Quotient Groups

Quotient groups are the mechanism behind modular arithmetic, where $\mathbb{Z}/n\mathbb{Z}$ underlies clock arithmetic, check-digit schemes, and the RSA and Diffie–Hellman cryptographic protocols, all of which compute inside quotient groups of $\mathbb{Z}$ or of multiplicative groups modulo $n$. In topology, the fundamental group of a quotient space and covering space theory use quotient groups of $\pi_1$ to classify covers. In physics, gauge symmetry breaking and the study of order parameters in condensed matter (e.g., the residual symmetry group after a phase transition) are literally computations of $G/N$ for a symmetry group $G$ and an unbroken subgroup $N$. In coding theory, cosets of a linear code (itself a subgroup of a vector space under addition) are used in syndrome decoding, an application of the coset-partition idea underlying quotient groups. In computational group theory and computer algebra systems, quotient groups are computed explicitly via coset enumeration algorithms (e.g. Todd–Coxeter) to determine finite presentations of groups.

Ask the tutor: How does the First Isomorphism Theorem generalize this construction to arbitrary homomorphisms $\varphi: G \to H$?

Think of $G$ as a set of objects and $N$ as a "unit" you are declaring negligible: two elements $a, b \in G$ get identified in the quotient exactly when $a^{-1}b \in N$, i.e., when they lie in the same coset of $N$. The quotient group is what remains after this identification, provided the identification respects multiplication. Here is the danger. If you pick representatives $a \in aN$ and $b \in bN$ and multiply them, you get $ab$, which lands in some coset. But there are many other valid representatives $a' = an_1$ and $b' = bn_2$ (with $n_1, n_2 \in N$) of the same two cosets, and $a'b' = a n_1 b n_2$. For the operation to be well-defined, this must lie in the same coset $(ab)N$ no matter which representatives were chosen. Expanding,

so this lands in $(ab)N$ precisely when $b^{-1} n_1 b \in N$ for every $n_1 \in N$ and every $b \in G$ — that is, precisely when $N$ is normal. Normality is not an extra convenience; it is the condition that makes "multiply the cosets" a meaningful sentence. Once well-defined, $G/N$ behaves like $G$ with all of $N$'s internal detail blurred into a single point (the identity coset). The picture below shows $G = \mathbb{Z}$ partitioned into cosets of $N = 4\mathbb{Z}$: four parallel "tracks" of integers, each track becoming one element of $\mathbb{Z}/4\mathbb{Z}$.

Example 1: $\mathbb{Z}/6\mathbb{Z}$, done with explicit non-symmetric elements. Let $G = \mathbb{Z}$ under addition and $N = 6\mathbb{Z} = \{\dots, -6, 0, 6, 12, \dots\}$. Since $\mathbb{Z}$ is abelian, every subgroup is normal (as $gng^{-1} = n$ for all $g, n$), so $N \trianglelefteq G$ automatically, and by the theorem $G/N$ is a group. The cosets are $0+N, 1+N, 2+N, 3+N, 4+N, 5+N$, six in total, matching $|G/N| = [G:N] = 6$ since $|N|$ is infinite but the index computation via representatives $\{0,1,2,3,4,5\}$ directly enumerates them (Lagrange's numeric form does not apply here since $G$ is infinite, but the coset-counting argument in Part 4 still identifies $|G/N|$ with the number of residue classes, which is $6$). Let's compute a genuinely asymmetric product: take $a = 37$ and $b = -14$. We want $(37+N) + (-14+N)$. First reduce representatives to standard residues mod 6: $37 = 6\cdot 6 + 1$, so $37 + N = 1+N$. And $-14 = 6\cdot(-3) + 4$, so $-14 + N = 4+N$ (check: $-14 - 4 = -18 = 6\cdot(-3) \in N$, confirming $-14+N = 4+N$). By the well-definedness guaranteed by the theorem, we may compute using any representatives, so use the originals:

Reduce: $23 = 6\cdot 3 + 5$, so $23+N = 5+N$. Alternatively, using the reduced representatives directly: $(1+N)+(4+N) = 5+N$. Both routes agree — this is the well-definedness theorem in action, verified numerically. The projection $\pi: \mathbb{Z} \to \mathbb{Z}/6\mathbb{Z}$ sends $37 \mapsto 1+N$ and $-14 \mapsto 4+N$, and indeed $\pi(37) + \pi(-14) = \pi(37 + (-14)) = \pi(23) = 5+N$, confirming $\pi$ is a homomorphism on these specific inputs.

Example 2: a non-abelian quotient, $S_4 / V_4$, with an explicit coset multiplication. Let $G = S_4$, the symmetric group on $\{1,2,3,4\}$, with $|G| = 24$. Let

the Klein four-subgroup consisting of the identity and the three double transpositions. It is a standard fact (verifiable directly by conjugating each element of $V_4$ by every element of $S_4$) that $V_4 \trianglelefteq S_4$: conjugation permutes the three double transpositions among themselves and fixes the identity, since conjugating a permutation by any $\sigma \in S_4$ preserves its cycle type, and the only elements of $S_4$ of cycle type $2+2$ are exactly the three nonidentity elements of $V_4$. Since $|N| = 4$ and $|G| = 24$, the theorem gives $|G/N| = 24/4 = 6$. In fact one can show $G/N \cong S_3$ (this can be checked by exhibiting an explicit isomorphism, though we do not need it here); for now we simply verify coset arithmetic directly, which is the point of this article. Take $a = (1234)$ and $b = (123)$, two specific, non-commuting, asymmetric elements of $S_4$. We compute the product of the cosets $aN$ and $bN$ two ways, to confirm well-definedness concretely. Direct way: $ab = (1234)(123)$. Applying right-to-left (apply $(123)$ first, then $(1234)$): $1 \mapsto 2 \mapsto 3$, $2\mapsto 3\mapsto 4$, $3 \mapsto 1 \mapsto 2$, $4 \mapsto 4 \mapsto 1$. So $ab = (1 3 2 4)$ in cycle notation (check: $1\to3, 3\to2, 2\to4, 4\to1$ ✓.) So $(aN)(bN) = (1324)N$. Using different representatives: Replace $a$ by $a' = a \cdot (12)(34) = (1234)(12)(34)$. Compute $a'$: apply $(12)(34)$ first then $(1234)$: $1\mapsto2\mapsto3$, $2\mapsto1\mapsto2$... wait, recompute carefully: $(12)(34)$ sends $1\to2, 2\to1, 3\to4, 4\to3$. Then $(1234)$ sends $1\to2,2\to3,3\to4,4\to1$. Composite (apply $(12)(34)$ then $(1234)$): $1\to2\to3$, $2\to1\to2$, $3\to4\to1$, $4\to3\to4$. So $a'$ sends $1\to3, 2\to2, 3\to1, 4\to4$, i.e. $a' = (13)$. Now replace $b$ by $b' = b \cdot (13)(24) = (123)(13)(24)$. Compute: $(13)(24)$ sends $1\to3,3\to1,2\to4,4\to2$. Then $(123)$ sends $1\to2,2\to3,3\to1$ (fixes 4). Composite (apply $(13)(24)$ then $(123)$): $1\to3\to1$, $2\to4\to4$, $3\to1\to2$, $4\to2\to3$. So $b'$ sends $1\to1, 2\to4,4\to3,3\to2$, i.e. $b' = (243)$. Check $a'N = aN$: $a'a^{-1}$ should lie in $N$; indeed by construction $a' = a n_1$ with $n_1=(12)(34)\in N$, so automatically $a'N=aN$. Similarly $b'N = bN$ since $b' = b n_2$ with $n_2 = (13)(24) \in N$. Now compute $a'b' = (13)(243)$. Apply $(243)$ first: $2\to4,4\to3,3\to2$ (fixes 1); then $(13)$: $1\to3,3\to1$ (fixes 2,4). Composite: $1\to1\to3$, $2\to4\to4$, $3\to2\to2$, $4\to3\to1$. So $a'b'$ sends $1\to3,\,2\to4,\,3\to2,\,4\to1$, i.e. $a'b' = (1 3 2 4)$. This is exactly $ab = (1324)$ computed above — same permutation exactly, not merely the same coset. This concretely confirms $(a'b')N = (ab)N = (1324)N$, exhibiting the well-definedness (Part 1 of the theorem) on a genuinely non-abelian, asymmetric example.