---
title: Semidirect Products
module: Products and Group Structure
moduleNumber: 5
lessonNumber: 2
order: 502
summary: >
  The semidirect product relaxes the direct product by requiring only one factor
  to be normal, with the other acting on it through a homomorphism into its
  automorphism group. This single twisting map lets abelian pieces assemble into
  non-abelian groups, realizes the dihedral groups as $\mathbb{Z}_n \rtimes
  \mathbb{Z}_2$, and, with a recognition theorem, classifies groups of several
  small orders.
topics: [Products and Group Structure]
sources:
  - book: Dummit & Foote
    ref: "Ch. 5 — Direct and Semidirect Products; §5.5 Semidirect Products, §5.3 Table of Groups of Small Order"
draft: false
---

The [direct
product](/abstract-algebra/products-and-group-structure/direct-products-and-finite-abelian-groups)
needs both factors normal, which forces them to commute and keeps the result
abelian whenever the factors are. Dropping normality of one factor removes that
constraint: the remaining normal factor is acted on by the other, with the
action recorded by a homomorphism into its automorphism group. The result can be
non-abelian even when the factors are abelian.

## Motivation from an existing product

Suppose a group $G$ already contains subgroups $H$ and $K$ with $H
\trianglelefteq G$ (but $K$ not necessarily normal) and $H \cap K = 1$. Then
$HK$ is a subgroup, and every element of $HK$ is written uniquely as $hk$. To
multiply two such elements we slide the middle factors past each other using
normality of $H$:

$$
(h_1 k_1)(h_2 k_2) = h_1 (k_1 h_2 k_1^{-1}) k_1 k_2 = h_3 k_3,
\qquad
h_3 = h_1 (k_1 h_2 k_1^{-1}), \quad k_3 = k_1 k_2.
$$

Because $H \trianglelefteq G$, the conjugate $k_1 h_2 k_1^{-1}$ lands back in
$H$, so $h_3 \in H$ and $k_3 \in K$. The whole multiplication depends only on
the operations in $H$ and $K$ and on how $K$ conjugates $H$. Writing the
conjugation action as $k \cdot h = k h k^{-1}$, the rule becomes

$$
(h_1 k_1)(h_2 k_2) = \bigl(h_1\,(k_1 \cdot h_2)\bigr)(k_1 k_2).
$$

Conjugation gives a homomorphism $\varphi : K \to \operatorname{Aut}(H)$, and
this formula shows the product is determined intrinsically by $H$, $K$, and
$\varphi$.[^df-motiv] Nothing in it requires that $G$ existed first — the three
ingredients alone define a group.

## The construction

> **Definition (Semidirect product).** Let $H$ and $K$ be groups and let
> $\varphi : K \to \operatorname{Aut}(H)$ be a homomorphism, with action written
> $k \cdot h = \varphi(k)(h)$. The **semidirect product** $H \rtimes_\varphi K$
> is the set of pairs $(h,k)$ under
> $$
> (h_1, k_1)(h_2, k_2) = \bigl(h_1\,(k_1 \cdot h_2),\; k_1 k_2\bigr).
> $$

The verification that this is a group uses only that $\cdot$ is a genuine action
of $K$ on $H$.[^df-thm10] Its structure mirrors the motivating picture exactly.

> **Theorem (Properties of $H \rtimes_\varphi K$).** In $G = H \rtimes_\varphi K$:
>
> - $|G| = |H|\,|K|$, and the maps $h \mapsto (h,1)$, $k \mapsto (1,k)$ embed
>   $H$ and $K$ as subgroups.
> - Under these identifications $H \trianglelefteq G$ and $H \cap K = 1$.
> - For all $h \in H$, $k \in K$, conjugation recovers the action:
>   $k h k^{-1} = \varphi(k)(h)$.

The notation $\rtimes$ points its open side toward the normal factor $H$. The
construction is not symmetric in $H$ and $K$: swapping them generally changes the
group, unlike the direct product.

$$
% caption: A semidirect product $H \rtimes_\varphi K$: the normal factor $H$ is
% the kernel of projection onto $K$, and $K$ acts on $H$ by conjugation through
% $\varphi$. The action twists the multiplication of the two abelian slabs.
\begin{tikzpicture}[scale=1.0, font=\small]
\definecolor{acc}{HTML}{4A6FA5}
% H slab (normal)
\draw[black, fill=acc!12] (0,0) rectangle (3.2,1.1);
\node[black] at (1.6,0.55) {$H$ normal};
% K slab
\draw[black, fill=acc!5] (0,1.5) rectangle (3.2,2.6);
\node[black] at (1.6,2.05) {$K$ complement};
% twisting action arrow from K onto H
\draw[acc, thick, ->] (3.55,2.05) .. controls (5.0,2.05) and (5.0,0.55) .. (3.55,0.55);
\node[acc, anchor=west] at (4.85,1.48) {acts by};
\node[acc, anchor=west] at (4.85,1.12) {conjugation};
% phi label
\node[black, anchor=west] at (0,3.0) {a homomorphism records the twist};
\end{tikzpicture}
$$

## Recovering the direct product

The direct product is the special case of trivial action: $H \rtimes_\varphi K$
collapses to $H \times K$ exactly when the action is trivial.

> **Proposition (Direct-product criterion).** For $\varphi : K \to
> \operatorname{Aut}(H)$ the following are equivalent:
>
> - $H \rtimes_\varphi K = H \times K$ (the identity map is an isomorphism);
> - $\varphi$ is the trivial homomorphism;
> - $K \trianglelefteq H \rtimes_\varphi K$.

A trivial action means $k \cdot h = h$ for all $k, h$, so the twisted product
formula degenerates to the componentwise one.[^df-prop11] Normality of $K$ is
equivalent: if $K$ is also normal then $[h,k] \in H \cap K = 1$, so $H$ and $K$
commute and the action is trivial. The direct product is the semidirect product
with no twist; every non-abelian example needs $\varphi \neq 1$.

| | Direct product $H \times K$ | Semidirect product $H \rtimes_\varphi K$ |
| --- | --- | --- |
| Normal factors | both $H$ and $K$ | $H$ only |
| Action $\varphi : K \to \operatorname{Aut}(H)$ | trivial | any homomorphism |
| $H$–$K$ elements | commute | $khk^{-1} = \varphi(k)(h)$ |
| Abelian if factors are | yes | not in general |
| Recovers | itself | $H \times K$ when $\varphi = 1$ |

## Building groups from a twist

The construction produces familiar groups and new ones from small abelian
inputs. In each example $K$ is cyclic, so $\varphi$ is fixed by naming a single
automorphism of $H$.

- **Dihedral groups.** Let $H$ be any abelian group and $K = \langle x\rangle
  \cong \mathbb{Z}_2$, with $x$ acting by inversion, $x \cdot h = h^{-1}$. Then
  $H \rtimes_\varphi K$ has $H$ as an index-$2$ subgroup inverted by $x$. When
  $H = \mathbb{Z}_n$ this is the dihedral group $D_{2n}$; when $H = \mathbb{Z}$
  it is the infinite dihedral group $D_\infty$.[^df-dihedral]
- **Non-abelian groups of order $12$.** Taking $H = \mathbb{Z}_3$ and $K =
  \mathbb{Z}_4$ with $x$ inverting $H$ (so $x^2$ centralizes it) gives a
  non-abelian group of order $12$ that is neither $A_4$ nor $D_{12}$, since its
  Sylow $2$-subgroup is cyclic of order $4$. This is the group $\mathbb{Z}_3
  \rtimes \mathbb{Z}_4$.
- **The holomorph.** For any $H$, taking $K = \operatorname{Aut}(H)$ with
  $\varphi$ the identity gives the **holomorph** $\operatorname{Hol}(H) = H
  \rtimes \operatorname{Aut}(H)$, the largest group in which $H$ is normal and
  $\operatorname{Aut}(H)$ acts as written. For example
  $\operatorname{Hol}(\mathbb{Z}_2 \times \mathbb{Z}_2) \cong S_4$.

In the dihedral realization, $D_{2n}$ is rotations $\mathbb{Z}_n$ acted on by a
reflection of order $2$ that inverts them.

$$
% caption: $D_{2n} \cong \mathbb{Z}_n \rtimes \mathbb{Z}_2$: the rotation
% subgroup $\langle r\rangle \cong \mathbb{Z}_n$ is normal; the reflection $s$
% has order $2$ and conjugates each rotation to its inverse, $s r s^{-1} = r^{-1}$.
\begin{tikzpicture}[scale=1.0, font=\small]
\definecolor{acc}{HTML}{4A6FA5}
% rotation ring
\draw[black] (0,0) circle (1.35);
\foreach \a/\lab in {90/{$1$}, 30/{$r$}, -30/{$r^2$}, -90/{$r^3$}, -150/{$r^4$}, 150/{$r^5$}} {
  \fill[acc] (\a:1.35) circle (2pt);
  \node[acc, font=\scriptsize] at (\a:1.68) {\lab};
}
\node[black] at (0,0) {rotations $\mathbb{Z}_n$};
% reflection s acting
\node[black, anchor=west] at (2.3,0.55) {$s$ has order $2$};
\node[black, anchor=west] at (2.3,0.1) {inverts every $r^k$};
\draw[acc, thick, ->] (2.25,0.3) .. controls (1.9,0.15) .. (30:1.35);
\end{tikzpicture}
$$

> **Worked example.** The non-abelian group of order $21$. Take $H = \mathbb{Z}_7
> = \langle a\rangle$ and $K = \mathbb{Z}_3 = \langle b\rangle$. The automorphism
> group $\operatorname{Aut}(\mathbb{Z}_7) \cong (\mathbb{Z}/7\mathbb{Z})^\times$
> is cyclic of order $6$, so it has a unique subgroup of order $3$, generated by
> $a \mapsto a^2$ (since $2^3 = 8 \equiv 1 \pmod 7$ and $2$ has order $3$ modulo
> $7$). Sending $b$ to this automorphism, $\varphi(b)(a) = a^2$, gives
> $$
> \mathbb{Z}_7 \rtimes_\varphi \mathbb{Z}_3 = \langle a, b \mid a^7 = b^3 = 1,\;
> b a b^{-1} = a^2\rangle,
> $$
> of order $21$. It is non-abelian, since $ba = a^2 b \neq ab$, and it is the
> smallest non-abelian group of odd order. The other order-$3$ automorphism $a
> \mapsto a^4$ gives an isomorphic group (replace $b$ by $b^2$), so there is
> exactly one non-abelian group of order $21$.

> **Worked example.** $\mathbb{Z}_3 \rtimes \mathbb{Z}_4$, non-abelian of order
> $12$. Take $H = \mathbb{Z}_3 = \langle a\rangle$ and $K = \mathbb{Z}_4 =
> \langle b\rangle$. Here $\operatorname{Aut}(\mathbb{Z}_3) \cong \mathbb{Z}_2$,
> generated by inversion $a \mapsto a^{-1}$. The only nontrivial homomorphism
> $\varphi : \mathbb{Z}_4 \to \mathbb{Z}_2$ sends $b$ to inversion, so
> $\varphi(b^2)$ is the identity and $b^2$ centralizes $a$:
> $$
> b a b^{-1} = a^{-1}, \qquad b^2 a b^{-2} = a.
> $$
> The result $\langle a, b \mid a^3 = b^4 = 1,\; b a b^{-1} = a^{-1}\rangle$ has
> order $12$. Its Sylow $2$-subgroup $\langle b\rangle$ is cyclic of order $4$,
> which separates it from $A_4$ (no element of order $4$) and from $D_{12}$
> (Sylow $2$-subgroup the Klein four-group).

## Split extensions and complements

The semidirect product is the content of a **split** short exact sequence. A
short exact sequence

$$
1 \longrightarrow H \longrightarrow G \longrightarrow K \longrightarrow 1
$$

records that $H$ is (a copy of) a normal subgroup of $G$ with quotient $K$. It
**splits** when there is a section $K \to G$, a homomorphism whose image is a
subgroup $\overline{K} \leq G$ meeting $H$ trivially and mapping isomorphically
to $K$. Such a $\overline{K}$ is a **complement** to $H$.

$$
% caption: A split short exact sequence $1 \to H \to G \to K \to 1$: the section
% $s$ (dashed) lifts $K$ back into $G$ as a complement, and $G = H \rtimes K$.
% Without a section the extension may be non-split.
\begin{tikzpicture}[scale=1.0, font=\small]
\definecolor{acc}{HTML}{4A6FA5}
\node (one)  at (0,0)   {$1$};
\node (H)    at (1.6,0) {$H$};
\node (G)    at (3.4,0) {$G$};
\node (K)    at (5.2,0) {$K$};
\node (one2) at (6.8,0) {$1$};
\draw[black, ->] (one) -- (H);
\draw[black, ->] (H) -- (G) node[midway, above, black, font=\scriptsize] {incl};
\draw[black, ->] (G) -- (K) node[midway, above, black, font=\scriptsize] {quot};
\draw[black, ->] (K) -- (one2);
% section
\draw[acc, thick, dashed, ->] (5.05,-0.28) .. controls (4.2,-0.95) and (4.2,-0.95) .. (3.55,-0.28);
\node[acc, font=\scriptsize] at (4.3,-1.1) {section $s$};
\end{tikzpicture}
$$

The recognition theorem says a complement is all that a semidirect
decomposition needs.

> **Theorem (Recognition of semidirect products).** Suppose $G$ has subgroups
> $H$ and $K$ with $H \trianglelefteq G$ and $H \cap K = 1$. Let $\varphi : K
> \to \operatorname{Aut}(H)$ send $k$ to conjugation by $k$ on $H$. Then $HK
> \cong H \rtimes_\varphi K$. In particular, if $G = HK$ then $G$ is the
> semidirect product of $H$ and $K$.

So a group factors as a semidirect product exactly when some proper normal
subgroup has a complement.[^df-thm12] Not every group qualifies. The quaternion
group $Q_8$ has no complement to any proper normal subgroup, so it is not a
semidirect product of proper subgroups — though it is a quotient of one. Simple
groups likewise admit no such decomposition.

## Classifying groups of small order

The recognition theorem drives a classification strategy. For a target order
$n$, the procedure is fixed.[^df-strategy]

```algorithm
caption: Classifying groups of order $n$ by semidirect products
find subgroups $H, K$ of every $G$ with $H \trianglelefteq G$, $H \cap K = 1$, $G = HK$
enumerate all isomorphism types for $H$ and for $K$
for each pair $(H, K)$ do
  list every homomorphism $\varphi : K \to \operatorname{Aut}(H)$
  form $H \rtimes_\varphi K$ for each $\varphi$
identify which of the resulting semidirect products are isomorphic
return the distinct isomorphism types
```

Sylow's theorem supplies the normal factor $H$ and its complement $K$; when
$|H|$ and $|K|$ are coprime, $H \cap K = 1$ is automatic by Lagrange. The count
of homomorphisms $\varphi$ is small, especially after accounting for the freedom
to reselect a generator of a cyclic $K$.

### Groups of order $pq$

Let $p < q$ be primes and $|G| = pq$. Sylow forces the Sylow $q$-subgroup $Q
\cong \mathbb{Z}_q$ to be normal, with complement $P \cong \mathbb{Z}_p$, so
$G \cong Q \rtimes_\varphi P$ for some $\varphi : P \to \operatorname{Aut}(Q)$.
Now $\operatorname{Aut}(Q)$ is cyclic of order $q-1$.

- If $p \nmid q-1$, the only homomorphism $\varphi$ is trivial, so $G \cong
  \mathbb{Z}_{pq}$ is cyclic — the unique group of order $pq$.
- If $p \mid q-1$, there is a nontrivial $\varphi$, and all nontrivial choices
  give isomorphic groups (they differ only by which generator of $P$ maps to a
  fixed automorphism). This yields one non-abelian group of order $pq$. When
  $p = 2$ it is $D_{2q}$.

So there are one or two groups of order $pq$, and the divisibility $p \mid q-1$
decides which. This is the smallest case where a number-theoretic condition on
the order controls the group count.

### Groups of order $30$

Order $30$ has a non-prime normal factor. Every group of order
$30 = 2 \cdot 3 \cdot 5$ has a normal subgroup $H$ of order $15$;
since $3 \nmid 5-1$, that $H$ is cyclic, $H \cong \mathbb{Z}_{15}$, by the
order-$pq$ case. A Sylow $2$-subgroup $K \cong \mathbb{Z}_2$ complements it, so
$G \cong \mathbb{Z}_{15} \rtimes_\varphi \mathbb{Z}_2$ for some $\varphi :
\mathbb{Z}_2 \to \operatorname{Aut}(\mathbb{Z}_{15})$.[^df-order30] Now

$$
\operatorname{Aut}(\mathbb{Z}_{15}) \cong (\mathbb{Z}/15\mathbb{Z})^\times \cong
\mathbb{Z}_4 \times \mathbb{Z}_2,
$$

which has exactly three elements of order $2$. Writing $H = \langle a\rangle
\times \langle b\rangle \cong \mathbb{Z}_5 \times \mathbb{Z}_3$, the three
involutions invert $a$ only, invert $b$ only, or invert both. Each nontrivial
$\varphi$ sends the generator of $K$ to one of them, giving three non-abelian
groups, plus the trivial action giving $\mathbb{Z}_{30}$:

| Action of $K$ on $H$ | Group | Center |
| --- | --- | --- |
| trivial | $\mathbb{Z}_{30}$ | order $30$ |
| invert $b$ (order $3$) only | $\mathbb{Z}_5 \times D_6 \cong \mathbb{Z}_5 \times S_3$ | order $5$ |
| invert $a$ (order $5$) only | $\mathbb{Z}_3 \times D_{10}$ | order $3$ |
| invert both | $D_{30}$ | order $1$ |

There are exactly four groups of order $30$, distinguished by their centers. The method builds nothing here that direct products could not, but it
proves the list is complete — which direct products alone cannot.

### Groups of order $12$

Order $12 = 2^2 \cdot 3$ runs the method in full. Either the Sylow $2$-subgroup
$V$ (isomorphic to $\mathbb{Z}_4$ or $\mathbb{Z}_2 \times \mathbb{Z}_2$) or the
Sylow $3$-subgroup $T \cong \mathbb{Z}_3$ is normal.[^df-order12] Working through
both cases:

| Normal factor | Complement | Action $\varphi$ | Group |
| --- | --- | --- | --- |
| $V = \mathbb{Z}_2 \times \mathbb{Z}_2$ | $\mathbb{Z}_3$ | trivial | $\mathbb{Z}_2 \times \mathbb{Z}_2 \times \mathbb{Z}_3 \cong \mathbb{Z}_6 \times \mathbb{Z}_2$ |
| $V = \mathbb{Z}_2 \times \mathbb{Z}_2$ | $\mathbb{Z}_3$ | order-$3$ cyclic | $A_4$ |
| $T = \mathbb{Z}_3$ | $\mathbb{Z}_4$ | trivial | $\mathbb{Z}_{12}$ |
| $T = \mathbb{Z}_3$ | $\mathbb{Z}_4$ | inversion | $\mathbb{Z}_3 \rtimes \mathbb{Z}_4$ |
| $T = \mathbb{Z}_3$ | $\mathbb{Z}_2 \times \mathbb{Z}_2$ | inversion | $D_{12} \cong S_3 \times \mathbb{Z}_2$ |

Deduplicating, there are exactly five groups of order $12$, three of them
non-abelian: $A_4$, $D_{12}$, and $\mathbb{Z}_3 \rtimes \mathbb{Z}_4$. The
alternating group $A_4$ is the case where a Klein four normal subgroup is
cyclically permuted by $\mathbb{Z}_3$ — a decomposition invisible without the
semidirect product.

## Direct versus semidirect on the same factors

The same $H$ and $K$ can produce different groups depending on the twist. Fixing
$H = \mathbb{Z}_3$ and $K = \mathbb{Z}_2$: the trivial action gives the abelian
$\mathbb{Z}_6$, while the inversion action gives the non-abelian $S_3 \cong
D_6$. Same underlying set of pairs, same factor orders, incompatible group
structures.

$$
% caption: One pair of factors, two products. With trivial action $\mathbb{Z}_3
% \times \mathbb{Z}_2 \cong \mathbb{Z}_6$ is cyclic; with inversion action
% $\mathbb{Z}_3 \rtimes \mathbb{Z}_2 \cong S_3$ is non-abelian. The twist alone
% separates them.
\begin{tikzpicture}[scale=1.0, font=\small,
  bx/.style={draw, black, inner sep=4pt}]
\definecolor{acc}{HTML}{4A6FA5}
\node[bx, fill=acc!5] (L) at (0,0) {$\mathbb{Z}_3$ and $\mathbb{Z}_2$};
\node[bx, fill=acc!10] (D) at (4.2,1.1) {$\mathbb{Z}_6$};
\node[bx, fill=acc!10] (S) at (4.4,-1.1) {$S_3$};
\draw[black, ->] (L) -- (D) node[midway, above, sloped, black, font=\scriptsize] {trivial action};
\draw[acc, thick, ->] (L) -- (S) node[midway, below, sloped, acc, font=\scriptsize] {inversion action};
\node[black, anchor=west, font=\scriptsize] at (2.5,1.55) {abelian};
\node[acc, anchor=west, font=\scriptsize] at (2.6,-1.6) {non-abelian};
\end{tikzpicture}
$$

## Limits of the construction

The semidirect product enlarges the reachable groups substantially, but it does
not reach everything. A group with no proper normal subgroup that has a
complement — a [simple
group](/abstract-algebra/group-actions-and-sylow/automorphisms-and-simple-groups)
is the clearest case — cannot be assembled this way. This limitation marks the boundary the
Hölder program draws: simple groups are the atoms, and semidirect products are
one tool for gluing atoms together, but the general gluing problem (the extension
problem) is harder.

Iterating the construction through a chain of abelian actions produces the
[solvable
groups](/abstract-algebra/products-and-group-structure/nilpotent-and-solvable-groups),
and the fact that solvability stops at $A_5$ is why the [quintic has no
formula](/abstract-algebra/galois-theory/solvability-by-radicals-and-the-quintic).

[^df-motiv]: **Dummit & Foote**, _Abstract Algebra_, §5.5 — Semidirect Products: the multiplication in $HK$ derived from $H \trianglelefteq G$, rewritten via the conjugation action to depend only on $H$, $K$, and $\varphi : K \to \operatorname{Aut}(H)$.
[^df-thm10]: **Dummit & Foote**, §5.5, Theorem 10: the pair-multiplication defines a group of order $|H||K|$ with $H \trianglelefteq H \rtimes K$, $H \cap K = 1$, and conjugation realizing $\varphi$.
[^df-prop11]: **Dummit & Foote**, §5.5, Proposition 11: the semidirect product equals the direct product iff $\varphi$ is trivial iff $K$ is normal.
[^df-dihedral]: **Dummit & Foote**, §5.5, Examples following Proposition 11: inversion actions realize $D_{2n} = \mathbb{Z}_n \rtimes \mathbb{Z}_2$ and $D_\infty = \mathbb{Z} \rtimes \mathbb{Z}_2$, and the holomorph $\operatorname{Hol}(H) = H \rtimes \operatorname{Aut}(H)$.
[^df-thm12]: **Dummit & Foote**, §5.5, Theorem 12: a normal subgroup with a complement gives a semidirect decomposition; $Q_8$ is cited as a group that is not a semidirect product of proper subgroups.
[^df-strategy]: **Dummit & Foote**, §5.5 — Some Classifications: the four-step strategy (find $H, K$; enumerate types; enumerate $\varphi$; deduplicate) and its application to orders $pq$ and $30$.
[^df-order30]: **Dummit & Foote**, §5.5, Example (Groups of Order 30): the normal cyclic subgroup of order $15$, the three involutions in $\operatorname{Aut}(\mathbb{Z}_{15})$, and the four resulting groups $\mathbb{Z}_{30}$, $\mathbb{Z}_5 \times S_3$, $\mathbb{Z}_3 \times D_{10}$, $D_{30}$.
[^df-order12]: **Dummit & Foote**, §5.5, Example (Groups of Order 12): the case analysis on which Sylow subgroup is normal, yielding five groups of order $12$, three non-abelian; and §5.3 — Table of Groups of Small Order.