2303 04843

GRAPHICALLY DISCRETE GROUPS AND RIGIDITY
ALEX MARGOLIS, SAM SHEPHERD, EMILY STARK, AND DANIEL WOODHOUSE
Abstract. We introduce the notion of graphical discreteness to group theory. A

finitely generated group is graphically discrete if whenever it acts geometrically on
a locally finite graph, the automorphism group of the graph is compact-by-discrete.
Notable examples include finitely generated nilpotent groups, most lattices in semisim-
ple Lie groups, and irreducible non-geometric 3-manifold groups. We show graphs of
groups with graphically discrete vertex groups frequently have strong rigidity proper-
arXiv:2303.04843v1 [math.GR] 8 Mar 2023
ties. We prove free products of one-ended virtually torsion-free graphically discrete

groups are action rigid within the class of virtually torsion-free groups. We also prove
quasi-isometric rigidity for many hyperbolic graphs of groups whose vertex groups
are closed hyperbolic manifold groups and whose edge groups are non-elementary
quasi-convex subgroups. This includes the case of two hyperbolic 3-manifold groups
amalgamated along a quasi-convex malnormal non-abelian free subgroup. We provide
several additional examples of graphically discrete groups and illustrate this property
is not a commensurability invariant.
1. Introduction
A foundational observation in geometry, dating back to at least the work of Klein,
is that the study of a metric space X is made far richer by studying X together with
the structure of its isometry group Isom(X). From the perspective of geometric group
theory, a finitely generated group Γ is studied not only as a metric space via word
metrics on the group, but via its geometric (i.e. properly discontinuous, cocompact, and
isometric) actions on proper metric spaces. If Γ acts geometrically on X, then modulo a
finite normal subgroup, Γ embeds as a uniform lattice into Isom(X), which is a locally
compact group when equipped with the compact-open topology.
This paper addresses the following broad problems:
Problem 1.1. (Lattice envelope problem.) Given a finitely generated group Γ, classify
the locally compact groups containing Γ as a uniform lattice.
A finitely generated group Γ is action rigid if whenever Γ and another finitely gener-
ated group Γ0 embed, modulo finite normal subgroups, as uniform lattices in the same
locally compact group, then Γ and Γ0 are virtually isomorphic.
Problem 1.2. (Action rigidity problem.) Classify the finitely generated groups that
are action rigid.
The lattice envelope problem has roots in Mostow’s Strong Rigidity [Mos73], which
classifies the Lie groups that contain the fundamental group of a closed hyperbolic man-
ifold as a uniform lattice. This classification was extended to lattice embeddings into lo-
cally compact groups and for a wider class of finitely generated groups by Furman [Fur01]
Date: March 10, 2023.

1
2 ALEX MARGOLIS, SAM SHEPHERD, EMILY STARK, AND DANIEL WOODHOUSE
and Bader–Furman–Sauer [BFS20]. Dymarz [Dym15] classified lattice envelopes of cer-

tain solvable groups, and the classification of lattice envelopes of virtually free groups
was given by Mosher–Sageev–Whyte [MSW02]. Well-studied groups in geometric group
theory have been shown to have only trivial lattice envelopes (see Definition 2.10), in-
cluding non-exceptional mapping class groups [Kid10], Out(Fn ) for n ≥ 3 [GH21], and
various 2-dimensional Artin groups [HH20].
Rigidity problems in geometric group theory frequently have the following form: if
two finitely generated groups share a common weak geometric structure, do they share
a common algebraic structure? Specifically, if two finitely generated groups are either
quasi-isometric or act geometrically on the same proper metric space, are they virtually
isomorphic? Two locally compact groups G and G0 are virtually isomorphic if there exist
open finite-index subgroups H ≤ G and H 0 ≤ G0 and compact normal subgroups K / H
and K 0 / H 0 so that H/K ∼ = H 0 /K 0 . A finitely generated group Γ is quasi-isometrically
rigid if any finitely generated group quasi-isometric to Γ is virtually isomorphic to Γ.
Quasi-isometric rigidity has been a central focus of geometric group theory, leading to
an abundance of new ideas with far-reaching consequences, including the relationship
between the topological notion of ends and algebraic splittings [Sta68, Dun85], the
development of quasi-conformal geometry [Tuk88, Gab92, CJ94, Sch95, BP00], and the
structure of asymptotic cones [Gro81, vdDW84, KL97].
Action rigidity can be equivalently stated in terms of group actions: a finitely gener-
ated group Γ is action rigid if whenever Γ and another finitely generated group Γ0 act
geometrically on the same proper metric space, then Γ and Γ0 are virtually isomorphic.
In the language of Problem 1.2, action rigidity makes precise the following comment of
Gromov.
Scholium 1.3. (Gromov; see [Mj12, Scholium 1.2]) If two discrete groups can be em-
bedded in the same locally compact group nicely, they are as good as commensurable.
In this paper we introduce the notion of graphically discrete groups, which are well-
suited for studying rigidity phenomena. A finitely generated group Γ is graphically
discrete if whenever Γ acts geometrically on a connected locally finite graph X, the au-
tomorphism group Aut(X) is compact-by-discrete, meaning Aut(X) contains a compact
normal subgroup so that the quotient group is discrete. In this case, Aut(X) is virtually
isomorphic to Γ, and Γ does not have any nontrivial totally disconnected uniform lattice
embeddings. Intuitively, a finitely generated group Γ is graphically discrete if it does
not act on any graph that has “too many more” symmetries than those coming from Γ.
See Sections 3.1 and 3.2 for other equivalent formulations of graphical discreteness and
Section 1.4, which motivates restricting to group actions on graphs. Crucially, there are
a large number of examples of groups that are graphically discrete; see Section 1.2.
1.1. Rigidity theorems. We begin by giving rigidity consequences for graphically dis-
crete groups. It follows readily from definitions that if Γ and Γ0 both act geometrically
on the same connected locally finite graph and Γ is graphically discrete, then Γ and Γ0
are virtually isomorphic; see Proposition 3.9. Work of Margolis [Mar22b, Theorem C]
characterized the class of finitely generated groups that embed, modulo a finite normal
subgroup, as a uniform lattice into a locally compact group that is not compact-by-
(totally disconnected). This can be used to deduce action rigidity for many graphically
discrete groups; see Theorem 1.10.
GRAPHICALLY DISCRETE GROUPS AND RIGIDITY 3
The central action rigidity results of this paper concern groups which are not typi-
cally graphically discrete, but decompose as a graph of groups whose vertex groups are
graphically discrete. Given a class C of finitely generated groups, we say a group Γ ∈ C
is action rigid within C if for all Γ0 ∈ C such that both Γ and Γ0 act geometrically on the
same proper metric space, then Γ and Γ0 are virtually isomorphic.
Theorem A. (Theorem 10.4.) Let Γ be a virtually torsion-free finitely presented infinite-

ended group such that each one-ended vertex group in a Stallings–Dunwoody decompo-
sition of Γ is graphically discrete. Then Γ is action rigid within the family of virtually
torsion-free groups.
The virtually torsion-free assumption in the theorem above is necessary: in Exam-

ple 10.14 we exhibit an amalgamated free product of graphically discrete groups over a
finite subgroup that is not action rigid. Theorem 10.4 actually holds within the class
of groups that command their finite-index subgroups, a more general class of groups
including both virtually torsion-free and residually finite groups.
By adding an assumption to the one-ended vertex groups of Γ (that these persistently
command their finite subgroups; see Definition 10.5), we prove that Γ is action rigid
in the absolute sense; see Theorem 10.10. As a consequence, we give the following
classification of action rigidity for 3-manifold groups.
Corollary B. (Theorem 10.12.) The fundamental group of a (possibly reducible) closed

3-manifold is action rigid if and only if the manifold does not admit H3 or Sol geometry.
Work of Whyte and Papasoglu–Whyte demonstrates that infinite-ended groups pos-

sess a remarkable degree of quasi-isometric flexibility [Why99, PW02]. For instance,
within the class of infinite-ended groups, Whyte produced the first examples to demon-
strate the sign of the Euler characteristic and the ratios of `2 -Betti numbers, which
are commensurability invariants, are not quasi-isometry invariants. In contrast, Theo-
rem A shows that while quasi-isometric rigidity does not typically hold for infinite-ended
groups, the weaker notion of action rigidity frequently does.
Theorem A generalizes previous work of Stark–Woodhouse [SW], who proved the
result when the one-ended vertex groups are residually finite and fundamental groups
of closed hyperbolic manifolds. A starting point to this project was to understand
the properties of closed hyperbolic manifolds that lead to these rigidity phenomena.
Graphical discreteness turned out to be an apt condition, interesting in its own right,
and satisfied in far greater generality.
Like Theorem A, the next action rigidity theorem concerns graphs of groups with
graphically discrete vertex groups; a key difference is that we now allow infinite edge
groups. See Section 11.2 for the relevant definitions.
Theorem C. (Theorem 11.8) Suppose Γ is the fundamental group of a minimal finite

graph of groups G satisfying the following properties:
(1) Every vertex group of G is graphically discrete, hyperbolic and cubulated.
(2) For each vertex group of G, the collection of images of incident edge maps is an
almost malnormal family of infinite quasi-convex subgroups.
(3) The associated Bass–Serre tree of spaces (X, T ) is preserved by quasi-isometries.
Then Γ is action rigid.
By specializing to the case of hyperbolic n-manifold vertex groups, and adding an

assumption about cohomological dimension, we use work of Mosher–Sageev–Whyte
[MSW11] and Biswas [Bis12] to upgrade the conclusion of Theorem C to quasi-isometric
rigidity.
Theorem D. (Theorem 11.16) Let Γ be the fundamental group of a minimal finite graph
of groups G satisfying the following properties:
(1) Each vertex group of G is cubulated and is the fundamental group of a closed
hyperbolic n-manifold for some n ≥ 3.
almost malnormal family of non-elementary quasi-convex subgroups with coho-
mological codimension at least two.
Then any finitely generated group quasi-isometric to Γ is abstractly commensurable to Γ.
We note that the vertex groups are automatically cubulated in the case n = 3 [KM12,
BW12], and that the cohomological codimension condition is satisfied if the edge groups
are all quasi-convex free groups (see also Corollary 11.18). Thus Theorem D yields the
following corollary:
Corollary E. Let Γ1 and Γ2 be fundamental groups of closed hyperbolic 3-manifolds,

and let F1 ≤ Γ1 and F2 ≤ Γ2 be isomorphic quasi-convex almost malnormal non-abelian
free subgroups. Then Γ1 ∗F1 =F2 Γ2 is quasi-isometrically rigid.
To prove Theorem D, we use the following strategy detailed in Section 11.1 for upgrad-
ing action rigidity to quasi-isometric rigidity. We show that if Γ is a finitely generated
tame group that is action rigid and has the uniform quasi-isometry property, then Γ is
quasi-isometrically rigid. If we weakened the hypotheses of Theorem D to allow two-
ended edge groups, then Γ would no longer possess the uniform quasi-isometry property
and the above strategy would break down.
1.2. Graphical discreteness. A vast family of well-studied groups can be shown to

be graphically discrete using the existing literature: virtually nilpotent groups [Tro84],
irreducible lattices in connected center-free real semisimple Lie groups without compact
factors [BFS20], uniform lattices in the isometry group of the 3-dimensional geome-
try Sol [Dym15], mapping class groups of non-exceptional finite-type surfaces [Kid10],
Out(Fn ) for n ≥ 3 [GH21], various 2-dimensional Artin groups of hyperbolic type [HH20],
hyperbolic surface-by-free groups [FM02], the Higman groups [HH22], and topologically
rigid hyperbolic groups [KK00]. See Theorem 3.16 for details. The results above, to-
gether with work of Kleiner–Leeb [KL01] imply that the fundamental group of a closed
geometric 3-manifold is graphically discrete; see Theorem 7.1.
We show fundamental groups of closed irreducible non-geometric 3-manifolds have
only trivial lattice envelopes, and are therefore graphically discrete.
Theorem F. (Theorem 7.9.) Let M be a closed irreducible, oriented 3-manifold with a

nontrivial geometric decomposition. Let G be a locally compact group containing π1 (M )
as a lattice. Then G contains a compact open normal subgroup K such that G/K is
isomorphic to the fundamental group of a compact 3-orbifold covered by M . In particular,
π1 (M ) has no nontrivial lattice embeddings, hence is graphically discrete.
Cannon’s Conjecture states that a hyperbolic group with boundary a 2-sphere is vir-
tually the fundamental group of a closed hyperbolic 3-manifold, hence must be graphi-
cally discrete as noted above. The next result says that any potential counterexamples
to Cannon’s Conjecture must also have this property.
Theorem 1.4. (Theorems 4.5, 4.8.) A hyperbolic group with visual boundary homeo-
morphic to an n-sphere with n ≤ 3 or to a Sierpinski carpet is graphically discrete.
The proof of the theorem above is an application of the resolution of the Hilbert–Smith
Conjecture concerning which locally compact groups act continuously and faithfully on
manifolds by homeomorphisms. The Hilbert–Smith Conjecture is open for n > 3 and
additional positive solutions would imply the result above holds for the same values of n.
Given a property of groups such as graphical discreteness, it is natural to ask how
robust the property is: is it a quasi-isometry or virtual isomorphism invariant? We show
graphical discreteness is neither via the following examples. A manifold is a minimal
element in its commensurability class if it does not nontrivially cover any orbifold. We
note that there exist closed hyperbolic 3-manifolds that are minimal elements in their
commensurability class; see Remark 5.2.
Theorem 1.5. (Theorem 5.4.) Let M and M 0 be closed hyperbolic 3-manifolds that are
minimal elements in their commensurability classes. Then Γ = π1 (M ) ∗ π1 (M 0 ) has no
nontrivial lattice embeddings, but has a finite-index subgroup that does have nontrivial
lattice embeddings. In particular, Γ is graphically discrete but contains a finite-index
subgroup that is not.
Corollary 1.6. The properties of being graphical discrete and possessing no nontrivial
lattice embeddings do not pass to finite-index subgroups. In particular, these properties
are not invariant under virtual isomorphisms nor under quasi-isometries.
We also provide additional examples of groups that are not graphically discrete. A
good source of such examples are graphs of groups that exhibit some sort of “flipping
symmetry”. We exploit this idea in the following setting. A simple surface amalgam is
a topological space constructed by taking a set of n ≥ 3 compact surfaces Σ1 , . . . , Σn
with negative Euler characteristic and a single boundary component, and identifying
all the boundary components by a homeomorphism. For fundamental groups of simple
surface amalgams, graphical discreteness is equivalent to having no nontrivial lattice
embeddings (Theorem 6.5). The following result characterizes when the fundamental
group of a simple surface amalgam has nontrivial lattice embeddings, and provides
examples of one-ended groups for which graphical discreteness is not a quasi-isometry
invariant.
Theorem 1.7. (Theorem 6.5) The fundamental group of a simple surface amalgam
has no nontrivial lattice embeddings if and only if the surfaces in the amalgam have
pairwise-distinct Euler characteristics.
In Proposition 3.17 we give a general criterion to show an HNN extension is not

graphically discrete. In particular, we use this to deduce a right-angled Artin group AΓ
is graphically discrete if and only if Γ is a complete graph.
If the automorphism group of a connected locally finite graph contains uniform lattices
that are not virtually isomorphic, then these uniform lattices are neither graphically
discrete nor action rigid. Results in [BM00, For22] then imply that any product of free
groups Fm × Fn and certain Baumslag–Solitar groups (including BS(k, kn) for coprime
k and n) are neither graphically discrete nor action rigid.
We also note that graphically discrete groups were considered by de la Salle–Tessera,
as groups in family C [dlST19, Section 6]. They proved that Cayley graphs of torsion-free
graphically discrete groups are uniquely strongly local-to-global rigid ; see the reference
for definitions.
1.3. Methods of proof. In this section we highlight some of the tools we develop to
prove the theorems in Sections 1.1 and 1.2. Many of these tools may be of independent
interest.
The rough strategy that we employ in the proofs of Theorems A and C is as follows.
Suppose that Γ and another finitely generated group Γ0 act geometrically on the same
proper metric space X. We divide the argument into three steps:
(1) Show that the isometry group G = Isom(X) acts continuously on a tree T that
corresponds (at least in some way) to the splitting of Γ.
(2) Upgrade the metric space X to a simply connected combinatorial cell complex Y
that decomposes as a tree of spaces over T .
(3) Build a common finite cover for the quotient spaces Y /Γ and Y /Γ0 .
Given the assumptions in Theorems A and C, Step (1) follows relatively easily from
results in the literature, so most of the work lies in Steps (2) and (3).
For Step (2) we use the following proposition (or one of its more technical versions
from Section 8). This result can be viewed as a variant of the Cayley–Abels graph for
a tdlc group G (see [KM08]) that is compatible with the action of G on a tree T .
Proposition 1.8. Let G be a locally compact group acting continuously, minimally and
cocompactly on a tree T . Suppose all vertex and edge stabilizers of T are compactly
generated and G contains a compact open subgroup. Then there exists a locally finite
connected graph X admitting a geometric action of G, and a surjective G-equivariant
map p : X → T such that the fibers of vertices in T are connected subgraphs of X.
For Step (3) we use Theorem 9.7. This theorem takes two compact graphs of spaces
X1 and X2 with a common universal cover X̃, satisfying certain additional assumptions,
and guarantees the existence of a common finite cover. The assumptions involved are
too technical to state here in full, but the rough idea for the most important assumptions
is as follows. First we assume that the vertex spaces in X̃ have discrete automorphism
groups. This allows us to construct common finite covers for pairs of vertex spaces in
X1 and X2 . In the context of Theorems A and C, this assumption can be deduced from
the graphical discreteness of vertex groups.
Then, to build a common finite cover of X1 and X2 , we want to glue together these
finite covers of vertex spaces along elevations of edge spaces. However, we need an
additional assumption in order to make these elevations of edge spaces match up. The
required assumption involves the notion of a group commanding a finite collection of
subgroups (see Definition 2.43). More precisely, we assume that each automorphism
group of a vertex space in X̃ commands a collection of stabilizers of orbit representatives
of incident edge spaces. For Theorem A, the commanding assumption derives from the
hypothesis that vertex groups are virtually torsion-free; while for Theorem C it derives
from the assumption that each vertex group of G is hyperbolic and cubulated with
quasi-convex incident edge groups.
To determine which groups are graphically discrete, we use a range of different argu-
ments and employ many results from the literature. In particular, we make use of the
following characterization of graphical discreteness.
Theorem 1.9. (Theorem 3.10) A finitely generated group Γ is graphically discrete if

and only if for every geometric action of Γ on a connected locally finite graph X and for
all vertices x ∈ X, the homomorphism (Aut(X))x → QI(X) from the stabilizer of x to
the quasi-isometry group of X has finite image.
An analogous statement holds for non-elementary hyperbolic groups replacing QI(X)

with Homeo(∂X); see Theorem 4.1. In a similar vein, one can characterize graphical
discreteness for one-ended hyperbolic groups with nontrivial JSJ decomposition using
the action of (Aut(X))x on the JSJ tree (Corollary 4.4). Another source of graphically
discrete groups are those with discrete quasi-isometry groups; more precisely, any finitely
generated tame group Γ for which the natural map Γ → QI(Γ) has finite-index image is
graphically discrete (see Theorem 3.16(6) and Theorem 3.26).
1.4. Groups acting on graphs. The study of locally compact groups is often split into
the Lie group setting and the totally disconnected locally compact group (tdlc) setting.
Indeed, a locally compact group G sits in a short exact sequence 1 → G0 → G →
G/G0 → 1, where the identity component G0 is an inverse limit of connected Lie groups
and G/G0 is a tdlc group; see [Tao14]. This paper primarily concerns the tdlc setting;
the automorphism group of a locally finite graph is a tdlc group and a partial converse
holds as well [KM08]. Restricting to groups acting on graphs is particularly natural in
light of recent work of Margolis [Mar22b], who characterized which finitely generated
groups can embed, modulo finite kernel, as uniform lattices into locally compact groups
that are not compact-by-(totally disconnected). In particular:
Theorem 1.10. [Mar22b] Let Γ be a finitely generated group with no infinite commen-
surated subgroup that is either a finite-rank free abelian group or a uniform lattice in
a connected semisimple Lie group with finite center. Then, if Γ and another finitely
generated group Γ0 act geometrically on the same proper metric space, then Γ and Γ0 act
geometrically on the same connected locally finite graph. In particular, if Γ as above is
graphically discrete, then Γ is action rigid.
Understanding the finitely generated groups with only trivial lattice envelopes is a
question of considerable interest in the field. We note that a finitely generated group
is graphically discrete exactly when the group has only trivial tdlc lattice envelopes
(Proposition 3.6). The theorem above implies that the only potential sources of examples
where these are different notions are finitely generated groups with infinite commensu-
rated subgroups that are either finite-rank free abelian or uniform lattices in connected
semisimple Lie groups.
1.5. Open questions. We make the following conjectures:
Conjecture 1.11. If Γ and Γ0 are graphically discrete, then Γ×Γ0 is graphically discrete.
In Theorem 4.9, we prove a special case of Conjecture 1.11 when Γ and Γ0 are non-
elementary hyperbolic groups. Conjecture 1.11 is open even in the special cases that Γ0
is Z, abelian, or nilpotent.
Uniform lattices in isometry groups of thick hyperbolic buildings are not graphically
discrete; see Corollary 4.2. These hyperbolic groups in dimension two have Menger
curve visual boundary and the failure of graphical discreteness is realized in the au-
tomorphism group of the building. On the other hand, we conjecture that hyperbolic
groups with Menger curve boundary whose standard 2-complexes have far less symmetry
are graphically discrete.
Conjecture 1.12. Random groups in the few-relator model and Gromov’s density
model are graphically discrete asymptotically almost surely as the length of the cho-
sen relators tends to infinity.
Conjecture 1.13. A one-ended, hyperbolic free-by-cyclic group Fn oφ Z with trivial
JSJ decomposition is graphically discrete.
We show hyperbolic groups with boundary homeomorphic to an n-sphere for n ≤ 3 are

graphically discrete. While the bound on dimension is used in the proof, we conjecture:
Conjecture 1.14. Hyperbolic groups with boundary the n-sphere for n ≥ 4 are graph-
ically discrete.
Acknowledgments. The authors are thankful for helpful discussions with Tullia Dy-
marz and with Genevieve Walsh, who explained Remark 5.2. The third author was
supported by NSF Grant No. DMS-2204339.
Contents
1. Introduction 1
1.1. Rigidity theorems 2
1.2. Graphical discreteness 4
1.3. Methods of proof 6
1.4. Groups acting on graphs 7
1.5. Open questions 7
Acknowledgments 8
2. Preliminaries 9
2.1. Graphs and cell complexes 9
2.2. Coarse geometry 9
2.3. Locally compact groups and lattices 10
2.4. Actions of locally compact groups on metric spaces 13
2.5. Action rigidity 15
2.6. Graphs of groups and spaces 16
2.7. Commanding subgroups 18
3. Introducing graphically discrete groups 19
3.1. Equivalent definitions of graphical discreteness 19
3.2. Characterization of compact-by-discrete automorphism groups 21
3.3. Examples of graphically discrete groups 23
3.4. Examples of non-graphically-discrete groups 24

3.5. Uniform lattices and quasi-actions 26
4. Graphical discreteness and non-discreteness for hyperbolic groups 28
4.1. Boundary characterization of graphical discreteness 28
4.2. Hyperbolic groups with sphere and Sierpinski carpet boundaries 30
4.3. Graphical discreteness of direct products of hyperbolic groups 31
5. Graphical discreteness is not a commensurability invariant 32
6. Simple surface amalgams 33
7. Graphical discreteness of closed 3-manifold groups 36
7.1. Geometric 3-manifolds 36
7.2. Non-geometric 3-manifolds 37
8. Blowups of groups acting on trees 40
9. Common Covers 44
10. Action rigidity 51
10.1. Ends and accessibility for locally compact groups 51
10.2. Action rigidity for infinite-ended groups 51
11. Quasi-isometric rigidity 54
11.1. From action rigidity to quasi-isometric rigidity 54
11.2. Action rigidity for graphs of groups 55
11.3. Quasi-isometric rigidity for graphs of hyperbolic manifold groups 58
References 63
2. Preliminaries
2.1. Graphs and cell complexes.
Definition 2.1. A graph X is a set of vertices V X and edges EX. An edge e is oriented
with an initial vertex ι(e) and a terminal vertex τ (e). Associated to e is the edge ē with
reversed orientation: ι(e) = τ (ē) and τ (e) = ι(ē). We have e 6= ē and e = ē¯ for all
e ∈ EX. We also consider graphs as geodesic spaces by making each edge isometric to
the unit interval. For v a vertex in a graph, the link of v is defined to be the following
set:
lk(v) = {e | τ (e) = v}
A tree is a connected graph with no cycles. In a tree, or more generally in a simplicial
graph, each edge is determined by its endpoints, so we will often write e = (u, v) to
mean u = ι(e) and v = τ (e).
Definition 2.2. A 2-dimensional cell complex is a space obtained by attaching 2-cells

to a graph, where the boundary of each cell is glued to a loop in the graph. We only deal
with 2-dimensional cell complexes in this paper and refer to them as cell complexes. A
cell complex (or graph) is locally finite if each vertex is incident to finitely many edges
and 2-cells.
2.2. Coarse geometry.

Definition 2.3. Let X and Y be metric spaces. Let K ≥ 1 and A ≥ 0. A function

f : X → Y is:
(1) (K, A)-coarse Lipschitz if for all x, x0 ∈ X, dY f (x), f (x0 ) ≤ KdX (x, x0 ) + A.

(2) (K, A)-quasi-isometry is a map f : X → Y such that:

(a) for all x, x0 ∈ X,
1
dX (x, x0 ) − A ≤ dY f (x), f (x0 ) ≤ KdX (x, x0 ) + A,

K
(b) for all y ∈ Y , there is some x ∈ X such that dY (f (x), y) ≤ A.
A map is coarse Lipschitz (resp. a quasi-isometry) if it is (K, A)-coarse Lipschitz (resp.
a (K, A)-quasi-isometry) for some K ≥ 1 and A ≥ 0. A map f : X → Y is coarsely
surjective if for some A, NA (f (X)) = Y .
Two quasi-isometries f, g : X → Y are A-close if supx∈X dY (f (x), g(x)) ≤ A, and are
close if they are A-close for some A ≥ 0.
If Γ is a finitely generated group, Γ can be equipped with the word metric with respect
to a finite generating set. This metric is well-defined up to quasi-isometry.
Definition 2.4. The quasi-isometry group QI(X) of a metric space X consists of self-
quasi-isometries of X modulo the equivalence relation f ∼ g if f and g are close. An
equivalence class in QI(X) containing a quasi-isometry f is denoted [f ]. The group
operation on QI(X) is given by composition [f ][g] = [f ◦ g], which can readily be shown
to be well-defined.
Notation 2.5. If G is a group, then we use 1G (or 1 in the absence of ambiguity) to
denote the identity element of G.
Definition 2.6. If G is a group and X is a metric space, then a (K, A)-quasi-action of
G on X is a collection of maps {fg }g∈G such that
(1) for every g ∈ G, fg is a (K, A)-quasi-isometry from X to X;
(2) for every g, k ∈ G, fgk is A-close to fg ◦ fk ;
(3) f1 is A-close to the identity on X.
A quasi-action of G on X is a (K, A)-quasi-action of G on X for some K ≥ 1 and A ≥ 0.
A quasi-action {fg }g∈G of G on X is said to be cobounded if there exists a B such that
for all x, x0 ∈ X, there is some g ∈ G such that d(fg (x), x0 ) ≤ B. Two quasi-actions
{fg }g∈G and {hg }g∈G of G on X and Y respectively, are quasi-conjugate if there is a
quasi-isometry f : X → Y and a constant C such that

d f (fg (x)), hg (f (x)) ≤ C
for all g ∈ G and x ∈ X.
Remark 2.7. A quasi-action {fg }g∈G of G on X induces a homomorphism G → QI(X)
given by g 7→ [fg ]. In particular, the natural isometric action of a finitely generated
group Γ on itself by left multiplication induces a map Γ → QI(Γ).
2.3. Locally compact groups and lattices. By convention, topological groups are
assumed to be Hausdorff. If G is topological group and H ≤ G is a closed subgroup,
G/H will always be assumed to be equipped with the quotient topology. The group
G/H is discrete if and only if H is open. It follows that G is a compact-by-discrete if
and only if G contains a compact open normal subgroup.
If X is a metric space, then Isom(X) denotes the isometry group of X, equipped with
the compact-open topology. If X is proper, i.e. closed balls are compact, then Isom(X)
is a second countable locally compact topological group; moreover, the compact-open
topology on Isom(X) coincides with the topology of pointwise convergence. If X is a
connected graph, then we identify its automorphism group Aut(X) with a subgroup of
the isometry group of the graph, metrized so each edge has length one.
Definition 2.8. A lattice of a locally compact group G is a discrete subgroup Γ ≤ G so
that Γ acts on G with a Borel fundamental domain of finite Haar measure. The lattice
is uniform if in addition Γ\G is compact. A (uniform) lattice embedding into a locally
compact group G is a monomorphism Γ → G whose image is a (uniform) lattice in G. A
virtual (uniform) lattice embedding into a locally compact group G is a homomorphism
Γ → G with finite kernel and with image a (uniform) lattice in G. If there exists a
virtual (uniform) lattice embedding Γ → G, we say Γ is a virtual (uniform) lattice in G.
We make frequent use of the following lemma:
Lemma 2.9 ([CM12, §2.C]). Let Γ be a (uniform) lattice in a locally compact group G.
If H ≤ G is open, then Γ ∩ H is a (uniform) lattice in H.
The following terminology is due to Bader–Furman–Sauer [BFS20]. We generalize
the definition given in [BFS20] from lattice embeddings to virtual lattice embeddings.
Definition 2.10 ([BFS20, Definition 3.1]). Let Γ and Γ0 be countable groups. Two
virtual lattice embeddings ρ : Γ → G and ρ0 : Γ0 → G0 are virtually isomorphic if there
exist:
(1) open finite-index subgroups H ≤ G and H 0 ≤ G0 ;
(2) compact normal subgroups K C H and K 0 C H 0 ;
(3) a commutative square
(Γ ∩ ρ−1 (H))/(Γ ∩ ρ−1 (K)) H/K
∼
= ∼
=
(Γ0 ∩ ρ0−1 (H 0 ))/(Γ0 ∩ ρ0−1 (K 0 )) H 0 /K 0

where the horizontal arrows are lattice embeddings induced by ρ and ρ0 and the
second vertical map is a topological isomorphism.
A virtual lattice embedding is trivial if it is virtually isomorphic to the identity map
IdΓ : Γ → Γ.
Remark 2.11. It follows from the preceding definition that a virtual lattice embedding
ρ : Γ → G is trivial if and only if either of the following equivalent conditions are
satisfied:
(1) G is compact-by-discrete, i.e. G contains a compact normal subgroup K such
that the quotient G/K is discrete.
(2) G contains a compact open normal subgroup.
We refer the reader to [BFS20, Lemmas 3.3–3.5] to verify that the preceding definition
makes sense and that the relation of virtual isomorphism is an equivalence relation.
The following definitions are analogues of finite generation and finite presentation in
the setting of locally compact groups.
Definition 2.12 ([CdlH16, §7]). A locally compact group G is compactly generated if

it admits a compact generating set. A locally compact group G is compactly presented
if it admits a presentation hS | Ri such that S is compact and the relations R are of
bounded length as words in S.
Lemma 2.13 ([CdlH16, Proposition 5.C.3 and Corollary 8.A.5]). Suppose Γ is a virtual
uniform lattice in a locally compact group G. Then:
(1) G is compactly generated if and only if Γ is finitely generated.
(2) G is compactly presented if and only if Γ is finitely presented.
A compact normal subgroup is maximal if it is not a proper subgroup of any other

compact subgroup. Since the product of compact normal subgroups is also a compact
normal subgroup, a maximal compact subgroup, if it exists, is unique and is the union
of all compact normal subgroups. The following criterion gives a sufficient criterion for a
locally compact group to possess a maximal compact normal subgroup. We remark that
the hypothesis of Proposition 2.14 is satisfied when the group Γ is virtually torsion-free.
Proposition 2.14 ([Cor15, Proposition 2.7]). Let Γ be a finitely generated group con-
taining a maximal finite normal subgroup. If Γ is a virtual uniform lattice in a locally
compact group G, then G contains a maximal compact normal subgroup.
The following remark allows us to restrict our attention to locally compact groups
that are second countable in many parts of this paper.
Remark 2.15. Suppose Γ is finitely generated and ρ : Γ → G is a (uniform) virtual
lattice embedding. Since G is compactly generated by Lemma 2.13, G contains a com-
pact normal subgroup K C G such that G/K is second countable [CdlH16, Remark
ρ
2.B.7.2]. Then Γ −
→ G → G/K is also (uniform) virtual lattice embedding. Moreover,
G is compact-by-discrete if and only if G/K is; see [HR63, Theorem 5.25].
Although we are primarily focused with uniform lattices, results of Bader–Furman–

Sauer show a large class of groups do not possess any non-uniform lattice embeddings:
Proposition 2.16 ([BFS20, Corollary 1.6]). Let Γ be an acylindrically hyperbolic group.
If Γ is not virtually isomorphic to a non-uniform lattice in a connected rank one simple
real Lie group, then all lattice embeddings of Γ are uniform.
Proof. Let Γ be an acylindrically hyperbolic group that is isomorphic to a non-uniform
lattice in locally compact group. As noted in [BFS20, §2], acylindrically hyperbolic
groups satisfy the hypotheses of [BFS20, Corollary 1.6]. Therefore, Γ is a virtually
isomorphic to either
• a non-uniform lattice in a connected, center-free, semisimple real Lie group with-
out compact factors;
• a non-uniform S-arithmetic lattice in a semisimple group with both real and
non-Archimedean factors present [BFS20, §4.1].
In particular, S-arithmetic lattices in the sense of [BFS20] always have S-rank at least
two. Margulis’ normal subgroup theorem asserts that for higher-rank lattices in semisim-
ple algebraic groups, all normal subgroups are finite or of finite-index [Mar91, Chapter
VIII]. However, since acylindrically hyperbolic groups are SQ-universal [Osi16], Γ can-
not satisfy the conclusions of the normal subgroup theorem, since there are uncountably
many countable groups. Thus Γ is a non-uniform lattice in a rank one simple Lie group
as required.
We make frequent use of the following useful lemma concerning continuous maps from
totally disconnected groups to Lie groups.
Lemma 2.17. If ρ : G → L is a continuous homomorphism from a totally disconnected
locally compact group to a Lie group, then ker(ρ) is open.
This follows from two well-known properties of Lie and totally disconnected groups:
Proposition 2.18 (No small subgroups). If G is a Lie group, then there exists a neigh-
borhood U of the identity such that every subgroup of G contained in U is trivial.
Theorem 2.19 ([VD36]). If G is a totally disconnected locally compact group, then
every neighborhood of the identity contains a compact open subgroup.
Proof of Lemma 2.17. Proposition 2.18 ensures there is an identity neighborhood U ⊆
L such that U contains no nontrivial subgroups. By Theorem 2.19, ρ−1 (U ) contains
a compact open subgroup V . Our choice of U ensures ρ(V ) ⊆ U is trivial, hence
V ≤ ker(ρ). Since V is open and ker(ρ) is a union of cosets of V , ker(ρ) is open.
We will also make use of the following related notions.
Definition 2.20. Let G be a group and let H, K ≤ G be two subgroups.
(1) H and K are commensurable if H ∩ K has finite index in both H and K.
(2) H and K are weakly commensurable if there exists some g ∈ G such that H and
gKg −1 are commensurable.
(3) H is commensurated if for all g ∈ G, H and gHg −1 are commensurable.
Two important and closely related sources of commensurable and commensurated
subgroups are the following:
Example 2.21. Two compact open subgroups of a topological group are commensu-
rable. In particular, every compact open subgroup is commensurated.
Example 2.22. If a group G acts on a locally finite graph X by graph automorphisms,
then for all v, w ∈ V X, the vertex stabilizers Gv and Gw are commensurable. In
particular, every Gv is commensurated in G.
2.4. Actions of locally compact groups on metric spaces.
Definition 2.23. Let X be a proper metric space and G a locally compact group. An
isometric action ρ : G → Isom(X) is:
(1) continuous if ρ is continuous, or equivalently, the map G × X → X given by
(g, x) 7→ ρ(g)(x) is continuous [CdlH16, Proposition 5.B.6];
(2) proper if for every compact K ⊆ X, the set {g ∈ G | gK ∩ K 6= ∅} has compact
closure in G;
(3) cocompact if there exists a compact K ⊆ X such that X = ρ(G)K;
(4) geometric if it is proper, cocompact, and continuous;
(5) discrete if ρ(G) is a discrete subgroup of Isom(X).
Remark 2.24. Throughout this paper, if Γ is a finitely generated group, we will assume
that Γ is equipped with the discrete topology unless stated otherwise.
Note that if X is a proper metric space, then Isom(X) acts properly and continuously
on X. We make extensive use of the following well-known lemma; see Proposition 2.34.
Lemma 2.25. If ρ : Γ → Isom(X) is a geometric action of a discrete group on a proper
metric space, then ρ is a virtual uniform lattice embedding.
The following equivalences are essential in the theory of totally disconnected groups
from a geometric viewpoint; see [KM08].
Lemma 2.26 (See [KM08] and [Mar22b, Proposition 2.13]). Let G be a compactly
generated locally compact group. The following are equivalent:
(1) G is compact-by-(totally disconnected);
(2) G contains a compact open subgroup;
(3) there exists a connected locally finite vertex-transitive graph X and a geometric
action G y X.
We now briefly discuss some elementary facts about compact subsets of locally com-
pact groups that act on metric spaces.
Proposition 2.27. Let G be a locally compact group acting continuously and properly
on a proper metric space X. A subset L ⊆ G has compact closure if and only if L has
bounded orbits in X.
Proof. Since the action of G on X is continuous, if L ⊆ X has compact closure, then
the orbit L · x is also compact, hence bounded. Conversely, suppose an orbit L · x is
bounded, hence K = L · x is compact. Then L ⊆ {g ∈ G | gK ∩ K 6= ∅} has compact
closure as the action is proper.
In order to discuss additional some properties concerning compact normal subgroups,
we require the following definitions:
Definition 2.28. A isometry φ of a metric space X is bounded if supx∈X d(x, φ(x)) < ∞.
The following can be thought of as a course analogue of a space having no nontrivial
bounded isometries:
Definition 2.29. A metric space X is tame if for every K ≥ 1 and A ≥ 0, there is a
constant C = C(K, A) such that if f, g : X → X are (K, A)-quasi-isometries such that
supx∈X d(f (x), g(x)) < ∞, then supx∈X d(f (x), g(x)) ≤ C.
A finitely generated group is tame if it is tame when equipped with the word metric.
Example 2.30. A proper geodesic metric space whose Morse boundary contains at
least three points is tame [Mar22a, Proposition 4.21]. In particular, an acylindrically
hyperbolic group is tame.
We use these definitions to describe maximal compact normal subgroups of certain
locally compact groups.
Proposition 2.31. Let X be a proper metric space, G a locally compact group and
ρ : G → Isom(X) a geometric action. If X is tame, then
B = {g ∈ G | ρ(g) is a bounded isometry}
is the maximal compact normal subgroup of G. In particular, if X has no nontrivial
bounded isometries, then ker(ρ) is the maximal compact normal subgroup of G.
Proof. It follows from the definition that B is normal. The tameness condition en-
sures there is a constant C such that for any two bounded isometries φ, ψ of X,
supx∈X d(φ(x), ψ(x)) ≤ C. Thus B has bounded orbits in X, hence has compact closure
by Proposition 2.27. The tameness condition ensures the limit of a convergent sequence
of bounded isometries is also bounded, hence B is closed. Thus B is a compact normal
subgroup.
We now show that any compact normal subgroup K C G is contained in B, which
shows B is maximal. The Milnor–Schwarz lemma for locally compact groups [CdlH16,
Theorem 4.C.5] ensures that the action of G on the Cayley graph of G/K is quasi-
conjugate to the action of G on X. Thus every element of K must be a bounded
isometry of X, hence K ≤ B.
We will make frequent use of the following lemma to deduce continuity of a group
action.
Lemma 2.32. Let X and Y be proper metric spaces, and assume Y is tame with
cocompact isometry group. Suppose H ≤ Isom(X) is a closed subgroup and ρ : H →
Isom(Y ) is an isometric action. Suppose also f : X → Y is a coarsely surjective, coarse
Lipschitz map such that supx∈X,h∈H d(ρ(h)f (x), f (hx)) < ∞. Then the composition
ρ q
H−
→ Isom(Y ) →
− Isom(Y )/K
is continuous, where K C Isom(Y ) is the maximal compact normal subgroup of Y and
q is the quotient map.
Proof. By Proposition 2.31, the subgroup K of Isom(Y ) consisting of bounded isometries

of Y is the maximal compact normal subgroup of Isom(Y ). Suppose (φi ) is a sequence
that converges to the identity in H. The hypotheses on f ensure there is a constant A
such that for all y ∈ Y , d(ρ(φi )(y), y) ≤ A for all i sufficiently large. It follows from the
Arzelá–Ascoli theorem that every subsequence of (ρ(φi )) has a convergent subsequence.
We wish to show (ρ(φi )K) converges to K in Isom(Y )/K. If this is not the case, then
there exists a convergent subsequence of (ρ(φi )) that converges to some λ ∈ Isom(Y )\K.
However, we know d(ρ(φi )(y), y) ≤ A for all y ∈ Y and all i sufficiently large, which
implies d(λ(y), y) ≤ A for all y ∈ Y . This contradicts our assumption λ ∈ / K.
In the case where Y has no nontrivial bounded isometries, Proposition 2.31 ensures
that the maximal compact normal subgroup of Y is trivial. We therefore deduce the
following important special case of Lemma 2.32.
Corollary 2.33. Let X, Y , H, f and ρ be as in Lemma 2.32. In addition, assume that

Y has no nontrivial bounded isometries. Then ρ is continuous.
2.5. Action rigidity. In this section we give equivalent characterizations of action rigid-
ity in terms of group actions on metric spaces and in terms of virtual lattice embeddings
into locally compact groups.
A metric space X is quasi-geodesic if it is quasi-isometric to a geodesic metric space.
This property is called large-scale geodesic in [CdlH16]; see Definition 3.B.1 and Lemma
3.B.6 of [CdlH16] for more details and equivalent formulations. For instance, a nontrivial
finitely generated group equipped with a word metric is not a geodesic metric space but
is a quasi-geodesic space, since it is quasi-isometric to the corresponding Cayley graph.
Proposition 2.34 ([MSW02, Corollary 7]). Let Γ1 and Γ2 be finitely generated groups.
The following are equivalent:
(1) Γ1 and Γ2 are virtual uniform lattices in the same locally compact group;
(2) Γ1 and Γ2 act geometrically on the same proper quasi-geodesic metric space;
(3) Γ1 and Γ2 act geometrically on the same proper metric space.
Proof. The equivalence of (1) and (2) was shown in [MSW02, Corollary 7], (2) =⇒ (3)
is obvious, and (3) =⇒ (1) follows from Lemma 2.25.
Definition 2.35. Two locally compact groups G1 and G2 are:
(1) abstractly commensurable if they have isomorphic finite-index subgroups;
(2) virtually isomorphic if there exist finite-index open subgroups H1 ≤ Γ1 and
H2 ≤ Γ2 and compact normal subgroups K1 / H1 and K2 / H2 so that H1 /K1 ∼ =
H2 /K2 .
Two groups that are abstractly commensurable are also virtually isomorphic. Al-
though the converse does not hold in general, there is the following partial converse:
Lemma 2.36. Suppose Γ1 and Γ2 are virtually torsion-free. Then Γ1 and Γ2 are ab-
stractly commensurable if and only if they are virtually isomorphic.
The equivalences in the next definition follow from Proposition 2.34.

Definition 2.37. A finitely generated group Γ is said to be action rigid if either of the
following equivalent conditions hold:
(1) whenever Γ0 is a finitely generated group such that Γ and Γ0 both act geometri-
cally on the same proper metric space, then Γ and Γ0 are virtually isomorphic;
(2) whenever Γ0 is a finitely generated group and there exist virtual uniform lattice
embeddings of Γ and Γ0 into the same locally compact group, then Γ and Γ0 are
virtually isomorphic.
The following lemma follows immediately from definitions.

Lemma 2.38. If all virtual uniform lattice embeddings of a finitely generated group Γ
are trivial, then Γ is action rigid.
2.6. Graphs of groups and spaces. In this subsection we define graphs of groups,
graphs of spaces, and related concepts. We refer to [Ser77, SW79] for further background.
Definition 2.39. A graph of spaces (X, Λ) is a graph Λ with the following data:
(1) a connected length space Xv for each v ∈ V Λ (called a vertex space)
(2) a connected length space Xe for each e ∈ EΛ (called an edge space) such that
Xē = Xe
(3) a π1 -injective path-isometric map φe : Xe → Xτ (e) for each e ∈ EΛ.
The total space X is the following quotient space:
!
G G .
X= Xv Xe × [0, 1] ∼,
v∈V Λ e∈EΛ
where ∼ is the relation that identifies Xe × {0} with Xē × {0} via the identification of
Xe with Xē , and the point (x, 1) ∈ Xe × [0, 1] with φe (x). Note that X inherits a length
space metric from the vertex and edge spaces. If Λ is a tree, then we call (X, Λ) a tree
of spaces.
Definition 2.40. A graph of groups G consists of the following data:
(1) a graph Λ (called the underlying graph),
(2) a group Gv for each v ∈ V Λ (called a vertex group),
(3) a group Ge for each e ∈ EΛ (called an edge group) such that Ge = Gē ,
(4) and injective homomorphisms Θe : Ge → Gτ (e) for each e ∈ EΛ (called edge
maps).
If G is a graph of groups with underlying graph Λ, a graph of spaces associated to
G consists of a graph of spaces (X, Λ) with points xv ∈ Xv and xe ∈ Xe so that
π1 (Xv , xv ) ∼
= Gv and π1 (Xe , xe ) ∼
= Ge for all v ∈ V Λ and e ∈ EΛ, and the maps
φe : Xe → Xτ (e) are such that (φe )∗ = Θe . The fundamental group of the graph of
groups G is π1 (X). The fundamental group is independent of the choice of X, so we will
often denote it by π1 (G).
Given a graph of spaces (X, Λ), any covering f : X̂ → X naturally has the structure of
a graph of spaces (X̂, Λ̂), where the edge and vertex spaces correspond to components
of the preimages of the edge and vertex spaces in (X, Λ). Moreover, f restricts to
coverings between edge and vertex spaces. We will usually denote such a covering by
f : (X̂, Λ̂) → (X, Λ) to emphasize that the graph of spaces structure is preserved.
Definition 2.41. An isometric action of a group G on a tree of spaces (X, T ) consists
of the following data:
(1) an action of G on T
(2) for each g ∈ G, v ∈ V T and e ∈ ET , isometries gv : Xv → Xgv and ge = gē :
Xe → Xge , such that if v = τ (e), the following diagram commutes
ge
Xe Xge
φe φge
gv
Xv Xgv .
Moreover, if the vertex and edge spaces are cell complexes, then we additionally
require the maps gv and ge to be cellular.
Note there is an induced isometric action of G on the total space X. We say that the
action of G on (X, T ) is geometric if the action on X is geometric.
We will often consider a group Γ acting on a tree of spaces (X, T ), with simply
connected edge and vertex spaces, free actions of the edge and vertex stabilizers on the
corresponding edge and vertex spaces, and no edge inversions in T . Then X/Γ has the
structure of a graph of spaces with underlying graph T /Γ. We call this the quotient graph
of spaces, and denote it by (X, T )/Γ. Moreover, the quotient map defines a covering
(X, T ) → (X, T )/Γ, and π1 (X/Γ) = Γ.
Let G be a locally compact group generated by a compact set S ⊆ G with Cayley
graph X. The number of ends e(G) of G is the supremum of the number of unbounded
components of X\B, where B is a bounded subset of X. This is independent of the choice
of S, since any two Cayley graphs of G with respect to different compact generating sets
are quasi-isometric. We remark that if the Cayley graph of G is not locally finite, then
the number of ends as defined above may be different from the standard definition as
defined in [DK18].
A Stallings–Dunwoody decomposition of a group Γ is a graph of groups decomposition
satisfying the conclusions of the next theorem.
Theorem 2.42 ([Dun85, Sta68]). If Γ is a finitely presented group, then Γ is the fun-
damental group of a finite graph of groups with finite edge groups and vertex groups that
have at most one end.
2.7. Commanding subgroups. This subsection concerns the notion of a group com-
manding a collection of subgroups. See [She23, Section 6] for more details.
Definition 2.43. A group Γ commands a collection of subgroups (Λ1 , ..., Λk ) if there
exist finite-index subgroups Λ̇i < Λi such that, for any choice of finite-index subgroups
Λ̂i < Λ̇i with Λ̂i / Λi , there exists a finite-index normal subgroup Γ̂ / Γ such that
Λi ∩ Γ̂ = Λ̂i for 1 ≤ i ≤ k.
If a group Γ is the fundamental group of a graph of groups such that each vertex
group commands its collection of incident edge groups, then it is possible to build a
finite cover of the graph of groups in a flexible manner by picking finite-index subgroups
of the vertex groups and using the commanding property to ensure that the incident
edge groups match up. We will employ a variation of this argument in Section 9 to a
setting where two groups act on a tree of spaces, and we wish to build a common finite
cover of the quotient graphs of spaces.
Remark 2.44. Note that the collection of subgroups (Λi ) from Definition 2.43 might
contain duplicates. Indeed, if Γ commands (Λi ), then we can add any number of du-
plicates of the trivial subgroup to our family of subgroups and Γ will still command
them.
Definition 2.45. We say that a group Γ commands its finite subgroups if it commands
any collection (Λ1 , ..., Λk ) of finite subgroups. Equivalently, for every finite subgroup
Λ ≤ Γ, there exists a finite-index subgroup Γ0 ≤ Γ so that Γ0 ∩ Λ = {1}.
For example, residually finite groups and virtually torsion-free groups command their
finite subgroups.
Example 2.46. A group is omnipotent in the sense of [Wis00] if and only if it commands
the subgroups {hgi i} for any finite set of independent elements {gi } [She23].
Example 2.47. The group Zn commands subgroups {Λi } if and only if they are in-
P
dependent in the sense that i λi = 0 with λi ∈ Λi implies λi = 0 for all i [She23,
Proposition 6.5].
A group is said to be cubulated if it acts geometrically on a CAT(0) cube complex.

A collection (Λi )i∈I of subgroups of G is almost malnormal if for all i, j ∈ I and g ∈ G,
the intersection gΛi g −1 ∩ Λj is finite unless i = j and g ∈ Λi .
Example 2.48. A cubulated hyperbolic group commands any finite almost malnormal
collection of quasi-convex subgroups. This follows from Wise’s Malnormal Special Quo-
tient Theorem [Wis21] combined with theorems of Agol [Ago13] and Osin [Osi07]; see
[She23, Theorem 1.4]. If hyperbolic groups are all residually finite, then we can extend
this to any hyperbolic group commanding any finite, almost malnormal collection of
quasi-convex subgroups [She23, Theorem 6.9].
Remark 2.49. In general one cannot choose Λ̇i = Λi in Definition 2.45. For instance,
let Γ be the free group with generators a, b, and consider Λ1 = hai, Λ2 = hbi, Λ3 = habi
and Λ̂1 = Λ1 , Λ̂2 = Λ2 , Λ̂3 = hababi. Then any Γ̂ / Γ with Λi ∩ Γ̂ = Λ̂i for i = 1, 2 must
contain a and b, so Γ̂ = Γ, and Λ3 ∩ Γ̂ = Λ3 6= Λ̂3 .
3. Introducing graphically discrete groups

3.1. Equivalent definitions of graphical discreteness. If the automorphism group
of a connected locally finite graph X is discrete, then two groups Γ and Γ0 acting
geometrically on that graph are virtually isomorphic. Indeed, modulo the finite kernels
of the action, Γ and Γ0 are finite-index subgroups of Aut(X). However, it is too much
to expect that for any action of a group Γ on a connected locally finite graph X that
Aut(X) is discrete: indeed, attaching a pair of leaves at every vertex of X produces
a graph with non-discrete automorphism group. The following definition accounts for
these compact obstructions to discreteness. See Figure 1.
Definition 3.1. If X is a graph and G ≤ Aut(X), then a G-imprimitivity system is an
equivalence relation ∼ on V X that is invariant under the action of G, i.e. g · v ∼ g · w
whenever v ∼ w. Equivalence classes of a G-imprimitivity system are called blocks. The
quotient graph X/ ∼ is the simplicial graph whose vertices are blocks, and two blocks [v]
and [w] are joined by an edge if and only if (v 0 , w0 ) ∈ EX for some v 0 ∈ [v] and w0 ∈ [w].
Remark 3.2. Even if the imprimitivity system ∼ is trivial, i.e. all blocks consist of a
single vertex, the graph X/ ∼ might not be equal to X. This is because X might not be
a simplicial graph, and quotienting out by the trivial imprimitivity system will remove
one-edge loops and multiple edges.
Remark 3.3. If G ≤ Aut(X) and ∼ is a G-imprimitivity system, then there is an
induced action of G on X/ ∼ given by g · [v] = [g · v].
X X/ ∼
Figure 1. The automorphism group of the graph X is not discrete. The

vertices are grouped in blue Aut(X)-invariant finite equivalence classes
so that the action of Aut(X) on the quotient graph X/ ∼ is discrete.
A topological group G is compact-by-discrete if G contains a compact normal subgroup

such that the quotient, equipped with the quotient topology, is discrete. Equivalently,
G is compact-by-discrete if it contains a compact open normal subgroup, since a normal
subgroup is open if and only if the corresponding quotient is discrete. We can charac-
terize compact-by-discrete automorphism groups using imprimitivity systems as follows.
See Theorem 3.10 for further characterizations.
Proposition 3.4. Let X be a connected locally finite graph and let G ≤ Aut(X) be a
closed subgroup that acts cocompactly on X. Then G is compact-by-discrete if and only
if there exists a G-imprimitivity system ∼ on X with finite blocks such that the induced
action G y X/ ∼ is discrete.
Proof. First, suppose there is a compact subgroup K C G such that G/K is discrete.
In particular, K is open. We define an equivalence relation ∼ on V X by v ∼ w if and
only if there exists k ∈ K such that kv = w. As K is normal, this is a G-imprimitivity
system. Since K is compact, every orbit K · v is finite by Proposition 2.27, and so ∼ has
finite blocks. The kernel of the induced action G y X/ ∼ is open because it contains the
open subgroup K, and this is equivalent to saying that the induced action G y X/ ∼
is discrete.
Conversely, suppose there exists a G-imprimitivity system ∼ on X with finite blocks
such that the induced action G y X/ ∼ is discrete. Since ∼ has finite blocks, the
kernel K of the action G y X/ ∼ is compact by Proposition 2.27. The quotient G/K is
discrete because the action G y X/ ∼ is discrete. Thus, G is compact-by-discrete.
Example 3.5. If G ≤ Aut(X) is cocompact and compact-by-discrete, then Proposi-
tion 3.4 tells us that there exists a G-imprimitivity system ∼ on X with finite blocks
such that the induced action G y X/ ∼ is discrete. However, it is not necessarily true
that Aut(X/ ∼) is a discrete group, as shown in Figure 3.1; note that Aut(X) does not
surject onto Aut(X/ ∼) in this example.
X/ ∼
Figure 2. An example where Aut(X) acts discretely on X/ ∼, but

the group Aut(X/ ∼) is not discrete. On the left, each set of colored
vertices indicates one equivalence class, and each black vertex is in its
own equivalence class.
We now turn to the definition of graphical discreteness.

Proposition 3.6. Let Γ be a finitely generated group. The following are equivalent:
(1) If Γ is a virtual uniform lattice in a tdlc group G, then G is compact-by-discrete.
(2) If ρ : Γ → G is a virtual uniform lattice embedding with G a tdlc group, then ρ
is trivial (in the sense of Definition 2.10).
(3) If Γ acts geometrically on a connected locally finite graph X, then Aut(X) is
compact-by-discrete.
If Γ satisfies any of these equivalent properties, we say Γ is graphically discrete.
Proof. (2) ⇐⇒ (1) is a consequence of Definition 2.10 and Remark 2.11.
(1) =⇒ (3): Suppose ρ : Γ → Aut(X) is a geometric action of Γ on a connected
locally finite graph X. By Lemma 2.25, ρ is a virtual uniform lattice embedding. Since
Aut(X) is totally disconnected, (1) implies Aut(X) is compact-by-discrete.
(3) =⇒ (1): Suppose Γ is a virtual uniform lattice in a tdlc group G. Then G is
compactly generated by Lemma 2.13, and, by Lemma 2.26, G admits a continuous,
proper, cocompact action ρ : G → Aut(X) on a connected locally finite graph X. In
particular, Γ acts geometrically on X. It then follows from (3) that Aut(X) has a
compact open normal subgroup K. Then N := ρ−1 (K) is clearly open and normal in
G, and is compact because the action of G on X is proper. Hence G is compact-by-
discrete.
Remark 3.7. Suppose Γ is graphically discrete and contains a maximal finite normal
subgroup. If Γ → G is a virtual uniform lattice embedding with G totally disconnected,
then the maximal compact normal subgroup of G, which exists by Proposition 2.14, is
open.
It thus follows from Proposition 3.6 that a finitely generated group Γ is graphically
discrete if all its virtual uniform lattice embeddings are trivial. Since much of the
literature is concerned with genuine (i.e. injective) lattice embeddings rather than virtual
lattice embeddings, we note the following consequence of Proposition 3.6.
Corollary 3.8. Let Γ be a finitely generated group. Suppose there exists a finite-index
torsion-free subgroup Γ0 ≤ Γ such that whenever ρ : Γ0 → G is a uniform lattice embed-
ding with G a tdlc group, then ρ is trivial. Then Γ is graphically discrete.
Proof. If ρ : Γ → G is virtual lattice embedding, then ρ|Γ0 is injective, hence a lattice

embedding of Γ0 .
One useful consequence of graphical discreteness, foreshadowing many of the rigidity

results to appear later, is the following:
Proposition 3.9. Let Γ1 and Γ2 be two finitely generated groups, at least one of which
is graphically discrete. If Γ1 and Γ2 act geometrically on the same connected locally
finite graph, then Γ1 and Γ2 are virtually isomorphic.
Proof. Suppose Γ1 is graphically discrete and that ρi : Γi → Aut(X) is a geometric

action on a connected locally finite graph X for i = 1, 2. As Γ1 is graphically discrete,
there exists a compact subgroup K C Aut(X) with Aut(X)/K discrete. For i = 1, 2,
Γi /ρ−1 −1
i (K) is isomorphic to a finite-index subgroup of Aut(X)/K, thus Γ1 /ρ1 (K) and
Γ2 /ρ−1 −1 −1
2 (K) are abstractly commensurable. Since ρ1 (K) and ρ2 (K) are finite, Γ1 and
Γ2 are virtually isomorphic.
3.2. Characterization of compact-by-discrete automorphism groups.
Theorem 3.10. Let X be a connected locally finite graph and let G ≤ Aut(X) be a
closed subgroup that acts cocompactly on X. With the topology on G induced from that
on Aut(X), the following are equivalent:
(1) G is compact-by-discrete.
(2) There exists a constant A > 0 such that |Gx · y| ≤ A for all x, y ∈ V X, where
Gx denotes the stabilizer of x.
(3) The homomorphism φ : G → QI(X) has open kernel.
(4) For some (equivalently every) x ∈ V X, the stabilizer Gx has finite image under
the homomorphism φ : G → QI(X).
(5) There exists a G-imprimitivity system ∼ on X with finite blocks such that the
induced action G y X/ ∼ is discrete.
Remark 3.11. As X is locally finite, all vertex stabilizers Gx are commensurable by

Example 2.22. This ensures that for φ : G → QI(X) as above, φ(Gx ) is finite for some
x ∈ V X if and only if φ(Gx ) is finite for every x ∈ V X.
The theorem above provides several additional ways to define graphical discreteness.
Note the equivalence of (1) and (5) was established in Proposition 3.4. The result is also
a useful tool for showing that certain groups are not graphically discrete. For instance,
non-abelian free groups act geometrically on the 4-valent tree, so one can deduce the
following corollary from Theorem 3.10(2) or (4).
Corollary 3.12. Non-abelian free groups are not graphically discrete.
See Section 3.4 for more examples of non-graphically-discrete groups whose proofs use
Theorem 3.10, and see Section 4 for further analysis in the setting of hyperbolic groups.
We first prove some of the easier implications in Theorem 3.10.
Lemma 3.13. The implications (1) ⇐⇒ (2) and (1) =⇒ (3) =⇒ (4) of Theorem 3.10
hold.
Proof. (1) =⇒ (2): Let K / G be a compact open normal subgroup. The vertex orbits
of K are finite because K is compact (Proposition 2.27), and this set of orbits is G-
invariant because K is normal. Since G acts cocompactly on X, the sizes of the vertex
orbits of K are bounded by some constant N . Given x ∈ V X, the stabilizer Gx is
compact and open (Proposition 2.27). Since the intersection Kx = Gx ∩ K is open, Kx
has finite index in Gx . Then given a second vertex y, we have
|Gx · y| ≤ [Gx : Kx ]|Kx · y| ≤ [Gx : Kx ]N.
Finally, the index [Gx : Kx ] can be bounded independently of x because K is normal
and G acts cocompactly on X.
(2) =⇒ (1): Schlichting’s Theorem [Sch80] (see also [BL89, Theorem 3]) states that
if G is a group with a subgroup H such that the conjugates of H are uniformly com-
mensurable (meaning there is a finite upper bound for the indices [H : H ∩ gHg −1 ] with
g ∈ G), then H is commensurable with a normal subgroup K / G. Moreover, we can
assume that K contains a finite intersection of conjugates of H (see the proof of [BL89,
Theorem 3]). With G being the group from Theorem 3.10, now suppose there exists a
constant A > 0 such that |Gx · y| ≤ A for all x, y ∈ V X. Since [Gx : Gx ∩ Gy ] = |Gx · y|,
all conjugates of a vertex stabilizer Gx are uniformly commensurable. So, we can apply
Schlichting’s Theorem with H being any vertex stabilizer Gx . The resulting normal sub-
group K / G contains a finite intersection of vertex stabilizers as a finite-index subgroup.
Hence, K is open by the definition of the compact-open topology, and K is compact by
Proposition 2.27.
(1) =⇒ (3): Let K / G be a compact open normal subgroup. The vertex orbits of K
are finite because K is compact (Proposition 2.27), and the set of orbits is G-invariant
because K is normal. Since G acts cocompactly on X we deduce that the diameters
of the vertex orbits of K are uniformly bounded, hence K lies in the kernel of the
homomorphism φ : G → QI(X). As K is open, we conclude that ker(φ) is open.
(3) =⇒ (4): Suppose ker(φ) is open and let x ∈ V X. As ker(φ) is an open neigh-
borhood of the identity automorphism, the pointwise G-stabilizer of some ball about x
is contained in ker(φ) (see [CdlH16, Example 5.B.11]). But X is locally finite, so this
pointwise stabilizer has finite index in Gx . Hence φ(Gx ) is finite.
To complete the proof of Theorem 3.10, it remains to prove the implication (4) =⇒ (1).
For this we need the following theorem of Trofimov. Recall a bounded automorphism
φ ∈ Aut(X) of a graph X is a graph automorphism of X such that supx∈V X d(x, φ(x)) <
∞. Note that a graph automorphism is bounded precisely if it lies in the kernel of the
natural map Aut(X) → QI(X).
Theorem 3.14. [Tro84, Proposition 2.3] Let X be a connected locally finite vertex-
transitive graph. There exists an Aut(X)-imprimitivity system ∼ on X with finite blocks,
such that no nontrivial bounded automorphism of the induced action Aut(X) y X/ ∼
stabilizes a vertex of X/ ∼.
We now complete the proof of Theorem 3.10 with the following lemma.
Lemma 3.15. The implication (4) =⇒ (1) from Theorem 3.10 holds. Namely, given a
connected locally finite graph X and G < Aut(X) a closed subgroup acting cocompactly
on X, if the stabilizer Gx has finite image under the homomorphism φ : G → QI(X) for
all x ∈ V X, then G is compact-by-discrete.
Proof. Fix x ∈ V X. Given d > 0, we define a graph Y with vertex set the G-orbit
G · x and an edge joining x1 , x2 ∈ G · x whenever d(x1 , x2 ) ≤ d. The graph Y is locally
finite for all d, connected for sufficiently large d, and the group G acts on Y with a
single vertex orbit. We pick d sufficiently large so that Y is connected. Let ∼ be the
Aut(Y )-imprimitivity system on Y provided by Proposition 3.14. In particular, ∼ is a G-
imprimitivity system with finite blocks such that no nontrivial bounded automorphism
of the induced action G y Y / ∼ stabilizes a vertex of Y / ∼.
Let K be the kernel of the homomorphism G → Aut(Y / ∼). Proposition 2.27 implies
that K is compact, so it remains to show that K is open. Let Kx = Gx ∩ K; since K is a
union of Kx -cosets, it suffices to prove Kx is open. The subgroup Gx stabilizes the vertex
[x] ∈ V (Y / ∼), so all elements of Gx − Kx act on Y / ∼ as unbounded automorphisms,
hence they also act on X as unbounded automorphisms. Thus, the elements of Gx − Kx
lie outside the kernel of φ : G → QI(X). We assumed that φ(Gx ) is finite, so we deduce
that Kx has finite index in Gx . Let Gx = Kx t g1 Kx t ... t gn Kx be the partition of Gx
into Kx -cosets. The gi are not in K, so gi [xi ] 6= [xi ] for some choice of ∼-block [xi ]. We
deduce that
\n
Kx = Gx ∩ G[xi ] ,
i=1
where G[xi ] is the stabilizer of the block [xi ]. A finite intersection of open subgroups is
open, therefore Kx is open. Thus, K is open, as required.
3.3. Examples of graphically discrete groups.

Theorem 3.16. Finitely generated groups in the following families are graphically dis-
crete:
(1) Virtually nilpotent groups [Tro84].
(2) Irreducible lattices in connected center-free, real semisimple Lie groups without
compact factors, other than non-uniform lattices in PSL2 (R) [Fur01, BFS20].
(3) Irreducible S-arithmetic lattices in the sense of [BFS20].
(4) Fundamental groups of closed irreducible 3-manifolds with nontrivial geometric
decomposition.
(5) Finitely generated groups Γ containing a finite-index torsion-free subgroup with

no nontrivial lattice embeddings. Examples include:
(a) The mapping class group of a non-exceptional finite-type orientable surface
[Kid10, Theorem 1.4]
(b) The pure braid group PBk (Sg ) of k strands on the closed surface of genus
g, where g, k ≥ 2 [CK15, Theorem B].
(c) Out(Fn ) for n ≥ 3 [GH21, Theorem 1.5]
(d) Various 2-dimensional Artin groups of hyperbolic type [HH20, Theorem
10.9]
(e) The Higman groups
Hign = ha1 , . . . , an | ai ai+1 a−1 2

i = ai+1 for i = 1, . . . , n mod n i
for n ≥ 5 [HH22, Corollary 1.3].

(f ) Fundamental groups of Riemannian n-manifolds for n ≥ 5, with pinched
1 2
negative curvature in the range [−(1 + n−1 ) , −1], that are not homeomor-
phic to closed hyperbolic manifolds [BFS20, Theorem E].
(6) Any finitely generated tame (see Definition 2.29) group Γ for which the natural
map Γ → QI(Γ) has finite-index image. Examples include:
(a) the mapping class group of non-exceptional finite-type orientable surface
[BKMM12];
(b) hyperbolic surface–by–free groups considered by Farb–Mosher [FM02, The-
orem 1.3];
(c) topologically rigid hyperbolic groups of Kapovich–Kleiner [KK00].
Proof. (1): Since Γ is virtually nilpotent, it has polynomial growth [Wol68]. Suppose Γ
acts geometrically on a locally finite vertex transitive graph X. Then Γ is quasi-isometric
to X so X has polynomial growth. A result of Trofimov, which generalizes Gromov’s
polynomial growth theorem, ensures there is an Aut(X)-imprimitivity system ∼ on X
with finite blocks such that the induced action Aut(X) y X/ ∼ is discrete [Tro84,
Theorem 1]. Thus Γ satisfies (1) of Proposition 3.6, hence is graphically discrete.
(2): Let H be a center-free, connected real semisimple Lie group without compact
factors. Let Γ ≤ H be an irreducible lattice, which is assumed to be uniform if H is
locally isomorphic to PSL2 (R). By Selberg’s lemma, Γ contains a finite-index torsion-
free subgroup Γ0 , which is also a lattice in H. Then [BFS20, Theorems B and D] ensure
every lattice embedding ρ : Γ0 ,→ G is trivial or virtually isomorphic to Γ0 ,→ H.
The latter is impossible if G is totally disconnected, thus Γ is graphically discrete by
Corollary 3.8.
(3): This also follows from [BFS20, Theorem B], similarly to (2) above (see [BFS20,
Remark 4.2]).
(4): This is shown in Theorem 7.9.
(5): Any such group satisfies the hypotheses of Corollary 3.8, hence is graphically dis-
crete. We note that Hign is torsion-free [HH22, Lemma 2.7], and every lattice embedding
of Hign is trivial; see [HH22, Corollary 5.20].
(6): This will follow from Theorem 3.26.
3.4. Examples of non-graphically-discrete groups.

Proposition 3.17. Let Γ be a finitely generated group and let Λ ≤ Γ be a proper

subgroup. Then the HNN extension
Γ∗Λ = hΓ, t | tλt−1 = λ, λ ∈ Λi
is not graphically discrete.
Proof. Let S be a finite generating set for Γ and let X be the Cayley graph of Γ∗Λ with
respect to the generating set S ∪{t}. Let T be the Bass–Serre tree of the HNN extension
Γ∗Λ . There is a natural map p : X → T given by collapsing all edges in X labeled by
generators in S.
We first define a certain automorphism g ∈ Aut(X) that descends to an automorphism
of T . Each element w ∈ Γ∗Λ can be expressed as a minimal length word in the generators
S ∪ {t} of the form w = tn w0 , where |n| is maximal. Define a map Γ∗Λ → Γ∗Λ by
tn w0 7→ t−n w0 . It is straightforward to check that this is a well-defined bijection of
Γ∗Λ . We also claim that it defines an automorphism g : X → X preserving edge labels.
Indeed, given w = tn w0 as above and a ∈ S ∪ {t} we have that wa 7→ t−n w0 a, unless
a = t and w0 ∈ Λ in which case wa = tn+1 w0 7→ t−n−1 w0 = t−n w0 t−1 ; so we always have
that the edge from w to wa maps to another edge in X with the same label. Since g
preserves edge labels, it descends to an automorphism ḡ of T . Moreover, g fixes the base
vertex 1 ∈ Γ∗Λ = V X, so ḡ fixes the base vertex x := p(1) ∈ T . Furthermore, ḡ swaps
the edges (tx, x) and (t−1 x, x) and fixes all other edges in lk(x). One can also show that
the support of ḡ is precisely the union of all geodesic rays that start at either tx or t−1 x
and avoid x. Finally, the support of g is the preimage of the support of ḡ.
We now define a family of automorphisms of X that each descend to an automorphism
of T and are pairwise at infinite distance from each other. Let w ∈ Γ − Λ and wi = ti w
for i ∈ Z. Each wi acts on Γ∗Λ by left multiplication, so defines an automorphism
of X that preserves edge labels. Hence we get sequences of automorphisms (wi gwi−1 )
in Aut(X) and (w̄i ḡ w̄i−1 ) in Aut(T ). Observe that the automorphisms (w̄i ḡ w̄i−1 ) fix x
(in fact they fix tj x for all j ∈ Z) and have disjoint supports; in particular, they are
at infinite distance from one another. Since the map p : X → T can only decrease
distances, we deduce that the automorphisms (wi gwi−1 ) are also at infinite distance
from one another, and they all fix the base vertex 1 ∈ V X. Therefore, we deduce from
Theorem 3.10(1) and (4) that Aut(X) is not compact-by-discrete, and that Γ∗Λ is not
graphically discrete.
We use Proposition 3.17 to classify the graphically discrete right-angled Artin groups.
Corollary 3.18. The right-angled Artin group AΓ is graphically discrete if and only if
Γ is a complete graph.
Proof. If Γ is complete, then AΓ is free abelian, and graphical discreteness follows from
Theorem 3.16(1). Otherwise, there exists v ∈ Γ so that {v} ∪ lk(v) 6= V Γ. Thus, AΓ can
be expressed as a nontrivial HNN extension
AΓ = AΓ−{v} ∗Alk(v) ,
where the generator v is the stable letter as in Proposition 3.17. Thus, the group AΓ is
not graphically discrete.
Proposition 3.17 also yields many examples of non-graphically-discrete infinite-ended
groups (beyond right-angled Artin groups).
Corollary 3.19. For all nontrivial finitely generated groups Γ, the free product Γ ∗ Z is
Corollary 3.20. If Γ1 , Γ2 are nontrivial finitely generated groups containing nontrivial

proper finite-index subgroups, the free product Γ = Γ1 ∗ Γ2 has a finite-index subgroup
that is not graphically discrete.
Proof. Since the groups Γ1 and Γ2 contain proper finite-index subgroups, there exists
a finite-index subgroup Γ̂ ≤ Γ whose Grushko decomposition contains a nontrivial free
factor. For example, let Y be the graph of spaces associated to Γ1 ∗ Γ2 . By taking covers
of the vertex spaces of Y , there exists a finite cover Ŷ → Y such that the underlying
graph of Ŷ is not a tree. Then Γ̂ := π1 (Ŷ ) is a finite-index subgroup of Γ containing a
nontrivial free factor. By Corollary 3.19, Γ̂ is not graphically discrete.
Yet more examples of non-graphically-discrete infinite-ended groups are given by the

following proposition.
Proposition 3.21. Let Γ1 , Γ2 be nontrivial finitely generated groups. Then the free
product Γ = Γ1 ∗ Γ2 ∗ Γ2 is not graphically discrete.
Proof. Let S1 , S2 be finite generating sets for Γ1 , Γ2 respectively. Let Γ02 be a copy of
Γ2 and let S20 be the copy of S2 in Γ02 . We can then write Γ = Γ1 ∗ Γ2 ∗ Γ02 , and consider
the Cayley graph X of Γ with respect to the finite generating set S = S1 t S2 t S20 .
Now construct a graph Y from X by replacing each vertex x ∈ V X with a tripod Tx .
That is, let x1 , x2 , x02 denote the endpoints of Tx , and for each edge incident to x in X
labeled by a generator in S1 (resp. S2 , S20 ) we connect this edge to x1 (resp. x2 , x02 ) in
Y . There is a natural geometric action of Γ on Y . Assuming |S1 |, |S2 | > 1, the centers
of the tripods are the only degree 3 vertices in Y , so all automorphisms of Y preserve
the collection of tripods. It is clear from the symmetry of the construction that for each
x ∈ V X, there is an automorphism of Y that fixes x1 and swaps x2 , x02 – call this the
flip at Tx .
By collapsing the edges labeled by S1 , S2 , S20 we get an Aut(Y )-equivariant map to
a tree p : Y → T (this is just a Bass–Serre tree for the splitting Γ = Γ1 ∗ Γ2 ∗ Γ02 ).
The flip at Tx induces an automorphism of T that fixes p(x1 ) and swaps p(x2 ), p(x02 ).
Given a vertex y ∈ Y , it is clear that the Aut(Y )-stabilizer of y contains infinitely
many flips, so it has arbitrarily large vertex orbits in both Y and T . It then follows
from Theorem 3.10(1) and (2) that Aut(Y ) is not compact-by-discrete, hence Γ is not
graphically discrete.
We remark Corollary 3.19 and Proposition 3.21 demonstrate that infinite-ended are
frequently not graphically discrete, and so action rigidity theorems for such groups, for
instance Theorem 10.4, cannot be deduced cheaply via Proposition 3.9.
3.5. Uniform lattices and quasi-actions. In this subsection, we first summarize

a construction due to Furman relating uniform lattice embeddings and quasi-actions
[Fur01, §3.2]. We conclude the subsection by showing that certain finitely generated
groups that have finite index in their quasi-isometry group are graphically discrete.
Recall that by Lemma 2.13 if a locally compact group G contains a virtual uniform
lattice, then G is compactly generated.
Definition 3.22. Let Γ be a finitely generated group, G a locally compact group, and
ρ : Γ → G a virtual uniform lattice embedding. Equip Γ and G with word metrics dΓ and
dG with respect to a finite and compact generating set, respectively (see Lemma 2.13).
The map ρ : Γ → G is then a quasi-isometry with coarse inverse ρ [CdlH16, Proposition
5.C.3]. The induced quasi-action of G on Γ is the collection of maps {fg }g∈G , where
fg : Γ → Γ is the quasi-isometry given by fg = ρ ◦ Lg ◦ ρ, where Lg is left multiplication
by g.
Remark 3.23. Although the topology on G induced by the metric dG does not coincide
with the topology on G (unless G is discrete), this metric is adapted in the sense of
[CdlH16]. That is, the metric dG is left-invariant, proper, and locally bounded, i.e.
every point in G has a neighborhood of finite diameter. In particular, the following
properties are satisfied:
(P1) L ⊆ G has compact closure if and only if it is bounded in (G, dG );
(P2) there is some sufficiently large R such that for every g ∈ G, the open ball BR (g)
contains an open neighborhood of g.
The next lemma relates the topology of G to the geometry of the induced quasi-action.
Lemma 3.24. Let ρ : Γ → G be as above, and let {fg }g∈G be the induced quasi-action
of G on Γ. There is a constant B ≥ 0 such that if a sequence (gi )i∈N converges to g in
G, then for all x ∈ Γ, there is some Nx ∈ N such that
dΓ (fgi (x), fg (x)) ≤ B
for all i ≥ Nx .
Proof. Pick K ≥ 1 and A ≥ 0 such that {fg }g∈G is a (K, A)-quasi-action, ρ and ρ are
(K, A)-quasi-isometries, and the compositions ρρ and ρρ are A-close to the identity.
Without loss of generality, we may assume ρ(1Γ ) = 1G and ρ(1G ) = 1Γ . Since {fg }g∈G
is a quasi-action, it is sufficient to prove the lemma in the case where g is the identity.
So, let (gi )i∈N be a sequence in G converging to 1G .
Using Property (P2) above, pick a constant R sufficiently large such that BR (1G )
contains an open neighborhood of 1G . Let x ∈ Γ. As gi converges to the identity, so
does ρ(x)−1 gi ρ(x). Therefore dG (ρ(x)−1 gi ρ(x), 1G ) ≤ R for all i sufficiently large. It
follows that
dΓ (fgi (x), x) = dΓ (ρ(gi ρ(x)), x) ≤ KdG (gi ρ(x), ρ(x)) + 2A
= KdG (ρ(x)−1 gi ρ(x), 1G ) + 2A ≤ KR + 2A
for all i sufficiently large.
We can also describe compact normal subgroups of G geometrically:
Lemma 3.25. Let ρ : Γ → G be as above, and let {fg }g∈G be the induced quasi-action
of G on Γ. If K C G is a compact normal subgroup, there is a constant B such that for
all k ∈ K, supx∈Γ d(fk (x), x) ≤ B.
Proof. The Milnor–Schwarz lemma for locally compact groups [CdlH16, Theorem 4.C.5]
says that the action of G on the Cayley graph of G/K is quasi-conjugate the action
of G on its Cayley graph. Therefore, the action of G on G/K is quasi-conjugate to
the induced quasi-action {fg }g∈G of G on Γ. Since K acts trivially on G/K, the result
follows.
Theorem 3.26. Let Γ be a finitely generated tame group such that the image of the
homomorphism φ : Γ → QI(Γ) induced by left-multiplication is a finite-index subgroup
of QI(Γ). Then every virtual uniform lattice embedding of Γ is trivial. In particular, Γ
is graphically discrete.
Proof. Suppose ρ : Γ → G is a uniform virtual lattice embedding. By Remark 2.15

we can restrict to the case where G is second countable. Let ψ : G → QI(Γ) be the
homomorphism induced by the quasi-action {fg }g∈G of G on Γ, where ψ(g) = [fg ]. We
will show ψ has compact open kernel. Equip QI(Γ) with the word metric with respect
to a finite generating set. The natural homomorphism φ : Γ → QI(Γ), which has finite-
index image by assumption and finite kernel since Γ is tame, is then a quasi-isometry.
We note the composition ψ ◦ ρ coincides with φ.
To set notation, pick K, A, R and B as in the proof and statement of Lemma 3.24.
Let F = {[f1 ], . . . , [fn ]} ⊆ QI(Γ) be a set of right φ(Γ) representatives. Increasing K and
A if necessary, we can assume each fi is a (K, A)-quasi-isometry. Thus every element
of QI(Γ) can be represented by a (K, A)-quasi-isometry of the form φ(h) ◦ fi for some
h ∈ Γ and 1 ≤ i ≤ n. Let C = C(K, A) be the tameness constant as in Definition 2.29.
We first show ker(ψ) is open, which implies ψ is continuous. Since G is second
countable, it is sufficient to show that if (gi ) is a sequence in G converging to 1G ,
then gi ∈ ker(ψ) for i sufficiently large. By Lemma 3.24, dΓ (fgi (1Γ ), 1Γ ) ≤ B for all
i sufficiently large. There are sequences (hi ) in Γ and (ki ) in {1, . . . , n} such that
[fgi ] = [φ(hi ) ◦ fki ] for all i ∈ N. Let R1 = maxni=1 dΓ (fi (1Γ ), 1Γ ). Then,
dΓ (hi , 1Γ ) ≤ dΓ (hi , hi fki (1Γ )) + dΓ (hi fki (1Γ ), 1Γ ) = dΓ (1Γ , fki (1Γ )) + d(hi fki (1Γ ), 1Γ )
≤ R1 + dΓ (hi fki (1Γ ), 1Γ ) ≤ dΓ (fgi (1Γ ), 1Γ ) + R1 + C ≤ B + R1 + C
for all i sufficiently large. Therefore, {ψ(gi ) | i ∈ N} is finite. If ψ(gi ) is not eventually
the identity, we can replace (gi )i∈N with a subsequence such that ψ(gi ) is constant and
equal to some [f ] 6= 1QI(Γ) . Without loss of generality, we may assume f is a (K, A)-
quasi-isometry. Since [f ] is not the identity, for all r ≥ 0, there is some x ∈ Γ such that
dΓ (f (x), x) > r. In particular, there is some x ∈ Γ such that dΓ (f (x), x) > B + C. Since
[fgi ] = [f ], dΓ (fgi (x), f (x)) ≤ C. Therefore, dΓ (fgi (x), x) > B for all i, contradicting
the choice of B.
It remains to show ker(ψ) is compact. Let g ∈ ker(ψ). Then dΓ (fg (1Γ ), 1Γ ) ≤ C, so
dG (g, 1G ) ≤ dG (ρ(ρ(g)), 1G ) + A ≤ KdΓ (fg (1Γ ), 1Γ ) + 2A ≤ KC + 2A.
Since the word metric on G is adapted, ker(ψ) has compact closure by Property (P1).
Since ker(ψ) is the kernel of a continuous map, it is closed. Thus, ker(ψ) is compact.
Question 3.27. Can the tameness assumption be removed from the previous theorem?
That is, if Γ is a finitely generated group such that the image of the homomorphism
φ : Γ → QI(Γ) induced by left-multiplication is a finite-index subgroup of QI(Γ), is every
virtual lattice embedding of Γ trivial?
4. Graphical discreteness and non-discreteness for hyperbolic groups

4.1. Boundary characterization of graphical discreteness. The next theorem is
particularly useful for exhibiting groups that are not graphically discrete. For example,
it follows immediately from the theorem that non-abelian free groups are not graphically
discrete. We refer to [BH99] for background on boundaries of hyperbolic groups. If Γ

is a hyperbolic group, we always assume Homeo(∂Γ) is equipped with the topology of
uniform convergence.
Theorem 4.1. A finitely generated non-elementary hyperbolic group Γ is graphically

discrete if and only if for every geometric action of Γ on a connected locally finite
graph X and for all vertices x ∈ X, the image of the induced action of the stabilizer
(Aut(X))x → Homeo(∂X) is finite.
Proof. Suppose Γ acts geometrically on a connected locally finite graph X, and let
G = Aut(X). The induced map G → Homeo(∂X) naturally factors through QI(X),
where the induced map QI(X) ,→ Homeo(∂X) is injective [DK18, Corollary 11.115].
Thus the image of Gx in QI(X) is finite if and only if the image of Gx in Homeo(∂X)
is finite. The result follows from Proposition 3.6 and Theorem 3.10.
We refer to [Ron89, Dav98] for background and the definition of hyperbolic buildings.
Corollary 4.2. Uniform lattices in thick hyperbolic buildings are not graphically dis-
crete. In particular, non-abelian free groups are not graphically discrete.
Proof. Let X be a thick hyperbolic building. By definition, each chamber of X is

contained in infinitely many apartments, and for any two such apartments there is
an automorphism of the building fixing the chamber pointwise and interchanging the
apartments. Each apartment is quasi-convex, hence its limit set embeds in the boundary
of X. Moreover, distinct apartments have distinct limit sets. Thus, a uniform lattice in
the isometry group of the building is not graphically discrete by Theorem 4.1.
A JSJ decomposition of a finitely presented group is a graph of groups decomposition

encoding how a group splits over a prescribed family of subgroups. In this article,
we only consider JSJ decompositions of one-ended hyperbolic groups over two-ended
subgroups, and we only consider the JSJ decomposition described by Bowditch [Bow98].
The associated Bass–Serre tree, called the JSJ tree, is canonically determined by the
local cut point structure of the boundary. We refer the reader to Bowditch’s article for
details of this [Bow98], and to the monograph of Guirardel–Levitt for a reference on JSJ
decompositions more generally [GL17].
Let Γ be a one-ended finitely generated hyperbolic group with nontrivial Bowditch
JSJ tree T . We equip Aut(T ) with the topology of pointwise convergence and note that
in non-degenerate cases, the tree T is not locally finite. Using the description of T in
terms of the topology of the boundary, we show the following:
Proposition 4.3. If Γ is a one-ended hyperbolic group with nontrivial JSJ tree T , there
exists a continuous monomorphism Ψ : Homeo(∂Γ) → Aut(T ) extending the action of
Γ on T .
Proof. We first summarize the construction of Bowditch’s tree. The tree T is defined
from a collection Ω of non-empty subsets of ∂Γ. In the terminology of [Bow98], Ω is
the collection of ∼-classes and ≈-pairs, although familiarity with this terminology is
not needed here. The set Ω is equipped with a certain betweenness relation, where ω is
between ω1 and ω2 if there exist a, b ∈ ω, c ∈ ω1 and d ∈ ω2 such that c and d lie in
different components of ∂Γ\{a, b}. The set Ω and this betweenness relation depend only
on the local cut point structure of ∂Γ, hence Homeo(∂Γ) acts on the set Ω preserving
the betweenness relation.
The tree T is constructed in a combinatorial way from Ω such that elements of Ω
are a subset of vertices of T , and the betweenness relation on T agrees with that of Ω.
There is no proper subtree of T containing Ω, hence any bijection of Ω preserving the
betweenness relation induces a unique automorphism of T . Thus there is a canonical
homomorphism Ψ : Homeo(∂Γ) → Aut(T ) determined by the action of Homeo(∂Γ) on
Ω. The action of Γ on its JSJ tree T is induced by the action of Γ on ∂Γ, hence Ψ
extends the action of Γ. All this is contained in [Bow98].
We now show Ψ is continuous. Let (φi )i∈I be a net in Homeo(∂Γ) converging to the
identity. We show for each ω ∈ Ω, there is some i0 ∈ I such that φi (ω) = ω for all i ≥ i0 .
By the above discussion, this will imply Ψ(φi ) converges pointwise on T . Since the tree
T is discrete and has no leaves, we can pick ω1 , ω2 ∈ Ω such that ω is the unique element
of Ω between ω1 and ω2 . Thus for all i ∈ I, φi (ω) is the unique element of Ω between
φi (ω1 ) and φi (ω2 ).
Since ω is between ω1 and ω2 , there exist a, b ∈ ω, c ∈ ω1 and d ∈ ω2 such that c and
d lie in different components of ∂Γ \ {a, b}. As (φi )i∈I converges to the identity, there
exists i0 ∈ I such that φi (c) and φi (d) lie in different components of ∂Γ \ {a, b} for all
i ≥ i0 . Hence ω is between φi (ω1 ) and φi (ω2 ) for all i ≥ i0 . Therefore, φi (ω) = ω for all
i ≥ i0 as required, showing Ψ is continuous.
We now show Ψ is injective. We fix φ ∈ ker(Ψ). Pick an infinite order element g ∈ Γ
fixing an edge e of T . Then ω = {g + , g − } is the cut pair in ∂Γ corresponding to e; see
[Bow98]. Then for all h ∈ Γ, φ must stabilize the edge he, hence φ(hω) = hω. We claim
that for every open U ⊆ ∂Γ, there is some h ∈ Γ with hω ⊆ U . Let U ⊆ ∂Γ. There
is some infinite order h ∈ Γ such that {h+ , h− } is disjoint from {g + , g − } [Gro87, 8.2.G]
and h+ ∈ U . Therefore hi (g + ), hi (g − ) → h+ as i → ∞. Thus hi ω ∈ U for i sufficiently
large, proving the claim. If z ∈ ∂Γ, let (Ui ) be a countable neighborhood basis of z. By
the preceding claim, there is a sequence (hi ) in Γ such that hi ω ∈ Ui . Then (hi g + ) and
(φ(hi g + )) both converge to z, thus φ(z) = z. Since φ(z) = z for all z ∈ ∂Γ, φ = Id.
Thus Ψ is injective.
Corollary 4.4. Let Γ be a finitely generated one-ended hyperbolic group Γ, with nontriv-
ial Bowditch JSJ tree T . Then Γ is graphically discrete if and only if for every geometric
action of Γ on a connected locally finite graph X and for all vertices x ∈ X, the image
of the induced action of the stabilizer (Aut(X))x → Aut(T ) is finite.
ψ
Proof. We consider the composition Aut(X) → Homeo(∂X) −
→ Aut(T ). The result
follows from Theorem 4.1 and Proposition 4.3.
4.2. Hyperbolic groups with sphere and Sierpinski carpet boundaries.
Theorem 4.5. If Γ is a hyperbolic group with boundary an n-sphere for n ≤ 3, then Γ

is graphically discrete.
To prove this, we recall the Hilbert–Smith Conjecture:
Conjecture 4.6. If a locally compact group G acts faithfully and continuously by

homeomorphisms on a connected n-manifold, then G is a Lie group.
It is well-known that this conjecture reduces to showing p-adic integers cannot act
faithfully on a connected n-manifold for any prime p. Although the general conjecture
is open, it is known in the case n = 1, 2; see Pages 233 & 249 of [MZ55]. The n = 3 case
was shown by Pardon [Par13].
Theorem 4.7 ([MZ55, Par13]). The cases of the Hilbert–Smith Conjecture where n ≤ 3
are true.
We use Theorem 4.7 to prove Theorem 4.5. We note that the same proof will work
for any n in which a positive solution to the n-dimensional Hilbert–Smith Conjecture is
known.
Proof of Theorem 4.5. Let Γ be a hyperbolic group with boundary an n-sphere for some
n ≤ 3. Suppose Γ acts geometrically on a connected locally finite graph X. Let
ρ : Aut(X) → Homeo(S n ) be the induced action on the boundary of S n . Then ρ is
continuous with compact kernel K; see [Fur01, Theorem 3.5]. As Aut(X)/K acts faith-
fully and continuously on S n by homeomorphisms, it is a Lie group by Theorem 4.7.
Since Aut(X) is totally disconnected, Lemma 2.17 implies ker(ρ) is open. Thus Aut(X)
contains a compact open normal subgroup, hence Γ is graphically discrete.
Theorem 4.8. If Γ is a hyperbolic group with visual boundary homeomorphic to a

Sierpinski carpet, then Γ is graphically discrete.
Proof. Let Γ be a hyperbolic group with visual boundary homeomorphic to the Sierpinski
carpet S. Suppose Γ acts geometrically on a connected locally finite graph X. As
above, there is a continuous homomorphism ρ : Aut(X) → Homeo(S) with compact
kernel. Moreover, there is a continuous homomorphism φ : Homeo(S) → Homeo(S 2 )
with trivial kernel. Indeed, the map φ extends a homeomorphism of the Sierpinski carpet
to a homeomorphism of the 2-sphere by extending the map on the peripheral circles to
maps on disks; see [JSV13, Theorem 2.1]. As in the above proof of Theorem 4.5, this
implies Γ is graphically discrete.
4.3. Graphical discreteness of direct products of hyperbolic groups.
Theorem 4.9. The direct product of finitely many graphically discrete non-elementary
hyperbolic groups is graphically discrete.
Proof. Suppose Γ = Πni=1 Γi is product of graphically discrete non-elementary hyperbolic

groups and that ρ : Γ → Isom(X) is a geometric action on a locally finite graph X.
Replacing Γ with its quotient by the finite normal subgroup ker(ρ), we can assume ρ
is injective and hence Γ can be identified with a subgroup of G = Isom(X). As in
Section 3.5, there is a quasi-isometry h : X → Γ quasi-conjugating the action of G on X
to the induced quasi-action {fg }g∈G of G on Γ. We proceed with an argument similar
to the proof of [Mar22a, Theorem 6.1].
There is a finite-index subgroup G∗ ≤ G, of index at most n!, such that all elements
of the restricted quasi-action {fg }g∈G∗ split up to uniform error as a product of n quasi-
isometries by [KKL98], see also [Mar22a, §2.3]. Note also ρ(Γ) ≤ G∗ . In particular,
there are quasi-actions of G∗ on each of Γ1 , . . . , Γn . Restricting to ρ(Γi ) ∼
= Γi we recover
the natural map Γi → QI(Γi ) induced by left multiplication, which does not fix a point
of ∂Γi . Thus the quasi-action of G∗ on Γi does not fix a point of ∂Γi either.
There is a quasi-isometry fi : Γi → Yi that quasi-conjugates the quasi-action of G∗

on Γi to an isometric action of G∗ on Yi , where Yi is either a rank one symmetric space
or a locally finite graph by [Mar22a, Corollary 5.7]. We now consider the composition
h i π fi
hi : X −
→ Γ −→ Γi −
→ Yi ,
where πi is the projection map. The map hi is a coarse surjective coarse-Lipschitz map
that is coarsely G-equivariant. We can thus apply Lemma 2.32 to deduce that the
composition ψi : G∗ → Isom(Yi )/Ki is continuous, where Ki C Isom(Yi ) is the maximal
compact normal subgroup. Let Hi ≤ Isom(Yi )/Ki be the closure of the image of ψi . We
claim that Hi is discrete. Indeed, in the case where Yi is a symmetric space, this follows
from Lemma 2.17, since Hi is a Lie group and G∗ is totally disconnected. Now suppose
Yi is a locally finite graph. Since fi quasi-conjugates the geometric action of Γi on itself
to a geometric action of Γi on Yi , we deduce Γi is a virtual uniform lattice in Isom(Yi ).
As Γi is graphically discrete and Isom(Yi ) is totally disconnected with Ki the maximal
compact normal subgroup, it follows Hi ≤ Isom(Yi )/Ki is discrete.
Let f = (f1 , . . . , fn ) : Γ = Πni=1 Γi → Πni=1 Yi be the product map. Then f ◦ h : X →
Πi=1 Yi quasi-conjugates the geometric action of G∗ on X to the isometric action of G∗
n
on Πni=1 Yi . It follows that the product map ψ = (ψ1 , . . . , ψn ) : G∗ → Πni=1 Isom(Yi )/Ki ,
which we already know is continuous with discrete image, also has compact kernel.
Therefore, G∗ is compact-by-discrete as required.
5. Graphical discreteness is not a commensurability invariant

Definition 5.1. A manifold is a minimal element in its commensurability class if it
does not nontrivially cover any orbifold.
We assume throughout the paper that 3-manifolds are connected.
Remark 5.2. (Existence of manifold minimal elements.) We thank Genevieve Walsh

for explaining the following examples. First, let N be a hyperbolic knot complement,
and let Γ = π1 (N ). Margulis [Mar91] proved that if Γ is non-arithmetic, then Γ has finite
index in its commensurator, CommPSL(2,C) (Γ) ≤ PSL(2, C). So, H3 / CommPSL(2,C) (Γ)
is the minimal orbifold in the commensurability class of N in this case. Showing that
Γ = CommPSL(2,C) (Γ) is equivalent to showing Γ has no symmetries (i.e. Γ is equal to its
normalizer) and no hidden symmetries (i.e. its normalizer is equal to its commensura-
tor). The figure-eight knot complement is the only arithmetic knot complement [Rei91].
For hyperbolic knots with up to 15 crossings, the only examples with hidden symme-
tries are the figure-eight knot and two dodecahedral knots [GHH08, CDH+ ]. See the
survey [Wal11] for more details. For example, the knots K1 = 9 32 and K2 = 9 33
with nine crossings in Rolfson’s Knot Table are hyperbolic knots with no symmetries,
as verified on KnotInfo. Thus, N = S 3 − Ki is a minimal knot complement in its
commensurability class.
Sufficiently large Dehn filling on these knot complements yields closed hyperbolic man-
ifolds that are minimal elements in their commensurability class, as shown in [CDH+ ,
Proposition 2.4].
The next lemma illustrates the advantage of working with manifolds rather than
orbifolds.
Lemma 5.3. Let M be a closed hyperbolic 3-manifold that is a minimal element in its
commensurability class. Let x ∈ H3 , and let Aut(H3 , π1 (M ) · x) denote the subgroup
of Isom(H3 ) that stabilizes the set π1 (M ) · x. Then Aut(H3 , π1 (M ) · x) = π1 (M ). In
particular, if g ∈ Aut(H3 , π1 (M ) · x) fixes x, then g is trivial.
Proof. The group Aut(H3 , π1 (M ) · x) acts geometrically on H3 and contains π1 (M ).
Hence, these groups are commensurable. Since M is a minimal element in its commen-
surability class, we deduce that Aut(H3 , π1 (M ) · x) = π1 (M ). The final sentence follows
since M is a manifold: the only element in π1 (M ) fixing x is the identity.
Theorem 5.4. Let M and M 0 be closed hyperbolic 3-manifolds that are minimal ele-
ments in their commensurability classes. Then Γ = π1 (M ) ∗ π1 (M 0 ) has no nontrivial
lattice embeddings, but has a finite-index subgroup that does have nontrivial lattice em-
beddings. In particular, Γ is graphically discrete but contains a finite-index subgroup that
is not.
Proof. Since π1 (M ) and π1 (M 0 ) are residually finite, they contain proper finite-index
subgroups. By Corollary 3.20, Γ contains a finite-index subgroup that is not graphically
discrete, so it suffices to show Γ has no nontrivial lattice embeddings. As Γ is hyperbolic
and not virtually free, it is not virtually isomorphic to a uniform lattice in a connected
rank-one semisimple Lie group. Therefore, Proposition 2.16 ensures that all lattice
embeddings of Γ are uniform. Suppose Γ acts geometrically on a proper quasi-geodesic
metric space X. We will prove Isom(X) is compact-by-discrete.
By [SW, Proposition 4.5], the Isom(X)-action on X is quasi-conjugate to an Isom(X)-
action on an ideal model geometry Z, which is a tree of spaces with vertex spaces either
a point or a copy of H3 and edge spaces that are points. Let ρ : Isom(X) → Isom(Z)
be the action on Z, and let f : X → Z be this quasi-conjugacy. Since Z is tame
ρ
(see Example 2.30), Lemma 2.32 implies the composition Φ : Isom(X) − → Isom(Z) →
Isom(Z)/K is continuous, where K is the maximal compact normal subgroup. (In fact,
K is trivial, although we do not need this.) Since f : X → Z is a quasi-conjugacy and
K is compact, ker(Φ) is a closed subgroup of Isom(X) with bounded orbits, hence is
compact by Proposition 2.27.
We now show Isom(Z) is discrete, which implies that Isom(X) is compact-by-discrete.
As Γ acts geometrically on Z, there are two orbits of one-ended vertex spaces, and by
Mostow rigidity, the quotient of one-ended vertex space by the corresponding vertex
stabilizer is isometric to either M orM 0 . After rescaling the length of the edge, Γ\Z is
isometric to Y = M t M 0 t [−1, 1] / ∼, where ∼ identifies −1 to some x ∈ M and 1
to some x0 ∈ M 0 . Let x e be a lift of x ∈ Y in a copy M f. If g ∈ Isom(Z) fixes x
f0 of M e,
then g fixes M0 by Lemma 5.3. Hence, g fixes all intervals attached to lifts of x in M0 .
f f
Continuing inductively, we see that g is the identity. Thus, Isom(Z) is discrete.
6. Simple surface amalgams

In this section we prove Theorem 6.5, characterizing graphically discrete simple sur-
face amalgams.
Notation 6.1. Let Γ = π1 (Z) be the fundamental group of a simple surface amalgam,
where Z is the union of surfaces Σ1 , . . . , Σk which each have a single boundary compo-
nent and negative Euler characteristic. The decomposition of Z into surfaces yields a
decomposition as a graph of spaces, where the underlying graph Λ consists of vertices

V Λ = {u0 , u1 , . . . , uk } and edges EΛ = {ei , ēi | 1 ≤ i ≤ k and ei = (ui , u0 )}. The vertex
spaces are Zu0 ∼ = S 1 and Zui ∼ = Σi for 1 ≤ i ≤ k. Each edge space is S 1 .
This decomposition gives the JSJ decomposition of Γ in the sense of [Bow98]. Let
T denote the JSJ tree, and let ρ : T → Λ denote the map given by quotienting by Γ.
There are two types of vertices in T :
(1) infinite-valence vertices, denoted V1 T , that correspond to the hanging Fuchsian
vertex groups;
(2) valence-k vertices, denoted V2 T , that correspond to the cosets of the cyclic sub-
group along which the surface groups are amalgamated.
So, ρ(V2 T ) = {u0 } and ρ(V1 T ) = {u1 , ..., uk }. Note that for all w ∈ V T t ET , the
stabilizer Γw is a quasi-convex subgroup of Γ ([Bow98, Proposition 1.2]), so ∂Γw ⊆ ∂Γ.
Proposition 6.2. If χ(Σi ) = χ(Σj ) for some i 6= j, then Γ is not graphically discrete
and hence has a nontrivial lattice embedding.
Proof. Suppose without loss of generality that χ(Σ1 ) = χ(Σ2 ). We can give the space
Z the structure of a simplicial complex such that Σ1 and Σ2 are isomorphic and there
is an automorphism of Z flipping Σ1 and Σ2 and fixing the rest of the complex. For
each v ∈ V2 T , this automorphism lifts to an elliptic automorphism of the universal cover
hv : Ze → Ze that fixes the vertex space Zev and induces the transposition on lk(v) that
fixes all edges except ee1 , ee2 ∈ lk(v) where ρ(e
ei ) = ei . The existence of these flipping
(0)
ev for some v ∈ V2 T and G = Aut Ze(1) , then the

automorphisms implies that if x ∈ Z
image of the stabilizer Gx in Aut(T ) is infinite. Hence, by Corollary 4.4, the group Γ is
Proposition 6.3. If χ(Σi ) 6= χ(Σj ) for all i 6= j, then Γ has no nontrivial lattice
embeddings, hence is graphically discrete.
In the proof of Proposition 6.3, we make use of the following elementary lemma:
Lemma 6.4. If G is a group and A, B, C are subgroups with B ≤ C, then:
(†) [AC : AB] = [C : (A ∩ C)B]
provided the products in (†) are all subgroups.
C AC
Proof. We consider the function φ : (A∩C)B → AB given by c(A ∩ C)B 7→ cAB. Since
B ≤ C, C ∩ AB = (A ∩ C)B, which shows φ is well-defined and injective. Surjectivity
of φ follows from the fact that as AC is a subgroup, AC = CA.
Proof of Proposition 6.3. As Γ is hyperbolic and not virtually free, it is not virtually
isomorphic to a non-uniform lattice in a connected simple rank one Lie group. Propo-
sition 2.16 thus implies every lattice embedding of Γ is uniform. As Γ is torsion-free,
every virtual lattice embedding is injective. So, suppose ρ : Γ → G is a uniform lattice
embedding into a locally compact group G. By Remark 2.15, we can assume that G is
second countable. We identify Γ with its image under ρ.
Consider the induced quasi-action of G on Γ, which gives a continuous homomor-
phism φ : G → Homeo(∂Γ) with compact kernel by [Fur01, Theorem 3.5]. If T is the
Bowditch JSJ tree of Γ, there is a continuous injective map Ψ : Homeo(∂Γ) → Aut(T )
by Proposition 4.3. We note that the action of G on T extends the natural action of Γ
on T , and that G is compact-by-(totally disconnected). As G acts continuously on T ,

for each w ∈ V T t ET , the stabilizer Gw is open. Therefore, Lemma 2.9 implies each Γw
is a uniform lattice in Gw . In particular, if w has infinite valence, Gw acts continuously
on ∂Γw with compact kernel [Fur01, Theorem 3.5]. For each edge e ∈ ET , as Γe ∼ = Z is
graphically discrete, the group Ge contains a maximal compact open subgroup Ke C Ge
by Remark 3.7. We will show Ke = Kf for all e, f ∈ ET , which proves Ke is a compact
open normal subgroup of G as required (see Remark 2.11).
Firstly, suppose v ∈ V1 T and e ∈ lk(v). Recall ∂Γv ⊆ ∂Γ is a cyclically ordered Cantor
set and that Gv preserves this cyclic ordering by [Bow98, Lemma 3.2]. Moreover, the
pair of points in ∂Γe are adjacent in this cyclic ordering. As Ke is compact and every
edge stabilizer is open, for any finite R ⊆ lk(v) containing e, the orbit Ke R ⊆ lk(v)
consists of finitely many edges. Since Ke preserves the cyclic order ∪f ∈Ke R ∂Γf and Ke
fixes the adjacent pair of points ∂Γe pointwise, Ke fixes ∪f ∈Ke R ∂Γf pointwise. Thus,
Ke fixes ∪f ∈lk(v) ∂Γf . Since ∪f ∈lk(v) ∂Γf is dense in ∂Γv , the group Ke fixes ∂Γv . Letting
Kv be the kernel of the action of Gv on ∂Γv , we see that Ke ≤ Kv for all e ∈ lk(v).
Since Kv ≤ Ge , we have Ke = Kv by maximality of Ke . In particular, Ke = Kf for all
e, f ∈ lk(v).
Now suppose v ∈ V2 T . We claim every g ∈ Gv fixes lk(v). This will imply Gv = Ge
for all e ∈ lk(v), and hence Ke = Kf for all e, f ∈ lk(v). Combined with the previous
paragraph, it will then follow that Ke = Kf for all e, f ∈ ET as required. Suppose for
contradiction there is some g ∈ Gv and e ∈ lk(v) with ge 6= e. Let ι(e) = w ∈ V1 T be the
hanging Fuchsian vertex incident to e. We will show that χ(Γw ) = χ(Γgw ), contradicting
our hypotheses.
As noted above, Kw = Ke is a compact open normal subgroup of Gw . Since Γw and
−1
g Γgw g are torsion-free uniform lattices in Gw , they inject to finite-index subgroups of
the discrete quotient Gw /Kw . As Euler characteristic is multiplicative, in order to show
χ(Γw ) and χ(g −1 Γgw g) = χ(Γgw ) are equal, it is sufficient to show [Gw : Γw Kw ] = [Gw :
(g −1 Γgw g)Kw ]. We emphasize that as g need not normalize Γ, the subgroups Γw and
g −1 Γgw g are not typically equal.
We first note that Γv = Γe = Γge ∼ = Z. Define Kv = ∩f ∈lk(v) Kf , which is compact
and normal in Gv . Since Ke is compact and Γe ∼ = Z is discrete and torsion-free, the
intersection Γe ∩ Ke is trivial. Lemma 6.4 now implies [Γe Ke : Γe Kv ] = [Ke : Kv ].
Therefore,
[Gv : Ge ][Ge : Γe Ke ][Ke : Kv ] = [Gv : Ge ][Ge : Γe Ke ][Γe Ke : Γe Kv ]
= [Gv : Γe Kv ] = [Gv : Γv Kv ].
Running the same argument with ge instead of e, we get
[Gv : Gge ][Gge : Γge Kge ][Kge : Kv ] = [Gv : Γv Kv ].
Conjugating by g, we have [Gv : Ge ] = [Gv : Gge ] and [Ke : Kv ] = [Kge : Kv ], so we
deduce that [Ge : Γe Ke ] = [Gge : Γge Kge ].
As Γw acts transitively on edges in lk(w), we have Gw = Γw Ge . Similarly, Ggw =
Γgw Gge . Therefore, applying Lemma 6.4 twice, we have
[Gw : Γw Kw ] = [Γw Ge : Γw Ke ] = [Ge : (Γw ∩ Ge )Ke ] = [Ge : Γe Ke ] = [Gge : Γge Kge ]
= [Gge : (Γgw ∩ Gge )Kge ] = [Γgw Gge : Γgw Kge ] = [Ggw : Γgw Kgw ].
It follows that [Gw : Γw Kw ] = [Gw : g −1 Γgw gKw ] as required.
Combining Propositions 6.2 and 6.3 gives:
Theorem 6.5. If Γ is the fundamental group of a simple surface amalgam, the following
are equivalent:
(1) Γ is graphically discrete;
(2) Γ has no nontrivial lattice embeddings;
(3) the surfaces in the amalgam have pairwise-distinct Euler characteristics.
Two simple surfaces amalgams are quasi-isometric if and only if the number of surfaces
in the unions are equal [Mal10]. Thus, we have the following corollary.
Corollary 6.6. Graphical discreteness is not a quasi-isometry invariant for one-ended

groups.
7. Graphical discreteness of closed 3-manifold groups

7.1. Geometric 3-manifolds. In this section we show the following:
Theorem 7.1. The fundamental group of a closed geometric 3-manifold is graphically

discrete.
Remark 7.2. Note that Z × F2 is simultaneously a uniform lattice in Z × Aut(T )

and a non-uniform lattice in R × H2 . Thus, it is the fundamental group of a compact
geometric 3-manifold with non-empty toroidal boundary that is not graphically discrete.
In particular, Theorem 7.1 does not generalize to fundamental groups of geometric 3-
manifolds with boundary.
Proposition 7.3. If Γ is a lattice in the Lie group Sol, then Γ is graphically discrete.
Proof. By [Rag72, Theorems 3.1 & 4.28] lattices in Sol are always uniform and torsion-
free. Suppose G is a totally disconnected locally compact group containing Γ as a lattice.
It is sufficient to show G is compact-by-discrete. By [Dym15, Theorem 1], there is a
continuous map ρ : G → Isom(Sol) with compact kernel K. It follows from Lemma 2.17
that K is open, hence G is compact-by-discrete.
We will now show a finitely generated group quasi-isometric to H2 × R is graphically

discrete. We need the following theorem concerning quasi-actions on H2 × R:
Proposition 7.4 ([KL01, §3,4,6]). If there is a cobounded quasi-action of a group G on

H2 × R, then there is an isometric action ρ : G → Isom(H2 ) such that
sup (π(gx), ρ(g)π(x)) < ∞,

x∈H2 ×R,g∈G
where π : H2 × R → H2 is the projection. Moreover, when G is finitely generated and

the quasi-action on H2 × R is proper, then ker(ρ) is two-ended.
Proposition 7.5. A finitely generated group quasi-isometric to H2 × R is graphically

discrete.
Proof. Suppose Γ is a finitely generated group quasi-isometric to H2 × R and Γ acts

geometrically on a connected locally finite graph X. We prove that G = Isom(X) is
compact-by-discrete.
The Milnor–Schwarz lemma implies there is a quasi-isometry f : X → H2 × R that
quasi-conjugates the isometric action of G on X to a cobounded quasi-action of G on
H2 × R. Proposition 7.4 implies there is an isometric action ρ : G → Isom(H2 ) such that
the coarse Lipschitz map π ◦ f is G-equivariant up to uniformly bounded error. Since
H2 has no nontrivial bounded isometries, the map ρ is continuous by Corollary 2.33.
By Lemma 2.17, ker(ρ) is open. Since Γ is a lattice in G, Lemma 2.9 implies Γ∩ker(ρ)
is a lattice in ker(ρ). As the quasi-action G y H2 × R restricts to a proper cobounded
quasi-action Γ y H2 × R, Proposition 7.4 implies Γ ∩ ker(ρ) is two-ended. Since ker(ρ) is
totally disconnected and contains a uniform lattice that is a two-ended subgroup, ker(ρ)
has a unique maximal compact open normal subgroup K with quotient infinite cyclic
or infinite dihedral [Cor18, Corollary 19.39]. Since K is a topologically characteristic
subgroup of ker(ρ), it is a normal compact open subgroup of G. Hence G is compact-
by-discrete.
Remark 7.6. The same argument holds without much modification for all the groups
in [KL01].
Since H2 × R and SL
^2 (R) are quasi-isometric [Ger92], Proposition 7.5 implies:
Corollary 7.7. A cocompact lattice in either H2 × R or SL

^2 (R) is graphically discrete.
Proof of Theorem 7.1. Recall (see [Thu97, Figure 4.22]) that the fundamental group of
a closed geometric three-manifold is either:
(1) virtually nilpotent (S 3 , S 2 × R, R3 or Nil),
(2) a cocompact lattice in Sol,
(3) a cocompact lattice in H2 × R or SL ^ 2 (R), or
(4) a cocompact lattice in H3 .
The virtually nilpotent and hyperbolic fundamental groups are graphically discrete by
Theorem 3.16. The remaining groups are graphically discrete by Proposition 7.3 and
Corollary 7.7.
7.2. Non-geometric 3-manifolds. In this section we use a result of Kapovich–Leeb
to classify lattice embeddings of non-geometric 3-manifolds [KL97].
Let M be a closed irreducible, oriented 3-manifold with a nontrivial geometric de-
composition. As a consequence of geometrization (see Theorem 1.14 in [AFW15]) the
JSJ tori give a decomposition of M into Seifert fibered and hyperbolic components. By
taking tubular neighborhoods of the JSJ tori we obtain a graph of spaces decomposition
f : M → Y where Y is a finite graph. The vertex spaces Mv = f −1 (v) are the Seifert
fibered or hyperbolic components, and the edge spaces Me = f −1 (me ), where me is
the midpoint of an edge, are homeomorphic to T2 . Let T denote the associated JSJ
tree, which is the Bass–Serre tree associated to the graph of spaces decomposition. The
universal cover M̃ is a graph of spaces over T with vertex and edge spaces denoted M̃v
and M̃e .
Let Γ = π1 (M ). We first note that Γ has no non-uniform lattice embeddings. Indeed,
Γ is acylindrically hyperbolic [MO15, Corollary 2.9]. Since Γ is non-geometric, it is
not isomorphic to a non-uniform lattice in a rank one simple Lie group, hence Proposi-
tion 2.16 implies it has no non-uniform lattice embeddings. Now suppose ρ : Γ → G is a
uniform lattice embedding. By Remark 2.15, we assume without loss of generality that
G is second countable. We identify Γ with its image in G. The uniform lattice embed-
ding ρ induces a quasi-action of G on Γ as in Section 3.5. Since Γ acts geometrically on
M̃ , this induces a quasi-action {fg }g∈G of G on M̃ .
Lemma 7.8. The quasi-action {fg }g∈G of G on M̃ induces a continuous action of G
on T such that for all g ∈ G, the Hausdorff distance between fg (M̃v ) and M̃gv is finite
for all v ∈ V T . Moreover, G is compact-by-(totally disconnected).
Proof. The main theorem of [KL97] proves that the quasi-action {fg }g∈G of G on M̃
induces an action σ : G → Aut(T ) such that fg (M̃v ) and M̃σ(g)(v) are at uniform finite
Hausdorff distance for all v ∈ V T and g ∈ G. To see this action is continuous, let (gi ) be
a sequence in G converging to the identity. By Lemma 3.24, there is a constant B such
that for each x ∈ M̃ , d(fgi (x), x) ≤ B for all i sufficiently large. Since no two distinct
vertex spaces of M̃ are at finite Hausdorff distance, for each v ∈ V T , the element σ(gi )
fixes v for i sufficiently large; hence, the action is continuous.
Since the action of Γ on T is acylindrical [WZ10, Lemma 2.4], there exist edges e and
0
e such that Γe ∩ Γe0 = Γ ∩ (Ge ∩ Ge0 ) is finite. Since the open subgroup Ge ∩ Ge0 contains
a finite group as a uniform lattice by Lemma 2.9, it is compact. Thus, G contains a
compact open subgroup, hence is compact-by-(totally disconnected) by Lemma 2.26.
We now show fundamental groups of closed irreducible non-geometric 3-manifolds

have no nontrivial lattice embeddings. The proof is motivated by the observation that
if M is such a manifold, then the isometry group of the universal cover of M is discrete;
in particular, if an isometry fixes a copy of the universal cover of a JSJ torus, then it
fixes the entire space. The main difficulty in the proof is that if G is a locally compact
group containing π1 (M ) as a uniform lattice, then it need not be the case that G acts
on M̃ .
Theorem 7.9. Let M be a closed irreducible, oriented 3-manifold with a nontrivial
geometric decomposition. Let G be a locally compact group containing π1 (M ) as a lattice.
Then G contains a compact open normal subgroup K such that G/K is isomorphic to
the fundamental group of a compact 3-orbifold covered by M . In particular, π1 (M ) has
no nontrivial lattice embeddings, hence is graphically discrete.
Proof. We begin by outlining the proof. Let M be as in the statement of the theorem,
and let G be a locally compact group containing Γ := π1 (M ) as a lattice. As noted
above, Γ has no non-uniform lattice embeddings, so G contains Γ as a uniform lattice.
We will show G has a compact open normal subgroup K and then apply arguments
analogous to those of Kapovich–Leeb [KL97, §5.2]. To exhibit such a K, we will use
an action of G on the JSJ tree, and prove each vertex and edge stabilizer Gv and Ge
of G has a maximal compact open normal subgroup, Kv and Ke , respectively, so that
Kv = Ke if e is incident to v. Taking K = Kv will then give the desired subgroup.
The group G is compact-by-(totally disconnected) and acts continuously on the JSJ
tree T by Lemma 7.8. Each vertex stabilizer Gv is open and compact-by-(totally dis-
connected). Since Γ is a uniform lattice in G and Gv is open, Γv := Γ ∩ Gv is a uniform
lattice in Gv by Lemma 2.9.
First suppose v is a vertex corresponding to a hyperbolic component. Then, Γv

is graphically discrete by Theorem 3.16, so Gv is compact-by-discrete. The discrete
quotient Qv of Gv contains Γv as a uniform lattice; hence, Qv is quasi-isometric to Γv .
By [Sch95], Qv acts on H3 by isometries with image a non-uniform lattice and with finite
kernel. Without loss of generality, replace Qv by its quotient by this finite kernel. Then,
let Kv be the kernel of the homomorphism Gv → Qv . Since Qv is discrete without finite
normal subgroups, Kv is the maximal compact normal open subgroup of Gv .
We will now show if e is an edge incident to v, then Kv is the unique maximal compact
normal subgroup of Ge ≤ Gv . The discrete quotient Qv of Gv acts geometrically on
truncated hyperbolic space Ωv ⊆ H3 . So, Gv acts on Ωv with kernel Kv . There is a
one-to-one correspondence between the stabilizers of the set of horospheres of Ωv under
the Gv -action and the set of edge stabilizers Ge ≤ G with e incident to v. Thus, if
e is incident to v, then the edge group Ge acts geometrically on a horosphere. Any
isometry of Ωv fixing this horosphere fixes Ωv , hence Kv is precisely the kernel of the
action of Ge on this horosphere. By [CdlH16, Example 2.E.23.5], Kv is contained in
a unique maximal compact normal subgroup of Ge (the compact radical of Ge ). Since
Ge /Kv is a finitely generated group acting geometrically and faithfully on E2 , it has
no nontrivial finite normal subgroups. Therefore, Kv is the unique maximal compact
normal subgroup of Ge .
Now suppose v is a vertex corresponding to a Seifert fibered component, and let
M̃v = Σv × R be a Seifert fibered vertex space, where Σv is a convex subset of H2 whose
boundary is a union of disjoint geodesics. Since Γv acts geometrically on M̃v and is a
uniform lattice in Gv , we deduce that Gv admits a cobounded quasi-action on M̃v .
We now summarize an argument identical to that given in [KL97, §5.2], which we
refer to for details. By reflecting along boundary flats, we can extend the cobounded
action of Gv on Σv × R to a quasi-action of Gv on H2 × R. This quasi-action preserves
fibers of the form {x} × R and so descends to a quasi-action of Gv on H2 . Since this
quasi-action preserves a discrete line pattern, it can be quasi-conjugated to an isometric
discrete cocompact action on H2 . Thus, there is a map φv : Gv → Isom(H2 ) whose
image is a discrete non-uniform lattice Qv in H2 and whose kernel is an open subgroup
Rv . Thus Qv is the fundamental group of a 2-orbifold Sv . The group Rv ∩ Γv is the
kernel of the restriction φv |Γv , which is a finitely generated two-ended group. Since Z is
graphically discrete and Rv ∩ Γv is a lattice in Rv , we deduce there is a compact open
normal subgroup Kv C Rv with quotient Lv isomorphic to Z or D∞ .
Since neither Qv nor Lv has finite normal subgroups, Kv is a maximal compact normal
open subgroup of Gv . Therefore Gv /Kv fits into the short exact sequence
1 → Lv → Gv /Kv → Qv = π1 (Sv ) → 1
and so Gv /Kv is the fundamental group of a Seifert fibered 3-orbifold Ov . Hence Gv acts
on Oev with kernel Kv . If e is an edge incident to v, then Ge is precisely the stabilizer
of the corresponding boundary component of O ev . The kernel of the action of Ge on the
corresponding boundary flat is precisely Kv , since any isometry of Oev fixing a boundary
flat is trivial. Therefore Kv is the unique maximal compact normal subgroup of Ge as
in the hyperbolic case.
We have shown that a maximal compact open normal subgroup of each vertex group
Gv is equal to the maximal compact open normal subgroup of each incident edge group.
Thus Kv = Kv0 for all vertices, and K = Kv is a compact open normal subgroup of G. By
the exact same reasoning as in [KL97, §5.3] (or alternatively, applying [KL97, Theorem
1.2] to G/K), we see G/K is the fundamental group of a compact non-geometric 3-
ρ
dimensional orbifold O. The image of Γ under the composition Γ − → G → G/K ∼ = π1 (O)
is a finite-index subgroup, hence π1 (M ) is isomorphic to a finite-index subgroup of π1 (O).
Therefore M is homeomorphic to a finite cover of O by Theorem 2.1.2 of [AFW15].
8. Blowups of groups acting on trees

This section provides a general construction to “blowup” the action of a locally com-
pact group on a tree to a geometric action on a locally finite graph with a tree of spaces
decomposition. The results will be utilized in this paper to show that if two finitely
generated groups act on the same proper metric space that admits a tree of spaces
decomposition with certain properties, then the groups act geometrically on the same
simplicial complex admitting a tree of spaces decomposition in a particularly nice way
(see Theorem 8.5).
Definition 8.1. Let G be a locally compact group acting continuously and minimally
on a tree T . A blowup of G y T is a surjective map p : X → T such that
(1) X is a locally finite connected graph, and G acts geometrically on X;
(2) p is G-equivariant and simplicial with respect to the first barycentric subdivision
of T ;
(3) for all v ∈ V T , Xv := p−1 (v) is a connected subgraph of X which we call a vertex
space;
(4) for all e ∈ ET , if me is the midpoint of e, then Xe := p−1 (me ) is a connected
subgraph of X which we call an edge space.
The following proposition can be thought of an interweaving of the construction of

the Cayley–Abels graph of a tdlc group, and the construction of the Bass–Serre tree
from a graph of groups decomposition.
Proposition 8.2 (Blowup existence). Let G be a locally compact group acting contin-
uously, minimally and cocompactly on a tree T . Suppose all vertex and edge stabilizers
of T are compactly generated and G contains a compact open subgroup. Then a blowup
p : X → T of G y T exists.
Proof. Let Ω0 ⊆ V T t ET be a set of representatives of the G-orbits of vertices and
edges of T . Let K be a compact open subgroup of G. We first show we can replace
K by a compact open subgroup of G contained in Gw for all w ∈ Ω0 . For every
w ∈ V T t ET , let Kw := K ∩ Gw . We claim every Kw is a finite-index subgroup of K.
Indeed, since K is compact, K acts elliptically on T . Hence, there is a vertex w0 ∈ V T
with K ≤ Gw0 . Since the action of G on T is minimal and cocompact, there exists a
g ∈ G such that w lies on the path P = [w0 , gw0 ]. Since K ∩ gKg −1 fixes P pointwise,
K ∩ gKg −1 ≤ Kw ≤ K. As K is a compact open subgroup of G, it is commensurated.
Since K ∩ gKg −1 is a finite-index subgroup of K, it follows Kw is also a finite-index
subgroup of K. Let K0 := ∩w∈Ω0 Kw . As Ω0 is finite, K0 is a compact open subgroup of
G. By replacing K with K0 if necessary, we can thus assume K ≤ Gw for every w ∈ Ω0 .
We also make the following additional choices on which the construction depends. For
each w ∈ Ω0 , the group Gw is compactly generated, so there exists a finite symmetric
set Sw ⊆ Gw such that Sw ∪ Kw generates Gw . Let Ω1 be the closure of Ω0 , i.e. Ω1

is the smallest subgraph of T (not necessarily connected) containing Ω0 . Pick a finite
symmetric set F ⊆ G such that Ω1 ⊆ F Ω0 .
We define a simplicial graph X as follows. The vertex set of X is defined to be
V X := G/K × Ω0 ,
where G/K is the collection of left cosets of K in G. Two vertices (g1 K, w1 ) and
(g2 K, w2 ) are joined by an edge in X if either of the following hold:
(I) w1 = w2 and g1−1 g2 ∈ KSw1 K;
(II) g1−1 g2 ∈ KF K and g1 w1 is an edge incident to the vertex g2 w2 or vice versa.
We call these type I and type II edges, respectively. Note that they are mutually
exclusive: an edge is a type II edge if and only if it is not a type I edge.
We define a map p : X → T that is simplicial with respect to B(T ), the first barycen-
tric subdivision of T , as follows:
• if v ∈ V T ∩ Ω0 , then (gK, v) 7→ gv;
• if e ∈ ET ∩ Ω0 , then (gK, e) 7→ mge where mge is the midpoint of the edge ge.
The map is well-defined because Kw = w for all w ∈ Ω0 . The map is simplicial since
type I edges of X are sent to vertices of B(T ), and type II edges of X are sent to edges
of B(T ).
We claim that Xy := p−1 (y) is a connected subgraph for all vertices y of B(T ). Indeed,
if (g1 K, w1 ) and (g2 K, w2 ) are two vertices in Xy , then w1 = w2 and g1−1 g2 ∈ Gw1 . Since
K ∪ Sw1 generates Gw1 , there exist s1 , s2 , . . . , sn ∈ Sw1 and k1 , . . . , kn ∈ K such that
s1 k1 s2 . . . sn kn = g1−1 g2 , and so
((g1 K, w1 ), (g1 s1 K, w1 )) , ((g1 s1 k1 K, w1 ), (g1 s1 k1 s2 K, w1 )) , . . .
is an edge path in Xy from (g1 K, w1 ) to (g2 K, w2 ) = (g2 K, w1 ) consisting only of type

I edges.
To show X is connected, note that since T is connected and each Xy is connected, it
is sufficient to show that if v ∈ V T is incident to e ∈ ET , then there is an edge in X
from Xv to Xe . Indeed, there exists g ∈ G such that g −1 e ∈ Ω0 and so g −1 v ∈ Ω1 = Ω0 .
Thus there exists f ∈ F such that f −1 g −1 v ∈ Ω0 . It follows that (gK, g −1 e) ∈ Xe and
(gf K, f −1 g −1 v) ∈ Xv are joined by a type II edge in X. Thus X is connected.
S
We now show X is locally finite. Consider the finite set J := w∈Ω0 Sw ∪F . Since K is
a commensurated subgroup of G, there is a finite subset J0 ⊆ G such that KJK ⊆ J0 K.
Therefore, every vertex adjacent to some (g1 K, w1 ) ∈ V X is of the form (g1 jK, w2 ) for
some j ∈ J0 and w2 ∈ Ω0 . Since there are only finitely many such vertices, X is locally
finite.
Define an action of G on X by g · (g1 K, w1 ) = (gg1 K, w1 ). It is clear that (g1 K, w1 )
and (g2 K, w2 ) are joined by an edge if and only if (gg1 K, w1 ) and (gg2 K, w2 ) are, and
that p is G-equivariant. The graph X is locally finite and G has |Ω0 | < ∞ orbits of
vertices in X, so G acts cocompactly on X. Since every vertex stabilizer is open, the
action of G on X is continuous. Moreover, the action of G on X is proper since X is
locally finite and vertex stabilizers are compact. We have thus shown that p : X → T is
a blowup.
The following lemmas allow us to upgrade the blowup construction to build a model
geometry with additional desirable properties needed to apply Theorem 9.7.
Lemma 8.3. Let G be a compactly presented group acting geometrically on a locally

finite connected graph X. For M ≥ 0, let XM be the 2-complex obtained by attaching
2-cells to every edge loop of length at most M in X. There exists a constant M0 such
that XM is simply-connected for all M ≥ M0 .
Proof. Equip the vertex set X (0) with a metric such that d(x, y) is the length of the
shortest edge path in X joining x and y. The metric space X (0) is quasi-isometric to
G so is coarsely simply connected ; see e.g. [CdlH16, §6,8]. Thus, there is a constant
(2)
N ≥ 1 such that the 2-skeleton of the Rips complex Y = PN (X (0) ) is simply connected
[CdlH16, Proposition 6.C.4]. Since N ≥ 1, there is a natural injection φ : X → Y .
Setting M = 3N yields a continuous map ψ : Y → XM sending a 1-simplex [x, y] ∈ Y
(1)
to a geodesic path in XM joining x and y, and sending a 2-simplex [x, y, z] to a 2-cell
in XM attached to edge path [x, y] ∪ [y, z] ∪ [z, x]. This can be done so that ψ ◦ φ is
(1)
the identity map on X = XM . To show XM is simply-connected, it is enough to show
any loop in X can be filled by a disc in XM . For any loop f : S 1 → X, there is a map
F : D2 → Y such that F |S 1 = φ ◦ f . Thus, (ψ ◦ F )|S 1 = f as required.
The following lemma will be essential for obtaining an imprimitivity system on a tree
of spaces given imprimitivity systems on vertex spaces. A continuous G-tree is a tree
admitting a continuous G-action. A finite refinement of a continuous G-tree T is a
continuous G-tree T 0 and an equivariant map p : T 0 → T obtained by collapsing finite
subtrees of T 0 to points.
Lemma 8.4 (Tree refinement lemma). Let G be a locally compact group acting contin-
uously on a tree T without edge inversions. Assume that for every vertex v ∈ V T , the
stabilizer Gv contains a compact open normal subgroup. Then there is a finite refinement
p : T 0 → T such that for each v ∈ V T 0 ,
• Gv contains a finite-index open subgroup that is the stabilizer of a vertex or edge
of T , and
• Gv contains a compact open normal subgroup fixing lk(v).
Proof. We first define the tree T 0 . For each v ∈ V T , pick a compact open normal
subgroup Kv C Gv such that gKv g −1 = Kgv for all g ∈ G. Define the vertex set of T 0
by
V T 0 := V T t (v, Kv e) | v ∈ V T, e ∈ lk(v) .

We call vertices in V T type I vertices and the remaining vertices type II vertices. Define
the set of edges of T 0 to be
ET 0 = v, (v, Kv e) | v ∈ V T, e ∈ lk(v) t

(v, Kv ē), (v 0 , Kv0 e) | e ∈ ET and e = (v, v 0 ) .

Define an action of G on V T 0 by extending the action on type I vertices and on type II

vertices by g(v, Kv e) = (gv, Kgv ge) = (gv, gKv e). This action on V T 0 extends to an
action of G on T 0 . Since G acts continuously on T and each Kv is open, the stabilizer of
each vertex of T 0 is open. Hence, the action of G on T 0 is continuous. For each v ∈ V T ,
let Yv be the induced subgraph of T 0 with vertex set {v} ∪ {(v, Kv e) | e ∈ lk(v)}. Each
Yv is a tree and is equal to Star(v). It is straightforward to verify that T 0 is a tree and

collapsing each subtree Yv to the vertex v yields a finite refinement p : T 0 → T .
The stabilizer Gv of a type I vertex v ∈ V T 0 is precisely the stabilizer of the corre-
sponding vertex v ∈ V T . Moreover, Gv contains the compact open normal subgroup
Kv that fixes lk(v), since all vertices in lk(v) are of the form (v, Kv e) for some e ∈ lk(v).
Let u := (v, Kv e) be a type II vertex, where e = (w, v). We first note that Gu contains
Ge as a finite-index open subgroup. It remains to show Gu contains a compact open
normal subgroup fixing lk(u). Since Gu ≤ Gv and Kv fixes u, we deduce that Kv is a
compact open normal subgroup of Gu . Since the action of G on T is continuous and Kv
is compact, there is a finite set F ⊆ G such that Kv e = F e. The vertices adjacent to u
are precisely v and the F translates of (w, Kw ē), so u has finite valence in T 0 . It now
follows from continuity of the G-action on T 0 that the kernel Ru of the action of Gu on
lk(u) is a finite-index open normal subgroup of Gu . Thus Kv ∩ Ru is a compact open
normal subgroup of Gu fixing lk(u).
The next theorem is the main result of this subsection. The conditions in the conclu-
sion of Theorem 8.5 will be needed to apply Theorem 9.7 to deduce action rigidity.
Theorem 8.5 (Model Geometry Theorem). Let G be a locally compact group acting
minimally, cocompactly and continuously on a tree T . Suppose every vertex and edge
stabilizer is compactly presented and every vertex stabilizer contains a compact open
normal subgroup. Then T has a finite refinement T 0 such that G acts geometrically on
a tree of spaces (X, T 0 ) and for every w ∈ V T 0 ∪ ET 0 ,
(1) Xw is simply-connected;
(2) the action of Gw on Xw is discrete;
(3) all edge and vertex spaces are locally finite cell complexes and the edge maps φe
are cellular isomorphisms onto their images;
(4) for each vertex v ∈ V T 0 , the subcomplexes {φe (Xe )}e∈lk(v) are disjoint.
Moreover, if each vertex stabilizer Gv contains a compact open normal subgroup fixing
lk(v), we can assume T = T 0 .
Proof. First apply Lemma 8.4 to obtain a finite refinement p : T 0 → T such that every
vertex or edge stabilizer of T 0 contains a finite-index open subgroup isomorphic to a
vertex or edge stabilizer of T . Since all vertex and edge stabilizers of T are compactly
presented, so are all vertex and edge stabilizers of T 0 [CdlH16, Corollary 8.A.5.]. More-
over, for every vertex v ∈ V T 0 , there is a compact open normal subgroup Kv C Gv
that fixes lk(v). In the case that every vertex stabilizer of the original tree T contains
a compact open normal subgroup fixing lk(v), we do not need to apply Lemma 8.4 and
can take T = T 0 .
Assume without loss of generality that Kgv = gKv g −1 for all g ∈ G and v ∈ V T 0 . For
each edge e = (v, w) ∈ ET 0 , define Ke = Kv Kw . Since Kv and Kw are compact open
normal subgroups of Ge , we deduce Ke is a compact open normal subgroup of Ge . Note
also that Kge = gKe g −1 for all g ∈ G.
Since G contains a compact open subgroup, we can apply Proposition 8.2 to deduce
a blowup p : Y → T 0 exists. For each vertex or edge w ∈ V T 0 ∪ ET 0 , as Gw is compact-
by-discrete and acts geometrically on the graph Yw , there is a Gw -imprimitivity system
∼w on Yw whose blocks are Kw -orbits of vertices. This can be done because every Kw
is a compact open normal subgroup of Gw ; see the proof of Proposition 3.6. The union
of all these imprimitivity systems gives an imprimitivity system ∼ on Y for which each
block is finite and contained in a single vertex or edge space. The choice of the Kw
ensure ∼ is G-invariant, hence p : Y → T 0 factors through a blowup pZ : Z → T 0 , where
Z = Y / ∼.
We now define a graph of spaces (X, T 0 ) satisfying the required properties. For each
vertex v ∈ V T 0 , let Wv denote the subgraph of Z induced by Zv ∪ e∈lk(v) Ze . Since
F
Kv ≤ Ke for all e ∈ lk(v), the group Kv acts trivially on Wv . Thus, the action of Gv
on Wv is discrete. Similarly, the action of Ge on the edge space We := Ze is discrete.
For each w ∈ V T 0 ∪ ET 0 , the group Gw is compactly presented and acts geometrically
on Ww , so by Lemma 8.3, there exists M ≥ 0 such that the complex (Ww )M is simply-
connected. As there are only finitely many G-orbits of Ww for w ∈ V T 0 ∪ ET 0 , the
number M can be chosen uniformly. Set Xw = (Ww )M , which is a locally finite cell
complex. There are natural inclusions φe : Xe ,→ Xv if e ∈ ET 0 is incident to v ∈ V T 0 .
It is clear that for each vertex v ∈ V T 0 , the subcomplexes φe (Xe ) for e ∈ lk(v) are
disjoint. The spaces {Xx | x ∈ V T 0 ∪ ET 0 } and maps {φe }e∈ET 0 assemble to form a tree
of spaces (X, T 0 ). Moreover, the G-action on Z induces a G-action on (X, T 0 ).
9. Common Covers
Assumption 9.1. In this section we work with a topological group G that acts con-
tinuously on a tree of spaces (X, T ). We will assume that the edge and vertex spaces
are locally finite cell complexes, that each edge map φe is a cellular isomorphism onto
its image, and that for each vertex space Xv the subcomplexes φe (Xe ) for e ∈ lk(v) are
disjoint.
Notation 9.2. For v ∈ V T , let Gv ≤ G be the subgroup stabilizing the vertex space
Xv . Let qv : Gv → Aut(Xv ) be the induced homomorphism, and let
Qv := qv (Gv ) ≤ Aut(Xv ).
Similarly, for e ∈ ET , let Ge ≤ G be the subgroup stabilizing the edge space Xe . Let
qe : Ge → Aut(Xe ) be the induced homomorphism, and let
Qe := qe (Ge ) ≤ Aut(Xe ).
Then, Qe = Qe . We will refer to Qv and Qe as the localized vertex and edge groups.
Note that any g ∈ Gv acts on lk(v) and, since the subcomplexes φe (Xe ) are disjoint for
e ∈ lk(v), we have that qv (g) determines this action. In other words, the action of Gv
on lk(v) descends to an action of Qv on lk(v).
Remark 9.3. We will consider subgroups Γ, Γ0 ≤ G that act faithfully and geometrically
on (X, T ). The vertex and edge stabilizers of Γ and Γ0 are given by Γv := Γ ∩ Gv ,
Γ0v := Γ0 ∩ Gv , Γe := Γ ∩ Ge , and Γ0e := Γ0 ∩ Ge . The motivation for considering the
localized vertex groups is that typically the intersection of the vertex groups Γv ∩ Γ0v
inside Gv will have trivial (or at least a very disappointing) intersection. However, under
the assumption that Qv is discrete as a subgroup of Aut(Xv ), the images qv (Γv ) and
qv (Γ0v ) are guaranteed to be commensurable, so their intersection has finite index in the
respective images.
Definition 9.4. For each e ∈ lk(v), let
Qev := StabQv (φe (Xe )) ≤ Qv .
The subcomplex φe (Xe ) ⊆ Xv is isomorphic to the edge space Xe , so restricting an

automorphism of Xv to φe (Xe ) yields a surjective homomorphism
Fve : Qev Qe .
Remark 9.5. The homomorphism Fve need not be injective. For example, let Xe ∼ = R,
let Xv ∼= H2 , and let φe (Xe ) be a geodesic line in H2 (or, more precisely, cell complex
versions of these spaces). Then, the reflection in H2 about this line is a nontrivial element
of Qev that is trivial in Qe . Nonetheless, when Qv acts discretely on Xv , it immediately
follows that the kernel of Fve is finite.
If the kernels of all Fve were trivial, then any g ∈ G fixing an edge space would also
fix the adjacent vertex spaces, and by fanning outward in T we would see that g fixes
the whole tree of spaces. If in addition, the action of each Qv on Xv was discrete, then
we would conclude that G acts discretely on (X, T ), and commensurability of Γ and Γ0
in Aut(X, T ) would be immediate.
Notation 9.6. Recall from Definition 2.41 that for each g ∈ G and v ∈ V T , the
restriction of g to the vertex space Xv yields an isomorphism gv : Xv → Xgv . If g ∈ Gv ,
then qv (g) = gv . Similarly, for e ∈ ET , there is an isomorphism ge : Xe → Xge . To each
g ∈ G and v ∈ V T we obtain an isomorphism Qv → Qgv induced by conjugating by gv .
Theorem 9.7 (Common Cover Theorem). Let G be a topological group acting con-
tinuously on a tree of spaces (X, T ) as in Assumption 9.1. Let Γ, Γ0 ≤ G be finitely
generated subgroups acting geometrically on (X, T ) and without edge inversions on T .
Suppose that the following conditions are satisfied.
(1) The edge and vertex stabilizers of Γ and Γ0 each act freely on the corresponding
edge and vertex spaces;
(2) The edge and vertex spaces are simply connected;
(3) The action of Qv on Xv is discrete for all v ∈ V T ;
(4) For any (finite) set of G-orbit representatives {ei }ki=1 ⊆ lk(v), the group Qv
commands the set of subgroups {Qevi }ki=1 .
Then, the quotient graphs of spaces X/Γ and X/Γ0 have a common finite cover. In
particular, Γ and Γ0 are weakly commensurable in Aut(X, T ).
Remark 9.8. The conclusion that Γ and Γ0 are weakly commensurable in Aut(X, T )
still follows even if Γ, Γ0 act with edge inversions on T . Indeed, if we consider T as a
bipartite graph with a partition of its vertex set, then Γ, Γ0 have subgroups of index
at most 2 that stabilize both halves of the partition, and hence do not contain edge
inversions.
To prove Theorem 9.7, we construct (X̂, S), a locally-finite tree of compact spaces so
that X̂ is a regular cover of both X/Γ and X/Γ0 . Then, we apply the following theorem
to obtain a common finite-sheeted cover of X/Γ and X/Γ0 . This theorem is obtained
from Leighton’s Theorem for Graphs of Objects [She22, Theorem 4.7] by taking C to be
the category of finite cell complexes.
Theorem 9.9. Let fi : (Z, T 0 ) → (Yi , Λi ) be coverings of graphs of spaces for i = 1, 2

such that:
(1) T 0 is a tree,
(2) all edge and vertex spaces are finite cell complexes,
(3) fi restricts to isomorphisms between edge and vertex spaces.
Then (Y1 , Λ1 ) and (Y2 , Λ2 ) have a common finite-sheeted cover.
Remark 9.10. (On Conditions (1)–(4) in Theorem 9.7.) If the edge and vertex groups
act freely on the corresponding edge and vertex spaces of (X, T ) (condition (1)), then qv
and qe restrict to injective homomorphisms Γv , Γ0v → Qv and Γe , Γ0e → Qe . In particular,
qv (Γv ) ∼
= Γv and so on. We also note that condition (1) implies that Fve is injective on
qv (Γv ) ∩ Qev and qv (Γ0v ) ∩ Qev .
We assume each vertex space Xv is simply connected, which together with condition
(1), implies π1 (Xv /Γv ) ∼ = Γv , and likewise for the edge spaces. We assume the group
Qv acts discretely on the vertex space Xv , so the vertex groups qv (Γv ) and qv (Γ0v ) are
commensurable inside Qv . Condition (4) is used to ensure that the common finite-index
subgroups of the vertex groups qv (Γv ) and qv (Γ0v ) guaranteed by condition (3) can be
chosen to agree along incident edge groups.
The common “locally finite” regular cover X̂ of X/Γ and X/Γ0 can be constructed by
specifying common finite-index subgroups of the vertex groups Q̂v /qv (Γv )∩qv (Γ0v ) ≤ Qv
for v ∈ V T whose induced actions on the edge spaces match up in the sense of the next
proposition.
Proposition 9.11. With the setup of Theorem 9.7, suppose that for each v ∈ V T there
exists a finite-index subgroup Q̂v / Qv so that the following hold:
(1) (Vertex space common covers) Q̂v ≤ qv (Γv ) ∩ qv (Γ0v ).
(2) (Equivariance) gv Q̂v gv−1 = Q̂gv for all g ∈ G.
(3) (Gluing condition) For each edge e = (u, v) ∈ ET with Q̂ev := Qev ∩ Q̂v and
Q̂ēu := Qēu ∩ Q̂u , we have
Q̂e := Fve (Q̂ev ) = Fuē (Q̂ēu ) ≤ Qe ,
and the restrictions Fve |Q̂e : Q̂ev → Q̂e and Fuē |Q̂ē : Q̂ēu → Q̂e are isomorphisms.
v u
Then, there exists a locally-finite tree of compact spaces (X̂, S) so that X̂ is a regular
cover of both X/Γ and X/Γ0 .
Definition 9.12. If f1 : A1 → B and f2 : A2 → B are continuous maps of topological

spaces, then the associated fiber product is the topological space C := {(a1 , a2 ) ∈ A1 ×
A2 | f1 (a1 ) = f2 (a2 )}, considered as a subspace of A1 × A2 . Note that the projection
maps π1 : C → A1 and π2 : C → A2 satisfy f1 π1 = f2 π2 .
If C 0 is another topological space, then a commutative diagram of continuous maps
π20
C0 A2
(1) π10 f2
f1
A1 B,
is called a fiber product diagram if there is a homeomorphism h : C → C 0 such that

π1 = π10 h and π2 = π20 h. We note that if f2 : A2 → B is a covering map of cell
complexes, then π10 : C 0 → A1 restricts to a covering map on each component of C 0 .
Remark 9.13. If the map π10 in diagram (1) restricts to a covering on each component
of C 0 , then any bijection h : C → C 0 with π1 = π10 h and π2 = π20 h will automatically
be a homeomorphism. So in this case, to check that (1) is a fiber product diagram, it
suffices to check that for each (a1 , a2 ) ∈ A1 × A2 with f1 (a1 ) = f2 (a2 ) there is a unique
c ∈ C 0 with π10 (c) = a1 and π20 (c) = a2 .
Fiber product diagrams occur naturally in the context of coverings of graphs of spaces.
Specifically, suppose f : (X̂, Λ̂) → (X, Λ) is a morphism of graphs of spaces in which
the maps between vertex spaces and the maps between edge spaces are covering maps.
Then f is a covering if and only if, for each vertex v ∈ Λ and lift v̂ ∈ Λ̂, the following is
a fiber product diagram:
F
ê∈lk(v̂) X̂ê X̂v̂
(2)
F
e∈lk(v) Xe Xv
Proof of Proposition 9.11. We first define the locally-finite tree of compact spaces (X̂, S).
Let S ⊆ T be a subtree so that for each v ∈ V S, the subtree S contains exactly one
edge from each Q̂v -orbit in lk(v). Since the group Γ acts cocompactly on X, the tree S
is locally finite. For v ∈ V S and e ∈ ES, define vertex and edge spaces by
X̂v := Xv /Q̂v and X̂e := Xe /Q̂e ,
where Q̂e is given in condition (3). The space X̂v is compact because Q̂v ≤ qv (Γv )∩qv (Γ0v )
is a finite-index subgroup of Qv , and qv (Γv ) ≤ Qv acts cocompactly on Xv . Similarly, X̂e
is compact. If e ∈ lk(v), then as the subspaces {φe0 (Xe0 )}e0 ∈lk(v) are disjoint, condition
(3) implies
φe (Xe )/Q̂ev ∼
= Xe /Q̂e = X̂e .
Thus, there is an embedding X̂e ,→ X̂v . The above data defines a tree of spaces (X̂, S)
as desired.
We now prove that (X̂, S) is a regular cover of (X, T )/Γ; the proof for (X, T )/Γ0 is
identical. There are finite-sheeted regular covers of the vertex and edge spaces
X̂v = Xv /Q̂v → Xv /qv (Γv ) and X̂e = Xe /Q̂e → Xe /qe (Γe )
since Q̂v is a finite-index normal subgroup of qv (Γv ) by assumption, and likewise for
Q̂e / qe (Γe ). The covering maps for vertex and edge spaces define a morphism of graphs
of spaces θ : (X̂, S) → (X, T )/Γ, as shown in the diagram below on the right. Indeed,
the commutativity of the diagram on the right results from the commutativity of the
diagram on the left, which follows from definitions.
Q̂e qe (Γe ) X̂e = Xe /Q̂e Xe /qe (Γe )

(Fve )−1
Q̂v qv (Γv ) X̂v = Xv /Q̂v Xv /qv (Γv )

We now show that θ is a covering of graphs of spaces by verifying that the following
is a fiber product diagram for any v ∈ V S and e ∈ lk(v):
F
ê∈S∩Γv ·e X̂ê X̂v
(3)
Xe /qe (Γe ) Xv /qv (Γv )
Note that, for ê ∈ S ∩ Γv · e, if g ∈ Γv with gê = e, then the map X̂ê → Xe /qe (Γe ) is
given by Q̂ê · x 7→ qe (Γe ) · gê (x).
By Remark 9.13, it suffices to consider points qe (Γe ) · y ∈ Xe /qe (Γe ) and Q̂v · x ∈ X̂v
that map to the same point in Xv /qv (Γv ), and prove that they have a unique common
lift in tê∈S∩Γv ·e X̂ê . Firstly, since they map to the same point in Xv /qv (Γv ), we know
that there is g ∈ Γv with gv φe (y) = x. Now by construction of S and condition (1)
of Proposition 9.11, there exists h ∈ Γv with qv (h) ∈ Q̂v and ê := hge ∈ S. Putting
ŷ := (hg)e (y), we have that Q̂ê · ŷ ∈ X̂ê maps to qe (Γe ) · y ∈ Xe /qe (Γe ). Now observe
φê (ŷ) = (hg)v φe (y) = hv (x),
and Q̂v · hv (x) = Q̂v · x since qv (h) ∈ Q̂v , hence Q̂ê · ŷ also maps to Q̂v · x ∈ X̂v .
Finally, we must argue that such a point Q̂ê · ŷ is unique. Indeed, the subcomplexes
φe0 (Xe0 ) ⊂ Xv for e0 ∈ lk(v) are disjoint, so Q̂v · x intersects only one Q̂v -orbit of these
subcomplexes, hence the edge space Xê containing ŷ is uniquely determined. Moreover,
the map X̂ê → X̂v is an embedding, so the point Q̂ê · ŷ ∈ X̂ê is also uniquely determined.
Thus θ is a covering as claimed.
Having constructed a covering map θ : (X̂, S) → (X, T )/Γ, it remains to show the
cover is regular. This requires showing that the deck transformations are transitive
on the fibers of θ. First note that for g ∈ Γ and v ∈ V T , condition (2) implies that
gv Q̂v gv−1 = Q̂gv , and so if v, gv ∈ V S then the map gv : Xv → Xgv descends to an
isomorphism ĝv : X̂v → X̂gv . Similarly, if e, ge ∈ ES then we get an isomorphism
ĝe : X̂e → X̂ge .
Now returning to the regularity of θ, suppose that x̂1 , x̂2 ∈ θ−1 (x) for some vertex x
in a vertex space of (X, T )/Γ. Then x̂i ∈ X̂vi , where vi ∈ V S. The point x̂i corresponds
to an orbit Q̂vi · xi for some xi ∈ Xvi , and θ(x̂i ) corresponds to the orbit Γ · xi . So
θ(x̂1 ) = θ(x̂2 ) implies that Γ · x1 = Γ · x2 . Hence, there exists g ∈ Γ such that gx1 = x2 .
Let v := v1 . We construct a deck transformation fˆ : (X̂, S) → (X̂, S) that extends the
map ĝv : X̂v = X̂v1 → X̂v2 , such that the restriction of fˆ to each vertex space X̂u is
a map fû = ĥu : X̂u → X̂fû for some h ∈ Γ, and similarly for the restrictions to edge
spaces. It is clear from these local maps that the covering (X̂, S) → (X, T )/Γ will be
invariant under fˆ. We will define fˆ inductively by fanning out from v, so we first set
fˆv = gv = v2 and fˆv = ĝv .
Given e ∈ lk(v) ∩ ES, it might be that ge ∈/ ES. However, by the construction of S,
there exists h ∈ Γgv such that qgv (h) ∈ Q̂gv and hge ∈ ES. We then define fê = hge
and fê = (hg)

d . We now consider the following commutative diagram.
e
ge hge
Xe Xge Xhge
gv hgv
Xv Xgv Xgv
Since qgv (h) ∈ Q̂gv , we see ĥgv : X̂gv → X̂gv is the identity, so the above commutative
diagram descends to the following commutative diagram.
fê =(hg)
d
e
X̂e X̂fê
fˆv =ĝv
X̂v X̂gv .
If u = ι(e) ∈ V S is the other end of e, then we continue the definition of fˆ by setting

fû = hgu and fû = (hg)
d . We then repeat the procedure for other edges in lk(v) ∩ ES.
u
Note that the map g : lk(v) → lk(gv) will send edges in different Q̂v -orbits to edges
in different Q̂gv -orbits (again using condition (2)), while composing with the various h
does not change the orbits; this means that fˆ : lk(v) ∩ ES → lk(gv) ∩ ES is an injection,
and hence a bijection.
At the next stage we fan out from u by doing a similar thing for edges in lk(u) ∩ ES
(other than ē), and so on. We have already noted that fˆ induces bijections on links
in S, so it must define an automorphism of S, and hence an automorphism of (X̂, S).
Finally, because we were free to choose the initial points x̂1 , x̂2 ∈ θ−1 (x), the collection
of all such fˆ must form the complete group of deck transformations of the covering
(X̂, S) → (X, T )/Γ, which shows that this covering is regular.
To prove Theorem 9.7, we will first find subgroups Q̂v E Qv that satisfy the common
cover, equivariance, and gluing condition hypotheses of Proposition 9.11, then we will
apply Leighton’s Theorem for graphs of objects to the cover (X̂, S) of (X, T )/Γ and
(X, T )/Γ0 provided by Proposition 9.11. The first step in finding the subgroups Q̂v
will be taking finite-index subgroups Q̌v 6 Qv that will satisfy the common cover and
equivariance conditions.
Definition 9.14. The localized cores of the localized vertex groups are the following
intersections:
\
gv−1 qu (Γu ) gv ∩ gv−1 qu (Γ0u ) gv ≤ Qv .

Q̌v :=
g∈G, u=gv
Lemma 9.15. If the localized vertex groups act discretely on their associated vertex
spaces, then the localized core Q̌v is a finite-index normal subgroup of Qv , and gv Q̌v gv−1 =
Q̌gv for all g ∈ G.
Proof. It is immediate from the definition that gv Q̌v gv−1 = Q̌gv , and this implies the
normality of Q̌v in Qv .
The groups Γ and Γ0 act geometrically on (X, T ), so there is a constant N such that for
all v ∈ V T , Γv and Γ0v each act on Xv with fundamental domain a subcomplex consisting
of at most N simplices. The same is then true for qv (g −1 Γgv g) and qv (g −1 Γ0gv g) acting
on Xv , and by discreteness of the Qv -action we see that these are subgroups of Qv of

uniformly bounded index. As Q̌v is an intersection of such subgroups, and Qv is finitely
generated (as it acts geometrically on Xv ), it follows that Q̌v has finite index in Qv .
In the proof of Theorem 9.7 below, we define Q̂v to be a G-equivariant choice of finite-
index subgroup of Q̌v , which will guarantee that the equivariance and common cover
conditions of Proposition 9.11 are still satisfied. We ensure that this choice of finite-index
subgroup satisfies the gluing condition by applying the commanding hypothesis.
Proof of Theorem 9.7. We will first construct groups {Q̂v ≤ Qv | v ∈ V T } satisfying the
conditions of Proposition 9.11 by using the commanding assumption and the localized
cores. Then, we apply Leighton’s Theorem for graphs of objects to the cover (X̂, S) of
X/Γ and X/Γ0 guaranteed by Proposition 9.11.
The outline of the construction of the desired groups {Q̂v ≤ Qv | v ∈ V T } is as
follows. We begin by using the commanding hypothesis on the groups {Qv | v ∈ V T } to
obtain finite-index subgroups Q̇ev ≤ Qev for all v ∈ V T and e ∈ lk(v). Then, we specify
finite-index subgroups Q̂ev ≤ Q̇ev that are normal in Qev , satisfy Fve (Q̂ev ) = Fuē (Q̂ēu ) for an
edge e = (u, v), and so that Fve |Q̂e is an isomorphism onto its image. Finally, we apply
v
the commanding hypothesis to obtain Q̂v so that Q̂v ∩ Qev = Q̂ev . We ensure that our
constructions are G-equivariant throughout.
We first specify, for each v ∈ V T , a choice of finite-index subgroups Q̇ev ≤ Qev for all
e ∈ lk(v). Let v ∈ V T . Let {ei }ki=1 be a finite set of Gv -orbit representatives of edges
in lk(v) (or equivalently Qv -orbit representatives). Let
Q̌ev := Q̌v ∩ Qev / Qev ,
where Q̌v is the localized core. The restriction of Fve to Q̌ev is injective since Q̌ev ≤
qv (Γev ), and Fve is injective on qv (Γe ) by Remark 9.10. By our hypotheses, Qv commands
{Qevi }ki=1 so there exist finite-index subgroups Q̇evi E Qevi as in Definition 2.43. We may
assume that Q̇evi / Q̌evi . Define Q̇ev / Q̌ev for all e ∈ lk(v) by conjugating. Furthermore,
define families {Q̇eu }e∈lk(u) for all u ∈ G · v by conjugating, so gu Q̇eu gu−1 = Q̇ge gu for all
g ∈ G, u ∈ G · v and e ∈ lk(u). We define {Q̇eu }e∈ET by repeating the above procedure
for each G-orbit of vertices. One upshot is that we can bound the indices [Qe : Fve (Q̇ev )]
by some constant M independent of e and v since there are finitely many vertex and
edge orbits.
To now define the subgroups Q̂ev ≤ Q̇ev ≤ Qev , let Q̂e E Qe be the intersection of all
subgroups of Qe of index at most M . Then, Q̂e has finite index in Qe , and Q̂e = Q̂ē .
Let Q̂ev = (Fve )−1 (Q̂e ) ∩ Q̇ev . The map Fve is injective on Q̇ev since Q̇ev ≤ Q̌ev , and observe
that by construction Q̂e ≤ Fve (Q̇ev ), so Fve will isomorphically identify Q̂ev with Q̂e . Since
ge Qe ge−1 = Qge for g ∈ G, the characteristic definition of Q̂e and the G-equivariance of
Q̌v and Q̌ev implies that ge Q̂e ge−1 = Q̂ge and that gv Q̂ev gv−1 = Q̂gegv .
Finally, apply the fact that Qv commands the subgroups {Qevi }ki=1 and the fact that
Q̂vi ≤ Q̇evi to obtain a finite-index normal subgroup Q̂v / Qv such that Q̂v ∩ Qevi = Q̂evi .
e
By normality of Q̂v and G-equivariance of the Q̂ev , we in fact have that Q̂v ∩ Qev = Q̂ev
for all e ∈ lk(v). By intersecting with Q̌v we can assume that Q̂v ≤ Q̌v . By the G-
equivariance of the subgroups Q̌v , Q̇ev and Q̂ev , we can choose the subgroups Q̂v as above
such that gv Q̂v gv−1 = Q̂gv for all v ∈ V T and g ∈ G. Hence, condition (2) is satisfied. By
construction, condition (1) of Proposition 9.11 is satisfied. Condition (3) holds because
Fve restricts to an isomorphism from Q̂ev to Q̂e .
There exists a pair of regular coverings (X̂, S) → (X, T )/Γ and (X̂, S) → (X, T )/Γ0
by Proposition 9.11. The covering (X̂, S) → (X, T )/Γ corresponds to a subgroup Γ̂ / Γ.
Since the deck transformation group Γ/Γ̂ acts properly, cocompactly and discretely on
the tree S (note this is the deck transformation action constructed in Proposition 9.11,
not the restriction of the original action of Γ on T ), there is a finite-index subgroup
Γ̄ ≤ Γ containing Γ̂ such that Γ̄/Γ̂ acts freely on S. The covering then factors as
(X̂, S) → (X̂, S)/Γ̄ → (X, T )/Γ, where the first covering restricts to isomorphisms
between vertex and edge spaces. Similarly, we have a composition of coverings (X̂, S) →
(X̂, S)/Γ̄0 → (X, T )/Γ0 . Finally, we obtain a common finite cover of (X̂, S)/Γ̄ and
(X̂, S)/Γ̄0 by applying Theorem 9.9. This completes the proof of Theorem 9.7, so Γ and
Γ0 are abstractly commensurable.
10. Action rigidity

10.1. Ends and accessibility for locally compact groups. A uniform lattice in a
locally compact compactly generated group G acts geometrically on a Cayley graph of
G with respect to a compact generating set, hence we have the following lemma:
Lemma 10.1. If a finitely generated group Γ is a uniform lattice in a locally compact
group G, then e(Γ) = e(G).
We make use of the following result of Houghton:
Theorem 10.2 ([Hou74],[Cor18, §19.4.2]). Let G be a compactly generated locally com-
pact group. If e(G) = ∞, then G is compact-by-(totally disconnected).
We will use the following reformulation of an accessibility result due to Cornulier,
building on work of Stallings, Abels, Dunwooody and Krön–Möller:
Theorem 10.3 ([Dun85, KM08, Cor18]). Let G be an infinite-ended compactly presented
locally compact group. Then G has a continuous cocompact action on a tree with compact
edge stabilizers and vertex stabilizers compactly generated and having at most one end.
Remark on the Proof. It follows from [Cor18, Corollary 19.39] that G does not admit
a continuous proper transitive isometric action on the real line. It now follows from
Definition 19.40 and Theorem 19.45 of [Cor18] that G admits a continuous cocompact
action on a tree T with compact edge stabilizers and vertex stabilizers having at most
one end. Although not explicitly stated in [Cor18], it follows from [KM08, Theorem
3.27] and [Cor18, Theorem 19.41] that vertex stabilizers are compactly generated.
10.2. Action rigidity for infinite-ended groups. Recall from Definition 2.45 that a
group Γ commands its finite subgroups if for every finite subgroup Λ ≤ Γ, there exists a
finite-index subgroup Γ0 ≤ Γ so that Γ0 ∩ Λ = {1}. For example, residually finite groups
and virtually torsion-free groups command their finite subgroups.
Theorem 10.4. Suppose Γ and Γ0 are finitely presented infinite-ended groups that
command their finite subgroups. Suppose every one-ended vertex group in a Stallings–
Dunwoody decomposition of Γ is graphically discrete. If Γ and Γ0 act geometrically on the
same proper metric space, then Γ and Γ0 are abstractly commensurable. In particular,
Γ is action rigid within the family of groups that command their finite subgroups.
Proof. Suppose that Γ and Γ0 act geometrically on a proper metric space Y . Then
G := Isom(Y ) is an infinite-ended compactly presented locally compact group that con-
tains Γ and Γ0 as virtual uniform lattices. By Theorem 10.3, G admits a continuous
cocompact action on a tree T with compact edge stabilizers and vertex stabilizers com-
pactly generated and with at most one end. After passing to a subtree if necessary, we
may assume the action is minimal.
Since Gv is open for all v ∈ V T , it contains Γv as a virtual uniform lattice by
Lemma 2.9. By Theorem 10.2, G is compact-by-(totally disconnected). Therefore, as Γv
is graphically discrete, the group Gv contains a compact open normal subgroup. Thus,
by Theorem 8.5, the tree T has a finite refinement T 0 such that G acts geometrically on
a tree of spaces (X, T 0 ) that satisfies Assumption 9.1. Moreover, for all a ∈ V T t ET ,
the space Xa is simply connected and the action of Ga on Xa is discrete. The groups Γ
and Γ0 are virtual uniform lattices in Aut(X, T 0 ).
Since Γ and Γ0 command all of their finite subgroups, we may replace these groups by
finite-index subgroups Γ̂ ≤ Γ and Γ̂0 ≤ Γ0 so that the edge and vertex stabilizers of Γ̂ and
Γ̂0 act freely on the corresponding edge and vertex spaces. In particular, Γ̂, Γ̂0 act freely
on Aut(X, T 0 ). The group Qv = qv (Gv ) as in Notation 9.2 is discrete, and hence contains
Γv as a finite-index subgroup. Thus, Qv commands all of its finite subgroups since Γv
does. Therefore, Γ̂ and Γ̂0 are weakly commensurable in Aut(X, T 0 ) by Theorem 9.7. In
particular, Γ̂ and Γ̂0 are abstractly commensurable, hence so are Γ and Γ0 .
By adding a stronger hypothesis to the group Γ, we obtain that Γ is action rigid.
Definition 10.5. A finitely generated group Γ persistently commands its finite sub-
groups if any group Γ0 virtually isomorphic to Γ commands its finite subgroups.
Example 10.6. Virtually polycyclic groups (e.g. finitely generated nilpotent groups)
persistently command their finite subgroups by [DK18, Theorem 13.77].
Example 10.7. The fundamental group of a compact 3-manifold persistently commands

its finite subgroups. Indeed, if a finitely generated group Γ is virtually isomorphic to the
fundamental group of a compact 3-manifold M , then Γ contains a finite-index subgroup
Γ0 isomorphic to the fundamental group of a compact 3-manifold N by [HL, Theorem
1.2]. The group π1 (N ) is residually finite by [Hem87] together with geometrization;
see [AFW15]. Hence, Γ is residually finite, and therefore commands its finite subgroups.
Example 10.8. Any cubulated hyperbolic group persistently commands its finite sub-
groups. This can be deduced from the fact that all cubulated hyperbolic groups are
residually finite, due to Agol [Ago13, Corollary 1.3]. Indeed, suppose Γ1 is a hyperbolic
group that acts geometrically on a CAT(0) cube complex X, and let Γ2 be virtually
isomorphic to Γ1 . We claim that Γ2 has a finite-index subgroup that also acts on X
geometrically, hence Γ2 is residually finite and commands its finite subgroups. As Γ1
and Γ2 are virtually isomorphic, for i = 1, 2 there exist finite-index subgroups Γ0i ≤ Γi
and finite normal subgroups Fi C Γ0i such that Γ01 /F1 ∼ = Γ02 /F2 . Since Γ01 is residually
00 0
finite, there is a finite-index subgroup Γ1 ≤ Γ1 intersecting F1 trivially. Thus there is a
finite-index subgroup Γ002 ≤ Γ02 containing F2 such that Γ002 /F2 ∼
= Γ001 F1 /F1 ∼
= Γ001 . Since
00 00
Γ1 acts geometrically on X, so does Γ2 as required.
Lemma 10.9. If Γ is a finitely generated infinite-ended group such that every one-ended
vertex group in a Stallings–Dunwoody decomposition of Γ commands its finite subgroups,
then Γ commands its finite subgroups.
Proof. This can be shown using elementary arguments about finite covers of graphs of
groups. The result also follows from [She23, Proposition 6.11(1)] and the fact that a
group commands its finite subgroups if and only if its finite subgroups are discrete in
the profinite topology.
Theorem 10.10. If Γ is a finitely presented infinite-ended group so that all one-ended

vertex groups in a Stallings–Dunwoody decomposition of Γ are graphically discrete and
persistently command their finite subgroups, then Γ is action rigid.
Proof. Suppose that Γ and a group Γ0 act geometrically on the same proper metric
space Y . Then, the group Γ0 is finitely presented and infinite-ended. We claim that Γ0
commands its finite subgroups.
The group G = Isom(Y ) acts continuously and compactly on a tree T with compact
edge stabilizers and vertex stabilizers that are compactly generated and have at most
one end by Theorem 10.3. As above, since Gv is open for all v ∈ V T , it contains Γv
and Γ0v as virtual uniform lattices by Lemma 2.9. Since Γv is graphically discrete and
G is compact-by-(totally disconnected) by Theorem 10.2, the group Gv is compact-by-
discrete. Hence, Γv and Γ0v are virtually isomorphic. Thus, Γ0v commands its finite
subgroups, so Γ0 commands its finite subgroups by Lemma 10.9. Therefore, Γ and Γ0
are abstractly commensurable by Theorem 10.4.
Remark 10.11. In fact, the proof of Theorem 10.10 says that if Γ is as above and
both Γ and Γ0 act geometrically on the same proper space, then they are abstractly
commensurable.
Theorem 10.12. The fundamental group of a (possibly reducible) closed 3-manifold is

action rigid if and only if the manifold does not admit H3 or Sol geometry.
Proof. Let M be a closed 3-manifold; after passing to a two-sheeted cover if necessary, we

may assume M is oriented. Suppose first that M is irreducible. If M is non-geometric,
then π1 (M ) is action rigid by Theorem 7.9 and Lemma 2.38. If M admits H3 or Sol
geometry, then M is not action rigid since there are infinitely many commensurability
classes of manifolds with each of these geometries; see [Neu97, Theorem A]. If M admits
one of the geometries S 3 , S 2 × R, E3 or Nil, then π1 (M ) is quasi-isometrically rigid,
and hence action rigid; see [Fri18, Theorem 1.7]. Finally, if M admits either H2 × R or
^
SL 2 (R) geometry and Γ is quasi-isometric to M , then Γ is virtually isomorphic to the
fundamental group of a closed geometric 3-manifold that admits the geometry of either
H2 × R or SL ^ 2 (R) [Rie01]. Action rigidity for these groups then follows from [DT16,
Corollary 1.2] and [Neu97, Theorem A].
Now suppose the fundamental group of M is infinite-ended and each one-ended vertex
group in a Stallings–Dunwoody decomposition of π1 (M ) is the fundamental group of an
irreducible closed 3-manifold. Then, each one-ended vertex group is graphically discrete
by Theorems 7.1 and Theorem 7.9 and persistently commands its finite subgroups by
Example 10.7. Thus, π1 (M ) is action rigid by Theorem 10.10.
If M is a 3-manifold with toroidal boundary, then π1 (M ) may have a one-ended vertex

group in a Stallings–Dunwoody decomposition that is isomorphic to Fn × Z, which is
not graphically discrete. The proof above does not extend to this setting, which includes
the case π1 (M ) ∼
= (F2 × Z) ∗ (F2 × Z), and we ask the following.
Question 10.13. Is the fundamental group of a 3-manifold with toroidal boundary
action rigid?
Example 10.14. This example, which follows the argument of [SW, Proposition 7.1],
illustrates that the commanding finite subgroup hypothesis is necessary in the action
rigidity results given in this section. By work of Raghunathan [Rag84, Main Theorem]
there exists a torsion-free uniform lattice Λ ≤ Spin(2, n) and a finite extension 1 → F →
Λ̂ → Λ → 1 so that Λ̂ is not virtually torsion-free. Thus, there exists an element f ∈ F
of order r that is contained in every finite-index subgroup of Λ̂. Let
∆ = ∆(r, 5, 5) = a, b, c | a2 = b2 = c2 = (ab)r = (bc)5 = (ac)5 = 1

be a hyperbolic triangle group. Then, the amalgamated free product H = Λ̂∗hf i=habi ∆ is
not action rigid, but Λ̂ and ∆ are both graphically discrete by Theorem 3.16. Indeed, Λ̂
acts geometrically on the symmetric space X associated to SO(2, n) because Spin(2, n) is
a double cover of the identity component of SO(2, n). The proof of [SW, Proposition 7.1]
can thus be applied by replacing vertex spaces isometric to HnF in the model geometry
with copies of the symmetric space X.
11. Quasi-isometric rigidity

In this section we outline a two-step method for proving quasi-isometric rigidity the-
orems; see Proposition 11.2. We then apply this approach in Theorem 11.16, giving
quasi-isometric rigidity for certain graphs of hyperbolic n-manifold groups.
11.1. From action rigidity to quasi-isometric rigidity.

Definition 11.1. A metric space X has the uniform quasi-isometry property if there
exist constants K ≥ 1 and A ≥ 0 such that for every quasi-isometry f : X → X, there
is a (K, A)-quasi-isometry f 0 : X → X such that supx∈X d(f (x), f (x0 )) < ∞.
We recall that a finitely generated group Γ is quasi-isometrically rigid if every finitely

generated group quasi-isometric to Γ is virtually isomorphic to Γ.
Proposition 11.2. Let Γ be a finitely generated tame group. Suppose that:
(1) Γ has the uniform quasi-isometry property;
(2) Γ is action rigid.
Then Γ is quasi-isometrically rigid.
Proof. Let G = QI(Γ). Since Γ is tame and has the uniform quasi-isometry property,
there is a cobounded quasi-action {fg }g∈G of G on Γ such that [fg ] = g; see e.g. [Mar22b,
Lemma 3.4]. It follows from work of Margolis that there is a proper quasi-geodesic
metric space X and a quasi-isometry f : Γ → X that quasi-conjugates the quasi-action
of G on Γ to an isometric action of G on X [Mar22a, Proposition 4.5 and Theorem
4.20]. Suppose Γ0 is a finitely generated group quasi-isometric to Γ. Since Γ0 admits a
proper cobounded quasi-action on Γ, the map f quasi-conjugates this quasi-action to a
geometric action of Γ0 on X. Thus, every finitely generated group quasi-isometric to Γ

acts geometrically on X. Since Γ is action rigid, this implies Γ is quasi-isometrically
rigid.
Remark 11.3. If Γ is tame and has infinite outer automorphism group, then the uniform
quasi-isometry property is never satisfied [Why10, Claim 1.26]. In particular, infinite-
ended groups do not have the uniform quasi-isometry property.
11.2. Action rigidity for graphs of groups. The aim of this subsection is to prove
the action rigidity result given in Theorem 11.8 by applying Theorems 8.5 and 9.7. We
begin with the relevant definitions.
Definition 11.4. If G is a finite graph of finitely presented groups, an associated Bass–

Serre tree of spaces (X, T ) is the universal covering of a graph of spaces (Z, Λ) associated
to G (in the sense of Definition 2.40) in which all vertex and edge spaces of (Z, Λ) are
compact CW complexes.
Notation 11.5. Throughout the remainder of this section, given a tree of spaces (X, T ),
we identify an edge space Xe with the subspace of the form Xe × {0} in the total space;
see Definition 2.39. In this way, Xe = Xē is a subspace of X for each e ∈ ET .
Work of Papasoglu and Mosher–Sageev–Whyte [Pap05, MSW11] gives many examples

of trees of spaces that are preserved by quasi-isometries in the following sense:
Definition 11.6. A tree of spaces (X, T ) is preserved by quasi-isometries if for every

(K, A)-quasi-isometry f : X → X, there is a constant B = B(X, K, A) such that the
following hold:
(1) if w ∈ V T ∪ ET , there is some w0 ∈ V T ∪ ET such that dHaus (f (Xw ), Xw0 ) ≤ B;
(2) if w0 ∈ V T ∪ ET , there is some w ∈ V T ∪ ET such that dHaus (f (Xw ), Xw0 ) ≤ B.
Notation 11.7. Suppose G = G(Λ, {Γv }, {Γe }, {Θe }) is a graph of groups. If v ∈ Λ,

the collection of images of incident edge maps {Θe (Γe )}e∈lk(v) is assumed to be indexed
by lk(v). In particular, distinct edges e1 , e2 ∈ lk(v) such that Θe1 (Γe1 ) = Θe2 (Γe2 )
correspond to distinct elements of {Θe (Γe )}e∈lk(v) .
We recall a graph of group G is minimal if the action of π1 (G) on the associated

Bass–Serre tree has no proper invariant subtree.
Theorem 11.8. Suppose Γ is the fundamental group of a minimal finite graph of groups
G satisfying the following properties:
(1) Every vertex group of G is graphically discrete, hyperbolic and cubulated.
almost malnormal family of infinite quasi-convex subgroups.
(3) The associated Bass–Serre tree of spaces (X, T ) is preserved by quasi-isometries.
Then Γ is action rigid.
Remark 11.9. Let G = G(Λ, {Γv }, {Γe }, {Θe }) be a minimal graph of groups such that
every vertex group Γv is infinite and the collection of images of incident edge maps
is an almost malnormal family of Γv . Then every edge monomorphism Γe → Γv has
infinite-index image in Γv .
We will use the following results in the proof of Theorem 11.8. Recall an action of a
group G on a tree T is k-acylindrical if any segment of length k in T has finite stabilizer.
Lemma 11.10. Suppose Γ is the fundamental group of a minimal finite graph of groups
G such that for each vertex group, the collection of images of incident edge maps is an
almost malnormal family. Then the action of Γ on the associated Bass–Serre tree is
2-acylindrical.
Moreover, if vertex groups of G are hyperbolic and every edge map has quasi-convex
image, then Γ is hyperbolic and all vertex and edge stabilizers are quasi-convex in Γ.
Proof. The fact that the action of Γ is 2-acylindrical follows from the condition that
images of edge maps form an almost malnormal family of Gv . Hyperbolicity can be
deduced from the Bestvina–Feighn combination theorem [BF92] by adapting a proof
of Kapovich, who considers a slightly more restrictive notion of acylindricity [Kap01].
Alternatively, the result follows verbatim from a much more general theorem of Martin–
Osajda [MO21].
We make extensive use of the next proposition to prove rigidity results in this section.
Informally, it gives a criterion on a tree of space (X, T ) which ensures the combinatorial
structure of T can be completely recovered from the coarse geometry of its vertex and
edge spaces.
Proposition 11.11. Let G be a minimal finite graph of finitely generated groups. Sup-
pose that for each vertex group, the collection of images of incident edge maps is an
almost malnormal family of infinite subgroups. Assume T is the Bass–Serre tree asso-
ciated to G, the corresponding tree of spaces is (X, T ). Then the following hold for all
w, w0 ∈ V T t ET :
(1) If dHaus (Xw , Xw0 ) < ∞, then w = w0 .
(2) If w 6= w0 , then Xw0 ⊆ NR (Xw ) for some R if and only if w is a vertex and w0
is an edge incident to w.
(3) For any distinct edges e1 , e2 ∈ ET , NR (Xe1 ) ∩ NR (Xe2 ) is bounded for all R.
Proof. Let Γ = π1 (G) and let Γw denote the stabilizer of w for each w ∈ V T t ET . By
the Milnor–Schwarz lemma, the action of Γ on X induces a quasi-isometry f : Γ → X
such that dHaus (f (Γw ), Xw ) < ∞ for each w ∈ V T t ET . It follows from [MSW11,
Lemma 2.2 and Corollary 2.4] that:
(a) Xw and Xw0 are at finite Hausdorff distance if and only if Γw and Γw0 are
commensurable;
(b) Xw ⊆ NR (Xw0 ) for some R if and only if Γw is commensurable to a subgroup
of Γw0 .
(c) NR (Xw ) ∩ NR (Xw ) is bounded for all R if and only if Γw ∩ Γw0 is finite.
Remark 11.9, Lemma 11.10 and the hypotheses on Γ ensure that (1)–(3) follow imme-
diately from (a)–(c) respectively.
We now use Proposition 11.11 to show that under suitable hypotheses, quasi-isometries
of X induce automorphisms of T in a functorial manner:
Corollary 11.12. Let (X, T ) be a tree of spaces preserved by quasi-isometries and

satisfying properties (1) and (2) in the conclusion of Proposition 11.11. Then for every
(K, A)-quasi-isometry f : X → X, there is a constant B = B(X, K, A) and a unique tree

automorphism f∗ : T → T such that dHaus (f (Xw ), Xf∗ (w) ) ≤ B for all w ∈ V T ∪ ET .
Moreover, the following are satisfied for all quasi-isometries f, g : X → X:
(1) If f and g are close, then f∗ = g∗ .
(2) If (g ◦ f )∗ = g∗ ◦ f∗ ;
(3) (IdX )∗ = IdT .
Proof. Let f : X → X be a (K, A)-quasi-isometry. Since (X, T ) is preserved by quasi-

isometries, there is a constant B = B(X, K, A) and a function f∗ : V T tET → V T tET
such that dHaus (f (Xw ), Xf∗ (w) ) ≤ B for all w ∈ V T tET . By (1) of Proposition 11.11, no
two distinct vertex or edge spaces are at finite Hausdorff distance, hence f∗ is uniquely
determined. Moreover, (2) of Proposition 11.11 ensures f∗ sends vertices to vertices,
edges to edges, and preserves incidence of edges and vertices. Therefore, f∗ must be a
tree morphism. Applying the same argument to a coarse inverse f¯ of f , we deduce f∗ is
invertible and hence an automorphism. The remaining properties follow easily from the
definition of f∗ and the fact that distinct vertex or edge spaces are at infinite Hausdorff
distance.
Proof of Theorem 11.8. Suppose Γ and Γ0 are virtual uniform lattices in the same locally
compact group G. As in Remark 2.15, we can assume without loss of generality that
G is second countable. Let (X, T ) be the associated Bass–Serre tree of spaces for Γ.
We first exhibit a quasi-action of G on the space X. Agol’s Theorem [Ago13] implies
vertex groups of G are virtually special. Thus, Γ has, in the terminology of Wise, an
almost malnormal quasiconvex hierarchy terminating in virtually special groups, so Γ is
virtually special [Wis21, Theorem 11.2]. Hence, Γ is virtually torsion-free. We can now
replace Γ by a finite-index torsion-free subgroup so that Γ is a uniform lattice in G. As
in Section 3.5, the uniform lattice embedding Γ → G induces a quasi-action of G of Γ.
Since Γ acts geometrically on X, the quasi-action of G on Γ can be quasi-conjugated to
a quasi-action {fg }g∈G of G on X.
We now show that G acts minimally, cocompactly, and continuously on the tree T . By
Proposition 11.11 and Corollary 11.12, there is a constant B such that for every g ∈ G
there is a tree automorphism g∗ ∈ Aut(T ) such that dHaus (fg (Xw ), Xg∗ (w) ) ≤ B. By the
functoriality properties of g 7→ g∗ given in Corollary 11.12, this defines an action of G on
T extending the action of Γ on T . In particular, this action is cocompact and minimal.
By Lemma 3.24, there exists a constant C such that for every x ∈ X and sequence
(gi ) → 1 in G, we have d(fgi (x), x) ≤ C for all i sufficiently large. Since distinct vertex
or edge spaces are at infinite Hausdorff distance by Proposition 11.11, it follows that if
a sequence (gi ) → Id in G, then the sequence (gi )∗ converges to the identity pointwise
in T . Hence, the action of G on T is continuous.
We claim each vertex group Gv contains a compact open normal subgroup fixing
lk(v). Let v ∈ V T . Since Γ is a uniform lattice in G and Gv is open, Lemma 2.9 implies
Γv is a uniform lattice in Gv . Since Γv is graphically discrete, Gv has a compact open
normal subgroup Kv C Gv . Now for each g ∈ Gv , we have dHaus (fg (Xv ), Xv ) ≤ B, hence
there is quasi-action {fgv }g∈Gv of Gv on Xv such that supx∈Xv d(fg (x), fkv (x)) ≤ B. This
quasi-action is cobounded since Γv ≤ Gv acts cocompactly on Xv . By Lemma 3.25, there
is a constant D such that supx∈Xv d(x, fkv (x)) ≤ D for all k ∈ Kv . Since no two distinct
edge spaces are at finite Hausdorff distance and dHaus fkv (φe (Xe )), φk∗ (e) (Xk∗ (e) ) < ∞

for all k ∈ Kv and e ∈ lk(v), it follows that Kv fixes lk(v) pointwise.

Finally, for each w ∈ V T ∪ ET , since Γw is finitely presented and Gw is open, it
follows from Lemmas 2.9 and 2.13 that Gw is compactly presented. Thus, Theorem 8.5
yields a geometric action of G on a tree of spaces (X 0 , T ) satisfying the conclusions of
Theorem 8.5.
To deduce that Γ and Γ0 are abstractly commensurable, we show Theorem 9.7 can
be applied. Each Gv is compact-by-discrete, so Γ0v is virtually isomorphic to Γv . Thus,
Γ0v is virtually special by the argument in Example 10.8, together with Agol’s Theorem.
Therefore, Γ0 is virtually torsion-free as in the argument at the beginning of this proof.
Since Γ and Γ0 are virtually torsion-free, we replace them with finite-index subgroups
that act freely on (X 0 , T ) and are uniform lattices (rather than virtual uniform lattices)
in G.
Define Qv and Qev as in Notation 9.2 and Definition 9.4. The group Qv is a hyperbolic
group containing Γv as a finite-index subgroup. In particular, Qv is virtually special. For
each e ∈ lk(v), the group Qev is a quasi-convex subgroup of Qv containing Γe as a finite-
index subgroup. This implies, for any set of G-orbit representatives {ei }ki=1 ⊆ lk(v),
the groups {Qevi }ki=1 form an almost malnormal family of quasi-convex subgroups in Qv .
Therefore, Qv commands {Qevi }ki=1 by Example 2.48. All the hypotheses of Theorem 9.7
are met, so Γ and Γ0 are abstractly commensurable.
While our main application of Theorem 11.8 appears in the next subsection, we include
two immediate corollaries. These corollaries use Lemma 11.10 and the fact that the
Bass–Serre tree of spaces corresponding to Bowditch’s JSJ tree of a one-ended hyperbolic
group that is not cocompact Fuchsian is preserved by quasi-isometries [Bow98]. The first
corollary is a special case of a more general theorem we provide in a forthcoming paper.
Corollary 11.13. If M and N are closed hyperbolic 3-manifolds, then π1 (M ) ∗Z π1 (N ),
where a generator of Z maps to primitive elements in π1 (M ) and π1 (N ), is action rigid.
Note that π(M ) and π1 (N ) are cubulated, as noted in Remark 11.17.
Manning–Mj–Sageev proved many hyperbolic surface-by-free groups are cubulated [MMS],
which, together with Theorem 3.16, yields the following corollary.
Corollary 11.14. If Γ and Γ0 are cubulated hyperbolic surface-by-free groups, then
Γ ∗Z Γ0 , where a generator of Z maps to primitive elements in Γ and Γ0 , is action rigid.
Quasi-isometric rigidity in the above examples is open.
We remark that Hruska–Wise proved that a separable quasi-convex subgroup of a hy-
perbolic group is virtually almost malnormal, giving other instances where Theorem 11.8
can be applied; see Corollary 11.18.
Proposition 11.15 ([HW09, Theorem 9.3]). Let Γ be a hyperbolic group, and let H ≤ Γ
be a quasi-convex separable subgroup. Then there exists a finite-index subgroup Γ0 ≤ Γ
containing H such that H is almost malnormal in Γ0 .
11.3. Quasi-isometric rigidity for graphs of hyperbolic manifold groups. The
aim of this section is to prove the following quasi-isometric rigidity theorem. If Γ is
a group of finite cohomological dimension cd(Γ), the cohomological codimension of a
subgroup Λ ≤ Γ is cd(Γ) − cd(Λ).
Theorem 11.16. Let Γ be the fundamental group of a minimal finite graph of groups
(1) Each vertex group of G is cubulated and is the fundamental group of a closed
hyperbolic n-manifold for some n ≥ 3.
almost malnormal family of non-elementary quasi-convex subgroups with coho-
mological codimension at least two.
Remark 11.17. Work of Kahn–Markovic and Bergeron–Wise implies that fundamental
groups of closed hyperbolic 3-manifolds are cubulated [KM12, BW12]. A theorem of
Bergeron–Haglund–Wise shows many additional hyperbolic n-manifolds are cubulated
[BHW11].
Before proving Theorem 11.16, we state a concrete special case.

Corollary 11.18. Let Γ1 and Γ2 be fundamental groups of closed hyperbolic 3-manifolds,
and let F1 ≤ Γ1 and F2 ≤ Γ2 be isomorphic quasi-convex non-abelian free subgroups.
There exist finite-index subgroups Γ01 ≤ Γ1 and Γ02 ≤ Γ2 containing F1 and F2 respec-
tively, such that any finitely generated group quasi-isometric to Γ = Γ01 ∗F1 =F2 Γ02 is
abstractly commensurable to Γ.
Proof. By Proposition 11.15, there are finite-index subgroups Γ01 ≤ Γ1 and Γ02 ≤ Γ2
containing F1 and F2 as malnormal subgroups. Let Γ = Γ01 ∗F1 =F2 Γ02 . By Remark 11.17,
Γ1 and Γ2 are cubulated. The rest of the hypotheses of Theorem 11.16 are satisfied, so
the theorem can be applied.
Fix Γ and G as in Theorem 11.16 for the remainder of this section. Let (X, T )
be the Bass–Serre tree of spaces associated to G. We first prove Γ is action rigid using
Theorem 11.8, and then we will prove Γ has the uniform quasi-isometry property. Quasi-
isometric rigidity then follows from Proposition 11.2, which allows us to upgrade action
rigidity to quasi-isometric rigidity.
11.3.1. Action Rigidity. To prove that Γ is action rigid, we use Theorem 11.8. To do so,
it remains to prove that the tree of spaces (X, T ) is preserved by quasi-isometries. We
note that the hypotheses of Theorem 11.16 ensure the conclusions of Proposition 11.11
hold. The next proposition is a consequence of work of Mosher–Sageev–Whyte [MSW11].
It will be used to prove both action rigidity and the uniform quasi-isometry property.
Proposition 11.19 ([MSW11, Theorem 1.6]). The tree of spaces (X, T ) is preserved
by quasi-isometries.
Proof. We apply Theorem 1.6 [MSW11], which says that if a graph of groups G satisfies
the following five conditions, then its associated Bass–Serre tree of spaces is preserved
by quasi-isometries.
(MSW1) G is finite type, irreducible, and of finite depth.
(MSW2) No depth zero raft of the Bass–Serre tree of G is a line.
(MSW3) Every depth zero vertex group of G is coarse PD.
(MSW4) For every one vertex, depth zero raft of the Bass–Serre tree, the crossing graph
condition holds.
(MSW5) Every vertex and edge group is coarse finite type.

Since some of these conditions are quite technical and will not be needed outside the
current proof, we refer the reader to [MSW11] for definitions of the above terms. We
now show that a graph of groups G satisfying the hypothesis of Theorem 11.16 satisfies
the preceding five conditions, which says (X, T ) is preserved by quasi-isometries.
Firstly, G is a finite graph of groups with hyperbolic vertex and edge groups, so Γ
is finitely presented and G has finite type. By Remark 11.9, every edge group is an
infinite-index subgroup of its adjacent vertex groups. Therefore, G is irreducible and all
vertex groups have depth zero, hence G satisfies (MSW1). All depth zero rafts consist
of a single vertex, so (MSW2) is satisfied. Each vertex group of G is the fundamental
group of a closed aspherical manifold, hence is coarse PD; thus (MSW3) is satisfied.
Since every vertex group of G is the fundamental group of a closed aspherical mani-
fold, it is of type F P . Furthermore, as edge groups are hyperbolic, hence of type F P∞ ,
and have finite cohomological dimension (as they are subgroups of their adjacent vertex
groups), they also have type F P . In particular, all vertex and edge groups have cohomo-
logical dimension equal to their coarse dimension as defined in [MSW11]. As a result,
[MSW11, Lemma 3.8] ensures that no edge group Xe coarsely separates an adjacent
vertex group, so each vertex group has empty crossing graph and (MSW4) vacuously
holds. As noted above, since every vertex and edge group of G has type F P , it is of
coarse finite type, hence (MSW5) holds.
Proposition 11.20. If Γ is as in Theorem 11.16, then Γ is action rigid.
Proof. It follows from the definition of Γ, Theorem 3.16, and Proposition 11.19 that all
the hypotheses of Theorem 11.8 are satisfied, hence Γ is action rigid.
Remark 11.21. Thus far, we have made no use of the fact that edge groups are non-
elementary. In particular, if Γ satisfies all the hypotheses of Theorem 11.16 except the
non-elementary edge group condition, then Γ is action rigid.
11.3.2. The uniform quasi-isometry property. In light of Proposition 11.2, Example 2.30
and Proposition 11.20, in order to show Γ is quasi-isometrically rigid it remains to
show Γ satisfies the uniform quasi-isometry property. We first recall relevant results of
Biswas [Bis12]; see [Sch97, BM12, Mj12] for related results.
Definition 11.22. Let Γ be a finitely generated group acting geometrically on a proper

geodesic hyperbolic space X. A Γ-symmetric pattern J in X is a Γ-invariant collection
of quasi-convex subsets of X such that:
(1) for every J ∈ J , the stabilizer StabΓ (J) acts cocompactly on J and is an infinite,
infinite-index subgroup of Γ;
(2) J contains only finitely many Γ-orbits.
A symmetric pattern in X is a Γ-symmetric pattern for some finitely generated group Γ
acting geometrically on X. We denote X together with a symmetric pattern J by
(X, J ). A pattern-preserving quasi-isometry f : (X, J ) → (X 0 , J 0 ) is a quasi-isometry
f : X → X 0 such that there exists a constant A ≥ 0 so that:
(1) for all J1 ∈ J , there exists a J2 ∈ J 0 such that dHaus (f (J1 ), J2 ) ≤ A;
(2) for all J2 ∈ J 0 , there exists a J1 ∈ J such that dHaus (f (J1 ), J2 ) ≤ A.
Let QI(X, J ) be the subgroup of QI(X) consisting of equivalence classes of pattern-

preserving quasi-isometries of J .
Remark 11.23. Suppose a group Γ acts geometrically on a proper hyperbolic space X

and H1 , . . . , Hn ≤ Γ is a collection of infinite, infinite-index quasi-convex subgroups. If
x ∈ X, then J := {gHi x | 1 ≤ i ≤ n, g ∈ Γ} is a Γ-symmetric pattern. Moreover, up
to modifying each element of J by uniformly bounded finite Hausdorff distance, every
Γ-symmetric pattern arises in this way.
Biswas proved that for n ≥ 3, pattern-preserving quasi-isometries of Hn are bounded

distance from isometries. More precisely, the theorem below follows from the discussion
on [Bis12, p590]. Note that Biswas states it in the more general framework of uniformly
proper maps.
Theorem 11.24 ([Bis12]). Suppose J and J 0 are symmetric patterns in Hn for some
n ≥ 3. If f : (Hn , J ) → (Hn , J 0 ) is a pattern-preserving quasi-isometry, then there is a
hyperbolic isometry f 0 : Hn → Hn such that supx∈Hn d(f (x), f 0 (x)) < ∞.
The following is a consequence of Theorem 11.24, which is used implicitly in the proof
of [Bis12, Corollary 1.3]; see also [Mj12, Corollary 1.5].
Corollary 11.25. Suppose a group Γ acts geometrically on Hn for some n ≥ 3 and J is

a Γ-symmetric pattern in Hn . Then, QI(Hn , J ) can be identified with a discrete subgroup
of Isom(Hn ), and under this identification Γ is a finite-index subgroup of QI(Hn , J ).
For completeness, we provide a proof of Corollary 11.25 from Theorem 11.24.
Proof. Theorem 11.24 ensures that QI(Hn , J ) ≤ Isom(Hn ). It is easy to see that
QI(Hn , J ) is closed, hence a Lie group by Cartan’s Theorem. Since QI(Hn , J ) coarsely
permutes the discrete pattern J , it is totally disconnected; this follows from [Mj12,
Proposition 3.5], noting that the compact-open topology on Isom(Hn ) coincides with
the topology on Isom(Hn ) as a subgroup of Homeo(∂Hn ) equipped with the topology
of uniform convergence. Lemma 2.17 applied to the inclusion map implies QI(Hn , J )
is discrete. As Γ is a uniform lattice contained in the discrete group QI(Hn , J ), it is a
finite-index subgroup of QI(Hn , J ).
Symmetric patterns arise naturally in the setting of Theorem 11.16. Indeed, let Γ be
as in Theorem 11.16 and let (X, T ) be the associated Bass–Serre tree of spaces. If v is
a vertex of T , then the vertex group Γv is a uniform lattice in Hnv for some nv ≥ 3.
Since Γv acts geometrically on both Xv and Hnv , the Milnor–Schwarz Lemma implies
there exists a coarsely Γv -equivariant quasi-isometry hv : Xv → Hnv . Let hv be a coarse
inverse to hv . Since there are finitely many Γ-orbits of vertices of T , we can without
loss of generality assume hv and hv have uniform quasi-isometry constants as v varies.
For each v ∈ V T , define a Γv -symmetric pattern Jv ⊆ Xv by
Jv := {φe (Xe ) | e ∈ lk(v)}.
Since the map hv is coarsely Γv -equivariant, Remark 11.23 yields a Γv -symmetric pattern
Jˆv ⊆ Hnv such that hv is a pattern-preserving quasi-isometry (Xv , Jv ) → (Hnv , Jˆv ).
The following lemma will be used to show quasi-isometries are pattern preserving.
Lemma 11.26. For K ≥ 1 and A ≥ 0, there is a constant B = B(K, A, X) such that

the following holds. If v, w ∈ V T and f : Xv → Xw is a (K, A)-quasi-isometry such that
dHaus (f (J), J 0 ) < ∞ for some J ∈ Jv and J 0 ∈ Jw , then dHaus (f (J), J 0 ) ≤ B.
Proof. For each v ∈ V T and J ∈ Jv , since StabΓv (J) is quasi-convex and acts cocom-
pactly on J, it follows from a result of Swenson that J has finite Hausdorff distance from
WCH(ΛJ), where ΛJ ⊆ ∂Xv is the limit set of J and WCH(ΛJ) is the weak convex
hull of J [Swe01]. Since there are only finitely many Γ orbits of such J, there exists a
constant A1 ≥ 0 such that dHaus (WCH(ΛJ), J) ≤ A1 for all J ∈ Jv .
Now suppose f : Xv → Xw is a (K, A)-quasi-isometry such that dHaus (f (J), J 0 ) < ∞
for some J ∈ Jv and J 0 ∈ Jw . Then f induces a homeomorphism ∂f : ∂Xv → ∂Xw
such that ∂f (ΛJ) = ΛJ 0 . The Extended Morse Lemma for δ-hyperbolic spaces implies
there is a constant A2 = A2 (K, A, X) such that dHaus (f (WCH(ΛJ)), WCH(ΛJ 0 )) ≤ A2 ,
hence dHaus (f (J), J 0 ) ≤ KA1 + A + A1 + A2 =: B as required.
We can now use Theorem 11.24 to prove the second hypotheses of Proposition 11.2.
Proposition 11.27. If Γ is as in Theorem 11.16, then Γ has the uniform quasi-isometry

property.
Proof. Since Γ is quasi-isometric to the space X, where (X, T ) is an associated Bass–

Serre tree of spaces, it is sufficient to show that X has the uniform quasi-isometry
property. Namely, we show there exist constants K and A such that every quasi-isometry
f : X → X is close to a (K, A)-quasi-isometry. Note that the optimal quasi-isometry
constants of f will typically be much larger than K and A.
Our strategy will be to use the induced map f∗ : T → T to construct a (K, A)-quasi-
isometry fˆ : X → X close to f . We first use Theorem 11.24 to show that for each
vertex v ∈ V T , there are pattern-preserving quasi-isometries fˆv : Xv → Xf∗ (v) whose
quasi-isometry constants are independent of f . Then, we glue these quasi-isometries of
vertex spaces together, using tameness of edge spaces, to yield a quasi-isometry fˆ.
We fix a quasi-isometry f : X → X. By Proposition 11.19, the tree of spaces (X, T ) is
preserved by quasi-isometries. Let f∗ : T → T be a corresponding tree automorphism as
in Corollary 11.12. Then, for every w ∈ V T ∪ ET , there is a quasi-isometry fw : Xw →
Xf∗ (w) such that supx∈Xw d(f (x), fw (x)) < ∞ by [FM00, Lemma 2.1]. Proposition 11.19
implies f coarsely preserves edge spaces of X, so for every vertex v ∈ V T , fv sends
elements of Jv to Jf∗ (v) up to finite Hausdorff distance. Applying a similar argument
to a coarse inverse f¯ of f yields a coarse inverse f¯v of fv that sends elements of Jf∗ (v)
to Jv up to finite Hausdorff distance. Thus, by Lemma 11.26, the map fv is a pattern-
preserving quasi-isometry (Xv , Jv ) → (Xf∗ (v) , Jf∗ (v) ).
Therefore, for each v ∈ V T , the composition hf∗ (v) fv hv is a pattern-preserving quasi-
isometry (Hn , Jˆv ) → (Hn , Jˆf∗ (v) ), where n = nv = nf∗ (v) and hv : Xv → Hnv is de-
fined above. Theorem 11.24 implies hf∗ (v) fv hv is close to an isometry φv : Hn →
Hn . Thus, there exist constants K1 ≥ 1 and A1 ≥ 0, which are independent of v,
such that the map fˆv := hf∗ (v) φv hv : Xv → Xf∗ (v) is a (K1 , A1 )-quasi-isometry and
supx∈Xv d(fv (x), fˆv (x)) < ∞. Crucially, K1 and A1 depend only on X and not on the
quasi-isometry constants of the original quasi-isometry f .
It follows from Lemma 11.26 that there is a constant B, depending only on X, such
that for each v ∈ V T and e ∈ lk(v),
dHaus (fˆv (φe (Xe )), φf∗ (e) (Xf∗ (e) )) ≤ B,
where φe : Xe → Xv is the edge map. Another application of [FM00, Lemma 2.1]
implies that for every v ∈ V T and e ∈ lk(v), there exists a (K2 , A2 )-quasi-isometry
fˆv,e : Xe → Xf∗ (e) such that
sup d φf∗ (e) (fˆv,e (x)), fˆv (φe (x)) ≤ B 0 ,

x∈Xe
where K2 , A2 and B 0 depend only on X. Since supx∈Xv d(fˆv (x), f (x)) < ∞, it follows
that supx∈Xe d(fˆv,e (x), f (x)) < ∞.
Now, for each e ∈ ET , we have two (K2 , A2 )-quasi-isometries fˆι(e),e , fˆτ (e),e : Xe →
Xf∗ (e) such that supx∈Xe d(fˆι(e),e (x), fˆτ (e),e (x)) < ∞. Since Xe is a non-elementary
hyperbolic space, it is tame; see Example 2.30. Hence there is a constant A3 =
A3 (K2 , A2 , X) such that supx∈Xe d(fˆι(e),e (x), fˆτ (e),e (x)) ≤ A3 . Again, it is crucial to
note all these constants depend only on X and not f . It now follows from [CM17,
Proposition 2.14] that the quasi-isometries {fˆv }v∈V T can be glued together with con-
trolled error to yield a (K4 , A4 )-quasi-isometry fˆ : X → X, where K4 and A4 depend
only on X, such that supx∈X d(fˆ(x), f (x)) < ∞.
Proof of Theorem 11.16. Since Γ is hyperbolic, it is tame by Example 2.30. The result
is thus a combination of Propositions 11.2, 11.20 and 11.27.
Provided an analogue of Theorem 11.24 holds, the proof of Theorem 11.16 goes
through virtually unchanged for uniform lattices in more general rank one symmet-
ric spaces. For quaternionic hyperbolic spaces and the Cayley hyperbolic plane, the
required statement follows from Pansu’s much stronger quasi-isometric rigidity theorem
[Pan89], where the conclusion of Theorem 11.24 holds without the pattern. For complex
hyperbolic spaces, Biswas–Mj prove an analogue of Theorem 11.24 provided the pattern
J arises as the translates of odd-dimensional Poincaré duality groups [BM12]. Applying
these results and following the proof of Theorem 11.16 gives the following:
Theorem 11.28. Let Γ be the fundamental group of a minimal finite graph of groups
(1) Each vertex group of G is cubulated and is a uniform lattice in a rank one sym-
metric space other than H2 .
almost malnormal family of non-elementary quasi-convex subgroups with virtual
cohomological codimension at least two. Moreover, for each vertex group that
is a lattice in complex hyperbolic space, all its incident edge groups are odd-
dimensional Poincaré duality groups.
References
[AFW15] Matthias Aschenbrenner, Stefan Friedl, and Henry Wilton. 3-manifold groups. EMS Series
of Lectures in Mathematics. European Mathematical Society (EMS), Zürich, 2015.
[Ago13] Ian Agol. The virtual Haken conjecture. Doc. Math., 18:1045–1087, 2013. With an appendix
by Agol, Daniel Groves, and Jason Manning.
[BF92] M. Bestvina and M. Feighn. A combination theorem for negatively curved groups. J. Dif-
ferential Geom., 35(1):85–101, 1992.
[BFS20] Uri Bader, Alex Furman, and Roman Sauer. Lattice envelopes. Duke Math. J., 169(2):213–
278, 2020.
[BH99] Martin R. Bridson and André Haefliger. Metric spaces of non-positive curvature, volume
319 of Grundlehren der Mathematischen Wissenschaften [Fundamental Principles of Math-
ematical Sciences]. Springer-Verlag, Berlin, 1999.
[BHW11] Nicolas Bergeron, Frédéric Haglund, and Daniel T. Wise. Hyperplane sections in arithmetic
hyperbolic manifolds. J. Lond. Math. Soc. (2), 83(2):431–448, 2011.
[Bis12] Kingshook Biswas. Flows, fixed points and rigidity for Kleinian groups. Geom. Funct. Anal.,
22(3):588–607, 2012.
[BKMM12] Jason Behrstock, Bruce Kleiner, Yair Minsky, and Lee Mosher. Geometry and rigidity of
mapping class groups. Geom. Topol., 16(2):781–888, 2012.
[BL89] George M. Bergman and Hendrik W. Lenstra, Jr. Subgroups close to normal subgroups. J.
Algebra, 127(1):80–97, 1989.
[BM00] Marc Burger and Shahar Mozes. Lattices in product of trees. Inst. Hautes Études Sci. Publ.
Math., (92):151–194 (2001), 2000.
[BM12] Kingshook Biswas and Mahan Mj. Pattern rigidity in hyperbolic spaces: duality and PD
subgroups. Groups Geom. Dyn., 6(1):97–123, 2012.
[Bow98] Brian H. Bowditch. Cut points and canonical splittings of hyperbolic groups. Acta Math.,
180(2):145–186, 1998.
[BP00] Marc Bourdon and Hervé Pajot. Rigidity of quasi-isometries for some hyperbolic buildings.
Comment. Math. Helv., 75(4):701–736, 2000.
[BW12] Nicolas Bergeron and Daniel T. Wise. A boundary criterion for cubulation. Amer. J. Math.,
134(3):843–859, 2012.
[CDH+ ] Eric Chesebro, Jason DeBlois, Neil Hoffman, Christian Millicap, Priyadip Mondal, and
William Worden. Dehn surgery and hyperbolic knot complements without hidden symme-
tries. arXiv:2009.14765; to appear IMRN.
[CdlH16] Yves Cornulier and Pierre de la Harpe. Metric geometry of locally compact groups, vol-
ume 25 of EMS Tracts in Mathematics. European Mathematical Society (EMS), Zürich,
2016. Winner of the 2016 EMS Monograph Award.
[CJ94] Andrew Casson and Douglas Jungreis. Convergence groups and Seifert fibered 3-manifolds.
Invent. Math., 118(3):441–456, 1994.
[CK15] Ionut Chifan and Yoshikata Kida. OE and W ∗ superrigidity results for actions by surface
braid groups. Proc. Lond. Math. Soc. (3), 111(6):1431–1470, 2015.
[CM12] Pierre-Emmanuel Caprace and Nicolas Monod. A lattice in more than two Kac-Moody
groups is arithmetic. Israel J. Math., 190:413–444, 2012.
[CM17] Christopher H. Cashen and Alexandre Martin. Quasi-isometries between groups with two-
ended splittings. Math. Proc. Cambridge Philos. Soc., 162(2):249–291, 2017.
[Cor15] Yves Cornulier. Commability and focal locally compact groups. Indiana Univ. Math. J.,
64(1):115–150, 2015.
[Cor18] Yves de Cornulier. On the quasi-isometric classification of locally compact groups. In New
directions in locally compact groups, volume 447 of London Math. Soc. Lecture Note Ser.,
pages 275–342. Cambridge Univ. Press, Cambridge, 2018.
[Dav98] Michael W. Davis. Buildings are CAT(0). In Geometry and cohomology in group theory
(Durham, 1994), volume 252 of London Math. Soc. Lecture Note Ser., pages 108–123. Cam-
bridge Univ. Press, Cambridge, 1998.
[DK18] Cornelia Druţu and Michael Kapovich. Geometric group theory, volume 63 of American
Mathematical Society Colloquium Publications. American Mathematical Society, Providence,
RI, 2018. With an appendix by Bogdan Nica.
[dlST19] Mikael de la Salle and Romain Tessera. Characterizing a vertex-transitive graph by a large
ball. Journal of Topology, 12(3):705–743, 2019.
[DT16] Kajal Das and Romain Tessera. Integrable measure equivalence and the central extension
of surface groups. Groups Geom. Dyn., 10(3):965–983, 2016.
[Dun85] M. J. Dunwoody. The accessibility of finitely presented groups. Invent. Math., 81(3):449–
457, 1985.
[Dym15] Tullia Dymarz. Envelopes of certain solvable groups. Comment. Math. Helv., 90(1):195–224,
2015.
[FM00] Benson Farb and Lee Mosher. On the asymptotic geometry of abelian-by-cyclic groups. Acta
Math., 184(2):145–202, 2000.
[FM02] B. Farb and L. Mosher. The geometry of surface-by-free groups. Geom. Funct. Anal.,
12(5):915–963, 2002.
[For22] Max Forester. Incommensurable lattices in baumslag-solitar complexes, 2022.
[Fri18] Roberto Frigerio. Quasi-isometric rigidity of piecewise geometric manifolds. In Handbook of
group actions. Vol. IV, volume 41 of Adv. Lect. Math. (ALM), pages 95–138. Int. Press,
Somerville, MA, 2018.
[Fur01] A. Furman. Mostow-Margulis rigidity with locally compact targets. Geom. Funct. Anal.,
11(1):30–59, 2001.
[Gab92] David Gabai. Convergence groups are Fuchsian groups. Ann. of Math. (2), 136(3):447–510,
1992.
[Ger92] S. M. Gersten. Bounded cocycles and combings of groups. Internat. J. Algebra Comput.,
2(3):307–326, 1992.
[GH21] Vincent Guirardel and Camille Horbez. Measure equivalence rigidity of Out(fn ), 2021.
[GHH08] Oliver Goodman, Damian Heard, and Craig Hodgson. Commensurators of cusped hyperbolic
manifolds. Experiment. Math., 17(3):283–306, 2008.
[GL17] Vincent Guirardel and Gilbert Levitt. JSJ decompositions of groups. Astérisque, (395):viii
+ 165, 2017.
[Gro81] Mikhael Gromov. Groups of polynomial growth and expanding maps. Inst. Hautes Études
Sci. Publ. Math., (53):53–73, 1981.
[Gro87] M. Gromov. Hyperbolic groups. In Essays in group theory, volume 8 of Math. Sci. Res. Inst.
Publ., pages 75–263. Springer, New York, 1987.
[Hem87] John Hempel. Residual finiteness for 3-manifolds. In Combinatorial group theory and topol-
ogy (Alta, Utah, 1984), volume 111 of Ann. of Math. Stud., pages 379–396. Princeton Univ.
Press, Princeton, NJ, 1987.
[HH20] Camille Horbez and Jingyin Huang. Boundary amenability and measure equivalence rigidity
among two-dimensional artin groups of hyperbolic type. arXiv preprint arXiv:2004.09325,
2020.
[HH22] Camille Horbez and Jingyin Huang. Measure equivalence rigidity among the higman groups.
arXiv preprint arXiv:2206.00884, 2022.
[HL] Peter Haı̈ssinsky and Cyril Lecuire. Quasi-isometric rigidity of three manifold groups.
arXiv:2005.06813.
[Hou74] C. H. Houghton. Ends of locally compact groups and their coset spaces. J. Austral. Math.
Soc., 17:274–284, 1974. Collection of articles dedicated to the memory of Hanna Neumann,
VII.
[HR63] Edwin Hewitt and Kenneth A. Ross. Abstract harmonic analysis. Vol. I: Structure of topolog-
ical groups. Integration theory, group representations. Die Grundlehren der mathematischen
Wissenschaften, Band 115. Academic Press, Inc., Publishers, New York; Springer-Verlag,
Berlin-Göttingen-Heidelberg, 1963.
[HW09] G. Christopher Hruska and Daniel T. Wise. Packing subgroups in relatively hyperbolic
groups. Geom. Topol., 13(4):1945–1988, 2009.
[JSV13] Abdelrazak Jmel, Ezzeddine Salhi, and Gioia Vago. Homeomorphisms of the Sierpinski
curve with periodic properties. Dyn. Syst., 28(2):203–213, 2013.
[Kap01] Ilya Kapovich. The combination theorem and quasiconvexity. Internat. J. Algebra Comput.,
11(2):185–216, 2001.
[Kid10] Yoshikata Kida. Measure equivalence rigidity of the mapping class group. Ann. of Math.
(2), 171(3):1851–1901, 2010.
[KK00] Michael Kapovich and Bruce Kleiner. Hyperbolic groups with low-dimensional boundary.
Ann. Sci. École Norm. Sup. (4), 33(5):647–669, 2000.
[KKL98] Michael Kapovich, Bruce Kleiner, and Bernhard Leeb. Quasi-isometries and the de Rham
decomposition. Topology, 37(6):1193–1211, 1998.
[KL97] Michael Kapovich and Bernhard Leeb. Quasi-isometries preserve the geometric decomposi-
tion of Haken manifolds. Invent. Math., 128(2):393–416, 1997.
[KL01] Bruce Kleiner and Bernhard Leeb. Groups quasi-isometric to symmetric spaces. Comm.
Anal. Geom., 9(2):239–260, 2001.
[KM08] Bernhard Krön and Rögnvaldur G. Möller. Analogues of Cayley graphs for topological
groups. Math. Z., 258(3):637–675, 2008.
[KM12] Jeremy Kahn and Vladimir Markovic. Immersing almost geodesic surfaces in a closed hy-
perbolic three manifold. Ann. of Math. (2), 175(3):1127–1190, 2012.
[Mal10] William Malone. Topics in geometric group theory. ProQuest LLC, Ann Arbor, MI, 2010.
Thesis (Ph.D.)–The University of Utah.
[Mar91] G. A. Margulis. Discrete subgroups of semisimple Lie groups, volume 17 of Ergebnisse der
Mathematik und ihrer Grenzgebiete (3) [Results in Mathematics and Related Areas (3)].
Springer-Verlag, Berlin, 1991.
[Mar22a] Alex Margolis. Discretisable quasi-actions i: Topological completions and hyperbolicity.
arXiv preprint arXiv:2207.04401, 2022.
[Mar22b] Alex Margolis. Model geometries of finitely generated groups. arXiv preprint
arXiv:2207.10509, 2022.
[Mj12] Mahan Mj. Pattern rigidity and the Hilbert-Smith conjecture. Geom. Topol., 16(2):1205–
1246, 2012.
[MMS] Jason F. Manning, Mahan Mj, and Michah Sageev. Cubulating surface-by-free groups.
arXiv:1908.03545.
[MO15] Ashot Minasyan and Denis Osin. Acylindrical hyperbolicity of groups acting on trees. Math.
Ann., 362(3-4):1055–1105, 2015.
[MO21] Alexandre Martin and Damian Osajda. A combination theorem for combinatorially non-
positively curved complexes of hyperbolic groups. Math. Proc. Cambridge Philos. Soc.,
170(3):445–477, 2021.
[Mos73] G. D. Mostow. Strong rigidity of locally symmetric spaces. Annals of Mathematics Studies,
No. 78. Princeton University Press, Princeton, N.J.; University of Tokyo Press, Tokyo, 1973.
[MSW02] Lee Mosher, Michah Sageev, and Kevin Whyte. Maximally symmetric trees. volume 92,
pages 195–233. 2002. Dedicated to John Stallings on the occasion of his 65th birthday.
[MSW11] Lee Mosher, Michah Sageev, and Kevin Whyte. Quasi-actions on trees II: Finite depth
Bass-Serre trees. Mem. Amer. Math. Soc., 214(1008):vi+105, 2011.
[MZ55] Deane Montgomery and Leo Zippin. Topological transformation groups. Interscience Pub-
lishers, New York-London, 1955.
[Neu97] Walter D. Neumann. Commensurability and virtual fibration for graph manifolds. Topology,
36(2):355–378, 1997.
[Osi07] Denis V. Osin. Peripheral fillings of relatively hyperbolic groups. Invent. Math., 167(2):295–
326, 2007.
[Osi16] D. Osin. Acylindrically hyperbolic groups. Trans. Amer. Math. Soc., 368(2):851–888, 2016.
[Pan89] Pierre Pansu. Métriques de Carnot-Carathéodory et quasiisométries des espaces symétriques
de rang un. Ann. of Math. (2), 129(1):1–60, 1989.
[Pap05] Panos Papasoglu. Quasi-isometry invariance of group splittings. Ann. of Math. (2),
161(2):759–830, 2005.
[Par13] John Pardon. The Hilbert-Smith conjecture for three-manifolds. J. Amer. Math. Soc.,
26(3):879–899, 2013.
[PW02] Panos Papasoglu and Kevin Whyte. Quasi-isometries between groups with infinitely many
ends. Comment. Math. Helv., 77(1):133–144, 2002.
[Rag72] M. S. Raghunathan. Discrete subgroups of Lie groups. Ergebnisse der Mathematik und ihrer
Grenzgebiete, Band 68. Springer-Verlag, New York-Heidelberg, 1972.
[Rag84] M. S. Raghunathan. Torsion in cocompact lattices in coverings of Spin(2, n). Math. Ann.,
266(4):403–419, 1984.
[Rei91] Alan W. Reid. Arithmeticity of knot complements. J. London Math. Soc. (2), 43(1):171–184,
1991.
[Rie01] Eleanor G. Rieffel. Groups quasi-isometric to H2 ×R. J. London Math. Soc. (2), 64(1):44–60,
2001.
[Ron89] Mark Ronan. Lectures on buildings, volume 7 of Perspectives in Mathematics. Academic
Press, Inc., Boston, MA, 1989.
[Sch80] G. Schlichting. Operationen mit periodischen Stabilisatoren. Arch. Math. (Basel), 34(2):97–
99, 1980.
[Sch95] Richard Evan Schwartz. The quasi-isometry classification of rank one lattices. Inst. Hautes
Études Sci. Publ. Math., (82):133–168 (1996), 1995.
[Sch97] Richard Evan Schwartz. Symmetric patterns of geodesics and automorphisms of surface
groups. Invent. Math., 128(1):177–199, 1997.
[Ser77] Jean-Pierre Serre. Arbres, amalgames, SL2 . Société Mathématique de France, Paris, 1977.
Avec un sommaire anglais, Rédigé avec la collaboration de Hyman Bass, Astérisque, No.
46.
[She22] Sam Shepherd. Two generalisations of Leighton’s theorem. Groups Geom. Dyn., 16(3):743–
778, 2022. With an appendix by Giles Gardam and Daniel J. Woodhouse.
[She23] Sam Shepherd. Imitator homomorphisms for special cube complexes. Trans. Amer. Math.
Soc., 376(1):599–641, 2023.
[Sta68] John R. Stallings. On torsion-free groups with infinitely many ends. Ann. of Math. (2),
88:312–334, 1968.
[SW] Emily Stark and Daniel J. Woodhouse. Action rigidity for free products of hyperbolic man-
ifold groups. arXiv:1910.09609. To appear in Annales de l’Institut Fourier.
[SW79] Peter Scott and Terry Wall. Topological methods in group theory. In Homological group
theory (Proc. Sympos., Durham, 1977), volume 36 of London Math. Soc. Lecture Note Ser.,
pages 137–203. Cambridge Univ. Press, Cambridge-New York, 1979.
[Swe01] Eric L. Swenson. Quasi-convex groups of isometries of negatively curved spaces. Topology
Appl., 110(1):119–129, 2001. Geometric topology and geometric group theory (Milwaukee,
WI, 1997).
[Tao14] Terence Tao. Hilbert’s fifth problem and related topics, volume 153 of Graduate Studies in
Mathematics. American Mathematical Society, Providence, RI, 2014.
[Thu97] William P. Thurston. Three-dimensional geometry and topology. Vol. 1, volume 35 of Prince-
ton Mathematical Series. Princeton University Press, Princeton, NJ, 1997. Edited by Silvio
Levy.
[Tro84] V. I. Trofimov. Graphs with polynomial growth. Mat. Sb. (N.S.), 123(165)(3):407–421, 1984.
[Tuk88] Pekka Tukia. Homeomorphic conjugates of Fuchsian groups. J. Reine Angew. Math., 391:1–
54, 1988.
[VD36] D. Van Dantzig. Zur topologischen Algebra. III. Brouwersche und Cantorsche Gruppen.
Compositio Math., 3:408–426, 1936.
[vdDW84] L. van den Dries and A. J. Wilkie. Gromov’s theorem on groups of polynomial growth and
elementary logic. J. Algebra, 89(2):349–374, 1984.
[Wal11] Genevieve S. Walsh. Orbifolds and commensurability. In Interactions between hyperbolic
geometry, quantum topology and number theory, volume 541 of Contemp. Math., pages 221–
231. Amer. Math. Soc., Providence, RI, 2011.
[Why99] Kevin Whyte. Amenability, bi-Lipschitz equivalence, and the von Neumann conjecture. Duke
Math. J., 99(1):93–112, 1999.
[Why10] Kevin Whyte. Coarse bundles. arXiv preprint arXiv:1006.3347, 2010.
[Wis00] Daniel T. Wise. Subgroup separability of graphs of free groups with cyclic edge groups. Q.
J. Math., 51(1):107–129, 2000.
[Wis21] Daniel T. Wise. The structure of groups with a quasiconvex hierarchy, volume 209 of Annals
of Mathematics Studies. Princeton University Press, Princeton, NJ, [2021] ©2021.
[Wol68] Joseph A. Wolf. Growth of finitely generated solvable groups and curvature of Riemannian
manifolds. J. Differential Geometry, 2:421–446, 1968.
[WZ10] Henry Wilton and Pavel Zalesskii. Profinite properties of graph manifolds. Geom. Dedicata,
147:29–45, 2010.
Alex Margolis, Department of Mathematics, Vanderbilt University, 1326 Stevenson

Center, Nashville, TN 37240, USA
Email address: alexander.margolis@vanderbilt.edu
Sam Shepherd, Department of Mathematics, Vanderbilt University, 1326 Stevenson Cen-

ter, Nashville, TN 37240, USA
Email address: samuel.shepherd@vanderbilt.edu
Emily Stark, Department of Mathematics and Computer Science, Wesleyan University,

Science Tower 655, 265 Church Street, Middletown, CT 06459, USA
Email address: estark@wesleyan.edu
Daniel Woodhouse
Email address: mr.daniel.woodhouse@gmail.com

2303 04843

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2303 04843

Uploaded by

Copyright:

Available Formats

GRAPHICALLY DISCRETE GROUPS AND RIGIDITY

ALEX MARGOLIS, SAM SHEPHERD, EMILY STARK, AND DANIEL WOODHOUSE

Abstract. We introduce the notion of graphical discreteness to group theory. A

ties. We prove free products of one-ended virtually torsion-free graphically discrete

Date: March 10, 2023.

and Bader–Furman–Sauer [BFS20]. Dymarz [Dym15] classified lattice envelopes of cer-

Theorem A. (Theorem 10.4.) Let Γ be a virtually torsion-free finitely presented infinite-

The virtually torsion-free assumption in the theorem above is necessary: in Exam-

Corollary B. (Theorem 10.12.) The fundamental group of a (possibly reducible) closed

Work of Whyte and Papasoglu–Whyte demonstrates that infinite-ended groups pos-

Theorem C. (Theorem 11.8) Suppose Γ is the fundamental group of a minimal finite

By specializing to the case of hyperbolic n-manifold vertex groups, and adding an

Corollary E. Let Γ1 and Γ2 be fundamental groups of closed hyperbolic 3-manifolds,

1.2. Graphical discreteness. A vast family of well-studied groups can be shown to

Theorem F. (Theorem 7.9.) Let M be a closed irreducible, oriented 3-manifold with a

In Proposition 3.17 we give a general criterion to show an HNN extension is not

Theorem 1.9. (Theorem 3.10) A finitely generated group Γ is graphically discrete if

An analogous statement holds for non-elementary hyperbolic groups replacing QI(X)

1.5. Open questions. We make the following conjectures:

We show hyperbolic groups with boundary homeomorphic to an n-sphere for n ≤ 3 are

3.4. Examples of non-graphically-discrete groups 24

Definition 2.2. A 2-dimensional cell complex is a space obtained by attaching 2-cells

2.2. Coarse geometry.

Definition 2.3. Let X and Y be metric spaces. Let K ≥ 1 and A ≥ 0. A function

(2) (K, A)-quasi-isometry is a map f : X → Y such that:

(Γ0 ∩ ρ0−1 (H 0 ))/(Γ0 ∩ ρ0−1 (K 0 )) H 0 /K 0

Definition 2.12 ([CdlH16, §7]). A locally compact group G is compactly generated if

A compact normal subgroup is maximal if it is not a proper subgroup of any other

Although we are primarily focused with uniform lattices, results of Bader–Furman–

Proof. By Proposition 2.31, the subgroup K of Isom(Y ) consisting of bounded isometries

Corollary 2.33. Let X, Y , H, f and ρ be as in Lemma 2.32. In addition, assume that

The equivalences in the next definition follow from Proposition 2.34.

The following lemma follows immediately from definitions.

A group is said to be cubulated if it acts geometrically on a CAT(0) cube complex.

3. Introducing graphically discrete groups

Figure 1. The automorphism group of the graph X is not discrete. The

A topological group G is compact-by-discrete if G contains a compact normal subgroup

Figure 2. An example where Aut(X) acts discretely on X/ ∼, but

We now turn to the definition of graphical discreteness.

Proof. If ρ : Γ → G is virtual lattice embedding, then ρ|Γ0 is injective, hence a lattice

One useful consequence of graphical discreteness, foreshadowing many of the rigidity

Proof. Suppose Γ1 is graphically discrete and that ρi : Γi → Aut(X) is a geometric

3.2. Characterization of compact-by-discrete automorphism groups.

Remark 3.11. As X is locally finite, all vertex stabilizers Gx are commensurable by

3.3. Examples of graphically discrete groups.

(5) Finitely generated groups Γ containing a finite-index torsion-free subgroup with

Hign = ha1 , . . . , an | ai ai+1 a−1 2

for n ≥ 5 [HH22, Corollary 1.3].

3.4. Examples of non-graphically-discrete groups.

Proposition 3.17. Let Γ be a finitely generated group and let Λ ≤ Γ be a proper

Corollary 3.20. If Γ1 , Γ2 are nontrivial finitely generated groups containing nontrivial

Yet more examples of non-graphically-discrete infinite-ended groups are given by the

3.5. Uniform lattices and quasi-actions. In this subsection, we first summarize

Proof. Suppose ρ : Γ → G is a uniform virtual lattice embedding. By Remark 2.15

4. Graphical discreteness and non-discreteness for hyperbolic groups

discrete. We refer to [BH99] for background on boundaries of hyperbolic groups. If Γ

Theorem 4.1. A finitely generated non-elementary hyperbolic group Γ is graphically

Proof. Let X be a thick hyperbolic building. By definition, each chamber of X is

A JSJ decomposition of a finitely presented group is a graph of groups decomposition

4.2. Hyperbolic groups with sphere and Sierpinski carpet boundaries.

Theorem 4.5. If Γ is a hyperbolic group with boundary an n-sphere for n ≤ 3, then Γ

To prove this, we recall the Hilbert–Smith Conjecture:

Conjecture 4.6. If a locally compact group G acts faithfully and continuously by