Direct Products and Finite Abelian Groups

The direct product builds a larger group from smaller ones $H$ and $K$; a recognition theorem reverses the process, splitting a group $G$ into subgroups that determine it. Together they reduce the classification of finite abelian groups to bookkeeping about prime powers.

The external direct product

The construction takes tuples and multiplies them slot by slot.

Each group axiom holds in the product because it holds in every factor: the identity is $(1_1,\dots ,1_n)$, the inverse of $(g_1,\dots ,g_n)$ is $(g_1^{-1},\dots ,g_n^{-1})$, and associativity is checked coordinate by coordinate.¹ The order multiplies.

The order of a single element also decouples. Writing $x=(x_1,\dots ,x_n)$, the $k$-th power is $x^k=(x_1^k,\dots ,x_n^k)$, so $x^k=1$ exactly when every $x_i^k=1$. Hence

infinite if any coordinate has infinite order.

Factors as coordinate axes

Inside the product sit faithful copies of each factor, together with projection maps onto each.²

The last point characterizes a direct product: elements drawn from different factors commute, because they overlap in no coordinate. The name coordinate axes is literal for $\mathbb{R}\times \mathbb{R}$, where the two factors are the $x$- and $y$-axes — though there they are not the only copies of $\mathbb{R}$, since every line through the origin is another.

Working inside ${\mathbb{Z}}_n\times {\mathbb{Z}}_m$, we let $x$ generate the first factor and $y$ the second and write a general element as $x^a{y}^b$ in place of the ordered pair, since $x$ and $y$ commute and each generates its own axis.

Recognizing a direct product internally

A single group $G$ may already be the product of two of its subgroups. The criterion is the three features isolated above: two normal subgroups that commute elementwise and meet only in the identity.

The proof runs through two facts.³ First, when $H,K⊴G$ with $H\cap K=1$, every $h\in H$ and $k\in K$ commute: the element $h^{-1}{k}^{-1}hk$ lies in $H$ (since $k^{-1}hk\in H$) and in $K$ (since $h^{-1}{k}^{-1}h\in K$), so it lies in $H\cap K=1$. Second, each element of $HK$ is written as $hk$ in exactly one way when $H\cap K=1$. The map $hk\mapsto (h,k)$ is then a well-defined bijective homomorphism.

The word internal direct product names $HK$ when $H,K⊴G$ and $H\cap K=1$; external names the tuple construction $H\times K$. The theorem says the distinction is only notational: one writes elements as products $hk$, the other as pairs $(h,k)$.

A more flexible statement replaces normality of both factors by a single normal factor with a complement; that is the semidirect product. Normality of both sides forces commuting, and commuting is what makes the product direct.

The commutator subgroup

The recognition theorem is one instance of a general tool. For $x,y\in G$ the commutator is $[x,y]=x^{-1}{y}^{-1}xy$, and it equals $1$ exactly when $x$ and $y$ commute. The subgroup $G^{\prime }=⟨[x,y]:x,y\in G⟩$ generated by all commutators is the commutator subgroup.⁴ Two facts are used repeatedly:

• $G^{\prime }⊴G$ (in fact $G^{\prime }$ is characteristic), and $G/G^{\prime }$ is abelian.
• $G/H$ is abelian if and only if $G^{\prime }\le H$; so $G/G^{\prime }$ is the largest abelian quotient of $G$.

For $G=D_8$, the center $Z(D_8)=⟨r^2⟩$ has order $2$ and the quotient $D_8/Z(D_8)$ is the Klein four-group, hence abelian, so $D_8^{\prime }\le Z(D_8)$; since $D_8$ is non-abelian, $D_8^{\prime }=Z(D_8)=⟨r^2⟩$. The commutator returns in the study of solvable groups, where iterating $G\mapsto G^{\prime }$ builds the derived series.

The Chinese Remainder Theorem for cyclic groups

A product of two cyclic groups is again cyclic exactly when the factor orders are coprime.

The proof is the order formula.⁵ With ${\mathbb{Z}}_m=⟨x⟩$ and ${\mathbb{Z}}_n=⟨y⟩$, the element $xy$ has order $\mathrm{lcm}(m,n)$. When $\gcd (m,n)=1$ this equals $mn=|G|$, so $xy$ generates and $G$ is cyclic; when $\gcd (m,n)>1$ the lcm is strictly less than $mn$, so no element has order $mn$ and $G$ is not cyclic.

This is the group-theoretic form of the Chinese Remainder Theorem: choosing a residue modulo $m$ and one modulo $n$ independently is the same as choosing a single residue modulo $mn$, precisely when $m$ and $n$ are coprime. For example, ${\mathbb{Z}}_2\times {\mathbb{Z}}_3\cong {\mathbb{Z}}_6$, laid out as a grid below.

When the factor orders share a prime, the product is not cyclic.

The Fundamental Theorem of Finitely Generated Abelian Groups

Every finitely generated abelian group is a product of cyclic pieces, and the list of pieces is unique once put in a canonical order. This is the complete classification.

Two finitely generated abelian groups are isomorphic if and only if they have the same free rank and the same list of invariant factors.⁶ The group is finite exactly when $r=0$, and then $|G|=n_1{n}_2\cdots n_s$. The divisibility chain forces the largest factor to the front: every prime dividing $|G|$ divides $n_1$, so $n_1$ is the exponent of $G$, the least $n$ with $x^n=1$ for all $x$.

The squarefree case follows at once.

Because a squarefree $n$ cannot support any nontrivial divisibility chain — the front factor $n_1$ must already be divisible by every prime, hence $n_1=n$.

The elementary divisor form

There is a second canonical form, obtained by splitting each cyclic factor into prime-power pieces via the coprime-splitting proposition.

The elementary divisors are the invariant factors of the Sylow subgroups. Their advantage is computational: listing all abelian groups of order $p^{\alpha }$ is the same as listing all partitions of $\alpha$, and this count does not depend on $p$.⁷ For a prime power $p^{\alpha }$, each partition $\alpha ={\beta }_1+{\beta }_2+\cdots +{\beta }_t$ with ${\beta }_1\ge \cdots \ge {\beta }_t\ge 1$ names the group ${\mathbb{Z}}_{p^{{\beta }_1}}\times \cdots \times {\mathbb{Z}}_{p^{{\beta }_t}}$.

The two forms describe the same groups, so an explicit conversion relates them by regrouping the prime-power blocks.

Converting between the forms

Take $G\cong {\mathbb{Z}}_{30}\times {\mathbb{Z}}_{30}\times {\mathbb{Z}}_2$, of order $2^3\cdot 3^2\cdot 5^2$. Splitting each ${\mathbb{Z}}_{30}\cong {\mathbb{Z}}_2\times {\mathbb{Z}}_3\times {\mathbb{Z}}_5$ and grouping by prime gives the elementary divisors. Reassembling into a divisibility chain gives the invariant factors.

Reading down the columns, the elementary divisors are $2,2,2,3,3,5,5$; reading across the rows, the invariant factors are $30,30,2$ (each divides the one above). The same group, two descriptions:

To decide whether two products of cyclic groups are isomorphic, compute either invariant. For instance ${\mathbb{Z}}_6\times {\mathbb{Z}}_{15}$ has elementary divisors $2,3,3,5$, while ${\mathbb{Z}}_{10}\times {\mathbb{Z}}_9$ has $2,5,9$; the lists differ, so the groups differ — visibly, the second has an element of order $9$ and the first does not.

Enumerating abelian groups of a fixed order

The elementary divisor form turns enumeration into a product of partition counts. If $n=p_1^{{\alpha }_1}\cdots p_k^{{\alpha }_k}$ and $q_i$ is the number of partitions of ${\alpha }_i$, then the number of abelian groups of order $n$ is $q_1{q}_2\cdots q_k$, since a group is chosen by independently picking a partition at each prime.

For $n=1800=2^3\cdot 3^2\cdot 5^2$: the exponent $3$ has three partitions, each of $2$ has two, so there are $3\cdot 2\cdot 2=12$ abelian groups of order $1800$. The choices at each prime:

Each of the $12$ groups is one choice from each row, multiplied together. The count depends only on the exponent pattern of $n$, not on the primes — every order of the form $p^3{q}^2{r}^2$ with distinct primes has exactly $12$ abelian groups.

The same partition data reappears with a twist in the structure theorem for modules over a PID, where the role of $\mathbb{Z}$ is played by an arbitrary principal ideal domain and finite abelian groups become the special case $R=\mathbb{Z}$.

The scope of the classification

For any $n$, the abelian groups of order $n$ are listed explicitly and finitely. Recovering which isomorphism type a given group belongs to may still require substantial computation when $n$ carries large prime powers, since the number of types grows with the partition counts.

The abelian case is closed; the non-abelian case is not. A direct product of abelian groups stays abelian, since a product is abelian if and only if every factor is. To build non-abelian groups from abelian pieces, we loosen the construction, keeping only one factor normal and letting the other act on it: the semidirect product.