The integers modulo $n$ were built by adding congruence classes through their representatives: $[a] + [b] = [a + b]$. The rule makes sense because the answer does not depend on which representatives are chosen. The quotient ring $R/I$ was built the same way from the cosets of an ideal. For a group $G$ and a subgroup $H$, the cosets $aH$ of the previous article invite the same construction, with the rule
$$(aH)(bH) = abH.$$
Does this rule make sense for every subgroup?
Sometimes it works. In the stop sign's group $D_8$, take the rotation subgroup $\langle r \rangle$. Its two cosets are the rotations and the reflections. The rule says that a rotation times a rotation is a rotation, a rotation times a reflection is a reflection, and a reflection times a reflection is a rotation. All three are true whichever rotations and reflections are chosen, so the cosets multiply like the two elements of $\{1, -1\}$.
Sometimes it fails. Take instead $H = \langle r^4, s \rangle = \{1, r^4, s, r^4 s\}$. Since $s \in H$, the cosets $1H$ and $sH$ are the same set, $H$ itself. Multiplying that coset by $rH$ should therefore give one answer, but the representatives disagree. The product $1 \cdot r = r$ lies in $rH$, while $s \cdot r = r^{-1}s = r^7 s$ lies in $r^3 H$. The script confirms that $rH = \{r, r^5, rs, r^5 s\}$ and $r^3 H = \{r^3, r^7, r^3 s, r^7 s\}$ are different cosets, so the product of the two cosets is not defined.
The difference between the two subgroups is that the left and right cosets of $\langle r \rangle$ agree, while those of $H$ do not (Cosets and Lagrange's Theorem). This article shows that the rule works exactly for the normal subgroups, those closed under conjugation. Their cosets then form a group, the quotient group $G/N$.
First we define normal subgroups and test examples in $D_8$, $S_8$ and the cube group. Then we prove that the coset product is well defined exactly for normal subgroups, and build the quotient group. Finally we meet the first quotients, among them the parity of the cube, and set normal subgroups beside ideals.
This is the second article of Cosets, Normal Subgroups and Quotient Groups, and the home of normal subgroups and quotient groups. It covers §§8.2 and 8.3 of the course text (Hungerford, 2014), with Judson (2026, §10.1) as a free companion. It uses cosets and, from Groups, conjugation and kernels.
Normal subgroups
What property of $\langle r \rangle$ made its left and right cosets agree? For elements $g$ and $x$ of a group $G$, the element $gxg^{-1}$ is the conjugate of $x$ by $g$. Conjugation by a fixed $g$ is an automorphism of $G$ (Group Homomorphisms and Isomorphisms), so it carries each subgroup $H$ to a subgroup $gHg^{-1} = \{ ghg^{-1} : h \in H \}$ of the same order. In the symmetric groups conjugation renames the objects being permuted, and on the cube it does the work of a move on other pieces.
Definition. A subgroup $N$ of a group $G$ is normal, written $N \trianglelefteq G$, if $gng^{-1} \in N$ for every $g \in G$ and every $n \in N$; that is, if $gNg^{-1} \subseteq N$ for every $g \in G$.
The course text defines a normal subgroup by the condition $Na = aN$ for every $a$, and Judson by $gN = Ng$. The next theorem shows that these definitions agree with this one. The idea: $gn = n'g$ and $gng^{-1} = n'$ are the same equation, rearranged.
Theorem. For a subgroup $N$ of a group $G$, the following are equivalent:
- $gNg^{-1} \subseteq N$ for every $g \in G$;
- $gNg^{-1} = N$ for every $g \in G$;
- $gN = Ng$ for every $g \in G$.
Proof. (1) implies (2). Let $g \in G$; one inclusion is (1). For the other, let $n \in N$. Applying (1) to the element $g^{-1}$ gives $g^{-1} n g \in N$, and then $n = g(g^{-1}ng)g^{-1}$ is an element of $gNg^{-1}$.
(2) implies (3). Let $n \in N$. By (2), $gng^{-1} = n'$ for some $n' \in N$, and then $gn = n'g \in Ng$; so $gN \subseteq Ng$. Applying (2) to $g^{-1}$ in the same way, $g^{-1}ng = n''$ for some $n'' \in N$, so $ng = gn'' \in gN$; so $Ng \subseteq gN$.
(3) implies (1). Let $n \in N$. Then $gn \in gN = Ng$, so $gn = n'g$ for some $n' \in N$, and $gng^{-1} = n' \in N$. $\blacksquare$
This is the course text's Theorem 8.11 and Judson's Theorem 10.1.3. A common misreading of condition (3) is that $g$ must commute with each element of $N$. It says only that $g$ commutes with $N$ as a whole: moving $g$ past an element of $N$ costs a change of that element. In $D_8$, $s r = r^7 s$: the reflection does not commute with the rotation $r$, but it carries it to another rotation, which is all that normality of $\langle r \rangle$ asks.
Examples
Abelian groups. In an abelian group $gng^{-1} = n$, so every subgroup is normal. All nine subgroups of $\mathbb{Z}_{36}$, all subgroups of $U(n)$ and every $n\mathbb{Z}$ in $\mathbb{Z}$ are normal. In every group, $\{e\}$ and $G$ are normal.
Kernels. If $f: G \to K$ is a homomorphism, its kernel is normal: for $k$ in the kernel, $f(gkg^{-1}) = f(g)\,e\,f(g)^{-1} = e$. This was observed in Group Homomorphisms and Isomorphisms, and it is the course text's Theorem 8.16. Three kernels met earlier are therefore normal:
- the alternating group $A_n$, the kernel of the sign, in $S_n$;
- the rotation subgroup $\langle r \rangle$, the kernel of the map $D_8 \to \{1, -1\}$ that tells rotations from reflections, in $D_8$;
- the moves of the Rubik's cube that leave every corner cubie in its own position, the kernel of the map from the cube group to the permutations of the eight corner positions.
The last has a direct meaning. If $M$ leaves the corners in place, then so does the conjugate $X M X^{-1}$ for any move $X$. The setup $X$ moves the corners somewhere, $M$ leaves them there, and $X^{-1}$ brings them back.
Subgroups of index 2.
Proposition. A subgroup of index 2 is normal.
The idea: with only two cosets on each side, the coset other than $N$ must be everything outside $N$, on the left and on the right alike.
Proof. Let $[G : N] = 2$ and $g \in G$. If $g \in N$, then $gN = N = Ng$. If $g \notin N$, then $gN \neq N$, and since the two left cosets $N$ and $gN$ cover $G$ without overlap, $gN$ is the complement $G \setminus N$. The same argument with right cosets gives $Ng = G \setminus N$. So $gN = Ng$ for every $g$, and $N$ is normal by the theorem. $\blacksquare$
So in a group of order 16 every subgroup of order 8 is normal, without a single conjugate being computed. In $D_8$ these are the rotations $\langle r \rangle$ and the two subgroups $\langle r^2, s \rangle$ and $\langle r^2, rs \rangle$. The script checks all three, and confirms that $A_4$, of index 2 in $S_4$, is normal as well. The same reasoning applies to $A_n$ in $S_n$, half of $S_n$ (Symmetric Groups and Permutations).
Checking generators is enough
Must a test of normality conjugate by every element of $G$? Conjugating by a generating set suffices.
Proposition. Let $N$ be a subgroup of a finite group $G$, and let $S$ be a set of generators of $G$. If $sNs^{-1} \subseteq N$ for every $s \in S$, then $N$ is normal.
The idea: the elements that conjugate $N$ into itself are closed under products, so they form a subgroup, and a subgroup that contains the generators is all of $G$.
Proof. Let $T$ be the set of elements $g \in G$ with $gNg^{-1} \subseteq N$. It is closed under products: if $g, h \in T$, then $(gh)N(gh)^{-1} = g(hNh^{-1})g^{-1} \subseteq gNg^{-1} \subseteq N$. Since $G$ is finite, $T$ is a nonempty finite subset closed under products, hence a subgroup by the finite subgroup test. It contains $S$ by hypothesis, so it contains the subgroup generated by $S$, which is $G$. $\blacksquare$
Without finiteness the same proof works if the hypothesis is required for the inverses of the generators too. For $D_8 = \langle r, s \rangle$ the two conjugations are
$$r \, r^a \, r^{-1} = r^a, \qquad s \, r^a \, s^{-1} = r^{-a},$$
so conjugation by either generator carries every rotation to a rotation in the same cyclic subgroup. (The subgroup $\langle r^d \rangle$ contains $r^{-a}$ whenever it contains $r^a$.) Hence each rotation subgroup $\langle r \rangle$, $\langle r^2 \rangle$, $\langle r^4 \rangle$ is normal in $D_8$. The script confirms this, and checks that for all 19 subgroups of $D_8$ the test with $r$ and $s$ alone gives the same answer as conjugating by all sixteen elements.
Subgroups that are not normal
How is a subgroup shown not to be normal? A single conjugate that leaves the subgroup is enough. A single conjugate that stays inside proves nothing.
In $D_8$. For $H = \langle r^4, s \rangle$, conjugation by $r$ sends a reflection $r^a s$ to $r^{a+2}s$, turning its mirror line through 45 degrees (Group Homomorphisms and Isomorphisms). So $r s r^{-1} = r^2 s$, which is not in $H$, and $rHr^{-1} = \{1, r^4, r^2 s, r^6 s\}$ is a different subgroup. This is the subgroup whose left and right cosets differ in the previous article's figure.
In $S_8$. Let $\sigma = (1\;5\;2\;8)(3\;6\;7)$, the running example of Symmetric Groups and Permutations, with the cyclic subgroup $\langle \sigma \rangle$ of order 12. Conjugating by a permutation renames the objects in the cycles. Conjugating by $(1\;2)$ gives $(1\;8\;2\;5)(3\;6\;7)$, which the script finds is $\sigma^7$, inside $\langle \sigma \rangle$.
Check your understanding. Does that conjugate show that $\langle \sigma \rangle$ is normal in $S_8$?
No. Normality needs every conjugate inside, and one conjugate inside settles nothing. Conjugating by $(1\;3)$ gives $(1\;6\;7)(2\;8\;3\;5)$, which sends 3 to 5. Every power of $\sigma$ sends 3 into the cycle $\{3, 6, 7\}$, so this conjugate lies outside, and $\langle \sigma \rangle$ is not normal.
On the cube. The 16 moves of the subgroup $\langle U, D \rangle$ turn only the top and bottom layers. The conjugate $R\,U\,R'$ of $U$ moves the two stickers of the front-right edge, which lies in the middle layer, so the script confirms that it is not in $\langle U, D \rangle$, and this subgroup is not normal in the cube group.
Multiplying cosets
Back to the question of the opening: for which subgroups does the rule $(aN)(bN) = abN$ make sense?
Theorem. Let $N$ be a subgroup of $G$. The rule $(aN)(bN) = abN$ is well defined, meaning that $aN = a'N$ and $bN = b'N$ always imply $abN = a'b'N$, if and only if $N$ is normal.
Here is the plan of the proof. For normal $N$, a change of representative adds a factor from $N$, which conjugation moves past $b$. Conversely, the two names $nN$ and $eN$ of one coset must give equal products, and that forces conjugates into $N$.
Proof. Suppose $N$ is normal, and let $aN = a'N$ and $bN = b'N$. Then $a' = an_1$ and $b' = bn_2$ for some $n_1, n_2 \in N$, and
$$a'b' = a\,n_1\,b\,n_2 = ab\,(b^{-1}n_1 b)\,n_2.$$
The element $b^{-1}n_1 b$ is a conjugate of $n_1$ (by $b^{-1}$), so it lies in $N$, and so does its product with $n_2$. Hence $a'b' \in abN$, and $a'b'N = abN$ by the coset test.
Conversely, suppose the rule is well defined, and let $g \in G$ and $n \in N$. Since $nN = N = eN$, the rule gives the same coset for the products $(nN)(g^{-1}N)$ and $(eN)(g^{-1}N)$, that is, $ng^{-1}N = g^{-1}N$. By the coset test, $(g^{-1})^{-1} n g^{-1} = gng^{-1}$ lies in $N$. So $N$ is normal. $\blacksquare$
The course text proves the first half as Theorem 8.12, in the language of congruence: for normal $N$, congruence modulo $N$ respects products (Theorem 8.10), just as congruence modulo $n$ respects products of integers. The second half explains why normal subgroups are the right ones to work modulo: no other subgroup allows the construction at all.
For a normal subgroup the coset $abN$ is also the product of the two cosets as sets: every product $xy$ with $x \in aN$ and $y \in bN$ lies in $abN$, by the computation in the proof, and every element $abn$ of $abN$ is such a product, $a \cdot bn$.
Check your understanding. In the right panel, take every $x$ in $H$ and every $y$ in $rH$. How many different products $xy$ are there, and which cosets do they fill?
Eight, filling the two cosets $rH$ and $r^3 H$, as the script confirms. The opening found one product in each: $1 \cdot r$ in $rH$ and $s \cdot r$ in $r^3 H$. For a normal subgroup the products of two cosets would fill exactly one coset.
The quotient group
When $N$ is normal, the cosets can be multiplied. Do they form a group?
Theorem. Let $N$ be a normal subgroup of $G$, and let $G/N$ be the set of left cosets of $N$. With the operation $(aN)(bN) = abN$, $G/N$ is a group. Its identity is $N = eN$, and the inverse of $aN$ is $a^{-1}N$. If $G$ is finite, then $\lvert G/N \rvert = [G : N] = \lvert G \rvert / \lvert N \rvert$. If $G$ is abelian, so is $G/N$.
The idea: each group axiom for cosets is the same axiom in $G$, read through representatives.
Proof. The operation is well defined by the previous theorem, and its values are cosets. Associativity is inherited from $G$:
$$\big((aN)(bN)\big)(cN) = (ab)c\,N = a(bc)\,N = (aN)\big((bN)(cN)\big).$$
$(eN)(aN) = aN = (aN)(eN)$, and $(aN)(a^{-1}N) = eN = (a^{-1}N)(aN)$. The number of cosets is the index, which is $\lvert G \rvert / \lvert N \rvert$ by Lagrange's theorem. If $ab = ba$ for all $a, b$, then $abN = baN$. $\blacksquare$
$G/N$ is the quotient group, or factor group, of $G$ by $N$, read "$G$ mod $N$". This is the course text's Theorem 8.13 and Judson's Theorem 10.1.4. Computing in $G/N$ means computing in $G$ and then ignoring factors that lie in $N$: two elements of $G$ become equal in $G/N$ when they differ by a factor from $N$. In particular a power $(aN)^k = a^k N$ is the identity of $G/N$ exactly when $a^k \in N$.
A quotient group is not a subgroup of $G$. Its elements are subsets of $G$, and its identity is the whole of $N$. In $D_8/\langle r \rangle$, the eight rotations together are one element. What connects $G/N$ back to $G$ is a homomorphism.
Theorem. If $N$ is a normal subgroup of $G$, the map $\pi: G \to G/N$, $\pi(a) = aN$, is a homomorphism of $G$ onto $G/N$ whose kernel is $N$.
The idea: the rule for multiplying cosets is exactly the statement that $\pi$ respects products.
Proof. $\pi(ab) = abN = (aN)(bN) = \pi(a)\pi(b)$, and every coset $aN$ is $\pi(a)$. The kernel consists of the $a$ with $aN = eN$, which by the coset test are the elements of $N$. $\blacksquare$
$\pi$ is the natural homomorphism (the course text's Theorem 8.18). With the observation that kernels are normal, it gives a characterization: the normal subgroups of $G$ are exactly the kernels of homomorphisms out of $G$. Every kernel is normal, and every normal subgroup is the kernel of its natural homomorphism. The first isomorphism theorem completes the picture by identifying the quotient by a kernel with the image of the homomorphism.
Examples of quotient groups
Which quotient groups has the course already met without the name?
The integers modulo $n$. In the additive group $\mathbb{Z}$, the subgroup $n\mathbb{Z}$ is normal, and its cosets $a + n\mathbb{Z}$ are the congruence classes $[a]$. The quotient operation $(a + n\mathbb{Z}) + (b + n\mathbb{Z}) = (a + b) + n\mathbb{Z}$ is the addition of $\mathbb{Z}_n$. So $\mathbb{Z}/n\mathbb{Z}$ is $\mathbb{Z}_n$, and the course's first construction of a new number system was the first quotient group. The natural homomorphism is reduction modulo $n$.
Even and odd permutations. $S_n/A_n$ has two elements, $A_n$ and the coset $(1\;2)A_n$ of odd permutations, each with $n!/2$ elements. The script checks $n = 4$ and $n = 5$. Its table is the familiar parity rule:
| even | odd | |
|---|---|---|
| even | even | odd |
| odd | odd | even |
Rotations and reflections. $D_8/\langle r \rangle$ has the two elements $\langle r \rangle$ and $s\langle r \rangle$ with the same table, rotations in the role of even and reflections in the role of odd, as in the opening. Judson (2026, Example 10.1.7) treats every $D_n$ this way. Both quotients have order 2 and so are copies of $\mathbb{Z}_2$, by the corollary on groups of prime order.
The parity of the Rubik's cube
A quarter turn of any face moves 20 stickers in five 4-cycles:
- the four stickers on the turned face that sit on corners;
- the four on edges;
- three 4-cycles of stickers on the sides of the layer.
The script checks this for all six quarter turns. A 4-cycle is odd, and five odd permutations multiply to an odd one (Symmetric Groups and Permutations), so each quarter turn is an odd permutation of the 48 stickers. Since the sign is a homomorphism, a move made of $k$ quarter turns, counting a half turn as two, permutes the stickers with sign $(-1)^k$.
The kernel $G^{+}$ of this sign, the moves that permute the stickers evenly, is a normal subgroup of index 2, of order
$$\frac{43{,}252{,}003{,}274{,}489{,}856{,}000}{2} = 21{,}626{,}001{,}637{,}244{,}928{,}000,$$
The quotient $G/G^{+}$ has two elements: the positions reachable with an even number of quarter turns and those reachable with an odd number. That makes a fact about every sequence of moves: no position can be reached both by an even number of quarter turns and by an odd number, because the parity of the count is the sign of the resulting permutation of the stickers, which depends only on the position. The commutator $R\,U\,R'\,U'$ (4 quarter turns) and the move $R\,U2\,D'\,B\,D'$ (6) lie in $G^{+}$; $R\,U2$ (3) does not.
Check your understanding. A cube is scrambled with 25 quarter turns. Could some sequence of exactly 30 quarter turns solve it?
No. A solving sequence is a move $Y$ with $XY = e$, where $X$ is the scramble, so $Y = X^{-1}$ has the sign of $X$, which is $(-1)^{25} = -1$. A sequence of 30 quarter turns has sign $+1$. Every solution of this scramble uses an odd number of quarter turns. The script confirms the sign $-1$ for one fixed scramble of 25 quarter turns and for its inverse.
The script also finds that each quarter turn permutes the eight corner cubies by a single 4-cycle, an odd permutation. Two homomorphisms that agree on a set of generators agree everywhere (Group Homomorphisms and Isomorphisms), so the sign of the sticker permutation equals the sign of the corner permutation for every move, and $G^{+}$ consists of the moves that permute the corners evenly. By the first invariant of the cube article, these are also the moves that permute the edges evenly.
Normal subgroups and ideals
The construction is the one that produced quotient rings, with one operation instead of two. In Rings the role of normality was played by absorption.
| Rings | Groups | |
|---|---|---|
| work modulo | an ideal $I$ | a normal subgroup $N$ |
| closure condition | $ra, ar \in I$ for $a \in I$, $r \in R$ | $gng^{-1} \in N$ for $n \in N$, $g \in G$ |
| classes | cosets $a + I$ | cosets $aN$ |
| same class when | $a - b \in I$ | $a^{-1}b \in N$ |
| quotient | $R/I$ | $G/N$ |
| natural map | $a \mapsto a + I$, kernel $I$ | $a \mapsto aN$, kernel $N$ |
| kernels are | ideals | normal subgroups |
Check your understanding. The course's work on rings never asked whether an ideal was "normal." Why not, and what did it ask instead?
The additive group of a ring is abelian, so every additive subgroup is normal, and the cosets $a + I$ can always be added. The absorption condition on an ideal is what makes their products well defined. In a group the single operation is usually not commutative, and conjugation takes over the part of absorption. The two proofs split the same way: well-definedness forces the condition, and the condition makes the quotient a structure of the same kind.
Simple groups
Can every group be divided by some normal subgroup? A group $G \neq \{e\}$ whose only normal subgroups are $\{e\}$ and $G$ is simple. It has no proper quotients other than the trivial one, so it cannot be broken into smaller pieces by this construction. Groups of prime order are simple, since by Lagrange's theorem they have no other subgroups at all. These are the only simple abelian groups (the course text, Theorem 8.25).
Nonabelian simple groups are rarer. The course text (§8.5) and Judson (2026, Theorem 10.2.4) prove that the alternating group $A_n$ is simple for every $n \ge 5$, so $A_5$, of order 60, cannot be divided by any normal subgroup, although it has many subgroups. Ruffini and Abel found that the general equation of degree 5 cannot be solved by a formula in radicals, and Galois explained which equations can be solved through their groups of permutations of the roots. In that theory the simplicity of $A_5$ is what makes $S_5$ fail the test (Judson, 2026, Chapter 23 and §23.3).
Where it leads
- The quotient by the kernel of a homomorphism is isomorphic to its image, which identifies quotients without listing cosets: The First Isomorphism Theorem for Groups, with the cube group's kernels and images worked out.
- The quotient rings of Ideals and Quotient Rings are, as additive groups, quotient groups of abelian groups, with a multiplication added.
What we have learned
Cosets can be multiplied through representatives exactly when the subgroup is normal, closed under conjugation. They then form the quotient group $G/N$, and the natural map $a \mapsto aN$ is a homomorphism with kernel $N$, so the normal subgroups are exactly the kernels. The rotations in $D_8$, $A_n$ in $S_n$ and the even moves of the cube are kernels of a sign.
Each of those quotients was recognized by hand, by matching its two cosets with $1$ and $-1$. Is there a general way to tell which group a quotient is? The next article, The First Isomorphism Theorem for Groups, gives one.
References
- Hungerford, T. W. (2014). Abstract Algebra: An Introduction (3rd ed.), §§8.2 to 8.5. Brooks/Cole, Cengage Learning. The course text: normal subgroups defined by $Na = aN$; congruence modulo a normal subgroup respects products (Theorem 8.10); equivalent conditions for normality (Theorem 8.11); the coset product is well defined (Theorem 8.12); the quotient group, its order, and quotients of abelian groups (Theorem 8.13); kernels are normal (Theorem 8.16); the natural homomorphism (Theorem 8.18); simple abelian groups (Theorem 8.25); the simplicity of $A_n$ for $n \ge 5$ (§8.5). record · 44 pages
- Judson, T. W. (2026). Abstract Algebra: Theory and Applications (annual edition, August 2026). Free online at judsonbooks.org. §10.1 (normal subgroups by $gN = Ng$; Theorem 10.1.3, equivalent conditions; Theorem 10.1.4, the factor group; Example 10.1.7, the rotations of $D_n$); §10.2 (Theorem 10.2.4, $A_n$ is simple for $n \ge 5$); Chapter 23 (the history of the quintic, from Ruffini and Abel to Galois) and §23.3 ($S_5$ is not solvable, by Theorem 10.2.4, so some quintics are not solvable by radicals). record · 44 pages