Algebra

Geometric group theory

Turns finitely generated groups into geometric objects — Cayley graphs and word metrics, studied up to quasi-isometry — powerful enough to prove Gromov's classification of groups of polynomial growth and to define hyperbolic groups.

IntuitionTurning a group into a shape you can walk around

Imagine a group whose only "moves" are a short list of allowed multiplications, applied over and over starting from a home position. Every finitely generated group secretly has this shape: pick a set of generators $S$ and draw the Cayley graph — one vertex for every group element, and an edge from $g$ to $gs$ for each generator $s \in S$ . An abstract algebraic object turns into a concrete picture you can draw and walk on. The word metric $d_S(g,h)$ is the fewest generators you need to multiply together to travel from $g$ to $h$ — exactly the length of the shortest path between them in the Cayley graph. Translating between algebra and geometry, and back again, is the whole idea of geometric group theory.

Example: The Cayley graph of $\mathbb{Z}/6\mathbb{Z}$ : a hexagon you can draw by hand

Take the cyclic group $\mathbb{Z}/6\mathbb{Z} = \{0,1,2,3,4,5\}$ under addition mod $6$ , with the generating set $S = \{1,5\}$ (that is, $+1$ and its inverse $-1 \equiv 5$ ). Draw the six elements as points and connect $g$ to $g+1$ and to $g-1$ for every $g$ . What shape appears, and what is the word-metric distance $d_S(0,3)$ ?

Solution

Every vertex $g$ connects only to its two neighbors $g \pm 1 \pmod 6$ , so the Cayley graph is exactly a hexagon (a $6$ -cycle) — six vertices arranged in a ring, matching the everyday picture of clock arithmetic. To reach $3$ from $0$ , walk clockwise ( $0 \to 1 \to 2 \to 3$ , length $3$ ) or counterclockwise ( $0 \to 5 \to 4 \to 3$ , also length $3$ ); no shorter route exists because $3$ sits exactly halfway around the hexagon. So $d_S(0,3) = 3$ , and in general $d_S(0,k) = \min(k,\, 6-k)$ .

UndergraduateFormalizing the picture: Cayley graphs and the word metric

Definition: Cayley graph

Let $G$ be a group and $S \subseteq G$ a symmetric generating set ( $S = S^{-1}$ , $e \notin S$ ). The Cayley graph $\mathrm{Cay}(G,S)$ has vertex set $G$ , with an edge joining $g$ and $gs$ for every $g \in G$ and $s \in S$ . Choosing a different finite generating set changes the picture up close, but — as we will see — never the coarse geometry.

Definition: Word metric

For $g,h \in G$ , the word metric $d_S(g,h)$ is the length of the shortest word in $S$ spelling $g^{-1}h$ :

d_S(g,h) = \min\{\, n : g^{-1}h = s_1 s_2 \cdots s_n,\ s_i \in S \,\}

This is exactly the graph distance in $\mathrm{Cay}(G,S)$ : $d_S(g,h)$ counts the edges of the shortest path from $g$ to $h$ . The metric is left-invariant, $d_S(kg,kh) = d_S(g,h)$ for every $k \in G$ , because left multiplication by $k$ permutes the vertices of $\mathrm{Cay}(G,S)$ while preserving every edge.

A finite group has finitely many possible Cayley graphs, but an infinite group's Cayley graph looks quite different for different generating sets — draw $\mathbb{Z}$ with $S = \{1\}$ and you get a bi-infinite line; draw it with $S = \{2,3\}$ and the local picture changes completely, yet from far away it still "looks like" the same line. Quasi-isometry makes "the same from far away" precise, and is the fundamental equivalence relation of geometric group theory.

Definition: Quasi-isometry

A map $f : X \to Y$ between metric spaces is a ** $(\lambda,\varepsilon)$ -quasi-isometric embedding** ( $\lambda \ge 1$ , $\varepsilon \ge 0$ ) if for all $x,y \in X$ :

\tfrac{1}{\lambda}\, d_X(x,y) - \varepsilon \;\le\; d_Y(f(x),f(y)) \;\le\; \lambda\, d_X(x,y) + \varepsilon

It is a quasi-isometry if in addition every point of $Y$ lies within distance $\varepsilon$ of the image $f(X)$ (the image is coarsely dense). Any two finite generating sets $S, S'$ of the same group $G$ give quasi-isometric Cayley graphs — the identity map on $G$ already works, for a suitable $\lambda$ — so every finitely generated group has a well-defined coarse geometry, independent of the chosen generators. This coarse geometry, not any single Cayley graph, is what geometric group theory actually studies.

UndergraduateBridging back to metric spaces: the Milnor–Švarc lemma

Theorem: Milnor–Švarc lemma

Let $G$ act by isometries on a proper, geodesic metric space $X$ , properly discontinuously and cocompactly (the quotient $X/G$ is compact). Then $G$ is finitely generated, and for any basepoint $x_0 \in X$ , the orbit map $g \mapsto g\cdot x_0$ is a quasi-isometry from $G$ , equipped with a word metric, to $X$ .

Why is it true?

Because $G$ acts by isometries, it cannot tell points of $X$ apart from any of their $G$ -translates; because the action is cocompact, one bounded piece of $X$ , copied by $G$ , already covers all of $X$ . So the orbit of a single point already captures the entire coarse shape of $X$ — studying the abstract group $G$ and studying the concrete space $X$ it acts on become interchangeable up to bounded error. This is the theorem that lets metric-space geometry and group theory trade places.

Proof

Fix $x_0 \in X$ and, using compactness of $X/G$ , choose $R$ large enough that the $G$ -translates of the closed ball $\bar B(x_0,R)$ cover $X$ . Let $S = \{\, g \in G : g \ne e,\ d_X(x_0,gx_0) \le 2R+1 \,\}$ , a finite set by proper discontinuity. To see $S$ generates $G$ : given $g \in G$ , mark points $x_0 = g_0 x_0, g_1 x_0, \dots, g_n x_0 = g x_0$ spaced at most $2R$ apart along a geodesic from $x_0$ to $g x_0$ ; each consecutive pair satisfies $d_X(g_{i-1}x_0, g_i x_0) \le 2R$ , so $g_{i-1}^{-1} g_i \in S$ , and multiplying these $n \le d_X(x_0, g x_0)/(2R) + 1$ elements of $S$ recovers $g$ . Hence $d_S(e,g) \le C_1\, d_X(x_0,gx_0) + C_1$ for a constant $C_1$ depending only on $R$ . Conversely, each generator moves $x_0$ by at most $2R+1$ , so $d_X(x_0,gx_0) \le (2R+1)\, d_S(e,g)$ . These two inequalities show $g \mapsto g x_0$ is a $(\lambda,\varepsilon)$ -quasi-isometric embedding for suitable $\lambda,\varepsilon$ , and cocompactness (every point of $X$ lies within $R$ of some $G$ -translate of $x_0$ ) makes its image coarsely dense — so it is a quasi-isometry, and since it is defined on all of $G$ , the finite set $S$ generates $G$ .

Example: The Cayley graph of the free group $F_2$ is an infinite $4$ -regular tree

Let $F_2 = \langle a,b \rangle$ be the free group on two generators: its elements are exactly the reduced words in $\{a,a^{-1},b,b^{-1}\}$ (no letter immediately followed by its own inverse). Take $S = \{a,a^{-1},b,b^{-1}\}$ . Show that $\mathrm{Cay}(F_2,S)$ has no cycles and that every vertex has degree $4$ , then count the elements of word-length exactly $n \ge 1$ .

Solution

A walk from $e$ that never immediately backtracks spells out a reduced word, and distinct reduced words always name distinct elements of $F_2$ — no cancellation is possible partway through, so no two different such walks can ever meet again. Hence $\mathrm{Cay}(F_2,S)$ has no cycles: it is a tree. Every vertex $g$ has exactly the $4$ distinct neighbors $ga, ga^{-1}, gb, gb^{-1}$ (distinct because $F_2$ has no relations to identify any of them), so the tree is $4$ -regular.

Counting by length: the identity is the unique element of length $0$ . A reduced word of length $n \ge 1$ is built by choosing its first letter ( $4$ choices) and then each later letter ( $3$ choices, since it must avoid being the inverse of the letter just before it). So there are exactly $4\cdot 3^{n-1}$ elements of word-length exactly $n$ , and the ball of radius $n$ has $b_S(n) = 1 + \sum_{k=1}^{n} 4\cdot 3^{k-1} = 2\cdot 3^{n} - 1$ elements — exponential growth, unlike the bounded growth of $\mathbb{Z}/6\mathbb{Z}$ .

This finite binary tree is only an analogy for the branching, ever-splitting shape of a free group's Cayley graph — it is not a picture of

\mathrm{Cay}(F_2,\{a,a^{-1},b,b^{-1}\})

itself. The true Cayley graph of

F_2

is infinite, and every one of its vertices has degree

4

(four ways to extend a reduced word); this finite tree instead has a degree-

2

root and degree-

1

leaves. Use it only to feel the no-cycles, always-branching character of a free group's geometry, not its exact shape.

AdvancedGrowth: from counting spheres to Gromov's theorem

The growth function of $G$ with respect to a finite generating set $S$ counts the size of metric balls:

b_S(n) = |B_S(e,n)| = \big|\{\, g \in G : d_S(e,g) \le n \,\}\big|

$G$ has polynomial growth if $b_S(n) \le C n^d$ for constants $C,d$ and all $n$ , and exponential growth if $b_S(n) \ge c\,\alpha^n$ for some $\alpha > 1$ — neither property depends on which finite $S$ is chosen, only on the group $G$ itself. We have already met both extremes: $\mathbb{Z}/6\mathbb{Z}$ has bounded growth (the whole group is one ball), $\mathbb{Z}^k$ has polynomial growth of degree $k$ , and the free group $F_2$ has exponential growth, with $b_S(n) = 2\cdot 3^n - 1$ .

Theorem: Gromov's theorem on groups of polynomial growth

(Gromov, 1981) A finitely generated group $G$ has polynomial growth if and only if $G$ is virtually nilpotent, i.e. $G$ has a nilpotent subgroup of finite index.

Why is it true?

Growth is a purely metric, large-scale invariant — you just count how many group elements fit in balls of the word metric — while "nilpotent" is a purely algebraic condition on iterated commutators. Gromov's theorem says these two utterly different-looking worlds, geometry and algebra, describe exactly the same groups here: the moment you can bound how fast a group's Cayley-graph balls grow by a polynomial, hidden algebraic structure (a nilpotent subgroup of finite index) is forced to exist. It answered a 1968 question of Milnor and Wolf and became the founding landmark result of geometric group theory.

Proof

One direction is classical algebra with a geometric flavor (Bass–Guivarc'h): if $N \trianglelefteq G$ is nilpotent of finite index with lower central series of ranks $r_1,\dots,r_c$ , a counting argument on normal forms of elements shows $b_S(n) \asymp n^d$ with $d = \sum_i i\cdot r_i$ , a polynomial bound — so virtually nilpotent groups do have polynomial growth. The converse, due to Gromov, is genuinely deep, and only its strategy is sketched here: rescale the Cayley graph by $1/n$ and take a limit (in the Gromov–Hausdorff sense, along a subsequence and an ultrafilter) as $n \to \infty$ ; polynomial growth is exactly the condition that keeps these rescaled balls from collapsing to a point or blowing up without bound, so a limiting metric space — the asymptotic cone of $G$ — exists and is a finite-dimensional, locally compact, geodesic space on which $G$ still acts transitively at the level of the limit. Montgomery and Zippin's structure theory of locally compact groups then forces the isometry group of this limit space to contain a Lie group, and a further argument bounds the degree of polynomial growth in terms of the dimension of that Lie group, ultimately pinning down a nilpotent subgroup of finite index inside $G$ itself. A later proof by Kleiner (2010) reroutes this same rescale-and-take-a-limit strategy through the finite-dimensionality of spaces of polynomial-growth harmonic functions, avoiding the Montgomery–Zippin machinery entirely, but the guiding idea — pass to a limit, extract a Lie group, and read off nilpotency — remains the same.

AdvancedCurvature without smoothness: Gromov-hyperbolic groups

Gromov's second landmark contribution (1987) distills the essential large-scale feature of negatively curved spaces — like the hyperbolic plane $\mathbb{H}^2$ — into a definition that makes sense for any geodesic metric space, smooth or not: thin triangles.

Definition: Gromov-hyperbolic space and hyperbolic group

A geodesic metric space $X$ is $\delta$ -hyperbolic ( $\delta \ge 0$ ) if every side of every geodesic triangle lies in the $\delta$ -neighborhood of the union of the other two sides. A finitely generated group $G$ is (Gromov-)hyperbolic if its Cayley graph $\mathrm{Cay}(G,S)$ , for some (equivalently, any) finite generating set $S$ , is $\delta$ -hyperbolic for some $\delta$ .

\text{for every geodesic triangle } [x,y,z]:\quad [x,y] \subseteq N_\delta\big([y,z] \cup [z,x]\big)

Trees — like the free-group Cayley graph above — are $0$ -hyperbolic: a geodesic triangle in a tree is literally a tripod, so each side sits on the union of the other two. The hyperbolic plane $\mathbb{H}^2$ is $\delta$ -hyperbolic for a universal $\delta$ , and so is the fundamental group of any closed surface of genus at least $2$ . By contrast, $\mathbb{Z}^2$ is not hyperbolic: a large square has sides that stay far apart in the middle no matter how big $\delta$ is, so no single $\delta$ works as the square grows — flat, Euclidean directions are exactly what hyperbolicity rules out.

The matrix

\begin{pmatrix} 2 & 1 \\ 1 & 1 \end{pmatrix}

has determinant

1

, so it belongs to

SL(2,\mathbb{R})

; the widget shows how it linearly stretches the Euclidean plane

\mathbb{R}^2

along its eigendirections. Every matrix in

SL(2,\mathbb{R})

also acts as an isometry of the hyperbolic plane

\mathbb{H}^2

via the Möbius transformation

z \mapsto \frac{az+b}{cz+d}

; a discrete group generated by such matrices is a Fuchsian group, a genuine and central object of geometric group theory (the fundamental group of a hyperbolic surface is exactly such a group). Note: this widget draws only the ordinary linear action on

\mathbb{R}^2

shown here, not the hyperbolic-plane Möbius action itself.

AdvancedBridges: mapping class groups, CAT(0) geometry, 3-manifolds, and Lie groups

Geometric group theory's ideas radiate outward. The mapping class group of a surface $\Sigma_g$ — isotopy classes of its self-homeomorphisms — acts on Teichmüller space, and Masur–Minsky's curve-complex machinery gives it a rich coarse geometry in the same Cayley-graph spirit developed above. ** $\mathrm{CAT}(0)$ spaces** — geodesic spaces where triangles are "no fatter" than their Euclidean comparison triangles — generalize non-positive curvature the way Gromov-hyperbolicity generalizes strictly negative curvature; groups acting properly and cocompactly on $\mathrm{CAT}(0)$ cube complexes, following Sageev, Wise, and Agol, were the decisive geometric-group-theoretic tool behind the resolution of several of Thurston's conjectures about $3$ -manifolds. This is a genuine bridge to the Poincaré conjecture: the fundamental group of a closed hyperbolic $3$ -manifold acts geometrically (properly discontinuously and cocompactly by isometries) on $\mathbb{H}^3$ , so by the Milnor–Švarc lemma above it is quasi-isometric to $\mathbb{H}^3$ itself, and understanding exactly which groups arise this way is inseparable from the geometrization of $3$ -manifolds underlying Perelman's proof. On the algebraic side, lattices — discrete, cocompact or finite-covolume subgroups of Lie groups, such as $SL(n,\mathbb{Z}) \subset SL(n,\mathbb{R})$ — are precisely the groups the Milnor–Švarc lemma lets us study geometrically, via the symmetric space the Lie group acts on. This is a natural doorway to Lie groups and Lie algebras, where rigidity theorems of Mostow and Margulis show that for higher-rank lattices, the abstract group alone remembers the entire Lie-theoretic structure it came from.

Historical note

Max Dehn's 1911 paper Über unendliche diskontinuierliche Gruppen formulated the word, conjugacy, and isomorphism problems for finitely presented groups and solved the word problem for surface groups with an explicit, geometrically motivated algorithm — tiling the hyperbolic plane by a fundamental polygon and shortening any word that traces too much of the polygon's boundary. Henri Poincaré had already, in his 1895 Analysis Situs, introduced the fundamental group of a space via loops up to homotopy — the first time a group was extracted directly from the geometry of a space, the conceptual seed that geometric group theory would later grow from in the opposite direction, by giving abstract groups back their geometry. The field's modern shape, however, is usually dated to two papers by Mikhail Gromov: 'Groups of polynomial growth and expanding maps' (1981), proving that polynomial growth characterizes virtually nilpotent groups, and 'Hyperbolic groups' (1987), which introduced $\delta$ -hyperbolicity and launched geometric group theory as a distinct field.

Henri Poincaré

ResearchThe frontier: today's open questions

Research frontier as of 2026

Where the field stands (as of 2026). Two currents from this page are still actively moving.

1. The Cannon conjecture remains open. It predicts that every Gromov-hyperbolic group whose boundary at infinity is homeomorphic to the $2$ -sphere $S^2$ is (virtually) the fundamental group of a closed hyperbolic $3$ -manifold, i.e. acts geometrically on $\mathbb{H}^3$ — a group-theoretic converse to the geometrization picture behind the Poincaré conjecture. Vladimir Markovic gave an influential criterion for it in 2013, reducing the conjecture to a statement about quasi-convex surface subgroups separating boundary points, and research since has pushed hardest on the cubulated case: Corey Bregman and Merlin Incerti-Medici's 2024 paper 'On a generalization of Cannon's conjecture for cubulated hyperbolic groups' extends Markovic-style techniques to boundary spheres of dimension $3$ or at least $5$ . The original $2$ -sphere case is still unresolved as of 2026. 2. Free groups and mapping class groups keep revealing open structure. Whether $\mathrm{Out}(F_3)$ , the outer automorphism group of the rank- $3$ free group, embeds into a linear group is a well-known open problem, in sharp contrast to Formanek and Procesi's proof that $\mathrm{Out}(F_n)$ is not linear for $n \ge 4$ (via a 'poison subgroup' construction). In parallel, the geometry of the free factor complex and free splitting complex, introduced by Bestvina and Feighn to mirror the curve complex of a surface, and the classification of 'big' mapping class groups of infinite-type surfaces remain very active fronts connecting this page to low-dimensional topology and Teichmüller theory.

In the Cayley graph of $\mathbb{Z}/6\mathbb{Z}$ with generating set $\{1,5\}$ (i.e. $\pm 1$ ), what is the word-metric distance $d_S(0,3)$ ?

Which of the following is quasi-isometric to $\mathbb{Z}$ (with its usual metric)?

According to Gromov's 1981 theorem, a finitely generated group has polynomial growth if and only if it is...

Which of the following groups (with a standard finite generating set) is NOT Gromov-hyperbolic?

The Milnor–Švarc lemma says that if $G$ acts properly discontinuously and cocompactly by isometries on a proper geodesic metric space $X$ , then...

References

Clara Löh (2017). Geometric Group Theory: An Introduction · DOI:10.1007/978-3-319-72254-2
Mikhael Gromov (1981). Groups of polynomial growth and expanding maps · DOI:10.1007/BF02698687
Mikhael Gromov (1987). Hyperbolic groups