Вы находитесь на странице: 1из 34

1

§2 Subgroups, homomorphisms, and quotients

So far, we’ve only studied properties of a single group H. In this section we study the
relationships between multiple groups.

• Subgroups: D3 is the set of rigid motions of a triangle, but there is a smaller


group H consisting of only rotations only. Since rotations are rigid symmetries,
we have H ⊆ D3 , so H is called a subgroup of D3 . Another example: Z is a
subgroup of R under addition.

• Homomorphisms: “Homos” and “morphe” are Greek for “same” and “shape”,
so a homomorphism is special type of function that transforms a group while
preserving its structure. We’ll also see isomorphisms which identify when two
groups have the same Cayley table.

• Quotients: In Z9 , you’ve seen that 5 + 7 ≡ 12 ≡ 3. Why is it that the addition


in Z9 is the same as in Z, with only difference being the “folding” mod 9? This
“folding” is a construction called a quotient that can be performed on any group.

2.1 Subgroups
Let G be a group, and let H be a nonempty subset of G. Then H is called a subgroup
of G, denoted H ≤ G, if it has the following two properties:

(i) [Closed under multiplication] If a, b ∈ H then ab ∈ H.

(ii) [Closed under inverses] If a ∈ H, then a−1 ∈ H.

If H is a subgroup of G, then H is a group too. Many of the groups we’ve encountered


have already been subgroups of each other.

Example 1. Show that Z is a subgroup of R under addition.

Example 2. Let 2Z = {2n : n ∈ Z} be the set of even integers. Show that 2Z is a


subgroup of Z under addition.

Example 3. Let 2Z + 1 = {2n + 1 : n ∈ Z} be the set of odd integers. Show that


2Z + 1 is not a subgroup of Z.
As a fun side note: we saw that 2Z is a subgroup but 2Z + 1 isn’t. What about
2Z + 2? ... Wait, that’s the same as 2Z. And 2Z + 3 = 2Z + 1. This looks eerily like
addition in Z2 , doesn’t it? There’s a quotient at play here.

Example 4. Let H := {1, −1}, which is a group under multiplication. Notice that H
is a subset of Z, but it’s not a subgroup of Z: while H is a group under multiplication,
Z is a group under addition. Lesson learned: a subgroup is not just a group inside
another group — they have to have the same operation.

Example 5. In S3 , let τ be the cyclic permutation


 
1 2 3
τ= .
2 3 1

Let H := {1, τ, τ 2 } and K := {1, τ }. Show that H is a subgroup of S3 , while K is not.


2

Example 6. If G is a group, then G ≤ G and {1} ≤ G. Do you see why?

Example 7. Let T be the unit circle in the complex plane (under multiplication), and
let E be the set of all roots of unity in C. Show that E is a subgroup of T .

By now you’ve realized that to check that H is a subgroup, you have to check that
it’s nonempty, in addition to the two closure properties. Here is an alternate way to
check that a subset is a subgroup.

Proposition 8. Let G be a group and let H ⊆ G. Then H ≤ G if and only if:

(a) 1 ∈ H; and

(b) If a, b ∈ H, then ab−1 ∈ H.

Proof. “=⇒”: Suppose H ≤ G. We will prove that H has properties (a) and (b).

(a) Since H is nonempty, it has at least one element a ∈ H. But then a−1 ∈ H since
H is closed under inverses, so aa−1 ∈ H since H is closed under multiplication.
But aa−1 = 1, so we’re done.

(b) Let a, b ∈ H. Then b−1 ∈ H since H is closed under inverses, so ab−1 ∈ H since
H is closed under multiplication.

Therefore H satisfies properties (a) and (b).


“⇐=”: Suppose H has properties (a) and (b). Then H is nonempty because 1 ∈ H.
We have to show that it is closed under (i) multiplication, and (ii) inverses. First we
prove (ii), then (i).

(ii) Closed under inverses: Let a ∈ H. Then 1 ∈ H by property (a), so a−1 =


1 · a−1 ∈ H by property (b).

(i) Closed under multiplication: Let a, b ∈ H. Then b−1 ∈ H by (ii) (which we just
proved), so ab = a(b−1 )−1 ∈ H by property (b).

Therefore H is a nonempty subset of G satisfying the two axioms, so H ≤ G. v

NOTE! In the above proof we used the fact that b = (b−1 )−1 . Do you know how
to prove this?

Example 9. Recall the setup of Example 5: the set K = {1, τ } is a subset of S3 , but
it’s not a subgroup. The reason is that it’s not closed under multiplication, because
(for instance) τ ∈ K but τ · τ ∈ / K. What if we just throw in τ · τ ? We get a bigger
set H := {1, τ, τ 2 }, which is indeed a subgroup of S3 .

Example 10. Let S := {−4, −2, 0, 2, 4} ⊆ Z. Then S is not a subgroup: for example
2+4 = 6 ∈ / S. To make it a subgroup, we could throw in 6 like we did above, but
then since a subgroup needs inverses, we’d have to throw in −6 too. Don’t forget
2 + 6 = 8 and 4 + 6 = 10. ... OK, so we need to throw in every even integer to make
it a subgroup. And indeed, 2Z (= the set of evens) is a subgroup of Z!
3

Any subset of a group G can be enlarged to obtain a subgroup — this is called


generating a subgroup. Usually we’d like to do this by adding in as few elements as
possible, but like we saw above, sometimes infinitely many are required. In general, let
G be a group and let S be a subset of G. If H is the smallest subgroup of G such that
S ⊆ H, then H is called the subgroup generated by S and is denoted H = hSi.
The set S is called a generating set for H.
Generating subgroups is similar to what you’ve seen before in vector spaces: given
a set S, you can take span(S) to get the smallest subspace containing S.

Example 11. If τ is the permutation described in Example 5, then hτ i = {1, τ, τ 2 }.

Example 12. In Z under addition, h2i = 2Z. Here are some more examples:

h−3, 6i = 3Z, h0i = {0}, h1i = Z.

Exercise: show that h2, 5i = Z. More generally, show that if m, n are coprime, then
hm, ni = Z.

Example 13. Let Zn be the integers mod n, under addition. Then Zn = h1i.

As we saw above, sometimes you can start with a finite set S, but it generates an
infinite subgroup. So how do you know when to stop adding elements? Wait, how
do you know that hSi even exists? This is a common situation in mathematics, and
we’ll see it many times in this course: even once you define an object, it’s important
to prove that it is well-defined.

Proposition 14. Let G be a group and let S ⊆ G. Then there exists a smallest
subgroup of G which contains S.

Proof sketch. So we’re looking for a subset H of G with the following properties: (a) H
is a subgroup of G, (b) H contains S, and (c) if H 0 is another subgroup of G containing
S, then H ⊆ H 0 . (Property (c) is what it means for H to be the “smallest”!)
Let X be the set of all subsets of G satisfying (a) and (b):

X := {X ⊆ G : S ⊆ X and X ≤ G}.

Then X is nonempty because G ∈ X (why?). So we can take the intersection of all


elements of X: \
H := X.
X∈X

Now it is left as an exercise for you to understand what H is, and show that it has
properties (a), (b), and (c). v

2.2 Cyclic groups


We saw subgroups generated by two or three elements. Now we investigate subgroups
generated by a single element.

Proposition 1. Let G be a group and let a ∈ G. Then

hai = {am : m ∈ Z}.


4

Proof. Let H := {am : m ∈ Z}; we have to show that H is the smallest subgroup of G
containing a. Clearly H contains a, since a = a1 ∈ H, so it’s nonempty.
Why is H a subgroup? We just have to check it’s closed under multiplication and
inverses.

(i) Closed under multiplication: Let x, y ∈ H. Then x = am and y = an for some


m, n ∈ Z, so xy = am+n ∈ H.

(ii) Closed under inverses: Let x ∈ H. Then x = am for some m ∈ Z, so x−1 =


a−m ∈ H.

This shows that H is a subgroup of G containing a — all that’s left is to show that
it’s the smallest one.
OK, let H 0 be a subgroup of G such that a ∈ H 0 ; we must show H ⊆ H 0 . Let x ∈ H.
Then x = am for some m ∈ Z. If m = 0 then x = 1 ∈ H 0 , so we’re done. If m is positive,
then since a ∈ H 0 and H 0 is closed under multiplication, we get x = a · · · a ∈ H 0 . If m
is negative, then since a−1 ∈ H 0 , we similarly get x = a−1 · · · a−1 ∈ H 0 . In all three
cases we see that x ∈ H 0 , so this shows H ⊆ H 0 , as required. v

Below we will see that hai can be finite or infinite.

Example 2. In Z8 , consider the subgroup generated by 2. Since we’re using additive


notation, we only need multiples of 2, so

h2i = {2n (mod 8) : n ∈ Z} = {0, 2, 4, 6}

since multiples of 2 cycle around mod 8.

Example 3. In Z, consider the subgroup generated by 2. Again since we’re using


additive notation, we get

h2i = {2n : n ∈ Z} = {. . . , −6, −4, −2, 0, 2, 4, 6, . . .}

which is an infinite set!

Example 4. In GL2 (R), let A and B be the matrices


   
1 0 1 2
A := , B :=
0 −1 0 1

Since A has order 2, the subgroup it generates is hAi = {I, A}. Here’s the Cayley
table:
I A
I I A
A A I

Looks like this subgroup is isomorphic to Z2 . How about hBi? Show that
 
n 1 2n
B = ,
0 1

and conclude hBi is infinite.


5

Let G be a group. Then G is called a cyclic group if there exists a ∈ G such that
G = hai. The element a is called a generator of G.
Example 5. Z8 is a cyclic group because Z8 = h3i. Notice also Z8 = h1i = h5i — so
generators aren’t unique.
Example 6. Z is a cyclic group because Z = h1i. The subgroup 2Z is also cyclic,
because 2Z = h2i.
In fact every subgroup of Z is cyclic. Why? Let H be a subgroup of Z; we want to
show H = hxi for some x ∈ Z. If H = {0} then H = h0i and we’re done. If H 6= {0},
then H has at least one nonzero element, but in fact you can show it has at least one
positive element. Prove it! Now let x be the least positive integer such that x ∈ H.
Claim: H = hxi.
I’ll leave “⊇” as an exercise for you, you should think about it. For “⊆”, let a ∈ H. We
want to show a is a multiple of x. By the division algorithm we can write a = qx + r
with 0 ≤ r < x. If r = 0, then a = qx ∈ hxi and we’re done; if r > 0, then

r = qx − a

so r ∈ H since H is a subgroup. But r < x, so this contradicts the choice of x as the


least positive element of H. So the only possibility is r = 0, so a = qx ∈ hxi. This
proves the claim, and we’re done!
If you really stare at the above proof and convert it to multiplicative notation, you’ll
see there’s nothing special about Z. You can copy that proof to prove the following
fact:
Proposition 7. A subgroup of a cyclic group is cyclic.
Moving on ... We’ve seen that cyclic groups can be either finite or infinite. Whether
or not the group hai is infinite depends on ord(a). For example, if ord(a) = 3, then
a4 = a3 a1 = a, a17 = (a3 )5 a2 = a2 , etc. — so we can get higher powers of a in terms
of just 1, a, a2 . Therefore hai = {1, a, a2 } only has three elements. More generally, we
have the following fact:
Proposition 8. Let G be a group and let a ∈ G. Then ord(a) = |hai|.
Proof. It’s possible that ord(a) is infinite, but I’ll leave that case as an exercise for
you. Instead assume ord(a) = n is finite. Then by Proposition 1.5.7, the elements
1, a, a2 , . . . , an−1 are distinct.
Claim: hai = {1, a, a2 , . . . , an−1 }.
If we have this claim we’re done, because the RHS has n elements, so we’d have
|hai| = n.
OK, so let’s prove the claim. Clearly “⊇” holds because each power of a belongs
to hai. For “⊆”, let ai be an arbitrary power of a with i ∈ Z. Like we saw in the n = 3
above case, we have to show that ai is amongst powers of a between 0 and n − 1. But
by the division algorithm, we can write i = qn + r with 0 ≤ r < n, and thus

ai = aqn+r = (an )q ar = 1q ar = ar

which belongs to the RHS because 0 ≤ r < n. QED! v


6

Now that we’ve studied groups with one generator, what about two generators?
Let’s say G = hx, yi. Then in general we’d have to include elements like xy, yx,
xyx, x2 y 3 x−23 y 4 x−1 , etc. You can imagine what three generators looks like! What’s
interesting is that a subgroup of a 2-generated group may not be 2-generated.

2.3 Lagrange’s Theorem


In this section we tackle the first important theorem in group theory, Lagrange’s The-
orem. I would call it the fundamental theorem of finite groups, and others would even
say it’s a cornerstone theorem of all of algebra. It states that if G is a group and H is
a subgroup of G, then |H| divides |G|. The proof of this theorem takes some work, so
here is an illustration before we start.

Example 1. Here is the Cayley table for a group G of order 6.

1 a b x y z
1 1 a b x y z
a a b 1 z x y
b b 1 a y z x
x x y z 1 a b
y y z x b 1 a
z z x y a b 1

The set H := {1, a, b} forms a subgroup of order 3, which is evident from the top-left
3 × 3 corner of the table. It separates the table evenly into four sections:

1 a b x y z
1 1 a b x y z
a a b 1 z x y
b b 1 a y z x
x x y z 1 a b
y y z x b 1 a
z z x y a b 1

Notice how opposite corners have matching symbols: the top-left and bottom-right are
{1, a, b}, and the top-right and bottom-left are {x, y, z}. The fact that H divides the
table so evenly is a consequence of Lagrange’s Theorem.
You can prove that G has no subgroups of order 4 — that’s because 4 is not a
divisor of 6. Lagrange’s Theorem is at play!

The proof of Lagrange’s Theorem is very tidy: we want to show that |H| divides
|G|, so if we split G into a disjoint union of five sets of size |H|, then |G| = 5|H|. QED!
Proving this rigorously requires the introduction of some basic concepts that we will
use throughout the course.

NOTE! Now is a good time to read the notes on equivalence relations — it’s
important!

Let G be a group, and let H be a subgroup. We define a relation on G as follows:

a∼b ⇐⇒ a−1 b ∈ H.
7

Proposition 2. The above relation ∼ defines an equivalence relation on G.

Proof. You can prove it! Show that ∼ is reflexive, symmetric, and transitive. v

Since you know about equivalence relations (!!), you know that the equivalence
classes partition G into disjoint subsets. Here’s the equivalence class:

[a] = {b ∈ G : a ∼ b} = {b ∈ G : a−1 b ∈ H}.

This is a bit opaque, but there’s a nice way to write it down. Notice a−1 b ∈ H if and
only if a−1 b = h for some h ∈ H, if and only if b = ah. This shows that the equivalence
class [a] is equal to the following set:

aH := {ah : h ∈ H}.

This is called a left coset of H. The punch line of this is: the equivalence classes of
∼ are exactly the left cosets of H. Notice that H itself is always a left coset, since
H = 1 · H = [1]. So if a ∈ G, then a ∈ H if and only if a ∼ 1.

Example 3. In S3 , let τ be the permutation that swaps 1 with 2 and fixes 3, and let
H := hτ i = {1, τ }. What are the left cosets of H? Here’s one: [1] = H of course. Any
element not in H would give another left coset, so we can let σ be the cycle 1 → 2 → 3
and get the left coset σH. How about σ 2 H? This is still different from the others,
since σ 2 ∈
/ H and σ 2 ∈/ σH (because otherwise σ −1 σ 2 = σ would be in H). So we have
at least three distinct left cosets

H = {1, τ }, σH = {σ, στ } σ 2 H = {σ 2 , σ 2 τ }.

These are actually all of them, and to see this, you just have to check that the above
six elements are all the elements of S3 . Therefore

S3 = H t σH t σ 2 H

Notice that all the cosets have the same size!

Example 4. Consider the group Z with subgroup 6Z = {6n : n ∈ Z}. What does the
equivalence relation look like? We know a ∼ b if and only if b − a ∈ 6Z, if and only if
a ≡ b (mod 6). The equivalence classes are [a] = a + 6Z. So we can actually list out
all the cosets:
[0], [1], [2], [3], [4], [5].

Look familiar? It’s Z6 ! Note Lagrange’s Theorem is not in play here, because Z is an
infinite group — but there are still only finitely many cosets. You should also notice
that since every integer has a remainder mod 6, we get that

Z = [0] t [1] t [2] t [3] t [4] t [5].

(Recall that “t” means “disjoint union”.) Notice that [2] = [7] = [13] = [19] = · · · ,
so even though there are infinitely many ways to write each coset, there are still only
finitely many distinct cosets. THAT is a good lesson to keep in mind and never forget.
8

If H ≤ G and C is a left coset of H, then any element a ∈ G such that C = aH is


called a coset representative of C. Any element a ∈ C will be a coset representative
of C. In the above example, [2] is a coset with representatives 7, 13, 19, or anything
equivalent to 2 mod 6. Usually we fix a set of representatives of all the cosets — in
that example, we picked {0, 1, 2, 3, 4, 5}, but you could have picked {12, 7, −4, 9, 16, 11}
if you really wanted to.
Example 5. It’s also possible for a subgroup to have infinitely many cosets. Example?
If G is a finite group and H ≤ G, then H can only have finitely many distinct
cosets in G. If a1 H, . . . , an H are all the distinct cosets, then they form a partition of
G, and therefore
G = a1 H t · · · t an H.
Lemma 6. Let G be a group and let H ≤ G. Then any two left cosets of H have the
same size.
Proof. Let aH and bH be two left cosets. To show that these have the same size, we
have to write down a bijection between them. Define a function f : aH → bH as
follows:
f (x) := ba−1 x.
First, why is f (x) ∈ bH? Since x ∈ aH, we get a−1 x ∈ H, so f (x) = ba−1 x ∈ bH. So
f is well-defined. We will show that f is a bijection by showing that it is injective and
surjective.

(i) Injective: Suppose that f (x) = f (x0 ) for some x, x0 ∈ X. Then ba−1 x = ba−1 x0 ,
so canceling off the ba−1 , we get x = x0 . So f is injective.

(i) Surjective: Let y ∈ bH be any element. Then b−1 y ∈ H, so x := ab−1 y is in aH,


and
f (x) = ba−1 x = ba−1 ab−1 y = y.
Thus f hits y, so it’s surjective.

This shows that f is a bijection between the cosets aH and bH, which means that
they have the same size. v

Since all cosets have the same size, in particular they all have the same size as the
coset [1] = H — so every coset is size |H|. Now we can prove Lagrange’s Theorem!
Theorem 7 (Lagrange’s Theorem). Let G be a finite group and let H ≤ G. Then |H|
divides |G|.
Proof. Let a1 , . . . , ad be the distinct left coset representatives of H. Then by Lemma
5, all cosets ai H have size |H|, and G = a1 H t · · · t ad H. So,

|G| = |a1 H| + · · · + |ad H| = |H| + · · · + |H| = d|H|.

This shows |H| divides |G|, as required! v

The number of distinct left cosets in H is called the index of H in G, denoted by


[G : H]. So for example [Z : 6Z] = 6, and in the above proof we had [G : H] = d. So
we actually showed something more specific than Lagrange’s Theorem!
9

Corollary 8 (of the proof). Let G be a finite group and let H ≤ G. Then

|G| = [G : H]|H|.

In particular we can see that the index of H also divides |G|.


Another corollary of Lagrange’s Theorem is that if G is a group of order 9, then it
can’t have any element of order 2. This is because if ord(a) = 2 then the order of the
subgroup hai would have order 2, which doesn’t divide 9!

Corollary 9. Let G be a finite group and let a ∈ G. Then ord(a) divides |G|.

Proof. We know that ord(a) is equal to the order of the subgroup hai, which must
divide |G| by Lagrange’s Theorem. v

Corollary 10. Let G be a finite group of order n. Then an = 1 for all a ∈ G.

Proof. Let k = ord(a). By Corollary 8 we know n = dk for some d ∈ Z. But then

an = (ak )d = 1. v

The converse of Lagrange’s Theorem states that if n is a divisor of |G|, then G


has a subgroup of order n. This is actually not true in general, but we won’t see a
counterexample till later. However, the converse of Lagrange’s Theorem is true for
abelian groups, and we’ll prove it in this course! For now we content ourselves with
the following lil’ nugget.

Corollary 11. A group of prime order is cyclic.

Proof. Let G be a group of prime order p. Since p ≥ 2, we know G has at least two
elements, so we can choose some a ∈ G \ {1}. Now what can we say about hai, the
subgroup generated by a? By Lagrange’s Theorem, |hai| divides p. But p is prime and
|hai| > 1 — the only possibility is |hai| = p. But then hai has the same number of
elements as G, which forces G = hai.
Notice we’ve proved something stronger — not only is G cyclic, but it’s generated
by any nontrivial element. v

Here’s one more application, something you might remember from Math 135. We
can prove it with group theory!

Corollary 12 (Fermat’s Little Theorem). Let a ∈ Z and let p be prime. Then

ap ≡ a (mod p).

Proof. If a ≡ 0 (mod p) there is nothing to prove. So suppose that a is nonzero mod


p, and let G := Zp \ {0}. Then G is a group under multiplication, and it has order
p − 1. So by Corollary 10 we have ap−1 ≡ 1 (mod p). Multiplying by a gives ap ≡ a
(mod p) as required. v
10

2.4 Homomorphisms
In Greek, “homos” means “same” and “morphe” means shape or structure. A ho-
momorphism is a way to transform a group while keeping the same structure. For
instance, consider the translation function

t : {one, two, three, four} → {un, deux, trois, quatre}

that takes a number and translates it to French: for example so t(one) = un, t(two) =
deux. The meaning of numbers and the concept of counting are the same — for
example t(one + one) = un + un — the only difference is that it’s translated to French.
So t preserves meaning while changing terminology. Here’s another example: consider
the English-to-French translation function

t : {big, large, horse} → {grand, cheval}.

It still preserves meaning, for example t(big horse) = grand cheval, but you also lose
some information because t(big) = t(large). Translation functions aren’t always injec-
tive, but they must preserve meaning.
Now let’s define things rigorously. Let G, H be two groups and let f : G → H be
a function. Then f is called a homomorphism if

f (ab) = f (a)f (b) for all a, b ∈ G.

Take a look at this definition and make sure it’s sensible: the product on the LHS
is multiplication in G, while the product on the RHS is in H. This property is just
saying that f is a transformation that preserves the multiplication operation. People
often say that homomorphisms are structure-preserving transformations.
Here are another two ways that homomorphisms preserve structure.

Proposition 1. Let f : G → H be a homomorphism. Then

(i) f (1) = 1.

(ii) f (x−1 ) = f (x)−1 for all x ∈ G.

Proof. (i) Observe that

f (1) = f (1 · 1) = f (1)f (1) = f (1)2

so by cancellation we get f (1) = 1.


(ii) Let y := f (x−1 ). We need to show that y is the inverse of f (x). Well,

f (x)y = f (x)f (x−1 ) = f (xx−1 ) = f (1) = 1

since f (1) = 1 by part (i). So y = f (x)−1 . v

Before doing examples, we first point out two important sets associated to a ho-
momorphism f : G → H.

• The kernel of f is ker(f ) := {a ∈ G : f (a) = 1}.

• The image of f is im(f ) := {f (a) : a ∈ G}.


11

So ker(f ) is a subset of G, and im(f ) is a subset of H. In fact, even more is true:


Proposition 2. Let f : G → H be a homomorphism. Then ker(f ) is a subgroup of G
and im(f ) is a subgroup of H.
Proof. Since f (1) = 1 we have 1 ∈ ker(f ). If a, b ∈ ker(f ), then f (ab) = f (a)f (b) =
1 · 1 = 1, so ab ∈ ker(f ). So ker(f ) ≤ G. The proof for im(f ) is similar, so I’ll leave it
to you! v

Example 3. Consider Z under addition. Show that the function f : Z → Z given by


f (x) = 5x is a homomorphism. What is its kernel? What is its image?
Example 4. Recall that R× is the multiplicative group of nonzero real numbers. Show
that the determinant det : GLn (R) → R× is a homomorphism. What is its kernel?
What is its image?
Example 5. Recall that R>0 is a multiplicative group and R is an additive group.
Show that the logarithm function ln : R>0 → R is a homomorphism. What is its
kernel? What is its image?
Example 6. Let f : R× → R>0 be the function f (x) = x2 . Show that this is a
homomorphism. What is its kernel? What is its image?
Example 7. Let f : Z → Z be the function f (x) = x2 . Show that this is NOT a
homomorphism.
Example 8. Let T be the unit circle in the complex plane, which is a multiplicative
group. Show that the complex conjugation function T → T , z 7→ z is a homomorphism.
What is its kernel? What is its image?
Example 9. For any n ≥ 1, the “reduction mod n” map Z → Zn is a homomorphism.
More curiously, define a map f : Z10 → Z5 by [a]10 7→ [a]5 — here [a]n just denotes the
equivalence class mod n. Show that f is a well-defined function, and a homomorphism!
What is its kernel? What is its image?
Example 10. Since D4 consists of symmetries of the square, we can think of these as
permutations of the corners of the square. Label the corners clockwise by {1, 2, 3, 4}.

Let ϕ : D4 → S4 be the function sending a symmetry to its corresponding permutation.


For example, if r is a clockwise rotation by 90◦ , then
 
1 2 3 4
ϕ(r) = .
2 3 4 1

And if f is a flip along the vertical axis, then


 
1 2 3 4
ϕ(f ) =
2 1 4 3

Show that ϕ is a homomorphism. What is its kernel? What is its image?


12

Example 11. Let G be any group. Show that the identity map id : G → G is a group
homomorphism. What is its kernel? What is its image?
Example 12. Let G, H be any groups and let f : G → H be the function given by
f (x) := 1 for all x ∈ G. Show that f is a homomorphism — it’s called the trivial
homomorphism. What is its kernel? What is its image?
Example 13. Let G be any group and fix g ∈ G. Define a map σg : G → G by
σg (x) := gxg −1 . Show that σg is a homomorphism and a bijection — this is called an
inner automorphism of G. What is its kernel? What is its image?
Example 14. Let G, H be two groups and let G × H be their direct product. Let π :
G × H → G be the map given by π(g, h) := g. Show that π is a group homomorphism.
It’s called the projection onto the first coordinate. What is its kernel? What is
its image? There is also a projection onto the second coordinate worth thinking about.
There are lots more homomorphisms out there, and they will be an incredibly
useful tool throughout this course.
Think of it this way: a homomorphism is a way to preserve the structure of a group
while transforming it into another. Just look at English–French translation, where you
can translate “big” and “large” to “grand”. Sometimes you lose information — e.g.
the trivial homomorphism — and sometimes you completely preserve information —
e.g. the map D4 → S4 and the projection G × H → H. But an important theme we
will see repeatedly is that you can’t gain information from a homomorphism, at least
not directly. A homomorphism f : G → H will transform G into a subgroup of H, and
viewing it as such can be helpful, but you still wouldn’t add any algebraic properties
to G.
Why do we lose information after applying a homomorphism f : G → H? Well, for
instance, if x, y ∈ ker(f ) then f (x) = 1 = f (y). So f can’t tell the difference between
two elements in the kernel!
Proposition 15. Let f : G → H be a homomorphism. Then f is injective if and only
if ker(f ) = {1}.
People say that the kernel is an obstruction to injectivity, or that the kernel mea-
sures injective-ness of a homomorphism. Those people are kind of pretentious but I
guess they’re right.

Proof. “=⇒”: Suppose that f is injective and let x ∈ ker(f ). Then f (x) = 1. But
then f (x) = 1 = f (1), so x = 1 by injectivity. This shows ker(f ) ⊆ {1}, but the
inclusion “⊇” is obvious — so we conclude that ker(f ) = {1}.
“⇐=”: Suppose that ker(f ) = {1}. If f (x) = f (y), then since f is a homomorphism,
we can use Proposition 16 to get

f (xy −1 ) = f (x)f (y)−1 = f (x)f (x)−1 = 1.

This shows that xy −1 is in ker(f ) = {1}, so xy −1 = 1, which implies x = y as


required. v

“Injectivity” is a function-theoretic property, and we have now magically connected


it with the group-theoretic properties of kernels. There are many more such connec-
tions throughout group theory! We conclude this section with two more.
13

Proposition 16. (a) Let f : G → H and g : H → K be homomorphisms. Then


g ◦ f is a homomorphism.

(b) Let f : G → H be a homomorphism. If f is a bijection, then f −1 is also a


homomorphism.

Proof. Exercise. v

2.5 Isomorphisms and automorphisms


In Chapter 1 we defined “isomorphism” between two groups G, H to mean that there
is a way to relabel the elements of G so that its Cayley table is the same as that of H.
In other words, it means that G and H have the same algebraic structure. But now
that we know about homomorphisms, the true definition of isomorphism is ready to
be revealed.
Let G, H be two groups. Then a function ϕ : G → H is called an isomorphism
if it is a homomorphism and a bijection. G is isomorphic to H, denoted G ' H, if
there exists an isomorphism ϕ : G → H.

Example 1. Let ϕ : 9Z → 16Z be given by ϕ(x) := 16 9 x. Show that ϕ is an isomor-


phism, and therefore 9Z ' 16Z. Makes sense — they’re both infinite cyclic groups.
All this isomorphism does is send 9 to 16, i.e. generator to generator.

Example 2. Consider the group G := {1, −1}, which is a subgroup of R× under


multiplication. Then the map ϕ : G → Z2 , given by 1 →
7 0 and −1 7→ 1, is an
isomorphism.

Example 3. R>0 is a group under multiplication, and R is a group under addition.


But they are isomorphic to one another! Indeed, an isomorphism R>0 → R is given
by x 7→ ln(x).

Example 4. Let S3 be the symmetric group on 3 letters, and let D3 be the symmetry
group of an equilateral triangle. We know that S3 and D3 both have order 6, and if
you write down their Cayley tables you’ll see that they’re the same — so these two
groups are isomorphic.
But let’s actually write down an isomorphism. ... ... ...

If G is isomorphic to H, then it should have all the same group-theoretic properties


as H. That means that if G has a subgroup of order 17 and H doesn’t, then there’s
no way they can be isomorphic! Nor if G is abelian while H is not.

Example 5. The groups G := Z2 × Z2 and H := Z4 are both abelian groups of order


4. But H has an element of order 4 while G does not. So they can’t be isomorphic to
each other, can they?
Let’s PROVE it. Suppose that G ' H, so that there exists an isomorphism
ϕ : G → H. Then since ϕ is surjective, there exists some g ∈ G such that ϕ(g) = 1.
But then 2g = 0 (that’s just how G works), so

0 = ϕ(0) = ϕ(2g) = 2ϕ(g) = 2 · 1 = 2

which gives 2 = 0 (mod 4) — NOT true. This contradiction proves that G is not
isomorphic to H, written G '
6 H.
14

If G 6' H, is it possible that H ' G? No way! Since isomorphism is the idea of


“equivalence”, it makes sense that ' defines an equivalence relation. Indeed:
Proposition 6. Let G, H, K be groups.
(a) G ' G.

(b) If G ' H, then H ' G.

(c) If G ' H and H ' K, then G ' K.

Proof. Exercise. v

In the case of cyclic groups, isomorphism is somewhat simplified: two cyclic groups
are isomorphic if and only if they have the same number of elements. This is immediate
from the following proposition.
Proposition 7. Let G be a cyclic group.
(a) If |G| = n is finite then G ' Zn .

(b) If |G| = ∞ then G ' Z.


Proof. (a) Let G = hgi. If |G| = n then ord(g) = n, and we define a homomorphism
ϕ : G → Zn as follows: for x ∈ G, write x = g i for some i ∈ Z, and define ϕ(x) := i.
First we have to check this is well-defined — what if x = g j ? Would that mean i = j?
Well, we know g i = x = g j , so g i−j = 1, which implies that n|(i−j). So i ≡ j (mod n).
Thus ϕ is well-defined.
Now we check that ϕ is a homomorphism: let x, y ∈ G and write x = g i , y = g j .
Then xy = g i+j , so
ϕ(xy) = i + j = ϕ(x) + ϕ(y)
so this is a homomorphism. To see that it’s an isomorphism we just need to see that
it’s a bijection. It’s surjective because ϕ(g i ) = i for any i ∈ Zn . And it’s injective
because its kernel is trivial: indeed, if ϕ(x) = 1, then x = g i with i ≡ 0 (mod n). But
then n|i, so we can write i = nd for some d ∈ Z, and then x = g i = (g n )d = 1. So
ker(ϕ) = {1} and ϕ is trivial.
(b) This one’s a bit easier. Let G = hgi; since G is infinite, ord(g) = ∞. Define a
function ϕ : G → Z by g i 7→ i. Again we first have to show this is well-defined, but
if g i = g j , then g i−j = 1, which contradicts the fact that ord(g) = ∞ unless i = j.
Showing that ϕ is a homomorphism is the same as in part (a), and seeing that it’s a
bijection is pretty simple too. Good luck!
Alternate proof for (b): you could define ϕ : Z → G by n 7→ g n . The isomorphism
goes the other way, but that’s still perfectly valid. Now there’s no trouble with well-
defined-ness! v

Corollary 8. Let G, H be two cyclic groups. Then G ' H if and only if |G| = |H|.
Proof. Why don’t you think about it as an exercise? I did all the leg work already
anyways! v

Corollary 9. Let G be a group of prime order p. Then G ' Zp .

Proof. This is a good one to think about yourself, too. v


15

2.5.1 Automorphisms
Obviously every group is isomorphic to itself, because the identity map G → G is
always an isomorphism. But it’s not the only isomorphism from G to itself. In general,
an isomorphism σ : G → G is called an automorphism of G.
Example 10. R2 is a group under addition. Consider the map σ : R2 → R2 given
by σ(a, b) := (b, a). Show that σ is an automorphism of R2 .
Example 11. Consider the map σ : Z8 → Z8 given by σ(x) := 3x. This is an
automorphism of Z8 . On the other hand, the map τ : Z8 → Z8 given by τ (x) := 4x is
not an automorphism, since it’s not injective: indeed, ker(τ ) = {0, 2, 4, 6} is nontrivial.
The automorphism group of G is the set
Aut(G) := {σ : G → G : σ is an automorphism}.
The reason it’s called the automorphism group is because, well, it’s a group.
Proposition 12. Let G be a group. Then Aut(G) is a group under function compo-
sition.
Proof. If σ, τ are automorphisms, then σ ◦ τ is also an automorphism (read section
2.4), so the operation makes sense. We just have to check that it’s associative, has
identity, and has inverses.
(i) Associative: Function composition is always associative!
(ii) Identity: The identity element is the identity map id : G → G, which is indeed
an automorphism.
(iii) Inverses: If σ : G → G is an automorphism, then so is σ −1 : G → G (exercise),
so that’s its inverse. v
Example 13. What is Aut(Z3 )? Bijections from Z3 to itself are the same as per-
mutations of {0, 1, 2} — but not all of these are homomorphisms! In order for a
bijection σ : Z3 → Z3 to be an automorphism, it at least has to satisfy σ(0) = 0.
Amongst these maps, the only ones that are homomorphisms are the identity map,
and the map σ(x) = 2x. Therefore Aut(Z3 ) = {id, σ}. (Notice σ 2 = id!). The map
ϕ : Aut(Z3 ) → Z2 , given by ϕ(id) = 0 and ϕ(σ) = 1, is an isomorphism — therefore
Aut(Z3 ) ' Z2 .
A better way to think of it: id is multiplication by 1, and σ is multiplication by 2 (mod
3 of course). So essentially Aut(Z3 ) ' {1, 2}, which is a group under multiplication
mod 3. In general, Aut(Zn ) = Z∗n is the group of invertible integers mod n under
multiplication. (Recall from Math 135 that a ∈ Z has an inverse mod n if and only if
gcd(a, n) = 1.)
Example 14. How about Aut(Z2 × Z2 )? Every automorphism fixes (0, 0), so we just
have to look at permutations of the elements a := (1, 0), b := (0, 1), and c := (1, 1). If
τ is the one that swaps a with b, while σ is the one that permutes in a cycle a → b → c,
these are actually both automorphisms. But every automorphism can be generated
from these two: for example, the swap b ↔ c is equal τ σ, and the cycle a → c → b is
equal σ 2 . We conclude that every permutation of {a, b, c} gives an automorphism of
Z2 × Z2 , so
Aut(Z2 × Z2 ) = hσ, τ i ' S3 .
16

Example 15. What is Aut(Z8 )? We could do the same thing as in the above ex-
amples — write down all the permutations fixing 0 and figure out which of them are
homomorphisms — but ... watch this instead.
Suppose σ : Z8 → Z8 is a homomorphism. If we let n := σ(1), then for any x ∈ Z8
we get
σ(x) = σ(1 + · · · + 1) = xσ(1) = xn.
So σ is just multiplication by n! Let’s call it σn . What’s the kernel? Well,

ker(σ) = {x ∈ Z8 : σ(x) = 0}
= {x ∈ Z8 : xn = 0}.

For which n is the kernel trivial? Obviously n = 0 means σ0 (x) = 0, so ... No dice
there. And n = 1 gives the identity map σ1 = id. How about n = 2? Well then
4 ∈ ker(σ2 ), so that’s not an automorphism. How about n = 3? Then no way you can
have 3x ≡ 0 (mod 8), unless x = 0! So σ3 (x) = 3x is an automorphism, as we saw in
Example 11.
Now you’re seeing a pattern: σn is an automorphism if and only if n is coprime to
8. So the only possibilities are n = 1, 3, 5, 7.

Aut(Z8 ) = {σ1 , σ3 , σ5 , σ7 }.

Note σ1 = id and all the others have order 2. Therefore

Aut(Z8 ) ' Z2 × Z2 .

Sweet going! I don’t think anyone actually says that.


There is a very special type of automorphism of a group. If G is a group and g ∈ G,
we define a function σg : G → G by

σg (x) := gxg −1 .

This is a homomorphism. Proof: if x, y ∈ G then

σg (xy) = gxyg −1 = gxg −1 gyg −1 = σg (x)σg (y).

It’s also a bijection, because its inverse is σg−1 = σg−1 . So σg ∈ Aut(G)! These are
called inner automorphisms of G. The operation σg is called conjugation by g.
Ready for things to get meta? The map σ : G → Aut(G) given by σ(g) := σg is a
homomorphism, because σgh = σg ◦ σh . Its kernel is called the center of G:

Z(G) := {g ∈ G : σg = id} = {g ∈ G : gx = xg for all x ∈ G}.

(Convince yourself of the last equality there!) In other words, Z(G) is the set of all
elements that commute with everything.
Example 16. If G is abelian, then it’s clear from the above definition that Z(G) = G
— that means σg = id for all g ∈ G, so the only inner automorphism is the identity
map. For example, the automorphism of Z3 given by x 7→ 2x is not inner.
Example 17. What is Z(S3 )? You’ll notice that no nontrivial element commutes with
every other element. For example, the swap 1 ↔ 2 doesn’t commute with anything
but itself (and the identity). So Z(S3 ) = {1}.
17

Example 18. What is Z(GLn (R))? This is trickier. But it turns out to be the set of
all nonzero scalar matrices: that is, scalar multiples of I.

Z(GLn (R)) = λI : λ ∈ R× .


We can get an isomorphism from R× to Z(GLn (R)) via λ 7→ λI, so

Z(GLn (R)) ' R× .

The center is a very important subgroup that we will see (and use!) many times
in this course.

2.6 Normal subgroups


There’s something deep going on with the homomorphism Z → Zn . The operation in
Zn is essentially the same as that in Z: it’s ordinary addition, but you reduce mod n
after. This “reduction mod n” step is a process called a quotient.

Example 1. What are the cosets of 4Z? They’re [0] = 4Z, [1] = 4Z + 1, [2] = 4Z + 2,
and [3] = 4Z + 3. We could do [4] = 4Z + 3, but that’s just the same as [0]. What
happens if we add two cosets?

[1] + [2] = (4Z + 1) + (4Z + 2) = 4Z + 3 = [3]


[2] + [3] = (4Z + 2) + (4Z + 3) = 4Z + 5 = 4Z + 1 = [1]

It’s the same as addition mod 4! This is not a coincidence: addition of cosets amounts
to addition in Z4 . We’re seeing here that the set of cosets itself forms a group.

Instead of Z and 4Z, we want to take ANY group G with ANY subgroup H, and
show that the cosets form a group. It is called the quotient group. Unfortunately,
there’s a problem if G is nonabelian: you can only form the quotient group if H is
what’s called a normal subgroup. I can define what makes a subgroup “normal”, but
it may be confusing at first. Wisdom will come with time. For now let’s treat it like
ripping off a bandaid — quick and easy.
Let G be a group and let N be a subgroup of G. Then N is a normal subgroup
of G, denoted N E G, if

gxg −1 ∈ N for all x ∈ N and g ∈ G.

What’s going on here? Inner automorphisms! The above property is saying that N is
closed under conjugation by every element of G. Not every subgroup is normal, as the
below example will show.

Example 2. Let D4 be the dihedral group of order 8. If r is a 90◦ clockwise rotation


and f is a flip, then the subgroup N := hri is normal. Why? We have to show it’s
closed under conjugation by every element of D4 . Well, remember that every element
g ∈ D4 all have the form g = f i rj for some i ∈ {0, 1} and some j ∈ {0, 1, 2, 3},
satisfying the rule rf = f r3 . So if we let x = rk be an element of N , we have to show
gxg −1 is in N too.

• Case 1: i = 0. Then g = rj , so gxg −1 = rk = x, which is in N .


18

• Case 2: i = 1. Then g = f rj , so

gxg −1 = (f rj )rk (f rj )−1 = f rj rk r−j f −1 = f rk f = f 2 r3k = r3k = x3

which is in N . Phew!
This shows that N E D4 .
How about the subgroup H generated by f ? So H = {1, f }. This is not a normal
subgroup of D4 , because it’s not closed under conjugation by r: indeed, rf r−1 =
rf r3 = r2 f which is NOT in H.
 
1 2 3
Example 3. In S3 , let σ = and let N := hσi. Let’s show that N E S3 .
2 3 1
The only other elements of S3 are swaps:
     
1 2 3 1 2 3 1 2 3
τ1 = , τ2 = , τ3 = .
2 1 3 1 3 2 3 2 1

So S3 = {1, σ, σ 2 , τ1 , τ2 , τ3 }. Obviously we don’t have to conjugate N by other elements


of N — we know N is closed under multiplication. So we just have to conjugate by
τ1 , τ2 , and τ3 , all of which are their own inverses. Let’s try:
     
−1 1 2 3 1 2 3 1 2 3 1 2 3
τ1 στ1 = = = σ2
2 1 3 2 3 1 2 1 3 3 1 2
which is in N . Now let’s conjugate σ by τ2 and τ3
     
−1 1 2 3 1 2 3 1 2 3 1 2 3
τ2 στ2 = = = σ2.
1 3 2 2 3 1 1 3 2 3 1 2
     
1 2 3 1 2 3 1 2 3 1 2 3
τ3 στ3−1 = = = σ2.
3 2 1 2 3 1 3 2 1 3 1 2
So all of these conjugates are in N . Great. NOW we have to conjugate σ 2 — but trust
me, you can show that

τ1 σ 2 τ1−1 = τ2 σ 2 τ2−1 = τ3 σ 2 τ3−1 = σ

which is in N . OK, technically since 1 ∈ N , we have to conjugate 1 too — but


τi 1τi−1 = 1 so there’s no point doing that. Great! So this shows N E S3 .
From the above examples, you can see how terribly painstaking it can be to show
that a given subgroup N is normal. You have to write down all its elements x, write
down all the elements g ∈ G, and show one-by-one that gxg −1 is in N .
Example 4. Recall that SLn (R) is the group of n × n real matrices with determinant
1. Show that SLn (R) E GLn (R).
Example 5. Consider GLn (R) and the subgroup N := {I, −I}. Show that N is a
normal subgroup of GLn (R).
Example 6. Let G be a group. Recall that the center of G is the subgroup

Z(G) := {z ∈ G : zg = gz for all g ∈ G}.

Show that Z(G) E G. In fact show that if N ≤ Z(G), then N E G.


19

Example 7. Show that if A is an abelian group, then every subgroup of A is normal.


Example 8. If G is any group, show that the trivial subgroup {1} is normal.
Here is a super handy way to check that a subgroup is normal! Get ready, this is
on the technical side of proofs.
Proposition 9. Let G be a group and let N be a subgroup of index 2. Then N is a
normal subgroup of G.
Proof. Recall that the index of a subgroup is the number of left cosets: so N has two
left cosets, say N and aN . These are the equivalence classes, so they partition G, i.e.

G = N t aN.

To check that N is normal, all we have to do is conjugate its elements, right? OK. Let
x ∈ N and g ∈ G; we want to show gxg −1 ∈ N . According to the above decomposition,
there are two cases: g ∈ N and g ∈ aN , but if g ∈ N then of course gxg −1 ∈ N because
N is a subgroup. Note to self: conjugating N by elements of N is always kind of dumb.
So we just have to deal with the case g ∈ aN . Write g = an for some n ∈ N . If
gxg −1 is not in N , we can apply the partition again: we must have gxg −1 ∈ aN . So
write gxg −1 = ay for some y ∈ N . In total we’re getting

ay = gxg −1 = (an)x(an)−1 = anxn−1 a−1

and canceling the a on the left gives

y = nxn−1 a−1 .

We know x, y, n ∈ N , so solving for a gives

a = y −1 nxn−1

which is an element of N ! We have deduced that a ∈ N , contradicting the fact that we


had two distinct left cosets N and aN . The contradiction arose from the assumption
that gxg −1 was not in N — we conclude gxg −1 ∈ N as required. v

Phew∼∼∼! Now we come across the most useful class of examples of normal
subgroups: kernels.
Proposition 10. Let f : G → H be a homomorphism. Then ker(f ) is a normal
subgroup of G.
Proof. We know that ker(f ) is a subgroup, so all that’s left is to show that it’s closed
under conjugation. We have to show that if x ∈ ker(f ) and g ∈ G, then gxg −1 ∈ ker(f ).
OK. Well, f (x) = 1, so

f (gxg −1 ) = f (g)f (x)f (g)−1 = f (g)f (g)−1 = 1.

Therefore gxg −1 ∈ ker(f ), as required. v

Conclusion: kernels are normal subgroups. Quite surprisingly, the converse is


true: if N is any normal subgroup of G, then N = ker(f ) for some homomorphism
f : G → H. How would one prove this? Not only would you have to construct a
homomorphism f , you’d have to construct a group H too. Not trivial! We’ll see that
the right group is H = G/N , the quotient.
20

2.7 Quotients
We’ve actually already seen the setup of quotients: recall that if H ≤ G, then the left
cosets of H are the sets
aH := {ah : h ∈ H}.
We used these to prove Lagrange’s Theorem. Now we define the quotient of G by H
is the set of left cosets:
G/H := {aH : a ∈ G}.
Recall that the left cosets are exactly the equivalence classes of the following relation
on G: x ∼ y if x−1 y ∈ H. So we’ll use the notation

[x] := xH.

Right now the quotient is just a set, but my claim is that if N E G, then G/N is
actually a group. It’s a group whose elements are sets! It’s crazy but true! The
operation is defined as follows: if [x] and [y] are two left cosets, then

[x] · [y] := [xy].

If you really think about this definition, there should be alarms going off — I’m trying
to multiply two sets, so I pick an element x from one, y from the other, and make a
new set containing xy. Well what if I pick different elements x0 and y 0 ? Would I still
get the same set containing x0 y 0 ? It’s not obvious that the above rule is well-defined.
But it is, because N is normal, and we prove this now.

Proposition 1. Let G be a group and N E G. Then G/N is a group under the


operation [x] · [y] := [xy].

Proof. First we show it’s well-defined. Here’s how you do it. Suppose that [x] = [x0 ].
Then x ∼ x0 , so x−1 x0 ∈ N , so x0 = xa for some a ∈ N . Similarly if [y] = [y 0 ] then
y 0 = yb for some b ∈ N . Now we have to check that [xy] = [x0 y 0 ], i.e. xy ∼ x0 y 0 . Well,

x0 y 0 = (xa)(yb)
= x(yy −1 )a(yb)
= xy(y −1 ay)b.

Now here’s where we use normality of N . Since a ∈ N and N is closed under conjuga-
tion, we see that y −1 ay ∈ N . We also know b ∈ N and N is closed under multiplication,
so the element c := y −1 ayb is in N . Thus we have shown that

(xy)−1 (x0 y 0 ) = c ∈ N

so that xy ∼ x0 y 0 , and we conclude [xy] = [x0 y 0 ]. Therefore the operation is well-


defined! Excellent!
Now let’s show that G/N is a group.

(i) Associativity: Let [x], [y], [z] be elements of G/N . Then

([x][y])[z] = [xy][z] = [(xy)z] = [x(yz)] = [x][yz] = [x]([y][z]).


21

(ii) Identity: The identity element is [1] = [N ]. Indeed, for any [x] ∈ G/N we get

[x][1] = [x · 1] = [x]

and similarly [1][x] = [x]. So [1] = 1G/N .

(iii) Inverses: Let [x] ∈ G/N . Then [x]−1 = [x−1 ]. Why? If we multiply [x] by [x−1 ]
we get
[x][x−1 ] = [xx−1 ] = [1]
which is the identity element of G/N .

This shows that G/N satisfies the axioms of a group. v

The group G/N is called the quotient group of G by N . Congratulations! We


just finished one of the most complicated, important constructions we’ll do in this
course. The idea of the construction is to systematically collapse all elements of N
into a single element [1], much like how in Z7 , all the elements 1, 8, 15, etc. are thought
of as the same.

Example 2. Here’s the classic example: what is Z/nZ? Well, the cosets are a + nZ =
{a + nk : k ∈ Z} which is the equivalence class of a mod n, and the operation is
[a] + [b] = [a + b] which is exactly addition mod n. So

Z/nZ ' Zn .

Example 3. Let V := Z2 × Z2 be the Klein 4-group, and consider the subgroup


N := Z2 × {0} = {(0, 0), (1, 0)}. Since V is abelian, N is automatically normal, so we
can look at the quotient V /N . What does it look like? Well (a, b) ∼ (c, d) if and only
if (a − b, c − d) ∈ N , if and only if a − b ∈ {0, 1} and c = d. So the only cosets are
[(0, 0)] and [(0, 1)]. Thus
G/N = {[(0, 0)], [(0, 1)]}
which is a group of two elements, and is therefore isomorphic to Z2 . A nice way to
write what we just proved is
Z2 × Z2
' Z2 .
Z2 × {0}
It’s like we canceled out one of the factors of Z2 ! More generally, we’ll show (later)
that for any groups G, H,
G×H
' H.
G × {1}
So quotients support a sort of “cancellation”.

If G is finite, what’s the cardinality of G/N ? Well, since G/N is just the set
of cosets, of which there are [G : N ] many. This is equal to |G|/|N | by Lagrange’s
Theorem (Corollary 2.3.7). So if G is finite, then

|G/N | = |G|/|N |.

Easy to remember, huh?


22

 
1 2 3
Example 4. In S3 , if we consider the permutation σ = and the group
2 3 1
N := hσi it generates, we showed in Example 2.6.3 that N E S3 . So we can consider
the quotient S3 /N . But we know that |S3 /N | = |S3 |/|N | = 6/3 = 2, so S3 /N is a
group of order 2, which implies
S3 /N ' Z2 .
A lot of Z2 going around here ... Just a coincidence though.

It’s possible for G/N to be finite, even if G is infinite. For example Z is infinite but
Z/nZ = Zn is finite. Is it possible for N to be finite but G/N to be infinite? Sure! For
example G could be any infinite group, and N := {1} could be the trivial subgroup.
Then N is finite, but G/N ' G (why?) which is infinite.
On the other hand, if both N and G/N are finite, then so must be G. This actually
has nothing to do with normal subgroups and quotients though.

Proposition 5. Let G be a group and let H ≤ G. If H is finite and G/H is finite,


then G is finite.

Proof. Since G/H is finite, there are only finitely many cosets mod H, so let a1 H, . . . , an H
be all of them. Then
G = a1 H t · · · t an H
which is a union of finitely many finite sets. The only possibility is that G has finitely
many elements. v

Let P be a property of groups. Then P is called extension-closed if, whenever


G is a group and N E G such that N satisfies P and G/N satisfies P , then G also
satisfies P . Therefore Proposition 5 is saying that finiteness is extension-closed.

Example 6. Abelian-ness, on the other hand, is not extension-closed. An easy ex-


ample is D4 : the subgroup N := {1, r, r2 , r3 } is normal and abelian, D4 /N ' Z2 is
abelian, and yet D4 is not abelian.

In the following exercise, we’ll flex our quotient muscles.

Proposition 7. Let G be a cyclic group and let H ≤ G. Then G/H is cyclic.

Proof. Let G = hai. Note that G is abelian, so any subgroup is normal, so it makes
sense to look at G/H. We claim that the coset [a] generates G/H.
Claim: G/H = h[a]i.
All we have to show is that every element of G/H is a power of [a]. Well, an element
of G/H has the form [x] for some x ∈ G, so we can write x = an for some n ∈ Z since
a generates G. Therefore
[x] = [an ] = [a]n
by definition of multiplication in G/N . QED! v

Remember why we defined quotients: we’re trying to show that every normal
subgroup must be the kernel of some homomorphism. We can prove this now. If
N E G, we can define a homomorphism

π : G → G/N
23

π(g) := [g].
Check that this is a homomorphism — it’s an important one, called the quotient
map. Some common notations for it include

π(g) = [g] = g.

I’m a fan of the third one.

Proposition 8. Let N E G and let π : G → G/N be the quotient map. Then π is


surjective and ker(π) = N .

Proof. The fact that π is surjective is interesting but obvious: the elements of G/N
have the form [g] = π(g), so they’re all in the image of π. What about the kernel?

ker(π) = {x ∈ G : π(x) = [1]}


= {x ∈ G : [x] = [1]}
= {x ∈ G : x ∼ 1}
= {x ∈ G : x ∈ N }
= N.

QED! v

2.7.1 Groups of order 2p: a fun exercise


In section 2.5, we classified groups of prime order — the only group of order p is Zp .
What about groups of order 2p? I can think of two: the dihedral group Dp , and the
cyclic group Z2p . These are actually all of them, and now we can prove it.

Proposition 9. Let G be a group of order 2p where p is an odd prime. Then G is


isomorphic to Z2p or Dp .

Before proving this, let’s think about HOW we would prove such a specific result.
We’d have to write down the Cayley table for G and show that it’s the same as Dn ,
and use that to construct an isomorphism. Well, what’s the Cayley table for Dn ? If
r is a rotation by 2π/p radians and f is a flip (across your favorite axis), then the
elements of Dn are all obtained by rotating then flipping:

Dn = {1, r, r2 , . . . , rp−1 , f, f r, f r2 , . . . , f rn−1 }.

In other words, Dn = hr, f i. Using the the three rules rn = 1, f 2 = 1, and rf = f rn−1 ,
you can write down the entire Cayley table of Dn ! These three rules are called the
relations of Dn .
In fact Dn is the largest group with these generators and relations. Here “largest”
means that if G is generated by two elements x, y satisfying the relations xn = 1,
y 2 = 1, and xy = yxn−1 , then G is the image of Dn under some homomorphism.

Lemma 10. Let G be a group and suppose that x, y ∈ G are two elements such
that xn = 1, y 2 = 1, and xy = yxn−1 . Then there exists a unique homomorphism
ϕ : Dn → G such that ϕ(r) = x and ϕ(f ) = y.
24

Proof. This statement has two aspects: not only do we have to show that ϕ exists,
but we have to show that there is only one such ϕ.
(i) Existence of ϕ. How do we define ϕ(s) for s ∈ Dn ? Well, we know that we HAVE
to have ϕ(r) = x and ϕ(f ) = y, so since ϕ is a homomorphism we would be forced to
have ϕ(rf r) = xyx. And ϕ(r2 f 3 r−1 ) = x2 y 3 x−1 . Right? The moral of the story is
that Dn is generated by r and f , so if we know what ϕ does to the generators, then
we know it completely.
So if s ∈ Dn , then s has the unique form s = f i rj for some i ∈ {0, 1}, j ∈
{0, . . . , n − 1}, so we have no choice but to define

ϕ(s) := y i xj .

This is well-defined — at first it looks like it depends on the choice of i, j, but since
these numbers are uniquely determined by s, there’s no problem.
Let’s check that ϕ is a homomorphism. If s = f i rj and t = f k r` , then the equation
rf = f r−1 can be used to computre the product st, as follows:

st = (f i rj )(f k r` ) = f i (f k r−j )r` = f i+k r`−j .

Therefore, ϕ(st) is equal to


 
ϕ(st) = ϕ f i+k r`−j
= y i+k x`−j
= (y i xj )(y k x` )
= ϕ(s)ϕ(t).

So ϕ is a homomorphism! Woo hoo! We can easily calculate to get ϕ(r) = ϕ(f 0 r1 ) = x


and ϕ(f ) = ϕ(f 1 r0 ) = y, so this ϕ satisfies the requirements.
(ii) Uniqueness of ϕ. Now we have to check the uniqueness of ϕ, i.e. that there’s no
other homomorphism ψ : Dn → G satisfying ψ(r) = x and ψ(f ) = y. Well, then if
s ∈ Dn , writing s = f i rj gives

ψ(s) = ψ(f i rj ) = ψ(f )i ψ(r)j = y i xj = ϕ(s)

which implies ψ = ϕ. Heh heh heh! So there’s only one such homomorphism after
all. v

OK, now armed with the lemma, we can prove Proposition 9!

Proof of Proposition 9. If G has an element of order 2p, then it must generate G, so G


is cyclic and G ' Z2p . So we proceed assuming G has no such element. Since |G| = 2p,
we know that each nontrivial element of a ∈ G must have order 2 or p by Lagrange’s
Theorem. We start by finding a “rotation” in G.
Claim 1: G has an element of order p.
Suppose not. Then every nontrivial element has order 2. By Q1 on Assignment 1, this
implies that G is abelian. But then if a, b ∈ G are two distinct elements, then the set
{1, a, b, ab} is a subgroup of G of order 4. Thus 4|2p by Lagrange’s Theorem, which is
a contradiction since p is prime. This proves Claim 1.
25

Therefore G has an element of order p, say x. Let A := hxi so that A is a subgroup


of G of index 2. Therefore A E G (exercise!), so we can consider the quotient group
G/A which has order |G|/|A| = 2p/p = 2. Finally, choose an element y ∈ G \ A, so
that the distinct cosets of A are A and yA.
Claim 2: y has order 2.
Why? Let n = ord(y). Then y n = 1. Applying the quotient map we get y n = 1. Thus
n divides |G/A| = 2 by Lagrange’s Theorem so either n = 1 or 2. Can’t have n = 1
because y 6= 1! So it’s gotta be y 2 = 1. This proves Claim 2.
OK! So now we have our “rotation” x, and our “flip” y. We have to show that
they generate G and obey the same rules as Dp .
Claim 3: G = hx, yi.
How do you prove this? Well, if it’s anything like Dp , the elements should be 1, x, . . . , xp−1
and y, yx, . . . , yxp−1 . These are all distinct (you can prove it!), so since there are 2p
elements here, that’s all of them. On the other hand, these are all elements of hx, yi.
Thus G and hx, yi both have 2p elements, which forces G = hx, yi.
Now we know xp = 1 and y 2 = 1. All that’s left is to show the last rule.
Claim 4: xy = yxp−1 .
Look at the element xy. The same argument we used for Claim 2 can be used to show
that any element outside A has order 2 — thus ord(xy) = 2, so

xy = (xy)−1 = y −1 x−1 = yxp−1

which is exactly Claim 4!


Now we’re home-free: by Lemma 10, we know that there exists a unique homo-
morphism ϕ : Dp → G such that ϕ(r) = x and ϕ(f ) = y. To conclude that Dp ' G,
we just have to show that ϕ is a bijection ... i.e. injective and surjective ...

• Surjective: We know that x = ϕ(r) and y = ϕ(f ) are in the image of ϕ, and
therefore im(ϕ) is a subgroup of G containing x and y. But by Claim 3, G is
generated by x, y, so this is only possible if im(ϕ) = hx, yi = G. Done!

• Injective: Well, actually, since Dp and G have the same size 2p, the Pigeonhole
Principle says that any surjective function Dp → G is automatically injective —
so injectivity is free. Sweet!

Conclusion: ϕ gives an isomorphism Dp ' G, which is what we set out to show.


QED! v

2.8 First Isomorphism Theorem


Quotients were introduced to relate kernels and normal subgroups: every kernel is a
normal subgroup, and every normal subgroup is the kernel of some homomorphism. So
if we start with a homomorphism f : G → H, then ker(f ) E G, and we can construct
the quotient G/ ker(f ) and the quotient map π : G → G/ ker(f ). Wait a second, now
we have two homomorphisms — f and π! The First Isomorphism Theorem solidifies
the relationship between these two maps.
26

Theorem 1 (FIT). Let f : G → H be a homomorphism. Then

G/ ker(f ) ' im(f ).

The First Isomorphism Theorem transcends mere theorem-hood. It’s a description of


nature. Here’s a diagram illustrating what’s going on:

G
π f

G/ ker(f ) im(f )
'

Proof. All we need to do is write down an isomorphism, right? Define a map by

f : G/ ker(f ) → im(f )

f (a) := f (a).
The main problem is here is well-defined-ness — we chose a representative of the
coset a and used that to define f . We have to show that it doesn’t matter which
representative we choose. Well, if a, b are representatives of the same coset, then a = b
which means that a−1 b ∈ ker(f ). So f (a−1 b) = 1, and since f is a homomorphism this
gives f (a) = f (b). Excellent! Also, note that f (a) is indeed an element of the image
of f . Conclusion: f is a well-defined mapping from G/ ker(f ) to im(f ).
Now we’ll show that f is an isomorphism.

• Homomorphism: This is easy: if a, b ∈ G, then

f (ab) = f (ab) = f (ab) = f (a)f (b) = f (a)f (b).

• Injective: Suppose that f (a) = f (b). Then

f (a) = f (b) =⇒ f (a) = f (b)


=⇒ f (a−1 b) = 1
=⇒ a−1 b ∈ ker(f )
=⇒ a = b.

Notice that this is the same proof we used for well-defined-ness, except backwards!

• Surjective: Let x ∈ im(f ). Then x = f (a) for some a ∈ G, so f (a) = f (a) = x.


Thus f hits x.

This shows that f is a bijective homomorphism, hence it is an isomorphism. QED! v

Corollary 2. Let f : G → H be a homomorphism where G is a finite group. Then

|G| = | ker(f )|| im(f )|.

Proof. The First Isomorphism Theorem tells us G/ ker(f ) ' im(f ). Then Lagrange’s
Theorem implies |G|/| ker(f )| = | im(f )|, which immediately gives the result. v
27

Example 3. Let C× be the multiplicative group of nonzero complex numbers, and


let T be the subgroup of unit complex numbers (i.e. those z ∈ C× such that |z| = 1).
We’ll show that
C× /T ' R>0 .
Define a map ϕ : C× → R× by a + bi 7→ a2 + b2 ; in other words, ϕ(z) := |z|2 . Show
that this is a homomorphism. What’s its kernel? Well, ϕ(z) = 1 if and only if |z| = 1,
if and only if z lies on the unit circle. Thus ker(ϕ) = T.
Now what’s its image? Of course ϕ(z) = |z|√2 must be a positive real; conversely,
any positive real number can be hit, because ϕ( t) = t. Therefore im(ϕ) = R>0 . The
First Isomorphism Theorem thus yields C× /T ' R>0 , as required.
Example 4. Let R× be the multiplicative group of real numbers, with subgroup R>0
of positive reals. We’ll show that

R× /R>0 ' {1, −1}

where {1, −1} is a group under multiplication. Well, we can define a map ϕ : R× → R×
by (
1 if t > 0,
ϕ(t) :=
−1 if t < 0.
This is a homomorphism with kernel ker(ϕ) = R>0 and image im(ϕ) = {1, −1}.
Therefore, by the First Isomorphism Theorem we get R× /R>0 ' {1, −1}. Of course,
this is in turn isomorphic to Z2 .
Example 5. Z is an additive subgroup of R, so it’s automatically normal since R is
abelian.. We’ll show that
R/Z ' T
where T is the unit circle. Define a map ϕ : R → T by

ϕ(t) := e2πit

— this is a homomorphism, because ϕ(s + t) = ei(s+t) = eis ei t = ϕ(s)ϕ(t). What’s the


image? Actually, ϕ is surjective: every complex number has the form z = reiθ where
r ≥ 0 is the radius, but every number in T has radius 1, so it must have the form
z = eiθ = ϕ(θ/2πi).
What’s the kernel? Well, remember that eiθ = 1 if and only if θ is an integer
multiple of 2π. So ϕ(t) = 1 if and only if e2πit = 1, if and only if t ∈ Z. So ker(ϕ) = Z.
Therefore R/Z ' T by the First Isomorphism Theorem.
Example 6. Enough with the real numbers already! Let’s try something else. In S4 ,
let τij be the permutation that swaps i with j and leaves everything else fixed. Note
that the disjoint τij ’s commute with each other, e.g. τ12 commutes with τ34 , but does
not commute with τ23 . Let H be the following subgroup:

H = {id, a := τ12 τ34 , b := τ13 τ24 , c := τ14 τ23 }.

You can call these elements “double swaps” since they do two swaps at once. Notice
they all have order 2, and H is isomorphic to the Klein 4-group. Also, it happens that
H is a normal subgroup of S4 — you can check this manually, but we’ll get a clever
way to do it in Chapter 3.
28

Taking that for granted, well, what’s S4 /H? We know it has order |S4 |/|H| =
24/4 = 6, so it could be Z6 or S3 or ... ?! This is a job for the First Isomorphism
Theorem.
Here’s a funny map from S4 into S3 .

ϕ : S4 → S3

Each σ ∈ S4 can be thought of as permuting the elements a, b, c as follows: if σ is the


4-cycle 1 → 2 → 3 → 4, then σ sends a to b, because a = τ12 τ34 and if you cyclically
shuffle the subscripts you get τ23 τ14 = c. If σ is the swap 1 ↔ 2, then σ fixes a = τ12 τ34
because if you swap the subscripts 1 and 2 nothing changes, and σ sends b = τ13 τ24 to
c = τ14 τ23 because if you swap 1 and 2 in the subscripts of b, you’ll end up with c.
So hopefully you’re convinced that each permutation σ ∈ S4 induces a permutation
of the three elements a, b, c. Let ϕ(σ) be this induced permutation. Then ϕ is a function
from S4 to Sym(a, b, c) (= the group of permutations of {a, b, c}), and it’s actually a
surjective homomorphism! What’s the kernel?

ker(ϕ) = {σ ∈ S4 : ϕ(σ) = id}.

So if σ is in the kernel, then ϕ(σ) fixes all of a, b, c. The only way this is possible is if
σ is one of id, a, b, or c — i.e. if σ ∈ H. Thus ker(ϕ) = H and we get

S4 /H ' S3

by the First Isomorphism Theorem!

Example 7. Let’s prove the “cancellation law” I mentioned before: that is, for any
groups G, H,
G×H
' H.
G × {1}
Indeed, define a homomorphism ϕ : G × H → H by ϕ((g, h)) := h. This is just the
projection onto the second coordinate! Notice ϕ is surjective since h = ϕ((1, h)) for
every h ∈ H, so actually im(ϕ) = H.
What about ker(ϕ)? Well, ϕ((g, h)) = 1 if and only if h = 1, so

ker(ϕ) = {(g, 1) : g ∈ G} = G × {1}.

So the First Isomorphism Theorem applies: G × H mod the kernel is isomorphic to


the image, so (G × H)/(G × {1}) ' H as required.

Example 8. Here’s a new group: the quaternion group. Ever heard of quaternions?
Just like the complex numbers introduce a unit i with i2 = −1, the quaternions
introduce three units i, j, k satisfying the following rules:

i2 = j 2 = k 2 = ijk = −1.

These rules are enough to figure out all possible products of quaternion units: for
example
ij = ij(−k 2 ) = −ijkk = −(−1)k = k
and
ji = ji(−ijk) = −ji2 jk = j 2 k = −k.
29

Notice ij 6= ji, so we see that quaternion multiplication is noncommutative.


The quaternion group is the following set:

Q8 := {±1, ±i, ±j, ±k}.

It is a nonabelian group of order 8. Notice −1 is the only element of order 2, and all
of ±i, ±j, ±k have order 4.
Let N := {±1}. Check that N is a normal subgroup (because 1 and −1 commute
with everything!). What’s the quotient Q8 /N ? Well we know it has order |Q8 |/|N | =
8/2 = 4, and in the quotient we have 1 = −1, so a = −a for all a ∈ Q8 — therefore
a2 = −1 = 1, so every nontrivial element has order 2. Can you think of a group of
order 4 like this? ... You got it: the Klein 4-group Z2 × Z2 !

Q8 /N ' Z2 × Z2 .

Let’s prove it. Define a function

f : Q8 → Z2 × Z2



(0, 0) if a = ±1,

(1, 0) if a = ±i,
f (a) :=
(0, 1) if a = ±j,



(1, 1) if a = ±k.

Check that f is a homomorphism, and it’s clearly surjective, so im(f ) = Z2 × Z2 .


What’s the kernel? The identity of Z2 × Z2 is (0, 0), and the only elements that map
there are ±1. So
ker(f ) = {±1} = N.
Thus by the First Isomorphism Theorem we get

Q8 /N = Q8 / ker(f ) ' im(f ) = Z2 × Z2 .

Huzzah!

2.8.1 Correspondence Theorem


Here’s one of my favorite theorems, the Correspondence Theorem. It gives a relation-
ship between subgroups of G and subgroups of G/N .

Example 9. If N := {±1}, we just showed that Q8 /N is isomorphic to Z2 × Z2 . The


latter group has three (nontrivial) subgroups:

{(0, 0), (1, 0)}, {(0, 0), (0, 1)}, {(0, 0), (1, 1)}.

All three of these are normal because Z2 × Z2 is abelian. This means that Q8 /N must
also have three nontrivial normal subgroups and no others. Via the isomorphism given
by i 7→ (1, 0), j 7→ (0, 1), and k 7→ (1, 1), we see that the subgroups of Q8 /N must be
  
1, i , 1, j , 1, k .
30

Now let’s ask an innocent question: which subgroups of Q8 contain N ? Obviously N


and Q8 itself, but what others? The above analysis gives us the hint that there are
only three:
{±1, ±i}, {±1, ±j}, {±1, ±k}.
Let’s prove that these are all of them. Let N 0 be a subgroup of Q8 containing N .
If π : Q8 → Q8 /N is the quotient map, then π(N 0 ) is a subgroup of Q8 /N , and is
therefore one of three subgroups listed above.
Here’s a picture: on the left are all subgroups of Q8 , and on the right are all
subgroups of Q8 /N .

Q8 Q8 /N

hii hji hki hii hji hki

h−1i {1}

{1}

You can see a correspondence between the two sides: h−1i corresponds to {1}, hii
corresponds to hii, etc. Notice that hii has four elements while hii has only two — I’m
not saying that corresponding subgroups are isomorphic, I’m just saying that the two
lattices are in bijection!

The set of subgroups of a group G is called the lattice of subgroups of G.


Lines are drawn to denote inclusions, with smaller subgroups at the bottom and larger
subgroups at the top. What we observed in the last example is that the lattice of
subgroups of Q8 /N is in bijection with the lattice of subgroups of Q8 which contain
N.
Of course, we can generalize this correspondence to all groups.

Theorem 10. Let G be a group, let N E G. Then the following correspondence gives
a bijection between subgroups of G/N and subgroups of G containing N :
{Subgroups of G containing N } −→ {Subgroups of G/N }
H 7−→ π(H)
−1
π (K) ←−[ K
Moreover

(a) The correspondence preserves order: if H ⊆ H 0 then π(H) ⊆ π(H 0 ), and if


K ⊆ K 0 then π −1 (K) ⊆ π −1 (K 0 ).

(b) The correspondence preserves normality: H is normal in G if and only if π(H)


is normal in G/N , and K is normal in G/N if and only if π −1 (K) is normal in
G.

(c) The correspondence preserves index: [π(H 0 ) : π(H)] = [H 0 : H].

Here π : G → G/N is the quotient map, and the notation π −1 (K) denotes inverse
image:
π −1 (K) := {x ∈ G : π(x) ∈ K}.
31

Caution: the notation π −1 (K) is not meant to insinuate that π is invertible!


Note that we use the following various notations:
π(H) = H/N = H.
Proof. If H is a subgroup of G containing N , then π(H) is a subgroup of G/N , so
there’s no problem there. We have to show that the two directions of the correspon-
dence are inverses. Well, to see that π −1 (π(H)) = H,
π −1 (π(H)) = {x ∈ G : π(x) ∈ π(H)}
= {x ∈ G : π(x) = π(h) for some h ∈ H}
= {x ∈ G : [x] = [h] for some h ∈ H}
= {x ∈ G : x ∈ H}
= H.
Showing that π(π −1 (K)) = K is very similar. This shows that the two operations are
inverse to one another, and therefore we have a bijective correspondence. (Alterna-
tively, you could show directly that H 7→ π(H) is injective and surjective.)
Properties (a) and (b) are left as exercises, but (c) is a bit trickier, so let’s do it.
Let’s use the notation π(H) = H, etc.
(c) Recall that H 0 /H is the set of left cosets of H in H 0 , and |H 0 /H| = [H 0 : H] by
definition of index.
H 0 /H := xH : x ∈ H 0 .


H 0 /H := yH : y ∈ H 0 .
To show that the index [H 0 : H] equals the index [H 0 : H], what better way than to
come up with a bijection between H 0 /H and H 0 /H? Nothing, I tell you!
So let’s do it. Define a function Φ : H 0 /H → H 0 /H by
Φ(xH) := π(x)H.
Notice that since x ∈ H 0 , we know that π(x)H is indeed an element of H 0 /H. But
we have to check that Φ is well-defined: if xH = yH, then Φ(xH) = Φ(yH). Well, if
xH = yH, then x−1 y ∈ H, so applying π shows that π(x−1 y) ∈ π(H) = H. But since
π is a homomorphism, this gives π(x)−1 π(y) ∈ H, which is exactly saying that π(x)
and π(y) are equivalent modulo H — therefore
π(x)H = π(y)H
as required. So Φ is well-defined.
Now we check that Φ is a bijection.

• Injective: Suppose Φ(xH) = Φ(yH). Then π(x)H = π(y)H, so π(x)−1 π(y) ∈


H. Since π is a homomorphism, this implies π(x−1 y) ∈ H = π(H). Thus
π(x−1 y) = π(h) for some h ∈ H. What does this mean? We know π is the
quotient map, so it maps two things to the same place if and only if they are
equivalent modulo ker(π) = N . Therefore x−1 y = hn for some n ∈ N . Now
here’s the trick: we know H ⊇ N , so actually n ∈ H too, so we conclude that
x−1 y = hn ∈ H
so that xH = yH, as required. Phew! So Φ is injective.
32

• Surjective: Let cH be a left coset of H, where c ∈ H 0 — we want to show that


cH = Φ(xH) for some x ∈ H 0 . We know that π is surjective onto G/N , so we
may write c = π(x) for some x ∈ G. In fact x ∈ H 0 . Why? We’ve actually
argued something like this above a number of times ... give it a shot!
Great! Since x ∈ H 0 , that means xH is an element of H 0 /H, and Φ(xH) =
π(x)H = cH as required.

This shows that Φ is a bijection between H 0 /H and H 0 /H, and therefore these two
sets have the same size. Conclusion: [H 0 : H] = [H 0 : H], which is what we wanted to
show! v

Example 11. What’s the lattice of subgroups of S3 ? There are three “swaps”, namely

τ1 := (1 ↔ 2), τ2 := (2 ↔ 3), and τ3 := (1 ↔ 3),

all of which generate subgroups of order 2. There are two 3-cycles, namely

σ := (1 → 2 → 3), and σ 2 = (2 → 1 → 3),

but the second is the square of the first, so they generate the same subgroup of order
3. What about the subgroup generated by a swap AND a 3-cycle? By Lagrange’s
Theorem it must have at least order 6, so you get all of S3 .
Conclusion: the lattice is as follows.

S3 = hσ, τ1 i

hσi hτ1 i hτ2 i hτ3 i

{1}

2.9 The Second and Third Isomorphism Theorems


You didn’t think there was just one did you? Actually, the second and third theorems
are very simple applications of the first, so going through their proofs is a good exercise.
First we need some notation. If G is a group and A, B ⊆ G, define the product
set AB by
AB := {ab : a ∈ A, b ∈ B}.

Note that this isn’t necessarily a subgroup of G, even if A and B are. However if A is
normal then this is a subgroup of G.

Lemma 1. Let H ≤ G and N E G. Then

(a) HN ≤ G.

(b) H ∩ N E H.
33

Proof. (a) The elements of HN have the form hn where h ∈ H and n ∈ N . Taking
h = n = 1 we get 1 ∈ HN , so HN is nonempty. Now suppose h1 n1 and h2 n2 are
elements of HN . Then

(h1 n1 )(h2 n2 ) = h1 (h2 h−1


2 )n1 h2 n2 = hn

where h = h1 h2 ∈ H and n = (h−1 2 n1 h2 )n2 ∈ N . Since this has the form hn, it’s a
member of HN , and therefore we’ve shown that HN is closed under multiplication.
Next we have to show it’s closed under inverses. Well, if h ∈ H and n ∈ N
then (hn)−1 = n−1 h−1 = h−1 (hn−1 h−1 ) has the correct form, since h−1 ∈ H and
hn−1 h−1 ∈ N . So (hn)−1 ∈ HN .
(b) Clearly H ∩N is a subset of H, and showing that it’s a subgroup is easy (try it!)
— so let’s figure out why it’s normal. But if x ∈ H ∩ N and h ∈ H, then hxh−1 ∈ H
since x ∈ H and hxh−1 ∈ N since N is normal. So hxh−1 ∈ H ∩ N , and H ∩ N is
normal in H. v

Theorem 2 (SIT). Let H ≤ G and N E G. Then HN ≤ G, H ∩ N E H, and


HN H
' .
N H ∩N
Proof. By Lemma 1, we know all the objects in question are groups, N E HN , and H ∩
N E H — so the quotients HN/N and H/H ∩ N are well-formed. We’ll use the First
Isomorphism Theorem: all we have to do is write down a surjective homomorphism
H
HN → H∩N whose kernel is N , right?
Define a homomorphism ϕ : HN → H/H ∩ N as follows:

ϕ(hn) := h(H ∩ N ).

Here h(H ∩ N ) is a left coset of H ∩ N , hence it is an element of the quotient H/H ∩ N .


Let’s just call it h. What’s going on? ϕ takes a product hn and peels off the h part.
Alarms are going off — why is ϕ well-defined? If h1 n1 = h2 n2 , we have to show that
h1 is equivalent to h2 mod H ∩ N . Well, h1 = n−1 1 h2 n2 , so

h−1 −1 −1 −1 −1
1 h2 = (n1 h2 n2 ) h2 = n2 h2 n1 h2

and thing is in H because the LHS is, and it’s also in N because the RHS is (think
about it!). So it’s in H ∩ N , which means that h1 ≡ h2 mod H ∩ N . Therefore h1 = h2 ,
so ϕ is well-defined.
Now let’s check that ϕ is a surjective homomorphism whose kernel is N .

(i) Homomorphism: Let x1 := h1 n1 and x2 := h2 n2 be members of HN . We’ll


use the formula we used to show Lemma 1: namely x1 x2 = (h1 h2 )(h−1
2 n1 h2 n2 ).
The h part of this is h1 h2 , so

ϕ(x1 x2 ) = ϕ((h1 h2 )(h−1


2 n1 h2 n2 ))
= h1 h2
= h1 · h2
= ϕ(x1 )ϕ(x2 ).

Sweet.
34

(ii) Surjective: This is easy: if x ∈ H/H ∩ N , then x = h(H ∩ N ) for some h ∈ H,


and so ϕ(h) = h(H ∩ N ) = x. So ϕ hits everything.

(iii) Kernel: Ho ho ho. Clearly N is in the kernel, since if n ∈ N then n has 1 as


its h part, so ϕ(n) = 1. Conversely, if ϕ(x) = 1, write x = hn for some h ∈ H,
n ∈ N . Then h = ϕ(x) = 1, so h ∈ N , and so x = hn ∈ N . This shows that
ker(ϕ) = N !

Now we can use the First Isomorphism Theorem — it says HN/ ker(ϕ) ' im(ϕ). But
the kernel is N and the image is H/H ∩ N , so we get
HN H
'
N H ∩N

as required! :) v

Now for the Third Isomorphism Theorem. It’s a cancellation law of sorts. First
we’ll need a lemma that we’ve actually been using implicitly for some time ...

Lemma 3. Let N be a normal subgroup of G and let f : G → H be some homomor-


phism. If N ⊆ ker(f ), then the map

f : G/N → H

f (x) := f (x)
is a well-defined homomorphism.

Proof. The proof is actually the same as the verification we did in the proof of the FIT
— try it! v

Ready for the Third Isomorphism Theorem? To state it, remember the Corre-
spondence Theorem. It says that subgroups of G/N correspond to subgroups of G
containing N , and moreover this correspondence preserves normality. So if we have a
normal subgroup K containing N , then it corresponds to a normal subgroup K/N of
G/N .
What happens if we take G/N and quotient out by K/N ? The Third Isomorphism
Theorem answers this question:

Theorem 4 (TIT). Let K, N be normal subgroups of G such that N ⊆ K ⊆ G. Then

G/N G
' .
K/N K
It’s saying you can “cancel out” the N from the top and bottom.

Proof. We’ll use the First Isomorphism Theorem again! Take the quotient map π :
G → G/K. Since N ⊆ K = ker(π), Lemma 2 applies — we get a well-defined
homomorphism π : G/N → G/K given by π(xN ) = π(x). It’s surjective because π is.
What’s the kernel of π? Well, π(xN ) = K if and only if π(x) = K, if and only
if x ∈ K, if and only if xN ∈ K/N . The Correspondence Theorem is at work here!
Therefore ker(π) = K/N . So by the First Isomorphism Theorem, (G/N )/(K/N ) '
G/K as required. v

Вам также может понравиться