Math 331-1: Abstract Algebra

Northwestern University, Lecture Notes

Written by Santiago Cañez

These are notes which provide a basic summary of each lecture for Math 331-1, the first quarter
of “MENU: Abstract Algebra”, taught by the author at Northwestern University. The book used
as a reference is the 3rd edition of Abstract Algebra by Dummit and Foote. Watch out for typos!
Comments and suggestions are welcome.

Contents

Lecture 1: Introduction to Groups 2

Lecture 2: Integers mod n 5
Lecture 3: Dihedral Groups 9
Lecture 4: Symmetric Groups 12
Lecture 5: Homomorphisms 17
Lecture 6: Group Actions 20
Lecture 7: Some Subgroups 23
Lecture 8: Cyclic Groups 27
Lecture 9: Generating Sets 29
Lecture 10: Zorn’s Lemma 32
Lecture 11: Normal Subgroups 36
Lecture 12: Cosets and Quotients 40
Lecture 13: Lagrange’s Theorem 44
Lecture 14: First Isomorphism Theorem 47
Lecture 15: More Isomorphism Theorems 51
Lecture 16: Simple and Solvable Groups 54
Lecture 17: Alternating Groups 58
Lecture 18: Orbit-Stabilizer Theorem 62
Lecture 19: More on Permutations 65
Lecture 20: Class Equation 69
Lecture 21: Conjugacy in Sn 72
Lecture 22: Simplicity of An 74
Lecture 23: Sylow Theorems 77
Lecture 24: More on Sylow 79
Lecture 25: Applications of Sylow 82
Lecture 26: Semidirect Products 85
Lecture 27: Classifying Groups 88
Lecture 28: More Classifications 92
Lecture 29: Finitely Generated Abelian 96
Lecture 30: Back to Free Groups 100
Lecture 1: Introduction to Groups

Abstract algebra is the study of algebraic structures, which are sets equipped with operations akin to
addition, multiplication, composition, and so on. Manipulating and solving equations—the basic
concepts you would have seen in a previous “algebra” course—play a role as well, but now our
focus is much more squarely on the underlying structures which allow such manipulations to work.
Ultimately, the goal is to use the various tools available—in particular the notions of “sub” and
“quotient” structures—to classify a given algebraic structure to the extent possible.
That is a very vague brief introduction, so let us say a bit more about what we mean by an
“algebraic structure”. In subsequent quarters we will be studying what are called rings and fields,
which in some sense provide vast generalizations of the notion of “numbers”. Both of these are sets
equipped with two operations which are assumed to satisfy some appropriate properties, but as a
first step towards understanding “algebra” we begin this quarter with the notion of a group, which
only involves a set equipped with a single operation. Historically groups arose in the following way.

Groups and polynomials. We are all familiar with the quadratic formula, which gives an explicit
description of the roots of a polynomial of degree 2, say with real coefficients:
√
2 −b ± b2 − 4ac
ax + bx + c = 0 =⇒ x =
2a
The key observation for us is that this expression for the roots involves only the coefficients of
the given polynomial and some basic algebraic operations: addition, subtraction, multiplication,
division, and taking a square root. If we go back a degree, it is also true that the roots of a linear
polynomial can be expressed in terms of its coefficients and basic algebraic operations (only division
is needed in this case!):
b
ax + b = 0 =⇒ x = − .
a
The search for analogous formulas for polynomials of higher degree is a problem which dates
back centuries, even millennia! The cubic formula for polynomials of degree 3 is much more
complicated than the quadratic formula above (it requires higher-order root extractions), and the
quartic formula for polynomials of degree 4 even more so (you can see what they look like on
Wikipedia!), but the point is that such formulas exist. However, in the 19th century it was proven
that no analogous quintic formula for polynomials of degree 5 existed. This seems quite surprising
at first, since it is not all all clear what breaks down when we make the jump from degree 4 to
degree 5. It turns out, as we’ll see fully in the spring, that the reason for this has to do with the
structure of the “group of permutations” of the roots of such polynomials.
We will avoid giving any formal definitions for now, but√ here√is the basic (and at this point quite
2
vague) idea. Consider the polynomial x − 2 with roots 2, − 2. There are two possible permu-
tations of the roots in this case—do nothing, or exchange them—which √ thus form a “permutation
group” with two elements. The polynomial x3 − 2 has three roots: 3 2 and two complex conjugate
roots. There are 3! = 6 ways of permuting these 3 roots, and so we get a√group √ of √
permutations
√
with 6 elements in this case. Now consider (x2 − 2)(x2 − 3). This has roots 2, − 2, 3, − 3, and
there are 4! = 24 ways of permuting these. However, in this case it turns out that not √ all of these
√
24 possible permutations should actually be allowed; the issue is that permuting, say, 2 and 3
exchanges roots of the two factors x2 − 2 and x2 − 3, and, for reasons we will leave for the spring,
this should not actually be√a valid permutation. We only get√four allowable permutations—the two
coming from permuting ± 2 and the two which permute ± 3—and thus a group with 4 elements.

2
 The upshot is that the question as to whether or not we can express the roots of a polynomial
in a certain way is intimately related to properties of these groups of root permutations, and that
properties of these groups reflect properties (or “symmetries”) of the roots. To give an answer as
to why a quintic formula does not exist—although not an answer which will make any sense as
this point—the fact is that there exist polynomials of degree 5 which have the so-called alternating
group A5 as its group of root permutations, and A5 has the property of being a simple, non-abelian
group; it is this property which prevents there from being a nice way of expressing the roots of such
a polynomial. (All groups which arise from polynomials of degree 4 or less—one example being
the alternating group A4 —are either non-simple or simple and abelian, so there is nothing which
stands in the way of there being a nice way of expressing the roots.) We will soon understand what
all of these terms above mean, but we mention this now in order to give some motivation for the
study of groups.

Symmetries of a square. Before giving the definition of a the term “group”, we give one example.
Let D8 be the set of rigid symmetries of a square, which are the rigid/physical movements we can
do to a square which result in the same square. (No “stretching” allowed!) For instance, we can
rotate the square counterclockwise by 90◦ , or by 180◦ , or by 270◦ . Rotating by something like 45◦
is not a symmetry since this technically results in a “diamond” and not the literal square you began
with. In order to distinguish between these rotations, we label the four vertices of the square and
use this labeling to keep track of which symmetry is which:

Rotating by 360◦ is the same as rotating by 0◦ , and any further rotations are the same as one
of the four we have so far, so we get four distinct rigid motions so far: 0, 90, 180, 270 ∈ D8 . (We’ll
drop the degree ◦ symbol from the notation.) We can compose such rotations with one another,
leading to equalities such as
90 · 180 = 270 and 904 = 0,
where we interpret 904 as 90 · 90 · 90 · 90. We are thinking of composition here as a type of
“multiplication”, so we are not literally saying that 90 times itself four times in the usual sense
results in 0.
In addition to rotations, there are four reflections which give rise to symmetries: a horizontal
reflection H, a vertical reflection V , reflection D across the main diagonal, and reflection A across
the “anti-diagonal”:

It turns out that these eight symmetries give all possible rigid motions of a square, so that
D8 = {0, 90, 180, 270, H, V, D, A}.
Now, we can consider compositions which involve reflections as well. For instance, we can compute
H · 90, which is the result of first rotating by 90 and then reflecting horizontally (as is usual with
compositions, we read from right to left when performing the required operation):

3
The same overall result can be obtained by performing the reflection D alone on the original square,
so the composition H · 90 is the same as D:

H · 90 = D.

In a similar way, you can work out that “multiplying” any two elements of D8 still results in an
element of D8 , so that D8 is closed under the operation of composition.
Here are some other key algebraic properties to notice. First, composition is associative, so that
it does not matter how we group elements:

(x · y) · z = x · (y · z) for all x, y, z ∈ D8 .

Second, 0 serves as an identity element for composition, meaning that “multiplying” 0 with any
element, in any order, results in that other element:

0 · x = x = x · 0 for all x ∈ D8 .

And finally, each element of D8 has an inverse, which is an element we can compose it with in
order to result in the identity element:

for all x ∈ D8 , there exists y ∈ D8 such that x · y = 0 = y · x.

Indeed, 90 and 270 are inverses of one another, and 0, 180, H, V, D, and A are their own inverses.
In this way, D8 shares some similarities with, say, the set of integers Z under addition: addition
is associative, the integer 0 serves as an identity element for addition, and any integer has an
“additive inverse”, namely its negative. The notion of a group—the fundamental object of study in
this course—was developed precisely to study such “similar” algebraic structures in a unified way.

Definition of a group. A group is a set G equipped with a binary operation · which

• is associative: (g · h) · k = g · (h · k) for all g, h, k ∈ G;

• admits a two-sided identity: there exists e ∈ G such that e · g = g = g · e for all g ∈ G; and
• admits two-sided inverses: for all g ∈ G, there exists g −1 ∈ G such that g · g −1 = e = g −1 · g.

(To say that · is a binary operation on G simply means that it acts on two elements of G and
produces a single element of G.) Thus, D8 under composition of rigid motions is a group, as is Z
under addition.
We will develop the basic properties of some standard examples over the next few days, but
for now we finish with a word about notation and terminology. It is common to think about the
operation · as a type of “multiplication” and to refer to it as such, so that we will usually speak of
the “product” of g and h for instance. Because of this, it is also common to drop the operation ·
from the notation altogether, and use gh alone to denote the product of g and h. But keep in mind

4
that this “multiplication” is meant to be computed using whatever binary operation our group is
equipped with, which in some cases might actually be “addition”. In the “additive” group Z for
instance, “mn” actually means m + n. Similarly, we use g −1 to denote the inverse of g due to
our “multiplicative” frame of mind, but in Z for example “m−1 ” actually means −m. In general,
we will default to using multiplicative notation when working with an abstract group, but will use
more standard symbols like + in the concrete examples which use them. In the abstract setting,
we will use e to denote the identity element of a group, or eG if we need to make the dependence
on the group G clear. (Technically, the symbol G alone cannot quite refer to a group alone, since
a group should be a set together with a binary operation, and a given set G might have multiple
binary operations which turn it into a group; each of these groups are considered to be different,
but nevertheless it is common to use the set G as the single notation for the entire group, with the
binary operation implicitly assumed to be present.)

Lecture 2: Integers mod n

Warm-Up 1. The definition of a group requires that inverses and an identity element exist, but it
does not outright state that these should be unique. However, this is a simple consequence of the
definition, as we now show. The basic point of this and the next Warm-Up is to provide some first
examples of working with abstract groups, relying solely on the definition of “group” itself. Recall
that when working with abstract groups in this way we default to using multiplicative notation.
Let G be a group and suppose e, e0 ∈ G are both identity elements. Then

e = ee0 = e0

where the first equality holds since e0 is an identity, and the second this e is an identity. Thus
e = e0 so that there is only one identity element.
Similarly, let g ∈ G and suppose h, k ∈ G are both inverses of g. Then

gh = e = gk

by definition of an inverse. Multiplying both sides on the left by h gives

h(gh) = h(gk).

By associativity this is the same as

(hg)h = (hg)k.
Since hg = e because h is an inverse of g, this becomes

eh = ek,

and by definition of the identity we thus get h = k. Hence there is only one inverse of g.
Note that in this latter proof we exploited all properties in the definition of “group”: associativ-
ity, the definition of identity, and the definition of inverse. This argument would not have worked
had we had a different definition of “group” in mind, so, as with all things in abstract mathematics,
the definition we gave is the way it is precisely so that we carry our arguments like the one above.
Moving forward we will not be so pedantic as we were here and explicit list each step in such an
argument—instead, we will say things like “multiplying both sides of gh = gk on the left by g −1
gives h = k” without directly mentioning where associativity and the other properties come in.
Now that we have proven in general that identities and inverses are unique, we do not have to
check that this is the case in any particular example we care about. Indeed, this is of course why

5
mathematicians care about proving things in the most general “abstract” way possible, so that we
know our argument will always apply.

Exercise. The associativity property given in the definition of a group is only stated for products
of three elements at a time, but in fact it extends to any number of elements so that there is never
any ambiguity in an expression like g1 g2 · · · gn . We will take this for granted going forward, but if
you have never thought about why this is true, it is a nice exercise in induction you can work out
for yourself if interested.
Here’s the statement. We will use the notation g1 g2 , . . . , gn to denote the specific product
obtained by multiplying g1 and g2 , and then the result by g3 , and then the result of that by g4 ,
and so on. So, for instance:
g1 g2 g3 g4 := ((g1 g2 )g3 )g4 .
The claim is that this specific grouping of the product is the same as any other grouping: fix ` ≥ 1
and show that for any m ≥ 1 and ` + m elements g1 , · · · , g`+m , we have

(g1 g2 . . . g` )(g`+1 g`+2 · · · g`+m ) = g1 g2 . . . g`+m .

(Proceed by induction on m. The induction step will use the base case of three elements—which
holds by definition—in addition to the induction hypothesis.)

Warm-Up 2. Suppose G is a group such that g 2 = e for all g ∈ G. We show that the group
multiplication on G is commutative, which means ab = ba for all a, b ∈ G. Let us also take this
opportunity to introduce some terminology. The order of an element g ∈ G is the smallest positive
integer n ∈ N such that g n = e, and we say that g has infinite order if no such n exists. Thus the
assumption on G in this Warm-Up says that all elements of G have order at most 2, or equivalently
that every non-identity element has order 2. (The identity is the only element of order 1.) We
will also use the term “order” in the following (seemingly) different way: the order of G itself is
the number of elements it has. So, for instance, D8 has order 8 and Z has infinite order. (Later
on we might distinguish between different types of “infinite order” coming from the notion of the
cardinality of a set.) We will see later that these two uses of the term “order” are not really
different, and that the order of an element is a special case of the order of a group.
One more piece of terminology: a group is abelian it its multiplication operation is commutative.
Thus, this Warm-Up is asking to show that any group for which every non-identity element has
order 2 must be abelian. (The term “abelian” comes from the name Abel, who in the early 1800’s
was the first person to prove that no general quintic formula existed.)
Let a, b ∈ G. Then ab is in G so (ab)2 = e by the assumption on G; in other words

abab = e.

(We omit parentheses in order to not clutter up our work, which we can do by associativity.) Since
a2 = e and b2 = e as well (so a and b are their own inverses), multiplying abab = e on the left by a
and on the right by b gives ba = ab, so G is abelian as claimed.

Some more examples. So far we have seen D8 under composition and Z under addition as
examples of groups. In D8 , 90 and 270 have order 4, while 180, H, V, D, A all have order 2. The
only element of Z with finite order is 0, so every nonzero integer has infinite order. Another standard
example is R× , the set of nonzero real numbers, under ordinary multiplication; the identity element
is 1, and the inverse of any element is its reciprocal. Here, only 1 and −1 have finite order.

6
 To give another example which you would have seen in another context, although likely not
using the language and notation of “groups”, we can take the group GLn (R) of invertible n × n
matrices with real entries:

GLn (R) := {A | A is an invertible n × n matrix over R}.

(The “GL” here stands for “general linear” group, which is a standard term used to refer to the set
of invertible linear transformations on some space.) The identity element is, of course, the identity
matrix. This group is non-abelian when n > 1 (i.e. matrix multiplication does not commute in
general), and you might recall formulas like (AB)−1 = B −1 A−1 for invertible matrices. In fact,
this type of formula is true in any group: (gh)−1 = h−1 g −1 , and the proof is the same as what it
is for matrices.
Consider now GLn (Z), which is the group of invertible n × n matrices with integer entries. The
key point here is that in order for an integer matrix A to be in GLn (Z) requires that its inverse
also be in GLn (Z), which means that it should have integer entries as well. For instance,
 
2 −1
4 6

is in GLn (R) but not in GLn (Z) since its inverse has non-integer entries. (In fact, you can show
that an invertible integer matrix has an inverse which is itself an integer matrix if and only if its
determinant is ±1. This comes from taking determinants of both sides of AA−1 = I, using the fact
that det(AA−1 ) = (det A)(det A−1 ), and using the fact that the determinant of an integer matrix is
an integer.) The point is that there is a difference in asking whether an integer matrix has inverse
over R versus over Z, and such subtleties will be important going forward. Later we will consider
replacing Z or R with other types of “numbers” and looking at other types of matrices.

Integers mod n. We now introduce a fundamental example of a group, which will play a key
role in understanding the structure of arbitrary groups, and in particular abelian groups. This
example is based on the notion of modular arithmetic, which we now define. Fix a positive integer
n ∈ N. We say that two integers a, b ∈ Z are equivalent (or congruent) mod n if their difference
a − b is divisible by n. For instance, 3 is equivalent to 9 mod 6, and to 27 mod 6. (The notation
a ≡ b mod n is commonly used to denote this.) The intuition is that upon dividing by 6, 9 and 27
both leave a remainder of 3, which is why they are equivalent to 3 mod 6. In general, this notion of
congruence says that we will only care about remainders when dividing by n, and “identify” things
which give the same remainder.
Any integer is equivalent to precisely one of 0, 1, 2, . . . , n − 1 mod n, so we define the set of
integers mod n to be
Z/nZ = {0, 1, . . . , n − 1}.
(The reason for the notation Z/nZ we are using will become clear later when we discuss quotient
groups, of which this is a basic example. The notation Zn is also commonly used, but we will prefer
Z/nZ since Zn also has other meanings, at least when n is prime.) We define addition mod n to be
addition as usual, only that we interpret the result as an element of Z/nZ depending on what is it
equivalent to. For instance, 3 + 5 = 8 ≡ 2 mod 6, so we would say that

3 + 5 = 2 in Z/6Z.

Here are a few more sums in Z/6Z: 1 + 5 = 0, 4 + 5 = 3, and 2 + 2 = 4. With this operation,
Z/nZ is an abelian group of order n. Indeed, 0 is the identity and the inverse of k ∈ {1, . . . , n − 1}

7
is n − k. In Z/6Z for instance, 1 and 5 are inverses, 2 and 4 are inverses, and 3 is its own inverse
(so it has order 2).

Direct products. Given two groups G and H, the direct product G × H is the group which as a
set is the usual Cartesian product:

G × H := {(g, h) | g ∈ G and h ∈ H}

and whose group operation is given by component-wise multiplication:

(g1 , h1 )(g2 , h2 ) = (g1 g2 , h1 h2 ).

To be clear, here g1 g2 is computed using the multiplication of G and h1 h2 using the multiplication
of H. The identity is (eG , eH ) and inverses are given by (g, h)−1 = (g −1 , h−1 ). We can easily extend
this idea in order to define the direct product of more than two groups.
We mention this construction now in order to state—without proof at this point—that every
finite abelian group is in fact the “same” as a direct product of groups of the form Z/nZ. (We will,
of course, also have to clarify what we mean by “same” here.) This and related facts will go a long
way towards understanding the structure of abstract groups in general, especially finite ones. For
instance, as we will see, it turns out that there are only two “distinct” groups of order 4: Z/4Z and
Z/2Z × Z/2Z.

Multiplication mod n. We define multiplication mod n on Z/nZ in the same way as addition:
multiply as normal but then interpret the result as an element of Z/nZ. For instance:

2 · 3 = 0, 4 · 5 = 2, and 5 · 5 = 1 in Z/6Z.

This first equality in fact implies that neither 2 nor 3 can have a multiplicative inverse in Z/6Z: if
2 had an inverse, we could multiply both sides of 2 · 3 = 0 on left by it in order to obtain 3 = 0,
which is not true, and similarly if 3 had an inverse. The final equality above shows that 5 is its
own multiplicative inverse in Z/6Z.
We define (Z/nZ)× to be the group of elements of Z/nZ which have a multiplicative inverse,
equipped with the operation of multiplication mod n. This is a group since, by definition, we are
only including things which do have an inverse. (The notion × is commonly used to extract elements
from a set which have a multiplicative inverse. For instance, we previously used R× to denote the
set of nonzero real numbers, which are precisely the real numbers which have a multiplicative
inverse, and we could also write Z× = {±1}.) In the n = 6 case, it turns out that

(Z/6Z)× = {1, 5}

since only 1 and 5 have multiplicative inverses mod 6. (The fact that 4 · 3 ≡ 1 mod 6 prevents 4
from having an inverse.)
It is no coincidence that in the n = 6 case the two elements 1, 5 which have multiplicative
inverses are precisely those which are relatively prime to 6—this is true in general:

(Z/nZ)× = {a ∈ Z/nZ | gcd(a, n) = 1}

where gcd denotes the greatest common divisor. The reason as to why comes from a result in
number theory known as Bezout’s Lemma:
The greatest common divisor of a, n ∈ Z is the smallest positive integer which can be
expressed as ax + ny for x, y ∈ Z.

8
(We will not prove this here since this is not a course in number theory, but it is a nice little exercise
to do on your own if you have never seen it before.) With this at hand, we see that there exists
x ∈ Z satisfying ax ≡ 1 mod n if and only if there exists k ∈ Z such that ax − 1 = nk (by definition
of equivalence mod n), which after rearranging as ax + n(−k) = 1 we see is true if and only if the
greatest common divisor of a and n is 1 by Bezout’s Lemma. Thus, we have a complete description
of the multiplicative group (Z/nZ)× ; in particular, if p is prime, then (Z/pZ)× = {1, 2, . . . , n − 1}
consists of all nonzero elements of Z/pZ, which will be an important observation later.

Lecture 3: Dihedral Groups

Warm-Up 1. We say that an element g of a group G generates G if everything in G can be
written as a product of copies of g or its inverse. In other words, G = hgi := {g n | n ∈ Z}, where
we interpret g 0 as the identity and g −k as (g −1 )k . If such g ∈ G exists, we say that G is cyclic. For
instance, all additive groups Z/nZ are cyclic, generated by 1 in each case.
We find all elements which generate the multiplicative groups (Z/7Z)× and (Z/9Z)× , which are
in fact cyclic. First, we have
(Z/7Z)× = {1, 2, 3, 4, 5, 6}.
The powers (computed mod 7) of each of these are:
g g2 g3 g4 g5 g6
1 1 1 1 1 1
2 4 1 2 4 1
3 2 6 4 5 1
4 2 1 4 2 1
5 4 6 2 3 1
6 1 6 1 6 1
Thus we see that 3 and 5 are the only elements which generate all of (Z/7Z)× . (They have order
6, which matches up with the order of (Z/7Z)× .) Note that once we knew 3 was a generator, we
could have immediately concluded that 5 would also be generator, since 5 = 3−1 : saying that all
elements of G can be written in terms of g and g −1 alone, is the same as saying that all elements
can be written in terms of g −1 and (g −1 )−1 = g alone.
For (Z/9Z)× = {1, 2, 4, 5, 7, 8}, we have:
g g2 g3 g4 g5 g6
1 1 1 1 1 1
2 4 8 7 5 1
4 7 1 4 7 1
5 7 8 4 2 1
7 4 1 7 4 1
8 1 8 1 8 1
Thus, 2 and 2−1 = 5 generate (Z/9Z)× , and no other elements do.
In general, it is true that for p prime, the multiplicative group (Z/pZ)× is cyclic. (We will see
this later—next quarter—as a special case of the general fact that “the group of units of a finite
field is cyclic”.) However, this is not an if and only if, as the case of (Z/9Z)× shows.

Warm-Up 2. We show that (Z/8Z)× is not cyclic, by showing that no element generates the
entire group. We have
(Z/8Z)× = {1, 3, 5, 7}.

9
In this case, we have: 32 = 1, 52 = 1, and 72 = 1 mod 7, so no element generates everything, since
such an element would necessarily have order 4 in this case. In particular, all odd powers of 3, 5, 7
result in the 3, 5, 7 respectively, and all even powers result in the identity.

Orders in products. We briefly stated last time that products of the cyclic groups Z/nZ in fact
give rise to “all” possible finite abelian groups, so let us just say a bit more about the structure of
such products. First, note that the (Z/8Z)× example above is an abelian group with 4 elements,
so if our claim about classifications of such groups is accurate, (Z/8Z)× should be the “same”
as either Z/4Z or Z/2Z × Z/2Z. Since (Zn8)× is not cyclic, it cannot be the same as Z/4Z—
which is cyclic—which leaves Z/2Z × Z/2Z as the only possibility. The idea is that if we think of
3 ∈ (Z/8Z)× as corresponding to the element (1, 0) ∈ Z/2Z × Z/2Z, 5 ∈ (Z/8Z)× as corresponding
to (0, 1) ∈ Z/2Z × Z/2Z, 1 as corresponding to (0, 0), and 7 to (1, 1), then the structure of these
two groups are essentially the same:

3 · 5 = 7 mimics (1, 0) + (0, 1) = (1, 1), (5 · 7) = 3 mimics (0, 1) + (1, 1) = (1, 0), etc.

(The correct language is that (Z/8Z)× is isomorphic to (Z/2Z) × (Z/2Z), which is a term we will
soon define.)
Second, note that orders in product groups in general are simple to compute. If g has order
n in G, and h has order m in H, then (g, h) will have order in G × H equal to the least common
multiple of n and m. The point is that if g has order n, then only powers of g which are multiples
of n will result in the identity, and similarly for h and multiples of m, so that both powers “match
up” in the correct way needed to give (eG , eH ) precisely when we take the least common multiple
and multiples of it. For instance, (2, 1) ∈ Z/6Z × Z/8Z has order 3 · 8 = 24 since 2 has order 3 in
the first factor and 1 has order 8 in the second, and (1, 2) has order lcm(6, 4) = 12.

Dihedral groups. For n ≥ 3 , we define the dihedral group D2n to be the set of symmetries/rigid
motions of a regular n-gon, which is a group under composition:

D2n := {symmetries of a regular n-gon}.

(The 2n refers to the number of elements overall. Be aware: another common notation for this
group is Dn where n is instead the number of vertices. Apparently, algebraists prefer to use D2n
whereas geometers prefer Dn . Being a geometer myself, I do prefer Dn , but will use D2n to align
with the book.) We looked at the n = 4 case back on the first day, and similar to that example it
is the case that D2n in general consists of n rotations and n reflections. For instance, D6 , the rigid
motions of a triangle, consists of counterclockwise rotations by 0, 120, 240, and reflections s1 , s2 , s3 ,
each across a line passing through a vertex and the midpoint of the opposite side:

10
Using the labeled vertices, we can compute products such as s1 · 120 = s2 , recalling that on the left
we first rotate by 120 and then reflect using s1 .

Rotations and reflections. Let us take a brief aside to recall/introduce some facts about rotations
and reflections from linear algebra. In the computation s1 · 120 = s2 above, it was in fact possible
to know beforehand that s1 · 120 would result in a reflection, using the observation that rotations
preserve orientation whereas reflections reverse orientation. That is, given a counterclockwise
orientation of the vertices 1 to 2 to 3 to 1, rotations maintain this counterclockwise nature while
reflections turn it into a clockwise ordering:

From this one can see that products of two orientation-preserving transformations are still orientation-
preserving, as are products of two orientation-reversing transformations, while the product of a
transformation which preserves orientation with one which reverses it will itself be orientation-
reversing. Thus, s1 · 120 must be a reflection. (Of course, to see that this reflection is s2 requires
actual computation.)
For another perspective, the rotations and reflections of D2n can each be described via 2 × 2
matrices; for instance  
cos θ − sin θ
sin θ cos θ
describes counterclockwise rotation by an angle of θ. The orientation-preserving or reversing nature
of each is reflected in the determinant: rotations by determinant 1 and reflections have determinant
−1. Then again we see from det(AB) = (det A)(det B) what to expect when composing two
elements of D2n , in terms of whether the result is a rotation or a reflection.

Generators and relations. Returning to D2n in general, using the discussion above we can come
up with a simpler way of describing its elements and group structure. We let r denote clockwise
rotation by 2π n . (We could have very well let r denote the counterclockwise rotation instead,
but again I’ll use the convention the book uses. So, in the case of a square, r denotes what we
previously called 270.) Then 1, r, r2 , . . . , rn−1 ∈ D2n give all the rotations, where we denote the
identity rotation by 1. If we let s denote reflection across the line segment through vertex 1 (which
passes through another vertex when n is even and the midpoint of a side when n is odd), we claim
that s, sr, sr2 , . . . , srn−1 ∈ D2n give all the reflections, so that explicitly

D2n = 1, r, . . . , rn−1 , s, sr, . . . , sr−1 .



Indeed, from above we know that sri in general will be a reflection (orientation-reversing times
orientation-preserving), and these are all distinct since sri = srj implies ri = rj .
The point is that r and s alone generate all of D2n , meaning that every element can be written
solely in terms of products of r’s and s’s and their inverses. Moreover, it turns out that all possible

11
products of elements in D2n can be derived solely from the following relations:

rn = 1, s2 = 1, rs = sr−1 .

The first and second come from the orders of r and s, while the third comes from a direct compu-
tation. (In the case of a square, with upper-right vertex being the one labeled 1, this third relation
becomes 270 · A = A · 90, which holds since both sides are H.) For instance, for something like
r2 sr, we can compute:

r2 sr = r(rs)r = r(sr−1 r) = (rs)r−1 r = rs = sr−1 = srn−1 .

Thus, the entire group structure can be determined from knowledge of r, s, and these relations.
We denote this via the notation:

D2n = r, s  rn = s2 = 1, rs = sr−1 ,

 

which is called a presentation of the group D2n . The first part of the notation “r, s” gives the
generators, and the second “rn = s2 = 1, rs = sr−1 ” the relations. Presentations will be a useful
way of encoding the data of various groups, where the key is that we have abstracted away any
geometric/numerical/whatever meaning, so that we can focus solely on the actual group structure
within.

Lecture 4: Symmetric Groups

Free groups. Before looking at some Warm-Up problems, we first introduce the following concept
in order to give some more context behind the notion of a “presentation”. Given a set S, the free
group on S (or generated by S) is the group hSi whose elements are the “words” made up of the
“letters” in S and their “inverses”:

hSi = {g1 g2 . . . gk | each gi is in S or is the “inverse” of something in S}.

By the “inverse” of something in S, we mean that for each s ∈ S we introduce a new symbol (or
“letter”) s−1 to use in our “words”. Thus, for instance, if S consists of the letters of the English
alphabet, then “algebra” is an element of hSi, as is “a−1 lgeb−1 ra”.
The point is that we give no other meaning to g1 g2 . . . gk and interpret it solely as a “word”
regardless of whether it makes sense to “multiply” gi and gj in some way. The group operation is
simply concatenation of words, were we literally just stick two words together:

(g1 . . . gk )(h1 . . . h` ) = g1 . . . gk h1 . . . h` .

The identity is the empty word consisting of no letters at all, and the only additional condition
we impose is that whenever we see a letter s and its inverse s−1 next to each other, we can get
rid of both them in order to force s−1 to behave like an inverse of s should. So, for instance, the
free group hx, yi on two letters x and y consists of expressions of the form g1 . . . gk where each gi is
x, y, x−1 , or y −1 . Thus, we have the following examples of elements:

x2 y 3 xy −3 x, yxyxyx3 y −2 , and so on.

An example of a product computation is given by:

(xy)(y −1 )x2 = x(yy −1 )x2 = x( |{z} )x2 = x3 .

empty word

12
All free groups on non-empty sets of generators have infinite order. The free group on a single
generator is essentially just Z, if we think of the generator as “1”. (More precisely, any free group
on a single generator is “isomorphic” to Z.)
With this in mind, we can think about a presentation of a group as the result of taking the
free group on the generators and imposing some additional relations on top of that. Thus, for the
dihedral group
D2n = r, s  rn = s2 = 1, rs = sr−1 ,

 

we take the (infinite) free group on two letters r and s, and impose the additional restrictions that
rn and s2 should be the identity, and rs should equal sr−1 . Note how these restrictions turn an
infinite group into one with only eight elements. After we discuss the notion of a quotient group
we will be able to give a more precise meaning to “taking the free group on the generators and
imposing some additional relations.”

Warm-Up 1. We identify, to the extent possible, the groups with the following presentations:

 3
x  x = x5 = 1 , x, y  x2 = y 2 = 1, xy = yx , x, y  x2 = y 2 = 1 .

 
 

In the first case, we note that the given relations imply the following:

x3 = 1 = x5 =⇒ 1 = x2 , so that then x2 = 1 = x3 =⇒ 1 = x.

Thus the presentation hx | x3 = x5 = 1i describes the trivial group whose only element is the
identity. Note that had we been given only a single relation involving a power of x, like hx | x3 = 1i,
we would get a finite cyclic group, so Z/3Z in this case, which is the “only” group generated by a
single element of order 3.
For the second presentation, we have two elements x and y of order 2. (Here, the given relations
will not force x = 1 nor y = 1 upon us, so we use the convention that in such a situation we assume
that the generators are not identities. The only time a generator will be allowed to be an identity
is when that requirement is required by the given relations, as was the case in the first presentation
above.) This at first elements which look like:

xyxyxyxyx, or more generally “words” alternating in x’s and y’s.

There is no need to have more than one x or y in a row since x2 = 1 = y 2 , which also shows there
is no need to an additional symbol for inverses since x−1 = x and y −1 = y. But now, the additional
relation xy = yx (so that x and y commute) allows us to group all the x’s together and all the y’s
together in order to simply this expression down to

xsome power y some power ,

which, depending on types of powers show up here, is either 1, x, y or xy since x2 = y 2 = 1. Thus

this group only has these four elements:

hx, y | x2 = y 2 = 1, xy = yxi = {1, x, y, xy},

and is in fact a presentation of Z/2Z × Z/2Z if we interpret x as (1, 0) and y as (0, 1). (Then xy is
(1, 1).) The observation here is that Z/2Z × Z/2Z is the ”only” group generated by two commuting
elements of order 2.
Finally, the third presentation is similar to the one above, only that we drop the restriction that
the generators commute. Thus we get elements which are alternating expressions in x and y like

xyxyxyxyx

13
only with no way of simplifying this down further. Hence, explicitly this group is:

{g1 . . . gk | each gi is either x or y, with no two x’s or y 0 s occurring consecutively},

with x2 = y 2 = 1 as the only restriction. If we think of x as given a “copy” of Z/2Z and y another
copy of Z/2Z in this group, the resulting group is called the “free product” of Z/2Z with itself, and
is denoted by Z/2Z ∗ Z/2Z. In general, the free product G ∗ H of G and H is similar to the notion
of a free group in that we take “words” formed by the “letters” in G and H, only that here we do
allow ourselves to multiply things in G together and things in H together, but not something thing
in G with something in H; so, we can rewrite something like g1 g2 hg as (g1 g2 )hg where g1 g2 is the
product as computed in G, but we cannot rewrite something like ghg in anyway. In Z/2Z ∗ Z/2Z,
even though both groups are the same, we still do not give any meaning to the “product” xy where
x ∈ Z/2Z and y ∈ Z/2Z since they come from different “copies” of Z/2Z. The notion of a free
product will not play a big role going forward, so no worries if the details are still unclear.

Warm-Up 2. We show that hx, y | x2 = y 2 = 1, (xy)4 = 1i is also a presentation of D8 . Comparing

to the standard presentation D8 = hr, s | r4 = s2 = 1, rs = sr−1 i, we see that neither x nor y behave
like the rotation r. Instead, both behave like reflections due to x2 = y 2 = 1. The claim is that
D8 can in fact be generated by (appropriately chosen) reflections. For instance, if we take x = A
and y = H, then xy = 90, and since 90 and A generate everything, so to do A and H. Moreover,
(xy)4 = 904 = identity holds here, so we see that the given presentation does indeed describe D8 .
The generators are not unique, and picking any two reflections such that xy is either 90 or 270
will work. (In a similar way, even though the book uses r = 270 and s = A in the presentation
hr, s | r4 = s2 = 1, rs = sr−1 i, taking r = 90 and s = H for instance is just as valid and still gives
the correct third relation rs = sr−1 .)
But, note that if we take x = H and y = V , we do not in fact get D8 . In this case x and y
commute since xy = 180 = yx, and xy has order 2 instead of 4. The point is that when we give
the relations as
x2 = y 2 = 1, (xy)4 = 1,
we are again working under the convention that the only additional relations we impose are the
ones we can derive from these, and in this case (xy)2 = 1 does not follow from the relations above
alone. Thus, x = H and y = V does not give a valid way to interpret x and y in the presentation
hx, y | x2 = y 2 = 1, (xy)4 = 1i, and neither does x = A and y = D for instance, since these give
an additional relation (xy)2 = 1 not derivable from the given ones. For x = H, y = V the correct
presentation is
hx, y | x2 = y 2 = 1, (xy)2 = 1i,
which is precisely the same as the second presentation in Warm-Up 1. (The second Warm-Up from
Lecture 2 shows that x2 = y 2 = (xy)2 = 1 is equivalent to x2 = y 2 = 1, xy = yx.) This reflects the
fact that {0, 180, H, V } is a subgroup of D8 which is “isomorphic” to Z/2Z × Z/2Z, and similarly if
we use x = A, y = D instead. We only get a valid description of D8 as hx, y | x2 = y 2 = 1, (xy)4 = 1i
by taking x = A, y = H or V (or vice-versa), or x = D, y = H or V (or vice-versa). Again, this
distinction will become clearer once we give a proper definition of a “presentation” in terms of
quotient groups.

Symmetric groups. Given a set Ω, the symmetric group SΩ on Ω (also called the permutation
group of Ω) is the group of all bijections from Ω to itself:

SΩ := {f : Ω → Ω | f is bijective}

14
under the operation of composition of functions. The identity is the identity function, and inverses
exist because a function is invertible if and only if it is bijective. A bijection will “permute” (i.e.
rearrange) the elements of Ω amongst themselves (hence the name “permutation group”), and we
think of this as being the most general type of “symmetry” possible: there are no geometric (as
in D2n ) nor other constraints on this type of symmetry. One of the basic things we will prove
later is that any (abstract) group can be viewed as a subgroup of a permutation group—in fact in
multiple different ways—which fits with the idea that groups in general are things which are used
to describe “symmetry” in various contexts.
When Ω is finite, say Ω = {1, 2, . . . , n}, we use Sn to denote the corresponding symmetric group
and call it the symmetric group on n letters. (The elements of Ω could have been n other things and
not necessarily the numbers 1, 2, . . . , n, but we can always relabel its elements using these numbers
without changing the actual group structure.) If we take a permutation (i.e. rearrangement) of 1234
like 3124 for instance (say n = 4 in this case), the corresponding function {1, 2, 3, 4} → {1, 2, 3, 4}
is the one which sends each number in 1234 to the number which is now in its location in 3124
after rearranging:
1 7→ 3, 2 7→ 1, 3 7→ 2, 4 7→ 4.
There are n! many such permutations, so Sn has order n!.

Cycles. Thinking of elements Sn as functions is definitely the way to go, but computing products
from this perspective seems challenging without having some better notation available to describe
these functions. (Thinking about elements as rearrangements of 123 . . . n instead makes products
even harder to wrap your head around.) So, we now introduce a convenient notation for describing
arbitrary permutations, which will make much of their structure clearer.
A k-cycle is a permutation (i.e. bijective function) given by the notation

(a1 a2 . . . ak ), where a1 , a2 , . . . , ak ∈ {1, 2, . . . , n} are distinct

which we read as saying that a1 is sent to a2 , a2 is sent to a3 , a3 to a4 , and so on, until at the
end ak is sent back (or, “cycles” back) to a1 . By convention, any number not appearing within the
cycle is assumed to be sent to itself under the corresponding function. For instance, (1234) ∈ S5
describes the function which sends:

1 7→ 2, 2 7→ 3, 3 7→ 4, 4 7→ 1, 5 7→ 5.

Viewed as an element of S6 instead, (1234) does the same as above except that it also sends 6 to
6. The identity permutation is usually denoted simply by (1). Note that a different ordering of the
numbers within a cycle can in fact describe the same cycle; for instance, (123) is the same as (231)
and as (312) since, as functions, they do the same thing. The inverse of a cycle is simply obtained
by reversing the ordering: (a1 a2 . . . ak )−1 = (ak . . . a2 a1 ).
It is a fact that any permutation in Sn can be written as a product of disjoint cycles, where
disjoint means that the cycles share no numbers in common. This product decomposition is unique
up to the ordering in which the cycles appear, as we will see in some examples. Looking at a few
examples makes clear that any permutation can indeed be expressed in this way, but we will give
a proper proof later after we have a few more concepts developed.
For instance, let us compute the product (123)(25)(324) by writing the resulting permutation as
a product a disjoint cycles. We begin by determining the cycle to which 1 belongs. Recalling that
in (123)(25)(324) we read from right to left, we first apply the permutation (324) to 1, resulting in
1 still. (Recall that a number not appearing within a cycle is assumed to be sent to itself.) Next

15
we apply (25) to the resulting 1, which still gives 1. Finally, we apply (123) to 1 to get 2, so overall
the permutation given by the product (i.e. composition) (123)(25)(324) sends 1 to 2:

(123)(25)(324) = (12 . . .) . . .

with the dots representing things still to-be-determined. We continue on with this same cycle, by
next determining what happens to 2: 2 is sent to 4 under (324), and then neither (25) nor (123)
affect 4, so overall 2 is sent to 4:

(123)(25)(324) = (124 . . .) . . .

Next, 4 is sent to 3 under (324), which is left alone by (25) and is then sent to 1 under (123), so
overall 4 is sent to 1, which closes off this first cycle:

(123)(25)(324) = (124) . . .

We begin to determine a next cycle by first looking at the smallest number which has not been
used up yet, in this case 3: 3 is sent to 2 under (324), which is sent to 5 under (25), which is left
alone by (123), so overall 3 is sent to 5:

(123)(25)(324) = (124)(35 . . .) . . . .

We can then work out that 5 is sent to 3, so the second cycle is complete and we have our desired
disjoint cycle expression:
(123)(25)(324) = (124)(35).
The manner in which we construct this disjoint cycle expression, by starting each new cycle with a
number not used up so far, does suggest that we will necessarily end up with disjoint cycles in the
end, but perhaps it is not clear at this point why it is that once a number is used up in a cycle,
it will not appear again as a non-initial point in any further cycle; again, we will prove this later.
Note that disjoint cycles commute, so that we could also write the above as

(123)(25)(324) = (124)(35) = (35)(124).

Finally, we note that orders are easy to compute in cycle form. The order of a k-cycle is precisely
k, since this is the fewest number of “advancements” needed to have each ai cycle back to itself
in (a1 . . . ak ). The order of a product of disjoint cycles is then the least common multiple of the
individual cycle lengths. For instance, the order of

(123)(25)(324) = (124)(35)

is 3 · 2 = 6. Indeed, we have

[(124)(35)]k = (124)(35)(124)(35) · · · (124)(35) = (124)k (35)k ,

| {z }
each (124)(35) k times

where we use the fact that disjoint cycles commute in order to group all the (124)’s together and
(35)’s together, and in order to get the identity k should be a multiple of 3 and 2, the individual
cycle lengths. But a warning: it is only with this disjoint cycle form that this least common multiple
fact works; for instance, (12)(234) has order 4 since it equals (1234), not order 2 · 3 = 6.

16
Lecture 5: Homomorphisms

Warm-Up 1. We show that Sn is generated by its transpositions, which is another name for
2-cycles. Indeed, note that

(a1 a2 . . . ak ) = (a1 a2 )(a2 a3 ) · · · (ak−1 ak ).

This comes from the fact that for i 6= k, ai first appears in (ai ai+1 ) in the product on the right
(reading right to left), and then ai+1 does not appear again as we keep reading to the left, so overall
ai goes to ai+1 for i 6= k. Then, the product on the right sends ak to ak−1 , which is then sent
to ak−2 , which is then sent to ak−3 , and so on until we end up with a1 as the result. Taking an
arbitrary permutation, writing it as a product of disjoint cycles, and then decomposing each cycle
into transpositions as above then proves our claim.
In fact, even more is true: Sn is generated by transpositions of the form (i i+1), which transpose
two consecutive numbers; that is, (12), (23), (34), . . . , (n − 1 n) alone generate Sn . This will be left
to the homework, and comes down to showing that an arbitrary transposition can be written as a
product of these.

Warm-Up 2. Cutting down the list of generators further, we claim that (12) and (123 . . . n)
generate Sn . Indeed, first note that

(12 . . . n)(12)(12 . . . n)−1 = (23),

which comes via a direct computation. (Recall that (12 . . . n)−1 = (n . . . 21).) Next:

(12 . . . n)(23)(12 . . . n)−1 = (34),

which is the same as (12 . . . n)2 (12)(12 . . . n)−2 = (34). In general, one can verify that products of
the form
(12 . . . n)k (12)(12 . . . n)−k
give all the specific transpositions (12), (23), . . . , (n − 1 n). Since these generate Sn , this shows that
(12) and (12 . . . n) generate Sn as well.
Set x = (12) and y = (12 . . . n), so that x2 = 1 and y n = 1. We can then work out a
relation expressing an alternate form for xy, say with y on the left instead. Doing so would yield
a presentation of Sn which looks like:

Sn = hx, y | x2 = y n = 1, some xy relationi.

However, in this case the presentation is unlikely to be so useful, for now at least. When thinking
about Sn we will for now always think of it in terms of explicit permutations, as opposed to a more
abstract generators and relations approach. For dihedral groups, on the other hand, the generators
and relations approach is often simpler to work with. (For n = 3, 4, the geometric definition is
probably just as simple, although for larger polygons the geometry is tougher to see.)

D2n as a subgroup of Sn . Here is another perspective on D2n . Each element in fact induces
a distinct permutation of the vertices of the n-gon, so that to each permutation we can associate
an element of Sn . For instance, in D8 where we take the standard numbering of the vertices of a
square we have used before (upper right vertex is 1, upper left is 2, lower left 3, and lower right
4), the rotation 90 has the following effect on the vertices: 1 7→ 2, 2 7→ 3, 3 7→ 4, 4 7→ 1. Thus,

17
90 corresponds to the 4-cycle (1234). The reflection H gives (23)(14), and other permutations are
simple to compute.
The upshot is that in this way we can view D2n as (isomorphic to) a subgroup of Sn . In fact,
when n = 3 we have that D6 gives every element of S3 since all permutations of the vertices are
realizable by rotations and reflections, but this is not the case for larger dihedral groups. We
thus have three perspectives on what D2n is: a group consisting of rotations and reflections of an
n-gon, an abstract group given by generators and relations, and a permutation (sub)group. Each
perspective will shed light on different aspects of the structure of D2n .

Some other groups. We mention a few more examples of groups which the book introduces at
this point. First, we have already seen the matrix groups GLn (R) and GLn (Z). We can replace R
and Z by other objects, and obtain for instance GLn (C), the group of invertible complex matrices,
and GLn (Z/mZ). (A matrix with entries in Z/mZ is invertible if and only if its determinant is
in (Z/mZ)× , a fact whose proof is similar to analogous fact for GLn (Z).) Of particular interest is
GLn (Z/pZ) for p prime, which we will look more carefully later.
The quaternion group is the group of order 8 given by the elements:

Q8 := {±1, ±i, ±j, ±k}

under multiplication. Here we interpret i, j, k all as “square roots” of −1, so that i2 = j 2 = k 2 = −1.
In addition, the products of these can be determined from

ij = k jk = i ki = h.

This group is non-abelian (for instance one can work out that ji = −k), and is distinct from
D8 . (We will look at more general quaternions a + bi + cj + dk next quarter, as a 4-dimensional
generalization of complex numbers.)

Homomorphisms. So far we have considered standalone groups on their own, but now we move
towards considering how groups relate to one another. The key notion which allows us to do so in
that of a homomorphism:

A homomorphism from a group G to a group H is a function φ : G → H such that

φ(g1 g2 ) = φ(g1 )φ(g2 ) for all g1 , g2 ∈ G. (We say that φ preserves multiplication.)

To be clear, the product g1 g2 on the left uses the operation of G and the product φ(g1 )φ(g2 ) on
the right uses the operation of H. The point is that φ relates these two operations to each other.
One basic consequence of this definition is that φ must send the identity eG of G to the identity
eH of H. Indeed, since eG eG = eG , we have:

φ(eG ) = φ(eG eG ) = φ(eG )φ(eG ),

and multiplying by φ(eG )−1 gives φ(eG ) = eH as claimed. In a similar way, one can show that
φ(g)−1 = φ(g −1 ) for any g ∈ G: φ(g −1 )φ(g) = φ(g −1 g) = φ(eG ) = eH and multiply by φ(g)−1 .

Examples. Here are a couple of first examples. First, for any n ∈ Z, “multiplication by n ” gives
a homomorphism from Z to Z: φ : Z → Z defined by φ(x) = nx for all x ∈ Z. The fact that this
preserves “multiplication” (which is actually addition in this case) is the distributive property:

n(x + y) = nx + ny.

18
 For another example, take det : GLn (R) → R× to be the map which sends an invertible matrix
to its determinant. It is a basic property of determinants that det(AB) = (det A)(det B), which is
the multiplication-preservation requirement.

Another example. Let us determine all homomorphisms Z/5Z → Z/4Z. There are 45 possible
functions between these sets, coming from the fact that each of the 5 elements can be sent to
one of 4 possible values, but the homomorphism condition places a strong restriction on which of
these functions are actually homomorphisms. The key observation is that since the domain Z/5Z
is cyclic, any homomorphism is completely determined by the value of φ(1), or more generally the
value of φ on any generator. Indeed, if we know φ(1), then we know
φ(2) = φ(1 + 1) = φ(1) + φ(1), and φ(k) = φ(1) + · · · + φ(1) .
| {z }
k times

So, we are down to figuring out the possible values of φ(1), of which at first glance there are at
most 4. But now, since 1 has order 5 in the domain, it must be true that
φ(1) + φ(1) + φ(1) + φ(1) + φ(1) = φ(5) = φ(0) = 0,
where at the end we use the fact that homomorphisms send identities to identities. Thus, the value
of φ(1) in Z/4Z must have order dividing 5, which leaves order 1 as the only possibility. Hence we
must have φ(1) = 0, which then forces everything in the domain to be sent to zero, so that there
is only one homomorphism Z/5Z → Z/4Z.

Isomorphisms. We can now make the idea that two groups are the “same” precise:
An isomorphism G → H is a homomorphism which is bijective, or equivalently invert-
ible. We say that two groups G and H are isomorphic if there exists an isomorphism
between them, and denote this by G ∼
= H.
An isomorphism G → H thus gives a way to move back and forth between elements of G and
H, all while preserving the group operations at every step along the way. It is in this sense that
should think of G and H as being the “same” group, since any property which G has H will have,
and vice-versa; the point is that group the structure of isomorphic groups is the same, even if the
elements and group operations “look different”. Various notions we have seen are preserved by
isomorphisms, such as the properties of being abelian or cyclic, orders of elements, the number of
elements of a given order, etc.
If φ : G → H is an isomorphism, then φ−1 : H → G is automatically a homomorphism as well,
which comes from applying φ−1 to both sides of
φ(g1 g2 ) = φ(g1 )φ(g2 )
and doing some rewriting.

Examples. Back in the first Warm-Up of Lecture 3 we showed that the multiplicative groups
(Z/7Z)× and (Z/9Z)× were both cyclic of order 6, which proves that each is isomorphic to the
additive Z/6Z:
(Z/7Z)× ∼
= Z/6Z and (Z/9Z)× ∼ = Z/6Z.
Explicit isomorphisms φ : (Z/7Z)× → Z/6Z and ψ : (Z/9Z)× → Z/6Z can be described by sending
generators to generators: set φ(3) = 1 and ψ(2) = 1, which determine everything else. For instance,
since 32 = 2 in (Z/7Z)× , we have
φ(2) = φ(3 · 3) = φ(3) + φ(3) = 1 + 1 = 2 in Z/6Z.

19
(To be clear, the inputs into φ uses multiplication mod 7, while the outputs use addition mod 6.)
As another example, which we have seen, the subgroup {0, 180, H, V } of D8 is isomorphic to
Z/2Z × Z/2Z. An explicit isomorphism φ : {0, 180, H, V } → Z/2Z × Z/2Z is given by

φ(0) = (0, 0) φ(H) = (1, 0) φ(V ) = (1, 0) φ(180) = (1, 1).

The homomorphism requirement is reflected in the fact that H · V = 180 corresponds to (1, 0) +
(0, 1) = (1, 1). Also note that all non-identity elements in both groups have order 2.

Classifying groups. And now with what we have developed so far, we claim that we can give a
complete list of all groups of order at most 11, meaning that any group of order at most 11 will be
isomorphic to exactly one in this list. To be clear, we do not yet have all the tools needed to prove
that this list is complete, but we have enough to describe the groups in the list.
First, there are the cyclic groups Z/nZ for n ≤ 11. There are also products of these, at least
products whose orders are at most 11. For instance, in the order 9 case, every group of order 9
turns out to be isomorphic to either Z/9Z or Z/3Z × Z/3Z. Next, there are the dihedral groups
D2n for n ≤ 5. The symmetric group S3 has order 6, but this is actually isomorphic to D6 (recall
how the rotations and reflections of a triangle induce every possible permutation of the vertices)
and so is already on the list.
The only remaining group occurs in the order 8 case, where the full list of groups of order 8 is:

Z/8Z Z/2Z × Z/4Z Z/2Z × Z/2Z × Z/2Z D8 Q8 ,

where Q8 is the quaternion group we introduced earlier in this lecture. (This is why we introduced
this group now, so that we could fully characterize all groups of order 8.) We will soon develop the
tools needed to justify all this, and will in fact work our way up towards larger orders, say order
at most 60 if we want a complete list.

Lecture 6: Group Actions

Warm-Up 1. We determine the number of homomorphisms from Z/2Z × Z/4Z to D8 . There are
88 total such functions, but the homomorphism conditions cuts down the possibilities much further.
The key observation is that Z/2Z × Z/4Z is generated by two elements, one (1, 0) of order 2 and
one (0, 1) of order 4, which commute. A homomorphism φ : Z/2Z × Z/4Z → D8 is completely
determined by its values on generators, since in this case for instance we have:

φ(m, n) = φ(m(1, 0) + n(0, 1)) = φ(1, 0)m φ(0, 1)n

to due to preservation of multiplication.

Now, since (1, 0) has order 2 in Z/2Z × Z/4Z, φ(1, 0) must have order dividing 2, meaning order
1 or 2. Thus the only possibilities are:

φ(1, 0) = 0, 180, H, V, D, A.

Since (0, 1) has order 4 in Z/2Z × Z/4Z, φ(0, 1) must have order dividing 4, but everything in D8
has order dividing 4 so this places no restrictions on what φ(0, 1) can be. However, since (1, 0) and
(0, 1) commute, it must be the case that φ(1, 0) and φ(0, 1) commute as well:

φ(1, 0)φ(0, 1) = φ((0, 1) + (1, 0)) = φ((1, 0) + (0, 1)) = φ(0, 1)φ(1, 0).

20
This restricts the possible values of φ(0, 1), since for instance φ(1, 0) = H and φ(0, 1) = D is not
valid since these do not commute.
Thus we can now count the possibilities. When φ(1, 0) = 0 or 180, φ(0, 1) can be anything since
every element of D8 commutes with both 0 and 180. This gives 8 + 8 = 16 homomorphisms so far.
When φ(1, 0) = H or V , then φ(0, 1) can only be 0, 180, H or V since these are the only things
which commute with H or V . This gives another 4 + 4 = 8 possibilities. Finally, when φ(1, 0) = D
or A, we get that φ(0, 1) can only be 0, 180, D, or A, giving 4 + 4 = 8 more homomorphisms. We
conclude that there are 16 + 8 + 8 = 32 homomorphisms Z/2Z × Z/4Z → D8 in total.

Warm-Up 2. Assuming the result that any finite abelian group is isomorphic to a direct prod-
uct of cyclic groups Z/nZ (to be proved later), we determine the product to which the abelian
multiplicative group (Z/15Z)× is isomorphic. Since this group has order 8:

(Z/15Z)× = {1, 2, 4, 7, 8, 11, 13, 14},

the possible products of cyclic groups are:

Z/8Z, Z/2Z × Z/4Z, Z/2Z × Z/2Z × Z/2Z.

(Note that Z/4Z × Z/2Z is isomorphic to Z/2Z × Z/4Z via the homomorphism (a, b) 7→ (b, a) which
swaps components, so it is in fact included in the list above.)
To determine which of these is our group, we compute the orders of elements in (Z/15Z)× .

element 1 2 4 7 8 11 13 14
order 1 4 2 4 4 2 4 2

Right away we can rule out Z/8Z, since there is no element of order 8 and hence (Z/15Z)× is not
cyclic, as would be required in order to be isomorphic to Z/8Z. Since Z/2Z × Z/2Z × Z/2Z has
no element of order 4, this cannot be isomorphic to (Z/15Z)× either since isomorphisms preserve
orders of elements.
Thus (Z/15Z)× must be isomorphic to Z/2Z × Z/4Z. Indeed, the orders of elements match up:
Z/2Z × Z/4Z has four elements of order 4:

(0, 1), (0, 3), (1, 1), (1, 3)

and three elements of order 2:

(0, 2), (1, 2), (1, 0).
If we assume, as we said we would, that any finite abelian group is isomorphic to a product of cyclic
groups, then we are done. But, we will construct an explicit isomorphism

φ : Z/2Z × Z/4Z → (Z/15Z)×

anyway. In general, constructing explicit isomorphisms is not always straightforward, but we will
develop enough theory later so that this is usually not necessary. In this case, let us first set

φ(0, 1) = 2

since we need φ(0, 1) to be an element of order 4 in (Z/15Z)× . (This is not the only possibility.)
Then, since φ should be a homomorphism, we get:

φ(0, 2) = φ(0, 1)2 = 4 and φ(0, 3) = φ(0, 1)3 = 8.

21
Now let us set
φ(1, 0) = 11.
Then everything else is determined:

φ(1, 1) = φ(1, 0)φ(0, 1) = 7 φ(1, 3) = φ(1, 0)φ(0, 1)3 = 13 φ(1, 2) = 14

and of course φ(0, 0) = 1. This then defines an explicit bijection Z/2Z × Z/4Z → (Z/15Z)× , which
is a homomorphism because we computed everything from knowledge of φ(1, 0) and φ(0, 1) alone
precisely in order to force it to be a homomorphism.

Products of cyclic groups. Let us briefly note that it is simple to determine when a product of
cyclic groups is itself cyclic. For instance, it is true that (Z/2Z) × (Z/3Z) is isomorphic to Z/6Z.
Indeed, (1, 1) in the former group has order equal to the least common multiple of 2 (the order of
1 in the first factor) and 3 (the order of 1 in the second factor), and so is 6. Since (Z/2Z) × (Z/3Z)
has order 6 and has an element (1, 1) of order 6, it is generated by this element and hence is
cyclic. Thus (Z/2Z) × (Z/3Z) ∼ = Z/6Z (sending a generator to a generator always gives rise to an
isomorphism), so that Z/6Z is in fact the only abelian group of order 6, up to isomorphism.
In general (Z/mZ)×(Z/nZ) is cyclic, and hence isomorphic to Z/mnZ if and only if m and n are
relatively prime. Indeed, being relatively prime is equivalent to having their least common multiple
be mn, which is the condition needed to have (1, 1)—or anything of the form (generator,generator)—
be of order mn. This extends to products of more than two cyclic groups, so that for instance:

Z/2Z × Z/3Z × Z/5Z ∼

= Z/6Z × Z/5Z ∼
= Z/30Z ∼
= Z/2Z × Z/15Z ∼
= Z/3Z × Z/10Z

are all isomorphic, so that there is only one abelian group Z/30Z of order 30.

Group actions. The notion of a group action gives a general way in which we can view a given
group as a group of “symmetries”, in multiple ways. First, we give some notation. Given a function

G×X →X

where G is a group and X a set, we will denote the element of X to which (g, x) is sent as g · x,
which we think of as “g acting on x”:
(g, x) 7→ g · x.
An action of G on X is then a map G × X → X such that:

• e · x = x for all x ∈ X, and

• g · (h · x) = (gh) · x for all g, h ∈ G and x ∈ X.

The first requirement says that the identity should “act as the identity”, and the second says that
“composing” actions should correspond to multiplication. We will often use the notation G y X
to mean that G acts on X, or in other words that there is a group action of G on X.

Examples. GLn (R) acts on Rn by the usual product of a matrix and a vector: GLn (R) y Rn
defined by A · x = Ax for A ∈ GLn (R) and x ∈ Rn . Certainly the identity matrix I satisfies Ix = x
for all x ∈ Rn , and the second requirement in the definition of an action comes from the way in
which the product of two matrices is defined: A(Bx) = (AB)x.
The definition we gave for the symmetric group Sn comes precisely from viewing Sn as acting
on {1, 2, . . . , n}: Sn y {1, 2, . . . , n} is defined by σ · x = σ(x) for all x ∈ {1, 2, . . . , n}, where σ(x)

22
is the result of applying the bijection σ to x. And as a final example, the fact that any rotation
or reflection of a regular n-gon permutes the vertices can be viewed as defining an action of D2n
on {1, 2, . . . , n}: label the vertices of the n-gon using 1, 2, . . . , n, and define D2n y {1, 2, . . . , n}
by having g ∈ D2n permute the vertices. For instance, if we take the standard counter-clockwise
labeling of the vertices of a square with 1 in the upper-right corner, the rotation 90 ∈ D8 acts as:
90 · 1 = 2, 90 · 2 = 3, 90 · 3 = 4, 90 · 4 = 1; in other words, 90 acts as the cycle (1234).

Homomorphisms to permutation groups. The data of an action of G on X can be rephrased

more compactly as follows. For each g ∈ G, acting on each element of X by g induces a function
x 7→ g · x from X to X. This function is in fact bijective, with inverse given by the function induced
by g −1 ; that is, the composition of the function induced by g with the function induced by g −1 , in
either order, is the identity function:

g −1 · (g · x) = (g −1 g) · x = e · x = x

and
g · (g −1 · x) = (gg −1 ) · x = e · x = x
for all x ∈ X. Thus we can view the action as giving a map G → SX , where SX is the group of
permutation (i.e. bijective functions) on X:

send g ∈ G to the permutation defined by x 7→ g · x.

The second requirement in the definition of an action then says precisely that this map G → SX
is a homomorphism, since multiplication in G corresponds to composition in SX . Conversely, given
a homomorphism G → SX , we can get an action of G on X by having g ∈ G act on X by
the corresponding element of SX . The upshot is that an action of G on X is the same as a
homomorphism G → SX , and it is in this way that we can view an action as giving a way to
interpret G as a group of permutations on X. The power of this idea will come from applying it to
different possible sets X in order to extract information about G.

Lecture 7: Some Subgroups

Warm-Up 1. Given an action of G on X and an element x ∈ X, we define two important sets

associated with the action: the orbit Gx through and the stabilizer Gx of x by:

Gx := {g · x | g ∈ G} and Gx := {g ∈ G | g · x = x}.

The orbit through x is thus the subset of X consisting of all things obtained by acting on x by
various group elements, and the stabilizer of x is the subset of G consisting of all group elements
which fix, or stabilize, x. In fact, stabilizers are subgroups of G, as can be seen from:

(gh) · x = g · (h · x) = g · x = x and g −1 · x = g −1 · (g · x) = (g −1 g) · x = e · x = x

when g · x = x and h · x = x.
Let SO3 (R) denote the group of 3×3 orthogonal matrices of determinant 1, which geometrically
represent 3-dimensional rotations. (The “SO” stands for “special orthogonal”; the term “special”
in this context refers to the determinant 1 condition.) Then SO3 (R) y R3 by matrix multiplication.
We describe the orbits and stabilizers of this action.

23
 For x ∈ R3 , applying any rotation to x results in a vector of the same length as x, and as we
vary the rotation used we obtain all possible such vectors. This says that the orbit through x is
the sphere centered at the origin of radius x:

SO3 (R)x = sphere of radius x centered at 0.

(We allow here for a sphere of radius 0, which arises when x = 0.) The orbit space of this action,
which is the set of orbits, is the collection of all spheres centered at the origin.
For x ∈ R3 , the stabilizer of x consists of all those rotations which leave the vector x unchanged:

SO3 (R)x = {A ∈ SO3 (R) | Ax = x}.

(In more linear-algebraic terms, this is the set of matrices which have x as an eigenvector of
eigenvalue 1, at least in the x 6= 0 case.) For x 6= 0, any such rotation can be viewed as occurring
around the “axis” spanned by x, or equivalently as a rotation of the plane passing through the
origin which is orthogonal to x. This subgroup of SO3 (R) is in fact isomorphic to SO2 (R), the
group of 2-dimensional rotations. For instance, if x is on the z-axis, then the plane orthogonal to x
is the xy-plane, which we can identity with R2 , so that rotations of the xy-plane around the z-axis
are indeed the same thing as elements of SO2 (R). An explicit isomorphism SO2 (R) → SO3 (R)x in
this case is given by:  
  cos θ − sin θ 0
cos θ − sin θ
7→  sin θ cos θ 0
sin θ cos θ
0 0 1
where θ is the angle of rotation. By viewing any plane through the origin as isomorphic (in the
linear-algebraic) sense to R2 , we can similarly view any stabilizer of this action as being isomorphic
to SO2 (R). (The isomorphism between the stabilizer and SO2 (R) in general is tougher to write
down, but can be obtained via a “change of basis”, which we will recall after the second Warm-Up.)

Warm-Up 2. We show that g·h := ghg −1 defines an action of a group G on itself, which we call the
conjugation action. We also show that this is actually an action by automorphisms, which means
that the bijection on G induced by each g ∈ G is actually a group isomorphism. (An isomorphism
G → G is called an automorphism of G, and the set Aut(G) of all such automorphisms is a
group under composition. The claim here is that the map G → SG obtained from the conjugation
action actually has image contained in Aut(G) ⊆ SG , and so can be viewed as a homomorphism
G → Aut(G). In general, actions coming from homomorphisms G → Aut(H), where H is some
group, will play a special role in understanding the structure of abstract groups.)
First we have:
e · h = ehe−1 = ehe = h for all h ∈ G,
so e acts as the identity. Second, for any g, k ∈ G, we have:

g · (k · h) = g · (khk −1 ) = gkhh−1 g −1 = (gk)h(gk)−1 = (gk) · h

for all h ∈ H. Thus conjugation is indeed an action of G on itself. (Note here we use the fact that
(ab)−1 = b−1 a−1 is true in any group.)
Finally, to say that this is an action by automorphisms means that:

g · (h1 h2 ) = (g · h1 )(g · h2 )

24
for all g, h1 , h2 ∈ G, since this is what it means for the induced map G → G defined by h 7→ g · h
to preserve multiplication. We check:

g · (h1 h2 ) = gh1 h2 g −1 = gh1 (g −1 g)h2 g −1 = (gh1 g −1 )(gh2 g −1 ) = (g · h1 )(g · h2 )

as required. Thus G acts on itself by automorphisms under conjugation.

Example. Let us give an example of the conjugation action, which has a well-known interpretation
in linear algebra. Take GLn (R) acting on itself by conjugation. Two matrices A, B ∈ GLn (R) are
conjugate—meaning one can be obtained from the other after conjugating by some matrix—if there
exists S ∈ GLn (R) such that B = SAS −1 . (In other words, A and B lie in the same orbit of the
conjugation action.) The standard term for this scenario in linear algebra is that A and B are
similar matrices, so similarity is just an instance of group conjugacy. In general, the orbits of the
conjugation action are called the conjugacy classes of the group, so here to say that two matrices
are similar just means that they lie in the same conjugacy class.
Now, you might recall the important fact from linear algebra that similarity corresponds exactly
to performing a change of basis; that is, similar matrices describe the same linear transformation
only with respect to possibly different bases. To make a connection with the first Warm-Up, recall
we noted above that the stabilizer of x under the multiplication action of SO3 (R) consisted of the
subgroup of rotations which rotated the plane orthogonal to x, and that this could be viewed as
being isomorphic to the group SO2 (R) of 2-dimensional rotations. Now we can be more precise:
any element of the stabilizer SO3 (R)x is similar to an element of the “copy” of SO2 (R) in SO3 (R)
consisting of those matrices of the form
 
cos θ − sin θ 0
 sin θ cos θ 0
0 0 1

which described rotations of the xy-plane. If we pick a basis for R3 consisting of x (in the x 6= 0
case) and two vectors orthogonal to x and to each other (so spanning the plane orthogonal to x),
the matrix of a linear transformation in the stabilizer with respect to this basis will indeed be
equal to one of the form above. Thus, by conjugating matrices of the form above, we can obtain
rotational matrices around any axis spanned by a nonzero vector, and these rotational matrices
make up the stabilizer of that vector. In this case then, the conjugacy class of a matrix of the form
above consists of all possible rotations of the same angle θ occurring around any axis.

A few more adjectives. The book describes certain properties an action may or may not posses,
such as being faithful, or free, or transitive, but we will put-off introducing these terms until they
are actually needed.

Subgroups. We have already been informally using the notion of a subgroup of a group G, and
now we will make this precise. A subgroup H of G is a nonempty subset of G which is:

• closed under multiplication: hk ∈ H for all h, k ∈ H where hk is the product in G, and

• closed under inversion: h−1 ∈ H for all h ∈ H, where h−1 is the inverse in G.

The point is that H is then itself a group in its own right, under the same operation as on G, only
sitting inside of the “larger” group G. (Note that eG ∈ H follows from the two properties above
and the property of H being nonempty: for h ∈ H, h−1 ∈ H, and thus hh−1 = eG is as well.) We
use H ≤ G notation for H being a subgroup of G.

25
 As the book shows, there are equivalent ways of rephrasing the subgroups conditions: both can
be combined into the single requirement that hk −1 ∈ H for h, k ∈ H, and in the finite case the
“closed under inversion” property can be dropped since the inverse of any element is a positive
power of that element due to the fact that it has finite order. But, we will note really use these
shorter descriptions, apart from in a few specific examples later.

Kernels and images. To any homomorphism φ : G → H we can associate two important

subgroups, one of G and one of H: the kernel ker φ ≤ G and the image φ(G) ≤ H defined by

ker φ := {g ∈ G | φ(g) = eH } and φ(G) := {φ(g) ∈ H | h ∈ G}.

So, in words, the kernel is the set of all things which map to the identity, and the image is the set of
all elements obtained as outputs. That these are indeed subgroups of G and H respectively follows
from basic properties of homomorphisms, namely: φ(g1 g2 ) = φ(g1 )φ(g2 ) and φ(g)−1 = φ(g −1 ). We
will omit the precise check here, but it should be straightforward. We will discuss the importance
of kernels and images later.

Centralizers and normalizers. We finish by introducing two more examples of subgroups, which
will play an important role in topics to come. For any subset A ⊂ G, we define its centralizer CG (A)
and normalizer NG (A) as follows:

CG (A) := {g ∈ G | gag −1 = a for all a ∈ A} and NG (A) := {g ∈ G | gAg −1 = A}.

(To clarify the notation in the second expression, gAg −1 denotes the set of all elements of A of the
form gag −1 for a ∈ A, and is called the conjugate of A by g.)
Since gag −1 = a is equivalent to ga = ag, we thus see that the centralizer of A in G is precisely
the set of all elements of g which commute with everything in A:

CG (A) = {elements of G which commute with all elements of A}.

To say that gag −1 = a is to say that g centralizes a, or in other words stabilizes a under the
conjugation action of G on itself. Now, the condition gAg −1 = A in the definition of the normalizer
does not mean that g commutes with all things in A, only that the set of things obtained by
conjugating elements of A by g is the same as the set A itself; the point is that gag −1 will be an
element of A, but will not necessarily equal a itself. (So, note that CG (A) is contained in NG (A).)
The check that these are both subgroups of G is straightforward, and we omit it here.
In the special case where A = G, the centralizer CG (G) of G in itself is most commonly called
the center of G and is denoted by Z(G). Thus, the center Z(G) is the subgroup of G consisting of
all elements which commute with every element of G. Note G is abelian if and only if Z(G) = G.

Example. A problem on Homework 1 computed the center of D2n . The result was that
(
{1} when n is odd
Z(D2n ) = n/2
{1, r } when n is even.

For instance, 180 commutes with everything in D8 , and so belongs to the center of D8 .

Another example. Take the subgroup hri of rotations in D2n . Certainly any rotation commutes
with any other rotation, so all rotations belong to the centralizer of hri. However, no reflection will
commute with all rotations, since in particular no reflection commutes with r itself:

r(sri ) = sr−1 ri = sri−1 6= (sri )r = sri+1 since ri−1 6= ri+1 .

26
Thus CD2n (hri) = hri only consists of the rotations.
However, the normalizer of hri is in fact all of D2n , which means that conjugating a rotation by
any element of D2n will always result in a rotation. One geometric reason is the fact that composing
two reflections and a rotation or three rotations always results in a rotation, by keeping track of the
orientation-reversing or orientation-preserving properties. Alternatively, we can compute directly
(using s−1 = s) that:

(sri )rj (sri )−1 = sri rj r−i s = sri+j sri = s2 r−i−j ri = r−j ∈ hri.

Thus ND2n (hri) = D2n as claimed. (This fact, that the normalizer is the entire group, is what it
means to say that hri is a normal subgroup of D2n . We will come back to normal subgroups, and
their crucial role, soon enough.)

Lecture 8: Cyclic Groups

Warm-Up 1. We give interpretations of centralizers and normalizers in terms of appropriately

chosen group actions. Let G be a group with A ⊆ G. Just as we introduced the action of G on
itself by conjugation, we can also define a conjugation action of G on its power set 2G , which is the
set of all subsets of G:
G y 2G := {S | S ⊆ G} via g · S := gSg −1 .
Recall here that gSg −1 is the set of all elements of G expressible as gsg −1 for some s ∈ S. (The
notation 2G for the power set of G is standard and comes from recognizing that to specify a subset
of G is the same as to specify a function from G to a 2-element set, but we will not elaborate on
this here.) That this is an action of G on 2G can be checked directly.
The point then is that the normalizer NG (A) of A in G is precisely the stabilizer of A under
this conjugation action of G on 2G :

g ∈ the stabilizer GA ⇐⇒ gAg −1 = A ⇐⇒ g ∈ NG (A).

This gives one perspective on why NG (A) is a subgroup of G, since stabilizers are always subgroups.
Next, consider now the action of this normalizer NG (A) on A, also by conjugation:

g ∈ NG (A) acts as g · a := gag −1 .

(Note here that we act by the normalizer NG (A) instead of the entire group G solely to guarantee
that the induced maps actually map A into A: each g ∈ NG (A) defines a bijection g : A → A by
conjugating elements. For g ∈ / NG (A)), the induced map would map A into some other set, and
so would not be considered a permutation of A.) Each stabilizer of this action consists of those
g ∈ NG (A) which stabilizer some a ∈ A, and so the centralizer CG (A) of A in G can be viewed as
the intersection of all such stabilizers:

g ∈ CG (A) ⇐⇒ gag −1 = a for all a ∈ A ⇐⇒ g ∈ the stabilizer NG (A)a for all a ∈ A

T
if and only if g ∈ a∈A NG (A)a . Since intersections of subgroups are always subgroups, this again
gives a reason as to why centralizers are always subgroups. It is common to use the notation CG (a)
to denote the centralizer of the singleton set {a}, which is the centralizer
T of a under the conjugation
action above. With this notation we can write the stabilizer of A as a∈A CG (a).
But here is one more perspective. The conjugation action of NG (A) on A can be viewed as
defining a homomorphism φ : NG (A) → SA , where SA is the permutation group of A. The

27
centralizer CG (A) is then the kernel of this homomorphism: g ∈ ker φ if and only if the map
induced by g on A is the identity function, which is true if and only if gag −1 = a for all a ∈ A.
Kernels are subgroups, and so centralizers are subgroups too.

Warm-Up 2. We determine the center of the group GL2 (R). (The answergeneralizes
 to GLn (R)
a b
for all n, but we use n = 2 just to simply the computation.) Suppose A = c d . This is in center
when AX = XA for all X ∈ GL2 (R). In particular, this requires that
     
a b 1 1 1 1 a b
= ,
c d 0 1 0 1 c d

which gives the conditions

a=a+c a+b=b+d c + d = d.
 
Hence we need c = 0 and a = d, so that A looks like A = a0 ab . Next, we also need
     
a b 1 0 1 0 a b
= ,
0 a 1 1 1 1 0 a

which boils down to a + b = a, so that b = 0. Thus A = [ a0 a0 ], so we conclude that the center

consists of nonzero scalar multiples of the identity:

Z(GL2 (R)) = {aI | a ∈ R× }.

The analogous statement holds for lager matrices, as can be seen by using other matrices in place
of [ 10 11 ] and [ 11 01 ], namely those which are the identity with a 1 in some other location.

Subgroups of cyclic groups. We now seek to give a complete description of all subgroups of
a given cyclic group. To get a sense for the claims we will make, let us list concretely all the
subgroups of Z/6Z:
{0}, {0, 2, 4}, {0, 3}, {0, 1, 2, 3, 4, 5}.
The first observation is that each of these are cyclic, generated by 0, 2 or 4, 3, and 1 or 5 respectively.
This holds true in general: any subgroup of a cyclic group is cyclic.
Indeed, suppose G = hxi is cyclic, generated by x, and let H ≤ G be a nontrivial subgroup.
Then all elements of H (and G) can be expressed as powers of x, so let xk for some k > 0 be the
smallest positive power of x which is in H. (Such a positive power exists, since if a negative power
is in H, its inverse will be positive power which is necessarily in H.) We claim that H is in fact
generated by xk . Pick any h ∈ H. Since h ∈ G we can express h as a power of x, say h = x` for
some ` ∈ Z. By the division algorithm there exists q ∈ Z and r ∈ {0, 1, . . . , k − 1} such that

` = kq + r.

But then we have:

xr = x`−kq = x` (xk )−q ,
which is a product of elements of H is thus in H. But xk was meant to be the smallest positive power
of x which was in H, so since 0 ≤ r < k we must have r = 0. Thus ` = kq, so x` = (xk )q ∈ hxk i,
which shows that H = hxk i.

28
Orders in cyclic groups. Next we describe the order of any element in a cyclic group. We will
restrict ourselves to finite cyclic groups, but the answer for an infinite cyclic group G = hxi is that
all non-identity elements have infinite order. Going back to the case of Z/6Z, we see that

2, 4 have order 3 3 has order 2 1, 5 have order 6.

The observation is that 3 is precisely 6 divided by the greatest common divisor of 6 and 2 (or 4),
2 is 6 divided by gcd(3, 6), and 6 is 6/gcd(6, 1) or 6/gcd(6, 5). The claim is that if G = hxi is finite
n
of order n, then xk (for k 6= 0) has order gcd(n,k) .
We will give the proof for completeness, but note that it really is more of a proof about divis-
ibility than it is about group theory. Set d = gcd(n, k) and write n and k as n = ds, k = d` for
s, ` ∈ Z which are relatively prime. (If they were not relatively prime, there would exist a common
n
divisor of n and k larger than d.) With this notation, gcd(n,k) = nd = s, so we first check:

(xk )s = xks = xd`s = (xn )l = 1

since x has order n. For now denote the order of xk by t, so that the above implies t divides s.
Next, since xk has order t we have
(xk )t = xkt = 1,
so that n (the order of x) divides kt. Using the expressions for n = ds and k = d` from above, this
gives that ds divides d`t, which implies s divides `t. Since s and ` are relatively prime, we have
n
that s divides t. Hence s and t are positive and divide each other, so we get gcd(n,k) = s = t = |xk |
as required.

Counting subgroups of cyclic groups. We know that all subgroups of a cyclic group are cyclic,
and the final fact we need in order to describe all such subgroups (in the finite case) more explicitly
is the following: if G is cyclic of order n, then for each divisor k of n there is a unique subgroup
of G of order k, and moreover these are the only subgroups of G. Thus, the subgroups of a finite
cyclic group correspond precisely to the divisors of n in a 1-to-1 manner. We will postpone the
proof of this to next time.

Lecture 9: Generating Sets

Warm-Up 1. We identity the automorphism group Aut(Z/nZ) in terms of a better known group.
Since the domain is cyclic, an automorphism ψ : Z/nZ → Z/nZ is fully determined by its value
on a generator, so in particular by the value of ψ(1). This element generates a subgroup of Z/nZ,
and so in order for ψ to be surjective (which is equivalent to injective since Z/nZ is finite) this
subgroup must be all of Z/nZ, meaning that ψ(1) must be a generator of Z/nZ. By what we know
about the structure of cyclic groups, k ∈ Z/nZ has order
n
,
gcd(k, n)

and thus this equals n if and only if gcd(k, n) = 1. We conclude that ψ(1) must be relatively prime
to n in order for ψ to be an automorphism.
The number of positive integers smaller than n which are relatively prime is commonly denoted
by φ(n), where φ is called Euler’s φ-function, so the order of Aut(Z/nZ) is φ(n). But (Z/nZ)× is
also a group of order φ(n), and indeed we claim that Aut(Z/nZ) ∼ = (Z/nZ)× . This can be seen

29
by determining the group structure of Aut(Z/nZ) explicitly: the automorphism determined by
ψ(1) = m is explicitly defined by
ψ(k) = mk,
and composition amounts to multiplying these “m”’s:

ψ1 (ψ2 (k)) = ψ1 (m2 k) = m1 m2 k

in a way which corresponds precisely to multiplication mod n. The identity automorphism cor-
responds to m = 1, and the inverse of ψ is the automorphism determined by ψ −1 (1) = m−1 , the
multiplicative inverse mod n. Thus Aut(Z/nZ) is indeed isomorphic to (Z/nZ)× .

Warm-Up 2. We prove the final claim leftover from last time if G = hxi is cyclic of order n, then
for each divisor k of n there is a unique subgroup of G of order k, and these are the only subgroups
n
of G. Indeed, if k divides n, then x k has order
n n
n = = k,
gcd(n, k ) `
n
where ` ∈ N satisfies n = k`. (Note gcd(n, `) = ` since ` itself divides n.) Thus hx k i has order k.
n
To show that hx k i is the only subgroup of G of order k, suppose H ≤ G also has order k. Then
H, being a subgroup of a cyclic group, is cyclic, so it is generated by some xs ∈ H. Then
n n
= k = |H| = |xs | = ,
` gcd(n, s)
so ` = gcd(n, s), and hence in particular ` is a divisor of s, say s = `b. But then any power of xs is
also a power of x` :
(xs )a = xas = xa`b = (x` )ab ,
n
so H = hxs i is contained in hx` i. Since H and hx` i = hx k i both have order k, we thus conclude
n n
that H = hx k i, so that hx k i is indeed the only subgroup of G of order k.
To see that there are no other subgroups of G (it is true in general that the order of any subgroup
has to divide the order of the entire group, but we do not know this yet), take an arbitrary subgroup
n
of G, which is thus cyclic. It is generated by some xk , so the order of this subgroup is then gcd(n,k) ,
which is in fact a divisor of n. Thus this subgroup is already among the ones determined above.

Subgroup lattices. The subgroup lattice of a (finite) group is a picture which contains all of
its possible subgroups, arranged according to which contains which. At this point we can thus
determine the full subgroup lattice of any cyclic group. For instance, the subgroups of Z/12Z are:

h0i = {0}, h1i = h5i = h7i = h11i = Z/12Z, h2i = h10i, h3i = h9i, h4i = h8i, h6i.

We draw them as follows, where a line segment indicates containment going upwards:

30
This is then the subgroup lattice of Z/12Z.
So far we can determine the full subgroup lattice of some other small groups of finite order,
such as D8 for instance, but this becomes more of a challenge as the order increases. But, at least
for finite cyclic groups, we can determine the full lattice with relative ease.

Why do we care? Let us now go back to a motivation from the first day of class, to see why the
2 2
subgroup
√ √ lattice is a useful thing to have. Recall that the polynomial (x − 2)(x − 3) has roots
± 2, ± 3, which can be permuted (in a still vague way which √ “preserves algebraic equations”
√ and
rules out other permutations) in four ways: fix or transpose ± 2, and fix or transpose ± 3. This
group of permutations is isomorphic to Z/2Z × Z/2Z, where each factor of Z/2Z controls whether
the same ± roots are transposed.
The fact is that each “branch” of the subgroup lattice of Z/2Z × Z/2Z corresponds to one way
of “introducing” the roots of (x2 − 2)(x2 − 3):

(Here the contains go from right to left.)

√ To start, note the coefficients of this polynomial
√ all lie in
Q. In order to construct the roots ± 2, which do not exist in Q, we need to “adjoin” 2 to Q to
construct a larger set of numbers:
√
Q → Q ∪ { 2} ∪ {anything else needed to get a “consistent” set of numbers}.

We will give a precise meaning to “anything else needed to get a consistent set of numbers”, √ and to
what “consistent” means in this context,
√ in the√spring, but the idea is that once we introduce 2, we
should also introduce things like 1+ 2, −4+2 2, and anything else obtained by adding/multiplying
the things we have so far. (The√ precise language we will come to later is that we want the √ smallest
field which contains Q and 2.) At this point we are able to describe the roots ± 2 of our
polynomial,
√ so in order to be able to describe all roots we need only go one step further and
“adjoin” 3: √ √ √
Q → Q ∪ { 2} ∪ {stuff} → Q ∪ { 2} ∪ { 3} ∪ {more stuff}.
We claim that that the Q at the start corresponds to the full permutation group Z/2Z × Z/2Z (in
a way which is to be determined later), and this specific way of “adjoining” elements corresponds
to passing through one branch of the subgroup lattice, as labeled
√ in the picture above.
√ The bottom
branch corresponds to the process where we first introduce 3 as a root, and then 2.
√ √ √
Q → Q ∪ { 3} ∪ {stuff} → Q ∪ { 3} ∪ { 2} ∪ {more stuff}.
√ √ √
(The middle branch in the picture above corresponds to adjoining 2 3 = 6 first, but what is
actually going on here requires clarification later.)
The upshot is that the structure of the subgroup lattice will eventually correspond to ways in
which we can introduce the roots of a given polynomial. The ability or non-ability of finding a
formula for all the roots will then be essentially controlled by the lattice itself.

31
Subgroups generated by sets. Cyclic subgroups are generated by single elements, but more
generally we can consider subgroups generated by arbitrary subsets of a group. If S ⊆ G, we define
the subgroup generated by S, denoted hSi, to be the smallest subgroup of G which contains S.
Here, “smallest” means with respect to set inclusion: if H is any subgroup of G which contains S,
then hSi ⊆ H. The point is that we include in hSi the set S and the bare minimum number of
things needed in order to guarantee we actually get a subgroup, but no more.
You will see that this a different definition than what the book gives, but is in fact equivalent
to it, as we will now prove. One remark is that the definition above does not actually say anything
about such a “smallest” subgroup in fact exists, and indeed this is not at all obvious. We will often
define objects in the coming quarters as being the “smallest” ones which have some property, but
will still need to argue that something satisfying that definition does exist. This is likely why the
book gives a more concrete definition for hSi, but I prefer to use definitions which highlight the
point of the object being defined (the point is that hSi is the smallest subgroup containing S), and
show their existence afterwards.
So, the claim is that the definition we have given is equivalent to the equality
\
hSi = H,
S⊆H≤G

which is the book’s definition. (This intersection is taken over all subgroups H of G which contain
S.) It is always true that the intersection of any number of subgroups is a subgroup, so the key
point is why this intersection is the smalleset subgroup containing S in the sense we defined above.
In this case, the answer is simple: if K is any subgroup of G containing S, then K occurs as one
of the subgroups H we are intersecting, and so
\
H⊆K
S⊆H≤G

simply
T because intersections are always subsets of each of the sets being intersected. This shows
that S⊆H≤G H satisfies the defining property of being the smallest subgroup of G containing S,
so we are done. Next time we will describe the elements of hSi more explicitly.

Lecture 10: Zorn’s Lemma

Warm-Up 1. We show that, explicitly, the subgroup generated by S ⊆ G is given by:

hSi = {sn1 1 · · · snk k | each si ∈ S, ni ∈ Z}.

That is, hSi consists of all products of elements of S and their inverses. This just mimics the
definition we gave of “free groups” earlier, only that here the group is not “free” since there is a
predetermined multiplication on the elements of S we are using—namely that of G—along with
whatever relations that entails.
First, the given set above is indeed a subgroup of G, simply because multiplying or inverting
products of elements of S and their inverses still results in a product of elements of S and their
inverses. Second, if H is a subgroup of G which contains S, then for each s1 , . . . , sk ∈ S and
n1 , . . . , nk ∈ Z, we have sni i ∈ H because H is a subgroup, and hence sn1 1 · · · snk k ∈ H for the same
reason. Thus H contains the subgroup defined above, so the subgroup defined above is the smallest
subgroup of G containing S as claimed.

32
Warm-Up 2. We show that the multiplicative group Q×
>0 of positive rational numbers is generated
by the reciprocals of the prime numbers:

Q× 1
>0 = { p | p is prime}.

(If we allowed negative reciprocals, we would get all of Q× .) This is simply a reflection of the fact
that any integer has a prime factorization. Indeed, let ab ∈ Q× >0 with a, b ∈ N, and write each of
a, b as a product of primes:
a = ps11 · · · pskk , b = q1t1 · · · q`t`
where each pi , qj is prime. Then:
 −s1  −sk  t1  t`
a 1 1 1 1 1
= ps11 · · · pskk t1 = ··· ··· ,
b q1 · · · q`t` p1 pk q1 q`

which is a product of reciprocals of primes as required.

Finite generation. We say that a group G is finitely generated if there is a finite subset S ⊆ G
which generates it. In general, finitely generated groups are simpler to work with, since having
a finite number of generators can help give explicit ways of performing various computations.
Certainly any finite group is finitely generated (it generates itself), but infinite groups can be
finitely generated as well: Z is generated by {1}, the free group on two letters hx, yi is generated by
{x, y}, and GLn (R) is generated by the finite set of so-called elementary matrices for instance. (An
elementary matrix is one which is obtained from the identity by performing one elementary row
operation. The fact that any invertible matrix can be written as a product of these alone follows
from the method commonly used in a linear algebra course to compute inverses by row-reducing.)
The product of an infinite number of copies of Z is not finitely-generated for instance, and
neither is the group Q× >0 from the second Warm-Up. Note that this is not obvious: it is not so
difficult to see that the generating set of the reciprocals of primes cannot be cut down to a finite
generating set—essentially because are not divisible by other primes—but the claim is that no finite
subset can generate the entire thing, which takes more work to prove.
In some sense, the notion of a finitely-generated group is an analog of that of a finite-dimensional
space in linear algebra. For instance, consider the subgroup of Rn generated by v1 , . . . , vk ∈ Rn .
Since the group operation here is addition, we will use additive instead of multiplicative notation.
Then we have concretely:

hv1 , . . . , vk i = {m1 v1 + · · · + mk vk | each mi ∈ Z}.

The observation is that this consists of linear combinations of v1 , . . . , vk , namely those with integer
coefficients. This subgroup is thus analogous to the span of v1 , . . . , vk , only not actually the entire
span since we are not using arbitrary real coefficients. But nevertheless, saying that a group is
finitely-generated is essentially the group-theoretic equivalent of saying that a space is spanned by
finitely many vectors in linear algebra. (We will consider these analogies more closely next quarter
in the context of modules over rings.)

Maximality. To motivate the discussion of what’s called Zorn’s Lemma we will now undertake,
we give the following definition: a proper subgroup H of G is said to be a maximal subgroup if it
is not contained in a strictly larger proper subgroup, i.e. if H ≤ K ≤ G, then H = K or K = G.
(A proper subgroup is one which is not equal to the entire group.) For instance, the maximal
subgroups of Z/nZ are those generated by the primes dividing n.

33
 A question we can ask is whether it is true that any group G contains subgroups which are
maximal, or, even better, whether any given prescribed subgroup H is contained in a maximal
subgroup. (That is, can H be “enlarged” to a maximal subgroup?) This specific question will not
be so important for us this quarter in the context of groups, but the idea of “maximality” in this
sense will be much more important in subsequent quarters in the context of rings and fields. We
introduce it now in order to do something interesting, and to set the tone for the rest of the quarter
as we now start to get into more intricate types of arguments.
The answer is that it is NOT true that any group must contain maximal subgroups, but it
is true for finitely-generated ones at least, which you will show on the homework. This will give
an example of why finite-generation can be a useful concept to have, and it will also illustrate an
important result in modern mathematics: Zorn’s Lemma. Zorn’s Lemma is a result which is at
core of many arguments in mathematics where we can show that a given object exists even when
we are not able to construct it explicitly. Many of the objects we will consider in later quarters will
arise in this way, and, as we will see, knowing existence alone is often all we actually need, so that
the actual construction of said object is not necessary. This quarter Zorn’s Lemma will not play a
big role beyond our discussion here and the one homework problem, but we are covering it now in
order to, as said above, set the tone for the rest of the course and because it is an amazing result
which every serious student of mathematics should see at least one time in their development.

Zorn’s Lemma. Here, then, is the statement of Zorn’s Lemma:

Suppose P is a nonempty, partially-ordered set in which every chain has an upper

bound. Then P has a maximal element.

Of course, there are many undefined terms in the statement, and even once we define them the
underlying meaning of Zorn’s Lemma will take some getting used to, but the various examples
we give of its use throughout the year should convince you that it truly is a remarkable tool.
Ultimately, as said above, the point is that it gives a way to show that some object we care about
exists (constructed as a “maximal” object within the appropriate context), even if we have no hope
of constructing it explicitly.

Partial orders. A partial order on a set P is a relation ≤ satisfying:

• (reflexivity) a ≤ a for all a ∈ P ,

• (transitivity) if a ≤ b and b ≤ c, then a ≤ c, and
• (anti-symmetry) if a ≤ b and b ≤ a, then a = b.

Here, ≤ is purely a symbol we use to denote the given relation, but the point is that these properties
suggest ≤ behaves as it if was an actual “ordering” on elements of P : anything should be “less
than or equal to” itself, the “less than or equal to” relation should be transitive, and the only way
in which two things can be “less than or equal to” each other is if they are the actually the same.
We also use the strict notation a < b to mean that a ≤ b and a 6= b.
Two key examples are the usual “less than or equal to” relation on R, where x ≤ y literally
means that x is less than or equal to y, and the partial order on a collection of subsets of a set
given by ⊆, where we interpret A ⊆ B as saying that A is “less than or equal to” B. However,
these examples have one important difference: in the case of R, all elements are comparable to
one another in the sense that given any x, y ∈ R, it is true that x ≤ y or y ≤ x, but this is not
necessarily true when considering collections of subsets. A chain in P is a subset whose elements are
all comparable to one another in this way. (A partial order in which all elements are comparable is

34
called a total order, so a chain in P is then a totally-ordered subset of P .) The term “chain” comes
from the idea that you can order all elements from “smaller” to “larger”, which in the countable
case looks like:
... ≤ a ≤ b ≤ c ≤ ....
An upper bound of a subset S of P is an element u ∈ P such that s ≤ u for all s ∈ S, which
is the same way the term “upper bound” is used, say, in analysis. Finally, a maximal element
of P is one for which there is nothing strictly larger: a ∈ P is maximal if whenever a ≤ b for
some b ∈ P , we have a = b. The usual (total) ordering on all of R has no maximal elements, but
subsets of R might have maximal elements. If we take all subsets of a set S, then under ⊆ the only
maximal element is S itself, but a subcollection of only certain subsets might have zero, one, or
more maximal elements.

Use in algebra. Zorn’s Lemma thus says that as long we know that any totally-ordered subset
can be bounded above by something, then we can conclude that at least one maximal element
exists. In the type of situation we care about, Zorn’s Lemma will be applied in the following way.
Take P to be a collection of subsets of some set. Suppose further P has the property that for any
subcollection C ⊆ P of sets such that any two are comparable via ⊆— meaning S that given A and B
in C it is always true that either A ⊆ B or B ⊆ A—we have that the union C of all things in C
also belongs to P . Then we can conclude that there is a set S in P which is not strictly contained
within anySlarger element of P . Here, the partial ordering on P is given by ⊆, C describes a chain
in P with C being its upper bound in P , and the resulting S is a maximal element of P . Such
maximal elements, as we’ll see, often have important properties we care about.

Rω has a basis. Here is one application of Zorn’s Lemma in linear algebra. Denote by Rω the set
of infinite sequences of real numbers:

Rω := {(x1 , x2 , x3 , . . .) | each xi ∈ R}.

(This is analogous to elements of Rn , only now with infinitely-many components.) Consider Rω

equipped with vector addition and scalar multiplication defined as one would expect:

(x1 , x2 , . . .) + (y1 , y2 , . . .) = (x1 + y1 , x2 + y2 , . . .) and r(x1 , x2 , . . .) = (rx1 , rx2 , . . .).

We aim to show that Rω has a basis in the sense of linear algebra: a linearly independent subset
of Rω which spans all of Rω . Now, the trouble is that it is not actually possible to write down
an explicit basis (!), so our proof is non-constructive. This is in stark contrast to the case of Rn ,
where bases are easy to write down. Note that the obvious candidate of taking the vectors ei which
have a 1 in the i-th location and 0 everywhere else (which work in the Rn case) do not work in
Rω , since it is not true that anything in Rω can be written as a linear combination of finitely many
of these ei , which is a technical requirement in the definition of “span” in the setting of infinite
dimensions; the issue is that any linear combination of finitely many of the ei ’s must eventually
end in all zeroes! So, in fact, the ei vectors only span the subspace of Rω consisting of sequences
which are eventually zero, and so do not form a basis for entirety of Rω .
Let I denote S the collection of all linearly independent subsets of vectors in Rω . Take any chain
C ⊆ I. Then C is still a collection S ω
of linearly independent vectors in R , and so is an upper
bound for this chain S in I. To see that C is still linearly independent, take any finite number of
vectors v1 , . . . , vn ∈ C. (To say that a set of vectors is “linearly independent” technically means
that any finite number of vectors taken from that set are linearly independent.) Each vi comes
from some Ci ∈ C. The fact that C is a chain implies that there exists C0 ∈ C which contains

35
 S
each of C1 , . . . , Cn , so v1 , . . . , vn ∈ C0 ⊆ I must be linearly independent. Hence C is a linearly
independent collection of vectors as claimed.
By Zorn’s Lemma there thus exists a maximal linearly independent set B of vectors in Rω .
If these vectors did not span Rω , picking x ∈ Rω not in their span gives a linearly independent
collection B ∪ {x} which is strictly larger than B, contradicting maximality of B. Thus B must
span Rω , so that B is a basis of Rω as desired. In general, the same reasoning shows that any vector
space (if you do not what this is yet, you will next quarter), even an infinite-dimensional one, has
a basis. Note again here that we have no idea what this basis is, but it exists!

Lecture 11: Normal Subgroups

Warm-Up. Take P to be the set of all subgroups of a given group G, partially-ordered by set
inclusion. Let C be a chain in P . We claim that the union of all subgroups in C is itself a subgroup
of G. Now, the union of subgroups of group is not necessarily a subgroup in general (for instance,
the union of h2i and h3i in Z is not closed under addition), so that the fact that this is true here
depends heavily on the S chain condition. S
Clearly the union H∈C H is nonempty since each H is nonempty. If g ∈ H∈C H, then g ∈ K
for some K ∈ C, in which case g −1 ∈ K since K is a subgroup of G, and thus g −1 is in the union as
well. Finally, if g1 , g2 are in the union, then g1 ∈ H1 and g2 ∈ H2 for some subgroups H1 , H2 ∈ C.
Since C is a chain, we have either H1 ⊆ H2 or H1 ⊇ H2 , so that both g1 , g2 are contained in the
S say g1 , g2 ∈ H1 . Then g1 g2 ∈ H1 since H1 is a subgroup, so g1 g2 is in the union, and
same one;
hence H∈C H is a subgroup of G as claimed.
Now, Zorn’s Lemma is applicable here since this union serves as an upper bound for the chain
in P . But, the conclusion of Zorn’s Lemma in this case is not so satisfying: a maximal subgroup
of G exists, but of course G is in fact maximal in itself, so that we didn’t actually need Zorn’s
Lemma to notice this. If we only consider proper subgroups in the definition of “maximal” (which
is common), the problem is that P should only be taken to consist of proper maximal subgroups as
well, but then the union of the elements in the chain above might not itself be a proper subgroup,
so that it does not serve as an upper bound within P . This is why we need some type of restriction
on G if we want to conclude that any such group has a (proper) maximal subgroup, such as being
finitely-generated.

Choice implies Zorn. We finish our current discussion of Zorn’s Lemma by saying something
about where it comes from, and why we should believe it to be true. In class we only pointed out
that it is equivalent to the Axiom of Choice, so that if we accept this axiom then we must accept
Zorn’s Lemma as well, but here we will elaborate more on this for those who are interested. (The
Axiom of Choice is the claim that given any collection of nonempty sets, we can pick an element
from each one simultaneously. The point is that there is no restriction on the number of nonempty
sets we consider, it could even be an uncountable collection of nonempty sets.)
Let us give a very rough (emphasis on the very) sketch of the proof that the Axiom of Choice
implies Zorn’s Lemma. As stated before, Zorn’s Lemma is actually equivalent to the Axiom of
Choice, but the direction we look at here (choice implies Zorn) is the one which justifies the use
of Zorn’s Lemma as a valid tool in mathematics. We will not discuss the converse direction (Zorn
implies choice), but if you ever take a topology course we’ll just point out that one approach to
the converse direction uses what’s called Tychonoff ’s Theorem in topology. Our proof sketch is
quite rough since we will get to a point where we would need to know much more advanced set
theory—in particular properties of cardinal and ordinal numbers—to make it precise, but the basic
idea will come across.

36
 Suppose P is a nonempty, partially-ordered set in which every chain has an upper bound, and
aiming for a contradiction suppose P did not contain any maximal elements. Then for any a ∈ P ,
we can always find some b ∈ B such that a < b since a is not maximal. Using the Axiom of Choice
we can thus pick such an element f (a) for any a ∈ P all at once. Fix a0 ∈ P , so that a0 < f (a0 ).
But by this construction we also have f (a) < f (f (a)), and so on we get:

a0 < f (a0 ) < f (f (a0 )) < f (f (f (a0 ))) < · · · .

This list gives a chain in P , so by the assumption of Zorn’s Lemma this chain has an upper bound,
call it a1 :
a0 < f (a0 ) < f (f (a0 )) < f (f (f (a0 ))) < · · · ≤ a1 .
But now we can consider the chain

a1 < f (a1 ) < f (f (a1 )) < f (f (f (a1 ))) < · · · ,

which itself has an upper bound a2 :

a1 < f (a1 ) < f (f (a1 )) < f (f (f (a1 ))) < · · · ≤ a2 .

Continuing in this way over and over (and over and over!) again gives a bunch of elements of P :

a0 < f (a0 ) < · · · ≤ a1 < · · · ≤ a2 < · · · ≤ a3 < · · · ≤ a4 < · · · .

In fact, there would be so many elements of P listed here that this would imply (and this is the
part which requires some pretty deep stuff which we will in no way attempt to make precise here)
that the cardinality of P would be larger than that of any other set, and in particular P would
have cardinality (strictly) larger that of P itself (or also of its power set), which is nonsense. Thus
we conclude that P must have had a maximal element after all!

The big three. As stated above, the Axiom of Choice not only implies but is actually implied
by Zorn’s Lemma, so that they are equivalent. Just for the sake of interest, we give the statement
of one more equivalent form of either of these: the Well-Ordering Theorem. A well-ordering on
a set P is a total order in which every nonempty subset of P has a least (i.e. smallest) element.
For instance, the usual ordering on N is a well-ordering, whereas the usual ordering on R is not.
The Well-Ordering Theorem says that every set can in fact be well-ordered. In the case of R, the
point is that the usual order is not the one which works, but that there is some way to “order” the
elements of R so that every nonempty subset does have a least element.
This is pretty surprising indeed, and the well-ordering on R which works would actually have
no relation to the usual ordering. An explicit such well-ordering on R is not possible to write
down, but nonetheless we know it must exist (if we accept the Axiom of Choice) since the Axiom
of Choice, the Well-Ordering Theorem, and Zorn’s Lemma are all equivalent to one another. These
types of surprising results are the main reason why the Axiom of Choice—as obvious as it may
seem—is viewed as quite controversial by many mathematicians: it has some seemingly paradoxical
consequences which often say that a certain objects exists without giving any sense as to how to
actually construct said object. There’s an old joke that says: the Axiom of Choice is clearly true,
the Well-Ordering Theorem is clearly false, and who knows about Zorn’s Lemma? The joke, of
course, is that the first of these seems obvious, the second seems like it could not possibly be true
(since we cannot even imagine what a well-ordering of R would actually look like), and the third
(Zorn) is such a complicated looking statement that no one really has any idea what it even means,
and yet all three are actually saying the same thing in the end.

37
Towards quotient groups. Now, after this brief diversion, we return to the actual content of this
course. We will begin to work towards the definition of a quotient group, and will start now with
the overarching point of the constructions we will see. I feel that all too often books, such as ours,
jump straight into the technical definitions required in order to construct quotient groups, but do
not say much about the point of it all is. Moreover, the technical definitions tend to obscure the
underlying idea and intent, but we will put this front and center.
Take a group G. The basic idea behind quotient groups is that we wish to “introduce” into G
some new relations, and hope to still get a group in the end. For instance, consider the construction
we gave earlier of the dihedral group D8 from the free groups hr, si on two generators: we take this
free group—consisting of all “words” in r, s and their inverses—and then impose the relations

r4 = 1, s2 = 1, and rs = sr−1 .

In this case, in the end, we get a group of order 8 as a result. This is the construction we wish to
mimic. So, let S be a subset of G. We use the elements of S to describe the new relations we want
to impose, by declaring that each element of S should “become” the identity; in other words, we
treat each element of S as being “equal” to the identity element. For instance, in the case of D8 ,
S would be the subset
S = {r4 , s2 , srsr},
whose elements are precisely the things which become equal to 1 in D8 . (Note that srsr = 1 is
equivalent to rs = sr−1 , and indeed it is always true in any group that a relation given by having
two expressions be equal to one another can always, after multiplying by inverses, be written as
one in which a single larger expression equals the identity.)
Now, setting the elements of S to be “equivalent” to the identity gives rise to new relations as
well, namely those which follow from the given ones; for instance, the three relations on D8 above
also give things like
r4 srsr = 1, (r4 s2 r4 )−1 = 1, and so on.
These additional relations come from the subgroup hSi generated by S, so we might as well assume
that S was already a subgroup H to begin with, and we will do so.

Equality in quotients. We use the notation G/H (pronounced “G mod H”), to denote the object
which results from declaring all elements of H to be equal to the identity. When this actually turns
out to be a valid group (there are some subtleties, as we’ll see), we call it a quotient group, in this
case the “quotient of G by the subgroup H”. The term “quotient” comes from a similarity which
exists with ordinary quotients of integers, where in that case we “divide” an integer into pieces,
just as (as we’ll see) quotient groups do as well in an appropriate sense.
Now, what does it mean to say that g = k in G/H? The point is that g and k might not
literally be the same element in G, but only become the “same” after introducing the relations
coming from H. Indeed, g = k is equivalent to g −1 k = e, but in the context of G/H the symbol
“e” really denotes any element of H (since we have declared all elements of H to be equal to the
identity), so that “g −1 k = e” really means g −1 k ∈ H. Thus, we have:

g = k in G/H ⇐⇒ g −1 k ∈ H.

Denoting this element of H by h = g −1 k, we have that gh = k in G. The point is that, if indeed

gh = k in G, then after “setting” h to be equal to the identity, the equation gh = k becomes
ge = k, or simply g = k. The set gH = {gh | h ∈ H} is called a coset of H in G (we’ll discuss

38
cosets in more detail next time), and consists of all the elements of G which become equal to g in
the quotient G/H:

g = k in G/H ⇐⇒ g −1 k ∈ H ⇐⇒ k = gh for some h ∈ H ⇐⇒ k ∈ gH.

Thus, two elements become equal upon quotienting out by H if they “differed” by an element of
H in G, in the sense above.

Example. Let us consider the following example, which finally clarifies the notation we’ve been
using Z/nZ for the additive group of integers mod n. For any n ∈ N, the set of integer multiples
of n—commonly denoted nZ—is a subgroup of Z. In the resulting quotient group Z/nZ (where we
“set” the elements of nZ to be equal to 0), we have that:

a = b in Z/nZ ⇐⇒ b − a ∈ nZ ⇐⇒ a ≡ b mod n.

(Note that b − a is “a−1 b“ written in additive notation.) Thus, equality in the quotient Z/nZ
corresponds precisely to equality mod n, as we’ve been using all along. Again, the point is that if
a = b + kn, then a and b become the same once we declare that nk ∈ nZ should be zero.

Well-definedness. Now, we want to endow G/H with a group structure. But it is clear how we
should proceed: we began with a group G at the start and imposed some new relations, so we can
still (try to) use the the same group operation we had on G on G/H as well. That is, we attempt
to define gk in G/H as simply gk ∈ G, where the only difference is that now gk might equal some
things in G/H that it did not equal in G, due to the new relations we’ve introduced. With this
definition, we automatically get associativity in G/H since associativity already held in G; there is
still an identity element, only now it is in fact represented by any element of H (which are all equal
in the quotient, so that the quotient does have only a single identity element); and the inverse of g
is still g −1 , again only that possibly g −1 is equal to some new things as well.
But, this attempt to define a group structure on G/H runs into a subtle problem. Suppose
that g = k in G/H. Then, if there is any justice in the world, it should be true that for all a ∈ G
we have ga = ka. After all, if g and k are supposed to be the “same”, then multiplying each by
literally the same a should indeed result in equal quantities:

g = k in G/H should imply ga = ka in G/H for all a ∈ G.

If we write out what these equalities mean in G/H, this becomes the statement that

g −1 k ∈ H should imply (ga)−1 (ka) ∈ H for all a ∈ G.

The product on the right is a−1 g −1 ka, and so the condition we need to hold is that:

g −1 k ∈ H =⇒ a−1 (g −1 k)a ∈ H for all a ∈ G,

or said another way with h = g −1 k ∈ H, we need that

h ∈ H =⇒ a−1 ha ∈ H for all a ∈ G.

Thus, in order for the proposed group operation on G/H to be well-defined and give a valid function
G/H × G/H → G/H (which is what multiplication should be), it should be true that conjugating
an element of H by an element of G should always result in an element of H. (Technically above we
are conjugating by a−1 , but of course if we replace a ∈ G by a−1 ∈ G—note the condition should

39
hold for all elements of G—we get aha−1 ∈ H as well.) This condition, because of its relation to
quotient groups, is so important that we give a special name: normality.

Normal subgroups. A subgroup H of G is normal in G if ghg −1 ∈ H for all g ∈ G. Using

notation we introduced previously in relation to normalizers, this says that H is normal in G if
gHg −1 ⊆ H for all g ∈ G. We use H E G to denote that H is a normal subgroup of G. The point
is that it is only for normal subgroups that the “quotienting” process described above works and
produces a well-defined group structure on G/H.
Now, if this is true, then in particular we also have g −1 Hg ⊆ H, so that upon multiplying by g
on the let and g −1 on the right we get H ⊆ gHg −1 as well. Thus, the condition of normality can
actually be written as an equality: gHg −1 = H for all g ∈ G. (But, nevertheless, to actually check
normality it is quicker to use gHg −1 ⊆ H since this only requires one containment.) This equality
says precisely that g ∈ NG (H) (the normalizer of H in G), so H E G if NG (H) = G; indeed, this
where the name “normalizer” comes from. Looking back to prior normalizer examples, you will
see that we have already come across examples of normal subgroups, such as hri E D2n . Note also
that every subgroup of an abelian group, such as nZ in Z, is normal.

Example. We finish with an example which explicitly shows what goes wrong when we do not
have normality. Take hAi = {0, A} ≤ D8 , where we use the rotation/reflection notation with A as
the anti-diagonal reflection. This subgroup is not normal in D8 , since for instance

90 · A · 90−1 = D ∈
/ hAi.

Thus, it should be true that the multiplication on the quotient D8 /hAi resulting from the one on
D8 is NOT well-defined, so that D8 /hAi is not actually a group under this operation.
Indeed, we have 180 = D in D8 /hAi since (180)−1 D = 180 · D = A ∈ hAi. But, now consider
180 · 90 = 270 and D · 90 = V . Since (270)−1 V = 90 · V = D ∈
/ hAi, we have that

180 · 90 = 270 6= D · 90 = V in D8 /hAi.

And so, even though 180 and D are the same in the quotient, 180 · 90 and D · 90 are not the
same in the quotient, so this attempt at defining multiplication in the quotient does not produce a
well-defined operation, so we do not get a valid group structure.

Lecture 12: Cosets and Quotients

Warm-Up. We show that h(123)i is not normal in S4 and find an example of elements σ, τ, γ ∈ S4
such that σ = τ but σγ 6= τ γ in the quotient S4 /h(123)i. (Thus we do not get a well-defined group
structure on this quotient, which is expected by the lack of normality.) First, we have:

(34)(123)(34)−1 = (124),

which is not in h(123)i so this subgroup is not normal in S4 . Now, we have (12) = (12)(123) = (23)
in S4 /h(123)i, but
(12)(34) = (12)(34) and (23)(34) = (234),
which are not equal in the quotient since (12)(34)(234)−1 = (124) ∈
/ h(123)i.

Cosets. Recall from last time that g = k in G/H if and only if there exists h ∈ H such that
gh = k. The set gH = {gh | h ∈ H} is called a (left) coset of H in G and thus (as we said before)

40
consists of all elements to which g becomes equal in G/H. We will then formally define G/H to
be precisely the set of cosets of H in G:

G/H := {gH | g ∈ H}.

The point is that G/H should, as a set, consist of distinct elements, and using this coset language
gives us a formal way of “identifying” the elements of G which become the “same” after imposing
the relations which come from H; in this setting, g and k are the “same” in G/H precisely when
the cosets they determine are literally the same:

g = k in G/H ⇐⇒ gH = kH.

Indeed, note that if gh = k, so that k ∈ gH, then we have equivalently g = kh−1 ∈ kH, which
does in fact imply that gH = kH since then gH = (kh−1 )H = k(h−1 H) = kH. (We use here the
fact that any h0 ∈ H can be written as h0 = h−1 (hh0 ) ∈ h−1 H.) It follows that two cosets are
thus either equal as sets or completely disjoint, which is what makes “grouping” elements of G into
cosets in this way work: each element only gets “grouped” into a single coset. (Next time we will
provide another perspective on this idea in terms of an equivalence relation.)
The group operation we have been informally using on G/H up until now can now formally be
defined in terms of coset multiplication:

(g1 H) · (g2 H) := (g1 g2 )H.

The point of course is that this group operation should simply mimic the group operation on G,
only that now we allow elements of G to be “equal” when they give equal cosets. The work we did
last time then shows that this operation indeed gives a well-defined group structure if and only if
H is a normal subgroup of G: we can pick an element k ∈ g2 H different from g2 to represent the
same coset g2 H = kH, and normality is what we need to say that using k instead of g2 to perform
the proposed group operation still gives the same answer:

(g1 H)(g2 H) = (g1 g2 )H is the same as (g1 H)(kH) = (g1 k)H

when H is normal. Note also that the notion of normality can be expressed in terms of cosets:
gHg −1 = H is equivalent to gH = Hg where Hg denotes the right coset corresponding to g, so
H is normal in G when the left coset corresponding to any g ∈ G is the same as the right coset
corresponding to g.

Quotient groups. Thus, for a normal subgroup H E G, we define formally define the quotient
group G/H to be the set of cosets of H in G under the group operation defined above. The identity
element is the coset eH = H (which equals hH for any h ∈ H) and the inverse of gH is g −1 H.
Now, we will abuse notation and use G/H to still denote the set of cosets of H in G even when H
is not normal in G, but in this case we will not refer to G/H as being a group and will not consider
multiplication of cosets, so that G/H will simply be a set without any extra structure.
The coset approach to defining quotient groups is the one most introductory books take, and
serves to give a formal construction. Our book approaches quotient groups via kernels and “fibers”,
which we will clarify in a bit. We will, for the most part, actually avoid coset notation altogether
when describing elements of G/H unless it is necessary to avoid confusion in various proofs. Instead,
we will think of G/H as we first described it before introducing cosets: as G itself with new relations
imposed. Thus, we will use g to denote both elements of G and elements of G/H, keeping in mind
that G/H we might have g = k without g = k being literally true in G.

41
 We feel that this better matches the way in which quotient groups are used in practice, and em-
phasizes the intent behind quotient groups (introduce new relations) better than the coset approach.
The downside of our approach and notation is that we will need to be more careful about noting
where various computations take place and where various equalities hold, but the trade-off will be
worth it. (This will be even more true in subsequent quarters when we consider other types of quo-
tient constructions.) Apart from using coset notation, if we truly do need to distinguish between
an element of G and the element it gives in a quotient, it is common to use g to denote the quotient
element corresponding to g ∈ G, which we think of as “g reduced mod H”. Indeed, the book has
been using this notation from the start when referring to elements of Z/nZ = {0, 1, . . . , n − 1}.
One more remark: if we are to think of quotient groups as introducing new relations into G, we
might think that the requirement of having subgroups be normal restricts the types of relations we
can consider. After all, using the motivation from last time, if S ⊆ G contains the new relations
we wish to introduce, the subgroup hSi they generate is not necessarily normal, so that G/hSi is
not defined as a group. However, this is not a problem: given hSi ≤ G, we can instead consider its
normal closure N in G, which is the subgroup of G generated by all the conjugates of hSi. This N
is then always normal in G, and is in a sense the “smallest” normal subgroup of G containing all
the relations in S; in turn, G/N is the “least trivial” group obtained G by introducing the relations
in S. We will not discuss normal closures further in this course, but they give a precise why of
being able to introduce whatever relations we want into G.

Example. We determine (somewhat informally for now) the better known group to which the
quotient group
(Z × Z)/h(1, 2)i
is isomorphic. (Note Z × Z is abelian, so every subgroup is normal.) In this quotient we have
(1, 2) = (0, 0), so the idea is to use this relation to simply the description of elements in the
quotient. From this equality (write it as (1, 0) + (0, 2) = (0, 0)) we get:

(1, 0) = (0, −2) in (Z × Z)/h(1, 2)i

Thus the point is that this cuts down the number of generators in the quotient, so that instead of
(1, 0) and (0, 1) being treated as “independent” generators, we can generate the entire quotient by
(0, 1) alone. Concretely, any element in the quotient is given by a pair whose first coordinate is 0:

(a, b) = (0, b − 2a) in (Z × Z)/h(1, 2)i.

No two such elements (0, m) and (0, n) are equal to each other, since (0, m) − (0, n) = (0, m − n)
is in h(1, 2)i only when m = n. Thus, we can see that the quotient is a cyclic group generated by
one element (0, 1) of infinite order, meaning that it is isomorphic to Z:

(Z × Z)/h(1, 2)i ∼
= Z.

Informally, a possible isomorphism between these two comes precisely from the realization that
(a, b) = (0, b − 2a): the entire data of an element (a, b) of the quotient is captured by a single
integer b − 2a, which as b and a vary can take on any integer value. That the map

(Z × Z)/h(1, 2)i → Z defined by (a, b) 7→ b − 2a

is indeed an isomorphism is something we will postpone proving, in order to derive it from the
general fact known as the First Isomorphism Theorem. For now, we note that this map is indeed

42
well-defined, meaning that we can choose a different element of Z×Z to represent the same thing as
(a, b) in the quotient without changing the value which the function above assigns to this element:
if (a, b) = (a0 , b0 ) in (Z × Z)/h(1, 2)i, then (a − a0 , b − b0 ) = (k, 2k) ∈ h(1, 2)i, so

a0 = a − k, b0 = b − 2k and thus b0 − 2a0 = (b − 2k) − 2(a − k) = b − 2a.

Again, this observation will become more general in the context of the First Isomorphism Theorem.

Kernels and quotients. Given a homomorphism φ : G → H, its kernel ker φ := {g ∈ G | φ(g) =

eH } is always normal in G. Indeed, for any g ∈ G and x ∈ ker φ, we have:

φ(gxg −1 )φ(g)φ(x)φ(g −1 ) = φ(g)eH φ(g)−1 = eH ,

so gxg −1 ∈ ker φ. Thus, the quotient group G/ ker φ always exists (as a group), and the First Iso-
morphism Theorem will tell us what group this quotient actually is. (Spoiler alert: it is isomorphic
to the image φ(G) of φ.) Moreover, in fact any normal subgroup H E G arises in this way: if H is
normal, G/H is a group, and the map

G → G/H defined by g 7→ g

(which really means g 7→ g = gH, but again we prefer to think of g in the quotient as literally
the same element as g in G, only subject to more relations) is a homomorphism whose kernel is
precisely H: h ∈ G gives the identity h = e in G/H if and only if h ∈ H. Thinking of normal
subgroups as kernels is how our book first introduces normality. The homomorphism G → G/H
sending g to itself (viewed as an element in G/H, or viewed as sending g to g = gH) is called the
canonical or natural projection of G onto G/H.
But now we ask: what does it mean to say that g1 = g2 in G/ ker φ, or equivalently what do
the cosets of ker φ in G actually measure? The equality g1 = g2 holds in this quotient if and only
if g1−1 g2 ∈ ker φ, which means

φ(g1−1 g2 ) = eH , or equivalently φ(g1 )−1 φ(g2 ) = eH , or equivalently φ(g1 ) = φ(g2 ).

Thus, we have that g1 = g2 in G/ ker φ precisely when g1 , g2 gives the same image φ(g1 ) = φ(g2 )
in H. In other words, g1 , g2 here should belong to the preimage φ−1 (h) of the same element
h = φ(g1 ) = φ(g2 ) in H, which is the set of all things in G which map to h:

φ−1 (h) := {g ∈ G | φ(g) = h}.

A preimage of a single element is also called a fiber of φ, and the conclusion is that the cosets of
ker φ are precisely the fibers of φ:

for g ∈ G, g(ker φ) gives all elements in G which map to the same thing in H as g.

The name “fiber” comes from the following picture. Visualize G as a plane, H as a line, and φ
as a “projection” map:

43
The fibers/preimages are then the vertical lines lying above points on the line. The entire fiber gets
“collapsed” into a single point upon taking the quotient, and the group operation on the quotient
can be viewed as “multiplying” these fibers: take a point in one fiber, a point in another fiber,
multiply them together in G, with the result being the fiber where the product lies. Normality
guarantees that it does not matter which points from each fiber we actually use here, we always
get the same fiber as a result.

Lecture 13: Lagrange’s Theorem

Warm-Up 1. We (informally) identity the quotient group (Z×Z)/h(5, 5)i as a simpler group. Note
first that (5, 5) = (0, 0) in the quotient, so that (5, 0) = (0, −5). This implies that any general (a, b)
will be equivalent to one which has first coordinate among 0, 1, 2, 3, 4, since we can subtract away
enough 5’s from a and “move” them over to the second coordinate, leaving the first coordinate as
one of 0, 1, 2, 3, 4. More precisely, if a = 5q + r where 0 ≤ r < 5 according to the division algorithm,
then
(a, b) = (5q + r, b) = (r, b − 5q).
(Note we can also write r here as a mod 5.) Alternatively, we have

(a, b) = (a − 5k, b − 5k) for any k ∈ Z,

and again for the appropriate k = q the first coordinate a − 5k will be in the range 0, 1, 2, 3, 4.
This all suggests that the first coordinate in an element of the quotient can actually be taken to
be in Z/5Z, and moreover the way in which addition works in the quotient—where adding multiples
of 5 to both coordinates does nothing—suggests the addition in this first coordinates should indeed
be taking place mod 5. Now, even though the same thing is happening in the second coordinate,
what happens there is dependent on the behavior in the first coordinate since it is the same 5k
which is subtracted from both above. So, we will not get that the second coordinate can also be
taken to be in Z/5Z, since this restriction would in return change the value of the first coordinate
again. (Getting Z/5Z × Z/5Z as the group to which the quotient is isomorphic would require that
we independently be able to reduce both coordinates mod 5.) Once we modify the first coordinate
to be in Z/5Z, the second coordinate is fixed as in

(a, b) = (r, b − 5q)

44
above. Since b is arbitrary, the value of b − 5q can be taken to be any integer, since any integer can
be written as 5 less than some integer. Thus, we expect that the second coordinate comes from the
entirety of Z, so this quotient should be isomorphic to Z/5Z × Z:

(Z × Z)/h(5, 5)i ∼
= Z/5Z × Z.

We could of course make the second coordinate take values in Z/5Z and allow the first coordinate
to be whatever instead, in which case we get Z × Z/5Z as the answer. We will prove formally that
this answer is correct later after we discuss the First Isomorphism Theorem.

Warm-Up 2. We show that if G/Z(G) is cyclic, then G is abelian, which is our first example of
using information about a quotient to derive information about a group. Recall that Z(G) denotes
the center of G—the subgroup of elements of G which commute with all elements of G—which is
normal in G since every element of G in fact centralizes (and hence normalizes) Z(G).
Let x ∈ G/Z(G) be a generator (recall that we are denoting elements of G/Z(G) simply by
elements of G and let g, h ∈ G. Then in the quotient we have g = xk and h = x` for some k, ` ∈ N,
which means that
g = xk a and h = x` b for some a, b ∈ Z(G).
Thus we compute:
gh = xk ax` b = xk x` ab = x` xk ba = x` bxk a = hg,
where we use the fact that a, b ∈ Z(G) commute with all other elements. This shows that G is
abelian as claimed. Note it was crucial here that G/Z(G) be cyclic and not simply abelian. If
G/Z(G) were only abelian, we could say that gh = hg in G/Z(G), but this gives gh = hga for
some a ∈ Z(G), which does not say that G is necessarily abelian. The point here is that Z(G)
and G/Z(G) being abelian is not enough to guarantee that G is abelian, so there is a limit to the
amount of information about G we can derive from knowledge of subgroups and quotients alone,
in general.

Cosets and actions. Let us note that cosets arise as the equivalence classes of a certain equivalence
relation. Indeed, given H ≤ G the relation

g ∼ k defined by g −1 k ∈ H

on G is an equivalence relation. (This is actually equivalent to H being a subgroup of G and not just
a subset.) Two elements are in then the same equivalence class precisely under the same condition
g −1 k ∈ H which says that the cosets gH and kH are the same, so the equivalence classes for this
equivalence relation are indeed the (left) cosets of H in G. As a consequence, we get immediately
that two cosets are either disjoint or identical since this is true of equivalence classes in general.
We now make two observations regarding cosets and actions. First, consider the right action
G x H of H on G given by right multiplication:

g · h := gh.

(A right action is similar to a left action, only that the property g · (k · x) = (gk) · x holds in the
“reverse” direction: g · (k · x) = (kg) · x. It is for this reason that it is more common to write the
action as occurring on the right, so that (x · k) · g = x · (kg) holds instead. Right actions will not
play a major role for us, essentially because there is a standard way of turning a right action into

45
a left action anyway. We’ll omit the details here.) For a fixed g ∈ G, the orbit of this action is
precisely the left coset of H corresponding to g:

gH = {g · h | h ∈ H} = {gh | h ∈ H}.

(Using the left action of H on G given by left multiplication would produce the right cosets of H
in G as the orbits.) Again, the fact that cosets arise as orbits implies immediately that they any
two are either disjoint or identical.
The second observation is that there is a natural action G y G/H of G on G/H—viewed here
as simply the set of cosets without reference to any potential group structure (H is not necessarily
normal in G)—again by left multiplication:

g · (kH) := (gk)H.

Equivalently this action can be viewed as a homomorphism G → SG/H , where SG/H as usual
denotes the group of permutations (i.e. self-bijections) of the set of cosets. The existence of such
a homomorphism for any subgroup H will be useful later in studying properties of G itself.

Lagrange’s Theorem. And now we come to the first truly nontrivial result in finite group theory:
Lagrange’s Theorem. By “nontrivial” we mean that this is the first result which really requires
some new insight or construction (in this case the language of cosets), and not the “basic” language
of groups alone. Even the facts we proved about cyclic groups previously—although certainly not
necessarily easy—only really used basic properties of powers and orders, and not anything too deep.
Of course, the use of the phrase “nontrivial” is subjective, and is meant to convey here the need to
consider objects which seem to come out of nowhere; after all, how likely is it that someone would
think of the idea of a “coset” in trying to prove what follows? It is this realization of the need to
use cosets which we mean by calling this a “nontrivial” result.
Here is the statement, which we might have guessed from all the examples of subgroups we have
seen so far: if G is a finite group and H a subgroup, then |H| divides |G|. Thus, we have nontrivial
constraint on what order of a subgroup could actually be. The basic idea of the proof is that given
g ∈ G, there |H| many elements in the coset gH, so that the number of elements of G gets reduced
by a factor of |H| when forming the set of cosets G/H. To be precise, write G as a disjoint union
of its cosets: G
G= gH.
g∈G/H

More simply, if we denote the distinct cosets of H in G by g1 H, . . . , gk H, then

G = g1 H t . . . t gk H.

Since this is a disjoint union, we have:

|G| = |g1 H| + · · · + |gk H|.

But each coset gi H has the same size as H since the map gi H → H defined by gi h 7→ h is a
bijection. (Note here it is crucial that gh and gh0 are distinct if h and h0 are distinct.) Thus we get

|G| = |H| + · · · + |H| = k|H|,

| {z }
k times

so that |H| divides |G| as claimed.

46
 This also shows that the number of distinct cosets—which is k in the notation above—also
divides |G|. This number of cosets of H in G is called the index of H in G, and is denoted by
[G : H]. (The book uses |G : H| to denote the index, but [G : H] is more common.) The proof of
Lagrange’s Theorem thus shows that

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

when G is finite. (This also holds in the infinite case, in the sense that if the left side is infinite, at
least one of the terms on the right is as well, and vice-versa.) Note that when H is normal in G,
the index is precisely the order of the quotient group, and we get:

|G|
|G/H| = [G : H] =
|H|

in the finite case. (This matches our intuition above, in that the number of elements of G gets
reduced by a factor of |H| when forming cosets.)

Consequences. And now we give a few basic consequences of Lagrange’s Theorem. First is the
fact that the order of any element g ∈ G (G finite) has to divide |G|, which just comes from the
fact that |g| = |hgi| and |hgi| divides |G| by Lagrange’s Theorem. As a second consequence, this
then implies that for any g ∈ G, g |G| = e since |G| is a multiple of |g|.
Third is the fact that any group of prime order must be cyclic! Indeed, if |G| = p is prime, the
order of any nonidentity x ∈ G must divide |G| = p, and hence must be p itself since p is prime.
This means hxi = G, so G is cyclic as claimed. Thus we now have our first true group classification
result: the only group (up to isomorphism) of prime order p is Z/pZ.

Lecture 14: First Isomorphism Theorem

Warm-Up 1. Suppose |G| = pq where p, q are distinct primes. We claim that either G is abelian
or G has trivial center Z(G). Indeed, by Lagrange’s Theorem, |Z(G)| divides |G| = pq, and hence
must be either 1, p, q, or pq. If |Z(G)| = pq, then Z(G) = G and G is abelian. If |Z(G)| = p or q,
then G/Z(G) has prime order q or p respectively, which means that G/Z(G) is cyclic, as we proved
at the end of last time. But this then implies that G is abelian, as we proved in a Warm-Up last
time. (Note that in this case Z(G) = G after all since G is abelian, so that Z(G) would have order
pq and not p or q, meaning that the case |Z(G)| = p or q case could not actually happen. But, we
were not able to see this until after this case was completed.) The remaining case |Z(G)| = 1 is
that Z(G) is trivial, so we are done.
As an application, this gives a quick proof that D2p has trivial center when p is prime, since
then 2 and p are distinct and D2p is certainly not abelian. Of course, D2n has trivial center for any
odd n, not just those that are prime, but the proof of this general fact (if you recall from the first
homework) is a lot more computationally involved than this simpler proof in the n prime case.

Warm-Up 2. Suppose K ≤ H ≤ G with G finite. We claim that the following equality among
indices holds:
[G : K] = [G : H][H : K],
so that the product of the number of cosets of K in H with the number of cosets of H in G is the
number of cosets of K in G. This type of equality will be useful in deriving information about the
various groups occurring in such a chain K ≤ H ≤ G.

47
 Let us denote the distinct elements of G/H and H/K by

G/H = {g1 , . . . , gm } and H/K = {h1 , . . . , hn }.

To be clear, we are saying here that not only are the gi distinct in G and the hj distinct in H, but
that they are also distinct in the respective quotients above. We claim that the mn many elements
gi hj give all the elements of and are distinct in G/K. If so, then [G : K] = mn, [G : H] = m, and
[H : K] = n in this notation, which gives the required equality. First, for any g ∈ G, there is some
gi ∈ G/H to which is equivalent, meaning that

g = gi h for some h ∈ H.

But then this h is equivalent to some hj ∈ H/K, so that

h = hj k for some k ∈ K.

Thus g = gi h = gi hj k, so that g = gi hj in G/K, and hence the gi hj do give all elements of G/K.
Now, to show that the gi hj give distinct elements of G/K, suppose some gi hj = gp hq in G/K.
Then
gi hj = gp hq k for some k ∈ K.
Manipulating this gives gi = gp (hq khj ), and since hq khj ∈ H (since K ≤ H and H is closed under
multiplication), this means that gi = gp in G/H. But g1 , . . . , gm ∈ G were assumed to give distinct
elements of G/H, so it must be the case that gi = gp in G as well. Hence the equality above
becomes
gi hj = gi hq k, which implies hj = hq k.
This now means that hj = hq in H/K, so that in fact hj = hq in H since h1 , . . . , hn ∈ H were
assumed to give distinct elements of H/K. Thus gi hj = gp hq in G/K implies that gi = gp in G and
hj = hq in H, so that the elements gi hj , 1 ≤ i ≤ m, 1 ≤ j ≤ n are all distinct in G/K as claimed.

Products of subgroups. As one more application of cosets, let us describe the size of the product
HK of two subgroups H, K ≤ G. This product is defined to be the set of all products hk with
h ∈ H and k ∈ K:
HK := {hk | h ∈ h, k ∈ K}.
Note (!!!) that this is not necessarily a subgroup of G, since it need not be closed under multipli-
cation: (h1 k1 )(h2 k2 ) cannot always be rewritten as (something in H)(something in K). Indeed, it
is true that HK is a subgroup if and only if HK = KH, since this is what is needed to “swap” the
terms in the k1 h2 in the middle of the product (h1 k1 )(h2 k2 ). (To be clear, by “swap” we mean to
rewrite it as some hk instead, but it is not necessarily true that k1 h2 will equal h2 k1 , which is why
we have equality HK = KH at the level of sets, not elements.) We will omit the details of this
claim here, but will point out that there are various conditions—perhaps most importantly when
at least one of H, K is normal in G—under which HK = KH will hold; we’ll come back to this
when discussing the Second Isomorphism Theorem.
But regardless of whether HK is actually a subgroup of G, we can still speak about its size
(assuming G is finite), and the claim is that
|H||K|
|HK| = .
|H ∩ K|
There idea is that there |H|-many choices for the first term in hk, and |K|-many for the second
term, giving at most |H||K|-many such expressions. But, some of these might create duplicate

48
products hk, and accounting for this reduces the size precisely by a factor of |H ∩ K|. Here are the
details. First, note that we can write HK as a union of those cosets of K which are determined by
the elements of H: [
HK = hK.
h∈H

To keep track of only distinct cosets—and hence get a disjoint union—we note that for h, h0 ∈ H:

h = h0 in G/K ⇐⇒ h−1 h0 ∈ K ⇐⇒ h−1 h0 ∈ H ∩ K ⇐⇒ h = h0 in H/(H ∩ K),

where the second ⇐⇒ comes from the fact that h, h0 were assumed to be in H to begin with.
Thus, we can specify the distinct cosets hK using the elements of H/(H ∩ K), so:
G
HK = hK.
h∈H/(H∩K)

Each hK has the same size as K, so taking sizes above gives

X
|HK| = |K| = [H : H ∩ K]|K|
h∈H/(H∩K)

since the size of H/(H ∩ K) is [H : H ∩ K]. But this also equals |H|/|H ∩ K| by the proof of
Lagrange’s Theorem, and substituting this into the equality above gives our claim.

Analog in linear algebra. We mentioned a while back—when discussing the idea of subgroups
generated by subsets—one analogy between group theory and linear algebra (generated subgroups
vs spans), and here is another. We will phrase this in the context of vector spaces in general, but
if you are unfamiliar with this concept you can simply think about Rn and its subspaces. (We we
will discuss vector spaces in this course next quarter.) If U, W are subspaces of a vector space V ,
then the following dimension equality holds:

dim(U + W ) = dim U + dim W − dim(U ∩ W ),

which mimics precisely “additive” version of the product subgroup equality derived above! Indeed,
the product equality in additive notation is

|H + K| = |H| + |K| − |H ∩ K|.

We will see a few more such analogies between group theory and linear algebra as we go, and will
see the proper context which underlies them next quarter. (This is also analogous to the basic
set-theoretic “inclusion/exclusion” equality: |A ∪ B| = |A| + |B| − |A ∩ B| for finite sets A, B.)

Isomorphism Theorems. We now move to a collection of results, called the isomorphism the-
orems, which help to more simply identity various quotient groups. Of these, the first of them—
appropriately called the First Isomorphism Theorem—is the most important and truly captures
much of the intent behind introducing quotients.
Here is the statement:

Suppose φ : G → H is a homomorphism of groups. Then φ induces an isomorphism

G/ ker φ ∼
= φ(G), where φ(G) denotes the image of φ.

49
This, as we will see, gives a quick way of identifying quotients of the form G/ ker φ, which really
covers all quotients G/N as long as we can find φ whose kernel is a given N . The proof goes as
follows. First, recall that cosets of ker φ in G correspond to the unique elements of the image φ(G):

g = g 0 in G/ ker φ if and only if φ(g) = φ(g 0 ).

In other words, ker φ measures the failure of φ to be be injective, so that by “collapsing” all of ker φ
down to the identity in G/ ker φ, we get an injective homomorphism:

φ : G/ ker φ → H.

We are abusing notation by continuing to refer to this map as φ, but the point is that it literally
does the same thing as the original φ: g 7→ φ(g). This is still well-defined on the domain G/ ker φ
since g = g 0 in this quotient means precisely that φ(g) = φ(g 0 ), so that the element of H we get
does not depend on the representative we pick for a given coset g(ker φ) in G/ ker φ.
This new map is injective precisely because elements g, g 0 for which φ(g) = φ(g 0 ) become the
same in G/ ker φ. Finally, by cutting down the codomain H to consist of only the image φ(G) ≤ H,
we guarantee that φ becomes surjective as well:
∼
=
φ : G/ ker φ → φ(G),

so that this map induced by φ is indeed an isomorphism. Thus, as long as we can identify a given
normal subgroup N E G as being the kernel of some well-chosen φ, we can identity the quotient
G/N = G/ ker φ by looking at the image of φ.

Another analog in linear algebra. Here is yet another analog between group theory and linear
algebra. Given a linear transformation T : V → W between finite-dimensional vector spaces V, W
(again, just think about a matrix transformation A : Rn → Rm if you are not familiar with the
language of vector spaces), the rank-nullity theorem asserts that the following holds:

dim V = dim im T + dim ker T.

(The dimension of im T is called the rank of T , and dim ker T is the nullity of T , hence the name
“rank-nullity” theorem. In the case where T = A : Rn → Rm is a matrix transformation, this
equality says that n = (# of pivots in rref A) + (# of free variables in rref A), where rref A denotes
the reduced row-echelon form of A.) In additive notation, taking orders in the First Isomorphism
Theorem above gives

|G| − | ker φ| = |φ(G)|, or equivalently |G| = |φ(G)| + | ker φ|,

and the analogy is clear. If you know about quotient vector spaces already (which we will study
next quarter from the more general point of view of quotient modules), you might recall that the
rank-nullity theorem is just a reflection of the underlying isomorphism between V / ker T (as a vector
space) and im T , which looks even more like the First Isomorphism Theorem here.

Example. Previously we argued informally that (Z × Z)/h(1, 2)i ∼

= Z, and now we give a concrete
proof. The goal is to find a surjective homomorphism

φ:Z×Z→Z

50
whose kernel is h(1, 2)i; if we can do so, the First Isomorphism Theorem immediately gives the
isomorphism we want. (Surjectivity means that φ(Z × Z) = Z.) To find φ we simply rely on the
work we did last time, where we determined that

(a, b) = (0, b − 2a) in (Z × Z)/h(1, 2)i

and used the remaining coordinate b − 2a as a hint that the data of an element in (Z × Z)/h(1, 2)i
should be determined by this single integer b − 2a. Thus, we use precisely this data to define our
homomorphism, sending (a, b) to b − 2a:

φ : Z × Z → Z is defined by φ(a, b) = b − 2a.

Now we check that this has the properties we want. First, φ is indeed a homomorphism:

φ((a, b) + (c, d)) = φ(a + c, b + d) = (b + d) − 2(a + c) = (b − 2a) + (d − 2c) = φ(a, b) + φ(c, d).

Next, we note φ is surjective since for any b ∈ Z we have φ(0, b) = b. Finally (a, b) is in ker φ if and
only if b − 2a = 0 if and only if b = 2a if and only if (a, b) = (a, 2a) ∈ h(1, 2)i. Thus, ker φ = h(1, 2)i,
so the First Isomorphism Theorem gives

(Z × Z)/h(1, 2)i = (Z × Z)/ ker φ ∼

= φ(Z × Z) = Z

as desired. The fact that the isomorphism (a, b) → b − 2a is well-defined and is injective on the
quotient is hidden within the machinery of the First Isomorphism Theorem.

Lecture 15: More Isomorphism Theorems

Warm-Up 1. We justify that (Z × Z)/h(5, 5)i ∼ = Z/5Z × Z, which we previously argued informally
should be true. To do so we find a surjective homomorphism

φ : Z × Z → Z/5Z × Z

whose kernel is h(5, 5)i. Recall when we argued informally the idea that we can always reduce the
first coordinate of (a, b) in the quotient mod 5 so that it takes on a value in Z/5Z: if a = 5q + r
with 0 ≤ r < 5, then

(a, b) = (5q + r, b) = (r, b − 5q) in (Z × Z)/h(5, 5)i.

This suggests we should take the first coordinate of φ(a, b) to be a mod 5: φ(a, b) = (a mod 5, ?).
The second coordinate should take on the role of any integer value if we want φ to be surjective,
but we also want it to be true that φ(a, b) = 0 if and only if (a, b) = h(5k, 5k)i. In our case, having
the first coordinate of φ(a, b) be a mod 5 already forces a = 5q to be a multiple of 5 when (a, b)
is to be in the kernel, and then the b − 5q second coordinate in the equivalent form of (a, b) above
becomes simply b − a, which suggests using b − a as the second coordinate, Then for (a, b) to be
in the kernel will require that b = a, so that b equals the same multiple of 5 as does a, which is
precisely what we need in for (a, b) to be the subgroup h(5, 5)i we are quotienting by.
So we define φ by
φ(a, b) = (a mod 5, b − a).
That this is a homomorphism follows from the fact that the mod 5 operation is a homomorphism
and that subtraction on Z × Z is a homomorphism (you can check the details). This φ is surjective

51
since any integer can be written as b − a for some integers a, b, and (a, b) ∈ ker φ if and only if
(a mod 5, b − a) = (0, 0), which is true if and only if b = a = 5q for some q ∈ Z. Hence the kernel
is h(5, 5)i, and so our claim follows from the First Isomorphism Theorem.

Warm-Up 2. Suppose A E G and B E H. Then it is true that A × B is normal in G × H. (We

will omit this verification here, but it is straightforward.) We claim that

(G × H)/(A × B) ∼
= (G/A) × (H/B).

Thus the “naive” answer of G×H G H

A×B = A × B we would expect if these symbols denoted literal fractions
is in fact true. (This is why the G/H notion for quotients is nice: in many ways it lines up with
how we expect “quotients” to behave!) To prove this, we consider the homomorphism

φ : G × H → (G/A) × (H/B) defined by φ(g, h) = (g, h).

To be clear, (g, h) on the right denotes its value as an element (i.e. in terms of cosets) in the
product of quotients (G/A) × (H/B) while (g, h) on the left denotes an actual element of G × H.
But the point is that essentially (g, h) should be sent to (a version of) itself!
That φ is a homomorphisms follows from the way in which multiplication in quotients is defined,
or in other words from properties of projection maps K → K/N , and it is surjective since projection
maps are surjective: any k ∈ K/N is (at least) the projection of k ∈ K. Also, (g, h) ∈ ker φ precisely
when (g, h) = (eG , eH ) in (G/A) × (H/B), which happens precisely when g ∈ A and h ∈ B. Thus
ker φ = A × B, so the First Isomorphism Theorem gives our result.
As an application, this immediately gives that (Z×Z)/(nZ×mZ) is isomorphic to Z/nZ×Z/mZ.

Example. Here is another basic example. The determinant map det : GLn (R) → R× which sends
an invertible matrix to its determinant is a homomorphism, with kernel equal to the special linear
group SLn (R) of matrices of determinant 1. This map is surjective since any x ∈ R× arises as the
determinant of some matrix (say the diagonal matrix with diagonal entries x, 1, . . . , 1), so the First
Isomorphism Theorem gives
GLn (R)/SLn (R) ∼
= R× .
This says that, in some sense, the only data an invertible matrix maintains if we ignore those
matrices which have determinant 1 is its determinant. Or, said another way, given the data of a
nonzero determinant value and a matrix of determinant 1, we can produce the data of an arbitrary
invertible matrix.

Second Isomorphism Theorem. The next isomorphism theorem we consider is a consequence of

the first, but is worth highlighting as a useful result in its own right. As with the second Warm-Up,
the statement makes sense if we think about quotients naively as “fractions”, so the point is that
this can actually be made precise. Here is the statement of the Second Isomorphism Theorem:
Suppose A, B ≤ G with B normal in G. Then AB is a subgroup of G, B is normal in
AB, A ∩ B is normal in A, and AB/B ∼
= A/(A ∩ B).
Let us digest this a bit. First, the final conclusion is the “naive” answer we expect when interpreting
the quotient notation as a literal fraction: in some sense, in AB/B the B’s “cancel out” leaving us
with A, except that we have to now take into account the fact that when “canceling out” elements
of B, those elements of A ∩ B also get “cancelled out”. In this context, “cancelling out’ means to
“set equal to the identity”. Thus, we do get A as the resulting quotient, only we cannot distinguish
between those elements of A which are also B since these are being set equal the identity.

52
 Second, apart from the final isomorphism alone, also important is the fact that AB will in fact
be a subgroup of G. Recall that this is not true in general, and is a crucial part of the statement
since otherwise AB/B would not make sense as a quotient group. That AB being a subgroup of
G is implied by B being normal in G follows from a fact we mentioned previously when discussing
products: HK is a group if and only if HK = KH. (We omitted the details previously, and will
still do so, but you should work it out on your own.) In fact, AB being a subgroup holds under
more general assumptions than B being normal in all of G as we have stated here—for instance,
it is enough to have A ≤ NG (B). (This is the assumption the book actually uses.) But, in the
practical examples we will encounter later on, it will be the case that B is in fact normal in G,
so we prefer to use this assumption in our phrasing of the Second Isomorphism Theorem without
worrying about more general ones which work. To give a brief sense of the context in which this
theorem will arise later, the setting of having a group AB equal to a product like this with B being
normal in it is the typical one which leads to consideration of semi-direct products (in the case
where A ∩ B = 1), which is a tool we will use to help classify groups. The notion of a “semi-direct
product” of groups was introduced in the Discussion 2 Problems handout, and we will review the
necessary details when we actually need them.
Finally, we point out that the other conclusions in the statement—B is normal in AB and A∩B
is normal in A—need to be there in order for the resulting quotients to in fact be groups. The
point is that assuming B is normal in G is actually enough to guarantee that everything showing
up in the quotient isomorphism AB/B ∼ = A/(A ∩ B) makes sense. Moreover, this gives an equality
between indices [AB : B] = [A : A ∩ B], which also follows (in the case where G is finite) from the
following equality we derived previously:

|A||B| |AB| |A|

|AB| = , or equivalently = .
|A ∩ B| |B| |A ∩ B|

Indeed, in some sense this previously inequality foreshadows the statement of the Second Isomor-
phism Theorem.

Proof of Second Theorem. The proof of “isomorphism” part of the Second Isomorphism The-
orem is an application of the First Isomorphism Theorem. (We will omit the check here that B is
normal in AB and that A ∩ B is normal in A under the assumption that B is normal in G, but
these are straightforward verifications.) The goal is to find a surjective homomorphism

φ : A → AB/B

whose kernel is A ∩ B. But essentially if we interpret A and AB/B in the right way, there is a clear
candidate: note A = A{1} is a subgroup of AB, and so elements of A can already be thought of
as elements of AB/B, and thus we can take φ to essentially be the map that sends everything to
itself: φ(a) = a, only with a on the right thought of as an element of AB/B.
It is clear than that this map is a homomorphism. If we take any ab ∈ AB/B, then a = ab in
AB/B, so φ(a) = a = ab and φ is surjective. Finally, the kernel of φ consists of those elements of
A which become trivial after modding out by B, but this is precisely those elements of A which
are actually in B. Hence ker φ = A ∩ B and the First Isomorphism Theorem gives the result. The
point is that the map which induces the isomorphism which is claimed to exist is an “obvious” one
in the sense that we just send each element to a “version of” itself.

Third Isomorphism Theorem. The third theorem we consider says naively that when “dividing
g/h
fractions with the same denominator, the denominators cancel out”: k/h = kg . To be precise:

53
 Suppose H ≤ K ≤ G with H, K both normal in G. Then K/H is normal in G/H and
we have an isomorphism (G/H)/(K/H) ∼
= G/K.

(Note the resulting isomorphism has on the left a “qoutient group of quotient groups”!) It should
be clear that K/H is a subgroup of G/H, since the fact that K was closed under multiplication in G
is not lost when introducing extra relations. That K/H is normal in G/H is also a straightforward
verification, so we omit it.
The isomorphism which is claimed to exist, as in the case of the Second Isomorphism Theorem,
is a consequence of the First Isomorphism Theorem. Define φ : G/H → G/K by sending anything
in G/H to itself: φ(g) = g. This makes sense—and gives a well-defined map—precisely because H
is assumed to be in K: g ∈ G/H still determines a unique element of G/K since introducing more
relations will not invalidate any equalities which existed beforehand. More concretely: if g = g 0 in
G/H, g will still equal g 0 in G/K because g = g 0 h for some h ∈ H ≤ K immediately gives g = g 0 h
for some h ∈ K. This is crucial, since otherwise φ would not be well-defined.
This φ then preserves multiplication, is surjective, and has kernel equal to the elements of G/H
that become the identity in G/K, which means those elements which were in K/H ≤ G/H to begin
with. Thus (G/H)/ ker φ ∼ = φ(G/H) is precisely the result we want. It is all just really a matter of
making sure that the various elements we consider are we we say they are, by chasing through the
definitions of the various quotients and groups we are considering.
Here is a simple application. For n dividing m, nZ contains mZ so nZ/mZ is a subgroup of
Z/mZ. Then the Third Isomorphism Theorem gives:

(Z/mZ)/(nZ/mZ) ∼
= Z/nZ.

Lecture 16: Simple and Solvable Groups

Warm-Up. Let C× I := {zI | z ∈ C× } denote the group of nonzero complex scalar multiples of
the n × n identity matrix. (This is actually the center of GLn (C), and so is normal in GLn (C).)
We identity the quotient GLn (C)/C× I as a quotient of SLn (C) instead. The key observation is
that any matrix in GLn (C) can be written as a product of elements of SLn (C) and C× I:
  √
1 n
A= √ n
det A
A ( det A · I).
√
Here, n det A denotes an n-th root (any will work, pick one!) of the complex number det A ∈ C× ,
and the fact that the first term has determinant 1 follows from the general result that for any scalar
c we have det(cA) = cn det A. Thus GLn (C) = SLn (C) · C× I, so since C× I is normal in GLn (C)
the Second Isomorphism Theorem gives

GLn (C)/C× I = (SLn (C) · C× I)/C× I ∼

= SLn (C)/(SLn (C) ∩ C× I).

A scalar multiple of I has determinant 1 if and only if the scalar used is an n-th root of unity (since
det(cI) = cn det I), so the intersection SLn (C) ∩ C× I consists of these specific matrices. Hence

GLn (C)/C× I ∼
= SLn (C)/{n-th roots of unity}.

Let us give some context behind this result, just for the sake of interest. The group GLn (C)/C×
is an example of what’s called a complex projective general linear group, and arises in the study
of complex projective (n − 1)-space. (It is the group of so-called projective transformations on
this space.) This result gives an alternate description of this group, and indeed shows that the

54
complex projective general linear group is isomorphic to the complex projective special linear group
of the same dimension. (In general, in projective geometry we study geometry “up to scalar
multiplication”, which is hence why we quotient GLn (C) and SLn (C) out by the appropriate
“scalar” subgroups of each. In other words, in projective geometry we cannot distinguish between
A and cA, so we should treat them as the “same”.)

Fourth Isomorphism Theorem. Let us mention one more “isomorphism theorem”, which gives
information about subgroups of a quotient G/N in terms of the subgroups of G. We should note
that this result is not commonly listed as one of the isomorphism theorems in other sources, but our
book does and so we do too. (Some other sources also switch the position of the Second and Third
Isomorphism Theorems; everyone agrees on what the First Isomorphism Theorem is though.)
The key observation is that any subgroup of G/N is of the form H/N where H is a subgroup
of G which contains N . Indeed, the preimage of any subgroup H 0 of G/N under the projection
map φ : G → G/N will be a subgroup H := φ−1 (H 0 ) of G containing N (since any n ∈ N maps
to φ(n) = e ∈ H 0 ≤ G/N , and so is in H), and its image under the same projection map is
H 0 = φ(H) = H/N , showing that H 0 = H/N is of the required form. Note that for any subgroup
A of G, its image φ(A) under projection is a subgroup of G/N regardless of whether A contained
N , but the point is that this same φ(A) ≤ G/N is also the image of its preimage φ−1 (φ(A)), which
contains the original A. Hence even φ(A) ≤ G/N will still be of the required form, it’s just that
the required form is φ−1 (φ(A))/N and not “A/N ”, which is not defined if A does not contain N .
The Fourth Isomorphism Theorem then describes, under this correspondence, properties of
subgroups of G/N in terms of those of G which contain N . For instance:
• H/N is normal in G/N if and only if H is normal in G;
• the index [G/N : H/N ] of H/N in G/N equals the index [G : H] of H in G;
• hH/N, K/N i = hH, Ki/N (angled brackets here referring to subgroups generated by);
and so on. We will not list all the properties here, but will point them out when needed. Ultimately,
this is about relating the lattice of subgroups of G/N to that of G.

Simple groups. We now define two types of groups which will play a role going forward: those
which are simple and those which are solvable. To set the stage, consider the problem of “recon-
structing” a group G from the data of a normal subgroup N and the corresponding quotient G/N .
We will visualize this scenario in the following way:
N ,→ G  G/N,
where ,→ denotes an injective homomorphism—in this case the inclusion of N into G—and  a
surjective homomorphism, in this case the projection of G onto G/N . The problem is then, given
N and Q, to find the groups ? which fit into the middle of:
N ,→ ?  Q
and thus will contain N ≤ ? as a subgroup and have Q ∼ = ?/N as a quotient. This is known as an
extension problem: can we “extend” Q by N to fill in the middle? The goal is to use information
about the “simpler” (say of smaller order in the finite case) groups N and Q = G/N to derive
information about G. In some sense, we ask if we can “break” G down into pieces, study those
pieces, and then put them back together to form G.
Now given the setup N ,→ G  G/N , if N itself had a normal subgroup N 0 E N , we could
then “break” N down in the same way:
N 0 ,→ N  N/N 0

55
and hope to use the even simpler pieces N 0 , N/N 0 to build up to N , and then use N, G/N to build
up to G. If N 0 itself has a normal subgroup we can do this again, and so on and so on.
A simple group is then one which cannot be “broken down” further in this way. To be precise, a
group G is simple if it has no nontrivial, proper normal subgroup: if N E G, then N = 1 or N = G.
Simple groups, as we will see, play a role analogous to prime numbers, in that they form—in the
finite case at least—the building blocks of all groups. So far, the only groups we have seen which
are simple are the cyclic groups Z/pZ of prime order.

Composition series. The problem of “breaking a group down into simple pieces” is then made
precise by the following notion. A composition series of G is a chain of subgroups:

1 = N0 E N1 E . . . E Nn−1 E Nn = G

from the trivial subgroup 1 up to G, were each Ni−1 is normal in the next Ni (it is NOT the case
that each Ni need be normal in G) and where each successive quotient Ni /Ni−1 is simple. It is this
“simple quotient” condition which says that the chain cannot be made any longer (i.e. broken down
further): Ni /Ni−1 is not simple if and only there exists Ni−1 ≤ K E N giving a nontrivial, proper
normal subgroup K/Ni−1 of N/Ni1 which would then make the chain above longer by inserting K
between Ni−1 and Ni . The successive quotients Ni /Ni−1 are called the composition factors of the
series. Note that N1 is itself simple since it is the first composition factor N1 /N0 .
Not all groups have a composition series; for instance, Z does not since it has no nontrivial
simple subgroups to play the role of N1 in the chain, because every subgroup of Z is also infinite
cyclic and hence not simple. But, it is true at least that all finite groups have a composition series,
as you will prove on a homework problem. For example, one composition series of Z/12Z is:

0 E 4Z/12Z E 2Z/12Z E Z/12Z.

| {z } | {z }
h4i h2i

The composition factors (respectively going left to right) are: Z/3Z, Z/2Z, and Z/2Z. (This is an
application of the Third Isomorphism Theorem; for instance (Z/12Z)/(2Z/12Z) ∼ = Z/2Z.) Another
composition series is given by:

0 E 6Z/12Z E 3Z/12Z E Z/12Z

and a third by
0 E 6Z/12Z E 2Z/12Z E Z/12Z.
These respectively (left to right) have composition factors Z/2Z, Z/2Z, Z/3Z and Z/2Z, Z/3Z, Z/2Z,
so even though it is not true that a composition series itself is unique (there are three here), what
is true is that the composition factors of each series are unique up to some rearrangement. (In fact,
the terms in the composition factors mimic the possible prime factorizations of 12—3 · 2 · 2, 2 · 2 · 3,
and 2 · 3 · 2—which will be true of all composition series of Z/nZ in general.)
This final observation is true in general for any finite group: any two composition series of a
finite group have the same length, and the composition factors in any such series are the same up
to permutation of the factors. This is the content of the Jordan-Hölder Theorem, which can be
proved using the Second Isomorphism Theorem. We will not give a proof here since composition
series will not play a big role in what we will be doing, but some of the exercises in the book outline
one proof if you’re interested in trying it out. (The idea is to first prove the special case where one
series has length three, and then to induct on the minimal length of two given series in general.
The length two case can only occur for simple groups and there is not much to say beyond that.)

56
Extensions. Given a composition series

1 = N0 E N1 E . . . E Nn−1 E Nn = G

of G, we can then hope to “build” from N1 all the way up to G in the manner described previously.
To be precise, the simple groups N1 and N2 /N1 fit into the picture

N1 ,→ N2  N2 /N1 ,

so if we have complete knowledge of these simple groups, we hope to obtain complete knowledge of
N2 . Then N2 and the simple group N3 /N2 fit into

N2 ,→ N2  N3 /N2 ,

so if we have built up N2 and have complete knowledge of the simple group N3 /N2 , we hope
to obtain complete knowledge of N3 . But then having built up to N3 , we hope to use complete
knowledge of the simple group N4 /N3 to next build up to N4 , then to N5 , and so on up to Nn = G.
We will say a bit next time about how part of this problem—having complete knowledge of all
simple groups—has actually been fully solved in the finite case. However, the remaining part—
knowing how to “build up” from simple groups to other groups—is intractable. The issue is that
in general knowing N and G/N is not enough to actually identity G itself, since there can in fact
be multiple groups which fit into the “extension problem” given by

N ,→ ?  G/N.

In general, such a group is in the middle is called an extension of G/N by N , of which there can
be many, so that even if all the simple groups needed can be well-understood, it might not be
possible to solve all the resulting extension problems to see how to put them together correctly
when building up to G. The best we can hope to do is to classify (maybe up to isomorphism) the
possible extensions we get along the way, so that we get a range of possibilities for what G can be.
We will not go into this much further, except for one scenario when we can say quite a bit more,
that being when an extension of Q = G/N by N can be described as a semi-direct product of Q
and N . Indeed, semi-direct products arise precisely when trying to solve such extension problems,
and were briefly previously when discussing the types of scenarios we’ll see later on where the
Second Isomorphism Theorem could prove useful. As we said back then, we’ll save giving the
formal definition until we actually need to do so later.

Solvable groups. Even though we will not do much with composition series, there is a related
notion which will be crucially important for us, at least in the spring. A group G is solvable if
there is a chain of normal subgroups from the trivial group 1 up to G:

1 = N0 E N1 E . . . E Nn−1 E Nn = G

with each successive quotient Ni /Ni−1 being abelian. For instance, any abelian group G is solv-
able, simply because the two-term chain 1 E G already satisfies the requirement in the definition.
Dihedral groups are also solvable since

1 E hri E D2n

has successive quotients hri and D2n /hri ∼

= Z/2Z, which are both abelian. In fact, any group of
order less than 60 is solvable, which we will essentially prove (although we might not phrase it

57
all completely in terms of the language of “solvability”) in the remaining time this quarter. The
smallest non-solvable group occurs in the order 60 case, and is a group we will define next time.
The notion of solvability above might seem to be quite random at first, but we will see in the
spring that this is precisely (!!!) what is needed in order to say that the roots of polynomial can
be described via an explicit algebraic formula. (Actually, this is a bit of a lie: what is actually
needed are cyclic quotients, but for finite groups having abelian quotients is equivalent to having
cyclic quotients, as we will see.) Indeed, the term “solvable” here comes exactly this relation to
solving polynomial equations. The fact that there is a non-solvable group of order 60 but not of
any smaller order is ultimately the reason why there exists a quartic (and quadratic and cubic)
formula, but not a “quintic formula”. We have much to look forward to!

Lecture 17: Alternating Groups

Warm-Up 1. We find all composition series of Z/30Z. First let us clarify an observation we
made last time when considering Z/12Z, that a composition series should correspond to a way of
expressing the prime factorization of 30. The second to last term pZ/30Z in a composition series will
be generated by some p, and in order this the corresponding quotient (Z/30Z)/(pZ/30Z) ∼ = Z/pZ
to be simple requires that p be prime. But then the same will be true of the term before pZ/30Z
in the composition series: pqZ/30Z (which is what a subgroup of pZ/30Z would have to look like)
will yield a simple quotient Z/qZ only when q is prime, and so on the further to the left we move in
the composition series. Thus we start by picking one prime dividing 30 to form the second-to-last
term, then another (not necessarily different) prime to form the term before this, and so on.
Thus since 30 = 2 · 3 · 5, there are 3 possible composition series, with composition factors
Z/2Z, Z/3Z, Z/5Z in some order. (So since 24 = 2 · 2 · 2 · 3, Z/24Z has 4 composition series.) The
composition series of Z/30Z are:
0 E 6Z/30Z E 2Z/30Z E Z/30Z
0 E 6Z/30Z E 3Z/30Z E Z/30Z
0 E 10Z/30Z E 5Z/30Z E Z/30Z
with composition factors {Z/5Z, Z/3Z, Z/2Z}, {Z/5Z, Z/2Z, Z/3Z}, and {Z/3Z, Z/2Z, Z/5Z} re-
spectively. (The fact that any finite group has a composition series and that composition factors
are unique essentially gives a very elaborate proof of the Fundamental Theorem of Arithmetic!)

Warm-Up 2. Suppose N E G. We show that if N and G/N are solvable, then G is solvable.
Thus, “solvability” is a property preserved under the taking of extensions. This will be useful later
when we come to a need to know whether a group is solvable or not, in that we can determine if it
is by considering the hopefully simpler groups N and G/N .
This comes down to the correspondence between subgroups of G/N and those of N together
with the Fourth Isomorphism Theorem. Since N is solvable there exists a chain
1 E N1 E . . . E Nn−1 E N
with each Ni /Ni−1 abelian, and since G/N is solvable there exists a chain
N/N E H1 /N E . . . E Hn−1 /N E G/N
with each successive quotient abelian. Here, each Hi is a subgroup of G containing N , and we have
used the fact that normal subgroups of Hi /N correspond to normal subgroups of Hi . This thus
lifts to a chain of the form
N E H1 E . . . E Hn−1 E G,

58
which together with the first chain above gives

1 E N1 E . . . E Nn−1 E N E H1 E . . . E Hn−1 E G.

We already know from above that the successive quotients for the terms up to N are abelian, and
for the remaining quotients the Third Isomorphism Theorem gives

(Hi /N )/(Hi1 /N ) ∼
= Hi /Hi−1 ,

so since the “quotients of quotients” on the left are abelian so are the quotients on the right. Hence
G is solvable if N and G/N are.

Classification of simple groups. We will briefly mention one last fact, simply for the sake of
interest and because it is truly a monumental achievement in modern mathematics of which any
student should be aware. Looking back to the use of composition series to “build up” to arbitrary
(finite) groups from simple groups, we mentioned that the process of “building up” in this way—
i.e. the extension problem—was intractable in general, but that the problem of understanding the
possible simple groups used in this process was in fact solved: the complete list of finite simple
groups (up to isomorphism) is known. The problem of classifying finite simple groups in this way
began in earnest in the 1950’s, and was thought to be completed at first in the 1980’s, only for
some holes to be discovered soon after, which were finally fully overcome in the early 2000’s.
Let us describe in a very rough way the list of finite simple groups. First, there are the cyclic
groups Z/pZ of prime order, which make up the first family of finite simple groups. Next, there is
a second family of groups denoted by An for n ≥ 5, which we will define after this brief digression.
Next we have the so-called groups of “Lie type” which in a very very rough sense are essentially
groups which can be described as certain matrix groups, only with entries being a more elaborate
type of “number”, like something in Z/pZ for instance. We will not give any more of a formal
definition of a group of “Lie type” than this, but the point is that this comprises an infinite number
of groups which can all be constructed or described in “similar” ways.
Finally, there are exactly 26 remaining finite simple groups which do not fall into any such nice
“family” as opposed to those above, and whose existence might appear at first to be a random
coincidence. These are the known as the sporadic simple groups. Of these, the one of largest order
is called the monster group (yes, that is indeed the official name), due to the fact that its order is:

808, 017, 424, 794, 512, 875, 886, 459, 904, 961, 710, 757, 005, 754, 368, 000, 000, 000.

(Holy cow is that a huge number!) It is quite astonishing that a group of this crazy larger order
could in fact be simple (this order is highly non-prime), let alone that anyone would have ever
thought to look for it, and be able to actually construct it. Even crazier perhaps is the fact that
monster group is now known to have connections to theoretical physics, but that is a story for
another time and place. It is truly a testament to the power of mathematical thinking that this
group, and indeed all finite simple groups, have indeed been found.

Parity of permutations. The groups An for n ≥ 5 in the second family mentioned in the
classification above are known as the alternating groups, which we will now work towards defining.
First, we need the notion of what it means for a permutation σ ∈ Sn to be even or odd :

σ ∈ Sn is an even permutation if it can be written as a product of an even number of

transpositions, and it is an odd permutation if it can be written as a product of an odd
number of transpositions.

59
(Recall that any permutation can be written as a product of some number of transpositions since
the transpositions generate all of Sn .) The key point is that this definition makes sense, in that a
permutation cannot be both even and odd simultaneously: a given permutation can be written as
a product of transpositions in multiple different ways, but all such ways will always contain either
an even number of 2-cycles overall or an odd number of 2-cycles overall. We will prove this fact
shortly. The term “parity of a permutation ” refers to this property of being either even or odd.
For instance, a single 2-cycle (ab) is odd, since it already consists of one transposition. A 3-cycle
(abc) is even since
(abc) = (ab)(bc)
consists of two transpositions. Again, this type of expression is not unique, since we also have for
instance
(abc) = (ab)(bc)(ac)(ac)(bc)(bc),
but what is unique is the parity of the number of transpositions used. A 4-cycle is odd:

(abcd) = (ab)(bc)(cd),

and so on in general a cycle of even length is odd and a cycle of odd length is even. (Note the
mismatch between the parity of the length and the parity of the cycle as a permutation!) From this,
it is easy to determine the parity of any permutation by looking at its disjoint cycle decomposition,
since odd · odd = even, odd · even = odd = even · odd, and even · even = even.

Alternating groups. The alternating group An is then defined to be the subgroup of Sn consisting
of the even permutations. That this is a subgroup follows from the fact that the identity permutation
(1) = (12)(12) is even, even times even is even, and inverting cycles preserve their lengths. The
order of An is exactly half the order of Sn : |An | = n! 2 . (Thus An has index 2 in Sn , and the
non-identity coset consists of the odd permutations.) This can be seen from the fact that there are
as many even permutations as there are odd ones, since the function

{even permutations} → {odd permutations} defined by σ 7→ σ(12)

is a bijection between finite sets (it equals its own inverse).

So, for instance, A4 is a group of order 4!2 = 242 = 12. It contains the identity, all 3-cycles (of
which there are 8), and 3 more elements which are products of two disjoint transpositions:

A4 = {(1), (123), (132), (134), (143), (124), (142), (234), (243), (12)(34), (13)(24), (14)(23)}.

On a previous set of discussion problems it was shown (although we were not using the language
of even and odd permutations at the time) that

N = {(1), (12)(34), (13)(24), (14)(23)}

is actually a normal subgroup of A4 , which shows that A4 is not simple. We will show later, after
we have a few more tools built up, that An is simple for n ≥ 5. (A5 has order 120
2 = 60, and is in
fact the smallest non-abelian simple group, and the smallest non-solvable group.) This is not to
say that An is simple only for n ≥ 5, since A1 = A2 = 1 and A3 ∼= Z/3Z are simple as well; so, in
fact, A4 is the only non-simple alternating group.

Inversions. We now work towards proving that our notion of even vs odd when it comes to per-
mutations is well-defined in that a given permutation can only be one of these. We give essentially

60
the same argument as the book, but phrased in a different (I think simpler) way. (In fact, the book
takes the concept we are about to introduce as the starting point in defining even vs odd, and then
later shows that this can be recast in terms of the number transpositions required in a product
expression. I prefer taking the latter approach as the definition, since we almost exclusively think
of elements of Sn in terms of cycles anyway)
Given σ ∈ Sn , an inversion occurs whenever we have i < j but σ(i) > σ(j). (Here σ(i) is the
result of evaluating the function σ at i ∈ {1, 2, . . . , n}.) As the name suggests, an inversion occurs
whenever σ “inverts” two elements in {1, 2, . . . , n} in the sense that it swaps their ordering relative
to one another. For instance, take the permutation σ = (1423) which as a function sends each of
1234 respectively to
4312.
Here, σ inverts 1 < 2 since σ(1) = 4 > σ(2) = 3, it inverts 1 < 3 since 4 > 1, and it also inverts
1 < 4, 2 < 3, and 2 < 4. It does not invert 3 < 4 since σ(3) = 1 < σ(4) = 2 still holds. (Visually, in
the notation above, an inversion occurs whenever a number is larger than a number to its right.)
Thus σ = (1423) has 5 inversions. We define the sign of a permutation σ, denoted sgn(σ), to be
(
1 number of inversions is even
(−1)# of inversions =
−1 number of inversions is odd.

The claim is that this notion of parity in terms of the number of inversions is the same as the one
we defined previously, so that even permutations have sign 1 and odd permutations have sign −1.
Before proving this, let us clarify the book’s approach to defining parity. Take the product of
polynomials xi − xj for i < j among 1, 2, . . . , n:
Y
(xi − xj ).
1≤i<j≤n

An element σ ∈ Sn acts on this by permuting the indices:

Y Y
(xi − xj ) 7→ (xσ(i) − xσ(j) ).
1≤i<j≤n 1≤i<j≤n

Each resulting factor xσ(i) − xσ(j) still occurs among our original factors xi − xj , only that the order
of the indices might be switched: if σ inverts i < j, xσ(i) − xσ(j) has the wrong index order, and
so actually equals the negative of the factor xσ(j) − xσ(i) in the original product. Thus the product
above with permuted indices equals ±1 the original product, with the sign coming from −1 raised
to the number of indices which were inverted. Hence we recover the definition of sgn(σ) we have
given above. You can decide for yourself whether our approach or the book’s is simpler to follow.

Parity makes sense. Let us finally then justify that the two notions of parity introduced above
(number of transpositions vs sign) agree. The key point is that the inversion/sign approach depends
only on the behavior of a permutation as a function, and hence whether we get sgn = 1 or −1
depends only on the permutation without any reference to what its cycle decompositions look like.
Take a transposition of adjacent terms (a a + 1), and consider the product τ = σ(a a + 1) in
relation to σ ∈ Sn itself. We claim that τ and σ differ by exactly 1 in the number of inversions in
each. First, τ and σ have the same effect on any i ∈ {1, 2, . . . , n} which is not a or a + 1, so the
number of inversions in each can only differ when considering a < a + 1. But:

τ (a) = σ(a + 1) and τ (a + 1) = σ(a),

61
so that if σ inverts a < a + 1, τ will not, while if σ does not invert a < a + 1, τ will. Hence σ
and σ(a a + 1) do differ in their number of inversions by exactly 1; in other words, each time we
multiply by an adjacent transposition (a a + 1), we change the number of inversions by 1.
We know from prior work that any σ ∈ Sn can be written as a product of adjacent transpositions.
(This was a previous homework problem: (12), (23), . . . , (n − 1 n) generate Sn .) An important
observation is that using only adjacent transpositions as opposed to arbitrary transpositions does
not alter whether we use an even or odd number overall: we have

(a a + k) = (a a + 1)(a + 1 a + k)(a a + 1),

and inductively continuing to break (a + 1 a + k) down into an expression with (a + 2 a + k), then
(a + 3, a + k), and so on will result in an odd number (the 2 broken off at each step plus the
(a + k − 1 a + k) leftover in the end) of adjacent transpositions overall. Thus if we write σ as a
product of transpositions, and then rewrite each of those in terms of only adjacent transpositions,
the parity of the number used remains the same.
But each time we introduce a new adjacent transposition, the parity (in terms of number of
inversions) changes as we showed above. Thus, suppose σ ∈ Sn and write it has a product of any
number of transpositions, which we can then take to be adjacent:

σ = τ1 . . . τk .

Since τ1 has one inversion, τ1 τ2 has an even number, then τ1 τ2 τ3 has an odd number, etc. Since
the number of inversions—and hence its parity—depends only on σ as a function, the right side
above must have this same parity, which means that k is even when sgn(σ) = 1 and odd when
sgn(σ) = −1. Thus the two notions of parity agree and hence a permutation cannot be both even
and odd (in the sense of our original definition) at the same time.

Lecture 18: Orbit-Stabilizer Theorem

Warm-Up 1. We find a composition series of A4 , and use it to show that A4 is solvable. Last
time we stated that
N = {(1), (12)(34), (13)(24), (14)(23)}
is a normal subgroup of A4 , as was shown on the Discussion 3 Problems handout. This N is abelian
(it is isomorphic to Z/2Z × Z/2Z), so any subgroup is normal. Thus for instance

1 E {(1), (12)(34)} E N E A4

is one possible chain of normal subgroups. The successive quotients going from left to right are:
Z/2Z, Z/2Z, Z/3Z, which are all simple, so this is a composition series. Other composition series
are obtained by using either (13)(24) or (14)(23) instead of (12)(34) to give the second term.
Since the quotients above are all abelian, this shows immediately that A4 is solvable.

Warm-Up 2. We show that Sn can be realized as a subgroup of An+2 . Define the map
(
σ if σ is even
φ : Sn → An+2 by φ(σ) =
σ(n + 1 n + 2) if σ is odd.

Then φ is a homomorphism, which can be checked by considering the different cases:

• if σ, τ ∈ Sn are both even, then στ is even and φ(στ ) = στ = φ(σ)φ(τ );

62
 • if one of σ, τ is even and the other odd, then στ is odd and φ(στ ) = στ (n+1 n+2) = φ(σ)φ(τ ),
where we use that (n + 1 n + 2) is disjoint from σ and τ to say it commutes with each; and
• if σ, τ are both odd, then στ is even and φ(στ ) = στ = σ(n + 1 n + 2)τ (n + 1 n + 2), which
is φ(σ)φ(τ ), where we use the fact that (n + 1 n + 2) has order 2.

Moreover, if φ(σ) = φ(τ ), then either both of these elements of An+2 fix n + 1 and n + 2, or they
both transpose them: in the first case, we have have φ(σ) = σ and φ(τ ) = τ , so σ = τ , while in the
second case φ(σ) = σ(n + 1 n + 2) and φ(τ ) = τ (n + 1 n + 2), so that σ(n + 1 n + 2) = τ (n + 1 n + 2)
also implies σ = τ . This shows that φ is injective, so Sn is isomorphic to the image of φ, which is
the desired subgroup of An+2 .

An is normal. We finish, for now, our discussion of An by pointing out that it is always a normal
subgroup of Sn . This comes immediately from fact An has index 2: in general, if [G : H] = 2, then
the left cosets of H are H, gH for some g ∈/ H, and the right cosets are H, Hg, so since the sets in
each of these pairs are disjoint and together give everything in G, we must have gH = Hg, so that
H is normal as claimed. (That index 2 implies normal also follows from the homework problem
saying that a subgroup H ≤ G whose index is the smallest prime dividing |G| must be normal, but
the argument above is a lot simpler than this general fact.)
We can also see that An is normal by noting that it is the kernel of the sign homomorphism:
sgn : Sn → {±1}. Here we view {±1} as a group under multiplication, and the fact that sgn
preserves multiplication comes from the fact that even times even is even, even times odd is odd,
and odd times odd is even, where here “even” and “odd” refer to permutations, not integers. The
identity of {±1} is 1, and those permutations with sgn = 1 are precisely the ones which are even.
As a consequence, we can now say that S4 is solvable:

1 E {(1), (12)(34)} E {(1), (12)(34), (13)(24), (14)(23)} E A4 E S4

has successive quotients which are all prime cyclic and hence abelian.

Orbit-stabilizer theorem. We will return to properties of permutation groups later, but for now
take a brief detour to highlight a useful fact about group actions: the orbit-stabilizer theorem. The
book states this as part of a proposition, but does not call it “orbit-stabilizer theorem” nor give it
the prominence it deserves.
Here is the claim. Suppose we are given an action G y X. Recall that for x ∈ X, Gx denotes
the orbit through x and Gx the stabilizer of x. The orbit-stabilizer theorem states that the index
of Gx in G is equal to the cardinality of Gx:

[G : Gx ] = |Gx|.

Thus, the size of any orbit for any group action of G can be obtained solely from information about
the group and its subgroups alone. (In a sense then, this says that studying arbitrary group actions
“essentially” comes down to studying the action of G on its subsets by left multiplication.) One of
the more important consequences is then, that, when G is finite, we have

|G| = |Gx ||Gx|

|G|
(this uses [G : Gx ] = |G x|
), so that the order of G can be derived solely from knowledge of a single
orbit and the corresponding stabilizer.
The orbit-stabilizer theorem follows from recognizing when two elements g, h ∈ G give the same
element in the orbit Gx: g · x = h · x if and only if x = (g −1 h) · x if and only if g −1 h ∈ Gx if and

63
only if g and h determine the same coset of Gx in G. Thus, the number of distinct elements in Gx
equals the number of distinct cosets of Gx (or, said another way so that this applies even in the
infinite setting, we have a bijection between the orbit Gx and the set G/Gx of cosets), which gives
our claim.

Example. Here is an example of the typical type of application of the orbit-stabilizer theorem.
Previously we argued that D2n has order 2n by stating that it consisted of n rotations and n
reflections, or by using the generators and relations to explicitly describe all elements. We can also
obtain this fact from the orbit-stabilizer theorem as follows, where the point is that we do not need
to have any a priori knowledge about generators and relations, nor precisely how many rotations
and reflections there will be ahead of time.
Consider the usual action of D2n on {1, 2, . . . , n}, viewed as the vertices of a regular n-gon.
Take vertex 1. For any other vertex, there is a rotation which will take 1 to it, so all vertices lie
in the orbit of 1 under the action of D2n . Also, the only non-identity element of D2n which fixes
1 is the reflection across the line passing through 1, so the stabilizer of 1 has order 2. Thus the
orbit-stabilizer theorem gives

|D2n | = |stabilizer||orbit| = 2n,

just as we would expect.

Symmetries of a tetrahedron. Here is another example, now in a setting where do not know
the answer ahead of time. Let G be the group of rigid (i.e. rotational) symmetries of a regular
tetrahedron. (Recall that “rigid” symmetries are ones which can be realized by motions in R3 , so
reflections of a 3-dimensional solid do not count.) Label the vertices of the tetrahedron by 1, 2, 3, 4:

For any vertex there is a rotation which will move vertex 1 to it, so the orbit through vertex 1 has
size 4. Also, the only rotations which fix vertex 1 are those which occur around the line passing
through 1 and the center of the opposite face, and there are 3 of these: the opposite face is an
equilateral triangle, and there are three rotations of this triangle. Thus, the orbit-stabilizer theorem
gives the order of G as:
|G| = |stabilizer||orbit| = 3 · 4 = 12.
Now, to determine which group of order 12 we have, use the action above to realize G as
a subgroup of S4 upon permuting the vertices. (Note that different rotations induce different
permutations, so the map sending an element of G to the element of S4 it induces is indeed
injective.) With the labeling above, the rotations which occur around the line passing through 1
and the center of the opposite face are:

(1), (234), (243).

64
The rotations which occur around the line passing through vertex 2 will only permute vertices
1, 2, 4, and we get:
(1), (124), (142).
And so on, considering the rotations which fix each of the other two vertices will give the remaining
3-cycles, so thus far we have:

(1), (234), (243), (124), (142), (134), (143), (123), (132) ∈ G.

Products of any of two of these will also give elements of G, and by computing them (and knowing
that G has 12 elements in order to avoid doing so much work), we get that G is exactly A4 :

G = {(1), (234), (243), (124), (142), (134), (143), (123), (132), (12)(34), (13)(24), (14)(23).}

(This essentially proves the fact that A4 is generated by 3-cycles, which you will show on the
homework is true for general n.)
This finishes justifying our claim, but we can now try to visualize the exact rotations that
produce (12)(34), (13)(24), (14)(23). These are tougher to see than the ones which give 3-cycles,
since the axes of rotations are not so straightforward to visualize. They arise by taking lines which
pass through the midpoint of one edge and the midpoint of the opposite edge (same colors in the
picture below):

If you think about the effect of rotating around these lines, you can convince yourself that such
rotations will indeed transpose two pairs of vertices.

Lecture 19: More on Permutations

Warm-Up. We determine the rotational symmetry group G of a cube. For convenience, we label
the vertices of the cube in the following way:

We first find the order of G. Each rotation in G induces a permutation of the vertices, so we get an
action of G on the set of 8 vertices. Pick a vertex, say vertex 1. Then the orbit through 1 consists
of all 8 vertices since any one vertex can be moved to any other via some rotation. (This is what
it means to say that this action is transitive: there is only one orbit.) The stabilizer of vertex 1

65
consists of those rotations which occur about the axis connecting vertices 1 and 7, and there are
three such rotations overall. (This is not to easy to see via the picture alone, but if you take an
actual physical cube and spin it while holding two opposite vertices fixed, you will be able to easily
see that there are three resulting rotations.) Thus the orbit-stabilizer theorem gives:

|G| = |G1 ||G · 1| = 3 · 8 = 24.

Let us also derive this by considering a different action, where the size of the stabilizer is possibly
simpler to identify. Any rotation also induces a permutation of the faces (i.e. sides) of the cube, so
G acts on the set of 6 faces. This action is also transitive, since any one face can be rotated into any
other, so there is only one orbit of size 6. The stabilizer of, say, the bottom face consists of those
rotations which occur about the vertical axis, and there are 4 such rotations overall: 90, 180, 270, 0.
Thus we get |G| = |Gbottom face ||G · (bottom face)| = 4 · 6 = 24 as well.
Now the goal is to determine which group of order 24 we actually have. One approach is to use
the action of G on the vertices to realize G as a subgroup of S8 : each rotation induces a different
permutation of the 8 vertices, so we get an injective homomorphism G ,→ S8 (this is what it means
to say that this action is faithful : different elements induce different permutations.) whose image
is a subgroup of S8 which is isomorphic to G. So one answer as to what group G actually is to
write down all of the resulting permutations. For instance, rotating around the vertical axis by 90
counterclockwise if viewed from above has the following effect on the vertices:

1 7→ 4 7→ 3 7→ 2 7→ 1 and 5 7→ 8 7→ 7 7→ 6 7→ 5,

so this rotation gives the permutation (1432)(5876). Writing down all 24 permutations of S8 we
get is tedious and will not help to easily identity what G actually is. Alternatively, we could
also consider the action (also faithful) on the faces to realize G as a subgroup of S6 , which gives
“simpler” description of G than the first approach, but still not satisfying.
So we look for a better action to consider. Note that G is not abelian, since composition of 3-
dimensional rotations is not commutative (composition of 2-dimensional rotations is commutative!),
and G is not D24 since D24 has 12 elements of order 2 whereas G only has 9. (This is not so easy
to see purely geometrically, but if nothing else you can write down G as a subgroup of S8 or S6 and
determine the elements of order 2 from there.) Apart from D24 , the only other non-abelian group
of order 24 we have worked with so far is S4 , so perhaps let us make a guess that G ∼ = S4 . If so,
and to prove this, we should be looking for an action of G on a 4-element set.
In fact, we can take the action of G on the four diagonals of the cube, or equivalently on the
set of pairs of vertices {1, 7}, {2, 8}, {3, 5}, {4, 6} which form the endpoints of the diagonals. Any
rotation induces a permutation of these 4 objects (diagonals or pairs), so we get a homomorphism
G → S4 . Different rotations will induce different permutations (it might take a little effort to see
this geometrically!), so this action is faithful, meaning that the map G → S4 is injective. Thus G
is isomorphic to the image, and since |G| = 24 this image has order 24 as well, meaning that the
image is all of S4 . Hence G ∼ = S4 . This same action could also have been used to determine the
order of G, although it might be a little tougher to do so here as opposed to using the action on the
faces for instance. (For the action on the diagonals/pairs, there is a single orbit of size 4 and each
stabilizer has order 6, which come from the three rotations which fix a specific vertex, and the three
rotations which transpose that vertex with its opposite vertex, and thus leave the corresponding
diagonal unchanged.) Thus this is a nice example where considering different actions is very useful:
one action to find the order, and a different to identity the group afterwards.

Rubik’s cube group. Let us take a brief digression to discuss a particularly interesting example

66
of a group which can studied using these tools, the so-called Rubik’s cube group, which everyone
should see at least once in their lifetime. Consider a standard Rubik’s cube:

with corners labeled with 1, . . . , 8 (corner 8 in the bottom, back, left corner is not visible in the
picture above) and “edge” blocks labeled with a, b, c, . . . , ` (12 total). Each “move” we can perform
on the cube (rotate a face) induces a permutation of the corners and edges, so that we can describe
such moves using cycle notation. (None the “center” pieces in the middle of each face ever move,
so we need not keep track of them.) For instance, the rotation U (for “upper”) of the top face by
90 clockwise if viewed from above is given by:
U = (1234)(abcd).
Then U 2 corresponds to rotating twice, U 3 is three times, and U 4 is the identity. Rotating each
other face by 90 in a similar way (clockwise if viewed the right way) gives more permutations,
commonly denoted by D, L, R, F, B for “down” (bottom face), “left”, “right”, “front”, and “back”.
The Rubik’s cube group is then (almost!) defined to be the group generated by these 6 rotations:
G = hU, D, L, R, F, Bi.
The point is that an element of G corresponds to a certain configuration of the cube: perform the
rotations dictated by the product expression for this element on the “trivial” identity configuration
(where everything is in the correct place) to obtain the desired configuration. This is mostly the
correct group but not quite, since we have made no mention of the orientation of each corner and
edge, meaning which color each side actually uses. (Of course, when solving the cube the colors are
important!) So, actually, G as defined above is only a quotient of the full Rubik’s cube group, where
we quotient out by the normal subgroup generated by all the configurations where each corner/edge
is in the correct location, only with the orientation of each possibly incorrect in relation to the trivial
configuration. (Try to convince yourself that this does give a normal subgroup! The effect of taking
such a quotient is to focus first only on the position of each corner/edge, not yet on what color
each side has.) The full Rubik’s cube group should take these orientations into account as well.
Nevertheless, the group G above already gives a group-theoretic approach towards studying
Rubik’s cubes. To “solve” a cube from an initial configuration is then equivalent to figuring out
how to express that configuration in terms of the generators U, D, L, R, F, B; if the configuration
can be written as g1 g2 · · · gk where each gi is one of these generators, to solve the cube then means
for perform the moves gk−1 · · · g2−1 g1−1 . The order of G can be found using the orbit-stabilizer
theorem for instance (not so straightforward though!), and much other information about G and
the full Rubik’s cube group is also known. For instance, the fact that from any initial configuration
it takes at most 20 moves to solve the cube is then a statement about how many generators it
takes to express a given configuration. Many modern algorithms for solving cubes come from
finding a minimal expression in terms of generators, and such considerations lead into the subject
of computational/combinatorial group theory. Good stuff!

Existence of cycle decompositions. Let us now clarify one property of permutations we had
deferred proving: the fact that any permutation σ ∈ Sn can be written as a product of disjoint

67
cycles, which is unique up to the order in which the cycles are written. This is in some sense obvious
if you simply compute the disjoint cycle expression for a given, explicit permutation (as we have
done numerous times before), but takes some thought to actually prove can always be done in a
way which avoids having explicit information about the permutation available. Given σ ∈ Sn , one
approach is to first consider the terms:

1, σ(1), σ 2 (1), . . . ∈ {1, 2, . . . , n}.

Since there are only n choices for what these σ k (1) values can be, there must be some overlap
so that σ k (1) = σ ` (1) for some k < `, and then σ `−k (1) = 1, which guarantees that the cycle
starting with 1 will in fact always “close up” on itself. Similarly for the cycle containing 2, or 3,
etc. The non-obvious part is to prove that disjoint cycles arise: in particular, if (1 σ(1) σ 2 (1) . . .)
is the cycle starting with 1 and m ∈ {1, 2, . . . , n} does not appear in this cycle, the claim is that
the cycle starting with m will be disjoint from the one above. This can be done directly via a
brute-force computation, which is not so bad at all, but then proving uniqueness of the disjoint
cycle decomposition takes more effort still.
But here is a cleaner approach which uses machinery we have built up. For σ ∈ Sn , consider
the action of hσi on {1, 2, . . . , n}. The orbits of this action are disjoint and have union equal to
all of {1, 2, . . . , n}. Since these orbits correspond to disjoint cycles (precisely because we are only
acting on {1, 2, . . . , n} by powers of σ), we have our claim. The uniqueness of the disjoint cycle
decomposition comes simply from the uniqueness of the orbits under the action above. (I think
this is clear enough, but the book goes into this in a more in-depth way if you are not convinced.)

Cayley’s Theorem. As we have seen in examples above, a group can be realized as a subgroup of
a permutation (symmetric) group via a faithful action. Every group has at least one such faithful
action—the action of the group on itself by left multiplication—so every group can be realized as
such a subgroup, which is the statement of Cayley’s Theorem:

Every group is isomorphic to a subgroup of a permutation group.

To be sure, the action of G on itself by left multiplication is faithful since g1 h = g2 h implies g1 = g2

by using the fact that h has an inverse. This action thus gives an injective homomorphism G → SG
(where SG is the group of permutations/bijections on G), and its image is the subgroup of SG to
which G is isomorphic. This applies to infinite groups just as well as finite ones, although in the
finite case |G| = n we are able to realize G as a subgroup of a more familiar symmetric group Sn .
Historically, groups were first studied in the 19th century exclusively as subgroups of permu-
tations groups. (Recall that groups arose via the problem of studying permutations of the roots
of a polynomial, so permutations groups and their subgroups were the only groups known at that
time.) The modern definition of a group as a set with a binary operation satisfying three properties
was not in widespread use until much later, so the real point of Cayley’s Theorem is to confirm
that this modern definition does agree with the original notion of what a “group” was. Cayley’s
Theorem will not be so crucial for us as stated, but rather is important because it emphasizes the
idea of thinking about abstract groups as groups of permutations, which will be crucial once we
realize that this can happen in many different ways, not just via the action of of G on itself by left
multiplication. For instance, the rotational symmetry group of a cube can in fact be realized as
a subgroup of S24 (24 is the order of the rotation group) as Cayley’s Theorem implies, but as we
have seen it can also be realized as a subgroup of S8 , or S6 , or finally S4 , and all such realizations
can shed light on some important aspect of that group.

68
Examples. Let us look at some basic examples. Under the action of Z/nZ on itself by left
multiplication (which is actually addition in this case), acting by 1 on each element does the
following:

1 + 0 = 1, 1 + 1 = 2, 1 + 2 = 3, . . . , 1 + (n − 2) = n − 1, 1 + (n − 1) = n = 0.

Thus, as a permutation, the action by 1 has the following effect:

0 7→ 1 7→ 2 7→ 3 7→ . . . 7→ n − 1 7→ 0,

so that 1 corresponds to the n-cycle (012 · · · n − 1) under the map Z/nZ → Sn used in Cayley’s
Theorem. (Here we are viewing Sn as describing permutations on the set {0, 1, . . . , n − 1}, but if
we use the usual symbols 1, 2, . . . , n instead and “relabel” 0 as “n”, then 1 ∈ Z/nZ corresponds to
the n-cycle (123 . . . n).) The action by 2 would be given by the permutation (012 . . . n − 1)2 , and
so on, so that Z/nZ is thus isomorphic to the cyclic subgroup h(012 . . . n − 1)i of Sn . (In the n = 6
for instance, acting by 2 has the following effect:

0 7→ 2 7→ 4 7→ 0, 1 7→ 3 7→ 5 7→ 1,

so that 2 gives the permutation (024)(135) ∈ S6 , or (246)(135) if we relabel 0 as 6 in order to use

the standard elements {1, 2, 3, 4, 5, 6} on which S6 acts.)

Other actions. Let us recall other standard actions which will be useful to consider. If H is a
subgroup of G, then G acts on the set of cosets G/H of H also by left multiplication. (Note that
this action is not faithful unless H is trivial, since otherwise g1 H = g2 H does not mean g1 = g2 .) In
particular, if [G : H] = k, this gives a homomorphism G → Sk , and considering different subgroups
of different indices gives multiple such maps which can be used to study G. For instance, the claim
from a recent homework problem that a subgroup whose index [G : H] = p is the smallest prime
dividing |G| comes from studying the kernel of the resulting map G → Sp .
G also acts on itself by conjugation, giving another homomorphism G → SG different from the
one in Cayley’s Theorem; in fact, in this case the image actually lies in the group of automorphisms
Aut(G), so that we have a map G → Aut(G). If H is normal in G, we can also act by G on H via
conjugation (normality is needed to guarantee that this is indeed an action on H), so get a map
G → Aut(H). It also makes sense to conjugate subgroups, so if X denotes the set of subgroups of
G, we have an action of G on X by g · H := gHg −1 and hence a map G → SX . All such actions
and maps into permutation groups will be useful in the coming weeks.

Lecture 20: Class Equation

Warm-Up 1. We find an explicit subgroup of S8 which is isomorphic to the quaternion group

Q8 , and we show that no such subgroup of S7 exists. First, we obtain a subgroup of S8 isomorphic
to Q8 via Cayley’s Theorem and the action of Q8 on itself by left multiplication. Let us label the
elements of Q8 using 1, 2, . . . , 8 as follows:

1 1, −1 2, i 3, −i 4, j 5, −j 6, k 7, −k 8.

Multiplication by −1 has the following effect:

1 7→ −1, −1 7→ 1, i 7→ −i, −i 7→ i, j 7→ −j, −j 7→ j, k 7→ −k, −k 7→ k,

69
so under the ordering above the permutation which corresponds to −1 is:

(12)(34)(56)(78).

Note this has order 2 in S8 , just as −1 does in Q8 . Left multiplication by i does the following:

1 7→ i 7→ −1 7→ −i 7→ 1, j 7→ k 7→ −j 7→ −k 7→ j,

so the corresponding permutation is

(1324)(5768).
(This has order 4 in S8 , just as i does in Q8 .) And so on, we can compute the permutations which
left multiplication by any element of Q8 induces to get:

1 = (1) −1 = (12)(34)(56)(78) i = (1324)(5768) −i = (1423)(5867)

j = (1526)(3847) −j = (1625)(3748) k = (1728)(3546) −k = (1827)(3645).

The subgroup of S8 containing these 8 elements is then one which is isomorphic to Q8 .

Now, a realization of Q8 as a subgroup of S7 would correspond to a faithful action of Q8 on
a 7-element set X = {1, 2, . . . , 7}. We are thus tasked with showing that no action of Q8 on this
X can be faithful, or equivalently that any no homomorphism Q8 → S7 can be injective. The key
realization is that any nontrivial subgroup of Q8 contains −1, since −1 is the square of each of
±i, ±j, ±k. Consider any action of Q8 on X. Since

8 = |Q8 | = |stabilizer||orbit|

and any orbit has at most 7 elements, each stabilizer must have more than 1 element and will thus
be nontrivial. (Said another way, for each a ∈ X the 8 expressions g1 · a, . . . , g8 · a, where g1 , . . . , g8
are the elements of Q8 , cannot all be distinct elements of X since X only has 7 elements, so some
gi · a = gj · a and then gj−1 gi is a nontrivial element of the stabilizer of a.)
Thus each stabilizer contains −1 ∈ Q8 , and then do does their intersection. But this intersection
is precisely the kernel of the induced map Q8 → S7 , so this kernel is nontrivial and hence this map
is not injective, and thus no subgroup of S7 is isomorphic to Q8 .

Warm-Up 2. We classify all groups of order 6, which we now have the tools needed to do so.
Suppose G has order 6. First, by Cauchy’s Theorem there exist elements x, y ∈ G of order 2 and
3 respectively, which generated subgroups hxi, hyi of orders 2 and 3. Since hyi has order 3, it has
index 36 = 2, so it is automatically normal in G since [G : hyi] = 2 is the smallest prime dividing
|G|. We now consider two cases, depending on whether or not hxi is also normal in G.
If hxi is normal, then since hxi ∩ hyi = 1 (this intersection is a subgroup of both hxi and hyi,
so it divides both of their orders), we have from a previous homework problem that the product
hxihyi is a subgroup of G, and is in fact isomorphic to hxi × hyi. Moreover, this product has the
following order:
|hxi||hyi| 2·3
|hxihyi| = = = 6,
|hxi ∩ hyi| 1
so in fact this product is all of G, and hence G = hxihyi ∼ = hxi × hyi ∼
= Z/2Z × Z/3Z ∼= Z/6Z.
If hxi is not normal, consider the action of G on the set G/hxi of cosets of hxi. (This set of
cosets is not a group since hxi is not normal, but it is certainly still a set on which G can act.) Since
|hxi| = 2, |G/hxi| = 3, so this is an action of G on a 3 element set and hence gives a homomorphism

G → S3 .

70
Any g in the kernel of this map acts trivially on each coset, so in particular it acts trivially on hxi:
ghxi = hxi. This requires that g ∈ hxi, so ker φ ≤ hxi. But hxi has order 2, so either ker φ = hxi or
ker φ = 1; the former is not possible since ker φ = hxi would force hxi to be normal in G, and we are
assuming it is not in this case, so we must have ker φ = 1. Thus φ is injective, so G is (isomorphic
to) a subgroup of S3 . Since |G| = 6 = |S3 |, we thus get G ∼ = S3 . Thus, there are precisely two
groups of order 6: Z/6Z and S3 .

Class equation. We now use the action of G on itself by conjugation to derive an equation which
expresses the order of G (when G is finite) in a useful way. Recall that the orbits of the action of
G on itself by conjugation are called the conjugacy classes of G; let us denote the conjugacy class
containing h ∈ G by Cl(h):
Cl(h) := {ghg −1 ∈ G | g ∈ G}.
Since these are orbits for an action, distinct conjugacy classes are disjoint, so that we can express
G as the disjoint union of these distinct classes:
G
G= Cl(hi )
h1 ,...,hr

where h1 , . . . , hr ∈ G are representatives of each conjugacy class. This then gives the equation
X
|G| = |Cl(hi )|.
h1 ,...,hr

But those conjugacy classes which have size 1 are precisely those which correspond to elements
of the center Z(G) of G:

|Cl(h)| = 1 ⇐⇒ ghg −1 = h for all g ∈ G ⇐⇒ h ∈ Z(G).

Thus we can rewrite the sum above by separating out those the sizes of those conjugacy classes
which have only one element, to get:
X
|G| = |Z(G)| + |Cl(hi )|
h1 ,...,hk

since there precisely |Z(G)|-many conjugacy classes of size 1, and where h1 , . . . , hk denote (after
possibly relabeling) those elements of G which give the distinct conjugacy classes of size larger than
1. The above equality is known as the class equation. By the orbit-stabilizer theorem, |Cl(hi )| =
[G : CG (hi )] where CG (hi ) is the centralizer of hi (which is the stabilizer for the conjugation action),
so the class equation can also be written as
X
|G| = |Z(G)| + [G : CG (hi )].
h1 ,...,hk

Note that each [G : CG (hi )] = |Cl(hi )| in fact divides the order of G by the orbit-stabilizer theorem,
which places restrictions on what this number can actually be.

Groups of prime-power order. As a first application of the class equation, we show that any
group of prime-power order has a nontrivial center. (That we can derive this combinatorially from
the class equation is quite amazing!) Suppose |G| = pn where p is prime. Since each

[G : CG (hi )] = |Cl(hi )|

71
in the class equation is larger than 1 (since those classes of size 1 are grouped into the |Z(G)| term
instead) and divides |G| = pn , we see that each such index/size must be divisible by p. But then p
divides the left side of X
|G| − |Cl(hi )| = |Z(G)|,
h1 ,...,hk

so it must divide |Z(G)| as well. But this order is certainly at least 1 because e ∈ Z(G), so we
conclude that |Z(G)| ≥ p. Thus Z(G) is nontrivial, and in fact has at least p elements.

Lecture 21: Conjugacy in Sn

Warm-Up 1. We classify all groups of order p2 , where p is prime. As a first step, we show that all
such groups must be abelian. Indeed, suppose |G| = p2 . Then the center Z(G) of G is non-trivial
by the consequence of the class equation we saw last time. Thus either |Z(G)| = p of |Z(G)| = p2 .
If |Z(G)| = p2 , then Z(G) = G and G is abelian. If |Z(G)| = p, then G/Z(G) has order p2 /p = p
and so is cyclic, which implies that G is abelian by a previous Warm-Up we did. (As we pointed
out back then, this then actually means that |Z(G)| = p is not possible, since if G is abelian then
G = Z(G) so Z(G) has order p2 . Cést la vie.)
Now, if there is an element in G of order p2 , then G is cyclic and isomorphic to Z/p2 Z. Otherwise
every non-identity element has order p. Pick two such elements x, y, each not in the cyclic generated
by the other. Then both hxi, hyi are normal in G since G is abelian, and hxihyi is a subgroup of
G. Since x, y are not powers of one another, hxi ∩ hyi is trivial (this intersection has order dividing
p, so if not trivial would have to equal both hxi = hyi), so |hxihyi| = |hxi||hyi| = p2 . Thus
G = hxihyi ∼
= hxi × hyi ∼
= Z/pZ × Z/pZ
in this case. Hence there are two groups of order p2 for p prime: Z/p2 Z and Z/pZ × Z/pZ.

Warm-Up 2. Suppose G has prime-power order pn . We show that G has a subgroup of each order
pi dividing pn . First, Z(G) is non-trivial, and of some prime-power order pk where 1 < k ≤ n.
Since Z(G) is abelian, a problem on a recent homework then shows that Z(G) has a subgroup of
any order pi dividing |Z(G)| = pk , so this produces a subgroup of G of any order pi for 1 ≤ i ≤ k.
So, suppose we now want a subgroup of order pi for k < i ≤ n. The quotient G/Z(G) has order
pn−k < pn , so by induction we may assume that our claim (about existence of subgroups of a given
order) is true for this group. In particular, G/Z(G) has a subgroup of order pi−k , which is of the
form H/Z(G) where H is a subgroup of G containing Z(G). But then Lagrange’s Theorem gives:
|H| = |H/Z(G)||Z(G)| = pi−k pk = pi ,
so H is our desired subgroup of G. We conclude that G has subgroups of all order pi for 1 ≤ i ≤ n.
(Groups of prime-power order will play a big role soon in the context of the Sylow Theorems.)

Conjugacy in Sn . We now wish to understand conjugacy classes in Sn , and then in An . But for
Sn we had a previous homework problem which gives us all the information we need: σ, τ ∈ Sn are
conjugate if and only if they have the same cycle type, meaning that they have the same number
of cycles of the same length appearing in their disjoint cycle decompositions. For instance, all
permutations of the form
(2-cycle)(3-cycle)
in Sn for n ≥ 5 will be conjugate to one another, and hence will all make-up one conjugacy class.
Thus, conjugacy classes in Sn are completely determined by the possible cycle types. In S5 for
example, there are thus ... conjugacy classes:

72
 • there one conjugacy class consisting of the identity alone (five 1-cycles);
• there is one conjugacy class consisting of transpositions (one 2-cycle and three 1-cycles);
• one class consists of products of two disjoint transpositions (two 2-cycles, one 1-cycle);
• one class consists of the 3-cycles (one 3-cycle, two 1-cycles);
• one class consists of products of disjoint 2- and 3-cycles; and
• one class consists of the 5-cycles

(Remark: the conjugacy classes of Sn thus correspond to the partitions of n, which are the ways of
expressing n as a sum of positive integers. The number 5 for instance has 6 partitions, corresponding
to the cycle types above: 1 + 1 + 1 + 1 + 1, 2 + 1 + 1 + 1, 2 + 2 + 1, 3 + 1 + 1, 3 + 2, 5. Partitions are
studied more carefully in a course on combinatorics, such as MATH 306.)

Counting permutations. We can easily determine the size of each conjugacy class above. For
instance, the number of 3-cycles (abc) in S5 can be found by making choices for each of a, b, c and
then taking into account over-counting: there are 5 choices for a, 4 for b, and 3 for c, but then we
have over-counted by a factor of 3 since (abc) gives the same permutation as (bca) and (cab). Thus
the number of 3-cycles in S5 is
5·4·3
= 20,
3
so the conjugacy class Cl((123)) of (123) (containing all 3-cycles) has size 20 in S5 .
Along the same lines, there are 5·4·3·2·1
5 = 24 cycles of length 5 in S5 , which comes from picking
each entry in (abcde) and then dividing by an over-counting factor of 5 since the permutation
(abcde) can also be described by the cyclic permutations of (abcde). Thus Cl((12345)) = 24. For
the conjugacy class of (12)(34), which consists of all products of disjoint transpositions, we count
the possibilities (ab)(cd) via
5·4·3·2
= 15.
2·2·2
where the over-counting factor of 2 · 2 · 2 comes from the fact that each 2-cycle used can by cyclically
permuted, but then also the fact that the order of the two transpositions themselves can be switched
since (ab)(cd) = (cd)(ab). Thus Cl((12)(34)) has 15 elements. The sizes of the remaining conjugacy
classes can be determined similarly.

Conjugacy in An . But now if we consider conjugacy in An , the answer is not as clean as it was
in Sn because conjugacy in Sn does not imply conjugacy in An . The issue is that the permutation
γ such that γσγ −1 = τ in Sn might not be even, as would be required in order to say that σ and τ
are still conjugate in An . For instance, in S4 we have:

(1234)(123)(1234)−1 = (234),

but (1234) is odd, so this equality does not imply that (123) and (234) are conjugate in A4 . Of
course, there might be some other permutation apart from (1234) which could also make (123)
conjugate to (234), but it turns out in this case that no such permutation will in fact be even, so
(123) and (234) are not conjugate in A4 . (This can be verified by a brute-force check.) Thus, (123)
and (234) will actually determine different conjugacy classes in A4 .
To determine the size of a conjugacy class in An , it is better to use the orbit-stabilizer theorem
and work with centralizers instead. For instance, the centralizer of a 5-cycle σ = (abcde) in A5
consists precisely of σ and its powers, so

CA5 ((abcde)) = h(abcde)i

73
has order 5. Thus |Cl(σ)| = [A5 : CA5 (σ)] = 60 5 = 12, which means that the 24 5-cycles in A5
(which make up one single conjugacy class in S5 ) actually split up into two distinct conjugacy
classes in A5 , each of size 12. (The same reasoning also works in S5 , with the same centralizer
here as in A5 , only that the order of the entire group S5 is 120 instead of |A5 | = 60, so we get
[S5 : CS5 (σ)] = 120
5 = 24 as the size of the conjugacy class, as expected.)
For 3-cycles in A5 however, the answer is indeed the same as in S5 since all 3-cycles are in fact
conjugate to one another within A5 . Actually, this is true for any n ≥ 5: for any (abc), pick σ ∈ Sn
such that σ(123)σ −1 = (abc); then if σ is even, this gives conjugacy in An too, while if σ is odd,
then σ(45) is even and realizes this conjugacy in An :

σ(45)(123)(σ(45))−1 = σ(45)(123)(45)σ −1 = σ(123)σ −1 = (abc).

(We use here that n ≥ 5 so that (45) makes sense, and the fact that (123) and (45) commute. This
fact about 3-cycles in An all being conjugate to one another is not true for n = 4, as shown by
the examples of (123) and (234) mentioned previously.) If we argue using centralizers, say in the
n = 5 case, the key point is that although the centralizer of σ = (123) in A5 is hσi and thus has
order 3, its centralizer in S5 also includes (45) (which is odd) and the products σ k (45). Thus this
centralizer has order 6 in S5 , so we get that the size of conjugacy class of (123) is:
120 60
[S5 : CS5 ((123))] = = 20 in S5 , and [A5 : CS5 ((123))] = = 20 in A5 ,
6 3
so that all 3-cycles do make up a single conjugacy class in A5 .
The remaining cycle type an element of A5 can have is as the product of two disjoint transposi-
tions, and here it also turns out that there is one single conjugacy class, just as there is in S5 . We
will save this argument for next time, after which we will have determined the sizes of all conjugacy
classes in A5 ; we will then use this to show that A5 is simple.

Lecture 22: Simplicity of An

Warm-Up. We determine the centralizer of (12)(34) in A5 , and then the size of conjugacy class
to which it belongs. In order for σ ∈ A5 to centralize (12)(34), we must have:

σ(12)(34)σ −1 = (σ(1) σ(2))(σ(3) σ(4)) = (12)(34).

(The first equality comes from the main computation carried out in the homework problem where
it was shown that conjugacy in Sn is determined by cycle type.) Thus σ must fix 5, and then there
are two possibilities:
• (σ(1) σ(2)) = (12) and (σ(3) σ(4)) = (34), so that σ must fix or transpose 1, 2, and fix or
transpose 3, 4. In S5 the possibilities are thus (1), (12), (34), and (12)(34), but in A5 only
σ = (1) and σ = (12)(34) work.
• σ(1) σ(2)) = (34) and (σ(3) σ(4)) = (12), so that ether σ(1) = 3, σ(2) = 4 or σ(1) = 4, σ(2) =
3. Thus σ = (13)(24) or (14)(23). (In S5 , (1324) and (1423) also work.)
Hence CA5 ((12)(34)) has order 4, so the conjugacy class containing (12)(34) has size
60
|Cl((12)(34))| = [A5 : CA5 ((12)(34))] = = 15.
4
Since there are 15 permutations in S5 with cycle type (2-cycle)(2-cycle) (as computed last time),
this means that they all remain conjugate to each other in A5 as well. (Note in S5 the centralizer

74
has order 8, so the conjugacy class has size [S5 : CS5 ((12)(34))] = 120
8 = 15, as expected. The drop
in a factor of 2 of the size of A5 vs S5 is cancelled out by the doubling in the size of the stabilizer.)
As a consequence of this and computations done last time, we now know there are 5 conjugacy
classes in A5 of sizes 1, 12, 12, 15, 20, corresponding to the identity, the two classes containing 5-
cycles, the class containing products of two disjoint transpositions, and the class containing 3-cycles.

A5 is simple. We can now give a proof that A5 is simple. The proof we give is not the only
possible proof, but is a nice one which avoids so many brute-force computations. We will outline
another brute-force proof afterwards. First, a general observation: if N is a normal subgroup of a
group G, then N is a union of conjugacy classes of G. Indeed, if h ∈ N , then ghg −1 ∈ N for all
g ∈ G by normality, so that the entirety of Cl(h) is contained in N . Thus N is the union of the
conjugacy classes Cl(h) ⊆ G with h ranging throughout N . (In other words, if a normal subgroup
intersects a conjugacy class of the larger group, then it must contain the entire class.)
Suppose N E A5 . Then |N | divides |A5 | = 60 by Lagrange’s Theorem, and since N is a union
of conjugacy classes of A5 , |N | must be a sum of some or all of 1, 12, 12, 15, 20 since these are all
the possible sizes of these conjugacy classes. One possibility is |N | = 1, in which case N is trivial.
Otherwise, N will contain the conjugacy class of the identity and at least one other class, so |N | > 1
is 1 plus at least one number among 12, 12, 15, 20. The only way such a resulting number |N | can
also divide 60 is for it to be 60 itself:

|N | = 1 + 12 + 12 + 15 + 20 = 60,

so we conclude that N = A5 in this case. Hence the only normal subgroups of A5 are 1 and A5 , so
A5 is simple as claimed.
Just to contrast this proof with what happens for A4 , which is not simple, we will note without
proof that the conjugacy classes of A4 have sizes: 1, 3, 3, 4. In this case, a nontrivial N E A4 can
have order |N | = 1 + 3 = 4 dividing |A4 | = 12 (without using all of 1, 3, 3, 4), highlighting the fact
that A4 does indeed have a normal subgroup of order 4

It’s all about the 3-cycles. Here is an outline of a different, more computational approach to
showing that A5 is simple. In general, note that if 1 6= N E A5 is normal, the existence of a single
3-cycle in N is enough to guarantee that N = A5 . Indeed, if N contains a 3-cycle, then it must
contain all 3-cycles since 3-cycles are all conjugate to one another in A5 and conjugates of things in
N are still in N by normality. But then since the 3-cycles generate A5 (from a homework problem),
we must have A5 = N as a result.
So, if we want to show that A5 is simple, we must show that any nontrivial N E A5 contains
some 3-cycle. Let σ ∈ N be different from the identity. The goal is to then—based on what the
cycle type of σ is—show that σ times some conjugate of σ is a 3-cycle, which is then in N by
normality and the fact that N is closed under multiplication. This can be done through some
brute-force computations: for instance, if σ = (abcde) is a 5-cycle, you can work out that

σ[(ab)(cd)σ[(ab)(cd)]−1 ]

is a 3-cycle; and if σ = (ab)(cd) is a product of disjoint transpositions, then

σ(abe)σ(abe)−1

will be a 3-cycle. In fact, a similar type of brute-force computation can be used to show that An
is simple for all n ≥ 5: take a non-identity σ in a nontrivial N E An and work out (depending on

75
the cycle type of σ) through some conjugations and multiplications that you can produce a 3-cycle
which will be in N . So, it is possible to prove that A5 and An (n ≥ 5) more generally is simple
without anything fancy, but certainly there will be nothing very enlightening about this brute-force
approach.

An is simple for n ≥ 6. Instead, we prove that An is simple for n ≥ 6 in a more clever way which
avoids as many computations. We essentially follow the book’s approach, although presented here
in a (hopefully) clearer way. (The book argues by contradiction, which is a bit confusing and
unnecessary if phrased slightly differently.) The one technical fact we will need is the following: for
any σ ∈ An , there exists a conjugate τ of σ in An which is different from σ but nevertheless has
the same effect as σ on some i ∈ {1, 2, . . . , n}, meaning σ(i) = τ (i) for some i ∈ {1, 2, . . . , n}. If
the disjoint cycle decomposition of σ contains a cycle (abc . . .) of length at least 3, then

τ = (cd)(ef )σ[(cd)(ef )]−1

(where d, e, f are different from a, b, c and each other) has this property since τ (a) = b = σ(a) and
σ 6= τ since τ (b) = d 6= σ(b); if the decomposition of σ contains only disjoint transpositions and
there are at least three (ab)(cd)(ef ), then

τ = (ab)(ce)σ[(ab)(ce)]−1

works since τ (a) = b = σ(a) but τ (e) = d 6= σ(e); and if σ = (ab)(cd), then

τ = (acb)σ(acb)−1

satisfies τ (a) = c 6= σ(a) and τ (e) = σ(e), where e is different from a, b, c, d.

So, for some n ≥ 6 denote An by G (for cleaner notation) suppose N is a nontrivial normal
subgroup of G. Pick a non-identity σ ∈ N . By the technical fact above there exists a conjugate τ
of σ in An different from σ such that σ(i) = τ (i) for some i ∈ {1, 2, . . . , n}. Then σ −1 τ (i) = i, so
σ −1 τ is a non-identity element of the stabilizer Gi . But τ is also N by normality, so σ −1 τ ∈ N ∩ Gi .
This intersection is thus a nontrivial normal subgroup of Gi (normality is straightforward to check).
But elements of Gi ≤ An can be viewed as permutations (still even!) of the n − 1 elements among
1, 2, . . . , n which are not i, so that Gi is in fact isomorphic to An−1 . By induction we may assume
that Gi is thus simple (the base case n = 5 was worked out previously), so that the intersection
N ∩ Gi above must actually equal Gi , which means that Gi ≤ N . But Gi contains a 3-cycle since
any 3-cycle (abc) where none of a, b, c are i will fix i, so we get that N contains a 3-cycle. Thus,
as explained before, N contains all 3-cycles of An and hence equals An since An is generated by
3-cycles. We therefore conclude that An is simple for n ≥ 6.

Sn is not solvable for n ≥ 5. And now, using the simplicity of An for n ≥ 5, we can show that Sn
is not solvable for n ≥ 5. (Just to hammer the point once again: this is the reason why no “quintic
formula” exists.) This was actually a problem from discussion section, but we will reproduce the
reason here. We use the fact, also covered in discussion, that for n ≥ 5 the alternating group An is
in fact the only nontrivial proper normal subgroup of Sn .
The chain of normal subgroups 1 E N1 E . . . E Nn−1 E Sn needed in the condition for
solvability can only be
1 E Sn or 1 E An E Sn
since An is simple and the only nontrivial proper normal subgroup of Sn . But in the first case
the quotient Sn /1 ∼
= Sn is not abelian, while in the second the first quotient An /1 ∼
= An is also
non-abelian. Hence no such chain with abelian quotients can exist, so Sn is not solvable.

76
Automorphisms of Sn . We will finish our current discussion of symmetric groups and their
properties with one curious fact. Recall that, for any group, conjugation by g ∈ G defines an
automorphism of G. These are called the inner automorphisms of G. (The group of inner auto-
morphisms of G in general is isomorphic to G/Z(G), which we will leave for you to see why.) So
we can ask whether any more automorphisms beyond these exist.
Here is the theorem: for n 6= 6, Aut(Sn ) consists of only the inner automorphisms, while for
n = 6, Aut(S6 ) contains non-inner automorphisms. It is strange at first that the n = 6 case should
be so different than every other n, but indeed it is. The proof of this fact is not something we will
cover, but you can read about it in various sources if interested. We will likely see, in the context
of the Sylow Theorems, how one might go about constructing the non-inner automorphism of S6 ,
although we will not give a complete construction.

Lecture 23: Sylow Theorems

Warm-Up. We show that for n ≥ 5, An is the only nontrivial, proper normal subgroup of Sn .
This was a problem from discussion section, but we reproduce it here since it is quite an important
result. In particular, it was the key step in the argument we gave last time to show that Sn is
not solvable for n ≥ 5. Suppose N E Sn is proper. Then N ∩ An is a subgroup of An , and is
moreover normal in An : for τ ∈ An and γ ∈ N ∩ An , τ γτ −1 ∈ N since N is normal in Sn , and
τ γτ −1 ∈ An since An is closed under multiplication. Thus since An is simple for n ≥ 5, we must
have N ∩ An = An or N ∩ A = 1. If N ∩ An = An , then An ≤ N , but since An already has index 2
and N is proper in Sn , this implies N = An . (Otherwise there would be a subgroup of Sn of index
between 1 and 2, which is nonsense.)
If N ∩ An = 1 and N 6= 1, then N must be of the form N = {1, τ } for some odd permutation τ
of order 2. (If N contained an odd permutation σ which was not of order 2, it would contain σ 2 6= 1,
which is even and hence N would intersect An non-trivially. If N contained two odd permutations
of order 2, it would contain their even product and hence again N would intersect An non-trivially.)
But then στ σ −1 ∈ {1, τ } for any σ ∈ Sn since N is normal, so that this product must in fact equal
τ . This means that any σ ∈ Sn commutes with τ , so that τ is in the center of Sn . But this center
is trivial (see the Discussion 3 Problems), so this is not possible and thus N must be trivial.

The Sylow Theorems. Now we come to some of the most fundamental results in finite group
theory: the Sylow Theorems. After Lagrange’s Theorem, the Sylow Theorems (which give a general
partial converse to Lagrange’s Theorem) provide the key tools we need to understand the structure
of finite groups in a general way, and are crucial to the problem of their classification. It is common
to present these results as a collection of three theorems:

Sylow 1 : Suppose |G| = pk m where p is prime and does not divide m, so that pk is the
largest power of p dividing |G|. Then G has a subgroup of order pk . (Such a subgroup
is called Sylow p-subgroup of G.)

Sylow 2 : Any two Sylow p-subgroups of G are conjugate to one another: if P, Q are
Sylow p-subgroups, then there exists g ∈ G such that gP g −1 = Q. Also, any subgroup
of G of prime-power order p` is contained in a Sylow p-subgroup.

Sylow 3 : The number np of Sylow p-subgroups is congruent to 1 mod p and divides the
factor m in |G| = pk m. Moreover, np = [G : NG (P )] where P is any Sylow p-subgroup.

77
Note that since a Sylow p-subgroup has prime-power order pk , it itself has subgroups of each order
pi for 1 ≤ i ≤ k by a Warm-Up we did previously, so G does as well. (In general, a group of
prime-power order pt is called a p-group.)
The first Sylow Theorem is thus an existence result, and produces subgroups we can work with.
The third Sylow Theorem places restrictions on how many such subgroups there can be, and the
second Sylow Theorem then tells us something about how these different subgroups are related to
one another. (Many, but not all, other sources transpose what we are calling Sylow 2 and Sylow
3. Everyone agrees, however, on which one is Sylow 1). We will use Sylp (G) to denote the set of
Sylow p-subgroups of G.

Groups of order 12. Before proving the Sylow Theorems, we give an example which starts to
illustrate their power and how to use effectively use them. Note how all the Sylow Theorems play
a role here. We seek to say as much as we can say about groups of order 12. We will not be able
to provide a complete classification yet, but we can already get pretty far.
Suppose G has order 12 = 22 · 3. Then G has Sylow 2-subgroups of order 22 = 4, and Sylow
3-subgroup of order 3. The number n2 of Sylow 2-subgroups then satisfies:

n2 ≡ 1 mod 2 and n2 | 3,

and the number n3 of Sylow 3-subgroups satifies:

n3 ≡ 1 mod 3 and n3 | 4.

This gives the following possibilities: n2 = 1 or 3, and n3 = 1 or 4. As a general observation,

note that np = 1 is equivalent to having a normal Sylow p-subgroup: if there is only one Sylow
p-subgroup P , then for any g ∈ G, gP g −1 must equal P (so P is normal) since gP g −1 is also a Sylow
p-subgroup; while if P is normal, then since any Sylow p-subgroup is conjugate to P , gP g −1 = P
gives all possible Sylow p-subgroups, so there is only one. Thus in the case where n2 = 1 and
n3 = 1, the Sylow 2-subgroup P and Sylow 3-subgroup Q ∼ = Z/3Z are each normal in G. Then
P Q is a subgroup of G, and since P ∩ Q = 1 because their orders are relatively prime, we have
|P Q| = |P ||Q| = 22 · 3 = 12. Hence in this case we have

G = PQ ∼
= P × Q (recall that this uses normality).

We previously classified the groups of order p2 , so that P ∼

= Z/4Z or P ∼= Z/2Z × Z/2Z, and thus
∼ ∼
we get G = Z/4Z × Z/3Z or G = Z/2Z × Z/2Z × Z/3Z.
Now consider the case where n3 = 4, so that there are four Sylow 3-subgroups of G. The action
of G on Syl3 (G)—a set with 4 elements—by conjugation (the conjugate of a Sylow p-subgroup
is still a Sylow p-subgroup) then produces a homomorphism φ : G → S4 . The kernel of this
homomorphism consists of those elements of G which conjugate each Sylow 3-subgroup to itself,
and thus is the intersection of the normalizers NG (P ) of the Sylow 3-subgroups:
\
ker φ = NG (P ).
P ∈Syl3 (G)

The third Sylow Theorem also gives n3 = [G : NG (P )] = |G|/|NG (P )|, so since n3 = 4 we get
|NG (P )| = 3. This means NG (P ) = P since P ≤ NG (P ). Thus the kernel above is the intersection
of the four Sylow 3-subgroups, and so is trivial since each Sylow 3-subgroup has order 3. Hence
G → S4 is injective, so that G ≤ S4 (or, more precisely, G is isomorphic to a subgroup of S4 ). But
the only subgroup of S4 of order 12 is A4 (S4 only has one subgroup of index 2 by a problem from

78
discussion), so G ∼= A4 in this case. (Here is an alternate argument: each Sylow 3-subgroup has
2 elements of order 3, and since any two Sylow 3-subgroups intersect trivially, this gives 4 · 2 = 8
distinct elements of order 3 in G. Thus G contains all eight 3-cycles of S4 , so it must be A4 since
the 3-cycles generate A4 .)
To summarize: if there is only one Sylow 2-subgroup and only one Sylow 3-subgroup, G is
abelian and isomorphic to Z/4Z × Z/3Z ∼ = Z/12Z or Z/2Z × Z/2Z × Z/3Z ∼ = Z/2Z × Z/6Z, while if
there are four Sylow 3-subgroups, G is isomorphic to A4 . This leaves the case where there is only
one Sylow 3-subgroup but three Sylow 2-subgroups. We will come back to this case later, to see
that there are only two possibilities left for what G can be. For now, we at least know that the
Sylow 3-subgroup must be normal in G (since there is only one). Both of the possibilities left will
arise as semidirect products of the Sylow 3-subgroup and one of the Sylow 2-subgroups.

Proof of Sylow 1. We will now give a proof of the first Sylow Theorem, and save the other two
for next time. We use the same proof as the book, but there are multiple other proofs available. All
essentially proceed by induction, and build from a subgroup of prime-order at the start (Cauchy’s
Theorem) up to a subgroup of maximal prime-power order.
Write |G| = pk m where p does not divide m. We consider two cases: p divides the order of the
center Z(G), and p does not divide the order of Z(G). If p divides |Z(G)|, pick using Cauchy’s
Theorem an element x ∈ Z(G) of order p. Since x is in the center, it commutes with all g ∈ G and
hence hxi is normal in G. Then the quotient G/hxi is a group of order pk−1 m. By induction, G/hxi
has a subgroup H/hxi of maximal prime-power order pk−1 , where H is a subgroup of G containing
hxi. This H then has order |H/hxi||hxi| = pk−1 p = pk , so that H is a Sylow p-subgroup of G.
If p does not divide |Z(G)|, consider the class equation
X
|G| = |Z(G)| + [G : CG (hi )]
hi

where the hi are representatives of the non-trivial conjugacy classes of G. Since p divides |G| but
not |Z(G)|, it cannot divide the sum of indices [G : CG (hi )], so there exists some hi such that p
does not divide [G : CG (hi )] specifically. But

|G| pk m
[G : CG (hi )] = = ,
|CG (hi )| |CG (hi )|

so in order for p to not divide this it must be the case that all powers of p in the numerator
of the fraction on the right also show up in the prime-factorization of the denominator. Thus
|CG (hi )| = pk s for some s, so by induction CG (hi ) has a Sylow p-subgroup H of order pk , which is
then also a Sylow p-subgroup of G.

Lecture 24: More on Sylow

Warm-Up 1. Suppose G is a simple group of order 168 = 23 · 3 · 7. We determine the number of

elements of G which have order 7. The key observation is that any element of order 7 generates a
subgroup of order 7, and so is contained in a Sylow 7-subgroup. The number n7 of such subgroups
satisfies
n7 ≡ 1 mod 7 and n7 | 23 · 3 = 24,
and thus the possibilities are n7 = 1 or n7 = 8. But G is simple so a Sylow 7-subgroup cannot be
normal, so we must have n7 = 8. (Recall that in general np = 1 if and only if the Sylow p-subgroup

79
is normal.) Now, any two Sylow 7-subgroups must intersect trivially since their intersection has
order dividing 7 and is not 7. This means that each contains 6 elements of order 7 not contained
in any other Sylow 7-subgroup, so we get 8 · 6 = 48 elements of order 7 overall.

Warm-Up 2. We show that no group of order 56 is simple. Since 56 = 23 · 7, a group G of order

56 has Sylow 2-subgroups of order 8 and Sylow 7-subgroups of order 7. The number of each satisfy:

n2 ≡ 1 mod 2 and divides 7 n7 ≡ 1 mod 7 and divides 8.

Thus the possibilities are n2 = 1, 7 and n7 = 1, 8. We claim that in fact at least one of these must
actually be 1, so that there will be either a normal Sylow 2-subgroup or a normal Sylow 7-subgroup,
and hence either way G is not simple.
Suppose n7 = 8. The Sylow 7-subgroups intersect trivially and each contain 6 elements of
order 7, so this gives 8 · 6 = 48 elements of order 7 in G. A single Sylow 2-subgroup gives another
8 elements of G (Sylow 2-subgroups and Sylow 7-subgroups intersect trivially because they have
relatively prime orders), so so far we get 48 + 8 = 56 elements in G. But this is the entire order of
G already, so there cannot be any other Sylow 2-subgroups since a second such group would give
at least one more elements in G. Thus in the case where n7 = 8 we must have n2 = 1, and hence
either n7 = 1 or n2 = 1, so G is not simple.

Sylows never normalize each other. We now move towards proving the remaining Sylow
Theorems. We will essentially follow the same proof as in the book, only we reorder some of the
arguments to make them (we think) easier to digest, at the expense of proving only parts of the
second and third Sylow Theorems at a time. Ultimately the proofs come down to looking at a
certain conjugation action of one Sylow subgroup and determining the sizes of its orbits.
We begin first with a lemma, which amounts to saying that one Sylow p-subgroup cannot never
normalize a different one. Proving this first is also the approach the book takes, although we will
give a somewhat different phrasing of this result than the book does. Here is our claim:

Suppose g ∈ G has prime-power order pi and P is a Sylow p-subgroup of G with

gP g −1 = P . Then g ∈ P .

So, if an element of prime-power order normalizes a Sylow p-subgroup, it must actually be in

that subgroup. As a consequence, if Q, P ∈ Sylp (G) are different Sylow p-subgroups, neither is
contained in the normalizer of the other, for if Q did normalize P , the fact that any g ∈ Q has
prime-power order (since Q has prime-power order) implies g ∈ P by the lemma, so that Q and P
would actually be the same.
Consider the subgroup hP, gi of G generated by P and g. Since g normalizes P (and certainly
elements of P do as well), we have that P is a normal subgroup of hP, gi, so that the quotient
hP, gi/P exists. Any element of hP, gi is a product of elements of P with powers of g:

p1 g k1 p2 g k2 p3 g k3 · · · pm g km

(note the first p1 or last g km terms could be identities), so that any such product becomes a power
of g alone in the quotient:

p1 g k1 p2 g k2 p3 g k3 · · · pm g km = eg k1 eg k2 eg k3 · · · eg km = g k1 +···+km in hP, gi/P.

Thus hP, gi/P is cyclic of some prime-power order p` dividing the order pi of g. By Lagrange’s
Theorem we then get |hP, gi| = pi |P | = pi+k . But P ≤ hP, gi has maximal prime-power order pk in

80
G since it is a Sylow p-subgroup, so that we must actually have P = hP, gi. (Otherwise hP, gi has
largest prime-power order than P in G.) But this means that g ∈ P as claimed.

Proof of first part of Sylow 3. We now prove that np ≡ 1 mod p. (Again, note we are structuring
the proofs a bit differently than the book does.) Pick a Sylow p-subgroup P and consider the action
of P on Sylp (G) by conjugation. (Elements of P conjugate Sylow p-subgroups to Sylow p-subgroups
since the orders remain unchanged.) This action breaks up Sylp (G) into the disjoint union of orbits:

Sylp (G) = (orbit 1) t . . . t (orbit k).

The size of each orbit divides |P | = pk , and so is either 1 or divisible by p. But there is precisely
one orbit of size 1: the orbit containing P has size 1 since all elements of P normalize P , and no
other orbit has size 1 since P does not normalize any other Sylow p-subgroup. (An orbit of size 1
occurs when gQg −1 = Q for all g ∈ P .) Thus every other orbit, apart from the one containing P ,
has size divisible by p. Thus:

|Sylp (G)| = 1 + (sum of things divisible by p),

so np := |Sylp (G)| ≡ 1 mod p as claimed. (The “sum of things divisible by p” goes away when
taking the result mod p.)

Proof of first part of Sylow 2. Now we prove that all Sylow p-subgroups are conjugate to one
another. Pick a Sylow p-subgroup P and let X denote the set of its conjugates in G:

X := {gP g −1 | g ∈ G}.

Each such conjugate is a Sylow p-subgroup of G, and we want to show X = Sylp (G) so that all
Sylow p-subgroups arise this way. P acts on X by conjugation, and we can break up X into a
disjoint union of orbits:
X = (orbit 1) t . . . t (orbit k).
The exact same reasoning as before implies that there is one orbit of size 1 (the one containing
eP e−1 = P ), and that the others have size divisible by p. Thus again we get |X| ≡ 1 mod p.
Now, if X was not all of Sylp (G), there would exist Q ∈ Sylp (G) not in X. This Q also acts on
X by conjugation, and we get:

X = (orbit 1) t . . . t (orbit t).

But now, there is no orbit of size 1: if Q normalizes an entire conjugate gP g −1 , then Q = gP g −1

by the technical lemma from before, so that Q would have been in X. Thus in this case all orbits
have sizes divisible by p, so |X| ≡ 0 mod p. This contradicts |X| ≡ 1 mod p from before, so no such
Q can exist. Hence X = Sylp (G), so all Sylow p-subgroups are conjugate to P and thus each other.

Proof of rest of Sylow 3. Fix P ∈ Sylp (G). Since all Sylow p-subgroups of G are conjugate to
P , the number np of Sylow p-subgroups is the number of such conjugates. But this is the size of
the orbit of P under the action of G on its subgroups by conjugation, so

np = [G : NG (P )]

by the orbit-stabilizer theorem. (The normalizer is the stabilizer for this action.) Moreover,

|G| pk m m
np = [G : NG (P )] = = k = .
|NG (P )| p ` `

81
where |NG (P )| = pk ` and we use the fact that P ≤ NG (P ) in order to say that |NG (P )| has at
least a factor of pk in its prime-factorization, and no larger p-power since pk is the largest power of
p dividing |G|. Thus np ` = m, so np divides m.

Proof of rest of Sylow 2. We finish our proof of the Sylow Theorems by proving that if H ≤ G
has prime-power order pi , then H is contained in some Sylow p-subgroup. The action of H on
Sylp (G) by conjugation (again) gives

Sylp (G) = (orbit 1) t . . . t (orbit k).

As before, each orbit has size a power of p, and since |Sylp (G)| ≡ 1 mod p, there must be at least
one orbit of size 1. (If all orbits had size larger than 1 we would have |Sylp (G)| ≡ 0 mod p.) But in
order to have an orbit of size 1 means that hP h−1 = P for all h ∈ H and some Sylow p-subgroup
P , which implies H ≤ P by the technical lemma from before.

Lecture 25: Applications of Sylow

Warm-Up 1. We classify groups of order |G| = pq where p < q are primes such that p does not
divide q − 1. Let P be a Sylow p-subgroup and Q a Sylow q-subgroup of G. Since Q has index p,
which is the smallest prime dividing |G|, Q is normal in G. (You can also see this by determining
that nq = 1.) The number of Sylow p-subgroups of G satisfies

np ≡ 1 mod p and divides q,

so we get np = 1 or np = q as possibilities. (Note q is prime.) But np = q is ruled out by the fact

that p does not divide q − 1, so that q is not congruent to 1 mod p, and hence np = 1 and thus P is
normal in G. Thus, since P and Q have trivial intersection (their orders are p and q respectively),
we get |P Q| = pq = |G|, and since P, Q are both normal in G we have:

G = PQ ∼
=P ×Q∼
= Z/pZ × Z/qZ ∼
= Z/pqZ.

Thus Z/pqZ is the only group with the given properties.

In particular, this applies to 15 = 3 · 5 since 3 does not divide 5 − 1, so the only group of order
15 is Z/15Z. This is then an example of a non-prime order (prime-order groups are always cyclic)
with the property that every group of that order must be cyclic. Similarly, every group of order
33 = 3 · 11 is cyclic, and there are many more examples of such orders.

Warm-Up 2. We show that every group G of order 30 contains Z/15Z as a subgroup. We can
actually give a quick answer to this based on a recent homework problem: since 30 = 2 · 15 where
15 is odd, a recent homework problem shows that G has a subgroup H of index 2, which then has
order 15 and is thus isomorphic to Z/15Z by the first Warm-Up.
Here is another method, which we look at only to clarify the book’s approach. (The book
makes this way more complicated than necessary.) Let P be a Sylow 3-subgroup and Q a Sylow
5-subgroup of G. Then we get
n3 = 1, 10 and n5 = 1, 6
as the possible numbers of such subgroups. But we cannot have both n3 = 10 and n6 = 6
simultaneously, since this would give 10 · 2 = 20 elements of order 3 (each Sylow 3-subgroup
contains 2 elements of order 3, and they all intersect trivially) and 6 · 4 = 24 elements of order
5 (each Sylow 5-subgroup contains 4 elements of order 5, and they all intersect trivially), which

82
already gives too many elements in G. Thus at least one of n3 or n5 is 1, so at least one of P or Q
is normal in G. Then P Q is a subgroup of G of order |P Q| = |P ||Q| = 15 (since P ∩ Q = 1), and
the first Warm-Up shows that P Q ∼ = Z/15Z.
Now, here is what the book says: since at least one of P or Q is normal in P Q, both P and
Q are characteristic subgroups of P Q, so both P and Q are normal in P Q, which implies that
PQ ∼ = P × Q is Z/15Z. We have not spoken about characteristic subgroups before, so here is a
definition: H is a characteristic subgroup of G if it is invariant under every automorphism of G,
i.e. if φ ∈ Aut(G), then φ(H) = H. (The fact that characteristic subgroups are always normal
comes from considering the automorphisms of G induced by conjugation by elements of G.) The
notion of a characteristic subgroup, however, is not we will really need going forward, and indeed I
think the book’s use of this concept here is confusing. In particular, the fact that here both P and
Q are characteristic in P Q is not a general type of phenomena for products of subgroups, and is
unique to the circumstances of this type of problem. The point is that in this case, even though P
is not the only Sylow 3-subgroup of G, it is the only Sylow 3-subgroup of P Q! Indeed, the number
np of Sylow 3-subgroups of P Q has to be congruent to 1 mod 3 and divide 15 (instead of 30 as
before), and n2 = 1 is now the only possibility. This is why both P and Q are normal in P Q in
this particular instance, so P Q ∼
= P × Q. I do not know what benefit the book gets from phrasing
this in terms of characteristic subgroups, but cést la vie.

Automorphisms of normal Sylows. Even if the book’s approach to the second Warm-Up above
is not necessarily the quickest way towards a proof, it does have the benefit of introducing the use
of automorphisms, in that case via the concept of a characteristic subgroup. This was not so crucial
in that specific problem, but now we outline the way in which automorphisms will be crucial in
applications of Sylow Theory in general.
Suppose G is a group with a unique (equivalently normal) Sylow p-subgroup P . The fact that
P is normal guarantees that conjugating P by elements of G still produces elements of P , so that
the map g : P → P induced by such a conjugation can be viewed as an automorphism of P . That
is, normality of P gives a homomorphism

G → Aut(P )

which sends g ∈ G to the automorphism which sends p ∈ P to gpg −1 ∈ P . The point is that
if we know something about the structure of this automorphism group Aut(P ), we can derive
information about the homomorphism G → Aut(P ) and thereby understand something more about
how elements of G relate to those of P . Even better: conjugating by elements of any subgroup
Q ≤ G (for instance, a Sylow q-subgroup) also gives a homomorphism Q → Aut(P ), which says
something about the commutativity or lack thereof between elements of Q and elements of P .

Example. Here is a first example of putting this idea into action. Suppose G is a group of order
231 = 3 · 7 · 11. We claim that every element of G of order 11 commutes with every other element
of G, and hence must be contained in the center Z(G). The structure of a certain automorphism
group here will be key.
Since n11 must divide 3 · 7 = 21 and is congruent to 1 mod 11, the only possibility is n11 = 1.
Thus there is a normal Sylow 11-subgroup of G; call it P ∼
= Z/11Z. Since P is normal, conjugation
by elements of P gives an automorphism of P , so we have a homomorphism

φ : G → Aut(P ).

Now, we previously determined the structure of the automophism group of a cyclic group, with
the result being Aut(Z/nZ) ∼
= (Z/nZ)× . (This comes from the fact that only those elements in

83
(Z/nZ)× can serve as generators of Z/nZ. Check earlier in these notes for more details.) Thus in
our case we have
Aut(P ) ∼= Aut(Z/11Z) ∼
= (Z/11Z)× ∼ = Z/10Z,
and so φ above maps G into a group of order 10. The image of φ must thus have order dividing
10 and dividing |G| = 231 (by the First Isomorphism Theorem), so since 10 and 231 are relatively
prime we get that φ(G) has order 1, and is thus trivial.
But saying that φ is trivial means that φ(g) is the identity automorphism for every g ∈ G, so
that the automorphism of P induced by conjugation by g is the identity:

gpg −1 = p for all p ∈ P.

This means that each p ∈ P commutes with all g ∈ G, so P ≤ Z(G). Every element of G of order
11 is contained in some Sylow 11-subgroup, so every such element is in Z(G) as claimed.

Another example. Let us consider one final example. Previously we considered those groups G
of order |G| = pq with p < q primes where p - q − 1 (G is cyclic in this case), and now we consider
the case where p does divide q − 1. Following the same logic as before, there is a unique normal
Sylow q-subgroup Q, and the number of Sylow p-subgroups is either np = 1 or np = q. In this case,
however, np = q is possible since q is congruent to 1 mod p because p | q − 1. If np = 1, then the
same reasoning as before applies and G will be cyclic.
Now consider the case where np = q, and let P be a Sylow p-subgroup of G. We show that in
this case there can only be at most p − 1 candidates for G, which will necessarily be non-abelian
since they have non-normal Sylow subgroups. (In fact, all of these “at most” p − 1 candidates
will be isomorphic to each other, so that there is actually only one group in this case, but we will
save this fact for later after we have discussed semidirect products.) As in the previous example,
normality of Q says that conjugating by elements of G gives an automorphism of Q, so in particular
we consider conjugating only by elements of P to get a homomorphism

φ : P → Aut(Q).

(To be clear, for x ∈ P , φ(x) is the automorphism of Q which sends y ∈ Q to xyx−1 ∈ Q.) Since
Q∼= Z/qZ since |Q| = q is prime, we have that

Aut(Q) ∼
= Aut(Z/qZ) ∼
= (Z/qZ)× ∼
= Z/(q − 1)Z.

The image φ(P ) of φ thus has order dividing q − 1 and |P | = p. Since p is prime, this gives the
possibilities |φ(P )| = 1 or p. (The latter is allowed since p does divide q − 1 in this example.) But
if |φ(P )| = 1, then φ is trivial, meaning that φ(x) is the identity automorphism for all x ∈ P :

xyx−1 = y for all y ∈ Q.

This would imply that elements of P commute with those of Q, so G = P Q (note |P Q| is still
pq = |G| in this case) would be abelian. This is not true in this case, so |φ(P )| = p instead.
Thus φ(P ) is cyclic (and φ induces an isomorphism between G and φ(G)), generated by any
element of Aut(Q) ∼ = Aut(Z/qZ) of order p. Now, to be clear, the isomorphism between Aut(Z/qZ)
and Aut(Q) sends the “multiplication by k” map (for k ∈ (Z/qZ)× ) to the “k-th power” map
y 7→ y k , which is a homomorphism of Q since Q is abelian. Hence, the possible generators of
φ(P ) ≤ Aut(Q) can be explicitly described as being the k-th power maps for those k ∈ (Z/qZ)× ∼ =
Z/(q − 1)Z of order p, of which there are p−1 many in total. (The elements of order p in Z/(q − 1)Z

84
are `(q − 1)/p for ` = 1, . . . , p − 1.) Pick generators x of P and y of Q. Then φ(x) ∈ φ(G) ≤ Aut(Q)
is one of these k-th power maps, meaning that

xyx−1 = y k for some k ∈ (Z/qZ)× of order p.

But this fully determines G, which can now be given in terms of generators and relations as:

G = hx, y | xp = 1 = y q , xyx−1 = y k i

for one of the values of k above. Since there are p − 1 choices for k, this at first glance gives at
most p − 1 possibilities for G as claimed. (As stated before, we will later show that in fact these
are all actually isomorphic to each other. Note that when p = 2, we can immediately see that we
only get one such group—the dihedral group—since p − 1 = 1 in this case: q − 1 ∈ (Z/qZ)× is the
only element of order 2, so the defining relation in the presentation above becomes xyx−1 = y q−1 ,
which is precisely the defining relation of D2q with x playing the role of s and y the role of r.)

Lecture 26: Semidirect Products

Warm-Up. We go back to considering groups G of order 12. Previously we argued that in the case
where the Sylow 2- and 3-subgroups are unique, hence normal, G is abelian and isomorphic to either
Z/12Z or Z/2Z × Z/6Z, while in the case where there are four Sylow 3-subgroups, G is isomorphic
to A4 . That leaves the case there are three Sylow 2-subgroups (so G is necessarily non-abelian)
and only one Sylow 3-subgroup. We claim that, for now, there are at most four possibilities for G.
(We will see later that there are actually only two: D12 and a group we have yet to describe.)
Let P be a Sylow 2-subgroup (of order 4) and Q the Sylow 3-subgroup (of order 3). Then Q is
normal in G, so conjugation by elements of P gives automorphism of Q:

φ : P → Aut(Q), φ(x) = conjugation by x for x ∈ P .

In order to obtain a non-abelian group, this map φ must be nontrivial since otherwise elements
of P would commute with elements of Q, in which case G = P Q ∼ = P × Q would be abelian.
We previously classified groups of order p for p prime, so that in this case P of order 4 = 22 is
2

isomorphic to either Z/4Z of Z/2Z × Z/2Z. Since Aut(Q) ∼ = Aut(Z/3Z) ∼ = (Z/3Z)× = {1, 2}, we
thus seek to classify nontrivial homomorphisms

φ : Z/4Z → (Z/3Z)× and φ : Z/2Z × Z/2Z → (Z/3Z)× .

In the first case, φ is determined by the value of φ(1). In order for φ to be nontrivial, φ(1)
should not be 1, so it must be φ(1) = 2. Thus we only get one possible group G in this case: if
x generates P and y generates Q, then φ(1) = 2 means that conjugation by x should act as the
“second-power” map, so that xyx−1 = y 2 and thus G = P Q has presentation

G = hx, y | x4 = 1 = y 3 , xyx−1 = y 2 i.

(This is actually not a group we have come across before!) In the case where P ∼
= Z/2Z × Z/2Z, a
homomorphism
φ : Z/2Z × Z/2Z → (Z/3Z)×
is determined by φ(1, 0) and φ(0, 1). Since (Z/3Z)× = {1, 2}, the non-trivial possibilities are:

φ(1, 0) = 1, φ(0, 1) = 2 φ(1, 0) = 2, φ(0, 1) = 1 φ(1, 0) = 2, φ(0, 1) = 2.

85
(Having φ(1, 0) = 1 = φ(0, 1) yields the trivial homomorphism, which gives G abelian.) Thus there
are at most three groups we get in this case, so four overall. (Again, we will see that the three we
get here are actually all isomorphic—it is D12 !—so there are actually only two distinct groups in
the case where n2 = 3 and n3 = 1.)

Rethinking products of subgroups. In many of the recent examples we have looked at (|G| = pq
and |G| = 12 for instance, and some others), we have come across the following setup: a group G
with a normal Sylow subgroup Q and some other subgroup P such that Q ∩ P = 1 and G = QP .
Let us consider this scenario in a bit more detail, since it is the prototype for the notion of a
semidirect product of groups, which is the final tool we need to carry out classifications.
If Q ∩ P = 1, then when writing an element of G = QP in the form yx with y ∈ Q and
x ∈ P , we see that there is a unique way of doing so: if yx = y 0 x0 are two such expressions, then
x(x0 )−1 = y −1 y 0 is in both P and Q, and hence must be 1, which implies x = x0 and y = y 0 . Thus to
get a handle on the entire structure of G = QP , we need only understand the group multiplication
better. In particular, if y1 x1 and y2 x2 are two elements of QP (with yi ∈ Q, xi ∈ P ), we seek to
write their product in the form required of elements of QP . Note that:

(y1 x1 )(y2 x2 ) = (yx1 y2 x−1

1 )(x1 x2 )

after inserting 1 = x−1

1 x1 in between y2 and x2 . But this is the form we want: since Q is normal,
x1 y2 x1 ∈ Q, so y1 x1 y2 x−1
−1
1 is indeed in Q and x1 x2 is in P . If we consider the action of P on Q
by conjugation, we can x1 y2 x−1
1 as x1 · y2 , so that the above product looks like

(y1 x1 )(y2 x2 ) = (y1 [x1 · y2 ])(x1 x2 ).

The point is that, as a set, we can think of QP as Q × P , since we can “separate” the elements
of Q from those of P due to the uniqueness of an expression yx with y ∈ Q, x ∈ P . Under
this identification QP = Q × P (not literally an equality), the multiplication above tells us how
to interpret the group operation on QP as an operation on Q × P instead: the Q-component is
y1 [x1 · y2 ] and the P -component is x1 x2 .

Semidirect Products. Thus, we make the following general definition. Let Q and P be groups
(not assumed a priori to be subgroups of any larger group) with an action of P on Q by automor-
phisms; equivalently a homomorphism φ : P → Aut(Q). The semidirect product determined by φ
is the group whose underlying set is Q × P , and whose group operation is given by

(y1 , x1 )(y2 , x2 ) := (y1 [x1 · y2 ], x1 x2 ).

That this does define a group structure is a straightforward (but non-trivial!) check, and was
actually carried out in the Discussion 2 Problems handout. (Of course, back then we had no
context for this construction. Also, back then we wrote the product as P × Q instead of Q × P ,
which requires a slight adjustment to the definition of the multiplication; the two constructions,
however, are easily seen to be isomorphic. The fact that P acts by automorphisms is crucial to
verifying the associativity of the multiplication above. The inverse of (y, x) is (x−1 · y −1 , x−1 ).)
The semidirect product is denoted by Q oφ P (it depends on the action φ), or simply Q o P if the
action is clear from context.
In the case of G = QP from before, we see thus see that with the action of P on Q being the
conjugation action, the resulting semidirect product Q o P is indeed isomorphic to QP . (After all,
this is what motivated the definition of a general semidirect product!) But in fact, this scenario
always holds, and all semidirect products arise in this way: set G = Q o P (recall that in the

86
general construction Q, P can be unrelated groups which are not assumed to be subgroups of an
already existing group), so that we can identity Q with the subgroup Q × {eP } and P with the
subgroup {eQ } × P . Then you can verify that Q is normal in G and Q ∩ P is trivial. Conjugating
an element of Q by an element of P in G = Q o P looks the following:

(eQ , x)(y, eP )(eQ , x)−1 = (x · y, x)(eQ , x−1 ) = (x · y, eP )

(so lo and behold Q ∼ = Q × eP is indeed normal!), which shows that the original action x · y
corresponds precisely to conjugation in the semidirect product, just as expected from the G = QP
case. Thus, thinking of semidirect products as taking place within an existing group G = QP with
a conjugation action or as Q o P with any action by automorphisms are equivalent. (Often the
terms inner and outer semidirect products are used to distinguish between these scenarios.)
Note also that the usual direct product Q × P is a special case: when P acts trivially on Q
(equivalently P → Aut(Q) is trivial), then

(y1 , x1 )(y2 , x2 ) = (y1 [x1 · y2 ], x1 x2 ) = (y1 y2 , x1 x2 )

is the usual group structure on Q × P . This corresponds precisely to what we have seen in G = QP
when both Q and P are normal, so that QP ∼ = Q × P.

Example. Consider the usual action of GLn (R) on Rn by matrix multiplication. If we consider Rn
as a group under vector addition, this is an action by automorphisms. (This just says that linear
transformations preserve addition.) Thus we get a semidirect product Rn o GLn (R) whose group
operation is:
(b, A)(c, B) = (b + Ac, AB).
This semidirect is called the n-dimensional general affine group, and is the group of so-called in-
vertible affine transformations of Rn , which are functions T : Rn → Rn of the form T (x) = Ax + b
for a fixed invertible matrix A and fixed vector b. (So, affine transformations are compositions of
linear transformations with translations. The defining data A, b of an affine transformation corre-
sponds precisely to (b, A) ∈ Rn × GLn (R).) The point is that composing two affine transformation
T (x) = Ax + b and S(x) = Bx + c gives:

T S(v) = T (Bx + c) = A(Bx + c) + b = ABx + [Ac + b],

where if we “isolate” the Rn -component b + Ac from the GLn (R)-component AB, we get precisely
the semidirect product element (b + Ac, AB). So, the semidirect group operation in this case
encodes composition of affine transformations.

Revisiting dihedral groups. Consider now the action of Z/2Z on Z/nZ where the non-identity
element of Z/2Z acts by inversion: 1 · x = −x, or in other words where Z/2Z → Aut(Z/nZ) sends
the generator 1 of the domain to the automorphism which send everything in Z/nZ to its inverse.
The semidirect product Z/nZ o Z/2Z has the group operation given explicitly by:

(a, b)(c, d) = (a + (−1)b c, b + d)

since (−1)b c = c when b = 0 and (−1)b c = −c when b = 1.

We claim that this is actually describing D2n ! This was shown back in the Discussion 2 Problems
where semidirect products were first introduced, but here is the idea anyway. Identify the generator
of Z/2Z with s ∈ D2n and the generator of Z/nZ with r ∈ D2n . Then we have

(0, s)(r, 0)(0, s)−1 = (r−1 , s)(0, s−1 ) = (r−1 , 0),

87
which says that srs−1 = r−1 . (Recall, after all, that the action Z/2Z → Aut(Z/nZ) should simply
correspond to conjugation in the semidirect product, so since we said that conjugation by s ∈ Z/2Z
should act as inversion, we should expect srs−1 = r−1 .) The elements of Z/nZ o Z/2Z, written
simply as elements in the product (Z/nZ)(Z/2Z) where Z/nZ and Z/2Z are viewed as subgroups
of Z/nZ o Z/2Z, are of the form ri sx where i = 1, . . . , n and x = 0, 1, which precisely correspond
to the elements of D2n . Using srs−1 = r−1 , and noting s−1 = s, we get the following presentation
for Z/nZ o Z/2Z:
hr, s | rn = 1 = s2 , srs = r−1 i,
which is indeed the standard presentation of D2n , so Z/nZ o Z/2Z ∼
= D2n as claimed.

Lecture 27: Classifying Groups

Warm-Up 1. We show that there are four homomorphisms Z/2Z → Aut(Z/3Z × Z/5Z), and that
they give rise to non-isomorphic semidirect products (Z/3Z × Z/5Z) o Z/2Z. (Honestly, this is
much more than a “Warm-Up”, since we will see in a bit that this amounts to classifying all groups
of order 30.) First, we claim that

Aut(Z/3Z × Z/5Z) ∼
= Aut(Z/3Z) × Aut(Z/5Z).

(More generally, Aut(Z/nZ × Z/mZ) = ∼ Aut(Z/nZ) × Aut(Z/mZ) when n and m are relatively
prime.) Given f ∈ Aut(Z/3Z × Z/5Z), the point is that f (1, 0) must actually be an element of
Z/3Z × 0 and f (0, 1) must be an element of 0 × Z/5Z. Indeed, (1, 0) has order 3 in Z/3Z × Z/5Z,
so f (0, 1) have have order 3 as well. But this rules out anything like

f (1, 0) = (a, nonzero),

since no nonzero element of Z/5Z has order dividing 3. Similarly, (0, 1) has order 5 in Z/3Z×Z/5Z,
so f (0, 1) must have order 5 as well, and hence must be of the form f (0, 1) = (0, b) since no nonzero
element of Z/3Z has order dividing 5. Thus f acting on Z/3Z ∼ = Z/3Z × 0 produces something in
Z/3Z, and similarly for f acting on Z/5Z ∼ = 0 × Z/5Z, so that the behavior of f can be “separated”
into its behavior on Z/3Z and its behavior on Z/5Z:

Aut(Z/3Z × Z/5Z) → Aut(Z/3Z) × Aut(Z/5Z) defined by f 7→ (f |Z/3Z , f |Z/5Z ),

where f |A denotes the restriction of f to the subgroup A, is an isomorphism. (In other words, an
automorphism of Z/3Z × Z/5Z does not “mix up” terms between the factors.)
Since Aut(Z/3Z) ∼ = (Z/3Z)× = {1, 2} and Aut(Z/5Z) ∼ = (Z/5Z)× = {1, 2, 3, 4}, we can now
get to work. A homomorphism φ : Z/2Z → Aut(Z/3Z × Z/5Z) ∼ = Aut(Z/3Z) × Aut(Z/5Z) is
determined by φ(1), which must be an element of order dividing 2. Both elements of Aut(Z/3Z)
have order dividing 2, but only 1, 4 ∈ Aut(Z/5Z) do (recall that by k ∈ Aut(Z/nZ) we mean the
map which is “multiplication by k”), so we get four possibilities:

φ(1) = (1, 1), (2, 1), (1, 4), (2, 4) ∈ Aut(Z/3Z) × Aut(Z/5Z).

To be even clearer: 1 ∈ Aut(Z/3Z) or Aut(Z/5Z) denotes the identity map id (multiply by 1) and
both 2 = −1 ∈ Aut(Z/3Z) and 4 = −1 ∈ Aut(Z/5Z) denote inversion inv (multiply by −1), so the
four possibilities are:
φ(1) = (id, id), (inv, id), (id, inv), (inv, inv).

88
The four semidirect product multiplications on (Z/3Z × Z/5Z) o Z/2Z are then explicitly given by:

(id, id) : ([a, b], x)([c, d], y) = ([a + c, b + d], x + y)

(inv, id) : ([a, b], x)([c, d], y) = ([a + (−1)x c, b + d], x + y)
(id, inv) : ([a, b], x)([c, d], y) = ([a + c, b + (−1)x d], x + y)
(inv, inv) : ([a, b], x)([c, d], y) = ([a + (−1)x c, b + (−1)x d], x + y)

The (−1)x term tells us whether or not we should invert. Note we can already tell that the
first case gives the abelian group Z/3Z × Z/5Z × Z/2Z; indeed, this is expected since in this case
Z/2Z → Aut(Z/3Z×Z/5Z) is trivial, so the semidirect product will be the ordinary direct product.
To show that the other three semidirect products (which are necessarily non-abelian) give
non-isomorphic groups, we determine the number of elements of order 2 in each. Using the multi-
plications above we can compute the following squares:

(inv, id) : ([a, b], x)2 = ([a + (−1)x a, 2b], 2x)

(id, inv) : ([a, b], x)2 = ([2a, b + (−1)x b], 2x)
(inv, inv) : ([a, b], x)2 = ([a + (−1)x a, b + (−1)b ], 2x)

Since x ∈ Z/2Z, we always have 2x = 0 so the third component tells us nothing. For the (inv, id)
case, 2b ∈ Z/5Z is zero if and only if b = 0. Also, if x = 0 here, then a + (−1)x a = 2a ∈ Z/3Z is
zero if and only if a = 0, so the only element whose square is zero when x = 0 is ([0, 0], 0), which
is just the identity. However, if x = 1, then a + (−1)x a = a − a = 0 for all a ∈ Z/3Z, so anything
of the form
([a, 0], 1) ∈ (Z/3Z × Z/5Z) o Z/2Z
has order 2. There are three choices for a, so this gives 3 elements of order 2 in the (inv, id) case.
In the (id, inv) case, similar reasoning shows that there are no elements of order 2 when x = 0,
and when x = 1 we have 2a = 0 ∈ Z/3Z if and only if a = 0 but b + (−1)x b = b − b = 0 for all
b ∈ Z/5Z, so anything of the form

([0, b], 1) ∈ (Z/3Z × Z/5Z) o Z/2Z

has order 2 in this case, and there are 5 such elements. Finally, in the (inv, inv) case, for x = 1 we
have that a + (−1)x a and b + (−1)x b are always zero, so

([a, b], 1) ∈ (Z/3Z × Z/5Z) o Z/2Z

all have order 2, and there are 15 such elements. Thus, we conclude that the four semidirect
products arising from Z/2Z → Aut(Z/3Z × Z/5Z) are all non-isomorphic. (The abelian case
Z/2Z × Z/3Z × Z/5Z only has one element of order 2.)

Warm-Up 2. But lest you believe that semidirect products can never be isomorphic, here is a
general scenario under which they are. Let φ : P → Aut(Q) be a homomorphism. If f ∈ Aut(P ),
then we get another action of P on Q via the homomorphism φ◦f : P → Aut(Q) which precomposes
with f , so that p · q under this new action is f (p) · q under the old. We claim that these two actions
of P on Q give isomorphic semidirect products.
Indeed, define a map Ψ : Q oφ◦f P → Q oφ P by Ψ(q, p) = (q, f (p)). This is bijective because
f : P → P is bijective. Moreover, we have:
using φ ◦ f
z }| {
Ψ((q1 , p1 )(q2 , p2 )) = Ψ(q1 [f (p1 ) · q2 ], p1 p2 )

89
 = (q1 [f (p1 ) · q2 ], f (p1 p2 ))
= (q1 [f (p1 ) · q2 ], f (p1 )f (p2 ))
= (q1 , f (p1 ))(q2 , f (p2 ))
= Ψ(q1 , p1 )Ψ(q2 , p2 ),

so Ψ preserves multiplication, and is thus an isomorphism. Hence Q oφ◦f P ∼

= Q oφ P as claimed.
We will make use of this fact in a bit.

Classifying groups. Now we come back to the problem of classifying groups, armed with the
notion of a semidirect product. Recall the following general type of setup: if a finite group G has a
normal Sylow subgroup Q (for some prime), and a subgroup P for which G = QP and Q ∩ P = 1
(equivalently |P | = |G|/|Q|), then G is a semidirect product of Q and P . Thus if we can classify
all such semidirect products, with maps P → Aut(Q), we can completely understand the structure
of G. (There is a limitation to how far this can get us, since a group in general might not have
any normal Sylow subgroups, so that not all groups can be described via semidirect products, but
it will get us quite far anyway. We will say something later about the problem of understanding
which groups are indeed obtainable as semidirect products.)

Groups of order 30. We classify all groups of order 30, which we now argue was done in the
first Warm-Up. Indeed, if |G| = 30, we know from a previous Warm-Up that G has Z/15Z as a
subgroup. This subgroup has index 2, and so is normal in G. Pick an element (using Cauchy’s
Theorem) x ∈ G of order 2. Then conjugation by x gives a homomorphism

hxi ∼
= Z/2Z → Aut(Z/15Z) ∼
= Aut(Z/3Z × Z/5Z) ∼
= Aut(Z/3Z) × Aut(Z/5Z).

Since Z/2Z∩Z/15Z = 1, G = (Z/2Z)(Z/15Z), so that G is a semidirect product of Z/15Z ∼ = Z/3Z×

Z/5Z and Z/2Z. The first Warm-Up then shows that there are four different possibilities for G: one
abelian group Z/2Z × Z/3Z × Z/5Z ∼ = Z/30Z, and three non-abelian groups (Z/3Z × Z/5Z) o Z/2Z,
characterized by having 3, 5, or 15 elements of order 2.
But actually, in this case we can give descriptions of these three non-abelian possibilities in
terms of better known groups. Indeed, D30 , Z/3Z × D10 , and Z/5Z × D6 all have order 30, are
non-abelian, and non-isomorphic, so these must be the three semidirect products we computed
before. To see which is which, we count elements of order 2:
• D30 has 15 elements of order 2, so D30 is the semidirect product in the φ(1) = (inv, inv) case,
• Z/3Z × D10 has 5 elements of order 2, so this is the φ(1) = (id, inv) case, and
• Z/5Z × D6 has 3 elements of order 2, so this is the φ(1) = (inv, id) case.
We can also determine this as follows. First, recall from last time that Z/nZ o Z/2Z with 1 ∈ Z/2Z
acting by inversion is isomorphic to D2n in general. In the φ(1) = (id, inv) case above, the point is
that this says 1 acts trivially on the Z/3Z factor of Z/3Z × Z/5Z, so that Z/3Z will then commute
with everything else and we can “break off” a Z/3Z factor from the semidirect product; but 1 acts
by inversion on the Z/5Z factor of Z/3Z × Z/5Z, so this portion of the semidirect product gives
Z/5Z o Z/2Z ∼ = D10 , which is why we get Z/3Z × D10 in this case. Similarly, in the φ(1) = (inv, id)
case, we act trivially on the Z/5Z factor—so this “breaks off”—but by inversion on the Z/3Z factor,
so we get
Z/5Z × (Z/3Z o Z/2Z) ∼ = Z/5Z × D6
in tis case. Finally, in the φ(1) = (inv, inv) case we act by inversion on both factors, which is
equivalent to acting by inversion on h(1, 1)i ∼
= Z/15Z, so we get Z/15Z o Z/2Z ∼
= D30 .

90
 To summarize, there are four groups of order 30: Z/30Z, D30 , Z/3Z × D10 , and Z/5Z × D6 .
The point is that the use of semidirect products gives us a way to prove that this list is complete.

Groups of order 12. Next we classify groups of order 12, for which we did the bulk of the work
before. Let us recall what we know, based on the numbers of Sylow subgroups:
• when n2 = 1 and n3 = 1, we have G ∼ = Z/12Z or G ∼= Z/2Z × Z/6Z,
• when n3 = 4, we have G ∼ = A4 (this was a good argument, go back and look at it!); and
• when n2 = 3 and n3 = 1, G is a semidirect product Q o P where Q is the normal Sylow
3-subgroup and P is a Sylow 2-subgroup. (We did not use the phrase “semidirect product” in
this case previously, but this is precisely what we had: G = QP , Q normal, and Q ∩ P = 1.)
Moreover, in this last case we also previously described all possible homomorphisms P → Aut(Q),
depending on whether P ∼ = Z/4Z or P ∼ = Z/2Z × Z/2Z, which are the only possibilities for |P | = 4:
∼
when P = Z/4Z, there is only one homomorphism 1 7→ 2 = inv (we previously called this is the
“second power” map, which is the same as inversion in Q ∼ = Z/3Z); and when P ∼= Z/2Z × Z/2Z,
we obtained three homomorphisms Z/2Z × Z/2Z → Aut(Z/3Z), and so at most three possible
semidirect products, which we will now show to all be isomorphic.
Recall that the possible nontrivial homomorphisms Z/2Z × Z/2Z → Aut(Z/3Z) are given on
generators by:

φ1 (1, 0) = inv, φ1 (0, 1) = id φ2 (1, 0) = id, φ2 (0, 1) = inv φ3 (1, 0) = inv, φ3 (0, 1) = inv.

(Again, we previously denoted id by 1 and inv by 2.) We now make use of the second Warm-Up:
if we can show that the second and third of these are obtained from the first by precomposing
with some automorphism of Z/2Z × Z/2Z, we will know that they give isomorphic semidirect
products. If we think of elements of Z/2Z × Z/2Z are column vectors [ xy ] where x, y ∈ Z/2Z, then
the automorphisms correspond precisely to invertible matrices with entries in Z/2Z:

Aut(Z/2Z × Z/2Z) ∼
= GL2 (Z/2Z).

(We will consider these types of matrices, and ones with other types of entries, in more detail next
quarter in the context of rings and modules.) Now, we have:
         
1 1 1 1 1 1 0 1
= and = .
0 1 0 1 0 1 1 0

Thus for f ∈ Aut(Z/2Z × Z/2Z) given by this specific matrix, we have:

φ1 (f (1, 0)) = φ1 (1, 1) = φ(1, 0)φ(0, 1) = id · inv = inv and φ1 (f (0, 1)) = φ1 (1, 0) = inv.

Hence φ1 ◦ f = φ3 , so Z/3Z oφ1 (Z/2Z × Z/2Z) ∼

= Z/3Z oφ3 (Z/2Z × Z/2Z). Since
         
0 1 1 0 0 1 0 1
= and = ,
1 1 0 1 1 1 1 1

for g the automorphism given by this matrix we have:

φ1 (g(1, 0)) = φ1 (0, 1) = id and φ1 (g(0, 1)) = φ1 (1, 1) = inv.

Thus φ1 ◦ g = φ2 , so Z/3Z oφ1 (Z/2Z × Z/2Z) ∼ = Z/3Z oφ2 (Z/2Z × Z/2Z). Hence, as claimed, all
semidirect products arising from nontrivial Z/2Z × Z/2Z → Aut(Z/3Z) are isomorphic.

91
 The group we get in this case is actually D12 , and here is one way of seeing this. Take the φ1
case. The second Z/2Z factor in the domain acts trivially, and so commutes with the Z/3Z factor
in Z/3Z o (Z/2Z × Z/2Z). If we take x ∈ Z/3Z, y ∈ Z/2Z, z ∈ Z/2Z to be generators of the three
factors, then we have:
xz = zx and yxy −1 = x−1 ,
where the second relation comes from having the first Z/2Z factor act by inversion on Z/3Z. But
these two together imply:

y(xz)y −1 = (yxy −1 )(yzy −1 ) = x−1 z −1 = (zx)−1 = (xz)−1 ,

where we have used the fact that yzy −1 = z since Z/2Z × Z/2Z is abelian, and that z −1 = z. The
element xz ∈ (Z/3Z)(Z/2Z) ∼ = Z/6Z has order 6 since |x| = 3 and |z| = 2, so since y ∈ Z/2Z acts
by inversion on this element xz ∈ Z/6Z, we can characterize Z/3Z o (Z/2Z × Z/2Z) instead as
Z/6Z o Z/2Z with the inversion action, which is precisely D12 .
The group we got in the Z/3Z o Z/4Z case, with 1 ∈ Z/4Z acting on Z/3Z by inversion, is
a brand new group we have not come across before, in the sense that it cannot be described in
terms of other groups we have seen: D2n , Sn , An , Z/nZ, etc. The group operation in this semidirect
product is explicitly given (in additive form) by:

(a, b)(c, d) = (a + (−1)b c, b + d).

Alternatively, in multiplicative form with Z/4Z = hxi and Z/3Z = hyi, we have xyx−1 = y −1
(x = 1 acts by inversion), so this group has presentation

Z/3Z o Z/4Z = hx, y | x4 = 1 = y 3 , xyx−1 = y −1 = y 2 i,

which had already derived previously. If you want an even more concrete description, it turns out
that we can realize this group as the subgroup of GL2 (C) generated by
   2πi/3 
0 i e 0
x= and y = ,
i 0 0 e−2πi/3

where e2π/3 = cos(2π/3) + i sin(2π/3) and e−2π/3 = cos(2π/3) − i sin(2π/3). You can check this by
verifying that these satisfy the relations in the presentation above.
To summarize, there are five groups of order 12: Z/12Z, Z/2Z×Z/6Z, A4 , D12 , and Z/3ZoZ/4Z
with 1 ∈ Z/4Z acting on Z/3Z by inversion.

Lecture 28: More Classifications

Warm-Up. We finish the classification of groups of order pq, where p < q are primes. Let us
refresh our memory of what we know:

• when p - q − 1, the only possibility is Z/qZ × Z/pZ ∼

= Z/qpZ;
• when p | q − 1, G = QP where Q is the normal Sylow q-subgroup and P a Sylow p-subgroup.
If P is also normal, then G ∼
=Q×P ∼ = Z/qZ × Z/pZ ∼= Z/qpZ is again cyclic.
• when p | q − 1 and P in the notation above is not normal, the structure of G is determined
by the relation xyx−1 = y k where P = hxi, Q = hyi, and k ∈ (Z/qZ)× has order p.

92
There are p − 1 elements of order p in (Z/qZ)× , giving at most p − 1 possibilities for G in the last
case, but we will now show that these are actually all isomorphic.
First, let us recast everything above in the language of semidirect products. In any case above we
have G = QP , where Q is the normal Sylow q-subgroup and P a Sylow p-subgroup (note Q∩P = 1),
so G is always isomorphic to a semidirect product Q o P for some action φ : P → Aut(Q).
In the case where p - q − 1, this action must be trivial since |φ(P )| divides both |P | = p and
| Aut(Q)| = |(Z/qZ)× | = q − 1, so that φ(P ) = 1. Hence the semidirect product is a direct product
in this case. If p | q − 1, φ(P ) is either trivial (giving again G ∼ = Q × P ), or a cyclic subgroup of
Aut(Q) of order p, so φ(x) (where x generates P ) is, as stated above, one of p − 1 elements of order
p in Aut(Q), and G ∼ = Q o P is the corresponding semidirect product.
Take k to be a generator of φ(P ) as a cyclic subgroup of order p in Aut(Q) ∼ = (Z/qZ)× . Then
the other elements of order p are k t for t = 1, . . . , p − 1. As an element of Aut(Z/qZ) ∼ = (Z/qZ)× ,
k denotes the “multiplication by k”, whereas in multiplicative form in Aut(Q) instead, this is the
“k-th power” map. This gives that the action of P = hxi on Q = hyi is determined by

xyx−1 = y k .

(This is just what we had said previously when we considered this problem.) For a different choice
of an element φ(x) of order p, we have φ(x) = k t (the “k t -th power” map) for some t = 1, . . . , p − 1,
so the semidirect product in this case is determined by the relation
t
xyx−1 = y k .

If φ1 denotes the first map P → Aut(Q) where φ1 (x) = k and φ2 the second where φ(x) = k t , then
for the automorphism f ∈ Aut(P ) given by the t-th power map, we have:

φ1 (f (x)) = φ1 (xt ) = φ1 (x)t = k t = φ2 (x),

so that φ2 = φ1 ◦ f . Thus by the second Warm-Up from last time, φ2 and φ1 determine isomorphic
semidirect products Q o P , so we conclude that all of the nontrivial choices for P → Aut(Q) give
the same nonabelian group G ∼ = Q o P . In terms of generators and relations
t
hx, y | xp = 1 = y q , xyx−1 = y k i v.s. hx, y | xp = 1 = y q , xyx−1 = y k i,

the isomorphism comes from using xt (also of order p) as a generator in the first case instead of x:
t
the defining relation xyx−1 = y k in the first then becomes the defining relation (xt )y(xt )−1 = y k
in the second, so the two presentations give the same group.
To summarize: when p - q −1 (p < q primes), the only group of order pq is Z/pqZ; whereas when
p | q − 1, there is one abelian group Z/pqZ of order pq and one non-abelian group Z/qZ o Z/pZ
where 1 ∈ Z/pZ acts on Z/qZ by multiplication by an element of order p in (Z/qZ)× . Note when
p = 2 that this non-abelian group Z/qZ o Z/2Z is precisely D2q , since here 1 ∈ Z/2Z acts on Z/qZ
by inversion because the only element of order 2 in (Z/qZ)× is q − 1 = −1.

Groups of order 18. Let us now classify all groups G of order 18. (We will have to make use
of some more advanced linear algebra here, namely linear algebra over what are called finite fields.
Indeed, one reason to look at this example is to get a sense for some of things we will be looking at
next quarter, such as the notion of a finite field and what it means to do linear algebra over them.)
Since 18 = 2 · 32 , there is a unique/normal Sylow 3-subgroup Q of order 9. (On the homework
you will classify groups of order pq 2 with p < q primes where p - q − 1. In this case, 2 does divide

93
3 − 1, so this example is not covered by the problem on the homework.) If P ∼
= Z/2Z is a Sylow
2-subgroup, then G = QP (Q ∩ P = 1 since they have relatively prime orders) so we get

G∼
= Q o Z/2Z for some action Z/2Z → Aut(Q).

Now, |Q| = 9 = 32 , so there are two possibilities for Q: Q ∼

= Z/9Z or Q ∼
= Z/3Z×Z/3Z. (Again,
we classified groups of prime-squared order earlier after discussing the class equation.) First we
consider the case Q ∼ = Z/9Z. But this is very similar to what we have done in other examples:
the only element of order 2 in Aut(Z/9Z) ∼ = (Z/9Z)× is 8 = −1, so Z/2Z → Aut(Z/9Z) is either
trivial or the action where 1 ∈ Z/2Z acts by inversion, and hence we get Z/2Z × Z/9Z ∼ = Z/18Z
∼
and Z/9Z o Z/2Z = D18 as the possibilities for G in this case.
Now consider Q ∼ = Z/3Z × Z/3Z, so that we need to understand homomorphisms

Z/2Z → Aut(Z/3Z × Z/3Z).

Thinking about elements of Z/3Z × Z/3Z as 2-dimensional column vectors with entries in Z/3Z
shows that we can obtain automorphisms via invertible 2 × 2 matrices with entries in Z/3Z, and
in fact all automorphisms arise in this way:

Aut(Z/3Z × Z/3Z) ∼
= GL2 (Z/3Z).

(As stated earlier, this—and some concepts which follow—is the kind of thing we will be able to
make clear next quarter.) So, the possible maps φ : Z/2Z → Aut(Z/3Z × Z/3Z) are determined
by φ(1) ∈ GL2 (Z/3Z) being a matrix of order dividing 2. But of course, different choices of such
matrices could give rise to isomorphic semidirect products, so we need to understand when precisely
that happens. Here is a fact: φ1 , φ2 give isomorphic semidirect products in this case if and only if
φ1 (1), φ2 (1) ∈ GL2 (Z/3Z) are conjugate matrices. We will not prove this here, but is something
which can be done using no more than we already know.
Given this, we must thus understand what conjugacy classes in GL2 (Z/3Z) look like. And here
is where linear algebra comes in: the conjugacy class of a matrix is characterized by its so-called
Jordan normal form. We will not develop nor define this concept at this point, but will do so next
quarter. (In the case of matrices with entries in R or C, Jordan normals forms are covered in Math
334, the course in abstract linear algebra.) We will say, however, that one class of Jordan normal
forms is given by the set of diagonal matrices (Jordan normal forms are related to the notion of
diagonalizability), which is all we need for our purposes. The fact is that any matrix in GL2 (Z/3Z)
of order dividing 2 is conjugate to exactly one of the following:
     
1 0 1 0 2 0
, , .
0 1 0 2 0 2

So, we get the possible semidirect products (Z/3Z × Z/3Z) o Z/2Z by considering the actions
determined by each of these. Note something special here: the action φ : Z/2Z → Aut(Z/3Z×Z/3Z)
determined by each of these has Z/2Z acting on each factor of Z/3Z × Z/3Z independently of one
another, due to the fact that these matrices are diagonal. In other words, taking each of these
times a vector [ ab ] in Z/3Z × Z/3Z does not  “mix up” the a and b coordinates. In general, a matrix
can mix these up, such as in [ 12 20 ] [ ab ] = a+2b
2a , so that a general action of Z/2Z on Z/3Z × Z/3Z
might not have Z/2Z acting independently on each factor, but the point is that all such actions will
determine the same semidirect product as one where Z/2Z does act independently on each factor.
For φ(1) = [ 10 01 ], the action φ : Z/2Z → Aut(Z/3Z × Z/3Z) is trivial, so G is abelian and
isomorphic to Z/3Z × Z/3Z × Z/2Z ∼ = Z/3Z × Z/6Z in this case. For φ(1) = [ 10 02 ], we have that

94
Z/2Z acts trivially on the first factor of Z/3Z × Z/3Z but by inversion (2 = −1 in Z/3Z) on the
second since:       
1 0 a a a
= = .
0 2 b 2b −b
The fact that Z/2Z acts trivially on the first factor says that the first Z/3Z factor commutes with
everything else (it already commutes with the second Z/3Z factor in Z/3Z × Z/3Z), so we can
“break off” a Z/3Z factor from all of G: G ∼ = Z/3Z × (something). We are left with Z/2Z acting
on the second Z/3Z by inversion, and this gives D6 , so overall:

G∼
= Z/3Z × (Z/3Z o Z/2Z) ∼
= Z/3Z × D6 .

That leaves the case where φ(1) = [ 20 02 ], where Z/2Z acts on each factor of Z/3Z × Z/3Z by
inversion. This semidirect product (Z/3Z × Z/3Z) o Z/2Z is not isomorphic to a group which can
be described in simpler terms using more familiar groups, so the semidirect product description is
the best we can do! Explicitly, the semidirect product multiplication is:

([a, b], x)([c, d], y) = ([a + (−1)x c, b + (−1)x d], x + y).

Written instead in multiplicative form with Z/2Z = hxi and Z/3Z × Z/3Z = hyi × hzi, this can be
characterized by the relations

xyx−1 = y −1 and xzx−1 = z −1 .

Thus this group has presentation hx, y, z | x2 = y 3 = z 3 = 1, yz = zy, xyx−1 = y −1 , xzx−1 = z −1 i.

In summary, there are five groups of order 18: Z/18Z, Z/3Z × Z/6Z, D18 , Z/3Z × D6 , and the
semidirect product (Z/3Z × Z/3Z) o Z/2Z where 1 ∈ Z/2Z acts on each factor by inversion.

Groups of order 8. We finish with the classification of groups of order 8. This requires—for the
final group at least—a different approach than what we’ve used so far since the final group we’ll
give is not obtainable as a semidirect product. Indeed, note that Sylow theory is completely useless
here since the only Sylow subgroup of a group G of order 8 = 23 is the entire group itself!
So, instead we argue by considering orders of elements. Any element of G will have order 1, 2, 4,
or 8. If there is an element of order 8, then G ∼= Z/8Z. If all non-identity elements have order 2,
then for three such elements x, y, z we have that the subgroups they generate have pairwise trivial
intersections, so
G = hxihyihzi ∼= hxi × hyi × hzi ∼
= Z/2Z × Z/2Z × Z/2Z.
That leaves the case where there is an element x of order 4, and no element of order 8. Note that
x2 then has order 2. If there is some other element y of order 2, then hxi ∩ hyi = 1 so G = hxihyi.
Since hxi is normal in G (it has index 2), in this case we do get G as a semidirect product hxi o hyi.
There are two homomorphisms Z/2Z → Aut(Z/4Z), where 1 ∈ Z/2Z acts trivially or by inversion,
so we get either G ∼
= Z/4Z × Z/2Z or G ∼ = Z/4Z o Z/2Z ∼ = D8 .
Thus finally we are left with the case where x2 (where |x| = 4 as above), is the only element of
order 2. Then all of the elements y not in hxi must have order 4. We then have x2 = y 2 and that
yxy −1 must be an element of order 4 in hxi (this subgroup is still normal), so we have

yxy −1 = x or yxy −1 = x3 .

If all such y satisfy the first relation, then x commutes with all elements not in hxi, which implies
that x is in the center of G. Then Z(G) has order 4 or 8, both of which imply G is abelian:

95
|Z(G)| = 8 immediately gives Z(G) = G, whereas |Z(G)| = 4 implies G/Z(G) ∼ = Z/2Z is cyclic,
which implies that G is abelian. But all the abelian cases were already worked out above (Z/8Z,
Z/4Z × Z/2Z, and Z/2Z × Z/2Z × Z/2Z; this probably depends on the fact that all finite abelian
groups are products of cyclic groups, which we will finally prove next time), so we are left with the
case where one y of order 4 satisfies yxy −1 = x3 . Our group then has presentation

G = hx, y | x4 = 1 = y 4 , yxy −1 = x3 i

(the other elements of order 4 are xy and (xy)−1 ), which in fact describes Q8 ! Indeed, we can take
x = i and y = j, and verify that i4 = 1 = y 4 and jij −1 = −i = i3 to see that this is the case. The
quaternion group Q8 is, in fact, a group which is not obtainable as a semidirect product, except
for something silly like Q8 o {e} or {e} o Q8 . (We will say something about why next time.)
To summarize, there are five groups of order 8: three Z/8Z, Z/4Z × Z/2Z, Z/2Z × Z/2Z × Z/2Z
abelian and two D8 , Q8 non-abelian.

Lecture 29: Finitely Generated Abelian

Warm-Up. We list all groups of order at most 15, which by now we have fully classified. Let us
record the list in the table below, where in the final column we recall some of the techniques used
in the classification:
order groups techniques
1 {e} :)
2 Z/2Z Lagrange
3 Z/3Z Lagrange
4 Z/4Z, Z/2Z × Z/2Z class equation
5 Z/5Z Lagrange
6 Z/6Z, S3 ∼= D6 G acting on cosets
7 Z/7Z Lagrange
8 Z/8Z, Z/4Z × Z/2Z, Z/2Z × Z/2Z × Z/2Z, D8 , Q8 semidirect, brute force
9 Z/9Z, Z/3Z × Z/3Z class equation
10 Z/10Z, D10 |G| = pq
11 Z/11Z Lagrange
12 Z/12Z, Z/6Z × Z/2Z, D12 , A4 , Z/3Z o Z/4Z G acting on Sylows
13 Z/13Z Lagrange
14 Z/14Z, D14 |G| = pq
15 Z/15Z |G| = pq

By “Lagrange” we mean the use of Lagrange’s Theorem to show that any group of prime order
is cyclic; by “class equation” we mean the use of the class equation to show that any group of
prime-power order has a nontrivial center, which then leads to the classification of groups of prime-
squared order; by “semidirect” we mean the use of semidirect products; by “|G| = pq” we mean the
classification of groups of order a product of two distinct primes; by “G acting on something” we
mean the action of G on cosets by left multiplication (in the order 6 case) or on Sylow subgroups
by conjugation (in the order 12 case), which leads to a map G → Sn ; and by “brute force” we
mean the brute force computation which constructs Q8 in the order 8 case. Of course, this list of
techniques is not exclusive, so that one technique might also show up elsewhere (such as the use of
semidirect products in the |G| = pq case or the order 12 case), and is only meant to highlight some
important ideas.

96
 We stop at order 15 because the order 16 case is where things get more complicated: there are 14
(!) non-isomorphic groups of order 16. Part of the reason why this is more complicated is because
the Sylow Theorems are of no use here: the only Sylow subgroup of a group of order 16 = 24 is
the entire group itself, so nothing can be said from Sylow theory alone. The same was true in the
order 8 = 23 case, and as there the order 16 case requires many more brute force computations. I
completely believe it is possible to work them all out by hand using the tools you have available
so far, but certainly it would involve a lot of work. After order 16 things calm down again for a
bit: 17, 19 and 23 are all prime; we did order 18 (and 30) previously; order 20 is on the homework,
and orders 21 and 22 are covered by |G| = pq. Then with order 24 we again get a large number
of possibilities—15 in this case—which involves a lot more work. And so on, we can keep going,
classifying many groups along the way, up to order 60 when we hit our first nonsolvable group A5 .
As a general comment, there exist many more tools we can use to classify even more groups, but
not ones we will discuss in this course.

What are semidirect products? Apart from the failure of the Sylow Theorems to give much
meaningful information (such as in the orders 8 and 16 cases), another main reason why the problem
of classifying groups becomes harder in general is the fact that not every group is obtainable as a
“non-silly” semidirect product. (By a “silly” semidirect product we mean something like
G∼
= G o {e} ∼
= {e} o G
with trivial actions of {e} on G or G on {e}, so by a “non-silly” one we mean not one of these, so
one which uses proper, nontrivial subgroups of G.) Let us say a bit more about this. If G ∼
= QoP
where Q E G and P ≤ Q with G = QP and Q ∩ P = 1, then the Second Isomorphism Theorem
gives (Q o P )/Q ∼= P/(Q ∩ P ) ∼= P . Thus, G ∼
= Q o P fits into the following sequence:
Q ,→ Q o P  P.
(Recall that the ,→ denotes an injecion, and the  a surjection.)
We briefly saw such sequences earlier, in the context of constructing finite groups from simple
groups via a composition series. In general, we referred to the problem of constructing G from such
a sequence
Q ,→ G  P
as an extension problem, and now we are saying that semidirect products provide one type of
solution to such a problem. But, semidirect products provide a special type of solution: if G in
the sequence above is in fact isomorphic to a semidirect product Q o P , then P ∼ = G/Q is itself
(isomorphic to) a subgroup of G. Said another way, in this case there is a map P → G whose
composition with the map G  P in the sequence is the identity on P (which forces P → G to
be injective), and via this map P → G we can realize P as a subgroup of G. When such a map
P → G exists, we say that the sequence above splits. The fact is that the converse is true: if the
sequence Q ,→ G  P splits, then G ∼ = Q o P for the action of P on Q by conjugation given by
∼
the realization of P = G/Q as a subgroup of G via the splitting.
Thus, G is a “non-silly” semidirect product if and only if it fits in the middle of a sequence
with nontrivial groups which splits. The problem of determining when a sequence splits is a hard
one in general, and indeed amounts to the same problem as that of building finite groups out of
simple groups. Consider now the example of Q8 . In this case, the only possible nontrivial normal
subgroups Q of G in the relevant sequence are Z/2Z or Z/4Z, so the only possible sequences are of
the form
Z/4Z ,→ Q8  Q8 /(Z/4Z) ∼
= Z/2Z or Z/2Z ,→ Q8  Q8 /(Z/2Z) ∼
= Z/2Z × Z/2Z.

97
(The Z/4Z in the first sequence is obtained by any of hii, hji, hki, and the Z/2Z ≤ Q8 in the second
sequence is h−1i. The quotient Q8 /h−1i ∼ = Z/2Z × Z/2Z is generated by i and j.) However, neither
of these sequences split: if the first did split, we would have Q8 ∼
= Z/4ZoZ/2Z, but the only possible
such semidirect products are Z/4Z × Z/2Z and D8 since the only actions Z/2Z → Aut(Z/4Z) are
the trivial one and the one where 1 acts by inversion; whereas if the second sequence split, there
would be a subgroup of Q8 isomorphic to Z/2Z × Z/2Z, which there is not. Thus Q8 cannot be
expressed a semidirect product of proper, nontrivial groups. In fact, Q8 was not the first group in
our list of groups of order at most 15 above which is not a semidirect product: Z/4Z is also not a
semidirect product, since the only semidirect product form which could work is Z/2Z o Z/2Z, and
the only such semidirect product is actually Z/2Z × Z/2Z.

Finitely generated abelian groups. And now we move towards our final result of the quarter:
the classification of finite abelian groups. We have made mention of this result in previously, where
the fact is that any finite abelian group is a direct product of cyclic groups. In fact, we will phrase
the result we want in the more general setting of finitely generated abelian groups, of which finite
abelian groups are a special case. (Actually, we will only prove the finite case at this point, and
will leave the general finitely generated case to next quarter where it will be a consequence of some
more general “linear algebraic” result.)
Recall that an abelian group G is finitely generated if there exist finitely many x1 , . . . , xn ∈ G
such that G = hx1 , . . . , xk i. Concretely, this means that anything in G is of the form n1 x1 +· · ·+nk xk
where each ni ∈ Z:
G = {n1 x1 + · · · + nk xk | ni ∈ Z}.
Here we use the convention (which will be in effect next quarter as well) that we use additive notation
when working with abelian groups (since addition in any way, shape, or form is always assumed
to be commutative, whereas multiplication is not), and that n1 x1 is assumed to be the result of
adding −x1 to itself |n1 | (absolute value) times when n1 is negative. (It is no coincidence that this
looks similar to “linear combination” or “span” notation from linear algebra, as we will see next
quarter.) Now, given these generators x1 , . . . , xk ∈ G we can construct a surjective homomorphism

φ : Zk  G

by sending (n1 , . . . , nk ) 7→ n1 x1 + · · · + nk xk . (Zk is the product of k-copies of Z.) Then by the

First Isomorphism Theorem we have:

G∼
= Zk / ker φ,

so that a finitely generated abelian group is a quotient of Zk for some k. Conversely, if there exists
a surjective map Zk → G—or equivalently if G is a quotient of some Zk —then G will be finitely
generated abelian with generators given by the images φ(ei ) where ei is the “vector” with 1 in the
i-th location and zeroes elsewhere.

Structure Theorem. The structure theorem for finitely generated abelian groups then classifies
finitely generated abelian groups G as being products of cyclic groups. But we can be more precise
about what this product structure looks like, and in fact there are two commonly used versions:

• G∼ k
= Zr × Z/pk11 Z × · · · × Z/pkt t Z for some r ≥ 0 and prime-powers pi i , and
• G∼ r
= Z × Z/d1 Z × · · · × Z/d` Z for some r ≥ 0 and di such that di | di+1 for each i.

98
To be clear, any G will admit both such expressions, and there is a uniqueness statement attached
to each. The first expression is called the primary factor decomposition of G, and the second is
called the invariant factor decomposition of G, with d1 , . . . , d` called the elementary divisors of
G. Note that the primes showing up in the primary factor decomposition are not assumed to be
distinct. The Zr is present (meaning r > 0) when G is infinite, so r = 0 if and only if G is finite.
We will prove this structure theorem (at least the existence part) next time in the case where
G is finite. The general version will be derived next quarter as a consequence of what’s called the
“structure theorem for finitely generated modules over a principal ideal domains” (we will see later
what all of these terms mean), of which finitely generated abelian groups are a key example. But
here is a hint at how the proof works in the general case: recall that G is finitely generated abelian
if and only if G ∼= Zk /K for some K ≤ Zk ; the proof then works by determining the structure of
any K ≤ Zk and then from here the structure of Zk /K.

Examples. Let us see some examples of the two forms of the structure theorem in action in the
case where G has order 72. Consider for instance Z/72Z, which is isomorphic to Z/8Z × Z/9Z. In
this case, Z/72Z is the invariant factor form (only one factor here) of this group, and Z/8Z × Z/9Z
is the primary factor form (8 and 9 are prime powers):

primary invariant
Z/8Z × Z/9Z Z/72Z.

Next, Z/2Z × Z/4Z × Z/9Z is isomorphic to Z/2Z × Z/36Z since Z/4Z × Z/9Z ∼
= Z/36Z:

primary invariant
Z/2Z × Z/4Z × Z/9Z Z/2Z × Z/36Z.

(Note 2, 4, 9 are all prime powers, and 2 | 36.) As another example, Z/2Z × Z/6Z × Z/6Z is
isomorphic to Z/2Z × Z/2Z × Z/2Z × Z/3Z × Z/3Z using Z/6Z ∼ = Z/2Z × Z/3Z twice, so:

primary invariant
Z/2Z × Z/2Z × Z/2Z × Z/3Z × Z/3Z Z/2Z × Z/6Z × Z/6Z.

The remaining abelian groups of order 72 are:

primary invariant
Z/2Z × Z/4Z × Z/3Z × Z/3Z Z/6Z × Z/12Z
Z/2Z × Z/2Z × Z/2Z × Z/9Z Z/2Z × Z/2Z × Z/18Z
Z/8Z × Z/3Z × Z/3Z Z/3Z × Z/24Z.

Rank and torsion. Let us note two more things. In the decompositions

G∼
= Zr × Z/pk11 Z × · · · × Z/pkt t Z ∼
= Zr × Z/d1 Z × · · · × Z/d` Z,

the exponent r ≥ 0 is unique. This number r is called the rank of G, and is in a sense analogous
to the notion of dimension in linear algebra. Rank zero corresponds to the finite case, so that the
Zr factor encodes the “infinite” part of G. (This is also called the free part of G, for reasons to be
made clear later.) Any element contained only in the free part has infinite order, so the elements
of finite order are encoded completely by the Z/pk11 Z × · · · × Z/pkt t Z ∼
= Z/d1 Z × · · · × Z/d` Z factors.
Previously on the homework we referred to the subgroup of G consisting of elements of finite
order as the torsion subgroup of G, so the point is that the “primary” and “invariant” factors
Z/pk11 Z × · · · × Z/pkt t Z ∼
= Z/d1 Z × · · · × Z/d` Z encode the torsion part of G.

99
 Thus, the structure theorem essentially says that a finitely generated abelian group can be
“broken” down into its free and torsion parts, and moreover the torsion part can be written in one
of two “nice” ways as a product of finite cyclic groups. We have been harping on this point quite a
bit, but will mention once more that this all fits in a general setting in module theory, where much
of the reasons for why these decompositions are useful will be made clear. (The answer, essentially,
is that they correspond to certain normal forms of matrices.)

Lecture 30: Back to Free Groups

Warm-Up. We show that any finite abelian group G is a product of groups of prime-power order.
This is actually fairly quick: suppose |G| = pk11 · · · pkmm is the prime factorization of the order of G,
and let A1 , . . . , Am be the Sylow subgroups of G, where |Ai | = pki i . Since G is abelian, each Ai is
normal in G, and the Ai have pairwise trivial intersections since they have relatively-prime orders.
Since |G| = pk11 · · · pkmm = |A1 | · · · |Am |, we thus have G = A1 . . . Am ∼
= A1 × · · · × Am as desired.

Finite abelian groups. We now show that any finite abelian group is a product of cyclic groups,
proving the finite case of the finitely generated abelian structure theorem from last time, or at least
the existence part of that theorem. (We save the uniqueness for the general theorem next quarter.)
By the Warm-Up, we are reduced to the case where |G| = pk has prime-power order. (If each Ai
above is a product of cyclic groups, so is G.) Let x ∈ G be an element of maximal order. Then hxi
is nontrivial and normal in G, and G/hxi is also finite and abelian. By induction, we may assume
that G/hxi is a product of cyclic groups:

G/hxi ∼
= hy1 i × · · · × hyt i
for some y1 , . . . , yt ∈ G. Denote the order yi in G/hxi by pmi , which divides the order of yi in G.
We claim that we can assume yi has order exactly pmi in G. To see this, note that since yi has
order pmi in G/hxi, we have mi
yip = xs in G
for some s ∈ N. If p does not divide s, then |xs | = |x| (the order of xs is |x|/gcd(s, |x|) and
gcd(s, |x|) = 1 since |x| is a power of p), so
mi |x|pmi
e = (xs )|x| = (yip )|x| = yi ,

which means that yi has order larger than |x|, contradicting the choice of x ∈ G. Thus p does
divide s, so s = apb for some a, b ∈ N. Then
mi b mi b b−mi mi
yip = xap =⇒ yip x−ap = (yi x−ap )p = e.
b−m
If yi x−ap i had order in G smaller than pmi , reversing the computation above shows there would
pn
be a smaller power of p such that yi is equal to a power of x, which contradicts the assumption
b−m
that pmi is the order of yi in G/hxi. Thus yi x−ap i ∈ hyi i has the same order pmi in G as yi has
in G/hxi, so by replacing yi with this element we may assume that yi has the same order as yi .
The subgroups hy1 i, . . . , hyt i have pairwise trivial intersection since the cosets their generators
determine have pairwise trivial intersection (since hy1 i · · · hyt i ∼= hy1 i × · · · × hyt i by the choice of
the yi ’s), and each hyi i has trivial intersection with hxi since yik ∈ hxi first when k is the order of
yi in G/hxi, which is the same as the order of yi in G meaning that yik = e. Thus, putting it all
together gives:

|G| = |hxi||G/hxi| = |hxi||hy1 i × · · · × hyt i|

100
 = |hxi||hy1 i| · · · |hyt i|
= |hxi||hy1 i| · · · |hyt i|,

which implies that G = hxihy1 i · · · hyt i ∼

= hxi × hy1 i × · · · × hyt i, so G is a product of cyclic
groups. (Phew! Since hxi and each hyi i has prime-power order, this gives the “primary factor
decomposition” form of G, but, as we saw in some examples last time, from this it is straightforward
to work out the “invariant factor decomposition” form of G.)

Back to finitely generated. So, any finite abelian group is indeed a product of cyclic groups.
But, as we saw, the proof is not for the faint of heart, and involves some brute-force computations
with quotients and orders. The proof in the general finitely generated case is even tougher, at least
until we have some better language to use in which to phrase it. As we saw last time, the proof
works by writing G as a quotient Z` /K of Z` for some subgroup K ≤ Zl , and then determining the
structure of K and of Z` /K. Here is the key fact which makes this work:
Any subgroup of a finitely generated abelian group is itself finitely generated!
In the case at hand, once we know K ≤ Z` is finitely generated, we can pick generates d1 , . . . , ds ∈ K
and then determine the structure of Z` /K using the First Isomorphism Theorem.
The fact that a subgroup of a finitely generated abelian group is itself finitely generated is
highly non-obvious and non-trivial, and indeed is not true in the non-abelian case in general. We
will give an example of how this can fail shortly, so the fact that we are working with abelian groups
here is absolutely crucial. Showing that a subgroup of a finitely generated abelian group is finitely
generated comes down to a linear algebraic type of computation, and indeed this fact should be
thought of as analogous to the fact in linear algebra that any subspace of a finite-dimensional space
is itself finite-dimensional,

Back to free groups. To give an example where a subgroup of a finitely generated group need not
be finitely generated, we return to the setting of free groups. Recall that the free group on a set S is
the group FS consisting of “words” of elements in S and their inverses with group operation given
by concatenation. The key point is that there are no non-trivial relations among the generators in
S. Let F2 = hx, yi be the free group on two generators and let H be the subgroup generated by
elements of the form y n xy −n :
H = hy n xy −n | n ∈ Zi.
Then in fact H is not finitely generated! To be clear, the generating elements y n xy −n are infinite in
number, and we can show that there are no nontrivial relations among these, so that this particular
generating set cannot be cut down to a finite generating set. However, this is still not enough to
show that H is not finitely generated, since we would have to also show that no finite subset of H—
possibly consisting of elements apart form the y n xy −n alone—can generate H either. This is hard
to prove using group theory alone, and amounts to a big, brute-force, and difficult computation.
Here is another seemingly “obvious” looking fact that is also highly non-trivial: any subgroup
of a free group is itself free. In this context, to say that a group is free just means that it has a
generating set whose elements satisfy no non-trivial relations among each other. More precisely,
H ≤ FS is free if there exists a set T such that H ∼ = FT . (You can view the fact that any subgroup
of a free is free as an analog of the fact that any vector space has a basis. The “no nontrivial
relations” is an analog of linear independence.) Producing the free generating set T for which
H∼ = FT is challenging to do in general, and again amounts to a big brute-force computation.
The upshot is that questions dealing with free groups are actually difficult to address directly
using group theory alone, so the fact that the analogous questions about free abelian groups (such as

101
the fact that any subgroup of a finitely generated one is still finitely generated) are easier to handle
really speaks to the benefit of having the abelian assumption. (A free abelian group is an abelian
group which satisfies a similar notion of “freeness”, namely that it has a generating set whose
elements satisfy no nontrivial relations, apart from the ones needed to say the group is abelian.
The “free abelian group on n generators” for instance is Zn .) The study of free groups in general
requires moving beyond group theory, and we will finish this quarter with a brief introduction to
some of the key tools in this area: fundamental groups.
But before moving on, we clarify that free groups are indeed important within group theory
itself, if for no other reason that free groups can be used to describe all groups. Just as a finitely
generated abelian group is a quotient of some Z` , it is true that any group whatsoever is a quotient
of a free group! Indeed, let G be a group and let FG be the free group generated by the elements
of G. (To be clear, even if elements of G satisfy some relations within G, they do not satisfy any
non-trivial relations in FG .) The map FG → G sending a “word” g1 g2 . . . gk to its actual value in
G as a product computed using the multiplication of G is a surjective homomorphism, so the First
Isomorphism Theorem gives:

G∼
= FG /K, where K is the kernel of FG → G.

(As a consequence, this shows that any group has a presentation: elements of G give the generators,
and elements of the kernel K the relations.) Thus, if we knew everything there was to know about
free groups (including K, which, as a subgroup of a free group, is free itself), we would everything
there was to know about all groups. Of course, knowing everything there is to know about free
groups is an intractable problem, but this at least highlights the role which free groups play in
group theory in general.

Fundamental groups. As mentioned above, fundamental groups are examples of groups which
on the one hand show up the study of free groups, but in fact are used much more broadly in
geometry and topology. They belong to the subject of algebraic topology, which deals with the use
of algebraic constructions (groups, rings, and modules—the latter two of which we will see next
quarter) to study geometric and topological spaces. The fundamental group is introduced in the
second quarter of the undergraduate topology course MATH 344-2, but here is a quick overview.
Given some space X (we will not clarify what we mean by “space” here, and just naively think of
a space as being some type of geometric object like a plane, sphere, or whatever), we can construct
a group out of the loops in X, which are paths which begin and end at the same point. We make
the declaration that two loops are thought of as being the “same” if one can be deformed into
the other. (We will not define what we mean by “deform” more formally, and again rely on pure
intuition. What we are really doing is defining an equivalence relation on the set of loops, and
considering its equivalence classes.) For instance, every loop in the plane R2 can be deformed to a
point, which we think of as being a constant loop:

Fix x0 ∈ X, and set π1 (X, x0 ) to be the set of (equivalence classes of) loops in this sense. Together
with the operation of “concatenation of paths”, where in a product γ1 γ2 we follow one loop and then

102
the other, π1 (X, x0 ) becomes a group called the fundamental group of X based at x0 . The identity
element is the constant loop at x0 , and the inverse of γ is visually the same loop but traversed in the
opposite direction. (So, all loops we consider are oriented with a particular direction. Showing that
going around γ one way followed by the other way does produce the identity requires the formal
definition of what “deform” means. Also, for nice enough spaces the basepoint x0 can essentially
be ignored since different choices give rise to isomorphic fundamental groups.)
Here are some examples. First, since all loops in R2 (no matter the point x0 we take) can be
deformed to the constant loop, we find that π1 (R2 , x0 ) = 1 is trivial. Next consider a circle S 1
(standard notation for the unit circle in R2 ):

The loop which goes around the circle once cannot be deformed to the constant loop, so this
fundamental group π1 (S 1 , p) is definitely non-trivial. It turns out that looping around the circle
twice, or three times, four times, etc produces all distinct elements in the fundamental group, so
that π1 (S 1 , p) is in fact isomorphic to Z! The integer which corresponds to a loop under this
isomorphism keeps track of how many times it wraps around the circle, with negative integers
corresponding to wrapping around in the opposite direction. It is true that all other loops can be
deformed into one of these, so that we get exactly Z as the fundamental group.
Finally, consider a figure eight:

As in the case of the circle, it turns out that the only thing which matters in the fundamental
group is how many times a loop wraps around each “leaf” of the figure eight. If we take a be
the loop which arounds the left leaf once and b the loop which wraps around the right leaf once,
then any element in the fundamental group looks like a word aabbbaabababbbababbab, so that the
fundamental group of the figure eight is the free group ha, bi on two generators! (Note that order
matters: wrapping around the left leaf and then the right is different than wrapping around the
right and the left, so ab 6= ba.) In general, if we “glue” n circles together at a single point, the
fundamental group of the resulting space will be the free group on n generators.

Geometric group theory. And thus we see free groups enter the realm of fundamental groups
and geometry/topology. This then leads to the fact that questions about free groups can be recast
as questions about certain spaces, such as ones obtained by gluing circles together. Subgroups of
fundamental groups correspond to what are called “covering spaces” of a space, so the claims that
any subgroup of a free groups is free, or that the subgroup hy n xy −n | n ∈ Zi of F2 = hx, yi is not
finitely generated, can be rephrased as questions about covering spaces instead. This allows for
the use of geometry/topology to study group theory, where there are new tools available and make
old, seemingly difficult problems much easier to handle. This is the topic of the subject known as
geometric group theory, and it is here that we will conclude our quarter. Thanks for reading!

103