Math 250A Lecture Notes

Groups, Rings, and Fields

Professor: Richard Borcherds
Scribe: Daniel Raban

Contents
1 Groups 7
1.1 Groups and Actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.1.1 The 8 actions of a group on itself . . . . . . . . . . . . . . . . . . . . 8
1.2 Lagrange’s theorem and consequences . . . . . . . . . . . . . . . . . . . . . 8
1.2.1 Applications of Lagrange’s theorem . . . . . . . . . . . . . . . . . . . 9
1.2.2 Geometric meaning of Lagrange’s theorem . . . . . . . . . . . . . . . 9
1.3 Groups of order 4 and product groups . . . . . . . . . . . . . . . . . . . . . 9

2 Groups of orders 6 and 8 11

2.1 Two groups of order 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.2 Quotient groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Other groups of order 6 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.4 Groups of order 8 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

3 Non abelian groups of order 8 15

3.1 The dihedral group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.2 Quaternions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

4 Groups of order 9, 10, and 12 17

4.1 Groups of order 9 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
4.2 Nilpotent groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.3 Groups of order 2p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
4.4 Groups of order 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

5 Groups of Order 12,. . . ,24 19

5.1 Groups of order 12 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.1.1 Sylow theorems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
5.2 Solvability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20

1
 5.3 Groups of order 13, 14, and 15 . . . . . . . . . . . . . . . . . . . . . . . . . 21
5.4 Groups of order 16 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
5.5 Finitely generated abelian groups . . . . . . . . . . . . . . . . . . . . . . . . 23
5.6 Groups of order 17,. . . ,24 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

6 Symmetric Groups 25
6.1 Basic definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6.2 The alternating group . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6.3 Sn , An , and platonic solids . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
6.4 Conjugacy classes of Sn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
6.5 Normal subgroups of Sn . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
6.6 Outer automorphisms of Sn . . . . . . . . . . . . . . . . . . . . . . . . . . . 27

7 Category Theory 29
7.1 Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
7.2 Functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
7.3 Natural transformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
7.4 Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
7.5 Equalizers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
7.6 Initial and final objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
7.7 Limits and pull-backs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
7.8 Coproducts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

8 Free Groups 35
8.1 Free abelian groups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
8.2 The free group on g1 , . . . , gn . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
8.2.1 Construction of the free group . . . . . . . . . . . . . . . . . . . . . 35
8.2.2 Subgroups of free groups . . . . . . . . . . . . . . . . . . . . . . . . . 36

9 Rings 39
9.1 Definition and examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
9.2 Analogies between groups and rings . . . . . . . . . . . . . . . . . . . . . . 40
9.3 Group rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
9.3.1 An alternative description of R[G] . . . . . . . . . . . . . . . . . . . 41
9.4 Ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
9.5 Generators and relations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

10 Euclidean Domains, Principal Ideal Domains, and Unique Factorization

Domains 43
10.1 Euclidean Domains and Principal Ideal Domains . . . . . . . . . . . . . . . 43
10.1.1 Euclidean Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
10.1.2 Principal Ideal Domains . . . . . . . . . . . . . . . . . . . . . . . . . 43

2
 10.2 Unique factorization domains . . . . . . . . . . . . . . . . . . . . . . . . . . 44
10.2.1 Definitions and relationship to principal ideal domains . . . . . . . . 44
10.2.2 Examples and Applications . . . . . . . . . . . . . . . . . . . . . . . 45

11 Prime Ideals and Maximal Ideals 47

11.1 Fields and integral domains . . . . . . . . . . . . . . . . . . . . . . . . . . . 47
11.2 Maximal ideals and Zorn’s lemma . . . . . . . . . . . . . . . . . . . . . . . . 48

12 Localization 49
12.1 What is localization? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
12.2 Construction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
12.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50

13 Modules 52
13.1 Basic notions and examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
13.1.1 Modules and homomorphisms . . . . . . . . . . . . . . . . . . . . . . 52
13.1.2 Exact sequences of modules . . . . . . . . . . . . . . . . . . . . . . . 53
13.1.3 Examples of modules . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
13.2 Free modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
13.3 Projective modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

14 More on Projective Modules 57

14.1 Projective modules as direct sums . . . . . . . . . . . . . . . . . . . . . . . 57
14.2 Eilenberg-Mazur swindle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

15 Tensor Products 58
15.1 Construction and universal property . . . . . . . . . . . . . . . . . . . . . . 58
15.2 Exact sequences and the tensor product . . . . . . . . . . . . . . . . . . . . 59
15.3 More examples and properties . . . . . . . . . . . . . . . . . . . . . . . . . . 61
15.4 Tensor products of noncommutative rings . . . . . . . . . . . . . . . . . . . 62

16 Duality 63
16.1 Notions of duality for algebraic objects . . . . . . . . . . . . . . . . . . . . . 63
16.1.1 Duality of vector spaces . . . . . . . . . . . . . . . . . . . . . . . . . 63
16.1.2 Duality of free modules . . . . . . . . . . . . . . . . . . . . . . . . . 63
16.1.3 Duality for finite abelian groups . . . . . . . . . . . . . . . . . . . . 63
16.2 Applications of duality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
16.2.1 Dirichlet characters . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
16.2.2 The Fourier transform . . . . . . . . . . . . . . . . . . . . . . . . . . 64
16.2.3 Existence of “enough” injective modules . . . . . . . . . . . . . . . . 65

3
17 Limits and Colimits 66
17.1 Colimits . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
17.1.1 Examples of colimits . . . . . . . . . . . . . . . . . . . . . . . . . . . 67
17.2 Exact sequences of colimits . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
17.3 Inverse limits and the p-adic integers . . . . . . . . . . . . . . . . . . . . . . 69

18 The Snake Lemma 71

18.1 Statement and proof of the snake lemma . . . . . . . . . . . . . . . . . . . . 71
18.2 Applications of the snake lemma . . . . . . . . . . . . . . . . . . . . . . . . 73
18.2.1 Exact sequences of tensor products of modules . . . . . . . . . . . . 73
18.2.2 The Mitag-Leffler condition . . . . . . . . . . . . . . . . . . . . . . . 75
18.3 Unrelated: Finitely generated modules over a PID . . . . . . . . . . . . . . 76

19 Polynomials and Divisibility 78

19.1 Polynomial division with remainder . . . . . . . . . . . . . . . . . . . . . . . 78
19.2 An application to field theory . . . . . . . . . . . . . . . . . . . . . . . . . . 79
19.3 Unique factorization in polynomial rings . . . . . . . . . . . . . . . . . . . . 80
19.4 Irreducibility tests in Z[x] (or Q[x]) . . . . . . . . . . . . . . . . . . . . . . . 81

20 More on Irreducibility Tests 84

20.1 Eisenstein’s criterion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
20.2 Rational roots . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84

21 Noetherian Rings and Hilbert’s Theorem 85

21.1 Noetherian rings and Noether’s theorem . . . . . . . . . . . . . . . . . . . . 85
21.2 Hilbert’s theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
21.3 Rings of invariants and symmetric functions . . . . . . . . . . . . . . . . . . 86

22 Symmetric Functions and Polynomial Invariants 88

22.1 Symmetric functions and Newton’s identities . . . . . . . . . . . . . . . . . 88
22.1.1 Newton’s identities . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
22.2 The discriminant . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
22.3 The ring of invariants, revisited . . . . . . . . . . . . . . . . . . . . . . . . . 90

23 Formal power series 93

23.1 Definition, inverse limit, and multiplicative inverses . . . . . . . . . . . . . . 93
23.2 Ideals of R JxK . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
23.3 Unique factorization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
23.4 Hensel’s lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

4
24 Field Extensions 98
24.1 Field extensions and algebraic elements . . . . . . . . . . . . . . . . . . . . 98
24.2 Splitting fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
24.3 Application to finite fields . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
24.4 Algebraic closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 102

25 Normal, Separable and Galois Extensions 104

25.1 Normal extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
25.2 Separable extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
25.3 Galois extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
25.3.1 Galois extensions and Galois groups . . . . . . . . . . . . . . . . . . 105
25.3.2 Galois groups and subextensions . . . . . . . . . . . . . . . . . . . . 107

26 The Fundamental Theorem of Galois Theory 109

26.1 Proof and an example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
26.2 Applications of the fundamental theorem . . . . . . . . . . . . . . . . . . . 110
26.2.1 Construction of a 17-sided regular polygon . . . . . . . . . . . . . . 110
26.2.2 Subextensions of a splitting field . . . . . . . . . . . . . . . . . . . . 110
26.3 Extensions corresponding to normal subgroups and factor groups . . . . . . 111
26.4 Finding extensions corresponding to a given group . . . . . . . . . . . . . . 112

27 Examples in Galois Theory and Primitive Elements 113

27.1 Galois group of an irreducible degree 3 polynomial . . . . . . . . . . . . . . 113
27.2 Algebraic closure of C . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
27.3 Primitive elements of separable extensions . . . . . . . . . . . . . . . . . . . 114
27.4 Primitive elements of extensions with Galois group Z/pZ . . . . . . . . . . . 115

28 Cyclic Extensions and Cyclotomic Polynomials 117

28.1 Cyclic extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 117
28.2 Cyclotomic polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
28.3 Applications of cyclotomic polynomials . . . . . . . . . . . . . . . . . . . . . 119
28.3.1 Primes modulo n . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
28.3.2 Galois extensions over Q . . . . . . . . . . . . . . . . . . . . . . . . . 120
28.3.3 Finite division algebras . . . . . . . . . . . . . . . . . . . . . . . . . 120
28.4 Norm and trace in finite extensions . . . . . . . . . . . . . . . . . . . . . . . 121

29 Norm and Trace 122

29.1 Norm and trace of finitely generated extensions . . . . . . . . . . . . . . . . 122
29.2 The integers of a quadratic field . . . . . . . . . . . . . . . . . . . . . . . . . 123
29.3 Discriminant of a field extension L/K . . . . . . . . . . . . . . . . . . . . . 125
29.4 Applications of the discriminant of a field extension . . . . . . . . . . . . . 126

5
30 Hilbert’s Theorem 90 and Galois Cohomology 128
30.1 Hilbert’s theorem 90 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128
30.2 Applications of Hilbert’s theorem 90 . . . . . . . . . . . . . . . . . . . . . . 128
30.3 Galois cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
30.3.1 Exact sequences . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
30.3.2 Lang’s definition of cohomology . . . . . . . . . . . . . . . . . . . . . 131
30.4 Hilbert’s theorem 90 for all Galois extensions . . . . . . . . . . . . . . . . . 132

31 Infinite Extensions and Galois Cohomology 133

31.1 Hilbert’s Theorem 90 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
31.2 Infinite Galois extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134
31.3 Abelian Kummer theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
31.4 Artin-Schrier extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137

6
1 Groups
1.1 Groups and Actions
Definition 1.1. A group is the set of symmetries of something.
Example 1.1. Consider the symmetries of a rectangle. You can:
1: Do nothing
a: Reflect horizontally
b: reflect vertically
c: rotate by π.
Definition 1.2. A group is a set G with a binary operation G × G → G (usually written
(a, b) → a + b, a × b, ab, or a · b) such that
(1) There is an identity (called e, 1, or 0) such that ea = a = ae
(2) Each element has inverse a−1 such that a−1 a = aa−1 = e
(3) The operation is associative (ab)c = a(bc).
We must reconcile the two definitions. The first, “concrete” definition can be thought
of in the second, “abstract” way by making the operation composition of symmetries. Is
the reverse true? Cayley proved that the answer is yes. To show this, we must construct
some S with G as the set of its symmetries. G has to act on S. This means we have an
action of G on S.
Definition 1.3. An action of G on S is a map G × S → S such that
(1) g(h · s) = (gh) · s for s ∈ S and g, h ∈ G
(2) e · s = s for s ∈ S.
Then what is S that satisfies this properties? Well, we can set S = G and make the
action on S multiplication by G. This represents G as a subgroup of the symmetries of
S; this is not necessarily the group of symmetries of S. But we have shown that G is
isomorphic to the set of permutations of a set. Recall:
Definition 1.4. A subgroup H of G is a subset of G closed under × and inverses, containing
the identity, e.
Definition 1.5. A homomorphism from G to H is a function from G to H preserving ×,
inverses, and the identity; i.e. f (ab) = f (a)f (b).
Definition 1.6. An isomorphism is a homomorphism that is a bijection.
If there is an isomorphism from G to H, then they are essentially the same, up to a
relabeling of the elements.
Example 1.2. Let G = (R, +), and let H = (R∗ , ×) (the nonzero reals). The isomorphism
G → H is the exponential map.

7
 So our new problem is to put some “structure” on S(= G) so that G is the set of all
symmetries preserving the structure. The structure is a right action of G on S.

Definition 1.7. A left action G × S → S is given by g · s. A right action S × G → S is

given by s · g.

Remark 1.1. The left and right actions of G on S are different, unless the operation is
commutative.

A symmetry f : S → S preserves a right action of G if f (s · g) = f (s) · g. Elements of G

acting on left preserve right actions as (g·s)·h = g·(s·h) (which follows from the associative
law). The right action commutes with the left action. Moreover, anything commuting with
the right action is of the form s 7→ g · s from some g ∈ G. Suppose f : S → S commutes
with a right action. Then f (e) = g for some g. Then f (s) = f (es) = f (e) · s = g · s. So f
is “the same” as g. So G is exactly the symmetries of G preserving the right action of G.
Picture G as a graph, where edges between elements are labeled by their right actions.
Then the left action of G is the symmetries of the graph.

1.1.1 The 8 actions of a group on itself

Suppose we have left action of G on S (g, s) 7→ g · s. We can get a right action by putting
s · g = g −1 s. Indeed, we have s(gh) = (gh)−1 s = h−1 (g −1 s) = (sg)h.
4 left actions of G on G:
1. g · s = s, the “trivial” action
2. g · s = gs, the standard left action
3. g · s = sg −1 , a right action “made into” a left action
4. g · s = gsg −1 , the adjoint action, or conjugation 1

1.2 Lagrange’s theorem and consequences

The way we will approach group theory is to list all groups and to prove theorems when
we need to study groups of a particular order (different from the treatment in the Lang
textbook).

I Order 1: the “trivial” group

I Order 2, 3 (prime order): There is just one group of any prime order p.

To prove the latter fact, we need Lagrange’s theorem.

Theorem 1.1 (Lagrange). If H is a subgroup of G, and the order (or number of elements)
of G, H is finite. Then the order of H divides the order of G.
1
Some people write conjugation as g −1 sg.

8
Definition 1.8. The left cosets of H in G are sets of the form gH := {gh : h ∈ H}. The
right cosets of H in G are sets of the form Hg := {hg : h ∈ H}.

Proof. Look at coests of H in G. Any two left cosets have H elements. Also, any two left
cosets are either the same or disjoint. If g1 h1 = g2 h2 , then for h ∈ H, g1 h = g2 (h2 h−1
1 h),
which is in g2 H. So |G| = |H| × number of left cosets.

A special case: if g ∈ G, look at the subgroup H of all powers of g; i.e. H = {g n : n ∈ Z}.

The order of g is the smallest n > 0 such that g n = e (if n exists). The order of subgroup
H is the order of g. If |G| is finite, g ∈ G, then order of g divides order of G. Suppose G
has order p (prime). Pick g ∈ G. Then g has order 1 or p; the first case is g = e, and the
second is for every other element. So G = H.

1.2.1 Applications of Lagrange’s theorem

Theorem 1.2 (Fermat). If g ∈ Z and p (prime) does not divide g, then g p−1 = 1 (mod p).

Proof. Look at group (Z/pZ)∗ . This is a group of order p − 1, so every element has order
dividing p − 1.

Theorem 1.3 (Euler). Suppose g, m ∈ Z are coprime. Then g φ(n) = 1 (mod m), where
φ(m) is the number of irreducible elements of Z/mZ.

Proof. Same as the proof of Theorem 1.2.

1.2.2 Geometric meaning of Lagrange’s theorem

Suppose G acts on a set S transitively. If s, t are in S, then s = g · t for some g. Fix some
s ∈ S. Put H = {h ∈ G : h · s = s ∀s ∈ S}, the elements of G fixing S. Then the points of
S are in bijection with the cosets of H, sending t 7→ {g ∈ G : s.t. g · s = t}. These are left
cosets since if g · s = t, then (gh) · s = t, as h · s = s.
Interpret |G| = |H| × number of left cosets in terms of the action. Then we have that
|G| = number of elements fixing s × number of elements of set S. For example, if G is the
group of rotations of an icosahedron, then

|G| = number of elements fixing center of a face × number of faces.

So in this case, |G| = 3 × 20 = 60.

1.3 Groups of order 4 and product groups

I Groups of order 4: 2 Examples.

I (Z/4Z, +) with elements {0, 1, 2, 3}

9
 I Symmetries of a rectangle {1, a, b, c}

To show that these two are not isomorphic, look a the orders of elements. The orders
of elements in the former are 1, 4, 2, and 4; the orders of elements in the latter are 1, 2, 2,
and 2. Order does not change under isomorphism, so these groups are not isomorphic.
Are these all the groups of order 4? Well, by Lagrange’s theorem, all elements have
order 1, 2, or 4. If a group has an element g of order 4, then the elements are 1, g, g 2 , g 3 with
the product being g a g b = g a+b (mod 4) . Then this is isomorphic to (Z/4Z, +). If all elements
have order 2, G is abelian (commutative). 1 = g 2 h2 = ghgh, so hg = h−1 g −1 = gh, making
G abelian. Writing G additively, G is a vector space over the field F2 with 2 elements. So
G is isomorphic to the unique 2-dimensional vector space over F2 . So, indeed, there are
just 2 groups of order 4.

Definition 1.9. The product of 2 groups G and H is G × H, where the operation is

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

The group {1, a, b, c} is isomorphic to a product of 2 subgroups. {1, a, b, c} ∼

= {1, a} ×
{1, b}, where 1 7→ (1, 1), a 7→ (a, 1), b 7→ (1, b), and c 7→ (a, b).

Example 1.3. R∗ ∼
= {1, −1} × R+ .

Example 1.4. The polar decomposition gives us C∗ = S 1 × R+ .

Example 1.5. If F is a field, the vector space Fn is a product of n copies of F, under

addition.

Example 1.6. Let G be the group  of all roots of 1 in C (contains square roots, cube roots,
fourthroots, etc). Then G = z ∈ C : z = e2πi(m/n) , m, n ∈ Z . Define the subgroups
H1 = z ∈ G : ∃n ∈ Z s.t. z 2n = 1 and H2 = z ∈ G : ∃n ∈ Z s.t. z 2n+1 = 1 . Then G ∼ =
H1 × H2 . In fact, we can separate in this way by any prime, not just by 2.

10
2 Groups of orders 6 and 8
2.1 Two groups of order 6
I Groups of order 6

I the cyclic group Z/6Z

I the symmetric group S3

The former is actually a product2 , Z/6Z ∼

= Z/2Z × Z/3Z. It has nontrivial proper
subgroups A = {0, 3} and B = {0, 2, 4}. G = AB, A ∩ B = {0}, and A, B commute, so
G∼= A × B.

Definition 2.1. The symmetric group is Sn = {permutations of n points 1, 2, . . . , n}

Notation for permutations: (a b c d) is the function taking a 7→ b → 7 c 7→ d 7→

a. The 6 elements are {e, (1 2), (2 3), (1 3), (1 2 3), (1 3 2)}. The proper subgroups are
{e, (1 2 3), (1 3 2)} , {e, (1 2)} , {e, (2 3)} , {e, (1 3)}, and {e}.

2.2 Quotient groups

Fundamental problem: Suppose H is a subgroup of G. We have a set of left cosets aH,
the set of such denoted by G/H. Is G/H a group? The most natural attempt is to define
the operation as (aH)(bH) = (ab)H. The operation we have defined implies that cosets
are equivalence classes for the relation a ≡ b iff aH = bH (meaning a−1 b ∈ H). Is this
well-defined?
Suppose b1 ≡ b2 , so b1 = b2 h for some h ∈ H. Then ab1 = ab2 h, so ab1 ≡ ab2 . Suppose
a1 ≡ a2 . We want a1 b ≡ a2 b. We have a2 hb = a2 b, so we would be done if the group is
commutative. In fact, the condition we need here is hb = bh0 for some h0 ∈ H; so this
operation is well defined if b−1 Hb = H.

Definition 2.2. A subgroup H is normal in G if gH = Hg for all g ∈ G.

Example 2.1. Let G = S3 and H = {e, (1 2 3), (1 3 2)}. Then H is normal.

Remark 2.1. In fact, any subgroup of index 2 is always normal. H is normal ⇐⇒ left
cosets are the same as right cosets. If H has index 2, left cosets are H and G \ H; these
are also right cosets, so H is normal. So G/H is a group of order 2.

Example 2.2. Let G = S3 and H = {e, (1 2)}. H is not normal because (23)H(23)−1 6= H;
we have (2 3)(1 2)(2 3)−1 = (1 3), which is not in H. In this case, the right cosets are not
equal to the left cosets.
2
Cayley once made the mistake of thinking these two were different groups, claiming that there were 3
groups of order 6.

11
2.3 Other groups of order 6
We want to classify the groups of order 6. The first step is to pick an element of order 3.
Why does this exist?
Theorem 2.1. Suppose p is prime and p divides |G|. The G has an element of order p.
Proof. Use induction on the order of the group. Assume this is true for all smaller groups.
First case: G is abelian. Pick some element g of some prime order q; this exists because
any element has order dividing G and if g has order mn, g m has order n. If q = p, we
are done. If q 6= p, then look at group G/ hgi; this has order less than G, so our inductive
hypothesis gives us that G/ hgi has ana elements h or order p. Now lift h to some a ∈ G.
ap ∈ hgi, so a has order p or pq. So a or aq has order p.
Second case: G is not abelian. Look at the adjoint action of G on itself; i.e. g·s = gsg −1 .
Decompose G into orbits under this action. The meaning of a, b being in the same orbit
is thatPa = gbg −1 for some g ∈ G. The orbits partition G into equivalence classes. So
|G| = |Orbit|. Lagrange’s theoremPsays that |Orbit| = |G| / |H|, where H is the stabilizer
of one point of the orbit. So |G| = orbits |G| / |H|. We now have 2 cases:
Case 1: Some H with |H| < |G| has order divisible by p. Then by induction, H has an
element of order p, so G does, as well.
Case 2: If |H| < |G| and |H| is not divisible by p, then |G| / |H| is divisible by p. So
X |G| X |G| X |G| X
|G| = + = + 1.
|{z} |H| |H| |H|
divisible by p orbits orbits orbits orbits
H(G H=G H(G H=G
| {z } | {z }
divisible by p divisible by p

Elements that commute with everything in G, the set of which is called the center of
G, is abelian and has order divisible by p because the term on the right is precisely the
order of the center of G. By the previous cases, the center of G has an element of order p,
so we are done.

Remark 2.2. This does not need to hold if p is not prime. G = Z/2Z × Z/2Z has no
element of order 4, but 4 divides |G|.
Suppose G has order 6. Pick element g of order 3. Then e, g, g 3 is a subgroup of


order 3. It is normal since it has index 2. Pick an element h of order 2. This gives a
subgroup {e, h}, which is not necessarily normal. Then G is a semidirect product of these
subgroups of orders 2 and 3.
Definition 2.3. A direct product of groups A and B is A × B where the operation is
(a1 , b1 )(a2 , b2 ) := (a1 a2 , b1 b2 ).
Here, A and B are both normal and commute. In the following definition, A and B
will not necessarily commute.

12
Definition 2.4. Suppose A is normal and B may not be normal. For each element b ∈ B,
a 7→ bab−1 is an automorphism of A. Suppose we have a automorphism ϕb of A for each
element of B where ϕb1 b2 = ϕb1 ϕb2 (this means we have a homomorphism from B into
Aut(A)). Then a semidirect product of groups A and B is A o B where the operation is
(a1 , b1 )(a2 , b2 ) := (a1 ϕb2 (a2 ), b1 b2 ).
So if we have a action of the group B on A, we can define the semidirect product A o B.
Example 2.3. Let A = Z/3Z, and let B = Z/2Z. The automorphisms of A are the
identity and a 7→ −a. There are 2 ways for B to act on A, the trivial action ϕb (a) = a,
and the nontrivial action ϕb (a) = −a if b 6= e. These produce the two groups of order 6:
Z/6Z and S3 , respectively.
There are no other groups of order 6.

2.4 Groups of order 8

Case 1: All elements have order 2. This implies the group is abelian (same argument as
last lecture), so it is really a vector space over F2 . So ∼
= F2 × F2 × F2 .
 it is G
Case 2: Some element g has order 4. Then H = 1, g, g , g 3 is a subgroup of index 2,
2

so it is normal. We write what is called an exact sequence:

injective surjective
1 → Z/4Z → G → Z/2Z → 1.
z}|{ z}|{
| {z } | {z }
∼
=H ∼
=G/H

f g
Definition 2.5. An exact sequence is a sequence of groups A − →B→ − C, where im(f ) =
ker(g). A short exact sequence is an exact sequence of the form 1 → A → B → C → 1.
Remark 2.3. A standard blunder is to assume that if we have an exact sequence 1 → H →
G → H/G → 1, then G is a direct or semidirect product of H and G/H. A counterexample
is G = Z/4Z and H = Z/2Z.
Remark 2.4. Given A, B ⊆ G with 1 → A → G → B → 1 exact, a common problem is to
find G. G is called the extension of B by A3 . This is hard even when A and B are abelian.
Pick some h ∈ H mapping to a nontrivial element of Z/2Z. So G contains g, h, g 4 = e,
h2 = e, g, or g 2 , and 1, g, g 2 , g 3 , so hgh−1 = g or g 3 .
So we get 6 cases. Note that hgh−1 = g iff G is abelian. We cannot have hgh−1 = g 3
and h2 = g, because then g and h commute, so the group is abelian and not abelian. If
h2 = g and hgh−1 = g, then G = Z/8Z. Otherwise, if hgh−1 = g, then G ∼ = Z/4Z × Z/2Z.
If hgh = g and h = e, we get the dihedral group of order 8. If h = g and hgh−1 = g 3 ,
−1 3 2 2 2

we have the quaternion group. This covers all the cases.

3
This is also sometimes called the extension of A by B.

13
Remark 2.5. The quaternions4 {a + bi + cj + dk : a, b, c, d ∈ R} form a 4 dimensional
division algebra containing C = {a + bi, ∈ R}.

We then have

I Groups of order 8

I the cyclic group Z/8Z

I the product group Z/2Z × Z/2Z × Z/2Z (∼
= F2 × F2 × F2 )
I the product group Z/4Z × Z/2Z
I the dihedral group D8
I the quaternion group Q8

4
The word quaternion actually means soldier. Quaternions (not the mathematical kind) are referenced
in the New Testament of Christianity.

14
3 Non abelian groups of order 8
3.1 The dihedral group
Last time we found two nonabelian groups of order 8, the dihedral group D8 (the symme-
tries of a square) and the quaternion group Q8 .
Problem: How many ways are there to arrange 8 non-attacking rooks on a chessboard?
Here, non-attacking means that two rooks cannot be placed in the same row or column.
There are 8 choices of where to put a rook in the first row, 7 places of where to put a rook
in the second row, and so on. So the number of ways is 8!.
Consider a modification to this problem: how many ways are there to do the above up
to symmetry? Well, D8 acts on the configurations by acting on the chessboard. How many
orbits are there?
Theorem 3.1 (Burnside5 ). Suppose a group G  S. Then the number of orbits under this
action is equal to the average number of fixed points of the action. That is,
1 X
|{Orbits}| = f (g),
|G|
g∈G

where f (g) is the number of elements of S fixed by g.

Proof. Look at the set of pair (g, s) withg · s = s. Count the number P of pairs in two ways:
Method 1: For each g, there are f (g) choices for s. So we get g∈G f (g).
Method 2: Look at one orbit of G of S. Say the orbit contains some s ∈ S. By
Lagrange’s theorem, the number of points in the orbit is |G| / |Gs |, where Gs is the stabilizer
of s. So |G| = |Orbit| × |number of elements in G fixing a point of the orbit|. This means
that the number of elements of G fixing a point in the orbit is the same for each point in
the orbit. Then
X
|pairs (g, s)| = |{pairs in orbit}|
orbits
X
= |Orbit| × |number of elements in G fixing a point of the orbit|
orbits
X
= |G|
orbits
= |G| × |{Orbits}| .
Dividing both our results by |G| gives us the desired equality.

Definition 3.1. Two elements s, b ∈ G are conjugate if there exists some g ∈ G such that
a = gbg −1 . Informally, elements are conjugate if they “sort of look the same.”
5
Many mathematicians have proved this independently of each other, so we could really put anyone’s
name here.

15
 The elements of D8 that are conjugate will have the same number of fixed points. To
calculate the number of configurations fixed by each conjugacy class, it is helpful to draw
pictures and eliminate rows based on the symmetries.

Conjugacy classes of G number of configurations fixed by element

identity 8! = 40320
reflections parallel to sides 2×0=0
switch both diagonals 8 × 6 × 4 × 2 = 384
rotation by π/2 2 × (6 × 2) = 24
reflection along a diagonal 2 × 764 = 1528
The most tricky of these is the last one; let cn be the desired number (not yet multiplied
by 2, the size of the conjugacy class), where the chessboard is n × n. then we have a few
possibilities: if we place a rook in the top left corner, then there are cn−1 ways to arrange
other rooks. If we place a rook elsewhere in row 1, we have cn−2 ways to arrange the other
rooks. So we get a recurrence relation cn = cn−1 + (n − 1)cn−2 , and we can solve to get
c8 = 764.
These sum up to be 42256, so using the above theorem, we have our final answer as
42256/8 = 5282 configurations.

3.2 Quaternions
We can represent quaternions using complex matrices:
       
1 0 i 0 0 i 0 −1
1= I= J= K=
0 1 0 −i i 0 1 0

Any nonzero quaternion has an inverse. Conjugate z̄ = a − bI − cJ − dK, so z z̄ =

a2 + b2 + c2 + d2 . Then z −1 = 1/z = z̄/(z z̄) = z̄/(a2 + b2 + c2 + d2 ), where the denominator
is nonzero if z 6= 0. So nonzero quaternions form a group under multiplication. Call
|z| = a2 + b2 + c2 + d2 . Letting H ∗ be the nonzero quaternions, we have a homomorphism
H ∗ → R∗ that takes z → |z|. This homomorphism has kernel S 3 . In fact, our quaternion
group is a subgroup of S 3 .
Identify R3 = {bI + cJ + dK ∈ H}. The map v 7→ g −1 vg (for g a nonzero quaternion)
maps R3 → R3 . It is a rotation of R3 .6 So we get a homomorphism S 3 → SO3 (R) .
| {z }
rotations of R3
This is not an isomorphism because the kernel has order 2 (exercise). We get a short exact
sequence
1 to {±1} → S 3 → SO3 (R) → 1.
6
In computer graphics, such as in video games, quaternions are used to compute rotations of R3 . They
are quicker to multiply than 3 × 3 matrices.

16
 Pick any finite group of rotations, a subset of SO3 (R). For example, pick rotations of
a rectangle in R3 , Z/2Z × Z/2Z, or rotations of an icosahedron (has 60 elements). The
inverse image of the previous homomorphism is a subgroup of S 3 of twice the order. In
our examples, we get the quaternions and the “binary icosahedral group.”

4 Groups of order 9, 10, and 12

4.1 Groups of order 9
There are two natural examples for the abelian cases:

I Groups of order 9

I Z/9Z
I Z/3Z × Z/3Z

These are the only abelian groups: if we have an element of order 9, then we have Z/9Z,
and if all elements are of order 3 an G is abelian, then it is a product of vector spaces over
3 elements, F3 × F3 ∼= Z/3Z × Z/3Z.
In fact, these two are the only groups of order 9 due to the following theorem:

Theorem 4.1. Let p be prime. All groups of order p2 are abelian.

We require a lemma:

Lemma 4.1. Any group of order pn has nontrivial center.

Proof. Sum over the conjugacy classes of G, picking some g in each Cg . Denote the center
of G as Z. If g is in the center of G, then its conjugacy class has only 1 element: g itself.
Then
X X |G| X |G|
|G| = |Cg | = = + |Z|
g g
|G g | |G g |
g ∈Z
/

Since p divides the order of G and the summation term, p divides the order of the
center. In particular, the center contains at least p elements and is nontrivial.

Proof. Suppose G has order p2 . By our lemma, the center is nontrivial, so the center has
order p or p2 .
However, the center cannot have order p. Suppose it does, and pick some g not in the
center. If g 2 ∈ Z, then g has order p2 , so the group is cyclic and hence abelian. Then
g2 ∈/ Z, so hgi ∩ Z = {e}. Then G = hgi Z, so every element can be written as g n a for some
n and a in the center of G. Then all elements commute with each other, so the center is
all of G, which is a contradiction.
So the center has order p2 and is hence all of G. So G is abelian.

17
4.2 Nilpotent groups
Suppose G0 has order pn . Take its center Z0 , and let G1 = G0 /Z0 . Keep quotienting out
by the center. One might think that the center would be trivial after quotienting out by
the center, but this is actually not true. Take G = {±1, ±i, ±j, ±k} and Z = {±1}; then
G/Z ∼= Z/2Z × Z/2Z has nontrivial center.
Definition 4.1. A group is nilpotent if it can be reduced to 1 element by repeatedly taking
the quotient of its center at each step.
All products of groups of prime power order are nilpotent. Later, we will prove a
converse: any finite nilpotent group is the product of group of order pn .

4.3 Groups of order 2p

For groups of order 10, and more generally order 2p for some prime p, we can generalize
the methods we used for groups of order 6:
1. Pick subgroup H of order p
2. H has index 2 so is normal
3. Pick subgroup S of order 2
As in the case of order 6, G ∼
= H o Z/2Z. This is classified by the ways Z/2Z can act
on Z/pZ. Later, we will show that the automorphisms are isomorphic to (Z/pZ)∗ , so we
get two groups:

I Groups of order 2p
I the abelian group Z/2pZ ∼ = Z/2Z × Z/pZ
I the dihedral group D2p , the symmetries of the regular 2p-gon (nonabelian)

4.4 Groups of order 12

Our list will be:

I Groups of order 12
I the abelian group Z/12Z ∼ = Z/3Z × Z/4Z
I the abelian group Z/6Z × Z/2Z ∼ = Z/3Z × Z/2Z × Z/2Z
I the nonabelian group D12 , the dihedral group of order 12 (∼
= D6 × Z/2Z)7
I A4 , the rotations of a tetrahedron (nonabelian)
I binary dihedral group (nonabelian)

7
D8n+4 splits as a product for any n ∈ N.

18
5 Groups of Order 12,. . . ,24
5.1 Groups of order 12
Last time we introduced A4 , the symmetries of a tetrahedron, and the binary dihedral
group as nonabelian groups of order 12. Recall, that if we have some homomorphism
S 3 → SO3 (R), then the inverse image has order 2 × |G|. If G = S3 , then we get A4 .
Look at the rotations of a tetrahedron; what are the conjugacy classes? We have

1. identity

2. rotation by 2π/3 (4 of these)

3. rotation by 4π/3 (4 of these)

4. pick opposite edges and reflect across them (3 of these).

5.1.1 Sylow theorems

Recall that if H is a subgroup of G, the |H| divides |G|. Suppose m divides |G|. Does G
have a subgroup of order m? In general, the answer is no. 6 divides the order of A4 , but
there is no subgroup of order 6; this is the smallest counterexample. However, the Sylow
theorems provide cases in which this must be true.

Theorem 5.1 (Sylow). Suppose p is prime, pn divides |G|, and pn+1 does not divide |G|.
Then

1. G has a subgroup of order pn (called a Sylow p-subgroup or p-Sylow subgroup).

2. All such subgroups are conjugate.

3. There are 1 (mod p) such subgroups (and this number divides |G|).

4. Any subgroup of order pm with m ≤ n is contained in some subgroup of order pn .

Proof. To prove part (1), we have 2 cases and proceed by induction on |G|. The first case
is when some proper subgroup H has index prime to p. Then pn divides H, so H has a
subgroup of order pn by induction. The second case is when all proper subgroups have
index divisible by p. Look at the adjoint action of G on itself. Then any orbit of G has 1
element (stabilizer = G) or a multiple of p elements (stabilizer of points 6= G). Then, as
we showed before, the order of the center is divisible by p. Pick g ∈ Z of order p. Then
G/ hgi has a subgroup of order pn−1 by induction. The inverse image of this subgroup has
order pn .
See Lang for parts (2),(3), and (4), or do them as an exercise.

19
 Applying this theorem to subgroups of order 3 of groups of order 12, the number of
such subgroups is 1 (mod 3) and divides 12. Then the number of is 1 or 4. If it is 1, then
the subgroup is normal. Also by the Sylow theorem, G has a subgroup of order 22 = 4.
In this case, G is a semidirect product of a normal subgroup of order 3 and a subgroup
of order 4. Look at the action of a group of order 4 on it. If Z/4Z acts trivially, we get
Z/12Z. If it acts nontrivially, we get the binary dihedral group. If we have Z/2Z, and it
acts trivially, we get Z/3Z × (Z/2Z)2 ; the nontrivial action gives us D12 .
If we have 4 subgroups of order 3, label them A1 , A2 , A3 , A4 , where Ai ∩ Aj = {e} if
i 6= j. So we get 8 = 4 × 2 elements of order 3. This leaves 4 elements not of order 3.
We know there is a subgroup of order 4 (by Sylow), so we get 3 elements of order 4, and
this subgroup is normal. So G is the semidirect product of the subgroup of order 4 by
subgroups Z/3Z. Z/3Z acts nontrivially on the subgroup of order 4. The only possibility
is Z/3Z acting on (Z/3Z)2 , so there is only 1 possible group. This group is A4 , since it has
4 subgroups of order 3 (fix one of the 4 vertices).

5.2 Solvability
So far we have shown that groups of order ≤ 12 can be split up into products with cyclic
groups.
Definition 5.1. A finite group is called solvable if either
1. it is cyclic

2. it has a normal subgroup N with N and G/N solvable.

Definition 5.2. G is callled simple if has no normal subgroup other than {e} and itself.
Example 5.1. The rotations of an icosahedron is a non-cyclic simple group. Look at the
conjugacy classes:
1. identity (order 1)

2. rotation by 2π/3 (order 3, 20 of them, each corresponding to a face)

3. rotation by 2π/5 (order 5, 12 of them, each corresponding to a vertex)

4. rotation by 4π/5 (order 5, 12 of them, each corresponding to a vertex)

5. rotation by π (order 2, 15 of them, (number of edges)/2).

Any normal subgroup must be a union of conjugacy classes. Suppose n is the order of a nor-
mal subgroup. Then n = 1+ some of {12, 12, 15, 20}, and n = 1, 2, 3, 5, 6, 10, 12, 15, 20, 30, 60.
Then the only solutions are n = 1 or n = 60, which shows that this group is simple.
Every finite group can be split up into simple groups.

20
Theorem 5.2 (Jordan-Holder). The set of simple groups we get does not depend on the
choice of splitting.

Proof. See Lang.8

Finite simple groups have been classified as 18 types in infinite series and 26 others
(sporadic).

Example 5.2. GLn (Fp ) gives rise to SLn (Fp ) by quotienting out by the kernel of the
determinant map, and SLn (Fp ) gives rise to PSLn (Fp ) by quotienting out by the center.

5.3 Groups of order 13, 14, and 15

13 is prime, and 14 is of order 2p, so our previous results give us:

I Groups of order 13

I Z/13Z

I Groups of order 14

I Z/14Z
I the dihedral group D14

For groups of order 15, we prove general results for groups of order p, q for primes p < q.
The Sylow theorems give us that G has a subgroup of order q. The number of conjugates
is 1 (mod q) and divides pq. So the only possibility is 1. So G has a normal subgroup
Z/qZ. SO G is a semidirect product of Z/qZ by Z/pZ. How can Z/pZ act on Z/qZ?
Aut(Z/qZ) = (Z/qZ)∗ , which has order q − 1. This is cyclic (will prove later when we
cover fields), so it has 1 subgroup of order p if p divides q − 1. So either p does not divide
q − 1 or p divides q − 1. In the first case, the only subgroup of order pq is cyclic, so we get
1 group of order 15. In the second case, there are 2 groups: the first is the cyclic group
(comes from the trivial action), and the second is Z/qZ o Z/pZ. We summarize this as

I Groups of order pq (p < q)

I If p divides q − 1
I Z/pqZ
I If p does not divide q − 1
I Z/pqZ
8
Professor Borcherds couldn’t really make sense of the proof in Lang, and he has never actually used
the Jordan-Holder theorem, which is why proof here has been omitted.

21
 I Z/qZ o Z/pZ.

Example 5.3. Let p = 2. 2 divides q − 1, so we get the cyclic and dihedral groups.

Example 5.4. Let p = 3 and q = 7. 3 divides 7 − 1, so we get a nonabelian group. This

is the smallest non-abelian group of odd order.

5.4 Groups of order 16

Groups of order 16 are a mess (same is true for pn , where n ≥ 4). We just list them and
not prove anything.

I Groups of order 16

I Abelian
I Z/16Z
I Z/8Z × Z/2Z
I Z/4Z × Z/4Z
I Z/4Z × (Z/2Z)2
I (Z/2Z)4
I Nonabelian, with an element of order 8
I Generalized quaternion: g 8 = 1, aga−1 = g −1 , a2 = g 4
I Dihedral: g 8 = 1, aga−1 = g −1 , a2 = 1
I Semidihedral: g 8 = 1, aga−1 = g 3 , a2 = 1
I (Nameless): g 8 = 1, aga−1 = g 5 , a2 = 1
I Products
I Q8 × Z/2Z
I D8 × Z/2Z
I Semidirect products
I Z/4Z o Z/4Z
I (Z/2Z)2 o Z/4Z
I Miscellaneous
I Pauli matrices, generated by the matrices
       
±1 0 ±i 0 0 ±1 0 ±i
, , , .
0 ±1 0 ±i ±1 0 ±i 0

22
5.5 Finitely generated abelian groups
So far, the finitely generated abelian groups we know about are finite products of Z and
Z/nZ for n ≥ 1. These are actually all the examples.

Theorem 5.3. Let G be a finitely generated abelian group. Then G is a finite product of
groups of the form Z or Z/nZ for n ≥ 1.

Proof. Suppose G is abelian (written additively), generated by g1 , . . . , gn . We have the

relations m1,1 g1 + m1,2 g2 + · · · + m1,n gn = 0, etc, which give us a (possibly infinite) matrix
of the coefficients. We can simplify the matrix by adding k times any column to any other;
this is a change of generators gi 7→ gi + kgj . We can add k times any row to any other
row; if rows R = S = 0, this is equivalent to R = 0 and kR + S = 0. We can apply
these operations to make m1,1 as small as possible. Subtract multiples of column 1 from
other columns to make row 1 have only 1 nonzero entry (m1,1 ). This is possible because
m1,1 divides m1,2 ; otherwise m1,2 = km1,1 + r for |r| < m1 , and we could subtract km1,1
from m1,2 and then subtract r = m1,2 from m1,1 to make m1,1 smaller. We can kill off the
first column in the same way, leaving m1,1 as the only nonzero entry in the first column.
Repeat this whole process with m2,2 and so on to get a matrix where only mi,i is nonzero
for 1 ≤ i ≤ n. So our group is now generated by g1 , . . . , gn with the relations m1,1 g1 = 0,
m2,2 g2 = 0,. . . . So G ∼
= Z/m1,1 Z × Z/m2,2 Z × · · · × Z/mn,n Z, where if mi,i = 0, we just
have Z in the product.

Remark 5.1. This decomposition is unique if we insist that mi,i divides mj,j for i < j or
if we insist that all mi,i are prime powers or 0, and order does not matter.

5.6 Groups of order 17,. . . ,24

17 is prime, so we have

I Groups of order 17

I Z/17Z

Groups of order 18 have a normal subgroup of order 32 . We can then classify the groups
by semidirect products to get 5 groups.

I Groups of order 18

I 5 semidirect product groups

I Groups of order 19

I Z/19Z

23
 Groups of order 20 have a normal subgroup of order 5. We can then classify the groups
by semidirect products to get 5 groups, as in the case of order 18.

I Groups of order 20

I 5 semidirect product groups

21 = pq for p = 3 and 7 = q (and 3 divides 7 − 1), so we have

I Groups of order 21

I Z/pqZ

22 = 2p, so we have

I Groups of order 22

I Z/22Z

23 is prime, so we have

I Groups of order 23

I Z/23Z

I Groups of order 24

I the symmetric group S4

I Binary dihedral group (inverse image of A4 under S 3 → SO3 )
I a dozen or so others. . .

24
6 Symmetric Groups
6.1 Basic definitions
Definition 6.1. The symmetric group Sn is the group of all permutations of the n points
{1, . . . , n}.

|Sn | = n! because there are n choices for the image of 1, then n − 1 choices for the
image of 2, etc. We denote elements using cycle notation: (a b c d) is the function taking
a 7→ b 7→ c 7→ d.

Definition 6.2. A transposition is a permutation that exchanges 2 elements and fixes all
others.

Proposition 6.1. Sn is generated by the transpositions (1 2), (2 3), (3 4), . . . , (n − 1 n).

Proof. This is “bubblesort,” the “2nd worst” sorting algorithm.9 In the worst case, bub-
blesort takes n(n − 1)/2 exchanges to sort a list of n elements.

6.2 The alternating group

Look at Sn acting on variables x1 , . . . , xn . It also acts on C[x1 , . . . , xn ], polynomials in n
variables. Look at the discriminant,
Y
∆ = (x1 − x2 )(x2 − x3 ) · · · (x1 − x3 ) · · · (xn−1−xn ) = (xi − xj ).
i<j

Any σ ∈ Sn maps ∆ to ∆ or −∆, so there exists some homomorphism ε : Sn → {±1}.

Definition 6.3. The alternating group An is the subgroup of elements σ ∈ Sn such that
σ(∆) = ∆); this is the kernel of ε.

So An is normall in Sn , of order n!/2.

6.3 Sn , An , and platonic solids

Symmetries of platonic solids are very closely related to the groups Sn and An .
The rotations and reflections of a tetrahedron is S4 , acting on the vertices; the rotations
are then A4 . The rotations of a cube or an octahedron are given by S4 acting on the
permutations of the diagonals; then the rotations and reflections are given by S4 × Z/2Z.
The number of rotations of a dodecahedron or an icosahedron is given by permutations of
the five inscribed “inner cubes,” which gives a homomorphism of rotations to S5 , and this
group is A5 ; then the rotations and reflections are given by A5 × Z/2Z.
9
The “worst” algorithm is called bogosort.

25
 We summarize the results in this table:

platonic solid number of rotations number of rotations and reflections

tetrahedron 12 (A4 ) 24 (S4 )
cube/octahedron 24 (S4 ) 48 (S4 × Z/2Z)
dodecahedron/icosahedron 60 (A5 ) 120 (A5 × Z/2Z ∼ 6 S5 )
=

These groups are “spherical reflection groups.”

6.4 Conjugacy classes of Sn

We can write any element of Sn as a product of disjoint cycles.

Definition 6.4. The cycle shape is the sizes of the cycles with multiplicities.

Example 6.1. The permutation (1 2 4)(5 7 8)(6 9)(10)(3) has cycle shape 32 , 2, 12 .

Two elements are conjugate if they have the same cycle shape. Given a, b, with the
same cycle shape, how can we find g with a = gbg −1 ? Write out the two permutations in
cycle notation and pair off elements:

(1 2 4)(5 7 8)(6 9)(10)(3)

↑↑↑ ↑↑↑ ↑↑ ↑ ↑
(2 4 5)(6 7 8)(1 3)(9)(10)
This gives us g = (1 6 5 4 2 1)(3 9 10)(7)(8).

Example 6.2. How many conjugacy classes are the of S4 ? This is the number of cy-
cle shapes, which is also the number of partitions of 4. Denoting Cσ as the conjugacy
class (viewing S4 as the rotations of a cube) and Gσ as the stabilizer under the action of
conjugation (also is the centralizer).
partition cycle shape Cσ |Gσ | |Cσ | = |G| / |Gσ |
1+1+1+1 14 identity 24 1
2+1+1 2, 1 2 rotation by π 4 6
3+1 3, 1 rotation by 2π/3 3 8
2+2 2 2 rotation by π 8 3
4 4 rotation by π/2 4 6
If σ has cycle shape 1n1 2n2 3n3 · · · , then the number of elements in the centralizer is
1n1 n1 ! · 2n2 n2 ! · · · .

26
6.5 Normal subgroups of Sn
What are the normal subgroups of Sn ? We already know of {e}, An , and Sn . Viewing S4
as the rotations of a cube, we have that S4 acts on 3 lines by permuting them; so we have a
homomorphism S4 → S3 , where the kernel is a normal subgroup of order 4 (the identity +
3 rotations by π). Following this pattern, we have homomorphisms S2 onto S1 , S3 onto S2 ,
and S4 onto S3 . However, the pattern breaks because there is no homomorphism from S5
onto S4 ; S5 has a simple subgroup A5 , the rotations of an icosahedron. If N is any normal
subgroup of S5 , N ∩ A5 is normal in A5 , so it is 1 or 5. So the only normal subgroups of
S5 are {e}, A5 , and S5 .

Theorem 6.1. An is simple for n ≥ 5.

Proof. We sketch a proof using induction on n. Suppose N is normal in Sn . Pick an

element g ∈ N with g 6= e. Find h so that ghg −1 h−1 fixes the point 1 (exercise). So
ghg −1 h−1 = g(hg −1 h−1 ) is also in N , which makes N have nontrivial intersection with
Sn−1 (things fixing 1). So N ∩ Sn−1 = An−1 or Sn−1 . So N contains all elements of An
fixing 1. Similarly, it contains all elements fixing i for any i. These generate An (also an
exercise).

Example 6.3. There are three groups of order 120 containing A5 and Z/2Z as composition
factors.

1. A5 × Z/2Z

2. S5 , which has a subgroup A5 and the quotient group Z/2Z

3. Binary icosahedral group10 , which has a quotient group A5 and a subgroup Z/2Z

6.6 Outer automorphisms of Sn

Conjugation is an automorphism of a group G, and we get an exact sequence

1 → Z(G) → conjugations → Aut(G) → outer automorphisms → 1.

For n ≥ 3 with n 6= 6, Aut(Sn ) ∼ = Sn , and all these automorphisms are inner automor-
phisms.
Let’s find a non-inner automorphism of S6 . Start with S5 . This has a subgroup of order
20. S5 acts on 0, 1, 2, 3, 4 ∈ F5 , and has the following subgroup: all permutations of the
form x 7→ ax + b for a, b ∈ F5 . So S5 has a subgroup of index 6, so it acts transitively on
6 points, giving us a homomorphism from S5 → S6 which is different from the usual such
10
Let G be this group. Then S 3 /G, the cosets of G in S 3 (not a group), has the same homology as S 3
but is not homeomorphic to S3 .

27
homomorphisms that fix some element (which are not transitive). S6 has 12 subgroups
∼
= S5 , not 6, as we might expect.
Any subgroup of index n in G products a homomorphism from G → Sn , where G acts
transitively on n points, so any subgroup of index 6 in S6 gives a homomorphism from
S6 → S6 . Pick one of the “funny” homomorphisms S5 → S6 to get a homomorphism from
S6 → S6 . Check that this is not an inner automorphism (exercise).

28
7 Category Theory
7.1 Categories
The purpose of category theory is to generalize common properties of existing structures
so we do not need to refer to the internal structure of our objects at all.

Definition 7.1. A category is a collection of objects and a set of morphisms such that

1. Each morphism has a domain and a range, both of which are objects

2. For each object a, there is an identity morphism 1a

3. For morphisms X : a → b and Y : b → c, there is a composite morphism Y ◦ X

4. (X ◦ Y ) ◦ Z = X ◦ (Y ◦ Z) if both are defined

5. 1b ◦ X = X ◦ 1a = X if X : a → b

Example 7.1. In the category of sets, the objects are sets, and the morphisms/arrows are
functions.

Example 7.2. In the category of groups, the objects are groups, and the morphisms/arrows
are group homomorphisms.

Example 7.3. In the category of topological spaces, the objects are topological spaces,
and the morphisms/arrows are continuous functions.

Example 7.4. Take a category with a single object, and let the morphisms be the elements
of a group G, where composition of the morphisms is the group operation. This is a group.

Example 7.5. Let S be a partially ordered set with ≤. We can make a category with
objects equal to the elements of S and morphisms from a → b such that there is 1 morphism
if a ≤ b and 0 otherwise.

7.2 Functors
Definition 7.2. A (covariant) functor F from a category C to a category D is defined by
the properties

1. F is a function from objects of C to objects of D

2. F is a function11 from morphisms of C to morphisms of D

3. F (1A ) = 1F (A)
11
Really, these are two separate functions, but we refer to them together as one function, the functor F .

29
 4. F (f ◦ g) = F (f ) ◦ F (g).
Let f : A → B be a morphism. The fourth condition makes it so F (f ) is a morphism
from F (A) → F (B); this is because we can set g = 1A .
Example 7.6. We can define a functor F from the category of groups to the category of
sets by F (G) = the underlying set of G. F sends group homomorphisms to themselves as
functions.
Example 7.7. The reason why functors were introduced was to study homology groups
Hi . Hi is a functor from topological spaces to abelian groups.
Example 7.8 (Abelianization of a group). Suppose G is a group. We can make G abelian
by quotienting out G/ ghg h−1 : g, h ∈ G to get an Abelian group Gab . This is a
−1


functor from Groups to Abelian groups. If f : G → H, we get a map Gab → H ab

(exercise).
Example 7.9. We have a functor from sets to abelian groups given by F (S) = Fab (S), the
· · such that P
free abelian group on S. This is the set of elements n1 s1 +n2 s2 +·P all but finitely
many ni = 0. If f : S → S 0 is a function, F (f 0 ) : S → S 0 sends α nα sα 7→ α nα s0α .
Example 7.10. Take a group G, viewed as a category with 1 object. A functor from the
group to sets will send the 1 object to some set and each g ∈ G to some function S → S.
So we get the action of G on a set S, the permutations of S.
Definition 7.3. A contravariant functor is a functor where F (f ◦ g) = F (g) ◦ F (f ).
Similarly to the note we made above, this property implies that if f : A → B is a
morphism, F (f ) is a morphism from F (B) → F (A).
Example 7.11. Let both categories be vector spaces over the same field K. We can
define a functor F (V ) = Hom(V, K); this is V ∗ , the dual of V . Suppose f : V → W is a
morphism; we must map it to some morphism F (f ) : W ∗ → V ∗ . We get the morphism
λ 7→ λ ◦ f .
Example 7.12. Suppose C is the category of a abelian groups. Look at Hom(A, B) for
abelian groups A, B. This is a bifunctor in 2 variables form C ×C → C. It is covariant in B
and contravariant in A. If f : B1 → B2 , we get a map F (f ) : Hom(A, B1 ) → Hom(A, B2 ).
If g : A1 → A2 , we get a map F (g) : Hom(A2 , B) → Hom(A1 , B).
Example 7.13. We can have the category of categories, where the objects are categories
and the morphisms are functors.
Remark 7.1. This does not actually exist because there is no set of all sets. Let R =
{x : x ∈
/ x}; then R ∈ R ⇐⇒ R ∈ / R. Similarly, the category of all groups does not exist,
either. We have a few possible solutions:

30
 1. Only work with groups whose elements are in some fixed large set

2. Work in set theory with “classes”

3. Grothendieck universes

4. Ignore it

We will adopt the 4th solution.

7.3 Natural transformations

What does natural mean? Look at finite dimensional vector spaces. We know that V ∼
= V ∗,
but there is no natural isomorphism. However, V ∼ ∗∗
= V with a “natural isomorphism”

v 7→ fv , where for each w ∈ W , fv (w) = w(v).

Definition 7.4. Suppose we have 2 categories C, D with functors F : C → D and G :

C → D. A natural transformation ϕ : F → G is a function ϕ such that

1. ϕ(a) is a morphism from F (a) → G(a)

2. if f : a → b, ϕ(b) ◦ F (f ) = G(f ) ◦ ϕ(a). That is, the following diagram commutes:

ϕ(a)
F (a) G(a)
F (f ) G(f )
ϕ(b)
F (b) G(b)

Example 7.14. Look at C = D = vector spaces over a field K. Let F be the identity from
C to D, and let G be the double dual, G(V ) = V ∗∗ . Then there is a natural transformation
from F → G. For each vector space V , we have a morphism (in fact an isomorphism since
it has an inverse) from F (V ) → G(V ) that satisfies the conditions above.

7.4 Products
The product is a typical construction in many spaces. Familiar examples include:

Example 7.15. The product of sets A, B is A × B = {(a, b) : a ∈ A, b ∈ B}.

Example 7.16. The product of groups A, B is A × B = {(a, b) : a ∈ A, b ∈ B} with a

group operation given by (a, b)(c, d) = (ac, bd).

Example 7.17. The product of topological spaces A, B is A × B = {(a, b) : a ∈ A, b ∈ B}

with the product topology.

31
Definition 7.5. Suppose X is any object with morphisms f : X → A and f : X → B.
Then a product A × B of A and B is an object with morphisms π1 : A × B → A and
π2 : A × B → B such that there exists a unique map ϕ : X → A × B such that ϕ ◦ π1 = f
and ϕ ◦ π2 = g.
X
ϕ
f g
A×B

π1 π2
A B
This property defines A × B up to canonical isomorphism (a morphism f : A → B such
that we can find g : B → A with f ◦ g = 1B and g ◦ f = 1A ). Suppose X, Y are both
products of A and B. Then the composition of the two maps ϕ, ψ between X and Y is the
identity by the uniqueness of the map defined above.
So we can define products in any category, and this definition ignores the internal
structure of the objects.

7.5 Equalizers
Definition 7.6. Let A and B be objects in a category. The equalizer of two morphisms
f, g : A → B is an object X and a morphism h : X → A such that
1. f ◦ h = g ◦ h

2. If Y is an object with i : Y → A such that f ◦ i = g ◦ i, then Y factors uniquely

through X.
That is, the followng diagram commutes:
f
h
X A B
g

Suppose A, B are groups with f : A → B. The kernel of f is the equalizer of f and 1,

the trivial map from A → B.

7.6 Initial and final objects

Definition 7.7. A is an initial object if there is a unique morphism from A to any other
object in the category.
Initial objects are unique up to isomorphism (exercise).

32
Example 7.18. The empty set is an initial object in the category of sets.
Example 7.19. The trivial group is an initial object in the category of groups.
Definition 7.8. A is a final object if there is a unique morphism from any other object in
the category to A.
Example 7.20. A 1-element set is a final object in the category of sets.
Example 7.21. The trivial group is a final object in the category of groups.

7.7 Limits and pull-backs

Definition 7.9. A limit of {Aα } is an object X with morphisms fα : X → Aα , character-
ized by the following properties:
1. If gα,α0 : Aα → Aα0 is a morphism, then fα0 = gα,α0 ◦ fα .

2. Any Y with this property factors through X.

Y
ϕ

X
f2
f1 f3
A1 g1,2 A2 g2,3 A3
g3,1

Example 7.22. A product is a limit of A and B.

Example 7.23. The equalizer is a limit of A and B with morphisms f, g : A → B.
Definition 7.10. The pull-back X is a limit of A and B with morphisms f : A → C and
g : B → C.
Y
ϕ

X
π1 π2
A B
f g

C
Example 7.24. The pull-back of sets A, B is {(a, b) ∈ A × B : f (a) = g(b)}.

33
7.8 Coproducts
If we reverse the arrows in a product, we get a coproduct.

X
ϕ
f g
AqB
π1 π2
A B
Example 7.25. In the category of sets, the coproduct is the disjoint union.

Example 7.26. In the category of abelian groups, the coproduct equals A × B, so the
coproduct equals the product.

In the category of groups, what is the coproduct of A and B? It is the free group on
two generators. We will discuss this next lecture.
We can also take infinite products and coproducts. The infinite product of abelian
groups is the usual infinite product, and the infinite coproduct of abelian groups is the
subgroup of the infinite product such that all but finitely many of the coordinates vanish.

34
8 Free Groups
8.1 Free abelian groups
Definition 8.1. The
Pn free abelian group on n generators g1 , g2 , . . . , gn is the group of ele-
ments of the form i=1 ni gi with ni ∈ Z.
There exists a unique isomorphism Zn ∼ = G taking ei 7→ gi . In categorical terms, the
free abelian group is the coproduct of n copies of Z.

G
π1 π4
π2 π3
Z Z Z Z

Call n, the exponent of Z, the rank of the free abelian group. Is the rank determined
by G = Zn ? Yes, because the number of homomorphisms from Zn → Z/2Z is 2n .
Proposition 8.1. Any subgroup of Zn is free of rank ≤ n.
Proof. Recall the proof that finitely generated abelian groups are products of cyclic groups.
We showed that if A is a subgroup of Zn , we can find generators g1 , . . . , gn of Zn . So A is
generated by n1 g1 , n2 g2 , . . . for some ni , making A free of rank ≤ n.

8.2 The free group on g1 , . . . , gn

8.2.1 Construction of the free group
Take all words in symbols g1 , g1−1 , g2 , g2−1 , . . . , gn , gn−1 , including the empty word. For ex-
ample 1, g1 , g1 g2 , g1 g2 g1−1 g2 , etc. These have an associative product, which is just concate-
nation of words. However, aa−1 is not the identity, so we still have some work to do. Take
the smallest equivalence relation such that g1 g1−1 ≡ 1, g2 g2−1 ≡ 1,g3 g3−1 ≡ 1, . . . and such
that if a ≡ b, then ac ≡ bc and ca ≡ cb. This second condition ensures that the product is
well-defined.
Definition 8.2. The free group on generators g1 , . . . , gn is the group of equivalence classes
of words in symbols g1 , g1−1 , g2 , g2−1 , . . . , gn , gn−1 under this equivalence relation, with the
group operation being concatenation of words.
What does the free group look like? It can be identified with “reduced” words in
g1 , g1−1 , . . . , where reduced means that we cancel out g1 g1−1 , g2 g2 −1, etc.
If A and B are different reduced words, are they different in the free group? Yes. It is
sufficient to show that AB −1 is empty. Then it is sufficient to show that if A is a reduced

35
word other than 1, the empty word, then A 6≡ 1. We show that if A 6= 1 is a reduced word,
we can find a finite group G and a map from the free group to G so that the image of A
under this map is nontrivial. This is the statement that free groups are residually finite
(nontrivial elements can be detected by finite groups).
Our finite group will be Sn for n = 1+length of the word A. We will illustrate the
argument with an example. Let A = b−1 a−1 ba−1 ba. Draw the following graph on n
vertices:
a b a−1 b a−1 b−1

Map a to an element of S11 that respects the arrows on the graph; here, we must send a
to a permutation σa with σa (1) = 2, σa−1 (3) = 4, and σa−1 (5) = 6, and we must send b
to a permutation σb with σb (2) = 3, σb (4) = 5, and σb−1 (6) = 7. The constraints on a−1
become constants on a by noting that σa−1 (x) = y ⇐⇒ σa (y) = x; the same holds for b−1 .
Then A gets mapped to the permutation σb−1 σa−1 σb σa−1 σb σa , and this permutation sends
the leftmost vertex, representing the element 1, to the rightmost vertex, representing the
element n = 7.
There are two cases we need to watch out for. The first case is when we have two
arrows labeled (without loss of generality) a going into the same vertex. The second case
is when we have two arrows labeled a leaving the same vertex. But these are the graphs

a a a a−1
=

a a a−1 a
=
We cannot have aa−1 or a−1 a, lest these be reduced to the empty word. So our construction
holds, and we are done.
For free abelian groups, a and b are isometries of the euclidean plane. For free groups,
a and b are isometries of the hyperbolic plane.

8.2.2 Subgroups of free groups

Subgroups of free groups are free, but may have larger rank.
Look at the set of left cosets of G in F , and F acts on the left cosets. This gives the
action of the generators g1 , g2 , . . . on the cosets. Pick one point, and call it the base point.
This action determines a subgroup of index n of things fixing the base point. The number
of subgroups of index n of F are in bijection with the connected graphs on n points with
G-colored cycles.

36
 a b a
b
b b
a
a
a
b

Free groups are the same as fundamental groups of connected graphs with a designated
base point. The fundamental group is the set of homotopy classes of loops from the base
point to itself, where loops follow the edges of the graph. Here, homotopy classes mean
that two paths are equivalent if the difference between the two are just edges traversed
that were immediately retraced backwards. The inverse of a path is the path traversed
backwards.
The fundamental group of the graph containing 1 point with n loops to itself is the free
group on n generators. Conversely, the fundamental group of any connected graph is a
free group. Why? Pick an edge with distinct vertices; then we can contract the two points
into one without changing the fundamental group. We can repeat this until there is only 1
point left, and we can then identify the fundamental group of the graph with a free group.

Example 8.1. Let G be a subgroup of index 2 of the free group on a, b.

a
b
b a

Contract the bottom edge, so we get a free group on 3 generators. What are these gener-
ators? Following the loops from the base point to itself, we get the generators b, a2 , and
a−1 ba.
a2

b a−1 ba

Not every subgroup of a finitely generated group is finitely generated.

Example 8.2. We will find a subgroup of the free group on 2 generators that is not finitely
generated. Consider the following graph:
a a a a a a a a
... ...

b b b b b b b

Contracting all the edges labeled a, we get 1 vertex

 kwith infinitely many loops going to
−k
itself. These loops are labeled with the generators a ba : k ∈ Z .

37
Example 8.3. Map the free group on two generators a, b to S3 so a and b permute the
vertices {1, 2, 3} as follows:

a a
b

S3 is generated by a, b with the relations a2 = 1, b2 = 1, and (ab)3 = 1. What is the kernel

of this map? The kernel is a subgroup of index 6.

a a
a b a
b a b
b b
a

Contract along the inner edges to get the desired free group.

If G has index n in the free group on m generators, then the graph of Fm has Euler
characteristic 1 − m; the Euler characteristic of a graph is |vertices| − |edges|. The graph
of G has n vertices and mn edges, so it has Euler characteristic m − nm = n(1 − m). So
the number of generators is n(1 − m).

38
9 Rings
9.1 Definition and examples
Definition 9.1. A ring is a set R along with two binary operations, + and ×, such that

1. R is an abelian group under +

2. × is associative

3. a(b + c) = ab + ac, (a + b)c = ac + bc.

We also have two optional axioms:

1. × has identity12 1 such that 1a = a1 = a.

2. ab = ba (commutative rings).

Example 9.1. The integers, Z, are a ring.

Example 9.2. The Gaussian integers, Z[i] = m + ni : m, n ∈ Z, i2 = −1 , are a ring.



Example 9.3. Polynomials over a field K, K[x], are a ring.

Example 9.4. The set of n × n matrices with entries in K, Mn (K), is a ring.

P
Example 9.5. The Burnside ring of a group G = S3 is the set of all sums ni Ai for
ni ∈ Z and Ai some transitive permutation representation of G (up to isomorphism). The
4 transitive permutation representations of S3 are conjugacy classes: {1, (1 2)}, {1, (1 3)},
{1, (2 3)}, {1, (1 2 3), (1 3 2)}. We get the adjoint representation on 6 points, 3 points, 2
points, and 1 point, so we get sums of the form aA1 + bA2 + cA3 + dA6 .
Any permutation representation is the union of transitive ones. So the set of all finite
permutation representations (up to isomorphism) is the elements of aA1 + bA2 + cA3 + dA6 .
This is not a ring, but we can force it to be by adding −.13
+ in this ring is the disjoint union of representations. × in this ring is the product of
permutation representations. In particular, we have the multiplication table

× A1 A2 A3 A6
A1 A1 A2 A3 A6
A2 A2 A2 ⊕ A2 A6 A6 ⊕ A6
A3 A3 A6 A ⊕ A6
3 A ⊕ A6 ⊕ A6
6

A6 A6 A ⊕ A6
6 A ⊕ A6 ⊕ A6
6 A ⊕ A ⊕ A6 ⊕ A6 ⊕ A6 ⊕ A6
6 6

12
It is sometimes common in analysis to consider rings that do not have an identity element.
13
This is the same thing one does in the construction of the integers from the natural numbers. Doing
this to any commutative monoid returns what is called the Grothendieck group.

39
9.2 Analogies between groups and rings
We can draw a parallel between groups and rings.

• A set S (in relation to groups) corresponds to the vector space with basis S (for
rings).

• The symmetric group Sn (symmetries of {1, 2, . . . , n}) corresponds to Mn (K) (linear

transformations of K n ).14

• We study G by making G act on some set. We study rings by making them act on
K n.

• Sets A, B have A q B and A × B with a + b and ab elements, respectively. Given

vector spaces V, W with respective dimensions a and b, V ⊕ W has dimension a + b;
the tensor product15 V ⊗ W has the property that if A is a basis for V and B is a
basis for W , then A × B is a basis for V ⊗ W , so V ⊗ W has dimension ab.

• |A ∪ B| = |A|+|B|−|A ∩ B|. Similarly, if V and W are vector spaces, dim(V ∪W ) =

dim(V ) + dim(W ) − dim(V ∩ W ).

Remark 9.1. If D = A ∪ B ∪ C, then |D| = |A| + |B| + |C| − |A ∩ B| − |A ∩ C| − |B ∩ C| +

|A ∩ B ∩ C|. This is not true for vector spaces. Let U, V, W be 2 dimensional vector spaces
in R3 containing some fixed line.

9.3 Group rings

Definition 9.2. Let G be a group and R a commutative ring. The group ring R[G] is the
free abelian group with basis G, where × is the group operation on G extended linearly.

Example 9.6. Let G be the Klein 4 group {1, a, b, c} with a2 = b2 = c2 = 1, ab = c, . . . .

So C[G] is a 4 dimensional vector space with basis a, b, c, d. It is a product of 4 copies of
the ring C.
Look at e1 = (1 + a + b + c)/4, e2 = (1 + a − b − c)/4, e3 = (1 − a + b − c)/4, and
e4 (1 − a − b + c)/4. Any product of two different ones of these is 0 and all have square
themselves. This is ei ej = 0(if i 6= j) and e2i = ei . This latter statement says that the ei
are idempotent.

More generally, for a ring R, suppose e ∈ R is idempotent. Then R = eR ⊕ (1 − e)R,

both of which are rings. Conversely, in A × B, (1, 0) is idempotent. So the presence of
idempotents is equivalent to the ring splitting as a product.
14
Sn is the Weyl group of GLn (K)
15
In older texts, this is sometimes referred to as the Kronecker product.

40
Example 9.7. Let G be the monoid G =  N. Then Z[G] is still a ring if we take our basis
to be x0 , x1 , x2 , . . . . This makes Z[G] = n0 x0 + n1 x1 + n2 x2 + · · · , the polynomial ring.
If we take G = Z, we get the Laurent polynomials in Z.

9.3.1 An alternative description of R[G]

We can think of elements of R[G] asPfunctions from G → R, where f (gi ) = ri . Then the
product of R[G] is given by f h(g) = g1 g2 =g f (g1 )h(g2 ), which is called the convolution of
f and h.
Let G = R, which is not finite.
R Consider the ring of all compactly supported continuous
functions f . Then f ∗ h(x) = f (y)h(x − y) dy, another type of convolution. This is a ring
under convolution, but it does not have an identity element for convolution.16

9.4 Ideals
Ideals correspond to normal subgroups (kernels of homomorphisms). We define ideals by
the properties we need for the kernel of a homomorphism.

Definition 9.3. An ideal I of a ring R is a subset of R such that

1. I contains 0R and is closed under addition and subtraction (I is a normal subgroup

of R with respect to addition)

2. If r ∈ I and t ∈ R then rt, tr ∈ I (stronger than saying that I is closed under ×).

We must check that the two conditions above are sufficient. Suppose I satisfies these.
Can we form R/I? Addition is well defined since I is a normal subgroup of R with respect
to addition. To see if multiplication is well defined, we first define multiplication to be
(aI)(bI) = (ab)I. We want that if a ≡ b (a − b ∈ I) and c ≡ d (c − d ∈ I), then ac ≡ bd
(ac − bd ∈ I). Let b = a + i1 and d = c + i2 . Then

ac − bd = ac − (a + i1 )(c + i2 ) = ac − ac − i1 c − i2 a − i1 i2 = −( i1 c + i2 a + i1 i2 ).
|{z} |{z} |{z}
∈I ∈I ∈I

be 0 by taking the smallest ideal I ⊇ S.

If S is any subset of a ring R, we can force S to P
In this case, I is the set of finite sums of the form si ∈S ri si ti with r1 , ti ∈ R.

9.5 Generators and relations

We form a free ring on a set S. We have 2 choices:
16
The Dirac δ distribution is actually an identity for convolution for a larger ring than this.

41
 1. Free commutative ring: First form the free commutative monoid on S. If S =
{x,
 y, z},a then this is {xn1 y n2 z n3 : ni ∈ N} The free commutative ring is the ring
na,b,c x y b z c : a, b, c ≥ 0 .
Say we have the elliptic curve y 2 = x3 − x. We can form the coordinate ring
Z[x, y]/(y 2 −x3 +x), where we are quotienting out by the ideal generated by y 2 −x3 +x.

2. Noncommutative free ring: Take the noncommutative free monoid on {x, y, z}. This
is all words in {x, y, z}. The noncommutative free ring is the group ring of the free
monoid.
Now we can construct rings such as Z[x, y, z]/(x2 + y 2 z − zy 2 ) (some ideal generated
by some elements), which is noncommutative.

Example 9.8. Suppose A and B are rings. We can construct the coproduct as follows:
assume A ∩ B = ∅, and form the free ring F on the set A ∪ B. Quotient out by an ideal
to force the map from A → F to be a homomorphism; we have I = (f (a + b) − f (a) −
f (b), f (ab) − f (a)f (b) ∀a, b, ∈ R) and so on (including all the relations we want). Then F/I
is the coproduct of A and B.

Example 9.9. The coproduct of Z[x] and Z[y] in the category of rings is the free non-
commutative ring on x, y. However, the coproduct of Z[x] and Z[y] in the category of
commutative rings is the polynomial ring Z[x, y].

42
10 Euclidean Domains, Principal Ideal Domains, and Unique
Factorization Domains
10.1 Euclidean Domains and Principal Ideal Domains
10.1.1 Euclidean Domains
Recall that every integer 6= 0 is a product of primes in an essentially unique way. 12 =
2 × 2 × 3 = 2 × 3 × 2 = (−2) × (−3) × 2. So the product is unique up to order and
multiplication by units.
This was essentially proved by Euclid. The key point he used was division with a
remainder. That is, given a, b with a 6= 0, we can write a = bq + r, where r is smaller than
b. Here, q is called the quotient, and r is the remainder.
What does smaller mean in this context? For integers, this means |r| < |b|. We can do
the same thing for polynomials a, b ∈ R[x]; a smaller than b means that deg(a) < deg(b)
(or a = 0).

Definition 10.1. A commutative ring R is a Euclidean domain if it has a function |·| :

R → N such that given a, b with b 6= 0, we can find r, q such that a = bq + r and |r| < |b|.17

Example 10.1. Let Z[i] = a + bi : a, b ∈ Z, i2 = −1 be the Gaussian integers. Z[i] is a



Euclidean domain. Define |a + bi| = a2 + b2 . This is the usual Euclidean norm but squared
to make sure we get an integer. Given a, b, we need to find r, q such that a = bq + r, which
means a/b = q + r/b, where |r/b| < 1. Given any a/b, we can find q ∈ Z[i] of distance < 1
from a/b. Draw an open disk of radius 1 around each elements of Z[i]. These cover C, so
we can find r, q.

10.1.2 Principal Ideal Domains

Definition 10.2. The ideal generated by elements g1 , g2 , . . . is the smallest ideal containing
these elements.

We denote (a, b, c, . . . ) as the ideal generated by a, b, c . . . .

Definition 10.3. A principal ideal domain is a commutative ring where all ideals are
generated by one element.

Example 10.2. Z is a principal ideal domain. In Z, we only have ideals of the form nZ.

Example 10.3. Here is an example of a commutative ring that is not a PID. Let R =
C[x, y], and let I = (x, y) be the set of all polynomials with constant term 0. If I = (f ),
then f divides x and f divides y. This means f = 1, but 1 ∈ / (x, y).
17
We don’t actually need the codomain of the norm function to be N; we just need it to be a well-ordered
set. In practice, however, the useful examples are all with sets that are basically N.

43
Theorem 10.1. Euclidean domains are principal ideal domains.

Proof. Let I be any ideal. Choose a ∈ I with a 6= 0 and |a| minimal. Then we claim that
I = (a). Suppose b ∈ I. Then b = aq + r with |r| < |a|. So r = b − aq means that r ∈ I,
and the minimality of |a| forces r = 0. So b = aq for some q, and this holds for any b ∈ I,
so I = (a).
√
Example 10.4. R = Z[(1 + −19)/2] is a PID that is not Euclidean. R is a PID; for
proof, see an algebraic number theory course. Here is a sketch that R is not Euclidean.
Let a ∈ R be nonzero and not a unit, with |a| minimal. Then look at R/(a). If b ∈ R,
b = aq + r with |r| < |a|. Then r is 0 or a unit. So every element of R/(a) is represented
by 0 or a unit. The only units of R are ±1, so R/(a) has ≤ 3 elements. If a 6= ±1, 0, then
R/(a) has ≥ 4 elements (actually |a|2 ).

10.2 Unique factorization domains

10.2.1 Definitions and relationship to principal ideal domains
Definition 10.4. Let a, b ∈ R. We say a divides b (denoted a|b) if there exists some c ∈ R
such that ac = b.

Definition 10.5. An element a is called irreducible if a 6= 0, a is not a unit, and a = bc

implies that either b or c is a unit.

Definition 10.6. An element a is called prime if a|bc implies that a|b or a|c.

For Z, these two definitions are equivalent, but this is not the case in all rings.

Lemma 10.1. In a principal ideal domain, irreducible elements are prime.

Proof. Suppose p is irreducible and p|ab. We want to show that p|a or p|b. Suppose that
p 6 |a. Then (p, a) = (c) since R is a principal ideal domain. Then c|p, so c is a unit or
is a unit times p. The second case is not possible because pu = c divides a, but a is not
divisible by p. So (c) contains 1 (by multiplying c by c−1 ) and is then equal to R. So
(p, a) = (1) = R.
We now have px + ay = 1 for some x, y ∈ R, which makes pbx + aby = b. Both terms
are divisible by p, so p|b. Hence, p is prime.

Definition 10.7. A unique factorization domain is a commutative ring in which every

element can be uniquely expressed as a product of irreducible elements, up to order and
multiplication by units.

Theorem 10.2. Every principal ideal domain is a unique factorization domain.

44
Proof. We first show existence of factorization into irreducibles. Given a ∈ R, first find
irreducible p dividing a if a is not a unit. Let a = bc; if b is irreducible, stop. Otherwise, let
b = de, and repeat the process until we get an irreducible element. Can this go on forever?
No. Suppose we have a, b, c, d, e, . . . with a = b0 b, b = c0 c, etc., where b0 , c0 , . . . are not
units. Then the ideal (a, b, c, d, . . . ) = (x), since we are in a PID. But then x ∈ (a, b, c, d, e)
(some finite sequence of the variables), so the sequence must stop after finitely many steps.
Now put a = bc with b irreducible, c = de where d is irreducible, e = f g, where f is
irreducible and so on. This stops after a finite number of steps by a similar argument. So
every nonzero element is a product of irreducibles.18
To prove uniqueness, suppose a = p1 · · · pm = q1 · · · qn with pi , qj irreducible. We want
to show that these factorizations are unique up to order and units. p1 is irreducible, so p1
divides some qi as p1 is prime. The qi are irreducible, so qi = p1 u for some unit u ∈ R. By
removing p1 and this qi from their respective sides (really we are bringing the two products
to the same side, factoring out the p1 , and asserting that the rest equals 0), we can repeat
this to eventually get our result.

Example 10.5. R be the set of polynomials in xq for rational q > 0; this is a set of terms
of elements like 3 + 3x5/7 + 2x17/3 . This argument goes wrong here because x = x1/2 x1/2 =
x1/4 x1/4 x1/4 x1/4 = · · · . The ideal (x1/2 , x1/4 , x1/8 , . . . ) is not principal.

10.2.2 Examples and Applications

Example 10.6. Suppose a + bi ∈ Z[i] is prime. Then (a + bi)(a − bi) = a2 + b2 ∈ Z. So
we can use this to factor elements in Z into elements in Z[i]. For example, 5 = 22 + 1 =
(2 + i)(2 − i).

65 = 5 × 13 = (2 + i)(2 − i)(3 + 2i)(3 − 2i) = (4 + 7i)(4 − 7i) = (8 − i)(8 + i)

This gives us 65 = 42 +72 = 82 +12 . So the different factorizations of x ∈ Z in the Gaussian

integers give us the ways to write x as a sum of two squares.
√
Example 10.7. Let R = Z[ −2]. Imagine this as a rectangular lattice in C. The √ circles
of radius 1 around these points cover C, so as we argued before with Z[i], Z[ −2] is a
euclidean domain and√ hence is a unique factorization domain. √
Now let R = Z[ −3]. The circles of radius 1 do not cover the point √ 1/2 + √−3/2. In
fact, R is not a unique factorization domain. We have 2 × 2 = (1 + 3i)(1 − 3i), and
the only units are ±1. These are all irreducible elements. If 2 = ab, then |a| |b| = |2| = 2,
which means |a| = ±1 or |b| = ±1.
Multiplying
√ z ∈ R by a multiplies |z| by |a| and rotates z by arg(a). So a principal ideal
in Z[ −3] looks like a rotated and rescaled rectangular lattice. What does a non-principal
18
This is still true if R has the following property: there is no strictly increasing sequence of ideals
I1 ( I2 ( I3 ( · · · . These are called Noetherian rings.

45
 √
ideal look like? Look at (2, 1 + −3); we get a “diamond” lattice instead of a rectangular
one.

Unique factorization domains need not be principal ideal domains.

Example 10.8. Z[x] is a UFD and has the non-principal ideal (2, x).

Example 10.9. Let K be a field. K[x, y] is a UFD and has the non-principal ideal (x, y).

We will see later that if R is a UFD, then so is R[x], the ring of polynomials over R.

Theorem 10.3 (Fermat). Any prime p ∈ Z with p > 0 and p ≡ 1 (mod 4) can be uniquely
expressed as a2 + b2 (up to sign differences in a, b).

Proof. (Z/pZ)∗ is cyclic of order p − 1 = 4n. It has an element −1 of order 2. Let g be

a generator, so g 4n = 1. So −1 ≡ g 2n (mod p), which means that −1 is a square mod
p. This gives us that −1 = a2 − np for some n, a. So np = a2 + 1 = (a + i)(a − i) in
Z[i]. p|(a + i)(a − i), but does not divide either of these two factors, so p is not prime and
hence is not irreducible in Z[i]. So p = (a + bi)(a − bi) for some a, b ∈ Z (we must have
this decomposition because a + bi times any other number would not be purely real). This
makes p = a2 + b2 .
For uniqueness, suppose that p = x2 + y 2 . Then p = (x + iy)(x − iy), which means
x + iy = u(a + bi) for some unit u because Z[i] is a unique factorization domain. Then
x = ±1 and b = ±b.

46
11 Prime Ideals and Maximal Ideals
11.1 Fields and integral domains
Definition 11.1. A field is a commutative ring where all nonzero elements have multi-
plicative inverses.
Definition 11.2. An integral domain is a ring where ab = 0 implies that a = 0 or b = 0.
Proposition 11.1. All fields are integral domains.
Proof. Let R be a field. Then for a, b ∈ R,

ab = 0 =⇒ a−1 ab = a−1 0 =⇒ b = 0.

Definition 11.3. Let I be an ideal of R. I is called maximal if R/I is a field.

Definition 11.4. Let I be an ideal of R. I is called prime if R/I is an integral domain.
Equivalently, I is prime if ab ∈ I implies that a ∈ I or b ∈ I.
Why are these definitions equivalent?

R/I is an integral domain ⇐⇒ [(a + I)(b + I) = I =⇒ a ∈ I or b ∈ I]

⇐⇒ [ab + I = I =⇒ a ∈ I or b ∈ I]
⇐⇒ [ab ∈ I =⇒ a ∈ I or b ∈ I].

We can see by the previous proposition that all maximal ideals are prime.
Definition 11.5. An ideal I 6= R is maximal if for any ideal J, I ⊆ J implies that I = J
or J = R.
Proposition 11.2. Let I be an ideal of a ring R. Then R/I is a field iff I is maximal.
Proof. Suppose I is maximal. Since I 6= R, 1 ∈/ I, so R/I contains an element 1 + I 6= I.
Letting x + I ∈ R/I, note that I + Ax = R, so there exists some y ∈ I and a ∈ R such
that y + ax = 1. Then ax + I = 1 + I, so (a + I) is the inverse of x + I in R/I. So R/I is
a field.
Conversely, suppose R/I is a field. Then for x ∈/ I, there exists some a ∈
/ I such that
ax + I = 1 + I. Then ax + y = 1 for some y ∈ I, so (1) ⊆ Ax + I, which makes Ax + I = R.
This holds for all x ∈
/ I, so I is maximal.

Example 11.1. Let R = Z. The ideals are of the form (n) for n = 0, 1, 2, 3, . . . . The
maximal ideals are (2), (3), (5), (7), . . . . The prime ideals are (0), (2), (3), (5), (7), . . . .
Example 11.2. Let R = C[x]; this is a PID. The ideals are (f ) for a polynomial f . The
maximal ideals are (x − a) for a ∈ C (any polynomial f of degree > 1 factorizes as f = gh,
so (f ) ( (g), making (f ) not maximal). The prime ideals are (x − a) for a ∈ C, and (0).

47
Example 11.3. Let R = C[x, y]. The ideal (x, y) is maximal because R/(x, y) = C, which
is a field. The ideals (x − a, y − b) are also maximal. These are the only maximal ideals.19
The prime ideals are (x − a, y − b), (0), and (f ) if f is any irreducible polynomial; this is
because C[x, y]/(f ) is an integral domain because C[x, y] is a UFD.

11.2 Maximal ideals and Zorn’s lemma

Definition 11.6. A partial order is a relation ≤ on a set S such that for all a, b, c ∈ S

1. a ≤ a (reflexivity).

2. If a ≤ b and b ≤ a, then a = b (antisymmetry).

3. If a ≤ b and b ≤ c, then a ≤ c (transitivity).

Example 11.4. Let S be the set of subsets of some set T . The ordering ≤ is inclusion.

Definition 11.7. Let S be a partially ordered set. A totally ordered subset T of S is a

subset such that for all a, b ∈ T , a ≤ b or b ≤ a.

Definition 11.8. Let S be a partially ordered set. An upper bound of a subset T is an

element a ∈ S such that b ≤ a for all b ∈ T .

Definition 11.9. Let S be a partially ordered set. An element a ∈ S is maximal 20 if a ≤ b

implies that b = a.

Lemma 11.1 (Zorn). Suppose S is a nonempty partially ordered set such that for any
totally ordered subset of S, there is an upper bound. Then S has a maximal element.

Proof. We will sketch a proof because a full proof requires some set theory. Suppose no
maximal element exists; we will find a contradiction.
Step 1: Pick s0 ∈ S since S is nonempty. Then {s0 } is totally ordered, so it has an
upper bound s1 . If s0 is not maximal, then s1 > s0 .
Step 2: Repeat this with {s0 , s1 }, which is totally ordered. And repeat this.
Step 3: We do this infinitely many times21 , and find sω , which is an upper bound of
{s0 , s1 , s2 , . . .}.
Step 4. We find an sα for every ordinal α. But the set of ordinals is a proper class, so
it must be bigger than S since S is a set. So we have a contradiction.

Corollary 11.1. If I is an ideal of R with I 6= R, I is contained in some maximal ideal.

19
See Hilbert’s Nullstellensatz. This word means zero position theorem.
20
You might think that maximal should mean that b ≤ a for all b ∈ S, but this is a very strong condition.
This implies a unique maximal element, which is not true for our definition of maximality.
21
Picking elements in this way requires the axiom of choice. As such, Zorn’s lemma was somewhat
controversial in the early 20th century.

48
Proof. Look at the set S of ideals 6= R containing I. It is partially ordered by ⊆ and is
nonempty
S because it contains I. Now suppose Iα is a totally ordered set of ideals; then
α Iα is an idealS
and is greater than Iα for each α. Why is this an ideal? The total ordering
is key. If a, b ∈ α Iα , then a ∈ Iα1 and b ∈ Iα2 ; without loss of generality, Iα1 ⊆ Iα2 , so
a + b ∈ Iα2 . This is the upper bound needed to satisfy the conditions of Zorn’s lemma.

Remark 11.1. You may be wondering why we need Zorn’s lemma. In general, there
exist nonempty ordered sets with no maximal elements. For example, take the open unit
interval, (0, 1).22

Corollary 11.2. The intersection of all prime ideals of a ring is the set of elements x with
xn = 0 for some n (called nilpotent).

Proof. Let p be a prime ideal. If xn = 0, then xn−1 x = xn = 0 ∈ p, so since p is prime,

xn−1 ∈ p or x ∈ p, and so on, so x ∈ p.
Suppose x is not nilpotent; we need to find a prime ideal P not containing x. Let
2

M = 1, x, x , . . . , which doesn’t contain 0 because x is not nilpotent. Let S be the set of
S is nonempty because (0) ∈ S.
ideals disjoint from M . S is partially ordered by inclusion. S
Any totally ordered subset {Iα } of S has an upper bound α Iα . So, by Zorn’s lemma, S
has a maximal element I; I is maximal in S, not a maximal ideal.
I is prime. Suppose a, b ∈/ S. Then (I, a) > I, so it contains an element of M xn = i1 +
sa. Likewise, (I, b) contains an element of M xn = i2 +tb. So i1 i2 +i2 sa+i1 tb+stab = xm+n
is an element of M , and the first 3 terms on the left hand side are in I. So ab ∈/ I because
otherwise the right hand side of this equation would be an element of I, which is impossible
because it is in M . So I is prime, as desired.

12 Localization
12.1 What is localization?
The integers do not have division. This is inconvenient, so we construct the rational
numbers Q = {m/n : m, n ∈ Z, n 6= 0}. Q is a field.
More generally, suppose R is a ring and S is a subset of R. We find a new ring R[S −1 ]
so that all elements of S have inverses. This is localization.

Example 12.1. If R is an integral domain and S is the set of nonzero elements of R, then
R[S −1 ] is a quotient field of R.
22
Assuming that ordered sets always have a maximal element has been the cause of numerous philosophical
blunders over the years, such as some attempted proofs of the existence of a god.

49
12.2 Construction
We may as well assume 1 ∈ S and S is closed under multiplication. If a, b have inverses,
then ab should, as well. First, Assume S has no zero divisors. We basically copy the
construction of Q from Z.
Take all pairs (r, s) with r ∈ R and s ∈ S. Call this r/s. We have an equivalence
relation r1 /s1 ≡ r2 /s2 means r1 s2 = r2 s1 . The subtle point of this construction is that we
need to check that this equivalence relation is transitive.
We first assume that S has no zero divisors. Suppose r1 /s1 ≡ r2 /s2 and r2 /s2 ≡ r3 /s3 .
We have r1 s2 = r2 s1 and r2 s3 = r3 s2 . So r1 s2 s3 = r2 s1 s3 = s1 r3 s2 . This makes s2 (r1 s3 =
r3 s1 ) = 0, and since s2 is not a zero divisor, r1 s3 = r3 s1 ; i.e. r1 /s1 ≡ r3 /s3 . The remaining
step is to check that the equivalence classes form a ring. We leave this as an exercise.
In this case, we have the map R → R[S −1 ] sending r 7→ r/1. This map is injective
because it has trivial kernel; r/1 = 0/1 means 1r = 0 · 1 = 0, which makes r = 0.
What if S has zero divisors? Then r1 /s1 ≡ r2 /s2 is not an equivalence relation. So let
I be the ideal of all elements with xs = 0 for some s ∈ S. Check that this is an ideal. Now
form R/I, and let S̄ be the image of S in R/I. Then S̄ has no zero divisors in R/I, so we
can form (R/I)[S̄ −1 ] as before.
So we get a ring R[S −1 ] with the following properties:

1. There is a homomorphism from R → R[S −1 ].

2. The images of all elements of S are invertible in R[S −1 ].

3. R[S −1 ] is the universal ring with these properties.

R R[S −1 ]s

The kernel of the map R → R[S −1 ] is I, the set of elements killed by something in S.
Then r1 /s1 ≡ r2 /s2 can be defined as ∃s3 such that s3 (r1 s2 − r2 s1 ) = 0.

12.3 Examples
Why is localization called localization?

Example 12.2. Let R = C[x], the set of polynomial functions on C. Suppose we want
to examine 0 ∈ C. What do the functions near 0 look like? An example is the rational
functions that are nonsingular at 0; this is an approximation to all holomorphic functions
in a neighborhood of 0. This is equal to R[S −1 ], where S is the set of polynomials that are
nonzero at 0. The map R → R[S −1 ] is injective but not surjective.

50
Example 12.3. Let R be the set of continuous functions on R. Focus on the point 0 ∈ R.
Look at the germs, functions that are equivalent in a neighborhood of 0. The ring of germs
is R[S −1 ], where S is the set of functions that are nonzero at 0. Here, the map R → R[S −1 ]
is surjective but not injective.

You may have noticed that in these two examples, S was the complement of a prime
ideal. In general, if p is any prime ideal, then the complement of p is multiplicatively
closed.

Example 12.4. Let R = Z, and suppose we are interested in (2). Let S = Z \ (2), the odd
numbers. So we get a ring Z(2) , the rationals a/b with b odd. In general, let Rp = R[S −1 ],
where S is the complement of a prime ideal p. The units of Z(2) are rationals of the form
a, b with a, b odd. 2 is a prime element of Z(2) . Anly element of Z(2) equals 2n u for some
unit u and a unique n ∈ N. So this is a UFD with only one prime: 2. We see that localizing
at 2 “kills off” all primes other than 2.

51
13 Modules
13.1 Basic notions and examples
13.1.1 Modules and homomorphisms
Informally, a module M over a ring R is like a vector space but over a ring.

Definition 13.1. A (left) module M over a ring R is an abelian group with a map R×M →
M sending (r, m) 7→ r · m such that for r, s ∈ R and x, y ∈ M

1. r · (x + y) = r · x + r · y.

2. (r + s) · x = r · x + s · x

3. (rs) · x = r · (s · x)

4. 1R · x = x (if R has 1).

A right module is the same thing, except the map is M × R → R, so the actions of R on
M is on the right.

Definition 13.2. Let M be an R-module. A submodule N is a subgroup of M such that

r · n ∈ N for each r ∈ R and n ∈ N .

Definition 13.3. A homomorphism of modules M1 , M2 is a map f : M1 → M2 such that

1. f (m1 + m2 ) = f (m1 ) + f (m2 )

2. f (r · m) = r · f (m).

Better (but not standard) notation would be that homomorphisms of left modules
should be written on the right (and vice versa for right modules). So we should write mf ,
not f m. This makes it so the second condition gives us that (rm)f = r(mf ), which gets
rid of the needless switching of the order of r and f . We will alternate between the two
notations.

Definition 13.4. Let M, N be modules over R. Then HomR (M, N ) is the set of module
homomorphisms from M to N .

If R is commutative, HomR (M, N ) is an R-module.

Definition 13.5. An endomorphism of M is a homomorphism from M to itself.

Definition 13.6. A bimodule is a left module over one ring and a right module over
another, where the left and right actions commute.

Example 13.1. R is an (R, R) bimodule.

52
13.1.2 Exact sequences of modules
Suppose we have the exact sequence

0 → A → B → C → 0.

Are the following two sequences exact?

0 → Hom(M, A) → Hom(M, B) → Hom(M, C) → 0

0 ← Hom(A, N ) ← Hom(B, N ) ← Hom(C, N ) ← 0

The answer is no.23 Look at
×2
0 → Z −−→ Z → Z/2Z → 0.

Then
×2
0 → Hom(Z/2Z, Z) −−→ Hom(Z/2Z, Z) → Hom(Z/2Z, Z/2Z) 
→0

| {z } | {z } | {z }
=0 =0 =Z/2Z

×2
0←
  Hom(Z, Z/2Z) ←
 −− Hom(Z, Z/2Z) ← Hom(Z/2Z, Z/2Z) ← 0.
Instead, we get exact sequences

0 → Hom(M, A) → Hom(M, B) → Hom(M, C)

Hom(A, N ) ← Hom(B, N ) ← Hom(C, N ) ← 0.

We leave this as an exercise.

13.1.3 Examples of modules

Example 13.2. Vector spaces over fields are modules.

Example 13.3. Abelian groups are modules over Z.

Example 13.4. Left ideals of R are the same as left submodules of a module R.

Example 13.5. Let G be a group acting on a set S. Form the vector space V over K
with basis S, and form the group ring K[G]. G acts on V by acting on the basis elements.
So V is a module over the ring K[G].24
23
The study of homological algebra is based on the fact that these sequences are not always exact in this
way.
24
The study of these modules is very important in representation theory.

53
Example 13.6. Suppose M is a left module over a ring R. Then HomR (M, M ), the
endomorphisms of M , is a ring, where the product is composition of endomorphisms.
M is a right module over HomR (M, M ). Furthermore, the right action of HomR (M, M )
commutes with the left action of R on M (follows from the definition of a homomorphism).
So M is a HomR (M, M ) bimodule.

HomR (M, M ) is analogous to the permutations of a set S. If we have a group, we can

represent it as the permutations of the set S. Similarly, a ring is often studied as a subring
of HomT (M, M ) for some T -module M .

Example 13.7. Take an algebraic number field such as Q[i], where i2 = −1. Think of Q[i]
as a vector space over Q, and think of the ring Q[i] as endomorphisms of this vector space.
So we can represent elements of Q[i] as matrices. Matrices are linear transformations of
vector spaces or equivalently homomorphisms of modules.
Pick a basis of Q[i]: {1, i}. The action of 1 is 1 → 1 and i → i and the action of u is
1 → i and i → −1. So we have the matrices
   
1 0 0 1
1= , i= .
0 1 −1 0

So Q[i] can be thought of as the matrices

 
a b
−b a

with a, b ∈ Q.
Look at the invariants of matrices, the trace and the determinant. Here, tr(a+bi) = 2a,
and det(a + bi) = |a + bi|.

13.2 Free modules

L
Definition 13.7. The direct sum of modules Mα over R is the abelian group Mα with
the action of R on each component α determined by the action of R on Mα .

Definition 13.8. A free module is a module that is a direct sum of copies of R.

In some sense, free modules are the simplest sort of module.

Example 13.8. Any vector space is a free module.

Example 13.9. Z is a free module over Z. However, Z/2Z is not free.

We want to define the rank of a free module as the number of copies of R in the sum.
Is this well defined? We must check that if Rm ∼= Rn , then m = n. However this is not
always true. When is this true?

54
 • This is true when R is a field.
• This is false if R is the 0 ring.
• This is true if R is commutative with R 6= 0.
Pick a maximal ideal I in R and suppose Rm ∼= Rn . Reduce mod I, so (R/I)m ∼
=
n
(R/I) as modules over a field R/I. So m = n because R/I is a field.
• This is sometimes true if R is not commutatative (see below).
• There exist rings R 6= 0 such that R ∼
= R ⊕ R as R modules (see below).
Example 13.10. Take R = Mn (K), the n × n matrices over a field K, and suppose
Ra ∼
= Rb . These are vector spaces of dimension an2 and bn2 , respectively, so a = b.
Example 13.11. Here is an example of a ring R 6= 0 such that R ∼ = R ⊕ R as R modules.
This is a possibly unsettling result. Homomorphisms from R to Rn can be identified with
m

m × n matrices, as in linear algebra. If R ∼

= R ⊕ R, we have a 1 × 2 invertible matrix!
Pick an abelian group A such that A ∼= A⊕A, such as Z⊕Z⊕Z⊕· · · . Put R = End(A);
in our example, this is the set of ∞ × ∞ matrices with only finitely many nonzero entries
in each row. Then R = Hom(A, A) = Hom(A, A ⊕ A) = R ⊕ R.
So the rank of a free R-module is not necessarily well-defined.

13.3 Projective modules

Given a free module M , we can recover the underlying set SM ; this is via a forgetful
functor F from the category of modules to the category of sets. Likewise, given a set S, we
can form the free module MS with basis S; this is also via a functor, F 0 . These functors
commute with morphisms in the following way:
F
M SM
f F (f )

N SN
F0

We say that the functors F and F 0 are adjoint. As a consequence, free modules are
projective.
Definition 13.9. A projective module P is a module with the following property. If the
sequence M → N → 0 is exact, then any map P → N lifts to a map P → M .

M N 0

55
Proposition 13.1. The following are equivalent:

1. P is projective.

2. P ⊕ Q is free for some module Q.

Proof. (1) =⇒ (2) : Pick a free module F so ϕ : F → P is onto. Then F → P → 0, so

we can find a map P → F .
ϕ
F P 0
id

P
But then F splits as P ⊕ ker(ϕ).
(2) =⇒ (1): Exercise.

Example 13.12. R = Z/6Z = Z/2Z ⊕ Z/3Z, so Z/2Z and Z/3/Z are projective over
Z/6Z but not free.

Example 13.13. Let R be the ring of continuous functions on a circle S 1 , and let M = R.
Then we can think of M as continuous functions S 1 → S 1 ×R. M is sections of S 1 ×R → S 1 ,
which equals the real valued functions on S 1 . This is a vector bundle 25 over S.
Consider a Möbius band, and view it as a vector bundle over S 1 , so each fiber is
isomorphic to R. Now define a module N to be the sections of this twisted vector bundle.
Then N is projective but not free.
N is not free because the orientations of the fibers change as you go around S 1 . It is
projective because N ⊕ N = M ⊕ M . At each point of S 1 , consider the normal bundle.
Now take the orthogonal complement. So we get 2 Möbius bands so at each point, and
their fibers intersect at every point. So we can think of N ⊕ N as the sum of 2 Möbius
bands.
In effect, we can think of projective modules as “twisted free modules.”
√
Example 13.14. √ Let R = Z[ −5]; we can think of this as a rectangular lattice in C. Let
M = (2, 1 + −5). The principal ideals here are rectangular with respect to this lattice
picture. Non-principal ideals are diamond shaped. Principal ideals here are free modules,
and nonprincipal ideals are not free.
We want to show that M is projective, and we do so by showing that M = R ⊕ R.
onto √
We map g : R ⊕ R −−→ M by sending (1, 0) 7→ 2 and (0, 1) 7→ 1 + −5. We want to
construct a section√f : M → R ⊕ R, where g(f (m)) = m. So R ⊕ R = M ⊕ ker(g). Let
f (x) = (−x, x(1 + −5)/2), and check that f (x) ∈ R ⊕ R. So M is projective.

25
We won’t be going over vector bundles in detail in this course. If you don’t know what a vector bundle
is, see a topology course.

56
14 More on Projective Modules
14.1 Projective modules as direct sums
Recall that P is a projective module if it satisfies the following commutative diagram for
exact sequences of modules M and N :
M N 0

P
We also showed that P projective iff P ⊕ Q is free for some Q. Are all submodules of
free modules projective? The answer is no.
Example 14.1. Here is a non-projective submodule of a free module. Let R = K[x, y],
where K is a field, and let I = (x, y), the ideal of polynomials with constant term 0. Look
g
at R ⊕ R → − I → 0 where (1, 0) 7→ x and (0, 1) 7→ y. If I is projective, then there exists
some map f : I → R ⊕ R such that gf (a) = a. Now suppose that f (x) = (a, b) and
f (y) = (c, d). Then ax + by = x and cx + dy = y. Then y(a, b) = x(c, d)), so ya = xc
and yb = xd. There are no polynomials satisfying this because ax + by = x implies that
a = 1 + yp (where p is a polynomial), and ya = xc implies that a cannot be 1 + yp.

14.2 Eilenberg-Mazur swindle

This is a technique useful for proving 1 = 0. Here is a basic example.
Example 14.2. Start with 1 + (−1) = 0. Then
0 = (1 + (−1)) + (1 + (−1)) + · · ·
1 = 1 + (−1 + 1) + (−1 + 1) + · · · ,
so we have shown that 1 = 0.
We assumed two things in the above example:
1. 1 has an additive inverse −1.
2. All infinite sums make sense.
The second condition is violated in Z, but we can use this technique to show that one of
these two conditions does not hold.
Example 14.3. Knots have no inverse. Suppose we have a closed loop with a knot in
it. Is there another knot we can put on the loop that will cancel out the first knot? The
answer is no. Apply the swindle: add infinite numbers of knots, making each successive
knot smaller so the knots all fit on the loop. Then the above contradiction would occur,
so a knot must not have an additive inverse.

57
Example 14.4. Suppose P is projective. Then P ⊕ Q = F , where F is free. Then Q is
also projective. We can take Q to be free (in fact equal to F ). Think of free modules as 0
in some sense. So P ⊕ Q is free means that Q is a sort of additive inverse of P (again, if
we ignore free modules). So infinite sums are defined, and we can use the swindle to get
that P = 0 if we ignore free modules. What we mean here is that P ⊕ Q is free for some
free module Q. The catch is that this free module Q is not finitely generated.

15 Tensor Products
This is covered in Chapter XVI in Lang, but we will cover it here. This is something you
really should know.

15.1 Construction and universal property

Definition 15.1. A bilinear map f : X × Y → Z is a map such that f (·, y) is linear for
fixed y and f (x, ·) is linear for fixed x.

Definition 15.2. Suppose R is a commutative26 ring, and suppose that M and N are
R-modules. The tensor product M ⊗ N is the module such that if f : M × N → P is
bilinear, then there exists a linear map f˜ : M ⊗ N → P such that the following diagram
commutes
ϕ
M ×N M ⊗N
f
f˜

P
To construct M ⊗ N , take the free module on elements m ⊗ n with m ∈ M and n ∈ N .
We get linear maps from this to P : m ⊗ n 7→ f (m, n). Take the quotient by all elements
of the form
(m1 + m2 ) ⊗ n − m1 ⊗ n − m2 ⊗ n
m ⊗ (n1 + n2 ) − m ⊗ n1 − m ⊗ n2
(rm) ⊗ n − r(m ⊗ n)
m ⊗ (rn) − r(m ⊗ n).
Taking the quotient by these elements enforces relations we want, such as

(rm) ⊗ n = r(m ⊗ n) = m ⊗ (rn),

so the tensor product exists.

26
This assumption is not necessary, but it simplifies things for now.

58
 Now that we have constructed the tensor product, what does it look like? We have the
identity
(M1 ⊕ M2 ) ⊗ N ∼= (M1 ⊗ N ) ⊕ (M2 ⊗ N ),
which says that a bilinear map (M1 ⊕ M2 ) ⊗ N → P is the same as a pair of bilinear maps
from (M1 ⊗ N ) → P and (M2 ⊗ N ) → P . Similarly, we have the identity
R⊗M ∼
= M,
which says that bilinear maps R × M → P are the same as linear maps from M → P .
Example 15.1.
Rm ⊗ Rn ∼
= Rm+n
If V, W are vector spaces with bases {vi } and {wj }, then V ⊗ W has basis vi ⊗ wj .

15.2 Exact sequences and the tensor product

Proposition 15.1. Suppose A → B → C → 0 is exact. Then so is
A ⊗ M → B ⊗ M → C ⊗ M → 0.
Remark 15.1. This does not hold if we put a 0 → before both of these sequences. We
say that ⊗M is right exact.
Proof. To prove things about the tensor product, forget the construction of the tensor
product using relations and instead use the universal property.
Homomorphisms A ⊗ B → C are bilinear maps A × B → C, which are linear maps
A → HomR (B, C). Think of this as an analogue of the fact that functions R × S → T are
the same as functions from R to the set of functions from S to T .
The key point of this proof is that A → B → C → 0 is exact if and only if
Hom(A, M ) ← Hom(B, M ) ← Hom(C, M ) ← 0
is exact. We leave this as an exercise.27
We want to show that A ⊗ N → B ⊗ N → C ⊗ N → 0 is exact. Then this is equivalent
to the following sequence being exact:
Hom(A ⊗ N, M ) ← Hom(B ⊗ N, M ) ← Hom(C ⊗ N, M ) ← 0.
Then, using our identification of homomorphisms A ⊗ N → M with linear maps A →
HomR (N, M ), this is equivalent to the following sequence being exact:
Hom(A, Hom(N, M )) ← Hom(B, Hom(N, M )) ← Hom(C, Hom(N, M )) ← 0.
And this is exact by applying the key point again.
27
This was an exercise from last lecture, but Professor Borcherds suspects that no one actually does
them.

59
 We can now calculate M ⊗ N . Pick Ra → Rb → M → 0, where Ra , Rb are free. Pick
relations generating ker(Rb → M ) and pick a set of b generators of M . Tensoring with N
gives us that
Ra ⊗ N → Rb ⊗ N → M ⊗ N → 0
is exact. So we get
N a → N b → M ⊗ N → 0,
which makes M ⊗ N = N b / im(N a → N b ).

Example 15.2. We can find M ⊗ N for finitely generated abelian groups M, N . Recall
that finitely generated abelian groups are direct sums of copies of Z and Z/nZ. Since
(A ⊕ B) ⊗ C = (A ⊗ C) ⊕ (B ⊗ C), it is enough to work out a few cases:

1. Z ⊗ Z = Z

2. Z ⊗ Z/mZ = Z.mZ

3. Z/mZ ⊗ Z = Z/mZ

4. Z/nZ ⊗ Z/mZ = Z/(gcd(m, n)Z).

To obtain this last result, take the exact sequence

×m
Z −−→ Z → Z/mZ → 0.

Then the sequence

×m
Z/nZ −−→ Z/nZ → (Z/mZ ⊗ Z/nZ) → 0
is exact, so
Z/mZ ⊗ Z/nZ ∼
= (Z/nZ)/m(Z/nZ) ∼
= Z/(gcd(m, n)Z).

Example 15.3.
Z/2Z ⊗ Z/2Z = Z/2Z
Z/2Z ⊗ Z/3Z = 0
Z/9Z ⊗ Z/12Z = Z/3Z
×2
The tensor product is not left exact. Look at 0 → Z −−→ Z → Z/2Z → 0. The following
sequence is not exact:
×2
0 → Z/2Z −−→ Z/2Z → Z/2Z → 0.

60
15.3 More examples and properties
Definition 15.3. An algebra S over a ring R is a commutative ring with a homomorphism
R → S that makes S an R-module.

You can think of algebras as modules with multiplication.

Example 15.4. Let S, T be algebras over R. Then S ⊗R T is a push-out of S, T over R.

R S

T S ⊗R T

Check that S ⊗R T is a commutative ring. We need a bilinear map (S ⊗ T ) × (S ⊗ T ) →

(S ⊗ T ). This is a linear map from S ⊗ T ⊗ S ⊗ T → S ⊗ T . This relies on associativity of
the tensor product; (A ⊗ B) ⊗ C ∼ = A ⊗ (B ⊗ C) because maps from each to M are trilinear
maps A × B × C → M . We have a map S ⊗ S → S given by the product on S. Same
for T ⊗ T → T . So we get a map S ⊗ T ⊗ S ⊗ T → S ⊗ S ⊗ T ⊗ T → S ⊗ T by sending
(s1 ⊗ t1 ) × (s2 ⊗ t2 ) → s1 s2 ⊗ t1 t2 . We leave verification of the pushout property as an
exercise.

Example 15.5. S = K[x] and T = K[y] with bases {xm } and {y n }, respectively. S ⊗R T
has a basis xm ⊗ y n . This can be identified as as the polynomial ring K[x, y] via the map
xm ⊗ y n → xm y n .

Example 15.6. C⊗R C is a ring. C has basis {1, i}, so C⊗C has basis {1 ⊗ 1, 1 ⊗ i, i ⊗ 1, i ⊗ i}.
Calculating a few products, we get

(i ⊗ i)(i ⊗ i) = i2 ⊗ i2 = −1 ⊗ −1 = 1 ⊗ 1

(1 ⊗ 1)(a ⊗ b) = (a ⊗ b)
(1 ⊗ 1 + i ⊗ i)2 = 2(1 ⊗ 1 + i ⊗ i).
Call e = (1 ⊗ 1 + i ⊗ i)/2. Then e2 = e, so e is idempotent. Then this ring splits as a
product, so C ⊗ C = e(C ⊗ C) × (1 − e)C(C ⊗ C) ∼= C × C.

Example 15.7. The tensor product satisfies all the axioms of a commutative semiring (a
ring but without subtraction)

1. (A ⊗ B) ⊗ C

2. (A ⊕ B) ⊗ C ∼
= (A ⊗ C) ⊕ (B ⊗ C)

3. A ⊗ B ∼
=B⊗A

61
 4. A ⊕ B ∼
=B⊕A

5. (A ⊕ B) ⊕ C ∼
= A ⊕ (B ⊕ C)

6. R ⊗ A ∼
= A.

If we want to construct a ring out of this structure, we have a few problems:

1. The set of all modules is not a set.

2. There is no subtraction.
This can be circumvented by constructing the set of all pairs M −N for M, N modules
under some equivalence relation.

3. By the swindle, M = 0 for any M .

We circumvent problems 1 and 3 by only considering finitely generated modules.28

Example 15.8. Take R = Z, the integers. The finitely generated modules are all of the
form Zn ⊕ (Z/2Z)n2 ⊕ (Z/4Z)n4 ⊕ (Z/8Z)n8 ⊕ · · · ⊕ (Z/3Z)n3 ⊕ · · · . So we get a basis
{ni bi }, where we allow the ni to be positive or negative. The product is b0 × bn = bn and
bpa × bpb = bpmin(a,b) .

15.4 Tensor products of noncommutative rings

When R is a noncommutative ring, M ⊗R N is only defined for M a right module and N
a left module. This is because we need

(mr) ⊗ n = m ⊗ (rn).

Secondly, M ⊗R N is only an abelian group, not an R-module. We have that

mr ⊗ n = m ⊗ rn,

but multiplying by s gives us

mrs ⊗ n = m ⊗ srn,
even though we want m(rs) ⊗ n = m ⊗ (rs)n.

28
This leads into K-theory, where you consider the ring of finitely generated modules over a ring R.

62
16 Duality
16.1 Notions of duality for algebraic objects
16.1.1 Duality of vector spaces
Definition 16.1. Let V be a vector space over a field K. Then we have the dual vector
space, V ∗ = Hom(V, K).

Recall from linear algebra that we have a natural map V → V ∗∗ taking v 7→ (f 7→ f (v))
for f ∈ Hom(V, k). Additionally, V ∗ is isomorphic to V if dim(V ) < L∞, but there is no
natural isomorphism. This does not hold in the general case; if V = ∞ n=1 K, then V has
∗
countable dimension, but dim(V ) is uncountable.
More generally, for objects in a category, we pick a “dualizing object,” and let the dual
be the set of homomorphisms to that object.

16.1.2 Duality of free modules

For free modules over a ring R, we take the dualizing object to be R. Then M ∗ =
Hom(M, R), and M ∗∗ ∼= M if M ∼ = Rn . This also holds if M is projective. We have M ⊕ N
is free, so M ⊕ N ∼
= (M ⊕ N )∗∗ ; then it is not difficult to obtain the property for M .

16.1.3 Duality for finite abelian groups

Since abelian groups are modules over Z, one might think that you should make Z the
dualizing object, but the only homomorphism from G → Z is the trivial one. So make the
dualizing object Q/Z.

Proposition 16.1. Let G be a finite abelian group. Then G ∼

= G∗ .

Proof. G is a direct sum of cyclic groups, so it is enough to check for when is G cyclic.
Whe have G ∼ = Z/nZ, which means that G∗ = Hom(G, Q/Z) ∼ = {qZ ∈ Q/Z : n(qZ) = Z} =
{0, 1/n, 2/n, . . . , (n − 1)/n}. This is cyclic of order n.

We also get that G ∼

= G∗∗ , and this isomorphism is considered natural.

16.2 Applications of duality

16.2.1 Dirichlet characters
Definition 16.2. A Dirichlet character is an element of the dual of (Z/N Z)∗ , the group
of units29 of the ring Z/N Z.
We are being sloppy here by using ∗ to both mean dual and the group of units. In the case of ((Z/N Z)∗ )∗ ,
29

we mean Hom((Z/N Z)∗ , S 1 ).

63
 Replace Q/Z by S 1 , unit circle in the complex numbers. We have the map Q/Z → S 1
sending x 7→ e2πix , so Q/Z ∼
= elements of finite order in S 1 .
Example 16.1. For N = 8, (Z/N Z)∗ = {1, 3, 5, 7} with 12 = 52 = 72 = 1. The characters
are
1 3 5 7
χ0 1 1 1 1
χ1 1 −1 1 −1
χ2 1 1 −1 −1
χ3 1 −1 −1 1
Dirichlet was interested in this because he defined the Dirichlet L-function
X χ(n)
,
ns
n≥1

where χ is a Dirichlet character. When N = 1 and χ is the trivial character, we get the
RIemann Zeta function.

Definition 16.3. Let χ1 , χ2 be Dirichlet characters for the same N . Then the inner
product of χ1 , χ2 is X
(χ1 , χ2 ) := χ1 (x)χ2 (x).
x∈(Z/N Z)∗

Proposition 16.2. Dirichlet characters are orthogonal.

Proof. Let χ1 6= χ2 , and define the homomorphism χ = χ1 χ2 . Then (χ1 , χ2 ) = (χ, 1),
where 1 is the trivial character (sends everything to 1). Since χ1 6= χ2 , χ 6= 1, so let
a ∈ Z/N Z with χ(a) 6= 1. Then
X X X
χ(x) = χ(ax) = χ(a) χ(x),
x∈(Z/N Z)∗ x∈(Z/N Z)∗ x∈(Z/N Z)∗

where multiplying by a just reindexes the elements of Z/N Z. So we have

X
(χ1 , χ2 ) = (χ, 1) = χ(x) = 0.
x∈(Z/N Z)∗

16.2.2 The Fourier transform

Definition 16.4. Suppose f is a complex function on a finite group G. The Fourier
transform f˜ is a function on G∗
X
f˜(χ) = (χ, f ) = χ(x)f (x).
x∈G

64
 Duality for infinite abelian groups (with a topology) follows a few rules:
1. The dualizing object is S 1 .

2. Groups should be locally compact

3. Homomorphisms should be continuous.

Example 16.2. Let G = Z. Then G∗ = Hom(Z, S 1 ) ∼ = S 1 . Let H = S 1 . Then H ∗ is the
continuous homomorphisms from S 1 → S 1 (z 7→ z n for n ∈ Z). These two groups are dual
to each other.
The fourier transform takes function on S 1 to a fourier series (a function on Z) by
sending Z
X
f 7→ cn e2πinz , cn = e−2πinz f (z) dz.
n z∈S 1

If G = R, then G∗ = Hom(R, S 1 ) ∼
= R. This gives the fourier transform on R.

16.2.3 Existence of “enough” injective modules

Definition 16.5. An injective module I is a module with the following property. If the
sequence 0 → B → A is exact, then any map B → I induces a homomorphism A → I.

0 B C

I
It is not immediately clear how we can find injective modules. The first step is to find
a divisible abelian group.
We want to say that every module is a submodule of an injective module.
Definition 16.6. A group G is divisible if given g ∈ G and n ∈ Z+ , there exists some
h ∈ G with nh = g.
Example 16.3. Q/Z is a divisible abelian group.
Finitely generated abelian groups are never divisible, except for the trivial group.
Proposition 16.3. Let I be a module. If it is divisible as an abelian group, it is injective
as a module.
Proof. Pick a ∈ A with a ∈ / B. We want to extend f to a. Pick the smallest n > 0 so that
na ∈ B if n exists. Extend f to a by putting f (a) = g, where g ∈ I satisfies ng = f (x). If
n does not exist, then put f (a) equals anything (it doesn’t matter what we put here). Now
extend f to all of A using Zorn’s lemma (choose the maximal extension from submodules
of A to I).

65
Proposition 16.4. Every abelian group is contained in an injective module.

Proof. By the previous proposition, Q/Z is injective, and given an abelian group G with
an element a 6= 0 in G, we can find a homomorphism f : G → Q/Z such that f (a) 6= 0. So
any abelian group G is a subset of a (possibly infinite) product of Q/Zs.

Proposition 16.5. Let R be a ring. Then the dual R∗ is an injective R-module

Proof. The key point is that HomZ (R, Q/Z) is an injective R-module. This is the dual
of R as a Z-module. Be careful; Q/Z is a Z-module but not necessarily an R-module. If
f ∈ Hom(R, Z) and r, s ∈ R, define f r by f r(x) = f (rs) This makes HomZ (R, Q/Z) a
right R-module.
The second key point is that HomR (M, Hom(R, Q/Z)) ∼ = HomZ (M, Q/Z); this is easy
but confusing to actually write out, so we leave it as an exercise. So finding an induced
homomorphism from A → Hom(R, Q/Z) is the same problem as finding an induced homo-
morphism from A → Q/Z, which is possible because Q/Z is injective.

0 B C 0 B C
=
Hom(R, Q/Z) Q/Z

So R∗ = Hom(R, Q/Z) is injective, as claimed.

17 Limits and Colimits

Recall from the lecture on category theory that a limit of a family {Gα } is a universal
object with morphisms from G → Gα for each α.

17.1 Colimits
Definition 17.1. A colimit G of the family {Gα } is universal object with morphisms from
Gα → G for each α. In other words, a colimit is the same concept as a limit, but the
arrows (morphisms) go the other way.

Y
ϕ

G
f2
f1 f3
G1 g1,2 G2 g2,3 G3
g3,1

66
 A special case is that if Gi → Gi+1 is injective for all i, then the colimit, G, is more or
less the union of the Gi .

G0 G1 G2 G3 ···

Example 17.1. Q/Z is the union of Z/Z ⊆ ( 12 Z)/Z ⊆ ( 16 Z)/Z ⊆ ( 24

1
Z)/Z ⊆ · · · .

17.1.1 Examples of colimits

Recall that the kernel f : A → B, where A, B are groups is the equalizer of f and 1, the
trivial map from A → B; this is the limit of A, B with the morphisms f, 1.

Definition 17.2. The cokernel X of A and B is the colimit of A, B with morphisms f, 1.

f
A B Y
1

This can also be thought of as the coequalizer of f, 1, where the coequalizer has the
same definition as the equalizer but with the arrows reversed.

Definition 17.3. The push-out X is the colimit of A and B with morphisms f : A → C

and g : B → C.

Y
ϕ

X
p1 p2
A B
f g

67
17.2 Exact sequences of colimits
When do colimits preserve exactlness? Say we have the following diagram with rows exact:

.. .. ..
. . .

0 Ai Bi Ci 0

0 Ai+1 Bi+1 Ci+1 0

.. .. ..
. . .

Then
0→
  colim A → colim B → colim C → 0
 i i i

is right exact but not left exact.

Example 17.2. Here is an example where the colimit is not left exact.
×2
0 Z Z Z/2Z 0
×2
×2
0 Z Z Z/2Z 0
×2 ×2 ×2

Z Z Z/2Z

×2
The colimit Z ⊕ Z/2Z −−→ Z ⊕ Z/2Z is not injective.

When do colimits preserve exactness, then?

Definition 17.4. A directed set S is a partially ordered set such that if a, b ∈ S, there
exists a c with a ≤ c and b ≤ c.

Example 17.3. The set N is directed under the usual ordering ≤.

Definition 17.5. A direct limit is a colimit of a family indexed by a directed set.

Proposition 17.1. Direct limits preserve exactness.

68
Proof. Suppose S is a directed set and we are taking the colimit over a family indexed by
S. We have modules Ai for i ∈ S with Ai → Aj with i < j. Every element of the colimit
is represented by some a ∈ Ai for some i. This is because any element of the colimit is
represented by some sum of elements aj ∈ Aj for various j ∈ S; then we can pick c ≥ all
these j, and take the sum of the images of aj in Ac .
Now suppose we have exact sequences 0 → Ai → Bi → Ci → 0 for i ∈ S. We want to
show that colim Ai → colim Bi is injective. Pick a ∈ colim Ai . Then a is represented by
some ai ∈ Ai for some i ∈ S. Now suppose that ai has image 0 in colim Bi . If bi is the
image of si , then bi = 0 in the colimit. So for some j, the image of bi in Bj is 0. So if aj is
the image of ai in Aj , then aj has image 0. Then aj = 0, which makes Aj → Bj = 0, and
so sj = 0 in the colimit.

17.3 Inverse limits and the p-adic integers

Look at G = Z[1/p]/Z ⊆ Q/Z. This is the colimit of Z/pZ ⊆ Z/p2 Z ⊆ Z/p3 Z ⊆ · · · . What
is G∗ ? We get

Hom(Z/pZ, S 1 ) ← Hom(Z/p2 Z, S 1 ) ← Hom(Z/p3 Z, S 1 ) ← · · ·

Definition 17.6. The inverse limit is the limit of a directed family {Aα }.
So the dual of a direct limit is the inverse limit of the duals. The dual for our example
above is the p-adic integers Zp . Look at

Z/pZ ← Z/p2 Z ← Z/p3 Z ← · · ·

Then the p-adic integers is the inverse limit of this. We get the set of sequences of base p
expansions going to the left an infinite distance. For example, if p = 3, such a sequence
would look like (. . . , 2, 1, 2, 2, 0, 1, 2). Addition and multiplication are indeed well-defined
componentwise.
Does taking inverse limits preserve exactness? The answer is no, even if the set is
directed.
Example 17.4. Take the following diagram, where the rows are exact:

×2
0 Z Z Z/2Z
×3 ×3
×2 ×2
0 Z Z Z/2Z 0
×3 ×3 ×2
×2
0 Z Z Z/2Z 0

69
The inverse limits give us 0 → 0 → 0 → Z/2Z → 0, but this is not exact.

However, there is hope! Taking inverse limits preserve exactness if the Ai preserve the
Mittag-Leffler30 condition.

30
This sounds like two people, but it is actually just one.

70
18 The Snake Lemma
18.1 Statement and proof of the snake lemma
Example 18.1. Consider the following commutative diagram with exact rows:

0 0 0

0 ker f ker g ker h

×2
0 Z Z Z/2Z 0
f ×2 g ×2 h ×2
×2
0 Z Z Z/2Z 0

×2
coker f coker g coker h 0

0 0 0

The map ker g → ker h is not surjective, and coker f → coker g is not injective. The snake
lemma says that these are the same problem.

Lemma 18.1 (Snake). Suppose we have the following commutative diagram with exact
rows:
A0 B0 C0 0
f g h

0 A1 B1 C1
Then there is a map ker h → coker f that makes the “snake sequence”

ker f → ker g → ker h → coker f → coker g → coker h

71
exact. This yields the commutative diagram31

ker f ker g ker h

A0 B0 C0 0

f g h

0 A1 B1 C1

coker f coker g coker h

Proof. We first construct the snake homomorphism by zigzaging through the diagram.
Take c ∈ ker h; then c ∈ C, so since B0 → C0 is surjective, we can lift c to an element
b ∈ B0 . Then we can map b to b0 ∈ B1 . Since c was in ker h and the diagram is commutative,
B1 → C1 sends b0 to 0. So b0 ∈ ker(B1 → C1 ) = im(A1 → B1 ), and we can lift b0 to a0 ∈ A1 .
Note that a0 is unique (given b) because A1 → B1 is injective. Finally, let a00 be the image
of a0 under the map (A1 → coker f ). So we map c 7→ a00 .
Is this well-defined? We have a choice of possibly different b. Suppose we picked some
b0 instead of b, and let a01 be the corresponding element of A1 we get. Note that B0 → C0
sends b − b0 to 0, so there exists some a ∈ A0 such that A0 → B0 maps a to b − b0 . Since
the diagram is commutative, the map A1 → B1 should send f (a) to g(b − b0 ). Then since
f is injective and A1 → B1 sends a0 − a00 to g(b − b0 ), we have that a0 − a00 = f (a); then we
have a0 − a00 ∈ im(f ), so a0 and a00 have the same image in coker f = A1 / im f .
We claim that the snake sequence is exact. The hard part is exactness at ker h and
coker f . Suppose we want to prove exactness at coker f . Suppose a00 ∈ coker f and is
in the kernel of the map coker f → coker g. Lift it to a0 ∈ A1 , and let b0 ∈ B1 be the
image of a0 . b0 maps to 0 in coker g by the definition of a00 (and because the diagram
commutes), so lift it to b ∈ B0 . Map b to c ∈ C0 . Now note that h(c) = 0 because
g(b) = b0 ∈ im(A1 → B1 ) = ker(B1 → C1 ). So c ∈ ker f , and the snake homomorphism
takes c to a00 , so the sequence is exact at coker f . The similar proof for ker h is left as an
exercise.
31
The code for this diagram was modified from an answer on this StackExchange post.

72
18.2 Applications of the snake lemma
18.2.1 Exact sequences of tensor products of modules
Recall that if 0 → A → B → C → 0 is exact, then so is

A ⊗ M → B ⊗ M → C ⊗ M → 0.

However, A ⊗ M → B ⊗ M is not always injective. What is the kernel? Choose free

modules Fi , Hi so that

0 → F1 → F0 → A → 0, 0 → H1 → H0 → C → 0.

Extend this to the following diagram:

0 F1 F1 + H1 H1 0
f g h
×2
0 F0 F0 + H0 H0 0

0 A B C 0

Tensor every row with M and put in the kernels to get the diagram

0 0 0

0 ker f ker g ker h

0 F1 ⊗ M (F1 ⊗ M ) + (H1 ⊗ M ) H1 ⊗ M 0
f g h

0 F0 ⊗ M (F0 ⊗ M ) + (H0 ⊗ M ) H0 ⊗ M 0

A⊗M B⊗M C ⊗M 0

0 0 0

Note that the bottom row is the row of cokernels of the vertical maps f, g, h, so by the
snake lemma, we get an exact sequence

0 → ker f → ker g → ker h → A ⊗ M → B ⊗ M → C ⊗ M → 0.

73
we can also call these
0 → Tor(A, M ) → Tor(B, M ) → Tor(C, M ) → A ⊗ M → B ⊗ M → C ⊗ M → 0.
Is Tor(A, M ) well-defined? It seems to depend on the choice of 0 → F1 → F0 → A → 0.
It is, in fact, well-defined.
Let’s calculate Tor(M, N ) for finitely generated abelian groups M, N . First, we have
Tor(M1 ⊕ M2 , N ) ∼ = Tor(M1 , N ) ⊕ Tor(M2 , N ), so it is enough to do the case where M, N
are cyclic. If M = N = Z, take the resolution 0 → F1 → F0 → M → 0. If M = Z and
N = Z/nZ, we have

0 F1 F0 M 0

0 0 Z Z 0

0 0 Z/nZ Z/nZ 0

So Tor(Z, Z/nZ) = 0.
If m = Z/mZ and N = Z/nZ, we have

0 F1 F0 M 0

×m
0 Z Z Z/mZ 0

×m
0 Z/nZ Z/nZ ··· 0
×m
Then Tor(Z/mZ, Z/nZ) = ker(Z/nZ −−→ Z/nZ) = Z/(m, n)Z.
So Tor(M,N) depends only on the torsion subgroups of M, N . In fact, if M, N are
finite, M ⊗ N ∼
= T or(M, N ), although this isomorphism is not natrual.
Example 18.2. Here is a historical example from algebraic topology. This is where the
idea of Tor came from. The universal coefficient theorem states that
Hi (M, G) = (Hi (M, Z) ⊗ G) ⊕ Tor(Hi−1 (M, Z), G),
where Hi (M, G) is the homology of the manifold M with coefficients in G.
Example 18.3. As a specific case of the previous example, let M = P 2 (2-dimensional
projective space). This is S 2 , where we identify opposite points. Suppose we know
H0 (M, Z) = Z, H1 (M, Z) = Z/2Z, and Hi (M, Z) = 0 for i > 1. Then
H0 (M, Z/2Z) = H0 (M, Z) ⊗ Z/2Z = Z/2Z

74
 H1 (M, Z/2Z) = H1 (M, Z) ⊗ Z/2Z ⊕ Tor(H0 (M, Z), Z, 2Z)
H2 (M, Z/2Z) = H2 (M, Z) ⊗ Z/2Z ⊕ Tor(H1 (M, Z), Z, 2Z),
which allows us to compute the homology32 H2 (M, Z/2Z).

18.2.2 The Mitag-Leffler condition

Look at · · · → A3 → A2 → A1 → A0 . Does the sequence of images stabilize? In other
words, does im Ai = im Ai+1 = · · · for some i?

Definition 18.1. Let · · · → A3 → A2 → A1 → A0 . The Mitag-Leffler condition is that

the sequence of images stabilizes for all An ; that is, for each n ∈ N, ther exists some i ≥ n
such that im Ai = im Ai+1 = · · · .

Example 18.4. The Mitag-Leffler condition holds if all Ai are finite.

Theorem 18.1. Suppose we have

.. .. ..
. . .

0 Ai+1 Bi+1 Ci+1 0

0 Ai Bi Ci 0

.. .. ..
. . .

If the Mitag-Leffler condition is satisfied, then

0 → lim Ai → lim Bi → lim Ci → 0.

Proof. We first do two easy cases:

1. Suppose all maps Ai+1 → Ai are onto (so ML condition is satisfied). We want to show
that lim Bi → lim Ci is onto. Pick some element of lim Ci , which looks like (c0 , c1 , . . . ) for
ci ∈ Ci , where ci is the image of ci+1 . We can lift the ci to bi . Is bi the image of bi+1 ? Pick
b0 ∈ B0 , and choose some b1 ∈ B1 . Then im(b1 ) − b0 ∈ ker(B0 → C0 ) = im(A0 → B0 ), so
let a0 ∈ A0 be its preimage. Then we can lift a0 to a1 ∈ A1 . Now replace b1 by b1 + im(a1 ).
Repeat this to find b2 , b3 , . . . . So bi maps to ci and bi−1 .
32
In the first edition of Lang’s book, there was an infamous exercise that said, “Take any book on
homological algebra, and prove all the theorems without looking at the proofs given in that book.” Professor
Borcherds seemed dismayed that the exercise was removed in a later edition of the book.

75
 2. Suppose for each i, we can find j so that Aj → Ai is 0 (this is the extreme opposite
condition to case 1). Then the ML condition holds. We want to show that lim Bi → lim Ci
iis onto. Pick Ai0 . Pick Ai1 so Ai1 → Ai0 is 0. Do the same over and over to get
→ Ai2 → Ai1 → Ai0 . Take the inverse limits over B0 , Bi1 , Bi2 , etc.. So we can assume all
maps Ai+1 → Ai are 0. Pick (c0 , c1 , c2 , . . . ), and pick bi mapping to ci . Is im(bi ) = bi−1 ?
The image of im(b2 ) is im(b1 ) because im(b2 ) − b1 is in the image of A1 , which is 0 in A0 .
So the sequence im(b1 ), im(b2 ), im(b3 ), . . . is in lim Bi , and has image (c0 , c1 , c2 , . . . ).
Now
T we combine the special cases 1 and 2. Suppose Ai satisfied the ML condition. Put
Xi = j≥i im(Aj → Ai ). So Xi ⊆ Ai , and we get exact sequences

0 Xi Ai Ai /Xi 0

0 Xi−1 Ai−1 Ai−1 /Xi−1 0

where the down maps for the Xi are surjective. For each i, we can find j so that
im(Aj /Xi → Ai /Xi ) = 0.
Use the snake lemma. Recall that 0 → A → B → C → 0 is exact implies that

A⊗M →B⊗M →C ⊗M →0

is exact and

Tor(A, M ) → Tor(B, M ) → Tor(C, M ) → A ⊗ M → B ⊗ M → C ⊗ M → 0

is exact.
Copy this argument since the limit is left exact. We do this by flipping all the arrows.
We constructed Tor by taking 0 → G1 → F0 → A → 0; this works when F is free or
projective. So we can flip the arrows by replacing the projective modules by injective
modules 0 → A → I0 → I1 → 0; this uses our fact that every module is contained in an
injective module.
So the analogue of Tor is lim1 (Ai ). We get a sequence

0 → lim Ai → lim Bi → lim Ci → lim1 Ai → lim1 Bi → lim1 Ci .

For this to be exact, we want lim1 Ai = 0. The proofs above show that this is true if either
of the special cases hold. Now look at 0 → Xi → Ai → Ai /Xi → 0. We have

0 → lim Xi → lim Ai → lim Ai /Xi → lim1 Xi → lim1 Ai → lim1 Ai /Xi → 0.

18.3 Unrelated: Finitely generated modules over a PID

Theorem 18.2. Any finitely generate modules over PID are sums of cyclic modules of the
form R/I.

76
Proof. We don’t have time in class to prove the whole theorem, so we will cheat and just
do the case of Euclidean domains. The proof is the same as the one we gave for Z. If
M is any submodule of Zn , we can find a basis b1 , . . . , bn of Zn . So M isLspanned by
d1 b1 , d2 b2 , . . . , dn bn for some di Then the finitely generated module Zn /m = Z/di Z.

77
19 Polynomials and Divisibility
19.1 Polynomial division with remainder
We start with some results you should already know.

Theorem 19.1. Suppose f, g are polynomials in R[x], where R is a commutative ring.

Also suppose that f has leading coefficient 1, so f (x) = xn + an−1 xn−1 + · · · + a0 . Then
g(x) = f (x)q(x) + r(x), where deg(r) < deg(f ).

Corollary 19.1. If K is a field, K[x] is a Euclidean domain.

Proof. We can make the leading coefficient of any polynomial 1 by multiplying by a unit.
Then apply the theorem.

Corollary 19.2. If K is a field, K[x] is a principal ideal domain.

Proof. All Euclidean domains are PIDs.

Corollary 19.3. If K is a field, K[x] is a unique factorization domain.

Proof. All PIDs are UFDs.

Example 19.1. How can we find the prime elements of F2 [x], where F2 = Z/2Z, the field
with 2 elements? Recall the sieve of Eratosthenes33 . List all numbers > 1, identify the
smallest number as prime, and cross out all multiples of it.

10,
 ...
2, 3, 4
, 5, 6
, 7, 8
, 9, 

Then, identify the first non-crossed out number as prime, and cross out all multiples of it.

10,
 ...
2, 3, 4
, 5, 6
, 7, 8
, 9
, 

If we repeat this process, we can find all the prime numbers.

For F2 [x], we list all elements (other than 0 or units) in order of degree.

x, x + 1, x2 , x2 + 1, x2 + x, x2 + x + 1, . . .

Cross out all multiples of x.

x, x + 1, x2 , x2 + 1, 
x2
+x, x2 + x + 1, . . .


The next element, x+1, is prime, so we cross out multiples of it. Note that x2 +1 = (x+1)2
in F2 [x].
x, x + 1, x2 , 
x2
 2
x + x, x2 + x + 1, . . .

+1, 
33
Erathosthenes was the first person to accurately calculate the circumference of the Earth.

78
The polynomials not divisible by x and x + 1 are

x2 + x + 1, x3 + x + 1, x3 + x2 + 1, x4 + x + 1, (
x4( 2
+ 1, x4 + x3 + 1, x4 + x3 + x2 + x + 1,
((
+(x(

and we can continue the process.

Proposition 19.1. Suppose a polynomial f ∈ R[x] has a root a (f (a) = 0). Then f (x) =
g(x)(x − a) for some g.

Proof. Apply division to get that f (x) = g(x)(x − a) + r. We have deg(r) < 1, so r is
constant. Put x = a to get f (a) = g(a)(a − a) + r = r, so r = 0.

Corollary 19.4. A polynomial f ∈ R[x] of degree n over an integral domain R has ≤ n

roots.

Proof. If a1 , . . . , ak are roots, then f (x) = (x−a1 ) · · · (x−ak )g(x), so k ≤ n. If the product
is 0, then so is some factor (x − ai ) because R is an integral domain.

Example 19.2. Let R = Z/8Z, which is not an integral domain. Let f (x) = x2 − 1, which
has degree 2. Then f (x) has 4 roots: 1, 3, 5, and 7.

Example 19.3. Let R be the quaternions (this is noncommutative), and look at f (x) =
x2 + 1. Then f has roots ±i, ±j, ±k, and roots ai + bj + ck for real a, b, c that satisfy
a2 + b2 + c2 = 1. This is an uncountable number of roots!

19.2 An application to field theory

We first prove a lemma.

Lemma 19.1. Any abelian group G with ≤ n elements of order n (∀n ≥ 1) is cyclic.

Proof. Recall that G ∼= Z/pn1 1 Z × Z/pn2 2 Z × · · · . Suppose that p1 = p2 ; then G has p2

elements x with xp = 1 (since G contains Z/pZ × Z/pZ). This is impossible, so all pi are
distinct. Then G is cyclic by the Chinese remainder theorem (Z/mZ × Z/nZ ∼ = Z/mnZ if
m, n are coprime).

Proposition 19.2. The group (Z/pZ)∗ of units mod p is cyclic if p is prime.

Proof. Since p is prime, Z/pZ is a field. So any polynomial in R[x] of degree n has ≤ n
roots. So xn − 1 has ≤ n roots for any n ≥ 1. Then G has ≤ n elements x with xn = 1
(for n ≥ 1). Using the lemma finishes off the proof.

Example 19.4. This need not hold if p is not prime. (Z/12Z)∗ ∼

= (Z/4Z)∗ × (Z/3Z)∗ , and
these are both cyclic of order 2.

Definition 19.1. A generator of (Z/pZ)∗ is called a primitive root.

79
 We have shown that primitive roots always exist when p is prime.

Example 19.5. Let’s find a primitive root of p = 23. The element should have order 22.
Check the elements −1, 1, 2, 3, 4, 5. We find that 5 is the primitive root because 52 , 511 6≡ 1
(mod 23).

The same argument shows that the following is true.

Theorem 19.2. If F is a field, any finite subgroup of F ∗ is cyclic.

Example 19.6. Let F = C, and take the subgroup of 8th roots of unity. This has primitive
root e2iπ/8 .

This also gives us the following corollary.

Corollary 19.5. If F is any finite field, then F ∗ is cyclic.

19.3 Unique factorization in polynomial rings

We want to show that Z[x] is a UFD, and we know that Z[x] ⊆ Q[x], which is a UFD
because Q is a field. We cannot do this as we usually do, because Z[x] is not a Euclidean
domain or a PID. For example, (2, x) is a non-principal ideal. So we use the fact that Q[x]
is a UFD.

Definition 19.2. Let f ∈ Q[x]. The content c(f ) is defined as follows: Suppose f (x) =
an xn + · · · + a0 . For each prime p, an = pmn bn , an−1 = pmn−1 bn−1 , . . . with mi ∈ Z and
bi not having any factors of p in the numerator or denominator. Let c(f ) = pmin(mi ) × b,
where b is some number with no factors of p.

Example 19.7. Let f (x) = (2/3)x2 + 4. Then c(f ) = 2/3.

Proposition 19.3. Z[x] is a unique factorization domain.

Proof. The key point of the proof is that c(f g) = c(f )c(g). We may assume that c(f ) =
c(g) = 1; otherwise, we multiply f and g by constants to make this so. We want to show
that c(f g) = 1. We know that f has integer coefficients, so c(f ) ∈ Z. Suppose p is any
prime in Z; we show that p does not divide c(f g).
Since c(f ) = c(g) = 1, p does not divide all coefficients of f or all the coefficients of g.
So f = an xn + · · · + ai xi + · · · + a0 and g = bm xm + · · · + bj xj + · · · + b0 where i and j are
the least indices such that ai and bj are not divisible by p. So the coefficient of xi+j in f g
is
a0 bi+j + a1 bi+j−1 + a2 bi+j−2 + · · · + ai bj + · · · + ai+j−1 b1 + ai+j b0 ,
which has all terms except ai bj divisible by p. This means that the coefficient of xi+j in
f g is not divisible by p. This is true for any prime p, so c(f g) = 1.

80
 We sketch the rest of the proof. The main point is that we need to show that irre-
ducible elements are prime. Recall that irreducible elements are such that f 6= gh with
deg(g), deg(h) < deg(f ); prime elements are such that if f divides g, h, then f divides g or
h.
The irreducibles of Z[x] are the primes 2, 3, 5, 7, . . . ∈ Z and the polynomials f (x) of
degree > 1 with c(f ) = 1.
We leave the following two statements as exercises:

1. These are all the irreducibles of Z[x].

2. Any element of Z[x] is a product of irreducibles.

If deg(f ) = 0, then f = p is prime in Z. If f divides gh, this means that c(gh) is

divisible by p. So c(g) or c(h) is fivisible by p (since c(gh) = c(g)c(h). So p divides gh.
The case of deg(f ) > 0 is similar and left as an exercise.

We have really proved the following theorem.

Theorem 19.3. If R is a UFD, then so is R[x].

Proof. Perform the same proof but with a few modifications. First, c(f ) is now only
defined up to multiplication by a unit. Also, irreducibles of R[x] are either irreducibles of
R (deg = 0) or irreducibles of K[x] with content 1, where K is the quotient field of R.

Corollary 19.6. Z[x1 , . . . , xn ] is a unique factorization domain.34

Corollary 19.7. If K is a field, K[x1 , . . . , xn ] is a unique factorization domain.

Proof. These two have the same proof: induction on the number of variables.

19.4 Irreducibility tests in Z[x] (or Q[x])

Given f ∈ Z[x], how do we factor f into irreducibles?

Example 19.8. Here is an algorithm, due to Kronecker:

Suppose that f − gh. We can assume g, h ∈ Z[x]. Then f (n) = g(n)h(n) for any
n ∈ Z. So we factor f (0), f (1), . . . , f (m), where m = deg(f ). Then g(0) divides f (0)
or g(1) divides f (1), (and so on), so there are only a finite number of possibilities for
g(0), . . . , g(m). But deg(g) ≤ m, so g is determined by g(0), . . . , g(m).

Kronecker’s algorithm is pretty slow. There are faster algorithms.

34
In fact, Z[x1 , x2 , . . . ] in infinitely many variables is a field, but we will not prove that here.

81
Example 19.9. The LLL algorithm35 is fast but not necessarily precise. We can write
f = af1 f2 · · · fn , where fi is irreducible with degree > 0 and a ∈ Z. We can do this in
polynomial time, but to find a, we must factor an integer, which may not be possible in
polynomial time.

To test for reducibility, we can use reduction mod p: If f (x) = g(x)h(x), then f (x) =
g(x)h(x) (mod p) for any prime p.

Example 19.10. Is 9x4 + 6x3 + 26x2 + 13x + 3 irreducible? Yes. It is x4 + x + 1 (mod 2),
and we saw that this was irreducible (mod 2).

Example 19.11. Let’s test if x4 − x2 + 3x + 1 is irreducible.

(mod 2) : x4 + x2 + x + 1 = (x + 1)(x3 + x2 + 1),

which are both irreducible (mod 2).

(mod 3) : x4 − x2 + 1 = (x2 + 1)2 .

which is also irreducible (mod 3).

Combine these results. The first one says that the only possible factorization is a
degree 1 polynomial times a degree 3 polynomial. The second says that the only possible
factorization is into 2 degree 2 polynomials. So the polynomial must be irreducible.

Theorem 19.4 (Eisenstein). Suppose f (x) has the following properties:

1. The leading coefficient is 1.

2. All other coefficients are divisible by p.

3. The constant term is not divisible by p2 .

Then f is irreducible.

We will not prove this right now. First, let’s see some examples.

Example 19.12. The polynomial x5 − 4x + 2 is irreducible by Eisenstein’s criterion.

Example 19.13. Look at the p-th roots of 1. These are the roots of the polynomial
xp − 1 = (x − 1)(xp−1 + xp−2 + · · · + x + 1). We want to show that the latter term is
irreducible by Eisenstein’s criterion. We need a trick to make this work. Put z = x − 1.
Then
xp − 1 (z + 1)p − 1
xp−1 + · · · + x + 1 = =
x−1 z
35
This stands for Lenstra, Lenstra, and Lovasz.

82
 p(p−1) p−2
(z p + pz p−1 + 2 z + · · · + pz + 1) − 1
=
z
= z p−1 + pz p−2 + · · · + p,

so Eisenstein applies, and z p−1 + pz p−2 + · · · + p is irreducible. So xp−1 + xp−2 + · · · + x + 1

is irreducible, as desired.
Why does this work? The prime p is totally ramified in Z[ζ], where Zp = 1. We have
that p factorizes in Z[ζ] as (1 − ζ)p−1 u, where u is a unit.

83
20 More on Irreducibility Tests
20.1 Eisenstein’s criterion
p −1
Last lecture, we were applying the Eisenstein criterion to xx−1 = xp−1 + xp−1 + · · · + x + 1.
We saw that if we set z = x − 1, this equaled z p−1 + pz p−2 + · · · + p.
Why does this work? Let ζ = e2πi/p , and look at the ring Z[ζ]. Then p pactorizes as
(1 − ζ)p−1 u for some unit u. In an algebraic number theory course, we would say that p is
“totally ramified,” so Eisenstein’s criterion applies. Notice that the polynomial has roots
ζ, ζ 2 , . . . , ζ p−1 , the p-th roots of unity. We also have that (ζ k − 1) = (ζ − 1)(ζ k−1 + · · · + 1).
Conversely, ζ − 1 is divisible by ζ k − 1, so ζ k is also a root of 1.

20.2 Rational roots

The only linear factors of xn + an−1 xn−1 + · · · + a0 are of the form x − b for b dividing a0 .
This is because (cx + b)(· · · ) = xn + · · · + a0 , so 1 = c × ∗ and a0 = b × ∗.

Example 20.1. It is not possible to trisect the angle of 120◦ with just a compass and
straightedge.36 We will show that we cannot construct 2 cos(40◦ ). We will not prove this
here, but any number that can be constructed cannot satisfy an irreducible polynomial of
degree n unless n is a power of 2. We want to show that 2 cos(40◦ ) satisfies an irreducible
polynomial in Z[x] of degree 3.
Look at z = e2πi/9 = cos(2π/9)+i sin(2π/9). This is an angle of 40◦ . Then 2 cos(40◦ ) =
z + z −1 . So we have the polynomial

0 = z 9 − 1 = (z 3 − 1)(z 6 + z 3 + 1),

which means that z 6 +z 3 +1 = 0. Rewriting this as z 3 +1+z −3 = 0 and letting c = (z+z −1 )

we get
c3 − 3c + 1 = 0.
To show that this is an irreducible polynomial, note that c3 −3c+1 has no linear factors
over Q. We just have to check that factors of the constant term 1 are not roots.
If a polynomial of degree ≤ 3 has no linear factors, it is irreducible.37 So c3 − 3c + 1 is
irreducible.

Example 20.2. The polynomials

x100 + 2, x100 + 3
36
Professor Borcherds gets a lot of emails from people claiming to have proven Fermat’s last theorem,
Goldbach’s conjecture, or that is is possible to trisect any angle.
37
The same method makes it easy to check polynomials of degree ≤ 3, but, in Professor Borcherd’s words
“degree ≥ 4 is painful.”

84
are both irreducible. This is in contrast to in general, where polynomials of the form xn + b
can have “unexpected” factorizations. For example,

x100 + 4 = (x50 + 2x25 + 3)(x50 − 2x25 + 2).

21 Noetherian Rings and Hilbert’s Theorem

21.1 Noetherian rings and Noether’s theorem
Definition 21.1. A ring is Noetherian 38 if all ideals are finitely generated.

Theorem 21.1. For a ring R, the following are equivalent:

1. R is Noetherian.

2. Every nonempty set of ideals has a maximal element.

3. Every strictly increasing chain I1 ( I2 ( I3 ( · · · of ideals is finite.

Proof. (2) ⇐⇒ (3): First note that (3) =⇒ (2) is just Zorn’s lemma in disguise. To get
(2) =⇒ (3), observe that if I1 ( I2 ( I3 ( · · · is infinite, then the set {I1 , I2 , I3 , . . .} has
no maximal element.39 S
(1) =⇒ (3): Suppose that I1 ⊆ I2 ⊆ I3 ⊆ · · · is a chain of ideals. Put I = i Ii .
Then I is an ideal. By condition (1), I = (x1 , . . . , xn ), so all xi are in some Im . Then
Im = Im+1 = · · · .
(3) =⇒ (1): Pick an ideal I. We want to show that I is finitely generated. Pick
any x1 ∈ / I. If I = (x1 ), we are finished. Otherwise, pick x2 ∈ I with x2 6= x1 and
check if I = (x1 , x2 ). If we are still not finished, continue abd we get (x1 ) ⊆ (x1 , x2 ) ⊆
(x1 , x2 , x3 ) ⊆ · · · . The infinite chain must stop by condition (3), so I = (x1 , · · · , xn ) is
finitely generated.

Example 21.1. Let R = K[x1 , x2 , . . . ]. Then (x1 ) ( (x1 , x2 ) ( (x1 , x2 , x3 ) ( · · · , so R is

not Noetherian.

Example 21.2. Let R = Z. We have the infinitely decreasing chain of ideals (2) ) (4) )
(8) ) (16) ) · · · . Rings without decreasing chains of ideals are called Artinian. It turns
out that all Artinian rings are Noetherian.

Theorem 21.2 (Noether). If R is Noetherian, so is R[x].

38
Emmy Noether found that a lot of theorems proven about polynomial rings using complicated techniques
could be simplified by using this condition.
39
You may notice that we did not use any properties of rings here. This part of the equivalence is just a
general fact about partially ordered sets.

85
Proof. Suppose I is an ideal of R[x]. Look at the chain of ideals of R given by I0 ⊆ I1 ⊆
I2 ⊆ · · · , where Ik is the set of leading coefficients of polynomials in I of degree ≤ 0.
R is Noetherian, so for some m, Im = Im+1 = Im+2 = · · · . Pick the set of polynomials
of degree 0 whose leading coefficients generate I0 (which is finitely generated because R is
Noetherian). Do this for polynomials of degree 1, 2, etc. We only need to do this finitely
many times because Im = Im+1 = Im+2 = · · · . We now leave it as an exercise to show that
these finite sets generate I.

21.2 Hilbert’s theorem

Theorem 21.3 (Hilbert). Any ideal of K[x1 , . . . , xn ] is finitely generated.

Proof. Use induction on number of variables and then use Noether’s theorem.

The following example shows why this is important.

Example 21.3. Recall that ideals of K[x] are generated by 1 element, but this need not
be true for K[x, y]. Look at the ideal (x3 , x2 y, xy 2 , y 2 ); this ideal must have at least 4
generators because no element in this set generates more than 1 of these 4 elements. In
general, ideals of K[x1 , . . . , xn ] need not be generated by n elements.

Example 21.4. This need not hold for infinitely many variables. K[x1 , x2 , x3 , . . . ] has the
ideal (x1 , x2 , x3 , . . . ), which cannot be generated by a finite number of elements.

Example 21.5. Look at the ideal (x) in K[x, y]. Then (x) is a ring (but without an
identity element) and is not finitely generated as a ring. For example, a generating set
could be x, xy, xy 2 , . . . . So we must pay attention to the distinction between being


finitely generated as an ideal of a ring and being finitely generated as a ring.

21.3 Rings of invariants and symmetric functions

Suppose a group G acts on a vector space V with basis {x1 , . . . , xn }. So for g ∈ G,

g · x1 = g1,1 x1 + g1,2 x2 + · · · + g1,n xn

G also acts on polynomials in x1 , . . . , xn by g · (p + q= g · p + g · q and g · (pq) = (g · p)(g · q).

Definition 21.2. The ring of invariants is the set of polynomials fixed by G (that is, the
polynomials p such that g · p = p for all g ∈ G).

Can we find a finite number of invariants so all invariants are polynomials in them with
coefficients in K? Hilbert showed that this is often true, and about 50 years later, Nagata
found a counterexample which showed that it is not always true.

86
Definition 21.3. Let V have basis {x1 , . . . , xn }, and let G be the symmetric group on
{x1 , . . . , xn }. The ring of symmetric functions is the ring of invariant polynomials.40

Example 21.6. Here are some examples of symmetric functions:

x1 + x2 + x3 + · · · xn

x1 x2 x3 · · · xn
x1 x2 + x1 x3 + · · · + x1 xn + x2 x3 + · · · + xn−1 xn
x1 x2 x3 + x1 x2 x4 + · · · + x1 x3 x4 + · · ·

Look at (x − x1 )(x − x2 ) · · · (x − xn ) = xn − ( xi )xn−1 + ( i<j xi xj )xn−2 + · · · ± xi .

P P Q
The coefficients of this polynomial are called the elementary symmetric functions.

Theorem 21.4. Any symmetric function is a polynomial in elementary symmetric func-

tions.

Proof. We produce an algorithm. The key point is to order the monomials in the right
way.41 We say xn1 1 xn2 2 · · · ≥ xm 1 m2
1 x2 · · · if (n1 , n2 , . . . ) ≥ (m1 , m2 , . . . ) in the lexicographic
order.
Suppose we have a symmetric polynomial p. Look at the biggest monomial in it, and
kill this monomial by subtracting the polynomial

q = (x1 + x2 + · · · )n1 −n2 (x1 x2 + · · · )n2 −n3 (x1 x2 x3 + · · · )n3 −n4 .

Note that all these terms are elementary symmetric functions. So p−q has a smaller largest
monomial. Repeating this process, we eventually get to 0 because it is not possible to have
an infinite sequence of strictly decreasing monomials (exercise).

40
Symmetric functions have a very rich combinatorial theory, showing up in places such as the irreducible
characters of the symmetric group and the number of Young tableau of a given shape. If you want to learn
more about symmetric functions, you should check out my notes on Math 249, Algebraic Combinatorics!
41
Ordering the monomials of a polynomial is very important in the study of Gröbner bases.

87
22 Symmetric Functions and Polynomial Invariants
22.1 Symmetric functions and Newton’s identities
Last time, we saw that any symmetric polynomial f is a polynomial in the elementary sym-
metric functions. We took the monomial xn1 1 xn2 2 · · · in f which is largest, and subtracted

(x1 + · · · + xn )n1 −n2 · · · .

The key point was that since f is symmetric, n1 − n2 , n2 − n3 and other terms are positive;
n n
if f has a term with xni i xj j with nj < ni , then f also has xi j xnj i .

22.1.1 Newton’s identities

What is x41 + x42 + x43 + · · · ? Look at

f (x) = (x − x1 )(x − x2 ) · · · (x − xn ) = xn − e1 xn−1 + e2 xn−2 + · · · .

d f 0 (x)
Take the logarithmic derivative, dx log f (x) = f (x) . The log derivative of f g is the log
derivative of f plus the log derivative of g.
So the log derivative of x − x1 is
1 1 x1 x1
= + 2 + 3 + ··· .
x − x1 x x x
And we get that the log derivative of f is

n x1 + x2 + · · · x21 + x22 + · · · p0 p1
+ 2
+ 3
= + 2 + ···
x x x x x
So f ( pm /xm+1 ) = f 0 gives us that
P

p0 p1
(xn − e1 xn−1 + · · · ))( + 2 ) = nxn−1 − (n − 1)e1 xn−2 + · · · .
x x
Equating the powers of x, we have

p0 = n, p1 − e1 p0 = −(n − 1)e1 , p2 − e1 p1 + e2 p0 = (n − 2)e2

Example 22.1. Let α, β, γ be the roots of z 3 + z + 1. What is α5 + β 5 + γ 5 ? We have

p0 = 3, p1 = 0, p2 + p0 = 1, p2 = −1, p3 = −3, p4 = 2.

and p5 + p3 + p2 = 0. These are the coefficients of the polynomial.42

42
In the 19th century, undergraduate students were expected to be able to calculate things like this
involving symmetric functions.

88
22.2 The discriminant
What about polynomials in x1 , . . . , xn invariant under the alternating group, An ?

Definition 22.1. A polynomials f in variables x1 , . . . , xn is antisymmetric if it changes

sign under elements σ ∈
/ An .

Proposition 22.1. Suppose f is invariant under An . Then f = g+h, where g is symmetric

and h is antisymmetric.

Proof. Set
f + σf f − σf
g= , h= .
2 2
The polynomial h changes sign if we switch xi and xj , so h is divisible by the polynomial
(x1 − x2 )(x1 − x3 )(x2 − x3 ) · · · . So let
Y
∆= (xi − xj ).
i<j

The invariant functions of An are generated by the symmetric functions e1 , . . . , en and

∆. Note that ∆2 is symmetric, so ∆2 is some polynomial in e1 , . . . , en . This is called
syzygy.43

Definition 22.2. The discriminant 44 of f (x) = an xn + an−1 xn−1 + · · · + a1 x + a0 is

an2n−2 ∆2 .

The discriminant vanishes iff f has multiple roots.

Proposition 22.2. A polynomial f has a multiple root iff f and f 0 have a common factor.

Proof. If f = (x − x1 )2 · · · , then f 0 = 2(x − x1 ) · · · + (x − x1 )2 · · · , so x − x1 is a common

factor. The converse is an exercise.

When do f (x), g(x) have a common factor?

f (x) = am xn + · · · + a0

g(x) = bn xn + · · · + b0
If f, g have a common factor, then f (x)p(x) − g(x)q(x) = 0 for some p, q with deg(p) < n
and deg(q) < m (set p = g/(x − α) and q = −f /(x − α)).
43
This comes from syn, which means together, and zygon, which means yoke. This is not the longest
word in the English language with no vowels; that honor goes to the word rhythms.
44
Invariants tend to end with -ant. For example, we have the determinany, the resultant, and the
catalecticant. Professor Borcherds is glad the last of these has fallen out of usage.

89
 This is a set of linear equations for coefficients of p, q. This has a nonzero solution if
some determinant vanishes. So the coefficients of linear equations are:

am am−1 · · · a0 0 0 0 0
 
0 am · · · a1 a0 0 0 0 
.
 
 . 
 . 
 
0
 0 · · · an · · · a2 a1 a0  
 bn bn−1 · · · b0 0 0 0 0
 
0
 bn · · · b1 b0 0 0 0 

 .
 ..


0 0 · · · bn · · · b2 b1 b0

This matrix with n + m rows is called the Sylvester matrix.

Definition 22.3. The resultant is the determinant of the Sylvester matrix.

Say f, g have a common root at ∞ if am = bm = 0. The resultant equals 0 iff f and

g have a common factor, possibly at ∞. This is the same as saying in geometry that the
projective line is complete.

Example 22.2. The polynomial f (x) = xn − e1 xn−1 + · · · has a multiple root if the
resultant of f, f 0 = 0. ∆ = 0 iff f has tahe multiple root, so ∆ should be a constant times
the resultant.

Example 22.3. When is the cubic curve y 2 = x3 + bx + c nonsingular? Curve f (x, y) is

nonsingular if g(x, y) = 0 = fx (x, y) = fy (x, y) has no solutions, where fx is the partial
derivative with respect to x. These are the conditions that 2y = 0 (so y = 0) and 3x2 +b = 0
(so g(x) = x3 bx + c = 0); then we need to check if g, g 0 have a common root x.
The resultant of x3 + bx + c and 3x2 + b, is
 
1 0 b c 0
0 1 0 b c 
 
det 
3 0 b 0 0 

0 3 0 b 0
0 0 3 0 b

which is 4b3 + 27c2 (up to a sign).

22.3 The ring of invariants, revisited

Suppose a finite group G acts on a complex vector space V spanned by {x1 , . . . , xn }. Recal
that the ring of invariant polynomials is the set of polynomials in x1 , . . . , xn invariant under
the action ofG. Is this ring finitely generated (over C)?

90
Example 22.4. If G = An and V = Cn , then the ring is generated by e1 , . . . , en , ∆.

In general this can be “mindbogglingly difficult.”45 Hilbert showed that the ring of
invariants is finitely generated over C.

Definition 22.4. The Reynolds operator 46 ρ is the average of the group elements,
1 X
ρ= g.
|G|
g∈G

The Reynolds operator takes polynomials in C[x,1 , . . . , xn ] to invariants.

x1 +x2 +···+xn
Example 22.5. Let G = Sn . Then if f = x1 , ρ(f ) = n .

Proposition 22.3. They Reynolds operator has the following properties:

1. ρ(f + g) = ρ(f ) + ρ(g)

2. ρ(1) = 1

3. ρ(f g) = ρ(f )ρ(g) if f = ρ(f )

Proof. Exercise.

Theorem 22.1 (Hilbert). If G is finite, the ring of invariants is always finitely generated
over C.

Proof. Look at the ring C[x1 , . . . , xn ]. This is graded by degree, where deg(xi ) = 1. Let
I be the ring of invariants. Then I = C ⊕ I1 ⊕ I2 ⊕ · · · , where Im is the set of invariants
homogeneous of degree m. Look at the ideal generated by I1 ⊕ I2 ⊕ I3 ⊕ · · · . By Hilbert’s
theorem, this ideal is finitely generated. Pick generators i1 , . . . , ik of this ideal. We show
that they generate the ring I.
Suppose the generate I1 , I2 , . . . Ik . We want to show that they generate Ik+1 . Pick
f ∈ Ik+1 . Then f is in an ideal J, so f = a1 i1 +a2 i2 +· · ·+an in for some an ∈ C[x1 , . . . , xn ]
with deg(ai ) > 0.
Apply the Reynolds operator. Then

ρ(f ) = ρ(a1 )i1 + ρ(a2 )i2 + · · · + ρ(an )in

because f is invariant. So deg(an ) < K as deg(in ) > 0, so ρ(an ) is a polynomial in i1 , . . . , in

by induction. So f is a polynomial in i1 , . . . , im .
45
Professor Borcherds showed us an invariant where the first generator took 13 pages to write out.
Someone in the 19th century had a lot of spare time.
46
Reynolds actually studied fluid dynamics. He showed that fluid flow averaged over time was a group.

91
 The following example illustrates the reason we need to be careful about showing that
i1 , . . . , ik generate I.

Example 22.6. Let R = C[x, y], and take the subring containing the ideal generated by
x and 1. This subring is not finitely generated as a ring.

Example 22.7. Let G = Z/nZ act on C[x, y]. Suppose that G is generated by σ, where
σ n = 1. Let σ(x) = ζx and σ(y) = ζy, where ζ = e2πi/n . The ring of invariants
is the polynomials with all terms of degree 0, n, 2n, . . . . A set of n + 1 generators is
xn , xn−1 y, xn−2 y 2 , . . . , y n . If we call these an , an−1 , . . . , a0 respectively, there are many
relations between the ai . For example, an an−2 = a2n−1 .
Are the collection of syzygies finitely generated? Yes. The ring of invariants is given
by a polynomial ring in generators a0 , . . . , an mod the ideal of syzygies. So the ideal of
syzygies is finitely generated by Hilbert’s theorem.

92
23 Formal power series
23.1 Definition, inverse limit, and multiplicative inverses
Definition 23.1. Let R be a ring. The ring of Formal power series, R JxK, is the ring of
series of the form
a0 + a1 x + a2 x2 + · · ·
with ai ∈ R.

When we say “formal,” we mean that we don’t care about convergence. So these usually
do not define a function.

Example 23.1. Consider the formal power series in C JxK

1 + 1!x + 2!x2 + 3!x3 + · · · .

This only converges for x = 0.

R JxK is the inverse limit of the rings R[x]/(xn ), the polynomial rings truncated at
degree n. The homomorphism R[x]/(xn ) → R[x]/(xn−1 ) just removes the xn term. We
also say that R JxK is the completion of R at the ideal (x). More generally, we can take
lim R/I n for any ideal I.
←−
Example 23.2. The map R → lim R/I n need not be injective. Let
←−
R = C[x1/n , all n > 0],

I = (x1/2 , x1/3 , x1/4 , . . . ).

So R/I n = R/I = C for all n, which makes lim R/I n = C.
←−
We also consider R Jx1 , x2 , . . . , xn K, the ring of formal power series in n variables. This
is just the ring defined recursively as R Jx1 , . . . , xn−1 K Jxn K.

Proposition 23.1. Let K JxK be a field. Suppose that f (x) = a0 + a1 x + · · · with a0 6= 0.

Then f has an inverse.

Proof. Put a0 = 1 for simplicity. Then f (x) = 1 + g(x), where g(x) = a1 x + a2 x2 + · · · .

Then
1/f = 1/(1 + g) = 1 − g + g 2 − g 3 + · · · ,
which makes sense because the coefficient of xn is a finite sum for every n.

Example 23.3. Let f = 1 + x + x2 . The inverse is 1 − x + x3 − x4 .

93
23.2 Ideals of R JxK
Proposition 23.2. The only ideals of K JxK are (0), (1), and (n) for n ≥ 1.

Proof. Any element an xn + an+1 xn+1 = xn (an + an+1 x + · · · ) = xn u for a unit u.

Corollary 23.1. K JxK is a PID, and a UFD.

What about K Jx, yK? This is not a principal ideal domain because it has the nonprin-
cipal ideal (x, y). However, we have the following result.

Theorem 23.1. If R is Noetherian, so is R JxK.

Proof. This is similar to the proof for polynomials. Let I be an ideal. Let In be the
ideal of the coefficients of xn in series with smallest term xn . Then I0 ⊆ I1 ⊆ I2 ⊆ · · · .
This stabilizes because R is Noetherian. Each of these is finitely generated, so I is finitely
generated.

Corollary 23.2. If R is Noetherian, so is R Jx1 , . . . , xn K.

Proof. Induct on n.

23.3 Unique factorization

Recall that K[x1 , . . . , xn ] is a UFD. We want to prove the corresponding fact for formal
power series. But this is not as straightforward to prove.
A bad attempt would be to try to show that if R is a UFD, so is R [x]; this is not true
in general.47
If we try to copy the proof for R[x], we need to define the content of a formal power
series. But this need not exist.

Example 23.4. Let R = Z and f = 1 + x/p + x2 /p2 + x3 /p3 + · · · , where p is prime. Then
the content would have to be p−∞ times something.

The following theorem lets us reduce formal power series proofs to polynomial proofs.
We treat the n = 2 case, but the proof for n variables is similar but with more bookkeeping.

Theorem 23.2 (Weierstrass preparation). Suppose f ∈ K Jx, yK, were K is a field. Then
f = ug, where u is a unit, g is a polynomial in y with coefficients in K JxK, and the leading
coefficient is a power of x.
47
Lang made this mistake in a previous version of the book. According to Professor Borcherds, there are
many papers that point out various errors in Lang’s book.

94
Proof. Pick the monomial xm y n so that am,n 6= 0 and if ab,c = 0, then b < m, or b = m
and b < n; this is the same as saying that (m, n) is least in the lexicographic ordering on
the degrees of polynomials with nonzero coefficients.
.. .. .. ..
. . . .
0 a1,0 a2,0 a3,0 ···
0 a1,3 a2,3 a3,3 ···
0 a1,2 a2,2 a3,2 ···
0 0 a2,1 a3,1 ···
0 0 a2,0 a3,0 ···

By multiplying by units, 1 + cxi y j , we can make the coefficients of every term xm y k zero
for k > n; we can do this infinitely many times because the infinite product just defines a
power series.
.. .. .. ..
. . . .
0 0 ∗ ∗ ···
0 0 ∗ ∗ ···
0 a1,2 ∗ ∗ ···
0 0 ∗ ∗ ···
0 0 ∗ ∗ ···
We can then kill all the coefficients xm+1 y k with k ≥ 1. Similarly, kill off the other
coefficients of x` y k with k ≥ m.
.. .. .. ..
. . . .
0 0 0 0 ···
0 0 0 0 ···
0 a1,2 0 0 ···
0 0 ∗ ∗ ···
0 0 ∗ ∗ ···

So f is a unit times xm y n + bi,j xi y j with i ≥ m + 1 and j ≤ m. Note that we have

P
to kill all the coefficients in this order; if you kill xi y j before you kill xi−k y j−` , when you
kill xi−k y j−` , you might make xi y j nonzero.

It turns out that the Weierstrass preparation theorem is what we needed.

Theorem 23.3. K Jx1 , . . . , xn K is a UFD.

Proof. We will treat the case of n = 2, R Jx, yK. We first show that every element has a
factorization into irreducibles. The proof we gave for R[x] works for any Noetherian ring,
and R JxK is Noetherian.

95
 To prove uniqueness, the key step is to show that irreducible elements are prime. Ir-
reducible means that g 6= gh with g, h not units, and prime means that if f divides gh,
then f divides g or h. This follows from the Weierstrass preparation theorem. Suppose
that f divides gh; we can assume f, g, h are polynomials in y with coefficients in K JxK.
By induction, K JxK is a UFD, so K JxK [y] is a UFD since it is a polynomial ring over a
UFD. So f divides g or h in K JxK [y] and hence in K JxK JyK.

Example 23.5. Let f (x, y) = y 2 − x2 − x3 . This is irreducible as a polynomial in K[x, y],

but it is not irreducible as a power series in K Jx, yK.
√ √
y 2 − x2 − x3 = (y + x 1 + x)(y − x 1 + x),
√
where 1 + x is the formal power series

√ 1 1
2 · −1
2 2
1+x=1+ x+ x + ··· .
2 2!
Geometrically, the curve y 2 = x2 − x3 only has 1 component. Near 0,√ the curve
looks reducible,
√ however, because it looks like two intersecting curves, y = x 1 + x and
y = −x 1 + x. So this polynomial is reducible in K Jx, yK iff the curve y 2 − x2 = x3 = 0
has two branches near x = y = 0 (the point where the ideal (x, y) vanishes).

23.4 Hensel’s lemma

Lemma 23.1 (Hensel). Suppose f (x, y) ∈ K Jx, yK, and suppose the smallest nonzero
coefficients are of degree d and form a polynomial fd (x, y). Suppose that fd (x, y) =
g(x, y)h(x, y) with g, h coprime. Then f (x, y) = G(x, y)H(x, y), where g and h are the
smallest degree terms of G and H, respectively.

We will not prove this. Instead, here are some examples.

Example 23.6. Let f (x) = y 2 − x2 − x3 . Then d − 2 and f2 = y 2 − x2 . So

f y 2 − x2 = (y − x)(y + x),

which lifts to
√ √
y 2 − x2 − x3 = (y − x 1 + x)(y − x 1 + x) = (y − x + · · · )(y + x + · · · ).

Example 23.7. Let f (x) = y 2 −x3 . Then d = 2 and fd = y 2 = y ·y. However, y 2 −x3 does
not factorize! This is because x3 has no square root as a formal power series. Geometrically,
y 2 − x3 = 0 looks like a cusp, so we don’t get two different curves around 0.

Here is an analogue of Hensel’s lemma in number theory.

96
Lemma 23.2 (Hensel (number theory version)). Suppose f (x) = (x−a)g(x), and f (x) = 0
around p, where f ∈ Z[x]. If f 0 (x) 6= 0 (mod p) has a root in Zp (f (x) ≡ 0 mod pn for
all n ≥ 1).

Example 23.8. Let f (x) = x2 = 7 and p = 3. Then f (1) = 12 − 7 ≡ 0 (mod 3), and
f 0 (1) = 2 6≡ 0 (mod 3). So x2 − 6 ≡ 0 (mod p)n has a root for all n ≥. We get
√ √
x2 − 7 = (x − 7)(x + 7)

Example 23.9. Let f (x) = x2 − 7 and p = 2. f (1) ≡ 0 (mod 2), and x2 − 7 has no roots
(mod 2)3 = 8. And f 0 (1) = 2 ≡ 0 (mod 2).

97
24 Field Extensions
24.1 Field extensions and algebraic elements
Definition 24.1. Let K be a field. A field extension L of K is a field such that K is a
subfield of L. This is written as K ⊆ L or L/K.
Example 24.1. C is a field extension of R.
Definition 24.2. The degree [L : K] of K/L is dim L as a vector space over K.
Example 24.2.
[C : R] = 2.
Definition 24.3. An element α ∈ L is called algebraic over K if α is a root of some
polynomial in K[x].
√
Example 24.3. The real number 5 2 is algebraic over Q, as a root of x5 − 2.
Example 24.4. Neither π nor e is algebraic over Q. The proof of this is hard.
In general, it is difficult to prove whether something is algebraic or not. The following
are still open problems:
1. Is e + π algebraic?
2. Is eπ algebraic?
Example 24.5. Let L = Q(x) be the rational functions in x. Then [L : Q] = ∞, and x is
not algebraic.
Theorem 24.1. α is algebraic over K iff α is contained in a finite extension K1 of K
([K1 : K] < ∞).
Proof. Suppose α ∈ K1 with [K1 : K] = n < ∞. Look at 1, α, α2 , . . . , αn . This is n + 1
elements in an n-dimensional vector space over K, so we get
a1 + a1 α + · · · + an αn = 0,
where ai ∈ K and the ai are not all 0. So α is algebraic.
Suppose that α is algebraic. Then p(α) = 0 for some p ∈ K[x]. We can assume p is irre-
ducible. So K[x]/(p) is a field, K1 . So [K1 : K] = deg(p), with basis 1, x, x2 , . . . , xdeg(p)−1 .
So we get a map K[x]/(p) → L.
x7→α
K[x]/(p) L

K
This map is injective since K[x] is a field, so the image of the map is a field of degree < ∞
containing α.

98
Lemma 24.1. Let K ⊆ K1 ⊆ K2 . Then

[K2 : K] = [K2 : K1 ][K1 : K].

Proof. Let x1 , . . . , xm be a basis of K1 over K, and let y1 , . . . , yn be a basis of K2 over K1 .

Then xi yj form a basis of K2 over K (exercise). So [K2 : K] = mn.

Proposition 24.1. Suppose α, β ∈ L are algebraic over K. Then so are α + β and αβ.
Proof. Say α ∈ K1 with [K1 : K] is finite. β satisfies an irreducible polynomial of degree
n < ∞ over K, so β satisfies an irreducible polynomial of degree ≤ n over K1 . Then β is
algebraic over K, say β ∈ K2 with [K2 : K1 ] < ∞. Then

[K2 : K] = [K2 : K1 ][K1 : K],

so [K2 : K] = [K2 : K1 ][K1 : K] < ∞. α + β ∈ K2 and αβ ∈ K2 , so they are algebraic.

√ √ √
Example 24.6. α = 2 + 3 2 + 5 2 is algebraic. The smallest degree polynomial p(x)
with p(α) = 0 has degree 30.
Example 24.7. All algebraic elements of C over Q form a field.48
In general, we have the following fact.
Proposition 24.2. K[x]/p(x) is a field if p is irreducible.
Proof. This is a quick consequence of a homework problem we have done, and should be
done as an exercise. Use the fact that K[x] is a PID.

Suppose that p is not irreducible. Then for p = f g for some coprime f, g. Then
K[x]/(p) ∼= K[x]/(f ) × K[x]/(g) by the Chinese remainder theorem. So if p does not have
multiple copies of the same factor, K[x]/(p) is a product of fields. If p has multiple copies
of a factor, K[x]/(p) can be strange.
Example 24.8. Let p = xn . Then K[x]/(xn ) is the ring of truncated polynomials of the
form a0 + a1 x + · · · + an−1 xn−1 with xn = 0 and ai ∈ K. This has nilpotent elements, so
it is not a product of fields.
Suppose that p is an irreducible polynomial in K[x]. We can find an extension field
L so that p has a root in L, L = K[x]/(p). Does P factorize into linear factors in L?
Sometimes.
Example 24.9. Let p(x) =√x3 −  2 in Q[x].√ This is√irreducible by Eisenstein’s criterion.
Let L = Q[x]/(x − 3) = Q[ 2] = a0 + a1 3 2 + a2 ( 3 2)2 : ai ∈ Q . Does x3 − 2 factor in
3 3

linear
√ factors in
√ L? It does not. L ⊆ R, and x3 − 2 only has 1 real root. The others are
3
2e2πi/3 and 2e4πi/3 .
3

48
This is called the field of algebraic numbers and is studied in algebraic number theory.

99
Example 24.10. Let p(x) = x4 + 1. This is irreducible; check by sending x 7→ x + 1. We
get x4 + 4x3 + 6x2 + 4x + 2, which is irreducible by Eisenstein. Look at the complex roots:
eπi/4 , e3πi/4 , e5πi/4 , e7πi/4 . So

L = Q[x]/(x4 + 1) ∼ = Q[ζ] = a0 ζ + z1 ζ + a2 ζ 2 + z3 ζ 3 : ai ∈ Q .


In this case, p factors as

p(x) = (x − ζ)(x − ζ 3 )(x − ζ 5 )(x − ζ 7 ).

24.2 Splitting fields

Definition 24.4. Suppose p ∈ K[x] with K ⊆ L. L is a splitting field of p if

1. The polynomial p factors into linear factors in L.

2. L is generated by roots of p.

Example 24.11. Q[ζ] is a splitting field of x4 + 1.

√
Example 24.12. Q[ 3 2] is not a splitting field of x3 − 2.
√
3 − 2. Form Q[ 3 2] =
How do we find a splitting field? Let’s √ find the splitting
√ field
√ 2 of x
Q[x]/(x3 − 2) = K1 . In K1 , x3 − 2 = (x − 3 2)(x2 + 3 2x + ( 3 2) √ ), where√ the latter factor
is in K1 [x]. Add the roots of this to K1 , forming K1 [x]/(x2 + 3 2x + ( 3 2)2 ).
Here is the general construction of the splitting field of p ∈ K[x]: Factor p. If there are
no factors of degree > 1, we are done. Otherwise, pick a factor q, where q is irreducible and
of degree > 1. Form a new field K[x]/(q). Over this field, p has one extra linear factor.
Repeat this with p/q. We get

K ⊆ K1 ⊆ K2 ⊆ K3 ⊆ · · · ⊆ Kn ,

where at degree k, we add the root αk of p/((x − α1 ) · · · (x − αk−1 )). So

[Kn : K] ≤ n!

using our lemma about degrees. So the splitting field has degree ≤ deg(p)!.
The splitting field is essentially unique.

Proposition 24.3. If L1 , L2 are 2 splitting fields of K, L1 → L2 , we can find an isomor-

phism from L1 → L2 , fixing all elements of K.

L1 L2

100
Proof. As before, construct the sequence of field extensions

K ⊆ K1 ⊆ K2 ⊆ K3 ⊆ · · · ⊆ Kn .

Suppose L is a splitting field of K. Then K1 → L because K1 = K[x]/q1 (x), and L is a

splitting field of P . We can form maps Ki → L for each i in this way.

K K1 K2 ··· Kn

Then the image of Kn is all of L since L is generated by the roots of p. So Kn ∼

= L.

This isomorphism is not necessarily unique.

√
Example 24.13. C is the splitting field of x2 + 1 over R. What is −1? It can be i or
−i, depending on which isomorphism you use.

24.3 Application to finite fields

Proposition 24.4. For each prime power pn , there is a unique finite field Fpn with pn
elements.
n
Proof. The main idea of the proof is that Fpn is the splitting field of xp − x.
n
We first show that the splitting field of xp − x has pn elements. This has pn roots
n n
because the derivative is pn xp −1 − 1, which is coprime to xp − x. The key point is is that
the roots form a field (closed under addition and multiplication) because (a + b)p = ap + bp
n
in characteristic p, and because the roots are 0 or roots to xp −1 = 1. So the roots form a
field of order pn .
n
For uniqueness, we want to check that any field of order pn is a splitting field of xp − x.
n
The key point here is that all elements are roots of xp − x. If x = 0, it is a root. If x 6= 0,
n
then x ∈ L∗ (order pn − 1 and is a group), so xp −1 = 1 by Lagrange’s theorem.

Example 24.14. Let’s construct the field of order 24 = 16. We have proved that it exists,
but the abstract proof is useless for construction. Find the irreducible factor p of x16 − x
of degree 4. Form F2 [x]/p. Any field of order 16 is a splitting field; for example F2 [x]/p
for any irreducible p of degree 4. Any irreducible polynomial in F [x] of degree 4 divides
x16 − x. So

x16 − x = (x4 + x + 1)(x4 + x3 + 1)(x4 + x3 + x2 + x + 1)(x2 + x + 1)(x + 1)x.

101
 2 1
Note that 1,2, and 4 are the factors of 4.49 This is divisible by x2 − x and x2 − 1. To
get an explicit construction of the field of order 24 , use F2 /(x4 + x + 1), or quotient out by
your favorite irreducible polynomial of degree 4 over F2 .50
Example 24.15. How many irreducible polynomials are there of degree 6 in F2 [x]? We
have that
6
x2 − x = (irred. polys of deg 6)(irred. polys of deg 3)(irred. polys of deg 2)(x + 1)x.

Using a kind of inclusion-exclusion argument, we get that the degree of the product of
polynomials of degree 6 is 26 − 23 − 22 + 21 . Each polynomial has degree 6, so the number
of polynomials is (26 − 23 − 22 + 21 )/6 = 9.

24.4 Algebraic closure

Definition 24.5. L is called the algebraic closure of K if the following conditions hold:
1. Any element of L is algebraic over K.

2. Any polynomial in L[x] has a root.

Example 24.16. C is the algebraic closure of R.
Proposition 24.5. Any field has an algebraic closure, unique up to isomorphism. More
generally, given any set of polynomials in K[x], we can find a splitting field such that:
1. All polynomials in the set factorize into linear factors.

2. L is generated by the roots of the polynomials.

Proof. Suppose there are a countable number of polynomials p1 , p2 , p3 , . . . . Form

K ⊆ K1 ⊆ K2 ⊆ · · · ,

where Kn is a splitting field for pn over Kn−1 . The union is a splitting field. If we have an
uncountable number of polynomials, use the magic words: Zorn’s lemma. So we have found
L ⊇ K such that all polynomials in K[x] have a root in L; we want that all polynomials
in L[x] have a root in L.
Suppose that p is irreducible in L[x], and form M = L[x]/p(x). Then the coefficients
of p are all in K, so they all lie in some finite extension of K. So α is contained in a finite
extension of K, so α is algebraic over K. This makes α ∈ L since any polynomial in K[x]
splits into linear factors in L.
Uniqueness of the algebraic closure is much like the uniqueness of splitting fields.
49
You may recall that these are the irreducible polynomials we computed in a previous lecture.
50
In general, there is no preferred element to quotient out by. This is troublesome, because the fields you
obtain are technically different, even though they are isomorphic.

102
 It’s difficult to find easy to explain examples of algebraic closures.

Example 24.17. Let K be the field of formal Laurent series over C. This has elements
· · · + a−n z −n + · · · + a0 + a1 z + · · · with ai ∈ C. The algebraic closure is
[
formal Laurent series in z 1/k .
k≥1

These are called Puiseux series.51

51
These date back to Newton, but they are not named after him because no one knew what algebraic
closures were back then.

103
25 Normal, Separable and Galois Extensions
25.1 Normal extensions
Recall that the splitting field L of a polynomial p over K is a field such that all roots of p
are in L, and L is generated by the roots.

Proposition 25.1. L is the splitting field of some family of polynomials (over K) iff any
irreducible p ∈ K[x] splits into linear factors in L.

Proof. Suppose p is irreducible in K[x] and has a root α ∈ L. Look at M , the algebraic
closure of L. Any homomorphism ϕ : K[α] → M extends to a homomorphism ψ : L → M
as M is algebraically closed. But im(ψ) must be L as L is the splitting field of some family
of polynomials; the splitting field is a uniquely determined subfield of M , as it is a subfield
generated by a family. So α is already in L.

Example 25.1.
√ Reducible polynomials need not split into linear factors in L. Let K = Q
and L = Q( 3 2). x3 − 2 has a root in L, but it does not split into linear factors.

Definition 25.1. A finite extension L/K is called normal if existence of 1 root of an

irreducible polynomial p implies that p factors into linear factors.

So L/K is normal iff it its the splitting field of some family of polynomials.

Proposition 25.2. Any degree 2 extension L/K is normal.

Proof. Suppose α is a root of (say) a2 + ax + b = (a − α)(a − β). We have that α + β = −a,

so β = −a − α. So β is already in the field K[α].
√ √ √ √
Example 25.2. Q[ 3 2]/Q is not normal. x3 − 2 = (x − 3 2)(x2 + 3 2x + ( 3 2)2 ).

Example √ 25.3. Normal extensions of√normal√extensions√need not be normal over the base
4
field. Q[ 2]/Q is not normal, but Q[ 4 2]/Q[ 2] and Q[ 2]/Q are.

25.2 Separable extensions

Definition 25.2. A polynomial p is called separable if it has no multiple roots, i.e. if p, p0
are coprime.

Definition 25.3. If L/K is an extension, α ∈ L is called separable if its irreducible

polynomial is separable.

Definition 25.4. A field extension L/K is called separable if all its elements are separable.

Theorem 25.1. L/K is separable if K has characteristic 0.

104
Proof. α is a root of an irreducible p. We have that deg(p0 ) < deg(p), so p, p0 have no
common factors since p is irreducible. So p and p0 are coprime.

Remark 25.1. Why does this only work for characteristic 0? The statement that p, p0
have no common factors does not hold if p0 = 0; in algebra, this does not imply that p is
constant if the characteristic of K is not 0.

Corollary 25.1. Any extension Fq /Fp of finite fields is separable.

Proof. Any element is a root of xq − x. This has derivative −1, so (f, f 0 ) = 1.

Example 25.4. Here is a non separable extension. Look at Fp (t)¡ the rational functions
with coefficients in Fp (contains Fp (tp )). Fp (tp ) ⊆ Fp (t), so t is a root of xp − tp . This
factors as (x − t)p because (a + b)p = ap + bp , so all roots are the same. So t cannot be the
root of any separable polynomial in Fp (tp )[x].

25.3 Galois extensions

25.3.1 Galois extensions and Galois groups
Definition 25.5. An extension is called Galois if it is separable and normal.

Definition 25.6. The Galois group Gal(L, K) of L/K is the group of automorphisms of
L fixing all elements of K.

In a sense, the main point of Galois theory is that Gal(L, K) controls the extension
L/K. So we can reduce facts about fields to facts about groups.

Lemma 25.1. Suppose L/K is an extension of degree n and M/K is any extension. Then
there are at most n ways to define a map L → M that acts as the identity on K.

Proof. Suppose L is generated by α, so L = K[α]. Then α is a root of a polynomial of

degree ≤ n. And f (α) is the root of a polynomial in M . This also have 6= n roots in M ,
so there are ≤ n possibilities for f (α). So there are ≤ n possibilities for f .
Now suppose that L is generated by α, β, γ, . . . . Look at

K ⊆ K[α] ⊆ K[α, β] ⊆ · · ·

There are at most [K[α, β], K[α] ways to extend a map from K[ α] to K[α, β]. So there
are ≤ [K[α] : K][K[α, β], K[α]][K[α, β, γ], K[α, β]] · · · ways to extend a map from K to L.
But this is just [L : K].

So if L/K is an extension of degree n, there are at most N automorphisms of L fixing

all elements of K.

Theorem 25.2. For a finite extension L/K, the following are equivalent:

105
 1. L is the splitting field of a separable polynomial.

2. L is Galois.

3. [L : K] = |G|, where G is the Galois group of L/K.

4. K = LG (the set of elements of L fixed by G).

Proof. (1) =⇒ (2): A splitting field is normal.

(2) =⇒ (3): Look at K ⊆ L ⊆ M , where M is the algebraic closure of K. Look at
maps l → M extending the identity map of K. Since L/K is separable, there are n such
extensions (n = [L : K]). Why? Suppose L is generated by α of degree n (root of p). We
can map α to any root of p in M , and p has n roots as it is separable. We leave the case
where L is not generated by 1 element as an exercise.
L/K is normal, so the image of any map L → M lies in L. So there are ≥ n maps from
L to L fixing K. From our lemma, we have that there are always ≤ [L : K] maps L to L,
so |g| = [L : K].
(3) =⇒ (4): Look at K ⊆ LG ⊆ L. There are ≥ n maps L to L extending LG . So
[L : LG ] ≥ n. But [L : K] = n¡ so K = LG .
(4) =⇒ (1): Let α ∈ L, Look at all conjugates of α under G = Gal(L/K). Look
at (x − α)(x − β)(x − γ) · · · . This is in K[x] as all coefficients are invaraiant under G,
since K = LG . So α is a root of a separabble polynomial as α, β, γ, . . . are distinct. The
polynomial splits into linear facts, which gives us normality.

By our lemma, the third statement means that L is “as symmetric as possible.”
√ √ √
Example 25.5. Take x3 − 2 over Q. This has 3 roots, 3 2, 3 2w, and 3 2w2 , where w is
a cube root of 1. √ √
Let L be the splitting field. Then [L : Q] = 6 because [L : Q[ 3 2]] = 2, and [Q[ 3 2] :
Q] = 3. So G = Gal(L, Q) has order 6 = [L : Q]. It acts as permutations of α, β, γ, so it is
the symmetric group S3 .

Example 25.6. Consider C/R. The Galois group has order 2, and is generated by complex
conjugation x + iy 7→ x − iy, which permutes the roots of z 2 + 1 = 0.

Example 25.7. Consider F16 /F2 . This is the splitting field of x16 − x, so it is Galois. So
the galois grou[ has order 4 = [F16 : F2 ]. What is it?
One element is the Frobenius element52 ϕ, which takes a 7→ a2 . Then ϕ(ab) = ϕ(a)ϕ(b),
and ϕ(a + b) = ϕ(a) + ϕ(b) since (a + b)2 = a2 + b2 in F2 . If a is fiixed by ϕ, then
a2 = a, so a = 1 or 0. So a ∈ F3 . So ϕ generates the Galois group, and ϕ4 = id.
ϕ4(a) = (((a2 )2 )2 )2 = a16 = a. So the Galois group is Z/4Z.
52
According to Professor Borcherds, the ϕ stands for Frobenius, even though Frobenius was German, not
Greek. I can’t tell if this was a joke or not.

106
25.3.2 Galois groups and subextensions
Theorem 25.3. Suppose M/K is a Galois extension with Galois group G. For any subex-
tension L (K ⊆ L ⊆ M ), Gal(M/L) is a subgroup of G. Conversely, any subgroup H ⊆ G
induces a subextension M H , the elements fixed by H.

In effect, we want to prove a bijection between subfields of M containing K and sub-

groups of G. We have a major problem: bigger subfields correspond to smaller subgroups.53
This can really be a source of confusion. Suppose that K ⊆ L ⊆ M , where L, M are
Galois extensions of K. Then Gal(M, K) is bigger than Gal(L, K).

Example 25.8. Let’s find all fields between Q and the splitting field of x3 − 2. Look at
the Galois group S3 . The subgroups of S3 are:

S3
2

3 3
3 h(1 2 3), (1 3 2)i

h(1 2)i h(2 3)i h(1 3)i

3
2 2
2

The subextensions of this splitting field are:

Q
2

3 3
3 Q(w)

√ √ √
Q( 3 2w2 ) Q( 3 2) Q( 3 2w)
3
2 2
2

√ √
Q( 3 2, 3 2w)

The indices of the subgroups will correspond to the degrees of the subextensions.
53
Professor Borcherds has been doing Galois theory for decades, but this still trips him up sometimes.

107
Example 25.9. Let ζ be the a 7th root of unity in C. Then ζ 7 = 1, and ζ 6 + ζ 5 + ζ 4 +
ζ 3 + ζ 2 + ζ + 1 = 0, where this polynomial is irreducible. This is (x − ζ)(x − ζ 2 ) · · · (z − ζ 6 ).
So Q[ζ] is normal of degree 6.
The Galois group has order 6 = [Q[ζ] : Q]. What is it? Suppose that σ is in the
Galois group. Then σ(ζ) is a root of x6 + x5 + x4 + x3 + x2 + x+ 1, so it is ζ k for some
1 ≤ k ≤ 6. Similarly, for τ , τ (ζ) = ζ ` , so στ (ζ) = ζ k` . So the Galoid group is the
group is (Z/7Z)∗ ∼ = Z/6Z, which is cyclic. There are 4 subgroups of orders 1, 2, 3, and 6,
respectively (of index 6, 3, 2, and 1), so there are 4 extension of Q contained in Q[ζ], of
degrees 6, 3, 2, and 1.

108
26 The Fundamental Theorem of Galois Theory
26.1 Proof and an example
Here is the example of the fundamental theorem that we started last time:
Example 26.1. Last time we had L = Q[ζ], where ζ = e2πi/7 . We wanted to find all
subfields of L. This had the Galois group (Z/7Z)∗ , which has subgroups

{1}
2

3
{1, 6}

{1, 2, 4} 3
2

{1, 2, 3, 4, 5, 6}
We should have 2 intermediate fields between Q and Q(ζ), of degree 2 and 3. What
are they?
Let’s find the degree 2 field. The elements are fixed by H = {1, 2, 4}. One fixed element
is a = ζ + ζ 2 + ζ 4 , which is not in Q. What is a? We must find a quadratic equation with
root a.

a2 = ζ + ζ 2 + 2ζ 3 + ζ 4 + 2ζ 5 + 2ζ 6
a2 + a = 2)ζ + ζ 2 + · · · + ζ 6
√ √
So a2 + a + 2 = 0, which makes a = −1+2 −7 . So the degree 2 field is Q[a] = Q[ −7].
Let’s find the degree 3 subfield. Let J = {1, 6}. Look for an invariant element; we choose
ζ + ζ 6 = ζ + ζ −1 . Note that ζ = e2πi/7 = cos(2π/7) + i sin(2π/7). So ζ + ζ −1 = 2 cos(2π/7).
Alternatively, we can find the irreducible equation it satisfies. We have

(ζ + ζ −1 )3 = ζ 3 + 3ζ + 3ζ −1 + ζ −3 .

(ζ + ζ −1 )2 = ζ 2 + 2 + ζ −2 .
Since ζ 3 + ζ 2 + ζ + · · · + ζ −3 = 0, we have that ζ − ζ −1 is a root of x3 + x2 − 2x − 1. The
3 roots of this polynomial are 2 cos(2π/7), 2 cos(4π/7), and 2 cos(8π/7).
Theorem 26.1 (Fundamental theorem of Galois theory). Let M/K be a Galois extension
with Galois group G. Then there is a correspondence of subextensions L of M with sub-
groups H of G given by L 7→ Gal(M/L). and H ⊆ G 7→ M H . Moreover, these maps are
inverses of each other.

109
Proof. We want to show that L = M Gal(M/L) . We have L ⊆ M Gal(M/L) , so it is enough to
show that they have the same size. We show that they have the same index in M .
Similarly, we have that H ⊆ Gal(M : M H )., so to show that they are the same, it also
suffices to show that they are the same size. So the theorem follows if we show:

1. |Gal(M : L)| = [M : L].

2. [M : M H ] = |H|.

The key point is to recall our lemma from last lecture: if K ⊆ L and K ⊆ M , there
are at most [L : K] maps L → M extending the identity map of K.
To prove the first statement, observe that |Gal(M/L)| ≤ [M : L] by the lemma. Now
suppose it is strictly less. Look at K ⊆ L ⊆ M . Byt the multiplicativity of indices, there
are < [L : K][M : L] = [M : K] maps from M → M . But since M/K is Galois, there are
exactly [M : K] maps M → M , which is a contradiction.
The proof of the second statement is similar, and we leave it as an exercise.

26.2 Applications of the fundamental theorem

26.2.1 Construction of a 17-sided regular polygon
We can use Galois theory to prove the existence of a construction of a 17-sided regular
polygon using a ruler and compass.54
−1 17
Example 26.2. We want to construct ζ, where ζ 17 = 1. We have ζζ−1 = 0. Recall that
this was an irreducible polynomial of degree 16. The idea is that we can find intermediate
fields Q ⊆ Q(α) ⊆ Q(β) ⊆ Q(γ) ⊆ Q(ζ). We can construct degree 2 extensions with a
ruler and compass because we can construct square roots with a ruler and compass.
Look at the Galois group (Z/17Z)∗ ∼ = Z/16Z. This has subgroups 0 ⊆ Z/2Z ⊆ Z/4Z ⊆
Z/8Z ⊆ Z/16Z, so we can find the desired field extensions. If we want to find out what
the fields are, we can proceed as earlier. Explicitly, the subgroups are

{0} ⊆ {1, 16} ⊆ {1, 4, 13, 16} ⊆ {1, 2, 4, 8, 9, 13, 15, 16} ⊆ Z/16Z,

so we can find the fixed fields of these subgroups:

Q(ζ), Q(ζ + ζ 1 6), Q(ζ + ζ 4 + ζ 13 + ζ 16 ), Q(ζ 1 + ζ 2 + ζ 4 + · · · ).

26.2.2 Subextensions of a splitting field

√
Example
√ √ 26.3. √
Let’s find all the subextensions
√ of x4 − 2 over√Q. This has
√ the roots 4 2,
4
2i, − 4 2, and 4 2i. We have that [Q( 4 2) : Q] = 4 and [Q( 4 2, i) : Q( 4 2)] = 2, so the
splitting field has degree 8 over Q. If we draw out the roots in the complex plane, we get
54
Gauss became famous as a teenager by becoming the first to give an explicit construction.

110
the vertices of the square. So the Galois group is the group of symmetries of the square,
D8 . Its subgroups are:

{1}

r2 r3 s hrsi r2 s hsi

hri r2 , rs r2 , s

D8

So the subextensions are:

√
4
Q(i, 2)

√ √ √ √ √
Q(i, 2) Q( 4 2(1 + i))) Q( 4 2(1 − i)) Q( 4 2) Q( 4 2i)

√ √
Q(i) Q( 2i) Q( 2)

26.3 Extensions corresponding to normal subgroups and factor groups

In the previous example, The 3 subgroups of order 4 and the first subgroup of order 2 are
normal. The other four subgroups of order 2 come in conjugate pairs. We can see that the
corresponding extensions are normal. This is true in general.

Proposition 26.1. Let H ⊆ Gal(L/K). Then H is normal iff LK /K is a normal exten-

sion.

Proof. H is normal iff all conjugates of H under G are the same as G. L/K is normal iff
all conjugates of L under the Galois group are the same as L.

Suppose L/K is a field extension corresponding to H and is normal. What is the Galois
group of L/K? A standard blunder is to think that it is H, which is actually Gal(M/L).
In fact, Gal(L/K) = G/H. If we have Aut(M ) → Aut(L), the kernel is everything fixing
all elements of L. This is H.

111
26.4 Finding extensions corresponding to a given group
Proposition 26.2. Let G be a finite group. Then there is a Galois extension L/K with
Galois group G.

Proof. First take G = Sn , and let L = Q(x1 , x2 , . . . , xn ), all rational functions in L vari-
ables. Now let K = LSn , the symmetric rational functions. If G is any finite group acting
on any field L, then L/LG is Galois with group G. So L/LSn is a Galois extension with
Galois group Sn . The same works for when G is a subgroup of Sn ; L/LG has Galois group
G. The result follows by Cayley’s theorem, that any finite group is a subgroup of some
permutation group.

This is very hard if you want a specific field K. The following is still an open problem:
“Given a finite group G, is there an extension of Q with Galois group G?”

Example 26.4. Let G = Z/5Z and let ζ 11 = 1. Notice that Q[ζ] has Galois groups
Z/11Z)∗ ∼= Z/10Z, which has the quotient, Z/5Z. Explicitly, if we take the field Q(ζ)Z/2Z ,
then its Galois group is (Z/10Z)/(Z/2Z) ∼
= Z/5Z.

Example 26.5. Let’s find an extension of Q with Galois group S5 (order 120). We take
the splitting field of x5 − 4x + 2. This is irreducible by Eisenstein’s criterion. If you look
at the graph, it has exactly 3 real roots (and hence 2 complex roots). The Galois group
is a subgroup of S5 , the permutations of the 5 roots. The Galois group contains a 5-cycle,
say (1 2 3 4 5), so its order is divisible by 5. The Galois group also contains a transposition
(complex conjugation). A 5 cycle and a transposition generate S5 (exercise). So the Galois
group of this splitting field is S5 .

This example generalizes into the following result:

Proposition 26.3. If p is prime ,we can find an extension L/Q with Galois group Sp .

Corollary 26.1. If G is finite, we can find extiensions L/K of Q with Gal(L/K) = G.

Proof. Let L be the extension with Gal(L/Q) = Sp for some large p. Take G ⊆ Sp , and
let K = LG .

112
27 Examples in Galois Theory and Primitive Elements
27.1 Galois group of an irreducible degree 3 polynomial
Consider an irreducible polynomial x3 + ax2 + bx + c = 0. The Galois group G ⊆ S3 ,
the permutations of the roots. 3 divides the order of the Galois group, so G = Z/3Z, so
Z = S3 .

Example 27.1. Take x3 − 2 over Q. The Galois group is S3 .

Example 27.2. Take x3 + x + 1 over F2 . The Galois group is Z/3Z.

We look at ∆ = (α − β)(β − γ)(γ − α), where α, β, and γ are the roots of the
polynomial. ∆ is fixed by Z/3Z, but changes sign under odd permutations of α, β, γ. If
the Galois group is Z/3Z, ∆ must be in the base field. If the Galois group is S3 , ∆ 7→ −∆
must be an automorphism. We must find if

∆2 = (α − β)2 (β − γ)2 (γ − α)2

has a square root in the base field. This is a symmetric function of α, β, γ, and we can
compute this as
∆2 = −4b3 − 27c2
if a = 0.

Example 27.3. Take x3 − 3x − 1 over Q. ∆2 = 81, which is a square in Q. So the Galois

group is Z/3Z.

27.2 Algebraic closure of C

We have enough tools to provide a mostly algebraic proof of the fundamental theorem of
algebra: that C is algebraically closed.

Theorem 27.1. C is algebraically closed.

Proof. We will use the following facts about R, C:

1. R has characteristic 0.

2. Any polynomial of odd degree over R has a real root (follows from intermediate value
theorem).

3. [C : R] = 2, and every element of C has a square root in C.

113
 Let L be a finite extension of C; we want to show that L = C. We may as well
extend L to a Galois extension (char(C) = 0, so L is automatically separable). So we have
R ⊆ C ⊆ L. Let G = Gal(L/R). We want to show that G has order 2. Fact 2 above gives
us that G has no subgroups of odd index > 1 as R has no extensions of odd degree. Let
H be a subgroup of C, so H has index 2 in G. Fact 3 gives us that H has no subgroups of
index 2 (since C has no extensions of index 2).
Let S be a 2-Sylow subgroup of G. S has odd index, so S = 6 by fact 2. So G = S
has order 2n for some n. So H has order 2n−1 . If n − 1 > 0, H has subgroups of index 2,
which would contradict fact 3, so |H| = 1, and |G| = 2. So C is algebraically closed.

27.3 Primitive elements of separable extensions

Lemma 27.1. Suppose V is a vector space over an infinite field K. Then V is not a union
of finitely many proper subspaces.

Proof. By induction. Let V1 , . . . , Vn be proper subspaces. Choose v no in V1 , . . . , Vn−1 by

induction. Choose w ∈ / Vn . Look at v + kq for k ∈ K. There is at most 1 value of k for
which this is in Vi for any given i. Since K is infinite, we can choose k so that v + kq is
not in any Vj .

Theorem 27.2. If L/K is a finite separable extension, L is generated by 1 element; i.e.

there exists some α ∈ L such that L = K(α).

Proof. There are only finitely many extensions between K and L. Let M be a Galois
extension containing L. Then there ares only finitely many extensions of K in M , as
these correspond to subgroups of the Galois group. Each extension is a vector space over
K. Suppose K is infinite. Then the vector space L is not a union of a finite number of
subspaces, so some element α ∈ L is not in any smaller extension of K. So L = K(α). If
K is finite, then L is finite, so L∗ is cyclic.

Example 27.4. Let Fp (tp , up ) ⊆ Fp (t, u). This has degree p2 because

[Fp (t, u) : Fp (t, up )] = [Fp (t, up ) : Fp (tp , up )] = (p)(p) = p2 .

Every element a of Fp (t, u) generates an extension of degree p or 1. In fact, ap ∈ Fp (tp , up )

for t or u since (x + y)p = xp + y p and (xy)p = xp y p . So this is true for all polynomials
in t, u. So Fp (t, u) is not generated by 1 element, and there are infinitely many extensions
between Fp (tp , up ) and Fp (t, u).
This is an example of a purely inseparable extension. These tend to be very weird and
break your intuition. [Jacobson: in some cases subfields iff subalgebras of Lie algebra]

114
27.4 Primitive elements of extensions with Galois group Z/pZ
Suppose L/K is a Galois extension with Galois group Z/pZ (cyclic). What can we say
√
about L? Suppose K = Q(ζ), where ζ is a primitive p-th root of unity. L = K( p a) for
√ √ √
some a ∈ K. This is a root of xp −a. The other roots are p a, p aζ, p aζ 2 , . . . . Any element
√ √
of the Galois group takes p a to p aζ i for some i. So the Galois groups is a subgroup of
Z/pZ, making it 1 or Z/pZ itself.
Suppose K contains all p-th roots of unity and K has characteristic 6= p. We want to
√
show that L = K( p a) for some a. How do we find this element? Let σ be a generator of
the Galois group Z/pZ, so σ p = 1. The key idea is to look at the action of σ on the vector
space L over K (forget that L is a field). σ is a linear transformation, so we can look at
its eigenvalues and eigenvectors. We hope to diagonalize σ.
σ p = 1, so its eigenvalues are the roots of xp = 1, which are contained in K. Now
let’s find eigenvectors. Pick any v ∈ L. Look at v + σv + σ 2 v + · · · + σ p−1 v, which has
eigenvalue 1. Similarly, v + ζσv + ζ 2 σ 2 v + · · · + ζ p−1 σ p−1 v has eigenvalue ζ −1 . We then
get v + ζ −1 σv + ζ −2 σ 2 v + · · · + ζ −(p−1) σ p−1 v is an eigenvector with eigenvalue ζ = ζ 1−p .
Note that v is the average of these, since v = 1 + ζ + ζ 2 + · · · + ζ p−1 = 0. So L is a direct
sum of p 1 dimensional subspaces, on which σ acts as 1, ζ, ζ 2 , ζ 3 , . . . .
Pick w to be any eigenvector of σ with eigenvalue 6= 1 (so q ∈ / K, where K is an
subspace with eigenvalue = 1). Then σw = ζw, say, which gives σwp = ζ p wp = wp . So
√
wp ∈ K as it is fixed by σ. Put a = wp ∈ K. Then L = K( p a). So we have shown that

Proposition 27.1. If L/K is a Galois extension such that

1. Gal(L/K) = Z/pZ,

2. K contains roots of 1 + x + · · · + xp−1 = 0,

3. K has characteristic 6= p,
√
then L = K( p a) for some a ∈ K.

What if K has characteristic p? Assume that L/K is Galois, [L : K] = p. Again, let

√
σ be a generator of the Galois group. L cannot be of the form K( p a) because xp − a
is inseparable (all roots are the same). So the splitting field is not Galois! Look at the
eigenvalues and eigenvectors of σ on the vector space L. σ p = 1, so σ − 1)p = 0. So σ − 1
is nilpotent and not diagonalizable! The only eigenvalue is 1, and the eigenspace is K.
Nilpotent matrices look something like this:
 
0 ∗ ∗ ∗
 0 ∗ ∗
M = 
 0 ∗
0

115
The eigenvectors of M are no use, but generalized eigenvectors, (M −λ)n = 0, are useful. So
try to find the easiest generalized eigenvector, (σ −1)2 v = 0. This means that (σ −1)v ∈ K,
as it is fixed by σ. So σv − v = a for some a ∈ K and v ∈ L. Changing v to v/a, we get
σv − v = 1. This is the simplest substitute for an eigenvector. Instead of σv = λv, we have
σv = λv + 1. So σv = v + 1, and σv p = v p + 1. Then σ(v p − v) = v p − v, so v p − v ∈ K.
So r is a root of xp − x − b = 0 for some b ∈ K. This is called an Artin-Schrier equation,
the analogue of xp − b. So L = K(v), where v is a root of an A-S polynomial.
Suppose v is a root of xp − x − b = 0 in characteristic p. What are the other roots?

(v + 1)p − (v + 1) − b = v p + 1 − v − 1 − b = v p − v − b = 0

So the other roots are v, v + 1, v + 2, . . . , v + (p − 1). This is p distinct roots. So K(v) is

Galois because it is separable (distinct roots) and normal (given one root, we can find the
others). The Galois group is a subgroup of Z/pZ.
Over characteristic p, there are 2 possibilities:

1. xp − x − b is irreducible, so it is a Galois extension with Galois group Z/pZ.

2. xp − x − b factors into linear factors (b is of the form cp − c for c ∈ K).

Example 27.5. We can apply this to the construction of finite fields. What was the issue
√
with order p2 ? Fp ( p a), a is not a square in Fp , but there is no neat way to write down
a in general. We can choose a choice of irreducible polynomial. What about pp ? In this
case, we can take a root of xp − x − 1. Check that this has no roots over Fp . xp − x = 0
for all x ∈ Fp .

Given a polynomial xn + an−1 xn−1 + · · · + an , a classical

√
problem is to find formulas
2 −b± b2 −4c
for its roots. For example, x + bx + c has roots x = 2 . There are no formulas for
5th degree polynomials; we will show this next time.

116
28 Cyclic Extensions and Cyclotomic Polynomials
28.1 Cyclic extensions
Definition 28.1. A cyclic extension is a Galois extension with a cyclic Galois group.
√
Last time, we determined that a cyclic extension L/K is K[ n a] if the characteristic
does not divide n and K[α] otherwise, where αp − α − b = 0; also note that the former
element is the solution to ap − a = 0. The nice thing about this is that if we know one
root, α, thenwe know other roots (αζ i and α + i, respectively).
Which polynomials can be “solved by radicals”? What we means is that roots can be
written using addition, subtraction, multiplication,
√ and k-th roots. For example, the roots
2 −b± b2 −4ac 55
to a quadratic equation ax + bx + c are x = 2a .
Theorem 28.1. The Galois group is solvable iff roots can be given using radicals and
Artin-Schrier equations (char > 0).
Proof. Suppose an equation is solvable by radicals. Assume that the base field K contains
all roots of 1 we need. Look at K0 ⊆ K1 ⊆ K2 ⊆ · · · ⊆ L, where L is the splitting field of
√
the polynomial. K1 = K0 ( n α1 ). Look at the Galois groups:

G ⊇ G1 ⊇ G2 ⊇ · · · ⊇ 1.

G2 is normal in G1 , and G1 /G2 is cyclic. G has a chain of subgroups, each normal in the
next, and all quotients are cyclic. So G is solvable.
Suppose G is solvable (and K contains all roots of 1). We have

G ⊇ G1 ⊇ G2 ⊇ · · · ⊇ 1,

where Gi is normal in Gi−1 , and Gi−1 /Gi is cyclic of prime order. Look at the fields

K ⊆ K1 ⊆ K2 ⊆ · · · ⊆ L.
|{z} |{z}
=LG1 =LG2
√
Ki+1 /Ki is a cyclic Galois extension, so Ki+1 = Ki ( n αn ) or Artin-Schrier.

Example 28.1. Consider x5 − 4x + 2. The Galois group is S5 , which has order 120. The
only normal subgroups are 1, A5 , and S5 . This polynomial is not solvable by radicals.
Example 28.2. x5 − 2 is irreducible and of degree 5, but it can be solve by√radicals.
The Galois group is solvable. The field extensions look like Q ⊆ Q(ζ) ⊆ Q(ζ, 5 2). The
corresponding groups of the wuotients of the Galois groups are Z/4Z and Z/5Z, which are
cyclic.
55
Mathematicians used to duel for money and prestige, presenting each other with difficult problems to
solve. Cardano came up with a general solution for finding roots of degree 4 polynomials, which became a
valuable asset for him in these duels.

117
Example 28.3. All polynomials of degree ≤ 4 can be solved by radicals (in characteristic
0), the Galois groups is a subgroup of S4 , so it is solvable. We have

S4 ⊇ A4 ⊇ (Z/2Z) × (Z/2Z) ⊇ 1.

28.2 Cyclotomic polynomials

Over Q, the roots of unity are the roots of xn −1 = 0. How does this factor into irreducibles?
Look at x12 − 1. This is divisible by x6 − 1, x4 − 1, x3 − 1, etc., but these have factors in
common.

Definition 28.2. The n-th cyclotomic polynomial Φn (x) is the polynomial with roots the
primitive n-th roots of unity (order exactly n).

Example 28.4. Let’s compute some examples:

n Φn (x)
1 x−1
2 x+1
3 −1
3 x + x + 1 = xx−1
3
4 −1
4 x2 + 1 = xx2 −1
x5 −1
5 x4 + x3 + x2 + x + 1 = x−1
(x6 −1)(x−1)
6 x2 − x + 1 = 3 2
(x −1)(x −1)

Example 28.5. We have to make sure we’re not dividing by factors multiple times, so we
must put an x − 1 in the numerator:

(x12 − 1)(x2 − 1)
Φ12 (x) = = x4 − x2 + 1
(x6 − 1)(x4 − 1)

x12 − 1 = Φ12 (x)Φ6 (x)Φ4 (x)Φ3 (x)Φ2 (x)Φ1 (x).

Example 28.6. Again, we make sure we don’t divide by factors multiple times.

(x15 − 1)(x − 1)
Φ15 (x) = = x8 − x7 + x5 − x4 + x2 − x + 1.
(x5 − 1)(x3 − 1)

If you want to really understand cyclotomic polynomials, try out the following exercise:
Find the smallest n such that Φn (x) has a coefficient not 0 or ±1.56

Theorem 28.2. Φn (x) is irreducible over Q. It’s Galois group is (Z/nZ)∗ .

56
You may have to check n > 100, but do not just do this brute force. You should do small cases and
notice some kind of pattern.

118
Proof. If b is prime, we have proved this using Eisenstein’s criterion. A similar proof works
for prime powers. For general n, we use a different argument. The first key idea is to
reduce (mod p) for primes p. The second key idea is to use the Frobenius map, F (t) = tp ,
where the field has characteristic p; F is an automorphism.
Suppose f is an irreducible factor of Φn (x) (over Q). Form Z[ζ] = Z[x]/f (x). This is
an integral domain, and the quotient field Q(ζ) is generated by a primitive n-th root ζ of 1.
Use Z, not Q to reduce mod p. Z[ζ] contains n distinct roots of xn − 1: 1, ζ, ζ 2 , . . . , ζ n−1 .
Now choose an irreducible factor g(x) of f (x) in Fp (x) (factor f (mod p)). In general,
d
deg g < deg f . The key point is that since xn − 1 has n distinct roots, nxn−1 = dx (xn − 1)
n
and x − 1 are coprime.
Since ζ is a root of g (which is irreducible), ζ p is also a root of g as t 7→ tp is an
automorphism of Fp (ζ). So in Z[ζ], ζ p is also a root of f . Then the map from roots of
unity in Z[s] to roots of unity in Fp [ζ] is bijective. So if p does not divide n, then the roots
of f are closed under the map ζ 7→ ζ p .
Now look at the Galois group of Z[ζ]. Automorphisms take ζ 7→ ζ k for k, n coprime,
so the Galois group is a subgroup of (Z/nZ)∗ . The Galois group contains ζ 7→ ζ p for p, n
coprime, which generate (Z/nZ)∗ . So the Galois group equals (Z/nZ)∗ , so f = Φn (x).

Definition 28.3. A cyclotomic57 field is a field generated by roots of unity.

28.3 Applications of cyclotomic polynomials

28.3.1 Primes modulo n
Theorem 28.3. Suppose n ∈ Z. There are infinitely many primes p > 0 with p ≡ 1
(mod n).58

Proof. The idea is to look at the primes P dividing Φn (a) for some a. Suppose p, n are
coprime. Then all roots of Φn (x) are distinct mod p. So Φn (x) is coprime to Φm (x) in
Fp (x) for m dividing n. So if p | Φn (a), p does not divide Φm (a) for m | n. This says that
if Φn (a) ≡ 0 (mod p), then Φm (a) 6≡ 0 (mod p) when m | n. So if an ≡ 1 (mod p), then
am 6= 1 (mod p) for m | n. So a has order exactly n (mod p), so n divides |(Z/pZ)∗ | = p−1,
so p = 1 (mod n).
So if p | Φn (a), then either p | n or p ≡ 1 (mod n). Suppose p1 , . . . , pk are 1 (mod n).
Choose p dividing Φn (np1 · · · pk ). Φn (x) = 1 + x + · · · , so this is 1 (mod n)p1 · · · pk , so p
does not divide p1 · · · pk . Then p does not divide n. So we have found p, a new prime ≡ 1
(mod n).
57
“Cyclo” means “circle,” and “tomic” means “cut.”
58
Dirichlet proved this for p ≡ a (mod n) for any a coprime to n, but the proof is not as nice. There
seems to be no known way to extend the nice proof to this more general case, which frustrates some people.

119
Example 28.7. Let n = 8. Then Φ8 (a) = a4 + 1. if a = 1, we get 2, which divides 8. If
a = 2, we get 9, which is 1 (mod 8). If a = 3, we get 82 = 41 × 2; 41 ≡ 1 (mod 8), and
2|8.

28.3.2 Galois extensions over Q

Recall the hard problem: given finite G, is G a Galois group of K/Q for some K?
Theorem 28.4. If G is abelian, there exists some K/Q, such that G is the Galois group
of K/Q.
Proof. Write G as a product of cyclic groups:

G = (Z/n1 Z) × (Z/n2 Z) × · · · .

Choose distinct primes p1 ≡ 1 (mod n)1 , p2 ≡ 1 (mod n)2 , · · · . (Z/n1 Z) is a quotient of

(Z/p + 1Z)∗ . So G is a quotient of Z/p1 Z × Z/p2 Z × · · · )∗ = (Z/p1 p2 · · · Z)∗ , which is the
Galois group of xp1 ,...pn − 1. So any quotient G/H is the Galois group of some extension
K/Q.

Here is a type of converse, which we will not prove.

Theorem 28.5 (Kronecker-Weber-Hilbert). If K is a Galois extension of Q with abelian
Galois group, then K ⊆ Q(ζ) for some root of unity ζ.

28.3.3 Finite division algebras

Can we find finite analogues of the quaternions H? This is a division algebra that is a
“non-commutative field.”
Theorem 28.6 (Wedderburn). Any finite division algebra is a field (commutative).
Proof. Recall that any group G is the union of its conjugacy classes, which have sizes
|G| / |H|, where H is a subgroup centralizing a representative element of a conjugacy class.
Let L be a finite division algebra, and let K be its center, a field Fq of order q for some
prime power q. Look at the group G = L∗ , which has order q − 1. Suppose a ∈ G. The
centralizer of a in L is a subfield of order q k for some k, so the centralizer of a in G is a
subfield of order q k − 1 (0 ∈
/ G). So
X qn − 1
q n−1 = q − 1 + ,
q ki −1
i

where the sum is over conjugacy classes of orders > 1. Note that k1 < n.
Now note that q n−1 is divisible n ki −1 ), as k < n.
Q by Φn (q). Alsoi note that so is (q −1)/(q 1
So q − 1 is divisible by Φn (x) = i∈(Z/nZ)∗ (q − ζ ). But observe that |q − ζi | > q − 1 unless
ζ i = 1. So n = 1. So L = K, which makes L commutative.

120
Definition 28.4. The Brauer group is the group of isomorphism classes of a finite dimen-
sional division algebras over a field K with center K.

Example 28.8. The Brauer group of R has 2 elements: R, and H.

If D1 , D2 are division algebras, D1 ⊗K D2 ∼

= Mn (D3 ) for some n, D3 , where D3 is the
product of D1 , D2 in the Brauer group.

Remark 28.1. Wedderburn’s theorem shows that the Brauer group of a finite field is
trivial.

28.4 Norm and trace in finite extensions

Let L/K be a finite extension, and choose a ∈ L. Multiplication by a is a linear transfor-
mation from L → L, where L is viewed as a vector space over K.

Definition 28.5. The trace of a is defined as the trace of a as a linear transformation.

The norm of a is the determinant of a as a linear transformation.

Definition 28.6. The norm of a is the determinant of a as a linear transformation.59

Example 28.9. Take C/R and a = x + iy ∈ C. A basis for C/R is {1, i}. a · 1 = x + iy,
and a · i = −y + ix. So a is given by the matrix
 
x y
.
−y x

So the trace of a is 2x, and the norm is x2 + y 2 .

59
Ignore Lang’s definition. Professor Borcherds thinks it is “silly.”

121
29 Norm and Trace
29.1 Norm and trace of finitely generated extensions
Let L/K be a field extension. The norm and the trace satisfy

N (ab) = N (a)N (b)

tr(a + b) = tr(a) + tr(b),

so we can think of the norm and trace as homomorphisms L∗ → K ∗ under × and L → K
under +, respectively.
Suppose a generates L/K (L = K(a)). a satisfies an irreducible polynomial xn +
bn−1xn−1 + · · · + b0 = 0. What are the trace and norm of a? Choose a basis for L over K,
say 1, a, a2 , . . . , an−1 . Then multiplying by a makes 1 7→ a, a 7→ a2 , . . . . So a is given
by the matrix
0 0 · · · 0 −b0
 
1 0 · · · 0 −b1 
.. .. 
 
 . . . .
0
 . . . . 
.
. . .
 .. .. . . 0 −b

n−2

0 ··· 0 1 −bn−1
The trace is −bn−1 ,and the norm is ±b0 .
Suppose the polynomial has roots a = a1 , a2 , . . . , an in an algebraic closure of L. Then
bn−1 = a1 + · · · + an , and b0 = ±a1 · · · an . So the trace is the sum of the roots of the
polynomial, and the norm is the product of the roots.

Example 29.1. In C/R, we have

tr(z) = z + z̄

N (z) = z z̄.

Suppose we have K ⊆ K(a) ⊆ L. Then N (a) in L is (N (a)(in K(a)))[L:K(a)] and tr(a)

in L is (tr(a)(in K(a))) · [L : K(a)] (exercise).
Suppose L : K is Galois with group G, then other roots are given by σi (a) for σ ∈ G,
so Y
N (a) = σ(a),
σ∈G
X
tr(a) = σ(a).
σ∈G

122
29.2 The integers of a quadratic field
√ √
Recall that Q[ −3] contains the ring Z[ −3],√ which is not a UFD since 4 = 2 × 2 =
√ √
(1 + −3)(1 − 3). It is also contained in Z[ −3+1 2 ], the Eisenstein integers, which is a
UFD.
Given a field L containing Q, what is a “nice” ring in it? The answer is that this is the
ring of algebraic integers in K.

Definition 29.1. The ring of algebraic integers in K is the ring of elements in a field K/Q
that are roots of polynomials in Z[x] with leading coefficient 1.

Proposition 29.1. Let L/Q be a finite extension. Then for α ∈ L, the following are
equivalent:

1. α is algebraically independent (root of xn + · · · = 0).

2. We can find a finitely generated Z-module A in L spanning L so that αA ⊆ A.

Proof. (1) =⇒ (2): Take A to be spanned by 1, α, α2 , . . . , an−1 . Then ααn−1 is a linear

combination of 1, α, α2 , . . . , an−1 .
(2) =⇒ (1): α is a linear transformation of a free Z-module A. α is a root of its
characteristic polynomial, which has leading coefficient 1 and other roots in Z.
√
Suppose L = Q( N ), where N is a squarefree √ integer. We want to find the algebraic
integers in L. The
√ easiest examples are m + n N , for m, n ∈ Z. Sometimes, there are
3+1
others, such as 2 . The key point is that if α is an algebraic integer, so are tr(α) and
N (α). So tr(α), N (α) ∈ Z.
√ n √ o
What are the norm and trace of m + n N ? Choose the basis 1, N for L/Q.
√ √ √
Multiplying
√ by √m makes 1 7→ m and N 7→ m N , and multiplying by n N makes
1 7→ n N and N 7→ nN . So we get the matrix
 
m nN
.
n m

These must be in Z. 2m ∈ Z makes m ∈ Z, so m2 −n2 ∈ Z. So n ∈ Z, as n is squarefree.

The other case is that m ∈ Z + 1/2, so m2 = k + 1/4. We need m2 − nN ∈ Z, which is
1/4 − 4n2 ∈ Z. So 1 ≡ (2n)2 N (mod 4). If N ≡ 2, √
3 (mod 4), we have no
√ solutions. So
we must have N ≡√ 1 (mod 4). The integerrs of Q( N ) are given by Z[ N ] if n ≡ 2, 3
(mod 4), and Z[ 1+2 N ] if n ≡ 1 (mod 4).
The trace gives us a bilinear form on L/K with (a, b) = tr(ab). This is either 0 or
nondegenerate.

123
Example 29.2. Here is an example when (·, ·) is zero. Let K − Fp (tp ) and L = Fp (t).
K ⊆ L and this is an inseparable extension. Any element of L is the root of an equation of
the form xp − a for a ∈ K, where the coefficient of xp−1 = 0. This coefficient in the trace,
so the trace is always 0.

For separable extensions L/K, the trace is not identically 0. This is trivial in charac-
teristic 0 because tr(1) = [L : k] 6= 0.

Definition 29.2. A character of a group G is a homomorphism from G → K ∗ (a “1-

dimensional representation” of G).

Lemma 29.1 (Artin). Suppose G is a group (or monoid) and K is a field. If χ1 , χ2 , . . . , χn

are distinct characters, they are linearly independent; i.e. if

a1 χ1 (g) + a2 χ2 (g) + · · · + an χn (g) = 0

for all g ∈ G, then a1 = a2 = · · · = an = 0.

Proof. Suppose a1 χ1 (g) + a2 χ2 (g) + · · · + an χn (g) = 0 for all g. Pick all ai to be not all
zero and n to be as small as possible. Since χ1 6= χ2 , pick h ∈ G with χ1 (h) 6= χ2 (h). Then

a1 χ1 (gh) + a2 χ2 (gh) + · · · + an χn (gh) = 0

for all g, which means that

a1 χ1 (g)χ1 (h) + a2 χ2 (g)χ2 (h) + · · · + an χn (g)χn (h) = 0.

If we multiply the original relation by χ1 (h), we get

a1 χ1 (g)χ1 (h) + a2 χ2 (g)χ1 (h) + · · · + an χn (g)χ1 (h) = 0

If we subtract these two equations, we get

a2 (χ1 (h) − χ2 (h))χ2 (g) + a3 (χ1 (h) − χ3 (h))χ3 (g) + · · · + an (χ1 (h) − χn (h))χn (g) = 0.

Note that χ1 (h) − χ2 (h) 6= 0. So we have a smaller nonzero linear relation between
χ1 , . . . , χn , which is a contradiction since we chose n to be as small as possible.

Proposition 29.2. For a Galois extension L/K, the trace is not identically zero.

Proof. We have that the trace is tr(a) = σ1 (a) + σ2 (a) + · · · + σn (a) with σi ∈ G. If
tr(a) = 0 for all a, we have a linear relation between σ1 , . . . , σn . This is not possible by
Artin’s lemma. So tr(a) 6= 0 for some a. Separable extensions are similar and we leave
that case as an exercise.

124
29.3 Discriminant of a field extension L/K
Definition 29.3. The discriminant of L/K is the discriminant of the bilinear form (a, b) =
tr(ab) on the vector space L.
Choose a basis {a1 , . . . , an } for L over K. The discriminant is det(B), where Bi,j =
(ai , aj ). This depends on the choice of basis. If {b1 , . . . , bn } is another basis, then some
matrix times A gives a change of basis from a1 , . . . , an to b1 , . . . , bn . The discriminant for
the bases is the discriminant for b1 , . . . , bn times the determinant of A. So the discriminant
is well-defined up to multiplication by squares of K. So disc(L/K) ∈ K ∗ /(K ∗ )2 .
Example 29.3. Suppose L = K(a). What is the discriminant of L/K? The element a
is a root of some irreducible polynomial p(a). Choose the basis 1, a, a2 , . . . , an−1 of L/K.
The discriminant is equal to the determinant of
tr(1) tr(a) tr(a2 ) · · · tr(an−1 )
 

tr(a) tr(a2 ) .. 
 . 

.. ..
. .

Assume L/K is Galois for simplicity. Then tr(ak ) = σ∈G σ(ak ), so we get
P

σ(1 · a2 ) · · · tr(an−1 )
P P P 
σ(1 · 1) σ(1 · a)
P 2) .. 
.
P
 σ(1 · a) σ(1 · a 
 
.. ..
. .
This is the product of the matrices
σ1 (1) σ2 (1) σ3 (1) · · · σn (1) σ1 (1) σ1 (a) σ1 (a2 ) · · · σ1 (an−1 )
  
 ..  .. 
σ1 (a) σ2 (a)
 .  σ2 (1) σ2 (a)
 . 

.. .. .. ..
. . . .
which are transposes of each other.
Recall the Vandemonde determinant
··· 1
 
1 1 1
 a
 2 b c ··· 
2 2

det  a
 b c ···   = ±(a − b)(a − c)(a − d)(a − e) · · · (b − c)(b − a) · · · ,
 . .. .. .
. .

 . . . . 
a n−1 bn−1 c n−1 ···

which is the product of the differences of different variables (where each difference is only
counted once). These are equal because the degrees are the same, and the left side is

125
divisible by a − b (and other terms) as if a = b, the the first two columns are the same. So
they differ up to a constant, which is 1. Q
So the discriminant is the square of the determinant ∆ = ± i<j (σi (a) − σj (a)). So
∆2 is the discriminant of the polynomial p(x). This means that the discriminant of the
field extension is just the discriminant of the irreducible polynomial of a.

29.4 Applications of the discriminant of a field extension

Example 29.4. Look at the fields Q[x]/(x2 +x+1), Q[x]/(x3 −x−1), and Q[x]/(x3 −x+1).
Which are isomorphic? The discriminants are −31, −31, and −23; remember to think of
these as elements of Q∗ /(Q∗ )2 . The third differs from the first two; −23/ − 31 is not a
square in Q∗ . The first two fields are isomorphic; change x to −x.

Example 29.5. Let’s find algebraic integers in L = Q(α), where α2 + α + 1 = 0. Look at

the discriminant of the basis 1, α, α2 . The discriminant is −31. Let A be the Z-linear


span of 1, α, α2 . Suppose B is the set of all algebraic integers. So A ⊆ B. disc(B) =

disc(A) × det(x)2 , where x is the matrix taking the basis of A to the basis of B. The
determinant is the order of the group B/A. Now note that −31 is squarefree. Then
det(x) = 1, so B = A.
√ √
Example 29.6. Take Q( −3) , so α = −3 √
and α3 + 3 = 0. This has discriminant −12,
−3+1
which is not squarefree. We have Z[α] ( Z[ 2 ], so you have to do more work.

Recall that the norm is a homomorphism L∗ → K ∗ . What are the kernel and the image
of this map? These can be quite complicated.

Example 29.7. Look at N : C∗ → R∗ given by N (z) = |z|2 = z z̄. The image is the
positive reals.

Example 29.8. Look at N : Q(i)∗ → Q∗ given by a + bi 7→ a2 + b2 . The image of the norm

is the rational number that are sums of 2 squares. As you can see, this gets complicated,
even in simple cases.

Theorem 29.1. If L, K are finite fields, then N : L∗ → K ∗ is onto.

Proof. Recall60 that the Galois group of L/K is cyclic, generated by the Frobenius element
x 7→ xq , where q = |K|. The Galois group is {} 1, F, F 2 , . . . , F n−1 , where n = [L : K].

N (a) = aF (a)F 2 (a) · · · F n−1 (a)

2 n−1
= aaq aq · · · aq
60
Maybe we should put “recall” instead. Professor Borcherds is unsure whether he actually remembered
to introduce this when we went over finite fields.

126
 n−1 /(q−1)
= aq .

So there are at most q n−1 /(q − 1) elements of norm 1. The image has at most q − 1
elements. The order of kernel times the order of the image is the order of L∗ (q n − 1), so
the kernel and image indeed have order q n−1 /(q − 1) and q − 1, respectively.

What is the kernel of N : L → K? Hilbert showed that if L/K is a cyclic extension

generated by σ, then N (a) = 1 iff a = σ(b)/b for some b ∈ L∗ .

127
30 Hilbert’s Theorem 90 and Galois Cohomology
30.1 Hilbert’s theorem 90
We will begin by proving this oddly named61 theorem we started last lecture.

Theorem 30.1 (Hilbert’s theorem 90). Suppose L/K us cyclic. Then N (a) = 1 iff a =
b/σb for some b ∈ L∗ .

Proof. If a = a/σb, we leave it as an exercise to show that N (a) = 1.

We want to solve aσb = b. Think of aσ as a linear transformation on the vector space
L; we want to find some b 6= 0 fixed by this linear transformation. Does aσ have finite
order? (aσ)2 = aσaσ, so it takes b 7→ aσ(aσ(b)) = aσ(a)σ 2 (b). So (aσ)2 = aσ(a)σ 2 . We
can continue this to get

(aσ)n = aσaσ
|
2
· · · σ n−1 a} |{z}
a{z σ n = 1.
N (a)=1 =1
P
A fixed vector of any G is given by g∈G g(v). So the vector fixed by (aσ) is given by
i ∈ Z(aσ)i (θ) for any θ ∈ L. So b solves the problem, except we do not know that
P
b=
b 6= 0. What is the correct choice of theta? Note that this is

θ + aσ(θ) + (aσ)2 θ + · · · = θ + aσθ + aσ(a)σ 2 (θ) + aσ(a)σ 2 (a)σ 3 (θ)

= (a0 σ 0 + a1 σ 1 + a2 σ 2 + · · · )(θ)

Use Artin’s lemma to get that the σi are linearly independent. We can then find a θ so
that the sum is 0.62

We will see later that this means that H −1 (L∗ ) = 0 for L/K cyclic. Here, H −1 (L∗ ) is
the Tate cohomology group.

30.2 Applications of Hilbert’s theorem 90

Example 30.1. Suppose K contains a primitive n-th root ζ of unity. Take a = ζ. Then
N (a) = ζζ · · · ζ = 1. So a = b/σb for some b. So σ(b) = ζb. This makes σ(bn ) = bn , so
√
bn ∈ K ∗ . So L = K( n ∗).
61
The name comes from Hilbert’s “Zahlbericht” (number report) in 1897
62
Professor Borcherds does not like the way Lang did this proof. Lang pulls out the second expression
out of nowhere. Professor Borcherds says it seems like a “deus ex machina.”

128
Example 30.2. Let’s solve x3 + x + 1 = 0. The discriminant is −31, which is not a square
in Q, so the Galois group of the splitting field of this polynomial over Q is S3 . This is a
solvable group because we have 1 ⊆ Z/3Z ⊆ S3 . This gives us the picture

L 1
3 3

K Z/3Z
2 2

Q(ω) S3

What is K? K is a subfield of L fixed by Z/3Z. S3 acts on α1 , α2 , α3 . Let σ be a

generator of Z/3Z. Then σ maps α1 7→ α2 7→ α3 7→ α1 . K is generated by some α, where α
is fixed by σ, but the elements of S3 are not in Z/3Z. Try α = (α1 − α2 )(α2 − α3 )(α3 − α1 )
(find some polynomial in α1 , α2 , α3 fixed by Z/3Z but not S3 . Now
α2 = (α1 − α2 )2 (α2 − α3 )2 (α3 − α1 )2
so it is in the base field. It is the discriminant of x3 + x + 1, which is
is symmetric in αi , √
−31. So K = Q(w, −31).
Next, we want to describe L in terms of K. L/K is a cyclic extension, so K contains
√
cube roots of 1. So by Hilbert’s theorem 90, L = K( 3 ∗), where ∗ is an eigenvector of σ
with eigenvalue equal to ω. Try α1 + ω −1 σ(α1 ) + ω −1 σ 2 (α1 ) = α1 + ω −1 α2 + ω −2 α3 . Call
this y. Let z = α1 + wα2 + w2 α3 . If we find y, z, 0, we can find α1 , α2 , α3 by linear algebra.
We know that y 3 , z 3 ∈ K and are fixe by σ.Expand these in polynomials in α1 , α2 , α3
to get that y 3 + z 3 = −27 and y 3 b3 =√−27.√ So we get that y 3 and z 3 are roots of
x2 + 27z − 27 = 0. So y 3 , z 3 = 27/2 ± 3 3i/2 −31, which means that y, z are given by
y = −3.04 . . . and z = 0.99 . . . . So α1 = (y + z)/3 ≈ −0.68 . . . .63
Example 30.3. Let’s solve degree 4 equations x4 +bx2 +cd+d by radicals. We will provide a
sketch. Look at the Galois group S4 , which is solvable because 1 ⊆ Z/2Z⊕Z/2Z ⊆ A4 ⊆ S4 .
We will have
M 1
4 4

L (Z/2Z)2
3 3

K A4
2 2
Q(ω, i) S4
63
Why do we put these approximate values? It’s so you can check the answer for yourself!

129
To get to K from Q(ω, i), we will adjoin a square root. Going up the diagram, we will then
adjoin a cube root and then another square root.
Suppose the roots are α1 , α2 , α3 , α4 . Note that α1 + α2 + α3 + α4 = 0. What is L?
It is generated by things fixed under (Z/2Z)2 . We wan to find a polynomial fixed by
(Z/2Z)2 ⊆ §4 . Try y1 = (α1 + α2 − α3 − α4 )2 /4 = −(α1 + α2 )(α3 + α4 ). It has conjugates

y2 = (α1 + α3 − α2 − α4 )2 /4

y3 = (α1 + α4 4 − α2 − α3 )2 /4
If we find y1 , y2 , y3 ,, we can find α1 , α2 , α3 , α4 using some algebra.
y1 , y2 , y3 generate a degree 6 extension of Q(ω, i). The Galois group is S3 = S4 /(Z/2Z)2 .
So y1 , y2 , y3 are the roots of some cubic over Q. In fact, there are the roots of y 3 − 2by 2 +
(b2 − d)yx2 = 0, which you can obtain via some messy algebra.64 We can solve this cubic to
find y1 , y2 , y3 and use those to find the αi .

30.3 Galois cohomology

30.3.1 Exact sequences
No one ever understands Galois cohomology the first time the encounter it.65
Suppose G is a group acting on some module M . Look at

1. M G , the subset of things fixed by G (the invariants of G on M ).

2. MG = M/ {m − gm : m ∈ M, g ∈ G}.

The former of these is the largest submodule of M where G acts trivially, and the latter is
the largest quotient of M where G acts trivially.
Suppose that 0 → A → B → C → 0 is an exact sequence. Act on it by G. Is this
exact? No, we get
0 → AG → B G → C G→0.


Similarly, we get that

0 →A
 

G → BG → CG → 0.

Example 30.4. Take 0 → Z → Z → Z/2Z → 0. with G = Z/2Z acting as −1 on Z. We

get
0 → 0 → 0 → Z/2Z
Z/2Z → Z/2Z → Z/2|Z → 0.
64
Mathematicians tried to find this for degree 5, but it turns out to be a degree 6 polynomial, which
is even worse than what you started with. The underlying fact driving this occurrence is that S5 is not
solvable.
65
Professor Borcherds says that no one ever understands Galois cohomology the first time they encounter
it. He even referred to this section as a “futile attempt” to explain it.

130
 Note that M G = HomZG (Z, M ), where ZG is the group ring of G and HomZG is the
homomorphisms preserving the action of G. So M is a module over ZG. Z is a module
over ZG iwth elements of G acting trivially (g · n = n).
We had earlier in the course that Hom(∗, ∗) does not preserve exactness, but the failure
was controlled by “Ext.” Similarly,

MG = Z ⊗ZG M.

The tensor product does not preserve exactness, but the failure is controlled by “Tor.”
Put H 0 (G, M ) = M G . The zeroth cohomology is HomZG (Z, M ). Put H i (G, M ) =
ExtiZG (Z, M ).
A long exact sequence of Ext gives us that if

0→A→B→C→0

is exact, then so is

0 → H 0 (A) → H 0 (B) → H 0 (C) → H 1 (A) → H 1 (B) → H 1 (C) → H 2 (A) → · · ·

Similarly, put H0 (G, M ) = MG and Hi (G, M ) = TorZG

i (Z, M ). We get

· · · → H1 (C) → H0 (A) → H0 (B) → H0 (C) → 0

So H 1 and H1 control the lack of exactness of M G and MG .

30.3.2 Lang’s definition of cohomology

How does this relate to Lang’s definition? Lang defines the first cohomology group as
follows:

Definition 30.1. A crossed homomorphism is a map G → M sending σ 7→ aσ with

aστ = aσ + σaτ .

This is a homomorphism from G → M except if G acts trivially on M , then this is just

Hom(G, M ) as groups.

Definition 30.2. A principal crossed homomorphism is a crossed homomorphism such

that aσ = b/σb for some fixed b.

Lang defines the first cohomology group as

crossed homomorphisms
H 1 (G, M ) = .
principal crossed homomorphisms

131
30.4 Hilbert’s theorem 90 for all Galois extensions
Theorem 30.2 (Hilbert’s theorem 90). Let L/K is a Galois extension with Galois group
G. Then H 1 (G, L∗ ) = 0.

Proof. We are given aσ ∈ L∗ with aστ = aσ · σaτ (multiply, not add, since we are dealing
with L∗ , which is a multiplicative group). We want to find b with aσ = b/σb for all σ.
What is a crossed homomorphism? Look at σ 7→ aσ σ. This is a linear map L → L, so
στ 7→ aστ στ = aσ σaτ τ = (aσ σ)(aτ τ ). So this map is a homomorphism G to End(L). We
will continue the proof next class.

132
31 Infinite Extensions and Galois Cohomology
31.1 Hilbert’s Theorem 90
Let’s introduce the notation Lang uses for his version of Hilbert’s theorem 90. Let G be a
group and A be an abelian group with G  A.

Definition 31.1. A 1-cocycle of G in A is a family of elements {ασ }σ∈G such that

αστ = ασ + σατ .

Definition 31.2. A 1-coboundary of G in A is a family of elements {ασ }σ∈G such that

there exists a fixed β ∈ A such that ασ = σβ − β for all σ ∈ G.

Theorem 31.1 (Hilbert’s Theorem 90). Let L/K be Galois with Galois group G. Then
H 1 (G, L∗ ) = 1.

Proof. A 1-cocycle gives a twisted action G  L given by σ 7→ aσ σ. So (aσ σ)(aτ τ ) = aστ στ

by the 1-cocycle condition. We want to find b with aσ σb = b for all σ; b is fixed by the
twisted action and b 6= 0. P
Find a fixed vector under G as σ∈GPσv, which is always fixed by G. A fixed vector
under the twisted action is given by b = σ∈G aσ · σv. We want to find v so b is nonzero.
This is possible by Artin’s theorem on the independence of σ, since otherwise, we could
find a nonero linear relation between these homomorphisms equal to 0.

Suppose G is cyclic, and let N (a) = 1 and a = b/σb, where σ generates G. What
is a 1-cocycle? Put a1 = 1, aσ = a, aσ2 = aσ σaσ = aσa, and in general, aσn =
aσ(a)σ 2 (a) · · · σ n−1 (a) = a1 = 1. So N (a) = 1 for this to give a 1 cocycle.
So since N (0) = 1, we get a 1-cocycle as above. Note that a = b/σb iff there is a cocycle
given by aσi = b/σ i b for all i, so a 1-cocycle is a 1-coboundary.

Theorem 31.2 (Hilbert’s theorem 90). H 1 (G, L) = 0, where L is considered as an additive

group.

Proof. As a module over K[H], L is isomorphic to K[G], so it is a free module. L has a

basis of the form {σw : σ ∈ G} for some fixed w; this is a result called the normal basis
theorem.66 This shows that H i (G, L) = 0 for i > 0.

Does H i (G, L∗ ) = 1 for i > 0? No. H 2 (G, L∗ ) is often nonzero. This is related to the
Brauer group. H 1 (G, L∗ ) is related to the Picard group. The Picard group of integers of a
number field is a class group.
Why is Lang’s definition of H 1 as cocycles/coboundaries (aστ = aσ + σ(aτ )) the same
as Borcherd’s definition Ext1Z[G] (Z, M )? Here is a sketch of a proof that they are the same.
66
Professor Borcherds never remembers the proof, so see Lang.

133
 To find Ext(A, B), Take the free resolution of A. So we want a free resolution of Q by
free Z-modules.
Z[G] ⊗ Z[G] ⊗ Z[G] → Z[G] ⊗ Z[G] → Z[G] → 0
These have respective Z-bases

g0 ⊗ g1 ⊗ g2 , g0 ⊗ g1 , g0 , 1

And we can map the basis elements by a map d, which sends a component to the identity.
G acts by acting on each component. You should check that d2 = 0 and that if da = 0,
then a = db for some b.
Now form the exact sequence

← Hom(F0 , B) ← Hom(F1 , B) ← Hom(F0 , B)

where Fi is the free resolution.

Check that d(aσ ) = 0 iff the aσ are a 1-cocycle (exercise). Then {aσ } = d(∗) iff the aσ s
are a 1-coboundary.

31.2 Infinite Galois extensions

We want to look at extensions that are algebraic, normal, and separable.

Example 31.1. Take Q̄/Q, where Q̄ is the algebraic closure.

Suppose L/K is an infinite Galois extension. What does the Galois group look like?
Any automorphism of L gives automorphisms of all finite extensions Li /K. An element
of Aut(L/K) is a set of elements of Aut(Li /K) that are compatible. So Gal(L/K) is the
inverse limit of the groups Gal(Li /K).

Example 31.2. Let K = Fp , and let L = F¯p . L = p≥1 Fpk . We have the following
S
picture:

Fp4 Fp6 Fp9

Fp2 Fp3

Fp

134
So the groups will look like this:

Z/4Z Z/6Z Z/9Z

Z/2Z Z/3Z

Z/Z

So Gal(F̄ /F ) = limn (Z/nZ). This is called the profinite completion of Z.

←−
Definition 31.3. A profinite group is an inverse limit of finite groups

Definition 31.4. The profinite completion of G is

lim G/Gi .
←−
Gi normal
G/Gi finite
Q
This is a subset of G/Gi , with the discrete topology. There is a universal map from
G to a profinite group. The image of G is dense in the Krull topology 67 , so lim G/Gi is a
←−
sort of completion of G.

Example 31.3. Recall that Z/nZ = ∼ Q Z/pni Z, where n = Q pki (by the Chinese remain-
i i
der theorem). Then lim Z/nZ = limk Z/pki i Z = p Zp , the p-adic integers.
Q Q
←− ←− i
For finite extensions, we get a 1 to 1 correspondence between extensions of K in L and
subgroups of Gal(L/K). Is the same true for infinite extensions? No. Suppose α ∈ L.
Look at K(α)/L. The set of things in the Galois group fixing α is closed in the Krull
topology; this is the set of things fixing α in M/K, where M is the normal closure of α. A
subgroup fixing any element α ∈ L is always closed in the Krull topology. So a subgroup
fixing all elements of an extension M is an intersection of closed subgroups and is hence
closed.
Instead, we get a 1 to 1 correspondence between extensions of K in L and closed
subgroups of Gal(L/K). We leave this as an exercise. The proof relies on the theorem for
finite Galois extensions and some bookkeeping.
67
Professor Borcherds expressed his displeasure with the fact that there is a Marvel villain named Krull.

135
Example
S 31.4. Let K = Q, and let L be the cyclotomic extension of Q (Q(all roots of unity).
L = Q(ζn ), where ζn is a primitive n-th root of unity. we get the picture

Q(ζ9 ) Q(ζ6 ) Q(ζ4 )

Q(ζ3 ) Q(ζ2 )

We know that Gal Q[ζn ]/Q = (Z/nZ)∗ . So Gal(Qcycl /Q) is given by the inverse limit of

(Z/9Z)∗ (Z/6Z)∗ (Z/4Z)∗

(Z/3Z)∗ (Z/2Z)∗

(Z/Z)∗

As before, (Z/nZ)∗ = (Z/pki i Z)∗ . So lim(Z/nZ)∗ = p (Z/pki i Z)∗ = p Z∗p . This is equal
Q Q Q
← −
to Z̄∗ , where Z̄ is the profinite completion of the ring Z . Nicely enough, it is abelian.

Example 31.5. Let K = Q and L = Q̄, the algebraic closure of Q. Let G = Gal(Q̄/Q).
G is not known. The abelianization of G is known. This is lim(Z/nZ)∗ = Gal(Qcycl /Q).
We have the exact sequence

0 → Gal(Q̄/Qcycl ) → Gal(Q̄/Q) → Gal(Qcycl /Q) → 0.

What is Gal(Q̄/Qcycl )? This is unknown. There is a conjecture of Shafarevich that this

is isomorphic to the profinite completion of a countable free group. Gal(Q̄/Q) is related
to the Langlands program and “automorphic forms.”68 Part of Andrew Wiles’ proof of
Fermat’s last theorem is about understanding some of the structure of Gal(Q̄/Q).
68
Professor Borcherds says that to understand what automorphic forms are, it takes a semester, and to
understand what “related to” means, it takes a lifetime of study.

136
31.3 Abelian Kummer theory
We want to find abelian extensions of K, given that K has enough roots of unity. Let K̄
be the separable algebraic closure of K, the largest separable extension in the algebraic
closure. Look at
1 → µn → K̄ ∗ → K̄ ∗ → 1,
where µn is the n-th roots of unity in K. This is an exact sequence of groups acted on by
Gal(K̄/K). Take the invariants under Gal(K̄/K).
x7→xn
1 → µn → K ∗ −−−−→ K ∗ → H 1 (G, µn ) → H 1 (G, K̄ ∗ ) → H 1 (G, K̄) → · · · .
| {z } | {z }
=1 =1

where these last two are 1 by Hilbert’s theorem 90. The definition of the first homomology
is the same as for when G is finite, except cocycles must be continuous.
So we get
x7→xn
k ∗ −−−−→ K ∗ → Hom(G, µn ) → 1,
and Hom(G, µn ) = H ∗ /(K ∗ )n , which is cyclic of order n. The kernels of homomorphisms
in this group are isomorphic to subgroups H of G with G/H cyclic and of order dividing
n. This is isomorphic to extensions L of K with Gal(L/K) cyclic and of order n. This is
√
the same as our previous description: cyclic extensions of the form K( n ∗).

31.4 Artin-Schrier extensions

Let L/K be cyclic of order p, where p is the characteristic of K. Then L = K(α), where α
is a root of xp − x − b = 0 for b ∈ K. Rewrite this in terms of infinite extenions and Galois
cohomology. Let K̄ be the separable closure of K. Use
x7→xp −x
0 → Fp → K̄ −−−−−→ K̄ → 0,

the exact sequence of modules acted on by Gal(K̄/K). Take the invariants

x7→xp −x
0 → Fp → K −−−−−→ K → H 1 (G, Fp → H 1 (G, K̄) → H 1 (G, K̄) → · · · .
| {z } | {z } | {z }
=Hom(G,Fp ) =0 =0

H i (G, K̄) = 0 for i > 0 by the normal basis theorem.

So Hom(G, Fp ) = K/ im(xp −x) correspond to normal subgroups of index p in Gal(K̄/K).
which correspond to cyclic extensions of degree p.
What about extensions L/K with group Z/pn Z and n > 1? The answer is to use Witt
vectors; see the exercises in Lang. We get

0 → Z/pn Z → W → W → 0,

where W is the ring of Witt vectors.

137