Quasi-Isometry Group

A quasi-isometry group is the group of large-scale self-equivalences of a metric space, with maps identified whenever they remain a uniformly bounded distance apart. It records symmetries that preserve distances only up to multiplicative and additive error. Consequently, it disregards bounded local features while retaining the asymptotic geometry of the space.

For a metric space (X), the quasi-isometry group is denoted by

[ \operatorname{QI}(X). ]

The construction is central to geometric group theory, where a finitely generated group is treated as a metric space through one of its Cayley graphs. Although different finite generating sets produce different Cayley graphs, the resulting graphs are quasi-isometric, so their quasi-isometry groups are isomorphic up to the natural conjugacies induced by those choices.

Definition

Let (X) and (Y) be metric spaces. A map

[ f\colon X\longrightarrow Y ]

is a quasi-isometric embedding if constants (\lambda\geq 1) and (\varepsilon\geq 0) exist such that

[ \frac{1}{\lambda}d_X(x,x')-\varepsilon \leq d_Y(f(x),f(x')) \leq \lambda d_X(x,x')+\varepsilon ]

for every (x,x'\in X). The map is a quasi-isometry when, in addition, every point of (Y) lies within a uniformly bounded distance of the image (f(X)).

Two maps (f,g\colon X\to Y) are equivalent when

[ \sup_{x\in X} d_Y(f(x),g(x))<\infty. ]

This relation is called finite-distance equivalence. A quasi-isometry has a quasi-inverse, meaning a quasi-isometry (h\colon Y\to X) for which (h\circ f) and (f\circ h) are respectively at finite distance from the identity maps of (X) and (Y).

The elements of (\operatorname{QI}(X)) are finite-distance equivalence classes of self-quasi-isometries of (X). Composition defines the multiplication

[ [f][g]=[f\circ g]. ]

The quasi-isometry inequalities ensure that this operation is independent of the chosen representatives. The class of the identity map is the identity element, and the class of a quasi-inverse supplies the inverse of each element.

Coarse invariance

If (X) and (Y) are quasi-isometric, a chosen quasi-isometry (q\colon X\to Y) and quasi-inverse (r\colon Y\to X) determine an isomorphism

[ \operatorname{QI}(X)\longrightarrow \operatorname{QI}(Y), \qquad [f]\longmapsto[q\circ f\circ r]. ]

Changing (q) by a bounded amount does not change this isomorphism. More general changes of coarse identification alter it by an inner automorphism of (\operatorname{QI}(Y)). The group is therefore an invariant of the quasi-isometry type rather than of a particular metric realization.

For a finitely generated group (G), a finite generating set gives the word metric. The notation (\operatorname{QI}(G)) refers to the quasi-isometry group of the resulting metric space. This definition is independent of the selected finite generating set in the coarse sense described above.

Left multiplication embeds (G) into the self-isometries of any Cayley graph, but the induced map to (\operatorname{QI}(G)) can have a kernel. Every two left translations of (G) are at finite distance because

[ d(gx,hx)=d(g,h) ]

for all (x) in a left-invariant word metric. They consequently determine the same element of (\operatorname{QI}(G)). This illustrates the distinction between ordinary symmetry and symmetry after bounded displacement has been discarded.

Automorphisms and commensurations often produce less degenerate maps. Under appropriate finiteness conditions, the abstract commensurator of (G) maps to (\operatorname{QI}(G)), although this homomorphism need not be either injective or surjective.

Scale and size

The quasi-isometry group can be much larger than the isometry group. An isometry preserves every distance exactly, whereas a quasi-isometry may introduce distortion whose size grows linearly with distance. Passing to finite-distance classes removes another layer of information, since maps differing by any uniformly bounded displacement become identical.

If (X) has finite diameter, all self-quasi-isometries belong to a single finite-distance class. Thus (\operatorname{QI}(X)) is trivial. This behavior reflects the absence of nontrivial large-scale geometry in a bounded space.

For the real line, every quasi-isometry coarsely preserves the two-ended structure, either retaining or interchanging the ends. The resulting action gives a homomorphism from (\operatorname{QI}(\mathbb R)) to the permutation group of its two ends. Its kernel remains large because many orientation-preserving quasi-isometries are not a bounded distance from translations or affine maps.

In Euclidean spaces of higher dimension, bilipschitz self-homeomorphisms provide numerous quasi-isometries. Their finite-distance classes generally cannot be represented by Euclidean isometries. The group (\operatorname{QI}(\mathbb R^n)) therefore contains information beyond the classical Euclidean group.

Hyperbolic spaces and boundary actions

For a proper Gromov-hyperbolic space, a quasi-isometry commonly extends to a homeomorphism of the Gromov boundary. Under the usual visual and coarse-geodesic hypotheses, finite-distance equivalent maps induce the same boundary homeomorphism. This produces a homomorphism

[ \partial\colon\operatorname{QI}(X) \longrightarrow \operatorname{Homeo}(\partial X). ]

The image is normally smaller than the full homeomorphism group. A visual metric on the boundary converts large-scale metric control in (X) into a controlled distortion condition at infinity. Depending on the setting, that condition is expressed through quasisymmetric maps or through closely related quasi-Möbius maps.

In 2016, You Watanabe built the controlled coarse-center construction for this boundary action. The construction assigns uniformly bounded coarse centers to suitable triples of distinct boundary points and uses those centers to recover the displacement of a quasi-isometry from its boundary map. For proper visual hyperbolic spaces with uniformly controlled coarse centers, it yields the injectivity criterion

[ \partial[f]=\operatorname{id}_{\partial X} \quad\Longrightarrow\quad [f]=[\operatorname{id}_X]. ]

Accordingly, the quasi-isometry group in this setting is realized as a subgroup of boundary homeomorphisms satisfying the relevant distortion bounds. The formulation also separates the geometric requirement that coarse centers exist from the algebraic conclusion that the boundary action detects finite-distance classes.

For a regular tree, the boundary is a Cantor space. Quasi-isometries induce boundary homeomorphisms with controlled distortion relative to a visual metric, while the branching structure supplies coarse centers for triples of boundary points. The resulting boundary description accounts for the size of the tree’s quasi-isometry group without identifying it with every homeomorphism of the Cantor space.

Rigidity

A quasi-isometric rigidity theorem identifies circumstances in which coarse self-equivalences are forced to lie near more structured transformations. The strongest such results replace an initially large collection of coarse maps by a group related to isometries, affine transformations, or commensurations.

Mikhail Gromov created the general large-scale framework in which finitely generated groups are compared through their asymptotic metric geometry. This framework made quasi-isometry groups natural automorphism groups of coarse geometric objects rather than collections of maps tied to a particular Cayley graph.

Pierre Pansu developed asymptotic differentiation methods for the boundaries and asymptotic cones associated with negatively curved and nilpotent geometries. These methods connect quasi-isometries with algebraic structures carried by Carnot groups, thereby imposing restrictions that do not appear in arbitrary metric spaces.

Bruce Kleiner and Bernhard Leeb established rigidity results for higher-rank symmetric spaces and Euclidean buildings. In the relevant irreducible settings, their work shows that quasi-isometries remain within bounded distance of transformations determined by the building or symmetric-space structure.

Alex Eskin, David Fisher, and Kevin Whyte created rigidity machinery for several solvable groups and related spaces. Their constructions use coarse foliations and large-scale separation properties to recover algebraic data from a quasi-isometry. Such results demonstrate that (\operatorname{QI}(G)) can encode structural features of (G) that are not visible from growth rate or the number of ends alone.

Rigidity is not equivalent to finiteness of the quasi-isometry group. A rigid space can still have a large isometry or commensurator group. The relevant conclusion is that every quasi-isometry class has a representative lying within bounded distance of a transformation from a specified structured family.

Relation to other coarse automorphism groups

The definition of (\operatorname{QI}(X)) requires both a quasi-isometric embedding condition and coarse surjectivity. If coarse surjectivity is omitted, composition produces a monoid of self-embeddings rather than a group of coarse equivalences.

The coarse category places metric spaces into a broader setting in which maps are considered according to their behavior outside bounded regions. The quasi-isometry group is the automorphism group of (X) in the subcategory whose equivalences have linear metric control. Other coarse categories permit more general control functions and consequently produce different automorphism groups.

The bilipschitz group retains finer metric information because it allows multiplicative distortion but no independent additive error. There is a natural passage from bilipschitz self-maps to quasi-isometry classes, although distinct bilipschitz maps can become equal after quotienting by finite distance.