Algebra 1

Matt Kerr
Lecture note
Algebra I
I. Sets A few preliminaries.
II. Groups Topics include: subgroups, Cayley and Lagrange the-

orems, group actions and Burnside’s lemma, orbits and con-
jugacy classes, cosets, normal subgroups, quotient groups,
homomorphisms, Sylow theorems.
III. Rings Topics include: homomorphisms,integral domains, ide-

als, fields, polynomial rings, Euclidean algorithm, multiplica-
tive group of a finite field, principal ideal domains, unique
factorization domains, Gauss’s lemma, irreducibility tests,
algebraic number rings.
IV. Modules Topics include: Schur’s Lemma, structure theorems

for finitely generated modules over a PID and abelian groups,
canonical forms, endomorphisms.
V. Algebras Topics include: exterior algebras, division algebras,

quaternions, Frobenius and Wedderburn theorems.
Virtual Fall Semester 2020

Contents
I Sets 1
I.A. Relations . . . . . . . . . . . . . . . . . . . . . . . . . 1
I.B. Integers . . . . . . . . . . . . . . . . . . . . . . . . . . 5
I.C. Posets . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
II Groups 9
II.A. Introduction . . . . . . . . . . . . . . . . . . . . . . . 9
II.B. Permutation groups . . . . . . . . . . . . . . . . . . . 12
II.C. Groups and subgroups . . . . . . . . . . . . . . . . . 17
II.D. Cosets and Lagrange’s theorem . . . . . . . . . . . . 22
II.E. Homomorphisms and isomorphisms . . . . . . . . . 26
II.F. Group actions and Cayley’s theorem . . . . . . . . . 34
II.G. Conjugacy and the orbit-stabilizer theorem . . . . . 39
II.H. Cauchy’s Theorem . . . . . . . . . . . . . . . . . . . 48
II.I. Normal subgroups and quotient groups . . . . . . . 53
II.J. Automorphisms . . . . . . . . . . . . . . . . . . . . . 66
II.K. Generators and relations . . . . . . . . . . . . . . . . 70
II.L. The Sylow theorems . . . . . . . . . . . . . . . . . . 77
II.M. Some results on finite groups . . . . . . . . . . . . . 83
II.N. ”Not-Burnside’s” counting lemma . . . . . . . . . . 92
i
III Rings 99
III.A.Examples of rings . . . . . . . . . . . . . . . . . . . . 99
III.B. Ring zoology . . . . . . . . . . . . . . . . . . . . . . . 107
III.C.Matrix rings . . . . . . . . . . . . . . . . . . . . . . . 111
III.D.Ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . 118
III.E. Homomorphisms of rings . . . . . . . . . . . . . . . 127
III.F. Fields . . . . . . . . . . . . . . . . . . . . . . . . . . . 136
III.G.Polynomial rings . . . . . . . . . . . . . . . . . . . . 142
III.H.Principal ideal domains . . . . . . . . . . . . . . . . 152
III.I. Unique factorization domains . . . . . . . . . . . . . 158
III.J. Greatest common divisors . . . . . . . . . . . . . . . 163
III.K.Gauss’s lemma and polynomials over UFDs . . . . . 172
III.L. Algebraic number rings . . . . . . . . . . . . . . . . 177
IV Modules 191
IV.A.Definition and examples . . . . . . . . . . . . . . . . 191
IV.B. Submodules and homomorphisms . . . . . . . . . . 199
IV.C. Modules over a PID . . . . . . . . . . . . . . . . . . . 211
IV.D.Applications to linear algebra . . . . . . . . . . . . . 232
IV.E. Endomorphisms . . . . . . . . . . . . . . . . . . . . . 240
V Remarks on Associative Algebras 249

V.A. Algebras over a field . . . . . . . . . . . . . . . . . . 249
V.B. Finite-dimensional division algebras . . . . . . . . . 254
ii
I. Sets
I.A. Relations
Recall that a set S is a collection of elements. If its order |S| (i.e.,

the number of elements) is finite, we may list its elements as in S =
{s1 , . . . , sn }; alternatively, we may write s1 ∈ S to say “s1 is an ele-
ment of S”. A collection T of some elements of S is called a subset,
written T ⊂ S. A proper subset T ( S is one which is not S itself. The
empty set ∅ contains no elements and is a subset of every set.
If S is the union of subsets {Si }i∈ I ,1 we will write S = i∈ I Si .
S
When these sets are disjoint (viz., Si ∩ S j = ∅ for i 6= j), writing

instead S = äi∈ I Si conveys that information. If the Si are also
nonempty, then this defines a partition of S.
Let S, T be sets. A map (or mapping, or function)
f: S→T
is a rule associating to each s ∈ S an element f (s) ∈ T; it has graph
Γ f = {(s, f (s)) | s ∈ S}.
A subset Γ ⊂ S × T is the graph of some map if and only if


∀s ∈ S ∃t ∈ T such that (s, t) ∈ Γ,

and
(s, t), (s, t0 ) ∈ Γ =⇒ t = t0 .


We say that:
• f is injective (written f : S ,→ T) if f (s) = f (s0 ) =⇒ s = s0 ;
• f is surjective (written f : S T) if f (S) = T; and
1Here I is called an index set; here, to each element of I there is associated a subset
Si ⊂ S.
1
2 I. SETS
∼
=
• f is bijective (written f : S → T) if f is injective and surjective
∼
=
(and we define its inverse map f −1 : T → S to send each f (s) 7→ t).
Composition of maps
f g
S / T /6 U
g◦ f (or g f )
is inherently associative. When inverses exist, we have
( g ◦ f ) −1 = f −1 ◦ g −1 .
I.A.1. D EFINITION . (i) A relation on S is a subset
∼ ⊂ S × S.
If ( a, b) ∈ ∼, then we write “a ∼ b”.
(ii) ∼ is an equivalence relation if

(reflexivity) a ∼ a

(symmetry) a ∼ b =⇒ b ∼ a

(transitivity) a ∼ b and b ∼ c =⇒ a ∼ c

hold for all a, b, c ∈ S.
I.A.2. E XAMPLES . Here are some random equivalence relations.

(i) P = set of all people; p1 ∼ p2 ⇐⇒ p1 , p2 reside in same country.2
(ii) “=” on any set S.
(iii) On R2 := R × R, p ∼ q ⇐⇒ p, q are equidistant from (0, 0);
or p ≡ q ⇐⇒ p, q lie on the same horizontal line.
(iv) Write N = {0, 1, 2, . . .} for the natural numbers. On N2 , say
( a, b) ∼ (c, d) ⇐⇒ a + d = b + c.
(v) On the integers Z, define “≡” (or “≡ (mod n)”) by
(n)
a ≡ b ⇐⇒ n | a−b.
(n)
(vi) Given f : S → T, define (on S)
a ∼ f b ⇐⇒ f ( a) = f (b).
2Note: “ ⇐⇒ ” means “iff” (i.e., if and only if). We use it here to define things.
I.A. RELATIONS 3
I.A.3. N ON - EXAMPLES . Here are some relations which are not

equivalence relations.
(i) On P , p1 ∼ p2 ⇐⇒ p1 , p2 are cousins.
(ii) On R, N, Q: >, ≥.
(iii) On Z, a ∼ b ⇐⇒ a relatively prime to b.
(iv) On Z, a | b ⇐⇒ a divides b.
Given an equivalence relation ∼ on S, the ∼-equivalence class

of a ∈ S is
(I.A.4) ā := {b ∈ S | b ∼ a} ⊂ S.
I.A.5. P ROPOSITION . The ∼-equivalence classes yield a partition of S,

and every partition arises in this way.
P ROOF. See Exercise (4) of Problem Set 1.
I.A.6. D EFINITION . (i) The quotient set
S/∼ := { ā | a ∈ S} ⊂ P (S)
is the set of ∼-equivalence classes.3

(ii) The natural map ν : S → S/∼ sends a 7→ ā.
We shall say two sets are isomorphic, written S ∼

= T, if there is a
bijective map between them.
I.A.7. E XAMPLES . Referring to I.A.2,

(i) P /∼ ∼ = set of all countries.
(ii) S/= ∼ = S.
(iii) R /∼ ∼
2
= {circles with center (0, 0)} ∼ = R≥0 ;
and R2 /≡ ∼ = {horizontal lines} ∼ = R.
(iv) N /∼ = Z.
2 ∼
(v) Z/≡ ∼ = {0̄, 1̄, 2̄, . . . , n − 1} is the set of residue classes mod n.
(n)
Finally, given sets S, T, an equivalence relation ∼ on S, and a map

f : S → T, we have:
3Here P (S) denotes the set of subsets of S, called its power set.
4 I. SETS
I.A.8. P ROPOSITION . Suppose that
a ∼ b =⇒ f ( a) = f (b)
for all a, b ∈ S. Then there is a unique f¯ : S/∼ → T such that f¯ ◦ ν = f .
P ROOF. Define f¯ to send ā ∈ S/∼ to f ( a) ∈ T. This is well-

defined (i.e., doesn’t depend on the choice of representative of the
equivalence class), and no other choice makes the diagram commute.

In the scenario of I.A.8, we say that f is well-defined mod(ulo) ∼, or
that the diagram
f
S / T
=
ν ! f¯
S/∼
commutes. As a simple example, consider the map f : Z → {1, −1}
sending n 7→ (−1)n , which is well-defined “mod 4”, i.e. modulo ≡.
(4)
So f¯ : {0̄, 1̄, 2̄, 3̄} → {−1, 1} sends 0̄, 2̄ 7→ 1 and 1̄, 3̄ 7→ −1. Obviously
this works for any other even integer; in particular, if we take ∼ to
be ≡, then f¯ is an isomorphism.
(2)
I.B. INTEGERS 5
I.B. Integers
We turn to some results of Euclid. A prime number p ∈ Z is one

not equal to 0, 1, −1 and whose only divisors are ± p, ±1.
I.B.1. F UNDAMENTAL T HEOREM OF A RITHMETIC . Any natural

number n ∈ N\{0, 1} has (up to order) a unique factorization
n = p1 p2 · · · p s ,
where the { pi } are (positive) primes, which are not necessarily distinct.
P ROOF. We use induction (n = 1 is clear). Assume the statement

holds for all n < m. Then m has a prime factorization: either it is
itself prime, or factors into m1 m2 with m1 , m2 < m.
As for uniqueness: if m = p1 · · · ps = q1 · · · qt with p1 = q1 , this
follows from induction. If instead p1 < q1 , then t > 1 (since q1 is
prime and m isn’t) and
1 < n0 := p1 ( p2 · · · ps − q2 · · · qt ) = (q1 − p1 )q2 · · · qt < m.

| {z }
m
Factoring the parentheticals into primes, the inductive hypothesis

says that the resulting factorizations of n0 must be the same (up to
order). So we either have
p1 | (q1 − p1 ) =⇒ p1 | q1 =⇒ p1 = q1 ,
which is a contradiction, or p1 is one of the q2 , . . . , qt . Reordering

puts us back in the p1 = q1 case.
I.B.2. P ROPOSITION . There are infinitely many primes.
P ROOF. Suppose p1 , . . . , ps is a complete list of positive primes;

then none of them divide p1 · · · ps + 1, contradicting I.B.1.
The FTA leads to the notion of the gcd (= greatest common divi-
sor) of m, n ∈ Z, written (m, n) and well-defined up to sign. To find
it, one traditionally employs the
6 I. SETS
I.B.3. D IVISION A LGORITHM . Given a, b ∈ Z, b 6= 0, there exist

q, r ∈ Z such that
0 ≤ r < |b| and a = bq + r.
P ROOF. We may assume b > 0; then M := {bn | n ∈ Z, bn ≤ a}

is nonempty and bounded above, hence4 has a largest element bq. So
a = bq + r (for some r ≥ 0) and b(q + 1) > a, from which b > r.
To find (m, n), we write as in I.B.3


 n = q0 m + r0

 m = q1 r0 + r1



r0 = q2 r1 + r2



 r1 = q3 r2 + r3
..


.

in which the gcd is the last nonzero remainder ri .5 This is best cov-
ered and proved later in a more general context (that of principal ideal
domains). For now, we shall just show:
I.B.4. P ROPOSITION . (m, n) = mu + nv for some u, v ∈ Z.
P ROOF. Let I := {mx + ny | x, y ∈ Z}, with least positive ele-

ment d = mu + nv ∈ I ∩ Z>0 . Writing m = dq + r (with 0 ≤ r < d),
one finds
r = m − dq = m − (mu + nv)q = m(1 − uq) − n(vq) ∈ I.
For this not to contradict leastness of d, we must have r = 0 and

thus d | m. Similarly, d | n. Moreover, any e dividing both m and n
divides d, which is therefore maximal among common divisors.
4This the the well-ordering principle; it is equivalent to the principle of induction.

5The idea: (n, m) = (n − q m, m) = (r , m) and so on. You eventually reach
0 0
(ri−1 , ri ), with ri−1 = qi+1 ri .
I.C. POSETS 7
I.C. Posets
I.C.1. D EFINITION . A partial order on a set S is a relation “≤”

such that 
x ≤ x

x ≤ y and y ≤ z =⇒ x ≤ z

x ≤ y and y ≤ x =⇒ x = y

for all x, y, z ∈ S. The pair (S, ≤) is called a poset.
An easy example is (P (S), ⊂).
I.C.2. D EFINITIONS . Let (S, ≤) be a poset.

(i) (S, ≤) is totally ordered ⇐⇒ x ≤ y or y ≤ x (∀ x, y ∈ S).
(ii) A chain is a subset C ⊂ S such that (C , ≤) is totally ordered.
(iii) An upper bound6 for a subset S0 ⊂ S is x ∈ S such that
y ∈ S0 =⇒ y ≤ x.
(iv) A maximal element7 of S is x ∈ S such that
x ≤ y and y ∈ S =⇒ x = y.
I.C.3. Z ORN ’ S L EMMA . If every chain in S has an upper bound, then

S has a maximal element.
This is needed for:

• ∃ of bases for ∞-dimensional vector spaces (i.e. a linearly inde-
pendent subset contained by no proper linear subspace);
• ∃ and ! of the algebraic closure of a field;8
• ∃ of transcendence bases for arbitrary field extensions;
• ∃ of maximal (proper) ideals containing a given proper ideal (for
rings with uncountably many elements); and
• (in analysis) stuff like the Hahn-Banach extension.
Zorn’s Lemma follows from (indeed, is equivalent to) the
6These need not exist or be unique in general: consider various subsets S0 ⊂ R of
the reals.
7This need not satisfy y ≤ x ∀y ∈ S, unless of course S is totally ordered.
8In mathematics, the symbol “!” stands for “unique” (or uniqueness, or uniquely).
8 I. SETS
I.C.4. A XIOM OF C HOICE . Given a family of nonempty sets {Xi }i∈ I ,

there exists a “choice function” f defined on I such that f (i ) ∈ Xi (∀i ).
Alternately, ∃ f (= { f (i )}i∈ I ) ∈ ∏i∈ I Xi – that is, the Cartesian product
is nonempty.
(Clearly, this is only needed when I is infinite.) People make a fuss

about using it because it renders your argument nonconstructive.
S KETCH OF PROOF THAT A O C =⇒ ZL. Let (S, ≤) be a poset in
which all chains have an upper bound (write “UB”). For each x ∈ S,
set
ϕ( x ) := {y ∈ S | y > x } ∈ P (S) ,
and assume no x is maximal (i.e. no ϕ( x ) = ∅). By I.C.4, there exists a
choice function f on ϕ(S) (a subset of P (S)), with f ( ϕ( x )) ∈ ϕ( x ).
Clearly x < f ( ϕ( x )).
Now, fixing x ∈ S, define a “sequence” in S by transfinite9 recur-
sion:
(
x0 := x,
xα+1 := f ( ϕ( xα )) (> xα ) for any ordinal number α
and more generally (since this won’t work for limit ordinals)
xα := f ( ϕ(UB{ x β | β < α})).
This “goes on forever”, so that α 7→ xα yields an injection Ord ,→ S

— which is impossible because Ord is not a set.
This was just to give an idea; if you want more than that, pick up
Halmos’s “Naive set theory” book.
9There are arguments that avoid transfinite induction, but they take longer to even
partially understand. You can think of an ordinal number as an isomorphism class
of well-ordered sets (which are totally ordered sets each of whose subsets has a
least element). The class Ord of ordinal numbers is not a set – it is “too big”.
II. Groups
II.A. Introduction
A group is a set G with a binary operation, i.e. a map1
•: G × G → G ,
satisfying

(i) [associativity] ( x · y) · z = x · (y · z)

(II.A.1) (ii) [identity] ∃ “1” ∈ G s.t. 1 · x = x = x · 1
(iii) [inverses] ∃“x −1 ” ∈ G s.t. x −1 · x = 1 = x · x −1


for all x, y, z ∈ G. Associativity means that there is no need for

parentheses in a product like a · b · c · d · e. The operator • is not
in general commutative; when it is, a group is said to be abelian,
and (only then) you will sometimes encounter the notation +, 0, − x
in lieu of •, 1, x −1 . If we drop hypothesis (iii) above, then (i)-(ii) de-
fine a monoid. We will often write groups and monoids in the form
“(set, binary operation, identity element)”, e.g. “( G, •, 1G )”.
Continuing our overview, a homomorphism is a map of groups
(or monoids) — i.e. of the underlying sets —
ϕ: G → H
respecting the binary operation
ϕ( x · y) = ϕ( x ) · ϕ(y) (∀ x, y ∈ G )
and with ϕ(1G ) = 1 H . (In the case of groups, ϕ( x −1 ) = ϕ( x )−1
follows at once.)
1We’ll use a large dot “•” when talking about the operation, and a small dot “·”
when using it.
9
10 II. GROUPS
Groups abound in mathematics and physics, e.g. via rotational

symmetries of a polyhedron or permutations of n particles. Not that
people in physics always liked this: Pauli talked about the “Grup-
penpest”. Here are some more interesting examples of this pest:
(1) Galois theory. By considering the structure of [roughly] the group

of permutations of the roots of a polynomial, one arrives at:
• the insolubility of a general quintic equation by radicals;
• the impossibility of trisecting an angle with a straightedge and
compass;
√
• the impossibility of “duplicating the cube” (constructing 3 2
using a unit grid).
(2) Quantum physics. The manner in which different atomic states
of an electron (eigenfunctions of the Schrödinger operator) come
packaged has to do with the irreducible representations of the
symmetry group O(3).
(3) Topology of manifolds. The (nonabelian) homotopy and the
(abelian) homology groups of a manifold can be utilized to de-
termine (for example):
• the impossibility of a continuous embedding of one manifold
into another; or
• the impossibility of giving a smooth hairstyle to a sphere.
(4) Diophantine equations. Integer solutions of algebraic equations
(e.g. x2 − y2 d = ±1) sometimes have a group structure — mean-
ing that by “taking powers” of one solution you get further solu-
tions. (Similar phenomena arise in algebraic geometry over the
complex numbers or “finite fields”.)
As to what the “representation theory” mentioned in (2) is all

about: suppose you have a group G of linear transformations of a
vector space V [or permutations of a set X]. This can profitably be
separated into two concepts: (i) the abstract group G; and (ii) a ho-
momorphism from that group into GL(V ) [or SX ] — called a group
representation [or group action]. There are extensive classification
results for abstract groups and their representations; one uses these
II.A. INTRODUCTION 11
to “recognize” G and to list possibilities for the representation (and

hence e.g. for the decomposition of V under the original set of trans-
formations). We’ll see some of the classification results for (finite
ϕ
and/or abelian) groups and of the theory of group actions G → SX
soon, and representation theory in Algebra II.
You are no doubt familiar with the noncommutativity of matrix
multiplication: for instance,
! ! ! ! ! !
1 1 1 0 2 1 1 1 1 0 1 1
= 6= =
0 1 1 1 1 1 1 2 1 1 0 1
is a computation in the infinite2 group GL2 (Q). Here is an example

in a finite group. Take a square sheet of paper, write “TOP” at the
top; then
• rotate 90◦ counterclockwise (“r”’),
• flip about its horizontal axis (“h”),
• rotate 90◦ clockwise (“r −1 ”), and
• flip again about the horizontal (“h−1 ”(= h)).
The end result of your flipping represents the element (∗) := h−1 r −1 hr
of the symmetry group of the sheet. If this group (the dihedral group
D4 , with elements 1, r, r2 , r3 , h, hr, hr2 , hr3 ) were abelian, then (∗) would
equal 1. But “TOP” doesn’t reappear at the top, and indeed (∗) = r2 .
2When we say a group is finite or infinite, we are referring to the order of the
group, which means its order as a set (i.e. the number of elements).
12 II. GROUPS
II.B. Permutation groups
Let X be a set; recall that (if finite) its order | X | is the number of
elements. A transformation of X is a map
τ: X → X ;
if τ is bijective (or equivalently, invertible), it is called a permutation.

Let
TX := set of all transformations of X,

and SX := set of all permutations of X.
The binary operation “composition of maps” makes TX into a monoid

and SX into a group, the symmetric group on X.
II.B.1. P ROPOSITION . If | X | = n < ∞, we have |TX | = nn and

|SX | = n!.
P ROOF. For each x ∈ X, there are n choices for τ ( x ); but if τ is to

be bijective, each choice removes an option for the next.
!
x1 ··· xn
Say X = { x1 , . . . , xn }. A useful notation is τ = .
τ ( x1 ) · · · τ ( xn )
II.B.2. E XAMPLE . Let X = { A, B}. We have

( ! ! ! !)
A B A B A B A B
TX = , , ,
A B B A A A B B
( ! !)
A B A B
and SX = , ,
A B B A
where the identity transformation is written first in each set. To re-
move reference to X and think of TX as an “abstract monoid”, write
{1, α, β, γ} for its 4 elements (in the same order) and produce a table
II.B. PERMUTATION GROUPS 13
1 α β γ
1 1 α β γ
α α 1 γ β
β β β β β
γ γ γ γ γ
which displays the abstract binary operation corresponding to the

compositions of these transformations. For instance, αβ = γ (shown
in the table) means, on the level of the transformations, that β fol-
lowed by α gives γ.
You can make such a table for any (finite order) group or monoid;
but conversely, given an arbitrary table of the form
1 α β γ
1 1 α β γ
α α
β β ?
γ γ
it need not yield a monoid: associativity does impose constraints.

Define the nth symmetric group by
Sn := S{1,...,n} .
II.B.3. P ROPOSITION . Any α ∈ Sn has, up to order, a unique com-

plete3 factorization into disjoint cycles (which commute).
II.B.4. E XAMPLE . In S9 , an example of a cycle is (3789), which

sends 3 7→ 7 7→ 8 7→ 9 7→ 3. (It is a 4-cycle because it involves 4 ele-
ments.) This is disjoint from (24) because the subsets of {1, 2, . . . , 9}
3Here, “complete” means that we formally include the 1-cycles (k ) that do noth-
ing, except to say that α sends k to itself, so that each element of {1, . . . , n} appears
exactly once in the product of cycles. (A 1-cycle is really just a way of writing the
identity element.)
14 II. GROUPS
involved are disjoint (which makes them commute). An example of

a (complete) factorization of a permutation into disjoint cycles is
!
1 2 3 4 5 6 7 8 9
= (16)(24)(3789)(5).
6 4 7 2 5 1 8 9 3
P ROOF OF II.B.3. The idea is to induce on the number of ele-

ments in {1, 2, . . . , n} that α moves. Say it moves the element i1 , viz.
i1 7 → i2 7 → i3 7 → · · · 7 → ir ,
α α α α
where r is the smallest integer for which ir ∈ {i1 , . . . , ir−1 }. (Clearly

2 ≤ r ≤ n + 1.)
In fact, we must have ir = i1 . (Otherwise, for some 2 ≤ j ≤ r − 1
we have α(ir−1 ) = i j = α(i j−1 ), and α is not injective, a contradic-
tion.) Hence α moves i1 , . . . , ir−1 in a cycle, and β := α · (i1 · · · ir−1 )−1
(which fixes each of them) moves r − 1 fewer elements than α. We
may view β as a permutation of4 {1, . . . , n} \ {i1 , . . . , ir−1 } and apply
induction to get a complete factorization into cycles. Throwing in
(i1 · · · ir−1 ) then gives the desired factorization of α.
To see the uniqueness, let γ1 · · · γs = α = β 1 · · · β t be two com-
plete factorizations. Since disjoint cycles commute, we may without
loss of generality assume that β 1 and γ1 contain i1 (and that no other
cycles in the two products do). Applying α repeatedly, we get

γ1 (i1 ) = i2 = β 1 (i1 )

..
 .
γ1 (ir ) = i1 = β 1 (ir−1 )

and so β 1 = γ1 . Cancel them and proceed inductively.

A transposition is a 2-cycle (ij); it sends i 7→ j 7→ i and fixes all
other elements.
II.B.5. P ROPOSITION . Any α ∈ Sn factors (nonuniquely) into a prod-

uct of (not necessarily disjoint) transpositions.
4Given sets T ⊂ S, S \ T denotes the set-theoretic complement (the elements of S
that aren’t in T). You can view this β as an element of Sn−r+1 , or (as we do here)
an element of Sn that fixes i1 , . . . , ir−1 .
II.B. PERMUTATION GROUPS 15
P ROOF. Factor α into disjoint cycles, then (for example) factor the
cycles via the formula (123 · · · r ) = (1r )(1 r −1) · · · (13)(12).
For each permutation α ∈ Sn , write c(α) for the number of dis-

joint cycles in its complete factorization,5 and define the sign
sgn(α) := (−1)n−c(α) .
Viewing {1, −1} as a group under multiplication, we have the
II.B.6. T HEOREM . The map sgn : Sn → {1, −1} is a homomorphism

of groups. That is, sgn(αβ) = sgn(α)sgn( β).
P ROOF. First observe that there are n − 1 cycles in the complete

factorization of a transposition τ; e.g., (12) = (12)(3)(4) · · · (n). So
sgn(τ ) = −1.
Writing β = σ1 · · · σc( β) for a complete factorization, consider
( ab) β. Without loss of generality, either (i) a, b occur in σ1 or (ii) a
occurs in σ1 and β in σ2 . Using
( ab)( ac1 · · · ck bd1 · · · d` )σ2 · · · σc( β) = ( ac1 · · · ck )(bd1 · · · d` )σ2 · · · σc( β)
| {z }
σ1
in case (i) and

( ab)( ac1 · · · ck )(bd1 · · · d` )σ3 · · · σc( β) = ( ac1 · · · ck bd1 · · · d` )σ3 · · · σc( β)
| {z }| {z }
σ1 σ2
in case (ii), we either gain or lose a cycle in the complete factorization

of ( ab) β. So for any transposition τ, we have sgn(τβ) = −sgn( β).
Finally, writing α = τ1 · · · τm by II.B.5, we have
sgn(αβ) = sgn(τ1 · τ2 · · · τm β) =
− sgn(τ2 · · · τm β) = · · · = (−1)m sgn( β), and
sgn(α)sgn( β) = sgn(τ1 · τ2 · · · τm )sgn( β) =

− sgn(τ2 · · · τm )sgn( β) = · · · = (−1)m sgn( β),
which completes the proof.
5It is essential to include the 1-cycles in this count!
16 II. GROUPS
II.B.7. C OROLLARY. The “number of transpositions” in α ∈ Sn is

well-defined mod 2.
P ROOF. sgn(α) = sgn(τ1 · · · τm ) = (−1)m , and we know sgn(α)

is well-defined. So m is well-defined mod 2.
The upshot is that we can unambiguously call α “even” or “odd”
according to whether it can be written as a product of an even or
odd number of transpositions. (To see which is the case, one instead
writes the complete factorization into disjoint cycles and computes
sgn(α).)
II.C. GROUPS AND SUBGROUPS 17
II.C. Groups and subgroups
Some further simple properties follow from the defining proper-

ties:
II.C.1. P ROPOSITION . Let G be a group, and a, b, x ∈ G.
(a) The cancellation laws hold: xa = xb (or ax = bx ) =⇒ a = b.
(b) The inverse of x is unique, and ( x −1 )−1 = x.
(c) ( an )m = anm , am an = am+n [laws of exponents]
(d) If a and b commute (ab = ba), then ( ab)n = an bn .
P ROOF. (a) Multiply on the left (resp. right) by x −1 .
(b) If x 0 x = 1 = xx 0 and x 00 x = 1 = xx 00 , then
x 00 = x 00 1 = x 00 xx 0 = 1x 0 = x 0 .
(c) Clear from the definition: an = a · · · a (n times).

(d) If a commutes with b, it commutes with powers of b. Now induce
on n: ( ab)n = ( ab)n−1 ab = an−1 bn−1 ab = an−1 abn−1 b = an bn .
II.C.2. R EMARK . (i) ab = ba is equivalent to the triviality of the
commutator [ a, b] := a−1 b−1 ab. (In algebra, an element being trvial
means it’s the identity element.)
(ii) For monoids: (a) is false, (c) and (d) hold. For those elements
of the monoid that have a (two-sided) inverse, (b) is true. (But those
elements form a group, so this doesn’t say much...)
II.C.3. E XAMPLES . (i) Abelian groups:
• (A, +, 0) where A = Z, Q, R, C.
• (V, +,~0) where V is a vector space.
• (Zn , +, 0̄) where Zn = Z/≡ = integers mod n.
(n)
• (Z∗n , •, 1̄) where Z∗n ⊂ Zn is the subset of elements possessing a
multiplicative inverse: b̄ ∈ Zn such that āb̄(= ab) = 1̄.
• (A∗ , •, 1) where A∗ = Q∗ , R∗ , C∗ (here Q∗ = Q \ {0} etc.).
2πik −1
• ({1, −1}, •, 1), and more generally ({e n }nk= 0 , •, 1).
• rotational symmetries of the (regular) n-gon.
Notes: (a) Z∗n = { ā | ( a, n) = 1}, since (by I.B.4) ( a, n) = 1 ⇐⇒
∃b, k ∈ Z with ab + nk = 1 ⇐⇒ ∃b such that ab = 1̄.
18 II. GROUPS
(b) Zn is an example of a cyclic group, i.e. a group on one gener-

ator: the notation
Zn = h1̄ | n · 1̄ = 0̄i
means that the elements comprise all of the “powers” 0̄, 1̄, 1̄ + 1̄, 1̄ +
1̄ + 1̄, etc. of the generator 1̄, subject to the relation shown (n · 1̄ =
1̄ + · · · + 1̄ [n times] = 0̄). Z = h1i is also a cyclic group (with no
relation), but (unlike Zn ) an infinite one.
(ii) Non-abelian groups:
• Sn = nth symmetric group, for n ≥ 3.

• Dn = nth dihedral group, for n ≥ 3: its elements comprise the n
rotational and n reflectional symmetries of a regular n-gon.
• GLn (A) general linear group, for n ≥ 2 (and A = Q, R, C): ele-
ments are invertible n × n matrices with entries in A.
• SL2 (Z) (integer 2 × 2 matrices with determinant 1) and other “arith-
metic groups”.
Notes: As suggested in (i), it can be useful to write groups in terms

and relations. For instance, for the “quotient of SL2 (Z)
of generators
1 0
by ± 0 1
”,
 !
0 −1
S =

1 0
PSL2 (Z) = hS, R | S2 = 1 = R3 i where 0 −1
!
1 1
!
R =
 = S·
1 1 0 1
says that the elements of PSL2 (Z) are arbitrary “words” in S and R
(and their inverses) subject only to the two relations written. For the
dihedral group, we have
Dn = hr, h | relations are a HW exercise!i
where r is counterclockwise rotation by 2π

n and h is a choice of reflec-
tion. We have also shown that Sn is generated by transpositions.
(iii) Monoids that are not groups:
• (N, +, 0), (Z>0 , •, 1), or (Z\{0}, •, 1).

• (P (S), ∪, ∅) for any nonempty set S.

• (σ, +, (0, 0)) where σ is a cone in R2 :
• the monoid of integral ideals in an algebraic number ring (which

we will meet later).
(iv) Direct products of (monoids or) groups: G1 × G2 , with group

operation ( g1 , g2 ) · (h1 , h2 ) := ( g1 h1 , g2 h2 ).
II.C.4. D EFINITION . A subgroup of G is a subset H ⊂ G satisfy-

ing:
(i) 1G ∈ H;
(ii) [closure under multiplication] x, y ∈ H =⇒ xy ∈ H; and
(iii) [closure under inversion] x ∈ H =⇒ x −1 ∈ H.
We write H ≤ G (or H < G for a proper subgroup — i.e. H 6= G), and
endow H with the operation “•” inherited from G (and hence with a
group structure).
II.C.5. E XAMPLES . (a) When α ∈ G is an element of a group,

we will use the notation hαi := {αn | n ∈ Z} to denote the cyclic
subgroup generated by α. (Though no relation is written, this can
certainly be finite since some power of α may be 1 in G.) Cyclic sub-
groups are clearly abelian.
(b) In Dn , we have cyclic subgroups hr i < Dn (resp. hhi) of order n

2πi
(resp. 2). In C∗ , he n i is the (cyclic) group of nth roots of unity. We
2πi
can intuitively think of he n i and hr i as copies of (Zn , +, 0̄) embed-
ded in C∗ and Dn , but we’ll need to employ homomorphisms and
isomorphisms to state this properly.)
20 II. GROUPS
(c) Intersections of subgroups are again subgroups: given H, K ≤ G,

we have H ∩ K ≤ G. (Why?)
(d) Generalizing (a), we can consider subgroups generated by a sub-

set S ⊂ G, denoted hSi ≤ G. There are three equivalent definitions
of this: as the smallest subgroup of G containing S; as the intersec-
tion of all subgroups containing S; or as all products of (powers of)
elements of S and their inverses.
(e) The centralizer of a subset S ⊂ G is defined by
CG (S) := { g ∈ G | gs = sg (∀s ∈ S)} ≤ G.
(To see that it is a subgroup, rewrite the condition in the braces as

sgs−1 = g. If also sg0 s−1 = g0 , then s( gg0 )s−1 = (sgs−1 )(sg0 s−1 ) =
gg0 , and sg−1 s−1 = (sgs−1 )−1 = g−1 .) In particular, we write CG ( a) :=
CG ({ a}) for the centralizer of one element, and C ( G ) := CG ( G ) for
the center of G. (Often “C” is written “Z” — this is the German her-
itage.)
(f) The cone in II.C.3(iii) is a submonoid of R2 .
(g) A submonoid of TX is called a monoid of transformations of X. A

subgroup of SX is a group of permutations of X. Here is an interesting
example.
Define An ⊂ Sn by
An := {α ∈ Sn | α is even} = {α ∈ Sn | sgn(α) = 1}.
We claim that, since sgn is a homomorphism, this is a subgroup:

indeed, 1 ∈ An ; and given α, β ∈ An ,
(
sgn(αβ) = sgn(α)sgn( β) = 1
sgn(α) = 1 = sgn( β) =⇒
sgn(α−1 ) = sgn(α)−1 = 1
so that (ii), (iii) in II.C.4 hold. This subgroup An ≤ Sn is called the
alternating group.
II.C.6. P ROPOSITION . If n ≥ 3, An is generated by 3-cycles.

P ROOF. α ∈ An =⇒ α is a product of an even number of trans-

positions. We can group these into pairs of distinct transpositions,
viz. α = (τ1 τ2 ) · · · (τ2q−1 τ2q ). For a pair ττ 0 , if the transpositions are
not disjoint, write
(ij)(ik) = (ikj);
while if they are disjoint, write
(ij)(k`) = (ij)( jk)( jk)(k`) = (ijk)( jkl ).

| {z }
1
This recasts α as a product of 3-cycles. (That, conversely, all 3-cycles

belong to An is clear from the first displayed formula.)
22 II. GROUPS
II.D. Cosets and Lagrange’s theorem
II.D.1. D EFINITION . The order of a group G is | G |, its order as a

set. The order of an element a ∈ G is |h ai|, the order of the cyclic
subgroup it generates.
To determine the relation between these orders (in the finite case),
we consider more generally | H | for H ≤ G and introduce (left) cosets
aH := { ah | h ∈ H } ⊂ G.
These are not subgroups.
II.D.2. P ROPOSITION . Distinct cosets are disjoint and have the same
number of elements.
P ROOF. First, we claim that
(II.D.3) aH = bH ⇐⇒ b −1 a ∈ H ⇐⇒ a ∈ bH.
The second “iff” is clear. To see the first, write
b −1 a ∈ H ⇐⇒ ∀h ∈ H, b−1 ah =: h0 ∈ H
⇐⇒ ∀h ∈ H, ah = bh0 for some h0 ∈ H
⇐⇒ aH ⊂ bH,
and similarly
bh ⊂ aH ⇐⇒ a−1 b ∈ H ( ⇐⇒ b−1 a ∈ H since ( a−1 b)−1 = b−1 a).
So if α ∈ aH and aH 6= bH, then (by (II.D.3)) αH = aH 6= bH, hence

(again by (II.D.3)) α ∈/ bH; and we conclude that aH ∩ bH = ∅.
Finally, the map (of sets) H → aH sending h 7→ ah is a bijection by
the cancellation law II.C.1(a).
Notice that what we have established is that
the left cosets are the partition of G formed by the

equivalence relation a ≡ b ⇐⇒ b−1 a ∈ H.
II.D.4. L AGRANGE ’ S T HEOREM . For H < G with | G | < ∞, we have

| H || G |. In particular, the order of any a ∈ G divides G.
II.D. COSETS AND LAGRANGE’S THEOREM 23
|G|
II.D.5. D EFINITION . [ G:H ] := | H | ∈ N is called the index of H in
G, and is the number of cosets (as will be clear from the next proof).
P ROOF OF II.D.4. We can write
G = a1 H q · · · q ar H
as a disjoint union. (Why? Every g is in some coset, namely gH.

Write G = ∪ g∈G gH and strike out repeated cosets. Once there is no
repetition, the remaining cosets are disjoint by Prop. II.D.2.) More-
over, we have that | ai H | = |1H | = | H | for all i (also by Prop. II.D.2).
So | G | = ∑ri=1 | ai H | = r | H |.
II.D.6. E XAMPLES . (a) G = S3 > H = h(12)i = {1, (12)}, (13) H =

{(13), (13)(12)} = {(13), (123)}, and (23) H = {(23), (132)}. Of
course, [ G:H ] = 3.
(b) If we take G = Dn > K = hr i = {1, r, r2 , . . . , r n−1 }, the only other
coset is hK = {h, hr, hr2 , . . . , hr n−1 }; and [ G:H ] = 2.
(c) Suppose p is prime. Since | D p | = 2p, the possible orders of el-
ements are 1, 2, p, and 2p (though in fact, no element of order 2p
exists).
Turning to consequences of Lagrange’s Theorem, first it should

be underscored why we call |h ai| the “order of a”: consider the se-
quence of powers 1, a, a2 , . . . , ak , with k the least power for which
one has a repetition (i.e. ak ∈ {1, a, a2 , . . . , ak−1 }). Then multiplying
ak = ai by a−i gives ak−i = 1, contradicting the leastness of k unless
i = 0. Hence ak = 1, and 1, a, a2 , . . . , ak−1 are distinct. Moreover, by
the Division Algorithm we may write (with 0 ≤ r ≤ k)
1
m
a =a kq+r
= (k q r
a7

) a = ar ∈ {1, a, . . . , ak−1 }
for any m ∈ Z; and so h ai = {1, a, a2 , . . . , ak−1 } =⇒ |h ai| = k.
Now we can deduce
II.D.7. C OROLLARY. Given a ∈ G, with | G | < ∞, we have: (i) the

smallest k ∈ Z>0 for which ak = 1 divides | G |; and (ii) a|G| = 1.
24 II. GROUPS
P ROOF. (i) is immediate from Lagrange and the discussion above;

1
and (ii) follows since a|G| = a[G:hai]·|hai| = (
a|h
ai| )[ G:h ai] = 1.
*

II.D.8. C OROLLARY. If | G | = p is prime, the G is cyclic (hence also
abelian).

P ROOF. Let a ∈ G \ {1}. Since |h ai| > 1 and |h ai|| G |= p, we
must have |h ai| = p. So a generates G.
Euler’s phi-function φ(m) counts the number of integers between
0 and m which are relatively prime to m; that is, φ(m) = |Z∗m |. So
applying Corollary II.D.7(ii) to G = Z∗m gives
II.D.9. E ULER ’ S T HEOREM . Let m ≥ 2. In Z∗m , we have āφ(m) = 1̄.

(That is, aφ(m) ≡ 1 for any a with ( a, m) = 1.)
(m)
A special case of this is Fermat’s little theorem:
(II.D.10) a p−1 ≡ 1 for p prime.

( p)
II.D.11. E XAMPLE . Some subgroups of S4 and their orders:

• V = {1, (12)(34), (13)(24), (14)(23)} “Klein 4-group”; |V | = 4.
• D4 < S4 : think of actions of symmetries of a square on the ver-
tices (numbered 1, 2, 3, 4); | D4 | = 8.
• A4 alternating group; |A4 | = 12.
To see the order of A4 , recall that |S4 | = 4! = 24; it suffices to show
that [S4 :A4 ] = 2. This is true for any n, not just 4: multiplying by any
transposition gives a bijection between An and Sn \An .
Since the elements of V have sgn 1 (why?), we have Sn > An >
V. These elements also arise from symmetries of the square (which
ones?), and so Sn > Dn > V. All of this agrees with Lagrange,
which also tells us that neither of A4 and D4 can contain the other.
II.D.12. D EFINITION . The exponent of a finite group G is
exp( G ) := min{e ∈ N | ge = 1 (∀ g ∈ G )}.
For example, exp(Sn ) = lcm[1, . . . , n]. When n = 4 this is 12: the

elements of S4 have orders 1, 2, 3, and 4; so the smallest power that
II.D. COSETS AND LAGRANGE’S THEOREM 25
makes all of them 1 is 12. There is no element of actual order 12. (You
will check all of this in HW.) The next result says that we can blame
this on the fact that S4 is nonabelian:
II.D.13. P ROPOSITION . Let G be finite abelian. Then there exists a

g ∈ G with order exp( G ).
II.D.14. L EMMA . Let G be abelian. Then for all g1 , g2 ∈ G,
(|h g1 i|, |h g2 i|) = 1 =⇒ |h g1 g2 i| = |h g1 i||h g2 i|.

P ROOF. As the intersection h g1 i ∩ h g2 i is a subgroup of both h g1 i
and h g2 i, its order divides them both, hence must be 1. Write o :=
|h g1 g2 i|. Since G is abelian, ( g1 g2 )o = 1 =⇒ g1o g2o = 1 =⇒
g1o = g2−o ∈ h g1 i ∩ h g2 i = {1}. Now g1o = 1 = g2o means that |h g1 i|
and |h g2 i| divide o (why?), and so their lcm — which in this case6 is
just |h g1 i||h g2 i| — must also divide o. Again using that G is abelian,
we have ( g1 g2 )|h g1 i||h g2 i| = 1, and it follows that o divides |h g1 i||h g2 i|.
So they are equal.
P ROOF OF II.D.13. Let g be an element of maximal order. Sup-
pose |h gi| 6= exp( G ), i.e. that there exists h ∈ G with h|h gi| 6= 1. Then
|hhi| does not divide |h gi|, and there exists a prime p with highest
powers p f resp. pe dividing |hhi| resp. |h gi|, such that f > e. Hence
by II.D.14
f e
γ := |h|hh{z
i|/p
gp
} · |{z} has order p f −e |h gi| > |h gi| ,
order p f |h gi|
order pe
in contradiction to the assumed maximality of |h gi|.

II.D.15. C OROLLARY. Let G be a finite group. Then
G is cyclic ⇐⇒ exp( G )=| G | and G is abelian.
P ROOF. ( =⇒ ) is clear: consider a generator of G. For ( ⇐= ),

II.D.13 provides g ∈ G with |h gi| = exp( G ) (= | G |). Conclude that
h gi = G.
6Recall that lcm( a, b) · gcd( a, b) = a · b.
26 II. GROUPS
II.E. Homomorphisms and isomorphisms
In §II.A it was mentioned that from the assumption
ϕ( ab) = ϕ( a) ϕ(b)
on the map ϕ : G → H (i.e., the defining property of a homomor-

phism) follow other properties:
cancel
• ϕ(1) = ϕ(1 · 1) = ϕ(1) ϕ(1) =⇒ 1 = ϕ(1)
ϕ (1)
• 1 = ϕ (1) = ϕ( xx −1 )
= ϕ ( x ) ϕ ( x −1 ) =⇒ ϕ( x −1 ) = ϕ( x )−1
n n
• ϕ( x ) = ϕ( x ) etc.
You can also use a homomorphism to construct subgroups of G and
H, called the kernel and image of ϕ:
• ker( ϕ) := { g ∈ G | ϕ( g) = 1 H } ⊂ G;
• im( ϕ) := {h ∈ H | h = ϕ( g) for some g ∈ G } ⊂ H.
(The image is also denoted ϕ( G ).)
II.E.1. P ROPOSITION . (i) ker( ϕ) ≤ G; and (ii) im( ϕ) ≤ H.
P ROOF. (i) ϕ( g) = 1 = ϕ( g0 ) =⇒ ϕ( gg0 ) = ϕ( g) ϕ( g0 ) = 1.

(ii) h = ϕ( g), h0 = ϕ( g0 ) =⇒ hh0 = ϕ( g) ϕ( g0 ) = ϕ( gg0 ).
II.E.2. E XAMPLES . (a) An = ker{sgn : Sn → {1, −1}}.

(b) SLn (C) = ker{det : GLn (C) → C∗ }.
2πi 2πia
(c) he n i = im{ξ n : Zn → C∗ }, where ξ n sends ā 7→ e n .
(d) hr i = im{ ϕn : Zn → Dn }, where ϕn sends ā 7→ r a .
(e) Γ( N ) := ker{SL2 (Z) → SL2 (Z/NZ)}. (The target of the map
means 2 × 2 matrices with entries in Zm and determinant 1̄. The
kernel can be thought of as integer matrices with determinant 1 and
equivalent to the identity matrix mod N, entry by entry.)
(f) 2πZ = ker{(R, +, 0) → (C∗ , •, 1)}, where the homomorphism
sends θ 7→ eiθ .
(g) C ( G ) = ker{ı : G → Aut( G )}. Here Aut( G ) is the group of auto-
morphisms of G, or isomorphisms7 from G to itself, under the binary
operation of composing maps. The homomorphism ı sends g 7→ ı g ,
7see II.E.3 just below
II.E. HOMOMORPHISMS AND ISOMORPHISMS 27
where ı g ( x ) := gxg−1 is the automorphism called conjugation by g.

(These are also written Ψ and Ψ g .) If G is abelian, then C ( G ) = G
and all ı g are just the identity map (sending g 7→ g).
Note that if G is a cyclic group hαi, a homomorphism ϕ : G → H

is completely determined by the image of α. (Why?)
II.E.3. D EFINITION . A homomorphism ϕ : G → H is called

• trivial if im( ϕ) = {1} (or {0} if the operation is “+”); equiva-
lently, ker( ϕ) = G.
• surjective (or “onto”), and written G H, if im( ϕ) = H; an
example is the reduction mod n homomorphism Z → Zn sending
a 7→ ā.
• injective (or “1-to-1”), and written G ,→ H, if ker( ϕ) = {1} (or
{0} if the operation is “+”); an example is the map Zn ,→ Zmn
sending ā 7→ ma.
∼
=
• an isomorphism, and written G → H, if it is both injective and
∼
=
surjective; the conjugation map ı g : G → G (for any g ∈ G) is
an example, as is the identity map. Another would be the map
2πi 2πia
Zn → he n i sending ā 7→ e n .
On one hand, a non-identity automorphism of a group (like con-

jugation by a non-central element in a non-abelian group) should be
thought of as a structural symmetry. On the other, given two groups
G and H, a priori differently presented and/or labeled, the existence
of an isomorphism ϕ between them reveals that they are really the
same group. We then say that G and H are isomorphic. Along these
lines there is the
II.E.4. P ROPOSITION . If G ∼ = H then G, H have:

(a) the same order (if finite);
(b) the same orders of subgroups and elements; and
(c) are either both abelian or both nonabelian.8
8One could also add (say) that G and H have the same minimal number of gener-
ators.
28 II. GROUPS
We will first prove two lemmas. The start with, we should justify
calling injective homomorphisms “1-to-1”.
II.E.5. L EMMA . For a homomorphism ϕ : G → H, the following are

equivalent:
(A) ϕ injective in the sense of II.E.3;
(B) ϕ is 1-to-1, i.e. injective in the set-theoretic sense; and
(C) ϕ is an isomorphism onto its image.
P ROOF. (A) ⇐⇒ (C): clear, since ϕ is always “surjective onto its

image”.
(A) =⇒ (B): suppose ϕ( x ) = ϕ(y). Then 1 = ϕ(y) ϕ( x )−1 = ϕ(yx −1 );
since the kernel is trivial, this gives yx −1 = 1 hence x = y.
(B) =⇒ (A): ϕ(1G ) = 1 H ; since ϕ is 1-to-1, no other element of G
goes to 1 H , so ker( ϕ) = ϕ−1 (1 H ) = {1}.
Part (ii) of the next lemma is useful for producing isomorphisms.
∼
=
II.E.6. L EMMA . (i) Any ϕ : G → H is invertible: “ϕ−1 : H → G” is
well-defined, a homomorphism and an isomorphism, with ϕ ◦ ϕ−1 = id H
and ϕ−1 ◦ ϕ = idG .
(ii) If homomorphisms ϕ : G → H and η : H → G are such that ϕ ◦ η =
id H and η ◦ ϕ = idG , then ϕ and η are isomorphisms.
P ROOF. (i) Let h ∈ H. Since ϕ is 1-to-1 [resp. onto], ϕ−1 (h) is ≤ 1

[resp. ≥ 1] element; i.e. ϕ−1 (h) ∈ G is exactly one element. Writing
h = ϕ( g) and h0 = ϕ( g0 ), applying ϕ−1 to ϕ( g) ϕ( g0 ) = ϕ( gg0 ) gives
ϕ−1 (hh0 ) = gg0 = ϕ−1 (h) ϕ−1 (h0 ). Finally, since ϕ is everywhere
defined (on G) [resp. well-defined], ϕ−1 is onto [resp. 1-to-1].
(ii) We check this for ϕ. For surjectivity: given h ∈ H, we have
h = idH (h) = ϕ(η (h)). For injectivity: if ϕ( g) = 1, then 1 = η (1) =
η ( ϕ( g)) = idG ( g) = g.
∼
=
P ROOF OF II.E.4. We have some ϕ : G → H.
(a) By II.E.6(i), ϕ is a bijection of sets; so the orders are the same.
(b) ϕ is a bijection, and for any G 0 ≤ G, we have ϕ( G 0 ) ≤ H (by
II.E.1(ii)) and G 0 ∼= ϕ( G 0 ) (given by restricting ϕ to G 0 ). Similarly,
taking H 0 ≤ H, ϕ−1 ( H 0 ) ≤ G and H 0 ∼ = ϕ−1 ( H 0 ). So orders of

subgroups (in particular, the cyclic groups generated by elements)
are the same.
(c) Applying ϕ to xy = yx yields ϕ( x ) ϕ(y) = ϕ(y) ϕ( x ); and any pair
of elements of H can be written as ϕ( x ), ϕ(y). So G abelian =⇒ H
abelian; and the converse holds by using ϕ−1 in the same way.
Here is a very useful way to construct isomorphisms for finite

groups (which saves work involved in II.E.6(ii)).
II.E.7. P ROPOSITION . If ϕ : G → H is an injective homomorphism

and | G | = | H | < ∞, then ϕ is an isomorphism.
P ROOF. To get surjectivity, apply the “pigeonhole principle”: you

have a map from an n-element set G to an n-element set H; no 2
elements of G go to the same element of H, and so every element of
H gets “hit”.
The contrapositive of II.E.4 says: if any of the structural proper-

ties (a), (b), (c) of 2 groups differ, they cannot be isomorphic. This will
be our first main application — telling groups apart (cf. (ii), (iii), (iv)
below). But let’s start with an isomorphism:
II.E.8. E XAMPLES . (i) The symmetries of a regular n-gon yield

permutations of the vertices (numbered 1 to n), which produces a
homomorphism ϕ : Dn → Sn . If vertices stay in place then clearly
there is no motion, and so ϕ is injective. (By II.E.5(c), you can think
of this as saying: there is (∀n) a subgroup of Sn isomorphic to Dn .) For
n = 3, | D3 | = 6 = |S3 | =⇒ ϕ is an isomorphism (by II.E.7);
numbering the vertices of the triangle counterclockwise, with “1”
fixed by the reflection h, we have ϕ(h) = (23) and ϕ(r ) = (123).
(ii) | D6 | = 12 = |A4 |. An isomorphism doesn’t “feel” natural, so

instinct tells us to look for a difference in structure: D6 has 2 elements
of order 3: r2 and r4 ; while A4 has 8 elements of order 3: the 8 3-cycles
(123), (132), (124), (142), (134), (143), (234), (243). So D6 A4 .
30 II. GROUPS
(iii) | D12 | = |S4 | = |Z24 | = 24. Z24 is abelian; the other two are
not: in S4 , (12)(23) = (123) 6= (132) = (23)(12), while in D12 ,
hr = r −1 h 6= rh. So Z24 D12 , S4 .
Now write out the cycle types for S4 :
form of decomp. how many

order
into disjoint cycles such elements?
(· · · ·) 4 6
(· · ·)(·) 3 8
(· ·)(· ·) 2 3
(· ·)(·)(·) 2 6
(·)(·)(·)(·) 1 1
The last row is just the identity element; the two rows above it indi-
cate that there are 3 + 6 = 9 elements of order 2 in S4 . Now D12 has
13 elements of order 2: the 12 reflections {hr a | a = 0, 1, . . . , 11}, and
one 180◦ -rotation r6 . So D12 S4 .
(iv) |V | = |Z4 | = 4. The orders of elements are 1, 2, 2, 2 for V, and

1, 4, 2, 4 for Z4 . So V Z4 .
(v) All cyclic groups of order N are isomorphic to (Z N , +). Just write
down the homomorphism from Z N → hαi sending 1̄ 7→ α hence
m 7→ αm .
We now formalize a construction touched on in II.C.3(iv):
II.E.9. D EFINITION . The direct product of two groups H and K is

(a group)
H × K := {(h, k) | h ∈ H, k ∈ K }
with (h, k ) · (h0 , k0 ) := (hh0 , kk0 ), (h, k )−1 = (h−1 , k−1 ), and 1 H ×K =
(1 H , 1K ). [If H, K are abelian, we will frequently write this additively:
(h, k) + (h0 , k0 ) = (h + h0 , k + k0 ), −(h, k) = (−h, −k), and 0 H ×K =
(0 H , 0K ).]
II.E.10. A LTERNATE D EFINITION . A group P is a direct product

of groups H and K if there exist homomorphisms p H : P → H and
pK : P → K such that for all groups G and homomorphisms f H : G → H

and f K : G → K, there exists a unique homomorphism f : G → P which
makes
fH
G / HO
f
fK pH

Ko pK
P
commute.
This kind of characterization of direct products is called universal,

and the italicized statement their universal property. In the HW, you
will check that P = H × K (from II.E.9) indeed is a direct product in
this sense (of II.E.10).
Now clearly | H × K | = | H | · |K |, which brings us to the
II.E.11. D IRECT P RODUCT T HEOREM . Let H, K ≤ G. Put HK :=

{hk | h ∈ H, k ∈ K }. (This is not necessarily a group!) Consider the
possible assumptions
(A) hk = kh (∀h ∈ H, k ∈ K )
(B) H ∩ K = {1G }.
Then
(i) (A) =⇒ HK ≤ G
(ii) (A) + (B) =⇒ HK ∼ = H×K
(iii) (A) + (B) + HK = G =⇒ G ∼ = H×K
(iv) (A) + (B) + | G | < ∞ + | H ||K |=| G | =⇒ G ∼
= H × K.
P ROOF. (i) We only need to check that 1 ∈ HK, (hk)(h0 k0 ) =

hh0 kk0 ∈ HK (by (A)), and (hk )−1 = (kh)−1 = h−1 k−1 ∈ HK (again
by (A)).
(ii) Define ϕ : H × K → HK by ϕ(h, k ) := hk. This is a homo-
morphism since ϕ(h, k ) ϕ(h0 , k0 ) = hkh0 k0 = hh0 kk0 = ϕ(hh0 , kk0 ) =
ϕ((h, k ) · (h0 , k0 )) (by (A)), injective because 1 = ϕ(h, k) = hk =⇒
k−1 = h ∈ H ∩ K = {1} =⇒ (h, k ) = (1, 1) (by (B)), and obviously
surjective by the description of HK.
32 II. GROUPS
(iii) is clear from (ii).

(iv) By (i), G ≥ HK, so
(ii)
| G | ≥ | HK | = | H × K | = | H ||K | = | G |
forces | G | = | HK |. Hence G = HK, whence (by (iii)) G ∼
= H × K.
II.E.12. E XAMPLE . Given r, s ∈ N, let ` := lcm(r, s), g := gcd(r, s).

Put s̃ := s/g ∈ N and G := Zr × Zs . Now let H denote the isomor-
phic image of Z` ,→ Zr × Zs (via9 ā 7→ ( ā, ā)), and K denote the iso-
morphic image of Zg ,→ Zr × Zs (via10 b̄ 7→ (0̄, bs̃)). Since ` g = rs,
we get | H ||K | = | G |.
Now in II.E.11, (A) holds since G is abelian. To see (B), we need
H ∩ K = {(0̄, 0̄)}. Take ( ā, ā) ≡ (0̄, bs̃) ∈ H ∩ K ⊂ Zr × Zs . It’s
enough to show that the left-hand side is zero, i.e. a ≡ 0 mod r and
mod s. We already have a ≡ 0 and a ≡ bs̃, which yield r | a and
(r ) (s)
s̃|s|( a − bs̃). Hence r, s̃| a; and since r and s̃ are relatively prime, we
get ` = r s̃| a. But r, s|`, and so r, s| a as desired. At this point, by
II.E.11(iv) we obtain H × K ∼ = G, or
Z` × Z g ∼
= Zr × Z s .
∼
=
II.E.13. E XAMPLE . The special case Zrs → Zr × Zs for (r, s) = 1
is also valid for multiplicative groups:
∗ ∼
=
ϕ : Zrs → Zr∗ × Z∗s
ā 7−→ ( ā, ā).
[This is clearly also a multiplicative homomorphism, and so invert-

ible congruence classes (mod rs) go to pairs of such. For surjectivity,
the point is to use the surjectivity of Zrs → Zr × Zs that we already
know. Given (b̄, c̄) ∈ Zr∗ × Z∗s , there is ( β̄, γ̄) ∈ Zr∗ × Z∗s with βb = 1̄
and γc = 1̄; and that surjectivity yields ā, ᾱ ∈ Zrs with ( ā, ā) = (b̄, c̄)
9In more detail, this sends a mod ` to (a mod r, a mod s). Since r, s|`, this makes
sense. The map is injective because if ā goes to (0̄, 0̄), this means that r, s| a, so that
their lcm `| a and the original ā was 0̄.
10Here g|b =⇒ s = gs̃|bs̃, so it is well-defined.
ϕ
and (ᾱ, ᾱ) = ( β̄, γ̄). So we get aα 7→ (bβ, cγ) = (1̄, 1̄). Since ϕ is
injective on a set-theoretic level, aα must be = 1̄, hence ā ∈ Zrs
∗ .]
This example has a beautiful number-theoretic application.
II.E.14. P ROPOSITION . The Euler phi-function
φ(n) = n ∏ (1 − 1p ).
p|n
p prime
P ROOF. Write the prime factorization of n
n = p1e1 · · · pet t .
Inductively applying II.E.13,
Z∗n ∼
= Z∗p1 e1 × · · · × Z∗pt et ,
and taking orders on both sides gives
∏ φ ( p i i ).
e
φ(n) =
i
Now, for a prime p, everything in {0, 1, . . . , pe − 1} is relatively prime

to pe except for multiples of p. As there are pe−1 such multiples,
φ( pe ) = pe − pe−1 = pe (1 − 1p ),
e 1 1
so φ(n) = ∏i pi i ∏i (1 − pi ) = n ∏ i (1 − pi ).
II.E.15. E XAMPLES . (i) D6 ∼ = D3 × Z2 : apply II.E.11(iv) to G =

D6 , H = hr i = Z2 , and K = hr2 , hi ∼
3 ∼ = D3 . (Think of a regular
triangle inside a regular hexagon, sharing 3 of its vertices.) Since
H = {1, r3 } and K = {1, r2 , r4 , h, hr2 , hr4 }, we have H ∩ K = {1};
| H ||K | = 2 · 6 = 12 = | D6 |; and r3 commutes with powers of r, and
also with h (in general, ri h = hr −i , but r3 = r −3 in D6 ).
(ii) V ∼
= Z2 × Z2 : use H = h(12)(34)i and K = h(14)(23)i, same
idea as above.
34 II. GROUPS
II.F. Group actions and Cayley’s theorem
II.F.1. D EFINITION . Let X be a set and G a group. An action of G

on X is a function
G×X → X
( g, x ) 7→ g.x
satisfying:
(i) ( gh).x = g.(h.x ) for all g, h ∈ G and x ∈ X; and
(ii) 1G .x = x for all x ∈ X.
A set X with G-action is called a G-set.
II.F.2. P ROPOSITION . A G-action on X is the same thing as a homo-

morphism ϕ : G → SX .
P ROOF. Given an action, x 7→ g.x is a permutation of X (i.e. bijec-

tion from X to itself), since
(i) (ii)
g−1 .( g.x ) = ( g−1 g).x = 1.x = x =⇒ g 1-to-1
(i) (ii)
g.( g−1 .x ) = ( gg−1 ).x = 1.x = x =⇒ g onto.
Setting ϕ( g) x := g.x therefore exhibits ϕ( g) as an element of SX .

This is a homomorphism because
(i)
ϕ( g) ϕ(h) x = g.(h.x ) = ( gh).x = ϕ( gh) x (∀ x )
=⇒ ϕ( g) ϕ(h) = ϕ( gh).
Conversely, given ϕ, define g.x = ϕ( g) x.
I find ϕ( g) x more notationally confusing than g.x, but viewing

an action as a homomorphism ϕ : G → SX is conceptually useful. If
ϕ is injective, we call the action faithful or effective. In that case the
action presents G as a subgroup of SX (cf. II.E.5).
II.F.3. D EFINITION . Let G act on X. The orbit of x is the subset
G ( x ) := { g.x | g ∈ G } ⊂ X
II.F. GROUP ACTIONS AND CAYLEY’S THEOREM 35
consisting of its “G-translates”, and the stabilizer of x is the sub-

group11
Gx := { g ∈ G | g.x = x } ≤ G
of elements “fixing” x. The action of G is transitive if G ( x ) = X (for
some, hence any, x).
II.F.4. E XAMPLES . (i) G = (Z, +) acts on X = R by translation:
n.r := r + n (r ∈ R, n ∈ Z).
[Check: (n1 + n2 ).r = r + n1 + n2 = (n1 .r ) + n2 = n2 .(n1 .r ); and

0.r = r + 0 = r.] Let x ∈ R. The orbit is G ( x ) = { x + n | n ∈ Z} and
the stabilizer is Gx = {0}.
(ii) “Tautological” examples:
• G = Sn acts on X = {1, . . . , n} by
σ.a := σ ( a) (σ ∈ Sn , a ∈ X).
We have G ( a) = X and Ga ∼ = Sn−1 , where the Sn−1 arises from

permutations of {1, . . . , n} \ { a}.
• GLn (R) acts on X = Rn by matrix multiplication.
(iii) D6 acts on X = {1, . . . , 6} by viewing X as the vertices of a regu-
lar hexagon:
3 2
4 1 h: flip
(r: rotation)
5 6
It’s helpful to use homomorphism notation here:
ϕ(r ) = (123456)
ϕ(h) = (26)(35).
Since D6 is generated by r, h, ϕ( D6 ) = h(123456), (26)(35)i ≤ S6 .
11[Jacobson]’s notation: Stab( x ).

36 II. GROUPS
(iv) The group ST of rotational symmetries of the regular tetrahedron

acts faithfully on its set of vertices X = {1, 2, 3, 4}. Viewing this as
an embedding (i.e. injective homomorphism) ϕ : ST ,→ S4 realizes
ST as A4 . This is because ϕ(ST ) contains all the 3-cycles, like (123),
which we may see as follows:
4 ← fix
3 2
Since 3-cycles already generate A4 , we don’t need more pictures, but

the other type of element — the products of disjoint transpositions
— can be visualized too, e.g. (13)(24):
4
180◦
3 2
On the other hand there are no single transpositions such as (12).

(We are not allowing reflections.) So
A4 ≤ ϕ(ST ) < S4 =⇒ A4 = ϕ(ST ),
using Lagrange’s theorem (how?).
(v) One can play the same game with the group SC of rotational sym-
metries of the cube, acting (faithfully) on . . .
• . . . the vertex set: SC ,→ S8
• . . . the edge set: SC ,→ S12
• . . . the face set: SC ,→ S6
• . . . the set of interior diagonals: SC ,→ S4
II.F. GROUP ACTIONS AND CAYLEY’S THEOREM 37
where “interior diagonals” connect antipodal points of the cube, as

shown. Let X be the 4-element set comprising these diagonals. The
following table describes the non-identity elements of SC :
# of total
possible
rotation type action on X possible # of
angles
axes elements
about facet midpoints
4-cycles, 90◦ , 180◦

3 3·3 = 9
(··)(··)’s 270◦
about edge midpoints
2-cycles 6 180◦ 6·1 = 6
about vertices
(on the diagonals)
3-cycles 4 120◦ , 240◦ 4·2 = 8
Adding the identity, we see that SC has at least 24 elements. Since

the action on X is faithful, it can have at most 4! = 24 elements.
Applying II.E.7 to the homomorphism ϕ : SC ,→ S4 , we see that
SC ∼
= S4 .
The example just concluded should have convinced you that there
are many natural ways of looking at some groups as subgroups of
permutation groups. But there is one “canonical” way:
II.F.5. C AYLEY ’ S T HEOREM . Every group G is a subgroup of the sym-

metric group SG . (In particular, if | G | = n is finite, then G is a subgroup
of Sn .)
38 II. GROUPS
P ROOF. Let G act on itself (X = G) by left translation:
g.g0 := gg0 .
Clearly ( gh).g0 = ghg0 = g.(h.g0 ) because group multiplication is as-

sociative; and also 1.g0 = 1g0 = g0 . By II.F.2, this yields a homomor-
phism ϕ : G → SG . It is injective because if g ∈ G has ϕ( g) = idG ,
then g = g1 = g.1 = ϕ( g)1 = idG (1) = 1. So ϕ gives an isomor-
phism from G onto its image ϕ( G ) ≤ SG .
II.F.6. R EMARK . We could also have used right translation in the

proof, i.e.
g.g0 := g0 g−1 .
This works because
g.(h.g0 ) = g.( g0 h−1 ) = g0 h−1 g−1 = g0 ( gh)−1 = ( gh).g0 .
Notice that the actions in the proof and remark aren’t so interest-
ing: the orbit of any element is the entire group. Fortunately, groups
also act on themselves in a more interesting way, which is our next
topic.
II.G. CONJUGACY AND THE ORBIT-STABILIZER THEOREM 39
II.G. Conjugacy and the orbit-stabilizer theorem
II.G.1. P ROPOSITION . Let g ∈ G. Mapping h 7→ ghg−1 defines an

isomorphism ı g : G → G.
P ROOF. HW. [Hint: use II.E.6(ii).]
II.G.2. D EFINITION . (i) ı g is called conjugation by g.

(ii) g0 , g00 ∈ G are said to be conjugate if there is a g ∈ G such that
g00 = ı g ( g0 ).
(iii) H 0 , H 00 ≤ G are said to be conjugate if there is a g ∈ G such that
H 00 = ı g ( H 0 ).
II.G.3. P ROPOSITION . (i) Conjugate groups are isomorphic.

(ii) Conjugate elements are of the same order.
P ROOF. The restriction of ı g gives the isomorphism in (i), and (ii)

follows from (i) by taking H 0 = h g0 i, H 00 = h g00 i.
Now consider the action of G on itself by conjugation
(II.G.4) g.g0 := gg0 g−1 .
The orbits of this action are called the conjugacy classes of G. There
are two notational ambiguities to get rid of here: first, since G can
act on itself in more than one way, we don’t write G ( g); second, if
an element lies in a subgroup H ≤ G, we need to distinguish G- and
H-orbits. (Even if the G-orbit lies in H, the H-orbit can be smaller.)
II.G.5. D EFINITION . Let g0 ∈ G. The conjugacy class of g0 in G

is
cclG ( g0 ) := { gg0 g−1 | g ∈ G }.
The conjugacy class of 1 ∈ G is always just the singleton {1}.
II.G.6. P ROPOSITION . G is abelian if and only if all of its conjugacy

classes have one element.
P ROOF. gh = hg (∀ g, h ∈ G) ⇐⇒ ghg−1 = h (∀ g, h ∈ G) ⇐⇒
cclG (h) = {h} (∀h ∈ G).
40 II. GROUPS
So let’s find the conjugacy classes in a couple of groups.
II.G.7. E XAMPLE . Let’s consider G = S3 . We know that cclS3 (1) =

{1}. Computing, one finds that
cclS3 ((12)) = {1(12)1−1 , (12)(12)(12)−1 ; (13)(12)(13)−1 ,
(123)(12)(123)−1 ; (132)(12)(132)−1 , (23)(12)(23)−1 }
= {(12); (23); (13)}
= cclS3 ((13)) = cclS3 ((23)),
consists of all the transpositions, while
cclS3 ((123)) = {(12)(123)(12)−1 , . . .}

= {(132), (123)} = cclS3 ((132))
contains both 3-cycles.
Now, rather than using brute force, we could cut down our work
by noticing that elements of cclS3 ((12)) must (like (12)) have order
2, hence be transpositions. But there is a still more powerful result.
II.G.8. D EFINITION . The cycle-structure of a permutation σ ∈ Sn

is the sequence


 b1 = # of fixed elements

 b2 = # of transpositions



b3 = # of 3-cycles
 ..
.





bn = # of n-cycles

in σ’s complete factorization into disjoint cycles. (More commonly,

we represent it symbolically, viz. (··)(··)(· · ·)(·).)
II.G.9. T HEOREM . cclSn (σ ) consists of all permutations with the same

cycle-structure as σ.
P ROOF. Write σ = ( a11 a12 · · · a1d1 ) · · · ( ak1 ak2 · · · akdk ) as a prod-

uct of disjoint cycles (of lengths d1 , . . . , dk ), with each element of
{1, . . . , n} appearing exactly once. For each η ∈ Sn , we have

ηση −1 = η ( a11 a12 · · · a1d1 )η −1 · · · η ( ak1 ak2 · · · akdk )η −1
= (η ( a11 )η ( a12 ) · · · η ( a1d1 )) · · · (η ( ak1 )η ( ak2 ) · · · η ( akdk ))
by your last HW. That is, we just apply η to all the “contents”, which
preserves the disjointness (η is bijective) and the lengths of the cy-
cles, hence the cycle structure. Finally, given any permutation with
the same cycle structure as σ
σ0 = (b11 b12 · · · b1d1 ) · · · (bk1 · · · bkdk )
then taking !
a11 a12 · · · akdk
η := ,
b11 b12 · · · bkdk
we have σ0 = ηση −1 .
II.G.10. E XAMPLE . Consider G = D5 . Recall that rh = hr −1 (and

h = h−1 , and r −1 = r4 ); that is,
rhr −1 = rrh = r2 h =⇒ r a hr −a = r2a h

hrh−1 = r −1 hh−1 = r −1 =⇒ hr a h−1 = r −a .
So cclD5 (h) = {h, r2 h, r4 h, r6 h = rh, r8 h = r3 h} = cclD5 (r2 h) = · · ·

and cclD5 (r ) = {r, r4 }, cclD5 (r2 ) = {r2 , r3 }, cclD5 (1) = {1}. In a
picture,
h rh r2 h r3 h r4 h
1 r r4 r2 r3
displays the four conjugacy classes in D5 .
The conjugacy classes in the last two examples partition G into

disjoint subsets. This is true in general:
x∼y ⇐⇒ ∃ g ∈ G s.t. y = gxg−1

( x and y are conjugate)
defines an equivalence relation on G. The equivalence classes are the
conjugacy classes; and if we take one representative gi of each, then
42 II. GROUPS
(by I.A.5) we have
(II.G.11) G= ä ccl( gi ).
i
More generally, if G acts on a set X and we define12
x∼y ⇐⇒ y ∈ G(x)
def.
⇐⇒ x, y in same G-orbit (∗)
then
∼ is reflexive: x ∈ G ( x ) =⇒ x ∼ x
∼ is symmetric: clear from (∗)
∼ is transitive: y = g.x and z = h.y =⇒ z = h.( g.x ) = hg.x.
hence defines an equivalence relation. Of course, X/∼ is the set of
orbits G ( x ), which by I.A.5 are disjoint with union all of X. If |X| <
∞, and we pick one element xi in each orbit, then
(II.G.12) X= ä G ( x i ).
i
Now we turn to our first counting result — a sort of analogue of

Lagrange’s Theorem for group actions.
II.G.13. E XAMPLE . First let’s look at the sizes of orbits and stabi-
lizers in the actions by conjugation from our last two examples:
(i) orbit: cclS3 ((12)) = {(12), (23), (13)} (3 elements)
stabilizer: (S3 )(12) = {1, (12)} (2 elements)
. . . and |S3 | = 6 = 3 · 2.
(ii) orbit: cclD5 (h) = {h, rh, r2 h, r3 h, r4 h} (5 elements)
stabilizer: ( D5 )h = {1, h} (2 elements)
. . . and | D5 | = 10 = 5 · 2.
It appears we are on to something. In the (big) statement that

follows, G/Gx will denote the set of left cosets of Gx in G.
12To see the second “ ⇐⇒ ” below: the forward implication is trivial, since x ∈
G ( x ) too. Conversely, suppose x, y are in the same G-orbit G (z), viz. x = g.z and
y = h.z. Then hg−1 .x = hg−1 g.z = h.z = y, so y ∈ G ( x ).
II.G.14. T HEOREM . Let x ∈ X be fixed.

(i) There is a 1-to-1 correspondence between points in the orbit of x and
cosets of its stabilizer — that is, a bijective map of sets:
(†)
G ( x ) −→ G/Gx
g.x 7−→ gGx .
(ii) [Orbit-Stabilizer Theorem] If | G | < ∞, then
| G ( x )| · | Gx | = | G |.
(iii) If x, x 0 belong to the same orbit, then Gx and Gx0 are conjugate as
subgroups of G (hence of the same order/etc.).
(iv) If g, g0 belong to the same (left) coset of Gx , then they act the same way
on x.
P ROOF. (i) We have g.x = g0 .x ⇐⇒ x = g−1 g0 .x ⇐⇒ g−1 g0 ∈

Gx ⇐⇒ g−1 g0 Gx = Gx ⇐⇒ gGx = g0 Gx , which proves (†) is
well-defined and injective. Surjectivity is obvious.
(ii) LHS(†) has size | G ( x )|; while RHS(†) has size | G/Gx | = # of
cosets of Gx , which by Lagrange’s Theorem is | G |/| Gx |.
(iii) The calculation is: if x 0 = g.x, then h ∈ Gx ⇐⇒ h.x = x
⇐⇒ gh.x = g.x ⇐⇒ ghg−1 .( g.x ) = g.x ⇐⇒ ghg−1 ∈ Gx0 . So
Gx 0 = ı g ( Gx ).
(iv) is pretty much a direct verbal translation of (i).
We want to apply II.G.14 to compute conjugacy classes. Recall

once more that in a group G, acting on itself by conjugation (and
x ∈ G), the orbit
G ( x ) = { gxg−1 | g ∈ G } =: cclG ( x )
is called the conjugacy class of x; while the stabilizer
Gx = { g ∈ G | gxg−1 = x } = CG ( x )
44 II. GROUPS
is called the centralizer of x. The Orbit-Stabilizer Theorem then says

that
(II.G.15) |cclG ( x )| · |CG ( x )| = | G |.

Next recall (Theorem II.G.9) that for σ ∈ Sn , cclSn (σ) consists of all
permutations with the same cycle-structure as σ. Since it is already the
cycle-structure which determines whether an element is in An , it fol-
lows that
(II.G.16) if σ ∈ An , then cclSn (σ) ⊂ An .
Here is a counting result for conjugacy classes in Sn .
II.G.17. P ROPOSITION . The number of permutations in Sn with cycle-

structure b1 , b2 , . . . , bn (cf. II.G.8) is
n!
.
∏nk=1 kbk bk !
P ROOF. First, lay out the “chambers” into which you are going
to insert the elements {1, . . . , n} to get a cycle:
(··)(··)(··) (· · ·) (· · ··)(· · ··) etc.

| {z } | {z } | {z }
b2 =3 b3 =1 b4 =2
Choose an ordering of {1, . . . , n} (there are n! possibilities) and plop

them down in that order. Now divide by the cyclic permutations
within each chamber (there are ∏nk=1 kbk = 2b2 3b3 4b4 · · · of these). Fi-
nally, divide out by permutations of chambers of the same length
(there are ∏nk=1 bk ! of these).
Before going on, you should reconceptualize this proof as an ap-

plication of (II.G.15).
6!
II.G.18. E XAMPLES . (i) |cclS6 ((12)(34)(56))| = 2·2·2·3! = 15.
6!
(ii) |cclS6 ((12345)(6))| = (5·1!)( 1·1!)
= 144.
6!
(iii) |cclS6 (1234)(56))| = (4·1!)(2·1!)
= 90.
6!
The order of the centralizer is, in each case, |cclS6 (··· )|
.
Now in spite of II.G.16, we may not have cclSn (σ ) = cclAn (σ) for
σ ∈ An :
II.G.19. T HEOREM . Given σ ∈ An , one has EITHER
(I) |cclAn (σ)| = |cclSn (σ )| ⇐⇒ CSn (σ) contains an odd permutation

equiv.
OR
(II) |cclAn (σ)| = 12 |cclSn (σ )| ⇐⇒ CSn (σ) ⊂ An .
equiv.
In the second case, one says that the conjugacy class “breaks” in An .
P ROOF. By (II.G.15) (applied twice),
(II.G.20) |cclAn (σ)||CAn (σ )| = |An | = 12 |Sn | = 12 |cclSn (σ )||CSn (σ )|.
If CSn (σ) ⊂ An , then CAn (σ) = CSn (σ) and so by (II.G.20) |cclAn (σ )| =
1
2 |cclSn ( σ )|.
Otherwise, CSn (σ ) contains an element of Sn \An (the odd per-
mutations), and CAn (σ) < CSn (σ), which by Lagrange means that
|CSn (σ)|
≥ 2.
|CAn (σ)|
But by (II.G.20) cclAn (σ) ⊆ cclSn (σ) =⇒ |cclSn (σ)||CAn (σ)| ≥
1
2 |cclSn ( σ )||CSn ( σ )| =⇒
|CSn (σ)|
≤2
|CAn (σ)|
(hence = 2). It follows that |cclSn (σ)| = |cclAn (σ)|.
II.G.21. E XAMPLES . (i) All 3-cycles are conjugate in A5 : since “all

3-cycles” is a conjugacy class (of some σ, say (123)) in S5 , we are
claiming cclS5 ((123)) = cclA5 ((123)). By II.G.19, it is enough to
show that CS5 ((123)) contains an odd permutation — i.e., that (123)
commutes with an odd permutation; and (45) does the job.
(ii) All 3-cycles are not conjugate in A4 : that is, cclA4 ((123)) is not
all the 3-cycles (= cclS4 ((123))), and we are in case (II) of II.G.19.
To check this, we need to compute CS4 ((123)): what permutations
η satisfy η (123)η −1 (= (η (1)η (2)η (3))) = (123)? Clearly just the
46 II. GROUPS
cyclic group h(123)i, which is indeed in A4 =⇒ |cclA4 ((123))| =

I I.G.19
1
2 |cclS4 ((123))|.
σ η
z }| { z }| {
(iii) How about cclA8 ((123)(4567)) and cclA8 ((123)(45678))?
• σ commutes with an odd permutation, namely (4567), and so σ

has the same conjugacy classes in A8 and S8 .
• η commutes with only elements of the group h(123), (45678)i which
consists of even permutations. So |cclA8 (η )| = 12 |cclS8 (η )|.
We mention in passing the conjugacy classes of a couple of other

groups: for D2n+1 (odd dihedral group) they are
{1}, {r, r −1 }, {r2 , r −2 }, . . . , {r n , r −n }; {h, rh, r2 h, . . . , r2n h}

and for D2n (even dihedral group)
{1}, {r, r −1 }, {r2 , r −2 }, . . . , {r n−1 , r −n+1 }, {r n };

{h, r2 h, r4 h, . . . , r2n−2 h}, {rh, r3 h, . . . , r2n−1 h}.
These are obtained by repeatedly applying rh = hr −1 as in II.G.10.
There is also Hamilton’s famous quaternion group:
II.G.22. D EFINITION . Q := {±1, ±i, ±j, ±k}, with ijk = i2 =

j2 = k2 = −1.
The conjugacy classes are {1}, {−1}, {i, −i}, {j, −j}, {k, −k}.
For example, jij−1 = −jij = jij(kk) = j(ijk)k = −jk = iijk = −i.
II.G.23. R EMARK . Hamilton arrived at the multiplication table for

Q by “formally dividing vectors in R3 ”, allowing himself
α~x α ~x ~y ~x ~x R (~x )
= , · = , and = θ
β~x β ~y ~z ~z ~y Rθ (~y)
for any rotation Rθ in the plane spanned by ~x and ~y. One is also
supposed to think “ ~~yx · ~y = ~x”, but don’t try to think too literally in
terms of linear transformations!
Taking I, J, K to be the standard basis of R3 , one can write (using

90◦ rotations13 about I, J, and K respectively)
K −J I −K J −I
i := J = K , j := K = I , k := I = J ,
which Hamilton encoded in a diagram:
K
k k
j i i −j
J j i
I i −j
-J
−k −k
-K
−J −J J J
Furthermore, we have i2 = K · KJ = J = −1, ij = −K · −IK = I = k,
I K I −J −K −I
ji = · = = −k = −ij, and ijk =
K J J · · = −1, and so on.
K I J
Now, to be honest, there are problems with the idea of “divid-
ing vectors in R3 ”, since at the end of the day there can be no “3-
dimensional division algebra over R” (as we’ll see later this semes-
ter). In any case, we get the right nonabelian group of order 8 and
that’s all we care about presently!
13One is supposed to think of i as “counterclockwise rotation around I” and so on.

48 II. GROUPS
II.H. Cauchy’s Theorem
By Lagrange, the order of an element g ∈ G divides | G |. The

converse statement, that for any positive integer n dividing G there ex-
ists g ∈ G of order n, is in general false. (Even for abelian groups:
Z2 × Z2 × Z2 contains no element of order 4.) But there is a pretty
application of the theory of group actions we have developed to the
case where n is prime. We’ll give two proofs; for the first you’ll have
to accept something that we will prove later.
We begin with some preliminaries: recall the
II.H.1. D EFINITION . The center of a group is
C ( G ) := { x ∈ G | gxg−1 = x ∀ g ∈ G } ,
the elements commuting with all the other elements of G.
Obviously we have:
(i) G is abelian ⇐⇒ G = C ( G );
(ii) C ( G ) is itself always abelian; and
(iii) |cclG ( x )| = 1 ⇐⇒ x ∈ C ( G ).
Recall also that if we take one representative xi in each conjugacy
class of G (| G | < ∞), then G = qi cclG ( xi ) and so
(II.H.2) | G | = ∑i |cclG ( xi )|.

Each element in C ( G ) has its own conjugacy class, and the right-
hand side of (II.H.2) becomes |C ( G )| + ∑i |cclG ( xi )|, where the sum
is now over representatives xi of conjugacy classes with more than
one element. Finally, by the Orbit-Stabilizer Theorem
|G|
|cclG ( xi )| = = [ G:CG ( xi )],
|CG ( xi )|
and we get the
II.H.3. C LASS E QUATION . | G | = |C ( G )| + ∑i [ G:CG ( xi )].
This will be used to prove

II.H. CAUCHY’S THEOREM 49
II.H.4. C AUCHY ’ S T HEOREM . If | G | < ∞ and p ∈ N is a prime

dividing | G |, then G contains an element of order p.
P ROOF (A). by induction on m ≥ 1, where | G | = mp.

base case (m = 1): We have | G | = p. Take any g ∈ G \{1}. Its order
is > 1 and divides p by Lagrange; hence |h gi| = p.
inductive step: [Assume we know the result for groups of order kp,

k < m.] Either (i) p|CG ( x )| for some x ∈ G \C ( G ), or (ii) p6 |CG ( x )|
for all x ∈ G \C ( G ).
In case (i), x ∈
/ C ( G ) =⇒ |cclG ( x )| > 1, and so
|G|
|CG ( x )| = < | G |.
|cclG ( x )|

By Lagrange, |CG ( x )|| G |; and so |CG ( x )| is a proper factor of | G | =
mp divisible by p. That is, |CG ( x )| = kp for some k < m (with
k | m); and we get an element in CG ( x ) of order p by the inductive
assumption.
In case (ii), let { xi } be a set of representatives of the conjugacy

classes outside the center; we have p6 |CG ( xi )| =⇒ p[ G : CG ( xi )]
for each i. So p divides the left-hand side of II.H.3 and the sum on
the right, hence also |C ( G )|. Now we use the
Fact: Any finite abelian group is a direct product of cyclic groups.
to write C ( G ) ∼
= Zm1 × · · · × Zmr . Clearly p must divide some m j ,
which gives a direct factor of C ( G ) of the form Zap . The element ā in
this factor has order p in C ( G ), thus also in G.
P ROOF (B). Inside G p = G × · · · × G consider the set
X := {( g0 , g1 , . . . , g p−1 ) ∈ G p | g0 g1 · · · g p−1 = 1}.
Having chosen entries g1 , . . . , g p−1 , we must take g0 = ( g1 · · · g p−1 )−1

to get an element of X, and so
| X | = | G | p −1 .
Introduce an action of Z p on X by cyclic permutation:
ā.( g0 , g1 , . . . , g p−1 ) := ( ga , . . . , g p−1 , g0 , g1 , . . . , ga−1 ).

50 II. GROUPS
This remains in X since g0 g1 · · · g p−1 = 1 =⇒
ga · · · g p−1 g0 g1 · · · ga−1 = ( g0 · · · ga−1 )−1 ( g0 g1 · · · g p−1 )( g0 · · · ga−1 )

= ( g0 · · · g a − 1 ) − 1 ( g0 · · · g a − 1 ) = 1
as required.
Now for given x ∈ X, the Orbit-Stabilizer Theorem gives
|Z p ( x )||(Z p ) x | = |Z p | = p
and so |Z p ( x )| = 1 or p (depending on x). Clearly,
|Z p ( x )| = 1 ⇐⇒ x invariant under cyclic permutations

⇐⇒ x = ( g, . . . , g) for some g ∈ G with g p = 1
Let α resp. β denote the number of 1- resp. p-element orbits in X;
since (1, . . . , 1) ∈ X is fixed, α > 0. If we can show that α > 1, then
there is some g 6= 1 with g p = 1, and we are done!
Finally, as X is a disjoint union of Z p -orbits, we have
| G | p−1 = |X| = α + pβ;

and since p| G |, this yields pα+ pβ =⇒ p | α > 0. So α ≥ p and we
are through.
We can use Cauchy’s Theorem to start classifying groups:
II.H.5. T HEOREM . Let p be an odd prime, | G | = 2p. Then G ∼

= Z2p
14
(cyclic) or D p (dihedral).
P ROOF. By Cauchy, there exist a, b ∈ G with |h ai| = 2 (hence

a = a−1 ) and |hbi| = p. Now a ∈/ hbi since the order of a doesn’t
divide p, and so
(II.H.6) ba ∈
/ hbi
14Note that Z × Z ∼ Z since (2, p) = 1.

2 p = 2p
II.H. CAUCHY’S THEOREM 51
since otherwise ba = br =⇒ a = br−1 ∈ hbi. Since [ G:hbi] = 2, there

are 2 cosets:
G = hbi q ahbi
= {1, b, b2 , . . . , b p−1 } q { a, ab, ab2 , . . . , ab p−1 }.
Thus
(II.H.6) =⇒ ba = abr (for some r ∈ [0, p − 1] ∩ Z)

=⇒ aba−1 = br
2
=⇒ b = abr a−1 = ( aba−1 )r = (br )r = br
2 −1
=⇒ br =1
=⇒ p | r2 −1 = (r + 1)(r − 1)
=⇒ p | r +1 or p | r −1
=⇒ 1 = br+1 or br−1
=⇒ b−1 = br or b = br
=⇒ aba−1 = b−1 or b .
(i) (ii)
In case (ii), a and b commute; use II.E.11 (on direct products) to de-
duce that G ∼ = Z p × Z2 . In case (i), we have just described the mul-
tiplication laws of D p .
II.H.7. D EFINITION . A group G with order | G | = pn ( p, n ∈

N, p prime) is called a p-group. (When we use this terminology, it is
understood that p is a prime.)
II.H.8. T HEOREM . Any p-group G has nontrivial15 center C ( G ).
P ROOF. We must show |C ( G )| 6= 1. Recall the class equation
| G | = |C ( G )| + ∑[ G:CG ( xi )],
i
15That is, C ( G ) 6= {1}.

52 II. GROUPS
where xi are representatives of those conjugacy classes with more

than one element. By the orbit-stabilizer theorem,
[ G:CG ( xi )] = |cclG ( xi )| > 1 ;

and by Lagrange’s theorem, [ G:CG ( xi )]| G |. Hence, p[ G:CG ( xi )]

for every i, and so (by the class equation and p| G |) it follows that

p|C ( G )|.
For G a non-p-group, trivial center is possible: e.g., C (Sn ) = {1}
for n ≥ 3.
II.H.9. C OROLLARY. If | G | = p2 , p prime, then G is abelian (and

∼
= Z p2 or Z p × Z p ).
P ROOF. By II.H.8, |C ( G )| > 1. By Lagrange, there are two cases:

Case (i): |C ( G )| = p. Taking h ∈ G \C ( G ),
1 < |cclG (h)| = [ G:CG (h)]| G | = p2 .

OST
Since 1 ∈ / cclG (h), we have |cclG (h)| = p (rather than p2 ) and thus
|CG (h)| = p; and since CG (h) ≥ C ( G ) > {1}, we must have CG (h) =
C ( G ). But h ∈ CG (h) (commutes with itself) and h ∈ / C ( G ), a contra-
diction. So the only possibility is . . .
Case (ii): |C ( G )| = p2 . We have |C ( G )| = p2 = | G | =⇒ C ( G ) =
G =⇒ G abelian. By Cauchy’s theorem, G 3 h of order p; let
H := hhi. Take g ∈ G \ H; it has order > 1 dividing p2 . If this order
is p2 then G ∼ = h gi ∼= Z p2 .
Otherwise, |h gi| = p; and setting K := h gi, we have:
• H ∩ K < K with order dividing |K | = p =⇒ H ∩ K = {1};
• hk = kh for every h ∈ H, k ∈ K because G is abelian; and
• | H ||K | = p2 = | G |.
Thus by II.E.11 G ∼= H×K ∼= Zp × Zp.
II.I. NORMAL SUBGROUPS AND QUOTIENT GROUPS 53
II.I. Normal subgroups and quotient groups
In our discussion of conjugation, we defined the centralizer of an

element x ∈ G: its elements are just those g ∈ G with gxg−1 = x.
Suppose we replace x by a subgroup H ≤ G. The new feature
which arises is that ı g ( H ) = gHg−1 (:= { ghg−1 | h ∈ H }) can equal
H without our having ghg−1 = h for each h ∈ H. So there are two
natural generalizations of CG ( x ): the centralizer (of H in G)
(II.I.1) CG ( H ) := { g ∈ G | ghg−1 = h ∀h ∈ H }
which we already encountered, and the normalizer (of H in G)
(II.I.2) NG ( H ) := { g ∈ G | gHg−1 = H }.
Given h ∈ H and g ∈ NG ( H ), we have only that ghg−1 ∈ H.

The orbit-stabilizer theorem (for conjugation) for an element x ∈
G said that the number of conjugates (= size of orbit) of x equals the
index of CG ( x ) in G. Similarly, recalling that the image of H under ı g
(also a subgroup of G) is called a conjugate of H, we have the
II.I.3. P ROPOSITION . The number of (distinct) conjugates of H in G

is [ G:NG ( H )].
P ROOF. Let G act by conjugation on the set X of subgroups of G.

We are interested in | G ( H )|, where G ( H ) means the orbit of H as an
element in the set X. By the general orbit-stabilizer theorem, this is
related to the stabilizer GH = NG ( H ) of H in X by
| G ( H )|| NG ( H )| = | G |,
or equivalently | G ( H )| = | G |/| NG ( H )| = [ G:NG ( H )].
II.I.4. D EFINITION . If NG ( H ) = G, H is normalized by all of G and

we say H is a normal subgroup of G. We write H E G (or H C G if
H is proper in G).
54 II. GROUPS
II.I.5. P ROPOSITION . For a subgroup H ≤ G, the following properties

are equivalent:
(i) NG ( H ) = G;
(ii)16 gHg−1 = H (∀ g ∈ G );
(iii) gH = Hg (∀ g ∈ G ); and
(iv) H is a union of (entire) conjugacy classes.
P ROOF. (i) ⇐⇒ (ii) is obvious, as NG ( H ) is just those g for which

gHg−1 = H.
(ii) ⇐⇒ (iii) looks clear, but let’s write out the details for one di-
rection: assume (ii), and let gh ∈ gH. We have h0 := ghg−1 ∈ H, so
that gh = ghg−1 g = h0 g ∈ Hg. So gH ⊂ Hg; the reverse inclusion is
similar.
(ii) =⇒ (iv): If H is not a union of conjugacy classes, then H contains
some but not all of a conjugacy class; i.e. there exist y ∈
/ H and x ∈ H
with y ∈ ccl( x ). But then for some g ∈ G, gxg = y ∈− 1 / H =⇒
− 1
gHg 6⊂ H.
(iv) =⇒ (ii): Let g ∈ G and h ∈ H. Since H is a union of conjugacy
classes, h ∈ H =⇒ ccl(h) ⊂ H =⇒ ghg−1 ∈ H. We conclude that
gHg−1 ⊆ H; moreover, every h ∈ H is g( g−1 hg) g−1 with g−1 hg ∈
ccl(h) ⊂ H, so the “⊆” is in fact an equality.
Note that if G is abelian, all its subgroups are normal. Here are
some more interesting
II.I.6. E XAMPLES . (a) In G = S4 : The Klein 4-group V4 is the union

of two conjugacy classes of S3 : the identity {1}, and the set of all el-
ements with cycle structure (··)(··). Hence V4 C S4 .
Consider next the cyclic subgroup h(123)i = {1, (123), (132)}.
Since (34)(123)(34)−1 = (124) ∈ / h(123)i, we find that h(123)i C6 S4 .
(b) In G = D5 : We have h(h)i = {1, h} C 6 D5 , as rhr −1 = r2 h ∈
/ h(h)i.
2 3 4 k − 1 − k
But hr i = {1, r, r , r , r } C D5 since hr h = r ∈ h(r )i.
16This is usually given as the definition of a normal subgroup.

(c) An C Sn for n ≥ 3: Conjugacy classes in Sn consist of all permu-

tations with a given cycle-structure. An consists of all permutations
with “even” cycle-structures (i.e., n − #{disjoint cycles} is even). So
An is a union of ccl’s in Sn , hence normal.
(d) C ( G ) E G for any group G: x ∈ C ( G ) =⇒ gxg−1 = x ∀ g ∈ G,
so gC ( G ) g−1 = C ( G ) (∀ g ∈ G). Alternatively: the center consists of
all 1-element ccl’s.
(e) [HW] [ G, G ] E G for any G: here [ G, G ] is the commutator sub-
group generated by all commutators [ g1 , g2 ] = g1−1 g2−1 g1 g2 of ele-
ments g1 , g2 ∈ G.
II.I.7. E XAMPLE . Find

(a) all normal subgroups of S4 other than {1} and S4 , and
(b) all the normal subgroups of each such H.
(a) We know the conjugacy clases correspond to the cycle-structures:
(i) (· · · ·), (ii) (· · ·)(·), (iii) (··)(··), (iv) (··)(·)(·), and (v) (·)(·)(·)(·)
[identity]. All subgroups contain the identity. If H contains ccl (iv),
then H = S4 : transpositions generate S4 by II.B.5. If H contains ccl
(ii) then H = A4 or S4 : 3-cycles generate A4 by II.C.6. If H contains
ccl (i), then H 3 (1234) hence (1234)2 = (13)(24); since it is normal,
H then contains ccl (iii), and the element (1234) · (14)(23) = (24)
=⇒ H contains ccl (iv) H = S4 . (We are not saying that there is
no proper subgroup of S4 containing a 4-cycle, just that there are no
proper normal subgroups!) Finally, if H ⊇ ccl (iii), there are 2 options:
contain also ccl (i), (ii), and/or (iv) (in which case we already know
the outcome); or don’t contain any of these. In the latter case, H =
V4 . So the (proper) normal subgroups of S4 are A4 and V4 .
(b) In V4 , the order-2 cyclic subgroups (e.g. {1, (12)(34)}) are normal
simply because V4 is abelian. Note that these are not normal in S4
since non-identity elements of V4 can be conjugated into one another.
In A4 = {1} ∪ ccl(ii) ∪ ccl(iii), “ccl (ii)” [3-cycles] splits into 2 ccl’s
(with respect to conjugation by A4 ) while “ccl (iii)” [(··)(··)] does not.
56 II. GROUPS
(Why? See II.G.19.) The 2 ccl’s into which the 3-cycles split are
{(123), (142), (134), (243)} and {(132), (124), (143), (234)}.

Obviously, including one in a subgroup forces inclusion of the other,
since squaring the first set of elements gives the second set and vice-
versa! But then you have included all 3-cycles and get all of A4 . The
only option for a normal subgroup of A4 (other than itself and {1})
is thus V4 = {1} ∪ ccl(iii).
Here are two more ways to produce normal subgroups. The

second is more important, and in fact characterizes all normal sub-
groups, as we will see.
II.I.8. P ROPOSITION . Any subgroup H ≤ G of index 2 is normal.

(Here we need not assume G finite.)
P ROOF. For any a ∈ G \ H, G = H q aH. Let h ∈ H and g ∈ G;

we must show that ghg−1 ∈ H (cf. II.I.5(ii)). If g ∈ H, this is clear; so
take g = ax ∈ aH.
Suppose ghg−1 ∈ / H. Then ghg−1 ∈ aH and (for some y ∈ H) we
have
ay = ( ax )h( ax )−1 = a(|xhx −1 −1 0 −1
{z }) a = ah a
∈H
=⇒ y = h 0 a −1
=⇒ a = y −1 h 0
∈ H, contradicting the choice of a.
− 1
So ghg ∈ H and we are done.
II.I.9. P ROPOSITION . Let ϕ : G → H be a homomorphism. Then

ker( ϕ) E G.
P ROOF. Let k ∈ ker( ϕ), i.e. ϕ(k) = 1 H . Then for g ∈ G
ϕ( gkg−1 ) = ϕ( g) ϕ(k ) ϕ( g)−1 = ϕ( g)1 H ϕ( g)−1 = 1 H
=⇒ gkg−1 ∈ ker( ϕ), done.
II.I.10. E XAMPLES .
(a) Both II.I.8 and II.I.9 give quick proofs that An E Sn :
• An = ker{sgn : Sn → Z2 } (identifying ({1, −1}, •) with (Z2 , +))
• [Sn :An ] = 2.
(b) SLn (F) = ker{det : GLn (F) → F∗ } C GLn (F), where F =

Q, R, C.
(c) hr i C Dn (index = 2).
As an application of normality we get a useful complement to

our earlier result on decomposing a group into a direct product of
subgroups II.E.11(iv). (Note that in (ii) below it isn’t enough to have
one of H or K normal in G — we need both.)
II.I.11. T HEOREM . Let H, K ≤ G (G finite) with H ∩ K = {1}.

(i) | H ||K | ≤ | G |.
(ii) If also H, K E G and equality holds in (i), then G ∼
= H × K.
P ROOF. (i) Define a map of sets
ϕ: H × K → G
(h, k) 7→ hk
This is 1-to-1: ϕ(h, k) = ϕ(h0 , k0 ) =⇒ hk = h0 k0 =⇒ (h0 )−1 h =
k0 k−1 ∈ H ∩ K = {1} =⇒ (h0 )−1 h = 1 = k0 k−1 =⇒ (h, k) = (h0 , k0 ).
Hence (by the pigeonhole principle) | H ||K | = | H × K | ≤ | G |.
(ii) By II.E.11(iv) we are done if (∀h ∈ H, k ∈ K) hk = kh. (Recall
in the proof of II.E.11 that this makes ϕ a homomorphism hence an
isomorphism.) Now K E G =⇒ (hkh−1 )k−1 ∈ K, while H E G
=⇒ h(kh−1 k−1 ) ∈ H. Hence hkh−1 k−1 ∈ H ∩ K = {1} =⇒
hk(kh)−1 = 1 =⇒ hk = kh.
II.I.12. D EFINITION . A group G is called simple if it contains no

normal subgroups apart from {1} and G.
II.I.13. E XAMPLE . Though we know that A4 contains V4 as a nor-

mal subgroup (hence is not simple), I claim that An is simple for
n ≥ 5.
P ROOF FOR A5 . (This gives an alternative approach to the method

of II.I.7 used in your HW to see this.) Let {1} 6= H E A5 , and
58 II. GROUPS
σ ∈ H \{1}. Write
σ = (123) , (12)(34) , (12345)

III II I
for the three non-identity cycle-types in A5 .

Case I: Set ρ := (132). Since H E A5 , H contains
(ρσρ−1 )σ−1 = (31245)(15432) = (134).

Case II: Set τ := (12)(35). Since H E A5 , H contains
(τστ −1 )σ−1 = (12)(54)(12)(34) = (354).

So in all cases (I, II, and III) H contains a 3-cycle. Since H E A5 and
the 3-cycles form a ccl17 in A5 , H contains all 3-cycles. But 3-cycles
generate A5 , and so H = A5 .
In light of this example and II.I.8, An can have no subgroups of

index 2 for n ≥ 5 even though 2|An | (= n!

2 ). This furnishes another
example of how the “converse of Lagrange” fails.
Now recall that for H ≤ G
G/H := set of left cosets of H in G

(with elements written gH).
We have | G/H | = | G |/| H | = [ G:H ].

If H E G, then left cosets equal right cosets, and we can make
G/H into a group, called a quotient group (or “factor group” in
some texts). Set
( aH )(bH ) := all elements of the form ahbh0 , h, h0 ∈ H.
II.I.14. P ROPOSITION . H E G ⇐⇒ ( aH )(bH )= abH (∀ a, b ∈ G ).
P ROOF. ( =⇒ ) : Using gH = Hg, one could write
( aH )(bH ) = HabH = abHH = abH.
17Recall that this is false for A , and is true for A because the stabilizer of a 3-cycle
4 5
contains a transposition.
Alternatively, and more explicitly,
ahbh0 = abb| −{z

1
hb} h0 = abh00 h0 ∈ abH
∈H
yields aHbH ⊂ abH; and conversely, abH = a1bH ⊂ aHbH yields

aHbH ⊃ abH.
( ⇐= ) : ( gH )( g−1 H ) = gg−1 H = H implies ghg−1 = ( gh)( g−1 1) ∈
H for all h ∈ H, so that gHg−1 ⊂ H (hence = H, by replacing g with
g−1 ).
II.I.15. R EMARK . This last Proposition is equivalent to [Jacobson,

def
Thm. 1.6], which states that the equivalence relation a ≡ b ≡ a−1 b ∈
H being compatible with multiplication is equivalent to normality
of H in G. Specifically, the “compatibility” requirement is that the
pairing and inversion be well-defined on equivalence classes (i.e. the
partition), and then “≡” is called a congruence.
II.I.16. C OROLLARY. If H E G, then G/H, together with coset mul-

tiplication, ( aH )−1 := a−1 H, and 1G/H := (1) H, forms a group. (The
|G|
order of this group is [ G:H ], and | H | if G is finite.)
P ROOF. By II.I.14, the set of cosets is closed under multiplica-

tion; associativity is automatic from associativity of the product on
G. Also, ( aH )(1H ) = aH and ( aH )( a−1 H ) = aa−1 H = 1H.
II.I.17. E XAMPLES . (a) We have nZ C Z (since Z is abelian), and

Z/nZ ∼ = Zn . (The elements of Z/nZ are of the form a + nZ, i.e.
cosets written additively.)
(b) The quotient group associated to An C Sn is just Sn /An ∼
= Z2 ,
with elements An and τAn , where τ is any transposition.
(c) H × {1} and {1} × K are both normal in H × K. The quotient
groups are K and H respectively.
(d) [HW] G/[ G, G ] yields an abelian group, called the “abelianiza-
tion” of G.
60 II. GROUPS
II.I.18. D EFINITION . Given H E G, the natural map
ν : G G/H
is the homomorphism obtained by sending g 7→ gH. [To check that

it is actually a homomorphism, write ν( g)ν( g0 ) = gHg0 H = gg0 H =
ν( gg0 ).]
Here is the “converse” of II.I.9:
II.I.19. C OROLLARY. Every normal subgroup of a group G is the kernel

of a homomorphism.
P ROOF. Given H E G, we have the natural map ν : G → G/H. I

claim that H = ker(ν):
• h ∈ H =⇒ ν(h) = hH = H = 1G/H =⇒ h ∈ ker(ν);
• k ∈ ker(ν) =⇒ kH = ν(k) = 1G/H (= H ) =⇒ k ∈ H.
II.I.20. F UNDAMENTAL T HEOREM OF G ROUP H OMOMORPHISMS .

Let ϕ : G → H be a group homomorphism, and write K := ker( ϕ). Then
K E G, and the map
ϕ̄ : G/K → ϕ( G ) (≤ H )
(∗)
gK 7−→ ϕ( g)
|G|
is an isomorphism of groups. (In particular, |K |
= | ϕ( G )|.)
P ROOF. We only need to check that ϕ̄ is an isomorphism.

• ϕ̄ is well-defined (as a map): Suppose gK = g0 K. (We must show
(∗)
that ϕ̄( gK ) = ϕ̄( g0 K ).) Then g0 = gk for some k ∈ K, and ϕ̄( g0 K ) =
(∗)
ϕ( g0 ) = ϕ( gk) = ϕ( g) ϕ(k ) = ϕ( g) = ϕ̄( gK ).
(∗)
• ϕ̄ is a homomorphism: Since ϕ is one, ϕ̄(( aK )(bK )) = ϕ̄( abK ) =
(∗)
ϕ( ab) = ϕ( a) ϕ(b) = ϕ̄( aK ) ϕ̄(bK ).
• ϕ̄ surjects onto ϕ( G ): Any ϕ( g) = ϕ̄( gK ).
• ϕ̄ is injective: ϕ̄( aK ) = 1 H =⇒ ϕ( a) = 1 H =⇒ a ∈ ker( ϕ) = K
=⇒ aK = K = 1G/H .
The following diagram nicely describes the situation, namely that

“ϕ factors through G/K”:
ϕ
G / H
<
ν "" . ϕ̄
G/K
It commutes (see the end of §I.A) in the sense that ϕ̄ ◦ ν = ϕ:
(∗)
ϕ̄(ν( g)) = ϕ̄( gK ) = ϕ( g).
II.I.21. C OROLLARY. If ϕ : G → H is a surjective homomorphism,

|G|
then G/ ker( ϕ) ∼= H. (In particular, | ker( ϕ)| = | H |.)
II.I.22. E XAMPLES . (a) We obtain Sn /An ∼

= Z2 again by using
sgn : Sn Z2 with kernel An .
(b) The map ψ : C∗ S1 := {z ∈ C∗ | |z| = 1} sending z 7→ z
|z|
has
ker(ψ) = R>0 . So C∗ /R>0 ∼
= S1 .
(c) Defining ϕ : R S1 by ϕ(r ) := e2πir , we have ker( ϕ) = Z, so
that R/Z ∼ = S1 .
(d) There is a homomorphism Φ : Q V4 with kernel ker(Φ) =
C ( Q) = {±1}; thus Q/{±1} ∼
= V4 . [HW]
(e) We construct a homomorphism φ : S4 S3 as follows: let S4 act
by conjugation on the ccl {(12)(34), (13)(24), (14)(23)}. Number-
ing its elements 1, 2, 3 in the order shown, we obtain φ, and calculate
that φ((12)) = (23) and φ((123)) = (132). Since φ(S4 ) ≤ S3 and
h(23), (132)i = S3 , we get surjectivity. By II.I.20 (or II.I.21),
|S4 | 24
= |S3 | =⇒ =6 =⇒ | ker(φ)| = 4.
| ker(φ)| | ker(φ)|
As ker(φ) E S4 , the only possibility is now ker(φ) = V4 . Conclude
that
(II.I.23) S4 /V4 ∼
= S3 .
62 II. GROUPS
The following is immediate from II.I.20 and Lagrange, and is use-

ful for ruling out homomorphisms between groups:
II.I.24. C OROLLARY. Let ϕ : G → H be a homomorphism, and | G |, | H |

finite. Then | ϕ( G )|| G |, | H |.
For a more serious application of the Fundamental Theorem, we

turn to the two isomorphism theorems for groups.
II.I.25. F IRST I SOMORPHISM T HEOREM . Let K E G, K ≤ H ≤ G.

Then:
(i) K E H
(ii) H/K ≤ G/K ( ) ( )
(†) subgroups of G subgroups
(iii) H 7→ H/K induces a bijection: ←→
containing K of G/K
(iv) H E G ⇐⇒ H/K E G/K.
(v) In case of (iv), G/H ∼ G/K
= H/K .
P ROOF. (i) is clear, and (ii) follows from II.I.20, viz.
h / hK
ϕ
H / G/K
;
ν !! - ϕ̄
H/K
(iii) injectivity of (†): Given H1 /K = H2 /K. Then for each h1 ∈ H1 ,
there exists h2 ∈ H2 such that h1 K = h2 K. But then h2−1 h1 ∈ K, and
so h1 = h2 k ∈ H2 . That is, we have shown that H1 ⊂ H2 . Similarly,
one has H2 ⊂ H1 ; and so H1 = H2 .
surjectivity of (†): Given H̄ ≤ G/K, H̄ is a collection of cosets.
Define H to be the union of these cosets (hence H ⊃ K), so that
h1 , h2 ∈ H =⇒ h1 K, h2 K ∈ H̄ =⇒ h1 h2 K = (h1 K )(h2 K ) ∈ H̄
=⇒ h1 h2 ∈ H (and similarly with inverses) =⇒ H ≤ G.

(iv) If H E G (and K E H, G), then
( gK )(hK )( g−1 K ) = ghg−1 K = h0 K ∈ H/K.

KEG H EG
The converse is similar.

(v) The composition
µ
&&
ν/ / ν̄ / / G/K
G G/K H/K
/ /
g gK ( gK )( H/K )
has ker(µ) = { g ∈ G | gK ∈ H/K } = H. [Check: gK ∈ H/K means

gK = hK for some h ∈ H, hence h−1 gK = K =⇒ h−1 g ∈ K =⇒
g = hk ∈ H.] Now apply II.I.21: in a diagram,
ϕ
G / / G/K
H/K
<
!! ∼
=
G/H
since H = ker(µ).
II.I.26. C OROLLARY. Given a homomorphism η : G G with kernel
K, let
Λ : = { H ≤ G | H ≥ K } ⊇ Λ 0 : = { H E G | H ≥ K }.
Then
(i) Sending H 7→ η ( H ) induces 1-to-1 correspondences
Λ ←→ {subgroups of G}
∪ ∪
Λ ←→ {normal sgps. of G}.
0
(ii) For H ∈ Λ0 , sending gH 7→ η ( g)η ( H ) induces

∼
=
G/H −→ G/η ( H ).
P ROOF. By II.I.21, G ∼
= G/K. Hence this is just parts (iii-iv) resp.
(v) of II.I.25.
II.I.27. S ECOND I SOMORPHISM T HEOREM . Let H ≤ G, K E G.
Then
(i) (K E) HK ≤ G.
64 II. GROUPS
(ii) H ∩ K E H.
∼
=
(iii) h(K ∩ H ) 7→ hK induces H/(K ∩ H ) → HK/K.
P ROOF. (i) HK = ∪h∈ H hK = ∪h∈ H Kh = KH implies that ( HK )2 =
H 2 K2 = HK, and also that (the set of all inverses of elements of HK)
( HK )−1 = K −1 H −1 = KH = HK. So HK is a subgroup of G.
(ii) Under ν : G G/K,
ν( H ) = {hK | h ∈ H } = {hkK | hk ∈ HK } = HK/K.
This image is a subgroup of G/K. So we get by restriction a homo-

morphism of groups ν| H : H HK/K, with ker(ν| H ) = { h ∈ H |
hK = K } = H ∩ K.
(iii) The diagram
ν| H
H / / HK/K
8
%% ∼
=
H/( H ∩ K )
provides the desired isomorphism, courtesy of II.I.21.

As an application, we finish off Example II.I.13:18
P ROOF THAT An IS SIMPLE FOR n ≥ 5. Having done n = 5 (the
base case) above, we induce on n (taking n ≥ 6). Suppose K E
An , and consider (for each i ∈ {1, . . . , n}) the subgroup Hi ≤ An of
even permutations fixing i; clearly Hi ∼ = An−1 , which is simple. By
II.I.27(ii), we have Hi ∩ K E Hi , hence Hi ∩ K = {1} or Hi . If it is Hi
for some i, then Hi ≤ K and so K contains a 3-cycle. But 3-cycles are
a ccl in An (since n > 4, by II.G.19), and these generate An , forcing
K = An .
So suppose K ∩ Hi = {1} for all i. Then any σ ∈ K \{1} must be
a product of r disjoint cycles of the same length k, with rk = n. (If
there were cycles of different lengths j < k in the decomposition of σ,
then σ j 6= 1 but fixes some i, so that Hi ∩ K 6= {1}, a contradiction.)
18There are also direct (but lengthier) arguments in the style of that example or
your HW.
Since n ≥ 6, we can choose τ = ( ab)(cd) ∈ An and i so that i, σ(i ) are

distinct from a, b, c, d, and so that τ and σ do not commute.19 Then
σ−1 (τστ −1 ) ∈ K since K E An ; but it also fixes i (hence belongs
to Hi ∩ K) and isn’t the identity, a contradiction. Thus there is no
σ ∈ K \{1} and K = {1}.
19This is easy, and left to you. Consider separately the cases k = 2 (which doesn’t
occur for n = 6) and k > 2.
66 II. GROUPS
II.J. Automorphisms
∼
=
II.J.1. D EFINITION . An isomorphism ϕ : G → G is called an auto-
morphism of G.
II.J.2. E XAMPLES . (i) The identity map idG is an automorphism

of any group.
∼
=
(ii) Conjugation by g ∈ G is denoted ı g : G → G; automorphisms of
this type are called inner. (The conjugation must be by an element of
G, not by an element of some larger group it sits in!) Abelian groups
have no non-identity inner automorphisms.
(iii) If G E G 0 , then conjugation by g0 ∈ G 0 does give an automor-
phism of G (but this may or may not be inner).
(iv) In Example II.I.22(e), S4 acted by conjugation on the ccl
{(12)(34), (13)(24), (14)(23)} = V4 \{1}.

That is, for each σ ∈ S4 , ıσ induces a permutation of V4 \{1} ( =⇒
element of S3 — we got all elements of S3 this way). In fact, each ıσ
induces an automorphism of V4 (since V4 E S4 ) and [except for the
identity] these are non-inner (as V4 is abelian).
Write
Aut( G ) := the set of automorphisms of G, and

Inn( G ) := the set of inner automorphisms of G.
II.J.3. P ROPOSITION -D EFINITION . Aut( G ) is a group under compo-

sition of maps, as is Inn( G ); and Inn( G ) E Aut( G ). So we can define the
group of outer automorphisms by Out( G ) := Aut( G )/Inn( G ). If G is
abelian, then Out( G ) = Aut( G ).
P ROOF. The composition of two isomorphisms is again an iso-

morphism; isomorphisms are invertible; and IdG is an isomorphism.
The same goes for inner automorphisms: e.g.,
(ı g ◦ ıh )( x ) = g(hxh−1 ) g−1 = ( gh) x ( gh)−1 = ı gh ( x ).

II.J. AUTOMORPHISMS 67
Finally, for x ∈ G and α ∈ Aut( G ),
(α ◦ ı x ◦ α−1 ( g) = α( xα−1 ( g) x −1 )
= α( x )α(α−1 ( g))α( x )−1
| {z }
=g
= ıα( x ) ( g )
=⇒ αInn( G )α−1 ⊆ Inn( G ).
II.J.4. E XAMPLES . (i) Aut(V4 ) ∼

= S3 , so we can see Ex. II.I.22(e)
ı(·)
in terms of a surjective homomorphism S4 Aut(V4 ) (with kernel
V4 ). So we see that the automorphism group of an abelian group
need not be abelian.
(ii)20 Aut(Zn ) ∼
= Z∗n . To see this, consider
µ : Z∗n → Aut(Zn )
ā 7−→ µ ā := multiplication by ā.
For injectivity of µ: suppose µ ā = idZn ; then µ ā (b̄) = b̄ for any

b̄ ∈ Zn , and taking b̄ = 1̄ gives ā = 1̄.
For surjectivity of µ: let α ∈ Aut(Zn ), and set ā = α(1̄). Now
(µ ā − α)(b̄) = µ ā (b̄) − α(b̄)

= āb̄ − α(1̄| + ·{z
· · + 1̄})
b times
= āb̄ − α(1̄) · b̄
|{z}
ā
= 0̄ (∀b̄)
=⇒ µ ā = α, so α ∈ im(µ).
We finish this section with a striking result.

20Here we recall that Z∗ = { ā | ( a, n) = 1} under multiplication mod n. It’s a
n
group because the gcd being 1 means that there exist r, s ∈ Z such that ra + sn = 1,
i.e. r̄ ā ≡ 1̄ and so r̄ = ā−1 . Similarly, µ ā below — which is a homomorphism from
(n)
Zn → Zn by the distributive law — has inverse µ ā−1 , making it an automorphism
of Zn .
68 II. GROUPS
II.J.5. T HEOREM . Let n > 2.

(i) Inn(Sn ) ∼ = Sn .
(ii) Assume n 6= 6. Then Aut(Sn ) ∼ = Inn(Sn ).
(iii) For n = 6, this is false (and Out(S6 ) ∼
= Z2 ).
P ROOF. (i) We want to show that ı : Sn → Aut(Sn ), the map

sending g 7→ ı g , is injective — in other words, that C (Sn ) = {1}. Let
σ ∈ Sn \{1} be given; it moves at least one number in {1, . . . , n}, say
a 7→ b. Take any c 6= a, b in {1, . . . , n}; then (bc)σ sends a 7→ c, while
σ(bc) sends a 7→ b. So σ ∈ / C (Sn ), done.
(ii) Any α ∈ Aut(Sn ) sends conjugate elements to conjugate ele-
ments (why?). Hence if α is going to move an element of one conju-
gacy class ccl1 into a different conjugacy class ccl2 , it must send all
of ccl1 into ccl2 , and its inverse does the reverse. So we would have
to have |ccl1 | = |ccl2 |, and moreover (since automorphisms send el-
ements of order k to elements of order k) that elements of ccl1 have
the same orders as those in ccl2 . The goal of this proof is to show
that these constraints on an automorphism messing with ccl’s are so
tight that it never happens except for n = 6.
Now the ccl’s in Sn with elements of order 2 are the
 
 σ has cycle structure
 
Ck := σ ∈ Sn (··) · · · (··)(·) · · · (·)


 | {z }| {z }  
k n−2k

(i.e. products of k disjoint transpositions) for 1 ≤ k ≤ b n2 c, with

n!
|Ck | = .
(n − 2k)!k!2k
We have
n! n!
|Ck | = |C1 | ⇐⇒ k
=
(n − 2k)!k!2 (n − 2)!2
( n − 2) !
⇐⇒ = k!2k−1
(n − 2k)!
n−2 k!2k−1

⇐⇒ = ;
2k − 2 (2k − 2)!
II.J. AUTOMORPHISMS 69
k!2k−1
but the binomial symbol is an integer, whereas (2k−2)!
is not an inte-
ger for k ≥ 4. Moreover, the k = 2 case (n− 2
2 ) = 2 is also impossible.
n −2
This leaves k = 3, and ( 4 ) = 1, which holds ⇐⇒ n = 6. We
conclude that for n 6= 6, α(C1 ) = C1 .
Now assume that n 6= 6, and let an automorphism α be given.
We have just shown that α sends transpositions to transpositions.
Suppose α((12)) = ( ab), and x ∈ {3, . . . , n}; then
(12)(1x ) = 3-cycle =⇒ α((12)(1x )) = ( ab)α((1x )) = 3-cycle

=⇒ α((1x )) = ( ac) or (bc) c 6= a, b
Without loss of generality (by swapping a and b if necessary) we may
assume α((1x )) = ( ac). With this assumption in place, we make the
Claim: α((1y)) = ( ad) (for some d 6= a) for any y ∈ {2, . . . , n}. [HW]
Taking this claim for granted, define a permutation of {1, . . . , n} by
σ(1) := a, σ (y) := this “d” for each y 6= 1, and compute ıσ−1 α((1y)) =
ıσ−1 (( ad)) = (1y). So (ıσ−1 ◦ α) is the identity on all (1y)’s. But trans-
positions generate Sn , and since (yy0 ) = (1y0 )(1y)(1y0 ), the (1y)’s
generate Sn all by themselves. It follows that ıσ−1 ◦ α = idSn , and so
α = ıσ−1 is an inner automorphism.
(iii) If α is inner, it has to stabilize ccl’s, not permute them. The com-
putation above suggests that there may be an automorphism α with
α(C1 ) = C3 , which would have to be outer. Constructing this will be
an application of Sylow theory, so we defer the proof of this part.
70 II. GROUPS
II.K. Generators and relations
The abelian case. Let G be an abelian group. We will write the

group operation as “+”. Note that for g ∈ G and a ∈ Z, the nota-
tion ag means adding g to itself a times (or, if a < 0, its inverse − g
to itself | a| times). So it is the equivalent of exponentiation in the
multiplicative notation.
II.K.1. P ROPOSITION . The following are equivalent:

(i) G = { a1 g1 + · · · + an gn | ai ∈ Z} for some g1 , . . . , gn ∈ G, called a
generating set for G.
(ii) G ∼= Zn /K for some n ∈ N, K ≤ Zn .
P ROOF. If (i) holds, define ϕ : Zn G to send a := ( a1 , . . . , an ) 7→

∑i ai gi . By the Fundamental Theorem, G ∼ = Zn / ker( ϕ).
Conversely, assuming (ii), write η for the composition
ν ∼
=
Zn Zn /K → G,
and set gi := η (ei ) (where ei is the ith standard basis vector). Every
element of Zn is of the form ∑i ai ei , and η is surjective; thus, every
element of G is of the form η (∑i ai ei ) = ∑i ai η (ei ) = ∑i ai gi .
II.K.2. D EFINITION . (i) If the equivalent conditions of II.K.1 hold,

G is finitely generated (f.g.).
(ii) K is called the relations subgroup for G.
(iii) If G ∼
= Zm (for some m), G is (f.g.) free abelian of rank m. The
image of the standard basis {ei }im=1 ⊂ Zm under the isomorphism is
called a basis of G.
II.K.3. E XAMPLES . (i) Zn is f.g. (with one generator: 1̄), and iso-
morphic to Z/nZ.
(ii) Q is not f.g.: if you pick rs11 , . . . , rsnn then any ∑in=1 ai rsi can be repre-
i
sented with denominator ∏i si — clearly not possible for an arbitrary
rational number.
II.K. GENERATORS AND RELATIONS 71
(iii) Suppose G ∼= Z3 /K, and K ∼ = Z2 with basis (11, −21, −10),

(1, −6, −5). Then we can write G in terms of “generators and rela-
tions”:21
Zh X, Y, Z i
G∼= .
h11X − 21Y − 10Z, X − 6Y − 5Z i
The key here is using the fact that K is free, and further, having a
basis for K. The next result and its proof generalize this:
II.K.4. T HEOREM . Every subgroup of a free f.g. abelian group is free
f.g.; more precisely, any K ≤ Zn is ∼
= Zm for some m ≤ n.
P ROOF. If n = 1, let a ∈ N ∩ K be as small as possible. If b ∈
K \{0} is not a multiple of a, then gcd( a, b) = `1 a + `2 b ∈ K, and is
less than a, a contradiction. So K = h ai ∼
= Z.
Now, assuming the statement for n − 1, consider the projection
π : K → Z to the first Z-factor. If π (K ) = {0}, we’re done by in-
duction (as ker(π ) ≤ Zn−1 ). Otherwise, π (K ) (≤ Z) consists of
multiples of some a = π (α), α ∈ K. Hence any β ∈ K is of the form
π ( β) π ( β)
(β − a α) + a α ∈ ker (π ) + hαi ,
and ker(π ) ∩ hαi = {0}. So by (say) II.E.11(iii), K ∼
= ker(π ) × hαi,
and applying the inductive assumption to ker(π ) ≤ Zn−1 , we are
done. (Note that the proof also yields a method for constructing a
basis, starting with α.)
In fact, the group in Ex. II.K.3(iii) is ∼
= Z45 × Z, which inspires
the next statement:
II.K.5. P ROPOSITION -D EFINITION . (Let G be abelian.) The subset
Gtor ⊆ G comprising elements of finite order is a subgroup, the torsion
part of G; while G/Gtor is a free abelian group (all nonzero elements are of
infinite order), the free part of G. (If G is f.g., this is ∼
= Zm for some m.)
P ROOF. Given g1 , g2 ∈ Gtor , we have ai ∈ N with ai gi = 0. Then
lcm( a1 , a2 ) · ( g1 + g2 ) = 0 =⇒ g1 + g2 ∈ Gtor . (So it’s closed under
addition — the rest is trivial.)
21The notation Zh X, Y, Z i means the free abelian group with basis X, Y, Z; the
denominator means the subgroup generated by those two elements.
72 II. GROUPS
Given g ∈ G \ Gtor , if ag ∈ Gtor for some a ∈ N, then there exists

b ∈ N such that 0 = b( ag) = (ba) g, making g ∈ Gtor , a contra-
diction. So g has infinite order in G/Gtor . (I skip the proof of the
parenthetical for now; we will return to f.g. abelian groups in the
context of modules.)
II.K.6. R EMARK . Prop. II.K.5 is false for nonabelian groups. There

is no reason, if g1 and g2 don’t commute, why g1a = 1 and g2b = 1
should imply that g1 g2 has finite order. One example is22 PSL2 (Z),
which is generated by R = − 11 0 −1 . These elements

−1 0 and S = 1 0
satisfy R3 = 10 01 = S2 (i.e. have finite order), but their product

RS = 10 11 has infinite order.

The general (non-abelian) case. We return to multiplicative no-

tation. Given a subset S ⊆ G, we defined the subgroup generated by
S as
hSi := smallest subgroup of G containing S.
For later use, also write
hhSii := smallest normal subgroup of G containing S.

A set of generators for G is a subset S such that hSi = G (and it
is minimal if for all S0 ( S, we have hS0 i < G). We say that G is
finitely generated iff there exists a finite set S with G = hSi. Having
a (small) generating set is useful because of the following
II.K.7. P ROPOSITION . A homomorphism ϕ : G → H is defined by its

behavior on a generating set. That is, if G = hSi and ϕ, η are homomor-
phisms with ϕ(s) = η (s) (∀s ∈ S), then ϕ = η.
P ROOF. Any g ∈ G may be written in the form g = s1 · · · s N with

si ∈ S (and possible repetitions). Hence, ϕ( g) = ϕ(s1 ) · · · ϕ(s N ) =
η ( s1 ) · · · η ( s N ) = η ( g ).
II.K.8. P ROPOSITION . Given ϕ : H → G, if ϕ( H ) ⊃ S and hSi = G,

then ϕ is surjective.

22SL (Z) quotiented by the normal 2-element subgroup generated by −1 0 .
2 0 −1
P ROOF. Since ϕ( H ) is a group, hSi ≤ ϕ( H ).
Now let S be a set, not a subset of a group, just a set. Consider

the set of words on S , by which we mean the set of expressions
m
s1m1 s2m2 · · · sk k (k ≥ 0, si ∈ S , mi ∈ Z)
subject only to the (equivalence) relation s a sb = s a+b (for each s ∈ S ).
Denote this set by23 hSi, and introduce the binary operation of “con-
−m
catenating words” together with the obvious inverses sk k · · · s1−m1
to put a group structure on it. (Clearly the subset S generates the
resulting group hSi!!) More intrinsically, we have the
II.K.9. P ROPOSITION -D EFINITION . There exists a unique group
FS ⊃ S
with the (universal) property that: for all groups G and maps f : S →
G, there exists a unique homomorphism ϕ : FS → G making the
diagram

S / F
S
f ~ ϕ
G
commute. In fact, FS ∼
= hSi. It is called the free group on S .
P ROOF. First we prove existence by showing that hSi has this

m
property. Define ϕ : hSi → G by ϕ(s1m1 · · · sk k ) = f (s1 )m1 · · · f (sk )mk .
This is clearly well-defined and a homomorphism, and any other
homomorphism η making the diagram commute must have η (s) =
f (s) for all s ∈ S , hence (by II.K.7) η = ϕ.
23This designation is temporary, as — while standard — it is likely to get confused

with the other meaning of hSi for a subset of a group. After II.K.9 we will be using
FS instead.
74 II. GROUPS
For uniqueness, suppose F and G are two groups containing S

as a subset and satisfying the universal property. Then there are ho-
momorphisms ϕ and η making
6 F 6 FO
( (
S v ϕ and S v η

( (
G G
commute. But then
6 F 6 G
( )
S v η◦ ϕ and S u ϕ◦η

( (
F G
commute as well, and then the uniqueness in the universal property
gives η ◦ ϕ = idF and ϕ ◦ η = idG . So F ∼
= G and we are done.
Henceforth I will drop hSi for free groups and use it only for
subgroups generated by a subset.
II.K.10. R EMARK . A similar characterization exists for the free

abelian group AS on S . In II.K.9, wherever “group(s)” occurs, re-
place it by “abelian group(s)”, and replace hSi by the group of finite
formal sums m1 s1 + · · · + mk sk with k ≥ 0, mi ∈ Z and si ∈ S . In
the (modified) first paragraph of the proof, ϕ(m1 s1 + · · · + mk sk ) :=
f (s1 )m1 · · · f (sk )mk is well-defined and a homomorphism precisely
because G is abelian.
Now let S ⊂ G be a finite generating set. We have by II.K.8-II.K.9

a (surjective) homomorphism
ϕ : FS G
with ϕ(s) = s for each s ∈ S . By the Fundamental Theorem,
G∼
= FS / ker( ϕ),
where of course ker( ϕ) is normal; and if ker( ϕ) = hhRii for some

subset R ⊂ FS , this becomes
(II.K.11) G∼
= FS /hhRii
— a presentation of G in terms of generators S and relations R. If
|R| < ∞, we say that G is finitely presented. We conclude with
some
II.K.12. E XAMPLES . (i) Dn ∼

= F{r,h} /hhr n , h2 , rhrhii.
(ii) PSL2 (Z) ∼
= F{S,R} /hhS2 , R3 ii.
(iii) [HW] AS ∼
= FS /[FS , FS ] for any set S .
The next two examples illustrate the role these concepts play in
algebraic topology and complex analysis.
(iv) A compact Riemann surface C of genus g is, topologically, the sur-
face of a sphere with g handles attached, or of a donut with g holes.
α
β
β β
α α
g=1 α2
α1 β2 β1
β1
β2 α2 α2 α1
g=2
β2 β1
α1
Choosing a point x ∈ C, its fundamental group π1 (C ) is the set of

closed curves starting and ending at x modulo the equivalence rela-
tion given by continuous deformation;24 the group operation is con-
catenating loops and inversion is reversing the direction. In fact, it
24More precisely, a closed curve is a continuous map γ : [0, 1] → C with γ(0) =

γ(1); and γ0 and γ1 are equivalent if there is a continuous map Γ : [0, 1] × [0, 1] →
C with γ0 (t) = Γ(0, t) and γ1 (t) = Γ(1, t).
76 II. GROUPS
is the quotient of a free group on certain loops (shown for g = 1, 2)

modulo a single relation:
g
π1 ( C ) ∼
= F{α1 ,β1 ,...,αg ,β g } /hh∏i=1 [αi , β i ]ii.
The relation arises from cutting open the surface as shown, then ob-
serving that the boundary can be continuously deformed to a point.
(To see that the boundary curve is the product of commutators shown,
start at the red dot with β resp. β 2 .)
N2 1
(v) Let N ≥ 3, and set κ := 2 ∏ p | N (1 − p2
) (where p is prime) and
N −6
g := 1 + 12 κ. Recall the congruence subgroups
Γ( N ) := ker{SL2 (Z) → SL2 (Z/NZ)}.
Let H := { x + iy | x ∈ R, y ∈ R>0 } denote the upper half-plane in

C. We let Γ( N ) act on H by fractional linear transformations, with
a b sending z 7 → az+b . The “quotient set”

c d cz+d
Y ( N ) := H/Γ( N )
obtained25 by identifying all points related by Γ( N ), is a genus g Rie-

mann surface with κ points removed. Moreover, writing γ1 , . . . , γκ
for loops around these points, we have
g
Γ( N ) ∼
= π1 (Y ( N )) ∼
= F{α1 ,β1 ,...,αg ,β g ,γ1 ,...,γκ } /hh∏i=1 [αi , β i ] ∏κj=1 γ j ii.
In particular, Γ( N ) is finitely presentable (in fact, it is also torsion-
free).
25That is, Y ( N ) is the set of orbits of the group action on H.

II.L. THE SYLOW THEOREMS 77
II.L. The Sylow theorems
Recall that a group of prime power order is called a p-group.

II.L.1. D EFINITION . Let G be a finite group with | G | = pk m, p
prime and p - m. A Sylow p-subgroup of G is a subgroup of order
pk , the maximum possible power of p.
We already know that these exist when k = 1, by Cauchy’s theo-
rem (which in fact guarantees an element of order p in G if k ≥ 1). To
attack the general case, we briefly recall what we will need on group
actions: if { xi } are representatives of the G-orbits in X with more
than one element, and XG denotes the fixed points (i.e. the union of
1-element orbits), then
|X| = |XG | + ∑ | G ( xi )|
i
= |X | + ∑[ G:Gxi ]
G
i
by the orbit-stabilizer theorem. Further, if X = G and G acts on itself

by conjugation, this becomes the class equation
(II.L.2) | G | = |C ( G )| + ∑[ G:CG ( xi )]
i
in which { xi } are representatives of ccl’s with > 1 element.26 The

main points here are that

• [ G:CG ( xi )]| G |, and
• [ G:CG ( xi )] 6= 1 (otherwise xi ∈ C ( G )).
We are now ready for
II.L.3. F IRST S YLOW T HEOREM . Every finite group has a Sylow p-
subgroup for each prime p dividing its order:
| G | = pk m =⇒ ∃ H ≤ G with | H | = pk .
P ROOF. Assume the theorem holds for all groups of order less
than | G |. (The base case is just the one-element group {1}.) Here is
the inductive step.
26So if G is abelian, there are no x ’s (and one is in case (2) in the next proof).
i
78 II. GROUPS
Clearly one of the following must be true: either

(1) pk |CG ( xi )| for some i; or

(2) pk does not divide |CG ( xi )| for any i.

In case (1), |CG ( xi )| = pk n is less than | G |. By the inductive hypoth-
esis, there exists H ≤ CG ( xi ) with | H | = pk . Since CG ( xi ) ≤ G,
H ≤ G.
In case (2), since |CG ( xi )| · [ G:CG ( xi )] = | G | = pk m, p divides
[ G:CG ( xi )] for every i. Hence, in the class equation (II.L.2), p di-
vides everything but |C ( G )|, and thus must also divide |C ( G )|. By
Cauchy’s theorem, there is a subgroup K ≤ C ( G ) of order p. Since
conjugation affects no element in C ( G ), K E G. We can thus speak of
the quotient group G/K with order pk−1 m, which by the inductive
hypothesis contains a subgroup H0 of order pk−1 . Let
ϕ : G G/K
be the quotient map. Since this map is p-to-1,27 the preimage ϕ−1 ( H0 ) ≤
G — which is a subgroup by II.I.25(iii) — has order pk as desired.
Recall that the normalizer of a subgroup H ≤ G is the largest

subgroup of G in which H is normal:
NG ( H ) := { g ∈ G | gHg−1 = H }.
If K ≤ NG ( H ), then KH = HK =⇒ KH is a group. We’ll need a

lemma for the proofs of the remaining Sylow theorems.
II.L.4. L EMMA . Let P1 , P2 ≤ G be Sylow p-subgroups, and suppose

P1 ≤ NG ( P2 ). Then P1 = P2 .
P ROOF. Write | G | = pk m, so | P1 | = | P2 | = pk and (since the

intersection of subgroups is a subgroup of each) | P1 ∩ P2 | = ps (for
some s ≤ k). Moreover,

P1 ≤ NG ( P2 ) =⇒ P1 P2 ≤ G =⇒ | P1 P2 || G |
27The quotient (natural map) ν : G G/H by a normal subgroup H E G of order
r is always an r-to-1 map: each coset of H has r elements, and you are collapsing
each coset to a single element.
and since P2 ≤ P1 P2 ,
pk = | P2 || P1 P2 | =⇒ | P1 P2 | = pk n

where n | m. Now use the product formula | P1 P2 || P1 ∩ P2 | = | P1 || P2 |

(from HW 4 #2), which gives pk n · ps = pk · pk . Since n was a factor of
m which was relatively prime to p, this yields a contradiction unless
n = 1. Conclude that | P1 P2 | = pk . But now P1 and P2 are subgroups
of P1 P2 and all three have the same order; hence all three are equal.

Let X denote the set of all Sylow p-subgroups of G, which is

nonempty by Sylow I. Write X = { P1 , P2 , . . . , PN }, so that N = |X|.
II.L.5. S ECOND S YLOW T HEOREM . N ≡ 1.

( p)
P ROOF. We let P1 act by conjugation on X. This makes sense be-

cause the conjugate gPi g−1 of a group of order pk still has order pk
(because conjugation is an isomorphism). This group action decom-
poses X into orbits; taking a system of representatives { Pj } of the
orbits of order > 1, we have
(II.L.6) XP1 | + ∑ | P1 ( Pj )|.

N = |X| = ||{z}
j
| {z }
fixed
> 1 elt.
pts.
orbits
By the orbit-stabilizer theorem, 1 < | P1 ( Pj )|| P1 | = pk . So p divides

all the terms in the RHS of (II.L.6) except possibly |XP1 |.

Now consider the fixed points: to have Pi ∈ XP1 means that
gPi g−1 = Pi (∀ g ∈ P1 ); and so by the definition of the normal-
izer, P1 ≤ NG ( Pi ). Lemma II.L.4 then tells us that P1 = Pi : that
is, i = 1 and P1 ∈ X is the only fixed point. So (II.L.6) reads N =
1 + {multiple of p} and we are done.
II.L.7. T HIRD S YLOW T HEOREM . All Sylow p-subgroups are conju-

gate. (That is, if P1 , P2 ≤ G are two such, then there exists g ∈ G such that
gP1 g−1 = P2 .)
80 II. GROUPS
P ROOF. Let Γ denote the set of left cosets of P2 , and consider

the action of P1 on Γ by left-multiplication. Suppose that the set of
fixed “points” Γ P1 is nonempty, and let gP2 be one of them: that is,
hgP2 = gP2 (∀h ∈ P1 ). Then
g−1 hgP2 = P2 =⇒ g−1 hg ∈ P2 =⇒ hg ∈ gP2
=⇒ h ∈ gP2 g−1 (∀h ∈ P1 ) =⇒ P1 ≤ gP2 g−1

=⇒ P1 = gP2 g−1 , since P1 and P2 have the same order. So we just
need to show |Γ P1 | 6= 0.
As before, we have (by counting orbits)
|Γ| = |Γ P1 | + ∑{ jth orbit}

size of
with the sum terms divisble by p (by the orbit-stabilizer theorem

and the fact that a p-group is acting). So on the one hand, we have
|Γ P1 | ≡ |Γ|. On the other, by Lagrange we have
( p)
|G| pk m
|Γ| = # of cosets of P2 = [ G:P2 ] = = k = m 6≡ 0.
| P2 | p ( p)
Hence, |Γ P1 | 6= 0.
Here are two more important results on p-groups and p-subgroups

(really, a refinement of Sylow I).
II.L.8. P ROPOSITION . Suppose | G | = pe . Then for any k ≤ e, there

exists a normal subgroup H E G of order pk .
P ROOF. (Assume true for pe−1 ; this is the inductive step.) We

know C ( G ) 6= {1} by II.H.8; so p|C ( G )| and by Cauchy, there exists
a ξ ∈ C ( G ) of order p. Since any subgroup of its center is normal
in G, hξ i E G; we may therefore consider G/hξ i a group of order
pe−1 . Applying the inductive hypothesis yields K E G/hξ i of order
pk−1 . We claim that its preimage under η : G G/hξ i, namely H :=
η −1 (K ), is the desired subgroup of G: indeed, H E G by II.I.25(iv);
and | H | = p · pe−1 since η is p-to-1.
II.L.9. C OROLLARY. Suppose | G | = pe m, where p - m. Then for any

1 ≤ k ≤ e, there exists a subgroup H ≤ G of order | H | = pk .
P ROOF. By Sylow I, G has a Sylow p-subgroup P ≤ G (of order

pe ). Applying II.L.8 to P yields H ≤ P of the correct order.
How might one use Sylow III, with its characterization of set of
Sylow p-subgroups of G as one big orbit under conjugation? The
orbit-stabilizer theorem says that the size of any orbit must divide
the order of G; so we get immediately that
the number N of Sylow p-subgroups

(II.L.10)
divides the order | G | of the group.
Together with Sylow II this frequently gives enough information to
determine N.
II.L.11. E XAMPLE . S5 has six 5-Sylow subgroups.

P ROOF. N |S5 | = 5! and N ≡ 1 =⇒ N = 6 or 1. But there’s
(5)
more than one, as h(12345)i and h(12354)i are distinct.
In fact, we can use this to show that S6 has an outer automor-

phism, thereby finishing off Theorem II.J.5:
P ROOF. The action of S5 (by conjugation) on its six 5-subgroups

is transitive by Sylow III, hence gives a map
ϕ : S5 → S6
|S |
with | ϕ(S5 )| ≥ 6 =⇒ | ker( ϕ)| = ϕ(S5 ) ≤ 20. But by HW 3, A5 (of
5
order 60) is the only nontrivial normal subgroup of S5 ; hence ker( ϕ),
being normal in S5 , must be trivial, and ϕ injective.
Now we claim that ϕ preserves parity (i.e. σ odd =⇒ ϕ(σ ) odd).
Suppose first of all that ϕ(S5 ) was contained in A6 ; then it has index
3 and the action of A6 (by left translation) on its cosets maps A6 →
S3 nontrivially hence (since A6 is simple and kernels are normal)
injectively. But this contradicts |A6 | > |S3 |, and so ϕ(S5 ) 6⊂ A6 ,
ϕ sgn
making the composition S5 → S6 → Z2 surjective. Again, A5 is
82 II. GROUPS
the only nontrivial proper normal subgroup of S5 , so it must be the

kernel of the composition. Thus the composition is sgn : S5 → Z2 ,
proving the claim.
Consider the homomorphism from S6 to itself obtained by let-
ting S6 act on the six cosets of ϕ(S5 ) (∼
= S5 ) by left translation. This
map
α : S6 → S6
has image of order ≥ 6, hence (by arguing as above) is injective, and
thus an isomorphism. That is, α ∈ Aut(S6 ).
Recall that to prove α is not inner, we only have to show that it
sends a transposition to a non-transposition.28 Suppose α((12)) =
( ab). Then α((12)) is swapping two cosets of ϕ(S5 ) and fixing the
other four. Let xϕ(S5 ) be one of the fixed cosets, so that (12) xϕ(S5 ) =
xϕ(S5 ) =⇒ x −1 (12) x ∈ ϕ(S5 ). Define σ ∈ S5 by ϕ(σ) = x −1 (12) x.
Since x −1 (12) x is odd, and ϕ preserves parity, σ is odd; it is also of
order 2 in S5 , and hence must be a transposition.
Since ϕ(σ) is a transposition, σ’s action on the 5-Sylow subgroups
of S5 swaps two and normalizes four. Let P = h(12345)i be one of
the latter (relabeling if needed). We may assume σ (1) = 1, and so
σ (12345)σ−1 = (1 σ(2) σ(3) σ(4) σ(5))

∈ P = {(12345), (13524), (14253), (15432), 1S5 } ,
which is visibly impossible if σ is a transposition. (No element of P
results from swapping two numbers in (12345).)
So α((12)) cannot be a transposition and we are done.
28In fact, this “non-transposition” had to be a product of three disjoint transpo-

sitions: in the notation of the proof of II.J.5(ii), a non-inner α (assuming it exists)
must exchange C1 and C3 . Notice that its square will then fix C1 and is thus in-
ner. In fact, its composition with any β exchanging C1 and C3 also fixes C1 , which
shows that Out(S6 ) ∼ = Z2 .
II.M. SOME RESULTS ON FINITE GROUPS 83
II.M. Some results on finite groups
Low order. Let’s review what we already know about classifying

these. By Lagrange’s theorem, if a group has prime order = p, then
any element 6= 1G has order p, hence generates the group. So
(II.M.1) (a) groups of prime order are cyclic (∼

= Z p ).
This covers orders 2, 3, 5, 7, 11, 13, . . .. Next,
(b) groups of order 2p are either cyclic (∼

= Z2p ) or dihedral (∼
= D p ).
This takes care of orders 6, 10, 14, . . .. Finally,
(c) groups of order p2 (p prime) are abelian and ∼

= Z p × Z p or Z p2 .
This finishes off orders 4, 9, . . .. Between 1 and 16 this leaves orders
8 , 12 , and 15.
II.M.2. T HEOREM . The groups of order 8 are (up to ∼

=)
Z8 , Z4 × Z2 , Z2 × Z2 × Z2 , Q, and D4 .
P ROOF. We begin with the abelian case. By II.L.8 there is a sub-

group H of order 4. Any g ∈ G \ H, together with H (∼ = Z2 × Z2 or
Z4 ) generates G.
If |h gi| = 2, G is Z2 × Z2 × Z2 or Z2 × Z4 . (Use II.E.11(iv).)
If |h gi| = 8, G ∼
= Z8 (and H ∼= Z4 ).
If |h gi| = 4: we need to show that there is an element g0 ∈ G
of order 2 and different from 2g. (Then II.E.11(iv) implies that G ∼ =
∼ ∼
h gi × h g i = Z4 × Z2 .) Under G G/H = Z2 , we have g 7→ 1̄ =⇒
0
2g 7→ 0̄ =⇒ 2g ∈ H (of order 2). If H = hhi ∼ = Z4 , then 2g = 2h

and we can take g := g − h. If H ∼
0
= Z2 × Z2 , then we take g0 to be
an element of H other than 1 and 2g.
Turning to the nonabelian case: clearly, we can’t have an element
of order 8. Also, were every non-identity element of order 2, we’d
have
1 = ( ab)2 = abab =⇒ ab = b−1 a−1 = ba =⇒ G abelian.

84 II. GROUPS
So there exists an element a ∈ G of order 4.

Since subgroups of index 2 are normal (cf. II.I.8), h ai E G. Pick
b ∈ G \h ai; then G = h a, bi, with b2 ∈ h ai. So b2 = aµ , where µ 6= 1
or 3 (otherwise |hbi| = 8); that is,
(II.M.3) b2 = a2 or 1.
By the normality, bab−1 ∈ h ai. Since bab−1 has the same order as
a, bab−1 = a±1 . But if bab−1 = a, then b and a commute and G is
abelian. So
(II.M.4) bab−1 = a−1 ,
i.e. ba = a−1 b.
Now (II.M.3) and (II.M.4) completely describe the multiplication
in a group of order 8 with elements 1, a, a2 , a3 , b, ba, ba2 , ba3 :
b2 = a2 or 1, b−1 = a2 b or b, ab = ba3 , a2 b = ba2 , etc.
There are two cases: first, if b2 = 1, then we clearly get an isomor-

∼
=
phism from D4 → G by sending r 7→ a and h 7→ b. Second, if b2 = a2 ,
∼
=
there is an isomorphism from the quaternions Q → G sending i 7→ a
and j 7→ b (and −1 7→ a2 = b2 ); the reader should check the remain-
ing details.
Next up would be 12, but this is harder — we’ll just list those:
D6 , A4 , Z2 × Z6 , Z12 , and the “third29 dicyclic group”
T := h a, b | a6 = 1, b2 = a3 = ( ab)2 i.
There is only one group of order 15, namely Z15 ; this will follow
from results below on groups of order pq. But there are 14 non-
isomorphic groups of order 16, so that’s a good place to stop this initial
mini-foray into group classification.
29This is a series of groups of order 4n: for n = 1, Z × Z ; for n = 2, Q; for n = 3,

2 2
T; etc.
High(er) order. Here we can really get going with the Sylow the-
orems, but first we require a few “counting” results. The first is im-
mediate from HW 4 #2:
| H ||K |
II.M.5. L EMMA . H, K ≤ G =⇒ | H ∩K |
≤ | G |.
e
For n = ∏it=1 pi i , recall that
Aut(Zn ) ∼
= Z∗n ∼
= Z∗pe1 × · · · × Z∗pet
1 t
has order φ(n) := {k ∈ Z>0 | k < n, (k, n) = 1}; e.g., φ( p) = p − 1,

φ( p2 ) = p( p − 1), φ( p3 ) = p2 ( p − 1) etc. for p prime. (One can also
prove that Z∗p ∼= Z p−1 — i.e. it is actually cyclic — which we’ll do
later on but won’t need here.) I will also use without proof
(II.M.6) Aut(Z p × · · · × Z p ) ∼
= GL(k, Z p ),
| {z }
k times
where the RHS means k × k invertible matrices with entries30 mod p.

For k = 1, this just says Aut(Z p ) ∼
= Z∗p (with order p − 1). For k = 2
it reads
Aut(Z p × Z p ) ∼
= GL(2, Z p )
with order ( p − 1)2 p( p + 1); you will prove this case in the next HW.
We don’t actually need much of this for our first result, on groups
of order pq:
II.M.7. T HEOREM . Let p and q be distinct primes, with p > q and

q - p − 1. Then the only group of order pq is Z pq .
P ROOF. Let | G | = pq and H, K be the subgroups of order p resp.

q guaranteed by Cauchy. By Lagrange’s theorem, (i) H ∩ K = {1};
and clearly (ii) | H ||K | = | G |.
30For this to make sense, you need to multiply and add in Z , which means to
p
consider it as a ring. Since we’re about to start that part of the course, it seems
fair to mention it! Of course, the automorphisms on the left of (II.M.6) are as an
abelian group, and GL(k, Z p ) itself is a (nonabelian) mutliplicative group. This
generalizes our earlier example Aut(Z2 × Z2 ) ∼ = S3 from II.J.4(i) of the group of
automorphisms of an abelian group being nonabelian, since S3 ∼ = GL(2, Z2 ).
86 II. GROUPS
Moreover, we claim that H E G. Otherwise, a distinct conjugate

H0 would be of the same order p, and (again by Lagrange) H ∩ H 0 =
{1}. Lemma II.M.5 then gives | G | ≥ | H || H 0 | =⇒ pq ≥ p2 =⇒
q ≥ p, a contradiction.
Now, consider the composition
∼
= ı(·) ∼
= / ∼
=
(II.M.8) Zq / K / Aut( H ) Aut(Z p ) 5/ Z∗p
induced by conjugating elements of H by elements of K (since H E

G). By the Fundamental Theorem,
|im(µ)||Z∗p |, |Zq |.

If q - p − 1, this is impossible unless |im(µ)| = 1. Hence, the map

(II.M.8) is trivial, and ık = id H (∀k ∈ K); i.e. khk−1 = h (∀h ∈ H, k ∈
K) or (iii) kh = hk (∀h, k).
The three hypothesis (i),(ii),(iii) imply (by the Direct Product The-
orem II.E.11(iv)) that G ∼ = H×K ∼ = Z p × Zq , which is ∼= Z pq since
( p, q) = 1.
Here is another instance of this type of argument:
II.M.9. E XAMPLE . We classify the groups G of order 52 · 372 . First,

there exist
• a Sylow 37-subgroup H (of order 372 ), and
• a Sylow 5-subgroup K (of order 52 ).
Clearly | H ||K | = | G |, and H ∩ K = {1}.
Moreover, H E G: if it had a distinct conjugate H 0 , then | H ∩ H 0 |
is either 37 or 1, so that
| H || H 0 | 372 · 372
52 · 372 = | G | ≥ = = 373 or 374
|H ∩ H0 | 37 or 1
yields a contradiction.
Thus K acts on H by conjugation, yielding a homomorphism
ϕ : K → Aut( H ).
Being a group of order p2 , p = 37 prime, H is one of the following

(cf. II.H.9):
• H∼ = Z p2 =⇒ Aut( H ) ∼
= Z∗p2 has order p( p − 1) = 37 · 36; or
• H∼ = Z p × Z p =⇒ Aut( H ) ∼
= GL(2, Z p ) has order p( p + 1)( p −
2 2
1) = 37 · 38(36) .
But for ϕ to be compatible with the Fundamental Theorem, we must

have |im( ϕ)||K |, |Aut( H )|; and 25 is relatively prime to 37, 36, and
38. We conclude that im( ϕ) = {1Aut( H ) }, so that kh = hk (∀k, h)
as in the last proof. Once again, G ∼ = H × K which yields the four
possibilities:
• G∼ = Z372 × Z52 ∼ = Z52 372 ;
• G∼ = (Z37 × Z37 ) × Z52 ∼ = Z37 × Z37·52 ;
∼ ∼
• G = Z372 × (Z5 × Z5 ) = Z372 ·5 × Z5 ; and
• G∼ = (Z37 × Z37 ) × (Z5 × Z5 ) ∼ = Z5·37 × Z5·37 .
In particular, G is abelian!
Recall that A5 was a simple group, i.e. had no normal subgroups

apart from itself and {1}. Also, |A5 | = 5!2 = 60. Here is a beautiful
application of Sylow theory which attests to the “specialness” of A5 .
II.M.10. T HEOREM . There is no nonabelian simple group of order less

than 60.
P ROOF. We know that

• groups of prime order are simple but abelian, and
• groups of prime power order are not simple (cf. II.L.8).
Suppose that there exists a nonabelian group G, simple of order
(II.M.11) | G | = pe m, m > 1, p prime, p - m, pe - (m − 1)!.

By Sylow I, G has a subgroup H of order pe and index m. Letting G
act on left cosets G/H by left-multiplication gives a homomorphism
ϕ : G → Sm which must be injective or trivial (so that ker( ϕ) doesn’t
furnish a normal subgroup other than {1} or G). In fact, it can’t be
trivial, as G acts transitively on H’s cosets. So ϕ is injective, and
88 II. GROUPS
ϕ( G ) ≤ Sm has order pe m. By Lagrange, pe m|m! =⇒ pe | (m − 1)!

in contradiction to our assumption.
Now the only positive integers less than 60 and not of the form
(II.M.11), and not a prime or prime power, are 30, 40, and 56.
We shall systematically rule these out as possible orders of simple
groups.
Suppose | G | = 30 = 2 · 3 · 5, and take P to be a Sylow 5-subgroup.
By Sylow II, the number of conjugates of P satisfies NP ≡ 1; while
(5)
by (II.L.10) NP | 30. Moreover, if G is to be simple, then P can’t be
normal, so NP > 1; and we deduce that NP = 6. Next, a 3-Sylow
subgroup Q has NQ ≡ 1, NQ |30, and NQ 6= 1, hence NQ = 10. By
(3)
Lagrange, none of the six conjugates of P and 10 conjugates of Q can
intersect outside of {1}. This requires G to have at least
(5 − 1) · 6 + (3 − 1) · 10 + 1 = 45 elements,
which, well, it doesn’t.
How about | G | = 40 = 23 · 5? Let P be a Sylow 5-subgroup. If N
is its number of conjugates, then N ≡ 1, N 6= 1, and N | 40. There’s
(5)
no such number N.
Finally, there’s | G | = 56 = 23 · 7 to take out. Write P for a Sylow 7-
subgroup, with N7 conjugates. We have N7 ≡ 1, N7 6= 1, and N7 | 56
(7)
=⇒ N7 = 8. Playing the same game with a Sylow 2-subgroup Q
gives N2 ≥ 3 conjugates. Now the conjugates of P can’t intersect
each other or those of Q outside of the identity element; not counting
the identity, this furnishes N7 · (7 − 1) = 8 · 6 = 48 distinct elements
of G. On the other hand, the conjugates of Q (which, remember, have
order 8) can intersect in order-4 subgroups. Without thinking too
hard, we at least get (counting the identity) 12 additional elements
of G by considering just Q and one conjugate. This again produces a
contradiction since 12 + 48 = 60 > 56, concluding the proof.
Miscellany. Before we leave the realm of finite group classifica-

tion, I would be remiss not to mention the famous classification of all
finite simple groups (completed in 2004) into:
• cyclic groups of prime order Z p
• alternating groups An , n ≥ 5 (Galois, 1832)
• simple groups of Lie type:
– This starts from É. Cartan’s classification (1894) of complex
simple Lie algebras into the Cartan types An (sln+1 ), Bn (so2n+1 ),
Cn (sp2n ), Dn (so2n ), G2 , F4 , E6 , E7 , E8 .
– C. Chevalley (1955) constructed integral bases for these, al-
lowing him to define the corresponding simple Lie groups
as algebraic groups over the integers, hence also over finite
fields (e.g. Z p , by reducing modulo p).
– e.g. PSLm (Z p ) is obvious; not so with G2 (Z p ).
– Steinberg, Suzuki, Ree filled in gaps (e.g. unitary groups).
• 26 sporadic simple groups
– the “Monster” is the largest, of order ' 8 × 1053
– Mathieu groups are the most approachable, as automorphism
groups of “Steiner systems”: e.g., S(4, 5, 11) denotes a set P
of 11 points, together with a set B of “blocks” of 5 points
each, such that each 4-point subset of P belongs to a unique
block; and M11 ≤ SP ∼ = S11 is the subgroup preserving blocks.
Another two topics oddly missing from [Jacobson] are group exten-
sions and semidirect products. They deserve a brief mention now, as
interesting constructions of (non-simple) finite groups.
We start with group extensions (of a group H by a group K).
These are short-exact sequences of groups
α β
(II.M.12) ε : = {1 → K → G → H → 1}.
That is, α is an injective homomorphism, β is a surjective homomor-

phism, and ker( β) = im(α). Hence,31 H ∼= G/α(K ) and α(K ) E G.
31Note that since α is injective, α(K ) ∼ K. The “1” on each end of the sequence
=
is a formality, which can be read as saying the kernel of α is the image of “1” (i.e.
trivial), and the image of β is the kernel of H → 1 (i.e. all of H).
90 II. GROUPS
II.M.13. E XAMPLE . The nth dicyclic group
Dicn := h a, b | a2n = 1, b2 = an , b−1 ab = a−1 i
sits in a short-exact sequence
1 / Z2n / Dicn / Z2 / 1.
/
b 1̄
/ /
1̄ a 0̄
Referring to (II.M.12), one says that
ε splits ⇐⇒ ∃ homomorphism γ : H → G with βγ = id H .
This displays H ∼= γ( H ) as a (not necessarily normal) subgroup of

G, since such a homomorphism is necessarily injective (why?). One
easy example: G = H × K is a split extension of H by K. Here is
another:
II.M.14. E XAMPLE . Recall the presentation
Dn = h a, b | an = 1, b2 = 1, b−1 ab = a−1 i
of the nth dihedral group. We have

γ
bo 1̄
1 / Zn / Dn / Z2 / 1
b
β
/ 1̄
1̄ / a / 0̄
so in this case the extension is split: γ yields a homomorphism be-
cause 1Dn = γ(0̄) = γ(1̄ + 1̄) = γ(1̄)2 = y2 does indeed hold. (This
won’t work in example II.M.13 since there’s no element of order 2 in
β−1 (1̄) ⊂ Dicn .)
So split extensions are more general than direct products, though
they are nicer than general group extensions. Can we characterize
them in some useful way?
II.M.15. D EFINITION . Let θ : H → Aut(K ) be a homomorphism,
sending h 7→ θh . The semidirect product K oθ H is the group with
underlying set K × H and product
(k, h) · (k0 , h0 ) := (k · θh (k0 ), h · h0 ).
II.M.16. P ROPOSITION . The extension ε splits ⇐⇒ G is a semidirect

product of K and H.
P ROOF. ( =⇒ ): Write H = γ( H ) and K = α(K ) as subgroups

of G. Define θ : H → Aut(K ) by h 7→ ıh ; this works since K E G.
(That is, θh := ıh is conjugation by h.) As H ∼
= G/K, the map of sets
µ : K × H → G sending (k, h) 7→ kh is bijective. Now compute:
µ(k, h)µ(k0 , h0 ) = khk0 h0 = khk0 h−1 hh0

= µ(kθh (k0 ), hh0 ).
The reverse direction ( ⇐= ) is HW.
II.M.17. E XAMPLE . The mod 3 Heisenberg group consists of matri-

ces  
1 0 0
 a 1 0
 
c b 1
with entries mod 3 (i.e. in Z3 ). While non-abelian, it has the same
number of elements of each order as Z3 × Z3 × Z3 . There is a natural
way to write this as an extension of Z3 × Z3 by Z3 . Does it split?
(This is a useful question to ask when looking at HW 5 #1.)
92 II. GROUPS
II.N. “Not-Burnside’s” counting lemma
Indeed, Burnside himself (1897) attributed it to Frobenius (1887),

though it was known much earlier to Cauchy as well. And of course,
it is known to you by HW #3. We begin by reviewing the statement
and proof.
II.N.1. N OTATION . Throughout, G denotes a finite group acting
on a finite set X, with:
• G ( x ) ⊂ X the G-orbit of x ∈ X (= { g.x | g ∈ G });
• X/G := the set of G-orbits;
• Xg := the fixed-point set of g ∈ G (= { x ∈ X | g.x = x }); and
• Gx ≤ G the stabilizer of x ∈ X (= { g ∈ G | g.x = x }).
1
II.N.2. T HEOREM (Burnside’s Lemma). |X/G | = |G| ∑ g ∈ G |X g |.
P ROOF. Consider the subset of G × X
S := {( g, x ) | g.x = x } ∼
=
set
ä Xg ∼
=
set
ä Gx .
g∈ G x ∈X
By the orbit-stabilizer theorem, | Gx || G ( x )| = | G |; hence

1
(II.N.3) ∑ |Xg | = |S| = ∑ |Gx | = |G| ∑ | G ( x )|
.
g∈ G x ∈X x ∈X
Taking { xi }ri=1 to be a system of representatives of the orbits, so that

X = qri=1 G ( xi ), we get (II.N.3) =
r
| G ( xi )|
|G| ∑ = | G |r = | G ||X/G |.
i =1
| G ( xi )|
II.N.4. R EMARK . This can be extended to an infinite group G (act-
ing on a finite set) by noticing that the homomorphism ϕ : G → SX
factors through ϕ̄ : G → SX where G := G / ker( ϕ) is finite. One has
that |X/G| = |X/G |, while the RHS of Burnside is replaced by an
“integral” over G .
Though the applications we shall give are indeed about counting
things, there are numerous theoretical corollaries and extensions of
this result:
II.N. “NOT-BURNSIDE’S” COUNTING LEMMA 93
• In combinatorics, a refinement called Pólya’s enumeration theorem

breaks the single number |X/G | out by “weights” that one at-
tributes to elements of X. That is, there is a weight function w : X →
N (or even Nk ) constant on G-orbits and one wants to count orbits
by weight.
• In topology, if f : T → S is a finite (connected, nonisomorphic)
covering of a topological space S , then there is a continuous map
from a circle into S that does not lift to (i.e. factor through) T .
(Here G is the “fundamental group” π1 (S ) and X is f −1 (s) for
some s ∈ S .)
• In number theory, if a polynomial F with integer coefficients (de-
gree ≥ 2, irreducible over Q) has Np roots mod p, the density of
primes for which Np = 0 is at least n1 . (Here G is the “Galois
group” of the polynomial, and X is the set of roots of F in the al-
gebraic closure Q̄. We will meet these notions in Algebra II.)
Before proceeding to the examples, here is an immediate theoret-
ical consequence for group actions (which is in fact related to the last
two bullet-points):
II.N.5. C OROLLARY. Given a finite group G acting transitively on a

finite set X (with at least 2 elements), there exists an element g ∈ G which
|G|
acts without fixed points.32 In fact, there are at least |X| such elements.
P ROOF. We can consider the actions of G on X, and also on X × X

by g.( x, x 0 ) := ( g.x, g.x 0 ). Write χ( g) := |Xg |, so that also χ2 ( g) =
|(X × X) g |; and for any function f on G and subset S ⊂ G, write
1
S f : = | G | ∑ g∈S f ( g ). We want to show that the subset of fixed-
R
| G0 |
point-free elements G0 := { g ∈ G | χ( g) = 0} ⊂ G has C := |G|
=
1
R
G0 1 ≥ |X| .
Burnside plus transitivity tell us that
Z Z
(II.N.6) 1 = |X/G | = χ and 2 ≤ |(X × X)/G | = χ2 ,
G G
32Up to this point, this Corollary is a theorem of Jordan from 1872. The last sen-
tence is due to Cameron-Cohen (1992) and its proof to Serre (2003).
94 II. GROUPS
as the “diagonal” ∆X := {( x, x ) | x ∈ X} ⊂ X × X is an orbit. If g ∈

/
R
G0 , then 1 ≤ χ( g) ≤ |X|; and so G\G (χ( g) − 1)(χ( g) − |X|) ≤ 0,
0
which can be rewritten as
Z Z
(χ( g) − 1)(χ( g) − |X|) ≤ (χ( g) − 1)(χ( g) − |X|) = C |X|.
G G0
By (II.N.6), the LHS becomes

Z Z Z
2
χ − (1 + |X|) χ + |X| 1 ≥ 2 − (1 + |X|) + |X| = 1.
G G G
1
Conclude that C ≥ |X|
as desired.
Here is a straightforward application of Burnside to a counting
problem.
II.N.7. E XAMPLE . How many inequivalent bracelets with five beads

can you make with only black and white beads?
Here two bracelets are equivalent if they are the same after rotat-
ing and flipping them, i.e. if they belong to the same orbit under D5 .
So we take X to be the set of B/W colorings of the sides of a regular
5-gon, and G := D5 (acting in the obvious way). We arrive at the
table
type of # of # of fixed
element of D5 such points |Xg |
1 (identity) 1 25 ( = | X | )
23 (edges flipped into each other must be
hr k (flip) 5
same color in order to be fixed)
2 (all edges must be same color to yield
r k (rotation) 4
a “fixed point”)
The number of bracelets (i.e. orbits) is then simply
|X/G | = 1
| D5 | ∑ |X g | = 1
10 {1 · 2
5
+ 5 · 23 + 4 · 2} = 8.
g∈ D5
In the HW, you will determine (up to rotational symmetry) how

many different ways one can paint the edges of a tetrahedron red, green, or
blue.
Crystallographic groups. We now turn to a much more interest-

ing example. By a lattice in R3 , we mean a subgroup of (R3 , +, 0)
isomorphic to Z3 . Denote by O(3) the orthogonal group comprising
3 × 3 matrices A with real entries and t AA = I3 . The determinant
defines a surjective homomorphism det : O(3) {±1}, with kernel
the special orthogonal group SO(3). Heuristically, this is the group of
rotations about the origin in R3 , while O(3) also includes reflections
and −I3 .
II.N.8. D EFINITION . A crystallographic group is a (nontrivial) finite
subgroup G ≤ SO(3) or O(3) that preserves a lattice in R3 .
The mathematical determination (Hessel 1830) of the existence of
exactly 32 geometric crystal classes, essentially via such groups, pre-
dated the actual ability to look inside crystals with X-rays by some-
thing like 80 years. We shall restrict ourselves here to the rotational
case, so that we can obtain a complete classification using Burnside.
So let G ≤ SO(3) be a finite subgroup. Each g ∈ G \{1} is a
rotation about some axis ` g , and we set { p g , p0g } := ` g ∩ S2 , where S2
is the unit 2-sphere centered at the origin. Let
{ p g , p0g }.
[
X :=
g∈ G \{1}
II.N.9. C LAIM . G acts on X.

P ROOF. We first clarify what we mean by this. Since SO(3) acts
(by matrix multiplication) on R3 , and in fact on S2 ,33 we need only
check that X is closed under the action of G. Given x ∈ X, x = p g0 (or
p0g0 ) for some g0 ∈ G. That is, g0 x = x.
Consider g.x = gx (where the RHS means the matrix g ∈ G times
the vector x). This is in X ⇐⇒ it is p g1 (or p0g1 ) for some g1 ∈ G. But
gg0 g−1 .gx = gg0 .x = g.g0 x = g.x = gx,
and so g1 := gg0 g−1 works.

33Given a vector ~v ∈ R3 and matrix A ∈ SO(3), we have that A~v · A~v = t~v t AA~v =
t~ = ~v · ~v =⇒ A preserves the length of ~v. Thus S2 is closed under the action of
v~v
SO(3) on R3 .
96 II. GROUPS
Now let n = | G | and r = |X/G |. Choose representatives x1 , . . . , xr

in each orbit and put ni := | Gxi | (≤ n). By the orbit-stabilizer theo-
rem, the orbit sizes are | G ( xi )| = nn , so that
i
r
n
|X| = ∑ ni .
i =1
Burnside yields |X/G | = |G1 | ∑ g∈G |Xg | =⇒

1 0
r = n |X| + ∑ g∈G\{1} |{ p g , p g }|

= n1 ∑ri=1 nni + (n − 1)2
= ∑ri=1 n1i + 2 − n2 .
For each x ∈ X, | Gx | ≥ 2 by definition (there is some non-identity
element stabilizing it)
=⇒ each ni ≥ 2
r 2
=⇒ r ≤ 2 +2− n
r 2
=⇒ 2 ≤ 2− n < 2,
2 1 1 1
so r < 4. If r = 1 then n = n1 +1 ≥ n + 1 =⇒ n ≥ 1, which is
absurd.
=⇒ r = 2 or 3.
1 1 2
Case r = 2: 2= n1 + n2 +2− n
2 1 1 2
=⇒ n = n1 + n2 ≥ n
2 1 1
=⇒ n = n1 + n2 (and n1 , n2 ≤ n)
=⇒ n1 = n = n2
=⇒ every g ∈ G stabilizes every x ∈ X.
The only way that even one non-identity element of G stabilizes ev-
ery x ∈ X, is if there are only 2 points p, p0 ∈ X; and then the elements
of G are rotations about the axis they span (by multiples of 2π n )
=⇒ G ∼
= Zn .
(∗) 1
Case r = 3: (1 <) n2 + 1 = n1 + 1
n2 + 1
n3 . We may assume 2 ≤
n1 ≤ n2 ≤ n3 ≤ n. Now
• if n1 > 2 then RHS(∗) ≤ 1, which is impossible; so n1 = 2.
• if n2 ≥ 4 then RHS(∗) ≤ 1, again impossible; so n2 = 2 or 3.
• if n2 = 3 then for the same reason, n3 < 6.
Thus our options are confined to the left-hand column of the follow-
ing table:
geometric realization:
(n1 , n2 , n3 ) n (= | G |) G
rotational symmetries of . . .
prism on
(2, 2, k)k≥2 2k Dk ···
regular k-gon
(2, 3, 3) 12 A4 tetrahedron
(2, 3, 4) 24 S4 cube
(2, 3, 5) 60 A5 icosahedron
Fix an orbit G ( xi ) ⊂ X (i = 1, 2, or 3). The stabilizer Gxi of the

representative xi comprises all the elements of G which are rotations
about xi (viewed as a vector in R3 ). So ni = | Gxi | is the order of
this axis of rotation, and also of the other axes of rotation in G ( xi ), of
which there are 12 | G ( xi )| = 2nn . Two examples:34
i
• (2, 2, 6) corresponds to the prism on the regular hexagon shown.
The orbits have sizes 6, 6, 2, corresponding to 3 axes (with 180◦
rotation, through the vertical edges), 3 axes (through the vertical
faces, again with 180◦ rotation), and 1 vertical axis (with 60◦ rota-
tion).
• (2, 3, 3) corresponds to the tetrahedron. The orbits have sizes 6, 4, 4,
so there are 3 axes (through pairs of nonintersecting edges) of
180◦ -rotation, and 2 + 2 = 4 axes (through each vertex and the
opposite face) of 120◦ -rotation.
34For the cube, see II.F.4(v).
98 II. GROUPS
This gives a flavor of how one derives the table.

We have at this point finished the classification of finite rotational
symmetry groups in space.
Next, impose the condition that we are not just stabilizing a sin-
gle “crystal unit” but a pattern which may be continued infinitely to
fill up R3 . So we ask for G to act on a lattice Λ ⊂ R3 , comprising all
Z-linear combinations of three linearly independent vectors ~u, ~v, w ~.
Now:
• If we think of g ∈ G as a 3 × 3 matrix written with respect to the
~ , then the trace tr( g) ∈ Z since g will have integer
basis ~u, ~v, w
entries.
• If we write g instead with respect to an orthonormal basis includ-
ing p g (a vector along the axis of rotation), it takes the form
 
1 0 0
0 cos(θ ) sin(θ ) 
 
0 − sin(θ ) cos(θ )
and so tr( g) = 2 cos(θ ) + 1.
• From matrix algebra, you know that the trace of g is independent
of the choice of basis. Hence, we must have 2 cos(θ ) ∈ Z. Since g
k , for some k ∈ Z. But 2 cos( k ) ∈
is a finite-order rotation, θ = 2π 2π
Z only for k = 1, 2, 3, 4, 6.
Therefore, for a crystallographic group G ≤ SO(3), we must have that
all of the ni belong to {1, 2, 3, 4, 6}. Throwing out all other groups in
our list, we are left with 10 nontrivial rotational crystallographic groups:
Z2 , Z3 , Z4 , Z6 , D2 (∼
= V4 ) , D3 (∼
= S3 ) , D4 , D6 , A4 , and S4 .
III. Rings
III.A. Examples of rings
The theory of rings and ideals grew out of several 19th and early
20th Century sources:
• polynomials (Gauss, Eisenstein, Hilbert, etc.);
• number rings (Dirichlet, Kummer [“ideal numbers”], Kronecker,
Dedekind [“ideals in number rings”], Hilbert, etc.); and
• matrix rings and hypercomplex numbers (Hamilton [quaternions],
Cayley [octonions], etc.).
Specifically, the term Zahlring showed up in the study of what we
would now call rings of integers in algebraic number fields; e.g. cy-
clotomic rings such as Z[ζ 5 ] (ζ 5 = a 5th root of 1) arose in the context
od attempts to prove Fermat’s last theorem, and ζ 5 “cycles back to
itself” (suggesting a ring) upon repeatedly taking powers. Here is
the modern definition, due to E. Noether (∼1920):
III.A.1. D EFINITION . A ring ( R, +, •, 0, 1) comprises a set R to-

gether with 2 binary operations and distinguished elements, satisfy-
ing:
(i) ( R, +, 0) is an abelian group;
(ii) ( R, •, 1) is a monoid; and
(iii) distributive laws:
r (s1 + s2 ) = rs1 + rs2 and (r1 + r2 )s = r1 s + r2 s.
Note that we do not assume the existence of multiplicative inverses.
III.A.2. R EMARK . (i) If we didn’t assume that “+” was commu-

tative, this would be forced upon us by the distributive laws as fol-
lows:
99
100 III. RINGS
• −( a + b) = (−b) + (− a) (not assuming ( R, +, 0) abelian)

• ∃ “additive” inverse −1 of 1 (since ( R, +, 0) is a group)
• adding −(0r ) on the left to 0r = (0 + 0)r = 0r + 0r gives 0 = 0r
• adding (−r ) on the right to (−r ) + r = 0 = 0r = (−1 + 1)r =
(−1)r + 1r = (−1)r + r gives −r = (−1)r
• −( a + b) = (−1)( a + b) = (−1) a + (−1)b = (− a) + (−b).
(ii) There is also the notion of a “rng” ( R, +, •, 0) where ( R, •) is
taken to be a “semigroup”, meaning that one doesn’t assume the
existence of a multiplicative “i”dentity (or inverses). However, we
can construct a ring containing R with underlying set S = Z × R,
operations
(
( n1 , r1 ) + ( n2 , r2 ) : = ( n1 + n2 , r1 + r2 ) and
( n1 , r1 ) · ( n2 , r2 ) : = ( n1 n2 , n1 r2 + n2 r1 + r1 r2 ) ,
and distinguished elements 1 := (1, 0) and 0 := (0, 0), by checking
that the associative and distributive laws hold. (R consists of the
elements (0, r ).)
(iii) A subring of R is a subset closed under +, −, and •. Hence the
intersection of subrings is a subring, and it makes sense to speak of
the subring generated by a subset S (= intersection of all subrings
containing S ).
(iv) A ring is called commutative if the multiplication “•” is. (We
don’t use the term “abelian” for rings.)
III.A.3. E XAMPLES . (i) (A, +, •, 0, 1), with A = Z, Q, R, C, or Zm .
( (
products ∏i ∈ I Ri
(ii) Direct of rings1 . If | I | < ∞ then these are
sums ⊕i ∈ I Ri
(
∏
the same. Otherwise, the consists of ∞-tuples
⊕
(
with no constraints
with all but finitely many entries zero.
1The products are also written ×
i ∈ I Ri , more typically when there are finitely many,
viz. R1 × · · · × Rk . We won’t use “⊕” for finite sums/products of rings.
III.A. EXAMPLES OF RINGS 101
(iii) Number rings. Let D be a squarefree integer, i.e. ± p1 · · · pd

where p1 , . . . , pd are distinct primes. Inside C (or R, if D > 0), it
is easy to see the closure properties for the (quadratic) number field
√ √
Q[ D ] := { a + b D | a, b ∈ Q}
and the (quadratic) number ring

√ √
Z[ D ] := { a + b D | a, b ∈ Z}.
What about
√ √
Z[ 1+2 D
] : = {m + n( 1+2 D
) | m, n ∈ Z}
√
= { a+b2 D
| a, b ∈ Z, a ≡ b} ?
(2)
a−b
(For the last equality, take m = 2 and n = b.) Of course, the issue
is multiplicative closure:
√ √
(m + n( 1+2 D
))(m0 + n0 ( 1+2 D
)) =
√ √
mm0 + (mn0 + nm0 )( 1+2 D
) + nn0 ( (1+ D)+
4
2 D
).
√
| {z }
nn0 ( D −1) 1+ D
4 +nn0 ( 2 )
Clearly closure holds ⇐⇒ 4 | D − 1 ⇐⇒ D ≡ 1. As we shall see,

(4)
√
the “ring of integers” in Q[ D ] is
 √
 Z[ 1+ D ] if D ≡ 1
2 (4)
√
Z[ D ] otherwise.
√
Two special cases of interest are Z[ 1+2 5 ] and Z[i].
(iv) Polynomial rings. Let R be a commutative ring. Set
R[ x ] := {sequences (r0 , r1 , . . . , rn , 0, 0, . . . ) | ri ∈ R}
| {z }
zero from
some point on
102 III. RINGS
and define, given a = ( ak )k≥0 and b = (bk )k≥0 ,
a + b := ( ak + bk )k≥0 and a · b := (∑kj=0 a j bk− j )k≥0 .
Also put 0 := (0, 0, 0, . . .) and 1 := (1, 0, 0, . . .). Then we have
( a + b) · c = (∑kj=0 ( a j + b j )ck− j )
= (∑kj=0 a j ck− j ) + (∑kj=0 b j ck− j ) = a · c + b · c
and
( a · b) · c = (∑ik=0 ai bk−i ) · c = (∑k`=0 (∑i`=0 ai b`−i )ck−` )

= (∑ik=0 ai ∑kj=−0i b j c(k−i)− j ) = a · (∑kj=0 b j ck− j )
`=i + j
= a · ( b · c ),
so that II.A.1(iii) is satisfied.
Now identify R with the subring {(r, 0, 0, . . .)} ⊂ R[ x ]. Taking
x := (0, 1, 0, 0, . . .), we have x n = (0, . . . , 0, 1, 0, 0, . . .) so that
| {z }
n
(r0 , r1 , r2 , . . . , rn , 0, 0, . . .) = rn x n + · · · + r1 x + r0 ,
which is obviously a much more appealing (and standard) notation.
We can also (inductively) define polynomial rings in several vari-
ables by
R[ x1 , . . . , xn ] := ( R[ x1 , . . . , xn−1 ]) [ xn ].
For any r ∈ R, we can consider the evaluation map
evr : R[ x ] −→ R
sending rn x n + · · · + r1 x + r0 7−→ rn r n + · · · + r1 r + r0 .
More generally, we can take the product
∏ evr : R[x] → ∏ R (= “RR ”)

r∈R R
of all such maps, sending a polynomial to (essentially) its “graph”.

This is not always surjective (e.g. if R = R) or injective (e.g. if R =
Z3 ).
(v) Quaternions. The ring version is built out of the group one: put
H := { a + i + cj + dk | a, b, c, d ∈ R} ,
where i, j, k have the same multiplicative properties as in the 8-element

group Q. Clearly this is noncommutative. The “H”, of course, is for
Hamilton.
(vi) Matrix rings. Let R be an arbitrary ring, n ∈ N. We define a ring
wth underlying set
n
Mn ( R) := {∑i,j =1 rij eij | rij ∈ R },
where the eij are formal symbols. Taking A = ∑i,j aij eij , B = ∑i,j bij eij ,
we set2 0 := ∑i,j
n n n
=1 0eij , 1 : = ∑i,j=1 δij eij = ∑i =1 eii , and
n n
A + B := ∑ (aij + bij )eij and AB := ∑ (∑nk=1 aik bkj )eij .
i,j=1 i,j=1
Associativity follows from

n
( AB)C = ∑ (∑nk,`=1 aik bk` c` j )eij = A( BC)
i,j=1
and the associativity of R; the rest is left to you.3 Of course, these can
be represented in the standard way as matrices
 
a11 · · · a1n
A =  ... ..
.
.. 

. 
an1 · · · ann
and you may think of eij as the matrix with a 1 at the (i, j)th place
and zeroes elsewhere. We have
(
0, j 6= k
eij ek` =
ei` , j = k.
The noncommutativity is highly visible this way.
2Here δ (= 1 if i = j, and 0 otherwise) is the Kronecker delta.

ij
3It is important to realize here that the order matters, not just of AB vs. BA, but of
aik bkj vs. bkj aik , because R may not be commutative.
104 III. RINGS
Here are some definitions which were clearly not possible (or not
interesting) for groups.
III.A.4. D EFINITION . Let R be a ring, r ∈ R an element.

(i) r is a left [resp. right] zero-divisor ⇐⇒ ∃ r 0 ∈ R\{0} such that
rr 0 = 0 [resp. r 0 r = 0].
(ii) r is nilpotent ⇐⇒ ∃ n ∈ N such that r n = 0.
(iii) r is idempotent ⇐⇒ r2 = r.
These are easily illustrated in M2 (R):

0 0
III.A.5. E XAMPLE . (i) In 11 00 00

01 = 00 = 0, the boxed
element is a left zero-divisor.
(ii) In 01 00
0 0
1 0 = 0, the boxed element is nilpotent.
(iii) In 10 00 1 0 = 1 0 , the boxed element is idempotent. (Think

00 00
projection.)
III.A.6. D EFINITION . The characteristic of a ring R is the (small-

est) number of times one has to add 1 (the multiplicative identity
element of R) to itself to obtain 0, unless this is not possible. In the
latter case, the characteristic is zero.
III.A.7. E XAMPLES . (i) R = Z, Q, R, C, H, M2 (R), Q[ x ] all have

char( R) = 0.
(ii) R = Zm , Mn (Zm ), Zm [ x ] have char( R) = m.
(iii) In a general commutative ring, we have
n
n k n−k
(III.A.8) ( x + y) = ∑
n
x y .
k =0
k
If char( R) = p, then p( kp) for 0 < k < p =⇒

(III.A.9) ( x + y) p = x p + y p ,
the so-called “Freshman’s dream”.
Next are some definitions analogous to those in groups or monoids:
III.A.10. D EFINITION . The center of R is
C ( R) := {r ∈ R | rs = sr ∀s ∈ R}.
III.A.11. E XAMPLES . (i) C (H) = R.

(ii) If R is commutative, C ( Mn ( R
)) = R,where R is identified with
r 0
the subring of diagonal matrices .. = r1 = “r”. More gener-
.
0 r
ally, C ( Mn ( R)) = C ( R).
P ROOF. Given A ∈ C ( Mn (R)),
n
0 = Aek` − ek` A = ∑ aij (eij ek` − ek` eij )
i,j=1
n n
= ∑ aik ei` − ∑ a` j ekj .
i =1 j =1
In particular, the (k, `)th entry of the last line is akk − a`` and the
(i, `)th entry (for i 6= k) is aik . So off-diagonal entries of A are 0 and
the diagonal ones are all equal. Finally, consider Ar − rA.
III.A.12. D EFINITION . r ∈ R is a unit (or invertible) ⇐⇒ ∃ r 0 ∈ R

such that rr 0 = 1 = r 0 r. (It is not enough in a general noncommutative
ring to have rr 0 = 1 or r 0 r = 1 for invertibility.) The units in R form
a group R∗ under multiplication.4
To begin with a few easy examples: for R = Q, R, C, H, and more

generally for division rings (see the next section), the units R∗ are all
nonzero elements. But that is not its general meaning. For instance,
we have Z∗ = {±1} and Z8∗ = {1̄, 3̄, 5̄, 7̄} ∼= Z2 × Z2 . Another ex-
ample is Mn (R) = GLn (R), which everyone knows is the matrices
∗
with determinant in R∗ = R\{0}. But for matrices over a more gen-

eral ring R? You’d think determinants might help, but not if R is
noncommutative:

III.A.13. E XAMPLE . Consider kj 1i ∈ M2 (H). The “determi-
nant” ki − 1j = j − j = 0, but
k j
k1 −2 −2
= 10 01 = 1.

j i 1 i
−2
2
4In Jacobson, R∗ means R\{0}, and U ( R) is the group of units. We will not use
this notation; the notation given above is more standard.
106 III. RINGS
So we can only hope for invertibility of matrices to be easily de-

tected via determinants when the entries are in a commutative ring.
Another key example of units in a commutative ring is problem
#7 from HW 1. Recall that this produced a group structure (∼ = Z×
Z2 ) on integer solutions to x − 5y = ±4. I claim that this can be
2 2
interpreted as an isomorphism
h √ i∗
Z × Z2 ∼ = Z 1+2 5
(III.A.14) √ a
( a, ±1) 7→ ± 1+2 5 .
√ √ √
x +y 5 1+ 5 x −y 5
Given α = 2 ∈ R := Z[ 2 ],
write α̃ := ∈ R. The com-
2
position law that led to the group structure on LHS(III.A.14) was ex-
actly multiplication in R. Moreover, ( x, y) solves the above equation
⇐⇒ α · (±α̃) =√ 1 =⇒ α ∈ R∗ . Conversely, if α ∈ R∗ , then there
x 0 +y0 5
exists α0 = 2 ∈ R with αα0 = 1, and then (αα̃)(α0 α̃0 ) = αα0 αα
f0 =
x2 −5y2
11̃ = 1. Since x ≡ y, we have that x2 ≡ 5y2 =⇒ αα̃ = 4 ∈ Z,
(2) (4)
and similarly for α0 α̃0 . So the only way the product of αα̃ and αα
f0 is
1, is if they are both ±1, and then α ∈ R∗ .
So far we have discussed only quadratic number fields and num-
ber rings. To give a brief glimpse ahead, a general result of Dirichlet
says that for a number field K with r1 distinct real embeddings and
r2 pairs of conjugate complex embeddings,5
(III.A.15) OK∗ ∼
= Zr1 +r2 −1 × {torsion group},
where OK ⊂ K is the ring of integers of K. The main point is that
(III.A.14) is a special case (with r1 = 2 and r2 = 0) of a much more
general result.
5All number fields can be viewed as vector spaces over Q of some finite dimension,
called the degree [K:Q]. In this case, that degree is r1 + 2r2 . (An embedding of fields
means an injective homomorphism, √ in this case into R or C. These notions will be
discussed later.) The case K = Q[ D ] has r1 = 0 and r2 = 1 if D < 0, or r1 = 2
and r2 = 1 if D > 0.
III.B. RING ZOOLOGY 107
III.B. Ring zoology
(III.B.1)
?_
COMMUTATIVE
RINGS
RINGS
?
Definition: ? COMMUTATIVE
No 0-divisors / DOMAINS ?_ (or “INTEGRAL”)
(left or right) DOMAINS
?
INTEGRALLY
(HW) CLOSED
DOMAINS
?
?
UNIQUE
Definition: /
DIVISION
FACTORIZATION
R∗ = R\{0} RINGS
DOMAINS(UFDs)
?
PRINCIPAL
IDEAL
DOMAINS(PIDs)
?
Definition: ?
EUCLIDEAN
commutative / FIELDS
DOMAINS
division ring
108 III. RINGS
We will define integrally closed domains, UFDs (also called “Fac-

torial” domains) and PIDs later.
III.B.2. D EFINITION . A Euclidean domain is a commutative do-

main which has a function
δ : R\{0} → N
with the following property: for all a ∈ R and b ∈ R\{0}, there exist
q, r ∈ R satisfying a = bq + r and either δ(r ) < δ(b) or r = 0. (This δ
is called a Euclidean function and is not unique.)
Clearly, these are just the domains to which we can generalize the
(Euclidean) division algorithm I.B.3.
III.B.3. R EMARK . The best choice for δ, when possible, is to have

δ(1) = 1 and δ−1 (1) = R∗ . This will be the case in all examples
below.
In the remainder of the section I simply comment on some of the

inclusions in (III.B.1) and give a few examples.
III.B.4. E XAMPLE . Given α = a + bi + cj + dk ∈ H\{0}, set ᾱ :=

a − bi − cj − dk. We have
αᾱ = a2 + b2 + c2 + d2 + ab(−i + i) + ac(−j + j) + ad(−k + k)

+ bc(−ij − ji) + bd(−ik − ki) + cd(−jk − kj)
=⇒ ᾱ
a2 + b2 + c2 + d2
= α−1 . This proves that
noncommutative division rings exist.
III.B.5. E XAMPLE . Z6 firnishes an example of a commutative ring

which is not a domain, due to the (obviously non-invertible) zero-
divisors 2̄, 3̄, and 4̄.
III.B.6. P ROPOSITION . Given a field F, (a) F and (b) F[ x ] are Eu-

clidean domains.6
6As will be seen very easily later, F[ x, y] is non-Euclidean.
III.B. RING ZOOLOGY 109
P ROOF. (a) Put δ(r ) := 1 ∀r ∈ F\{0}. Set q = b−1 a, r = 0.

(b) Put δ( P( x )) := 2deg( P( x)) . Use polynomial long division to con-
struct q, r.
√
III.B.7. E XAMPLE . Q[i] is a field. (Here, and elsewhere, i := −1.)
To see this, simply write
1 a − bi a b
= = 2 2
− 2 i.
a + bi ( a + bi)( a − bi) a +b a + b2
√
Similarly, we can show Q[ d] is a field for any d ∈ Z.
III.B.8. P ROPOSITION . (a) Z and (b) Z[i] are Euclidean domains.
P ROOF. (a) Put δ(m) := |m| and use the division algorithm.
(b) Writing α = a + bi, put δ(α) := αᾱ = |α|2 = a2 + b2 . Let α ∈ Z[i]
and β ∈ Z[i]\{0}. We will find µ and ρ in Z[i] such that α = βµ + ρ
and δ( β) > δ(ρ).
Working in Q[i], we have αβ−1 = x + yi; pick m, n ∈ Z such that
e := x − m and η := y − n have |e|, |η | ≤ 12 . Then
α = β{(m + e) + (n + η )i} = β{m + ni} + β{e + ηi}.

| {z } | {z }
=:µ =:ρ
Clearly µ ∈ Z[i], and so ρ = α − βµ ∈ Z[i] also. Now
δ(ρ) = |ρ|2 = | β|2 |e + ηi|2 = δ( β){e2 + η 2 }

≤ δ( β) · { 41 + 41 } < δ( β),
and we are done.
III.B.9. R EMARK . The δ in the proof of (b) is an example of a

Galois norm. This is easy to generalize to quadratic number rings
√ √
1+ D
√
R = Z[ D ] and (if D ≡ 1) Z[ 2 ]. Given α = a + b D, write
(4)
√
α̃ := a − b D (which is the complex conjugate ᾱ if D < 0); then
we define the norm by N (α) := αα̃. When this gives a Euclidean
function, a number ring is called norm-Euclidean. For imaginary qua-
dratic (D < 0), in which case Euclidean and norm-Euclidean are
110 III. RINGS
equivalent, the complete list is

√ √ √ √ √
Z[ −1], Z[ −2], Z[ 1+ 2 −3 ], Z[ 1+ 2 −7 ], and Z[ 1+ 2−11 ].
| {z }
(HW)
In the real quadratic case, the list of norm-Euclidean cases is much

longer (but finite) and strictly smaller than the list of Euclidean cases
(which is conjectured to be infinite).
We should also mention that for a ring R,

( ) !
ab = ac or ba = ca
(III.B.10) R is a domain ⇐⇒ AND =⇒ b = c .
a 6= 0
P ROOF. If R is a domain, suppose a(b − c) = 0 with a 6= 0; then

as there are no zero-divisors, b − c = 0.
Conversely, assume the condition on RHS(III.B.10), and suppose
ab = 0 with a 6= 0. Then ab = a0 =⇒ b = 0, and no left zero-
divisors exist. (Now reverse a and b.)
Finally, note that
(III.B.11) R is a domain =⇒ char( R) is prime or 0.

III.C. MATRIX RINGS 111
III.C. Matrix rings

n
Let R be a commutative ring; then a matrix A = ∑i,j =1 aij eij ∈
Mn ( R) has a determinant:
III.C.1. D EFINITION . det( A) := ∑σ∈Sn sgn(σ ) ∏in=1 ai,σ(i) ∈ R.
As immediate consequences of this definition,
det is alternating and multilinear
in the columns { a j } of A, viewed as elements of Rn :

(III.C.2)



 (i) det( a1 , . . . , rak + r 0 a0k , . . . , an ) = r det( a1 , . . . , ak , . . . , an )
+r 0 det( a1 , . . . , a0k , . . . , an )




 (ii) det( a1 , . . . , ak , . . . , a` , . . . , an ) = − det( a1 , . . . , a` , . . . , ak , . . . , an )

(iii) = 0 if ak = a` .
(Note that if 2|char( R), then (iii) does not follow from (ii), but it fol-
lows directly from III.C.1.) Moreover, since sgn(σ−1 ) = sgn(σ),
det(t A) = det( A) =⇒ (III.C.2) holds for rows.
This means that the elementary row operations (EROs)


(I)
 adding r times the jth row (r ∈ R) to the ith row
(III.C.3) (II) swapping ith and jth rows
(III) multiplying the ith row by r (r ∈ R∗ ),


which are invertible, have the following effects on det( A): none;
multiply by −1; multiply by r (respectively). EROs correspond to
multiplying A on the left by the elementary matrices



 elementary matrix det
1 + reij

(I) 1
(III.C.4)


 (II) 1 + eij + e ji − eii − e jj −1
(III) 1 + (r − 1)e

r.
ii
112 III. RINGS
When R is a field, such as R (as in the standard linear algebra

course), EROs can be used to put A into a unique reduced row ech-
elon form:




• each row has a “leading 1” as its first nonzero entry
th
• in row j, this occurs in the µ( j) entry, where µ is



(III.C.5) a strictly increasing Z>0 -valued function
• for each j, all entries in the µ( j)th column are zero,






 except for the leading 1.
(If µ( j) > n, the jth row is zero.) The resulting matrix
(III.C.6) rref( A) = EN · · · E1 A
| {z }
as in (III.C.4)
is either 1 or has last row 0. Moreover, by (III.C.3)-(III.C.4) it is clear

that

N
(III.C.7) det(rref( A)) = ∏i=1 det( Ei ) det( A).
Since the det( Ei ) 6= 0 and the Ei are invertible, this yields the
III.C.8. P ROPOSITION . When R is a field,
det( A) 6= 0 ⇐⇒ rref( A) = 1 ⇐⇒ A is invertible.
It also gives a way to compute the inverse: assuming rref( A) = 1,
(III.C.9) EN · · · E1 ( A | 1) = ( EN · · · E1 A | EN · · · E1 ) ,
| {z } | {z } | {z }
n×2n 1 A −1
i.e. computing rref( A | 1) gives (1 | A−1 ). Moreover, (III.C.9) shows

that any matrix B with nonzero determinant is a product of elemen-
tary matrices ∏iM=1 Ei , and so by (III.C.3)-(III.C.4) we get that first
M
det( B) = ∏i=1 det( Ei ) then
(III.C.10) det( BC ) = det( B) det(C ) ,
arguing by induction on M.
Turning to the next level of generality, suppose that R is a Eu-
clidean domain. To produce an analogue of the rref, find the nonzero
entry b in the first (nonzero) column with the lowest δ(b), then apply
a = bq + r to the other entries in that column and use a type (I) ERO
to kill the bq’s. Repeat this step on the column until all but one en-
try is zero; swap it to the top position. Restricting to the (n − 1) × n
submatrix below the first row, we repeat the algorithm to arrive at
 
0 · · · 0 r1 ∗ · · · ∗ s
 0 · · · · · · · · · · · · 0 r2
 





 ?
0
 

where by a type (I) ERO together with the Euclidean algorithm we

can assume δ(s) < δ(r2 ). Using type (I) and (II) EROs, we eventually
produce a matrix in Hermite normal form
(III.C.11)  
0 · · · 0 α1,µ(1) ∗ · · · ∗ α1,µ(2) ∗ · · · ∗ α1,µ(3)
0 · · · · · · · · · · · · · · · 0 α2,µ(2) ∗ · · · ∗ α2,µ(3)
 

Herm( A) = (αij ) = 
 etc. 
 0 · · · · · · · · · · · · · · · · · · · · · · · · · · · · 0 α3,µ(3)


..
.
where δ(αi,µ( j) ) < δ(α j,µ( j) ) for all i < j, and all entries below the
diagonal are 0. Clearly we still have (by (III.C.3)-(III.C.4)) that
(
Herm( A) = EN · · · E1 A
(III.C.12)
det(Herm( A)) = (∏i det( Ei )) det( A)
where ∏i det( Ei ) ∈ R∗ , and so
(III.C.13)
det( A) 6= 0 ⇐⇒ det(Herm( A)) 6= 0 ⇐⇒ µ( j) = j (∀ j).
| {z }
=∏in=1 αii
Assume the αii 6= 0 (∀i). We need the following
III.C.14. L EMMA . ∏in=1 αii ∈ R∗ ⇐⇒ αii ∈ R∗ (∀i ).
P ROOF. If n = 2, then this says that rs ∈ R∗ ⇐⇒ r, s ∈ R∗ .

Suppose rs = u ∈ R∗ ; then u−1 exists, and r (su−1 ) = 1 =⇒ r ∈ R∗ .
Now induce on n.
114 III. RINGS
Putting this all together,
det( A) ∈ R∗ ⇐⇒ det(Herm( A)) ∈ R∗

(III.C.12)
⇐⇒
(III.C.13)
∏ αii ∈ R∗ (III.C.14)
⇐⇒ αii ∈ R∗ (∀i ).
But if the αii ∈ R∗ , we may (if necessary) kill all the off-diagonal
entries (by type (I) EROs) and scale the diagonal ones to 1 (by type
(III) EROs). Since elementary matrices are invertible, we arrive at the
III.C.15. T HEOREM . For R Euclidean, the following are equivalent:

(i) Herm( A) has diagonal entries in R∗ ;
(ii) det( A) ∈ R∗ ; and
(iii) A is invertible.
In this case, (III.C.9) still gives a way to compute A−1 .
Now here’s a problem: we can’t prove det( AB) = det( A) det( B)

as we did above if neither A nor B satisfies these equivalent condi-
tions – let alone if R isn’t a Euclidean domain (or a domain!).
Fortunately, the solution is straightforward (if a bit ugly). Sup-
pose, once again, that R is a general commutative ring.
III.C.16. P ROPOSITION . det( AB) = det( A) det( B).
P ROOF. Write C = AB, so that c j` = ∑nk=1 a jk bk` . We compute

n n
det( A) det( B) = ∑ sgn(σ) ∏ a j,σ( j) × ∑ sgn(η ) ∏ bk,η (k)
σ ∈Sn j =1 η ∈Sn k =1
n n
[taking ρ = η ◦ σ] = ∑ sgn(ρ) ∑ ∏ a j,σ( j) ∏ bk,ρ(σ−1 (k))
ρ ∈Sn σ ∈Sn j =1 k =1
n
[identify k = σ ( j)] = ∑ sgn(ρ) ∑ ∏ a j,σ( j) bσ( j),ρ( j)
ρ ∈Sn σ ∈Sn j =1
n n
[see below] = ∑ sgn(ρ) ∏ ∑ a j` b`,ρ( j)
ρ ∈Sn j=1 `=1
n
= ∑ sgn(ρ) ∏ c j,ρ( j) = det(C ).
ρ ∈Sn j =1
As for the boxed equality, first observe7 that

!
n n n
(III.C.17) ∏ ∑ a j` b`,ρ( j) = ∑ ∏ a j,µ( j) bµ( j),ρ( j)
j =1 `=1 µ ∈Tn j =1
where Tn denotes all maps from {1, . . . , n} to itself (not just permu-
tations). For each µ ∈ Tn , let B(µ) denote the n × n matrix whose ith
row is the µ(i )th row of B for each i. Applying a variant of the first 3
steps above in reverse gives
n n n
∑ sgn(ρ) ∑ ∏ a j,µ( j) bµ( j),ρ( j) = ∑ ∑ ∏ a j,µ( j) ∏ bµ(ρ−1 (k)),k
ρ ∈Sn µ ∈Tn \ Sn j =1 ρ ∈Sn µ ∈Tn \ Sn j =1 k =1
!
n n
= ∑ ∏ a j,µ( j) ∑ sgn(ρ) ∏ bµ(ρ−1 (k)),k
µ ∈Tn \ Sn j =1 ρ ∈Sn k =1
!
n n
∑ ∏ a j,µ( j) ∑ sgn(ρ) ∏ bρ−1 (k),k
(µ)
=
µ ∈Tn \ Sn j =1 ρ ∈Sn k =1
!
n
= ∑ ∏ a j,µ( j) det B(µ) = 0
µ ∈Tn \ Sn j =1
since B(µ) has repeated rows for µ ∈ Tn \Sn . This shows that after
multiplying (III.C.17) by sgn(ρ) and summing over ρ ∈ Sn we can
omit terms with µ ∈/ Sn , proving the boxed equality.
What about invertibility? Well, you may recall that adjugate ma-
trices and Cramer’s rule were always the most horrible approach to
computing inverses and solving systems of equations in linear alge-
bra, unless the matrix entries were (say) transcendentals, polynomi-
als, etc. Since our entries are now in a general commutative ring,
let’s try this approach. Defining the cofactor
(n−1)×(n−1)
z }| {
(III.C.18) Aij := (−1)i+ j det ( ak` )k 6= i ,
` 6= j
7When expanding the LHS, the µth term on the RHS is obtained by choosing the
µ( j)th term (of the sum in parentheses) from the jth factor, as j runs from 1 to n.
116 III. RINGS
as an immediate (computational) consequence of III.C.1 we have the

Laplace expansions
n n
(III.C.19) det( A) = ∑ ai` Ai` = ∑ aki Aki .
`=1 k =1
Now let A0 (resp. A00 ) be the matrices obtained by deleting the ith
row (resp. column) and replacing it by the jth row (resp. column).
The repeated row (resp. column) makes the determinant zero:
)
0 = det A0 = ∑n`=1 a j` Ai`
(III.C.20) ( j 6 = i ).
0 = det A00 = ∑nk=1 akj Aki
Overall, (III.C.19)-(III.C.20) =⇒
n n
(III.C.21) ∑ a j` Ai` = (det A)δij = ∑ akj Aki .
`=1 k =1
Defining the adjugate matrix (or “classical adjoint”) by
(III.C.22) adj( A) := n × n matrix with (i, j)th entry A ji ,
we have the
III.C.23. P ROPOSITION . (adjA) A = det( A)1 = A(adjA).
P ROOF. The (i, j)th entry of (adjA) A is

n n
∑ [adjA]ik akj = ∑ akj Aki = (det A)δij .
k =1 k =1
This leads to the
III.C.24. T HEOREM . A ∈ Mn ( R) belongs to Mn ( R)∗ (=: GLn ( R))

if and only if det( A) ∈ R∗ .
adjA adjA
P ROOF. ( =⇒ ): by III.C.23, ( det A ) A = 1 = A( det A ).
( ⇐= ): by III.C.16, AB = 1 =⇒ det( A) det( B) = 1.
Even for R a field, for n > 1 Mn ( R) is not a division ring; even the
subring of diagonal matrices isn’t! On the other hand, the diagonal
matrices with all entries equal yield a subring which is isomorphic
to R. Are there other sorts of “in-between” subrings that are still

division rings,8 i.e. have all nonzero elements invertible?
III.C.25. E XAMPLE . Consider the subset of M2 (C) comprising ma-

trices of the form
! !
α γ a + bi c + di
M= = .
−γ̄ ᾱ −c + di a − bi
Since
! ! !
α γ α0 γ0 αα0 − γγ̄0 αγ0 + γᾱ0
=
−γ̄ ᾱ −γ̄0 ᾱ0 −γ̄α0 − ᾱγ̄0 −γ̄γ0 + ᾱᾱ0
is still such a matrix, and closure is obvious for addition, this is a
subring. Moreover,
0 = det( M ) = |α|2 + |γ|2 ⇐⇒ α = γ = 0 ⇐⇒ M = 0.
Hence, all nonzero entries of this type are invertible, and this subset
is a division ring, which we will later identify with H.
8Keep in mind here that GL ( R) is not a subring, since it’s not closed under addi-
n
tion. It’s just a group under multiplication.
118 III. RINGS
III.D. Ideals
Let R be a commutative domain. We say, given s, r ∈ R, that

defn.
(III.D.1) s|r (“s divides r”) ⇐⇒ r = st for some t ∈ R,
/ R ∗ ∪ {0})
and (for r ∈
!
defn. r = ab ( a, b ∈ R)
(III.D.2) r is irreducible ⇐⇒ .
=⇒ a or b ∈ R∗
If u ∈ R∗ and r = su, one writes r ∼ s and says that r and s are
associate;9 since s = ru−1 , this is an equivalence relation. The irre-
ducibles of Z are clearly the (±)primes.
√
Consider R = Z[ d], equipped with the “norm map”
N: R→Z
(III.D.3)
r 7→ rr̃,
√ √
where is r = m + n d, r̃ = m − n d.
III.D.4. L EMMA . R∗ = N −1 ({±1}).
P ROOF. Since rse = r̃ s̃, N is a homomorphism of multiplicative

monoids; and so N ( R∗ ) ⊂ Z∗ = {±1} ( =⇒ R∗ ⊂ N −1 ({±1})). If
N (r ) = ±1, then r̃ = ±r −1 =⇒ r ∈ R∗ .
√ √
III.D.5. P ROPOSITION . Let r ∈ Z[ d]\(Z[ d]∗ ∪ {0}), and sup-
pose N (r ) ∈ Z has no nontrivial (6= ±1) proper (6= ±N (r )) factors of
the form m2 − n2 d. Then r is irreducible.
P ROOF. If r = ab, then N (r ) = N ( a)N (b). By hypothesis, N ( a)

or N (b) = ±1. Hence a or b is a unit, by III.D.4.
√
III.D.6. E XAMPLE . In Z[ 10],
√
N (±1 + 10) = −9 and N (3) = 9 ;
9Alternatively, define r and s to be associate ⇐⇒ r |s and s|r; this is equivalent

(why?). If s|r and r - s, then s is a proper factor of r.
III.D. IDEALS 119
√
±3 are not of the form m2 − 10n2 (HW). Hence, ±1 + 10 and 3 are
irreducible. But
√ √
(III.D.7) (1 + 10)(−1 + 10) = 9 = 3 · 3 ,
and so the analogue of the Fundamental Theorem of Arithmetic I.B.1
fails.
This sort of ambiguity was a big problem for attempts to prove

Fermat’s Last Theorem in the mid-19th Century, or for solving Dio-
phantine equations more generally. A way out was proposed by
Kummer, who postulated “ideal elements” into which numbers in
the ring augmented by their inclusion would then decompose. For
√
instance, in the case of Z[ 10], these “ideal elements” π1 , π2 would
satisfy10

3 = √
 π1 π2
(III.D.8) 1 + 10 = π12
√
−1 + 10 = π22 .


Then (III.D.7) becomes π12 π22 = (π1 π2 )2 . Kummer showed that one
could construct a theory in which such elements would formally re-
spect divisibility and distributive properties. (Later it was realized
that they could be represented by actual elements in the “Hilbert
√
class field of Q( 10)”.) But Dedekind had the even nicer idea of
√
characterizing an “ideal number” π by its “shadow” in Z[ 10], con-
sisting of everything (formally) divisible by π. This is essentially our
modern notion of an ideal (in a number ring — the notion in gen-
eral is due to E. Noether). Indeed, the “shadows” of π1 and π2 in
the above example will be (in the notation about to be defined) the
ideals
√ √
(III.D.9) (3, 1 + 10) and (3, −1 + 10).
We will return to this example below.
Turning to some generalities, we have the
10To be clear, no actual elements in the ring satisfy these equations.

120 III. RINGS
III.D.10. D EFINITION . A right (resp. left) ideal I in a ring R is an

additive subgroup which is closed under right (resp. left) multipli-
cation by all elements of R:

 a, b ∈ I =⇒ a + b ∈ I

• a ∈ I =⇒ − a ∈ I

0∈I

• a ∈ I, r ∈ R =⇒ ar ∈ I (resp. ra ∈ I).
An ideal I ⊂ R is a left and right ideal.11
Given ideals I, J ⊂ R, I ∩ J is clearly an ideal. If S ⊂ R is a subset,

we define the ideal generated by S by
\
(III.D.11) (S) := I.
I ⊂ R ideal
I⊃S
III.D.12. P ROPOSITION . The ideal (S) consists of all finite sums
r1 s1 r10 + r2 s2 r20 + · · · + rk sk rk0
where ri , ri0 ∈ R, si ∈ S , and k ∈ N.
P ROOF. By the closure properties of III.D.10, all such finite sums

must belong to (S). By associativity and distributivity, the set of
such sums is itself closed under addition and multiplication by R,
hence is one of the ideals being intersected in RHS(III.D.11), and as
such contains (S).
III.D.13. D EFINITION . Given I ⊂ R an ideal, I is

• finitely generated ⇐⇒ I = (S) for some finite subset S ⊂ R.
• principal ⇐⇒ I = ( a) for some element a ∈ R.
Note that if R is commutative, then ( a) = {ra | r ∈ R}, and
( a1 , . . . , a m ) = {r1 a1 + · · · r m a m | r1 , . . . , r m ∈ R }.
11Note that this is a stronger notion than being a “subrng” because of the closure
under multiplication by elements of R. And yes, I mean “subrng” not “subring”:
except for R itself, ideals in R do not contain 1.
III.D. IDEALS 121
We can also consider “sums” and “products” of ideals: define

(
I + J := ( I ∪ J ) = { a + b | a ∈ I, b ∈ J }
(III.D.14)
I J := ( I J ) = {∑ik=1 ai bi | ai ∈ I, bi ∈ J, k ∈ N},
where I J is the set of products { ab | a ∈ I, b ∈ J }. To state the
obvious:
III.D.15. P ROPOSITION . Suppose I = (S) and J = (T ).

(i) I + J = (S ∪ T ).
(ii) If R is commutative, then I J = ({st | s ∈ S , t ∈ T }) = (S T ).
(iii) In particular, if I = ( a) and J = (b), then I + J = ( a, b) and (for R
commutative) I J = ( ab).
Furthermore, if R is commutative and a, b ∈ R, we have
III.D.16. P ROPOSITION (“Caesar’s lemma”). To divide is to con-

tain:12
a|b ⇐⇒ ( a) ⊇ (b).
P ROOF. If ra = b, then
(b) = (ra) = {r 0 ra | r 0 ∈ R} ⊂ {r 00 a | r 00 ∈ R} = ( a).

Conversely, ( a) ⊃ (b) =⇒ b ∈ ( a) =⇒ b = ra for some r ∈ R.
√
III.D.17. E XAMPLE . Returning to III.D.6 ff and R = Z[ 10], we
compute
√ √ √
(3, 1 + 10)2 = 9, 3 + 3 10, 11 + 2 10
√ √ √ √ √
= (1 + 10)(−1 + 10), (1 + 10)3, (1 + 10)(1 + 10)
√
⊂ (1 + 10),
12A rough translation into algebra-ese of J. Caesar’s famous maxim “divide et

impera”. I jest, but this is useful as a mnemonic device for remembering the rule.
122 III. RINGS
making use of III.D.15(ii) to square the ideal.13 Similarly one shows

√ √
that (3, −1 + 10)2 ⊂ (−1 + 10) and
√ √ √ √
(3, 1 + 10)(3, −1 + 10) = 9, 3 + 3 10 − 3 + 3 10
⊂ (3).
For the reverse inclusions,14

√ √ √ √
1 + 10 = −(11 + 2 10) + 9 + (3 + 3 10) ∈ (3, 1 + 10)2
√ √
=⇒ (1 + 10) ⊂ (3, 1 + 10)2 ,
√ √
and similarly (−1 + 10) ⊂ (3, −1 + 10)2 ; while
√ √ √ √
3 = 9 − (3 + 3 10) + (−3 + 3 10) ∈ (3, 1 + 10)(3, −1 + 10)
√ √
=⇒ (3) ⊂ (3, 1 + 10)(3, −1 + 10).
√ √
So if we set I1 = (3, 1 + 10) and I2 = (3, −1 + 10), we indeed
have
√ √
I1 I2 = (3) , I12 = (1 + 10) , and I22 = (−1 + 10)
and the ideals serve their intended function, recovering an analogue

of (III.D.8).
Returning to the setting of a general ring R, let I ( R be a proper

ideal. Clearly, this is a normal subgroup of the additive (abelian)
group, and so we can construct the (additive) quotient group R/I.
Its elements are the equivalence classes defined by the equivalence
relation
a ≡ b ⇐⇒ a − b ∈ I.
That is, they are the cosets a + I, with the addition rule
(III.D.18) ( a + I ) + (b + I ) = ( a + b) + I.
13
This is an important point: the product ( a, b)(c, d) is the ideal generated by the
set of products { a, b} {c, d} := { ac, ad, bc, bd}.
14The basic principle being applied here (in the case of 1-element sets) is that if a
set S is contained in an ideal I, then (S) ⊂ I.
III.D. IDEALS 123
We now define a multiplicative structure on R/I by the rule
(III.D.19) ( a + I )(b + I ) := ab + I,
with identity coset 1 + I. The main check required is that (III.D.19) is
well-defined: given a0 = a + α ∈ a + I and b0 = b + β ∈ b + I,
( a0 + I )(b0 + I ) = a0 b0 + I = ( a + α)(b + β) + I
= ab + αb + aβ + αβ + I
| {z }
∈I
= ab + I.
Distributivity is clear from (III.D.18)-(III.D.19) and distributivity in
R. Hence, R/I has the structure of a ring.
III.D.20. R EMARK . In our study of groups, we had two “stupid
quotients”, G/h1i(∼= G ) and G/G (∼ = {1}). Here, the only stupid
quotient ring is R/(0) = R; because {0} is not a ring, we cannot con-
sider R/R, and accordingly the definition of quotient ring requires a
proper ideal.
III.D.21. E XAMPLES . (i) (n) = nZ ⊂ Z is a proper ideal (n > 1),
and Z/(n) (or Z/nZ) is just Zn (viewed as a ring).
(ii) In Z[ x ]/( x2 − 10), any element is of the form P( x ) + ( x2 − 10)
(where ( x2 − 10) is the principal ideal). Applying polynomial divi-
sion, this equals { x2 − 10} · Q( x ) + R( x ) + ( x2 − 10) = R( x ) + ( x2 −
10), where R( x ) = ax + b.
(iii) In the ring C0 (M) of continuous functions on a manifold M, the
subset IS of functions identically zero on a subset S ⊂ M is an ideal.
In C0 (M)/IS , cosets f + IS and g + IS are the same ⇐⇒ f − g ∈ IS
⇐⇒ f and g have the same restriction to S . So the quotient can be
thought of as a ring of functions on S of some sort.
√ √
(iv) We can consider Z[ 10] modulo the ideals (3), (±1 + 10), and
√
(3, ±1 + 10).
(v) Let R be commutative. While there are left [resp. right] ideals in
Mn ( R) (e.g. matrices with last column [resp. row] zero) that “take
124 III. RINGS
advantage of the matrix structure”, there are no (2-sided) ideals that

do this:
III.D.22. P ROPOSITION . If I ⊂ R is an ideal, then Mn ( I ) ⊂ Mn ( R)

is an ideal.15 In fact, all ideals of Mn ( R) arise in this way.
P ROOF. If A ∈ Mn ( I ), B ∈ Mn ( R), then entries ∑k aik bkj of AB

are obviously in I, hence AB ∈ Mn ( I ).
Let J ⊂ Mn ( R) be an ideal, and let
I := { a ∈ R | a is an entry in some matrix belonging to J }.
Then J ⊂ Mn ( I ).
To show I is an ideal: given A ∈ J, J contains
eki Ae j` = eki (∑m,n amn emn )e j` = ∑m,n amn δim δnj ek` = aij ek` .
Hence for all a ∈ I for all k, `, we have aek` ∈ J. Now for α, β ∈ I,

r ∈ R,
(
(α + β)e11 = αe11 + βe11 ∈ J =⇒ α + β ∈ I
αe11 , βe11 ∈ J =⇒
αre11 = αe11 · re11 ∈ J =⇒ αr ∈ I
(and similarly for rα, −α).
To show J ⊃ Mn ( I ): given elements αij ∈ I, each αij eij ∈ J (by the
last paragraph). Thus, a general element ∑i,j αij eij of Mn ( I ) belongs
to J.
What about ideals in Q, Q[i], R, C?
III.D.23. T HEOREM . Let R be a commutative ring. Then
R is a field ⇐⇒ R has no nontrivial proper ideals.
P ROOF. ( =⇒ ): Let I ⊂ R be a nontrivial ideal, a ∈ I \{0}. Given

any b ∈ R, b = a( a−1 b) ∈ I, so I = R.
( ⇐= ): Let a ∈ R\{0}; then ( a) = { ar | r ∈ R} contains { a}
hence is nontrivial. By hypothesis, it must be R. Hence for any b ∈ R,
there is an r ∈ R such that ar = b; take b = 1.
15e.g., M ( pZ) ⊂ M (Z)
n n
III.D. IDEALS 125
Notice where the proof of “( ⇐= )” breaks down for something

like R = Mn (C) (which satisfies the hypothesis on ideals by III.D.22):
we get that {∑i ri ari0 | ri , ri0 ∈ R} = R, which doesn’t imply that a is
invertible.
At this point, we should mention the key example:
III.D.24. P ROPOSITION . Zm is a field ⇐⇒ m is prime.
P ROOF. ( =⇒ ): obvious, since m composite =⇒ Zm not a

domain.
( ⇐= ): Given a + (m) (or “ā”) in Zm \{0}, with a ∈ {1, . . . , m −
1}, we know that the gcd of m and a is 1 (as m is prime). So there
exist k, ` ∈ Z such that ka + `m = 1 =⇒ (k + (m))( a + (m) =
1 − ` m + ( m ) = 1 + ( m ).
Before turning to homomorphisms, here is one more
III.D.25. D EFINITION . An ascending chain of ideals is a nested

sequence
I1 ⊆ I2 ⊆ I3 ⊆ · · · ⊆ R
of ideals. Note that this is a chain in the set-theoretic sense (totally
ordered), while the set of all ideals is partially ordered by inclusion.
III.D.26. L EMMA . The union ∪ j≥1 Ij ⊂ R is an ideal. More generally,

for any chain C in the set of ideals of R, ∪ J ∈C J is an ideal of R.
P ROOF. Any element (or pair of elements) of the union is con-

tained in some member J0 ∈ C , because of the total ordering. By the
closure properties III.D.10 of J0 , the sum of these elements and their
products by elements of R are contained in J0 hence in ∪ J ∈C J. So this
union satisfies the closure properties too.
III.D.27. T HEOREM . Let I ( R be a proper ideal. Then there exists a

maximal proper ideal I0 which contains I. (Here “maximal” means merely
that there is no ideal J with I0 ( J ( R. )
P ROOF. Let P denote the set of proper ideals of R containing I,

partially ordered by ⊆, and let C be a chain in P . Consider the set
126 III. RINGS
IC := ∪ J ∈C J, which by the Lemma is an ideal. Clearly, since every J

contains I and doesn’t contain 1, IC ⊃ I and 1 ∈ / IC , which implies
IC ∈ P .
We have shown that every chain in P has an upper bound (in P ).
So Zorn produces a maximal element in P , which must be a maximal
proper ideal containing I.
III.E. HOMOMORPHISMS OF RINGS 127
III.E. Homomorphisms of rings
Let R and S be rings.
III.E.1. D EFINITION . (i) A ring homomorphism ϕ : R → S is a

map which is both a homomorphism of additive groups and multi-
plicative monoids: ϕ(r1 + r2 ) = ϕ(r1 ) + ϕ(r2 ), ϕ(r1 r2 ) = ϕ(r1 ) ϕ(r2 ),
and ϕ(1R ) = 1S .
(ii) A ring isomorphism is a homomorphism of rings which is
injective and surjective. (Equivalently: there exists a homomorphism
η : S → R such that η ◦ ϕ = idR and ϕ ◦ η = idS .)
III.E.2. WARNING . In contrast to the case of groups, it is essential

to include “ϕ(1R ) = 1S ” in III.E.1(i). This not only prohibits (say)
multiplication-by-2 from giving a ring homomorphism from Z to Z;
it means that there is no such thing as a trivial (zero) ring homomor-
·2 0
phism. Both Z → Z and Z → Z are “rng homomorphisms”.
III.E.3. P ROPOSITION . (i) ϕ( R) is a subring of S, and

(ii) ker( ϕ) (:= ϕ−1 ({0})) is a proper ideal in R.
P ROOF. (i) ϕ( R) contains 1, and given α = ϕ(r1 ), β = ϕ(r2 ) ∈

ϕ( R), we have α + β = ϕ(r1 + r2 ) ∈ ϕ( R) and αβ = ϕ(r1 r2 ) ∈ ϕ( R).
(ii) Given r ∈ R and κ1 , κ2 ∈ ker( ϕ), we have ϕ(κ1 + κ2 ) =
ϕ(κ1 ) + ϕ(κ2 ) = 0 + 0 = 0 =⇒ κ1 + κ2 ∈ ker( ϕ), and ϕ(rκ1 ) =
ϕ(r ) ϕ(κ1 ) = ϕ(r ) · 0 = 0 etc. =⇒ rκ1 , κ1 r ∈ ker( ϕ). In particular,
−κ1 and 0κ1 = 0 are in ker( ϕ). Finally, ker( ϕ) is proper because it
doesn’t contain 1.
III.E.4. E XAMPLES . (i) “Evaluation” maps evr : R[ x ] → R send-

ing P( x ) 7→ P(r ) (or their products, as in III.A.3(iv)) are homomor-
phisms.
(ii) An injective homomorphism (or embedding) ϕ : H ,→ M2 (C) is
obtained by sending 1 7→ 10 01 , i 7→ 0i −0i , j 7→ −01 01 , and k 7→

0 i . This gives an isomorphism of H with a subring of M (C)

i 0 2
(specifically, the one from III.C.25). The only thing to check is that
the matrices behave “the same” as i, j, k under multiplication.
128 III. RINGS
(iii) The natural map ν : R R/I sending r 7→ r + I (or “r̄”), where

I ⊂ R is a proper ideal, is clearly consistent with III.E.3.
(iv) det : Mn (C) → C is not a ring homomorphism. (Why?)
III.E.5. F UNDAMENTAL T HEOREM OF R ING H OMOMORPHISMS .

Given ϕ : R → S, with K := ker( ϕ), there exists a unique ring homomor-
phism ϕ̄ : R/K ,→ S making the diagram
ϕ
R / S
=
ν !! . ϕ̄
R/K
commute. In particular, the image ϕ( R) ∼
= R/K.
P ROOF. By III.E.3(ii), R/K is well-defined as a ring; and by II.I.20,
there exists a unique additive group homomorphism ϕ̄ such that
ϕ̄ ◦ ν = ϕ. It is only left to check that ϕ̄ is a ring homomorphism:
ϕ̄(r̄1 r̄2 ) = ϕ̄(ν(r1 )ν(r2 )) = ϕ̄(ν(r1 r2 )) = ϕ(r1 r2 ) = ϕ(r1 ) ϕ(r2 ) =
ϕ̄(ν(r1 )) ϕ̄(ν(r2 ) = ϕ̄(r̄1 ) ϕ̄(r̄2 ).
III.E.6. E XAMPLES . (continuing III.D.21)

(i) Consider the evaluation map
√
ev√10 : Z[ x ] −→→ Z[ 10]
√
sending P( x ) 7−→ P( 10)
and x2 − 10 7−→ 0.
Clearly x2 − 10 ∈ K and thus ( x2 − 10) ⊂ K := ker(ev√10 ).

√
Conversely, if P( 10) = 0 and P is even, then P( x ) = Q( x2 )
for some polynomial Q(y), hence Q(10) = 0 =⇒ y − 10 | Q(y)
=⇒ x2 − 10 | P( x ) in Z[ x ]. If P isn’t even, then P = P1 + xP2
√
where Pi ( x ) = Qi ( x2 ) and 0 = Q1 (10) + 10Q2 (10) =⇒ again
x2 − 10 | P( x ). Invoking III.D.16 (“Caesar”), we get ( x2 − 10) ⊃ K.
Conclude that
Z[ x ] ∼ √
= Z[ 10].
( x2 − 10)
(ii) If M is a manifold with submanifold16 S , then the restriction map
C0 (M) −→
→ C0 (S)
f 7−→ f |S
is a surjective homomorphism, with kernel K = IS . So

C0 (M)
C0 (S) ∼
= .
IS
Similar isomorphisms show up in mathematics everywhere from co-
ordinate rings (in algebraic geometry) to multiplier algebras (in op-
erator theory).
(iii) Let’s look at the map
√
α : Z[ 10] −→
→ Z9
√
defined by a + b 10 7−→ a − b
√
(which sends 1 + 10 7−→ 0̄).
Is this a homomorphism? It sends 1 7→ 1̄, respects “+”, and satisfies

√ √ √
α ( a + b 10)(c + d 10) = α ( ac + 10bd) + ( ad + bc) 10
= ac + 10bd − ( ad + bc)
= ac + bd − ad − bc
= ( a − b)(c − d)
√ √
= α( a + b 10) · α(c + d 10),
√
so yes. Clearly (1 + 10) ⊂ ker(α). Conversely,
√
a + b 10 ∈ ker(α) =⇒ a = b + 9n (n ∈ Z)
√ √
=⇒ a + b 10 = b(1 + 10) + 9n
√ √
= b + n(−1 + 10) (1 + 10)
16We will not get surjectivity if S is an arbitrary subset.

130 III. RINGS
√
shows that ker(α) ⊂ (1 + 10). Conclude that
√
Z[ 10] ∼
√ = Z9 ;
(1 + 10)
√ √
by a similar argument, we can replace (1 + 10) by (−1 + 10).
(iii’) What about
√
β : Z[ 10] −→
→ Z3 × Z3
√
a + b 10 7−→ ( a + b, a − b)
3 7−→ (0̄, 0̄) ?
This sends 1 7→ (1̄, 1̄) and ( a + b, a − b) · (c + d, c − d) =

√ √
( ac + bd + ad + bc, ac + bd − ( ad + bc)) = β(( a + b 10)(c + d 10)).
So β is a homomorphism with ker( β) ⊃ (3). Moreover, a ≡ b and
(3)
√
a ≡ −b =⇒ a ≡ 0 ≡ b =⇒ a + b 10 ∈ (3). So
(3) (3) (3)
√
Z[ 10] ∼
= Z3 × Z3 .
(3)
(iii”) Finally, for

√
γ : Z[ 10] −→
→ Z3
√
a + b 10 7−→ a − b
3 7−→ 0̄
√
1+ 10 7−→ 0̄
√
the general element of ker(γ) is 3n + b(1 + 10)
√
√ Z[ 10] ∼
=⇒ ker(γ) = (3, 1 + 10) =⇒ √ = Z3 .
(3, 1 + 10)
(iv) For an example of a more general sort, consider (for any ring R)
η : Z −→ R
0 7−→ 0R
1 7−→ 1R
Z>0 3 n 7−→ 1R + · · · + 1R (n times)
−n 7−→ −(1R + · · · + 1R ).
Clearly η (n + m) = η (n) + η (m), and η (nm) = η (n)η (m) (using
distributivity). The image η (Z) is called the prime ring, and is the
smallest subring of R. Any ideal of Z is of the form (n), since these
are (as we checked before) the additive subgroups. Conclude that
η (Z) ∼
= Z if char( R) = 0, and η (Z) ∼
= Zm if char( R) = m is finite.
III.E.7. R EMARK . Given a homomorphism ϕ : R → S, we have
Z
ηR ηS
ϕ
R / S,
ϕ
with n̄ 7→ n̄. On the one hand, this implies char(S) | char( R), which
could rule out some homomorphisms. If char( R) = 0 it won’t rule
out anything, but here is something which could: if α ∈ R satisfies a
polynomial equation 0 = ∑k ak αk , ak ∈ Z (i.e. ηR (Z)), we must have
(writing β := ϕ(α)) that 0 = ∑k āk βk . One could then try to show
that S doesn’t contain such a β.
With essentially no work, the two isomorphism theorems from

§II.I lift to the ring setting:
132 III. RINGS
III.E.8. F IRST I SOMORPHISM T HEOREM . Let ϕ : R S be a surjec-

tive ring homomorphism with kernel K. Then ϕ induces a 1-to-1 correspon-
dence
( ) ( )
ideals I ⊂ R ideals
←→
containing K J⊂S
via I 7−→ ϕ( I ) ,
∼
=
and isomorphisms R/I → S/ϕ( I ).
P ROOF. We only need to check that ϕ( I ) and ϕ−1 ( J ) are closed

under multiplication by R; the rest follows from II.I.25 and III.E.5.
Given I ⊂ R, Sϕ( I ) = ϕ( R) ϕ( I ) = ϕ( RI ) = ϕ( I ) =⇒ ϕ( I ) is an
ideal.
Given J ⊂ S, α ∈ ϕ−1 ( J ), and r ∈ R, we have ϕ(rα) = ϕ(r ) ϕ(α) ∈
SJ = J =⇒ rα ∈ ϕ−1 ( J ) =⇒ ϕ−1 ( J ) is an ideal.
III.E.9. S ECOND I SOMORPHISM T HEOREM . Given I ⊂ R an ideal

and S ⊂ R a subring. Then:
(i) S + I ⊂ R is a subring having I as an ideal;
(ii) S ∩ I is an ideal in S; and
∼
=
(iii) s + (S ∩ I ) 7→ s + I induces an isomorphism S/(S ∩ I ) → (S + I )/I.
P ROOF. Left to you.
III.E.10. E XAMPLE . (i) Referring to Example III.E.6(iii), we can ap-

√ √
ply III.E.8 to α : Z[ 10] Z9 to determine ideals in R := Z[ 10].
Since S := Z9 contains one nontrivial proper ideal (namely (3̄)), R
√
contains one proper ideal containing (but 6=) (1 + 10). Clearly, this
√
is (3, 1 + 10), and so we get for free
√
Z[ 10] ∼ Z9 ∼
√ = = Z3 .
(3, 1 + 10) Z3
√
(ii) With the same R, take S := Z ⊂ R and I = (1 + 10) ⊂ R.
Clearly S + I = R, and applying III.E.9 gives
√
Z ∼ Z [ 10]
√ = √ ,
Z ∩ (1 + 10) (1 + 10)
√
which we know is ∼ = Z9 . Conclude that Z ∩ (1 + 10) = (9).
Here is a more interesting application of the Fundamental Theo-
rem III.E.5.
III.E.11. D EFINITION . We say that two ideals I, J ⊂ R are rela-
tively prime (or coprime) if I + J = R, or equivalently that there
exist elements ı ∈ I and  ∈ J such that ı +  = 1. (You should
check that in Z, (m) and (n) are relatively prime iff m and n are, i.e.
gcd(m, n) = 1.)
III.E.12. C HINESE R EMAINDER T HEOREM . Let I1 , . . . , Im be pair-
wise relatively prime ideals in a ring R; that is, for each i 6= j, Ii + Ij = R.
Then
R/(∩m ∼
j=1 I j ) = R/I1 × · · · × R/Im .
P ROOF. Clearly
ϕ : R −→ R/I1 × · · · × R/Im
r 7−→ (r + I1 , . . . , r + Im )
is a homomorphism. We must show that it is surjective with kernel

∩m j=1 I j = : I, and then the Fundamental Theorem does the rest of the
work.
Suppose the result is known for less than m ideals (with m ≥ 3).
Then setting I 0 := ∩m 0 ∼ m
j=2 I j , we have R/I = × j=2 R/I j . By assumption,
for each pair I1 and Ij we have elements α j ∈ I1 and β j ∈ Ij such that
α j + β j = 1. Hence,17
m m
1= ∏(α j + β j ) ∈ ∏( I1 + Ij ) ⊂ I1 + I2 · · · Im ⊂ I1 + I 0
j =2 j =2
17Note that all terms of the product

∏m
j=2 ( I1 + I j ) are contained in I1 except for the
term I2 · · · Im .
134 III. RINGS
=⇒ I1 + I 0 = R. Hence R/I ∼ = R/I 0 × R/I1 as desired.

What remains is to check the m = 2 case. First, ker( ϕ) consists of
those r ∈ R with ϕ(r ) = (0 + I1 , 0 + I2 ), or equivalently, r ∈ I1 ∩ I2 .
For surjectivity of ϕ: given a := ( a + I1 , b + I2 ) ∈ R/I1 × R/I2 ,
I1 + I2 = R =⇒ there exist ı1 ∈ I1 , ı2 ∈ I2 such that a − b = −ı1 + ı2
=⇒ a + ı1 = b + ı2 =: r, with ϕ(r ) = a.
III.E.13. R EMARK . (i) More explicitly, the Theorem is saying that

if r1 , . . . , rm are elements of R, and I1 , . . . , Im pairwise coprime, then:
• there exists an r ∈ R such that r ≡ ri mod Ii for every i; and
• this r is unique up to the addition of elements in I1 ∩ · · · ∩ Im .
(ii) If R is commutative and I1 and I2 are relatively prime, with α ∈ I1

and β ∈ I2 such that α + β = 1, then a ∈ I1 ∩ I2 =⇒ a = a(α + β) =
αa + bβ ∈ I1 I2 . Conversely, it is immediate that I1 I2 ⊂ I1 ∩ I2 ; and so
I1 I2 = I1 ∩ I2 . From here, its obviously true for m > 2 as well: if R is
commutative and the Ij are pairwise coprime, then
I1 ∩ · · · ∩ Im = I1 · · · Im .
The original form of III.E.12 is a result about congruences in num-

ber theory, a version of which of which was discovered by Sun Tzu
in the 3rd Century.
III.E.14. C OROLLARY. Let k1 , . . . , k m be pairwise coprime integers;

that is, (k i , k j ) = 1 (∀i 6= j). Then18
∼
=
Z/k1 · · · k m Z −→ Z/k1 Z × · · · × Z/k m Z.
Taking units on both sides recovers the results on units in Z/mZ from
II.E.13-II.E.14.
But one needn’t apply the Chinese Remainder Theorem only to

integers:
18or if you prefer, Z ∼

k1 ···k m = Zk 1 × · · · × Zk m .
√ √
III.E.15. E XAMPLE . In R = Z[ 10], the ideals I1 = (1 + 10) and
√
I2 = (−1 + 10) are coprime, in view of
√ √ √ √
(1 + 10)(−1 + 10) − 4(1 + 10) + 4(−1 + 10) = 1.
Moreover, I1 I2 = (9). So
√ √ √
Z[ 10] ∼ Z[ 10] Z[ 10] ∼
= √ × √ = Z9 × Z9 .
(9) (1 + 10) (−1 + 10)
136 III. RINGS
III.F. Fields
Given a field F, the intersection of all its subfields is called the

prime subfield. Clearly, this contains the prime ring η (Z), which is
isomorphic to Z p (p prime) or to Z. In the first case, Z p is the prime
subfield; in the latter, we may extend η : Z ,→ F to Q by η ( rs ) :=
η ( r ) η ( s ) −1 .
0
This extension is well-defined since given rs0 = rs , we have r 0 s =
rs0 =⇒ η (r 0 )η (s) = η (r )η (s0 ) =⇒ η (r 0 )η (s0 )−1 = η (r )η (s)−1 . To
see that it is injective, recall from III.D.23 that a field has no nontrivial
proper ideals. Hence
all (ring) homomorphisms from a field to a ring

(III.F.1)
are injective.
We conclude
III.F.2. P ROPOSITION . The prime subfield of a field F is isomorphic to
Q or Z p .
Also note the following about ring homomorphisms ϕ : F → R
(in addition to (III.F.1)): given f ∈ F (with inverse f −1 ), we have
ϕ( f ) ϕ( f −1 ) = ϕ( f f −1 ) = ϕ(1) = 1 =⇒ ϕ( f −1 ) = ϕ( f )−1 .
One way to construct fields (beyond the usual suspects) is via
quotient rings. For the remainder of this section, let R denote a com-
mutative ring.
III.F.3. T HEOREM . If I ( R denotes a proper ideal, then
R/I is a field ⇐⇒ I is maximal.

P ROOF. ( ⇐= ): Given a proper ideal J ( R/I, its preimage un-
der ν : R R/I is a proper ideal containing I (and equal to I iff
J = {0}) by III.E.8. Hence if I is maximal, the only possibility for J is
{0}. By III.D.23, R/I is a field.
( =⇒ ): Assume R/I is a field, and let J ⊂ R be an ideal with
I ( J. We will show that J = R so that I is maximal.
Given any r ∈ J \ I, the ideal ( I, r ) generated by I and r is con-
tained in J. Since r ∈
/ I, we have ν(r ) 6= 0. As ν is onto, there exists
III.F. FIELDS 137
r 0 ∈ R with ν(r 0 ) = ν(r )−1 ; and then
ν(1 − rr 0 ) = ν(1) − ν(r )ν(r 0 ) = 1 − 1 = 0 =⇒ a := 1 − rr 0 ∈ I.
This means 1 = a + rr 0 ∈ ( I, r ) hence ( I, r ) = J = R.

III.F.4. E XAMPLES . (i) Similarly to III.E.6(i), we have (by the Fun-
Q[ x ] ∼
= √
damental Theorem III.E.5) ( x2 −10) → Q[ 10], which we know is a
field. Hence ( x2 − 10) is maximal.
(ii) Given a submanifold S ⊂ M, when is C0 (S) a field? It can only
consist of one point — otherwise there are obvious zero-divisors. So
IS is maximal ⇐⇒ S is a point.
√ √
Z[ 10] ∼
(iii) Since √ = Z3 , the ideal (3, 1 + 10) is maximal. None of
(3,1+ 10) √ √
the principal ideals (1 + 10), (−1 + 10), (3) are.
Briefly veering off topic, there is an important variant of III.F.3.
III.F.5. D EFINITION . An ideal I ( R is prime if
ab ∈ I =⇒ a ∈ I or b ∈ I.
III.F.6. T HEOREM . R/I is a domain ⇐⇒ I is prime.
P ROOF. I is not prime ⇐⇒ ∃ a, b ∈ R\ I such that ab ∈ I. Equiv-
alently, taking ā = a + I etc., ∃ ā, b̄ ∈ ( R/I )\{0} such that āb̄ = 0̄;
that is to say, R/I is not a domain.
Since fields are domains . . .
III.F.7. C OROLLARY. Maximal ideals are prime.
Turning back to the beginning of this section, note that in a sense
Q was the subfield of F generated by Z (in the characteristic zero
case). We want to generalize this.
III.F.8. P ROPOSITION . Let R be a subring of a field F. Then the inter-
section of all subfields containing R (the “subfield generated by R”) is
R × R\{0}
(III.F.9) {αβ−1 | α ∈ R, β ∈ R\{0}} ∼
= ,
≡
where (α, β) ≡ (γ, δ) ⇐⇒ αβ−1 = γδ−1 in F ⇐⇒ αδ = βγ in R.
138 III. RINGS
P ROOF. We only need to check that III.F.9 is a subfield, since any

field containing R clearly contains it. The only remotely nontriv-
ial check is closure under addition: αβ−1 + γδ−1 = αδβ−1 δ−1 +
βγβ−1 δ−1 = (αδ + βγ)( βδ)−1 .
Going further, we can perform this construction without a “ref-
erence field” F.
III.F.10. T HEOREM . Any commutative domain R can be embedded in

a field.
P ROOF. Again we define an equivalence relation

def.
(III.F.11) ( a, b) ∼ (c, d) ⇐⇒ ad = bc
on R × R\{0}. This is
• reflexive: ab = ba
• symmetric: ad = bc ⇐⇒ cb = da
• transitive: ad = bc and c f = de =⇒ ad f = bc f = bde =⇒
d( a f − be) = 0 (and d 6= 0) =⇒ a f = be (since R is a domain).
Define (as a set)
R × R\{0}
F{ R } : = ,
∼
with 1F{ R} := (1, 1), 0F{ R} := (0, 1),
( a, b) · (c, d) := ( ac, bd) , and ( a, b) + (c, d) := ( ad + bc, ad).

These operations are well-defined: for instance, if ( a, b) ∼ ( a0 , b0 ), i.e.
ab0 = ba0 , then ( a0 d + b0 c)bd = b0 d( ad + bc) hence
( a0 , b0 ) + (c, d) = ( a0 d + b0 c, b0 d) = ( ad + bc, bd).

(The other checks in this vein are left to you.)
Next, we check the properties of a ring: we have
• (0, 1) + ( a, b) = (0b + 1a, 1b) = ( a, b)
• (1, 1) · ( a, b) = ( a, b)
• (− a, b) + ( a, b) = (− ab + ba, b2 ) = (0, b2 ) = (0, 1)
• ( a, b) · ((c, d) + (e, f )) = ( a(c f + de), b(d f )) = ( acb f + abde, b2 d f )
= ( ac, bd) + ( ae, b f )
III.F. FIELDS 139
and the other distributive and associative laws can also be checked.
Moreover, if ( a, b) 6= 0F{ R} (i.e. a 6= 0), then
(b, a) · ( a, b) = (ba, ab) = (1, 1) = 1F{ R}

and so F{ R} is a field.
Finally, we need to show that
φ : R → F{ R }
r 7→ (r, 1)
is an injective homomorphism, embedding R as a subring. We have

φ ( 1 ) = 1F { R } , φ ( r 1 + r 2 ) = ( r 1 + r 2 , 1 ) = ( r 1 , 1 ) + ( r 2 , 1 ) = φ ( r 1 ) +
φ(r2 ), etc.; and if φ(r ) = 0F{ R} then (r, 1) = (0, 1) =⇒ r · 1 = 1 · 0
=⇒ r = 0, done.
III.F.12. D EFINITION . F{ R} is called the field of fractions of R.
We can put together III.F.8 and III.F.10 as follows:
III.F.13. P ROPOSITION . Given a commutative domain R, any injective

ring homomorphism ϕ : R ,→ F factors through R’s field of fractions

R p
ϕ
/ F;
<
φ ! - ϕ̃
F{ R }
and if the only subfield of F containing ϕ( R) is F itself, then F ∼

= F{ R }.
P ROOF. The second statement is obvious (since F{ R} ∼ = ϕ̃(F{ R})

is a subfield containing ϕ( R)), so what we need to do is check that
ϕ̃(( a, b)) := ϕ( a) ϕ(b)−1
is well-defined and a homomorphism (easy and left to you), as well

as injective: if ϕ( a) ϕ(b)−1 = 0 then ϕ( a) = 0 =⇒ a = 0 =⇒
( a, b) = (0, 1).
140 III. RINGS
√ √
III.F.14. E XAMPLES . (i) Consider ϕ : Z[ d] ,→ Q[ d]. Any sub-
field containing its image contains (∀ a, b, c ∈ Z, c 6= 0) c−1 and
√ √ √ √
( a + b d)c−1 hence Q[ d]. So Q[ d] ∼ = F{Z[ d]}.
(ii) Let F be a field, R = F[ x ]. Then F( x ) := F{F[ x ]} consists of
“rational functions” in x.
Associated to the field of fractions is a different notion of ideal.

(We continue to take R a commutative domain.)
III.F.15. D EFINITION . (i) A fractional ideal of R is a subset J ⊂

F{ R} of the form f I := f · I = { f a | a ∈ I } for some f ∈ F{ R} and
ideal I ⊂ R.
(ii) J is principal if I is.
(iii) J is invertible if there exists a fractional ideal J 0 with J J 0 = R.
Principal fractional ideals are invertible since they are of the form
f R ⊂ F{ R} and we have f R · f −1 R = R2 = R. Denote by
• J ( R) := the set of fractional ideals
• J ( R)∗ := the set of invertible fractional ideals
• P J ( R) := the set of principal fractional ideals.
Under the obvious multiplication f I · f 0 I 0 = f f 0 I I 0 , J ( R)∗ forms
an abelian group with identity element R, and (normal) subgroup
P J ( R ).
III.F.16. D EFINITION . C `( R) := J ( R)∗ /P J ( R) is the ideal class

group.
We shall discuss its relation to uniqueness of factorization later.
III.F.17. E XAMPLE . Assume d ∈ Z\{0} squarefree, with d 6≡ 1,

(4)
√
and consider an ideal of the form I = (α, β) inside R = Z[ d].
^ √ √
Writing m+n d := m − n d, and Ĩ = (α̃, β̃), we will compute I Ĩ.
√
But first, we need a little “lemma”. Suppose that a + b d (a, b ∈
Q) solves an integer equation of the form x2 + Bx + C = 0. Then
√ √
2
a + b d = − B± 2B −4C =⇒ B2 − 4C = A2 d for some A ∈ Z.
III.F. FIELDS 141
Since d 6≡ 1, we get B2 − 4C 6≡ 1, which forces B (and thus A) to be

(4) (4)
√
even, whence a, b ∈ Z. What this shows is that an element of Q[ d]
belongs to R if it solves a monic integral quadratic equation.
Returning to the computation: as the norm map sends R → Z,
and α, β ∈ R, we have
I Ĩ = (αα̃, β β̃, α β̃, βα̃) = (αα̃, β β̃, α β̃ + βα̃, βα̃) = ( g, βα̃).

| {z }
in Z, with gcd =: g
βα̃
Since g is a root of
(x − βα̃
g )( x − α β̃
g ) = x2 − ( βα̃+g α β̃ ) x + αgα̃ · βgβ̃ ,
| {z } | {z }
∈Z ∈Z
βα̃
our “lemma” tells us that g ∈ R hence g | βα̃ in R. So we conclude
that
I Ĩ = ( g) ,
a very useful result called
√
Hurwitz’s Theorem (which also works for
d ≡ 1 and R = Z[ 2 ]). I say it is useful because it comes with the
1+ d
(4)
presciption for how to calculate g, as the gcd of three integers.
What this all means for fractional ideals is that
1
g Ĩ furnishes an inverse to I.
This gives examples of non-principal ideals that have an (explicit!)

inverse. Later we will see that all nontrivial ideals in R are invertible.
Our discussion of fraction fields is not complete without men-
tioning one case where there is nothing to do, a result sometimes
called “Wedderburn’s little theorem”:
III.F.18. T HEOREM (Wedderburn). Let R be a commutative domain,
with | R| < ∞. Then R is a field.
P ROOF. Let r ∈ R\{0}. Since R is finite, there exists a power
n ∈ Z>0 such that r n ∈ {1, r, . . . , r n−1 }, say r n = r k . Then r k (r n−k −
1) = 0, and since R is a domain, we have r n−k = 1 and r is a unit. So
R\{0} = R∗ and R is a field.
142 III. RINGS
III.G. Polynomial rings
Throughout we shall assume that R, S denote commutative rings.

We defined polynomial rings over R in an indeterminate x (and in in-
dependent indeterminates x1 , . . . , xn ) in III.A.3(iv). From the induc-
tive construction there it is clear that (writing I = (i1 , . . . , in ) ∈ Nn
and x I := x1i1 · · · xnin )
(III.G.1) 0= ∑ a I x I ∈ R [ x1 , . . . , x n ] ⇐⇒ all a I = 0.
I
Write ı : R ,→ R[ x ] (or R[ x1 , . . . , xn ]).

III.G.2. T HEOREM . Given ϕ : R → S and u ∈ S, there exists a unique
homomorphism ϕ̃ : R[ x ] → S such that ϕ̃( x ) = u and ϕ̃ ◦ ı = ϕ. (More
generally, given u1 , . . . , un ∈ S, there exists a unique ϕ̃n : R[ x1 , . . . , xn ] →
S such that ϕ̃n ( xi ) = ui (∀i) and ϕ̃n ◦ ı = ϕ.)
P ROOF. Uniqueness follows from the fact that ϕ̃ [resp. ϕ̃n ] is
specified on generators of R[ x ], namely R and x [resp. x1 , . . . , xn ].
For existence of ϕ̃, define ϕ̃(∑k ak x k ) := ∑k ϕ( ak )uk . We have
ϕ̃(∑k ak x k ) ϕ̃(∑` b` x ` ) = ∑ (∑k+`=n ϕ(ak ) ϕ(b` )) un

n
= ∑ ϕ(∑k+`=n ak b` )un [since ϕ homom.]
n
= ϕ̃ (∑n (∑k+`=n ak b` ) x n )
= ϕ̃ (∑k ak x k )(∑` b` x ` ) ,

so ϕ̃ is a homomorphism (the other checks being trivial).

For existence of ϕ̃n , apply induction: at each stage, we extend
ϕ̃n−1 : R[ x1 , . . . , xn−1 ] → S to ϕ̃n : R[ x1 , . . . , xn−1 ][ xn ] → S restricting
to ϕ̃n−1 and sending xn 7→ un .
III.G.3. D EFINITION . If S ⊃ R and ϕ is the inclusion, ϕ̃ [resp ϕ̃n ]
is denoted evu [resp. evu ], and the image by
evu ( R[ x ]) =: R[u]
[resp. evu ( R[ x1 , . . . , xn ]) =: R[u1 , . . . , un ])]. Note that this image con-

sists of polynomials in u [resp. the {ui }].
III.G. POLYNOMIAL RINGS 143
III.G.4. C OROLLARY. Writing Iu := ker(evu ), we have
R[u] ∼
= R[ x ]/Iu
and Iu ∩ R = {0} (and the obvious analogues for u).
P ROOF. Use the Fundamental Theorem together with injectivity

of evu | R (= ϕ).
III.G.5. C OROLLARY. Given σ ∈ Sn , there exists a unique automor-

phism ζ (σ ) of R[ x1 , . . . , xn ] sending xi 7→ xσ(i) .
P ROOF. Put S := R[ x1 , . . . , xn ], ui := xσ(i) , and ζ (σ) := ϕ̃n . An

inverse is provided by ζ (σ−1 ).
III.G.6. D EFINITION . As in III.G.3, let u or u1 , . . . , un be elements

of a ring S containing R.
(i) u is transcendental over R ⇐⇒ evu is injective.
(ii) Otherwise, u is algebraic over R. In this case there exists f ( x ) ∈
Iu \{0}, so that f (u) = 0 in S. (That is, u satisfies a polynomial equa-
tion with coefficients in R.)
(iii) u1 , . . . , un are algebraically independent over R ⇐⇒ evu is
injective; otherwise, they are algebraically dependent.
As a consequence of (III.G.1), u1 , . . . , un are algebraically inde-

pendent if, and only if,
(III.G.7) ∑ rI uI = 0 =⇒ all r I = 0.
I
On the other hand, if R = F and S are fields,19 and each ui algebraic

over F, then F[u1 , . . . , un ] is called an algebraic extension20 of F.
III.G.8. P ROPOSITION . An algebraic extension (of a field F) is a field.

Moreover, every element of this field is algebraic over F.
19The argument below works for S a domain. We will give a “higher-level” ap-
proach to III.G.8 when we study PIDs.
20This is a provisional (somewhat nonstandard) definition. The (standard) termi-
nology algebraic field extension, used later in these notes, refers to something more
general: a field containing F, all of whose elements are algebraic over F. (This
need not be generated by a finite number of elements.)
144 III. RINGS
P ROOF. We only have to prove this for F[u], u algebraic (since in-
duction then yields it for F[u1 , . . . , un ]). Let f ( x ) = ∑nk=0 ak x k ∈ F[ x ]
be a (nonzero) polynomial of minimal degree with f (u) = 0. (Note that
this degree is n.) Since S has no zero-divisors, f ( x ) is irreducible. In
particular, a0 6= 0 and (rescaling) we may assume a0 = 1. Then
(− ∑nk=1 ak uk−1 ) · u = 1 shows that u is invertible in F[u].
Now let v ∈ F[u] be arbitrary. If there exists some polynomial
g( x ) = ∑k bk x k ∈ F[ x ] with g(v) = 0 in S, then the same argument
(taking g of minimal degree, b0 = 1, etc.) produces an inverse for v
in F[u], namely − ∑k>0 bk vk−1 . So this will prove both statements of
the Proposition.
Notice that F[u] is a vector space over F of dimension n. Indeed,
−1 a k k
using f (u) = 0 ( =⇒ un = − ∑nk= 0 an u ) we can reduce the degree
of any polynomial in u (i.e. element of F[u]) to ≤ n − 1. Moreover, if
−1 n −1 0 k
∑nk= 0 ck u = ∑k=0 ck u ∈ F[ u ] then ck = ck : otherwise the difference
k 0
of the two sides gives a polynomial of degree < n with u as a root,

contradicting minimality of n.
So to find the desired polynomial g, consider the linear transfor-
mation µv : F[u] → F[u] given by multiplication by v. (This is calcu-
lated in the basis 1, u, . . . , un−1 by using f (u) = 0.) Taking g to be the
characteristic polynomial of µv , by Cayley-Hamilton 0 = g(µv ) =
µ g(v) . As S hence F[u] has no zero-divisors, g(v) is itself zero.
III.G.9. E XAMPLE . An algebraic extension F of Q is called a num-

ber field. By III.G.8, every α ∈ F has f ( x ) ∈ Q[ x ] such that f (α) = 0.
The ring of integers O F ⊂ F comprises those α with an f of the form
(III.G.10) x m + a m −1 x m −1 + · · · + a 0 , a j ∈ Z.
(Such a polynomial, with top coefficient 1, is called monic.) Check-

ing directly that O F is a ring is too messy. We postpone that to when
we have the tools for a better approach, which will show in addi-
tion that the characteristic polynomial of multiplication by α ∈ O F
(as in the above proof) is itself monic integral. Since that polynomial
has degree n := dimQ ( F ) (from the proof), we only need to consider

equations (III.G.10) with m = n.
√
Consider F = Q[ d] ∼ = Q[ x ]/( x2 − d). What is O F ? (We assume
d squarefree, so that d 6≡ 0.)
(4)
√
Since the above “n” is just 2 in this case, an element a + b d
(a, b ∈ Q) of F belongs to O F if and only if it satisfies
√ √
0 = ( a + b d)2 + m( a + b d) + n for some m, n ∈ Z.
√
Then 0 = ( a2 + b2 d + ma + n) + (2ab + mb) d, and so either
(i) b = 0 and a2 + ma + n = 0 ( =⇒ a ∈ Z)
or
(ii) −2a = m ( =⇒ a = A2 , A ∈ Z) and
2
b2 = − A +2mA
4d
+4n
( =⇒ b = B2 , B ∈ Z).
A2 + B2 d+2mA
In case (ii), 4 (= −n) ∈ Z =⇒ A2 + B2 d + 2mA ≡ 0.
(4)
Thus:
• if A is even, then B2 d ≡ 0 (and d 6≡ 0) hence B is even; while
(4) (4)
• if A is odd, then m is odd and (noting 33 , 12 ≡ 1)
(4)
1 + B2 d + 2 ≡ 0 =⇒ B2 d ≡ 1 =⇒ B odd and d ≡ 1.
(4) (4) (4)
This gives the “⊆” half of

 √
Z [ 1+ d ], d ≡ 1
2
(III.G.11) OF = √ (4)
Z[ d ], otherwise.
The reverse inclusion “⊇” is more straightforward: given α = a +
√ √
b d on the RHS, consider ( x − α)( x − α̃), where α̃ = a − b d as
usual.
Polynomial division. Earlier we made assertions about polyno-

mial division in F[ x ], F a field. Now it is time to be more precise.
Given f ( x ) = ∑dj=0 a j x j with a j ∈ R (an arbitrary commutative ring)
and ad 6= 0, write deg( f ) := d. We set deg(0) := −∞. Then
(III.G.12)
deg( f g) ≤ deg( f ) + deg( g) (with equality if R is a domain)
146 III. RINGS
and
(III.G.13) deg( f + g) ≤ max (deg( f ), deg( g)) .
III.G.14. P ROPOSITION . R domain =⇒ R[ x1 , . . . , xn ] domain and

R [ x1 , . . . , x n ] ∗ = R ∗ .
P ROOF. For n = 1, f g = 0 =⇒ deg( f ) + deg( g) = deg( f g) =

−∞ =⇒ f or g = 0; while f g = 1 =⇒ deg( f ) + deg( g) = 0 =⇒
deg( f ) = 0 = deg( g) =⇒ f , g ∈ R∗ . For n > 1, use induction.
For R not a domain, we need not have R[ x ]∗ equal to R∗ : e.g. in
Z9 [ x ], (1 + 3x )(1 − 3x ) = 1.
Now let R be any commutative ring, and
n m
f = ∑ ai x i , g = ∑ bj x j ∈ R [ x ].
i =0 j =0
III.G.15. T HEOREM (Polynomial long division). There exist k ∈ N

and q, r ∈ R[ x ] such that deg(r ) < deg( g) and (bm )k f = qg + r. If
bm ∈ R∗ then we have f = qg + r, and q, r are unique.
P ROOF. Assume (n =) deg( f ) ≥ deg( g) (= m) (since otherwise

we’re done). Writing21
f 1 := bm f − an x n−m g (noting n1 := deg( f 1 ) < deg( f ))

| {z }
p1
(1)
f 2 : = bm f 1 − a n 1 x n 1 − m g = : ( bm ) 2 f − p 2 g
..
.
we eventually
reach
k
r : = f k : = bm f − pk g of degree < deg( g)
For the uniqueness statement, we are assuming bm ∈ R∗ . If q1 g +

r1 = q2 g + r2 , then deg((q1 − q2 ) g) = deg(r2 − r1 ) < m. If q1 − q2 6=
0, then (since bm is not a zero-divisor) deg((q1 − q2 ) g) ≥ m yields a
contradiction. So q1 = q2 , and thus r1 = r2 .
21Note: a( j) denote coefficients of f .
k j
III.G.16. C OROLLARY. Given f ∈ R[ x ] and a ∈ R, there exist unique

q, r ∈ R[ x ] such that f ( x ) = ( x − a)q( x ) + f ( a). Hence, ( x − a) | f ( x )
⇐⇒ f ( a) = 0. (Such an “a” is called a root of f .)
All of this is for a general commutative ring. More narrowly:
III.G.17. C OROLLARY. If R is a domain, then a polynomial f ∈ R[ x ]
of degree n := deg( f ) has at most n roots.
P ROOF. Let a1 , . . . , ar be distinct roots of f . We have ( x − a1 ) | f
by III.G.16. Assume inductively ( x − a1 ) · · · ( x − ak−1 ) | f . Then
f ( x ) = ( x − a 1 ) · · · ( x − a k −1 ) g ( x )
=⇒ 0 = f ( ak ) = ( ak − a1 ) · · · ( ak − ak−1 ) g( ak )
| {z }
6 =0
=⇒ 0 = g( ak ) (since R is a domain)
=⇒ g( x ) = ( x − ak )h( x ) (for some h ∈ R[ x ])
=⇒ ( x − a1 ) · · · ( x − ak ) | f .
So in fact, f ( x ) = H ( x ) ∏rj=1 ( x − ai ) (for some H ∈ R[ x ]) hence
n ≥ r.
What if R is not a domain? Consider, say, polynomials over Z6 :
f ( x ) = 3x has 0̄, 2̄, and 4̄ as roots. So III.G.17 fails.
Turning to the case where R is a field, we have the famous
III.G.18. T HEOREM . The multiplicative group of a finite field is cyclic.
More generally, any finite subgroup G of the multiplicative group of a field
F is cyclic.
P ROOF. Recall from II.D.15 that since G is abelian, G is cyclic
⇐⇒ exp( G ) = | G |. This was based on the fact that there exists
an element of order exp( G ) := min{e ∈ N | ge = 1 (∀ g ∈ G )}. In
general, exp( G ) ≤ | G | since g|G| = 1 for all g ∈ G.
Now every g ∈ G satisfies gexp(G) − 1 = 0. But III.G.17 =⇒
xexp(G) − 1 has at most exp( G ) roots. So | G | ≤ exp( G ).
∗ ∼ Z , and not Z×4 , Z × Z ,
III.G.19. E XAMPLE . This says Z17 = 16 2 8 2
etc. — this beats trying to find a generator!
148 III. RINGS
III.G.20. R EMARK . Assuming the structure theorem for finitely

generated abelian groups,22 we can give a different proof of III.G.18
as follows. The structure theorem tells us that G ∼ = Zm1 × · · · × Zm k
where m1 > 1 and m1 | m2 | · · · | mk . So every g ∈ G is a root23 of
x mk − 1, hence | G | ≤ mk (by III.G.17), whence k = 1.
As we shall see later,24 there exist finite fields of prime power

order (for any prime power).
III.G.21. C OROLLARY. If F is a finite field, then F ∼

= Z p [u] where Z p
is its prime subfield and u is algebraic over Z p .
P ROOF. Let u be a generator of the multiplicative group F∗ =

F\{0}.
Polynomial functions. Let F be a field, Fn := F × · · · F (n times).

Consider a different kind of evaluation map:
(III.G.22)
!
Fn ring of F-valued
Φn,F : F[ x1 . . . . , xn ] −→ F = ∏n F =:
functions over Fn
n ∈F
f ( x ) 7−→ { f (u)}u∈Fn
The image Φn,F (F[ x1 , . . . , xn ]) =: Pn (F) is called the ring of (F-valued)
polynomial functions over Fn . We write si for Φn,F ( xi ), the ith coor-
dinate function, and clearly Pn (F) = F[s1 , . . . , sn ]. Two questions
arise:
• Are all functions polynomial functions? (i.e. is Φn,F surjective?)
• Do distinct polynomials yield distinct functions? (i.e. is Φn,F in-
jective? Note that this would imply that Pn (F) ∼= F[ x1 , . . . , xn ].)
We can give a surprisingly clear answer to both questions with the
aid of the following
22This will be discussed and proved in the context of modules where it belongs.
23Note that the group operation is being written multiplicatively, because G is a
multplicative group inside a field. In “additive” terms, gmk − 1 = 0 reads mk g = 0.
24Obviously Z n isn’t a field, so that won’t cut it!
p
III.G.23. L EMMA . Assume |F| = ∞. Then for each f ∈ F[ x1 , . . . , xn ]

other than the zero polynomial, there exists u ∈ Fn with f (u) 6= 0.
P ROOF. For n = 1: any f ∈ F[ x ] has at most deg( f ) (< ∞) roots,

so Φn,F ( f ) 6= 0. Next, assuming the result for n − 1 indeterminates,
let f n ∈ F[ x1 , . . . , xn−1 ][ xn ]. Writing f n = g0 + g1 xn + · · · gd xnd , let
u0 ∈ Fn−1 be such that gd (u0 ) 6= 0. Then f n (u0 , xn ) is a nontrivial
polynomial in xn , and we get un ∈ F such that f n (u0 , un ) 6= 0.
III.G.24. T HEOREM . Φn,F is injective ⇐⇒ |F| = ∞.
P ROOF. If |F| = q < ∞, then |F∗ | = q − 1 and so αq−1 = 1 =⇒

q
αq = α (∀α ∈ F) =⇒ x1 − x1 ∈ ker(Φn,F ).
If |F| = ∞, the lemma implies that no nonzero f ∈ F[ x1 , . . . , xn ]
is sent to the zero function.
III.G.25. T HEOREM . If |F| < ∞, then Φn,F is surjective.
P ROOF. The proof of III.G.23 shows that when degxi ( f ) < q :=

|F| for all i, there exists u ∈ Fn such that f (u) 6= 0. This is because
at each stage of the induction, the number of roots of f n in xn is less
than the number of elements of F.
q
On the other hand, the functions xi − xi in the proof of III.G.24
belong to ker(Φn,F ). By the division algorithm, for every k ≥ q we
q
get xik = ( xi − xi ) Q( xi ) + R( xi ) with deg( R) < q, and so any f ∈
F[ x1 , . . . , xn ] is of the form
n
∑ gi (x)(xi − xi ) + g(x),
q
with degxi ( g) < q (∀i ).
i =1
Hence f ∈ ker(Φn,F ) ⇐⇒ g( x ) = 0, which yields

q q
(III.G.26) Pn ( F ) ∼
= F[ x1 , . . . , x n ] / ( x1 − x1 , . . . , x n − x n ).
But |FF | = qq , and
n n
q −1 n
|Pn ( F )| = #{choices for g( x ) = ∑i1 ,...,in =0 a I x I } = qq
as well.
150 III. RINGS
Symmetric polynomials. Looking back at III.G.5, the automor-

phisms ζ (σ) of F[ x1 , . . . , xn ] produce a group homomorphism
ζ : Sn → Aut(F[ x1 , . . . , xn ]).
We will write F[ x1 , . . . , xn ]Sn for the subring of ζ (Sn )-invariant ele-

ments, i.e. the symmetric polynomials. Also note that a polynomial
is called homogeneous if all its monomial terms have the same total
degree (= sum of exponents).
III.G.27. D EFINITION . (i) The elementary symmetric polynomi-

25
als are
e1 ( x ) = ∑ xi , e2 ( x ) = ∑ xi x j , . . . , e n ( x ) = x1 . . . x n .
i i< j
(ii) The Newton symmetric polynomials are
s1 ( x ) = ∑ xi , s2 ( x ) = ∑ xi2, . . . , sn ( x ) = ∑ xin .
i i i
Both sets belong to F[ x1 , . . . , xn ]Sn , which is easiest to see for the

{ei } by writing formally
n n
(III.G.28) ∏ ( y − xi ) = ∑ (−1) j e j (x)yn− j .
i =1 j =0
We shall prove below that the ei “span” F[ x1 , . . . , xn ]Sn . (More pre-

cisely, III.G.29 means that there is one and only one way to write
each symmetric polynomial in the form ∑ D∈Nn a D e D , where e D :=
e1 ( x )d1 · · · en ( x )dn .) As you will show in HW, the si also “span the
symmetric polynomials” if n! 6= 0 in F.
Consider the ring homomorphism
En : F[ x1 , . . . , xn ] −→ F[ x1 , . . . , xn ]Sn
xi 7−→ ei ( x )
with image F[e1 , . . . , en ].
III.G.29. T HEOREM . En is an isomorphism.

25Note that e ( x ) has (n) monomial terms.
k k
P ROOF. We begin with surjectivity. Since every symmetric poly-

nomial is a sum of homogeneous symmetric polynomials, it suffices
to prove that every homogeneous symmetric polynomial is a poly-
nomial in the {ei }.
Under the lexicographic ordering on monomials, let aK x1k1 · · · xnkn
be the highest-order term in some given symmetric f ; since f con-
tains all permutations of each monomial, we have k1 ≥ k2 ≥ · · · ≥
k n . The highest monomial in e1k1 −k2 e2k2 −k3 · · · enkn is
( x1 )k1 −k2 ( x1 x2 )k2 −k3 ( x1 x2 x3 )k3 −k4 · · · ( x1 · · · xn )kn = x1k1 x2k2 · · · xnkn .
Hence f − aK e1k1 −k2 · · · enkn has lower highest monomial than f , and
continuing on in this manner we eventually reach the zero polyno-
mial.
Turning to injectivity, consider a finite sum ∑ D a D e D (with not all
a D zero). For each D ∈ Nn , write (for i = 1, . . . , n) k i = di + · · · + dn ,
and consider those (nonzero) a D e D with largest |K | := ∑i k i . The
highest monomial in each is a D x1k1 · · · xnkn , and these are all distinct
(D 6= D 0 =⇒ K 6= K 0 ). Taking the (unique) a D e D with “highest
highest” monomial, we see that this monomial occurs once, with a
nonzero coefficient. Hence ∑ D a D e D 6= 0.
152 III. RINGS
III.H. Principal ideal domains
Let R be a commutative domain.
III.H.1. D EFINITION . R is a principal ideal domain (PID) if every

ideal I ⊆ R is principal.
Regardless of whether R is a PID, note that we have

(
r | s ⇐⇒ (r ) ⊇ (s)
(III.H.2)
r ∼ s ⇐⇒ (r ) = (s)
for r, s ∈ R.
III.H.3. E XAMPLES (of PIDs).

(A) R = Z (consider the additive subgroups).
(B) Euclidean domains (which of course √ includes (A)).
1+ −11
(C) F[ x ] (F any field), Z[i], and Z[ 2 ] (HW) are Euclidean, hence
PIDs by (B).√
(D) Z[ 1+ 2−19 ], while non-Euclidean (HW), is a PID.
P ROOF OF (B). Given I ⊆ R an ideal in a Euclidean domain R, let

β ∈ I \{0} be of minimal δ( β) (∈ N), and take α ∈ I to be arbitrary.
Then
α = βq + r (q, r ∈ R)
with (i) δ(r ) < δ( β) and r = α − βq ∈ I \{0}, or (ii) r = 0. Since (i)
contradicts minimality of δ( β), we have (ii) and α = βq ∈ ( β) =⇒
I ⊆ ( β). Since β ∈ I, we have ( β) ⊆ I; thus I = ( β) is principal.
√
P ROOF OF (D). Write α := 1+ 2−19 and R := Z[α]. Let I be any
nonzero ideal of R, and take x ∈ I \{0} of minimal norm x x̃ = | x |2
(i.e. minimal | x |). We will show that I = xR (= ( x )). Equivalently,
working in the field of fractions K = Q[α], we can try to show that
the fractional ideal J := x −1 I is R. (Clearly, from I ⊃ xR we have
J ⊃ R.)
Step 1 Any element γ ∈ J \ R has imaginary part =(γ) differing from any
√ √ √ √ √ √
3 3 19− 3
2 Z.
19 19
integral multiple of 2 by at least 2 , i.e. =(γ) ∈ [ 2 , 2 ] +
III.H. PRINCIPAL IDEAL DOMAINS 153
Given γ ∈ J, suppose |γ − r | < 1 for some r ∈ R. Since γ = x −1 r0

for some r0 ∈ I, we have 1 > | x −1 r0 − r | =⇒ | x | > |r0 − rx |. Since
| x | is minimal, r0 − rx ∈
/ I \{0}. But r0 − rx ∈ I as r0 , x ∈ I. So the
only possibility is for r0 − rx to be 0, i.e. γ = x −1 r0 = r ∈ R.
Conclude that any γ ∈ J \ R has |γ − r | ≥ 1 (∀r ∈ R). Represent-
ing elements of R in the complex plane by red dots, the following
picture explains why the above claim holds:
imaginary axis
√ 1
2
−ᾱ 19
2
α
√
3
2
√ √
19− 3
2
√
3
2
−2 − 23 −1 − 12 1
2 1 3
2 real axis
since being outside the

√
circles forces γ inside the union of translates
19
of the grey strip by 2 iZ. In fact, since we can translate (in J \ R) by
elements√of R,√
this
√
shows: if J \ R 6= ∅, then there exists γ ∈ J \ R with
=(γ) ∈ [ 23 , 192− 3 ] and <(γ) ∈ (− 12 , 21 ].
Step 2 For such a γ, we have γ = α2 or − ᾱ2 .
√ √ √
We have =(2γ) ∈ [ 3, 19 − 3] and <(2γ) ∈ (−1, 1]. In particu-
lar, <(2γ) is within 12 of either 12 or − 21 . Accordingly, either |2γ − α|2
or |2γ + ᾱ|2 is
√ √ √
≤ ( 219 − 3)2 + ( 21 )2 = 8 − 57 < 8 − 7 = 1,
154 III. RINGS
i.e. 2γ is within 1 of α or −ᾱ — hence cannot be in J \ R by Step 1.

Conclude that 2γ ∈ R. But the only elements of R in the rectangle
to which 2γ is confined are α, −ᾱ. Hence γ = α2 or − ᾱ2 .
Step 3 J does not contain either of these elements.
α
Since J is closed under multiplication by elements of R, if γ = 2 or
− ᾱ2 , then α2ᾱ ∈ J. But
√ √
1+ −19 1− −19
αᾱ 2 · 2 5
1+19
2 = 2 = = ,8
2
which is within 1 of an element (say, 3) of R so cannot be in J \ R. On
/ R. So 52 ∈
the other hand, 25 ∈ / J, a contradiction.
Thus there exists no γ ∈ J \ R; that is, J = R. Hence I = ( x ) is
principal as desired.
III.H.4. E XAMPLES (of non-PIDs).

√
(A) Z[ 10] is not a PID.
√
P ROOF. Writing I := (3, 1 + 10), Hurwitz gives
√ √
I Ĩ = (3, 1 + 10)(3, 1 − 10) = (gcd(9, −9, 6)) = (3).
√ √
Suppose I = ( β) for some β = a + b 10 ∈ Z[ 10]. Then (3) = I Ĩ =
( β β̃) = ( a2 − 10b2 ) =⇒ a2 − 10b2 ∼ 3. Since Z∗ = {±1}, we get
a2 − 10b2 = ±3, which by a recent HW problem is impossible.
(B) R[ x ], where R is a PID, need not be a PID. In particular, F[ x, y]

(for F a field) is not.
P ROOF. Consider the proper ideal I := {∑i,j aij xi y j | a00 = 0} =

( x, y) in F[ x, y]. If I = ( f ) then f | x, y.
Now I claim that x is irreducible. To show this, suppose x = gh.
Since F[y] is a domain, the degrees (as polynomials in x over F[y])
satisfy degx g + degx h = 1. Swapping g and h if needed, we have
degx g = 0 and degx h = 1 hence g ∈ F∗ = (F[ x, y])∗ is a unit.
Likewise, y is irreducible.
So f = ax or a, for a ∈ (F[ x, y])∗ = F∗ ; and f = by or b, with

b ∈ (F[ x, y])∗ = F∗ . Obviously then f ∈ F∗ , which gives I = R[ x, y],
a contradiction. We conclude that I is not principal.
(C) Z[ x ] is not a PID: consider I = (3, x3 − x2 + 2x − 1) (HW) or,

more simply, I = (3, x ).
√ √
(D) Two more non-examples are (i) Z[ −17] and (ii) Z[ 1+ 2−23 ]. I
won’t prove this, but rather just say where the argument in the proof
of III.H.3(D) goes wrong: for (i), the bounding argument in Step 2 —
i.e. getting |2γ − α| or |2γ + ᾱ| < 1 — fails because (viewed as a lat-
tice) R is now too “spread out” vertically. For (ii), Step 2 still works,
but α2ᾱ = 238+1 = 3 belongs to R hence fails to yield a contradiction.
We now turn to some remarks on principal ideals generated by

irreducible elements. To begin, let R be a commutative domain, and
α ∈ R\{0}. Notice that in general
/ R∗ =⇒ α - 1
α irreducible =⇒ α ∈
(III.H.5)
=⇒ (α) 63 1 =⇒ (α) ∈ P P ,
where “P P ” denotes the set of proper principal ideals of R.
III.H.6. T HEOREM . α is irreducible ⇐⇒ (α) is maximal in25 P P .
P ROOF. ( =⇒ ): Suppose ( β) ∈ P P and ( β) ⊇ (α). Then β ∈ / R∗

and α = βr (for some r ∈ R). Since α is irreducible, r must belong to
R∗ . So (α) = ( β).
( ⇐= ): Let (α) be maximal in P P , and write α = βγ, with β ∈ /
∗
R . Then ( β) ∈ P P and ( β) ⊇ (α). By maximality of (α) in P P ,
( β) = (α) hence we can write β = αδ. This gives α = αδγ =⇒
δγ = 1 =⇒ γ ∈ R∗ . Thus α is irreducible.
In general, for a principal ideal (α), “maximality in P P ” is quite

a bit weaker than “maximality”. Of course, when R is a PID these
are equivalent, and we get the
25The RHS contains two assertions: (α) ∈ P P , and (α) is maximal there.
156 III. RINGS
III.H.7. C OROLLARY. Let R be a PID, α ∈ R\{0}. Then26
α is irreducible ⇐⇒ (α) is maximal amongst proper ideals.
III.H.8. C OROLLARY. Let R be a PID, α ∈ R\( R∗ ∪ {0}). Then:

(i) R/(α) is a field ⇐⇒ α is irreducible; and
(ii) otherwise, R/(α) isn’t a domain.
P ROOF. (i) Follows at once from III.H.7 and III.F.3.

(ii) If α is not irreducible, then there exist β, γ ∈ R\( R∗ ∪ {0}) such
that α = βγ. Suppose β ∈ (α); then β = αr (r ∈ R) =⇒ α = αrγ
=⇒ rγ = 1 =⇒ γ ∈ R∗ , a contradiction.
So β, γ ∈/ (α) =⇒ β̄, γ̄ 6= 0̄ in R/(α) but β̄γ̄ = ᾱ = 0̄.
Now let F be a field and S ⊃ F a ring, with u ∈ S. Recall from

III.G.3 the evaluation map evu : F[ x ] → S sending x 7→ u, with image
=: F[u], and kernel =: Iu . Since F[ x ] is a PID, Iu = ( g) for some
g ∈ F[ x ], and Iu ∩ F = {0} =⇒ g ∈ / F∗ (= F[ x ]∗ ). If g = 0, then
u is transcendental over F; otherwise, deg( g) > 0 and u is algebraic
over F.
Henceforth assume that u is algebraic; then as F is a field, we
may also assume that g is monic. In fact, since any two generators of
Iu are associate, this uniquely determines g.
III.H.9. D EFINITION . The (unique) monic generator mu of Iu is

called the minimal polynomial of u over F.
III.H.10. L EMMA . This mu is the lowest-degree polynomial in F[ x ]\{0}

having u as a root.
P ROOF. f (u) = 0 =⇒ f ∈ Iu = (mu ) =⇒ f = mu q =⇒

deg( f ) ≥ deg(mu ) or f = 0.
III.H.11. C OROLLARY. F[u] is a field ⇐⇒ mu is irreducible in F[ x ].

Otherwise, F[u] is not a domain.
P ROOF. Immediate from F[u] ∼

= F[ x ]/Iu and III.H.8.
26The RHS here means that (α) is a maximal ideal (in the standard sense).
III.H.12. R EMARK . The following construction of F[u] appears

tautological but is actually the most useful one. Let g( x ) ∈ F[ x ]
be a monic polynomial of positive degree; we put S := F[ x ]/( g( x ))
and u := x + ( g( x )) ∈ S. Then the evaluation map evu : F[ x ]
F[ x ]/( g( x )) is just the natural map, with kernel Iu = ( g( x )). Hence
F[u] = F[ x ]/( g( x )), and g( x ) = mu ( x ). The construction yields a
field if and only if g( x ) is irreducible. Regardless of that, every ele-
−1
ment of F[u] can be written in exactly one way as a sum ∑dk= 0 ak u
k
with ak ∈ F and d := deg( g). This makes it a vector space over F of

dimension d with basis 1, u, . . . , ud−1 .
This construction is important, for instance, when studying num-
ber fields. Let F = Q. Rather than starting with u ∈ S = C and
sending x to that, we start with an irreducible polynomial and need
never make any reference to C. So to take one example, we can define
√
Q[ −3] := Q[ x ]/( x2 + 3). This is both practically superior (when
studying polynomials for which we don’t know an “explicit” root)
and theoretically superior (as we don’t have to invoke the funda-
mental theorem of algebra). In Q[u] := Q[ x ]/( g( x )) one still thinks
of u as an abstract root of g. If desired, we can map Q[u] into C in
multiple ways by sending u to any root of g in C.
III.H.13. E XAMPLE . I claim that F = Q[θ ] := Q[ x ]/( x3 − x + 2) is

a field. Suppose otherwise; then g := x3 − x + 2 is reducible, which
means it has a linear and quadratic factor. The linear factor obviously
would have a root Q P
∈ Q (written in lowest terms). But
P3
Q3
− P
Q +2 = 0 =⇒ Q=1 =⇒ P3 − P + 2 = 0, P ∈ Z;
and reducing mod 5, multiplying by P̄ and using P̄4 = 1̄ gives 1̄ −

P̄2 + 2̄ P̄ = 0̄ =⇒ P̄( P̄ − 2̄) = 1̄. Since 1̄−1 = 1̄, 2̄−1 = 3̄, 3̄−1 = 2̄,
and 4̄−1 = 4̄, this is impossible.
As the polynomial has degree 3, F is a vector space over Q of
dimension 3, with basis 1, θ, θ 2 . The field F is called a cubic field.
158 III. RINGS
III.I. Unique factorization domains
Let R be a commutative domain, and α ∈ R\{0}. We recall (cf.

III.H.6) that
(III.I.1) α is irreducible ⇐⇒ (α) is maximal in P P ( R).

We are interested in
(a) when r ∈ R\{0} can be expressed as a product of irreducibles,
and
(b) when (and in what sense) such a factorization is unique.
III.I.2. D EFINITION . R satisfies the ascending chain condition for

principal ideals (ACCPI) iff for each chain I1 ⊆ I2 ⊆ · · · in P P ( R),
there exists n ∈ N such that Im = In for all m ≥ n.
III.I.3. R EMARK . If Ik = ( ak ), this says that
· · · | a3 | a2 | a1 =⇒ ∃n ∈ N such that am ∼ an (∀m ≥ n).

That is, there are no infinite sequences { ai } ⊆ R where each ai+1 is a
proper factor of ai (ai+1 | ai but ai - ai+1 ). In this form, the ACCPI is
known as the divisor chain condition (DCC), which is the terminol-
ogy I’ll use for both.
III.I.4. L EMMA . DCC holds =⇒ every I ∈ P P ( R) is contained in a

maximal element.
P ROOF. ( a) ∈ P P ( R) =⇒ ( a) maximal or ( a) ( ( a0 ). Rinse and

repeat; DCC implies this terminates.
III.I.5. T HEOREM . DCC holds =⇒ any r ∈ R\( R∗ ∪ {0}) is a finite

product of irreducibles.
P ROOF. Clearly (r ) ∈ P P ( R). Assume r is not itself irreducible.

Then (r ) is not maximal in P P ( R), so that III.I.4 gives a proper con-
tainment (r ) ( ( a1 ) in a maximal element ( a1 ) ∈ P P ( R); we thus
have r = a1 r1 , with r1 ∈ R\( R∗ ∪ {0}). If r1 is not irreducible, repeat
to get (r1 ) ( ( a2 ) maximal in P P ( R), which gives r1 = a2 r2 .
III.I. UNIQUE FACTORIZATION DOMAINS 159
Suppose this process doesn’t terminate. Then we obtain sequences

(
a1 , a2 , a3 , . . . of irreducible elements
r1 , r2 , r3 , . . . of elements of R\( R∗ ∪ {0})
such that r = a1 a2 · · · an rn (∀n). Hence rn = rn+1 an+1 , with an+1 ∈/
∗
R , so that (rn ) ( (rn+1 ) (∀n), a contradiction by the DCC.
Conclude that for some n, rn is irreducible, and r = a1 a2 · · · an rn
presents r as a product of irreducibles.
III.I.6. D EFINITION . (i) Let r ∈ R. Two factorizations
r1 · · · r m = r = s1 · · · s n
into irreducibles are essentially equivalent if
m = n and ∃ σ ∈ Sn such that si ∼ rσ(i) (i = 1, . . . , n).
(ii) R is a unique factorization domain (UFD) if

(
(a) every r ∈ R\( R∗ ∪ {0}) is a product of irreducibles, and
(b) this product is essentially unique.
(iii) Given a UFD R and r = r1 · · · rn ∈ R\( R∗ ∪ {0}) (with r1 , . . . , rn
irreducible), we define the length `(r ) to be n. (The length of a unit
is defined to be 0.) Clearly `(rs) = `(r ) + `(s) for all r, s ∈ R\{0}.
Continuing for the time being with a general commutative do-

main R, we have the
III.I.7. D EFINITION . An element a ∈ R\( R∗ ∪ {0}) is prime if
a | bc =⇒ a | b or a | c.
(Note that this is the same as saying that ( a) is a prime ideal.)
III.I.8. L EMMA . For a ∈ R\( R∗ ∪ {0}), a prime =⇒ a irreducible.
P ROOF. Given a ∈ R prime, suppose a = bc. Then a | b or a | c.

If a | b, we have b = ar (for some r ∈ R) =⇒ a = arc =⇒ rc = 1
=⇒ c ∈ R∗ . Likewise, if a | c, then b ∈ R∗ . So a is irreducible.
The converse does not hold in general:
160 III. RINGS
√
III.I.9. E XAMPLE . In Z[ 10], 3 is irreducible (by a norm argu-
ment, cf. III.D.6). But 3 is not prime:
√ √
3 | 9 = (1 + 10)(−1 + 10),
√ √
but 3 divides neither 1 + 10 nor −1 + 10.
One way to think of all this is that for a principal ideal ( a),
+3 +3
( a) maximal
(III.I.10) ( a) maximal ( a) prime
O in P P ( R)
O

a prime +3 a irreducible.
III.I.11. D EFINITION . R satisfies the primeness condition (PC) if

every irreducible element is also prime.
III.I.12. T HEOREM . Let R be a commutative domain. Then
R is a UFD ⇐⇒ R satisfies DCC and PC.
P ROOF. ( =⇒ ): Suppose given an ascending chain ( a1 ) ⊆ ( a2 ) ⊆

· · · in P P ( R); without loss of generality we may assume ( a1 ) 6= {0}.
Then `( a1 ), `( a2 ), . . . is a non-increasing27 sequence in N. So there
exists an n ∈ N such that (∀m ≥ n) `( am ) = `( an ) =: `. Now
( am ) ⊇ ( an ) =⇒ am | an =⇒ an = am r
and factoring into irreducibles gives
an,1 · · · an,` = am,1 · · · am,` (r1 , · · · r j u)
(where u ∈ R∗ and the rest are irreducible). By (essential) unique-

ness, j = 0 (i.e. r ∈ R∗ ) and after reordering an,i ∼ am,i =⇒ am ∼ an
=⇒ ( am ) = ( an ) (∀m ≥ n). So DCC holds.
Next, if r is irreducible and r | ab, write ab = rc. If a ∈ R∗
then r | b, and if b ∈ R∗ then r | a; otherwise, write a = a1 · · · ak ,
b = b1 · · · b` , c = c1 · · · cm (for factorizations into irreducibles), which
27e.g. factor both sides of a = a r into irreducibles to see `( a ) ≤ `( a ).
1 2 2 1
III.I. UNIQUE FACTORIZATION DOMAINS 161
gives a1 · · · ak b1 · · · b` = rc1 · · · cm . By (essential) uniqueness, r ∼

some ai or b j =⇒ r | a or b. So r is prime, i.e. PC holds.
( ⇐= ): Let r ∈ R\( R∗ ∪ {0}) be given. Since DCC holds, r is a
product of irreducibles by III.I.5. To check the (essential) uniqueness,
let µ(r ) denote the minimum number of irreducible factors in such a
product. If µ(r ) = 1, then r is irreducible, and can’t split as a product
of more than one, so clearly uniqueness holds.
Suppose we have uniqueness for all r with µ(r ) < M, and let
µ(r ) = M; write r = r1 · · · r M for a (minimal length) factorization
into irreducibles. By PC, the ri are prime. If r = s1 · · · s N is another
factorization into irreducibles, then r M | s1 · · · s N =⇒ r M | some s j ,
say s N . Since s N is irreducible (and r M ∈ / R∗ ), we get s N = r M u (for
some u ∈ R∗ ), i.e. r M ∼ s N . But now r 0 = (u−1 r1 )r2 · · · r M−1 has
µ(r 0 ) < M and r 0 = s1 · · · s N −1 . By induction, M − 1 = N − 1 and
(permuting factors if needed) s j ∼ r j (j = 1, . . . , M − 1) and we are
done.
In particular, in a UFD, prime and irreducible elements are the

same thing. So we get the following analogue of III.H.8:
III.I.13. C OROLLARY. Let R be a UFD, α ∈ R\( R∗ ∪ {0}). Then

R/(α) is a domain ⇐⇒ α is irreducible.
P ROOF. Combine III.F.6 with the fact that (α) is prime iff α is.
III.I.14. E XAMPLES .
(A) All PIDs (and hence all Euclidean domains) are UFDs.
(B) F[ x, y] and Z[ x ] are UFDs but (as we know) not PIDs.
(C) There is no number ring that is a UFD but not a PID.
We will prove (B) and (C) in §§III.K-III.L; for now, here is the
P ROOF OF (A). Consider an ascending chain I1 ⊆ I2 ⊆ · · · in

P P ( R), and consider the ideal J = ∪ j≥1 Ij ⊂ R. Since R is a PID,
J = ( a) for some a ∈ J. But then a ∈ In for some n, so J = ( a) ⊂ In
=⇒ Im = In for all m ≥ n. So DCC holds.
162 III. RINGS
Next suppose that a ∈ R is irreducible, and a | bc but a - b. Then

b∈/ ( a) =⇒ ( a, b) ) ( a). By (III.I.1), ( a) is maximal in P P ( R). Since
R is a PID, ( a, b) is principal. So ( a, b) = R. It follows that there exist
p, q ∈ R such that ap + bq = 1; multiplying by c gives apc + bcq = c.
Since a | bc, we therefore have a | c. Conclude that a is prime. So PC
holds, and III.I.12 finishes the job.
III.J. GREATEST COMMON DIVISORS 163
III.J. Greatest common divisors
III.J.1. D EFINITION . Let R be a commutative ring, and S ⊂ R a

nonempty subset. Then γ ∈ R is a GCD of S if
(
(i) γ | s (∀s ∈ S), and
(ii) δ | s (∀s ∈ S) =⇒ δ | γ.
If 1 is a GCD of S , then S is relatively prime.28
III.J.2. R EMARKS . (a) In terms of ideals: (i) S ⊂ (γ); and (ii) S ⊂

(δ) =⇒ (γ) ⊂ (δ). If S is relatively prime, then S (or (S)) is not
contained in a proper principal ideal.
(b) If γ, γ0 are two GCDs of S , then (a) =⇒ (γ) = (γ0 ) =⇒ γ ∼ γ0 .
That is, if a GCD exists, it is unique up to units.
(c) R PID =⇒ (S) = (γ) for some γ ∈ R, which is clearly a GCD
for S , and γ = s1 r1 + · · · + sn rn for some s j ∈ S , r j ∈ R.
(c’) Conversely to (c), if every S ⊂ R has a GCD of the form γ = ∑i si ri ,
then we have (S) ⊃ (γ) ⊃ (S) =⇒ (S) is principal. Since any
I = ( I ), R is then a PID. (As every UFD is not a PID, the italicized
property cannot hold for UFDs in general.)
(d) Dually, we have the notion of least common multiple (LCM): `
is a LCM of S if (i) s | ` (∀s ∈ S ) and (ii) s | κ (∀s ∈ S ) =⇒ ` | κ.
(e) For two elements: γ is a GCD of a, b ∈ R if: (i) γ | a and γ | b, and
(ii) δ | a, b =⇒ δ | γ.
√
Of course, a GCD need not always exist: in Z[ 10], let a = 3 +
√ √
3 10, b = 9. Then δ = 3 and δ0 = 1 + 10 both divide a and b
properly (i.e. the quotient is not a unit). Moreover, we have a - b,
b - a, δ - δ0 , and δ0 - δ. Were there a GCD γ of a and b, we’d have29
28As for ideals, a pair of relatively prime elements (|S| = 2) is said to be coprime.
29We write α k β for “α is a proper divisor of β”, which is to say that α | β and β
α
is not a unit. The reason we’d have (say) δ k γ here is that, were γ
δ a unit, δ0 | γ
would become δ0 | δ, which is false.
164 III. RINGS
δ0 , δ k γ k a, b =⇒
N (δ) = N (δ0 ) = 9 k N (γ) k 81 = N ( a) = N (b) =⇒

√
N (γ) = 27 =⇒ γ = c + d 10 with c2 − 10d2 = 27 =⇒ c2 ≡ 7, a
(10)
contradiction since 7 is not a square mod 10.
√
But Z[ 10] (as we know) is not a UFD, and in the UFD case the
situation changes:
III.J.3. T HEOREM . Any nonempty subset S of a UFD R has a GCD.

P ROOF. Write D := r ∈ R r | s (∀s ∈ S) ⊂ R for the set of all
divisors. Clearly 1 ∈ D =⇒ D 6= ∅. Recalling the length function
`(r ) (= # of irreducible factors in r) for a UFD, r ∈ D =⇒ `(r ) ≤ `(s)
(∀s ∈ S ) =⇒ ∃ γ ∈ D of maximal length `(γ).
Let a ∈ D be arbitrary. We claim that a | γ, which will establish
that γ is a GCD of S .
Write D 0 for the common divisors of γ and a. Arguing as above,
there exists c ∈ D 0 of maximal length `(c); and we may write a = cd,
γ = cδ. Clearly it is enough to show that d ∈ R∗ , since then c | γ =⇒
a | γ.
Suppose this is not so — i.e., that d ∈/ R∗ , with irreducible factor f .
Then `(c f ) = `(c) + 1, while c f | a. By maximality of `(c), we must
have c f - γ, hence f - δ.
Now for every s ∈ S , we have a, γ | s =⇒ c f | s = γξ = cδξ
=⇒ f | δξ. By III.I.12, since R is a UFD and f is irreducible, f is
prime. Since f - δ, it follows that f | ξ, hence γ f | s. Since s was
arbitrary, γ f ∈ D . But `(γ f ) = `(γ) + 1, contradicting maximality
of `(γ).
III.J.4. D EFINITION . R satisfies the GCD condition (GCDC) if ev-

ery pair a, b ∈ R has a GCD.
When the GCDC holds, we shall write gcd( a, b) (which is then

well-defined up to a unit).
III.J.5. R EMARKS . (i) Note that (by III.J.3) UFDs satisfy the GCDC;
and (by III.J.2(c)) for a PID we have ( a, b) = (gcd( a, b)).
(ii) The GCDC implies the existence of GCDs for all nonempty finite
subsets S ⊂ R. [P ROOF: given S = {s1 , . . . , sn }, inductively assume
that there exists a GCD γ0 for {s1 , . . . , sk−1 }. Then γ := gcd(γ0 , sk )
has γ | γ0 | s1 , . . . , sk−1 and γ | sk . Moreover, if γ0 | s1 , . . . , sk then
γ | s1 , . . . , sk−1 =⇒ γ0 | γ0 , which together with γ0 | sk yields γ0 | γ.]
So “gcd(S)” makes sense.30
(iii) If γ1 = gcd(S1 ), γ2 = gcd(S2 ) for two nonempty finite subsets,
the same argument gives gcd(S1 ∪ S2 ) = gcd(γ1 , γ2 ).
(iv) If γ = gcd(S) (for a finite subset S = {s1 , . . . , sn }) and r ∈ R,
then (writing r S := {rs1 , . . . , rsn }) we have rγ = gcd(r S).
III.J.6. P ROPOSITION . Let R be a commutative domain. Then GCDC

=⇒ PC (primeness condition).
P ROOF. Assume GCDC, and let π ∈ R be irreducible; we claim

that π is prime. First note that
(
π, if π | a
gcd(π, a) ∼
1, if π - a.
Let π | αβ and π - α. We must show π | β.
Suppose otherwise: π - β. Then (writing ( , ) for gcd( , ))
1 ∼ (π, α)(π, β) ∼ ((π, α)π, (π, α) β)

∼ (π , πα), (πβ, αβ) ∼ (π 2 , πα, πβ, αβ)
2
∼ (π (π, α, β), αβ) ∼ (π, αβ) ∼ π,

a contradiction since π is not a unit.
III.J.7. C OROLLARY. Let R be a commutative domain, with GCDC and

DCC. Then R is a UFD.
30The reader may wonder about infinite subsets, since their GCDs exist in III.J.3 for
UFDs. But the GCDC doesn’t imply R is a UFD, and can’t handle infinite subsets,
without also assuming the DCC. For example, if you are feeling adventurous, try
to show that if R ⊂ C is the ring of all algebraic integers (i.e. roots of monic
polynomials, which we √will show yield a ring later on), then the GCDC holds, but
S := {2q | q ∈ Q, q > 2} ⊂ R has no GCD.
166 III. RINGS
P ROOF. Combine III.J.6 and III.I.12.
This leads to a second proof that PIDs are UFDs, since the GCDC
obviously holds for PIDs (cf. III.J.2(c)).
III.J.8. R EMARK . In some of the remarks and computations above,

we have treated some aspects of GCDs in terms of ideals. Before
proceeding, we want to emphasize that when R is not a PID, some
caution is warranted.
Consider that we have two notions of coprimality for a, b in a
commutative ring R:
(i) The ideals ( a), (b) are coprime if ( a) + (b) (= ( a, b)) = R
(ii) The elements a, b are coprime if gcd( a, b) = 1.
Clearly (i) =⇒ (ii). But (ii) doesn’t imply (i) in a non-PID, e.g. in
the UFD F[ x, y], ( x ) and (y) are not coprime as ideals, but x and y are
coprime as elements.
So far we have said a lot about the theory of GCDs, and nothing
about effectively computing them (when they are not visibly obvi-
ous).
III.J.9. E UCLID ’ S A LGORITHM . In a PID, gcd(α, β) is the princi-

pal generator of (α, β), which gives a clue how to find it. If R is
Euclidean, this leads to a (very efficient) algorithm. Define qi and ri
recursively by
α = q1 β + r1 δ(r1 ) < δ( β) [or r1 = 0]

β = q2 r1 + r2 δ(r2 ) < δ(r1 ) [or r2 = 0]
r1 = q3 r2 + r3 δ(r3 ) < δ(r2 ) [or r3 = 0]
.. ..
. .
As δ doesn’t take negative values, eventually some rn+1 = 0 (where

rn 6= 0):
rn−1 = qn+1 rn + 0.
Now look at this in terms of ideals:
(α, β) = (q1 β + r1 , β) =
( β, r1 ) = (q2 r1 + r2 , r1 ) =
(r1 , r2 ) = ( q3 r2 + r3 , r2 ) =
(r2 , r3 ) = · · · =
( r n −1 , r n ) = ( q n +1 r n , r n ) = ( r n ).
This proves the
III.J.10. T HEOREM . For α, β in a Euclidean domain R, gcd(α, β) is the

last nonzero remainder in the Euclidean algorithm.
We now turn to a couple of applications of Euclid’s algorithm

and GCDs in Z.
Application 1: The RSA cryptosystem.
III.J.11. P ROPOSITION . Suppose k, k0 , m ∈ Z>1 , gcd( a, m) = 1, and

0
kk0 ≡ 1. Then akk ≡ a.
φ(m) (m)
P ROOF. Since a ∈ Z∗m , we have aφ(m) ≡ 1 by Euler’s theorem

(m)
0
II.D.9, and so akk = a· a Nφ(m) = a ( aφ(m) ) N ≡ a.
(m)
As k is invertible mod φ(m) provided they are coprime, we have
III.J.12. C OROLLARY. The map (·)k : Z∗m → Z∗m is an isomorphism if

0
gcd(k, φ(m)) = 1, and has inverse (·)k .
Say you want to be able to receive secure communications from

me over a public channel:
You Pick two large primes p, q, put m = pq. Then φ(m) = ( p −

1)(q − 1). Now
let k ∈ (0, φ(m)) ∩ Z be large,

(III.J.13)
with (k, φ(m)) = 1, and find k0 .
Make m, k public; and keep p, q, φ(m), k0 secret.

168 III. RINGS
Me I take a message, encode it as a single number a ∈ (0, m) ∩ Z,

and send you
b := āk ∈ Zm .
0
You Compute b̄k ∈ Zm , recovering (by III.J.11) my message a.
Suppose someone overhears m, k, b and wants to break the code to re-
cover a. They must find k0 , which requires knowing φ(m), for which
they will need to be able to factor m (into p and q). Unless they have
a quantum computer, this could take centuries.
As for us, how do we manage III.J.13? By using Euclid: first, to
check gcd(k, φ(m)) = 1; but less obviously, to solve the congruence
kk0 ≡ 1:
φ(m)
φ(m) = kq0 + r0 =⇒ r0 = φ(m) − kq0 ≡ −kq0

φ(m)
k = r0 q1 + r1 =⇒ r1 = k − r0 q1 ≡ k + kq0 q1 = k(1 + q0 q1 )
φ(m)
r0 = r1 q2 + r2 =⇒ r2 = r0 − r1 q2 ≡ −kq0 − k (1 + q0 q1 )q2
φ(m)
..
. = − k ( q0 + q2 + q0 q1 q2 )
Eventually, some rn = 1 and so the algorithm gives
1 ≡ k · (big mess).
φ(m)
The big mess is our k0 .
Application 2: Prime factorization in quadratic fields. Let p be

√
an odd prime number (∈ N), and K = Q[ d] a quadratic number
field (d squarefree). Below, ( p) will mean pOK , i.e. the ideal ( p) ⊂
OK . Denote by I(K ) the monoid of ideals31 in OK . An element
I ∈ I(K ) is irreducible if we cannot write I = I1 I2 , with both I1 , I2
proper in OK . We would like to factor ( p) in I(K ) as a product of
irreducibles.
31cf. Problem Set 7 #4. Here we take this to consist of all nonzero ideals.
We know that OK is often not a UFD, and that GCDs may not
exist. So we are not going to take them in OK . Rather, the connection
of this section to GCDs comes from Hurwitz’s theorem (cf. III.F.17).
Recall that given I = (α, β) ⊂ OK and Ĩ = (α̃, β̃), it says that
• αα̃, β β̃, and α β̃ + βα̃ belong to Z, and
• if g is their GCD in Z, then I Ĩ = ( g) = gOK .
It is the main tool in the proof of the following
III.J.14. T HEOREM . As an element of I(K ), the ideal ( p) ⊂ OK de-

composes into irreducibles as follows:
√
(i) d ≡ 0 =⇒ ( p) = ( p, d)2 =: ℘p 2 , and we say p ramifies.
( p)
√ √
(ii) d ≡ m2 6≡ 0 =⇒ ( p) = ( p, m − d)( p, m + d) =: ℘p ℘˜p (where
( p) ( p)
℘p 6= ℘˜p ), and we say p splits.
(iii) d 6≡ square =⇒ ( p) is irreducible in OK , and we say p is inert.

( p)
P ROOF. Introduce the ideal norm N : I(K ) → N\{0}, sending a

nonzero ideal I ⊂ OK to the unique generator in N of I Ĩ. (That is,
I Ĩ = (N( I )).) This is well-defined by Hurwitz, and is a multiplica-
tive monoid homomorphism. Moreover, it is useful for detecting
irreducibles: if N( I ) = 1, then
I Ĩ = (N( I )) = (1) = OK ⊆ I = I OK ⊆ I Ĩ
forces I = OK . So if N( I ) is prime, then I is irreducible in I(K ).

For (i), combining Hurwitz with the fact that d is squarefree and
√ √
divisible by p, we get ℘p 2 = ( p, d)( p, d) = (gcdZ ( p2 , 0, d)) = ( p)
=⇒ N( ℘p ) = p =⇒ pr irreducible.
For (ii), again by Hurwitz we have
√ √
℘p ℘˜p = ( p, m − d)( p, m + d) = (gcdZ ( p2 , 2pm, m2 − d))
= ( p · gcdZ ( p, 2m, n)) = ( p)
since p odd and m 6≡ 0 =⇒ p, 2m coprime. Again N( ℘p ) = p =
( p)
N( ℘˜p ), and so both ℘p and ℘˜p are irreducible.

170 III. RINGS
Finally, for (iii), begin by noting that N(( p)) = p2 , and suppose
that ( p) is not irreducible. Then there exists an ideal I of norm p with
I ) ( p) (as ( p) must break into two such). Assume the following
III.J.15. FACT. Every I ∈ I(K ) is generated by 2 elements of OK .
which will be proved in a moment. Then I = (α, β) =⇒ p =

gcdZ (αα̃, β β̃, α β̃
√
+ βα̃). Since I ) ( p), p cannot divide both α and β;
r +s d
say p - α = 2 (where r ≡ s).
(2)
r 2 − s2 d
On the other hand, p | αα̃ = 4 =⇒ r2 ≡ s2 d. If p | s
(4p)
then p | r and so (writing r = pr 0 , s = ps0 , with r 0 ≡ s0 ) we have
(2)
√ √
pr 0 + ps0 d r 0 +s0 d
α = 2 = p( 2 ) hence p|α, a contradiction. Therefore
p - s, and there exists an inverse s−1 ∈ Z p . We then find that
d ≡ (ss−1 )2 d ≡ s2 d(s−1 )2 ≡ r2 (s−1 )2 ≡ (rs−1 )2 ,

( p) ( p) ( p) ( p)
in contradiction to our hypothesis in (iii). Conclude that I cannot

exist, and ( p) is irreducible.
Here is a standard bit of notation attached to III.J.14.
III.J.16. D EFINITION . Define the Legendre symbol by



d  0 in case (i)

:= 1 in case (ii)
p 
−1 in case (iii)

It won’t be used until a later section.

We now prove Fact III.J.15 — actually a bit more. Recall:
• any subgroup K ≤ Zn is ∼ = Zm for some m ≤ n (cf. II.K.4); and
• any quadratic number field is of the form Q[ x ]/( x2 − d) (with el-
ements of the form q1 + q2 x) hence a Q-vector space of dimension
2. In fact, for a general number field F,32 we’ll show in the Galois
theory unit that F ∼ = Q[ x ]/(mu ) for some minimal polynomial mu
of degree n, so that dimQ F = n =: [ F:Q].
32We already saw this for number fields of the form Q[u] (cf. III.G.9 and its proof,
and III.H.13); the point here is that even those which appear to require multiple
generators really have just one.
III.J.17. P ROPOSITION .
(a) Let F be a number field, O F its ring of integers.33
(i) Every nonzero ideal I ⊂ O F contains a basis for F as a Q-vector
space, hence a subgroup ∼ = Z[ F:Q] .
(ii) Assuming that O F ∼ = Z[ F:Q] (and F ∼
= Q[ F:Q] ),34 we have that
I∼= Z[ F:Q] , with basis spanning F/Q.
√
(b) Let K = Q[ d], and I ⊂ OK be a nonzero ideal. Then as an additive
abelian group, I = hγ, δi, for some γ, δ ∈ OK ; and, moreover, I = (γ, δ).
P ROOF. (a) (i) If β 1 , . . . , β n (n = [ F:Q]) is a basis for F/Q, then I

claim that there exists b ∈ Z such that bβ i ∈ O F (∀i). To see this, note
that each β i satisfies some monic rational polynomial equation, as F
is algebraic over Q. Taking b to be the product of all denominators
of the coefficients of this equation, the bβ i will satisfy equations with
integer coefficients:
a1 d −1 a2 d −2 ad
βd + b1 β + b2 β +···+ bd =0
(bβ)d a1 ad
=⇒ bd
+ b1 bd−1
(bβ)d−1 + a2
b2 bd−2
(bβ)d−2 +···+ bd =0
2 d
=⇒ (bβ)d + a1 bb1 (bβ)d−1 + a2 bb2 (bβ)d−2 + · · · + ad bb = 0.
d
Next, taking any α ∈ I \{0}, each bβ i α ∈ I; and since (in F) mul-

tiplication by αb is invertible, the {bβ i α} cannot satisfy a nontrivial
Q-linear relation (without contradicting linear independence of the
{ β i }). So I contains a Zn .
(ii) Applying II.K.4 to I ≤ O F ∼= Zn gives I ∼ = Zm for some
m ≤ n. Applying it to the result of (i) (that I contains a subgroup
isomorphic to Zn ) gives n ≤ m. So m = n.
√ √
∼ 1+ d
(b) We have OK = h1, di or h1, 2 i, in either case isomorphic
to Z2 as an abelian group. So (a)(ii) yields I ∼ = Z2 ; and writing
I = hγ, δi (Z-linear combinations of γ, δ), we clearly have I ⊂ (γ, δ)
(OK -linear combinations). Since γ, δ ∈ I and I is an ideal, we also
have I ⊃ (γ, δ).
33We have yet to check that this is a ring, except for F = Q[ d].
√
34These will turn out to be always true (as we already know for quadratic fields).
172 III. RINGS
III.K. Gauss’s lemma and polynomials over UFDs
Let R be a UFD, and F := F( R) its field of fractions. Recall that

R[ x ]∗ = R∗ and F [ x ]∗ = F ∗ = F \{0}.
III.K.1. D EFINITION . (i) Given f = ∑nk=0 ak x k ∈ R[ x ], the content

of f (defined up to units) is c( f ) := gcd({ ak }) ∈ R.
(ii) f is primitive if c( f ) ∼ 1. Notice that monic polynomials are
primitive.
Clearly in general f = c( f ) · g, with g primitive, since
c( f ) = gcd({ ak }) = c( f ) · gcd({ c(akf ) }) =⇒ gcd({ c(akf ) }) = 1.
III.K.2. P ROPOSITION . Given f ∈ F [ x ]\{0}, we have

(
g ∈ R[ x ] primitive
(III.K.3) f = αg, with ,
α ∈ F∗
in which g is unique up to multiplication by units (i.e. R∗ ).
III.K.4. R EMARK . One way we will apply this is via


 f = αg

(III.K.5) f , g both primitive ∈ R[ x ] =⇒ α ∈ R∗ .
α ∈ F∗


This follows from III.K.2 since 1 · f = f = α · g gives two decom-

positions of the form (III.K.3), so that the uniqueness implies that
f = g · unit. More loosely, (III.K.5) says that “two primitive polyno-
mials which are associate in F [ x ] are associate in R[ x ].”
ak k
P ROOF OF III.K.2. Write f = ∑nk=0 bk x , ak ∈ R, bk ∈ R\{0}. Let
β
β := ∏k bk , so that β f ∈ R[ x ], and γ := c( β f ). Then g := γ f ∈ R[ x ]
is primitive and f = γβ g. If α0 g0 = f = αg with g, g0 primitive, then
III.K. GAUSS’S LEMMA AND POLYNOMIALS OVER UFDS 173
∃b ∈ R such that
αb, α0 b ∈ R =⇒ (αb) g = (α0 b) g0
| {z } | {z }
content αb content α0 b
0
=⇒ αb ∼ α b
=⇒ uαb = α0 b (u ∈ R∗ )
=⇒ αbg = uαbg0
=⇒ g = ug0
=⇒ g ∼ g0 ,
which completes the proof.
The following basic result goes back to Gauss’s Disquisitiones

Arithmeticae (c. 1800).
III.K.6. G AUSS ’ S L EMMA (v. 1.0). f , g ∈ R[ x ] primitive =⇒ f g

primitive.
m+n
P ROOF. Write f = ∑in=0 ai xi , g = ∑m j k
j=0 b j x , f g = ∑k=0 ck x , and
suppose that c( f g) ∈/ R∗ (aiming for a contradiction). Let r | c( f g) be
irreducible. Since R is a UFD, r is also prime.
As f [resp. g] is primitive, r cannot divide all the ai [resp. b j ], and
so there exists a least i0 [resp. j0 ] such that r - ai0 [resp. r - b j0 ]. Since
r is prime, we have r - ai0 b j0 . On the other hand, r | ∑`<i0 a` bi0 + j0 −`
and r | ∑`>i0 a` bi0 + j0 −` , so that
r - ∑`<i0 a` bi0 + j0 −` + ai0 b j0 + ∑`>i0 a` bi0 + j0 −` = ci0 + j0 .

This contradicts the assumption that r divides c( f g). Conclude that

c( f g) ∈ R∗ and f g is primitive.
Now let h ∈ R[ x ]\ R be a polynomial of positive degree.
III.K.7. G AUSS ’ S L EMMA (v. 2.0). h is irreducible in R[ x ] ⇐⇒ h is

primitive (in R[ x ]) and irreducible in F [ x ].
P ROOF. ( ⇐= ): If h is reducible in R[ x ], then we have h = f g

/ R[ x ]∗ = R∗ . Assume deg( f ) ≤ deg( g). Then either
with f , g ∈
174 III. RINGS
deg( f ) = 0 and f | c(h) =⇒ c(h) 1, or deg( f ) > 0 =⇒ h

reducible in F [ x ].
( =⇒ ): If h is irreducible in R[ x ], then obviously h is primitive.
Let h = f g in F [ x ], with f , g both of positive degree. By III.K.2,
f = α f 0 , g = βg0 (with f 0 , g0 ∈ R[ x ] primitive, and α, β ∈ F ∗ ) =⇒
h = αβ f 0 g0 . By III.K.6, f 0 g0 is primitive. By (III.K.5), f 0 g0 ∼ h =⇒
αβ ∈ R∗ =⇒ h = (αβ f 0 ) g0 is reducible in R[ x ], a contradiction.
Recall that we are assuming R is a UFD.
III.K.8. T HEOREM . R[ x ] is a UFD. (In particular, Z[ x ] is one.)
So uniqueness of factorization is stable under adjoining indeter-

minates, unlike the property of having all ideals be principal.
III.K.9. C OROLLARY. R[ x1 , . . . , xn ] is a UFD. (So for F any field,

F[ x1 , . . . , xn ] is one.)
In particular, F [ x1 , . . . , xn ] is a UFD, which is fortunate since oth-

erwise algebraic geometry would have no chance of working!
P ROOF OF III.K.9. Recall that F [ x ] is a UFD. Given f ∈ R[ x ]\{0},

we have
f = c( f ) g ( g ∈ R[ x ] primitive)
= c ( f ) g1 · · · g k ( g j ∈ F [ x ] irreducibles)
= c( f )( β 1 f 1 ) · · · ( β k f k ) ( β j ∈ F ∗ , f j ∈ R[ x ] primitive)
( f 1 · · · f k primitive by III.K.6,
= c( f ) β f 1 · · · f k
hence β ∈ R∗ by (III.K.5))
= α1 · · · α ` f 1 · · · f k (αi ∈ R irreducible)
where the last step is possible because R is a UFD. Clearly the αi are
irreducible in R[ x ], and by III.K.7, so are the f j .
Now we must show the essential uniqueness of this factoriza-
tion. If f = α10 · · · α0`0 f 10 · · · f k00 (deg(αi0 ) = 0, deg( f j0 ) > 0) is an-
other factorization into irreducibles in R[ x ], then III.K.7 =⇒ the
f j0 are irreducible in F [ x ] and primitive, whence (by III.K.6) f 10 · · · f k00
III.K. GAUSS’S LEMMA AND POLYNOMIALS OVER UFDS 175
is primitive. So we get α1 · · · α` ∼ α10 · · · α0`0 and f 10 · · · f k00 ∼ f 1 · · · f k

by III.K.2. Since R is a UFD, ` = `0 and αi0 ∼ ασ(i) (in R, hence in R[ x ])
for some σ ∈ S` . And because F [ x ] is a UFD, k = k0 and f j0 ∼ f π ( j)
(in F [ x ], hence in R[ x ] by III.K.2) for some π ∈ Sk .
III.K.10. C OROLLARY. Let f ∈ R[ x ] be primitive, g ∈ R[ x ]\{0}, and

f | g in F [ x ]. Then f | g in R[ x ].
P ROOF. Using III.K.9, write g = α1 · · · α j g1 · · · gk , with αi ∈ R

irreducible and g j ∈ R[ x ] irreducible of positive degree. By III.K.7,
the g j are primitive, and irreducible in F [ x ]. Hence we may write
g = (α1 · · · α j g1 ) g2 · · · gk as a product of irreducibles in F [ x ].
Since f | g in F [ x ] (and F [ x ] is a UFD), we have f = βgi1 · · · gir
for some β ∈ F ∗ and {i1 , . . . , ir } ⊆ {1, . . . , k }; note that gi1 · · · gir is
primitive by III.K.6. Since f is also primtive, applying III.K.5 gives
β ∈ R∗ . So f | g in R[ x ].
III.K.11. C OROLLARY. Given g ∈ R[ x ] monic, f ∈ F [ x ] monic divid-

ing g (in F [ x ]). Then f ∈ R[ x ].
P ROOF. Write (by III.K.2) f = αh, with h ∈ R[ x ] primitive and

α ∈ F ∗ . Then h| g in F [ x ], and so (by III.K.10) h| g in R[ x ]. Accordingly,
we write g = hG, with G ∈ R[ x ]. Since the highest coefficient of g is
1, the highest coefficients of h and G must be units, say uh , uG ∈ R∗ .
But then f monic =⇒ α = u− 1
h , and so f ∈ R [ x ].
The main application of these results for now is to proving irre-
ducibility for polynomials over Q.
III.K.12. C OROLLARY. If f ∈ Z[ x ] is monic, then all rational roots

are integers.
P ROOF. If q ∈ Q is a root, then (by III.G.16) x − q divides f in

Q[ x ]. By III.K.11, x − q must belong to Z[ x ], i.e. q ∈ Z.
III.K.13. E XAMPLE . We claim that f = x3 − 3x − 1 is irreducible in

Q[ x ]. By III.K.7, it suffices to show irreducibility in Z[ x ]. If it factored
there, it would have a linear factor, necessarily x + 1 or x − 1 (why?).
But f (1) = −3 and f (−1) = 1 are both nonzero.
176 III. RINGS
III.K.14. E ISENSTEIN ’ S I RREDUCIBILITY C RITERION . If f ( x ) = a0 +

a1 x + · · · + an x n ∈ Z[ x ], and there exists a prime p such that p| ai (for
i = 0, . . . , n − 1), p - an and p2 - a0 , then f is irreducible in Q[ x ].
P ROOF. First notice that if f is not primitive, then p - c( f ), and

f
f˜ := c( f ) is primitive and still satisfies the hypotheses. Moreover, if
f˜ is irreducible in Q[ x ], so is f . So we may assume for the rest of the
proof that f is primitive.
By III.K.7, it suffices to show that f is irreducible in Z[ x ]. Suppose
that f = gh where g = b0 + · · · + br xr and h = c0 + · · · + cs x s . Since
f is primitive, r and s are both positive, and the assumptions yield:
• p | b0 c0 but p2 - b0 c0 hence (swapping g and h if needed) p - c0
and p | b0 ; and
• p - br cs hence p - br .
Let i0 denote the least integer i for which p - bi . Since 0 < i0 ≤ r < n
we have

p ai0 = c0 bi0 + c1 bi0 −1 + · · · + ci0 b0
|{z} | {z }
p- p|
which is a contradiction.
III.K.15. E XAMPLE . To see that f = x n − p is irreducible in Q[ x ],

simply note that the hypotheses of III.K.14 hold: p does not divide
the coefficient of x n , but divides all other coefficients, with p2 not
dividing the constant term.
The last two examples show that if θ ∈ R satisfies θ 3 − 3θ − 1

[resp. θ n = p] then
Q[ θ ] ∼
= Q[ x ]/( x3 − 3x − 1) [resp. ∼
= Q[ x ] / ( x n − p )]
is a field, using the fact that Q[ x ] is a PID (cf. III.H.8). Since Z[ x ] is a
UFD, the corresponding quotients of Z[ x ] are domains by III.I.13.
III.L. ALGEBRAIC NUMBER RINGS 177
III.L. Algebraic number rings
Let F = Q[u1 , . . . , un ] be an algebraic field extension of Q, and

O F ⊂ F the subset of algebraic integers in F, i.e. elements which are
roots of monic polynomials with coefficients in Z. We begin this sec-
tion by making good on a promise from III.E.8, namely showing that
O F is a ring. One has to be more√clever than to attack this directly;
√
try to check directly that 3 5 + 1+ 2 17 − 3i is an algebraic integer!
Consider an element α ∈ F, with minimal polynomial mα ∈ Q[ x ].
Recall that this is the unique monic generator of Iα := ker{evα : Q[ x ] →
F }, or equivalently the lowest-degree (nontrivial, monic) polynomial
over Q having α as a root. Here it is crucial that Q[ x ] is a PID.
III.L.1. T HEOREM . The following are equivalent:

(i) α ∈ O F
(ii) mα ∈ Z[ x ]
(iii) Z[α] is a finitely generated abelian group
(i.e., Z[α] = Z + αZ + α2 Z + · · · + αn−1 Z for some n ∈ N)
(iv) There exists a nontrivial f.g. abelian subgroup G ≤ Q[α] closed under
multiplication by α.
P ROOF. We do this “merry-go-round” style:

(i) =⇒ (ii): By definition of O F , there exists a monic f ∈ Z[ x ] with
f (α) = 0. Then f ∈ Iα = (mα ) ⊂ Q[ x ] =⇒ f = mα g for some
g ∈ Q[ x ]. But now since f and mα are monic, f ∈ Z[ x ], and mα | f , we
have mα ∈ Z[ x ] by III.K.11.
(ii) =⇒ (iii): Let n = deg(mα ), so that
m α ( x ) = x n + a n −1 x n −1 + · · · + a 0 , ai ∈ Z.
Then mα (α) = 0 =⇒
αn = − an−1 αn−1 − · · · − a0 ∈ h1, α, α2 , . . . , αn−1 i,
where the RHS denotes the additive abelian subgroup of F generated

by these elements. Inductively let m > n, and assume we know that
178 III. RINGS
αk ∈ h1, α, . . . , αn−1 i for k ≤ m − 1. Then
αm = αm−n · αn ∈ hαm−n , αm−n+1 , . . . , αm−1 i ≤ h1, α, α2 , . . . , αn−1 i.
Hence Z[α] = h1, α, . . . , αn−1 i as a group.

(iii) =⇒ (iv): Take G = Z[α]. Then
αG = αZ[α] = hα, α2 , . . . , αn i ≤ h1, α, . . . , αn−1 i = Z[α] = G.
(iv) =⇒ (i): Let G = hγ1 , . . . , γr i ≤ Q[α] be a finitely generated

abelian subgroup. By assumption on G, we can express
r
αγi = ∑ µij γj (i = 1, . . . , r ) with µij ∈ Z.
j =1
Rewriting this in matrix form35 gives

    
γ1 µ11 · · · µ1r γ1
 ..   .. .. ..   .. 
α .  =  . . .  . 
γr µr1 · · · µrr γr
| {z }
µ(α)
and we see that α is an eigenvalue of µ(α), hence a root of the char-

acteristic polynomial f ( x ) := det( xIr − µ(α)). Now simply observe
that f is monic and belongs to Z[ x ].
III.L.2. C OROLLARY. O F is a subring of F, called the ring of integers

of F (or simply an algebraic number ring).
P ROOF. We need only check closedness of O F under addition

and multiplication. Let α, β ∈ O F . Then Z[α] and Z[ β] are finitely
generated, from which it follows that Z[α, β] is also finitely gener-
ated. More concretely, if Z[α] = Z + αZ + · · · + αn−1 Z and Z[ β] =
Z + βZ + · · · + βm−1 Z, then Z[α, β] = Z[α][ β] = ∑in=−01 ∑m −1 i j
j=0 α β Z.
Both Z[α + β] and Z[αβ] are additive subgroups of Z[α, β], and so
35The vectors here belong to the vector space Q[α]r over the field Q[α], and the re-
sult we are using from linear algebra works over any field: given M~v = λ~v, clearly
~v is in the kernel of left-multiplication by λIr − M, which means the columns of
the latter are dependent and hence that its determinant is zero.
are themselves finitely generated (cf. II.K.4). By III.L.1, α + β and αβ

belong to O F .
III.L.3. E XAMPLE . In HW, you’ll show that the pth cyclotomic

polynomial x p−1 + x p−2 + · · · + 1 is irreducible (for p an odd prime).
p −1 2πi
In C[ x ], this factors as ∏k=1 ( x − ζ kp ) where ζ p = e p . So all powers
of ζ p are algebraic integers, and Z[ζ p ] ⊆ OQ[ζ p ] . The field Q[ζ p ] ∼
=
Q[ x ]/( x p−1 + x p−2 + · · · + 1) is called the pth cyclotomic field.
p −2
Given an arbitrary element α = a0 + a1 ζ p + · · · + a p−2 ζ p ∈
OQ[ζ p ] , we know from III.L.1 that the minimal polynomial has inte-
ger coefficients. In fact, one can use Galois theory to show that all
the ai must be integers, hence that Z[ζ p ] = OQ[ζ p ] . We will prove
this next semester, along with the
III.L.4. P ROPOSITION (Kummer). If u ∈ Z[ζ p ]∗ , then u/ū is a root

of unity.
I am stating these results now because we will refer to them in an

application at the end of the section.
Integral ideals. Now write, given a number field K, I(K ) for its
monoid of integral ideals (i.e. nonzero ideals I ⊂ OK ). Slightly chang-
ing notation,36 we write J (K ) for the fractional ideals a = λI (λ ∈ K ∗ ,
I ∈ I(K )). Recall that a ∈ J (K ) is invertible iff a · b = OK for some
b ∈ J ( K ).
We have seen that
(a) factorization into irreducibles is not necessarily unique, and
(b) irreducibles need not be prime
in a non-UFD OK . As we shall now see, replacing {OK∗ , OK , K } by
{(1), I(K ), J (K )} makes these problems disappear.
The proof of the next result requires Galois theory if K is a gen-
eral number field (as it involves introducing discriminants and ideal-
norms in general), so we shall assume it. However, I will explain
how it follows from what we already know when K is quadratic.
36This is instead of writing J (O ).
K
180 III. RINGS
III.L.5. T HEOREM . (i) Maximal ideals I ∈I(K ) are invertible in J (K ).

(ii) There exists a homomorphism of monoids N : I(K ) → N\{0} strictly
respecting inclusions: I ⊇ J =⇒ N( I ) ≤ N( J ) with equality iff I = J.
(iii) Every ideal I ∈ I(K ) is finitely generated as an abelian group.
√
P ROOF FOR K = Q[ d] . (i) We proved J (K ) is a group when we
used Hurwitz’s theorem to show N1( I ) Ĩ · I = (1).
(ii) We know N( I · J ) = N( I ) · N( J ). Now if I ⊃ J then Ĩ ⊃ J̃
=⇒ I Ĩ ⊃ J J̃ =⇒ (N( I )) ⊃ (N( J )) =⇒ N( I ) | N( J ) (which is
in fact stronger than N( I ) ≤ N( J )). If also N( I ) = N( J ) =: m, then
Ĩ ⊇ J̃ =⇒
J = (1) J = m1 I Ĩ J ⊇ m1 J J̃ I = (1) I = I
hence I = J.
(iii) See Fact III.J.15, proved in III.J.17(b).
III.L.6. R EMARK . (a) From (i), it follows that any product of max-
imal ideals is invertible in J (K ).
(b) In (iii), I is of rank [K:Q] as an abelian group, at least according
to unproved assertions in III.J.17.
(c) For any I ∈ I(K ), the quotient OK /I is a finite abelian group.
One can define an ideal norm by N( I ) := |OK /I |, which agrees with
the definition in the quadratic case.
III.L.7. L EMMA . Given I, J ∈ I(K ), with I ⊃ J, and I invertible37 in

J (K ). Then:
(i) I −1 J ∈ I(K );
(ii) I | J in I(K ); and
(iii) I −1 J ⊃ J, with equality iff I = OK .
P ROOF. (i) I ⊃ J =⇒ OK = I −1 I ⊃ I −1 J =⇒ I −1 J ∈ I(K ).

(ii) I | I · I −1 J = J.
(iii) OK ⊃ I =⇒ (OK ·) I −1 J ⊃ I · I −1 J = J. If I −1 J = J then αJ ⊂
J for each α ∈ I −1 . Since J is finitely generated, say = h β 1 , . . . , β n i,
37This is in fact always true. See III.L.10.
we can write multiplication by α in this basis: [α]{ β} =: µ(α), with

entries in Z. Set f (λ) := det(λI − µ(α)), which as before is monic
and integral. By Cayley-Hamilton, 0 = f (µ(α)) = [ f (α)]{ β} =⇒
f (α) = 0 =⇒ α ∈ OK . Since α ∈ I −1 was arbitrary, we have
I −1 ⊂ OK =⇒ OK = I I −1 ⊂ I OK = I =⇒ I = OK .
III.L.8. R EMARK . Note that I(K )∗ — the invertible elements with

inverse in I(K ) — is trivial (= {OK }). This is because if both I, I −1 ∈
I(K ) then I I −1 = OK ⊃ I = I OK ⊃ I I −1 =⇒ I = OK . Hence the
natural definition of “irreducible element” I ∈ I(K ),
“I = I1 I2 =⇒ I1 or I2 is invertible in I(K ),”
becomes
“I = I1 I2 =⇒ one of I1 and I2 is just OK ”
— no factoring at all. As mentioned at the end of §III.J, this is what

we will mean by an irreducible (integral) ideal.
III.L.9. T HEOREM . Any J ∈ I(K ) is a product of maximal ideals.
P ROOF. Suppose otherwise, and choose J ∈ I(K ) a non-product-

of-maximals of smallest possible N( J ). Observe that J non-maximal
=⇒ ∃ I ∈ I(K ) such that OK ) I ) J =⇒ N( I ) < N( J ) by
III.L.5(ii). By “minimality” of N( J ), I must be a product of maximal
ideals; according to III.L.6(a), it is then invertible in J (K ).
By III.L.7, since I is invertible and contains J, we must have I −1 J ∈
I(K ), with I −1 J ) J (since I 6= OK ) hence N( I −1 J ) < N( J ). Again
by “minimality” of N( J ), I −1 J must be a product of maximal ideals,
which presents J = I · ( I −1 J ) itself as a product of maximals, a con-
tradiction.
III.L.10. C OROLLARY. (i) Any I ∈ I(K ) is invertible in J (K ).

(ii) J (K ) is a group (abelian, of course).
P ROOF. (i) Use III.L.5(i) and III.L.9.

(ii) Given a = λI, a−1 = λ−1 I −1 gives an inverse.
182 III. RINGS
III.L.11. R EMARK . The Corollary implies that Lemma III.L.7 doesn’t

need the invertibility hypothesis on I. So III.L.7(ii) simply reads
I⊃J ⇐⇒ I | J,
that is, “to divide is to contain”. This is a different result than III.D.16,
but we will call it Caesar’s lemma as well.
Note in addition that for I ⊃ J, III.L.7 now gives J 0 := I −1 J ∈
I(K ), so that J = I J 0 . By multiplicativity of N, we get N( J ) =
N( I )N( J 0 ) hence N( I ) | N( J ).
Before stating the next (extremely important) result, recall that a

priori “ ℘ is a prime ideal” means
(III.L.12) ℘ 3 ab =⇒ ℘ 3 a or ℘ 3 b.
Suppose that ℘ contains I J but not I, and let ı0 ∈ I \( I ∩ ℘). Then

ı0  ∈ I J ⊂ ℘ (∀  ∈ J), hence all  ∈ J are in ℘ by (III.L.12); conclude
that ℘ ⊃ J. This gives an alternate characterization
(III.L.13) ℘ ⊃ IJ =⇒ ℘ ⊃ I or ℘ ⊃ J
of primality of ℘, which is more suitable for the present context.
III.L.14. P ROPOSITION . For ℘ ∈ I(K ) proper ( ℘ ( OK ), the follow-

ing are equivalent:
(a) ℘ is irreducible (in I(K ): i.e., doesn’t factor at all);
(b) ℘ is a maximal ideal;
(c) ℘ is a prime ideal; and
(d) ℘ is a prime element in I(K ) ( ℘ | I J =⇒ ℘ | I or ℘ | J).
P ROOF. (a) =⇒ (b): If ℘ is non-maximal, it is a product of (multiple)

maximal ideals by III.L.9, and so is not irreducible.
(b) =⇒ (c): If ℘ is maximal, then OK / ℘ is a field hence a domain,
and so ℘ is a prime ideal.
(c) =⇒ (d): Caesar.
(d) =⇒ (a): Suppose ℘ is a prime element of I(K ), and that ℘ = I J
(I, J ∈ I(K )). Then ℘ | I or ℘ | J, say the former: I = ℘Q (Q ∈ I(K ))
=⇒ ℘ = I J = ℘Q J =⇒ N( ℘) = N( ℘)N(Q )N( J ) in N. So
N(Q ) = 1 = N( J ), whence Q = OK = J by III.L.7(iii). So ℘ is
irreducible.
Finally we come to the main point:
III.L.15. C OROLLARY. Any ideal I ∈ I(K ) has a unique factorization

(up to order) into prime ideals (hence into primes/irreducibles in I(K )).
P ROOF. Existence of such a factorization follows from III.L.9 and

III.L.14, and one can give a direct proof of uniqueness using Cae-
sar and III.L.7. A more intuitive approach is to use [Jacobson, Thm.
2.21] extending our results on UFDs to unique factorization monoids.
We want to show I(K ) is a UFM, so it suffices to check DCC and PC.
For DCC, use the norm N and III.L.5(ii); and PC follows immediately
from III.L.14.
Here is a somewhat obvious but useful result:
III.L.16. C OROLLARY. Let J ∈ I(K ) have prime norm N( J ) ∈ N.

Then J satisfies the equivalent conditions of III.L.14.
P ROOF. We need only prove that J is maximal. To this end, sup-

pose otherwise and let J ( I ( OK . Then by III.L.5(ii), N( J ) >
N( I ) > 1 (= N(OK )). But by III.L.7 (cf. Remark III.L.11), we have
N( I )|N( J ), a contradiction since N( J ) is prime.
The ideal class group. Next, we denote by P J (K ) ≤ J (K ) the

subgroup of principal fractional ideals, i.e. those of the form (λ) :=
λOK , for λ ∈ K ∗ , and by
C `(K ) := J (K )/P J (K )
the ideal class group.38
III.L.17. D EFINITION . The class number of K is the order
hK := |C `(K )|
38As with J (K ), this is a slight change in notation from III.F.16.
184 III. RINGS
of the ideal class group.39
III.L.18. T HEOREM . OK is a PID ⇐⇒ hK = 1.
P ROOF. By definition, OK is a PID if and only if all integral ideals

are principal, which is to say (i) I(K ) = I(K ) ∩ P J (K ). The class
number is 1 exactly when (ii) J (K ) = P J (K ). Clearly (ii) implies (i)
by intersecting both sides with I(K ). Moreover, given a ∈ J (K ), we
have a = λI for some I ∈ I(K ); if (i) holds, then I is principal, and
then so is a. Hence (i) implies (ii).
Write [a] := a · P J (K ) for the coset (ideal class) of a fractional

ideal a. The identity element is [OK ] = [(1)] =: e. Here are some
(mostly obvious) rules for working in C `(K ):
III.L.19. P ROPOSITION . Let I, J ∈ J (K ).

(i) [ I ] = e ⇐⇒ I ∈ P J (K ) (I is principal).
(ii) [ I ] = [ J ] ⇐⇒ I · P J (K ) = J · P J (K ) ⇐⇒ I = (λ) J for some
λ ∈ K ∗ ⇐⇒ (α) I = ( β) J for some α, β ∈ OK \{0}.
(iii) [ I ][ J ] = [ I J ] (multiplication of cosets).
(iv) [ I ]−1 = [ I −1 ].
(v) [ I ]m = e ⇐⇒ I m is principal.
(vi) I J = (α) ⇐⇒ [ I ]−1 = [ J ].
P ROOF OF ( VI ). e = [(α)] = [ I J ] = [ I ][ J ].
(i) (iii)
This is all very useful for solving (or showing insoluble) Dio-
phantine equations like X 2 = Y 3 − 14, as you will see in Problem
Set 10.
III.L.20. T HEOREM . If the ring of integers OK of an algebraic number

field is a UFD, then it is a PID.
III.L.21. C OROLLARY. hK = 1 ⇐⇒ OK PID ⇐⇒ OK UFD.
39This is always finite, a fact which we will not be able to prove (but see III.L.27
for the idea).
√
P ROOF OF III.L.20 FOR K = Q[ d] . Suppose that OK is a UFD.
To show that it is a PID (every ideal is principal), it will suffice to
prove that its prime ideals are principal, since (by III.L.14) every
ideal is a product of maximal ideals, and maximal ideals are prime.
So let ℘ ∈ I(K ) be a prime ideal, and write N( ℘) = ∏i σi for the
(unique) decomposition of its norm into irreducibles in OK . Since
OK is a UFD, these irreducibles σi are prime elements of OK , so that
the (σi ) are prime ideals, and thus irreducible in I(K ) by III.L.14.
By Hurwitz, ℘ ℘ ˜ = (N( ℘)) =⇒ ℘ | (N( ℘)) = ∏ (σi ) =⇒ ℘ |
i
(σi ) for some i (since ℘ ∈ I(K ) is a prime element). By irreducibility
of (σi ), we have ℘ = (σi ), so that ℘ is principal as desired.
III.L.22. R EMARK . (a) In a non-UFD OK , the irreducible σi need

not be prime, and so the (σi ) need not be irreducible as elements of
I(K ). These principal ideals can and do split up into products of
(necessarily) non-principal prime ideals.
(b) The observation that ℘ | (N( ℘)) does generalize to arbitrary
number fields; therefore, so does the above proof.
For all this to be useful for number theory, we need to be able to

compute class groups, which requires being able to find all the prime
√
ideals and then all the ideals of a given norm. Consider K = Q[ d]:
III.L.23. L EMMA . Let ℘ ∈ I(K ) be a prime ideal. Then there exists a

unique prime p ∈ N such that ℘ | ( p). Hence, if p 6= 2 then
℘p or ℘˜p if ( d ) = 0 or 1
(
℘= p
(III.L.24)
( p) if ( dp ) = −1
√ √
where in the first line ℘p := ( p, m − d) (and ℘˜p = ( p, m + d)) are
determined from d ≡ m2 .
( p)
n
P ROOF. Let N( ℘) = ∏i pi i be a prime factorization in N. As ℘ is
prime and ℘ | (N( ℘)), we must have ℘ | ( pi ) for some pi =: p. So
N( ℘) | N(( p)) = p2 (for K quadratic) =⇒ N( ℘) = p or p2 .
If N( ℘) = p, then (by Hurwitz) ℘ ℘ ˜ = ( p), whence ℘ = ℘p or ℘˜p
since I(K ) is a UFM. If N( ℘) = p2 , then (by III.L.5(ii)) ℘ = ( p).
186 III. RINGS
For the uniqueness, if ℘ | (q) for some other prime q, we get

N( ℘) = q or q2 ; hence q = p.
If p = 2, the possibilities are a bit more complicated and depend

on the congruence class mod 8 (see Problem Set 10).
Continuing to assume K quadratic, we have the
III.L.25. P ROPOSITION . Let I ∈ I(K ) and suppose

mj
∏ pi i ∏ 0 p j ∏ 00 pk k
` n
N( I ) =
i j k
is a prime factorization (in N) with ( pd ) = 1, ( 0 dp ) = 0, and ( 00 dp ) = −1.

i j k
Then the {nk } are even, and40

n
a ` −a
∏ ℘ i ℘˜ i i i ∏ 0 ℘j j ∏(00 pk ) 2
m k
I= i
i j k
with 0 ≤ ai ≤ ì .
P ROOF. We have (00 pk ) irreducible, (0 p j ) = 0 ℘2j , ( pi ) = ℘i ℘

˜ i , and
` ˜ ì 0 ℘2m j
I | (N( I )) = ∏i ℘i i ℘ i ∏j j ∏k (00 pk )nk . By uniqueness of fac-
a b
˜ i ∏ 0 ℘ ∏ (00 pk )dk where cj
torization in I(K ), we have I = ∏i ℘i i ℘ i j j k
` mj n
ai , bi ≤ ì , c j ≤ 2m j , dk ≤ nk , and ∏i pi i ∏ j 0 p j ∏k 00 pk k = N( I ) =
a +b cj 2d
∏i pi i i ∏ j 0 p j ∏k 00 pk k . By uniqueness of factorization in N, ai +
bi = ì , c j = m j , and 2dk = nk .
√ √
III.L.26. E XAMPLE . Let K = Q[ −29]. I claim that OK = Z[ −29]
has an ideal of norm 5 and order 3 in C `(K ).
Consider the integer prime p = 5: since −29 ≡ 12 , we have
(5)
√ √
˜
(5) = (5, 1 − −29)(5, 1 + −29) = ℘5 ℘5 ; and by the Proposition,
℘5 , ℘˜ 5 are the only ideals of norm 5. Pell’s equation a2 + 29b2 = 5 is
insoluble, so ℘5 is non-principal and [ ℘5 ] is nontrivial.
On the other hand, a2 + 29b2 = 125 has solutions (±3, ±2), and
√
so ( β) := (3 + 2 −29) has norm 53 . This gives ( β) | (N(( β))) =
3 3− a
(5)3 = ℘3 ℘˜ =⇒ ( β) = ℘a ℘˜
5 5 for some a ∈ {0, 1, 2, 3}.
5 5
40The notation here means for instance ℘

˜i = ℘
˜ p and 0 ℘ j = ℘ 0 .
i pj
√
Now β = 5 − 2(1 − −29) ∈ ℘5 , so that (by Caesar) ℘5 | ( β).
√
If also ℘ ˜ 5 | ( β), then (5) = ℘5 ℘ ˜ 5 | ( β) hence 5 | 3 + 2 −29, which
is visibly false.41 So we conclude that ( β) = ℘35 , hence that [ ℘5 ]3 =
[ ℘35 ] = [( β)] = e as claimed. Note also that [ ℘˜ 5 ] = [ ℘5 ]−1 = [ ℘5 ]2
since [ ℘5 ][ ℘ ˜ 5 ] = [(5)] = e.
˜ 5 ] = [ ℘5 ℘
III.L.27. R EMARK . In order to compute C `(K ) completely, one

uses the Minkowski bound: for each class τ ∈ C `(K ), there exists a
representative I ∈ I(K ) (i.e. [ I ] = τ) satisfying
r2
4 n!
q
N( I ) ≤ BK := | ∆ K |,
π nn
where n = [K:Q] is the degree, r2 the number of pairs of conjugate
complex embeddings, and ∆K is the discriminant.42 By III.L.25 (and
its generalization to arbitrary number fields), it follows that there are
only finitely many ideal classes, so that hK < ∞.
Fermat’s equation. The foregoing is useful for treating Diophan-
tine equations, which are polynomial equations in one or more vari-
ables with integer coefficients, to which integer solutions are sought.
A particularly famous example is
(III.L.28) xp + yp = zp , p = prime > 3.
Of course, Fermat’s Last Theorem states that for any exponent n > 2,
the only solutions to x n + yn = zn are the “trivial” ones, with x or
y = 0. The cases n = 4 (Fermat) and 3 (Euler) were proved by
Fermat’s method of descent; and if one has the theorem for some n,
one has it for all exponents divisible by n (why?).
As you may know, the proof was ultimately completed by Wiles
in 1995, building on decades of work by many people on modularity
and Galois representations. What I want to discuss here is Kum-
mer’s big advance in the mid-19th Century, which led to the devel-
opment of ideals.
41That is, 3 + 2 √−29 does not belong to Z[√−29].
5 5√
42Say K = Q[ d]. Then n = 2; and r is 1 for d < 0 and 0 for d > 0. The
2
discriminant is 4d unless d ≡ 1, in which case it is d.
(4)
188 III. RINGS
Suppose there exists a solution to (III.L.28) in relatively prime

x, y, z ∈ Z\{0}, none divisible by p.43 (In fact, x and y must also be
coprime; otherwise m | x, y =⇒ m p | z p =⇒ gcd(m, z) 6= 1 violates
the relative primality of x, y, z.) We will obtain a contradiction by
passing to the “cyclotomic” number ring Z[ζ ], where ζ denotes a
primitive pth root of 1, and considering the equation
(III.L.29) ( x + y)( x + yζ ) · · · ( x + yζ p−1 ) = z p .

We split the argument up into two cases.
Case 1: Z[ζ ] a UFD. As the pth cyclotomic polynomial

tp − 1
( t − ζ ) · · · ( t − ζ p −1 ) = = 1 + t + · · · + t p −1
t−1
evaluates to p at t = 1,
( p ) ⊂ (1 − ζ a ) for each a = 1, . . . , p − 1.
Since it is irreducible over Q, it is the minimal polynomial of ζ, and

thus any element of K := Q[ζ ] has a unique representation as a0 +
a 1 ζ + · · · + a p −2 ζ p −2 .
Let ω ∈ Z[ζ ] = OK be a prime factor of x + yζ. By unique fac-
torization and (III.L.29), ω | z. If ω also divides x + yζ a+1 (for some
a ∈ {1, . . . , p − 1}) then it divides the Z[ζ ]-linear combination
ζ −1 ( x + yζ ) − ζ −1 ( x + yζ a+1 ) = y(1 − ζ a )
hence yp. Now in Z, gcd(z, yp) | gcd(z, y) · gcd(z, p) = 1 · 1 = 1

=⇒ zm + ypm = 1 for some n, m ∈ Z =⇒ ω | 1 =⇒ ω ∈ Z[ζ ]∗ , a
contradiction. So ω divides no other factor in LHS(III.L.29).
Since ω divides z, ω p | z p . No ω-factor can divide other factors
(of LHS(III.L.29)), so ω p | x + yζ. By uniqueness of the decompo-
sition of x + yζ into prime factors, and repeating the argument for
43There is a case where one of x, y, z is divisible by p, which (while more compli-

cated) can be treated by similar methods.
each prime factor, we find that

(
α ∈ Z[ ζ ]
x + yζ = uα p ,
u ∈ Z[ ζ ] ∗ .
Write α = a0 + a1 ζ + · · · + a p−2 ζ p−2 .

Now we apply Kummer’s result III.L.4 that u/ū is a root of 1 in
Z[ζ ], i.e. ±ζ k for some k (we may assume u/ū = ζ k ). Modulo p, in
Z[ζ ]/( p), we have by the “freshman’s dream”
p −2
∑ ai
p p p p
p
α ≡ a0 + a1 ζ p +···+ a p −2 ζ ( p −2) p = =: a ∈ Z p .
i =0
Applying complex conjugation (which preserves the integer prime

( p)) to x + yζ = uα p ≡ ua gives x + yζ −1 ≡ ūa hence
u
ζ k ( x + yζ −1 ) = ( x + yζ −1 ) ≡ ua ≡ x + yζ mod ( p).
ū
That is, p divides x + yζ − ζ k x − ζ k−1 y in Z[ζ ]. By uniqueness of the
representation of elements of Z[ζ ], this is impossible unless k = 1.
So
p | ( x − y) + ζ (y − x ) =⇒ p | x−y =⇒ x ≡ y.
( p)
Writing x p + (−z) p = (−y) p , we obtain similarly x ≡ −z. But then

( p)
2x p ≡ x p + y p = z p ≡ − x p mod ( p)
=⇒ p | 3x p , a contradiction.
Case 2: Z[ζ ] not a UFD? Well, we aren’t going to prove Fermat’s

Last Theorem for all odd primes, so there must be a catch. But we
can still show non-existence of (nontrivial) solutions in some cases,
by reinterpreting (III.L.29) as an equation

p −1
(III.L.30) (( x + y)) (( x + yζ )) · · · ( x + yζ ) = (z) p
of ideals in Z[ζ ]. We may further (uniquely!) factor both sides of

(III.L.30) into prime ideals. If some prime ideal ℘ ⊃ (( x + yζ )) (i.e.
℘ | (( x + yζ ))), then it can’t contain/divide any other of the ideals
190 III. RINGS
on LHS(III.L.30). (Otherwise ℘ ⊃ (z, yp) = Z[ζ ] as before.) Since

I(K ) is a UFM, ℘ | (z) =⇒ ℘ p | (z) p =⇒ ℘ p | (( x + yζ )) and so
(( x + yζ )) = I p , I not necessarily principal.
Now suppose that p is a regular prime: that is,
p - hK (= hQ[ζ p ] ).
In this case, if [ I ] 6= e ∈ C `(K ), then by Lagrange we would have

[ I ] p 6= e ∈ C `(K ), contradicting principality of (( x + yζ )). Therefore
I is principal: I = (α) for some α ∈ Z[ζ ]. So (( x + yζ )) = (α p ) hence
x + yζ = uα p and we proceed as in Case 1.
The first irregular prime is 37. The method described here essen-
tially settles Fermat for any smaller exponent (prime or not). Note
how deeply we dug into the ideal structure of Z[ζ ] to deal with an
equation ostensibly in rational integers!
IV. Modules
IV.A. Definition and examples
Modules over a ring arose from algebraic number theory and rep-
resentation theory. The definition we use now, a simultaneous gen-
eralization of vector spaces over a field and the action of a group on
a set, is another contribution of E. Noether. The main immediate ap-
plications will be to the structure theory of finitely generated abelian
groups and to the canonical forms of a linear transformation on a
vector space.
IV.A.1. D EFINITION . Let R be a ring.

A left (resp. right) R-module is
• an abelian group M
together with a “scalar multiplication” map
• R × M → M resp. M × R → M
(r,m) 7−→ rm (m,r ) 7−→ mr
satisfying the axioms (∀ m, m0 ∈ M and r, r 0 ∈ R)

(i) r (m + m0 ) = rm + rm0  (m + m0 )r = mr + m0 r
 
 

(ii) (r + r 0 )m = rm + r 0 m  m(r + r 0 ) = mr + mr 0
 

resp.
(iii) (rr 0 )m = r (r 0 m) 




 m(rr 0 ) = (mr )r 0
(iv) 1R m = m m1R = m.
 
If R is commutative, then we use the terminology “R-module” as left

vs. right turn out to yield equivalent structures.
IV.A.2. E XAMPLES . (a) Given a field F, an F-module is the same

thing as an F-vector space (we can take this as the definition).
(b) A Z-module is the same thing as an abelian group.
191
192 IV. MODULES
(c) Any ring R is a (left and right) module over itself. Any left [resp.
right] ideal I ⊂ R is a left [resp. right] R-module.
(c’) Given any subring R0 ⊂ R, R is a (left and right) R0 -module, and
any R-module M has the structure of an R0 -module.
(c”) Given a ring homomorphism θ : S → R, an R-module M has the
structure of an S-module via sm := θ (s)m.
(d) Given a ring R, the map R × Rn → Rn sending (r, (r1 , . . . , rn )) 7→
(rr1 , . . . , rrn ) makes Rn into a (left) R-module. This is the prototype
for free R-modules. (“Direct summands” of Rn will be the prototype
for projective R-modules, and “quotients” of Rn for finitely gener-
ated R-modules.)
(e) For those who are familiar with manifolds, a finitely generated
projective C ∞ (M)-module is the same thing as a smooth vector bun-
dle over M.
(f) Rn is a left Mn (R)-module.
(g) Let G be a finite group. A representation of G on an F-vector
space V is a map
ρ
G×V → V
( g, v) 7→ ρ( g)v (or “g.v”)

0 0
 g.(v + v ) = g.v + g.v

satisfying g.( f v) = f ( g.v) ( f ∈ F)
( gg0 ).v = g.( g0 .v), 1G .v = v.


We can “linearize” this action to get a left-module: let F[ G ] be the

ring consisting of elements ∑i f i [ gi ] with multiplication law gener-
ated by [ g][ g0 ] := [ gg0 ], the so-called group ring of G over F. Then
we define
(∑i f i [ gi ])v := ∑i f i ( gi .v)
and check axioms (i)-(iv). So a representation of G has the structure
of an F[ G ]-module.
IV.A. DEFINITION AND EXAMPLES 193
(h) Given an F-vector space V, an endomorphism
T: V → V
is an F-linear homomorphism of abelian groups; that is, we have

T ( f v) = f T (v) and T (v + v0 ) = T (v) + T (v0 ) (∀ f ∈ F, v, v0 ∈
V). Denoting the collection of all such by EndF (V), we consider the
evaluation map
F[λ] −→ EndF (V )
θ
P(λ) 7−→ P( T ),
where λ is an indeterminate.
Now, we can add and compose endomorphisms, making EndF (V )
into a ring and V into an EndF (V )-module. It also makes θ a ring
homomorphism, with image
im(θ ) =: F[ T ].
By (c”), this gives V the structure of an F[λ]-module, which leads to

the theory of canonical forms for T.
(i) Let F be a number field, and a ⊂ F be a fractional ideal. Then a
has the structure of O F -module. Indeed, F is also an O F -module; but
it is not finitely generated as an abelian group (why?), whereas a is.
Conversely, we claim that any finitely generated abelian subgroup
of F with O F -module structure is a fractional ideal. Let a ≤ F be f.g.
and closed under multiplication by O F ; then we ask: does there exist
an element f ∈ F such that f a ⊂ O F ? If this is true, then f a =: I is an
ideal of O F , and a = f −1 I a fractional ideal.
To see this, let α1 , . . . , αk be a generating set for a (as abelian
group), and write αi = bai , ai , bi ∈ O F , using the fact that F is the
i
fraction field of O F . Then (∏ j b j )αi ∈ O F (∀i ) =⇒ (∏ j b j )a ⊂ O F .
Now consider the
IV.A.3. D EFINITION . A module M over a ring R is finitely gener-

ated (as an R-module) if there exists a finite subset S ⊆ M such that
M = {∑s∈S rs s | rs ∈ R}.
194 IV. MODULES
Since O F is f.g. as an abelian group, a is f.g. as an abelian group

iff a is f.g. as an O F -module, and so we have the
IV.A.4. P ROPOSITION . The fractional ideals of F are precisely the finitely

generated O F -submodules of F.
(I’ll discuss submodules at greater length later.)
The similarities between Defn. IV.A.1 ((iii) and (iv) in particular)

and the definition of a monoid G acting on a set X,1 suggest recasting
the definition of module as a homomorphism of rings — just as we
can recast the monoid action as a homomorphism of monoids G →
TX (where TX is the monoid of transformations). In the remainder of
the section we work this out.
IV.A.5. D EFINITION . Given an abelian group ( M, +, 0), the set of

endomorphisms End( M) is the set of homomorphisms η : M → M.
(The defining properties are η ( x + y) = η ( x ) + η (y) and η (0) = 0,
consequences of which include η (− x ) = −η ( x ), η (nx ) = nη ( x ), and
the determination of η by its behavior on a generating set for M.)
IV.A.6. P ROPOSITION . End( M ) is a ring under addition and compo-

sition of endomorphisms.
S KETCH . I’ll summarize some key points:

• 1End( M) = id M
• 0End( M) = zero-map (sending everything to 0)
• End( M) is closed under addition since
(η + ζ )( x + y) = η ( x + y) + ζ ( x + y)
= η ( x ) + η (y) + ζ ( x ) + ζ (y)
[M abelian =⇒ ] = η ( x ) + ζ ( x ) + η (y) + ζ (y)
= (η + ζ )( x ) + (η + ζ )(y).
1Same as Defn. II.F.1, with G only taken to be a monoid.

• Distributivity properties hold, e.g.

((η + ζ )ρ)( x ) = (η + ζ )(ρ( x )) = η (ρ( x )) + ζ (ρ( x ))
= (ηρ)( x ) + (ζρ)( x ) = (ηρ + ζρ)( x ).
What is the group of units (End( M))∗ ? These are, naturally,

the invertible endomorphisms — the automorphisms Aut( M). Note
that this is a subgroup of the multiplicative monoid of End( M) and is not
usually closed under addition.
IV.A.7. E XAMPLE . (i) Let M = (Z, +, 0). Then we have End( M) =

(Z, +, •, 0, 1). Why? M is generated by 1, so any endomorphism is
determined by where 1 is sent. Of course, Aut( M ) = {±1} ∼ = Z2 (as
a ring).
(ii) Let M = (Zn , +, 0). Again (for the same reason) End( M ) =
(Zn , +, •, 0, 1), but Aut( M) ∼
= Z∗n .
(iii) Let M = Zn . I claim that End( M ) ∼
= Mn ( Z ) :
P ROOF. Write e1 , . . . , en for the standard basis (column) vectors

in Zn . We define φ : End(Zn ) → Mn (Z) by φ(µ) := (µ(e1 ) | · · · |
µ(en )), so φ(idZn ) = 1n and φ(0) = 0n ; φ clearly respects “+”. As for
“•”: for any µ ∈ End(Zn ) and v ∈ Zn , matrix-vector multiplication
yields
v1 n
φ(µ)v = (µ(e1 ) | · · · | µ(en )) ... = ∑ vi µ(ei ) = µ (∑in=1 vi ei )
vn i =1
= µ ( v ).
So for η, ζ ∈ End(Zn ), we have φ(η )ζ (ei ) = ηζ (ei ) hence
φ(ηζ ) = (ηζ (e1 ) | · · · | ηζ (en )) = φ(η ) · (ζ (e1 ) | · · · | ζ (en ))

= φ ( η ) · φ ( ζ ),
where the dot is matrix multiplication. Injectivity and surjectivity
are clear, since the {ei } freely generate Zn .
196 IV. MODULES
One should compare the following to “Cayley for monoids”:2
IV.A.8. T HEOREM . Any ring R is isomorphic to a ring of endomor-

phisms of an abelian group, i.e. to a subring of End( M ) for some abelian
group M.
P ROOF. Let M = ( R, +, 0), and denote by `r : M → M the group

m 7−→ rm
homomorphism given by left-multiplication by an element r ∈ R.
We obtain a homomorphism of rings by
` : R → End( M)
r 7 → `r
(since rs 7→ `rs = `r `s
and r + s 7→ `r+s = `r + `s ).
We only need to show that `( R) ∼

= R, i.e. that ` presents R as a sub-
ring of End( R). That is, we must check injectivity. If `r = 0End( M) ,
then rm = 0 (∀m ∈ M) =⇒ r = r1 = 0, done.
If we try the same thing for right multiplication, we run into the
problem
rrs (m) = m(rs) = (mr )s = (rr (m))s = rs (rr (m)) = (rs rr )(m).
IV.A.9. D EFINITION . The opposite ring of R is ( R, +, •op , 0, 1) =:

Rop , where r ·op s := sr.
So r gives a homomorphism r : Rop → End( M), where M con-

tinues to denote the abelian group ( R, +, 0). We can write (with
[Jacobson])
Rr := im(r) ⊆ End( M ) , R` := im(`) ⊆ End( M).
Recalling that C A ( B) denotes the centralizer of B in A, we have
IV.A.10. P ROPOSITION . Rr = CEnd( M) ( R` ), and R` = CEnd( M) ( Rr ).

2i.e. the statement that every monoid G is a submonoid of a monoid of transfor-
mations of a set (in particular, G itself).
P ROOF. `r rs = rs `r is clear, so Rr ⊂ CEnd( M) ( R` ) etc. Conversely,

suppose η ∈ End( M) is such that η `r = `r η for every r ∈ R. Then
η (m) = η (m1) = η (`m (1)) = `m (η (1)) = m · η (1) = rη (1) (m) (∀m)
=⇒ η = rη (1) ∈ Rr . (Note that η (1) need not be 1 since η is merely

a homomorphism of abelian groups.)
The basis of the discussion above is viewing R as left and right
R-module. If we instead let M be an arbitrary left R-module, we see
that
L : R −→ End( M)
r 7−→ {m 7→ rm}
yields a ring homomorphism. Conversely, given a ring homomor-

phism
θ : R → End( M),
with M an abelian group, one verifies IV.A.1(i)-(iv) as follows:
• θ lands in End( M ) =⇒ (i): r (m + m0 ) = rm + rm0 ;
• θ sends r + s to θ (r ) + θ (s) =⇒ (ii): (r + s)m = rm + sm;
• θ sends rs to θ (r ) ◦ θ (s) =⇒ (iii): (rs)m = r (sm); and
• θ sends 1R to 1End( M) =⇒ (iv): 1R m = m.
Similarly, if M is a right R-module, then
R : Rop −→ End( M )
r 7−→ {m 7→ mr }
produces a ring homomorphism; and the converse is left to the reader.

This proves the
IV.A.11. T HEOREM . Let R be a ring, M an abelian group. A left

R-module structure on M is equivalent to a ring homomorphism R →
End( M). A right R-module structure on M is equivalent to a ring homo-
morphism Rop → End( M).
From this point of view, the two notions are “the same” for a
commutative ring R because R = Rop .
198 IV. MODULES
For representations of G (cf. IV.A.2(g)), the homomorphism in

IV.A.11 takes the specific form of a ring homomorphism
F[ G ] −→ EndF (V )
which is induced by linearizing a group homomorphism
G → AutF (V ).
The right-hand sides here denote F-linear endo/auto-morphisms;

this constraint on the F[ G ]-module structure/G-action comes from
the assumption that g.( f v) = f ( g.v) in IV.A.2(g). If V is finite (say,
n) dimensional, then AutF (V ) ∼= GL(n, F).
IV.B. SUBMODULES AND HOMOMORPHISMS 199
IV.B. Submodules and homomorphisms
Let M be a (left) R-module.
IV.B.1. D EFINITION . An R-submodule N ⊆ M is an additive sub-

group closed under the “scalar multiplication” action of R.
IV.B.2. E XAMPLES . (continuation of IV.A.2)

(a) Given an F-vector space (= F-module), an F-submodule is a sub-
space (defined over F).
(b) Given an abelian group A (= Z-module), a Z-submodule is just
a subgroup.
(c) Regarding R as (left) R-module, the (left) R-submodules are pre-
cisely the (left) ideals.
(d) cf. IV.B.8(ii) below.
(e) Sub-vector bundles of a vector bundle V → M yield C ∞ (M)-
submodules.
(f) There are no proper nontrivial Mn (R)-submodules of Rn . (Why?)
(g) The sub-F[ G ]-modules of a representation V of G are the sub-
representations W ⊂ V — i.e. sub-F-vector spaces stabilized by G
(G (W ) ⊂ W).
(h) Given T ∈ EndF (V ) and regarding V as F[λ]-module via λv :=
T (v), an F[λ]-submodule is a subspace W ⊂ V stabilized by T (that
is, T (W ) ⊂ W).
IV.B.3. D EFINITION . Given a subset S ⊂ M, the R-submodule

generated by S is3
RhSi := {∑finite
s∈S rs s | rs ∈ R },
or equivalently the intersection of all R-submodules containing S .

Just as for ideals, we define sums by
∑α Nα := Rh{ Nα }i = {∑finite
α nα | nα ∈ Nα }.
3See IV.A.3 for finite generation.
200 IV. MODULES
“Finite” means that, while the index set may be infinite, only finitely
many terms in each sum can be nonzero.
IV.B.4. P ROPOSITION -D EFINITION (Quotient R-modules). Given

an R-submodule N ⊂ M, the quotient group M/N has the structure of an
R-module.
P ROOF. Define r m̄ = r (m + N ) := rm + N = rm. This is well-

defined since for m0 − m ∈ N, RN ⊂ N =⇒ r (m − m0 ) ∈ N =⇒
rm + N = rm0 + rm − rm0 + N = rm0 + r (m − m0 ) + N = rm0 + N.
Now check the properties in IV.A.1 for M/N, e.g. (rs)m̄ = (rs)m =
r (sm) = r (sm) = r (sm̄).
IV.B.5. D EFINITION . A homomorphism of R-modules η : M →

M0 is a homomorphism of abelian groups intertwining the action of
R: η (rm) = rη (m). The set of all such is denoted HomR ( M, M0 ),
and EndR ( M ) := HomR ( M, M ). The usual words on injective and
surjective homomorphisms and isomorphisms apply.
IV.B.6. P ROPOSITION . EndR ( M) is a ring, and HomR ( M, M0 ) an

abelian group.
P ROOF. The sum and (if defined) composite of two R-intertwining

homomorphisms also intertwine the action of R.
Just as in the case of groups and rings, kernels and images define
subobjects.
IV.B.7. P ROPOSITION . The kernel ker(η ) ⊆ M and image im(η ) ⊆

M0 are R-submodules.
P ROOF. R ker(η ) ⊂ ker(η ) since η (m) = 0 =⇒ η (rm) = rη (m) =

0; and Rim(η ) ⊂ im(η ) since rη (m) = η (rm).
IV.B.8. E XAMPLES . (i) The inclusion ı : N ,→ M of a submod-

ule and the projection ν : M M/N to the quotient module are
R-module homomorphisms.
def.
(ii) A (f.g.) R-module M is free ⇐⇒ M ∼ = Rn (as R-module) for
some n ∈ N. (If R is noncommutative, n need not be unique.)
A submodule of a “free” R-module need not be free unless R is a
PID: for instance, Z6 has 3Z6 as sub-Z6 -module.
(iii) Consider the cyclic (sub)module
Rx := {rx | r ∈ R} ⊂ M
for some x ∈ M. We have
µ x : R Rx
r 7→ rx ,
which satisfies r 0 µ x (r ) = r 0 rx = µ x (r 0 r ) and µ x (r + r 0 ) = (r + r 0 ) x =

rx + r 0 x = µ x (r ) + µ x (r 0 ) hence is an R-module homomorphism.
Define the annihilator
ann( x ) := ker(µ x ) ⊆ R,
which is a (left) R-submodule of R hence a left ideal. If R = Z, ann( x )

is the principal ideal of Z generated by ord( x ), the order of x in M.
(iv) An F-module homomorphism between two F-vector spaces is
an F-linear transformation.
In another similarity to groups and rings, the various isomor-

phism theorems hold for R-modules. In particular, we have the
IV.B.9. F UNDAMENTAL T HM . OF R-M ODULE H OMOMORPHISMS .

Any given R-module homomorphism η : M → M0 factors as follows:
η
M / M0
9
ν $$ , η̄
M/ ker(η )
In particular, im(η ) ∼
= M/ ker(η ) as R-modules.
202 IV. MODULES
P ROOF. As usual, η̄ is well-defined and the diagram commutes

because of the abelian group result. We need to check that η̄ is an R-
module homomorphism: by definition, η̄ (m̄) := η (m), and r η̄ (m̄) =
¯ ) = η̄ (r m̄).
rη (m) = η (rm) = η̄ (rm
IV.B.10. E XAMPLE . Let x ∈ M be given. Applying this Theorem

to IV.B.8(iii) (with η := µ x ) gives Rx ∼
= R/ann( x ). Note that, if M is
free, and R is a domain, then ann( x ) = {0} and so Rx ∼ = R (is free).
Free R-modules. Let’s go into some more detail on these. First
note that Rn = Rhe1 , . . . , en i, where as before ei is the ith standard
r1

basis (column) vector. Moreover, if we write 0Rn = ∑i ri ei = .. ,
.
rn
then ri = 0 for all i. Consequently, there is exactly one way of ex-
pressing each element of Rn as ∑in=1 ri ei ; and so given any R-module
M and u1 , . . . , un ∈ M,
θ : Rn −→ M
(IV.B.11)
∑i ri ei 7−→ ∑i ri ui
is a well-defined R-module homomorphism. [Check: rθ (∑ ri ei ) =
r ∑ ri ui = ∑ rri ui = θ (∑ rri ei ) = θ (r ∑ ri ei ).]
IV.B.12. D EFINITION . A base for a (f.g., left) R-module is an or-

dered generating set m1 , . . . , mn (for some n ∈ N) such that
∑in=1 ri mi = 0 =⇒ ri = 0 (∀i ).
IV.B.13. P ROPOSITION . A (f.g.) R-module M is free ⇐⇒ M has a

base.
P ROOF. ( =⇒ ) is clear: use the standard base. For ( ⇐= ): given

a base {m1 , . . . , mn } ⊂ M, the homomorphism θ : Rn → M (sending
∑ ri ei 7→ ∑ ri mi ) is injective and surjective (by definition of “base”),
so that M ∼= Rn .
IV.B.14. T HEOREM -D EFINITION . Let M be a (f.g.) free R-module. If

R is commutative, then
rank( M) := “# of elements in a base for M”

is well-defined.
P ROOF. Let { f 1 , . . . , f n } and {e1 , . . . , em } (n ≥ m) be bases of M.

We have e j = ∑nk=1 a jk f k and
m n m
fi = ∑ bij e j = ∑ ∑ bij a jk f k
j =1 k =1 j =1
for some a jk , bij ∈ R. Since { f j } is a base, the displayed equality

gives ∑m j=1 bij a jk = δik (i, k ∈ {1, . . . , n }). Adding n − m columns
[resp. rows] of zeroes to the n × m matrix (bij ) [resp. m × n matrix
( a jk )] therefore yields n × n matrices satisfying
 a11 ··· a1n
 
b11 · · · b1m 0 · · · 0  .. ..

 .. . . 
.. .. .. am1 ··· amn 
BA :=  .  = 1n
 
. . .
 0. ··· 0. 
 
bn1 · · · bnm 0 · · · 0 .. ..
0 ··· 0
whence det( B) det( A) = 1 as R is commutative (cf. III.C.16). If n >

m then the rows/columns of zeroes make det( A) and det( B) zero, a
contradiction. So we have n = m.
IV.B.15. C OROLLARY. Let R be commutative, M a f.g. free R-module.

Then GLn ( R) acts transitively on the set of bases of M.
P ROOF. In the proof of IV.B.14, A sends { f j } to {ei } and det( A) ∈

R∗ =⇒ A ∈ GLn ( R) (invertible).
Conversely, if A is invertible (∃ B s.t. AB = 1n = BA) and { f j } is
a base, I claim that ei := ∑nj=1 aij f j is a base. First,
BA = 1n =⇒ ∑i bki ei = ∑ j (∑i bki aij ) f j = ∑ j δkj f j = f k

=⇒ {ei } generate M.
Second, if ∑i ri ei = 0 then
AB = 1n =⇒ 0 = iri ∑ j aij f j = ∑ j (∑i ri aij ) f j

[{ f j } base] =⇒ ∑i ri aij = 0 (∀ j)
=⇒ 0 = ∑i,j ri aij b jk = ∑i ri δik = rk (∀k),
so that {ei } is a base by IV.B.12.
204 IV. MODULES
Now let R be general and M, N be free right4 R-modules with

bases e = {e1 , . . . , em } resp. f = { f 1 , . . . , f n }.
IV.B.16. D EFINITION . The matrix of η ∈ HomR ( M, N ) relative to

e, f is
f [ η ]e : = ( aij ) i = 1, . . . , n
j = 1, . . . , m
where η (e j ) = ∑in=1 f i aij (with aij ∈ R). Also write

x1
e [ x ] :=
..
.
xm
5
for x = ∑m
j=1 e j x j ∈ M (with x j ∈ R), and similarly for y ∈ N.
IV.B.17. P ROPOSITION . f [η ]e · e [ x ] = f [η ( x )], where the dot is com-

puted by matrix-vector multiplication and the R-module structure.
P ROOF. This says that if the aij and x j are as above, then
(IV.B.18) ∑i f i (∑ j aij x j ) = η ( x ).
To check this, write LHS(IV.B.18) = ∑ j (∑i f i aij ) x j = ∑ j η (e j ) x j =
η ( ∑ j e j x j ) = η ( x ).
Writing [·] as a shorthand for f [·]e when the bases are understood,
we have
IV.B.19. P ROPOSITION . [·] : HomR ( M, N ) → Mn×m ( R) is an iso-

morphism of abelian groups.
P ROOF. It is clear that [η + η 0 ] = [η ] + [η 0 ]. Any A ∈ Mn×m ( R)

can be used to define η on the e, and this gives a homomorphism
with [η ] = A, proving surjectivity. Finally, [·] is injective since [η ]
defines η via IV.B.17.
4If R is commutative, these are just left R-modules by rm := mr. The main point
here is that I want the transpose of what [Jacobson] gets.
5In contrast, [Jacobson] uses row vectors.
Generalizing IV.B.17, if L is a free R-module with base {δ1 , . . . , δ` }

then the diagram
compose
(IV.B.20) HomR ( M, N ) × HomR ( L, M) / HomR ( L, N )
∼
= f [·]e ×e [·]δ ∼
= f [·]δ

matrix mult. /
Mn×m ( R) × Mm×` ( R) Mn×` ( R)
commutes. Since composition of maps is associative, (IV.B.20) =⇒

matrix multiplication is too.
P ROOF OF (IV.B.20). Let η ∈ HomR ( M, N ) and ζ ∈ HomR ( L, M).
Writing ( A)ik for the (i, k )th entry of a matrix A, we have
∑in=1 f i ( f [ηζ ]δ )ik = (ηζ )(δk ) = η (ζ (δk ))

= η (∑m m
j=1 e j (e [ ζ ]δ ) jk ) = ∑ j=1 η ( e j )(e [ ζ ]δ ) jk
= ∑m n
j=1 ( ∑i =1 f i ( f [ η ]e )ij )(e [ ζ ]δ ) jk
= ∑in=1 f i {∑m
j=1 ( f [ η ]e )ij (e [ ζ ]δ ) jk }
= ∑in=1 f i ( f [η ]e · e [ζ ]δ )ik .
Now use that { f i } is a base.
Applying this in the case M = N = L ( =⇒ m = n = `), we
have proved
IV.B.21. T HEOREM . Given a free right R-module M, we have an iso-
∼
=
morphism of rings e [·] := e [·]e : EndR ( M) −→ Mm ( R).
IV.B.22. R EMARK . A natural question is whether HomR ( M, N )
has the structure of an R-module. Let’s consider this in the left-R-
module case. If R is non-commutative, the answer is in general no:
given f ∈ HomR ( M, N ), we can try to define r f by6
(IV.B.23) (r f )(m) := r · f (m)

(where the RHS = f (rm) since f ∈ HomR ( M, N )). The problem
is that, in order for this r f to still lie in HomR ( M, N ), we need r 0 ·
6We occasionally use a “·” to indicate some (but not all) R-module actions, so as to
clarify the order of operations.
206 IV. MODULES
(r f )(m) = (r f )(r 0 m). But by (IV.B.23), this becomes r 0 · (r · f (m)) =

r · f (r 0 m), hence (using f ∈ HomR ( M, N )) r 0 r · f (m) = rr 0 · f (m),
which is clearly not true in general. On the other hand, if M and N
are right R-modules, and N also has a left R-module structure, then
Homright R-mod ( M, N ) will have a left R-module structure, defined
by7 (IV.B.23). Needless to say, all these delicate issues vanish if R is
commutative; but if you want to think about modules over matrix
rings or group rings, you have to face them.
Direct summands. We have been examining the special case of

M a “sum” of copies of R (as R-module). Let’s consider more general
“sums”:8
IV.B.24. D EFINITION . (i) Given R-modules { Mi }in=1 , their direct

sum is the R-module9 M1 ⊕ · · · ⊕ Mn with underlying abelian group
M1 × · · · × Mn and R-action by r (m1 , . . . , mn ) := (rm1 , . . . , rmn ).
(ii) Given R-module homomorphisms ηi : Mi → N and µi : Mi →
Ni , we define R-module homomorphisms
• ∑i ηi : M1 ⊕ · · · ⊕ Mn → N, by (m1 , . . . , mn ) 7→ ∑ ηi (mi ); and
• ⊕i µi : M1 ⊕ · · · ⊕ Mn → N1 ⊕ · · · ⊕ Nn , by
(m1 , . . . , mn ) 7→ (µ1 (m1 ), . . . , µn (mn )).
The following is (for n = 2, at least) reminiscent of the direct

product theorem for groups.
IV.B.25. T HEOREM . Let { Mi }in=1 be R-submodules of M. If

(i) [spanning] M = ∑ Mi and
(ii) [independence] M j ∩ ∑i6= j Mi = {0} (∀ j),
then M ∼ = ⊕in=1 Mi .
7Note that RHS(IV.B.23) will no longer be given by f (rm), since f is not assumed to
be a left R-module homomorphism (and I haven’t even assumed a left R-module
structure on M.
8For simplicity, all modules are henceforth left modules; though the results also
hold for right modules.
9Also written ⊕n M or just ⊕ M or ⊕ M .
i =1 i i i i
P ROOF. Let ηi : Mi ,→ M be the inclusions. By (i),
η : = ∑ i η i : ⊕ Mi → M
is surjective. By (ii), given xi ∈ Mi with 0 = ∑in=1 xi , we have
x j = ∑i6= j (− xi ) ∈ M j ∩ ∑i6= j Mi =⇒ xj = 0
for all j; hence η is injective.

∼
=
IV.B.26. R EMARK . (a) In this setting, where ∑i ηi : ⊕ Mi → M for
submodules Mi ⊂ M satisfying (i) and (ii), we shall write M = ⊕i Mi ,
and call M an internal direct sum (of these submodules).
(b) [Jacobson] has a converse result, which says that if M ∼ = ⊕ Mi
then (i) and (ii) hold (for the submodules arising from M1 × {0} and
{0} × M2 on the RHS); he also has an “associativity” result for ⊕.
(c) Applying the Fund. Thm. IV.B.9 to the projection π : M ⊕ N M
((m, n) 7→ m) produces an isomorphism ( M ⊕ N )/N ∼ = M.
(d) We can take infinite ⊕’s indexed by a set I . Elements are I -tuples
with all but finitely many entries zero. (Axiom of choice plays no role
here.)
We now consider the question of when it is possible to use direct

sums to “atomize” a given module. What is an “atom”?
IV.B.27. D EFINITION . A nonzero R-module M is irreducible (or

simple) if {0} and M are its only submodules.
IV.B.28. P ROPOSITION . M is irreducible ⇐⇒ M is cyclic with every

nonzero element as generator.
P ROOF. ( ⇐= ): no proper subset of M but {0} is closed under

the action of R.
( =⇒ ): If some x ∈ M\{0} has Rx 6= M then Rx is a nontrivial
proper submodule.
IV.B.29. C OROLLARY. M irreducible ⇐⇒ M ∼

= R/I with I a maxi-
mal (left) ideal of R.
208 IV. MODULES
P ROOF. Use an isomorphism theorem. (HW)
IV.B.30. S CHUR ’ S L EMMA . Given M1 , M2 irreducible R-modules, any

nonzero R-module homomorphism θ : M1 → M2 is an isomorphism.
P ROOF. (Assume M1 6= {0}.) Since ker(θ ) ⊂ M1 is a submodule,

either ker(θ ) = {0} or M1 , hence θ is injective or zero. If θ is injective,
then θ ( M1 ) ⊂ M2 is a nonzero submodule, so equals M2 , making θ
also surjective.
IV.B.31. C OROLLARY. If M is irreducible, then EndR ( M ) is a division

ring.
P ROOF. Every nonzero θ ∈ EndR ( M) is invertible, by Schur’s

Lemma.
You may wonder what happens if we have an irreducible sub-

module N ⊂ M — is it a direct summand, i.e. is there a “comple-
mentary” submodule N 0 so that M = N ⊕ N 0 ? In general, this is false
— consider 2Z ⊂ Z (as Z-module), which has no such “comple-
ment” — but it’s obviously true for finite-rank modules over a field
(i.e. vector spaces).
IV.B.32. D EFINITION . An R-module is semisimple if every sub-

module of M is a direct summand.
We now jump into a bit of deep water:
IV.B.33. T HEOREM . The following are equivalent for an R-module M:

(a) M is semisimple;
(b) M is isomorphic to a direct sum of irreducible R-modules; and
(c) M is the internal direct sum of some irreducible R-submodules.
P ROOF. (c) =⇒ (b): obvious

(b) =⇒ (a): Say M ∼ = ⊕i∈I Mi , with Mi irreducible, with N ⊂ M a
submodule. Invoking Zorn’s lemma, we let K ⊂ I be maximal with
respect to the property that (∑i∈K Mi ) ∩ N = {0}.
Given i0 ∈ K, Mi0 ⊂ (∑i∈K Mi ) + N. (Duh.)
Given i0 ∈/ K, maximality =⇒ ( Mi0 + ∑i∈K Mi ) ∩ N 6= {0}. So

there exist mi0 ∈ Mi0 and mK ∈ ∑i∈K Mi such that mi0 + mK =: n ∈
N \{0}. Note that necessarily mi0 6= 0, and so Mi0 = Rmi0 . Since
mi0 = n − mK , we find Mi0 ⊂ RmK + Rn ⊂ (∑i∈K Mi ) + N.
Thus for every i0 ∈ I , Mi0 is contained in (∑i∈K Mi ) + N, hence
that M = ∑i∈K Mi + N. By IV.B.25, we have M ∼ = (∑i∈K Mi ) ⊕ N.
Conclude that M is semisimple.
(a) =⇒ (c): Let N ⊂ M be a submodule, and n ∈ N \{0}. Invok-
ing Zorn again, we let L ⊂ N be maximal with n ∈ / L. Since M
is semisimple, we have M = L ⊕ L0 ; intersecting with10 N gives
0
N = L ⊕ ( L00 ∩ N ) =: L ⊕ L0 .
Suppose L0 is not simple: then it has a proper nonzero submodule
L10 ; applying semisimplicity of M and “intersecting” as above, we
get L0 = L10 ⊕ L20 . Hence N = L ⊕ L10 ⊕ L20 . If n ∈ L ⊕ Li0 (i = 1, 2)
then we can write n = `1 + `10 = `2 + `20 (with `1 , `2 ∈ L). But then
`1 − `2 = `20 − `10 ∈ L ∩ L0 = {0} =⇒ `10 = `20 ∈ L10 ∩ L20 = {0} =⇒
n ∈ L, a contradiction. So n ∈ / L ⊕ L20 (swapping 1 and 2 if needed),
which violates the maximality of L, another contradiction! Conclude
that L0 is simple.
So we have shown that every submodule N of M contains a simple
direct summand.
Next, let { Mi | i ∈ I} be a set of simple submodules of M, maxi-
mal (Zorn again) with respect to the property that
MI : = ∑ Mi = ⊕i∈I Mi .
i ∈I
By semisimplicity of M, M = MI ⊕ M0 . Suppose M0 6= {0}. Then

the italicized statement above produces a direct sum decomposition
M0 = L ⊕ L0 with L0 simple. But this contradicts maximality of
10One has to be a bit careful here: the classic example from linear algebra is that
R2 is the direct sum of the two coordinate axes, a decomposition that you certainly
can’t “intersect” with (say) the diagonal. The difference here is that N contains one
of the summands (namely, L). So given m = ` + `00 ∈ L ⊕ L00 = M, if it happens
that m ∈ N then `00 = m − ` ∈ N (since m, ` ∈ N). So `00 ∈ L00 ∩ N as desired.
210 IV. MODULES
{ Mi }i∈I (since you can now throw in L0 ). So M = MI = ⊕i∈I Mi

is a direct sum of irreducibles as desired.
This brings us to one of the topics we shall explore next semester:
IV.B.34. D EFINITION . R is a (left) semisimple ring ⇐⇒ all (left)

R-modules are semisimple.
In particular, in representation theory it is paramount to know

when the representations of a group G are “completely reducible”
(to direct sums of irreducible representations).
IV.C. MODULES OVER A PID 211
IV.C. Modules over a PID
Let R be a principal ideal domain, and M an R-module. (Since R

is commutative, left vs. right is immaterial.) We begin with a simple
statement about generators and relations (which indeed has nothing
to do with R being a PID).
IV.C.1. P ROPOSITION . M is finitely generated ⇐⇒ M ∼

= Rn /K
(with K an R-submodule of Rn ).
P ROOF. ( ⇐= ): M = Rhē1 , . . . , ēn i.

( =⇒ ): If M = Rh x1 , . . . , xn i (i.e. M is f.g.), then define η : Rn M
by ∑i ri ei 7→ ∑i ri xi ; by the Fundamental Thm., M ∼ = Rn / ker(η ).
The following generalizes II.K.4 (Z-module case) and a standard

linear algebra result (F-module case).
IV.C.2. T HEOREM . Any submodule K of Rn is isomorphic to Rn0 , for

some n0 ≤ n.
P ROOF. The result holds trivially for n = 0.

Assume it “for n − 1” and consider the projection π : Rn R
sending ∑i ri ei 7→ r1 , with ker(π ) ∼
= R n −1 .
If π (K ) = {0} then we’re done by induction. (Why?)
Otherwise, as an R-submodule of R, π (K ) is an ideal — in a PID.
So we have π (K ) = (r) for some r ∈ R\{0}, and moreover this
r = π (κ ) for some κ ∈ K. Observe that ann(κ ) = {0} since κ ∈ Rn
and R is a domain.
Now any k ∈ K can be written in the form
π (k) π (k)
k = (k − r κ) + r κ ∈ (ker(π ) ∩ K ) + Rκ,
π (k) π (k)
since π (k − r κ ) = π (k ) − r r = 0. Moreover, we have that
(ker(π ) ∩ K ) ∩ Rκ = {0}, as rκ ∈ ker(π ) =⇒ 0 = π (rκ ) = rπ (κ ) =
rr =⇒ r = 0 =⇒ rκ = 0. By the direct-sum theorem IV.B.25, it
now follows that
K = (ker(π ) ∩ K ) ⊕ Rκ.
212 IV. MODULES
Applying the inductive assumption to the submodule (ker(π ) ∩ K ) ⊂

Rn−1 , it takes the form Rm0 for some m0 ≤ n − 1. Finally, since
ann(κ ) = {0}, Rκ ∼= R; and K ∼
= R m0 +1 .
We want to get from “ugly” presentations M ∼ = Rn /K to “nice”
ones like Rx1 ⊕ · · · ⊕ Rxk ⊕ Rr . The starting point is to write K with
respect to a base. More precisely, given a submodule K ⊂ Rn , we
∼
=
may compose the isomorphism Rn0 → K guaranteed by IV.C.2 —
or, more generally, any surjective homomorphism Rm K — with the
inclusion K ,→ Rn to get an R-module homomorphism
Rm → Rn
θ
(IV.C.3)
e0j 7→ θ (e0j ) =: a j ( j = 1, . . . , m)
whose image is K.
IV.C.4. D EFINITION . The n × m matrix of θ with respect to the

standard bases ({e0j } of Rm , {ei } of Rn ) is
↑ ↑

e [ θ ] e0 : = A : = a1 ··· am .
↓ ↓
A is called a relations matrix for M := Rn /K, and we can write11

Rhe1 , . . . , en i
M∼
in
= Rn /θ ( Rm ) = Rn /A · Rm = .
bases Rh∑i a1i ei , . . . , ∑i aim ei i
Our hopes are pinned on transforming A into something nice, for
which we have to revisit the elementary matrices from §III.C. Recall
that
GLn ( R) := invertible n × n matrices with entries in R

= n × n matrices with entries in R and det ∈ R∗ ,
e.g. for R = Z we need det = ±1.
IV.C.5. E XAMPLE . The elementary matrices of (III.C.4) belong to

GLn ( R). We will need some notation for these:
11The notation Rh· · ·i simply means all R-linear combinations of the elements in-
j
side the angle brackets; ai means the ith entry of a j .
(n)
• Tij ( a) := 1n + aeij , where a ∈ R, has inverse Tij (− a):
(n) (n)
• Pij := 1n + eij + e ji − eii − e jj = Pji is its own inverse.
(n) (n)
• Di (u) := 1n + (u − 1)eii , where u ∈ R∗ , has inverse Di (u−1 ).
IV.C.6. P ROPOSITION . Let A be an n × m relations matrix for (a f.g.

R-module) M. Let P ∈ GLn ( R), Q ∈ GLm ( R). Then PAQ is a relations
matrix for M.
P ROOF. P corresponds to a change of basis {ei } 7→ {ẽi } for Rn ,

and Q to a change of basis {e0j } 7→ {ẽ0j } for Rm : that is, P = ẽ [idRn ]e
(i.e. ek = ∑i pik ẽi ), while Q = e0 [idRm ]ẽ0 (i.e. ẽ0` = ∑ j q j` e0j ). So
PAQ = ẽ [idRn ]e · e [θ ]e0 · e0 [idRm ]ẽ0 = ẽ [θ ]ẽ0
is just a matrix of θ with respect to different bases of Rm and Rn .
In practice, you may not need to keep track of how the bases
change, but just to find some PAQ which is in a nice form (the nor-
mal form below). At the risk of beating elementary matrices into the
ground:
IV.C.7. E XAMPLE . Let’s see how to compute various “PAQ’s”.

Here A is any n × n matrix over R.
To get this matrix from A do the following (to A)

(n)
Tij ( a) · A add a×(row j) to (row i)
(m)
A · Tij ( a) add a×(column i ) to (column j)
(n)
Pij · A swap rows i and j
(m)
A · Pij swap columns i and j
(n)
Di ( u ) · A multiply row i by u
(m)
A· Di ( u ) multiply column i by u
The operations on the RHS of the table will be called elementary

operations (EOs).
214 IV. MODULES
The structure theorem for Z-modules. We are now going to state

and prove the main results for abelian groups (R = Z). Later we will
generalize the proof, first to the case where R a Euclidean domain,
and then to the general PID case.
IV.C.8. L EMMA -D EFINITION . Every A ∈ Mn×m (Z) can be trans-

formed by EOs into a matrix in normal form:
! ! )
D 0 D henceforth summarized
, ( D | 0) , , D, or 0 D 0
!
0 0 0 by “
0 0
”,
with D = diag(d1 , d2 , . . . , dk ) a diagonal matrix and d1 | d2 | · · · | dk .
IV.C.9. T HE F UNDAMENTAL T HEOREM OF F INITELY G ENERATED

A BELIAN G ROUPS /Z- MODULES (FTFGAG). Any finitely generated
abelian group G may be expressed uniquely in the form
(IV.C.10) Zr
Zd1 × · · · × Zdk × |{z}
| {z }
Gtor G/Gtor
where di ≥ 2 and d1 | d2 | · · · | dk .
E ASY PART OF PROOF ( ASSUMING L EMMA IV.C.8). Putting every-

thing together:
• G finitely generated =⇒ G ∼
= Zn /K with relations matrix A.
• Lemma IV.C.8 =⇒ EOs convert A to normal form.
• Example IV.C.7 =⇒ the resulting matrix is of the form PAQ with
P ∈ GLn (Z) and Q ∈ GLm (Z).
• Prop. IV.C.6 =⇒ PAQ is a relations matrix for G.
Conclude that
!
G ∼
= Zn /( PAQ)(Zm ) = Zn /
D 0
Zm
0 0
Z h X1 , . . . , X n i
= = Zd1 × · · · × Zd k × Zn − k ,
h d 1 X1 , . . . , d k X k i
where r = n − k is the number of complete rows of zeroes in the

normal form.12
IV.C.11. D EFINITION . In IV.C.9, r is the rank (of the free part) of

G, and d1 , . . . , dk the torsion exponents (or invariant factors) of G.
(G is finite ⇐⇒ r = 0.)
IV.C.12. E XAMPLE . Consider

Zh X, Y, Z i Z3
G := = .
h11X − 21Y − 10Z, X − 6Y − 5Z i K
Clearly K ∼
= Z2 , and in the “standard” bases (cf. IV.C.4) we have
 
11 1
A = −21 −6 .
 
−10 −5
Applying EOs, we reduce to normal form:
add (−11)×(col. 2) 0 1
subtract (row 2)
0 1
subtract (row 1)
0 1

/ 45 −6 / 45 −6 / 45 −6
to (col. 1) 45 −5 from (row 3) 0 1 from (row 3) 0 0
 
1 0
add 6×(row 1)
0 1
swap
/ 45 0 / 
0 45,

to (row 2) 0 0 cols. 1 and 2
0 0
concluding that d1 = 1, d2 = 45, r = 1, and

Zh X̃, Ỹ, Z̃ i
G ∼
= = Z45 × Z.
h X̃, 45Ỹ i
We now return to the proofs.
12Here I am writing { X } for the base of Zn corresponding to the {e } in the first

i i
line.
216 IV. MODULES
P ROOF OF IV.C.8. Let A ∈ Mn×m (Z), and write
aij := (i, j)th entry of A, Rs := sth row of A,
and Ct := tth column of A.

A row or column will be said to be cleared if it has only one nonzero
entry. As we change A by EOs, it will (at intermediate steps) have
the form  
d1 0

 ..
. 0


,
 
 0
 d k 
A0

0

with d1 | d2 | · · · | dk , and A0 not of the form

 
d ←0→
 
 ↑ 

 0
 ∗ .


↓
We will write ast | A0 if ast divides all entries of A0 . Recall that for
q ∈ Q, the floor function bqc is defined to be the greatest integer less
than or equal to q.
On the next page, we present an algorithm for reducing A to nor-
mal form. The goal is to reach (4) and reduce the size of A0 (i.e.
increase k by 1). Since one either progresses all the way around the
outer semicircle ((1) → (2) → (4)) or reduces | ast | upon return-
ing to (0) (which cannot reduce indefinitely!), the algorithm termi-
nates.
EO Normalization Algorithm (R = Z):
Input A
(n × m integer
matrix)
(1)
Output
! “Reduce Ct mod ast ”:
D 0 replace each Ri by
0 0 A0
=0
a
Ri − b astit c Rs .
d
re
c lea
t
(0) no
Ct
Take ast :=
nonzero entry of A0
Ct cleared
with smallest absolute
value. Multiply Rs by
0 )
A sign of ast (to make
ce
pla C
it positive).
(re d, R t c
are s no leare
l e tc d
(4) c 0 lea ,
Rs red
, |A (2)
Ct , Rs cleared, ast - A0
Ct a st
Move ast to
upper left of A0 : swap
“Reduce Rs mod ast ”:
Rs with 1st row of A0 ,
replace each Cj by
then swap Ct with 1st a
Cj − b astsj cCt .
column of A0 .
,
ed
ar
(3)
e
0
cl
-A
s
R
t
as
ast - some aij ,

so reduce aij mod ast :
add Rs to Ri , and then

a
subtract b astij cCt
from Cj
Rs cleared, ast | A0
218 IV. MODULES
P ROOF OF UNIQUENESS IN IV.C.9. Suppose that

(†)
= Z d 1 × · · · × Z d k × Zr ∼
G ∼ = Z e1 × · · · × Z e ` × Z s ,
where d1 | · · · | dk and e1 | · · · | e` (with di , e j ≥ 2). We must show
that r = s, k = `, and d j = e j (∀ j = 1, . . . k).
First, because the LHS and RHS of (†) are isomorphic groups.
they have isomorphic torsion and free parts:
(a) Zd1 × · · · × Zdk ∼

= Z e1 × · · · × Z e ` , (b) Zr ∼
= Zs .
Now (b) =⇒ the “cokernels” of multiplication by 2 are the same:
Zr ∼ Z s
= =⇒ (Z/2Z)r ∼
= (Z/2Z)s =⇒ 2r = 2s
2 · Zr 2 · Zs
whence r = s.13
Next, let Am ( G ) denote the number of elements of order dividing
m; then (a) =⇒ Ae1 (Ze1 × · · · × Ze` ) = Ae1 (Zd1 × · · · × Zdk ). By
an easy calculation, this yields
gcd(e1 , e1 ) · gcd(e1 , e2 ) · · · gcd(e1 , e` ) = gcd(e1 , d1 ) · · · gcd(e1 , dk )
hence
k
e1` = ∏ gcd(e1 , d j ) ≤ e1k ,
j =1
from which we conclude that ` ≤ k. A symmetric argument shows
` ≥ k, so ` = k; in particular, the above inequality is an equality
so that gcd(e1 , d j ) = e1 (∀ j) =⇒ e1 | d j (∀ j). Again, a symmetric
argument (taking Ad1 on both sides of (a)) shows d1 | e j (∀ j). But
then d1 | e1 and e1 | d1 =⇒ e1 = d1 .
Repeating the argument starting with
A e2 ( Z e1 × · · · × Z e k ) = A e2 ( Z d 1 × · · · × Z d k )
gives
k k
gcd(e2 , e1 ) · ∏ gcd(e2 , e j ) = gcd(e2 , d1 ) · ∏ gcd(e2 , d j )
j =2 (=e1 ) j=2
13If you prefer, you can argue using II.K.4 that r ≤ s and s ≤ r.
k k
=⇒ e2k−1 = ∏ gcd(e2 , e j ) = ∏ gcd(e2 , d j ) ≤ e2k−1 .
j =2 j =2
Clearly the inequality is an equality, and so gcd(e2 , d j ) = e2 hence
e2 | d j for each j. On the other hand, taking Ad2 of both sides gives
d2 | e j . So d2 | e2 and e2 | d2 =⇒ d2 = e2 .
Continue in this manner until you get all d j = e j .
Using the Chinese Remainder Theorem to decompose the Zd j
factors in (IV.C.10) yields the
IV.C.13. C OROLLARY (p-primary version of FTFGAG). Any finitely
generated abelian group G may be expressed (uniquely up to rearrangement
of factors) in the form
Z p r 1 × · · · × Z p r k × Zr ,
1 k
where the { pi } are not-necessarily-distinct primes.

IV.C.14. R EMARK . The abelian groups of order pn (p prime) are
in 1-to-1 correspondence with the partitions of n:
n = n1 + · · · + n k ( n1 ≤ · · · ≤ n k ) ←→ Z pn1 × · · · × Z pnk .
Together with IV.C.13, this allows you to find all abelian groups of a
given order: e.g., for order 360 = 23 32 5, we have
G∼
= {Z23 or (Z21 × Z22 ) or (Z21 × Z21 × Z21 )}
× {Z32 or (Z31 × Z31 )} × Z5 .
IV.C.15. E XAMPLE . Let’s see how to transform a more compli-
cated matrix than the one in IV.C.12 into normal form, by applying
the EO Normalization Algorithm. (You won’t need to follow the
algorithm this precisely in working problems. The point of going
through this example is to know what to do if you get stuck!)
 
4 −10 −2 20 30
0 28 −2 −60 90 
A= .
 
3 −3 −2 6 −9 
7 −7 −4 14 −21
220 IV. MODULES
(0) A0 = A, ast = a33 = −2. Changing the sign of R3 yields

!
4 −10 −2 20 30
0 28 −2 −60 90
.
−3 3 2 −6 9
7 −7 −4 14 −21
(1) Reduce C3 mod 2 (replace R1 , R2 , R4 by R1 + R3 , R2 + R3 , R4 + 2R3 ):

!
1 −7 0 14 39
−3 31 0 −66 99
.
−3 3 2 −6 9
1 −1 0 2 3
(2) Reduce R3 mod 2:

!
1 −7 0 14 39
−3 31 0 −66 99 .
1 1 2 0 1
1 −1 0 2 −3
Since R3 is not cleared, we must return to (0):

(0) ast = a11 = 1.
(1) Reduce C1 mod 1:
!
1 −7 0 14 39
0 10 0 −24 216
.
0 8 2 −14 −38
0 6 0 −12 −42

 
1 0 0 0 0
 
 0
 10 0 −24 216 
.
 0

8 2 −14 −38 

0 6 0 −12 −42
which displays our new 3 × 4 A0 . Step (4) does nothing.

(0) ast = a33 = 2.
(1) done.
 
1 0 0 0 0
 
 0
 10 0 −24 216 
.
 0 0 2 0 0 
 
0 6 0 −12 −42
(4) Swap m33 to top left position in A0 :

 
1 0 0 0 0
 
 0 2 0 0 0

 
 0

0 10 −24 216 

0 0 6 −12 −42
and reset A0 to be the smaller 2 × 3 matrix.

(0) ast = a43 = 6.
(1) Reduce C3 mod 6 (subtract R4 from R3 ):
 
1 0 0 0 0
 
 0 2 0 0 0 .


 0

0 4 −12 258 

0 0 6 −12 −42
Since C3 is not cleared, we return to

(0) ast = a33 = 4.
(1) Reduce C3 mod 4 (subtract R3 from R4 ):
 
1 0 0 0 0
 
 0 2 0 0 0 .


 0

0 4 −12 258


0 0 2 0 −300
Good grief! C3 is still not cleared!

(0) ast = a43 = 2.
(1) Reduce C3 mod 2:
 
1 0 0 0 0
 
 0 2 0 0 0 .


 0

0 0 −12 858


0 0 2 0 −300

 
1 0 0 0 0
 
 0 2 0 0 0 .


 0

0 0 −12 858 

0 0 2 0 0
(4) Swap a43 to the top left position in A0 :

 
1 0 0 0 0
 
 0 2 0 0 0 
 
 0 0 2 0 0 
 
0 0 0 −12 858
and reset A0 to be the smaller 1 × 2 matrix.

(0) ast = a44 = −12. Change the sign, bypass (1), and
222 IV. MODULES

 
1 0 0 0 0
 
 0 2 0 0 0 
 .
 0 0 2 0 0 
 
0 0 0 12 6
Since R4 is not cleared, we return to

(0) ast = a45 = 6.
 
1 0 0 0 0
 
 0 2 0 0 0 
 .
 0 0 2 0 0 
 
0 0 0 0 6
(4) Swap the last two columns and replace A0 :

 
1 0 0 0 0
 
 0 2 0 0 0 

 0
 = ( D | 0).
 0 2 0 0 

0 0 0 6 0
At last, we arrive at the normal form!

The structure theorem in the general case. Again let R be a PID.
IV.C.16. L EMMA -D EFINITION . Every A ∈ Mn×m ( R) can be trans-

formed by invertible row and column operations14 into a matrix in normal
form  
d1



..
.

 0
 =: nf( A)

dk

 
0
 
0
where the invariant factors d1 | · · · | dk are unique up to units. (The
matrix nf( A) itself is thus well-defined up to units.)
P ROOF. We break this into two parts: existence and uniqueness.

Step 1A : Reduction to normal form for R a Euclidean domain.
Let δ : R\{0} → Z>0 be a Euclidean function. We describe how
to modify the EO Normalization Algorithm above:
(0’) Take ast to the nonzero entry of A0 with smallest δ.
Ct Cj
· · · · ·
 
·
Ri  ait · aij ·
· · · · ·
 
 
Rs· ast · asj ·
· · · · ·
(1’) Subtract (for each i) qRs from Ri , where ait = qast + r (replaces
ait by r, with δ(r ) < δ( ast )).
(2’) Subtract (for each j) q̃Ct from Cj , where asj = q̃ast + r̃ (replaces
asj by r̃, with δ(r̃ ) < δ( ast )).
(3’) (a) Add Rs (cleared) to Ri ; then (b) subtract q0 Ct from Cj where
aij = q0 ast + r 0 (replaces aij by r 0 , with δ(r 0 ) < δ( ast )).
(4’) Swap ast to the upper left of A0 .
14i.e. A 7→ PAQ, P and Q invertible over R. EOs will not in general be enough,
but suffice for Euclidean domains.
224 IV. MODULES
Again one either proceeds all the way around the outer semicircle
((10 ) → (20 ) → (40 )), or reduces δ( ast ), so the process must termi-
nate.
Step 1B : Reduction to normal form in general.
Let ` : R\{0} → N be the length function. (Since R is a PID,
R is a UFD, and this is well-defined.) For (0”), we take ast to be the
nonzero entry of A0 with smallest ` (e.g. a unit, if there is one). (4”) is
the same as (4’). We need replacements for (1’), (2’), and (3’)(b) when
ast - ait (resp. asj , aij ), since the Euclidean algorithm isn’t available.
In fact, EOs won’t suffice. Though (1”) [resp. (2”) and (3”)(b)]
will still be given by row [resp. column] operations, or (equiva-
lently) left- [resp. right-]multiplication by invertible matrices, the
operations/matrices involved are of a slightly more general nature.
For (2”), here is what we can do.15 Set a := ast , b := asj , and let
x, y ∈ R be such that
xa + yb = d := gcd( a, b),
z := db , w := − da . Right-multiplication by
Ct Cj
1 
..
 . 
 1 
Rt  x z
 

 1 
..
 
.
 
 
 1 
Rj  y w
 

1
 
..
 
.
1
replaces . . . Ct Cj ast asj

by . . . xCt + yCj zCt + wCj xast + yasj = d zast + wasj = 0
15The analogues for (1”) and (3”)(b) are essentially the same and left to you.
which may “undo” our clearing of Cj .16 But this is not a problem,
as it creates a new entry (“d” in the (s, t) place) with length `(d) <
`( ast ) (where `( ast ) was the previous shortest length). So as before,
the minimal length is reduced each time we return to (0”) without
passing through (4”) and reducing the size of A0 .
Step 2 : Uniqueness of the invariant factors.
Define ∆i ( A) := gcd{i × i minors of A} and
r( A) := max{i | ∆i ( A) 6= 0} (“determinantal rank”).
By multilinearity of determinants, any i × i minor of PAQ (where

P ∈ Mn×n ( R), Q ∈ Mm×m ( R)) is an R-linear combination of i × i
minors of A. Hence ∆i ( A) | ∆i ( PAQ) in R. But if P, Q are invertible,
this applies in reverse and
∆i ( A) ∼ ∆i ( PAQ).
Now suppose
   
d1 d10


PAQ = 

..
. 0  , 0 0


P AQ = 

..
. 0  .
 dk   d0k 
0 0
   
0 0
On the one hand, direct computation implies
(
∆i ( PAQ) = d1 · · · di
(∀i ).
∆i ( P0 AQ0 ) = d10 · · · di0
On the other, ∆i ( PAQ) ∼ ∆i ( A) ∼ ∆i ( P0 AQ0 ) (∀i). We conclude that
di ∼ di0 (∀i) and k = r( A) = `.
IV.C.17. E XAMPLE . For an n × n matrix A over F[λ] (F a field),

we can take
∆n−1 ( A) = monic gcd of entries of adj( A)
16This was already a feature of the original Step (3), though not of Step (2).
226 IV. MODULES
and
∆n ( A) = det( A)/(coefficient of highest power of λ)
since we are free to multiply the ∆i by units.

Suppose B is an n × n matrix over F, and A = λ1n − B. The
characteristic polynomial of B is
n
p B (λ) = ∆n ( A) = ∏ d i ( A ).
i =1
We will show (later) that

∆n ( A)
dn ( A) =
∆ n −1 ( A )
is the minimal polynomial m B (λ) of B. For instance, consider
 
1 1 1
B = 1 1 1 .
 
1 1 1
We reduce A to normal form with row and column operations:
   
λ − 1 −1 −1 −1 λ − 1 −1
A = λ13 − B =  −1 λ − 1 −1  7→ λ − 1 −1 −1 
   
−1 −1 λ − 1 −1 −1 λ − 1
     
1 0 0 1 1
7→ 0 λ2 − 2λ −λ 7→  −λ λ2 − 2λ 7→  λ .
     
0 −λ λ 0 λ2 − 3λ λ2 − 3λ
| {z }
=nf( A)
Conclude that the invariant factors of A are d1 ( A) = 1, d2 ( A) = λ,
and d3 ( A) = λ2 − 3λ; the last of these is indeed the minimal polyno-
mial of B:
   
3 3 3 1 1 1
B2 − 3B = 3 3 3 − 3 1 1 1 = 0.
   
3 3 3 1 1 1
We are finally ready to state and prove our main result:

IV.C.18. T HE S TRUCTURE T HEOREM FOR F INITELY G ENERATED

M ODULES O VER A PID. Any f.g. module M over R may be expressed
(up to isomorphism) uniquely in the form
(IV.C.19) R/(δ1 ) ⊕ · · · ⊕ R/(δ` ) ⊕ Rt ,
/ R∗ and δ1 | · · · |δ` .
where the δi ∈
More precisely, M is an internal direct sum of cyclic modules:
(
M = Rz1 ⊕ · · · ⊕ Rzs (zi ∈ M)
(IV.C.20)
where ann(z1 ) ⊃ · · · ⊃ ann(zs ) ;
and the annihlator ideals (hence also the number s) are uniquely deter-
mined.
As we saw in the Z-module case, the uniqueness part does not

follow from the uniqueness of the di in the normal form for A. (There
are obviously many presentations Rn /K of M, with different n.) What
we can do immediately is the existence part:
P ROOF OF (IV.C.19)-(IV.C.20) (E XISTENCE OF DECOMPOSITION ).

We have
M∼
= Rn /K ∼
= Rn /θ ( Rm ) ∼
= Rn /A· Rm ∼
= Rn /PAQ· Rm ,
n m
R , R. By IV.C.16
with the last step given by change of bases for
D 0
we may arrange to have PAQ = nf( A) = 0 0
(with D =
diag(d1 , . . . , dk )) hence

M∼ n D 0
=R / 0 0
· Rm .
That is, there is an R-module homomorphism
ρ : Rn M
ei 7 → ρ ( ei ) = : x i
with kernel K = Rhd1 e1 , . . . , dk ek i ⊆ Rn , where d1 | · · · |dk ( =⇒

(d1 ) ⊃ · · · ⊃ (dk )).
228 IV. MODULES
Since ρ is surjective, M = ∑in=1 Rxi . We describe these sum-

mands: for any i, we have 0 = rxi (= rρ(ei ) = ρ(rei )) ⇐⇒ rei ∈ K.
• If i > k, rei ∈ K ⇐⇒ r = 0 hence ann( xi ) = {0} and Rxi ∼ = R.
∼
• If i < k, rei ∈ K ⇐⇒ di |r. So ann( xi ) = (di ) and Rxi = R/(di ).
Finally, 0 = ∑i ri xi = ρ(∑i ri ei ) =⇒ ∑i ri ei ∈ K (∀i ) =⇒ di |ri (∀i )
=⇒ each ri xi = 0. So the homomorphism ⊕i Rxi M is injective,
and M = ⊕in=1 Rxi ∼ = R/(d1 ) ⊕ · · · ⊕ R/(dk ) ⊕ Rn−k .
Now it may be that none, some, or all of the {di } are units; as-
sume that the units are d1 , . . . , dk0 (here 0 ≤ k0 ≤ k). Then Rxi = {0}
for i = 1, . . . , k0 . Writing ` := k − k0 , δi := dk0 +i , t := n − k,
s := n − k0 , and zi := xk0 +i yields the specific forms of the decom-
positon shown in IV.C.19 and (IV.C.20).
Uniqueness considerations. Finishing the proof of the structure

theorem requires some preliminary results about decomposing tor-
sion modules.
IV.C.21. D EFINITION . The torsion submodule of an R-module M

is
tor( M) := { x ∈ M | rx = 0 for some r ∈ R\{0}}.
M is a torsion module if M = tor( M).
IV.C.22. P ROPOSITION . A f.g. module M over a PID R is an internal

direct sum of the form tor( M ) ⊕ Rt .
P ROOF. By the existence part of the structure theorem (that we

have now proved),
M = Rz1 ⊕ · · · ⊕ Rzs ∼
= R/(d1 ) ⊕ · · · ⊕ R/(d` ) ⊕ |R ⊕ ·{z
· · ⊕ R.}
t copies
Given m = ∑i`=1 ri zi + ∑is=`+1 ri zi ∈ tor( M), there exists r ∈ R\{0}

such that 0 = rm = ∑i`=1 rri zi + ∑is=`+1 rri zi . Since M is a direct sum,
0 = (rri )zi for i = 1, . . . , s. But for i > `, ann(zi ) = {0} =⇒ rri = 0
=⇒ ri = 0 (as R is a domain). So tor( M) ⊂ R/(d1 ) ⊕ · · · ⊕ R/(d` ).
The reverse inclusion is clear.
IV.C.23. D EFINITION . Let p ∈ R be a prime. The p-primary com-

ponent of M is
A p ( M) := { x ∈ M | pk x = 0 for some k ∈ N}.

IV.C.24. L EMMA . Let p1 , . . . , p` be a list of distinct17 primes in R.
Then ∑i`=1 A pi ( M) = ⊕i`=1 A pi ( M ) (⊂ tor( M)).
P ROOF. By induction, it suffices to show that
A p1 ∩ ∑i`=2 A pi ( M) = {0}.
k
Given x in the LHS, we have p1k1 x = 0 = p2k2 · · · p` ` x for some k i ∈ N.
k
But as the primes are distinct, gcd( p1k1 , p2k2 · · · p` ` ) = 1. So there exist
k
m, n ∈ R such that x = 1x = (mp1k1 + np2k2 · · · p` ` ) x = 0.
IV.C.25. T HEOREM . Assume M is a f.g. torsion module over a PID R.
Then M = ⊕ p∈ R prime A p ( M ) ∼
e
= ⊕i R/( pi i ), where pi are not necessar-
ily distinct primes in R and ei ∈ Z>0 . Both direct sums are finite, which
is to say that A p ( M ) is nonzero for only finitely many18 primes.
P ROOF. We know M = ⊕kj=1 R/(d j ), d j ∈ R\{0} and d1 | · · · |dk .
e j`
Moreover, d j = ∏m `=1 p` (∀ j) for some list of distinct primes { p` }
(and {e j` ∈ N}). So we will almost be through if we can check that
e j`
R/(d j ) = ⊕m
`=1 R/ ( p` ).
(Note that d1 | · · · |dk =⇒ e1` ≤ e2` ≤ · · · ≤ ek` for each `.) By

induction, this reduces to the following module-theoretic version of
the Chinese Remainder Theorem:
(IV.C.26) R/( f g) ∼
= R/( f ) ⊕ R/( g) if ( f , g) = R.
To see this, let x be a generator of the LHS and rx an arbitrary el-
ement. Then ( f , g) = R =⇒ ∃ ri ∈ R with r1 f + r2 g = r =⇒
rx = r1 f x + r2 gx =⇒
R/( f g) = Rx = R f x + Rgx.
17This means non-associate: they don’t generate the same ideal.
18Again, we are thinking of primes “up to units”; or equivalently, in terms of the
corresponding prime ideals.
230 IV. MODULES
Next, g( f x ) = ( f g) x = 0 =⇒ ( g) ⊂ ann( f x ); while 0 = r ( f x )

=⇒ r f ∈ ( f g) =⇒ r f = r 0 f g =⇒ r = r 0 g =⇒ r ∈ ( g). So
Rfx ∼= R/( g), and similarly Rgx ∼
= R/( f ). Finally, y ∈ R f x ∩ Rgx
=⇒ gy = 0 = f y =⇒ y = 1y = (r10 f + r20 g)y = 0, finishing off
(IV.C.26).
So we have proved
finite
M∼
M e
= R/( pi i ),
i
and moreover the proofs of (IV.C.26) and (IV.C.19) show that the di-
rect sum is internal. Therefore, we are reduced to
(IV.C.27)
e j` e j`
If M = ⊕ j,` Rx j` = ⊕ j,` R/( p` ), then A p` ( M) = ⊕ j R/( p` ) (∀`).
Clearly one has “⊇” on the right. To see the reverse inclusion “⊆”,
e j`
we need A p` ( M) ∩ ⊕ j,`6=`0 R/( p` ) = {0}. But the “⊕ j,`6=`0 ” here
0
belongs to ∑`6=`0 A p` ( M), so we are done by the proof of Lemma
IV.C.24.
IV.C.28. R EMARK . We can view the isomorphism in the last the-

e
orem as an internal direct sum. The summands R/( pi i ) are called
e
primary cyclic submodules of M, and the pi i are the elementary di-
visors of M.
We are at last ready for the
P ROOF OF UNIQUENESS IN IV.C.18. Assume
M = Rz1 ⊕ · · · ⊕ Rzs = Rw1 ⊕ · · · ⊕ Rwr ,
with annihilators (invariant factors) d1 | · · · |ds resp. d10 | · · · |dr0 . The

last few annihilators in each list may be zero. The number of these
trivial annihilators is the same on each side, as M/tor( M ) has well-
defined rank (R is commutative). So we may assume that M = tor( M )
and all the di , d0j are nonzero.
Next, decompose all the Rzi resp. Rw j into sums of primary cyclic
submodules, viz.
e j` e0
M = ⊕` ⊕sj=1 R/( p` ) = ⊕` ⊕rk=1 R/( p`k` ).
If these factors are the same, there is only one way to put them back
together to get d1 | · · · |ds and d10 | · · · |dr0 , and this will prove they are
the same set of divisors. Since
e e0
A p` ( M) = ⊕sj=1 R/( p`j` ) = ⊕rk=1 R/( p`k` ),
we may assume that M = A p ( M) for a single prime p ∈ R.
Considering the filtration19 by R-submodules
M ⊃ pM ⊃ p2 M ⊃ · · · ,
pn M
each =: M(n) is an R/( p)-module (since pM(n) = 0). Since
p n +1 M
( p) is prime and R is a PID, ( p) is in fact maximal, and R/( p) a field,
making M(n) a vector space. Writing
0
M = A p ( M) = ⊕sj=1 R/( pe j ) = ⊕rk=1 R/( pek ) ,
we have
(
s s
( pn )/( pe j ) 0, if e j ≤ n
M (n) =
M M
ej = ( pn )
j =1
( p n +1 ) / ( p ) j =1 ( p n +1 )
, otherwise
(and also the same with s resp. e j replaced by r resp. ek0 ).

Let Dn resp. Dn0 be the number of e j resp. ek0 greater than n. Since
R/( p) → ( pn )/( pn+1 )

r̄ 7−→ r̄pn
is an isomorphism of R/( p)-modules, we find that

0
M (n) ∼
= ( R/( p)) Dn ∼
= ( R/( p)) Dn
as a vector space over the field R/( p). Hence Dn = Dn0 . Since n was
arbitrary, we conclude that (up to reordering) the e j and ek0 are the
same.
19a nested sequence of submodules, usually indexed by a set of integers.

232 IV. MODULES
IV.D. Applications to linear algebra
Let T ∈ EndF (V )\{0} be a nontrivial linear transformation of a

finite-dimensional vector space V over a field F. Take { xi }in=1 ⊂ V to
be a basis and B := x [ T ] to be the corresponding matrix, with entries
bij ∈ F. We have that V = ⊕in=1 Fxi = ∑in=1 F[λ] xi , where V has the
structure of an F[λ]-module by P(λ)v := P( T )v (for any polynomial
P(λ) ∈ F[λ]). Since F[λ] is not f.g. as an F-module, V must be a (f.g.)
torsion F[λ]-module.
We have a short-exact sequence
η
K := ker(η ) ,→ F[λ]n V
ei 7 → x i
of F[λ]-modules, in which K must be free with generators { f i }in=1 . To

obtain the (d j ) which will be annihilators of the F[λ]z j in the struc-
ture theorem decomposition, we must find (then put in normal form)
a matrix whose columns express the { f j } in terms of the {ei }. To wit:
IV.D.1. L EMMA . A := λ1n − B is a relations matrix for V.
P ROOF. We need to specify the { f j }. Put

n
f j := λe j − ∑ bij ei .
i =1
Clearly η ( f j ) = λη (e j ) − ∑i bij η (ei ) = T ( x j ) − ∑i bij xi = 0, by defi-

nition of B. So f j ∈ K (∀ j).
To see that they generate K, suppose 0 = η (∑ j Pj (λ)e j ) for some
polynomials Pj . By repeatedly applying λk e j = λk−1 λe j = λk−1 f j +
∑i bij λk−1 e j , we may rewrite this as 0 = η (∑ j Q j (λ) f j + ∑i β i ei ) with
0
β i ∈ F. That is, 0 = ∑ j Q j ( T ) ) + ∑i β i xi =⇒ β i = 0 (∀i). Hence
η (f j
*

∑ j Pj (λ)e j = ∑ j Q j (λ) f j ∈ F[λ]h f 1 , . . . , f n i.
IV.D.2. R EMARK . In fact, we can prove that { f j } is a base for K over

F[λ]: given ∑ j h j (λ) f j = 0, we have
(∑ hi (λ)λei =) ∑ h j (λ)λej = ∑ h j (λ)bij ei (in F[λ]n )

i j i,j
IV.D. APPLICATIONS TO LINEAR ALGEBRA 233
=⇒ hi (λ)λ = ∑ j h j (λ)bij (in F[λ]) for each i. But this is impossi-

ble: consider i such that hi is of maximal degree: then deg(LHS) >
deg(RHS).
Apply the normal form algorithm to obtain bases {ei0 } and { f j0 }

(for F[λ]n resp. K) related by
Q(λ1n − B) P = diag(d1 , . . . , dk0 , dk0 +1 , . . . , dn )
(IV.D.3)
= diag(1, . . . , 1, δ1 , . . . , δs )
in the notation of the structure theorem and its proof (with k = n
and ` = s since V is torsion). That is, f i0 = di ei0 is our new base for
K = ker(η ).
Now put η (ei0 ) =: xi0 . This is not a basis for V as a vector space (F-
module), since x10 , . . . , xk0 0 = 0. However, the remaining nonzero ele-
ments xk0 0 +1 =: z1 , . . ., xn0 =: zs must generate V as an F[λ]-module;
and indeed by the structure theorem we have
(IV.D.4) V = F[λ]z1 ⊕ · · · ⊕ F[λ]zs ∼

= F[λ]/(δ1 ) ⊕ · · · ⊕ F[λ]/(δs ),
with δ1 | · · · | δs nonzero nonunits, i.e. polynomials of positive
degree.
The canonical forms. The direct sum decomposition in (IV.D.4)

also expresses V as an internal direct sum of s subspaces, whose di-
mensions obviously must add to n. We start by examining the matrix
of the restriction of T to one such subspace.
Pick an i ∈ {1, . . . , s} and write
δi =: F (λ) = λm + Fm−1 λm−1 + · · · + F0 .
Since δi has degree m,
(IV.D.5) zi , Tzi , . . . , T m−1 zi
are linearly independent over F and span F[λ]zi . Moreover,
0 = F ( λ ) zi = F ( T ) zi =⇒ T ( T m−1 zi ) = T m zi = ( T m − F ( T ))zi
= − F0 zi − F1 Tzi − · · · − Fm−1 T m−1 zi .
234 IV. MODULES
We conclude that in the basis (IV.D.5) of F[λ]zi , the restriction T |F[λ]zi

has matrix
 
0 − F0
1 0 − F1 
 
 
 1 0 − F2 
(IV.D.6) CF : =  ..  ,
 
. . .
 1 . 
...
 
0 − Fm−2 
 

1 − Fm−1
which is called the companion matrix of the monic polynomial F.
Doing the same thing for each of the subspaces in (IV.D.3) we find
first that
z̃ := {z1 , Tz1 , . . . , T deg(δ1 )−1 ; . . . ; zs , Tzs , . . . , T deg(δs )−1 zs }
is a basis of V; in particular, ∑is=1 deg(δs ) = n. Writing T in this basis

produces a block diagonal matrix with deg(δi ) × deg(δi ) blocks
(IV.D.7) z̃ [ T ] = diag(Cδ1 , . . . , Cδs )

which is called the rational canonical form of the original matrix B.
The point, of course, is that this new matrix is similar to B: taking
S := z̃ [idV ] x the change-of-basis matrix, we have SBS−1 = (IV.D.7).
Next, assume that the {δi } can be completely factored into lin-
ear factors20 over F. In this case we get more useful bases for each
subspace F[λ]zi by decomposing it into primary cyclic submodules.
For example, if δi = (λ − α1 )e1 (λ − α2 )e2 , take y = (λ − α2 )e2 zi
and w = (λ − α1 )e1 zi , and observe that by (IV.C.26),
F[ λ ] F[ λ ]
(IV.D.8) F[ λ ] z i = F[ λ ] y ⊕ F[ λ ] w ∼
= ⊕ .
((λ − α1 ) ) ((λ − α2 )e2 )
e 1
Clearly {y, ( T − α1 )y, ( T − α1 )2 y, . . . , ( T − α1 )e1 −1 y} is an F-basis for

F[λ]y, and similarly for F[λ]w. Writing the restriction of T to F[λ]y
20This is always true if F is algebraically closed, which is to say that every polyno-
mial over F has a root in F. For instance, C is algebraically closed.
with respect to this basis gives the e1 × e1 matrix

 
α1
 1 α1
 

.
 
(IV.D.9) Je1 (α1 ) :=  1 ..
 

..
 

 . α1


1 α1
since 
 Ty = ( T − α1 )y + α1 y

T (( T − α1 )y) = ( T − α1 )e1 y + α1 ( T − α1 )y

etc.

Repeating this process for each δi yields, as before, a basis for V.

Writing T with respect to this basis produces a block diagonal matrix
(IV.D.10) diag( Je1 (α1 ), Je2 (α2 ), . . . )
called the Jordan canonical form, which is again similar to B.
IV.D.11. D EFINITION . The Jordan form reveals the generalized

eigenspaces Eα of V with respect to T. We set
Eα ( T ) := A(λ−α) (V ) = {v ∈ V | (λ − α)k v = 0 for some k ∈ N}.
Clearly this is the span of the basis elements corresponding to the

blocks Jei (αi ) in (IV.D.10) with αi = α, so that
dim( Eα ( T )) = ∑ ei .
i : αi = α
To summarize, Jordan canonical form corresponds to the primary

cyclic decomposition of V as an F[λ]-module, and the rational canon-
ical form to the (less refined) decomposition in the structure theorem.
Let’s try a basic
IV.D.12. E XAMPLE . We recall from Example IV.C.17, that for

 
1 1 1
B = 1 1 1 ,
 
1 1 1
236 IV. MODULES
we have nf(λ13 − B) = diag(1, λ, λ2 − 3λ). Hence
V = F[ λ ] z1 ⊕ F[ λ ] z2 ∼
= F[λ]/(λ) ⊕ F[λ]/(λ2 − 3λ),
and with respect to the basis z̃ = {z1 , z2 , Tz2 } we get the rational
canonical form  
0 0 0
 0 0 0 
 
0 1 3
since T (z1 ) = 0, T (z2 ) = Tz2 , and T ( Tz2 ) = T 2 z2 = 3Tz2 (from
T 2 − 3T = 0 on F[λ]z2 ).
For the Jordan form, we factor δ2 (λ) = λ2 − 3λ = λ(λ − 3) to
further decompose V into primary cyclic modules:
V = F[λ]z1 ⊕ F[λ]( T − 3 idV )z2 ⊕ F[λ] Tz2

∼
= F[ λ ] / ( λ ) ⊕ F[ λ ] / ( λ ) ⊕ F[ λ ] / ( λ − 3).
Of course, T kills the first two generators and T ( Tz2 ) = 3( Tz2 ) so
the Jordan form is  
0 0 0
 0 0 0 
 
0 0 3
and dim( E0 ( B)) = 2, dim( E3 ( B)) = 1. After all, if a matrix can be
diagonalized, the Jordan form is diagonal. This happens precisely
when the δi (taken individually) have no repeated linear factors.
One thing you may wonder is how to find the basis (or change-
of-basis matrix) which puts B in rational or Jordan canonical form.
We have (writing θ : K ,→ F[λ]n for the inclusion)
e0 [ θ ] f 0 = nf(λ13 − B) = Q(λ13 − B) P
= e0 [idF[λ]n ]e · e [ θ ] f · f [idK ] f 0 ,
so that Q = e0 [id]e =⇒ columns of Q−1 = e [id]e0 yield the e0 -basis
(written in the e-basis). One builds the basis z̃ for the rational (or
Jordan) form from the xi0 := η (ei0 ) for i = k0 + 1, . . . , n. Noting that
x j := η (e j ), if (say) the last column of Q−1 is

 
  (k)
p1 ( λ ) a1
 ..   . 
.  = ∑λ 
0 k . 
e [ en ] =   .  ,
(k)
pn (λ) an
then applying η yields21

   
(k) (k)
a1 a1
k  . 
 . 
x [ xn ] = ∑ x [ T ]  ..  = ∑ Bk 
0
 
. 
 . .
k (k) k (k)
an an
But this is a bit ugly and there are often better ways to proceed:
IV.D.13. E XAMPLE . The matrix

 
2 −1 1 −1
 −1 2 −2 1 
B=
 

 0 1 1 1 
0 −1 1 0
has characteristic polynomial
p B (λ) = det(λ14 − B) = (λ − 1)3 (λ − 2).
This guides the selection of our basis: this is straightforward for

eigenvalue 2, as
−2
E2 ( B) = ker( B − 214 ) = 1 = : h v1 i.
1
0
For eigenvalue 1, first find bases for kernels of powers of ( B − 1):

0
ker( B − 1) = −1
0
1
0
−1
⊂ ker(( B − 1)2 ) = −1 ,
0
0
1
1 0
0 −1 0
⊂ E1 ( B) = ker(( B − 1)3 ) = −1 ,
0
0
1 , 00 .
1 0 1
21This is essentially what [Jacobson] does in the Example on his pp. 198-199,
though as usual his convention is the transpose of that used in these notes.
238 IV. MODULES
(So far, the bases for kernels are easily computed by taking rref of
B − 21, B − 1, ( B − 1)2 , and ( B − 1)3 .) It is the last basis vector for
E1 ( B) that generates it as a Q[λ]-module, and we choose its cyclic
images as our remaining basis vectors for V:
0 −1 0
v2 := 00 7→ v3 := ( B − 1)v2 = 1
1 7→ v4 := ( B − 1) v2 = −01 .
2
1 −1 1
Taking S to be the matrix with columns given by the {vi }, we get

 
2
1
 
 −1
B = S S .

 1 1 
1 1
The minimal polynomial. We previously used this term for an

element of an algebraic extension of a field. But it makes sense for
any finitely generated torsion module M over a PID R, by the struc-
ture theorem. In the notation of IV.C.18, since the free part is zero
(t = 0 and ` = s), each direct summand is annihilated by some
δi . Since all of these divide δs , we have δs M = {0}. Conversely, if
rM = {0} for some r ∈ R, then r ∈ (δ1 ) ∩ · · · ∩ (δs ) = (δs ). So
(δs ) ⊂ R is the set of all elements annihilating M.
So in the special case under study here (cf. (IV.D.4)), (dn ) ⊂ F[λ]
is the annihilator of V. An immediate consequence is the
IV.D.14. T HEOREM . dn ( T ) (= δs ( T )) is the zero transformation, and

if F ∈ F[λ] satisfies F ( T ) = 0, then dn | F. The same holds with “B”
[resp. “matrix”] replacing “T” [resp. “transformation”].
P ROOF. For the second part, just note that F (λ)V = {0} ⇐⇒
F ( T )v = 0 (∀v ∈ V) ⇐⇒ F ( T ) xi = 0 (∀i) ⇐⇒ F ( B) = 0.
IV.D.15. D EFINITION . dn (λ) is the minimal polynomial of T (or

B). We will henceforth write this m T (or m B ).
det(λ1− B)
IV.D.16. P ROPOSITION . (a) m B (λ) = ( ).
monic gcd of (n−1) × (n−1)
minors of λ1 − B
(b) m B and p B := det(λ1 − B) are invariant under similarity.
P ROOF. (a) This is just dn = ∆n /∆n−1 .

(b) If B0 = SBS−1 , S ∈ GLn (F), then B and B0 are matrices of the
same T (with respect to different bases of V). The invariant factors di
in F[λ] are defined for the F[λ]-module V, which itself depends only
on T.
Notice that the coefficients of powers of λ in p B (λ) are therefore
polynomials in the entries of B that are invariant under similarity transfor-
mation (conjugation by an invertible S). These include the trace and
determinant.
Finally we have the
IV.D.17. C OROLLARY (Cayley-Hamilton). p B ( B) = 0.
P ROOF. We have p T (λ) := det(λidV − T ) = ds+1 (λ) · · · dn (λ),

hence p T ( T ) = ds+1 ( T ) · · · dn ( T ) = 0 (since dn ( T ) = 0).
This looks much simpler than the proofs in linear algebra courses,
because we have already proved a more difficult result using mod-
ule theory. In any case, writing p B ( B) = det( B1 − B) = det(0) = 0
is still wrong, because you have to first expand the determinant and
then substitute in the matrix, not the other way around!
240 IV. MODULES
IV.E. Endomorphisms
Recall from IV.B.21-IV.B.22 that for a free module M of rank n over

a commutative ring R, sending endomorphisms to their matrix (with
respect to some base) yields a map
∼
=
EndR ( M) −→ Mn ( R)
which is in fact an isomorphism of rings and of R-modules. What

happens if M is no longer free? In this section we will give an answer
to this question in the case (henceforth assumed) that R is a PID. We
begin with some easy
IV.E.1. E XAMPLES . (a) Suppose M = Rz ∼ = R/(d) is a cyclic R-

module, and note that rz corresponds to r̄ under the isomorphism.
The map
EndR ( M) −→ R/(d)
(IV.E.2)
η 7−→η (1̄)
is an isomorphism of rings and R-modules. [Why? Clearly (IV.E.2) is
an R-module homomorphism. It is injective because η is determined
by where it sends a generator; and surjective because it sends
µr := {multiplication by r } 7−→ r̄
for any r̄ ∈ R/(d). So then EndR ( M ) consists entirely of µr ’s, and

(IV.E.2) sends composition to multiplication.]
(b) If M ∼ = ( R/(d))⊕n , then writing ēi for the “standard” generators
(ē1 = (1̄, 0̄, . . . , 0̄), etc.), writing η (ē j ) = ∑i r̄ij ēi defines a map
EndR ( M) → Mn ( R/(d))
η 7→ (r̄ij )
which one also shows is an isomorphism (of rings and R-modules),

by combining the approach for free modues with that in (a).
(c) On the other hand, if M ∼
= ⊕i R/( pi ) with pi distinct primes of R,
then by Schur’s Lemma IV.B.32, HomR ( R/( pi ), R/( p j )) = {0} for
IV.E. ENDOMORPHISMS 241
i 6= j. (Why?) Combining this with (a) yields
EndR ( M) ∼
= ⊕i EndR ( R/( pi )) ∼
= ⊕i R/( pi ).
Alternatively, one can use the Chinese Remainder Theorem (see the
proof of IV.C.25) to write M ∼
= R/(∏ pi ), apply (a), and use the CRT
again on the RHS.
(d) Finally, if M ∼
= ⊕i ( R/( pi ))⊕ni , then combining Schur’s Lemma
with (b) yields
EndR ( M) ∼ = ⊕i Mni ( R/( pi )),
which is again an isomorphism as rings and as R-modules.
Now we turn to the general case: let
M = Rz1 ⊕ · · · ⊕ Rzs ∼
= Rz1 ⊕ · · · ⊕ Rz` ⊕ Rt ,
| {z }
tor( M)
where ` + t = s, ann(zi ) = (δi ), δ1 | · · · | δ` , and δ`+1 = · · · = δs = 0.

We can present M in terms of generators and relations as
R h e1 , . . . , e s i
M∼
= Rs /K = .
hδ1 e1 , . . . , δ` e` i
Our aim is to get a description of the endomorphism ring
S := EndR ( M)
in the spirit of the above examples, but in terms of the {δi }.

Recall the matrix description of endomorphisms of Rs
∼
=
θ : EndR ( Rs ) −→ Ms ( R)
η̃ 7−→ e [η̃ ] =: (nij ) =: N,
where η̃ (e j ) = ∑i nij ei . Given η̃ ∈ EndR ( Rs ), we can ask when it

makes sense modulo K, as an endomorphism of M (= Rs /K ). Evi-
dently,
• η̃ defines an element η ∈ S ⇐⇒ η̃ (K ) ⊆ K; and
• η̃ defines the zero element in S ⇐⇒ η̃ ( Rs ) ⊆ K.
242 IV. MODULES
For x̃ ∈ Rs , we have
!
δ1
x̃ ∈ K ⇐⇒ x̃ = ∑i`=1 di ri ei ⇐⇒ e [ x̃ ] ∈ ... Rs =: DRs
(for some ri ∈ R) δs
(thinking of Rs as column vectors on the RHS). Hence
η̃ (K ) ⊆ K ⇐⇒ η̃ ( x̃ ) ∈ K (∀ x̃ ∈ K )
[apply e [ ] ] ⇐⇒ NDv ∈ DRs (∀v ∈ Rs )
[apply to v = e1 , . . . , es ] ⇐⇒ ND ⊂ DMs ( R)
⇐⇒ N ∈ MS ,
def.
and
η̃ ( Rs ) ⊆ K ⇐⇒ Nv ∈ DRs (∀v ∈ Rs )
⇐⇒ N ∈ DMs ( R) =: JS .
def.
Note that MS is a subring of Ms ( R): given N, N 0 ∈ MS , we can write
( N 0 N ) D = N 0 ( ND ) = N 0 ( DM0 ) = ( N 0 D )M0 = ( DM)M0 = DM00

with M, M0 , M00 ∈ Ms ( R); and so N 0 N ∈ MS . Furthermore, JS ⊂
MS is a (two-sided) ideal: given N ∈ MS ,
N JS = NDMs ( R) ⊂ DMs ( R) = JS
and JS N = DMs ( R) N ⊂ DMs ( R) = JS .
So MS /JS is a ring (and an R-module!); and we have the
IV.E.3. T HEOREM . θ induces an isomorphism

∼
=
θ̄ : S −→ MS /JS
of rings (and R-modules).
P ROOF. We just did it! To briefly recapitulate: applying θ = e [ ]

to the numerator and denominator of the RHS of
{η̃ ∈ EndR ( Rs ) | η̃ (K ) ⊆ K }
S = EndR ( M) = EndR ( Rs /K ) =
{η̃ ∈ EndR ( Rs ) | η̃ ( Rs ) ⊆ K }
yields exactly MS /JS .
IV.E.4. R EMARK . Note that we can thnk of θ̄ as “taking the ma-

trix with respect to z1 , . . . , zs ” even though this is not a base in the
standard sense.
Now consider the conditions defining MS if s = 2: keeping in

mind that δ1 |δ2 (and denoting by rij arbitrary elements of R), we have

n11 n12 n11 n12 δ1 δ1 r11 r12
N = n21 n22 ∈ MS ⇐⇒ n21 n22 δ2 = δ2 r21 r22

⇐⇒ δδ1 nn11 δδ22 nn22
1 21
12
= δδ12 rr11 δδ12 rr12
21 22
⇐⇒ n21 ∈ ( δδ21 );
so n21 = n210 δ2 , with n0 and the other n arbitrary elements of R.

δ1 21 ij
(Note that if δ2 = 0 6= δ1 , this would make n21 = 0.) Furthermore,
we have

N ∈ JS ⇐⇒ nn11 n12
21 n22
= δ1
δ2
r11 r12
r21 r22 = δ1 r11 δ1 r12
δ2 r δ2 r22
21
⇐⇒ n11 , n12 ∈ (δ1 ) and n21 , n22 ∈ (δ2 ).

The upshot is that, for elements of θ̄ (S) = MS /JS , we need to
consider n11 and n12 as elements of R/(δ1 ), n21 as an element of
( δδ12 )/(δ2 ), and n22 in R/(δ2 ).
More generally, for any s, this analysis leads to the following
specifications for entries in the “regions” of the s × s matrix N (cor-
responding via θ̄ to elements of S) as shown:

 (I) i ≤ j, ` : nij ∈ R/(δi ) 
...

(I)


(II) j < i ≤ ` : nij ∈ ( δδi )/(δi )

. . . (I) 
 
j
(II)


 (III) i > `; j ≤ ` : nij = 0  
(III) (IV)


(IV) i, j > ` : nij ∈ R
so we can write nij := nij0 δδi in (II) as above, with nij0 ∈ R/(δj ). In the
j
event that M is torsion, ` = s and we don’t have regions (III) and
(IV).
An immediate consequence is
IV.E.5. C OROLLARY. The center of S = EndR ( M) is R.

244 IV. MODULES
P ROOF. Let ε is ∈ S be the endomorphism with matrix given by22

θ̄ (ε is ) = eis . (Note that this is possible because the (i, s)th entry lies
in region (I) or (IV), never (II).) This endomorphism sends zs 7→ zi
and kills all other z j . So given η ∈ C (S) (in the center), and writing
N = θ̄ (η ), we have
η (zs ) = η (ε ss (zs )) = ε ss (η (zs )) = ε ss (∑i nis zi ) = ∑i nis ε ss zi = nss zs
and
η (z j ) = η (ε js zs ) = ε js (η (zs )) = ε js (∑i nis zi ) = ∑i nis ε js (zi ) = nss z j ,
so that η is simply multiplication by ns s — which, being in region (I)

or (IV), can be any element of R.
Assume henceforth that M is torsion. As S is an R-module:

(a) if R = Z, then M = G is a finite abelian group, and S = EndZ ( G )
also has the structure of a finite abelian group, with a (finite) order;
while
(b) if R = F[λ], then M = V is an F-vector space on which λ acts by
a linear transformation T ∈ EndF (V ), and S = EndF[λ] (V ) itself
has the structure of an F-vector space, with a (finite) dimension.
So we can take the theory for a test-drive to see if we can compute
the italicized numbers. For (a), we have the
IV.E.6. C OROLLARY. Consider any finite abelian group, written in the

form G ∼
= Zm1 × · · · × Zms with m1 | · · · | ms . Then the number of group
homomorphisms from G to itself is
s
2s−2j+1
|EndZ ( G )| = ∏ m j .
j =1
P ROOF. With S = EndZ ( G ), one counts the possible choices for

the nij in a matrix N ∈ MS /JS . For (I) i ≤ j, nij ∈ Z/(mi ) = Zmi ;
mi
while for (II) i > j, nij = nij0 m with nij0 ∈ Z/(m j ) = Zm j . So to
j
compute |S| = |MS /JS |, we simply have to take the product of all
22Recall that e is the matrix with (i, j)th entry 1 and all other entries 0.
ij
entries of the matrix

 
m1 m1 m1 · · · m1
m1 m2 m2 · · · m2
 
 
 
 m1 m2 m3 · · · m3 
.. .. ..
 
..
. . . .
 
 
m1 m2 m3
which gives the result.
For (b), notice that
S = EndF[λ] (V ) = {η ∈ EndF (V ) | ηT = Tη }
is the centralizer of T. Writing x [ T ] = B and x [η ] = Z with respect to

some basis of V, S is identified with23
(S ∼
=) EndF[λ] (Fn ) = { Z ∈ Mn (F) | ZB = BZ },
the ring of matrices commuting with B.
IV.E.7. C OROLLARY. Let B ∈ Mn (F), with normal form
nf(λ1n − B) = diag(1, . . . , 1, δ1 (λ), . . . , δs (λ)).
Then dimF (S) = ∑sj=1 (2s − 2j + 1) deg(δj (λ)).
P ROOF. Once again, we use θ̄ to identify S with s × s matrices

δ (λ)
N with entries (I) nij ∈ F[λ]/(δi (λ)) or (II) nij = nij0 δi (λ) (and nij0 ∈
j
F[λ]/(δj (λ))). So these nij ’s each lie in a vector space of dimension
(I) deg(δi ) resp. (II) deg(δj ), and we can record these degrees in a
matrix exactly like that in the last proof. Only this time, to get the
dimension of S, we add these entries rather than multiplying them.

Call the transformation T cyclic if its action on V makes the latter

into a cyclic F[λ]-module (that is, s = 1).
IV.E.8. C OROLLARY. A linear transformation T ∈ EndF (V ) is cyclic

⇐⇒ the only transformations commuting with T are polynomials in T.
23Here λ acts on Fn via B.
246 IV. MODULES
P ROOF. First let T be an arbitrary transformation, and take d =

deg(m T ) = deg(δs ) to be the degree of the minimal polynomial. The
polynomials in T certainly commute with T, and so
(IV.E.9) F[λ]/(m T ) =: F[ T ] ,→ EndF[λ] (V ).

−1
We have dim(RHS) = d + ∑sj= 1 (2s − 2j + 1) deg( δj ) by IV.E.7, and
dim(LHS) = d. But then V is cyclic ⇐⇒ s = 1 ⇐⇒ dim(RHS) is
d ⇐⇒ (IV.E.9) is an isomorphism ⇐⇒ the centralizer of T consists
of polynomials in T.
IV.E.10. E XAMPLES . (i) The matrices commuting with a Jordan

block are polynomials in the Jordan block.
(ii) Consider the matrix
 
−1
1 −1
B=
 
−1

 1
1 −1
acting on V = Q4 . This is in rational canonical form, hence the com-

panion matrix for δ = δ1 (s = 1), and we accordingly write
V = Q[λ]/(δ(λ)), δ(λ) = λ4 + λ3 + λ2 + λ + 1.
This is cyclic, and so IV.E.8 applies.

But we can also recognize δ as the 5th cyclotomic polynomial,
and thus V ∼ = Q[ζ 5 ] as the corresponding cyclotomic number field.
So IV.E.8 tells us that EndQ[λ] (V ) ∼
= Q[ζ 5 ] realizes the multiplicative
action of the number field on itself via 4 × 4 rational matrices that
are polynomials in B. In particular, B corresponds to ζ 5 itself.
If we replace V by VC = C4 ,
j
VC = C[λ]/(δ(λ)) = ⊕4j=1 C[λ]/(λ − ζ 5 ) ∼
= C4
=⇒ EndC[λ] (VC ) = C × C × C × C is represented by diagonal
matrices with respect to the (complex) eigenbasis for B.
Notice that in going from Q to C, the dimension as a vector space

(over Q resp. C) does not change, but the ring structure does dra-
matically — from a field to a non-domain!
(iii) Let V = C3 . Recall from Example IV.D.12 that
 
1 1 1
B= 1 1 1 
 
1 1 1
is similar to its rational and Jordan forms
   
0 0
B0 =  0 0  and B00 =  0 .
   
1 3 3
From B0 , we see that s = 2, δ1 = λ and δ2 = λ2 − 3λ, from which

IV.E.7 yields
dimC (EndC[λ] (V )) = 3 deg(δ1 ) + 1 deg(δ2 ) = 5.
But what the ring structure of S = EndC[λ] (V ) is like, is much

clearer from B00 , which yields the decomposition into primary cyclic
submodules V ∼ = (C[λ]/(λ))⊕2 ⊕ C[λ]/(λ − 3). From there, we
can use IV.E.3(d) to compute S ∼ = M2 (C) × C as a ring, since both
C[λ]/(λ) and C[λ]/(λ − 3) are isomorphic to C as rings.
V. Remarks on Associative Algebras
V.A. Algebras over a field
Let F be a field. What would you call an F-vector space where

you can multiply vectors?
V.A.1. D EFINITION . An F-algebra (or algebra over F) is a ring

A, together with a scalar multiplication by F which makes A into an
F-vector space and satisfies
(V.A.2) f · ( a1 a2 ) = ( f · a1 ) a2 = a1 ( f · a2 ).
V.A.3. E QUIVALENT D EFINITION . A ring A together with an em-

bedding ε : F ,→ C ( A).
P ROOF THAT V.A.1 IMPLIES V.A.3. Let A be an F-algebra, and

(for each f ∈ F) set ε( f ) := f · 1 A ∈ A. Then:
• ε is a homomorphism (from F to A): since A is a vector space over
F, we have ( f 1 + f 2 ) · a = f 1 · a + f 2 · a and ( f 1 f 2 ) · a = f 1 · ( f 2 · a),
and setting a = 1 A (and using (V.A.2)) does the job;
• ε is injective because F is a field; and
• ε( f ) ∈ C ( A) because aε( f ) = a( f · 1 A ) = f · ( a1 A ) = f · (1 A a) =
( f · 1 A ) a = ε( f ) a.
[The other direction is left to you.]
We will now stop writing the “·”. Also, notice that ε embeds F as
a subring of A whose elements commute with everything; so we can
identify F with this subring, and drop “ε”.
V.A.4. E XAMPLES . (i) Field extensions E/F: these are, by defini-

tion, fields containing F.
(ii) Polynomial algebras F[ x1 , . . . , xn ].
249
250 V. REMARKS ON ASSOCIATIVE ALGEBRAS
(iii) Product algebras F × · · · × F.

(iv) Matrix algebras Mn (F) (= the ring of endomorphisms of an n-
dimensional F-vector space).
(v) Group algebra F[ G ] (of a finite group G).
(vi) Ring of endomorphisms of an F[λ]-module.
(vii) Quaternion algebras (e.g., H as an R-algebra; or the rational
quaternion algebras from HW 6 #6).
(viii) Exterior algebras (defined below).
V.A.5. D EFINITION . A0 ⊂ A is an F-subalgebra if A0 is a sub-

F-vector space and subring. The F-subalgebra generated by a set
S ⊂ A is
( )
elements of A that can be written
F[S] := A0
\
= .
as polynomials over F in {1} ∪ S
A0 ⊂ A subalg.
A0 ⊃ S
(So there is an obvious notion of finitely generated F-algebra; this is

much weaker than finite-dimensionality of A as vector space over F.)
V.A.6. D EFINITION . Officially, I ⊂ A is an (algebra) ideal if I is

an ideal in the ring A which is an F-vector subspace. But in point of
fact, since F ⊂ A, any ring-ideal of A is already closed under multi-
plication by F, hence also an algebra-ideal; so there’s no difference.
Given an ideal I ⊂ A, the quotient A/I has an F-algebra struc-

ture: the composition F ,→ A A/I is still injective by (III.F.1).
V.A.7. D EFINITION . A map α : A → B of F-algebras is an F-

algebra homomorphism if it is a ring homomorphism which is F-
linear (i.e., α( f a) = f α( a) for all f ∈ F and a ∈ A).
As usual, we get a “Fundamental Theorem” to the effect that I :=

ker(α) is an (algebra) ideal and there exists ᾱ such that
A
α / B
=
η !! . ᾱ
A/I
V.A. ALGEBRAS OVER A FIELD 251
commutes.
There is also an algebro-theoretic version of Cayley’s theorem:
V.A.8. T HEOREM . Any F-algebra A is isomorphic to a subalgebra of

an algebra of endomorphisms of a vector space.
P ROOF. Consider A as an F-vector space, and map
` : A −→ EndF ( A)
a 7−→ ` a := left-multiplication by a.
Since A is an algebra,
` a ( f α) = a( f α) = f ( aα) = f ` a (α)
=⇒ ` a ∈ EndF ( A). Moreover, we know that ` is an injective ring
homomorphism.1 Finally, ` is a homomorphism of F-vector spaces,
since by (V.A.2) we have ` f a (α) = ( f a)(α) = f ( aα) = f ` a (α) and
thus ` f a = f ` a .
Exterior algebras. For this extended example, start with a vector

space V/F of dimension n (without a multiplication law, of course).
We would like an algebra A generated by V such that
(V.A.9) v2 = 0 (∀v ∈ V ).
Let {u1 , . . . , un } ⊂ V be a basis. Then (V.A.9) gives
0 = (ui + u j )2 − u2i − u2j = ui u j + u j ui =⇒
(V.A.10) ui u j = − u j ui .
If we take i1 < · · · < ik , then this yields
(V.A.11) uiσ(1) · · · uiσ(k) = sgn(σ )ui1 · · · uik (∀σ ∈ Sk )
since sgn(σ ) = (−1)# of transpositions in σ and σ can be built from adja-

cent transpositions (as in (V.A.10)). Henceforth, we shall write “∧”
for our product and make the formal
1Why? Recall ` = 0 =⇒ 0 = ` 1 = a · 1 = a.
a a
V•
V.A.12. D EFINITION . FV is the F-algebra with
• (F-vector space) basis consisting of monomials2
u I : = u i1 ∧ · · · ∧ u i k ( i1 < · · · < i k )
where I = {i1 , . . . , ik } ranges over subsets of {1, . . . , n},
• product
(
0, if I ∩ J 6= ∅
uI ∧ uJ =
sgn(σIJ )uI∪J , if I ∩ J = ∅
where σIJ shuffles I and J together, and
• identity u∅ = 1.
V•
We have dimF ( F V) = ∑nk=0 (nk) = (1 + 1)n = 2n , and
V• L Vk
(V.A.13) FV = k F V,
Vk
where FV is the subspace spanned by monomials of degree k.
V.A.14. E XAMPLE . We illustrate the product: taking I = {1, 3, 6}

and J = {2, 5}, we “shuffle” them together by jumping 2 over 3 and
6, then 5 over 6, for a total of three transpositions. Hence
(u1 ∧ u3 ∧ u6 ) ∧ (u2 ∧ u5 ) = (−1)3 u1 ∧ u2 ∧ u3 ∧ u5 ∧ u6 = −uI∪J .

| {z } | {z }
uI uJ
V.A.15. P ROPOSITION . Let B = (bij ) ∈ Mn (F). Then

(b11 u1 + · · · + bn1 un ) ∧ · · · ∧ (bn1 u1 + · · · + bnn un ) = (det( B)) u1 ∧ · · · ∧ un .
P ROOF. Expanding the LHS gives
∑i1 ,...,in (bi1 ,1 · · · bin ,n )ui1 ∧ · · · ∧ uin .

Since v ∧ v = 0, terms with i j = ik for j 6= k vanish, and this becomes
∑σ∈Sn (bσ(1),1 · · · bσ(n),n )uσ(1) ∧ · · · ∧ uσ(n)

= ∑σ (∏i bσ(i),i )sgn(σ) u1 ∧ · · · ∧ un ,
where the parenthetical quantity is just det( B).

2The degree of the monomial is k = |I|.
V.A. ALGEBRAS OVER A FIELD 253
V.A.16. T HEOREM . Assume |F| = ∞, let Q ∈ F[ x11 , x12 , . . . , xnn ] =:

F[{ xij }] be a homogeneous polynomial of degree q in n2 variables, and de-
fine Q( B) := Q({bij }) for matrices B ∈ Mn (F). Assume that Q(1n ) = 1
and Q( BB0 ) = Q( B) Q( B0 ). Then Q is a power of det.
S KETCH . Since Q is homogeneous,

(V.A.17)
Q( B) Q(adjB) = Q((det B)1n ) = (det B)q Q(1n ) = (det B)q .
Writing X = ( xij ), set P( X ) := Q( X ) Q(adjX ) − (det X )q ∈ F[{ xij }].

Since nontrivial polynomials over an infinite field do not evaluate
to zero everywhere, but P( B) = 0 for all B by (V.A.17), we must
have P = 0. Hence Q( X ) Q(adjX ) = (det X )q in F[{ xij }] and so
Q( X ) | (det X )q . But F[{ xij }] is a UFD, and so the result is clear
if we know that det X is irreducible in F[{ xij }]. This is proved in
[Jacobson].
Exterior algebras are ubiquitous in algebra (esp. representation
theory) and geometry (differential forms).
V.B. Finite-dimensional division algebras
What about a vector space where you can multiply and divide vec-
tors?
V.B.1. D EFINITION . A division algebra over a field F is an F-
algebra A whose underlying ring is a division ring.
This rules out most of the examples in V.A.4; for example, prod-
ucts like F × F contain zero-divisors, as do matrix algebras.
V.B.2. E XAMPLES . (i) Field extensions are division algebras: e.g.,
C is an R-division algebra; and Q[ζ 5 ] is a Q-division algebra.
(ii) Quaternion algebras give some non-commutative examples: H
(Hamilton’s quaternions) is an R-division algebra; while the non-
split (i.e., division ring) cases in HW 6 #6 give Q-division algebras.
We are particularly interested in division algebras which are finite-
dimensional (as F-vector spaces). While number fields (viewed as
field extensions) easily yield an endless list of such examples over
Q, you may find it difficult to recall seeing any finite-dimensional
field extensions of C. That is because they don’t exist!
V.B.3. D EFINITION . (i) An algebraic field extension3 of F is one
whose every element is algebraic (cf. III.G.6(ii)) over F.
(ii) Call a field F algebraically closed if it has no algebraic field
extensions (other than itself).
V.B.4. E XAMPLE . The Fundamental Theorem of Algebra states
that every polynomial over C has a root (hence all roots) in C. (This
theorem is proved in complex analysis.) Since any element α of a
field extension which is algebraic over C satisfies a polynomial equa-
tion P(α) = 0, α actually belongs to C. So C is algebraically closed.
Clearly division algebras are the simplest kind of F-algebra after
field extensions; so we shall do a rough classification for F = R, C,
and finite fields. To begin with, we finish off C with the
3Warning: these need not be finite-dimensional (though they certainly are if they
are finitely generated).
V.B. FINITE-DIMENSIONAL DIVISION ALGEBRAS 255
V.B.5. T HEOREM . Let F be an algebraically closed field, and A a finite-

dimensional division algebra over F. Then A = F.
P ROOF. Let a ∈ A, and consider the ring homomorphism
eva : F[λ] F[ a] ⊂ A
f ( λ ) 7 → f ( a ).
This cannot be injective, since A (hence F[ a]) is finite-dimensional

and F[λ] is not. So we have F[ a] ∼ = F[λ]/(m a ), where m a is the
minimal polynomial of a over F. Were this reducible, F[ a] wouldn’t
be a domain, which is impossible since A is a division algebra.
Hence m a is irreducible, and F[ a] is a field, all of whose elements
are algebraic over F (cf. III.G.8). Since F is algebraically closed,
F[ a] = F. So, in particular, a ∈ F; and since a ∈ A was arbitrary,
A = F.
Given p(λ) ∈ R[λ] monic, we have
p(λ) = ∏nj=1 (λ − α j ) = ∏nj=1 (λ − ᾱ j )
in C[λ], by the Fundamental Theorem of Algebra. We can rewrite

this as
p(λ) = ∏ri=1 (λ − ai ) × ∏sk=1 (λ − β k )(λ − β̄ k )

= ∏ri=1 (λ − ai ) × ∏sk=1 (λ2 − 2<( β k ) + | β k |2 ),
with ai ∈ R and β k ∈ / R. Hence no polynomial of degree > 2 is
irreducible in R[λ].
Let A be a finite-dimensional division algebra over R. Given α ∈
A\R, we consider as usual
evα : R[λ] R[α] ⊂ A,
which as above has a nontrivial kernel K since dimF ( A) < ∞. Since

R[λ] is a PID, K = (mα ) with mα irreducible (also as above); and as
α∈/ R, deg(mα ) > 1. So deg(mα ) = 2, and mα (λ) = λ2 − 2aλ + b,
with a2 < b. We may thus write α = β + a, where β ∈ A\R and
β2 = a2 − b < 0.
Now consider the subset
A 0 : = { α ∈ A | α 2 ∈ R≤0 } ⊂ ( A \ R ) ∪ { 0 } .
From the last paragraph it is clear that if A\R 6= ∅, then A0 6= {0}

(and the converse is obvious).
V.B.6. L EMMA . A0 is an R-subspace of A.
P ROOF. Given r ∈ R, α ∈ A0 , we have (αr )2 = α2 r2 ≤ 0 =⇒

αr ∈ A0 . So A0 is closed under multiplication and we only need to
check sums of elements. So let u, v ∈ A0 \{0} be linearly independent
over R in A. (If they are dependent, u + v is a multiple of u and we
are done.) By assumption, we have u2 , v2 ∈ R<0 .
Suppose first that u = av + b, with a, b ∈ R. Then in
u2 = ( av + b)2 = a2 v2 + 2abv + b2 ,
the RHS terms are real except for 2abv, which forces ab = 0. But we
can’t have a = 0, for then u = b ∈ R; and if b = 0, then u = av
contradicts the independence.
So u is not of the form av + b, which means that u, v, and 1 are
independent over R. Hence u + v, u − v ∈ A\R; and so as above
(for α), they satisfy irreducible quadratic equations
0 = (u + v)2 − p(u + v) − q and 0 = (u − v)2 − r (u − v) − s.
Writing c = u2 , d = v2 , these become
0 = c + d + (uv + vu) − p(u + v) − q

and 0 = c + d − (uv + vu) − r (u − v) − s,
and adding gives
0 = ( p + r )u + ( p − r )v + (q + s − 2c − 2d)1.
By independence of {u, v, 1} it now follows that p = r = 0. So for the

original equations to have been irreducible, we must have q, s < 0 ;
in particular, (u + v)2 = q ∈ R<0 . Hence u + v ∈ A0 as desired.
For u ∈ A0 , set
Q(u) := −u2 ∈ R.
V.B.7. L EMMA . Q is a positive-definite quadratic form on A0 .
P ROOF. Since A is a domain, Q(u) = 0 ⇐⇒ u = 0. Moreover,

for a ∈ R, Q( au) = a2 Q(u), so Q is quadratic. Finally, Q(u) ≥ 0 for
all u ∈ A0 (by definition of A0 ).
Write
B(u, v) := Q(u + v) − Q(u) − Q(v) = −(uv + vu)
for the associated positive-definite symmetric bilinear form. Now

suppose A0 6= {0}, i.e. A ) R, and pick i ∈ A0 such that Q(i) = 1;
we can do this by rescaling any element in A0 \{0} by a real number.
Then i2 = −1, and we fix the copy of C = R + iR = R[i] ⊂ A.
Next, suppose that A ) C; then A0 ) iR, and we pick ̂ ∈ A0 \iR
1
*
and take j̃ := ĵ − iB(i, ĵ). This gives B(j̃, i) = B(ĵ, i) − B(i, ĵ)
B(i,
i) =
2
0, and rescaling j̃ gives j with j = −1 and j ⊥ i (i.e. 0 = B(i, j) =
ij + ji). Setting k = ij, we compute

2
k = ijij = −iijj = −(−1)(−1) = −1

ik = i2 j = −j = jii = −iji = −ki

jk = · · · = −kj


0
k ∈ A , k ⊥ i, j

=⇒ 1, i, j, k R-linearly independent
R + iR + jR + kR = H ⊂ A.


Finally, suppose A ) H. Then there exists ` ∈ A0 with Q(`) = 1

and ` ⊥ i, j, k. As above, this gives ì = −i`, `j = −j`, and `k =
−k`; substituting k = ij in the last of these gives
−(ij)` = `(ij) = (ì)j = −(i`)j = −i(`j) = i(j`) = (ij)`,
a contradiction. This proves the famous
V.B.8. T HEOREM (Frobenius, 1877). Let A be a finite-dimensional

division algebra over R. Then A = R, C, or H.
V.B.9. R EMARK . If one allows A to be nonassociative, then there

is one more (8-dimensional) option, Cayley’s octonions O = H × H
with the multiplication law
(q, r ) · (s, t) = (qs − r ∗ t, q∗ t + rs)

where “∗” denotes “quaternionic conjugation” (i 7→ −i, j 7→ −j,
k 7→ −k). More or less, this mimics the way you get H from C × C
and C from R × R. The octonions play a starring role in the explicit
construction of the exceptional Lie groups G2 , F4 , E6 , E7 , E8 in Cartan’s
classsification of simple Lie groups over C.
V.B.10. R EMARK . There are lots of non-isomorphic 4-dimensional

“quaternion algebras” over Q, and there are lots of algebraic field
extensions. But one might have held out hope that, say, there is an
upper bound on the dimension of non-commutative Q-division al-
gebras. Alas, this is not the case: for instance, if γ is an even integer
not divisible by 8, the Q-algebra generated by x, y subject to the rela-
tions
x3 + x2 − 2x − 1 = 0, xy = y( x2 − 2), y3 = γ
is a division algebra of dimension 9. A classification of such exam-
ples was carried out by Dickson.
Finally, we consider the case of a division algebra A over a finite

field F (i.e. |F| < ∞), with n := dimF A < ∞. Clearly | A| = |F|n ,
and so (forgetting the F-action) A is a finite division ring. Con-
versely, if A a finite division ring, then C ( A) is a finite field and A is
an algebra over it (cf. V.A.3), necessarily finite-dimensional.
V.B.11. T HEOREM (Wedderburn, 1905). Any finite division ring is

commutative, hence a field.
V.B.12. R EMARK . The theorem means that algebraic field exten-

sions furnish the only examples of finite-dimensional F-division al-
gebras when |F| < ∞.
P ROOF OF V.B.11. Set F = C ( A), q = | F |, n = dimF A. We need

to show that n = 1, since this is equivalent to A = F.4
Applying the class equation to the group A∗ = A\{0} gives
(V.B.13) | A∗ | = ∑i |ccl( xi )| = ∑i [ A∗ : stab( xi )]

where xi is a set of representatives for the conjugacy classes in A∗ .
In particular, there are q − 1 one-element conjugacy classes, given by
the elements x1 , . . . , xq−1 of F ∗ ; each has stabilizer equal to all of A∗ .
Each xi ∈ A∗ \ F ∗ , on the other hand, is stabilized by the nonzero
elements of a proper subring Ai ⊂ A containing F. (Why?) These Ai
are F-algebras, and so | Ai | = qmi with 1 ≤ mi < n, and |stab( xi )| =
qmi − 1. Thus (V.B.13) becomes
q n −1
(V.B.14) q n − 1 = | A | − 1 = | A ∗ | = ( q − 1 ) + ∑ i ≥ q q mi −1 .
Now regard, for each i, A as a module over Ai . Clearly, it is free

(A has no zero-divisors), of some finite rank di . Moreover, Ai is a
mi -dimensional vector space over F. So as F-vector spaces,
F n = A = A i ⊕ · · · ⊕ A i = F mi ⊕ · · · ⊕ F mi
| {z } | {z }
di di
=⇒ n = mi di =⇒ mi | n (∀i).
Finally, define the dth cyclotomic polynomial
∏
j
f d (λ) := ( λ − ζ d ),
1 ≤ j ≤ d−1
(d, j) = 1
with f 1 (λ) = 1 by convention; then we have
λn − 1 = ∏ f d ( λ ),
1≤d≤n
d|n
4I have changed font for the field, because we want to think of A = F as a field
extension of some original field F.
and similarly for λmi − 1. So

λn − 1
mi | n (∀i ≥ q) =⇒ ∈ ( f n (λ)) ⊂ Z[λ] (∀i ≥ q)
λ mi − 1
qn − 1
=⇒ qn − 1, m ∈ ( f n (q)) ⊂ Z (∀i ≥ q)
q i −1
=⇒ f n (q) | q − 1
(V.B.15)

=⇒ | f n (q)|q − 1.
But
∏
j
| f n (q)| = |q − ζ n | > q − 1,
( j,n)=1
and we have a contradiction, unless n = 1.
V.B.15. C OROLLARY. Any finite domain R is a field.
P ROOF. For any r ∈ R, left-multiplication `r gives a map R →

R. This map is injective since there R has no zero-divisors. By the
pigeonhole principle, it is therefore surjective, and there exists r 0 ∈ R
with rr 0 = 1R . So R \ {0} = R∗ and R is a division ring, and we are
done by V.B.11.

Algebra 1

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Algebra 1

Uploaded by

Copyright:

Available Formats

Matt Kerr

II. Groups Topics include: subgroups, Cayley and Lagrange the-

III. Rings Topics include: homomorphisms,integral domains, ide-

IV. Modules Topics include: Schur’s Lemma, structure theorems

V. Algebras Topics include: exterior algebras, division algebras,

Virtual Fall Semester 2020

V Remarks on Associative Algebras 249

Recall that a set S is a collection of elements. If its order |S| (i.e.,

When these sets are disjoint (viz., Si ∩ S j = ∅ for i 6= j), writing

is a rule associating to each s ∈ S an element f (s) ∈ T; it has graph

Γ f = {(s, f (s)) | s ∈ S}.

A subset Γ ⊂ S × T is the graph of some map if and only if

is inherently associative. When inverses exist, we have

I.A.1. D EFINITION . (i) A relation on S is a subset

hold for all a, b, c ∈ S.

I.A.2. E XAMPLES . Here are some random equivalence relations.

(vi) Given f : S → T, define (on S)

I.A.3. N ON - EXAMPLES . Here are some relations which are not

Given an equivalence relation ∼ on S, the ∼-equivalence class

I.A.5. P ROPOSITION . The ∼-equivalence classes yield a partition of S,

P ROOF. See Exercise (4) of Problem Set 1.

I.A.6. D EFINITION . (i) The quotient set

is the set of ∼-equivalence classes.3

We shall say two sets are isomorphic, written S ∼

I.A.7. E XAMPLES . Referring to I.A.2,

Finally, given sets S, T, an equivalence relation ∼ on S, and a map

I.A.8. P ROPOSITION . Suppose that

for all a, b ∈ S. Then there is a unique f¯ : S/∼ → T such that f¯ ◦ ν = f .

P ROOF. Define f¯ to send ā ∈ S/∼ to f ( a) ∈ T. This is well-

We turn to some results of Euclid. A prime number p ∈ Z is one

I.B.1. F UNDAMENTAL T HEOREM OF A RITHMETIC . Any natural

P ROOF. We use induction (n = 1 is clear). Assume the statement

1 < n0 := p1 ( p2 · · · ps − q2 · · · qt ) = (q1 − p1 )q2 · · · qt < m.

Factoring the parentheticals into primes, the inductive hypothesis

which is a contradiction, or p1 is one of the q2 , . . . , qt . Reordering

I.B.2. P ROPOSITION . There are infinitely many primes.

P ROOF. Suppose p1 , . . . , ps is a complete list of positive primes;

I.B.3. D IVISION A LGORITHM . Given a, b ∈ Z, b 6= 0, there exist

0 ≤ r < |b| and a = bq + r.

P ROOF. We may assume b > 0; then M := {bn | n ∈ Z, bn ≤ a}

I.B.4. P ROPOSITION . (m, n) = mu + nv for some u, v ∈ Z.

P ROOF. Let I := {mx + ny | x, y ∈ Z}, with least positive ele-

r = m − dq = m − (mu + nv)q = m(1 − uq) − n(vq) ∈ I.

For this not to contradict leastness of d, we must have r = 0 and

4This the the well-ordering principle; it is equivalent to the principle of induction.

I.C.1. D EFINITION . A partial order on a set S is a relation “≤”

for all x, y, z ∈ S. The pair (S, ≤) is called a poset.

An easy example is (P (S), ⊂).

I.C.2. D EFINITIONS . Let (S, ≤) be a poset.

(iv) A maximal element7 of S is x ∈ S such that

I.C.3. Z ORN ’ S L EMMA . If every chain in S has an upper bound, then

This is needed for:

I.C.4. A XIOM OF C HOICE . Given a family of nonempty sets {Xi }i∈ I ,

(Clearly, this is only needed when I is infinite.) People make a fuss

xα := f ( ϕ(UB{ x β | β < α})).

This “goes on forever”, so that α 7→ xα yields an injection Ord ,→ S

A group is a set G with a binary operation, i.e. a map1

for all x, y, z ∈ G. Associativity means that there is no need for

respecting the binary operation

Groups abound in mathematics and physics, e.g. via rotational

(1) Galois theory. By considering the structure of [roughly] the group

As to what the “representation theory” mentioned in (2) is all

to “recognize” G and to list possibilities for the representation (and