Semidirect Products

Picture $N$ as a set of "translations" and $H$ as a set of "linear symmetries" acting on the same space. A direct product forces the translations and symmetries to be indifferent to each other's order of application. A semidirect product instead says: to move a symmetry $h$ past a translation $n$, you must first transform $n$ by $h$'s action, then swap. This is exactly the rule you use every day when composing rigid motions of the plane: first rotate, and a subsequent translation must be measured in the rotated frame. The dihedral group $D_n$ (symmetries of a regular $n$-gon) is the paradigm case. Let $N = \langle r \rangle \cong \mathbb{Z}/n\mathbb{Z}$ be the rotation subgroup and let $H = \langle s\rangle \cong \mathbb{Z}/2\mathbb{Z}$ be generated by a single reflection. The relation $srs^{-1} = r^{-1}$ says exactly that conjugation by $s$ acts on $N$ as the automorphism $n \mapsto n^{-1}$. This is not the trivial action, so $D_n$ is not $\mathbb{Z}/n \times \mathbb{Z}/2$ in general — it is the semidirect product $\mathbb{Z}/n \rtimes_\varphi \mathbb{Z}/2$ with $\varphi$ sending the nontrivial element of $\mathbb{Z}/2$ to inversion.

Example 1: The dihedral group $D_{5}$ as a semidirect product, computed concretely. Let $N = \langle r \rangle \cong \mathbb{Z}/5\mathbb{Z}$ (rotations by multiples of $72°$ of a regular pentagon) and let $H = \langle s \rangle \cong \mathbb{Z}/2\mathbb{Z}$, where $s$ is one fixed reflection, inside $G = D_5$, the symmetry group of the regular pentagon, $|G|=10$. Checking the hypotheses of the recognition theorem. First, $N \trianglelefteq G$: the rotation subgroup has index $2$ in $D_5$ (order $5$ inside order $10$), and every subgroup of index $2$ is normal, since its unique nontrivial coset is automatically closed under conjugation (conjugating a coset representative outside $N$ by any $g\in G$ again lands outside $N$, as conjugation is a bijection on $G$ preserving cosets of a normal-index-2 candidate — concretely here, $srs^{-1}=r^{-1}\in N$ verifies it directly). Second, $N \cap H$: $N$ has order $5$, $H$ has order $2$, and $\gcd(5,2)=1$, so by Lagrange's theorem any common subgroup has order dividing both $5$ and $2$, forcing $N\cap H = \{e\}$. Third, $NH = G$: $|NH| = |N||H|/|N\cap H| = 5\cdot 2/1 = 10 = |G|$, so $NH=G$ since $NH \subseteq G$ and both have size $10$. Computing $\varphi$. Conjugation gives $\varphi_s(r) = srs^{-1}$. In $D_5$ the defining relation is $srs^{-1} = r^{-1}$ (a reflection reverses the sense of rotation). So $\varphi_s$ is the automorphism $r^k \mapsto r^{-k}$ of $\mathbb{Z}/5\mathbb{Z}$, i.e. under the identification $N \cong \mathbb{Z}/5\mathbb{Z}$, $\varphi_s$ is multiplication by $-1 \equiv 4 \pmod 5$. Building the isomorphism explicitly. By the theorem, $\Psi: \mathbb{Z}/5\mathbb{Z} \rtimes_\varphi \mathbb{Z}/2\mathbb{Z} \to D_5$, $\Psi(k,\epsilon) = r^k s^\epsilon$, is an isomorphism. Let's verify one nontrivial product both ways: take $(k,\epsilon)=(2,1)$ and $(k',\epsilon') = (3,1)$. In the semidirect product,

since $-3 \equiv 2 \pmod 5$. So $\Psi((2,1)(3,1)) = \Psi(4,0) = r^4$. Directly in $D_5$: $\Psi(2,1)\Psi(3,1) = (r^2 s)(r^3 s) = r^2 (sr^3) s = r^2 (r^{-3}s)s = r^{2-3}s^2 = r^{-1} = r^4$, using $sr^3 = r^{-3}s$ (from $srs^{-1}=r^{-1}$ applied three times) and $s^2=e$. Both computations give $r^4$, matching perfectly, and $4 \equiv -1 \pmod 5$ is exactly consistent.

Example 2: A semidirect product with a non-cyclic acting group, $S_3 \cong \mathbb{Z}/3 \rtimes \mathbb{Z}/2$ realized inside $\operatorname{Hol}$ data, then a genuinely asymmetric case: $(\mathbb{Z}/7\mathbb{Z}) \rtimes (\mathbb{Z}/3\mathbb{Z})$. To keep the numbers non-trivial and asymmetric (not the familiar order-6 or order-8 cases), build $G$ as a semidirect product of $N = \mathbb{Z}/7\mathbb{Z}$ by $H = \mathbb{Z}/3\mathbb{Z}$, using the fact that $\operatorname{Aut}(\mathbb{Z}/7\mathbb{Z}) \cong (\mathbb{Z}/7\mathbb{Z})^\times \cong \mathbb{Z}/6\mathbb{Z}$, which has a unique subgroup of order $3$. Finding the automorphism of order 3. We need an element of order dividing $3$ in $(\mathbb{Z}/7\mathbb{Z})^\times = \{1,2,3,4,5,6\}$ under multiplication mod $7$. Compute powers of $2$: $2^1=2,\ 2^2=4,\ 2^3=8\equiv 1$. So $2$ has order $3$ in $(\mathbb{Z}/7\mathbb{Z})^\times$. Define $\varphi: \mathbb{Z}/3\mathbb{Z} \to \operatorname{Aut}(\mathbb{Z}/7\mathbb{Z})$ by sending the generator $1 \in \mathbb{Z}/3\mathbb{Z}$ to the automorphism $\alpha(x) = 2x \bmod 7$. Since $\alpha^3(x) = 2^3 x = 8x \equiv x \pmod 7$, $\alpha$ has order exactly $3$ (it is not the identity since $\alpha(1)=2\ne 1$, and $\alpha^2(1) = 4 \ne 1$), so this extends to a well-defined homomorphism $\varphi$ on all of $\mathbb{Z}/3\mathbb{Z}$ by $\varphi(k) = \alpha^k$. The resulting group. $G := \mathbb{Z}/7\mathbb{Z}\rtimes_\varphi \mathbb{Z}/3\mathbb{Z}$ has order $7\cdot 3 = 21$. Its multiplication is

A concrete non-commutativity check. Take $(1,1)$ and $(1,0)$ (writing $N$-part additively mod 7, $H$-part additively mod 3). Then

These differ ($(3,1)\ne(2,1)$), so $G$ is non-abelian, as expected since $\varphi$ is non-trivial (Corollary part (4) shows conjugation genuinely reproduces $\varphi$, so a non-trivial $\varphi$ forces non-commuting elements between the $N$- and $H$-parts). Order check via Corollary. $|G| = |N||H| = 21$ matches $|N \rtimes_\varphi H| = |N\times H| = 7\cdot 3$ directly from the underlying set, confirming this is a group of order $21$ that is not $\mathbb{Z}/21\mathbb{Z}$ (which is abelian) — it is the unique non-abelian group of order $21$, a standard example arising later from Sylow analysis of groups of order $21$.

The semidirect product, as an explicit named construction unifying the direct product and phenomena like the dihedral groups, crystallized during the structural development of group theory in the late nineteenth and early twentieth centuries, alongside Otto Hölder's work on group extensions and Otto Schmidt's later systematic extension theory. The underlying idea — that a normal subgroup together with a complementary subgroup and a conjugation action reconstruct the whole group — traces to the general program of understanding groups via their normal subgroups, a program driven forward by Camille Jordan and Otto Hölder in their study of composition series, and formalized in the extension-theoretic language associated with Otto Schreier and Heinz Hopf in the 1920s.

Semidirect products are the standard mechanism for building and recognizing symmetry groups throughout mathematics and its applications. The Euclidean group of rigid motions of $\mathbb{R}^n$ is the semidirect product of the translation group $\mathbb{R}^n$ by the orthogonal group $O(n)$, with $O(n)$ acting on translations by its defining linear action; this is the group used throughout classical mechanics, robotics, and computer graphics to describe rigid-body motion. The affine group of a vector space, $\operatorname{Aff}(V) = V \rtimes GL(V)$, is used identically in projective geometry and in computer vision for camera transformations. In Classification via Sylow Analysis, semidirect products are exactly the tool used to build every group whose Sylow subgroups admit a normal complement: once a group of order $p^aq^b$ (or more generally with a normal Sylow subgroup $N$) is shown to have a complementary subgroup $H$, the recognition theorem here says the whole group is determined by $N$, $H$, and the conjugation action, reducing the classification problem to classifying homomorphisms $H \to \operatorname{Aut}(N)$ up to appropriate equivalence. In cryptography, wreath products (iterated semidirect products) underlie certain hard combinatorial group-theoretic problems proposed for group-based cryptosystems. In physics, semidirect products describe the Poincaré group (Lorentz transformations acting on spacetime translations) and the structure of crystallographic space groups, where a point group acts on a lattice of translations exactly as $H$ acts on $N$ here.

Ask the tutor: How does the classification of homomorphisms $\varphi: H \to \operatorname{Aut}(N)$ up to equivalence let us count all semidirect products of two given groups?