Direct Products

Suppose you are handed two groups HH and KK that you understand completely. You want to build a new, larger group out of them. How should the elements interact? The simplest possible answer — let elements of HH and elements of KK ignore each other entirely — produces the external direct product. This is a construction so natural that it silently underlies the coordinate plane R2\mathbb{R}^2, the Klein four group, and the structure of every finite abelian group. The direct product matters for two reasons. First, it is the first and most important way to manufacture new groups from old ones. Second, it runs in reverse: given a mysterious group GG, we often want to certify that GG is secretly a direct product of two smaller, better-understood pieces sitting inside it. This reverse direction is the internal direct product. The bridge between the external construction (built from scratch) and the internal recognition (found inside an existing group) is the heart of this article. The big picture is that direct products are a decomposition tool. Just as an integer factors into primes, a finite group often factors into a product of subgroups. Its internal arithmetic is then easier to analyze separately. This idea reaches its full power in the classification of finite abelian groups. There, every such group is shown to be a direct product of cyclic groups of prime-power order. It also foreshadows the semidirect product, which relaxes the "ignore each other" condition. That relaxation allows controlled interaction between the factors. This article gives the precise external construction. It proves its basic properties (order, abelianness, universal mapping property). Then it gives the internal recognition criterion. This is the theorem that tells you, from purely internal conditions on two subgroups H,K≤GH, K \le G, that G≅H×KG \cong H \times K. Both directions are proved in full. Worked examples show the criterion applied correctly. A counterexample shows exactly where it can fail.

Key takeaways
  • The external direct product H×KH \times K is the Cartesian product with coordinatewise operation; it is a group of order ∣H∣⋅∣K∣|H|\cdot|K| when both are finite.
  • H×{e}H \times \{e\} and {e}×K\{e\}\times K are normal subgroups of H×KH\times K isomorphic to HH and KK, respectively, intersecting only in the identity.
  • The internal direct product recognition criterion certifies G≅H×KG \cong H \times K from three purely internal conditions on subgroups H,K≤GH,K \le G.
  • The three conditions are: H,K⊴GH, K \trianglelefteq G, H∩K={e}H \cap K = \{e\}, and HK=GHK = G.
  • Under these conditions every element of HH commutes with every element of KK, which is the mechanism that makes the isomorphism work.
  • Key idea: direct products let you reconstruct a whole group exactly from two normal, complementary subgroups whose only common element is the identity.

Where this lives today

The direct product is the group-theoretic version of "independent coordinates," and this idea recurs constantly. In algebra, the Fundamental Theorem of Finite Abelian Groups expresses every finite abelian group as a direct product of cyclic groups of prime-power order, turning classification into pure arithmetic on the prime factorization of the order. In crystallography and physics, the symmetry group of a crystal lattice is frequently analyzed as (or built from) direct products of point-group and translation-group symmetries, and in quantum mechanics, the state space of a composite system with independent degrees of freedom is a tensor product whose symmetry group often factors as a direct product of the symmetry groups of each subsystem. In coding theory and cryptography, the Chinese Remainder Theorem isomorphism Zn≅Zp1a1×⋯×Zpkak\mathbb{Z}_n \cong \mathbb{Z}_{p_1^{a_1}}\times\cdots\times\mathbb{Z}_{p_k^{a_k}} (an instance of the internal/external direct product correspondence) underlies fast modular arithmetic algorithms used in RSA implementations. In computer science, direct products of finite groups appear in the design of error-correcting codes and in state-space decompositions for concurrent systems, where independent components correspond exactly to direct-product factors with no cross-interaction — precisely the condition that elements of one factor commute with elements of the other.

Ask the tutor: Can you show me how the internal direct product criterion is used as a lemma inside the proof of the Fundamental Theorem of Finite Abelian Groups?

DefinitionExternal and Internal Direct Product

External direct product. Let HH and KK be groups (written multiplicatively, with identities eH,eKe_H, e_K). The external direct product H×KH \times K is the set

H×K={(h,k):h∈H, k∈K}H \times K = \{(h,k) : h \in H,\ k \in K\}

with componentwise operation

(h1,k1)(h2,k2)=(h1h2, k1k2).(h_1,k_1)(h_2,k_2) = (h_1h_2,\ k_1k_2).

This operation is associative because multiplication in HH and in KK is associative in each coordinate separately; the identity is (eH,eK)(e_H,e_K); the inverse of (h,k)(h,k) is (h−1,k−1)(h^{-1},k^{-1}). Thus H×KH\times K is a group. This construction extends verbatim to any finite list H1,…,HnH_1,\dots,H_n, giving H1×⋯×HnH_1\times\cdots\times H_n with entrywise operations.

Internal direct product. Let GG be a group and let H,K≤GH, K \le G be subgroups. We say GG is the internal direct product of HH and KK, written G=H×KG = H \times K (by abuse of notation identifying GG with the external product), if: 1. H⊴GH \trianglelefteq G and K⊴GK \trianglelefteq G (both are normal subgroups), 2. H∩K={e}H \cap K = \{e\}, 3. HK={hk:h∈H,k∈K}=GHK = \{hk : h\in H, k \in K\} = G.

Intuition

Think of H×KH \times K as laying out two independent "axes" of freedom: moving in the HH-coordinate never disturbs the KK-coordinate and vice versa. This is exactly the picture of the coordinate plane R2=R×R\mathbb{R}^2 = \mathbb{R} \times \mathbb{R}: horizontal and vertical motion are independent, and every point is uniquely a pair (horizontal amount, vertical amount). The group-theoretic direct product is the abstract version of "independent axes," and the order formula ∣H×K∣=∣H∣⋅∣K∣|H\times K| = |H|\cdot|K| is the group-theoretic analogue of counting lattice points in a grid. The internal criterion asks: given a single group GG, can we find two "independent axes" already living inside it? Normality of HH and KK ensures that conjugation cannot mix the axes; the trivial intersection ensures no element is redundantly counted on both axes; and HK=GHK = G ensures the axes together span everything. The surprising and crucial fact — proved below — is that these three conditions force HH and KK to commute elementwise, which is exactly what is needed to recover the direct-product multiplication rule from data that, on its face, only talks about subgroups and cosets.

Figure
The internal direct product as two independent axes inside GG
H-axis K-axis e h k hk = (h,k) G = HK, with H ∩ K = {e}, each element hk unique
TheoremExternal Direct Product: Basic Properties

Let HH and KK be groups and let G=H×KG = H \times K be their external direct product.

(a) If HH and KK are finite, ∣G∣=∣H∣⋅∣K∣|G| = |H| \cdot |K|.

(b) The subsets

Hˉ=H×{eK},Kˉ={eH}×K\bar H = H \times \{e_K\}, \qquad \bar K = \{e_H\} \times K

are subgroups of GG, each normal in GG, with Hˉ≅H\bar H \cong H, Kˉ≅K\bar K \cong K, Hˉ∩Kˉ={(eH,eK)}\bar H \cap \bar K = \{(e_H,e_K)\}, and HˉKˉ=G\bar H \bar K = G.

(c) Every element of Hˉ\bar H commutes with every element of Kˉ\bar K.

(d) GG is abelian if and only if both HH and KK are abelian.

(e) (Universal property) For any group LL and homomorphisms φ:L→H\varphi: L \to H, ψ:L→K\psi: L \to K, there is a unique homomorphism θ:L→H×K\theta: L \to H \times K with πH∘θ=φ\pi_H \circ \theta = \varphi and πK∘θ=ψ\pi_K \circ \theta = \psi, where πH,πK\pi_H, \pi_K are the coordinate projections.

Proof

(a) As a set, H×KH \times K is the Cartesian product of the underlying sets, so ∣H×K∣=∣H∣⋅∣K∣|H\times K| = |H|\cdot|K| by the multiplication principle for finite sets.

(b) First check Hˉ\bar H is a subgroup: it is nonempty (contains (eH,eK)(e_H,e_K)), and for (h1,eK),(h2,eK)∈Hˉ(h_1,e_K),(h_2,e_K) \in \bar H we have (h1,eK)(h2,eK)−1=(h1h2−1,eK)∈Hˉ(h_1,e_K)(h_2,e_K)^{-1} = (h_1 h_2^{-1}, e_K) \in \bar H, so Hˉ≤G\bar H \le G by the subgroup criterion. The same argument gives Kˉ≤G\bar K \le G. The map ιH:H→Hˉ\iota_H : H \to \bar H, ιH(h)=(h,eK)\iota_H(h) = (h,e_K), is a bijection by construction, and

ιH(h1h2)=(h1h2,eK)=(h1,eK)(h2,eK)=ιH(h1)ιH(h2),\iota_H(h_1 h_2) = (h_1h_2, e_K) = (h_1,e_K)(h_2,e_K) = \iota_H(h_1)\iota_H(h_2),

so ιH\iota_H is an isomorphism; hence Hˉ≅H\bar H \cong H. Symmetrically Kˉ≅K\bar K \cong K. For normality, take (a,b)∈G(a,b) \in G arbitrary and (h,eK)∈Hˉ(h,e_K) \in \bar H. Then

(a,b)(h,eK)(a,b)−1=(aha−1, beKb−1)=(aha−1,eK)∈Hˉ,(a,b)(h,e_K)(a,b)^{-1} = (aha^{-1},\ b e_K b^{-1}) = (aha^{-1}, e_K) \in \bar H,

since aha−1∈Haha^{-1} \in H. So Hˉ⊴G\bar H \trianglelefteq G; symmetrically Kˉ⊴G\bar K \trianglelefteq G. If (h,k)∈Hˉ∩Kˉ(h,k) \in \bar H \cap \bar K then the second coordinate is eKe_K (from membership in Hˉ\bar H, since Hˉ\bar H's elements have second coordinate eKe_K) and the first coordinate is eHe_H (from membership in Kˉ\bar K). Hence (h,k)=(eH,eK)(h,k) = (e_H,e_K), so Hˉ∩Kˉ={(eH,eK)}\bar H \cap \bar K = \{(e_H,e_K)\}. For any (h,k)∈G(h,k) \in G we can write (h,k)=(h,eK)(eH,k)∈HˉKˉ(h,k) = (h,e_K)(e_H,k) \in \bar H \bar K, so G⊆HˉKˉG \subseteq \bar H\bar K; the reverse inclusion is automatic since Hˉ,Kˉ≤G\bar H, \bar K \le G. Hence HˉKˉ=G\bar H \bar K = G.

(c) For (h,eK)∈Hˉ(h,e_K) \in \bar H and (eH,k)∈Kˉ(e_H,k) \in \bar K:

(h,eK)(eH,k)=(heH, eKk)=(h,k),(eH,k)(h,eK)=(eHh, keK)=(h,k).(h,e_K)(e_H,k) = (h e_H,\ e_K k) = (h,k), \qquad (e_H,k)(h,e_K) = (e_H h,\ k e_K) = (h,k).

The two products agree, so (h,eK)(h,e_K) and (eH,k)(e_H,k) commute.

(d) (⇒\Rightarrow) Suppose GG is abelian. Since Hˉ≤G\bar H \le G and subgroups of abelian groups are abelian, Hˉ\bar H is abelian, and Hˉ≅H\bar H \cong H forces HH abelian (isomorphisms preserve commutativity: if φ\varphi is an isomorphism and φ(x)φ(y)=φ(y)φ(x)\varphi(x)\varphi(y)=\varphi(y)\varphi(x) always holds because GG abelian, then applying φ−1\varphi^{-1} gives xy=yxxy=yx). Symmetrically KK is abelian. (⇐\Leftarrow) Suppose H,KH,K abelian. For (h1,k1),(h2,k2)∈G(h_1,k_1),(h_2,k_2)\in G,

(h1,k1)(h2,k2)=(h1h2,k1k2)=(h2h1,k2k1)=(h2,k2)(h1,k1),(h_1,k_1)(h_2,k_2) = (h_1h_2, k_1k_2) = (h_2h_1, k_2k_1) = (h_2,k_2)(h_1,k_1),

using commutativity in HH and in KK coordinatewise. So GG is abelian.

(e) Define θ(x)=(φ(x),ψ(x))\theta(x) = (\varphi(x), \psi(x)) for x∈Lx \in L. This is a homomorphism since

θ(xy)=(φ(xy),ψ(xy))=(φ(x)φ(y),ψ(x)ψ(y))=(φ(x),ψ(x))(φ(y),ψ(y))=θ(x)θ(y),\theta(x y) = (\varphi(xy),\psi(xy)) = (\varphi(x)\varphi(y), \psi(x)\psi(y)) = (\varphi(x),\psi(x))(\varphi(y),\psi(y)) = \theta(x)\theta(y),

using that φ,ψ\varphi,\psi are homomorphisms. Clearly πH(θ(x))=φ(x)\pi_H(\theta(x)) = \varphi(x) and πK(θ(x))=ψ(x)\pi_K(\theta(x)) = \psi(x) for all xx, so the required diagram commutes. For uniqueness, if θ′\theta' also satisfies πHθ′=φ\pi_H\theta' = \varphi, πKθ′=ψ\pi_K \theta' = \psi, then for each xx, θ′(x)\theta'(x) is a pair whose first coordinate is φ(x)\varphi(x) and second coordinate is ψ(x)\psi(x), forcing θ′(x)=(φ(x),ψ(x))=θ(x)\theta'(x) = (\varphi(x),\psi(x)) = \theta(x). Hence θ\theta is unique. ■\blacksquare

∎
TheoremInternal Direct Product Recognition Criterion

Let GG be a group and let H,K≤GH, K \le G be subgroups satisfying: 1. H⊴GH \trianglelefteq G and K⊴GK \trianglelefteq G, 2. H∩K={e}H \cap K = \{e\}, 3. HK=GHK = G. Then:

(a) Every h∈Hh \in H commutes with every k∈Kk \in K.

(b) Every element g∈Gg \in G can be written uniquely as g=hkg = hk with h∈Hh \in H, k∈Kk \in K.

(c) The map

Φ:H×K→G,Φ(h,k)=hk\Phi : H \times K \to G, \qquad \Phi(h,k) = hk

is a group isomorphism from the external direct product H×KH\times K onto GG. Consequently G≅H×KG \cong H \times K.

Proof

We use conditions (1)–(3) exactly as given; the proof technique is direct verification via the commutator trick, followed by construction of an explicit isomorphism.

(a) Elementwise commuting. Fix h∈Hh \in H, k∈Kk \in K. Consider the commutator

c=hkh−1k−1=(hkh−1) k−1.c = hkh^{-1}k^{-1} = (hkh^{-1})\,k^{-1}.

Since K⊴GK \trianglelefteq G, conjugating kk by h−1∈Gh^{-1}\in G... more directly: since K⊴GK \trianglelefteq G and h∈Gh \in G, we have hkh−1∈Khkh^{-1} \in K (as KK is normal, conjugates of elements of KK by any group element stay in KK). Hence c=(hkh−1)k−1∈Kc = (hkh^{-1})k^{-1} \in K, being a product of two elements of KK. On the other hand, write c=h (kh−1k−1)c = h\,(kh^{-1}k^{-1}). Since H⊴GH \trianglelefteq G and k∈Gk \in G, we have kh−1k−1∈Hkh^{-1}k^{-1} \in H. Hence c=h⋅(kh−1k−1)∈Hc = h\cdot(kh^{-1}k^{-1}) \in H, being a product of two elements of HH. Thus c∈H∩K={e}c \in H \cap K = \{e\} by hypothesis (2), so c=ec = e, i.e.

hkh−1k−1=e  ⟺  hk=kh.hkh^{-1}k^{-1} = e \iff hk = kh.

This proves every h∈Hh\in H commutes with every k∈Kk \in K.

(b) Existence and uniqueness of the factorization. Existence is immediate from hypothesis (3): HK=GHK = G means every g∈Gg \in G has the form g=hkg = hk for some h∈Hh \in H, k∈Kk\in K. For uniqueness, suppose g=h1k1=h2k2g = h_1 k_1 = h_2 k_2 with h1,h2∈Hh_1,h_2 \in H, k1,k2∈Kk_1,k_2\in K. Then

h2−1h1=k2k1−1.h_2^{-1} h_1 = k_2 k_1^{-1}.

The left side lies in HH (subgroup, closed under products/inverses), and the right side lies in KK (same reason). So both sides lie in H∩K={e}H \cap K = \{e\} by hypothesis (2). Hence h2−1h1=eh_2^{-1}h_1 = e and k2k1−1=ek_2 k_1^{-1} = e, giving h1=h2h_1 = h_2 and k1=k2k_1 = k_2. This proves uniqueness.

(c) Φ\Phi is an isomorphism. First, Φ\Phi is a homomorphism from H×KH \times K (with its direct-product operation) to GG: for (h1,k1),(h2,k2)∈H×K(h_1,k_1), (h_2,k_2) \in H\times K,

Φ((h1,k1)(h2,k2))=Φ(h1h2, k1k2)=h1h2k1k2.\Phi\big((h_1,k_1)(h_2,k_2)\big) = \Phi(h_1h_2,\ k_1k_2) = h_1h_2k_1k_2.

By part (a), h2k1=k1h2h_2 k_1 = k_1 h_2 (since h2∈H,k1∈Kh_2\in H, k_1\in K), so

h1h2k1k2=h1k1h2k2=Φ(h1,k1) Φ(h2,k2).h_1h_2k_1k_2 = h_1k_1h_2k_2 = \Phi(h_1,k_1)\,\Phi(h_2,k_2).

Hence Φ((h1,k1)(h2,k2))=Φ(h1,k1)Φ(h2,k2)\Phi\big((h_1,k_1)(h_2,k_2)\big) = \Phi(h_1,k_1)\Phi(h_2,k_2), so Φ\Phi is a homomorphism. Φ\Phi is surjective: by hypothesis (3), HK=GHK = G, so every g∈Gg \in G equals hk=Φ(h,k)hk = \Phi(h,k) for some (h,k)∈H×K(h,k) \in H\times K. Φ\Phi is injective: suppose Φ(h,k)=e\Phi(h,k) = e, i.e. hk=ehk = e, i.e. h=k−1h = k^{-1}. Then h∈Hh \in H and h=k−1∈Kh = k^{-1} \in K (since KK is a subgroup, k∈K⇒k−1∈Kk \in K \Rightarrow k^{-1}\in K), so h∈H∩K={e}h \in H \cap K = \{e\}, giving h=eh = e and hence k=h−1=ek = h^{-1} = e. So ker⁡Φ={(e,e)}\ker \Phi = \{(e,e)\}, and a homomorphism with trivial kernel is injective (if Φ(x)=Φ(y)\Phi(x) = \Phi(y) then Φ(xy−1)=e\Phi(xy^{-1}) = e so xy−1∈ker⁡Φ={e}xy^{-1}\in\ker\Phi = \{e\}, so x=yx=y). A bijective homomorphism is an isomorphism (its set-theoretic inverse is automatically a homomorphism, by the standard argument: if Φ(a)=g1,Φ(b)=g2\Phi(a)=g_1,\Phi(b)=g_2 then Φ−1(g1g2)=Φ−1(Φ(a)Φ(b))=Φ−1(Φ(ab))=ab=Φ−1(g1)Φ−1(g2)\Phi^{-1}(g_1g_2)=\Phi^{-1}(\Phi(a)\Phi(b))=\Phi^{-1}(\Phi(ab))=ab=\Phi^{-1}(g_1)\Phi^{-1}(g_2)). Hence Φ:H×K→G\Phi: H\times K \to G is an isomorphism, and G≅H×KG \cong H\times K. ■\blacksquare

∎

Worked examples

Example 1: A genuinely internal direct product inside Z12\mathbb{Z}_{12} (asymmetric orders). Let G=Z12={0,1,…,11}G = \mathbb{Z}_{12} = \{0,1,\dots,11\} under addition mod 1212. Let

H=⟨4⟩={0,4,8},K=⟨3⟩={0,3,6,9}.H = \langle 4 \rangle = \{0,4,8\}, \qquad K = \langle 3 \rangle = \{0,3,6,9\}.

Note ∣H∣=3|H| = 3 and ∣K∣=4|K| = 4 — deliberately asymmetric, unlike a naive 2×22\times 2 example.

  • Normality: GG is abelian, so every subgroup is normal. Hence H,K⊴GH, K \trianglelefteq G automatically.
  • Trivial intersection: H∩KH \cap K consists of common multiples of both 44 and 33 inside Z12\mathbb{Z}_{12}; since gcd⁡(4,3)=1\gcd(4,3)=1 and H={0,4,8}H = \{0,4,8\}, K={0,3,6,9}K=\{0,3,6,9\}, direct inspection shows the only common element is 00. So H∩K={0}H \cap K = \{0\}.
  • HK=GHK = G: ∣H∣∣K∣=3⋅4=12=∣G∣|H||K| = 3\cdot 4 = 12 = |G|. Since the map Φ:H×K→G\Phi: H\times K \to G, Φ(h,k)=h+k\Phi(h,k)=h+k, has trivial kernel by the same argument as in the theorem (using H∩K={0}H\cap K=\{0\}) it is injective; an injective map between finite sets of equal size (∣H×K∣=12=∣G∣|H\times K| = 12 = |G|) is automatically surjective. Hence HK=GHK = G.

All three hypotheses hold, so by the recognition criterion,

Z12≅H×K≅Z3×Z4.\mathbb{Z}_{12} \cong H \times K \cong \mathbb{Z}_3 \times \mathbb{Z}_4.

Concretely, every n∈Z12n \in \mathbb{Z}_{12} decomposes uniquely as n=h+kn = h + k with h∈{0,4,8}h \in \{0,4,8\}, k∈{0,3,6,9}k\in\{0,3,6,9\}: for instance 7=4+37 = 4 + 3, 10=4+610 = 4+6, 11=8+311 = 8+3, matching the uniqueness clause of part (b) of the theorem. This is the group-theoretic shadow of the Chinese Remainder Theorem arithmetic fact Z12≅Z3×Z4\mathbb{Z}_{12}\cong \mathbb{Z}_3\times\mathbb{Z}_4, and it previews how the Fundamental Theorem of Finite Abelian Groups decomposes any finite abelian group into cyclic prime-power pieces.

Example 2: S3×Z2S_3 \times \mathbb{Z}_2 built externally, with an explicit non-abelian factor. Let H=S3H = S_3 (order 6, nonabelian) and K=Z2={0,1}K = \mathbb{Z}_2 = \{0,1\}. Form G=S3×Z2G = S_3 \times \mathbb{Z}_2, of order 1212 by Theorem part (a). Take specific elements σ=(1 2 3)∈S3\sigma = (1\,2\,3) \in S_3 and τ=(1 2)∈S3\tau = (1\,2) \in S_3, and consider a=(σ,1)a = (\sigma, 1), b=(τ,0)b = (\tau, 0) in GG. Compute

ab=(σ,1)(τ,0)=(στ,1),ba=(τ,0)(σ,1)=(τσ,1).ab = (\sigma,1)(\tau,0) = (\sigma\tau, 1), \qquad ba = (\tau,0)(\sigma,1) = (\tau\sigma, 1).

Since στ≠τσ\sigma\tau \ne \tau\sigma in S3S_3 (a direct check: στ=(1 2 3)(1 2)=(1 3)\sigma\tau = (1\,2\,3)(1\,2) = (1\,3), while τσ=(1 2)(1 2 3)=(2 3)\tau\sigma = (1\,2)(1\,2\,3) = (2\,3), and these are different transpositions), we get ab≠baab \ne ba. So GG is nonabelian — consistent with Theorem part (d), since one factor (S3S_3) is nonabelian. Now identify the canonical copies Hˉ=S3×{0}\bar H = S_3 \times \{0\} and Kˉ={e}×Z2\bar K = \{e\}\times \mathbb{Z}_2 inside GG. By Theorem part (b) these are normal, meet trivially, and multiply out to all of GG; by part (c) every element of Hˉ\bar H commutes with every element of Kˉ\bar K — for instance (σ,0)(e,1)=(σ,1)=(e,1)(σ,0)(\sigma, 0)(e,1) = (\sigma,1) = (e,1)(\sigma,0), confirming the commuting mechanism concretely even though GG itself is nonabelian, because the noncommutativity of GG lives entirely within the Hˉ\bar H copy, never across the two factors.

CounterexampleDropping normality of one factor breaks the criterion

Consider G=S3G = S_3, and let

H={e,(1 2 3),(1 3 2)}=⟨(1 2 3)⟩(the unique subgroup of order 3, normal in S3),H = \{e, (1\,2\,3), (1\,3\,2)\} = \langle (1\,2\,3)\rangle \quad(\text{the unique subgroup of order }3,\text{ normal in }S_3),
K={e,(1 2)}(a subgroup of order 2).K = \{e, (1\,2)\} \quad(\text{a subgroup of order }2).

Check the hypotheses of the recognition criterion:

  • H⊴S3H \trianglelefteq S_3: true, since HH has index 22 and subgroups of index 22 are always normal.
  • K⊴S3K \trianglelefteq S_3: false. Conjugating (1 2)(1\,2) by (1 2 3)(1\,2\,3) gives
(1 2 3)(1 2)(1 2 3)−1=(2 3)∉K,(1\,2\,3)(1\,2)(1\,2\,3)^{-1} = (2\,3) \notin K,

so KK is not closed under conjugation by all of GG, hence not normal.

  • H∩K={e}H \cap K = \{e\}: true, since ∣H∣=3,∣K∣=2|H|=3,|K|=2 are coprime, forcing trivial intersection by Lagrange's theorem.
  • HK=S3HK = S_3: true, since ∣H∣∣K∣=6=∣S3∣|H||K| = 6 = |S_3| and H∩KH\cap K trivial forces HKHK to have 66 distinct elements (same counting argument as Example 1), filling up S3S_3.

So conditions (2) and (3) hold, but condition (1) fails for KK. Does the conclusion still hold? Test elementwise commuting (the key mechanism from part (a) of the proof): take h=(1 2 3)∈Hh = (1\,2\,3) \in H and k=(1 2)∈Kk = (1\,2) \in K. Then

hk=(1 2 3)(1 2)=(1 3),kh=(1 2)(1 2 3)=(2 3).hk = (1\,2\,3)(1\,2) = (1\,3), \qquad kh = (1\,2)(1\,2\,3) = (2\,3).

Since (1 3)≠(2 3)(1\,3) \ne (2\,3), hh and kk do not commute. So the commuting conclusion of part (a) fails, and indeed S3S_3 is nonabelian while H×K≅Z3×Z2≅Z6H\times K \cong \mathbb{Z}_3\times \mathbb{Z}_2 \cong \mathbb{Z}_6 is abelian (an abelian group of order 6), so

S3≇H×K,S_3 \not\cong H \times K,

even though H∩K={e}H\cap K = \{e\} and HK=GHK=G both hold. This shows normality of both subgroups is essential, not merely a convenience: without it the map Φ(h,k)=hk\Phi(h,k)=hk still exists and is bijective (a bijection of sets), but it fails to be a homomorphism, because the commutator argument in the proof of part (a) used normality of KK (to get hkh−1∈Khkh^{-1}\in K) and of HH (to get kh−1k−1∈Hkh^{-1}k^{-1}\in H) simultaneously.

Scope, edge cases & common traps

SituationWhat holdsWhat can fail
H⊴GH \trianglelefteq G, H∩K={e}H\cap K=\{e\}, HK=GHK=G, but KK not normalΦ\Phi is a bijection of setsΦ\Phi need not be a homomorphism; G≇H×KG\not\cong H\times K in general (see counterexample)
H,K⊴GH,K \trianglelefteq G, HK=GHK=G, but H∩K≠{e}H\cap K \ne \{e\}Elementwise commuting still holds (proof of part (a) only used normality)Factorization g=hkg=hk not unique; Φ\Phi not injective; get G≅(H×K)/DG \cong (H\times K)/D for a diagonal-type subgroup DD, not G≅H×KG\cong H\times K
Infinite H,KH, KExternal product still a group; parts (a)–(e) of the first theorem hold verbatim (order formula (a) only needs finiteness)Some finiteness-dependent counting shortcuts (like "injective + equal finite size ⇒\Rightarrow surjective") are unavailable
More than two factors H1,…,HnH_1,\dots,H_nCriterion generalizes: pairwise trivial intersection with the product of the others, all normal, product spans GGMust check Hi∩(H1⋯Hi^⋯Hn)={e}H_i \cap (H_1\cdots \widehat{H_i}\cdots H_n) = \{e\}, not merely pairwise trivial intersections Hi∩Hj={e}H_i\cap H_j=\{e\}

The mathematician behind it

Origins · 1 minCamille Jordan and how this was discovered
CJ
Camille Jordan
1838–1922 · France, structural group theory in the wake of Galois

Jordan's Traité des substitutions et des équations algébriques (1870) systematized the study of permutation groups arising from Galois theory, and with it the language of decomposing groups into simpler normal pieces — the ancestor of direct-product and composition-series thinking that later authors (notably Hölder, and then Dummit–Foote-style modern treatments) sharpened into the internal/external direct product formalism used today.

History
- 1832 — Galois's manuscripts (published posthumously 1846) introduce the group-theoretic study of equation solvability, laying groundwork for structural decomposition ideas.
- 1870 — Jordan publishes his Traité, systematically studying normal subgroups and quotient structure in permutation groups.
- 1889 — Otto Hölder refines Jordan's decomposition results, contributing to what becomes the Jordan-Hölder Theorem on composition series, a close cousin of direct-product decomposition.
- 1900s–1920s — abstract group theory (Weber, Burnside, and later Emmy Noether's structural school) formalizes direct products and internal/external equivalence in the modern axiomatic language.
- 1930s–present — the internal direct product criterion becomes a standard stepping stone toward the Fundamental Theorem of Finite Abelian Groups and, with relaxed commuting hypotheses, toward semidirect products.
Why do we need H∩K={e}H \cap K = \{e\} in the internal direct product criterion?

This condition guarantees that the factorization g=hkg = hk of each element is unique, which is essential for Φ(h,k)=hk\Phi(h,k) = hk to be injective. If H∩KH \cap K contained some nonidentity element xx, then for any g=hkg = hk we could also write g=(hx)(x−1k)g = (hx)(x^{-1}k), giving two different-looking pairs mapping to the same gg, so Φ\Phi would not be injective and could not be an isomorphism from H×KH \times K. Trivial intersection is precisely the "no double counting" condition that makes the correspondence between GG and the Cartesian product H×KH \times K bijective, matching the uniqueness clause proved in part (b) of the recognition theorem.

Why do we need both HH and KK to be normal, not just one?

The proof that elements of HH commute with elements of KK uses both normalities simultaneously: for h∈H,k∈Kh \in H, k \in K, the commutator c=hkh−1k−1c = hkh^{-1}k^{-1} is shown to lie in KK by writing c=(hkh−1)k−1c = (hkh^{-1})k^{-1} and using normality of KK to place hkh−1hkh^{-1} in KK, and is shown to lie in HH by writing c=h(kh−1k−1)c = h(kh^{-1}k^{-1}) and using normality of HH to place kh−1k−1kh^{-1}k^{-1} in HH. Both memberships are needed to conclude c∈H∩K={e}c \in H \cap K = \{e\}. If only one factor is normal, the counterexample with G=S3G = S_3 shows the commuting conclusion, and hence the isomorphism G≅H×KG \cong H \times K, can genuinely fail.

Is the internal direct product criterion a definition or a theorem?

Both, depending on convention, but logically it functions as a theorem here. We define "GG is the internal direct product of HH and KK" by the three listed conditions on subgroups of GG. We then prove, as a theorem, that this internal condition is equivalent to GG being isomorphic to the external direct product H×KH \times K via the explicit map Φ(h,k)=hk\Phi(h,k) = hk. Some textbooks instead define the internal direct product directly as "G≅H×KG \cong H\times K via this multiplication map" and then prove the three conditions are equivalent to that; either way, the mathematical content proved is identical.

Does the direct product criterion generalize to more than two subgroups?

Yes. For subgroups H1,…,Hn≤GH_1, \dots, H_n \le G, one says GG is the internal direct product of the HiH_i if each Hi⊴GH_i \trianglelefteq G, the product H1H2⋯HnH_1 H_2 \cdots H_n equals GG, and each HiH_i intersects the product of all the other subgroups trivially: Hi∩(H1⋯Hi^⋯Hn)={e}H_i \cap (H_1\cdots \widehat{H_i} \cdots H_n) = \{e\} for each ii. This last condition is strictly stronger than merely requiring Hi∩Hj={e}H_i \cap H_j = \{e\} for all i≠ji\ne j pairwise; without the stronger condition, uniqueness of factorization can fail even when all pairwise intersections are trivial.

How does the direct product relate to the semidirect product?

The direct product is the special case of the semidirect product where the "action" of one factor on the other is trivial. In a semidirect product H⋉KH \ltimes K, elements of HH are allowed to act nontrivially on KK by an automorphism-valued homomorphism φ:H→Aut⁡(K)\varphi: H \to \operatorname{Aut}(K), and the multiplication rule becomes (h1,k1)(h2,k2)=(h1h2, φ(h2−1)(k1)⋅k2)(h_1,k_1)(h_2,k_2) = (h_1h_2,\ \varphi(h_2^{-1})(k_1)\cdot k_2) or a similar twisted formula (conventions vary). When φ\varphi is the trivial homomorphism (every hh acts as the identity automorphism), this formula collapses exactly to the direct product rule (h1h2,k1k2)(h_1h_2,k_1k_2). So the direct product criterion's requirement that KK be normal (not just HH) is precisely what forces the trivial-action, "fully symmetric" case rather than a genuine semidirect twist.

If ∣G∣=∣H∣⋅∣K∣|G| = |H|\cdot|K|, H∩K={e}H\cap K=\{e\}, and HK=GHK=G, is HH automatically normal?

No, not in general — normality is a separate hypothesis that must be checked, as the S3S_3 counterexample demonstrates: there ∣S3∣=∣H∣∣K∣=3⋅2|S_3| = |H||K| = 3\cdot 2, H∩K={e}H \cap K = \{e\}, and HK=S3HK = S_3, yet KK fails to be normal and the direct product conclusion fails. However, if HH has index equal to ∣K∣|K| and that index is 22, index-22 subgroups are automatically normal (a distinct, narrower fact); this coincidence is why H=⟨(1 2 3)⟩H=\langle(1\,2\,3)\rangle in the counterexample happens to be normal even though we did not separately verify it via a general "automatic normality" theorem for all HK=GHK=G situations.

Can HH and KK overlap nontrivially and still give something related to a direct product?

If H,K⊴GH, K \trianglelefteq G and HK=GHK = G but H∩K≠{e}H \cap K \ne \{e\}, the elementwise commuting conclusion (part (a)) still holds, since its proof only used normality of both subgroups, not trivial intersection. However uniqueness of the factorization fails, so GG is not isomorphic to H×KH \times K directly; instead one gets G≅(H×K)/DG \cong (H \times K)/D where D={(x,x−1):x∈H∩K}D = \{(x,x^{-1}) : x \in H\cap K\} is a "diagonal" subgroup encoding the redundancy, a fact provable using the Isomorphism Theorems.

Before this

Where this leads