Chapter 10
Hard

Free Groups & Presentations

00 · Symbol Glossary

$S$A generating set (just a set, not yet a group)

An arbitrary set of symbols — think of them as letters of an alphabet — that will be used to build a group from nothing but formal combinations of themselves.

$F(S)$Free group on S

The group of all reduced words in SS and formal inverses of SS, under concatenation. The "most unconstrained" possible group containing SS — no relation holds among its elements except what the group axioms force.

$s^{-1}$Formal inverse symbol

For each sSs\in S, a new formal symbol s1s^{-1}, distinct from ss itself, destined to become ss's actual group inverse once F(S)F(S) is built.

$\langle S \mid R\rangle$Group presentation

Read "the group generated by SS subject to relations RR." Shorthand for F(S)/NF(S)/N, where NN is the smallest normal subgroup containing every word in RR.

$w=1$A relation

A single equation, written as a relator word ww set equal to the identity. "aba1b1=1aba^{-1}b^{-1}=1" is the relation forcing aa and bb to commute, since it rearranges to ab=baab=ba.


01 · Building a Group From Nothing but Symbols

Every group so far has been built from already-existing mathematical objects: numbers, permutations, matrices. This chapter runs the process in reverse: start with a bare set of symbols and manufacture the least constrained group that could possibly contain them.

Construction — The Free Group F(S)

Given a set SS, form a new set of symbols S1={s1:sS}S^{-1} = \{s^{-1} : s\in S\} (one formal inverse per generator). A word is a finite string of symbols from SS1S\cup S^{-1}. A word is reduced if it contains no adjacent pair ss1s\,s^{-1} or s1ss^{-1}s (which would obviously "cancel" once these symbols become genuine group inverses). The free group F(S)F(S) is the set of all reduced words (including the empty word, playing the role of ee), with operation: concatenate two words, then repeatedly cancel any adjacent inverse pairs until the result is reduced again.

Well-definedness is a genuine theorem, taken as given here

Checking that this operation is associative — that reducing "eagerly" during concatenation never depends on which adjacent pair you cancel first — is a real theorem (classically proved via van der Waerden's trick: represent each generator as a permutation of the set of all reduced words, and verify associativity there instead, where it's automatic). We take this construction as given and spend this chapter's effort on what makes it useful.

Example — Reducing a word in $F(\{a,b\})$

The word aba1ab1bab a^{-1} a b^{-1} b reduces: the middle a1aa^{-1}a cancels, giving abb1bab\,b^{-1}b; then b1bb^{-1}b cancels, giving abab. The fully reduced form is abab — a genuinely different (shorter) element of F({a,b})F(\{a,b\}).

Common mistake — Free groups are almost never abelian

In F({a,b})F(\{a,b\}), is ab=baab=ba? As reduced words, abab and baba are different strings of symbols, with no cancellation available to relate them — they are genuinely different elements. Unless SS has at most one element, F(S)F(S) is always nonabelian: nothing in the construction forces any two distinct generators to commute, and the "least constrained" philosophy of free groups means nothing is ever assumed beyond the bare group axioms.


02 · The Universal Property

The reduced-word construction is a means to an end. What actually characterizes F(S)F(S) — and what makes it useful without ever thinking about cancellation again — is a single mapping property.

Theorem 10.1 — Universal Property of Free Groups

Let ι:SF(S)\iota: S\to F(S) send each generator to the corresponding length-1 word. For any group GG and any function f:SGf: S\to G (assigning an arbitrary target element to each generator, with no constraints), there exists a unique homomorphism φ:F(S)G\varphi: F(S)\to G with φ(ι(s))=f(s)\varphi(\iota(s)) = f(s) for every sSs\in S.

Proof

Existence. For a reduced word w=s1ϵ1s2ϵ2skϵkw = s_1^{\epsilon_1}s_2^{\epsilon_2}\cdots s_k^{\epsilon_k} (each siSs_i\in S, each ϵi=±1\epsilon_i=\pm1), define:

φ(w):=f(s1)ϵ1f(s2)ϵ2f(sk)ϵkG\varphi(w) := f(s_1)^{\epsilon_1}f(s_2)^{\epsilon_2}\cdots f(s_k)^{\epsilon_k} \in G

(with φ(empty word)=eG\varphi(\text{empty word}) = e_G). This is a legitimate function of ww, since ww is already given in reduced form. To see φ\varphi is a homomorphism, first extend the same formula to unreduced concatenations of words — it clearly satisfies φ(uv)=φ(u)φ(v)\varphi(uv)=\varphi(u)\varphi(v) for the naive (unreduced) concatenation, by simply combining the two symbol lists. Cancellation of an adjacent pair ss1s\,s^{-1} (or s1ss^{-1}s) inside a word changes nothing about φ\varphi's value, since f(s)f(s)1=eGf(s)f(s)^{-1}=e_G (respectively f(s)1f(s)=eGf(s)^{-1}f(s)=e_G) contributes nothing to the product. So φ\varphi takes the same value on a word before and after reduction, and φ(w1w2)=φ(w1)φ(w2)\varphi(w_1\ast w_2) = \varphi(w_1)\varphi(w_2) holds for F(S)F(S)'s actual operation \ast (concatenate, then reduce) as well.

Uniqueness. Suppose ψ:F(S)G\psi: F(S)\to G is any homomorphism with ψ(ι(s))=f(s)\psi(\iota(s))=f(s) for all sSs\in S. For a reduced word w=s1ϵ1skϵkw=s_1^{\epsilon_1}\cdots s_k^{\epsilon_k}, viewed as the product ι(s1)ϵ1ι(sk)ϵk\iota(s_1)^{\epsilon_1}\cdots\iota(s_k)^{\epsilon_k} inside F(S)F(S), the homomorphism property (repeatedly applied) together with Theorem 6.1(b) (homomorphisms preserve inverses, handling the ϵi=1\epsilon_i=-1 cases) forces:

ψ(w)=ψ(ι(s1))ϵ1ψ(ι(sk))ϵk=f(s1)ϵ1f(sk)ϵk=φ(w)\psi(w) = \psi(\iota(s_1))^{\epsilon_1}\cdots\psi(\iota(s_k))^{\epsilon_k} = f(s_1)^{\epsilon_1}\cdots f(s_k)^{\epsilon_k} = \varphi(w)

So ψ=φ\psi=\varphi on every element of F(S)F(S): the extension is forced, hence unique.

Why 'free' — no relation is ever assumed

The universal property says exactly this: to define a homomorphism out of F(S)F(S), you may send the generators absolutely anywhere in the target group, with zero constraints, and the rest of the homomorphism is then completely determined. No other group has this much freedom — any group with even one nontrivial relation among its generators would restrict which target assignments are allowed. This is the precise sense in which F(S)F(S) is the "least constrained" group containing SS.


03 · Presentations

Most groups are not free — they satisfy extra relations (rn=er^n=e in Zn\mathbb{Z}_n, srs=r1srs=r^{-1} in DnD_n). A presentation builds exactly such a group by starting from F(S)F(S) and forcing chosen words to become the identity.

Definition — Presentation

Given a set SS and a set of words RF(S)R\subseteq F(S) (called relators), let NF(S)N\trianglelefteq F(S) be the smallest normal subgroup of F(S)F(S) containing every element of RR (called the normal closure of RR — the subgroup generated by all conjugates gwg1gwg^{-1} for wRw\in R, gF(S)g\in F(S); a normal subgroup, by construction, since it's closed under conjugation by every element of F(S)F(S)). The presented group is:

SR:=F(S)/N\langle S \mid R\rangle := F(S)/N

Example — Recovering Z_n as a presentation

aan\langle a \mid a^n\rangle: here S={a}S=\{a\}, R={an}R=\{a^n\}. Since F({a})F(\{a\}) is already abelian (a single generator has nothing to fail to commute with), every subgroup is normal (Chapter 05), so N=anN=\langle a^n\rangle (the ordinary cyclic subgroup generated by ana^n inside F({a})ZF(\{a\})\cong\mathbb{Z}). The quotient F({a})/NZ/nZ=ZnF(\{a\})/N \cong \mathbb{Z}/n\mathbb{Z} = \mathbb{Z}_n — a presentation recovering exactly the group from Chapter 01, built entirely from the single relation "aa has order (dividing) nn."

Example — A presentation for $D_n$

r,srn,s2,(sr)2\langle r,s \mid r^n,\, s^2,\, (sr)^2\rangle. The relator (sr)2=srsr(sr)^2 = srsr set equal to the identity rearranges (multiply both sides on the right by r1s1r^{-1}s^{-1}, using s1=ss^{-1}=s since s2=es^2=e) to exactly srs=r1srs=r^{-1} — the relation identified geometrically back in Chapter 02, Section 05.

Example — Forcing commutativity with a relator

a,baba1b1\langle a,b \mid aba^{-1}b^{-1}\rangle: the single relator aba1b1aba^{-1}b^{-1} set to the identity rearranges to ab=baab=ba — forcing exactly one relation, that aa and bb commute, and nothing more. This presented group turns out to be Z×Z\mathbb{Z}\times\mathbb{Z} (Chapter 07): free on two commuting generators, with no bound on either generator's order.


04 · The Mapping Property of Presentations

Theorem 10.2 — Von Dyck's Theorem

Suppose GG is generated by elements {gs}sS\{g_s\}_{s\in S} (one chosen target per generator), and every relator wRw\in R becomes the identity of GG when each ss is replaced by gsg_s throughout ww. Then there is a surjective homomorphism φˉ:SRG\bar\varphi: \langle S\mid R\rangle \to G sending sNgssN \mapsto g_s.

Proof

By Theorem 10.1 (universal property), the function f:SGf: S\to G, f(s)=gsf(s)=g_s, extends to a unique homomorphism φ:F(S)G\varphi: F(S)\to G. By hypothesis, every relator wRw\in R satisfies φ(w)=eG\varphi(w)=e_G (substituting gsg_s for each ss is exactly what φ\varphi computes), so RkerφR\subseteq\ker\varphi. By Theorem 6.3, kerφF(S)\ker\varphi\trianglelefteq F(S), and since NN is the smallest normal subgroup containing RR, NkerφN\subseteq\ker\varphi.

Define φˉ:F(S)/NG\bar\varphi: F(S)/N \to G by φˉ(wN)=φ(w)\bar\varphi(wN) = \varphi(w). Well-defined: if w1N=w2Nw_1N=w_2N, then w1w21Nkerφw_1w_2^{-1}\in N\subseteq\ker\varphi, so φ(w1)=φ(w2)\varphi(w_1)=\varphi(w_2) (the same style of argument used in Theorem 6.5's First Isomorphism Theorem proof). Homomorphism: inherited directly from φ\varphi. Surjective: since {gs}\{g_s\} generate GG, every element of GG is a product of the gsg_s and their inverses, and φˉ(sN)=φ(s)=gs\bar\varphi(sN)=\varphi(s)=g_s hits each generator, so products of sNsN's under φˉ\bar\varphi hit every element of GG.

Example — D_n really is the group its presentation describes

The genuine rotation r0r_0 and reflection s0s_0 in DnD_n satisfy r0n=er_0^n=e, s02=es_0^2=e, (s0r0)2=e(s_0r_0)^2=e (Chapter 02). By Theorem 10.2, there is a surjection φˉ:r,srn,s2,(sr)2Dn\bar\varphi: \langle r,s\mid r^n,s^2,(sr)^2\rangle \to D_n.

Von Dyck's Theorem only guarantees a surjection, not automatically an isomorphism — a presentation could, in principle, describe a strictly larger group that merely maps onto the target. Showing the surjection above is actually injective requires a separate counting argument: every reduced word in r,s\langle r,s\mid \ldots\rangle can be rewritten, using only the three relations, into the form risjr^is^j with 0i<n0\le i<n, j{0,1}j\in\{0,1\} — giving at most 2n2n distinct elements in the presented group. Since φˉ\bar\varphi is a surjection onto DnD_n (which has exactly 2n2n elements), the presented group has at least 2n2n elements too. Combining both bounds forces exactly 2n2n, and a surjection between finite sets of equal size is a bijection: φˉ\bar\varphi is an isomorphism, confirming Dnr,srn,s2,(sr)2D_n \cong \langle r,s\mid r^n,s^2,(sr)^2\rangle exactly, not merely as a quotient.

Common mistake — Assuming every presentation's surjection is automatically injective

Von Dyck's Theorem by itself never proves a presentation equals a specific target group — only that the target is some quotient of the presented group. Confirming an isomorphism (as in the DnD_n example) always requires an independent argument bounding the presented group's size from above, matched against the target's known size. Skipping this step is one of the most common errors when working with presentations.


05 · Exercises

EXERCISE 10.1

Concatenate the two words directly, then repeatedly cancel any adjacent symbol-inverse pairs until no more cancellation is possible.

Concatenating aba1aba^{-1} and ab1a1ab^{-1}a^{-1} gives the string aba1ab1a1aba^{-1}ab^{-1}a^{-1}. The middle a1aa^{-1}a cancels: abb1a1ab\,b^{-1}a^{-1}. Then bb1b\,b^{-1} cancels: aa1aa^{-1}. Then aa1a\,a^{-1} cancels: the empty word.

The product is the identity — meaning ab1a1ab^{-1}a^{-1} is precisely the inverse of aba1aba^{-1} in F({a,b})F(\{a,b\}), consistent with Theorem 2.3's general formula (xyz)1=z1y1x1(xyz)^{-1}=z^{-1}y^{-1}x^{-1} applied to x=a,y=b,z=a1x=a,y=b,z=a^{-1}.

In F({a,b})F(\{a,b\}), compute the reduced form of (aba1)(ab1a1)(aba^{-1})(ab^{-1}a^{-1}).

EXERCISE 10.2

Apply the universal property directly: since F({a,b})F(\{a,b\}) places no constraints on where generators go, just specify images and read off what φ\varphi must do to a word.

By Theorem 10.1, define f(a)=(1101)f(a)=\begin{pmatrix}1&1\\0&1\end{pmatrix}, f(b)=(1001)f(b)=\begin{pmatrix}1&0\\0&-1\end{pmatrix} (any two elements of GL2(R)GL_2(\mathbb{R}), since the universal property imposes no constraints). This extends uniquely to a homomorphism φ:F({a,b})GL2(R)\varphi: F(\{a,b\}) \to GL_2(\mathbb{R}) with φ(a)=f(a)\varphi(a)=f(a), φ(b)=f(b)\varphi(b)=f(b), and, e.g., φ(a2b1)=f(a)2f(b)1\varphi(a^2b^{-1}) = f(a)^2f(b)^{-1}, computed by ordinary matrix multiplication.

Using Theorem 10.1, explain how to define a homomorphism from F({a,b})F(\{a,b\}) into GL2(R)GL_2(\mathbb{R}), and state what φ(a2b1)\varphi(a^2b^{-1}) must equal.

EXERCISE 10.3

Rearrange the relator a2a^2 set to the identity: what does a2=ea^2=e say about the order of aa?

aa2\langle a \mid a^2\rangle: the relation a2=ea^2=e means aa has order dividing 2. Since F({a})ZF(\{a\})\cong\mathbb{Z} (Chapter 03's infinite cyclic group), quotienting by a2\langle a^2\rangle gives Z/2Z=Z2\mathbb{Z}/2\mathbb{Z} = \mathbb{Z}_2 — the same style of computation as the Zn\mathbb{Z}_n example in Section 03, with n=2n=2.

aa2Z2\langle a \mid a^2\rangle \cong \mathbb{Z}_2.

Identify the group aa2\langle a \mid a^2\rangle.

EXERCISE 10.4

Apply Von Dyck's Theorem: does Z4\mathbb{Z}_4 have generators satisfying the two given relations?

Try g=1Z4g=1\in\mathbb{Z}_4 (additive notation, so "g4g^4" means 41=404\cdot1=4\equiv0, and "g2g^2" means 21=22\cdot1=2). Check: g4=0=eZ4g^4=0=e_{\mathbb{Z}_4} ✓ (satisfies the first relator). But g2=20g^2=2\neq0fails the second relator a2=ea^2=e.

Since Z4\mathbb{Z}_4 requires an element of order exactly 4 to be generated by a single element (Chapter 03), and any element satisfying both a4=ea^4=e and a2=ea^2=e has order dividing gcd(4,2)=2\gcd(4,2)=2, no generator of Z4\mathbb{Z}_4 can satisfy both relations simultaneously. Von Dyck's Theorem requires the relations to hold for the chosen generating elements — here, no valid choice exists, so the theorem does not produce a surjection onto Z4\mathbb{Z}_4 for this presentation. (In fact aa4,a2Z2\langle a\mid a^4,a^2\rangle \cong \mathbb{Z}_2, since a2=ea^2=e already forces order dividing 2, making a4=ea^4=e automatic and redundant.)

Does Von Dyck's Theorem give a surjection from aa4,a2\langle a \mid a^4, a^2\rangle onto Z4\mathbb{Z}_4 (using a1a\mapsto1)? Explain why or why not.


06 · Chapter Summary

ConceptStatement
Free group F(S)F(S)Reduced words in SS1S\cup S^{-1} under concatenation-and-cancel
Universal propertyAny f:SGf:S\to G extends to a unique homomorphism F(S)GF(S)\to G (Thm 10.1)
Presentation SR\langle S\mid R\rangleF(S)/NF(S)/N, where NN = normal closure of the relators RR
Von Dyck's TheoremGenerators of GG satisfying RR give a surjection SRG\langle S\mid R\rangle \twoheadrightarrow G (Thm 10.2)
Presentation vs. isomorphismVon Dyck gives only a surjection; equality needs an independent size bound
Zn\mathbb{Z}_naan\langle a\mid a^n\rangle
DnD_nr,srn,s2,(sr)2\langle r,s\mid r^n,s^2,(sr)^2\rangle, confirmed exact via a counting argument
Z×Z\mathbb{Z}\times\mathbb{Z}a,baba1b1\langle a,b\mid aba^{-1}b^{-1}\rangle

Next: Chapter 11 — Rings begins the second major arc of the course: structures with two interacting operations, addition and multiplication, generalizing Z\mathbb{Z} the way groups generalized a single operation.