Bourbaki

§ 6. Extensions, solvable groups, nilpotent groups

Throughout this paragraph, the group laws are, unless expressly mentioned otherwise, written multiplicatively.

1. Extensions

Definition 1

Let FF and GG be two groups. An extension of GG by FF is a triple E=(E,i,p)\mathscr{E} = (E, i, p), where EE is a group, ii an injective homomorphism of FF into EE and pp a surjective homomorphism of EE onto GG such that Im⁡(i)=Ker⁡(p)\operatorname{Im}(i) = \operatorname{Ker}(p). A homomorphism s:G→Es : G \to E (resp. r:E→Fr : E \to F) such that p∘s=Id⁡Gp \circ s = \operatorname{Id}_G (resp. r∘i=Id⁡Fr \circ i = \operatorname{Id}_F) is called a section (resp. retraction) of the extension E\mathscr{E}.

An extension E=(E,i,p)\mathscr{E} = (E, i, p) of GG by FF is often denoted by the diagram E:F→iE→pG\mathscr{E} : F \xrightarrow{i} E \xrightarrow{p} G, in which ii and pp are sometimes omitted if no confusion can arise. It is sometimes said simply that the group EE is an extension of GG by FF.

For a group EE to be an extension of GG by FF, it is necessary and sufficient that it contain a normal subgroup F′F' isomorphic to FF such that the quotient group E/F′E/F' is isomorphic to GG.

An extension E:F→iE→pG\mathscr{E} : F \xrightarrow{i} E \xrightarrow{p} G is called central if the image i(F)i(F) is contained in the centre of EE; this is only possible if FF is commutative.

Let E:F→iE→pG\mathscr{E} : F \xrightarrow{i} E \xrightarrow{p} G and E′:F→i′E′→p′G\mathscr{E}' : F \xrightarrow{i'} E' \xrightarrow{p'} G be two extensions of GG by FF. A morphism of E\mathscr{E} into E′\mathscr{E}' is a homomorphism u:E→E′u : E \to E' such that p′∘u=pp' \circ u = p and u∘i=i′u \circ i = i', or, in other words, such that the following diagram is commutative:

EiF→i′E′↓↓EpG \begin{array}{ccc} & & E \\ & i & \\ F & \xrightarrow{i'} & E' \\ & \downarrow & \downarrow \\ & & E \\ & p & \\ & & G \end{array}
u↑andp′↑ u \uparrow \quad \text{and} \quad p' \uparrow

Proposition 1

Let E:F→iE→pG\mathscr{E}: F \xrightarrow{i} E \xrightarrow{p} G and E′:F→i′E′→p′G\mathscr{E}': F \xrightarrow{i'} E' \xrightarrow{p'} G be extensions of GG by FF. If u:E→E′u: E \to E' is a morphism of E\mathscr{E} into E′\mathscr{E}', uu is an isomorphism of EE onto E′E' and u−1u^{-1} is a morphism of E′\mathscr{E}' into E\mathscr{E}.

Let x∈Ex \in E be such that u(x)=eu(x) = e. Then p(x)=p′(u(x))=ep(x) = p'(u(x)) = e, whence x∈i(F)x \in i(F). Let y∈Fy \in F be such that x=i(y)x = i(y); then i′(y)=u(i(y))=ei'(y) = u(i(y)) = e. As i′i' is injective, y=ey = e and x=ex = e. Therefore uu is injective. By virtue of § 4, no. 6, Corollary 1 to Proposition 7, uu is surjective since u(i(F))=i′(F)u(i(F)) = i'(F). The last assertion is immediate.

In other words, the extensions E\mathscr{E} and E′\mathscr{E}' are isomorphic if and only if there exists a morphism of E\mathscr{E} into E′\mathscr{E}'.

Let FF and GG be two groups and write E0=F×GE_0 = F \times G; let i:F→E0i: F \to E_0 be the canonical injection and p:E0→Gp: E_0 \to G be the canonical projection. Every extension of GG by FF isomorphic to the extension E0:F→iE0→pG\mathscr{E}_0: F \xrightarrow{i} E_0 \xrightarrow{p} G is called a trivial extension.

Proposition 2

Let E:F→iE→pG\mathscr{E}: F \xrightarrow{i} E \xrightarrow{p} G be an extension of GG by FF. The following conditions are equivalent*:

(i) E\mathscr{E} is a trivial extension; (ii) E\mathscr{E} has a retraction rr; (iii) E\mathscr{E} has a section ss such that s(G)s(G) is contained in the centralizer of i(F)i(F).

Clearly (i) implies (ii) and (iii). If (ii) holds, the mapping (r,p):E→F×G(r, p): E \to F \times G is a morphism of E\mathscr{E} into E0\mathscr{E}_0, whence (i). If (iii) holds, the homomorphism of F×GF \times G into EE corresponding to (i,s)(i, s) (§\S 4, no. 9, Proposition 12) is a morphism of E0\mathscr{E}_0 into E\mathscr{E}, whence (i).

It may be that an extension E:F→E→G\mathscr{E}: F \to E \to G is not trivial and yet the group EE is isomorphic to F×GF \times G (Exercise 6).

Definition 2

Let FF and GG be two groups and τ\tau a homomorphism of GG into the automorphism group of FF. Write τ(g)(f)=gf\tau(g)(f) = gf for g∈Gg \in G and f∈Ff \in F. The set F×GF \times G with the law of composition

((f,g),(f′,g′))↦(f,g)⋅τ(f′,g′)=(f⋅gf′,gg′) ((f, g), (f', g')) \mapsto (f, g) \cdot_{\tau} (f', g') = (f \cdot gf', gg')

is called the external semi-direct product of GG by FF relative to τ\tau.

The external semi-direct product of GG by FF relative to τ\tau is denoted by F×τGF \times_{\tau} G.

Proposition 3

The external semi-direct product F×τGF \times_{\tau} G is a group. The mappings i:F→F×τGi: F \to F \times_{\tau} G defined by i(f)=(f,e)i(f) = (f, e), p:F×τG→Gp: F \times_{\tau} G \to G defined by p(f,g)=gp(f, g) = g, and s:G→F×τGs: G \to F \times_{\tau} G defined by s(g)=(e,g)s(g) = (e, g) are group homomorphisms. The triple (F×τG,i,p)(F \times_{\tau} G, i, p) is an extension of GG by FF and ss is a section of the extension.

We have:

((f,g)⋅τ(f′,g′))⋅τ(f′′,g′′)=(f.gf′,gg′)⋅τ(f′′,g′′)=(f.gf′.gg′f′′,gg′g′′); ((f, g) \cdot_{\tau} (f', g')) \cdot_{\tau} (f'', g'') = (f.gf', gg') \cdot_{\tau} (f'', g'') = (f.gf'.gg'f'', gg'g'');
(f,g)⋅τ((f′,g′)⋅τ(f′′,g′′))=(f,g)⋅τ(f′g′f′′,g′g′′)=(f.g(f′.g′f′′),gg′g′′). (f, g) \cdot_{\tau} ((f', g') \cdot_{\tau} (f'', g'')) = (f, g) \cdot_{\tau} (f'g'f'', g'g'') = (f.g(f'.g'f''), gg'g'').

Now g(f′.g′f′′)=gf′.gg′f′′g(f'.g'f'') = gf'.gg'f'', which shows that the law of composition defined by (1) is associative. The element (e,e)(e, e) is the identity under this law. The element (f,g)(f, g) admits as inverse (g−1f−1,g−1)(g^{-1}f^{-1}, g^{-1}). Hence the law of composition on F×τGF \times_{\tau} G is a group law. The other assertions are immediate.

Using the notation of Proposition 3, Eτ\mathscr{E}_{\tau} will denote the extension

F→iF×τG→pG. F \xrightarrow{i} F \times_{\tau} G \xrightarrow{p} G.

Let E′:F→i′E′→p′G\mathscr{E}' : F \xrightarrow{i'} E' \xrightarrow{p'} G be an extension of GG by FF and s′:G→E′s' : G \to E' a section of E′\mathscr{E}'. We define an operation τ\tau of GG on the group FF by:

(2)

i′(τ(g,f))=s′(g)i′(f)s′(g)−1=Int⁡(s′(g))(i′(f)). i'(\tau(g, f)) = s'(g)i'(f)s'(g)^{-1} = \operatorname{Int}(s'(g))(i'(f)).

Proposition 4

With the above notation, there exists one and only one isomorphism u of Eτ\mathscr{E}_{\tau} onto E′\mathscr{E}' such that u∘s=s′u \circ s = s'.

(f,g)=(f,e)⋅τ(e,g)=i(f)⋅τs(g).(f, g) = (f, e) \cdot_{\tau} (e, g) = i(f) \cdot_{\tau} s(g).
Therefore, if uu is a solution to the problem, of necessity u(f,g)=i′(f).s′(g)u(f, g) = i'(f).s'(g), whence the uniqueness of uu. We prove the existence. We write u(f,g)=i′(f).s′(g)u(f, g) = i'(f).s'(g). Then

u(f,g).u(f′,g′)=i′(f)s′(g)i′(f′)s′(g′)=i′(f)(s′(g)i′(f′)s′(g)−1)s′(g)s′(g′)=i′(f)i′(τ(g,f′)).s′(g)s′(g′)=i′(f.τ(g,f′)).s′(gg′)=u((f,g)⋅τ(f′,g′)). \begin{align*} u(f, g).u(f', g') &= i'(f)s'(g)i'(f')s'(g') \\ &= i'(f)(s'(g)i'(f')s'(g)^{-1})s'(g)s'(g') \\ &= i'(f)i'(\tau(g, f')).s'(g)s'(g') \\ &= i'(f.\tau(g, f')).s'(gg') \\ &= u((f, g) \cdot_{\tau} (f', g')). \end{align*}

Therefore, uu is a homomorphism of F×τGF \times_{\tau} G into E′E'. Obviously u∘i=i′u \circ i = i', p′∘u=pp' \circ u = p and u∘s=s′u \circ s = s'.

Remark

The definition of the operation τ\tau by formula (2) depends on the extension E′\mathscr{E}' and the section s′s'. When FF is commutative, the operation τ\tau does not depend on s′s'. For Int⁡(s′(g))∣i′(F)\operatorname{Int}(s'(g)) | i'(F) depends then only on the coset of s′(g)s'(g) mod. i′(F)i'(F).

More generally, let E:F→E→G\mathscr{E} : F \to E \to G be an extension of GG by a commutative group FF (it is not assumed that E\mathscr{E} admits a section). The group EE operates on FF by inner automorphisms, this image is trivial on the image of FF and hence defines an operation of GG on FF. If E\mathscr{E} admits a section, this operation is that defined by formula (2).

Corollary

Let GG be a group and HH and KK two subgroups of GG such that HH is normal, H∩K={e}H \cap K = \{e\} and H.K=GH.K = G. Let τ\tau be the operation of KK on HH by inner automorphisms of GG. The mapping (h,k)↦hk(h, k) \mapsto hk is an isomorphism of H×τKH \times_{\tau} K onto GG.

Under the hypotheses of this corollary, GG is said to be the semi-direct product of KK by HH.

Example

(1) Let GG be a group and EE a homogeneous principal GG-set; let Γ\Gamma denote the automorphism group of GG. Let AA be the set of permutations ff of EE with the following property: There exists γ∈Γ\gamma \in \Gamma such that ff is a γ\gamma-morphism of EE into EE (that is, f(gb)=γ(g)f(b)f(gb) = \gamma(g)f(b) for b∈Eb \in E and g∈Gg \in G). The above formula f(gb)=γ(g)f(b)f(gb) = \gamma(g)f(b) shows that if f∈Af \in A there exists a unique γ∈Γ\gamma \in \Gamma such that ff is a γ\gamma-morphism, we shall denote it by p(f)p(f). Let f,f′f, f' be in AA, γ=p(f),γ′=p(f′)\gamma = p(f), \gamma' = p(f'). Then, for all b∈Eb \in E and all g∈Gg \in G,

(f′∘f)(gb)=f′(γ(g)f(b))=γ′(γ(g))f′(f(b)) (f' \circ f)(gb) = f'(\gamma(g)f(b)) = \gamma'(\gamma(g))f'(f(b))
which proves that f′∘f∈Af' \circ f \in A and that p(f′∘f)=p(f′)p(f)p(f' \circ f) = p(f')p(f). On the other hand, f(γ−1(g)f−1(b))=gbf(\gamma^{-1}(g)f^{-1}(b)) = gb, whence f−1(gb)=γ−1(g)f−1(b)f^{-1}(gb) = \gamma^{-1}(g)f^{-1}(b) and f−1∈Af^{-1} \in A. Thus AA is a subgroup of SE\mathscr{S}_E and pp is a homomorphism of AA into Γ\Gamma. The kernel of pp is the set AutG(E)\mathrm{Aut}_G(E) of automorphisms of the GG-set EE. We fix a∈Ea \in E. We have defined in § 5, no. 6 an isomorphism ψa\psi_a of G0G^0 onto AutG(E)\mathrm{Aut}_G(E) such that ψa(x)(ga)=gxa\psi_a(x)(ga) = gxa for all g,xg, x in GG. On the other hand, for γ∈Γ\gamma \in \Gamma, let sa(γ)s_a(\gamma) be the permutation of EE defined by sa(γ)(ga)=γ(g)as_a(\gamma)(ga) = \gamma(g)a for all g∈Gg \in G; it is immediately verified that sas_a is a homomorphism of Γ\Gamma into AA such that p∘sa=Idτp \circ s_a = \mathrm{Id}_{\tau}. Thus G0→ψaA→pΓG^0 \xrightarrow{\psi_a} A \xrightarrow{p} \Gamma is an extension of Γ\Gamma by G0G^0 and sas_a is a section of this extension. This extension and this section define an operation of Γ\Gamma on G0G^0, sa(Γ)s_a(\Gamma) acting on ψa(G0)\psi_a(G^0) by inner automorphisms; we write this operation exponentially. We show that this operation is the natural operation (§3§ 3, no. 1, Example 3): for x,gx, g in GG and γ∈Γ\gamma \in \Gamma,
(ψa(γx))(ga)=(sa(γ)∘ψa(x)∘sa(γ)−1)(ga)=(sa(γ)∘ψa(x))(γ−1(g)a)=sa(γ)(γ−1(g)xa)=gγ(x)a=ψa(γ(x))ga (\psi_a(\gamma x))(ga) = (s_a(\gamma) \circ \psi_a(x) \circ s_a(\gamma)^{-1})(ga) = (s_a(\gamma) \circ \psi_a(x))(\gamma^{-1}(g)a) = s_a(\gamma)(\gamma^{-1}(g)xa) = g\gamma(x)a = \psi_a(\gamma(x))ga
whence γx=γ(x)^\gamma x = \gamma(x). Proposition 4 then shows that AA is isomorphic to the semidirect product of Γ=Aut(G)\Gamma = \mathrm{Aut}(G) by G0G^0 under the natural operation of Aut(G)\mathrm{Aut}(G) on G0G^0. Note that the isomorphism which we have constructed depends in general on the choice of the element a∈Ea \in E. (2) *Let AA be a commutative ring. The upper triangular group T(n,A)T(n, A) is the semi-direct product of the diagonal subgroup D(n,A)D(n, A) by the upper strict triangular group T1(n,A)T_1(n, A).

2. Commutators

Definition 3

Let GG be a group and xx and yy two elements of GG. The element x−1y−1xyx^{-1}y^{-1}xy of GG is called the commutator of xx and yy.

(x, y) is used to denote the commutator of x and y. Then obviously

(y,x)=(x,y)−1.(y, x) = (x, y)^{-1}.
For x and y to commute it is necessary and sufficient that (x,y)=e(x, y) = e. More generally,
xy=yx(x,y).xy = yx(x, y).
On the other hand we write (3)
xy=y−1xy=x(x,y)=(y,x−1)x.x^y = y^{-1}xy = x(x, y) = (y, x^{-1})x.
As the mapping x↦xyx \mapsto x^y is the inner automorphism Int⁡(y−1)\operatorname{Int}(y^{-1}), (x,y)z=(xz,yz)(x, y)^z = (x^z, y^z) for all x,y,z∈Gx, y, z \in G.

For x,y,z∈Gx, y, z \in G, we prove the following relations: (4)

(x,yz)=(x,z).(x,y)z=(x,z).(z,(y,x)).(x,y)(x, yz) = (x, z).(x, y)^z = (x, z).(z, (y, x)).(x, y)
(4 bis)
(xy,z)=(x,z)y.(y,z)=(x,z).((x,z),y).(y,z)(xy, z) = (x, z)^y.(y, z) = (x, z).((x, z), y).(y, z)
(5)
(xy,(y,z)).(yz,(z,y)).(zx,(x,y))=e(x^y, (y, z)).(y^z, (z, y)).(z^x, (x, y)) = e
(6)
(x,yz).(z,xy).(y,zx)=e(x, yz).(z, xy).(y, zx) = e
(6 bis)
(xy,z).(yz,x).(zx,y)=e.(xy, z).(yz, x).(zx, y) = e.

Now

(x,yz)=x−1z−1y−1xyz=(x,z)z−1x−1y−1xyz=(x,z)(x,y)z=(x,z)(z,(x,y)−1)(x,y) \begin{align*} (x, yz) &= x^{-1}z^{-1}y^{-1}xyz = (x, z)z^{-1}x^{-1}y^{-1}xyz = (x, z)(x, y)^z \\ &= (x, z)(z, (x, y)^{-1})(x, y) \end{align*}
by (3), which proves (4). Formula (4 bis) follows similarly. On the other hand,
(xy,(y,z))=(xy)−1(z,y)(xy)(y,z)=y−1x−1yz−1y−1zyy−1xyy−1z−1yz=(yzy−1xy)−1(zxz−1yz). \begin{align*} (x^y, (y, z)) &= (x^y)^{-1}(z, y)(x^y)(y, z) \\ &= y^{-1}x^{-1}yz^{-1}y^{-1}zyy^{-1}xyy^{-1}z^{-1}yz \\ &= (yzy^{-1}xy)^{-1}(zxz^{-1}yz). \end{align*}
Then writing u=yzy−1xy,v=zxz−1xyu = yzy^{-1}xy, v = zxz^{-1}xy and w=xyx−1zxw = xyx^{-1}zx, we obtain
(xy,(y,z))=u−1v.(x^y, (y, z)) = u^{-1}v.
By cyclically permuting x,y,zx, y, z, we deduce (yz,(z,x))=v−1w(y^z, (z, x)) = v^{-1}w and
(zx,(x,y))=w−1u,(z^x, (x, y)) = w^{-1}u,
which immediately imply (5). Finally, (6) follows by multiplying together the two sides in the three formulae obtained by cyclically permuting x,y,zx, y, z in the formula (x,yz)=x−1z−1y−1xyz=(yzx)−1(xyz)(x, yz) = x^{-1}z^{-1}y^{-1}xyz = (yzx)^{-1}(xyz), and similarly for (6 bis).

If A and B are two subgroups of G, (A, B) denotes the subgroup generated by the commutators (a,b)(a, b) with a∈Aa \in A and b∈Bb \in B.† Then (A,B)={e}(A, B) = \{e\} if and only if A centralizes B. (A,B)⊂A(A, B) \subset A if and only if B normalizes A. If A and B are normal (resp. characteristic), so is (A,B)(A, B).

† We here reject the notational convention made in § 1, no 1 of extending a law of composition to subsets.

Proposition 5

Let A,B,CA, B, C be three subgroups of GG.

(i) The subgroup AA normalizes the subgroup (A,B)(A, B).

(ii) If the subgroup (B,C)(B, C) normalizes AA, the subgroup (A,(B,C))(A, (B, C)) is generated by the elements (a,(b,c))(a, (b, c)) with a∈A,b∈Ba \in A, b \in B and c∈Cc \in C.

(iii) If A,BA, B and CC are normal, then

(A,(B,C))⊂(C,(B,A)).(B,(C,A)). (A, (B, C)) \subset (C, (B, A)).(B, (C, A)).

By (4 bis), for a,a′∈Aa, a' \in A and b∈Bb \in B,

(a,b)a′=(aa′,b).(a′,b)−1, (a, b)^{a'} = (aa', b).(a', b)^{-1},

whence (i). Suppose now that (B,C)(B, C) normalizes AA. For a∈A,b∈B,c∈Ca \in A, b \in B, c \in C and x∈Gx \in G, (4) implies

(a,(b,c).x)=(a,x).(x,((b,c),a))(a,(b,c)) (a, (b, c).x) = (a, x).(x, ((b, c), a))(a, (b, c))

and ((b,c),a)∈A((b, c), a) \in A since (B,C)(B, C) normalizes AA, whence by induction on pp the fact that (a,∏i=1p(bi,ci))\left( a, \prod_{i=1}^p (b_i, c_i) \right), for bi∈B,ci∈Cb_i \in B, c_i \in C, belongs to the subgroup generated by the elements of the form (a,(b,c))(a, (b, c)). If finally A,BA, B and CC are normal, so are the subgroups (A,(B,C))(A, (B, C)), (C,(B,A))(C, (B, A)) and (B,(C,A))(B, (C, A)). It therefore suffices by (ii) to show that

(a,(b,c))∈(C,(B,A)).(B,(C,A)) (a, (b, c)) \in (C, (B, A)).(B, (C, A))

for all a∈A,b∈Ba \in A, b \in B and c∈Cc \in C. Now by (5), writing ab−1=ua^{b^{-1}} = u

(a,(b,c))=((ub),(b,c))=(cu,(u,b))−1.(bc,(c,u))−1, (a, (b, c)) = ((u^b), (b, c)) = (c^u, (u, b))^{-1}.(b^c, (c, u))^{-1},

whence (iii).

Definition 4

Let GG be a group. The subgroup generated by the commutators of elements of GG is called the derived group of GG.

The derived group of GG is thus the subgroup (G,G)(G, G). It is also denoted by D(G)D(G). By an abuse of language, it is sometimes called the commutator group of GG although it is in general distinct from the set of commutators of elements of GG (Exercise 16). D(G)={e}D(G) = \{e\} if and only if GG is commutative.

Proposition 6

Let f:G→G′f : G \to G' be a group homomorphism. Then f(D(G))⊂D(G′)f(D(G)) \subset D(G'). If ff is surjective, the homomorphism of D(G)D(G) into D(G′)D(G') the restriction of ff is surjective.

The image under ff of a commutator of elements of GG is a commutator of elements of G′G'. If ff is surjective, the image under ff of the set of commutators of GG is the set of commutators of G′G'. The proposition thus follows from § 4, no. 3, Corollary 3 to Proposition 2.

Corollary 1

The derived group of a group GG is a characteristic subgroup of GG. In particular it is a normal subgroup of GG.

Corollary 2

Let G be a group. The quotient group G/D(G) is commutative. Let π:G→G/D(G)\pi : G \to G/D(G) be the canonical homomorphism. Every homomorphism f of G into a commutative group G' can be expressed uniquely in the form f=fˉ∘πf = \bar{f} \circ \pi, where fˉ:G/D(G)→G′\bar{f} : G/D(G) \to G' is a homomorphism.

Now π(D(G))={e}\pi(D(G)) = \{e\}. As π\pi is surjective, it follows that D(G/D(G))={e}D(G/D(G)) = \{e\}, whence the first assertion. The second follows from § 4, no. 4, Proposition 5.

Corollary 3

Let H be a subgroup of G. The following conditions are equivalent:

(i) H⊃D(G)H \supset D(G); (ii) H is a normal subgroup and G/HG/H is commutative. (ii) ⇒\Rightarrow (i) by Corollary 2 and (i) ⇒\Rightarrow (ii) by § 4, no. 7, Theorem 4, since every subgroup of a commutative group is normal.

Corollary 4

Let G be a group and X a subset of G which generates G. The group D(G) is the normal subgroup of G generated by the commutators of elements of X.

Let H be the normal subgroup of G generated by the commutators of elements of X and ϕ:G→G/H\phi : G \to G/H the canonical homomorphism. The set ϕ(X)\phi(X) generates G/HG/H. The elements of ϕ(X)\phi(X) are pairwise permutable and hence H is commutative (§ 4, no. 3, Corollary 2 to Proposition 2). Hence (Corollary 3) H contains D(G). On the other hand, obviously H⊂D(G)H \subset D(G).

Remark

(1) Corollary 2 can also be expressed by saying that G/D(G)G/D(G), together with π\pi, is a solution of the universal mapping problem for G, relative to commutative groups and homomorphisms from G to commutative groups.

(2) Under the hypotheses of Corollary 4, the subgroup generated by the commutators of elements of X is contained in D(G) but is not in general equal to D(G) (cf. Exercise 15e).

Example

(1) If G is a non-commutative simple group, then D(G)=GD(G) = G. Therefore every homomorphism of G into a commutative group is trivial.

(2) The derived group of the symmetric group SnS_n is the alternating group AnA_n. For AnA_n is generated by the products of two transpositions; if τ=τx,y\tau = \tau_{x,y} and τ′=τx′,y′\tau' = \tau_{x',y'} are two transpositions, let σ\sigma be a permutation such that σ(x′)=x\sigma(x') = x and σ(y′)=y\sigma(y') = y. Then τ′=σ−1τσ\tau' = \sigma^{-1} \tau \sigma and ττ′=τ−1τ′=τ−1σ−1τσ\tau \tau' = \tau^{-1} \tau' = \tau^{-1} \sigma^{-1} \tau \sigma is a commutator. Hence An⊂D(Sn)A_n \subset D(S_n). As Sn/AnS_n/A_n is commutative, An⊃D(Sn)A_n \supset D(S_n) (Corollary 3).

3. Lower central series, nilpotent groups

Let G be a group, H a subgroup of G and K a normal subgroup of G. The image of H in G/K is contained in the centre of G/K if and only if (G,H)⊂K(G, H) \subset K.

Definition 5

Let G be a group. The lower central series of G is the sequence (Cn(G))n≥1(C^n(G))_{n \geq 1} of subgroups of G defined inductively by:

C1(G)=G,Cn+1(G)=(G,Cn(G)). C^1(G) = G, \quad C^{n+1}(G) = (G, C^n(G)).

Let f:G→G′f : G \to G' be a group homomorphism. It is seen, by induction on nn, that f(Cn(G))⊂Cn(G′)f(C^n(G)) \subset C^n(G') and that, if ff is surjective, f(Cn(G))=Cn(G′)f(C^n(G)) = C^n(G'). In particular, for all n≥1n \geq 1, C2(G)C^2(G) is a characteristic (and hence normal) subgroup of GG. For all n≥1n \geq 1, Cn(G)/Cn+1(G)C^n(G)/C^{n+1}(G) is contained in the centre of G/Cn+1(G)G/C^{n+1}(G).

Let (G1,G2,…)(G_1, G_2, \ldots) be a decreasing sequence of normal subgroups of GG such that (1) G1=GG_1 = G; (2) for all ii, Gi/Gi+1G_i/G_{i+1} is contained in the centre of G/Gi+1G/G_{i+1}. Then Ci(G)⊂GiC^i(G) \subset G_i, as is seen by induction on ii.

Now

(Cm(G),Cn(G))⊂Cm+n(G). (C^m(G), C^n(G)) \subset C^{m+n}(G).
For, if this relation is denoted by (Fm,n)(F_{m,n}), it follows from (Fm,n)(F_{m,n}), by no. 2, Proposition 5, that
(Cm(G),Cn+1(G))⊂(G,(Cm(G),Cn(G))).(Cn(G),(G,Cm(G)))⊂Cm+n+1(G).(Cm+1(G),Cn(G)). \begin{align*} (C^m(G), C^{n+1}(G)) &\subset (G, (C^m(G), C^n(G))).(C^n(G), (G, C^m(G))) \\ &\subset C^{m+n+1}(G).(C^{m+1}(G), C^n(G)). \end{align*}
Hence ((Fm,n) and (Fm+1,n))⇒(Fm,n+1)((F_{m,n}) \text{ and } (F_{m+1,n})) \Rightarrow (F_{m,n+1}). As (Fm,1)(F_{m,1}) and (F1,n)(F_{1,n}) are obvious (Fm,n)(F_{m,n}) follows by induction.

Definition 6

A group GG is called nilpotent if there exists an integer nn such that Cn+1(G)={e}C^{n+1}(G) = \{e\}. The least integer nn such that Cn+1(G)={e}C^{n+1}(G) = \{e\} is called the nilpotency class of a nilpotent group GG.

If n∈Nn \in \mathbf{N}, a group of nilpotency class nn is called a nilpotent group of class nn. It is sometimes said that the nilpotency class of a group GG is finite if GG is nilpotent.

Example

(1) A group is nilpotent of class 0 (resp. ≤1\leq 1) if and only if it consists of the identity element (resp. is commutative).

(2) *For every commutative ring AA and every integer n≥1n \geq 1, the upper strict triangular group T1(n,A)T_1(n, A) is nilpotent of class ≤n−1\leq n - 1 (and exactly of class n−1n - 1 if A≠{0}A \neq \{0\}).

(3) Let GG be a nilpotent group of class nn. Every subgroup (resp. every quotient group) of GG is nilpotent of class ≤n\leq n. For, if HH is a subgroup of GG, then Cn(H)⊂Cn(G)C^n(H) \subset C^n(G). If G′G' is a quotient group of GG and π:G→G′\pi : G \to G' is the canonical homomorphism, then Cn(G′)=π(Cn(G))C^n(G') = \pi(C^n(G)).

(4) A finite product of nilpotent groups is nilpotent.

Proposition 7

Let GG be a group and nn an integer. The following conditions are equivalent:

(a) GG is nilpotent of class ≤n\leq n. (b) There exists a series of subgroups of GG:

G=G1⊃G2⊃…⊃Gn+1={e} G = G^1 \supset G^2 \supset \ldots \supset G^{n+1} = \{e\}
such that (G,Gk′)⊂Gk′+1(G, G^{k'}) \subset G^{k'+1} for all k∈{1,n}k \in \{1, n\}.

(c) There exists a subgroup AA of GG contained in the centre of GG such that G/AG/A is nilpotent of class ≤n−1\leq n - 1.

(a) ⇒\Rightarrow (b): it suffices to take Gk=Ck(G)G^k = C^k(G). (b) ⇒\Rightarrow (a): by induction on kk, Ck(G)⊂GkC^k(G) \subset G^k. (a) ⇒\Rightarrow (c): it suffices to take A=Cn(G)A = C^n(G). (c) ⇒\Rightarrow (a): let π:G→G/A\pi : G \to G/A be the canonical homomorphism; then π(Cn(G))=Cn(G/A)={e}\pi(C^n(G)) = C^n(G/A) = \{e\} and hence Cn(G)⊂AC^n(G) \subset A, whence Cn+1(G)={e}C^{n+1}(G) = \{e\}.

More briefly: a group is nilpotent of class ≤n\leq n if it can be obtained from the group {e}\{e\} by nn successive central extensions.

Corollary

A central extension of a nilpotent group (by a necessarily commutative group) is nilpotent.

Proposition 8

Let GG be a nilpotent group of class ≤n\leq n and let HH be a subgroup of GG. There exists a sequence of subgroups

G=H1⊃H2⊃…⊃Hn+1=H, G = H^1 \supset H^2 \supset \ldots \supset H^{n+1} = H,

such that Hk+1H^{k+1} is normal in HkH^k and Hk/Hk+1H^k/H^{k+1} is commutative for all k≤nk \leq n.

Choose a sequence (Gk)(G^k) of subgroups of GG satisfying the conditions of Proposition 7 (b) for all kk; GkG^k is normal in GG. Write:

Hk=H.Gk. H^k = H.G^k.

It is necessary to verify that Hk+1H^{k+1} is normalized by Hk=H.GkH^k = H.G^k; as it is normalized by HH, it suffices to verify that it is by GkG^k. Now, if s∈Gks \in G^k and h∈Hh \in H,

shs−1=shs−1h−1.h∈(G,Gk).H shs^{-1} = shs^{-1}h^{-1}.h \in (G, G^k).H

and (G,Gk).H(G, G^k).H is contained in Gk+1.H=Hk+1G^{k+1}.H = H^{k+1}; hence s.Hk+1.s−1=Hk+1s.H^{k+1}.s^{-1} = H^{k+1}, which shows that Hk+1H^{k+1} is normal in HkH^k.

Finally, the canonical homomorphism Gk/Gk+1→Hk/Hk+1G^k/G^{k+1} \to H^k/H^{k+1} is obviously surjective; as the first group is commutative, so is the second.

Corollary 1

Let GG be a nilpotent group and HH a subgroup of GG. If HH is distinct from GG, the normalizer NG(H)N_G(H) of HH in GG is distinct from HH.

Let kk be the largest index such that Hk≠HH^k \neq H. The group HkH^k normalizes HH and is distinct from HH.

Corollary 2

Let GG be a nilpotent group and HH a subgroup of GG. If HH is distinct from GG, there exists a normal subgroup NN of GG, containing HH, distinct from GG and such that G/NG/N is commutative.

Let kk be the least index such that Hk≠GH^k \neq G. The group HkH^k satisfies the required conditions.

Corollary 3

Let G be a nilpotent group and H a subgroup of G. If G = H.(G, G), then G = H.

Every subgroup N of G which contains H and such that G/N is commutative contains H.(G, G). Corollary 3 thus follows from Corollary 2.

Corollary 3 can also be formulated thus: let X be a subset of G. For X to generate G, it is necessary and sufficient that the image of X in G/D(G) generate G/D(G).

Corollary 4

Let f: G' → G be a group homomorphism. Suppose that* (a) G is nilpotent. (b) The homomorphism f₁: G'/(G', G') → G/(G, G), derived from f by passing to the quotients, is surjective. *Then f is surjective.

This follows from Corollary 3 applied to the subgroup H = f(G').

Proposition 9

Let G be a nilpotent group of class ≤ n and let N be a normal subgroup of G. There exists a series of subgroups

N=N1⊃N2⊃…⊃Nn+1={e} N = N^1 \supset N^2 \supset \ldots \supset N^{n+1} = \{e\}

such that (G, N^k) ⊂ N^{k+1} for k = 1, ..., n.

If (G^k) satisfies condition (b) of Proposition 7, then take

Nk=Gk∩N. N^k = G^k \cap N.

Corollary 1

Let G be a nilpotent group, Z the centre of G and N a normal subgroup of G. If N ≠ {e}, then N ∩ Z ≠ {e}.

Let k be the largest index such that N^k ≠ {e}. The group N^k is contained in N. On the other hand, (G, N^k) ⊂ N^{k+1} = {e}; hence N^k is contained in the centre Z of G.

Corollary 2

Let f be a homomorphism from a nilpotent group G to a group G'. If the restriction of f to the centre of G is injective, f is injective.

This is Corollary 1 applied to Ker(f).

4. Derived series, solvable groups

Definition 7

Let G be a group. The derived series of G is the series (D^n(G))_{n \in \mathbf{N}} defined inductively by:

D0(G)=G;Dn+1(G)=D(Dn(G))for n∈N. D^0(G) = G; \quad D^{n+1}(G) = D(D^n(G)) \quad \text{for } n \in \mathbf{N}.

Then D0(G)=C1(G)=G, D1(G)=C2(G)=D(G)=(G,G)D^0(G) = C^1(G) = G,\ D^1(G) = C^2(G) = D(G) = (G, G). For all n \in \mathbf{N}, D^n(G) \subset C^{2^n}(G), as is seen by induction on n using formula (7) of no. 3.

Let f:G→G′f : G \to G' be a group homomorphism. It is seen, by induction on n, that f(Dn(G))⊂Dn(G′)f(D^n(G)) \subset D^n(G') and that, if ff is surjective, f(Dn(G))=Dn(G′)f(D^n(G)) = D^n(G'). In particular, for all n∈Nn \in \mathbf{N}, Dn(G)D^n(G) is a characteristic (and therefore normal) subgroup of G. For all n∈Nn \in \mathbf{N}, the group Dn(G)/Dn+1(G)D^n(G)/D^{n+1}(G) is a commutative normal (but not in general central) subgroup of G/Dn+1(G)G/D^{n+1}(G).

Let (G0,G1,…)(G_0, G_1, \ldots) be a decreasing sequence of subgroups of G such that: (1) G0=GG_0 = G; (2) for all ii, Gi+1G_{i+1} is normal in GiG_i and Gi/Gi+1G_i/G_{i+1} is commutative. Then Di(G)⊂GiD^i(G) \subset G_i for all ii, as is seen by induction on ii.

Definition 8

A group G is called solvable if there exists an integer n such that Dn(G)={e}D^n(G) = \{e\}. If G is a solvable group, the least integer n such that Dn(G)={e}D^n(G) = \{e\} is called the solvability class of G.

A solvable group of solvability class nn is called a solvable group of class nn. A group is sometimes said to be of finite solvability class if it is solvable.

Example

(1) A group is solvable of class 0 (resp. ≤1\leq 1) if and only if it is reduced to {e}\{e\} (resp. is commutative).

(2) Every nilpotent group of class ≤2n−1\leq 2^n - 1 is solvable of class ≤n\leq n; this follows from the relation Dn(G)⊂C2n(G)D^n(G) \subset C^{2^n}(G) proved above.

(3) Let G be a solvable group of class ≤n\leq n. Every subgroup (resp. quotient group) of G is solvable of class ≤n\leq n (proof analogous to that of no. 3, Example 3).

(4) If G is a solvable group of class pp and F is a solvable group of class qq, every extension E of G by F is a solvable group of class ≤p+q\leq p + q. For, let π:E→G\pi : E \to G be the projection; then π(Dp(E))⊂Dp(G)={e}\pi(D^p(E)) \subset D^p(G) = \{e\} and therefore Dp(E)⊂FD^p(E) \subset F; it follows that Dp+q(E)=Dq(Dp(E))⊂Dq(F)={e}D^{p+q}(E) = D^q(D^p(E)) \subset D^q(F) = \{e\}.

(5) The symmetric group SnS_n is solvable if and only if n<5n < 5 (cf. § 5, Exercises 10 and 16).

(6) *If A is a commutative ring, the upper triangular group T(n, A) is solvable but not in general nilpotent.

Proposition 10

Let G be a group and n an integer. The following conditions are equivalent:

(i) G is solvable of class ≤n\leq n. (ii) There exists a series of normal subgroups of G

G=G0⊃G1⊃…⊃Gn={e} G = G^0 \supset G^1 \supset \ldots \supset G^n = \{e\}
such that the groups Gk/Gk+1G^k/G^{k+1} are commutative. (iii) There exists a series of subgroups of G
G=G0⊃G1⊃…⊃Gn=e G = G^0 \supset G^1 \supset \ldots \supset G^n = e
such that, for all k, Gk+1G^{k+1} is a normal subgroup of GkG^k and Gk/Gk+1G^k/G^{k+1} is commutative.

(iv) There exists a normal commutative subgroup AA of GG such that G/AG/A is solvable of class ≤n−1\leq n - 1.

For (i) ⇒\Rightarrow (ii) it suffices to take GkG^k equal to Dk(G)D^k(G). (ii) ⇒\Rightarrow (iii) trivially. (iii) ⇒\Rightarrow (i) for Dk(G)D^k(G) is necessarily contained in GkG^k. The equivalence of (ii) and (iv) is immediate by induction on nn.

More briefly: a group is solvable of class ≤n\leq n if it can be obtained by successive extensions of nn commutative groups.

Corollary

Let GG be a finite group and

G=G0⊃G1⊃…⊃Gn={e} G = G^0 \supset G^1 \supset \ldots \supset G^n = \{e\}
a Jordan-Hölder series of GG For GG to be solvable, it is necessary and sufficient that the quotients Gk/Gk+1G^k/G^{k+1} be cyclic of prime order.

If the quotients of a composition series of GG are cyclic and hence commutative, GG is solvable by Proposition 10. Conversely, if GG is solvable, the group Gk/Gk+1G^k/G^{k+1} is, for all kk, solvable and simple (§4§ 4, no. 7, Proposition 9). Now, every solvable simple group HH is cyclic of prime order. For D(H)D(H) is a normal subgroup of HH; D(H)=HD(H) = H is impossible for in that case Dk(H)=HD^k(H) = H for all kk; then D(H)={e}D(H) = \{e\} and HH is commutative. The corollary then follows from §4§ 4, no. 10, Corollary to Proposition 20.

5. M646-groups

In this number and the following, the letter pp denotes a prime number (§4§ 4, no 10, Proposition 16).

Definition 9

A finite group whose order is a power of pp is called a pp-group.

Let GG be a pp-group of order prp^r. Every divisor of prp^r is a power of pp (§4§ 4, no. 10, Corollary to Theorem 7). Therefore every subgroup and every quotient group of GG is a pp-group (§4§ 4, no. 4, Corollary to Proposition 4); the cardinal of every homogeneous space of GG is a power of pp (§5§ 5, no. 5, Theorem 1).

An extension of a pp-group by a pp-group is a pp-group.

Examples. (1) A commutative pp-group is isomorphic to a product of cyclic groups Z/pnZ\mathbf{Z}/p^n\mathbf{Z} (cf. Exercise 19 and also VII, §4§ 4, no. 7, Proposition 7). (2) Let kk be a finite field of characteristic pp. The strict triangular group T1(n,k)T_1(n, k) is a pp-group. (3) The quaternionic group {±1,±i,±j,±k}\{ \pm 1, \pm i, \pm j, \pm k \} is a 2-group (cf. Exercise 4).

Proposition 11

Let EE be a finite set and GG a pp-group operating on EE. Let EGE^G denote the set of x∈Ex \in E such that gx=xgx = x for all g∈Gg \in G (the fixed points). Then

Card⁡(EG)≡Card⁡(E)(modp). \operatorname{Card}(E^G) \equiv \operatorname{Card}(E) \pmod{p}.

E — E^G is a disjoint union of orbits not reduced to a point. The cardinal of such an orbit is a power of pp distinct from p0=1p^0 = 1 and hence a multiple of pp.

Corollary

Let G be a p-group. If G is not reduced to e, its centre is not reduced to e.

Let G operate on itself by inner automorphisms. The set of fixed points is the centre Z of G. By Proposition 11,

Card⁡(Z)≡Card⁡(G)≡0(modp), \operatorname{Card}(Z) \equiv \operatorname{Card}(G) \equiv 0 \pmod{p},

whence Card⁡(Z)≠1\operatorname{Card}(Z) \neq 1 and Z≠{e}Z \neq \{e\}.

Theorem 1

Let G be a p-group and prp^r its order. There exists a sequence of subgroups of G

G=G1⊃G2⊃…⊃Gr+1={e} G = G^1 \supset G^2 \supset \ldots \supset G^{r+1} = \{e\}

such that (G,Gk)⊂Gk+1,1≤k≤r,(G, G^k) \subset G^{k+1}, 1 \leq k \leq r, and Gk/Gk+1,1≤k≤r,G^k/G^{k+1}, 1 \leq k \leq r, is cyclic of order pp.

The theorem is true for G={e}G = \{e\}. We prove it by induction on Card⁡(G)\operatorname{Card}(G). Let Z be the centre of G, x≠ex \neq e an element of Z (Corollary to Proposition 11) and ps,s≠0,p^s, s \neq 0, the order of x. Then xps−1x^{p^s - 1} is an element of order pp and therefore Z contains a subgroup G′G' which is cyclic of order pp. By the induction hypothesis, the group G′=G/G′G' = G/G' has a series of subgroups (G′k)1≤k≤r({G'}^k)_{1 \leq k \leq r} with the required properties. Let π:G→G′\pi : G \to G' be the canonical homomorphism. The sequence of subgroups of G defined by Gk=π−1(G′k),1≤k≤r,Gr+1={e}G^k = \pi^{-1}({G'}^k), 1 \leq k \leq r, G^{r+1} = \{e\} is a solution for Gk/Gk+1G^k/G^{k+1} is isomorphic to G′k/G′k+1{G'}^k/{G'}^{k+1} for 1≤k≤r1 \leq k \leq r (§ 4, no. 7, Theorem 4).

Corollary

Every p-group is nilpotent.

This follows from no. 3, Proposition 7.

Proposition 12

Let G be a p-group and H a subgroup of G distinct from G. Then:

(a) The normalizer NG(H)N_G(H) of H in G is distinct from G. (b) There exists a normal subgroup N of G of index p in G, which contains H.

Assertion (a) follows from no. 3, Corollary 1 to Proposition 8. We prove (b). By no. 3, Corollary 2 to Proposition 8, there exists a normal subgroup N′N' of G containing H, distinct from G and such that G/N′G/N' is commutative. Let N be a maximal subgroup distinct from G containing N′N'. Then N is normal (no. 2, Corollary 3 to Proposition 6) and G/NG/N is a simple commutative pp-group and hence cyclic of order pp (§ 4, no. 10, Corollary to Proposition 20).

Corollary

Let G be a p-group. Every subgroup of G of index p is normal.

6. Sylow subgroups

Definition 10

Let G be a finite group. A Sylow p-subgroup of G is any subgroup P of G satisfying the two following conditions:

(a) P is a p-group. (b) (G : P) is not a multiple of p.

If the order of G is written in the form prmp^r m, where m is not a multiple of pp, conditions (a) and (b) are equivalent to Card(P)=pr\mathrm{Card}(P) = p^r.

Example

(1) In the group Sp\mathfrak{S}_p let ζ\zeta be a cycle of order pp. The subgroup generated by ζ\zeta is a Sylow pp-subgroup of Sp\mathfrak{S}_p for pp does not divide (p−1)!(p - 1)!. (2) *Let k be a finite field of characteristic pp and let n be a positive integer. The strict triangular group T1(n,k)T_1(n, k) is a Sylow pp-subgroup of the group GL(n,k)\mathbf{GL}(n, k).

Theorem 2

Every finite group contains a Sylow pp-subgroup.

The proof depends on the following lemma.

Lemma

Let n=prmn = p^r m, where m is an integer which is not a multiple of pp. Then

(npr)≢0(modp). \binom{n}{p^r} \not\equiv 0 \pmod{p}.

Let S be a group of order prp^r (for example Z/prZ\mathbf{Z}/p^r \mathbf{Z}) and T a set with m elements. Write X=S×TX = S \times T and let E be the set of subsets of X with prp^r elements. Then Card(X)=n\mathrm{Card}(X) = n, whence Card(E)=(npr)\mathrm{Card}(E) = \binom{n}{p^r} (Set Theory, III, § 5, no. 8, Corollary 1 to Proposition 11). Let S operate on X by s.(x,y)=(sx,y)(s,x∈S,y∈T)s.(x, y) = (sx, y) (s, x \in S, y \in T) and consider the canonical extension of this operation to E. In the notation of no. 5, Proposition 11, the set ESE^S is the set of orbits of X, that is the set of subsets Y⊂XY \subset X of the form S×{t},t∈TS \times \{ t \}, t \in T, whence Card(ES)=m\mathrm{Card}(E^S) = m. By no. 5, Proposition 11,

(npr)=Card(E)≡Card(ES)=m≢0(modp), \binom{n}{p^r} = \mathrm{Card}(E) \equiv \mathrm{Card}(E^S) = m \not\equiv 0 \pmod{p},
which proves the lemma.

We now prove the theorem. Let G be a finite group and n its order; we write n=prmn = p^r m, where m is not a multiple of pp. Let E be the set of subsets of G with prp^r elements. Then

Card(E)=(npr); \mathrm{Card}(E) = \binom{n}{p^r};
whence, by virtue of the lemma, Card(E)≢0(modp)\mathrm{Card}(E) \not\equiv 0 \pmod{p}. Consider the extension to E of the operation of G on itself by left translation. There exists X∈EX \in E whose orbit has non-zero cardinal mod. pp. If HXH_X denotes the stabilizer of X, then

(G:Hx)≢0(modp)(G:H_x) \not\equiv 0 \pmod{p}, which means that prp^r divides Card(Hx)\mathrm{Card}(H_x). But HxH_x consists of the s∈Gs \in G such that sX=XsX = X; if x∈Xx \in X, then Hx⊂X.x−1H_x \subset X.x^{-1}, whence Card(Hx)⩽Card(X)=pr\mathrm{Card}(H_x) \leqslant \mathrm{Card}(X) = p^r. Hence Card(Hx)=pr\mathrm{Card}(H_x) = p^r.

Corollary

If the order of G is divisible by p, the group G contains an element of order p.

By virtue of Theorem 2, this is reduced to the case where G is a pp-group ≠{e}\neq \{e\}; if x∈Gx \in G is different from e, the cyclic group generated by x is then of order pnp^n with n⩾1n \geqslant 1 and it therefore contains a subgroup of order pp.

Remark

For every prime number q dividing Card(G), let P_q be a Sylow q-subgroup of G. Then the subgroup H of G generated by the P_q is of order a multiple of Card(P_q) for each q and of order a divisor of Card(G), hence it is equal to G.

Theorem 3

Let G be a finite group.* (a) The Sylow p-subgroups of G are conjugate to one another. Their number is congruent to 1 mod. p. (b) *Every subgroup of G which is a p-group is contained in a Sylow p-subgroup.

Let P be a Sylow pp-subgroup of G (Theorem 2) and let H be a pp-subgroup of G. Let E = G/P and consider the operation of H on G/P. As Card(E) ≢0\not\equiv 0 mod. pp, Proposition 11 of no. 5, shows that there exists x∈G/Px \in G/P such that hx=xhx = x for all h∈Hh \in H. If g is a representative of x in G, this means that H⊂gPg−1H \subset gPg^{-1}, whence assertion (b).

If H is a Sylow pp-subgroup, then Card(H) = Card(P) = Card(gPg−1gPg^{-1}), whence H=gPg−1H = gPg^{-1}, which proves the first assertion of (a).

We now prove the second assertion of (a). Let S\mathscr{S} be the set of Sylow pp-subgroups of G and let P operate on S\mathscr{S} by inner automorphisms. The element P∈SP \in \mathscr{S} is a fixed point under this operation, we show that it is the only one. Let Q∈SQ \in \mathscr{S} be a fixed point; Q is a Sylow subgroup of G normalized by P and hence P is contained in the normalizer N of Q. The groups P and Q are Sylow pp-subgroups of N; hence there exists n∈Nn \in N such that P=nQn−1=QP = nQn^{-1} = Q. By no. 5, Proposition 11, Card(SP\mathscr{S}^P) ≡\equiv Card(SPP\mathscr{S}^{P^P}) = 1 (mod. pp).

Corollary 1

Let P be a Sylow p-subgroup of G, let N be its normalizer in G and let M be a subgroup of G containing N. The normalizer of M in G is equal to M.

Let s∈Gs \in G be such that sMs−1=MsMs^{-1} = M. The subgroup sPs−1sPs^{-1} of M is a Sylow pp-subgroup of M. There thus exists t∈Mt \in M such that sPs−1=tPt−1sPs^{-1} = tPt^{-1}; then t−1s∈Nt^{-1}s \in N, whence s∈tN⊂Ms \in tN \subset M.

Corollary 2

Let f:G1→G2f : G_1 \to G_2 be a homomorphism of finite groups. For every Sylow p-subgroup P1P_1 of G1G_1 there exists a Sylow p-subgroup P2P_2 of G2G_2 such that f(P1)⊂P2f(P_1) \subset P_2.

This follows from Theorem 3 (b) applied to the subgroup f(P1)f(P_1) of G2G_2.

Corollary 3

(a) Let HH be a subgroup of GG. For every Sylow pp-subgroup PP of HH there exists a Sylow pp-subgroup QQ of GG such that P=Q∩HP = Q \cap H.

(b) Conversely, if QQ is a Sylow pp-subgroup of GG and HH is normal in GG, the group Q∩HQ \cap H is a Sylow pp-subgroup of HH.

(a) The pp-group PP is contained in a Sylow pp-subgroup QQ of GG and Q∩HQ \cap H is a pp-subgroup of HH containing PP and is hence equal to PP.

(b) Let P′P' be a Sylow pp-subgroup of HH. There exists an element g∈Gg \in G such that gP′g−1⊂QgP'g^{-1} \subset Q. As HH is normal, P=gP′g−1P = gP'g^{-1} is contained in HH and hence in Q∩HQ \cap H. As Q∩HQ \cap H is a pp-subgroup of HH and PP is a Sylow pp-subgroup of HH, P=Q∩HP = Q \cap H.

Corollary 4

Let NN be a normal subgroup of GG. The image in G/NG/N of a Sylow pp-subgroup of GG is a Sylow pp-subgroup of G/NG/N and every Sylow pp-subgroup of G/NG/N is obtained in this way.

Let G′=G/NG' = G/N and P′P' be the image in G′G' of a Sylow pp-subgroup PP of GG. The group GG operates transitively on G′/P′G'/P' and hence G′/P′G'/P' is equipotent to G/SG/S, where SS is a subgroup of GG containing PP. Therefore (G′:P′)(G':P') divides (G:P)(G:P), is thus not a multiple of pp and the pp-group P′P' is a Sylow pp-subgroup of G′G'. Let Q′Q' be another Sylow pp-subgroup of G′G'; then Q′=g′P′g′−1Q' = g'P'{g'}^{-1} for some g′∈G′g' \in G'; if g∈Gg \in G is a representative of g′g', the group Q′Q' is the image of Q=gPg−1Q = gPg^{-1}.

7. Finite nilpotent groups

Theorem 4

Let GG be a finite group. The following conditions are equivalent:

(a) GG is nilpotent. (b) GG is a product of pp-groups. (c) For every prime number pp there exists a normal Sylow pp-subgroup of GG. (b) ⇒\Rightarrow (a) (no. 5, Corollary to Theorem 1).

Suppose (a) holds and let PP be a Sylow pp-subgroup of GG. If NN is the normalizer of PP in GG, Corollary 1 to Theorem 3 shows that NN is its own normalizer. By § 6, no. 3, Corollary to Proposition 8, this shows that N=GN = G. Hence (a) ⇒\Rightarrow (c).

Suppose (c) holds and let II be the set of prime numbers dividing Card(G)\mathrm{Card}(G). For all p∈Ip \in I, let PpP_p be a normal Sylow pp-subgroup of GG. For all p≠qp \neq q, Pp∩PqP_p \cap P_q is reduced to ee for it is both a pp-group and a qq-group, hence PpP_p and PqP_q centralize one another (§4§ 4, no. 9, Proposition 15). Let ϕ\phi be the canonical homomorphism (§4§ 4, no. 9, Proposition 12) of ∏p∈IPp\prod_{p \in I} P_p into GG. The homomorphism ϕ\phi is surjective by the Remark of no. 6. As Card(∏p∈IPp)=Card(G)\mathrm{Card}\left(\prod_{p \in I} P_p\right) = \mathrm{Card}(G), it follows that ϕ\phi is bijective.

Remark

(1) Let G be a finite group and pp a prime number. By no. 6, Theorem 3 (a) and no. 6, Theorem 2, the following conditions are equivalent: (i) there exists a normal Sylow pp-subgroup of G; (ii) every Sylow pp-subgroup of G is normal; (iii) there exists only one Sylow pp-subgroup of G.

(2) Let G be a nilpotent finite group. Let I be the set of prime divisors of Card(G). By Theorem 4 and Remark 1, G=∏p∈IGpG = \prod_{p \in I} G_p, where GpG_p is the unique Sylow pp-group of G.

(3) Applied to commutative groups, Theorem 4 gives the decomposition, of commutative finite groups as a product of primary components, which will be studied from another point of view in Chapter VII.

Example

The group S3\mathfrak{S}_3 is of order 6. It contains a normal Sylow 3-subgroup of order 3: the group A3A_3. It contains three Sylow 2-subgroups of order 2: the groups {e,τ}\{e, \tau\}, where τ\tau is a transposition. The group S3\mathfrak{S}_3 is thus not nilpotent.

Exercises

See the exercises for § 6.

Transcribed from the printed edition, 1998, Springer. Not translated.