Skip to content

Normal Subgroups and Quotient Groups: The Condition for Multiplying Cosets

Prerequisite:Subgroups and Cosets: Why Lagrange's Theorem Produces Divisibility

Raw
  • If we try to equip the set G/HG/H of left cosets of a subgroup HGH \le G with the law “multiply the representatives”, the answer may change when the representatives are changed. The pair G=S3G = S_3, H={e,(12)}H = \{e, (1\,2)\} is an actual example.
  • A necessary and sufficient condition for this law to be defined is that gH=HggH = Hg for every gGg \in G, that is, that HH be a normal subgroup (Theorem 3.2). Normality is not some odd requirement that left and right cosets coincide; it is a restatement of the very wish to divide.
  • For a normal subgroup NN, the set G/NG/N becomes a group (the quotient group), and π:GG/N, ggN\pi : G \to G/N,\ g \mapsto gN is a surjective homomorphism with kernel NN. Conversely the kernel of a homomorphism is always a normal subgroup, so “normal subgroup” and “kernel of a homomorphism” are two faces of one concept.
  • The subgroups of Z\mathbb{Z} are exactly the nZn\mathbb{Z}, and the quotient group Z/nZ\mathbb{Z}/n\mathbb{Z} is cyclic of order nn. The order of aˉ\bar{a} is n/gcd(a,n)n/\gcd(a,n), and aˉ\bar{a} is a generator precisely when gcd(a,n)=1\gcd(a,n) = 1.
  • Passing to a quotient is an operation that discards information. The group G/NG/N need not be a subgroup of GG, and all distinctions inside NN are forgotten completely.

1. Motivation: giving the set of cosets a law of composition

Section titled “1. Motivation: giving the set of cosets a law of composition”

The hands of a clock come full circle at 12. Five hours after 9 o’clock is not 14 o’clock but 2 o’clock. What we are doing here is identifying integers under the relation “same remainder upon division by 12” and then adding in the identified world. That this computation is consistent is something we normally never question. Pause for a moment, however, and it is far from obvious. Behind it lies the fact that we may replace 99 by 2121 and 55 by 1717 without changing the answer, that is,

9+5=142,21+17=382(mod12)9 + 5 = 14 \equiv 2, \qquad 21 + 17 = 38 \equiv 2 \pmod{12}

so that the answer is determined independently of the choice of representatives. Only because of this does clock arithmetic mean anything. Gauss opened his Disquisitiones Arithmeticae (1801) by introducing the congruence symbol ab(modm)a \equiv b \pmod{m} and setting out that congruences may be “computed with directly” as far as addition, subtraction and multiplication are concerned. This is the starting point of the idea of a quotient in modern algebra.

Group theory treats this situation in far greater generality. Given a group GG and a subgroup HH, we saw in Subgroups and cosets (Lagrange’s theorem) that GG is partitioned into the left cosets aHaH (Lemma 5.2[Subgroups and Cosets]). We write G/HG/H for the set of all cosets. This is the situation corresponding to Z\mathbb{Z} and 12Z12\mathbb{Z} in the case of the integers. It is then tempting to ask the naive question:

Can G/HG/H be given a law of composition by multiplying representatives? That is, may we set (aH)(bH):=(ab)H(aH) \cdot (bH) := (ab)H?

If this were possible, G/HG/H would itself be a group, and we would obtain a new group that views GG “coarsely, up to HH”. The basic strategy of group theory — break a large group into smaller ones and study those — begins here.

The definition, however, fails in general. It can happen that aH=aHaH = a'H and bH=bHbH = b'H while (ab)H(ab)H(ab)H \ne (a'b')H (Example 3.1). Isolating the condition under which it does not fail yields exactly the notion that Galois called an “invariant subgroup” and that we call today a normal subgroup. The aim of this article is to derive that condition, to construct the quotient group, and to verify the most basic example Z/nZ\mathbb{Z}/n\mathbb{Z} down to the last detail.

The idea of dividing by an equivalence relation to build a new object is of course not confined to group theory. The general theory at the level of sets is collected in Relations and equivalence relations: what does “the same” mean?. It clarifies matters to read this article as asking: when does a law of composition descend to the quotient?

2. Preliminaries: cosets and homomorphisms

Section titled “2. Preliminaries: cosets and homomorphisms”

Let us fix notation. Throughout, GG is a group with identity element ee, and HGH \le G means that HH is a subgroup of GG.

Definition 2.1Left coset, right coset, index

For HGH \le G and aGa \in G, the sets

aH:={ah:hH},Ha:={ha:hH}aH := \{ah : h \in H\}, \qquad Ha := \{ha : h \in H\}

are called respectively the left coset and the right coset of HH containing aa. The set of all left cosets is written G/HG/H, the set of all right cosets H\GH \backslash G. The number of elements of G/HG/H (a cardinal, not necessarily finite) is called the index of HH in GG and is written [G:H][G : H].

We collect the basic properties of left cosets, since we shall use them repeatedly. The proof merely reconfirms the content of the previous chapter, but everything that follows rests on this proposition, so we write it out in full.

Proposition 2.2Basic properties of left cosets

Let HGH \le G and a,bGa, b \in G.

  1. hHh \in H if and only if hH=HhH = H.
  2. aH=bHaH = bH if and only if a1bHa^{-1}b \in H.
  3. If aHbHaH \cap bH \ne \varnothing then aH=bHaH = bH. In particular the left cosets form a partition of GG.
  4. The map HaH, hahH \to aH,\ h \mapsto ah is a bijection. In particular aH=H|aH| = |H|.
Proof(Proposition 2.2)

(1) Suppose hH=HhH = H. Since HH contains the identity ee, we get h=hehH=Hh = he \in hH = H, so hHh \in H. Conversely let hHh \in H. As HH is closed under products, hHHhH \subseteq H. For the reverse inclusion take xHx \in H. Since HH is closed under inverses we have h1Hh^{-1} \in H, hence h1xHh^{-1}x \in H, and x=h(h1x)hHx = h(h^{-1}x) \in hH. Thus HhHH \subseteq hH, and hH=HhH = H follows.

(2) Suppose aH=bHaH = bH. Then b=bebH=aHb = be \in bH = aH, so there is hHh \in H with b=ahb = ah. Multiplying on the left by a1a^{-1} gives a1b=hHa^{-1}b = h \in H. Conversely suppose a1b=hHa^{-1}b = h \in H. Then b=ahb = ah and

bH=(ah)H=a(hH)=aHbH = (ah)H = a(hH) = aH

where the second equality is associativity of the product and the third uses (1).

(3) Take xaHbHx \in aH \cap bH, so that x=ah1=bh2x = ah_1 = bh_2 for some h1,h2Hh_1, h_2 \in H. Then a1b=h1h21Ha^{-1}b = h_1h_2^{-1} \in H (using that HH is a subgroup), so aH=bHaH = bH by (2). Moreover eHe \in H gives a=aeaHa = ae \in aH, so every element lies in at least one left coset. Hence the left cosets partition GG into pairwise disjoint subsets.

(4) Surjectivity is the definition of aHaH. Injectivity follows because ah=ahah = ah' gives h=hh = h' after multiplying on the left by a1a^{-1} (cancellation in a group).

Remark 2.3

Parts (2) and (3) of Proposition 2.2 say that the binary relation on GG defined by ab    defa1bHa \sim b \overset{\text{def}}{\iff} a^{-1}b \in H is an equivalence relation(Definition 3.1)[関係と同値関係] whose equivalence classes are exactly the left cosets. For Z\mathbb{Z} with H=nZH = n\mathbb{Z}, the relation aba \sim b reads banZb - a \in n\mathbb{Z}, that is, ab(modn)a \equiv b \pmod{n}. Congruences are a special case of left cosets.

Discussing quotient groups requires the language of homomorphisms, so we set it up here. The subject is treated properly in the next chapter, The isomorphism theorems for groups.

Definition 2.4Homomorphism and kernel

Let G,GG, G' be groups and let ee' denote the identity of GG'. A map f:GGf : G \to G' satisfying

f(xy)=f(x)f(y)(x,yG)f(xy) = f(x)f(y) \qquad (\forall x, y \in G)

is called a group homomorphism. In that case

kerf:={xG:f(x)=e},imf:=f(G)\ker f := \{x \in G : f(x) = e'\}, \qquad \operatorname{im} f := f(G)

are called respectively the kernel and the image of ff. A bijective homomorphism is an isomorphism, and we write GGG \cong G'.

Remark 2.5

A homomorphism ff preserves the identity and inverses. Indeed, from f(e)=f(ee)=f(e)f(e)f(e) = f(ee) = f(e)f(e) we obtain e=f(e)e' = f(e) upon multiplying both sides by f(e)1f(e)^{-1}. Furthermore e=f(e)=f(xx1)=f(x)f(x1)e' = f(e) = f(xx^{-1}) = f(x)f(x^{-1}) gives f(x1)=f(x)1f(x^{-1}) = f(x)^{-1}. Also kerf\ker f is a subgroup of GG: from f(e)=ef(e) = e' we get ekerfe \in \ker f; if x,ykerfx, y \in \ker f then f(xy)=ee=ef(xy) = e'e' = e'; and if xkerfx \in \ker f then f(x1)=f(x)1=(e)1=ef(x^{-1}) = f(x)^{-1} = (e')^{-1} = e'.

3. When is the product of representatives defined?

Section titled “3. When is the product of representatives defined?”

Now to the heart of the matter. Given HGH \le G, we wish to put a law of composition on G/HG/H. The rule we want to lay down is

(aH)(bH):=(ab)H(aH) \cdot (bH) := (ab)H

The left-hand side takes two sets as input, while the right-hand side produces its answer by way of the representatives aa and bb. For a fixed set aHaH there are H|H| possible choices of representative aa, so unless we check that the answer does not depend on that choice, the formula defines no map at all. Let us first write down precisely what must be checked.

(W)aH=aH and bH=bH  (ab)H=(ab)H(a,a,b,bG)\text{(W)} \qquad aH = a'H \ \text{and}\ bH = b'H \ \Longrightarrow\ (ab)H = (a'b')H \qquad (\forall a, a', b, b' \in G)

When condition (W) holds, the rule above defines a map G/H×G/HG/HG/H \times G/H \to G/H, and we say that the operation is well defined. As the next example shows, (W) does not hold unconditionally.

Example 3.1Failure of the product of representatives in the symmetric group on three letters

Let G=S3G = S_3 (all permutations of {1,2,3}\{1,2,3\}), with product the composition of maps, (στ)(x)=σ(τ(x))(\sigma\tau)(x) = \sigma(\tau(x)). Take the subgroup

H={e, (12)}H = \{e,\ (1\,2)\}

so that H=2|H| = 2 and [S3:H]=3[S_3 : H] = 3.

First we compute the left cosets. For (13)(12)(1\,3)(1\,2) we have 1221 \mapsto 2 \mapsto 2, 2132 \mapsto 1 \mapsto 3, 3313 \mapsto 3 \mapsto 1, giving (123)(1\,2\,3). Likewise (23)(12)(2\,3)(1\,2) sends 1231 \mapsto 2 \mapsto 3, 2112 \mapsto 1 \mapsto 1, 3323 \mapsto 3 \mapsto 2, giving (132)(1\,3\,2). Hence

eH={e,(12)},(13)H={(13),(123)},(23)H={(23),(132)}.eH = \{e, (1\,2)\},\qquad (1\,3)H = \{(1\,3), (1\,2\,3)\},\qquad (2\,3)H = \{(2\,3), (1\,3\,2)\}.

On the other hand (12)(13)(1\,2)(1\,3) sends 1331 \mapsto 3 \mapsto 3, 2212 \mapsto 2 \mapsto 1, 3123 \mapsto 1 \mapsto 2, giving (132)(1\,3\,2), so

H(13)={(13),(132)}(13)HH(1\,3) = \{(1\,3), (1\,3\,2)\} \ne (1\,3)H

and the left and right cosets do not agree.

Now we test (W). Since (13)H=(123)H(1\,3)H = (1\,2\,3)H, the elements a=b=(13)a = b = (1\,3) and a=b=(123)a' = b' = (1\,2\,3) are representatives of the same coset. Yet

ab=(13)(13)=e  (ab)H=H,ab = (1\,3)(1\,3) = e \ \Longrightarrow\ (ab)H = H,ab=(123)(123)=(132)  (ab)H=(23)Ha'b' = (1\,2\,3)(1\,2\,3) = (1\,3\,2) \ \Longrightarrow\ (a'b')H = (2\,3)H

(here (123)(123)(1\,2\,3)(1\,2\,3) sends 1231 \mapsto 2 \mapsto 3, 2312 \mapsto 3 \mapsto 1, 3123 \mapsto 1 \mapsto 2, that is, it equals (132)(1\,3\,2)), and H(23)HH \ne (2\,3)H. Merely changing the representatives changed the resulting coset. For this HH there is no way to put a law of composition on G/HG/H by multiplying representatives.

Looking at the reason for the failure, one sees that the discrepancy between left and right cosets is what matters. Indeed, the following theorem holds. It is the most important statement in this article.

Theorem 3.2Necessary and sufficient condition for the product of cosets to be defined

Let GG be a group and HGH \le G. The following four conditions are equivalent.

  1. Condition (W) holds; that is, the operation (aH)(bH):=(ab)H(aH)(bH) := (ab)H on G/HG/H is well defined.
  2. aHa1HaHa^{-1} \subseteq H for every aGa \in G, where aHa1:={aha1:hH}aHa^{-1} := \{aha^{-1} : h \in H\}.
  3. aHa1=HaHa^{-1} = H for every aGa \in G.
  4. aH=HaaH = Ha for every aGa \in G.
Proof(Theorem 3.2)

We prove (1)(2)(3)(4)(1)(1) \Rightarrow (2) \Rightarrow (3) \Rightarrow (4) \Rightarrow (1) in that order.

(1) \Rightarrow (2). Take arbitrary aGa \in G and hHh \in H. Since hHh \in H, part (1) of Proposition 2.2 gives hH=H=eHhH = H = eH. Also a1H=a1Ha^{-1}H = a^{-1}H holds trivially. Applying (W) with representatives hh and ee in the first slot and a1a^{-1} and a1a^{-1} in the second, we obtain

(ha1)H=(ea1)H=a1H(h a^{-1})H = (e a^{-1})H = a^{-1}H

By part (2) of Proposition 2.2, this is equivalent to (ha1)1a1H(ha^{-1})^{-1} \cdot a^{-1} \in H, that is, to

ah1a1Ha h^{-1} a^{-1} \in H

As hh ranges over all of HH and hh1h \mapsto h^{-1} is a bijection of HH onto itself, h1h^{-1} also ranges over all of HH. Hence aha1Haha^{-1} \in H for every hHh \in H, that is, aHa1HaHa^{-1} \subseteq H.

(2) \Rightarrow (3). Applying (2) to aa gives aHa1HaHa^{-1} \subseteq H. Next, applying (2) to a1a^{-1} gives a1HaHa^{-1}Ha \subseteq H. Conjugating both sides by aa on the left and a1a^{-1} on the right yields

H=a(a1Ha)a1aHa1H = a(a^{-1}Ha)a^{-1} \subseteq aHa^{-1}

(we used that XYX \subseteq Y implies aXa1aYa1aXa^{-1} \subseteq aYa^{-1}). Combining the two inclusions gives aHa1=HaHa^{-1} = H. Note that the universally quantified hypothesis — that (2) holds for every element — is what does the essential work here.

(3) \Rightarrow (4). Multiplying both sides of aHa1=HaHa^{-1} = H on the right by aa gives aH=HaaH = Ha. Since this is an equality of sets, let us be more careful: if xaHx \in aH then x=ahx = ah and x=(aha1)aHax = (aha^{-1})a \in Ha; conversely if xHax \in Ha then x=hax = ha, and applying (3) to a1a^{-1} gives a1haHa^{-1}ha \in H, so x=a(a1ha)aHx = a(a^{-1}ha) \in aH.

(4) \Rightarrow (1). Suppose aH=aHaH = a'H and bH=bHbH = b'H. By part (2) of Proposition 2.2 we may write a=ah1a' = ah_1 and b=bh2b' = bh_2 with h1,h2Hh_1, h_2 \in H. Then

ab=ah1bh2.a'b' = a h_1 b h_2 .

Now h1bHbh_1 b \in Hb, and applying hypothesis (4) to bb gives Hb=bHHb = bH, so there is h3Hh_3 \in H with h1b=bh3h_1 b = b h_3. Therefore

ab=a(bh3)h2=(ab)(h3h2)(ab)Ha'b' = a (b h_3) h_2 = (ab)(h_3h_2) \in (ab)H

(here h3h2Hh_3h_2 \in H because HH is closed under products). On the other hand ab(ab)Ha'b' \in (a'b')H, so (ab)H(ab)H(ab)H \cap (a'b')H \ne \varnothing, and part (3) of Proposition 2.2 gives (ab)H=(ab)H(ab)H = (a'b')H.

flowchart LR
A["(1) the product (aH)(bH)=(ab)H is independent of representatives"] --> B["(2) aHa⁻¹ ⊆ H for every a"]
B --> C["(3) aHa⁻¹ = H for every a"]
C --> D["(4) aH = Ha for every a"]
D --> A
Equivalence of the four conditions on normality (proved in the direction of the arrows)

The value of this theorem lies in the fact that condition (2) is easy to check. Condition (1) is a universal statement about four elements, whereas (2) is nothing more than an inclusion between two subsets, expressing that conjugation does not carry elements of HH out of HH. From now on we shall verify normality mainly in the form (2).

Definition 4.1Normal subgroup

A subgroup NN of a group GG satisfying

gNg1=NgNg^{-1} = N

for every gGg \in G is called a normal subgroup of GG, written NGN \trianglelefteq G. By Theorem 3.2, this is equivalent both to ”gNg1NgNg^{-1} \subseteq N for every gGg \in G” and to ”gN=NggN = Ng for every gGg \in G”.

The element gxg1g x g^{-1} is called the conjugate of xx by gg. Setting φg(x):=gxg1\varphi_g(x) := gxg^{-1}, we have φg(xy)=gxyg1=(gxg1)(gyg1)=φg(x)φg(y)\varphi_g(xy) = gxyg^{-1} = (gxg^{-1})(gyg^{-1}) = \varphi_g(x)\varphi_g(y), so φg\varphi_g is a homomorphism, and since φg1\varphi_{g^{-1}} is its inverse it is an isomorphism of GG onto itself (an inner automorphism). A normal subgroup is therefore a subgroup fixed by every inner automorphism. This is the view from which Galois spoke of an “invariant subgroup”. Think of conjugation as changing the coordinates from which one looks: in SnS_n it amounts to relabelling the letters, in GLn\mathrm{GL}_n to changing the basis. Normal subgroups are those unaffected by such relabelling — subgroups intrinsic to the group.

Proposition 4.2Tests for normality

Let GG be a group.

  1. If GG is abelian, then every subgroup of GG is normal.
  2. If HGH \le G satisfies [G:H]=2[G : H] = 2, then HGH \trianglelefteq G.
  3. The centre Z(G):={zG:zx=xz (xG)}Z(G) := \{z \in G : zx = xz \ (\forall x \in G)\} of GG is a normal subgroup of GG.
  4. If f:GGf : G \to G' is a group homomorphism, then kerfG\ker f \trianglelefteq G.
  5. If {Nλ}λΛ\{N_\lambda\}_{\lambda \in \Lambda} is a family of normal subgroups of GG (with Λ\Lambda \ne \varnothing), then λΛNλG\bigcap_{\lambda \in \Lambda} N_\lambda \trianglelefteq G.
Proof(Proposition 4.2)

(1) Let HGH \le G, gGg \in G and hHh \in H. Since GG is abelian, ghg1=hgg1=hHghg^{-1} = hgg^{-1} = h \in H. Hence gHg1HgHg^{-1} \subseteq H for every gg, and part (2) of Theorem 3.2 gives HGH \trianglelefteq G.

(2) Take gGg \in G. If gHg \in H, then part (1) of Proposition 2.2 gives gH=HgH = H, and the same argument for right cosets gives Hg=HHg = H, so gH=HggH = Hg. Now suppose gHg \notin H. The left cosets partition GG (part (3) of Proposition 2.2) and there are [G:H]=2[G:H] = 2 of them. One of them is H=eHH = eH, so the other is GHG \setminus H. Since ggHg \in gH and gHg \notin H we have gHHgH \ne H, whence gH=GHgH = G \setminus H. Exactly the same argument for right cosets shows that the right cosets partition GG into two pieces, one being HH and the other GHG \setminus H, and since HgHHg \ne H we get Hg=GHHg = G \setminus H. Therefore gH=HggH = Hg, and part (4) of Theorem 3.2 gives HGH \trianglelefteq G.

(3) First we check that Z(G)GZ(G) \le G. The identity commutes with every element, so eZ(G)e \in Z(G). For z,wZ(G)z, w \in Z(G) and xGx \in G,

(zw)x=z(wx)=z(xw)=(zx)w=(xz)w=x(zw)(zw)x = z(wx) = z(xw) = (zx)w = (xz)w = x(zw)

so zwZ(G)zw \in Z(G). If zZ(G)z \in Z(G), multiplying zx=xzzx = xz by z1z^{-1} on both sides gives xz1=z1xxz^{-1} = z^{-1}x, so z1Z(G)z^{-1} \in Z(G). As for normality, for gGg \in G and zZ(G)z \in Z(G) we have gzg1=zgg1=zZ(G)gzg^{-1} = zgg^{-1} = z \in Z(G), so gZ(G)g1Z(G)gZ(G)g^{-1} \subseteq Z(G), and Theorem 3.2 gives Z(G)GZ(G) \trianglelefteq G.

(4) By Remark 2.5 we have kerfG\ker f \le G. For gGg \in G and nkerfn \in \ker f,

f(gng1)=f(g)f(n)f(g)1=f(g)ef(g)1=ef(gng^{-1}) = f(g)f(n)f(g)^{-1} = f(g)\,e'\,f(g)^{-1} = e'

(the first equality uses the homomorphism property together with f(g1)=f(g)1f(g^{-1}) = f(g)^{-1} from Remark 2.5). Hence gng1kerfgng^{-1} \in \ker f, that is, g(kerf)g1kerfg(\ker f)g^{-1} \subseteq \ker f, and Theorem 3.2 gives kerfG\ker f \trianglelefteq G.

(5) Put N:=λNλN := \bigcap_\lambda N_\lambda. Each NλN_\lambda contains ee, so eNe \in N. If x,yNx, y \in N then x,yNλx, y \in N_\lambda for every λ\lambda, hence xyNλxy \in N_\lambda and therefore xyNxy \in N. Similarly x1Nx^{-1} \in N. Thus NGN \le G. For gGg \in G and xNx \in N, we have xNλx \in N_\lambda and NλGN_\lambda \trianglelefteq G for every λ\lambda, so gxg1Nλgxg^{-1} \in N_\lambda. As this holds for every λ\lambda, we get gxg1Ngxg^{-1} \in N. Hence gNg1NgNg^{-1} \subseteq N and NGN \trianglelefteq G.

Example 4.3Standard examples of normal subgroups

(a) The special linear group. Let KK be a field and let det:GLn(K)K×\det : \mathrm{GL}_n(K) \to K^{\times} be the determinant. Multiplicativity of the determinant, det(AB)=detAdetB\det(AB) = \det A \det B (see Theorem 6.1[Determinants and Their Properties]), says exactly that det\det is a group homomorphism, and its kernel is SLn(K)={A:detA=1}\mathrm{SL}_n(K) = \{A : \det A = 1\}. Part (4) of Proposition 4.2 gives SLn(K)GLn(K)\mathrm{SL}_n(K) \trianglelefteq \mathrm{GL}_n(K).

(b) The alternating group. The sign map sgn:Sn{±1}\mathrm{sgn} : S_n \to \{\pm 1\} is a homomorphism whose kernel is the alternating group AnA_n. Hence AnSnA_n \trianglelefteq S_n. For n2n \ge 2 we have [Sn:An]=2[S_n : A_n] = 2, so this also follows from part (2) of Proposition 4.2.

(c) Trivial examples. In any group GG we have {e}G\{e\} \trianglelefteq G and GGG \trianglelefteq G, since g{e}g1={geg1}={e}g\{e\}g^{-1} = \{geg^{-1}\} = \{e\} and gGg1=GgGg^{-1} = G. A nontrivial group with no normal subgroups other than these two is called a simple group.

(d) A non-example. The subgroup H={e,(12)}H = \{e, (1\,2)\} of Example 3.1 is not normal in S3S_3. Indeed, computing (13)(12)(13)1(1\,3)(1\,2)(1\,3)^{-1} and using (13)1=(13)(1\,3)^{-1} = (1\,3), we get 13311 \mapsto 3 \mapsto 3 \mapsto 1, 22132 \mapsto 2 \mapsto 1 \mapsto 3, 31223 \mapsto 1 \mapsto 2 \mapsto 2, that is, (23)H(2\,3) \notin H. The index is 3, so part (2) of Proposition 4.2 does not apply, and in fact the subgroup is not normal.

Remark 4.4

Normality is not transitive. From NHN \trianglelefteq H and HGH \trianglelefteq G it does not follow that NGN \trianglelefteq G. Let us verify this in the dihedral group of order 8,

D4=r,sr4=s2=e, srs1=r1.D_4 = \langle r, s \mid r^4 = s^2 = e,\ srs^{-1} = r^{-1} \rangle .

Put H:={e,r2,s,r2s}H := \{e, r^2, s, r^2s\}. From sr2s1=(srs1)2=r2=r2sr^2s^{-1} = (srs^{-1})^2 = r^{-2} = r^2 we get sr2=r2ssr^2 = r^2s, so HH is closed under products; every element has order at most 2, so HD4H \le D_4 with H=4|H| = 4 and [D4:H]=2[D_4 : H] = 2. Part (2) of Proposition 4.2 therefore gives HD4H \trianglelefteq D_4. Likewise N:={e,s}N := \{e, s\} is a subgroup of index 2 in HH, so NHN \trianglelefteq H. But srs1=r1srs^{-1} = r^{-1} yields sr1=rssr^{-1} = rs, and

rsr1=r(sr1)=r(rs)=r2sNrsr^{-1} = r(sr^{-1}) = r(rs) = r^2s \notin N

so NN is not normal in D4D_4. The fact that a normal subgroup of a normal subgroup need not be normal in the whole group must always be kept in mind when decomposing a group in stages, that is, when working with composition series.

Theorem 5.1Quotient group

Let GG be a group and NGN \trianglelefteq G. The operation

(aN)(bN):=(ab)N(a,bG)(aN)(bN) := (ab)N \qquad (a, b \in G)

on the set G/NG/N of left cosets is well defined, and G/NG/N is a group under it. Its identity element is eN=NeN = N, and the inverse of aNaN is a1Na^{-1}N. When GG is finite, G/N=[G:N]=G/N|G/N| = [G : N] = |G| / |N|. This group G/NG/N is called the quotient group (or factor group) of GG by NN.

Proof(Theorem 5.1)

Well-definedness. Since NGN \trianglelefteq G, condition (4) of Theorem 3.2 holds, and the implication (4) \Rightarrow (1) of that theorem gives condition (W). Thus (ab)N(ab)N is determined independently of the choice of representatives.

Associativity. For a,b,cGa, b, c \in G, applying the definition of the operation twice on each side gives

((aN)(bN))(cN)=((ab)N)(cN)=((ab)c)N,\bigl((aN)(bN)\bigr)(cN) = \bigl((ab)N\bigr)(cN) = \bigl((ab)c\bigr)N,(aN)((bN)(cN))=(aN)((bc)N)=(a(bc))N.(aN)\bigl((bN)(cN)\bigr) = (aN)\bigl((bc)N\bigr) = \bigl(a(bc)\bigr)N .

These agree by associativity (ab)c=a(bc)(ab)c = a(bc) in GG.

Identity. For every aGa \in G we have (eN)(aN)=(ea)N=aN(eN)(aN) = (ea)N = aN and (aN)(eN)=(ae)N=aN(aN)(eN) = (ae)N = aN. Hence eN=NeN = N is the identity.

Inverses. (aN)(a1N)=(aa1)N=eN=N(aN)(a^{-1}N) = (aa^{-1})N = eN = N and (a1N)(aN)=(a1a)N=eN=N(a^{-1}N)(aN) = (a^{-1}a)N = eN = N. Hence the inverse of aNaN is a1Na^{-1}N.

Therefore G/NG/N is a group. As for the order, G/N|G/N| is by definition the number of left cosets, that is, the index [G:N][G:N]. If GG is finite, part (3) of Proposition 2.2 shows that GG is partitioned into [G:N][G:N] left cosets, and part (4) shows that each coset has N|N| elements, so G=[G:N]N|G| = [G:N]\,|N|, that is, [G:N]=G/N[G:N] = |G|/|N| (Lagrange's theorem(Theorem 6.1)[Subgroups and Cosets]).

Note that each element of the quotient group is a subset of GG. The coset aNaN is a bag holding together all elements that differ from aa only by something in NN, and the operation in the quotient group is the multiplication of such bags. Here is a picture.

GG / NNaNbNeNaNbNπ
The natural projection π collapses each coset to a single point. All distinctions inside N are forgotten completely

Proposition 5.2The natural projection

Let NGN \trianglelefteq G. The map

π:GG/N,π(g)=gN\pi : G \longrightarrow G/N, \qquad \pi(g) = gN

is a surjective homomorphism with kerπ=N\ker \pi = N. It is called the natural projection (or canonical surjection).

Proof(Proposition 5.2)

Homomorphism. For a,bGa, b \in G, the very definition of the operation in the quotient group gives

π(ab)=(ab)N=(aN)(bN)=π(a)π(b).\pi(ab) = (ab)N = (aN)(bN) = \pi(a)\pi(b).

Surjectivity. By definition, every element of G/NG/N can be written as gNgN for some gGg \in G, and this is π(g)\pi(g).

Kernel. By Theorem 5.1 the identity of G/NG/N is N=eNN = eN. Hence

gkerπ    gN=eN    g1eN    g1N    gNg \in \ker\pi \iff gN = eN \iff g^{-1}e \in N \iff g^{-1} \in N \iff g \in N

where the second equivalence is part (2) of Proposition 2.2 and the last holds because NN is closed under inverses. Therefore kerπ=N\ker\pi = N.

Corollary 5.3

For a subgroup NN of a group GG, the following are equivalent.

  1. NGN \trianglelefteq G.
  2. There exist a group GG' and a homomorphism f:GGf : G \to G' with N=kerfN = \ker f.
Proof(Corollary 5.3)

(1)(2)(1) \Rightarrow (2): the map π:GG/N\pi : G \to G/N of Proposition 5.2 satisfies kerπ=N\ker \pi = N, so we may take G=G/NG' = G/N and f=πf = \pi. (2)(1)(2) \Rightarrow (1): this is precisely part (4) of Proposition 4.2.

This corollary matters conceptually. The definition of a normal subgroup, "gNg1=NgNg^{-1} = N", looks at first like a contrived condition, but it in fact means “the part that is crushed by a homomorphism”. Determining all homomorphisms out of GG and determining all normal subgroups of GG are one and the same task. The refinement of this correspondence is the isomorphism theorem of the next chapter, G/kerfimfG/\ker f \cong \operatorname{im} f (Theorem 5.1[群の準同型定理]), treated in The isomorphism theorems for groups.

Remark 5.4

If gGg \in G has finite order and gm=eg^m = e, then (gN)m=gmN=eN=N(gN)^m = g^mN = eN = N, so the order of gNgN divides the order of gg. It may, however, be strictly smaller. With G=ZG = \mathbb{Z}, N=12ZN = 12\mathbb{Z} and g=1g = 1, the element gg has infinite order while gNgN has order 12. Passing to a quotient can only lower orders, never raise them.

6. Example: the congruence group Z/nZ in full detail

Section titled “6. Example: the congruence group Z/nZ in full detail”

Let us examine the most basic quotient group down to the last detail. The stage is the additive group (Z,+)(\mathbb{Z}, +), so we write everything additively: cosets are a+Na + N and the operation is (a+N)+(b+N)=(a+b)+N(a+N) + (b+N) = (a+b)+N.

Lemma 6.1

If HH is a subgroup of the additive group Z\mathbb{Z}, then there is exactly one integer n0n \ge 0 with H=nZ={nk:kZ}H = n\mathbb{Z} = \{nk : k \in \mathbb{Z}\}.

Proof(Lemma 6.1)

Existence. If H={0}H = \{0\}, take n=0n = 0, so that H=0ZH = 0\mathbb{Z}. Assume from now on that H{0}H \ne \{0\}. Then there is mHm \in H with m0m \ne 0, and since HH is closed under inverses (here m-m), at least one of mm and m-m is positive. Hence HH contains positive integers. By the well-ordering of the natural numbers there is a least positive integer contained in HH; call it nn.

We show nZHn\mathbb{Z} \subseteq H. We have nHn \in H, and HH is closed under addition and inverses, so by induction nkHnk \in H for every k0k \ge 0, and also (nk)=n(k)H-(nk) = n(-k) \in H (formally one applies the principle of mathematical induction(Theorem 3.2)[Techniques of Proof] directly). Hence nZHn\mathbb{Z} \subseteq H.

We show the reverse inclusion. Take any aHa \in H and write a=qn+ra = qn + r with q,rZq, r \in \mathbb{Z} and 0r<n0 \le r < n by the division algorithm. Since qnnZHqn \in n\mathbb{Z} \subseteq H and aHa \in H, we get r=aqnHr = a - qn \in H. If r>0r > 0, then rr is a positive integer in HH with r<nr < n, contradicting the minimality of nn. Hence r=0r = 0, that is, a=qnnZa = qn \in n\mathbb{Z}. Therefore H=nZH = n\mathbb{Z}.

Uniqueness. Suppose nZ=nZn\mathbb{Z} = n'\mathbb{Z} with n,n0n, n' \ge 0. If n=0n = 0 then nZ={0}n\mathbb{Z} = \{0\}, so nZ={0}n'\mathbb{Z} = \{0\} and n=0n' = 0. If n>0n > 0, then nn is the least positive integer in nZn\mathbb{Z}, and the same holds for nn', so n=nn = n'.

Since Z\mathbb{Z} is abelian, part (1) of Proposition 4.2 shows that all its subgroups are normal. Hence Theorem 5.1 yields, for every n0n \ge 0, the quotient group

Z/nZ={a+nZ:aZ}\mathbb{Z}/n\mathbb{Z} = \{\,a + n\mathbb{Z} : a \in \mathbb{Z}\,\}

We abbreviate a+nZa + n\mathbb{Z} by aˉ\bar{a}. Rewriting part (2) of Proposition 2.2 additively gives

aˉ=bˉ    a+bnZ    nba    ab(modn)\bar{a} = \bar{b} \iff -a + b \in n\mathbb{Z} \iff n \mid b - a \iff a \equiv b \pmod{n}

so the elements of Z/nZ\mathbb{Z}/n\mathbb{Z} are exactly the congruence classes modulo nn. And the assertion that the quotient operation aˉ+bˉ=a+b\bar{a} + \bar{b} = \overline{a+b} is well defined is nothing but the familiar statement of elementary number theory,

aa, bb(modn)  a+ba+b(modn)a \equiv a',\ b \equiv b' \pmod{n} \ \Longrightarrow\ a + b \equiv a' + b' \pmod{n}

(proved directly by deducing n(a+b)(a+b)n \mid (a+b) - (a'+b') from naan \mid a - a' and nbbn \mid b - b'). What Theorem 3.2 tells us is that this naive fact held because nZn\mathbb{Z} is a normal subgroup.

Example 6.2The elements of Z/nZ and their orders

Let n1n \ge 1.

Number of elements. For any aZa \in \mathbb{Z} the division algorithm gives a=qn+ra = qn + r with 0r<n0 \le r < n, so aˉ=rˉ\bar{a} = \bar{r}. Hence Z/nZ={0ˉ,1ˉ,,n1}\mathbb{Z}/n\mathbb{Z} = \{\bar{0}, \bar{1}, \ldots, \overline{n-1}\}. Moreover if 0r<r<n0 \le r < r' < n then 0<rr<n0 < r' - r < n, so nrrn \nmid r' - r and therefore rˉrˉ\bar{r} \ne \bar{r'}. Thus these nn elements are distinct and Z/nZ=n|\mathbb{Z}/n\mathbb{Z}| = n, in agreement with the computation of [G:N][G:N] in Theorem 5.1.

Cyclicity. Since aˉ=a1ˉ\bar{a} = a\bar{1} (add 1ˉ\bar{1} to itself aa times; for a<0a < 0, add 1ˉ-\bar 1 to itself a|a| times), we have Z/nZ=1ˉ\mathbb{Z}/n\mathbb{Z} = \langle \bar{1} \rangle, a cyclic group of order nn.

Orders of elements. The order of aˉ\bar{a} is n/gcd(a,n)n/\gcd(a, n). Indeed, put d:=gcd(a,n)d := \gcd(a,n), a=daa = da', n=dnn = dn' (so that gcd(a,n)=1\gcd(a', n') = 1). Then for k1k \ge 1,

kaˉ=0ˉ    nka    dnkda    nka    nkk\bar{a} = \bar{0} \iff n \mid ka \iff dn' \mid k d a' \iff n' \mid ka' \iff n' \mid k

where the last equivalence uses gcd(a,n)=1\gcd(a', n') = 1 together with Euclid’s lemma (if nkan' \mid ka' and gcd(n,a)=1\gcd(n', a') = 1 then nkn' \mid k). The least positive integer kk satisfying this is k=n=n/dk = n' = n/d.

Generators. It follows that aˉ\bar{a} generates Z/nZ\mathbb{Z}/n\mathbb{Z} (that is, has order nn) precisely when gcd(a,n)=1\gcd(a,n) = 1. Hence the number of generators equals Euler’s totient φ(n)\varphi(n).

Example 6.3The clock group Z/12Z

Let us compute explicitly for n=12n = 12. Using the formula ord(aˉ)=12/gcd(a,12)\mathrm{ord}(\bar a) = 12/\gcd(a,12) of Example 6.2 we obtain the following table.

aa01234567891011
gcd(a,12)\gcd(a,12)1212341614321
order of aˉ\bar a1126431221234612

The generators are 1ˉ,5ˉ,7ˉ,11\bar1, \bar5, \bar7, \overline{11}, four in number, matching φ(12)=12(11/2)(11/3)=4\varphi(12) = 12(1 - 1/2)(1-1/3) = 4.

The subgroups are dˉ\langle \bar{d}\rangle (of order 12/d12/d), one for each positive divisor dd of 1212, six in all. For instance

3ˉ={0ˉ,3ˉ,6ˉ,9ˉ},4ˉ={0ˉ,4ˉ,8ˉ}.\langle \bar{3}\rangle = \{\bar{0}, \bar{3}, \bar{6}, \bar{9}\}, \qquad \langle \bar{4}\rangle = \{\bar{0}, \bar{4}, \bar{8}\} .

Since Z/12Z\mathbb{Z}/12\mathbb{Z} is abelian these are all normal, so we may form quotients again. With N:=3ˉN := \langle\bar 3\rangle we get (Z/12Z)/N=12/4=3|(\mathbb{Z}/12\mathbb{Z})/N| = 12/4 = 3, and the cosets are the three sets

0ˉ+N={0ˉ,3ˉ,6ˉ,9ˉ},1ˉ+N={1ˉ,4ˉ,7ˉ,10},2ˉ+N={2ˉ,5ˉ,8ˉ,11}\bar0 + N = \{\bar0,\bar3,\bar6,\bar9\},\quad \bar1 + N = \{\bar1,\bar4,\bar7,\overline{10}\},\quad \bar2 + N = \{\bar2,\bar5,\bar8,\overline{11}\}

Since aˉ+N\bar{a} + N depends only on the remainder of aa upon division by 3, this quotient group is isomorphic to Z/3Z\mathbb{Z}/3\mathbb{Z}; as an operation it corresponds to “take the remainder modulo 12, then take the remainder modulo 3”, which is the same as “take the remainder modulo 3 from the start”. This phenomenon, that a quotient of a quotient is again a quotient, is formulated in general as the third isomorphism theorem in the next chapter (Corollary 7.2[群の準同型定理]).

Example 7.1Quotients of the symmetric group, the general linear group and the real line

(a) S3/A3S_3/A_3. The subgroup A3={e,(123),(132)}A_3 = \{e, (1\,2\,3), (1\,3\,2)\} has index 2, so part (2) of Proposition 4.2 gives A3S3A_3 \trianglelefteq S_3. The quotient group consists of the two elements {A3, (12)A3}\{A_3,\ (1\,2)A_3\}, where A3A_3 is the set of all even permutations and (12)A3(1\,2)A_3 the set of all odd ones. The operation in the quotient is exactly the rule

(even)(even)=even,(even)(odd)=odd,(odd)(odd)=even(\text{even})(\text{even}) = \text{even},\quad (\text{even})(\text{odd}) = \text{odd},\quad (\text{odd})(\text{odd}) = \text{even}

and S3/A3Z/2ZS_3/A_3 \cong \mathbb{Z}/2\mathbb{Z}. The “contents” of a permutation are forgotten entirely; only the parity survives.

(b) GLn(R)/SLn(R)\mathrm{GL}_n(\mathbb{R})/\mathrm{SL}_n(\mathbb{R}). By Example 4.3 (a) we have SLn(R)GLn(R)\mathrm{SL}_n(\mathbb{R}) \trianglelefteq \mathrm{GL}_n(\mathbb{R}). Writing S:=SLn(R)S := \mathrm{SL}_n(\mathbb{R}), part (2) of Proposition 2.2 gives

AS=BS    A1BS    det(A1B)=1    detA=detBAS = BS \iff A^{-1}B \in S \iff \det(A^{-1}B) = 1 \iff \det A = \det B

so the cosets correspond bijectively to the values of the determinant. The operation is (AS)(BS)=(AB)S(AS)(BS) = (AB)S, and determinants multiply as det(AB)=detAdetB\det(AB) = \det A \det B, so GLn(R)/SLn(R)R×\mathrm{GL}_n(\mathbb{R})/\mathrm{SL}_n(\mathbb{R}) \cong \mathbb{R}^{\times}. The quotient group is the world in which a matrix is seen through its determinant alone.

(c) R/Z\mathbb{R}/\mathbb{Z}. Since (R,+)(\mathbb{R}, +) is abelian we have ZR\mathbb{Z} \trianglelefteq \mathbb{R} and may form R/Z\mathbb{R}/\mathbb{Z}. As x+Z=y+Zx + \mathbb{Z} = y + \mathbb{Z} is equivalent to xyZx - y \in \mathbb{Z}, each coset has exactly one representative in the interval [0,1)[0,1), namely the fractional part of xx. The operation is “add and discard the integer part”, that is, addition mod 1\mathrm{mod}\ 1. The map x+Ze2πixx + \mathbb{Z} \mapsto e^{2\pi i x} is well defined (replacing xx by x+mx + m does not change the value, since e2πim=1e^{2\pi i m} = 1) and is a bijective homomorphism, so R/Z\mathbb{R}/\mathbb{Z} is isomorphic to the multiplicative group of the unit circle in the complex plane. Winding an infinite line into a circle, one unit of length at a time, is in group-theoretic terms exactly the passage to a quotient.

What these examples have in common is that the quotient group represents what remains after the information we wish to discard has been discarded. In (a) the fine detail of a permutation is discarded and only the parity kept; in (b) the fine detail of a matrix is discarded and only the determinant kept; in (c) the integer part of a real number is discarded and only the fractional part kept. In each case the totality of the discarded information is precisely the normal subgroup NN, and what remains is the quotient group G/NG/N. What Corollary 5.3 said is that this splitting into “discarded” and “retained” is the same thing as a single homomorphism.

The general theory of groups starts from Introduction to group theory: the definition and examples, but the basic policy when facing the great problem of classifying groups is: find a normal subgroup NN and decompose into NN and G/NG/N. The groups for which this policy finally stops working, the simple groups, are the “elementary particles” of the decomposition. Galois theory, which decides the solvability of an equation by decomposing a group (An invitation to Galois theory), is likewise an argument tracing a chain of normal subgroups (Theorem 5.4[ガロア理論への招待]).

Exercise 8.1Easy

In Z/18Z\mathbb{Z}/18\mathbb{Z}, find the order of 12\overline{12} and list all elements of 12\langle \overline{12}\rangle. Then find the order of the quotient group (Z/18Z)/12(\mathbb{Z}/18\mathbb{Z})/\langle\overline{12}\rangle.

Solution

By Example 6.2 the order of 12\overline{12} is 18/gcd(12,18)=18/6=318/\gcd(12,18) = 18/6 = 3. Hence

12={0ˉ, 12, 24}={0ˉ, 12, 6ˉ}\langle \overline{12}\rangle = \{\bar 0,\ \overline{12},\ \overline{24}\} = \{\bar 0,\ \overline{12},\ \bar 6\}

(since 24=18+624 = 18 + 6, we have 24=6ˉ\overline{24} = \bar 6). As Z/18Z\mathbb{Z}/18\mathbb{Z} is abelian, part (1) of Proposition 4.2 shows this subgroup is normal, and the order formula in Theorem 5.1 gives

(Z/18Z)/12=183=6.\bigl|(\mathbb{Z}/18\mathbb{Z})/\langle\overline{12}\rangle\bigr| = \frac{18}{3} = 6 .

Note that 12=6ˉ\langle\overline{12}\rangle = \langle\bar 6\rangle, the subgroup corresponding to the divisor 66 of 1818.

Exercise 8.2Standard

Let N,MGN, M \trianglelefteq G. Show that NMN \cap M and NM:={nm:nN, mM}NM := \{nm : n \in N,\ m \in M\} are both normal subgroups of GG.

Solution

The case of NMN \cap M. Apply part (5) of Proposition 4.2 with Λ={1,2}\Lambda = \{1,2\}, N1=NN_1 = N, N2=MN_2 = M; the claim follows at once.

The case of NMNM. First we show it is a subgroup. We have e=eeNMe = ee \in NM. To see closure under products, take n1m1, n2m2NMn_1m_1,\ n_2m_2 \in NM. Since MGM \trianglelefteq G we have n21m1n2Mn_2^{-1}m_1n_2 \in M; calling this element mm', we get m1n2=n2mm_1 n_2 = n_2 m'. Hence

(n1m1)(n2m2)=n1(m1n2)m2=n1n2mm2NM(n_1m_1)(n_2m_2) = n_1(m_1n_2)m_2 = n_1n_2m'm_2 \in NM

(as n1n2Nn_1n_2 \in N and mm2Mm'm_2 \in M). As for inverses, NGN \trianglelefteq G gives m1n1mNm^{-1}n^{-1}m \in N, so

(nm)1=m1n1=(m1n1m)m1NM.(nm)^{-1} = m^{-1}n^{-1} = (m^{-1}n^{-1}m)\,m^{-1} \in NM .

Therefore NMGNM \le G.

Next, normality. For gGg \in G, nNn \in N and mMm \in M, inserting g1g=eg^{-1}g = e gives

g(nm)g1=(gng1)(gmg1)NMg(nm)g^{-1} = (gng^{-1})(gmg^{-1}) \in NM

(here gng1Ngng^{-1} \in N because NGN \trianglelefteq G, and gmg1Mgmg^{-1} \in M because MGM \trianglelefteq G). Hence g(NM)g1NMg(NM)g^{-1} \subseteq NM for every gg, and part (2) of Theorem 3.2 gives NMGNM \trianglelefteq G.

Note that if the normality of MM is dropped and MM is merely a subgroup, then NMNM is still a subgroup (use NGN \trianglelefteq G in place of MGM \trianglelefteq G in the computation above) but need not be normal.

Exercise 8.3Standard

Let GG be a group. Show that if the quotient group G/Z(G)G/Z(G) is cyclic, then GG is abelian. Using this, show that if G=p2|G| = p^2 with pp prime, then GG is abelian. You may use Lagrange’s theorem and the fact that a group of prime order is cyclic. You may also assume Z(G){e}Z(G) \ne \{e\} (that the centre of a pp-group is nontrivial).

Solution

Write Z:=Z(G)Z := Z(G). By part (3) of Proposition 4.2 we have ZGZ \trianglelefteq G, so the quotient group G/ZG/Z is defined.

First part. Suppose G/Z=gZG/Z = \langle gZ\rangle. Take any xGx \in G. Then xZG/ZxZ \in G/Z, so xZ=(gZ)k=gkZxZ = (gZ)^k = g^kZ for some integer kk. By part (2) of Proposition 2.2 we have (gk)1xZ(g^k)^{-1}x \in Z, so putting z:=gkxZz := g^{-k}x \in Z we may write x=gkzx = g^kz. Similarly any yGy \in G can be written y=glwy = g^lw with wZw \in Z. Since zz and ww commute with every element of GG,

xy=gkzglw=gkglzw=gk+lzw,yx=glwgkz=glgkwz=gk+lwz.xy = g^kz\,g^lw = g^kg^lzw = g^{k+l}zw, \qquad yx = g^lw\,g^kz = g^lg^kwz = g^{k+l}wz .

From z,wZz, w \in Z we get zw=wzzw = wz (apply the definition of the centre to zZz \in Z and wGw \in G, say), so xy=yxxy = yx. As xx and yy were arbitrary, GG is abelian.

Second part. Suppose G=p2|G| = p^2. Since ZGZ \le G, Lagrange’s theorem gives Z{1,p,p2}|Z| \in \{1, p, p^2\}, and Z1|Z| \ne 1 by hypothesis. If Z=p2|Z| = p^2 then Z=GZ = G and GG is abelian. We rule out the case Z=p|Z| = p. In that case Theorem 5.1 gives G/Z=p2/p=p|G/Z| = p^2/p = p, so G/ZG/Z is cyclic, a group of prime order being cyclic. By the first part GG is then abelian, that is, Z=GZ = G and Z=p2|Z| = p^2, contradicting Z=p|Z| = p. Hence Z=p2|Z| = p^2 is the only possibility and GG is abelian.

Exercise 8.4Hard

Show that the alternating group A4A_4 (of order 12) has no subgroup of order 6. In particular, the converse of Lagrange’s theorem is false.

Solution

Suppose there were HA4H \le A_4 with H=6|H| = 6. The order formula of Theorem 5.1 gives [A4:H]=12/6=2[A_4 : H] = 12/6 = 2, so part (2) of Proposition 4.2 gives HA4H \trianglelefteq A_4. Hence the quotient group A4/HA_4/H is defined and has order 2.

In a group of order 2, u2u^2 is the identity for every element uu (clear if uu is the identity; otherwise uu has order 2 by Lagrange’s theorem). Let π:A4A4/H\pi : A_4 \to A_4/H be the natural projection (Proposition 5.2). Then for every xA4x \in A_4,

π(x2)=π(x)2=H\pi(x^2) = \pi(x)^2 = H

so x2kerπ=Hx^2 \in \ker\pi = H.

Now let σA4\sigma \in A_4 be a 3-cycle. Since σ3=e\sigma^3 = e, we have (σ2)2=σ4=σ(\sigma^2)^2 = \sigma^4 = \sigma, and applying the fact above with x=σ2x = \sigma^2 gives σ=(σ2)2H\sigma = (\sigma^2)^2 \in H. Thus every 3-cycle in A4A_4 lies in HH.

The number of 3-cycles on {1,2,3,4}\{1,2,3,4\} is 4×2=84 \times 2 = 8: there are (43)=4\binom{4}{3} = 4 ways to choose the three letters moved, and 22 ways to cycle the chosen three. A 3-cycle is a product of two transpositions ((abc)=(ab)(bc)(a\,b\,c) = (a\,b)(b\,c)), hence even, so all of them lie in A4A_4. Therefore H8+1=9|H| \ge 8 + 1 = 9 (counting the identity as well), contradicting H=6|H| = 6.

Hence no subgroup of order 6 exists. Since 6126 \mid 12 and yet there is no subgroup of order 6, the converse of Lagrange’s theorem — that for every divisor dd of G|G| there is a subgroup of order dd — is false.

  • Matsuzaka Kazuo, Daisūkei Nyūmon (Introduction to Algebraic Systems), Iwanami Shoten, 1976 (in Japanese) — Chapter 3 (group theory) gives a detailed account of cosets, normal subgroups and quotient groups. One of the most careful introductions available in Japanese.
  • Katsura Toshiyuki, Daisūgaku I: Gun to Kan (Algebra I: Groups and Rings), University of Tokyo Press, 2004 (in Japanese) — Chapter 1. A concise treatment of normal subgroups and the isomorphism theorems.
  • Yukie Akihiko, Daisūgaku 1: Gunron Nyūmon (Algebra 1: Introduction to Group Theory), Nippon Hyoron Sha, 2010 (in Japanese) — Chapter 2. Rich in concrete examples, well suited to practising computations in SnS_n and DnD_n.
  • S. Lang, Algebra, 3rd revised ed., Springer GTM 211, 2002 — Chapter I (Groups). Organizes quotient groups and the isomorphism theorems in categorical language.
  • D. S. Dummit and R. M. Foote, Abstract Algebra, 3rd ed., Wiley, 2004 — Chapter 3 (Quotient Groups and Homomorphisms). Contains the tests for normality together with a wealth of exercises.
  • C. F. Gauss, Disquisitiones Arithmeticae, 1801 — Chapter 1. The original source for the notation of congruences and for their compatibility with addition, subtraction and multiplication.

Appendix: A normal subgroup is a union of conjugacy classes

Section titled “Appendix: A normal subgroup is a union of conjugacy classes”

The partition by conjugacy. The relation xy    gG, y=gxg1x \approx y \iff \exists g \in G,\ y = gxg^{-1} on a group GG is an equivalence relation: it is reflexive because x=exe1x = exe^{-1}; symmetric because y=gxg1y = gxg^{-1} gives x=g1y(g1)1x = g^{-1}y(g^{-1})^{-1}; and transitive because y=gxg1y = gxg^{-1} and z=hyh1z = hyh^{-1} give z=(hg)x(hg)1z = (hg)x(hg)^{-1}. Its equivalence classes are called conjugacy classes. Directly from the definition, for a subgroup NGN \le G,

NG    N is a union of conjugacy classesN \trianglelefteq G \iff N \text{ is a union of conjugacy classes}

Indeed, if NGN \trianglelefteq G then all conjugates gxg1gxg^{-1} of xNx \in N lie in NN, so NN contains the whole conjugacy class of xx; conversely if NN is a union of conjugacy classes then gNg1NgNg^{-1} \subseteq N for every gg, so NN is normal by Theorem 3.2. This reformulation is a practical tool when listing all normal subgroups of a finite group, because the constraint that the sizes of the classes must sum to N|N| and that N|N| must divide G|G| (Lagrange’s theorem) bites hard.

The case of S3S_3. The conjugacy classes of S3S_3 are three: {e}\{e\} (one element), the class of transpositions {(12),(13),(23)}\{(1\,2),(1\,3),(2\,3)\} (three elements), and the class of 3-cycles {(123),(132)}\{(1\,2\,3),(1\,3\,2)\} (two elements). The unions containing ee whose size divides 66 are 11, 1+2=31+2 = 3 and 1+3+2=61+3+2 = 6, three possibilities in all (note that 1+3=41+3 = 4 does not divide 6). The corresponding subsets are {e}\{e\}, A3A_3 and S3S_3, all of which are subgroups. Hence the normal subgroups of S3S_3 are exactly {e}, A3, S3\{e\},\ A_3,\ S_3. That the subgroup {e,(12)}\{e,(1\,2)\} of Example 3.1 was not normal is now immediate: it contains only part of the class of transpositions.

The case of A4A_4. Consider V:={e, (12)(34), (13)(24), (14)(23)}V := \{e,\ (1\,2)(3\,4),\ (1\,3)(2\,4),\ (1\,4)(2\,3)\}. Conjugation obeys the rule σ(ab)(cd)σ1=(σ(a)σ(b))(σ(c)σ(d))\sigma (a\,b)(c\,d) \sigma^{-1} = (\sigma(a)\,\sigma(b))(\sigma(c)\,\sigma(d)), so conjugating (12)(34)(1\,2)(3\,4) by σ=(123)\sigma = (1\,2\,3) yields (23)(14)(2\,3)(1\,4), and by σ=(132)\sigma = (1\,3\,2) yields (31)(24)(3\,1)(2\,4). Thus the three double transpositions are conjugate to one another inside A4A_4, and VV is the union of the class {e}\{e\} with this three-element class. Once one checks that VV is a subgroup (that the product of two double transpositions is the third, and that every element has order 2), the criterion above gives VA4V \trianglelefteq A_4. By Theorem 5.1 we have A4/V=12/4=3|A_4/V| = 12/4 = 3, and this quotient group is cyclic of order 3. Combined with Exercise 8.4, this reveals an asymmetric picture: A4A_4 has no subgroup of order 6, yet it has a normal subgroup of order 4.

Report an error in this article ・Operated by: Mugen Giken LLCPricingTermsLegal notice

© 2026 夢現技研合同会社 ・Feeding the text to an LLM is welcome. Code samples are MIT licensed.