A BSTRACT A LGEBRA

Romyar Sharifi
 Contents

Introduction 3

Part 1. A First Course 7

Chapter 1. Set theory 9
1.1. Sets and functions 9
1.2. Relations 11
1.3. Binary operations 15
Chapter 2. Group theory 21
2.1. Groups 21
2.2. Subgroups 26
2.3. Cyclic groups 28
2.4. Generators 31
2.5. Direct products 33
2.6. Groups of isometries 37
2.7. Symmetric groups 38
2.8. Homomorphisms 42
2.9. The alternating group 47
2.10. Cosets 48
2.11. Conjugation 52
2.12. Normal subgroups 54
2.13. Quotient groups 57
Chapter 3. Ring theory 61
3.1. Rings 61
3.2. Families of rings 64
3.3. Units 67
3.4. Integral domains 69
3.5. Ring homomorphisms 71
3.6. Subrings generated by elements 74
3.7. Ideals and quotient rings 76
3.8. Principal ideals and generators 81
3.9. Polynomial rings over fields 85
3.10. Maximal and prime ideals 88
3.11. Fields of fractions 91
3
4 CONTENTS

Chapter 4. Advanced group theory 97

4.1. Isomorphism theorems 97
4.2. Commutators and simple groups 100
4.3. Automorphism groups 103
4.4. Free abelian groups 106
4.5. Finitely generated abelian groups 109
4.6. Group actions on sets 113
4.7. Permutation representations 117
4.8. Burnside’s formula 118
4.9. p-groups 120
4.10. The Sylow theorems 122
4.11. Applications of Sylow theory 125
4.12. Simplicity of alternating groups 129
4.13. Free groups and presentations 131
Chapter 5. Advanced ring theory 137
5.1. Unique factorization domains 137
5.2. Polynomial rings over commutative rings 142
5.3. Irreducibility of polynomials 145
5.4. Euclidean domains 147
5.5. Vector spaces over fields 152
5.6. Modules over rings 157
5.7. Free modules and generators 161
5.8. Matrix representations 165
5.9. Associative algebras 167
Chapter 6. Field theory and Galois theory 171
6.1. Extension fields 171
6.2. Finite extensions 174
6.3. Composite fields 178
6.4. Constructible numbers 180
6.5. Finite fields 183
6.6. Cyclotomic fields 185
6.7. Field embeddings 188
6.8. Algebraically closed fields 190
6.9. Transcendental extensions 193
6.10. Separable extensions 195
6.11. Normal extensions 200
6.12. Galois extensions 202
6.13. Permutations of roots 209

Part 2. A Second Course 215

Chapter 7. Topics in group theory 217
 CONTENTS 5

7.1. Semidirect products 217

7.2. Composition series 219
7.3. Solvable groups 223
7.4. Nilpotent groups 225
7.5. Groups of order p3 228

Chapter 8. Topics in ring theory 231

8.1. Localization of commutative rings 231
8.2. Finitely generated and noetherian modules 237
8.3. Homomorphism groups 240
8.4. Tensor products 242
8.5. Exterior powers 250
8.6. Graded rings 252
8.7. Determinants 255
8.8. Torsion and rank 259
8.9. Modules over PIDs 261
8.10. Canonical forms 265

Chapter 9. Category theory 273

9.1. Categories 273
9.2. Functors 275
9.3. Natural transformations 278
9.4. Limits and colimits 281
9.5. Adjoint functors 287
9.6. Representable functors 290
9.7. Equalizers and images 292
9.8. Additive and abelian categories 295

Chapter 10. Topics in Galois theory 301

10.1. Norm and trace 301
10.2. Discriminants 304
10.3. Extensions by radicals 307
10.4. Linearly disjoint extensions 310
10.5. Normal bases 312
10.6. Profinite groups 313
10.7. Infinite Galois theory 319

Chapter 11. Commutative algebra 323

11.1. Localization of modules 323
11.2. Radicals of ideals 326
11.3. Primary decomposition 327
11.4. Integral extensions 332
11.5. Going up and going down 340
11.6. Hilbert’s Nullstellensatz 342
6 CONTENTS

11.7. Spectra of rings 345

11.8. Krull dimension 348
11.9. Dedekind domains 350
11.10. Discrete valuation rings 354
11.11. Ramification of primes 357
Chapter 12. Homological algebra 361
12.1. Exact sequences 361
12.2. The snake and five lemmas 363
12.3. Homology and cohomology 366
12.4. Projective and injective objects 369
12.5. Exact functors 373
12.6. Projective and injective resolutions 377
12.7. Derived functors 380
12.8. Tor and Ext 387
12.9. Group cohomology 393
12.10. Galois cohomology 400
Chapter 13. Representation theory 405
13.1. Semisimple modules 405
13.2. Representations of groups 410
13.3. Maschke’s theorem 413
13.4. Characters 416
13.5. Character tables 420
13.6. Induced representations 425
13.7. Applications to group theory 430
 Introduction

In mathematics, we often encounter objects that are sets with various operations that can be
performed on them. For instance, one may add and multiply integers, and one can do the same
with rational numbers, real numbers, and even complex (or imaginary) numbers. Or, given two
functions that input and output real numbers, we can compose them. We can add vectors, or
multiply them by scalars.
In abstract algebra, we attempt to provide lists of properties that common mathematical ob-
jects satisfy. Given such a list of properties, we impose them as “axioms”, and we study the
properties of objects that satisfy these axioms. The objects that we deal with most in the first part
of these notes are called groups, rings, and fields.
Groups, rings, and fields all sets with binary operations. A binary operation inputs two
elements of the set and outputs a third such element. Addition and multiplication of integers,
for instance, are binary operations, as is composition of real-valued functions of a real number.
Scalar multiplication of a vector in the plane is not however, since it starts not with two vectors,
but rather a scalar (i.e., a real number) and a vector.
We often require our binary operations to have certain properties like associativity or com-
mutativity. If we call our operation “?”, then associativity reads
(x ? y) ? z = x ? (y ? z),
and commutativity reads
x ? y = y ? x.
In imprecise terms, they tell you that the order in which you perform the operations doesn’t
matter. Though the most typically-encountered binary operations tend to be associative, many
are not commutative (i.e., are “noncommutative”). For instance, you may recall that the order of
composition of functions matters: e.g., sin(x2 ) and sin2 (x) are two different things.
The integers Z, the rational numbers Q, the real numbers R, and the complex numbers C
are all rings, as is the set of n by n matrices with entries in any of these. A ring is a set with
two binary operations called addition and multiplication. In order to be a ring, we require asso-
ciativity of both operations, commutativity of addition, and distributivity of the two operations.
Distributivity can be expressed as follows:
(x + y) · (z + w) = x · z + x · w + y · z + y · w.
Every ring must have an element called “0” (satisfying 0 + x = x for any x) and, for every number
x, there should be another −x which when added to x, gives you 0. Typically, a ring also has an
element called “1”, which satisfies 1 · x = x = x · 1. For instance, in any ring of n by n matrices,
the element “1” is actually the identity matrix.
7
8 INTRODUCTION

You may recall that multiplication of square matrices is noncommutative (if they are at least
2 by 2 in size). For instance, we have
           
1 1 1 0 1 −1 1 1 1 0 1 1
· = 6= = · .
0 1 0 −1 0 −1 0 −1 0 −1 0 1
On the other hand, the multiplications in Z, Q, R, and C are all commutative. A ring in which
multiplication is a commutative binary operation is a called a commutative ring.
Once we have rings, fields are simple to describe. Fields are commutative rings with one
extra property. That is, a field has inverses under multiplication: if x is in the field and isn’t 0,
then there must be an element x−1 = 1/x as well, and it satisfies x · x−1 = 1. In particular, Q, R,
and C are fields as well as rings, but Z is not a field. In a field, fractions add and multiply in the
familiar way:
x z xw + yz x z xz
+ = and · = .
y w yw y w yw
Some rings have nonzero elements x and y with product xy equal to 0. These are called
zero-divisors. For instance,      
1 0 0 0 0 0
· = ,
0 0 0 1 0 0
and so we can have that the product of two nonzero matrices is the zero matrix. If a commutative
ring has no zero divisors, then we can construct its field of fractions artificially. Its elements
consisting of elements denoted x/y, where x and y are in the original rings. The field of fractions
of Z is Q, and here we have our first example of a construction that is well-known for the simplest
ring of all, the integers, but can be performed more generally (for instance to polynomials),
starting from the axioms of a ring and a few extra properties.
Groups may seem a bit less familiar, but they are also in a sense simpler. Groups have only
one binary operation. Call it whatever you like: addition, multiplication, or just “?”. A group
and its binary operation ? must satisfy just three properties: associativity of ?, the existence of
an identity element e, and the existence of inverses. The identity element e is like the number 1
is under multiplication, or like 0 is under addition, in the rings that are familiar to us. It satisfies
e?x = x = x?e
for all x in the group. The inverse of an element x is normally denoted x−1 , but it is written −x if
our operation is addition. It satisfies
x−1 ? x = e = x ? x−1 .
In particular, rings are groups if we forget about the multiplication and just consider the operation
of addition. Fields are groups under multiplication if we throw out 0.
Many less familiar but interesting mathematical objects are groups. The rotations of a cir-
cle form a group under composition (following one rotation by another), and the permutations
(switches of positions) of five balls between five slots are a group under composition as well. The
n by n real matrices with nonzero determinant form a group under multiplication too. The set of
“moves” of a Rubik’s cube (compositions of rotations of sides by 90 degree multiples) form a
 INTRODUCTION 9

group too: a very complicated one, in fact. So, groups are in some sense a less refined but much
broader class of objects than the rings, with more exotic members.
In our examples, some of the groups have finitely many elements and hence are known as
finite groups. Here’s an interesting property of every finite group. Suppose that a finite group G
has n elements, and let x be one of them. Then xn , which is x ? x ? · · · ? x with x appearing n times,
is the identity element e. For instance, if I permute the position of 5 balls in five slots in a certain
manner, over and over, the balls will wind up in the position they started after 120 steps, since
that is the order of the group. In fact, this exaggerates the number of repetitions needed: the balls
end up at the starting point in six or fewer. The same goes with the Rubik’s cube: repeat the
same sequence of moves enough times, and, if you have enough patience (meaning watch out for
carpal tunnel syndrome), you will end back up where you started. This is something that, a priori,
may not seem obvious at all. Yet, this property of finite groups is a very general phenomenon,
derived solely from the group axioms.
Hopefully this encourages you to believe that abstract algebra may be of serious use both in-
side and outside mathematics, and indeed, it is so, in addition to being a fascinating and beautiful
theory in its own right for those so inclined. In the next chapter, we begin our study of abstract
algebra at a much more leisurely pace.
 Part 1

A First Course
 CHAPTER 1

Set theory

1.1. Sets and functions

In these notes, we assume some basic notions from set theory, for which we give only the
briefest of reviews. We won’t attempt to define a set formally here. Instead, we simply make
some remarks about them. Vaguely, a set is a collection of objects. Not every collection of objects
is a set: the “collection” of all sets is not a set. On the other hand, most reasonable collections of
objects are sets: the integers, the real numbers, the movies in your DVD collection (seemingly, a
soon-to-be dated notion), those are sets.
Sets consist of elements. If X is a set, we write x ∈ X to mean that x is an element of X (or “x
is in X”). Similarly, x ∈/ X means that x is not an element of X (which only really makes sense if
both x and the elements of X are elements of some common larger set so they can be compared.)
E XAMPLES 1.1.1.
a. The empty set ∅ is the set with no elements.
b. The set consisting of elements called 1, 2, and 3 is denoted {1, 2, 3}, and this notation
extends to any finite collection of objects.
c. The set {1, 2, 3, . . .} of positive integers is again a set.
d. The real numbers R form a set.
Any collection of elements of a set X is called a subset of X and is a set itself. We write
Y ⊆ X to mean that Y is a subset of X. If Y and Z are different subsets of X, then we write Y 6= Z
and we say that Y and Z are distinct subsets.
A property P that only some elements of X satisfy allows us to specify a subset of X consisting
of elements of X that satisfy P, which we denote in set-theoretic notation by
{x ∈ X | x satisfies P},
or just {x | x satisfies P} if X is understood.
E XAMPLE 1.1.2. The subset {n ∈ Z | 2 divides n} of Z is the set of even integers.
D EFINITION 1.1.3. Let X be a set and Y be a subset of X. Then X −Y denotes the complement
of Y in X, which is defined as
X −Y = {x ∈ X | x ∈ / Y }.
If Y is a subset of X that is not X itself, then it is called a proper subset, and we write Y ⊂ X
(or Y ( X). Given two subsets Y and Z of a larger set X, we can form their union Y ∪ Z and their
intersection Y ∩ Z, which are also subsets of X.
13
14 1. SET THEORY

D EFINITION 1.1.4. Given sets X and Y , the direct product X ×Y is the set of pairs (x, y) with
x ∈ X and y ∈ Y .
In set-theoretic notation, we may write this as
X ×Y = {(x, y) | x ∈ X, y ∈ Y }.
D EFINITION 1.1.5. A function f : X → Y from a set X to a set Y is a rule that to each x ∈ X
associates a single element f (x) ∈ Y , known as the value of f at x.
N OTATION 1.1.6. We sometimes refer to a function as a map, and we sometimes write f : x 7→
y to indicate that f (x) = y, or in other words that f maps (or sends) x to y.
We can, of course, compose functions, as in the following definition.
D EFINITION 1.1.7. Let X, Y , and Z be sets and f : X → Y and g : Y → Z functions. The
composition (or composite function) g◦ f : X → Z of g with f is the function such that (g◦ f )(x) =
g( f (x)) for all x ∈ X.
D EFINITION 1.1.8. Let f : X → Y be a function.
a. The function f is one-to-one (or injective, or an injection) if for every x, y ∈ X such that
f (x) = f (y), one has x = y.
b. The function f is onto (or surjective, or a surjection) if for every y ∈ Y , there exists an
x ∈ X such that f (x) = y.
c. The function f a one-to-one correspondence (or bijective, or a bijection) if it is both one-
to-one and onto.
R EMARK 1.1.9. In other words, to say a function f : X → Y is one-to-one is to say that it
sends at most one element of X to any given element of Y . To say that it is onto is to say that
it sends at least one element of X to any given element of Y . So, of course, to say that it is a
one-to-one correspondence is to say that it sends exactly one element of X to each element of Y .
E XAMPLES 1.1.10.
a. The map f : Z → Z defined by f (x) = 2x for every x ∈ Z is one-to-one, but not onto.
b. The function f : R → R defined by f (x) = x3 is a bijection.
c. The function f : R → R defined by f (x) = x sin(x) is onto, but not one-to-one.
D EFINITION 1.1.11.
a. A set X is finite if X has only a finite number of elements, and it is infinite otherwise.
b. If X is a finite set, then the order |X| of X is the number of elements it has.
P ROPOSITION 1.1.12. Let X and Y be finite sets of the same order, and let f : X → Y be a
function. Then f is one-to-one if and only if it is onto.
P ROOF. Let n = |X|, and denote the elements of X by x1 , . . . , xn . If f (xi ) = f (x j ) for some
i 6= j, then the subset { f (x1 ), . . . , f (xn )} of Y has fewer than n elements, hence cannot equal Y .
Conversely, if { f (x1 ), . . . , f (xn )} has fewer than n elements, then f (xi ) = f (x j ) for some i 6= j.
Therefore f is not one-to-one if and only if it is not onto, as desired. 
 1.2. RELATIONS 15

Note that every bijection has an inverse.

D EFINITION 1.1.13. If f : X → Y is a bijection, then we define the inverse of f to be the
function f −1 : Y → X satsifying f −1 (y) = x for the unique x such that f (x) = y.
Given a bijection f : X → Y , note that
f −1 ( f (x)) = x and f ( f −1 (y)) = y
for all x ∈ X and y ∈ Y . In other words, f −1 ◦ f (resp., f ◦ f −1 ) is the function that takes every
element of Y (resp., X) to itself.
E XAMPLE 1.1.14. The function f : R → R defined by f (x) = x3 has inverse f −1 (x) = x−1/3 .
Often, it is useful to use what is called an indexing set I to define a collection, which is just
some given set, like the natural numbers. Given objects xi for each i ∈ I, we can use set-theoretic
notation to define a set consisting of them
{xi | i ∈ I}
that is in one-to-one correspondence with I via the map that takes i to xi .
D EFINITION 1.1.15. Let X be a set and {Yi | i ∈ I} be a collection of subsets of X indexed by
a set I.
a. The intersection and union of the sets Yi are defined as
\ [
Yi = {x ∈ X | x ∈ Yi for all i ∈ I} and Yi = {x ∈ X | x ∈ Yi for some i ∈ I},
i∈I i∈I
respectively.
b. If Yi ∩Y j = ∅ for every i, j ∈ I with i 6= j, we say that the sets Yi are disjoint.
c. If the collection of Yi is disjoint, then their union is called a disjoint union and is often
written as a
Yi .
i∈I

D EFINITION 1.1.16. Let {Xi | i ∈ I} be a collection of sets. The direct product ∏i∈I Xi of the
Xi is the set of tuples
∏ Xi = {(xi)i∈I | xi ∈ Xi}.
i∈I
In other words, an element of ∏i∈I Xi is a choice of one element of Xi for each i ∈ I.

1.2. Relations
In this section, we consider several types of a very general construct called a relation.
D EFINITION 1.2.1. A relation R is a subset of X × Y . We often write xRy to indicate that
(x, y) ∈ R.
E XAMPLES 1.2.2.
16 1. SET THEORY

a. The circle S1 = {x2 + y2 = 1} forms a relation in R × R. As is well-known, a pair (x, y) is

in S1 if and only if (x, y) = (cos θ , sin θ ) for some θ ∈ [0, 2π).
b. The relation ≤ on R × R is given by {(x, y) | x ≤ y}.
As a first example, we see that functions can be considered as relations.
R EMARK 1.2.3. A function f : X → Y gives rise to a relation
Γ f = {(x, f (x)) | x ∈ X} ⊆ X ×Y,
known as the graph of f . Equivalently, each relation R in X × Y such that for each x ∈ X there
exists a unique y ∈ Y with xRy gives rise to a function f defined by f (x) = y (where xRy).
E XAMPLE 1.2.4. The relation in R2 corresponding to f : R → R is the graph of f in the usual
sense.
We will consider two other types of relations.
D EFINITION 1.2.5. An equivalence relation ∼ on X is a relation in X × X that satisfies the
following properties.
a. (reflexivity) For all x ∈ X, we have x ∼ x.
b. (symmetry) For any x, y ∈ X, we have x ∼ y if and only if y ∼ x.
c. (transitivity) If x, y, z ∈ X satisfy x ∼ y and y ∼ z, then x ∼ z.
E XAMPLES 1.2.6.
a. Equality is an equivalence relation = on any set X. As a relation, it defines the subset
{(x, x) | x ∈ X} of X × X.
b. The relation ≤ on R is not an equivalence relation, as it is not symmetric.
c. Let n be a positive integer, and consider the relation ≡n on Z defined by a ≡n b if a − b
is divisible by n. This is an equivalence relation known as congruence modulo n. We will write
a ≡ b mod n in place of a ≡n b, as is standard.
D EFINITION 1.2.7. Let ∼ be an equivalence relation on a set X. The equivalence class of
x ∈ X is the set {y ∈ X | x ∼ y}.
E XAMPLES 1.2.8.
a. The equivalence classes under = on a set X are just the singleton sets {x} for x ∈ X.
b. The equivalence class of 3 under ≡7 on Z is {. . . , −11, −4, 3, 10, 17, . . .}.
D EFINITION 1.2.9. We refer to the set of equivalence classes of Z under congruence modulo
n as the integers modulo n, and denote it Z/nZ. (Note that number theorists usually denote this
set Z/nZ.) A typical member a has the form
a = {a + bn | b ∈ Z}
for some integer a. An equivalence class a is known as a congruence class modulo n.
 1.2. RELATIONS 17

L EMMA 1.2.10. The distinct equivalence classes of X under an equivalence relation ∼ are
disjoint, and X is the disjoint union of its distinct equivalence classes.
P ROOF. The second statement follows from the first once we known that different equiva-
lence classes are disjoint, since every x ∈ X is in some equivalence class. For the first statement,
suppose that x and y are elements of X, and let Ex and Ey denote their respective equivalence
classes under ∼. If Ex and Ey are not disjoint, then there exists z ∈ Ex ∩ Ey , so x ∼ z and y ∼ z.
But then z ∼ x by symmetry of ∼, and so for any w ∈ X, we have x ∼ w implies z ∼ w by transi-
tivity of ∼. Given that and using y ∼ z, we then have y ∼ w, again by transitivity. Hence Ex ⊆ Ey .
But since x and y are interchangeable in the last sentence, we have Ey ⊆ Ex as well. Therefore,
Ex = Ey , which is to say any two equivalence classes of X are either distinct or equal. 
D EFINITION 1.2.11. Let X be a set and ∼ an equivalence relation on X.
a. For any equivalence class E of ∼, a representative of E is just an element of E.
b. A set of representatives (of the equivalence classes) of X under ? is a subset S of X such
that each equivalence class of X contains exactly one element of S.
E XAMPLE 1.2.12. The set {0, 1, 2, . . . , n−1} is a set of representatives of Z under congruence
modulo n.
D EFINITION 1.2.13. A partial ordering on a set X is a relation ≤ in X × X that satisfies the
following properties.
i. (reflexivity) For all x ∈ X, we have x ≤ x.
ii. (antisymmetry) If x, y ∈ X satisfy x ≤ y and y ≤ x, then x = y.
iii. (transitivity) If x, y, z ∈ X satisfy x ≤ y and y ≤ z, then x ≤ z.
A set X together with a partial ordering ≤ is referred to as a partially ordered set.
E XAMPLES 1.2.14.
a. The relation ≤ on R is a partial ordering, as is ≥.
b. The relation < on R is not a partial ordering, as it is not reflexive.
c. The relation ⊆ on the set of subsets PX of any set X, which is known as the power set of
X, is a partial ordering.
d. The relation = is a partial ordering on any set.
e. The relation ≡n is not a partial ordering on Z, as 0 and n are congruent, but not equal.
Given a partial ordering ≤ on a set X, we can speak of minimal and maximal elements of X.
D EFINITION 1.2.15. Let X be a set with a partial ordering ≤.
a. A minimal element in X (under ≤) is an element x ∈ X such that if z ∈ X and z ≤ x, then
z = x.
b. A maximal element y ∈ X is an element such that if z ∈ X and y ≤ z, then z = y.
Minimal and maximal elements need not exist, and when they exist, they need not be unique.
Here are some examples.
18 1. SET THEORY

E XAMPLES 1.2.16.
a. The set R has no minimal or maximal elements under ≤.
b. The interval [0, 1) in R has the minimal element 0 but no maximal element under ≤.
c. The power set PX of X has the minimal element ∅ and maximal element X under ⊆.
d. Under = on X, every element is both minimal and maximal.
e. Consider the set S of nonempty sets of the integers Z, with partial ordering ⊆. The minimal
elements of S are exactly the singleton sets {n} for n ∈ Z.
One can ask for a condition under which maximal (or minimal) elements exist. To phrase
such a condition, we need two more notions.
D EFINITION 1.2.17. Let X be a set with a partial ordering ≤. A chain in X is a subset C of
X such that if x, y ∈ C, then either x ≤ y or y ≤ x.
That is, a chain is a subset under which every two elements can be compared by the partial
ordering.
E XAMPLE 1.2.18. The power set PX of X = {1, 2, 3} is not a chain, as we have neither {1, 2}
contained in {2, 3}, nor {2, 3} contained in {1, 2}. However, its subset {∅, {1}, {1, 2}, {1, 2, 3}}
does form a chain.
E XAMPLE 1.2.19. Any subset of R forms a chain under ≤.
The previous example leads us to the following definition, which we mention primarily as a
remark.
D EFINITION 1.2.20. If X is itself a chain under ≤, then ≤ is said to be a total ordering on X.
We need the notion of bounds on subsets of a partially ordered set.
D EFINITION 1.2.21. Let X be a set with a partial ordering ≤. Let A be a subset of X. An
upper bound on A under ≤ is an element x ∈ X such that a ≤ x for all a ∈ A.
That is, an upper bound on a subset is an element of the set that is at least as large as every
element in the subset. Note that the upper bound need not, but can, be contained in the subset
itself. (And, of course, lower bounds could have been defined similarly.)
E XAMPLES 1.2.22.
a. The subset [0, 1) of R has an upper bound 1 ∈ R under ≤. In fact, any element x ≥ 1 is an
upper bound for [0, 1). The subset [0, 1] has the same upper bounds.
b. The subset Q of R has no upper bound under ≤.
We now come to Zorn’s lemma, which is equivalent to the so-called “axiom of choice”, and
as such, is as much an axiom as it is a theorem (and more of a theorem than it is a lemma). Some,
though far from most, mathematicians prefer not to include the axiom of choice among the ax-
ioms of set theory, fearing that the resulting collection of axioms may be logically incompatible.
For the purposes of this book, we will have no such qualms, and we state Zorn’s lemma without
proof: the reader may take it as an axiom.
 1.3. BINARY OPERATIONS 19

T HEOREM 1.2.23 (Zorn’s lemma). Let X be a set with a partial ordering ≤, and suppose
that every chain in X has an upper bound. Then X contains a maximal element.
Later on in these notes, we will see a couple of examples where Zorn’s lemma can be used
to produce the existence of maximal elements in situations of use to algebraists. Zorn’s lemma
is the form of the axiom of choice considered most conducive to applications in algebra.
Finally, let us consider the notion of generation. We have the following rather obvious lemma.
L EMMA 1.2.24. Let X be a set and S be a subset. Let P be a subset of PX containing X
such that P is closed under intersection, and let PS be the (nonempty) subset of elements of P
containing S. Then the intersection of the elements of PS is the unique minimal element of PS .
That is, it is the smallest subset of X in P containing S.
We think of P of some property of certain subsets of X that X itself satisfies, where a subset
of X is in P it has the property. As P is closed under intersection, for any subset S of X, we may
speak of the smallest subset that contains S and has property P. We then think of this subset as
the subset of X with property P generated by S. For instance, we have the following.
E XAMPLE 1.2.25. Let X be a set and S ⊆ X × X be a relation on X. The set of equivalence
relations on X is closed under intersection, as one may easily check, and X × X is an equivalence
relation. Thus, the intersection all equivalence relations containing S is the smallest equivalence
relation ∼S containing S. We call ∼S the equivalence relation generated by S.
Two elements x, y ∈ X are equivalent under ∼S if and only if there exist a sequence of ele-
ments z0 , . . . , zn ∈ X with x = z0 and y = zn for some n ≥ 1 such that zi = zi+1 , (zi , zi+1 ) ∈ S, or
(zi+1 , zi ) ∈ S for every 0 ≤ i ≤ n − 1. To see this, one checks two things: first, that what we have
just described defines an equivalence relation on S, and secondly, that any equivalence relation
on S must contain every such pair (x, y).

1.3. Binary operations

To give context to the term “binary operation”, which we study in this section, here is what
one might refer to simply as an “operation”.
D EFINITION 1.3.1. A (left) operation ? of a set X on a set Y is a function ? : X ×Y → Y .
N OTATION 1.3.2. The value ?(x, y) of (x, y) ∈ X × Y under an operation ? : X × Y → Y is
denoted x ? y. It is often referred to as the product of x and y under ? (when confusion does not
arise from this language).
E XAMPLE 1.3.3. The set R acts on Rn for any n ≥ 1 by left diagonal multiplication. That is,
we have
a · (x1 , . . . , xn ) = (ax1 , . . . , axn )
n
for a ∈ R and (x1 , . . . , xn ) ∈ R . Geometrically, this is the action of scaling of a vector.
If Z is a subset of Y , we can ask if the values x ? z for x ∈ X and z ∈ Z land in Z.
D EFINITION 1.3.4. Let ? : X ×Y → Y be a (left) operation of X on Y . A subset Z of Y is said
to be closed under ? (or, left multiplication by ?) if x ? z ∈ Z for all x ∈ X and z ∈ Z.
20 1. SET THEORY

E XAMPLE 1.3.5. Consider the operation · : Z × Z → Z of multiplication. The subset E of

even numbers in Z is closed under this operation, which is to say left multiplication by integers.
That is, if a ∈ Z and b ∈ E, then ab ∈ E. However, the subset O of odd numbers is not closed
under this operation. For instance, 2 ∈ Z and 1 ∈ O, but 2 · 1 = 2 ∈
/ O.
D EFINITION 1.3.6. Let ? : X ×Y → Y be an operation and Z be a subset of Y that is closed
under ?. Then the restriction of ? to an operation of X on Z is the operation ?Z : X × Z → Z
defined by x ?Z z = x ? z for all x ∈ X and z ∈ Z.
In this text, we will most often encounter binary operations.
D EFINITION 1.3.7. A binary operation on a set X is an operation of X on itself.
R EMARKS 1.3.8. Let X be a set.
a. A binary operation ? on X is simply a function ? : X × X → X.
b. We often refer to a binary operation on X more simply as an “operation” on X, the fact
that X is operating on itself being implied.
E XAMPLES 1.3.9. The following are binary operations.
a. Addition (or subtraction) + on Z, Q, R, C, Rn , and m-by-n matrices Mmn (R) with entires
in R for any m, n ≥ 1.
b. Multiplication · on Z, Q, R, C, and square n-by-n matrices Mn (R) for any n ≥ 1.
c. Composition ◦ on the set Maps(X, X) of maps from a set X to itself, e.g., X = R.
d. Union ∪ and intersection ∩ on power set PX of any set X.
E XAMPLES 1.3.10.
a. Exponentiation is not a binary operation on C, as (−1)1/2 , for instance, has two possible
values. It is therefore not well-defined.
b. Addition is not a binary operation on the set R× of nonzero real numbers, as −1 + 1 = 0,
/ R× . We say that R× is not closed under addition.
and 0 ∈
c. Division in not a binary operation on R, as division by 0 is not defined, but division is a
binary operation on R× .
We can define a binary operation on a finite set via a multiplication table.
E XAMPLE 1.3.11. Consider the set X = {a, b, c}. The following table defines a binary oper-
ation ? on X:
? a b c
a b c a
b a c c
c b a c
The entry in row b and column a is, by way of example, b ? a, and therefore, b ? a = a. On the
other hand, a ? b is located in the row corresponding to a and column of b, and hence a ? b = c.
 1.3. BINARY OPERATIONS 21

In the previous example, we could have filled in the nine entries in the bottom right 3-by-
3 square arbitrarily with elements of X, as there are no conditions of the values of a binary
operation. Often, it is useful to impose conditions that give additional structure.
D EFINITION 1.3.12. Let X be a set.
a. A binary operation ? on X is associative if
(x ? y) ? z = x ? (y ? z)
for all x, y, z ∈ X.
b. A binary operation ? on X is commutative if
x?y = y?x
for all x, y ∈ X.
E XAMPLES 1.3.13.
a. Addition is associative and commutative on Z, Q, R, C, Rn , Mmn (R), and Maps(R, R).
b. Subtraction is neither associative nor commutative on the sets Z, Q, R, C, Rn , Mmn (R),
and Maps(R, R).
c. Multiplication is associative and commutative on Z, Q, R, C, Rn , Maps(R, R), and is
associative but not commutative on Mn (R) for n ≥ 2.
d. Union and intersection are associative and commutative binary operations on PX .
e. Composition on Maps(X, X) is an associative binary operation, but it is not commutative
if X has at least 3 elements.
D EFINITION 1.3.14. Let X be a set and ? a binary operation on X. Two elements x, y ∈ X are
said to commute under ? if x ? y = y ? x.
Commutativity of a binary operation on a finite set can be seen on its mutliplication table,
as the table is then symmetric across the diagonal. Associativity is hard to see, but it is a strong
condition. Here are some examples.
E XAMPLES 1.3.15. The following are tables of binary operations on the set {a, b}:
? a b ∗ a b  a b
a b a a a b a b a
b b a b b a b a a
Of these, only ∗ is associative, while only ∗ and  are commutative, since a and b do not commute
under ?.
E XAMPLE 1.3.16. We can define operations + and · on Z/nZ as follows. Let a, b ∈ Z, and
recall that we denote their equivalence classes under congruence modulo n by a, b ∈ Z/nZ. We
define a + b = a + b and a · b = a · b. These are well-defined, as if c and d are congruent modulo
n to a and b, respectively, then c + d ≡ a + b mod n and c · d ≡ a · b mod n.
22 1. SET THEORY

D EFINITION 1.3.17. A set X together with a binary operation ? : X × X → X is called a

binary structure, and we write it as a pair (X, ?).
R EMARK 1.3.18. If (X, ?) is a binary structure, then we often refer to X as the underlying
set.
E XAMPLE 1.3.19. The pair (Z/nZ, +) is a binary structure, as is (Z/nZ, ·).
D EFINITION 1.3.20. Let (X, ?) be a binary structure. A subset A is said to be closed under
the binary operation ? if a ? b ∈ A for all a, b ∈ A.
D EFINITION 1.3.21. Let (X, ?) be a binary structure and A a closed subset of X. Then the
restriction of ? to A is a binary operation ?A : A × A → A defined by a ?A b = a ? b for all a, b ∈ A.
We usually denote ?A more simply by ?.
R EMARK 1.3.22. If (X, ?) is a binary operation and A is a closed subset of X, then (A, ?) is a
binary structure as well.
E XAMPLES 1.3.23.
a. The subsets Z, Q, and R of C are closed under the binary operation +.
b. The subset [−1, 1] of R is not closed under +, though it is under ·.
c. The set of all nonempty subsets of a set X is closed under the binary operation ∪ on PX ,
but not under the operation ∩ (unless X has fewer than two elements).
d. The matrices in Mn (R) with determinant 1 are closed under multiplication. The resulting
binary structure is denoted SLn (R).
R EMARK 1.3.24. If ? : X × X → X is a binary operation, then we can also think of it as an
operation. However, the notions of a subset A of X being closed under ? as a binary operation and
being closed under ? as an operation do not in general coincide. The first says that ? restricts to a
binary operation ? : A × A → A, while the second says that ? restricts to an operation ? : X × A →
A. In other words, the first requires only that the product of any two elements of A lands in A
(under ?), while the second says that the product x ? a lands in A for any x ∈ X and any a ∈ A,
which is a stronger condition.
E XAMPLE 1.3.25. Consider the set Z and the binary operation · of multiplication on it. The
set of odd numbers E is closed under multiplication thought of as a binary operation, since the
product of any two odd numbers is odd. However, it is not closed · thought of as an operation of Z
on itself, since the product of an even number and an odd number is not odd (as in Example 1.3.5).
Look for similarities in the following tables of binary structures with underlying sets of order
3.
+ 0 1 2 ? a b c
0 0 1 2 a a b c
1 1 2 0 b b c a
2 2 0 1 c c a b
 1.3. BINARY OPERATIONS 23

In fact, if we replace + by ?, 0 by a, 1 by b, and 2 by c, the first table becomes the second. In

a sense, these binary operations are the “same”. We give this notation of sameness a technical
definition. Note that to be the same in this sense, there must exist a bijection between the sets:
i.e., if they are finite, they must have the same number of elements.
D EFINITION 1.3.26. Let (X, ?) and (Y, ∗) be binary structures. We say that they are isomor-
phic if there exists a bijection f : X → Y such that
f (a ? b) = f (a) ∗ f (b)
for all a, b ∈ X. We then say that f is an isomorphism.
R EMARK 1.3.27. If we remove the condition of bijectivity in Definition 1.3.26, then the map
f : X → Y with f (a ? b) = f (a) ∗ f (b) is called a homomorphism of binary structures.
In the above example f (0) = a, f (1) = b, and f (2) = c, and the condition that f (x + y) =
f (x) ? f (y) for all x, y ∈ {0, 1, 2} is exactly that the multiplication tables match.
E XAMPLES 1.3.28.
a. The map f : Z → Z defined by f (n) = −n defines an isomorphism from (Z, +) to itself.
b. The map f : Z → Z defined by f (n) = 2n is not an isomorphism from (Z, +) to itself. It
satisfies f (m + n) = f (m) + f (n), but it is not onto.
c. Let R>0 = {x ∈ R | x > 0}. Define f : R → R>0 by f (x) = ex . This is an isomorphism
from (R, +) to (R>0 , ·), since
ex+y = ex ey
for all x, y ∈ R.
d. The map f : R → R defined by f (x) = x3 is not an isomorphism from (R, +) to itself, as
f (1 + 1) = 8 6= 2 = f (1) + f (1).
On the other hand, the same map does define an isomorphism from (R, ·) to (R, ·).
We have the following lemma.
L EMMA 1.3.29. Suppose that f is an isomorphism from (X, ?) to (Y, ∗). Then the inverse
f −1 of f is an isomorphism from (Y, ∗) to (X, ?).
P ROOF. Let y1 , y2 ∈ Y . Then there exists x1 , x2 ∈ X with f (x1 ) = y1 and f (x2 ) = y2 We have
f −1 (y1 ) ? f −1 (y2 ) = x1 ? x2 ,
and
f (x1 ? x2 ) = f (x1 ) ∗ f (x2 ) = y1 ∗ y2 ,
so
x1 ? x2 = f −1 (y1 ∗ y2 ),
as desired. 
24 1. SET THEORY

E XAMPLE 1.3.30. The inverse of f : R → R>0 with f (x) = ex is f −1 (x) = log(x), which
satisfies
log(x · y) = log(x) + log(y)
for x, y ∈ R>0 .
In fact, the properties of being isomorphic puts an equivalence relation on any set of binary
structures.
E XAMPLE 1.3.31. The set of representatives for the isomorphism classes (i.e., equivalence
classes under isomorphism) of binary structures on the set {a, b} has 10 elements. That is, one
can construct at most 10 tables for binary operations on {a, b} that give binary structures, no two
of which are isomorphic, as the reader can check.
 CHAPTER 2

Group theory

2.1. Groups
In this section, we introduce groups, which can briefly be defined as associative binary struc-
tures with identities and inverses. We begin by defining the two latter terms.
D EFINITION 2.1.1. Suppose that (X, ?) is a binary structure.
a. A left (resp., right) identity element of X is an element e ∈ X that satisfies e ? x = x (resp.,
x ? e = x).
b. If e ∈ X is both a left and a right identity element of X, we say that it is an identity element
of X.
E XAMPLES 2.1.2.
a. Under addition, 0 is a left and right identity element in Z, Q, R, C, Rn , Mn (R), and
Maps(R, R), with 0 in the latter three examples being the zero vector, zero matrix, and constant
function with value 0. Similarly, under multiplication, 1 is a left and right identity element in all
of the latter sets.
b. Under subtraction on the sets from part a, the element 0 is a right identity but there is no
left identity element.
c. Under composition, f (x) = x is an identity element in Maps(R, R).
d. Under union, ∅ is an identity element in PX .
e. Multiplication is a binary operation on the even integers 2Z but 2Z has no left and no right
identity elements.
f. For the binary structure defined on {a, b} by the table
? a b
a a b
b a b
a and b are both left identity elements, but there is no right identity element.
One could ask whether or not there can be more than one (left and right) identity element in
a binary structure. The following provides the answer.
L EMMA 2.1.3. Let (X, ?) be a binary structure. Suppose that e ∈ Xis a left identity element
and that f ∈ X is a right identity element. Then e = f , and in particular e is an identity element
in X.
25
26 2. GROUP THEORY

P ROOF. If f is a right identity element, we have e ? f = e. On the other hand, since e is a left
identity element, we have e ? f = f . Therefore, we have e = f . 
The following is an immediate corollary.
C OROLLARY 2.1.4. Let (X, ?) be a binary structure that contains an identity element e. Then
every (left or right) identity element in X is equal to e.
D EFINITION 2.1.5. Suppose that (X, ?) is a binary structure with an identity element e ∈
X.
a. A left (resp., right) inverse of x ∈ X is an element y ∈ X such that y ? x = e (resp., x ? y = e).
b. An element that is both a left and a right inverse to x ∈ X is called an inverse of x ∈ X.
E XAMPLES 2.1.6.
a. In Z, Q, R, C, Mn (R), and Maps(R, R), the negative −x of an element called x is the
inverse under addition. Under multiplication, x−1 = 1/x is the inverse of any x 6= 0 in Q, R, and
C. The elements that have multiplicative inverses in Z are ±1, in Mn (R) they are the matrices
with nonzero determinant, and in Maps(R, R) they are the nowhere vanishing functions.
b. Under subtraction on the sets of part a, an element x is its own left and right inverse.
c. Under composition, an element f ∈ Maps(R, R) has an inverse f −1 if and only if it is a
bijection.
d. Under union on PX , only ∅ has an inverse, which is itself.
e. For the binary structure defined on {a, b, c} by the table
? a b c
a a b c
b b a a
c c b c
a is an identity element and is its own inverse, b is an inverse of itself, c is a right inverse of b
and therefore b is a left inverse of c, but c has no right inverse.
L EMMA 2.1.7. Let (X, ?) be a binary structure with an identity element e. Suppose that x ∈ X
has a left inverse y and a right inverse z. Then y = z.
P ROOF. We need only write down the chain of equalities
y = y ? e = y ? (x ? z) = (y ? x) ? z = e ? z = z.

With the concepts of identity elements and inverses in hand, we now give the full definition
of a group.
D EFINITION 2.1.8. A group is a set G together with a binary operation ? : G × G → G such
that
i. ? is associative,
 2.1. GROUPS 27

ii. there exists an element e ∈ G such that e ? x = x = x ? e, and

iii. for every x ∈ G, there exists an element y ∈ G such that x ? y = e = y ? x.
In other words, a group is a set with an associative binary operation, an identity element, and
inverses with respect to that identity element.
Here are some examples of groups.
E XAMPLES 2.1.9.
a. Under addition, Z, Q, R, C, Mn (R), and Maps(R, R) are all groups.
b. For X = Q, R, or C, we set X × = X − {0}. Under multiplication, Q× , R× , and C× are
groups.
c. Under multiplication, the set GLn (R) of invertible n by n-matrices (i.e., those with nonzero
determinant) forms a group, known as the general linear group.
d. Under multiplication, the set of nowhere vanishing functions in Maps(R, R) forms a
group.
e. The set {e} consisting of a single element is a group under the binary operation ? defined
by e ? e = e. This group is known as the trivial group.
On the other hand, here are some of many binary structures that are not groups.
E XAMPLES 2.1.10.
a. The integers are not a group under multiplication, nor are Q, R, or C before removing 0.
b. The set Maps(R, R) is not a group under composition, as not every function has an inverse.
c. The set PX of subsets of a set is not a group under union.
The following theorem is used in showing the uniqueness of inverses.
P ROPOSITION 2.1.11 (Cancellation theorem). Let G be a group, and let x, y, z ∈ G be such
that
x?y = x?z (resp., y ? x = z ? x).
Then y = z.
P ROOF. We prove the first statement. Let x0 be any (left) inverse to x. Under the given
assumption, we have
y = e ? y = (x0 ? x) ? z = x0 ? (x ? y) = x0 ? (x ? z) = (x0 ? x) ? z = e ? z = z.

The following is now quickly derived.
L EMMA 2.1.12. Let G be a group. If y, z ∈ G are both inverses to x ∈ G on either the left or
the right (or both), then y = z.
P ROOF. Suppose first that y and z are right inverses to x. Then we have
x ? y = e = x ? z,
28 2. GROUP THEORY

and the result now follows from the cancellation theorem. A similar argument holds if both y and
z are right inverses. In fact, even if y is a left inverse and z is a right inverse, there is by definition
of the group a third element x0 in the group that is both a left an a right inverse, and so equals
both y and z by what we have just proven. So y and z must be equal. 
N OTATION 2.1.13. Let G be a group and x ∈ G an element. Suppose the operation on G is
not denoted +. Then we (almost invariably) use the following notation.
a. The unique inverse to x is written x−1 .
b. Let n ∈ Z. We set x0 = e. If n ≥ 1, we usually write xn for x ? x ? · · · ? x, the product being
of n copies of x, which is unambiguously defined by the associativity of ?.
If the binary operation on the group is denoted +, then we write the inverse of x as −x and nx
instead of xn .
R EMARK 2.1.14. Let G be a group and e an element for which the operation is not denoted
as +. The reader should be able to check that for x ∈ G and m, n ∈ Z, one has
xm+n = xm xn , xmn = (xm )n , xn = (x−1 )−n = (x−n )−1 , and en = e.
D EFINITION 2.1.15. Let G be a group.
a. We say that G is abelian if its binary operation is commutative.
b. We say that G is nonabelian if its binary operation is not commutative.
E XAMPLES 2.1.16.
a. All of Z, Q, R, C, Mmn (R), and Maps(R, R) are abelian groups under addition.
b. The groups Q× , R× , and C× are abelian (under multiplication).
c. The group GLn (R) is nonabelian if n ≥ 2.
R EMARK 2.1.17. From now on, we will drop the use of ? for an arbitrary binary operation,
and simply use the more conventional symbol ·. However, the reader should keep in mind that
this does not mean that the operation in question is multiplication. Moreover, we shall often write
x · y more simply as xy.
L EMMA 2.1.18. Let G be a group. For x, y ∈ G, we have (xy)−1 = y−1 x−1 .
P ROOF. We have
(y−1 x−1 )(xy) = y−1 (x−1 (xy)) = y−1 ((x−1 x)y) = y−1 (ey) = y−1 y = e.
Therefore y−1 x−1 is left inverse to xy, and so by Lemma 2.1.12 it equals (xy)−1 . 
We end this section with a few more examples of groups.
E XAMPLE 2.1.19. The set Z/nZ of congruence classes modulo n forms a group under the
addition law
a + b = a + b.
The identity is 0, and the inverse of a is −a.
Clearly, Z/nZ is an abelian group.
 2.1. GROUPS 29

R EMARK 2.1.20. Usually, we simply write a for a. We have kept up the distinction to this
point to make clear the difference between a and its equivalence class. From now on, however, if
we understand that we are working in Z/7Z, e.g., from context, we will write equations such as
5 + 2 = 0, with the fact that we are working with equivalence classes as above being understood.
D EFINITION 2.1.21. The symmetric group SX on a set X is the set
SX = { f : X → X | f is bijective}
with the binary operation ◦ of composition.
D EFINITION 2.1.22. Let X be a set. An element of SX is referred to as a permutation of X.
We say that σ ∈ SX permutes the elements of X.
R EMARK 2.1.23. The group SX is alternately referred to as the group of permutations of a
set X.
R EMARK 2.1.24. The group SX is nonabelian if X has at least three elements.
E XAMPLE 2.1.25. If X = R, then f (x) = x + 1 and g(x) = x3 both lie in SR , but do not
commute.
D EFINITION 2.1.26. When X = {1, 2, . . . , n}, then we set Sn = SX , and we refer to Sn as the
symmetric group on n letters.
R EMARK 2.1.27. The notion of isomorphism of binary structures carries over to groups. An
isomorphism of groups is just an isomorphism of the underlying binary structures, i.e., a bijection
f : G → G0 between groups G and G0 such that
f (x · y) = f (x) · f (y)
for each x, y ∈ G. If G and G0 are isomorphic, we write G ∼ = G0 (noting that the property of being
isomorphic forms an equivalence class on any set of groups).
E XAMPLES 2.1.28.
a. The group GL1 (R) is isomorphic to R× via the map f : R× → GL1 (R) defined by f (a) =
(a).
b. Let X be a set with exactly n elements, say X = {x1 , x2 , . . . , xn }. Then we define an iso-
morphism
f : Sn −∼→ SX , f (σ )(xi ) = xσ (i) ,
which is to say that f takes a permutation σ ∈ Sn that takes i to some other number j to the
permutation in SX that maps xi to x j . In other words, it doesn’t matter whether we’re permuting
n cars or n apples: the groups are isomorphic.
To every group, we have an associated opposite group.
D EFINITION 2.1.29. The opposite group Gop of a group G is the set G together with the
operation x ? y = yx for x, y ∈ G.
E XAMPLE 2.1.30. The opposite group of an abelian group is the original group.
30 2. GROUP THEORY

2.2. Subgroups
D EFINITION 2.2.1. A subset H of a group G is a subgroup if it is closed under the binary
operation on G and is a group with respect to the restriction of that operation to a binary operation
on H. If H is a subgroup of G, we write H 6 G.
More succinctly, a subset of a group is a subgroup if it is a group with respect to the operation
on the group.
R EMARK 2.2.2. The relation 6 is a partial ordering on the set of subgroups of a group.
D EFINITION 2.2.3.
a. The set {e} containing only the identity element of G is a subgroup of G known as the
trivial subgroup (as it is a trivial group that is also a subgroup).
b. A subgroup H of G that is not the trivial subgroup is called nontrivial.
D EFINITION 2.2.4. If H is a subgroup of G with H 6= G, then we say that H is a proper
subgroup of G, and we write H < G.
E XAMPLES 2.2.5. The groups Z, Q, and R under addition are all subgroups of C.
To check that a group is a subgroup, one usually employs the following criteria.
T HEOREM 2.2.6. A subset H of a group G is a subgroup under the restriction on the binary
operation · on G if and only if
(0) e ∈ H,
(1) H is closed under ·,
(2) if h ∈ H, then h−1 ∈ H.
P ROOF. If H is a subgroup of G with respect to ·, then it is by definition closed under ·.
Since H is a group under ·, there exists an element f ∈ H with f · h = h for all h ∈ H. By the
cancellation theorem, we then have f = e, so e ∈ H. Also, for each h ∈ H, we have an element
h0 ∈ H with h · h0 = e. As e = h · h−1 , the cancellation theorem again tells us that h0 = h−1 , so
h−1 ∈ H. Therefore, the conditions (0)-(2) hold.
Conversely, if conditions (0)-(2) hold, then H is a binary structure under · by (1) and (0)
and (2) leave us only to verify associativity in the definition of a group. However, this follows
automatically on H from the associativity of · on the larger set G. 
E XAMPLES 2.2.7. The subset 2Z of Z is a subgroup under +. To see this, note that 0 is even,
the sum of two even integers is even, and the negative of an even integer is also even.
E XAMPLE 2.2.8. The subset
SLn (R) = {A ∈ GLn (R) | det(A) = 1}
of GLn (R) is a subgroup under ·, known as the special linear group. We use Theorem 2.2.6 to
check this:
(0) We have det In = 1, so In ∈ SLn (R).
 2.2. SUBGROUPS 31

(1) If A, B ∈ SLn (R), then

det(A · B) = det(A) · det(B) = 1
so A · B ∈ SLn (R).
(2) If A ∈ SLn (R), then
det(A−1 ) = det(A)−1 = 1
so A−1 ∈ SLn (R).
E XAMPLE 2.2.9. Let
S1 = {z ∈ C | |z| = 1} = {e2πiθ | θ ∈ R}.
Here e2πiθ corresponds to the point (cos θ , sin θ ) on the unit circle in the usual model of the
complex plane. In fact, recall that
e2πiθ = cos θ + i sin θ ∈ C
and q
| cos θ + i sin θ | = cos2 (θ ) + sin2 (θ ) = 1.
Then S1 is a subgroup of C× under ·. To see this, we check:
(0) We have |1| = 1.
(1) If z, w ∈ S1 , then z = e2πiθ and w = e2πiψ for some θ , ψ ∈ R. We have
zw = e2πi(θ +ψ) ∈ S1 .
(2) If z = e2πiθ , then
z−1 = e2πi(−θ ) ∈ S1 .
Theorem 2.2.6 has the following shorter formulation.
C OROLLARY 2.2.10. A nonempty subset H of a group is a subgroup under the restriction of
the binary operation · on G if and only if h · k−1 ∈ H for all h, k ∈ H.
P ROOF. If H is a subgroup of G and h, k ∈ H, then k−1 ∈ H and, consequently, hk−1 ∈
H by Theorem 2.2.6. Conversely, suppose hk−1 ∈ H for all h, k ∈ H. As H is nonempty, let
h ∈ H. Using this critersion, we have successively that e = hh−1 ∈ H, k−1 = ek−1 ∈ H, and
hk = h(k−1 )−1 ∈ H, so Theorem 2.2.6 implies that H is a subgroup. 
D EFINITION 2.2.11.
a. A group G is finite if its underlying set is finite. Otherwise, we say that G is infinite.
b. The order |G| of a finite group G is the order (number of elements in) of the underlying
set. If G is infinite, we say that its order is infinite.
The following provides an interesting example.
L EMMA 2.2.12. Let n ≥ 1. The group Sn is finite of order n!.
32 2. GROUP THEORY

P ROOF. For an arbitrary element σ ∈ Sn , we have n choices for the value σ (1). Then σ (2)
can be any of the remaining n − 1 values, and σ (3) is one of the then remaining n − 2 values, and
so forth, until one value is left for σ (n). Therefore, the order of Sn is n · (n − 1) · · · 1 = n!. 
E XAMPLE 2.2.13. As a further subgroup of S1 (so also a subgroup of C× ), we have
µn = {z ∈ C× | zn = 1} = {e2πik/n | k ∈ Z}.
To see the equality of the latter two sets, note that (e2πik/n )n = 1. On the other hand zn = 1
implies that |z|n = 1, so |z| = 1, which means that z = e2πiθ for some θ ∈ R. But the only way
that (e2πiθ )n = 1 can hold is for nθ to be an integer, which means exactly that θ = k/n for some
n ∈ Z. Note that |µn | = n, since e2πik/n = e2πi j/n if and only if j ≡ k mod n. That this order
equals |Z/nZ| = n is no coincidence. In fact, these two groups are isomorphic, as well shall see
in the following section.

2.3. Cyclic groups

D EFINITION 2.3.1. Let G be a group, and let g ∈ G. The cyclic subgroup of G generated by
g is the group
hgi = {gn | n ∈ Z}.
L EMMA 2.3.2. Let g ∈ G. Then hgi is the smallest subgroup of G containing g.
P ROOF. Since the smallest subgroup of G containing G is itself a group, it must contain gn
for all n ∈ Z, so it contains hgi. On the other hand, we see that hgi is a subgroup of G since
it contains e = g0 , is closed under multiplication (as gm · gn = gm+n ), and contains inverses (as
(gn )−1 = g−n ). Being that hgi is a subgroup of G contained in the smallest subgroup containing
g, it is itself the smallest subgroup. 
E XAMPLES 2.3.3.
a. The cyclic subgroup h2i of Z generated by 2 is 2Z.
b. The cyclic subgroup of GL2 (R) generated by
 
0 −1
A=
1 0
is
hAi = {I2 , A, −I2 , −A}.
D EFINITION 2.3.4.
a. A group G is called cyclic if there exists g ∈ G with G = hgi.
b. An element of g of a group G is called a generator if G = hgi. We then say that g generates
G and that G is generated by g.
R EMARK 2.3.5. Of course, any cyclic subgroup of a group G is itself a cyclic group.
E XAMPLES 2.3.6.
a. The group Z is cyclic, generated by 1.
 2.3. CYCLIC GROUPS 33

b. The group Z/nZ is cyclic for any n ≥ 0, again generated by 1.

c. The group µn is cyclic with generator e2πi/n .
d. The trivial group is a cyclic group of order 1.
R EMARK 2.3.7. Every cyclic group is abelian, since powers of a generator commute.
D EFINITION 2.3.8. Let G be a group. The order of an element g ∈ G is the smallest positive
integer n such that gn = e, if it exists. If such an n exists, then g is said to have finite order, and
otherwise g is said to have infinite order.
P ROPOSITION 2.3.9. Let g be an element in a group. Then the order of hgi and the order of
g are equal if either is finite (and both infinite otherwise). Moreover, for any i, j ∈ Z, we have
gi = g j if and only if
• i ≡ j mod n, if g is finite of order n, and
• i = j, if g has infinite order.
P ROOF. First, suppose that g has finite order n. If gi = g j , then gi− j = e. Note that gn = e as
well. Dividing i − j by n, we have
i − j = qn + r
for some quotient q ∈ Z and remainder 0 ≤ r ≤ n − 1. We then have
e = gi− j = gqn+r = (gn )q gr = gr ,
but r < n and n is minimal, so r = 0. That is, i − j is a multiple of n, so i ≡ j mod n. In particular,
the distinct elements of hgi are exactly e, g, . . . , gn−1 , so hgi has order n.
If g has infinite order, then for gi = g j to hold, one must have gi− j = e, which forces i = j.
Therefore, all of the powers of g are distinct, and hgi is infinite. 
L EMMA 2.3.10. Suppose that G is a cyclic group. If G is infinite, then G is isomorphic to Z.
Otherwise, G is isomorphic to Z/nZ, where n = |G|.
P ROOF. Let g be a generator of G. Suppose first that G is infinite. We define a map
f : Z → G, f (i) = gi for all i ∈ Z.
This is one-to-one since f (i) = f ( j) implies gi = g j , which can only happen if i = j by Propo-
sition 2.3.9. It is onto as every element of hgi has the form gi = f (i) for some i. It is then an
isomorphism of groups as
f (i + j) = gi+ j = gi g j = f (i) f ( j).
If |G| = n, then we define
f : Z/nZ → G, f (i) = gi for all i ∈ Z.
This is well-defined as f (i + qn) = gi+qn = gi , so it is independent of the choice of representative
of i modulo n. It is one-to-one as gi = g j implies i = j in Z/nZ by Proposition 2.3.9. It is then
onto and an isomorphism for the same reasons as in the infinite case. 
34 2. GROUP THEORY

As a result, the group µn is isomorphic to Z/nZ (under the map taking e2πik/n to k). The
groups nZ for n ≥ 1 are all isomorphic to Z itself, but for this one must take the map Z → nZ
that is multiplcation by n.
T HEOREM 2.3.11. Every subgroup of a cyclic group is cyclic.
P ROOF. Let H be a subgroup of a cyclic group G with generator g. Let k ≥ 1 be minimal
such that gk ∈ H. We claim that H = hgk i. Since H is closed under multiplication and inverses,
it must contain every power of gk , so it contains the subgroup hgl i. Now suppose that gi ∈ H
for some i ∈ Z. Again, divide i by k and get q ∈ Z and 0 ≤ r ≤ k − 1 with i = qk + r. Then
gi = (gk )q gr , so
gr = gi (gk )−q ∈ H,
in that H is a subgroup. But minimality forces r = 0, so i is a multiple of k, proving the claim. 
C OROLLARY 2.3.12. The subgroups of Z are exactly the nZ = hni with n a nonnegative
integer.
Let’s consider the subgroups of Z/nZ for some n ≥ 1, which we now know to be cyclic.
Recall that the greatest common divisor gcd(i, j) of two integers i and j that are not both zero is
defined to be the smallest positive integer dividing both i and j. We also set gcd(0, 0) = 0.
L EMMA 2.3.13. Given i, j ∈ Z, we have
hgcd(i, j)i = {ai + b j | a, b ∈ Z}.
P ROOF. In the case that i = j = 0, we have that both sides equal h0i, so the lemma holds,
and therefore we may assume that at least one is nonzero. Since gcd(i, j) divides both i and j,
we have i, j ∈ hgcd(i, j)i. As a subgroup, the latter group is closed under addition and taking
of negatives, so ai + b j is in it as well. In other words, H = {ai + b j | a, b ∈ Z} is contained in
hgcd(i, j)i.
Conversely, note that the set H is a (nontrivial) subgroup of Z in that it satisfies all of the
properties of one, so it equals hdi for some d ≥ 1. Since i, j ∈ hdi by definition, we have that
d divides both i and j, and therefore is less than or equal to gcd(i, j). On the other hand, we
know that d ∈ hgcd(i, j)i, so gcd(i, j) ≤ d, and therefore d = gcd(i, j). In other words, we have
H = hgcd(i, j)i. 
P ROPOSITION 2.3.14. Every subgroup of Z/nZ has the form hdi for some d ≥ 1 dividing n.
In fact, for any j ∈ Z, we have h ji = hgcd( j, n)i.
P ROOF. The second statement implies the first, so we focus on it. Since gcd( j, n) divides j,
we have that h ji 6 hgcd( j, n)i. On the other hand, we have by Lemma 2.3.13 that
gcd( j, n) ∈ {a j + bn | a, b ∈ Z}
inside Z, which means that gcd( j, n) ≡ a j mod n for some a ∈ Z. In other words, in Z/nZ, we
have gcd( j, n) ∈ h ji, so hgcd( j, n)i 6 h ji, as desired. 
R EMARK 2.3.15. The subgroup hni of Z/nZ is just the trivial subgroup h0i = {0}.
Recall that two integers are said to be relatively prime if their greatest common divisor is 1.
 2.4. GENERATORS 35

C OROLLARY 2.3.16. Let G be a group and g ∈ G an element of order n.

a. For i ∈ Z, the order of gi is n/d, where d = gcd(i, n), and hgi i = hgd i.
b. The generators of hgi are the gi with i relatively prime to n.
P ROOF. Consider the isomorphism φ : G → Z/nZ under which gi is taken to i. This carries
the subgroup hgi i bijectively to the subgroup hii, which by Proposition 2.3.14 equals hdi. But
the latter group has elements 0, d, 2d, . . . , (n/d − 1)d, so has order n/d. As φ is a bijection, part
a is then seen to hold. Part b then follows immediately from part a, as the i for which hgi i = hgi
are the i with gcd(i, n) = 1. 
D EFINITION 2.3.17. The Euler phi-function is the map ϕ : Z>0 → Z>0 such that ϕ(n) is the
number of relatively prime integers to n between 1 and n.
R EMARK 2.3.18. The Euler phi-function ϕ has the properties that ϕ(mn) = ϕ(m)ϕ(n) when-
ever gcd(m, n) = 1 and that ϕ(pr ) = pr−1 (p − 1) for a prime number p and r ≥ 1. Its values on
1, 2, 3, 4, 5, . . . are 1, 1, 2, 2, 4, 2, 6, 4, 6, 4, 10, 4, 12, 6, 8, 8, 16, . . ..
R EMARK 2.3.19. It follows from Corollary 2b that the number of generators of a cyclic group
G of order n is exactly ϕ(n), where ϕ is the Euler phi-function.
2.4. Generators
The relation 6 is a partial ordering on any set of subgroups of a group. The following
proposition asserts the existence of minimal elements of certain such subsets. It is a consequence
of Lemma 1.2.24, but prove it here for convenience.
P ROPOSITION 2.4.1. Let G be a group, and let S be a nonempty subset of G. Then there
exists a smallest subgroup hSi of G containing S.
P ROOF. The set PS of subgroups of G containing S is nonempty, for it contains G itself. Set
\
hSi = H.
H∈X
As each H ∈ X contains G, so does hSi. Moreover, an arbitrary intersection of subgroups of G is
easily verified to itself be a subgroup of G, so hSi is a subgroup. Finally, if H is any subgroup of
G containing S, then H ∈ X, so hSi 6 H by definition of the intersection, so hSi is the smallest
such subgroup (i.e., the unique minimal element of X). 
D EFINITION 2.4.2. The smallest subgroup hSi containing a set S is the subgroup of G gener-
ated by S.
While this definition is rather abstract, we do have the following more concrete description
of the elements of hSi.
P ROPOSITION 2.4.3. Let S be a nonempty subset of G. An element g ∈ G is contained in hSi
if and only if g may be written as a product of powers of elements of S: i.e.,
mk
g = sm1 m2
1 s2 · · · sk
for some k ≥ 0, si ∈ S and mi ∈ Z for 1 ≤ i ≤ k.
36 2. GROUP THEORY

P ROOF. First, let g be an element that is a product of powers of elements of S. Since hSi is a
subgroup, it is closed under integer powers and products, so g ∈ hSi.
Conversely, note that the set H of elements that are products of powers of elements of S is
a subgroup of G, as it contains e = s0 for s ∈ S, is closed under products by definition, and is
closed under inverses as
(sm1 m2 mk −1
1 s2 · · · sk ) = s−m
k
k
· · · s−m2 −m1
2 s1 .
As H is a subgroup of G containing S but contained in hSi and hSi is the minimal such subgroup,
we have H = hSi. Thus, any element of hSi may be written as a product of powers of elements
of S, as desired. 
D EFINITION 2.4.4. We say that a subset S of G generates G if G = hSi, and then S is said to
be a set of generators of G.
D EFINITION 2.4.5. We say that a group G is finitely generated if there exists a finite set of
generators of G.
R EMARK 2.4.6. If G can be generated by a finite set {g1 , g2 , . . . , gn }, we usually write
hg1 , g2 , . . . , gn i
instead of
h{g1 , g2 , . . . , gn }i,
and we say that G is generated by g1 , g2 , . . . , gn .
E XAMPLE 2.4.7. A cyclic group is finitely generated: in fact, it is generated by a single
element.
E XAMPLE 2.4.8. Any finite group is finitely generated, as it is generated by itself.
E XAMPLE 2.4.9. Consider the subgroup G of GL2 (R) that is
(−1)i
  
b
G= i, j, b ∈ Z .
0 (−1) j
It can be generated by the set
     
−1 0 1 0 1 1
, , .
0 1 0 −1 0 1
To see this, note that
 b  
1 1 1 b
= ,
0 1 0 1
and j  i 
(−1)i
  
1 0 1 b −1 0 b
= .
0 −1 0 1 0 1 0 (−1) j
The group G is not cyclic as it is infinite but contains elements of order 2, but all infinite cyclic
groups are isomorphic to Z. In fact, G cannot be generated by any two of its elements: the proof
of this more tricky fact is left to the reader.
 2.5. DIRECT PRODUCTS 37

E XAMPLE 2.4.10. The group Q can be generated by the set { 1n | n ≥ 1}. However, Q is not
finitely generated. For, any integer N ≥ 1 and nonzero integers ai , bi with bi > 0 for 1 ≤ i ≤ N.
Any element of  
a1 a2 aN
, ,...,
b1 b2 bN
must have denominator, when put in reduced form, that is a divisor of b1 b2 · · · bN . But clearly not
every fraction has such a denominator, so Q cannot be finitely generated.
2.5. Direct products
Given any two groups, we can form a new group out of them, known as the direct product,
whose underlying set is in fact exactly the direct product of the underlying sets of the groups in
question.
D EFINITION 2.5.1. Let G and G0 be groups. The direct product of G and G0 is the binary
structure G × G0 that is the direct product of the sets G and G0 together with the binary operation
defined by
(a, a0 ) · (b, b0 ) = (a · b, a0 · b0 )
for a, b ∈ G and a0 , b0 ∈ G0 .
One might expect the direct product of G and G0 to be a group, and in fact it is. The straight-
forward check is left to the reader.
L EMMA 2.5.2. The direct product G × G0 of two groups is a group.
Of course, using this construction, we can think up more examples of new groups than we
can mention, e.g., Sm × GLn (R) for any m, n ≥ 1. The following remarks are easily verified from
the definition of the direct product.
R EMARK 2.5.3. The group G × G0 is abelian if and only if both G and G0 are abelian.
R EMARK 2.5.4. If f : G → H is a group isomorphism and G0 is another group, then the map
f 0 : G × G0 → H × G0
given by f 0 (g, g0 ) = ( f (g), g0 ) for g ∈ G and g0 ∈ G0 is an isomorphism as well.
R EMARK 2.5.5. Direct product forms an associative and commutative binary operation on
any set of isomorphism classes of groups. That is, for any groups G1 , G2 , and G3 , we have
(G1 × G2 ) × G3 ∼
= G1 × (G2 × G3 ) and G1 × G2 ∼ = G2 × G1 .
In particular, the associativity means it makes sense to speak of the group
G1 × G2 × · · · × Gn
for any groups G1 , G2 , . . . Gn .
R EMARK 2.5.6. If each of the groups G1 , G2 , . . . , Gn is finite, then
n
|G1 × G2 × · · · × Gn | = ∏ |Gi |.
i=1
38 2. GROUP THEORY

N OTATION 2.5.7. We write Gn for the direct product G × G × · · · × G of n copies of G.

R EMARK 2.5.8. More generally, for any collection
{Gi | i ∈ I}
of groups Gi for i in some indexing set I, we can put a binary operation on the direct product set
∏ Gi
i∈I
given by coordinate-wise multiplication
(ai )i∈I · (bi )i∈I = (ai · bi )i∈I ,
and the resulting group is known as the direct product of the Gi .
Let n ≥ 1, and let Gi be a group for each 1 ≤ i ≤ n. Let
G = G1 × G2 × · · · × Gn .
For g ∈ Gi , let g(i) ∈ G denote the element
g(i) = (e, . . . , e, g, e, . . . , e) ∈ G
that is nontrivial in only the ith coordinate of G and g in the ith coordinate.
P ROPOSITION 2.5.9. Suppose that Si is a generating set of Gi for each 1 ≤ i ≤ n. Then
n
{g(i) | g ∈ Si }
[
S=
i=1
is a generating set of G.
P ROOF. Suppose gi ∈ Gi for each 1 ≤ i ≤ n. Then
(1) (2) (n)
(g1 , g2 , . . . , gn ) = g1 g2 · · · gn ∈ hSi.

(i)
For example, if each Gi is cyclic with generator gi , then the set {gi | 1 ≤ i ≤ n} generates
G. While it is immediate from Proposition 2.5.9 that finite direct products of finitely gener-
ated groups are finitely generated, infinite direct products of nontrivial groups are never finitely
generated.
E XAMPLE 2.5.10. The group
∞
G = ∏(Z/2Z)
i=1
is not finitely generated. We give a very brief sketch of the proof: one checks that any finite set
of elements X in G must have the property that there exist positive integers j and k such that for
each x = (xi ) ∈ X, we have x j = xk . Then every element in hXi has this property, and since not
every element of G has this property, we have hXi 6= G.
The following result gives a general recipe for determining the order of (g1 , g2 , . . . , gn ).
 2.5. DIRECT PRODUCTS 39

T HEOREM 2.5.11. Suppose that gi ∈ Gi for each 1 ≤ i ≤ n. The order of g = (g1 , g2 , . . . , gn )

is the least common multiple of the orders of the gi if each of the elements gi has finite order, and
otherwise g has infinite order.
P ROOF. We have
gm = (gm m m
1 , g2 , . . . , gn ),
and this is the identity if and only if m is a multiple of the orders of each of the gi , so infinite if
any one of them is infinite, and otherwise a multiple of the least common multiple. 
E XAMPLE 2.5.12. Let G = Z/2Z × Z/3Z × Z/4Z × Z/4Z. Then every element of G has
order dividing lcm(2, 3, 4, 4) = 12.
The latter example illustrates a more general phenomenon.
D EFINITION 2.5.13. The exponent of a group G is the smallest integer n ≥ 1 such that gn = e
for all g ∈ G, if it exists. Otherwise, it is infinite.
C OROLLARY 2.5.14. If Gi has exponent ni for each 1 ≤ i ≤ n, then the exponent of G is the
least common multiple of the ni .
We mention the following result, the proof of which we leave to the reader.
P ROPOSITION 2.5.15. Suppose that {Gi | i ∈ I} is a collection of groups and, for each i ∈ I,
we are given Hi 6 Gi . Then we have
∏ Hi 6 ∏ Gi.
i∈I i∈I

Note, however, that not all subgroups of a direct product are direct products of subgroups.
E XAMPLE 2.5.16. There are 5 subgroups of the Klein four group Z/2Z × Z/2Z:
{0}, Z/2Z × Z/2Z, h(1, 0)i, h(0, 1)i, andh(1, 1)i.
The first four sit inside Z/2Z × Z/2Z as direct products of subgroups in the two individual
coordinates, while the final subgroup does not.
Finally, we note the following interesting fact.
T HEOREM 2.5.17. Let m and n be relatively prime positive integers. Then the natural map
θmn : Z/mnZ → Z/mZ × Z/nZ
induced by a 7→ (a, a) is an isomorphism. On the other hand, if m and n are not relatively prime,
then Z/mnZ and Z/mZ × Z/nZ are not isomorphic.
P ROOF. Suppose that m and n are relatively prime. Note that
θmn (a + b) = (a + b, a + b) = (a, a) + (b, b) = θmn (a) + θmn (b),
so θmn preserves the operation. If (a, a) = (b, b) in Z/mZ × Z/nZ, then m and n both divide
a − b, so mn does, as they are relatively prime. Therefore, θmn is injective. Since both groups
have the same order mn, it is surjective as well.
40 2. GROUP THEORY

If m and n are not relatively prime, then their least common multiple is
mn
lcm(m, n) = < mn.
gcd(m, n)
Corollary 2.5.14 then implies that the exponent of Z/mZ × Z/nZ is less than mn, the exponent
of Z/mZ × Z/nZ. As the exponent of a group is preserved by an isomorphism, the two groups
in question cannot be isomorphic. 

The following equivalent corollary is known as the Chinese remainder theorem (CRT).

C OROLLARY 2.5.18 (Chinese Remainder Theorem). Let k ≥ 2 and m1 , . . . , mk be mutually

relatively prime positive integers, which is to say that every pair of them is relatively prime. For
any b1 , . . . , bk ∈ Z, there exists an integer a, unique up to congruence modulo m1 m2 · · · mk , such
that a ≡ bi mod mi for each 1 ≤ i ≤ k.

P ROOF. The existence in the case k = 2 is equivalent to the surjectivity of θm1 m2 in The-
orem 2.5.17, while the uniqueness is its injectivity. The case of general k follows by an easy
induction on k. 

R EMARK 2.5.19. We can give an explicit recipe for the construction of solutions of congru-
ences modulo relatively prime integers (in the case of two congruences, and then by recursion).
The construction is contained in the following direct proof that the map θmn in Theorem 2.5.17
is surjective:
Suppose that b ∈ Z/mZ and c ∈ Z/nZ. Let x, y ∈ Z be such that mx + ny ≡ 1 mod mn, which
we can find since gcd(m, n) = 1. Then x is inverse to m in Z/nZ, and y is inverse to n in Z/mZ.
Therefore, we have that

θmn (cmx + bny) = (b, c) ∈ Z/mZ × Z/nZ.

E XAMPLE 2.5.20. Suppose we want to find a ∈ Z with

a ≡ 2 mod 7 and a ≡ 1 mod 5.

We note that 3 is an inverse of 7 modulo 5, and it is also an inverse of 5 modulo 7. So, in the
proof of surjectivity in Theorem 2.5.17, we have m = 7, n = 5, b = 2, c = 1, x = 3, and y = 3, so

cmx + bny = 1 · 7 · 3 + 2 · 5 · 3 = 51 ≡ 16 mod 35.

Therefore a = 16 is the unique integer satisfying the two congruences. Moreover, note that x and
y are independent of b and c, so we can use these x and y in solving any two congruences modulo
7 and 5.

E XAMPLE 2.5.21. We can use Theorem 2.5.17 to find isomorphisms between direct products
of cyclic groups. For instance, using this and Remarks 2.5.4 and 2.5.5, we have that

Z/12Z × Z/8Z ∼
= Z/4Z × Z/3Z × Z/8Z ∼
= Z/4Z × Z/24Z.
 2.6. GROUPS OF ISOMETRIES 41

2.6. Groups of isometries

D EFINITION 2.6.1. Let n ≥ 1. An isometry of Rn is a bijection f : Rn → Rn that preserves
distances, which is to say, for every x, y ∈ Rn , we have
| f (x) − f (y)| = |x − y|,
where the absolute value denotes the distance from 0 in Rn .
Since the property of preserving distances is preserved by composition, the following lemma
is easily seen.
L EMMA 2.6.2. The set In of isometries of Rn forms a group under composition.
P ROPOSITION 2.6.3. If f ∈ I1 , then there exists a ∈ R such that f (x) = a + x or f (x) = a − x.
R EMARK 2.6.4. The function f (x) = a + x is known as a translation, while the function
f (x) = a − x is known as a reflection (about the point x = a).
The group of isometries of R2 is a much more complicated group. We state, without proof,
the following theorem.
T HEOREM 2.6.5. Every isometry f of R2 has one of the following four forms.
i. f is a translation: there exists a ∈ R2 such that f (x) = x + a.
ii. f is a reflection: there exists a line L in R2 such that f (x) is the reflection of x across L.
iii. f is a rotation: there exist a ∈ R2 and θ ∈ [0, 2π) such that f is given by counterclockwise
rotation by θ radians about the center a.
iv. f is a glide reflection: f is the composition of a reflection and followed by a translation
by a nonzero distance in a direction parallel to the line of reflection.
R EMARK 2.6.6. Isometries of R2 are either orientation-preserving (i.e., the translations and
the rotations) or orientation-reversing (i.e., the reflections and glide-reflections). An isome-
try that preserves orientation will map the letter “S” drawn in the plane to another letter than
looks like an “S”, while an orientation-reversing isometry will map it to a backwards “S”. The
composition of two orientation-preserving or two orientation-reversing isometries is orientation-
preserving, while the composition of an orientation-preserving isometry with an orientation-
reversing isometry (in either order) is orientation-reversing.
D EFINITION 2.6.7. Suppose that X ⊆ Rn . We say that f ∈ In is a symmetry of X if for every
x ∈ Rn one has f (x) ∈ X if and only if x ∈ X.
The condition of f being a symmetry of X insures that the restriction of f to a map from X
to X is a bijection.
D EFINITION 2.6.8. For n ≥ 3, the dihedral group Dn is the group of symmetries of a regular
n-gon, which we can take to be inscribed about the unit circle around the origin of R2 with a
vertex at (1, 0).
Note that a different choice of regular n-gon in R2 simply leads to an isomorphic group.
42 2. GROUP THEORY

P ROPOSITION 2.6.9. For n ≥ 3, the dihedral group Dn is a group of order 2n, consisting of
n rotations about the origin by multiples of 2π n radians and n reflections. In the case that n is
odd, these reflections are through lines through a vertex and a midpoint of the opposite side. In
the case that n is even, n2 of these reflections are through two opposite vertices and the other n2 of
them are through midpoints of two opposite sides.
P ROOF. The above-described rotations and reflections are all easily seen to be symmetries of
the regular n-gon in question. We must see that these are the only ones. Any nonzero translation
or glide reflection moves the origin of R2 , and therefore moves the center of the polygon, hence
gannot be a symmetry. Any rotation must be about the origin, or it too will move the center, and
any rotation about the origin must take a vertex to a vertex, hence be by an angle that is a multiple
of 2π/n. Any reflection must for the same reason be a reflection across a line through the origin.
If the line determining such a reflection does not cross a midpoint or vertex, then it will move the
closest vertex on either side it passes through to a point which is less than the distance of a side
of the polygon away, hence not to another vertex. Therefore, it must pass through the origin and
either a midpoint or a vertex, and all such lines of reflection are described in the statement of the
proposition. 
P ROPOSITION 2.6.10. Let r ∈ Dn be counterclockwise rotation about the origin by 2π n radi-
ans, and let s ∈ Dn be the reflection across the x-axis. These two elements satisfy r = 1, s2 = 1,
n

and
sr = r−1 s,
and every element in Dn may be written uniquely in the form r j sk with 0 ≤ j ≤ n−1 and 0 ≤ k ≤ 1.
P ROOF. The proposition boils down to the assertions that the r j are all of the rotations and
the r j s are all of the reflections. Since r j is exactly rotation counterclockwise by 2πn j radians,
the first of these assertions holds. It is easy to see that if we first rotate across the x-axis and
then rotate counterclockwise by 2πn j radians, it is the same as reflecting across the line that is
πj
n radians counterclockwise from the x-axis. These lines pass alternately through vertices and
midpoints for even and odd j, respectively, and hence are all of the reflections. 
C OROLLARY 2.6.11. We have Dn = hr, si, for r and s as in Proposition 2.6.10.
There are many other interesting objects of which one can consider the symmetries, even in
the plane. A pattern of finite, nonzero width and height that is repeated over an over infinitely in
one direction has a symmetry group that is known as a “frieze group”, while a pattern of finite,
nonzero width and height that is repeated over and over in two non-parallel directions is known
as a “wallpaper group”.
2.7. Symmetric groups
Let n be a positive integer. In this section, we study the symmetric group Sn . Recall that an
element of Sn is a bijection σ : Xn → Xn , where Xn is the set {1, 2, . . . , n}. It is common to denote
the element σ of Sn by  
1 2 ··· n
σ= .
σ (1) σ (2) · · · σ (n)
 2.7. SYMMETRIC GROUPS 43

E XAMPLE 2.7.1. The permutation

 
1 2 3 4 5
σ=
3 5 2 1 4
is the element of S5 with values σ (1) = 3, σ (2) = 5, σ (3) = 2, σ (4) = 1, and σ (5) = 4.
This notation for permutations is amenable to composition.
E XAMPLE 2.7.2. Let σ be as in Example 2.7.1, and let
 
1 2 3 4 5
τ= .
2 1 4 3 5
To compute σ τ, we write a three-by-three matrix with the top two rows given by the notation
for τ and the next row determined by where σ takes the elements 1 through 5, i.e., we put σ (i)
below i for each i in the second row. This reads
 
1 2 3 4 5
2 1 4 3 5  .
5 3 1 2 4
The first and third rows of the latter matrix then yield σ τ:
 
1 2 3 4 5
στ = .
5 3 1 2 4
Taking inverses is even easier: one merely switches the two rows.
E XAMPLE 2.7.3. Let σ be as in Example 2.7.1. Switching its two rows, we obtain
 
3 5 2 1 4
,
1 2 3 4 5
and reordering the top row in the order 1 through 5, while preserving the columns by reordering
the bottom row in the same fashion, we obtain
 
−1 1 2 3 4 5
σ = ,
4 3 1 5 2
D EFINITION 2.7.4. Let σ ∈ Sn . The orbit of x ∈ Xn under σ is
Oσ (x) = {σ j (x) | j ∈ Z}.
E XAMPLE 2.7.5. Let σ be as in Example 2.7.1 and τ be as in Example 2.7.2. Then
Oσ (1) = {1, 2, 3, 4, 5},
while
Oτ (1) = {1, 2}, Oτ (3) = {3, 4}, Oτ (5) = {5}.
P ROPOSITION 2.7.6. The relation ∼ on Xn given by x ∼ y if and only if y = σ i (x) for some
i ∈ Z is an equivalence relation.
44 2. GROUP THEORY

P ROOF. For x ∈ X, we have x = σ 0 (x), so x ∼ x. For x, y ∈ X with x ∼ y, there exists i ∈ Z

with y = σ i (x), and then x = σ −i (y), so y ∼ x. For x, y, z ∈ X with x ∼ y and y ∼ z, we have
i, j ∈ Z such that y = σ i (x) and z = σ j (y). We then have
z = σ j (y) = σ j (σ i (x)) = σ i+ j (x),
so x ∼ z. 
By definition, Oσ (x) is the equivalence class of x under the equivalence relation defined in
Propositiion 2.7.6.
D EFINITION 2.7.7.
a. For k ≥ 2, a k-cycle in Sn is an element of Sn that has one orbit with k elements, and for
which all the other orbits have only one element each.
b. A cycle is a permutation that is a k-cycle for some k ≥ 2.
c. The length of a cycle σ is the integer k ≥ 2 such that σ is a k-cycle.
E XAMPLE 2.7.8. The element σ of Example 2.7.1 is a 5-cycle, but τ as in Example 2.7.2 is
not a cycle.
We have another notation for permutations, which depends on their orbit decomposition. We
begin with the case of a cycle.
N OTATION 2.7.9. Suppose that σ is a k-cycle, and let x be an element in its largest orbit. We
use
(x σ (x) · · · σ k−1 (x))
to denote the element σ .
R EMARK 2.7.10. When a group G is viewed as a subgroup of a symmetric group (i.e., as
consisting of permutations of some set), any non-identity element of G is called a nontrivial
element of G, and the identity element is called trivial, or the trivial element.
There are k different ways to write a k-cycle in the form of Definition 2.7.9.
E XAMPLE 2.7.11. The 5-cycle σ of Example 2.7.1 is equal to
(1 3 2 5 4) = (3 2 5 4 1) = (2 5 4 1 3) = (5 4 1 3 2) = (4 1 3 2 5).
E XAMPLE 2.7.12. Every nontrivial element of S3 is a cycle: these elements are (12), (13),
(23), (123), and (132).
However, not every nontrivial element of S4 is a cycle, as a permutation in S4 can have two
orbits of order 2.
D EFINITION 2.7.13. We say that two cycles σ and τ in Sn are disjoint if the largest orbit of
σ has empty intersection with the largest orbit of τ.
We prove the following lemma.
L EMMA 2.7.14. Any two disjoint cycles commute.
 2.7. SYMMETRIC GROUPS 45

P ROOF. Let σ , τ ∈ Sn be disjoint cycles. Let Oσ and Oτ denote their largest orbits. If x ∈ Oσ ,
then x, σ (x) ∈
/ Oτ , so
σ τ(x) = σ (x) = τσ (x).
Similarly, σ and τ commutate on elements of Oτ . Finally, if x lies in the complement of Oσ ∪ Oτ ,
then σ τ(x) = x = τσ (x). So τ and σ commute. 
R EMARK 2.7.15. We see from the proof of Lemma 2.7.14 that in any product τ of disjoint
cycles, an element x ∈ Xn will be fixed by all but at most one of the cycles, and if there is such a
cycle, the value τ(x) will equal the value of that cycle on x (and otherwise τ(x) = x).
We next see that the cycles generate Sn .
P ROPOSITION 2.7.16. Every nontrivial permutation in Sn may be written as a product of
disjoint cycles in a unique way, up to the order of the cycles.
P ROOF. Let σ ∈ Sn , and suppose that σ has m orbits of order greater than 1. Choose repre-
sentatives xi of each of these orbits for 1 ≤ i ≤ m, and set ki = |Oσ (xi )|. We then consider the
product
τ = (x1 σ (x1 ) · · · σ (x1 )k1 −1 )(x2 σ (x2 ) · · · σ (x2 )k2 −1 ) · · · (xm σ (xm ) · · · σ (xm )km −1 )
of disjoint cycles, and we claim that τ = σ . For x ∈ Xn , we have either x = σ j (xi ) for some
1 ≤ i ≤ m and 0 ≤ j ≤ ki − 1, or x lies in an orbit of order 1. In the former case, we have
τ(x) = σ j+1 (xi ) = σ (x). In the latter, we have τ(x) = x = σ (x) as well. Hence, we see that σ
may be written as a product of disjoint cycles.
We leave uniqueness primarily to the reader. We merely note that, first, the elements ap-
pearing in the individual cycles above are the elements in the orbits and so must be in any such
decomposition of σ . Given that, the individual cycles are forced to be as above by the values of
σ. 
E XAMPLE 2.7.17. Consider the permutation
 
1 2 3 4 5 6
λ= .
2 6 4 3 5 1
Then λ (1) = 2, λ (2) = 6, λ (6) = 1, so one of the cycles in the decomposition of λ is (1 2 6).
Also, λ (3) = 4 and λ (4) = 3, so another is (3 4). On the other hand, λ (5) = 5, so 5 is not moved
(or is “fixed”) by λ . We therefore have
λ = (1 2 6)(3 4).
Proposition 2.7.16 has the following interesting application to orders of elements.
P ROPOSITION 2.7.18. The order of an element of Sn is the least common multiple of the
orders of the disjoint cycles of which it is a product.
P ROOF. Suppose that σ ∈ Sn decomposes as a product of m disjoint cycles τ1 , τ2 , . . . , τm of
length k1 , k2 , . . . , km . Then Lemma 2.7.14 implies that
σ i = τ1i · · · τmi
46 2. GROUP THEORY

for every i ∈ Z, and by disjointness, the only way for σ i = e to occur is if τ ij = e for all 1 ≤ j ≤ m.
But τ j has order k j , so this will happen if and only if i is a multiple of each k j , and therefore of
the least common multiple of the k j . Hence the order of σ is this least common multiple. 
E XAMPLE 2.7.19. In S7 , the element
 
1 2 3 4 5 6 7
σ = (1 4 2 7)(3 6) =
4 7 6 2 5 3 1
has order 4. Note that
σ 2 = (1 2)(4 7) and σ 3 = (1 7 2 4)(3 6).
E XAMPLE 2.7.20. The exponent of a finite group is the least common multiple of the orders
of its elements. Since the order of an element is the least common multiple of the orders of its
cycles, which have orders 2, . . . , n, the exponent of Sn is lcm(1, 2, . . . , n). E.g., the exponent of S7
is 3 · 4 · 5 · 7 = 420.
In fact, G has a smaller generating set than the cycles, which is to say the set of transpositions.
D EFINITION 2.7.21. A transposition in Sn is a 2-cycle.
P ROPOSITION 2.7.22. Every element of Sn is a product of transpositions.
P ROOF. As every permutation is a product of cycles, we need only show that every cycle is
a product of transpositions. In fact,
(x1 x2 . . . xk ) = (x1 x2 )(x2 x3 ) · · · (xk−1 xk ),
as is easily checked. 
R EMARK 2.7.23. In fact, the symmetric group Sn is generated by transpositions of the form
(x x + 1) with 1 ≤ x ≤ n − 1. That is, the previous proposition tells us that
(x x + 1 · · · y) = (x x + 1)(x + 1 x + 2) · · · (y − 1 y)
for any 1 ≤ x < y ≤ n, and then for such x and y we have
(x y) = (y − 1 y) · (y − 2 y − 1 y) · · · (x x + 1 · · · y),
so all transpositions are contained in the subgroup generated by transpositions of the form (x x +
1), which again by Proposition 2.7.22 is all that we need.
2.8. Homomorphisms
In order to compare groups, it is useful to consider a generalization of the concept of isomor-
phism that actually has a simpler definition, as the condition of bijectivity is removed.
D EFINITION 2.8.1. Let G and G0 be groups. A homomorphism φ from G to G0 is a function
φ : G → G0
such that
φ (ab) = φ (a)φ (b)
for all a, b ∈ G.
 2.8. HOMOMORPHISMS 47

E XAMPLES 2.8.2.
a. Let ψn : Z → Z be the multiplication-by-n map, defined by ψn (a) = na for all a ∈ Z. Then
ψn is a homomorphism since
ψn (a + b) = n(a + b) = na + nb = ψn (a) + ψn (b).
b. The reduction map φn : Z → Z/nZ defined by φn (a) = a + nZ is a surjective homomor-
phism.
c. The determinant map
det : GLn (R) → R×
satisfies
det(AB) = det(A) det(B)
for all A, B ∈ GLn (R), so is a (surjective) homomorphism.
d. For m < n, we have an (injective) homomorphism ι : Sm → Sn that takes a permutation
σ of Xm = {1, 2, . . . , m} to the permutation τ ∈ Sn that satisfies τ(i) = σ (i) for 1 ≤ i ≤ m and
τ( j) = j for m < j ≤ n.
e. For A ∈ Mrs (R), we define a left-multiplication-by-A map
ψA : Mst (R) → Mrt (R)
by ψA (B) = AB for B ∈ Mst (R). By distributivity of multiplication of matrices, this is a homo-
morphism. It need not in general be injective or surjective.
f. The set
C1 (R) = { f : R → R | f is everywhere differentiable}
forms a group under addition. In fact, it is a subgroup of Maps(R, R). The derivative map
∂ : C1 (R) → Maps(R, R), ∂(f) = f0
is a homomorphism.
Here are several standard homomorphisms between groups.
D EFINITION 2.8.3. Let G and G0 be groups with identity elements e and e0 , respectively.
a. The trivial homomorphism φ : G → G0 is given by φ (g) = e0 , the identity of G0 , for all
g ∈ G.
b. The identity homomorphism idG : G → G on any group G, given by idG (g) = g for all
g ∈ G.
c. For H 6 G, we the inclusion map ιH : H → G with ιH (h) = h for all h ∈ H.
The following easily-proven lemma is useful to know.
L EMMA 2.8.4. Let G, G0 , and G00 be groups, and let φ : G → G0 and ψ : G0 → G00 be homo-
morphisms. Then
ψ ◦ φ : G → G00
is also a homomorphism.
48 2. GROUP THEORY

P ROOF. For a, b ∈ G, we have

ψ ◦ φ (ab) = ψ(φ (a)φ (b)) = ψ(φ (a))ψ(φ (b)) = ψ ◦ φ (a) · ψ ◦ φ (b).

In the following, G and G0 will be groups, and we will use e and e0 to denote their respective
identity elements.
L EMMA 2.8.5. Let φ : G → G0 be a homomorphism. Then φ (e) = e0 , and φ (g−1 ) = φ (g)−1
for all g ∈ G.
P ROOF. We have φ (e) = φ (e)φ (e) by the defining property of a homomorphism, and the
cancellation theorem then implies that φ (e) = e0 . Moreover,
φ (g) · φ (g−1 ) = φ (e) = e0 ,
again by the homomorphism property, and uniqueness of right inverses in a group then implies
that φ (g−1 ) = φ (g)−1 . The last statement then follows easily from these and the homomorphism
property. 
We have the following easy consequence.
L EMMA 2.8.6. Let φ : G → G0 be a homomorphism. Then
φ (gr11 gr22 · · · grkk ) = φ (g1 )r1 φ (g2 )r2 · · · φ (gk )rk
for any k ≥ 1, g1 , g2 , . . . , gk ∈ G, and r1 , r2 , . . . , rk ∈ Z.
P ROOF. For g ∈ G and r ∈ Z, we have gr = (g−1 )−r and
φ (g)r = φ (g−1 )−r ,
so it suffices to assume that each ri is nonnegative in the theorem. But then, by writing out the
powers as products, the result amounts simply to proving the result when each ri = 1. On the
other hand,
φ (g1 g2 · · · gk ) = φ (g1 )φ (g2 · · · gk ) = · · · = φ (g1 )φ (g2 ) · · · φ (gk )
by iterative use of the defining property of a homomorphism. 
D EFINITION 2.8.7. Let φ : G → G0 be a homomorphism.
a. The kernel of φ is the subset of G that is
ker φ = {g ∈ G | φ (g) = e0 }.
b. The image of φ is the subset of G0 that is
im φ = {φ (g) | g ∈ G}.
P ROPOSITION 2.8.8. Let φ : G → G0 be a homomorphism. Then ker φ is a subgroup of G
and im φ is a subgroup of G0 .
 2.8. HOMOMORPHISMS 49

P ROOF. Since φ (e) = e0 , we have e ∈ ker φ . Moreover, if a, b ∈ ker φ then

φ (ab) = φ (a)φ (b) = e0 · e0 = e0 ,
so ab ∈ ker φ , and if a ∈ ker φ then
φ (a−1 ) = φ (a)−1 = (e0 )−1 = e0 ,
so a−1 ∈ ker φ . It follows that ker φ 6 G.
Next, note that e0 = φ (e), so e0 ∈ G. Also, if φ (a), φ (b) ∈ im φ for some a, b ∈ G, then
φ (a)φ (b) = φ (ab) ∈ im φ
and
φ (a)−1 = φ (a−1 ) ∈ im φ .
Hence, we have that im φ 6 G0 . 
Clearly, a homomorphism φ : G → G0 is surjective if and only if im φ = G0 . On the other
hand, we have the following less obvious criterion for injectivity of φ in terms of its kernel.
P ROPOSITION 2.8.9. A homomorphism φ : G → G0 is injective if and only if ker φ = {e}.
P ROOF. If φ is injective and a ∈ ker φ , then φ (a) = e0 = φ (e), so a = e by injectivity of φ .
On the other hand, if ker φ is trivial and φ (a) = φ (b) for some a, b ∈ G, then
φ (ab−1 ) = φ (a)φ (b)−1 = e0 ,
so ab−1 = e, and therefore a = b. 
E XAMPLES 2.8.10.
a. The multiplication-by-n map ψn is injective, as ψn (a) = na = 0 if and only if a = 0. Its
image is nZ.
b. The inclusion map ιH : H → G of a subgroup H in a group G is obviously injective, and
its image is H.
c. The reduction map φn : Z → Z/nZ is surjective, and its kernel is nZ.
d. The determinant map det : GLn (R) → R× is surjective with kernel SLn (R).
e. The derivative map ∂ : C1 (R) → Maps(R, R) has kernel equal to the subgroup of constant
functions. Its image is difficult to describe explicitly, but it is not surjective.
We can also speak of the image of a subgroup under a homomorphism.
D EFINITION 2.8.11. Let H be a subgroup of G. Then image of H under a homomorphism
φ : G → G0 is
φ (H) = {φ (h) | h ∈ H}.
R EMARK 2.8.12. The set φ (H) is a subgroup of G0 , as it is the image of composition φ ◦ ιH
of the inclusion map ιH : H → G with φ .
50 2. GROUP THEORY

D EFINITION 2.8.13. The restriction of a homomorphism φ : G → G0 to H 6 G is the homo-

morphism
φ |H : H → G0
that is the composition φ ◦ ιH , where ιH is the inclusion map. In other words, φ |H (h) = φ (h) for
all h ∈ H.
We can also speak of the inverse image of a subgroup under a homomorphism.
D EFINITION 2.8.14. Let φ : G → G0 be a homomorphism. Let H 0 6 G0 . The inverse image
of H 0 under φ is
φ −1 (H 0 ) = {h ∈ H | φ (h) ∈ H 0 }.
P ROPOSITION 2.8.15. Let φ : G → G0 be a homomorphism, and let H 0 6 G0 . Then φ −1 (H 0 )
is a subgroup of G.
P ROOF. Note that φ (e) = e0 ∈ H 0 , so e ∈ φ −1 (H 0 ). Also, if a, b ∈ φ −1 (H 0 ), then φ (ab−1 ) =
φ (a)φ (b)−1 ∈ H 0 , since H 0 is a group, so ab−1 ∈ φ −1 (H 0 ). In other words, φ −1 (H 0 ) is closed
under multiplcation and inverses, so is a subgroup of G. 
E XAMPLE 2.8.16. Consider the multiplication-by-n map ψn : Z → Z. We have
m
ψn−1 (mZ) = {a ∈ Z | na ∈ mZ} = Z.
gcd(n, m)
A homomorphism is completely determined by its values on a generating set.
P ROPOSITION 2.8.17. Let S be a generating set of G, and let φ , ψ : G → G0 be homomor-
phisms. Suppose that φ (s) = ψ(s) for all s ∈ S. Then φ = ψ.
P ROOF. Since S generates G, every element of G has the form
sr11 sr22 · · · srkk
for some k ≥ 0, s1 , s2 , . . . , sk ∈ G, and r1 , r2 , . . . , rk ∈ Z. We have
φ (sr11 sr22 · · · srkk ) = φ (s1 )r1 φ (s2 )r2 · · · φ (sk )rk = ψ(s1 )r1 ψ(s2 )r2 · · · ψ(sk )rk = ψ(sr11 sr22 · · · srkk ),
as desired. 
This is a very useful property for checking whether or not two homomorphisms are equal.
On the other hand, one might be tempted to try to use it to specify a homomorphism by setting its
values on a generating set arbitrarily. This in general does not work. For instance, the only homo-
morphism φ : Z/nZ → Z is the trivial homomorphism, since one must have nφ (1) = φ (n) = 0.
That is, one can’t simply take φ (1) to be an arbitrary value of Z.
We leave it to the reader to check the following easy assertion, which in particular explains
the problem just described.
L EMMA 2.8.18. Suppose that g ∈ G has finite order and φ : G → G0 is a homomorphism.
Then the order of φ (g) divides the order of g.
 2.9. THE ALTERNATING GROUP 51

2.9. The alternating group

In this section, we study a certain subgroup of Sn , known as the alternating group. Let us
begin with its definition.
D EFINITION 2.9.1. The alternating group An is the subgroup of Sn consisting of permutations
that can be written as a product of an even number of transpositions.
E XAMPLE 2.9.2. We have (1 2 3) ∈ A3 , as (1 2 3) = (1 2)(2 3). Even more obviously, we
have (1 2)(3 4) ∈ A4 .
The assertion that An is a subgroup of Sn contained in Definition 2.9.1 is easy to verify, and
we leave it to the reader. What is not so immediate is that not every element in Sn can be written
as a product of an even number of transpositions, which is to say that An 6= Sn . For this reason
and others, we give an alternate characterization of Sn .
D EFINITION 2.9.3. A permutation matrix in GLn (R) is a matrix for which the entries are all
zero aside from one entry in each row and each column, which is 1.
P ROPOSITION 2.9.4. The set Pn of permutation matrices forms a subgroup of GLn (R) that is
isomorphic to Sn . Explicitly, define
κ : Sn → GLn (R)
by taking κ(σ ) to be the matrix with entry
(
1 if σ ( j) = i
κ(σ )i j =
0 if σ ( j) 6= i.
in the ith row and jth column. Then κ is injective with image Pn .
P ROOF. Let σ ∈ Sn . We first remark that κ(σ ) is a permutation matrix: since σ is a function,
each j is taken to exactly one i, so each column has exactly one 1, and since σ is a bijection, each
i comes from exactly one j, so each row has exactly one 1. Moreover, we have
(
n
1 if σ (τ(k)) = i
(κ(σ )κ(τ))ik = ∑ κ(σ )i j κ(τ) jk = κ(σ )iτ(k) κ(τ)τ(k)k =
j=1 0 if σ (τ(k)) 6= i,
and the latter term is exactly κ(σ τ)ik . Therefore, κ is a homomorphism. It is also clearly one-
to-one, since κ(σ ) will have a non-diagonal entry that is nonzero if σ 6= e. Finally, for any
A = (ai j ) ∈ Pn , we have A = κ(σ ), where σ ( j) is defined as the unique i such that ai j = 1, so κ
is onto. 
We next determine the image of An under the map κ of Proposition 2.9.4.
L EMMA 2.9.5. The image of An under κ is equal to the subgroup of permutation matrices
that have determinant 1.
P ROOF. Then κ((a b)), where 1 ≤ a < b ≤ n, is exactly the identity matrix after one row
operation, which is switching the ith and jth rows. As switching two rows changes the sign of a
matrix, we have det κ((a b)) = −1. As det ◦κ is a homomorphism, we have that the determinant
52 2. GROUP THEORY

of a product of a product of an even number of transpositions is 1, and the determinant of the

product of an odd number of transpositions is −1. 
For the following definition, we note that the determinant of any permutation matrix is either
1 or −1.
D EFINITION 2.9.6.
a. We define the sign function on Sn by
sign = det ◦κ : Sn → {±1},
with κ as in Proposition 2.9.4. Its value on a permutation is the sign of the permutation.
b. We say that σ ∈ Sn is even if sign(σ ) = 1 and odd if sign(σ ) = −1.
R EMARK 2.9.7. By Lemma 2.9.5, we have that An is exactly the subgroup of even permuta-
tions in Sn .
E XAMPLE 2.9.8. Cycles of even length are odd, while cycles of odd length are even.

2.10. Cosets
D EFINITION 2.10.1. Let H be a subgroup of a group G, and let a be an element of G. The
left H-coset of a is the subset of G that is
aH = {ah | h ∈ H}.
The right H-coset of a is the set
Ha = {ha | h ∈ H}.
R EMARK 2.10.2. If G is abelian, then aH = Ha for any H 6 G and a ∈ G, so we may speak
simply of cosets (as opposed to left and right cosets). If the operation on G is addition, we write
a + H for the H-coset of a.
E XAMPLE 2.10.3. The 2Z-coset 1 + 2Z of 1 in Z is the set of odd integers.
E XAMPLE 2.10.4. Let H = h(12)i 6 S3 . Then the left cosets of H are
H = (1 2)H = {(1 2), e}
(1 2 3)H = (1 3)H = {(1 3), (1 2 3)}
(1 3 2)H = (2 3)H = {(2 3), (1 3 2)}.
The property of two cosets being equal provides an equivalence relation on a group G, as
expressed in the following lemma, the proof of which follows directly from the definitions of left
and right cosets.
L EMMA 2.10.5. Let H be a subgroup of a group G, and let a, b ∈ H. The relation a ∼l b
(resp., a ∼r b) if and only if aH = bH (resp., Ha = Hb) is an equivalence relation on G, and the
equivalence class of a ∈ H under this relation is aH (resp., Ha).
C OROLLARY 2.10.6. If H 6 G, then G is the disjoint union of its distinct left (or right)
H-cosets.
 2.10. COSETS 53

We make the following remark.

L EMMA 2.10.7. Let H 6 G, and let a, b ∈ H. Then aH = bH if and only if a−1 b ∈ H, and
Ha = Hb if and only if ab−1 ∈ H.
P ROOF. Suppose a−1 b ∈ H, and set h = a−1 b. Then b = ah, so bk = a(hk) ∈ aH for every
k ∈ H, which implies bH ⊆ aH. Moreover, bh−1 = a, so aH ⊆ bH as well. Conversely, if
aH = bH, then there exists h ∈ H such that b = ah, so a−1 b ∈ H. The case of right cosets is
similar, noting that ab−1 ∈ H if and only if a = hb for some h ∈ H. 
N OTATION 2.10.8. For H 6 G, we let G/H denote the set of left cosets of H in G, and we let
H\G be the set of right cosets of H in G.
E XAMPLE 2.10.9. One might notice that the set of left cosets of nZ in Z is given the notation
Z/nZ by Corollary 2.10.8, which could in theory lead to some confusion with the group Z/nZ.
However, this is no coincidence. The cosets of nZ in Z are the a + nZ with 0 ≤ a ≤ n − 1, which
are exactly the elements of the group Z/nZ. So, Z/nZ as a group is just the set Z/nZ (of cosets)
with a particular binary operation.
D EFINITION 2.10.10. We refer to a set of representatives for the left (resp., right) H-cosets
in G as a set of left (resp., right) coset representatives.
E XAMPLE 2.10.11. The elements e, (1 2 3), (1 3 2) form a set of left coset representatives
for h(1 2)i in S3 .
E XAMPLE 2.10.12. The coset shri = hris consists of all reflections in Dn . The coset rk hsi =
{rk , rk s}
consists of the counterclockwise rotation rk about the origin by 2πk/n radians and the
reflection rk s across the line through the origin at an angle πk/n radians counterclockwise from
the x-axis. Note that
hsirk = {rk , r−k s},
and this is not rk hsi unless 2k = n.
P ROPOSITION 2.10.13. There is a canonical bijection
φ : G/H → H\G
given by φ (aH) = Ha−1 for a ∈ G.
P ROOF. First, we check that the map φ is well-defined. For a subset A of G, let us use A−1
to denote
A−1 = {a−1 | a ∈ A}.
If aH = bH, then
(2.10.1) Ha−1 = {ha−1 | h ∈ H} = {(ah−1 )−1 | h ∈ H} = {(ak)−1 | k ∈ H} = (aH)−1 .
Since aH = bH, we have that (aH)−1 = (bH)−1 , which implies noting (2.10.1) for both a and b
that
Ha−1 = (aH)−1 = (bH)−1 = Hb−1 .
Therefore, φ is well-defined.
54 2. GROUP THEORY

Next, define ψ : H\G → G/H by ψ(Ha) = a−1 H for a ∈ G. This is also well-defined, as
Ha = Hb implies that
a−1 H = (Ha)−1 = (Hb)−1 = b−1 H,
and it is clearly inverse to φ , so φ is a bijection. 
E XAMPLE 2.10.14. Let H = h(1 2)i 6 S3 . Then the bijection φ of Proposition 2.10.13 is
given by φ (H) = H,
φ ((1 2 3)H) = H(1 3 2), and φ ((1 3 2)H) = H(1 2 3).
D EFINITION 2.10.15. Let H be a subgroup of G.
a. If there are finitely many left cosets of H in G, then we say that H is of finite index in G,
and otherwise H is of infinite index.
b. If H is of finite index in G, then we define the index [G : H] of H in G to be the number of
left cosets of H in G.
E XAMPLES 2.10.16.
a. The index [S3 : H] of H = h(12)i in S3 is 3.
b. The index [Z : nZ] of nZ in Z is n.
c. The group Z is not of finite index in Q.
d. We have [Dn : hsi] = n and [Dn : hri] = 2.
e. We have [Sn : An ] = 2 for n ≥ 2, and the nonidentity coset is the set of odd permutation
which equals, e.g., (1 2)An .
E XAMPLE 2.10.17. For any group G, we have [G : G] = 1, and if G is finite, we have [G :
hei] = |G|
R EMARK 2.10.18. By Proposition 2.10.13, we could just as well have used right cosets
instead of left cosets in the definition of the index.
T HEOREM 2.10.19 (Lagrange’s theorem). Let H be a subgroup of a finite group G. Then we
have
|G| = [G : H]|H|.
In particular, the order of H divides the order of G.
P ROOF. Since G is finite, so is H, and every coset aH is in bijection with H via the map
θ : H → aH with θ (h) = ah. As G is the disjoint union of its left cosets, we have
|G| = ∑ |aH| = ∑ |H| = [G : H]|H|.
aH∈G/H aH∈G/H


We can use Lagrange’s theorem to determine the indices of subgroups when we know both
the orders of the group and of the subgroup. Here is an example.
 2.10. COSETS 55

E XAMPLE 2.10.20. There is an injective homomorphism

ι : Sn−1 → Sn
as in Example 2.8.2 that takes a permutation of the set Xn−1 ⊂ Xn to the permutation that has
the same values on the elements of Xn−1 and which fixes n. Using ι, we may identify Sn−1 with
the isomorphic subgroup ι(Sn−1 ) of Sn consisting of elements σ ∈ Sn with σ (n) = n. Under this
identification, we have
n!
[Sn : Sn−1 ] = = n.
(n − 1)!
E XAMPLE 2.10.21. Since [Sn : An ] = 2 for n ≥ 2, we have |An | = 21 n! for such n.
C OROLLARY 2.10.22. Let G be a finite group. Then the order of every element of G divides
the order of G.
E XAMPLE 2.10.23. We have already seen that the orders of the subgroups of Z/nZ are ex-
actly the positive divisors of n.
E XAMPLE 2.10.24. According to the corollary, every element of Sn should have order divid-
ing n!. In fact, we already know from Example 2.7.20 that every element has order dividing the
least common multiple of 1, 2, . . . , n, which clearly divides n!.
Finally, we mention the following interesting corollary of Lagrange’s theorem.
C OROLLARY 2.10.25. Every group of prime order is cyclic.
P ROOF. Let G be a group of order a prime p. If g ∈ G is not the identity, it must generate
a nontrivial subgroup of G, which can only have order p by Lagrange’s theorem, and therefore
must be G. That is, G = hgi, finishing the proof. 
We note that the index satisfies the following multiplicative property.
P ROPOSITION 2.10.26. Let H and K be subgroups of G with K 6 H. Then K has finite index
in G if and only if H has finite index in G and K has finite index in H. Moreover, if K has finite
index in G, then we have
[G : K] = [G : H][H : K].
P ROOF. Let S be a set of H-coset representatives in G and T be a set of K-coset representa-
tives in H. Consider the set
U = {st | s ∈ S,t ∈ T }.
We claim that U is a set of left K-coset representatives in G. For this, note that if g ∈ G, we may
choose s ∈ S with g = sh for some h ∈ H, and we may choose t ∈ T with h = tk for some k ∈ K.
In other words, g ∈ stK, so gK = stK. This proves the claim.
Next, note that if stK = s0t 0 K with s, s0 ∈ S and t,t 0 ∈ T , then since tK ⊂ H and t 0 K ⊂ H, we
have sH = s0 H, so s = s0 . But then stK = st 0 K, so tK = t 0 K, and therefore t = t 0 . In other words,
we have shown that the map S × T → U given by (s,t) 7→ st is a bijection. Thus, S and T are
finite if and only if U is, and if they are, then |U| = |S||T |, as desired. 
56 2. GROUP THEORY

R EMARK 2.10.27. Proposition 2.10.26 implies Lagrange’s theorem by taking the subgroup
K to be the trivial subgroup. That is, for a finite group G and subgroup H, we have
|G| = [G : hei] = [G : H][H : hei] = [G : H]|H|.

2.11. Conjugation
D EFINITION 2.11.1. Let G be a group.
a. Let a, x ∈ G. Then axa−1 is known as the conjugate of x by a.
b. We say that an element x ∈ G is conjugate to an element y ∈ G if there exists a ∈ G with
y = axa−1 .
R EMARK 2.11.2. One might recall the related notion of similar matrices in Mn (R).
L EMMA 2.11.3. The relation ∼ on G given by x ∼ y if and only if x is conjugate to y is an
equivalence relation on G.
P ROOF. We have x = exe−1 , so x ∼ x. If x ∼ y, then there exists a ∈ G with y = axa−1 , which
implies
x = a−1 ya = a−1 y(a−1 )−1 ,
so y ∼ x. Finally, if x ∼ y and y ∼ z, then there exist a, b ∈ G with y = axa−1 and z = byb−1 , so
z = byb−1 = b(axa−1 )b−1 = (ba)x(ba)−1 ,
and z ∼ x. 
D EFINITION 2.11.4. The set
Cx = {axa−1 | a ∈ G}
of elements of that are conjugate to x ∈ G is called the conjugacy class of x.
As a consequence of the fact that conjugacy forms an equivalence relation, any two conjugacy
classes are either disjoint or equal.
E XAMPLE 2.11.5. Let a = ri and b = ri s in Dn . Then we have
ar j a−1 = ri · r j · r−i = r j , a(r j s)a−1 = ri · r j s · r−i = r2i+ j s,
br j b−1 = ri s · r j · sr−i = r− j , b(r j s)b−1 = ri s · r j s · sr−i = r2i− j s.
Therefore, we have Cri = {ri , r−i } for all i ∈ Z, while
Cs = {r2i s | i ∈ Z},
which is all reflections if n is odd, but only half of them if n is even, in which case the remaining
conjugacy class is Crs .
D EFINITION 2.11.6. For any a ∈ G, the conjugation map is the function defined by
γa : G → G, γa (x) = axa−1
for x ∈ G.
 2.11. CONJUGATION 57

R EMARK 2.11.7. The process of applying a map γa to an element of G is referred to as

conjugation.
L EMMA 2.11.8. For a ∈ G, the conjugation map γa is an isomorphism.
P ROOF. For x, y ∈ G, we have
γa (x)γa (y) = (axa−1 )(aya−1 ) = a(xy)a−1 = γa (xy).
Also, γa−1 is the inverse function to γa , so γa is bijective. 
In particular, we have
a(x1 x2 · · · xs )a−1 = ax1 a−1 · ax2 a−1 · · · · · axs a−1 and (axa−1 )−1 = ax−1 a−1
for any elements of G.
One very interesting example is conjugation in Sn . We describe this in the case of a cycle.
L EMMA 2.11.9. Let σ ∈ Sn , and let τ = (x1 x2 · · · xk ) ∈ Sn be a k-cycle. Then
σ τσ −1 = (σ (x1 ) σ (x2 ) · · · σ (xk )).
P ROOF. We check this as functions. Let x ∈ Xn . Then
 
xi+1
 if σ −1 (x) = xi , 1 ≤ i ≤ k − 1 xi+1 if x = σ (xi ), 1 ≤ i ≤ k − 1
−1 −1
τ(σ (x)) = x1 if σ (x) = xk = x1 if x = σ (xk )
σ −1 (x) otherwise
 σ −1 (x) otherwise


It follows that 
σ (xi+1 ) if x = σ (xi ), 1 ≤ i ≤ k − 1

−1
σ (τ(σ (x))) = σ (x1 ) if x = σ (xk )

x otherwise,
but the latter just the value of the cycle (σ (x1 ) σ (x2 ) · · · σ (xk )) on x. 
R EMARK 2.11.10. We can use Lemma 2.11.9 to compute the conjugate of any permutation
τ by a permutation σ , as τ can be written as a product of cycles, τ = τ1 τ2 · · · τs and
σ τσ −1 = σ τ1 σ −1 · σ τ2 σ −1 · · · σ τs σ −1 .
E XAMPLE 2.11.11. In S7 , we have
(1 2 3 4) · (2 3 7)(4 5) · (1 2 3 4)−1 = (3 4 7)(1 5).
That is, we have replaced the entries in the permutation (2 3 7)(4 5) that (1 2 3 4) moves to the
values it takes them to.
D EFINITION 2.11.12. Write τ ∈ Sn as a product of disjoint cycles τ = τ1 τ2 · · · τs with lengths
2 ≤ k1 ≤ k2 ≤ · · · ks ≤ n. Then (k1 , k2 , . . . , ks ) is said to be the cycle type of τ.
We may use Lemma 2.11.9 to prove the following.
P ROPOSITION 2.11.13. Two elements in Sn are conjugate if and only if they have the same
cycle type.
58 2. GROUP THEORY

P ROOF. By Lemma 2.11.9 and Remark 2.11.10, any two conjugate permutations must have
the same cycle type. On the other hand, we will show that every permutation with cycle type
(k1 , k2 , . . . , ks ) is conjugate to a particular permutation α depending only on (k1 , k2 , . . . , ks ). Since
conjugacy of elements forms an equivalence relation, we will then have the result.
We first define α. Let mi = k1 + k2 + · · · + ki−1 for each 1 ≤ i ≤ s + 1. In particular, m1 = 0.
Define αi ∈ Sn by
αi = (mi + 1 mi + 2 · · · mi + ki ).
Then the αi are disjoint cycles, and we set α = α1 α2 · · · αs .
Now suppose that τ has cycle type (k1 , k2 , . . . , ks ). We must show that τ is conjugate to α.
Write
τ = τ1 τ2 · · · τs ,
where the τi are disjoint cycles:
τi = (xmi +1 xmi +2 · · · xmi +ki ).
Now choose any σ ∈ Sn such that σ (i) = xi for each 1 ≤ i ≤ ms . (For each ms < i ≤ n, we are free
to successively choose the σ (i) as i increases to be any values between 1 and n not yet chosen.)
Then
σ αi σ −1 = (σ (mi + 1) σ (mi + 2) · · · σ (mi + ki )) = (xmi +1 xmi +2 · · · xmi +ki )
for each 1 ≤ i ≤ s, so σ ασ −1 = τ, as desired. 
R EMARK 2.11.14. In other words, the conjugacy class of a permutation is all permutations
with that same cycle type.

2.12. Normal subgroups

We now focus our attention on a very special class of subgroups of a group.
D EFINITION 2.12.1. A subgroup N of a group G is said to be normal if aN = Na for every
a ∈ G. We also say that N is normal in G, and we write N P G to indicate this.
N OTATION 2.12.2. If N is a proper normal subgroup of a group G, then we write N C G.
Of course, the trivial subgroup and the improper subgroup of a group G are, by this definition,
normal subgroups. If G is abelian, then every subgroup is normal. We also have the following.
L EMMA 2.12.3. If H is an index 2 subgroup of a group G, then H is normal in G.
P ROOF. Since G has just two left H-cosets, one of which is H, the other must be the comple-
ment of H in G. The same holds for the right H-cosets, hence the result. 
Here are a couple more examples, the second of which we can see from Lemma 2.12.3.
E XAMPLES 2.12.4.
a. We have hri i C Dn for all n ≥ 3 and i ∈ Z. To see this, note that ri commutes with every
r j , while
r j s · ri = r−i · r j s,
and r−i ∈ hri i.
 2.12. NORMAL SUBGROUPS 59

b. We have that An C Sn for all n ≥ 2.

R EMARK 2.12.5. If H is a subgroup of G that is not normal, then aH 6= Ha for some a ∈ G.
But note that a ∈ aH ∩ Ha, and distinct right cosets are disjoint, so aH cannot equal any right
coset of G. Therefore, there exists a left coset that is not equal to a right coset.
We can give another characterization of normal subgroups using conjugation. For this, we
use the following definition.
D EFINITION 2.12.6. Let H be a subgroup of G. For a ∈ G, the conjugate subgroup of H by
a is the set of conjugates of a:
aHa−1 = {aha−1 | h ∈ H}.
That the conjugate subgroup is, in fact, a subgroup is a corollary of Lemma 2.11.8:
C OROLLARY 2.12.7. For H 6 G and a ∈ G, we have that aHa−1 6 G.
P ROOF. We have aHa−1 = γa (H), and the latter is the image of a (sub)group under a homo-
morphism to G, hence a subgroup of G. 
E XAMPLES 2.12.8.
a. In Dn , we have
shris−1 = hri and rhsir−1 = hr2 si.
b. In S4 , we have
(2 3 4)h(1 2), (3 4)i(2 3 4)−1 = h(1 3), (2 4)i.
L EMMA 2.12.9. A subgroup N of G is normal if and only if aNa−1 = N for all a ∈ G.
P ROOF. Let a ∈ G. The function θa : G → G given by right multiplication by a, i.e., θa (g) =
ga is a bijection by the cancellation theorem. Moreover, θa restricts to bijections aNa−1 → aN
and N → Na, so aNa−1 = N if and only if aN = Na. 
C OROLLARY 2.12.10. A subgroup N of G is normal if and only if ana−1 ∈ N for all a ∈ G
and n ∈ N.
P ROOF. The only if direction follows from Lemma 2.12.9. On the other hand, the condition
ana−1 ∈ N for all a ∈ G and n ∈ N clearly implies that aNa−1 6 N, which we have seen implies
aN 6 Na. But it also means Na−1 6 a−1 N for all a ∈ G, and this equation for a−1 reads Na 6 aN,
which means aN = Na, as desired. 
The following proposition gives an extremely useful criterion for a group to be normal.
P ROPOSITION 2.12.11. Let φ : G → G0 be a homomorphism of groups. Then ker φ is a
normal subgroup of G.
P ROOF. Let a ∈ G and n ∈ ker φ . Then we have
φ (ana−1 ) = φ (a)φ (n)φ (a)−1 = φ (a)φ (a)−1 = e0 ,
where e0 is the identity of G0 . In other words, ana−1 ∈ N, so N is normal by Corollary 2.12.10.

60 2. GROUP THEORY

E XAMPLE 2.12.12. The special linear group SLn (R) is a normal subgroup of the general
linear group GLn (R), as it is the kernel of the determinant map.
Here are two other examples.
E XAMPLE 2.12.13. Consider the group
  
a b ×
Aff(R) = a ∈ R ,b ∈ R ,
0 1
which is a subgroup of GL2 (R). Then the set
  
1 t
N= t ∈R
0 1
is a normal subgroup of G. To see this is either the following calculation
−1    −1
−a−1 b
     
a b 1 t a b a at + b a 1 at
= = ∈ N,
0 1 0 1 0 1 0 1 0 1 0 1
or much more simply, that N is the kernel of the restriction of the determinant map to G.
On the other hand, the subgroup
  
x 0 ×
H= |x∈R
0 1
is not a normal subgroup of G. In fact,
−1    −1
−a−1 b
     
a b x 0 a b ax b a x (1 − x)b
= = ,
0 1 0 1 0 1 0 1 0 1 0 1
and the latter element is not in H if x 6= 1 and b 6= 0.
E XAMPLE 2.12.14. Let n ≥ 3. Let τ ∈ Sn be a k-cycle with k ≥ 2. Then hτi is not normal
in Sn unless n = k = 3. If hτi were normal in Sn , then every conjugate of τ would have to be a
nontrivial power of τ, of which there are k − 1. On the other hand, the conjugates of τ are exactly
the k-cycles, of which there are
n! (n − 1)!
≥ ≥ n − 1,
k(n − k)! (n − k)!
which forces n = k in order that hτi might possibly be normal. But for n = k, we have
n!
= (n − 1)!,
k(n − k)!
and the latter term is greater than n if n > 3, so n = k = 3. On the other hand, we have already
seen that
A3 = h(1 2 3)i = h(1 3 2)i C S3 .
 2.13. QUOTIENT GROUPS 61

2.13. Quotient groups

The sets of left and right cosets of a normal subgroup are of course the same set, and in this
section we prove that this set can be given the structure of a group.
D EFINITION 2.13.1. Let A and B be subsets of a group G. Then we define the product of A
and B as
AB = {ab | a ∈ A, b ∈ B}
Moreover, if g is an element of G, we define
gA = {ga | a ∈ A} and Ag = {ag | a ∈ A}.
R EMARK 2.13.2. If H is a subgroup of G, then HH = H.
T HEOREM 2.13.3. Let N be a normal subgroup of a group G. Then the product of cosets as
subsets of G provides a binary operation on G that satisfies
aN · bN = aNbN = abN.
Moreover, G/N is a group under this operation.
P ROOF. Let a, b ∈ G. Since N is normal in G, we have Nb = bN. Therefore, as sets we have
aNbN = a(Nb)N = a(bN)N = abNN = abN,
as desired. The associativity of the operation is a direct consequence of the associativity of the
operation on G, as (ab)cN = a(bc)N for any a, b, c ∈ G. Then N = eN is easily seen to be the
identity element of G/N, and the inverse of aN is a−1 N, since
aN · a−1 N = aa−1 N = N = a−1 N · aN.
Therefore G/N is a group under this operation. 
D EFINITION 2.13.4. Let N be a normal subgroup of a group G. The quotient group of G by
N is the group that is the set G/N with the binary operation aN · bN = abN for a, b ∈ G.
E XAMPLES 2.13.5.
a. The set of cosets of nZ in Z is a group under addition of cosets, and it is exactly the group
Z/nZ constructed before, since
(a + nZ) + (b + nZ) = (a + b) + nZ
by definition.
b. The quotient groups Dn /hri for n ≥ 3 and Sn /An for n ≥ 2 are all cyclic groups of order 2.
c. Suppose that n ≥ 4 is even. Let H = hr2 i C Dn . Then [Dn : H] = 4, and the four distinct
cosets are H, rH, sH, and rsH. Since the square of each of these cosets is H, we have an
isomorphism between Dn /H and the Klein four-group
Dn /H ∼
= Z/2Z × Z/2Z
that takes rH to (1, 0) and sH → (0, 1).
62 2. GROUP THEORY

R EMARK 2.13.6. If N is not normal in G, then G/N is not a group under the product of left
cosets. In fact, for a, b ∈ N, the set aNbN will not in general be a left coset. E.g., if one takes
b = a−1 and a to be such that aNa−1 contains an element not in N, then aNa−1 N will contain but
not equal N, so it is not a left coset.
Moreover, if one simply tries to define aN ·bN = abN, then the resulting operation is not well-
defined, as it depends on the choice of coset representatives. E.g., assuming it were well-defined
and again taking a and b as above, we can find n ∈ N such that ana−1 ∈ / N, so
N = aN · a−1 N = anN · a−1 N = ana−1 N 6= N,
which contradicts well-definedness.
Note that the function G → G/N that takes g ∈ G to its N-coset gN is a homomorphism by
definition of the quotient group. We give it a name.
D EFINITION 2.13.7. Let N be a normal subgroup of a group G. The quotient map πN : G →
G/N is the homomorphism defined by πN (g) = gN for g ∈ G.
R EMARK 2.13.8. The kernel of the quotient map πN : G → G/N is N.
C OROLLARY 2.13.9. A subgroup N of G is normal if and only if there exists a group G0 and
a group homomorphism φ : G → G0 such that N = ker φ .
We end with the following result on the subgroups of quotient groups.
P ROPOSITION 2.13.10. Let G be a group and N be a normal subgroup of G. Then the
subgroups of G/N are exactly the quotient groups H/N, where H is a subgroup of G containing
N. Moreover, such a subgroup H of G is normal in G if and only if H/N is normal in G/N.
P ROOF. We first note that if H is a subgroup of G containing N, then N is normal in H, so
we may form the quotient group H/N. Its binary operation agrees with the restriction of the
operation on G/N (multiplication of N-cosets), so it is a subgroup of G/N.
Conversely, if Q is a subgroup of G/N, then set
H = {h ∈ G | hN ∈ Q}.
Then nN = N ∈ Q, in that it is the identity element of G/N and Q is a subgroup, so N ⊆ H.
That H is a subgroup of G follows directly from the fact that Q is a subgroup of G/N, since if
hN, kN ∈ Q, then hN · (kN)−1 ∈ Q, so hk−1 N ∈ Q, which means that hk−1 ∈ H.
Finally, let H be a subgroup of G containing N. Then, for a ∈ G, we have
aN · (H/N) = {ahN | h ∈ H} = aH
and
(H/N) · aN = {Nha | h ∈ H} = Ha,
so H/N P G/N if and only if H P G. 
T HEOREM 2.13.11 (First Isomorphism Theorem). Let φ : G → G0 be a homomorphism of
groups. Then the function
φ̄ : G/ ker φ −∼
→ im φ , φ̄ (a ker φ ) = φ (a)
for a ∈ G is a well-defined group isomorphism.
 2.13. QUOTIENT GROUPS 63

P ROOF. First, note that if a, b ∈ G are such that a ker φ = b ker φ , then a = bk for some
k ∈ ker φ , so
φ̄ (a ker φ ) = φ (bk) = φ (b)φ (k) = φ (b) = φ̄ (b ker φ ),
and hence φ̄ is well-defined. Moreover, if φ̄ (a ker φ ) = 0, then φ (a) = 0, so a ∈ ker φ , and
therefore we have that φ̄ is injective. Since φ has image im φ , so does φ̄ , and hence φ̄ is surjective
by definition. 
R EMARK 2.13.12. We have that φ = ιim φ ◦ φ̄ ◦ πker φ , as is represented in the following dia-
gram:
G
φ
/ G0 .
=
πker φ # ιim φ
G/ ker φ
φ̄
/ im φ

E XAMPLE 2.13.13. The determinant map det : GLn (R) → R× induces an isomorphism
det : GL (R)/SL (R) −∼ → R× .
n n

To give another example, we make the following definition.

D EFINITION 2.13.14. Let G1 , G2 , . . . , Gk be groups. The ith projection map is the surjective
homomorphism
k
πi : ∏ G j → Gi , πi (g1 , g2 , . . . , gk ) = gi .
j=1

R EMARK 2.13.15. The projection map πi of Definition 2.13.14 has kernel

( )
k
ker πi = (g1 , g2 , . . . , gk ) ∈ ∏ G j g j = e j for all j 6= i ,
j=1
where ei is the identity element of Gi . By the first isomorphism theorem, πi induces an isomor-
phism
 k 
π̄i : ∏ G j / ker πi −∼ → Gi .
i=1
For instance, if k = 2 and i = 1, we can think of π̄i as an isomorphism
G1 × G2 ∼
−→ G1 .
{e1 } × G2
E XAMPLE 2.13.16. We have
Z/4Z × Z/4Z ∼
= Z/2Z × Z/2Z.
h2i × h2i
To see this, define a map φ : Z/4Z × Z/4Z → Z/2Z × Z/2Z by φ (a, b) = (a, b). Then φ is
surjective with kernel h2i × h2i, so the first isomorphism theorem applied to φ provides the
isomorphism.
 CHAPTER 3

Ring theory

3.1. Rings
In this section, we define rings and fields. These are sets with two binary operations, known
as addition and multiplication.
D EFINITION 3.1.1. Let R be a set with a pair (+, ·) of binary operations. We say that R
satisfies the left distributive law (with respect to + and ·) if
a · (b + c) = (a · b) + (a · c)
for all a, b, c ∈ R, and we say that R satisfies the right distributive law if
(a + b) · c = (a · c) + (b · c)
for all a, b, c ∈ R.
The distributive law being one of the standard axioms of arithmetic, it is satisfied by many
common objects, such as Z, Q, R, C, and so on. We give one less standard example.
E XAMPLE 3.1.2. The set Maps(R, R) satisfies the left and right distributive laws with respect
to the pair of operations (+, ·). It satisfies the right distributive law with respect to (+, ◦) and
(·, ◦), where ◦ is composition.
We now define a ring.
D EFINITION 3.1.3. A set R with a pair (+, ·) of binary operations is a ring if
i. R is an abelian group under +,
ii. the binary operation · is associative,
iii. R has an identity element 1 under ·, and
iv. R satisfies the left and right distributive laws.
R EMARK 3.1.4. When + and · are used to denote the binary operations of a ring, we refer +
as addition and · as multiplication. Unless otherwise stated, the operations of R will be denoted
+ and ·.
R EMARK 3.1.5. As in the case of groups, we often write ab for a · b for a, b in a ring R. We
also use a · b + c to denote (a · b) + c for a, b, c ∈ R.
E XAMPLES 3.1.6.
a. The sets Z, Q, R, and C are all rings with respect to the usual operations of addition and
multiplication.
65
66 3. RING THEORY

b. The sets Maps(R, R) and Mn (R) for n ≥ 1 are also rings with respect to addition and
multiplication.
c. The set nZ for n ≥ 1 is a ring with respect to addition and multiplication.
d. The set Z/nZ is a ring with respect to its operations of addition and multiplication.
R EMARK 3.1.7. Since the first binary operation on a ring R is denoted +, the identity element
is denoted 0 as usual, and the additive inverse of a ∈ R is denoted −a. The sum of n copies of a
is denoted na for n ≥ 1, and −(na) is also denoted −na.
We have the following properties in any ring.
L EMMA 3.1.8. Let R be a ring, and let a, b ∈ R. Then we have
a. 0 · a = a · 0 = 0,
b. a · (−b) = (−a) · b = −ab, and
c. (−a) · (−b) = ab.
P ROOF.
a. We have
0 · a + b · a = (0 + b) · a = b · a
by the right distributive law and the fact that 0 is an additive identity. Therefore, the Cancellation
theorem tells us that 0 · a = 0. Similarly, a · 0 = 0 using the left distributive law instead of the
right.
b. We have
a · (−b) + ab = a · (−b + b) = a · 0 = 0
by the left distributive law, the definition of the additive inverse, and part a. The other equality is
similar.
c. This follows from part b, which tells us that
(−a) · (−b) = −(a · (−b)) = −(−(ab)) = ab.

On a set with one element, there is only one possible binary operation, and using it as both
addition and multiplication turns that set into a ring.
D EFINITION 3.1.9. The zero ring is the ring {0}. We say that a ring R is a nonzero ring if R
has more than one element.
That a ring R has an identity under · is to say exactly that there is an element 1 ∈ R with
1 · a = a · 1 = a for all a ∈ R. By Lemma 2.1.4, the multiplicative identity 1 in a ring is unique.
E XAMPLES 3.1.10. The rings Z, Q, R, C, Maps(R, R), Mn (R) for n ≥ 1, Z/nZ for n ≥ 1,
and {0} are all rings. However, nZ is not a ring for n ≥ 2.
R EMARK 3.1.11. One easily checks that (na) · (mb) = (nm)ab for n, m ∈ Z and a, b ∈ R for
any ring R. One has (n · 1) · (m · 1) = nm · 1. We often denote n · 1 by n, though we remark that it
is possible that n · 1 = m · 1 for n 6= m, as will happen in any finite ring, for instance.
 3.1. RINGS 67

R EMARK 3.1.12. If R is a ring with 1 = 0, then x = 1 · x = 0 · x = 0 for all x ∈ R, so R is the

zero ring.
We now introduce the notion of a subring of a ring, which does not play quite as prominent
of a role in ring theory as does the notion of a subgroup of a group in group theory.
D EFINITION 3.1.13. A subring S of a ring R is a subset of R containing 1 that is a ring with
respect to the restrictions to S of the binary operations of addition and multiplication on R.
We leave it to the reader to check the following.
L EMMA 3.1.14. A subset S of a ring R is a subring if it is closed under the operations of
addition and multiplication on R, contains 0 and 1, and contains −a for all a ∈ S.
Clearly, the property of being a subring is a transitive one.
E XAMPLES 3.1.15.
a. The ring nZ is not a subring of Z, as it does not contain 1.
b. The ring Z is a subring of Q, which is in turn a subring of R, which is in turn a subring of
C.
Most of the study of ring theory is focused on commutative rings.
D EFINITION 3.1.16. A ring R is a commutative ring if multiplication on R is commutative.
We then say that the ring R is commutative.
D EFINITION 3.1.17. A ring R that is not commutative is a noncommutative ring.
E XAMPLES 3.1.18. The rings Z, Q, R, C, Maps(R, R), Z/nZ for n ≥ 1, and {0} are all
commutative rings. However, Mn (R) is a noncommutative ring for all n ≥ 2.
The notion of a field is really just a special case of the notion of a ring, but it is an important
one.
D EFINITION 3.1.19. A field is a nonzero commutative ring for which every nonzero element
has a multiplicative inverse.
In other words, a field is a nonzero commutative ring for which the nonzero elements form a
group under multiplication (in fact, an abelian group).
D EFINITION 3.1.20. A subfield of a field F is a subring of F that is a field.
E XAMPLES 3.1.21.
a. The rings Q, R, and C are fields. Of course, Q is a subfield of R and C, and R is a subfield
of C.
b. The ring Z is not a field.
The analogous object to a field in the more general theory of possibly noncommutative rings
is known as a division ring.
68 3. RING THEORY

D EFINITION 3.1.22. A division ring (or skew field) D is a nonzero ring such that every
nonzero element is invertible under multiplication.
Clearly, all fields are division rings. As with fields, we have multiplicative groups of division
rings, which no longer need be abelian.
D EFINITION 3.1.23. The group of nonzero elements in a division ring D is known as the
multiplicative group of D and is denoted D× .
We end this section with one example of a noncommutative division ring.
D EFINITION 3.1.24. The ring of quaternions H is the set of distinct elements a + bi + c j + dk
with a, b, c, d ∈ R, together with addition defined by
(a + bi + c j + dk) + (a0 + b0 i + c0 j + d 0 k) = (a + a0 )i + (b + b0 ) j + (c + c0 )k
and multiplication defined by
(a + bi + c j + dk) · (a0 + b0 i + c0 j + d 0 k) = (aa0 − bb0 − cc0 − dd 0 )
+ (ab0 + ba0 + cd 0 − dc0 )i + (ac0 − bd 0 + ca0 + db0 ) j + (ad 0 + bc0 − cb0 + da0 )k
for a, b, c, d, a0 , b0 , c0 , d 0 ∈ R.
R EMARK 3.1.25. The ring H is an R-vector space with basis 1, i, j, k, where
α · (a + bi + c j + dk) = αa + (αb)i + (αc) j + (αd)k
for α, a, b, c, d ∈ R. Note that we have i j = k = − ji, jk = i = −k j, ki = j = −ik, and i2 = j2 =
k2 = −1 in H.
T HEOREM 3.1.26. The quaternion algebra is a division ring.
P ROOF. We give only a sketch. Distributivity is a direct consequence of the definitions of
the operations of addition and multiplication. In fact, it is also easy to see that α · xy = x · (αy) =
(αx) · y for α ∈ R and x, y ∈ H. Using the distributive law and the latter fact, associativity of
multiplication follows from a check of associativity on the subset {i, j, k} of H× . Finally, any
nonzero a + bi + c j + dk ∈ H has inverse
(a + bi + c j + dk)−1 = (a2 + b2 + c2 + d 2 )−1 (a − bi − c j − dk),
so H is a division ring. 

3.2. Families of rings

In this section, we consider various sorts, or families, of rings one can construct out of other
rings. We begin with matrix rings.
D EFINITION 3.2.1. If R is a nonzero ring, the matrix ring Mn (R) consisting of n-by-n matrices
with entries in R is the set with the addition (ai j ) + (bi j ) = (ai j + bi j ) and multiplication
!
n
(ai j ) · (bi j ) = ∑ aik bk j .
k=1
 3.2. FAMILIES OF RINGS 69

We leave the proof of the following to the reader.

L EMMA 3.2.2. Let R be a ring and n ≥ 1. Then Mn (R) is a ring.
L EMMA 3.2.3. The ring Mn (R) is noncommutative if R is a ring and n ≥ 2.



P ROOF. Suppose n = 2. Let A = 00 10 and B = 01 00 . Then AB = I, the identity matrix in
M2 (R), while BA = 0. The general case follows from the case n = 2 by taking matrices that
contain the same entries as A and B in their upper lefthand corners and are zero in all other
entries. 
Another important class of rings is the polynomial rings.
D EFINITION 3.2.4. Let R be a ring, and fix an indeterminate (i.e., a symbol) x. The poly-
nomial ring R[x] with R-coefficients is the set of finite formal (i.e., two are different if they are
written differently) sums of powers of x with coefficients in R, i.e.,
( )
∞
R[x] = ∑ aixi | ai ∈ R for all i ≥ 0, ai = 0 for all i > N for some N ≥ 0 .
i=0
together with the binary operations of addition and multiplication given by
∞ ∞ ∞
∑ aixi + ∑ b j x j = ∑ (ai + bi)xi,
i=0 j=0 i=0
! ! !
∞ ∞ ∞ k
∑ ai x i · ∑ b jx j = ∑ ∑ aibk−i xk .
i=0 j=0 k=0 i=0

An element f = ∑∞ i
i=0 ai x of R[x] is called a polynomial, the ai are referred to as coefficients, and
x is called a variable.
R EMARK 3.2.5. If ai = 0 for all i > N, then we more commonly write ∑N i
i=0 ai x for f =
i
∑∞
i=0 ai x . We will also sometimes write
f = a0 + a1 x + a2 x2 + · · · + an xn ,
identifying x0 with “1” and x1 with “x”.
D EFINITION 3.2.6. The degree deg f of a nonzero polynomial f = ∑∞ i
i=0 ai x ∈ R[x] is the
smallest integer N such that ai = 0 for all i > N. When needed, we consider the degree of 0 to be
−∞.
R EMARK 3.2.7. A polynomial of degree 0 is said to be constant, a polynomial of degree 1 is
linear, a polynomial of degree 2 is quadratic, followed by cubic, quartic, quintic, and so forth.
D EFINITION 3.2.8. If f is a polynomial of degree n ≥ 0, then its leading coefficent is the
coefficient of xn in f . If constant coefficient is the coefficient of x0 = 1.
E XAMPLE 3.2.9. The polynomials 2 + 3x − x2 and 1 + x are elements of Z[x]. One has, as
usual,
(2 + 3x − x2 ) · (1 + x) = 2 + (2 + 3)x + (3 − 1)x2 − x3 = 2 + 5x + 2x2 − x3 .
70 3. RING THEORY

The following is a direct consequence of the definitions of addition and multiplication in

polynomial rings.
L EMMA 3.2.10. Let R be a ring, and let f , g ∈ R[x] be polynomials. Then deg f g ≤ deg f ·
deg g. Moreover, we have
deg( f + g) ≤ max{deg f , deg g},
and equality holds in the last statement if deg f 6= deg g.
D EFINITION 3.2.11. The polynomials a = a + 0 · x + 0 · x2 + · · · for a ∈ R are referred to as
constant polynomials. These are exactly 0 and the polynomials of degree 0. The set of constant
polynomials forms a subring of R[x], which we also denote R.
We leave it to the reader to check the following.
L EMMA 3.2.12. Let R be a ring. Then the polynomial ring R[x] is in fact a ring.
R EMARK 3.2.13. The ring R[x] is commutative if and only if R is commutative. The 1 is a
multiplicative identity in R, then 1 is a multiplicative identity in R[x] as well.
We may also consider polynomial rings in several variables.
D EFINITION 3.2.14. Let n ≥ 1 and x1 , x2 , . . . , xn be indeterminates. The polynomial ring in n
variables over a ring R is defined to be
R[x1 , x2 , . . . , xn ] = (((R[x1 ])[x2 ]) · · · )[xn ].
We write an element of this ring as
N1 N2 Nn
∑ ∑ ··· ∑ ai1i2...iN x1i1 x2i2 · · · xnin ,
i1 =0 i2 =0 in =0

where the coefficients lie in R. The elements x1i1 x2i2 · · · xnin are called monomials.
We will see below that this construction is independent, up to isomorphism, of the ordering
of the variables.
R EMARK 3.2.15. In multiplying in R[x1 , x2 , . . . , xn ], the variables xi all commute with each
other and the elements of R. A quantity such as x2 x1 x2 equals x1 x22 .
E XAMPLE 3.2.16. In the ring Z[x, y], we have polynomials like x2 + 2xy and 1 − x + y, and
we have
(x2 + 2xy)(1 − x + y) = x2 + 2xy − x3 + x2 y + 2xy2 .
Finally, we consider direct products.
D EFINITION 3.2.17. Let I be an indexing set, and let {Ri | i ∈ I} be a nonempty collection
of rings. Then the direct product ∏i∈I Ri of the Ri over i ∈ I is the binary structure is the di-
rect product of the sets Ri together with the binary operations of coordinate-wise addition and
multiplication. If I = Xn = {1, 2, . . . , n}, we write
∏ Ri = R1 × R2 × · · · × Rn.
i∈I
 3.3. UNITS 71

That the direct product of rings is a ring is a simple consequence of its definition, and we
state it without proof.
L EMMA 3.2.18. Any direct product of rings is a ring.
R EMARKS 3.2.19. Let {Ri | i ∈ I} be a nonempty collection of rings, and set R = ∏i∈I Ri .
a. The ring R is commutative if and only if each Ri is commutative.
b. The zero element of R is the element (0)i∈I .
c. The element (1)i∈I is the multiplicative identity in R.
d. The element ei which is 0 in every coordinate but the ith, where it is 1, satisfies e2i = ei ,
but ei is not the multiplicative identity of R (unless I has only one element).
E XAMPLE 3.2.20. If R is any ring, then Rn is the product of n copies of R.
3.3. Units
Not all rings are fields, but one can still ask which elements are invertible under multiplica-
tion. These elements are known as units.
D EFINITION 3.3.1. A unit in a ring is a nonzero element u ∈ R such that u has a multiplicative
inverse in R. We also say that u is invertible.
E XAMPLES 3.3.2.
a. The element 1 is a unit in every nonzero ring.
b. The units in a field F are the elements of F × .
c. The only units in Z are 1 and −1.
P ROPOSITION 3.3.3. The units in a nonzero ring R form a group under multiplication.
P ROOF. Let R× denote the set of units in R. If u, v ∈ R× , then let u0 , v0 ∈ R× be multiplicative
inverses to u and v respectively. We have
uv · (v0 u0 ) = 1 = (v0 u0 ) · uv,
so multiplication is a binary operation on R× , which we already know to be associative. Clearly,
1 is a unit and an identity in R× , and by definition, every unit has an inverse in R× , so R× is a
group. 
D EFINITION 3.3.4. The group of units in a nonzero ring R is denoted R× .
R EMARK 3.3.5. If F is a field, then its unit group and its multiplicative group coincide, and
hence the notation F × for both is unambiguous.
E XAMPLE 3.3.6. The group of units in Mn (R) for a ring R is its subset GLn (R) of invertible
matrices. E.g., if R = R, then these are the matrices with nonzero determinant.
E XAMPLE 3.3.7. If R = ∏i∈I Ri is a direct product of rings Ri over an indexing set I, then
R× = ∏ R×
i .
i∈I
72 3. RING THEORY

P ROPOSITION 3.3.8. The units in Z/nZ for n ≥ 1 are exactly the images of those i ∈ Z
relatively prime to n.
P ROOF. Let i ∈ Z. By Proposition 2.3.14, we have hii = hgcd(i, n)i as subgroups of Z/nZ.
The set of i j with j ∈ Z/nZ are exactly the elements of hii. Therefore, i is a unit in Z/nZ if and
only if 1 is an integer multiple of gcd(i, n) in Z/nZ. Since gcd(i, n) is a divisor of n, this can and
will only happen if gcd(i, n) = 1, which is to say that i is relatively prime to n. 
C OROLLARY 3.3.9. The group (Z/nZ)× has order φ (n), where φ is the Euler φ -function.
C OROLLARY 3.3.10. For n ≥ 1, the ring Z/nZ is a field if and only if n is prime.
We now have the following corollaries by the corollary of Lagrange’s theorem that the order
of an element of a group divides the order of the group. What is remarkable is that they are
nonobvious statements of simple arithmetic.
C OROLLARY 3.3.11 (Euler’s theorem). Let n ≥ 1. Then
aφ (n) ≡ 1 mod n
for every a ∈ Z relatively prime to n.
Note that every nonzero element of Z/pZ is relatively prime to p. Hence we also also have
the following special case of Euler’s theorem.
C OROLLARY 3.3.12 (Fermat’s little theorem). Let p be a prime number. Then
a p−1 ≡ 1 mod p
for every a ∈ Z not divisible by p.
These raise the following questions. What is the order of a unit in Z/nZ? We know it to be
a divisor of φ (n), but is there a simple formula for it in terms of a and n? This is one of many
questions in the field of mathematics known as number theory. Let us give a few examples of
arithmetic in Z/nZ.
E XAMPLE 3.3.13. Suppose we wish to calculate 3362 in Z/11Z. Fermat’s little theorem tells
us that 310 ≡ 1 mod 11, so
3362 ≡ (310 )36 32 ≡ 32 ≡ 9 mod 11.
In other words, 310 = 9 in Z/11Z.
E XAMPLE 3.3.14. What is the order of 2 in (Z/101Z)× ? Since 101 is prime, the order of 2
must be a divisor of 100. We have 25 < 101, and 210 = 1024 ≡ 14 mod 101. Moreover, we have
220 ≡ (14)2 ≡ 196 ≡ −6 mod 101,
225 = 220 25 ≡ −6 · 32 ≡ −192 ≡ 10 mod 101,
250 ≡ 102 ≡ −1 mod 101.
Therefore, the order of 2 in Z/101Z must be 100.
 3.4. INTEGRAL DOMAINS 73

3.4. Integral domains

D EFINITION 3.4.1. A left (resp., right) zero divisor in a ring R is a nonzero element a ∈ R
such that there exists a nonzero element b ∈ R with ab = 0 (resp., ba = 0). A zero divisor in a
ring R is an element that is either a left or a right zero divisor.
R EMARK 3.4.2. Note that 0 is never considered to be a zero divisor (at least under our con-
ventions). In fact, 1 is never a zero divisor either, as 1 · b = b for all b ∈ R.
E XAMPLE 3.4.3. The ring M2 (R) has zero divisors. For instance, we have
       
1 0 0 0 0 0 1 0
· = · = 0.
0 0 0 1 0 1 0 0
E XAMPLE 3.4.4. If R = R1 × R2 for some nonzero rings R1 and R2 , then R has zero divisors,
since if a ∈ R1 is nonzero and b ∈ R2 is nonzero, we have (a, 0) · (0, b) = (0, b) · (a, 0) = 0. For
instance, Zn has zero divisors for n ≥ 2, though Z does not.
One might ask for a ring that contains a left zero divisor that is not a right zero divisor. For
this, let us make the following general definition.
D EFINITION 3.4.5. Let A be an abelian group under addition. The endomorphism ring of A
is the set
End(A) = { f : A → A | f is a group homomorphism}
under addition and composition of functions.
R EMARK 3.4.6. If A is an abelian group, then End(A) is a ring, with 1 being the identity
function on A. In general, End(A) may be a noncommutative ring.
E XAMPLE 3.4.7. Let A = ∏∞
i=1 Z, an abelian group under addition. Define L, R ∈ End(A) by
L(a1 , a2 , a3 , . . .) = (a2 , a3 , a4 , . . .) and R(a1 , a2 , a3 , . . .) = (0, a1 , a2 , . . .).
Moreover, let M ∈ End(A) be defined by
M(a1 , a2 , a3 , . . .) = (a1 , 0, 0, . . .).
Then
LM(a1 , a2 , a3 , . . .) = L(a1 , 0, 0, . . .) = 0 and MR(a1 , a2 , a3 , . . .) = M(0, a1 , a2 , . . .) = 0,
so L is a left zero divisor and R is a right zero divisor. On the other hand,
LR(a1 , a2 , a3 , . . .) = L(0, a1 , a2 , . . .) = (a1 , a2 , a3 , . . .),
so LR = 1. Therefore, L cannot be a right zero divisor, for if XL = 0 for some X ∈ End(A), then
0 = (XL)R = X(LR) = X. Similarly, R is not a left zero divisor.
E XAMPLE 3.4.8. In the ring Z/6Z, the elements 2, 3, and 4 are zero divisors, since 2 · 3 =
3 · 4 = 0.
More generally, we have the following.
74 3. RING THEORY

L EMMA 3.4.9. For n ≥ 1, the zero divisors in Z/nZ are exactly its nonzero elements that are
not relatively prime to n.
P ROOF. Let ā ∈ Z/nZ be nonzero, and let b = n/ gcd(a, n). Then āb̄ = 0, and we know that
b̄ 6= 0 if and only if gcd(a, n) 6= 1. On the other hand, if āb̄ = 0, then ab is a multiple of n, so
b is a multiple of n/ gcd(a, n). Therefore, a is a zero divisor if and only if gcd(a, n) 6= 1, which
occurs if and only if a is not relatively prime to n 
As a corollary, if p is a prime number, then Z/pZ has no zero divisors. In fact, we shall see
momentarily that every field has no zero divisors.
D EFINITION 3.4.10. A nonzero commutative ring R is called an integral domain if R contains
no zero divisors.
L EMMA 3.4.11. Every field is an integral domain.
P ROOF. Let F be a field, and let a ∈ F be such that there exists a nonzero element b ∈ F with
ab = 0. Then 0 = (ab)b−1 = a. 
By definition, any subring of an integral domain is also an integral domain.
E XAMPLES 3.4.12. The fields Q, R, C, and Z/pZ for any prime p are all integral domains.
That Z is an integral domain is either an easy check or the fact that it is a subring of Q. Since
Z/nZ contains zero divisors for composite n ≥ 1, it is not an integral domain.
P ROPOSITION 3.4.13. Let R be an integral domain. Then R[x] is an integral domain. More-
over, if f , g ∈ R[x] are nonzero, then deg f g = deg f + deg g, and the units in R[x] are exactly the
units in R.
P ROOF. Let f , g ∈ R[x] be nonzero polynomials of degree N and M respectively. Write f =
∑N
i=0 ai x and g = ∑M
i j
j=0 b j x . Then
N+M k
fg = ∑ ck xk , ck = ∑ ai bk−i .
k=0 i=0
If 0 ≤ i ≤ N + M, then ai = 0 if i > N and bM+N−i = 0 if i < N, so cN+M = aN bM . Since R is
an integral domain, we then have cN+M 6= 0, so f g 6= 0. Therefore, we have deg f g = N + M. If
f g = 1, then this forces N = M = 0, and therefore f = a0 , g = b0 , and a0 b0 = 1, which means
that f ∈ R× . 
One particularly nice use of integral domains is that they obey cancellation laws.
L EMMA 3.4.14. Let R be an integral domain, and let a, b, c ∈ R be such that ab = ac. Then
either a = 0 or b = c.
P ROOF. If ab = ac, then a(b − c) = 0 by the distributive law (and Lemma 3.1.8), so as R
contains no zero divisors, at least one of a and b − c must be 0. 
We have already seen that Z/nZ is an integral domain if and only if n is prime, and so if and
only if Z/nZ is a field. We have the following stronger result.
 3.5. RING HOMOMORPHISMS 75

T HEOREM 3.4.15. If R is a finite integral domain, then R is a field.

P ROOF. Let a ∈ R be nonzero. Lemma 3.4.14 tells us that the elements ab with b ∈ R are all
distinct. Since there are then |R| of them, the set {ab | b ∈ R} is R itself. In particular, there exists
b ∈ R with ab = 1, proving that a has a multiplicative inverse. 
Finally, we introduce the notion the characteristic of a ring.
D EFINITION 3.4.16. Let R be a ring. The characteristic char(R) of R is the smallest n ≥ 1
such that na = 0 for all a ∈ R if such an n exists, and otherwise we set char(R) = 0.
E XAMPLES 3.4.17.
a. The ring Z/nZ has characteristic n, while Z, Q, R, and C all have characteristic 0.
b. The characteristic of Mn (R) for a ring R is equal to the characteristic of R for every n ≥ 1.
L EMMA 3.4.18. The characteristic of a nonzero ring R is the smallest n > 1 such that n = 0
in R if such an n exists, and is 0 otherwise.
P ROOF. We cannot have 1 · a = 0 unless a = 0, so char(R) 6= 1 as R is nonzero. Recall that
n ∈ R is considered to be n · 1. If n = 0 in R, then clearly na = 0 for all a ∈ R. On the other hand,
that n = 0 is the special case of na = 0 with a = 1. If n = n · 1 6= 0 for all n ≥ 1, then by definition,
we have char(R) = 0. 
P ROPOSITION 3.4.19. The characteristic of an integral domain is either 0 or prime.
P ROOF. We employ Lemma 3.4.18. If R is an integral domain and n = 0 in R for some com-
posite n > 1, then n = mm0 = 0 for some prime m and m0 dividing n, which by the nonexistence
of zero divisors implies that either m or m0 is zero. In other words, the smallest n > 1 with n 6= 0
in R cannot be composite, so must be prime. 

3.5. Ring homomorphisms

In this section, we introduce the notion of a ring homomorphism, which is a function from
one ring to another that is compatible with both addition and multiplication: in other words, it is
a homomorphism of binary structures both for (R, +) and for (R, ·).
D EFINITION 3.5.1. Let R and Q be rings. A function φ : R → Q is a ring homomorphism if
φ (1) = 1 and it satisfies
φ (a + b) = φ (a) + φ (b) and φ (ab) = φ (a)φ (b)
for all a, b ∈ R.
We give some examples of ring homomorphisms.
E XAMPLES 3.5.2.
a. The reduction map φn : Z → Z/nZ with φn (a) = a is a surjective ring homomorphism.
b. The multiplication-by-n map ψn : Z → Z with ψn (a) = na is not a ring homomorphism
unless n = 1.
76 3. RING THEORY

Here are several standard ring homomorphisms.

D EFINITION 3.5.3. Let R be a ring.
a. The identity homomorphism idR : R → R is the ring homomorphism given by idR (a) = a
for all a ∈ R.
b. If R is a subring of a ring Q, we have the inclusion map ιR : R → Q with ιR (a) = a for all
a ∈ R.
An inclusion map is always injective, but will only be surjective if the subring is the whole
ring. Here are some other examples.
E XAMPLES 3.5.4. Let R be a nonzero ring.
a. There is an injective ring homomorphism ι : R → R[x] that sends a ∈ R to the constant
polynomial a ∈ R[x].
b. There is a surjective ring homomorphism π : R[x] → R that sends f ∈ R[x] to its constant
coefficient. Note that π ◦ ι = idR , but ι ◦ π 6= idR .
We mention another useful class of ring homomorphisms of polynomial rings, arising from
maps on coefficients.
E XAMPLES 3.5.5. Let R and Q be rings, and let φ : R → Q be a ring homomorphism. This
induces maps on polynomial rings and matrix rings, as follows.
a. The map φ̃ : R[x] → Q[x] induced by φ on polynomial rings is given by the formula
 N  N
φ̃ ∑ ai xi = ∑ φ (ai )xi
i=0 i=0
for ai ∈ R for 0 ≤ i ≤ N for some N ≥ 0.
b. The map φ̃ : Mn (R) → Mn (Q) induced by φ on matrix rings is given by the formula
φ̃ ((ai j )) = (φ (ai j ))
for (ai j ) ∈ Mn (R).
R EMARK 3.5.6. If R is a subring of Q, then we may use the map of polynomial rings induced
by the inclusion map of R into Q to view R[x] as a subring of Q[x].
R EMARK 3.5.7. The product of ring homomorphisms φi : Ri → Qi over an index set i ∈ I is
a ring homomorphism between the corresponding products.
L EMMA 3.5.8. Let R be a ring and Q be an integral domain, and let φ : R → Q be a nonzero
homomorphism. If u ∈ R× , then φ (u) ∈ Q× .
P ROOF. Let v be a multiplicative inverse to u in R. By the previous lemma
φ (u)φ (v) = φ (uv) = φ (1) = 1,
and, similarly, we have φ (v)φ (u) = φ (vu) = 1. 
We also have the following.
 3.5. RING HOMOMORPHISMS 77

D EFINITION 3.5.9. If R = ∏i∈I Ri is a product of rings, then there are projection maps
π i : R → Ri , πi ((ai )i∈I ) = ai
which are ring homomorphisms.
R EMARK 3.5.10. If R = ∏i∈I Ri is a product of rings, the inclusion maps ιi : Ri → R for i ∈ I
given by taking a ∈ Ri to the element with ith coordinate a and jth coordinate 0 for j 6= i are not
ring homomorphisms if at least two Ri are nonzero rings, since ι(1) 6= 1 ∈ R.
As with group homomorphisms, we have notions of kernel and image of a ring homomor-
phism.
D EFINITION 3.5.11. Let φ : R → Q be a ring homomorphism. Then the kernel of φ is
ker φ = {r ∈ R | φ (r) = 0},
and the image of φ is
im φ = {φ (r) | r ∈ R}.
One can check very easily that im φ is a subring of Q for any ring homomorphism φ : R → Q.
However, while ker φ is a subgroup of R closed under multiplication, it will not contain 1 unless
φ = 0.
E XAMPLES 3.5.12. Let R be a ring. We consider the homomorphisms of Example 3.5.4.
a. The inclusion ι : R → R[x] has ker ι = 0 and im ι the subring of constant polynomials in
R[x], which we also denote R.
b. The projection π : R[x] → R has im ι = R and kernel consisting of the polynomials with 0
constant coefficient, which is the to say, the multiples of x.
Note that since any ring homomorphism is, in particular, a homomorphism of abelian groups
under addition, we have the following.
L EMMA 3.5.13. A ring homomorphism φ : R → Q is injective if and only if ker φ = {0}.
We will have much more to say about kernels later. For now, let us finish with a corollary for
fields.
L EMMA 3.5.14. Let φ : F → F 0 be a ring homomorphism, where F and F 0 are fields. Then
φ is injective and φ (x)−1 = φ (x−1 ) for all x 6= 0.
P ROOF. For any x ∈ F × , we have
φ (1) = φ (x · x−1 ) = φ (x) · φ (x−1 ),
so φ (x) is nonzero and has multiplicative inverse φ (x−1 ). In particular, Lemma 3.5.13 tells us
that φ is injective. 
As usual, we can speak about injective and surjective ring homomorphisms, as well as iso-
morphisms.
D EFINITION 3.5.15. A ring homomorphism φ : R → Q is an isomorphism if it is bijective.
78 3. RING THEORY

For instance, let us check that a polynomial ring in two variables is independent of the or-
dering of the variables, up to an isomorphism. We leave it to the reader to treat the case of more
than two variables using the following lemma and the construction in Example 1a.
L EMMA 3.5.16. Let x and y be indeterminates. The map σ : (R[x])[y] → (R[y])[x] satisfying
! ! !
M N N M
i j
(3.5.1) σ ∑ ∑ ai j x y =∑ ∑ ai j y j xi ,
j=0 i=0 i=0 j=0
where the ai j are elements of R. is an isomorphism.
P ROOF. Note that every element of (R[x])[y] may be expressed in the form on the left of
(3.5.1), since a polynomial in y with coefficients in R[x] has finite degree (at most M), and each
of the finitely many nonzero coefficients then has a degree, and we choose N to be at least the
maximum of these degrees. Similarly, every element of (R[y])[x] may be written in the form on
the right of (3.5.1), so the map is onto. By definition, it is one-to-one, and we leave it to the
reader to check that it is a ring homomorphism. 
As usual, the inverse of an isomorphism of rings is an isomorphism of rings.
3.6. Subrings generated by elements
D EFINITION 3.6.1. Let R be a subring of a ring Q, and let X be a set of elements of Q. The
subring of Q generated over R by X is the smallest subring of Q containing R and X.
Since the intersection of subrings containing a given set of elements is a subring, Defini-
tion 3.6.1 makes sense. When we have a finite set X, we often speak of the subring generated
over R by the elements of X, as opposed to X itself. We will only be interested in a special case
in which the elements we are adding to the subring commute with every element in that subring.
We note the following, which we leave to the reader to verify.
D EFINITION 3.6.2. Let R be a subring of a ring Q, and let α ∈ Q commute with every element
of R. The ring given by adjoining α to R is
( )
N
R[α] = ∑ riα i | ri ∈ R for all 0 ≤ i ≤ N for some N ≥ 0 .
i=0

R EMARK 3.6.3. We often read R[α] as “R adjoin α.”

We leave it to the reader to check the following.
L EMMA 3.6.4. Let R be a subring of a ring Q, and let α ∈ Q commute with every element of
R. The R[α] is the subring generated over R by α.
D EFINITION 3.6.5. Let R be a subring of Q. If α1 , α2 , . . . , αn ∈ Q commute with each other
and every element of R, we set
R[α1 , α2 . . . , αn ] = (((R[α1 ])[α2 ]) · · · )[αn ].
R EMARK 3.6.6. The ring R[α1 , . . . , αn ] in Definition 3.6.5 is the smallest subring of Q con-
taining R and each αi , so generated over R by the αi .
 3.6. SUBRINGS GENERATED BY ELEMENTS 79

E XAMPLES 3.6.7.
a. The ring Z[i] = {a + bi | a, b ∈ Z} is a subring of C known as the Gaussian integers. Note
that since i2 = −1, it is unnecessary to consider polynomials of higher degree.
b. The ring
( )
√
n
n−1 √
n
Z[ 2] = ∑ ai 2 | ai ∈ Z, 0 ≤ i ≤ n − 1
i=0
is a subring of R.
c. The ring
√ √ √
Z[i, 2] = {a + bi + c 2 + d 2i | a, b, c, d ∈ Z}
is a subring of C.
d. The ring Q[x2 ] is a subring of Q[x] consisting of polynomials of the form
N
∑ aix2i
i=0
with every ai ∈ Q.
We may relate this to the evaluation of polynomial rings at ring elements.
D EFINITION 3.6.8. Let R be a subring of a ring Q, and let α ∈ Q commute with every element
of R. For f = ∑N i
i=0 ci x ∈ R[x], we define the value of f at α ∈ R to be
N
f (α) = ∑ ci α i .
i=0
For any a ∈ R, the evaluation-at-α map is defined by
eα : R[x] → R[α], eα ( f ) = f (α)
for all f ∈ R[x].
The following is a result of the definitions of addition and multiplication in R[x].
L EMMA 3.6.9. Let R be a subring of a ring Q, and let α ∈ Q commute with every element of
R. The evaluation-at-α map eα : R[x] → R[α] is a ring homomorphism.
P ROOF. Let f = ∑N i M i
i=0 ci x ∈ R[x], and let g = ∑i=0 di x ∈ R[x] for some n ≥ 0. Then we have
D D D
eα ( f + g) = ( f + g)(α) = ∑ (ci + di )α i = ∑ ci α i + ∑ di α i = eα ( f ) + eα (g),
i=0 i=0 i=0

where D = max{M, N}, and

!
M+N k
eα ( f g) = ∑ ∑ cidk−i α k.
k=0 i=0
80 3. RING THEORY

Since α commutes with every element of R, we have ci dk−i α k = ci α i dk−i α k−i for all i ≤ k, so
the latter term equals !
N  M
∑ ciα i · ∑ d j α j = eα ( f ) · eα (g).
i=0 j=0

R EMARK 3.6.10. The evaluation-at-zero map is none other than the ring homomorphism
constructed in Example 3.5.4a that takes a polynomial to its constant term.
E XAMPLE 3.6.11. If X is a set and R is a ring, then the set Maps(X, R) of functions from X to
R forms a ring under the usual operations of pointwise addition and multiplication on R. Given
a ∈ X, we again have an evaluation-at-a map
εa : Maps(X, R) → R,
given by εa ( f ) = f (a) for f ∈ Maps(X, R) and a ∈ X, which is a ring homomorphism.
E XAMPLE 3.6.12. Let R be a commutative ring, and let a ∈ R. The evaluation map ea on R[x]
can be viewed as the composition εa ◦ κ, where
κ : R[x] → Maps(R, R), κ( f )(a) = f (a)
for f ∈ R[x] and a ∈ R. In other words, κ takes a polynomial to the function it defines. It is a ring
homomorphism since R is commutative.
Note that even if R is commutative, κ is not always injective. For instance, if R = Z/pZ for a
prime number p, then f = x p − x is a nonzero polynomial in R[x], but p(a) = 0 for all a ∈ Z/pZ,
so κ(p) = 0.

3.7. Ideals and quotient rings

In this section, we introduce the notion of an ideal of a ring. An ideal plays the role that a
normal subgroup does in group theory, which is to say that we can take a quotient of a ring by an
ideal and obtain another ring. The issue with simply using a subring can be seen in the following
example.
E XAMPLE 3.7.1. Consider the quotient group Q/Z under addition. The multiplication in Q
does not induce a well-defined multiplication on Z. To see this, note that one would like
(a + Z) · (b + Z) = (ab + Z)
for any a, b ∈ Q. But then we would have
1  1  1
0 + Z = (0 + Z) · + Z = (1 + Z) · + Z = + Z,
2 2 2
which is clearly not the case.
To fix this, we introduce the notion of an ideal. We begin with left and right ideals.
D EFINITION 3.7.2. A subset I of a ring R that is a subgroup under addition is called a left
(resp., right) ideal if R · I ⊆ I (resp., I · R ⊆ I).
 3.7. IDEALS AND QUOTIENT RINGS 81

D EFINITION 3.7.3. A two-sided ideal, or more simply, an ideal, of a ring R is any subset of
R that is both a left and a right ideal.
In other words, a left ideal J of R is an additive subgroup for which r · b ∈ J for all r ∈ R and
b ∈ J, and a right ideal K is one for which c · r ∈ K for all r ∈ R and c ∈ K. An ideal I of R is an
additive subgroup for which both r · a ∈ I and a · r ∈ I for all r ∈ R and a ∈ I.
R EMARK 3.7.4. Note that I ⊆ R · I, so the condition that R · I ⊆ I (resp., I · R ⊆ I) amounts to
R · I = I (resp., I · R = I).
In fact, we have the following simple criterion for a nonempty subset to be an ideal.
L EMMA 3.7.5. Let R be a ring, and let I be a nonempty subset of R. Then I is a left (resp.,
right ideal) if and only if the following hold:
i. I is closed under addition: if a, b ∈ I, then a + b ∈ I, and
ii. I is closed under left (resp., right) multiplication by elements of R: if r ∈ R and a ∈ I, then
ra ∈ I (resp., ar ∈ I).
P ROOF. We need only see that a set I satisfying (i) and (ii) is a subgroup. For this, we must
show that it contains 0, which it does since 0 = 0 · a for any a ∈ I, and that it contains additive
inverses, which it does since −a = −1 · a for any a ∈ I. 
R EMARK 3.7.6. Every left and every right ideal in a commutative ring R is an ideal of R.
E XAMPLES 3.7.7.
a. The subset nZ of Z is an ideal of Z for each n ∈ Z. That is, any integer multiple of an
integer multiple of n is an integer multiple of n.
b. The subset Z of Q is not an ideal, as 1/2 · Z 6⊆ Z, for instance.
c. Let R be a nonzero ring. Consider the set of matrices in Mn (R) that are 0 in all entries
outside their first columns. This is a left ideal of Mn (R), but it is not a right ideal for n ≥ 2.
Similarly, the set of matrices in Mn (R) that are 0 in all entries outside their first rows is a right
ideal of Mn (R).
d. Let R be a ring. The set of all polynomials with zero constant coefficient is an ideal of R,
equal to the set of multiples of x in R[x].
D EFINITION 3.7.8. The zero ideal of a ring R is the subset {0}. The improper ideal of R is
the ring R itself. An ideal is said to be nonzero if it is not equal to zero, and an ideal is said to be
proper if it is not equal to R.
We note the following.
L EMMA 3.7.9. Let R be a ring, and let I be a left (or right) ideal of R. Then I = R if and only
if I contains a unit, and in particular if and only if I contains 1.
P ROOF. If I = R, then clearly I contains 1 and therefore a unit. If u ∈ I is a unit, then u−1 ∈ R,
so 1 = u−1 · u ∈ I. And if 1 ∈ I, then a = a · 1 ∈ I for all a ∈ R. 
82 3. RING THEORY

The following classifies, as a special case, all ideals in a field.

C OROLLARY 3.7.10. The only left and only right ideals in a division ring are {0} and D.
P ROOF. If I is a nonzero left or right ideal of D, it then contains a unit, so is D. 
We shall see later that the converse to Corollary 3.7.10 also holds. We give one more example.
L EMMA 3.7.11. Let R and Q be rings. Then any left ideal of R × Q has the form I × J, where
I is a left ideal of R and J is a left ideal of Q.
P ROOF. Let K be an ideal of R × Q. Let
I = {a ∈ R | (a, 0) ∈ K} and J = {b ∈ Q | (0, b) ∈ K},
which are left ideals of R and of Q, respectively. If (a, b) ∈ K, then (1, 0) · (a, b) = (a, 0), so
a ∈ I and (1, 0) · (a, b) = (0, b), so b ∈ J. Therefore, K ⊆ I × J. Conversely, if (a, b) ∈ I × J, then
(a, 0) ∈ K and (0, b) ∈ K, so (a, b) = (a, 0) + (0, b) ∈ K, so I × J ⊆ K. 
The following is the ring-theoretic analogue of Proposition 2.12.11.
P ROPOSITION 3.7.12. Let φ : R → Q be a homomorphism of rings. Then ker φ is an ideal of
R.
P ROOF. We know from Proposition 2.8.8 that ker φ is a subgroup of R under addition. More-
over, if r ∈ R and a ∈ ker φ , then
φ (ra) = φ (r)φ (a) = φ (r) · 0 = 0,
so ra ∈ ker φ . Similarly, we have φ (ar) = 0, so ar ∈ ker φ as well. 
We may now construct the analogue of a quotient group, known as a quotient ring.
T HEOREM 3.7.13. Let R be a ring, and let I be a two-sided ideal of R. Then the quotient
group R/I has a well-defined multiplication on it, given by
(r + I) · (s + I) = (rs + I)
for r, s ∈ R. Moreover, with the usual addition of cosets and this multiplication, R/I becomes a
ring.
P ROOF. Suppose that a, a0 , b, b0 ∈ R with a + I = a0 + I and b + I = b0 + I. Then there exist
x, y ∈ I with a0 = a + x and b0 = b + y. We have
(a0 b0 + I) = (a + x)(b + y) + I = ab + ay + xb + xy + I = ab + I,
since ay, xb, xy ∈ I in that I is a two-sided ideal. Therefore, the multiplication on R/I is well-
defined. That it is associative is a direct consequence of the associativity of multiplication on R.
Distributivity is again a consequence of distributivity on R, but we write out the proof of the left
distributive law:
(a + I) · ((b + I) + (c + I)) = (a + I) · (b + c + I) = a(b + c) + I = (ab + ac) + I
= (ab + I) + (ac + I) = (a + I) · (b + I) + (a + I) · (c + I).

 3.7. IDEALS AND QUOTIENT RINGS 83

D EFINITION 3.7.14. The quotient of a ring R by an ideal I is the ring R/I defined by Theo-
rem 3.7.13. We say that R/I is the quotient ring of R by I (or the factor ring of R by I).
E XAMPLES 3.7.15.
a. The quotient of the ring Z by the ideal nZ is the ring Z/nZ.
b. The quotient of any ring R by the zero ideal is isomorphic to R. The quotient of any ring
R by R is isomorphic to the zero ring.
The following is immediately verified.
D EFINITION 3.7.16. The map πI : R → R/I defined by πI (a) = a + I is called the quotient
map from R to R/I.
R EMARK 3.7.17. For R a ring, I and ideal of R, and a, b ∈ R, we may sometimes write a ≡ b
mod I to mean that a + I = b + I, or simply just a = b when it is understood that we are working
with the images of a and b under πI , i.e., in the ring R/I.
The following is easily verified.
L EMMA 3.7.18. Let I be an ideal in a ring R. The quotient map πI : R → R/I is a surjective
ring homomorphism with kernel I.
We have the analogue of the first isomorphism theorem.
T HEOREM 3.7.19. Let φ : R → Q be a homomorphism of rings. Then the map
φ̄ : R/ ker φ → im φ
defined by φ̄ (a + ker φ ) = φ (a) for all a ∈ R is an isomorphism of rings.
P ROOF. We know that φ̄ is an isomorphism of additive groups by Theorem 2.13.11. Let
I = ker φ . For a, b ∈ R, we have
φ̄ ((a + I)(b + I)) = φ̄ (ab + I) = φ (ab) = φ (a)φ (b) = φ̄ (a + I)φ̄ (b + I),
so φ̄ is a ring homomorphism as well, therefore, a ring isomorphism. 
E XAMPLE 3.7.20. The kernel of the homomorphism π : R[x] → R of Example 3.5.4a is the
ideal I of polynomials with zero constant coefficient. In that it is onto, e0 induces an isomorphism
between R[x]/I and R.
Note that if φ is a surjective map, then φ carries ideals to ideals.
P ROPOSITION 3.7.21. Let φ : R → Q be a surjective homomorphism of rings. If I is a left
(resp., right) ideal of R, then φ (I) is a left (resp., right) ideal of Q.
P ROOF. We show this for left ideals I of R. Let q ∈ Q. Then q = φ (r) for some r ∈ R, and if
a ∈ I, then sφ (a) = φ (ra) ∈ φ (I), so φ (I) is a left ideal of Q. 
R EMARK 3.7.22. If φ : R → Q is a ring homomorphism that is not surjective, φ (R) is not an
ideal of Q, That is, since φ (1) = 1 ∈ φ (R), the ideal generated by φ (R) is Q, which is strictly
larger than φ (R).
84 3. RING THEORY

The following is a straightforward generalization of Proposition 3.7.12.

P ROPOSITION 3.7.23. Let φ : R → Q be a ring homomorphism. Let J be a left (resp., right)
ideal of Q. Then φ −1 (J) is a left (resp., right) ideal of R.
P ROOF. We prove this for left ideals of Q. Let J be a left ideal of Q. If a, b ∈ R with
φ (a), φ (b) ∈ J, then φ (a + b) = φ (a) + φ (b) ∈ J. If moreover r ∈ R, then φ (ra) = φ (r)φ (a) ∈ J,
so part a holds. 
We can now classify the ideals in quotient rings.
T HEOREM 3.7.24. Let R be a ring, and let I be an ideal of R. Then the quotient map
πI : R → R/I induces a one-to-one correspondence between the left, right, and two-sided ideals
of R containing I and the left, right, and two-sided ideals of R/I, respectively.
P ROOF. We prove this for left ideals. If J is a left ideal of R containing I, then πI (J) is a
left ideal of R/I by Proposition 3.7.21. If πI (J) = πI (K) for some left ideal of R containing K,
then any j ∈ J satisfies j = k + i for some k ∈ K and i ∈ I, and therefore j ∈ K since I ⊆ K. We
therefore have J ⊆ K, and similarly K ⊆ J, so J = K. On the other hand, if N is any left ideal of
R/I, then J = π −1 (N) is a left ideal of R, and it contains I since I = π −1 ({0}). Since πI (J) = N,
we are done. 
For later use, let us make the following definitions for left ideals, with the definitions for right
(resp., two-sided) ideals coming from replacing the word “left” by “right” (resp., “two-sided”).
D EFINITION 3.7.25. Let φ : R → Q be a ring homomorphism.
a. For any left ideal I of R, the extension of I by φ is the left ideal generated by φ (I).
b. For any left ideal J of Q, the contraction of J by φ is φ −1 (J).
c. We say that a left ideal of Q is extended by φ if it is the extension of some left ideal of R
by φ .
d. We say that a left ideal of R is contracted by φ if it is the contraction of some left ideal of
Q by f .
We mention the following interesting proposition with a rather subtle proof.
P ROPOSITION 3.7.26. Let φ : R → Q be a homomorphism of rings. The maps given by
contraction and extension of left ideals by φ restrict to mutually inverse bijections

{contracted left ideals of R} o

extend / {extended left ideals of Q}.
contract

The same holds with right ideals, or two-sided ideals, replacing left ideals.
P ROOF. We prove this for left ideals. Let I = φ −1 (J) for some ideal J of Q. Note that
φ (φ −1 (J)) = J ∩ φ (R). Then
φ −1 (Qφ (I)) = φ −1 (Qφ (φ −1 (J))) = φ −1 (Q(J ∩ φ (R))),
 3.8. PRINCIPAL IDEALS AND GENERATORS 85

and the latter set contains φ −1 (J ∩ φ (R)) = I but also is contained in I = φ −1 (J) since
Q(J ∩ φ (R)) ⊆ J.
Next, let J = Qφ (I) for some ideal I of R. Then
Qφ (φ −1 (J)) = Qφ (φ −1 (Qφ (I)))
contains Qφ (I) = J since φ −1 (Qφ (I)) ⊇ I, while
Qφ (φ −1 (J)) ⊆ QJ = J
as well. 

3.8. Principal ideals and generators

D EFINITION 3.8.1. A left ideal J of a ring R is said to be principal if there exists an element
a ∈ R such that
J = Ra = {ra | r ∈ R}.
Similarly, a right ideal K of a ring R is principal if there exists an element a ∈ R such that
K = aR = {ar | r ∈ R}. We then say that J (resp., K) is the left (resp, right) ideal generated by a.
R EMARK 3.8.2. Note that Ra for a ∈ R is always a left ideal of R, since ra − sa = (r − s)a for
r, s ∈ R, so Ra is an additive subgroup, and s(ra) = (sr)a, so Ra is closed under left multiplication
by elements of R.
We also have the notion of a principal ideal.
D EFINITION 3.8.3. An ideal I of a ring R is principal if there exists an element a ∈ R such
that ( )
N
I= ∑ riasi | ri, si ∈ R for 1 ≤ i ≤ N and N ≥ 0 .
i=1
We then say that I is generated by a and write I = (a).
R EMARK 3.8.4. The set RaR = {ras | r, s ∈ R} will not in general be a two-sided ideal, as
ras + r0 as0 for r, r0 , s, s0 ∈ R need not itself be an element of RaR.
E XAMPLES 3.8.5.
a. For each n ≥ 1, the ideal nZ is the principal ideal (n).
b. For every ring R, the zero ideal is the principal ideal (0).
c. For every ring R, we have R = (1), so R is a principal ideal of R, known as the imp
d. The ideal (x) in R[x] is the ideal consisting of all polynomials with nonconstant coefficient.
e. The ideal generated by (a, b) ∈ Z × Z is equal to the set {(ax, by) | x, y ∈ Z}.
E XAMPLE 3.8.6. Let R be a nonzero ring, and let n ≥ 1. For integers s,t with 1 ≤ s,t ≤ n, let
Est = (ei j ) ∈ Mn (R) be the matrix with est = 1 and ei j = 0 for (i, j) 6= (s,t). If A = (ai j ) ∈ Mn (R)
86 3. RING THEORY

is any matrix, then the (i, j)th entry of AEss is ais if j = s and 0 otherwise. Therefore, the left
ideal generated by Ess is
Mn (R)Ess = {(bi j ) ∈ Mn (R) | bi j = 0 for 1 ≤ i, j ≤ N, j 6= s},
the set of matrices that are zero outside of the sth column. Similarly, the (i, j)th entry of Ess A is
as j if i = s and 0 otherwise, so Ess Mn (R) is the right ideal of matrices that are zero outside of the
sth row.
The two-sided ideal (Ess ) of Mn (R) is in fact all of Mn (R). To see this, note that E js Ess Es j =
E j j for any j ∈ Z. We then have
n n
A= ∑ AE j j = ∑ (AE js)EssEs j ∈ (Ess).
j=1 j=1
Note, however, that the set X = Mn (R)Ess Mn (R) is not Mn (R), since each column of a matrix in
X has entries which are all equal to each other.
D EFINITION 3.8.7. A nonzero ring R is simple if its only ideals are 0 and R.
R EMARK 3.8.8. The reader can check using Example 3.8.6 that if D is a division ring and
A ∈ Mn (D) is nonzero, then the ideal (A) is all of Mn (D). So, Mn (D) is simple, but note that it is
not a division ring if n ≥ 2, and it does have proper, nonzero left ideals.
The following three results also clearly have analogues for right ideals that we leave unstated.
P ROPOSITION 3.8.9. Let D be a nonzero ring that contains no nonzero, proper left ideals.
Then D is a division ring.
P ROOF. Let u ∈ D be nonzero. By assumption, we have Du = D, so there exists v ∈ D such
that vu = 1. Then Dv = D, so there exists w ∈ D such that wv = 1. We then have w = wvu = u,
so u = v−1 ∈ D× . 
L EMMA 3.8.10. Let R be a ring, and let a, b ∈ R. Then Ra ⊆ Rb if and only if there exists
r ∈ R such that a = rb.
P ROOF. If Ra ⊆ Rb, then since a ∈ Rb, we have a = rb for some r ∈ R. Conversely, if a = rb
and r0 ∈ R, then r0 a = (r0 r)b ∈ Rb, so Ra ⊆ Rb. 
L EMMA 3.8.11. Let R be a ring that has no zero divisors. Let a, b ∈ R. Then Ra = Rb if and
only if b = ua for some u ∈ R× .
P ROOF. Note that a = 0 if and only if b = 0, so we may suppose that a and b are nonzero
with Ra = Rb. Since b ∈ Ra, we have that there exists u ∈ R with b = ua. Similarly, there exists
v ∈ R with a = vb. But then a = vua and b = uvb, so (1 − vu)a = (1 − uv)b. Since R has no left
zero divisors, we have uv = vu = 1. Conversely, if b = ua, then clearly b ∈ Ra, so Rb ⊆ Ra. On
the other hand, a = u−1 b, so Ra ⊆ Rb as well. 
E XAMPLE 3.8.12. In Q[x], we have ( f ) = (g) if and only if f = cg for some c ∈ Q× , since
Q[x]× = Q× .
We have various operations that can be performed on ideals.
 3.8. PRINCIPAL IDEALS AND GENERATORS 87

L EMMA 3.8.13. Let I and J be left ideals (resp., right ideals) of a ring R.
a. The set
I + J = {a + b | a ∈ I, b ∈ J}
is a left ideal (resp., right ideal) of R.
b. The intersection I ∩ J is a left (resp., right ideal) of R.
P ROOF.
a. If a ∈ I, b ∈ J, and r ∈ R, then r(a + b) = ra + rb, and ra ∈ I, rb ∈ J since I and J are
ideals, so r(a + b) ∈ I + J. Moreover, I + J is a subgroup of R under addition by Lemma 4.1.4.
b. If a, b ∈ I ∩ J and r ∈ R, then clearly a − b ∈ I ∩ J and ra ∈ I ∩ J, so I ∩ J is a left ideal of
R.

R EMARK 3.8.14. The argument of Lemma 3.8.13b carries over to show that an arbitrary
intersection of left (resp., right) ideals of a ring R is a left (resp., right) ideal of R.
Clearly, addition of ideals forms an associative and commutative binary operation on the set
of ideals of a ring. More generally, we have the following result.
L EMMA 3.8.15. Let T be an indexing set, and let {It | t ∈ T } be a collection of left (resp.,
right ideals) of a ring R. Then the set
( )
N
∑ It = ∑ ati | ti ∈ T, ati ∈ Iti for each 1 ≤ i ≤ N for some N ≥ 0
t∈T i=1
of finite sums of elements of the ideals It is an ideal of R, equal to the intersection of all ideals of
R containing It for every t ∈ T .
P ROOF. Note that ∑t∈T It consists exactly of finite sums of elements in the union ∪t∈T It . It
is a subgroup, as the sum of two finite sums is a finite sum, and the negative of two finite sums is
as well. Moreover, it is an ideal, as for any N ≥ 0, ti ∈ T and ati ∈ Iti for 1 ≤ i ≤ N, we have
N N
r · ∑ ati = ∑ rati ,
i=1 i=1
and rati ∈ Iti since Iti is a left ideal of R. Therefore ∑t∈T It is a left ideal, and similarly, it is a right
ideal.
Finally, note that if J is any ideal of R containing each It , then it must contain any finite
sum of elements in these ideals, i.e., in ∪t∈T It . Therefore, J contains ∑t∈T It . Therefore, the
intersection of all ideals of R containing each It is an ideal of R containing ∑t∈It It , and ∑t∈It It is
itself an ideal of R containing each It , so it equals the intersection. 
D EFINITION 3.8.16. Let R be a ring, and let {It | t ∈ T } be a collection of left (resp., right)
ideals. The sum of the ideals It with t ∈ T is the left (resp., right) ideal ∑t∈T It of R.
We will define generators solely for two-sided ideals, though they have obvious analogues
for left and right ideals.
88 3. RING THEORY

D EFINITION 3.8.17. Let X be a subset of a ring R. The ideal (X) generated by X is the sum
of the ideals (x) for x ∈ X. If I is an ideal of R and I = (X), we say that X is a set of generators of
I, and X generates I. The elements of X are called generators. If X = {a1 , a2 , . . . , an } is a finite
set, then we write (a1 , a2 , . . . , an ) for (X).
R EMARKS 3.8.18.
a. Every ideal is generated by the set of all of its elements.
b. Using Lemma 1.2.24, we could equivalently have defined (X) to be the smallest ideal
containing X.
Since the set-theoretic product of two ideals will not in general be closed under addition, we
depart from earlier notation to make the following definition.
D EFINITION 3.8.19. Let I and J be ideals of a ring R. Then the product IJ of I and J is the
ideal of R generated by all ab with a ∈ I and b ∈ J.
In particular, we may speak of powers I n = II · · · I of an ideal I for any n ≥ 1. Products are
easily calculated in terms of generators, as seen in the following examples.
E XAMPLES 3.8.20.
a. If R is a ring and x, y ∈ R, then (x) · (y) = (xy).
b. In the ring Q[x, y], we have
(x, y) · (x2 , x + y) = (x3 , x2 y, x2 + xy, xy + y2 ).
D EFINITION 3.8.21. We say that two ideals I and J of a ring R are coprime if I + J = R.
D EFINITION 3.8.22. For k ≥ 1, we say that ideals I1 , . . . , Ik of a ring R are pairwise coprime
if Ii + I j = R for all 1 ≤ i < j ≤ k.
We prove a general form of the Chinese Remainder Theorem.
T HEOREM 3.8.23 (Chinese Remainder Theorem). Let I1 , . . . , Ik be pairwise coprime two-
sided ideals of a ring R for some k ≥ 1. Then there is an isomorphism
R/(I ∩ I ∩ · · · ∩ I ) −∼
1 2 → R/I × R/I × · · · × R/I
k 1 2 k
that sends the coset of a ∈ R to (a + I1 , a + I2 , . . . , a + Ik ).
P ROOF. The kernel of the map R → ∏ki=1 R/Ii induced by the diagonal map is clearly I1 ∩
I2 ∩ · · · ∩ Ik . We need only see that it is surjective. Consider the case that k = 2. Let a, b ∈ R.
Then there exist d ∈ I1 and c ∈ I2 such that a + I1 = c + I1 and b + I2 = d + I2 . If we set x = c + d,
then x + I1 = c + I1 and x + I2 = d + I2 , so x maps to (a + I1 , b + I2 ).
For any k ≥ 3, suppose by induction we know the result for k − 1, so R/(I2 ∩ · · · ∩ Ik ) ∼ =
R/I2 × · · · R/Ik . We therefore need only see that I1 and I2 ∩ · · · ∩ Ik are coprime. Note that
I2 ∩ · · · ∩ Ik contains the product I2 · · · Ik . For each 2 ≤ i ≤ k, let ai ∈ I1 and bi ∈ Ii be such
that ai + bi = 1. Then 1 = (a2 + b2 ) · · · (ak + bk ) is an element of I1 plus b1 . . . bk ∈ I2 · · · Ik , as
needed. 
 3.9. POLYNOMIAL RINGS OVER FIELDS 89

D EFINITION 3.8.24. An ideal I of a ring R is said to be finitely generated if it has a finite set
of generators, which is to say that I = (a1 , a2 , . . . , an ) for some n ≥ 1 and a1 , a2 , . . . , an ∈ I.
E XAMPLE 3.8.25. If R[x, y], the ideal (x, y) is the ideal of elements with 0 constant term, as
every monomial other than 1 is either divisible x or y. It is not principal, since no element of
R[x, y] not in R× divides both x and y, but it is finitely generated.
E XAMPLE 3.8.26. Let n ≥ 2. The ideal (n, x) of Z[x] is the set of all sums n f + xg with
f , g ∈ Z[x], which is equal to the set of polynomials with Z-coefficients and constant coefficient
divisible by n. This is not principal, since n and x are both multiples only of ±1, which are not
contained in (n, x)
E XAMPLE 3.8.27. Consider the ideal (4, 6) of Z. It contains 2 = 6 − 4, so (2) ⊆ (4, 6) and
we have 4, 6 ∈ (2), so (4, 6) ⊆ (2). Therefore, (4, 6) is a principal ideal of Z, equal to the ideal
(2).
In fact, note the following.
L EMMA 3.8.28. The ideals of Z are exactly the subgroups of Z under addition, i.e., the nZ
with n ≥ 0. In particular, every ideal of Z is principal.
P ROOF. Ideals are by definition subgroups under addition, and if I is an ideal of Z, the
condition that Z · I ⊆ I is a consequence of this, since it merely says that Z-multiples of elements
of I are contained in I. That the subgroups of Z have the form nZ is Corollary 2.3.12. 
This leads to the following definition.
D EFINITION 3.8.29. An integral domain R is a principal ideal domain, or PID, if every ideal
in R is principal.
So far, we have the following examples.
E XAMPLES 3.8.30.
a. The ring Z is a principal ideal domain.
b. Every field is a principal ideal domain.
c. If R and Q are principal ideal domains, then every ideal R × Q is principal, though it is not
a domain.

3.9. Polynomial rings over fields

We consider polynomial rings over a field. One of the key properties of polynomial rings over
a field is that we can divide any polynomial by any nonzero polynomial, obtaining a remainder.
More generally, we can divide a polynomial over a commutative ring by another so long as the
leading coefficient of the divisor is a unit.
T HEOREM 3.9.1 (Division algorithm). Let R be a commutative ring. Suppose that f , g ∈ R[x]
are polynomials and that the leading coefficient of g is a unit in R. Then there exist unique
polynomials q, r ∈ R[x] such that f = qg + r and deg r < deg g.
90 3. RING THEORY

P ROOF. The case that f = 0 is trivial, so we assume that f is nonzero. We verify this by
induction on the degree n of f . Note that if n ≤ deg g, and in particular if n = 0, then we may
take q = 0 and r = f if deg g > 0 and q = f g−1 and r = 0 if deg g = 0 (recalling that we consider
the degree of 0 to be less than that of every nonzero polynomial). So suppose that n ≥ m = deg g.
Let an be the nonzero coefficient of xn in f and bm ∈ R× be the coefficient of xm in g. Then
f 0 = f − an b−1
m x
n−m g has degree at most n, and the coefficient of xn is a − a b−1 · b = 0, so in
n n m m
fact we have deg f 0 < n. By induction, therefore, there exist q0 and r in R[x] such that f 0 = q0 g + r
and deg r < m. Setting q = an b−1 0
m + q , we have

f = an b−1 0 −1 0
m g + f = (an bm + q )g + r = qg + r,
as desired.
If qg + r = q0 g + r0 for some q0 , r0 ∈ R[x] with deg r0 < deg g, then we have
(3.9.1) (q − q0 )g + (r − r0 ) = 0.
If q 6= q0 , we would have
deg(q − q0 )g ≥ deg g > deg(r − r0 ),
in contradiction to (3.9.1). So, we must have q = q0 , and then (3.9.1) yields r = r0 , establishing
uniqueness. 
We next show that polynomial rings in one variable over a field form another class of principal
ideal domains.
T HEOREM 3.9.2. Let F be a field. Then F[x] is a principal ideal domain. In fact, any
nonzero ideal I of F is generated by any nonzero polynomial that has minimal degree among all
polynomials in I.
P ROOF. By Theorem 3.4.13, F[x] is an integral domain. Let I be a nonzero ideal in F[x],
and let g be a nonzero polynomial in F[x] of minimal degree. We claim that I = (g). Let f ∈ I.
Using the division algorithm, we write f = qg + r with q, r ∈ F[x] with deg r < deg g. Then
r = f − qg ∈ I, which by the minimality of the degree of g forces r = 0. Thus f ∈ (g), and as f
was arbitrary, we have I = (g). 
D EFINITION 3.9.3. Let F be a field. A nonconstant polynomial f ∈ F[x] is irreducible if there
does not exist any g ∈ F[x] with 0 < deg g < deg f that divides f . A nonconstant polynomial that
is not irreducible is called reducible. A noncontant divisor of a polynomial is referred to as a
factor.
E XAMPLE 3.9.4. By definition, any polynomial of degree 1 is irreducible in F[x]. The poly-
nomial x2 + 1 is irreducible in Q[x] but not in C[x], where we have
x2 + 1 = (x + i)(x − i).
On the other hand, x2 is reducible for any F, since x2 = x · x.
D EFINITION 3.9.5. Let R be a ring. We say that a ∈ R is a root (or zero) of a polynomial
f ∈ R[x] if f (a) = 0.
 3.9. POLYNOMIAL RINGS OVER FIELDS 91

D EFINITION 3.9.6. In a commutative ring R, we say that an element b divides an element a in

R if there exists some c ∈ R such that a = bc. Equivalently, b divides a if a ∈ (b). We sometimes
write b | a to denote that b divides a.
We note the following.
P ROPOSITION 3.9.7. Let F be a field, and let f ∈ F[x]. Then a ∈ F is a root of f if and only
if x − a divides f .
P ROOF. If x − a divides f , then there exists g ∈ F[x] with f = (x − a)g. We then have
f (a) = (a − a)g(a) = 0, noting Lemma 3.6.9. Conversely, if a is a zero of f , then the division
algorithm implies that there exists some q ∈ F[x] and c ∈ F such that f = q(x − a) + c. We then
have
0 = f (a) = q(a)(a − a) + c = c,
so x − a divides f . 
We obtain the following corollaries.
C OROLLARY 3.9.8. Let F be a field and f ∈ F[x] be a polynomial of degree greater than 1.
If f has a root in F, then f is reducible.
P ROOF. If a ∈ F is a root of f , then Proposition 3.9.7 implies that f = g(x − a) for some
g ∈ F[x] with deg g = deg f − 1 > 0, so f is not irreducible. 
Since a reducible polynomial of degree 2 or 3 must have a linear factor, we therefore have
the following.
C OROLLARY 3.9.9. Let F be a field and f ∈ F[x] be a polynomial of degree 2 or 3. Then f
is reducible if and only if it has a root in F.
C OROLLARY 3.9.10. Let F be a field, and let f ∈ F[x] be a nonzero polynomial. Then f has
at most deg f distinct roots in F.
P ROOF. Suppose that f = (x − a1 ) . . . (x − am )g, where g ∈ F[x] has no roots, and a1 , a2 , . . .,
am ∈ F. Clearly, we may write f in this form, as otherwise we can factor out from g a linear
term x − b for some β with g(b) = 0. Moreover, we must have m ≤ n by degree considerations.
Finally, if f (c) = 0 for some c ∈ F, then since F is an integral domain, we must have c − ai = 0
for some i, which is to say that the ai are the only roots of f . 
E XAMPLES 3.9.11.
a. The polynomial x2 has 0 as its only root.
b. The polynomial x2 + 1 has no roots in Q, but it has two roots, ±1, in C.
c. The polynomial
x4 − x2 − 2x − 1 = (x2 + x + 1)(x2 − x − 1)
is not irreducible in Q[x], but it has no roots in Q.
92 3. RING THEORY

3.10. Maximal and prime ideals

Recall that Z/nZ is a field for n prime, but Z/nZ is not a field for n composite. In this section,
we shall see how we can interpret this as a property of the ideal nZ.
D EFINITION 3.10.1. An ideal m of a ring R is maximal if it is a proper ideal of R that is not
properly contained in any proper ideal of R.
In other words, a proper ideal m of R is maximal if there does not exist an ideal N of R such
that m ( N ( R.
E XAMPLES 3.10.2.
a. The maximal ideals of Z are exactly the pZ for p prime, as mZ contains nZ if and only if
m divides n. In particular, as p is a prime number, pZ is not contained in nZ for any n ≥ 2 with
n 6= p.
b. In a field, the unique maximal ideal is (0).
c. In Z × Z, the maximal ideals have either the form pZ × Z or Z × pZ for some prime
number p.
P ROPOSITION 3.10.3. Let F be a field. The maximal ideals of F[x] are exactly the ideals of
the form ( f ) with f ∈ F[x] irreducible.
P ROOF. Let f ∈ F[x]. If f = 0, then ( f ) = 0, which is not maximal. If f is a nonzero
constant, then ( f ) = (1) = F[x]. If f is reducible, then f = gh with g, h ∈ F[x] nonconstant, and
then ( f ) ⊆ (g), but g ∈
/ ( f ) since deg g < deg f , so ( f ) is not maximal.
If f is irreducible and I is an ideal containing ( f ), then I = (g)for some g ∈ F[x] as F[x] is a
PID. There then exists h ∈ F[x] such that f = gh. Since f is irreducible, we then have that either
g or h is constant, which is to say that I = (g) = F[x] or I = (g) = ( f ). In other words, ( f ) is
maximal. 

The following gives an alternate characterization of maximal ideals of rings.

T HEOREM 3.10.4. A proper ideal m in a commutative ring R is maximal if and only if R/m
is a field.
P ROOF. By Theorem 3.7.24, the ideals in R/m are in one-to-one correspondence with the
ideals in R containing m, which are just m and R. Since R/m has just two ideals, they must be
0 and R/m. Therefore, every nonzero element of R/m generates the ideal R/m, so is a unit. It
follows that R/m is a field. 
R EMARK 3.10.5. The same argument can be applied to noncommutative rings R to conclude
that if m is maximal then R/m has no nonzero proper ideals. However, as we have remarked
above, this does not imply that R/m is a division ring.
E XAMPLE 3.10.6. Recall that Z/nZ is a field if and only if n ≥ 2 is prime, which is to say if
and only if nZ is a maximal ideal of Z.
 3.10. MAXIMAL AND PRIME IDEALS 93

E XAMPLE 3.10.7. Since x2 + 1 is irreducible over Q, the ideal m = (x2 + 1) is maximal in

Q[x]. Clearly, m is contained in the kernel of the evaluation map ei : Q[x] → Q(i) defined by
ei ( f ) = f (i), but then it must be the entire kernel as the kernel is proper and m is maximal.
By the first isomorphism theorem for rings, the field Q[x]/(x2 + 1) is isomorphic to Q[i]. In
particular, Q[i] is equal to the subfield Q(i) of C consisting of fractions a+bi
c+di with a, b, c, d ∈ Q and
c d
(c, d) 6= (0, 0). One can also see this directly: the multiplicative inverse of c+di is c2 +d 2 − c2 +d 2 i.

E XAMPLE 3.10.8. The ring Q[x]/(x2 ) is not a field, or even an integral domain, since x · x ∈
(x2 ).
E XAMPLE 3.10.9. In Z[x], the ideals (p, x), where p is a prime number, are maximal. To see
this, consider the homomorphism
φ : Z[x] → Z/pZ
given by φ ( f ) = f (0) + pZ. This is surjective with kernel consisting of those f with constant
coefficient a multiple of p, which is to say the ideal (p, x).
Given a proper ideal I of a ring R: is I necessarily even contained in a maximal ideal? As-
suming the axiom of choice, the answer is yes. We require a preliminary lemma.
L EMMA 3.10.10. Let C be a chain of ideals in a ring R, ordered with respect to inclusion of
subsets of R. Then the ideal [
N= J
J∈C
is an ideal of R.
P ROOF. If x, y ∈ N, then x ∈ J and y ∈ K for some J, K ∈ C . Then J ∪ K is either J or K, so
is in C , and we then have x − y ∈ J ∪ K, so x − y ∈ N. Thus, N is a subgroup of R under addition.
For a ∈ R and x ∈ N, we have that x ∈ J for some J ∈ C , and then ax and xa are elements of J,
since J is an ideal. In particular, they are also elements of N. Therefore, N is an ideal. 
T HEOREM 3.10.11. Let I be a proper ideal of a ring R. Then there exists a maximal ideal m
of R that contains I.
P ROOF. Let X be the set of proper ideals of R containing I, which we endow with the usual
partial ordering ⊆. Suppose that C ⊂ X is a chain. Consider the ideal
[
N= J
J∈C
of R. Note that 1 ∈ / J for all J ∈ C , so N is proper. In other words, N ∈ X, and it is
/ N since 1 ∈
an upper bound for C . Zorn’s lemma then tells us that X contains a maximal element, which is
necessarily a maximal ideal of R. 
In commutative rings, maximal ideals are part of a broader class of ideals known as prime
ideals.
D EFINITION 3.10.12. Let R be a commutative ring. A proper ideal p of R is said to be a
prime ideal (or prime) if for all b, c ∈ R with bc ∈ p, either b ∈ p or c ∈ p.
94 3. RING THEORY

E XAMPLES 3.10.13.
a. If A is an integral domain, then (0) is a prime ideal.
b. In Z, the prime ideals are exactly (0) and the pZ for p prime. That is, if ab ∈ (p) with p
prime, then p divides ab, so p divides a or p divides b, and hence either a ∈ (p) or b ∈ (p).
We have the following analogue of Theorem 3.10.4.
T HEOREM 3.10.14. Let R be a commutative ring. Then a proper ideal p of R is prime if and
only if R/p is an integral domain.
P ROOF. The ideal p is prime if and only if ab ∈ p implies than a ∈ p or b ∈ p, which translates
to the fact that ab = 0 imples a = 0 or b = 0 in the ring R/p. 
C OROLLARY 3.10.15. Let R be a commutative ring. Then every maximal ideal of R is prime.
P ROOF. If m is a maximal ideal of R, then Theorem 3.10.4 then tells us that R/m is a field.
Theorem 3.10.14 yields that m is prime. 
As for polynomial rings over fields, we have the following theorem.
P ROPOSITION 3.10.16. Let F be a field. The prime ideals in F[x] are exactly (0) and those
( f ) such that f ∈ F[x] is irreducible.
P ROOF. Note that if f is nonconstant and reducible, then f = gh for some nonconstant g, h ∈
F[x] of degree less than deg f , so g, h ∈
/ ( f ). Therefore, ( f ) is not prime.
On the other hand, if f is nonconstant and irreducible, then Proposition 3.10.3 tells us that
( f ) is maximal, and Corollary 3.10.15 then tells us that ( f ) is prime. 
E XAMPLE 3.10.17. In Z[x], the ideal (x) is prime, since Z[x]/(x) ∼ = Z, but (x) is not maximal.
This follows either from the fact that Z is not a field, or the fact that (x) is properly contained in
(p, x) for any prime p.
Prime ideals are preserved under contraction.
L EMMA 3.10.18. Let φ : R → Q be a homomorphism of commutative rings. Then any con-
traction of a prime ideal of Q by φ is a prime ideal of R.
P ROOF. For a prime ideal q of Q, this amounts to the fact that if a, b ∈ R with φ (ab) ∈ q, then
φ (a) ∈ q or φ (b) ∈ q, since q is prime. 
The following lemma is also very useful.
L EMMA 3.10.19.
a. Let p1 , . . . , pk be prime ideals of R. If an ideal I is contained in ki=1 pi , then I is contained
S

in some pi .
Tk
b. Let a1 , . . . , ak be ideals of R. If a prime ideal p contains ∏ki=1 ai (resp., equals i=1 ai ),
then p contains (resp., equals) some ai .
 3.11. FIELDS OF FRACTIONS 95

P ROOF. We prove part a by induction on k, it being clearly true for k = 1. Suppose that I is
not contained in any pi but I is contained in the union of the pi . By induction, for each i, we can
find ai ∈ I such that ai ∈
/ p j for all j 6= i. By assumption, we then have ai ∈ pi for each i. The
element
k k
b = ∑ ∏ ai
i=1 j=1
j6=i
of I has image in R/pi equal to the image of its ith term, which is nonzero by the primality of i.
/ ki=1 pi , which is a contradiction.
That is, b ∈
S

As for part b, let a = ∏ki=1 ai . Suppose that p does not contain any ai , and choose ai ∈ ai with
/ p for each i. Then a = ∏ki=1 ai ∈ a, but a ∈
ai ∈ / p by primality of p. Therefore, p does not contain
Tk
a. If on the other hand p = i=1 ai , then p contains some ai by what we have shown, so must
equal it in that p ⊆ ki=1 ai by assumption.
T


3.11. Fields of fractions

As is seen by the most basic case of the integers Z, not all rings are fields. Yet, Z is contained
in many fields, the smallest being Q, the rational numbers. The field Q consists exactly of
fractions ba , where a and b are integers and b is nonzero. One can ask more generally, given an
ring R, does one have a good notion of a fraction ba with a, b ∈ R and b 6= 0? And, if so, can one
form a field out of them? As we shall, see in the case of an integral domain, the answer is yes.
L EMMA 3.11.1. Let R be an integral domain, and set
X = {(a, b) ∈ R × R | b 6= 0}.
The relation ∼ on X given by (a, b) ∼ (c, d) if and only if ad = bc is an equivalence relation.
P ROOF. For (a, b) ∈ X, we have ab = ba, so (a, b) ∼ (a, b), so ∼ is reflexive. If (c, d) ∈ X
with (a, b) ∼ (c, d), then ad = bc implies cb = da, so (c, d) ∼ (b, a) as well, and ∼ is symmetric.
Finally, if (e, f ) ∈ X as well and (a, b) ∼ (c, d) while (c, d) ∼ (e, f ), we have ad = bc and c f = de.
Multiplying the former equality by f and then applying the latter, we obtain
ad f = bc f = bde.
Since d 6= 0 and R is an integral domain, this implies a f = be, which means that (a, b) ∼ (e, f ),
and therefore ∼ is transitive. 
Note that the last step shows the need for having an integral domain in order to have an
equivalence relation in Lemma 3.11.1.
D EFINITION 3.11.2. Let R be an integral domain. We let Q(R) denote the set of equivalence
classes of elements of X under the relation ∼ of Lemma 3.11.1. The equivalence class of (a, b)
with a, b ∈ R and b 6= 0 will be denoted ab , and it is called the quotient of a by b. By using the
symbol ab , we are implicitly representing the quotient by (a, b), and this representative is called a
fraction. We then refer to a as the numerator of ba and b as the denominator of ba .
The following is immediate.
96 3. RING THEORY

L EMMA 3.11.3. Let R be an integral domain, and let a, b, x ∈ R with b and x nonzero. Then
we have
a ax
=
b bx
in Q(R).
L EMMA 3.11.4. Let R be an integral domain. There are well-defined operations + and · on
Q(R) given by
a c ad + bc
+ =
b d bd
and
a c ac
· = .
b d bd
P ROOF. Let X be as in Lemma 3.11.1. Define + on X by
(a, b) + (c, d) = (ad + bc, bd)
and · on X by
(a, b) · (c, d) = (ac, bd).
To prove the proposition, we must show that if (a, b) ∼ (a0 , b0 ) and (c, d) ∼ (c0 , d 0 ), we have
(a, b) + (c, d) ∼ (a0 , b0 ) + (c0 , d 0 ) and (a, b) · (c, d) = (a0 , b0 ) · (c0 , d 0 ).
We check that
(ad + bc)b0 d 0 = ab0 dd 0 + cd 0 bb0 = ba0 dd 0 + dc0 bb0 = bd(a0 d 0 + b0 c0 )
and
acb0 d 0 = ab0 cd 0 = ba0 c0 d = bda0 c0 ,
as desired. 
C OROLLARY 3.11.5. Let R be an integral domain and a, a0 , b ∈ R with b 6= 0. In Q(R), one
has
a a0 a + a0
+ = .
b b b
P ROOF. Noting Lemma 3.11.3, we have
ab + a0 b (a + a0 )b a + a0
= = .
b2 b2 b

T HEOREM 3.11.6. Let R be an integral domain. Under the operations + and · of Lemma 3.11.4,
the ring Q(R) is a field.
P ROOF. First, we note that addition is commutative since
a c ad + bc cb + da c a
+ = = = + ,
b d bd db d b
 3.11. FIELDS OF FRACTIONS 97

and it is associative since

 a c  e ad + bc e (ad + bc) f + bde
+ + = + =
b d f bd f bd f
ad f + b(c f + de) a c f + de a  c e 
= = + = + + .
bd f b df b d f
Next, we note that
0 a a·1+0·b a
+ = = ,
1 b 1·b b
0
so 0 = 1 in Q(R). We also have
−a a −ab + ab 0 0
+ = 2
= 2= ,
b b b b 1
2
the latter step by noting that 0 · 1 = b · 0 = 0. Hence, Q(R) is an abelian group under addition.
We note that multiplication is associative, as
a c  e ac e ace a ce a c e
· · = · = = · = · · .
b d f bd f bd f b df b d f
We check distributivity as follows:
a  c e  a c f + de ac f + ade ac f ade ac ae a c a e
· + = · = = + = + = · + · .
b d f b df bd f bd f bd f bd b f b d b f
Note that
1 a a
· = ,
1 b b
so 1 = 11 in Q(R). Finally, note that ba 6= 0 = 01 if and only if a 6= 0, and in this case we can form
b
a . We then have
b a ab 1
· = = = 1,
a b ab 1
so ab = ( ab )−1 . Therefore, Q(R) is a field. 
D EFINITION 3.11.7. Let R be an integral domain. The field Q(R) is called the quotient field,
or the field of fractions, of R.
R EMARK 3.11.8. The field Q(R) is not a quotient of R in the sense it is the set of equivalence
classes for an equivalence relation on R itself. Rather, it is a set of quotients of elements of R in
the sense of division, and in fact it contains R. That is, quotient rings and quotient fields are quite
different should not be confused with each other.
D EFINITION 3.11.9. Let F be a field. The field F(x) of fractions of F[x] is called the field of
rational functions in one variable over F.
x+1 x2 −x
E XAMPLE 3.11.10. The fraction x2 +1
is an element of Q(x), as is x2 +1
, and
x + 1 x2 − x x2 + 1
+ = = 1.
x2 + 1 x2 + 1 x2 + 1
98 3. RING THEORY

The following theorem says, in essence, that Q(R) is the smallest field containing R.
T HEOREM 3.11.11. Let R be an integral domain.
a. The map ιR : R → Q(R) given by ιR (r) = 1r is an injective ring homomorphism. We use it
to identify R with a subring of Q(R), setting r = 1r .
b. If F is any field containing R, then is a unique injective ring homomorphism Q(R) → F
that restricts to the inclusion map R → F.
P ROOF. That ιR is a ring homomorphism is easily checked, and it is injective since 1r = 01
implies by definition that r = 0. Now, suppose that R is contained a field F. Define θ : Q(R) → F
by a
θ = ab−1 .
b
This is well-defined, as if ad = bc for some c, d ∈ R with d 6= 0, then ab−1 = cd −1 . Moreover,
for any quotients ab and dc in Q(R), we have
a c  a c
θ + = (ad + bc)(bd)−1 = (ad + bc) · (bd)−1 = ab−1 + cd −1 = θ +θ
b d b d
and a c  a  c 
−1 −1 −1
θ · = ac · bd = ab cd = θ ·θ ,
b d b d
so θ is a ring homomorphism. If θ ( ba ) = 0, then ab−1 = 0, which implies that a = 0, and hence
a
b = 0. Therefore, θ is injective. Also, note that
a
θ (a) = θ = a · 1−1 = a.
1
Finally, if χ : Q(R) → F is any homomorphism with which restricts to the inclusion map R → F,
then we have a a 1 a
χ =χ · = χ(a) · χ(b)−1 = ab−1 = θ ,
b 1 b b
so χ = θ . 
Let us make our comment prior to the theorem more precise.
C OROLLARY 3.11.12. Let R be an integral domain and K a field containing it. Then there is
a smallest subfield F of K containing R, and it is isomorphic to the field of fractions of R via a
map Q(R) → K that extends the identity map on R.
P ROOF. The smallest field F containing R is simply the intersection of all fields contained
in K and containing R. We then apply Theorem 3.11.11 to the inclusion map ιR : R → F. The
image of the induced map θ : Q(R) → F is a field containing R and contained in K, so must be
F itself. 
Corollary 3.11.12 allows us think more concretely about fields of fractions by speaking of
fields of fractions inside a given field.
D EFINITION 3.11.13. Let R be an integral domain and K a field containing it. The field of
fractions of R in K is the smallest subfield F of K containing R.
 3.11. FIELDS OF FRACTIONS 99

The next corollary tells us that it’s okay to think of elements of a field F of the form ab−1
with a ∈ F and b ∈ F × as fractions ba .
C OROLLARY 3.11.14. If F is a field, then it is isomorphic to its own field of fractions.
P ROOF. By Corollary 3.11.12, there is a field containing F in F, which of course is F itself,
that is isomorphic to the field of fractions Q(F). 
In other words, the field of fractions of F is F.
C OROLLARY 3.11.15. Let R and Q be integral domains, and let φ : R → R0 be an injective
ring homomorphism. Then there is a unique homomorphism Q(φ ) : Q(R) → Q(R0 ) such that
Q(φ )(a) = φ (a) for all a ∈ R.
P ROOF. As the composite map ιR0 ◦ φ : R → Q(R0 ) is injective, Theorem 3.11.11 tells us that
there is a unique injective homomorphism Q(φ ) : Q(R) → Q(R0 ) with Q(φ ) ◦ ιR = ιR0 ◦ φ , as
desired. 
E XAMPLE 3.11.16. The quotient field of Z[i] is isomorphic to Q(i). To see this, note that
Q(i) is a field containing Z[i], and so there is an inclusion homomorphism Q(Z[i]) → Q(i) that
takes a fraction of the form a+bi
c+di with a, b, c, d ∈ Z and (c, d) 6= (0, 0) to itself, but every element
in Q(i) has the form q + ri with q, r ∈ Q, and any such element can be written as such a fraction
with d = 0.
E XAMPLE 3.11.17. The quotient field of Z[x] is Q(x), the quotient field of Q[x]. To see this,
note that the inclusion map α : Z[x] → Q[x] sending a polynomial to itself induces an injective
homomorphism Q(α) : Q(Z[x]) → Q(x) by Corollary 3.11.15. Moreover, for f , g ∈ Z[x], we
have Q(α)( gf ) = gf by definition. If f , g ∈ Q[x], then there exists a nonzero a ∈ Z such that
a f , ag ∈ Z[x]. (Here, a is the least common multiple of the denominators of the coefficients of P
and Q, written as fractions in lowest terms.) Then QP = aQ aP
in Q(x), so QP is in the image of the
map Q(α). Therefore, Q(α) is an isomorphism.
 CHAPTER 4

Advanced group theory

4.1. Isomorphism theorems

We have already proven the first isomorphism theorem. In this section, we shall use it in
deriving two others.
D EFINITION 4.1.1. Let H and K be subgroups of a group G. We define the join H ∨ K of H
and K to be the subgroup of G generated by H ∪ K.
R EMARK 4.1.2. Note that the join H ∨ K contains (and is generated by) the set
HK = {hk | h ∈ H, k ∈ K}
and H ∨ K = HK and only if HK 6 G.
E XAMPLE 4.1.3. Take H = h(1 2)i and K = h(1 3)i as subgroups of S3 . We have
HK = {e, (1 2), (1 3), (1 3 2)},
which is not a subgroup of S3 , while H ∨ K = S3 .
L EMMA 4.1.4. We have that HK 6 G if and only if HK = KH.
P ROOF. Suppose first that HK 6 G. Let h ∈ H and k ∈ K. Since h, k ∈ HK, we have kh ∈ HK,
as HK 6 G. Thus KH ⊆ HK. On the other hand, we have (hk)−1 = k−1 h−1 ∈ KH, so the inverse
of every element of HK is contained in KH. But every element of HK is the inverse of some
element of HK since HK 6 G, so HK ⊆ KH as well. Thus, we have HK = KH.
Now suppose that HK = KH. We always have e = e · e ∈ HK. Moreover, if h, h0 ∈ H and
0
k, k ∈ K, then
hk · h0 k0 = h(kh0 )k0 ,
and since KH = HK, there exists h00 ∈ H, k00 ∈ K such that kh0 = h00 k00 , so
h(kh0 )k0 = h(h00 k00 )k0 = hh00 · k00 k0 ∈ HK.
Moreover, we have
(hk)−1 = k−1 h−1 ∈ KH,
but KH = HK, so (hk)−1 ∈ HK. Thus, we have that HK 6 G. 
C OROLLARY 4.1.5. Suppose that H and N are subgroups of G with N normal. Then we have
HN 6 G. If H is normal in G as well, then we have HN P G.
101
102 4. ADVANCED GROUP THEORY

P ROOF. By Lemma 4.1.4, it suffices to show that HN = NH. But N P G, so hN = Nh for all
h ∈ H, which means that [ [
HN = hN = Nh = NH.
h∈H h∈H
Moreover, if H P G, then for any g ∈ G, we have
gHNg−1 = gHg−1 · gNg−1 = HN.

T HEOREM 4.1.6 (Second Isomorphism Theorem). Let H be a subgroup of a group G, and
let N be a normal subgroup of G. Then we have an isomorphism
H/(H ∩ N) −∼→ HN/N.
P ROOF. Define
φ : H → HN/N, φ (h) = hN.
Then
ker φ = {h ∈ H | h ∈ N} = H ∩ N.
Moreover, if h ∈ H and n ∈ N, then hnN = hN = φ (h), so φ is surjective. The result therefore
follows by the first isomorphism theorem. 
E XAMPLE 4.1.7. Consider the subgroups H = Z × Z × {0} and N = {0} × Z × Z of G = Z3 .
We have HN = G, and H ∩ N = {0} × Z × {0}. Note that
Z × Z × {0} ∼
H/(H ∩ N) = −→ Z
{0} × Z × {0}
via the map that takes (a, b, 0)(H ∩ N) to a. On the other hand, we have
Z×Z×Z ∼
HN/N = −→ Z
{0} × Z × Z
via the map that takes (a, b, c)N to a.
R EMARK 4.1.8. Suppose that H and K are subgroups of a group G with K 6 H. If K and H
are both normal subgroups of G, then K P H. On the other hand, the property of being a normal
subgroup is not transitive. One may have K P H and H P G but K 6P G!
E XAMPLE 4.1.9. Take G = A4 ,
H = h(1 2)(3 4), (1 3)(2 4)i,
and K = h(1 2)(3 4)i. Since
(1 2)(3 4) · (1 3)(2 4) = (1 4)(2 3) = (1 3)(2 4) · (1 2)(3 4),
the group H is an abelian group of order 4 (isomorphic to the Klein four-group) consisting of the
three cycles of cycle type (2, 2) and the identity. Now K C H since H is abelian, and H C G since
conjugation preserves cycle type. On the other hand, K is not a normal subgroup of G since
(1 2 3) · (1 2)(3 4) · (1 2 3)−1 = (1 4)(2 3) ∈
/ K.
 4.1. ISOMORPHISM THEOREMS 103

T HEOREM 4.1.10 (Third Isomorphism Theorem). Let H and K be normal subgroups of a

group G with K 6 H. Then we have an isomorphism
G/H −∼ → (G/K)/(H/K).
P ROOF. We first remark that H/K 6 G/K since it is a subset of G/K that is a group under
the operation on G/K. Moreover, H/K P G/K since if h ∈ H and a ∈ G, then aha−1 = h0 for
some h0 ∈ H, so
aK · hK · a−1 K = h0 K ∈ H/K.
We may now define
θ : G → (G/K)/(H/K)
by
θ (a) = (aK)(H/K).
By the group laws on G/K and (G/K)/(H/K), we have
θ (ab) = (abK)(H/K) = (aK · bK)(H/K) = (aK)(H/K) · (bk)(H/K) = θ (a)θ (b).
Then θ (a) = H/K if and only if aK = hK for some h ∈ H, so if and only if a ∈ H. Thus
ker θ = H. On the other hand, θ is surjective by definition. The result now follows from the first
isomorphism theorem. 
R EMARK 4.1.11. For H, K and G as in the third isomorphism theorem, the composite map
G → G/K → (G/K)/(H/K) → G/H,
where the first and second maps are quotient maps and the third is the inverse of the isomorphism
in the third isomorphism theorem, is exactly the quotient map G → G/H.
E XAMPLE 4.1.12. Let G = Z, H = mZ and K = nZ, where m, n ≥ 1 and m divides n, so that
KhH. Then G/H = Z/mZ, G/K = Z/nZ, and H/K = mZ/nZ. We note that
Z/nZ Z/nZ ∼
(G/K)/(H/K) = = −→ Z/mZ = G/H,
mZ/nZ hmi
the map in the last isomorphism being induced by the natural reduction-modulo-m map from
Z/nZ to Z/mZ and the first isomorphism theorem.
E XAMPLE 4.1.13. Let G = Z3 , H = Z × Z × {0} and K = Z × {0} × {0}. Then G/H ∼ = Z via
the map that takes (a, b, c)H to c, while G/K ∼
= Z × Z via the map that takes (a, b, c)K to (b, c),
and H/K has image Z × {0} under this map. Then
Z×Z ∼
(G/K)/(H/K) −∼
→ −→ Z,
Z × {0}
where the latter map takes (b, c)(Z × Z) to c.
We also have the following, known as the butterfly (or Zassenhaus) lemma, which we state
without proof.
104 4. ADVANCED GROUP THEORY

T HEOREM 4.1.14 (Butterfly lemma). Let H, K, A, B be subgroups of a group G with A P H

and B P K. Then there is a canonical isomorphism
A(H ∩ K) ∼ B(H ∩ K)
= .
(A(H ∩ B) B(A ∩ K)

4.2. Commutators and simple groups

D EFINITION 4.2.1. Let G be a group and a, b ∈ G. The commutator of a and b is
[a, b] = aba−1 b−1 .
D EFINITION 4.2.2. The commutator subgroup [G, G] of a group G is the subgroup of G
generated by its commutators, which is to say
[G, G] = h[a, b] | a, b ∈ Gi.
R EMARK 4.2.3. If G is an abelian group, then [G, G] = {e}.
E XAMPLE 4.2.4. In Dn , we have
[ri , r j s] = ri (r j s)(r−i )(sr− j ) = r2i ,
[ri s, r j s] = (ri s)(r j s)(sr−i )(sr− j ) = ri− j sr j−i s = r2(i− j) .
Therefore, we have that
[Dn , Dn ] = hr2 i,
which has index 2 and 4 in Dn in the cases that n is odd and even, respectively.
E XAMPLE 4.2.5. We have
[GLn (R), GLn (R)] ⊆ SLn (R).
since det(ABA−1 B−1 ) = 1 for any A, B ∈ GLn (R). The opposite equality also holds, but we shall
not prove it here.
L EMMA 4.2.6. The commutator subgroup of G is a normal subgroup of G.
P ROOF. Let a, b, g ∈ G. We have
g[a, b]g−1 = gaba−1 b−1 g−1 = (ga)b(ga)−1 b−1 · bgb−1 g−1 = [ga, b][b, g] ∈ [G, G].
Since every element of [G, G] is a product of powers of elements of the form [a, b] with a, b ∈ G
and every such element is sent to another element of [G, G] by the conjugation homomorphism
γg , the image of γg is contained in [G, G]. Since this holds for all g ∈ G, we have that [G, G] is
normal. 
E XAMPLE 4.2.7. Let n ≥ 3. We claim that [Sn , Sn ] = An . Note that
[σ , (a b)] = σ (a b)σ −1 (a b) = (σ (a) σ (b))(a b)
for σ ∈ Sn and a, b ∈ Xn with a < b. It follows that [G, G] contains all products of two trans-
posiitons in Sn . Moreover, these generate An by definition, so An ⊆ [Sn , Sn ]. Furthermore, every
 4.2. COMMUTATORS AND SIMPLE GROUPS 105

element of [Sn , Sn ] is even as every such element is a product of elements of the form [σ , τ] with
σ , τ ∈ Sn , and these satisfy
sign([σ , τ]) = sign(σ ) sign(τ) sign(σ )−1 sign(τ)−1 = 1.
T HEOREM 4.2.8. Let N be a normal subgroup of G. Then G/N is an abelian group if and
only if [G, G] 6 N.
P ROOF. Let a, b ∈ G. We have abN = baN if and only if a−1 b−1 ab ∈ N, so if and only if
[a−1 , b−1 ] ∈ N.
But [G, G] is the smallest subgroup of G containing [c, d] for every c, d ∈ G, so N
is normal if and only if [G, G] is contained in N. 
D EFINITION 4.2.9. The maximal abelian quotient, or abelianization, Gab of a group G is the
quotient group
Gab = G/[G, G].
We have the following consequence of Theorem 4.2.8.
C OROLLARY 4.2.10. Let G be a group and H be an abelian group, and suppose φ : G → H
is a homomorphism. Then there exists a homomorphism φ̄ : Gab → H with φ̄ ◦ π[G,G] = φ , where
π[G,G] : G → Gab is the quotient map.
P ROOF. By the first isomorphism theorem, there exists a unique map ψ : G/ ker φ → H with
ψ ◦ πker φ = φ . By Theorem 4.2.8, we have that [G, G] ⊆ ker φ . and now the third isomorphism
theorem provides a composite map
β : Gab = G/[G, G] → (G/[G, G])/(ker φ /[G, G]) −∼→ G/ ker φ
such that πker φ = β ◦ π[G,G] . Set φ̄ = ψ ◦ β . Then
φ̄ ◦ π[G,G] = ψ ◦ β ◦ π[G,G] = ψ ◦ πker φ = φ ,
as desired. 
E XAMPLE 4.2.11. The abelianization of Dn is Dn /hr2 i, and if n is even this group is isomor-
phic to Z/2Z × Z/2Z under the map φ̄ induced by the homomorphism
φ : Dn → Z/2Z × Z/2Z
that takes ri s j to (i, j). If n is odd, then Dn /hr2 i is isomorphic to Z/2Z via the map from Dn that
takes ri s j to j.
E XAMPLE 4.2.12. We have Snab = Sn /An , and so Snab ∼ = Z/2Z via the map induced by the sign
map.
Here is another nice class of normal subgroups.
D EFINITION 4.2.13. The center of a group G is the subgroup
Z(G) = {z ∈ G | za = az for all a ∈ G}.
Since az = za for all z ∈ Z(G) and a ∈ G, we clearly have that aZ(G) = Z(G)a for all a ∈ G,
and hence we have that Z(G) P G. We leave the verification of the following examples to the
reader.
106 4. ADVANCED GROUP THEORY

E XAMPLES 4.2.14.
a. If G is abelian, then Z(G) = G.
b. For n ≥ 3, we have Z(Sn ) = {e}.
c. For n ≥ 3, we have Z(Dn ) = hrn/2 i if n is even and Z(Dn ) = {e} if n is odd.
d. For n ≥ 2, we have that Z(GLn (R)) is the subgroup of scalar matrices.
D EFINITION 4.2.15. A nontrivial group G is called simple if it has no nontrivial, improper
normal subgroups.
E XAMPLE 4.2.16. An abelian group G is simple if and only if it is cyclic of prime order,
since otherwise it will have a nontrivial, improper subgroup, which is automatically normal since
G is abelian.
E XAMPLES 4.2.17. The groups Sn and Dn for n ≥ 3 are not simple, since they contain im-
proper, nontrivial normal subgroups. Moreover, A4 is not simple, as it contains the normal sub-
group h(1 2)(3 4), (1 3)(2 4)i of order 8, as seen in Example 4.1.9.
We have the following easy lemma.
L EMMA 4.2.18. If G is simple and nonabelian, then Z(G) = {e}.
P ROOF. If G is nonabelian, then Z(G) 6= G, and if G is also simple, then since Z(G) is normal,
we must have Z(G) = {e}. 
We note that if a group is not simple, we can find a nontrivial quotient of it by a nontrivial
normal subgroup that is.
D EFINITION 4.2.19. We say that a normal subgroup M of a group G is a maximal normal
subgroup if it is not contained in any larger proper normal subgroup of G.
E XAMPLE 4.2.20. Any subgroup of index 2 in a group is a maximal normal subgroup, since
such a subgroup is normal and is not contained in a larger proper normal subgroup, being that its
index would have to be smaller than 2, but greater than 1.
P ROPOSITION 4.2.21. Let N be a normal subgroup of G. Then G/N is simple if and only if
N is maximal.
P ROOF. This is an immediate consequence of Proposition 2.13.10, since G/N is simple if
and only if it has no proper normal subgroups, which are in bijection with the proper normal
subgroups of G containing N. 
E XAMPLE 4.2.22. Since An is a maximal normal subgroup of Sn , the quotient Sn /An is simple
(which we already knew since it is isomorphic to Z/2Z).
Finite simple groups are considered the building blocks of all finite groups. Their classifica-
tion was the major project in group theory during the 20th century, and it was finally finished at
the beginning of the 21st. Many examples of nonabelian finite simple groups are simple enough
to give, though proving they are simple is another matter.
 4.3. AUTOMORPHISM GROUPS 107

E XAMPLE 4.2.23. The groups An are simple for all n ≥ 5. In fact, A5 is a finite simple group
with the smallest possible order, which is 60, and it is the unique such group up to isomorphism.
We defer the proofs of these facts until later.
We mention one more broad class of examples of finite simple groups.
E XAMPLE 4.2.24. We remark that Z/pZ has two binary operations of addition and multipli-
cation, and these satsify the distributive property. Hence we may consider the set GLn (Z/pZ) of
invertible n by n matrices with entries in Z/pZ, and this forms a group under multiplication. We
also have its subgroup SLn (Z/pZ) of matrices with determinant 1. It is not necessarily simple,
as it is possible that it can have nontrivial center: the group of scalar matrices with determinant
1. I.e., aI ∈ SLn (Z/pZ) if and only if an = 1 in Z/pZ. The quotient of SLn (Z/pZ) by its center
is called PSLn (Z/pZ). It turns out that PSLn (Z/pZ) is simple for all primes p for all n ≥ 3 and
for all primes p ≥ 5 and n = 2.
4.3. Automorphism groups
D EFINITION 4.3.1. An automorphism of a group G is an isomorphism φ : G → G.
The subgroup test shows quickly that the set of automorphisms of G forms a subgroup of G
under composition.
P ROPOSITION 4.3.2. The set of automorphisms Aut(G) of a group G forms a group under
composition.
P ROOF. Since composition of functions is associative, to check that Aut(G) is a group, we
need only check that it contains an identity element, which it clearly does, and that it contains
inverses, which is does since the inverse of an isomorphism is an isomorphism. 
D EFINITION 4.3.3. The automorphism group Aut(G) of a group G is the group of automor-
phisms of G under conjugation.
E XAMPLES 4.3.4.
a. We have Aut(Z/nZ) ∼ = (Z/nZ)× for n ≥ 1 via the map that takes φ ∈ Aut(Z/nZ) to φ (1).
In fact, we have φ (a) = aφ (1) for any a ∈ Z/nZ, and so φ is multiplication by φ (1). This can be
invertible if and only if a is a unit in Z/nZ.
b. The same discussion as in part a tells us that Aut(Z) = Z× = h−1i.
c. We have Aut(Zn ) ∼ = GLn (Z). That is, if φ ∈ Aut(Zn ) and ei is the ith element in the
n
standard basis of Z (with a 1 in its ith entry and 0 in the others), then φ (ei ) determines the ith
column of a matrix in GLn (Z). The inverse map is given by allowing GLn (Z) to act on Zn by
left multiplication, viewing an element of Zn as a column vector.
We give the example of the automorphisms of the dihedral group as a proposition.
P ROPOSITION 4.3.5. For n ≥ 3, the group Aut(Dn ) is isomorphic to the subgroup Aff(Z/nZ)
of GL2 (Z/nZ) given by   
a b ×
a ∈ (Z/nZ) , b ∈ Z/nZ .
0 1
108 4. ADVANCED GROUP THEORY

P ROOF. The isomorphism f : Aff(Z/nZ) → Aut(Dn ) is given by

 
a b
f: 7→ φa,b ,
0 1
where φa,b (r) = ra and φa,b (s) = rb s. Since Dn is generated by r and s, there exists at most one
element of Aut(Dn ) taking these values on r and s. Since Fr,s is free, we can define Φ : Fr,s → Dn
by Φ(r) = ra and Φ(s) = rb s for a and b as above. Note that Φ(rn ) = ran = e, Φ(s2 ) = (rb s)2 = e,
and Φ((rs)2 ) = (ra+b s)2 = e, so by the presentation Dn ∼
= hr, s | rn , s2 , rsrsi, we have the existence
of φa,b . In that a is invertible modulo n, we have
hra , rb si = hr, rb si = hr, si = Dn ,
so φa,b is onto and hence in Aut(Dn ) as Dn is finite.
Now, any φ ∈ Aut(Dn ) must send r to another element of order n, so ra with a ∈ Z prime
to n. It must also send s to an element of order 2 that cannot be in the subgroup hφ (r)i = hri,
since φ is surjective. Thus, φ (s) = rb s for some b ∈ Z. Thus, f is onto, and it is one-to-one by
definition. To see it is a homomorphism, note that
  0 0  0
aa b + ab0
 
a b a b
= ,
0 1 0 1 0 1
while
0 0 0 0
φa,b (φa0 ,b0 (r)) = φa,b (ra ) = raa and φa,b (φa0 ,b0 (s)) = φa,b (rb s) = rab +b s.

D EFINITION 4.3.6. An automorphism of G is called an inner automorphism (or inner) if it is
equal to a conjugation map γa : G → G for some a ∈ G.
L EMMA 4.3.7. The set Inn(G) of inner automorphisms of G is a subgroup of Aut(G) under
composition.
P ROOF. That the inner automorphisms form a subgroup amounts to the facts that γe = idG ,
that γab = γa γb , and that γa−1 = γa−1 for a ∈ G. 
D EFINITION 4.3.8. The inner automorphism group Inn(G) is the subgroup of G consisting
of inner automorphisms.
L EMMA 4.3.9. For a group G, the inner automorphism group Inn(G) is a normal subgroup
of G.
P ROOF. For φ ∈ Aut(G) and g, x ∈ G, we have
(φ ◦ γg ◦ φ −1 )(x) = φ (gφ −1 (x)g−1 ) = φ (g)xφ (g)−1 = γφ (g) (x),
so φ γg φ −1 = γφ (g) lies in Inn(G). 
D EFINITION 4.3.10. The outer automorphism group of G is the quotient group Out(G) =
Aut(G)/ Inn(G).
 4.3. AUTOMORPHISM GROUPS 109

R EMARK 4.3.11. An automorphism is sometimes called outer if it is not inner. However, the
outer automorphism group is not a group of automorphisms, but rather cosets thereof.
R EMARK 4.3.12. If G is an abelian group, then every inner automorphism of G is trivial, so
Out(G) ∼
= Aut(G).
E XAMPLE 4.3.13. The group Inn(Dn ) for n ≥ 3 is generated by the images γr and γs of r
and s under γ : Dn → Aut(Dn ). We have γr (s) = r2 s and γs (r) = r−1 , and of course γr (r) = r
and γs (s) = s. Using the isomorphism of Proposition 4.3.5, we that Inn(G) is isomorphic to the
subgroup of GL2 (Z/nZ) given by
  
a b
K= a = ±1, b ∈ 2Z/nZ .
0 1
The quotient group Out(Dn ) is then in bijection with pairs (i, j) ∈ (Z/nZ)× /h−1i × Z/(n, 2)Z.
We leave it to the reader to check that
(
×
Out(Dn ) = ∼ (Z/nZ) /h−1i if n is odd,
×
(Z/nZ) /h−1i × Z/2Z if n is even
 
using the fact that 10 n/2
1
is in the center of the group H of Proposition 4.3.5 if n is even.

D EFINITION 4.3.14. A subgroup N of a group G is characteristic if φ (N) = N for every

φ ∈ Aut(G).
L EMMA 4.3.15. Let G be a group.
a. If H is the unique subgroup of G of a given order, then H is characteristic.
b. The center Z(G) of a group G is characteristic.
c. The commutator subgroup [G, G] of a group G is characteristic.
P ROOF. Let φ ∈ Aut(G). For part a, note that φ (H) has the same order as H. For part b,
note that φ (a)φ (x) = φ (x)φ (a) for any a ∈ Z(G) and x ∈ G, but φ is onto, so φ (a) commutes
with every element of G. So, φ : Z(G) → Z(G), and φ −1 ∈ Aut(G) has the same property, so
φ (Z(G)) = Z(G). For part c, note that [G, G] is generated by commutators [a, b] with a, b ∈ G,
and φ ([a, b]) = [φ (a), φ (b)] ∈ [G, G]. We can see that φ ([G, G]) actually equals [G, G] by noting
that φ is onto. 
L EMMA 4.3.16. If K is a characteristic subgroup of a normal subgroup N of a group G, then
K P G. If, moreover, N is characteristic in G, then K is characteristic in G.
P ROOF. Let a ∈ G. Then the restriction of γa to N provides an element of Aut(N) as N is
normal, and so aKa−1 = γa (K) = K as K is characteristic. Thus K P G.
If N is characteristic in G and φ ∈ Aut(G), then the restriction of φ to N is an automorphism of
N as N is characteristic in G, and so φ (K) = K as K is characteristic in N. Thus K is characteristic
in G. 
110 4. ADVANCED GROUP THEORY

4.4. Free abelian groups

The theory of free abelian groups is the analogue of the theory of vector spaces when the
scalars are taken to be not real or complex numbers, but rather integers. In this section, we
briefly explore this theory.
D EFINITION 4.4.1. An abelian group G (under addition) is said to be a free abelian group if it
has a generating set X of G such that for any n ≥ 1, distinct x1 , x2 , . . . , xn ∈ X, and c1 , c2 , . . . , cn ∈ Z
with
n
∑ cixi = 0,
i=1
one has c1 = c2 = · · · = cn = 0. Such a set X is called a basis of G, and G is said to be free on X.
E XAMPLE 4.4.2. The group Zn is free on the set {e1 , e2 , . . . , en }, where ei ∈ Zn is the tuple
that is 0 in every coordinate but the ith, where it is 1.
E XAMPLE 4.4.3. The group Z/nZ is not free for n ≥ 1, since one has na = 0 for every
a ∈ Z/nZ.
R EMARK 4.4.4. Much as in linear algebra, freeness of an abelian group G on a set X implies
that there is a unique way to represent any nonzero element a ∈ G as a sum
n
a = ∑ ci xi
i=1
for some n ≥ 1, distinct elements x1 , x2 , . . . , xn of X, and nonzero elements c1 , c2 , . . . , cn of Z.
D EFINITION 4.4.5. If x1 , x2 , . . . , xn ∈ G, where G is a free abelian group, then we refer to a
sum
n
∑ cixi
i=1
with c1 , c2 , . . . , cn ∈ Z as an integral linear combination of elements of G.
Let us begin with a very general construction of a direct sum of groups, which we will then
specialize immediately to the case of interest that the groups are all Z.
D EFINITION 4.4.6. Let I be an indexing set and {Gi | i ∈ I} a collection of abelian groups.
Let M n o
Gi = (ai )i∈I ∈ ∏ Gi | ai ∈ Gi , ai = 0 for all but finitely many i ∈ I .
i∈I i∈I
L
Then i∈I Gi is a subgroup of ∏i∈I Gi known as the direct sum of the groups Gi .
L
R EMARK 4.4.7. When I is finite, we have i∈I Gi = ∏i∈I Gi .
N OTATION 4.4.8. The symbol δi, j (or δi j ), for i and j in some set I, is taken to mean
(
1 if i = j
δi j =
0 if i 6= j.
 4.4. FREE ABELIAN GROUPS 111

E XAMPLE 4.4.9. For any indexing set I, the direct sum

M n o
Z = (ai )i∈I ∈ ∏ Z | ai ∈ Z, ai = 0 for all but finitely many i ∈ I .
i∈I i∈I

is a free group under coordinate-wise addition with basis {e j = (δi j )i∈I | j ∈ I}.
D EFINITION 4.4.10. The basis {ei | i ∈ I} in Example 4.4.9 is known as the standard basis
L
of i∈I Z.

Free groups have the property that homomorphisms are defined uniquely by their values on
a basis, as we now show.
P ROPOSITION 4.4.11. Let G be an abelian group. Then G is free on a subset X if and only if,
for every function φ̄ : X → G0 , where G0 is an abelian group, there exists a unique homomorphism
φ : G → G0 with φ (x) = φ̄ (x) for all x ∈ X.
P ROOF. Suppose first that G is free on a basis X. Then for c1 , c2 , . . . , cn ∈ Z and distinct
elements x1 , x2 , . . . , xn ∈ X, define
 n  n
φ ∑ ci xi = ∑ ci φ̄ (xi ).
i=1 i=1
The map φ is then a well-defined map on all of G by Remark 4.4.4, and it is easy to check that
it is a homomorphism. Moreover, if ψ : G → G0 is any homomorphism with ψ(x) = φ̄ (x) for all
x ∈ X, then
 n  n n  n 
ψ ∑ ci xi = ∑ ci ψ(xi ) = ∑ ci φ̄ (xi ) = φ ∑ ci xi .
i=1 i=1 i=1 i=1
Conversely, suppose that G and X have the property of the proposition. We claim that G is
free on X. First, suppose that xi ∈ X and ci ∈ Z for 1 ≤ i ≤ n and some n ≥ 1 are such that
n
a = ∑ ci xi = 0.
i=1

Define φ̄ : X → Zn by φ̄ (xi ) = ei . Then

n
0 = φ̄ (a) = ∑ ci ei ,
i=1
which forces ci = 0 for all i, as the ei form a basis of Zn .
Next, let H be the subgroup of G generated by X. We define two homomorphisms G → G.
One is given by idG , while is the composition of the map π : G → H uniquely determined by
π(x) = x for all x ∈ X with the inclusion map ι : H → G. By assumption, then, we must have
idG = ι ◦ π, and as the latter map has image H, we have G = H. Thus, G is free on X. 
R EMARK 4.4.12. The existence of unique homomorphisms of a free abelian group G with
prescibed values on a basis X, as found in Proposition 4.4.11, is often referred to as the universal
property of G.
112 4. ADVANCED GROUP THEORY

C OROLLARY 4.4.13. Suppose that G is a free abelian group on a basis X = {xi | i ∈ I}, where
I is an indexing set. Then there is a unique isomorphism
Z −∼
M
ψ: →G
i∈I
such that ψ(ei ) = xi for all i ∈ I, where {ei | i ∈ I} is the standard basis of G.
P ROOF. We can define ψ as in the statement of the corollary by Proposition 4.4.11, which
also implies the existence of a unique homomorphism φ : G → i∈I Z such that φ (xi ) = ei for
L

all i ∈ I. Since φ ◦ ψ(ei ) = ei and ψ ◦ φ (xi ) = xi for all i ∈ I, the same proposition implies that
φ ◦ ψ and ψ ◦ φ are the identity homomorphisms. In particular, ψ is an isomorphism. 
D EFINITION 4.4.14. The general linear group GLn (Z) of degree n is the group of n-by-n ma-
trices with integer entries which have inverses with integer entries, with respect to the operation
of matrix multiplication.
R EMARK 4.4.15. A n-by-n matrix A with integer entries has an inverse with integer entries
if and only if det(A) = ±1.
For two free abelian groups to be isomorphic, their bases must have the same cardinality.
Equivalently, an abelian group cannot have bases of two different cardinalities. We prove this
only in the special case of finitely generated abelian groups.
T HEOREM 4.4.16. Suppose that G is a free abelian group with basis X having n elements.
Then every basis of G has n elements.
P ROOF. By Corollary 4.4.13, we have that G ∼
= ∏ni=1 Z. We then have that
n
G/2G ∼
= ∏ Z/2Z,
i=1
and so has order 2n .
If G had a different basis with a finite number of elements m, then G/2G
would have order 2m , forcing m = n. On the other hand, if G had an infinite basis indexed by a
set I, then the same argument would tell us that
G/2G ∼
M
= Z/2Z,
i∈I
which is infinite, so impossible. 
D EFINITION 4.4.17. If G is a finitely generated, free abelian group, we refer to the number
of elements in any basis of it as its rank.
We have the following analogue of the change-of-basis theorem in linear algebra.
P ROPOSITION 4.4.18. Let G be a free abelian group with basis X = {x1 , x2 , . . . , xn }. Then
X0 = {x10 , x20 , . . . , xn0 } is also a basis of G if and only if there exists a matrix A = (ai j ) ∈ GLn (Z)
such that
n
xi0 = ∑ ai j x j
j=1
for each 1 ≤ i ≤ n.
 4.5. FINITELY GENERATED ABELIAN GROUPS 113

P ROOF. Since X generates G, we may write each xi0 as

n
xi0 = ∑ ai j x j
j=1

for some ai j ∈ Z and then form an n-by-n matrix A = (ai j ). If X 0 also generates G, then we may
write
n n n
xi = ∑ bi j x0j = ∑ ∑ bi j a jk xk
j=1 j=1 k=1
for some bi j ∈ Z and then form B = (bi j ). Since X is a basis, this tells us that BA = I, so
A ∈ GLn (Z).
Conversely, if there exists a B with BA = I, then
n  n  n
xi = ∑ ∑ bi j a jk xk = ∑ bi j x0j ,
k=1 j=1 j=1

so the x0j generate G, and moreover they form a basis as, if

n
∑ cixi0 = 0,
i=1
we then have
n n
∑ ∑ ai j cix j = 0,
i=1 j=1
so
n
∑ ai j ci = 0
i=1
for each i, or in other words the vector c = (c1 , c2 , . . . , cn ) satisfies Ac = 0, which means BAc = 0,
or c = 0. 

4.5. Finitely generated abelian groups

We begin with the following corollary of Theorem 4.4.18.
L EMMA 4.5.1. Suppose that X = {x1 , x2 , . . . , xn } is a basis of a free abelian group G, and let
ci ∈ Z for 2 ≤ i ≤ n. Then X 0 = {x10 , x2 , . . . , xn } with
x10 = x1 + c2 x2 + · · · + cn xn
is also a basis of G.
P ROOF. Take A ∈ GLn (Z) to be A = I + ∑nk=2 c j E1 j , where Ei j is the n-by-n matrix with
exactly one nonzero entry, which is a 1 in the ith row and jth column. It is easy to see that
det(A) = 1, so A ∈ GLn (Z). We then apply Proposition 4.4.18. 
We are now ready to prove the following result.
114 4. ADVANCED GROUP THEORY

L EMMA 4.5.2. Let G be a finitely generated, free abelian group of rank n, and let H be a
nontrivial subgroup. Then there is an isomorphism
ϕ : G → Z × G0 ,
where G0 is a subgroup of G that is free abelian of rank n − 1, such that
ϕ(H) = dZ × H 0 ,
for some d ≥ 1, where H 0 = H ∩ G0 .
P ROOF. Consider the set B of all bases of G. Let d ≥ 1 be minimal such that there exists
X 0 = {x0 , x2 , . . . , xn } ∈ B such that
n
y = dx0 + ∑ d j x j ∈ H
i=2
for some d2 , . . . , dn ∈ Z, and fix such an X0 and y. We may divide each di for 2 ≤ i ≤ n by d to
obtain
di = qi d + ri
with qi ∈ Z and 0 ≤ ri < n. Then
y = d(x0 + q2 x2 + · · · + qn xn ) + r2 x2 + · · · + rn xn .
Let x1 = x0 + q2 x2 + · · · + qn xn . Then X = {x1 , x2 , . . . , xn } is a basis by Lemma 4.5.1. The mini-
mality of d now forces r2 = · · · = rn = 0. In other words, we have y = dx1 ∈ H.
Let G0 = hx2 , . . . , xn i, which is free abelian of rank n − 1, and define a homomorphism
ϕ : G → Z × G0 , ϕ(a) = (c1 , a − c1 x1 ),
for a ∈ G, where c1 ∈ Z is such that
n
a = ∑ ci xi
i=1
for some c2 , . . . , cn ∈ Z. We have that ϕ(a) = 0 if and only if c1 = 0 and a−c1 x1 = 0 by definition,
which occurs exactly when a = 0 as X is a basis. Therefore, ϕ is injective. Moreover, for a0 ∈ G0
and c ∈ Z, we clearly have that
ϕ(cx1 + a0 ) = (c, a0 ),
so ϕ is surjective. Therefore, ϕ is an isomorphism.
Finally, we compute ϕ(H). Suppose a ∈ H is written as above, and let c1 = qd + r with q ∈ Z
and 0 ≤ r < d. Then b = a − qdx1 ∈ H, and
n
b = rx1 + ∑ ci xi .
i=2
By the minimality of d, we must have r = 0. In other words, we have b ∈ H 0 and
a = q(dx1 ) + b,
so ϕ(a) ∈ dZ × H 0 . Conversely, if (m, b) ∈ dZ × H 0 , then
(m, b) = ϕ(mx1 + b),
 4.5. FINITELY GENERATED ABELIAN GROUPS 115

and mx1 ∈ H since d divides m, so mx1 + b ∈ H. Therefore, ϕ(H) = dZ × H 0 , as desired. 

We also note the following easy corollary of Theorem 2.5.17, obtained by applying it recur-
sively.
C OROLLARY 4.5.3. Let m be a positive integer, and for some k ≥ 0, write
m = pr11 pr22 · · · prkk
for distinct prime numbers p1 , p2 , . . . , pk and r1 , r2 , . . . , rk ≥ 2. Then
Z/mZ ∼ r r
= Z/p 1 Z × Z/p 2 Z × · · · × Z/p k Z.
1 2 k
r

We can now classify the finitely generated abelian groups up to isomorphism.

T HEOREM 4.5.4 (Structure theorem for finitely generated abelian groups). Let G be a finitely
generated abelian group. Then there exist k, r ≥ 0 and positive integers d1 , d2 , . . . , dk ≥ 2 such
that there is an isomorphism
G∼= Zr × (Z/d1 Z) × · · · × (Z/dk Z).
In fact, the integers di may be chosen so that di+1 divides di for each 1 ≤ i ≤ k − 1, and then
these are the unique r, k, and d1 , d2 , . . . , dk with those properties. Alternatively, we may choose
the isomorphism so that each di is a power of a prime number, in which case the decomposition
is again unique up to reordering.
P ROOF. We prove the result by induction on the number of elements n in a finite generat-
ing set of G, where we may consider the trivial group to be generated by the empty set. The
case n = 0 is then just the case that r = k = 0, and we have the result. Suppose we know the
result for all abelian groups that can be generated by n elements. Let G be an abelian group for
which X = {x1 , x2 , . . . , xn+1 } is a minimal set of generators. Then there exists a unique surjective
homomorphism ψ : Zn+1 → G such that ψ(ei ) = xi for 1 ≤ i ≤ n + 1. Let H = ker ψ 6 Zn+1 .
By Lemma 4.5.2, we have an isomorphism
ϕ : Zn+1 → Z × Zn
such that ϕ(H) = dZ × H 0 for some H 0 6 Zn and d ≥ 1. By the first isomorphism theorem, we
have
Z × Zn ∼
G∼= Zn+1 /H ∼ = = Z/dZ × Zn /H 0 .
dZ × H 0
Now, since Zn /H 0 can be generated by n elements, it may be written by induction as
Zn /H 0 ∼
= Zr × (Z/d1 Z) × · · · × (Z/dk−1 Z)
for some d1 , d2 , . . . , dk−1 ≥ 2 with r +k −1 ≤ n. Letting dk = d if d ≥ 2 and noting that G ∼
= Zn /H 0
if d = 1, we therefore have the first statement of the theorem.
By Corollary 4.5.3, we may can decompose each Z/di Z into a finite direct product of groups
of the form Z/pk Z with p prime and k ≥ 2, proving the last decomposition. On the other hand,
suppose we have decomposed G up to isomorphism as
G∼= Zr × P1 × P2 · · · × Pt ,
116 4. ADVANCED GROUP THEORY

where p1 , p2 , . . . , pt are distinct prime numbers and each Pi for 1 ≤ i ≤ t is a finite abelian pi -
group, which in turn we have written as
misi
Pi ∼ m m
= Z/pi i1 Z × Z/pi i2 Z × · · · Z/pi Z
for some si ≥ 1 and mi1 ≥ mi2 ≥ · · · ≥ misi ≥ 1. Let k = max{si | 1 ≤ i ≤ t}, set mi j = 0 if j > si ,
and let
t
mi j
d j = ∏ pi
i=1
for each 1 ≤ j ≤ k. Then d1 , d2 , . . . , dk ≥ 2, and di+1 divides di for each 1 ≤ i ≤ k − 1, as desired.
Moreover, Theorem 2.5.17 implies that
m1 j m2 j mt j
Z/d j Z ∼
= (Z/p1 Z) × (Z/p2 Z) × · · · × (Z/pt Z),
which yields the desired decomposition of G by gathering terms and applying these isomor-
phisms.
Finally, we address uniqueness of the latter two decompositions. First, we claim that in any
decomposition of G (without restriction on the di ), we must have the same r. For this, let
Gtor = {a ∈ G | na = 0 for some n ≥ 1}.
If we have written
G = Zr × (Z/d1 Z) × (Z/d2 Z) × · · · × (Z/dk Z)
for some r, k ≥ 0 and d1 , d2 , . . . , dk ≥ 2, then
Gtor = {0} × (Z/d1 Z) × (Z/d2 Z) × · · · × (Z/dk Z),
and G/Gtor ∼ = Zr . But Zr ∼
6 Zs for s 6= r, so the r in the decomposition must be unique. Moreover,
=
if Gtor = {0}, which is to say that |Gtor | = 1, then uniqueness of the decomposition is simply that
r is unique such that G ∼ = Zr , which we have just proven.
Now, suppose k ≥ 1 and we have chosen the di either to be prime powers, which we list in
descending order, or such that di+1 divides di for all 1 ≤ i ≤ k − 1 (so also in descending order).
In the former case, d1 is the largest order of any element of prime power order in G, and in the
latter, d1 is the exponent of G. Therefore, if we have a second decomposition,
G = Zr × (Z/d10 Z) × (Z/d20 Z) × · · · (Z/dk0 0 Z),
written in the same form as the first, then we must have d1 = d10 . Take the quotient, therefore, by
the subgroup Zr × Z/d1 Z × {0} × {0} in each decomposition. Then we have
Z/d2 Z × · · · × Z/dk Z ∼
= Z/d 0 Z × · · · × Z/d 0 0 Z.
2 k
By induction on the order of Gtor , assuming that we have proven the uniqueness for all G0 with
|G0tor | < |Gtor |, we must have k = k0 and di = di0 for all 2 ≤ i ≤ k, proving uniqueness. 
In the course of the proof of , we considered a subgroup of an abelian group G that we denote
Gtor . We give it a formal definition.
D EFINITION 4.5.5. Let G be an abelian group.
 4.6. GROUP ACTIONS ON SETS 117

a. The torsion subgroup of G is the subgroup

Gtor = {a ∈ G | na = 0 for some n ≥ 1}
of G.
b. An element of Gtor is called a torsion element of G.
R EMARK 4.5.6. If G is a finite abelian group, then G = Gtor .
We leave the proof of the following direct corollary of Theorem 4.5.4 to the reader.
C OROLLARY 4.5.7. If G is a finitely generated abelian group, then G ∼ = Zr × Gtor for some
r ≥ 0. Moreover, Gtor is a finite abelian group, and it is isomorphic to a direct product of cyclic
groups.
4.6. Group actions on sets
D EFINITION 4.6.1. An action of a group G on a set X is an operation
?: G×X → X
satisfying the following properties
i. e ? x = x for all x ∈ X,
ii. a ? (b ? x) = (ab) ? x for all a, b ∈ G and x ∈ X.
We then say that G acts on X and that the operation ? is a G-action, and we refer to X as a G-set.
R EMARK 4.6.2. As with binary operations, we typically denote “?” more simply by “·”.
E XAMPLES 4.6.3.
a. The symmetric group SX acts on X. In particular, Sn acts on Xn = {1, 2, . . . , n}.
b. The group of isometries of Rn acts on Rn .
c. The wallpaper group of a tiling of the plane acts on R2 .
d. For n ≥ 3, the group Dn acts on the set of vertices of the regular n-gon of which it is the
symmetry group, as well as the set of its edges.
e. The group GLn (R) acts on Rn by left multiplication of column vectors.
Here are a couple of more abstract examples.
E XAMPLES 4.6.4.
a. A group G acts on itself by left multiplication: a ? x = ax for a, x ∈ G.
b. A group G acts on itself by conjugation: a ? x = axa−1 for a, x ∈ G.
R EMARK 4.6.5. A group G does not act on itself by right multiplication. If we defined
a ? x = xa, then
a ? (b ? x) = a ? (xb) = (xb)a = x(ba),
while (ab) ? x = x(ab). The action by right multiplication is an example of what is known as a
right action (as opposed to a usual, or left, action).
118 4. ADVANCED GROUP THEORY

D EFINITION 4.6.6. We say that an action of a group G on a set X is transitive if for every
x, y ∈ X, there exists a ∈ G with ax = y. We then say that G acts transitively on X.
E XAMPLES 4.6.7.
a. The group SX acts transitively on X.
b. The group Dn acts transitively on the set of vertices of a regular n-gon, as well as the set
of edges.
c. The group of isometries of Rn acts transitively on Rn .
d. The group GLn (R) does not act transitively on Rn , as an invertible matrix times a nonzero
vector is always nonzero.
e. Any group G acts transitively on itself by left multiplication. This is simply the cancella-
tion theorem: if x, y ∈ G, then a = yx−1 satisfies ax = y.
f. The action of G on itself by conjugation is not transitive if G is nontrivial. For example,
the identity element is not a conjugate of any other element.
g. The group SX acts on the power set of X,
σ ?Y = {σ (y) | y ∈ Y }
for Y ⊆ X, but this action is not transitive if X is nonempty. For instance, σ ? Y always has the
same cardinality as Y , so it cannot be the empty set if Y is nonempty.
h. The group G acts on the set G/H of left cosets a subgroup H of G by left multiplication:
a · bH = abH,
and this is a transitive action.
D EFINITION 4.6.8. Let G be a group and X be a G-set. The orbit of x ∈ X is the set
G · x = {gx | g ∈ G}.
R EMARK 4.6.9. Recall that the orbit of x ∈ Xn under σ ∈ Sn was defined as
Oσ (x) = {σ i x | i ∈ Z},
and we can reinterpret this orbit as the orbit hσ i · x.
The following lemma is nearly immediate.
L EMMA 4.6.10. An action of a group G on a set X is transitive if and only if G · x = X for
every (equivalently, some) x ∈ X.
E XAMPLES 4.6.11.
a. Since SX acts transitively on X, we have that SX · x = X for every x ∈ X.
b. Consider the action of Sn on the power set of Xn . We have
Sn · {1, 2, . . . , k} = {Y ⊂ Xn | |Y | = k}.
c. The orbit of v ∈ Rn under GLn (R) is Rn − {0} if v 6= 0 and {0} if v = 0.
 4.6. GROUP ACTIONS ON SETS 119

d. The orbit of x ∈ G under the action of G on itself by conjugation is the conjugacy class Cx
of x.
e. The orbit of H 6 G under the action of G of its set of subgroups by conjugation is the set
of all conjugate subgroups to G:
{aHa−1 | a ∈ G}.
We remark that the property of being in the same orbit is an equivalence relation on a G-set
X, and therefore we obtain a partition of X as a disjoint union of its orbits.
P ROPOSITION 4.6.12. Let X be a G-set. The relation x ∼G y if and only if G · x = G · y for
x, y ∈ X is an equivalence relation on G, and the equivalence class of x ∈ X under ∼G is the orbit
Gx of x. Therefore, G is the disjoint union of its distinct orbits.
P ROOF. That ∼G is an equivalence relation is checked immediately. We remark that for
x, y ∈ X, we have G · x = G · y if and only if y ∈ G · x, since ax = by for some a, b ∈ G if and only
if cx = y for some c ∈ G (that c being b−1 a). Therefore, the equivalence class of x ∈ X is exactly
the orbit G · x, and the final statement is just Lemma 1.2.10. 
D EFINITION 4.6.13. Let X be a G-set for some group G. Let a ∈ G and x ∈ X. We say that a
fixes x if ax = x.
D EFINITION 4.6.14. We say that an action of a group G on a set X is faithful if the only
element a ∈ G that fixes all x ∈ X is the identity element. We then say that G acts faithfully on X,
and X is a faithful G-set.
In other words, G acts faithfully on X if ax = x for all x ∈ X implies a = e.
E XAMPLES 4.6.15.
a. The group SX acts faithfully on X, since a nontrivial permutation of X does not fix every
element of X.
b. For n ≥ 3, the group Dn acts faithfully on the set of vertices of the regular polygon, as well
as the set of edges.
c. The group G acts on itself faithfully by left multiplication, since if ax = x for any x ∈ G,
then a = e.
d. The action of a group G on itself by conjugation is faithful if and only if the group has
trivial center. To see this, note that
Z(G) = {a ∈ G | axa−1 = x for all x ∈ G}.
D EFINITION 4.6.16. Let G be a group and X be a G-set. The stabilizer, or isotropy subgroup,
of Gx of an element x ∈ X is the set of elements of G that fix x. That is, we have
Gx = {a ∈ G | ax = x}.
R EMARK 4.6.17. The stabilizer Gx is indeed a subgroup of G, since e ∈ Gx , and for a, b ∈ Gx ,
we have (ab)x = a(bx) = ax = x, so ab ∈ Gx , while
a−1 x = a−1 (ax) = ex = x,
120 4. ADVANCED GROUP THEORY

so a−1 ∈ Gx .
L EMMA 4.6.18. A group G acts faithfully on a set X if and only if
\
Gx = {e}.
x∈X
P ROOF. We have a ∈ Gx if and only if ax = x. Thus a ∈ Gx for all x ∈ X if and only if ax = x
for all x ∈ X, andTthe action of G on X is not faithful if and only if the latter occurs for some a ∈ G
with a 6= e. So, x∈X Gx contains a non-identity element if and only if G acts non-faithfully on
X. 
E XAMPLES 4.6.19. We give some examples of stabilizers.
a. The stabilizer of n under the action of Sn is the image of Sn−1 under the homomorphism
ι : Sn−1 → Sn of Example 2.10.20.
b. The stabilizer of a vertex under the action of Dn on a regular n-gon consists exactly of
the subgroup of order 2 generated by the unique reflection in Dn for which the line of reflection
passes through the vertex.
c. The stabilizer of 4 under the action of hσ i 6 S5 , where σ = (1 2 3)(4 5), is hσ 2 i.
d. The stabilizer of x ∈ G under the action of G on itself by left multiplication is trivial:
Gx = {a ∈ G | ax = x} = {e}
The following definition gives an interesting class of examples of stabilizers.
D EFINITION 4.6.20. The stabilizer of x ∈ G under the action of G on itself by conjugation is
the subgroup of elements in G that commute with x and is known as the centralizer Zx of x
Zx = {a ∈ G | ax = xa}.
E XAMPLE 4.6.21. The centralizer Z(1 2 3) in S5 is
Z(1 2 3) = h(1 2 3), (4 5)i.
We end by comparing orbits and stabilizers.
T HEOREM 4.6.22. Let X be a G-set, and let x ∈ X. Then there is a bijection
ψx : G/Gx → G · x,
given by ψx (aGx ) = ax for any a ∈ G.
P ROOF. First, we note that ψx is well-defined, since if b ∈ aGx , then b = ag for some g ∈ Gx ,
and
ψx (bGx ) = bx = agx = a(gx) = ax = ψx (aGx ).
Moreover, it is one-to-one since, if ax = bx, then x = a−1 bx, so a−1 b ∈ Gx , and therefore aGx =
bGx . Finally, it is onto by definition. 
C OROLLARY 4.6.23. If G is a finite group, then every element x ∈ X has a finite orbit, and
|G · x| = [G : Gx ].
In particular, the number of elements in the orbit of x divides |G|.
 4.7. PERMUTATION REPRESENTATIONS 121

E XAMPLE 4.6.24. The centralizer of (1 2 3) in S5 has order 6, while the orbit of (1 2 3) is

the set of 3-cycles in S5 , of which there are 20, and we note that |S5 | = 20 · 6.
We find an application in the class equation.
P ROPOSITION 4.6.25 (The class equation). Let G be a finite group. Then
|G| = |Z(G)| + ∑ [G : Zx ],
x∈X
where X is set of representatives of the conjugacy classes in G with more than one element.
P ROOF. By Corollary 4.6.23, we have that [G : Zx ] = |Cx | for x ∈ X. Moreover, |Cx | = 1 if
x ∈ Z(G). The equality we wish to prove is therefore reduced to the known fact that |G| is the
sum of the orders of its distinct conjugacy classes. 
Here is one application.
P ROPOSITION 4.6.26. The group A5 is simple.
P ROOF. Any normal subgroup of a group is a disjoint union of conjugacy classes in that
group including the conjugacy class {e}. Let us determine the conjugacy classes in A5 . The con-
jugacy classes in S5 of nontrivial elements in A5 are the products of 2 transpositions, the 3-cycles,
and the 5-cycles. The centralizer of (1 2 3) in A5 is the group h(1 2 3)i, so |C(1 2 3) | = 603 = 20,
and C(1 2 3) is thus the set of 3-cycles. The centralizer of (1 2)(3 4) is h(1 2)(3 4), (1 3)(2 4)i, so
|C(1 2)(3 4) | = 15, and C(1 2)(3 4) is thus the set of products of two transpositions. The centralizer
of (1 2 3 4 5) is h(1 2 3 4 5)i, so |C(1 2 3 4 5) | = 12 and so there are two conjugacy classes of 5
cycles, each with 12 elements. We then have that the distinct conjugacy classes of elements in
A5 have 1, 12, 12, 15, and 20 elements. Aside from 1 and 60, no sum of these numbers including
1 divides 60, so A5 has no nontrivial, improper normal subgroups. 
4.7. Permutation representations
In this brief section, we give a characterization of group actions on sets as permutation rep-
resentations.
T HEOREM 4.7.1. Let X be a G-set, and define σa : X → X by σa (x) = ax for any a ∈ G. Then
σa ∈ SX , and the map
ρ : G → SX
such that ρ(a) = σa for all a ∈ G is a homomorphism. Conversely, if ρ : G → SX is a homomor-
phism, then the operation defined by
a ? x = ρ(a)(x)
on a ∈ G and x ∈ G is a G-action.
P ROOF. We have
(σa−1 ◦ σa )(x) = x = (σa ◦ σa−1 )(x),
so σa is a bijection, which is to say σa ∈ SX . For a, b ∈ G and x ∈ X, we have
ρ(ab)(x) = σab (x) = (ab)x = a(bx) = σa (σb (x)) = (σa ◦ σb )(x) = (ρ(a) ◦ ρ(b))(x),
122 4. ADVANCED GROUP THEORY

so ρ is a homomorphism.
Conversely, given ρ, we check that e ? x = ρ(e)(x) = idX (x) = x, while
a ? (b ? x) = ρ(a)(ρ(b)(x)) = ρ(ab)(x) = (ab) ? x,
so ? is a G-action. 
D EFINITION 4.7.2. If X is a G-set, then the homomorphism ρX associated to X by Theo-
rem 4.7.1 is called its permutation representation.
E XAMPLE 4.7.3. The action of SX on X gives rise to a permutation representation ρ : SX →
SX satisfying ρ(σ )(x) = σ (x) for all x ∈ X. In other words, we have ρ = idSX .
E XAMPLE 4.7.4. Consider the action of GLn (R) on Rn . The permutation representation
ρ : GLn (R) → SRn
takes A ∈ GLn (R) to a map TA : Rn→ Rn that satisfies TA (v) = Av. In other words, the image of
ρ is the subgroup of SRn consisting of invertible linear transformations.
The following lemma is almost immediate.
L EMMA 4.7.5. A group G acts faithfully on a set X if and only if ρX is injective. In fact, the
kernel of ρX is the intersection of the stabilizers Gx over all x ∈ X.
P ROOF. We have ρX (a) = idX if and only if ax = ρX (a)(x) = x for all x ∈ X. 
E XAMPLE 4.7.6. The permutation representation attached to G acting on itself by conjuga-
tion is a homomorphism γ : G → SG given by a 7→ γa , and its image is the inner automorphism
group, a subgroup of Aut(G) 6 SG . The kernel of γ is Z(G).
We now prove Cayley’s theorem, which tells us that every group is a subgroup of a symmetric
group.
T HEOREM 4.7.7 (Cayley). Every group G is isomorphic to a subgroup of SG .
P ROOF. Consider the permutation representation ρG : G → SG associated to the action of G
on itself by left multiplication. By Example 4.6.4, ρG is injective, and therefore, G is isomorphic
to im ρG 6 SG . 
R EMARK 4.7.8. Note that if G has order n, Cayley’s theorem tells us that G is isomorphic
to a subgroup of Sn . However, this is not always the smallest permutation group in which it is
contained. For example, the action of Dn on its vertices is faithful, and so a choice of numbering
of these vertices identifies Dn with a subgroup of Sn , as opposed to S2n . Even more simply, Sn is
obviously a subgroup of itself, and not just isomorphic to a subgroup of Sn! .

4.8. Burnside’s formula

N OTATION 4.8.1. Let X be a G-set, and let S be a subset of G. Then we set
X S = {x ∈ X | ax = x for all a ∈ S}.
If S = {a} for some a ∈ X, we sometimes write X a for X {a} .
 4.8. BURNSIDE’S FORMULA 123

R EMARK 4.8.2. For x ∈ X and a ∈ G, where X is a G-set, the statement that ax = x is

equivalent both to a ∈ Gx and to x ∈ X a .
E XAMPLES 4.8.3.
a. Take σ = (1 3)(2 5) ∈ S6 , and let X = X6 . We have X σ = {4, 6}.
b. Let X be the Dn -set that is the set of vertices of the regular n-gon inscribed on the unit
circle in R2 with a vertex at (1, 0). Then X s = {(1, 0)} if n is odd and X s = {(−1, 0), (0, 1)} if n
is even, while X r = ∅.
c. Let X = Rn , and let A ∈ GLn (R). Then X A = {v ∈ Rn | Av = v} is the eigenspace of A with
eigenvalue 1 (so {0} if 1 is not an eigenvalue). We have X GLn (R) = {0}.
We now state Burnside’s formula.
T HEOREM 4.8.4 (Burnside). Let G be a finite group, and let X be a finite G-set. Let r be the
number of distinct orbits in X under G. Then
1
r= ∑ |X a |.
|G| a∈G
P ROOF. We will count the set of pairs
S = {(a, x) | a ∈ G, x ∈ X, ax = x} ⊆ G × X
in two different ways. First, note that
a
S= {(a, x) | x ∈ X a },
a∈G
so we have
|S| = ∑ |X a|.
a∈G
On the other hand, note that a
S= {(a, x) | a ∈ Gx },
x∈X
so we have
|G|
|S| = ∑ |Gx | = ∑ |G · x| ,
x∈X x∈X
the latter statement being Corollary 4.6.23 and Lagrange’s theorem. If O is an orbit in X, then it
is the orbit of all x ∈ O, so we have
|G| 1
∑ |G · x| = |G| ∑ |O| = |G|.
x∈O x∈O
Since X is the disjoint union of its orbits O and there are r orbits, we obtain |S| = r|G|, and
therefore we have
r|G| = ∑ |X a |,
a∈G
as desired. 
124 4. ADVANCED GROUP THEORY

Burnside’s formula has an amusing use in certain problems involving counting.

E XAMPLE 4.8.5. Suppose we have a table with which is a regular octagon, with one chair
placed at each side. Let us consider two seatings of eight people at the table to be the equivalent
if and only if every person has the same two neighbors under both seatings (though possibly on
different sides). We can ask: how many equivalence classes of seatings are there?
Let X denote the set of all seatings, so |X| = 8!. Two seatings are equivalent if and only if
there is an element of D8 that takes the positions of the people under one seating to the their
positions under the other. In other words, the seatings are in one-to-one correspondence with
the orbits under the action of D8 on X. We note that X e = X, and X a = ∅ if a 6= e since any
nontrivial element of D8 will change the position of at least one person (in fact, at least six
people). Applying Burnside’s formula, we have that the number of equivalence classes r of
seatings is 8!/16 = 2520.
E XAMPLE 4.8.6. How many different ways are there to color the faces of a cube either red
or blue that actually look different? (Here: two colorings are the same if one is a rotation of
another.)
Let X denote the set of all colorings, so |X| = 26 = 64. The group of rotations (orientation-
preserving isometries) of a die has 5 types of elements: the identity, 6 rotations of order 4 through
the centers of opposite faces, 3 rotations of order 2 of the same form, 6 rotations of order 2
through the centers of opposite edges, and 8 rotations of order 3 through the centers of opposite
vertices. Respectively, these elements a have |X a | = 64, 8, 16, 8, and 4. We then have
1 240
r= (64 + 6 · 8 + 3 · 16 + 6 · 8 + 8 · 4) = = 10.
24 24
This means there are exactly 10 different-looking colorings. Note that, as is often the case with
these sorts of problems, it would have been easier to simply count them directly.

4.9. p-groups
D EFINITION 4.9.1. A group G is said to be a p-group if every element of G is finite of order
a power of p.
Note that we have already classified the finite abelian p-groups up to isomorphism.
E XAMPLE 4.9.2. The group D4 is a nonabelian 2-group of order 8.
All finite groups of p-power order are clearly p-groups. We shall see that the converse is true
as well. For this, we require the following useful lemma.
L EMMA 4.9.3. Let G be a finite group of p-power order, and let X be a G-set. Then
|X| ≡ |X G | mod p.
P ROOF. By Corollary 4.6.23, every orbit in X has order dividing |G|, hence a power of p.
Note that the orbits of order 1 are exactly the {x} with x ∈ X G . On the other hand, the other orbits
 4.9. p-GROUPS 125

all have order divisible by p, so if Y is a set of representatives of the orbits of G, then it contains
X G , and we have
|X| = ∑ |G · y| ≡ ∑ |{x}| mod p,
y∈Y x∈X G

In general, if the order of a finite group G is n, then while we know that every element of
G has order dividing n, we do not have the converse (unless G is cyclic). On the other hand,
Cauchy’s theorem, which we now prove, tells us that G contains elements of every prime order
dividing n.
T HEOREM 4.9.4 (Cauchy). Let p be a prime number, and let G be a finite group of order
divisible by p. Then G contains an element of order p.
P ROOF. We consider the set
X = {(a1 , a2 , . . . , a p ) ∈ G p | a1 a2 · · · a p = e}.
Note that if (a1 , a2 , . . . , a p ) ∈ X, then a1 , a2 , . . . , a p−1 ∈ G can be chosen arbitrarily, and then a p =
(a1 a2 . . . a p−1 )−1 is determined by those ai . It follows that |X| = |G| p−1 . Let τ = (1 2 . . . p) ∈ S p .
We let τ, and hence hτi, act on X by
τ · (a1 , a2 , . . . , a p ) = (a2 , . . . , a p , a1 ).
Note that this is an action, as
(a2 · · · a p )a1 = e,
since a1 being left inverse to a2 · · · a p implies that it is also right inverse to a2 · · · a p . Then
X hτi = X τ = {(a, a, . . . , a) ∈ G p | a p = e}.
By Lemma 4.9.3, we have
|X| ≡ |X hτi | mod p.
Since p divides |G|, it divides |X|, and hence it divides the order of X hτi . But X hτi is in bijection
with the set of elements of G of order dividing p, and e is such an element. So, we must have at
least p distinct elements in G of order dividing p, hence at least p − 1 of order p. 
Cauchy’s theorem has the above-mentioned corollary.
C OROLLARY 4.9.5. Every finite p-group has p-power order.
P ROOF. If G is a finite group and ` is a prime dividing |G|, then G has an element of order `
by Cauchy’s theorem. So if G is a p-group, then by definition the only prime that can divide |G|
is p. 
The following result is very useful in the study of p-groups.
P ROPOSITION 4.9.6. The center Z(G) of a nontrivial finite p-group is nontrivial.
126 4. ADVANCED GROUP THEORY

P ROOF. Consider the action of G on itself by conjugation. The set of elements of G fixed by
every element of G under conjugation is exactly the center of G. By Lemma 4.9.3, we therefore
have that |G| ≡ |Z(G)| mod p. Since |G| is a nontrivial power of p, this means that Z(G) is not
the trivial subgroup. 
We give an application of Proposition 4.9.6 to the study of the structure of p-groups of order
p2 .
T HEOREM 4.9.7. Every group of order p2 , where p is a prime, is abelian.
P ROOF. Let G be a group of order p2 . By Proposition 4.9.6, we have that Z(G) is nontrivial,
so has either order p or p2 . We must show that it is the latter, since Z(G) = G if and only if G is
abelian. So, suppose by way of contradiction that |Z(G)| = p, and let b ∈ G be an element that
is not in the center of G. Then H = Z(G)hbi has order greater than p, hence is all of G. But H
is abelian, since b commutes with every element of Z(G) and certainly every element of Z(G)
commutes with itself. So, G = H is abelian as well, contradicting |Z(G)| = p. 
In particular, this tells us that there are only two isomorphism classes of groups of order p2 ,
those of Z/p2 Z and Z/pZ × Z/pZ.
4.10. The Sylow theorems
D EFINITION 4.10.1. Let G be a group, and let p be a prime number.
a. A subgroup of G is called a p-subgroup if it is a p-group.
b. A p-subgroup of G is called a Sylow p-subgroup of G if it is not properly contained in any
p-subgroup of G.
R EMARK 4.10.2. If G is a finite group and pn is the largest power of p dividing |G|, then
every p-subgroup of G has order dividing pn .
E XAMPLES 4.10.3.
a. In S5 , the Sylow 5-subgroups are the subgroups generated by the 5-cycles, the Sylow
3-subgroups are the subgroups generated by the 3-cycles, and the Sylow 2-subgroups are the
subgroups of order 8 the form
h(a b c d), (a c)i
with a, b, c, d distinct elements of X5 .
b. If n ≥ 3 and n = 2k m with m odd, then every Sylow 2-subgroup of Dn has the form Dn =
hri s, rm i for some 0 ≤ i < m.
The Sylow theorems, which we now state in one compact result that we refer to as Sylow’s
theorem, constitute an extremely useful tool for the study of finite groups.
T HEOREM 4.10.4 (Sylow). Let G be a finite group, let p be a prime number, and let n be
exponent of the highest power of p dividing |G|. Then the following hold.
a. Every Sylow p-subgroup of G has order pn .
b. Every two Sylow p-subgroups of G are conjugate.
 4.10. THE SYLOW THEOREMS 127

c. The number of Sylow p-subgroups divides |G| and is congruent to 1 modulo p.

We defer the proof of the Sylow’s theorem to below.
N OTATION 4.10.5. Let G be a finite group and p a prime number. We let Syl p (G) denote the
set of Sylow p-subgroups of G, and we let n p (G) = | Syl p (G)|.
The fact that n p (G) both divides |G| and is congruent to 1 modulo P can be very useful in
determining the possible isomorphism classes of groups of a given order.
E XAMPLE 4.10.6. It is easy to see from our description of the Sylow p-subgroups of S5
that every two Sylow p-sugroups of S5 are conjugate, as the elements generating such groups
are conjugate. We have n5 (S5 ) = 6 ≡ 1 mod 5. We also have n3 (S5 ) = 10 ≡ 1 mod 3, and
n2 (S5 ) = 15, which is odd.
To understand the second part of Sylow’s theorem, we introduce the concept of a normalizer.
D EFINITION 4.10.7. Let G be a group and H be a subgroup. The normalizer NG (H) of H in
G is the subgroup
NG (H) = {a ∈ G | aHa−1 = H}
of G.
R EMARKS 4.10.8.
a. By definition, NG (H) is the stabilizer of H under the action of G on its set of subgroups
by conjugation, so in particular is a subgroup.
b. We have H P NG (H) and NG (H) is the largest subgroup of G in which H is normal.
c. We have NG (H) = G if and only if H P G.
E XAMPLES 4.10.9.
a. We have NDn (hri) = Dn and NDn (hsi) = hsi if n is odd and hs, rn/2 i if n is even.
b. The normalizer of h(1 2 3 4)i in S5 is h(1 2 3 4), (1 3)i.
The following lemma is crucial for proving the first part of Sylow’s theorem.
L EMMA 4.10.10. Suppose that G is a finite group, and let H be a subgroup of G of order a
power of a prime p. Then
[G : H] ≡ [NG (H) : H] mod p.
P ROOF. Let L = G/H. Then H acts on L by left multiplication: h · (aH) = haH for h ∈ H,
a ∈ G. We have that aH ∈ L H if and only if h(aH) = aH for every h ∈ H, which is to say that
aha−1 ∈ H for every h ∈ H, which means exactly that aHa−1 = H. In other words,
L H = {aH | a ∈ NG (H)} = NG (H)/H.
By Lemma 4.9.3, we have that
|L | ≡ |L H | mod p,
which is exactly the statement of the lemma. 
128 4. ADVANCED GROUP THEORY

Let us fix a prime p throughout the rest of this section. We prove a strengthening of the first
part of Sylow’s theorem.
T HEOREM 4.10.11 (First Sylow theorem). Let G be a group, and let n be the exponent of the
highest power of p dividing G. Every subgroup of G of order pk with k < n is a normal subgroup
of a subgroup of G of order pk+1 .
P ROOF. Suppose that H < G has order pk . By Lemma 4.10.10, its index in its normalizer is
congruent to [G : H] modulo p, so is divisible by p. But then NG (H)/H has order divisible by p,
and so by Cayley’s theorem there exists a subgroup of it of order p. By Proposition 2.13.10, there
then exists a subgroup K of NG (H) in which H is normal and such that |K/H| = p. Lagrange’s
theorem then implies that |K| = pk+1 , as desired. 
Theorem 4.10.11 tells us, in particular, that every p-subgroup of G of order less than pn is
not maximal, so part a of Sylow’s theorem holds. In fact, recursion tells us that:
C OROLLARY 4.10.12. Every p-subgroup of a finite group G is contained in a Sylow p-
subgroup of order pn , where n is the exponent of the highest power of p dividing G. In particular,
every Sylow p-subgroup has order pn .
The first Sylow theorem also has the following simple corollary.
C OROLLARY 4.10.13. Let G be a group, and let n be the exponent of the highest power of p
dividing G. The G has subgroups of order pk for every 1 ≤ k ≤ n.
P ROOF. Suppose without loss of generality that p divides |G|. By Cauchy’s theorem, G has
an element of order p, so it has a subgroup of order p. By recursion, the first Sylow theorem then
tells us that G has subgroups of every p-power order dividing G. 
We next prove the second part of Sylow’s theorem, which we state as a separate result.
T HEOREM 4.10.14 (Second Sylow theorem). If P and Q are Sylow p-subgroups of a finite
group G for some prime p, then P and Q are conjugate subgroups of G.
P ROOF. We consider the action of Q on the set of left cosets G/P via h · aP = (ha)P for h ∈ Q
and a ∈ G. By Lemma 4.9.3, we have that |(G/P)Q | ≡ |G/P| mod p. Since p does not divide
|G/P|, we therefore have that p does not divide |(G/P)Q |. In particular, there exists an element
bP ∈ (G/P)Q . Since hbP = bP for all h ∈ Q, we have b−1 hb ∈ P for all h ∈ Q, so b−1 Qb 6 P.
Since P and Q have the same order, we therefore have that Q = bPb−1 is a conjugate of P in
G. 
C OROLLARY 4.10.15. Suppose that G is a finite group, and let P be a Sylow p-subgroup of
G. Then P is normal in G if and only if n p (G) = 1.
P ROOF. We know that every conjugate of a Sylow p-subgroup is a Sylow p-subgroup, as it
has the same order, so it is an immediate corollary of Theorem 4.10.14 that P is normal in G if
and only if n p (G) = 1. 
The second Sylow theorem also has, after a short argument, the following consequence.
 4.11. APPLICATIONS OF SYLOW THEORY 129

P ROPOSITION 4.10.16. Let G be a finite group, let p be a prime number, and let P be a Sylow
p-subgroup. Then n p (G) = [G : NG (P)]. In particular, the number of Sylow p-subgroups of G
divides |G|.
P ROOF. Consider the action of G on the set Syl p (G) of Sylow p-subgroups of G by conjuga-
tion By Theorem 4.10.4b, we have that Syl p (G) has just one orbit under this action, which is all
of Syl p (G). Since the stabilizer of P is NG (P), the result follows from Corollary 4.6.23. 
We will require a special case of the following lemma.
L EMMA 4.10.17. Let Q be a Sylow p-subgroup and P be a p-subgroup of a finite group G.
Then P ∩ NG (Q) = P ∩ Q.
P ROOF. Let H = P ∩ NG (Q). We need only show that H 6 Q. Since H 6 NG (Q), we have
that hQh−1 = Q for all h ∈ H, so HQ = QH, and therefore HQ is a subgroup of G with Q as a
normal subgroup. The second isomorphism theorem implies that
|H||Q|
|HQ| = ,
|H ∩ Q|
and yields in particular that HQ is a p-subgroup of G. On the other hand, Q is a Sylow p-
subgroup of G, so HQ cannot be larger, and therefore must equal Q. Thus, we have the required
containment H 6 Q. 
Finally, we prove the third part of Sylow’s theorem.
T HEOREM 4.10.18 (Third Sylow theorem). The number n p (G) of Sylow p-subgroups of a
finite group G divides |G| and is congruent to 1 modulo p.
P ROOF. The first part is just Proposition 4.10.16 and Lagrange’s theorem. For the second
part, we assume that p divides |G|, as the result is otherwise trivial. Let P be a Sylow p-subgroup
of G, and let P act on Syl p (G) by conjugation: if Q ∈ Syl p (G) and a ∈ P, then a takes Q to
aQa−1 . By Lemma 4.9.3, we then have
n p (G) ≡ | Syl p (G)P | mod p.
Let Q ∈ Syl p (G)P , which tells us that P 6 NG (Q). By Lemma 4.10.17, we then have that
P = P ∩ NG (Q) 6 Q,
which forces P = Q as P and Q have the same order by Corollary 4.10.12. Thus, we have that
| Syl p (G)P | = 1, and so n p (G) ≡ 1 mod p. 

4.11. Applications of Sylow theory

We can use Sylow’s theorem to classify, or simply to give information on, the structure of
groups of a given order. For instance, Sylow’s theorem can be used to show that there are no
simple groups of certain small orders or of orders with certain sorts of prime factorizations, as
we see in the following examples.
130 4. ADVANCED GROUP THEORY

E XAMPLE 4.11.1. There are no simple groups of order 42. If G is a group of order 42 , then
n7 (G) divides 42 and is 1 modulo 7, which forces n7 (G) = 1. By Corollary 4.10.15, we have that
the unique subgroup of G of order 7 is normal, so G is not simple.
E XAMPLE 4.11.2. Let G be a group of order 30. Suppose that n3 (G) > 1 and n5 (G) > 1.
Then the third part of Sylow’s theorem tells us that n5 (G) = 6 and n3 (G) = 10. Now, any Sylow
5-subgroup has order 5, hence is cyclic with 4 elements of order 5, and any two distinct Sylow
5-subgroups have trivial intersection. Therefore, G contains 24 elements of order 5. On the other
hand, the same argument with 3 replacing 5 tells us that G contains 20 elements of order 3. This
is clearly impossible. In particular, there are no simple groups of order 30.
E XAMPLE 4.11.3. There are no simple groups of order pn , where p is a prime and n ≥ 2.
This follows from Corollary 4.10.13, which tells us that such a group has a subgroup of order
pn−1 and the first Sylow theorem, which tells us that the subgroup is normal in a subgroup of
order pn , which is necessarily the whole group.
We can also study groups with orders having a particularly nice form. The following result
is useful for that.
P ROPOSITION 4.11.4. Let G be a group. Suppose that H and K are normal subgroups of G
with HK = G and H ∩ K = {e}. Then the function ψ : H × K → G given by ψ(h, k) = hk for
h ∈ K and k ∈ K is an isomorphism.
P ROOF. Let h ∈ H and k ∈ K. Then [h, k] equals both (hk−1 h−1 )k−1 , from which it is seen to
be an element of K, as K C G, and h(kh−1 k−1 ), which is similarly seen to be an element of H. As
H ∩ K = {e}, we therefore have [h, k] = e, and therefore elements of H commute with elements
of K. It follows that ψ as defined is a homomorphism. It is onto as G = HK and one-to-one as
hk = e implies h, k ∈ H ∩ K, so (h, k) = (e, e). 
Proposition 4.11.4 has the following application in conjunction with Sylow’s theorems.
T HEOREM 4.11.5. Suppose that p and q are prime numbers with p < q. Then every group of
order pq has a normal subgroup of order q and is in fact cyclic if q 6≡ 1 mod p.
P ROOF. Note that nq (G) divides p and is 1 modulo q, which forces nq (G) = 1 since p < q.
By Corollary 4.10.15, G has a unique, normal Sylow q-subgroup K of order q. On the other hand,
n p (G) divides q and is 1 modulo p. Supposing that q 6≡ 1 mod p. then we must have n p (G) = 1
as q is a prime. Let H be the unique, normal subgroup of order p. Now, both H and K are cyclic,
so let h, k ∈ G with H = hhi and K = hki. By Proposition 4.11.4, we have that
G∼= H ×K ∼ = Z/pZ × Z/qZ ∼ = Z/pqZ,
the last step being the Chinese remainder theorem. 
We can use this to give a more complicated example of a proof that all groups of a given
order are cyclic.
E XAMPLE 4.11.6. All groups of order 255 are cyclic. Since 255 = 3 · 5 · 17, the structure
theorem for finitely generated abelian groups tells us that every abelian group of order 255 is
 4.11. APPLICATIONS OF SYLOW THEORY 131

cyclic. So, we must show that every group G of order 255 is abelian. By the third Sylow
theorem, n17 (G) = 1, so G has a unique subgroup N of order 17, which is normal. Then G/N has
order 15 and so is cyclic by Theorem 4.11.5. By Theorem 4.2.8, the subgroup N must contain
the commutator subgroup [G, G] of N.
Again by the third Sylow theorem, we have either n3 (G) = 1 or n3 (G) = 85 and n5 (G) = 1 or
n5 (G) = 51. If n3 (G) = 85, then G has at least 170 elements of order 3, and if n5 (G) = 51, then
G has at least 204 elements of order 5. Clearly, both of these cannot hold at the same time, so
either n3 (G) = 1 or n5 (G) = 1. But then G has either a normal subgroup of order 3 or a normal
subgroup of order 5. Call this subgroup Q. Then G/Q has order 3 · 17 or 5 · 17, and in either
case, Theorem 4.11.5 tells us that it is cyclic. As before, we then have that Q contains [G, G], but
Q ∩ N is trivial since Q and N have relatively prime order, so [G, G] = {e}, which is to say that
G is abelian.
Let us also expand our study of groups of order 30.
E XAMPLE 4.11.7. Every group G of order 30 has a normal subgroup of order 5. To see this,
let P be a subgroup of order 3 and Q a subgroup of order 5. By Example 4.11.2, either P or Q
is normal, and therefore PQ is a subgroup of G of order 15. By Theorem 4.11.5, it is cyclic. We
thus have that |NG (Q)| is either 15 or 30, so nG (Q) = 1 or 2, and 2 is impossible by the third
Sylow theorem. Thus, Q is normal.
We can also rule out a whole class of possible orders of simple groups with the following
result.
P ROPOSITION 4.11.8. There are no simple groups of order p2 q, where p and q are distinct
prime numbers.
P ROOF. Let G be a group of order p2 q. If p > q, then n p (G) = 1 by the third Sylow theorem,
so G has a normal Sylow p-subgroup (which is abelian of order p2 ). If q > p, then nq (G) = 1
or p2 . We need only check the latter case. In this case, p2 ≡ 1 mod q, so q divides p2 − 1, but it
does not divide p − 1 as q > p, so q divides p + 1, which forces p = 2 and q = 3. Then G has
order 12. Now, if n3 (G) = 4, a simple element count shows that one cannot have n2 (G) = 3, so
G has a normal Sylow 2-subgroup. 

Another method for exhibiting the non-simplicity of groups of a given order comes from the
use of permutation representations.
P ROPOSITION 4.11.9. Let G be a finite simple group of order properly divisible by p. Then
G is isomorphic to a subgroup of Sd where d = n p (G). In particular, |G| divides d!.
P ROOF. Let G act on the set L of left cosets of the normalizer NG (P) of some Sylow p-
subgroup P of G by left multiplication. This is a transitive action, so the permutation represen-
tation G → SL is nontrivial, hence injective as G is simple. As |L | = d by the third Sylow
theorem, we have SL ∼= Sd , so G is isomorphic to a subgroup of Sd . 

We provide a couple of examples.

132 4. ADVANCED GROUP THEORY

E XAMPLE 4.11.10. There are no simple groups G of order 160. That is, if G were such a
group, then n2 (G) = 5 by the third Sylow theorem, and therefore G is isomorphic to a subgroup
of S5 . But 120 does not divide 5!.
E XAMPLE 4.11.11. There are no simple groups of order 396 = 22 32 11. If G were such a
group, then n11 (G) = 12, and the normalizer of a Sylow 11-subgroup P in G has order 33 = 396 12
by the third Sylow theorem. It follows by Proposition 4.11.9 that G is isomorphic to and thus
may be identified with a subgroup of S12 . By definition NG (P) is contained in NS12 (P). But P is
generated by an 11-cycle, and the number of such Sylow 11-subgroups of S12 is easily counted
to be 12 · 9! (as there are 12!
11 such cycles and 10 per subgroup), which again by the third Sylow
theorem implies that the order NP (G) is 110, which is not a multiple of 33.
The following weakening of the second isomorphism theorem to allow arbitrary finite sub-
groups is a useful tool.
L EMMA 4.11.12. Let H and K be finite subgroups of a group G. Then we have
|H||K|
|HK| = .
|H ∩ K|
P ROOF. By definition, we have |HK| ≤ |H||K|. If h, h0 ∈ H and k, k0 ∈ K are such that hk =
h0 k0 ,
then setting a = (h0 )−1 h = k0 k−1 , we have that a ∈ H ∩ K. Moreover, note that h0 = ha−1
and k0 = ak. Conversely, given h ∈ H, k ∈ K, and a ∈ H ∩ K, then defining h0 = ha−1 and k0 = ak,
we see that hk = h0 k0 . Therefore, if we define an equivalence relation on the set H × K by
(h, k) ∼ (h0 , k0 ) if and only if hk = h0 k0 , the number of pairs in each equivalence class is |H ∩ K|,
and as a result there are |H||K|/|H ∩ K| equivalence classes. On the other hand, the number of
equivalence classes is by definition |HK|, proving the desired equality. 
Let us apply Lemma 4.11.12 to an example.
E XAMPLE 4.11.13. There are no simple groups of order 48. Suppose G is a group of order
48. By the third Sylow theorem, we have n2 (G) = 1 or 3. If n2 (G) = 1, then the unique Sylow
2-subgroup is normal. If n2 (G) = 3, let H and K be distinct subgroups of G of order 16. Then
the fact that |HK| ≤ 48 and Lemma 4.11.12 force |H ∩ K| = 8. Then H ∩ K has index 2 in H and
K, hence is normal in both, so its normalizer NG (H ∩ K) contains HK, which has order 32, so
equals G. But then H ∩ K is normal in G, so again G is not simple.
The latter example used a special case of the following, which tells us that a group G of order
48 with n2 (G) = 3 6≡ 1 mod 4 has two Sylow 2-subgroups with intersection of order 8.
P ROPOSITION 4.11.14. Let G be a finite group, and let n be the exponent of the highest power
of p dividing |G|. Let P be a Sylow p-subgroup of G. Let r ≤ n be a positive integer such that
|P ∩ Q| ≤ pn−r for every Sylow p-subgroup Q of G with Q 6= P. Then we have n p (G) ≡ 1 mod pr .
P ROOF. Let P be a Sylow p-subgroup, and consider the action of P on Syl p (G) by conju-
gation. Let Q ∈ Syl p (G) with Q 6= P. Lemma 4.10.17 tells us that the elements in P that fix Q
under conjugation (i.e., the a ∈ P such that aQa−1 = Q) are exactly those in P ∩ Q. For i such
that pi = [P : P ∩ Q], this implies that there are exactly pi conjugates of Q by elements of P, so
 4.12. SIMPLICITY OF ALTERNATING GROUPS 133

the order of the P-orbit of Q is pi . Under the assumption of the proposition, we have that |P ∩ Q|
divides pn−r , so i ≥ r. Therefore, P-orbit of Syl p (G) other than the singleton orbit {P} has order
divisible by pr , which implies that n p (G) = | Syl p (G)| ≡ 1 mod pr . 

4.12. Simplicity of alternating groups

Before we proceed to simplicity, we first show that we can use group actions to give an
alternate definition of the sign of a permutation (and therefore of alternating groups) that does
not use the determinant map, which we did not define above.
P ROPOSITION 4.12.1. Let Sn act on the set X of polynomials p = p(x1 , x2 , . . . , xn ) in n vari-
ables x1 , x2 , . . . , xn by
σ · p = p(xσ (1) , xσ (2) , . . . , xσ (n) )
for σ ∈ Sn and p ∈ X. Take
∆ = ∏ (x j − xi ) ∈ X
1≤i< j≤n
The function ε : Sn → {±1} given by
σ · ∆ = ε(σ )∆
for all σ ∈ Sn is equal to the homomorphism sign.
P ROOF. For σ , τ ∈ Sn and p ∈ X, we have
σ τ · p = p(xσ (τ(1)) , xσ (τ(2)) , . . . , xσ (τ(n)) ) = σ · p(xτ(1) , xτ(2) , . . . , xτ(n) ) = σ · (τ · p),
and clearly e · p = p for the identity e, so Sn acts on X. Note that since σ ∈ Sn takes each term
x j − xi with i < j to plus or minus another term of the same form in a one-to-one fashion (in that
σ is one-to-one), we do indeed have σ · ∆ = ±∆. That ε is a homomorphism follows from the
fact that Sn acts on X, since
ε(σ τ)∆ = σ τ(∆) = σ (ε(τ)∆) = ε(τ)σ (∆) = ε(σ )ε(τ)∆.
It remains only to check that ε(τ) = −1 for any transposition τ = (k `) with 1 ≤ k < ` ≤ n. For
this, note that τ(x j − xi ) = x j − xi unless {i, j} ∩ {k, l} = ∅. If {i, j} = {k, `}, then τ(x` − xk ) =
−(x` − xk ). The remaining terms have the form ±(x` − xm ) or ±(xk − xm ) for some m 6= k, `. We
consider these in pairs. If m < k, then we have
τ((x` − xm )(xk − xm )) = (xk − xm )(x` − xm ) = (x` − xm )(xk − xm ).
If m > `, we have
τ((xm − xk )(xm − x` )) = (xm − x` )(xm − xk ) = (xm − xk )(xm − x` ),
and if k < m < `, we have
τ((xm − xk )(x` − xm )) = (xm − x` )(xk − xm ) = (xm − xk )(x` − xm ).
Therefore, the product of the contributions to ε(τ) from the various terms is −1, as required. 
We once again exhibit that A5 is simple, and moreover, that it is the only simple subgroup of
order 60, up to isomorphism.
134 4. ADVANCED GROUP THEORY

L EMMA 4.12.2. If G is a group of order 60 with n5 (G) > 1, then G is simple.

P ROOF. The assumption forces n5 (G) = 6 by the third Sylow theorem. So, the normalizer
of any Sylow p-subgroup has order 10. Let N be a proper normal subgroup of G. If 5 | |N|,
then N contains a Sylow 5-subgroup of G and hence all Sylow 5-subgroups of G by the second
Sylow theorem. But then N has at least 1 + 6 · 4 = 25 elements, so is of order 30. But N has
a unique subgroup of order 5 by Example 4.11.7, which is normal in G by Lemma 4.3.15a and
Lemma 4.3.16. This contradicts n5 (G) = 6. It follows that 5 - |N|. Now, if N has order 6 or 12,
then again it has a normal Sylow subgroup which is then by the same reasoning itself normal in
G. So, we may assume that |N| ∈ {2, 3, 4}. Then |G/N| ∈ {15, 20, 30}, and in all of these cases,
G/N has a normal subgroup of order 5 by the third Sylow theorem and Example 4.11.7. But then
G itself has a normal subgroup with order divisible by 5, which we have already shown is not the
case. Thus N must be the trivial subgroup. 
As a corollary, we recover Proposition 4.6.26 that A5 is simple. Let us prove that this is the
only subgroup of order 60.
P ROPOSITION 4.12.3. The group A5 is isomorphic to every simple group of order 60.
P ROOF. Let G be a simple group of order 60, which we know exists by Proposition 4.6.26.
We show that G is isomorphic to A5 . From the third Sylow theorem, the possibilities for n2 (G)
are 3, 5, and 15, which is also the index of the normalizer N of a Sylow 2-subgroup P. Since G
is not isomorphic to a subgroup of S3 , we can eliminate n2 (G) = 3.
If n2 (G) = 5, then G is isomorphic to a subgroup of S5 , so G may be identified with a normal
subgroup of S5 of index 2. It follows that G ∩ A5 is a normal subgroup of A5 which is either A5 or
of index 2 in A5 . The latter being impossible by the simplicity of A5 , we must have that G = A5 .
Suppose now that n2 (G) = 15. Since 15 6≡ 1 mod 4, Proposition 4.11.14 tells us that |P∩Q| =
2 for some Q ∈ Syl2 (G) with Q 6= P. Set M = NG (P ∩ Q), which is not G since G is simple. Since
|M| is a multiple of 4 that is greater than 8 by Lemma 4.11.12, we must have |M| = 12 or 20, from
which it follows that M has index at most 5, and therefore G is isomorphic to a subgroup of S5 .
The same argument as before would tell us that G ∼ = A5 , but note that we assumed n2 (G) = 15,
so we reach a contradiction. 
We now prove that the alternating groups on at least 5 elements are simple.
T HEOREM 4.12.4. The groups An for n ≥ 5 are simple.
P ROOF. We prove this by induction on n ≥ 5, the case n = 5 having been proven in Propo-
sition 4.6.26. Let G = An for some n ≥ 6. For any i ∈ Xn , the stabilizer Gi is isomorphic to
An−1 , which is simple by induction. Suppose that N is a nontrivial normal subgroup of G. If
there exists i ∈ Xn and τ ∈ N − {e} with τ(i) = i, then N ∩ Gi is a nontrivial normal subgroup
of Gi , and it follows that Gi 6 N by the simplicity of Gi . For any j ∈ Xn , we can find σ ∈ An
with σ (i) = j, and then G j = σ Gi σ −1 6 N by normality of N. As every element of An can be
written as a product of an even number of transpositions, every element of An may be written as
a product of products of two transpositions, and any product of two transpositions lies in G j for
some j ∈ Xn since n > 4. Thus, we must have that N = G = An .
 4.13. FREE GROUPS AND PRESENTATIONS 135

Now, we show that N must contain a permutation that fixes some element of Xn . Let τ ∈ N.
If τ(i) = τ 0 (i) for any τ 0 ∈ N and i ∈ Xn , then τ 0 τ −1 fixes i. If the cycle decomposition of τ
contains a k-cycle with k ≥ 3, say (a1 a2 . . . ak ), then we may choose σ ∈ An that fixes a1 and
a2 but not a3 . If τ is a product of disjoint transpositions that does not fix any element, then write
τ = (a1 a2 )(a3 a4 ) . . . (am−1 am ) and take σ = (a1 a2 )(a3 a5 ) (using the fact that n ≥ 6). It follows
in both cases that τ 0 = σ τσ −1 6= τ, but τ 0 (a1 ) = τ(a1 ), as desired. 

4.13. Free groups and presentations

We begin with a general definition of a free group by its “universal property”.
D EFINITION 4.13.1. A group F is free on a subset X if, whenever f : X → G is a function,
where G is a group, there exists a unique homomorphism
φf : F → G
such that φ f (x) = f (x) for all x ∈ G. The existence of this unique homomorphism is referred to
as the universal property of F.
P ROPOSITION 4.13.2. Let F be free on a set X and F 0 be free on a set X 0 , and suppose
f : X → X 0 is a bijection. Then the homomorphism φ f : F → F 0 given by the universal property
is an isomorphism.
P ROOF. Let g be the inverse to f , and let φg0 : F 0 → F be the homorphism given by the
universal property for F 0 . Then φg0 ◦ φ f (x) = x for all x ∈ X and φ f ◦ φg0 (x0 ) = x0 for all x0 ∈ X.
Since the identity homomorphisms of F and F 0 also take elements of X and X 0 to themselves,
respectively, the the universal property for F and for F 0 imply that φg0 ◦φ f = idF and φ f ◦φg0 = idF 0 ,
respectively. Therefore, we have that φg0 = φ −1 f , so φ f is an isomorphism. 
E XAMPLE 4.13.3. The integers Z are a free group on the subset {1}, since for any group G
and element x ∈ G, we can define φ : Z → G with φ (1) = x by φ (n) = xn for all n ∈ Z, and this
is the unique homomorphism taking 1 to x.
E XAMPLE 4.13.4. The group Zn , although a free abelian group, is not a free group. For
example, take n = 2. Then the map f : {(1, 0), (0, 1)} → D3 with f (1, 0) = r and f (0, 1) = s
cannot be extended to a homomorphism φ : Z2 → D3 , for such a function would have to satisfy
rs = φ (0, 1)φ (1, 0) = φ (1, 1) = φ (1, 0)φ (0, 1) = sr,
which does not hold in D3 .
To show the existence of free groups on larger sets, we construct them explicitly.
D EFINITION 4.13.5. A word on in a set X is a symbol
x1n1 x2n2 · · · xknk
with x1 , x2 , . . . , xk ∈ X and n1 , n2 , . . . , nk ∈ Z, where k ≥ 0. If k = 0, we sometimes denote this
word by e, and it is called the empty word.
R EMARK 4.13.6. We write the word x1 for x ∈ X more simply as x.
136 4. ADVANCED GROUP THEORY

D EFINITION 4.13.7. The product of two words w = x1n1 x2n2 · · · xknk and v = ym 1 m2 ml
1 y2 · · · yl in X
is the concatenation
w · v = wv = x1n1 x2n2 · · · xknk ym 1 m2 ml
1 y2 · · · yl .
Clearly, concatenation is an associative binary operation on the set of words, and e is an
identity element for this operation.
Recall from Example 1.2.25 that there is a smallest equivalence relation containing any rela-
tion on a set. So, let us define an equivalence relation on the set of words by a set of generators.
D EFINITION 4.13.8. The standard equivalence relation ∼ on the set WX of words on X is the
smallest equivalence relation such that
(4.13.1) wv ∼ wx0 v
and
(4.13.2) wxm+n v ∼ wxm xn v
for all w, v ∈ WX , x ∈ X, and m, n ∈ Z.
Two words are then equivalent if and only if one can be obtained from the other by a finite
sequence of operations on the word consisting each of adding or removing an x0 for some x ∈ X
or changing xm+n in a word to xm xn or changing xm xn in a word to xm+n for some x ∈ X and
m, n ∈ Z.
D EFINITION 4.13.9. We say that a word x1n1 x2n2 · · · xknk in X is reduced if xi 6= xi+1 for all
1 ≤ i ≤ k − 1 and ni 6= 0 for all 1 ≤ i ≤ k.
P ROPOSITION 4.13.10. Every word is equivalent to a unique reduced word.
P ROOF. The relation (4.13.2) tells us recursively for any n ≥ 1 that
(4.13.3) wxn v ∼ w(x · x · · · x)v and wx−n v ∼ w(x−1 · x−1 · · · x−1 )v,
for all w, v ∈ WX and x ∈ X, with n symbols “x” appearing on the right-hand sides. If we start
with a word w in X, we may use (4.13.3) to expand it and (4.13.1) to remove any 0-powers of
elements of X, to obtain an equivalent word of the form
x1±1 x2±1 · · · xk±1 .
−1
We may use (4.13.1) and (4.13.3) to remove terms of the form xi xi+1 or xi−1 xi+1 with xi = xi+1 ,
relabeling after each step, until no such terms exist. We may then gather terms by again applying
(4.13.2) to obtain a reduced word equivalent to w.
The process we have described does not change a reduced word. Moreover, the operations
of adding in or removing an x0 from a word or changing xm+n to xm xn for some m, n ∈ Z or
vice-versa do not change the result of the process. Therefore, each word is equivalent to a unique
reduced word. 
N OTATION 4.13.11. The set of equivalence classes of words on a set X is denoted FX .
P ROPOSITION 4.13.12. The set FX is a group under concatenation of words, and it is gener-
ated by the set X.
 4.13. FREE GROUPS AND PRESENTATIONS 137

P ROOF. We give only a sketch. First, we must check that if w ∼ w0 and v ∼ v0 are two
pairs of equivalent words in X, then wv ∼ w0 v0 . This follows quickly from the definition of the
equivalence relation ∼. So, the binary operation is well-defined, associative, and has identity e.
Moreover, the inverse of the equivalence class of a word x1n1 x2n2 · · · xknk is the equivalence class of
the word
xk−nk · · · x2−n2 x1−n1 .
That X generates FX is simply Proposition 2.4.3. 
R EMARK 4.13.13. It is typical to denote an element of FX by any word representing it, which
means that we will use the symbol “=” instead of “∼” when interpreting these words as elements
of FX .
E XAMPLES 4.13.14.
a. The free group F{x} consists exactly of all xn for n ∈ Z, and only x0 = e, so F{x} ∼
= Z.
b. The free group F{x,y} with x 6= y consists of all words
xn1 ym1 xn1 ym2 · · · xnk ymk ,
where we can take ni 6= 0 for i ≥ 2 and m j 6= 0 for j < k. We have, e.g.,
x2 y−1 x−3 y−1 · yx3 y2 x5 = x2 yx5 .
L EMMA 4.13.15. The group FX is a free group on the set X.
P ROOF. By Definition 4.13.1, we must show that for any group G and function f : X → G,
the function
φ f (x1n1 x2n2 · · · xknk ) = f (x1 )n1 f (x2 )n2 · · · f (xk )nk
is the unique well-defined homomorphism such that φ f (x) = f (x) for all x ∈ X. That φ f is a
homomorphism is a direct consequence of its definition and the definition of multiplication of
words by concatenation, once it is seen to be well-defined. That it is well-defined is a conse-
quence of the fact that the only relations that are imposed on words are those that exist in any
group. That is, for words v and w, x ∈ X, and m, n ∈ Z, we have
φ f (vx0 w) = φ f (v)φ f (x0 )φ f (w) = φ f (v) f (x)0 φ f (w) = φ f (v)φ f (w),
φ f (vxm+n w) = φ f (v) f (x)m+n φ f (w) = φ f (v) f (x)m f (x)n φ f (w) = φ f (vxm xn w),
so φ f is constant on equivalent words. 
D EFINITION 4.13.16. The group FX of Proposition 4.13.12 is the free group on a set X.
Proposition 4.13.2 then immediately implies the following.
C OROLLARY 4.13.17. If X and Y are sets with the same cardinality, then FX and FY are
isomorphic.
The following then provides an object that is well-defined up to isomorphism.
N OTATION 4.13.18. The free group on a set with n elements is denoted Fn .
The following gives the relationship between free groups and free abelian groups.
138 4. ADVANCED GROUP THEORY

P ROPOSITION 4.13.19. The free abelian group on a set X is isomorphic to the abelianization
of the free group on X.
P ROOF. Define π : FX → x∈X Z by π(x) = ex , where ex is the standard basis element of
L
L
x∈X Z corresponding to x. This is a surjective homomorphism. AsL the image of π is abelian,
ab
the map π factors through a sujrective homomorphism π̄ : FX → x∈X Z, and therefore the
maximal abelian quotient of FX surjects onto x∈X Z. In FXab , we may rearrange the terms of the
L

image of x1n1 x2n2 · · · xknk for xi ∈ X and ni ∈ Z with 1 ≤ i ≤ k so that it is the image of a like element
with x1 , . . . , xk are all distinct. Now, such an element is sent to ∑ki=1 ni exi under π̄, and so it is 0 if
and only if all ni = 0. If follows that π̄ is an isomorphism. 
T HEOREM 4.13.20. The free group Fn on n elements cannot be generated by fewer than n
elements.
P ROOF. If Fn could be generated by n − 1 elements, then Fnab could be generated by n − 1
elements by Proposition 4.13.19. But Fnab is isomorphic to the free abelian group on n elements,
so Theorem 4.4.16 tells us that Fnab cannot be generated by n − 1 elements. 
We omit the proof of the following theorem.
T HEOREM 4.13.21. Every subgroup of a free group is also a free group.
One of the most important uses of free groups is to give presentations of groups. First, we
make the following definition, recalling Lemma 1.2.24 to see that it is well-defined.
D EFINITION 4.13.22. The normal closure of a subset S of a group G is the smallest normal
subgroup of G containing S, equal to the intersection of all normal subgroups of G containing S.
D EFINITION 4.13.23. A presentation of a group G on a set S and a subset T of FS is a
surjective homomorphism FS → G with kernel equal to the normal closure of the set T . We say
that G is presented by the generating set S and the relation set T , and we write G ∼
= hS | T i.
R EMARK 4.13.24. If G is presented by S and T , then the first isomorphism theorem tells us
that G ∼
= FS /R by an isomorphism sending s ∈ S to the coset of s in FS /R, where R is the normal
closure of T .
In fact, we have the following stronger result.
P ROPOSITION 4.13.25. Suppose that G and G0 are groups with G presented by S and T .
Suppose we are given a subset {xs | s ∈ S} of G0 . Then there exists a homomorphism φ : G → G0
with φ (s) = xs for all s ∈ S if and only if the unique homomorphism Φ : FS → G0 with Φ(s) = xs
for all s ∈ S satisfies T ⊆ ker Φ.
P ROOF. The existence and uniqueness of Φ is by the universal property of FS . If T ⊆ ker Φ,
then the normal closure R of T is contained in ker Φ since ker Φ is a normal subgroup of Fr,s con-
taining T . In this case, the first isomorphism theorem implies that Φ induces a map Φ̄ : FS /R → T
with Φ̄(sR) = xs for all s ∈ S. Since G ∼ = FS /R by an isomorphism sending s to sR, the compo-
0
sition φ : G → G is the desired map. Similarly, if φ exists, then we may compose it with the
surjection FS −∼ → G taking s ∈ S to s to obtain a map Φ : FS → G0 with R, and hence T , in its
kernel. 
 4.13. FREE GROUPS AND PRESENTATIONS 139

D EFINITION 4.13.26. If G is presented by finite sets S = {s1 , s2 , . . . , sk } and T = {r1 , r2 , . . . , rd },

then G is said to be finitely presented.
N OTATION 4.13.27. We write
G = hs1 , s2 , . . . , sk | r1 , r2 , . . . , rd i.
to denote that G has a presentation by sets S = {s1 , s2 , . . . , sk } and T = {r1 , r2 , . . . , rd }.
E XAMPLES 4.13.28. We give several examples of presentations:
a. FS ∼
= hS | ∅i,
b. Z2 ∼
= ha, b | aba−1 b−1 i,
c. Z/nZ ∼
= ha | an i,
d. Dn ∼
= hr, s | rn , s2 , rsrsi.
R EMARK 4.13.29. One sometimes writes
hs1 , s2 , . . . , sk | r1 = r10 , r2 = r20 , . . . , rd = rd0 i
for a finite presentation
hs1 , s2 , . . . , sk | r1−1 r10 , r2−1 r20 , . . . , rd−1 rd0 i.
E XAMPLE 4.13.30. We have
Zn ∼
= hx1 , . . . , xn | xi x j = x j xi for 1 ≤ i < j ≤ ni.
Note that we can start with the presentation, rather than a group, in order to define new
groups.
E XAMPLE 4.13.31. The quaternion group Q8 is the group of order 8 with the presentation
Q8 = hi, j | i4 = e, i2 = j2 , i j = ji−1 i.
The elements of Q8 are usually labelled {±1, ±i, ± j, ±k}, with k = i j, −1 = i2 , −i = i3 , − j = j3 ,
and −k = ji = k3 . We remark that Q8 ∼6 D4 .
=
Sometimes, we just end up with complicated presentations of familiar groups.
E XAMPLE 4.13.32. Consider the group
G = hx, y | x2 y, x4 yi.
Then e = (x2 y)−1 x4 y = y−1 x2 y, which forces x2 = e, and then y = x2 y = e. Since in fact x2 = e
and y = e imply x2 y = x4 y = e, the group G also has a presentation
G = hx, y | x2 , yi,
and so is just hxi ∼
= Z/2Z.
140 4. ADVANCED GROUP THEORY

R EMARK 4.13.33. To determine whether or not two presentations yield isomorphic groups
is a very difficult question. So difficult, in fact, that it has been shown that there does not exist
a single recursive computer algorithm into which one can input any two finite group presenta-
tions that will always output whether or not they are yield isomorphic groups. Moreover, there
exist presentations of groups G from which one cannot write a computer algorithm to determine
whether or not a word in the generators of G is equal to the identity of G.
D EFINITION 4.13.34. Let G1 and G2 be groups. A free product G1 ∗ G2 of G1 and G2 is a
group for which there exist homomorphisms ιi : Gi → G1 ∗ G2 for i ∈ {1, 2} such that for any
group K and group homomorphisms φi : Gi → K with i ∈ {1, 2}, there exists a unique homomor-
phism Φ : G1 ∗ G2 → K with Φ ◦ ιi = φi for i ∈ {1, 2}.
P ROPOSITION 4.13.35. Let G1 and G2 be groups. Then the free product of G1 and G2 exists
and is unique up to isomorphism. Moreover, if Gi has a presentation Gi ∼= hSi | Ti i for each
i ∈ {1, 2}, then the free product is isomorphic to
G1 ∗ G2 ∼
= hS1 q S2 | T1 q T2 i,
where q denotes the disjoint union.
P ROOF. We verify that the group N = hS1 q S2 | T1 q T2 i is a free product of G1 and G2 . We
leave the uniqueness of the free product up to isomorphism as an exercise for the reader. Let
πi : FSi → Gi be the surjections defining the presentation of Gi for i ∈ {1, 2}. Define homomor-
phisms χi : FSi → N by letting χi (s) equal the image of s in N for all s ∈ Si . By definition of N,
we have that Ti is contained in the kernel of χi , so the first isomorphism theorem provides maps
ιi : Gi → N such that ιi (πi (s)) is the image of s in N for any s ∈ Si .
Now, for i ∈ {1, 2}, let φi : Gi → K be a homomorphism to some group K. Then we have a
unique map Ψ : FSi qS j → K determined by Ψ(s) = φi (πi (s)) for all s ∈ Si for i ∈ {1, 2}. If t ∈ Ti
for some i, then Ψ(t) = φi (e) = e, so the the first isomorphism theorem yields a homomorphism
Φ : N → K such that Φ(ιi (g)) = φi (g) for all g = πi (s) for some s ∈ S for i ∈ {1, 2}. However,
the elements of πi (Si ) generate Gi , so we have that Φ ◦ ιi = φi for each i. Moreover, Φ is unique,
as its values on the images of the elements of S1 q S2 are determined by the latter equalities. 
R EMARK 4.13.36. An element of the free product of groups G and H is an equivalence class
of words g1 h1 g2 h2 · · · gk hk with gi ∈ Gi and hi ∈ Hi for 1 ≤ i ≤ k (under an equivalence relation
under which the identity elements of the two groups are each identified with the identity element
of the free product and which otherwise only imposes the relations of the original groups within
words), with multiplication induced by concatenation.
 CHAPTER 5

Advanced ring theory

5.1. Unique factorization domains

In this section, we investigate the role that prime numbers play in the integers in greater
generality. Recall that every nonzero integer can be written as plus or minus a product of distinct
prime powers, and these prime powers are unique. Note that the units in Z are ±1, so we can
say that every nonzero integer can be written as a product of prime powers times a unit. In this
section, we investigate this property for a larger class of integral domains.
First, we introduce an analogue of prime numbers.
D EFINITION 5.1.1. Let R be an integral domain. A nonunit and nonzero element p ∈ R is
said to be an irreducible element if for every a, b ∈ R with p = ab, either a or b is a unit.
D EFINITION 5.1.2. Two elements a and b of a nonzero commutative ring R are said to asso-
ciates if a = ub with u ∈ R× .
Of course, the property of being associate is an equivalence relation on an integral domain R.
The equivalence class of 0 is {0} and that of 1 is R× . We have the following simple lemma, which
tells us that the equivalence class of an irreducible element consists of irreducible elements.
L EMMA 5.1.3. If R is an integral domain, and p ∈ R is irreducible, then so is every associate
of p.
P ROOF. That is, if u ∈ R× and up = ab for a, b ∈ R, then p = (u−1 a)b. As p is irreducible,
either u−1 a ∈ R× or b ∈ R× . Finally, if u−1 a ∈ R× , then a ∈ R× . 
E XAMPLES 5.1.4.
a. The irreducible elements of Z are ±p for prime numbers p. The elements p and −p are
associates.
b. The irreducible elements of F[x], for a field F, are the irreducible polynomials of F, since
the units of F[x] are the nonzero constant polynomials. Every nonzero polynomial has a unique
associate with leading coefficient equal to 1.
c. In the subring of C that is
√ √
Z[ −2] = {a + b −2 | a, b ∈ Z},
√ √
a number of prime integers√are no longer √ irreducible. For instance
√ 2 = −(√ −2)2 , and −2 is
Also, 3 = (1 + −2)(1
not a unit. √ √ − −2), and neither√ 1 + −2 nor 1 − −2 is a unit, for if,
e.g., u ∈ Z[ −2] with u(1 + −2) = 1, then 3u = 1 − −2, which is clearly impossible. On the
other hand, it turns out that 5 is irreducible, though we do not prove this now.
141
142 5. ADVANCED RING THEORY

D EFINITION 5.1.5. An integral domain R is a unique factorization domain, or a UFD, if

every nonzero, nonunit element a ∈ R can be written as a product
a = p1 p2 · · · pr
with p1 , p2 , . . . , pr irreducible elements of R for some r ≥ 1, and moreover, this expression is
unique in the sense that if
a = q1 q2 · · · qs
with q1 , q2 , . . . , qs irreducible for some s ≥ 1, then s = r and there exists a permutation σ ∈ Sr
such that qσ (i) and pi are associates for all 1 ≤ i ≤ r.
R EMARK 5.1.6. If one wants to allow units, one can rephrase Definition 5.1.5 to read that
every nonzero element a ∈ R can be written as a = up1 · · · pr with u ∈ R× and p1 , p2 , . . . , pr
irreducible in R for some r = 0 in a unique manner such that any such decomposition of a =
vq1 · · · qs has s = r and, after a reordering of the irreducibles, each qi is an associate of pi .
E XAMPLE 5.1.7. The ring Z is a unique factorization domain.
As we shall see later, F[x] for a field F is a unique factorization domain as well.
E XAMPLE 5.1.8. Consider the subring F[x2 , xy, y2 ] of F[x, y]. It consists exactly of the poly-
nomials in F[x, y] that can be written as polynomials in x2 , xy, and y2 . These latter three elements
are irreducible in F[x2 , xy, y2 ], but we have
x2 · y2 = xy · xy,
so factorization is not unique.
A more standard example is the following.
√
E XAMPLE 5.1.9. Consider the subring Z[ −5] of C. We have
√ √
6 = 2 · 3 = (1 + −5)(1 − −5).
√
The element√ 2 divides only
√ elements of the form a + b −5 with a, b ∈ Z even,√so it does not
divide 1 + −5 or 1 − −5. On the other hand, 2 is irreducible since if a + b −5 divides 2,
then so does its complex conjugate, and then
√ √
(a + b −5)(a − b −5) = a2 + 5b2
√
divides 2, which happens only if a = ±1 and b = 0. Therefore, Z[ −5] is not a unique factor-
ization domain.
One advantage of unique factorization domains is that they allow us to define a concept of
greatest common divisor.
D EFINITION 5.1.10. Let R be a UFD. Let a1 , a2 , . . . , ar ∈ R be nonzero. A principal ideal (d)
for d ∈ R is said to be the greatest common divisor, or GCD, of a1 , a2 , . . . , ar if d divides ai for
each 1 ≤ i ≤ r and if d 0 also divides each ai , then d 0 divides d.
The element d in the definition of GCD, if it exists, is only defined up to unit. On the other
hand, (d) is independent of this choice.
 5.1. UNIQUE FACTORIZATION DOMAINS 143

L EMMA 5.1.11. Let R be a UFD. Then every collection a1 , a2 , . . . , ar of nonzero elements of

R has a GCD.
P ROOF. We sketch the proof. Factor each ai into a unit times a product of irreducibles.
If there exists an irreducible element p1 that divides each ai , an associate of it is one of the
irreducibles appearing in the factorization of ai . We then have bi ∈ R with ai = pbi for each
i, and the factorization of bi has one fewer irreducible element than that of ai . We repeat this
process until the collection no longer has a common irreducible divisors, obtaining irreducibles
p1 , p2 , . . . , pk such that d = p1 p2 · · · pk divides every ai .
We claim that (d) is the GCD of a1 , a2 , . . . , ak . If not, then there exists d 0 that does not divide
d which divides every ai . This means that there exists an irreducible element q ∈ R and some
n ≥ 1 such that qn divides d 0 but not d. Then qn divides every ai , which means since qn does
not divide d that q actually divides each ci such that ai = dci , in contradiction to the definition of
d. 
One advantage of having the notion of a GCD is that in quotient fields, it allows us to talk
about fractions being in lowest terms.
a
D EFINITION 5.1.12. Let R be a UFD, and let a, b ∈ R with b 6= 0. We say that the fraction b
is reduced, or in lowest terms, if the GCD of a and b is (1).
L EMMA 5.1.13. Let R be a UFD. Every fraction in Q(R) may be written in lowest terms.
P ROOF. Let a, b ∈ R with b 6= 0. Let (d) be the GCD of a and b. Then there exist a0 , b0 ∈ R
with a = da0 and b = db0 , and we have that the GCD of a0 and b0 is (1). We therefore have that
a0 a
b0 = b , and the former form of the fraction is in lowest terms. 
Let us study factorization in principal ideal domains.
D EFINITION 5.1.14. Let X be a set, and let ≤ be a partial ordering on X.
a. An ascending chain in X is a sequence (ai )i≥1 of elements of X such that ai ≤ ai+1 for all
i ≥ 1.
b. We say that X satisfies the ascending chain condition, or ACC, if every ascending chain
(ai )i≥1 in X is eventually constant: i.e., there exists j ≥ 1 such that ai = a j for all i ≥ j.
The following is an equivalent characterization of the ACC.
P ROPOSITION 5.1.15. A nonempty set X with a partial ordering ≤ satisfies the ACC if and
only if every nonempty subset of X contains a maximal element.
P ROOF. If every nonempty subset of X contains a maximal element, then clearly ascending
chains are eventually constant: i.e., their underlying sets are finite. For the other direction, it
suffices to show that if X satisfies the ACC, then it contains a maximal element. Let C be a
nonempty chain in X, and suppose it does not have an upper bound. For each x ∈ C, there exists
y ∈ C with y > x, as otherwise x would be an upper bound. We may therefore recursively pick
ai ∈ X with ai < ai+1 for each i, but this is impossible. Thus C has an upper bound, and therefore
X has a maximal element by Zorn’s lemma. 
144 5. ADVANCED RING THEORY

D EFINITION 5.1.16. We say that a commutative ring R is noetherian if the set of its ideals
satisfies the ascending chain condition with respect to containment of ideals.
R EMARK 5.1.17. We may rephrase the condition that R be noetherian by saying that if
(In )n≥1 is an ascending chain of ideals, then there exists m ≥ 1 such that the union I of the
In with n ≥ 1 equals Ii for all i ≥ m.
R EMARK 5.1.18. One may define a noncommutative ring to be left noetherian (resp., right
noetherian rings) if it satisfies the ACC on left ideals (resp., right ideals). In general, a noetherian
ring is taken to be one that is both left and right noetherian.
T HEOREM 5.1.19. A commutative ring R is noetherian if and only if every ideal of R is finitely
generated.
P ROOF. Suppose that every ideal of R is finitely generated. Let (In )n≥1 be a chain of ideals of
R. Let I be the union of the In for n ≥ 1, which is an ideal by Lemma 3.10.10. Since I is finitely
generated, I = (a1 , a2 , . . . , ar ), with ak ∈ I with 1 ≤ k ≤ r for some r ≥ 1. For each k, there exists
mk ≥ 1 with ak ∈ Imk , and if we let m be the maximum of the mk , then ak ∈ Im for every ak . Since
I is the smallest ideal of R containing each ak , we have I ⊆ Im , which forces I = Im .
Conversely, suppose R is noetherian, and let I be an ideal of R. Let x1 ∈ I, and sup-
pose inductively that we have constructed x1 , x2 , . . . , xn ∈ R with the property that if we set
Ik = (x1 , x2 , . . . , xk ) for every 1 ≤ k ≤ n, then Ik ⊆ Ik+1 for every 1 ≤ k ≤ n − 1. If In 6= I, then
let xn+1 ∈ I with xn+1 ∈ / In . Then In+1 = (x1 , x2 , . . . , xn+1 ) properly contains In . If this process
repeats indefinitely, then we have constructed an ascending chain (In )n≥1 that is not eventually
constant, which would contradict the assumption that R is noetherian. Therefore, there exists
m ≥ 1 such that Im = I, and so I = (a1 , a2 , . . . , am ) is finitely generated. 
C OROLLARY 5.1.20. Every principal ideal domain is noetherian.
P ROPOSITION 5.1.21. Let R be a noetherian domain. Then every nonzero, nonunit a ∈ R
may be written as a = p1 p2 · · · pr with the pi ∈ R irreducible for all 1 ≤ i ≤ r and some r ≥ 1.
P ROOF. Let
X = {(a) | a ∈ R − (R× ∪ {0}) is not a product of irreducible elements}.
which we order by inclusion of ideals. If X 6= ∅, then X contains a maximal element (a) by
Proposition 5.1.15. Since a is not irreducible, we may write a = bc for some b, c ∈ / R× . Then
(b) and (c) properly contain (a) as R is a domain. By maximality of (a), we may write both b
and c as products of irreducible elements. Since a = bc, we have that a is product of irreducible
elements as well, which is a contradiction. Thus X is empty, as required. 
L EMMA 5.1.22. Let R be a PID, and let a ∈ R be nonzero. Then (a) is maximal if and only if
a is an irreducible element.
P ROOF. Clearly, a cannot be a unit for either condition to hold. If a = bc with b and c
non-units, then (a) ( (b) ( R, so (a) is not maximal. And if (a) is not maximal, then there
exists an proper ideal I = (c) of R properly containing (a), so we may write a = bc with b ∈ R.
Since the containment is proper, b is not a unit, and c is not a unit by definition. Therefore, a is
reducible. 
 5.1. UNIQUE FACTORIZATION DOMAINS 145

D EFINITION 5.1.23. An element p of a domain R is prime if for every a, b ∈ R such that

p | ab, either p | a or p | p.
L EMMA 5.1.24. In a domain, prime elements are irreducible.
P ROOF. Let R be a domain. Let p ∈ R be prime and suppose p = ab for a, b ∈ R. Then p | a
or p | b as p is prime. Suppose then without loss of generality that a = pc for some c ∈ R. We
then have p = pbc, and by cancellation this forces bc = 1, so b ∈ R× . Thus, p is irreducible. 
L EMMA 5.1.25. In a UFD, irreducible elements are prime.
P ROOF. Let R be a UFD, and let π ∈ R be irreducible. Suppose that a, b ∈ R with π | ab.
Write ab = πc for c ∈ R. Write a, b, and c as products of irreducible elements times a unit. Since
R is a UFD, we must then have that π is associate to one of the irreducible elements of which a
and b are products. That is, π divides a or b. 
√
E√XAMPLE 5.1.26. In Example 5.1.9, we showed that 2 is irreducible in Z[ −5], and 6 =
√ √
(1 + −5)(1 − −5), but 2 - (1 ± −5). Thus, 2 is not prime.
P ROPOSITION 5.1.27. Let R be a domain in which every nonzero, non-unit element a has a
factorization a = p1 p2 · · · pr with pi prime for 1 ≤ i ≤ r and r ≥ 1. Then R is a UFD.
P ROOF. Let a ∈ R be a nonzero, nonunit element, and write a = p1 p2 · · · pr with p1 , p2 , . . . , pr
prime. By Lemma 5.1.24, this is also a factorization of a into irreducible elements. Suppose that
a = q1 q2 · · · qs
with q1 , q2 , . . . , qs irreducible. If s = 1, then a is irreducible, so r = 1 and p1 = q1 . Suppose by
induction we have proven uniqueness whenever there is a decomposition of a with fewer than
s ≥ 2 irreducibles.
As pr is prime, we have that pr divides some qi for some 1 ≤ i ≤ s. Since qi is irreducible,
this means that qi = wpr with w ∈ R× . Since R is an integral domain, we then have
p1 p2 · · · pr−1 = wq1 q2 · · · qi−1 qi+1 · · · qs .
As s ≥ 2 by assumption, note that wq1 is an associate to q1 and the expression on the right is a
product of s − 1 irreducible elements. By induction, we have r = s, and there exists a bijective
function
σ : {1, 2, . . . , r − 1} → {1, 2, . . . , i − 1, i + 1, . . . , r}
with qσ (i) and pi associates for each 1 ≤ i ≤ r − 1. We may extend σ to an element of Sr by
setting σ (r) = i, and then qσ (r) = qi is an associate of pr as well, proving uniqueness. 
In a principal ideal domain, irreducible elements are in fact prime.
L EMMA 5.1.28. Let R be a PID. If p ∈ R is irreducible, then p is prime.
P ROOF. Let a, b ∈ R with p | ab. Then ab ∈ (p), and (p) is maximal by Lemma 5.1.22. Since
every maximal ideal of R is prime, we have that (p) is prime, and therefore either a ∈ (p) or
b ∈ (p). 
The following key theorem is now a consequence of results we have already proven.
146 5. ADVANCED RING THEORY

T HEOREM 5.1.29. Every principal ideal domain is a unique factorization domain.

P ROOF. Suppose that R is a PID. As R is noetherian, Proposition 5.1.21 states that we may
factor every nonzero, nonunit element in R as a product of irreducible elements. As irreducible
elements of a PID are prime, Proposition 5.1.27 tells us that R is a UFD. 
Given that every polynomial ring over a field is a PID, we have the following corollary. It is
an interesting exercise to prove it directly.
C OROLLARY 5.1.30. For any field F, the ring F[x] is a unique factorization domain.
Corollary 3.9.2 tells us what we may already have known from experience, that we can factor
one-variable polynomials into irreducible factors over a field, and there is only one way to do
this.

5.2. Polynomial rings over commutative rings

Now that we know that every PID is a UFD, the question arises: is every UFD also a PID?
The answer, in fact, is no. For this, let us examine polynomial rings over integral domains in a
bit more detail.
D EFINITION 5.2.1. Let R be an integral domain. A polynomial f ∈ R[x] is said to be primitive
if the only elements of R that divide all of the coefficients of f are units.
In a UFD, we can actually talk about the GCD of the coefficients of a polynomial.
D EFINITION 5.2.2. Let R be a UFD. The content of the a polynomial in R[x] is the GCD of
its coefficients.
R EMARK 5.2.3. If R is a UFD, then a polynomial in R[x] is primitive if and only if the GCD
of its coefficients is (1).
D EFINITION 5.2.4. A polynomial in R[x] for a nonzero ring R is said to be monic is its leading
coefficient is 1.
R EMARK 5.2.5. Monic polynomials in R[x], where R is a UFD, are primitive.
L EMMA 5.2.6. Let R be a UFD. If (c) is the content of f ∈ R[x] for, then there exists a
primitive polynomial g ∈ R[x] with f = cg.
P ROOF. By definition, c divides each coefficient of f , so f = cg for some g ∈ R[x]. Let d ∈ R
be such that (d) is the content of g. Then g = dh for some h ∈ R[x], so we have f = cdh. But this
implies that cd divides every coefficient of f , so cd divides the content c, forcing d to be a unit
as R is a domain. Therefore, g is primitive. 
E XAMPLE 5.2.7. The polynomial f = 25x2 + 10x − 15 in Z[x] has content 5, and so it is not
primitive. In fact, f = 5g, where g = 5x2 + 2x − 3, and g is primitive.
L EMMA 5.2.8 (Gauss’s Lemma). Let R be a UFD. Then the product of any two primitive
polynomials in R[x] is primitive.
 5.2. POLYNOMIAL RINGS OVER COMMUTATIVE RINGS 147

P ROOF. Let
n m
f = ∑ ai xi and g = ∑ b jx j
i=0 j=0

be primitive polynomials in R[x]. The kth coefficient of f g is ck = ∑ki=0 ai bk−i . If p is an irre-

ducible element of R, then since f and g are primitive, there exist minimal nonnegative integers
r and s such that p - ar and p - bs . Since p | ai for i < r and p | b j for j < s, which is to say that
p | br+s−i for r < i ≤ r + s, we have that p divides every term of cr+s except ar bs , which it does
not divide. Therefore, p does not divides cr+s . Since p was arbitrary, f g is primitive. 
Note that we can speak about polynomials being irreducible in R[x] for any integral domain
R, since we have a notion of irreducible element in such a ring. For a field F, this coincides with
the usual notion of an irreducible polynomial.
P ROPOSITION 5.2.9. Let R be an integral domain, and let F = Q(R).
a. If f ∈ R[x] is a primitive polynomial that is irreducible as an element of F[x], then f is
irreducible in R[x]. In particular, if f is primitive and cannot be written as a product of two
nonconstant polynomials in R[x], then it is irreducible in R[x].
b. Suppose that R is a UFD. If f ∈ R[x] is irreducible, then it is irreducible as an element of
F[x] as well. In fact, if f ∈ R[x] and f = gh for nonconstant g, h ∈ F[x], then there exists α ∈ F ×
such that g0 = αg and h0 = α −1 h are in R[x] and therefore f = g0 h0 in R[x].
P ROOF. First, we treat part a. If f ∈ R[x] is primitive and reducible (which is to say, not
irreducible and not a unit or zero), then we can write f = gh for nonunits g, h ∈ R[x]. If g or h is
constant, then f is not primitive, so neither is constant, and therefore f is reducible in F[x].
Next, we turn to part b. Suppose that f ∈ R[x] can be written as f = gh with g, h ∈ F[x]
nonconstant. Let (d) (resp., (e)) be a multiple of all of the denominators of the coefficients of
g (resp., h), written in lowest terms. Then de f = g0 h0 , where g0 , h0 ∈ R[x] are nonconstant. The
content of de f is contained in (de), so the content of g0 h0 is as well. By unique factorization in
R, we may write de = d 0 e0 , where d 0 ∈ R divides the content of g0 and e0 divides the content of
h0 , and we may then divide g0 by d 0 and h0 by e0 to obtain g00 and h00 in R[x] such that f = g00 h00 .
Therefore, f is reducible in R[x], and the remaining statement of the lemma holds as well. 
We are now ready to prove the following.
T HEOREM 5.2.10. If R is a UFD, then R[x] is a UFD as well.
P ROOF. Let f ∈ R[x] be a nonzero element that is not a unit. Write
f = f1 f2 . . . fr
with fi ∈ R[x] nonconstant, where r is maximal such that this can be done. Note that such a
maximal r exists as the degree of f is finite. For 1 ≤ i ≤ r, let (ci ) be the content of fi , and
define gi ∈ R[x] by fi = ci gi . Set c = c1 c2 · · · cr , and set g = g1 g2 · · · gr . Now, if any gi were not
irreducible in F[x] for F = Q(R), then it would not be irreducible in R[x] by Proposition 5.2.9b.
Moreover, since gi is primitive, it would then be written as a product of two nonconstant poly-
nomials in R[x], which would contradict the maximality of r. Therefore, each gi is irreducible.
148 5. ADVANCED RING THEORY

Since R is a UFD, we may also write c = p1 p2 · · · pk with pi ∈ R irreducible for 1 ≤ i ≤ k and

some k ≥ 0, and so
f = p1 p2 · · · pk g1 g2 · · · gr
is a factorization of f into irreducibles in R[x].
Now, if
f = q1 q2 · · · ql h1 h2 · · · hs
with qi ∈ R irreducible and hi ∈ R[x] irreducible and nonconstant, then (q1 q2 · · · ql ) is the content
of f by Gauss’s lemma, and so q1 q2 · · · ql agrees with c up to unit in R. Since R is a UFD, it
follows that l = k and there exists σ ∈ Sk such that each qσ (i) is an associate of pi . Next, we have
g1 g2 · · · gr = uh1 h2 · · · hs
for some unit u ∈ R× , and by uniqueness of factorization in F[x], we have that s = r, and there
exists τ ∈ Sr such that hτ(i) = vi gi for some vi ∈ F × for each 1 ≤ i ≤ r. But the content of each gi
and each h j is (1), since these elements are irreducible in R[x], and therefore writing vi = abii with
ai , bi ∈ R, the fact that bi hτ(i) = ai gi implies that (ai ) = (bi ), since both sides must have the same
content. In other words, vi ∈ R× , and so hτ(i) and gi are associates in R[x], finishing the proof of
uniqueness. 
E XAMPLES 5.2.11.
a. Since Z is a UFD, so is Z[x]. However, Z[x] is not a PID, since (p, x) is not principal.
b. Since Q[x] is a UFD, so is Q[x, y]. Again, Q[x, y] is not a PID, since (x, y) is not principal.
c. If R is any UFD, then R[x1 , x2 , · · · xn ] is a UFD for any n ≥ 1.
We next prove that a polynomial ring over a noetherian commutative ring is noetherian.
T HEOREM 5.2.12 (Hilbert’s basis theorem). The polynomial ring R[x] over a commutative
noetherian ring R is noetherian.
P ROOF. Let I be an ideal of R[x]. We must show that I is finitely generated. Let L be the set
the leading coefficients of the elements of I. Then L is clearly an ideal: if a ∈ L is the leading
coefficient of f ∈ I and r ∈ R, then ra is the leading coefficient of r f ∈ I, and if a, b ∈ L are
the leading coefficients of f and g, respectively, then xdeg g f + xdeg f g ∈ I has leading coefficient
a + b. Since R is noetherian, there exist a1 , a2 , . . . , ak ∈ L such that L = (a1 , a2 , . . . , ak ). Let fi ∈ I
of degree ni ≥ 0 have leading coefficient ai for 1 ≤ i ≤ k. Let n = max{ni | 1 ≤ i ≤ k}.
Next, for m ≥ 0, let Lm be the set of all leading coefficients of polynomials in I of degree m.
This, again, is clearly an ideal of R, so we have Lm = (bm,1 , bm,2 , . . . , bm,lm ) for some lm ≥ 1 and
bm,i ∈ Lm for 1 ≤ i ≤ lm . For each such i, let gm,i ∈ I be a polynomial of degree m with leading
coefficient bm,i . We claim that I is generated by
n
[
X = { fi | 1 ≤ i ≤ k} ∪ {gm,i | 1 ≤ i ≤ lm }.
m=0
Let J be the ideal of R[x] generated by X, which is contained in I. Let h ∈ I, and let c ∈ L be
its leading coefficient. We want to show that h ∈ J. Write c = ∑ki=1 ri ai with ri ∈ R. If d ≥ n, then
 5.3. IRREDUCIBILITY OF POLYNOMIALS 149

c is the leading coefficient of

k
h0 = ∑ ri xd−ni fi ∈ J,
i=1
so h − h0 has degree less than d. We can then replace h by h − h0 and repeat the process until
d < n.
We are reduced to showing that if h ∈ I has degree d < n and leading coefficient c, then h ∈ J.
In this case, we have c = ∑li=1
d
si bd,i with si ∈ S, and c is the leading coefficient of
ld
0
h = ∑ si gd,i ∈ J
i=1
Then h − h0∈ J has degree less than d. Replacing h − h0 by h and repeating the process, we see
that h ∈ J. 
C OROLLARY 5.2.13. Let R be a noetherian ring. Then R[x1 , x2 , . . . , xn ] is noetherian for
every n ≥ 1.

5.3. Irreducibility of polynomials

In this section, we investigate criteria for determining if a polynomial is irreducible or not.
D EFINITION 5.3.1. Let R be an integral domain, and let p be a prime ideal of R. We say that
a nonconstant polynomial f = ∑ni=0 ai xi in R[x] is an Eisenstein polynomial (with respect to p) if
an ∈ / p2 .
/ p, ai ∈ p for all 0 ≤ i ≤ n − 1, and a0 ∈
T HEOREM 5.3.2 (Eistenstein criterion). Let R be an integral domain, and let f ∈ R[x] be an
Eiseinstein polynomial.
a. If R is a UFD, then f is irreducible in Q(R)[x].
b. If f is primitive, then it is irreducible in R[x].
P ROOF. Suppose f = ∑ni=0 ai xi is of degree n and Eisenstein with respect to a prime ideal
p of R. By Proposition 5.2.9, it suffices for each part to show that f is not a product of two
nonconstant polynomials in R[x]. So, let g = ∑si=0 bi xi and h = ∑tj=0 c j x j be polynomials in R[x]
with f = gh, where s + t = n. We then have
k
ak = ∑ bi ck−i
i=0

for all 0 ≤ k ≤ n. In particular, a0 = b0 c0 is an element of p but not p2 . Since p is prime, at least

/ p2 , at least one does not lie in p as well.
one of b0 and c0 lies in p, but as a0 ∈
Without loss of generality, suppose that b0 ∈ p and c0 ∈ / p. As an = bs ct ∈
/ p, we have bs ∈/ p.
Let k ≥ 1 be minimal such that bk ∈ / p. If k < n, then ak ∈ p and bi ∈ p for i < k, so we have
bk c0 ∈ p, which therefore forces c0 ∈ p by the primality of p. Therefore, k = n, which means that
h is constant, proving the result. 
We will most commonly be concerned with the Eisenstein criterion in the case that R = Z.
150 5. ADVANCED RING THEORY

E XAMPLE 5.3.3. For any prime number p and integer n ≥ 1, the polynomial xn − p is irre-
ducible by the Eisenstein criterion. That is, we take our prime ideal to be (p) in the ring Z.
E XAMPLE 5.3.4. For a prime number p, set
xp − 1
Φp = = x p−1 + x p−2 + · · · + 1.
x−1
This polynomial has as its roots in C the distinct pth roots of unity that are not equal to 1. Over
Q, we claim it is irreducible. For this, consider the polynomial
(x + 1) p − 1 p−1
 
p
Φ p (x + 1) = = ∑ xi ,
x i=0 i + 1
which has coefficents divisible by p but not p2 except for its leading coefficient a p−1 , which is 1.
Therefore, Φ p (x + 1) is Eisenstein, hence irreducible. But if Φ p were to factor into g and h, then
Φ p (x + 1) would factor into g(x + 1) and h(x + 1), which have the same leading coefficients as g
and h, and hence are nonconstant if and only if g and h are. In other words, Φ p is irreducible as
well.
R EMARK 5.3.5. The condition in the Eisenstein criterion that the constant coefficient not lie
in the square of the prime ideal is in general necessary. For instance, x2 − p2 ∈ Z[x] is never
irreducible for a prime p.
Often, we can tell if a polynomial is irreducible by considering its reductions modulo ideals.
P ROPOSITION 5.3.6. Let R be an integral domain, and let p be a prime ideal of R. Let
f ∈ R[x] with leading coefficient not in p. Let f¯ denote the image of f in (R/p)[x] given by
reducing its coefficients modulo p.
a. If R is a UFD and f¯ is irreducible in Q(R/p)[x], then f is irreducible in Q(R)[x].
b. If f is primitive and f¯ is irreducible in R/p[x], then f is irreducible in R[x].
P ROOF. If R is a UFD and f is reducible in Q(R)[x], then by Proposition 5.2.9, we have
that f = gh for some nonconstant g, h ∈ R[x]. Similarly, if f is primitive and reducible in R[x],
then f = gh for nonconstant g, h ∈ R[x]. In either case, since the leading coefficient of f is not
in p and p is prime, we have that the leading coefficients of g and h are not in p as well. That
is, the images of g and h in (R/p)[x] are nonconstant, which means that f¯ is a product of two
nonconstant polynomials, hence reducible in Q(R/p)[x]. 
R EMARK 5.3.7. For R = Z, Proposition 5.3.6 tells us in particular that if f ∈ Z[x] is monic
and its reduction f¯ ∈ F p [x] modulo p is irreducible for any prime p, then f is irreducible.
E XAMPLE 5.3.8. Let f = x4 + x3 + 1001 ∈ Z[x]. We claim that f is irreducible in Q[x]. For
this, consider its reduction modulo 2. The polynomial f¯ = x4 + x3 + 1 ∈ (Z/2Z)[x] is either
irreducible, has a root in (Z/2Z)[x], or is a product of two irreducible polynomials of degree 2.
But f¯(0) = f¯(1) = 1, and x2 + x + 1 is the only irreducible polynomial of degree 2 in (Z/2Z)[x],
and (x2 + x + 1)2 = x4 + x2 + 1 6= f¯, so f¯ is irreducible. By Proposition 5.3.6, f is irreducible in
Q[x].
 5.4. EUCLIDEAN DOMAINS 151

E XAMPLE 5.3.9. The converse to Proposition 5.3.6 does not hold. For instance, x2 + x + 1 is
irreducible in Q[x], but it has a root in (Z/3Z)[x].
We also have the following simple test for the existence of roots of polynomials over UFDs.
P ROPOSITION 5.3.10. Let R be a UFD and f = ∑ni=0 ai xi ∈ R[x] with a0 , an 6= 0. Suppose
that α ∈ Q(R) is a root of f , and write α in reduced form as α = dc for some c, d ∈ R. Then c
divides a0 and d divides an in R.
P ROOF. Since x − dc divides f in Q(R)[x] and dc is in reduced form, it follows from Proposi-
tion 5.2.9 that f = (dx − c)g for some g ∈ R[x]. Writing g = ∑n−1 i
i=0 bi x , we see that a0 = −cb0
and an = dbn−1 . 
E XAMPLE 5.3.11. Let f = 2x3 − 3x + 5 ∈ Z[x]. We check that f (1) = 4, f (−1) = 6, f (5) ≡
−10 mod 25, f (−5) ≡ 20 mod 25, and f ( 21 ), f (− 21 ), f ( 52 ), and f (− 52 ) are all represented by
reduced fractions with denominators equal to 4. Proposition 5.3.10 therefore tells us that f has
no roots in Q, hence is irreducible, being of degree 3.

5.4. Euclidean domains

D EFINITION 5.4.1. A norm f on an ring R is a function f : R → Z≥0 with f (0) = 0. We say
that f is positive if the only a ∈ R for which f (a) = 0 is a = 0.
D EFINITION 5.4.2. Let R be an integral domain. A Euclidean norm ν on R is a norm on R
such that for all nonzero a, b ∈ R, one has
i. ν(a) ≤ ν(ab), and
ii. there exist q, r ∈ R with a = qb + r and either ν(r) < ν(b) or r = 0.
R EMARK 5.4.3. Property (ii) of Definition 5.4.2 is known as the division algorithm.
D EFINITION 5.4.4. A Euclidean domain R is an integral domain such that there exists a
Euclidean norm on R.
E XAMPLES 5.4.5.
a. The integers Z are a Euclidean domain with Euclidean norm ν(a) = |a| for any nonzero
a ∈ Z.
b. Every polynomial ring F[x] over a field F is a Euclidean domain, the degree function
providing a Euclidean norm on F[x] (if we take the degree of 0 to be 0). Note that this is not a
positive norm.
L EMMA 5.4.6. In a Euclidean domain R with Euclidean norm ν, the minimal value of ν on
all nonzero elements of R is ν(1), and ν(u) = ν(1) for u ∈ R if and only if u ∈ R× .
P ROOF. By the definition of a Euclidean norm, we have ν(1) ≤ ν(a·1) = ν(a) for all nonzero
a ∈ R. If u ∈ R× , then ν(u) ≤ ν(u · u−1 ) = ν(1), so ν(u) = ν(1). Conversely, if b ∈ R with
ν(b) = ν(1), then we may write 1 = qb + r for some q, r ∈ R with either ν(r) < ν(1) or r = 0.
By what we have shown, the latter holds, so qb = 1, and b is a unit. 
152 5. ADVANCED RING THEORY

E XAMPLE 5.4.7. In F[x], the units are exactly the nonzero constant polynomials, i.e., those
with degree 0.
While we will explain below that not every PID is a Euclidean domain, it is the case that
every Euclidean domain is a PID.
T HEOREM 5.4.8. Every Euclidean domain is a PID.
P ROOF. Let I be a nonzero ideal in a Euclidean domain R with Euclidean norm ν. We must
show that I is principal. Let b ∈ I be a nonzero element with minimal norm among all elements
of I. For any a ∈ I, we may write a = qb + r with q, r ∈ R and either ν(r) < ν(b) or r = 0.
Note that a, b ∈ I, so r ∈ I as well, which precludes the possibility of ν(r) < ν(b), since ν(r) is
minimal among norms of elements of I. Therefore, we have r = 0, so a ∈ (b). As a was arbitrary
and b ∈ I, we have I = (b). 
The key property of Euclidean domains is the ability to perform the Euclidean algorithm,
which we see in the following.
T HEOREM 5.4.9 (Euclidean algorithm). Let R be a Euclidean domain with Euclidean norm
ν, and let a, b ∈ R be nonzero elements. Let r−1 = a and r0 = b. Suppose recursively that we are
given elements r j ∈ R for −1 ≤ j ≤ i and some i ≥ 0. If ri 6= 0, write
(5.4.1) ri−1 = qi+1 ri + ri+1
with qi+1 , ri+1 ∈ R and either ν(ri+1 ) < ν(ri ) or ri+1 = 0. If ri+1 6= 0, repeat the process with i
replaced by i + 1. The process terminates with d = rn 6= 0 and rn+1 = 0 for some n ≥ 1, and (d)
is the GCD of a and b. Moreover, we may use the formulas in (5.4.1) and recursion to write d as
d = xa + yb for some x, y ∈ R.
P ROOF. We note that the process must terminate, as the values of the ν(ri ) for i ≥ 0 are
decreasing. Moreover, the result d = rn satisfies rn−1 = qn+1 rn , so it divides rn−1 by definition,
and then we see by downward recursion using (5.4.1) that d divides every ri−1 . Finally, if c is
any common divisor of a and b, then it again recursively divides each ri (this time by upwards
recursion and (5.4.1)), so c divides d. Therefore, (d) is the GCD of a and b.
Note that d = rn−2 − qn rn−1 , and suppose that we may write d = zr j + wr j+1 for some −1 ≤
j ≤ n − 2. If j = −1, we are done. Otherwise, note that r j+1 = r j−1 − q j+1 r j , so
d = zr j + w(r j−1 − q j+1 r j ) = wr j−1 + (z − q j+1 w)r j ,
and we have written d as an R-linear combination of r j−1 and r j . Repeat the process for j − 1.
The final result is the desired R-linear combintation of a and b. 
E XAMPLE 5.4.10. Take Z and its usual Euclidean norm. We take a = 550 and b = 154. Then
550 = 3 · 154 + 88, so we set r1 = 88. Then 154 = 88 + 66, so we set r2 = 66, and 88 = 66 + 22,
so we set r3 = 22, and 66 = 3 · 22, so we stop at d = r3 = 22, which is therefore the greatest
common divisor of a and b. Working backwards, we obtain
22 = 88 − 66 = 88 − (154 − 88) = 2 · 88 − 154 = 2 · (550 − 3 · 154) − 154 = 2 · 550 − 7 · 154.
That is, we have written d as 2a − 7b.
 5.4. EUCLIDEAN DOMAINS 153

Often Euclidean norms come in the form of multiplicative norms.

D EFINITION 5.4.11. A multiplicative norm N : R → Z≥0 on a commutative ring R is a posi-
tive norm such that for all N(ab) = N(a)N(b) for all a, b ∈ R.
R EMARK 5.4.12. Note that the existence of a multiplicative norm N on a commutative ring R
forces R to be an integral domain, for if ab = 0, then N(a)N(b) = N(ab) = 0, so either N(a) = 0
or N(b) = 0, and therefore either a = 0 or b = 0.
E XAMPLE 5.4.13. The absolute value on Z is a multiplicative norm, as well as a Euclidean
norm.
E XAMPLE 5.4.14. The function N on the Gaussian integers Z[i] given by N(a + bi) = a2 + b2
is a multiplicative norm. Clearly, a2 + b2 = 0 if and only if a + bi = 0. Given a, b, c, d ∈ Z, we
have
N((a + bi)(c + di)) = (ac − bd)2 + (ad + bc)2 = (ac)2 + (bd)2 + (ad)2 + (bc)2
= (a2 + b2 )(c2 + d 2 ) = N(a + bi)N(c + di).
P ROPOSITION 5.4.15. The ring Z[i] of Gaussian integers is a Euclidean domain with respect
to the Euclidean norm N(a + bi) = a2 + b2 for a, b ∈ Z.
P ROOF. Since N is a multiplicative norm, we need only check the division algorithm. Extend
N to a function on C by defining N(a + bi) = a2 + b2 for a, b ∈ R. Let a, b, c, d ∈ Z with (c, d) 6=
(0, 0). Then we have
a + bi
= s + ti
c + di
for some s,t ∈ Q, and let e, f ∈ Z be integers with |s − e| ≤ 1/2 and |t − f | ≤ 1/2. Then we have
N(a + bi − (e + f i)(c + di)) = N(c + di)N((s − e) + (t − f )i)
 2  2 !
1 1
≤ N(c + di) + = N(c + di)/2 < N(c + di),
2 2
so the division algorithm is satisfied: a + bi = q(c + di) + r with q = e + f i and N(r) < N(c +
di). 
C OROLLARY 5.4.16. The units in Z[i] are exactly 1, −1, i, −i.
P ROOF. Since N is a Euclidean norm on Z[i], the units are exactly those nonzero elements of
norm N(1) = 1. We have a2 + b2 = 1 if and only if (a, b) = (±1, 0) or (a, b) = (0, ±1). 
L EMMA 5.4.17. If a, b, c, d ∈ Z and c + di divides a + bi in Z[i], then c − di divides a − bi in
Z[i].
P ROOF. Write a + bi = (c + di)(e + f i) for some e, f ∈ Z. Then a = ce − d f and b = cd + de,
so
(c − di)(e − f i) = (ce − d f ) − (c f + de)i = a − bi.

154 5. ADVANCED RING THEORY

We can completely determine the irreducible elements in Z[i] as follows.

P ROPOSITION 5.4.18. The irreducible elements in Z[i] are, up to multiplication by a unit,
1 + i, primes p ∈ Z with p ≡ 3 mod 4, and a + bi for a, b ∈ Z such that p = a2 + b2 ≡ 1 mod 4
is a prime in Z. Moreover, the primes in Z that can be written in the form a2 + b2 are exactly 2
and those that are 1 modulo 4.
P ROOF. First, note that if a + bi divides c + di in Z[i] for integers a, b, c, d, then N(a + bi)
divides N(c + di), since N is multiplicative. So, 1 + i is irreducible since N(1 + i) = 2.
Let p be an odd prime in Z. If p is divisible by some irreducible element π = a + bi with
a, b ∈ Z, then since p is prime, only one of two things can happen. Either ab = 0, or a and b
are relatively prime in Z, noting Corollary 5.4.16. Suppose ab 6= 0. By Lemma 5.4.17, we have
that a − bi divides p, and π̄ = a − bi is irreducible. If π̄ were associate to π, then π would divide
2a = (a + bi) + (a − bi) and 2b = −i((a + bi) − (a − bi)). Then π divides 2, but that is impossible.
Thus, π and π̄ both dividing p implies that p is divisible by N(π) = a2 + b2 . As p is prime, we
have p = a2 + b2 .
So, we have shown that either our odd prime p is irreducible in Z[i] or p = a2 + b2 for some
a, b ∈ Z. Note that the squares in Z/4Z are 0 and 1, so any integer of the form a2 + b2 is 0, 1, or
2 modulo 4. In particular, if p ≡ 3 mod 4, then p is irreducible in Z[i].
If p ≡ 1 mod 4 is prime in Z, then (Z/pZ)× has order divisible by 4. As Z/pZ contains
only two roots of x2 − 1, which are −1 and 1, so (Z/pZ)× contains an element of order 4. In
particular, there exists n ∈ Z such that n2 ≡ −1 mod p, which is to say that p divides n2 + 1. If p
were irreducible in Z[i], then p would divide either n + i or n − i, but then it would divide both,
being an integer. Thus p would divide 2i, which it does not. So, p is reducible, which means
equals a2 + b2 for some a, b ∈ Z. 
L EMMA 5.4.19. Let N be a multiplicative norm on an integral domain R. Then N(u) = 1 for
all u ∈ R× .
P ROOF. We have N(1) = N(1)2 , and R is an integral domain, so N(1) = 1. Moreover, since
N(u−1 )N(u) = N(1) = 1,
we have that N(u−1 ) = N(u)−1 , and therefore N(u) = 1. 
√
E XAMPLE 5.4.20. Consider the multiplicative norm N on Z[ −5] given by
√
N(a + b −5) = a2 + 5b2 .
√
We have a2 + 5b2 = 1 if and only if a = ±1√and b = 0, so the only units in Z[ −5] are ±1.
Now, if 2 = αβ for some nonunits α, β ∈ Z[ −5], then 4 = N(2) = N(α)N(β ), so N(α) = 2,
but 2 is√clearly not a value of N. Therefore, 2 is irreducible, and so is 3.√ Also, we have that
N(1 ± −5) = 6, and since 2 and 3 are not values of N, we have that 1 ± −5 is irreducible as
well. As these elements are all non-associates, the existence of the two factorizations
√ √
6 = 2 · 3 = (1 + −5)(1 − −5)
√
proves that Z[ −5] is not a UFD.
 5.4. EUCLIDEAN DOMAINS 155

Not all principal ideal domains are Euclidean. We give most of the outline of how one
produces an example.
D EFINITION 5.4.21. An nonzero, non-unit element b of an integral domain R is called a
universal side divisor if every element a ∈ R may be written in the form a = qb + r for some
q, r ∈ R with r = 0 or r ∈ R× .
L EMMA 5.4.22. Let R be a Euclidean domain with Euclidean norm ν. Let b ∈ R be a nonzero,
non-unit element such that ν(b) is minimal among nonzero, non-unit elements of R. Then b is a
universal side divisor of R.
P ROOF. Let a ∈ R. By definition of ν, we may write a = qb + r with ν(r) < ν(b) or r = 0.
By the minimality of ν(b), we must have that r is a unit or 0. 
√
E XAMPLE 5.4.23. We claim that the ring R = Z[(1 + −19)/2] is not Euclidean. Suppose
by contradiction that it is a Euclidean domain, and let ν be a Euclidean norm on R. We also have
the multiplicative norm N on R given by
√ ! √ ! √ !
1 + −19 1 + −19 1 − −19
(5.4.2) N a + b = a+b a+b = a2 + ab + 5b2 .
2 2 2
Note that if α ∈ R − Z, then N(α) ≥ 5, so the only units in R are ±1.
Let β ∈ R be a universal side divisor, which exists as R is Euclidean, and write 2 = qβ + r
for q ∈ R and r ∈ {0, 1, −1}. We then have that N(β ) divides N(2 − r) as N is multiplicative, so
N(β ) divides 4 or 9, and this implies β ∈ {±2, ±3} by the formula for N. Now take α = (1 +
√
−19)/2, and set α = q0 β + r0 with q0 ∈ R and r0 ∈ {0, 1, −1}. We have N(α) = N(α − 1) = 5
and N(α + 1) = 7, which are not multiples of N(β ) ∈ {4, 9}, so we obtain a contradiction.
D EFINITION 5.4.24. A Dedekind-Hasse norm on an integral domain R is a positive norm µ
on R such that for every a, b ∈ R, either a ∈ (b) or there exists a nonzero element c ∈ (a, b) such
that µ(c) < µ(b).
P ROPOSITION 5.4.25. An integral domain R is a PID if and only if there exists a Dedekind-
Hasse norm on R.
P ROOF. Suppose first that µ is a Dedekind-Hasse norm on R. Let I be a nonzero ideal of R,
and let b ∈ I − {0} with minimal norm under µ. If a ∈ I, then since there does not exist a nonzero
element c ∈ (a, b) ⊆ I with µ(c) < µ(b) by the minimality of µ(b), we have by definition of a
Dedekind-Hasse norm that a ∈ (b). Thus I = (b).
Suppose on the other hand the R is a PID. Define µ : R → Z≥0 by µ(0) = 0, µ(u) = 1 for
u ∈ R× , and µ(p1 p2 · · · pk ) = 2k if p1 , . . . , pk are irreducible elements of R. This is well-defined
as R is a UFD. Given a, b ∈ R, we have (a, b) = (d) for some d ∈ R, since R is a PID. Since d
divides b, we have µ(d) ≤ µ(b). If µ(d) = µ(b), then a and b have the same number of divisors
as d and therefore are associates, so a ∈ (b). Thus, µ is a Dedekind-Hasse norm. 
√
E XAMPLE 5.4.26. We have already seen that R = Z[(1 + −19)/2] is not a Euclidean do-
main. To see that R is a PID, it suffices to show that the multiplicative norm N on R given by
(5.4.2) is a Dedekind-Hasse norm on R. We outline the standard unenlightening verification.
156 5. ADVANCED RING THEORY

Let α, β ∈ R with α ∈ / (β ). We claim that there exist s,t ∈ R with 0 < N(sα − tβ ) < N(β ).
Note that we can extend N to a map N : Q(R) → Z≥0 by the formula (5.4.2), allowing a, b ∈ Q.
Our condition that N on R be a Dedekind-Hasse norm is then that 0 < N(s αβ − t) < 1. We will
find s and t. For this, write √
α a + b −19
=
β c
for a, b, c ∈ Z with no common divisor and c > 1.
√ the cases with c ≤ 4. If2c = 2, 2then either a or b 2is odd, 2then take s = 1
First one considers
and t = ((a − 1) + b −19)/2. √ If c = 3, then a + 19b 6≡ 0 mod 3, so a + 19b = 3q + r with
r ∈ {1, 2}. Take s = a − b −19 and t = q. If c = 4, then again either √ a or b is odd. If only one
is, then write a2 + 19b2 = 4q + r with 1 ≤ r ≤ 4, √ and take s = a − b −19 and t = q. If both are,
2 2 1
write a + 19b = 8q + 4, and take s = 2 (a − b −19) and t = q.
Now suppose that c ≥ 5. Since (a, b, c) = (1), we have x, y, z ∈ Z√such that xa + yb + √ zc = 1.
Write ay − 19bx = qc + r, with q ∈ Z and |r| ≤ c/2. Take s = y + x −19 and t = q − z −19.
The reader will check that
√  r2 + 19
 
α 
N s − t = c−2 N s(a + b −19) − tc = ,
β c2
which is at most 41 + 19
36 =
7
9 if c ≥ 6 and at most 4
25 + 19
25 =
23
25 if c = 5.

5.5. Vector spaces over fields

In this section, we give a very brief discussion of the theory of vector spaces over fields, as it
shall be subsumed by the sections that follow it.
D EFINITION 5.5.1. Let F be a field. A vector space V over F is an abelian group under
addition that is endowed with an operation · : F × V → V of scalar multiplication such that for
all a, b ∈ F and v, w ∈ V , one has
i. 1 · v = v,
ii. a · (b · v) = (ab) · v,
iii. (a + b) · v = a · v + b · v,
iv. a · (v + w) = a · v + a · w.
R EMARK 5.5.2. In a vector space V over a field F, we typically write av for a · v, where a ∈ F
and v ∈ V .
E XAMPLE 5.5.3. If F is a field, then F n is a vector space over F under the operation
a · (α1 , α2 , . . . , αn ) = (aα1 , aα2 , . . . , aαn )
for a, α1 , α2 , . . . , αn ∈ F.
D EFINITION 5.5.4. An element of a vector space V over a field F is called a vector, and the
elements of F under in the operation · are referred to as scalars.
 5.5. VECTOR SPACES OVER FIELDS 157

E XAMPLE 5.5.5. In every vector space V , there is an element 0, and it is called the zero
vector.
D EFINITION 5.5.6. The zero vector space 0 is the vector space over any field F that is the set
{0} with the operation a · 0 = 0 for all a ∈ F.
E XAMPLE 5.5.7. If F is a field, then F[x] is a vector space over F with a · f for a ∈ F and
f ∈ F[x] defined to be the usual product of polynomials in F[x]. I.e., the operation of scalar
multiplication is just multiplication by a constant polynomial.
E XAMPLE 5.5.8. The field C is an R-vector space, as well as a Q-vector space. The field
R is a Q-vector space. The operations of scalar multiplication are just restrictions of the usual
multiplication map on C.
The reader will easily check the following.
L EMMA 5.5.9. If V is a vector space over a field F, then for a ∈ F and v ∈ V , we have
a. 0 · v = 0,
b. a · 0 = 0,
c. −(av) = (−a)v = a(−v).
D EFINITION 5.5.10. Let V be a vector space over a field F. A subspace W of V is a subset that
is closed under the operations of addition and scalar multiplication to W (i.e., to maps W ×W → V
and F ×W → V , respectively) and is a vector space with respect to these operations.
The following is easily proven.
L EMMA 5.5.11. A subset W of a vector space V is a subspace if and only if it is a subgroup
under addition and closed under scalar multiplication.
E XAMPLES 5.5.12.
a. The zero subspace {0} and V are both subspaces of any vector space V .
b. The field F is a subspace of F[x].
D EFINITION 5.5.13. Let V be a vector space over a field F, and let S be a subset of V . A
linear combination of elements of S is any sum
n
∑ aivi
i=1
with v1 , v2 , . . . , vn distinct vectors in S and a1 , a2 , . . . , an ∈ F for some n ≥ 0. We say that such a
linear combination is nontrivial if there exists a j with 1 ≤ j ≤ n and a j 6= 0.
D EFINITION 5.5.14. Let V be a vector space over a field F and S be a set of vectors in V .
The subspace spanned by S, also known as the span of V , is the set of all linear combinations of
elements of S, or simply the zero subspace if S is empty.
E XAMPLE 5.5.15. For any vector space V , the set V spans V .
158 5. ADVANCED RING THEORY

D EFINITION 5.5.16. We say that a set S of vectors in a vector space V over a field F spans
V if V equals the subspace spanned by S.
That is, S spans an F-vector space V if, for every v ∈ V , there exist n ≥ 0, vi ∈ V , and ai ∈ F
for 1 ≤ i ≤ n such that
n
v = ∑ ai vi .
i=1
D EFINITION 5.5.17. We say that a set of S of vectors in a vector space V over a field F is
linearly independent if every nontrivial linear combination of vectors in S is nonzero. Otherwise,
S is said to be linearly dependent.
That is, a set S of vectors in an F-vector space V is linearly independent if whenever n ≥ 1,
vi ∈ V and ai ∈ F for 1 ≤ i ≤ n and
n
∑ aivi = 0,
i=1
then ai = 0 for all 1 ≤ i ≤ n.
L EMMA 5.5.18. Let S be a linearly independent subset of a vector space V over a field F,
and let W be the span of F. If v0 ∈ V −W , then S ∪ {v0 } is also linearly independent.
P ROOF. Let v1 , v2 , . . . , vn ∈ S and c0 , c1 , . . . , cn ∈ F for some n ≥ 1, and suppose that
n
∑ civi = 0.
i=0
We cannot have c0 6= 0, as then
n
v0 = −c−1
0 ∑ ci vi ∈ W.
i=1
On the other hand, the fact that c0 = 0 implies that ci = 0 for all 1 ≤ i ≤ n by the linear indepen-
dence of V . Thus, S ∩ {v0 } is linearly independent. 
E XAMPLE 5.5.19. In any vector space V , the empty set is linearly independent. If v ∈ V is
nonzero, then {v} is also a linearly independent set.
D EFINITION 5.5.20. A subset B of a vector space V over a field F is said to be a basis of V
over F if it is linearly independent and spans V .
E XAMPLE 5.5.21. The set {e1 , e2 , . . . , en } of F n , where ei is the element of F n that has a 1 in
its ith coordinate and 0 in all others, is a basis of F n .
E XAMPLE 5.5.22. The set {xi | i ≥ 0} is a basis of F[x]. That is, every polynomial can be
written as a finite sum of distinct monomials in a unique way.
R EMARK 5.5.23. For a field F, it is very hard to write down a basis of ∏∞
i=0 F. In fact, the
proof that it has a basis uses the axiom of choice.
D EFINITION 5.5.24. A vector space V is said to be finite dimensional if it has a finite basis
(i.e., a basis with finitely many elements). Otherwise V is said to be infinite dimensional.
 5.5. VECTOR SPACES OVER FIELDS 159

The following theorem employs Zorn’s lemma.

T HEOREM 5.5.25. Let V be a vector space over a field F. Every linearly independent subset
of V is contained in a basis of V .
P ROOF. Let S be a linearly independent subset of V , and let X denote the set of linearly
independent subsets of VSthat contain S. We order X by containment of subsets. If C is a chain
in X, then its union U = T ∈C T is linearly independent since if v1 , v2 , . . . , vn ∈ U for some n ≥ 1,
then each vi is contained in some Ti ∈ X for each 1 ≤ i ≤ n, and one of the sets T j contains the
others, since C is a chain. Since T j is linearly independent, any nontrivial linear combination
of the elements vi with 1 ≤ i ≤ n is nonzero. Therefore, U is linearly independent as well, so is
contained in X.
By Zorn’s Lemma, X now contains a maximal element B, and we want to show that B spans
V , so is a basis of V containing S. Let W denote the span of B. If v ∈ V −W , then B0 = B ∪ {v}
is linearly independent by Lemma 5.5.18, so an element of X, which contradicts the maximality
of B. That is, V = W , which is to say that B spans V . 

In particular, the empty set is contained in a basis of any vector space, so we have the follow-
ing:
C OROLLARY 5.5.26. Every vector space over a field contains a basis.
A similar argument yields the following.
T HEOREM 5.5.27. Let V be a vector space over a field F. Every subset of V that spans V
contains a basis of V .
P ROOF. Let S be a spanning subset of V . Let X denote the set of linearly independent subsets
of S, and order X by containment. As seen in the proof of Theorem 5.5.25, any union of a chain
of linearly independent subsets is linearly independent, so has an upper bound. Thus, Zorn’s
lemma tells us that X contains a maximal element B. Again, we want to show that B spans V , so
is a basis. If it were not, then there would exist some element of V which is not in the span of B,
but is in the span of S. In particular, there exists an element v ∈ S that is not in the span of B. The
set B ∪ {v} is linearly independent, contradicting the maximality of B. 

We also have the following, which can be generalized to a statement on cardinality.

T HEOREM 5.5.28. Let V be a vector space over a field F. If V is finite dimensional, then
every basis of V contains the same number of elements, and otherwise every basis of V is infinite.
P ROOF. Let B1 = {v1 , v2 , . . . , vn } be a basis of V with a minimal number n of elements, and
let B = {w1 , w2 , . . . , wm } be another basis of V with m ≥ n. Then B1 spans V , so w1 is a nontrivial
linear combination of the vi for 1 ≤ i ≤ n:
n
(5.5.1) w1 = ∑ ai vi
i=1
160 5. ADVANCED RING THEORY

for some ai ∈ F. Letting j be such that a j 6= 0, we may write v j as a linear combination of w1

and the vi with i 6= j. In other words, B2 = (B1 − {v j }) ∪ {w1 } spans V . Suppose
n
(5.5.2) c j w1 + ∑ ci vi = 0
i=1
i6= j

for some ci ∈ F. Using (5.5.1), we may rewrite the sum in (5.5.2) as a linear combination of the
vi , the coefficient of v j in which is a j c j , which forces c j = 0 as B1 is a linearly independent set.
But then we see from (5.5.2) that all ci = 0 as B − {v j } is linearly independent. So, B2 is a basis
of V .
Suppose by recursion that, for k ≤ m, we have found a basis Bk of order n of V that contains
only w1 , . . . , wk−1 and elements of B. Then wk is a nontrivial linear combination of the elements
of Bk , and the coefficient of some vl is nonzero in this linear combination by the linear indepen-
dence of B. We therefore have that Bk+1 = (Bk − {vl }) ∪ {wk } spans V , and a similar argument
to the above shows that it is a basis. Finally, we remark that the basis Bm+1 must be B1 itself,
since it contains B1 , so we have m = n, as desired. 
D EFINITION 5.5.29. The dimension of a finite-dimensional vector space V over a field F is
the number of elements in a basis of V over F. We write dimF (V ) for this dimension.
E XAMPLE 5.5.30. The space F n is of dimension n over F.
The maps between vector spaces that respect the natural operations on the spaces are called
linear transformations.
D EFINITION 5.5.31. A linear transformation T : V → W of F-vector spaces is a function
from V to W satisfying
T (v + v0 ) = T (v) + T (v0 ) and T (av) = aT (v)
for all a ∈ F and v, v0 ∈ V
R EMARK 5.5.32. In other words, a linear transformation is a homomorphism of the underly-
ing groups that “respects scalar multiplication.”
D EFINITION 5.5.33. A linear transformation T : V → W of F-vector spaces is an isomor-
phism of F-vector spaces if it is there exists an linear transformation T −1 : W → V that is inverse
to it.
Much as with group and ring homomorphisms, we have the following:
L EMMA 5.5.34. A linear transformation is an isomorphism if and only if it is a bijection.
E XAMPLES 5.5.35. Let V and W be F-vector spaces.
a. The identity map idV : V → V is an F-linear transformation (in fact, isomorphism).
b. The zero map 0 : V → W is an F-linear transformation.
 5.6. MODULES OVER RINGS 161

5.6. Modules over rings

D EFINITION 5.6.1. Let R be a ring. A left R-module, or left module over R, is an abelian
group M together with an operation · : R × M → M such that for all a, b ∈ R and m, n ∈ M, one
has
i. 1 · m = m,
ii. (a · b) · m = (ab) · m,
iii. (a + b) · m = a · m + b · n,
iv. a · (m + n) = a · m + a · n.
D EFINITION 5.6.2. Let R be a commutative ring. We refer more simply to a left R-module as
a R-module, or module over R.
R EMARK 5.6.3. When one speaks simply of a module over a ring R, one means by default a
left R-module.
N OTATION 5.6.4. When an abelian group M is seen as a left module over a ring R via the
extra data of some operation R×M → M, we say that this operation endows M with the additional
structure of a left R-module.
E XAMPLE 5.6.5. The definition of a module over a field coincides with the definition of a
vector space over a field. In other words, to say that M a module over a field F is exactly to say
that M is a vector space over F.
E XAMPLE 5.6.6. The modules over Z are exactly the abelian groups. That is, suppose that A
is a Z-module, which by definition is an abelian group with an additional operation · : Z × A → A.
We show that this additional operation satisfies n · a = na for n ∈ Z and a ∈ A, where na is the
usual element of the abelian group A. So, let a ∈ A. By axiom (i), we have 1 · a = a, and then the
distributivity of axiom (iii) allows us to see that n · a = na for all n ≥ 1. Using axioms (iv) and
(ii), we have
0 · a = 0 · (2a − a) = 0 · 2a − 0 · a = (0 · 2) · a − 0 · a = 0 · a − 0 · a = 0,
and then finally we have
(−n) · a + n · a = (n − n) · a = 0 · a = 0,
so (−n) · a = −na for n ≥ 1.
E XAMPLE 5.6.7. For a ring R and n ≥ 1, the direct product Rn is a left Mn (R)-module via
matrix multiplication (A, v) 7→ A · v for A ∈ Mn (R) and v ∈ Rn ,viewing elements of Rn as column
vectors.
We also have the notion of a right R-module.
D EFINITION 5.6.8. Let R be a ring. A right R-module, or right module over R, is an abelian
group M together with an operation · : M × R → R such that for all a, b ∈ R and m, n ∈ M, one
has
i. m · 1 = m,
162 5. ADVANCED RING THEORY

ii. m · (a · b) = m · (ab),
iii. m · (a + b) = m · a + n · b,
iv. (m + n) · a = m · a + n · a.
E XAMPLE 5.6.9. Every left ideal I over a ring R is a left R-module with respect to the restric-
tion R × I → I of the multiplication on R. Every right ideal over R is a right module with respect
to the restriction I × R → I of the multiplication on R.
D EFINITION 5.6.10. Let R be a ring. The opposite ring Rop to R is the ring that is the abelian
group R together with the multiplication ·op : R × R → R given by a ·op b = ba, where the latter
product is taken in R.
R EMARK 5.6.11. The identity map induces an isomorphism R → (Rop )op of rings.
The reader will easily check the following.
L EMMA 5.6.12. A right module M over R also has the structure of a left module over Rop ,
where the latter operation ·op : Rop : M → M is given by a ·op m = ma, where the latter product is
that given by the right R-module structure of M.
E XAMPLE 5.6.13. For a field F, the map T : Mn (F) → Mn (F) given by transpose (that is,
A 7→ AT for A ∈ Mn (F)) is a ring isomorphism between Mn (F) and Mn (F)op .
We also have the notion of a bimodule.
D EFINITION 5.6.14. Let R and S be rings. An abelian group M that is a left R-module and a
right S-module is called an R-S-bimodule if
(r · m) · s = r · (m · s)
for all r ∈ R, s ∈ S, and m ∈ M.
E XAMPLES 5.6.15.
a. Any left R-module M over a commutative ring R is an R-R-bimodule with respect to given
left operation and the (same) right operation m · r = rm for m ∈ M and r ∈ R.
b. A two-sided ideal of a ring R is an R-R-bimodule with respect to the operations given by
the usual multiplication on R.
c. For m, n ≥ 1, the abelian group Mmn (R) of m-by-n matrices with entries in R is an Mm (R)-
Mn (R)-bimodule for the operations of matrix multiplication.
Let us return our focus to R-modules, focusing on the case of left modules, as right modules
are just left modules over the opposite ring by Lemma 5.6.12.
D EFINITION 5.6.16. An R-submodule (or, submodule) N of a left module M over a ring R
is a subset of N that is closed under addition and the operation of left R-multiplication and is an
R-module with respect to their restrictions + : N × N → N and · : R × N → N to N.
L EMMA 5.6.17. Let R be a ring, M be a left R-module, and N be a subset of M. Then N is
an R-submodule of M if and only if it is nonempty, closed under addition, and closed under left
R-multiplication.
 5.6. MODULES OVER RINGS 163

P ROOF. Clearly, it suffices to check that if N is nonempty and closed under addition and
left R-multiplication, then it is an R-submodule. The condition of being closed under left R-
multiplication assures that 0 and inverses of elements of N lies in N, so N is an abelian group
under + on M. The axioms for N to be an R-module under · are clearly satisfied as they are
satisfied by elements of the larger set M. 
E XAMPLES 5.6.18.
a. The subspaces of a vector space V over a field F are exactly the F-submodules of V .
b. The subgroups of an abelian group are the Z-submodules of that group.
c. Any left ideal I of R is a left R-submodule of R viewed as a left R-module.
d. Any intersection of R-submodules is an R-submodule as well.
e. For an R-module M and a left ideal I, the abelian group
nn o
IM = ∑ ai mi | ai ∈ I, mi ∈ M for 1 ≤ i ≤ n
i=1
is an R-submodule of M.
We also have the following construction.
D EFINITION 5.6.19. Let M be an R-module and {Ni | i ∈ I} be a collection of submodules
for an indexing set I. The sum of the submodules Ni is the submodule ∑i∈I Ni of M with elements
∑i∈I ni for ni ∈ Ni and all but finitely many ni equal to 0.
If M is an R-module and N is a submodule, we may speak of the quotient abelian group
M/N. It is an R-module under the action r · (m + N) = rm + N for r ∈ R and m ∈ M. This
is well-defined, as a different representative m + n of the coset m + N for n ∈ N will satisfy
r(m + n) + N = rm + rn + N = rm + N.
D EFINITION 5.6.20. Let M be a left R-module and N be an R-submodule of M. The quotient
module M/N of M by N is the abelian group of cosets together with the multiplication R ×
M/N → M/N given by r · (nN) = (rn)N.
E XAMPLE 5.6.21. For an R-module M and a left ideal I, we have the quotient module M/IM.
In particular, note that R/I is a left R-module with respect to r(s + I) = rs + I, even if it is not a
ring (i.e., if I is not two-sided).
We can also speak of homomorphisms of R-modules.
D EFINITION 5.6.22. Let M and N be left modules over a ring R. A left R-module homomor-
phism φ : M → N is a function such that φ (r · m) = rφ (m) and φ (m + n) = φ (m) + φ (n) for all
r ∈ R and m, n ∈ M.
N OTATION 5.6.23. If R is commutative (or it is understood that we are working with left
modules), we omit the word “left” and speak simply of R-module homomorphisms.
R EMARK 5.6.24. A right R-module homomorphism φ : M → N is just a left Rop -module
homomorphism.
164 5. ADVANCED RING THEORY

D EFINITION 5.6.25. Let M and N be left modules over a ring R.

a. An isomorphism f : M → N of left R-modules is a bijective homomorphism.
b. An endomorphism of a left R-module M is a homomorphism f : M → M of left R-modules.
c. An automorphism of a left R-modules M is an isomorphism f : M → M of left R-modules.
N OTATION 5.6.26. Sometimes, we refer to an R-module homomorphism as an R-linear map,
and an endomorphism of R-modules as an R-linear endomorphism.
E XAMPLES 5.6.27.
a. The zero map 0 : M → M and the identity map id : M → M are endomorphisms of an
R-module M, with id being an automorphism.
b. Let V and W be vector spaces over a field F. A left F-module homomorphism φ : V → W
is just an F-linear transformation.
c. Let N be an R-submodule of a left R-module M. The inclusion map ιN : N → M is an
R-module homomorphism, as is the quotient map πN : M → M/N.
d. If M is an R-S-bimodule, then right multiplication ψs : M → M by an element s ∈ S defines
a left R-module endomorphism. In particular, if R is a commutative ring, then multiplication by
r ∈ R defines an R-module endomorphism. Note that if R is noncommutative, then the condition
that left multiplication by r ∈ R be a left module homomorphism M → M is that r(sm) = s(rm)
for all r, s ∈ R and m ∈ M, which need not hold.
e. The identity map F n → F n provides an isomorphism between F n viewed as a left Mn (F)-
module via (A, v) 7→ Av for A ∈ Mn (F) and v ∈ F n (viewing F n as column vectors) and F n viewed
as a left Mn (F)-module via (A, v) 7→ vT AT (viewing F n as row vectors).
Note that we may speak of the kernel and the image of a left R-module, as an R-module ho-
momorphism is in particular a group homomorphism. The reader will easily verify the following.
L EMMA 5.6.28. Let φ : M → N be a left R-module homomorphism. Then ker φ and im φ are
R-submodules of M and N, respectively.
We also have analogues of all of the isomorphism theorems for groups. Actually, these
are virtually immediate consequences of said isomorphism theorems, as the fact that one has
isomorphisms of groups follows immediately from them, and then one need only note that these
isomorphisms are actually homomorphisms of R-modules.
T HEOREM 5.6.29. Let R be a ring. Let φ : M → N be an homomorphism of left R-modules.
Then there is an isomorphism φ̄ : M/ ker φ → im φ given by φ̄ (m + ker φ ) = φ (m).
T HEOREM 5.6.30. Let R be a ring, and let N be left R-submodules of an R-module M. Then
there is an isomorphism of R-modules
M/(M ∩ N) −∼→ (M + N)/N, m + (M ∩ N) 7→ m + N.
T HEOREM 5.6.31. Let R be a ring, let M be an R-module, and let Q ⊆ N be R-submodules
of M. Then there is an isomorphism
M/N −∼ → (M/Q)/(N/Q), m + N 7→ (m + Q) + (N/Q).
 5.7. FREE MODULES AND GENERATORS 165

We also have the following analogue of Theorems 2.13.10 and 3.7.24.

T HEOREM 5.6.32. Let R be a ring, let M be an R-module, and let N be an R-submodule
of M. Then the map P 7→ P/N gives a bijection between submodules P of M containing N and
submodules of M/N. This bijection has inverse Q 7→ πN−1 (Q) on submodules Q of M/N, where
πN : M → M/N is the quotient map.

5.7. Free modules and generators

D EFINITION 5.7.1. Let S be a subset of an R-module M.
a. The submodule of M generated by S is the intersection of all submodules of M containing
S.
b. We say that S generates M, or is a set of generators or generating set of M, if no proper
R-submodule of M contains S.
R EMARK 5.7.2. The R-submodule of M generated by S consists of the elements ∑ni=1 ai mi
with mi ∈ S and ai ∈ R for 1 ≤ i ≤ n and some n ≥ 1. The proof is much as before.
R EMARK 5.7.3. The sum ∑i∈I Ni of submodules Ni of M is the submodule generated by
∪i∈I Ni .
N OTATION 5.7.4. The R-submodule of an R-module M generated by for a single element
m ∈ M (or, more precisely, by {m}) is denoted R · m.
D EFINITION 5.7.5. We say that an R-module is finitely generated if it has a finite set of
generators.
D EFINITION 5.7.6. We say that an R-module is cyclic if it can be generated by a single
element.
E XAMPLE 5.7.7. A cyclic R-submodule of R is just a principal left ideal.
L EMMA 5.7.8. An R-module is cyclic if and only if it is isomorphic to R/I for some left ideal
I of R.
P ROOF. Let M be a cyclic R-module, generated by an element m. Let I = {r ∈ R | rm = 0}.
The surjective R-module homomorphism R → M sending r ∈ R to rm has kernel the left ideal I,
so by the first isomorphism theorem for modules, we have R/I ∼ = M. Conversely, if I is an left
ideal of R, then the quotient R-module R/I is generated by the coset of 1. 
We can define direct sums and direct products of modules.
D EFINITION 5.7.9. Let (Mi )i∈I be a collection of left modules over a ring R.
a. The direct product ∏i∈I Mi is the R-module that is the direct product of the abelian groups
Mi together with the left R-multiplication r · (mi )i∈I = (rmi )i∈I for r ∈ R and mi ∈ Mi for all i ∈ I.
L
b. The direct sum i∈I Mi is the R-module that is the direct sum of the abelian groups Mi
together with the left R-multiplication r · (mi )i∈I = (rmi )i∈I for r ∈ R and mi ∈ Mi for all i ∈ I
with all but finitely many mi = 0.
166 5. ADVANCED RING THEORY

R EMARK 5.7.10. If I is a finite set, then the canonical injection

M
Mi → ∏ Mi
i∈I i∈I
is an isomorphism. In this case, the two concepts are often used interchangeably.
N OTATION 5.7.11. A direct sum (resp., product) of two R-modules M and N is denoted
M ⊕ N.
D EFINITION 5.7.12. We say that an R-submodule A of an R-module B is a direct summand
of C if there exists an R-module C such that B = A ⊕C. In this case, C is called a complement to
A in B.
D EFINITION 5.7.13. Let R be a ring.
a. An R-module M is free on a subset X of M if for any R-module N and function φ̄ : X → N of
elements of N, there exists a unique R-module homomorphism φ : M → N such that φ (x) = φ̄ (x)
for all x ∈ X.
b. A basis of an R-module M is a subset of M on which it is free.
R EMARK 5.7.14. An abelian group A is free on a set X if and only if it is a free Z-module on
X, as follows from Proposition 4.4.11.
In fact, we have the following alternative definition of a free R-module. The proof is nearly
identical to Proposition 4.4.11, so omitted.
P ROPOSITION 5.7.15. An R-module M is free on a basis X if and only if the set X generates
M and, for every n ≥ 1 and x1 , x2 , . . . , xn ∈ X, the equality
n
∑ cixi = 0
i=1
for some c1 , c2 , . . . , cn ∈ R implies that ci = 0 for all i.
R EMARK 5.7.16. We might refer to the property that a set X generates an R-module M as
saying that M is the R-span of X. The property that ∑ni=1 ci xi = 0 implies ci = 0, where ci ∈ R
and xi ∈ X for 1 ≤ i ≤ n and some n ≥ 1 can be referred to as saying that the set X is R-linearly
independent.
L
C OROLLARY 5.7.17. For any set X, the R-module x∈X R is free on the standard basis
{ex | x ∈ X}, where ex for x ∈ X is the element which is nonzero only in its x-coordinate, in which
it is 1.
L
P ROOF. The ex span x∈X R by its definition and are clearly R-linearly independent. 
C OROLLARY 5.7.18. Every R-module is a quotient of a free R-module.
P ROOF. Let M be an R-module, and choose a generating set X of M (e.g., M itself). Take the
unique R-module homomorphism M
ψ: R→M
x∈X
which satisfies ψ(ex ) = x for all x ∈ X. It is onto as X generates M. 
 5.7. FREE MODULES AND GENERATORS 167

Noting Corollary 5.5.26, we also have the following.

C OROLLARY 5.7.19. Every vector space over a field F is a free F-module.
The following is also a consequence of the universal property.
T HEOREM 5.7.20. Let R be a commutative ring. A free module M on a set X is isomorphic
to a free module N on a set Y if and only if X and Y have the same cardinality.
P ROOF. If X and Y have the same cardinality, then any bijection f : X → Y gives an injection
X → N which extends uniquely to a homomorphism φ : M → N. Similarly, the inverse of f
extends uniquely to a homomorphism ψ : N → M, and ψ ◦ φ (resp., φ ◦ ψ) is then the unique
extension to a homomorphism of the inclusion X → M (resp., Y → N), therefore the identity.
That is, φ and ψ are inverse isomorphisms.
For the converse, we first suppose that Y is infinite and that there is an isomorphism M → N.
Let B denote the image of X in N, which is then necessarily an R-basis of N. Each element y ∈ Y
is contained in the span of a finite subset By of B. The union B0 of these sets By spans N. For any
v ∈ B − B0 , the set B0 ∪ {v} is R-linearly dependent, which cannot happen as B is a basis. Thus,
B = B0 . Now, the cardinality |B| of B is at most the cardinality of the disjoint union of the sets By
for y ∈ Y , each of which is finite. In particular, we have
|X| = |B| ≤ |Y × Z| = |Y |,
the latter equality holding as Y is infinite. If X is also infinite, then by reversing the roles of X
and Y , this forces |X| = |Y |.
Finally, suppose that Y is finite, without loss of generality. Let m be a maximal ideal of R.
Consider the field F = R/m, and observe that
! !
M/mM ∼ R ∼
M M M
= R /m = F,
x∈X x∈X x∈X
and similarly for Y . An isomorphism M −∼ → N induces an isomorphism of F-vector spaces
∼
M/mM −→ N/mN, which by the above isomorphisms have bases of cardinality |X| and |Y |
respectively. Since Y is finite, Theorem 5.5.28 tells us that X must be finite of order |Y |. 
The following is immediate.
C OROLLARY 5.7.21. Let R be a commutative ring, and let M be a free R-module on a set of
n elements. Then every basis of M has n elements.
By Theorem 5.7.23, we may make the following definition.
D EFINITION 5.7.22. The rank of a free module M over a commutative ring R is the unique
n ≥ 0 such that M ∼
= Rn if it exists. Otherwise, M is said to have infinite rank.
For an integral domain, we can do somewhat better with a bit of work. In fact, the following
result does not require this assumption, but the proof we give does.
T HEOREM 5.7.23. Let R be an integral domain. Let M be a free R-module on a set of n
elements, and let Y be a subset of M. Then:
168 5. ADVANCED RING THEORY

i. if Y generates M, then Y has at least n elements,

ii. if Y is R-linearly independent, then Y has at most n elements, and
iii. Y is a basis if and only if it generates M and has exactly n elements.
Moreover, a free module on an infinite set cannot be generated by a finite set of elements.
P ROOF. Suppose that M is free on n elements. A choice of basis defines an isomorphism
∼
M −→ Rn of R-modules, so we may assume that M = Rn . Note that Rn is contained in the Q(R)-
module Q(R)n via the canonical inclusion, and any generating set Y of Rn spans Q(R)n . But by
Theorems 5.5.28 and 5.5.27, this forces Y to have at least n elements. If Y has n elements, then
Y would similarly be a basis of Q(R)n . So, if we had ∑ni=1 ci yi = 0 for some ci ∈ R and distinct
yi ∈ Y , then each ci = 0, which means that Y is an R-basis of Rn .
On the other hand, if Y has more than n elements, then by Theorem 5.5.25, the set Y cannot
be linearly independent in Q(R)n . That is, there exist αi ∈ Q(R)n and distinct yi ∈ Y for 1 ≤ i ≤ m
−1
and m ≥ 1 with ∑m i=1 αi yi = 0 and not all αi = 0. For each i, write αi = ci di with ci , di ∈ R and
di 6= 0. Taking d to be the product of the di , we then have ai = dαi ∈ R and not all ai = 0. Since
∑mi=1 ai yi = 0, it follows that Y is not a basis.
Finally, if N is a free module on anLinfinite set X, then N ∼
L
= x∈X R, and so we take N to be
the latter module. We then have that x∈X Q(R) is a Q(R)-vector space with an infinite basis.
But then Theorem 5.5.28 tells us that every basis is infinite, which by Theorem 5.5.27 tells us
that a finite set cannot span. 
R EMARK 5.7.24. The full analogues of Theorems 5.5.25 and 5.5.27 do not hold for modules
over arbitrary rings, over even abelian groups. That is, take the free Z-module Z. The set {2}
does not span it and is not contained in a basis of Z, and the set {2, 3} does span it and does not
contain a basis.
E XAMPLE 5.7.25. The polynomial ring R[x] is a free R-module on the basis {xi | i ∈ Z≥0 }.
R EMARK 5.7.26. Consider the ideal I = (2, x) of Z[x]. It is not a free Z[x]-module. To
see this, first note that it is not a principal ideal so cannot be generated by a single element.
As I can be generated by the two elements 2 and x, if I were free, then it would follow from
Theorem 5.7.23 that {2, x} would be a basis for I. On the other hand, x · 2 − 2 · x = 0, which
would contradict Proposition 5.7.15.
P ROPOSITION 5.7.27. Let M be an R-module, and let π : M → F be a surjective R-module
homomorphism, where F is R-free. Then there exists an injective R-module homomorphism
ι : F → M such that π ◦ ι = idF . Moreover, we have M = ker(π) ⊕ ι(F).
P ROOF. Let X be an R-basis of F, and for each x ∈ X, choose mx ∈ M with π(mx ) = x. We
take ι : F → M to be the unique R-module homomorphism with ι(x) = mx for all x ∈ X, which
exists as F is free. Then π ◦ ι(x) = x for all x ∈ X, so π ◦ ι = idF by uniqueness, and ι must be
injective.
Finally, let A = ker π. Note that any m ∈ M satisfies m − ι ◦ π(m) ∈ A, so M = A + ι(F). If
m ∈ A ∩ ι(F), then m = ι(n) for some n ∈ F and n = π ◦ ι(n) = π(m) = 0, so m = 0. In other
words, we have M = A ⊕ ι(F). 
 5.8. MATRIX REPRESENTATIONS 169

In particular, every free quotient of an R-module M is isomorphic to a direct summand of M.

5.8. Matrix representations
We work in this section with (nonzero) homomorphisms of free modules over a ring R. Most
of the time, the case of interest is that of linear transformations of vector spaces over fields, but
there is no additional restriction caused by working is full generality.
L EMMA 5.8.1. Let R be a ring. Let A ∈ Mmn (R) be a matrix for some m, n ≥ 1. Then there is
a unique R-module homomorphism T : Rn → Rm satisfying T (v) = Av for all v ∈ Rn , where Av is
matrix multiplication, viewing elements of Rm and Rn as column vectors.
P ROOF. Define T (e j ) = ∑m i=1 ai j f i , where e j (resp., f i ) is the jth (resp., ith) standard basis
element of Rn (resp., Rm ). If v = ∑nj=1 c j e j for some c j ∈ F with 1 ≤ j ≤ n, then
n m  n 
T (v) = ∑ c j T (e j ) = ∑ ∑ ai j c j fi = Av.
j=1 i=1 j=1
The uniqueness follows from the fact that Rn is free, so any R-module homomorphism from it is
determined by its values on a basis 
D EFINITION 5.8.2. An ordered basis is a basis of a free R-module together with a total
ordering on the basis.
R EMARK 5.8.3. We refer to a finite (ordered) basis on a free R-module as a set {v1 , v2 , . . . , vn }
and take this implicitly to mean that the set has cardinality n and that the basis is ordered in the
listed order (i.e., by the ordering vi ≤ vi+1 for all 1 ≤ i < n).
E XAMPLE 5.8.4. The standard basis {e1 , e2 , . . . , en } on Rn is ordered in the order of positions
of the nonzero coordinate of its elements.
N OTATION 5.8.5. If B = {v1 , v2 , . . . , vn } is an ordered basis of a free R-module V , then we
let ϕB : Rn → V denote the R-module isomorphism satisfying ϕB (ei ) = vi for all i.
Given ordered bases of free R-modules V and W , an R-module homomorphism T : V → W
can be described by a matrix.
D EFINITION 5.8.6. Let V and W be free modules over a ring R with ordered bases B =
{v1 , v2 , . . . , vn } and C = {w1 , w2 , . . . , wm }, respectively. Let T : V → W be an R-module homo-
morphism. We say that a matrix A = (ai j ) ∈ Mnm (R) represents T with respect to the bases B
and C if
m
T (v j ) = ∑ ai j wi
i=1
for all 1 ≤ j ≤ n.
R EMARK 5.8.7. Given ordered bases B = {v1 , . . . , vn } of a free module V and C = {w1 , . . . , wm }
of a free module W , the composition
ϕB T ϕ −1
ϕC−1 ◦ T ◦ ϕB : Rn −→ V − C
→ W −−→ Rm ,
170 5. ADVANCED RING THEORY

is given by multiplication by a matrix A by Lemma 5.8.1. This A is the matrix representing T

with respect to B and C.
T ERMINOLOGY 5.8.8. Let V be a free R-module with finite basis B, and let T : V → V be
an R-module homomorphism. We say say that a matrix A represents T with respect to B if A
represents T with respect to B and B. If V = Rn and B is the standard basis, we simply say that A
represents T .
L EMMA 5.8.9. Let T 0 : U → V and T : V → W be homomorphisms of finite rank free R-
modules. Let B, C, and D be bases of U, V , and W , respectively. Suppose that A0 represents T 0
with respect to B and C and that A represents T with respect to C and D. Then AA0 represents
T ◦ T 0 : U → W with respect to B and D.
P ROOF. We have that A represents ϕD−1 ◦ T ◦ ϕC and A0 represents ϕC−1 ◦ T 0 ◦ ϕB . In other
words, the maps are left multiplication by the corresponding matrices. The map

ϕD−1 ◦ T ◦ T 0 ◦ ϕB = (ϕD−1 ◦ T ◦ ϕC ) ◦ (ϕC−1 ◦ T 0 ◦ ϕB ),

is then left multiplication by AA0 , which is to say that it is represented by AA0 . 
D EFINITION 5.8.10. Let B = {v1 , . . . , vn } and B0 = {v01 , . . . , v0n } be ordered bases of a free
R-module V . The change-of-basis matrix from B to B0 is the matrix QB,B0 = (qi j ) that represents
the R-module homomorphism TB,B0 : V → V with TB,B0 (vi ) = v0i for 1 ≤ i ≤ n with respect to B.
R EMARK 5.8.11. If v0j = ∑ni=1 qi j vi for all i, then the change-of-basis matrix QB,B0 of Defini-
tion 5.8.10 is the matrix (qi j ). It is invertible, and QB0 ,B = Q−1
B,B0 .

R EMARK 5.8.12. Let V be free of rank n with bases B and B0 . By definition, the change-
of-basis matrix QB,B0 represents ϕB−1 ◦ TB,B0 ◦ ϕB . On the other hand, we also have that that
ϕB0 = TB,B0 ◦ ϕB . Thus, see that

ϕB−1 ◦ ϕB0 = ϕB−1 ◦ TB,B0 ◦ ϕB ,

is represented by QB,B0 .
T HEOREM 5.8.13 (Change of basis theorem). Let T : V → W be a linear transformation of
free R-modules of finite rank. Let B and B0 be ordered bases of V and C and C0 be ordered bases
−1
of W . If A is the matrix representing T with respect to B and C, then QC,C 0 AQB,B0 is the matrix
0
representing T with respect to B and C .0

P ROOF. We have that A represents ϕC−1 ◦ T ◦ ϕB , and we wish to compute the matrix repre-
senting ϕC−1
0 ◦ T ◦ ϕB0 . We have

ϕC−1 −1 −1 −1
0 ◦ T ◦ ϕB0 = (ϕC0 ◦ ϕC ) ◦ (ϕC ◦ T ◦ ϕB ) ◦ (ϕB ◦ ϕB0 ),

−1
and these three matrices are represented by QC,C 0 , A, and QB,B0 , respectively. 
 5.9. ASSOCIATIVE ALGEBRAS 171

5.9. Associative algebras

Much as with groups, we may speaker of the center of a ring.
D EFINITION 5.9.1. The center Z(R) of a ring R is the subring of R given by the subset
Z(R) = {a ∈ R | ab = ba for all b ∈ R}.
We now define the notion of an algebra over a commutative ring.
D EFINITION 5.9.2. Let R be a commutative ring. An (associative) R-algebra A is the pair of
an R-module A and a binary operation · on A which, together with the addition on A, makes A
into a ring, and which satisfies
r(a · b) = (ra) · b = a · (rb)
for all r ∈ R and a, b ∈ A.
R EMARK 5.9.3. An R-algebra A comes endowed with a homomorphism φ : R → Z(A), given
by φ (r) = r · 1 for r ∈ R and the element 1 ∈ A. In fact, to give an R-algebra A is to give a ring
A and a ring homomorphism φ : R → Z(A) with φ (1) = 1, for then this provides an R-module
structure on A given by r · a = φ (r)a, which makes A into an R-algebra.
R EMARK 5.9.4. Often, it is supposed that the map φ : R → Z(A) that defines the R-algebra
structure on A is injective. This is automatically the case if R is a field.
E XAMPLES 5.9.5.
i. The polynomial ring R[x1 , . . . , xn ] over a commutative ring R is an R-algebra for any n ≥ 1.
ii. The matrix ring Mn (R) over a commutative ring R is an R-algebra for any n ≥ 1.
iii. If F is a field and E is a field containing F as a subring, then F is an E-algebra.
iv. Every ring R is a Z-algebra.
E XAMPLE 5.9.6. The ring H = {a + bi + c j + dk | a, b, c, d ∈ R} of quaternions has center
Z(H) = R, hence is an algebra over R. Note that C = {a + bi | a, b ∈ R} is contained in H, but
H is not a C-algebra, as C is not contained in the center of H.
D EFINITION 5.9.7. An R-algebra homomorphism f : A → B is a ring homomorphism of
R-algebras A and B that is also a homomorphism of R-modules.
E XAMPLE 5.9.8. For any R-algebra A, the structure homomorphism φ : R → A with image
in Z(A) is a homomorphism of R-algebras.
We provide three other classes of examples.
D EFINITION 5.9.9. Let R be a commutative ring and M an R-module. Then endomorphism
ring EndR (M) of M over R is the R-algebra of R-linear endomorphisms of M. It is a ring under
addition and composition of endomorphisms, and has the R-module structure given by multipli-
cation of scalars: that is,
(r · f )(m) = r · f (m)
for all f ∈ EndR (M), r ∈ R, and m ∈ M.
172 5. ADVANCED RING THEORY

R EMARK 5.9.10. The map φ : R → Z(EndR (M)) defining the R-algebra structure on EndR (M)
takes r ∈ R to left multiplication by r on M.
D EFINITION 5.9.11. The automorphism group AutR (M) of an R-module M is the group of
R-automorphisms of M under composition.
R EMARK 5.9.12. The unit group of EndR (M) is AutR (M).
E XAMPLE 5.9.13. We have an isomorphism of R-algebras EndR (Rn ) −∼ → Mn (R) by taking
T ∈ EndR (R ) to the matrix A representing it with respect to the standard basis of Rn . We have
n

that T ∈ AutR (Rn ) if and only if A is invertible.

P ROPOSITION 5.9.14. Let M be an R-module and A be an R-algebra with structure map
φ : R → Z(A). There is a bijection between operations · : A × M → M which make M into a left
A-module such that φ (r) · m = r · m for all r ∈ R and m ∈ M and R-algebra homomorphisms
ψ : A → EndR (M) determined by ψ(a)(m) = a · m for all a ∈ A and m ∈ M.
We mention another class of algebras known as group rings. The reader will easily check the
following.
L EMMA 5.9.15. Let R be a commutative ring an G be a group. The set R[G] of elements
∑g∈G ag g with ag ∈ R for all g ∈ G and almost all ag = 0 and addition and multiplication of
multiplication are given respectively by the formulas
! ! !
∑ agg + ∑ bgg = ∑ (ag + bg)g and ∑ agg · ∑ bg g = ∑ ∑ ahbh−1g g
g∈G g∈G g∈G g∈G g∈G g∈G h∈G
is an R-algebra with R-module structure given by the scalar multiplication
r· ∑ agg = ∑ (rag)g.
g∈G g∈G

R EMARK 5.9.16. As an R-module, R[G] is in bijection with the free R-module g∈G R ·
L

g with basis G. The multiplication on R[G] is the unique multiplication that restricts to the
multiplication on G and makes R[G] into an R-algebra. The identity element 1 in R[G] is the
identity of G.
D EFINITION 5.9.17. The group ring R[G] of a group G with coefficients in a commutative
ring R is the unique R-algebra that is free as an R-module with basis G and has multiplication
that restricts to the multiplication on G.
E XAMPLE 5.9.18. For n ≥ 1, there is an isomorphism
n−1 n−1
R[x]/(x − 1) −∼
n
→ R[Z/nZ], ∑ aix i
7→ ∑ ai[i],
i=0 i=0
of R-modules, where [i] denotes the group element corresponding to i ∈ Z/nZ. Similarly, one
has R[x, x−1 ] ∼
= R[Z].
We also have the notion of a free R-algebra over a commutative ring R.
 5.9. ASSOCIATIVE ALGEBRAS 173

D EFINITION 5.9.19. Let R be a commutative ring, and let X be a set. The free R-algebra
RhXi on X is the R-algebra with underlying additive group the free R-module on the words in X
and multiplication the unique R-bilinear map given on words in X by concatenation.
R EMARK 5.9.20. Another way of describing the free R-algebra on a set X is that it’s a non-
commutative polynomial ring with variables in X.
N OTATION 5.9.21. If X = {x1 , . . . , xn } has n elements, then we write Rhx1 , . . . , xn i for RhXi.
 CHAPTER 6

Field theory and Galois theory

6.1. Extension fields

D EFINITION 6.1.1. A field E is an extension field (or extension) of F if F is a subfield of E.
We write E/F (which reads “E over F”) to denote that E is an extension field of F, and we say
that E/F is a field extension, or an extension of fields.
E XAMPLES 6.1.2. We have that R is an extension field of Q, and C is an extension of both
Q and R. We have that Q(i) is an extension of Q of which C, but not R, is an extension field.
We will often have cause to deal with the field Z/pZ, where p is a prime. When we think of
Z/pZ as a field, we make a change of notation.
D EFINITION 6.1.3. For a prime p, the field of p elements, F p , is Z/pZ.
L EMMA 6.1.4. Let F be a field. If F has characteristic 0, then F is an extension of Q. If F
has characteristic equal to a prime p, then F is an extension of F p .
P ROOF. If F has characteristic 0, we define ι : Q → F by ι(ab−1 ) = (a · 1) · (b · 1)−1 , where
a, b ∈ Z and b 6= 0. Since ι is a ring homomorphism and Q is a field, it is injective, so Q
sits isomorphically inside F. If F has characteristic p, then we define ι : F p → F by the same
equation, where now a, b ∈ Z and b 6≡ 0 mod p. Since F has characteristic p, this is a ring
homomorphism, and again it is injective. 
D EFINITION 6.1.5. An intermediate field of a field extension E/F is a subfield E 0 of E con-
taining F. The extension E 0 /F is said to be a subextension of F in E.
D EFINITION 6.1.6. The ground field (or base field) of a field extension E/F is the field F.
D EFINITION 6.1.7. Let E/F be a field extension. Let A ⊂ E. The field generated over F by
the set A (or its elements) is the smallest subfield K of E containing F and A, often denoted F(A).
We say that the elements of A generate K as an extension of F and that K is given by adjoining
the elements of A to F.
N OTATION 6.1.8. Let E/F be a field extension and α1 , α2 , . . . , αn ∈ E for some n ≥ 0. We
write F(α1 , α2 , . . . , αn ) for the subfield of E generated by the set α1 , α2 , . . . , αn over F.
R EMARK 6.1.9. One often says “F adjoin α” to refer to a field F(α).
R EMARK 6.1.10. Note that the field generated over F by a set of elements A of E is well-
defined, equal to the intersection of all subfields of E containing both F and A.
175
176 6. FIELD THEORY AND GALOIS THEORY

R EMARK 6.1.11. Note that we distinguish between the field F(x1 , x2 , . . . , xn ) of rational
functions, where x1 , x2 , . . . , xn are indeterminates, and F(α1 , α2 , . . . , αn ), where α1 , α2 , . . . , αn
are elements of an extension field of F. These fields can be quite different. However, in that
F(α1 , α2 , . . . , αn ) is the quotient field of F[α1 , α2 , . . . , αn ], every element of F(α1 , α2 , . . . , αn ) is
a rational function in the elements αi with 1 ≤ i ≤ n.
√
E XAMPLE 6.1.12. The fields Q( 2) and Q(i) are extension fields of Q inside R and C,
respectively.
P ROPOSITION 6.1.13. Let E/F be a field extension, and let α ∈ E. Then F(α) is isomorphic
to the quotient field of F[α].
P ROOF. Since F[α] is the smallest subring of E containing F and α and F(α) is the smallest
subfield of E containing F and α, inclusion provides an injective homomorphism
ι : F[α] → F(α).
Since F(α) is a field, ι induces an injective map Q(ι) : Q(F[α]) → F(α). Since the image of
Q(ι) is a subfield of F(α) containing F and α and F(α) is the smallest such field in E, we have
that Q(α) is surjective as well. 
In many cases, an extension field generated by an element is actually equal to the ring gener-
ated by the element. We see this holds in a couple of simple examples.
√ √
E XAMPLE 6.1.14. The fields Q( 2)√and Q(i) equal Q[ 2] and Q[i] as subrings √ of R and
C respectively. E.g., the elements of Q( 2) all may be written in the form a + b 2 for some
a, b ∈ Q.
√
The key in this example is that 2 and i are roots of polynomials with coefficients in Q. Let
us examine this further.
T HEOREM 6.1.15. Let E be an extension field of a field F, and let f ∈ F[x] be an irreducible
polynomial that has a root α ∈ E. Then the evaluation map eα : F[x] → F(α) given by eα ( f ) =
f (α) induces an isomorphism
eα : F[x]/( f ) −∼
→ F(α)
of fields such that eα (a) = a for all a ∈ F.
P ROOF. First, note that we have the inclusion map F → F[x] and the quotient map F[x]/( f ),
inducing a nonzero, and hence injective, map of fields F → F[x]/( f ). This allows us to view F
as a subfield of F[x]/( f ). The map eα has kernel containing f , and if eα (g) = g(α) = 0 for some
g ∈ F[x], then g also has α as a root. Since any GCD of f and g will then have α as a root, we
have that the GCD of f and g is ( f ), and in particular, f divides g, so g ∈ ( f ). Therefore, eα is
injective. Since the image of eα is a field containing F and α, it must then be equal to F(α). 
E XAMPLE 6.1.16. The field Q(i) is isomorphic to the quotient ring Q[x]/(x2 + 1).
We now obtain the following theorem as a corollary.
T HEOREM 6.1.17 (Kronecker). Let F be a field, f ∈ F[x] a polynomial. Then there exists a
field extension E of F and an element α ∈ E such that f (α) = 0.
 6.1. EXTENSION FIELDS 177

P ROOF. First, we may assume that f is irreducible by replacing f by an irreducible poly-

nomial dividing it which has α as a root. We then set E = F[x]/( f ), which is a field. Let
α = x + ( f ). Then f (α) is the image of f in E, and so f (α) = 0. 
D EFINITION 6.1.18. Let E/F be a field extension. A nonconstant polynomial f ∈ F[x] is
said to split (or factor completely) in E if it can be written as a product of linear polynomials in
E[x].
D EFINITION 6.1.19. Let F be a field. A splitting field E for f ∈ F[x] over F is an extension
of F such that f splits in E but not any proper subextension of F in E.
E XAMPLES 6.1.20.
√
a. The field Q( 2) is the splitting field of x2 − 2, since it contains both of its roots. It is also
the splitting field of (x − a)2 − 2b2 for any a, b ∈ Q.
√ √
b. The field Q( 3 2) is not the splitting field of x3 − 2, since it contains 3 2√but not
√ its other
√
two roots. On the other hand, if ω ∈ C is a primitive cube root of unity, then Q( 2, ω √2, ω 2 3 2)
3 3

is a splitting field for x3 − 2 inside C. This field may be written more simply as Q(ω, 3 2).
As a corollary of Kronecker’s theorem, we have the following.
C OROLLARY 6.1.21. Let F be a field, and let f ∈ F[x]. Then there exists a splitting field for
f over F.
P ROOF. Let n = deg f . The result is clearly true for n = 1. Set K = F[x]/( f ). Then f has
a root α in K by Kronecker’s theorem. Set g(x) = (x − α)−1 f (x) ∈ E[x]. Then there exists a
splitting field E of g over K which is generated by the roots of g over F by induction. This E is
a splitting field of f over F, since it is generated by the roots of f . 
We next distinguish two types of elements of extension fields of F: those that are roots of
polynomials and those that are not.
D EFINITION 6.1.22. Let E/F be a field extension. An element α ∈ E is called algebraic over
F if there exists a nonzero f ∈ F[x] such that f (α) = 0. Otherwise, α is said to be transcendental
over F.
When speaking of elements of extensions of Q, we speak simply of algebraic and transcen-
dental numbers.
D EFINITION 6.1.23. An element of C is said to be an algebraic number if it is algebraic over
Q and a transcendental number if it is transcendental over Q.
√
E XAMPLES 6.1.24. The number 2 is an algebraic number, since it is a root of x2 − 2.
Similarly, i is algebraic, being a root of x2 + 1. However, the real number π is transcendental,
and the real number e such that log e = 1 is transcendental as well. We do not prove the latter
two facts here.
E XAMPLE 6.1.25. A real number given by repeated square roots
r
√
q
a1 + a2 + · · · + an
178 6. FIELD THEORY AND GALOIS THEORY

with the ai positive rational numbers is algebraic: it is a root of

(· · · ((x2 − a1 )2 − a2 )2 · · · )2 − an .
E XAMPLE 6.1.26. If F is a field and α ∈ F, then α is algebraic over F, being a root of x − α.
Note the following.
P ROPOSITION 6.1.27. Let E/F be a field extension, and let α ∈ E. Then α is transcendental
over F if and only if the evaluation homomorphism eα : F[x] → E is injective.
P ROOF. By definition, α ∈ E is transcendental if and only if g(α) 6= 0 for every g ∈ F[x] that
is nonzero. But g(α) = eα (g), so we are done. 
This allows us to give the prototypical example of a transcendental element.
C OROLLARY 6.1.28. Let F be a field. The element x of the field F(x) of rational functions of
F is transcendental over F.
P ROOF. Let y be an indeterminate. Consider eα : F[y] → F(x) given by eα (g) = g(x). We
have that the polynomial g(x) is zero in F(x) if and only if it is zero in F[x], and therefore if and
only if g = 0 in F[y]. 
T HEOREM 6.1.29. Let E/F be a field extension, and let α ∈ E be algebraic over F. Then
there exists a unique monic, irreducible polynomial f ∈ F[x] such that f (α) = 0.
P ROOF. Since α is algebraic over F, there exists a polynomial g ∈ F[x] such that g(α) = 0.
Since g factors as a product of irreducible polynomials, and E is in particular an integral domain,
one of the irreducible factors must have α as a root, and by multiplying it by a constant, we may
take it to be monic. So suppose that f is a monic irreducible polynomial in F[x] with f (α) = 0.
Without loss of generality, we may assume that deg f is minimal among all such polynomials. If
f 0 ∈ F[x] satisfies f 0 (α) = 0, then the division algorithm provides q, r ∈ F[x] with r = f 0 − q f
and either deg r < deg f or r = 0. Since r(α) = 0, we must have r = 0, but then f 0 = q f , so f
divides f 0 . If f 0 were monic and irreducible, this would force f 0 = f , as desired. 
D EFINITION 6.1.30. Let E/F be a field extension, and let α ∈ E be algebraic over F. The
minimal polynomial of α over F is the unique monic irreducible polynomial in F[x] which has α
as a root.
E XAMPLES 6.1.31.
a. If F is a field and α ∈ F, then x − α is the minimal polynomial of α over F.
b. The polynomial x2 + 1 is the minimal polynomial of i over Q.

6.2. Finite extensions

R EMARK 6.2.1. If E/F is an extension of fields, then E is an F-vector space via the restric-
tion of the multiplication in E to a map F × E → E, which is then given by a · α = aα for a ∈ F
and α ∈ E. Moreover, E actually contains F, so F is a F-subspace of E.
 6.2. FINITE EXTENSIONS 179

D EFINITION 6.2.2. An extension E/F of fields is finite if E is a finite-dimensional field

extension of F. Otherwise, E/F is said to be an infinite extension.
√ √
E XAMPLE
√ 6.2.3. The field Q( 2) is an extension of Q with basis {1, 2} and hence is a
basis of Q( 2) over Q.
E XAMPLE 6.2.4. The field F(x) of rational functions over a field F is infinite. More gener-
ally, if E/F is an extension that contains an element that is transcendental over F, then E is an
infinite extension of F.
D EFINITION 6.2.5. The degree [E : F] of a finite extension E of a field F is defined to be the
dimension dimF E of E as a vector space over F. If E/F is an infinite extension, we say that the
degree of E over F is infinite.
√ √
E XAMPLE
√ 6.2.6. The degree
√ [Q( 2) : Q] is 2, as 2 has minimal polynomial x2 − 2. The
set {1, 2} forms a basis of Q( 2) as a Q-vector space.
The following is essential to our studies.
T HEOREM 6.2.7. Let F be a field, and let f ∈ F[x] be an irreducible polynomial of degree n.
Then the field F[x]/( f ) has degree n over F.
P ROOF. Since ( f ) contains only multiples of f , it contains no nontrivial linear combinations
of the monomials xi with 0 ≤ i ≤ n − 1. In other words, the xi + ( f ) with 0 ≤ i ≤ n − 1 are
linearly independent over F. On the other hand, if g ∈ F[x], then g = q f + r with q, f ∈ F[x] and
deg r < n, so g + ( f ) = r + ( f ), and therefore g + ( f ) may be written as the image in the quotient
of a linear combination of the monomials xi with 0 ≤ i ≤ n − 1. That is, the elements xi + ( f )
with 0 ≤ i ≤ n − 1 form a basis of F[x]/( f ). 
The proof of Theorem 6.2.7, when taken together with Theorem 6.1.15, yields the following.
C OROLLARY 6.2.8. Let E/F be a field extension, and let α ∈ E be algebraic over F. Let n
be the degree of the minimal polynomial of F. Then [F(α) : F] = n, and {1, α, . . . , α n−1 } is a
basis of F(α) over F.
P ROPOSITION 6.2.9. If E/F is a finite extension and α ∈ E, then α is algebraic over F.
P ROOF. Since [E : F] is finite, there is an n ≥ 1 such that the set {1, α, . . . , α n } is F-linearly
dependent. We then have
n
∑ ciα i = 0
i=0
for some ci ∈ F with 0 ≤ i ≤ n. Setting f = ∑ni=0 ci xi ∈ F[x], we see that f (α) = 0, so α is
algebraic. 
C OROLLARY 6.2.10. Every finite extension E of a field F has the form E = F(α1 , α2 , . . . , αn )
for some algebraic αi ∈ E for 1 ≤ i ≤ n.
P ROOF. One may simply take {αi | 1 ≤ i ≤ n} to be a basis of E over F. Since each αi ∈ E,
we have F(α1 , α2 , . . . , αn ) ⊆ E, and since every element in E is a linear combination of the αi ,
we have the opposite containment. 
180 6. FIELD THEORY AND GALOIS THEORY

The following theorem, while stated for arbitrary field extensions, has a number of applica-
tions to finite extensions.
T HEOREM 6.2.11. Let E be an extension of a field F, and let K be an extension of E. If A is
a basis of E over F and B is a basis of K over E, then
AB = {αβ | α ∈ A, β ∈ B}
is a basis of K over F, and the map A × B → AB given by multiplication in K is a bijection.
P ROOF. We first show that AB spans K. By definition of B, any γ ∈ K can be written as
m
γ= ∑ c jβ j
j=1
with c j ∈ E and β j ∈ B for 1 ≤ j ≤ m and some m ≥ 1. Each c j is in the F-span of some finite
subset of A. By taking the union of these subsets, we see that there is a single finite subset of A
such that every c j with 1 ≤ j ≤ m is in its span. That is, we may write
n
c j = ∑ d i j αi
i=1
for some di j ∈ F and αi ∈ E for 1 ≤ i ≤ n and some n ≥ 1. Plugging in, we obtain
n m
γ=∑ ∑ d i j αi β j ,
i=1 j=1
so the set AB spans K over F.
Now, if some F-linear combination of the elements of AB equals zero, then in particular (by
throwing in terms with zero coefficients if needed) we may write
n m
∑ ∑ ai j αiβ j = 0
i=1 j=1
for some αi ∈ A, β j ∈ B, and ai j ∈ F for 1 ≤ i ≤ n and 1 ≤ j ≤ m, for some m and n. Since the
β j are E-linearly independent, this implies that
n
∑ a i j αi = 0
i=1
for all 1 ≤ j ≤ m. Since the αi are F-linearly independent, we then have that ai j = 0 for all i and
j. Therefore, the set AB is a basis of K over F.
Note that we may also conclude that the surjection A × B → AB given by multiplication in K
is injective. If it were not, then we would have two distinct pairs (α, β ), (α 0 , β 0 ) ∈ A × B such
that αβ − α 0 β 0 = 0, contrary to what we have shown. 
Theorem 6.2.11 has the following almost immediate corollary.
C OROLLARY 6.2.12. Let E be a finite extension of a field F, and let K be a finite extension
of E. Then K/F is a finite extension, and we have
[K : F] = [K : E][E : F].
 6.2. FINITE EXTENSIONS 181

P ROOF. Theorem 6.2.11 tells us that any basis of K over F has [K : E][E : F] elements, hence
the result. 
This corollary has in turn the following two corollaries.
C OROLLARY 6.2.13. Let E be a finite extension of a field F, and let K be a finite extension
of E. Then [K : E] and [E : F] divide [K : F].
C OROLLARY 6.2.14. Let K/F be a finite extension, and let E be a subfield of K containing
F. Then K/E and E/F are finite extensions.
E XAMPLE 6.2.15. By Corollary 6.2.12, we have
√ √ √ √
[Q(i, 2) : Q] = [Q(i, 2) : Q( 2)][Q( 2) : Q],
√ √ √ √
/ Q(√2), we have that x2 + 1 is irreducible in Q( 2)[x], so [Q(i, 2) : Q( 2)] = 2.
and since i ∈
Therefore, [Q(i, 2) : Q] = 4.
We give another corollary of Corollary 6.2.12 that is a converse to Corollary 6.2.10.
C OROLLARY 6.2.16. Let K/F be a field extension, and let α1 , α2 , . . . , αn ∈ K be algebraic.
Then F(α1 , α2 , . . . , αn ) is a finite extension of F.
P ROOF. The corollary is true for n = 1 by definition of an algebraic element. Suppose by
induction we know it for n − 1, and let E = F(α1 , α2 , . . . , αn−1 ), which is a finite extension of F
by induction. Note that αn is algebraic over E in that it is algebraic over F. Since K = E(αn ),
we therefore have that K is a finite extension of E. That K/F is a finite extension now follows
from Corollary 6.2.12. 
D EFINITION 6.2.17. A field extension E/F is said to be algebraic if every element of E is
algebraic over F. Otherwise, E/F is said to be a transcendental extension.
P ROPOSITION 6.2.18. Every finite extension is algebraic.
P ROOF. If α ∈ E, then F(α) ⊆ E, so F(α)/F is finite. Hence, α is algebraic over F. 
In fact, we can do better.
P ROPOSITION 6.2.19. Let E be an intermediate field in a field extension K/F. Then K/F is
algebraic if and only if both K/E and E/F are algebraic.
P ROOF. Suppose that K/E and E/F are algebraic. Let α ∈ K, and let f = ∑ni=0 ai xi ∈ E[x]
be its minimal polynomial over E. Since E/F is algebraic, the field E f = F(a1 , . . . , an ) is finite
over of F, and therefore so is E f (α). In particular, α is algebraic over F, and therefore K/F is
algebraic. The other direction is immediate. 
R EMARK 6.2.20. A transcendental field extension can never be finite.
E XAMPLES 6.2.21.
a. The field R is a transcendental extension of Q.
√ √ √ √
b. The field K = Q( 2, 3 2, 4 2, 5 2, . . .)√
is an algebraic extension of Q, as the field generated
by any finite list of these roots is equal to Q( n 2) for n ≥ 2, every element of K is contained such
a field, and each of these fields is algebraic over Q.
182 6. FIELD THEORY AND GALOIS THEORY

6.3. Composite fields

D EFINITION 6.3.1. Let E1 and E2 be subfields of a field K. The compositum, or composite
field, E1 E2 of E1 and E2 is the smallest subfield of K containing both E1 and E2 .
R EMARK 6.3.2. The compositum E1 E2 of subfields E1 and E2 of a field K is the intersection
of all subfields of K containing both E1 and E2 .
E XAMPLE 6.3.3. Let K/F be a field extension, and let α, β ∈ K. Then
F(α, β ) = F(α)F(β ).
More generally, if E is any subfield of K containing F, then
EF(α) = E(α).
We prove the following in the case of finite extensions. Note, though, that this finiteness is
not needed, as seen through Corollary 6.3.11 below.
P ROPOSITION 6.3.4. Let E1 and E2 be finite extensions of a field F contained in a field K.
Suppose that A and B are bases of E1 and E2 as F-vector spaces, respectively. Then E1 E2 is
spanned by the set AB.
P ROOF. Set m = [E1 : F] and n = [E2 : F], and let A = {α1 , α2 , . . . , αm } and B = {β1 , β2 , . . . , βn }.
Clearly, we have
E1 E2 = F(α1 , α2 , . . . , αm , β1 , β2 , . . . , βn ).
As the elements of A and B are algebraic, we have E(αi ) = E[αi ] and E(βi ) = E[βi ] for any field
E containing F, for all i and j. We then see by a simple recursion that every element of E1 E2
may actually be expressed as a polynomial in the elements of A and B with coefficients in F, not
just a rational function. However, any monomial in the elements of A lies in E1 , hence may be
written as a linear combination of the elements of A. Similarly, any monomial in the elements
of B lies in E2 , hence may be written as a linear combination of the elements of B. Therefore,
every monomial is the elements of A and B may be written as a product of a linear combination
of elements of A with a linear combination of elements of B, which is the a linear combination of
elements of AB. Since every polynomial is a linear combination of monomials, we are done. 
C OROLLARY 6.3.5. Let E1 and E2 be finite extensions of a field F that are contained in a
field K. Then we have
[E1 E2 : F] ≤ [E1 : F][E2 : F].
P ROOF. Let A (resp., B) be a basis of E1 (resp., E2 ) over F. Then AB has at most [E1 : F][E2 :
F] elements and spans E1 E2 over F. 
C OROLLARY 6.3.6. Let E1 and E2 be finite extensions of a field F that are contained in a
field K, and suppose that [E1 : F] and [E2 : F] are relatively prime. Then we have
[E1 E2 : F] = [E1 : F][E2 : F].
P ROOF. Both [E1 : F] and [E2 : F] divide [E1 E2 : F], so by their relative primality, their prod-
uct [E1 : F][E2 : F] does as well. So we have [E1 E2 : F] ≥ [E1 : F][E2 : F], while Corollary 6.3.5
provides the opposite inequality. 
 6.3. COMPOSITE FIELDS 183

D EFINITION 6.3.7. Let n ≥ 1. An nth root of unity is an element of order dividing n in the
multiplicative group of a field.

That is, if F is a field, ζ ∈ F is an nth root of unity if and only if ζ n = 1.

√ in C that is not equal to 1. Note that [Q(ω) :

E XAMPLE 6.3.8. Let ω be a third√root of unity
Q] = 2, since ω 2 + ω + 1 = 0. Then 3 2 and ω 3 2 are both cube roots of 2, and we have
√3
√
3
√3
Q( 2, ω 2) = Q(ω, 2)

We then see that

√
3
√
3
[Q( 2) : Q][Q(ω 2) : Q] = 9,
while
√
3
√
3
[Q(ω, 2) : Q] = [Q( 2) : Q][Q(ω) : Q] = 6
by Corollary 6.3.6.

More generally, we may define the compositum of a collection of fields.

D EFINITION 6.3.9. Let {Ei | i ∈ I} be a collection of subfields of a field K for some indexing
set I. Then the compositum of the fields Ei for i ∈ I is smallest subfield of K containing all Ei .

Let us give an alternate description of the compositum.

L EMMA 6.3.10. Let {Ei | i ∈ I} be a collection of intermediate fields in an extension K/F

for some indexing set I. Then the compositum E of the Ei is equal to the union of its subfields
F(α1 , α2 , . . . , αn ), where n ≥ 0 and each α j with 1 ≤ j ≤ n is an element of Ei for some i ∈ I.

P ROOF. Clearly the above-described union U is contained in E and contains each Ei . How-
ever, we must show that U is a field, hence equal to E. If a, b ∈ U are nonzero, then a ∈
F(α1 , α2 , . . . , αn ) and b ∈ F(β1 , β2 , . . . , βm ), where n, m ≥ 0 and the α j and βk are elements
of the Ei . Then
a − b, ab−1 ∈ F(α1 , α2 , . . . , αn , β1 , β2 , . . . , βm ),
and the latter field is a subset of U, so U = E. 

We have the following corollary.

C OROLLARY 6.3.11. Let {Ei | i ∈ I} be algebraic extensions of a field F that are contained
in a field K, where I is an indexing set. Then the compositum E of the fields Ei is an algebraic
extension of F.

P ROOF. By Lemma 6.3.10, any α ∈ E is an element of a subfield F(α1 , α2 , . . . , αn ) of E,

where each α j ∈ Ei for some i ∈ I. Since Ei is algebraic, F(α j )/F is finite for all 1 ≤ j ≤ n, and
therefore F(α1 , α2 , . . . , αn )/F is finite by Corollary 6.3.5. 
184 6. FIELD THEORY AND GALOIS THEORY

6.4. Constructible numbers

In this section, we discuss a classical problem of the ancient Greeks, which we present as a
game. The game begins with a line segment of length 1 that has already been drawn on the plane.
One is given two tools: a straightedge and a compass. At any step of the game, one can either use
the straightedge to draw a line segment or the compass to draw a circle, in ways we will shortly
make more specific. The goal of the game is to draw a line segment of a given desired length in
a finite number of steps.
At any step, we consider a point to have been marked if it is either the endpoint of an already
drawn line segment or the intersection of a drawn line segment or circle with another drawn line
segment or circle. The straightedge allows us to draw a line segment between any two marked
points and also to extend any previously drawn line segment until it meets any point that has
already been drawn on the plane. The compass allows us to draw a circle that contains a given
marked point and has as its center any other marked point.
Given these rules, we may now make the following definition.
D EFINITION 6.4.1. A real number α is said to be constructible if one can draw a line segment
of length |α| in the plane, starting from a line segment of length 1, using a straightedge and
compass, in a finite number of steps.
We will denote a line segment between two distinct points A and B in R2 by AB. Its length
will be denoted by |AB|. We prove a few preliminary results.
L EMMA 6.4.2. Suppose that a line segment AB has been drawn in the plane.
a. We may draw a line segment bisecting AB.
b. We may draw a line segment AC perpendicular to AB.
c. Given a point D in the plane, we may draw a line segment DE parallel to AB.
P ROOF. For part a, draw circles with center A and center B, both of radius |AB|. These
intersect at two points, and the line segment between them is perpendicular to AB and passes
through a midpoint F of that segment.
For part b, by drawing the circle with center A and radius |AF|, we may mark a point G on the
line that contains |AB| that is on the opposite side of A from F and is such that |AF| = |AG|. As
we have already shown, we may then draw a line segment CH bisecting FG and passing through
A, which provides us with AC.
For part c, if BD is perpendicular to AB, we set H = D. Otherwise, we draw a circle with
center D and passing through B. It intersects AB in a second point H. We draw a line segment
through D bisecting BH using part a. We then use part b to draw a perpendicular DE to BH, and
it is by definition parallel to AB. 
We also have the following.
L EMMA 6.4.3. Suppose we have drawn either a line segment of length α or a circle of radius
α in the plane.
a. We may draw a line segment of length α with any marked point as an endpoint, along any
line that contains at least one other marked point.
 6.4. CONSTRUCTIBLE NUMBERS 185

b. We may draw a circle of radius α with center any marked point.

P ROOF. First we note that the two assumptions are equivalent. Given a line segment of length
α, we may use its endpoints to draw a circle of radius α. Given a circle of radius α and center
A, since we have at least one marked point other than its center in the plane, we can by drawing
the line segment from the center to that point mark a point B on the circle. The resulting line
segment B then has radius α.
Suppose then that we are given a line segment AB of length α and another line segment CD.
Me make two constructions using Lemma 6.4.2. We draw a line segment CE parallel to AB. We
draw the line segment AC and then the parallel to AC passing through B. It intersects the line
through C and E at a point F such that |AF| = α. The circle with center A passing through F
then determines a point G on the line through A and C such that |AG| = α. We thus have both
parts. 
We prove the following.
T HEOREM 6.4.4. The set of constructible numbers is a subfield of R.
P ROOF. Suppose that α and β are constructible and positive. Then we may draw a line
segment AB of length α in the plane, and we may then draw a line segment BC of length β along
the line defined by AB. If we do this so that it overlaps with AB, then we have constructed a line
segment AC of length |α − β |.
On the other hand, given AB of length α, draw a line segment AC of length β that is perpen-
dicular to AB, and let E be the point on the ray defined by AC such that AE has length 1. Draw
the line segment CB, and use it to draw a parallel line segment from E to a point D on the ray
defined by the segment AB. We then have that the triangle ABC is similar to the triangle ADE, so
|AD| |AB| α
|AD| = = = .
|AE| |AC| β
Therefore, αβ −1 is constructible. 
T HEOREM 6.4.5. The field of constructible numbers consists exactly of the real numbers that
can be obtained from 1 by applying a finite sequence of the operations of addition, subtraction,
multiplication, division (with nonzero denominators), and the taking of square roots (of positive
numbers), using numbers already obtained from 1 at an earlier point in the sequence.
P ROOF. We first show that the square root of a constructible positive number α is con-
structible. For this, draw a line segment AD of length 1 + α and mark a point B at distance
1 from A and α from D along the segment. Find the midpoint O of AD, and draw a circle with
center O and radius |AO| = (1 + α)/2. Draw a perpendicular to AD at the point B, and let C be a
point where it intersects the drawn circle. Then the triangle ABC is similar to the triangle CBD,
and therefore we have
|BC| |BD| α
|BC| = = = ,
|AB| |BC| |BC|
√
and hence |BC| = α.
186 6. FIELD THEORY AND GALOIS THEORY

For the converse, we merely give a sketch. Let E be the set (or actually, field) of numbers that
can be constructed from 1 using field operations and square roots. Suppose that our initial line
segment was between (0, 0) and (1, 0) on the plane. Suppose that all previously marked points
have coordinates in E. These points have been marked as the intersection points of lines and
circles, where the lines are determined by previously marked points with E-coordinates and the
circles have centers previously marked points with E-coordinates and are chosen to pass through
marked points with E-coordinates. Every drawn line thus has the form ax + by + c = 0 with
a, b, c ∈ E, and every drawn circle has the form x2 + y2 + dx + ey + f = 0 with d, e, f ∈ E. The
intersection of two such lines has coordinates obtained by field operations on the coefficients of
the two lines in question. The coordinates of the intersection points of a line and a circle coming
from the solution of a quadratic equation with coefficients are obtained by field operations on
the coefficients of the line and the circle. Finally, the intersection of two circles can be reduced
to the latter case by considering a common chord (or tangent line). It follows that any new line
segment or circle created with these operations has two marked points in E, and therefore every
constructible length lies in E as well. 
Since the square root of a field element defines an extension of degree dividing 2 of the field
in which it lies, we have the following.
C OROLLARY 6.4.6. Let α be a constructible number. Then α is an algebraic number, and
[Q(α) : Q] is a power of 2.
C OROLLARY 6.4.7. The field of constructible numbers is an algebraic extension of Q.
The ancient Greeks were in particular very concerned with three problems that they could
not solve with a straightedge and compass. This was for good reason: they involved constructing
line segments of unconstructible length. Yet, the Greeks never managed to prove this, and it was
not until the 19th century that proofs were finally given. We list these three problems now.
E XAMPLES 6.4.8.
a. It is impossible to “double the cube.” That is, given a line segment, one cannot construct
from it a new line segment such that a cube with the new line segment as one of its sides would
have twice the volume of a cube with the original line segment as its side. Assuming the initial
line
√ segment had√ a constructible length α, the new line segment would have to have length
3
2α, but then 3 2 would be constructible, yet it defines an extension of degree 3 over Q, in
contradiction to Corollary 6.4.6.
b. It is impossible to “square the circle.” That is, given a drawn circle, it is impossible to
construct
√ a square with the same area.√If the circle had radius r, then the square would have side
length πr, which would mean that π would be constructible, and hence π would be as well,
in contradiction to Corollary 6.4.6, since π is transcendental.
c. It is impossible to “trisect all angles.” That is, given an arbitrary angle between two drawn
line segments with a common endpoint in a plane, it is not always possible to draw a line segment
with the same endpoint having an angle with one of the line segments that is a third of the
original angle. Note that an initial such angle θ exists if and only if cos θ is constructible, as
seen by drawing a perpendicular from one line segment at point a distance one from the point
 6.5. FINITE FIELDS 187

of intersection until it intersects the line defined by the other. Therefore, the problem is, given
a constructible number α = cos θ , to show that cos(θ /3) is constructible. However, we have a
trigonometric identity
cos θ = 4 cos3 (θ /3) − 3 cos(θ /3).
Suppose that θ = π/3. Then cos(π/3) = 12 , and cos(π/9) would be a root of the polynomial
8x3 − 6x − 1, which is irreducible over Q since it is irreducible in Z[x]. (It has no roots, even
modulo 2.) But then cos(π/9) would define a degree 3 extension of Q, contradicting Corol-
lary 6.4.6 again.
6.5. Finite fields
In this section, we classify all finite fields, which is to say, fields of finite order.
N OTATION 6.5.1. We use F p to denote Z/pZ when we consider it as a field.
P ROPOSITION 6.5.2. Every finite field contains pn elements for some n ≥ 1.
P ROOF. Let F be a finite field. Since it is finite, it has characteristic p for some prime number
p, which means that it contains the field F p , and moreover is a finite dimensional vector space
over F p . Therefore, F has a finite Z/pZ-basis {α1 , α2 , . . . , αn }, so that the elements of F are
exactly the elements c1 α1 + c2 α2 + · · · + cn αn with c1 , c2 , . . . , cn ∈ Z/pZ. We therefore have
|F| = pn . 
D EFINITION 6.5.3. Let F be a field and n be a positive integer. The group µn (F) of nth roots
of unity in F is the subgroup of F × with elements the nth roots of 1 in F.
L EMMA 6.5.4. Let F be a field and n be a positive integer. Then µn (F) is a cyclic group of
order dividing n.
P ROOF. Every element in µn (F) has order dividing n. Let m be the exponent of µn (F). Then
every element of µn (F) is an mth root of unity, so is a root of xm − 1, and hence the order of
µn (F) is at most m. On the other hand, since m is the exponent, the classification of finite abelian
groups tells us that µn (F) contains an element of order m, so therefore µn (F) is cyclic of order
m, which divides n. 
P ROPOSITION 6.5.5. Let F be a finite field of order pn for some prime p and n ≥ 1. Then F ×
is cyclic, and its multiplicative group is equal to µ pn −1 (F).
n −1
P ROOF. Since |F × | = pn − 1, every element of F × is a root of the polynomial x p − 1, and
conversely. Therefore, it follows from Lemma 6.5.4 that F × = µ pn −1 (F) is cyclic. 
C OROLLARY 6.5.6. The group (Z/pZ)× of units in Z/pZ is cyclic of order p − 1.
E XAMPLE 6.5.7. The cyclic group (Z/17Z)× of order 16 is generated by 3.
L EMMA 6.5.8. Let F be a field of characteristic a prime p, and let α, β ∈ F. Then we have
n n n
(α + β ) p = α p + β p
for all n ≥ 0.
188 6. FIELD THEORY AND GALOIS THEORY

P ROOF. It is easy to see that pi ≡ 0 mod p for 1 ≤ i ≤ p − 1, and so we have the result for


n = 1 by the binomial theorem. By induction, the result for general n follows immediately. 
T HEOREM 6.5.9. Let n be a positive integer. There exists a field F pn of order pn containing
F p , and it is unique up to isomorphism. Moreover, if E is a finite field extension of F p of degree
a multiple of n, then E contains a unique subfield isomorphic to F pn .
n n
P ROOF. Let F be the set of roots of x p −x in a splitting field Ω of x p −x over F p . If α, β ∈ F
n n
are nonzero, then clearly (αβ −1 ) p = αβ −1 , so αβ −1 ∈ F. Moreover, we have (α − β ) p =
n n n n
α p − β p by Lemma 6.5.8, so (α − β ) p = α − β . It follows that F is a field in which x p − x
splits, so it equals Ω.
Now, F has at most pn elements by definition. We must show that has exactly pn elements,
n
so that its degree is n over F p . Clearly x factors into x p − x exactly once. Let a ∈ F × , and set
n n
x p − x p −1 i−1 pn −i
g(x) = = ∑ a x .
x−a i=1
Then we have
pn −1
n −1 n −1
g(a) = ∑ ap = (pn − 1)a p = −1 6= 0,
i=1
n
so x − a is not a factor of g(x), and therefore all roots of x p − x are distinct.
We prove the remaining claims. First, any finite field extension of F p of degree a multiple m
m
of n has pm elements, and Proposition 6.5.5 then implies that it consists of roots of x p − x. In
n
particular, it contains a unique subfield of degree n consisting of the roots of x p − x. Next, note
that F ∼= F[x]/( f ), where f is the minimal polynomial of a generator of µ pn −1 (F). Given any
n
other field F 0 of order pn , it also consists of the roots of x p − x, so contains a root of f . This root
then generates F 0 , being a primitive (pn − 1)th root of unity, so F 0 ∼ = F[x]/( f ) as well. 
R EMARK 6.5.10. Since F pn has order pn and is an F p -vector space, we have [F pn : F p ] = n.
C OROLLARY 6.5.11. The field F pn contains a subfield isomorphic to F pm if and only if m
divides n.
From now on, for a prime p and a positive integer n, we will speak of F pn as being the unique
(up to isomorphism) field of order pn .
E XAMPLE 6.5.12. The field F9 consists of 0 and 8th roots of unity. We have F9 = F3 (ζ ),
where ζ is a primitive 8th root of unity (or even a primitive fourth root of unity), so a root of
x4 + 1. Since [F9 : F3 ] = 2, the minimal polynomial of ζ must be of degree 2. Over F3 , we have
only three irreducible polynomials of degree two: x2 +1, x2 +x −1 and x2 −x −1. The product of
the latter two is x4 + 1, which is to say that the 2 of the primitive 8th roots of unity have minimal
polynomial x2 + x − 1 and the other two x2 − x − 1. On the other hand, we have F9 = F3 (ζ 2 ) as
well, and ζ 2 is a primitive 4th root of unity with minimal polynomial x2 + 1.
The following result is rather useful.
 6.6. CYCLOTOMIC FIELDS 189

P ROPOSITION 6.5.13. Let q be a power of a prime p. Let m ≥ 1, and let ζm be a primitive

mth root of unity in an extension of F p . Then [Fq (ζm ) : Fq ] is the order k of q in (Z/mZ)× . In
other words, we have Fq (ζm ) = Fqk .

P ROOF. Let k = [Fq (ζm ) : Fq ]. Then Fq (ζm ) = Fqk , and so m divides qk − 1, and then q has
order dividing k in (Z/mZ)× . On the other hand, since Fq (ζm ) is not contained in Fq j for any
j < k, we have that q j is not 1 in (Z/mZ)× . That is, q has the desired order k modulo m. 
In order to apply the previous result, it is useful to understand the structure of the unit group
of Z/mZ.
P ROPOSITION 6.5.14. Let m be a positive integer, and write m = pr11 pr22 · · · prkk for distinct
primes pi and positive integers ri for 1 ≤ i ≤ k, for some k ≥ 1. Then
k
(Z/mZ) ∼×
= ∏(Z/pi i Z)× .
r

i=1

Moreover, if p is a prime number and r is a positive integer, we have

(
Z/(p − 1)Z × Z/pr−1 Z if p is odd
(Z/pr Z)× ∼
=
Z/2Z × Z/2r−2 Z if p = 2 and r ≥ 2.

P ROOF. The first statement is a corollary of the Chinese remainder theorem for Z. The
reduction map (Z/pr Z)× → (Z/pZ)× ∼ = Z/(p − 1)Z (noting Corollary 6.5.6) then has kernel
i−1
the multiplicative group (1 + pZ)/(1 + pr Z) of order pr−1 . If p is odd, then (1 + p) p − 1 ≡
pi mod pi+1 by the binomial theorem, so 1 + p has order pr−1 in the group. If p = 2 and r ≥ 2,
then 5 = 1 + 4 similarly generates the subgroup (1 + 4Z)/(1 + 2r Z) of order 2r−2 . Clearly, this
group does not contain −1, which has order 2. That is, (1 + 2Z)/(1 + 2r Z) is generated by the
images of −1 and 5 and so is isomorphic to Z/2Z × Z/2r−2 Z. 

6.6. Cyclotomic fields

Let us explore the extensions of Q generated by roots of unity, known as cyclotomic fields.
N OTATION 6.6.1. Let n ≥ 1. We will use ζn to denote a primitive nth root of unity in an
n/m
extension of Q. We can and therefore do choose these so that ζn = ζm if m divides n. For
instance, one could take ζn = e2πi/n ∈ C.
D EFINITION 6.6.2. Let n ≥ 1. Then nth cyclotomic field is the extension of Q generated by
a primitive nth root of unity ζn .
R EMARK 6.6.3. The nth cyclotomic field Q(ζn ) is Galois over Q. That is, Q(ζn ) is the
splitting field of xn − 1 in that all of the roots of xn − 1 are powers of ζn .
D EFINITION 6.6.4. The nth cyclotomic polynomial Φn is the unique monic polynomial in
Q[x] with roots the primitive nth roots of unity.
190 6. FIELD THEORY AND GALOIS THEORY

Note that Φn lies in Q[x] since every conjugate of a primitive nth root of unity is also a root
of Φn . In Example 5.3.4, we saw that every
Φ p = 1 + x + · · · + x p−1 ,
where p is prime, is irreducible using the Eisenstein criterion.
R EMARKS 6.6.5. Let n be a positive integer.
a. We have
xn − 1 = ∏ Φd ,
d|n
with the sum taken over positive divisors of n.
b. Every conjugate to a primitive nth root of unity is also necessarily a root of xn − 1 that is
not a root of its divisor xm − 1 for any m dividing n, which is to say another primitive nth root of
unity. Therefore Φn as defined lies in Q[x].
c. We have
n
Φn (x) = ∏ (x − ζni ),
i=1
gcd(i,n)=1
and therefore Φn has degree ϕ(n), where ϕ is the Euler-phi function. In particular, we have
deg Φn = ϕ(n).
D EFINITION 6.6.6. The Möbius function µ : Z>0 → {−1, 0, 1} is defined by
(
(−1)k if n is a product of k distinct primes,
µ(n) =
0 otherwise.
We note the following.
L EMMA 6.6.7. For any n ≥ 2, one has ∑d|n µ(d) = 0.
P ROOF. Since µ(d) is zero if d is divisible by a square of a prime, we have ∑d|n µ(d) =
∑d|m µ(d), where m is the product of the primes dividing n. If there are k such primes, then there
are kj products of j of them, each of which contributes (−1) j to the sum. In other words,

k  
k
∑ µ(d) = ∑ j (−1) j = (1 − 1)k = 0,
d|n j=0

since k ≥ 1. 
T HEOREM 6.6.8 (Möbius inversion formula). Let A be an abelian group and f : Z>0 → A a
function. Define g : Z>0 → A by
g(n) = ∑ f (d)
d|n
for n ≥ 1. Then
f (n) = ∑ µ(d)g( dn ).
d|n
 6.6. CYCLOTOMIC FIELDS 191

P ROOF. We calculate
∑ µ( dn )g(d) = ∑ ∑ µ( dn ) f (k) = ∑ ∑ µ( dn ) f (k) = ∑ ∑n µ( kcn ) f (k) = f (n),
d|n d|n k|d k|n d|n k|n c|
k|d k

the last step by Lemma 6.6.7. 

Taking A = Q(x)× , we have the following.
L EMMA 6.6.9. Let n ≥ 1. Then
Φn = ∏ (X n/d − 1)µ(d).
d|n
d≥1
The lemma can be used to calculate cyclotomic polynomials explicitly.
E XAMPLES 6.6.10.
a. We have Φ1 (x) = x − 1.
b. For a prime p and k ≥ 1, we have
k p−1
xp − 1 k−1
Φ pk (x) = k−1 = ∑ xip .
x p − 1 i=0
c. For p and q distinct primes, we have
(x pq − 1)(x − 1) Φq (x p )
Φ pq (x) = p =
(x − 1)(xq − 1) Φq (x)
For instance, taking q = 2 we obtain
xp + 1
Φ2p (x) = = Φ p (−x),
x+1
and we have
Φ15 (x) = x8 − x7 + x5 − x4 + x3 − x + 1.
The nth cyclotomic polynomial is in fact irreducible over Q.
T HEOREM 6.6.11. Let n ≥ 1. Then the cyclotomic polynomial Φn is irreducible in Q[x].
P ROOF. Write Φn = f g with f , g ∈ Z[x] and f monic irreducible with ζ as a root. Take any
prime p not dividing n, and note that ζ p is also a root of Φn .
If ζ p is a root of g, then g(x p ) is divisible by the minimal polynomial f (x) of ζ . Let f¯ and ḡ
denote the reductions modulo p of f and g respectively. Then ḡ(x p ) = ḡ(x) p ∈ F p [x] is divisible
by f¯(x), so ḡ and f¯ have a common factor. The reduction φn = f¯ḡ of Φn modulo p therefore has
a multiple root in F p . In particular, xn − 1 has a multiple root, but we know that it does not. That
is, if we choose k ≥ 1 so that pk ≡ 1 mod n, then the cyclic group F× pk
of order pk − 1 contains n
distinct nth roots of unity.
Thus, ζ p is a root of f for any prime p and any root ζ of f . Since any integer a prime to
n can be written as a product of primes not dividing n, it follows that ζ a is a root of f for all a
prime to p. This forces f = Φn , so Φn is irreducible. 
192 6. FIELD THEORY AND GALOIS THEORY

6.7. Field embeddings

D EFINITION 6.7.1. Let E and E 0 be extensions of a field F, and let ϕ : E → E 0 be an isomor-
phism of fields. We say that ϕ fixes F if ϕ(α) = α for all α ∈ F.
D EFINITION 6.7.2. Let α and β be elements of field extensions of a field F. We say that α
and β are conjugate over F if there exists a field isomorphism ϕ : F(α) → F(β ) fixing F such
that ϕ(α) = β .
P ROPOSITION 6.7.3. Let E and E 0 be extensions of a field F, and let α ∈ E, β ∈ E 0 be
algebraic over F. Then α and β are conjugate over F if and only if the minimal polynomials of
α and β in F[x] are equal.
P ROOF. Suppose that α and β are conjugate over F, and let ϕ : F(α) → F(β ) be a field
isomorphism such that ϕ(α) = β and ϕ restricts to the identity map on F. Then ϕ(g(α)) = g(β )
for all g ∈ F[x]. Let f ∈ F[x] be the minimal polynomial of α. Then we have
0 = ϕ(0) = ϕ( f (α)) = f (β ),
so β is a root of f . As f is irreducible, it must be the minimal polynomial of β .
Conversely, suppose that α and β have the same minimal polynomial f ∈ F[x]. Then we
have isomorphisms from F[x]/( f ) to F(α) and F(β ) as in Theorem 6.1.15, and composing the
inverse of the first with the latter yields the desired isomorphism F(α) → F(β ). 
E XAMPLE 6.7.4. Since i and −i are both roots of the irreducible polynomial x2 + 1 over R,
they are conjugate elements of C. Therefore, there is a field isomorphism C → C that takes i to
−i and fixes R. Such an isomorphism must take a + bi to its complex conjugate
a + bi = a − bi
and is therefore the usual complex conjugation. In particular, complex conjugation is an isomor-
phism of fields, which is also easily verified directly. Moreover, we see that if f ∈ R[x] has a root
α, then ᾱ is a root as well, since f (α) = f (ᾱ).
D EFINITION 6.7.5. An embedding of fields, or field embedding, is an injective ring homo-
morphism ϕ : F → F 0 , where F and F 0 are fields.
R EMARK 6.7.6. Any nonzero ring homomorphism between fields is injective, so “injective”
can be replaced by “nonzero” in the definition of a field embedding.
D EFINITION 6.7.7. Let ϕ : F → M be a field embedding, and let E/F be an extension field.
We say that a field embedding Φ : E → M extends ϕ, and is an extension of ϕ, if Φ|E = ϕ.
E XAMPLE
√ 6.7.8. We have a field embedding ι : Q√→ R. There√are two field embeddings
ι0 :Q( √ 2) → R extending
√ ι. Either we take ι 0 (a + b 2) = a + b 2 for a, b ∈ Q, or we set
0
ι (a + b 2) = a − b 2. On the other hand, there is no field embedding κ : Q(i) → R extending
ι, since there is no element of i that would satisfy κ(i)2 + 1 = 0, but there is no element of R
with this property.
Let us give a slight extension of one direction of Proposition 6.7.3.
 6.7. FIELD EMBEDDINGS 193

T HEOREM 6.7.9. Let E/F be a field extension, and let α ∈ E be algebraic over F. Let
ϕ : F → M be a field embedding, and consider the induced map ϕ̃ : F[x] → M[x]. Let f ∈ F[x]
be the minimal polynomial of α. Then there is a bijection between the set of field embeddings
F(α) → M extending ϕ and the set of roots of ϕ̃( f ) in M taking an extension Φ of ϕ to Φ(α).
P ROOF. Suppose that β is a root of ϕ̃( f ). Let eβ : M[x] → M denote the evaluation map
at β . The composition eβ ◦ ϕ̃ has kernel containing ( f ), and the kernel then equals ( f ) by the
maximality of ( f ) and the fact that the composition is nonzero. The first isomorphism theorem
yields a field embedding F[x]/( f ) → M sending the coset of x to β . The map Φ is then obtained
by composing with the isomorphism F(α) → F[x]/( f ) of Theorem 6.1.15, and it sends α to β .
Moreover, if κ is any other lift of ϕ such that κ(α) = β , we have
! !
deg f −1 deg f −1 deg f −1
κ ∑ ci α i = ∑ ϕ(ci )β i = Φ ∑ ci α i
i=0 i=0 i=0
for all ci ∈ F for 1 ≤ i < deg f , so κ = Φ.
Conversely, suppose Φ : F(α) → M is an extension of ϕ. Then we have ϕ̃( f )(Φ(α)) =
Φ( f (α)) = 0. 
C OROLLARY 6.7.10. Let E/F be a field extension, let α ∈ E be algebraic over F, and let
ϕ : F → M be a field embedding. Let ϕ̃ : F[x] → M[x] denote the induced map on polynomial
rings. The number of extensions of ϕ to an embedding Φ : F(α) → M is the number of distinct
roots of ϕ̃( f ) in M, where f ∈ F[x] is the minimal polynomial of α.
R EMARK 6.7.11. In the setting of Corollary 6.7.10, we may identify F with its isomorphic
image ϕ(F). This allows us to think of F as a subfield of M. In this case, f ∈ F[x] may be
thought of as itself having roots in M, and the number of embeddings of F(α) in M is the
number of distinct roots of f in M.
In general, for finite extensions, we have the following.
C OROLLARY 6.7.12. Let E/F be a finite extension of fields. Let ϕ : F → M be a field em-
bedding. Then the number of extensions Φ : E → M of F is finite, less than or equal to [E : F].
P ROOF. Since any finite extension E/F is finitely generated, it suffices by the multiplicativity
of degrees of field extensions in Corollary 6.2.12 and recursion to prove the result in the case that
E = F(α) for some α ∈ E. In this case, the degree of the minimal polynomial of α is equal
to [E : F] and is greater than or equal to the number of distinct roots in M of the image of the
minimal polynomial of α. The result is therefore a consequence of Corollary 6.7.10. 
√
E XAMPLE 6.7.13. As seen in Example 6.7.8, there are exactly two embeddings of Q( 2) in
R, but no embeddings of Q(i) in R.
√ √
E XAMPLE 6.7.14. √There are√four embeddings
√ of
√ Q( 2, 3) in R. If ϕ is such an embed-
ding, then we have ϕ( 2) = ± 2 and ϕ( 3) = ± 3, and the signs determine the embedding
uniquely.
Finally, we note the following.
194 6. FIELD THEORY AND GALOIS THEORY

P ROPOSITION 6.7.15. Let E/F be an algebraic field extension, and let σ : E → E be a field
embedding fixing F. Then σ is an isomorphism.
P ROOF. Let β ∈ E, and let f ∈ F[x] be its minimal polynomial. By Proposition 6.7.3, every
root of f in E is sent by σ to another root of f in E. As σ is injective and the set of roots of
f in E is finite, σ permutes these roots. In particular, there exists a root α of f in E such that
σ (α) = β . Therefore, we have σ (E) = E, as desired. 

6.8. Algebraically closed fields

We begin with the notion of an algebraically closed field.
D EFINITION 6.8.1. A field L is algebraically closed if contains a root of every nonconstant
polynomial f ∈ L[x].
The following theorem has analytic, topological, geometric, and algebraic proofs (though all
in a sense require some very basic analysis).
T HEOREM 6.8.2 (Fundamental theorem of algebra). The field C of complex numbers is alge-
braically closed.
We defer an algebraic proof of this theorem until after our treatment of Galois theory. For the
reader’s enjoyment, here are sketches of three proofs which require some knowledge of subjects
outside of this course. The first two use complex analysis:
R EMARK 6.8.3. if p ∈ C[x] has no roots, then p−1 is homolorphic and bounded as a function
on C, hence constant by the maximum modulus principle.
R EMARK 6.8.4. A nonconstant polynomial p ∈ C[x] defines a nonconstant continuous map
from the Riemann sphere P1 (C) to itself. Its image is closed as P1 (C) is compact Hausdorff,
while its image is open by the holomorphicity of p and the open mapping theorem, so the image
is P1 (C).
Next, algebraic topology:
R EMARK 6.8.5. Suppose that p ∈ C[x] is monic of degree n, and choose r > 0 such that
|p(z) − zn | < rn for all z ∈ C with |z| = r. The map F : S1 → S1 with
p(rz)
F(z) =
|p(rz)|
is homotopic to z 7→ zn by a homotopy H : [0, 1] × S1 → S1 given by
(1 − t)(p(rz) − (rz)n ) + (rz)n
H(t, z) = .
|t p(rz) + (1 − t)(rz)n |
Now F extends to a map C → S1 on the simply connected space C by the same formula, so
induces the zero map on π1 (S1 ) ∼
= Z. But z 7→ zn induces multiplication by n on π1 (S1 ), so n = 0.
P ROPOSITION 6.8.6. Let L be an algebraically closed field, and let f ∈ L[x] be nonconstant.
Then f splits in L.
 6.8. ALGEBRAICALLY CLOSED FIELDS 195

P ROOF. We prove this by induction, as it is clear for deg f = 1. Suppose we know the result
for all polynomials of degree less than n = deg f . Since f has a root α in L, we have f = (x − α)g
for some g ∈ L[x] of degree n − 1. By induction, g factors into linear terms. 
C OROLLARY 6.8.7. Let M be an algebraic extension of an algebraically closed field L. Then
M = L.
P ROOF. Let α ∈ M. As M is algebraic over L, there exists a nonconstant f ∈ L[x] with
f (α) = 0, and by Proposition 6.8.6, the polynomial f is divisible by x − α in L[x] (recalling that
M[x] is a UFD). Therefore, we have α ∈ L. 
We next show that extensions of field embeddings into algebraically closed fields always
exist, when the extension is algebraic.
T HEOREM 6.8.8. Let E/F be an algebraic extension of fields. Let ϕ : F → M be a field
embedding, where M is an algebraically closed field. Then there exists a field embedding Φ : E →
M extending ϕ.
P ROOF. Let X denote the nonempty set of all pairs (K, σ ), where K is an intermediate sub-
field of E/F and σ : K → M is an extension of ϕ. We say that (K, σ ) ≤ (K 0 , σ 0 ) for (K, σ ) and
(K 0 , σ 0 ) ∈ X if K 0 contains K and σ 0 |K = σ . Let C be a chain in X, set
[
L= K
(K,σ )∈C

and define τ : L → M by τ|K = σ for all (K, σ ) ∈ C. It is easy to see that τ is a well-defined field
embedding, since C is a chain, and therefore, (L, τ) ∈ X is an upper bound for C.
By Zorn’s lemma, we therefore have that X contains a maximal element, which we call
(Ω, λ ). We claim that E = Ω. To see this, let α ∈ E, and let f ∈ Ω[x] be the minimal polynomial
of α over Ω. If f = ∑ni=0 ai xi with ai ∈ Ω for 0 ≤ i ≤ n and n = deg f , then we set
n
g = ∑ λ (ai )xi .
i=0
Since M is algebraically closed, g has a root β in M. By Proposition 6.7.9, we may then extend
λ to an embedding λ 0 : Ω(α) → M. We then have (Ω, λ ) ≤ (Ω(α), λ 0 ), and the maximality of
Ω forces E = Ω. Setting Φ = λ , we are done. 
P ROPOSITION 6.8.9. The set of all algebraic elements over a field F in an extension E is a
subfield of E, and it is equal to the the largest intermediate extension of E/F that is algebraic
over F.
P ROOF. Let M denote the set of all algebraic elements over F in E, and let α, β ∈ M. Then
F(α, β )/F is a finite extension, so every element of it is algebraic. In particular, α − β and, if
β 6= 0, the element αβ −1 are elements of F(α, β ), so they are algebraic elements over F, hence
contained in M. Therefore, M is a field. The second statement is then an immediate consequence
of the definition of M. 
C OROLLARY 6.8.10. The set Q of algebraic numbers in C forms a field.
196 6. FIELD THEORY AND GALOIS THEORY

D EFINITION 6.8.11. An algebraic closure of a field F is an algebraically closed, algebraic

extension of F.
R EMARK 6.8.12. Since an algebraic closure is algebraic, every element of the the algebraic
closure of a field F has to be the root of a polynomial with F-coefficients. On the other hand,
since F is algebraically closed, it contains all roots of every polynomial with coefficients in F
(i.e., every polynomial in F[x] factors completely). Thus, if an algebraic closure exists, and we
shall see that it does, it consists exactly of all roots of polynomials in F, and every root of a
polynomial with coefficients in F is actually the root of a polynomial with coefficients in F.
In fact, if a field is contained in an algebraically closed field, then we can see that it does in
fact have an algebraic closure quite directly.
P ROPOSITION 6.8.13. Let F be a field, and suppose that M is an algebraically closed exten-
sion field of F. Then M contains a unique algebraic closure of F, equal to the field of elements
of M that are algebraic over F.
P ROOF. Let F denote the field consisting of all elements of M that are algebraic over F. We
claim that F is algebraically closed. For this, suppose that f ∈ F[x], and α ∈ M be a root. As α
is algebraic over F, we have by Proposition 6.2.19 that α is also algebraic over F. That is, α is
an element of F. 
C OROLLARY 6.8.14. The field Q of algebraic numbers in C is an algebraic closure of Q.
Using Zorn’s lemma, we may prove that every field has an algebraic closure. This is the first
result on extension fields in which we do not have a previously given field that contains the field
of interest, which makes the proof rather more tricky.
T HEOREM 6.8.15. Every field F has an algebraic closure.
P ROOF. Let F be a field. Let Ω be a set that is the disjoint union of finite sets R f for each
monic irreducible f ∈ F[x], where the number of elements in R f is the number of distinct roots of
f in a splitting field. (We will end up identifying the elements of R f with roots of f , but they do
not start as such.) We may view F as a subset of Ω by identifying a ∈ F with the unique element
of Rx−a . Let X be the nonempty set of all algebraic extensions of F, the underlying sets of which
are contained in Ω in the sense if E ∈ X, then every α ∈ E lies in R f for f ∈ F[x] the minimal
polynomial of α. We put a partial ordering on X by E ≤ E 0 for E, E 0 ∈ X if and only if E ⊆ E 0
and E 0 /E is a field extension.
Let C be chain in X, and let K be the union of the fields in C . Then K is a field, as any two
elements α, β ∈ K satisfy α, β ∈ E for some E ∈ C (taking the larger of the two fields in which
α and β are contained by definition), and then α − β ∈ E and αβ −1 ∈ E if β 6= 0. Since K ∈ X,
the chain C has an upper bound, and we may apply Zorn’s lemma to the set X to find a maximal
element F ∈ X.
Let f ∈ F[x], and let g ∈ F[x] be a monic, nonconstant irreducible polynomial dividing f .
We then have that E = F[x]/(g) is an extension of F that is algebraic over F. We may view
the underlying set of E as being contained in Ω as follows. If h ∈ F[x] is a monic irreducible
polynomial with a root in E, we may identify those of its distinct roots in E that are not contained
 6.9. TRANSCENDENTAL EXTENSIONS 197

in F with distinct elements of Rh that are not in F. Since F ∈ X is maximal, we must have E = F.
In particular, f must factor completely in F[x].
It remains only to show that F is algebraically closed. Any root β of an irreducible polyno-
mial g ∈ F[x] in an extension of F is algebraic over F by Proposition 6.2.19. Therefore, g divides
the minimal polynomial f ∈ F[x] of β , which by the argument we have just given splits over F.
In particular, we have β ∈ F. 
We next remark that the algebraic closure of any field is in fact unique up to isomorphism.
P ROPOSITION 6.8.16. Let M and M 0 be algebraic closures of a field F. Then there exists an
isomorphism Φ : M → M 0 fixing F.
P ROOF. Theorem 6.8.8 applied to the case that ϕ = idF , E = M, and M = M 0 implies that
there exists a field embedding Φ : M → M 0 extending F. To see that it is an isomorphism, note
that the image of Φ is algebraic over F, being contained in M 0 , and algebraically closed over
F since a root of a polynomial in F[x] in M maps under Φ to a root of the same polynomial.
Therefore, Φ(M) is an algebraic closure of F contained in M 0 , and hence must be M 0 itself. 
R EMARK 6.8.17. As any two algebraic closures of a field F are isomorphic via an isomor-
phism that fixes F, we usually refer to “the” algebraic closure of F, denoting it by F.
R EMARK 6.8.18. If E is an algebraic extension of F and E is the algebraic closure of E, then
it is also an algebraic closure of F. In particular, there exists an algebraic closure of F containing
E.
6.9. Transcendental extensions
D EFINITION 6.9.1. A field extension E/F is totally transcendental if every element of E − F
is transcendental over E.
T ERMINOLOGY 6.9.2. For a ring R and an indexing set I, we may speak of the polynomial
ring R[(xi )i∈I ] in the variables xi for i ∈ I. It is simply the union over all finite lists i1 , . . . , in of
distinct elements of I of the polynomial rings R[xi1 , . . . , xin ], with the operations being induced
by the operations on these rings. If R is commutative, then the rational function field R((xi )i∈I )
is the fraction field of R[(xi )i∈I ]. This field is itself the union of the rational function fields
R(xi1 , . . . , xin ).
P ROPOSITION 6.9.3. For any indexing set I, the field F((ti )i∈I ) of rational functions in the
variables ti for i ∈ I is purely transcendental over F.
P ROOF. Consider first the extension F(t)/F given by the F-rational function field in a single
variable t. Let α = gf ∈ F(t) − F, where f , g ∈ F[t] and g 6= 0. We may view f (x) and g(x) as ele-
ments of F[x]. Then α · g(x) ∈ F(α)[x] is not an element of F[x], so the polynomial f (x) − αg(x)
is nonzero but does have a root t, which is then algebraic over F(α). Since t is transcendental
over F, this forces α to be as well. This gives the result for a single variable, and the case of
finitely many variables follows immediately by induction. Since F((ti )i∈I ) is the union of the
rational function fields F(ti1 , . . . ,tin ) with i1 , . . . , in ∈ I, the case of finitely many variables yields
the general case. 
198 6. FIELD THEORY AND GALOIS THEORY

P ROPOSITION 6.9.4. Every extension of fields is a totally transcendental extension of an

algebraic extension.
P ROOF. Given a field extension K/F, we may consider its subfield E of elements algebraic
over F. If α ∈ K − F, then α cannot be algebraic over E. That is, if it were, then it would also
be algebraic over F in that E/F is algebraic. 
D EFINITION 6.9.5. Let K/F be a field extension.
a. We say that a collection (αi )i∈I of elements of K, is algebraicaly independent over F, if
f (αi1 , . . . , αin ) 6= 0 for all nonzero polynomials f ∈ F[x1 , . . . , xn ] and distinct elements i1 , . . . , in
for some n ≥ 1.
b. We say that a subset S of K is algebraically independent over F, or F-algebraically in-
dependent, if f (s1 , . . . , sn ) 6= 0 for all nonzero polynomials f in n variables over F and distinct
s1 , . . . , sn ∈ S for some n ≥ 1.
Here are a couple of straightfoward lemmas.
L EMMA 6.9.6. Let K/F be a field extension, and let S ⊂ K be algebraically independent over
F. Then t ∈ K is transcendental over the field F(S) generated by S over F if and only if S ∪ {t}
is algebraically independent over F.
L EMMA 6.9.7. . A subset S of a field extension K of F is algebraically independent over F if
and only if each s ∈ S is transcendental over K(S − {s}).
D EFINITION 6.9.8. A subset S of an extension K of a field F is a transcendence basis of K/F
if and only if S is algebraically independent over F and K is algebraic over F(S).
The following equivalent conditions for being a transcendence basis nearly mimic the usual
equivalent conditions for a subset of a vector space to be a basis. (That is, a subset is a basis if
and only if it is a maximal linearly independent subset and if and only if it is a minimal spanning
set.)
P ROPOSITION 6.9.9. Let S be a subset of an extension K of a field F. The following are
equivalent:
i. S is a transcendence basis of K/F,
ii. S is a maximal F-algebraically independent subset of K,
iii. S is a minimal subset of K such that K is algebraic over F(S).
P ROOF. The equivalence of (i) and (ii) is a direct consequence of Lemma 6.9.6, and the
equivalence of (i) and (iii) is a direct consquence of Lemma 6.9.7. 
T HEOREM 6.9.10. Every F-algebraically independent subset of an extension K/F is con-
tained in a transcendence basis, and every subset of K that generates an extension over which K
is algebraic contains a transcendence basis.
P ROOF. Let A be an F-algebraically independent subset of K. Let X be the set of F-
algebraically independent subsets of K containing A, ordered by inclusion. We may take the
 6.10. SEPARABLE EXTENSIONS 199

union of any chain C in X, and it is F-algebraically independent in that any finitely many ele-
ments of the union on which we would test algebraic independence is contained in some element
of the chain. This union is an upper bound, and thus by Zorn’s lemma, X contains a maximal
element B. To finish the proof, we need only see that K is algebraic over F(B). But this is
clear, since if t ∈ K − B is transcendental over F(B), then B ∪ {t} is F-algebraically independent,
contradicting the maximality of B.
Now, let S ⊆ K be such that K/F(S) is algebraic. Consider the set Y of F-algebraically
independent subsets of K contained in S, again ordered by inclusion. Every chain has an upper
bound as before, so Y contains a maximal element T . We need only see that K is algebraic over
F(T ). If not, then since K is algebraic over F(S) and T ⊂ S, we must have that there exists
s ∈ S − T that is transcendental over F(T ), and then T ∩ {s} ∈ Y , contradicting the maximality
of T . 
C OROLLARY 6.9.11. Every extension of fields K/F has an intermediate field E such that
K/E is algebraic and E/F is totally transcendental.
We omit a proof of the following.
T HEOREM 6.9.12. If S and T are transcendence bases of an extension K/F, then S and T
have the same cardinality.
In particular, we may make the following definition.
D EFINITION 6.9.13. We say that a field extension K/F has finite transcendence degree if it
has a finite transcendence basis, in which case the number of elements in a transcendence basis is
called the transcendence degree. Otherwise, we say that K/F has infinite transcendence degree.

6.10. Separable extensions

D EFINITION 6.10.1. Let F be a field. Let f ∈ F[x] be nonzero, and let α ∈ F be a root of f .
The multiplicity of α as a root of f is the largest positive integer m such that (x − α)m divides f
in F[x].
E XAMPLE 6.10.2. Let f = x p − t ∈ F p (t)[x], which is irreducible. In F p (t)[x], we have
f = x p − t = (x − t 1/p ) p ,
so t 1/p has multiplicity p as a root of f .
L EMMA 6.10.3. Let F be a field, and let f ∈ F[x] be irreducible. Then every root of f in an
algebraic closure F of F has the same multiplicity.
P ROOF. Let α, β ∈ F be roots of f . Fix an field isomorphism σ : F(α) → F(β ) taking α to
β , and extend it to an embedding τ : F → F. Let τ̃ : F[x] → F[x] map induced by τ. If m denotes
the multiplicity of α, then
τ̃((x − α)m ) = (x − β )m .
Since (x − α)m divides f in F[x] and τ̃( f ) = f , the multiplicity of β is then at least m, but this
was independent of the choice of α and β , so α and β have the same multiplicity. 
200 6. FIELD THEORY AND GALOIS THEORY

C OROLLARY 6.10.4. Let F be a field. The number of distinct roots of an irreducible polyno-
mial f ∈ F[x] in an algebraic closure F of F divides the degree of f .
D EFINITION 6.10.5. Let F be a field. We say that a nonconstant polynomial f ∈ F[x] is
separable if every root of f has multiplicity 1.
D EFINITION 6.10.6. Let F be a field and F be an algebraic closure of F. An element α ∈ F
is separable over F if and only if its minimal polynomial is separable over F.
D EFINITION 6.10.7. We say that an algebraic extension E/F is separable if every α ∈ E is
separable over F.
N OTATION 6.10.8. Let K and L be extensions of a field F. We will denote the set of field
embeddings of K into L that fix F by EmbF (K, L). If K is algebraic over F and L is taken to be
a fixed algebraic closure of F, we will simply write EmbF (K) (despite the dependence on the
algebraic closure).
L EMMA 6.10.9. Let E/F be a field extension, and let α ∈ E be algebraic over F. Then α is
separable over F if and only if F(α)/F is separable.
P ROOF. We prove the nontrivial direction, which results from several applications of Theo-
rem 6.7.9. Fix an algebraic closure F of F. For a given β ∈ F(α), the number e = | EmbF (F(β ))|
is at most the degree [F(β ) : F], with equality if and only if β is separable. Since α is separable
over F, we have
| EmbF (F(α))| = [F(α) : F].
Moreover, α is separable over F(β ) as well, since its minimal polynomial over F(β ) divides its
minimal polynomial over F. Thus, the number of embeddings of F(α) in F extending a given
embedding of F(β ) into F is exactly [F(α) : F(β )]. Therefore, we have that
[F(α) : F] = [F(α) : F(β )]e,
which means that e = [F(β ) : F], so β is separable. 
L EMMA 6.10.10. Let E be an algebraic extension of a field F, and let K be an algebraic
extension of E. If K/F is separable, then so are K/E and E/F.
P ROOF. Suppose that K/F is separable. By definition, if α ∈ E, then α ∈ K, so its minimal
polynomial over F is separable. Moreover, the minimal polynomial of any β ∈ K over E divides
the minimal polynomial of β over F, so β is separable over E. 

We also have the following.

P ROPOSITION 6.10.11. Let E/F be a finite extension. Fix an algebraic closure F of F.
a. The number of embeddings of E into F that fix F divides [E : F].
b. The number of embeddings of E into F that fix F is equal to [E : F] if and only if E/F is
separable.
 6.10. SEPARABLE EXTENSIONS 201

P ROOF. Let e denote the number of embeddings of E in F. Write E = F(α1 , α2 , . . . , αn ), and

let Ei = F(α1 , α2 , . . . , αi−1 ) for 1 ≤ i ≤ n+1. Then Ei+1 = Ei (αi ) for i ≤ n, and by Theorem 6.7.9,
the number ei of embeddings of Ei+1 into F extending an embedding ϕi of Ei into F is the number
of distinct roots of the minimal polynomial of αi over Ei . This number, in turn, is a divisor of
[Ei+1 : Ei ], with equality if and only if αi is separable over Ei . Since
n
e = ∏ ei
i=1
and
n
[E : F] = ∏[Ei+1 : Ei ],
i=1
we therefore have that e divides [E : F], with equality if and only if ei = [Ei+1 : Ei ] for each i, and
in particular, noting Lemma 6.10.10, if E/F is separable.
Conversely, suppose that e = [E : F]. For β ∈ E, the number of distinct roots c of its minimal
polynomial is the number of embeddings of F(β ) into F fixing F. By the above argument, the
number of embeddings d of E into F extending one of those embeddings divides [E : F(β )], and
we have e = cd, so we must have c = [F(β ) : F]. That is, E/F is separable. 
P ROPOSITION 6.10.12. Let K be an algebraic extension of a field F, and let E be an inter-
mediate field in K/F. Then K/F is separable if and only if K/E and E/F are.
P ROOF. By Lemma 6.10.10, we are reduced to showing that if K/E and E/F are separable,
then K/F is separable. Proposition 6.10.11 implies this immediately if K/F is finite. In general,
take α ∈ K, and note that any minimal polynomial g of α over E actually has coefficients in some
finite subextension E 0 of E, in that E/F is algebraic. Then E 0 (α)/E 0 is separable since g is, and
E 0 /F is separable by Lemma 6.10.10. As E 0 (α)/F is finite, we have the result. 
D EFINITION 6.10.13. We say an extension E/F is purely inseparable if E contains no non-
trivial separable subextensions of F.
Proposition 6.10.12 tells us that it suffices to check the separability of an extension on a
generating set. It also implies the following.
C OROLLARY 6.10.14. Let K/F be an algebraic extension. The set E of all separable ele-
ments in K/F is a subfield of K. Moreover, the extension K/E is purely inseparable.
D EFINITION 6.10.15. Let K/F be a finite extension, and let E be the maximal separable
subextension of F in K.
i. The degree of separability [K : F]s of K/F is [E : F].
ii. The degree of inseparability [K : F]i of K/F is [K : E].
Finally, let us investigate circumstances under which all finite extensions of a given field are
separable.
D EFINITION 6.10.16. A field F is perfect is every finite extension of it is separable.
202 6. FIELD THEORY AND GALOIS THEORY

E XAMPLE 6.10.17. The field F p is perfect. To see this, recall the field F pn for n ≥ 1 is equal
n
to the set of roots of the polynomial x p − x, which are all distinct (since there need to be pn
n
of them). Since the minimal polynomial of any α ∈ F pn , divides x p − x, that polynomial is
separable, and therefore F pn /F p is separable.
L EMMA 6.10.18. Let E/F be an algebraic field extension. Let f ∈ E[x] be monic, and let
m ≥ 1 be such that f m ∈ F[x]. Then, either m = 0 in F or f ∈ F[x].
P ROOF. Suppose that f ∈ / F[x]. Write f = ∑ni=0 ai xi with n = deg f and an = 1. Let i ≤
n − 1 be maximal such that ai ∈ / F. The coefficient c of x(m−1)n+i in f m is a polynomial in
the coefficients ai , ai+1 , . . . , an−1 such that c − mai is a polynomial in ai+1 , . . . , an−1 , which are
elements of F. Since c ∈ F, we have mai ∈ F, which forces either m = 0 in F or ai ∈ F. 
T HEOREM 6.10.19. Let F be a field of characteristic 0. Then F is perfect.
P ROOF. If f ∈ F[x] is irreducible, then every root of f in an algebraic closure F occurs with
some multiplicity m ≥ 1. It follows that
d
f = ∏(x − αi )m
i=1
for some d ≥ 1 and distinct α1 , α2 , . . . , αm ∈ F, so f = gm for some g ∈ F[x]. Since the charac-
teristic of F is zero, Lemma 6.10.18 tells us that m = 1. 
The following tells us that the degree of inseparability of a finite field extension is the power
of the characteristic of the fields.
P ROPOSITION 6.10.20. Let F be a field of characteristic p. If E/F is purely inseparable
k
and α ∈ E, then α p ∈ F for some minimal k ≥ 0, and the minimal polynomial of α over F is
k k k
x p − α p = (x − α) p .
P ROOF. Fix an algebraic closure F of F containing E. Let f ∈ F[x] be the minimal polyno-
mial of some element α of E not in F. Again we have
d
f = ∏(x − αi )m
i=1
for some d ≥ 1 and distinct α1 , α2 , . . . , αm ∈ F, so f = gm for some g ∈ F[x]. We must show that
m is a p-power and d = 1.
k k
Write m = pkt with p - t and k ≥ 1. The fact that f = (g p )t ∈ F[x] forces g p ∈ F[x] by
Lemma 6.10.18. Since f is irreducible, we have t = 1.
k
Now set ai = αip and write
d
k
f = ∏(x p − ai ).
i=1
k
Then f (x) = h(x p )
for h = ∏di=1 (x − ai ).
The polynomial h lies in F[x] since it has the same set
of coefficients as f , it is irreducible as any factorization of h would give rise to a factorization
 6.10. SEPARABLE EXTENSIONS 203

k
of f , and it has α p as a root. Also, the ai are distinct elements, since there are no nontrivial
pk th roots of unity in a field of characteristic p, which tells us that raising to the pk th power is
injective. As E/F is purely inseparable and any root of h generates a separable extension of F,
we must have d = 1. 
C OROLLARY 6.10.21. Let F be a field of characteristic p, and let E/F be a finite extension.
Then [E : F]i is a power of p.
We then have the following.
P ROPOSITION 6.10.22. The degree of separability [K : F]s of a finite extension K/F is equal
to the number of embeddings of K fixing F into a given algebraic closure of F.
P ROOF. Let E be the maximal separable subextension of F in K. We know that there are
n
[K : F]s elements of EmbF (E). Any α ∈ K − E has minimal polynomial (x − α) p over E for
some n ≥ 1, so α has only one conjugate over E in K. Thus, any ϕ ∈ EmbF (E) extends uniquely
to an embedding of E(α) in F. Replacing E by E(α) and repeating this last argument, we obtain
recursively that ϕ has a unique extension to all of K. Since every element of EmbF (K) is an
extension of its restriction to E, the number of such elements is [K : F]s . 
We have the following multiplicativity of separable and inseparable degrees.
L EMMA 6.10.23. Let K/F be a finite extension and E an intermediate field in K/F. Then
[K : F]s = [K : E]s [E : F]s and [K : F]i = [K : E]i [E : F]i .
P ROOF. By the multiplicativity of degrees of field extensions, it suffices to consider separable
degrees. It also suffices by recursion to consider this in the case that K can be generated over E
by a single element α. Fix an algebraic closure F̄ of F. Given a field embedding of E into F̄
fixing F, the number of extensions of it to K = E(α) for any α ∈ K is [K : E]s by Corollary 6.7.10
and Proposition 6.10.22. The number of such embeddings being [E : F]s , we have the result. 
Finally, we show that finite separable extensions can be generated by a single element.
D EFINITION 6.10.24. We say that a finite field extension E/F is simple if there exists α ∈ E
such that E = F(α). In that case, α is said to be a primitive element for E/F.
T HEOREM 6.10.25 (Primitive element theorem). Every finite, separable field extension is
simple.
P ROOF. Note that if F is finite, then it is isomorphic to F pn for some prime p and n ≥ 1, and
by Proposition 6.5.5, it equals F p (ξ ) for some primitive (pn − 1)th root of unity in F. So we may
assume that F is infinite.
Since every finite extension is finitely generated by Corollary 6.2.10, it suffices by recursion
to show that if E/F is a finite field extension with E = F(α, β ) for some α, β ∈ E, then there
exists c ∈ F such that E = F(α + cβ ).
Since F is infinite, we can and do choose c ∈ F such that
α0 − α
c 6= − 0
β −β
204 6. FIELD THEORY AND GALOIS THEORY

for all conjugates α 0 of α over F with α 0 6= α and all conjugates β 0 of β over F with β 0 6= β . Set
γ = α + cβ . Then γ 6= α 0 + cβ 0 for all α 0 and β 0 as above. Let f be the minimal polynomial of α,
and let h(x) = f (γ − cx) ∈ F(γ)[x]. Then h(β ) = f (α) = 0 and h(β 0 ) 6= 0 for β 0 . Since h shares
the root β with the minimal polynomial g of β over F, but not any other root, and the minimal
polynomial q of β over F(γ) divides both of the latter polynomials, we must have q = x − β ,
which is to say that β ∈ F(γ), which then implies that α ∈ F(γ) as well. We therefore have
F(γ) = F(α, β ), as desired. 
R EMARK 6.10.26. Much as with algebraic closure, we have the notion of a separable closure
of a field. A field L is separably closed if it contains a root of every monic, separable polynomial
with coefficients in L. Algebraically closed fields are therefore separably closed. A separable
closure of a field F is a separable extension F sep of F that is separably closed. If F is a subfield
of any separably closed field L, the set of all roots in L of all monic, separable polynomials in
F[x] is a subfield that is a separable closure of F. Separable closures exist: in fact, given a field
F, take an algebraic closure F of F, and it then contains a separable closure F sep , which is the
union of all finite separable subextensions of F in F. Of course, if F is perfect, then the notions
of separable closure and algebraic closure of F coincide.

6.11. Normal extensions

We extend the definition of a splitting field to include sets of polynomials.
D EFINITION 6.11.1. Let F be a field, and let S be a subset of F[x] consisting of nonconstant
polynomials. A splitting field E for S over F is an extension of F such that every polynomial in
S splits in E and which contains no proper subextension of F in which this occurs.
√ √
E XAMPLE 6.11.2. The field Q( 2, 3) is the splitting field of {x2 − 2, x2 − 3}. It is then
also the splitting field of (x2 − 2)(x2 − 3).
E XAMPLE 6.11.3. An algebraic closure F of a field F is a splitting field of the set of all
nonconstant polynomials in F[x].
R EMARK 6.11.4. An algebraic closure F of a field F will always contain a unique splitting
field for any subset S of F[x]. This field is equal to the intersection of all subfields of F in which
every polynomial in S splits.
D EFINITION 6.11.5. We say that an algebraic field extension E/F is normal if E is the
splitting field of some set of polynomials in F[x].
L EMMA 6.11.6. If E/F is normal, then so is E/F 0 , where F 0 is any intermediate field in
E/F.
P ROOF. If E is the splitting field of a set S of polynomials in F[x], then by definition it
contains the splitting field of the set S over F 0 . If there were a proper subfield of E containing
F 0 in which all the polynomials in S split, then E would not be a splitting field over F, so E is a
splitting field of S over F 0 as well. 
 6.11. NORMAL EXTENSIONS 205

T HEOREM 6.11.7. An algebraic field extension E/F is normal if and only if every field em-
bedding Φ of E that fixes F into an algebraic closure F of F containing E satisfies Φ(E) = E.
Moreover, under these conditions, E is equal to the splitting field over F of the set of minimal
polynomials over F of every element of E.
P ROOF. Suppose first that E/F is normal, and let S ⊆ F[x] be a set of polynomials of which
E is a splitting field. Let Φ ∈ EmbF (E). By definition, E is generated over F by the roots of
all polynomials in S. Let f ∈ S, and let α ∈ E be a root. By Theorem 6.7.9, we must have
that Φ(α) = β , where β is a root of f in F. But every root of f in F lies in the subfield E,
since f splits in E, so Φ(α) ∈ E. As every element of E may be written as a rational function
in the roots of polynomials in S with coefficients in F, we therefore have Φ(E) ⊆ E. Noting
Proposition 6.7.15, we then have that Φ(E) = E.
Conversely, suppose that Φ(E) = E for every Φ ∈ EmbF (E). Let α ∈ E, and let f be its
minimal polynomial over F. Then for any root β ∈ F of f , we have an isomorphism ϕ : F(α) →
F(β ) sending α to β . We may then extend the resulting embedding F(α) → F to an embedding
Φ : E → F. Since Φ(E) = E, we therefore have β ∈ E. So E contains the splitting field of every
polynomial of F that has a root in E. Since E is algebraic and therefore consists entirely of roots
of polynomials in F, it is therefore equal to said splitting field. 
C OROLLARY 6.11.8. Let E/F be a normal field extension, and let f ∈ F[x] be an irreducible
polynomial that has a root in E. Then f splits in E.
P ROOF. This follows directly from the final statement of Theorem 6.11.7. 
For composite extensions, we have the following.
P ROPOSITION 6.11.9. Let F be a field and F an algebraic closure of F. Suppose that E and
K are subfields of F that are normal over F. Then EK/F is normal as well.
P ROOF. We note that any ϕ ∈ EmbF (EK) restricts to embeddings of E and of K into F.
Since E/F and K/F are normal, we have ϕ(E) = E and ϕ(K) = K. Every element in EK is a
rational function in the elements of E ∪ K, so ϕ(EK) is contained in EK (and thus equal to EK)
as well. By Theorem 6.11.7, EK/F is normal. 
D EFINITION 6.11.10. Let E be a field. An automorphism of E is an isomorphism of rings
from E to itself.
E XAMPLES 6.11.11.
a. The identity map idF is an automorphism of any field F, known as the trivial automor-
phism. It is the identity element in Aut(F), and it is often denoted by 1.
b. Complex conjugation is an automorphism of C fixing R.
√ √ √ √
c. The map
√ φ : Q( 2) → Q( 2) sending a + b 2 to a − b 2 for all a, b ∈ Q is an automor-
phism of Q( 2).
d. The only automorphism of Q is the trivial automorphism, as the fact that φ (1) = 1 forces
φ (a) = a for all a ∈ Q using the properties of a ring homomorphism.
206 6. FIELD THEORY AND GALOIS THEORY

e. The Frobenius map ϕ p : F p → F p defined by ϕ p (x) = x p is an automorphism of F p fixing

F p.
R EMARK 6.11.12. The set of automorphisms of a field form a group under composition.
That is, the composition of two automorphisms is also an automorphism, as is the inverse of one.
D EFINITION 6.11.13. The automorphism group Aut(E) of a field E is the group of automor-
phisms of E with the operation of composition.
Often, we are interested in automorphisms fixing a subfield F of E. It is easy to see that these
form a subgroup of Aut(E).
N OTATION 6.11.14. We let AutF (E) denote the subgroup of Aut(E) for a field E consisting
of automorphisms that fix a subfield F.
R EMARK 6.11.15. If E is of characteristic 0, then AutQ (E) = Aut(E).
E XAMPLE 6.11.16. Note that C = R(i), and i has minimal polynomial x2 + 1. Any automor-
phism of C fixing R must take i to i or −i, which then determines the automorphism uniquely.
That is, the group AutR (C) consists of exactly two elements, the trivial automorphism and com-
plex conjugation.
The following is an immediate corollary of Theorem 6.11.7, Proposition 6.10.11a, and Propo-
sition 6.10.22.
C OROLLARY 6.11.17. Let E be a finite normal extension of a field F. Then EmbF (E) =
AutF (E), and the order [E : F]s of this group divides [E : F].
√
E XAMPLE 6.11.18. Consider the splitting field E = Q(ω, 3 2) of x3 − 2, where ω is a prim-
itive cube root of unity. Since E is normal, any embedding of E in an algebraic closure of Q
containing E has image E, so gives rise to an automorphism of E. Theorem 6.7.9 then tells us
that we may choose such an automorphism uniquely as follows. First, we choose another root
of the minimal polynomial x2 + x + 1 of ω and send ω to it, i.e., to ω or ω 2 . This yields an au-
tomorphism
√ of Q(ω). Then, we extend this automorphism to an automorphism of E by sending
3
2 to a root of its
√ minimal polynomial over Q(ω). √Since the degree of Q(ω), i.e. 2, is prime to
3 3
the degree of Q( 2), i.e. 3, over Q, √we have [Q(ω,
√ 2)√: Q(ω)] = 3,
√ so x3 − 2 is still irreducible
3 3 3 2 3
over Q(ω). √ Therefore, we can send 2 to any √ of 2, ω 2, and ω 2. That is, there are exactly
6, or [Q(ω, 2) : Q], elements of AutQ (Q(ω, 3 2)).
3

6.12. Galois extensions

D EFINITION 6.12.1. An algebraic field extension is said to be Galois if it is both normal and
separable.
R EMARK 6.12.2. By Theorem 6.10.19, an algebraic extension of a field of characteristic 0 is
Galois if and only if it is normal.
E XAMPLES 6.12.3.
√
a. The extensions Q( 2) and Q(i) of Q are Galois.
 6.12. GALOIS EXTENSIONS 207
√
b. The extension Q( 3 2)/Q is not Galois. It is separable but not normal.
c. The extension F p (t 1/p )/F p (t) is not Galois. It is normal but not separable.
d. The field Q is a Galois extension of Q.
e. For any n ≥ 1, the field F pn is a Galois extension of F p .
D EFINITION 6.12.4. Let E/F be a Galois extension. The Galois group Gal(E/F) of E/F is
the group of automorphisms of E that fix F.
R EMARK 6.12.5. The group Gal(E/F) is just AutF (E) in our earlier notation. The nota-
tion Gal(E/F) is used only for Galois extensions, whereas AutF (E) can be used for arbitrary
extensions.
N OTATION 6.12.6. We often write
E

F
to indicate that E is a field extension of F, and if E/F is Galois with Galois group G, we indicate
this by the diagram
E
G

F.
Drawings such as these are known as field diagrams and are useful in illustrating examples.
We will be concerned here only with finite Galois extensions. The following is immediate
from Corollary 6.11.17 and Proposition 6.10.11b.
P ROPOSITION 6.12.7. Let E/F be a finite Galois extension of fields. Then Gal(E/F) is a
finite group of order [E : F].
L EMMA 6.12.8. Let E be a field, and let G be a subgroup of Aut(E). Then the set of elements
of E that are fixed by every element of G is a subfield of E.
P ROOF. Let a, b ∈ E with b 6= 0. Let σ ∈ G. Then we have
σ (a − b) = σ (a) − σ (b) = a − b and σ (ab−1 ) = σ (a)σ (b)−1 = ab−1 ,
so a − b and ab−1 are elements of E fixed by G. 
With Lemma 6.12.8 in hand, we may make the following definition.
D EFINITION 6.12.9. Let G be a subgroup of Aut(E). The fixed field E G of E under G is the
largest subfield of E fixed by G.
Note the following.
L EMMA 6.12.10. Let K/F be a Galois extension, and let E be an intermediate field in K/F.
Then K is a Galois extension of E. Moreover, E/F is Galois if and only if it is normal.
208 6. FIELD THEORY AND GALOIS THEORY

P ROOF. The extension K/E is normal by Lemma 6.11.6 and separable by Lemma 6.10.10.
The extension E/F is also separable by Lemma 6.10.10, hence the second claim. 
P ROPOSITION 6.12.11. Let K/F be a finite Galois extension. Then the fixed field of K under
Gal(K/F) is F.
P ROOF. Let E = F Gal(K/F) . Clearly F ⊆ E, and we must show the other containment. By
Lemma 6.12.10, the extension K/E is Galois. On the other hand, every element of Gal(K/F)
fixes E, so Gal(K/F) is equal to its subgroup Gal(K/E) of automorphisms fixing E. By Propo-
sition 6.12.7, we have that
[K : F] = | Gal(K/F)| = | Gal(K/E)| = [K : E],
which means that [E : F] = 1, and therefore E = F. 
N OTATION 6.12.12. If K/F is a finite Galois extension and E is an intermediate field, then
the restriction of σ ∈ K to an embedding of E into K is denoted σ |E .
R EMARK 6.12.13. If K/F is a finite Galois extension and E is an intermediate field in K/F
such that E/F is Galois, then σ |E is an automorphism of E, so σ |E ∈ Gal(E/F).
D EFINITION 6.12.14. Let K/F be a finite Galois extension, and let E be an intermediate
field in K/F such that E/F is Galois. Then the restriction map from K to E (over F) is the
homomorphism of groups Gal(K/F) → Gal(E/F) takes σ ∈ Gal(K/F) to σ |E .
L EMMA 6.12.15. Let K/F be a Galois extension, and let E be an intermediate field in K/F.
Then there exists a bijection of sets
resE : Gal(K/F)/ Gal(K/E) → EmbF (E), resE (σ Gal(K/E)) = σ |E
for σ ∈ Gal(K/F), where F is an algebraic closure of F containing K.
P ROOF. Let σ , τ ∈ Gal(K/F). We have that σ |E = τ|E if and only if σ −1 τ fixes E, or
equivalently, is an element of Gal(K/E). In other words, σ |E = τ|E if and only if σ Gal(K/E) =
τ Gal(K/E). Therefore, resE is both well-defined and one-to-one.
Given an embedding τ of E into F fixing F, we may extend it to an embedding σ of K into
F. Since K/F is normal, σ is an automorphism of K. That is, σ is an element Gal(K/F) with
σ |E = τ, so resE is surjective. 
P ROPOSITION 6.12.16. Let K/F be a Galois extension, and let E be an intermediate field in
K/F. Then E/F is Galois if and only if Gal(K/E) is a normal subgroup of Gal(K/F). If E/F is
Galois, then restriction induces an isomorphism
res : Gal(K/F)/ Gal(K/E) −∼
E → Gal(E/F).
P ROOF. If E/F is Galois, then the restriction map from Gal(K/E) to Gal(K/F) is a surjec-
tive homomorphism with kernel exactly Gal(K/E) by Lemma 6.12.15. So, Gal(K/E) is normal
in Gal(K/F), and we have the stated isomorphism.
Conversely, suppose that Gal(K/E) is a normal subgroup of Gal(K/F). We already know
that E/F is separable by Lemma 6.10.10. To show that E/F is normal, it suffices by Theo-
rem 6.11.7 to show that ϕ(α) ∈ E for all α ∈ E and field embeddings ϕ : E → F fixing F, where
 6.12. GALOIS EXTENSIONS 209

F is an algebraic closure of F containing E. Since K/E is Galois, and since E is the fixed field of
Gal(K/E), we have ϕ(α) ∈ K, and we will have ϕ(α) ∈ E if we can show that σ (ϕ(α)) = ϕ(α)
for all σ ∈ Gal(K/E). Since K/F is Galois, we may lift ϕ to τ ∈ Gal(K/F). The desired equality
then amounts to σ τ(α) = τ(α), or τ −1 σ τ(α) = α. Since Gal(K/E) is normal in Gal(K/F), we
have that τ −1 σ τ fixes E, and in particular α. 
The final ingredient we need is as follows.
P ROPOSITION 6.12.17. Let K/F be a finite Galois extension, and let H be a subgroup of
Gal(K/F). Then we have Gal(K/K H ) = H.
P ROOF. By definition, H fixes K H , so we have H 6 Gal(K/K H ). Since K/F is separable, so
is K/K H , and the primitive element theorem tells us that K = K H (α) for some α ∈ K. Define
f= ∏ (x − σ (α)) ∈ K[x].
σ ∈H

For σ ∈ H, let σ̃ : K[x] → K[x] denote the induced homomorphism. We then have σ̃ ( f ) = f
for all σ ∈ H, which means that f ∈ K H [x]. In particular, the minimal polynomial of α over
K H divides f , and the degree of that polynomial is [K : K H ], while the degree of f is |H|. This
implies that [K : K H ] ≤ |H|, which since H 6 Gal(K/K H ), forces equality on both counts. 
D EFINITION 6.12.18. Let P and Q be sets of subsets of a set X and a set Y , respectively, and
suppose that φ : P → Q is a function. We say that φ is inclusion-reversing if whenever A, B ∈ P
with A ⊆ B, one has φ (B) ⊆ φ (A).
We may now state the fundamental theorem of Galois theory, which is essentially just a
combination of results we have proven above.
T HEOREM 6.12.19 (Fundamental theorem of Galois theory). Let K/F be a finite Galois
extension. Then there are inverse inclusion-reversing bijections

{intermediate fields in K/F} o

ψ
/ {subgroups of Gal(K/F)}
θ

defined on intermediate fields E in K/F and subgroups H of Gal(K/F) by

ψ(E) = Gal(K/E) and θ (H) = K H .
Moreover, for such E and H, we have
[K : E] = | Gal(K/E)| and |H| = [K : K H ].
These correspondences restrict to bijections

{normal extensions of F in K} o
ψ
/ {normal subgroups of Gal(K/F)}.
θ

Moreover, if E is normal over F (resp., H P Gal(K/F)), then restriction induces an isomorphism

Gal(K/F)/ Gal(K/E) −∼ → Gal(E/F) (resp., Gal(K/F)/H −∼ → Gal(K H /F)).
210 6. FIELD THEORY AND GALOIS THEORY

P ROOF. Let E and H be as in the statement of the theorem. We have that

θ (ψ(E)) = θ (Gal(K/E)) = K Gal(K/E) = E
by Proposition 6.12.11 and
ψ(θ (H)) = ψ(K H ) = Gal(K/K H ) = H
by Proposition 6.12.17, so θ and ψ are inverse bijections. The inclusion-reversing properties of
θ and ψ are immediate from the definitions of Galois groups and fixed fields. The statements
on orders and indices then follow immediately from Proposition 6.12.7, and the statements on
normal extensions and subgroups then become simply Proposition 6.12.16. 
√
E XAMPLE 6.12.20. The extension Q( 2, i)/Q is Galois with Galois group isomorphic to
the Klein four group. We have the complete field diagram
√
Q( 2, i)

Z/2Z Z/2Z
Z/2Z

√ √
Q( 2) (Z/2Z)2 Q( −2) Q(i)

Z/2Z
Z/2Z Z/2Z

Q.
√ √ √
That is, Gal(Q(
√ 2,
√ i)/Q) is abelian with two generators σ and τ such that σ ( 2) = − 2,
σ (i) = i, τ( 2) = 2, and τ(i) = −i.
√
E XAMPLE 6.12.21. Let G = Gal(Q(ω, 3 2)/Q), where ω is a primitive 3rd root of unity. We
have the field diagram
√
Q(ω, 3 2)
Z/3Z Z/2Z
Z/2Z Z/2Z
√ √ √
Q(ω) G Q( 3 2) Q(ω 3 2) Q(ω 2 3 2)

Z/2Z
Q
As a consequence of the fundamental theorem of Galois theory, we have G ∼ = S3 , since there are
only two groups of order 6 up to isomorphism and the cyclic one has a unique subgroup
√ of order
3
3. lt follows that our field
√ diagram contains all of the intermediate fields in Q(ω, 2)/Q. √ The fact
√
that the extension Q( 2) is not Galois for 0 ≤ i ≤ 2 corresponds to the fact Gal(Q(ω, 2)/Q( 3 2))
3 3
 6.12. GALOIS EXTENSIONS 211

is not a normal subgroup of G, and the other two non-normal intermediate fields correspond to
conjugate subgroups. √
3
One can also
√ see this
√ explicitly: note that
√ Gal(Q(ω,
√ 2)/Q(ω)) is generated by an element
3 3 3 3
τ such that τ( 2) = ω 2, and Gal(Q(ω, 2)/Q( 2)) is generated by an element σ such that
σ (ω) = ω 2 . Then τ 3 = 1, σ 2 = 1, and
σ τσ −1 (ω) = σ τ(ω 2 ) = σ (ω 2 ) = ω = τ −1 (ω)
√ √ √ √ √
σ τσ −1 ( 3 2) = σ τ( 3 2) = σ (ω 3 2) = ω 2 3 2 = τ −1 ( 3 2),
so σ τσ −1 = τ −1 , and G = hσ , τi is a nonabelian group of order 6, isomorphic to D3 ∼
= S3 .
More generally, we have the following results on Galois groups of composite fields.
P ROPOSITION 6.12.22. Let L/F be an algebraic extension, and let K and E be extensions
of F in L such that K/F is finite Galois. Then EK/E and K/(E ∩ K) are finite Galois, and the
restriction map
resK : Gal(EK/E) → Gal(K/(E ∩ K)), resK (σ ) = σ |K for σ ∈ Gal(EK/E)
is an isomorphism.
P ROOF. First, note that EK/E is normal as it is the splitting field of the same set of polyno-
mials in F[x] that K is over F. Since K/F is finite and separable, we have K = F(β ) for some
β ∈ K, so EK = E(β ), and the fact that the minimal polynomial of β is separable over F tells
us that it is over E as well, and therefore EK/E is separable as well. Thus, EK/E is Galois, and
K/(E ∩ K) is Galois by Lemma 6.12.10.
Now, suppose that σ ∈ Gal(EK/E) and resK (σ ) = σ |K = 1. By definition, we have σ |E = 1
as well, so σ fixes every rational function over E in β , and therefore σ fixes EK, which is to say
that σ = 1, or resK is injective. Now, let H be the image of resK . The elements of K fixed by H
are exactly the elements of K fixed by Gal(EK/E), so we have
K H = K Gal(EK/E) = (EK)Gal(EK/E) ∩ K = E ∩ K,
and therefore H = Gal(K/K H ) = Gal(K/(E ∩ K)), so resK is surjective as well. 
P ROPOSITION 6.12.23. Let L/F be an algebraic extension, and let K and E be finite Galois
extensions of F in L. Then EK/F and E ∩ K/F are Galois, and the product of restriction maps
π : Gal(EK/F) → Gal(K/F) × Gal(E/F), π(σ ) = (σ |K , σ |E ) for σ ∈ Gal(EK/F)
is an injective homomorphism that is an isomorphism if and only if E ∩ K = F.
P ROOF. That EK/F is separable is Corollary 6.10.12 applied to EK/E and E/F, and that
it is normal is Proposition 6.11.9. If β ∈ E ∩ K, then both E and K contain all roots in F of its
minimal polynomial, so E ∩ K also contains these roots, hence is normal over K. That E ∩ K is
separable over F follows from the fact that E is.
The kernel of π is exactly those elements of Gal(EK/F) that fix both K and E, and hence fix
all of EK, since every element of EK is a rational function in the elements of E and K. Thus π is
212 6. FIELD THEORY AND GALOIS THEORY

injective. Since π is injective, it is surjective if and only if the orders of its domain and codomain
are the same, which is to say if and only if
[EK : F] = [E : F][K : F].
By Proposition 6.12.22, we have
[EK : F] = [EK : K][K : F] = [E : E ∩ K][K : F],
so π is surjective if and only if E ∩ K = F. 
D EFINITION 6.12.24. Let K/F be a Galois extension.
a. We say that K/F is abelian if Gal(K/F) is abelian.
b. We say that K/F is cyclic if Gal(K/F) is cyclic.
E XAMPLES 6.12.25. We revisit Examples 6.12.20 and 6.12.21.
√ √
a. The field Q(i, 2) is the compositum of the √ normal extensions Q(i) and Q( 2), which
both have√Galois group Z/2Z and satisfy Q(i) ∩ Q( √ 2) = Q. By Proposition 6.12.23, we have
∼ 2
Gal(Q(i, 2)/Q) = (Z/2Z) . The extension Q(i, 2)/Q is abelian.
√ √
b. Take G = Gal(Q(ω, 3 2)/Q). Take K = Q(ω) and E = Q( 3 2). Then G = Gal(EK/Q),
and we set N = Gal(EK/K) and H = Gal(EK/E). The map resK : G → Gal(K/Q) is a surjection
with kernel N that restricts to an isomorphism on H by Proposition 6.12.22. In particular, H is a
complement to N, and G is a semidirect product N o H, nontrivial as E/Q is not normal. In our
case, N ∼
= Z/3Z and H ∼ = Z/2Z, so G is nonabelian of order 6, isomorphic to S3 .
The following example is worth being stated as a proposition, as it tells us that all Galois
groups of all extensions of finite fields are cyclic.
P ROPOSITION 6.12.26. Let q be a prime power and n ≥ 1. Then Fqn /Fq is cyclic of degree
n.
P ROOF. The group Gal(Fqn /Fq ) contains the Frobenius element ϕq with ϕq (α) = α q for all
r
α ∈ Fqn . For ϕqr (α) = α q to equal α would mean that that α is a (qr − 1)th root of unity, which
in turn could only happen for all α ∈ Fqn if and only if r is a multiple of n. That is, the order of
ϕq is n. Therefore, Gn must be cyclic of order n, generated by ϕq . We have that Fqm is a subfield
of Fqn if and only if m divides n, in which case Gal(Fqn /Fqm ) = hϕqm i is a cyclic group of order
n/m. In particular, every finite Galois extension of finite fields is cyclic. 
We can also determine the structure of the Galois groups of cyclotomic extensions of Q. We
note that the extension Q(ζn )/Q is Galois in that Q(ζn ) is the splitting field of xn − 1.
T ERMINOLOGY 6.12.27. For n ≥ 1 and a ∈ (Z/nZ)× , we will take ζna to be ζnã for any ã ∈ Z
with a = ã + nZ.
D EFINITION 6.12.28. For every n ≥ 1, the nth cyclotomic character is the unique map
χn : Gal(Q(ζn )/Q) → (Z/nZ)×
χ (σ )
such that σ (ζn ) = ζn n for all σ ∈ Gal(Q(ζn )/Q).
 6.13. PERMUTATIONS OF ROOTS 213

P ROPOSITION 6.12.29. The nth cyclotomic character is an isomorphism for every n ≥ 1.

P ROOF. We note first that χn is a homomorphism. That is, for σ , τ ∈ Gal(Q(ζn )/Q), we
have
χ (σ τ) χ (τ) χ (σ )χn (τ)
ζn n = σ τ(ζn ) = σ (ζn n ) = σ (ζn )χn (τ) = ζn n .
Next, note that χn is injective since an element of Gal(Q(ζn )/Q) is determined by its value on
the generator ζn of the extension. Finally, Theorem 6.6.11 implies that [Q(ζn ) : Q] = ϕ(n), so
the orders of the two groups are the same. 
C OROLLARY 6.12.30. The nth cyclotomic field is a finite abelian extension of Q.
R EMARK 6.12.31. The Kronecker-Weber theorem, a proof of which is beyond the scope of
these notes, states that every finite abelian extension of Q is contained inside some cyclotomic
field.

6.13. Permutations of roots

We first recall that every finite Galois extension is the splitting field of some polynomial (and
in fact we may take that polynomial to be irreducible by the primitive element theorem).
T HEOREM 6.13.1. Let K/F be the splitting field of a separable degree n polynomial in F[x].
Then Gal(K/F) is isomorphic to a subgroup of Sn .
P ROOF. Let K be the splitting field of f ∈ F[x], and let X be the set of n roots of f . For α ∈ X
and σ ∈ Gal(K/F), we have f (σ (α)) = σ ( f (α)) = 0, so σ (α) ∈ X. In other words, Gal(K/F)
acts on X, and thus we have an induced permutation representation ρ : Gal(K/F) → SX . Note
that K is given by adjoining the elements of X to F, so if σ ∈ Gal(K/F) fixes every element of X,
it fixes every element of K and is therefore tirival. Thus, the action of Gal(K/F) on X is faithful,
so ρ is injective. 
C OROLLARY 6.13.2. Let K/F be the splitting field of a separable degree n polynomial in
F[x]. Then [K : F] divides n!.
E XAMPLES 6.13.3. Again, we revisit Examples 6.12.20 and 6.12.21.
√ 2 2
√ 2, i) is the splitting field of (x − 2)(x + 1) over Q, which has 4 roots. The
a. The field Q(
image of Gal(Q( 2, i)/Q) under any permutation representation on these roots is conjugate to
h(1 2), (3 4)i.
√ √ √
b. If we label the roots of x3 − 2 in the order 3 2, ω 3 2, ω 2 3 2, then we have a permutation
representation
√
ρ : G = Gal(Q(ω, 2)/Q(ω)) −∼
3
→S . 3
We have G = hσ , τi as in Example 6.12.21 with ρ(σ ) = (2 3) and ρ(τ) = (1 2 3).
One might ask if every subgroup of Sn , and therefore every finite group, occurs as the Galois
group of some extension of fields. As we shall see, the answer is yes.
214 6. FIELD THEORY AND GALOIS THEORY

D EFINITION 6.13.4. Let F be a field, and let x1 , x2 , . . . , xn be indeterminates. For 1 ≤ k ≤ n,

the kth elementary symmetric polynomial sn,k in F[x1 , x2 , . . . , xn ] is
sn,k (x1 , . . . , xn ) = ∑ xi1 xi2 · · · xik .
1≤i1 <i2 <···<ik ≤n

R EMARK 6.13.5. Put differently, sk is the sum over the subsets of Xn = {1, 2, . . . , n} of order
k of the products of variables with indices in the sets. That is,
sn,k (x1 , . . . , xn ) = ∑ ∏ xi.
P⊂Xn i∈P
|P|=k
n

As a consequence, sk,n is a sum of k monomials.
E XAMPLES 6.13.6. We have sn,1 = x1 + x2 + · · · + xn and sn,n = x1 x2 · · · xn . For n = 3, we
also have s3,2 = x1 x2 + x1 x3 + x2 x3 , and for n = 4, we have
s4,2 = x1 x2 + x1 x3 + x1 x4 + x2 x3 + x2 x4 + x3 x5 and s4,3 = x1 x2 x3 + x1 x2 x4 + x1 x3 x4 + x2 x3 x4 .
P ROPOSITION 6.13.7. The function field F(x1 , x2 , . . . , xn ) is a finite Galois extension of its
subfield F(sn,1 , sn,2 , . . . , sn,n ), with Galois group isomorphic to Sn .
P ROOF. Let E = F(sn,1 , sn,2 , . . . , sn,n ) and K = Q(x1 , x2 , . . . , xn ). The polynomial
n n
f (y) = ∏(y − xi ) = ∑ sn,i yi ∈ F[y]
i=1 i=0
has roots xi with 1 ≤ i ≤ n. Thus K is the splitting field of f over E. To ρ ∈ Sn , we can associate
a unique φ (ρ) ∈ AutF (K) by
φ (ρ)(h(x1 , x2 , . . . , xn )) = h(xρ(1) , xρ(2) , . . . , xρ(n) )
for h ∈ K. As Sn acts on the set of subsets of Xn of order k, Remark 6.13.5 implies that
φ (ρ)(sn,k ) = sn,k for all k, so φ (ρ) ∈ Gal(K/E). The map φ : Sn → Gal(K/E) is a homomor-
phism that is injective by definition and surjective by Theorem 6.13.1. 
We have the following consequence.
C OROLLARY 6.13.8. Every finite group is isomorphic to the Galois group of some field ex-
tension.
P ROOF. Let G be a group, and choose n such that H is isomorphic to a subgroup of Sn ,
which exists by Cayley’s theorem. Proposition 6.13.7 yields an extension K/E of fields with
Gal(K/E) ∼= Sn . Then G is isomorphic to some subgroup H of Gal(K/E), and we have H ∼ =
H
Gal(K/K ). 
D EFINITION 6.13.9. Let F be a field. The discriminant of a monic, degree n polynomial
f ∈ F[x] is
D( f ) = ∏ (αi − α j )2 ,
1≤i< j≤n
where f = ∏ni=1 (x − αi ) in a splitting field of F.
 6.13. PERMUTATIONS OF ROOTS 215

The following lemma is obvious from the definition of the discriminant.

L EMMA 6.13.10. The discriminant of a monic polynomial f is 0 if and only if f is insepara-
ble.
In fact, the discriminant of a monic polynomial lies in the ground field of the extension, from
which it easily follows that it is well-defined independently of the choice of splitting field in its
definition.
P ROPOSITION 6.13.11. The discriminant of a monic polynomial f ∈ F[x] lies in F.
P ROOF. By Lemma 6.13.10, we may suppose that f is separable. Let K be a splitting field of
F, and let σ ∈ Gal(K/F). As σ permutes the roots αi of f , it induces an element ρ ∈ Sn such that
σ (αi ) = αρ(i) . Taking ∆ = ∏1≤i< j≤n (xi − x j ), we know by Proposition 4.12.1 that ρ(∆) = ±∆
for the standard action of Sn on polynomials in variables x1 , x2 , . . . , xn . But then ρ(∆2 ) = ∆2 , so
plugging in αi for xi , we obtain σ (D( f )) = D( f ). Since D( f ) is fixed by Gal(K/F), it lies in
F. 
R EMARK 6.13.12. The proof of Proposition 6.13.11 shows that an element of Gal(K/F) for
the splitting field K of a separable polynomial f of degree n induces an even permutation of the
roots of f if and only if it fixes ∏1≤i< j≤n (αi − α j ).
As a direct consequence of Remark 6.13.12, we have the following.
P ROPOSITION 6.13.13. The discriminant D( f ) of a monic, separable polynomial f ∈ F[x] is
a square in F × if and only if the Galois group of its splitting field has image a subgroup of An
via its permutation representation on the roots of f .
We explore the consequences of Proposition 6.13.13 for polynomials of low degree.
E XAMPLE 6.13.14. Let f = x2 + ax + b ∈ F[x]. Let α, β be the roots of F in an algebraic
closure of F. The extension F(α)/F is normal, being that it is of degree 1 or 2, so F(α) = F(β ).
Note that −a = α + β and b = αβ , so
D( f ) = α 2 + β 2 − 2αβ = a2 − 4b.
If char F = 2, then a2 − 4b = a2 , so Proposition 6.13.13 tells us that F(α)/F is trivial if a 6= 0
and inseparable (possibly trivial) if a = 0. If char F 6= 2, the extension F(α)/F is separable, so
Proposition 6.13.13 again tells us that a2 − 4b is a square if and only if α √
∈ F. This can also be
seen by the quadratic formula, which tells us in particular that F(α) = F( D) if char F 6= 2.
The case of degree 3 polynomials is rather more involved.
E XAMPLE 6.13.15. Suppose char F 6= 3. Let f = x3 + ax2 + bx + c ∈ F[x]. Setting y = x + 3a ,
we obtain
f = (y − a3 )3 + a(y − 3a )2 + b(y − a3 ) + c
2 3 2 3
= (y3 − ay2 + a3 y − 27
a
) + (ay2 − 2a3 y + a9 ) + (by − ab
3 )+c
2 3
= y3 + (− a3 + b)y + ( 2a ab
27 − 3 + c).
216 6. FIELD THEORY AND GALOIS THEORY

Set p = 13 (−a2 + 3b) and q = 27 1

(2a3 − 9ab + 27c), and let g = x3 + px + q ∈ F[x].
Let K be a splitting field of f over F, and let α, β , γ ∈ K be the roots of g. Then α +β +γ = 0,
s3,2 (α, β , γ) = p, and −αβ γ = q. Note that this implies that
(6.13.1) 0 = (α + β + γ)2 = α 2 + β 2 + γ 2 + 2p
and
(6.13.2)
p2 = (αβ + αγ + β γ)2 = 2αβ γ(α + β + γ) + α 2 β 2 + α 2 γ 2 + β 2 γ 2 = α 2 β 2 + α 2 γ 2 + β 2 γ 2 .
Note that the formal derivative of g is
3x2 + p = s3,2 (x − α, x − β , x − γ),
and we can plug α into this, for instance, to obtain
3α 2 + p = (α − β )(α − γ).
Doing this also for β and γ and taking the ordering of the differences into account, we obtain by
(6.13.1) and (6.13.2) that
−D(g) = (3α 2 + p)(3β 2 + p)(3γ 2 + p)
= 27α 2 β 2 γ 2 + 9p(α 2 β 2 + α 2 γ 2 + β 2 γ 2 ) + 3p2 (α 2 + β 2 + γ 2 ) + p3
= 27q2 + 9p3 − 6p3 + p3 = 27q2 + 4p3 .
That is, D(g) = −4p3 − 27q2 . Since the roots of f and g differ by 3a , the differences of the roots
of the two are the same, so
D( f ) = −4p3 − 27q2 = a2 b2 − 4a3 c + 18abc − 4b3 − 27c2 .
Now, suppose that f is irreducible. Then Gal(K/F) is isomorphic to a subgroup of S3 of
order divisible by 3, so it is either isomorphic to A3 ∼ = Z/3Z or S3 , depending on whether D( f )
is a square or not, respectively. If D( f ) ∈ F ×2 , then K is given by adjoining any single root of
f . If D( f ) 6∈ F ×2 , then K has a unique intermediate extension F(D( f )1/2 ) of degree 2, and K is
given by adjoining to this any root of f .
We go into a bit less detail for polynomials of degree 4.
E XAMPLE 6.13.16. Let K be the splitting field of a monic, irreducible, separable polynomial
f of degree 4 in F[x]. If D( f ) ∈ F ×2 , then Gal(K/F) is isomorphic to a subgroup of A4 of degree
divisible by 4, so A4 or the Klein 4-group V4 . If D( f ) 6∈ F ×2 , then Gal(K/F) is isomorphic to
Z/4Z, D8 , or S4 .
Let α1 , α2 , α3 , α4 be the roots of f , and set β3 = (α1 +α2 )(α3 +α4 ), β2 = (α1 +α3 )(α2 +α4 ),
and β1 = (α1 + α4 )(α2 + α3 ). The set {β1 , β2 , β3 } is a union of orbits under Gal(K/F), so we
can set
g(x) = (x − β1 )(x − β2 )(x − β3 ) ∈ F[x]
and let E be the splitting field of g over F.
 6.13. PERMUTATIONS OF ROOTS 217

Let ρK : Gal(K/F) → S4 (resp., ρE : Gal(E/F) → S3 ) be the permutation map for the given
ordering of the αi (resp., βi ). Then we have π : S4 → S3 with kernel h(1 2)(3 4), (1 3)(2 4)i and
restricting to the identity on h(1 2), (1 2 3)i such that π(ρK (σ )) = ρE (σ |E ) for all σ ∈ Gal(K/F).
If g splits, then Gal(K/F) ∼ = V4 . If g factors as a linear polynomial times an irreducible
quadratic, then Gal(K/F) ∼ = 8 if f is irreducible over F(D( f )1/2 ) and Gal(K/F) ∼
D = Z/4Z oth-
erwise. If g is irreducible and D(g) ∈ F , then Gal(E/F) = Z/3Z, which forces Gal(K/F) ∼
×2 ∼ = A4
since 4 divides [K : F]. If g is irreducible and D(g) is not a square in F, then Gal(E/F) ∼ = S3 ,
∼
which forces Gal(K/F) = S4 .
We next present a proof of the fundamental theorem of algebra that uses Galois theory. We
will use the fact that every polynomial of odd degree has a real root (by the intermediate value
theorem). We also recall that quadratic polynomials in C[x] split completely, as is seen via the
quadratic formula and the fact that complex numbers have square roots in C.
P ROOF OF THE FUNDAMENTAL THEOREM OF ALGEBRA . First, let f ∈ C[x] be monic and
irreducible, and let f¯ ∈ C[x] given by applying complex conjugation to its coefficients. The
polynomial g = f f¯ lies in R[x] since complex conjugation permutes f and f¯, and it suffices to
show that g has a root in C. So, we can and do assume that f ∈ R[x].
Let n = deg f , and write n = 2k m for some odd m and k ≥ 0. If k = 0, then f has odd degree
and hence a real root, so we suppose k ≥ 1. By induction, suppose we know that all polynomials
in R[x] of degree 2k−1 times an odd number have a root in C. Let α1 , α2 , . . . , αn be the roots of f
in a splitting field Ω of f over C.
For t ∈ R, define
ht (x) = ∏ (x − (αi + α j + tαi α j )) ∈ Ω[x].
1≤i< j≤n
Any permutation of the αi ’s preserves ht , so Gal(Ω/R) fixes ht , and thus ht ∈ R[x]. Note that
n

deg ht = 2 = 2 m0 for some odd m0 , and thus by induction ht has a root in C, which necessarily
k−1

has the form αi + α j + tαi α j for some i < j. In fact, we have such a root for every t ∈ R, and
since that is an infinite set of t, there exist i < j and s,t ∈ R such that αi + α j + sαi α j and
αi + α j + tαi α j are both in C from which it follows that αi + α j ∈ C and αi α j ∈ C. But then
(x − αi )(x − α j ) ∈ C[x], which being quadratic, has a root in C. 
 Part 2

A Second Course
 CHAPTER 7

Topics in group theory

7.1. Semidirect products

P ROPOSITION 7.1.1. Let N and H be groups and ϕ : H → Aut(N) be a homomorphism. Then
there exists a group G with underlying set N × H and group operation
(n, h) · (n0 , h0 ) = (nϕ(h)(n0 ), hh0 )
for all n, n0 ∈ N and h, h0 ∈ H. Moreover, H = {e} × H 6 G and N = N × {e} P G. In fact, in G
we have ϕ(h)(n) = hnh−1 for all h ∈ H and n ∈ N.
P ROOF. We note that (e, e) ∈ G is an identity, that (ϕ(h−1 )(n−1 ), h−1 ) is inverse to (n, h),
and we leave it to the reader to check associativity. Clearly H, N 6 G by definition of the multi-
plication, and we check that for h ∈ H and n ∈ N, we have
hnh−1 = (e, h)(n, e)(e, h−1 ) = (e, h)(n, h−1 ) = (ϕ(h)(n), e) = ϕ(h)(n) ∈ N.

D EFINITION 7.1.2. For groups N and H and a homomorphism ϕ : H → Aut(N), the group
defined by Proposition 7.1.1 is known as the semidirect product of N and H relative to ϕ and is
denoted by N oϕ H.
E XAMPLE 7.1.3. If H and N are groups and ϕ : H → Aut(N) satisfies ϕ(h) = idN for all
h ∈ H, then H oϕ N is the direct product H × N.
E XAMPLE 7.1.4. Let ϕ : (Z/nZ)× → Aut(Z/nZ) be the isomorphism taking a ∈ (Z/nZ)
×
to multiplication by a. Set G = Z/nZ oϕ (Z/nZ)× . Then G −∼
→ Aff(Z/nZ) via (b, a) 7→ a0 b1 ,
so G is also isomorphic to Aut(Dn ) by Proposition 4.3.5.
P ROPOSITION 7.1.5. Let G be a group with normal subgroup N and subgroup H such that
N ∩ H = {e} and NH = G. Define a homomorphism ϕ : H → Aut(N) by ϕ(h)(n) = hnh−1 . Then
we may define an isomorphism of groups by
ψ : N oϕ H → G, ψ(n, h) = nh
for all n ∈ N and h ∈ H.
P ROOF. Any g ∈ G can be written as nh for some n ∈ N and h ∈ H by assumption, so f is
onto. For n, n0 ∈ N and h, h0 ∈ H, we have
ψ((n, h)(n0 , h0 )) = ψ(nϕ(h)(n0 ), hh0 ) = nϕ(h)(n0 )hh0 = nhn0 h0 = ψ((n, h)) f ((n0 , h0 )).
If ψ(n, h) = nh = e, then n = h−1 ∈ N ∩ H, so n = h = e. 
221
222 7. TOPICS IN GROUP THEORY

The proposition we have just proven has Proposition 4.11.4 as a corollary.

A LTERNATE PROOF OF P ROPOSITION 4.11.4. By Proposition 7.1.5, we need only that ϕ : H →
Aut(N) given by ϕ(h)(n) = hnh−1 for h ∈ H and n ∈ N is the trivial map. That is, we need
that hnh−1 = n, or [h, n] = e, for all such h and n. This follows as H and N are normal, so
[h, n] ∈ H ∩ N = {e}. 
D EFINITION 7.1.6. If G is a group with subgroups N and H such that G ∼ = N oϕ H for
ϕ : H → Aut(N) given by ϕ(h)(n) = hnh−1 , then we say that G is the internal semidirect product
of N and H and write G = N o H to denote this.
D EFINITION 7.1.7. Let G be a group with normal subgroup N. A complement to N in G is a
subgroup H such that G is the internal semidirect product N o H of N and H.
R EMARK 7.1.8. There are often many complements to a normal subgroup. In particular, if
G = N o H and n ∈ N, then nHn−1 is also a complement to N. If N is abelian, then we have the
equality γnhn−1 = γh of conjugation maps, but if N is nonabelian, then these may not be equal.
The map ϕ : H → Aut(N) given by ϕ(h) = γnhn−1 would then satisfy G ∼ = N ×ϕ H, though not
necessarily internally. Rather, this is simply an expression of the fact that G = N o nHn−1 .
The following rather general result can be used to show that two semidirect products are
isomorphic.
P ROPOSITION 7.1.9. Let H, H 0 , N, and N 0 be groups and ϕ : H → Aut(N) and ϕ 0 : H 0 →
Aut(N 0 ) be homomorphisms. Suppose that there exist isomorphisms ψ : H → H 0 and θ : N → N 0 ,
and define
Θ : Aut(N) → Aut(N 0 )
by Θ(α) = θ ◦ α ◦ θ −1 for any α ∈ Aut(N). If Θ ◦ ϕ = ϕ 0 ◦ ψ, then the map
f : N oϕ H → N 0 oϕ 0 H 0
defined by f (n, h) = (θ (n), ψ(h)) for all n ∈ N and h ∈ H is an isomorphism.
P ROOF. Note that f has an inverse given by f −1 (n0 , h0 ) = (θ −1 (n0 ), ψ −1 (h0 )) for all n0 ∈ N
and h0 ∈ H, so we need only show that f is a homomorphism. Letting n1 , n2 ∈ N and h1 , h2 ∈ H,
we calculate:
f ((n1 , h1 )(n2 , h2 )) = f (n1 ϕ(h1 )(n2 ), h1 h2 ) = (θ (n1 )θ (ϕ(h1 )(n2 )), ψ(h1 )ψ(h2 )),
f (n1 , h1 ) f (n2 , h2 ) = (θ (n1 ), ψ(h1 ))(θ (n2 ), ψ(h2 )) = (θ (n1 )ϕ 0 (ψ(h1 ))(θ (n2 )), ψ(h1 )ψ(h2 )).
To see that the first coordinates of these expressions are equal, we check that
ϕ 0 (ψ(h1 ))(θ (n2 )) = Θ(φ (h1 ))(θ (n2 )) = (θ ◦ φ (h1 ) ◦ θ −1 )(θ (n2 )) = θ (φ (h1 )(n2 )).
Thus, f is a homomorphism. 
We can use this to completely classify groups of order a product of two distinct primes,
completing the study begun in Theorem 4.11.5.
T HEOREM 7.1.10. Let p and q be distinct primes with q ≡ 1 mod p. Then there exists a
unique isomorphism class of nonabelian groups of order pq.
 7.2. COMPOSITION SERIES 223

P ROOF. Let G be a nonabelian group. By Theorem 4.11.5, we have that it has a unique
normal subgroup Q of order q, and let P be a subgroup of order p. By Proposition 7.1.5, we have
that G = Q o P. We have P ∼= Z/pZ and Q ∼ = Z/qZ. Fixing such isomorphisms and recalling the
∼
canonical isomorphism Aut(Z/qZ) = (Z/qZ)× , we are reduced to showing that there is a unique
isomorphism class of semi-direct product (Z/qZ) oϕ (Z/pZ), where ϕ : Z/pZ → (Z/qZ)× is a
nontrivial homomorphism. The group (Z/qZ)× is cyclic by Corollary 6.5.5. Let a ∈ (Z/qZ)×
be a generator.
Any nontrivial homomorphism ϕ : Z/pZ → (Z/qZ)× must send 1 to an element of order p
in (Z/qZ)× . If we set b = a(q−1)/p , then ϕ(1) = bi for some i ∈ Z with i 6≡ 0 mod p. Let us
denote this particular homomorphism by ϕi , and define ψi : Z/pZ → Z/pZ to be multiplication
by i. Then ϕi = ϕ1 ◦ ψi since both maps send 1 to bi . Proposition 7.1.9 then tells us that the
semidirect products (Z/qZ) o (Z/pZ) defined by ϕ1 and ϕi are isomorphic. That is, there is a
unique isomorphism class of nonabelian semidirect product of order pq. 

7.2. Composition series

First, we explain how simple groups may be used in building arbitrary finite groups, starting
with the following definition.
D EFINITION 7.2.1. Any collection (Hi )i∈Z of subgroups of a group G with Hi−1 6 Hi for
i ∈ Z is called a series of subgroups of G.
D EFINITION 7.2.2. Let C = (Hi )i∈Z be a series of subgroups of a group G.
a. We say that C is an ascending series if Hi = 1 for i sufficiently small.
b. We say that C is a descending series if Hi = G for i sufficiently large.
c. We say that C is a finite series if it is both ascending and descending.
d. The length of a finite series C is difference j − i of the smallest integer j such that H j = G
and largest integer i such that Hi = 1.
N OTATION 7.2.3. We use the notation
1 = H0 6 H1 6 · · · 6 Ht−1 6 Ht = G
to denote a finite series of subgroups Hi of a group G with H0 = 1, Ht = G. It has length t if
H0 6= H1 and Ht−1 6= Ht .
R EMARK 7.2.4. To say that a series (Hi )i∈Z of subgroups of G is finite is stronger than simply
saying it has only finitely many terms. For instance, if G is nontrivial, then Hi = 1 for all i ∈ Z
provides a series with only one distinct subgroup, but it is not finite as no Hi equals G.
R EMARK 7.2.5. A descending series in G is often taken to be a list (Hi )i∈Z of subgroups of
G with Hi 6 Hi−1 for all i and Hi = G for i sufficiently small. This agrees with the usual notion
in the sense that letting Ki = H−i will provide a descending series (Ki )i∈Z in the sense of the
original definition.
224 7. TOPICS IN GROUP THEORY

D EFINITION 7.2.6. A finite series

1 = H0 6 H1 6 · · · 6 Ht−1 6 Ht = G
of subgroups of G is said to be a subnormal series if Hi−1 P Hi for all 1 ≤ i ≤ t. It is called a
normal series if Hi P G for all 0 ≤ i ≤ t − 1.
D EFINITION 7.2.7. Two subnormal series (Hi )ti=0 and (Ki )ti=0 are equivalent if there exists
σ ∈ St such that Hi /Hi−1 ∼
= Kσ (i) /Kσ (i)−1 for all 1 ≤ i ≤ t.
D EFINITION 7.2.8. A refinement of a subnormal series (Hi )ti=0 in a group G is a subnormal
series (Ki )si=0 such that there exists an increasing function f : {0, . . . ,t} → {0, . . . , s} such that
Hi = K f (i) for 0 ≤ i ≤ t.
T HEOREM 7.2.9 (Schreier refinement theorem). Any two subnormal series in a group G have
refinements that are equivalent.
P ROOF. Let (Hi )ti=0 and (Ki )si=0 be subnormal series in G. For 0 ≤ i < t and 0 ≤ j < s, let
Msi+ j = Hi (Hi+1 ∩ K j ) and Nt j+i = K j (K j+1 ∩ Hi ).
Set Mmn = Nmn = G as well. Then Msi+ j P Msi+ j+1 for 0 ≤ i ≤ t − 1 and 0 ≤ j ≤ s − 2 as
K j P K j+1 , and Msi+s−1 P Hi+1 = Ms(i+1) for 0 ≤ i ≤ t − 1 as Hi P Hi+1 and Hs−1 P G. Thus
(Mi )mn mn
i=0 is a subnormal series, as is (Ni )i=0 . In fact, we see from this that (Mi )i refines (Hi )i and
(Ni )i refines (Ki )i .
It remains to see that (Mi )i and (Ni )i are equivalent. For 0 ≤ i ≤ t − 1 and 0 ≤ j ≤ s − 2, note
that
Msi+ j+1 ∼ Hi (Hi+1 ∩ K j+1 ) ∼ K j (K j+1 ∩ Hi+1 ) ∼ Nt j+i+1
= = =
Msi+ j Hi (Hi+1 ∩ K j ) K j (K j+1 ∩ Hi ) Nt j+i
by the butterfly lemma, and
Ms(i+1) Hi+1 Ks−1 Hi+1 ∼ Ks−1 (Ks ∩ Hi+1 ) ∼ K(s−1)t+i+1
∼
= ∼
= = =
Ms(i+1)−1 Hi (Hi+1 ∩ Ks−1 ) Ks−1 Hi Ks−1 (Ks ∩ Hi ) K(s−1)t+i
for 0 ≤ i ≤ t − 1 since Ks = G. Thus, the two refinements are equivalent. 
D EFINITION 7.2.10. A subnormal series of subgroups
1 = H0 C H1 C · · · C Ht−1 C Ht = G
of a group G is called a composition series for G if Hi /Hi−1 is simple for each 1 ≤ i ≤ t. The
simple groups Hi /Hi−1 are referred to as the composition factors of the series.
L EMMA 7.2.11. Let
1 = H0 C H1 C · · · C Ht−1 C Ht = G
be a composition series for G, and let N be a proper normal subgroup of G.
a. There exists s ≤ t and an increasing function f : {0, . . . , s} → {0, . . . ,t} with f (0) = 0 such
that
1 C H f (1) ∩ N C H f (2) ∩ N C · · · C H f (s) ∩ N = N
 7.2. COMPOSITION SERIES 225

is a composition series for N with composition factors

H f (i) ∩ N H f (i)
∼
= .
H f (i−1) ∩ N H f (i)−1
b. Set H̄i = Hi /(Hi ∩ N) for 0 ≤ i ≤ t. There exists an r ≤ t and an increasing function
f0: {0, 1, . . . , r} → {0, 1, . . . ,t} with f (0) = 0 such that
1 = H̄0 C H̄ f 0 (1) C · · · C H̄ f 0 (r−1) C H̄ f 0 (r) = G/N
is a composition series for G/N with composition factors
H̄ 0 /H̄ 0 ∼
= H 0 /H
f (i) f (i−1) f (i) f 0 (i)−1 .
c. In the notation of parts a and b, the images of f and f 0 to be complementary away from
0,x and r + s to equal t.
P ROOF. Let 0 ≤ i ≤ t −1. The quotient (Hi ∩N)/(Hi−1 ∩N) is a subgroup of the simple group
Hi /Hi−1 and therefore necessarily trivial or improper. Let s be the number of simple quotients.
Let f (0) = 0, and for 1 ≤ j ≤ s, let f ( j) be the smallest positive integer greater than f ( j − 1) and
such that (H f ( j) ∩ N)/(H f ( j)−1 ∩ N) is simple. Then H f ( j)−1 ∩ N = H f ( j−1) ∩ N, and the result
follows.
Similarly, by the third isomorphism theorem, we have
H̄i ∼ Hi
= ,
H̄i−1 Hi−1 (Hi ∩ N)
which is a quotient of Hi /Hi−1 by the image of Hi+1 ∩ N in it. Since Hi+1 /Hi is simple, this
image is either trivial or Hi+1 /Hi . That is, H̄i+1 /H̄i is either trivial or simple. Let s be the number
of simple terms. Set f 0 (0) = 0, and for 1 ≤ j ≤ r, take f 0 ( j) to be the smallest integer greater
than f 0 ( j − 1) such that H̄ f 0 ( j) /H̄ f 0 ( j)−1 is simple. Then H̄ f 0 ( j)−1 = H̄ f 0 ( j−1) .
Note that H̄i 6= H̄i−1 if and only if Hi ∩ N = Hi−1 ∩ N. Then r + s = t and the images of f and
f 0 are complementary by construction. 
We leave the straightforward proof of the following lemma to the reader.
L EMMA 7.2.12. Let G be a group, and let N be a normal subgroup. Suppose that N has a
composition series
1 = H0 C H1 C · · · C Ht−1 C Hs = N
and G/N has a composition series
1 = Q0 C Q1 C · · · C Qr−1 C Qr = G/N.
For 1 ≤ i ≤ r, let Hs+i denote the unique subgroup of G containing N and such that Hs+i /N = Qi ,
which exists by Proposition 2.13.10. Then the series
1 = H0 C H1 C · · · C Ht−1 C Ht = G
is a composition series of G with composition factors satisfying Hs+i /Hs+i−1 = Qi /Qi−1 for
1 ≤ i ≤ r.
226 7. TOPICS IN GROUP THEORY

C OROLLARY 7.2.13. Let G be a group and N a normal subgroup. If N and G/N have
composition series, then G has a composition series. Moreover, its list of composition factors
consists of the concatenation of the list of composition factors of N by the list of composition
factors of G.
T HEOREM 7.2.14 (Jordan-Hölder theorem).
a. Every finite group has a composition series.
b. Let G be a nontrivial group with composition series
1 = N0 C N1 C · · · C Ns−1 C Ns = G
and
1 = H0 C H1 C · · · C Ht−1 C Ht = G.
Then s = t and there exists a permutation σ ∈ St such that
H /H ∼
= Ni /Ni−1
σ (i) σ (i)−1
for all 1 ≤ i ≤ t.
P ROOF. To show part a, we work by induction on the order n of the group G. It is clear in
the case that G is trivial, with t = 0. Now, if G is nontrivial of order n, then either it is simple,
and the composition series is 1 6 G, or it is not, and there exists a nontrivial normal subgroup
K C G, and then K and G/K have composition series by induction. The result is then immediate
from Lemma 7.2.13.
To see that the composition series is unique in the stated sense of part b, start with two
composition series as in the statement of the theorem. We work by induction on the minimal
length s of a composition series for G. If s = 0, then G is trivial. If s = 1, then G is simple,
so it cannot have a nontrivial normal subgroup, and all composition series must have length 1.
Consider N = Ns−1 , which has the composition series
1 C N1 C · · · C Ns−2 C Ns−1 = N,
as well as a composition series
1 C H f (1) ∩ N C · · · C H f (r−1) ∩ N C H f (r) ∩ N = N
for some r ≤ t and increasing f : Xr → Xt by Lemma 7.2.11a. Since the minimal length of
a composition series of N is less than s, we have by induction that r = s − 1 and there exists
σ ∈ Ss−1 such that
Ni /Ni−1 ∼
= (H ∩ N)/(H
f (σ (i)) ∩ N) ∼
=H
f (σ (i)−1) /H f (σ (i)) f (σ (i))−1
for all i, again by Lemma 7.2.11a.
Let k < t be maximal such that Hk−1 6 N. Then
Hk−1 ∩ N = Hk−1 6 Hk ∩ N < Hk .
Since Hk /Hk−1 is simple, this forces Hk−1 = Hk ∩ N. In particular, k is not in the image of f .
Moreover, we have
Hk /Hk−1 ∼= Hk /(Hk ∩ N) ∼
= Hk N/N ∼= G/N,
 7.3. SOLVABLE GROUPS 227

the latter step as N is a maximal normal subgroup of G and Hk 66 N. As we have found the final
composition factor in the series (Ni )i among those of the series (Hi )i , it remains only to show
that s = t. If (Hi ∩ N)/(Hi−1 ∩ N) nontrivial for any for i 6= k, then by Lemma 7.2.11c, the group
G/N has a composition series of length at least 2, but G/N is simple, so this is impossible. Thus,
r = t − 1 as well, as needed. 
D EFINITION 7.2.15. The Jordan-Hölder factors of a group G are the terms in a list of the
isomorphism classes of the composition factors in a composition series for G.
E XAMPLES 7.2.16.
a. The group Z/pn Z for a prime p has n copies of Z/pZ as its Jordan-Hölder factors, which
arise from its unique composition series
0 C hpn−1 i C hpn−2 i C · · · C hpi C Z/pn Z.
b. The group Z/6Z has two composition series
0 C h2i C Z/6Z and 0 C h3i C Z/6Z,
both of which have Jordan-Hölder factors Z/2Z and Z/3Z.
c. Let G be a nonabelian group order pq with p and q distinct primes and q ≡ 1 mod p. Then
G has a unique composition series 1 C Q C G, where Q has order q, and it has Jordan-Hölder
factors Z/pZ and Z/qZ.
d. For n ≥ 6, the group Sn has a unique composition series 1 C An C G with Jordan-Hölder
factors An and Z/2Z.
R EMARK 7.2.17. The set of Jordan-Hölder factors of a group tell us a great deal about the
structure of a group, but they do not tell us the group. For instance, Z/n2 Z and (Z/nZ)2 have
the same Jordan-Hölder factors for any n ≥ 2.

7.3. Solvable groups

D EFINITION 7.3.1. Let G be a group. The derived series of G is the unique descending series
(G(i) )i≥0
of subgroups of G with G(0) = G and G(i) = [G(i−1) , G(i−1) ] for all i ≥ 1.
N OTATION 7.3.2. Often, one writes G0 for G(1) = [G, G] and G00 for G(2) = [[G, G], [G, G]].
D EFINITION 7.3.3. A group G is solvable if its derived series is finite.
E XAMPLES 7.3.4.
a. The derived series of an abelian group satisfies G(i) = 1 for all i ≥ 1. Hence, abelian groups
are solvable.
b. The derived series of a nonabelian simple group G satisfies G(i) = G for i ≥ 0. Hence,
nonabelian simple groups are not solvable.
E XAMPLE 7.3.5. Let R be a commutative ring. Consider the group
n 1 a c  o
T = Heis(R) = 1 b | a, b, c ∈ R 6 GL3 (R).
1
228 7. TOPICS IN GROUP THEORY

The reader should verify that this group has commutator subgroup equal to its center, which is
n 1 0 c  o
Z(T ) = [T, T ] = 1 0 | c ∈ R 6 GL3 (R).
1
In particular, T (2)
= [Z(T ), Z(T )] = 0. In fact, the reader might find a rather canonical isomor-
phism from Heis(Z) to the group presented by hx, y, z | [x, y] = z, [x, z] = [y, z] = ei.
L EMMA 7.3.6. The groups G(i) for i ≥ 1 are characteristic subgroups of a group G.
P ROOF. First, Lemma 4.3.15c tells us that G(i) is characteristic in G(i−1) for each i ≥ 1, and
then the result follows recursively from Lemma 4.3.16. 
We can now prove the following equivalence of definitions of solvability.
P ROPOSITION 7.3.7. The following statements regarding a group G are equivalent:
i. G is solvable,
ii. G has a normal series with abelian composition factors, and
iii. G has a subnormal series with abelian composition factors.
P ROOF. That (i) implies (ii) is a consequence of the facts that the group G(i) are characteristic,
hence normal, and that G(i−1) /G(i) is the quotient of G(i−1) by its commutator subgroup, hence
abelian. That (ii) implies (iii) is obvious. So, suppose (iii) and let
G = N0 B N1 B · · · B Nt−1 B Nt = 1
be a subnormal series of length t. (Note the reversed indexing, as in Remark 7.2.5.) We claim that
G(i) 6 Ni for each i ≥ 0. For i = 0, we have G = G(0) = N0 . In general, suppose inductively that
G(i−1) 6 Ni−1 . Then G(i) 6 [Ni−1 , Ni−1 ] by definition, and we have [Ni−1 , Ni−1 ] 6 Ni as Ni−1 /Ni
is abelian. Therefore, we have that Gt = Nt = 1, and G is solvable. 
We also have the following.
P ROPOSITION 7.3.8.
a. Every subgroup of a solvable group is solvable.
b. Every quotient group of a solvable group is solvable.
c. If G is a group and N is a normal subgroup of G such that N and G/N are both solvable,
then G is solvable as well.
P ROOF. Let G be a group and N a normal subgroup. If G is solvable, then it has a composition
series with abelian factors, so N and G/N are solvable by Lemma 7.2.11. Part (iii) is a corollary
of Corollary 7.2.13, since the derived series of N and G/N have abelian composition factors. 
P ROPOSITION 7.3.9. A group G with a composition series is solvable if and only if it is finite
and its Jordan-Hölder factors are all cyclic of prime order.
P ROOF. If G has cyclic Jordan-Hölder factors, then G is solvable by Proposition 7.3.7. If G
is solvable and has a composition series, then the composition factors are abelian by Proposi-
tion 7.3.7 and the uniqueness in Theorem 7.2.14. As composition factors, they are also simple,
hence cyclic of prime order, from which it follows that G is finite. 
 7.4. NILPOTENT GROUPS 229

E XAMPLE 7.3.10. All groups of order pq for distinct primes p and q are solvable, as their
Jordan-Hölder factors are Z/pZ and Z/qZ.
D EFINITION 7.3.11. A Hall subgroup of a finite group G is a subgroup H such that |H| and
[G : H] are relatively prime.
7.4. Nilpotent groups
D EFINITION 7.4.1. Let G be a group. The lower central series of G is the unique descending
series (Gi )i≥1 of G with G1 = G and
Gi+1 = [G, Gi ] = h{[a, b] | a ∈ G, b ∈ Gi }i
for each i ≥ 1.
R EMARK 7.4.2. By convention, Gi starts with G1 = G, while G(i) starts with G(0) = G.
R EMARK 7.4.3. For a group G, we have G0 = G2 , but G00 = [G0 , G0 ] can be smaller than
G3 = [G, G0 ]. In fact, we clearly have G(n+1) 6 Gn for all n ≥ 1.
The reader will easily verify the following by induction.
L EMMA 7.4.4. The groups Gi in the lower central series of a group G are characteristic
subgroups of G.
D EFINITION 7.4.5. A group G is nilpotent if its lower central series is finite.
D EFINITION 7.4.6. The nilpotency class of a nilpotent group is the length of its lower central
series, which is to say the smallest n ≥ 0 such that Gn+1 = 1.
L EMMA 7.4.7. Let G be a group. Then G(i) 6 Gi for all i.
P ROOF. This is almost trivial by induction, as
G(i) = [G(i−1) , G(i−1) ] 6 [G, G(i−1) ] 6 [G, Gi−1 ] = Gi .

C OROLLARY 7.4.8. Nilpotent groups are solvable.
E XAMPLES 7.4.9.
a. The lower central series of an abelian group satisfies Gi = 1 for all i ≥ 1.
b. Let T be as in Example 7.3.5. Then T1 = Z(T ) and T2 = 1, so T is nilpotent.
c. Let G be the group Aff(R) of upper-triangular matrices in GL2 (R) with lower-right entry
1, as in Example 2.12.13). We have
     
a 0 1 b 1 b(1 − a)
, =
0 1 0 1 0 1
for all a ∈ R× and b ∈ R. It follows easily from this that
  
0 1 b
G2 = G = b∈R
0 1
230 7. TOPICS IN GROUP THEORY

and then Gi = G2 for all i ≥ 1. On the other hand, G0 is abelian, so G00 = 1. Thus, Aff(R) is
solvable but not nilpotent. (Note that R can be replaced by any nonzero commutative ring R in
which R× and 1 − R× have nontrivial intersection.)
We can give an alternative characterization of nilpotent groups through the ascending series
of the following definition.
D EFINITION 7.4.10. The upper central series of a group G is the unique ascending series
with Z 0 (G) = 1 and Z i+1 (G) equal to the inverse image of Z(G/Z i (G)) under the
(Z i (G))i≥0
projection map G → G/Z i (G) for all i ≥ 0.
R EMARK 7.4.11. For any group G, we have Z 1 (G) = Z(G), and Z i (G) 6 Z i+1 (G) for all
i ≥ 0.
L EMMA 7.4.12. If G is a nontrivial nilpotent group, then Z(G) 6= 1.
P ROOF. Let n be the nilpotency class of G. Then Gn is nontrivial but central in G since
[G, Gn ] = 1. 
P ROPOSITION 7.4.13. A group G is nilpotent if and only if Z i (G) = G for i sufficiently large.
In this case, the nilpotency class of G equals the smallest n such that Z n (G) = G, and we have
Gn+1−i 6 Z i (G) for all 1 ≤ i ≤ n.
P ROOF. The result is clear for abelian groupts, which are the nilpotent groups of nilpotency
class 1. Let Ḡ = G/Z(G) for brevity of notation.
Suppose that G is nilpotent of nilpotency class n ≥ 2. As Gn+1 = [G, Gn ] = 1, we have
1 6= Gn 6 Z(G). Since Z(G) is central in G, it follows that
Ḡi = Gi Z(G)/Z(G)
for all i. Since Gn−1 is not central in G by definition, we have that Ḡn−1 6= 1. By induction on n,
we then have
Z i−1 (Ḡ) 6= Ḡn−i 6 Z i (Ḡ)
for 1 ≤ i ≤ n − 1, and Z n−1 (Ḡ) = Ḡ. Taking the inverse images of these groups under the quotient
map from G, we obtain
Z i (G) 6= Gn−i 6 Z i+1 (G)
for 1 ≤ i ≤ n − 1 as well, and in particular Z n−1 (G) < Z n (G) = G.
Conversely, if Z n (G) = G for some minimial n ≥ 2, then Z n−1 (Ḡ) = Ḡ, so Ḡ is nilpotent of
nilpotency class n − 1 by induction on n. This means that Gn 6 Z(G), and therefore Gn+1 = 1,
so G is nilpotent. 
The following corollary can also be seen directly, using Proposition 4.9.6.
C OROLLARY 7.4.14. Finite p-groups are nilpotent.
P ROOF. This follows by induction on the order of a nontrivial p-group P, since Z(P) 6= 1,
and P/Z(P) is a p-group which we suppose by induction to be nilpotent. Then Z i+1 (P) is the
inverse image of Z i (P/Z(P)) in P, so P is nilpotent as well. 
 7.4. NILPOTENT GROUPS 231

T HEOREM 7.4.15 (Frattini’s argument). Let G be a finite group and N a normal subgroup.
Let P be a Sylow p-subgroup of N. Then G = N · NG (P).
P ROOF. For any g ∈ G, we have gPg−1 6 N, as N is normal, so gPg−1 is also a Sylow p-
subgroup of N. As such, it is conjugate to P in N, which is to say there exists a ∈ N such that
agP(ga)−1 = P, or ag ∈ NG (G). In other words, g ∈ N · NG (P). 

We are now ready to prove the following equivalent conditions for nilpotency.
T HEOREM 7.4.16. Let G be a finite group. Then the following are equivalent:
i. the group G is nilpotent,
ii. every proper subgroup of G is a proper subgroup of its normalizer in G,
iii. every Sylow p-subgroup of G is normal,
iv. G is the direct product of its Sylow p-subgroups,
v. every maximal proper subgroup of G is normal.
P ROOF. Suppose that G is nilpotent of nilpotence class n, and let H be a proper subgroup
of G. If HZ(G) = G, then H is a proper subgroup of NG (H) = G. Thus, we may suppose
that HZ(G) 6= G. As we always have that NG (HZ(G)) = NG (H), we may further assume that
Z(G) 6 H in proving (ii). In this case, H/Z(G) is a proper subgroup of G/Z(G), which has
nilpotence class less than n as Gn 6 Z(G), so (G/Z(G))n = 1. Thus, H/Z(G) is a proper subgroup
of NG (H/Z(G)) by induction, but the latter group is NG (H)/Z(G) since Z(G) P NG (H), and
therefore H/Z(G) is a proper subgroup of NG (H)/Z(G). Thus, (i) implies (ii).
Next, suppose (ii). If G is a p-group, (iii) obviously holds, so suppose this is not the case. Let
P be a Sylow p-subgroup of G for some p | |G|, and note that P < G. Let N = NG (P). By part (ii),
we have that P < N. Note also that P is a normal subgroup of NG (N) in that it is characteristic in
N, which forces NG (N) = N. Since (ii) holds, N cannot be proper in G, and thus P is normal in
G. Hence, (ii) implies (iii).
Suppose (iii). Let s be the number of primes dividing |G|, and let P1 , P2 , . . . , Ps be the distinct
Sylow subgroups of G. If s = 1, then we are done. In general, we set H = P1 . . . Ps−1 C G, and by
induction we have that H ∼ = P1 × P2 × · · · × Ps−1 . We then note that H ∩ Ps = 1 and HPs = G, so
G∼ = H × Ps . Hence, (iii) implies (iv).
Suppose (iv), and let M be a maximal proper subgroup of G. Let P1 , P2 , . . . Ps be the distinct
Sylow subgroups of G. If M ∩ Pi 6= Pi for some i, then M ∩ Pj = Pj for all j 6= i, since otherwise
M < MPi < G. Thus, M is the direct product of M ∩ Pi and the Pj for j 6= i. By the first Sylow
theorem, M ∩ Pi is normal in Pi , so M is normal in G. Thus, (iv) implies (v).
Suppose (v). Let P be a Sylow p-subgroup of G, and suppose it is not normal. Let M
be a maximal proper subgroup of G containing NG (P). Then M is normal in G, and Frattini’s
argument implies that G = MNG (P) = M, a contradiction. So, P is normal and (iii) holds, and so
(iv) holds. It then suffices to note that finite p-groups are nilpotent by Corollary 7.4.14, as this
tells us that (v) implies (i). 

The following is a useful fact regarding nilpotent groups.

232 7. TOPICS IN GROUP THEORY

P ROPOSITION 7.4.17. Let G be a nilpotent group, and let S be a subset of G with image in
Gab a generating set. Then S generates G.
P ROOF. We prove this by induction on the nilpotence class n of nilpotent groups G. It is clear
if G is abelian, or n = 1. For n ≥ 2, consider G/Gn , and note that its abelianization is Gab , so by
induction G/Gn is generated by the image of S. Thus, if we let H = hSi, we have G = Gn H. This
implies that H is normal in G since Gn 6 Z(G) and thus elements of Gn and of H normalize H.
We have that
Gn = [Gn H, Gn−1 ] = [H, Gn−1 ] 6 H,
the first equality as Gn H = G, the second as Gn 6 Z(G), and the third as H P G. It follows that
G = Gn H = H = hSi,
as claimed. 

7.5. Groups of order p3

We note the following useful fact.
L EMMA 7.5.1. Let G be a group such that G/Z(G) is cyclic. Then G is abelian, so G = Z(G).
P ROOF. Any b ∈ G − Z(G) has image generating G/Z(G), so G = Z(G)hbi. As b commutes
with itself and every element of Z(G), it is in the center of G, a contradiction. 
Let us classify the groups of order p3 for a prime number p.
T HEOREM 7.5.2. Let p be a prime number. There are exactly two isomorphism classes of
nonabelian groups of order p3 . These are represented by:
a. if p = 2, the dihedral group D4 and the quaternion group Q8 , and
b. if p is odd, the Heisenberg group Heis(Z/pZ) and the group
  
a b 2
K= ∈ Aff(Z/p Z) a ≡ 1 mod p .
0 1
P ROOF. Let G be a nonabelian group of order p3 . By Lemma 7.5.1, the quotient G/Z(G)
cannot be cyclic. This eliminates the possibility that |Z(G)| = p2 , since then G/Z(G) would be
cyclic of order p. Also, |Z(G)| = 6 1 as G is a p-group. Thus, we have |Z(G)| = p, and G/Z(G)
is a direct product of two cyclic groups of order p. Note that [G, G] 6 Z(G) since G/Z(G) is
abelian, which forces [G, G] = Z(G) since G is nonabelian. Let a, b ∈ G with images together
generating G/Z(G). Then G = ha, bi by Proposition 7.4.17, since finite p-groups are nilpotent.
Moreover, z = [b, a] generates Z(G), and note that this means ab = baz.
Now, suppose that G has an element of order p2 . Without loss of generality, we may suppose
that it is b. Then b p generates Z(G), so we have b pi = z for some i ∈ Z with 0 < i < p. We then
have ab = b1+pi a, which in particular tells us that hbi is a normal subgroup of G. Suppose that
we can also choose a to have order p. Then G = hbi o hai, and in fact G has a presentation
2
G∼
= ha, b | a p = b p = e, ab = b1+pi ai.
 7.5. GROUPS OF ORDER p3 233

If p is odd, then we have an isomorphism

   
1 + pi 0 1 1
f : G → K, f (a) = , f (b) = ,
0 1 0 1
and if p = 2, then f : G → D4 defined by f (a) = s, f (b) = r works as well.
Suppose now that we cannot choose a and b with G = ha, bi and either a or b of order p. If
p = 2, then this implies that a2 = b2 = (ab)2 = z, and from this one checks that G is a quotient
of the group with presentation
hx, y | x2 = y2 = (xy)2 , x4 = ei,
and the latter group is isomorphic to Q8 under the map which takes x to i and y to j. If p is odd,
then one sees that
(ba) p = b p a p z p(p−1)/2 = b p a p = z pi z p j = z p(i+ j)
for some i, j ∈ Z prime to p. Note that we have used p is odd here, since otherwise p(p − 1)/2 =
1, which is not a multiple of p = 2. If we replace a by ak where k ∈ Z with ik ≡ − j mod p, then
(ba) p = e, which yields a contradiction.
Finally, suppose that G has no element of order p2 . If p = 2, then z = [b, a] = (ba)2 = e,
which is a contradiction. If p is odd, then G is a quotient of the group that has presentation
hx, y | x p , y p , [x, y] p , [x, [x, y]], [y, [x, y]]i,
but Heis(Z/pZ) has this presentation, so it is isomorphic to G in that G has order p3 . 
 CHAPTER 8

Topics in ring theory

8.1. Localization of commutative rings

In this section, we let R be a commutative ring.
D EFINITION 8.1.1. A subset S of a commutative ring is multiplicatively closed if it is closed
under multiplication, 1 ∈ S, and 0 ∈
/ S.
To describe our first class of examples, we first make the following definition.
D EFINITION 8.1.2. An element a of a ring is nilpotent if there exists n ≥ 1 such that an = 0.
E XAMPLE 8.1.3. Given a ∈ R, the set {an | n ≥ 1} is multiplicatively closed if and only if a
is not nilpotent.
L EMMA 8.1.4. Let p be a prime ideal of R. Then Sp = R − p is a multiplicatively closed
subset of R.
P ROOF. If a, b ∈ S, then since a, b ∈
/ p and p is prime, we have ab ∈
/ p, so ab ∈ S. Moreover
0∈
/ S and 1 ∈ S by definition. 
Next, we consider rings of fractions, or localizations, for arbitrary commutative rings. First,
we prove a strengthening of Lemma 3.11.1.
L EMMA 8.1.5. Let S be a multiplicatively closed subset of R. The relation ∼ on R × S given
by (a, s) ∼ (b,t) if and only if there exists r ∈ S such that rat = rbs is an equivalence relation.
P ROOF. Let a, b, c ∈ R and s,t, u ∈ S. That ∼ is reflexive is the fact that ras = ras for any
r ∈ S, that it is symmetric is the fact that rat = rbs implies rbs = rat. If q, r ∈ S are such that
rat = rbs and qbu = qct, then multiplying the former equality by qu and then applying the latter,
we obtain
(rqt)au = q(rat)u = q(rbs)u = r(qbu)s = r(qct)s = (rqt)cs.
We have rqt ∈ S since S is multiplicatively closed, so (a, s) ∼ (c, u). Therefore, ∼ is transitive.

R EMARK 8.1.6. Let S be a multiplicatively closed subset of R. If S contains no zero divisors,
then the relation ∼ on R × S is more simply defined by (a, s) ∼ (b,t) if and only if at = bs. That
is, this implies rat = rbs for all r ∈ R, and likewise, the latter implies at = bs since r is not a zero
divisor.
D EFINITION 8.1.7. Let S be a multiplicatively closed subset of R. The equivalence class as
of a pair (a, s) ∈ R × S is called an fraction of R with denominator in S (or S-fraction), and the set
of such S-fractions is denoted S−1 R.
235
236 8. TOPICS IN RING THEORY

a at
R EMARK 8.1.8. Let S be a multiplicatively closed subset of R. By definition, we have s = st
for any a ∈ R and s,t ∈ S. We denote the fraction 1a more simply by a.
R EMARK 8.1.9. If we were to allow 0 ∈ S, then S−1 R would have just one element 0. The
condition that 1 ∈ S is not strictly necessary so long as S is nonempty, as we can set a = ass for
any s ∈ S anyway.
It is worth describing, for additional clarity, S−1 R in the case that R is a domain, prior to
treating the general case. In this case, S−1 R is a subset of the field of fractions Q(R), as ∼ on
R × S is the restriction of the previously defined relation on R × (R − {0}).
P ROPOSITION 8.1.10. Let S be a multiplicatively closed subset of a domain R. Then
na o
S−1 R = | a ∈ R, s ∈ S
s
is the smallest subring of Q(R) containing R in which every element of S is a unit.
P ROOF. We note that S−1 R ⊆ Q(R) is closed under addition and multiplication as S is mul-
tiplicatively closed, and it clearly contains additive inverses. It also contains all r ∈ R as r = rss
for any s ∈ S, so S−1 R is a subring of Q(R) containing R. We then have s · 1s = 1 in this ring, so
S ⊆ (S−1 R)× . Finally, as 1s is the inverse to s ∈ S in Q(R), any subring B of Q(R) containing R
with S ⊆ B× contains a · 1s = as for all a ∈ R, s ∈ S. That is, any such subring B contains S−1 R. 
We now turn to the general case.
T HEOREM 8.1.11. Let S be a multiplicatively closed subset of a commutative ring R. The set
S−1 R is a ring under addition and multiplication of fractions:
a b at + bs a b ab
+ = and · = .
s t st s t st
P ROOF. Suppose that (a, s) ∼ (a0 , s0 ), and let r ∈ R be such that ras0 = ra0 s. Then
r(at + bs)s0t = ras0tt + rbss0t = ra0 stt + rbss0t = r(a0t + bs0 )st,
so addition is well-defined, noting its symmetry. Similarly, we have
rabs0t = ra0 bst,
so multiplication is well-defined. By definition, addition and multiplication are commutative,
and associativity and distributivity of the two are exactly as in the proof of Theorem 3.11.6.
Moreover, for any s ∈ S, we have
a a 0 a a a 1 a
+0 = + = and ·1 = · = .
s s 1 s s s 1 s
Also, we have
a −a as + (−a)s
+ = = 0,
s s s2
so − as = −a −1
s . Thus, S R is a ring under addition and multiplication. 
 8.1. LOCALIZATION OF COMMUTATIVE RINGS 237

D EFINITION 8.1.12. The ring S−1 R consisting of S-fractions for a multiplicatively closed
subset S of a commutative ring R is called the ring of S-fractions of R, or the localization of R at
S.
T HEOREM 8.1.13. Let S be a multiplicatively closed subset of a commutative ring R.
a. There is a canonical ring homomorphism R → S−1 R given by φS (a) = a for all a ∈ R, and
its kernel is {a ∈ R | sa = 0 for some s ∈ S}. In particular, φS is injective if and only if S contains
no zero divisors.
b. Every element of S maps to a unit in S−1 R under φS . Moreover, φS is universal with this
property in the sense that if Q is a commutative ring and f : R → Q is a homomorphism such that
f (S) ⊂ Q× , then there is a unique homomorphism θ : S−1 R → Q such that f = θ ◦ φS .
P ROOF. That φS is a homomorphism is simply that
a b ab
φS (a)φS (b) = = = φS (ab)
11 1
for any s,t ∈ S. We have φS (a) = 0 for a ∈ R if and only if (a, 1) ∼ (0, 1), and therefore if and
only if ra = 0 for some r ∈ S. Thus, we have part a.
Note that 1s ∈ S−1 R is clearly a multiplicative inverse of φS (s) ∈ S−1 R. Define θ : S−1 R → Q
by θ ( as ) = f (s)−1 f (a). It is easily checked to be a homomorphism. It also restricts to f by
definition. To see that it is well-defined, note that if as = bt , then rat = rbs for some r ∈ R, and so
f (s)−1 f (a) = f (rst)−1 f (rat) = f (rst)−1 f (rbs) = f (t)−1 f (b).
Thus, we have part b. 
R EMARK 8.1.14. Theorem 8.1.13 tells us that if S has no zero divisors, then S−1 R is the
smallest commutative ring containing R in which every element of S is a unit. In particular, if
every element of S already is a unit in R, then S−1 R = R. If S has zero divisors, then the map
from S to the unit group of S−1 R is still injective.
D EFINITION 8.1.15. The total ring of fractions Q(R) of a commutative ring R is the local-
ization of R at the set of nonzero elements of R that are not zero divisors.
R EMARK 8.1.16. If R is an integral domain, then its total ring of fractions is its field of
fractions Q(R).
E XAMPLES 8.1.17. Let us examine a few localizations of the ring R = Z × Z, employing
Theorem 8.1.13 to determine their isomorphism classes.
a. The total ring of fractions of R is given by inverting the set
{(c, d) | c, d ∈ Z − {0}}.
The homomorphism induced by the inclusion of Z2 in Q2 is a ring isomorphism
Q(Z × Z) −∼→ Q × Q, (a,b)
7→ ( a , b ).
(c,d) c d
b. Take S = ((Z − {0}) × {0}) ∪ {(1, 1)}. Then the homomorphism f : R → Q given by
∼
f (a, b) = a for a, b ∈ Z induces an isomorphism S−1 R −
→ Q.
238 8. TOPICS IN RING THEORY

We study extensions and contractions of ideals by φS .

N OTATION 8.1.18. For an ideal I of a commutative ring R and a multiplicative set S in R, let
S−1 I denote the extension of I through φS .
P ROPOSITION 8.1.19. Let R be a commutative ring, and let S be a multiplicative subset of
R.
a. For any ideal I of R, we have
S−1 I = { as | a ∈ R, s ∈ S}
and
φS−1 (S−1 I) = {a ∈ R | Sa ∩ I 6= ∅}.
b. Every ideal of S−1 R is equal to S−1 I for some ideal I of R.
P ROOF. By definition, every element of S−1 I is an S−1 R-linear combination of fractions a1
with a ∈ I, so an R-linear combination of fractions as with a ∈ I and s ∈ S. But we can take
common denominators and use the fact that I is an R-ideal to write every such fraction as bt for
some b ∈ I and t ∈ S.
Let a ∈ R be such that we have s ∈ S with x = sa ∈ I. Then φS (a) = xs ∈ S−1 I, so a ∈
φS−1 (S−1 I). Conversely, if φS (a) = xs for some x ∈ I and s ∈ S, then rsa = rx for some r ∈ S, from
which it follows that Sa ∩ I 6= ∅. This proves part a.
Let J be an ideal of S−1 R, and set I = φS−1 (J). Then S−1 I ⊆ J by definition. On the other
hand, as ∈ J for some a ∈ R and s ∈ S, then 1a ∈ J, so a ∈ I, and therefore as ∈ S−1 I. Thus J = S−1 I,
and we have part b. 
P ROPOSITION 8.1.20. Let R be a commutative ring, and let S be a multiplicatively closed
subset of R. Then extension and contraction give mutually inverse maps between the set of prime
ideals of R disjoint from S and the set of prime ideals of S−1 R. Moreover, the extension of any
prime ideal of R that intersects S is S−1 R.
P ROOF. Let p be a prime ideal of R. Note that if p is not disjoint from S, then 1 = ss ∈ S−1 p
for any s ∈ S ∩ p, which means that S−1 p = S−1 R.
So, suppose that p ∩ S = ∅. We claim that its extension S−1 p is prime. For this claim, it
suffices to check that if as · bt = uc for some a, b ∈ R, c ∈ p and s,t, u ∈ S, then either a ∈ p or b ∈ p.
The equality implies that there exists r ∈ S such that rabu = rcst, and the right-hand side is in p
but ru ∈
/ p, so ab ∈ p, and we then need merely note that p is prime.
The ideal
I = φS−1 (S−1 p) = {a ∈ R | Sa ∩ p 6= ∅}
clearly contains p. On the other hand, if a ∈ I, then there exists s ∈ S such that sa ∈ p, so a ∈ p
as p ∩ S = ∅. That is, I = p.
The contraction of a prime ideal is always prime, and from what we have proven every prime
ideal of R that is disjoint from S is a contraction by φS of a prime ideal of S−1 R. On the other hand,
we have already argued that the extension of a prime ideal of R disjoint from S is a prime ideal of
S−1 R. Moreover, any prime ideal of S−1 R is the extension of an ideal of R by Proposition 8.1.19
 8.1. LOCALIZATION OF COMMUTATIVE RINGS 239

and then of a prime ideal as it is the extension of its contraction by Proposition 3.7.26. Such a
prime ideal must be disjoint from S by the above. Thus, we obtain the inverse bijections as the
restriction of the inverse bijections in Proposition 3.7.26. 
D EFINITION 8.1.21. Let R be a commutative ring, and let x ∈ R be a non-nilpotent element.
Then the localization of R with respect to x, denoted by R[x−1 ], is the ring S−1 R for S = {xn | n ≥
0}.
E XAMPLE 8.1.22. Let n ∈ Z. Then the ring Z[ n1 ] may be identified with the subset of Q
consisting of reduced fractions with denominator a product of powers of primes dividing n, or
equivalently, with denominator dividing a power of n. The distinct ideals of Z[ 1n ] are generated
by nonnegative a ∈ Z with (a, n) = 1.
E XAMPLE 8.1.23. Let R = Z × Z and x = (1, 0), and consider R[x−1 ]. Since x · (0, 1) = 0,
and x is invertible in R[x−1 ], we have (0, 1) = 0 in R[x−1 ]. Note also that (1, 0)n (a, 0) = (a, 0),
(a,0) −1 ] given by a 7→ (a, 0) is an
so (1,0)n = (a, 0). It follows that the ring homomorphism Z → R[x

isomorphism.
D EFINITION 8.1.24. Let R be a commutative ring, and let p be a prime ideal of R. Then
Rp = Sp−1 R, where Sp−1 = R − p, is referred to as the localization of R at p.
E XAMPLES 8.1.25. Let p ∈ Z be prime.
a. The localization Z(p) is the subring of Q consisting of reduced fractions with denominators
not divisible by p.
b. The ring Q[x](x) consists of rational functions with denominator not divisible by x.
c. The ring Z[x](x) equals Q[x](x) inside Q(x).
d. The ring Z[x](p,x) is the subring of Q(x) of rational functions with denominator having
nonzero constant term modulo p.
E XAMPLE 8.1.26. Consider R = Z × Z and its prime ideal p = pZ × {0}. Then Rp ∼
= Z(p) ,
where the localization map Z × Z → Z(p) sends (a, b) to a.
D EFINITION 8.1.27. A commutative ring R is local if it has a unique maximal ideal.
D EFINITION 8.1.28. The residue field of a local ring R with maximal ideal m is the field
R/m.
The first part of the following explains something of the meaning of the terminology “local-
ization.”
P ROPOSITION 8.1.29. Let p be a prime ideal of a commutative ring R.
a. The ring Rp is a local ring with maximal ideal pRp .
b. The proper ideals of Rp are exactly those of the form IRp for some ideal I of R contained
in p.
240 8. TOPICS IN RING THEORY

P ROOF. By Proposition 8.1.19b, every ideal of Rp has the form IRp for some ideal I of R. If
a ∈ R − p, then by definition a is invertible in Rp , hence aRp = Rp . Thus, IRp = Rp for every ideal
of R not contained in p. On the other hand IRp ⊆ pRp if I ⊆ p, so pRp is the unique maximal ideal
of Rp . 

We note the following lemmas.

L EMMA 8.1.30. Let R be a local ring and m be its maximal ideal. Then R× = R − m.
P ROOF. If (a) 6= R, then a is contained in a maximal ideal, which must be m. Conversely, if
a ∈ R× , then (a) = R, so a is not contained in m. 
L EMMA 8.1.31. Let m be a maximal ideal of a commutative ring R. Then the canonical ring
homomorphism R/m → Rm /mRm is an isomorphism.
P ROOF. Since nonzero maps of fields are injective, it suffices to see that the map is onto. If
r ∈ R and u ∈ R − m, then let v ∈ R be such that (u + m)(v + m) = 1. Then ur + mRm is the image
of vr + m. 
R EMARK 8.1.32. If p is a non-maximal prime ideal in a commutative ring R, then R/p →
Rp /pRp is injective but cannot be an isomorphism, since R/p is not a field but Rp /pRp is.
We mention the following result on ideals contained in the complements of multiplicative
sets.
T HEOREM 8.1.33. Let R be a commutative ring, let S be a multiplicative set, and let I be an
ideal of R disjoint from S. Then there exists an ideal of R containing I and maximal with respect
to the property that it is disjoint from S. Moreover, any such ideal is prime.
P ROOF. Let X denote the set of ideals of R that contain I and are disjoint from S. Since I ∈ X,
it is nonempty. If C is a chain in X, then the ideal J∈C C of R is clearly contained in X, hence
S

is an upper bound on C . By Zorn’s lemma, X contains a maximal element p.

If p is not prime, then we may choose ab ∈ p with a, b ∈ / p. Set I = (a) + p and J = (b) + p
so that IJ ⊆ p. By maximality of p, we have that there exist c ∈ I ∩ S and d ∈ J ∩ S. Since S
is a multiplicative set, we then have cd ∈ IJ ∩ S ⊆ p ∩ S, contradicting p ∈ X. Thus, p must be
prime. 

We next give an interesting characterization of unique factorization domains. First, a defini-

tion and a lemma.
D EFINITION 8.1.34. A multiplicative subset S of a commutative ring R is saturated if for all
a, b ∈ R with ab ∈ S, we have a, b ∈ S.
L EMMA 8.1.35. Let R be a domain. The set
S = {a ∈ R − {0} | a = up1 . . . pk for u ∈ R× and pi prime for 1 ≤ i ≤ k for some k ≥ 0}
is a saturated multiplicative set.
 8.2. FINITELY GENERATED AND NOETHERIAN MODULES 241

P ROOF. The set S is clearly multiplicative. We show it is saturated. Let a, b ∈ R such that
ab ∈ S. If ab ∈ R× , then a, b ∈ R× (as, for instance a · b(ab)−1 = 1). Suppose by induction
that if the product element ab can be written as a product of a unit times k − 1 prime elements
for k ≥ 1, the result holds. Write ab = p1 . . . pk with each pi prime. As pk is prime, it divides
either a or b. Without loss of generality, we suppose that pk | a, and we write a = pk a0 for some
a0 ∈ R. Then a0 b = p1 . . . pk−1 . By induction, we have a0 , b ∈ S. Then a = a0 pk ∈ S as well, as S
is multiplicative. 
T HEOREM 8.1.36. A domain R is a UFD if and only if every nonzero prime ideal of R contains
a prime element.
P ROOF. Suppose first that R is a UFD and p is a prime ideal of R. If a ∈ p − {0}, then we
may write a = π1 · · · πk with πi ∈ R irreducible, hence prime, for 1 ≤ i ≤ k. As p is prime, we
have π j ∈ p for some 1 ≤ j ≤ k.
Conversely, suppose that every nonzero prime ideal of R contains a prime element. Let S
be the saturated multipilicative set of Lemma 8.1.35. Suppose that a ∈ R − S is nonzero. Then
for any r ∈ R, we have ra ∈ / S as S is saturated. In other words, (a) is disjoint from S. By
Theorem 8.1.33, there exists a prime ideal p containing (a) with p ∩ S = ∅. By assumption, p
contains a prime element p. But p ∈ S by definition, a contradiction. Thus, S = R−{0}, and every
nonzero element in R factors as a unit times a product of prime elements. By Proposition 5.1.27,
we conclude that R is a UFD. 

8.2. Finitely generated and noetherian modules

D EFINITION 8.2.1. Let R be a ring. A left R-module M is said to be noetherian if its set of
submodules satisfies the ascending chain condition.
P ROPOSITION 8.2.2. A module M over a ring R is noetherian if and only if every submodule
of M is finitely generated over R.
P ROOF. If every submodule of M is finitely generated, then the union of every ascending
chain {Ni | i ≥ 1} of submodules of M is finitely generated, and each one of these generators is
contained in some Nk , so they are all contained in the largest Nk among these. Thus, the union is
actually equal to Nk , so the ACC holds.
On the other hand, if the ACC holds for M and N is an R-submodule of M, then we can
pick n1 ∈ N − {0} and then, if it exists, ni+1 ∈ N with n ∈ / Ni with Ni = ∑ij=1 Rni for each i. By
definition, Ni is properly contained in Ni+1 , so by the ACC, eventually we cannot continue the
process, which is to say that for some k, we have Nk = N, or in other words that M is generated
by {n1 , n2 , . . . , nk }. 
R EMARK 8.2.3. Finitely generated modules need not be noetherian. A ring is left noetherian
(i.e., satisfies the ascending chain condition on left ideals) if and only if it is noetherian as a left
module over itself. Yet, any ring is finitely generated as a left module over itself, being that it is
generated by 1.
242 8. TOPICS IN RING THEORY

L EMMA 8.2.4. Let R be a ring, let M be an R-module, and let N be an R-submodule of M.

If S is a generating set of N and T is a subset of M with image generating M/N, then S ∪ T
generates M.
P ROOF. If m ∈ M, then there exist n ∈ N, mi ∈ T , and bi ∈ R for 1 ≤ i ≤ j for some j such
that
j
m = n + ∑ bi mi ,
i=1
and then there exist ni ∈ S and ci ∈ R for 1 ≤ i ≤ k for some k such that
k
n = ∑ ci ni ,
i=1
so n is an R-linear combination of the mi and the ni . That is, M is generated by S ∪ T . 
C OROLLARY 8.2.5. Let R be a ring, let M be an R-module, and let N be a finitely generated
R-submodule of M such that M/N is also finitely generated. Then M is finitely generated.
L EMMA 8.2.6. Let R be a ring and N be a submodule of an R-module M. Then M is noether-
ian if and only if both N and M/N are noetherian.
P ROOF. If M is noetherian, then N is noetherian by definition. Moreover, the inverse image P
of any submodule O of M/N under the quotient map π : M → M/N is a submodule of M, hence
generated by some finite set S. Then π(S) generates O, so we conclude that M/N is noetherian.
If N and M/N are both noetherian and P is a submodule of M, then P ∩ N and P/(P ∩ N)
are finitely generated as submodules of N and M/N, respectively, so P is finitely generated by
Corollary 8.2.5. That is, M is noetherian. 
C OROLLARY 8.2.7. Finite direct sums of noetherian modules are noetherian.
P ROOF. If M = N ⊕ N 0 for R-modules M, N, and N 0 , then N 0 = M/N, so by the lemma, M is
noetherian if N and N 0 are. The result then follows by induction on the number of summands. 
P ROPOSITION 8.2.8. Every finitely generated left module over a left noetherian ring is noe-
therian.
P ROOF. Let M be a finitely generated left module over a noetherian ring R, and let N be a
submodule of M. Since M is finitely generated, there is a surjective R-module homomorphism
Rn → M for some n. Let P be the inverse image of N in Rn . The module N is generated by the
image of any set of generators of P under the quotient map P → N. So, we need only show that
any submodule P of Rn is finitely generated, which is to say that Rn is left noetherian. This is
true as R is a left noetherian R-module, and Rn is the direct sum of n copies of R. 
We next consider modules that satisfy the descending chain condition.
D EFINITION 8.2.9. Let X be a set with a partial ordering ≤. A descending chain on X is an
ascending chain with respect to the opposite partial ordering ≥ defined by x ≥ y if and only if
y ≤ x for x, y ∈ X.. We say that X satisfies the descending chain condition, or DCC, if it satisfies
the ACC with respect to ≥.
 8.2. FINITELY GENERATED AND NOETHERIAN MODULES 243

D EFINITION 8.2.10. We say that a module over a ring R is artinian if its set of submodules
satisfies the descending chain condition.
E XAMPLE 8.2.11. Any finite-dimension vector spaces over a field F is an artinian F-module.
L EMMA 8.2.12. Let R be a ring and N be a submodule of an R-module M. Then M is artinian
if and only if both N and M/N are artinian.
P ROOF. That M being artinian implies N and M/N are artinian is straightforward. If N and
M/N are artinian and (Mi )i≥1 is a descending chain in M, then there exists k ≥ 0 such that
Mi ∩ N = Mk ∩ N and (Mi + N)/N = (Mk + N)/N for all i ≥ k. But this can only happen if
Mi = Mk for all i ≥ k as well: if m ∈ Mk , then m ∈ Mi + N, so m = m0 + n for some m0 ∈ Mi and
n ∈ N, but then n = m − m0 ∈ Mk ∩ N ⊆ Mi , and therefore m = m0 + n ∈ Mi . 
L EMMA 8.2.13. If m is a maximal ideal in a noetherian ring, then R/mn is an artinian R-
module.
P ROOF. Note that R/m is a field, hence artinian as an R-module. By induction on n ≥ 1,
we may suppose that R/mn−1 is artinian as an R-module. By Lemma 8.2.12, it suffices to show
that mn−1 /mn is Artinian over R. Since R is noetherian, mn−1 is a finitely generated R-module,
and the images in mn−1 /mn of any list of generators span it as an R/m-vector space. Since it is
artinian as an R/m-module, it is also artinian as an R-module. 
D EFINITION 8.2.14. The Jacobson radical J(R) of a (possibly noncommutative) ring R is
the intersection of all left maximal ideals of R.
The following extends Lemma 8.1.30.
L EMMA 8.2.15. Let x ∈ R. Then x ∈ J(R) if and only if 1 − rx ∈ R× for all r ∈ R.
/ R× , then there exists a left maximal ideal m containing 1 − rx. Then
P ROOF. If 1 − rx ∈
rx ∈
/ m, so rx ∈
/ J(R), and therefore x ∈/ J(R). Conversely, if x ∈/ J(R), then there exists a left
maximal ideal m such that x ∈/ m. Then there exist r ∈ R and y ∈ m such that 1 = rx + y. Then
/ R× .
1 − rx = y ∈ 
T HEOREM 8.2.16 (Nakayama’s lemma). Let M be a finitely generated module over a com-
mutative ring R, and suppose that J(R)M = M. Then M = 0.
P ROOF. Let {m1 , m2 , . . . , mk } be a set of generators of M with k ≥ 1. Since m1 ∈ J(R)M, we
can find ai ∈ J(R) for 1 ≤ i ≤ k such that
k
m1 = ∑ ai mi .
i=1
Since (1 − a1 )m1 is contained in the submodule M 0 generated by m2 , . . . , mk . On the other hand,
1 − a1 ∈ R× by Lemma 8.2.15. But then m1 itself is contained in M 0 , which tells us that M 0 = M
and k is not minimal. That is, the minimal number of generators of M is zero. 
C OROLLARY 8.2.17. Let M be a finitely generated module over a local ring R, and suppose
that mM = M, where m is the maximal ideal of M. Then M = 0.
244 8. TOPICS IN RING THEORY

C OROLLARY 8.2.18. Let M be a finitely generated module over a local ring R with maximal
ideal m, and let X be a set of elements of M such that {m + mM | m ∈ X} generates M/mM as a
vector space over the residue field R/m. Then X generates M.
P ROOF. Let N be the submodule of M generated by X. Then N + mM = M, so every element
in M/N is the N-coset of some element of mM, which is to say that m(M/N) = M/N. By
Nakayama’s lemma, we have M/N = 0, so X generates M. 
E XAMPLE 8.2.19. Take the set of tuples (111, 107, 50), (23, −17, 41), and (30, −8, 104).
Suppose that we want to see if they generate the Q-vector space Q3 . It suffices, then, to see that
they generate the Z(p) -module Z3(p) for some prime p. Moreover, the map F p → Z(p) /pZ(p) is
an isomorphism, so by Corollary 8.2.18, it suffices to see that these tuples generate F3p . Modulo
2, they are (1, 1, 0), (1, 1, 1), and (0, 0, 0), so they do not generate F32 . However, modulo 3, they
are (0, −1, −1), (−1, 1, −1), and (0, 1, −1), which do in fact generate F33 , and thus the original
tuples generate Q3 .

8.3. Homomorphism groups

R EMARK 8.3.1. Let M be a left module over an R-algebra A. Then M is an R-module under
r · m = φ (r)m, where φ : R → Z(A) is given by the structure of A as an R-algebra.
D EFINITION 8.3.2. Let M and N be left modules over an R-algebra A. The homomorphism
group HomA (M, N) is the R-module of homomorphisms φ : M → N under the usual addition of
maps and the scalar multiplication (r · φ )(m) = r · φ (m) for r ∈ R and m ∈ M.
R EMARK 8.3.3. It is traditional to call HomA (M, N) a homomorphism group, even when it
has an additional R-module structure (for when we simply take R = Z, it is just a Z-module, or
abelian group).
E XAMPLE 8.3.4. Let R be a commutative ring. Then HomR (Rm , Rn ) is a free R-module of
rank mn, isomorphic to Mnm (R) via φ 7→ A with φ (e j ) = ∑ni=1 ai j e j .
E XAMPLE 8.3.5. Let m, n ≥ 1. Then HomZ (Z/mZ, Z/nZ) ∼ = Z/(m, n)Z. That is, an element
this group is completely determined of φ (1), and φ (1) has to be an element of order dividing m
n
in Z/nZ, so a multiple of gcd(m,n) .
In general, if M and N are A-modules with an additional right module structures that turn
them into bimodules, then we can consider transfer these structures to HomA (M, N), as we briefly
explore.
D EFINITION 8.3.6. Let A and B be algebras over a commutative ring R. We say that an
A-B-bimodule M is R-balanced if rm = mr for all r ∈ R and m ∈ M.
E XAMPLES 8.3.7.
a. If A is an R-algebra, then A is an R-balanced A-A-bimodule.
b. For a commutative ring R, the Mm (R)-Mn (R)-bimodule Mmn (R) is R-balanced.
 8.3. HOMOMORPHISM GROUPS 245

P ROPOSITION 8.3.8. Let A, B, and C be R-algebras, let M be an R-balanced A-B-bimodule,

and let N be an R-balanced A-C-bimodule. Then HomA (M, N) is an R-balanced B-C-bimodule
under the actions given by
(b · φ )(m) = φ (mb) and (φ · c)(m) = φ (m)c
for b ∈ B, c ∈ C, m ∈ M, and φ ∈ HomA (M, N).
Homomorphism groups behave well with respect to direct sums and products, as made pre-
cise in the following proposition.
P ROPOSITION 8.3.9. Let A be an R-algebra.
a. Let M be a left A-module, and let {N j | j ∈ J} be a collection of left A-modules. Then there
is a canonical isomorphism of left R-modules
 
HomA M, ∏ N j ∼ = ∏ HomA (M, N j ).
j∈J j∈J
b. Let N be a left A-module, and let {Mi | i ∈ I} be a collection of left A-modules. Then there
is a canonical isomorphism of left R-modules
M 
HomA Mi , N ∼ = ∏ HomA (Mi , N).
i∈I i∈I

P ROOF. Given a collection of A-module homomorphisms φ j : M → N j for j ∈ J, we define

Φ : M → ∏ j∈J N j by Φ(m) = (φ j (m)) j∈J , which is clearly an A-module homomorphism. Con-
versely, given Φ, we define φ j = π j ◦ Φ where π j : ∏k∈J Nk → N j is the projection map, and φ j
is then an A-module homomorphism. The bijection (φ j ) j∈J 7→ Φ is clearly a map of R-modules.
Thus, we have part a.
Now, given a collection of A-module homomorphisms ψi : Mi → N for i ∈ I, we define
Ψ : i∈I Mi → N by Ψ((mi )i∈I ) = ∑i∈I mi , which is well-defined as all but finitely many mi = 0
L

by definition of the direct sum. The map Ψ is then an A-module homomorphism. Conversely,
given Ψ, we define ψi (m) = Ψ(ιi (m)), where ιi : M → i∈I Mi is the inclusion. These are by
L

definition inverse associations, and the bijection (ψi )i∈I → Ψ is again clearly an R-module ho-
momorphism. 
Let us consider the example of a dual vector space.
D EFINITION 8.3.10. Let V be a vector space over a field K. The dual vector space is V ∗ =
HomK (V, K).
R EMARK 8.3.11. Note that V ∼
L
= i∈I K for any choice of basis, so
M 
V∗ ∼
= HomK K, K ∼= ∏ HomK (K, K) ∼ = ∏K
i∈I i∈I i∈I

by part b of Proposition 8.3.9. That is, V and V ∗ are not in general isomorphic, but they will be
so if V is finite-dimensional. However, this isomorphism is not canonical: it depends on a choice
of basis, which we next make explicit.
246 8. TOPICS IN RING THEORY

D EFINITION 8.3.12. Let V be an n-dimensional vector space over a field K, and let B =
{e1 , e2 , . . . , en } be a basis of V . The dual basis to B is the basis of V ∗ given by B∗ = { f1 , f2 , . . . , fn },
where for 1 ≤ i, j ≤ n, we have
fi (e j ) = δi j
We next consider the double dual V ∗∗ = (V ∗ )∗ of an arbitrary vector space. For a finite-
dimensional vector space, it is canonically isomorphic to V .
P ROPOSITION 8.3.13. Let V be a vector space over a field K. There is a canonical injec-
tion Φ : V → V ∗∗ of K-vector spaces given by F(v)( f ) = f (v) for v ∈ V and f ∈ V ∗ . It is an
isomorphism if V is finite-dimensional.
P ROOF. Let v ∈ V and f ∈ V ∗ . First, note that
Φ(v)( f + a f 0 ) = f (v) + a f (v0 ) = Φ(v)( f ) + aΦ(v)( f 0 ),
so Φ(v) ∈ V ∗∗ . Second, note that
Φ(av + v0 )( f ) = f (av + v0 ) = a f (v) + f (v0 ) = a · Φ(v)( f ) + Φ(v0 )( f ),
so Φ is a K-linear transformation. Third, note that if Φ(v) = 0, then Φ(v)( f ) = f (v) = 0 for
all f ∈ V ∗ . If v 6= 0, we can extend {v} to a basis B of V and define f ∈ V ∗ by f (v) = 1 and
f (w) = 0 for all w ∈ B − {v}. Thus, the fact that f (v) = 0 for all f ∈ V ∗ implies that v = 0, so Φ
is injective.
Now, suppose that V is n-dimensional, let {e1 , e2 , . . . , en } be a basis, and let { f1 , f2 , . . . , fn }
be its dual basis in V ∗ . If ϕ ∈ V ∗∗ , then set c j = ϕ( f j ) for each 1 ≤ j ≤ n. Then
 n   n 
Φ ∑ ϕ( fi )ei ( f j ) = f j ∑ ϕ( fi )ei = ϕ( f j )
i=1 i=1

for all j, so ϕ ∈ Φ(V ). That is, Φ is an isomorphism. 

8.4. Tensor products

D EFINITION 8.4.1. Let A be ring, let M be a right A-module, and let N be a left A-module.
The tensor product M ⊗A N of M and N over A is the abelian group that is the quotient of the free
abelian group with basis M × N by its subgroup generated by
i. (m + m0 , n) − (m, n) − (m0 , n) for all m, m0 ∈ M and n ∈ N,
ii. (m, n + n0 ) − (m, n) − (m, n0 ) for all m ∈ M and n, n0 ∈ N, and
iii. (ma, n) − (m, an) for all m ∈ M, n ∈ N, and a ∈ A.
The image of (m, n) in M ⊗A N is denoted m ⊗ n.
D EFINITION 8.4.2. Let A be ring, let M be a right A-module, and let N be a left A-module.
An element of M ⊗A N of the form m ⊗ n for some m ∈ M and n ∈ N is called a simple tensor.
R EMARK 8.4.3. Any tensor product M ⊗A N is generated as an abelian group by simple
tensors m ⊗ n for m ∈ M and n ∈ N. It is not in general equal to the set of such tensors.
 8.4. TENSOR PRODUCTS 247

P ROPOSITION 8.4.4. Let A be an algebra over a commutative ring R, let M be a right A-

module, and let N be a left A-module. The tensor product M ⊗A N is an R-module under the
unique action that satisfies
r(m ⊗ n) = mr ⊗ n = m ⊗ rn.
for all r ∈ R, m ∈ M, and n ∈ N.
P ROOF. If we consider the free abelian group on M × N as an R-module via r(m, n) = (mr, n)
for r ∈ R, m ∈ M, and n ∈ N, then the elements providing the relations in Definition 8.4.1 define
an R-submodule. Therefore, the quotient becomes an R-module under this action. 
R EMARK 8.4.5. If A is an R-algebra, the tensor product M ⊗A N is isomorphic to the quotient
of the free R-module on M × N by the submodule generated by the elements of Definition 8.4.1,
along with the elements r(m, n) − (mr, n) for r ∈ R, m ∈ M, and n ∈ N.
D EFINITION 8.4.6.
a. Let L, M, and N be abelian groups. A map φ : M × N → L is said to be bilinear if
φ (m + m0 , n) = φ (m, n) + φ (m0 , n) and φ (m, n + n0 ) = φ (m, n) + φ (m, n0 ).
for all m, m0 ∈ M and n, n0 ∈ N. Here, the first equality (for all m, m0 , and n) is referred to as left
linearity (or linearity in the first variable) and the second as right linearity.
b. Let L, M, and N be left modules over a commutative ring R. A bilinear map φ : M × N → L
satisfying
rφ (m, n) = φ (rm, n) = φ (m, rn)
for all r ∈ R, m ∈ M, and n ∈ N, then φ is said to be R-bilinear.
D EFINITION 8.4.7. Let A be a ring, let M be a right A-module, and let N be a left A-module.
A function φ : M × N → L is said to be A-balanced if φ (ma, n) = φ (m, an) for all a ∈ A.
R EMARK 8.4.8. Let A be an algebra over a commutative ring R, let M be a right A-module,
and let N be a left A-module. The tensor product M ⊗A N is endowed with an A-balanced R-
bilinear map
ιM,N : M × N → M ⊗A N, φ (m, n) = m ⊗ n,
as seen directly from the relations defining M ⊗A N.
The tensor product enjoys a universal property, exhibited in the following proposition.
P ROPOSITION 8.4.9. Let A be an algebra over a commutative ring R, let M be a right A-
module, let N be a left A-module, and let L be an R-module. Let φ : M × N → L be R-bilinear
and A-balanced. Then there exists a unique R-module homomorphism Φ : M ⊗A N → L such that
Φ(m ⊗ n) = φ (m, n) for all m ∈ M and n ∈ N.
P ROOF. We use the alternate construction of M ⊗A N of Remark 8.4.5. The map φ induces a
unique R-module homomorphism
M
F: R(m, n) → L, F((m, n)) = φ (m, n),
(m,n)∈M×N
248 8. TOPICS IN RING THEORY

since the direct sum is free. The R-bilinearity of φ tells us that the elements (m + m0 , n) − (m, n) −
(m0 , n), (m, n + n0 ) − (m, n) − (m, n0 ), and r(m, n) − (mr, n) lie its the kernel. The fact that φ is
A-balanced similarly tells us that the elements (ma, n) − (m, an) are contained in its kernel. The
first homomorphism theorem then provides an R-module homomorphism Φ : M ⊗A N → L with
Φ(m ⊗ n) = φ (m, n) for all m ∈ M and n ∈ N.
If Ψ : M ⊗A N → L is an R-module homomorphism also satisfying Ψ ◦ ιM,N = φ , then Ψ(m ⊗
n) = φ (m, n) = Φ(m ⊗ n) for all m ∈ M and n ∈ N, but the symbols m ⊗ n generate M ⊗A N as
an R-module, since the tensor product is defined as the quotient of the free R-module on M × N.
Therefore, we must have Φ = Ψ. 
R EMARK 8.4.10. The defining property of the map Φ : M ⊗A N → L of Proposition 8.4.9 is
stated more succinctly as Φ ◦ ιM,N = φ .
The reader may check the following, which gives the uniqueness of the tensor product up to
unique isomorphism as a module satisfying the universal property of the tensor product.
P ROPOSITION 8.4.11. Let A be an algebra over a commutative ring R, let M be a right
A-module, and let N be a left A-module. Let P be an R-module, and let λ : M × N → P be an R-
bilinear map such that for any R-bilinear, A-balanced map φ : M × N → L, there exists a unique
R-module homomorphism Φ : P → L such that Φ ◦ λ = φ . Then there is a unique isomorphism
ψ : P −∼
→ M ⊗A N such that ψ ◦ λ = ιM,N .
R EMARK 8.4.12. Let A be an algebra over a commutative ring R, let M be a right A-module,
and let N be a left A-module. For any m ∈ M and n ∈ N, we have m ⊗ 0 = 0 = 0 ⊗ n. For the first
equality, note that m ⊗ 0 = 0(m ⊗ 0) = 0.
We give an example by way of a proposition.
P ROPOSITION 8.4.13. Let m, n ≥ 1. Then (Z/mZ) ⊗Z (Z/nZ) ∼
= Z/(m, n)Z.
P ROOF. Let d = gcd(m, n), and write d = am + bn for some a, b ∈ Z. Note that x ⊗ y =
xy(1 ⊗ 1), so T = (Z/mZ) ⊗Z (Z/nZ) is cyclic, and moreover,
d(1 ⊗ 1) = (am + bn)(1 ⊗ 1) = a(m ⊗ 1) + b(1 ⊗ n) = 0,
so the order of T divides d.
We can define a bilinear map φ : Z/mZ×Z/nZ → Z/dZ by φ (x, y) = xy mod d for x ∈ Z/mZ
and y ∈ Z/nZ. We then have a homomorphism Φ : T → Z/(m, n)Z with Φ(1 ⊗ 1) = 1, and it is
therefore surjective. This forces |T | = d and Φ to be an isomorphism, as desired. 
P ROPOSITION 8.4.14. Let A be an R-algebra, let M be a right A-module, and let {Ni | i ∈ I}
be a collection of left A-modules. Then
M  M
M ⊗A Ni ∼= (M ⊗A Ni ).
i∈I i∈I
P ROOF. First, define an R-bilinear, A-balanced map
M  M
φ: M× Ni → (M ⊗A Ni ), φ (m, ∑ ni ) = ∑ m ⊗ ni .
i∈I i∈I i∈I i∈I
 8.4. TENSOR PRODUCTS 249

This induces an R-module homomorphism

M  M
Φ : M ⊗A Ni ∼
= (M ⊗A Ni )
i∈I i∈I
with Φ(m ⊗ ni ) = m ⊗ ni for m ∈ M and ni ∈ I for some i ∈ I.
Next, define R-bilinear, A-balanced maps
M 
ψi : M × Ni → M ⊗A Ni , ψi (m, ni ) = m ⊗ ni .
i∈I
The collection (ψi )i∈I gives rise to a unique R-module homomorphism
M M 
Ψ: (M ⊗A Ni ) → M ⊗A Ni ,
i∈I i∈I
satisfying Ψ(m ⊗ ni ) = m ⊗ ni for m and ni as above by Proposition 8.3.9b. By definition, the
maps Φ and Ψ are inverse to each other. 
P ROPOSITION 8.4.15. Let M and N be modules over a commutative ring R. Then there is a
unique isomorphism of R-modules
M ⊗ N −∼
→ N ⊗ M,
R Rm ⊗ n 7→ n ⊗ m.
P ROOF. Consider the R-bilinear map φ : M × N → N ⊗R M given by φ (m, n) = n ⊗ m. It
induces an R-module homomorphism Φ : M ⊗R N → N ⊗R M satisfying Φ(m ⊗ n) = n ⊗ m by the
universal property of the tensor product. It then has inverse the similarly defined map Ψ : N ⊗R
M → M ⊗R N with Ψ(n ⊗ m) = m ⊗ n. 
R EMARK 8.4.16. We can allow a tensor product over an arbitrary R-algebra A in Propo-
sition 8.4.15, but we obtain M ⊗A N ∼
= N ⊗Aop M as R-modules (noting that Aop has the same
R-module structure as A).
E XAMPLE 8.4.17. Let R be a commutative ring. The tensor product Rm ⊗R Rn is a free
R-module of rank mn with basis {ei ⊗ e j | 1 ≤ i ≤ m, 1 ≤ j ≤ n}. This follows immediately
from Proposition 8.4.14 (and Proposition 8.4.15) and the fact that R ⊗R R ∼
= R. Here, the latter
isomorphism is induced the R-bilinear map (x, y) 7→ xy, its inverse being the map R → R ⊗R R
with x 7→ x ⊗ 1.
P ROPOSITION 8.4.18. Let A and B be R-algebras. Let L be a right A-module, let M be an
R-balanced A-B-bimodule, and let N be a left B-module. Then there is a unique isomorphism of
R-modules
(L ⊗ M) ⊗ N −∼
A B → L ⊗ (M ⊗ N),A (l ⊗ m) ⊗ n 7→ l ⊗ (m ⊗ n).
B

P ROOF. Let φ : (L ⊗A M) × N → P be an R-bilinear, B-balanced map to some R-module P.

This gives rise to a map
ψ = φ ◦ (ιL,M × idN ) : L × M × N → P
which is R-linear in each variable separately and satisfies
ψ(la, m, n) = ψ(l, am, n) and ψ(l, mb, n) = ψ(l, m, bn)
250 8. TOPICS IN RING THEORY

for all a ∈ A, b ∈ B, l ∈ L, m ∈ M, and n ∈ N. In particular, for each l ∈ L, we obtain an R-module

homomorphism Ψl : M ⊗B N → P with Ψl (m ⊗ n) = ψ(l, m, n) by the fact that ψl : M × N → P
with ψl (m, n) = ψ(l, m, n) is R-bilinear and B-balanced. We then obtain an R-bilinear, A-balanced
map
θ : L × (M ⊗B N) → P, θ (l, m ⊗ n) = ψl (m ⊗ n) = ψ(l, m, n)
which in turn induces an R-module homomorphism
Θ : L ⊗A (M ⊗B N) → P, Θ(l ⊗ (m ⊗ n)) = ψ(l, m, n).
Since the elements l ⊗ (m ⊗ n) generate L ⊗A (M ⊗B N), this is the unique homomorphism that
agrees with ψ on these simple tensors. Since ψ(l, m, n) = φ (l ⊗ m, n), the R-module L ⊗A (M ⊗B
N) satisfies the universal property of the tensor product L ⊗A (M ⊗B N), hence is canonically
isomorphic to it via the indicated map, as in Proposition 8.4.11. 
L EMMA 8.4.19. Let A be an R-algebra. Let M and M 0 be right A-modules, and let N and
N0 be left A-bmodules. Let φ : M → M 0 and ψ : N → N 0 be homomorphisms of left and right
A-modules, respectively. Then there exists a homomorphism of R-modules
φ ⊗ ψ : M ⊗A N → M 0 ⊗A N 0 , (φ ⊗ ψ)(m ⊗ n) = φ (m) ⊗ ψ(n).
P ROOF. The map θ : M × N → M 0 × N 0 with θ (m, n) = φ (m) ⊗ ψ(n) is immediately seen to
be R-bilinear, and it is A-balanced since
φ (ma) ⊗ ψ(n) = φ (m)a ⊗ ψ(n) = φ (m) ⊗ aψ(n) = φ (m) ⊗ ψ(an).
Thus, it induces an R-module homomorphism M ⊗A N → M 0 ⊗A N 0 with the desired property. 
We can also form the tensor product of R-algebras.
P ROPOSITION 8.4.20. Let A and B be algebras over a commutative ring R. The tensor
product A ⊗R B is an R-algebra under the unique multiplication satisfying
(a ⊗ b)(a0 ⊗ b0 ) = aa0 ⊗ bb0 .
for a, a0 ∈ A and b, b0 ∈ B.
P ROOF. First, we should check the desired multiplication on A ⊗R B is well-defined. To start,
given a ∈ A and b ∈ B, we claim that the map A × B → A ⊗R B given by (a0 , b0 ) 7→ aa0 ⊗ bb0 is
R-bilinear (and therefore R-balanced). To see this, we merely note that a(ra0 )⊗bb0 = r(aa0 ⊗bb0 )
and a(a0 + a00 ) ⊗ bb0 = aa0 ⊗ bb0 + aa00 ⊗ bb0 . Therefore, we obtain a well-defined map
ψ : A × B → EndR (A ⊗R B), ψ(a, b)(a0 ⊗ b0 ) = aa0 ⊗ bb0 .
Note also that ψ is R-bilinear as well, so we obtain an R-module homomorphism A ⊗R B →
EndR (A ⊗R B), which we may rewrite then as a well-defined operation
(A ⊗R B) × (A ⊗R B) → A ⊗R B, (a ⊗ b, a0 ⊗ b0 ) 7→ aa0 ⊗ bb0 .
This operation is R-bilinear by what we have said. As it clearly satisfies (1 ⊗ 1)(a0 ⊗ b0 ) = a0 ⊗ b0 ,
so we need only observe its associativity to finish the proof of the result. This can be checked on
simple tensors, for which it is in an immediate consequence of the associativity of the operations
on A and B. 
 8.4. TENSOR PRODUCTS 251

P ROPOSITION 8.4.21. Let A and B be algebras over a commutative ring R. An abelian group
M that is a left A-module and a right B-module is an R-balanced A-B-bimodule if and only if it
is an A ⊗R Bop -module under the action (a ⊗ b)m = (am)b.
P ROOF. Let M be an R-balanced A-B-bimodule. We endow it with an A × Bop -action by
(a, b)m = amb. This is action is R-bilinear, so it factors through an action of A ⊗R Bop that clearly
satisfies (1 ⊗ 1)m = m and (a ⊗ b)(m + m0 ) = (a ⊗ b)m + (a ⊗ b)m0 and therefore makes M into
an A ⊗R Bop -module.
Conversely, if M is an A ⊗R Bop -module, it is in particular an R-balanced A-B-bimodule via
the actions am = (a ⊗ 1)m and mb = (1 ⊗ b)m, as the reader may quickly verify. 
When M and N have R-balanced bimodule structures, we can also attain a bimodule structure
on their tensor product.
P ROPOSITION 8.4.22. Let A, B, and C be R-algebras over a commutative ring R. Let M
be an A-B-bimodule and N be an R-balanced B-C-bimodule. Then M ⊗B N is an R-balanced
A-C-bimodule with respect to actions satisfying
a(m ⊗ n) = (am) ⊗ n and (m ⊗ n)c = m ⊗ (nc)
for all a ∈ A, c ∈ C, m ∈ M, and n ∈ N.
P ROOF. For a ∈ A and c ∈ C, we can define an R-bilinear map
φ : M × N → M ⊗B N, (m, n) 7→ (am) ⊗ (nc),
noting that
φa,c (rm + m0 , n) = (a(mr + m0 )) ⊗ (nc) = r(am ⊗ nc) + am0 ⊗ nc = rφ (m, n) + φ (m, n0 )
and similarly for the second variable. We thus have an induced map
Φa,c : M ⊗B N → M ⊗B N, m ⊗ n 7→ (am) ⊗B (nc)
of R-modules. The map
A ×Cop → EndR (M ⊗B N), (a, c) → Φa,c
then defines an R-algebra homomorphism. In other words, this gives M ⊗B N the structure of a
left A ⊗R Cop -module. 
We give an application.
P ROPOSITION 8.4.23. Let A be a ring, let M be a left A-module, and let I be a two-sided
ideal of A. Then there is an isomorphism of left A-modules
M/IM −∼→ A/I ⊗ M, A m + IM 7→ 1 ⊗ m.
P ROOF. Note that A/I is an A-A-bimodule, so A/I ⊗A M has the structure of an A-module
by Proposition 8.4.22. In one direction, we can define an A-module homomorphism Ψ : M →
A/I ⊗A M by Ψ(m) = 1⊗m. In the other, we can define an left A-linear, A-balanced map φ : A/I ×
M → M/IM by φ (a + I, m) = am + IM, which induces an A-module homomorphism Φ : A/I ⊗A
M → M/IM which is clearly inverse to Ψ. 
252 8. TOPICS IN RING THEORY

We have the following direct corollary.

C OROLLARY 8.4.24. Let A be a ring, and let M be a left A-module. Then M ∼
= A ⊗A M as
A-modules.
R EMARK 8.4.25. Note that Proposition 8.4.23 requires I to be a two-sided ideal, though the
definition of M/IM only requires I to be a left ideal. That is, we need a right A-action on A/I in
order to define A/I ⊗A M. We cannot take M ⊗A A/I either, as M is a left A-module.
C OROLLARY 8.4.26. Let M and N be free modules over a commutative ring R with bases B
and C, respectively. Then M ⊗R N is a free R-module with basis {m ⊗ n | m ∈ B, n ∈ C}.
P ROOF. Note that M ∼= m∈B R and N ∼
L L
= n∈C R via the isomorphisms given by the bases B
and C. Since tensor products and direct sums commute, and R ⊗R R ∼
= R by Corollary 8.4.24, we
have isomorphisms
M ⊗R N ∼ R ⊗R N ∼ R ⊗R R ∼
M MM M
= = = R
m∈B m∈B n∈C (m,n)∈B×C

such that the composite identifies m ⊗ n with 1 in the (m, n)-coordinate of the direct sum. In
particular, {m ⊗ n | m ∈ B, n ∈ C} maps to the standard basis of the direct sum, so is a basis of
M ⊗R N. 
The following corollary is also useful.
C OROLLARY 8.4.27. Let R be a commutative ring, let I be an ideal of R, let M and N be left
R-modules. Suppose that IM = 0, so M can also be viewed as an R/I-module. Then there is an
isomorphism of R-modules
∼
M ⊗R N −
→ M ⊗R/I N/IN, m ⊗ n 7→ m ⊗ (n + IN).
P ROOF. By Proposition 8.4.23, we have M ∼ = R/I ⊗R M and N/IN ∼
= R/I ⊗R N, so by asso-
ciativity and commutativity of tensor products over R, we have
M ⊗R N ∼ = R/I ⊗R M ⊗R N ∼ = M ⊗R N/IN
via the R-modules homomorphism which sends m ⊗ n to m ⊗ (n + IN) for m ∈ M and n ∈ N.
Finally, note that the canonical maps from M × N to simple tensor products in M ⊗R N/IN and
M ⊗R/I N/IN are both R/I-bilinear, and therefore induce inverse maps between M ⊗R N/IN and
M ⊗R/I N/IN. 
Here is an interesting comparison of tensor products and homomorphism groups in the case
of vector spaces.
L EMMA 8.4.28. Let V and W be finite-dimensional vector spaces over a field F. Then we
have an F-linear isomorphism
Ψ : V ∗ ⊗ W −∼
F→ Hom (V,W ), FΨ(φ ⊗ w)(v) = φ (v)w
for φ ∈ V ∗ , v ∈ V , and w ∈ W .
 8.4. TENSOR PRODUCTS 253

P ROOF. One checks directly that the map ψ : V ∗ × W → HomF (V,W ) with ψ(φ , w)(v) =
φ (v)w is F-bilinear, thus induces a map on the tensor product. Let B be a basis of V and C be a
basis of W . For each v ∈ B, and ϕ ∈ HomF (V,W ), write

ϕ(v) = ∑ av,ww.
w∈C

Define φw ∈ V ∗ for w ∈ C by φw (v) = av,w for v ∈ B. We can then define Θ : HomF (V,W ) →
V ∗ ⊗F W by
Θ(ϕ) = ∑ φw ⊗ w.
w∈C

By definition, Ψ(Θ(ϕ))(v) = φ (v) and

Θ(Ψ(φ ⊗ w)) = Θ(v 7→ φ (v)w) = φ ⊗ w.

T HEOREM 8.4.29. Let A and B be algebras over a commutative ring R. Let M be an R-
balanced A-B-bimodule, let N be a left B module, and let L be a left A-module. Then there is an
isomorphism of R-modules
Ξ : HomA (M ⊗B N, L) −∼
→ HomB (N, HomA (M, L))
given by
Ξ( f )(n)(m) = f (m ⊗ n)
for all f ∈ HomA (M ⊗B N, L), m ∈ M and n ∈ N.
P ROOF. First, define Ξ as in the statement of the theorem. Note that

Ξ( f )(n)(am + m0 ) = f ((am + m0 ) ⊗ n) = f (a(m ⊗ n) + (m0 ⊗ n))

= a f (m ⊗ n) + f (m0 ⊗ n) = aΞ( f )(n)(m) + Ξ( f )(n)(m0 ),

so Ξ( f )(n) is a homomorphism of A-modules. Moreover,

Ξ( f )(bn0 + n00 )(m) = f (m ⊗ (bn0 + n00 )) = f (mb ⊗ n0 ) + f (m ⊗ n00 )

= Ξ( f )(mb)(n0 ) + Ξ( f )(m)(n00 ) = (bΞ( f ))(m)(n0 ) + Ξ( f )(m)(n00 ),

so Ξ( f ) is a homomorphism of B-modules. Thus, Ξ is well-defined. In addition,

Ξ(r f )(n)(m) = (r f )(m ⊗ n) = f (r(m ⊗ n)) = f (mr ⊗ n)

= f (m ⊗ rn) = Ξ( f )(rn)(m) = (rΞ( f ))(n)(m)

so Ξ(r f ) = rΞ( f ), and since Ξ is also clearly a homomorphism of abelian groups, Ξ is a homo-
morphism of R-modules.
254 8. TOPICS IN RING THEORY

To finish the proof, we must exhibit an inverse to Ξ. For this, suppose we are given λ ∈
HomB (N, HomA (M, L)) and define φ : M × N → L by φ (m, n) = λ (n)(m). This map satisfies
φ (rm + m0 , n) = λ (n)(rm + m0 ) = rλ (n)(m) + λ (n)(m0 ) = rφ (m, n) + φ (m0 , n),
φ (m, rn + n0 ) = (rλ (n))(m) + λ (n0 )(m) = rλ (n)(m) + λ (n0 )(m) = rφ (m, n) + φ (m, n0 ),
φ (mb, n) = λ (mb)(n) = (λ (m)b)(n) = λ (m)(bn) = φ (m, bn),
φ (am, n) = λ (am)(n) = (aλ (m))(n) = aλ (m)(n) = aφ (m, n)
for all m, m0 ∈ M, n, n0 ∈ N, r ∈ R, a ∈ A, and b ∈ B. Thus, φ induces a unique map Φ : M ⊗B N →
L of A-modules with Φ(m ⊗ n) = λ (n)(m). The map λ → Φ is then by definition inverse to Ξ,
which tells us that Ξ is a bijection, hence an isomorphism. 
R EMARK 8.4.30. If we suppose in Theorem 8.4.29 that N is an R-balanced B-C-bimodule
and L is an R-balanced A-D-bimodule for R-algebras C and D, then the isomorphism Ξ is one of
R-balanced C-D-bimodules.

8.5. Exterior powers

In this section, R will denote a commutative ring.
D EFINITION 8.5.1. Let M be an R-module. For a nonnegative integer k, the kth tensor power
M ⊗k of M over R is the tensor product M ⊗R M ⊗R · · · ⊗R M of k copies of M if k is positive and
R if k = 0.
D EFINITION 8.5.2. Let M1 , M2 , . . . , Mk and N be R-modules for some k ≥ 1. A map f : M1 ×
M2 × · · · × Mk → N is said to be R-multilinear if it is R-linear in each of its k variables, which is
to say that
f (m1 , m2 , . . . , mi−1 , rmi + m0i , mi+1 , . . . , mk )
= f (m1 , m2 , . . . , mi−1 , rmi , mi+1 , . . . , mk ) + f (m1 , m2 , . . . , mi−1 , m0i , mi+1 , . . . , mk )
for r ∈ R and all m j and m0j ∈ M j for 1 ≤ j ≤ k.
The reader will quickly check the following.
P ROPOSITION 8.5.3. Let M1 , M2 , . . . , Mk and N be R-modules for some k ≥ 1. For an R-
multilinear map θ : ∏ki=1 Mi → N, there exists a unique R-module homomorphism
Θ : M1 ⊗R M2 ⊗R · · · ⊗R Mk → N
such that Θ(m1 ⊗ m2 ⊗ · · · ⊗ mk ) = θ (m1 , m2 , . . . , mk ) for all mi ∈ Mi with 1 ≤ i ≤ k.
D EFINITION 8.5.4. LetVM be a module over a commutative ring R. For a nonnegative integer
k, the kth exterior power k M is the quotient of M ⊗k by the R-submodule generated by the
elements of the form m1 ⊗ m2 ⊗ · · · ⊗ mk , where mi = m j for some 1 ≤ i < j ≤ k. The image of a
tensor m1 ⊗ m2 ⊗ · · · ⊗ mk in k M is denoted m1 ∧ m2 ∧ · · · ∧ mk .
V

R EMARK 8.5.5. The kth exterior power of a module M is often referred to as the wedge
product of M with itself k times.
 8.5. EXTERIOR POWERS 255

D EFINITION 8.5.6. Let M and N be abelian groups. A multilinear map f : M k → N is said to

be alternating if
f (m1 , m2 , . . . , mk ) = 0
for any m j ∈ M for 1 ≤ j ≤ k such that mi = mi+1 for some 1 ≤ i ≤ k − 1.
R EMARK 8.5.7. There is an alternating, R-bilinear map κ : M k → k M for any k ≥ 0 such
V

that κ(m1 , m2 , . . . , mk ) = m1 ∧ m2 ∧ · · · ∧ mk for all m j ∈ M for 1 ≤ j ≤ k

P ROPOSITION 8.5.8. Let M and N be R-modules, and let ψ : M k → N be R-multilinear and
alternating. Then there exists a unique R-module homomorphism Ψ : k M → N such that
V

Ψ(m1 ∧ m2 ∧ · · · ∧ mk ) = ψ(m1 , m2 , . . . , mk )
for all mi ∈ M for 1 ≤ i ≤ k.
P ROOF. Since ψ is R-multilinear, there exists by Proposition 8.5.3 a unique R-module ho-
momorphism Θ : M ⊗k → N with Θ(m1 ⊗ m2 ⊗ · · · ⊗ mk ) = ψ(m1 , m2 , . . . , mk ) for all mi ∈ M for
1 ≤ i ≤ k. If mi = mi+1 for some 1 ≤ i ≤ k − 1, then
Θ(m1 ⊗ m2 ⊗ · · · ⊗ mk ) = ψ(m1 , m2 , . . . , mk ) = 0
so Θ factors through the desired map Ψ : k M → N.
V
as ψ is alternating,
0 Vk
If Ψ : M → N also has the property of the proposition, then we may compose Ψ with the
quotient map κ : M ⊗k → N to obtain a map Θ0 = Ψ ◦ κ : M ⊗k → N that satisfies the universal
property of Proposition 8.5.3, hence is equal to Θ. This then forces the equality Ψ0 = Ψ for the
induced maps on the exterior product. 
We leave the following to the reader.
L EMMA 8.5.9. Let ϕV: M → N be aVhomomorphism of R-modules. Then for any k ≥ 0, there
exists a homomorphism k ϕ : k M → k N satisfying
V
^
k
( ϕ)(m1 ∧ m2 ∧ · · · ∧ mk ) = ϕ(m1 ) ∧ ϕ(m2 ) ∧ · · · ∧ ϕ(mk ).
L EMMA 8.5.10. Let M be an R-module. Then we have
m1 ∧ m2 ∧ · · · ∧ mk = −mτi (1) ∧ mτi (2) ∧ · · · ∧ mτi (k) ,
where τi = (i i + 1) ∈ Sk , for all 1 ≤ i ≤ k and all m j ∈ M for 1 ≤ j ≤ k.
P ROOF. The proof in the general case amounts to the following calculation in the case k = 2.
For any m, n ∈ M, we have
0 = (m + n) ∧ (m + n) = m ∧ m + m ∧ n + n ∧ m + n ∧ n = m ∧ n + n ∧ m,
so m ∧ n = −n ∧ m. 
R EMARK 8.5.11. The property that m ∧ n = −n ∧ m for all m, n ∈ M tells us directly that
m ∧ m = −m ∧ m, and so 2m ∧ m = 0, by taking m = n. In other words, if 2 is invertible in R, the
submodule of M ⊗R M generated by tensors of the form m ⊗ n + n ⊗ m contains the tensors of the
form m ⊗ m.
256 8. TOPICS IN RING THEORY
Vk
T HEOREM 8.5.12. Let M be a free
R-module of rank n. Then the kth
exterior power M of
M over R is a free R-module of rank nk for any k ∈ Z, where we take nk = 0 for k > n.
P ROOF. Let m1 , . . . , mn be a basis of M. The 0th exterior power is just R, so the result
holds for k = 0. For k ≥ 1, we know that M ⊗k is R-free with a basis of elements of the form
mi1 ⊗ mi2 ⊗ · · · ⊗ mik with 1 ≤ i j ≤ n for each 1 ≤ j ≤ k. Since we can switch the orders of
the terms of elements of k M with only a change of sign, we have that k M is generated by the
V V

mi1 ∧mi2 ∧ · · ·∧mik with 1 ≤ i1 ≤ i2 ≤ · · · ≤ ik ≤ n. But by definition of the exterior product, those
elements with i j = i j+1 for some j are
0, so it is generated by those with 1 ≤ i1 < i2 < · · · < ik ≤ n.
The number of such elements is nk .
It remains only to see R-linear independence. For this, fix 1 ≤ i01 < i02 < · · · < i0k ≤ n, and define
f : M k → R as the unique R-multilinear map satisfying that f (mi1 , mi2 , . . . , mik ) equals 0 unless
{i1 , . . . , ik } = {i01 , . . . , i0k }, in which case it is sign(σ ) for σ ∈ Sn such that σ (i j ) = i0j for 1 ≤ j ≤ k
and σ fixes every other element of {1, 2, . . . , n}. (Recall from Proposition 4.12.1 that the sign
map can be defined independently of the definition of the determinant, so as to avoid circularity
in our argument.) That this map is alternating can be easily checked: let m = ∑ni=1 ri mi ∈ M, and
consider
n n
f (. . . , m, m, . . .) = ∑ ∑ rir j f (. . . , mi, m j , . . .).
i=1 j=1
We then note that
f (. . . , mi , m j , . . .) + f (. . . , m j , mi , . . .) = 0
for i 6= j and f (. . . , mi , mi , . . .) = 0 for all i to see that the sum is trivial. The map f then induces
an element F ∈ HomR ( k M, R). Given some nontrivial R-linear combination x in k M of the
V V

generators mr1 ∧ mr2 ∧ · · · ∧ mrk with r1 < r2 < · · · < rk , the value F(x) is also the coefficient of
mi1 ∧ mi2 ∧ · · · ∧ mik in the linear combination x. So, if x = 0, then the linear combination must be
the zero linear combination, which verifies R-linear independence. 
C OROLLARY 8.5.13. The R-module n Rn is one-dimensional with basis vector e1 ∧ e2 ∧ · · · ∧
V

en , where {e1 , e2 , . . . , en } is the standard basis of Rn such that ei = (δi, j ) j with δi, j ∈ {0, 1} equal
to 1 if and only if i = j.
8.6. Graded rings
D EFINITION 8.6.1. A graded ring A is a ring determined by a sequence of abelian groups Ai
for i ≥ 0 and biadditive maps φi, j : Ai × A j → Ai+ j for i, j ≥ 0 satisfying
φi+ j,k (φi, j (ri , r j ), rk ) = φi, j+k (ri , φ j,k (r j , rk ))
for ri ∈ Ai , r j ∈ A j , rk ∈ Ak andL
i, j, k ≥ 0 and such that A0 is a ring with multiplication φ0,0 ,
where the additive group of A is ∞ i=0 Ai and the multiplication on A is given by
! !
∞ ∞ ∞ k
∑ ri · ∑ si = ∑ ∑ φi,k−i(ri, sk−i),
i=0 i=0 k=0 i=0
where the sums are finite and ri , si ∈ A for all i. The group Ai is called the degree i part, or ith
graded piece, of A, and an element of Ai is said to be homogeneous of degree i.
 8.6. GRADED RINGS 257

D EFINITION 8.6.2. A graded algebra over a commutative ring R is an R-algebra A that is a

graded ring with structure map R → A0 ∩ Z(A).
D EFINITION 8.6.3. For a commutative ring R, an homomorphism of graded R-algebras
ψ : A → B is a homomorphism of rings such that ψ(Ai ) ⊆ Bi for each i ≥ 0.
Clearly if R has a grading, then R is a graded ring with respect to the resulting subgroups and
maps.
D EFINITION 8.6.4. A grading on L∞
a ring R is a sequence of additive subgroups Ri with i ≥ 0
such that R0 is a subring and R = i=0 Ri such that the multiplication on R restricts to maps
φi, j : Ri × R j → Ri+ j for all i, j ≥ 0. We say that R is graded by the Ri .
E XAMPLE 8.6.5. Any commutative or noncommutative polynomial ring R on a set X has
a grading under which the nth graded piece is the R-span of of the words in X of length n. In
fact, there are many possible gradings by assigning arbitrary choices of positive degrees to the
different elements of X.
E XAMPLE 8.6.6. Given a ring A and an ideal I, we may form the graded ring grI A =
L∞ n n+1 , where the maps I i /I i+1 ×I j /I j+1 → I i+ j /I i+ j+1 are given by (x+I i+1 , y+I j+1 ) 7→
n=0 I /I
xy + I i+ j+1 . If A is an R-algebra, then grI A is a graded R-algebra via the map R → A/I.
We can form an algebra out of the tensor powers of a module.
D EFINITION 8.6.7. For a commutative ring R and n ≥ 0, the nth tensor power of an R-module
M is T n (M) = M ⊗n = M ⊗R · · · ⊗R M, the n-fold R-tensor product of M with itself, which is taken
to be R if n = 0.
D EFINITION 8.6.8. For a commutative ring R and nonzero R-module M, the tensor algebra
TR (M) of M is the graded R-algebra with ith graded piece T i (M) together with the unique R-
bilinear maps φi, j : T i (M) × T j (M) → T i+ j (M) satisfying
φi, j (m1 ⊗ · · · ⊗ mi , n1 ⊗ · · · n j ) = m1 ⊗ · · · ⊗ mi ⊗ n1 ⊗ · · · n j ,
where the R-algebra structure map is the identity R → T 0 (M)
E XAMPLE 8.6.9. The R-tensor algebra of R is isomorphic to R[x] as a graded R-algebra.
That is, we have an isomorphism ψ : R[x] → TR (R) of graded R-algebras uniquely determined by
ψ(x) = 1 ∈ T 1 (R). More generally, the R-tensor algebra of Rn is isomorphic to Rhx1 , . . . , xn i as a
graded algebra (where the xi have degree 1).
D EFINITION 8.6.10. A graded ideal of a graded ring is an ideal that has a homogeneous
generating set.
The reader can verify the following.
L∞
L EMMA 8.6.11. An ideal I of a graded ring A is homogeneous if and only if I = n=0 In ,
where In = An ∩ I for all n ≥ 0.
L EMMA 8.6.12. The quotient of a graded R-algebra A by a homogeneous ideal I is a graded
R-algebra with ith graded piece Ai /(Ai ∩ I), where Ai is the ith graded piece of A.
258 8. TOPICS IN RING THEORY

D EFINITION 8.6.13. Let R be a commutative ring and M be an R-module.

a. The symmetric algebra SR (M) on a R-module M is the quotient of TR (M) by the homoge-
neous ideal generated by the elements m ⊗ n − n ⊗ m with m, n ∈ R.
b. The nth graded piece Sn (M) of SR (M) is called the nth symmetric power of M.

N OTATION 8.6.14. For M an R-module and x, y ∈ TR (M), the image of their product x ⊗ y in
SR (M) is denoted x · y.

E XAMPLE 8.6.15. The symmetric algebra SR (Rn ) is isomorphic to R[x1 , . . . , xn ].

D EFINITION
V
8.6.16. Let R be a commutative ring and M be an R-module. The exterior
algebra R M on M is the quotient of TR (M) by the homogeneous ideal generated by the elements
m ⊗ m with m ∈ R.

N OTATION 8.6.17. For M an R-module and x, y ∈ TR (M), the image of their product x ⊗ y in
SR (M) is denoted x ∧ y.
V
L EMMA 8.6.18. V The multiplication on RM for an R-module M satisfies x∧x = 0 and x∧y =
−y ∧ x for all x, y ∈ R M.

P ROOF. For any x, y ∈ TR (M), we have

(x + y) ⊗ (x + y) = x ⊗ x + x ⊗ y + y ⊗ x + y ⊗ y,

which reduces the problem to proving that x ⊗ x lies in the homogeneous ideal I generated by the
m ⊗ m for m ∈ R. By the distributive property of multiplication, the result is further reduced to
the case of simple tensors. For m1 , . . . , mn ∈ R, we claim that

m1 ⊗ · · · ⊗ mn ⊗ m1 ⊗ · · · ⊗ mn ∈ I,

and for this it suffices to show that

m1 ⊗ (m2 ⊗ · · · ⊗ mn ) ⊗ m1 ∈ I,

This is clear if n = 1. For n ≥ 2, from the case n = 1 it follows that m1 ⊗ m2 − m2 ⊗ m1 ∈ I, which

reduces us to showing that
m1 ⊗ (m3 ⊗ · · · ⊗ mn ) ⊗ m1 ∈ I,
which now follows by induction. 

The reader can now verify the following.

8.6.19. For an R-module M and n ≥ 0, the nth graded piece of R M is isomor-

V
L EMMA
phic to n M under the R-linear map that takes the image of m1 ⊗ · · · ⊗ mn to m1 ∧ · · · ∧ mn for
V

m1 , . . . , mn ∈ M.
 8.7. DETERMINANTS 259

8.7. Determinants
In this section, R denotes a commutative ring.
D EFINITION 8.7.1. Let n ≥ 1.
a. The determinant det(A) of a matrix A ∈ Mn (R) with columns v1 , . . . , vn ∈ Rn is the unique
element of R such that
v1 ∧ v2 ∧ · · · ∧ vn = det(A) · e1 ∧ e2 ∧ · · · ∧ en ,
where ei denotes the ith element in the standard basis of Rn .
b. The determinant map
det : Mn (R) → R
is the map that takes a matrix to its determinant.

Vn n
R EMARK 8.7.2. The determinant map is an alternating, multilinear map if we view M (R) as n
R by taking a matrix to the wedge product v1 ∧ v2 ∧ · · · ∧ vn of its columns v1 , v2 , . . . , vn .
P ROPOSITION 8.7.3. The determinant map det : Mn (R) → R satisfies
det(A) = ∑ sign(σ )a1σ (1) a2σ (2) · · · anσ (n)
σ ∈Sn

for any A = (ai j ) ∈ Mn (R).

P ROOF. Let v1 , v2 , . . . , vn denote the columns of A = (ai j ). Then v j = ∑ni=1 ai j ei . We have
n n n
v1 ∧ v2 ∧ · · · ∧ vn = ∑ ∑ ··· ∑ ai11ai22 · · · ainn · ei1 ∧ ei2 ∧ · · · ∧ ein ,
i1 =1 i2 =1 in =1
but note that all the terms such that the j1 , j2 , . . . , jn are not all distinct are zero. The remaining
nonzero terms correspond to permutations σ ∈ Sn with σ ( j) = i j for each 1 ≤ j ≤ n. We then
have
v1 ∧ v2 ∧ · · · ∧ vn = ∑ aσ (1)1 aσ (2)2 · · · aσ (n)n · eσ (1) ∧ eσ (2) ∧ · · · ∧ eσ (n)
σ ∈Sn
= ∑ sign(σ )aσ (1)1 aσ (2)2 · · · aσ (n)n · e1 ∧ e2 ∧ · · · ∧ en
σ ∈Sn
= ∑ sign(σ )a1σ (1) a2σ (2) · · · anσ (n) · e1 ∧ e2 ∧ · · · ∧ en ,
σ ∈Sn

the latter step coming from rearranging the terms and replacing σ by σ −1 . 
L EMMA 8.7.4. Let A, B ∈ Mn (R) for some n ≥ 1. Then
det(AB) = det(A) det(B).
P ROOF. Let vi be the ith column of A, let wi be the ith column of B, and let zi be the ith
column of AB. Then Awi = zi for all i. By Lemma 8.5.9, we then obtain
z1 ∧ z2 ∧ · · · ∧ zn = det(A)w1 ∧ w2 ∧ · · · ∧ wn .
260 8. TOPICS IN RING THEORY

Since
w1 ∧ w2 ∧ · · · ∧ wn = det(B)e1 ∧ e2 ∧ · · · ∧ en ,
the result holds. 
D EFINITION 8.7.5. Let R be a ring. Two matrices A and A0 in Mn (R) for some n ≥ 1 are
called similar if there exists a matrix Q ∈ GLn (R) such that A0 = Q−1 AQ.
R EMARK 8.7.6. Let T : Rn → Rn be a linear transformation represented by the matrix A with
respect to the standard basis of Rn . If A0 = Q−1 AQ for Q ∈ GLn (R), then A0 represents T with
respect to the basis B = {v1 , . . . , vn } with v j = ∑ni=1 qi j ei for 1 ≤ j ≤ n. Conversely, any two
matrices that each represent T with respect to some basis are similar.
Lemma 8.7.4 has the following corollary.
C OROLLARY 8.7.7. Let R be a commutative ring.
a. For any A ∈ GLn (R), we have det(A) det(A−1 ) = 1.
b. Let A and B be similar matrices in Mn (R). Then det(A) = det(B).
L EMMA 8.7.8. Let A ∈ Mn (R), and let AT denote its transpose. Then det(AT ) = det(A).
P ROOF. Write A = (ai, j ). By Proposition 8.7.3, we have
det(AT ) = ∑ sign(σ )aσ (1)1 aσ (2)2 · · · aσ (n)n ,
σ ∈Sn
but aσ ( j) j = aσ ( j)σ −1 (σ ( j)) , so
aσ (1)1 aσ (2)2 · · · aσ (n)n = a1σ −1 (1) a2σ −1 (2) · · · anσ −1 (n) .
Noting that sign(σ ) = sign(σ −1 ) for all σ ∈ Sn , we have
det(AT ) = ∑ sign(σ −1 )a1σ −1 (1) a2σ −1 (2) · · · anσ −1 (n) = det(A).
σ ∈Sn

We also have the following standard properties of the determinant.
L EMMA 8.7.9. Let A ∈ Mn (R).
a. Let B be a matrix obtained by switching either two rows or two columns of A. Then
det(B) = − det(A).
b. Let C be a matrix obtained by adding an R-multiple of one row (resp., column) of A to
another row (resp., column). Then det(C) = det(A).
c. Let D be a matrix obtained by multiplying one row or column of A by some c ∈ R. Then
det(D) = c det(A).
P ROOF. By Lemma 8.7.8, it suffices to prove these for columns. Part a follows from the
more general fact that
vσ (1) ∧ vσ (2) ∧ · · · ∧ vσ (n) = sign(σ )v1 ∧ v2 ∧ · · · ∧ vn ,
 8.7. DETERMINANTS 261

and part b follows from

v1 ∧ · · · ∧ vi ∧ · · · ∧ (v j + rvi ) ∧ · · · ∧ vn
= (v1 ∧ · · · ∧ vi ∧ · · · ∧ v j ∧ · · · ∧ vn ) + r(v1 ∧ · · · ∧ vi ∧ · · · ∧ vi ∧ · · · ∧ vn )
= v1 ∧ · · · ∧ vi ∧ · · · ∧ v j ∧ · · · ∧ vn .

Part c follows from the multilinearity of the exterior product. 

L EMMA 8.7.10. Let A ∈ Mn (R) be a block diagonal matrix with A ∈ Mni (R) for 1 ≤ i ≤ m
and some m ≥ 1. Then det(A) = ∏m
i=1 det(Ai ).

P ROOF. We have

Ae1 ∧ Ae2 ∧ · · · ∧ Aen = det(A1 )(e1 ∧ · · · ∧ eni ) ∧ · · · ∧ det(Am )(en−nm +1 ∧ · · · ∧ en ),

as required. 

D EFINITION 8.7.11. For A ∈ Mn (R) and 1 ≤ i, j ≤ n, the (i, j)-minor of A is the matrix
Ai j ∈ Mn−1 (R) obtained by removing the ith row and jth column from A. The (i, j)-cofactor of
A is (−1)i+ j det(Ai j ).

P ROPOSITION 8.7.12 (Cofactor expansion). Let A = (ai j ) ∈ Mn (R). Then for any i with
1 ≤ i ≤ n, we have
n
det(A) = ∑ (−1)i+ j ai j det(Ai j ),
j=1

and for any j with 1 ≤ j ≤ n, we have

n
det(A) = ∑ (−1)i+ j ai j det(Ai j ).
i=1

P ROOF. The first follows from the second by taking the transpose. So, fix j. Denote the ith
column of A by vi . Set
(i)
wk = vk − aik ei

for each k, which is the column vector given by replacing the ith entry of vk by zero. We may
(i) (i) (i) (i)
then view w1 , . . . , w j−1 , w j , . . . , wn as the column vectors of the minor Ai j in the ordered basis
e1 , . . . , ei−1 , ei+1 , . . . , en . In particular, we have

(i) (i) (i) (i)

w1 ∧ · · · ∧ w j−1 ∧ w j+1 ∧ · · · ∧ wn = det(A)e1 ∧ · · · ∧ ei−1 ∧ ei+1 ∧ · · · ∧ en .
262 8. TOPICS IN RING THEORY

We then have
v1 ∧ v2 ∧ · · · ∧ vn = (−1) j−1 vi ∧ v1 ∧ · · · ∧ v j−1 ∧ v j+1 ∧ · · · ∧ vn
n
j−1
= (−1) ∑ ai j ei ∧ v1 ∧ · · · ∧ v j−1 ∧ v j+1 ∧ · · · ∧ vn
i=1
n
(i) (i) (i) (i)
= (−1) j−1 ∑ ai j ei ∧ w1 ∧ · · · ∧ w j−1 ∧ w j+1 ∧ · · · ∧ wn
i=1
n
= (−1) j−1 ∑ ai j det(Ai j )ei ∧ e1 ∧ · · · ∧ ei−1 ∧ ei+1 ∧ · · · ∧ en
i=1
n
= (−1)i+ j ∑ ai j det(Ai j )e1 ∧ e2 ∧ · · · ∧ en ,
i=1
where in the third equality we have applied Lemma 8.7.9(b). 
D EFINITION 8.7.13. Let A ∈ Mn (R). The adjoint matrix to A is the matrix with (i, j)-entry
(−1)i+ j det(A ji ).
T HEOREM 8.7.14. Let A ∈ Mn (R), and let B be its adjoint matrix. Then AB = det(A)In .
P ROOF. The (i, j)-entry of AB is ∑nk=1 (−1)k+ j aik det(A jk ). If i = j, this is just det(A) by
Proposition 8.7.12. If i 6= j, then the same proposition tells us that this equals the determinant
of a matrix which has the ith row of the matrix obtained by replacing the jth row of A by the ith
row of A. Since this matrix has two rows which are the same, its determinant is 0. 
C OROLLARY 8.7.15. A matrix A ∈ Mn (R) is invertible if and only if det(A) ∈ R× , in which
case its inverse is A−1 = det(A)−1 B, where B is the adjoint matrix to A.
As any two similar matrices in Mn (R) have the same determinant and any two matrices rep-
resenting a linear transformation are similar, the following is well-defined.
D EFINITION 8.7.16. Let V be a free R-module of finite rank. The determinant of an R-
module homomorphism T : V → V is the determinant of a matrix representing T with respect to
an R-basis of V .
R EMARK 8.7.17. Let T : V → V be a homomorphism of free R-modules, and let A be an
R-algebra. Then we have an A-module homomorphism idA ⊗T : A ⊗F V → A ⊗F V , which we
usually denote more simply by T . It satisfies T (a ⊗ v) = a ⊗ T (v) for any a ∈ A and v ∈ V .
D EFINITION 8.7.18.
a. The characteristic polynomial of a matrix A ∈ Mn (R) is cA (x) = det(xI − A).
b. The characteristic polynomial of an R-module homomorphism T : V → V with V a free
R-module of finite rank is cT (x) = det(x id −T ), where id denotes the identity map on F[x] ⊗F V .
D EFINITION 8.7.19. The trace of a matrix A = (ai j ) ∈ Mn (R) is
n
tr(A) = ∑ aii ∈ R.
i=1
 8.8. TORSION AND RANK 263

The trace is a homomorphism of additive groups.

L EMMA 8.7.20. If A, B ∈ Mn (R), then tr(A + B) = tr(A) + tr(B).
L EMMA 8.7.21. Let A ∈ Mn (R). The constant coefficient of cA (x) is (−1)n det(A), and the
coefficient of xn−1 is − tr(A).
P ROOF. We have cA (0) = det(−A) = (−1)n det(A). The second part is an easy consequence
of the permutation formula for the determinant applied to xIn − A, from which it is seen that
only the term corresponding to the identity of Sn has degree at least n − 1. This term is equal to
(x − a11 )(x − a22 ) · · · (x − ann ), and its xn−1 -coefficient is − tr(A). 
C OROLLARY 8.7.22. If A and B are similar matrices in Mn (R), then tr(A) = tr(B).
The reader may also verify the following directly.
L EMMA 8.7.23. Let A, B ∈ Mn (R). Then tr(AB) = tr(BA).
8.8. Torsion and rank
D EFINITION 8.8.1. Let M be a module over an integral domain R. We say that m ∈ M is an
R-torsion element if there exists a nonzero element a ∈ R with am = 0.
D EFINITION 8.8.2. Let M be a module over an integral domain R. Then M is said to be a
torsion module if all of its nonzero elements are R-torsion elements.
L EMMA 8.8.3. Let M be an module over an integral domain R. The set of R-torsion elements
of M is an R-submodule of M.
P ROOF. If m ∈ M and r ∈ R−{0} are such that rm = 0, then clearly r(am) = 0 for all a ∈ R as
well. If moreover m0 ∈ M and r0 ∈ R − {0} with r0 m0 = 0, then rr0 (m + m0 ) = r0 (rm) + r(r0 m0 ) =
0, and rr0 ∈ R − {0} as R is a domain. Thus, the set of R-torsion elements in M is indeed a
submodule. 
D EFINITION 8.8.4. Let M be a module over an integral domain R. The R-torsion submodule
Mtor of M is the set of R-torsion elements of M.
D EFINITION 8.8.5. Let R be a ring and M a left R-module. The annihilator of M in R is the
left ideal
Ann(M) = {r ∈ R | rm = 0 for all m ∈ M}
of R.
The reader will easily verify the following.
L EMMA 8.8.6. The annihilator Ann(M) of a left R-module over a ring R is a two-sided ideal
of R.
D EFINITION 8.8.7. Let R be a ring and M a left R-module. We say that an R-module M is
faithful if Ann(M) = 0.
R EMARK 8.8.8. Let R be an integral domain and M an R-module. If Ann(M) 6= 0, then M is
R-torsion since any nonzero r ∈ Ann(M) satisfies rm = 0 for all m ∈ M.
264 8. TOPICS IN RING THEORY

L EMMA 8.8.9. Let R be an integral domain, and let M be a finitely generated R-module.
Then Ann(M) 6= 0 if and only if M is R-torsion.
P ROOF. We may suppose that M is R-torsion. Let m1 , . . . , mn generate M, and let r1 , . . . , rn ∈
R − {0} be such that ri mi = 0 for 1 ≤ i ≤ n. Then r1 r2 · · · rn is a nonzero element of Ann(M). 
L∞
E XAMPLE 8.8.10. The abelian group n=1 Z/nZ is both faithful and torsion as a Z-module.
Let us introduce a general notion of rank for modules over integral domains.
D EFINITION 8.8.11. Let R be an integral domain, and let M be an R-module. The rank of
M over R, or R-rank of M, is the largest nonnegative integer n = rankR M such that M contains n
elements that are linearly independent over R, if it exists. If rankR M exists, then R is said to have
finite rank, and otherwise it has infinite rank.
For free modules over integral domains, this agrees with the notion of rank defined above.
We can give an alternative characterization of the rank. For this, we introduce the following
lemma.
L EMMA 8.8.12. Let ι : M → Q(R) ⊗R M be the R-module homomorphism defined by ι(m) =
1 ⊗ m for m ∈ M. Then ker ι = Mtor .
P ROOF. By Proposition 11.1.9, the module Q(R) ⊗R M is canonically isomorphic to the lo-
calization of M by S = R − {0}. The map ι becomes identified with the map M → S−1 M given
by m 7→ m1 . The definition of S−1 M tells us that m1 = 0 if and only if there exists r ∈ R − {0} such
that rm = 0, which is to say m ∈ Mtor . 
P ROPOSITION 8.8.13. Let R be an integral domain, and let M be an R-module. Then M has
finite rank over R if and only if Q(R) ⊗R M is finite-dimensional over Q(R), in which case
rankR M = dimQ(R) Q(R) ⊗R M.
P ROOF. First, suppose that m1 , m2 , . . . , mn are n elements of M. First, suppose that the ele-
ments mi are R-linearly dependent. Let ai ∈ R not all 0 be such that ∑ni=1 ai mi = 0. Then
n n n
∑ ai(1 ⊗ mi) = ∑ ai ⊗ mi = ∑ 1 ⊗ aimi = 0,
i=1 i=1 i=1
so the elements 1 ⊗ mi are Q(R)-linearly dependent.
Conversely, suppose that the elements 1 ⊗ mi are Q(R)-linearly dependent. Let αi ∈ Q(R)
with ∑ni=1 αi ⊗ mi = 0 and not all αi = 0. Let d ∈ R be such that ai = dαi ∈ R for all i, and set
m = ∑ni=1 ai mi . We then have
n n
1 ⊗ m = ∑ ai ⊗ mi = d ∑ αi ⊗ mi = 0,
i=1 i=1
so there exists r ∈ R−{0} with rm = 0 by Lemma 8.8.12. That is, the mi are R-linearly dependent.
In particular, if M has no maximal Q(R)-linearly independent subset, then neither does
Q(R) ⊗R M, so R has finite rank if and only if Q(R) ⊗R M is infinite dimensional.
 8.9. MODULES OVER PIDS 265

If M has a finite linear independent set X = {m1 , . . . , mn } of maximal order n, then the set of
1 ⊗ mi is a Q(R)-linearly independent subset of Q(R) ⊗R M of the same order. We have seen that
for any y ∈ Q(R) ⊗R M, there exists c ∈ R such that cy has the form 1 ⊗ n for some n ∈ M, and
we have then seen that y and the elements 1 ⊗ mi are linearly dependent, as X is maximal. Since
{1 ⊗ mi | 1 ≤ i ≤ n} can be extended to a basis of Q(R) ⊗R M, it must already then be a basis, so
n = dimQ(R) Q(R) ⊗R M. 
E XAMPLE 8.8.14. Consider any nonzero ideal I of a domain R. The usual method shows the
existence of a Q(R)-linear map Φ : Q(R) ⊗R I → Q(R) satisfying Φ( f ⊗ x) = x f for x ∈ I and
f ∈ Q(R). This map is clearly onto since Φ( 1x ⊗ x) = 1 for any nonzero x ∈ I, so
rankR I = dimQ(R) Q(R) ⊗R I ≥ 1.
On the other hand, if x ∈ I is nonzero, then any simple tensor g ⊗ y ∈ Q(R) ⊗R I can be rewritten
as g ⊗ y = gy
x ⊗ x. The Q(R)-linear transformation ψ : Q(R) → Q(R) ⊗R I given by ψ( f ) ⊗ f ⊗ x
is therefore onto, so rankR I ≤ 1. Thus, I has R-rank 1, but it is a free R-module if and only if it
is principal.
Though a bit off of the topic of this section, we also note the following consequence of
Nakayama’s lemma for free modules, which we shall employ later. In the case of local domains,
it is a considerable strengthening of the statement that the rank of a free module is the dimension
of its reduction modulo a maximal ideal.
L EMMA 8.8.15. Let M be a finitely generated free module over a local ring R with maximal
ideal m, and let X be a subset of M. If the image of X in M/mM is R-linearly independent, then
X is R-linearly independent and can be extended to a basis of M.
P ROOF. Let X̄ denote the image of X in M/mM. Extend X̄ to a basis B̄ of M/mM, and let
B ⊂ M be a lift of B̄ to M with X ⊆ B. Then B spans M by Corollary 8.2.18. To see that it is
linearly independent, suppose that B has n elements m1 , . . . , mn and consider the sum ∑ni=1 ai mi
for some ai ∈ R. Suppose that not all ai are zero, and let k ≥ 0 be minimal such that ai ∈ mk for
all i. Note that the map
mk /mk+1 ⊗R M → mk M/mk+1 M
induced by the R-action on M is an isomorphism by the freeness of M, since tensor products
commute with direct sums and it is clearly true for M = R.
By Corollary 8.4.27, we also have an isomorphism
∼
mk /mk+1 ⊗R M −
→ mk /mk+1 ⊗R/m M/mM
via the map induced by the identity in the first variable and the quotient map in the second. But
∑ni=1 ai ⊗ mi 6= 0 in the right-hand side since B̄ is a basis of M/mM and the tensor product is of
R/m-vector spaces. Therefore ∑ni=1 ai mi 6= 0. In other words B is a basis of M, and X is R-linearly
independent. 

8.9. Modules over PIDs

L EMMA 8.9.1. Let R be a PID. Any finitely generated R-submodule of Q(R) is cyclic.
266 8. TOPICS IN RING THEORY

P ROOF. Let M be an R-module generated by some subset {α1 , . . . , αn } of Q(R). Let d ∈

R − {0} be such that dαi ∈ R for all i. Then d : M → dM is an isomorphism, and dM is an ideal
of R, hence principal. That is, dM is cyclic as an R-module, so M is as well. 
P ROPOSITION 8.9.2. Let R be a PID. Let V be an n-dimensional Q(R)-vector space, and let
M be a finitely generated R-submodule of V . Then there exists a basis {v1 , v2 , . . . , vn } of V and
k ≤ n such that M is a free R-module with R-basis {v1 , v2 , . . . , vk }.
P ROOF. We suppose without loss of generality that M is nonzero. Pick a nonzero element
m1 ∈ M. Recall that R is noetherian as it is a PID. Then Q(R)m1 is a 1-dimensional Q(R)-vector
space, and M is noetherian being that it is R-finitely generated, so M ∩ Q(R)m1 is R-finitely
generated. Since Q(R)m1 is a 1-dimensional Q(R)-vector space with R-submodule M ∩ Q(R)m1 ,
Lemma 8.9.1 tells us that M ∩ Q(R)m1 = Rv1 for some v1 ∈ Q(R)m1 . Set M̄ = M/Rv1 . This is
an R-submodule of the (n − 1)-dimensional vector space V̄ = V /Q(R)v1 , since if x ∈ M is such
that x + Rv1 is in the kernel of M̄ → V̄ , then there exists α ∈ Q(R) such that x = αv1 . But then
x ∈ M ∩ Q(R)m1 , which is to say x ∈ Rv1 .
Now, by induction on n, there exist v2 , . . . , vk ∈ M for some k ≥ 1 such that v2 + Rv1 , . . . , vk +
Rv1 form an R-basis of M̄. Then v1 , v2 , . . . , vk ∈ M generate M by Lemma 8.2.4, and we claim
they are R-linearly independent. That is, if ∑ki=1 ci vi = 0 for some ci ∈ R, then
k
∑ ci(vi + Rm1) = 0,
i=2
and so ci = 0 for 2 ≤ i ≤ k. As v1 6= 0, this forces c1 = 0 as well. To finish, we merely extend
{v1 , . . . , vk } to a Q(R)-basis {v1 , . . . , vn } of V , noting that an R-linearly independent subset of V
is also Q(R)-linearly independent. 
C OROLLARY 8.9.3. Every finitely generated, torsion-free module over a principal ideal do-
main is free.
P ROOF. Let M be a finitely generated, torsion-free module over a PID R. We have seen in
Lemma 8.8.12 that the canonical map M → Q(R) ⊗R M is injective, in that Mtor = 0. The result
is then immediate from Proposition 8.9.2, as Q(R) ⊗R M is a finite-dimensional Q(R)-vector
space. 
C OROLLARY 8.9.4. Any submodule of a free module of rank n over a principal domain is
free of rank at most n.
P ROOF. Let R be a PID. Let M be a free R-module of rank n, and let N be an R-submodule
of M. Then N → Q(R) ⊗R M is injective, so we can apply Proposition 8.9.2. 
P ROPOSITION 8.9.5. Let M be a finitely generated module over a principal ideal domain R.
Then M ∼
= Rr ⊕ Mtor , where r is the rank of M.
P ROOF. Note that M/Mtor is free of finite rank by Corollary 8.9.3, so isomorphic to Rr for
some r. By Proposition 5.7.27, we have that M ∼= Mtor ⊕ Rr . 
Every Z-submodule of Z is free, but 0 and Z are the only direct summands of Z. On the
other hand, if we consider Z/nZ as a (Z/nZ)-module for some n ≥ 1, then 0 and Z/nZ are its
 8.9. MODULES OVER PIDS 267

only free subbodules, and of course they are direct summands. The following lemma implies
(via an application of the Chinese remainder theorem) that free submodules of finitely generated
modules over quotients of PIDs by nonzero elements are always direct summands.

L EMMA 8.9.6. Let R be a principal ideal domain, let π ∈ R be an irreducible element. Let
k ≥ 1 and set R̄ = R/(π k ). Then any free R̄-submodule F of a finitely generated R̄-module M is a
direct summand of M. If F is maximal, then M ∼ = F ⊕C, where π k−1C = 0.

P ROOF. We work by induction on k. Let M be an R̄-module and F a free submodule of M.

If k = 1, then M is a finite-dimensional R̄-vector space. Then F is a direct summand of M, since
any basis of it extends to a basis of M.
Now take k ≥ 2. Suppose first that A is a maximal free R̄-submodule of M. Consider the
subgroup

N = {m ∈ M | π k−1 m = 0}.

We have πA ⊆ N, which is an R/(π k−1 )-module. By induction on k, the free R/(π k−1 )-module
πA is a direct summand of N. We have N = πA ⊕C for some R̄-submodule C of A.
Note that N contains πM, and therefore any R/(π)-linearly independent subset of M/N lifts
an R/(π)-linearly independent subset of M/πM. Note that R̄ is a local ring with maximal ideal
generated by π. Any set of representatives in M of an R/(π)-basis of M/N is then R̄-linearly
indepedent by Lemma 8.8.15.
The canonical map f : A/πA → M/N is injective as A ∩ N = {a ∈ A | π k−1 A = 0} = πA by
the freeness of A. In particular, we have A ∩C = 0. The map f is also surjective: if it were not,
then by what we have shown, we could extend the image of a basis of A to a basis of M/N and
lift to obtain a free R̄-submodule of M properly containing A. We therefore have A + N = M, so
A +C = M. Thus M = A ⊕C.
It remains to show that an arbitrary free R̄-module F is a direct summand of M. It suffices to
show that F is a direct summand of a maximal free R̄-submodule A. For this, note that the map
F/πF → A/πA is injective, since if a ∈ πA ∩ F, then π k−1 a = 0, so a ∈ πF by the freeness of
F. We may then choose a set X that is a basis of a complement of F/πF in A/πA. Again by
Lemma 8.8.15, any lift of X to a linearly independent subset of A will span an R̄-complement to
F. Thus F is a direct summand of A. 

We are now ready to prove the structure theorem for finitely generated modules over principal
ideal domains.

T HEOREM 8.9.7 (Structure theorem for finitely generated modules over PIDs). Let R be a
PID, and let M be a finitely generated R-module.
a. There exist unique nonnegative integers r and k and nonzero proper principal ideals I1 ⊆
I2 ⊆ · · · ⊆ Ik of R such that

M∼
= Rr ⊕ R/I1 ⊕ R/I2 ⊕ · · · ⊕ R/Ik .
268 8. TOPICS IN RING THEORY

b. There exist unique nonnegative integers r and l, and for 1 ≤ i ≤ l, distinct nonzero prime
ideals pi of R and positive integers vi,1 ≥ vi,2 ≥ · · · ≥ vi,mi for some mi ≥ 1 such that
mi
l M
v
M∼
M
(8.9.1) = Rr ⊕ R/pi i, j .
i=1 j=1

Moreover, r and l are unique, and the tuple (pi , (vi, j ) j )i is unique up to ordering in i.
P ROOF. By Proposition 8.9.5, it suffices to consider the case that M is torsion. Note that the
uniqueness of r in parts (a) and (b) follows from the fact that r = dimQ(R) Q(R) ⊗R M = rankR M.
So, let M be a finitely generated torsion R-module.
We first demonstrate the existence of a decomposition as in part b. Let c ∈ R be a generator
of the annihilator Ann(M) of M. Since we have unique factorization in R, we may write c =
uπ1k1 · · · πlkl with u ∈ R× and with π1 , . . . , πr distinct irreducible elements of M and k1 , . . . , k;
positive integers for some k ≥ 0. By the Chinese remainder theorem, we have an isomorphism
l
R/(c) ∼ k
= ∏ R/(πi i )
i=1

of rings, which in turn provides a direct sum decomposition

l l
M = M/cM ∼
= M ⊗R R/(c) ∼ M ⊗R R/(πiki ) ∼ M/πiki M.
M M
= =
i=1 i=1

In particular, we are reduced to the case that M is a module over the local ring R/(π k ) for some
irreducible element π of R.
For the moment, suppose that M is a nonzero finitely generated R/(π k )-module for some
k ≥ 1. Note that if k = 1, then M is simply a finite dimensional R/(π)-vector space, so a choice
of basis gives a direct sum decomposition M ∼ = m
L
i=1 R/(π) for some m. For general k, let F be
k
a maximal free R/(π )-submodule of M. By Lemma 8.9.6, we have M = F ⊕ C, where C is a
finitely generated R/(π k−1 )-module. By induction on k, this gives the decomposition of part b.
We next prove the existence in part a using the decomposition in part b. Let us take πi to be
v v v
an irreducible element generating pi for each i. For j ≥ 1, set b j = π1 1, j π2 2, j · · · πl l, j , where we
take vi, j = 0 for j > mi . By construction, we have that b j+1 | b j for each j ≥ 1. Set I j = (b j ), and
let k be maximal such that Ik 6= R or zero if all Ik = R. Applying the Chinese remainder theorem
again to see that
l
v
R/πi i, j ∼
M
= R/I j
i=1
we obtain a decomposition as in part a.
Next, we exhibit uniqueness. If p = (π) is a nonzero prime ideal of R and v ≥ 0, then the map
R/p → pv /pv+1 induced by multiplication by π v is an isomorphism. Let N = R/qw for a nonzero
prime ideal q and w ≥ 0. Note that pv + qw = R if p 6= q, being that p and q are generated by
 8.10. CANONICAL FORMS 269

coprime elements. We therefore have

(
R/p if p = q and v < w
= pv /((pv+1 + qw ) ∩ pv ) ∼
pv N/pv+1 N ∼ =
0 if p 6= q or v ≥ w.
For any v ≥ 1, we then have that
pvi M/pv+1 ∼ u
i M = (R/pi ) ,
where u is the number of j such that v < vi, j . Thus, the pi and vi, j in any decomposition of M as
in part b are the same.
Finally, we reduce the uniqueness in part a to the known uniqueness of part b. Given any
decomposition M ∼ = R/I1 ⊕ · · · ⊕ R/Ik as in part a, we can again obtain a decomposition of M into
R modulo power of prime ideals, applying CRT to expand out each R/I j . Since I j ⊇ I j+1 , this
v v v
decomposition then satisfies I j = p11, j p22, j · · · pl l, j with vi, j ≥ vi, j+1 for each j ≥ 0. Thus, the de-
composition is as in part b, and by its uniqueness, we obtain the uniqueness of the decomposition
in part a. 
R EMARK 8.9.8. The structure theorem for finitely generated abelian groups is the special
case of the structure theorem for finitely generated modules over a PID for the PID Z.
D EFINITION 8.9.9. Let R be a PID, and let M be a finitely generated R-module.
a. The ideals I1 , I2 , . . . , Ik associated to M by Theorem 8.9.7a are called the invariant factors
of M.
v
b. The prime powers pi i, j associated to M by Theorem 8.9.7b are called the elementary divi-
sors of M.

8.10. Canonical forms

D EFINITION 8.10.1. Let V be a vector space over a field F, and let T : V → V be an F-linear
transformation.
a. An eigenvector v of T with eigenvalue λ ∈ F is an element v ∈ V − {0} such that T v = λ v.
b. An element λ ∈ F is called an eigenvalue of T if there exists an eigenvector in V with
eigenvalue λ .
c. The eigenspace of F for λ ∈ F of T is the nonzero subspace
Eλ (T ) = {v ∈ V | T (v) = λ v}
of V .
Note that Eλ (T ) = ker(T − λ idV ).
L EMMA 8.10.2. Let V be a vector space over a field F. The following are equivalent for an
F-linear transformation T : V → V and λ ∈ F:
i. Eλ (T ) 6= 0,
ii. λ is an eigenvalue of T , and
270 8. TOPICS IN RING THEORY

iii. cT (λ ) = 0.
P ROOF. The first two are equivalent by definition. Moreover, T − λ idV has a nonzero kernel
if and only if cT (λ ) = det(λ idV −T ) = 0. 
T ERMINOLOGY 8.10.3. We may speak of eigenvectors, eigenvalues, and eigenspaces Eλ (A)
of a matrix in A ∈ Mn (F), taking them to be the corresponding objects for the linear transforma-
tion T : F n → F n that A represents.
The following is the key to the application of the structure theorem for modules over PIDs to
linear algebra.
N OTATION 8.10.4. If T : V → V is a linear transformation and f = ∑ki=1 ci xi ∈ F[x], then we
set
n
f (T ) = ∑ ci T i : V → V,
i=1

where T i : V → V denotes the i-fold composition of T with itself.

R EMARK 8.10.5. If T is represented by a matrix A and f = ∑ki=1 ci xi ∈ F[x], then f (T ) is
represented by
k
f (A) = ∑ ci Ai ∈ Mn (F).
i=1

D EFINITION 8.10.6. Let T : V → V be an F-linear endomorphism of an F-vector space V .

The F[x]-module structure endowed on V by T is that which satisfies f (x) · v = f (T )v for all
T ∈ F[x].
This construction gives us one way to define the minimal polynomial of a linear transforma-
tion.
D EFINITION 8.10.7.
a. Let V be a finite-dimensional F-vector space. The minimal polynomial mT (x) of a linear
transformation T : V → V is the unique monic generator of the annihilator Ann(V ) under the
F[x]-module structure on V induced by T .
b. The minimal polynomial mA (x) is the minimal polynomial of the linear transformation
T : F n → F n that A represents with respect to the standard basis of F n .
L EMMA 8.10.8. The minimal polynomial of an endomorphism T of a finite-dimensional vec-
tor space V divides the characteristic polynomial of T .
P ROOF. Let v ∈ V . Then det(xI −T )·v = det(T −T )·v = 0 by definition, so cT (x) ∈ Ann(V ).

L EMMA 8.10.9. If A and B are similar matrices in Mn (F), then cA (x) = cB (x) and mA (x) =
mB (x).
 8.10. CANONICAL FORMS 271

P ROOF. Suppose Q ∈ GLn (F) is such that B = QAQ−1 . Then xI − B = Q(xI − A)Q−1 , so
cB (x) = det(xI − B) = det(xI − A) = cA (x).
Moreover, if g ∈ F[x] is such that g(A)v = 0 for all v ∈ F n , then
g(B)Qv = Qg(A)v = 0
for all v ∈ F n , so g(B) annihilates F n as well. By symmetry, we have mA (x) = mB (x). 
L EMMA 8.10.10. Let A be a block diagonal matrix with blocks Ai ∈ Mni (F) for 1 ≤ i ≤ m
and some m ≥ 1. Then
m
cA (x) = ∏ cAi (x),
i=1
while mA (x) is the least common multiple of the mAi (x) with 1 ≤ i ≤ m.
Suppose that we endow a finite-dimensional F-vector space V with the structure of an F[x]-
module through a linear transformation T : V → V . Since F[x] is a PID, the structure theorem for
modules over a PID tells us that there exists an F[x]-module isomorphism
m
V∼
M
= F[x]/( fi ),
i=1
where m ≥ 0 and the fi ∈ F[x] are monic, nonconstant polynomials for 1 ≤ i ≤ m such that fi+1 | fi
for 1 ≤ i < m.
L EMMA 8.10.11. Let f = ∑ni=0 ci xi ∈ F[x] be a monic polynomial of degree n ≥ 1. With
respect to the ordered basis {1, x, . . . , xn−1 } of V = F[x]/( f ) as an F-vector space, the linear
transformation given by multiplication by x on V is represented by the matrix

0 0 · · · 0 −c0

1 0 0 −c1 
. . .. .. 

A f = 0 1
 . . . .

. .
 .. . . . . . 0 −cn−2 


0 · · · 0 1 −cn−1
(This matrix is taken to be (−c0 ) if n = 1.)
P ROOF. Let T f : V → V be the linear transformation given by left multiplication by x. Note
that T f (xi ) = xi+1 for 0 ≤ i ≤ n − 2 and
n−1
T f (xn−1 ) = xn = − ∑ ci xi .
i=0
Thus, if A f = (ai, j ), we have ai+1,i = 1 for 1 ≤ i ≤ n − 1 and ai,n = −ci−1 for 1 ≤ i ≤ n, and all
other entries are zero. 
D EFINITION 8.10.12. For any monic, nonconstant f ∈ F[x], the matrix A f of Lemma 8.10.11
is known as the companion matrix to f .
272 8. TOPICS IN RING THEORY

L EMMA 8.10.13. If f ∈ F[x] is nonconstant and monic, then cA f (x) = mA f (x) = f .

P ROOF. The case n = 1 is clear. Write f = xg + c0 for some g ∈ F[x]. By induction of the
degree of f , we have
cA f = det(xI − A f ) = x det(xI − Ag ) + (−1)n−1 c0 det(−In−1 ) = xg + c0 = f .

We can make the following definition as a consequence of the structure theorem for F[x]-
modules.
D EFINITION 8.10.14. Let V L be a finite-dimensional F-vector space and T : V → V an F-
∼
linear transformation. Write V = m i=1 F[x]/( f i ) for some m ≥ 0 and monic f i ∈ F[x] with f i | f i+1
for all i < m. The rational canonical form of T is the block-diagonal matrix
A f1
 
 A f2 
 ... ,
 
A fm
where A fi is the companion matrix of fi .
R EMARK 8.10.15. The rational canonical form represents T with respect to the basis of V
determined by taking the image under the isomorphism i=1 F[x]/( fi ) −∼
Lm
→ V of the ordered basis
of the direct sum given by concatenating the bases {1, x, . . . , xdeg( fi )−1 } of the ith summands in
order of increasing i.
We also note the following.
R EMARK 8.10.16. By definition of rational canonical form, a matrix in rational canonical
form in one field is already in rational canonical form in any extension field.
D EFINITION 8.10.17. The rational canonical form of a matrix A ∈ Mn (F) is the rational
canonical form of the linear transformation that A represents with respect to the standard basis of
F n.
R EMARK 8.10.18. By Remark 8.10.15 and the change of basis theorem, A is similar to its
rational canonical form. Moreover, two matrices are similar if and only if they have the same
rational canonical form, since similar n-by-n matrices give rise to isomorphic F[x]-module struc-
tures on F n and conversely.
D EFINITION 8.10.19. The invariant factors of A ∈ Mn (F) are the monic generators in F[x]
of the invariant factors of F n viewed as an F[x]-module via the linear transformation represented
by A with respect to the standard basis.
As a simple consequence of Lemmas 8.10.13 and 8.10.10, we have the following.
L EMMA 8.10.20. Let f1 , f2 , . . . , fm be the invariant factors of a matrix A (with fi | fi+1 for
i < m). Then mA (x) = fm (x) and cA (x) = ∏m i=1 f i (x).
 8.10. CANONICAL FORMS 273

As a consequence, the irreducible divisors of cA (x) and mA (x) are the same. In particular, we
have:
C OROLLARY 8.10.21. Let A ∈ Mn (F) and λ ∈ F. The following are equivalent:
i. An element λ ∈ F is an eigenvalue of A ∈ Mn (F).
ii. The polynomial x − λ divides the minimal polynomial mA (x).
iii. The polynomial x − λ divides the characteristic polynomial cA (x).
This lemma is at times enough to calculate the rational canonical form of a matrix.
E XAMPLES 8.10.22. Let A ∈ Mn (F).
a. If f = cA (x) is a product of distinct monic, irreducible polynomials, then the rational
canonical form of A is A f .
b. If f = mA (x) has degree n, then the rational canonical form of A is A f .
c. If f = cA (x) = mA (x)d and g = mA (x) is irreducible of degree dn , then A has d invariant
factors of the form Ag .
d. Note that
  

0 0 0

 0  and 1 0
  

 0 0   0 0
1 0 1 0
are both 4-by-4 matrices in rational canonical form with characteristic polynomial x4 and mini-
mal polynomial x2 .
Recall that we have a second decomposition of V for the F[x]-module structure given by T .
That is, there exist distinct monic, irreducible polynomials p1 (x), p2 (x), . . . , pl (x) and positive
integers vi, j for 1 ≤ j ≤ mi for some mi ≥ 1 for 1 ≤ i ≤ l such that
mi
l M
V∼
M
= F[x]/(pi (x)vi, j ).
i=1 j=1
If the field F contains all the roots of cA (x), then it contains all the roots of the pi , so being
irreducible, these polynomials must be linear. This occurs, for instance, if F is algebraically
closed. Let us assume this is the case and write pi (x) = x − λi for some λi ∈ F.
L EMMA 8.10.23. Let V = F[x]/((x − λ )n ) for some λ ∈ F and n ≥ 0. The linear transfor-
mation given by multiplication by x on V is represented by the matrix
 
λ 1
.
λ ..
 
Jλ ,n = 
 
 . .. 1


λ
with respect to the ordered basis {(x − λ )n−1 , (x − λ )n−2 , · · · , x − λ , 1} of V .
274 8. TOPICS IN RING THEORY

P ROOF. The linear transformation T : V → V that is multiplication by x satisfies

T ((x − λ ) j ) = (x − λ ) j+1 + λ (x − λ ) j
for all j, with (x − λ ) j = 0 in V for j ≥ n. The result follows. 
D EFINITION 8.10.24. A matrix Jλ ,n of the form Lemma 8.10.23 is called a Jordan block of
dimension n for λ .
D EFINITION 8.10.25. Let V be a finite-dimensional F-vector space, and let T : V → V an
∼ Lm
F-linear transformation such that cT (x) splits in F. Write V = i=1 F[x]/((x − λi )ni ) for some
λi ∈ F and ni ≥ 1 for 1 ≤ i ≤ m and some m ≥ 1. The Jordan canonical form of T is a block-
diagonal matrix
Jλ1 ,n1
 
 Jλ2 ,n2 
 .. 
 . 
Jλm ,nm
whereJλi ,ni is the Jordan block of dimension ni for λi .
T ERMINOLOGY 8.10.26. If the characteristic polynomial of cT (x) splits in F, we say that T
has a Jordan canonical form over F.
The Jordan canonical form is unique up to ordering of the Jordan blocks.
D EFINITION 8.10.27. The Jordan canonical form of a matrix A ∈ Mn (F) is the Jordan canon-
ical form of the linear transformation that A represents with respect to the standard basis of F n .
R EMARK 8.10.28. Every (square) matrix has a rational canonical form, while every matrix
over an algebraically closed field has a Jordan canonical form.
P ROPOSITION 8.10.29. Suppose that T : V → V has a Jordan canonical form over F. Then
λ ∈ F is an eigenvalue of T if and only if it is a diagonal entry of the Jordan canonical form of
T.
P ROOF. Consider the isomorphism V ∼ = m ni
i=1 F[x]/((x − λi ) ) The image v ∈ V of (x −
L

λi )ni−1 in the ith term of the right-hand side of the above isomorphism is an eigenvector with
eigenvalue λi . On the other hand, if λ 6= λi , then (x − λ ) f 6= 0 for nonzero f ∈ F[x]/((x − λi )ni ),
so λ is not an eigenvalue of T . 
E XAMPLES 8.10.30. Let A ∈ Mn (F), and suppose that the characteristic polynomial of A
splits in F.
a. If cA (x) is a product of distinct linear factors, then the Jordan canonical form of A is
diagonal with entries the distinct eigenvalues of A.
b. If mA (x) is a product of distinct linear factors, then the Jordan canonical form of A is
diagonal with entries that are all distinct eigenvalues of A.
c. If mA (x) has degree n, then the rational canonical form of A is block-diagonal with Jordan
blocks Jλi ,ni , where the λi are all distinct.
 8.10. CANONICAL FORMS 275

d. If cA (x) = (x − λ )n for some λ ∈ F and mA (x) = x − λ , then A = Jλ ,n .

D EFINITION 8.10.31. Suppose that T : V → V . The generalized eigenspace of λ ∈ F for T
is the F[x]-submodule
{v ∈ V | (T − λ )n (v) = 0 for some n ≥ 0}
of V .
R EMARK 8.10.32. The generalized eigenspace of λ ∈ F under T : V → V contains the
eigenspace Eλ (T ) of λ .
E XAMPLE 8.10.33. The generalized eigenspace of
Jλ1 ,n1
 
 Jλ2 ,n2 
A= ..  ∈ Mn (F)
 . 
Jλm ,nm
is the span of the elements ei of the standard basis of F n for which the ith diagonal entry of A is
λi .
The following is an easy consequence of the example just given.
P ROPOSITION 8.10.34. Let T : V → V be a linear transformation. Then V is the direct sum
of its nontrivial generalized eigenspaces if and only if cT (x) splits in F.
We provide one example of how to obtain the rational and Jordan canonical forms of a matrix.
E XAMPLE 8.10.35. Let
 
2 −2 8
A = 0 3 −5 ∈ M3 (Q).
0 0 2
We have  
x−2 2 −8
cA (x) = det  0 x−3 5  = (x − 2)2 (x − 3).
0 0 x−2
So, Q3 is the direct sum of its generalized eigenspaces for 2 and 3. We compute that
   
0 −2 8 0 −2 10
A − 2 = 0 1 −5 and (A − 2)2 = 0 1 −5 ,
0 0 0 0 0 0
so ker(A − 2) = he1 i and ker((A − 2)2 ) = he1 , 5e2 + e3 i, where we use angle brackets to denote
the Q-span. Note that (A − 2)(5e2 + e3 ) = −2e1 . Similarly, we compute ker(A − 3) = h2e1 − e2 i.
The Jordan canonical form of the matrix A is then
 
2 1
 2 
3
276 8. TOPICS IN RING THEORY

with respect to the basis h−2e1 , 5e2 + e3 , 2e1 − e2 i.

By the Chinese remainder theorem, we have F 3 ∼ = F[x]/((x − 2)2 (x − 3)) under the F[x]-
3
module structure on F induced by A. Note that
(x − 2)2 (x − 3) = (x2 − 4x + 4)(x − 3) = x3 − 7x2 + 16x − 12.
To find a basis of the rational canonical form
 
0 0 12
1 0 −16
0 1 7
of A, we pick a vector v that generates F 3 as an F[x]-module, and then A is in rational canonical
form with respect to the basis {v, Av, A2 v}. To find v, note that
    
0 −2 8 −1 −2 8 0 0 2
(A − 2)(A − 3) = 0 1 −5  0 0 −5 = 0 0 0 ,
0 0 0 0 0 −1 0 0 0
so v = e3 works, and one possible basis is {e3 , 8e1 − 5e2 + e3 , 42e1 − 25e2 + 4e3 }.
D EFINITION 8.10.36. We say that a matrix A ∈ F n is diagonalizable if it is similar to a diag-
onal matrix. A linear transformation T : V → V is diagonalizable if and only if T is representable
by a diagonal matrix with respect to some basis of V .
Clearly, a linear transformation T : V → V is diagonalizable if and only if V is the direct sum
of its distinct eigenspaces. The following is then a special case of Proposition 8.10.34.
P ROPOSITION 8.10.37. A linear transformation T : V → V is diagonalizable if and only if
mT (x) splits in F.
 CHAPTER 9

Category theory

9.1. Categories
The extremely broad concept of a “category” allows us to deal with many of the constructions
in mathematics in an abstract context. We begin with the definition. We will mostly ignore set-
theoretical considerations that can be used to put what follows on a firmer basis, but note that a
class is a collection of objects that can be larger than a set, e.g., the class of all sets, in order that
we might avoid Russell’s paradox.
D EFINITION 9.1.1. A category C is
(1) a class of objects Obj(C ),
(2) for every A, B ∈ Obj(C ), a class HomC (A, B) of morphisms from A to B, where we often
use the notation f : A → B to indicate that f is an element of HomC (A, B), and
(3) a composition map
HomC (A, B) × HomC (B,C) → HomC (A,C)
for each A, B,C ∈ Obj(C ) that takes ( f , g) for f : A → B and g : B → C to the composition g ◦ f ,
subject to the properties that
i. for each A ∈ Obj(C ), there exists an identity morphism idA : A → A such that, for all
f : A → B and g : B → A with B ∈ Obj(C ), we have
f ◦ idA = f and idA ◦g = g,
and
ii. composition is associative, i.e.,
h ◦ (g ◦ f ) = (h ◦ g) ◦ f
for any three morphisms h : C → D, g : B → C, and f : A → B between objects A, B,C, D ∈
Obj(C ).
D EFINITION 9.1.2. We say that a category is small if its class of objects is a set.
R EMARK 9.1.3. What we call a category is often referred to as a locally category, and a
category in that terminology allows the morphisms between a pair of objects to form a class, not
just a set.
E XAMPLES 9.1.4.
a. The category Set which has sets as its objects and maps of sets as its morphisms.
277
278 9. CATEGORY THEORY

b. The category Gp which has groups as its objects and group homomorphisms as it mor-
phisms.
c. Similarly, we have categories Ring, the objects of which we take to be the (possibly zero)
rings with 1 and with morphisms the ring homomorphisms that preserve 1, and Field.
d. If R is a ring, then the category R-mod has objects the left R-modules and morphisms the
left R-module homomorphisms.
e. The category Top which has topological spaces as its objects and continuous maps as its
morphisms.
We may construct new categories out of old. The following provides a useful example.
D EFINITION 9.1.5. Let C and D be categories. The product category C × D is the category
with objects the pairs (C, D) with C ∈ Obj(C ) and D ∈ Obj(D) and morphisms ( f , g) : (C, D) →
(C0 , D0 ) for any f : C → C0 in C and g : D → D0 in D.
D EFINITION 9.1.6. Given a category C , we define the opposite category C op to have the
same class of objects as C and
HomC op (A, B) = HomC (B, A)
for A, B ∈ Obj(C ).
D EFINITION 9.1.7. A monoid G is a set with an associative binary operation and an identity
element for the operation.
E XAMPLE 9.1.8. Any monoid G gives rise to a category with one object, morphisms equal
to the elements of G, and composition law given by multiplication. Then Gop is again a monoid
with the same elements but the multiplication reversed. A category with one object is also called
a monoid, and we have a one-to-one correspondence between monoids and these categories.
We will often have cause to single out a particular class of morphisms in a category known
as isomorphisms.
D EFINITION 9.1.9. Let C be a category.
a. A morphism f : A → B in C is an isomorphism if there exists morphism g : B → A in C
such that g ◦ f = idA and f ◦ g = idB .
b. Two objects A and B in C are said to be isomorphic if there exists an isomorphism f : A →
B in C .
c. If f : A → B is a morphism and g ◦ f = idA (resp., f ◦ g = idB ), then we say that g is a right
inverse (resp., a left inverse) to f . If both g ◦ f = idA and f ◦ g = idB , then we say that g is (an)
inverse to f , or that f and g are inverse to each other (or mutually inverse, or inverses).
E XAMPLES 9.1.10.
a. The isomorphisms in Set are the bijections.
b. The isomorphisms in Gp are the isomorphisms of groups.
 9.2. FUNCTORS 279

c. The isomorphisms in Top are the homeomorphisms.

D EFINITION 9.1.11.
a. A morphism f : A → B in a category C is a monomorphism if for any g, h : C → A with
C ∈ Obj(C ), the property that f ◦ g = f ◦ h implies g = h.
b. A morphism f : A → B in a category C is an epimorphism if for any g, h : B → C with
C ∈ Obj(C ), the property that g ◦ f = h ◦ f implies g = h.
E XAMPLES 9.1.12.
a. In Set and R-mod, a morphism is a monomorphism (resp., epimorphism) if and only if it
is injective (resp., surjective).
b. The natural injection Z → Q in Ring is an epimorphism, since a ring homomorphism
Q → R is completely determined by its value on 1.
R EMARK 9.1.13. A morphism f : A → B in a category C is a monomorphism if and only if
the opposite morphism f op : B → A in C op is an epimorphism.
We have the following.
L EMMA 9.1.14. Let f : A → B and g : B → A be morphisms in a category C such that g ◦ f =
idA . Then f is a monomorphism and g is an epimorphism.
P ROOF. Let h, k : C → A be morphisms such that f ◦ h = f ◦ k. Then
k = g ◦ f ◦ k = g ◦ f ◦ h = h.
Thus f is a monomorphism. Similarly, g is an epimorphism, or apply Remark 9.1.13. 
In other words, right inverses are monomorphisms and left inverses are epimorphisms.
D EFINITION 9.1.15. Let C be a category and C ∈ Obj(C ).
a. A subobject of C is a pair (A, ι) consisting of an object A and a monomorphism ι : A → C.
b. A quotient of C is a pair (B, π) consisting of an object B and an epimorphism π : C → B.
D EFINITION 9.1.16. A subcategory C of a category D is a category with objects consisting
of a subclass of Obj(D) and morphisms HomC (A, B) for A, B ∈ Obj(C ) consisting of a subset
of HomD (A, B) containing idA for A = B and such that composition maps in C agree with the
restriction of the composition maps in D between the same objects.
E XAMPLES 9.1.17.
a. The category Ab of abelian groups with morphisms the group homomorphisms between
abelian groups is a subcategory of Gp.
b. The category Field is a subcategory of Ring.
9.2. Functors
To compare two categories, we need some notion of a map between them. Such maps are
referred to as functors. There are two basic types.
280 9. CATEGORY THEORY

D EFINITION 9.2.1. Let C and D be categories.

a. A covariant functor (or simply functor) F : C → D between two categories C and D is a
map of objects F : Obj(C ) → Obj(D) and a map of morphisms
F : HomC (A, B) → HomD (F(A), F(B))
for each A, B ∈ Obj(C ) such that F(idA ) = idF(A) and F(g ◦ f ) = F(g) ◦ F( f ) for all f : A → B
and g : B → C for each A, B,C ∈ Obj(C ).
b. As with a covariant functor, a contravariant functor F : C → D is again a map on objects,
but with maps between sets of morphisms of the form
F : HomC (A, B) → HomD (F(B), F(A))
that satisfies F(idA ) = idF(A) and F(g ◦ f ) = F( f ) ◦ F(g).
We give some examples of functors.
E XAMPLES 9.2.2.
a. We have the forgetful functors Gp → Set, Ring → Set, and Top → Set, which take objects
to their underlying sets and morphisms to the corresponding set-theoretic maps.
b. We have another forgetful functor from R-mod to the category Ab of abelian groups.
c. A homomorphism of monoids G → G0 induces a functor of the corresponding categories,
and conversely.
d. The opposite functor op : C → C op that is the identity on objects and takes a morphism
f : A → B to its opposite morphism f op : B → A in C op is contravariant.
R EMARK 9.2.3. A contravariant functor F : C → D may also be viewed as a covariant func-
tor C → D op , in particular by composing F with the opposite functor op : D → D op .
R EMARK 9.2.4. A subcategory C of a category D is endowed with a canonical inclusion
functor that takes an object of C to the same object of D and is the identity map on morphism.
D EFINITION 9.2.5. Let F : C → D be a functor.
a. The functor F is called faithful if it is one-to-one on morphisms.
b. The functor F is called full if it is onto on morphisms.
c. A functor F is fully faithful if it is both faithful and full.
d. A subcategory is called a full subcategory if it the corresponding inclusion functor is full.
R EMARK 9.2.6. Every functor takes isomorphisms to isomorphisms.
R EMARK 9.2.7. The inclusion functor attached to a subcategory is always faithful.
R EMARK 9.2.8. A fully faithful functor is sometimes referred to as an embedding of cate-
gories, or sometimes a full embedding (and when so, a faithful but not necessarily full functor
might instead be referred to as an embedding).
E XAMPLES 9.2.9.
 9.2. FUNCTORS 281

a. The category Ab is a full subcategory of Gp.

b. The category Field is a full subcategory of Ring.
c. The above-described forgetful functors to sets are faithful but not full.
d. The category in which the objects are sets but the morphisms are bijections of sets is a
subcategory of Set that has the same objects but is not full.
D EFINITION 9.2.10. A directed graph G is a collection consisting of
(1) a set VG of vertices of G and,
(2) for every v, w ∈ VG , a set EG (v, w) of edges from v to w in G .
T ERMINOLOGY 9.2.11. In category theory, we often refer to the vertices of a directed graph
as dots and the edges as arrows.
E XAMPLE 9.2.12. The following picture provides the data of a directed graph with 4 vertices
and edge sets with between 0 and 2 elements each:
· / · 

· / ·
D EFINITION 9.2.13. The category (freely) generated by a directed graph G is the category I
with Obj(I) = VG and, for v, w ∈ Obj(I), with HomI (v, w) equal to the set of all words en en−1 · · · e1
for some n ≥ 0 (with n = 0 providing the empty word) with ei ∈ EG (vi−1 , vi ) for vi ∈ VG for
1 ≤ i ≤ n, with v0 = v and vn = w, together with the composition given by concatenation of
words.
E XAMPLE 9.2.14. Consider the directed graph G given by

v1
e1
/ v2 e2
/ v3 .

The category I generated by G has three objects v1 , v2 , v3 and morphism sets

HomI (vi , vi ) = {idvi }, HomI (vi , vi+1 ) = {ei },
HomI (v1 , v3 ) = {e2 e1 }, and HomI (vi , v j ) = ∅ if j < i.
E XAMPLE 9.2.15. Consider the directed graph G given by
e1
"
v1 e v2 .
e2

Let I be the category generated G . For i, j ∈ {1, 2} the set HomI (vi , v j ) consists of the words
with alternating letters e1 and e2 that start with ei and end with e j (including the empty word if
i = j).
D EFINITION 9.2.16. A diagram in C is a functor from a category generated by a graph to C .
282 9. CATEGORY THEORY

R EMARK 9.2.17. Let G be a directed graph, let I be the category generated by G , and let
C be a category. Given a map F : VG → C and functions F : EG (v, w) → HomC (F(v), F(w))
for each v, w, ∈ VG , there exists a unique functor F : I → C that agrees with F on VG and on
EG (v, w) ⊆ HomI (v, w) for every v, w ∈ VG .
R EMARK 9.2.18. Often, we consider finite graphs, in which every collection of vertices and
edges is finite. The resulting diagrams are known as finite diagrams.
D EFINITION 9.2.19. A commutative diagram in C is a diagram F : I → C , where I is the
category generated by a graph, which is a constant function on every set of morphisms.
E XAMPLE 9.2.20. To give a functor from I as in Example 9.2.14 to a category C is to pro-
scribe three objects A, B,C in C and two morphisms f : A → B and g : B → C. Thus, such a
diagram may be represented by
f g
A−
→B→
− C,
and it is automatically commutative.
E XAMPLE 9.2.21. To give a functor from I as in Example 9.2.15 to a category C is to pro-
scribe two objects A, B in C and two morphisms f : A → B and g : B → A. The diagram
f
#
Aa B
g

is commutative if and only if f ◦ g = idB and g ◦ f = idA .

9.3. Natural transformations
D EFINITION 9.3.1. Let F, G : C → D be two (covariant) functors. A natural transforma-
tion η : F G is a class of morphisms ηA : F(A) → G(A) for each A ∈ Obj(C ) subject to the
condition that
F(A)
ηA
/ G(A)
F( f ) G( f )
 
F(B)
ηB
/ G(B)
commutes for every f : A → B and A, B ∈ Obj(C ). If instead F and G are contravariant functors,
then the direction of the vertical arrows in the diagram are reversed.
E XAMPLE 9.3.2. Consider the functor T : Ab → Ab that sends an abelian group A to its
torsion subgroup Ator (i.e., the subgroup of elements of finite order) and takes a homomorphism
f : A → B to its restriction T ( f ) : Ator → Btor . Let I : Ab → Ab denote the identity functor. For
each abelian group A, we can define ιA : Ator → A to be the inclusion map. We clearly have
f ◦ ιA = ιB ◦ T ( f ) for all f : A → B, so ι : T I is a natural transformation.
E XAMPLE 9.3.3. If we think of groups G and G0 as monoids, so that functors G → G0 are
homomorphisms, then a natural transformation f f 0 between two homomorphisms f , f 0 : G →
G0 is given simply by an element x ∈ G0 such that f 0 (g) = x f (g)x−1 for all g ∈ G.
 9.3. NATURAL TRANSFORMATIONS 283

D EFINITION 9.3.4. Let F, G : C → D be functors. A natural transformation η : F G is

said to be a natural isomorphism if each ηA for A ∈ C is an isomorphism.
R EMARK 9.3.5. Every natural isomorphism η : F G has an inverse η −1 : G F with
ηA−1= (ηA )−1 for A ∈ Obj(C ).
D EFINITION 9.3.6. Let F, G : C → D be functors, and let ηA : F(A) → G(A) be morphisms
for each A ∈ Obj(C ). We say that these morphisms are natural if the ηA form a natural transfor-
mation η : F → G.
D EFINITION 9.3.7. Two categories C and D are said to be equivalent if there exist functors
F : C → D and G : D → C and natural isomorphisms η : G ◦ F idC and η 0 : F ◦ G idD .
Two such functors F and G are said to be quasi-inverse, and F and G are said to be equivalences
of categories.
E XAMPLE 9.3.8. A category C with one object 0 and one morphism is equivalent to the
category D with two objects 1, 2 and four morphisms, the identity morphisms of 1 and 2 and
isomorphisms 1 → 2 and 2 → 1. We have quasi-inverse functors F and G with F(0) = 1 and
F(id0 ) = id1 and G(1) = G(2) = 0 and G( f ) = id0 for all f . To see naturality, note that every
morphism between two objects in either category is unique.
The following theorem provides a standard example of equivalence of categories.
T HEOREM 9.3.9 (Morita equivalence). The category of left modules over a ring R is equiva-
lent to the category of left modules over Mn (R) for every n ≥ 1.
P ROOF. Let A be the R-Mn (R)-bimodule of row vectors of length n with R-entries. Let B be
the Mn (R)-R-bimodule of column vectors of length n with R-entries. Define
F : R-mod → Mn (R)-mod, F(M) = B ⊗R M, F( f ) = idB ⊗ f
for left R-modules M and M 0 and f ∈ HomR (M, M 0 ). Also, define
G : Mn (R)-mod → R-mod, G(N) = A ⊗Mn (R) N, G(g) = idA ⊗g
for left Mn (R)-modules N and N 0 and g ∈ HomMn (R) (N, N 0 ). Since multiplication induces iso-
morphisms
A⊗ B −∼
Mn (R) → R and B ⊗ A −∼ R → M (R),
n

both G ◦ F and F ◦ G are naturally isomorphic to identity functors. 

D EFINITION 9.3.10. Given two categories C and D with C small, the functor category
Func(C , D) has objects the functors C → D and morphisms the natural transformations between
functors, defining composition of natural transformations via composition of the morphisms de-
termining them.
D EFINITION 9.3.11. Let C be a category and A ∈ Obj(C ) be an object.
a. We have a functor hA : C → Sets given by
hA (B) = HomC (A, B)
284 9. CATEGORY THEORY

and, for g : B → C,
hA (g)( f ) = g ◦ f
for all f : A → B.
b. We have a contravariant functor hA : C → Set with
hA (B) = HomC (B, A) and hA (g)( f ) = f ◦ g
for B,C ∈ Obj(C ), g : B → C, and f : C → A.
D EFINITION 9.3.12. Let C be a small category. The Yoneda embedding is the functor
hC : C → Func(C op , Set)
defined by hC (A) = hA for A ∈ Obj(C ) and hC ( f ) : hA hB for f : A → B in C given by
hC ( f )C (g) = f ◦ g
for each g : C → A in C and any C ∈ Obj(C ).
R EMARK 9.3.13. The reader should check that the Yoneda embedding is a well-defined func-
tor.
T HEOREM 9.3.14. Let C be a small category. The Yoneda embedding hC is fully faithful.
P ROOF. We first show faithfulness. Let f , g : A → B be two morphisms with hC ( f ) = hC (g).
Then
f = f ◦ idA = hC ( f )A (idA ) = hC (g)A (idA ) = g ◦ idA = g.
As for fullness, suppose that η : hA hB for some A, B ∈ Obj(C ). We claim that η = h(e),
where e = ηA (idA ). To see this, note that the fact that η is a natural transformation means, in
particular, that the diagram
hA (A)
ηA
/ hB (A)

hA ( f ) hB ( f )
 
hA (C)
ηC
/ hB (C)
commutes for any f : C → A. Applying both compositions to the identity morphism of A, we get
the two equal terms
hB ( f ) ◦ ηA (idA ) = hB ( f )(e) = f ◦ e = h(e)C ( f )
and
ηC ◦ hA ( f )(idA ) = ηC (idA ◦ f ) = ηC ( f ),
and therefore, the desired equality. 
R EMARK 9.3.15. Similarly, we have a fully faithful contravariant functor
hC : C → Func(C , Set)
given by the hA for A ∈ Obj(C ) and natural transformations between them. This is just the
Yoneda embedding for the category C op .
 9.4. LIMITS AND COLIMITS 285

Theorem 9.3.14 can be thought of as a more general version of the following standard theo-
rem of group theory.
C OROLLARY 9.3.16 (Cayley’s theorem). Every group G is isomorphic to a subgroup of the
symmetric group SG on G.
P ROOF. Consider the monoid G formed by G. Recall that in G, morphisms are elements of
G. As h : G → Func(Gop , Set) is a functor, Yoneda’s lemma provides an injective function
h : G → HomFunc(Gop ,Set) (hG , hG )
on morphisms with the properties that h(e) = idhG and h(xy) = h(x) ◦ h(y) for x, y ∈ G. Since G
has only the object G, and hG (G) = G, this induces a one-to-one function ρ : G → Maps(G, G)
with ρ(x) = h(x)G and satisfying ρ(xy) = ρ(x) ◦ ρ(y) and ρ(e) = idG . In particular, we have
ρ(x−1 ) ◦ ρ(x) = idG for every x ∈ G, so its image lands in SG , and the resulting map G → SG is
an injective homomorphism. 
We shall later require the following strengthening of Theorem 9.3.14.
T HEOREM 9.3.17 (Yoneda’s lemma). For any object A of a small category C and contravari-
ant functor F : C → Set, there is a bijection
∼
HomFunc(C op ,Set) (hA , F) −
→ F(A)
given by η 7→ ηA (idA ) that is natural in A and F.
P ROOF. Let B ∈ Obj(C ). Given x ∈ F(A), consider the composition
F x ev
HomC (B, A) −
→ HomSet (F(A), F(B)) −−→ F(A),
where evx is evaluation at x. This defines a natural transformation ξ x : hA F. If η : hA F
and f : B → A, then
F( f ) ◦ ηA (idA ) = ηB (idA ◦ f ) = ηB ( f )
by the naturality of η. On the other hand, if x ∈ F(A), then
ξAx (idA ) = evx (F(idA )) = evx (idF(A) ) = x.
Hence the maps η 7→ ηA (idA ) and x 7→ ξ x are inverse to each other. 

9.4. Limits and colimits

In this section, C denotes a category, and I denotes a small category.
N OTATION 9.4.1. We write i ∈ I to denote, more simply, that i is an object in I.
D EFINITION 9.4.2. Let F : I → C be a functor. When it exists, the limit of F is a pair
(lim F, (φi )i∈I ) consisting of an object lim F in C and morphisms
φi : lim F → F(i)
for each i ∈ I such that φ j = F(κ) ◦ φi for all morphisms κ : i → j in I and with the universal
property that if X is any object of C together with morphisms ψi : X → F(i) for which ψ j =
286 9. CATEGORY THEORY

κ ◦ ψi for all morphisms κ : i → j, then there exists a unique morphism f : X → lim F such that
ψi = φi ◦ f for all i ∈ I.
N OTATION 9.4.3. We usually use lim F to refer more simply to a pair (lim F, (φi )i∈I ) that is
a limit of F : I → C , with the maps understood.
R EMARK 9.4.4. The universal property of the limit of a functor F as in Definition 9.4.2 may
be visualized by commutative diagrams
X
f
ψi  ψj
lim F
φi φj
  F(κ) 
F(i) / F( j).

L EMMA 9.4.5. If (X, (ψi )i∈I ) and (lim F, (φi )i∈I ) are limits of a functor F : I → C , then there
is a unique isomorphism f : X → lim F such that ψi = φi ◦ f for all i ∈ I.
P ROOF. There are morphisms f : X → lim F and g : lim F → X that are unique with the
respective properties that ψi = φi ◦ f and φi = ψi ◦ g for all i ∈ I. Note that we have φi = φi ◦ f ◦ g
for all i ∈ I. On the other hand, the universal property of X implies that the identity idX is the
unique morphism h such that ψi ◦ h = ψi for all i ∈ I, so f ◦ g = idX . Similarly, g ◦ f is the identity
of lim F by its universal property. Therefore, the unique map f is an isomorphism. 
R EMARK 9.4.6. Lemma 9.4.5 says that a limit, when it exists, is unique up to unique isomor-
phism (respecting the universal property) and for that reason, we refer to “the”, rather than “a”,
limit.
If I has only identity morphisms, then the limit of a functor F : I → C is determined entirely
by the image objects Ai = F(i) for all i ∈ I. Hence the notation in the following definition makes
sense.
D EFINITION 9.4.7. Let I be a category with only identity morphisms, and let F : I → C be a
functor. Set Ai = F(i) for each i ∈ I.
a. The limit ∏i∈I Ai of F, when it exists, is called the product of the Ai .
b. The maps
pi : ∏ Ai → Ai
i∈I
resulting from the universal property of the product are known as projection maps.
E XAMPLES 9.4.8. The product coincides with direct product in the categories Set, Gp, Top,
Ring and R-mod. Products of more than one object do not exist in the category Field.
R EMARK 9.4.9. A commutative diagram in a category C arises from a functor F : I → C ,
where I is a category generated by a directed graph. Therefore, we may speak of the limit of the
diagram.
 9.4. LIMITS AND COLIMITS 287

D EFINITION 9.4.10. The limit A1 ×B A2 of a diagram

(9.4.1) A1
f1
f2 
A2 /B

in C , when it exists, is called the pullback of the diagram.

R EMARK 9.4.11. The pullback of (9.4.1) is endowed with morphisms p1 and p2 that make

A1 ×B A2
p1
/ A1
p2 f1
 f2 
A2 /B
commute.
E XAMPLE 9.4.12. In Set, Gp, Top, and R-mod, the pullback is the subobject (i.e., subset,
subgroup, subspace, or submodule) with underlying set
{(a1 , a2 ) ∈ A1 × A2 | f1 (a1 ) = f2 (a2 )}.
We also have the dual notion to limits:
D EFINITION 9.4.13. Let F : I → C be a functor. When it exists, the colimit of F is a pair
(colim F, (αi )i∈I ) consisting of an object colim F ∈ C together with morphisms
αi : F(i) → colim F
for each i ∈ I such that α j ◦ F(κ) = αi for all morphisms κ : i → j and with the universal property
that if X is any object of C together with morphisms βi : X → F(i) for which β j ◦ κ = βi for all
morphisms κ : i → j, then there exists a unique morphism f : colim F → X such that βi = f ◦ αi
for all i ∈ I.
N OTATION 9.4.14. A colimit of a functor F : I → C is usually denoted simply by the object
colim F, with the morphisms omitted.
R EMARK 9.4.15. The properties of the colimit expressed in Definition 9.4.13 may be sum-
marized by the commutativity of the diagrams
F(κ)
F(i) / F( j)

αi ! } αj
colim F
βi βj
f
#  {
X
for all κ : i → j in I.
288 9. CATEGORY THEORY

We have the obvious analogue of Lemma 9.4.5, which again tells us that we may speak of
“the” colimit.
L EMMA 9.4.16. If (X, (βi )i∈I ) and (colim F, (αi )i∈I ) are colimits of a functor F : I → C , then
there is a unique isomorphism f : colim F → X such that αi = f ◦ βi for all i ∈ I.
R EMARK 9.4.17. When it exists, the colimit of F : I → C in C satisfies
colim F = op (lim(op ◦ F)) ,
so its underlying object is an limit in C op .
D EFINITION 9.4.18. The colimit of a functor F : I → C from a category I with only identity
morphisms is called a coproduct, and it is denoted qi∈I F(i).
E XAMPLES 9.4.19.
a. The coproduct in Set and Top of two objects X1 and X2 is the disjoint union X1 q X2 .
b. The coproduct in Gp of two groups G1 and G2 is the free product G1 ∗ G2 .
c. The coproduct in R-mod (and in particular Ab) of two R-modules A1 and A2 is the direct
sum A1 ⊕ A2 .
d. The coproduct in the category of commutative rings R1 and R2 is the tensor product R1 ⊗
R2 .
R EMARK 9.4.20. Examples 9.4.19(a-d) generalize directly to arbitrary collections of objects.
R EMARK 9.4.21. Much as with limits, we may speak of a colimit of a diagram in a category.
D EFINITION 9.4.22. The colimit of a diagram

(9.4.2) B
g1
/ A1
g2

A2 ,
in C is called the pushout A1 qB A2 .
R EMARK 9.4.23. The pushout of the diagram (9.4.2) fits into a diagram

B
g1
/ A1
g2 ι1
 
A2
ι2
/ A1 qB A2 ,

where ι1 and ι2 are induced by the universal property of the colimit.

E XAMPLE 9.4.24. In Set and Top, the pushout is the quotient (set or topological space) of
the disjoint union of A1 and A2 under the equivalence relation identifying g1 (b) with g2 (b) for
all b ∈ B.
 9.4. LIMITS AND COLIMITS 289

D EFINITION 9.4.25. We say that a category C admits the limit (resp., colimit) of a functor
F : I → C if the limit (resp., colimit) exists in C .
R EMARK 9.4.26. More generally, we may speak of C admitting the limits (or colimits) of
any collection of functors from small categories to C .
D EFINITION 9.4.27. A category is called complete if it admits all limits.
E XAMPLE 9.4.28. The category of finite sets is not complete.
D EFINITION 9.4.29. A category is called cocomplete if it admits all colimits.
R EMARK 9.4.30. To say that C is complete is to say that C op is cocomplete.
P ROPOSITION 9.4.31. The category Set is both complete and cocomplete.
P ROOF. Let F : I → Set be a functor. We merely describe the limit and colimit of F and
leave the rest to the reader. The limit is
( )
lim F = (ai )i ∈ ∏ F(i) | F(φ )(ai ) = a j if φ : i → j in I ,
i∈I
and the colimit is
colim F = qi∈I F(i)/ ∼
where ∼ is the minimal equivalence relation satisfying ai ∼ a j for ai ∈ F(i) and a j ∈ F( j) if
there exists φ : i → j with F(φ )(ai ) = a j . 
R EMARK 9.4.32. In fact, the categories Set, Top, Gp, Ab, Ring, and R-mod admit all limits
and colimits.
The reader will easily verify the following.
P ROPOSITION 9.4.33. Let I be a small category and C a (co)complete category. Then the
category Func(I, C ) is (co)complete.
C OROLLARY 9.4.34. Let I be a small category and C be (co)complete. Let F : I → C be a
functor, and supposing that C is small, consider the Yoneda embedding hC : C → Func(C op , Set).
Then hC ◦ F has a (co)limit in Func(C op , Set).
D EFINITION 9.4.35. A directed set I is a set I together with a partial ordering ≤ on I such
that for any i, j ∈ I, there exists k ∈ I with i ≤ k and j ≤ k.
D EFINITION 9.4.36.
a. The limit of a diagram
· · · → A3 → A2 → A1
in a category C is referred to as the sequential limit of the objects Ai .
b. The colimit of a diagram
A1 → A2 → A3 → · · ·
in a category C is referred to as the sequential colimit of the objects Ai .
290 9. CATEGORY THEORY

E XAMPLE 9.4.37. In Ab, the sequential limit of the groups Z/pn Z with respect to homomor-
phisms Z/pn+1 Z → Z/pn Z given by reduction modulo pn is the group Z p of p-adic integers. The
sequential colimit of these same groups with respect to the maps Z/pn Z → Z/pn+1 Z induced by
multiplication modulo p is the group Q p /Z p , equal to the p-power torsion in Q/Z.
The sequential limit (resp., sequential colimit) is just a special case of the notion of an inverse
limit (resp., direct limit), which is a more usual terminology.
D EFINITION 9.4.38.
a. A directed category I is a category with a nonempty directed set I of objects and at most
one morphism i → j for any i, j ∈ I, which exists if and only if i ≤ j.
b. A codirected category is a category I such that I op is directed.
D EFINITION 9.4.39. Let I be a codirected category. The limit of a functor F : I → C is
referred to the inverse limit of the objects F(i) over the inverse system of objects F(i) for i ∈ I
and morphisms F(κ) for κ : i → j in I, and it is denoted limi∈I F(i).
←−
D EFINITION 9.4.40. Let I be a directed category. The colimit of a functor F : I → C is the
direct limit of the objects F(i) over the directed system of objects F(i) for i ∈ I and morphisms
F(κ) for κ : i → j in I and is denoted limi∈I F(i) (or sometimes just lim F).
−→ −→
E XAMPLES 9.4.41.
a. The inverse limit of the (commutative) rings Z/pn Z with respect to homomorphisms
Z/pn+1 Z → Z/pn Z given by reduction modulo pn is the ring Z p known as the p-adic integers.
b. The direct limit of the abelian groups Z/pn Z with respect to the multiplication-by-p maps
Z/pn Z → Z/pn+1 Z is equal to the subgroup of elements of p-power order (under addition) in
Q/Z.
c. The absolute group Gal(Q/Q) of all automorphisms of the field of algebraic numbers Q is
isomorphic to the inverse limit of the collection of all Gal(K/Q) for K a finite, normal extension
of Q inside Q, with respect to the maps Gal(L/Q) → Gal(K/Q) given by restriction with K ⊆ L.
Note that the set of field extensions is directed as the compositum of two normal extensions is
normal, and the resulting category is codirected as we use only the restriction morphisms.
For certain diagrams, limits and colimits can be quite boring, especially when the diagram
contains an initial object in the case of limits, or terminal object in the case of colimits.
D EFINITION 9.4.42.
a. An initial object A in C is an object such that for each B ∈ Obj(C ), there is a unique
morphism A → B in C .
b. A terminal object X in C is an object such that for each B ∈ Obj(C ), there is a unique
morphism B → X in C .
c. An zero object 0 in C is an object that is both initial and terminal.
R EMARK 9.4.43. Terminal and initial objects are unique up to unique isomorphism when
defined.
 9.5. ADJOINT FUNCTORS 291

We provide some examples.

E XAMPLES 9.4.44.
a. The empty set ∅ is the initial object in the category Set, while any set with one element is
a terminal object.
b. The trivial group is a zero object in the category Gp.
c. The zero ring is a terminal object and Z is an initial object in Ring.
We omit the proof of the following easy lemma.
L EMMA 9.4.45. Let I be a small category and F : I → C a functor.
a. Suppose that I has an initial object i. Then
lim F = F(i).
b. Suppose that I has a terminal object j. Then
colim F = F( j).

9.5. Adjoint functors

The following definition allows us to weaken the property of being quasi-inverse to one of
“adjointness”.
D EFINITION 9.5.1. We say that F : C → D is left adjoint to G : D → C if there exist bijec-
tions
φC,D : HomD (F(C), D) −∼ → HomC (C, G(D))
for each C ∈ Obj(C ) and D ∈ Obj(D) such that the φC,D form a natural transformation of functors
C op × D → Set.
We also say that G is right adjoint to F, and we say that F and G are adjoint functors.
R EMARK 9.5.2. To say that η is a natural transformation in Definition 9.5.1 is a fancier way
of saying that given morphisms f : C0 → C in C and g : D → D0 in D, we have a commutative
diagram
η(C,D)
HomD (F(C), D) / HomC (C, G(D))
t7→g◦t◦F( f ) u7→G(g)◦u◦ f
 η(C0 ,D0 ) 
HomD (F(C0 ), D0 ) / HomC (C0 , G(D0 )).

R EMARK 9.5.3. If F : C → G and G : D → C are quasi-inverse functors, then we have

bijections
φ : Hom (F(C), D) −∼
C,D D → Hom (C, G(D)), Cφ (g) = G(g) ◦ η , C,D C
for C ∈ Obj(C ) and D ∈ Obj(D), where g : F(C) → D in D and η is a natural isomorphism
G◦F idC . These form a natural transformation φ between functors C op × D → Set, so G is
right adjoint to F. Similarly, using η 0 : F ◦ G idD , we see that G is left adjoint to F.
292 9. CATEGORY THEORY

E XAMPLE 9.5.4. The forgetful functor Gp → Set is right adjoint to the functor Set → Gp that
takes a set X to the free group FX on X and a map f : X → Y of sets to the unique homomorphism
φ f : FX → FY with φ f |X = f . That is, the restriction map
Hom(FY , G) → Maps(Y, G)
is inverse to the map Maps(Y, G) → Hom(FY , G) that takes φ f : Y → G to F( f ), and these bijec-
tions are easily seen to be natural.
The following is perhaps the most standard example of adjointness: that of Hom and ⊗ in
categories of modules.
E XAMPLE 9.5.5. Let A and B be algebras over a commutative ring R. Fix an R-balanced A-
B-bimodule M. Define a functor tM : B-mod → A-mod by tM (N) = M ⊗B N on B-modules N and
tM (g) = idM ⊗g on B-module homomorphisms g : N → N 0 . Define another functor hM : A-mod →
B-mod by hM (L) = HomA (M, L) on A-modules L and hM ( f )(h) = f ◦ h for f ∈ HomA (L, L0 ) and
h ∈ HomA (M, L). Then the isomorphism of Theorem 8.4.29 is the adjunction map
Hom (t (N), L) −∼
A M → Hom (N, h (L)).
B M

These isomorphisms are natural in N and L, and hence tM is left adjoint to hM .

P ROPOSITION 9.5.6. Fix categories I and C , and suppose that all limits F : I → C exist. The
functor lim has a left adjoint ∆ given by taking A ∈ Obj(C ) to the constant functor cA , where
cA (i) = A for all i ∈ I, and taking g : A → B for A, B ∈ Obj(C ) to the natural transformation
cA cB given by g : cA (i) = A → cB (i) = B for all i ∈ I.
P ROOF. We must describe natural isomorphisms
Hom (cA , F) ∼
Func(I,C ) = HomC (A, lim F)
for A ∈ Obj(C ) and F : I → C . I.e., given a natural tranformation η : cA F, we must associate
a map f : A → lim F, and conversely. Such a natural transformation η consists of maps
ηi : cA (i) = A → F(i)
that are compatible in the sense that η j = F(κ) ◦ ηi for all κ : i → j. Thus, the existence of a
unique f is simply the universal property of the limit. On the other hand, if we have f , then we
have maps
φi ◦ f : A → F(i),
where φi is the map lim F → F(i) arising in the definition of the limit. These maps then define
the universal transformation η. 
We now see exactly how adjointness weakens inverseness.
D EFINITION 9.5.7. Two categories C and D are said to be equivalent if there exist functors
F : C → D and G : D → C and natural isomorphisms η : G ◦ F idC and η 0 : F ◦ G idD .
Two such functors F and G are said to be quasi-inverse, and F and G are said to be equivalences
of categories.
 9.5. ADJOINT FUNCTORS 293

E XAMPLE 9.5.8. A category C with one object 0 and one morphism is equivalent to the
category D with two objects 1, 2 and four morphisms, the identity morphisms of 1 and 2 and
isomorphisms 1 → 2 and 2 → 1. We have quasi-inverse functors F and G with F(0) = 1 and
F(id0 ) = id1 and G(1) = G(2) = 0 and G( f ) = id0 for all f . To see naturality, note that every
morphism between two objects in either category is unique.
N OTATION 9.5.9. Let η : F F 0 be a natural transformation between functors F, F 0 : C →
D.
a. If G : D → E is a functor, then we define a natural transformation G(η) : G ◦ F G ◦ F0
by
G(η)C = G(ηC ) : G(F(C)) → G(F 0 (C))
for all objects C of C .
b. If H : B → C is a functor, then we define a natural transformation η(H) : F ◦ H F0 ◦ H
by
η(H)B = ηH(B) : F(H(B)) → F 0 (H(B))
for all objects B of C .
D EFINITION 9.5.10. Let F : C → D and G : D → C be functors.
a. A unit for the pair (F, G) is a natural transformation idC G ◦ F.
b. A counit for the pair (F, G) is a natural transformation F ◦ G idD .
c. A unit-counit adjunction is a pair (F, G), a unit η for (F, G), and a counit η 0 for (F, G)
satisfying
idF = η 0 (F) ◦ F(η) : F F
as morphisms in Func(C , D) and
idG = G(η 0 ) ◦ η(G) : G G
as morphisms in Func(D, C ).
P ROPOSITION 9.5.11. A functor F : C → D is left adjoint to a functor G : D → C if and
only if there exists a unit-counit adjunction for the pair (F, G).
P ROOF. Suppose that F is left adjoint to G. We define η : idC G ◦ F as follows. For
C ∈ Obj(C ), we have bijections
∼
HomD (F(C), F(C)) −
→ HomC (C, G ◦ F(C))
by adjointness, and we define ηC to be the image of idF(C) . For D ∈ Obj(D), we also have
∼
HomD (F ◦ G(D), D) −
→ HomC (G(D), G(D))
and define η 0 : F ◦ G idD by taking ηD0 to be the image of idG(D) under the inverse of this map.
We leave it to the reader to check that these are natural and form a unit-counit adjunction. The
converse is left to the reader as well. 
294 9. CATEGORY THEORY

9.6. Representable functors

D EFINITION 9.6.1. Let F : C → Set be a contravariant functor. Then F is said to be repre-
sentable if there exists a natural isomorphism hB F for some B ∈ Obj(C ). (In other words, we
have natural bijections
∼
HomC (A, B) −→ F(A)
for all objects A of C .) We then say that B represents F.
Using Yoneda’s lemma and assuming C to be small, we can reword Definition 9.6.1 as saying
that there exists B ∈ Obj(C ) such that there are compatible bijections between the set of natural
transformations hA F and the set of morphisms A → B for each A ∈ Obj(C ).
E XAMPLE 9.6.2. Consider the contravariant functor P : Set → Set which takes a set S to its
power set P(S), the set of all subsets of S and a map f : S → T to the map P( f ) : P(T ) → P(S)
by mapping U ⊂ T to f −1 (U). Then P is represented by the set {0, 1} via the isomorphism
∼
Maps(S, {0, 1}) −
→ P(S)
by φ 7→ φ −1 ({1}). These isomorphisms form a natural transformation:
Maps(T, {0, 1})
∼ / P(T )

h{0,1} ( f ) P( f )
 
Maps(S, {0, 1})
∼ / P(S)

for f : S → T . Here, the lefthand vertical map takes φ to φ ◦ f and the righthand vertical map
takes a subset X of T to f −1 (X). We check that
(φ ◦ f )−1 ({1}) = f −1 (φ −1 ({1})).
The following is a corollary of Yoneda’s lemma.
L EMMA 9.6.3. A representable functor is represented by a unique object up to isomorphism.
If B and C represent a contravariant functor F : C → Set, then such an isomorphism f : B → C
is unique making the diagrams
HomC (A, B)
∼ / F(A)

hA ( f )

HomC (A,C)
∼ / F(A)

commute for all A ∈ Obj(C ).

P ROOF. Let F : C → Set be a representable (contravariant) functor represented by B ∈ Obj(C )
and C ∈ Obj(C ). Then we have natural isomorphisms ξ : hB F and ξ 0 : hC F. The compo-
0 −1 B C C
sition ξ ◦ ξ : h → h is equal to h ( f ) for a unique f : B → C by the weak form of Yoneda’s
lemma. 
T HEOREM 9.6.4. Let F : C → D be a functor between small categories.
 9.6. REPRESENTABLE FUNCTORS 295

a. The functor F has a right adjoint if and only if the functor hD ◦ F is representable for each
D ∈ Obj(D). If G is right adjoint to F, then hD ◦ F is representable by G(D).
b. If F has a right adjoints G and G0 , then there exists a unique natural isomorphism ξ : G
G0 such that diagrams
η(C,D)
HomD (F(C), D) / HomC (C, G(D))
f 7→ξD ◦ f
0
η(C,D) 
HomD (F(C), D) / HomC (C, G0 (D))

commute for all C ∈ Obj(C ) and D ∈ Obj(D), where the horizontal morphisms are the adjunction
isomorphisms.
P ROOF. Assume that F has a right adjoint G, and consider the adjunction morphisms
∼
η(C,D) : HomD (F(C), D) −
→ HomC (C, G(D)).
In other words,
hD ◦ F(C) ∼ = hG(D) (C),
so G(D) represents hD ◦ F. In this case, the uniqueness in part b is an immediate consequence of
Lemma 9.6.3.
Now suppose that hD ◦ F is representable for each D by some object G(D) (chosen using
the axiom of choice). Then there exist isomorphisms η(C,D) that are natural in C. We must also
0
define G on morphisms f : D → D0 in D. Such an f induces a natural transformation hD hD
which provides morphisms
0
hD ◦ F(C) → hD ◦ F(C)
0
for all C ∈ Obj(C ) and thus induces hG(D) (C) → hG(D ) (C), and these are natural in C. Thus, we
0
have a natural transformation hG(D) hG(D ) . Since the Yoneda embedding is fully faithful, we
have a unique morphism G(D) → G(D0 ) inducing this natural transformation, which we define
to be G( f ). We leave to the reader the check that G as defined is a functor. 
D EFINITION 9.6.5. Let F : C → Set be a covariant functor. We say that F is representable
if there exists a natural isomorphism hA F for some A ∈ Obj(C ). (That is, there are natural
isomorphisms
∼
F(B) − → HomC (A, B)
in B ∈ Obj(C ).) In this case, we say that A represents F.
R EMARK 9.6.6. A covariant functor F : C → Set is representable if and only if the con-
travariant functor F ◦ op : C op → Set is representable. The same object of C will represent both
objects.
E XAMPLE 9.6.7. Let F : Gp → Set be the forgetful functor. Then F can be represented by
Z. To see this, we define the set map
G → HomGp (Z, G)
by a 7→ (1 7→ a) for a ∈ G. Naturality is clear.
296 9. CATEGORY THEORY

E XAMPLE 9.6.8. Let F : Gp → Set be the functor which sends a group to its subset G[n] of
elements of order dividing n. Then F can be represented by Z/nZ.
E XAMPLE 9.6.9. Consider a functor F : I → C between small categories. To say that the
contravariant functor lim hC ◦ F : C → Set is representable is exactly to say that there exists an
object X in C such that one has natural isomorphisms
∼ ∼
→ lim(hC ◦ F)(A) −
HomC (A, X) − → lim(HomC (A, F(·)))
for A ∈ C . In other words, lim hC ◦ F is representable if and only if lim F exists in C .
E XAMPLE 9.6.10. Consider a functor F : I → C between small categories. View hC as a co-
variant functor C → Hom(C , Set). To say that the functor lim(hC ◦F) : C → Set is representable
is exactly to say that there exists an object X ∈ C such that one has natural isomorphisms
∼ ∼
HomC (X, A) −
→ lim(hC ◦ F)(A) −
→ lim(HomC (F(·), A))
for A ∈ C . In other words, lim hC ◦ F is representable if and only if colim F exists in C .
9.7. Equalizers and images
D EFINITION 9.7.1. Let C be a category, and let
f
(9.7.1) A // B.
g

be a diagram in C .
a. The limit eq( f , g) of the diagram (9.7.1), when it exists, is called its equalizer.
b. The colimit coeq( f , g) of (9.7.1) is called its coequalizer.
We have a commutative diagram:
f
/A / / coeq( f , g).
eq( f , g) /B
g

E XAMPLES 9.7.2.
a. Let X,Y be sets, and consider maps f , g : X → Y . In Set, we have
eq( f , g) = {x ∈ X | f (x) = g(x)}
and coeq( f , g) is the quotient of Y by the minimal equivalence relation ∼ generated by f (x) ∼
g(x) for all x ∈ X.
b. In R-mod, the equalizer is expressed as in Set. For an R-module homomorphism f : A →
B, we have
coeq( f , g) = B/{( f − g)(a) | a ∈ A}.
L EMMA 9.7.3. Let f , g : A → B be morphisms in a category C .
a. Suppose that eq( f , g) exists. Then the induced map h : eq( f , g) → A is a monomorphism.
b. Suppose that coeq( f , g) exists. Then the induced map k : B → coeq( f , g) is an epimor-
phism.
 9.7. EQUALIZERS AND IMAGES 297

P ROOF. Suppose that α, β : C → eq( f , g) are morphisms in C such that h ◦ α = h ◦ β . Let

h0 = h ◦ α, and note that f ◦ h0 = g ◦ h0 . But then α : C → eq( f , g) is unique such that h0 = h ◦ α
by the universal property of eq( f , g). Since h0 = h ◦ β as well, we have α = β . Part b follows
from part a by working in the opposite category. 
T HEOREM 9.7.4. A category that admits all products and equalizers is complete, and a cat-
egory that admits all coproducts and coequalizers is cocomplete.
P ROOF. For the second statement, by taking the opposite category, we are reduced to the
first statement. Let C be a category that admits all products and equalizers. Let F : I → C be a
functor from a small category I, and consider the equalizer eq( f , g) of the two morphisms
f,g: ∏ F(i) → ∏ F(φ (i))
i∈I φ : i→φ (i)

defined via the universal property of the second product as the unique morphisms satisfying
pφ ◦ f = pφ (i) : ∏ F( j) → F(φ (i))
j∈I
and
pφ ◦ g = F(φ ) ◦ pi : ∏ F( j) → F(φ (i))
j∈I
for morphisms φ : i → φ (i) in C , where pφ denotes projection onto the φ -coordinate in the
second product and pi denotes projection to the i-coordinate in the first.
Let ι : eq( f , g) → ∏i∈I F(i) be the morphism defining the equalizer. We claim that the equal-
izer eq( f , g), together with the maps pi ◦ ι : eq( f , g) → F(i) for i ∈ I, satisfies the univeral prop-
erty of lim F. By definition, for any morphism φ in C as above, we have
F(φ ) ◦ (pi ◦ ι) = pφ ◦ g ◦ ι = pφ ◦ f ◦ ι = pφ (i) ◦ ι.
Moreover, given X ∈ C and morphisms ψi : X → F(i) for i ∈ I such that F(φ ) ◦ ψi = φφ (i) for all
i ∈ I, we have a product map ψ : X → ∏i∈I F(i) such that
pφ ◦ f ◦ ψ = pφ (i) ◦ ψ = ψφ (i)
and
pφ ◦ g ◦ ψ = F(φ ) ◦ pi ◦ ψ = F(φ ) ◦ ψi = ψφ (i) ,
so there exists a unique morphism θ : X → eq( f , g) with pi ◦ ι ◦ θ = ψi for all i ∈ I. That is,
eq( f , g) satisfies the universal property of the limit. 
D EFINITION 9.7.5. In a category C with a zero object 0, the zero morphism 0 : A → B be-
tween objects A, B ∈ Obj(C ) is the composition of A → 0 → B of morphisms proscribed by the
fact that 0 is both initial and terminal.
D EFINITION 9.7.6. Let C be a category with a zero object. Let f : A → B be a morphism in
C , and let 0 : A → B be the zero morphism.
a. The kernel ker f of f is the equalizer of f and 0.
b. The cokernel coker f of f is the coequalizer of f and 0.
298 9. CATEGORY THEORY

E XAMPLES 9.7.7.
a. In R-mod, the categorical notions of kernel and cokernel agree with the usual ones.
b. In Gp, kernel is the usual notion, and the cokernel of a group homomorphism f : G → H
with the quotient of H by the normal closure of f (G).
There are different notions of image and coimages in categories. We use the following.
D EFINITION 9.7.8. Let f : A → B be a morphism in a category C .
a. The image of f is an object im f of C together with a monomorphism ι : im f → B such
that there exists a morphism π : A → im f such that π ◦ ι = f and such that if g : C → B us
a monomorphism and e : A → C is a morphism such that e ◦ g = f , then there exists a unique
morphism ψ : im f → C such that g = ι ◦ ψ.
b. The coimage of f is an object coim f together with an epimorphism π : A → coim f such
that there exists a morphism ι : coim f → B with ι ◦ π = f and such that if e : A → C is an
epimorphism and g : C → B is a morphism such that f = e ◦ g, the there exists a unique morphism
θ : coim f → C such that e = θ ◦ f .
R EMARK 9.7.9. In the definition of the image of f : A → B, then π is uniquely determined as
ι is a monomorphism and f = ι ◦ π. Moreover, if g ◦ ψ = ι as stated, then g ◦ ψ ◦ π = ι ◦ π = g ◦ e,
and g is a monomorphism, so ψ ◦ π = e. That is the diagram
f
A /
= BJ
π ! ι

e
im f g
ψ
 
C
commutes. Finally, note that ψ is forced to be a monomorphism since ι is. The analogous
statements hold for the coimage.
P ROPOSITION 9.7.10. Let f : A → B be a morphism in a category C .
a. If C admits equalizers, then the canonical morphism π : A → im f is an epimorphism.
b. If C admits coequalizers, then the canonical morphism ι : coim f → B is a monomor-
phism.
P ROOF. We prove only part a, as part b is proven similarly. Suppose that α, β : im f →
D satisfy α ◦ π = β ◦ π. Then by definition of the equalizer, there exists a unique morphism
ρ : A → eq(α, β ) such that for the canonical monomorphism c : eq(α, β ) → im f , we have π =
c ◦ ρ. On the other hand, consider the composite monomorphism g = ι ◦ c : eq(α, β ) → B, where
ι : im f → B is the morphism of the image. Note that g ◦ ρ = f , so by definition of the image,
there exists a unique morphism d : im f → eq(α, β ) such that g ◦ d = ι. Then ι ◦ (c ◦ d) = ι, so
c ◦ d = id. On the other hand,
g ◦ (d ◦ c) = ι ◦ c ◦ d ◦ c = ι ◦ c = g.
 9.8. ADDITIVE AND ABELIAN CATEGORIES 299

As g is a monomorphism, we have d ◦ c = id as well. 

R EMARK 9.7.11. If C has finite products, finite coproducts, equalizers, and coequalizers, we
may also make the following definition of the image an coimage of a morphism f : A → B.
a. The image of f is the equalizer of the two morphisms ιi : B → B qA B.
b. The coimage of f is the coequalizer of the two projection morphisms pi : A ×B A → A.
This definition agrees with that already given if every morphism in C factors through an equalizer
morphism and C admits finite limits and colimits. We omit the nontrivial proof.
L EMMA 9.7.12. For any f : A → B in a category C that admits equalizers (or coequalizers)
and for which coim f and im f exist, there is a unique morphism u : coim f → im f such that the
composition
s u t
A→ − coim f → − im f →
− B
of induced morphisms is f .
P ROOF. We suppose that C admits equalizers. By Proposition 9.7.10, the canonical mor-
phism π : A → im f with t ◦ π = f is an epimorphism. By the definition of coim f , there then
exists a unique morphism u : coim f → im f such that π = u ◦ s. Then t ◦ u ◦ s = t ◦ π = f . If
v : coim f → im f also satisfies t ◦ v ◦ s = f , then t ◦ v ◦ s = t ◦ π, and t : im f → B is a monomor-
phism, so v ◦ s = π. Thus u = v by the uniqueness of u. 
D EFINITION 9.7.13. We say that a morphism f : A → B in a category that admits an image
and coimage of f is strict if the induced morphism coim f → im f is an isomorphism.
E XAMPLE 9.7.14. Every morphism in the category of R-modules is strict.

9.8. Additive and abelian categories

D EFINITION 9.8.1. An additive category C is a category with the following properties:
i. for A, B ∈ Obj(C ), the set of morphisms HomC (A, B) in C has an abelian group law (ad-
dition) with the property that for any diagram
f
/B
g1
// C h / D,
A
g2

in C , we have
h ◦ (g1 + g2 ) ◦ f = h ◦ g1 ◦ f + h ◦ g2 ◦ f ,
ii. C has a zero object 0,
iii. C admits finite coproducts.
In an additive category C , there always exists the zero morphism is the identity element in
the abelian group HomC (A, B).
E XAMPLES 9.8.2.
a. The categories Ab and R-mod are additive categories, with the usual addition of homo-
morphisms.
300 9. CATEGORY THEORY

b. The full subcategory R-mod of finitely generated R-modules is an additive category.

In an additive category, we denote the coproduct of two objects A1 and A2 by A1 ⊕ A2 .
L EMMA 9.8.3. Finite products exist in an additive category, and there are natural isomor-
phisms
A1 q A2 ∼ = A1 × A2
for A1 , A2 ∈ Obj(C ). The resulting inclusion morphisms ιi : Ai → A1 q A2 and projection mor-
phisms and pi : A1 q A2 → Ai obtained by viewing A1 q A2 as a product and coproduct, respec-
tively, satisfy pi ◦ ιi = idAi and pi ◦ ι j = 0 for i 6= j, while
ι1 ◦ p1 + ι2 ◦ p2 = idA1 qA2 .
P ROOF. We have morphisms ιi : Ai → A1 q A2 by definition. We also have maps pi : A1 q
A2 → Ai defined by (
idAi if i = j
pi ◦ ι j =
0 6 j
if i =
and the universal property of the coproduct. We then have
(ι1 ◦ p1 + ι2 ◦ p2 ) ◦ ιi = ιi ,
and hence ι1 ◦ p1 + ι2 ◦ p2 = idA1 qA2 , again by the universal property.
Given an object B ∈ Obj(C ) and morphisms gi : B → Ai , we then have a morphism
ψ = ι1 ◦ g1 + ι2 ◦ g2 : B → A1 q A2 ,
which is unique such that
pi ◦ ψ = gi .
Hence A1 q A2 satisfies the universal property of the product. 
D EFINITION 9.8.4. An object A in an additive category C together with objects Ai , inclusion
morphisms ιi : Ai → A, and projection morphisms pi : A → Ai for i ∈ {1, 2} for which pi ◦ ι j is
zero if i 6= j and idAi if i = j and for which ι1 ◦ p1 + ι2 ◦ p2 = idA is called a biproduct of the
objects A1 and A2 , and we write it as A1 ⊕ A2 .
The notion of a biproduct allows us to reinterpret addition in an additive category. First, note
the following definitions.
D EFINITION 9.8.5. Let A be an object in an additive category C .
a. The diagonal morphism ∆A : A → A ⊕ A in C is the unique morphism induced by two
copies of idA : A → A and the universal property of the product.
b. The codiagonal morphism ∇A : A ⊕ A → A in C is the unique morphisms induced by two
copies of idA : A → A and the universal property of the coproduct.
D EFINITION 9.8.6. Let C be an additive category, and let f1 : A1 → B1 and f2 : A2 → B2 be
morphisms in C . The biproduct, or direct sum, f1 ⊕ f2 of the maps f1 and f2 is the morphism
A1 ⊕ A2 → B1 ⊕ B2 induced as the (morphism defined by the universal property of the) coproduct
of the composite maps Ai → Bi → B1 ⊕ B2 , the latter morphisms being inclusions.
 9.8. ADDITIVE AND ABELIAN CATEGORIES 301

R EMARK 9.8.7. Equivalently, the direct sum of f1 and f2 as in Definition 9.8.6 is induced as
the product of the composite maps A1 ⊕ A2 → Ai → Bi , the initial morphisms being projections.
Of course, we could make these definitions in an arbitrary category using products and co-
products.
L EMMA 9.8.8. Let f , g : A → B be two morphisms in an additive category C . Then we have
f + g = ∇B ◦ ( f ⊕ g) ◦ ∆A .
P ROOF. Let ιiA and pAi respectively denote the inclusion maps and projection maps for the
biproduct A ⊕ A, and similarly for B. We have

∇B ◦ ( f ⊕ g) ◦ ∆A = ∇ ◦ ( f ⊕ g) ◦ (ι1A ◦ pA1 + ι2A ◦ pA2 ) ◦ ∆A

= ∇B ◦ ( f ⊕ g) ◦ ι1A ◦ pA1 ◦ ∆A + ∇B ◦ ( f ⊕ g) ◦ ι2A ◦ pA2 ◦ ∆A .
Taking the first term without loss of generality, we have
∇B ◦ (( f ⊕ g) ◦ ι1A ) ◦ (pA1 ◦ ∆A ) = ∇B ◦ (ι1B ◦ f ) ◦ idA = idB ◦ f = f .

D EFINITION 9.8.9. A functor F : C → D between additive categories is called additive if for
each A, B ∈ Obj(C ), the map
HomC (A, B) → HomD (F(A), F(B))
is a group homomorphism.
E XAMPLE 9.8.10. Let C be an additive category. Then for any A ∈ Obj(C ), the functors hA
and hA may be considered as functors to Ab, rather than Set. The resulting functors are additive.
L EMMA 9.8.11. A functor F : C → D of additive categories is additive if and only if F
preserves biproducts, which is to say that the natural morphisms F(A1 ) ⊕ F(A2 ) → F(A1 ⊕ A2 )
and F(A1 ⊕ A2 ) → F(A1 ) ⊕ F(A2 ) are inverse isomorphisms for all objects A1 , A2 in C .
P ROOF. Suppose first that F is an additive functor. Note that F(ιi ◦ pi ) = idF(Ai ) and F(ιi ◦
p j ) = F(0) for i 6= j, but F(0) = 0 by additivity of F. Again by additivity of F, we have
F(ι1 ) ◦ F(p1 ) + F(ι2 ) ◦ F(p2 ) = F(idA1 ⊕A2 ) = idF(A1 ⊕A2 ) .
It follows that F(A1 ⊕ A2 ) is a biproduct of F(A1 ) and F(A2 ) in D, so in particular it is a coprod-
uct.
On the other hand, if F preserves biproducts and f , g : A → B are morphisms in C , then it is
easy to see that F( f ⊕ g) = F( f ) ⊕ F(g), and Lemma 9.8.8 tells us that
F( f + g) = F(∇B ◦ ( f ⊕ g) ◦ ∆A ) = ∇F(B) ◦ (F( f ) ⊕ F(g)) ◦ ∆F(A) = F( f ) + F(g).

C OROLLARY 9.8.12. Let F : C → D be a fully faithful functor of additive categories. Then
F is an additive functor.
302 9. CATEGORY THEORY

R EMARK 9.8.13. For additive functors, we may consider a finer notion of representability
than we previously studied. If F : C → Ab is an additive contravariant (resp., covariant) functor
of additive categories, then we may consider it to be representable if there exists an object X ∈
Obj(C ) and a natural isomorphism η : hX F (resp., η : hX F). In this case, the morphisms
ηA for A ∈ Obj(C ) will be isomorphisms of groups.
L EMMA 9.8.14. A morphism in an additive category that admits kernels is a monomorphism
if and only if it has zero kernel. A morphism in an additive category that admits cokernels is an
epimorphism if and only if it has zero cokernel.
P ROOF. Let f : A → B be a monomorphism, and let h : ker f → A be the induced morphism.
Since f ◦ h = 0 by definition of the kernel, we have h = 0, as f is a monomorphism. This forces
ker f to be 0, since h factors through 0. (Or, one could just apply Lemma 9.7.3.) On the other
hand, suppose that f has trivial kernel, and let g, h : C → A be maps with f ◦ g = f ◦ h. Then
f ◦ (g − h) = 0, and by universal property of the kernel, g − h factors through 0, i.e., is 0.
The proof for cokernels is similar, or is the result on kernels in the opposite (additive) cate-
gory. 
P ROPOSITION 9.8.15. Let C be an additive category that admits kernels and cokernels. Let
f : A → B be a morphism in C . Then
im f ∼
= ker(B → coker f )
and
coim f ∼
= coker(ker f → A).
P ROOF. We prove the first isomorphism. Let g : B → coker f . By Yoneda’s lemma, it suffices
to show that him f and hker g are naturally isomorphic. For C ∈ Obj(C ), we have a map
∼
HomC (C, im f ) −
→ {α : C → B | ι1 ◦ α = ι2 ◦ α}
that takes a morphism C → im f and composes it with the morphism im f → B given by definition
of the equalizer of the maps ιi : B → B qA B. It is a bijection by the universal property of the
equalizer.
For any D ∈ Obj(C ) and morphisms φ1 , φ2 : B → D such that φ1 ◦ f = φ2 ◦ f , note that there
exists a unique morphism k : B qA B → D with φi = k ◦ ιi . Any α : C → B such that ι1 ◦ α = ι2 ◦ α
then satisfies φ1 ◦α = φ2 ◦α for any such φi : B → D and any D. On the other hand, note that the ιi
themselves satisfy the property that ι1 ◦ f = ι2 ◦ f and are morphisms ιi : B → D with D = B qA B.
In other words, we have
{α : C → B | ι1 ◦ α = ι2 ◦ α}
= {α : C → B | φ1 ◦α = φ2 ◦α if φ1 ◦ f = φ2 ◦ f for some φ1 , φ2 : B → D (for some D ∈ Obj(C ))}.
Now, we are in an additive category, so this equals
(9.8.1) {α : C → B | φ ◦ α = 0 if φ ◦ f = 0 for some φ : B → D}.
By the universal property of coker f , for any φ : B → D with φ ◦ f = 0, there is a morphism
j : coker f → D with j ◦ g = φ . If g ◦ α = 0, then φ ◦ α = j ◦ g ◦ α = 0, and this works for any
 9.8. ADDITIVE AND ABELIAN CATEGORIES 303

φ : B → D with φ ◦ f = 0. On the other hand, g itself satisfies g ◦ f = 0 so is such a φ . It follows

that the set in (9.8.1) equals
{α : C → B | g ◦ α = 0}.
By the universal property of ker g, this is in bijection with HomC (C, ker g), taking an α in the
set to the unique morphism to ker g through which it factors. Clearly, the composition of these
bijections is natural in C, so we have the desired natural isomorphism. 
D EFINITION 9.8.16. An abelian category is an additive category C in which
i. every morphism in C admits a kernel and a cokernel and
ii. every morphism in C is strict.
E XAMPLES 9.8.17.
a. The category R-mod is abelian.
b. The full subcategory C of R-mod of finitely generated R-submodules is not necessarily
abelian. E.g., when R is commutative and non-noetherian, we can take I to be an ideal of R that
is not finitely generated, and so the kernel of R → R/I is not in C .
R EMARK 9.8.18. Note that if C is an abelian category, then so is C op . The roles of mono-
and epimorphisms, kernels and cokernels, and images and coimages switch in C and C op .
P ROPOSITION 9.8.19. The functor category Func(C , D) from a small category C to an
abelian category D is abelian.
P ROOF. We sketch the proof. First, note that it is additive: we have the zero functor which
sends all objects to the zero object and all morphisms to the zero (identity) morphism of the zero
object, and if F, G : C → D are functors, then F ⊕ G is given by (F ⊕ G)(C) = F(C) ⊕ G(C)
and (F ⊕ G)( f ) = F( f ) ⊕ G( f ) for f : A → B in C. This can be used to define the addition on
morphisms (i.e., natural transformations) as before.
Next, the kernel of a natural transformation η : F G is defined by (ker η)(C) = ker ηC and
(ker η)( f ) for f : A → B is the kernel of the induced morphism ker ηA → ker ηB . The cokernel is
defined similarly. Note that
(coim η)A ∼= coker(ker η η)A ∼= coker(ker ηA ηA ) ∼ = coim ηA ,
and similarly for images. Finally, since D is abelian, the natural map coim η im η is an
isomorphism coim ηA → im ηA on objects A in C , hence has a natural inverse determined by the
inverses of these morphisms. 
T ERMINOLOGY 9.8.20. In an abelian category C , we typically refer to a coproduct (when it
exists) as a direct sum, and we write i∈I Ai in place of qi∈I Ai .
L
 CHAPTER 10

Topics in Galois theory

10.1. Norm and trace

D EFINITION 10.1.1. Let E/F be a finite extension of fields. For α ∈ E, let mα : E → E
denote the F-linear transformation defined by left multiplication by α.
a. The norm map NE/F : E → F is defined by NE/F (α) = det mα for α ∈ E.
b. The trace map TrE/F : E → F is defined by TrE/F (α) = tr mα for α ∈ E.
R EMARK 10.1.2. For a finite field extension E/F, the trace map TrE/F is a homomorphism,
and the norm map NE/F is a homomorphism to F × upon restriction to E × .
P ROPOSITION 10.1.3. Let E/F be a finite extension of fields, and let α ∈ E. Let f ∈ F[x]
be the minimal polynomial of α over F, let d = [F(α) : F], let s = [E : F(α)], and let F be an
algebraic closure of F. Suppose that f factors in F[x] as
d
f = ∏(x − αi )
i=1

for some α1 , . . . , αd ∈ F. Then the characteristic polynomial of mα is f s , and we have

d d
NE/F (α) = ∏ αis and TrE/F (α) = s ∑ αi .
i=1 i=1

P ROOF. If {β1 , . . . , βs } is a basis for E/F(α), then {βi α j | 1 ≤ i ≤ s, 0 ≤ j ≤ d − 1} is

a basis for E/F. The matrix A representing mα with respect to this basis (with the lexico-
graphical ordering on the pairs (i, j)) is block diagonal with s blocks all equal to the matrix
for mα : F(α) → F(α) for the ordered basis {1, α, . . . , α d−1 }. As we have an isomorphism of
fields F(α) ∼
= F[x]/( f ) fixing F under which α is sent to the coset of x, the latter matrix is the
companion matrix A f .
By Lemma 8.10.10, we have char mα = f s . Lemma 8.7.21 tells us that
f s = xd − tr(mα )xd−1 + · · · + (−1)d det(mα ),
By expanding out the factorization of f s in F[x], we see that NE/F α and TrE/F α are as stated in
this case. 
We can also express the norm as a power of a product of conjugates and the trace as a multiple
of a sum of conjugates.
305
306 10. TOPICS IN GALOIS THEORY

P ROPOSITION 10.1.4. Let E/F be a finite field extension, and let t = [E : F]i be its degree of
inseparability. Then, for α ∈ E, we have
NE/F (α) = ∏ σ α t and TrE/F (α) = t ∑ σ α.
σ ∈EmbF (E) σ ∈EmbF (E)

P ROOF. The distinct conjugates of α in a fixed algebraic closure F of F are exactly the τα
for τ ∈ EmbF (F(α)). These τα are the distinct roots of the minimal polynomial of α over F,
each occuring with multiplicity the degree [F(α) : F]i of insparability of F(α)/F. Now, as in the
proof of Lemma 6.10.23, each of these embeddings extends to [E : F(α)]s distinct embeddings
of E into F, and each extension σ ∈ EmbF (E) of τ sends α to τ(α). By Proposition 10.1.3, we
have
NE/F α = ∏ (τα)[E:F(α)][F(α):F]i = ∏ σ α [E:F(α)]i [F(α):F]i = ∏ σ αt ,
τ∈EmbF (F(α)) σ ∈EmbF (E) σ ∈EmbF (E)
and similarly for the trace. 
We have the following immediate corollary.
C OROLLARY 10.1.5. Let E/F be a finite separable extension of fields. Then, for α ∈ E, we
have
NE/F (α) = ∏ σ α and TrE/F (α) = ∑ σ α.
σ ∈EmbF (E) σ ∈EmbF (E)

We also have the following.

P ROPOSITION 10.1.6. Let K/F be a finite field extension and E be an intermediate field in
the extension. Then we have
NK/F = NE/F ◦ NK/E and TrK/F = TrE/F ◦ TrK/E .
P ROOF. Since [K : F]i = [K : E]i [E : F]i , it suffices by Proposition 10.1.4 to show that
!
∏ δα = ∏ σ ∏ τα .
δ ∈EmbF (K) σ ∈EmbF (E) τ∈EmbE (K)

We extend each σ to an automorphism σ̃ of F fixing F. We then have

!
(10.1.1) ∏ σ ∏ τα = ∏ ∏ (σ̃ ◦ τ)α.
σ ∈EmbF (E) τ∈EmbE (K) σ ∈EmbF (E) τ∈EmbE (K)
For the trace map, we simply replace the products by sums.
Let σ , σ 0 ∈ EmbF (E) and τ, τ 0 ∈ EmbE (K), and suppose that
(10.1.2) σ̃ ◦ τ = σ̃ 0 ◦ τ 0 ∈ EmbF (K)
Since σ̃ ◦ τ|F = σ |F , we have that σ = σ 0 . Since σ̃ is an automorphism, we then apply its inverse
to (10.1.2) to obtain τ = τ 0 . As there are
| EmbF (E)|| EmbE (K)| = [E : F]s [K : E]s = [K : F]s = | EmbF (K)|
terms of the product in (10.1.1), we have the result. 
 10.1. NORM AND TRACE 307
√
E XAMPLE 10.1.7. The norm for the extension Q( d)/Q, where d is a square-free integer,
is given by
√ √ √
NQ(√d)/Q (x + y d) = (x + y d)(x − y d) = x2 − dy2
for x, y ∈ Q.
E XAMPLE 10.1.8. For a, b, c ∈ Q, we have
√
3
√
3
NQ( √
3
2)/Q
(a + b 2 + c( 2)2 )
√3
√
3
√
3
√
3
√
3
√
3
= (a + b 2 + c( 2)2 )(a + bω 2 + cω 2 ( 2)2 )(a + bω 2 2 + cω( 2)2 )
= a3 + 2b3 + 4c3 − 6abc,
for ω a primitive cube root of unity. The trace is simpler:
√
3
√
3
TrQ( √3 2)/Q (a + b 2 + c( 2)2 ) = 3a.

D EFINITION 10.1.9. A E-valued linear character of a group G is a group homomorphism

χ : G → E × , where E is a field.
D EFINITION 10.1.10. We say that a set of E-valued linear characters X of a group G is E-
linearly independent if it linearly independent as a subset of the E-vector space of functions
G → E.
T HEOREM 10.1.11. Any set of E-valued linear characters G → E × of a group G is E-linearly
independent.
P ROOF. Let X be a set of linear characters G → E × . Suppose by way of contradiction that
m ≥ 1 is minimal such that there m distinct, linearly dependent elements of G. Choose ai ∈ E
and χi ∈ X with 1 ≤ i ≤ m for which
m
∑ ai χi = 0.
i=1
Also, let h ∈ G be such that χ1 (h) 6= χm (h). Set bi = ai (χi (h) − χm (h)) for 1 ≤ i ≤ m − 1. For
any g ∈ G, we then have
m−1 m m m
∑ bi χi(g) = ∑ ai(χi(h) − χm(h))χi(g) ∑ ai χi(hg) − χm(h) ∑ ai χi(g) = 0.
i=1 i=1 i=1 i=1

Since b1 6= 0 and ∑m−1

i=1 bi χi has only m − 1 terms, this contradicts the existence of m. 
In the case of cyclic extensions, the kernels of the norm map bears a simple description.
T HEOREM 10.1.12 (Hilbert’s Theorem 90). Let E/F be a finite cyclic extension of fields, and
let σ be a generator of its Galois group. Then
 
σ (β ) ×
ker NE/F = |β ∈E .
β
308 10. TOPICS IN GALOIS THEORY

P ROOF. Set n = [E : F]. Let β ∈ E, and note that

 n−1 i+1
σ (β ) NE/F (β )

σ (β )
NE/F =∏ i = = 1.
β i=0 σ (β ) NE/F (β )
Next, suppose that α ∈ ker NE/F , and set
xγ = γ + ασ (γ) + ασ (α)σ 2 (γ) + · · · + ασ (α) · · · σ n−2 (α)σ n−1 (γ)
for γ ∈ E. The elements of Gal(E/F), which is to say the powers of σ , are distinct E-valued
characters on E × , and therefore they are E-linearly independent. Thus, there exists γ ∈ E × such
that xγ 6= 0. We then note that
ασ (xγ ) = ασ (γ) + ασ (α)σ 2 (γ) + · · · + ασ (α) · · · σ n−2 (α)σ n−1 (γ) + NE/F (α)γ = xγ ,
σ (xγ−1 )
so α = , finishing the proof. 
xγ−1

There is also an additive form of Hilbert’s Theorem 90, which describes the kernel of the
trace. We leave the proof to the reader.
P ROPOSITION 10.1.13 (Additive Hilbert’s Theorem 90). Let E/F be a finite cyclic extension
of fields, and let σ be a generator of its Galois group. Then
ker TrE/F = {σ (β ) − β | β ∈ E} .

10.2. Discriminants
In this section, we give a second treatment of discriminants.
D EFINITION 10.2.1. Let F be a field and V a finite-dimensional F-vector space. A F-bilinear
form is a F-bilinear map ψ : V ×V → F.
D EFINITION 10.2.2. A F-bilinear form ψ on a F-vector space V is said to be symmetric if
ψ(v, w) = ψ(w, v) for all v, w ∈ V .
E XAMPLE 10.2.3. Given a matrix Q ∈ Mn (F), we can define a bilinear form on F n by
ψ(v, w) = vT Qw
for v, w ∈ F n which is symmetric if and only if Q is.
E XAMPLE 10.2.4. If E/F is a finite extension of fields, then ψ : E × E → F defined by
ψ(α, β ) = TrE/F (αβ )
for α, β ∈ E is a symmetric F-bilinear form on E.
D EFINITION 10.2.5. The discriminant D(ψ) of a bilinear form ψ on a finite dimensional
F-vector space V relative to an ordered basis (v1 , . . . , vn ) of V is the determinant of the matrix
with (i, j) entry equal to ψ(vi , v j ).
 10.2. DISCRIMINANTS 309

L EMMA 10.2.6. Let ψ : V ×V → F be a F-bilinear form on a finite-dimensional vector space

V of dimension n ≥ 1. Let v1 , . . . , vn ∈ V , and let T : V → V be a F-linear transformation. Then
det(ψ(T vi , T v j )) = (det T )2 · det(ψ(vi , v j ))
P ROOF. It suffices to show this in the case that the vi form a basis of V , since in this case, for
any w1 , . . . , wn ∈ V there exists a linear transformation U : V → V with U(vi ) = wi for all i. We
then have
det(ψ(Twi , Tw j )) = det(ψ(TUvi , TUv j )) = det(TU)2 det(ψ(vi , v j )) = det(T )2 det(ψ(wi , w j )).
So, assume that the vi form a basis of V , and let A = (ai j ) denote the matrix of T with respect
to the ordered basis (v1 , . . . , vn ) of V . We have T vi = ∑nk=1 aki vk for each i, and therefore we have
n n
ψ(T vi , T v j ) = ∑ aki ∑ al j ψ(vk , vl ).
k=1 l=1
As matrices, we then have
(ψ(T vi , T v j )) = AT (ψ(vi , v j ))A,
and the result follows as det T = det A = det AT . 
R EMARK 10.2.7. It follows from Lemma 10.2.6 that the discriminant of a bilinear form with
respect to a basis is independent of its ordering, since a permutation matrix has determinant ±1.
D EFINITION 10.2.8. Let E/F be a finite extension of fields. The discriminant D(β1 , . . . , βn )
of E/F relative to an ordered basis (β1 , . . . , βn ) of E as a F-vector space is the discriminant of
the bilinear form
(α, β ) 7→ TrE/F (αβ )
relative to the basis.
P ROPOSITION 10.2.9. Let E/F be a finite separable extension of fields. Then the discrimi-
nant of E/F relative to an ordered basis (β1 , . . . , βn ) of E satisfies
D(β1 , . . . , βn ) = (det(σi β j ))2 ,
where {σ1 , . . . , σn } is the set of embeddings of E in an algebraic closure of F that fix F.
P ROOF. Note that
n
TrE/F (βi β j ) = ∑ σk (βi )σk (β j ),
i=1
so the matrix (TrE/F (βi β j )) equals QT Q, where Q ∈ Mn (E) satisfies Qi j = σi (β j ). 
D EFINITION 10.2.10. Let F be a field, and let α1 , . . . , αn ∈ F. The Vandermonde matrix for
α1 , . . . , αn is  
1 α1 · · · α1n−1
1 α · · · α n−1 
2 2 
Q(α1 , . . . , αn ) =  . . .

.
. . . .. 
. 
1 αn · · · αnn−1
310 10. TOPICS IN GALOIS THEORY

L EMMA 10.2.11. Let F be a field, and let Q(α1 , . . . , αn ) be the Vandermonde matrix for
elements α1 , . . . , αn of F. Then
det Q(α1 , . . . , αn ) = ∏ (α j − αi ).
1≤i< j≤n

P ROOF. We work by induction on n ≥ 1, the case n = 1 asserting the obvious fact that
det Q(α) = 1 for any α ∈ F. To compute the determinant of Q = Q(α1 , . . . , αn ), in order of
descending i ≤ n − 1 subtract α1 times the ith column of Q from the (i + 1)th column, which
leaves the determinant unchanged. We then obtain
1 0 ··· 0
α2 − α1 · · · α2n−2 (α2 − α1 )
1 α2 − α1 · · · α2n−2 (α2 − α1 ) .. ..
det Q = .. .. .. = . .
. . . n−2
αn − α1 · · · αn (αn − α1 )
1 αn − α1 · · · αnn−2 (αn − α1 )
n
= ∏(αi − α1 ) · Q(α2 , . . . , αn ),
i=2
and the result now follows by induction. 
P ROPOSITION 10.2.12. Suppose that E/F is a separable extension of degree n, and let α ∈ E
be such that E = F(α). Then
D(1, α, . . . , α n−1 ) = D( f ),
where f ∈ F[x] is the minimal polynomial of α.
P ROOF. Let σ1 , σ2 , . . . , σn be the embeddings of E in a fixed algebraic closure of F, and set
j−1
αi = σi (α). Then σi (α j−1 ) = αi , so Lemmas 10.2.9 and 10.2.11 tell us that
D(1, α, . . . , α n−1 ) = det Q(α1 , α2 , . . . , αn )2 = ∏ (α j − αi )2 .
1≤i< j≤n

The latter term is just D( f ). 

D EFINITION 10.2.13. Let F be a field and f = ∑ni=0 ai xi ∈ F[x]. The derivative f 0 ∈ F[x] of
f is f 0 = ∑ni=1 iai xi−1 .
R EMARK 10.2.14. An irreducible polynomial f ∈ F[x] is inseparable if and only if f 0 = 0.
P ROPOSITION 10.2.15. Suppose that E/F is a separable extension of degree n, and let α ∈ E
be such that E = F(α), and let f ∈ F[x] be the minimal polynomial of α. Then
n(n−1)
D( f ) = (−1) 2 NE/F ( f 0 (α)),
where f 0 ∈ F[x] is the derivative of f .
P ROOF. Let α1 , . . . , αn be the conjugates of α in an algebraic closure F of F. Then
n n
0
f (x) = ∑ ∏ (x − α j ),
i=1 j=1
j6=i
 10.3. EXTENSIONS BY RADICALS 311

so we have
n
f 0 (αi ) = ∏ (αi − α j )
j=1
j6=i
for each i, and the conjugates of f 0 (α) in F are the f 0 (αi ). We then have
n n n(n−1)
NE/F ( f 0 (α)) = ∏ ∏ (αi − α j ) = (−1) 2 D( f )
i=1 j=1
j6=i

C OROLLARY 10.2.16. Let L/F be a finite separable extension of fields. Then the discrimi-
nant of L/F relative to an ordered basis (β1 , β2 , . . . , βn ) of L is nonzero.
P ROOF. Since L/F is separable, there exists α ∈ L such that L = F(α). Then (1, α, . . . , α n−1 )
is an ordered basis of L/F, and there exists an invertible F-linear transformation T : L → L with
T (α i−1 ) = βi for 1 ≤ i ≤ n. By Lemma 10.2.6, we have that
D(β1 , β2 , . . . , βn ) = (det T )2 D(1, α, . . . , α n−1 ).
It follows Proposition 10.2.12 that D(1, α, . . . , α n−1 ) 6= 0, so we have the result. 
R EMARK 10.2.17. Together, Lemma 10.2.6 and Corollary 10.2.16 tell us that the discrimi-
nant of a finite separable field extension L/F (relative to an ordered basis) reduces to a element
of F × /F ×2 that is independent of the choice of basis.
10.3. Extensions by radicals
D EFINITION 10.3.1. Let F be a field. A Kummer extension E of F is one that is given by
adjoining roots of elements of F.
N OTATION 10.3.2. For a field F of characteristic not dividing n ≥ 1, we let µn denote the
group of nth roots of unity in a given algebraic closure of F.
P ROPOSITION 10.3.3. Let F be a field, and let F be a fixed algebraic closure of F. Let n ≥ 1
and a ∈ F. Let E = F(α) for α ∈ F with α n = a, and let d ≥ 1 be minimal such that α d ∈ F.
a. The extension E/F is Galois if and only if char F does not divide d and E contains µd .
b. If E/F is Galois and µd ⊂ F, then the map
σ (α)
χa : Gal(E/F) −∼
→ µd , χa (σ ) = .
α
is an isomorphism of groups.
P ROOF. The minimal polynomial f of α divides xd − α d but not xm − α m for any m dividing
d. If µd has order d, then f is separable as xd − α d is. If µd has order m for some m properly
dividing d, then f divides xd − α d = (xm − α m )d/m but not xm − α m so is inseparable. Note that
char F does not divide d if and only if µd has order d, so we suppose for the remainder of the
proof that this holds.
312 10. TOPICS IN GALOIS THEORY

Any field embedding σ of E in F fixing F must send α to ζ α for some ζ ∈ µd . If µd ⊂ E,

then every such element lies in E, so E/F is Galois. Conversely, if E/F is Galois, then since f
does not divide any xm − α m with m properly dividing d and µd has order d, it has a root of the
form ζ α with ζ ∈ µd of order d. Then ζ = (ζ α) · α −1 ∈ E, so µd ⊂ E.
Finally, if E/F is Galois and µd ⊂ F, then the map χa as defined in the statement is bijective
by what we have already said, and it satisfies
!
σ τ(α) σ (α) τ(α)
χa (σ τ) = = ·σ = χa (σ ) · σ (χa (τ)) = χa (σ )χa (τ)
α α α
for σ , τ ∈ Gal(E/F), noting that σ fixes µd . 
D EFINITION 10.3.4. Let F be a field containing µn for some n ≥ 1 that is not divisible by
char F. For any a ∈ F × and extension E/F containing an nth root of a, the Kummer character
attached to a is the homorphism χa : Gal(E/F) → µn given by
√
σ ( n a)
χa (σ ) = √ n
a
for σ ∈ Gal(E/K).
P ROPOSITION 10.3.5. Let F be a field of characteristic not dividing n ≥ 1, and suppose √that
F contains the nth roots of unity. Let E be a cyclic extension of F of degree n. Then E = F( n a)
for some a ∈ F × .
P ROOF. Let ζn be a primitive nth root of unity in F. Note that NE/F (ζ ) = ζ n = 1, so Hilbert’s
σ (α)
Theorem 90 tells us that there exists α ∈ E and a generator σ of Gal(E/F) with α = ζ . Note
that
n n n(n−1)
NE/F (α) = ∏ σ i α = ∏ ζ i α = ζ 2 α n = (−1)n−1 α n ,
i=1 i=1
so setting a = −NE/F (−α), we have α n = a. Since α has n distinct conjugates in E, we have
that E = F(α). 
N OTATION 10.3.6. Let ∆ be √ a subset of a field K, and let n ≥ 1 be such that K contains the
nth roots of unity in K. Then K( n ∆) is the field given by adjoining an nth root of each element
of ∆ to K.
T HEOREM 10.3.7 (Kummer duality). Let F be a field of characteristic not dividing n ≥ 1, and
suppose that F contains the nth roots of unity. Let E√
be a finite abelian extension of F of exponent
dividing n, and set ∆ = E ×n ∩ F × . Then E = F( n ∆), and there is a perfect bimultiplicative
pairing
h , i : Gal(E/F) × ∆/F ×n → µn
given by hσ , ai = χa (σ ) for σ ∈ Gal(E/F) and a ∈ ∆.
P ROOF. Since µn ⊂ F, Proposition 10.3.3 tells us that the map taking a ∈ ∆ to its Kummer
cocycle χa yields an injection
ψ : ∆/F ×n → Hom(Gal(E/F), µn ).
 10.3. EXTENSIONS BY RADICALS 313

This gives rise to the bimultiplicative Kummer pairing h , i, and it implies that any a ∈ ∆/F ×n
of order d dividing n pairs with some element of Gal(E/F) to a dth root of unity.
We claim that ψ is surjective. Let χ : Gal(E/F) → µn be a homomorphism, and let H =
ker χ, which by the fundamental theorem of Galois theory corresponds to some cyclic extension
K/F of degree dividing n. By Proposition 10.3.5, we have that K = F(α) for some α with
a = α n ∈ ∆, and then χ = χak for some k ≥ 1. That is, χ = χb with b = ak ∈ ∆, so ψ(b) = χ.
Since ∆/F ×n is therefore finite of degree [E : F], we have that the map
Gal(E/F) → Hom(∆/F ×n , µn )
induced by the pairing is an isomorphism as well, and thus the Kummer pairing is perfect. 
R EMARK 10.3.8. One may replace ∆ in Theorem 10.3.7 by any Γ ⊆ ∆ with ∆ = ΓK ×n . Then
∆/K ×n should be replaced by the isomorphic Γ/(Γ ∩ K ×n ).
D EFINITION 10.3.9. A finite field extension E/F is solvable by radicals
√ if there exists s ≥ 0
and fields Ei for 0 ≤ i ≤ s with E0 = F, E ⊆ Es , and Ei+1 = Ei ( ni αi ) for some αi ∈ Ei and
integers ni ≥ 1 for 0 ≤ i < s. If we can take Es = E, then we say that E is a radical extension of
F.
T HEOREM 10.3.10. Let F be a field. Let f ∈ F[x] be nonconstant, and suppose that its
splitting field K has degree over F not divisible by char F. Then K is solvable by radicals if and
only if Gal(K/F) is a solvable group.
P ROOF. First, take n = [K : F], and consider L = K(ζn ) and E = F(ζn ) for a primitive nth
root of unity ζn (which exists by assumption on char F). Then K/F is solvable by radicals if and
only if L/E is, since ζn is an nth root of 1. Moreover, L = KE is Galois over E, and Gal(L/E) is
solvable if and only if Gal(K/F) is, since Gal(L/E) is isomorphic to the subgroup Gal(K/K ∩ E)
of Gal(K/F) by restriction, and (K ∩ E)/F is abelian, hence solvable, in that E/F is abelian.
Thus, we have reduced to the case that F contains the nth roots of unity.
If K/F is solvable by radicals, then there exists a field L containing F that is a radical exten-
sion of F. We claim√ that we may take L/F to be Galois. Suppose that K = Ks where K0 = F
and Ki+1 = Ki ( i αi ) with ni dividing n and αi ∈ Ki for i < s. Then let L0 = F and let Li+1 be
n

the field given by adjoining to Li an ni th root of each conjugate of αi over √ F. Then Li+1 is the
compositum of Li and the splitting field of the minimal polynomial of i αi over F, so is Galois
n

over F for each i. Moreover, L = Ls is by definition a radical extension of F. Now, Gal(L/F) is

solvable, since Gal(Li+1 /Li ) is abelian in that Li contains the ni th root of unity. Since Gal(L/F)
is solvable, so is Gal(K/F).
Conversely, if Gal(K/F) is solvable, then we have intermediate fields Ki with K0 = F, Ks = K,
and Ki ⊂ Ki+1 such that Ki+1 /F is Galois and Gal(Ki+1 /Ki ) is cyclic of degree dividing n. But
then Ki+1 is a Kummer extension of Ki , given by adjoining the ni th root of some αi ∈ Ki , where
ni = [Ki+1 : Ki ]. So, K/F is in fact a radical extension. 
C OROLLARY 10.3.11. If F is a field of characteristic not dividing 6 and K is the splitting
field over F of a polynomial of degree at most 4, then K/F is solvable by radicals.
P ROOF. We know that Gal(K/F) is isomorphic to a subgroup of Sn for n equal to the degree
of the polynomial defining K, and Sn is solvable for n ≤ 4, so Gal(K/F) is solvable as well. 
314 10. TOPICS IN GALOIS THEORY

E XAMPLE 10.3.12. The splitting field K of the polynomial f = 2x5 − 10x + 5 ∈ Q[x] has
Galois group isomorphic to S5 , and S5 is insolvable, so K/Q is not solvable by radicals. To see
this, note first that the polynomial is irreducible by the Eisenstein criterion for the prime 5. So,
5 divides [K : Q], and hence the image G in S5 of Gal(K/Q) under a permutation representation
of the roots contains a 5 cycle. Moreover, f 0 = 10(x4 − 1) has real roots at ±1 and f (−1) > 0
while f (1) < 0, so f has exactly three real roots. In particular, if τ ∈ Gal(K/Q) is the restriction
of complex conjugation, then τ fixes the three real roots and transposes the two imaginary roots,
so G contains a transposition. But S5 is generated by any five cycle and any transposition, so
Gal(K/Q) ∼ = G = S5 .

10.4. Linearly disjoint extensions

P ROPOSITION 10.4.1. Let K be a field, and let f ∈ K[x] be monic and irreducible. Let M be
ei
a field extension of K, and suppose that f factors as ∏m
i=1 f i in M[x], where the f i are irreducible
and distinct and each ei is positive. Then we have an isomorphism
m
∼
→ ∏ M[x]/( fiei )
κ : K[x]/( f ) ⊗K M −
i=1
of M-algebras such that if g ∈ K[x], then κ((g + ( f )) ⊗ 1) = (g + ( fiei ))i .
∼
P ROOF. Note that we have a canonical isomorphism K[x] ⊗K M −
→ M[x] that gives rise to the
first map in the composition
m
∼ ∼
K[x]/( f ) ⊗K M − → ∏ M[x]/( fiei ),
→ M[x]/( f ) −
i=1
the second isomorphism being the Chinese remainder theorem. The composition is κ. 
We have the following consequence.
L EMMA 10.4.2. Let L/K be a finite separable extension of fields, and let M be an alge-
braically closed field containing K. Then we have an isomorphism of M-algebras
∼
κ : L ⊗K M −
→ ∏ M,
σ : L,→M
where the product is taken over field embeddings of L in M fixing K, such that
κ(β ⊗ 1) = (σ β )σ
for all β ∈ L.
P ROOF. Write L = K(θ ), and let f ∈ K[x] be the minimal polynomial of θ . Then we define
κ as the composition
∼ M[x] ∼
L ⊗K M − → −
→ ∏ M,
(x − σ (θ )) σ : L,→M
where the first isomorphism is that of Proposition 10.4.1 and the second takes x to σ (θ ) in the
coordinate corresponding to σ . Any β ∈ L has the form g(θ ) for some g ∈ K[x], and since any
σ : L ,→ M fixing K fixes the coefficients of g, we have κ(β ⊗ 1) is as stated. 
 10.4. LINEARLY DISJOINT EXTENSIONS 315

R EMARK 10.4.3. If we compose κ of Lemma 10.4.2 with the natural embedding L ,→ L⊗K M
that takes α ∈ L to α ⊗ 1, then the composition
ιM : L → ∏ M
σ : L,→M

is the product of the field embeddings σ of L in M fixing K.

D EFINITION 10.4.4. Let K be a field and L and M be extensions of K both contained in some
field Ω. We say that L and M are linearly disjoint over K if every K-linearly independent subset
of L is M-linearly independent.
L EMMA 10.4.5. Let K be a field and L and M be extensions of K both contained in some field
Ω. If L and M are linearly disjoint over K, then L ∩ M = K.
P ROOF. If x ∈ L ∩ M with x ∈/ K, then x and 1 are elements of L that are K-linearly indepen-
dent but not M-linearly independent, so L and M are not linearly disjoint over K. 
From the definition, it may not be clear that the notion of linear disjointness is a symmetric
one. However, this follows from the following.
P ROPOSITION 10.4.6. Let K be a field and L and M be extensions of K both contained in
some field Ω. Then L and M are linearly disjoint over K if and only if the map ϕ : L ⊗K M → LM
induced by multiplication is an injection.
P ROOF. Suppose that γ1 , . . . , γs ∈ M are L-linearly dependent, and write ∑si=1 βi γi = 0 for
some βi ∈ L. If ϕ is injective, then we must have ∑si=1 βi ⊗ γi = 0, which means that the γi are
K-linearly dependent.
Conversely, let L and M be linearly disjoint over K. Suppose that we have a nonzero
s
x = ∑ βi ⊗ γi ∈ ker ϕ
i=1

for some βi ∈ L and γi ∈ M, with s taken to be minimal. If x 6= 0, then the γi are L-linearly
dependent, so they are K-linearly dependent. In this case, without loss of generality, we may
suppose that
s−1
γs + ∑ αi γi = 0
i=1
for some αi in K. Then
s−1
x= ∑ (βi − αiβs) ⊗ γi,
i=0
contradicting minimality. Thus ker ϕ = 0. 
C OROLLARY 10.4.7. Let K be a field and L and M be extensions of K both contained in a
given algebraic closure of K. Then L and M are linearly disjoint over K if and only if L ⊗K M is
a field.
316 10. TOPICS IN GALOIS THEORY

P ROOF. Note that LM is a union of subfields of the form K(α, β ) with α ∈ L and β ∈ M.
Since α and β are algebraic over K, we have K(α, β ) = K[α, β ], and every element of the
latter ring is a K-linear combination of monomials in α and β . Thus ϕ of Proposition 10.4.6 is
surjective, and the result follows from the latter proposition. 
C OROLLARY 10.4.8. Let K be a field and L and M be finite extensions of K inside a given
algebraic closure of K. Then [LM : K] = [L : K][M : K] if and only if L and M are linearly disjoint
over K.
P ROOF. Again, we have the surjection ϕ : L ⊗K M → LM given by multiplication which is
an injection if and only if L and M are linearly disjoint by Proposition 10.4.6. As L ⊗K M has
dimension [L : K][M : K] over K, the result follows. 
R EMARK 10.4.9. Suppose that L = K(θ ) is a finite extension of K. To say that L is linearly
disjoint from a field extension M of K is by Propostion 10.4.1 exactly to say that the minimal
polynomial of θ in K[x] remains irreducible in M[x].
We prove the following in somewhat less generality than possible.
L EMMA 10.4.10. Let L be a finite Galois extension of a field K inside an algebraic closure
Ω of K, and let M be an extension of K in Ω. Then L and M are linearly disjoint if and only if
L ∩ M = K.
P ROOF. We write L = K(θ ) for some θ ∈ L, and let f ∈ K[x] be the minimal polynomial of
θ . As Gal(LM/M) ∼ = Gal(L/(L ∩ M)) by restriction, we have L ∩ M = K if and only if [LM :
M] = [L : K]. Since LM = M(θ ), this occurs if and only if f is irreducible in M[x]. The result
then follows from Remark 10.4.9. 

10.5. Normal bases

D EFINITION 10.5.1. A normal basis of a finite Galois extension L/K is a basis of L as a
K-vector space of the form {σ (α) | σ ∈ Gal(L/K)} for some α ∈ L.
The goal of this section is to prove E. Noether’s theorem that every finite Galois extension
has a normal basis. We start with the following lemma.
L EMMA 10.5.2. Let L/K be a finite Galois extension with Galois group {σ1 , . . . , σn }, where
n = [L : K]. Let {α1 , . . . , αn } be a basis of L as a K-vector space. Then the set
{(σ1 (α j ), . . . , σn (α j )) | 1 ≤ j ≤ n}
is an L-basis of Ln .
P ROOF. Let W be the L-span of the subset of Ln in question. Set W ∨ = HomL (W, L), and
let ϕ ∈ (Ln )∨ be such that ϕ(W ) = 0. It suffices to show that ϕ = 0. Note that there exists
u = (a1 , . . . , an ) ∈ Ln such that ϕ(v) = uT v for all v ∈ Ln , so ∑ni=1 ai σi (α j ) = 0 for all 1 ≤ j ≤ n.
As {α1 , . . . , αn } is a K-basis of L, we therefore have that ∑ni=1 ai σi vanishes on L. Since the σi
are L-linearly independent, we have ai = 0 for all i, and therefore ϕ = 0. 
L EMMA 10.5.3. Every finite cyclic extension of fields has a normal basis.
 10.6. PROFINITE GROUPS 317

P ROOF. Let L/K be finite cyclic of degree n, generated by an element σ . Then K[Gal(L/K)]
is isomorphic to K[x]/(xn − 1) via the unique K-algebra homomorphism that takes σ to x. As L is
a K[Gal(L/K)]-module, it becomes a K[x]-module annihilated by xn − 1. If f = ∑n−1 i
i=0 ci x ∈ K[x]
annihilates L, then ∑n−1 i
i=0 ci σ (α) = 0 for all α ∈ L, which by the linear independence of the σ
i

forces f to be zero. Thus, the annihilator of L is (xn − 1), and by the structure theorem for
finitely generated modules over the PID K[x], this means that L has a K[x]-summand isomorphic
to K[x]/(xn − 1), generated by some α ∈ L. Since the latter module has K-dimension n, as does
L, the elements {α, σ (α), . . . , σ n−1 (α)} form a K-basis of L.

T HEOREM 10.5.4 (Normal basis theorem). Every finite Galois extension of fields has a nor-
mal basis.
P ROOF. Let L/K be a finite Galois extension of degree n. Since any finite extension of
finite fields is cyclic, we may by Lemma 10.5.3 suppose that K is infinite. Write Gal(L/K) =
{σ1 , . . . , σn } and σ1 = 1. Let {α1 , . . . , αn } be a basis of L as a K-vector space. It suffices to find
β ∈ L with D(σ1 (β ), . . . , σn (β )) 6= 0 by Corollary 10.2.16.
Define an element p ∈ L[x1 , . . . , xn ] by
!2
n
p(x1 , . . . , xn ) = det ∑ σ −1
j σi (αk )xk .
k=1

Note that the coefficients of p are fixed by the elements of Gal(L/K), since they permute the
columns of the matrix. By Lemma 10.5.2, we can find β j ∈ L for 1 ≤ j ≤ n be such that
n
∑ β j (σ1(α j ), σ2(α j ), . . . , σn(α j )) = (1, 0, . . . , 0).
j=1

Then for all 1 ≤ i, j ≤ n, we have

n
∑ σ −1
j σi (αk )βk = δi, j ,
k=1

so p(β1 , . . . , βn ) = det(In )2 = 1, so p 6= 0. Since K is infinite, there exist a1 , . . . , an ∈ K with

p(a1 , . . . , an ) 6= 0. For γ = ∑nj=1 ai αi , we have by Proposition 10.2.9 the first equality in
D(σ1 (γ), . . . , σn (γ)) = det(σ −1 2
j σi (γ)) = p(a1 , . . . , an ) 6= 0.


10.6. Profinite groups

D EFINITION 10.6.1. A topological group G is a group endowed with a topology with respect
to which both the multiplication map G × G → G and the inversion map G → G that takes an
element to its inverse are continuous.
E XAMPLES 10.6.2.
318 10. TOPICS IN GALOIS THEORY

a. The groups R, C, R× , and C× are continuous with respect to the topologies defined by
their absolute values.
b. Any group can be made a topological group by endowing it with the discrete topology.
R EMARK 10.6.3. We may consider the category of topological groups, in which the maps
are continuous homomorphisms between topological groups.
D EFINITION 10.6.4. A homomorphism φ : G → G0 between topological groups G and G0 is
a topological isomorphism if it is both an isomorphism and a homeomorphism.
The following lemma is almost immediate, since elements of a group are invertible.
L EMMA 10.6.5. Let G be a topological group and g ∈ G. Then the map mg : G → G with
mg (a) = ga for all a ∈ G is a topological isomorphism.
We also have the following.
L EMMA 10.6.6. A group homomorphism φ : G → G0 between topological groups is continu-
ous if and of only, for each open neighborhood U of 1 in G0 with 1 ∈ U, the set φ −1 (U) contains
an open neighborhood of 1.
P ROOF. We consider the non-obvious direction. Let V be an open set in G0 , and suppose
that g ∈ G is such that h = φ (g) ∈ V . Then h−1V is open in G0 as well, by Lemma 10.6.5. By
assumption, there exists an open neighborhood W of 1 in G contained in φ −1 (h−1V ), and so gW
is an open neighborhood of g in G such that φ (gW ) ⊆ V . Hence, φ is continuous. 
L EMMA 10.6.7. Let G be a topological group.
a. Any open subgroup of G is closed.
b. Any closed subgroup of finite index in G is open.
P ROOF. If H is an open (resp., closed) subgroup of G, then its cosets are open (resp., closed)
as well. Moreover, G − H is the union of the nontrivial cosets of H. Therefore, G − H is open if
G is open and closed if G is closed of finite index, so that there are only finitely many cosets of
H. 
L EMMA 10.6.8. Every open subgroup of a compact group G is of finite index in G.
P ROOF. Let H be a open subgroup of G. Note that G is the union of its distinct H-cosets,
which are open and disjoint. Since G is compact, there can therefore only be finitely many cosets,
which is to say that H is of finite index in G. 
We leave it to the reader to verify the following.
L EMMA 10.6.9.
a. A subgroup of a topological group is a topological group with respect to the subspace
topology.
b. The quotient of a topological group G by a normal subgroup N is a topological group with
respect to the quotient topology, and it is Hausdorff if G is Hausdorff and N is closed.
 10.6. PROFINITE GROUPS 319

c. A direct product of topological groups is a topological group with respect to the product
topology.
R EMARK 10.6.10. The category of topological Hausdorff abelian groups is not abelian,
though it is additive and admits kernels and cokernels. For instance, consider the inclusion map
ι : Q → R with R having its usual topology and Q having the subspace topology. Then ker ι = 0
and coker ι = 0 (since Q is dense in R, and thus every continuous map from R is determined by
its values on Q). By Proposition 9.8.15, we have im ι ∼
= R but coim ι ∼= Q.
Recall that an inverse system of groups is covariant functor from a codirected set to the
category of groups. Here, we shall view it as a contravariant functor from a directed set I, or
more concretely, a collection (Gi , φi, j ) of groups Gi for i ∈ I and homomorphisms φi, j : Gi → G j
for every i, j ∈ I with j ≤ i. The following simple proposition gives a direct construction of the
inverse limit of an inverse system.
P ROPOSITION 10.6.11. Let (Gi , φi, j ) be an inverse system of groups over a directed indexing
set I. Then the an inverse limit of the system is given explicitly by the group
( )
G= (gi )i ∈ ∏ Gi | φi, j (gi ) = g j
i∈I
and the maps πi : G → Gi for i ∈ I that are the compositions of the G → ∏i∈I Gi → Gi of inclusion
followed by projection.
We may endow an inverse limit of groups with a topology as follows.
D EFINITION 10.6.12. Let (Gi , φi, j ) be an inverse system of topological groups over an in-
dexing set I, with continuous maps. Then the inverse limit topology on the inverse limit G of
Proposition 10.6.11 is the subspace topology for the product topology on ∏i∈I Gi .
L EMMA 10.6.13. The inverse limit of an inverse system (Gi , φi, j ) of topological groups (over
a directed indexing set I) is a topological group under the inverse limit topology.
P ROOF. The maps
∏ Gi × ∏ Gi → ∏ Gi and ∏ Gi → ∏ Gi
i∈I i∈I i∈I i∈I i∈I
given by componentwise multiplication and inversion are clearly continuous, and this continuity
is preserved under the subspace topology on the inverse limit. 
R EMARK 10.6.14. In fact, the inverse limit of an inverse system of topological groups and
continuous maps, when endowed with the product topology, is an inverse limit in the category of
topological groups.
When we wish to view it as a topological group, we typically endow a finite group with the
discrete topology.
D EFINITION 10.6.15. A profinite group is an inverse limit of a system of finite groups, en-
dowed with the inverse limit topology for the discrete topology on the finite groups.
320 10. TOPICS IN GALOIS THEORY

Recall the following definition.

D EFINITION 10.6.16. A topological space is totally disconnected if and only if every point
is a connected component.
We leave the following as difficult exercises.
P ROPOSITION 10.6.17. A compact Hausdorff space is totally disconnected if and only if it
has a basis of open neighborhoods that are also closed.
P ROPOSITION 10.6.18. A compact Hausdorff group that is totally disconnected has a basis
of neighborhoods of 1 consisting of open normal subgroups (of finite index).
We may now give a topological characterization of profinite groups.
T HEOREM 10.6.19. A profinite topological group G is compact, Hausdorff, and totally dis-
connected.
P ROOF. First, suppose that G is profinite, equal to an inverse limit of a system (Gi , φi, j ) of
finite groups over an indexing set I. The direct product ∏i∈I Gi of finite (discrete) groups Gi is
compact Hausdorff (compactness being Tychonoff’s theorem). As a subset of the direct product,
G is Hausdorff, and to see it is compact, we show that G is closed. Suppose that
(gi )i ∈ ∏ Gi
i∈I
with (gi )i ∈
/ G, and choose i, j ∈ I with i > j and φi, j (gi ) 6= g j . The open subset
( )
(hk )k ∈ ∏ Gk | hi = gi , h j = g j
k∈I
of the direct product contains (gi )i and has trivial intersection with G. In that the complement of
G is open, G itself is closed. Finally, note that any open set ∏i∈I Ui with each Ui open in Gi (i.e.,
an arbitrary subset) and Ui = Gi for all but finitely many i is also closed. That is, its complement
is the intersection !
\
(G j −U j ) × ∏ Ui
j∈I i∈I−{ j}
of open sets, which is actually equal to the finite intersection over j ∈ I with Ui 6= Gi . It is
therefore open, and by Proposition 10.6.17, the group G is totally disconnected. 
R EMARK 10.6.20. We leave it to the reader to check that the converse to Theorem 10.6.19
also holds. They key is found in the proof of part a of the following proposition.
P ROPOSITION 10.6.21. Let G be a profinite group, and let U be the set of all open normal
subgroups of G. Then the following canonical homomorphisms are homeomorphisms:
a. G → limN∈U G/N,
←−
b. H → limN∈U H/(H ∩ N), for H a closed subgroup of G, and
←−
c. G/K → limN∈U G/NK, for K a closed normal subgroup of G.
←−
 10.6. PROFINITE GROUPS 321

P ROOF. We prove part a. The continuous map φ from G to the inverse limit Q of its quotients
has closed image, and φ is injective since U is a basis of 1 in G as in Proposition 10.6.18.
Suppose that (gN N)N∈U is not in the image of φ , which is exactly to say that the intersection
of the closed sets gN N is empty. Since G is compact this implies that some finite subset of the
{gN N | N ∈ U } is empty, and letting M be the intersection of the N in this subset, we see that
gM M = ∅, which is a contradiction. In other words, φ is surjective. 
The following is a consequence of Proposition 10.6.21a. We leave the proof to the reader.
C OROLLARY 10.6.22. Let G be a profinite group and V a set of open normal subgroups of
G that forms a basis of open neighborhoods of 1. Then the homomorphism
G → lim G/N
←−
N∈V
is a homeomorphism.
The following lemma will be useful later.
L EMMA 10.6.23. The closed subgroups of a profinite group are exactly those that may be
written as intersections of open subgroups.
P ROOF. In a topological group, an open subgroup is also closed, an arbitrary intersection of
closed sets is closed, and an arbitrary intersection of subgroups is a subgroup, so an intersection
of open subgroups is a closed subgroup. Let U denote the set of open subgroups of a profinite
group G. Let H be a closed subgroup of G. It follows from Proposition 10.6.21b and the second
isomorphism theorem that the set of subgroups of the norm NH with N open normal in G has
intersection H. Note that each NH is open as a union of open subgroups, so it is open. 
We may also speak of pro-p groups.
D EFINITION 10.6.24. A pro-p group, for a prime p, is an inverse limit of a system of finite
p-groups.
We may also speak of profinite and pro-p completions of groups.
D EFINITION 10.6.25. Let G be a group.
a. The profinite completion Ĝ of G is the inverse limit of its finite quotients G/N, for N a
normal subgroup of finite index in G, together with the natural quotient maps G/N → G/N 0 for
N ≤ N 0.
b. The pro-p completion G(p) of G, for a prime p, is the inverse limit of the finite quotients
of G of p-power order, i.e., of the G/N for N P G with [G : N] a power of p, together with the
natural quotient maps.
R EMARK 10.6.26. A group G is endowed with a canonical homomorphism to its profinite
completion Ĝ by the universal property of the inverse limit.
R EMARK 10.6.27. We may also speak of topological rings and fields, where multiplication,
addition, and the additive inverse map are continuous, and in the case of a topological field, the
multiplicative inverse map on the multiplicative group is continuous as well. We may speak of
322 10. TOPICS IN GALOIS THEORY

profinite rings as inverse limits by quotients by two-sided ideals of finite index (or for pro-p
rings, of p-power index).
The next proposition shows that Z p is the pro-p completion of Z.
P ROPOSITION 10.6.28. Let p be a prime. We have an isomorphism of rings
!
∞ k−1
∼
→ lim Z/pk Z,
ψ : Zp − ∑ ai pi 7→ ∑ ai pi ,
←−
k≥1 i=0 i=0 k
where the maps Z/pk+1 Z → Z/pk Z in the system are the natural quotient maps. Moreover, ψ is
a homeomorphism.
P ROOF. The canonical quotient map ψk : Z p → Z/pk Z is the kth coordinate of ψ, which is
then a ring homomorphism by the universal property of the inverse limit. The kernel ψ is the
intersection of the kernels of the maps ψk , which is exactly
\
pk Z p = 0.
k
Moreover, any sequence of partial sums modulo increasing powers of p has a limit in Z p , which
maps to the sequence under ψ. The open neighborhood pn Z p of 0 in the p-adic topology is sent
to the intersection ! !
n ∞
∏ {0} × ∏ Z p /pk Z p ∩ lim Z/pk Z ,
←−
k=1 k=n+1 k≥1
which is open in the product topology. On the other hand, the inverse image of a basis open
neighborhood ! !
n ∞
∏ Uk × ∏ Z p /pk Z p ∩ lim Z/pk Z
←−
k=1 k=n+1 k≥1
with 0 ∈ Uk for all 1 ≤ k ≤ n under ψ clearly contains n
p Z p . It then follows from Lemma 10.6.6
that ψ is a homeomorphism. 
D EFINITION 10.6.29. The Prüfer ring Ẑ is the profinite completion of Z. That is, we have
Z∼= lim Z/nZ
←−
n≥1
with respect to the quotient maps Z/nZ → Z/mZ for m | n.
Since Z/nZ may be written as a direct product of the Z/pk Z for primes p with pk exactly
dividing n, we have the following.
L EMMA 10.6.30. We have an isomorphism of topological rings
Ẑ ∼
= Z p. ∏
p prime

E XAMPLE 10.6.31. The free profinite (or pro-p) group on a generating set S is the profinite
(resp., pro-p) completion of the free group on S.
 10.7. INFINITE GALOIS THEORY 323

R EMARK 10.6.32. As with free groups, closed subgroups of free profinite (or pro-p) groups
are free profinite (or pro-p) groups. Moreover, every profinite (resp., pro-p) group is a topological
quotient of the free group on a set of its generators, so we may present such groups via generators
and relations much as before.
D EFINITION 10.6.33. A subset S of a topological group G is said to be a topological gener-
ating set of G if G is the closure of the subgroup generated by S.
D EFINITION 10.6.34. We say that a topological group is (topologically) finitely generated if
it has a finite set of topological generators.
R EMARK 10.6.35. If G is a free profinite (or pro-p) group on a set S, then it is topologically
generated by S.
We leave a proof of the following to the reader.
L EMMA 10.6.36. Let G be a topological group, and let H be a (normal) subgroup. Then the
closure H of H is also a (normal) subgroup of G.

10.7. Infinite Galois theory

Recall that an algebraic extension of fields L/K is Galois if it is normal, so that every polyno-
mial in K[x] that has a root in L splits completely, and separable, so that no irreducible polynomial
in K[x] has a double root in L. The Galois group Gal(L/K) of such an extension is the group of
automorphisms of L that fix K.
In the setting of finite Galois extensions L/K, the subfields E of L containing F are in one-
to-one correspondence with the subgroups H of Gal(L/K). In fact, the maps E 7→ Gal(L/E) and
H 7→ LH give inverse bijections between these sets. This is not so in the setting of infinite Galois
extensions, where there are rather more subgroups than there are subfields. To fix this issue, we
place a topology on Gal(L/K) and consider only the closed subgroups under this topology. The
above-described correspondences then work exactly as before.
P ROPOSITION 10.7.1. Let L/K be a Galois extension of fields. Let E denote the set of finite
Galois extensions of K contained in L, ordered by inclusion. This is a directed set. Let ρ be the
map
ρ : Gal(L/K) → lim Gal(E/K)
←−
E∈E
defined by the universal property of the inverse limit, with the maps Gal(E 0 /K) → Gal(E/K) for
E, E 0 ∈ E with E ⊆ E 0 and the maps Gal(L/K) → Gal(E/K) for E ∈ E being restriction maps.
Then ρ is an isomorphism.
P ROOF. Let σ ∈ Gal(L/K). If σ |E = 1 for all E ∈ E , then since
[
L= E,
E∈E
we have that σ = 1. On the other hand, if elements σE ∈ Gal(E/K) for each E ∈ E are compatible
under restriction, then define σ ∈ Gal(L/K) by σ (α) = σE (α) if α ∈ E. Then, if α ∈ E 0 for some
324 10. TOPICS IN GALOIS THEORY

E 0 ∈ E as well, then
σE 0 (α) = σE∩E 0 (α) = σE (α),
noting that E ∩ E 0 ∈ E . Therefore, σ is well-defined, and so ρ is bijective. 
Proposition 10.7.1 gives us an obvious topology to place on the Galois group of a Galois
extension.
D EFINITION 10.7.2. Let L/K be a Galois extension of fields. The Krull topology on Gal(L/K)
is the unique topology under which the set of Gal(L/E) for E/K finite Galois with E ⊆ L forms
a basis of open neighborhoods of 1.
R EMARK 10.7.3. The Krull topology agrees with the inverse limit topology induced by the
isomorphism of Proposition 10.7.1, since
1 → Gal(L/E) → Gal(L/K) → Gal(E/K) → 1
is exact. Therefore, if L/K is Galois, then Gal(L/K) is a topological group under the Krull
topology.
L EMMA 10.7.4. Let L/K be a Galois extension of fields. The open subgroups in Gal(L/K)
are exactly those subgroups of the form Gal(L/E) with E an intermediate field in L/K of finite
degree over K.
P ROOF. First, let E be an intermediate field in L/K of finite degree. Let E 0 be the Galois
closure of E in L, which is of finite degree over K. Then Gal(L/E 0 ) is an open normal sub-
group under the Krull topology, contained in Gal(L/E). Since Gal(L/E) is then a union of left
Gal(L/E 0 )-cosets, which are open, we have that Gal(L/E) is open.
Conversely, let H be an open subgroup in Gal(L/K). Then H contains Gal(L/E) for some
finite Galois extension E/K in L. Any α ∈ LH , where LH is the fixed field of H in L, is contained
in M Gal(L/E) , where M is the Galois closure of K(α). Since the restriction map Gal(L/E) →
Gal(M/E) is surjective, we then have α ∈ M Gal(M/E) . But M/K is finite, so M Gal(M/E) = E by
the fundamental theorem of Galois theory. Thus LH ⊆ E.
Let H̄ be the image of H under the restriction map π : Gal(L/K) → Gal(E/K). As Gal(L/E) ≤
H, we have that π −1 (H̄) = H. We remark that H̄ = Gal(E/LH ), since H̄ = Gal(E/E H̄ ) by the
fundamental theorem of Galois theory for finite extensions and LH = E H = E H̄ . But π −1 (H̄) is
then Gal(L/LH ) as well. 
From this, we may derive the following.
L EMMA 10.7.5. Let L/K be a Galois extension of fields. The closed subgroups of Gal(L/K)
are exactly those of the form Gal(L/E) for some intermediate field E in the extension L/K.
P ROOF. Under the Krull topology on Gal(L/K), the open subgroups are those of the form
Gal(L/E) with E/K finite. By Lemma 10.6.23, we have therefore that the closed subgroups are
those that are intersections of Gal(L/E) over a set S of finite degree over K intermediate fields
E. Any such intersection necessarily fixes the compositum E 0 = ∏E∈S E, while if an element
 10.7. INFINITE GALOIS THEORY 325

of Gal(L/K) fixes E 0 , then it fixes every E ∈ S, so lies in the intersection. That is, any closed
subgroup has the form
Gal(L/E 0 ) =
\
Gal(L/E).
E∈S

T HEOREM 10.7.6 (Fundamental theorem of Galois theory). Let L/K be a Galois extension.
Then there are inverse one-to-one, inclusion reversing correspondences

{intermediate extensions in L/K} o

ψ
/ {closed subgroups of Gal(L/K)}
θ

given by ψ(E) = Gal(L/E) for any intermediate extension E in L/K and θ (H) = LH for any
closed subgroup H of Gal(L/K). These correspondences restrict to bijections between the nor-
mal extensions of K in L and the closed normal subgroups of Gal(L/K), as well as to bijections
between the finite degree (normal) extensions of K in L and the open (normal) subgroups of
Gal(L/K). For any E of finite degree and the corresponding closed of finite index H, we have
[L : E] = Gal(L/E) and |H| = [L : LH ].
Moreover, if E is normal over K (resp., H P Gal(L/K) is closed), then restriction induces a
topological isomorphism
∼
Gal(L/K)/ Gal(L/E) −→ Gal(E/K)
∼
→ Gal(LH /K)).
(resp., Gal(L/K)/H −
P ROOF. We will derive this from the fundamental theorem of Galois theory for finite Galois
extensions. Let E be an intermediate extension in L/K. Then E ⊆ LGal(L/E) by definition. Let
x ∈ LGal(L/E) . The Galois closure M of E(x) in L is of finite degree over E. But every element of
Gal(M/E) extends to an element of Gal(L/E), which fixes x. So x ∈ M Gal(M/E) , which equals E
by fundamental theorem of Galois theory for finite Galois extensions. Since x was arbitrary, we
have E = LGal(L/E) . In other words, θ (ψ(E)) = E.
Let H be a closed subgroup of Gal(L/K). In Lemma 10.7.5, we saw that H = Gal(L/E) for
some intermediate E in L/K. Since E = LGal(L/E) = LH from what we have shown, we have
that H = Gal(L/LH ). Therefore, ψ(θ (H)) = H. It follows that we have the desired inclusion-
reserving one-to-one correspondences. The other claims are then easily checked, or follow from
the case of finite degree, and are left to the reader. 
D EFINITION 10.7.7. A separable closure of a field L is any field that contains all roots of all
separable polynomials in L.
N OTATION 10.7.8. We typically denote a separable closure of L by Lsep .
R EMARK 10.7.9. If one fixes an algebraically closed field Ω containing L, then there is a
unique separable closure of L in Ω, being the subfield generated by the roots of all separable
polynomials in L[x].
326 10. TOPICS IN GALOIS THEORY

D EFINITION 10.7.10. The absolute Galois group of a field K is the Galois group
GK = Gal(K sep /K),
where K sep is a separable closure of K.
R EMARK 10.7.11. The absolute Galois group, despite the word “the”, is not unique, but
rather depends on the choice of separable closure. An isomorphism of separable closures gives
rise to a canonical isomorphism of absolute Galois groups, however.
E XAMPLE 10.7.12. Let q be a power of a prime number. Then there is a unique topological
∼
isomorphism GFq − → Ẑ sending the Frobenius automorphism ϕq : x 7→ xq to 1. To see this, note
that Gal(Fqn /Fq ) → Z/nZ given by sending ϕq to 1 is an isomorphism, and these give rise to
compatible isomorphisms in the inverse limit
∼ ∼ ∼
GFq −
→ lim Gal(Fqn /Fq ) −
→ lim Z/nZ −
→ Ẑ.
←− ←−
n n

E XAMPLE 10.7.13. Let Q(µ p∞ ) denote the field given by adjoining all p-power roots of unity
to Q. Then
Gal(Q(µ p∞ )/Q) ∼= lim Gal(Q(µ pn )/Q) ∼ = lim (Z/pn Z)× ∼ = Z×p
←− ←−
n n
the middle isomorphisms arising from the pn th cyclotomic characters.
T ERMINOLOGY 10.7.14. The isomorphism Gal(Q(µ p∞ )/Q) → Z×
p of Example 10.7.13 called
the p-adic cyclotomic character.
Since the compositum of two abelian extensions of a field inside a fixed algebraic closure is
abelian, the following makes sense.
N OTATION 10.7.15. Let K be a field. The maximal abelian extension of K inside an algebraic
closure of K is denoted K ab .
R EMARK 10.7.16. The abelianization Gab
K of the absolute Galois group GK of a field K canon-
ab
ically isomorphic to Gal(K /K) via the map induced by restriction on GK .
 CHAPTER 11

Commutative algebra

In this chapter, all rings are commutative unless otherwise stated.

11.1. Localization of modules

We now turn to localization of modules.
N OTATION 11.1.1. Let M be a module over a commutative ring R, and let S be a multiplica-
tively closed subset of R. The set of equivalence classes of S × M under ∼S is denoted S−1 M,
and the equivalence class of (s, m) is denoted s−1 m or ms . We write m1 more simply as m.
L EMMA 11.1.2. Let M be a module over a commutative ring R, and let S be a multiplicatively
closed subset of R. Then the relation ∼S on S × M defined by (s, m) ∼S (t, n) if there exists r ∈ S
such that r(sn − tm) = 0 is an equivalence relation.
P ROOF. The relation ∼S is clearly reflexive and symmetric, so we only need to check transi-
tivity. For this, let (s, m) ∼S (s0 , m0 ) and (s0 , m0 ) ∼S (s00 , m00 ) in S × M. Then there exist r, r0 ∈ S
such that r(sm0 − s0 m) = r0 (s0 m00 − s00 m0 ) = 0. We then have
0 = rr0 s00 (sm0 − s0 m) + rr0 s(s0 m00 − s00 m0 ) = rr0 s0 (sm00 − s00 m),
so (s, m) ∼S (s00 , m00 ). 
We omit the easy but nonetheless tedious proof of the following.
P ROPOSITION 11.1.3. Let M be a module over a commutative ring R, and let S be a multi-
plicatively closed subset of R. The set S−1 M of equivalence classes of S × M under the equiva-
lence relation ∼S is an S−1 R-module under the operations
m n tm + ns a m am
+ = and · =
s t st s t st
for a ∈ R, m, n ∈ M, and s,t ∈ S. There is a canonical map ι : M → S−1 M of R-modules given by
ι(m) = m1 .
E XAMPLE 11.1.4. Let S be a multiplicatively closed subset of a commutative ring R. Then
the localization S−1 R of R viewed as a left R-module is just the ring S−1 R viewed as a module
over itself.
E XAMPLE 11.1.5. Let R be an integral domain and S = R − {0}. If M is an R-module, then
S−1 M is a Q(R)-vector space.
327
328 11. COMMUTATIVE ALGEBRA

L EMMA 11.1.6. Let S be a multiplicatively closed subset of a commutative ring R. Let {Mi |
i ∈ I} be a collection of R-modules. Then
M 
S−1 Mi −∼ S−1 Mi
M
→
i∈I i∈I

via the canonical map that takes s−1 (mi )i∈I to (s−1 mi )i∈I .
E XAMPLE 11.1.7. Let p be a prime number, and let S p be the multiplicatively closed subset
of Z that is the complement of the prime ideal (p). For n ≥ 1, the localization S−1p (Z/nZ) is
k k
isomorphic to Z/p Z, where p is the highest power of p dividing n.
To see this, note that Z/nZ ∼
= Z/mZ × Z/pk Z, where n = pk m, so we have
S−1 −1 −1 k
p (Z/nZ) = S p (Z/mZ) × S p (Z/p Z).

Now, for any x ∈ Z/mZ, we have x = m−1 (mx) = m−1 0 = 0 in S−1 p (Z/mZ). It follows that
−1
S p (Z/mZ) = 0. On the other hand, if y ∈ Z/p Z and a ∈ S p are such that a−1 y = 0, then there
k

exists b ∈ S p such that by = 0, which means that y = 0 since b is prime to p. Furthermore, any
a−1 y ∈ S−1 k k k × −1
p (Z/p Z) is in the image of Z/p Z as a ∈ (Z/p Z) . Thus, we have S p (Z/p Z) =
k

Z/pk Z.
R EMARK 11.1.8. Given a commutative ring R and a multiplicatively closed set S, localiza-
tion provides a functor S−1 : R-mod → S−1 R-mod. That is, if f : M → N is an R-modules homo-
morphism, then we have an induced R-module homomorphism S−1 f : S−1 M → S−1 N given by
f (s−1 m) = s−1 f (m).
We may describe the localization of a module over a commutative ring as a tensor product,
as follows.
P ROPOSITION 11.1.9. Let M be a module over a commutative ring R, and let S be a multi-
plicatively closed subset of R. Then S−1 M ∼
= S−1 R ⊗R M as S−1 R-modules.
P ROOF. Define a map θ : S−1 R × M → S−1 M by θ (s−1 a, m) = am s . To see that it is well-
a b
defined, note that if s = t , then we have r ∈ S with r(ta − sb) = 0, and then r(tam − sbm) = 0, so
am bm −1
s = t . The map θ is easily checked to be left S R-linear, right R-linear, and R-balanced. We
then obtain a map of S−1 R-modules Θ : S−1 R ⊗R M → S−1 M satisfying Θ( as ⊗ m) = am s by the
−1
universal property of the tensor product. (That it is an S R-module homomorphism, rather than
just an R-module homomorphism, follows directly from the left S−1 R-linearity.) For bijectivity,
it suffices to exhibit an inverse function.
Define a function ψ : S × M → S−1 R ⊗R M by ψ(s, m) = s−1 ⊗ m. If (s, m) ∼S (t, n), then let
r ∈ S be such that r(sn − tm) = 0. We then have
s−1 ⊗ m = (rst)−1 ⊗ rtm = (rst)−1 ⊗ rsn = t −1 ⊗ n,
so we obtain a well-defined map Ψ : S−1 M → S−1 R ⊗ M given by Ψ( ms ) = s−1 ⊗ m. (In fact, Ψ is
a homomorphism of S−1 R-modules, but it is not necessary to check this to finish the proof, since
 11.1. LOCALIZATION OF MODULES 329

the inverse of a module isomorphism is one as well.) By definition, Θ(Ψ( ms )) = ms , and we have
Ψ(Θ(s−1 a ⊗ m)) = Ψ( am
s )=s
−1
⊗ am = s−1 a ⊗ m.

N OTATION 11.1.10. Let p be a prime ideal of a commutative ring R, and let M be an R-
module. Then the localization of the Rp -module Sp−1 M is denoted Mp .
P ROPOSITION 11.1.11. Let R be a commutative ring and M be an R-module. Then the
following are equivalent:
i. M = 0,
ii. Mp = 0 for every prime ideal p of R, and
iii. Mm = 0 for every maximal ideal m of R.
P ROOF. Clearly, (i) implies (ii) and (ii) implies (iii). Let m ∈ M be nonzero. Let I be anni-
hilator of m in R, which is to say I = Ann(Rm). Then I is a proper ideal, hence contained in a
maximal ideal m of R. If ur ∈ Rm annihilates m, then rsm = 0 for some s ∈ R − m. Thus rs ∈ m,
so r ∈ m as m is prime. This implies that the annihilator of m in Rm in a proper ideal, so m is
nonzero in Mm . Thus, we have the contrapositive to (iii) implies (i). 
P ROPOSITION 11.1.12. Let R be a commutative ring and f : M → N a homomorphism of
R-modules. Then the following are equivalent:
i. f is injective,
ii. the maps fp : Mp → Np induced by f are injective for all prime ideals p of R, and
iii. the maps fm : Mm → Nm are injective for all maximal ideals m of R.
We have the following.
P ROPOSITION 11.1.13. Let m be a maximal ideal of a commutative ring R. For n ≥ 0, then
canonical map
mn /mn+1 → mn Rm /mn+1 Rm
is an isomorphism of R/m-vector spaces.
P ROOF. Note that Rm ⊗R mn ∼ = mn Rm by Lemma 11.1.9, the isomorphism being induced by
the R-linear map ( as , b) 7→ ab
s . Therefore, we have an isomorphism
∼
Rm ⊗R mn /mn+1 −
→ mn Rm /mn+1 Rm
by right exactness of tensor product with Rm .
Now consider the composite isomorphism
∼
→ mn /mn+1 → mn Rm /mn+1 Rm
R/m ⊗R mn /mn+1 −
∼ ∼ ∼
→ Rm ⊗R mn /mn+1 −
− → Rm /mRm ⊗R mn /mn+1 −
→ R/m ⊗R mn /mn+1 ,
where in the first and last steps we have used that the maximal ideal annihilates the successive
quotients of powers of the maximal ideal in both the case of R and Rm and in the last step we have
330 11. COMMUTATIVE ALGEBRA

used the n = 0 case found in Lemma 8.1.31. An element of the form 1 ⊗ x with x ∈ mn /mn+1 is
sent under the composition to 1 ⊗ x. The composite is also easily seen to be a map of R/m-vector
spaces, so it is the identity being that it is the identity on simple tensors. In particular, the map
mn /mn+1 → mn Rm /mn+1 Rm is an isomorphism. 
11.2. Radicals of ideals
Let R be a commutative ring.
√
D EFINITION 11.2.1. The radical I of an ideal I of R is the set
√
I = {a ∈ R | ak ∈ I for some k ≥ 1}.
√
L EMMA 11.2.2. For any ideal I of R, the radical I of I is an ideal of R.
√ k k k k
√ . If r ∈ R and a ∈ I, then there
P ROOF √ existsl k ≥ 1 with a ∈ I, and then (ra) = r a ∈ I,
so ra ∈ I as well. If we also have b ∈ l with b ∈ I, then
k+l  
k+l k + l i k+l−i
(a + b) = ∑ ab ∈I
i=0 i
√
since either i ≥ k or k + l − i ≥ l if i ∈ Z. Thus a + b ∈ I. 
The nilradical of R is the radical of the ideal (0) of R.
D EFINITION 11.2.3. The nilradical of R is the ideal of nilpotent elements in R.
E XAMPLE 11.2.4. The nilradical of F[x]/(xn ) for a field F and n ≥ 2 is generated by x.
D EFINITION 11.2.5. A commutative ring is reduced if it has no nonzero nilpotent elements.
E XAMPLES 11.2.6.
a. Domains are reduced.
b. The quotient of a commutative ring by its nilradical is reduced.
c. The ring Z/6Z is reduced, though it is not a domain.
In fact, the following is easily verified.

√L EMMA 11.2.7. If π : R → R/I is the projection of R onto its quotient by an ideal I, then
π( I) is the nilradical of R/I.
D EFINITION 11.2.8. An ideal is radical, or a semiprime ideal, if it is its own radical.
E XAMPLES 11.2.9.
a. Prime ideals are radical. That is, if an ∈ p for some n ≥ 1, then a ∈ p by primality.
b. Let F be a field, f1 , . . . , fr ∈ F[x] be irreducible, and k1 , . . . , kr be positive integers. Then
q
( f1k1 · · · frkr ) = ( f1 · · · fr ).
Thus, the nonzero radical ideals of F[x] are exactly the ideals generated by products of distinct
irreducible elements.
 11.3. PRIMARY DECOMPOSITION 331

c. Radicals of ideals are radical.

√
P ROPOSITION 11.2.10. Let I be a proper ideal of R. Then I is the intersection of all prime
ideals of R containing I.
√ √ √
P ROOF. If p is a prime ideal containing I, then p = p ⊇ I. Thus, I is contained in the
intersection of all prime ideals
√ J containingnI.
Suppose that a ∈ J − I. Then S = {a | n ≥ 0} is a multiplicative set disjoint from I. By
Theorem 8.1.33, there exists a prime ideal p containing I and disjoint √
from S. Then a cannot be
in the intersection J, which is a contradiction. Thus, we have that J = I. 

√ n 11.2.11. Let R be noetherian and I be an ideal of R. Then there exists n ≥ 1

P ROPOSITION
such that ( I) ⊆ I.
√
P ROOF. Let a1 , . . . , am ∈ R be such that I = (a1 , . . . , am ). For 1 √
≤ i ≤ m, let ki ≥ 1 be such
that aki i ∈ I, and let k = max{ki | 1 ≤ i ≤ m}. For any x = ∑m r a
i=1 i i ∈ I, we have
xkm ∈ ({ai11 · · · aimm | i j ≥ 0 for all j with i1 + · · · + im = km}) ⊆ (ak1 , . . . , akm ) ⊆ I.
Thus, we may take n = km in the statement. 
D EFINITION 11.2.12. An ideal I of R is nilpotent if there exists n ≥ 1 such that I n = 0.
C OROLLARY 11.2.13. The nilradical of a noetherian commutative ring is nilpotent.
It is easy to see why this can fail in a noncommutative ring.
E XAMPLE 11.2.14. Consider the polynomial ring R = F[x√ 1 , x2 , . . .] in countably many vari-
k
ables over a field F and its ideal I = (xk | k ≥ 1). Its radical is I = (xk | k ≥ 1) but no power of
√
I is contained in I.
The taking of radicals behaves well with respect to localization.
L EMMA
√ 11.2.15.
√ Let S be a multiplicatively closed subset of R, and let I be an ideal of R.
Then S−1 I = S−1 I.

11.3. Primary decomposition

D EFINITION 11.3.1. A proper ideal q of R is primary if for any a, b ∈ R with ab ∈ q, one has
either a ∈ q or bn ∈ q for some n ≥ 1.
√
That is, an ideal q is primary if whenever ab ∈ q, either a ∈ q or b ∈ q. Of course, prime
ideals are primary. The following is just a rephrasing of the definition of primary.
L EMMA 11.3.2. A proper ideal q of R is primary if and only if every zero divisor in R/q is
nilpotent.
The following is a key property of primary ideals.
P ROPOSITION 11.3.3. The radical of any primary ideal is a prime ideal.
332 11. COMMUTATIVE ALGEBRA
√
P ROOF. Let q be a primary ideal of R. If a, b ∈ R with ab ∈ q, then ak bk ∈ q for some k ≥ 1,
√
and therefore either ak ∈ q or there exists n ≥ 1 such that bkn ∈ q. In the first, case a ∈ q, and
√ √
in the second, b ∈ q, so q is prime. 
In particular, the radical of a primary ideal q is the smallest prime ideal of R containing q,
given that it is also the intersection of all prime ideals containing q.
D EFINITION 11.3.4. The radical p of a primary ideal q of R is called the associated prime to
q, and we say that q is p-primary.
E XAMPLES 11.3.5. Let F be a field.
a. The ideal (x2 , y) of F[x, y] is primary since F[x, y]/(x2 , y) ∼ = F[x]/(x2 ), and every zero
divisor in the latter ring is nilpotent. Its associated prime is (x, y).
b. Consider R = F[x, y, z]/(xy − zp 2 ) and its ideal p = (x, z), which is prime since R/p ∼ F[y].
=
We have xy ∈ p2 , but x ∈ / p2 and y ∈/ p2 = p. Thus p2 is not primary, even though p is prime.
√
L EMMA 11.3.6. If I is an ideal of R such that I is maximal, then I is primary. In particular,
any power of a maximal ideal m is primary with associated prime m.
√
P ROOF. Suppose that m = I is maximal. The image of m in R/I is the nilradical of R/I,
which means that the nilradical is the only prime ideal of R/I. In particular, R/I is local, and
every element of R/I that is not nilpotent is a unit. In particular, every zero divisor of R/I is
nilpotent. Thus, I is m-primary. 
The following is easily checked.
L EMMA 11.3.7. A finite intersection of primary ideals with the same associated prime is
primary.
D EFINITION 11.3.8. Let I be an ideal of R.
a. A primary decomposition of I is a finite collection {q1 , . . . , qk } of primary ideals of R such
that I = ki=1 qi .
T

b. We say that I is decomposable if it has a primary decomposition.

√
c. A primary decomposition {q1 , . . . , qk } of I is minimal if the radicals qi are all distinct
and no proper subset of the primary decomposition is also a primary decomposition of I.
Every ideal with a primary decomposition has a minimal such decomposition.
L EMMA 11.3.9. Let I be a decomposable ideal of R. Then I has a minimal primary decom-
position.
P ROOF. From this decomposition, we may first remove one at a time any primary ideals that
contain the intersection of the others. By Lemma 11.3.7, we may then replace the subcollection
of those ideals in the decomposition with the same associated prime by the single primary ideal
that is its intersection. The resulting collection is minimal. 
 11.3. PRIMARY DECOMPOSITION 333

E XAMPLE 11.3.10. Let F be a field. The ideal (xy2 ) of F[x, y] has a minimal primary de-
composition (xy2 ) = (x) ∩ (y2 ), and the associated primes of these primary ideals are (x) and
(y).
D EFINITION 11.3.11. A proper ideal I of R is irreducible if for any ideals a and b of R with
I = a ∩ b, either I = a or I = b.
P ROPOSITION 11.3.12. Let R be noetherian. Then every irreducible ideal of R is primary.
P ROOF. Let I be an irreducible ideal of R, and let a, b ∈ R with ab ∈ I but b ∈ / I. For each
n ≥ 1, let Jn = {r ∈ R | an r ∈ I}, and note that Jn is an ideal of R. Then Jn form an ascending chain
of ideals containing I, and since R is noetherian, this chain is eventually constant, say Jn = Jn+1
for all n ≥ N with N ≥ 1. Consider the ideals a = (aN ) + I and b = (b) + I containing I. We
claim that a ∩ b = I. Let c ∈ a ∩ b. Then c = aN r + q for some r ∈ R and q ∈ I. Since c ∈ (b) + I,
we have ac ∈ (ab) + I = I. In other words, aN+1 r + qa ∈ I, so aN+1 r ∈ I, so r ∈ JN+1 = JN .
Therefore aN r ∈ I as well, so c ∈ I, and the claim holds. Since I is irreducible and b ∈ / I, we must
have I = a, which means that aN ∈ I. Therefore, I is primary. 
E XAMPLES 11.3.13. Let F be a field, and consider the ring F[x, y].
The ideal (x, y)2 is (x, y)-primary, but (x, y)2 = (x2 , y) ∩ (x, y2 ), so (x, y)2 is not irreducible.
Thus, the converse to Proposition 11.3.12 does not hold.
The ideal (x2 , y) is irreducible, and therefore primary, but it is not prime.
P ROPOSITION 11.3.14. Let R be noetherian. The every proper ideal of R is a finite intersec-
tion of irreducible ideals.
P ROOF. Let X be the set of proper ideals of R that cannot be written as a finite intersection
of irreducible ideals of R. Since R is noetherian, either X is empty or X has a maximal element
m. Since m ∈ X, it is not irreducible, so there exist ideals a and b properly containing m with
m = a ∩ b. Since m is maximal in X, both of a and b can be written as a finite intersection of
irreducible ideals, so m may be as well, which contradicts the existence of m. Therefore X is
empty, as desired. 
Combining Propositions 11.3.12 and 11.3.14, we have the following.
T HEOREM 11.3.15 (Primary decomposition theorem). Every proper ideal of a noetherian
commutative ring R is decomposable.
D EFINITION 11.3.16. For ideals I and J of a commutative ring R, the ideal quotient of I by
J is (I : J) = {r ∈ R | rJ ⊆ I}.
N OTATION 11.3.17. For an ideal I of R and a ∈ I, we set Ia = (I : (a)).
Note that ideal quotients are ideals.
L EMMA 11.3.18. Let q be a p-primary ideal of R. For a ∈ R, we have qa = R if a ∈ q and qa
is p-primary otherwise.
334 11. COMMUTATIVE ALGEBRA

P ROOF. If a ∈ q, then 1 ∈ qa , so we are done. If a 6∈ q, let b, c ∈ R with bc ∈ qa and b ∈

/ qa .
√
Then abc ∈ q but ab ∈/ q, so b ∈
/ q, and c ∈ q as q is primary. Hence qa is primary. Moreover,
we have √ √
qa = {r ∈ R | rn a ∈ q for some n ≥ 1} = q = p.

C OROLLARY 11.3.19. Let I be a decomposable ideal of R. Let {q1 , . . . , qn } be a minimal
√
primary decomposition of I, and let pi = qi for each i. For a ∈ R, we have
√ n
\
Ia = pi .
i=1
a∈q
/ i
√
In particular, for each i, there exists an a ∈ R such that Ia = pi .
Tn
P ROOF. For a ∈ R, we have Ia = i=1 (qi )a . By Lemma 11.3.18, we have
√ \n p n
\
Ia = (qi )a = pi .
i=1 i=1
a∈q
/ i
We may choose a ∈ q j for
√ j 6= i such that a ∈
/ qi by the minimality of the decomposition. For
such an a, we then have Ia = pi . 
In fact, for noetherian R, we have the following refinement of the last statement of Corol-
lary 11.3.19.
L EMMA 11.3.20. Suppose that R is noetherian, and let I be a proper ideal of R. Let p be an
associated prime of I. Then there exists a ∈ R such that Ia = pi .
P ROOF. Let q be a p-primary ideal in a minimal primary decomposition of I, and let J be
the intersection of the other primary ideals set decomposition. √ Consider the nonempty set X =
{Ia | a ∈ J, a ∈
/ q}, and note that by Corollary 11.3.19, we have Ia = p for each Ia ∈ X. As R is
noetherian, the set X contains a maximal element Ib for some b ∈ J, b ∈ / q. It will suffice to show
that Ib is prime, since in that case Ib is radical with radical p.
Now suppose that c, d ∈ R are such that cd ∈ Ib but c ∈ / Ib , which is to say bcd ∈ I but bc ∈ / I.
We know that bc ∈ J since b ∈ J, so bc ∈ / q. But then Ibc ∈ X contains the maximal element Ib ,
so Ibc = Ib , and therefore d ∈ Ib . That is, Ib is prime. 
This lemma has the following interesting consequence.
P ROPOSITION 11.3.21. Let R be noetherian, and let I be proper. An element x ∈ R reduces
to a zero divisor in R/I if and only if it is contained in an associated prime of I.
P ROOF. Let Z be the set of elements of R that reduce to zero divisors modulo I. Then Z =
a∈R−I Ia . Let p be an associated primes of I. By Lemma 11.3.20, we may choose b ∈ R with
S

Ib = p Since b ∈/ I as Ib 6= R, have p ⊆ Z.
Next take r ∈ Z, and define Xr = {Ia | a ∈/ I, r ∈ Ia }, which is nonempty by definition of Z.
Since R is noetherian, it has a maximal element Ib for some b ∈ / I. It suffices to show that Ib is
 11.3. PRIMARY DECOMPOSITION 335

prime, as it contains r. For this, let c, d ∈ R with cd ∈ Ib but c ∈ / Ib . Then bcd ∈ I but bc ∈ / I.
Since r ∈ Ibc , we have Ibc ∈ Xr , and it contains Ib , so Ibc = Ib by maximality of Ib . As d ∈ Ibc , we
then have d ∈ Ib . Thus, Ib is prime. 
We now consider uniqueness of primary decompositions, given the existence of one.
T HEOREM 11.3.22. Let I be a decomposable ideal of R. The set of associated primes to the
primary ideals in a minimal primary decomposition of I is uniquely determined by I.
√
P ROOF. Let {q1 , . . . , qn } be a minimal primary√decomposition of I, with pi = qi for each i.
By Corollary 11.3.19,
√ we may choose a such that Ia = pi for any i. On the other hand,√for any
a ∈ R such that Ia is prime, part b of Lemma 3.10.19 and Corollary 11.3.19 tell us that Ia = pi
for some i (with a ∈
/ qi ). Thus, the set of associated primes {p1 , . . . , pn } is uniquely determined
by I. 
D EFINITION 11.3.23. Let I be a decomposable ideal of R. A prime ideal is called an associ-
ated prime of I if it is the associated prime of an element of a minimal primary decomposition of
I.
D EFINITION 11.3.24. Let I be a proper ideal of R. An isolated prime of I is a minimal
element in the set of prime ideals of R containing I, ordered by inclusion.
R EMARK 11.3.25. If I is a proper ideal of a noetherian ring R, then every prime ideal of R/I
is contained in a minimal prime, and the minimal primes of R/I are exactly the images of the
isolated primes of I.
P ROPOSITION 11.3.26. Let I be a decomposable ideal of R. A prime ideal of R is an isolated
prime of I if and only if it is a minimal element under inclusion in the set of associated primes of
I.
P ROOF. Let I = ni=1 qi with each qi primary, and let pi be the associated prime of qi . If p is
T

a prime ideal of R containing I, then

√ n n
√ \ √ \
p= p⊇ I= qi = pi .
i=1 i=1
By part b of Lemma 3.10.19, we have that p contains some pi for some i, so it contains some
minimal prime in the set of associated primes of I. 
R EMARK 11.3.27. If (0) is decomposable, then R has finitely many minimal primes. These
are exactly the isolated primes in a minimal primary decomposition of (0).
D EFINITION 11.3.28. Let I be a decomposable ideal of R. An embedded prime of I is an
associated prime of I that is not isolated.
E XAMPLE 11.3.29. Let F be a field. The ideal I = (xy, y2 ) of F[x, y] has a minimal primary
decomposition I = (x, y)2 ∩ (y), so it has associated primes (x, y) and (y). The ideal (y) is the
unique isolated prime of I. Note that I also has the primary decomposition I = (x, y2 ) ∩ (y).
In fact, the following uniqueness result also holds.
336 11. COMMUTATIVE ALGEBRA

P ROPOSITION 11.3.30. Let I be a decomposable ideal of R. Let p1 , . . . , pn be distinct isolated

primes of I. Let Q be a minimal primary decomposition of I, and let qi ∈ Q be pi -primary for
each 1 ≤ i ≤ n. Then the ideal ni=1 qi is independent of the choice of Q.
T

P ROOF. Consider the multiplicative set S = R − ni=1 pi . For any associated prime ideal p of
S

I, the intersection S ∩ p is empty if and only if p = pi for some i by Proposition 8.1.20. Given a
primary decomposition Q, suppose that q ∈ Q is p-primary for some prime p. If p = pi for some
i, then S−1 p is a prime of S−1 R and S−1 qi is S−1 p-primary, and otherwise we have S−1 p = S−1 R.
We then have
n
S−1 I = S−1 q = S−1 qi ,
\ \

q∈Q i=1
and the contraction of S−1 I
to R is the intersection of the contractions of the primary ideals
S−1 qi . By Lemma 8.1.19(b), the contraction of S−1 qi is the ideal of a ∈ qi such that there exists
r ∈ S with ra ∈ qi . Since S is contained in the complement of pi , we have r ∈/ qi , which as qi is
primary forces such an a to lie in qi . In other words, ni=1 qi is the contraction of S−1 I, and this
T

is independent of the choice of Q. 

C OROLLARY 11.3.31. If p is any isolated prime of a decomposable ideal I, then the unique
p-primary ideal in any minimal primary decomposition of I is independent of the choice of de-
composition.
11.4. Integral extensions
D EFINITION 11.4.1. We say that B/A is an extension of commutative rings if A and B are
commutative rings such that A is a subring of B.
D EFINITION 11.4.2. Let B/A be an extension of commutative rings. We say that β ∈ B is
integral over A if β is the root of a monic polynomial in A[x].
E XAMPLES 11.4.3.
a. Every element a ∈ A is integral over A, in that a is a root of x − a.
b. If L/K is a field extension and α ∈ L is algebraic over K, then α is integral over K, being
a root of its minimal polynomial, which is monic.
c. If L/K is a field extension and α ∈ L is transcendental over K, then α is not integral over
K.
√ √
d. The element 2 of Q( 2) is integral over Z, as it is a root of x2 − 2.
√ √
e. The element α = 1−2 5 of Q( 5) is integral over Z, as it is a root of x2 − x − 1.
P ROPOSITION 11.4.4. Let B/A be an extension of commutative rings. For β ∈ B, the follow-
ing conditions are equivalent:
i. the element β is integral over A,
ii. there exists n ≥ 0 such that {1, β , . . . , β n } generates A[β ] as an A-module,
iii. the ring A[β ] is a finitely generated A-module, and
 11.4. INTEGRAL EXTENSIONS 337

iv. there exists a faithful A[β ]-submodule of B that is finitely generated over A.
P ROOF. Suppose that (i) holds. Then β is a root of a monic polynomial g ∈ A[x]. Given
any f ∈ A[x], the division algorithm tells us that f = qg + r with q, r ∈ A[x] and either r = 0
or deg r < deg g. It follows that f (β ) = r(β ), and therefore that f (β ) is in the A-submodule
generated by {1, β , . . . , β deg g−1 }, so (ii) holds. Condition (ii) is clearly at least as strong as (iii).
Suppose that (iii) holds. Then we may take the A[β ]-submodule A[β ] of B, which being free over
itself has trivial annihilator.
Finally, suppose that (iv) holds. Let
n
M = ∑ Aγi ⊆ B
i=1

be a faithful A[β ]-module, and suppose without loss of generality that β 6= 0. We have
n
βγj = ∑ ci j γi
j=1

for some ci j ∈ A with 1 ≤ i, j ≤ n. The characteristic polynomial c(x) ∈ A[x] of the matrix
C = (ci j ) is monic, and as β In −C takes (γ1 , . . . , γn ) ∈ M n to zero, c(β ) acts as zero on M. Since
M is a faithful A[β ]-module, we must have c(β ) = 0. Thus, β is integral. 
E XAMPLE 11.4.5. The element 12 ∈ Q is not integral over Z, as Z[1, 2−1 , . . . , 2−n ] for n ≥ 0
is equal to Z[2−n ], which does not contain 2−(n+1) .
D EFINITION 11.4.6. Let B/A be an extension of commutative rings. We say that B is an
integral extension of A if every element of B is integral over A.
√ √
E XAMPLE 11.4.7. The ring Z[ 2] is an integral extension of Z. Given α = a + b 2 with
a, b ∈ Z, note that α is a root of x2 − 2ax + a2 − 2b2 .
The integral extensions of a field are its algebraic field extensions.
L EMMA 11.4.8. Let B/A be an integral extension of domains. Then B is a field if and only if
A is a field.
P ROOF. Suppose that A is a field. Every b ∈ B ⊆ Q(B) is the root of a monic polynomial
with coefficients in A, so A(b) = A[b] ⊆ B. That is, b ∈ B× ∪ {0}, and thus B is a field.
Now suppose that B is a field. Then for any a ∈ A − {0}, the element a−1 is integral over A,
so there is a monic polynomial f ∈ A[x] with f (a−1 ) = 0. Write f (x) = xn + g(x) with deg g < n
for some n ≥ 1. Then a−1 = −an−1 g(a−1 ) ∈ A, so a ∈ A× . 
R EMARK 11.4.9. An extension of fields is integral if and only if it is algebraic.
L EMMA 11.4.10. Suppose that B/A is an extension of commutative rings such that B is
finitely generated as an A-module, and let M be a finitely generated B-module. Then M is a
finitely generated A-module.
338 11. COMMUTATIVE ALGEBRA

P ROOF. Let {m1 , . . . , mn } be a set of generators of M as a B-module, and let {β1 , . . . , βk } be

a set of generators of B as an A-module. We claim that {βi m j | 1 ≤ i ≤ k, 1 ≤ j ≤ n} is a set of
generators of M as an A-module. To see this, let m ∈ M and write
n
m= ∑ b jm j
j=1
with b j ∈ B for 1 ≤ j ≤ n. For 1 ≤ j ≤ n, we then write
k
b j = ∑ ai j β i
i=1
with ai j ∈ A for 1 ≤ i ≤ k. We then have
k n
m= ∑ ∑ ai j βim j ,
i=1 j=1
as desired. 
We now give a criterion for a finitely generated algebra over a ring to be finitely generated as
a module.
P ROPOSITION 11.4.11. Let B/A be an extension of commutative rings and suppose that
B = A[β1 , β2 , . . . , βk ]
for some k ≥ 0 and βi ∈ B with 1 ≤ i ≤ k. Then the following are equivalent:
i. the ring B is integral over A,
ii. each βi with 1 ≤ i ≤ k is integral over A, and
iii. the ring B is finitely generated as an A-module.
P ROOF. Clearly, (i) implies (ii), so suppose that (ii) holds. By definition, each βi is then
integral over any commutative ring containing A. By Proposition 11.4.4, each A[β1 , . . . , β j ]
n
with 1 ≤ j ≤ k is a finitely generated A[β1 , . . . , β j−1 ]-module, generated by {1, β j , . . . , β j j } for
some n j ≥ 0. Assuming recursively that A[β1 , . . . , β j−1 ] is finitely generated as an A-module,
Lemma 11.4.10 implies that A[β1 , . . . , β j ] = A[β1 , . . . , β j−1 ][β j ] is finitely generated as an A-
module as well. Therefore, (iii) holds. Finally, if (iii) holds and β ∈ B, then since B is a faith-
ful A[β ]-module (as 1 ∈ B), the element β is integral over a by Proposition 11.4.4. Thus (i)
holds. 
We derive the following important consequence.
P ROPOSITION 11.4.12. Suppose that C/B and B/A are integral extensions of commutative
rings. Then C/A is an integral extension as well.
P ROOF. Let γ ∈ C, and let f ∈ B[x] be a monic polynomial which has γ as a root. Let B0 be
the subring of B generated over A by the coefficients of f , which is integral over A as B is. By
Proposition 11.4.11, the ring B0 is then finitely generated as an A-module. As B0 [γ] is a finitely
generated B0 -module as well, we have B0 [γ] is finitely generated as an A-module. Hence, B[γ] is
 11.4. INTEGRAL EXTENSIONS 339

itself an integral extension of A. By definition of an integral extension, the element γ is integral

over A. Since γ ∈ C was arbitrary, we conclude that C is integral over A. 
D EFINITION 11.4.13. Let B/A be an extension of commutative rings. The integral closure
of A in B is the set of elements of B that are integral over A.
P ROPOSITION 11.4.14. Let B/A be an extension of commutative rings. Then the integral
closure of A in B is a subring of B.
P ROOF. If α and β are elements of B that are integral over A, then A[α, β ] is integral over A
by Proposition 11.4.11. Therefore, every element of A[α, β ], including −α, α + β , and α · β , is
integral over A as well. That is, the integral closure of A in B is closed under addition, additive
inverses, and multiplication, and it contains 1, so it is a ring. 
E XAMPLE 11.4.15. The integral closure of Z in Z[x] is Z, since if f ∈ Z[x] is of degree at
least 1 and g ∈ Z[x] is nonconstant, then g( f (x)) has degree deg g · deg f in x, hence cannot be 0.
D EFINITION 11.4.16.
a. The ring of algebraic integers is the integral closure Z of Z inside C.
b. An algebraic integer is an element of Z.
D EFINITION 11.4.17. Let B/A be an extension of commutative rings. We say that A is
integrally closed in B if A is its own integral closure in B.
D EFINITION 11.4.18. We say that an integral domain A is integrally closed if it is integrally
closed in its quotient field.
More generally, we may make the following definition.
D EFINITION 11.4.19. A commutative ring is normal if it is integrally closed in its total ring
of fractions.
In particular, a domain is is integrally closed if and only if it is normal.
E XAMPLE 11.4.20. Every field is integrally closed.
P ROPOSITION 11.4.21. Let A be an integrally closed domain, let K be the quotient field, and
let L be a field extension of K. If β ∈ L is integral over A with minimal polynomial f ∈ K[x], then
f ∈ A[x].
P ROOF. Since β ∈ L is integral, it is the root of some monic polynomial g ∈ A[x] such that
f divides g in K[x]. As g is monic, every root of g in an algebraic closure K containing K is
integral over A. As every root of f is a root of g, the same is true of the roots of f . Write
f = ∏ni=1 (x − βi ) for βi ∈ K integral over A. As the integral closure of A in K is a ring, it follows
that every coefficient of f is integral over A, being sums of products of the elements βi . Since
f ∈ K[x] and A is integrally closed, we then have f ∈ A[x]. 
In particular, we have the following result on the norm and trace on quotient fields on integral
extensions of domains.
340 11. COMMUTATIVE ALGEBRA

C OROLLARY 11.4.22. Let B/A be an integral extension of domains, and suppose that A is
integrally closed in its quotient field K. Let L denote the quotient field of B, and suppose that
L/K is finite. Then NL/K (β ) and TrL/K (β ) are elements of A for every β ∈ B.
P ROOF. Since β is integral over A, so are all of its conjugates in an algebraic closure of
L, since they are also roots of the monic polynomial of which β is a root. It follows from
Proposition 10.1.4 that NL/K (β ) and TrL/K (β ) are elements of B ∩ K, which is A since A is
integrally closed. 
The following holds in the case of UFDs.
P ROPOSITION 11.4.23. Let A be a UFD, let K be the quotient field of A, and let L be a field
extension of K. If β ∈ L is integral over A with minimal polynomial f ∈ K[x], then f ∈ A[x].
P ROOF. Let β ∈ L be integral over A, let g ∈ A[x] be a monic polynomial of which it is a
root, and let f ∈ K[x] be the minimal polynomial of β . Since f divides g in K[x] and A is a UFD
with quotient field K, there exists d ∈ K such that d f ∈ A[x] and d f divides g in A[x]. Since f is
monic, d must be an element of A (and in fact may be taken to be a least common denominator of
the coefficients of f ). The coefficient of the leading term of any multiple of d f will be divisible
by d, so this forces d to be a unit, in which case f ∈ A[x]. 
C OROLLARY 11.4.24. Every unique factorization domain is integrally closed.
P ROOF. The minimal polynomial of an element a of the quotient field K of a UFD A is x − a.
If a ∈
/ A, it follows from Proposition 11.4.23 that a is not integral over A. 
E XAMPLES 11.4.25. The ring Z is integrally closed.
√ √
1+ 17
E XAMPLE 11.4.26. The ring Z[ 17] is not integrally closed, since α = is a root of
√ 2
the monic polynomial x2 − x − 4. In particular, Z[ 17] is not a UFD.
P ROPOSITION 11.4.27. Let B/A be an extension of commutative rings, and suppose that B
is an integrally closed domain. Then the integral closure of A in B is integrally closed.
P ROOF. Let A denote the integral closure of A in B, and let Q denote the quotient field of A.
Let α ∈ Q, and suppose that α is integral over A. Then A[α] is integral over A, so A[α] is integral
over A, and therefore α is integral over A. That is, α is an element of A, as desired. 
E XAMPLE 11.4.28. The ring Z of algebraic integers is integrally closed.
P ROPOSITION 11.4.29. Let A be an integral domain with quotient field K, and let L be an
algebraic extension of K. Then the integral closure B of A in L has quotient field equal to L inside
L. In fact, every element of L may be written as db for some d ∈ A and b ∈ B.
P ROOF. Any β ∈ L is the root of a nonconstant polynomial f = ∑ni=0 ai xi ∈ K[x], with an = 1.
Let d ∈ A be such that d f ∈ A[x]. Then
n
n −1
d f (d x) = ∑ ai d n−i xi ∈ A[x]
i=0
is both monic and has dβ as a root. In other words, dβ is contained in B, as desired. 
 11.4. INTEGRAL EXTENSIONS 341

E XAMPLE 11.4.30. The quotient field of Z is Q.

D EFINITION 11.4.31. A number field (or algebraic number field) is a finite field extension of
Q.
D EFINITION 11.4.32. The ring of integers (or integer ring) of a number field K is the integral
closure of Z in K.
In other words, the ring of integers of a number field is the subring of algebraic integers it
contains. The prototypical examples of rings of integers arise in the setting of quadratic fields.
√
T HEOREM 11.4.33. Let d 6= 1 be a square-free integer. The ring O of integers in Q( d) is
( h √ i
Z 1+2 d if d ≡ 1 mod 4,
O= √
Z[ d] if d ≡ 2, 3 mod 4.
√
P ROOF. Suppose that α = a + b d is integral for a, b ∈ Q. If b = 0, then we must have
a ∈ Z. If b 6= 0, then the minimal polynomial of α is f = x2 − 2ax + a? ?2 − b2 d. Since α
is integral, then we must have f ∈ Z[x], so 2a ∈ Z. If a ∈ Z, then since a2 − b2 d ∈ Z and d is
/ Z, then 2a = a0 and 2b = b0 for some odd a0 , b0 ∈ Z, and
square-free, we have b ∈ Z as well. If a ∈
0 2 0 2 2
(a ) ≡ (b ) d mod 4. As (Z/4Z) = {0, 1}, this is impossible if d 6≡ 1 mod 4. If d ≡ 1 mod 4,
then clearly we can take a0 = b0 = 1. 
D EFINITION 11.4.34. Let B/A be an integral extension of domains such that A is integrally
closed, and suppose that B is free of rank n as an A-module. Let (β1 , . . . , βn ) be an ordered
basis of B as a free A-module. The discriminant B over A relative to the basis (β1 , . . . , βn ) is
D(β1 , . . . , βn ).
L EMMA 11.4.35. Let A be an integrally closed domain with quotient field K. Let L be a
finite separable extension of K, and let B denote the integral closure of A in L. Let (α1 , . . . , αn )
be any ordered basis of L as a K-vector space that is contained in B. Let β ∈ L be such that
TrL/K (αβ ) ∈ A for all α ∈ B. Then
n
D(α1 , . . . , αn )β ∈ ∑ Aαi .
i=1
P ROOF. Since β ∈ L, we may write
n
β = ∑ ai αi
i=1
for some ai ∈ K for 1 ≤ i ≤ n. For any i, we have that
n
(11.4.1) TrL/K (αi β ) = ∑ a j TrL/K (αiα j ).
j=1
The right-hand side of (11.4.1) is the ith term of the product of the matrix Q = (TrL/K (αi α j ))
times the column vector with ith entry ai . Since the determinant of Q is d = D(α1 , . . . , αn ), letting
Q∗ ∈ Mn (A) denote the adjoint matrix to Q, we have Q∗ Q = dIn . Thus, we have dai ∈ A for each
i. In other words, dβ lies in the A-module generated by the αi , so we are done. 
342 11. COMMUTATIVE ALGEBRA

P ROPOSITION 11.4.36. Let A be an integrally closed domain with quotient field K. Let L be
a finite separable extension of K, and let B denote the integral closure of A in L. There exists
an ordered basis (α1 , . . . , αn ) of L as a K-vector space contained in B. Moreover, for any such
basis, we have
n n
∑ Aαi ⊆ B ⊆ ∑ Ad −1αi,
i=1 i=1
where d = D(α1 , . . . , αn ).
P ROOF. First, take any ordered basis (β1 , . . . , βn ) of L/K. By Proposition 11.4.29, there
exists a ∈ A − {0} such that αi = aβi ∈ B for each 1 ≤ i ≤ n. Clearly, (α1 , . . . , αn ) is a basis of
L/K, so in particular, the A-module generated by the αi is free and contained in B. The other
containment is simply a corollary of Lemma 11.4.35 and the fact that TrL/K (B) ⊆ A. 
The following notion of rank is most interesting for finitely generated modules, though we
shall have occasion to use it without this assumption.
D EFINITION 11.4.37. The rank of a module M over a domain A is
rankA (M) = dimK (K ⊗A M).
C OROLLARY 11.4.38. Let A be an integrally closed noetherian domain with quotient field K.
Let L be a finite separable extension of K, and let B denote the integral closure of A in L. Then B
is a finitely generated, torsion-free A-module of rank [L : K].
P ROOF. By Proposition 11.4.36, we have free A-modules M and M 0 of rank n = [L : K] such
that M ⊆ A ⊆ M 0 . Since M 0 has no A-torsion, neither does B. We have
K ⊗A M ⊆ K ⊗A B ⊆ K ⊗A M 0 .
As M and M 0 are both isomorphic to An , their tensor products over A with K are n-dimensional
K-vector spaces, which forces K ⊗A B to have K-dimension n as well. Moreover, B is finitely
generated being a submodule of a finitely generated module over A, as A is noetherian. 
P ROPOSITION 11.4.39. Let A be an integrally closed noetherian domain with quotient field
K. Let L be a finite separable extension of K, and let B denote the integral closure of A in L. Then
any finitely generated, nonzero B-submodule of L is a torsion-free A-module of rank [L : K].
P ROOF. Let M be a finitely generated, nonzero B-submodule of L. If β ∈ L× , then the
multiplication-by-β map B → Bβ is an isomorphism of B-modules, so Bβ has rank [L : K] as an
A-module. In particular, rankA (M) ≥ rankA (B), taking β ∈ M. Since M is B-finitely generated
and contained in the quotient field of B, there exists α ∈ B such that αM ⊆ B. Since multiplication
by α is an isomorphism, rankA (M) ≤ rankA (B). The result now follows from Corollary 11.4.38.

C OROLLARY 11.4.40. Let A be a PID with quotient field K, let L be a finite separable ex-
tension of K, and let B denote the integral closure of K in L. Then any finitely generated B-
submodule of L is a free A-module of rank [L : K].
P ROOF. By the structure theorem for modules over a PID, any torsion-free rank n module
over A is isomorphic to An . The result is then immediate from Proposition 11.4.39. 
 11.4. INTEGRAL EXTENSIONS 343

We have the following application to number fields.

L EMMA 11.4.41. Let K be a number field. Then the discriminant of OK over Z is independent
of the choice of ordered basis of OK as a free Z-module.
P ROOF. By Corollary 11.4.40, the ring OK is free of rank n = [K : Q] over Z. If β1 , . . . , βn
and α1 , . . . , αn are bases of OK as a free Z-module, then there exists a Q-linear homomorphism
T : K → K such that T (αi ) = βi for all i. Then
D(β1 , . . . , βn ) = det(T )2 D(α1 , . . . , αn ),
and det(T ) is a unit in Z, so in {±1}, which is to say that det(T )2 = 1. 
D EFINITION 11.4.42. If K is a number field, the discriminant disc(K) of K is the discriminant
of OK over Z relative to any basis of OK as a free Z-module.
Noting Theorem 11.4.33, the case of quadratic fields is immediately calculated.
√
P ROPOSITION 11.4.43. Let K = Q( d), where d 6= 1 is a square-free integer. Then
(
d d ≡ 1 mod 4,
disc(K) =
4d d ≡ 2, 3 mod 4.
The following theorem will be useful to us later.
T HEOREM 11.4.44 (Noether’s normalization lemma). Let F be a field, and let A be a finitely
generated commutative F-algebra with generators z1 , . . . , zr ∈ A. Then there exists s ≤ r and
F-algebraically independent elements t1 , . . . ,ts ∈ A such that A is integral over F[t1 , . . . ,ts ].
P ROOF. The result is obvious for r = 0, so suppose r ≥ 1. If the elements z1 , . . . , zr are
algebraically independent over F, then we may take s = r and ti = zi for all i, so suppose not.
In this case, there exists a nonzero polynomial f ∈ F[x1 , . . . , xr ] such that f (z1 , . . . , zr ) = 0. Let
d be the maximum of the degrees of f viewed as a polynomial in each of the xi . Since f is
nonconstant, without loss of generality we may take it to be nonconstant as a polynomial in x1
with coefficients in F[x2 , . . . , xr ].
Consider the polynomial
(d+1)r−1
 
d+1
g(x1 , . . . , xr ) = f x1 , x2 + x1 , . . . , xr + x1 .

Each monomial x1k1 · · · xrkr in f has ki ≤ d for all i and gives rise to a sum of monomials in g,
exactly one of which has the form a constant in F times x1 to the power ∑ri=1 ki (1 + d)i−1 . Each
of these powers for the different monomials in f is distinct, so the highest degree term in g
viewed as a polynomial in x1 has a nonzero coefficient c that lies in F. That is, c−1 g is monic as
a polynomial in x1 with coefficients in F[x2 , . . . , xr ].
Set
(d+1)i−1
wi = zi − z1
for 2 ≤ i ≤ r, and note that g(z1 , w2 , . . . , wr ) = f (z1 , . . . , zr ) = 0. It follows that z1 is integral over
B = F[w2 , . . . , wr ]. By induction, there exist s ≤ r and elements t1 , . . . ,ts ∈ B that are algebraically
344 11. COMMUTATIVE ALGEBRA

independent over F and for which B is integral over F[t1 , . . . ,ts ]. Then A = B[z1 ] is integral over
F[t1 , . . . ,ts ] by the transitivity of integral extensions, proving the theorem. 
C OROLLARY 11.4.45. Let K be an extension of a field F that is finitely generated as an
F-algebra. Then K is a finite extension of F.
P ROOF. By Noether’s normalization lemma, K is an integral extension of F[t1 , . . . ,ts ] for
some algebraically independent elements t1 , . . . ,ts ∈ K. However, K is a field so contains the
quotient field K(t1 , . . . ,ts ). Since no ti−1 is integral over F[t1 , . . . ,ts ], we must have s = 0. Thus
K is integral over F, which is to say it is an algebraic extension of F, but then it is clearly finite
being that it is generated by finitely many elements. 
11.5. Going up and going down
We use B/A to denote an extension of commutative rings.
D EFINITION 11.5.1. Let B/A be an extension of commutative rings. We say that an ideal b
of B lies over an ideal a of A if b ∩ A = a.
We begin by noting the following simple lemma.
L EMMA 11.5.2. Let A be an integral domain, and let B be a commutative ring extension of
A that is integral over A. If b is an ideal of B that contains a nonzero element which is not a zero
divisor, then b lies over a nonzero ideal of A.
P ROOF. That b ∩ A is an ideal is clear, so it suffices to show that b ∩ A is nonzero. Let β ∈ b
be nonzero and not a zero divisor. The element β is a root of some monic polynomial g ∈ A[x].
Write g = xn f for some nonzero f ∈ A[x] with nonzero constant term. Since β ∈ b, we have
f (β ) − f (0) ∈ b, and as f (β ) = 0 given that β is not a zero divisor, we have f (0) ∈ b. But
f (0) 6= 0, so b has a nonzero element. 
The following are also easily verified.
L EMMA 11.5.3. If B/A is integral and b is an ideal of B that lies over a, then B/b is integral
over A/a.
L EMMA 11.5.4. Let S be a multiplicatively closed subset of A. If B/A is integral, then so is
S−1 B/S−1 A.
P ROPOSITION 11.5.5. Let B/A be an integral extension. If p is a prime ideal of A, then there
exists a prime ideal q of B lying over p.
P ROOF. The ring Bp = Sp−1 B is integral over Ap by Lemma 11.5.4. Let M be a maximal ideal
of Bp . Set m = M ∩ Ap . Since the field Bp /M is an integral extension of Ap /m, Lemma 11.4.8
tells us that Ap /m is a field as well, so m is a maximal ideal of Ap . Since Ap is local, we have
m = pAp . Let ιB : B → Bp and ιA : A → Ap be the localization maps so that q = ιB−1 (M) is prime,
and
q ∩ A = ιB−1 (M) ∩ A = ιA−1 (m) = ιA−1 (pAp ) = p.

 11.5. GOING UP AND GOING DOWN 345

T HEOREM 11.5.6 (Going up). Let B/A be an integral extension. Suppose that p1 ⊆ p2 are
prime ideals of A and q1 is a prime ideal of B lying over p1 . Then there exists a prime ideal q2 of
B containing q1 and lying over p2 .
P ROOF. Let Ā = A/p1 and B̄ = B/p2 , and let π : B → B̄ be the quotient map. Let p̄2 be the
image of p2 in Ā. By Proposition 11.5.5, there exists a prime ideal q̄2 of B̄ lying over p̄2 . Then
q2 = π −1 (q̄2 ) contains q1 and satisfies
q2 ∩ A = π −1 (q̄2 ∩ Ā) = π −1 (p̄2 ) = p2 ,
since p2 contains p1 . 
P ROPOSITION 11.5.7. Let B/A be an integral extension of domains. Let a be an ideal of A.
B is a root of f ∈ A[x] with f = xn + ∑n−1
An element β ∈ √ i
i=0 ai x with ai ∈ a for all 0 ≤ i ≤ n − 1 if
and only if β ∈ aB.

√ ∈ B is a root of f as in the statement of the proposition. Then

P ROOF. First, suppose that β
n−1
βn i
= − ∑i=0 ai β ∈ aB, so β ∈ aB.
Conversely, suppose that β n ∈ aB for some n ≥ 1. Let a1 , . . . , ak ∈ a and b1 , . . . , bk ∈ B for
k ≥ 1 be such that β n = ∑ki=1 ai bi . Set M = A[b1 , . . . , bk ], which is a finitely generated faithful
A-module as a finitely generated integral A-algebra with the property that β n M ⊆ aM. Given a
generating set of M with n elements, as in the proof of Lemma 11.4.4, we may form a matrix
in Mn (A) given by the action of β n on the generating set by multiplication, and it has entries in
a. Then its characteristic polynomial c(x) = ∑ni=1 ci xi ∈ A[x] has ci ∈ a for i < n. Since c(β n )
annihilates M, we have c(β n ) = 0 as M is faithful. We then have β n = − ∑ni=1 ci β i ∈ aB. 
P ROPOSITION 11.5.8. Let B/A be an integral extension of domains such that A is integrally
closed, and suppose that β ∈ B is the root of a monic polynomial in A[x] with non-leading coef-
ficients in a radical ideal a of A. Then the minimal polynomial of β is also such a polynomial.
P ROOF. Let K = Q(A) and f ∈ K[x] be the minimal polynomial of β . Note that f ∈ A[x] by
Proposition 11.4.21. Let L be an extension of K that contains all of the roots of f , and let C be
the integral closure of A in L. √
By Proposition 11.5.7, we have β ∈ aC, so there exists g ∈ A[x] monic with g(β ) = 0 and
which has image xn in A/a[x] for n = deg g. Then√f divides g in K[x], so in particular g(γ) = 0
for every root γ of f , which again tells us that γ ∈ aC.
Since the non-leading
√ coefficients of f are symmetric polynomials in the roots of f , we now
have that they lie in aC. Once again, such a coefficient a is a root of a monic g ∈ A[x] which
reduces to xdeg g modulo a, but then adeg g ∈ a as a ∈ A. As we have assumed that a is radical, we
then have a ∈ a, so f has the desired form. 
L EMMA 11.5.9. Let p be a prime ideal of A. There exists a prime ideal q of B lying over p if
and only if pB ∩ A = p.
P ROOF. If q ∩ A = p, then q contains pB, and then
p = q ∩ A ⊇ pB ∩ A ⊇ p,
so pB ∩ A = p.
346 11. COMMUTATIVE ALGEBRA

Conversely, if pB ∩ A = p, then pB is disjoint from Sp , so there exists a maximal ideal M of

Bp with pB contained in M. Let q be the inverse image of M in B. Then m = q ∩ A is a prime
ideal containing p with m ∩ Sp = ∅, which forces m = p. 
T HEOREM 11.5.10 (Going down). Let B/A be an integral extension of integral domains with
A integrally closed. Suppose that p2 ⊆ p1 are prime ideals of A and q1 is a prime ideal of B lying
over p1 . Then there exists a prime ideal q2 of B contained in q1 and lying over p2 .
P ROOF. Note that the maps B → Bq1 and A → Ap1 are injective as B is a domain. By
Lemma 11.5.9, it is enough to show that p2 Bq1 ∩ A = p2 . That is, in this case there exists a
prime ideal Q2 of Bq1 lying over p2 Ap1 , and then we can take q2 = Q2 ∩ B.
If β = bs ∈ p2 Bq1 with b ∈ p2 B and s ∈ B − q1 , then by Proposition 11.5.8, the minimal
polynomial f = xn + ∑n−1 i
i=0 ai x of b has non-leading coefficients in p2 . If β is also in A, then s =
β −1 b has minimal polynomial β −n f (β x) ∈ Q(A)[x]. Since s is integral over A, this polynomial
lies in A[x], and therefore β i−n ai ∈ A for all i. If β ∈
/ p2 , then for all i we have β i−n ai ∈ p2 since
n
ai ∈ p2 . But then s ∈ p2 B, so s ∈ q1 , a contradiction. Thus, β ∈ p2 , as required. 

11.6. Hilbert’s Nullstellensatz

We use K to denote a fixed algebraically closed field in this section. Much but certainly not
all of what is done here can be generalized to fields which are not algebraically closed as well, but
for this brief introduction, we feel it suffices to focus on the more specific setting. This section
assumes some basic knowledge of topological spaces.
Fix a nonnegative integer n.
D EFINITION 11.6.1. Let S be a subset of K[x1 , . . . , xn ]. The zero set, or vanishing locus, of S
is
V (S) = {(a1 , . . . , an ) ∈ K n | f (a1 , . . . , an ) = 0 for all f ∈ S}.
An algebraic set in K n is any subset of K n that is a zero set of some set of polynomials in
K[x1 , . . . , xn ].
From now on, let us set R = K[x1 , . . . , xn ] for brevity.
N OTATION 11.6.2. If S = { f1 , . . . , fn } ⊂ R, we also write V ( f1 , . . . , fn ) for V (S). At times,
for a = (a1 , . . . , an ) ∈ K n , we write f (a) for f (a1 , . . . , an ).
E XAMPLE 11.6.3. We have V (∅) = K n and V (R) = V (1) = ∅.
E XAMPLE 11.6.4. Consider f (x, y) = x−y and g(x, y) = x2 +y2 −2 in C[x, y]. Then V ( f , g) =
V ( f ) ∩V (g) = {(−1, −1), (1, 1)}.
R EMARK 11.6.5. For any subset S of R, the zero set V (S) equals the zero set of the ideal (S)
generated by S.
P ROPOSITION 11.6.6.
a. The intersection of any collection of algebraic sets in K n is also an algebraic set.
b. The union of any finite collection of algebraic sets in K n is also an algebraic set.
 11.6. HILBERT’S NULLSTELLENSATZ 347

P ROOF. Let {Si | i ∈ I} be a collection of subsets of R. Then i∈I V (Si ) = V (S) is algebraic,
T

so we have part a. If S and T are subsets of R, set I = (S) and J = (T ). We clearly have
V (S) ∪V (T ) = V (I) ∪V (J) ⊆ V (I ∩ J).
If a ∈ V (I ∩ J) and a ∈
/ V (I), then there exists f ∈ I with f (a) 6= 0. If g ∈ J, then f g ∈ I ∩ J, so
f (a)g(a) = 0, so g(a) = 0. Thus a ∈ V (J), and we have part b. 
It follows from the proposition that the following definition does in fact yield a topology.
D EFINITION 11.6.7. The Zariski topology on K n is the topology {K n − V (S) | S ⊆ R} with
closed sets the algebraic sets in K n .
D EFINITION 11.6.8. For n ≥ 0, the affine n-space over K is the set AnK = K n endowed with
the Zariski topology.
R EMARK 11.6.9. For any (a1 , . . . , an ) ∈ AnK , we have V (x1 −a1 , . . . , xn −an ) = {(a1 , . . . , an )},
so points in AnK are closed. However, it is not a Hausdoff topology: for instance, for n = 1, the
only closed sets other than X are finite, so any two nonempty open sets will intersect as K is
infinite.
N OTATION 11.6.10. Let Z ⊆ AnK . Then
I(Z) = { f ∈ K[x1 , . . . , xn ] | f (a) = 0 for all a ∈ Z}.
The set I(Z) is clearly an ideal: it is the ideal of R of elements that vanish on all of Z.
R EMARK 11.6.11. Note that if f ∈ R satisfies f k ∈ I(Z) for some subset Z of AnK and k ≥ 1,
then f (a)k = 0 for all a ∈ Z, so f vanishes on Z, which is to say that f ∈ I(Z). Hence, I(Z) is a
radical ideal.
E XAMPLE 11.6.12. For a = (a1 , . . . , an ) ∈ AnK , we have I({a}) = (x1 − a1 , . . . , xn − an ).
In particular I and V provide bijections between the points of AnK and a subset of the maximal
ideals of R, i.e., those of the form (x1 − a1 , . . . , xn − an ) for some a ∈ AnK . The statement the
latter maximal ideals are all of the maximal ideals of R is known as the weak form of Hilbert’s
Nullstellensatz.
T HEOREM 11.6.13. Every maximal ideal of K[x1 , . . . , xn ] has the form (x1 − a1 , . . . , xn − an )
for some (a1 , . . . , an ) ∈ AnK .
P ROOF. Let m be a maximal ideal of R = K[x1 , . . . , xn ], and consider L = R/m, which is
a field containing K that is finitely generated over K. By Corollary 11.4.45, the field L is an
algebraic extension of the algebraically closed field K, so it is equal to K. Under the quotient
map R → L = K, each xi is sent to some ai ∈ L, so xi − ai ∈ m. Since (x1 − a1 , . . . , xn − an ) is
maximal, it equals m. 
In other words, V and I give inverse bijections between the maximal ideals of R and the
singleton subsets of AnK . We now prove
√ the stronger form of this statement, one which boils
down to the statement that I(V (a)) = a for ideals a of R.
348 11. COMMUTATIVE ALGEBRA

T HEOREM 11.6.14 (Hilbert’s Nullstellensatz). The maps I and V provide mutually inverse,
inclusion-reversing bijections

{radical ideals of K[x1 , . . . , xn ]} o

V / {algebraic sets in An }.
K
I

P ROOF. The operation I is by definition inclusion-reversing on subsets of AnK , and the opera-
tion V is inclusion-reversing on subsets of R = K[x1 , . . . , xn ]. It is immediate
√ from the definitions
and Remark 11.6.11 that if a is an ideal of R, then I(V (a)) contains a, and if Z is a subset of R,
then V (I(Z)) contains Z. If Z = V (a) for some ideal a, then
V (I(Z)) = V (I(V (a))) ⊆ V (a) = Z,
since V is inclusion-reversing. Thus, on √algebraic sets Z, we have V (I(Z)) = Z.
It remains to show that I(V (a)) ⊆ a for any ideal a of R. Let f ∈ I(V (a)). For an indeter-
minate y, let J be the ideal of R[y] generated by a and 1 − f y. We view R[y] as K[x1 , . . . , xn , y] and
consider the vanishing set of J in An+1
K . If a = (a1 , . . . , an+1 ) ∈ V (J), then (a1 , . . . , an ) ∈ V (a), in
which case we have
(1 − f y)(a1 , . . . , an+1 ) = 1 − f (a1 , . . . , an )an+1 = 1 6= 0.
Thus V (J) = ∅. By the weak form of the Nullstellensatz, if J were a proper ideal, then its
vanishing locus would contain the point in the vanishing locus of a maximal ideal containing it,
so J = R[y].
In the quotient R[y]/(1 − f y), which we identify with R[√f −1 ] via y 7→ f −1 , it follows that
1 ∈ aR[ f −1 ]. In other words, f N ∈ a for some N ≥ 0, so f ∈ a, as was desired. 
R EMARK 11.6.15. We record the following simple consequences of the Nullstellensatz.
p p
a. For any S ⊆ R, we have V (S) = V ( (S)), and I(V (S)) = (S).
b. For any Z ⊆ AnK , we have I(Z) = I(Z), where Z is the closure of Z in the Zariski topology
(i.e., the smallest algebraic set containing Z), and V (I(Z)) = Z.
D EFINITION 11.6.16. We say that an algebraic set is irreducible if it is not a union of two
proper algebraic subsets.
We have seen that maximal ideals correspond to singleton sets under V and I. Hilbert’s
Nullstellensatz tells us that prime ideals correspond to irreducible algebraic sets.
C OROLLARY 11.6.17. The maps I and V restrict to mutually inverse, inclusion-reversing
bijections

{prime ideals of K[x1 , . . . , xn ]} o

V / {irreducible algebraic sets in An }.
K
I

P ROOF. An algebraic set Z is by definition the vanishing locus of some radical ideal I of R.
By the Nullstellensatz, such a set Z is irreducible if and only if I = I(Z) cannot be written as
an intersection of two radical ideals properly containing I. Note that if I were the intersection
of two arbitrary ideals, then it would also be the intersection of their radicals. Conversely, if I
is an irreducible ideal, then so is its radical, and then its vanishing locus is irreducible as well.
 11.7. SPECTRA OF RINGS 349

Since the irreducible ideals in R are exactly the primary ideals, and those which are radical are
the prime ideals, irreducible algebraic sets correspond exactly to the prime ideals of R. 
R EMARK 11.6.18. By the primary decomposition theorem and Corollary 11.6.17, every al-
gebraic set is a finite union of irreducible algebraic sets.
R EMARK 11.6.19. An irreducible algebraic set Z ⊆ AnK together with its subspace topology
is also what is called an (affine) algebraic variety. It has an associated coordinate ring K[Z] =
R/I(Z). Note that the ring R/I(Z) is a domain, since I(Z) is prime. The radical ideals of K[Z]
correspond to algebraic subsets of Z, and via this bijection the maximal ideals of K[Z] correspond
to the points (or more precisely, singleton subsets) of Z.

11.7. Spectra of rings

In the previous section, we saw that the points of AnK for an algebraically closed field K
correspond to the maximal ideals of K[x1 , . . . , xn ]. Making this identification, we may think of
the Zariski topology as endowing the set of maximal ideals of K[x1 , . . . , xn ] with a topology. We
now aim to mimic this for the larger set of prime ideals, in an arbitrary commutative ring R.
D EFINITION 11.7.1. The spectrum Spec R of a commutative ring R is the set of prime ideals
of R.
E XAMPLE 11.7.2. For a PID R, we have Spec R = {(0)} ∪ {( f ) | f irreducible}.
N OTATION 11.7.3. For any subset T of R, we set
V (T ) = {p ∈ Spec R | T ⊆ p}.
For any subset Y of Spec R, we set \
I(Y ) = p.
p∈Y

R EMARK 11.7.4. We have V√(T ) = V ((T )) for the ideal

√ (T√) generated by T . In fact, for any
ideal I of R, we have V (I) = V ( I) since if I ⊆ p, then I ⊆ p = p.
The following lemma is easily verified.
L EMMA 11.7.5.
a. We have V ((0)) = Spec R and V (R) = ∅.
b. If a and b are ideals of R, then V (a) ∪V (b) = V (a ∩ b) = V (ab).
c. If {a j | j ∈ X} is a collection of ideals of R, then
! !
\ [
V (a j ) = V aj =V ∑ aj .
j∈X j∈X j∈X

In particular, the sets V (I) for I an ideal of R form a topology on Spec R.

D EFINITION 11.7.6. The Zariski topology on Spec R is the unique topology with closed sets
the V (I) with I an ideal of R.
350 11. COMMUTATIVE ALGEBRA

R EMARK 11.7.7. In Spec R, the singleton sets {m} with m maximal are closed, since V (m) =
{m}. However, points in general need not be closed. The closure of {p} with p prime is the
smallest closed subset containing p, which is exactly V (p), the set of prime ideals containing p.
So, {p} is closed if and only if p is maximal. E.g., in an integral domain, the closure of (0) is
Spec R!
D EFINITION 11.7.8. The closed points of Spec R are the maximal ideals of R.
D EFINITION 11.7.9. The closure of a subset Y of Spec R in the Zariski topology on R is
known as the Zariski closure of Y .
The analogue of the Nullstellensatz for Spec R is considerably less difficult.
P ROPOSITION 11.7.10. The maps I and V provide mutually inverse, inclusion-reversing bi-
jections
V /
{radical ideals of R} o {closed subsets of Spec R}.
I
√
In fact, for any ideal a of R, we have I(V (a)) = a, and for any subset Y of Spec R, the set
V (I(Y )) is the Zariski closure of Y .
P ROOF. That V and I are inclusion-reversing is clear. Let a be an ideal of R. Then
\ √
I(V (a)) = p= a
a⊆p

by Proposition 11.2.10. Conversely, if Y is a subset of Spec R, then its closure Y is V (a) for some
ideal a of R, and
√
V (I(Y )) ⊆ V (I(Y )) = V (I(V (a))) = V ( a) = V (a) = Y ,
but V (I(Y )) is closed an contains Y , so V (I(Y )) = Y . 
C OROLLARY 11.7.11. The Zariski closure of a subset Y of Spec R is the set of all prime
ideals containing some element of Y .
P ROOF. The set V (I(Y )) consists of the prime ideals p containing the intersection of all prime
ideals containing Y . By Lemma 3.10.19b, these are exactly the ideals that contain some element
of Y . 
Let us compare V and I of Definition 11.7.3 with our prior maps with these notations for the
polynomial ring R = K[x1 , . . . , xn ] over an algebraically closed field K.
P ROPOSITION 11.7.12. Let R = K[x1 , . . . , xn ] for some n ≥ 0 and algebraically closed field
K. Let us use V 0 and I 0 to denote the maps which take vanishing loci of algebraic sets in AnK and
the ideal of vanishing of subsets of R, respectively.
a. The injective map ι : AnK → Spec R given by taking a point to its corresponding maximal
ideal is a homeomorphism onto its image, which we use to identify AnK with a subspace of Spec R.
b. For any ideal a of R, we have V 0 (a) = V (a) ∩ AnK .
c. For any Zariski closed subset Y of Spec R, we have I(Y ) = I 0 (Y ∩ AnK ).
 11.7. SPECTRA OF RINGS 351

P ROOF. If Y = V (a) for some ideal a of Spec R, then Y ∩ AnK is the set of maximal ideals
of R containing a, which equals V 0 (a), proving part a. This implies that the intersection of Y
with AnK is closed and that the image of a closed set Z = V 0 (a) in the Zariski topology on AnK
is closed under the subspace topology on AnK from the Zariski topology on Spec R. Thus, ι is a
homeomorphism onto its image, proving part b.
Finally, if Y = V (a) is a closed subset of Spec R with a radical, and if Z = Y ∩ AnK = V 0 (a),
then a = I 0 (Z) is the intersection of all prime ideals containing a. That is, a is the intersection of
all prime ideals in V (a) = Y , so a = I(Y ) as well, and we have part c. 
D EFINITION 11.7.13. If a ∈ R, then Ua = Spec R −V ((a)) is called a principal open set of R.
P ROPOSITION 11.7.14. The sets Ua for a ∈ R form a basis for the Zariski topology on Spec R.
P ROOF
T
. Let U be an open
S
set in Spec R. Then U = Spec R − V (I) for some ideal I, and
V (I) = a∈I V ((a)), so U = a∈I Ua . 
We have the following simple lemma.
L EMMA 11.7.15. Let ϕ : R → S be a ring homomorphism.
a. If q is a prime ideal of S, then ϕ −1 (q) is a prime ideal of R.
b. Suppose that ϕ is surjective and p is a prime ideal of R containing the kernel of ϕ. Then
ϕ(p) is a prime ideal of S.
P ROOF. Set p = ϕ −1 (q) for q a prime ideal of S. Then a, b ∈ R satisfy ab ∈ p if and only if
ϕ(ab) ∈ q, so if and only if either ϕ(a) ∈ q or ϕ(b) ∈ q, i.e., a ∈ p or b ∈ p.
If p is a prime ideal of R containing ker ϕ, then q = ϕ(p) is an ideal of S by the surjectivity of
ϕ, For a, b ∈ S with ab ∈ q, write a = ϕ(c) and b = ϕ(d) for some c, d ∈ R. Then cd ∈ ϕ −1 (q) = p
since ker ϕ ⊆ p, so c ∈ p or d ∈ p, and therefore a ∈ q or b ∈ q. 
By Lemma 11.7.15, the following definition makes sense.
D EFINITION 11.7.16. Let ϕ : R → S be a ring homomorphism. The pullback map
ϕ ∗ : Spec S → Spec R
is the function given by ϕ ∗ (q) = ϕ −1 (q) for q ∈ Spec S.
The following lemma is simple.
L EMMA 11.7.17. If ϕ is a ring homomorphism, then the pullback map ϕ ∗ is continuous with
respect to the Zariski topologies.
R EMARK 11.7.18. The map that takes a ring to its spectrum and ring homomorphism to the
corresponding pullback map is a contravariant functor from Ring to Top.
E XAMPLE 11.7.19. Let π : R → F be a surjective ring homomorphism, where F is a field.
Then π ∗ ((0)) = ker π is the maximal ideal that is the kernel of f .
E XAMPLE 11.7.20. Let p be a prime ideal. The localization map ϕ : R → Rp has pullback
ϕ ∗ (qRp ) = q for prime ideals q of R contained in p.
352 11. COMMUTATIVE ALGEBRA

E XAMPLE 11.7.21. Consider the map ϕ : C[x] → C[x] given by ϕ( f )(x) = f (x2 ). Then
ϕ ∗ ((0)) = (0), and for a ∈ C irreducible,
ϕ ∗ ((x − a)) = {g ∈ C[x] | g(a2 ) = 0} = (x − a2 ).
In particular, ϕ ∗ is 2-to-1 on closed points, taking both (x − a) and (x + a) to (x − a2 ), except for
a = 0, in which case only (x) is carried to (x).
11.8. Krull dimension
We continue to use R to denote a commutative ring.
D EFINITION 11.8.1. The length of an ascending chain (pi )ni=0 of distinct prime ideals is n.
We often refer to such a finite strictly ascending chain more simply as a chain of prime ideals,
where minimal confusion can arise.
E XAMPLE 11.8.2. If R is an integral domain and n ≥ 0, then the ring R[x1 , . . . , xn ] contains a
chain of primes of length n:
(0) ⊂ (x1 ) ⊂ (x1 , x2 ) ⊂ · · · ⊂ (x1 , . . . , xn ).
D EFINITION 11.8.3. The Krull dimension, or dimension, dim R of a commutative ring R is
the length of the longest ascending chain of distinct prime ideals in R, if it exists, and is otherwise
said to be infinite.
R EMARK 11.8.4. In set-theoretic terms, if finite, dim R is one less than the maximum of the
cardinalities of all chains in Spec R.
E XAMPLES 11.8.5. Let F be a field.
a. The Krull dimension of F is 0: its only prime ideal is (0). In fact, the Krull dimension of
F[x]/(xn ) for n ≥ 0 is 1, since its unique prime ideal is (x).
b. The Krull dimension of Z is 1: the longest chains are all of the form (0) ⊂ (p) for some
prime number p. In fact, dim R = 1 for every PID R that is not a field.
c. The Krull dimension of F[(xi )i≥1 ] is infinite, since
(0) ⊂ (x1 ) ⊂ (x1 , x2 ) ⊂ · · ·
is an ascending chain of prime ideals that is not eventually constant.
L EMMA 11.8.6. Let π : R → S be a surjective map of rings. Then dim R ≥ dim S.
P ROOF. Let (qi )ni=0 be a chain of primes in S of length n, and set pi = ϕ −1 (qi ) for each i.
Then qi = ϕ(pi ) for each i, so each pi is distinct. 
L EMMA 11.8.7. Let p be a non-minimal prime of R. Then dim R ≥ dim R/p + 1.
P ROOF. Any chain in R/p of maximal length has inverse image in R of the same length, and
such a chain can be extended by adding in a minimal prime properly contained in p. 
P ROPOSITION 11.8.8. If B/A is an integral extension of domains, then A has finite Krull
dimension if and only if B does, in which case dim B = dim A.
 11.8. KRULL DIMENSION 353

P ROOF. The the going up theorem tells us that dim B ≥ dim A. Suppose that dim B = n, let
(qi )ni=0
be a maximal ascending chain of prime ideals of B, and set pi = qi ∩ A for all i. We
have p1 ∩ A 6= (0) by Lemma 11.5.2, so dim B/q1 = dim B − 1, and dim A/p1 ≤ dim A − 1 by
Lemma 11.8.7. By induction on dim B, we then have the remaining inequality dim B ≤ dim A. 

P ROPOSITION 11.8.9. Let F be a field and n be a nonnegative integer. The ring F[x1 , . . . , xn ]
has Krull dimension n.

P ROOF. Example 11.8.2 tells us that R = F[x1 , . . . , xn ] has dimension at least n. We may
suppose that n ≥ 1. Let (pi )mi=0 be an ascending chain of prime ideals in R. We may suppose that
p0 = (0) and that p1 is minimal, generated by an irreducible element g ∈ R, as otherwise we may
extend the chain to contain such primes.
Consider the quotient R = S/p1 = S/(g). Since the images of the xi in S satisfy an equation
of algebraic dependence over F, and these images generate S as an F-algebra, the quotient field
of S has transcendence degree at most n − 1 over F. Thus, no set of more than n − 1 elements of
S can be algebraically independent.
By Noether’s normalization lemma, there exist algebraically independent elements t1 , . . . ,ts ∈
R such that S is integral over F[t1 , . . . ,ts ]. From what we have already shown, we must have
s ≤ n − 1. Then dim S = s ≤ n − 1 by Proposition 11.8.8 and induction. On the other hand, the
images p̄i in S of the ideals pi with 1 ≤ i ≤ m remain prime in S by Lemma 11.7.15, and they are
distinct, so m − 1 ≤ s ≤ n − 1. Therefore, m = n. 

In fact, the following result, for which we omit the proof, holds more generally.

T HEOREM 11.8.10. Let R be a noetherian domain of finite Krull dimension. Then

dim R[x] = dim R + 1.

D EFINITION 11.8.11. The height ht(p) of a prime ideal p of R is the length of the longest
chain of primes of R contained in p. A prime of height 0 is called a minimal prime of R.

Note that a minimal prime of R is just an isolated prime of (0).

E XAMPLES 11.8.12.
a. In F[x1 , . . . , xn ] for a field F, the height of (x1 , . . . , xk ) for k ≤ n is k.
b. In a UFD, the primes of height one are principal, generated by the irreducible elements.
c. In a product of fields R = ∏ni=1 Fi , the minimal primes are the maximal ideals, the kernels
of projection maps R → Fk for some 1 ≤ k ≤ n.

R EMARK 11.8.13. Suppose R = K[x1 , . . . , xn ] with K algebraically closed. The prime ideals p
of R correspond to algebraic sets in AnK . The dimension of the algebraic set V that is the vanishing
locus of p is defined to be n − ht(p). In particular, AnK has dimension n, as one would expect. We
often refer to ht(p) as the codimension of V in AnK . In particular, the vanishing locus of a single
nonconstant polynomial in R has codimension 1.
354 11. COMMUTATIVE ALGEBRA

11.9. Dedekind domains

D EFINITION 11.9.1. A Dedekind domain is a noetherian, integrally closed domain of Krull
dimension at most 1.
The condition of having Krull dimension at most 1 is the same as every nonzero prime ideal
being maximal. We have the following class of examples.
L EMMA 11.9.2. Every PID is a Dedekind domain.
P ROOF. A PID is noetherian, and it is a UFD, so it is integrally closed. Its nonzero prime
ideals are maximal, generated by its irreducible elements. 
P ROPOSITION 11.9.3. Let A be a Dedekind domain, and let B be the integral closure of A in
a finite, separable extension of the quotient field of A. Then B is a Dedekind domain.
P ROOF. Note that B is a finitely generated A-module by Corollary 11.4.38. If b is an ideal of
B, then b is an A-submodule of B, and as A is noetherian, it is therefore finitely generated. Thus,
B is noetherian. That B is integrally closed is just Proposition 11.4.27. That every nonzero prime
ideal in A is maximal follows from Lemma 11.8.8. 
We have the following immediate corollary.
C OROLLARY 11.9.4. The ring of integers of any number field is a Dedekind domain.
More examples of Dedekind domains can be produced as follows.
P ROPOSITION 11.9.5. Let A be a Dedekind domain, and let S be a multiplicatively closed
subset of A. Then S−1 A is also a Dedekind domain.
P ROOF. Given an ideal b of S−1 A, set a = A ∩ b. Then a is an ideal of A, and b = S−1 a. It
follows that any set of generators of a as an ideal of A generates S−1 a as an ideal of S−1 A. Hence
S−1 A is noetherian. If, moreover, b is a nonzero prime, then clearly a is as well, and a is maximal
since A is a Dedekind domain. Then S−1 A/b ∼ = A/a is a field, so b is maximal as well.
Let K be the quotient field of A. Any α ∈ K that is integral over S−1 A satisfies a monic
polynomial f with coefficients in S−1 A. Set n = deg f . If d ∈ S is the product of the denominators
of these coefficients, then d n f (d −1 x) ∈ A[x] is monic with dα ∈ K as a root. Since A is integrally
closed, we have dα ∈ A, so α ∈ S−1 A. That is, S−1 A is integrally closed. 
L EMMA 11.9.6. Let A be a noetherian domain, and let a be a nonzero ideal of A.
a. There exist k ≥ 0 and nonzero prime ideals p1 , . . . , pk of A such that p1 · · · pk ⊆ a.
b. Suppose that dim A ≤ 1. If p1 , . . . , pk are as in part a and p is a prime ideal of A containing
a, then p = pi for some positive i ≤ k.
P ROOF. Consider the set X of nonzero ideals of A for which the statement of the first part of
the lemma fails, and order X by inclusion. Suppose by way of contradiction that X is nonempty.
As A is noetherian, X must contain a maximal element a by Proposition 5.1.15. Now a is not
prime since it lies in X, so let a, b ∈ A − a with ab ∈ a. Then a + (a) and a + (b) both properly
 11.9. DEDEKIND DOMAINS 355

contain a, so by maximality of a, there exist prime ideals p1 , . . . , pk and q1 , . . . , ql of A for some

k, l ≥ 0 such that p1 · · · pk ⊆ a + (a) and q1 · · · ql ⊆ a + (b). We then have
p1 · · · pk q1 · · · ql ⊆ (a + (a))(a + (b)) ⊆ a,
a contradiction of a ∈ X. This proves part a.
Now, suppose that a is proper, and let p be a prime ideal containing a. Assume that dim A ≤ 1.
If no pi equals p, then since pi is maximal, there exist bi ∈ pi − p for each 1 ≤ i ≤ k. We then
have b1 · · · bk ∈
/ p as p is prime, so b1 · · · bk ∈
/ a, a contradiction. Hence we have part b. 
D EFINITION 11.9.7. A fractional ideal of a domain A is a nonzero A-submodule a of the
quotient field of A for which there exists a nonzero d ∈ A such that da ⊆ A.
R EMARK 11.9.8. Every nonzero ideal in a domain A is a fractional ideal, which is sometimes
referred to as an integral ideal. Every fractional ideal of A that is contained in A is an integral
ideal.
E XAMPLE 11.9.9. The fractional ideals of Z are exactly the Z-submodules of Q generated
by a nonzero rational number.
L EMMA 11.9.10. Let A be a noetherian domain. An A-submodule of the quotient field of A
is a fractional ideal if and only if it is finitely generated.
P ROOF. If a is a finitely generated A-submodule of the quotient field of A, then let d ∈ A
denote the product of the denominators of a set of generators. Then da ⊆ A. Conversely, suppose
that a is a fractional ideal and d ∈ A is nonzero and satisfies da ⊆ A. Then da is an ideal of A,
hence finitely generated. Moreover, the multiplication-by-d map carries a isomorphically onto
da. 
D EFINITION 11.9.11. Let A be a domain with quotient field K, and let a and b be fractional
ideals of A.
a. The inverse of a is a−1 = {b ∈ K | ba ⊆ A}.
b. The product of a and b is the A-submodule of K generated by the set {ab | a ∈ a, b ∈ b}.
L EMMA 11.9.12. Let A be a domain, and let a and b be fractional ideals of A. Then a−1 ,
a + b, ab, a ∩ b are fractional ideals of A as well.
P ROOF. Let K denote the quotient field of A. Let c, d ∈ A be nonzero such that ca ⊆ A and
db ⊆ A. Then c(a ∩ b) ⊆ A, cd(a + b) ⊆ A, and cdab ⊆ A.
Note that a−1 is an A-submodule of K which is nonzero since there exists d ∈ A with da ⊆ A in
that a is a fractional ideal. Let a ∈ a be nonzero, and let e ∈ A be its numerator in a representation
of a as a fraction, so e ∈ a as well. For any t ∈ a−1 , we have te ∈ A by definition, so ea−1 ⊆ A,
and therefore a−1 is a fractional ideal. 
R EMARK 11.9.13. By definition, multiplication of fractional ideals is an associative (and
commutative) operation, so the set I(A) of fractional ideals in A is a monoid.
D EFINITION 11.9.14. We say that a fractional ideal a of a domain A is invertible if there
exists a fractional ideal b of A such that ab = A.
356 11. COMMUTATIVE ALGEBRA

L EMMA 11.9.15. A fractional ideal a of a domain A is invertible if and only if a−1 a = A.

P ROOF. For the nonobvious direction, suppose that a is invertible. Then we must have b ⊆
a−1 by definition of a−1 . On the other hand,
A = ba ⊆ a−1 a ⊆ A,
so we must have a−1 a = A. 
E XAMPLE 11.9.16. Consider the maximal ideal (x, y) of Q[x, y]. If f ∈ Q(x, y)× is such
that f x ∈ Q[x, y] (resp., f y ∈ Q[x, y]) then its denominator is a divisor of x (resp., y). Therefore
(x, y)−1 = Q[x, y], and we have
(x, y) · (x, y)−1 = (x, y) 6= Q[x, y].
Thus, (x, y) is not invertible as a fractional ideal.
D EFINITION 11.9.17. A principal fractional ideal of A is an A-submodule (a) generated by
a nonzero element a of the quotient field of A.
L EMMA 11.9.18. Let a be a fractional ideal of a PID. Then a is principal.
b
P ROOF. There exists d ∈ A such that da = (b) for some b ∈ A. Then d ∈ a and given any
c ∈ a, we have dc = ba for some a ∈ A, so c = a db . That is, a = ( db ). 
L EMMA 11.9.19. Let A be a domain, and let a be a nonzero element of its quotient field.
Then (a) is invertible, and (a)−1 = (a−1 ).
P ROOF. If x ∈ (a)−1 , then xa = b for some b ∈ A, so x = ba−1 ∈ (a−1 ). If x ∈ (a−1 ), then
x = a−1 b for some b ∈ A. On other hand, any z ∈ (a) has the form z = ya for some y ∈ A, and we
have xz = a−1 bya = by ∈ A, so x ∈ (a)−1 . We then have
(a)(a)−1 = (a)(a−1 ) = (aa−1 ) = A,
completing the proof. 
L EMMA 11.9.20. Let A be a Dedekind domain, and let p be a nonzero prime ideal of A. Then
pp−1 = A.
P ROOF. Let a ∈ p be nonzero. Noting Lemma 11.9.6a, we let k ≥ 1 be minimal such that
there exist nonzero prime ideals p1 , . . . , pk of A with p1 · · · pk ⊆ (a). By Lemma 11.9.6b, we
may without loss of generality suppose that pk = p. By the minimality of k, we may choose
/ (a). Then a−1 b ∈
b ∈ p1 · · · pk−1 be such that b ∈ / A, but we have
a−1 bp ⊆ a−1 p1 · · · pk ⊆ A,
which implies that a−1 b ∈ p−1 . Moreover, if p−1 p = p, then a−1 bp ⊆ p. Since p is finitely
generated, Proposition 11.4.4 tells us that a−1 b is integral over A. But A is integrally closed,
so we have a contradiction. That is, we must have p ( p−1 p ⊆ A, from which it follows that
p−1 p = A by maximality of p. 
 11.9. DEDEKIND DOMAINS 357

T HEOREM 11.9.21. Let A be a Dedekind domain, and let a be a fractional ideal of A. Then
there exist k ≥ 0 and distinct nonzero prime ideals p1 , . . . , pk and r1 , . . . , rk ∈ Z − {0} such that
a = pr11 · · · prkk , and this decomposition is unique up to ordering. Moreover, a is an ideal of A if
and only if every ri is positive.
P ROOF. First suppose that a is a nonzero ideal of A. We work by induction on a minimal
nonnegative integer m such that there are nonzero prime ideals q1 , . . . , qm of a (not necessarily
distinct) with q1 · · · qm ⊆ a, which exists by Lemma 11.9.6a. If m = 0, then A ⊆ a, so a = A. In
general, for m ≥ 1, we know that a is proper, so there exists a nonzero prime ideal p that contains
a and p = qi for some i ≤ m. Without loss of generality, we take i = m. Then
q1 · · · qm−1 ⊆ q1 · · · qm p−1 ⊆ ap−1 ⊆ A.
By induction, there exist nonzero prime ideals q01 , . . . , q0` of A for some ` < m such that ap−1 =
q01 · · · q0` . The desired factorization is given by multiplying by p, applying Lemma 11.9.20, and
gathering together nondistinct primes.
In general, for a fractional ideal a, we let d ∈ A be such that da ⊆ A. We write da = q1 · · · qm
for some m ≥ 0 and prime ideals qi for 1 ≤ i ≤ m. We also write (d) = l1 · · · ln for some n ≥ 0
and prime ideals li for 1 ≤ i ≤ n. By Lemma 11.9.20, we then have
a = (d)−1 (da) = l−1 −1
1 · · · ln q1 · · · qm .

If qi = l j for some i and j, then we may use Lemma 11.9.20 to remove qi l−1 j from the product.
Hence we have the desired factorization.
Now suppose that
a = pr11 · · · prkk
for some k ≥ 0, distinct primes p1 , . . . , pk and nonzero r1 , . . . , rk . For each prime p of A, consider
the localization Ap , which is a Dedekind domain with unique nonzero prime ideal pAp . Note that
qAp = Ap if q is a nonzero prime of A other than p. We therefore have
aAp = pr11 . . . , prkk Ap = pr Ap ,
where r = ri if p = pi for some i, and r = 0 otherwise. Moreover, if pk Ap = pl Ap for some integers
k ≤ l, then pl−k Ap = Ap , which since p is nonzero, can only happen if k = l. Therefore, the primes
pi and corresponding integers ri are uniquely determined by a. 
We have the following immediate corollary of Theorem 11.9.21.
C OROLLARY 11.9.22. The set of fractional ideals I(A) of a Dedekind domain A is a group
under multiplication of fractional ideals with identity A, the inverse of a ∈ I(A) being a−1 .
D EFINITION 11.9.23. Let A be a Dedekind domain. The group I(A) of fractional ideals of A
is called the ideal group of A.
D EFINITION 11.9.24. Let A be a Dedekind domain. Then we let P(A) denote the set of its
principal fractional ideals. We refer to this as the principal ideal group.
C OROLLARY 11.9.25. Let A be a Dedekind domain. The group P(A) is a subgroup of I(A).
358 11. COMMUTATIVE ALGEBRA

D EFINITION 11.9.26. The class group (or ideal class group) of a Dedekind domain A is
Cl(A) = I(A)/P(A), the quotient of the ideal group by the principal ideal group.
L EMMA 11.9.27. A Dedekind domain A is a PID if and only if Cl(A) is trivial.
P ROOF. Every element of I(A) has the form ab−1 where a and b are nonzero ideals of A. If
A is a PID, then both a and b are principal and, therefore, so is ab−1 . On the other hand, if a is a
nonzero ideal of A with a = (a) for some a ∈ K, then clearly a ∈ A, so Cl(A) being trivial implies
that A is a PID. 
N OTATION 11.9.28. Let K be a number field. We let IK , PK , and ClK denote the ideal group,
principal ideal group, and class group of OK , respectively. We refer to these as the ideal group
of K, the principal ideal group of K, and the class group of K, respectively.
√ √ √
E XAMPLE 11.9.29. Let K = Q( −5). Then OK = Z[ −5]. √ The ideal a = (2, 1 + −5) is
non-principal. To see this, note that NK/Q (2) = 4 and NK/Q (1 + −5) = 6, so any generator x
of a must satisfy NK/Q (x) ∈ {±1, ±2}. But
√
NK/Q (a + b −5) = a2 + 5b2
√
for a, b ∈ Z, which forces√x = ±1. This would mean √ that a = Z[ −5]. To see that this can-
not happen, define φ : Z[ −5] → Z/6Z by φ (a + b −5) = a − b for a, b ∈ Z. This is a ring
homomorphism as
√ √ √
φ ((a + b −5)(c + d −5)) = φ (ac − 5bd + (ad + bc) −5) = ac − 5bd − ad − bc
= ac + bd − ad − bc = (a − b)(c − d).
√ √
Moreover, φ (1 + −5) = 0, so the kernel of φ contains (and is in fact equal to) (1 + −5).
Therefore, φ induces a surjection (in fact, isomorphism),
√
Z[ −5]/a → Z/6Z/(2) → Z/2Z,
√
so a 6= Z[ −5], and x does not exist. Therefore, ClQ(√−5) is nontrivial.
We end with the following important theorem.
T HEOREM 11.9.30. A Dedekind domain is a UFD if and only if it is a PID.
P ROOF. We need only show that a Dedekind domain that is a UFD is a PID. Let A be such
a Dedekind domain. By Theorem 11.9.21, it suffices to show that each nonzero prime ideal p
of A is principal. Since p is prime and A is a UFD, any nonzero element of p is divisible by an
irreducible element in p. If π is such an element, then (π) is maximal and contained in p, so
p = (π). 

11.10. Discrete valuation rings

D EFINITION 11.10.1. A discrete valuation ring, or DVR, is a principal ideal domain that has
exactly one nonzero prime ideal.
L EMMA 11.10.2. The following are equivalent conditions on a principal ideal domain A.
 11.10. DISCRETE VALUATION RINGS 359

i. A is a DVR,
ii. A has a unique nonzero maximal ideal,
iii. A has a unique nonzero irreducible element up to associates.
P ROOF. This is a simple consequence of the fact that in a PID, every nonzero prime ideal is
maximal generated by any irreducible element it contains. 
D EFINITION 11.10.3. A uniformizer of a DVR is a generator of its maximal ideal.
Moreover, we have the following a priori weaker but in fact equivalent condition for a domain
to be a DVR.
P ROPOSITION 11.10.4. A domain A is a DVR if and only if it is a local Dedekind domain
that is not a field.
P ROOF. A DVR is a PID, hence a Dedekind domain, and it is local by definition. Conversely,
suppose that A is noetherian, integrally closed, and has a unique nonzero prime ideal p. We must
show that A is a PID. Since nonzero ideals factor uniquely as products of primes in A, every ideal
of A has the form pn for some n. In particular, p = (π) for any π ∈ p − p2 , and then pn = (π n ) for
all n. Therefore, A is a PID and hence a DVR. 
T HEOREM 11.10.5. A noetherian domain A is a Dedekind domain if and only if its localiza-
tion at every nonzero prime ideal is a DVR.
P ROOF. We have seen in Proposition 11.9.5 that Ap is a Dedekind domain for all nonzero
prime ideals p. By Proposition 11.10.4, each such localization is therefore a DVR.
Conversely, if A is a noetherian integral domain
T
such that Ap is a DVR for every nonzero
prime ideal p, we consider the intersection B = p Ap over all nonzero prime ideals p of A, taken
inside the quotient field K of A. Clearly, B contains A, and if dc ∈ B for some c, d ∈ A, then we set
a = {a ∈ A | ac ∈ (d)}.
By definition of B, we may write = rs with r ∈ A and s ∈ A − p, and we see that sc = rd, so
c
d
s ∈ a. In other words, we have a 6⊆ p for all prime ideals p of A, which forces a = A. This implies
that c ∈ (d), so dc ∈ A.
Next, suppose that q is a nonzero prime ideal of A, and let m be a maximal ideal containing
it. Then qAm is a nonzero prime ideal of Am , which is a DVR, so qAm = mAm . Since q and m are
prime ideals contained in m, we therefore have
q = A ∩ qAm = A ∩ mAm = m.
Thus, A has Krull dimension at most 1.
Finally, each Ap is integrally closed in K by Corollary 11.4.24, and then the intersection A is
as well, since any element of K that is integral over A is integral over each Ap , hence contained
in each Ap . That is, A satisfies the conditions in the definition of a Dedekind domain. 
To make some sense of the name “discrete valuation ring”, we define the notion of a discrete
valuation. For this purpose, we adjoin an element ∞ to Z which is considered larger than any
element of Z, and we set x + y = ∞ if x, y ∈ Z ∪ {∞} and either x or y equals ∞.
360 11. COMMUTATIVE ALGEBRA

D EFINITION 11.10.6. Let K be a field. A discrete valuation on K is a surjective map v : K →

Z ∪ {∞} such that
i. v(a) = ∞ if and only a = 0,
ii. v(ab) = v(a) + v(b), and
iii. v(a + b) ≥ min(v(a), v(b))
for all a, b ∈ K.
D EFINITION 11.10.7. If v is a discrete valuation on a field K, then the quantity v(a) for a ∈ K
is said to be the valuation of a with respect to v.
The following are standard examples of discrete valuations.
E XAMPLE 11.10.8. Let p be a prime number. Then the p-adic valuation v p on Q is defined
by v p (0) = ∞ and v p (a) = r for a ∈ Q× if a = pr a0 for some r ∈ Z and a0 ∈ Q× such that p
divides neither the numerator nor denominator of a0 in reduced form.
E XAMPLE 11.10.9. Let F be a field, and consider the function field F(t). The valuation at ∞
on F(t) is defined by v∞ ( gh ) = deg h − deg g for g, h ∈ F[t] with h 6= 0, taking deg 0 = ∞.
More generally, we have the following.
D EFINITION 11.10.10. Let A be a Dedekind domain with quotient field K, and let p be a
nonzero prime ideal of A. The p-adic valuation vp on K is defined on a ∈ K × as the unique
integer such that (a) = pv p (a) bc−1 for some nonzero ideals b and c of A that are not divisible by
p.
E XAMPLE 11.10.11. For the valuation at ∞ on F(t), where F is a field, we may take A =
K[t −1 ]and p = (t −1 ). Then the valuation v∞ on F(t) is the (t −1 )-adic valuation. To see this, note
that for nonzero g, h ∈ F[t], one has
g(t) G(t −1 )
= (t −1 )deg h−deg g ,
h(t) H(t −1 )
where G(t −1 ) = t − deg g g(t) and H(t −1 ) = t − deg h h(t) are polynomials in t −1 which have nonzero
constant term.
L EMMA 11.10.12. Let A be a Dedekind domain with quotient field K, and let p be a prime
ideal of A. The p-adic valuation on K is a discrete valuation.
P ROOF. Let a, b ∈ K be nonzero (without loss of generality). Write (a) = pr a and (b) = ps b
for r = vp (a) and s = vp (b) and fractional ideals a and b of A. Note that (ab) = pr+s ab, so
v p (ab) = r + s. We have
(a + b) = pr a + ps b = pmin(r,s) (pr−min(r,s) a + ps−min(r,s) b),
so
vp (a + b) = min(r, s) + vp (pr−min(r,s) a + ps−min(r,s) b) ≥ min(r, s).

 11.11. RAMIFICATION OF PRIMES 361

L EMMA 11.10.13. Let v be a discrete valuation on a field K. Then we have v(−a) = v(a) for
all a ∈ K.
P ROOF. Note that 2v(−1) = v(1) = 0, so we have v(−a) = v(−1) + v(a) = v(a). 
L EMMA 11.10.14. Let v be a discrete valuation on a field K. Then we have
v(a + b) = min(v(a), v(b))
for all a, b ∈ K with v(a) 6= v(b).
P ROOF. If v(a) < v(b), then
v(a) = v((a + b) − b) ≥ min(v(a + b), v(b)) ≥ min(v(a), v(b)) = v(a),
so we have v(a) = min(v(a + b), v(b)), which forces v(a + b) = v(a). 
D EFINITION 11.10.15. Let K be a field, and let v be a discrete valuation on K. Then
Ov = {a ∈ K | v(a) ≥ 0}
is called the valuation ring of v.
L EMMA 11.10.16. Let K be a field, and let v be a discrete valuation on K. Then Ov is a DVR
with maximal ideal
mv = {a ∈ K | v(a) ≥ 1}.
P ROOF. That Ov is a ring follows from the fact that if a, b ∈ Ov , then v(ab) = v(a) + v(b) ≥ 0,
v(−a) = v(a) ≥ 0, and v(a + b) ≥ min(v(a), v(b)) ≥ 0. For a ∈ Ov and x, y ∈ mv , we have
v(x + y) ≥ min(v(x), v(y)) ≥ 1 and v(ax) = v(a) + v(x) ≥ 1, so mv is an ideal. It is also the unique
maximal ideal: given a ∈ Ov −mv , we have v(a−1 ) = v(a)+v(a−1 ) = v(1) = 0, so a ∈ Ov× . Given
an ideal a of Ov , let a ∈ a be an element of minimal valuation n. Let π ∈ Ov with v(π) = 1, and
write a = π n u for some u ∈ Ov× . Then v(u) = 0, so u ∈ Ov× . Therefore, (π n ) ⊆ a. On the other
hand, since n is minimal, we have a ⊆ (π n ), and therefore a is a principal. By Lemma 11.10.2,
we conclude that Ov is a DVR. 
E XAMPLE 11.10.17. In Q, we have
na o
Ov p = Z(p) = | a, b ∈ Z such that p - b .
b

11.11. Ramification of primes

The integral closure B of a Dedekind domain A in a finite extension L of its quotient field K
is also a Dedekind domain. If p is a nonzero prime ideal of A, then we can consider the ideal pB
of B. This ideal may no longer be prime. Instead, it has a factorization
e
(11.11.1) pB = Pe11 · · · Pgg
for some distinct nonzero prime ideals Pi of B and positive integers ei , for 1 ≤ i ≤ g for some
g ≥ 1. We make the following definitions.
362 11. COMMUTATIVE ALGEBRA

D EFINITION 11.11.1. Let B/A be an extension of commutative rings. We say that a prime
ideal P of B lies over (or above) a prime ideal p of A if p = P ∩ A. We then say that p lies under
(or below) P.
In (11.11.1), the prime ideals of B lying over p are exactly the Pi for 1 ≤ i ≤ g.
D EFINITION 11.11.2. Let A be a Dedekind domain, and let B be the integral closure of A in
a finite extension L of the quotient field K of A. Let p be a nonzero prime ideal of A.
a. We say that p ramifies (or is ramified) in L/K if pB is divisible by the square of a prime
ideal of B. Otherwise, it is said to be unramified.
b. We say that p is inert in L/K if pB is a prime ideal.
c. We say that p is split in L/K if there exist two distinct prime ideals of B lying over p.
Otherwise, p is non-split.
It follows directly that p is ramified in L/K if some ei in (11.11.1) is at least 2. On the other
hand, p is inert in L/K if there is exactly one prime ideal of B lying over p and its ramification
index is 1, which is to say that g = 1 and e1 = 1 in (11.11.1). Finally, p is split in L/K if g > 1.
√
E XAMPLE 11.11.3. Let A = Z and L = Q( 2). The integral closure of A in L is B = OL =
√ √
Z[ 2]. The prime p = (2) ramifies in Q( 2)/Q, since
√ √
2Z[ 2] = ( 2)2 .
√ √ √ √
Moreover, √ P = ( 2) is a prime ideal of Z[ 2], since Z[ 2]/( 2) ∼ = Z/2Z via the map that
takes a + b 2 to a mod 2. Therefore, p is ramified and non-split.
√ √
∼ ∼
√ consider the prime ideal (3) of Z. We have Z[ 2]/(3)
Next, √ = F3 [ 2] = F9 , so (3) is inert
in Q( 2)/Q. On the other hand, the prime factorization of 7Z[ 2] is exactly
√ √ √
7Z[ 2] = (3 + 2)(3 − 2),
√ √ √
since Z[ 2]/(3√ ± 2) is isomorphic to Z/7Z via the map that takes a + b 2 to a ∓ 3b. That is,
(7) splits in Q( 2)/Q.
D EFINITION 11.11.4. Let A be a Dedekind domain, and let p be a nonzero prime ideal of A.
The residue field of p is A/p.
R EMARK 11.11.5. Let A be a Dedekind domain, and let B be the integral closure of A in a
finite extension L the quotient field K of A. Let p be a nonzero prime ideal of A, and let P be a
prime ideal of L lying over K. Then B/P is a field extension of A/p via the natural map induced
on quotients by the inclusion A ,→ B.
D EFINITION 11.11.6. Let A be a Dedekind domain, and let B be the integral closure of A in
a finite extension L of the quotient field K of A. Let p be a nonzero prime ideal of A, and let P
be a prime ideal of B lying over p.
a. The ramification index eP/p of P over p is the largest e ≥ 1 such that Pe divides pB.
b. The residue degree fP/p of a prime ideal of P lying over p is [B/P : A/p].
 11.11. RAMIFICATION OF PRIMES 363

R EMARK 11.11.7. It follows quickly from the definitions that ramification indices and residue
degrees are multiplicative in extensions. That is, if A ⊆ B ⊆ C are Dedekind domains with the
quotient field of C a finite extension of that of A and P is a prime ideal of C lying over P of B
and p of A, then
eP/p = eP/P eP/p and fP/p = fP/P fP/p .
√
E XAMPLE√11.11.8. In Example 11.11.3, the residue degree of√( 2) over 2Z is 1, the residue
degree of 3Z[ 2] over 3Z is 2, and the residue degrees of (3 ± 2) over 7Z are each 1. The
ramification indices are 2, 1, and 1, repsectively.
We shall require the following lemmas.
L EMMA 11.11.9. Let p be a nonzero prime ideal in a Dedekind domain A. For each i ≥ 0,
the A/p-vector space pi /pi+1 is one-dimensional.
P ROOF. Let x ∈ pi − pi+1 for some i ≥ 0. (Such an element exists by unique factorization of
ideals.) We need only show that the image of x spans pi /pi+1 . For this, note that (x) = pi a for
some nonzero ideal of A not divisible by p. Then
(x) + pi+1 = pi (a + p) = pi ,
the last step by the Chinese remainder theorem. 
L EMMA 11.11.10. Let A be a Dedekind domain and P be a set of nonzero prime ideals of A.
Let S a multiplicatively closed subset of A such that S ∩ p = ∅ for all p ∈ P. Let a be a nonzero
ideal of A that is divisible only by prime ideals in P. Then the natural map
A/a → S−1 A/S−1 a
is an isomorphism.
P ROOF. Suppose that b ∈ S−1 a ∩ A, and write b = as for some a ∈ a and s ∈ S. Then a = bs,
and since a divides (a) while (s) is relatively prime to a, we must have that a divides (b). In
other words, b ∈ a, and therefore the map is injective. Given c ∈ A and t ∈ S, the ideals (t) and a
have no common prime factor, so in that A is a Dedekind domain, satisfy (t) + a = A. Thus, there
exists u ∈ A such that ut − 1 ∈ a. Then cu + a maps to ct + S−1 a, so the map is surjective. 
The ramification indices and residue degrees of the primes over p satisfy the following degree
formula.
T HEOREM 11.11.11. Let A be a Dedekind domain, and let B be the integral closure of A in
a finite separable extension L the quotient field K of A. Let p be a nonzero prime ideal of A, and
write
e
pB = Pe11 · · · Pgg
for some distinct nonzero prime ideals Pi of B and positive integers ei , for 1 ≤ i ≤ g and some
g ≥ 1. For each i, let fi = fPi /p . Then
g
∑ ei fi = [L : K].
i=1
364 11. COMMUTATIVE ALGEBRA

P ROOF. We prove that dimA/p B/pB equals both quantities in the desired equality. By the
Chinese remainder theorem, we have a canonical isomorphism
g
B/pB ∼ e
= ∏ B/Pi i ,
i=1
of A/p-vector spaces, so
g g ei −1
dimA/p B/pB = ∑ dimA/p B/Pei i =∑ ∑ dimA/p Pij /Pij+1.
i=1 i=1 j=0
j j+1
By Lemma 11.11.9, each Pi /Pi is a 1-dimensional B/Pi -vector space, and we therefore have
g g
dimA/p B/pB = ∑ ei dimA/p B/Pi = ∑ ei fi .
i=1 i=1
Let S denote the complement of p in A. Then S−1 A = Ap and S−1 B are Dedekind domains,
and Ap is a DVR, hence a PID. Moreover, S−1 B is the integral closure of Ap in L, being both
integrally closed and contained in said integral closure. Thus, Corollary 11.4.40 tells us that S−1 B
is free of rank [L : K] over Ap . In particular, S−1 B/pS−1 B is an [L : K]-dimensional Ap /pAp -vector
space. On the other hand, note that
S ∩ Pi = S ∩ A ∩ Pi = S ∩ p = ∅
for each 1 ≤ i ≤ g. Therefore, Lemma 11.11.10 tells us that
S−1 B/pS−1 B =∼ B/pB
and Ap /pAp ∼
= A/p. We thus have that dimA/p B/pB = [L : K], as required. 
In other words, Theorem 11.11.11 tells us that the sum over all primes lying over p of the
products of their ramification indices with their residue degrees equals the degree of the field
extension L/K.
 CHAPTER 12

Homological algebra

We work in this chapter largely in an abelian category. At times, proofs of statements that
hold true in arbitrary abelian categories will be given only in categories of modules over a ring.
This choice, which simplifies the exposition, will be given a more rigorous justification in the
course of the chapter.

12.1. Exact sequences

Though we’ve managed to suppress them to this point, exact sequences are ubiquitous in
algebra. Let’s give the definitions.
D EFINITION 12.1.1. Let I be the set of integers in an interval in R. A diagram A· in a category
C of the form
A
di+1 i dA
· · · → Ai+1 −−→ Ai −→ Ai−1 → · · ·
is a sequence, where the Ai are defined for i ∈ I and the morphisms diA : Ai → Ai−1 are defined
for i ∈ I with i − 1 ∈ I. We will refer to I as the defining interval of the sequence A· .
N OTATION 12.1.2. In an abelian category, if A is a subobject of an object B, we write A ⊆ B
to denote this and B/A for the cokernel of the inclusion morphism A → B. For f : B → C, we
will let f (A) denote the image of the composite of the inclusion with f .
We are particularly interested in exact sequences.
D EFINITION 12.1.3. We say that a diagram
f g
A−
→B→
−C
in an abelian category C is exact if g ◦ f = 0 and the induced monomorphism im f → ker g is an
isomorphism.
D EFINITION 12.1.4. A sequence A· = (Ai , diA ) with defining interval I in an abelian category
C is exact, or an exact sequence, if the subdiagram
di di−1
Ai+1 −
→ Ai −−→ Ai−1
is exact for each i ∈ (I + 1) ∩ (I − 1).
That is, A· is exact if di ◦ di+1 = 0 and the canonical morphism im di+1 → ker di is an isomor-
phism for all i for which di+1 and di are defined: we write this more simply as the identification
im di+1 = ker di for subobjects of Ai .
365
366 12. HOMOLOGICAL ALGEBRA

R EMARK 12.1.5. One can make the same definitions of exact sequences in the category
of groups, or more generally in any “semi-abelian” category, and much of the discussion that
follows remains the same.
R EMARK 12.1.6. If the interval I of definition of A· has a left (resp., right) endpoint N such
that AN = 0, then one can extend A· to the left (resp., right) by taking Ai = 0 for all i < N (resp.,
i > N). In fact, we could do this for any sequence for which I has an endpoint (without the
condition AN = 0), but we do not as the operation does not preserve exactness (nor do the to-be-
defined morphisms between two sequences defined on different intervals extend to morphisms
under this operation).
D EFINITION 12.1.7. Let C be an abelian category.
a. A short exact sequence in C is an exact sequence in C of the form
0 → A → B → C → 0.
b. A left short exact sequence in C is an exact sequence in C of the form
0 → A → B → C.
c. A right short exact sequence in C is an exact sequence in C of the form
A → B → C → 0.
R EMARK 12.1.8. To say that
f g
0→A−
→B→
− C→0
is exact is to say that f is a monomorphism, im f = ker g, and g is an epimorphism.
E XAMPLE 12.1.9. Multiplication-by-n provides a short exact sequence of abelian groups
n
0→Z→
− Z → Z/nZ → 0.
R EMARK 12.1.10. If f : A → B is any morphism in an abelian category, then we have an
exact sequence
f
0 → ker f → A −→ B → coker f → 0.
Note that this provides two short exact sequences
0 → ker f → A → im f → 0 and 0 → im f → B → coker f → 0,
since coim f → im f is an isomorphism. We can “splice these back together” to get the 4-term
sequence by taking the composite A → im f → B, which is f .
D EFINITION 12.1.11. A long exact sequence in an abelian category C is an exact sequence
A· = (Ai , diA )i∈Z .
Frequently, a long exact sequence is expressed in the form
· · · → A2 → A1 → A0 → 0,
and we can extend it to all integers by setting Ai = 0 for all i ≤ −1.
 12.2. THE SNAKE AND FIVE LEMMAS 367

E XAMPLE 12.1.12. The sequence

f f f
· · · → Q2 −
→ Q2 −
→ Q2 −
→ Q2 → · · ·

with f : Q2 → Q2 defined by f (a, b) = (0, a) for a, b ∈ Q is a long exact sequence of Q-vector

spaces. The sequence
ι f f
0→Q−
→2
Q2 −
→ Q2 −
→ Q2 → · · ·
of Q-vector spaces with ι2 (b) = (0, b) is also a long exact sequence.

We next study maps between sequences. Let us make a formal definition.

D EFINITION 12.1.13. Let A· = (Ai , diA ) and B· = (Bi , diB ) be sequences in a category C with
defining intervals I and J, respectively. A morphism of sequences f· : A· → B· in a category
is a collection ( fi )i∈I∩J of morphisms fi : Ai → Bi in C such that diB ◦ fi = fi−1 ◦ diA for all
i ∈ I ∩ J ∩ (I + 1) ∩ (J + 1).

R EMARK 12.1.14. We can view the condition for a sequence of maps fi : Ai → Bi between
the terms of sequences A· and B· defined at all i ∈ Z to be a map of sequences as saying that the
diagram
A
di+1 diA
··· / Ai+1 / Ai / Ai−1 / ···
fi+1 fi fi−1
 B
di+1  diB 
··· / Ai+1 / Ai / Ai−1 / ···

commutes.

12.2. The snake and five lemmas

The following result on maps between short exact sequences is the key to much of homolog-
ical algebra.

T HEOREM 12.2.1 (Snake lemma). Let C be an abelian category, and let

0 /A f
/B g
/C /0
α β γ
 f0  g0 
0 / A0 / B0 / C0 / 0.

be a commutative diagram in C with exact rows. Then there is an exact sequence

→ coker α → coker β → coker γ → coker g0 → 0

δ
0 → ker f → ker α → ker β → ker γ −
368 12. HOMOLOGICAL ALGEBRA

such that the resulting diagram

/ ker α / ker β / ker γ

   δ
0 / ker f /A f
/B g
/C /0
α β γ

 f0  g0 
0 / A0 / B0 / C0 / coker g0 /0
O

  
/ coker α / coker β / coker γ

with the natural inclusion and quotient maps is commutative.

P ROOF. We work in the category of modules over a ring R. We define δ as follows. For
c ∈ ker γ, find b ∈ B with g(b) = c. Then g0 ◦ β (b) = γ(c) = 0, so β (b) = f 0 (a0 ) for some a0 ∈ A0 .
Let δ (c) denote the image ā0 of a0 in coker α. To see that this is well-defined, note that if b2 ∈ B
also satisfies g(b2 ) = c, then g(b − b2 ) = 0, so b2 − b = f (a) for some a ∈ A. We then have
β (b2 ) = β (b) + β ◦ f (a) = β (b) + f 0 ◦ α(a),
so b2 = f 0 (a0 + α(a)). But a0 + α(a) has image ā0 in coker α, so δ is well-defined. That δ is an
R-module homomorphism follows easily from the construction.
We now check that the other maps are well-defined. Since
β ◦ f (ker α) = f 0 ◦ α(ker α) = 0,
we have f (ker α) ⊆ ker β . Similarly, g(ker β ) ⊂ ker γ. Also, if ā0 ∈ coker α, then we may lift
it to a0 ∈ A0 , map to b0 ∈ B0 , and then project to b̄0 ∈ coker β . This is well-defined as any other
choice of a0 differs by some a ∈ A, which causes b̄0 to change by the image of β ( f (a)), which is
zero. Thus f 0 induces a well-defined homomorphism
f¯0 : coker α → coker β .
Similarly, we have a well-defined surjection
ḡ0 : coker β → coker γ.
We next check that our sequence is a complex. Note that g ◦ f = 0, so the same is true on
ker α, and g0 ◦ f 0 = 0, so ḡ0 ◦ f¯0 = 0 as well. Let b ∈ ker β . Then δ (g(b)) is given by considering
β (b) = 0, lifting it to some a0 ∈ A0 , which we may take to be 0, and projecting to coker α. Hence
δ (g(ker β )) = 0. On the other hand, if c ∈ ker γ, then f¯0 (δ (c)) is given by definition by projecting
β (b) to coker β , where g(b) = c, hence is zero. Hence, the image of one map is contained in the
kernel of the next at each term of the six term sequence.
Finally, we check exactness at each term. If a ∈ ker α, then f (a) = 0 implies a ∈ ker f .
Inclusion then provides a map ker f → ker α that is by definition injective. If b ∈ ker β ∩ ker g,
 12.2. THE SNAKE AND FIVE LEMMAS 369

then there exists a ∈ A with f (a) = b. Since f 0 ◦ α(a) = β ◦ f (a) = 0 and f 0 is injective, we have
α(a) = 0, or a ∈ ker α. Hence
f (ker α) = ker(ker β → ker γ),
and we have exactness at ker β .
If c ∈ ker δ , then whenever g(b) = c and f 0 (a0 ) = β (b), we have ā0 = 0, letting ā0 denote the
image of a0 ∈ coker α. We then have a0 = f (a) for some a ∈ A, so b2 = b − f (a) still satisfies
f (b2 ) = c, but β (b2 ) = 0. So b2 ∈ ker β , and we have exactness at ker γ.
If ā0 ∈ coker α is the image of a0 ∈ A0 and f¯0 (ā0 ) = 0, then there exists b ∈ B with β (b) =
f 0 (a0 ). Now
γ(g(b)) = g0 (β (b)) = g0 ( f 0 (a0 )) = 0,
so g(b) ∈ ker γ, and δ (g(b)) = ā0 . Hence, we have exactness at coker α.
If b̄0 ∈ coker β is the image of b0 ∈ B0 and ḡ0 (b̄0 ) = 0, then there exists c ∈ C with g(b0 ) = γ(c).
Now c = g(b) for some b ∈ B. And b02 = b0 − b has image b̄0 in coker β . On the other hand,
f 0 (b02 ) = 0, so b02 = f 0 (a0 ) for some a0 ∈ A0 . If ā0 ∈ coker α is the image of a0 , then f¯0 (ā0 ) = b̄0
as the image of b02 in coker β . Thus, we have exactness at coker β . Finally, if c̄0 ∈ coker γ is the
image of c0 ∈ C0 with trivial image in coker g0 , then c0 = g0 (b0 ) for some b0 ∈ B0 , then the image
b̄0 ∈ coker β of b0 satisfies ḡ0 (b̄0 ) = c̄0 . 
Next, we state another useful result on maps between exact sequences, known as the five
lemma.
T HEOREM 12.2.2 (Five lemma). Let

A
e /B f
/C g
/D h /E
α β γ δ ε
  f0  g0  
e0 / B0 / C0 / D0 h0 / E 0.
A0
be a commutative diagram with exact rows in an abelian category C .
a. If β and δ are epimorphisms and ε is a monomorphism, then γ is an epimorphism.
b. If β and δ are monomorphisms and α is an epimorphism, then γ is a monomorphism.
c. If β and δ are isomorphisms, α is an epimorphism, and ε is a monomorphism, then γ is
an isomorphism.
P ROOF. We work in the category of modules over a ring R. It is immediate that parts a and
b imply part c (the actual five lemma). We prove part a and note that it, if proven in an arbitrary
abelian category, implies b in the opposite category, which is also abelian. Suppose that β and
δ are surjective and ε is injective. Let c0 ∈ C0 , and note that g0 (c0 ) = δ (d) for some d ∈ D by
surjectivity of δ . Also,
ε(h(d)) = h0 (δ (d)) = h0 (g0 (c0 )) = 0,
so h(d) = 0 by injectivity of ε. By exactness of the top row at D, we then have c ∈ C such that
g(c) = d. Now,
g0 (γ(c) − c0 ) = δ (g(c)) − g0 (c0 ) = δ (d) − δ (d) = 0,
370 12. HOMOLOGICAL ALGEBRA

so by exactness of the bottom row at C, there exists b0 ∈ B0 such that f 0 (b0 ) = c0 − γ(c). As β is
surjective, there also exists b ∈ B such that β (b) = b0 . Set x = c + f (b) ∈ C. Then
γ(x) − γ(c) = γ( f (b)) = f 0 (β (b)) = f 0 (b0 ) = c0 − γ(c)
so γ(c) = c0 . Thus, γ is surjective and part a is proven. 

12.3. Homology and cohomology

D EFINITION 12.3.1. Let C be an abelian category.
a. A chain complex, or more simply complex, in C is a sequence A· = (Ai , diA )i∈Z in C such
A = 0 for all i ∈ Z.
that diA ◦ di+1
b. For a chain complex A· and i ∈ Z, the morphism diA : Ai → Ai−1 is called the ith differential
in the complex A· .
N OTATION 12.3.2. Unless otherwise specified, the ith object in a chain complex A· will be
denoted Ai and the ith differential by di : Ai → Ai−1 . If we have multiple complexes, we will use
diA to specify the differential on A· .
R EMARK 12.3.3. Unlike with sequences in general (or exact sequences in particular), if
the terms and morphisms of complex are specified only for some interval of integers, then we
complete it to a complex by declaring all remaining objects and morphisms to be zero.
D EFINITION 12.3.4. A morphism of complexes in an abelian category is a morphism of se-
quences between complexes.
D EFINITION 12.3.5. The category of chain complexes Ch(C ) for an abelian category C
is the category with objects the complexes (Ai , diA )i∈Z in C and morphisms the morphisms of
complexes in C .
R EMARK 12.3.6. A sequence of complexes is exact in Ch(C ) if and only if it the resulting
sequence of objects in each fixed degree is exact in C . For instance, a sequence of complexes
f· g·
0 → A· −
→ B· −
→ C· → 0
in Ch(C ) is short exact if and only if each
fi gi
0 → Ai −
→ Bi −
→ Ci → 0
is a short exact sequence.
Note that the category of chain complexes in C is a fully faithful subcategory of the category
of sequences with defining interval Z.
D EFINITION 12.3.7. A complex is often said to be acyclic if it is an exact sequence.
The reader can easily check the following.
P ROPOSITION 12.3.8. Let C be an abelian category. Then the category Ch(C ) is an abelian
category as well.
 12.3. HOMOLOGY AND COHOMOLOGY 371

D EFINITION 12.3.9. The ith homology of a complex A· in an abelian category is the object
ker diA
Hi (A) = A
.
im di+1
R EMARK 12.3.10. A complex A· is exact if and only if Hi (A) = 0 for all i ∈ Z.
E XAMPLE 12.3.11. The complex
4 4
· · · → Z/8Z →
− Z/8Z →
− Z/8Z → · · ·
of abelian groups has ith homology group 2Z/4Z ∼= Z/2Z for every i.
L EMMA 12.3.12. Any morphism f· : A· → B· of complexes in C induces natural morphisms
fi∗ : Hi (A) → Hi (B)
for each i ∈ Z. More specifically, these satisfy
ῑiB ◦ fi∗ ◦ π̄iA = πiB ◦ fi ◦ ιiA ,
where πiB : Bi → coker di+1
B and π̄ A : ker d A → H (A) are the canonical epimorphisms, and where
i i i
A A B B
ιi : ker di → Ai and ῑi : Hi (B) → coker di+1 are the canonical monomorphisms.
P ROOF. We prove this in the case that C is a category of R-modules. If a ∈ ker diA , then
A (a0 ), then f (a) = d B ( f
diB ( fi (a)) = fi−1 (diA (a)) = 0, so fi (a) ∈ ker diB . If a = di+1 0
i i+1 i+1 (a )) ∈
B . Thus, the composite map
im di+1
fi
→ ker diB  Hi (B)
ker diA −
factors through Hi (A), inducing the stated map fi∗ . 
T HEOREM 12.3.13. Let C be an abelian category, and let
f· g·
0 → A· −
→ B· −
→ C· → 0
be a short exact sequence in Ch(C ). There there are morphisms δi : Hi (C) → Hi−1 (A) for all
i ∈ Z, natural in the short exact sequence, that fit into a long exact sequence
f∗
i i i g∗ δ
· · · → Hi (A) −→ Hi (B) −→ Hi (C) −
→ Hi−1 (A) → · · · .
P ROOF. First, the snake lemma applied to the diagram

0 / Ai fi
/ Bi gi
/ Ci /0
diA diB diC
 fi−1  gi−1 
0 / Ai−1 / Bi−1 / Ci−1 /0

provides exact sequences

0 → ker diA → ker diB → ker diC and coker diA → coker diB → coker diC → 0.
372 12. HOMOLOGICAL ALGEBRA

Then, we wish to apply the snake lemma to the diagram

f¯i
A
coker di+1 / coker d B ḡi
/ coker dC /0
i+1 i

d¯iA d¯iB d¯iC

 fi−1
 gi−1

0 / ker d A / ker d B / ker dC
i−1 i−1 i−1

with exact rows (where the “bars” denote morphisms induced on quotients). By the snake lemma,
we have an exact sequence
δi
ker d¯iA → ker d¯iB → ker d¯iC −
→ coker d¯iA → coker d¯iB → coker d¯iC .
Note that for X ∈ {A, B,C}, we have
ker d¯iX = ker(Xi / im di+1
X
→ Xi−1 ) ∼
= Hi (A) and coker d¯iX ∼ X
= coker(Xi → ker di−1 )∼
= Hi−1 (A),
and the morphisms induced by fi and gi (resp., fi−1 and gi−1 ) on the first (resp., second) of these
are just the morphisms fi∗ and g∗i (resp., fi−1
∗ and g∗ ). Our exact sequence then becomes
i−1
∗ g∗i−1
i f∗ i g∗ i δ fi−1
Hi (A) −→ Hi (B) −→ Hi (C) −
→ Hi−1 (A) −−→ Hi−1 (B) −−→ Hi−1 (C).
Taken for all i, these yield the long exact sequence. 
E XAMPLE 12.3.14. Consider the complexes A· , B· , and C· of abelian groups with Ai = Z/4Z,
Bi = Z/8Z, and Ci = Z/2Z for all i and diA = diB = 0 for all i and diC equal to multiplication by
4 for all i. Then Hi (A) = Z/4Z, Hi (B) = 2Z/4Z ∼
= Z/2Z, and Hi (C) = Z/2Z for all i. We have
a short exact sequence
f· g·
0 → A· −
→ B· −
→ C· → 0
for maps f· : A· → B· induced by multiplication by 2 on each term and g· : B· → C· given by
reduction modulo 2 on each term. The resulting long exact sequence has the form
δ 0 δ 0 δ
··· −
→ Z/4Z  Z/2Z →
− Z/2Z ,→ Z/4Z  Z/2Z →
− Z/2Z −
→ ··· .
We briefly describe cochain complexes and cohomology, which simply amount to a change
of indexing from decreasing to increasing.
D EFINITION 12.3.15. A cochain complex is a collection A· = (Ai , dAi )i∈Z of objects Ai and
morphisms dAi : Ai → Ai+1 such that dAi ◦ dAi−1 = 0 for all i ∈ Z. The morphism dAi : Ai → Ai+1 is
called the ith differential of the cochain complex A· .
We then have the notion of cohomology.
D EFINITION 12.3.16. The ith cohomology of a cochain complex A· in an abelian category is
the object
ker dAi
H i (A) = .
im dAi−1
 12.4. PROJECTIVE AND INJECTIVE OBJECTS 373

R EMARK 12.3.17. Much as with complexes, we can speak of morphisms of cochain com-
plexes f · : A· → B· , which are collections of morphisms f i : Ai → Bi such that dBi ◦ f i = f i+1 ◦ dAi
for all i ∈ Z. Again, short exact sequences
f· g·
0 → A· − → C· → 0
→ B· −
of cochain complexes give rise to long exact sequences in cohomology, but now of the form
fi gi δi
· · · → H i (A) −
→∗ ∗
H i (B) −→ → H i+1 (A) → · · ·
H i (C) −
D EFINITION 12.3.18. Let A· = (Ai , diA ) and B· = (Bi , diB ) be chain complexes. Let f· , g· : A· →
B· be morphisms of chain complexes.
a. A chain homotopy from f· to g· is a sequence s· = (si )i∈Z of morphisms si : Ai → Bi+1
satisfying
B
fi − gi = di+1 ◦ si + si−1 ◦ diA
for all i ∈ Z.
b. We say that f· and g· are chain homotopic, and write f· ∼ g· , if there exists a homotopy
from f· to g· .
c. If f· is (chain) homotopic to 0, then f· is said to be null-homotopic.
R EMARK 12.3.19. For cochain complexes A· and B· and morphisms f · , g· : A· → B· , a chain
homotopy from f · to g· is a sequence s· = (si )i∈Z of morphisms si : Ai → Bi−1 such that f i − gi =
dBi−1 ◦ si + si+1 ◦ dAi .
The morphisms s· defining a null-homotopy fit into a diagram
A
di+1 diA
··· / Ai+1 / Ai / Ai−1 / ···
si si−1
fi+1 fi fi−1

 ~ B
di+1  ~ diB 
··· / Bi+1 / Bi / Bi−1 / ··· .

P ROPOSITION 12.3.20. Assume that f· and g· are chain homotopic morphisms A· → B· . Then
the morphisms fi∗ and g∗i on homology are equal for all i ∈ Z.
D EFINITION 12.3.21. A morphism of complexes f· : A· → B· is a homotopy equivalence if
there exists a morphism g· : B· → A· such that g· ◦ f· ∼ idA· and f· ◦ g· ∼ idB· .

12.4. Projective and injective objects

We continue to work in an abelian category C .
D EFINITION 12.4.1.
a. We say that an epimorphism g : B → C is split if there exists a morphism t : C → B with
g ◦ t = idC . In this case, we say that t is a splitting of g.
374 12. HOMOLOGICAL ALGEBRA

b. We say that a monomorphism f : A → B is split if there exists a morphism s : B → A with

s ◦ f = idA . In this case, we say that s is a splitting of f .
c. We say that a short exact sequence
f g
(12.4.1) 0→A− →B→ − C→0
splits if there exists an isomorphism w : A ⊕C −∼
→ B with w(a, 0) = f (a) and g(w(0, c)) = c for
all a ∈ A and c ∈ C.
E XAMPLE 12.4.2. The exact sequence of abelian groups
·2 mod 2
0 → Z/3Z −
→ Z/6Z −−−−→ Z/2Z → 0
is split, but
·2 mod 2
0 → Z/2Z −
→ Z/4Z −−−−→ Z/2Z → 0
is not.
P ROPOSITION 12.4.3. The following conditions on a short exact sequence
f g
0→A−
→B→
− C→0
are equivalent:
i. The sequence splits.
ii. The monomorphism f : A → B splits.
iii. The epimorphism g : B → C splits.
P ROOF. We prove this in the category of R-modules.
(iii) ⇒ (ii): Suppose we have a splitting map t : C → B. Then define s : B → A by s(b) = a
where f (a) = b − t(g(b)). This is well-defined as f is injective, and such an a exists since
g(b − t(g(b))) = g(b) − g(t(g(b))) = g(b) − g(b) = 0.
It splits f as
s( f (a)) = s(b) − s(t(g(b))) = s(b),
the latter step using the fact that s ◦ t = 0, which follows in turn from
f (s(t(c))) = t(c) − t(g(t(c))) = t(c) − t(c) = 0.
(ii) ⇒ (iii): Suppose that we have a splitting map s : B → A. Then define t : C → B by t(c) =
b − f (s(b)) where g(b) = c. To see this is well defined, note that f (a) − f (s( f (a))) = 0 for any
a ∈ A.
(ii)+(iii) ⇒ (i): Define w(a, c) = f (a) + t(c). Its inverse is w0 (b) = (s(b), g(b)). To see this,
we check that
w0 ◦ w(a, c) = (s( f (a) + t(c)), g( f (a) + t(c))) = (s( f (a)), g(t(c))) = (a, c)
for a ∈ A and c ∈ C, and note that
w ◦ w0 (b) = w(s(b), g(b)) = f (s(b)) + t(g(b)).
 12.4. PROJECTIVE AND INJECTIVE OBJECTS 375

for b ∈ B. Set c = g(b), so that g(b − t(c)) = 0, and let a ∈ A be such that f (a) = b − t(c). Then
( f ◦ s + t ◦ g)(b) = f (s( f (a))) + f (s(t(c))) + t(g( f (a))) + t(g(t(c))) = f (a) + t(c) = b.
(i) ⇒ (iii) Set t(c) = w(0, c). Then g(t(c)) = g(w(0, c)) = c.

D EFINITION 12.4.4. An object P of an abelian category is projective if for epimorphism
π : B → C and every morphism g : P → C, there exists an morphism f : P → B such that g = π ◦ f .
R EMARK 12.4.5. The property of P being projective is represented by the existence of g in
the commutative diagram
P
f
g
 
B
π /C /0
with exact lower row.
P ROPOSITION 12.4.6. Free R-modules are projective.
P ROOF. Let F be a free R-module with basis X. Let π : B → C be a surjective R-module
homomorphism, and suppose that g : F → C is an R-module homomorphism. For each x ∈ X,
let b be an element of B such that π(b) = g(x). Since F is free, we may defined f : F → B by
f (x) = b for each x ∈ X. Then π( f (x)) = g(x) for all x ∈ X, so π ◦ f = g as X generates F. 
E XAMPLE 12.4.7. Not every projective module need be free. For example, consider R =
Z/6Z. We claim that P = Z/3Z is a projective R-module. To see this, suppose that B is a
Z/6Z-module and g : B → Z/3Z is surjective. Take any b ∈ B with g(b) = 1. Then the Z/6Z
submodule generated by b is isomorphic to Z/3Z, and hence 1 7→ b defines a splitting of g.
Note that P is not projective as a Z-module (abelian group) since the quotient map Z → Z/3Z
does not split. In fact, every projective Z-module is free.
We describe some equivalent conditions for projectivity.
P ROPOSITION 12.4.8. The following conditions on an R-module P are equivalent:
i. P is projective,
ii. every surjection π : M → P of R-modules is split, and
iii. P is a direct summand of a free R-module.
P ROOF. If P is projective and π : M → P is a surjection, then the identity map idP : P → P
lifts to a homomorphism f : P → M such that idP = π ◦ f , so (i) implies (ii). If (ii) holds, then
choose a set of generators X of P, and let F be the free R-module on X, which comes equipped
with a surjection π : F → P that restricts to idX . This surjection is split, so P is a direct summand
of F by Proposition 12.4.3, and therefore (iii) holds.
If P is a direct summand of a free module F with complement Q, then π : B → C is a sur-
jection, and g : P → C is a homomorphism of R-modules, then we can extend g to g̃ : F → C by
setting g̃(q) = 0 for all q ∈ Q. We then have f˜ : F → B such that π ◦ f˜ = g̃ by the projectivity of
F, and the restriction f = f˜|P satisfies π ◦ f = g. Thus, (iii) implies (i). 
376 12. HOMOLOGICAL ALGEBRA

E XAMPLE 12.4.9. For n ≥ 1, the left Mn (R)-module L of column vectors under left multipli-
cation is a direct summand of Mn (R), which is isomorphic to Ln as a left R-module. Hence, L is
projective, though it is not free for n ≥ 2.
In the case that R is a principal ideal domain, we have the following.
C OROLLARY 12.4.10. If R is a principal ideal domain, then every projective R-module is
free.
P ROOF. By the classification of finitely generated modules over a principal ideal domain, it
suffices for finitely generated R-modules to show that any R-module of the form
A = R/(a1 ) ⊕ R/(a2 ) ⊕ · · · R/(an )
for nonzero and nonunit a1 , a2 , . . . , an ∈ R is not projective. Consider the obvious quotient map
Rn → A. That it splits means that each R → R/(ai ) splits. Then R ∼ = (ai ) ⊕ (x) for some x ∈ R,
which means that R is free of rank 2 over itself, which is impossible (e.g., by the classification
theorem).
The general case is left as an exercise. 
D EFINITION 12.4.11. An object I in an abelian category is injective if for every monomor-
phism ι : A → B and every morphism f : A → I, there exists a morphism g : B → I such that
f = g ◦ ι.
R EMARK 12.4.12. The property of I being injective is represented by the existence of g in
the commutative diagram
0 /A ι /B
f
  g
I
with exact upper row.
Dually to the analogous result projective modules, we have the following.
L EMMA 12.4.13. As R-module I is injective if and only if every monomorphism ι : I → R is
split.
We also have the following interesting criterion, which employs Zorn’s lemma.
P ROPOSITION 12.4.14 (Baer’s criterion). A left R-module I is injective if and only if every
homomorphism J → I with J a left ideal of R may be extended to a map R → I.
P ROOF. Let A be an R-submodule of an R-module B and f : A → I be an R-module homo-
morphism. It suffices to show that we can extend f to g : B → I with g|A = f . Let X be the set
of pairs (C, h) with C an R-submodule of B containing A and h : C → I an R-module homomor-
phism. We have a partial ordering on X given by (C, h) < (C0 , h0 ) if C S
is contained in C0 and
h0 |C = h. Given a chain C in X, we have an upper bound (C, h) with C = (D,k)∈C D such that if
d ∈ D for (D, k) ∈ C , then h(d) = k(d). By Zorn’s lemma, C has a maximal element (M, l).
 12.5. EXACT FUNCTORS 377

Suppose first that M 6= B, and let b ∈ B − M. Consider the left ideal

J = {r ∈ R | rb ∈ M}
of R. Define s : J → I by s(r) = l(rb) for r ∈ J. This R-module homomorphism may be extended
to t : R → I by assumption. Then define N = M + Rb and let q : N → I be the unique R-module
homomorphism such that q|M = l|M and q(rb) = t(r) for all r ∈ R. This exists as M ∩ Rb = Jb,
and l(rb) = s(r) = t(r) for r ∈ J. (Also, if rb = 0, then r ∈ J, so q(rb) = l(rb) = 0 in this
instance.) The existence of q gives a contradiction of the maximality of M. Thus M = B, and we
are done. 
E XAMPLE 12.4.15.
a. Q is an injective Z-module.
b. Z/nZ is an injective Z/nZ-module for any n ≥ 1.
c. Z/3Z is an injective Z/6Z-module, but not an injective Z/9Z-module.
We have a very nice description of injective objects in Ab.
D EFINITION 12.4.16. An abelian group D is called divisible if multiplication by n is surjec-
tive on D for every natural number n.
P ROPOSITION 12.4.17. An abelian group is injective if and only if it is divisible.
P ROOF. Let D be injective, and take d ∈ D. Then there exists a group homomorphism
φ : Z → D with 1 7→ d. We also have the multiplication-by-n map on Z, which is injective.
By injectivity of D, we have a map θ : Z → D with φ = nθ . Then d = nθ (1), so D is divisible.
Conversely, let D be divisible. By Baer’s criterion, it suffices to show that every homomor-
phism φ : nZ → D with n ≥ 1 extends to a homomorphism θ : Z → D. Such a φ is determined
by d = φ (n). Let d 0 ∈ D be such that nd 0 = d. Set θ (1) = d 0 . 

12.5. Exact functors

Despite the fact that additive functors preserve direct sums, they may not preserve exact
sequences. We make the following definitions.
D EFINITION 12.5.1. Let F : C → D be an additive functor of abelian categories.
a. We say that F is left exact if for every left short exact sequence 0 → A → B → C in C , the
sequence 0 → F(A) → F(B) → F(C) is exact in D.
b. We say that F is right exact if for every right short exact sequence A → B → C → 0 in C ,
the sequence F(A) → F(B) → F(C) → 0 is exact in D.
c. We say that F is an exact functor if for every short exact sequence 0 → A → B → C → 0
in C , the sequence 0 → F(A) → F(B) → F(C) → 0 is exact in D.
R EMARK 12.5.2. A contravariant additive functor F : C → D is left exact if the resulting
covariant functor C op → D is left exact.
E XAMPLE 12.5.3. The functor F : Ab → Ab by F(A) = A ⊕ A with F( f ) = f ⊕ f is exact.
378 12. HOMOLOGICAL ALGEBRA

T ERMINOLOGY 12.5.4. We (somewhat loosely) say a functor a certain structure if its takes
structures of one sort in a given category (induced from the source category) to those of the
same sort in another (induced from the target category). For instance, exact functors are additive
functors that preserve short exact sequences.
L EMMA 12.5.5. Let F : C → D be an additive functor of abelian categories. The following
are equivalent:
i. F is exact,
ii. F is both left and right exact,
iii. F preserves all three-term exact sequences A → B → C, and
iv. F preserves all exact sequences.
P ROOF. It is immediate that (iv) implies the other statements and also that (ii) implies (i).
Suppose that F that preserves all three-term exact sequences. To say that
0→A→B→C →0
is short exact is equivalent to saying that the three three-term sequences 0 → A → B, A → B → C,
and B → C → 0 are all exact. Since exactness of these is preserved by F, so is exactness of the
original short exact sequence. So, (iii) implies (i).
If F is an exact functor, take any exact sequence A· . Then 0 → ker di → Ai → im di → 0 is
short exact, so
F(di )
0 → F(ker di ) → F(Ai ) −−−→ F(im di ) → 0
is exact as well. It F(di ) follows that image F(im di ) and kernel F(ker di ) for all i, so we have
im F(di ) = F(im di ) = F(ker di−1 ) = ker F(di−1 ).
Thus, F(A· ) is exact. Thus, (i) implies (iv), which finishes the proof. 
R EMARK 12.5.6. The reader may also check that an additive functor of abelian categories is
left (resp., right) exact if and only if it sends short exact sequences to left (resp., right) short exact
sequences.
Recall that for an additive category C , the functors hX (and hX ) may be viewed as taking
values in Ab, and clearly such functors are additive. In fact, they are also left exact.
L EMMA 12.5.7. Let C be an abelian category, and let X be an object of C .
a. The functor hX : C → Ab is left exact.
b. The functor hX : C op → Ab is left exact.
P ROOF. Let
f g
0→A− →B→ − C→0
be an exact sequence in C . Applying hX , we obtain homomorphisms
hX ( f ) hX (g)
0 → HomC (X, A) −−−→ HomC (X, B) −−−→ HomC (X,C)
 12.5. EXACT FUNCTORS 379

of abelian groups, and we claim this sequence is exact. If hX ( f )(α) = 0, then f ◦ α = 0, but f is a
monomorphism, so α = 0. Since hX is a functor, we have hX (g)◦hX ( f ) = 0, and if β ∈ ker hX (g),
then g ◦ β = 0. Naturality of the kernel implies that β factors through a morphism X → ker g.
But we have canonical isomorphisms
∼ ∼ ∼
A−
→ coim f −
→ im f −
→ ker g,
the first as f is a monomorphism, and the composite of the composite of these with the canonical
morphism ker g → B is g. Therefore, we obtain a morphism α : X → A satisfying f ◦ α = g. This
proves part a, and part b is just part a with C replaced by C op . 
L EMMA 12.5.8. Let R be a ring, and let N be a right R-module. The tensor product functor
tN : R-mod → Ab given on objects by tN (M) = N ⊗R M and on morphisms by tN (g) = idN ⊗g is
right exact.
P ROOF. Since tensor products commute with direct sums, tN is additive. Let
f g
A−
→B→
− C→0
be a right short exact sequence of R-modules. The group N ⊗R C is generated by simple tensors
n ⊗ c with n ∈ N and c ∈ C and
n ⊗ c = n ⊗ g(b) = (idN ⊗g)(n ⊗ b)
for any b ∈ B with g(b) = c, we have that tN (g) is surjective. We need then only define an inverse
to the surjection ḡ : cokertN ( f ) → N ⊗R C. For this, we consider the map θ : N ×C → cokertN ( f )
given on (n, c) ∈ N ×C by picking b ∈ B with g(b) = c and then setting θ (n, c) = n⊗b+imtN ( f ).
If g(b0 ) = c, then g(b−b0 ) = 0, so b−b0 = f (a) for some a ∈ A, and then n⊗(b−b0 ) = tN ( f )(n⊗
a), so θ is well-defined and then easily seen to be biadditive and R-balanced. The induced map
Θ : N ⊗R C → cokertN ( f ) is inverse to ḡ by definition. 
L EMMA 12.5.9. Let C be an abelian category. A sequence
f g
0→A−
→B→
−C
is exact if every sequence
hX ( f ) hX (g)
0 → HomC (X, A) −−−→ HomC (X, B) −−−→ HomC (X,C)
is exact.
P ROOF. For X = A, we get
g ◦ f = hX (g) ◦ hX ( f )(idA ) = 0,
so we have a monomorphism s : im f → ker g. For X = ker g and β : ker g → B the natural
monomorphism defined by the kernel, we have hX (g)(β ) = g ◦ β = 0, so there exists α : ker g →
A with f ◦ α = β . We then have that β factors a morphism t : ker g → im f inverse to s. 
P ROPOSITION 12.5.10. Any right (resp., left) adjoint to ia functor between abelian categories
is left (resp., right) exact.
380 12. HOMOLOGICAL ALGEBRA

P ROOF. We treat the case of left exactness, the other case simply being the corresponding
statement in opposite categories. Let G : C → D be an additive functor of abelian categories that
admits a left adjoint F. Suppose that
f g
0→A−
→B→
−C
is a left exact sequence in C . Then for any D ∈ Obj(D), the sequence
hF(D) ( f ) hF(D) (g)
0 → hF(D) (A) −−−−−→ hF(D) (B) −−−−−→ hF(D) (C)
is left exact. Since F is left adjoint to G, this sequence is isomorphic to
hD (G( f )) hD (G(g))
0 → hD (G(A)) −−−−−→ hD (G(B)) −−−−−→ hD (G(C))
as a sequence of abelian groups. Since this holds for all D, the sequence
G( f ) G(g)
0 → G(A) −−−→ G(B) −−→ G(C)
is exact. 
P ROPOSITION 12.5.11. Let R be a ring, and fix an R-module M.
a. The covariant homomorphism functor hM : R-mod → Ab is exact if and only if M is R-
projective.
b. The contravariant homomorphism functor hM : R-mod → Ab is exact if and only if M is
R-injective.
P ROOF. We prove part a. Suppose that the functor is exact. Then for any epimorphism
g : B → P we have an epimorphism
HomR (P, B) → HomR (P, P),
and any inverse image t of idP is the desired splitting map of g.
On the other hand, suppose that P is projective. Consider an exact sequence
f g
0→A−
→B→
− C → 0.
Then we have a diagram
hP ( f ) hP (g)
HomR (P, A) −−−→ HomR (P, B) −−−→ HomR (P,C) → 0.
That this is a complex is immediate. Surjectivity of hP (g) follows immediately from the def-
inition of a projective module. Finally, let h ∈ ker hP (g), so h : P → ker g. Then A → ker g is
an epimorphism, and we have by projectivity of P a map j : P → A with with f ◦ j = h, i.e.,
hP ( f )( j) = h. 
R EMARK 12.5.12. If R is a commutative ring, then HomR (A, B) for R-modules A and B may
be viewed as an R-module under (r · f )(a) = r · f (a). It follows easily that HomR (A, · ) is an
additive functor from the category of R-modules to itself which is exact if A is projective.
 12.6. PROJECTIVE AND INJECTIVE RESOLUTIONS 381

The following embedding theorem, the proof of which is beyond the scope of these notes,
allows us to do most of the homological algebra that can be done in the category of R-modules
for any R in an arbitrary abelian category.
T HEOREM 12.5.13 (Freyd-Mitchell). If C is a small abelian category, then there exists a
ring R and an exact, fully faithful functor C → R-mod.
In other words, C is equivalent to a full, abelian subcategory of R-mod for some ring R.
We can use this as follows: suppose there is a result we can prove about exact diagrams in R-
modules for all R, like the snake lemma. We then have the result in all abelian categories, since
we can take a small full, abelian subcategory containing the objects in which we are interested
and embed it into some category of left R-modules. If the result holds in that category, then by
exactness of the embedding, the result will hold in the original category.
12.6. Projective and injective resolutions
D EFINITION 12.6.1. Let C be an abelian category, and let A be an object in C .
a. A resolution of A is a complex C· of objects in C together with an augmentation morphism
ε C : C0 → A such that the augmented complex
dC
1 εC
· · · → C1 −→ C0 −→ A → 0
is exact.
b. A projective resolution of A is a resolution of A by a complex of projective objects.
D EFINITION 12.6.2. An abelian category C is said to have sufficiently many (or enough) pro-
jectives if for every A ∈ Obj(C ), there exists a projective object P ∈ Obj(C ) and an epimorphism
P → A.
Since free modules are projective, R-mod has enough projectives.
R EMARK 12.6.3. If C has enough projectives, then every object in C has a projective reso-
lution. (We leave the proof as an exercise.)
E XAMPLES 12.6.4. We have the following examples of projective resolutions, all of which
are in fact resolutions by free modules:
a. In Ab, the abelian group Z/nZ has a projective resolution
·n
0→Z−
→ Z → Z/nZ → 0.
b. Consider R = Z[X], and let A = Z[X]/(n, X 2 + 1). Then we have a projective resolution
(·(X 2 +1),·n) (·n)−(·(X 2 +1))
0 → Z[X] −−−−−−−→ Z[X] ⊕ Z[X] −−−−−−−−−→ Z[X] → Z[X]/(n, X 2 + 1) → 0.
c. Consider the ring R = Z[X]/(X n − 1). (This is isomorphic to the group ring Z[Z/nZ].) We
have a projective resolution of Z:
N X−1 N X−1
··· → R −
→ R −−→ R −
→ R −−→ R → Z → 0,
where N = ∑n−1 i
i=0 X .
382 12. HOMOLOGICAL ALGEBRA

P ROPOSITION 12.6.5. Let P· → A and Q· → B be projective resolutions in an abelian cate-

gory, and suppose that g : A → B is an R-module homomorphism. Then g extends to a morphism
f· : P· → Q· of chain complexes such that
··· / P2 / P1 / P0 /A /0
f2 f1 f0 g
   
··· / Q2 / Q1 / Q0 /B /0
commutes. Furthermore, any other lift of g is chain homotopic to f· .
P ROOF. Let P = (Pi , di ) and Q = (Qi , di0 ), and let ε and ε 0 denote the respective augmentation
maps. Then g ◦ ε : P0 → B. Since ε 0 is an epimorphism, we have a map f0 : P0 → Q0 lifting g ◦ ε.
Now f0 induces a map
f¯0 : ker ε → ker ε 0 ,
and since im d1 = ker ε and ker d10 = ker ε 0 , we have an epimorphism Q1 → ker d10 , and we again
use projectivity, this time of Q1 , to lift f¯0 ◦ d1 to a map f1 as in the diagram. We continue in this
manner to obtain f· .
Now, for uniqueness up to chain homotopy, it suffices to show that if g = 0, then f· is chain
homotopic to zero. Well, d00 ◦ f0 = g ◦ d0 = 0, so f0 (P0 ) ⊆ im d10 . By projectivity of P0 , we have
s0 : P0 → Q1 with
f0 = d10 ◦ s0 + s−1 ◦ d0 = d10 ◦ s0 ,
where we have set si = 0 for i < 0 (and di = 0 for i ≤ 0). Now h1 = f1 − s0 ◦ d1 satisfies
d10 ◦ h1 = d10 ◦ f1 − d10 ◦ s0 ◦ d1 = f0 ◦ d1 − f0 ◦ d1 = 0,
so im h1 ⊆ im d20 . Thus, we have s1 : P1 → Q2 lifting h1 , i.e., so that
d20 ◦ s1 = h1 = f1 − s0 ◦ d1 ,
as desired. We continue in this fashion to obtain all si . 
P ROPOSITION 12.6.6 (Horseshoe lemma). Suppose that
f g
0→A−
→B→
− C→0
is a short exact sequence in an abelian category and that (P·A , ε A ) and (P·C , ε C ) are projective
resolutions of A and C respectively. Then there exists a projective resolution (P·B , ε B ) of B with
PiB = PiA ⊕ PiC for each i and such that the diagram

(12.6.1) 0 / P·A ι·
/ P·B p·
/ P·C /0

εA εB εC
  
0 /A f
/B g
/C /0

  
0 0 0
commutes, where ι· and p· are the natural maps on each term.
 12.6. PROJECTIVE AND INJECTIVE RESOLUTIONS 383

P ROOF. Choose a lift t0 of ε C to P0C → B, and let

ε B = f ◦ ε A + t0 ◦ p0 .
Then ε B is clearly surjective, and we have the desired commutativity of the “first two” squares.
Next, letting d·X denote the boundary maps with X = A, C, we may define the boundary map d1B
for B similarly. That is, consider a lift of d1C : P1C → ker ε C to a map t1 : P1C → ker ε B , and define
d1B = ι0 ◦ d1A + t1 ◦ p1 .
Then d1B maps onto ker d1B and makes the next two squares commute. We then continue in this
fashion. 
L EMMA 12.6.7. Let C be an abelian category and P· a split long exact sequence of projec-
tives with Pi = 0 for i < 0. Then P· is a projective object in Ch(C ).
P ROOF. Let P· be a split exact sequence of projectives in C . In other words, we may write
each P0 = Q0 and Pi = Qi ⊕Qi−1 for i ≥ 1, where Qi is a projective object in C , and the morphism
Pi → Pi−1 is simply the composition of the projection Pi → Qi−1 with the inclusion Qi−1 → Pi .
Suppose that π· : A· → P· is a epimorphism of complexes. Since Qi is projective, there exists a
splitting si : Qi → Ai of the composition of πi with projection to Qi . Then
ti = si ⊕ si−1 : Pi = Qi ⊕ Qi−1 → Ai
is a splitting of πi . Since t· is a morphism of complexes, it is a splitting of π· . 
R EMARK 12.6.8. Every split exact complex is the cone of a complex with zero differentials.
R EMARK 12.6.9. Though we shall not prove it, every projective object in the category of
chain complexes over an abelian category is a split exact sequence of projectives. Also, the
projective objects in the category of bounded below chain complexes (or those in nonnegative
degrees) are the bounded below exact sequences of projectives (which automatically split).
D EFINITION 12.6.10. We say that a functor F : C → D between categories preserves pro-
jectives if it takes projective objects in C to projective objects in D.
P ROPOSITION 12.6.11. Let C and D be an abelian category. Let F : C → D be a functor
that is left adjoint to an exact functor G : D → C . Then F preserves projectives.
P ROOF. Let P be a projective object in C . Let f : A → B be a epimorphism in D. We must
show that hF(P) ( f ) : F(A) → F(B) is an epimorphism. Note that we have a commutative diagram
hP (G( f ))
HomC (P, G(A)) / HomC (P, G(B))
o o
 hF(P) ( f ) 
HomD (F(P), A) / HomD (F(P), B),

Exactness of G tells us that G( f ) is an epimorphism, and the upper horizontal map is then an
epimorphism by the projectivity of P, hence the result. 
384 12. HOMOLOGICAL ALGEBRA

12.7. Derived functors

Suppose that F : C → D is a right exact functor between abelian categories C and D and
that C has enough projectives. Then we could try to define the ith left derived functor of F (for
i ≥ 0) on an object A of C by Hi (F(P· )), where P· → A is a projective resolution of A. Of course,
we must check that this definition is independent of the projective resolution chosen, and that we
obtain induced maps on morphisms so that our map becomes a functor.
In the following, one may suppose that C is the category of R-mod, but it is often the case
that D is some other category, like Ab or S-mod for some ring S.
P ROPOSITION 12.7.1. Let F : C → D be an additive functor between abelian categories C
and D. For i ≥ 0, there are functors Li F : C → D given on A ∈ C by Li F(A) = Hi (F(P· )) for P· →
A a projective resolution of A ∈ C and given on g : A → B in C by Li F(g) = F( f )∗ : Hi (F(P· )) →
Hi (F(Q· )) for P· → A and Q· → B projective resolutions and f· : P· → Q· a morphism of com-
plexes f : P· → Q· compatible with the augmentations to A and B. These functors are dependent
on the choices made up only to unique isomorphism.
P ROOF. The key point is that Proposition 12.6.5 tells us that F( f )∗ is independent of the
choice of f , since any two choices are chain homotopic. In particular, given any two choices of
P· and Q· of projective resolutions of A, and any choices of f· → P· → Q· f·0 → Q· → P· augmenting
the identity morphism on A gives rise to morphisms on cohomology, the resulting maps F( f )∗
and F( f 0 )∗ must be mutually inverse, since f ◦ f 0 and f 0 ◦ f augment the identity on A, as to the
identity morphisms on Q· and P· . Thus, Hi (F(P· )) is unique up to unique isomorphism, so Li F is
well-defined (up to unique isomorphism), and Li F(idA ) = idLi F(A) . Similarly, it is easy to check
that the uniqueness also implies that Li F is compatible with compositions. 
D EFINITION 12.7.2. For an additive functor F : C → D, the ith left derived functor Li F : C →
D of F is the functor defined by Proposition 12.7.1.
We have the following obvious corollaries of Lemma 12.7.1.
L EMMA 12.7.3. Let F : C → D be a right exact functor of abelian categories. Then we have
∼
a canonical, natural isomorphism L0 F −
→ F of functors.
P ROOF. Since F is right exact, the sequence
F(P1 ) → F(P0 ) → F(A) → 0
is exact. Hence, we have
L0 F(A) = H0 (F(P· )) ∼
= F(A).
The reader will easily check the independence of the choice of resolution and naturality, as in
Lemma 12.7.1. 
C OROLLARY 12.7.4. If P is a projective object, then Li F(P) = 0 for i ≥ 1.
P ROOF. Consider the projective resolution that is P in degree zero and 0 elsewhere, where
the augmentation map P → P is the identity. This has the desired homology. 
Next, we prove that the Li F are functors.
 12.7. DERIVED FUNCTORS 385

P ROPOSITION 12.7.5. To each morphism f : A → B in C , we can associate morphisms

Li F( f ) : Li F(A) → Li F(B)
for all i ≥ 0 in such a way that Li F : C → D becomes a functor and L0 F( f ) = F( f ). Further-
more, each Li F is additive.
P ROOF. The unique morphism Li F( f ) is induced on homology by the morphism of chain
complexes given in Proposition 12.6.5. Functoriality follows by canonicality of the map of ho-
mology.
To see additivity, note that Li F(0A ) is induced by the zero morphism of chain complexes
and hence is is zero map on Li F(A). Similarly Li F( f + g), for f , g : A → B, can be given by
the sum of the induced maps on chain complexes, hence is given by the sum of the maps on
homology. 
D EFINITION 12.7.6. For a right exact functor F : C → D of abelian categories, the functor
Li F is called ith left derived functor of F.
We see that L0 F and F are canonically naturally isomorphic functors.
D EFINITION 12.7.7. A homological δ -functor is a sequence of additive functors Fi : C → D
for i ∈ Z, together with, for every exact sequence
f g
0→A−
→B→
− C→0
in C , morphisms δi : Fi (C) → Fi−1 (A) fitting in a long exact sequence
Fi ( f ) Fi (g)
i δ
· · · → Fi (A) −−−→ Fi (B) −−−→ Fi (C) −
→ Fi−1 (A) → · · ·
which are natural in the sense that if we have a morphism of short exact sequences in C ,
0 /A /B /C /0

  
0 / A0 / B0 / C0 / 0,

then we obtain a morphism of long exact sequences in D,

··· / Fi (A) / Fi (B) / Fi (C) / Fi−1 (A) / ···

   
··· / Fi (A0 ) / Fi (B0 ) / Fi (C0 ) / Fi−1 (A0 ) / ··· .

E XAMPLE 12.7.8. Define functors F0 , F1 : Ab → Ab by F0 (A) = A/pA and

F1 (A) = A[p] = {a ∈ A | pa = 0}
for any abelian group A, and set Fi = 0 otherwise. Given an exact sequence
0→A→B→C →0
386 12. HOMOLOGICAL ALGEBRA

in Ab, we obtain a long exact sequence

1 δ
0 → A[p] → B[p] → C[p] −→ A/pA → B/pB → C/pC → 0
from the snake lemma applied to the diagram
0 /A /B /C /0
·p ·p ·p
  
0 /A /B /C / 0.
This defines a δ -functor.
T HEOREM 12.7.9. For every short exact sequence
0→A→B→C →0
in C , there exist morphisms δi : Li F(C) → Li−1 F(A) such that the functors L· F together with the
maps δ· form a homological δ -functor.
P ROOF. By the Horseshoe lemma, we have a projective resolution P·X → X for X = A, B, C
fitting in a diagram (12.6.1). Now, applying F to the resolutions, we have split exact sequences
0 → F(PiA ) → F(PiB ) → F(PiC ) → 0
for each i. The resulting exact sequence of complexes (which need not be split) yields a long
exact sequence in homology
1 δ
· · · L1 F(B) → L1 F(C) −→ F(A) → F(B) → F(C) → 0,
as desired.
It remains to check naturality. Consider a morphism of short exact sequences

0 /A f
/B g
/C /0
qA qB qC
 f0  g0 
0 / A0 / B0 / C0 / 0.
0
By Proposition 12.6.5, can extend qA and qC to maps of complexes qA· : P·A → P·A and qC· : P·C →
0 0
P·C . Suppose we have constructed P·B and P·B via the Horseshoe lemma. We fit this all into a
commutative diagram

(12.7.1) 0 / P·A ι·
/ P·B p·
/ P·C /0
εA
  
0 qA· /A f
/B g
qC· /C /0
 
ι·0 p0·
0 / P·A0 / P·B0 / P·C0 qC /0

  f0   g0  
0 / A0 / B0 / C0 / 0.
 12.7. DERIVED FUNCTORS 387

We also have splitting maps ji : PiC → PiB and ki : PiB → PiA for each i (and, similarly, maps ji0 and
ki0 ). For each X, let us denote the augmentation map by ε X .
0
We must define a map qB· : P·B → P·B making the entire diagram (12.7.1) commute. We first
note that
0 0
g0 ◦ (qB ◦ ε B − ε B ◦ j00 ◦ qC0 ◦ p0 ) = qC ◦ g ◦ ε B − ε C ◦ p00 ◦ j0 ◦ qC0 ◦ p0
0 0
= qC ◦ ε C ◦ p0 − ε C ◦ qC0 ◦ p0 = (qC ◦ ε C − ε C ◦ qC0 ) ◦ p0 = 0.
Hence, there exists a map β0 : P0B → A0 with
0
f 0 ◦ β0 = qB ◦ ε B − ε B ◦ j00 ◦ qC0 ◦ p0 .
0 0 0
Since ε A is an epimorphism, we may choose α0 : P0B → P0A with ε A ◦ α0 = β0 . Now set
qB0 = ι00 ◦ qA0 ◦ k0 + ι00 ◦ α0 ◦ p0 + j00 ◦ qC0 ◦ p0 .
0
The trickiest check of commutativity is that ε B ◦ qB0 = qB ◦ ε B . We write this mess out:
0 0 0 0
ε B ◦ qB0 = ε B ◦ ι00 ◦ qA0 ◦ k0 + ε B ◦ ι00 ◦ α0 ◦ p0 + ε B ◦ j00 ◦ qC0 ◦ p0
0 0 0
= f 0 ◦ ε A ◦ qA0 ◦ k0 + f 0 ◦ ε A ◦ α0 ◦ p0 + ε B ◦ j00 ◦ qC0 ◦ p0
0
= f 0 ◦ qA ◦ ε A ◦ k0 + f 0 ◦ β0 ◦ p0 + ε B ◦ j00 ◦ qC0 ◦ p0
0 0
= f 0 ◦ qA ◦ ε A ◦ k0 + ( j00 ◦ qC ◦ ε C − ε B ◦ j00 ◦ qC0 ) ◦ p0 + ε B ◦ j00 ◦ qC0 ◦ p0
= f 0 ◦ qA ◦ ε A ◦ k0 + j00 ◦ qC ◦ ε C ◦ p0
= qB ◦ ε B .
0
The other qBi are defined similarly. For instance, one can see there exists a map β1 : P1C → P0A
such that
0
ι0 ◦ β1 = ι0 ◦ β0 ◦ d0C + j10 ◦ qC0 ◦ d0C − d0B ◦ j10 ◦ qC1 ◦ p1 ,
and we set
qB1 = ι10 ◦ qA1 ◦ k1 + ι10 ◦ α1 ◦ p1 + j10 ◦ qC1 .

D EFINITION 12.7.10. A (homological) universal δ -functor is a δ -functor F· = (Fi , δi ) with
Fi : C → D such that if G· = (Gi , δi0 ) is any other δ -functor with Gi : C → D for which there
exists a natural transformation η0 : G0 F0 , then η0 extends uniquely to a morphism of δ -
functors, i.e., a sequence of natural transformations ηi : Gi Fi such that
δi0
Gi (C) / Gi−1 (A)

(ηi )C (ηi−1 )A
 
Fi (C)
δi
/ Fi−1 (A)
388 12. HOMOLOGICAL ALGEBRA

commutes for any short exact sequence in C :

0 → A → B → C → 0.
(That is, we get a morphism of the associated long exact sequences.)
The δ -functor of left derived functors of F is universal, which will follow as a corollary of
Theorem 12.7.14 below.
T HEOREM 12.7.11. The δ -functor (Li F, δi ) is universal.
D EFINITION 12.7.12. Let F : C → D be a left exact functor between abelian categories. We
say that an object Q in C is F-acyclic if Li F(Q) = 0 for all i ≥ 1.
Note that the Li F(A) for any A ∈ Obj(C ) may be computed using resolutions by F-acyclic
objects, as opposed to just projectives.
P ROPOSITION 12.7.13. Let F : C → D be a left exact functor between abelian categories,
and let A be an object of C . Suppose that C· → A is a resolution of A by F-acyclic objects. Then
Li F(A) ∼
= Hi (F(C· )) for each i ≥ 0.
P ROOF. Note that we have an exact sequence
F(dC ) F(ε C )
F(C1 ) −−−1→ F(C0 ) −−−→ F(A) → 0,
so F(A) ∼
= H0 (F(C· )). Set K0 = ker ε C . We then have an exact sequence
0 → L1 F(A) → F(K0 ) → F(C0 ) → F(A) → 0,
which yields
ker F(d1C ) ∼
L1 F(A) ∼= ker(coker(F(C2 ) → F(C1 )) → F(C0 )) ∼ = = H1 (F(C· )).
im F(d2C )
We also have isomorphisms Li F(A) ∼ = Li−1 F(K0 ) for each i ≥ 2.
For i ≥ 1, set Ki = ker diC ∼ C . The exact sequences
= im di+1
0 → Ki → Ci → Ki−1 → 0,
then yield isomorphisms used in the following for i ≥ 2:
ker F(diC ) ∼
Li F(A) ∼
= Li−1 F(K0 ) ∼
= ··· ∼
= L1 F(Ki−2 ) ∼
= ker(F(Ki−1 ) → F(Ci−1 )) ∼
= C )= i
H (F(C· )).
im F(di+1

More generally, we have the following characterization of universal δ -functors.
T HEOREM 12.7.14. Let C and D be abelian categories such that C has enough projectives.
Suppose that (Fi , δi ) form a δ -functor Fi : C → D and Fi (P) = 0 for every projective P ∈ Obj(C )
and i ≥ 1. Then (Fi , δi ) is universal.
 12.7. DERIVED FUNCTORS 389

P ROOF. Suppose that (Gi , δi0 ) is another δ -functor and that we have a natural transformation
G0 F0 . Let A ∈ Obj(C ) and let π : P → A be an epimorphism with P projective. Let K = ker π.
Let i ≥ 1, and suppose that we have constructed a natural transformation Gi−1 Fi−1 . Since
Fi (P) = 0 is projective, we have a commutative diagram
Gi (A) / Gi−1 (K) / Gi−1 (P)

  
0 / Fi (A) / Fi−1 (K) / Fi−1 (P).

The morphism Gi (A) → Fi (A) is the unique map which makes the diagram commute.
Now let f : A → B be a morphism in C . We create a diagram as follows:
0 /K /P /A /0
f
  
0 / K0 / P0 /B / 0,

by taking P0 to be projective, P to be any projective with an epimorphism to the pullback of the

diagram P0 → B ← A, and K and K 0 to be the relevant kernels. We then have a diagram
Gi (A) / Gi−1 (K)

# &
Fi (A) / Fi−1 (K)

 
Gi (B) / Gi−1 (K 0 )

#  & 
Fi (B) / Fi−1 (K 0 ).

We need only see that the leftmost square commutes, but this follows easily from a diagram chase
and the fact that the two horizontal maps on the frontmost square are monomorphisms.
Hence, we have constructed a sequence of natural transformations Gi Fi . It remains only
to see that these form a morphism of δ -functors. This being an inductive argument of the above
sort, we leave it to the reader. 
As a corollary, we have a natural isomorphism of δ -functors between the left derived functors
Li F0 of a right exact functor F0 and any δ -functor (Fi , δi ) with Fi (P) = 0 for P projective and i ≥ 1.
We next wish to study right derived functors of left exact functors.
D EFINITION 12.7.15. An injective resolution of an object A of an abelian category is a
cochain complex I · of injective objects with I i = 0 for i < 0 and a morphism A → I 0 such that the
resulting diagram
0 → A → I0 → I1 → I2 → · · ·
is exact.
390 12. HOMOLOGICAL ALGEBRA

D EFINITION 12.7.16. We say that an abelian category C has enough (or sufficiently many)
injectives if for every A ∈ Obj(C ), there exists an injective object I ∈ Obj(C ) and a monomor-
phism A → I.
R EMARK 12.7.17. An abelian category has enough injectives if and only if every object of it
has an injective resolution.
P ROPOSITION 12.7.18. The category R-mod has enough injectives.
P ROOF. First take the case that R = Z. Let A be an abelian group, and write it as a quotient
of a free abelian group
M
A = ( Z)/T
j∈J
L
for some indexing set J and submodule T of j∈J Z. Then we may embed A in
M
I=( Q)/T,
j∈J

which is divisible as a quotient of a divisible group.

Next, let A be a left R-module. We have an injection of left R-modules,
φ : A → HomZ (R, A),
by φ (a)(r) = ra. Now, embed A in a divisible group D, so that the resulting map
HomZ (R, A) → HomZ (R, D)
is an injection. The proof that HomZ (R, D) is an injective R-module is left to the reader. 
We also have the analogues of Propositions 12.6.5 and 12.6.6 for injective resolutions.
Suppose now that F : C → D is a left exact functor between abelian categories and that C
has enough injectives. For each i ≥ 0, we define additive functors Ri F : C → D by
Ri F(A) = H i (F(I · )),
where A → I · is any injective resolution of A ∈ Obj(C ) and, for f : A → B in C , by
Ri F( f ) : Ri F(A) → Ri F(B)
to be the map on homology induced by any morphism of chain complexes I · → J · extending f ,
where A → I · and B → J · are injective resolutions. We have R0 F = F. The functors R· F are
called the right-derived functors of F.
D EFINITION 12.7.19. A cohomological δ -functor is a sequence of additive functors F i : C →
D for i ∈ Z, together with, for every exact sequence
f g
0→A−
→B→
− C→0
in C , morphisms δi : F i (C) → F i+1 (A) fitting in a long exact sequence
F i( f ) F i (g) δ
· · · → F i (A) −−−→ F i (B) −−−→ F i (C) −
→i
F i+1 (A) → · · ·
 12.8. TOR AND EXT 391

which are natural in the sense that if we have a morphism of short exact sequences in C ,
0 /A /B /C /0

  
0 / A0 / B0 / C0 / 0,
then we obtain a morphism of long exact sequences in D,
··· / F i (A) / F i (B) / F i (C) / F i+1 (A) / ···

   
··· / F i (A0 ) / F i (B0 ) / F i (C0 ) / F i+1 (A0 ) / ··· .

R EMARK 12.7.20. A cohomological δ -functor (F i , δ i ) is universal if there exists a unique

extension of any natural transformation F 0 G0 , where (Gi , (δ 0 )i ) is another δ -functor, to a
morphism of δ -functors.
T HEOREM 12.7.21. The functors R· F form a cohomological universal δ -functor.
The proof is dual to that of Theorems 12.7.9 and 12.7.11. We also have the following.
T HEOREM 12.7.22. Let C be an abelian category that has enough injectives. Then the
cohomology functors H i : Ch≥0 (C ) → C for i ≥ 0 on complexes in nonnegative degrees together
with the connecting homomorphisms δ i attached to a short exact sequence of complexes form a
universal δ -functor.

12.8. Tor and Ext

E XAMPLE 12.8.1. Take the abelian group M = Z/nZ. If we apply the functor tM : Ab → Ab
n
to the exact sequence of abelian groups 0 → Z →
− Z → Z/nZ → 0 for some n ≥ 2, we obtain the
right, but not left, exact sequence
− Z/nZ −∼
0
Z/nZ → → Z/nZ → 0.
If we apply hM : Ab → Ab to the same exact sequence, we obtain the left, but not right, exact
sequence
0 → 0 → 0 → Z/nZ,
noting that HomZ (Z/nZ, Z) = 0.
D EFINITION 12.8.2. A right R-module N is R-flat, or just flat, if the tensor product functor
tN : R-mod → Ab is exact.
R EMARKS 12.8.3.
a. An S-R-bimodule N is flat as a right R-module if and only if the functor tN : R-mod →
S-mod is exact.
b. A left R-module M is defined to be flat if it is flat as a right Rop -module (which is equivalent
to the right tensor product functor with M being exact on Rop -mod).
392 12. HOMOLOGICAL ALGEBRA

P ROPOSITION 12.8.4. Projective right R-modules are R-flat.

P ROOF. Let P be a projective R-module, and let Q be a complement in a free R-module F on
a basis X. Let f : A → B be an injection of R-modules. We have a commutative diagram
idP ⊗ f
P ⊗R _ A / P ⊗R B
_

 idF ⊗ f 
F ⊗R A / F ⊗R B
o o
L   ( f )x∈X L

x∈X A
/ x∈X B,
the vertical isomorphisms following from the commutativity of direct sums and tensor products.
Since f is injective, so is the lowermost vertical map. Since P is a direct sum of F, the map
P ⊗ A → (P ⊗ A) ⊕ (Q ⊗ A) −∼
R R →F⊗ A R R
is injective, and similarly with A replaced by B. Thus, commutativity of the diagram yields the
injectivity idP ⊗ f . Since left tensor product with P preserves injective homomorphisms and right
exact sequences, it preserves short exact sequences and is therefore exact. 
For modules over a principal ideal domain, we can characterize flat modules as follows.
P ROPOSITION 12.8.5. Let R be a PID. An R-module M is flat if and only if M is R-torsion-
free.
P ROOF. Let M be a flat R-module. Let r ∈ R be a nonzero element, and let φr : R → R be the
injective map φr (x) = rx for x ∈ R. The tensor product map idM ⊗R φr is injective as A is R-flat.
Under the identification M ⊗R R ∼ = M of Corollary 8.4.24, determined by m ⊗ r 7→ rm, the map
idM ⊗R φr becomes identified with the map ψr : M → M that is left multiplication by r. Since ψr
is then injective for every nonzero r, we see that M has no nonzero R-torsion.
Next, let M be R-torsion free. It is the union (which is also the direct limit) of its finitely
generated, necessarily torsion-free R-submodules. We omit here a check of the fact that direct
limits and tensor products commute. Given this, we may assume that M is finitely generated, in
which case it follows from Proposition 12.8.5 that M is free, hence projective, and hence flat. 
R EMARK 12.8.6. It follows from Corollary 8.9.3 and Proposition 12.8.5 that finitely gener-
ated flat modules over a PID R are R-free.
D EFINITION 12.8.7. Let R and S be rings, and let A be an S-R-bimodule. For i ≥ 0, the ith
Tor-functor
TorRi (A, · ) : R-mod → S-mod
is the ith left derived functor of tA .
R EMARK 12.8.8. If R is a commutative ring, then an R-module A provides functors
TorRi (A, · ) : R-mod → R-mod
since R-modules are automatically R-R-bimodules.
 12.8. TOR AND EXT 393

R EMARK 12.8.9. As TorRi (A, B) = Hi (A ⊗R Q· ) for any projective resolution Q· of B by R-

modules, the composition of the functor
TorRi (A, · ) : R-mod → S-mod
with the forgetful functor F : R-mod → Ab agrees with the functor
TorRi (F(A), · ) : R-mod → Ab,
hence the omission of the notation for S in the definition of TorRi (A, · ).
E XAMPLE 12.8.10. In Ab, consider the projective resolution
n
0→Z→
− Z → Z/nZ → 0
n
of B. Computing the homology of 0 → A →
− A → 0, we obtain

A/nA
 if i = 0
Tori (A, Z/nZ) ∼
Z
= A[n] = {a ∈ A | na = 0} if i = 1

0 if i ≥ 2.
L EMMA 12.8.11. Let R be a ring. The following conditions on a right R-module A are
equivalent:
i. A is flat,
ii. TorR1 (A, · ) = 0,
iii. TorRi (A, · ) = 0 for all i ≥ 1.
P ROOF. Clearly, (iii) implies (ii). If
0 → B1 → B2 → B3 → 0
is an exact sequence of right R-modules, then we have a long exact sequence for any R-module
that ends with
TorR1 (A, B3 ) → A ⊗R B1 → A ⊗R B2 → A ⊗R B3 → 0,
from which it is clear that (ii) implies (i).
Finally, if (i) holds and Q· is a projective resolution of B in R-mod, then the complex
· · · → A ⊗R Q1 → A ⊗R Q0 → A ⊗R B → 0
is exact by the flatness of A. It follows that
TorRi (A, B) = Hi (A ⊗R Q· ) = 0
for all i ≥ 1. 
P ROPOSITION 12.8.12. Let A be a right R-module and B a left R-module. Let P· → A be a
resolution of A by projective right R-modules. Then
TorR (A, B) ∼
i = Hi (P· ⊗R B)
for all i ≥ 0. In particular, the functors TorRi ( · , B) are the left derived functors of R-tensor product
with B.
394 12. HOMOLOGICAL ALGEBRA

P ROOF. We sketch a proof. Form projective resolutions P· → A and Q· → B. We then have a

double complex P· ⊗R Q· , and we can consider homology of the total complex
M
Tot(P· ⊗R Q· )k = Pi ⊗R Q j ,
i+ j=k
where the boundary maps from each term Pi ⊗R Q j are given by the sums
diA ⊗ idQ j +(−1)i idPi ⊗d Bj .
We claim that the homology of this chain complex is isomorphic to the homology of the com-
plexes P· ⊗R B and A ⊗R Q· , from which the lemma follows.
We have maps of complexes
(12.8.1) Tot(P· ⊗R Q· ) → P· ⊗R B
and
(12.8.2) Tot(P· ⊗R Q· ) → A ⊗R Q·
induced by augmentation morphisms (up to sign, and zero maps otherwise). The double complex
P· ⊗R Q· → P· ⊗R B (i.e., with Pi ⊗R B in the (i, −1)-position) has exact columns, since each
projective module is flat. One can show that this implies that the total complex of this cpomplex
is exact. This says precisely that the map in (12.8.1) induces an isomorphism on homology.
Similarly, so does the map in (12.8.2). 
We have the following almost immediate corollary, since left and right tensor product with a
module over a commutative ring are naturally isomorphic functors.
C OROLLARY 12.8.13. Let R be commutative. We have TorRi (A, B) ∼ = TorRi (B, A) for all R-
modules A, B and i ≥ 0.
We now give an alternate proof of Proposition 12.8.12.
P ROOF. Let Q· → B be a projective resolution of A by right R-modules. Suppose that
0 → A1 → A2 → A3 → 0
is an exact sequence. Then
0 → A1 ⊗R Q· → A2 ⊗R Q· → A3 ⊗R Q· → 0
is exact. This yields a long exact sequence in homology of the form
· · · → TorRi (A1 , B) → TorRi (A2 , B) → TorRi (A3 , B) → TorRi−1 (A1 , B) → · · · ,
so the functors TorRi ( · , B) do in fact form a δ -functor. Futhermore, since any projective right
R-module P is flat, we have that TorRi (P, B) = 0 for all i ≥ 1. By Theorem 12.7.14, it follows
that the TorRi ( · , B) are a universal δ -functor extending tB . The proposition therefore follows by
Theorem 12.7.11. 
R EMARK 12.8.14. It follows from Proposition 12.8.12 and Proposition 12.7.13 that the
TorRi (A, B)can be computed via a flat resolution of either A or B.
The following explains something more of the name “Tor”.
 12.8. TOR AND EXT 395

1 (A, · ) = 0 if and only if A is torsion-free.

L EMMA 12.8.15. The functor TorZ
P ROOF. We prove this for finitely generated abelian groups. (The general result then follows
from the fact that left derived functors commute with colimits.) By Proposition 12.8.13, we may
compute TorZ 1 (A, B) by finding a projective resolution of A. Say
A∼= Zm ⊕ Z/n1 Z ⊕ · · · Z/nr Z
with r ≥ 0 and the ni ≥ 2. Then we have a projective resolution of the form
(1,··· ,1,n1 ,··· ,nr )
0 → Zm+r −−−−−−−−−→ Zm+r → A → 0.
Tensoring with B and computing H1 , we obtain B[n1 ] ⊕ · · · ⊕ B[nr ]. This will always be trivial if
and only if r = 0. 
By Lemma 12.8.15, a Z-module is flat if and only if it is torsion-free. This is seen to hold in
the same manner with Z replaced by any PID. Note that this does not hold for all commutative
rings.
E XAMPLE 12.8.16. Consider R = Q[x, y]. Then the exact sequence
(y,−x) (a,b)7→ax+by
0 → R −−−→ R2 −−−−−−−→ R → Q → 0
is a free resolution of Q. Let J be the ideal (x, y) of R, so Q ∼
= R/J. Then we have isomorphisms
0
TorR1 (J, Q) ∼
= TorR2 (Q, Q) ∼ − Q2 ) = Q.
= ker(Q →
Thus J is not flat as an R-module, even though it is torsion-free.
Here is another class of examples.
L EMMA 12.8.17. Let S be a subset of R that is multiplicatively closed. Then the localization
S−1 R is a flat R-module.
P ROOF. Recall that we have natural isomorphisms S−1 A ∼
= S−1 R ⊗R A for R-modules A. Sup-
pose that f : A → B is an injection of R-modules. Then we obtain an induced R-module homo-
morphism f˜ : S−1 A → S−1 B, which we must show is an injection. Suppose f˜(s−1 a) = 0. Then
0 = s f˜(s−1 a) = f˜(a) = f (a),
so a = 0. 
D EFINITION 12.8.18. Let R and S be rings, and let A be an R-S-bimodule. For i ≥ 0, the ith
Ext-functor
ExtiR (A, · ) : R-mod → S-mod
is the ith right derived functors of hA .
E XAMPLE 12.8.19. For R = Z, we may consider the injective resolution
n
0 → Z/nZ → Q/Z →
− Q/Z → 0
396 12. HOMOLOGICAL ALGEBRA

of Z/nZ. For any abelian group B, we write B∨ = Hom(B, Q/Z). We must compute the coho-
n
mology of A∨ →
− A∨ . This yields

∨
A [n]
 if i = 0
i ∼ ∨
ExtZ (A, Z/nZ) = A /nA ∨ if i = 1

0 if i = 2.

One has that ExtiR (P, B) = 0 for all B and all i ≥ 1 if P is a projective module, as follows from
the exactness of HomR (P, · ). We have the analogous result to Proposition 12.8.13 for Ext-groups,
which says that such groups may be computed using projective resolutions.
P ROPOSITION 12.8.20. We have ExtiR (A, B) ∼
= H i (HomR (P· , B)), where P· → A is any pro-
jective resolution of A.
We end with a characterization of Ext1R in terms of extensions.
D EFINITION 12.8.21. An extension of an R-module A by an R-module B is an exact sequence
0 → B → E → A → 0, where E is an R-module. Two extensions of A by B are called equivalent
if there is an isomorphism of exact sequences between them that is the identity on A and B.
Note that all split extensions (i.e., those with split exact sequences) are split.
E XAMPLE 12.8.22. There are p equivalence classes of extensions of Z/pZ by Z/pZ as Z-
modules:
pi mod p
0 → Z/pZ − → Z/p2 Z −−−→ Z/pZ → 0
with 1 ≤ i ≤ p − 1, and
0 → Z/pZ → Z/pZ ⊕ Z/pZ → Z/pZ → 0.
T HEOREM 12.8.23. There is a one-to-one correspondence between equivalence classes of
extensions of A by B and Ext1R (A, B).
P ROOF. Suppose that E is an equivalence class of extensions of A by B, representative by an
exact sequence
(12.8.3) 0 → B → E → A → 0.
We then have an exact sequence
E ∂
HomR (E, B) → HomR (B, B) −→ Ext1R (A, B),
and we set Φ(E ) = ∂E (idB ). This is clearly independent of the choice of representative.
Conversely, suppose u ∈ Ext1R (A, B). Fix an exact sequence
ι
0→K→
− P→A→0
with P projective. We then have an exact sequence
∂
→ Ext1R (A, B) → 0.
HomR (P, B) → HomR (K, B) −
 12.9. GROUP COHOMOLOGY 397

Let t ∈ HomR (K, B) with ∂ (t) = u. Let E be the pushout

E = P qK B = P ⊕ B/{(ι(k),t(k)) | k ∈ K}.
We have a commutative diagram
(12.8.4) 0 /K /P /A /0
t
 
0 /B /E /A / 0.
Here, the map E → A is defined by universality of the pushout (via the map P → A and the zero
map B → A). We define Ψ(u) to be the equivalence class E 0 of the extension given by the lower
row. Though it is not immediately clear that this is independent of the choice of t with ∂ (t) = u,
this follows if we can show that Ψ and Φ as constructed are mutually inverse.
To see that Φ(Ψ(u)) = u, set E = Ψ(u), again choosing any t with ∂ (t) = u. The diagram

(12.8.5) HomR (B, B)

∂E
/ Ext1 (A, B)
R

hB (t)

HomR (K, B)
∂ / Ext1 (A, B),
R

commutes. Hence, we have

Φ(Ψ(u)) = Φ(E ) = ∂E (idB ) = ∂ (t) = u,
as desired.
On the other hand, suppose given E with exact sequence (12.8.3). By projectivity of P, the
map P → A lifts to a map P → E. Hence, we have a diagram as in (12.8.4). Furthermore, the
map t in the diagram (12.8.4) satisfies ∂ (t) = ∂E (idB ) by the commutativity of (12.8.5). Now,
there exists a map P qK B → E by universality of the pushout, and it is the identity on A and B,
hence an isomorphism by the 5-lemma. It follows by construction that
Ψ(Φ(E )) = Ψ(∂E (idB )) = E .


12.9. Group cohomology

In this section, we let G denote a group.
D EFINITION 12.9.1. The augmentation map ε : Z[G] → Z is the unique ring homomorphism
with ε(g) = 1 for all g ∈ G.
D EFINITION 12.9.2. The augmentation ideal IG of Z[G] is the kernel of the augmentation
map.
L EMMA 12.9.3. The augmentation ideal IG is generated by {g − 1 | g ∈ G}.
398 12. HOMOLOGICAL ALGEBRA

P ROOF. We have ( )
IG = ∑ agg ∈ Z[G] | ∑ ag = 0 .
g∈G g∈G
For α = ∑g∈G ag g ∈ IG , we have
α =α− ∑ ag = ∑ ag(g − 1).
g∈G g∈G

D EFINITION 12.9.4. Let A be an Z[G]-module.
a. The G-invariant group of A is the Z-module
AG = {a ∈ A | ga = a for all g ∈ G},
the maximal Z[G]-submodule of A on which all elements of G act trivially.
b. The G-coinvariant group of A is the Z-module AG = A/IG A, the maximal Z[G]-quotient
of A on which all elements of G act trivially.
E XAMPLES 12.9.5.
a. If we view Z as a Z[G]-trivial module, we have ZG = Z and ZG ∼
= Z.
b. We have Z[G]G ∼ = Z via the augmentation map, and
(
Z · NG if G is finite
Z[G]G =
0 otherwise,
where NG = ∑g∈G g is the norm element in a finite group G. The computation of the invariant
group follows from the fact that the action of G on itself by left multiplication is transitive, so for
an element of Z[G] to be G-fixed, its coefficients must all be equal.
c. Let K/F be a finite Galois extension of fields, and let G = Gal(K/F). Then K G = F and
(K × )G = F × .
E XAMPLE 12.9.6. For n ≥ 2, let Sn act on A = Z[x1 , x2 , . . . , xn ] by
σ · p(x1 , x2 , . . . , xn ) = p(xσ (1) , xσ (2) , . . . , xσ (n) )
for σ ∈ Sn and p ∈ A . This action is Z-bilinear so it gives A the structure of a left A [Sn ]-
module. Then A Sn is the Z-module of symmetric polynomials in A , which is the Z-module
generated by the elementary symmetric polynomials (see Definition 6.13.4). On the other hand,
ASn ∼
= Z[x], with the isomorphism induced by the Z-linear map A → Z[x] taking each xi to x.
R EMARK 12.9.7. We have a left exact invariant functor A 7→ AG as a functor Z[G]-mod →
Ab, with the map on homomorphisms being the restriction to invariant subgroups. This functor
is naturally isomorphic to the functor hZ , where Z is viewed as the trivial Z[G]-module. In
particular
ηA : HomZ[G] (Z, A) → AG , ηA (φ ) = φ (1)
for φ ∈ HomZ[G] (Z, A) is a natural isomorphism. Thus, the invariant factor is left exact.
 12.9. GROUP COHOMOLOGY 399

Similarly, A 7→ AG defines a right exact coinvariant functor which is isomorphic to the functor
tZ , in that we have natural isomorphisms
Z⊗ A −∼
Z[G] →A , 1 ⊗ a 7→ a + I A.
G G
In particular, the coinvariant functor is right exact.
D EFINITION 12.9.8.
a. The cohomology H ∗ (G, · ) of G is the δ -functor given by the right derived functors of
the G-invariant functor. The ith cohomology group of G with coefficients in a Z[G]-module A is
H i (G, A).
b. The homology H∗ (G, · ) of G is the δ -functor given by the left derived functor of the
G-coinvariant functor. The ith homology group of G with coefficients in a Z[G]-module A is
Hi (G, A).
R EMARK 12.9.9. By definition, we have natural isomorphisms
H i (G, A) ∼
= ExtiZ[G] (Z, A) and Hi (G, A) ∼
Z[G]
= Tori (Z, A)
for i ≥ 0 and Z[G]-modules A.
Let us give a more explicit description of group cohomology.
D EFINITION 12.9.10. The bar resolution of Z as a Z[G]-module is the complex C· with
Ci = Z[Gi+1 ] for i ≥ 0, differentials di : Ci → Ci−1 given on (g0 , . . . , gi ) ∈ Gi+1 by
i
di ((g0 , . . . , gi )) = ∑ (−1) j (g0, . . . , g j−1, g j+1, . . . , gi)
j=0
and augmentation ε : C0 → Z the augmentation map.
R EMARK 12.9.11. As follows from Remark 12.9.9, the group H i (G, A) is the ith cohomology
group of the complex
D0
0 → HomZ[G] (Z[G], A) −→ HomZ[G] (Z[G2 ], A) → · · ·
Di−1
→ HomZ[G] (Z[Gi ], A) −−→ HomZ[G] (Z[Gi+1 ], A) → · · ·
with HomZ[G] (Z[G], A) ∼= A in degree 0. Similarly, Hi (G, A) is the ith homology group of the
complex
· · · → Z[Gi+1 ] ⊗Z[G] A → · · · → Z[G2 ] ⊗Z[G] A → Z[G] ⊗Z[G] A → 0,
with Z[G] ⊗Z[G] A ∼= A in degree 0.
There is another complex which computes the cohomology of G, that of the inhomogeneous
G-cocycles, which has a more complicated differential but is more amenable to computation.
D EFINITION 12.9.12. Let A be a G-module, and let i ≥ 0.
a. The group of i-cochains of G with coefficients in A is the set of functions from Gi to A:
Ci (G, A) = { f : Gi → A}.
400 12. HOMOLOGICAL ALGEBRA

b. The ith differential d i = dAi : Ci (G, A) → Ci+1 (G, A) is the map

d i ( f )(g0 , g1 , . . . , gi ) = g0 · f (g1 , . . . gi )
i
+ ∑ (−1) j f (g0 , . . . , g j−2 , g j−1 g j , g j+1 , . . . , gi ) + (−1)i+1 f (g0 , . . . , gi−1 ).
j=1

We remark that C0 (G, A) is taken simply to be A, as G0 is a singleton set. The proof of the
following, which tells us that C· (G, A) is a cochain complex, is left to the reader.
L EMMA 12.9.13. For any i ≥ 0, one has d i+1 ◦ d i = 0.
We consider the cohomology groups of C· (G, A).
D EFINITION 12.9.14. Let i ≥ 0.
a. We set Z i (G, A) = ker d i , the group of i-cocycles of G with coefficients in A.
b. We set B0 (G, A) = 0 and Bi (G, A) = im d i−1 for i ≥ 1. We refer to Bi (G, A) as the group
of i-coboundaries of G with coefficients in A.
T HEOREM 12.9.15. The maps
ψ i : HomZ[G] (Z[Gi+1 ], A) → Ci (G, A)
defined by
ψ i (ϕ)(g1 , . . . , gi ) = ϕ(1, g1 , g1 g2 , . . . , g1 g2 · · · gi )
are isomorphisms for all i ≥ 0. This provides isomorphisms of complexes in the sense that ψ i+1 ◦
Di = d i ◦ ψ i for all i ≥ 0. Moreover, these isomorphisms are natural in the G-module A.
P ROOF. If ψ i (ϕ) = 0, then
ϕ(1, g1 , g1 g2 , . . . , g1 g2 · · · gi ) = 0
for all g1 , . . . , gi ∈ G. Let h0 , . . . , hi ∈ G, and define g j = h−1
j−1 h j for all 1 ≤ j ≤ i. We then have

ϕ(h0 , h1 , . . . , hi ) = h0 ϕ(1, h−1 −1

0 h1 , . . . , h0 hi ) = h0 ϕ(1, g1 , . . . , g1 · · · gi ) = 0.
Therefore, ψ i is injective. On the other hand, if f ∈ Ci (G, A), then defining
ϕ(h0 , h1 , . . . , hi ) = h0 f (h−1 −1
0 h1 , . . . , hi−1 hi ),
we have
ϕ(gh0 , gh1 , . . . , ghi ) = gh0 f ((gh0 )−1 gh1 , . . . , (ghi−1 )−1 ghi ) = gϕ(h0 , h1 , . . . , hi )
and ψ i (ϕ) = f . Therefore, ψ i is an isomorphism of groups.
That ψ · forms a map of complexes is shown in the following computation:
ψ i+1 (Di (ϕ))(g1 , . . . , gi+1 ) = Di (ϕ)(1, g1 , . . . , g1 · · · gi+1 )
= ϕ ◦ di+1 (1, g1 , . . . , g1 · · · gi+1 )
i+1
= ∑ (−1) j ϕ(1, g1, . . . , g1 · · · g j−2, g1 · · · g j , . . . , g1 · · · gi+1).
j=0
 12.9. GROUP COHOMOLOGY 401

The latter term equals

i
g1 ψ i (ϕ)(g2 , . . . , gi+1 ) + ∑ (−1) j ψ i (ϕ)(g1 , . . . , g j−2 , g j−1 g j , g j+1 , . . . , gi+1 )
j=1

+ (−1)i+1 ψ i (ϕ)(g1 , . . . , gi ),
which is d i (ψ i (ϕ)).
Finally, suppose that α : A → B is a G-module homomorphism. We then have
α ◦ ψ i (ϕ)(g1 , . . . , gi ) = α ◦ ϕ(1, g1 , . . . , g1 · · · gi ) = ψ i (α ◦ ϕ)(g1 , . . . , gi ),
hence the desired naturality. 
C OROLLARY 12.9.16. The ith cohomology group of the complex (HomZ[G] (Z[Gi+1 ], A), DiA )
is naturally isomorphic to H i (G, A).
C OROLLARY 12.9.17. The ith cohomology group of G with coefficients in A is
H i (G, A) = Z i (G, A)/Bi (G, A).
The cohomology groups measure how far the cochain complex C· (G, A) is from being exact.
We give some examples of cohomology groups in low degree.
L EMMA 12.9.18. We have
Z 1 (G, A) = { f : G → A | f (gh) = g f (h) + f (g) for all g, h ∈ G}
and B1 (G, A) is the subgroup of f : G → A for which there exists a ∈ A such that f (g) = ga−a for
all g ∈ G. In particular, if A is a Z[G]-module with trivial G-action, then H 1 (G, A) = Hom(G, A).
P ROOF. Let a ∈ A. Then d 0 (a)(g) = ga − a for g ∈ G, so ker d 0 = AG . That proves part a,
and part b is simply a rewriting of the definitions. Part c follows immediately, as the definition
of Z 1 (G, A) reduces to Hom(G, A), and B1 (G, A) is clearly (0), in this case. 
We remark that, as A is abelian, we have Hom(G, A) = Hom(Gab , A), where Gab is the maxi-
mal abelian quotient of G (i.e., its abelianization).
We turn briefly to an interesting use for second cohomology groups.
D EFINITION 12.9.19. A group extension of G by a G-module A is a short exact sequence of
groups
− E −
ι π
0→A→ →G→1
such that, choosing any section s : G → E of π, one has
s(g)as(g)−1 = g · a
for all g ∈ G, a ∈ A. Two such extensions E → E 0 are said to be equivalent if there is an isomor-
∼
phism θ : E − → E 0 fitting into a commutative diagram
0 /A /E /G /0
θ

0 /A / E0 /G / 0,
402 12. HOMOLOGICAL ALGEBRA

We denote the set of equivalence classes of such extensions by E (G, A).

D EFINITION 12.9.20. A factor set of a group G valued in a Z[G]-module A is a 2-cocycle
f : G2 → A satisfying f (1, g) = f (g, 1) = 0 for all g ∈ G.
L EMMA 12.9.21. Every 2-cocycle of a group is cohomologous to, i.e., has the same coho-
mology class as, a factor set.
P ROOF. The condition that F : G2 → A is a 2-cocycle is that
gF(h, k) + F(g, hk) = F(gh, k) + F(g, h)
for all g, h, k ∈ G. In particular, taking g = h = e, we have F(e, k) = F(e, e) and taking h = k = e,
we have gF(e, e) = f Fg, e). Note that for a 1-cochain c, we have
dc(g, h) = gc(h) − c(gh) + c(g).
In particular, if we set c(g) = c for all g ∈ G some fixed c ∈ A, then dc(g, h) = gc for all g ∈ G,
so if we take c = F(e, e) and replace F by f = F − dc, then f (e, k) = 0 and f (g, e) = 0 for all
g, k ∈ G. 
T HEOREM 12.9.22. The group H 2 (G, A) is in canonical bijection with E (G, A) via the map
induced by that taking a factor set f : G2 → A to the extension E f = A × G with multiplication
given by
(a, g) · (b, h) = (a + gb + f (g, h), gh)
This identification takes the identity to the semi-direct product A n G determined by the action of
G on A.
P ROOF. We check that E f so defined is a group. That it has identity (0, e) is clear from the
definition. Associativity is as follows:
(a + gb + f (g, h), gh) · (c, k) = (a + gb + f (g, h) + ghc + f (gh, k), ghk)
= (a + g(b + hc) + g f (h, k) + f (g, hk), ghk) = (a, g) · (b + hc + f (h, k), hk).
The inverse of (a, g) clearly has the form (b, g−1 ) for some b ∈ A, and we then must have a +
gb + f (g, g−1 ) = 0, so b = −g−1 a − g−1 f (g, g−1 ), and the inverse exists. That E f is a group
extension of G by A is now nearly immediate. Note also that E f is split if f = 0, ]since in that
case (0, g)(0, h) = (0, gh), so G is a subgroup.
Let c be a 1-cochain with dc(e, g) = dc(g, e) = 0, the latter property occurring if and only if
c(e) = 0. Consider the map ψ f ,c : E f → E f +dc given by
ψ f ,c (a, g) = (a − c(g), g).
We have
ψ f ,c (a + gb + f (g, h), gh) = (a + gb + f (g, h) − c(gh), gh)
= (a + gb + f (g, h) + dc(g, h) − gc(h) + c(g), gh) = ψc (a, g)ψc (a, h),
so ψc is a homomorphism, and it is clearly has inverse ψ f +dc,−c . Thus, we have a well-defined
map from H 2 (G, A) to E (G, A).
 12.9. GROUP COHOMOLOGY 403

It remains to construct an inverse, which we sketch as the computations all follow from
what we have already done. Given a group extension, we indeed always have a 2-cochain
f (g, h) : G2 → A defining the multiplication. We claim that f is a factor set. For this, asso-
ciativity again tells us that f is a 2-cocycle, and the fact that (0, e) is a two-sided identity forces
f (e, g) = f (g, e) = 0 for all g ∈ G. The resulting association is clearly inverse on the level of
extensions and cochains. If θ : E → E 0 is an isomorphism of group extensions of G by A, then
θ (0, g) = (−c(g), g) for some c : G → A that has the property that if the factor set is associated
to E is dc plus the factor set associated to E 0 . 
R EMARK 12.9.23. Theorem 12.9.22 tells us that E (G, A) also has a group structure, which
may also be given an explicit description. Given E and E 0 extensions of G by A, their product is
E ∗ E 0 = (E ×G E 0 )/h(a, −a) | a ∈ Ai.
This product is known as the Baer sum of the two extensions.
Let’s give a group-theoretic application of this description of H 2 (G, A).
P ROPOSITION 12.9.24 (Schur). Let G be a group of order mn, where m and n are relatively
prime positive integers. Then every abelian normal subgroup of order n has a complement in G
of order m.
P ROOF. Let N be an abelian normal subgroup of G, and set H = G/N. Let f : G2 → N be a
factor set corresponding to G as an extension of H by N by Theorem 12.9.22. For every h ∈ H,
let
t(h) = ∏ f (h, k),
k∈H
which makes sense as H is abelian. For h, h0 ∈ H, we have
∏ f (h, h0k) = ∏ f (h, k) = t(h).
k∈H k∈H
Now
f (h, h0 )mt(hh0 ) = ∏ f (h, h0)m f (hh0, k) = ∏ h f (h0, k)h−1 f (h, h0k) = ht(h0)h−1 · t(h).
k∈H k∈H
Let a, b ∈ Z be such that am + bn = 1. We then have
f (h, h0 ) = f (h, h0 )am+bn = dt(h, h0 )a .
Thus, f is a coboundary, so G is a split extension by Theorem 12.9.22. In particular, it contains
a subgroup of order n, isomorphic to H. 
R EMARK 12.9.25. Though we do not prove it, we have H i (G, A) = 0 for all i ≥ 1 whenever
G and A are finite of relatively prime order. In fact, for any finite group G and G-module A, the
exponent of H i (G, A) divides the order of G.
D EFINITION 12.9.26. A Hall subgroup of a finite group is a subgroup with relatively prime
order and index.
We can extend Proposition 12.9.24 from abelian to arbitrary normal subgroups.
404 12. HOMOLOGICAL ALGEBRA

T HEOREM 12.9.27 (Schur-Zassenhaus). Every normal Hall subgroup of a finite group has a
complement.
P ROOF. Let N be a normal Hall subgroup of G of order n and index m. If N is abelian, then
the result follows from Proposition 12.9.24. Suppose the result holds true in the case of normal
subgroups of order less than n ≥ 2. Let p be a prime dividing n. Let P be a Sylow p-subgroup of
G. Then PN/N has p-power order dividing m. Since m and n are relatively prime, this forces P
to be contained in N. In other words, the Sylow p-subgroups of N and G are the same. Now
[G : NG (P)] = n p (G) = n p (N) = [N : NN (P)],
so [NG (P) : NN (P)] = m. On the other hand, NN (P)/P has order prime to m and less than n, being
properly contained in N/P. Furthermore, NN (P) = N ∩ NG (P) is normal in NG (P). By induction
on n, we have that there exists a subgroup K of NG (P) with K/P isomorphic to NG (P)/NN (P)
and |K/P| = m.
Since P is a p-group, its center Z = Z(P) is nontrivial. It is also a characteristic subgroup of
P, so it is normal in K. By induction, P/Z has a has a complement in K/Z, equal to H/Z for
some subgroup H of K, which necessarily has order m. This group H is the desired complement
to N. 

12.10. Galois cohomology

We briefly consider the cohomology of finite Galois extensions. We have the following gen-
eralization of Hilbert’s Theorem 90, which also has the same name.
T HEOREM 12.10.1 (Hilbert’s Theorem 90). Let L/K be a finite Galois extension with Galois
group G. Then H 1 (G, K × ) = 0.
P ROOF. Let f : G → L× be a 1-cocycle. We view the elements σ ∈ G as abelian characters
L× → L× . As distinct characters of L× , these characters form a linearly independent set. The sum
∑σ ∈G f (σ )σ is therefore a nonzero map L× → L. Let α ∈ L× be such that z = ∑σ ∈G f (σ )σ (α) 6=
0. For any τ ∈ G, we have

τ −1 (z) = ∑ τ −1( f (σ )) · τ −1σ (α) = ∑ τ −1( f (τσ ))σ (α)

σ ∈G σ ∈G
−1
= ∑τ ( f (τ) · τ f (σ ))σ (α) = τ −1 ( f (τ)) ∑ f (σ )σ (α) = τ −1 ( f (τ))z.
σ ∈G σ ∈G

Thus,
z
f (τ) = ,
τ(z)
so f is the 1-coboundary of z−1 . 

To see how this implies Hilbert’s theorem 90 in the case of finite cyclic extensions, we prove
the following result on the cohomology of cyclic groups.
 12.10. GALOIS COHOMOLOGY 405

P ROPOSITION 12.10.2. Let G be a finite cyclic group and A be a Z[G]-module. Then for
i ≥ 1, we have (
AG /NG A if i is even,
H i (G, A) ∼
=
A[NG ]/IG A if i is odd,
where A[NG ] is the kernel of multiplication by NG on A.
P ROOF. Let g be a generator of G, and consider the augmented resolution of Z given by
N G g−1 G N g−1 ε
· · · → Z[G] −→ Z[G] −−→ Z[G] −→ Z[G] −−→ Z[G] → − Z → 0,
where ε is the augmentation maps. Note that HomZ[G] (Z[G], A) ∼ = A by evaluation at 1, and the
map NG (resp., g − 1) on Z[G] induces NG (resp., g − 1) on A via these isomorphisms. The groups
H i (G, A) are then the cohomology groups of the complex
g−1 G N g−1
A −−→ A −→ A −−→ A → · · · ,
which have the desired form. 
R EMARK 12.10.3. Suppose that L/K is finite cyclic with Galois group G having generator
σ . Proposition 12.10.2 implies that
ker NL/K
H 1 (G, K × ) ∼
= ,
{σ (α)/α | α ∈ L× }
which is trivial by Theorem 12.10.1. This is exactly the statement of Hilbert’s Theorem 90 for
finite cyclic extensions.
We next see how we can use Galois cohomology to study Kummer theory.
P ROPOSITION 12.10.4. Suppose that L/K is a finite extension with Galois group G. Let n be
a positive integer not divisible by the characteristic of K. Then there is an isomorphism
K × ∩ L×n
H 1 (G, µn ) ∼
= ,
K ×n
where the class of an element a ∈ K × ∩ L×n in the quotient corresponds to the class of the cocycle
σ (α)
χa (σ ) = ,
α
where α ∈ L× with α n = a
P ROOF. The short exact sequence
n
1 → µn (L) → L× →
− L×n → 1
of G-modules gives rise to a long exact sequence
n
0 → µn (K) → K × →
− K × ∩ L×n −
∂
→ H 1 (G, µn ) → 0,
where Hilbert’s Theorem 90 gives the final equality. That a ∈ K × ∩ L×n is sent to the class of χa
follows from the definition of the connecting homomorphism by the Snake lemma. 
406 12. HOMOLOGICAL ALGEBRA

This leads to the following definition.

D EFINITION 12.10.5. Let n be a positive integer not divisible by the characteristic of K. For
a ∈ K × , a Kummer cocycle attached to K is a map χa : GK → µn given by
σ (α)
χa (σ ) = ,
α
where α ∈ (K sep )× with α n = a.
R EMARK 12.10.6. The Kummer cocycle χa in Definition 12.10.5 actually depends on the
choice of nth root of a up to a 1-coboundary of an element of µn . If µn ⊆ K, however, it is unique
and is the Kummer character of a. In this case, Proposition 12.10.4 reduces to
Hom(G, µn ) ∼
= ∆/K ×n ,
where ∆ = K × ∩ L×n . This in turn yields the perfect pairing of Kummer duality.
We next turn to the question of the structure of H 2 (G, L× ) for a finite Galois extension L/K
with Galois group G. We fix such an extension L/K with Galois group G in what follows.
D EFINITION 12.10.7. A central simple algebra over a field K is a simple K-algebra with
center equal to K.
E XAMPLE 12.10.8. Any matrix algebra Mn (D) over a division algebra D is a central simple
algebra over the center Z(D), which is a field. For instance, if H denotes the ring of quaternions,
then Mn (H) is a central simple R-algebra.
P ROPOSITION 12.10.9. Let f ∈ Z 2 (G, L× ) be a factor set. Let B f be an L-vector space with
basis bσ for g ∈ G. Define a multiplication on B f as the unique binary operation extending the
scalar multiplication L × B f → B f and satisfying
bσ α = σ (α)bσ and bσ bτ = f (σ , τ)bσ τ .
for σ , τ ∈ G and α ∈ L. Then B f is a central simple K-algebra with identity b1 .
P ROOF. We have
(bσ bτ )bρ = f (σ , τ)(bσ τ bρ ) = f (σ , τ) f (σ τ, ρ)bσ τρ = σ ( f (τ, ρ)) f (σ , τρ)bσ τρ
= σ ( f (τ, ρ))(bσ bτρ ) = bσ ( f (τρ)bτρ ) = bσ (bτ bρ ).
It follows that B f is an associative L-algebra with K in its center. Note that b1 = 1 in the ring B f
since f (σ , 1) = f (1, σ ) = 1 for all σ ∈ G.
Let β ∈ L× generate L/K. Let
z= ∑ ασ bσ ∈ Z(G).
σ ∈G
. Then zβ = β z, so
∑ (ασ σ (β ) − β ασ )bg = 0,
σ ∈G
and therefore σ (β ) = β for all σ ∈ G with ασ = 0. Since β is a generator of L/K, this forces
ασ = 0 for all σ 6= 1, so z = α1 ∈ K. Thus, Z(B f ) = K.
 12.10. GALOIS COHOMOLOGY 407

Next, let I be a nonzero ideal of B f , and let x ∈ B f be an element with a minimal number
k of nonzero coefficients in its expression as an L-linear combination of elements of G. If σ
and τ are distinct elements of G for which x has nonzero coefficients, then σ (β ) 6= τ(β ). Then
x − τ(β )xβ −1 ∈ I, but its bτ -coefficient is now zero, while its bσ -coefficient is not, and it has no
nonzero coefficients that x does not have. This contradicts the minimality of k, forcing it to be 1.
Thus, x = αbσ for some α ∈ L× and σ ∈ G. But such an x is a unit in B f , so I = B f . Thus, B f is
a simple ring. 
D EFINITION 12.10.10. For a factor set f : G → L× , the K-algebra B f of Proposition 12.10.9
is the crossed product algebra of f .
 CHAPTER 13

Representation theory

13.1. Semisimple modules

The following definitions will be of special interest in the case of a group ring over a field.
D EFINITION 13.1.1. A module M over a ring R is simple, or irreducible, if it has no nonzero,
proper R-submodules. Otherwise, M is said to be reducible.
D EFINITION 13.1.2. A module M over a ring R is indecomposable if it is not the direct sum
of two proper submodules.
D EFINITION 13.1.3. A module M over a ring R is semisimple, or completely reducible, if it
is a direct sum of irreducible submodules.
R EMARK 13.1.4. By definition, a module is simple if and only if it is both semisimple and
indecomposable.
E XAMPLES 13.1.5.
a. Any division ring D is simple as a left module over itself, as it has no nontrivial left ideals.
b. Any vector space V over a field F is semisimple as an F-module, in that it has a basis that
allows us to express it (up to isomorphism) as a direct sum of copies of F.
c. The ring Z is indecomposable as a Z-module, but it is not simple, as it contains proper,
nontrivial submodules nZ for n ≥ 2.
d. Any simple Z-module is a simple abelian group, so isomorphic to Z/pZ for some prime
p.
e. The Z-module Z ⊕ Z is neither semisimple nor indecomposable, as it is not a direct sum
of simple Z-modules.
f. Let R be the ring of upper-triangular matrices in M2 (F) for F a field, and consider the
R-module M = F 2 under left multiplication of column vectors. Then M has a simple submodule
N = F · e1 , so M is not simple. Moreover, M is not semisimple, as M = F · v for any v ∈
/ N, so N
has no complement in M.
Semisimple modules have the following equivalent characterizations.
P ROPOSITION 13.1.6. Let M be an R-module. The following are equivalent:
i. M is semisimple.
ii. M is a sum of simple submodules.
409
410 13. REPRESENTATION THEORY

iii. Every submodule of M is a direct summand.

P ROOF. That (i) implies (ii) is clear. As for (ii) implies (iii), let N be a submodule of
M
M= Mi ,
i∈I
where the Mi are simple. Then N ∩ Mi is either 0 or Mi for each i, and N is the direct sum of the
Mi for which N ∩ Mi = Mi .
That (iii) implies (i) is proven as follows. We first claim that any nonzero R-module M
contains a nonzero simple submodule. To see this, choose m ∈ M, and replace M with Rm
without loss of generality. Let N be a maximal R-submodule of M not containing n, which exists
by Zorn’s Lemma. Then M = N ⊕N 0 for some nonzero R-submodule N 0 . Now N 0 must be simple,
since any Q ⊆ N 0 has N ⊕ Q containing a and therefore equals M.
Now consider the nonempty set X of semisimple submodules of M under inclusion. The
union of any chain C in X is semisimple (as the reader may check), so X has a maximal element
N by Zorn’s lemma. Let N 0 be a complement to N in M, so M = N ⊕ N 0 . If N 0 is nonzero,
then N 0 contains a simple submodule Q by the claim, and N ⊕ Q is semisimple, contradicting the
maximality of M. So, M = N is semisimple. 
We define semisimple rings in a manner that does not obviously relate to simple rings.
D EFINITION 13.1.7. A nonzero ring is semisimple if it is semisimple as a left module over
itself.
The following contains equivalent conditions for a ring to be semisimple.
T HEOREM 13.1.8. The following conditions on a nonzero ring R are equivalent:
i. R is semisimple,
ii. every R-module is semisimple,
iii. every R-module is projective,
iv. every R-module is injective.
P ROOF. Suppose that R is semisimple, and let M be an R-module. Then M is a sum of its
cyclic submodules, so by Lemma 13.1.6, it suffices to see that quotients Q of R are semisimple.
Again employing Lemma 13.1.6, the kernel I of the quotient map R → Q is a direct summand,
so Q is isomorphic to a left ideal of R, which is semisimple as R is.
That (ii) implies (iii) is an immediate consequence of Lemma 13.1.6. Every surjection of R-
modules is split if and only if every injection of R-modules is split by Proposition 12.4.3, so (iii)
and (iv) are equivalent, noting Proposition 12.4.8 and Lemma 12.4.13. Finally, (iv) tells us that
every R-submodule of an R-module is a direct summand, so is semisimple by Lemma 13.1.6. 
We claim that simple rings are indeed semisimple, so long as we assume that descending
chains of left or right ideals terminate. This can be seen directly for matrix rings over division
rings, using Morita equivalence.
D EFINITION 13.1.9. A ring R is left artinian (resp., right artinian if it satisfies the descending
chain condition on left ideals (resp., right ideals).
 13.1. SEMISIMPLE MODULES 411

L EMMA 13.1.10. If a ring R is the sum of a collection of its nonzero left ideals, then it is also
a sum of a finite subcollection.
P ROOF. If {Ix | x ∈ X} is a set of nonzero left ideals of R such that R ∼ = ∑x∈X Ix as left-
modules, then we can write 1 = ∑nj=1 a j for some n ≥ 1, where a j ∈ Ix j for some x j ∈ I. But then
the left ideals Ix j with 1 ≤ j ≤ n generate R as a left R-module. 
C OROLLARY 13.1.11. A semisimple ring R is left artinian, isomorphic as an R-module to the
direct sum of its finitely many minimal left ideals.
P ROOF. By definition, R is isomorphic to the direct sum of its minimal ideals. Since the sum
is direct, no proper subcollection of the minimal ideals generates R. Lemma 13.1.10 then tells us
that the collection of minimal ideals must be finite. It follows that R is left artinian. 
P ROPOSITION 13.1.12. Let R be a left (or right) artinian simple ring. Then R is semisimple.
P ROOF. Let R be left artinian and simple. We first claim that R has a simple R-submodule
(i.e., left ideal). For this, construct a possibly finite sequence of left ideals Ji of R recursively,
starting with J1 = R, and then for i ≥ 1, taking Ji+1 to be a proper simple submodule if Ji is
not simple. Since R is left artinian, we must have that the process terminates, so R has a simple
submodule.
Now, consider the nonzero sum M of all distinct simple R-submodules of R. Let N be a
simple submodule of R, and let r ∈ R. Then Nr is isomorphic to a quotient of n, so is either 0
or simple. In particular, Nr is contained in M, and therefore Mr ⊆ M. Thus, M is not only a left
ideal of R, but a right ideal as well, and therefore M = R.
By Lemma 13.1.10, the R-module R is then a finite sum of distinct simple left ideals: say R
is the sum of Ni simple for 1 ≤ i ≤ k, where k is minimal. If the intersection Ni with the sum Mi
of the N j for v 6= i is nonzero, then it must equal Ni , as Ni is simple. But then Ni ⊆ Mi , so Mi = R,
which contradicts the minimality of k. So, R is in fact the direct sum of the Ni , as required. 
The following is an easy but very useful fact regarding homomorphisms of simple modules.
L EMMA 13.1.13 (Schur’s lemma). Let R be a ring, and let M and N be simple R-modules.
Then any nonzero homomorphism f : M → N is an isomorphism.
P ROOF. The kernel of f is a proper R-submodule of M, hence zero, and the image of f is a
nonzero R-submodule of N, hence N. Thus f is bijective. 
Since every nonzero R-linear endomorphism of a simple module is invertible by Schur’s
lemma, we have the following corollary.
L EMMA 13.1.14. Let R be a nonzero ring. The ring EndR (M) of R-linear endomorphisms of
a simple module M is a division ring.
Recall that a ring is simple if it has no nonzero ideals, and by Remark 3.8.8, matrix rings over
division algebras are simple. We have the following consequence of Schur’s lemma
L EMMA 13.1.15. Let M be a simple R-module, and let n ≥ 1. Then EndR (M n ) ∼ = Mn (D),
n
where D is the division ring EndR (M). In particular, EndR (M ) is a simple ring.
412 13. REPRESENTATION THEORY

P ROOF. We define a homomorphism

Φ : Mn (D) −∼
→ EndR (M n )
on a matrix C = (φi j ) ∈ Mn (D) by
!
n n
Φ(C)(m1 , . . . , mn ) = ∑ φ1 j (m j ), . . . , ∑ φn j (m j ) .
j=1 j=1
Every endomorphism φ ∈ EndR (M n ) is determined uniquely by the collection of maps φi j =
π j ◦ φ ◦ ιi ∈ EndR (M), where πi and ιi denote the ith projection and inclusion maps, so this is
one-to-one and onto. 
We can improve this lemma to treat a finite direct sum of arbitrary simple modules.
L EMMA 13.1.16. Let R be a nonzero ring. Let M be an R-module that is isomorphic to a
direct sum N1n1 ⊕ N2n2 ⊕ · · · ⊕ Nknk with the Ni mutually nonisomorphic simple modules and ni ≥ 1
for 1 ≤ i ≤ k. Then we have an isomorphism of rings
k
EndR (M) ∼
= ∏ Mni (Di ),
i=1
where Di is the division ring EndR (Ni ).
P ROOF. Let πi : M → Nini and ιi : Nini → M be the projection and inclusion maps. Any R-
module endomorphism f of M determines and is determined by the homomorphisms fi, j = π j ◦
n n
f ◦ ιi : Nini → N j j for 1 ≤ i, j ≤ n. But HomR (Ni , N j ) = 0 for i 6= j, so HomR (Nini , N j j ) = 0 for
i 6= j as well. Therefore, the product of restriction maps to Nini yields the first of the isomorphisms
in
k k
EndR (M) ∼
= ∏ End R (N ni ∼
) = iMn (Di ),
∏ i
i=1 i=1
where the second isomorphism is by Lemma 13.1.15 
Evaluation at 1 gives the isomorphism in the following lemma.
L EMMA 13.1.17. We have EndR (R) → Rop as rings.
We now come to the Artin-Wedderburn theorem, which classifies semisimple rings.
T HEOREM 13.1.18 (Artin-Wedderburn theorem). A nonzero ring is semisimple if and only if
it is isomorphic to a direct product of matrix algebras over division rings.
P ROOF. For any nonzero ring R, we have an isomorphism
Rop −∼
→ End (R), R r 7→ (s 7→ sr).
Supposing that R is semisimple, we have by Corollary 13.1.11 that R ∼ n n n
= N1 1 ⊕ N2 2 ⊕ · · · ⊕ Nk k
with the Ni mutually nonisomorphic simple left R-modules and ni ≥ 1 for 1 ≤ i ≤ k. Noting
Lemma 13.1.17, we then have
k
= EndR (R) ∼
Rop ∼ = ∏ Mni (Di ),
i=1
 13.1. SEMISIMPLE MODULES 413

where Di = EndR (Ni ) is a division ring. By taking the opposite ring of both sides, we obtain
k k
R∼
op
= ∏ Mni (Di )op = ∏ Mni (Di ),
i=1 i=1
op
and Di is a division ring as well.
On the other hand, suppose that R ∼ = ∏ki=1 Mni (Ei ) for some division algebras Ei . The matrix
rings Mni (Ei ) are semisimple left modules over Mni (Ei ), isomorphic to a direct sum of the simple
submodules of column vectors. They are also then semisimple as modules for the larger ring
R, since the action of R on Mni (Ei ) by left multiplication factors through Mni (Ei ). Thus, R is a
semisimple ring as a direct sum of these as a left R-module. 
Here are some corollaries. The first follows directly from Proposition 13.1.12 and the Artin-
Wedderburn theorem.
C OROLLARY 13.1.19. A nonzero ring is left artinian and simple if and only if it is isomorphic
to a matrix ring over a division ring.
Consequently, we have the following, which explains the relationship between simple and
semisimple rings.
C OROLLARY 13.1.20. A nonzero ring is semisimple if and only if it is isomorphic to a finite
direct product of left artinian simple rings.
For algebras over a field, we obtain Wedderburn’s theorem.
C OROLLARY 13.1.21 (Wedderburn). An algebra over a field F is semisimple if and only if it
is a product of finite-dimensional simple F-algebras, and these simple algebras are isomorphic
to matrix rings over finite-dimensional division algebras over F.
The following greatly limits the choice of finite-dimensional division algebras over alge-
braically closed fields.
P ROPOSITION 13.1.22. Let D be a finite-dimensional division algebra over an algebraically
closed field F. Then D = F.
P ROOF. Let γ ∈ D. Note that γ commutes with every element of F, so F(γ) is a field. Since
D is finite-dimensional over F, the elements γ i for i ≥ 0 are linearly dependent over F, and
therefore γ is algebraic over F. Thus F(γ) = F, which is to say γ ∈ F. 
C OROLLARY 13.1.23. Let A be a finite-dimensional, semsimple F-algebra, where F is an
algebraically closed field. Then A is isomorphic to a direct product of matrix algebras with
F-entries.
For commutative rings, we have this:
C OROLLARY 13.1.24. A commutative semisimple ring is a finite direct product of fields.
A finite-dimensional commutative semisimple algebra over a field F is isomorphic to a direct
product of finite field extensions of F.
414 13. REPRESENTATION THEORY

D EFINITION 13.1.25. Let R be a ring. An idempotent in R is a nonzero element e ∈ R such

that e2 = e.
D EFINITION 13.1.26. Let R be a ring. We say two idempotents e, f ∈ R are orthogonal if
e f = f e = 0.
R EMARK 13.1.27. Any finite sum of orthogonal idempotents is also an idempotent.
D EFINITION 13.1.28. We say that an idempotent e in a ring R is primitive if eR is a subring
of R that is not a product of two subrings of R.
L EMMA 13.1.29. Let R be a nonzero ring and k ≥ 1. Then R = R1 × R2 × · · · × Rk with Ri
rings for 1 ≤ i ≤ k if and only if there exist mutually orthogonal idempotents e1 , e2 , . . . , ek in Z(R)
such that e1 + e2 + · · · + ek = 1. These may be chosen so that Ri = (ei ) in R.
P ROOF. If R = ∏ki=1 Ri , then let ei be the identity in Ri . The ei are then clearly mutually
orthogonal, central idempotents. Set e = ∑ki=1 ei . If 1 = (r1 , r2 , . . . , rk ) ∈ R, then
e = e · 1 = (e1 r2 , e2 r2 , . . . , ek rk ) = r.
Conversely, given e1 , e2 , . . . , ek , set Ri = Rei in R. For any r ∈ R, we have r = ∑ki=1 rei , so
R = ∑ki=1 Ri . If ri ∈ Ri for each i, then set r = ∑ki=1 ri . This satisfies re j = r j for each 1 ≤ j ≤ k,
so r = 0 if and only if each r j = 0, and thus R = ki=1 Ri as left R-modules. Since each ei is
L

central in R, each Ri is also a right ideal and a ring with unit element ei (though not a subring of
R if k ≥ 2, since then ei 6= 1). 
E XAMPLE 13.1.30. If R is a direct product of matrix rings, then it has a set of mutually
orthogonal idempotents consisting of the identity matrices in those rings.
L EMMA 13.1.31. Let R = ∏ki=1 Ri be a direct product of rings Ri , and let ei be identity element
of Ri . Let M be a left R-module. Then M = ki=1 ei M as an R-module.
L

P ROOF. Any m ∈ M can be written as m = e1 m + e2 m + · · · + ek m, so M = ∑ki=1 ei M. If

m1 + m2 + · · · + mk = m with mi ∈ ei M for 1 ≤ i ≤ k, then
mi = ei (m1 + m2 + · · · + mk ) = ei m
for each i, so the representation of m as an element of the sum is unique. Therefore, M is the
direct sum of the ei M. 

13.2. Representations of groups

Let G be a group.
P ROPOSITION 13.2.1. Let R be a commutative ring, let G be a group, and let M be an R-
module. There is a one-to-one correspondence between
i. homomorphisms ρ : G → AutR (M),
ii. R[G]-module structures given by R-bilinear maps φ : R[G] × M → M
such that ρ corresponds to a unique φ with ρ(g)(m) = φ (g, m) for all g ∈ G and m ∈ M.
 13.2. REPRESENTATIONS OF GROUPS 415

P ROOF. If ρ : G → AutR (M) is a homomorphism, then we define

!
(13.2.1) ∑ ag g ·m = ∑ agρ(g)(m)
g∈G g∈G

for ∑g∈G ag g ∈ R[G] and m ∈ M. For a fixed m, this provides the unique R[G]-module homo-
morphism R[G] → M that sends g to ρ(g)(m) by the R-freeness of R[G]. In other words, the
operation R[G] × M → M is left distributive. Since ρ(g)(m + m0 ) = ρ(g)(m) + ρ(g)(m0 ) for
g ∈ G and m, m0 ∈ M in that ρ(g) ∈ AutR (M), right distributivity follows from the definition in
(13.2.1) as well.
Conversely, given an R[G]-module M, we define ρ : G → AutR (M) by ρ(g)(m) = g · m for
g ∈ G and m ∈ M. Note that
ρ(g)(m + m0 ) = g(m + m0 ) = gm + gm0 = ρ(g)(m) + ρ(g)(m0 )
and ρ(g)(rm) = g(rm) = r(gm) = rρ(g)(m), so ρ(g) is indeed an element of AutR (M). More-
over,
ρ(gg0 )(m) = (gg0 )m = g(g0 m) = ρ(g)(g0 m) = ρ(g)(ρ(g0 )(m)) = (ρ(g) ◦ ρ(g0 ))(m),
so ρ is a homomorphism. 
R EMARK 13.2.2. To give an R[G]-module structure on an R-module M, it suffices to give an
operation G × M → M such that the map g : M → M defined by left multiplication is R-linear.
R EMARK 13.2.3. The trivial R[G]-module R on which g · r = r for all g ∈ G and r ∈ R
corresponds to the homomorphism ρ : G → AutR (R) ∼
= R× with ρ(g) = 1 for all g ∈ G.
E XAMPLE 13.2.4. Let R be a commutative ring and G be a group. We may view R[G] as
a left R[G]-module under left multiplication. This corresponds to the homomorphism ρ : G →
AutR (R[G]) that takes g to left multiplication by g on R[G].
We now focus on the special case that R is a field, which yields group representations. From
now on in this section, we let F denote a field.
D EFINITION 13.2.5. A representation, or group representation, of a group G over a field F
is an F-vector space V , together with a homomorphism ρ : G → AutF (V ). We also say that V is
an F-representation of G.
R EMARK 13.2.6. By Proposition 13.2.1, to make an F-vector space V into an F[G]-module V
is equivalent to providing a homomorphism ρ : G → AutF (V ) that makes it into a representation
of G.
D EFINITION 13.2.7. We say that a representation ρ : G → AutF (V ) is finite-dimensional if
V is a finite-dimensional F-vector space, in which case dimF V is its dimension, also known as
its degree.
Representations form one of the most important tools in the study of the structure of groups.
416 13. REPRESENTATION THEORY

E XAMPLE 13.2.8. Let G be a subgroup of GLn (F). Then the inclusion ρ : G → GLn (F)
defines a representation of G, and this turns F n into an F[G]-module, where g ∈ G acts on v ∈ F n
by left multiplication of the column vector v by the matrix corresponding to g.
E XAMPLE 13.2.9. The representation ρ : R → GL2 (R) given by
 
cos θ sin θ
ρ(θ ) = .
− sin θ cos θ
is a two-dimensional real representation of the additive group R.
D EFINITION 13.2.10.
a. The trivial representation of G over F is F with the trivial G-action.
b. The regular representation of G over F is F[G] with the action of F[G] on itself by left
multiplication.
R EMARK 13.2.11. Two F-representations V and W of G are isomorphic if V and W are
isomorphic as F[G]-modules. Phrased in terms of the corresponding homomorphisms ρV and
ρW , this says that ρV and ρW are conjugate by the isomorphism ϕ : V → W : that is, ρW (g) =
ϕ ◦ ρV (g) ◦ ϕ −1 for all g ∈ G.
E XAMPLES 13.2.12. Let V and W be F-representations of a group G.
a. The F-vector space V ⊗F W is a representation of G with respect to the diagonal G-action
g · (v ⊗ w) = gv ⊗ gw for g ∈ G, v ∈ V and w ∈ W .
b. The F-vector space HomF (V,W ) is a representation of G with respect to the G-action
(g · ϕ)(v) = gϕ(g−1 v) for g ∈ G, ϕ ∈ HomF (V,W ), and v ∈ V .
As a special case, we have the following.
D EFINITION 13.2.13. Let V be an F-representation of a group G. The dual representation to
V is V ∗ = HomF (V, F).
The reader will easily check the following.
L EMMA 13.2.14. Let V and W be F-representations of a group G, and suppose that W is
finite-dimensional. Then HomF (V,W ) ∼
= V ∗ ⊗F W .
T ERMINOLOGY 13.2.15. We speak of F-representations of a group G as being simple, inde-
composable, and so forth, if the F[G]-modules that define them have these properties.
D EFINITION 13.2.16. An F-representation V of a group G is called faithful if ρV : G →
AutF (V ) is injective.
D EFINITION 13.2.17. A subrepresentation W of an F-representation V of a group G is an
F[G]-submodule of V .
R EMARK 13.2.18. An irreducible (i.e., simple) representation is one that has no nonzero,
proper subrepresentations.
E XAMPLES 13.2.19.
 13.3. MASCHKE’S THEOREM 417

a. All one-dimensional representations of a group are irreducible.

b. Let D p = hr, si be the dihedral
group of prime
order p, and let ρ : D p → GL2 (F p ) be
−1 0 1 1
the representation with ρ(s) = 0 1 and ρ(r) = 0 1 . Then ρ is indecomposable but not irre-
ducible, since the F p -submodule W of V = F2p spanned by e1 is left stable by (i.e., is closed under)
the action of D p , so is a subrepresentation. On the other hand, W does not have a complement in
V (i.e., the only line in F2p that is stabilized by D p is W ).
c. The regular representation of a finite group is faithful, whereas the trivial representation is
not faithful unless the group is trivial.
Let us rephrase Schur’s lemma in the context of representations.
L EMMA 13.2.20. Let V be an irreducible F-representation of G. Then EndF[G] (V ) is a divi-
sion algebra over F.
P ROOF. This is an immediate consequence of Lemma 13.1.14, noting that the the endomor-
phisms given by multiplication by elements of F are contained in the center of EndF[G] (V ). 
By Proposition 13.1.22, this has the following corollary.
C OROLLARY 13.2.21. Let V be a finite-dimensional irreducible F-representation of a group
G, where F is algebraically closed. Then EndF[G] (V ) ∼
= F.
D EFINITION 13.2.22. Let V and W be representations of G over a field F, with V semisimple
and W irreducible. The multiplicity of W in V is the largest nonnegative integer n such that W n
is isomorphic to a subrepresentation of V . We say that W occurs with multiplicity n in V .
L EMMA 13.2.23. Let V be an F-representation of a finite group G. Let E/F be a field
extension. Then E ⊗F V is an E[G]-module under the action g · (α ⊗ v) = α ⊗ gv with the same
character as V .
P ROOF. Note that E[G] ∼= E ⊗F F[G], and the action described is just the usual action of a
tensor product of algebras on a tensor product of modules over them. 
D EFINITION 13.2.24. For an F-representation V of a group G and a field extension E/F, the
E-representation E ⊗F V is called the base change of V from F to E.

13.3. Maschke’s theorem

In this section, we let G be a finite group, and we let F be a field of characteristic not dividing
the order of G.
T HEOREM 13.3.1 (Maschke’s theorem). Let G be a finite group, let F be a field of character-
istic not dividing |G|, and let V be a representation of G over F. Then every subrepresentation
of V is a direct summand of V as an F[G]-module.
P ROOF. Let W be an F[G]-submodule of V . As F-modules, we know that we can find a
basis B0 of W contained in a basis B of V . We then have a projection map p : V → W given by
418 13. REPRESENTATION THEORY

p(∑v∈B av v) = ∑w∈B0 aw w, where av ∈ F equals zero for almost all v. This is an F-linear transfor-
mation that restricts to the identity on W , but it is not necessarily a F[G]-module homomorphism.
So, define
1
π : V → V, π(v) = ∑ g−1 p(gv)
|G| g∈G
for v ∈ V . Then π is clearly F-linear, and moreover
1 1
π(hv) = ∑ g−1 p(ghv) = ∑ (kh−1)−1 p(kv) = hp(v),
|G| g∈G |G| k∈G
so π is an F[G]-module homomorphism. Since W is an F[G]-submodule of V , the image of π is
contained in W , and for w ∈ W , we have
1 1
π(w) = ∑ g−1 p(gw) = ∑ g−1 gw = w.
|G| g∈G |G| g∈G
In particular, the inclusion of W in V splits π, so W is a direct summand of V as an F[G]-
module. 
As a consequence of Maschke’s theorem and Wedderburn theory, or more specifically, The-
orem 13.1.8, we have the following corollary.
C OROLLARY 13.3.2. The group ring F[G] is a semisimple F-algebra, which is to say iso-
morphic to a finite direct product of matrix rings over finite-dimensional division algebras over
F.
This in turn yields the following corollaries. For the first, see Corollary 13.1.24.
C OROLLARY 13.3.3. Let G be a finite abelian group, and let F be a field of characteristic
not dividing |G|. Then F[G] is a direct product of finite field extensions of F.
E XAMPLE 13.3.4. By the Chinese remainder theorem, we have
Q[Z/pZ] =∼ Q[x]/(x p − 1) ∼
= Q[x]/(x − 1) × Q[x]/Φ p (x) ∼
= Q × Q(ζ p ),
where ζ p is a primitive pth root of unity in C. Note, however, that if we take F p in place of Q,
then we obtain
F p [Z/pZ] ∼
= F p [x]/(x p − 1) ∼
= F p [x]/(x − 1) p ∼
= F p [y]/(y p )
for y = x − 1, which is not a direct product of matrix rings over fields.
For the following, see Corollary 13.1.23.
C OROLLARY 13.3.5. If F is algebraically closed, then F[G] is isomorphic to a direct product
of matrix algebras over F.
P ROPOSITION 13.3.6. Suppose that F is algebraically closed, and write
k
F[G] ∼
= ∏ Mni (F)
i=1
 13.3. MASCHKE’S THEOREM 419

for some k ≥ 1 and ni ≥ 1 for 1 ≤ i ≤ k. Then k is equal to the number of conjugacy classes of
|G|.
P ROOF. First, we remark that Z(Mni (F)) = F, so dimF Z(F[G]) = k. For any g ∈ G, we form
out of its conjugacy class Cg the sum Ng = ∑h∈Cg h. If we let G act on F[G] by conjugation, then
the G-invariant module for this action is Z(F[G]). Moreover, Ng lies in this invariant group. That
is, the action restricts to an action on G which preserves conjugacy classes, so
kNg k−1 = ∑ khk−1 = ∑ h = Ng,
h∈Cg h∈Cg

The elements Ng , where g runs over a set S of representatives for the conjugacy classes of G, are
linearly independent as they are sums over disjoint sets of group elements. And if z = ∑g∈G ag g ∈
Z(F[G]) and k ∈ G, then

∑ agg = ∑ agkgk−1 = ∑ ak−1gk g,

g∈G g∈G g∈G

so ag = ak−1 gk for all k, so ah = ag for all h ∈ Cg . Thus, z is in the F-span of the elements Ng .
Thus, we have k = |S|, the number of conjugacy classes. 
R EMARK 13.3.7. If F is algebraically closed, then we may by Corollary 13.3.5 write
k
F[G] ∼
= ∏ Mni (F)
i=1
for some k ≥ 1 and ni ≥ 1 for 1 ≤ i ≤ k. Then G has k isomorphism classes of irreducible
representations of dimensions n1 , n2 , . . . , nk . Let Vi be the ith of these, with dimF Vi = ni . Then
Vi occurs with multiplicity ni in the regular representation R[G], which is to say that R[G] ∼ =
V1n1 ⊕V2n2 ⊕· · ·⊕Vknk . Under this isomorphism, each copy of Vi is identified with one of the simple
left ideals in Mni (F), isomorphic to the module F ni of column vectors for this ring. Counting
dimensions tells us that
k
∑ n2i = |G|.
i=1

E XAMPLE 13.3.8. The group S3 has 3 conjugacy classes, so there are 3 isomorphism classes
of irreducible representations of G, and the sum of the squares of their dimensions are 6, so they
have dimensions 1, 1, and 2. Thus, we have
C[S3 ] ∼
= C × C × M2 (C).
The two one-dimensional representations correspond to homomorphisms G → C× , factoring
through Gab ∼ = Z/2Z. There are exactly two of these, the trivial homomorphism and the sign
map sign : S3 → {±1}. These correspond to the trivial F[G]-module F and the F[G]-module F
on which σ ∈ S3 acts by σ · v = sign(σ )v for v ∈ F.
The irreducible two-dimensional representation W of S3 is a subrepresentation of the 3-
dimensional permutation representation ρV : S3 → GL3 (C). That is, consider the standard basis
{e1 , e2 , e3 } of the corresponding F[S3 ]-module V = F3 on which S3 acts by permuting the indices
420 13. REPRESENTATION THEORY

of the basis elements. Then W is spanned by e1 − e2 and e2 − e3 . With respect to this basis, the
corresponding homomorphism ρW : S3 → GL2 (C) satisfies
   
−1 1 0 −1
ρW ((1 2)) = and ρW ((1 2 3)) = .
0 1 1 −1
E XAMPLE 13.3.9. Since all of the C-representations of S3 take values in GLn (Q) for some
n, we have Q[S3 ] ∼
= Q × Q × M2 (Q). In other words, the irreducible C-representations of S3 are
obtained from the irreducible Q-representations of S3 by base change.
13.4. Characters
Recall from Lemma 8.7.22 that the traces of similar matrices are equal.
D EFINITION 13.4.1. Let V be a finite-dimensional vector space over a field F. The trace of
ϕ ∈ AutF (V ) is the trace of the matrix representing ϕ with respect to any choice of ordered basis
of V .
D EFINITION 13.4.2. The character of a representation ρ : G → AutF (V ) of a group G on a
finite-dimensional vector space V over a field F is a map χ : G → F defined by
χ(g) = tr ρ(g).
T ERMINOLOGY 13.4.3. We say that χ is a character of G if it is the character of a represen-
tation of G.
N OTATION 13.4.4. Given an F[G]-module V , we denote the corresponding representation
(i.e., homomorphism) by ρV and and its character by χV .
E XAMPLES 13.4.5.
a. The character χ : G → F of a one-dimensional representation ρ : G → F × satisfies χ(g) =
ρ(g) for all g ∈ G.
b. The character of the permutation representation ρ : Sn → GLn (F) satisfies ρ(σ ) = |Xnσ |
for every σ ∈ Sn , where Xn = {1, 2, . . . , n}.
c. Let W be as in Example 13.3.8. Then the character χW : S3 → C satisfies χW ((1 2)) = 0
and χW ((1 2 3)) = −1.
d. The character χ of the regular representation F[G] satisfies χ(1) = |G| and χ(g) = 0 for
all g ∈ G − {1}.
D EFINITION 13.4.6. The character of the trivial representation is called the trivial character,
or principal character of G.
D EFINITION 13.4.7. A character χV : G → F of an F[G]-module V is irreducible if V is
irreducible.
D EFINITION 13.4.8. The degree of a character χV of an F-representation V of G is dimF (V ).
D EFINITION 13.4.9. A class function of G is a function G → F, for F a field, that is constant
on conjugacy classes in G.
 13.4. CHARACTERS 421

L EMMA 13.4.10. Let G be a group, let F be a field, and let V and W be F-representations of
G. Then
a. χV (e) = dimF V ,
b. χV ⊕W = χV + χW ,
c. χV = χW if V and W are isomorphic representations, and
d. χV is a class function on G.
P ROOF. Since ρV (e) is the identity transformation, we have part a. Part b follows by choos-
ing a basis of V ⊕W that is a union of bases of V and W and noting that the matrix representing
ρV ⊕W (g) for g ∈ G with respect to that basis is block diagonal with blocks ρV (g) and ρW (g). Part
c holds as ρV (g) and ρW (g) are represented by similar matrices if V and W are isomorphic. Part
d also holds as ρV (g) and ρV (g0 ) are represented by similar matrices if g and g0 are conjugate in
G. 
P ROPOSITION 13.4.11. Let G be a finite group, and let F be a field of characteristic zero.
Let V and W be finite-dimensional F-representations of G. Then V and W are isomorphic if and
only if χV = χW .
P ROOF. By Lemma 13.4.10c, we know that V ∼
= W implies χV = χW . Write
r
F[G] = ∏ Mn (Di ).
i=1

For 1 ≤ i ≤ r, let ei denote the identity of Mni (Di ), let Vi be the irreducible F-representation Dni i
of G, and let χi denote its character. Then there exist mi for 1 ≤ i ≤ r such that
r
mi
M
V= Vi .
i=1

Extend χV by F-linearity to a map χV : F[G] → F. Then

r
χV (e j ) = ∑ mi χi (e j ) = mi dimF Vi ,
i=1
so the multiplicities mi of the Vi in V are uniquely determined by χV . That is, χV determines the
isomorphism class of V . 
The following easy lemma, which we will use implicitly, is also quite useful for passing
between groups.
L EMMA 13.4.12. Let V be an F-representation of G.
a. If H is a subgroup of G, then V may be considered as an F-representation of H, and its
character is the restriction χV |H .
b. If N is a normal subgroup of G and N acts trivially on V , then for π : G → G/N the
quotient map, the character of V as an F-representation of G/N is π ◦ ψV .
422 13. REPRESENTATION THEORY

For the remainder of this section, we suppose that G is finite and F is algebraically closed of
characteristic not dividing |G|.
P ROPOSITION 13.4.13. Suppose that G is finite and F is algebraically closed. Let V be a
finite-dimensional F-representation of G. Then ρV (g) is diagonalizable.
P ROOF. By restricting ρV to the cyclic subgroup generated by G, we may suppose that G
is cyclic, say of order n. In this case, F[G] is a direct sum of 1-dimensional representations
Vi on which g acts by multiplication by ζni for ζn a choice of primitive nth root of unity in F.
As V is semisimple, this tells us that V is a direct sum of 1-dimensional representations. The
automorphism ρV (g) is then diagonal with respect to any basis of V consisting of one basis
element of each of these summands. 
Since g has finite order dividing |G|, the following corollary is immediate.
C OROLLARY 13.4.14. Let V be a finite-dimensional F-representation of G. Then the eigen-
values of ρV (g) for g ∈ G are all roots of unity of order dividing |G|.
L EMMA 13.4.15. Let V and W be finite-dimensional F-representations of G. Set χV (g) =
χV (g−1 ) for all g ∈ G. Then we have
a. χV ⊗F W = χV χW and
b. χHomF (V,W ) = χV χW .
P ROOF. By the commutativity of the tensor product and direct sums and the semisimplicity
of F[G], part a reduces to the case that V and W are irreducible. Through a simple application of
Lemma 13.2.23, we may assume that F is algebraically closed. By Proposition 13.4.13, we may
then diagonalize the matrices ρV (g) and ρW (g) for g ∈ G with respect to choices of ordered bases
(v1 , . . . , vn ) of V and (w1 , . . . , wn ) of W . We then have that ρV ⊗F W is diagonal with respect to the
basis of elements vi ⊗ w j with respect to the lexicographical ordering. The diagonal coordinate
corresponding to vi ⊗ w j is the product of the (i, i)-entry of ρV (g) and the ( j, j)-entry of ρW (g).
That is,
tr ρV ⊗W (g) = (tr ρV (g)) · (tr ρW (g)),
as desired.
For part b, we recall the isomorphism
HomF (V,W ) ∼ = V ∗ ⊗F W
of Lemma 13.2.14. We are then reduced by part a to the case that W = F, the trivial F[G]-module.
Again replacing F by its algebraic closure, we may diagonalize ρV (g) for g ∈ G with respect to
a basis B of V . Let B∗ be its dual basis. For φ ∈ B∗ and v ∈ B, we have φ (g−1 v) = αv φ (v), where
g−1 v = αv v. Thus, the trace of ρV ∗ (g) agrees with the trace of ρV (g−1 ), as desired. 
R EMARK 13.4.16. Let G be a finite group and F be a field of characteristic not dividing |G|.
Since χV ⊕W = χV + χW and χV ⊗F W = χV · χW , the set of F-valued characters of G form a ring
with identity the trivial character.
P ROPOSITION 13.4.17. The irreducible F-characters of G form a basis for the F-vector
space of F-valued class functions on G.
 13.4. CHARACTERS 423

P ROOF. Let g1 , . . . , gr be representatives of the r conjugacy classes of G. The space of F-

valued class functions of G has a basis consisting of the maps θi : G → F for 1 ≤ i ≤ r such
that θi (g) = 1 if g ∈ Cgi and θi (g) = 0 otherwise. On the other hand, there are also r irreducible
F-representations Vi for 1 ≤ i ≤ r of G by Proposition 13.3.6, so it suffices to see that their
characters χi = χVi are linearly independent.
Write F[G] ∼ = ∏ri=1 Mni (F) in such a way that Vi is the isomorphic to the simple module F ni
of Mni (F). Let ei denote the idempotent of F[G] corresponding to the identity of Mni (F). We
may extend χi F-linearly to a map χi : F[G] → F. Then χi (x) for x ∈ F[G] is the trace of the
endomorphism of Vi defined by left multiplication by x. Since left multiplication by ei on Vi
(resp., V j for j 6= i) is the identity map (resp., zero map), we have χi (ei ) = ni (resp., χi (e j ) = 0
for j 6= i). Given any linear combination φ = ∑ri=1 ai χi with ai ∈ F, we have φ (e j ) = a j n j , so
φ = 0 if and only if ai = 0 for all i. 
We can identify the idempotents in F[G] that correspond to identity matrices in terms of
characters.
P ROPOSITION 13.4.18. Let χi for 1 ≤ i ≤ r denote the irreducible F-characters of G, and let
ni denote the degree of χi Then the elements
ni
ei = ∑ χi(g−1)g
|G| g∈G

are the primitive, central, orthogonal idempotents of F[G].

P ROOF. Let fi denote the primitive central idempotent in F[G] that acts on the identity on the
irreducible representation Vi with character χi . Write fi = ∑g∈G ag g with ag ∈ F for g ∈ G. For
any g ∈ G, we have
χF[G] ( fi g−1 ) = ∑ χF[G] (ah hg−1 ) = ag |G|.
h∈G
r
On the other hand, we have χF[G] = ∑i=1 ni χi , where ni = dimF Vi , so
r
χF[G] ( fi g−1 ) = ∑ n j χ j ( fig−1).
j=1

If ρi : G → EndF (Vi ) is the F-linear map restricting to ρVi , then

ρ j ( fi g−1 ) = ρ j ( fi )ρ j (g−1 ) = δi j ρ j (g−1 ),
so χ j ( fi g−1 ) = δi j χ j (g−1 ). Thus, we have
ag |G| = ni χi (g−1 ).
It follows that
ni
fi = ∑ χi(g−1)g.
|G| g∈G

424 13. REPRESENTATION THEORY

13.5. Character tables

In this section, we focus on the theory of C-valued characters of a finite group G.
D EFINITION 13.5.1. A character table of a finite group G is a matrix in Mr (C), where r is the
number of conjugacy classes of G with (i, j)-entry χi (g j ), where χi for 1 ≤ i ≤ r are the distinct
characters of the irreducible C-representations of G and gi for 1 ≤ i ≤ r are representatives of the
distinct conjugacy classes in G.
Usually, a character table is written in a table format, as in the following example.
E XAMPLE 13.5.2. Take the group S3 . Let χ1 denote the trivial character, χ2 denote the sign
character, and χ3 the irreducible character of dimension 2. By Example 13.3.8, the character
table of S3 is then as follows:
S3 1 (1 2) (1 2 3)
χ1 1 1 1
χ2 1 −1 1
χ3 2 0 −1
Recall that α denotes the complex conjugate of a complex number α.
L EMMA 13.5.3. Let χ be a C-valued character of degree d of a finite group G of order n.
For g ∈ G, we have χ(g) ∈ Z[ζn ], |χ(g)| ≤ d, and χ(g−1 ) = χ(g).
P ROOF. Let ρ : G → AutC (V ) be the representation corresponding to χ. By Corollary 13.4.14,
the value ρ(g) can be diagonalized to matrix with entries in µn . Since the inverse of a root of
unity is its complex conjugate, ρ(g−1 ) = ρ(g)−1 may be then be represented by the diagonal
matrix A−1 = (ai j ). Then χ(g) is a sum of d roots of unity of order dividing n, which the first
two statements, and we have
χ(g−1 ) = tr(A−1 ) = tr(A) = χ(g).

Let us set ζn = e2πi/n ∈ C for n ≥ 1. The following lemma is useful for producing new
characters out of old.
L EMMA 13.5.4. Let G be a group of order n, and let σ ∈ Gal(Q(ζn )/Q), where ζn = e2πi/n ∈
C. If χ is a character of G, then so is χ σ : G → F defined by χ σ (g) = σ (χ(g)) for all g ∈ G.
Moreover, if a ∈ (Z/nZ)× is such that σ (ζn ) = ζna , then χ σ (g) = χ(ga ).
P ROOF. By Proposition 13.4.13, if G is finite of order n, then every C-representation of G is
the base change of a Q(ζn )-representation. Let V be the Q(µn )-representation of G with character
χ. As a vector space, V ∼ = Q(µn )d for some d ≥ 0, and so σ induces an automorphism σ : V → V
as the direct sum of the automorphisms Q(µn ) → Q(µn ). Then σ ◦ ρV is again a representation,
and its character is χ σ . Note that σ ◦ ρV (g) = ρV (ga ), since in diagonalized form, the entries of
ρV (g) are all elements of µn upon which σ acts by raising to the ath power. 
We pause for a moment to discuss inner products on C-vector spaces.
 13.5. CHARACTER TABLES 425

D EFINITION 13.5.5. An inner product on an C-vector space V is a map h , i : V × V → C

that satisifes
hαv + v0 , wi = αhv, wi + hv0 , wi and hv, β w + w0 i = β̄ hv, wi + hv, w0 i
for all v, v0 , w, w0 ∈ V and α, β ∈ C.
T ERMINOLOGY 13.5.6. That h , i is an inner product on a C-vector space V may be ex-
pressed as saying that it is left C-linear (or just linear) and right conjugate linear.
D EFINITION 13.5.7. An inner product h , i on a C-vector space V is positive definite if
hv, vi ≥ 0 for all v ∈ V , with equality only for v = 0.
D EFINITION 13.5.8. An inner product h , i on a C-vector space V is Hermitian if it is positive
definite and hv, wi = hw, vi for all v, w ∈ V .
D EFINITION 13.5.9. A basis B of a C-vector space V with a Hermitian inner product h , i is
orthonormal if hv, wi = δv,w for all v, w ∈ B.
D EFINITION 13.5.10. A complex inner product space is a pair consisting of a C-vector space
V and a Hermitian inner product on V .
E XAMPLE 13.5.11. The dot product on Cn defined by
n
(a1 , a2 , . . . , an ) · (b1 , b2 , . . . , bn ) = ∑ ai bi
i=1

is a positive definite, Hermitian inner product on Cn . The standard basis of Cn is orthonormal

with respect to the dot product, so Cn is an inner product space with respect to the dot product.
D EFINITION 13.5.12. An inner product h , i on a C-representation V of G is said to be
G-invariant, or an invariant inner product, if hgv, gwi = hv, wi for all v, w ∈ V .
The following provides a useful example.
L EMMA 13.5.13. Let h , i be a Hermitian inner product on a C-representation V of G. Then
the map [ , ] : V ×V → C defined on v, w ∈ V by
1
[v, w] = ∑ hgv, gwi
|G| g∈G

is a G-invariant inner product on V .

P ROOF. As a positive real scalar multiple of a sum of Hermitian inner products on V , the
pairing [ , ] is also Hermitian. The invariance by an element of G follows by reindexing the
sum. 

The next lemma contains the definition of an inner product on the space of C-valued class
functions of G.
426 13. REPRESENTATION THEORY

L EMMA 13.5.14. The function which assigns to a pair (θ , ψ) of C-valued class function of
G the value
1
hθ , ψi = ∑ θ (g)ψ(g)
|G| g∈G
is a positive definite, Hermitian inner product on the space of C-valued class functions of G.
We consider the space of class functions as a Hermitian inner product space with respect to
the Hermitian inner product of Lemma 13.5.14.
R EMARK 13.5.15. If we let h ∈ G act on the space of class functions on G by (h · θ )(g) =
θ (h−1 g), then the resulting inner product is G-invariant.
R EMARK 13.5.16. The inner product of the characters of any two C-representations V and
W is real by Lemma 13.5.3, since

∑ χV (g)χW (g) = ∑ χV (g−1)χW (g−1) = ∑ χV (g)χW (g).

g∈G g∈G g∈G

L EMMA 13.5.17. Let V be a C[G]-module of finite dimension. Then

1
dimC V G = ∑ χV (g).
|G| g∈G
1
P ROOF. Let z = |G| NG ∈ C[G]. Since NG2 = |G|, the element z is an idempotent. The C-linear
endomorphism T of V defined by left multiplication by z therefore has minimal polynomial
dividing x2 − x = x(x − 1). In particular, it is diagonalizable. The trace of T is then the sum of
its nontrivial eigenvalues, which is the dimension of the eigenspace E1 (T ) of 1. It remains then
only to show that E1 (T ) = V G . We check this on v ∈ V : if zv = v, then gv = gzv = zv = v for all
1
g ∈ G, while if gv = v for all g ∈ G, then zv = |G| |G|v = v. 

P ROPOSITION 13.5.18. Let V and W be complex G-representations. Then

hχV , χW i = dimC HomC[G] (V,W ).

P ROOF. Note that HomC[G] (V,W ) = HomC (V,W )G , where g ∈ G acts on φ ∈ HomC (V,W )
by (g · φ )(v) = gφ (g−1 v) for every v ∈ V . Thus,
1 1
dimC HomC[G] (V,W ) = ∑ χHomC (V,W ) (g) = ∑ χV (g)χW (g),
|G| g∈G |G| g∈G

the last step by Lemma 13.4.15b. 

We may now prove the orthogonality of the basis of characters.

T HEOREM 13.5.19 (First orthogonality relation). The set of irreducible complex characters
of a finite group G forms an orthonormal basis of the space of C-valued class functions of G.
 13.5. CHARACTER TABLES 427

P ROOF. Let Vi for 1 ≤ i ≤ r be the distinct irreducible C[G]-modules, and let χi denote the
character of Vi . For any 1 ≤ i, j ≤ r, we have
hχi , χ j i = dimC HomC[G] (Vi ,V j ),
by Proposition 13.5.18. The result then follows by Schur’s lemma. 

This orthogonality also gives us a sort of orthogonality of rows of the character table.
R EMARK 13.5.20. Let {g1 , g2 , . . . , gr } be a set of representatives of the conjugacy classes of
a finite group G. Let χ1 , χ2 , . . . , χr be the complex irreducible character. By Theorem 13.5.19, the
rows ri of the character table with (i, j)-entry χi (g j ) are orthogonal with respect to the weighted
dot product
1 r
ri · ri0 = ∑ c j χi(g j )χi0 (g j ) = hχi, χi0 i,
|G| j=1
where the weight c j is the order of the conjugacy class of g j .
We also have an orthogonality relation for columns.
T HEOREM 13.5.21 (Second orthogonality relation). Let G be a finite group, and let χ1 , χ2 , . . . , χr
be its distinct irreducible, complex characters. For any g, h ∈ G, we have
(
r
|Zg | if g and h are conjugate,
∑ χi(g)χi(h) = 0 otherwise,
i=1

where Zg denotes the centralizer of g in G.

P ROOF. Let g1 , g2 , . . . , gr represent the distinct conjugacy classes in G, and let A ∈ Mr (C) be
the matrix with (i, j)-entry χi (g j ). Let C be the diagonal matrix with (i, i)-entry ci = |Cgi |. Then
r
(ACĀt )i j = ∑ χi(gk )ck χ j (gk ) = δi j |G|,
k=1

the last step by Remark 13.5.20. In particular, ACĀt is a scalar multiple of the identity matrix, so
ACĀt = Āt AC, which tells us that
r
δi j |G| = (Āt AC)i j = ∑ χk (gi)χk (g j )c j .
k=1

Since |Zg j | = |G|c−1

j by the orbit-stabilizer theorem, we are done. 

Let us study character tables in some examples.

E XAMPLE 13.5.22. Let n ≥ 1, and let ζ = e2πi/n ∈ C. Every character of Z/nZ is a power
of the character χ : Z/nZ → C× given by χ(i) = ζ i . Since Z/nZ is abelian, every element of
Z/nZ is the lone element in its conjugacy class. The character table of Z/nZ is as follows:
428 13. REPRESENTATION THEORY

Z/nZ 0 1 2 ··· n−1

1 1 1 1 ··· 1
χ 1 ζ ζ2 · · · ζ n−1
χ2 1 ζ2 ζ4 · · · ζ n−2
.. .. .. .. .. ..
. . . . . .
χ n−1 1 ζ n−1 ζ n−2 ··· ζ.

R EMARK 13.5.23. In general, the number of 1-dimensional complex characters of a finite

group G is |Gab |, since these are exactly the irreducible representations of C[Gab ], which has
C-dimension |Gab |.

E XAMPLE 13.5.24. Let G be any nonabelian group of order 8. By Theorem 7.5.2, there
are two up to isomoprhism, D4 and Q8 . The center Z of G has order 2 (it is nontrivial since
G is a 2-group and if it had order at least 4, it is easy to see that the group would be abelian).
Furthermore, G/Z is abelian since all groups of order 4 are abelian. This also means that G/Z
is the abelianization of G. It is also easy to see that this implies G/Z ∼ = Z/2Z ⊕ Z/2Z. So G
has four characters χ1 , · · · , χ4 of degree 1 and therefore one character χ5 of degree 2 to make
8 = 22 + 1 + 1 + 1 + 1. Pick representatives g and h in G of the two summands Z/2Z. Then g, h,
and gh must be representatives of distinct conjugacy classes, which are then forced to have order
2 since g, h, gh ∈
/ Z. Finally, let z generate the center. The character table is

G 1 z g h gh
χ1 1 1 1 1 1
χ2 1 1 1 −1 −1
χ3 1 1 −1 1 −1
χ4 1 1 −1 −1 1
χ5 2 −2 0 0 0

The last row is determined by orthogonality of the columns, since its first entry must be 2. Note
that this implies that the isomorphism type of a group is not determined by its character.

E XAMPLE 13.5.25. Note that S4 has 5 conjugacy classes, and the sums of the squares of the
degrees ni of the 5 irreducible characters χi equals |24|. Also S4 → Z/2Z via the sign map, so
there are (at least) two 1-dimensional characters: the trivial character χ1 and the sign character
χ2 . Since n23 + n24 + n25 = 22, we have n3 = 2 and n4 = n5 = 3 (if put in increasing order). The
quotient of S4 by the normal subgroup h(1 2)(3 4), (1 3)(2 4)i is isomorphic to S3 , so we obtain
by composition with the irreducible two-dimensional representation ρ : S3 → GL2 (C) a two-
dimensional representation χ3 : S4 → GL2 (C), which is nonabelian and hence irreducible, being
semisimple. Whatever χ4 is, note that if we tensor its representation V4 with the representation
V2 of the sign character χ2 , we obtain another irreducible character χV4 ⊗CV2 = χ4 χ2 of dimension
3, which we call χ5 . (We will see that it is actually different from χ4 .) The character table is
 13.6. INDUCED REPRESENTATIONS 429

S4 e (1 2) (1 2 3) (1 2 3 4) (1 2)(3 4)
χ1 1 1 1 1 1
χ2 1 −1 1 −1 1
χ3 2 0 -1 0 2
χ4 3 1 0 −1 -1
χ5 3 −1 0 1 -1.
The entries in blue are determined from the character table for S3 . The entries in red are obtained
by noting that χ5 = χ4 χ2 and using orthogonality of columns. (That is, the third and fifth columns
are determined using orthogonality with the first and the second and fourth, up to sign, using
orthogonality with the first and each other.)
Note that we can restrict representations and characters to subgroups: for H 6 G and a G-
representation V , this amounts to considering V as a module over the group ring of H and re-
stricting the function χV to H.
E XAMPLE 13.5.26. The group A4 has 4 conjugacy classes with representatives e, (1 2 3),
(1 3 2), and (1 2 3 4). We have Aab ∼
4 = Z/3Z, generated by the image of (1 2 3), so there are
three abelian characters which are the powers of the character χ such that χ((1 2 3)) = ω, where
ω = e2πi/3 . Since |A4 | = 12, there is one more character ψ, which has degree 3. Its values can
be calculated by orthogonality of columns, yielding the character table
A4 e (1 2 3) (1 3 2) (1 2)(3 4)
1 1 1 1 1
χ 1 ω ω 2 1
2
χ 1 ω 2 ω 1
ψ 3 0 0 -1.
13.6. Induced representations
Let G be a finite group and H a subgroup. For a commutative ring R we can view a R[G]-
module A as an R[H]-module in the obvious fashion. When thinking of A as an R[H]-module, it
is often helpful to give it a new name and symbol
D EFINITION 13.6.1. An R[G]-module A viewed as an R[H]-module is called the restriction
of A from G to H and is denoted by ResG
H (A).
Together with the obvious definition on morphisms, restriction defines an exact functor
ResG
H : R[G]-mod → R[H]-mod.
The natural question arises as to whether or not ResG
H has an adjoint, and indeed, it has a left
adjoint. We first give the construction.
D EFINITION 13.6.2. Let H be a subgroup of a group G, and let R be a commutative ring.
The induced module from H to G of an R[H]-module B is the R[G]-module
IndG
H (B) = HomR[H] (R[G], B)

where for ϕ ∈ IndG

H (B), we let g ∈ G act by (g · ϕ)(x) = ϕ(xg) for all x ∈ F[G].
430 13. REPRESENTATION THEORY

R EMARK 13.6.3. Since R[G] is R[H]-free, IndG H provides an exact functor from R[H]-mod to
R[G]-mod, since is the functor hR[G] in our earlier notation.
If H is of finite index in G, we have an alternate description of the induced module. That is,
there is another way in which to produce an R[G]-module from an R[H]-module B using tensor
products. That is, we can take the R[G]-module R[G] ⊗R[H] B, where g ∈ G acts on x ⊗ b in the
tensor product by g(x ⊗ b) = gx ⊗ b.
P ROPOSITION 13.6.4. Let H be a finite index subgroup of G, and let R be a commutative
ring. Given an R[H]-module B, there is natural isomorphism
∼
κ : IndG
H (B) −
→ R[G] ⊗R[H] B
given on ϕ ∈ IndG
H (B) by
κ(ϕ) = ∑ g−1 ⊗ ϕ(g),
ḡ∈H\G
where for each ḡ ∈ H\G, the element g ∈ G is a choice of representative of ḡ.
P ROOF. First, we note that χ is a well-defined map, as
(hg)−1 ⊗ ϕ(hg) = g−1 h−1 ⊗ hϕ(g) = g ⊗ ϕ(g)
for ϕ ∈ R[G] ⊗R[H] B, h ∈ H, and g ∈ G. Next, we see that χ is an R[G]-module homomorphism,
as
χ(g0 ϕ) = ∑ g−1 ⊗ ϕ(gg0 ) = g0 ∑ (gg0 )−1 ⊗ ϕ(gg0 ) = g0 χ(ϕ)
ḡ∈H\G ḡ∈H\G
g0
for ∈ G. As the coset representatives form a basis of R[G] as a free R[H]-module, we may
define an inverse to χ that maps
∑ g−1 ⊗ bg ∈ IndG
H (B)
ḡ∈H\G

to the unique R[H]-linear map ϕ that takes the value bg on g for the chosen representative of
ḡ ∈ H\G. 
P ROPOSITION 13.6.5. Let H be a finite index subgroup of G. Then IndG
H is left adjoint to
ResG
H .
P ROOF. Using the alternate characterization of IndG
H U of Proposition 13.6.4 and the adjoint-
ness of Hom and ⊗, we have
Hom (R[G] ⊗
R[G] U,V ) ∼
R[H] = Hom (U, Hom (R[G],V )) ∼
R[H] = Hom
R[G] (U,V ),
R[H]
the latter isomorphism being induced by evaluation at 1 in the second variable. 
D EFINITION 13.6.6. Let H be a subgroup of a group G, let W be an F-representation of
H.
a. The induced representation from H to G of W is IndG
H (W ).
b. If H has finite index in G and W is finite-dimensional with character ψ, then the induced
character IndG G
H (ψ) of ψ is the character of IndH (W ).
 13.6. INDUCED REPRESENTATIONS 431

E XAMPLE 13.6.7. Let H be a finite index subgroup of G. For the trivial representation F of
H, we have
IndG ∼ ∼
H (F) = F[G] ⊗F[H] F = F[G/H],
where the latter module is the F[G]-module with F-basis the left G-set G/H. That is, IndG H (F)
is the permutation representation for the left action of G on G/H. Thus, the induced character
χ = IndG H (ψ1 ) of the trivial character ψ1 on H has the property that χ(g) is the number of left
H-cosets fixed by left multiplication by g ∈ G.
R EMARK 13.6.8. Let H be a finite index subgroup of G. The induced representation of
the regular representation F[H] of H is the regular representation F[G] of G. In particular, all
irreducible F-representations of G are summands of induced representations of the irreducible
F-representations of H.
R EMARK 13.6.9. In finite group theory, one often uses the alternate tensor product charac-
terization provided by Proposition 13.6.4 as the definition of the induced representation.
N OTATION 13.6.10. For a subgroup H of a group G, we denote the restriction of a character
χ of G to H by ResG
H χ, or more simply χ|H .
N OTATION 13.6.11. For the inner product of Lemma 13.5.14, we use h , iG to indicate its
dependence on the group G.
D EFINITION 13.6.12. For a character χ of a finite-dimensional C-representation of a group,
the multiplicity of an irreducible character ψ in it is the multiplicity of the irreducible represen-
tation with character ψ in the representation with character χ.
We have the following corollary of Proposition 13.6.4.
C OROLLARY 13.6.13 (Frobenius reciprocity). Let G be a finite group and H a subgroup. Let
ψ be a C-valued character of H and χ be a C-valued character of G. Then
hIndG G
H ψ, χiG = hψ, ResH χiH .
P ROOF. Let ψ = χU and χ = χV for representations U and V of H and G, respectively. By
Propositions 13.5.18 and 13.6.5, we have
hIndG G G
H ψ, χiG = dimC HomC[G] (IndH U,V ) = dimC HomC[H] (U,V ) = hψ, ResH χiH .

We can construct tables that contain these values of the pairings in Corollary 13.6.13.
D EFINITION 13.6.14. The induction-restriction table of a subgroup H of a finite group G is
the matrix with rows indexed by the complex irreducible characters ψi of H and columns by the
complex irreducible characters χ j of G with (i, j)-entry hψi , ResG
H χ j iH .

E XAMPLE 13.6.15. Again let χi for 1 ≤ i ≤ 5 and 1, χ, χ 2 , and ψ be the characters of S4

and A4 , respectively of Examples 13.5.25 and 13.5.26. From the character tables of G = S4 and
H = A4 , we see that χ1 |H = χ2 |H = 1, χ3 |H = χ + χ 2 , and χ4 |H = χ5 |H = ψ. The induction-
restriction table is
432 13. REPRESENTATION THEORY

χ1 χ2 χ3 χ4 χ5
1 1 1 0 0 0
χ 0 0 1 0 0
χ2 0 0 0 1 0
ψ 0 0 0 1 1.
Frobenius reciprocity tells us that IndG G G 2 G
H 1 = χ1 + χ2 , IndH χ = IndH χ = χ3 , and IndH ψ = χ4 +
χ5 .
P ROPOSITION 13.6.16. Let H be a finite index subfroup of a group G, and let g1 , . . . , gk be
a system of left coset representatives of H in G. For a character ψ of H, extend ψ to a function
ψ̃ : G → C by setting ψ̃(g) = 0 if g ∈
/ H. For g ∈ G, we then have
k
−1
IndG
H (ψ)(g) = ∑ ψ̃(gi ggi ).
i=1
P ROOF. Let W be an m-dimensional representation of H with character ψ, and let B =
(w1 , . . . , wm ) be an ordered F-basis of W . Recall that
IndG ∼
H (W ) = F[G] ⊗F[H] W
We have a basis gi ⊗ w j of F[G] ⊗F[H] W for 1 ≤ i ≤ k and 1 ≤ j ≤ m with the lexicographical
ordering.
For a given 1 ≤ i ≤ k, any g ∈ G satisfies
ggi = gσ (i) hi
for some σ ∈ Sk and hi ∈ H. Then
g(gi ⊗ w j ) = gσ (i) ⊗ hi w j .
With respect to the given basis, the matrix of ρW (g) is a k-by-k matrix of blocks in Mm (F) with
one nonzero block in each row and each column, i.e., the blocks with coordinates (i, σ (i)), which
are those representing ρW (hi ) with respect to the basis B.
Adding up the diagonal entries in the (i, i)-block, we get 0 if i0 6= i and ψ(h) = ψ(g−1
i ggi )
0 −1
if i = i. By definition of ψ̃, this equals ψ̃(gi ggi ) in all cases. Summing over i, we obtain the
result. 
C OROLLARY 13.6.17. Let G be a finite group and H be a subgroup of G. Let ψ be a character
of H. For g ∈ G, we then have
1
IndG
H (ψ)(g) = ∑ ψ(k−1gk).
|H| k∈G
kgk−1 ∈H

P ROOF. This follows from the formula of Proposition 13.6.16. To see that, take ψ̃ as in its
statement, and note that for any h ∈ H, we have
ψ̃((gi h)−1 g(gi h)) = ψ̃(h−1 (g−1
i ggi )h),
since conjugation by h−1 preserves H and G − H, and ψ is a class function on H. 
 13.6. INDUCED REPRESENTATIONS 433

The following corollary is immediate.

C OROLLARY 13.6.18. Let H be a finite index normal subgroup of G, and let χ be a character
of H. Then IndG
H (ψ)(g) = 0 for all g ∈
/ H.
E XAMPLE 13.6.19. Consider the dihedral group G = D2n of order 4n with n ≥ 3. It has
abelianization the Klein 4-group generated by the images of r and s, so G has four degree 1
characters χ1 , χ2 , χ3 , χ4 with χ1 trivial, χ2 (r) = χ3 (s) = −1 and χ2 (s) = χ3 (r) = 1, and χ4 =
χ2 χ3 . Now, consider the cyclic subgroup H = hri, which has characters ψ i for 0 ≤ i ≤ 2n − 1
with ψ(r) = ζ2n . The induced character θi of ψ i is trivial on all reflections and satisfies
ij −i j
θi (r j ) = ζ2n + ζ2n .
The characters θi = IndG i
H ψ with 1 ≤ i ≤ n − 1 are all distinct degree 2 characters which are
clearly not sums of the χi , so they are irreducible characters of G. The sum of the squares of the
dimensions of these characters is 4 · 12 + (n − 1) · 22 = 4n = |G|, so these are all of the irreducible
characters on G. Setting
k −k
ξk = ζ2n + ζ2n = 2 cos(kπ/n) ∈ R,
the character table is then as follows.
D2n e s rs r r2 ··· rn−1 rn
χ1 1 1 1 1 1 ··· 1 1
χ2 1 1 −1 −1 1 · · · (−1) n−1 (−1)n
χ3 1 −1 −1 1 1 ··· 1 1
χ4 1 −1 1 −1 1 · · · (−1) n−1 (−1)n
θ1 2 0 0 ξ1 ξ2 ··· ξn−1 −2
θ2 2 0 0 ξ2 ξ4 · · · ξ2(n−1) 2
θ3 2 0 0 ξ3 ξ6 · · · ξ3(n−1) −2
.. .. .. .. .. .. .. .. ..
. . . . . . . . .
θn−1 2 0 0 ξn−1 ξ2(n−1) · · · ξ(n−1)2 (−1)n−1 2.
Here, one might note that ξ0 = 2, and ξk = ξk+2n = −ξk+n = −ξn−k for all k. We remark that
θi |H = ψ i + ψ −i for all 1 ≤ i ≤ n − 1, while θ0 = χ1 + χ3 and θn = χ2 + χ4 , consistent with
Frobenius reciprocity.
We also give a formula which tells us explicitly how to determine the induced character to G
of an H-character from the character table for H and knowledge of conjugacy classes.
P ROPOSITION 13.6.20. Let G be a finite group and H be a subgroup of G. Let ψ be a
character of G, let g ∈ G, and let Cg be the conjugacy class of g in G. Write H ∩Cg as a possibly
empty disjoint union of conjugacy classes T1 , . . . , Tl of H. For 1 ≤ i ≤ l, let hi be a representative
of Ti . Then
l
|Ti |
IndGH ψ(g) = [G : H] ∑ ψ(hi ).
i=1 |Cg |
P ROOF. This is a matter of counting. That is, by Corollary 13.6.17, we must show that
the number of k ∈ G such that k−1 gk is conjugate to hi in H is |Zg ||Ti |, where Zg denotes the
434 13. REPRESENTATION THEORY

centralizer of g in G. We know that there are |Zg | elements of G that conjugate g to any particular
element of Cg . Thus, there are |Zg ||Ti | elements in G that conjugate g to one of the elements in
Ti , as desired. 

We may use Proposition 13.6.20 to determine the induced characters on a group from the
characters on its subgroup.
E XAMPLE 13.6.21. Take H = S3 , which we view as a subgroup of G = S4 . Recall that
the conjugacy classes of S4 are determined by cycle type, with conjugacy classes C1 , · · · ,C5
corresponding to cycle types e, (1 2), (1 2 3), (1 2 3 4), and (1 2)(3 4) having orders 1, 6, 8,
6, 3, respectively. Now, C4 and C5 contain no elements of S3 , while C1 , C2 , and C3 contain the
conjugacy classes T1 , T2 , and T3 in S3 of 1, (1 2), and (1 2 3). Note that |Ci | = |Ti |, |C2 | =
2|T2 |, |C3 | = 4|T3 |, and [G : H] = 4. Let ψ1 , ψ2 , and ψ3 be the trivial, sign, and irreducible 2-
dimensional characters of S3 , respectively. By Proposition 13.6.20, we obtain the following table
from the character table of S3 :
S4 e (1 2) (1 2 3) (1 2 3 4) (1 2)(3 4)
φ1 4 2 1 0 0
φ2 4 -2 1 0 0
φ3 8 0 -1 0 0.
We will use this to determine the characters of S4 once again. Assume we have already found
its abelian characters, the trivial character χ1 and the alternating character χ2 . Since
hIndG G
H φi , IndH φi i = dim HomC[G] (Vi ,Vi ),

where Vi is the G-representation induced by φi , and these values are 2, 2, 3, respectively, we have
that the Vi break up into these respective numbers of irreducible representations. But note that
hφ1 , χ1 i = 1 and hφ1 , χ2 i = 0, so φ1 − χ1 so is an irreducible degree 3 character of G, which we
previously called χ4 . Similarly, φ2 − χ2 is an irreducible degree 3 character, which we called χ5 .
We compute that hφ3 , χ4 i = hφ3 , χ5 i = 1, and φ3 − χ4 − χ5 is an irreducible character of degree
2, which we called χ3 .

13.7. Applications to group theory

Let G be a group of order n, and let C1 , . . . ,Cr be the conjugacy classes in G, choose g j ∈ C j
and set c j = |C j | for each 1 ≤ j ≤ r. Let χ1 , . . . , χr be the irreducible complex characters of G,
and set ni = deg χi for 1 ≤ i ≤ r. Let Vi denote the irreducible representation with character χi ,
let ρi : C[G] → EndC (Vi ) be the C-algebra homomorphism restricting to the representation ρVi .
We also use χi to denote its C-linear extension to map χi : C[G] → C.
P ROPOSITION 13.7.1. Set
Ni = {g ∈ G | χi (g) = χi (1)}
for 1 ≤ i ≤ r. The normal subgroups of G are exactly the intersections
T
j∈J N j , where J is a
subset of {1, . . . , r}.
 13.7. APPLICATIONS TO GROUP THEORY 435

P ROOF. First, Ni is a normal subgroup, since χi (g) = χi (1) if and only if g acts as the identity,
recalling that the eigenvalues of g are all roots of unity, so Ni = ker ρVi . It follows that every
intersection of the Ni ’s is normal.
Now suppose N is normal in G. Let V = C[G/N]. Let χV be the character of V as a C[G]-
module. Since ker ρV = N, we have χV (g) = χV (1) if and only if g ∈ N. If g ∈ / N, then ghN 6= hN
for any h ∈ G, so χV (g) = 0.
Now χV is a sum of irreducible characters with nonnegative integer coefficients, say χV =
r
∑i=1 ai χi . We claim that N = i∈J Ni , where J is the set of i with ai ≥ 1. Note that for any
T

character ψ and g ∈ G, we have |ψ(g)| ≤ ψ(1) since ψ(g) is a sum of ψ(1) roots of unity. For
g ∈ G, we have
r r
χV (g) = |χV (g)| = ∑ ai χi(g) ≤ ∑ ai χi(1) = χV (1),
i=1 i=1
with equality of the first and last term holding if and only if g ∈ N. However, the middle in-
equality is an equality if and only if all χi (g) for i ∈ J are equal and have absolute value χi (1).
This condition holds if and only if all χi (g) = χi (1), since one of these characters is the trivial
character. 
Next, we will show how to find the center of G.
P ROPOSITION 13.7.2. Set
Zi = {g ∈ G | |χi (g)| = χi (1)}
Tr
for 1 ≤ i ≤ r. Then Zi is a normal subgroup of G, and the center of G is equal to i=1 Zi .

P ROOF. We have that Zi is a normal subgroup since the condition that g ∈ Zi is exactly that g
acts as a scalar multiple of the identity, in other words that g is in the inverse image of the center
of ρVi (G).
We claim that Zi /Ni is the center of G/Ni . Note that ρVi has kernel Ni defined as in Proposi-
tion 13.7.1, and the elements of Zi are mapped to scalar matrices in the center of ρVi (G) ∼ = G/Ni .
So Zi /Ni is contained in the center of G/Ni . Now suppose that gNi is in the center of G/Ni .
Then ρVi (g) commutes with all ρVi (h) for h ∈ G, which implies that left multiplication by g is a
C[G]-module isomorphism of Vi . But Vi is simple, so HomC[G] (Vi ,Vi ) ∼ = C. In other words, g acts
as scalar multiplication by some element, hence is contained in Zi .
Given the claim, we have that Z(G)Ni /Ni ≤ Zi /Ni , and so Z(G) ⊆ Zi for T all i. Now suppose
−1 −1
that z ∈ Zi for all i. Let g ∈ G. Then gzg z ∈ Ni by our earlier claim. But ri=1 Ni is trivial by
Proposition 13.7.1. So, gzg−1 z−1 = 1, for all g ∈ G, so z ∈ Z(G), as desired. 
P ROPOSITION 13.7.3. For each pair (i, j) of integers with 1 ≤ i, j ≤ r and each g ∈ G, we
have
cj
χi (g j ) ∈ Z[µn ].
ni
P ROOF. Set N j = ∑g∈C j g ∈ Z(C[G]). Multiplication by N j defines a C[G]-linear map N j : Vi →
Vi which by Schur’s lemma is a scalar multiple of the identity, say N j = α j idVi for α j ∈ C, which
436 13. REPRESENTATION THEORY

tells us that χi (N j ) = α j χi (1) = α j ni . But we also have

χi (N j ) = ∑ χi (g) = c j χi (g j ),
g∈C j
c
so α j = nij χi (g j ). We show that eac
Next, set
a jkl = |{(g, h) | g ∈ C j , h ∈ Ck , gh = gl }| ∈ Z≥0 ,
and note that this number is independent of the choice of gl , since sgs−1 · shs−1 = sghs−1 for any
s ∈ G. Then
r r
ρi (N j )ρi (Nk ) = ∑ ∑ ρi(gh) = ∑ a jkl ∑ ρi(q) = ∑ a jkl ρi(Nl ).
g∈C j h∈Ck l=1 q∈Cl l=1

Since N j acts on Vi by multiplication by the scalar α j , this implies

r
α j αk = ∑ a jkl αl .
l=1
In particular the subring Z[{α j | 1 ≤ j ≤ k}] of C has finite Z-rank, so it is integral over Z. In
particular, each α j is integral over Z. The result now follows as α j ∈ Q(µn ) for each j, and Z[µn ]
is the integer ring of Q(µn ). 
C OROLLARY 13.7.4. The dimension of an irreducible complex representation of a finite
group G divides |G|.
P ROOF. Let 1 ≤ i ≤ r, and consider the quotient of interest
r cj
n n
= hχi , χi i = ∑ ni χi(g j )χi(g j ),
ni ni j=1
cj
which is a Z[µn ]-linear combination of the algebraic integers ni χi (g j ), hence an algebraic integer.
Since the fraction also lies in Q, we have that ni divides n.