Free Groups and Group Presentations

Free groups are groups with no relations among their generators beyond those forced by the group axioms; every group is a quotient of a free group. A group presentation

⟨ X ∣ R ⟩

specifies a group by generators

X

and relations

R

, encoding it as a free group modulo the relations. They apply throughout combinatorial group theory, topology (fundamental groups), and algebra. The foundations were laid by Walther von Dyck (1882) and developed by Nielsen, Schreier, and others in the early 20th century.

A free group $F (X)$ on a set $X$ consists of all reduced words — finite strings of symbols from $X$ and their formal inverses, with no adjacent cancelling pairs — under concatenation followed by reduction. It is "free" because the only equalities holding among its elements are those mandated by the group axioms; no generator satisfies any unexpected relation. Free groups matter because they are the universal groups: any function from $X$ into a group $G$ extends uniquely to a homomorphism $F (X) \to G$ , making free groups the building blocks from which all groups arise via presentations. A surprising fact is that the free group on two generators contains, as subgroups, free groups of every finite and countably infinite rank — a single object houses copies of infinitely many "larger" free groups.

The Core Insight

Start with an alphabet $X$ . Allow yourself to write down every word you can, refusing to assume any coincidences. The only simplifications permitted are $x x^{- 1} = e$ and $x^{- 1} x = e$ . The resulting group is as "large" and "unconstrained" as possible on $X$ . To build any specific group, you then impose relations — equations declaring certain words equal to the identity. This two-step recipe (free group, then quotient by relations) describes every group whatsoever.

Formal Statement

Words, Reduction, and the Free Group

Let $X$ be a set (the alphabet or generating set). Form the set of symbols $X^{\pm} = X \cup X^{- 1}$ , where $X^{- 1} = {x^{- 1} : x \in X}$ is a disjoint copy. A word is a finite sequence

w = x_{1}^{ϵ_{1}} x_{2}^{ϵ_{2}} \dots x_{n}^{ϵ_{n}}, x_{i} \in X, ϵ_{i} \in {+ 1, - 1},

⟳a word is just an element of the free monoid on

X^{\pm}

;

n \geq 0

, with

n = 0

giving the empty word

e

A word is reduced if it contains no adjacent pair $x x^{- 1}$ or $x^{- 1} x$ . The free group $F (X)$ is the set of all reduced words, with the product of $u$ and $v$ defined by concatenating and then cancelling cancelling pairs until reduced:

u \cdot v = reduce (uv) .

⟳well-definedness requires that reduction yields a unique result — guaranteed by the Diamond Lemma / van der Waerden's trick below.

The rank of $F (X)$ is the cardinality $∣ X ∣$ . We write $F_{n}$ for the free group of rank $n$ , and $F_{\infty}$ for countably infinite rank.

Group Presentation

Given a set $X$ and a set $R$ of words in $X^{\pm}$ (the relators), the presentation $⟨ X ∣ R ⟩$ denotes the group

⟨ X ∣ R ⟩ = F (X) / N (R),

⟳

N (R)

is the smallest normal subgroup of

F (X)

containing

R

— the

normal closure, formed by taking all conjugates $w r^{\pm 1} w^{- 1}$ of relators and all their products.

The normal closure is explicitly

N (R) = {\prod_{i = 1}^{k} w_{i} r_{i}^{δ_{i}} w_{i}^{- 1} : w_{i} \in F (X), r_{i} \in R, δ_{i} = \pm 1, k \geq 0} .

⟳this set is closed under product, inverse, and conjugation, and any normal subgroup containing

R

must contain all such elements; hence it is the intersection of all normal subgroups containing

R

A group $G$ is finitely generated if it has a presentation with $∣ X ∣ < \infty$ , and finitely presented if both $∣ X ∣, ∣ R ∣ < \infty$ . We often write relations as equations: a relator $r = ab a^{- 1} b^{- 1}$ corresponds to the relation $ab = ba$ .

Universal Property of Free Groups

Let $ι : X ↪ F (X)$ be the inclusion of generators. For every group $G$ and every function $f : X \to G$ , there exists a unique group homomorphism $\overset{ˉ}{f} : F (X) \to G$ with $\overset{ˉ}{f} \circ ι = f$ :

\overset{ˉ}{f} (x_{1}^{ϵ_{1}} \dots x_{n}^{ϵ_{n}}) = f (x_{1})^{ϵ_{1}} \dots f (x_{n})^{ϵ_{n}} .

⟳define

\overset{ˉ}{f}

on words by applying

f

symbol-by-symbol; reduction is respected because

f (x) f (x)^{- 1} = e

G

; uniqueness follows since

X

generates

F (X)

This property characterizes $F (X)$ up to unique isomorphism, and is the defining feature of "freeness" in category theory: $F$ is left adjoint to the forgetful functor $Grp \to Set$ .

Universal Property of Presentations — von Dyck's Theorem

Let $G = ⟨ X ∣ R ⟩$ with quotient map $π : F (X) \to G$ . If $H$ is any group and $f : X \to H$ a function such that $\overset{ˉ}{f} (r) = e_{H}$ for every $r \in R$ (i.e., the relations hold in $H$ ), then $f$ induces a unique homomorphism $ϕ : G \to H$ with $ϕ (π (x)) = f (x)$ :

ϕ (π (w)) = \overset{ˉ}{f} (w), w \in F (X) .

⟳

\overset{ˉ}{f}

kills

R

, hence kills

N (R) = ker π

, so

\overset{ˉ}{f}

factors through

F (X) / N (R) = G

by the first isomorphism theorem.

Why It Holds / Derivation

Reduction is Well-Defined (van der Waerden's Trick)

We must show that the group operation [2] is associative and that reduced words form a group. The subtle point is that a word may be reducible in several orders; we need the reduced form to be unique.

Step 1 — Define an action on reduced words. Let

W

be the set of reduced words. For each symbol

a \in X^{\pm}

, define a map

σ_{a} : W \to W

σ_{a} (x_{1} \dots x_{n}) = {x_{2} \dots x_{n} a x_{1} \dots x_{n} if x_{1} = a^{- 1}, otherwise .

⟳prepending

a

either cancels a leading

a^{- 1}

or extends the word, always producing a reduced word.

Step 2 — Each $σ_{a}$ is a bijection with inverse

σ_{a^{- 1}}

: one checks

σ_{a^{- 1}} σ_{a} = id_{W}

directly by cases. Hence each

σ_{a}

lies in the symmetric group

Sym (W)

. Step 3 — Map words to permutations. Extend to any word

w = a_{1} \dots a_{n}

Σ_{w} = σ_{a_{1}} \circ \dots \circ σ_{a_{n}}

. Then

Σ_{w}

depends only on the reduced form of

w

, because inserting a cancelling pair

a a^{- 1}

inserts

σ_{a} σ_{a^{- 1}} = id

, which changes nothing. Step 4 — Recover the word. Applying

Σ_{w}

to the empty word

e

returns the reduced form of

w

Σ_{w} (e) = reduce (w) .

⟳build up

w

symbol by symbol from the right; each

σ_{a_{i}}

prepends or cancels exactly as reduction would.

Step 5 — Conclude. The map

w \mapsto Σ_{w}

sends reduced words injectively into

Sym (W)

(by Step 4) and is a homomorphism for concatenation. Hence reduced words inherit a well-defined, associative product, and form a group: the identity is

e

and the inverse of

x_{1} \dots x_{n}

x_{n}^{- 1} \dots x_{1}^{- 1}

■

Every Group Has a Presentation

Let $G$ be any group. Choose $X = G$ as the generating set (or any generating set). The universal property [5] gives a surjective homomorphism $π : F (X) ↠ G$ (surjective because $X$ generates $G$ ). Let $R$ be a set of words generating $ker π$ as a normal subgroup; e.g. take $R = ker π$ itself. By the first isomorphism theorem,

G ≅ F (X) / ker π = ⟨ X ∣ R ⟩ .

⟳surjectivity gives

im π = G

; quotienting

F (X)

by the kernel recovers

G

; choosing

R

to normally generate

ker π

realizes the kernel as

N (R)

Thus every group — finite or infinite — arises as a quotient of a free group. $■$

Walther von Dyck | 1856–1934

A student of Felix Klein, von Dyck introduced group presentations in his 1882 paper "Gruppentheoretische Studien." He was the first to study groups abstractly via generators and defining relations, rather than as concrete permutation or matrix groups, and proved the universal mapping property now bearing his name (Theorem [6]). His work bridged Klein's geometric program with the emerging abstract theory; the "von Dyck groups" $⟨ a, b ∣ a^{p}, b^{q}, (ab)^{r} ⟩$ governing tessellations of the sphere, plane, and hyperbolic plane carry his name. He also edited Kepler's and Galileo's collected works and founded the Deutsches Museum in Munich.

Building Intuition

Picture the free group $F_{2} = F ({a, b})$ as an infinite tree — specifically the Cayley graph, a 4-valent tree with no loops. Standing at the identity vertex, you may step in four directions: $a$ , $a^{- 1}$ , $b$ , $b^{- 1}$ . Each reduced word is a unique non-backtracking path from the root; "non-backtracking" is the reduced condition (don't immediately undo your last step). Because the graph is a tree, there is exactly one such path to each vertex — this is precisely why each group element equals exactly one reduced word.

A relation corresponds to declaring that some loop in this tree should actually close up. Imposing $ab = ba$ glues paths together so that "go $a$ then $b$ " lands at the same place as "go $b$ then $a$ ," collapsing the tree into the grid $Z^{2}$ . The more relations you add, the more the tree folds onto itself; the free group is the unfolded extreme where nothing is identified beyond the forced cancellations.

The Cayley graph of $F_2$ is an infinite 4-valent tree; reduced words are non-backtracking paths from the identity.

<svg viewBox="0 0 520 380" xmlns="http://www.w3.org/2000/svg">
<rect width="520" height="380" fill="#0a0e17"/>
<!-- edges from origin -->
<g stroke="#94a3b8" stroke-width="2" fill="none">
<line x1="260" y1="190" x2="380" y2="190"/>
<line x1="260" y1="190" x2="140" y2="190"/>
<line x1="260" y1="190" x2="260" y2="90"/>
<line x1="260" y1="190" x2="260" y2="290"/>
<!-- second level (shorter, thinner suggestion) -->
<line x1="380" y1="190" x2="450" y2="190"/>
<line x1="380" y1="190" x2="380" y2="135"/>
<line x1="380" y1="190" x2="380" y2="245"/>
<line x1="140" y1="190" x2="70" y2="190"/>
<line x1="140" y1="190" x2="140" y2="135"/>
<line x1="140" y1="190" x2="140" y2="245"/>
<line x1="260" y1="90" x2="205" y2="90"/>
<line x1="260" y1="90" x2="315" y2="90"/>
<line x1="260" y1="90" x2="260" y2="30"/>
<line x1="260" y1="290" x2="205" y2="290"/>
<line x1="260" y1="290" x2="315" y2="290"/>
<line x1="260" y1="290" x2="260" y2="350"/>
</g>
<!-- highlighted reduced path: e -> a -> ab -->
<g stroke="#d4a853" stroke-width="3.5" fill="none">
<line x1="260" y1="190" x2="380" y2="190"/>
<line x1="380" y1="190" x2="380" y2="135"/>
</g>
<!-- vertices -->
<g fill="#0a0e17" stroke="#94a3b8" stroke-width="2">
<circle cx="260" cy="190" r="7"/>
<circle cx="380" cy="190" r="5"/>
<circle cx="140" cy="190" r="5"/>
<circle cx="260" cy="90" r="5"/>
<circle cx="260" cy="290" r="5"/>
</g>
<circle cx="380" cy="135" r="5" fill="#0a0e17" stroke="#d4a853" stroke-width="2"/>
<!-- labels -->
<g fill="#e2e8f0" font-family="serif" font-size="15" font-style="italic">
<text x="248" y="215">e</text>
<text x="388" y="208">a</text>
<text x="105" y="208">a⁻¹</text>
<text x="268" y="80">b</text>
<text x="262" y="312">b⁻¹</text>
<text x="388" y="128" fill="#d4a853">ab</text>
</g>
<text x="20" y="30" fill="#d4a853" font-family="serif" font-size="14">gold: reduced path e → a → ab</text>
</svg>

Worked Examples

Computing in $\langle a, b \mid a^2,\, b^3,\, (ab)^2 \rangle$

Given: The presentation

G = ⟨ a, b ∣ a^{2}, b^{3}, (ab)^{2} ⟩

. We claim

G ≅ S_{3}

, the symmetric group on 3 letters (order 6). Step 1 — Read the relations as equations.

a^{2} = e, b^{3} = e, (ab)^{2} = e \Rightarrow abab = e .

⟳a relator

r

being trivial means

r = e

G

Step 2 — Derive a commutation rule. From

abab = e

and

a^{2} = e

(so

a^{- 1} = a

abab = e \Rightarrow ba = (ab)^{- 1} = b^{- 1} a^{- 1} = b^{- 1} a = b^{2} a .

⟳invert

ab

using

(x y)^{- 1} = y^{- 1} x^{- 1}

; then

a^{- 1} = a

b^{- 1} = b^{2}

Step 3 — Enumerate normal forms. Every element can be pushed to the form

b^{i} a^{j}

with

i \in {0, 1, 2}

j \in {0, 1}

, using

ab = b^{2} a

to move

a

's rightward:

G = {e, b, b^{2}, a, ba, b^{2} a}, ∣ G ∣ \leq 6.

⟳any word reduces to

b^{i} a^{j}

; the relations cap the count at

3 \times 2 = 6

Step 4 — Realize in $S_{3}$ . Map

a \mapsto (12)

b \mapsto (123)

. Check:

(12)^{2} = e

(123)^{3} = e

, and

((12) (123))^{2} = (23)^{2} = e

. All relations hold, so von Dyck's theorem gives a surjection

G ↠ S_{3}

, hence

∣ G ∣ \geq 6

. Answer:

G ≅ S_{3}, ∣ G ∣ = 6

The Free Abelianization

Given: Compute the abelianization of the free group

F_{2} = ⟨ a, b ⟩

, i.e.

F_{2} / [F_{2}, F_{2}]

. Step 1 — Add commutator relations. The commutator subgroup is normally generated by

[a, b] = ab a^{- 1} b^{- 1}

. Adjoining this relator:

F_{2}^{ab} = ⟨ a, b ∣ ab a^{- 1} b^{- 1} ⟩ = ⟨ a, b ∣ ab = ba ⟩ .

⟳setting the commutator to

e

forces

a, b

to commute.

Step 2 — Recognize the structure. Once

a, b

commute, every word equals

a^{m} b^{n}

uniquely:

F_{2}^{ab} ≅ Z^{2}, a^{m} b^{n} \mapsto (m, n) .

⟳commuting generators of infinite order give the free abelian group of rank 2; verify the map is a bijective homomorphism.

Step 3 — General rank. The same argument shows

F_{n}^{ab} ≅ Z^{n}

. Since

Z^{m} ≅ Z^{n}

iff

m = n

, this proves rank is an invariant: Answer:

F_{n}^{ab} ≅ Z^{n}, so F_{m} ≅ F_{n} ⟺ m = n

Common Misconceptions

Relations are not unique, and "no relations" ≠ "no equalities"

A common error is to think a group has a canonical presentation. It does not: $⟨ a ∣ a^{6} ⟩$ , $⟨ x, y ∣ x^{2}, y^{3}, x y x^{- 1} y^{- 1} ⟩$ , and others can all present $Z /6 Z$ . Different presentations of the same group are related by Tietze transformations (adding/removing redundant generators or relators).

A second error: students say free groups have "no relations." More precisely, free groups have no relations beyond those required by the group axioms. The equation $a a^{- 1} = e$ certainly holds in $F ({a})$ — that is forced. "Free" means no extra relations, not no equalities whatsoever.

Subgroups of free groups can have larger rank

One might guess that a subgroup of $F_{n}$ has rank $\leq n$ . False. The commutator subgroup $[F_{2}, F_{2}]$ of $F_{2}$ is itself a free group of infinite rank. By the Nielsen–Schreier theorem, every subgroup of a free group is free, but the rank can balloon: a subgroup of index $k$ in $F_{n}$ is free of rank

rank = k (n - 1) + 1,

⟳this is the Schreier index formula, derivable from Euler characteristic

χ

multiplicativity for covering spaces of a wedge of

n

circles.

So a finite-index subgroup of $F_{2}$ ( $n = 2$ ) of index $k$ has rank $k + 1$ , which grows without bound. Free groups are infinitely "self-similar."

Connections

Free groups and presentations sit at a crossroads of algebra, topology, and logic.

Topology: By the Seifert–van Kampen Theorem, the fundamental group of a wedge of $n$ circles is exactly $F_{n}$ , and a CW-complex with one 0-cell, generators as 1-cells, and relators as 2-cells has fundamental group $⟨ X ∣ R ⟩$ . Presentations are 2-complexes.
Universal algebra: The free-forgetful adjunction (Theorem [5]) is the prototype of Free Objects in a Category; free monoids, free abelian groups, and free modules all follow the same pattern.
Quotient structure: Presentations are an instance of the First Isomorphism Theorem applied to $F (X) \to G$ , with relators normally generating the kernel.
Geometric group theory: Cayley Graphs turn group elements into vertices and generators into edges, making free groups into trees and giving a geometric notion of group structure.
Combinatorics: The Word Problem for Groups asks whether a given word equals the identity; for free groups it is trivially solvable by reduction, but for general presentations it can be undecidable.

Real-World Applications

Free groups and presentations are working tools across mathematics and beyond:

Topology — fundamental groups. Computing $π_{1}$ of surfaces and knot complements yields explicit presentations; e.g. the trefoil knot group is $⟨ x, y ∣ x^{2} = y^{3} ⟩$ , used to prove knots are distinct.
Cryptography. Non-abelian group-based protocols (e.g. the Anshel–Anshel–Goldfeld key exchange) use braid groups, given by presentations, where the difficulty of the conjugacy search problem provides security.
Crystallography & tilings. Von Dyck and triangle groups $⟨ a, b ∣ a^{p}, b^{q}, (ab)^{r} ⟩$ classify the 17 wallpaper groups and hyperbolic tessellations, directly informing the analysis of crystal symmetry.
Robotics & motion planning. The fundamental groups of configuration spaces (often free or free-products) encode whether a robotic path can be continuously deformed past obstacles.
Computer algebra systems. GAP and Magma implement the Todd–Coxeter coset enumeration algorithm to compute the order and structure of finitely presented groups automatically.
Formal language theory. The reduction of words to normal form connects to rewriting systems and confluent term-rewriting (Knuth–Bendix completion).
Physics — anyons & topological computing. Braid group presentations describe the statistics of anyonic quasiparticles, underpinning proposals for fault-tolerant quantum computation.

Historical Development

1856 (Hamilton): The icosian calculus gives an early concrete presentation-like description of the symmetry group of the icosahedron via generators and relations.
1878 (Cayley): Arthur Cayley introduces the idea of representing a group by its multiplication structure and graphs (later Cayley graphs), and abstractly defines groups.
1882 (von Dyck): Walther von Dyck formalizes generators and defining relations, proving the universal property and effectively defining free groups as the relation-free case.
1908 (Tietze): Heinrich Tietze introduces Tietze transformations, showing any two finite presentations of the same group are connected by elementary moves, and poses the isomorphism problem.
1911 (Dehn): Max Dehn formulates the three fundamental decision problems — the word, conjugacy, and isomorphism problems — and solves the word problem for surface groups.
1921–1927 (Nielsen, Schreier): Jakob Nielsen and Otto Schreier prove that every subgroup of a free group is free, with Schreier supplying the index formula [15].
1936 (Todd, Coxeter): The Todd–Coxeter coset enumeration algorithm provides a systematic method to compute finite quotients from presentations.
1955 (Novikov), 1958 (Boone): Pyotr Novikov and (independently) William Boone prove the word problem is undecidable for general finitely presented groups — a landmark linking group theory and computability.
1987 (Gromov): Mikhael Gromov's theory of hyperbolic groups recasts free groups as the simplest hyperbolic groups, launching modern geometric group theory.

Subtleties & Limits

The transition from a presentation to the group it presents hides deep undecidability. The word problem (does a given word equal $e$ ?), conjugacy problem, and isomorphism problem are all undecidable for general finitely presented groups (Novikov–Boone, Adyan–Rabin). Thus there is no algorithm that, given two presentations, decides whether they present isomorphic groups — even deciding whether a presented group is trivial is undecidable.

Open and active areas include: the Andrews–Curtis conjecture (can every balanced presentation of the trivial group be reduced to the empty one by elementary moves and stabilizations? — widely believed false but unresolved); the precise classification of which groups are hyperbolic or automatic; and the profinite rigidity question of which finitely presented groups are determined by their finite quotients. The geometry of "random" presentations (Gromov's random groups) remains a fertile research frontier.