Rings and Modules

Old Syllabus for O4

T. W. K orner
October 5, 2004
Small print The syllabus for the course is dened by the Faculty Board Schedules (which
are minimal for lecturing and maximal for examining). Please note that, throughout, ring
means commutative ring with one. I should very much appreciate being told of any
corrections or possible improvements and might even part with a small reward to the rst
nder of particular errors. This document is written in L
A
T
E
X2e and stored in the le
labelled ~twk/1B/Rings.tex on emu in (I hope) read permitted form. My e-mail address
is twk@dpmms.
Contents
1 Rings 2
2 Ideals, quotients and the isomorphism theorem 4
3 Integral domains, elds and fractions 6
4 Unique factorisation, Euclidean and principal ideal domains 11
5 Polynomials over rings 14
6 Unique factorisation for polynomials 18
7 Fields and their simple extensions 22
8 Splitting elds of polynomials 26
9 Finite elds 28
10 Modules 31
1
11 Linear relations in modules 35
12 Matrices and modules 38
13 The module decomposition theorems 42
14 Applications to endomorphisms 47
15 Reading and further reading 51
1 Rings
The same ideas and proofs occur in the study of the integers (number theory),
polynomials (leading to algebraic geometry), parts of the theory of matrices
and in the theory of Abelian groups. They may be unied by using the
theory of commutative rings and modules following a programme laid out by
Emmy Noether and others. We start by looking at commutative rings with
one.
Denition 1 We say that (R, +, .) is a commutative ring with a one if
(i) (R, +) is an Abelian group.
(ii) a(bc) = (ab)c for all a, b, c R. [Associative law of multiplication.]
(iii) a(b +c) = ab +ac, (b +c)a = ba +ca for all a, b, c R. [Distributive
law.]
(iv) There exists a 1 R such that 1a = a1 = a for all a R. [Existence
of a multiplicative identity.]
(v) ab = ba for all a, b R. [Commutative law of multiplication.]
Rules (iii) and (iv) could be shortened using rule (v). We usually write 0 for
the identity of the group (R, +) and call 0 the zero of R.
Rule (iv) is made easier to use by the following simple remark.
Lemma 2 (Uniqueness of multiplicative identities) If (M, .) is an ob-
ject with multiplication and 1, 1

M are identities in the sense that

1a = a1 = a and 1

a = a1

= a for all a M,
then 1 = 1

.
Thus R has a unique multiplicative identity 1. (We shall usually refer to 1
as one. It is sometimes called the unit element of R but the word unit
means something dierent in the context of this course, see Denition 42.)
2
There are important examples of non-commutative rings (that is systems
obeying all the rules in Denition 1 except (v) the commutative law of mul-
tiplication) such as the set of n n matrices with the usual addition and
multiplication [n 2]. However, there are many beautiful results which are
only true for commutative rings. Rule (iv) (the existence of a one) is less
important. It gives some of our theorems and proofs a more elegant form
but commutative rings without one are not much harder to deal with.
In this course we shall only deal with commutative rings with 1 and ring
will mean commutative ring with 1.
Rings have many of the properties of the ordinary number systems with
which we are familiar from school. The integers Z with the usual operations
form one of the most important examples. Note that the equation 2m = 1
has no solution in Z (in other words 2 has no multiplicative inverse). The
system (Z
n
, +, ) of the integers modulo n is another example. (The reader
is certainly familiar with this system but denition freaks will nd a neat
denition using ring theory in Denition 16.) Note that in Z
12
3 ,= 0 and 4 ,= 0 yet 3 4 = 0
(we call 3 and 4 divisors of zero) and
2 ,= 6 yet 2 3 = 6 3
(thus we can not use cancellation to get from a b = a c to b = c). In Z
81
we have 3 ,= 0, 3
2
,= 0, 3
3
,= 0 yet 3
4
= 0 (we say that 3 is nilpotent). These
examples suggest that when dealing with rings we should rst try methods
and ideas which work for ordinary number systems but be prepared to
modify or, if the worst comes to the worst, abandon those parts which depend
on division or cancellation.
However we have access to another fertile source of inspiration. We have
already met two examples of abstract algebraic systems:- groups and vector
spaces. Techniques and ideas which were useful for these are likely to be
useful for rings.
Here are a few denitions and results along familiar lines.
Denition 3 Let (R, +, .) be a ring. If S is a subset of R such that
(i) a b S and ab S whenever a, b S,
(ii) 1 S,
then we call S a subring of R.
(Condition (ii) excludes the possibility S = 0.)
Lemma 4 Let (R, +, .) be a ring and S subring of R. Then S equipped with
the addition and multiplication inherited from R is itself a ring.
3
Lemma 5 Let (A, +
A
,
A
) and (B, +
B
,
B
) be rings. If we dene addition
and multiplication on A B by
(a
1
, b
1
) + (a
2
, b
2
) =(a
1
+
A
a
2
, b
1
+
B
b
2
)
(a
1
, b
1
) (a
2
, b
2
)a =(a
1

A
a
2
, b
1

B
b
2
)
then (A B, +, ) is a ring.
We often write AB for the ring just dened and call it the external direct
sum.
Denition 6 Let R and S be rings with multiplicative identities 1
R
and 1
S
.
We say that a map : R S is a homomorphism (more precisely a ring
homomorphism) if
(i) (r
1
+r
2
) = (r
1
) +(r
2
), (r
1
r
2
) = (r
1
)(r
2
) for all r
1
, r
2
R
(ii) (1
R
) = 1
S
.
(Condition (ii) excludes the possibility (r) = 0 for all r R.)
Lemma 7 Let R and S be rings and : R S a homomorphism. Then
(R) is a subring of S.
We often write im = (R) and call it the image of .
Denition 8 Let R and S be rings and : R S a homomorphism. If
is a bijection we say that is an isomorphism (more exactly a ring isomor-
phism) and that R and S are isomorphic. We write R

= S (R is isomorphic
to S by the map ) and R

= S (R is isomorphic to S).
Lemma 9 Isomorphism is an equivalence relation. That is
(i) R

= R.
(ii) If R

= S, S

= T then R

= T.
(iii) If R

= S then S

= R.
2 Ideals, quotients and the isomorphism the-
orem
In many ways subrings are less important for ring theory than ideals.
Denition 10 Let (R, +, .) be a ring. If I is a non-empty subset of R such
that
(i) a b I whenever a, b I,
(ii) ab I whenever a R and b I,
then we call I an ideal of R.
4
We observe that I is a subgroup of (R, +) the ring R considered as an
Abelian group under addition. We take over from group theory the idea of
a coset
r +I = r +s : s I
and observe that the rst part of the proof of Lagranges theorem shows that
the cosets form a disjoint cover of R.
Lemma 11 Let I be an ideal of a ring R. Then
(i)

rR
(r +I) = R.
(ii) If r, s R then either (r +I) (s +I) = or r +I = s +I.
The remarkable thing is that we can dene addition and multiplication
of cosets in a natural way.
Lemma 12 If I is an ideal of a ring R and
r
1
+I = r
2
+I, s
1
+I = s
2
+I
then
(r
1
+s
1
) +I = (r
2
+s
2
) +I, r
1
s
1
+I = r
2
s
2
+I.
Denition 13 If I is an ideal of a ring R we write R/I for the set of cosets
of I and dene addition and multiplication on R/I by
(r +I) + (s +I) = (r +s) +I, (r +I)(s +I) = rs +I.
Lemma 14 If I is an ideal of a ring R then R/I with addition and multi-
plication as in the previous denition is a ring.
We call R/I a quotient ring.
The idea of a quotient ring gives a clean denition of arithmetic modulo
m.
Lemma 15 If m Z then
mZ = mr : r Z
is an ideal of Z.
Denition 16 If m 2 we write
Z
m
= Z/mZ.
The reader will readily identify Z/mZ for all m Z.
The next example warns us to stick to Denition 13.
5
Example 17 The set I = 0, 2 is an ideal of the ring Z
4
. We have
(0 +I)(0 +I) = 0 +I
but rs : r, s I = 0 ,= I.
Quotient rings as closely linked with homomorphisms.
Denition 18 If R and S are rings and : R S is a homomorphism we
write
ker =
1
(0) = r R : (r) = 0
and call ker the kernel of .
Lemma 19 Suppose that R and S are rings and : R S is a homomor-
phism. Then
(i) ker is an ideal of R.
(ii) (r) = s has a solution r R if and only if s im.
(iii) If (r) = s then (r

) = s if and only if r

r + ker .
We have just shown that every kernel of a homomorphism is an ideal.
The next remark shows that every ideal is the kernel of a homomorphism.
Lemma 20 Let I be an ideal of the ring R. Then the map : R R/I
given by
(r) = r +I
is a homomorphism with kernel I.
The machinery is now in place to state and prove our rst key theorem.
Theorem 21 (The isomorphism theorem) Suppose that R and S are
rings and : R S is a homomorphism. Then
R/ ker

= im.
3 Integral domains, elds and fractions
The fact that we can not necessarily cancel or divide in rings means that
they are too general for many purposes.
Denition 22 A ring (D, +, .) is called an integral domain if, whenever
ab = 0, we can deduce that a = 0 or b = 0.
6
Denition 23 A ring (F, +, .) is called a eld if (F 0, .) is an Abelian
group.
Thus a ring (F, +, .) is a eld if, whenever a F and a ,= 0 we can nd
a
1
with aa
1
= 1. The element a
1
(unique by a simple argument from
elementary group theory) is called the multiplicative inverse of a.
Lemma 24 (i) If (D, +, .) is an integral domain and ab = ac with a ,= 0
then b = c.
(ii) Every eld is an integral domain.
(iii) Every subring of an integral domain is an integral domain.
Lemma 26 below is not in the printed syllabus but this has not deterred
examiners from setting it in the past. We need denitions which, important
though they are in a more general context, are only included here in order
to allow us to state the lemma.
Denition 25 (i) We say that an ideal I of a ring R is maximal if I ,= R
but if J is an ideal with J I and J ,= I then J = R.
(ii) We say that an ideal P in a ring R is prime if ab P implies a P
or b P.
Lemma 26 Suppose that I is an ideal in a ring R.
(i) I is maximal if and only if R/I is a eld.
(ii) I is prime if and only if R/I is an integral domain.
We already know quite a lot of elds including R, C and Q. We also know
some nite elds.
Lemma 27 (i) If p is a prime then Z
p
is a eld.
(ii) If m is not a prime then Z
m
is not an integral domain. [m 2]
We digress briey to discuss characteristics. If R is a ring, n a strictly
positive integer and a an element of R let us write
na = a +a + +a
. .
n
,
(n)a = na and 0a = a.
Lemma 28 Let R be a ring with multiplicative identity 1
R
.
(i) The map : Z R given by (m) = m1
R
is a homomorphism.
(ii) The set im of all elements of the form m1
R
is isomorphic to Z or
Z
n
for some n 2.
7
Denition 29 With the notation of Lemma 28, if im is isomorphic to Z
n
we say that R has characteristic n. If im is isomorphic to Z we say that R
has characteristic (or, in some texts characteristic 0).
There is another way of viewing this idea.
Lemma 30 If A is a subset of a ring R then there is a smallest subring B
containing A. (In other words there exists a subring B of R such that B A
and if C is any subring of R with C A then C B.)
We call B the ring generated by A. If A = so that B is the smallest ring
in R we call B the prime subring of R. (Here prime is used as in primal
scream, the rst or underlying scream.)
Lemma 31 With the notation of Lemma 28, im is the smallest subring of
R. Thus the primal ring of R is isomorphic to Z or Z
n
for some n 2.
It is natural to identify the prime subring with Z or Z
n
and write m =
(m) = m1
R
.
The notion of characteristic is most useful when applied to integral do-
mains.
Lemma 32 (i) The characteristic of an integral domain is either a prime
or .
(ii) The prime subring of an integral domain may be identied with Z or
Z
p
where p is a prime.
(iii) If (R, +, .) is an integral domain then every non-zero element of the
additive group (R, +) has order the characteristic of the integral domain.
Later on we shall see that polynomials provide important examples of
integral domains which are not elds. For the moment the only obviously
interesting example we know of an integral domain which is not a eld is Z.
However this is such an important example that it justies by itself all the
work we shall do in the remainder of this section.
From the point of view of late nineteenth century mathematics we shall
be showing that the rationals can be constructed from the integers. God
created the integers, all the rest is the work of man. As a bonus we nd that
the same proof gives the more modern sounding result that every integral
domain can be embedded in a eld. (From the point of view of the plain
man we are just describing fractions with a great deal of caution.)
Lemma 33 If (D, +, .) is an integral domain write D

= D 0. The
relation dened on D D

by
(r
1
, s
1
) (r
2
, s
2
) if r
1
s
2
= r
2
s
1
8
is an equivalence relation.
If (r
1
, s
1
) (r
2
, s
2
) and (u
1
, v
1
) (u
2
, v
2
) then
(r
1
v
1
+s
1
u
1
, s
1
v
1
) (r
2
v
2
+s
2
u
2
, s
2
v
2
) and (r
1
u
1
, s
1
v
1
) (r
2
u
2
, s
2
v
2
).
Lemma 34 Continuing with the assumptions and notation of Lemma 33 let
us write k for the set D/ of equivalence classes
r
s
= (r

, s

) D D

: (r

, s

) (r, s).
Then we may dene addition and multiplication on k by
r
s
+
u
v
=
rv +su
sv
and
r
s
u
v
=
ru
sv
.
With this addition and multiplication, (k, +, .) is a eld.
If we dene : D k by
(r) =
r
1
then is an injective homomorphism and so

D = im is a subring of k
isomorphic to D.
It is natural to identify

D with D by writing
r =
r
1
for each r D. We call k the eld of fractions of D.
We have thus characterised integral domains.
Lemma 35 A ring D is an integral domain if and only if it is isomorphic
to a subring of a eld.
If we use the natural identication of

D with D we can restate Lemma 35 in
a more striking manner.
Lemma 36 A ring D is an integral domain if and only if it embeds in a
eld.
The naturalness of our construction is emphasised by the Lemma 38 be-
low. We need a preliminary remark.
Lemma 37 If A is a subset of a eld F then there is a smallest subeld B
containing A. (In other words there exists a subeld B of F such that B A
and if C is any subeld of F with C A then C B.)
9
We call B the eld generated by A. If A = so that B is the smallest eld
in R we call B the prime subeld of R.
Lemma 38 Suppose that (F, +, .) is a eld and D a subring of F. Let Q
be the smallest subeld of F containing D. Then there is an isomorphism
: Q k such that (r) =
r
1
for all r D.
Lemma 32 (ii) tells us that the prime subring of a eld may be identied
either with Z or Z
p
where p is a prime. If the prime subring is Z
p
then it is
also a eld and so the prime subeld of F. If the prime subring is Z we may
use Lemma 38 to identify the prime subeld.
Lemma 39 The prime subeld of a eld may be identied in a natural man-
ner with Q or Z
p
where p is a prime.
So far as the syllabus is concerned this concludes the section. What
follows is easy but not on the syllabus.
If we start with D = Z the construction above yields k = Q as a eld. But
mathematicians are also interested in order. Recall that there is a relation
> on Z. We say that a > b if b a > 0. The properties of > follow from the
following rules
(A) If a Z then exactly one of the following is true: a = 0 or a > 0 or
a > 0.
(B) If a, b Z, a > 0 and b > 0 then a +b > 0 and ab > 0.
Lemma 40 Let D = Z in Lemma 33. If (r
1
, s
1
) (r
2
, s
2
) and r
1
s
1
> 0 then
r
2
s
2
> 0.
Lemma 41 Let D = Z in Lemma 34. Then we may dene a relation > on
k = Q by the conditions
r
s
>
u
v
if
r
s

u
v
> 0
and
r
s
> 0 if rs > 0.
The following results hold
(A) If a Q then exactly one of the following is true: a = 0 or a > 0 or
a > 0.
(B) If a, b Q, a > 0 and b > 0 then a +b > 0 and ab > 0.
In the language of the analysis course C9, Q is an ordered eld.
10
4 Unique factorisation, Euclidean and prin-
cipal ideal domains
In this section I shall give a rather cold blooded and abstract treatment of
factorisation in rings. Historically the subject was an exciting and confusing
one. There are several theorems in number theory and elsewhere, in par-
ticular the Wiles-Taylor theorem (formerly Fermats last theorem), which
looked easy to prove provided the obvious factorisation theorem holds and
very distinguished mathematicians fell into the trap of assuming that which
is obvious is true. On the other hand when unique factorisation did indeed
hold, it provided a very powerful tool. We give a simple example by proving
an elegant theorem of Fermat (Theorem 57) via unique factorisation at the
end of this section.
There are two immediate problems, the rst obvious and easily overcome,
the second less so. The easy problem is illustrated when we try to extend
the unique factorisation theorem from N (which is, of course, not a ring) to
the ring Z. We observe that
15 = (3) 5 = 3 (5)
and that
15 = (3) (5) = 3 5
so some restatement of the theorem is necessary. We set up the machinery
to deal with this in the next denition and the lemma that follows.
Denition 42 Let R be a ring. We say that u R is a unit if there exists
an v R such that uv = 1. (Thus u is a unit if it has a multiplicative
inverse). We say that r and s are associates if there exists a unit u with
r = su.
We extend a standard notation of elementary number theory to any ring R.
If a, b, c R and a = bc we say that b divides a and write b[a.
Lemma 43 (i) Consider a ring R. The relation r is an associate of s is an
equivalence relation on R.
(ii) Consider an integral domain D. Two elements a, b D are associates
if and only if a[b and b[a.
As examples we note that all non-zero elements in a eld are units and so all
pairs of non-zero elements are associates. In Z the units are 1 and 1 and
the only associate of n is n.
The second problem is clearly marked by the two denitions that follow
together with Example 47
11
Denition 44 Let R be a ring. We say that q R is irreducible if it is not
a unit and whenever a[q then a is either a unit or an associate of q.
Denition 45 Let R be a ring. We say that p R is prime if it is neither
0 nor a unit and whenever p[ab [a, b R] then p[a or p[b.
Lemma 46 Any prime is irreducible.
Unfortunately there exist rings in which not all irreducible elements are
prime.
Example 47 Let
r = n +m

(5) : n, m Z
and let N : R Z
+
be given by
N(n +m

(5)) = [n +m

(5)[
2
= n
2
+ 5m
2
.
(i) R is a subring of C so an integral domain.
(ii) N(ab) = N(a)N(b) for all a, b R.
(iii) The units of R are 1 and 1.
(iv) 6 = 2 3 = (1 +

(5)) (1

(5)).
(v) The elements 2, 3, (1 +

(5)) and (1

(5)) are irreducible.

In the development of the theory of factorisation for Z (strictly speaking
for N, which is not a ring) carried out in Course C3 we showed that every
irreducible element is prime by using Bezouts theorem. Fortunately there
exist a large class of integral domains for which something rather close to
Bezouts theorem holds the so called principal ideal domains.
Denition 48 If R is a ring we say that an ideal I of R is principal if it is
generated by a single element a, in other words
I = aR = ar : r R.
We also write I = (a).
Denition 49 An integral domain D is said to be a principal ideal domain
if every ideal I of D is principal.
Lemma 50 In a principal ideal domain every irreducible element is prime.
Lemma 51 In a principal ideal domain every element which is neither a
unit nor 0 is the product of a nite number of irreducible elements.
12
Once we have Lemmas 50 and 51 the same easy, if slightly tedious, argu-
ments that we used to prove unique factorisation for the integers in Course C3
give us a unique factorisation theorem for principal ideal domains.
Theorem 52 Let D be a principal ideal domains.
(i) If r D is non-zero we can nd a unit u and irreducible elements a
1
,
a
2
, . . . a
n
such that
r = ua
1
a
2
. . . a
n
.
(ii) Suppose that u and v are units and a
1
, a
2
, . . . a
n
, b
1
, b
2
, . . . b
m
are
irreducible with
ua
1
a
2
. . . a
n
= vb
1
b
2
. . . b
m
.
Then m = n and by renumbering we can ensure that a
j
and b
j
are associates
for all 1 j n.
I said that principal ideal domains are common but I have given no tech-
nique for proving that a domain is a principal ideal domain. Not surprisingly,
one way is to seek an analogue of Euclids algorithm from Course C3.
Denition 53 We say that an integral domain D is a Euclidean domain if
we can nd a function : D 0 Z
+
(called a Euclidean function) such
that
(i) if a[b then (a) (b),
(ii) given a R and b R with b ,= 0 we can nd q and r such that
a = qb +r and either r = 0 or (r) < (b).
Lemma 54 If D is a Euclidean domain with Euclidean function then u is
a unit of D if and only if u ,= 0 and (u) = (1).
Theorem 55 Every Euclidean domain is a principal ideal domain.
We are now in position to give the reader a genuinely novel example of a
domain with unique factorisation.
Example 56 (The Gaussian integers) Consider
R = n +mi : n, m Z
and let : R Z
+
be given by
(n +mi) = [n +mi[
2
= n
2
+m
2
.
(i) R is a subring of C so an integral domain (called the Gaussian inte-
gers).
(ii) is a Euclidean function, so R is a Euclidean domain.
13
It is quite hard to give examples of a principal ideal domains which are
not Euclidean (presumably, not because they are uncommon but because it
is hard to show that no Euclidean function could possibly exist). However,
they exist and are given, or at least referenced, in the heavier algebra texts.
The remainder of this section is not on the syllabus. In it we use factori-
sation in the Gaussian integers to prove a theorem of Fermat.
Theorem 57 (Fermat) We work in N. An odd prime p can expressed as
the sum of the squares of two integers
p = n
2
+m
2
if and only if p is of the form 4N + 1 for some integer N.
The only if part is easy, but to prove the if part we need the following lemma
on Gaussian integers.
Lemma 58 We work in N except in part (i). Suppose that p is a prime such
that we can nd integers x and y and an integer c coprime to p such that
x
2
+y
2
= cp. Then
(i) p is not a prime for the Gaussian integers,
(ii) there exist integers n and m such that p = n
2
+m
2
.
Combining this with the following simple consequence of Wilsons theorem
(Course C3) we obtain Fermats theorem (Theorem 57).
Lemma 59 Suppose that p is of the form 4N + 1 for some integer N.
(i) We can solve the congruence x
2
1 mod p.
(ii) We can nd an integer x with 1 x p/2 such that x
2
+ 1
2
0
mod p.
5 Polynomials over rings
The denition of polynomials over rings is complicated by the phenomenon
illustrated in the next example.
Example 60 Let f : Z
2
Z
2
be dened by f(x) = x
2
+ x. Then f(x) = 0
for all x.
We must thus decide whether to dene a polynomial by its values (which is
what an analyst would do) or by its coecients. As algebraists we decide to
dene it by its coecients and enshrine our choice in the following denition.
14
Denition 61 The polynomial ring R[X] over R is the collection of se-
quences
r = (r
0
, r
1
, r
2
, . . . )
where each r
j
R and only nitely many of the r
j
are non-zero. We dene
r +s = (r
0
+s
0
, r
1
+s
1
, r
2
+s
2
, . . . )
and
rs = t
where t
j
=

j
k=0
r
j
s
kj
.
Neither the next lemma nor its proof present any surprises.
Lemma 62 The polynomial ring R[X] over R is a ring.
Finally we remove the mask of the mysterious stranger and write
r =

j=0
r
j
X
j
=
N

j=0
r
j
X
j
where N is any integer suciently large that r
j
= 0 for all j N. Of course,
the X
j
are simple place holders (we call X an indeterminate). If we want to
talk about the value of a polynomial we need a simple homomorphism (the
pont evaluation map).
Lemma 63 (Point evaluation) If x R the map
x
: R[X] R given by

x
_
N

j=0
r
j
X
j
_
=
N

j=0
r
j
x
j
is a homomorphism.
As might be expected, we write
x
p = p(x). The degree of a polynomial is
dened in the obvious manner.
Denition 64 If
p(X) =
N

j=0
r
j
X
j
and r
N
,= 0 then we say that p has degree N and write p = N. If p = 0 we
write p = .
In this course we conne ourselves to polynomials over integral domains.
15
Lemma 65 If D is an integral domain then so is the polynomial ring D[X].
Lemma 66 If p and q are polynomials over an integral domain then
(i) (p +q) max(p, q),
(ii) (pq) = p +q.
If we restrict ourselves still further to elds we can use a very powerful
result.
Theorem 67 (Euclidean division) If a and b are polynomials over a eld
F and b ,= 0 then we can nd polynomials q and r such that a = qb + r and
r < a.
As an immediate corollary we have a key result.
Lemma 68 The polynomial ring over a eld is a Euclidean domain and so
a principal ideal domain.
Notice that Lemma 68 does not extend even to such a well behaved integral
domain as Z.
Example 69 The ideal generated by 2 and X is not principal in Z.
In the next section we shall see how this problem can be partially overcome by
embedding the integral domain in its quotient eld. A rather trivial example
of this technique is used to derive Lemma 71 from Lemma 70 (iii) below.
Lemma 70 Let us work in the ring of polynomials over a eld F.
(i) If p is a polynomial and p(a) = 0 for some a F then we can nd a
polynomial q such that p(X) = (X a)q(X).
(ii) If p is a polynomial and p(a
1
) = p(a
2
) = = p(a
m
) = 0 for some
distinct a
1
, a
2
, . . . , a
m
F then we can nd a polynomial q such that
p(X) = (X a
1
)(X a
2
) . . . (X a
m
)q(X).
(iii) A polynomial of degree n has at most n zeros in F.
Lemma 71 Suppose that D is an integral domain and p is a polynomial in
D[X] of degree n 0. Then there are at most n distinct solutions of p(x) = 0
with x D.
16
So far as the syllabus is concerned this concludes the section. What
follows is easy but not on the syllabus.
Suppose we consider the particular eld R. We know that the polynomials
on R form an integral domain but in this special case we can also dene an
order. If p(X) =

n
j=0
a
j
X
j
with a
n
,= 0 we say that p > 0 if a
n
> 0. If p is
the zero polynomial we say that p 0. The following two rules are easy to
check.
(A) If p R[X] then exactly one of the following is true: p = 0 or p > 0
or p > 0.
(B) If p, q R[X], p > 0 and q > 0 then p +q > 0 and pq > 0.
If p, q R[X] we write p > q if p q > 0.
Exactly as Lemmas 40 and 41 we can extend this order to the eld of
quotients.
Lemma 72 Let D = R[X] in Lemma 33. If (r
1
, s
1
) (r
2
, s
2
) and r
1
s
1
> 0
then r
2
s
2
> 0.
Lemma 73 Let D = R[X] in Lemma 34. Then we may dene a relation >
on k = K by the conditions
r
s
>
u
v
if
r
s

u
v
> 0
and
r
s
> 0 if rs > 0.
The following results hold
(A) If a K then exactly one of the following is true: a = 0 or a > 0 or
a > 0.
(B) If a, b K, a > 0 and b > 0 then a +b > 0 and ab > 0.
In the language of the analysis course C9, K is an ordered eld but of a type
rather dierent from Q and R.
Remember that Q and R obeyed the axiom of Archimedes. If a, b > 0
then we can nd an n Z
+
such that
na = a +a + +a
. .
n
> b.
However, in K, we have X, 1 > 0 yet
n = n1 = 1 + 1 + + 1
. .
n
X
17
for all n. In a more striking, but equivalent, formulation
1
n
>
1
X
for all n Z with n 1. Thus we have an ordered eld containing Z
for which 1/n 0. Ordered elds like K which do not obey the axiom of
Archimedes are called non-Archimedean.
6 Unique factorisation for polynomials
Once we have a denition for the ring R[X] of polynomials over a ring R it is
easy to dene the ring R[X
1
, X
2
, . . . , X
n
] of polynomials in n indeterminates
X
1
, X
2
, . . . , X
n
by using the inductive denition
R[X
1
, X
2
, . . . , X
k+1
] = R[X
1
, X
2
, . . . , X
k
][X
k+1
].
It is not hard to see that this abstract denition corresponds to our intu-
itive picture of polynomials in several variables provided that we dene the
polynomial by its coecients rather than its values. The typical element of
R[X
1
, X
2
, . . . , X
n
] can be written
P(X
1
, X
2
, . . . , X
n
) =
N

i
1
=1
N

i
2
=1

N

in=1
a
i
1
,i
2
,...,in
X
i
1
X
i
2
. . . X
in
and addition, multiplication and point evaluation
P(x
1
, x
2
, . . . , x
n
) =
N

i
1
=1
N

i
2
=1

N

in=1
a
i
1
,i
2
,...,in
x
i
1
x
i
2
. . . x
in
for x
1
, x
2
, . . . , x
n
R. The details which echo the previous section are just
as trivial here as they were there and I shall omit them.
Simple induction using Lemma 62 and Lemma 65 gives the appropriate
version of those lemmas.
Lemma 74 If R is ring then so is R[X
1
, X
2
, . . . , X
n
].
Lemma 75 If D is an integral domain then so is D[X
1
, X
2
, . . . , X
n
].
Unfortunately, although Lemma 68 tells us that the polynomial ring F[X]
over a eld F is a Euclidean domain and so a principal ideal domain this
result does not extend to polynomials in several indeterminates.
18
Example 76 If F is a eld and we work in the ring F[X
1
, X
2
] then the ideal
generated by X
1
and X
2
is not principal.
In spite of this, it turns out that unique factorisation still holds for
F[X
1
, X
2
, . . . , X
n
]. If we reect on how we might prove this, it seems natural
to use induction on n. In order to set out the induction it is natural to make
the following denition based on the statement of Theorem 52
Denition 77 Let D be an integral domain. We say that D is a unique
factorisation domain if the following two statements hold.
(i) If r D is non-zero we can nd a unit u and irreducible elements a
1
,
a
2
, . . . a
n
such that
r = ua
1
a
2
. . . a
n
.
(ii) Suppose that u and v are units and a
1
, a
2
, . . . a
n
, b
1
, b
2
, . . . b
m
are
irreducible with
ua
1
a
2
. . . a
n
= vb
1
b
2
. . . b
m
.
Then m = n and by renumbering we can ensure that a
j
and b
j
are associates
for all 1 j n.
The following point should be noted.
Lemma 78 In a unique factorisation domain every irreducible is a prime
(so the two terms are synonymous).
Our aim would be achieved if we could prove the following theorem.
Theorem 79 If D is a unique factorisation domain then so is D[X].
Simple induction gives the next result.
Theorem 80 If D is a unique factorisation domain then so is D[X
1
, X
2
, . . . , X
n
].
By Theorem 52 every principal ideal domain is a unique factorisation domain
and we have a very strong result.
Theorem 81 If D is a principal ideal domain then D[X
1
, X
2
, . . . , X
n
] is a
unique factorisation domain.
How might we prove Theorem 79? Consider the special case when D = Z.
We know nothing about Z[X] but we do know that Z embeds naturally in
its eld of fractions Q and that unique factorisation holds for Q[X] (by The-
orem 52). Since Z[X] embeds naturally in Q[X] we can proceed as follows.
Suppose we have a polynomial 6X
3
+24X
2
+24X +6 in Z[X]. We may not
19
be able to factorise it in Z[X] but we can certainly factorise it in Q[X]. Take
one such factorisation
6X
3
+ 24X
2
+ 24X + 6 =
42
25
_
5
2
X +
5
2
__
10
7
X
2
+
30
7
X +
10
7
_
.
By clearing fractions and cancelling (Z is, after all, the quintessential integer
domain) we arrive at
6X
3
+ 24X
2
+ 24X + 6 = 2.3(X + 1)(X
2
+ 3X + 1)
and a little thought shows that if (
5
2
X +
5
2
) and (
10
7
X
2
+
30
7
X +
10
7
) were
irreducible in Q[X] then (X +1) and (X
2
+3X +1) are irreducible in Z[X].
Although the proof of Theorem 79 given below is quite complicated it
is my belief that any one seeking to develop the idea just given into a cast
iron proof of the uniqueness of factorisation for Z[X] would be lead almost
inevitably to something like it. One the proof is written down it is a simple
matter to replace Z by a general unique factorisation domain.
Denition 82 Let A be a subset of a ring R, such that A contains a non-zero
element. We say that a is a highest common factor of A if
(i) a[x for all x A,
(ii) if a

[x for all x A then a

[a.
Lemma 83 Any nite subset A of a unique factorisation domain such that
A contains a non-zero element has a highest common factor.
In fact the following is true though we shall not use it.
Lemma 84 Any subset A of a unique factorisation domain such that A con-
tains a non-zero element has a highest common factor.
In what follows we work under the following standing hypothesis.
Standing hypothesis We have a unique factorisation domain D embedded
in its eld of fractions F. We use the natural embeddings of D in F, F in
F[X], D[X] in F[X] and D in D[X].
We say that a polynomial p(X) =

n
j=1
a
j
X
j
in D[X] is primitive if 1 is
a highest common factor of a
j
: 0 j n. We observe that any q D[X]
can be written as q = p with D and p primitive.
Lemma 85 Under our standing hypothesis,
(i) The units of D[X] are precisely the units of D.
(ii) Any q F[X] can be written as q = p with F and p a primitive
polynomial in D[X].
(iii) If p, p

are primitive polynomials in D[X] and p =

for some
,

F then p and p

are associates in D[X], that is there exists a unit

D such that p = p

.
20
Lemma 86 Under our standing hypothesis, if p and q are primitive polyno-
mials in D[X] so is pq.
Lemma 87 (Gauss lemma) Under our standing hypothesis, a polynomial
p D[X] is irreducible if and only if it is either (a) an irreducible element
of D or (b) it is primitive in D[X] and irreducible in F[X].
Theorem 88 Under our standing hypothesis, D[X] is a unique factorisation
domain.
Theorem 88 is just Theorem 79 so we are done. We cease working under the
standing hypothesis.
The reader may suspect that it is hard to establish if a particular poly-
nomial is irreducible. She is right
1
. One useful tool is due to Eisenstein. We
give it for Q[X] though it can be generalised.
Lemma 89 (Eisensteins criterion) Suppose that
P(X) = a
0
+a
1
X +a
2
X
2
+ +a
n
X
n
is a polynomial in Z[X] (i.e. P has integral coecients). If there exists a
prime number p such that p,[a
n
, p[a
n1
, p[a
n2
, . . . , p[a
0
but p
2
,[a
0
, then P
is irreducible over Q[X].
As an example of how it used consider the following.
Lemma 90 If p is prime then 1 + X + X
2
+ + X
p1
is irreducible over
Q[X].
The trick here is to make the substitution Y = X1 and to base our algebra
on the recollected formula
1 +x +x
2
+ +x
p1
=
x
n
1
x 1
=
(y 1)
p
1
y
,
from the days before we did abstract algebra.
The formula
(X 1)(X
3
+X
2
+X + 1) = X
4
1
= (X
2
1)(X
2
+ 1)
= (X 1)(X + 1)(X
2
+ 1)
shows us that (X
3
+ X
2
+ X + 1) = (X + 1)(X
2
+ 1) and suggests how to
prove the converse.
Lemma 91 If n is composite then 1+X+X
2
+ +X
n1
is not irreducible
over Q[X].
1
At least, as far as human beings are concerned. There is an algorithm which will
always work and computer algebra programs can handle quite complicated cases.
21
7 Fields and their simple extensions
We already know that it may be useful to embed a eld in a larger eld. Not
all polynomials are soluble in R but they are in the larger eld C. In this
section we study other extensions.
We begin with a couple of examples.
Example 92 Consider Q as a subeld of C.
(i) Let be a transcendental number. Then Q() the smallest subeld of
C containing Q and is isomorphic to the eld of fractions of Q[X].
(ii) Let be a root of z
2
+ z + 1 = 0 in C. Then each element of Q()
the smallest subeld of C containing Q and may be written in exactly one
way as a +b with a, b Q.
Of course, we may not be in the happy position of Example 92 and nd
our extension ready made as a subeld of some larger eld.
Denition 93 We say that L is an extension of a eld K if there is an
injective homomorphism : K L (i.e. if K is isomorphic to a subeld of
L).
Having made this denition we shall usually ignore it and treat K as a
subeld of L with the natural identication k = (k) for k K. However,
there are one or two points where we need to act more cautiously.
Denition 94 We say that L is a simple extension of a eld K if we can
nd an element u L such that u and K generate L. We write L = K(u).
We now see that choices of Example 92 are, in some sense, typical.
Denition 95 Suppose that L is a simple extension of K with L = K(u).
If u satises a polynomial equation
u
n
+a
n1
u
n1
+a
n2
u
n2
+ +a
0
= 0
with a
j
K we say that u is algebraic and that L = K(u) is an algebraic
extension of K. If not we say that u is transcendental and that L = K(u)
is a transcendental extension of K.
Lemma 96 If K(u) is a transcendental extension of a eld K then K(u) is
isomorphic to k the eld of fractions of K[X] under the natural isomorphism
: k K(u) which has (a) = a for all a K and (X) = u.
The more interesting case of algebraic extension is dealt with in a series
of simple but important lemmas.
22
Lemma 97 If K(u) is an algebraic extension of a eld K then u is the zero
of one and only one monic irreducible polynomial p in K[X]. If q K[X]
and q(u) = 0 then q = hp for some h K[X] (that is q is in the ideal (p)
generated by p).
Denition 98 With the notation and hypotheses of Lemma 97 we say that
p is the minimal polynomial of u. If p has degree n we say that u has degree
over K of value n. We also write [u : K] = n.
Lemma 99 With the notation and hypotheses of Lemma 97 the mapping
: K[X] K(u) given by
(f) = f(u)
is a surjective homomorphism with kernel (p).
Lemma 100 With the notation and hypotheses of Lemma 97 K(u) is iso-
morphic to K[X]/(p). Thus every algebraic extension of K is isomorphic to
the quotient of K[X] by the ideal generated by some irreducible polynomial.
It is interesting to ask what happens to p when we factorise it in K(u)[X].
Since p(u) = 0 we know that X u is a factor of p(X) (by the remainder
theorem Lemma 70 (i)) so p will have linear factors. We shall discuss this
in detail in the next section but for the moment we just give an example
to show that, even in K(u)[X], p may not factorise completely into linear
factors.
Example 101 Consider Q as a subeld of C. Let p Q[X] be given by
p(X) = X
4
3.
(i) The polynomial p is monic and irreducible over Q[X].
(ii) If L is the eld generated by Q and 3
1/4
(the positive fourth root of 3
then L = Q(3
1/4
) and p(3
1/4
) = 0. In Q(3
1/4
), p factors into irreducibles as
p(X) = (X 3
1/4
)(X + 3
1/4
)(X
2
+ 3
1/2
).
(iii) If L is the eld generated by Q and 3
1/4
i then L = Q(3
1/4
i) and
p(3
1/4
i) = 0. In Q(3
1/4
i), p factors into irreducibles as
p(X) = (X 3
1/4
i)(X + 3
1/4
i)(X
2
3
1/2
).
We complete the unstarred part of this section with another simple but
useful observation.
23
Lemma 102 (i) If K is a subeld of L then L can be considered as a vector
space of K in a natural manner.
(ii) If L is a transcendental extension of K then L is innite dimensional
as a vector space over K.
(iii) If L = K(u) and u is algebraic of degree n then L has dimension n
as a vector space over K. The elements 1, u, . . . , u
n1
form a basis for L.
If K is a subeld of L we write [L : K] for the dimension (possibly ) of L
as a vector space over K. We call [L : K] the degree of L over K.
Lemma 103 (Tower law) If K is a subeld of L and L is a subeld of M
then [M : K] = [M : L][L : K].
The rest of this section is not on the syllabus and, even if time allows
to be covered, will only be sketched. Details may be found in the opening
chapters of most texts on Galois theory (e.g. Chapter 6 of [4]).
We know that we stand on the shoulders of giants. The only question to
be answered is whether we see any further. Our work so far enables us to
solve two geometric problems that the Greeks were unable to solve. Both deal
with ruler and compass constructions. The Greeks asked which constructions
were possible with a ruler and compass alone. More prosaicly, but essentially
equivalently we ask which points (x, y) R
2
can be constructed starting
from (0, 0) and (0, 1) using ruler and compass alone.
Lemma 104 Consider a ruler and compass construction starting from (0, 0)
and (0, 1) in which the point (x
j
, y
j
) is obtained at the jth step. If we write
R
0
= Q and R
j
= R
j1
(x
j
)(y
j
) (that is R
j
is the smallest subeld of R
2
containing R
j1
, x
j
and y
j
) then [R
j
, R
j1
] takes the value 1 or 2. Thus, by
the tower law, [R
j
, Q] = 2
r
for some integer r 0.
Theorem 105 (The Delian problem) (i) The polynomial X
3
2 = 0 is
irreducible over Q.
(ii) If in Lemma 104 we have (x
j
, , y
j
) = (0, 2
1/3
) then [R
j
, Q] must be
divisible by 3.
(iii)We can not construct the point (0, 2
1/3
) by ruler and compass con-
struction starting from (0, 0) and (0, 1).
(iv) It is impossible using ruler and compass alone to construct a cube
whose volume is double that of a cube of given edge.
There are many people for whom only the useful is worthwhile. Hogben
dismisses Plato, Eudoxus and Euclid as men who who treated mathematics
as a respectable form of relaxation for the opulently idle. Even Kline in his
24
magisterial history Mathematical Thought from Ancient to Modern Times [6]
sometimes reminds one of a school teacher in charge of a class of brilliant
pupils who will persist in chasing the butteries of pure mathematics rather
than applying themselves to the stern task of understanding the real world.
Come on master Gauss stop looking at those cyclotomic polynomials you
have three orbits to compute before bedtime! Even if they understand the
thrill of seeing a problem solved that has baed mankind for 2000 years they
see that thrill as a sinful diversion.
According to a story current in antiquity the Delians, suering from pesti-
lence, sent to the oracle who told them to double the size of a particular cubic
altar to Apollo. They did as they were told by doubling the length of each
of its sides. When the plague continued they consulted Plato who explained
that the god wished his altar doubled in volume (preserving the cubic shape).
The god, continued Plato, demanded this not because he wanted or needed
such an altar but in order to censure the Greeks for their indierence to math-
ematics and lack of respect for geometry. The gods no longer punish societies
which reject the pursuit of knowledge for its own sake quite so directly but
perhaps such societies punish themselves.
Theorem 106 (The trisection problem) (i) We can construct the point
(cos /3, sin /3) by ruler and compass construction starting from (0, 0) and
(0, 1).
(ii) If we could trisect every angle by ruler and compass construction we
could construct (cos /9, sin /9) by ruler and compass construction starting
from (0, 0) and (0, 1).
(iii) If = cos /9 then 4
3
3
1
2
= 0. If = 2 then
3
3 1 = 0.
(iv) The polynomial X
3
3X 1 = 0 is irreducible over Q.
(v)We can not trisect every angle by a ruler and compass construction.
The credit for these two theorems goes to Wantzel. Possibly if he had done
something romantic like being killed in a duel mathematicians would have
had the courtesy to attach his name to his theorems.
Suppose we could prove the following theorem.
Theorem 107 (Lindeman) The number is transcendental.
Then we would be able to solve a third great problem of antiquity.
Theorem 108 (Impossibility of circle squaring) (i) It is impossible that
R
j
.
(ii) We cannot construct a square of area equal to a given circle by a ruler
and compass construction.
25
There are now fairly short proofs of Theorem 107 (see, for example, Ian
Stewarts beautiful Galois Theory [8] Chapter 6) but, so far as I know, no
easy ones.
If we consider a regular polygon with n sides inscribed in the unit circle
in such a way that one vertex is at (0, 1) we see that the vertices are at points
(x
r
, y
r
) given by x
r
+ iy
r
=
r
where = exp(2i/n) (so the
r
are the rth
roots of unity. The constructibility of a regular polygon with n sides by a
ruler and compass construction is thus closely linked to the polynomial
X
n
1 = (X 1)(1 +X +X
2
+ +X
n1
)
and so to the cyclotomic polynomial 1 +X +X
2
+ +X
n1
. In particular,
though we shall not do it, it is not hard to get from Lemma 90 to the
statement that the regular p-gon (with p a prime) is only constructible by
a ruler and compass construction if p 1 is a power of 2. As a very young
man, Gauss showed the reverse (if p 1 is a power of 2 the regular p-gon
is constructible). It is said that it was this discovery that decided him on a
mathematical career. The details of the mathematics involved may be found
in [8], Chapter 17.
8 Splitting elds of polynomials
In Lemmas 96 to 100 we derived the properties of simple extensions but took
the simple extensions as given. Clearly, there always exists a transcendental
extension of a given eld K since the eld of fractions of K[X] is such an
extension. Moreover, Lemma 96 tells us that (up to isomorphism) this exten-
sion is unique. Does there always exist an algebraic extension corresponding
to a given irreducible polynomial and is it unique (up to isomorphism)?
The obvious way forward is pointed out by Lemma 100.
Lemma 109 If K is a eld and p is irreducible in K[X] then L = K[X]/(p)
is a eld containing (an isomorphic copy of ) K. We can nd u L such
that L = K(u) is simple algebraic extension of K and X u is a factor of
p(X) in L.
The only problem here is to show that K[X]/(p) is a eld and this follows
from the analogue of Bezouts theorem for principal ideal domains. Unique-
ness is simple.
Lemma 110 Suppose that K is a eld and p is irreducible in K[X]. If
K(u
1
) and K(u
2
) are simple algebraic extensions of K such that X u
j
is
a factor of p(X) in K(u
j
) then there is an isomorphism : K(u
1
) K(u
2
)
with (a) = a for a K and (u
1
) = u
2
.
26
Thus in Example 101 we know without further computation that Q(3
1/4
)

=
Q(3
1/4
i).
Repeated use of Lemma 109 gives the theorem which the last section lead
us to expect.
Theorem 111 If K is a eld and p K[X] there exists a eld L containing
(an isomorphic copy of ) K such that [L : K] < and we can nd A K,

1
,
2
, . . . ,
n
L such that
p(X) = A(X
1
)(X
2
) . . . (X
n
)
We say that p splits over L. In order to obtain a uniqueness result we need
to tighten up the conditions of the theorem.
Denition 112 If K is a subeld of the eld L and p K[X] we say that
L is a splitting eld for p over K if
(i) p factorises into linear factors
p(X) = A(X
1
)(X
2
) . . . (X
n
)
over L.
(ii) If p factorises into linear factors over a subeld L

of L then L

= L.
Observe that condition (ii) can be replaced by the statement L = K(
1
,
2
, . . . ,
n
)
the eld generated by K,
1
,
2
, . . .
n1
and
n
.
The uniqueness theorem is now easy to state.
Theorem 113 Suppose that K is a eld and p K[X]. If L and L

are
splitting elds of p then there is an isomorphism : L L

with (a) = a
for all a K.
There may be many dierent ways to go from K to a splitting eld by
adjoining roots and Theorem 113 is slightly harder to prove than might be
expected.
The following lemma contains the key idea.
Lemma 114 Let K be a eld, p K[X] and let L be a splitting eld for
p over K. Suppose that L

is a eld containing a subeld K

isomorphic to
K under the isomorphism i such that i(p) splits in L

. (Here, if p(X) =

n
r=0
a
r
X
r
we write i(p)(X) =

n
r=0
i(a
r
)X
r
.) Then there is an injective
homomorphism j : L L

such that j[
K
= i.
This is as far as we shall go with the study of splitting elds but the
following remark (which is not on the syllabus) seems worth making. We
need results on countability from course C3.
27
Lemma 115 (i) If K is a countable subeld of L and [L : K] < then L
is countable.
(ii) If K is a countable eld we can nd a countable eld L containing
(an isomorphic copy of ) K such that every polynomial p K[X] splits in
L[X].
(iii) If K is a countable eld we can nd countable elds K
j
with
K = K
0
K
1
K
2
. . .
with K
j1
a subeld of K
j
such that every polynomial p K
j1
[X] splits in
K
j
[X].
(iv) If K is a countable eld we can nd a countable eld L containing
(an isomorphic copy of ) K such that every polynomial p L[X] factors
completely into linear factors.
The same idea gives the following more striking result.
Lemma 116 There is a countable subeld F of C with F Q such that
every polynomial in F[X] has a root in F.
Thus, from the point of view of a dyed in the wool algebraist, the construction
of the uncountable eld C in order to have the fundamental theorem of
algebra is a reckless extravagance.
9 Finite elds
In this short but interesting section we nd all nite elds explicitly.
Our rst step is already substantial.
Lemma 117 If F is a nite eld then F has characteristic p a prime (that is
F has prime eld (an isomorphic copy of ) Z
p
). The eld F has p
n
elements
where [F : Z
p
] = n.
The second step is also remarkable.
Lemma 118 Let (F, +, .) be a eld. If G is a nite subgroup of the multi-
plicative group (F 0, .) then G is a cyclic group.
Notice that this result applies to general elds. The reader should identify
all possible G in the cases F = C and F = R. Our proof depends on a simple
result from the theory of commutative groups.
28
Lemma 119 If G is a nite Abelian group there exists an integer N and an
element h such that
(i) g
N
= e for all g G,
(ii) h has order exactly N.
Combining Lemmas 117 and 118, we see that all nite elds have a very
simple structure.
Theorem 120 If F is a nite eld then F is (isomorphic to) the splitting
eld of X
p
n
1
1 over Z
p
for some prime p and some integer n 1.
(We can refer to the splitting eld since Theorem 113 tells us that splitting
elds are unique up to isomorphism.)
Theorem 120 tells us the structure of a given nite eld, if it exists, but
does not tell us if such a eld exists. To obtain existence results we need to
investigate the polynomial X
p
n
1
1 Z
p
[X]. We use a general result on
repeated roots.
Lemma 121 Let K be a eld. Suppose that p(X) =

n
j=0
a
j
X
j
K[X]
splits over K. Then p has (X a)
2
as a factor for some a K if and only
if the formal derivative
p

(X) =
n

j=1
ja
j
X
j
and p[X] have a non-trivial common factor.
Lemma 122 Let K be a eld of characteristic p a prime. If X
p
n
1
1 splits
over K then all the linear factors are distinct.
Our results look nicer when stated in terms of X
p
n
X.
Theorem 123 If p is a prime and n an integer the splitting eld of X
p
n
X
over Z
p
contains p
n
elements consisting of the p
n
distinct roots of X
p
n
X.
We have now proved existence and uniqueness so we may make the fol-
lowing denition.
Denition 124 The nite eld of order p
n
(p prime, n 1) is called the
Galois eld of order p
n
and written GF(p
n
).
This triumph completes that part of this section which is on the syllabus.
However (strictly o the syllabus) we must admit that the triumph is not
quite as complete as it appears. Observe that Lemma 118 tells us that
the non-zero elements of GF(p
n
) form a cyclic group generated by a single
29
element x say. As temporary notation let us call x a multiplicative generator
of GF(p
n
). Surely, we can not claim to understand GF(p
n
) unless we have
some short algorithm for nding a multiplicative generator for it. So far as
I know, no such algorithm has been found.
Of course since GF(p
n
) is nite, exhaustive search will eventually turn
up such a generator. We note also that quite a large proportion of the
elements of GF(p
n
) must be multiplicative generators (can you make this
statement more precise?) so properly random trial and error
2
will rapidly
nd a multiplicative generator x with arbitrarily low probability of failure.
Let us choose a basis u
1
, u
2
, . . . , u
n
for GF(p
n
) as a vector space over Z
p
. We
then have
x
r
= a
1
(r)u
1
+a
2
(r)u
2
+ +a
n
(r)u
n
.
The n-tuple in
a
r
= (a
1
(r), a
2
(r), . . . , a
n
(r))
thus runs through each element of Z
p
0 exactly once as r runs from 0 to
p
n
1.
In a telepathy experiment, Albert and Bertha are placed in separate sealed
rooms. The experiment has already been running for a time 5N minutes
where N is unknown to them. A bell rings each 5 minutes and (supposing
it to be 5r minutes since they entered the room) they are asked to guess
an n-tuple of integers a
r+N
= (a
1
(r + N), a
2
(r + N), . . . , a
n
(r + N)) with
0 a
j
(r + N) p 1. If one of them guesses right he or she is told so and
presented with a paper star. Bertha has the advantage that she knows how a
r
is constructed and in particular knows x. It is easy to see that initially Albert
and Bertha can only guess at random but that once Bertha has guessed right
she can lock in and give the correct answer each time.
One way of trying to hide a radio signal is to spread it as a large number
of weak signals at dierent frequencies and to change the choice of frequencies
at regular intervals. Of course the enemy may make a lucky choice of listening
frequencies and catch a brief part of the signal but the change of frequencies
should stymie him. On the other hand, our own side may not be able to
keep their timekeepers suciently synchronised with the transmitter during
long periods of silence. We begin to see how military men and others might
develop a deep interest in Galois elds.
2
The ghastly modern educationalists jargon seeks to replace trial and error by trial
and improvement but here the failure of a guess results in no improvement.
30
10 Modules
The theory of vector spaces is a well developed and powerful one. We have
seen examples of its use in this course in Lemma 117 which helped us classify
nite elds and in the denition of the degree [L : K] of an extension which
helped resolve the classical ruler and compass problems. From time to time
we come across structures like the lattice Z
2
which have a vector space
avour without being vector spaces. It is thus natural to seek a theory
which generalises the notion of a vector space though though we may expect
the development of such a theory to be more intricate and the general results
to be less neat.
We proceed in the obvious way by replacing eld by ring in the deni-
tion of a vector space.
Denition 125 Let R be a ring. We say that (M, R, +, .) is a module over
R if the following is conditions hold.
(i) (M, +) is an Abelian group.
(ii) There is a map : R M M written (r, m) = rm such that
(a) r(m
1
+m
2
) = rm
1
+rm
2
,
(b) (r
1
+r
2
)m = r
1
m+r
2
m,
(c) (r
1
r
2
)m = (r
1
(r
2
m),
(d) 1m = m,
for all r, r
1
, r
2
R and m, m
1
, m
2
M.
We say that M is a module over R. Since the syllabus requires it to be
explicitly stated, we remark that a vector space over a eld F is automatically
a module over F.
We have an immediate pleasant surprise.
Lemma 126 Let (G, +) be a commutative group. If we write
na = a +a + +a
. .
n
,
(n)a = na and 0a = a [a G] then G is a module over Z.
However, this example shows us that the behaviour of modules, even over very
nice rings, is very dierent from that of vector spaces. (Precisians will worry
that not all the terms in the next example have been dened, everybody else
will welcome early warning of trouble.)
Example 127 Let C
6
be the cyclic group generated by [1] and write n[1] =
[n]. Then if we take C
6
as a module over Z, [1] is a minimal generating
set but so is [2], [3].
31
This should be contrasted with the theory of nite dimensional vector spaces
where every minimal generating set (in the language of Course P1, every
minimal spanning set) has the same number of elements. The reader may
care to reect on the importance of division in the proof of the Steinitz
replacement lemma. For the moment we note that results which involve the
notion of basis or dimension explicitly or implicitly are unlikely to carry over
from vector spaces to general modules.
Our next example is not surprising.
Lemma 128 If S is a subring of a ring R then R is a module over S with
module multiplication dened to be ring multiplication in R. In particular R
is a module over itself.
Our nal introductory example may seem a little strange but much of the
strangeness will vanish on reection.
Lemma 129 Let V be a vector space over a eld F and let be an endo-
morphism of V (that is a linear map from V to V ). Then V is a module
over the ring of polynomials F[X] with module multiplication dened by the
following rule.
If p(X) =

n
j=0
a
j
X
j
and v V then pv = p()v, that is
pv = a
0
v +a
1
(v) +a
2

2
(v) + +a
n

n
(v).
The reader should note the implied convention
0
= . She should then
examine the denition when F = C and is the linear map given, in turn,
by the matrices
_
0 0
0 0
_
,
_
1 0
0 0
_
,
_
1 0
0 2
_
,
_
2 0
0 2
_
,
_
0 1
0 0
_
.
If the situation described in Lemma 129 holds we talk of the F[X] module
constructed from V via .
It is natural to ask whether a concept so general that it includes both
Abelian groups and the eect of polynomials of a given endomorphism on
a vector space is not too general to produce interesting mathematics. The
object of the last part of this course is to produce a theorem on modules
(Theorem 171) so powerful that it gives both a complete classication of
nite Abelian groups and of endomorphisms on nite dimensional vector
spaces over C.
Before moving directly to this topic we rst produce some standard alge-
braic denitions, theorems and constructions parallelling those already pro-
duced in our studies of groups, vector spaces and rings.
32
Denition 130 If M and N are modules over a ring R we say that :
M N is a (module) homomorphism if
(r
1
m
1
+r
2
m
2
) = r
1
(m
1
) +r
2
(m
2
)
for all r
1
, r
2
R and m
1
, m
2
M. If is a bijection we say that it is a
(module) isomorphism and that M and N are isomorphic.
Denition 131 If (M, R, +, .) is a module over a ring R we say that a subset
N of M is a submodule if N is a subgroup of (M, +) and rn N whenever
r R and n N.
The process of quotienting is familiar from our work with rings in Section 2
which we shall follow almost exactly. Since N is a subgroup of (M, +) we
may work with cosets u +N of N.
Lemma 132 Let N be a submodule of a module M over a ring R. Then
(i)

uM
(u +N) = M.
(ii) If u, v M then either (u +N) (v +N) = or u +N = v +N.
Lemma 133 If N is a submodule of a module M over a ring R and
u
1
+N = u
2
+N, v
1
+I = v
2
+I
then
(u
1
+v
1
) +I = (u
2
+v
2
) +I, ru
1
+I = ru
2
+I
for all r R.
Denition 134 If N is a submodule of a module M over a ring R we write
M/N for the set of cosets of N and dene addition and multiplication on
M/N by
(u +N) + (v +N) = (u +v) +N, r(u +N) = ru +N.
Lemma 135 If N is a submodule of a module M over a ring R then M/N
with module addition and multiplication as in the previous denition is a
module over R.
We call M/N a quotient module.
We continue along the sequence of Section 2.
33
Denition 136 If M and N are modules over a ring R and : M N is
a homomorphism we write
ker =
1
(0) = r R : (r) = 0
and call ker the kernel of .
Lemma 137 If M and N are modules over a ring R and : M N is a
homomorphism then
(i) ker is a submodule of M.
(ii) (u) = v has a solution u M if and only if v im.
(iii) If (u) = v then (u

) = v if and only if u

u + ker .
Lemma 138 Let N be an submodule of a module M over a ring R. Then
the map : M M/N given by
(u) = u +N
is a homomorphism with kernel N.
Theorem 139 (The isomorphism theorem for modules) Suppose that
M and N are modules over a ring R and : R S is a homomorphism.
Then
R/ ker

= im.
We have followed the same path to obtain the same isomorphism theorem
for rings and modules. There is a similar result for groups (but the key notion
is that of a normal subgroup that is of a subgroup H of a group G such that
g
1
Hg = H for all g G). Clearly we ought to seek some master theorem
from which all these results could be derived. Such concerns are the subject
of Universal Algebra and its younger cousin Category Theory. In the context
of the present course, most readers will nd the generalisation of vector
spaces to modules suciently hard without seeking to study a concept of
algebraic system which will include objects with a single non-commutative
multiplication (groups), objects with two commutative multiplications linked
by a distributive law (rings) and products of such objects with further links
(modules).
The research supervisor of the great probabilist Feller told him that the
best mathematics consists of the general embedded in the concrete. Feller
claimed that it was some years before he realised this was not an anti-
militarist slogan. Most mathematicians would agree with Fellers supervisor.
Unfortunately they would dier widely on the proportion of general and con-
crete required and still more widely on what, precisely, is general and what
concrete.
34
11 Linear relations in modules
So far, the results we have proved on modules have had a general algebraic
avour. However, we deliberately chose the axioms for modules to echo those
for vector spaces and from now on we shall try to exploit that fact.
Lemma 140 If M is a module over a ring R and A a non empty subset of
M then the set N of elements
k

j=1
r
j
a
j
with r
j
R, a
j
A and k a positive integer is a submodule of M. If N

is
any submodule of M with N

A then N

N.
We call N the submodule generated by A.
Denition 141 If M is a module over a ring R and M generated by a single
element m we say that M is a cyclic module and write M = Rm.
If M
1
, M
2
, . . . , M
n
are submodules of module M we write M
1
+ M
2
+
+ M
n
for the submodule generated by

n
r=1
M
r
. We recall from vector
space theory that direct sums are more useful than sums.
Denition 142 If M is a module over a ring R and M
1
, M
2
, . . . , M
n
are
submodules of M we say that M
1
+ M
2
+ + M
n
is a direct sum (more
specically an internal direct sum) of M
1
, M
2
, . . . , M
n
and write
M
1
+M
2
+ +M
n
= M
1
M
2
M
n
if the only solution to the equation m
1
+ m
2
+ + m
n
= 0 with m
j
M
j
[j = 1, 2, . . . n] is m
j
= 0 [j = 1, 2, . . . n].
Lemma 143 Let M be a module over a ring R and M
1
, M
2
, . . . , M
n
sub-
modules of M. The following conditions are equivalent.
(i) M
1
+M
2
+ +M
n
is a direct sum.
(ii) (

i=j
M
i
) M
j
= 0 for each 1 j n.
(iii) Each m M
1
+ M
2
+ + M
n
can be written in only one way as
m =

n
j=1
m
j
with m
j
M
j
[j = 1, 2, . . . n].
We can also dene an external direct sum (analogous to the direct sum
of rings in Lemma 5).
35
Lemma 144 Let M
1
, M
2
, . . . , M
n
be modules over a ring R. If we dene
addition and module multiplication on

n
j=1
M
j
by
(m
1
, m
2
, . . . , m
n
) + (m

1
, m

2
, . . . , m

n
) = (m
1
+m

1
, m
2
+m

2
, . . . , m
n
+m

n
)
r(m
1
, m
2
, . . . , m
n
) = (rm
1
, rm
2
, . . . , rm
n
)
for m
j
, m

j
M
j
, r R then

n
j=1
M
j
is a module over R.
We write M
1
M
2
M
n
for the ring just dened and call it the external
direct sum. If the M
j
are all submodules of the same module M then there
is a natural isomorphism between the internal and external direct sums and
no problems arise if we identify the two objects.
We shall need the following simple result.
Lemma 145 If M
1
and M
2
are submodules of a module M over a ring R
and M
1
+M
2
is a direct sum then
(M
1
M
2
)/M
2

= M
1
.
Our programme in the nal part of the course is to show that reasonably
well behaved modules M over reasonably well behaved rings can be written
as the direct sum M
1
M
2
M
n
of submodules M
j
each of which is well
behaved (in particular cyclic, so that M
j
= Rm
j
). To place this programme
in context, note that that one of the fundamental theorems of vector space
theory can be written as follows.
Theorem 146 If V is module over a eld F generated by a nite set then
V = V
1
V
2
V
n
where each submodule V
j
is cyclic (and is isomorphic to F as a module over
F). Further, the number n is an invariant of V (that is, every such decom-
position requires exactly n submodules of the stated type).
We cannot evade consideration of one of the most striking ways that a
module like Z
7
or Z
21
over Z diers from a vector space.
Lemma 147 Let M be a module over a ring R. If m M the set
o(m) = r R : rm = 0
is an ideal of R.
36
Denition 148 (i) We call the ideal o(m) dened in Lemma 147 the order
ideal of m.
(ii) If o(m) ,= 0 we say that m is a torsion element.
(iii) If a module has no non-zero torsion elements we say that it is torsion
free.
Lemma 149 If M is a module over R and T is the set of torsion elements
in M then T is a submodule of M and M/T is a torsion free module.
We adopt a denition of linear independence which is taken directly from
vector spaces.
Denition 150 If M is a module over R we say that elements m
1
, m
2
, . . . ,
m
n
are linearly independent if the equation
n

j=1
r
j
m
j
= 0
with r
j
R [j = 1, 2, . . . , n] only has the solution r
1
= r
2
= = r
n
= 0.
The next denition parallels the idea of a basis for a vector space
Denition 151 If M is a module over R generated by linearly independent
elements m
1
, m
2
, . . . , m
n
we say that the elements form a basis for M and
that they generate M freely. We say that M is a nitely generated free
module.
(More generally M is freely generated if it has a subset X which generates
M and is such that any non-empty nite subset of X is linearly independent.
We shall not make use of this idea.)
Lemma 152 If M is a module over R and m
1
, m
2
, . . . , m
t
M the fol-
lowing four statements are equivalent.
(i) The elements m
1
, m
2
, . . . , m
t
form a basis for M.
(ii) Any element m of M can be written in one and only one way as
m =
t

j=1
r
j
m
j
with r
j
R.
(iii) The elements m
1
, m
2
, . . . , m
t
generate M and the following condi-
tion holds. If N is an R module and n
j
N then there exists a homomor-
phism : M N with (m
j
) = n
j
[1 j t].
(iv) Each m
j
is torsion free (i.e. not a torsion element) and
M = m
1
R m
2
R m
t
R.
37
Algebraists would prefer to use condition (iii) or something like it as the
denition of freely generated since it chimes in with their predeliction for
universal objects.
The following remark is more or less obvious.
Lemma 153 The module M over a ring R is freely generated by t elements
if and only if
M

= R R R R
. .
t
.
The next remark is almost as obvious but will play a key role in the proof of
our module decomposition theorem (Theorem 171).
Lemma 154 If a module M over a ring R is nitely generated then we
can nd a nitely generated free module F and an injective homomorphism
: F M. (In other words, every nitely generated module is the image of
some nitely generated free module.)
In the case of a cyclic module, Lemma 154 can be sharpened.
Lemma 155 Suppose M is a cyclic module over a ring R generated by m.
Then
M

= R/o(m).
In particular two cyclic modules over R are isomorphic if and only if their
generating elements have the same order ideal.
Thus if M is a cyclic module generated by m it is natural to call o(m) the
order ideal of M.
12 Matrices and modules
There is no problem in extending the notion of an r s matrix together with
the denitions of matrix addition, matrix multiplication and so forth from
elds to rings.
Lemma 156 Let M and N be nitely generated free modules over a ring R.
Suppose that M has basis m
1
, m
2
, . . . , m
r
and that N has basis n
1
, n
2
, . . . ,
n
s
. Then there is bijection A between homomorphisms : M N and
r s matrices A = (a
ij
) over R given by
(m
j
) =
s

i=1
a
ij
n
i
.
38
I repeat my warning that generalising results from vector spaces is the natural
way forward but that we must act as though we were walking on eggs. The
care required may not be obvious to the reader who looks only at the theorems
we do prove but will be obvious to anyone who asks about the theorems we
do not prove.
Is there any other point to which you would wish to draw my
attention? To the curious incident of the dog in the night-time.
The dog did nothing in the night-time. That was the curious
incident. remarked Sherlock Holmes.
Denition 157 We say that an s s matrix A over R is invertible if there
exists an s s matrix

A with A

A =

AA = I.
The standard uniqueness argument shows that

A, if it exists, is unique.
Lemma 158 The product of s s invertible matrices is itself invertible.
(Thus the s s invertible matrices over R form a group.)
Example 159 The matrix
_
a b
c d
_
over Z is invertible if and only if ad bc = 1.
Lemma 160 Suppose that M is a nitely generated free module over a ring
R and that M has basis m
1
, m
2
, . . . , m
s
. If A = (a
ij
) is an s s invertible
matrix A over R and
m

j
=
s

i=1
a
ij
m
i
then m

1
, m

2
, . . . , m

s
is also a basis for M.
When we dealt with matrices over elds we used elementary row and
column operations and their associated matrices. We can do the same thing
here. First let us set out the corresponding elementary s s matrices.
(i) F
ij
is the matrix obtained from the identity matrix by interchanging
row i and row j.
(ii) G
i
(u) is the matrix obtained from the identity matrix by multiplying
row i by the unit u.
(iii) H
ij
(r) is the matrix obtained from the identity matrix by adding r
times row j to row i [i ,= j, r R].
(iv)

H
ij
(r) is the matrix obtained from the identity matrix by adding r
times column j to column i [i ,= j, r R].
39
We shall not use G
i
(u) but we include it for completeness. Observe that

H
ij
(r) = H
ji
(r). Exactly as in the eld case we have the following easy
remarks.
Lemma 161 (i) The eect of pre-multiplying a matrix of the appropriate
size
(1) by F
ij
is to interchange row i and row j,
(2) by G
i
(u) is to multiply row i by the unit u,
(3) by H
ij
(r) is to add r times row j to row i.
(ii) The eect of post-multiplying a matrix of the appropriate size
(1) by F
ij
is to interchange column i and column j,
(2) by G
i
(u) is to multiply column i by the unit u,
(3) by

H
ij
(r) is to add r times column j to column i.
(iii) The matrices F
ij
, G
i
(u), H
ij
(r) and

H
ij
(r) are all invertible.
When we worked over elds we where able to reduce matrices to very
special forms by pre- and post-multiplication by invertible matrices.
Denition 162 Let A and B be s t matrices over a ring R. We say that
A and B are equivalent if we can nd an invertible s s matrix P and an
invertible t t matrix Q such that B = PAQ.
Lemma 163 (i) Equivalence of matrices is an equivalence relation.
(ii) Let M and N be nitely generated free modules over a ring R. Suppose
that M has basis m
1
, m
2
, . . . , m
s
and that N has basis n
1
, n
2
, . . . , n
t
.
Suppose that the homomorphism : M N corresponds to the matrix A
for these bases. If A is equivalent to B then we can nd bases m

1
, m

2
, . . . ,
m

s
for M and n

1
, n

2
, . . . , n

t
for N such that : M N corresponds to
the matrix B for these bases.
We can not do very much over general rings but we can do a great deal
over Euclidean domains.
Lemma 164 If A is a non-zero s t matrix over a Euclidean domain we
can nd a sequence of elementary row and column operations which reduce
A to a matrix B with b
i1
= 0 for 2 i s, b
1j
= 0 for 2 j t 1 and b
11
dividing every element b
ij
of B.
Lemma 165 If A is a s t matrix over a Euclidean domain we can nd
a sequence of elementary row and column operations which reduce A to a
matrix D with d
ij
= 0 for i ,= j (that is D is diagonal) and d
ii
[d
(i+1)(i+1)
for
all 1 i min(s, t) 1.
40
We restate Lemma 165 as a theorem.
Theorem 166 If A is a s t matrix over a Euclidean domain then A is
equivalent to a diagonal matrix D with d
ii
[d
(i+1)(i+1)
for all 1 i min(s, t)
1.
This result is ultimately due to Henry Smith who proved it for integer valued
matrices. Smith was a major pure mathematician at a time and place (19th
century Oxford) not particularly propitious for such a talent. He seems
to have been valued more as a good College and University man than for
anything else
3
.
In the next section we obtain the module decomposition theorem (The-
orem 171) as a direct consequence of Theorem 166 but for the moment we
just note a simple corollary.
Lemma 167 Let M be a nitely generated free module over a Euclidean
domain then all bases of M contain the same number of elements.
We call the number of elements in a basis of M the rank of M.
There are two important remarks to make.
(1) The results which we obtain for Euclidean domains can be extended
with a little more work to principal ideal domains. The details are given
in [5] Chapters 7 and 8. However all our applications will be to Euclidean
domains. (I remarked earlier on the diculty of nding simple examples of
principal ideal domains which are not Euclidean.)
There is a further point. Our applications will be to modules over a
domain R where R is Z and C[X]. For both of these the Euclidean function
is such that given a R and a non-zero b R there is an algorithm for nding
c, r R such that a = cb + r and (r) < (b). The proof of Theorem 166
is thus algorithmic, that is we can actually calculate P, Q invertible and D
of the correct form such that PAQ = D. We are thus not doing abstract
algebra but concrete algebra which can be (and is) programmed for electronic
computers.
(2) We shall not give general uniqueness theorems corresponding to our
general decomposition theorems. Such results will again be found in [5]
Chapters 7 and 8. They are not very hard but a 24 hour course cannot
contain everything. In the concrete examples that we give uniqueness will
be more or less obvious.
3
He even supervised on Sunday afternoon, telling his students that It was lawful on
the Sabbath day to pull an ass out of the ditch.
41
13 The module decomposition theorems
We are now within sight of our module decomposition theorems. We need
three preliminary lemmas. The rst is a simple consequence of Lemma 147.
Lemma 168 If M is a cyclic module over a principal ideal domain D then
M

= D/(d) for some d D. If D/(d)

= D/(d

) then d and d

are associates.
We say that M is of order d.
The second requires a little work.
Lemma 169 Every submodule G of a nitely generated free module F over
a principal ideal domain D is itself a nitely generated free module. The rank
of G is no greater than the rank of F.
The third is routine abstract algebra.
Lemma 170 Let M be the internal direct sum
M = M
1
M
2
M
s
of submodules M
i
. Suppose N
i
is a submodule of N
i
for each i and N =
N
1
+N
2
+ +N
s
. If is the natural homomorphism : M M/N then
M/N = (M) = (M
1
) (M
2
) (M
s
)
and (M
i
)

= M
i
/N
i
.
Theorem 166 now gives us our rst decomposition theorem.
Theorem 171 (Basic module decomposition theorem) If M is a nitely
generated module over a Euclidean domain D then M may be written as an
internal direct sum
M = M
1
M
2
M
s
where M
i
is a non-trivial cyclic submodule of order d
i
[1 i s] and d
i
[d
i+1
[1 i s 1].
Let us note the following consequences.
Lemma 172 If M is a nitely generated module over a Euclidean domain
D then M = T F where T is the torsion submodule and F is a nitely
generated free module.
Lemma 173 If M is a nitely generated torsion free module over a Eu-
clidean domain D then M is a nitely generated free module.
42
Turning from the general to the concrete we obtain a structure theorem
for nitely generated Abelian groups.
Theorem 174 If G is a nitely generated Abelian group then (as a group)
G

= Z
d
1
Z
d
2
Z
dr
Z
t
where d
i
[d
i+1
[1 i r 1].
(Note that this result can be stated entirely in group theoretic terms.)
Lemma 175 (We work with groups and group isomorphism.)
(i) If Z
t

= Z
t

then t = t

.
(ii) If
Z
d
1
Z
d
2
Z
dr

= Z
d

1
Z
d

2
Z
d

r
with d

i
[d

i+1
[1 i r

1] and d
i
[d
i+1
[1 i r 1] then r = r

and d

i
= d
i
for 1 i r.
(iii) The decomposition in Theorem 174 is unique.
Notice that we have provided an algorithm which presented with gener-
ators for an Abelian group together with relations between them can decide
if the largest group compatible with these relations is nite or innite. It
has been shown (though the proof is book length) that no such algorithm
can exist for the non-Abelian case that is there exists no computer program
which presented with generators for a group together with relations between
them can decide if the largest group compatible with these relations is nite
or innite. (This subject is known as the word problem for groups.)
Of course, the group Z
6
can be decomposed still further as Z
6
= Z
2
Z
3
.
This fact suggests that we develop our decomposition theorem, Theorem 171,
as follows. Recall that a Euclidean domain is a principal ideal domain and so
a unique factorisation domain. Our main result echos the Chinese remainder
theorem.
Lemma 176 Let M be a cyclic module of order d over a principal ideal
domain D. If d has the prime factorisation d = up

1
1
p

2
2
. . . p
s
s
with u a unit,
the p
i
non-associate primes and
i
1 then
M = M
1
M
2
M
s
where M
j
is cyclic of order p

j
j
.
Cyclic modules of order p

with p a prime are called primary modules.

43
Theorem 177 (Primary decomposition theorem) If M is a nitely gen-
erated module over a Euclidean domain D then M may be written as an
internal direct sum
M = M
1
M
2
M
s
where M
i
are primary modules or free cyclic modules.
It is worth noting that no further splitting is possible.
Denition 178 A non-trivial module over a ring R is called indecomposable
if whenever M = M
1
M
2
with M
1
, M
2
submodules then either M
1
= 0
or M
2
= 0.
Lemma 179 (i) A primary module over a principal ideal domain is inde-
composable.
(ii) A free cyclic module over an integral domain is indecomposable.
Theorem 177 immediately gives a structure theorem for nitely generated
Abelian groups.
Theorem 180 If G is a nitely generated Abelian group then (as a group)
G

= Z
p

1
1
Z
p

2
2
Z
p
r
r
Z
t
where p
i
is a prime and
i
1 [1 i r]. If we add the condition p

i
i
p

i+1
i+1
the decomposition is unique.
The following example shows that things are not so simple for non-nitely
generated Abelian groups (and so, certainly, for modules in general) as one
might at rst imagine.
Example 181 (i) Consider the Abelian group Q. Any non-trivial nitely
generated subgroup is generated by a single element and is thus isomorphic
to Z. However Q is not nitely generated.
(ii) Consider the Abelian group Q/Z. Any non-trivial nitely generated
subgroup is a nite cyclic group. However Q/Z is not nitely generated.
Since a nite Abelian group is automatically nitely generated we have
a complete classication of all nite Abelian groups.
Theorem 182 If G is a nite Abelian group then
G

= Z
p

1
1
Z
p

2
2
Z
p
r
r
where p
i
is a prime and
i
1 [1 i r]. If we add the condition p

i
i
p

i+1
i+1
the decomposition is unique.
44
The rest of this section is very much o the syllabus but gives a striking
application of Theorem 182. We write
T = C : [[ = 1
and note that T is an Abelian group under multiplication. We write D
n
for
the subgroup of T dened by
D
n
= T :
n
= 1.
(Thus D
n
is the multiplicative group of nth roots of unity.) We observe that
D
n
is group isomorphic to Z
n
.
Denition 183 If G is a nite Abelian group and : G T is a group
homomorphism we say that is a character of G.
Lemma 184 The collection

G of characters of a nite group G form an
Abelian group under the multiplication rule (
1

2
)(g) =
1
(g)
2
(g).
We call

G the dual group of G. Once we have the classication theorem for
nite Abelian groups we can use the following easy result to give a corre-
sponding classication for their dual groups.
Lemma 185 If
G = Z
p

1
1
Z
p

2
2
Z
p
r
r
with p
s
is a prime and
s
1 [1 s r] then we may identify

G with
D
p

1
1
D
p

2
2
D
p
r
r
as follows. If
= (
1
,
2
, . . . ,
r
) D
p
1
1
D
p
2
2
D
p
r
r
and
n = (n
1
, n
2
, . . . , n
r
) Z
p

1
1
Z
p

2
2
Z
p
r
r
then
(n) =
r

s=1

ns
s
.
We shall only make use of the following consequences.
Lemma 186 If G is a nite Abelian group then

G is a nite group with the
same number of elements. If g G there exists a

G with (g) ,= 0.
45
Lemma 187 Let G be a nite Abelian group with N elements. If g G but
g ,= e then

G
(g) = 0.
If g = e

G
(g) = N.
If G is a nite Abelian group let us write C(G) for the set of functions
f : G C. If f, h C(G) we write
f, h) = [G[
1

xG
f(x)h(x)

where [G[ is the number of elements of G and z

denotes the complex conju-

gate of z.
Lemma 188 If G is a nite Abelian group then C(G) equipped with the
usual pointwise addition and scalar multiplication is a vector space over C.
The operation , ) is an inner product on C(G). The characters of G form
an orthonormal basis for G.
It is thus natural to write

f() = f, h),
and call

f :

G C the Fourier transform of an f C(G). Lemma 187 gives
us the required representation theorem.
Lemma 189 If G is a nite Abelian group and f C(G) then
f =

f().
This small but perfectly formed Fourier theory for nite Abelian groups
is used in number theory and machine computation. Even more importantly
it suggests that we should look at Fourier theory in the context of groups
and this gives rise to representation theory both for nite non-Abelian groups
and for innite groups satisfying reasonable continuity conditions.
46
14 Applications to endomorphisms
Throughout this section V will be a nite dimensional vector space over a
eld F and an endomorphism of V . We recall from Lemma 129 that V is a
module over the ring of polynomials F[X] with module multiplication dened
by pv = p()v. We observe that F[X] is a Euclidean domain. Further if V has
basis e
1
, e
2
, . . . e
n
as vector space then e
1
, e
2
, . . . e
n
generate V as a module
(though, of course they may not be linearly independent when V is considered
as a module). We note that every v V is a torsion element. We can thus
apply our module decomposition theorem (Theorem 171). Translated into
the language of vector spaces it takes the following form.
Lemma 190 Let V be a nite dimensional vector space over a eld F and
an endomorphism of V . Then V may be expressed a the direct sum of
subspaces
V = V
1
V
2
V
s
where each V
i
is associated with a monic polynomial P
i
F[X] of degree n
i
and a vector v
i
as follows.
(i)

Vectors of the form

k
v
i
span V
i
.
(ii) We have P
i
()(v) = 0 for all v V
i
.
(iii) If P F[X] and P()(v) = 0 for all v V
i
then P
i
[P.
(iv) P
i
[P
i+1
for all 1 i s 1.
We can immediately improve the form of this result.
Theorem 191 As for Lemma 190 but with (i)

replaced by
(i) v
i
, (v
i
),
2
(v
i
) . . . ,
n
i
1
(v
i
) is a basis for V
i
.
It is worth noting that the subspace V
i
are invariant in the sense that (V
i
)
V
i
.
Forgetting about modules for the moment and working entirely in stan-
dard vector space theory we can translate Theorem 191 into statement about
matrices by choosing the obvious basis for V .
Theorem 192 (Rational canonical form) Let V be a nite dimensional
vector space over a eld F and an endomorphism of V . Then there is basis
for V such that has matrix A (relative to this basis) which consists of zeros
except for s blocks consisting of n
i
n
i
square matrices A
i
[1 i s] down
47
the diagonal satisfying the following conditions. Each
A(i) =
_
_
_
_
_
_
_
_
_
0 0 0 0 0 0 a
0
(i)
1 0 0 0 0 0 a
1
(i)
0 1 0 0 0 0 a
2
(i)
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0 0 1 0 a
n
i
2
(i)
0 0 0 0 0 1 a
n
i
1
(i)
_
_
_
_
_
_
_
_
_
and, if we write
P
i
(X) = X
n
i
+
n
i
1

k=0
a
k
X
k
,
we have P
i
[P
i+1
for all 1 i s 1.
Denition 193 An n n matrix of the form
A =
_
_
_
_
_
_
_
_
_
0 0 0 0 0 0 a
0
1 0 0 0 0 0 a
1
0 1 0 0 0 0 a
2
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0 0 1 0 a
n2
0 0 0 0 0 1 a
n1
_
_
_
_
_
_
_
_
_
is called the companion matrix of the monic polynomial
p(X) = X
n
+
n1

k=0
a
k
X
k
.
Lemma 194 If A is the companion matrix of a monic polynomial P then
P(X) = det(XI A).
(N.B. we have here a relation between coecients with X an indeterminate.)
In other words P is the characteristic polynomial of A.
We can now grasp some of the implications of our results on endomor-
phisms.
Theorem 195 Let V be a nite dimensional vector space over a eld F and
an endomorphism of V . Let p
i
[1 i s] be the polynomials which appear
in Theorems 191 and 192.
(i)

s
i=1
p
i
is the characteristic polynomial of A (and so of ).
(ii) The polynomial p
s
is the minimal polynomial (more exactly the min-
imal annihilating polynomial) of (and so of A). In other words p
s
() = 0
and if q F[X] satises q() = 0 then p
s
[q.
48
Incidentally we have proved the Cayley Hamilton theorem for general elds.
(The proof via triangular matrices in Course P1 only works for C though
we can deduce the result for real matrices by considering them as complex
matrices.)
Theorem 196 (Cayley Hamilton) If V is a nite dimensional vector space
over a eld F and an endomorphism of V then satises its own charac-
teristic equation.
We can also prove that the rational canonical decomposition is indeed canon-
ical.
Lemma 197 The matrix in Theorem 192 is uniquely determined by the given
conditions.
What about the Primary Decomposition Theorem (Theorem 177)? Work-
ing along the same lines as Theorem 191, we obtain the following result.
Lemma 198 Let V be a nite dimensional vector space over a eld F and
an endomorphism of V . Then V may be expressed a the direct sum of
subspaces
V = V
1
V
2
V
s
where each V
i
is associated with a monic polynomial P
i
F[X] of degree n
i
and a vector v
i
as follows.
(i) v
i
, (v
i
),
2
(v
i
) . . . ,
n
i
1
(v
i
) is a basis for V
i
.
(ii) We have P
i
()(v) = 0 for all v V
i
.
(iii) If P F[X] and P()(v) = 0 for all v V
i
then P
i
[P.
(iv) P
i
is a power of an irreducible polynomial (that is P
i
= Q
m
i
i
where
Q
i
is irreducible and m
i
1).
Even such a simple eld as R has both linear and quadratic irreducible
polynomials and even rather weighty tomes on algebra do not seek any use
for Lemma 198 in this case. However if the eld is complete, that is every
polynomial has a root, then the only irreducible polynomials are linear and
we get a relatively simple result.
Theorem 199 Let V be a nite dimensional vector space over a complete
eld F and an endomorphism of V . Then V may be expressed a the direct
sum of subspaces
V = V
1
V
2
V
s
where each V
i
is associated with a polynomial (X
i
)
n
i
F[X] and a vector
w
i
as follows.
49
(i) w
i
, (
i
)w
i
, (
i
)
2
w
i
, . . . , (
i
)
n
i
1
w
i
form a basis for V
i
.
(ii) We have (
i
)
n
i
(v) = 0 for all v V
i
.
(iii) If P F[X] and P()(v) = 0 for all v V
i
then (X
i
)
n
i
[P.
As with Theorem 191 the obvious choice of basis for V gives us a theorem
about matrices. We write J(, n) for the n n matrix with s down the
diagonal, 1s immediately below and zero every where else so that
J(, n) =
_
_
_
_
_
_
_
_
_
0 0 0 0 0 0
1 0 0 0 0 0
0 1 0 0 0 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
0 0 0 0 1 0
0 0 0 0 0 1
_
_
_
_
_
_
_
_
_
.
We call J(, n) a Jordan matrix.
Theorem 200 (Jordan normal form) Let V be a nite dimensional vec-
tor space over a complete eld F and an endomorphism of V . Then there
is basis for V such that has matrix A (relative to this basis) which consists
of zeros except for Jordan matrices J(
i
, n
i
) [1 i s] down the diagonal.
Standard vector space techniques complete the result.
Lemma 201 The matrix associated with in Theorem 200 is unique up to
reordering the diagonal blocks.
Two and a half years ago in Course C1 we noted that the matrix
_
0 1
0 0
_
showed that not all square matrices are diagonalisable even over C. By ad
hoc techniques we showed that any 2 2 matrix was conjugate to a matrix
of the form
_
1
0
_
or
_
0
0
_
.
We noted that these forms were particularly useful in the study of dierential
equations.
We complete the course by giving the full solution of the problem of
classifying square matrices under conjugation over any complete eld and, in
particular, over C.
50
Theorem 202 (Jordan normal form for matrices) If A is a nn ma-
trix over a complete eld then we can nd an invertible n n matrix P
such that J
A
= PAP
1
consists of zeros except for Jordan matrices J(
i
, n
i
)
[1 i s] down the diagonal. The matrix J
A
so associated with A is unique
up to reordering the diagonal blocks.
15 Reading and further reading
Not all theorems in mathematics are hard to prove though some are. I would
hope that the reader will be able to prove many of the results in the notes
as exercises. Where she cannot, the results on rings, integral domains and
factorisation (sections 1 to 5) will be found in the standard algebra texts in
her College library and in the book of Hartley and Hawkes Rings, Modules
and Linear Algebra [5]. I have tried (but not very hard and with only partial
success) to follow the notation of Hartley and Hawkes. Whichever text she
follows she should note that our decision to use ring to mean commutative
ring with 1 is not standard.
The material in sections 6 to 9 belong to Galois theory. Garlings A
Course in Galois Theory [4] is, not surprisingly, very much in tune with the
approach adopted in Cambridge but, again, most of the standard algebra
texts cover the material. The book of Hartley and Hawkes covers the re-
mainder of the course on modules and their decomposition theorems. (Since
we aim to get to the decomposition theorems as fast as possible and we do not
deal with uniqueness, Hartley and Hawkes contains somewhat more material.
Since most British algebraists under the age of 50 learnt their module theory
from Hartley and Hawkes the close relation between book and syllabus is no
accident.)
Turning specically to some of the more general algebra texts we note that
Volume 1 of Cohns Algebra [2] covers most of the material including modules
is a typically ecient manner. Those who like to proceed from the general
to the particular will nd their tastes catered for in MacLane and Birkhos
Algebra [7]. Those who prefer the other direction will also prefer Birkho
and MacLanes Introduction to Modern Algebra [1] but this covers much less
of the course. The syllabus also commends Fraleighs A First Course in
Modern Algebra [3] but, I must confess that like many American textbooks
it reminds me of the vegetables in an American supermarket, whose splendid
appearance does not compensate for their bland taste. In any case the reader
will do better to browse through several general texts rather than concentrate
on one.
The reader who wants to learn more about the topics treated in the course
51
is in an unusually fortunate position. Most mathematicians treat textbook
writing in the same way that lawyers treat drafting legal documents and
believe that, once they have covered every possible contingency in the most
precise manner possible, their job is done. However, Ian Stewarts (yes, the
man you saw on TV) rst book Galois Theory [8] is a brilliantly written text
on a fascinating subject and a pleasure to read. He joined David Tall to
write Algebraic Number Theory [9] which gives the concrete number theory
which partners our abstract treatment of factorisation. Klines Mathematical
Thought from Ancient to Modern Times provides a picture of mathematical
progress from antiquity and thus a context for this course, and indeed the
whole Tripos.
References
[1] G. Birkho and S. MacLane A Survey of Modern Algebra (3rd Ed)
Macmillan, New York, 1965.
[2] P. M. Cohn Algebra (2nd Ed) Vol 1, Wiley, 1982.
[3] J. B. Fraleigh A First Course in Modern Algebra (5th Ed) Addison-
Wesley, 1989.
[4] D. J. H. Garling A Course in Galois Theory CUP, 1986.
[5] B. Hartley and T. O. Hawkes Rings, Modules and Linear Algebra Chap-
man and Hall, 1970.
[6] M. Kline Mathematical Thought from Ancient to Modern Times OUP,
1972.
[7] S. MacLane and G. Birkho Algebra (2nd Ed) Macmillan, New York,
1979.
[8] I. Stewart Galois Theory Chapman and Hall, 1973.
[9] I. Stewart and D. Tall Algebraic Number Theory Chapman and Hall,
1979.
52