Subgroups
Once you know what a group is, the very next question is: do groups hide smaller groups
inside them? They do — constantly. The even integers sit inside all the integers; the pure
rotations of a polygon sit inside its full set of rotations-and-flips; a single "do nothing" move
sits inside every group there is. Each of these is a subgroup — a subset that is
not just along for the ride, but is a fully-fledged group in its own right, using the exact same
operation as its parent.
This is one of the most powerful moves in algebra. Instead of studying a big, complicated group
all at once, we break it into the little groups living inside it and study those — the way you
understand a machine by understanding its parts. Whether a subset counts as a subgroup is not a
matter of taste: there is a crisp checklist, and this page is about learning to run it quickly.
(If the word "group" itself is still hazy, revisit the
examples of groups
first — everything below assumes you have a group already in hand.)
What exactly is a subgroup?
Let G be a group with operation \ast and
identity e. A subset H \subseteq G is a
subgroup, written H \le G, when H
is itself a group under the very same operation \ast inherited
from G. That means H has to satisfy all the
group axioms on its own:
- Identity: e \in H.
- Closure: if a, b \in H then a \ast b \in H (you can't escape H by combining its members).
- Inverses: if a \in H then a^{-1} \in H.
Associativity comes for free — it is already true everywhere in G, so it
is automatically true on the smaller set H. That is the whole reason
subgroups are cheap to check: you never re-verify associativity, only that the subset is
closed, has the identity, and holds its inverses.
The subgroup tests — the checklist made short
Nobody wants to verify three separate conditions every time. Two theorems compress the work.
A nonempty subset H \subseteq G is a subgroup if and only if:
- H is closed under the operation: a, b \in H \implies a \ast b \in H; and
- H is closed under inverses: a \in H \implies a^{-1} \in H.
Why does "nonempty" replace "contains the identity"? Because it hands the identity to you for free:
pick any a \in H (there is one, since H is
nonempty); inverses give you a^{-1} \in H; closure then gives you
a \ast a^{-1} = e \in H. So the identity is a consequence, not a
separate thing to check.
A nonempty subset H \subseteq G is a subgroup if and only if
a, b \in H \implies a \ast b^{-1} \in H.
That single condition secretly bundles closure and inverses together.
The one-step test is the connoisseur's tool: one line does everything. Take a = b
and it hands you a \ast a^{-1} = e \in H (the identity). Take
a = e and it hands you e \ast b^{-1} = b^{-1} \in H
(inverses). Once you know inverses live in H, replacing
b by b^{-1} turns
a \ast b^{-1} back into a \ast b — so you get
closure too. Everything from one clean inequality-free line.
Worked example: the even integers 2\mathbb{Z} \le \mathbb{Z}
Take the group (\mathbb{Z}, +) and the subset
2\mathbb{Z} = \{\dots, -4, -2, 0, 2, 4, \dots\} of even integers. Here the
operation is +, so "a \ast b^{-1}" reads as
a - b (the inverse of b under addition is
-b). Run the one-step test: take two evens
a = 2m and b = 2n. Then
a - b = 2m - 2n = 2(m - n),
which is even, hence back in 2\mathbb{Z}. The set is nonempty (it holds
0), so the single line is passed and 2\mathbb{Z} \le \mathbb{Z}.
Exactly the same argument shows k\mathbb{Z} \le \mathbb{Z} for
every integer k — and, remarkably, these turn out to be
the only subgroups of \mathbb{Z} there are.
Worked example: \{0, 2, 4\} \le \mathbb{Z}_6
Now a finite one. The group \mathbb{Z}_6 = \{0,1,2,3,4,5\} adds modulo
6. Is the subset of even residues
H = \{0, 2, 4\} a subgroup? The cleanest way to see closure is to
read it straight off the group's Cayley table — the addition table where cell
(i, j) holds (i + j) \bmod 6. Reveal the steps
below: the 3 \times 3 block sitting where both the row and the column
belong to \{0, 2, 4\} never once produces an odd number — every entry
stays inside H. That visual "the block stays inside itself" is
closure.
Closure holds (the block stays put), inverses hold (0 pairs with
0, and 2 pairs with 4
since 2 + 4 = 6 \equiv 0), and the set is nonempty — so
\{0, 2, 4\} \le \mathbb{Z}_6. Notice its size, 3,
divides the size of the whole group, 6. That is not a coincidence, as we
will see in a moment.
Worked example: rotations inside a symmetry group C_n \le D_n
The dihedral group D_n is all
2n symmetries of a regular n-gon:
n rotations and n reflections. The rotations
alone,
C_n = \{\, r^0, r^1, r^2, \dots, r^{n-1} \,\},
form a subgroup. Compose two rotations and you get another rotation
(r^i \ast r^j = r^{(i+j) \bmod n}) — closure; the inverse of a rotation is
the rotation the other way ((r^i)^{-1} = r^{n-i}) — inverses; and
r^0 is the identity. So C_n \le D_n, a group of
size n living inside one of size 2n. The
reflections, by contrast, do not form a subgroup — but that is a trap worth
its own box below.
The two subgroups every group owns
Two subgroups come free with every group G, no work required:
-
the trivial subgroup \{e\} — just the identity. It is
closed (e \ast e = e), it holds its inverse
(e^{-1} = e), and it is nonempty. The smallest possible subgroup.
-
the whole group G itself — every group is a subgroup of
itself. The largest possible subgroup.
Any subgroup that is neither of these is called a proper, nontrivial subgroup — and
those are where the interesting structure lives. Ordering all of a group's subgroups by inclusion
draws its subgroup lattice. For \mathbb{Z}_6 it is a tidy
diamond: at the bottom the trivial \{0\}; above it the two proper
subgroups \{0, 3\} (size 2) and
\{0, 2, 4\} (size 3); and at the top all of
\mathbb{Z}_6.
\{0\} \;\subset\; \{0,3\},\ \{0,2,4\} \;\subset\; \mathbb{Z}_6.
Read off the sizes — 1, 2, 3, 6 — and every one of them divides
6. That pattern is no accident.
Because a subgroup H tiles its parent group perfectly.
Slide H around by multiplying every element by a fixed
g, giving the set gH = \{\, g \ast h : h \in H \,\} — a
coset. Two beautiful facts hold: every coset has exactly the same number of elements as
H (they are just H shunted along), and any two
cosets are either identical or completely disjoint — no partial overlaps. So the cosets chop
G into equal-sized, non-overlapping tiles, each the size of
H.
If there are k such tiles, then
|G| = k \cdot |H| — so |H| must divide
|G|. That is Lagrange's theorem, and it explains at a
glance why \{0, 2, 4\} (size 3) can live inside
\mathbb{Z}_6 (size 6) but a subgroup of size
4 or 5 never could. The subgroup you found
was, in effect, one tile of a floor.
The classic blunder is to check closure, see it pass, and declare victory — forgetting the
identity and inverses. In an infinite group closure alone is not enough. Take the
natural numbers \mathbb{N} = \{0, 1, 2, \dots\} sitting inside
(\mathbb{Z}, +). Add two naturals and you get a natural — perfectly
closed! But 3 has no additive inverse in
\mathbb{N} (there is no natural number -3), so
\mathbb{N} is not a subgroup of
\mathbb{Z}. Closure got you nowhere on its own.
The reflections of D_n fail even harder: they miss the identity entirely
(the "do nothing" move is a rotation, not a reflection), and composing two reflections gives a
rotation — so they are not even closed.
There is one place the shortcut is genuinely safe: in a finite group, a nonempty
subset that is closed under the operation is automatically a subgroup. Why? Keep multiplying
a by itself: a, a^2, a^3, \dots all stay inside
by closure, but there are only finitely many elements, so the powers must eventually repeat — and
that forces both the identity and a^{-1} to appear among them. Finiteness
does the extra work for you; infinity does not. Never transplant the finite shortcut into an infinite
group.