that description, transient classes correspond to conical points and non-transient classes correspond to parabolic cone points [2406].

Explicit boundary constructions also exist for combination theorems. For finite graphs of relatively hyperbolic groups with parabolic edge groups, the Bowditch boundary of the total group can be constructed from the vertex Bowditch boundaries together with the boundary of the Bass–Serre tree, with additional identifications reflecting the edge-parabolic data [2104]. This provides a concrete gluing model rather than an abstract existence theorem.

## 6. Rigidity, measures, and current uses

The Bowditch boundary is now a standard input in rigidity problems. In the theory of relatively Anosov representations, it is the domain of continuous \(\rho\)-equivariant limit maps
\[
(\xi_\rho^k,\xi_\rho^{d-k}) : \partial(\Gamma,\mathcal P)\to \mathrm{Gr}_k(\mathbb R^d)\times \mathrm{Gr}_{d-k}(\mathbb R^d),
\]
and the topology of \(\partial(\Gamma,\mathcal P)\) imposes strong restrictions on which groups can admit such representations [2401]. One general estimate is
\[
\dim(\partial(\Gamma,\mathcal P))\le d-k-1,
\]
except in the exceptional cases
\[
(d,k)\in \{(2,1),(4,2),(8,4),(16,8)\}
\quad\text{and}\quad
\partial(\Gamma,\mathcal P)\cong S^{d-k},
\]
and several classification results for \(\mathrm{SL}_3(\mathbb R)\), \(\mathrm{Sp}_{2m}(\mathbb R)\), and \(\mathrm{SL}_4(\mathbb R)\) are driven by this boundary topology [2401].

The Bowditch boundary also supports a probabilistic theory. For groups hyperbolic relative to virtually nilpotent subgroups, one can construct a random walk on a cusped graph whose Martin boundary is the Bowditch boundary, and the associated harmonic measure is a conformal density for the Green metric [2112]. In that setting the harmonic measure is exact dimensional on the Bowditch boundary, and with the visual distance induced by the Green metric the boundary becomes an Ahlfors-regular metric measure space [2112].

Quantitative boundary invariants are beginning to be studied in explicitly relative settings. For certain relatively hyperbolic Coxeter groups, upper and lower bounds on the conformal dimension of the Bowditch boundary have been established, using embedded round trees for lower bounds and geometrically finite actions on \(CAT(-1)\) spaces for upper bounds [2504]. This is used there to distinguish infinitely many quasi-isometry classes [2504].

Generic-topology phenomena can also be nontrivial. For a non-abelian free group with random cyclic peripheral structure, there is no generic Bowditch-boundary homeomorphism type; instead, the minimal size of cut sets in the boundary increases as the lengths of the random peripheral words go to infinity [2310]. This suggests that the relative boundary can vary widely even inside a fixed ambient group.

Finally, the boundary can change under refinement of the peripheral structure. If \((G,\mathcal P)\) is relatively hyperbolic with
\[
\partial_B(G,\mathcal P)\cong S^n,
\]
then under dimension-dependent hypotheses and for a proper subset \(\mathcal Q\subsetneq \mathcal P\) whose removed peripheral groups are hyperbolic, one obtains
\[
\partial_B(G,\mathcal Q)\cong \text{the }(n-1)\text{-dimensional Sierpiński carpet}
\]
[2208]. This suggests that the Bowditch boundary is sensitive not only to the ambient group but also to the chosen peripheral structure, a theme that runs throughout the theory.

Source: https://www.emergentmind.com/topics/bowditch-boundary