A set of products is not automatically a subgroup
Definition 5.1.1
(Product of two subgroups)
Let be subgroups of a group . The product set is
The order matters in a nonabelian group: and may differ. Even though both factors are subgroups, need not be closed under multiplication. We will first count this set when the subgroups are finite and then identify exactly when it is a subgroup. The set is a product inside one group; it is not automatically the external direct product .
Count the fibers, not the expressions
Proposition 5.1.2
(Size of a product set)
If and are finite subgroups of , then
Proof
(Every element has exactly representations)
The multiplication map , , is onto by definition, but different pairs can represent the same element. Fix one representation . Every gives another:
Conversely, if , then belongs to both and , and , . These choices of are distinct because implies . Thus every fiber has exactly pairs, and dividing by the common fiber size gives the formula.
This counts a set whether or not it is a subgroup; the formula does not itself prove closure.
Worked example 5.1.3
(A sum of subgroups of )
In the additive group , let and . Their intersection is , so . Listing the sums gives
the subgroup of even residue classes. Addition is commutative, so and the closure criterion below also applies. This is an internal sum, not a claim that every subgroup product is a direct product.
Worked example 5.1.4
(The degenerate case )
If , then and the formula gives . This agrees with the direct description: , since a subgroup is closed under products and contains the identity. More generally, whenever one finds and .
Quick check
(If finite has 8 elements, has 6, and their intersection has 2, how many elements are in ? Does this prove that is a subgroup?)
Apply the counting formula, then recall what it does not decide.
Solution · Answer
elements. The count alone does not prove closure.
Closure is equivalent to reversing the order
Theorem 5.1.5
(Closure criterion for a product of subgroups)
For subgroups , the product set is a subgroup of if and only if . No finiteness assumption is needed.
Proof
(Inverse set and swapping the middle factor)
If is a subgroup, it equals its inverse set, and
Conversely, suppose . The set contains the identity. For and , the middle product belongs to , so it can be written . Then
Also , so inverses remain in the set. Thus is a subgroup.
Proof X-ray
Where the hypothesis $HK=KH$ is used
The key move is the single rewriting of the middle product as : that is the only place where the order of the two subgroups has to be reversed, and it is exactly what provides. Closure under inverses uses the other inclusion, because and we need it to land back in . For the forward direction, nothing is computed: a subgroup is its own inverse set, which forces . The argument never counts elements, so it is valid for infinite groups as well; the counting formula is a separate fact that applies even when is not a subgroup.
Corollary 5.1.6
(A normal factor suffices)
If is normal in , then , so is a subgroup.
Proof
(Moving a conjugate across)
For and , , giving . Conversely , so . Therefore , and the criterion applies.
Worked example 5.1.7
(Neither subgroup normal, yet the product is a subgroup)
In let and . Neither is normal in : conjugating by gives . But the two transpositions are disjoint and commute, so . This set is a subgroup of order , as the criterion predicts. Normality is a convenient sufficient condition, not part of the criterion.
Counterexample mode
Two subgroups can have a non-subgroup product
Let be the ten symmetries of a regular pentagon, with and . Take and . Both are subgroups because , but
The element is in , while its square is not: the five rotations are distinct, and reflections are not rotations. Thus is not closed. The counting formula still gives because the subgroups meet only at the identity. As a further check, does not divide , so by Lagrange's theorem could not be a subgroup.
Common mistake
(The counting formula does not decide closure)
The formula is valid for every pair of finite subgroups, so it cannot distinguish the case where is a subgroup from the case where it is not. In the pentagon example it returns 4 although is not a subgroup. To test closure, use (or the divisibility check as a quick necessary condition).
Quick check
(In the pentagon example, why is ?)
Compare with each of the four listed elements.
Solution · Answer
It is neither nor , and it is a rotation rather than either listed reflection or .
Quick check
(In let and . How many elements does have, and is it a subgroup?)
Count with the formula, then compare with .
Solution · Answer
, so , namely with under rightmost-first composition. Since does not divide , Lagrange's theorem shows that is not a subgroup.