Mathematics · Book 5 · Bachelor Year 3

University Mathematics — Year 3

University Mathematics — Year 3 · Bachelor Year 3

1Group Theory

The Year 2 volume met groups as bookkeeping devices: Lagrange’s theorem, cyclic groups, the symmetric group and its signature. This chapter turns group theory into a method. The engine is the notion of a group acting on a set: counting orbits and fixed points yields the class equation, Cauchy’s theorem and the three Sylow theorems — the fundamental local-to-global principle of finite group theory. We then learn to assemble groups (direct and semidirect products) and to disassemble them (composition series, solvable groups), and we prove the theorem that, in Chapter 4, will close a three-century-old question about polynomial equations: the alternating group AnA_n is simple for n5n \geq 5.

1.1 Quotient groups and the isomorphism theorems

Throughout, GG is a group written multiplicatively, ee its identity. Recall from the Year 2 volume: subgroups, cosets gHgH, Lagrange’s theorem (G=[G:H]H\abs G = [G:H]\,\abs H for finite GG), the order of an element, cyclic groups, and the symmetric group SnS_n with its signature morphism ε ⁣:Sn{±1}\varepsilon \colon S_n \to \{\pm1\}.

Definition 1.1

A subgroup NN of GG is normal (written NGN \trianglelefteq G) when gNg1=NgNg^{-1} = N for every gGg \in G — equivalently, when left and right cosets coincide: gN=NggN = Ng for all gg.

Theorem 1.2 (Quotient group)

Let NGN \trianglelefteq G. The set G/NG/N of cosets, with the multiplication (gN)(hN)=ghN(gN)(hN) = ghN, is a well-defined group, the quotient group, and the canonical projection π ⁣:GG/N\pi \colon G \to G/N, ggNg \mapsto gN, is a surjective morphism with kernel NN. Conversely, every kernel of a group morphism is normal: normal subgroups are exactly the kernels.

Proof. Well-definedness is the whole point. If gN=gNgN = g'N and hN=hNhN = h'N, write g=gng' = gn, h=hmh' = hm with n,mNn, m \in N. Then gh=gnhm=gh(h1nh)mghNg'h' = gnhm = gh\,(h^{-1}nh)\,m \in ghN since h1nhNh^{-1}nh \in N by normality: the product of cosets does not depend on the representatives. Associativity, identity eN=NeN = N and inverses (gN)1=g1N(gN)^{-1} = g^{-1}N are inherited from GG. Clearly π\pi is a surjective morphism and π(g)=N    gN\pi(g) = N \iff g \in N.

If f ⁣:GHf \colon G \to H is a morphism and kkerfk \in \ker f, then f(gkg1)=f(g)f(k)f(g)1=ef(gkg^{-1}) = f(g)f(k)f(g)^{-1} = e: kernels are normal.

Theorem 1.3 (Universal property; first isomorphism theorem)

Let f ⁣:GHf \colon G \to H be a morphism and NGN \trianglelefteq G with NkerfN \subseteq \ker f. There is a unique morphism fˉ ⁣:G/NH\bar f \colon G/N \to H with f=fˉπf = \bar f \circ \pi. In particular, taking N=kerfN = \ker f:

G/kerf        imf,gNf(g).G/\ker f \;\xrightarrow{\;\sim\;}\; \operatorname{im} f, \qquad gN \mapsto f(g).

Proof. Uniqueness: fˉ(gN)\bar f(gN) must be f(g)f(g). Existence: if gN=gNgN = g'N then g1gNkerfg^{-1}g' \in N \subseteq \ker f, so f(g)=f(g)f(g) = f(g') and fˉ(gN)=f(g)\bar f(gN) = f(g) is well defined; it is a morphism because ff is. For N=kerfN = \ker f: fˉ\bar f is injective, since fˉ(gN)=e\bar f(gN) = e means gkerfg \in \ker f, i.e. gN=NgN = N; its image is that of ff.

Theorem 1.4 (Second and third isomorphism theorems)

Let HGH \leq G and NGN \trianglelefteq G.

  1. HN={hn:hH,nN}HN = \{hn : h \in H,\, n \in N\} is a subgroup, NHNN \trianglelefteq HN, HNHH \cap N \trianglelefteq H, and

    H/(HN)    HN/N.H/(H \cap N) \;\cong\; HN/N .
  2. If moreover NKGN \subseteq K \trianglelefteq G, then K/NG/NK/N \trianglelefteq G/N and (G/N)/(K/N)G/K(G/N)\big/(K/N) \cong G/K.

Proof. (1) HNHN is a subgroup: (hn)(hn)=hh(h1nh)nHN(hn)(h'n') = hh'\,(h'^{-1}nh')n' \in HN and (hn)1=h1(hn1h1)HN(hn)^{-1} = h^{-1}(hn^{-1}h^{-1}) \in HN, using normality of NN. Compose HHNπHN/NH \hookrightarrow HN \xrightarrow{\pi} HN/N: this morphism is surjective (hnN=hNhnN = hN) with kernel {hH:hN}=HN\{h \in H : h \in N\} = H \cap N; apply Theorem 1.3.

(2) The projection G/NG/KG/N \to G/K, gNgKgN \mapsto gK, is well defined (NKN \subseteq K), surjective, with kernel K/NK/N; apply Theorem 1.3 again.

Theorem 1.5 (Correspondence theorem)

Let NGN \trianglelefteq G. The map HH/NH \mapsto H/N is a bijection between subgroups of GG containing NN and subgroups of G/NG/N, preserving inclusions, indices and normality (in both directions).

Proof. Its inverse is Hˉπ1(Hˉ)\bar H \mapsto \pi^{-1}(\bar H). Both maps send subgroups to subgroups, and are mutually inverse: π1(H/N)=HN=H\pi^{-1}(H/N) = HN = H since NHN \subseteq H, and π(π1(Hˉ))=Hˉ\pi(\pi^{-1}(\bar H)) = \bar H by surjectivity of π\pi. Inclusions are clearly preserved; [G:H]=[G/N:H/N][G:H] = [G/N : H/N] because gH(gN)(H/N)gH \mapsto (gN)(H/N) is a well-defined bijection between coset spaces; and gHg1=HgHg^{-1} = H for all gg iff (gN)(H/N)(gN)1=H/N(gN)(H/N)(gN)^{-1} = H/N for all gNgN, again by surjectivity of π\pi.

Example 1.6

ε ⁣:Sn{±1}\varepsilon \colon S_n \to \{\pm 1\} gives Sn/An{±1}S_n/A_n \cong \{\pm1\}; det ⁣:GLn(K)K×\det \colon GL_n(K) \to K^\times gives GLn(K)/SLn(K)K×GL_n(K)/SL_n(K) \cong K^\times; te2iπtt \mapsto \eu^{2\iu\pi t} gives R/ZU\R/\Z \cong \mathbb U, the circle group. The first isomorphism theorem is how quotients are computed in practice: find a surjection with the right kernel.

Method 1.7

To prove NGN \trianglelefteq G, in decreasing order of elegance: exhibit NN as the kernel of a morphism defined on GG; check gNg1NgNg^{-1} \subseteq N for all gg (this suffices: applying it to g1g^{-1} and conjugating gives the reverse inclusion); verify that NN is a union of conjugacy classes; or note that [G:N]=2[G:N] = 2 (then gN=NggN = Ng is forced — Exercise 1.1).

1.2 Group actions

Definition 1.8

An action of GG on a set XX is a morphism φ ⁣:GS(X)\varphi \colon G \to \mathfrak{S}(X) into the group of bijections of XX; one writes gxg \cdot x for φ(g)(x)\varphi(g)(x). Equivalently: a map G×XXG \times X \to X with ex=xe \cdot x = x and g(hx)=(gh)xg \cdot (h \cdot x) = (gh) \cdot x. The orbit of xx is Ox={gx:gG}\mathcal O_x = \{g \cdot x : g \in G\}, its stabilizer is the subgroup Gx={g:gx=x}G_x = \{g : g\cdot x = x\}, and XG={x:g, gx=x}X^G = \{x : \forall g,\ g \cdot x = x\} is the set of fixed points. The action is transitive if there is exactly one orbit, faithful if φ\varphi is injective, free if all stabilizers are trivial.

Example 1.9

Five actions run all of finite group theory:

  1. GG on itself by left translation gx=gxg \cdot x = gx: free and transitive.
  2. GG on itself by conjugation gx=gxg1g \cdot x = gxg^{-1}: orbits are the conjugacy classes, stabilizers the centralizers ZG(x)={g:gx=xg}Z_G(x) = \{g : gx = xg\}, fixed points the center Z(G)Z(G).
  3. GG on the coset space G/HG/H by gxH=gxHg \cdot xH = gxH: transitive, with stabilizer of the coset HH equal to HH. Every transitive action is of this form (Exercise 1.8).
  4. GG on its set of subgroups by conjugation: the stabilizer of HH is the normalizer NG(H)={g:gHg1=H}N_G(H) = \{g : gHg^{-1} = H\}, the largest subgroup of GG in which HH is normal.
  5. SnS_n on [ ⁣[1,n] ⁣]\intint{1}{n}: the mother of all examples.

Theorem 1.10 (Orbit–stabilizer)

The map gGxgxgG_x \mapsto g \cdot x is a well-defined bijection G/GxOxG/G_x \to \mathcal O_x. In particular, for GG finite,

Ox=[G:Gx]divides G,\abs{\mathcal O_x} = [G : G_x] \quad\text{divides } \abs G,

and, the orbits partitioning XX (they are the classes of the equivalence xy    yOxx \sim y \iff y \in \mathcal O_x),

X=i[G:Gxi](xi ⁣:one point per orbit).\abs X = \sum_{i} \,[G : G_{x_i}] \qquad (x_i\colon \text{one point per orbit}).

Proof. Well defined and injective: gGx=hGx    h1gGx    h1gx=x    gx=hxgG_x = hG_x \iff h^{-1}g \in G_x \iff h^{-1}g \cdot x = x \iff g \cdot x = h \cdot x; read the chain in both directions. Surjectivity is the definition of the orbit. The counting statements follow from Lagrange’s theorem and the partition of XX into orbits.

Corollary 1.11 (Class equation)

For a finite group GG, picking one representative xix_i in each conjugacy class with more than one element:

G=Z(G)+i[G:ZG(xi)],each [G:ZG(xi)]>1 dividing G.\abs G = \abs{Z(G)} + \sum_i \,[G : Z_G(x_i)], \qquad\text{each } [G : Z_G(x_i)] > 1 \text{ dividing } \abs G.

Proof. Apply Theorem 1.10 to the conjugation action: singleton orbits are exactly the elements of Z(G)Z(G).

Theorem 1.12 (Fixed points of pp-groups)

Let pp be prime. A pp-group is a finite group whose order is a power of pp. If a pp-group GG acts on a finite set XX, then

XGX(modp).\abs{X^G} \equiv \abs X \pmod p .

Consequences: a nontrivial pp-group has nontrivial center, and every group of order p2p^2 is abelian.

Proof. Each orbit has cardinality [G:Gx][G:G_x], a power of pp; this power is 11 exactly on fixed points and otherwise divisible by pp. Summing over orbits gives the congruence. For the center: conjugation action of GG on itself has XG=Z(G)X^G = Z(G), so Z(G)G0(modp)\abs{Z(G)} \equiv \abs G \equiv 0 \pmod p, and Z(G)eZ(G) \ni e forces Z(G)p\abs{Z(G)} \geq p. Order p2p^2: if Z(G)GZ(G) \neq G then Z(G)=p\abs{Z(G)} = p and G/Z(G)G/Z(G) is cyclic of order pp, which forces GG abelian (Exercise 1.2) — contradiction.

Theorem 1.13 (Cauchy)

If a prime pp divides G\abs G, then GG contains an element of order pp.

Proof (McKay). Let X={(g1,,gp)Gp:g1g2gp=e}X = \{(g_1, \dots, g_p) \in G^p : g_1 g_2 \cdots g_p = e\}. Choosing g1,,gp1g_1, \dots, g_{p-1} freely determines gpg_p: X=Gp1\abs X = \abs G^{p-1}, divisible by pp. The cyclic group Z/pZ\Z/p\Z acts on XX by cyclic shift (g1,,gp)(g2,,gp,g1)(g_1, \dots, g_p) \mapsto (g_2, \dots, g_p, g_1) — this preserves XX, since g2gpg1=g11(g1gp)g1=eg_2 \cdots g_p g_1 = g_1^{-1}(g_1 \cdots g_p)g_1 = e. By Theorem 1.12, XZ/pZX0(modp)\abs{X^{\Z/p\Z}} \equiv \abs X \equiv 0 \pmod p. Fixed points are the constant tuples (g,,g)(g, \dots, g) with gp=eg^p = e; the tuple (e,,e)(e, \dots, e) is one of them, so there are at least pp of them, hence at least one geg \neq e with gp=eg^p = e: its order is exactly pp.

Theorem 1.14 (Cayley)

Every group of order nn embeds in SnS_n.

Proof. Left translation φ ⁣:GS(G)Sn\varphi \colon G \to \mathfrak S(G) \cong S_n is a morphism; φ(g)=id\varphi(g) = \mathrm{id} forces g=ge=eg = ge = e: it is faithful.

Method 1.15

Fixed-point counting is the universal opening move of finite group theory. To prove that something exists (a central element, an element of order pp, a normal subgroup, a fixed point), make a well-chosen group act on a well-chosen finite set, then compare XG\abs{X^G} with X\abs X modulo pp, or let orbit sizes divide the group order. The proofs of Theorems 1.12 and 1.13 and of all three Sylow theorems below are five variations on this single idea.

1.3 The Sylow theorems

Lagrange’s theorem says the order of a subgroup divides G\abs G; the converse fails (A4A_4, of order 1212, has no subgroup of order 66Exercise 1.1). The Sylow theorems salvage the converse for prime powers, and their counting clause is the sharpest general tool we have for producing normal subgroups.

Definition 1.16

Write G=pam\abs G = p^a m with pmp \nmid m. A Sylow pp-subgroup of GG is a subgroup of order pap^a — a pp-subgroup of the largest conceivable order. The number of Sylow pp-subgroups of GG is denoted npn_p.

Lemma 1.17

If G=pam\abs G = p^a m with pmp \nmid m, then (pampa)m(modp)\dbinom{p^a m}{p^a} \equiv m \pmod p.

Proof. In Fp[X]\mathbb F_p[X], the freshman’s dream (1+X)p=1+Xp(1+X)^p = 1 + X^p (the coefficients (pk)\binom pk, 0<k<p0<k<p, are divisible by pp: pp divides the numerator of p!k!(pk)!\frac{p!}{k!(p-k)!} but not the denominator) iterates to (1+X)pa=1+Xpa(1+X)^{p^a} = 1 + X^{p^a}, whence

(1+X)pam=(1+Xpa)m=k=0m(mk)Xkpain Fp[X].(1+X)^{p^a m} = \bigl(1 + X^{p^a}\bigr)^m = \sum_{k=0}^{m} \binom mk X^{k p^a} \quad\text{in } \mathbb F_p[X].

Identify the coefficient of XpaX^{p^a}: on the left (pampa)modp\binom{p^a m}{p^a} \bmod p, on the right (m1)=m\binom m1 = m.

Theorem 1.18 (Sylow I: existence)

For every prime pp, Sylow pp-subgroups of GG exist.

Proof (Wielandt). Let Ω\Omega be the set of subsets of GG of cardinality pap^a; GG acts on Ω\Omega by left translation gS=gSg \cdot S = gS. By Lemma 1.17, Ω=(pampa)m≢0(modp)\abs\Omega = \binom{p^a m}{p^a} \equiv m \not\equiv 0 \pmod p, so some orbit OS\mathcal O_S has size prime to pp (if pp divided every orbit size, it would divide Ω\abs\Omega). Let H=GSH = G_S be the stabilizer of such an SS. Since [G:H]=OS[G : H] = \abs{\mathcal O_S} is prime to pp and paG=[G:H]Hp^a \mid \abs G = [G:H]\,\abs H, we get paHp^a \mid \abs H. Conversely, fix sSs \in S: the map HSH \to S, hhsh \mapsto hs, is injective and lands in SS because hS=ShS = S; hence HS=pa\abs H \leq \abs S = p^a. So H=pa\abs H = p^a.

Theorem 1.19 (Sylow II: domination and conjugacy)

Let PP be a Sylow pp-subgroup and QQ any pp-subgroup of GG. Then QgPg1Q \subseteq gPg^{-1} for some gGg \in G. In particular all Sylow pp-subgroups are conjugate, and PG    np=1P \trianglelefteq G \iff n_p = 1.

Proof. Let QQ act on the coset space X=G/PX = G/P, of cardinality m≢0(modp)m \not\equiv 0 \pmod p. By Theorem 1.12 applied to the pp-group QQ, XQm≢0(modp)\abs{X^Q} \equiv m \not\equiv 0 \pmod p: there is a fixed coset gPgP, i.e. QgP=gPQgP = gP, i.e. g1QgPg^{-1}Qg \subseteq P. If QQ is itself a Sylow subgroup, equality of orders turns QgPg1Q \subseteq gPg^{-1} into an equality. Finally PGP \trianglelefteq G iff its conjugates {gPg1}\{gPg^{-1}\} — which by the above are all the Sylow pp-subgroups — reduce to {P}\{P\}.

Theorem 1.20 (Sylow III: counting)

np1(modp)n_p \equiv 1 \pmod p, and np=[G:NG(P)]n_p = [G : N_G(P)], which divides mm.

Proof. Let Sylp\mathrm{Syl}_p be the set of Sylow pp-subgroups; GG acts on it transitively by conjugation (Theorem 1.19), with stabilizer of PP the normalizer NG(P)PN_G(P) \supseteq P: np=[G:NG(P)]n_p = [G : N_G(P)], and m=[G:P]=[G:NG(P)][NG(P):P]m = [G:P] = [G:N_G(P)]\,[N_G(P):P] shows npmn_p \mid m.

Now restrict the action to PP and count fixed points. If QSylpQ \in \mathrm{Syl}_p is fixed by PP, then PNG(Q)P \subseteq N_G(Q); both PP and QQ are Sylow pp-subgroups of the group NG(Q)N_G(Q), hence conjugate in it (Theorem 1.19 applied to NG(Q)N_G(Q)); but QNG(Q)Q \trianglelefteq N_G(Q), so QQ is its only conjugate there: P=QP = Q. Thus the only fixed point is PP itself, and Theorem 1.12 gives np=SylpSylpP=1(modp)n_p = \abs{\mathrm{Syl}_p} \equiv \abs{\mathrm{Syl}_p^P} = 1 \pmod p.

Method 1.21

To analyse a group of given order n=pamn = p^a m: list the divisors of mm congruent to 11 mod pp — these are the candidates for npn_p. If the only candidate is 11, the Sylow pp-subgroup is normal. If np>1n_p > 1 is forced to be small, act by conjugation on Sylp\mathrm{Syl}_p to obtain a morphism GSnpG \to S_{n_p} with small kernel. And count elements: distinct Sylow pp-subgroups of prime order pp intersect trivially, so they carry np(p1)n_p(p-1) elements of order exactly pp; overlapping counts for different primes often force a contradiction (Exercise 1.7).

Example 1.22

Let G=pq\abs G = pq with p<qp < q primes and pq1p \nmid q - 1. Then nqpn_q \mid p and nq1modqn_q \equiv 1 \bmod q force nq=1n_q = 1 (as p<qp < q); npqn_p \mid q and np1modpn_p \equiv 1 \bmod p force np=1n_p = 1 (as q≢1modpq \not\equiv 1 \bmod p). Let P,QP, Q be the two normal Sylows: PQ={e}P \cap Q = \{e\} (coprime orders), so PQ=pq\abs{PQ} = pq (Exercise 1.4) and GP×QZ/pZ×Z/qZZ/pqZG \cong P \times Q \cong \Z/p\Z \times \Z/q\Z \cong \Z/pq\Z by Proposition 1.24 below. Every group of order 1515, 3333, 3535, … is cyclic. The excluded case pq1p \mid q - 1 produces exactly one more group, nonabelian — see the weekend problem (Problem 1.1).

Example 1.23 (A complete Sylow census: S4S_4)

Let us run the method on G=S4G = S_4, G=24=233\abs G = 24 = 2^3\cdot3. Sylow 33: n38n_3 \mid 8, n31mod3n_3 \equiv 1 \bmod 3, so n3{1,4}n_3 \in \{1, 4\}; since (123)\langle(123)\rangle and (124)\langle(124)\rangle are distinct, n3=4n_3 = 4 — the four subgroups (abc)\langle(abc)\rangle, one for each 33-element subset {a,b,c}\{a, b, c\}, accounting for the 88 three-cycles. By Sylow II they are conjugate, and the conjugation morphism S4SSyl3S4S_4 \to S_{\mathrm{Syl}_3} \cong S_4 is an isomorphism here (its kernel is contained in N=NG((123))N = N_G(\langle(123)\rangle) of order 24/4=624/4 = 6, and a normal subgroup of S4S_4 inside S3S_3-like NN must be trivial: it would consist of even permutations fixing the four Sylows, and only ee does). Sylow 22: n23n_2 \mid 3, n21mod2n_2 \equiv 1 \bmod 2: n2{1,3}n_2 \in \{1, 3\}. The subgroup D=(1234),(13)D = \langle(1234), (13)\rangle has order 88 (a dihedral D4D_4: the symmetries of the square with vertices 1,2,3,41, 2, 3, 4), is not normal ((12)(1234)(12)=(2134)(12)(1234)(12) = (2134) generates a different 44-cycle subgroup), so n2=3n_2 = 3: the three copies of D4D_4 correspond to the three ways of pairing 44 points into a “square”. Note the moral of the census: S4=24\abs{S_4} = 24 leaves room for either Sylow to fail normality, and both do — compare order 1212, where the count forces one of them normal (Part IV of Problem 1.1).

1.4 Products, direct and semidirect

Proposition 1.24 (Recognizing a direct product)

Let H,KGH, K \trianglelefteq G with HK={e}H \cap K = \{e\} and HK=GHK = G. Then (h,k)hk(h,k) \mapsto hk is an isomorphism H×KGH \times K \to G.

Proof. For hHh \in H, kKk \in K, the commutator hkh1k1hkh^{-1}k^{-1} lies in KK (read it as (hkh1)k1(hkh^{-1})k^{-1}, using normality of KK) and in HH (read it as h(kh1k1)h(kh^{-1}k^{-1})): it is ee, so HH and KK commute elementwise and the map is a morphism. It is surjective since HK=GHK = G, and injective since hk=ehk = e gives h=k1HK={e}h = k^{-1} \in H \cap K = \{e\}.

Normality of both factors is what fails most often: in S3=(123)(12)S_3 = \langle (1\,2\,3)\rangle \,\langle(1\,2)\rangle both factors intersect trivially and generate, yet S3≇Z/3Z×Z/2ZS_3 \not\cong \Z/3\Z \times \Z/2\Z. The right notion when only one factor is normal:

Definition 1.25

Let HH, KK be groups and φ ⁣:KAut(H)\varphi \colon K \to \operatorname{Aut}(H) a morphism. The semidirect product HφKH \rtimes_\varphi K is the set H×KH \times K equipped with

(h,k)(h,k)=(hφ(k)(h),  kk).(h, k)\,(h', k') = \bigl(h\,\varphi(k)(h'),\; kk'\bigr).

Proposition 1.26

HφKH \rtimes_\varphi K is a group; H×{e}H \times \{e\} is a normal subgroup isomorphic to HH, {e}×K\{e\} \times K a subgroup isomorphic to KK; they intersect trivially and generate. Conversely, if G=NKG = NK with NGN \trianglelefteq G, KGK \leq G and NK={e}N \cap K = \{e\}, then GNφKG \cong N \rtimes_\varphi K for φ(k)=(nknk1)\varphi(k) = (n \mapsto knk^{-1}).

Proof. Direct verification: associativity reduces to φ(kk)=φ(k)φ(k)\varphi(kk') = \varphi(k)\circ\varphi(k') and each φ(k)\varphi(k) being a morphism; the identity is (e,e)(e,e) and (h,k)1=(φ(k1)(h1),k1)(h,k)^{-1} = \bigl(\varphi(k^{-1})(h^{-1}), k^{-1}\bigr). The projection (h,k)k(h,k) \mapsto k is a morphism onto KK with kernel H×{e}H \times \{e\}, which is therefore normal. For the converse: every gGg \in G writes uniquely as nknk with nNn \in N, kKk \in K (existence: G=NKG = NK; uniqueness: nk=nknk = n'k' gives n1n=kk1NKn'^{-1}n = k'k^{-1} \in N \cap K), and

(nk)(nk)=n(knk1)  kk(nk)(n'k') = n\,(kn'k^{-1})\;kk'

shows that nk(n,k)nk \mapsto (n, k) transports the law of GG to that of NφKN \rtimes_\varphi K.

Example 1.27

(a) The dihedral group DnD_n (n3n \geq 3) of the 2n2n symmetries of a regular nn-gon: the rotations form a normal subgroup of index 22, any reflection generates a complement, and conjugating a rotation by a reflection inverts it: DnZ/nZφZ/2ZD_n \cong \Z/n\Z \rtimes_\varphi \Z/2\Z with φ(1)=(xx)\varphi(1) = (x \mapsto -x). (b) The affine group of a line, {xax+b:aK×,bK}KK×\{x \mapsto ax + b : a \in K^\times,\, b \in K\} \cong K \rtimes K^\times: translations normal, homotheties a complement. (c) SnAnZ/2ZS_n \cong A_n \rtimes \Z/2\Z (complement: any transposition). (d) The quaternion group Q8Q_8 is not a semidirect product of proper subgroups: every nontrivial subgroup contains 1-1 (Problem 1.1), so no two proper subgroups intersect trivially.

1.5 Solvable groups; simplicity of AnA_n

Definition 1.28

The commutator of x,yGx, y \in G is [x,y]=xyx1y1[x,y] = xyx^{-1}y^{-1}; the derived subgroup D(G)D(G) is the subgroup generated by all commutators. The derived series is D0(G)=GD^0(G) = G, Di+1(G)=D(Di(G))D^{i+1}(G) = D(D^i(G)), and GG is solvable if Dn(G)={e}D^n(G) = \{e\} for some nn.

Proposition 1.29

D(G)D(G) is normal (indeed stable under every automorphism), G/D(G)G/D(G) is abelian, and for NGN \trianglelefteq G: G/NG/N abelian     D(G)N\iff D(G) \subseteq N. Moreover GG is solvable iff there is a chain G=G0G1Gn={e}G = G_0 \trianglerighteq G_1 \trianglerighteq \dots \trianglerighteq G_n = \{e\} with each Gi+1GiG_{i+1} \trianglelefteq G_i and each quotient Gi/Gi+1G_i/G_{i+1} abelian. Subgroups and quotients of solvable groups are solvable; conversely, if NN and G/NG/N are solvable, so is GG.

Proof. An automorphism α\alpha maps [x,y][x,y] to [αx,αy][\alpha x, \alpha y]: it permutes the commutators, so preserves the subgroup they generate; conjugations are automorphisms, whence normality. In G/D(G)G/D(G), xˉyˉxˉ1yˉ1=[x,y]=eˉ\bar x\bar y \bar x^{-1}\bar y^{-1} = \overline{[x,y]} = \bar e: the quotient is abelian. If G/NG/N is abelian then every [x,y]N[x,y] \in N, so D(G)ND(G) \subseteq N; conversely if D(G)ND(G) \subseteq N then G/NG/N, a quotient of the abelian G/D(G)G/D(G) by the third isomorphism theorem, is abelian.

If GG is solvable, the derived series is such a chain. Conversely, given a chain, Di(G)GiD^i(G) \subseteq G_i by induction: Gi/Gi+1G_i/G_{i+1} abelian gives D(Gi)Gi+1D(G_i) \subseteq G_{i+1}, so Di+1(G)=D(DiG)D(Gi)Gi+1D^{i+1}(G) = D(D^i G) \subseteq D(G_i) \subseteq G_{i+1}; hence Dn(G)={e}D^n(G) = \{e\}.

Heredity: Di(H)Di(G)D^i(H) \subseteq D^i(G) for HGH \leq G (induction), and Di(G/N)=π(Di(G))D^i(G/N) = \pi(D^i(G)) since π\pi maps commutators onto commutators; this gives the statements for subgroups and quotients. Extension: if Dm(G/N)={e}D^m(G/N) = \{e\} then Dm(G)ND^m(G) \subseteq N, and Dn(N)={e}D^n(N) = \{e\} gives Dm+n(G)=Dn(Dm(G))Dn(N)={e}D^{m+n}(G) = D^n(D^m(G)) \subseteq D^n(N) = \{e\}.

Example 1.30

Abelian groups are solvable. pp-groups are solvable, by induction on the order: Z(G){e}Z(G) \neq \{e\} and G/Z(G)G/Z(G) is a smaller pp-group. S3S_3 and S4S_4 are solvable: S4A4V{e}S_4 \trianglerighteq A_4 \trianglerighteq V \trianglerighteq \{e\}, where V={e,(12)(34),(13)(24),(14)(23)}V = \{e, (1\,2)(3\,4), (1\,3)(2\,4), (1\,4)(2\,3)\} is the Klein group of double transpositions (normal in S4S_4: a union of conjugacy classes), with abelian quotients Z/2Z\Z/2\Z, Z/3Z\Z/3\Z, VV. In Chapter 4, “the general equation of degree nn is solvable by radicals” will literally meanSnS_n is a solvable group”. Whence the importance of the next definition.

Definition 1.31

A group G{e}G \neq \{e\} is simple if its only normal subgroups are {e}\{e\} and GG. A nonabelian simple group is not solvable: D(G)GD(G) \trianglelefteq G is not {e}\{e\} (else GG abelian), so D(G)=GD(G) = G and the derived series is constant. The abelian simple groups are exactly the Z/pZ\Z/p\Z, pp prime (an abelian group is simple iff it has no proper nontrivial subgroup, iff it is cyclic of prime order by Lagrange).

Lemma 1.32

For n3n \geq 3, AnA_n is generated by 33-cycles; for n5n \geq 5, all 33-cycles are conjugate in AnA_n.

Proof. An element of AnA_n is a product of an even number of transpositions; pair them up and use (composition right to left)

(ab)(cd)=(acb)(acd),(ab)(bc)=(abc),(ab)(ab)=e(a\,b)(c\,d) = (a\,c\,b)(a\,c\,d), \qquad (a\,b)(b\,c) = (a\,b\,c), \qquad (a\,b)(a\,b) = e

for disjoint, overlapping and equal pairs respectively: each pair of transpositions is a product of 33-cycles.

Conjugacy: σ(abc)σ1=(σa  σb  σc)\sigma(a\,b\,c)\sigma^{-1} = (\sigma a\; \sigma b\; \sigma c), so any two 33-cycles are conjugate by some σSn\sigma \in S_n. If σ\sigma is odd, replace it by σ=σ(de)\sigma' = \sigma (d\,e) where d,ed, e are two points outside {a,b,c}\{a, b, c\} — they exist since n5n \geq 5; then σ\sigma' is even and σ(abc)σ1=σ(abc)σ1\sigma'(a\,b\,c) \sigma'^{-1} = \sigma(a\,b\,c)\sigma^{-1}, since (de)(d\,e) commutes with (abc)(a\,b\,c).

Theorem 1.33 (Simplicity of the alternating group)

AnA_n is simple for n5n \geq 5.

Proof. Let NAnN \trianglelefteq A_n, N{e}N \neq \{e\}. By Lemma 1.32 it suffices to show that NN contains one 33-cycle: normality and conjugacy of 33-cycles in AnA_n then put all 33-cycles in NN, so N=AnN = A_n.

For ρSn\rho \in S_n let F(ρ)={x:ρ(x)x}F(\rho) = \{x : \rho(x) \neq x\} be its support and f(ρ)=F(ρ)f(\rho) = \abs{F(\rho)}. Choose σN{e}\sigma \in N \setminus \{e\} with f(σ)f(\sigma) minimal. Note that a nontrivial even permutation has f3f \geq 3, and that f(σ)=4f(\sigma) = 4 is impossible for σAn\sigma \in A_n unless σ\sigma is a double transposition (a 44-cycle is odd). We show σ\sigma is a 33-cycle.

Case A: σ\sigma is a product of disjoint transpositions, say σ=(ab)(cd)\sigma = (a\,b)(c\,d)\cdots with f(σ)4f(\sigma) \geq 4. Pick e{a,b,c,d}e' \notin \{a, b, c, d\} (possible: n5n \geq 5), set τ=(cde)\tau = (c\,d\,e') and

σ=τστ1σ1N(τστ1N by normality).\sigma' = \tau\sigma\tau^{-1}\,\sigma^{-1} \in N \qquad (\tau\sigma\tau^{-1} \in N \text{ by normality}).

Since στ1σ1=(σc  σe  σd)=(d  σe  c)\sigma\tau^{-1}\sigma^{-1} = (\sigma c\;\sigma e'\;\sigma d) = (d\;\sigma e'\;c) (using σc=d\sigma c = d, σd=c\sigma d = c), we get σ=(cde)(d  σe  c)\sigma' = (c\,d\,e')(d\;\sigma e'\;c).

If σe=e\sigma e' = e' (which holds in particular when f(σ)=4f(\sigma) = 4, i.e. σ=(ab)(cd)\sigma = (a\,b)(c\,d)): then (dec)=(cde)(d\,e'\,c) = (c\,d\,e') and σ=(cde)2=(ced)\sigma' = (c\,d\,e')^2 = (c\,e'\,d), a 33-cycle lying in NN, with f(σ)=3<4f(σ)f(\sigma') = 3 < 4 \leq f(\sigma) — contradicting minimality.

If σee\sigma e' \neq e': then σe{a,b,c,d,e}\sigma e' \notin \{a, b, c, d, e'\} (σ\sigma swaps a,ba,b and c,dc,d, and e{a,b,c,d}e' \notin \{a,b,c,d\} with σ\sigma injective), so σ\sigma moves the six points a,b,c,d,e,σea, b, c, d, e', \sigma e': f(σ)6f(\sigma) \geq 6. On the other hand σ\sigma', a product of two 33-cycles with supports in {c,d,e,σe}\{c, d, e', \sigma e'\}, satisfies f(σ)4f(\sigma') \leq 4; and σe\sigma' \neq e, since σ(d)=τστ1(c)=τσ(e)=σed\sigma'(d) = \tau\sigma\tau^{-1}(c) = \tau\sigma(e') = \sigma e' \neq d (τ\tau fixes σe{c,d,e}\sigma e' \notin \{c,d,e'\}). So σN{e}\sigma' \in N \setminus\{e\} with f(σ)4<f(σ)f(\sigma') \leq 4 < f(\sigma): minimality is contradicted.

Case B: some cycle of σ\sigma has length 3\geq 3, say σ(a)=b\sigma(a) = b, σ(b)=c\sigma(b) = c with a,b,ca, b, c distinct. If σ\sigma is exactly this 33-cycle, we are done. Otherwise f(σ)5f(\sigma) \geq 5 (the case f(σ)=4f(\sigma) = 4 with a 3\geq3-cycle is the odd 44-cycle, excluded), so we may pick d,eF(σ){a,b,c}d, e' \in F(\sigma) \setminus \{a, b, c\}. Set τ=(cde)\tau = (c\,d\,e') and σ=τστ1σ1N\sigma' = \tau\sigma\tau^{-1}\sigma^{-1} \in N. As before σ=(cde)(σc  σe  σd)\sigma' = (c\,d\,e')\,(\sigma c\;\sigma e'\;\sigma d) moves only points of

M={c,d,e}{σc,σd,σe}F(σ)M = \{c, d, e'\} \cup \{\sigma c, \sigma d, \sigma e'\} \subseteq F(\sigma)

(images of moved points are moved: σ(x)x\sigma(x) \ne x implies σ(σx)σx\sigma(\sigma x) \neq \sigma x, σ\sigma being injective). Also bMb \notin M: the five points a,b,c,d,ea, b, c, d, e' are distinct, so b{c,d,e}b \notin \{c, d, e'\}; and b{σc,σd,σe}b \in \{\sigma c, \sigma d, \sigma e'\} would force a{c,d,e}a \in \{c, d, e'\} (apply σ1\sigma^{-1}, using σa=b\sigma a = b), which is false. Hence σ\sigma' fixes bb, while σ\sigma moves bb; and F(σ)F(σ)F(\sigma') \subseteq F(\sigma). Finally σe\sigma' \neq e: σ1(c)=b\sigma^{-1}(c) = b, τ1(b)=b\tau^{-1}(b) = b, σ(b)=c\sigma(b) = c, τ(c)=d\tau(c) = d, so σ(c)=dc\sigma'(c) = d \neq c. Thus σN{e}\sigma' \in N \setminus \{e\} with f(σ)f(σ)1f(\sigma') \leq f(\sigma) - 1, contradicting minimality.

Both cases being impossible, σ\sigma is a 33-cycle.

Corollary 1.34

For n5n \geq 5: AnA_n and SnS_n are not solvable, and the only normal subgroups of SnS_n are {e}\{e\}, AnA_n and SnS_n.

Proof. AnA_n is nonabelian simple, hence not solvable (Definition 1.31); a group containing a non-solvable subgroup is not solvable (Proposition 1.29). Let NSnN \trianglelefteq S_n: then NAnAnN \cap A_n \trianglelefteq A_n equals {e}\{e\} or AnA_n. If NAn=AnN \cap A_n = A_n, then AnNA_n \subseteq N and N{An,Sn}N \in \{A_n, S_n\} by the index. If NAn={e}N \cap A_n = \{e\}, the restriction to NN of the projection SnSn/AnZ/2ZS_n \to S_n/A_n \cong \Z/2\Z is injective, so N2\abs N \leq 2; if N={e,σ}N = \{e, \sigma\}, normality makes the conjugacy class of σ\sigma equal to {σ}\{\sigma\}, i.e. σZ(Sn)\sigma \in Z(S_n). But Z(Sn)={e}Z(S_n) = \{e\} for n3n \geq 3: if σe\sigma \ne e moves aa to bab \neq a, pick c{a,b}c \notin \{a, b\}; then (bc)σ(bc)1(b\,c)\sigma(b\,c)^{-1} sends aa to cbc \ne b, so it differs from σ\sigma. Hence N={e}N = \{e\}.

Theorem 1.35 (Jordan–Hölder)

Every finite group G{e}G \neq \{e\} admits a composition series

{e}=G0G1Gr=G,Gi/Gi1 simple,\{e\} = G_0 \trianglelefteq G_1 \trianglelefteq \cdots \trianglelefteq G_r = G, \qquad G_{i}/G_{i-1} \text{ simple},

and the multiset of composition factors Gi/Gi1G_i/G_{i-1}, up to isomorphism, does not depend on the chosen series. A finite group is solvable iff all its composition factors are cyclic of prime order.

Proof. Existence: induction on G\abs G. If GG is simple, take {e}G\{e\} \trianglelefteq G. Otherwise pick a maximal proper normal subgroup NN (there are finitely many subgroups); G/NG/N is simple by the correspondence theorem (a proper nontrivial normal subgroup of G/NG/N would lift to a normal subgroup of GG strictly between NN and GG). Append NGN \trianglelefteq G to a composition series of NN.

Uniqueness: induction on G\abs G, the case GG simple being clear. Take two composition series, with penultimate terms MGM \trianglelefteq G and NGN \trianglelefteq G (so G/MG/M, G/NG/N are simple). If M=NM = N, conclude by induction applied to MM. Otherwise MNMN, normal in GG and strictly containing MM, equals GG (MM is maximal normal: any normal MLGM \subsetneq L \subsetneq G would map to a proper nontrivial normal subgroup of the simple G/MG/M). The second isomorphism theorem gives

G/M=MN/MN/(MN),G/N=MN/NM/(MN).G/M = MN/M \cong N/(M \cap N), \qquad G/N = MN/N \cong M/(M \cap N).

Set K=MNK = M \cap N (G\trianglelefteq G) and fix a composition series of KK. Then MM carries two composition series: its original one, and the series of KK followed by KMK \trianglelefteq M (the quotient M/KG/NM/K \cong G/N is simple). By induction (applied to MM), the factors of the original series of MM are {factors of K}{G/N}\{\text{factors of } K\} \cup \{G/N\}; likewise for NN. Hence both series of GG have factors

{factors of K}    {G/N,  G/M},\{\text{factors of } K\} \;\cup\; \{\,G/N,\; G/M\,\},

the same multiset.

Solvability: if all factors are Z/piZ\Z/p_i\Z, the series is a chain with abelian quotients, so GG is solvable (Proposition 1.29). Conversely, a composition factor of a solvable group is solvable (a quotient of a subgroup) and simple; a solvable simple group is abelian (D(G)GD(G) \ne G forces D(G)={e}D(G) = \{e\}), hence some Z/pZ\Z/p\Z.

Remark 1.36

Jordan–Hölder says every finite group is built from simple groups, with a well-defined parts list — an arithmetic of groups in which the simple groups are the primes, and where how the parts are glued (extension data, as in the semidirect product) replaces mere multiplication. The classification of the finite simple groups — the cyclic Z/pZ\Z/p\Z, the alternating An5A_{n \geq 5}, sixteen families of Lie type, and 2626 sporadic groups — is one of the monuments of twentieth-century mathematics; its proof, spread over some ten thousand journal pages, is very far beyond this course.

The ten subgroups of the dihedral group D_4 = r, s r4 = s2 = e,\ srs-1 = r-1. The three subgroups of index 2 (middle row) are normal, as is the center r2 (highlighted); the four reflection subgroups fall into two conjugacy classes of two. Chains from bottom to top give composition series, e.g. \e\ r2 r D_4: factors ℤ/2ℤ, ℤ/2ℤ, ℤ/2ℤ — always the same multiset, as Jordan–Hölder demands.
The ten subgroups of the dihedral group D4=r,sr4=s2=e, srs1=r1D_4 = \langle r, s \mid r^4 = s^2 = e,\ srs^{-1} = r^{-1}\rangle. The three subgroups of index 22 (middle row) are normal, as is the center r2\langle r^2\rangle (highlighted); the four reflection subgroups fall into two conjugacy classes of two. Chains from bottom to top give composition series, e.g. {e}r2rD4\{e\} \trianglelefteq \langle r^2\rangle \trianglelefteq \langle r\rangle \trianglelefteq D_4: factors Z/2Z,Z/2Z,Z/2Z\Z/2\Z, \Z/2\Z, \Z/2\Z — always the same multiset, as Jordan–Hölder demands.

1.6 Exercises

Exercise 1.1

(a) Show that every subgroup of index 22 is normal. (b) Show that if [G:H]=2[G : H] = 2, then x2Hx^2 \in H for every xGx \in G. (c) Deduce that A4A_4 has no subgroup of order 66: Lagrange’s converse fails. (Count the squares of 33-cycles.)

Solution

Solution of Exercise 1.1.

(a) Let [G:H]=2[G:H] = 2. For gHg \in H, gH=H=HggH = H = Hg. For gHg \notin H: the two left cosets are HH and gHgH, so gH=GHgH = G \setminus H; likewise Hg=GHHg = G \setminus H. Hence gH=HggH = Hg for all gg: HGH \trianglelefteq G.

(b) By (a), G/HG/H is a group of order 22; the class xˉ\bar x satisfies xˉ2=eˉ\bar x^2 = \bar e, i.e. x2Hx^2 \in H.

(c) Suppose HA4H \leq A_4 with H=6\abs H = 6, hence of index 22. By (b), σ2H\sigma^2 \in H for every σA4\sigma \in A_4. Every 33-cycle is such a square: if σ3=e\sigma^3 = e then σ=σ4=(σ2)2\sigma = \sigma^4 = (\sigma^2)^2. So HH contains all eight 33-cycles of A4A_4: H8>6\abs H \geq 8 > 6, a contradiction. (Lagrange’s converse fails at the very first opportunity: 6126 \mid 12.)

Exercise 1.2

Show that if G/Z(G)G/Z(G) is cyclic then GG is abelian. Deduce again that every group of order p2p^2 is abelian, and exhibit, for each prime pp, a nonabelian group of order p3p^3. (Think of upper triangular matrices with unit diagonal over Fp\mathbb F_p.)

Solution

Solution of Exercise 1.2.

Say G/Z(G)=gZ(G)G/Z(G) = \langle gZ(G) \rangle. Every xGx \in G then writes x=gkzx = g^k z with kZk \in \Z, zZ(G)z \in Z(G). For x=gkzx = g^kz, y=glzy = g^l z':

xy=gkzglz=gk+lzz=glzgkz=yx,xy = g^k z\, g^l z' = g^{k+l} z z' = g^l z'\, g^k z = yx,

central elements commuting with everything: GG is abelian.

Order p2p^2: Z(G){e}Z(G) \neq \{e\} (Theorem 1.12), so Z(G){p,p2}\abs{Z(G)} \in \{p, p^2\}. If it were pp, then G/Z(G)G/Z(G) would have order pp, hence be cyclic, forcing GG abelian and Z(G)=GZ(G) = G of order p2p^2 — contradiction. So Z(G)=GZ(G) = G.

Nonabelian of order p3p^3: the Heisenberg group

Hp={(1ac01b001):a,b,cFp}GL3(Fp),H_p = \left\{ \begin{pmatrix} 1 & a & c\\ 0 & 1 & b\\ 0 & 0 & 1 \end{pmatrix} : a, b, c \in \mathbb F_p \right\} \leq GL_3(\mathbb F_p),

of order p3p^3 (free choice of a,b,ca, b, c; closure and inverses by direct computation). It is nonabelian: the two elementary matrices I+E12I + E_{12} and I+E23I + E_{23} have commutator I+E13II + E_{13} \neq I.

Exercise 1.3

(a) Show that Aut(Z/nZ)(Z/nZ)×\operatorname{Aut}(\Z/n\Z) \cong (\Z/n\Z)^\times. (b) Show that the inner automorphisms ιg ⁣:xgxg1\iota_g \colon x \mapsto gxg^{-1} form a normal subgroup Inn(G)Aut(G)\operatorname{Inn}(G) \trianglelefteq \operatorname{Aut}(G), with Inn(G)G/Z(G)\operatorname{Inn}(G) \cong G/Z(G).

Solution

Solution of Exercise 1.3.

(a) A morphism f ⁣:Z/nZZ/nZf \colon \Z/n\Z \to \Z/n\Z is determined by k=f(1ˉ)k = f(\bar 1) (then f(mˉ)=mkˉf(\bar m) = m\bar k), and every kˉ\bar k defines one. It is bijective iff kˉ\bar k generates Z/nZ\Z/n\Z, iff gcd(k,n)=1\gcd(k,n) = 1, iff kˉ(Z/nZ)×\bar k \in (\Z/n\Z)^\times. Composition corresponds to multiplication: fkfl=fklf_k \circ f_l = f_{kl}. Hence Aut(Z/nZ)(Z/nZ)×\operatorname{Aut}(\Z/n\Z) \cong (\Z/n\Z)^\times.

(b) The map ι ⁣:GAut(G)\iota\colon G \to \operatorname{Aut}(G), gιgg \mapsto \iota_g, is a morphism: ιgιh=ιgh\iota_g \circ \iota_h = \iota_{gh}. Its image is Inn(G)\operatorname{Inn}(G); its kernel is {g:gxg1=x x}=Z(G)\{g : gxg^{-1} = x\ \forall x\} = Z(G). The first isomorphism theorem gives Inn(G)G/Z(G)\operatorname{Inn}(G) \cong G/Z(G). Normality in Aut(G)\operatorname{Aut}(G): for αAut(G)\alpha \in \operatorname{Aut}(G),

(αιgα1)(x)=α(gα1(x)g1)=α(g)xα(g)1=ια(g)(x).(\alpha \circ \iota_g \circ \alpha^{-1})(x) = \alpha\bigl(g\,\alpha^{-1}(x)\,g^{-1}\bigr) = \alpha(g)\, x\, \alpha(g)^{-1} = \iota_{\alpha(g)}(x).

Exercise 1.4 ★★

Let H,KH, K be subgroups of a finite group GG. (a) Prove the product formula HKHK=HK\abs{HK}\,\abs{H\cap K} = \abs H\, \abs K, by counting the fibers of the map H×KHKH \times K \to HK, (h,k)hk(h,k) \mapsto hk. (b) Show that HKHK is a subgroup iff HK=KHHK = KH (automatic when one of the two is normal). (c) If H,KGH, K \trianglelefteq G and HK={e}H \cap K = \{e\}, show that hk=khhk = kh for all hHh \in H, kKk \in K.

Solution

Solution of Exercise 1.4.

(a) Consider μ ⁣:H×KHK\mu \colon H \times K \to HK, (h,k)hk(h,k) \mapsto hk, surjective by definition. Fix h0k0HKh_0k_0 \in HK: then hk=h0k0    h01h=k0k1HKhk = h_0k_0 \iff h_0^{-1}h = k_0 k^{-1} \in H \cap K. Writing u=h01hu = h_0^{-1}h, the fiber of h0k0h_0k_0 is {(h0u,u1k0):uHK}\{(h_0u,\, u^{-1}k_0) : u \in H \cap K\}, of cardinality HK\abs{H \cap K}. Hence HK=H×K=HKHK\abs H\,\abs K = \abs{H\times K} = \abs{HK}\,\abs{H \cap K}.

(b) If HKHK is a subgroup: KHHKKH \subseteq HK because kh=(h1k1)1(HK)1=HKkh = \bigl(h^{-1}k^{-1}\bigr)^{-1} \in (HK)^{-1} = HK; and HKKHHK \subseteq KH by taking inverses in HK=(HK)1(KH)1HK = (HK)^{-1} \subseteq (KH)^{-1}\dots more directly, for hkHKhk \in HK, (hk)1=k1h1KH(hk)^{-1} = k^{-1}h^{-1} \in KH, so HK=(HK)1KHHK = (HK)^{-1} \subseteq KH; both inclusions give HK=KHHK = KH. Conversely if HK=KHHK = KH: closure, (hk)(hk)=h(kh)kh(HK)k=(hH)(Kk)HK(hk)(h'k') = h(kh')k' \in h(HK)k' = (hH)(Kk') \subseteq HK; inverses, (hk)1=k1h1KH=HK(hk)^{-1} = k^{-1}h^{-1} \in KH = HK; and eHKe \in HK: subgroup. If, say, KGK \trianglelefteq G, then hK=KhhK = Kh for all hh, so HK=KHHK = KH automatically.

(c) For hHh \in H, kKk \in K, the commutator [h,k]=hkh1k1[h,k] = hkh^{-1} k^{-1} equals (hkh1)k1K(hkh^{-1})k^{-1} \in K (KK normal) and h(kh1k1)Hh(kh^{-1}k^{-1}) \in H (HH normal), hence lies in HK={e}H \cap K = \{e\}: hk=khhk = kh.

Exercise 1.5 ★★

(Burnside’s counting lemma) A finite group GG acts on a finite set XX. Show that the number of orbits is the average number of fixed points:

#{orbits}=1GgGFix(g),Fix(g)={xX:gx=x},\#\{\text{orbits}\} = \frac{1}{\abs G}\sum_{g \in G} \abs{\operatorname{Fix}(g)}, \qquad \operatorname{Fix}(g) = \{x \in X : g \cdot x = x\},

by counting the set {(g,x):gx=x}\{(g,x) : g\cdot x = x\} in two ways. Application: Z/pZ\Z/p\Z (pp prime) acts by rotation on necklaces of pp beads with aa available colors; deduce Fermat’s little theorem apa(modp)a^p \equiv a \pmod p.

Solution

Solution of Exercise 1.5.

Count E={(g,x)G×X:gx=x}E = \{(g,x) \in G \times X : g \cdot x = x\} two ways:

E=gGFix(g)=xXGx=xXGOx=GO orbitxO1O=G#{orbits},\abs E = \sum_{g \in G} \abs{\operatorname{Fix}(g)} = \sum_{x \in X} \abs{G_x} = \sum_{x \in X} \frac{\abs G}{\abs{\mathcal O_x}} = \abs G \sum_{\mathcal O \text{ orbit}} \sum_{x \in \mathcal O} \frac{1}{\abs{\mathcal O}} = \abs G \cdot \#\{\text{orbits}\},

using orbit–stabilizer (Gx=G/Ox\abs{G_x} = \abs G/\abs{\mathcal O_x}) and the partition into orbits.

Necklaces: let XX be the set of maps Z/pZ{1,,a}\Z/p\Z \to \{1, \dots, a\} (colorings of pp positions), X=ap\abs X = a^p, with Z/pZ\Z/p\Z acting by rotation. The identity fixes all apa^p colorings. A rotation kˉ0ˉ\bar k \neq \bar 0 generates Z/pZ\Z/p\Z (pp prime), so a coloring it fixes is invariant under all rotations, hence constant: aa fixed colorings. Burnside:

#{orbits}=ap+(p1)apN,\#\{\text{orbits}\} = \frac{a^p + (p-1)a}{p} \in \N ,

so pap+(p1)ap \mid a^p + (p-1)a, i.e. papap \mid a^p - a: Fermat’s little theorem, by pure counting.

Exercise 1.6

Using the Sylow theorems, show that every group of order 1515 is cyclic and that every group of order 4545 is abelian.

Solution

Solution of Exercise 1.6.

Order 15=3515 = 3 \cdot 5: n35n_3 \mid 5 and n31(mod3)n_3 \equiv 1 \pmod 3 force n3=1n_3 = 1; n53n_5 \mid 3 and n51(mod5)n_5 \equiv 1 \pmod 5 force n5=1n_5 = 1. The Sylows P3,P5P_3, P_5 are normal, intersect trivially (coprime orders), and P3P5=15\abs{P_3P_5} = 15 (Exercise 1.4(a)): by Proposition 1.24, GZ/3Z×Z/5ZZ/15ZG \cong \Z/3\Z \times \Z/5\Z \cong \Z/15\Z (Chinese remainder).

Order 45=32545 = 3^2 \cdot 5: n35n_3 \mid 5, n31(mod3)n_3 \equiv 1 \pmod 3 give n3=1n_3 = 1; n59n_5 \mid 9, n51(mod5)n_5 \equiv 1 \pmod 5 give n5=1n_5 = 1. So GP3×P5G \cong P_3 \times P_5 with P3=9=32\abs{P_3} = 9 = 3^2 and P5=5\abs{P_5} = 5: both abelian (Theorem 1.12 for p2p^2; prime order is cyclic), hence so is GG.

Exercise 1.7 ★★

Show that no group of order 3030, and none of order 5656, is simple. (For 3030: if n31n_3 \neq 1 and n51n_5 \neq 1, count the elements of orders 33 and 55. For 5656: count the elements of order 77.)

Solution

Solution of Exercise 1.7.

Order 30=23530 = 2 \cdot 3 \cdot 5. n56n_5 \mid 6, n51(mod5)n_5 \equiv 1 \pmod 5: n5{1,6}n_5 \in \{1, 6\}; n310n_3 \mid 10, n31(mod3)n_3 \equiv 1 \pmod 3: n3{1,10}n_3 \in \{1, 10\}. Suppose GG simple, so n5=6n_5 = 6 and n3=10n_3 = 10. Two distinct subgroups of prime order pp intersect trivially (the intersection is a proper subgroup of Z/pZ\Z/p\Z), so the six Sylow 55-subgroups carry 6×4=246 \times 4 = 24 elements of order 55, and the ten Sylow 33-subgroups carry 10×2=2010 \times 2 = 20 elements of order 33: 24+20=44>30124 + 20 = 44 > 30 - 1 non-identity elements — absurd. So n5=1n_5 = 1 or n3=1n_3 = 1: a normal Sylow exists.

Order 56=23756 = 2^3 \cdot 7. n78n_7 \mid 8, n71(mod7)n_7 \equiv 1 \pmod 7: n7{1,8}n_7 \in \{1, 8\}. If n7=8n_7 = 8, the Sylow 77-subgroups carry 8×6=488 \times 6 = 48 elements of order 77, leaving exactly 5648=856 - 48 = 8 other elements. A Sylow 22-subgroup has order 88 and consists of such elements, so it is the set of them: n2=1n_2 = 1. Either n7=1n_7 = 1 or n2=1n_2 = 1: never simple.

Exercise 1.8 ★★

(a) Let HGH \leq G of index nn. Show that the action of GG on G/HG/H yields a morphism GSnG \to S_n whose kernel gGgHg1\bigcap_{g \in G} gHg^{-1} is the largest normal subgroup of GG contained in HH. (b) Deduce: if GG is finite and pp is the smallest prime divisor of G\abs G, every subgroup of index pp is normal. (c) Show that every transitive action of GG on a set XX is isomorphic to the action on a coset space: there is a bijection XG/GxX \to G/G_x commuting with the actions.

Solution

Solution of Exercise 1.8.

(a) The action gxH=gxHg \cdot xH = gxH gives a morphism ρ ⁣:GS(G/H)Sn\rho \colon G \to \mathfrak S(G/H) \cong S_n. Its kernel is

kerρ={g:xG, gxH=xH}={g:x, x1gxH}=xGxHx1,\ker\rho = \{g : \forall x \in G,\ gxH = xH\} = \{g : \forall x,\ x^{-1}gx \in H\} = \bigcap_{x \in G} xHx^{-1},

a normal subgroup (a kernel) contained in HH (take x=ex = e). If NGN \trianglelefteq G and NHN \subseteq H, then for every xx: N=xNx1xHx1N = xNx^{-1} \subseteq xHx^{-1}, so NkerρN \subseteq \ker\rho: the kernel is the largest such.

(b) Let [G:H]=p[G:H] = p, smallest prime dividing G\abs G, and K=kerρHK = \ker\rho \subseteq H. Then G/KG/K embeds in SpS_p, so [G:K][G:K] divides p!p!. Also [G:K]=[G:H][H:K]=p[H:K][G:K] = [G:H]\,[H:K] = p\,[H:K], so [H:K][H:K] divides (p1)!(p-1)!. But [H:K][H:K] divides G\abs G, whose prime divisors are all p\geq p, while the prime divisors of (p1)!(p-1)! are all <p< p: hence [H:K]=1[H:K] = 1, i.e. H=K=kerρH = K = \ker \rho is normal.

(c) Let the action be transitive and xXx \in X. The map Φ ⁣:G/GxX\Phi \colon G/G_x \to X, gGxgxgG_x \mapsto g \cdot x, is well defined and bijective (orbit–stabilizer; the orbit is all of XX), and it intertwines the actions: Φ(hgGx)=Φ(hgGx)=(hg)x=hΦ(gGx)\Phi(h \cdot gG_x) = \Phi(hgG_x) = (hg) \cdot x = h \cdot \Phi(gG_x).

Exercise 1.9 ★★

(a) Show that D(G)D(G) is the smallest normal subgroup of GG with abelian quotient, and that every morphism from GG to an abelian group factors uniquely through the abelianization Gab=G/D(G)G^{\mathrm{ab}} = G/D(G). (b) Compute D(Sn)D(S_n) and SnabS_n^{\mathrm{ab}} for n2n \geq 2, and D(Q8)D(Q_8) and Q8abQ_8^{\mathrm{ab}}.

Solution

Solution of Exercise 1.9.

(a) D(G)D(G) is normal with abelian quotient (Proposition 1.29); and if NGN \trianglelefteq G has G/NG/N abelian, the same proposition gives D(G)ND(G) \subseteq N: D(G)D(G) is the smallest. Universal property: let f ⁣:GAf \colon G \to A with AA abelian. Then f([x,y])=[f(x),f(y)]=ef([x,y]) = [f(x), f(y)] = e, so D(G)kerfD(G) \subseteq \ker f, and Theorem 1.3 factors f=fˉπf = \bar f \circ \pi through GabG^{\mathrm{ab}}, uniquely since π\pi is surjective.

(b) Commutators are even permutations, so D(Sn)AnD(S_n) \subseteq A_n. Conversely every 33-cycle is a commutator:

[(ab),(ac)]=(ab)(ac)(ab)(ac)=(abc),\bigl[(a\,b),\,(a\,c)\bigr] = (a\,b)(a\,c)(a\,b)(a\,c) = (a\,b\,c),

(direct check on a,b,ca, b, c), and 33-cycles generate AnA_n (Lemma 1.32): D(Sn)=AnD(S_n) = A_n for n3n \geq 3, and SnabSn/AnZ/2ZS_n^{\mathrm{ab}} \cong S_n/A_n \cong \Z/2\Z. (For n=2n = 2: S2S_2 is abelian, D(S2)={e}D(S_2) = \{e\}, S2ab=S2Z/2ZS_2^{\mathrm{ab}} = S_2 \cong \Z/2\Z — the formula SnabZ/2ZS_n^{\mathrm{ab}} \cong \Z/2\Z holds for all n2n \geq 2.)

Q8Q_8: the quotient Q8/{±1}Q_8/\{\pm 1\} has order 44, hence is abelian, so D(Q8){±1}D(Q_8) \subseteq \{\pm 1\}; and [i,j]=iji1j1=ij(i)(j)=(ij)2=k2=1[\mathrm i, \mathrm j] = \mathrm i \mathrm j \mathrm i^{-1}\mathrm j^{-1} = \mathrm i\mathrm j(-\mathrm i)(-\mathrm j) = (\mathrm i \mathrm j)^2 = \mathrm k^2 = -1, so D(Q8)={±1}D(Q_8) = \{\pm 1\} and Q8ab(Z/2Z)2Q_8^{\mathrm{ab}} \cong (\Z/2\Z)^2 (order 44, exponent 22: the classes of i,j\mathrm i, \mathrm j square to 1ˉ\bar 1).

Exercise 1.10 ★★

Let GG be a pp-group and HGH \subsetneq G a proper subgroup. Show that HNG(H)H \subsetneq N_G(H) (“normalizers grow”), and deduce that every maximal subgroup of a pp-group is normal of index pp. (Induction on G\abs G, using Z(G){e}Z(G) \neq \{e\}: treat separately Z(G)HZ(G) \subseteq H and Z(G)⊈HZ(G) \not\subseteq H.)

Solution

Solution of Exercise 1.10.

Induction on G\abs G; for G=p\abs G = p the only proper subgroup is H={e}H = \{e\}, and NG({e})=G{e}N_G(\{e\}) = G \supsetneq \{e\}. Let Z=Z(G){e}Z = Z(G) \neq \{e\} (Theorem 1.12).

If Z⊈HZ \not\subseteq H: pick zZHz \in Z \setminus H; zz commutes with HH, so zHz1=HzHz^{-1} = H and zNG(H)Hz \in N_G(H) \setminus H.

If ZHZ \subseteq H: pass to Gˉ=G/Z\bar G = G/Z, a pp-group of smaller order, and Hˉ=H/ZGˉ\bar H = H/Z \subsetneq \bar G (correspondence theorem). By induction, NGˉ(Hˉ)HˉN_{\bar G}(\bar H) \supsetneq \bar H; pick gˉNGˉ(Hˉ)Hˉ\bar g \in N_{\bar G}(\bar H) \setminus \bar H and a lift gg. Then gHg \notin H, and gHg1HZ=HgHg^{-1} \subseteq HZ = H: indeed ghg1=gˉhˉgˉ1Hˉ\overline{ghg^{-1}} = \bar g \bar h \bar g^{-1} \in \bar H means ghg1HZ=Hghg^{-1} \in HZ = H (as ZHZ \subseteq H). So gNG(H)Hg \in N_G(H)\setminus H.

Maximal subgroups: if MM is maximal, NG(M)MN_G(M) \supsetneq M forces NG(M)=GN_G(M) = G: MGM \trianglelefteq G. Then G/MG/M is a pp-group with no proper nontrivial subgroup (correspondence + maximality). Take xˉeˉ\bar x \neq \bar e in G/MG/M, of order pkp^k; then xˉpk1\bar x^{p^{k-1}} generates a subgroup of order pp, which must be everything: G/M=p\abs{G/M} = p.

Exercise 1.11 ★★★

(Simplicity of A5A_5, hands on) (a) Show that the conjugacy classes of A5A_5 have cardinalities 11, 1515, 2020, 1212, 1212. Pay attention to the splitting of the S5S_5-class of 55-cycles: for a 55-cycle σ\sigma, compare the centralizers of σ\sigma in S5S_5 and in A5A_5. (b) Deduce that A5A_5 is simple: a normal subgroup is a union of conjugacy classes, contains ee, and has cardinality dividing 6060. (c) Show that a simple group of order 6060 necessarily has n5=6n_5 = 6.

Solution

Solution of Exercise 1.11.

(a) A5=60\abs{A_5} = 60. Cycle types in A5A_5: ee; double transpositions, 12(51)(42)1=15\frac{1}{2}\binom{5}{1}\binom{4}{2}\cdot 1 = 15 of them (535 \cdot 3 ways: choose the fixed point, then pair up); 33-cycles, 5433=20\frac{5 \cdot 4 \cdot 3}{3} = 20; 55-cycles, 4!=244! = 24.

A class of S5S_5 contained in A5A_5 either stays one A5A_5-class or splits in two, according to whether the S5S_5-centralizer of an element contains an odd permutation (class in A5=60/ZA5(σ)\abs{\text{class in }A_5} = 60/\abs{Z_{A_5}(\sigma)} and ZA5=ZS5A5Z_{A_5} = Z_{S_5} \cap A_5). For σ=(12)(34)\sigma = (1\,2)(3\,4): ZS5(σ)=120/15=8\abs{Z_{S_5}(\sigma)} = 120/15 = 8, and (12)ZS5(σ)(1\,2) \in Z_{S_5}(\sigma) is odd, so ZA5=4\abs{Z_{A_5}} = 4 and the class has 60/4=1560/4 = 15 elements: no split. For σ=(123)\sigma = (1\,2\,3): ZS5(σ)σ×(45)Z_{S_5}(\sigma) \supseteq \langle \sigma \rangle \times \langle (4\,5)\rangle, of order 6=120/206 = 120/20, hence equal; it contains the odd (45)(4\,5): class of 60/3=2060/3 = 20: no split. For σ\sigma a 55-cycle: ZS5(σ)=σZ_{S_5}(\sigma) = \langle \sigma\rangle (order 120/24=5120/24 = 5), all even: ZA5(σ)=σZ_{A_5}(\sigma) = \langle \sigma\rangle and the A5A_5-class has 60/5=1260/5 = 12 elements — the 2424 five-cycles split into two classes of 1212. Class sizes: 1,15,20,12,121, 15, 20, 12, 12.

(b) A normal subgroup NN is a union of conjugacy classes including {e}\{e\}, with N60\abs N \mid 60. The possible sums 1+(subset of {15,20,12,12})1 + (\text{subset of } \{15, 20, 12, 12\}) are

1, 13, 13, 16, 21, 25, 28, 28, 33, 36, 40, 40, 45, 48, 48, 60;1,\ 13,\ 13,\ 16,\ 21,\ 25,\ 28,\ 28,\ 33,\ 36,\ 40,\ 40,\ 45,\ 48,\ 48,\ 60 ;

the only divisors of 6060 in the list are 11 and 6060: N={e}N = \{e\} or A5A_5.

(c) Let GG be simple with G=60\abs G = 60. n512n_5 \mid 12, n51(mod5)n_5 \equiv 1 \pmod 5: n5{1,6}n_5 \in \{1, 6\}. n5=1n_5 = 1 would make the Sylow 55-subgroup normal, contradicting simplicity (1<5<601 < 5 < 60). Hence n5=6n_5 = 6.

Exercise 1.12 ★★

(Normalizers of Sylow subgroups are self-normalizing) Let PP be a Sylow pp-subgroup of a finite group GG and H=NG(P)H = N_G(P). (a) Show that PP is the unique Sylow pp-subgroup of HH. (b) Deduce NG(H)=HN_G(H) = H. (For gNG(H)g \in N_G(H): gPg1gPg^{-1} is a Sylow pp-subgroup of HH, so gPg1=PgPg^{-1} = P.) (c) Conclude that no Sylow normalizer is contained in a proper normal subgroup of GG, and that a maximal subgroup containing NG(P)N_G(P) is self-normalizing.

Solution

Solution of Exercise 1.12.

(a) PP is normal in H=NG(P)H = N_G(P) by definition of the normalizer, and it is a Sylow pp-subgroup of HH (its order is already the full pp-part of G\abs G, a fortiori of H\abs H). A normal Sylow subgroup is unique: any other would be conjugate to it (Sylow II in HH), hence equal to it.

(b) Let gNG(H)g \in N_G(H). Then gPg1gHg1=HgPg^{-1} \subseteq gHg^{-1} = H is a subgroup of HH of the same order as PP: a Sylow pp-subgroup of HH, so gPg1=PgPg^{-1} = P by (a). Thus gNG(P)=Hg \in N_G(P) = H: NG(H)HN_G(H) \subseteq H, and the reverse inclusion is trivial.

(c) Suppose HNGH \subseteq N \trianglelefteq G with NN proper. PP is a Sylow pp-subgroup of NN; for any gGg \in G, gPg1NgPg^{-1} \subseteq N is another, so gPg1=nPn1gPg^{-1} = nPn^{-1} for some nNn \in N (Sylow II in NN), giving n1gNG(P)Nn^{-1}g \in N_G(P) \subseteq N and gNg \in N: N=GN = G, contradiction (this is the Frattini argument). For a maximal subgroup MNG(P)M \supseteq N_G(P): NG(M)MN_G(M) \supseteq M is either MM or GG; if GG, then MGM \trianglelefteq G is a proper normal subgroup containing NG(P)N_G(P) — excluded by the previous point. So NG(M)=MN_G(M) = M.

1.7 Problem: the groups of order at most 15

Problem 1.1

Weekend problem — classification of small groups

The aim is a complete classification, with full proofs, of the groups of order 15\leq 15 up to isomorphism. Orders 1,2,3,5,7,11,131, 2, 3, 5, 7, 11, 13 are settled by Lagrange (cyclic), and orders 44 and 99 by Theorem 1.12 plus the analysis below of p2p^2: there remain 6,8,10,12,14,156, 8, 10, 12, 14, 15.

Part I — Tools.

  1. Show that a group in which every element satisfies x2=ex^2 = e is abelian; deduce that such a finite group has order 2k2^k and is isomorphic to (Z/2Z)k(\Z/2\Z)^k. (View it as a vector space over F2\mathbb F_2.)
  2. Show that a group of order p2p^2 is isomorphic to Z/p2Z\Z/p^2\Z or (Z/pZ)2(\Z/p\Z)^2. List the abelian groups of order 88 up to isomorphism: Z/8Z\Z/8\Z, Z/4Z×Z/2Z\Z/4\Z \times \Z/2\Z, (Z/2Z)3(\Z/2\Z)^3 — prove the list is complete and irredundant without the structure theorem of Chapter 3 (discuss by the maximal order of an element).
  3. Let φ,φ ⁣:KAut(H)\varphi, \varphi' \colon K \to \operatorname{Aut}(H) be two actions. Show that if φ=φα\varphi' = \varphi \circ \alpha with αAut(K)\alpha \in \operatorname{Aut}(K), then HφKHφKH \rtimes_{\varphi} K \cong H \rtimes_{\varphi'} K.
  4. Determine Aut(Z/nZ)\operatorname{Aut}(\Z/n\Z) for n=3,4,5,7n = 3, 4, 5, 7 explicitly, and show Aut((Z/2Z)2)S3\operatorname{Aut}\bigl((\Z/2\Z)^2 \bigr) \cong S_3.

Part II — Orders 2p2p (66, 1010, 1414) and pqpq.

  1. Let G=2p\abs G = 2p with pp an odd prime. Show that GG has a normal subgroup N=rN = \langle r \rangle of order pp and an element ss of order 22 outside NN.
  2. Deduce GZ/pZφZ/2ZG \cong \Z/p\Z \rtimes_\varphi \Z/2\Z, where φ(1)Aut(Z/pZ)\varphi(1) \in \operatorname{Aut}(\Z/p\Z) is an involution, and conclude: GZ/2pZG \cong \Z/2p\Z or GDpG \cong D_p; check these two are not isomorphic. This settles orders 66, 1010, 1414.
  3. More generally, let G=pq\abs G = pq with p<qp < q primes. Show that if pq1p \nmid q-1 then GG is cyclic (Example 1.22), and that if pq1p \mid q - 1 there is, besides Z/pqZ\Z/pq\Z, exactly one nonabelian group Z/qZZ/pZ\Z/q\Z \rtimes \Z/p\Z up to isomorphism — use question 3 and the fact that Aut(Z/qZ)(Z/qZ)×\operatorname{Aut}(\Z/q\Z) \cong (\Z/q\Z)^\times is cyclic of order q1q - 1, admitted here and proved in Chapter 4 (cyclicity of Fq×\mathbb F_q^\times). Conclude for order 1515.

Part III — Order 88. Let GG be nonabelian of order 88.

  1. Show that GG has an element rr of order 44 (use question 1) and that N=rN = \langle r\rangle is normal.
  2. Let sNs \notin N. Show that s2Ns^2 \in N (Exercise 1.1(b)), that srs1=r1srs^{-1} = r^{-1} (examine the possible images of rr under conjugation, which must have order 4, and exclude srs1=rsrs^{-1} = r), and that s2{e,r2}s^2 \in \{e, r^2\} (what happens if s2=rs^2 = r or r3r^3? and why must s2s^2 commute with ss?).
  3. In the case s2=es^2 = e, show GD4G \cong D_4.
  4. In the case s2=r2s^2 = r^2, show that the multiplication table is entirely determined; the resulting group is the quaternion group Q8={±1,±i,±j,±k}Q_8 = \{\pm 1, \pm \mathrm i, \pm \mathrm j, \pm \mathrm k\}, i2=j2=k2=ijk=1\mathrm i^2 = \mathrm j^2 = \mathrm k^2 = \mathrm i\, \mathrm j\,\mathrm k = -1 (set r=ir = \mathrm i, s=js = \mathrm j). Verify that Q8Q_8 exists, e.g. inside GL2(C)GL_2(\C) via

    i(i00i),j(0110).\mathrm i \mapsto \begin{pmatrix} \iu & 0\\ 0 & -\iu \end{pmatrix}, \qquad \mathrm j \mapsto \begin{pmatrix} 0 & 1\\ -1 & 0 \end{pmatrix}.
  5. Show that every nontrivial subgroup of Q8Q_8 contains 1-1; deduce that every subgroup of Q8Q_8 is normal, that D4≇Q8D_4 \not\cong Q_8 (count elements of order 22), and that Q8Q_8 is not a semidirect product of two proper subgroups.

Part IV — Order 1212. Let G=12\abs G = 12, P3Syl3(G)P_3 \in \mathrm{Syl}_3(G), P2Syl2(G)P_2 \in \mathrm{Syl}_2(G).

  1. Show n3{1,4}n_3 \in \{1, 4\}, n2{1,3}n_2 \in \{1, 3\}, and that n3=4n_3 = 4 forces n2=1n_2 = 1 (count elements of order 33).
  2. Suppose n3=4n_3 = 4. The conjugation action on Syl3\mathrm{Syl}_3 gives ρ ⁣:GS4\rho \colon G \to S_4. Show that kerρ\ker \rho, contained in every NG(P3)N_G(P_3) and hence of order dividing 33, is trivial (why can it not have order 33?); that the image, a subgroup of order 1212 of S4S_4, is necessarily A4A_4 (index 2 subgroups are normal and contain all squares — Exercise 1.1; count the squares in S4S_4); and conclude GA4G \cong A_4.
  3. Suppose n3=1n_3 = 1, so GZ/3ZφP2G \cong \Z/3\Z \rtimes_\varphi P_2 with φ ⁣:P2Aut(Z/3Z)Z/2Z\varphi \colon P_2 \to \operatorname{Aut}(\Z/3\Z) \cong \Z/2\Z. Enumerate the cases: φ\varphi trivial yields Z/12Z\Z/12\Z and Z/6Z×Z/2Z\Z/6\Z \times \Z/2\Z; P2=Z/4ZP_2 = \Z/4\Z with φ\varphi surjective yields the dicyclic group Dic3=Z/3ZZ/4Z\mathrm{Dic}_3 = \Z/3\Z \rtimes \Z/4\Z; P2=(Z/2Z)2P_2 = (\Z/2\Z)^2 with φ\varphi surjective yields, up to the equivalence of question 3, a single group — show it is D6D_6, e.g. by exhibiting an element of order 66 and a reflection-like involution.
  4. Show that Z/12Z\Z/12\Z, Z/6Z×Z/2Z\Z/6\Z \times \Z/2\Z, D6D_6, A4A_4, Dic3\mathrm{Dic}_3 are pairwise non-isomorphic (count elements of order 22, or use n3n_3). This settles order 1212.

Part V — Synthesis.

  1. Assemble the classification table: for each order n15n \leq 15, the complete list of groups up to isomorphism, with the counts 1,1,1,2,1,2,1,5,2,2,1,5,1,2,11, 1, 1, 2, 1, 2, 1, 5, 2, 2, 1, 5, 1, 2, 1.

Part VI — Beyond: the groups of order p3p^3, pp odd. The order-88 analysis of Part III has a beautiful odd-prime analogue, with one genuinely new phenomenon. Let pp be an odd prime and GG nonabelian of order p3p^3.

  1. Show that Z(G)=p\abs{Z(G)} = p, that G/Z(G)(Z/pZ)2G/Z(G) \cong (\Z/p\Z)^2 (a cyclic quotient by the center forces abelianity: Exercise 1.2), and that D(G)=Z(G)D(G) = Z(G) (for D(G)Z(G)D(G) \subseteq Z(G), use that G/Z(G)G/Z(G) is abelian; for equality, GG is nonabelian and D(G){e}D(G) \neq \{e\}). Deduce that every commutator [x,y]=xyx1y1[x, y] = xyx^{-1}y^{-1} is central and of order dividing pp.
  2. (The key identity) Let x,yGx, y \in G and z=[y,x]z = [y, x], central. Prove by induction on kk:

    (xy)k=xkykzk(k1)/2.(xy)^k = x^k y^k z^{k(k-1)/2} .

    (Move each yy past each xx; every crossing costs one central factor zz.)

  3. Deduce that for pp odd the map θ ⁣:xxp\theta\colon x \mapsto x^p is a group morphism from GG to Z(G)Z(G) (why is xpx^p central? why does zp(p1)/2=ez^{p(p-1)/2} = e need pp odd?), and conclude that GG has exponent pp or p2p^2, the two cases being distinguished by whether θ\theta is trivial.
  4. (Exponent pp) Suppose every element satisfies xp=ex^p = e. Pick x,yx, y whose classes generate G/Z(G)G/Z(G) and set z=[y,x]z = [y, x]. Show that zez \neq e, that every element of GG is uniquely xaybzcx^ay^bz^c (0a,b,c<p0 \leq a, b, c < p), and that the multiplication is entirely determined by the relations xp=yp=zp=ex^p = y^p = z^p = e, zz central, [y,x]=z[y, x] = z. Verify that the Heisenberg group

    Hp={(1ac01b001):a,b,cFp}GL3(Fp)H_p = \left\{ \begin{pmatrix} 1 & a & c\\ 0 & 1 & b\\ 0 & 0 & 1 \end{pmatrix} : a, b, c \in \mathbb F_p \right\} \subseteq GL_3(\mathbb F_p)

    realizes these relations and has exponent pp (compute (I+N)p(I + N)^p with NN strictly upper triangular, using N3=0N^3 = 0 and p3p \geq 3): every exponent-pp nonabelian group of order p3p^3 is isomorphic to HpH_p.

  5. (Exponent p2p^2) Suppose some rGr \in G has order p2p^2, and set N=rN = \langle r\rangle, normal (index pp: Exercise 1.10). Show there is sNs \notin N with sp=es^p = e (take any tNt \notin N; using question 20, correct it: θ(t)=tpZ(G)N\theta(t) = t^p \in Z(G) \subseteq N — justify Z(G)=rpZ(G) = \langle r^p\rangle — and choose aa with s=tras = tr^{a} satisfying sp=es^p = e; where is pp odd used?). Show srs1=r1+psrs^{-1} = r^{1+p} up to replacing ss by a power, and conclude: there is exactly one nonabelian group of order p3p^3 and exponent p2p^2, namely Z/p2ZφZ/pZ\Z/p^2\Z \rtimes_\varphi \Z/p\Z with φ(1) ⁣:rr1+p\varphi(1)\colon r \mapsto r^{1+p} (use question 3; Aut(Z/p2Z)\operatorname{Aut}(\Z/p^2\Z) is cyclic of order p(p1)p(p-1), admitted here, so it has a unique subgroup of order pp).
  6. Conclude the count: for odd pp there are exactly 55 groups of order p3p^3 (three abelian, two nonabelian), just as for p=2p = 2 — but the two nonabelian ones are no longer D4D_4 and Q8Q_8. Pinpoint exactly where the odd-pp argument breaks for p=2p = 2: in the identity of question 19, zk(k1)/2z^{k(k-1)/2} for k=p=2k = p = 2 is z1ez^1 \neq e, so squaring is not a morphism — and indeed Q8Q_8 has a unique element of order 22 while exponent-44 D4D_4 has five.

Part VII — Complements.

  1. For pp odd, count the elements of order pp in each of the two nonabelian groups of order p3p^3: show that HpH_p has exactly p31p^3 - 1 of them, while Mp=Z/p2ZZ/pZM_p = \Z/p^2\Z \rtimes \Z/p\Z has exactly p21p^2 - 1 (use the morphism θ\theta of question 20: identify its image, then the order of its kernel). Verify numerically for p=3p = 3: 2626 against 88. Explain why no squaring-morphism argument of this kind can separate D4D_4 from Q8Q_8, and which count does separate them.
  2. Call an integer n1n \geq 1 cyclic if every group of order nn is cyclic. Show that if p2np^2 \mid n for some prime pp, or if nn has prime divisors p<qp < q with pq1p \mid q - 1, then nn is not cyclic (in each case exhibit a noncyclic group of order nn, using Part II for the second). Deduce that nn cyclic forces gcd(n,φ(n))=1\gcd(n, \varphi(n)) = 1, where φ\varphi is Euler’s totient, and check against the table of question 17: among n15n \leq 15, the orders carrying a single group are exactly n{1,2,3,5,7,11,13,15}n \in \{1, 2, 3, 5, 7, 11, 13, 15\}, precisely those with gcd(n,φ(n))=1\gcd(n, \varphi(n)) = 1.
Solution

Solution of Problem 1.1.

1. For x,yGx, y \in G: (xy)2=e(xy)^2 = e gives xy=(xy)1=y1x1=yxxy = (xy)^{-1} = y^{-1}x^{-1} = yx (each element is its own inverse): abelian. Such a GG, written additively, is a vector space over F2\mathbb F_2 (2x=02x = 0, and the axioms are the abelian group axioms); if finite, it has a finite basis: G(Z/2Z)kG \cong (\Z/2\Z)^k, of order 2k2^k.

2. Order p2p^2: GG is abelian (Theorem 1.12). If some element has order p2p^2, GG is cyclic. Otherwise all xex \neq e have order pp; additively GG is then a vector space over Fp\mathbb F_p (px=0px = 0), of dimension 22 (p2p^2 elements): G(Z/pZ)2G \cong (\Z/p\Z)^2.

Abelian of order 88, by the maximal order mm of an element: m=8m = 8: cyclic Z/8Z\Z/8\Z. m=2m = 2: (Z/2Z)3(\Z/2\Z)^3 by question 1. m=4m = 4: let xx have order 44 and yxy \notin \langle x \rangle; y2xy^2 \in \langle x\rangle (index 22). y2{x,x3}y^2 \in \{x, x^3\} would give yy order 88; so y2{e,x2}y^2 \in \{e, x^2\}. If y2=x2y^2 = x^2, replace yy by xyxy: (xy)2=x2y2=x4=e(xy)^2 = x^2y^2 = x^4 = e (GG abelian) and xyxxy \notin \langle x\rangle. So we may assume y2=ey^2 = e: then xy={e}\langle x \rangle \cap \langle y \rangle = \{e\}, both normal (abelian), xy=8\abs{\langle x\rangle \langle y\rangle} = 8: Proposition 1.24 gives GZ/4Z×Z/2ZG \cong \Z/4\Z \times \Z/2\Z. Irredundant: the numbers of solutions of x2=ex^2 = e are 2,4,82, 4, 8 in the three groups.

3. Define ψ ⁣:HφKHφK\psi \colon H \rtimes_{\varphi'} K \to H \rtimes_{\varphi} K by ψ(h,k)=(h,α(k))\psi(h, k) = (h, \alpha(k)), a bijection. Morphism:

ψ((h,k)(h,k))=(hφ(k)(h),α(kk))=(hφ(αk)(h),α(k)α(k))=ψ(h,k)ψ(h,k).\psi\bigl((h,k)(h',k')\bigr) = \bigl(h\,\varphi'(k)(h'),\, \alpha(kk')\bigr) = \bigl(h\,\varphi(\alpha k)(h'),\, \alpha(k)\alpha(k')\bigr) = \psi(h,k)\,\psi(h',k').

4. Aut(Z/nZ)(Z/nZ)×\operatorname{Aut}(\Z/n\Z) \cong (\Z/n\Z)^\times (Exercise 1.3): for n=3n = 3: {±1}Z/2Z\{\pm 1\} \cong \Z/2\Z; n=4n = 4: {1ˉ,3ˉ}Z/2Z\{\bar 1, \bar 3\} \cong \Z/2\Z; n=5n = 5: {1ˉ,2ˉ,3ˉ,4ˉ}\{\bar 1, \bar 2, \bar 3, \bar 4\}, cyclic of order 44 generated by 2ˉ\bar 2 (2,4,3,12, 4, 3, 1); n=7n = 7: cyclic of order 66 generated by 3ˉ\bar 3 (3,2,6,4,5,13, 2, 6, 4, 5, 1). For V=(Z/2Z)2V = (\Z/2\Z)^2: an automorphism is F2\mathbb F_2-linear (it preserves addition, and scalars are 0,10, 1), so Aut(V)=GL2(F2)\operatorname{Aut}(V) = GL_2(\mathbb F_2), of order (41)(42)=6(4-1)(4-2) = 6; it acts faithfully on the 33 nonzero vectors, giving an injective morphism to S3S_3 between groups of order 66: Aut(V)S3\operatorname{Aut}(V) \cong S_3.

5. Cauchy provides rr of order pp; N=rN = \langle r \rangle has index 22, hence is normal (Exercise 1.1). Cauchy also provides ss of order 22, and sNs \notin N (all non-identity elements of NN have odd order pp).

6. Ns={e}N \cap \langle s \rangle = \{e\} and Ns=2p\abs{N\langle s\rangle} = 2p (Exercise 1.4(a)): by Proposition 1.26, GZ/pZφZ/2ZG \cong \Z/p\Z \rtimes_\varphi \Z/2\Z with φ(1)=(xsxs1)\varphi(1) = (x \mapsto sxs^{-1}) an automorphism of order dividing 22. In (Z/pZ)×(\Z/p\Z)^\times, k2=1k^2 = 1 has only the solutions k=±1k = \pm 1 (X21X^2 - 1 has at most two roots in the field Fp\mathbb F_p). If φ(1)=id\varphi(1) = \mathrm{id}: the product is direct, GZ/pZ×Z/2ZZ/2pZG \cong \Z/p\Z \times \Z/2\Z \cong \Z/2p\Z. If φ(1)=id\varphi(1) = -\mathrm{id}: G=r,srp=s2=e, srs1=r1DpG = \langle r, s \mid r^p = s^2 = e, \ srs^{-1} = r^{-1}\rangle \cong D_p (Example 1.27). They are not isomorphic: DpD_p is nonabelian for p3p \geq 3 (srs1=r1rsrs^{-1} = r^{-1} \neq r).

7. nq1(modq)n_q \equiv 1 \pmod q divides p<qp < q: nq=1n_q = 1, so NZ/qZN \cong \Z/q\Z is normal. Let PZ/pZP \cong \Z/p\Z be a Sylow pp-subgroup: NP={e}N \cap P = \{e\}, NP=GNP = G (order pqpq), so GZ/qZφZ/pZG \cong \Z/q\Z \rtimes_\varphi \Z/p\Z with φ ⁣:Z/pZAut(Z/qZ)Z/(q1)Z\varphi \colon \Z/p\Z \to \operatorname{Aut}(\Z/q\Z) \cong \Z/(q-1)\Z (cyclic, admitted). If pq1p \nmid q - 1: the image of φ\varphi has order dividing both pp and q1q-1, hence is trivial, and GZ/pqZG \cong \Z/pq\Z (Example 1.22). If pq1p \mid q - 1: besides the trivial φ\varphi, any nontrivial φ\varphi is injective (its kernel, a subgroup of Z/pZ\Z/p\Z, is trivial) with image the unique subgroup CC of order pp of the cyclic group Z/(q1)Z\Z/(q-1)\Z. Two nontrivial actions φ,φ\varphi, \varphi' are then two isomorphisms Z/pZC\Z/p\Z \to C, so α=φ1φAut(Z/pZ)\alpha = \varphi^{-1}\circ\varphi' \in \operatorname{Aut}(\Z/p\Z) satisfies φ=φα\varphi' = \varphi\circ\alpha: by question 3 the two semidirect products are isomorphic. Hence exactly one nonabelian group of order pqpq (nonabelian since φid\varphi \neq \mathrm{id} makes some conjugation nontrivial). Order 1515: p=3p = 3, q=5q = 5, 343 \nmid 4: cyclic only.

8. Not every element has order 2\leq 2 (else abelian by question 1), and no element has order 88 (else cyclic, abelian): some rr has order 44, and N=rN = \langle r\rangle, of index 22, is normal.

9. s2Ns^2 \in N by Exercise 1.1(b). The conjugate srs1Nsrs^{-1} \in N has order 44, so srs1{r,r3}srs^{-1} \in \{r, r^3\}; if srs1=rsrs^{-1} = r then rr and ss commute and G=r,sG = \langle r, s\rangle is abelian — excluded. So srs1=r1srs^{-1} = r^{-1}. If s2=rs^2 = r or r3r^3, then ss has order 88: excluded. (Alternatively: s2s^2 commutes with ss, but srs1=r1s r s^{-1} = r^{-1} and sr3s1=r3=rs r^3 s^{-1} = r^{-3} = r: neither rr nor r3r^3 is fixed by conjugation by ss.) So s2{e,r2}s^2 \in \{e, r^2\}.

10. If s2=es^2 = e: G=r,sr4=s2=e, srs1=r1G = \langle r, s \mid r^4 = s^2 = e,\ srs^{-1} = r^{-1}\rangle. The eight elements risjr^is^j (0i<40 \leq i < 4, 0j<20 \leq j < 2) are distinct (srs \notin \langle r\rangle) and the relations determine all products: the assignment rr \mapsto (rotation by π/2\pi/2), ss \mapsto (a reflection) defines a surjective morphism onto D4D_4, between groups of order 88: an isomorphism.

11. If s2=r2s^2 = r^2: again G={risj}G = \{r^i s^j\} and the relations r4=er^4 = e, s2=r2s^2 = r^2, srs1=r1srs^{-1} = r^{-1} force the whole table. With i=r\mathrm i = r, j=s\mathrm j = s, k=rs\mathrm k = rs, 1=r2-1 = r^2: i2=j2=1\mathrm i^2 = \mathrm j^2 = -1, k2=rsrs=rr1ss=s2=1\mathrm k^2 = rsrs = r r^{-1} s s = s^2 = -1 (using sr=r1ssr = r^{-1}s), and ijk=rsrs=rr1ss=s2=1\mathrm i \mathrm j \mathrm k = r\,s\,rs = r\,r^{-1}s\,s = s^2 = -1. Existence: the matrices

A=(i00i),B=(0110)A = \begin{pmatrix} \iu & 0\\ 0 & -\iu \end{pmatrix}, \qquad B = \begin{pmatrix} 0 & 1\\ -1 & 0 \end{pmatrix}

satisfy A4=IA^4 = I, B2=I=A2B^2 = -I = A^2 and BAB1=A1BAB^{-1} = A^{-1} — for the last one, check

BA=(0ii0)=A1B.BA = \begin{pmatrix} 0 & -\iu\\ -\iu & 0 \end{pmatrix} = A^{-1}B .

So {±I,±A,±B,±AB}\{\pm I, \pm A, \pm B, \pm AB\} is a group of order 88 realizing the table: Q8Q_8 exists.

12. Let H{e}H \neq \{e\} be a subgroup and xH{e}x \in H \setminus \{e\}. If x1x \neq -1 then x{±i,±j,±k}x \in \{\pm\mathrm i, \pm\mathrm j, \pm\mathrm k\} and x2=1Hx^2 = -1 \in H. So 1H-1 \in H always. The subgroups are {e}\{e\}, {±1}\{\pm 1\} (the center), i,j,k\langle \mathrm i\rangle, \langle \mathrm j\rangle, \langle \mathrm k\rangle (index 22) and Q8Q_8: all normal ({e}\{e\} and the center trivially, index 22 by Exercise 1.1, Q8Q_8 itself). D4D_4 has five elements of order 22 (r2r^2 and the four reflections), Q8Q_8 only one (1-1): not isomorphic. A semidirect product HKH \rtimes K with H,K{e}H, K \neq \{e\} requires HK={e}H \cap K = \{e\}, impossible since both contain 1-1.

13. n34n_3 \mid 4, n31(mod3)n_3 \equiv 1 \pmod 3: n3{1,4}n_3 \in \{1, 4\}; n23n_2 \mid 3, odd: n2{1,3}n_2 \in \{1, 3\}. If n3=4n_3 = 4: the four Sylow 33-subgroups pairwise intersect trivially (prime order), giving 4×2=84 \times 2 = 8 elements of order 33; the remaining 44 elements must constitute the unique Sylow 22-subgroup: n2=1n_2 = 1.

14. kerρ\ker\rho normalizes every Sylow 33-subgroup, so kerρNG(P3)\ker \rho \subseteq N_G(P_3), which has index n3=4n_3 = 4, i.e. order 33: kerρ{1,3}\abs{\ker\rho} \in \{1, 3\}. Order 33 would make kerρ\ker\rho a normal Sylow 33-subgroup, contradicting n3=4n_3 = 4. So ρ\rho is injective and its image HS4H \leq S_4 has order 1212, index 22: HS4H \trianglelefteq S_4 and HH contains all squares (Exercise 1.1(b)). The squares of S4S_4 include ee and all eight 33-cycles (σ=(σ2)2\sigma = (\sigma^2)^2 for a 33-cycle), which generate A4A_4 (they lie in A4A_4, and together with their products give all twelve elements; or: Lemma 1.32 for n=4n = 4’s generation part, which only needs n3n \geq 3). So A4HA_4 \subseteq H and A4=H\abs{A_4} = \abs H: GH=A4G \cong H = A_4.

15. P3GP_3 \trianglelefteq G, P3P2={e}P_3 \cap P_2 = \{e\}, P3P2=GP_3P_2 = G: GZ/3ZφP2G \cong \Z/3\Z \rtimes_\varphi P_2 (Proposition 1.26), φ ⁣:P2Aut(Z/3Z)={±id}Z/2Z\varphi \colon P_2 \to \operatorname{Aut}(\Z/3\Z) = \{\pm\mathrm{id}\} \cong \Z/2\Z.

  • φ\varphi trivial: direct products Z/3Z×Z/4ZZ/12Z\Z/3\Z \times \Z/4\Z \cong \Z/12\Z and Z/3Z×(Z/2Z)2Z/6Z×Z/2Z\Z/3\Z \times (\Z/2\Z)^2 \cong \Z/6\Z \times \Z/2\Z.
  • P2=Z/4ZP_2 = \Z/4\Z, φ\varphi surjective: necessarily φ(1)=id\varphi(1) = -\mathrm{id} (the only nontrivial choice): one group, Dic3=Z/3ZZ/4Z\mathrm{Dic}_3 = \Z/3\Z \rtimes \Z/4\Z.
  • P2=(Z/2Z)2P_2 = (\Z/2\Z)^2, φ\varphi surjective: kerφ\ker\varphi is one of the three subgroups of order 22; the three resulting φ\varphi differ by automorphisms of (Z/2Z)2(\Z/2\Z)^2 permuting these subgroups (question 4: AutS3\operatorname{Aut} \cong S_3 acts transitively on the three involutions), so by question 3 they give one isomorphism class. It is D6D_6: pick tt generating kerφ\ker\varphi and xx generating Z/3Z\Z/3\Z; the element ρ=(x,t)\rho = (x, t) satisfies ρ2=(2x,0)\rho^2 = (2x, 0), ρ3=(0,t)\rho^3 = (0, t), ρ6=e\rho^6 = e and no smaller power is ee: order 66; for s=(0,u)s = (0, u) with ukerφu \notin \ker\varphi: s2=es^2 = e and sρs1=(x,t)=ρ1s\rho s^{-1} = (-x, t) = \rho^{-1}. As ρ,s\langle \rho, s\rangle has order 1212, GD6G \cong D_6.

16. Counting elements of order 22: Z/12Z\Z/12\Z has 11; Z/6Z×Z/2Z\Z/6\Z\times\Z/2\Z has 33; D6D_6 has 77 (six reflections and the half-turn ρ3\rho^3); A4A_4 has 33; Dic3\mathrm{Dic}_3 has 11 (only (0,2)(0, 2): an element (h,k)(h, k) with kk of order 44 in Z/4Z\Z/4\Z has order 44). This separates all but the pairs {Z/12Z,Dic3}\{\Z/12\Z, \mathrm{Dic}_3\} and {Z/6Z×Z/2Z,A4}\{\Z/6\Z\times\Z/2\Z, A_4\}: the first members are abelian, the second not (Dic3\mathrm{Dic}_3: the action is nontrivial; A4A_4: (123)(1\,2\,3) and (12)(34)(1\,2)(3\,4) do not commute). Five distinct groups; parts II–IV show the list is complete.

17. The classification table:

nngroups of order nn#
11{e}\{e\}11
22Z/2Z\Z/2\Z11
33Z/3Z\Z/3\Z11
44Z/4Z\Z/4\Z, (Z/2Z)2(\Z/2\Z)^222
55Z/5Z\Z/5\Z11
66Z/6Z\Z/6\Z, S3=D3S_3 = D_322
77Z/7Z\Z/7\Z11
88Z/8Z\Z/8\Z, Z/4Z×Z/2Z\Z/4\Z{\times}\Z/2\Z, (Z/2Z)3(\Z/2\Z)^3, D4D_4, Q8Q_855
99Z/9Z\Z/9\Z, (Z/3Z)2(\Z/3\Z)^222
1010Z/10Z\Z/10\Z, D5D_522
1111Z/11Z\Z/11\Z11
1212Z/12Z\Z/12\Z, Z/6Z×Z/2Z\Z/6\Z{\times}\Z/2\Z, D6D_6, A4A_4, Dic3\mathrm{Dic}_355
1313Z/13Z\Z/13\Z11
1414Z/14Z\Z/14\Z, D7D_722
1515Z/15Z\Z/15\Z11

Orders 6,10,146, 10, 14 are Part II with p=3,5,7p = 3, 5, 7; order 1515 is question 7; order 88 is Part III together with question 2; order 1212 is Part IV; prime orders are Lagrange; orders 44 and 99 are question 2.

18. Z=Z(G)Z = Z(G) is nontrivial (Theorem 1.12) and ZGZ \neq G (nonabelian), so Z{p,p2}\abs Z \in \{p, p^2\}. If Z=p2\abs Z = p^2, then G/ZG/Z is cyclic of order pp and Exercise 1.2 makes GG abelian: excluded, so Z=p\abs Z = p and G/Z=p2\abs{G/Z} = p^2. By question 2, G/ZG/Z is Z/p2Z\Z/p^2\Z or (Z/pZ)2(\Z/p\Z)^2; cyclic is again excluded by Exercise 1.2: G/Z(Z/pZ)2G/Z \cong (\Z/p\Z)^2. Since G/ZG/Z is abelian, every commutator lies in ZZ: D(G)ZD(G) \subseteq Z; and D(G){e}D(G) \neq \{e\} (GG nonabelian), so D(G)=ZD(G) = Z (Z=p\abs Z = p leaves no room). Commutators are central of order dividing Z=p\abs Z = p.

19. Induction on kk, the case k=1k = 1 being trivial. Using yx=xyz1yx = xyz^{-1}\cdot — precisely, z=[y,x]=yxy1x1z = [y, x] = yxy^{-1}x^{-1} gives yx=zxyyx = zxy, i.e. moving one yy leftward past one xx produces one factor zz, which is central and can be parked anywhere. Then

(xy)k+1=(xy)kxy=xkykzk(k1)/2xy=xk(ykx)yzk(k1)/2=xk+1yk+1zk(k1)/2+k,(xy)^{k+1} = (xy)^k\,xy = x^ky^kz^{k(k-1)/2}\,xy = x^k\,(y^kx)\,y\,z^{k(k-1)/2} = x^{k+1}y^{k+1}\,z^{k(k-1)/2 + k},

since carrying xx past yky^k costs kk factors of zz (ykx=zkxyky^kx = z^kxy^k, by kk applications of yx=zxyyx = zxy); and k(k1)/2+k=k(k+1)/2k(k-1)/2 + k = k(k+1)/2.

20. With k=pk = p: (xy)p=xpypzp(p1)/2(xy)^p = x^py^pz^{p(p-1)/2}. For pp odd, (p1)/2(p-1)/2 is an integer, so zp(p1)/2=(zp)(p1)/2=ez^{p(p-1)/2} = (z^p)^{(p-1)/2} = e (question 18: zz has order dividing pp): θ(xy)=θ(x)θ(y)\theta(xy) = \theta(x)\theta(y), a morphism. Its values are central: the class of xx in G/Z(Z/pZ)2G/Z \cong (\Z/p\Z)^2 has order dividing pp, so xpZx^p \in Z. If θ\theta is trivial, every element has order dividing pp: exponent pp (not 11: G{e}G \neq \{e\}). Otherwise some xpex^p \neq e, and xx has order p2p^2 (order divides p3p^3, and xx cannot have order p3p^3: GG would be cyclic, hence abelian): exponent p2p^2.

21. Classes xˉ,yˉ\bar x, \bar y generating G/ZG/Z: their commutator z=[y,x]z = [y, x] is e\neq e, else x,y,Zx, y, Z would generate an abelian GG (their classes generate the quotient and ZZ is central) — and zz generates ZZ (Z=p\abs Z = p). Every gGg \in G has class xˉayˉb\bar x^a\bar y^b for unique 0a,b<p0 \leq a, b < p, so g=xaybzcg = x^ay^bz^c with a unique 0c<p0 \leq c < p: p3p^3 elements, all accounted for. Products of such normal forms are computed using only yx=zxyyx = zxy, zz central, and xp=yp=zp=ex^p = y^p = z^p = e: the table is forced, so any two exponent-pp nonabelian groups of order p3p^3 are isomorphic (match the generators). The Heisenberg group realizes the relations: with X=I+E12X = I + E_{12}, Y=I+E23Y = I + E_{23}, one computes [Y,X]=IE13[Y, X] = I - E_{13} (central in HpH_p), and for any strictly upper triangular NN, N3=0N^3 = 0 gives

(I+N)p=I+pN+(p2)N2=Iin characteristic p, p3,(I + N)^p = I + pN + \binom p2N^2 = I \qquad\text{in characteristic } p,\ p \geq 3,

since ppp \mid p and p(p2)p \mid \binom p2 for odd pp: exponent pp. So the exponent-pp group is HpH_p.

22. Z(G)=rpZ(G) = \langle r^p\rangle: indeed rpr^p is central (question 20 argument: the class of rr in the exponent-pp quotient G/ZG/Z gives rpZr^p \in Z) and is e\neq e, so it generates the order-pp center. Take any tNt \notin N. If tp=et^p = e, set s=ts = t. Otherwise θ(t)=tpZ=rp\theta(t) = t^p \in Z = \langle r^p\rangle, say tp=rpbt^p = r^{pb}; set s=trbs = tr^{-b}: by question 20 (θ\theta a morphism, pp odd), sp=tprpb=es^p = t^pr^{-pb} = e, and sNs \notin N. Conjugation: srs1Nsrs^{-1} \in N (NN normal) has order p2p^2, so srs1=rmsrs^{-1} = r^m with pmp \nmid m; also sp=es^p = e forces mpm(modp2)m^p \equiv m \pmod{p^2} — conjugating pp times returns rr, so mp1(modp2)m^p \equiv 1 \pmod {p^2}, and mmp1(modp)m \equiv m^p \equiv 1 \pmod p (Fermat): m=1+apm = 1 + ap. Nontriviality (GG nonabelian) gives a≢0a \not\equiv 0; replacing ss by the power sas^{a'} with aa1(modp)aa' \equiv 1 \pmod p turns the action into rr1+pr \mapsto r^{1+p}. This presents GG as Z/p2ZφZ/pZ\Z/p^2\Z\rtimes_\varphi\Z/p\Z with φ(1) ⁣:rr1+p\varphi(1)\colon r \mapsto r^{1+p}; by question 3, any two nontrivial morphisms Z/pZAut(Z/p2Z)\Z/p\Z \to \operatorname{Aut}(\Z/p^2\Z) with the same image — and the image is the unique subgroup of order pp of the cyclic Aut(Z/p2Z)\operatorname{Aut}(\Z/p^2\Z) — give isomorphic semidirect products: uniqueness.

23. Abelian: Z/p3Z\Z/p^3\Z, Z/p2Z×Z/pZ\Z/p^2\Z\times\Z/p\Z, (Z/pZ)3(\Z/p\Z)^3 (question 2’s argument, one degree up: classify by maximal order). Nonabelian: exactly HpH_p (exponent pp, question 21) and Z/p2ZZ/pZ\Z/p^2\Z\rtimes\Z/p\Z (exponent p2p^2, question 22), distinguished by their exponents. Total: five. For p=2p = 2 the morphism argument of question 20 collapses: zp(p1)/2=z1=zez^{p(p-1)/2} = z^{1} = z \neq e, squaring is not a morphism, and indeed both nonabelian groups of order 88 have exponent 44 — the invariant that separates D4D_4 from Q8Q_8 is the number of elements of order 22 (five against one), not the exponent. The odd-pp world is, for once, tidier than characteristic 22.

24. In HpH_p every element e\neq e has order pp (exponent pp, question 21): p31p^3 - 1 elements of order pp. In MpM_p, the map θ ⁣:xxp\theta \colon x \mapsto x^p is a morphism MpZ(Mp)=rpM_p \to Z(M_p) = \langle r^p\rangle (question 20, pp odd); θ(r)=rpe\theta(r) = r^p \neq e, so the image is the whole order-pp center and kerθ={x:xp=e}\ker\theta = \{x : x^p = e\} has order p3/p=p2p^3/p = p^2. The elements of order pp are the nonidentity elements of this kernel: p21p^2 - 1 of them. For p=3p = 3: H3H_3 has 271=2627 - 1 = 26 elements of order 33, and M3=Z/9ZZ/3ZM_3 = \Z/9\Z \rtimes \Z/3\Z has 91=89 - 1 = 8. For p=2p = 2 the argument dies at the start: squaring is not a morphism on a nonabelian group of order 88 (question 23), and indeed the set {x:x2=e}\{x : x^2 = e\} has 66 elements in D4D_4 — not the order of a subgroup of D4D_4. The count that does separate the pair is the number of elements of order 22: five in D4D_4, one in Q8Q_8 (question 11).

25. If p2np^2 \mid n, the group Z/pZ×Z/(n/p)Z\Z/p\Z \times \Z/(n/p)\Z has order nn and is not cyclic: every element’s order divides lcm(p,n/p)=n/p<n\operatorname{lcm}(p, n/p) = n/p < n, since pn/pp \mid n/p. If p<qp < q are primes dividing nn with pq1p \mid q - 1, question 7 provides a nonabelian group Z/qZZ/pZ\Z/q\Z \rtimes \Z/p\Z of order pqpq; then (Z/qZZ/pZ)×Z/(n/pq)Z(\Z/q\Z \rtimes \Z/p\Z) \times \Z/(n/pq)\Z has order nn and is nonabelian, hence not cyclic. Now suppose gcd(n,φ(n))>1\gcd(n, \varphi(n)) > 1 and pick a prime pp dividing both. Writing φ(n)=qanqa1(q1)\varphi(n) = \prod_{q^a \parallel n} q^{a-1}(q - 1), the divisibility pφ(n)p \mid \varphi(n) means either p2np^2 \mid n (the factor qa1q^{a-1} with q=pq = p, a2a \geq 2) or pq1p \mid q - 1 for some prime qnq \mid n, qpq \neq p: in both cases nn is not cyclic by the above. By contraposition, nn cyclic forces gcd(n,φ(n))=1\gcd(n, \varphi(n)) = 1. Check for n15n \leq 15: the values φ(n)\varphi(n) for n=1,,15n = 1, \dots, 15 are 1,1,2,2,4,2,6,4,6,4,10,4,12,6,81, 1, 2, 2, 4, 2, 6, 4, 6, 4, 10, 4, 12, 6, 8, and gcd(n,φ(n))=1\gcd(n, \varphi(n)) = 1 exactly for n=1,2,3,5,7,11,13,15n = 1, 2, 3, 5, 7, 11, 13, 15 — exactly the entries of the table of question 17 with a single group. The other orders are witnessed noncyclic as above: 4,8,9,124, 8, 9, 12 by a square factor, 6,10,12,146, 10, 12, 14 by 2q12 \mid q - 1. (The converse — gcd(n,φ(n))=1\gcd(n, \varphi(n)) = 1 implies nn cyclic — is also true; question 7 proves its first nontrivial case, n=pqn = pq with pq1p \nmid q - 1.)