Algebra II: Rings and modules.

Lecture Notes
(based on notes by Kevin McGerty)

Hilary Term, 2024.

Contents
1 Rings: Definition and examples. 3
1.1 Polynomial Rings. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Basic properties. 7
2.1 Integral Domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.2 The field of fractions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

3 Ideals and Quotients. 9

3.1 The quotient construction. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3.2 Images and preimages of ideals. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

4 Prime and maximal ideals, Euclidean domains and PIDs. 17

5 An introduction to fields. 20

6 Unique factorisation. 23
6.1 Irreducible polynomials. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29

7 Modules: Definition and examples. 30

7.1 Submodules, generation and linear independence. . . . . . . . . . . . . . . . . . 32

8 Quotient modules and the isomorphism theorems. 33

9 Free, torsion and torsion-free modules. 35

9.1 Homorphisms between free modules. . . . . . . . . . . . . . . . . . . . . . . . . 38

10 Canonical forms for matrices over a Euclidean Domain. 40

11 Presentations and the canonical form for modules. 43

12 Application to rational and Jordan canonical forms. 47

12.1 Remark on computing rational canonical form. . . . . . . . . . . . . . . . . . . . 49

13 Appendix A: Polynomial rings and convolution. 50

13.1 Convolution algebras. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51

14 Appendix B: Unique Factorization for Z. 53

14.1 Highest common factors. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
14.2 Characterising prime numbers. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
14.3 Unique factorization. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

1
15 Appendix C: A PID which is not a ED 57
15.1 R is not a Euclidean Domain . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
15.2 R is a PID . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57

2
1 Rings: Definition and examples.
The central characters of this course are algebraic objects known as rings. Informally speak-
ing, a ring is any mathematical object which has a notion of addition and multiplication (the
precise definition will be given shortly). As such it is a very general notion. The most basic
example is Z, the set of integers, and in this course we will largely focus on a class of rings
(known as principal ideal domains or PIDs) which are in some sense very similar to Z. By
seeing how many properties of the integers naturally extend to PIDs we will not only gain
a better understanding of topics like factorization, but also give us new tools with which to
study other topics such as linear algebra and number theory. For example, if k is a field then
k[t] the ring of polynomials with coefficients in k, is a PID, and the theory we develop for any
PID applied to k[t] yields a canonical form for matrices over an arbitrary field1 . As an example
of the applications to number theory, factorization in PIDs will allow us to determine which
primes p ∈ N can be written as the sum of two (integer) squares, i.e. for which primes p the
equation x2 + y2 = p has a solution (x, y) ∈ Z2 .

Definition 1.1. A ring is a datum (R, +, ×, 0, 1) where R is a set, 1, 0 ∈ R and +, × are binary
operations on R such that

1. R is an abelian group under + with identity element 0.

2. The binary operation × is associative and 1 × x = x × 1 = x for all x ∈ R.2

3. Multiplication distributes over addition:

x × (y + z) = (x × y) + (x × z),
(x + y) × z = (x × z) + (y × z), ∀x, y, z ∈ R.

Just as for multiplication of real numbers or integers, we will tend to suppress the symbol
for the operation ×, and write “.” or omit any notation at all. If the operation × is commutative
(i.e. if x.y = y.x for all x, y ∈ R) then we say R is a commutative ring3 . Sometimes4 people
consider rings which do not have a multiplicative identity. It is also worth noting that some
texts require an additional axiom asserting that 1 , 0. In fact it’s easy to see from the other
axioms that if 1 = 0 then the ring has only one element. We will refer to this ring as the “zero
ring”. While it is a somewhat degenerate object, it seems unnecessary to me to exclude it.

Example 1.2. i) The integers Z form the fundamental example of a ring. As mentioned
before, in some sense much of the course will be about finding an interesting class of
rings which behave a lot like Z. Modular arithmetic gives another example: if n ∈ Z then
Z/nZ, the integers modulo n, form a ring with the usual addition and multiplication.

ii) The subset Z[i] = {a + ib ∈ C : a, b ∈ Z} is easily checked to be a ring under the normal
operations of addition and multiplication of complex numbers. It is known as the Gaus-
sian integers. We shall see later that it shares many of the properties with the ring Z of
ordinary integers.

iii) Any field, e.g. Q, R, C, is a ring – the only difference between the axioms for a field and for
a ring is that in the case of a ring we do not require the existence of multiplicative inverses
(and that, for fields one insists that 1 , 0, so that the smallest field has two elements).
1
The Jordan form you learned last term only applies to fields like C which are algebraically closed.
2
That is, R is a monoid under × with identity element 1 if you like collecting terminology.
3
We will try and use the letter R as our default symbol for a ring, in some books the default letter is A.
4
In Algebra 1 last term, the definition of a ring did not demand a multiplicative identity, nevertheless in this
course we will require it. For more on this see www-math.mit.edu/∼ poonen/papers/ring.pdf.

3
 iv) If k is a field, and n ∈ N, then the set Mn (k) of n × n matrices with entries in k is a ring,
with the usual addition and multiplication of matrices.

v) Saying the previous example in a slightly more abstract way, if V is a vector space over
a field k then End(V) the space of linear maps from V to V, is a ring. In this case the
multiplication is given by composition of linear maps, and hence is not commutative.
We will mostly focus on commutative rings in this course.

vi) Example iv) also lets us construct new rings from old, in that there is no need to start
with a field k. Given any ring R, the set Mn (R) of n × n matrices with entries in R is again
a ring.

vii) Polynomials in any number of indeterminates form a ring: if we have n variables t1 , t2 , . . . , tn

and k is a field then we write k[t1 , . . . , tn ] for the ring of polynomials in the variables
t1 , . . . , tn with coefficients in k.

viii) Just as in v), there is no reason the coefficients of our polynomials have to be a field –
if R is a ring, we can build a new ring R[t] of polynomials in t with coefficients in R in
the obvious way. What is important to note in both this and the previous example is
that polynomials are no longer functions: given a polynomial f ∈ R[t] we may evaluate
it at an r ∈ R and thus associate it to a function from R to R, but this function may not
determine f . For example if R = Z/2Z then clearly there are only finitely many functions
from R to itself, but R[t] still contains infinitely many polynomials. We will construct R[t]
rigorously shortly.

ix) If we have two rings R and S , then we can form the direct sum of the rings R ⊕ S : this
is the ring whose elements are pairs (r, s) where r ∈ R and s ∈ S with addition and
multiplication given componentwise.

x) Another way to construct new rings from old is to consider, for a ring R, functions on
some set X taking values in R. The set of all such functions RX = { f : X → R inherits a ring
structure from R by defining addition and multiplication pointwise, i.e. ( f + g)(x) = f (x) +
g(x), ( f.g)(x) = f (x).g(x) for all x ∈ X (exactly as we do for R and C-valued functions). The
simplest example of this is when X = {1, 2, . . . , n} when you get5 Rn = {(a1 , . . . , an ) : ai ∈ R},
where we add and multiply coordinatewise.

xi) To make the previous example more concrete, the set of all functions f : R → R is a
ring. Moreover, the set of all continuous (or differentiable, infinitely differentiable,...)
functions also forms a ring by standard algebra of limits results.

Definition 1.3. If R is a ring, a subset S ⊆ R is said to be a subring if it inherits the structure

of a ring from R, thus we must have 0, 1 ∈ S and moreover S is closed under the addition and
multiplication operations in R. It is then straight-forward to check that (S , +, ×, 0, 1) satisfies
the axioms for a ring.

For example, the integers Z are a subring of Q, the ring of differentiable functions from R
to itself is a subring of the ring of all functions from R to itself. The ring of Gaussian integers
is a subring of C, as are Q,R (the latter two being fields of course). Recall that for a group
G containing a subset H, the subgroup criterion says that H is a subgroup if and only if it is
nonempty and whenever h1 , h2 ∈ H we have h1 h−1 2 ∈ H (here I’m writing the group operation
on G multiplicatively). We can use this to give a similar criterion for a subset of a ring to be a
subring.
5
Recall, for example, that sequences of real numbers are defined to be functions a : N → R, we just tend to write
an for the value of a at n (and refer to it as the n-th term) rather than a(n).

4
Lemma 1.4 (Subring criterion). Let R be a ring and S a subset of R, then S is a subring if and only
if 1 ∈ S and for all s1 , s2 ∈ S we have s1 s2 , s1 − s2 ∈ S .

Proof. The condition that s1 − s2 ∈ S for all s1 , s2 ∈ S implies that S is an additive subgroup by
the subgroup test (note that as 1 ∈ S we know that S is nonempty). The other conditions for
a subring hold directly. □

When studying any kind of algebraic object6 it is natural to consider maps between those
kind of objects which respect their structure. For example, for vector spaces the natural class
of maps are linear maps, and for groups the natural class are the group homomorphisms. The
natural class of maps to consider for rings are defined similarly:

Definition 1.5. A map f : R → S between rings R and S is said to be a (ring) homomorphism if

1. f (1R ) = 1S ,

2. f (r1 + r2 ) = f (r1 ) + f (r2 ),

3. f (r1 .r2 ) = f (r1 ). f (r2 ),

where, strictly speaking, we might have written +R and +S for the addition operation in the
two different rings R and S , and similarly for the multiplication operation. Partly because
the meaning is clear from context and partly because otherwise the notation becomes hard to
read, we will (as is conventional) use the same notation for the addition and multiplication in
all rings. Note that it follows from (2) that f (0) = 0.

Exercise 1.6. Check which of our examples of rings above are subrings of another example,
e.g. R and Z[i] are both subrings of C.

If f : R → S is a ring homomorphism, it is easy to see that its image

im( f ) = f (R) = {s ∈ S : ∃r ∈ R, f (r) = s}

is a subring of S . If it is all of S we say that f is surjective. We say that f : R → S is an

isomorphism if there is a homomorphism g : S → R such that f ◦ g = idS and g ◦ f = idR . It is
easy to check that f is an isomorphism if and only if it is a bijection (that is, to check that the
set-theoretic inverse of f is automatically a ring homomorphism – you probably did a similar
check for linear maps between vector spaces before.)

Example 1.7. i) For each positive integer n, there is a natural map from Z to Z/nZ which
just takes an integer to its equivalence class modulo n. The standard calculations which
show that modular arithmetic is well-defined exactly show that this map is a ring homo-
morphism.

ii) Let V be a k-vector space and let α ∈ Endk (V). Then ϕ : k[t] → Endk (V) given by ϕ( ni=0 ai ti ) =
P

i=0 ai α is a ring homomorphism. Ring homomorphisms of this type will reveal the
Pn i

connnection between the study of the ring k[t] and linear algebra. (In a sense you saw
this last term when defining things like the minimal polynomial of a linear map, but we
will explore this more fully in this course.)

iii) The inclusion map i : S → R of a subring S into a ring R is a ring homomorphism.

6
Or more generally any mathematical structure: if you’re taking Topology this term then continuous maps
are the natural maps to consider between topological spaces, similarly in Integration you consider measurable
functions: loosely speaking, you want to consider maps which are compatible with the structures your objects
have, be that a topology, a vector space structure, a ring structure or a measure.

5
 !
a −b
iv) Let A = { : a, b ∈ R}. It is easy to check this A is a subring of Mat2 (R). The map
b a
!
a −b
ϕ : C → A given by a + ib 7→ is a ring isomorphism. (This homomorphism arises
b a
by sending a complex number z to the map of the plane to itself given by multiplication
by z.)
The first of the above examples has an important generalisation which shows that any ring
R in fact has a smallest subring: For n ∈ Z≥0 set nR = 1 + 1 + . . . + 1 (that is, 1, added to itself
n times), and for n a negative integer nR = −(−n)R . The problem sheet asks you to check that
{nR : n ∈ Z} is a subring of R, and indeed that the map n 7→ nR gives a ring homomorphism from
ϕ : Z → R. Since a ring homomorphism is in particular a homomorphism of the underlying
abelian groups under addition, using the first isomorphism theorem for abelian groups we see
that {nR : n ∈ Z}, as an abelian group, is isomorphic to Z/dZ for some d ∈ Z≥0 . Since any subring
S of R contains 1, and hence, since it is closed under addition, nR for all n ∈ Z, we see that S
contains the image of ϕ, so that the image is indeed the smallest subring of R.
Definition 1.8. The integer d defined above is called the characteristic of the ring R.

1.1 Polynomial Rings.

The remark above that in general polynomials with coefficients in a ring cannot always be
viewed as functions might have left you wondering what such a polynomial actually is. In
other words, what do we mean when we say k[t] is a ring where “t is an indeterminate.”
To make sense of this, suppose that R is a ring, and consider the set7 RN = { f : N → R}. We
already saw this set has the structure of a ring using “pointwise” addition and multiplication,
but it also has a more interesting multiplication operation, ( f, g) 7→ f ∗ g where
X n
X
( f ∗ g)(n) = f (k).g(l) = f (k)g(n − k), n ∈ N.
k+l=n k=0
Since k, l ∈ N, the above sum is finite, so we get a well defined function f ∗g (c.f. convolution
of real-valued functions.). It follows directly from the definitions that (RN , +, ∗, 1, 0) is a ring,
where 0 is the constant function taking value 0, while 1(n) = 1 if n = 0 and is zero otherwise.
We will write R[[t]] for this ring.
Let R[t] = { f ∈ RN : ∃N > 0, f (n) = 0 ∀n > N}. R[t] is a subring of R[[t]] (check this) and it
is the ring of polynomials with coefficients in R which we wanted to construct. To see why, let
t ∈ RN be the function such that t(1) = 1 and t(n) = 0 for n , 1. By induction it is easy to check
that tk = t ∗ . . . ∗ t (k times) is such that tk (n) = 1 if k = n and is zero otherwise. Now suppose
that f ∈ R[t] and say f (n) = 0 for all n > N. Then we see that f = n=0
PN
f (n)tn . In fact in general
for f ∈ R[[t]] we have8 f = n≥0 f (n)tn .
P

Remark 1.9. Our definition of ∗ makes sense on a bigger set than RN : If we take the set S =
{ f : Z → R : ∃N ∈ Z, f (n) = 0 ∀n < N} then you can check that if f, g ∈ S the function
X
( f ∗ g)(n) = f (k)g(n − k)
k∈Z

is well-defined (in that only finitely many terms on the right are non-zero for any given integer
n. This ring is denoted R((t)), and it turns out that if R is a field, so is R((t)).
7
Here N = Z≥0 the set of non-negative integers. In some places (though hopefully not other parts of these notes)
N denotes the strictly positive integers.
8
At first sight the right hand side of this expression looks like it might not make sense because it is an infinite
sum. However it does give a well-defined function on N because on any element of N only finitely many terms (in
fact exactly one) in the infinite sum are nonzero.

6
 Note also that we have a ring homomorphism ιR : R → R[t] given by ιR (a) = a.1 which is
injective, thus we can view R as a subring of R[t].
The fundamental property of a polynomial ring R[t] is that ring homomorphisms ϕ : R[t] →
S are given simply by "evaluating t”. That is, to specify a homomorphism from a polynomial
ring R[t] to a ring S you only need to say what happens to the elements of R (the coefficients)
and what happens to t. We formalise this in the following Lemma.
Lemma 1.10. (Evaluation homomorphisms.) Let R, S be rings. A homomorphism Φ : R[t] → S is
determined by the pair (ϕ, s) where ϕ = Φ|R and s = Φ(t). Conversely, any such pair determines a
unique ring homomorphism and hence the set of all ring homomorphisms {Φ : R[t] → S } is in bijective
correspondence with the set of all pairs {(ϕ, s) : ϕ : R → S , s ∈ S } consisting of a ring homomorphism ϕ
and an element of S .
Proof. Any element of R[t] has the form ni=0 ai ti , (ai ∈ R), hence if Θ is any homomorphism
P
satisfying Θ ◦ i = ϕ and Θ(t) = s we see that
n
X n
X n
X n
X
Θ( ai t i ) = Θ(ai ti ) = Θ(ai )Θ(ti ) = ϕ(ai )si ,
i=0 i=0 i=0 i=0

Hence Θ is uniquely determined. To check that, for any pair (ϕ, s) there is a corresponding
homomorphism Φ : R[t] → S , we just have to check that the function Φ( ni=0 ai ti ) = ni=0 ϕ(ai )si
P P
is indeed a homomorphism, but this is straight-forward from the definitions. □

From now on, unless we explicitly state otherwise, all rings will be assumed to be commutative.

2 Basic properties.
2.1 Integral Domains
Now that we have seen some examples of rings, we will discuss some basic properties of rings
and their elements. Note that it is a routine exercise9 in axiom grubbing to check that, for any
ring R, we have a.0 = 0 for all a ∈ R. The next definition records the class of rings for which
this is the only case in which the product of two elements is zero.
Definition 2.1. If R is a ring, then an element a ∈ R\{0} is said to be a zero-divisor if there is some
b ∈ R\{0} such that a.b = 0. A ring which is not the zero ring and has no zero-divisors is called
an integral domain. Thus if a ring is an integral domain and a.b = 0 then one of a or b is equal
to zero.
Another way to express the fact that a ring is an integral domain is observe that it is exactly
the condition which permits cancellation10 , that is, if x.y = x.z then in an integral domain
you can conclude that either y = z or x = 0. This follows immediately from the definition of
an integral domain and the fact that x.y = x.z ⇐⇒ x.(y − z) = 0, which follows from the
distributive axiom.
Example 2.2. If R is a ring, then R2 is again a ring, and (a, 0).(0, b) = (0, 0) so that (a, 0) and (0, b)
are zero-divisors. The (noncommutative) ring of n × n matrices Mn (k) for a field k also has lots
of zero divisors (even though a field k has none). The integers modulo n have zero-divisors
whenever n is not prime.
9
It’s a good idea to try and check that the axioms for a ring do indeed imply that you can perform the standard
algebraic manipulations you are used to, so things like 0.x = 0 hold in any ring. None of the checks you have to do
are very exciting, so it’s best to pick a few such statements. One operation you have to be careful about however, is
cancellation (but then again you already should be aware of this issue from matrix algebra).
10
Except for the assertion the ring is not the zero ring, the zero ring having cancellation vacuously.

7
 On the other hand, it is easy to see that a field has no zero-divisors. The integers Z are an
integral domain (and not a field). Slightly more interestingly, if R is an integral domain, then
R[t] is again an integral domain. Moreover, the same is true of R[[t]].

Recall the characteristic of a ring defined in the last lecture.

Lemma 2.3. Suppose that R is an integral domain. Then any subring S of R is also an integral domain.
Moreover, char(R), the characteristic of R, is either zero or a prime p ∈ Z.

Proof. It is clear from the definition that a subring of an integral domain must again be an
integral domain. Now from the definition of the characteristic of a ring, if char(R) = n > 0 then
Z/nZ is a subring of R. Clearly if n = a.b where a, b ∈ Z are both greater than 1, then aR .bR = 0
in R with neither aR nor bR zero, thus both are zero divisors. It follows that if R is an integral
domain then char(R) is zero or a prime. □

Remark 2.4. Note that in particular, the characteristic of a field is always zero or a prime.

Recall that in a ring we do not require that nonzero elements have a multiplicative inverse11 .
Nevertheless, because the multiplication operation is associative and there is a multiplicative
identity, the elements which happen to have multiplicative inverses form a group:

Definition 2.5. Let R be a ring. The subset

R× = {r ∈ R : ∃s ∈ R, r.s = 1},

is called the group of units in R – it is a group under the multiplication operation × with identity
element 1.

Example 2.6. The units in Z form the group {±1}. On the other hand, if k is a field, then the
units k× = k\{0}. If R = Mn (k) then the group of units is GLn (k). Note that these examples all
show the relationship between R and R× can vary significantly from one ring to another: when
R = Z, the units form a finite group, if R = k then R⋆ = k\{0} contains all but one elements of k.

Remark 2.7. In our example of Z/nZ notice that this ring either has zero-divisors (when n is
composite) or is a field (when n is prime). In fact this is dichotomy holds more generally: a
finite integral domain is always a field. (See the problem sheet for more details.)

2.2 The field of fractions.

We first describe the construction of the rational numbers from the integers: A rational number
is, of course, a ratio of two integers. To say formally what this means we start with the set
Q(Z) = {(a, b) ∈ Z2 : b , 0}. Then we define a relation ∼ on Q(Z) by setting (a, b) ∼ (c, d) if
ad = bc. (To see where this comes from, notice that it expresses the fact that a/b = c/d without
using division). It is clear that ∼ is reflexive and symmetric, and an easy calculation shows
that it is transitive, so that it is an equivalence relation: Indeed suppose that (a, b) ∼ (c, d) and
(c, d) ∼ (e, f ). Then we have ad = bc and c f = de and need to check that (a, b) ∼ (e, f ), that is,
a f = be. But since d , 0 we have

a f − be = 0 ⇐⇒ d.(a f − be) = 0 ⇐⇒ (ad). f − b.(de) = 0 ⇐⇒ (bc). f − b.(c f ) = 0,

as required
As noted above, the axioms for a ring imply that 0.x = 0 for all x ∈ R, thus the additive identity cannot have a
11

multiplicative inverse, hence the most we can ask for is that every element of R\{0} does – this is exactly what you
demand in the axioms for a field.

8
 The set of equivalence classes12 is denoted Q, and we write a
b for the equivalence class
containing (a, b). You can then check that the formulas
a c ad + bc a c ac
+ = , . = ,
b d bd b d bd
give a well-defined addition and multiplication on the set of equivalence classes, forming the
field Q. The fact that it is a field and not just a ring follows because ba = 0 exactly when a = 0,
and thus if ab , 0 it has inverse ba . The details of verifying the operations are independent of the
representatives you chose in the equivalence classes take some time to write down rigorously,
but there are no surprises in the process.
The interesting thing to notice here is that this construction also makes sense for an arbitrary
integral domain: given an integral domain R, the relation on Q(R) = {(a, b) ∈ R2 : b , 0} given
by (a, b) ∼R (c, d) if ad = bc is an equivalence relation and the same formulas give the set of
equivalence classes F(R) the structure of a field. At various points you need to use cancellation
(for example in showing the relation ∼ is transitive) which is why the construction only works
for integral domains, and not more general rings13 .
Definition 2.8. The field F(R) is known as the field of fractions of R. The ring R embeds into F(R)
via the map r 7→ 1r , thus an integral domain is naturally a subring of its field of fractions.
The rational numbers are the smallest field which contain the integers, in the sense that
any field which contains Z automatically contains the rationals (essentially because if you are
a field and contain m, n ∈ Z then you contain n1 and so mn ). This is in fact the characterising
property of the field of fractions, which can be formalised as follows:
Proposition 2.9. Let k be a field and let θ : R → k be an embedding (that is, an injective homomor-
phism). Then there is a unique injective homomorphism θ̃ : F(R) → k extending θ (in the sense that
θ̃|R = θ where we view R as a subring of F(R) via the above embedding).
Proof. (non-examinable): Suppose that f : F(R) → k was such a homomorphism. Then by
assumption f ( 1a ) = θ(a), and since homomorphism of rings respect multiplicative inverses this
forces f ( a1 ) = θ(a)−1 . But then, again because f is supposed to be a homomorphism, we must
have f ( ba ) = f ( a1 . 1b ) = f ( 1a ). f ( b1 ) = θ(a).θ(b)−1 . Thus if f exists, it has to be given by this formula.
The rest of the proof consists of checking that this recipe indeed works: Given (a, b) ∈
R × R\{0} first define Θ(a, b) = θ(a).θ(b)−1 . Then it is easy to check that Θ is constant on the
equivalence classes of ∼ the relation defining F(R), so that it induces a map θ̃ : F(R) → k. Finally
it is straight-forward to see that this map is a homomorphism extending θ as required. □

Remark 2.10. Notice that this Proposition implies that any field k of characteristic zero contains
exactly one copy of the rationals. Indeed any isomorphism ϕ : Q → F between Q and a subfield
F of k yields a ring homomorphism from Z to k (by restriction). Since there is a unique ring
homomorphism from c : Z → k, and by Proposition 2.9 c extends uniquely to a homomorphism
c : Q → k, we must have c = ϕ and hence F = c(Q).

3 Ideals and Quotients.

From now on we will assume all our rings are commutative. In this section we study the
basic properties of ring homomorphisms, and establish an analogue of the ”first isomorphism
theorem” which you have seen already for groups. Just as for homomorphisms of groups,
homomorphisms of rings have kernels and images.
12
Notice that we have thus defined a ratio to be the set of all pairs of integers (a, b) (with b non-zero) which are in
that ratio.
13
There is a more sophisticated construction which works for more general rings however, but we leave that for
later courses.

9
Definition 3.1. Let f : R → S be a ring homomorphism. The kernel of f is

ker( f ) = {r ∈ R : f (r) = 0},

and the image of f is

im( f ) = {s ∈ S : ∃r ∈ R, f (r) = s}.
Just as for groups, the image of a homomorphism is a subring of the target ring. For ker-
nels the situation is a little different. In the case of groups, kernels of homomorphisms are
subgroups, but not any subgroup is a kernel – the kernels are characterised intrinsically by
the property of being normal (i.e. perserved by the conjugation action of the group). We will
show that the kernels of ring homomorphisms can similarly be characterised intrinsically, but
the situation, because we have two binary operations, is slightly different: a kernel is both more
and less than a subring. Indeed since homomorphisms are required to send 1 to 1, the kernel
never contains 1 unless it is the entire ring, thus a kernel is not a subring unless the target of
the homomorphism is the zero ring14 . However, it is closed under addition and multiplication
(as is straight-forward to check) and because 0.x = 0 for any x, it in fact obeys a stronger kind
of closure with respect to multiplication15 : If x ∈ ker( f ) and r ∈ R is any element of R, then
f (x.r) = f (x). f (r) = 0. f (r) = 0 so that x.r ∈ ker( f ). This motivates the following definition:
Definition 3.2. Let R be a ring. A subset I ⊆ R is called an ideal if it is a subgroup of (R, +) and
moreover for any a ∈ I and r ∈ R we have a.r ∈ I. We write I ◁ R to denote the fact that I is an
ideal in R.
Lemma 3.3. If f : R → S is a homomorphism, then ker( f ) is an ideal. Moreover if I ⊆ R then I is an
ideal if and only if it is nonempty, closed under addition, and closed under multiplication by arbitrary
elements of R.
Proof. This is immediate from the definitions. For the moreover part, we just need to check
that I is closed under taking additive inverses. But this follows from the fact that it is closed
under multiplication by any element of R since −x = (−1).x for any x ∈ R. □

Note that if I is an ideal of R which contains 1 then I = R. We will shortly see that in fact
any ideal is the kernel of a homomorphism. First let us note a few basic properties of ideals:
First we need some notation:
Definition 3.4. If X, Y are any subsets of R let AR (X) = {A : X ⊆ A ⊆ R, (A, +) ≤ (R, +)} be
the collection of subgroups of the abelian group (R, +) which contain X. Then for an arbitrary
subset X of R we define
 n 
\ 
X 
⟨X⟩ab = A= ,
 
x : n ∈ x ∈ X, ∀k, 1 ≤ k ≤ n

 k Z ≥0 k 

 
A∈AR (X) k=1

so that ⟨X⟩ab is the additive subgroup of R generated by the set X. This notion

X + Y = ⟨X ∪ Y⟩ab = ⟨X⟩ab + ⟨Y⟩ab

XY = ⟨X · Y⟩ab , where X · Y = {x.y : x ∈ X, y ∈ Y}.
Note that if either of X or Y is empty then X ·Y is also empty, and hence XY = ⟨X ·Y⟩ = ⟨∅⟩Z = {0}.
In general, the additive group XY generated by the pairwise products, i.e. the elements of X · Y,
is not closed under the multiplication operation of the ring R, however if one supposes that
both X and Y are closed under that operation, then XY is also.
14
It’s also worth noticing that if R is the zero ring, and θ : R → S is a ring homomorphism we must have S = R,
since we insist that a ring homomorphism preserves the additive and multiplicative identities.
15
This is analogous to the fact that kernels of group homomorphisms, being normal, are loosely speaking “more
closed" than arbitrary subgroups.

10
Lemma 3.5. Let R be a ring, and I, J ideals in R and X any subset of R. Then I + J, I ∩ J and IX are
ideals. Moreover we have I J ⊆ I ∩ J and I, J ⊆ I + J.
Proof. For I + J it is clear that this is an abelian subgroup of R, while if i ∈ I, j ∈ J and r ∈ R,
then r(i + j) = (r.i) + (r. j) ∈ I + J as both I and J are ideals, hence I + J is an ideal. Checking
I ∩ J is an ideal is similar but easier. To see that IX is an ideal, note that it is clear that the sum
of two elements of IX is of the same form, and if nk=1 ik xk ∈ IX then
P

n
X n
X
r. ik x k = (r.ik ).xk ∈ IX.
k=1 k=1

Thus by the moreover part of Lemma 3.3, IX is an ideal16 . The containments are all clear once
you note that if i ∈ I and j ∈ J then i j in in I ∩ J because both I and J are ideals. □
In fact given a collection of ideals {Iα : α ∈ A} in a ring R, their intersection α∈A Iα is easily
T
seen to again be an ideal. This easy fact is very useful for the following reason:
Definition 3.6. Given any subset T of R, one can define
\
⟨T ⟩ = I
T ⊆I

(where I is an ideal) the ideal generated by T . We can also give a more explicit “from the ground
up” description of the ideal generated by a subset X:
Lemma 3.7. Let T ⊆ R be a nonempty subset. Then we have
⟨T ⟩ = R.T.
Proof. We have already seen that R.T is an ideal (since R itself is an ideal). We first check that
R.T is contained in any ideal I which contains T . But if {x1 , . . . , xk } ⊆ T ⊆ J and r1 , . . . , rk ∈ R,
then since J is an ideal certainly rk xk ∈ J and hence nk=1 rk xk ∈ J. Since the xk , rk and n ∈ N
P
were arbitrary it follows that R.T ⊆ J.
T
It follows that R.T ⊆ I◁R,T ⊆R I, but since R.T is itself an ideal containing T , clearly the
intersection lies in R.T also, so we have the desired equality.
□
This is completely analogous to the notion of the “span” of a subset in a vector space. If I
and J are ideals, it is easy to see that I + J = ⟨I ∪ J⟩. In the case where T = {a} consists of a
single element, we often write aR or17 Ra for ⟨a⟩.
Remark 3.8. Note that in the above, just as for span in a vector space, there is no need for the
set X to be finite.
Remark 3.9. Note that if T ⊂ R is a subset of a ring R we can also consider the subring which
it generates: the intersection of subrings is again a subring18 , so we may set
\
⟨T ⟩ s = S,
T ⊆S

where the subscript “s” is supposed to denote subring. I leave it as an exercise to find a “ground
up” description of ⟨T ⟩ s .
16
This is one reason for the convention that X.Y = {0} if either of X or Y is empty – it ensures I.X an ideal even
when X is empty
17
Since R is commutative Ra = {r.a : r ∈ R} = {a.r : r ∈ R} = aR.
18
Note also that this is a pretty general way of defining the widget “generated” by a subset of a given object:
whatever a widget is, provided the intersection of widgets is again a widget, then if S is some subset of your object,
the widget it “generates” is the intersection of all widgets which contain S – the stability under taking intersections
ensures this intersection is still a widget, and it is thus the smallest widget containing S . The closure of sets in
topological spaces, the ideal generated by a set in a ring and the subring generated by a set in a ring are all defined
in this way.

11
Definition 3.10. If an ideal is generated by a single element we say it is principal. Two elements
a, b ∈ R are said to be associates if there is a unit u ∈ R× such that a = u.b. (This is an equivalence
relation on the elements of R).

If I = ⟨a⟩ then just knowing I does not quite determine a, but it almost does, at least if R is
an integral domain. The notion of associate elements lets us make this precise.

Lemma 3.11. Let R be a domain. Then if I is a principal ideal, the generators19 of I are associates, and
any associate of a generator is again a generator. Thus the generators of a principal ideal form a single
equivalence class of associate elements of R.

Proof. If I = {0} = ⟨0⟩ the claim is immediate, so assume I , {0} and hence any generator
is nonzero. Let a, b ∈ R be generators of I, so I = ⟨a⟩ = ⟨b⟩. Since a ∈ ⟨b⟩, there is some
r ∈ R with a = r.b, and similarly as b ∈ ⟨a⟩ there is some s with b = s.a. It follows that
a = r.b = r(s.a) = (r.s)a, hence a(1 − r.s) = 0, and so since a , 0 and R is an integral domain,
r.s = 1, that is, r and s are units.
Finally if I = ⟨a⟩ and b = u.a where u ∈ R× , then certainly b ∈ ⟨a⟩ = I so that ⟨b⟩ ⊆ I, but also
if x ∈ I, then x = r.a for some r ∈ R and hence x = r.(u−1 .b) = (r.u−1 ).b so that x ∈ ⟨b⟩, and hence
I ⊆ ⟨b⟩. It follows I = ⟨b⟩ as required. □

For example in Z we will see that the ideals are all of the form ⟨n⟩ and the integer n is
determined up to sign by the ideal ⟨n⟩ (the units in Z being exactly {±1}).

3.1 The quotient construction.

In order to show that ideals and kernels of ring homorphisms are the same thing, we now study
a notion of quotient for rings, similar to the quotients of groups and vector spaces which you
have already seen. This is one of the most important constructions in the course. The notion
of a quotient object is a fundamental one (not just in algebra, but also in topology and many
other subjects). It is a subtle thing that usually takes some time to become accustomed to, but
it is absolutely essential.
Suppose that R is a ring and that I is an ideal in R. Then since (I, +) is a subgroup of the
abelian group (R, +), we may form the quotient group (R/I, +). We briefly recall the construc-
tion of the group (R/I, +): The relations r ∼ s if r − s ∈ I is easily checked to be an equivalence
relation,20 and the equivalence classes are the cosets {r + I : r ∈ R}. We want to endow R/I with
the structure of an abelian group in such a way that the map q : R → R/I sending an element
r ∈ R to its equivalence class r + I ∈ R/I is a group homomorphism. This condition forces us to
require
(r1 + I) + (r2 + I) = (r1 + r2 ) + I,
so the only thing to do is to check that this formula makes sense, that is, to check that it is
independent of the choice of representatives r1 .r2 . To see this, suppose that s j ∈ r j + I for
j = 1, 2 are two other choices of representatives. Then there are elements i1 , i2 ∈ I such that
s j = r j + i j ( j = 1, 2), and hence s1 + s2 = (r1 + r2 ) + (i1 + i2 ) ∈ (r1 + r2 ) + I since i1 + i2 ∈ I.
Now that we have shown R/I inherits a binary operation from R, it is easy to see that that
operation makes R/I into an abelian group with identity 0 + I. We now want to show that R/I
has the structure of a ring where the multiplication is again induced from that on R, so that
the map q : R → R/I is actually a ring homomorphism. Again there is at most one possibility
for the multiplication on R/I which could satisfy this condition: we must have

(r1 + I) × (r2 + I) = r1 .r2 + I,

19
i.e. the elements a ∈ R such that I = ⟨a⟩.
20
And you have done similar checks for groups and vector spaces before.

12
or said in terms of the quotient map q : R → R/I we must have q(r1 ).q(r2 ) = q(r1 .r2 ). Again
the issue is whether this is indeed a well-defined operation, i.e. independent of the choice of
representatives r1 , r2 . Thus as before take s j = r j + i j some other representatives for the cosets
r j + I, ( j = 1, 2). Then we have
s1 .s2 = (r1 + i1 ).(r2 + i2 ) = r1 .r2 + (i1 r2 + r1 i2 + i1 i2 ) ∈ r1 .r2 + I
since i1 r2 , r1 i2 , i1 i2 all lie in I since I is an ideal. It follows that we have a well-defined binary
operation on R/I coming from the multiplication in R also.
Theorem 3.12. The datum (R/I, +, ×, 0 + I, 1 + I) defines a ring structure on R/I and moreover the
map q : R → R/I given by q(r) = r + I is a surjective ring homomorphism. Moreover the kernel of q is
the ideal I.
Proof. Checking each axiom is an easy consequence of the fact that the binary operations +, ×
on R/I are defined by picking arbitrary representatives of the cosets, computing up in the ring
R and then taking the coset of the answer (the important part of the definitions being that this
last step is well-defined). Thus for example, to check × is associative, let C1 , C2 , C3 be elements
of R/I and choose r1 , r2 , r3 ∈ R such that Ci = q(ri ) = ri + I for i = 1, 2, 3. Then
C1 × (C2 × C3 ) = q(r1 ) × (q(r2 ) × q(r3 ))
= q(r1 ) × q(r2 r3 ) = q(r1 .(r2 r3 ))
= q((r1 r2 ).r3 )) = q(r1 r2 ) × q(r3 )
= (q(r1 ) × q(r2 )) × q(r3 ) = (C1 × C2 ) × C3 .
where in going from the second to the third line we use the associativity of multiplication
in R. Checking the other axioms is similarly straight-forward. Finally, the map q : R → R/I
is clearly surjective, and that it is a homomorphism is also immediate from the definitions.
Clearly q(r) = 0 ∈ R/I precisely when q(r) = r + I = 0 + I, that is precisely when r ∈ I. Thus
ker(q) = I as required. □
The map q : R → R/I is called the quotient homomorphism (or quotient map). The next
corollary establishes the answer to the question we started this section with: what are the
subsets of R which are kernels of homomorphisms from R? We already noted that any kernel
is an ideal, but the construction of quotients now gives us the converse:
Corollary 3.13. The ideals in R are exactly the kernels of the set of homomorphisms with domain R.
Proof. We have already seen that the kernel of a ring homomorphism is always an ideal so it
only remains to show that any ideal is the kernel of some homomorphism. But this is exactly
what the previous theorem shows: If I is an ideal and q : R → R/I is the quotient map then q is
a ring homomorphism and ker(q) = I. □
For our next result about quotient rings, it may be helpful to compare with the following
result from last term’s linear algebra about quotient vector spaces: If T : V → W is a linear map,
and U < V is a subspace, then T induces a linear map T̄ : V/U → W on the quotient space V/U
if and only if U ⊆ ker(T ).
Theorem 3.14. (Universal Property of Quotients.) Suppose that R is a ring, I is an ideal of R, and
q : R → R/I the quotient homomorphism. If ϕ : R → S is a ring homomorphism such that I ⊆ ker(ϕ),
then there is a unique ring homomorphism ϕ̄ : R/I → S such that ϕ̄ ◦ q = ϕ. That is, the following
diagram commutes:
ϕ
RA /S
AA ||>
AA ||
A
q AA ||ϕ̄
||
R/I
Moreover ker(ϕ̄) is the ideal ker(ϕ)/I = {m + I : m ∈ ker(ϕ)}.

13
Proof. Since q is surjective, the formula ϕ̄(q(r)) = ϕ(r) (r ∈ R) uniquely determines the values
of ϕ̄, so that ϕ̄ is unique if it exists. But if r − r′ ∈ I then since I ⊆ ker(ϕ) it follows that 0 =
ϕ(r − r′ ) = ϕ(r) − ϕ(r′ ) and hence ϕ is constant on the I-cosets, and therefore induces a map
ϕ̄(m + I) = ϕ(m). The fact that ϕ̄ is a homomorphism then follows directly from the definition of
the ring structure on the quotient R/I: For example, to see that ϕ̄ respects multiplication note
that if C1 , C2 ∈ R/I then picking r1 , r2 such that C1 = q(r1 ), C2 = q(r2 ) we have

ϕ̄(C1 .C2 ) = ϕ̄(q(r1 )q(r2 )) = ϕ̄(q(r1 r2 )) = ϕ(r1 r2 ) = ϕ(r1 ).ϕ(r2 )

= ϕ̄(q(r1 ))ϕ̄(q(r2 )) = ϕ̄(C1 ).ϕ̄(C2 ),

where in the above equalities we just use the defining property ϕ̄ ◦ q = ϕ of ϕ̄ and the fact that
q and ϕ are homomorphisms. To see what the kernel of ϕ̄ is, note that ϕ̄(r + I) = ϕ(r) = 0 if and
only if r ∈ ker(ϕ), and hence r + I ∈ ker(ϕ)/I as required. □

Remark 3.15. The name “universal property” is perhaps overly grand, but you should think
of it as analogous to the fact that the ideal generated by a set is characterized as the smallest
ideal containing that set: The quotient R/I is the largest quotient of R which sends all of I to 0,
in the strong sense that if ϕ : R → S is any surjective homomorphism such that ϕ(I) = 0, then
R/I surjects onto S (and thus is “at least as large” as S ).

This theorem has important corollaries

Corollary 3.16. We have the following isomorphisms:

1. (First isomorphism theorem.) If ϕ : R → S is a homomorphism then ϕ induces an isomorphism

ϕ̄ : R/ker(ϕ) → im(ϕ).

2. (Second isomorphism theorem.) If R is a ring, A a subring of R, and I an ideal of R, then

(A + I)/I  A/(A ∩ I),

3. (Third isomorphism theorem.) Suppose that I1 ⊆ I2 are ideals in R. Then we have

(R/I1 )/(I2 /I1 )  R/I2 .

Proof. For the first isomorphism theorem, apply the universal property to I = ker(ϕ). Since in
this case ker(ϕ̄) = ker(ϕ)/ker(ϕ) = 0 it follows ϕ̄ is injective and hence induces an isomorphism
onto its image which from the equation ϕ̄ ◦ q = ϕ must be exactly im(ϕ).
For the second isomorphism theorem, note first that if A is a subring and I is an ideal, it is
easy to check21 that A + I is again a subring of R which contains I as an ideal. Let q : R → R/I
be the quotient map. It restricts to a homomorphism p from A to R/I, whose image is clearly
(A + I)/I, so by the first isomorphism theorem it is enough to check that the kernel of p is A ∩ I.
But this is clear: if a ∈ A has p(a) = 0 then a + I = 0 + I so that a ∈ I, and so a ∈ A ∩ I.
(Note this argument automatically shows that A ∩ I is an ideal of A since it is the kernel of the
homomorphism p).
For the third isomorphism theorem, let qi : R → R/I j for j = 1, 2. By the universal property
for q2 we see that there is a homomorphism q̄2 : R/I1 → R/I2 induced by the map q2 : R → R/I2 ,
with kernel ker(q2 )/I1 = I2 /I1 and q̄2 ◦ q1 = q2 . Thus q̄2 is surjective (since q2 is) and hence the
result follows by the first isomorphism theorem. □
21
See the problem set.

14
Example 3.17. Suppose that V is a k-vector space and α ∈ End(V). Then we saw before that
ϕ : k[t] → End(V) given by ϕ( f ) = f (α). It is easy to see that this map is a homomorphism, and
hence we see that im(ϕ) is isomorphic to k[t]/I where I = ker( f ) is a principal ideal. The monic
polynomial generating I is the minimal polynomial of α as studied in Algebra I.

Another useful application of these results is a general version22 of the “Chinese Remainder
Theorem”. To state it recall from Example 1.2 ix) the direct sums construction for rings: if R
and S are rings, then R ⊕ S is defined to be the ring of ordered pairs (r, s) where r ∈ R, s ∈ S ,
with addition and multiplication done componentwise.

Theorem 3.18. Let R be a ring, and I, J ideals of R such that I + J = R. Then

R/I ∩ J  R/I ⊕ R/J.

Proof. We have quotient maps q1 : R → R/I and q2 : R → R/J. Define q : R → R/I ⊕ R/J by
q(r) = (q1 (r), q2 (r)). By the first isomorphism theorem, it is enough to show that q is surjective
and that ker(q) = I ∩ J. The latter is immediate: if q(r) = 0 then q1 (r) = 0 and q2 (r) = 0, whence
r ∈ I and r ∈ J, that is, r ∈ I ∩ J. To see that q is surjective, suppose (r + I, s + J) ∈ R/I ⊕ R/J.
Then since R = I + J we may write r = i1 + j1 and s = i2 + j2 , where i1 , i2 ∈ I, j1 , j2 ∈ J. But then
r + I = j1 + I and s + J = i2 + J, so that q( j1 + i2 ) = (r + I, s + J). □

Remark 3.19. Suppose that R = I + J where I and J are ideals as above and moreover that
I ∩ J = {0}. Then each r ∈ R can be written uniquely in the form i + j where i ∈ I and j ∈ J
(the proof is exactly the same as it is for subspaces in a vector space). In this situation we
write23 R = I ⊕ J. Note that since I.J ⊆ I ∩ J it follows that i. j = 0 for any i ∈ I, j ∈ I, thus if
i1 , i2 ∈ I and j1 , j2 ∈ J we see (i1 + j1 ).(i2 + j2 ) = i1 i2 + j1 j2 . Writing 1 = e1 + e2 where e1 ∈ I
and e2 ∈ J if follows (I, +, ×, 0, e1 ) is a ring as is (J, +, ×, 0, e2 ), and it is easy to see that these
rings are isomorphic to R/J and R/I respectively. This gives a more explicit description of the
isomorphism R  R/I ⊕ R/J provided by the Chinese Remainder Theorem in this case.
Note also that if we start with two rings S 1 , S 2 , and define R = S 1 ⊕ S 2 as in Example
1.2 ix), then the copies S 1R , S 2R of S 1 and S 2 inside R (that is, the elements {(s, 0) : s ∈ S 1 }
and {(0, t) : t ∈ S 2 } respectively) are ideals in R (not subrings because they do not contain the
multiplicative identity element (1, 1)) and clearly their intersection is {(0, 0)}, so that R = S 1R ⊕S 2R ,
thus the “external” notion of direct sum we saw in lecture 1 is compatible with the “internal”
direct sum notation we used above (that is, when we write R = I ⊕ J to denote that I, J are
ideals in R with I + J = R and I ∩ J = {0}).
When R = Z and I = nZ = {nd : d ∈ Z}, J = mZ, then you can check that I + J = Z precisely
when n and m are coprime, and then it also follows that I ∩ J = (n.m)Z (the problem sheet
asks you to work out the details of this), and so we recover the classical “Chinese Remainder
Theorem”: if m, n are coprime integers, then Z/(nm)Z  (Z/nZ) ⊕ (Z/mZ). For example, if R =
Z/6Z then R = 3̄R⊕ 4̄R (writing n̄ for n+6Z etc.) and this gives the identification R = Z/2Z⊕Z/3Z.

3.2 Images and preimages of ideals.

Next we want to compare ideals in a quotient ring with ideals in the original ring.

Lemma 3.20. Let ϕ : R → S be a surjective homomorphism of rings. If I ◁ R then

ϕ(I) = {s ∈ S : ∃i ∈ I, s = ϕ(i)}
22
The classical Chinese Remainder Theorem shows that if m, n ∈ Z are coprime then for any a, b ∈ Z there is a
solution to the pair of equations x = a mod m and x = b mod n, moreover this solution is unique modulo m.n.
Check you see why the general version stated above implies this.
23
The notation is compatible with the direct sum notation used in the first lecture – see the next paragraph.

15
is an ideal in S . Similarly if J ◁ S then ϕ−1 (J) = {r ∈ R : ϕ(r) ∈ J} is an ideal in R. Thus ϕ induces a
pair of maps:
ϕ
.
{ Ideals in R } n { Ideals in S }
ϕ−1

Proof. Let I ◁ R. Then ϕ(I) is certainly an additive subgroup of S since ϕ is a homomorphism

of additive groups, and if s ∈ S and j = ϕ(i) ∈ ϕ(I), since ϕ is surjective, we may find r ∈ R such
that ϕ(r) = s. It follows that s. j = ϕ(r).ϕ(i) = ϕ(r.i) ∈ ϕ(I) since r.i ∈ I because I is an ideal in R.
If J ◁ S we can consider the subset ϕ−1 (J) = {x ∈ R : ϕ(x) ∈ J}. We claim this is an ideal of
R. Indeed if x, y ∈ ϕ−1 (J) then ϕ(x + y) = ϕ(x) + ϕ(y) ∈ J since J is an additive subgroup, so that
ϕ−1 (J) is an additive subgroup, and if r ∈ R, x ∈ ϕ−1 (J), then ϕ(r.x) = ϕ(r).ϕ(x) ∈ J since ϕ(x) ∈ J
and J is an ideal in S . □

The next proposition shows that these maps can be used to identify the ideals of S with the
subset of the ideals of R consisting of those ideals which contain the kernel of the homomor-
phism ϕ.

Proposition 3.21. Let ϕ : R → S be a surjective ring homomorphism and let K = ker(ϕ) ◁ R. Then

1. If J ◁ S then we have ϕ(ϕ−1 (J)) = J;

2. If I ◁ R then we have ϕ−1 (ϕ(I)) = I + K.

In particular the maps J 7→ ϕ−1 (J) and I 7→ ϕ(I) induce bijections between the set of ideals in S and the
set of ideals in R which contain K:
ϕ
.
{ Ideals in R containing K } o { Ideals in S }
ϕ−1

Proof. For the first part, note that if f : X → Y is any map of sets and Z ⊆ Y then f ( f −1 (Z)) =
Z ∩ im( f ). Thus because ϕ is surjective we see that for any subset J ⊆ S (and in particular for
any ideal) ϕ(ϕ−1 (J)) = J.
For the second part, note that if I ◁ R then 0 ∈ ϕ(I), and so K = ker(ϕ) = ϕ−1 (0) ⊆ ϕ−1 (ϕ(I)).
Since I ⊆ ϕ−1 (ϕ(I)) also, it follows that I + K, the ideal generated by I and K, must lie in the
ideal ϕ−1 (ϕ(I)). To see the reverse inclusion, note that if x ∈ ϕ−1 (ϕ(I)) then by definition there
is some i ∈ I with ϕ(x) = ϕ(i), and hence ϕ(x − i) = 0. But then x = i + (x − i) ∈ I + K, so that
ϕ−1 (ϕ(I)) ⊆ I + K as required.
Finally, to see the bijective correspondence, note we have already seen that for an ideal J ◁S
we have ϕ(ϕ−1 (J)) = J, and since K ⊆ ϕ−1 (J) it follows that J 7→ ϕ−1 (J) is an injective map whose
image lands in the set of ideals of R which contain K. On the other hand, if I ⊇ K is an ideal in
R the I + K = I and so ϕ−1 (ϕ(I)) = I, so that I 7→ ϕ(I), when restricted to the set of ideals of R
which contain K, is the inverse map to J 7→ ϕ−1 (J) as required. □

In particular, if I ◁ R and we take ϕ = q to be the canonical quotient homomorphism q : R →

R/I we get the following:

Corollary 3.22. Let R be a ring, I an ideal in R and q : R → R/I the quotient map. If J is an ideal then
q(J) is an ideal in R/I, and if K is an ideal in R/I then q−1 (K) = {r ∈ R : q(r) ∈ K} is an ideal in R which
contains I. Moreover, these correspondences give a bijection between the ideals in R/I and the ideals in
R which contain I.

16
4 Prime and maximal ideals, Euclidean domains and PIDs.
The quotient construction gives us a powerful way to build new rings and fields. The properties
of the rings we obtain as quotients depend on the properties of the ideals we quotient by, and
this leads us to the study of certain classes of ideals. In this section we begin studying two
important such classes.

Definition 4.1. Let R be a ring, and I an ideal of R. We say that I is a maximal ideal if it is not
strictly contained in any proper ideal of R. We say that I is a prime ideal if I , R and for all
a, b ∈ R, whenever a.b ∈ I then either a ∈ I or b ∈ I. If a prime I is principal any generator of I
is said to be a prime element.

Lemma 4.2. An ideal I in a ring R is prime if and only if R/I is an integral domain24 . It is maximal if
and only if R/I is a field. In particular, a maximal ideal is prime.

Proof. Suppose that a, b ∈ R. Note that (a + I)(b + I) = 0 + I if and only if a.b ∈ I. Thus if R/I is
an integral domain, (a + I)(b + I) = 0 forces either a + I = 0 or b + I is zero, that is, a or b lies in
I, which shows I is prime. The converse is similar.
For the second part, note that a field is a ring which has no nontrivial ideals (check this!).
The claim then follows immediately from the correspondence between ideals in the quotient
ring and the original ring given in Lemma 3.22. Since fields are obviously integral domains,
the “in particular” claim follows immediately. □

Remark 4.3. You can also give a direct proof that a maximal ideal is prime. Indeed if I is
maximal and a.b ∈ I, and suppose that b < I. Then the ideal J = I + bR generated by I and b is
strictly larger than I, and so since I is maximal it must be all of R. But then 1 = i + br for some
i ∈ I and r ∈ R, and hence a = a.1 = a.i + (a.b)r ∈ I since i, a.b ∈ I as required.

Example 4.4. Let R = Z. Since an ideal I in Z is in particular an subgroup of the abelian group
Z, we know it must be cyclic, that is I = dZ for some integer d. Thus every ideal in Z is principal.
An ideal dZ is prime exactly when d is prime, and since in that case Z/dZ is a field provided
d , 0 it follows the maximal ideals are exactly the nonzero prime ideals.

We now consider a more substantial example, that of polynomials in one variable over a
field. Although the case of field coefficients is the only one we really need for the moment, the
following lemma captures, for polynomials with coefficients in a general ring, when you can
do “long division with remainders” in polynomial rings. For this we first need to recall the
notion of the degree of a nonzero polynomial:

Definition 4.5. If R is a ring and f ∈ R[t] is nonzero, then we may write f = ni=0 ai ti , where
P
an , 0. We set the degree deg( f ) of f to be n, and say an is a the leading coefficient of f . If R is an
integral domain, then for any f, g ∈ R[t] you can check that deg( f.g) = deg( f ) + deg(g) (and so
in particular this implies R[t] is also an integral domain).

Lemma 4.6. (Division Algorithm). Let R be a ring and f = ni=0 ai ti ∈ R[t], where an ∈ R× . Then if
P
g ∈ R[t] is any polynomial, there are unique polynomials q, r ∈ R[t] such that either r = 0 or deg(r) <
deg( f ) and g = q. f + r.

Proof. This is straight-forward to prove by induction on deg(g). Since the an ∈ R× , if h ∈ R[t]\{0}

it is easy to see25 that deg( f.h) = deg( f ) + deg(h). It follows that if deg(g) < deg( f ) we must take
24
Note that this is “why” one wants to exclude R from being a prime ideal – I defined an integral domain to be a
ring which was not the zero ring and had no zero divisors.
25
The key here is that a unit is never a zero-divisor: if a.b = 0 and a is a unit, then b = (a−1 .a).b = a−1 .(a.b) = a−1 .0 =
0.

17
q = 0 and thus r = g. Now suppose that g = mj=0 b j t j where bm , 0 and m = deg(g) ≥ n = deg( f ).
P

Then since a−1 m−n . f has leading term b tm the polynomial

n bm t m

h = g − a−1
n bm t
m−n
. f,

has deg(h) < deg(g). It follows by induction that there are unique q′ , r′ with h = q′ . f + r′ . Setting
q = a−1
n bn t
m−n + q′ and r = r ′ it follows g = q. f + r. Since q and r are clearly uniquely determined

by q and r′ they are also unique as required.

′ □

It follows from the previous lemma that if k is a field, then we have the division algorithm
for all non-zero polynomials. This allows us to prove that all ideals in k[t] are principal.

Lemma 4.7. Let I be a nonzero ideal in k[t]. Then there is a unique monic polynomial f such that
I = ⟨ f ⟩. In particular, all ideals in k[t] are principal.

Proof. Since I is nonzero we may pick an f ∈ I of minimal degree, and rescale it if necessary to
make it monic. We claim I = ⟨ f ⟩. Indeed if g ∈ I, then using the division algorithm, we may
write g = q. f + r where either r = 0 or deg(r) < deg( f ). But then r = g − q. f ∈ I, and thus
by the minimality of the degree of f ∈ I we must have r = 0 and so g = q. f as required. The
uniqueness follows26 from the fact that if I = ⟨ f ⟩ and I = ⟨ f ′ ⟩ then we would have f = a. f ′ and
f ′ = b. f , for some polynomials a, b ∈ k[t]. But then f = a. f ′ = (ab). f so that a and b must have
degree zero, that is, a, b ∈ k. Since we required f and f ′ to be monic, it follows that a = b = 1
and so f = f ′ as required. □

The division algorithm also allows to give a reasonably explicit description of the rings
we obtain quotient of a polynomial ring k[t]: We have just seen that any nonzero ideal I is
of the form ⟨ f ⟩ for a monic polynomial f . By the division algorithm, any polynomial g can
be written uniquely as g = q. f + r where deg(r) < deg( f ). Thus the polynomials of degree
strictly less that d = deg( f ) form a complete set of representatives for the I-cosets: every coset
contains a unique representative r of degree strictly less than deg( f ). Since {1, t, . . . , tdeg( f )−1 }
form a basis of the k-vector space of polynomials of degree less than deg( f ) this means that if
we let q : k[t] → k[t]/I be the quotient map, and α = q(t), then {1, α, . . . , αd−1 } form a k-basis
for k[t]/I, and we multiply in k[t]/I using the rule αd = −a0 − a1 α − . . . − ad−1 αd , where f (t) =
td + d−1 i
P
i=0 ai t . In particular, k[t]/⟨ f ⟩ is a k-vector space of dimension deg( f ). We can therefore
interpret the quotient construction k[t]/⟨ f ⟩ as a way of building a new ring out of k and an
additional element α which satisfies the relation f (α) = 0, or rather, the quotient construction
gives us a rigorous way of doing this. The following example shows how one can use this to
give a new construction of the complex numbers.

Example 4.8. When k = R, intuitively we build C out of R and an element “i” which satisfied
i2 + 1 = 0. The quotient construction lets us make this intuition rigorous: we simply define C to
be the quotient ring R[t]/⟨t2 + 1⟩. Indeed this is a field because t2 + 1 is irreducible27 in R[t] (see
Lemma 4.17 below for more on this) and if we let i denote the image of t under the quotient
map from R[t] to C, then C = R[t]/⟨t2 + 1⟩ is a two-dimensional R-vector space with basis {1, i}
and i satisfies i2 + 1 = 0.

Remark 4.9. In fact with a little more care28 it is straight-forward to check that if R is any ring
and f ∈ R[t] is a monic polynomial of degree d, and we let Q = R[t]/⟨ f ⟩ and α = q(t) (where
q : R[t] → R[t]/⟨ f ⟩ is the quotient map as before) then any element of Q can be written uniquely
26
This also follows from the fact that generators of a principal ideal are all associates, and the fact that the units
in k[t] are exactly k× .
27
In general it is not so easy to decide if a polynomial f ∈ k[t] is irreducible, but in the case where deg( f ) ≤ 3, f is
reducible if and only if it has a root in k, which can (sometimes) be easy to check.
28
In particular, one needs to use the general statement of the division algorithm as given in Lemma 4.6.

18
in the form r0 + r1 α + . . . + rd−1 αd−1 , where the multiplication in Q is given by the same rule as
above. Of course for a general ring, not all ideals in R[t] will necessarily be principal, and even
if I = ⟨ f ⟩, if the leading coefficient of f is not a unit, we cannot apply the division algorithm.

Notice that the argument we used in the proof of Lemma 4.7 runs exactly the same way
as the proof that every subgroup of (Z, +) is cyclic (or that any ideal in Z is principal). This
suggests it might be useful to abstract the division algorithm for a general integral domain.

Definition 4.10. Let R be an integral domain and let N : R\{0} → N be a function. We say that
R is a Euclidean domain if given any a, b ∈ R with b , 0 there are q, r ∈ R such that a = b.q + r
and either r = 0 or N(r) < N(b).

Remark 4.11. Some texts require that the norm N satisfies additional properties, and in practice
these additional properties are often very useful. For example sometimes the norm satisfies
N(a.b) = N(a).N(b) (in which case the norm is said to be multiplicative) or N(a.b) = N(a) + N(b).
The most general additional property one often asks for is that N(a) ≤ N(a.b) for all a, b ∈ R\{0}.
You can check that if R is a Euclidean domain satisfying this last property then the group of
units R× is precisely the set {a ∈ R : N(a) = N(1)}. However, if one just wants to know the ring
is a PID the only condition one needs is the division algorithm.

Both Z and k[t], for any field k, are Euclidean domains with the norm given by the absolute
value and the degree function respectively. We now show that the Gaussian integers Z[i] gives
another example:

Lemma 4.12. Let R = Z[i] and let N : R → N be the function N(z) = a2 + b2 , where z = a + ib ∈
Z[i], a, b ∈ Z. Then (R, N) is an Euclidean Domain.

Proof. Note that N is the restriction of the square of the modulus function on C, so in particular
N(z.w) = N(z).N(w). We write |z|2 instead of N(z) when z ∈ C\Z[i]. Suppose that s, t ∈ Z[i] and
t , 0. Then s/t ∈ C, and writing s/t = u + iv where u, v ∈ Q we can clearly take a, b ∈ Z such that
|u − a|, |v − b| ≤ 1/2 and so q = a + ib we have |s/t − q|2 ≤ 14 + 14 = 12 , and so N(s − qt) ≤ 21 N(t) (since
N(z1 z2 ) = N(z1 ).N(z2 )) and hence if r = s − qt ∈ Z[i] we see that either r = 0 or N(r) ≤ 12 N(t) < N(t)
as required. Note that r is not necessarily unique in this case. □

Lemma 4.13. Let (R, N) be an Euclidean domain. Then any ideal in R is principal.

Proof. The proof that any ideal is principal is exactly the same as for k[t]: If I is a nonzero ideal,
take d ∈ I such that N(d) is minimal. Then if m ∈ I we may write m = q.d + r, where r = 0 or
N(r) < N(d). But r = m − q.d ∈ I so that the minimality of N(d) forces r = 0 and so m = q.d. It
follows that I ⊆ R.d, and since d ∈ I clearly Rd ⊆ I, hence I = Rd as required. □

Definition 4.14. An integral domain in which every ideal is principal, that is, generated by
a single element, is called a Principal Ideal Domain. This is usually abbreviated to PID. The
previous Lemma shows that any Euclidean Domain is a PID.

Remark 4.15. It is also possible to consider rings in which every ideal is principal but which are
not necessarily integral domains29 . Such rings are called Principal Ideal Rings. As we mostly
focus on integral domains, we will not however use this term in this course.

We would like to calculate which ideals in a Euclidean domain are prime and which are
maximal. In fact we can give an answer for any PID not just any Euclidean domain.
29
As an exercise, you might try to find an example of such a ring.

19
Definition 4.16. Let R be an integral domain. A nonzero element r ∈ R is said to be irreducible
if whenever r = a.b then exactly one of a or b is a unit (so that in particular r is not a unit). We
will say an element R ∈ R\({0} ∪ R× ) is reducible if it is not irreducible30 .

Lemma 4.17. Let R be a PID, and let d ∈ R\{0}. Then the following are equivalent:

1. R.d = ⟨d⟩ is a prime ideal.

2. d is irreducible in R.

3. R.d is a maximal ideal in R.

Proof. 1) implies 2): If d = a.b then as d ∈ R.d is prime we must have a ∈ R.d or b ∈ R.d. By
symmetry we may assume a ∈ R.d (and hence, since R.d is a proper ideal and R.a ⊆ R.d we see
that a is not a unit31 ). But then there is some r ∈ R with a = r.d, and so d = a.b = (r.b).d and
hence (1 − r.b).d = 0 and so since R is an integral domain and d , 0 we must have r.b = 1, that
is b ∈ R× .
2) implies 3): Suppose that d is irreducible, and that R.d ⊆ I ◁ R. Since R is a PID, we must
have I = R.a for some a ∈ R, and R.d ⊆ R.a shows that d = a.b for some b ∈ R. But then as d
is irreducible we must have one of a or b a unit. But if a is a unit, then R.a = R, while if b is
a unit d and a are associates and so generate the same ideal, that is R.d = I. It follows R.d is a
maximal ideal as claimed.
3) implies 1): We have already seen that in any ring a maximal ideal must be prime.
□

Remark 4.18. Note that the implication “1) implies 2)” holds in any integral domain, while “3)
implies 1)” holds in any commutative ring. In a general ring d ∈ R irreducible is equivalent to
the ideal R.d being maximal amongst principal ideals in R.
It is also worth pointing out that the Lemma reduces the problem classifying prime and
maximal ideals in a PID R to the problem of finding irreducible elements in R. When R is
say C[t], this is easy: by the fundamental theorem of algebra a monic polynomial p ∈ C[t] is
irreducible if and only if p = t − λ for some λ ∈ C. On the other hand if R = Q[t] then it is in
general very difficult to decide if a polynomial p ∈ Q[t] is irreducible. For the ring R = Z[i] it is
possible to give a fairly complete description of the irreducibles: see the problem sheet.

5 An introduction to fields.
In the previous section we saw how to construct C from R as the quotient C  R[t]/⟨t2 + 1⟩. This
example generalises substantially, and in this section we use the quotientswe have developed
to construct some examples of fields, and develop a little of their basic properties. Our main
tool is Lemma 4.17, which shows that if f ∈ k[t] is any irreducible polynomial then k[t]/⟨ f ⟩ is a
field, and moreover by the above discussion it is clearly a k-vector space of dimension deg( f ).

Example 5.1. Suppose that E be a finite field (i.e. a field with finitely many elements). Then E
has characteristic p for some prime p ∈ N (since otherwise E contains a copy of Z and is hence
infinite). Thus E contains the subfield F p  Z/pZ. In particular we can view it as an F p -vector
space, and since it is finite, it must certainly be finite-dimensional. But then if d = dimF p (E),
clearly there are pd elements in E. Thus we see that a finite field must have prime-power order.
30
On the one hand, since units have no prime factors, it seems reasonable to consider them not to be reducible,
but on the other hand, we do not want them to be irreducible: we will show any nonzero element of a PID is a
product of irreducibles in an essentially unique way, and this uniqueness would not make sense if we allow units
to be irreducible.
31
Check you see that an element r of a ring R is a unit if and only if R.r = R.

20
 Let’s see an explicit example: Take for example p = 3. Then it is easy to check that t2 + 1
is irreducible in F3 [t] (you just need to check it does not have a root in F3 , and there are only 3
possibilities!). But then by our discussion above E = F3 [t]/⟨t2 + 1⟩ is field of dimension 2 over
F3 , and hence E is a finite field with 9 elements.
More generally, if we can find an irreducible polynomial f of degree d in F p [t] the quotient
F p [t]/⟨ f ⟩ will be a finite field of order pd . In the Problem sheets we will show that for each
d there is an irreducible polynomial of degree d in F p [t], hence showing that finite fields of
any prime-power order exist. In fact there is only one field (up to isomorphism) of any fixed
prime-power order, but we will not prove that in this course.

Definition 5.2. If E, F are any fields and F ⊆ E we call E a field extension of F and write E/F.
The inclusion of F into E gives E the structure of an F-vector space. If E is finite dimensional
as an F-vector space, we write [E : F] = dimF (E) for this dimension and call it the degree of the
field extension E/F.

Although it probably seems a very crude notion, as it forgets alot of the structure of E, the
degree of a field extension is nevertheless very useful. One reason for this is the following
Lemma:

Lemma 5.3. Let E/F be a field extension and let d = [E : F] < ∞. Then if V is an E-vector space, we
may view V as an F-vector space, and V is finite dimensional as an F-vector space if and only if it is as
an E vector space, and moreover dimF (V) = [E : F] dimE (V).

Proof. Certainly if V is an E-vector space then by restricting the scalar multiplication map to
the subfield F it follows that V is an F-vector space. Moreover, if V is finite dimensional as
an F-vector space it is so as an E-vector space (a finite F-spanning set will certainly be a fi-
nite E-spanning set). Conversely, suppose that V is a finite dimensional E-vector space. Let
{x1 , x2 , . . . , xd } be an F-basis of E, and let {e1 , . . . , en } be an E-basis of V. To finish the proof it
is enough to check that {xi e j : 1 ≤ i ≤ d, 1 ≤ j ≤ n} is an F-basis of V: Indeed if v ∈ V, then
since {e1 , . . . , en } is an E-basis of V there are λi ∈ E (1 ≤ i ≤ n) such that v = ni=1 λi ei . Moreover,
P
since {x1 , . . . , xd } is an F-basis of E then for each λi there are elements µij (1 ≤ j ≤ d) such that
λi = dj=1 µij x j . Thus we have
P

n
X n
X d
X X
v= λi ei = µij x j ei = µij (x j ei ),


i=1 i=1 j=1 1≤i≤n,1≤ j≤d

whence the set {x j ei : 1 ≤ i ≤ n, 1 ≤ j ≤ d} spans V as an F-vector space (and in particular we

have already established that V is finite dimensional as an F-vector space). To see that this set is
linearly independent, and hence establish the dimension formula, just notice that in the above
equation, v = 0 if and only if each λi = 0 by the linear independence of the vectors {e1 , . . . , en },
j
and λi = 0 if and only if each µi = 0 for 1 ≤ j ≤ d by the linear independence of the x j s. □

Example 5.4. Let V be a C vector space with basis {e1 , . . . , en }. Then since {1, i} is an R-basis of
C, it follows {e1 , . . . , en , ie1 , . . . , ien } is an R-basis of V.

We record a particularly useful case of the above Lemma:

Corollary 5.5. (Tower Law) Let F ⊂ E ⊂ K be fields, then [K : F] is finite if and ony if both degrees
[E : F], [K : E] are, and when they are finite we have [K : F] = [E : F][K : E].

Proof. Apply the previous Lemma to the E-vector space K. □

We now use these tools to study finite extensions of Q inside the field of complex numbers.
The problem sheets also study finite fields, that is, finite extensions of Z/pZ.

21
Definition 5.6. Let α ∈ C. We say that α is algebraic over Q if there is a field E which is a finite
extension of Q containing α. Otherwise we say that α is transcendental. Notice that since the
intersection of subfields is again a subfield32 , given any set T ⊆ C there is always a smallest
subfield which contains it. This is called the field generated by T , and is denoted Q(T ) (recall
that any subfield of C contains Q, since it contains Z and hence Q because it is the field of
fractions of Z). In the case where X has just a single element α we write Q(α) rather than Q({α})
and we say the field extension is simple. Note that an element α ∈ C is algebraic if and only if
Q(α) is a finite extension of Q. Slightly more generally, if F is any subfield of C and α ∈ C we
let F(α) = Q(F ∪ {α}) be the smallest subfield of C containing both F and α, and one says α is
algebraic over F if F(α)/F is a finite extension.

The next Lemma shows that simple extensions are exactly the kind of fields our quotient
construction builds.

Lemma 5.7. Suppose that E/F is a finite extension of fields (both say subfields of C) and let α ∈ E.
Then there is a unique monic irreducible polynomial f ∈ F[t] such that the evaluation homomorphism
ϕ : F[t] → E given by sending t to α induces an isomorphism F(α)  F[t]/⟨ f ⟩.

Proof. The field K = F(α) is a finite extension of F since it is a subfield of E (and hence a sub-
F-vector space of the finite dimensional F-vector space E). Let d = [K : F] = dimF (K). As the
set {αi : 0 ≤ i ≤ d} has d + 1 elements, it must be linearly dependent, and hence there exist
λi ∈ F (0 ≤ i ≤ d), not all zero, such that di=0 λi αi = 0. But then if g = di=0 λi ti ∈ F[t]\{0},
P P
we see that g(α) = 0. It follows that the kernel I of the homomorphism ϕ : F[t] → E given by
ϕ( mj=0 c j t j ) = mj=0 c j α j is nonzero. Now any nonzero ideal in F[t] is generated by a unique
P P
monic polynomial, thus we have I = ⟨ f ⟩, where f is monic and f is uniquely determined by
ϕ (and so by α). By the first isomorphism theorem, the image S of ϕ is isomorphic to F[t]/I.
Now S is a subring of a field, so certainly an integral domain, hence ⟨ f ⟩ must be a prime ideal,
and by our description of prime ideals in F[t] it must therefore in fact be maximal, so that S
is actually a field. Finally, any subfield of C containing F and α must clearly contain S (as the
elements of S are F-linear combinations of powers of α) so it follows S = F(α). □

Definition 5.8. Given α ∈ C, the polynomial f associated to α by the previous Lemma, that is,
the irreducible polynomial for which Q(α)  Q[t]/⟨ f ⟩, is called the minimal polynomial of α over
Q. Note that our description of the quotient Q[t]/⟨ f ⟩ shows that [Q(α) : Q] = deg( f ), hence the
degree of the simple field extension Q(α) is just the degree of the minimal polynomial of α.

Remark 5.9. (Non-examinable) For simplicity let’s suppose that all our fields are subfields of
C. It is in fact the case that any finite extension E/F is simple, that is E = F(α) for some
α ∈ E (this is known as the primitive element theorem, which is proved in next year’s Galois
theory course). Moreover it turns out that given any finite extension E/F of a field F there
are in fact only finitely many fields K between E and F. Neither statement is obvious, but you
should think about how the two facts are clearly closely related: if you accept the statement
about finitely many subfields between E and F then it is not hard to believe the primitive
element theorem – you should just pick an element of E which does not lie in any proper
subfield, and to see such an element exists one just has to show that the union of finitely many
proper subfields of E cannot be the whole field E. On the other hand, if E/F is a finite field
extension and we know that E = F(α) for some α ∈ E, then we have E  F[t]/⟨ f ⟩ where
f ∈ F[t] is the minimal polynomial of α over F. If K is a field with F ⊆ K ⊆ E, then certainly
E = K(α) also, and it follows E  K[t]/⟨g⟩, where g ∈ K[t] is irreducible. But now you can check
(using the tower law) that if g = ki=0 ci ti ∈ K[t], then the ci s actually generate K over F, that is
P
K = F({ci : 0 ≤ i ≤ k}), thus the the possible subfields of F(α) are all determined already by the
roots of f (as the ci s are just polynomial functions of the roots).
32
Just as for subspace of vector space, subrings of a ring, ideals in a ring etc.

22
 √
Example 5.10. 1. Consider
√ 3 ∈ C. There is a unique ring homomorphism ϕ : Q[t] → C
such that ϕ(t) = 3. Clearly the ideal ⟨t2 − 3⟩ lies in ker(ϕ), and since t2 − 3 is irreducible
in Q[t] so that ⟨t2 − 3⟩ is a maximal ideal, we see that ker ϕ = ⟨t2 − 3⟩, and hence im(ϕ) 
Q[t]/⟨t2 − 3⟩. Now the quotient 2
√ Q[t]/⟨t − 3⟩ is field, hence im(ϕ) is also. Moreover, √ any
subfield of C which contains 3 clearly contains im(ϕ), so we see that imϕ = Q( 3). In
particular, since the images of {1, t} form a basis of the quotient Q[t]/⟨t2 −3⟩ by our descrip-
tion of quotients of polynomial rings
√ in the previous section, and√under the isomorphism
√
induced by ϕ these map to 1 and 3 respectively, we see that Q( 3) = {a + b 3 : a, b ∈ Q},
a degree two extension of Q. (Note that one can also just directly check that the right-
hand side of this equality is a field – I didn’t do that because I wanted to point out the
existence of the isomorphism with Q[t]/(t2 − 3).)

2. Exactly the same strategy33 shows that Q(21/3 ) is isomorphic to Q[t]/⟨t3 − 2⟩, and hence
Q(21/3 ) is a 3-dimensional Q-vectors space with basis {1, 21/3 , 22/3 }, again given by the
image of the standard basis we defined in the 3
√ quotient Q[t]/⟨t − 2⟩. Note that while its
relatively easy to check directly that {a + b 3 : a, b ∈ Q} is a subfield of C, it’s already
noticeably harder to see directly that {a + b21/3 + c22/3 : a, b, c ∈ Q} is a subfield of C: one
needs to show that for any a, b, c ∈ Q not all zero, the reciprocal (a + b21/3 + c22/3 )−1 can
be written as a Q-linear combination of {1, 21/3 , 22/3 }.
√ 1/3
Example 5.11. Now let T = { √ 3, 2 }. Let us figure out what E = Q(T ) looks like. Certainly it
contains the subfields E1 = Q( 3) and E2 = Q(21/3 ).
Using the tower law and the above examples, we see that [E : Q] = [E : E1 ].[E1 : Q] = 2[E :
E1 ], and similarly [E : Q] = 3[E √ : E2 ]. It follows that 6 = l.c.m.{2, 3} divides [E : Q]. On the
other hand, consider E/E2 . If 3 ∈ E2 , then clearly√E = E2 , which would mean [E : Q] = 3,
which is not divisible by 6, so that we must have 3 < E2 . But then√ arguing exactly as we
did above, there is a unique homomorphism E2 [t] → C sending t to 2, with 2 − 3⟩,
√ √ kernel ⟨t
1/3
a maximal ideal since 3 < E2 , so that im(ϕ) must be the field generated by 2 and 2 , and
thus [E : E2 ] = dimE2 (E2 [t]/⟨t2 − 3⟩) = 2, and so [E : Q] = [E : E2 ][E2 : Q] = 2.3 = 6. Moreover,
using the proof of the tower law, you can check that the arguments above even show that E
has a Q-basis given by {2a/3 3b/2 : 0 ≤ a ≤ 2, 0 ≤ b ≤ 1}.

6 Unique factorisation.
Throughout this section unless otherwise explicitly stated all rings are integral domains.
For the integers Z, any integer can be written as a product of prime numbers in an essentially
unique way. (See the Appendix for a direct proof of this, which may be useful to review before
reading this section.) We will show in this section that this property holds for any Principal
Ideal Domain.

Definition 6.1. Let R be an integral domain. If a, b ∈ R we say that a divides b, or a is a factor of b,

and write a|b, if there is some c ∈ R such that b = a.c. Note that we can also write this in terms
of the ideals a and b generate: in fact a|b if and only if bR ⊆ aR, as you can see immediately
from the definitions.

It also makes sense to talk about least common multiples and highest common factors in
any integral domain:

Definition 6.2. Let R be an integral domain. We say c ∈ R is a common factor of a, b ∈ R if c|a and
c|b, and that c is the highest common factor, and write c = h.c.f.(a, b), if whenever d is a common
factor of a and b we have d|c. In the same way, we can define the least common multiple of
33
We just need to check that t3 − 2 ∈ Q[t] is irreducible, but this follows because it does not have a root in Q.

23
a, b ∈ R: a common multiple is an element k ∈ R such that a|k and b|k, and the least common
multiple is a common multiple which is a factor of every common multiple.
Note that these definitions can be rephrased in terms of principal ideals: c is a common
factor of a, b if and only if {a, b} ⊆ cR. An element g is the highest common factor of {a, b} if and
only if gR is minimal among principal ideals containing {a, b}, that is, if {a, b} ⊆ cR then gR ⊆ cR.
Similarly the l is the least common multiple of {a, b} if it lR is maximal among principal ideals
which lie in aR ∩ bR.

Lemma 6.3. If a, b ∈ R where R is an integral domain, then if a highest common factor h.c.f{a, b} exists,
it is unique up to units. Similarly when it exists, the least common multiple is also unique up to units.
Moreover if R is a PID then the highest common factor and least common multiple alway exist.

Proof. This is immediate from our description of the highest common factor in terms of ideals.
Indeed if g1 , g2 are two highest common factors, then we must have g1 R ⊆ g2 R (since g1 is a
highest common factor and g2 is a common factor) and symmetrically g2 R ⊆ g1 R. But then
g1 R = g2 R, and so since R is an integral domain this implies g1 , g2 are associates, i.e. they differ
by a unit. The proof for least common multiples is analogous.
If R is a PID then the ideal ⟨a, b⟩ is principal, and so is clearly the minimal principal idea
containing a, b, and so any generator of it is a highest common factor. Similarly Ra ∩ Rb is
principal and any generator of it will be a least common multiple. □

Recall that an nonzero element c in an integral domain R is irreducible if whenever c = a.b

exactly one of a or b is a unit34 , and that a nonzero element p ∈ R is prime if the ideal Rp is
prime. More explicitly p ∈ R is prime if whenever p|a.b either p|a or p|b (or possibly both).
Note that it follows by induction on m that if p is prime and p|a1 .a2 . . . am then p|a j for some
j (1 ≤ j ≤ m). Moreover, by Lemma 4.17 and the remark immediately following it, if R is an
integral domain then prime elements are always irreducible, and for elements of a PID the two
notions are equivalent.
We now want to study factorisation in an integral domain, and in particular the question of
when one can uniquely factor elements into a product of irreducibles. We formalise this with
the following definition.

Definition 6.4. An integral domain R is said to be an unique factorisation domain (or UFD) if
every element of R\{0} is either a unit, or can be written as a product of irreducible elements,
and moreover the factorization into irreducibles is unique up to reordering and units. More
explicitly, if R is a UFD and r ∈ R is nonzero and not a unit, then there are irreducible elements
p1 , . . . , pk such that r = p1 p2 . . . pk and whenever r = q1 q2 . . . ql is another such factorization
for r, then k = l and the q j s can be reordered so that q j = u j p j , where u j ∈ R is a unit. If,
as is normal, we interpret an empty product in a ring to be 1, then we can rephrase this too
include the units in the assertion so that any nonzero element can be expressed as a product
of irreducibles uniquely up to order and units.

Lemma 6.5. Suppose that R is an integral domain. Then the following are equivalent:

1. R is a UFD.

2. Both of the following hold:

i) Every irreducible element is prime,

ii) Every nonzero non-unit a ∈ R can be written as a product of irreducibles.

3. Every nonzero non-unit a ∈ R can be written as a product of prime elements.

34
We also say an element a of an integral domain R is reducible if it is non-zero, not a unit, and not irreducible, i.e.
if we can write it as a product a = b.c where b, c ∈ R\{0} and neither b or c is a unit.

24
Proof. We first show R is a UFD if and only if i) and ii) of (2) holds: Suppose that R is a UFD and
p is an irreducible. If p divides a.b, where a, b ∈ R, then if either a or b is zero or a unit we are
done. Otherwise by assumption they can be written as a product irreducibles, say a = q1 . . . qk
and b = r1 . . . rl for some k, l ≥ 1. But we have a.b = p.d by definition, and writing d = s1 . . . sm
as a product of irreducibles, by uniqueness of the factorization of a.b into irreducibles we see
that that up to units p must be one of the qi s or r j s, and hence p divides a or b as required.
For the converse, we use induction on the minimal number M(a) of irreducibles (or equiv-
alently, primes) in a factorization of a into irreducibles. If M(a) = 1 then a irreducible and
uniqueness is clear by the definition of an irreducible element35 . Now suppose that M =
M(a) > 1 and a = p1 p2 . . . p M = q1 q2 . . . qk for irreducibles pi , q j and k ≥ M. Now it follows
that p1 |q1 . . . qk , and so since p1 is prime there is some q j with p1 |q j . Since q j is irreducible, this
implies that q j = u1 p1 for some unit u1 ∈ R. Reordering the ql s if needed we can assume j = 1,
and so we see that (u−1 1 p2 ) . . . p M = q2 q2 . . . qk , and by induction it follows that k − 1 = M − 1,
i.e. k = M, and moreover the irreducibles occuring are equal up to reordering and units as
required.
To see that condition (2) is equivalent to condition (3), note that since prime elements are
always irreducible, we need only check that irreducibles are prime. But if a ∈ R is irreducible
and a is a product of primes, say a = p1 p2 . . . pk , then by the definition of irreducibility we must
have k = 1 and hence a is prime as required. □

We are now going to show that unique factorisation holds in any PID. By the above, since
we already know that irreducibles are prime in a PID, it is enough to show that any element
has some factorization into irreducibles. At first sight this seems like it should be completely
obvious: if an element a ∈ R is irreducible, then we’re done, otherwise it has a factorisation
a = b.c. where b, c are proper factors (that is, b|a and c|a and neither are associates of a). If either
of b or c is not irreducible then we can find a proper factorisation of them and keep going until
we reach a factorisation of a into irreducibles. The trouble with this argument is that we need
to show the process we describe stops after finitely many steps. Again intuitively this seems
clear, because the proper factors of a should be “getting smaller”, but again a priori they might
just keep getting “smaller and smaller”. The key to showing that this cannot happen is to
rephrase things in terms of ideals: Recall that b|a if and only if aR ⊆ bR and b is a proper factor
of a (i.e. b divides a and is not an associate of a) if and only if aR ⊊ bR, that is, aR is strictly
contained in bR. Thus if R, our PID, contained an element which could be factored into smaller
and smaller factors this would translate this into a nested sequence of ideals each of which
strictly contained the previous ideal. The next Proposition shows that this cannot happen in a
PID.

Proposition 6.6. Let R be a PID and suppose that {In : n ∈ N} is a sequence of ideals such that In ⊆ In+1 .
Then the union I = n≥0 In is an ideal and there exists an N ∈ N such that In = IN = I for all n ≥ N.
S

Proof. Let I = n≥1 In . Given any two elements p, q ∈ I, we may find k, l ∈ N such that p ∈ Ik
S
and q ∈ Il . It follows that for any r ∈ R we have r.p ∈ Ik ⊂ I, and taking n = max{k, l} we see that
r, s ∈ In so that r + s ∈ In ⊂ I. It follows that I is an ideal. Since R is a PID, we have I = ⟨c⟩ for
some c ∈ R. But then there must be some N such that c ∈ IN , and hence I = ⟨c⟩ ⊆ IN ⊆ I, so that
I = IN = In for all n ≥ N as required. □

Remark 6.7. A ring which satisfies the condition that any nested ascending chain of ideals
stabilizes is called a Noetherian ring. The condition is a very important “finiteness” condition
in ring theory. (Note that the proof that the chain of ideals stabilizes generalises readily if you
just know every ideal is generated by finitely many elements, rather than a single element.)
Polynomial rings in any number of indeterminates have this property by a theorem know as
35
Or if you prefer, including units, we can start with M(a) = 0, so that a is already a unit.

25
Hilbert’s Basis Theorem, which you can learn more about in the Commutative Algebra course
in Part B.
Theorem 6.8. Let R be a PID. Then R is a UFD.
Proof. As discussed above, it follows from the fact that irreducibles are prime in a PID and
Lemma 6.5 that we need only show any element can be factored as a product of irreducible
elements. Thus suppose for the sake of a contradiction that there is some a = a1 ∈ R which is
not a product of irreducible elements. Clearly a cannot be irreducible, so we may write it as
a = b.c where neither b nor c is a unit. If both b and c can be written as a product of prime
elements, then multiplying these expressions together we see that a is also, hence at least one
of b or c cannot be written as a product of prime elements. Pick one, and denote it a2 . Note that
if we set Ik = ⟨ak ⟩ (for k = 1, 2) then I1 ⊊ I2 . As before a2 cannot be irreducible, so we may find
an a3 such that I2 = ⟨a2 ⟩ ⊊ ⟨a3 ⟩ = I3 . Continuing in this fashion we get a nested sequence of
ideals Ik each strictly bigger than the previous one. But by Proposition 6.6 this cannot happen
if R is a PID, thus no such a exists.
□

Remark 6.9. (Non-examinable). The annoying “up to units” qualification for prime factorisation
in a PID vanishes if you are willing to live with ideals rather than elements: in a PID any
proper ideal I can be written as a product of nonzero prime ideals I = P1 P2 . . . Pk where the
prime ideals occuring in this factorisation are unique up to reordering. Indeed this is just the
statement that two elements of an integral domain are associates if and only if they generate
the same principal ideal. However, if you do Algebraic Number Theory next year you’ll see
this idea extended to rings where unique factorization of elements fails (in particular the rings
are not PIDs!) but where nevertheless unique factorization of ideals continues to hold.
Remark 6.10. In special cases the proof that any element is a product of irreducibles can be
simplified: more precisely, suppose that R is an Euclidean domain with a norm N which sat-
isfies the condition that N(a) ≤ N(a.b) for all a, b ∈ R\{0}. We will call36 such a norm weakly
multiplicative. (This holds for example if the norm satisfies something like N(a.b) = N(a).N(b)
or N(a.b) = N(a) + N(b).) In this case we can replace the use of Proposition 6.6 with a more con-
crete inductive argument. In order to make the induction work however, we will need to know
that when we factorise an element as a product of two proper factors (i.e. so neither factor is a
unit) then the norms of the factors are strictly smaller than the norm of the element. Of course
if have an explicit description of the norm (as we do say for k[t] or Z) this may be easy to check
directly, but it is in fact a consequence of the weakly multiplicative property. More precisely
we have:
Claim: Let R be an ED with a weakly multiplicative norm. If a, b ∈ R\{0} satisfy b|a and N(a) =
N(b) then a and b are associates.
Proof : To prove the claim, suppose that N(a) = N(b) and a = b.c. We must show that c is a unit.
By the division algorithm we have b = q.a + r where r = 0 or N(r) < N(a) = N(b). Substituting
a = b.c and rearranging we get b(1 − q.c) = r, and hence if r , 0 then N(r) = N(b.(1 − q.c)) ≥
N(b) = N(a) which is a contradiction. Thus r = 0 and so since b , 0, 1 − q.c = 0 and so c is a unit
as required.
We now show how, in any Euclidean Domain R with a weakly multiplicative norm a nonunit
a ∈ R\{0} is a product of irreducibles using induction on N(a) the norm. Note that N(1) ≤
N(1.a) = N(a) for all a ∈ R\{0}, so that the minimum value of N is N(1). But by what we have
just done, if N(a) = N(1) then a is a unit (since 1 divides any a ∈ R). If N(a) > N(1) then either
a is an irreducible element, in which case we are done, or a = b.c, where neither b nor c is a
unit. But then by the claim we must have N(b), N(c) < N(a), and hence by induction they can
36
I don’t know if there is a standard name for this property – “multiplicative” would suggest something like
N(a.b) = N(a).N(b). “Submultiplicative” might be another reasonable term, but it sounds pretty awful.

26
be expressed as a product of irreducibles and so multiplying these expressions together we see
so can a. It follows every a ∈ R\{0} is unit or a product of irreducibles as required.

A ring may be a UFD without being a PID: in fact we will now show that Z[t] is a UFD,
even though it is not a PID.
The idea is to use the fact that, since Z and Q[t] are PIDs, unique factorisation holds in each.
We can then show Z[t] is a UFD by studying the inclusion of Z[t] into Q[t]. The next definition
and Lemma are the key to our understanding of factorisation in Z[t].

Definition 6.11. If f ∈ Z[t] then define the content c( f ) of f to be the highest common factor of
the coeficients of f . That is, if f = ni=0 ai ti then we set c( f ) = h.c.f.{a0 , a1 , . . . , an }. Note that in a
P
general integral domain the highest common factor is only defined up to units, but in the case
of Z if we insist c( f ) > 0 then it is unique (since the units in Z are just {±1}). In particular, given
f ∈ Z[t] nonzero, c( f ) is the unique positive integer such that f = c( f ). f1 where f1 has content
1, that is, its coefficients generate the whole ring Z.

Lemma 6.12. (Gauss). Let f, g ∈ Z[t]. Then c( f.g) = c( f ).c(g).

Proof. Suppose first f, g ∈ Z[t] have c( f ) = c(g) = 1. Then let p ∈ N be a prime. We have for each
such prime a homomorphism Z[t] → F p [t] given by ϕ p ( ni=0 ai ti ) = ni=0 āi ti , where āi denotes
P P
ai + pZ ∈ F p . It is immediate that ker(ϕ p ) = pZ[t], so that we see p|c( f ) if and only if ϕ p ( f ) = 0.
But since F p is a field, F p [t] is an integral domain, and so as ϕ p is a homomorphism we see that

p|c( f.g) ⇐⇒ ϕ p ( f.g) = 0 ⇐⇒ ϕ p ( f ).ϕ p (g) = 0

⇐⇒ ϕ p ( f ) = 0 or ϕ p (g) = 0 ⇐⇒ p|c( f ) or p|c(g),

whence it is clear that c( f.g) = 1 if c( f ) = c(g) = 1.

Now let f, g ∈ Z[t], and write f = a. f ′ ,g = b.g′ where f ′ , g′ inZ[t] have c( f ′ ) = c(g′ ) = 1,
(so that c( f ) = a, c(g) = b). Then clearly f.g = (a.b).( f ′ g′ ) and since c( f ′ g′ ) = 1 it follows that
c( f.g) = c( f ).c(g) as required.
□

Alternative proof. If you found the above proof of the fact that c( f.g) = 1 if c( f ) = c(g) = 1
a bit too slick, then a more explicit version of essentially the same argument goes as follows:
Let f = ni=0 ai ti and37 g = ni=0 bi ti , and write f.g = 2n k
P P P
k=0 ck t . Suppose that d divides all the
coefficients of f.g and d is not a unit. Since c( f ) = 1, there must be a smallest k such that d does
not divide ak and similarly since c(g) = 1 there is a smallest l such that d does not divide bl .
Consider X
ck+l = ai b j ,
i+ j=k+l

Now d divides every term on the right-hand side except for ak bl , since every other term has
one of i < k or j < l, but then d does not divide the sum, contradicting the assumption that d
divides ck+l . Thus we have a contradiction and thus c( f.g) = 1 as required.
We can now extend the definition of content to arbitrary nonzero elements of Q[t].

Lemma 6.13. Suppose f ∈ Q[t] is nonzero. Then there is an unique α ∈ Q>0 such that f = α f ′ where
f ′ ∈ Z[t] and c( f ′ ) = 1. We write c( f ) = α. Moreover, if f, g ∈ Q[t] then c( f.g) = c( f ).c(g).

Proof. Let f = ni=0 ai ti where ai = bi /ci for bi , ci ∈ Z and h.c. f {bi , ci } = 1 for all i, 0 ≤ i ≤ n.
P
Pick d ∈ Z>0 such that dai ∈ Z for all i, (1 ≤ i ≤ n) so that d f ∈ Z[t] (for example you can take
d = l.c.m.{ci : 0 ≤ i ≤ n} or ni=0 ci ). Set c( f ) = c(d. f )/d (where the righthand side is already
Q
defined because d. f ∈ Z[t]). Then f = c( f ). f ′ where f ′ = (d. f )/c(d. f ) is clearly a polynomial
37
Note that so long as we do not assume that both ba and an are nonzero we may take the same upper limit in the
sums.

27
in Z[t] with content one. To check c( f ) is well-defined we must show that if f = α1 . f1 = α2 . f2
where f1 , f2 have content one and α1 , α2 ∈ Q>0 then α1 = α2 . But writing αi = mi /ni for positive
integers mi , ni , we find n2 .(m1 . f1 ) = n1 .(m2 f2 ) ∈ Z[t]. Taking content and using the trivial fact
that if n ∈ Z\{0} ⊂ Z[t] then c(d) = |d|, we see that

c(n2 .(m1 f1 )) = c(n2 ).c(m1 f1 ) = n2 .m1 , c(n1 .(m2 f2 )) = c(n1 ).c(m2 f2 ) = n1 m2 .

Equating it follows α1 = m1 /n1 = m2 /n2 = α2 as required.

It is now easy to check multiplicativity: if f, g ∈ Q[t]/\{0} then if d1 , d2 ∈ Z are such that
d1 . f, d2 g ∈ Z[t] then clearly d1 d2 .( f.g) ∈ Z[t] so that
c(d1 d2 f.g) c((d1 f ).(d2 g)) c(d1 f ) c(d2 .g)
c( f.g) = = = . = c( f ).c(g),
d1 d2 d1 .d2 d1 d2
as required. □

Remark 6.14. Note in particular it follows immediately from the previous Lemma that if f ∈
Q[t]\{0} then f ∈ Z[t] if and only if c( f ) ∈ Z.
We now relate factorization in Z[t] and Q[t], and obtain a description of some prime ele-
ments in Z[t].
Lemma 6.15. 1. Suppose that f ∈ Z[t] ⊂ Q[t] is nonzero, and that f = g.h where g, h ∈ Q[t]. Then
there exist α ∈ Q such that (α.g), (α−1 .h) ∈ Z[t]. Thus f = (α.g)(α−1 h) is a factorisation of f in
Z[t].

2. Suppose that f ∈ Q[t] is irreducible and c( f ) = 1. Then f is a prime element of Z[t].

3. Let p ∈ Z be a prime number. Then p is a prime element in Z[t].

Proof. For the first part, by Lemma 6.13 we may write g = c(g).g1 and h = c(h).h1 where g1 , h1 ∈
Z[t] have content 1. Then c( f ) = c(g)c(h) so that as f ∈ Z[t] we have c(g).c(h) ∈ Z. Setting α = c(h)
we see that f = (α.g).(α−1 .h) where α.g = (c(g).c(h)).g1 and α−1 h = h1 both lie in Z[t] as required.
For the second part, first note that if f ∈ Q[t] has c( f ) = 1 then by definition f must lie in
Z[t] (and has content 1). To see that such an f is prime, we need to show that if g, h ∈ Z[t] and
f |g.h in Z[t] then f |g or f |h in Z[t]. Now if f |g.h in Z[t], certainly it does so in Q[t]. Since Q[t] is
a PID, irreducibles are prime and so either f |g or f |h in Q[t]. Suppose that f |g (the argument
being identical for h). Then we have g = f.k for some k ∈ Q[t]. Now by Lemma 6.13 we may
write k = c(k).k′ where k′ ∈ Z[t], and moreover by the same lemma, c(g) = c( f ).c(k) = c(k) since
c( f ) = 1. But g ∈ Z[t], hence c(g) = c(k) ∈ Z and hence k ∈ Z[t] so that f divides g in Z[t] as
required.
For the final part, we have already seen that the homomorphism ϕ p : Z[t] → F p [t] has kernel
pZ[t], and so since F p [t] is an integral domain, the ideal pZ[t] is prime, that is, p is a prime
element of Z[t].
□

Theorem 6.16. The ring Z[t] is a UFD.

Proof. Since Z[t] is an integral domain (as Z is), by Lemma 6.5 it is enough to show that any
element of Z[t] is a product of primes. Let f ∈ Z[t]. We may write f = a. f ′ where c( f ′ ) = 1,
and since Z is a UFD we may factorise a into a product of prime elements of Z which we have
just seen are prime in Z[t]. Thus we may assume c( f ) = 1. But then viewing f as an element
of Q[t] we can write it as a product of prime elements in Q[t], say f = p1 p2 . . . pk . Now using
Lemma 6.13, each pi can be written as ai qi where ai ∈ Q and qi ∈ Z[t] and c(qi ) = 1. But then by
the Lemma 6.15, qi is prime in Z[t], and f = (a1 . . . ak )q1 . . . qk . Comparing contents we see that
(a1 . . . ak ) = 1 and so we are done. □

28
Remark 6.17. It is easy to see from this that in fact all primes in Z[t] are either primes in Z or
primes (equivalently irreducibles) in Q[t] which have content 1.

Remark 6.18. In fact one can show directly (see the problem set) that if R is a UFD then highest
common factors exist (that is, given elements a1 , . . . , an ∈ R there is an element d such that d|ai
for all i, (1 ≤ i ≤ n) and if c|ai for all i also, then c|d). It follows that if R is a UFD then
we can define the content of a nonzero element of R[t] to be the highest common factor of
its coefficients just as we did for Z[t]. This observation implies the following theorem, whose
proof is not examinable.

Theorem 6.19. If R is a UFD then the polynomial ring R[t] is also a UFD. More generally, if R is a
UFD then R[t1 , . . . , tn ] the ring of polynomials in n variables with coefficients in R, is a UFD.

Proof. (Nonexaminable.) The proof follows the same strategy as for Z[t]: if f ∈ Rt] then as saw
in the previous remark, the content of f makes sense (though now we cannot use positivity to
make it unique, so it is only defined up to units). Let F be the field of fractions of R. Then F[t]
is a PID and hence a UFD, and the content lets us understand the relation of factorization in
R[t] to factorization in F[t] using the analogue of Gauss’s Lemma. You can then check that the
primes in R[t] are then the primes in R and the irreducibles in F[t] with content 1, and the fact
that any element of R[t] has a prime factorization then follows exactly as for Z[t].
For the final part, since R[t1 , . . . , tn ] = S [tn ] where S = R[t1 , . . . , tn−1 ] the result follows from
the first part and induction on n. □

The previous theorem shows that, for example, Q[x, y] is a UFD. It is not hard to see that
neither Z[t] nor Q[x, y] are PIDs38 , so the class of rings which are UFDs is strictly larger than
the class of PIDs. In fact not every PID is a Euclidean domain either, so there are strict contain-
ments: EDs ⊊ PIDs ⊊ UFDs. Finding a PID which is not a Euclidean domain is a bit subtle, so
we wont do it here, but see the Appendix.

6.1 Irreducible polynomials.

In this section we develop some techniques for deciding when a polynomial f ∈ Q[t] is irre-
ducible. By what we have done above, if f ∈ Q[t] is irreducible, we may write f = c( f ).g where
g ∈ Z[t] has content 1 and is a prime in Z[t]. Since f and g are associates in Q[t] it follows that to
understand irreducible elements in Q[t] it is enough to understand the prime elements in Z[t]
of positive degree (or equivalently, the irreducibles f ∈ Q[t] with c( f ) = 1.)
This is useful for the following reason: Recall that for any prime p ∈ Z we have the homo-
morphism39 ϕ p : Z[t] → F p [t]. This allows us to transport questions about factorisation in Z[t]
to questions about factorisation in F p [t]: If f ∈ Z[t] with c( f ) = 1, and f = g.h, then g and h have
content 1. In particular, if f = g.h is a factorization of f with neither of g, h a unit, then both
g and h have positive degree. If we pick p a prime not dividing the leading coefficient of f ,
then deg(ϕ p ( f )) = deg( f ) and so40 we must have deg(ϕ p (g)) = deg(g) and deg(ϕ p (h)) = deg(h), and
hence we obtain a proper factorization of ϕ p ( f ). It follows that if ϕ p ( f ) is irreducible in F p [t],
then f must be irreducible in Z[t]. Since the rings F p [t] are “smaller” than either Z[t] or Q[t]
this can give us ways of testing irreducibility (indeed notice since the coefficient field is finite,
it is in principle a finite check to see if a given element in F p [t] is irreducible).

Example 6.20. Suppose that f = t3 −349t+19 ∈ Z[t]. If f is reducible in Q[t], it is reducible in Z[t]
and hence its image under ϕ p in F p [t] will be reducible. But since f has degree 3 it follows it is
38
In fact for Z[t] this follows from Lemma 6.15 – do you see why?
39
Note that there is no homomorphism from Q[t] to F p [t] for any prime p. This is why we have to pass through
Z[t].
40
Because the degree of a product of polynomials is the sum of the degrees whenever the coefficient ring is an
integral domain.

29
reducible if and only if it has a degree 1 factor, and similarly for its image in F p [t], which would
therefore mean it has a root in F p . But taking p = 2 we see that ϕ2 ( f ) = f¯ = t3 + t + 1 ∈ F2 [t]
and so it is easy to check that f¯(0) = f¯(1) = 1 ∈ F2 , so f¯ does not have a root, and hence f
must be irreducible. Note on the other hand t2 + 1 is irreducible in Z[t] but in F2 [t] we have
t2 + 1 = (t + 1)2 , so ϕ p ( f ) can be reducible even when f is irreducible.

Lemma 6.21. (Eisenstein’s criterion.) Suppose that f ∈ Z[t] has c( f ) = 1, and f = an tn + an−1 tn−1 +
. . . a1 t + a0 . Then if there is a prime p ∈ Z such that p|ai for all i, 0 ≤ i ≤ n − 1 but p does not divide an
and p2 does not divide a0 then f is irreducible in Z[t] and Q[t].

Proof. Since c( f ) = 1, we have already seen that irreducibility in Z[t] and Q[t] are equivalent.
Let us write ϕ p : Z[t] → F p [t] for the quotient map. Suppose that f = g.h was a factorisation
of f in Z[t] where say 0 < deg(g) = k < n. Then we have ϕ p ( f ) = ϕ p (g).ϕ p (h). By assumption
ϕ p ( f ) = ān tn (where for m ∈ Z we write m̄ for m + pZ, the image of m in F p ). Since F p [t] is a UFD,
t is irreducible, and ān is a unit, it follows that up to units we must have ϕ p (g) = tk , ϕ p (h) = tn−k .
But then (since k and n−k are both positive) the constant terms of both g and h must be divisible
by p, and hence a0 must be divisible by p2 , contradicting our assumption. □

Example 6.22. This gives an easy way to see that 21/3 < Q: if it was t3 − 2 would be reducible,
but we see √this√is not the case by applying Eisenstein’s criterion with p = 2. (It also gives a
proof that 2, 3 are irrational).

7 Modules: Definition and examples.

As usual, all rings are assumed to be commutative unless the contrary is explicitly stated.
In this section we begin the study of “linear algebra over rings”. Recall a vector space is
just an abelian group with an action of a field of “scalars” obeying some standard rules. The
definition of a module is exactly the same, except now we allow our scalars to belong to an
arbitrary ring, rather than insisting they belong to a field. Formally, we say the following:

Definition 7.1. Let R be a ring with identity 1R . A module over R is an abelian group (M, +)
together with a multiplication action a : R × M → M of R on M written (r, m) 7→ r.m which
satisfies:

1. 1R .m = m, for all m ∈ M;

2. (r1 .r2 ).m = r1 .(r2 .m), for all r1 , r2 ∈ R, m ∈ M

3. (r1 + r2 ).m = r1 .m + r2 .m for all r1 , r2 ∈ R and m ∈ M;

4. r.(m1 + m2 ) = r.m1 + r.m2 for all r ∈ R and m1 , m2 ∈ M.

Remark 7.2. Just as with vector spaces, we write the addition in the abelian group M and the
addition in the ring R as the same symbol “+”, and similarly the multiplication action of R on
M is written in the same way as the multiplication in the ring R, since the axioms ensure that
there is no ambiguity in doing so.

Remark 7.3. Note that the definition makes perfectly good sense for a noncommutative ring
(when it would normally be described as a left module since the action of the ring is on the left).
Next year’s course on Representation Theory will study certain noncommutative rings called
group algebras, and modules over them. In this course we will focus on modules over integral
domains and all our main results will be for modules over a PID, though even then, in some
cases we will only give proofs for the case where our ring is a Euclidean domain.

30
Lemma 7.4. Let M be an abelian group and R a ring.
i) The set End(M) = Hom(M, M) of group homomorphisms from M to itself is naturally a (in
general noncommutative) ring where addition is give pointwise and multiplication is given by
composition.

ii) Giving M the structure of an R-module, that is an action R × M → M, is equivalent to giving a

ring homomorphism ϕ : R → End(M).
Proof. For the first part, since M is abelian, if f1 , f2 ∈ End(M) then f1 + f2 is again a group ho-
momorphism, and clearly f1 + f2 = f2 + f1 so addition in End(M) is commutative. Composition
of functions gives End(M) a multiplication which it is easy to check distributes over addition,
thus End(M) is a (not necessarily commutative) ring.
For the second part, if the action map a : R × M is denoted by (r, m) 7→ r.m as usual, then the
equation ϕ(r)(m) = r.m defines ϕ in terms of a and conversely. It is routine to check the equiva-
lences – property (4) of an action shows that m 7→ r.m is a homomorphism of the abelian group
M, property (1) corresponds to requiring ϕ(1R ) = 1End(M) , property (2) to the compatibility of ϕ
with multiplication, and property (3) to the compatibility of ϕ with addition. □

Remark 7.5. You should compare this Lemma with the corresponding result from group ac-
tions: if G is a group and X is a set, then giving an action of G on X is the same as giving a group
homomorphism from G to the group of permutations of the set X (i.e. the group of bijections
from X to itself). You could define vector spaces this way, but in Prelims we tell you what a
vector space is before we tell you what a group action is (or indeed what a ring is!)
Example 7.6. Let’s give a few examples:
1. As mentioned above, if R is a field, the definition is exactly that of a vector space over R,
so modules over a field are just vector spaces over that field.

2. At the other end of the spectrum in a sense, if A is an abelian group, then it has a natural
structure of Z-module: if n is a positive integer, then set n.a = a+a+. . .+a (n times) and if n
is a negative integer, set n.a = −(a+a+. . .+a) (where this time we add a to itself −n times).
It’s easy to check this makes A a Z-module, and moreover, the conditions (1), (2), (3), (4)
in fact force this definition on us, so that this Z-module structure is unique41 . Thus we
see that Z-modules are just abelian groups.

3. Suppose that R is a ring. Then R is a module over itself in the obvious way.

4. If R is a ring and I is an ideal in R, then it follows directly from the definitions that I is an
R-module.

5. Again if I is an ideal in R then R/I is naturally an R-module where the multiplication

action is given via the quotient homomorphism q : R → R/I, that is, if m ∈ R/I and r ∈ R
we set r.m = q(r).m (the multiplication on the righthand side being inside the ring R/I).
Indeed the properties (1), (2) (3) and (4) all follow immediately from the fact that q is a
ring homomorphism.

6. Generalising the previous example somewhat, if ϕ : R → S is a homomorphism of rings,

and M is an S -module, then we can give M the structure of an R-module by setting r.m =
ϕ(r).m (where the action on the right-hand side comes from the S -module structure. Thus
for example any if I is an ideal of R then any R/I-module automatically has the structure
of an R-module via the quotient map q : R → R/I.
41
Writing down all the details of a proof of this is very similar to the exercise in the problem sheets in which
you showed that given any ring R there is a unique homomorphism from Z to R. The reason for this is because the
module structure corresponds to the unique ring homomorphism Z → End(M).

31
 7. Generalising the example of R being an R-module over itself in a slightly different way,
given our ring R and a positive integer n, we may consider the module Rn = {(r1 , r2 , . . . , rn ) :
ri ∈ R} of n-tuples of elements of R (written as row vectors or column vectors – different
books prefer different conventions), where the addition and the multiplication by scalars
is done componentwise. (This is exactly the way we define the vector space Rn for the
field R). Such a module is an example of a free module over R.

8. To give a more substantial example, suppose that V is a vector space over a field k and
ϕ : V → V is a linear map. Then we can make V into a k[t]-module by setting p(t).v =
p(ϕ)(v) for any v ∈ V and p(t) ∈ k[t] (that is just evaluate the polynomial p on the linear
map ϕ). Indeed a homomorphism from k[t] to Endk (V) is uniquely determined by its
restriction to the scalars k and the image of t. Here we define ϕ by the conditions that it
sends the complex number λ ∈ k ⊆ k[t] to λ.idV , and t to ϕ. The fact that the assignment
f.v = ϕ( f )(v) for v ∈ V, f ∈ k makes V into a k[t]-module follows directly from the fact
that ϕ is a homomorphism. Conversely, if we are given a k[t]-module M, we can view
it as a k-vector space where the multiplication by scalars is given to us by viewing the
elements of k as degree zero polynomials. The action of multiplication by t is then a k-
linear map from M to itself. Thus k[t]-modules are just k-vector spaces equipped with an
endomorphism.

7.1 Submodules, generation and linear independence.

Definition 7.7. If M is an R-module, a subset N ⊆ M is called a submodule42 if it is an abelian
subgroup of M and whenever r ∈ R and n ∈ N then r.n ∈ N.
T
If {Ni : i ∈ I} is a collection of submodules then their intersection i∈I Ni is also a submodule.
This allows us to define (just as we did for ideals, subrings, subfields etc.) for a set X ⊂ M the
submodule generated by X,

Definition 7.8. If X is any subset of an R-module M then the submodule generated or spanned
by X is defined to be: \
⟨X⟩R = ⟨X⟩ = N,
N⊇X

where N runs over the submodules of M which contain X. (Provided it is clear from the context
which ring M is being viewed as a module for, we will omit the subscript “R”.) Explicitly, it is
the subset R.X = { ki=1 ri xi : ri ∈ R, xi ∈ X} (where this is by convention understood to be {0} if
P
X = ∅). The proof is exactly the same43 as the proof for ideals in a ring.

If N1 , N2 are submodules then the submodule they generate is their sum N1 + N2 = {m + n :

m ∈ N1 , n ∈ N2 }. To prove this one first checks that the righthand side is indeed a submodule
and then that any submodule containing N1 and N2 must contain all the elements of N1 + N2
(and these two steps both follow directly from the definitions). Note that this generalises the
fact which we have already seen that the ideal generated by the union of two ideals I ∪ J is just
their sum I + J.

Definition 7.9. If M is a module over R, we say a set S ⊆ M is linearly independent if whenever we

have an equation r1 s1 + r2 s2 + . . . rk sk = 0 for ri ∈ R, si ∈ S (1 ≤ i ≤ k) we have r1 = r2 = . . . rk = 0.
We say that a set S is a basis for a module M if and only if it is linearly independent and it
spans M. Any module which has a basis is called a free module. Finally we say that a module
is finitely generated if it is generated by some finite subset.
42
Note the definitions in this subsection are exactly the same as for the case of a vector space.
43
As should come as no surprise given example (3) above: a subset of R viewed as an R-module is a submodule
if and only if it is an ideal.

32
8 Quotient modules and the isomorphism theorems.

Just as for vector spaces, given a module together with a submodule there is a natural notion
of a quotient module. (If you’ve understood quotients of rings and quotients of vectors space,
everything here should look very familiar, as the constructions mimics those cases, in fact they
are word for word the same as for quotient vector spaces).

Definition 8.1. If N is a submodule of M, then in particular it is a subgroup of an abelian group,

so we can form the quotient M/N. The condition that N is a submodule then is precisely what
is needed for the multiplication on M to induce a module structure on M/N: If r ∈ R and
m + N ∈ M/N then define r.(m + N) = r.m + N. This is well defined because if m1 + N = m2 + N
we have m1 − m2 ∈ N, and so r.(m1 − m2 ) ∈ N, whence r.m1 + N = r.m2 + N. The module M/N is
called the quotient module of M by N.

Definition 8.2. There is also a natural analogue of linear maps for modules: if M1 , M2 are R-
modules, we say that ϕ : M1 → M2 is an R-module homomorphism (or just homomorphism) if:

1. ϕ(m1 + m2 ) = ϕ(m1 ) + ϕ(m2 ), for all m1 , m2 ∈ M1 ,

2. ϕ(r.m) = r.ϕ(m), for all r ∈ R, m ∈ M1 ,

that is, ϕ respects the addition and multiplication by ring elements. An isomorphism of R-
modules is a homomorphism which is a bijection (and you can check, just as for groups, that
this implies the inverse map of sets is also a homomorphism of modules). Just as the kernel
and image of a linear map between vector spaces are subspaces, it is easy to see that ker(ϕ) =
{m ∈ M1 : ϕ(m) = 0} and im(ϕ) = {ϕ(m) : m ∈ M1 } are submodules of M1 and M2 respectively.

Remark 8.3. It is easy to check that if M1 , M2 are R-modules then ϕ : M1 → M2 is an R-module

homomorphism if and only if ϕ(v + tw) = ϕ(v) + tϕ(w) for all v, w ∈ M1 and t ∈ R.

Example 8.4. When R is a field, module homomorphisms are exactly linear maps. When R =
Z, a Z-module homomorphism is just a homomorphism of the abelian groups. As another
important example, it is easy to see that if M is an R-module and N is a submodule of M then
the definition of the module structure on M/N ensures precisely that the map q : M → M/N
given by q(m) = m + N is a (surjective) module homomorphism.

Lemma 8.5. (Submodule correspondence:) Let M be an R-module and N a submodule. Let q : M →

M/N be the quotient map. If S is a submodule of M then q(S ) is a submodule of M/N, while if T
is a submodule of M/N then q−1 (T ) is a submodule of M. Moreover the map T 7→ q−1 (T ) gives an
injective map from submodules of M/N to the submodules of M which contain N, thus submodules of
M/N correspond bijectively to submodules of M which contain N.

Proof. The proof works precisely the same way as the proof of the correspondence between
ideals given by a surjective ring homomorphism ϕ : R → S . Indeed that result is a special
case of this Lemma, since ϕ makes S into an R-module, and ideals in S are precisely the R-
submodules of S since ϕ is surjective.
To check that q(S ) and q−1 (T ) are submodules of N and M respectively follows directly
from the definitions. We give the argument for q−1 (T ), the argument for q(S ) follows exactly
the same pattern. If m1 , m2 ∈ q−1 (T ) then q(m1 ), q(m2 ) ∈ T and it follows since T is a submodule
that q(m1 ) + q(m2 ) = q(m1 + m2 ) ∈ T which says precisely that m1 + m2 ∈ q−1 (T ). Similarly if r ∈ R
then q(r.m1 ) = r.q(m1 ) ∈ T since q(m1 ) ∈ T and T is a submodule, so that r.m1 ∈ q−1 (T ). Thus
q−1 (T ) is a submodule of M as required.
Now if T is any subset of M/N we have q(q−1 (T )) = T simply because q is surjective. Since
we have just checked q−1 (T ) is always a submodule in M, this immediately implies that the

33
map S 7→ q(S ) is a surjective map from submodules in M to submodules in M/N and that
T 7→ q−1 (T ) is an injective map44 , and moreover since q(N) = {0} ⊆ T for any submodule T of
M/N we have N ⊆ q−1 (T ) so that the image of the map T 7→ q−1 (T ) consists of submodules of M
which contain N. Hence it only remains to check that the submodules of M of the form q−1 (T )
are precisely these submodules. To see this suppose that S is an arbitrary submodule of M,
and consider q−1 (q(S )). By definiton this is

q−1 (q(S )) = {m ∈ M : q(m) ∈ q(S )}

= {m ∈ M : ∃s ∈ S such that m + N = s + N}
= {m ∈ M : ∃s ∈ S such that m ∈ s + N}
= S + N.

But if S contains N then we have S + N = S and hence q−1 (q(S )) = S and so any submodule S
which contains N is indeed the preimage of a submodule of M/N as required. □

Remark 8.6. If N ⊆ M is a submodule and q : M → M/N is the quotient map, for a submodule
Q of M containing N we will usually write Q/N for the submodule q(Q) of M/N.

Theorem 8.7. (Universal property of quotients.) Suppose that ϕ : M → N is a homomorphism of

R-modules, and S is a submodule of M with S ⊆ ker(ϕ) and let q : M → M/S be the quotient homomor-
phism. Then there is a unique homomorphism ϕ̄ : M/S → N such that ϕ = ϕ̄ ◦ q, that is, such that the
following diagram commutes:
ϕ
MD /N
DD z=
DD zzz
D
q DD zz
" zz ϕ̄
M/S
Moreover ker(ϕ̄) is the submodule ker(ϕ)/S = {m + S : m ∈ ker(ϕ)}.

Proof. The proof exactly mirrors the case for rings. Since q is surjective, the formula ϕ̄(q(m)) =
ϕ(m) uniquely determines the values of ϕ̄, so that ϕ̄ is unique if it exists. But if m − m′ ∈ S then
since S ⊆ ker(ϕ) it follows that 0 = ϕ(m − m′ ) = ϕ(m) − ϕ(m′ ) and hence ϕ is constant on the S -
cosets, and therefore induces a map ϕ̄(m + S ) = ϕ(m). The fact that ϕ̄ is a homomorphism then
follows directly from the definition of the module structure on the quotient M/S , and clearly
ϕ = ϕ̄ ◦ q by definition. To see what the kernel of ϕ̄ is, note that ϕ̄(m + S ) = ϕ(m) = 0 if and only
if m ∈ ker(ϕ), and hence m + S ∈ ker(ϕ)/S as required. □

Corollary 8.8. Let M be an R-module. We have the following isomorphisms.

i) (First isomorphism theorem.) If ϕ : M → N is a homomorphism then ϕ induces an isomorphism

ϕ̄ : M/ker(ϕ) → im(ϕ).

ii) (Second isomorphism theorem.) If M is an R-module and N1 , N2 are submodules of M then

(N1 + N2 )/N2  N1 /N1 ∩ N2 ,

iii) (Third isomorphism theorem.) Suppose that N1 ⊆ N2 are submodules of M. Then we have

(M/N1 )/(N2 /N1 )  M/N2 .

44
Again this is just set theory: if f : X → Y and g : Y → X are functions such that g ◦ f = idX then f is injective and
g is surjective.

34
Proof. The proofs again are exactly the same as for rings. For the first isomorphism theorem,
apply the universal property to S = ker(ϕ). Since in this case ker(ϕ̄) = ker(ϕ)/ker(ϕ) = 0 it fol-
lows ϕ̄ is injective and hence induces an isomorphism onto its image which from the equation
ϕ̄ ◦ q = ϕ must be exactly im(ϕ).
For the second isomorphism theorem, let q : M → M/N2 be the quotient map. It restricts
to a homomorphism p from N1 to M/N2 , whose image is clearly (N1 + N2 )/N2 , so by the first
isomorphism theorem it is enough to check that the kernel of p is N1 ∩ N2 . But this is clear: if
n ∈ N1 has p(n) = 0 then n + N2 = 0 + N2 so that m ∈ N2 , and so n ∈ N1 ∩ N2 .
For the third isomorphism theorem, let qi : M → M/Ni for i = 1, 2. By the universal property
for q2 with S = N1 we see that there is a homomorphism q̄2 : M/N1 → M/N2 induced by the
map q2 : M → M/N2 , with kernel ker(q2 )/N1 = N2 /N1 and q̄2 ◦ q1 = q2 . Thus q̄2 is surjective
(since q2 is) and hence the result follows by the first isomorphism theorem. □

9 Free, torsion and torsion-free modules.

All rings R in this section are integral domains unless otherwise stated.
The fact that the nonzero elements of a ring do not have to be invertible means that modules
behave less uniformly than vector spaces do. For example, you know that any vector space has
a basis, and hence in the above terminology it is free. However, over Z there are many modules
which are not free: indeed a finite abelian group is certainly a Z-module, but it cannot be free
since a free module must contain infinitely many elements (if an element m is part of a basis,
it is easy to check that the elements n.m must all be distinct for n ∈ Z). In fact if M is a finite
abelian group, every element is of finite order by Lagrange’s theorem, which means that for
every element m ∈ M there is an integer n ∈ N such that n.m = 0. This is one important way in
which a module over a general ring can be different from the case of vector spaces: we may have
a nonzero scalar r and a nonzero element m of our module M whose product r.m is nevertheless
equal to zero. This is similar to the fact that a general ring may contain zero-divisors.

Definition 9.1. Let M be an R-module and suppose that m ∈ M. Then the annihilator of m,
denoted AnnR (m) is {r ∈ R : r.m = 0}. A direct check shows that AnnR (m) is an ideal in R. When
AnnR (m) is nonzero we say that m ∈ M is a torsion element.
We say that a module M is torsion if every m ∈ M is a torsion element. On the other hand,
if a module M has no nonzero torsion elements we say that M is torsion-free. Note that a ring is
an integral domain if and only if it is torsion-free as a module over itself, i.e. torsion elements
in the R-module R itself are exactly the zero-divisors in R.

Remark 9.2. If M is an R-module, and m ∈ M then the submodule R.m of M generated by m is

isomorphic as an R-module to R/AnnR (m). Indeed the map r 7→ r.m defines an R-module homo-
morphism from R to M whose image is exactly R.m. Since the kernel of the map is evidently
AnnR (m) the isomorphism follows from the first isomorphism theorem. (Note this also shows
AnnR (m) is an ideal, though this is also completely straight-forward to see directly.)

Definition 9.3. A module which is generated by a single element is known as a cyclic module.
It follows from what we have just said that any cyclic module is isomorphic to a module of the
form R/I where I is an ideal of R (corresponding to the annihilator of a generator of the cyclic
module).

Recall from above that we say a module M is free if it has a basis S . The case where S is
finite is the one of most interest to us. Then, just as picking a basis of a vector space gives you
coordinates45 for the vector space, the basis S allows us to write down an isomorphism ϕ : M →
45
That is, if V is an n-dimensional R-vector space, the fact that a choice of basis for V gives you an isomorphism
from V to Rn is just a formal way of saying that picking a basis gives you coordinates for V.

35
Rn where n = |S |. Indeed if S = {s1 , s2 , . . . , sn } and m ∈ M then we may write m = ni=1 ri si for a
P
unique n-tuple (r1 , r2 , . . . , rn ) ∈ Rn , and we set ϕ(m) = (r1 , . . . , rn ). It is straight-forward to check
that ϕ is then an isomorphism of modules.
It is easy to see that when R is an integral domain a free module must be torsion free,
but the converse need not be true in general, as the next example shows. On the other hand,
for principal ideal domains, whose modules will be our main focus, we will shortly see that
torsion-free modules are actually free.

Example 9.4. Let R = C[x, y] be the ring of polynomials in two variables. Then the ideal I = ⟨x, y⟩
is a module for R. It is torsion-free because R is an integral domain (and I is a submodule of R)
but it is a good exercise to see that it is not free. The study of modules over a polynomial ring
with many variables is a basic ingredient in algebraic geometry, and the commutative algebra
course in Part B focuses largely the study of these rings and their quotients.

Recall we also had the notion of a torsion element in a module.

Lemma 9.5. Let M be an R-modules, and let M tor = {m ∈ M : AnnR (m) , {0}} is a submodule of M.
Moreover, the quotient module M/M tor is a torsion-free module.

Proof. Let x, y ∈ M tor . Then there are nonzero s, t ∈ R such that s.x = t.y = 0. But then s.t ∈ R\{0},
since R is an integral domain, and (s.t)(x + y) = t.(s.x) + s.(t.y) = 0, and clearly if r ∈ R then
s.(r.x) = r.(s.x) = 0, so that it follows M tor is a submodule of M as required.
To see the moreover part, suppose that x + M tor is a torsion element in M/M tor . Then there
is a nonzero r ∈ R such that r.(x + M tor ) = 0 + M tor , that is, r.x ∈ M tor . But then by definition
there is an s ∈ R such that s.(r.x) = 0. But then s.r ∈ R is nonzero (since R is an integral domain)
and (s.r).x = 0 so that x ∈ M tor and hence x + M tor = 0 + M tor so that M/M tor is torsion free as
required. □

We will study finitely generated modules for a PID via the study of free modules. The free
modules are, in a sense, the ones whose behaviour is closest to that of vector spaces over a field.
In particular we will be able to understand maps between free modules in terms of matrices
just like we do in linear algebra.
We first show that there is an analogue of the notion of dimension for a free module: Just
as for vector spaces, the size of a basis for a free module is uniquely determined (even though
a free module may have many different bases, just as for vector spaces).

Lemma 9.6. Let M be a finitely generated free R-module. Then the size of a basis for M is uniquely
determined and is known as the rank rk(M) of M.

Proof. Let X = {x1 , . . . , xn } be a basis of M. Pick a maximal ideal46 I in R. Let I M be the sub-
module generated by the set {i.m : i ∈ I, m ∈ M} and let MI = { ni=1 ri xi : ri ∈ I}. Since I is an
P
ideal it is easy to check that MI is a submodule of M. We claim that MI = I M. In fact, since X
generates M, any element of the from r.m where r ∈ I and m ∈ M lies in MI , so that I M ⊆ MI .
On the other hand, certainly I M contains ri xi for any ri ∈ I, i ∈ {1, 2, . . . , n}, and so all sums of
the form ni=1 ri xi , and so MI ⊆ I M and hence MI = I M as required. Notice that in particular
P
this means the submodule MI = I M does not depend on the choice of a basis of X.
Let q : M → M/I M be the quotient map. The quotient module M/I M is module for not just
R, but in fact47 for the quotient field k = R/I, via the action (r + I).q(m) = q(r.m). Indeed we just
need to check this definition does not depend on the choice of r ∈ r + I. But if r − r′ ∈ I then
r.m − r′ .m = (r − r′ ).m ∈ I M and so q(r′ .m) = q(r.m) as claimed.
46
In a PID we know that maximal ideals exist – if R is a field then we take I = 0, otherwise we take aR for a ∈ R
an irreducible element. In a general ring maximal ideals also always exist if you assume the axiom of choice.
47
This is exactly what the submodule I M is cooked up to do – if you like M/I M is the largest quotient of M on
which R/I acts naturally.

36
 We now claim that if X is a basis for M then q(X) is a basis for the k-vector space M/I M.
Note that if we assume the claim then |X| = dimk (M/I M) and the right-hand side is clearly
independent of X (since we have checked that the submodule I M is) so this will finish the
proof of the Lemma. To prove the claim first note that since X generates M and q is surjective it
follows that q(X) generates (i.e. spans) M/I M. Now suppose we have ni=1 ci q(xi ) = 0 ∈ M/I M,
P
where ci ∈ k. Picking any representatives ri ∈ R for the ci ∈ R/I we see that
n
X n
X n
X
0= ci q(xi ) = q(ri xi ) = q( ri xi )
i=1 i=1 i=1

where the second equality follows from the definition of the R/I-action, and the lasts from the
fact that q is an R-module homomorphism. But then it follows that y = ki=1 ri xi ∈ ker(q) = I M.
P
But since I M = MI this means that ri ∈ I for each i, that is ci = 0. It follows X̄ is linearly
independent and hence a k-basis of M/I M as required. □

We will shortly see that any finitely generated module is a quotient of a free module Rn for
some n. It will therefore be important to understand submodules of free modules. If R is a PID
(as we will from now on assume) then the submodules of free modules are particularly well
behaved.

Proposition 9.7. Let M be a finitely generated free module over R a PID, and let X = {e1 , . . . , en } be a
basis. Then if N is a submodule of M, N is also free and has rank at most n elements.

Proof. We prove this by induction on n = |X|. If n = 1, then if ϕ : R → M is the homomorphism

defined by ϕ(r) = re1 , the first isomorphism theorem shows that M  R. Now a submodule of
R is just an ideal, and hence the condition that R is a PID exactly ensures any submodules N of
R is cyclic, say N = Ra. Since R is an integral domain, it follows that {a} is a basis of N unless
a = 0, so that N is free of rank 0 or 1 and the n = 1 case is thus established.
Now suppose that n > 1. Let W = Re1 + Re2 + . . . Ren−1 , and let N1 = N ∩ W. Now N1 is
a submodule of the free module W,and W has rank n − 1, so by induction N1 is free of rank
k ≤ n − 1. Let {v1 , . . . , vk } be a basis of N1 . The second isomorphism theorem shows that N/N1 
(N + W)/W ⊆ M/W. But M/W clearly has basis {en + W}, so that N/N1 is either zero or free of
rank 1. In the former case N = N1 and we are done, while in the latter we may pick vk+1 so that
vk+1 + N1 is a basis of N/N1 . We claim {v1 , . . . , vk+1 } is a basis of N. If m ∈ N then since {vk+1 + N1 } is
a basis of N/N1 , we may find rk+1 ∈ R such that m + N1 = rk+1 vk+1 + N1 . But then m − rk+1 vk+1 ∈ N1 ,
and so since N1 has basis {v1 , . . . , vk } there are r1 , . . . , rk ∈ R such that m − rk+1 vk+1 = ki=1 rk vk ,
P
that is, m = i=1 ri vi . To see {v1 , . . . , vk+1 } is linearly independent, suppose that i=1 si vi = 0.
Pk+1 Pk+1
Then 0 + N1 = k+1 i=1 si vi + N = sk+1 vk+1 + N1 , so that we must have sk+1 = 0 since {vk+1 + N1 } is
P
P1
a basis of N/N1 . But then ki=1 sk vk = 0 and hence si = 0 for all i, 1 ≤ i ≤ k since {v1 , . . . , vk } is a
basis of N1 . It follows {v1 , . . . , vk+1 } is a basis of N and since k + 1 ≤ (n − 1) + 1 = n we are done.
□

Remark 9.8. Although it is noted in the proof above, it is worth emphasising that if R is an
integral domain, then the submodules of a free module of rank d are free of rank at most d if
and only if R is a PID, because the case of a free module of rank 1 requires that ideals of R must
be principal.

Example 9.9. Suppose that N ⊂ Z3 is the submodule

N = {(a, b, c) ∈ Z3 : a + b + c ∈ 2Z}.

Proposition 9.7 tells us that N must be free of rank at most 3, but let’s use the strategy of proof to
actually find a basis. Let {e1 , e2 , e3 } be the standard basis of Z3 and let Mi = Re1 + Re2 + . . . Rei , so
that each of Mi /Mi−1 is isomorphic to R via the map induced by projecting to the i-th coordinate.

37
Similarly let Ni = N ∩ Mi , so that N1 ⊆ N2 ⊆ N3 = N. Now N1 = {(2a, 0, 0)}, so that (2, 0, 0) is
obviously a generator. For N2 = {(a, b, 0) : a + b ∈ 2Z} we have N2 /N1  (N2 + M1 )/M1 and this
is clearly all of M2 /M1 (since the map is given by (a, b, c) 7→ b and b is clearly arbitrary), so
has a basis e2 + M1 . We can lift this to an element of N2 by taking (99, 1, 0) say. Finally taking
N3 /N2 = N/N2 we again see that it is all of M/M2 so that we can pick (0, 89, 1) as a generator,
and so {(2, 0, 0), (99, 1, 0), (0, 89, 1)} is a basis of N

9.1 Homorphisms between free modules.

In this section we want to study “linear maps” (or module homomorphisms) between free
modules. The advantage of working with free modules here is that by choosing bases, we can
record such a map using a matrix, just like you do in linear algebra over a field. Indeed all
the results of this section have exactly the same proofs as the corresponding results for vector
spaces, which were discussed in Prelims Linear algebra I (see Section §6.5 of the online notes
for that course for example, or the Part A Linear Algebra notes for Lectures 1 and 2). We begin
with a little notation.

Definition 9.10. If M, N are R-modules, let HomR (M, N) denote the set of module homomor-
phisms from M to N. It is an R-module: if ψ, ϕ ∈ HomR (M, N) then ψ + ϕ is a module homomor-
phism (where (ψ + ϕ)(m) = ψ(m) + ϕ(m)) and r.ψ is defined by (r.ψ)(m) = r.(ψ(m)).

Notice that the scalar multiplication gives a module structure only when R is commutative.

Lemma 9.11. Let M and N be R-modules, and let ϕ : M → N be an R-module homomorphism

i) Let X be a spanning set for M, then ϕ is uniquely determined by its restriction to X.

ii) If X is a basis of M then given any function f : X → N there is a unique R-module homomorphism
ϕ f : M → N.

Proof. If v ∈ F, then since X spans M, there are elements x1 , . . . , xn ∈ X and r1 , . . . rn ∈ R such that
v = ni=1 ri xi . Since ϕ is an R-homomorphism it follows ϕ(v) = ni=1 ri ϕ(xi ), so ϕ(v) is uniquely
P P
determined by the {ϕ(xi ) : 1 ≤ i ≤ n}.
If X is also a basis of M we can reverse this process: given f : X → N since the expression
for v ∈ M in terms of the elements of X is unique, we get a well-defined function ϕ f : M → N
by setting, for v = ni=1 ri xi (where ri ∈ R, xi ∈ X, 1 ≤ i ≤ n), ϕ(v) = ni=1 ri f (xi ). This function is
P P
R-linear again because of uniqueness: if v = ni=1 ri xi and48 w = ni=1 si xi then for t ∈ R we have
P P
v + tw = i=1 (ri + tsi )xi , hence
Pn

n
X n
X n
X
ϕ f (v + tw) = (ri + tsi )xi = ri xi + t si xi = ϕ(v) + tϕ(w),
i=1 i=1 i=1

as required. □

Corollary 9.12. Let ϕ : F1 → F2 be a homomorphism of free modules with bases X1 = {e1 , . . . , em } and
X2 = { f1 , . . . , fn } respectively. Then ψ is determined by the matrix A = (ai j ) ∈ Matn,m (R) given by
n
X
ϕ(ei ) = a ji f j . (9.1)
j=1

Conversely given a matrix A ∈ Matn,m (R) the above formula determines a unique R-homomorphism
ϕA : F 1 → F 2 .
48
We can assume v, w lie in the span of some finite subset {x1 , . . . , xn } of X – by definition each of them does and
the union of two finite sets is finite!

38
Proof. This follows immediately from the above, since the matrix A records (once we know the
bases X1 and X2 ) since the map ϕ completely as it records the values of ϕ on X1 . Similarly, if we
are given a matrix A, we may define a function f : X1 → N using Equation (9.1), which extends
uniquely to an R-module homomorphism ϕA : F1 → F2 . □

Exactly as in linear algebra, composition of R-module homomorphisms corresponds to ma-

trix multiplication: As we will use change of bases matrices (at least the special ones corre-
sponding to elementary row and column operations) we review briefly the details now, but
the arguments are exactly the same as in Prelims Linear Algebra.

Lemma 9.13. Let F1 , F2 , F3 be free modules with bases X1 = {e1 , e2 , . . . , en } and X2 = { f1 , f2 , . . . , fm }

and X3 = {g1 , g2 , . . . , gl } respectively. If ϕ : F1 → F2 and ψ : F2 → F3 , the matrix of ϕ with respect to
the bases X1 and X2 is A and the matrix of ψ with respect to the bases X2 and X3 is B, then the matrix of
the homomorphism ψ ◦ ϕ with respect to the bases X1 and X3 is BA.

Proof. We just need to compute ψ ◦ ϕ(ei ) in terms of the basis {g1 , . . . , gl }. But we have
m
X m
X
ψ ◦ ϕ(ei ) = ψ( a ji f j ) = a ji ψ( f j )
j=1 j=1
m
 l 
X X 
= a ji  bk j gk 
j=1 k=1
 
l
X X m 
=  bk j a ji  gk ,

 
k=1 j=1

Thus the matrix of ψ ◦ ϕ has (k, i) entry

Pm
j=1 bk j a ji as required.
□

Corollary 9.14. If F is a free module with basis X = {e1 , . . . , en }, then the set of isomorphisms ψ : F → F
corresponds under the above map to GLn (R) = {A ∈ Matn (R) : ∃B ∈ Matn (R), A.B = B.A = In }, that is,
the group of units in the (noncommutative) ring Matn (R). Moreover, given two bases X = {x1 , . . . , xn }
and Y = {y1 , . . . , yn } there is a unique isomorphism ψ : F → F such that ψ(xi ) = yi .

Proof. The first statement follows from the fact that composition of morphisms corresponds to
matrix multiplication, so that the map sending a homomorphism to the corresponding n × n
matrix is a ring map. For the moreover, note that if X and Y are bases, there is a unique module
homomorphism ψ : F → F such that ψ(xi ) = yi and a unique module homomorphism ϕ : F → F
such that ϕ(yi ) = xi . The composition ϕ ◦ ψ satisfies ϕ ◦ ψ(xi ) = xi , and hence (again by the
uniqueness property) ϕ ◦ ψ = id. □

Exercise 9.15. Let A ∈ Matn (R). The determinant function makes sense for square matrices
with entries in any commutative ring. Characterize the group of invertible matrices GLn (R) in
terms of the determinant function.

Remark 9.16. Lemma 9.13 makes it easy to see how changing the bases of the free modules
effects the matrix we associate to a homomorphism. If F1 and F2 are free modules with bases
X1 and X2 respectively, write X2 [ϕ]X1 for the matrix of the homomorphism ϕ with respect to the
bases X1 , X2 . Let Y1 , Y2 be another pair of bases for F1 and F2 respectively. If A = X2 [ϕ]X1 we
would like to calculate Y2 [ϕ]Y1 in terms of A. To do this, let Q = Y1 [idF1 ]X1 , and let P = X2 [idF2 ]Y2 .
Then it follows from Lemma 9.13 and the fact that ϕ = idF2 ◦ ϕ ◦ idF1 that

Y2 [ϕ]Y1 = PAQ.

39
Definition 9.17. The matrices P and Q are called the change of bases matrices for the pairs of
bases X2 , Y2 and X1 , Y1 respectively. They are readily computed: if F is a free module with two
bases X and Y, the matrix Y [idF ]X has columns given by the “Y-coordinates” of the elements
of the basis X: If Y = { f1 , . . . , fn } and X = {e1 , . . . , en } then e j = ni=1 pi j fi where P = (pi j ) is
P
the change of basis matrix. For example, if F = Rn with standard basis {e1 , . . . , en } (that is,
ei = (0, . . . , 1, . . . 0) where the 1 is in position i) and Y = { f1 , . . . , fn } is any other basis, then the
change of basis matrix from Y to the standard basis is just the matrix with columns the basis
vectors fi , and thus the change of basis matrix from the standard basis to the basis Y is given
by the inverse of this matrix.

The above discussion shows that if ϕ : F1 → F2 is a homomorphism between free modules

F1 and F2 of rank m and n with bases X1 , X2 respectively, we may associate to ϕ a matrix A, and
by picking other possible bases we obtain matrices of the form PAQ where P ∈ GLn (R) and
Q ∈ GLm (R) are change of bases matrices.
Thus the homomorphism ψ corresponds to the equivalence class of A in Matn,m (R) where X
and Y are equivalent if there are invertible matrices P and Q such that Y = PXQ, that is, if they
are in the same orbit of the natural action49 of GLn (R) × GLm (R).
It follows that in order to find a “canonical form” for the homomorphism ϕ, that is, a matrix
representing ϕ which is as simple as possible (and preferably that it be as unique as possible),
we need to find a canonical element of each orbit of the action of GLn × GLm (R) on Matn,m (R).
We will solve this problem in the next section when R is a Euclidean domain.

10 Canonical forms for matrices over a Euclidean Domain.

Before we begin finding this canonical form, it is worth recalling the situation for vector spaces:
in this case the statement we want is essentially the rank-nullity theorem: if R is a field and
ψ : Rm → Rn , then you can pick a basis of ker(ψ) ⊆ Rm and extend it to a basis of Rm . The image
of the vectors you add in to obtain a basis of Rm give a basis of im(ψ), which can in turn be
extended to a basis of Rn . The matrix of ψ with respect to the resulting bases of Rm and Rn is
then diagonal with 1s and 0s on the diagonal, where the rank of ψ is the number of 1s and the
nullity the number of zeros. The key to this argument is the Steinitz Exchange Lemma, which
is that allows you to show that in a vector space you can extend any linearly independent set
to a basis. However, this lemma is obviously false for modules: for example 2Z is free inside Z,
with basis {2}, but the only bases for Z are {1} and {−1}, thus we cannot “extend” the basis {2}
to a basis of Z. We will show in this section that, in a sense, this is the only thing that fails for
modules over a Euclidean domain: that is, we will show that if N is a submodule of Rn there is
always a basis of {e1 , . . . , en } of Rn for which we can obtain a basis of N by taking appropriate
multiples of a subset of the basis. Explicitly, perhaps after reordering the basis {e1 , . . . , en }, we
will show that there are elements c1 , . . . , ck ∈ R and some k ≤ n, such that {c1 e1 , . . . , ck ek } is a
basis for N.
The most explicit way to prove rank-nullity for linear maps between vector spaces is to use
row and column operations (which correspond to particularly simple changes of the basis of
the source and target of the linear map respectively). We will use the same idea for modules
over a Euclidean domain.

Definition 10.1. Let A ∈ Mm,n (R) be a matrix, and let r1 , r2 , . . . , rm be the rows of A, which are
row vectors in Rn . An elementary row operation on a matrix A ∈ Mm,k (R) is an operation of the
form
49
Note that this action, if we want it to be a left action, should be (P, Q).X = PXQ−1 , but the inverse is not too
important since we are only interested in the orbits of the action: A and B are in the same orbit if and only if there
are invertible matrices P and Q such that B = PAQ.

40
 1. Swap two rows ri and r j .

2. Replace one row, row i say, with a new row ri′ = ri + cr j for some c ∈ R, and j , i.

In the same way, viewing A as a list of n column vectors, we define elementary column oper-
ations.
Note that the row operations correspond to multiplying A by elementary matrices on the
left and the column operations correspond to multiplying A by elementary matrices on the
right. Indeed if we let Ei j denote the matrix with (i, j)-th entry equal to 1 and all other entries
zero, then the matrix corresponding to the first row operation is S i j = Ik − Eii − E j j + Ei j + E ji ,
while second elementary row operation is given by multiplying on the left by Xi j (c) = Ik + cEi j .
The column operations are given by multiplying on the right by these matrices.

1
 

 0 1 
 
 1 
S i j =  ..
 
 . 


 1 0 

1

1
 

 . .
.

 

1 c

Xi j (c) = 
 
..


 . 

 1 
 
1
Definition 10.2. If A, B ∈ Matn,m (R) we say that A and B are equivalent if B = PAQ where P ∈
Matn,n (R) and Q ∈ Matm,m (R) are invertible matrices. Thus two matrices are equivalent if and
only if they lie in the same orbit of the GLn (R) × GLm (R) action defined above.
We will say that A and B are ERC equivalent if one can be obtained from the other by a
sequence of elementary row and column operations. Since row and column operations corre-
spond to pre- and pos-multiplying a matrix by elementary matrices, it is clear that two ERC
equivalent matrices are equivalent. (In fact, if you also allow the elementary row and column
operations which simply rescale a row or column by a unit then you can show the converse
too, but we do not need that here.)
For the remainder of this section we assume that R is a Euclidean domain. Recall that we write
N : R\{0} → N for the norm function of our Euclidean domain R.
Theorem 10.3. (Smith normal form) Suppose that A ∈ Matn,m (R) is a matrix. Then A is ERC equiva-
lent (and hence equivalent) to a diagonal matrix D where if k = min{m, n} then50

 d1 0 . . . 0 
 
.
 0 d2 . . 0 
 
 .
 . . . . . . .

D =  . 0


 0 . . . 0 dk 
 . .. .. 
 .
 . . . 
0 0 ... 0


50
The displayed matrix shows the case where k = m, and so there are (n − m) rows below dk consist entirely of
zeros. If k = n then there are (m − n) columns consisting entirely of zeros to the right of dk .

41
and each successive di divides the next (thus possibly d s = d s+1 = . . . dk = 0, for some s, 1 ≤ s ≤ k).
Moreover, the sequence of elements (d1 , d2 , . . . , dk ) is unique up to units.

Proof. We will not prove the uniqueness statement (though see Problem Sheet 4 for how one
can do this). We claim that by using row and column operations we can find a matrix equiva-
lent to A which is of the from
 b11 0 . . . 0 
 
 0 b22 . . . b2m 
B =  . .. . . .. 
 (10.1)
 .. . . . 
0 bn2 . . . bnm
 

where b11 divides all the entries bi j in the matrix. Factoring out b11 from each entry, we may
then applying induction (on n say) to the submatrix B′ = (bi j /b11 )i, j≥2 , to obtain the proposition.
(Note that row and column operations on B′ correspond to row and column operations on B
because b11 is the only nonzero entry in the first row and column of B.)
We are thus reduced to proving the claim. For this we use induction on N(A) = min{N(ai j ) :
1 ≤ i ≤ n, 1 ≤ j ≤ m}. Using row and column swaps we may assume N(a11 ) = N(A).
Step 1: If any ai1 or a1 j is not divisible by a11 , say ai1 , then ai1 = qi1 a11 + ri1 , so taking qi1 times
row 1 from row i we get a new matrix A′ with entry ri1 and N(A1 ) ≤ N(ri1 ) < N(a11 ) = N(A), so
we are done by induction.
Step 2: If all the ai1 s and a1 j s are divisible by a11 we my subtract appropriate multiples of
the first row from the other rows to get a matrix A2 with all entries in the first column below
a11 equal to zero and similarly using column operations we can then get a matrix A3 with all
entries on the first row after a11 equal to zero.
Step 3: Thus A3 has the form we require except perhaps it has an entry not divisible by a11 .
Thus either we are done, or letting (A3 ) = (a3i j ) we have for some i, j > 1, the entry a3i j is not
divisible by a11 = a311 . Then add row i of A3 to row 1, to get a matrix A4 where now and we see
that we are back in the situation of step 1, and we are done by induction.
The claim and hence the theorem are thus proved. □

Example 10.4. The above proposition is really an algorithm, so lets use it in an example, taking
R = Z: Let  
 2 5 3 
A =  8 6 4 
 
 
3 1 0
The entry of smallest norm is the (3, 2) entry, so we swap it to the (1, 1) entry (by swapping
rows 1 and 3 and then columns 1 and 2 say) to get
 
 1 3 0 
A1 =  6 8 4 
 
 
5 2 3

Now since the (1, 1) entry is a unit, there will be no remainders when dividing so we get
 
 1 0 0 
A2 =  0 −10 4 
 

0 −13 3


Next we must swap the (3, 3)-entry to the (2, 2)-entry to get:
 
 1 0 0 
A3 =  0 3 −13 
 

0 4 −10


42
Dividing and repeating our row and column operations now on the second row and column
(this time we do get remainders) gives:
   
 1 0 0   1 0 0 
A4 =  0 3 −13  ∼ A5 =  0 3 2 
   
   
0 1 3 0 1 8

(where ∼ is to denote ERC equivalence). Now moving the (3, 2) entry to the (2, 2)-entry and
dividing again gives:
     
 1 0 0   1 0 0   1 0 0 
A6 =  0 1 8  ∼ A7 =  0 1 0 ∼ A =  0 1 0
     
 8 
  
0 3 2 0 3 −22 0 0 −22

which is in the required normal form.

11 Presentations and the canonical form for modules.

In this section all rings R are PIDs.
So far we have concentrated on the study of free modules. Aside from the fact that these are
the simplest modules to study, we do this because we can understand any (finitely generated)
module in terms of free modules. The key to this is the notion of a presentation of a module.
This is a notion which is important in other parts of algebra also – if you take the Group Theory
option next term for example you will see how it arises in that subject.
The goal is to describe a (finitely generated) module M in concrete terms: explicitly we
will show that any finitely generated module can be described in terms of a finite generating
set and a matrix recording the “relations” amongst the generators. To do this we will use
matrix algebra we developed for homomorphisms of free modules in the previous section.
Once we have this explicit description of M, we can use the canonical form theorem for the
matrix describing M to obtain a canonical form for the module M.

Proposition 11.1. i) Let M be a nonzero finitely generated module. Then there is an n ∈ N and a
surjective morphism ϕ : Rn → M. In particular, Rn /ker(ϕ)  M.

ii) Let M and ϕ be as in i). There exists a free module Rm with m ≤ n and an injective homomorphism
ψ : Rm → Rn such that im(ψ) = ker(ϕ).
In particular, M is isomorphic to Rn /im(ψ).

Proof. For the first part, given any finite subset {m1 , m2 , . . . , mn } of M, if {e1 , . . . , en } is a basis of
Rn (say the standard basis consisting of elements ei = (0, . . . , 1, . . . 0) all of whose coordinates
are zero except for the i-the entry which is equal to 1) then the map ei 7→ mi (1 ≤ i ≤ n) extends,
by Lemma 9.11ii), to a homomorphism ϕ : Rn → M.
Clearly the condition that {m1 , m2 , . . . , mn } is a generating set is then equivalent to the map ϕ
being surjective, since both assert that any element of M can be written in the form ni=1 ri mi =
P
ϕ( i=1 ri ei ) (ri ∈ R, 1 ≤ i ≤ n). The surjectivity of ϕ and the first isomorphism theorem then
Pn
show that Rn / ker ϕ  M.
For the second part, note that since R is a PID, the submodule ker(ϕ) is a free submodule of
rank m ≤ n. Pick a basis {x1 , . . . , xm } of ker(ϕ), and define ψ by sending the standard basis of Rm
to {x1 , . . . , xm }. This map is then clearly injective and has image exactly ker(ϕ) as required. □

Definition 11.2. Let M be a finitely generated R-module. The pair of maps ϕ, ψ of the previous
Lemma, so that im(ϕ) = M and ψ : Rm → Rn has image im(ψ) = ker(ϕ) is called a presentation of

43
the finitely generated modules M. When the map ψ can be chosen to be injective, the presenta-
tion is called a resolution of the module M. It is a special feature of modules over a PID that, for
a finitely generated module, there are presentations which are also resolutions. Future courses
in commutative algebra and what is called homological algebra study what happens to these
two notions for more general rings.

Remark 11.3. (Non-examinable.) The properties of the above homomorphisms ψ and ϕ can be
captured by noticing that
ψ ϕ
0 / Rm / Rn /M /0
is what is called a short exact sequence: An exact sequence is a sequence of homomorphisms where
the image of each map is the kernel the next map in the sequence. A short exact sequence is
one with exactly five terms, the outermost two terms both being 0. Exact sequences play an
important role in algebraic topology and homological algebra.

To see why, in more concrete terms, one calls this a presentation, lets make explicit what
we have done. If {e1 , . . . , em } is the standard basis of Rm and { f1 , . . . , fn } is the standard basis of
Rn , then just as in linear algebra, we may write
n
X
ψ(e j ) = ai j fi
i=1

for some ai j ∈ R, and the resulting matrix A = (ai j )1≤i≤n,1≤ j≤m encodes the homomorphism ψ.
Describing a module M as the quotient Rn /im(ψ) says that M has generators m1 , . . . , mn (the
images of the elements fi + im(ϕ) ∈ Rn /im(ψ) under the isomorphism from Rn /im(ψ) → M
induced by ϕ) and the R-linear dependencies these generators satisfy are all consequences of
the m equations:
X n
ai j mi = 0 ( j = 1, 2, . . . , m).
i=1

Thus the map ϕ : → M picks out the generators we use for M and the map ψ records the rela-
Rn
tions, or linear dependencies, among these generators: that they are R-linear relations among
the generators follows because ϕ◦ψ = 0, while the fact that all other relations are a consequence
of these follows because the elements ( ni=1 ai j fi )mj=1 are a basis for ker(ϕ) = im(ψ). Indeed if we
P
have a relation i=1 ri mi = 0, then it follows that ϕ( ni=1 ri fi ) = 0, that is ni=1 ri fi ∈ im(ψ), which
Pn P P

i=1 ri fi = i=1 si ψ(ei ) =

n m Pm Pn
means that there are are scalars (si )m
P P
i=1 such that i=1 si ( i=1 ai j fi ). In
other words, the R-linear relation given by the scalars (ri )ni=1 can be obtained as a linear com-
bination (given by the (s j )mj=1 ) of the relations ni=1 ai j mi = 0 (1 ≤ j ≤ m). Up to isomorphism
P
then, the structure of the module M is captured by the matrix of relations A = (ai j ). It is this
which allows us to use our canonical form theorem for matrices over a Euclidean Domain to
obtain a canonical form for finitely generated modules over such rings.
For the rest of this section, we will assume all rings are Euclidean Domains, although all results stated
here actually also hold more generally for PIDs.

Theorem 11.4. Suppose that M is a finitely generated module over a Euclidean domain R. Then there
is an integer s and nonzero nonunits c1 , c2 , . . . , cr ∈ R such that c1 |c2 | . . . |cr such that:
r
M
M( R/ci R) ⊕ R s .
i=1

Proof. Since R is a PID we may find a presentation for M, that is, an injection ψ : Rm → Rn
(so that m ≤ n) and a surjection ϕ : Rn → M with ker(ϕ) = im(ψ), so that M  Rn /im(ψ).
Now if A is the matrix of ψ with respect to the standard bases of Rm and Rn , by Theorem 10.3,

44
which gives a normal form for matrices over a Euclidean domain, we know we can transform
A into a diagonal matrix D with diagonal entries d1 |d2 | . . . dm using elementary row and column
operations. But since row and column operations correspond to pre- and post-multiplying A
by invertible matrices, and these correspond to changing bases in Rn and Rm respectively, it
follows that there are bases of Rn and Rm with respect to which ψ has matrix D. But then if
{ f1 . . . , fn } denotes the basis of Rn , we see that the image of ψ has basis {d1 f1 , . . . , dm fm }. Now
m
define a map θ : Rn → n−m by setting for any m = Pn a f ∈ M,
L

i=1 R/di R ⊕ R i=1 i i

n
X
θ( ai fi ) = (a1 + d1 R, . . . am + dm R, am+1 , . . . , an ).
i=1

It is the clear that θ is surjective and ker(θ) is exactly the submodule generated by {di fi : 1 ≤
i ≤ m}, that is, im(ψ). It follows by the first isomorphism theorem that M  Rk /im(ψ) 
L m k−m as required.
i=1 (R/di R) ⊕ R
Finally, since ψ is injective it follows that each of the di are nonzero. On the other hand if di
is a unit (and so all d j for j ≤ i are also) then R/di R = 0, so this summand can be omitted from
the direct sum. The result now follows. □

Remark 11.5. The sequence of elements {c1 , c2 , . . . , cr } are in fact unique up to units. We won’t
have time to show this here (the problem sheets asks you to show uniqueness for c1 and c1 . . . cm
at least given a presentation.). The integer s is also unique, which we now show as a conse-
quence of the important corollary to the structure theorem which says that a finitely generated
torsion-free R-module is free.

Corollary 11.6. Let M be a finitely generated torsion-free module over R. Then M is free. In general
if M is a finitely generated R-module, the rank s of the free part of M given in the structure theorem is
rk(M/M tor ) and hence it is unique.
Lr
Proof. By the above structure theorem, M is isomorphic to a module of the form R s ⊕( i=1 R/ci R),
thus we can assume M isL actually equal to a module of this form.
r
Let F = R s and N = i=1 R/ci R, so that M = F ⊕ N. We claim that N = M . Certainly if
tor

a ∈ R/ci R then since ci |ck we see that ck (a) = 0. But then if m ∈ N, say m = (a1 , . . . , ak ) where
ai ∈ R/ci R it follows ck (a1 , . . . , am ) = (ck a1 , . . . , ck ak ) = (0, . . . , 0) so N is torsion. On the other
hand if m = ( f, n) where f ∈ F and n ∈ N then r( f, n) = (r. f, r.n) = (0, 0) we must have f = 0
since a free module is torsion-free. Thus M tor = N as claimed. It follows that M is torsion-free
if and only if M = F is free. Moreover, by the second isomorphism theorem F  M/M tor (or
more directly, just by noting that the restriction of the quotient map q : M → M/N = M/M tor to
F is an isomorphism since it is readily seen to be injective and surjective) so that s = rk(F) =
rk(M/M tor ). □

(Note that Problem sheet 4 gives an alternative proof that a torsion-free module over a PID is free
using just Proposition 9.7.)
Just to make it explicit, notice that since an abelian group is just a Z-module, our structure
theorem gives us a classification theorem for finitely generated abelian groups.

Corollary 11.7. (Structure theorem for finitely generated abelian groups) Let A be a finitely generated
abelian group. Then there exist an integer r ∈ Z≥0 and integers c1 , c2 , . . . , ck ∈ Z greater than 1 such
that c1 |c2 | . . . |ck and
A  Zr ⊕ (Z/c1 Z) ⊕ . . . ⊕ (Z/ck Z).
Moreover the integers s, c1 , . . . , ck are uniquely determined.

Proof. This is simply a restatement of the previous theorem, except that once we insist the ci
are positive the ambiguity caused by the unit group Z× = {±1} is removed. □

45
 We can give an alternative formulation of the canonical form theorem, known as the primary
decomposition form, using the Chinese Remainder Theorem, or the following slight generaliza-
tion of it.
Lemma 11.8. Let d1 , . . . , dk ∈ R be a set of pairwise coprime elements of a PID, that is h.c.f{di , d j } = 1
if i , j. Then
k
M
R/(d1 d2 . . . . dk )R = R/di R.
i=1

Proof. The condition that the di s are pairwise coprime means that if we set, for i < k, ci =
di+1 . . . dk then for each i we have h.c.f{di , ci } = 1 (indeed if p is a prime element dividing di
and ci then since p is prime it divides one of the factors di+1 , . . . , dk of ci , say d j where j > i.
But then p divides h.c.f{di , d j } contradicting our assumption). Thus we see that for each i with
1 ≤ i ≤ k − 1 we have di R + ci R = R and di R ∩ ci R = di ci R. Thus by the Chinese Remainder
Theorem and induction on k we see that
k
M
R/(d1 . . . dk )R = R/(d1 c1 R)  R/d1 R ⊕ R/(c1 )R  R/di R.
i=1

where in the last isomorphism we may use induction on the factor R/c1 R since c1 = d2 . . . dk is
a product of k − 1 pairwise coprime factors. □

In particular if c = pn11 . . . . pnk k is the prime factorisation of c where the pi are distinct primes
we can apply the previous Lemma (with di = pni i ) to see that
r
M
R/cR  R/pri i R. (11.1)
i=1

This allows us to give an alternative statement of the structure theorem:

Theorem 11.9. (Structure theorem in primary decomposition form): Let R be a Euclidean domain and
suppose that M is a finitely generated R-module. Then there are irreducibles p1 , . . . , pk ∈ R and integers
s, ri , 1 ≤ i ≤ k, such that:
M k
M (R/pri i R) ⊕ R s .
i=1
Moreover, the pairs (pi , ri ) are uniquely determined up to units (where the units act on the pi only).
(Note however that the pi s are not necessarily distinct.)
Proof. This follows immediately using the decomposition (11.1) on each of the cyclic modules
R/ci R in the statement of our first structure theorem. □

Example 11.10. Suppose that A  Z/44Z ⊕ Z/66Z. Then the first structure theorem would write
A as:
A  Z/22Z ⊕ Z/132Z.
Indeed the generators corresponding to the direct sum decomposition give a presentation of
A as Z2 → Z2 → A where the first map is given by the matrix
!
44 0
0 66
and as 66 = 1.44 + 22 we see that row and column operations allow us to show this matrix is
equivalent to:
! ! ! ! !
44 0 44 44 44 44 22 −44 22 0
∼ ∼ ∼ ∼ .
0 66 0 66 −44 22 44 44 0 132

46
On the other hand, for the primary decomposition (since 44 = 22 .11 and 66 = 2.3.11) we would
write A as:
A  (Z/2Z) ⊕ (Z/22 Z) ⊕ (Z/3Z) ⊕ (Z/11Z)⊕2

Notice that the prime 2 appears twice raised to two different powers. Intuitively you should
think of the primary decomposition as decomposing a module into a direct sum of as many
cyclic summands as possible, while the canonical form decomposes the module into a direct
sum with as few cyclic summands as possible.

Remark 11.11. Note that the first structure theorem gives a canonical form which can be ob-
tained algorithmically, while the second requires one to be able to factorise elements of the
Euclidean domain, which for example in C[t] is not an automatically computable operation.

12 Application to rational and Jordan canonical forms.

The structure theorem also allows us to recover structure theorems for linear maps: If V is
a k-vector space and T : V → V is a linear map, then we view V as a k[t]-module by setting
t.v = T (v) for v ∈ V.

Lemma 12.1. Let M be a finitely generated k[t]-module. Then M is finite dimensional as a k-vector
space if and only if M is a torsion k[t]-module. Moreover, a subspace U of V is a k[t]-submodule if and
only if U is T -invariant, i.e. T (U) ⊆ U.

Proof. Given M we can apply the structure theorem to see that:

M  k[t] s ⊕ k[t]/⟨c1 ⟩ ⊕ . . . ⊕ k[t]/⟨ck ⟩,

where s ∈ Z≥0 and the ck are nonconstant51 polynomials. Now k[t] is infinite dimensional as
a k-vector space while k[t]/⟨ f ⟩ is deg( f )-dimensional as a k-vector space, so if follows that M is
torsion if and only if s = 0 if and only if M is finite dimensional as a k-vector space. For the
final statement, notice that a subspace U of M is T -invariant if and only if it is p(T )-invariant
for every p ∈ k[t]. □

The Lemma shows that pairs (V, T ) consisting of a finite dimensional k-vector space V and
a linear map T : V → V correspond to finitely generated torsion k[t]-modules under our corre-
spondence above. We can use this to give structure theorems for endomorphisms52 of a vector
space. Note that the ambiguity about units in the statement of the canonical form theorem can
be removed in the case of k[t]-modules by insisting that the generators ci of the annihilators of
the cyclic factors k[t]/⟨ci ⟩ are taken to be monic.

Definition 12.2. For a monic polynomial f = tn + n−1 i

P
i=0 ai t ∈ k[t] of degree n ≥ 1, the n × n
matrix53
0 . . . . . . 0 −a0 
 
 .. 
1 0 . −a 1 

C( f ) = 0 1 . . . ... ..  .
 
. 
 .. . . . .
 
 . . . 0 −an−2 
0 . . . 0 1 −an−1


is called the companion matrix of f .

51
i.e. nonzero nonunit elements if k[t].
52
Recall that an endomorphism of a vector space V is just a linear map from V to itself.
53
If n = 1 the matrix is C( f ) = (−a0 ).

47
 If λ ∈ k and n ∈ N, then let Jn (λ) be the n × n matrix

λ 1 . . . . . .
 
0
0 λ 1 . . . 0

 . . .. 

. .
Jn (λ) =  .. . . . . . . . 
 . .. ..
 ..

. . 1
0 ... λ
 
0

Lemma 12.3. Let k be any field, and suppose that f ∈ k[t] is a monic polynomial and λ ∈ k.
1. If f = tn + n−1
k=0 ak t then the k[t]-module k[t]/⟨ f ⟩ has basis {t + ⟨ f ⟩ : 0 ≤ i ≤ n − 1 and the matrix
k i
P
for the action of t with respect to this basis is given by C( f ) the companion matrix of f .

2. If g = (t − λ)n ∈ k[t] for some λ ∈ k and n ∈ Z>0 , then the k[t]-module k[t]/⟨g⟩ has basis
{(t − λ)k + ⟨g⟩ : 0 ≤ k ≤ n − 1}. With respect to this basis, the action of t is given by the Jordan
block matrix Jn (λ) (where we order the basis by decreasing powers of (t − λ) in order to get an
upper triangular matrix).
Proof. Recall that by the division algorithm for polynomials, each coset of ⟨ f ⟩ has a unique
representative of degree strictly smaller than that of f . The assertion that the two sets in parts
(1) and (2) are k-bases then follows because (t − λ)k : 0 ≤ k ≤ n − 1} is clearly a basis for the
space of polynomials of degree at most n − 1 for any λ ∈ k. For the assertions about the action
of t, note for (1) that t.(ti + ⟨ f ⟩) = ti+1 + ⟨ f ⟩ for i < n − 1, while if i = n − 1, t.tn−1 + ⟨ f ⟩ = tn + ⟨ f ⟩ =
− n−1k=0 a t + ⟨ f ⟩. For (2) note that t acts with matrix Jn (λ) if and only if (t − λ) acts by Jn (0),
P k k

which is clear: as noted in the statement of the Lemma, in order to get an upper triangular,
rather than a lower triangular, matrix, we need to order the basis by decreasing degree rather
than increasing degree. □

Theorem 12.4. (Rational Canonical Form.) Suppose that V is a nonzero finite dimensional k-vector
space and T : V → V is a linear map. Then there are unique nonconstant monic polynomials f1 , . . . , fk ∈
k[t] such that f1 | f2 | . . . | fk and a basis of V with respect to which T has matrix which is block diagonal
with blocks C( fi ):
...
 
 C( f1 ) 0 0 
 .. 
 0 C( f2 ) 0 . 
 . .. 
 . ..
 . 0 . . 
0 ... 0 C( fk )


Lk
Proof. By the canonical form theorem and Lemma 12.1, there is an isomorphism θ : V → i=1 k[t]/⟨ fi ⟩
of k[t]-modules, where54 f1 | f2 | . . . | fk and the fi are monic nonunits (hence nonconstant polyno-
mials). The fi are unique (rather than unique up to units) since we insist they are monic. Now
Lk
the direct sum i=1 k[t]/⟨ fi ⟩ has a basis B given by the union of the bases in Lemma 12.3, and
the preimage θ−1 (B) is thus a basis of V. The matrix of T with respect to this basis is thus the
Lk
same as the matrix of the action of t on the direct sum i=1 k[t]/⟨ fi ⟩, and again by Lemma 12.3
this is clearly block diagonal with blocks C( fi ) (1 ≤ i ≤ k) as required.
□

This matrix form for a linear map given by the previous theorem is known as the Rational
Canonical Form of T . Notice that this form, unlike the Jordan canonical form, makes sense for
a linear map on a vector space over any field, not just an algebraically closed field like C.
We can also recover the Jordan canonical form for linear maps of C-vector spaces from
the second, primary decomposition, version of our structure theorem, which expresses each
54
Note that k , 0 since we are assuming that V is not {0}.

48
module in terms of cyclic modules k[t]/⟨ f k ⟩ where f is irreducible. The monic irreducibles over
C are exactly the polynomials t−λ for λ ∈ C. Thus the second structure theorem tells us that, for
V a finite dimensional complex vector space and T : V → V, we may write V = V1 ⊕ V2 ⊕ . . . ⊕ Vk
where each Vi isomorphic to C[t]/⟨(t − λ)r ⟩ for some λ ∈ C, and r ∈ N. The Jordan canonical
form now follows exactly as in the proof of the rational canonical form, replacing the use of the
canonical form for modules with the primary decomposition, and the use of part (1) of Lemma
12.3 with part (2) of the same Lemma.

12.1 Remark on computing rational canonical form.

It is also worth considering how one can explicitly compute the decomposition that yields the
rational canonical form: our proof of the existence of the canonical form is constructive, so if
we can find a presentation of the k[t] module given by a linear map acting on a k-vector space
V then we should be able to compute. The following proposition shows one way to do this.

Proposition 12.5. Let V be an n-dimensional k-vector space and ϕ : V → V a linear map. If ϕ has
matrix A ∈ Matn (k) with respect to a basis {e1 , . . . , en } of V, then the k[t]-module corresponding to (V, ϕ)
has a presentation
/ k[t]n f / V
k[t]n
r

where the homomorphism r between the free k[t]-modules is given by the matrix tIn − A ∈ Matn (k[t]),
and the map from f : k[t]n → V is given by ( f1 , . . . , fn ) 7→ ni=1 fi (A)(ei ).
P

Proof. Sketch: Since t acts by ϕ on V, and ϕ has matrix A, it follows that the image N of the map r
lies in the kernel of the f . It thus suffices to check that this map is injective and its image is the
whole kernel. To see that it is the whole kernel, let F = kn ⊂ k[t]n be the copy of kn embedded
as the degree zero polynomials. It follows immediately from the definitions that f restricts to
an k-linear isomorphism from F to V, and thus it is enough to show that N + F = k[t]n and
N ∩ F = {0} (where the former is the vector space sum). Both of these statements can be checked
directly: the intersection is zero because f restricts to an isomorphism on F and N ⊆ ker( f ).
The sum N + F must be all of k[t]n since it is easy to check that it is a submodule and it contains
F which is a generating set. Finally, since the quotient k[t]n /N is torsion, N must have rank n
and hence r does not have a kernel (since the kernel would have to be free of positive rank, and
hence the image would have rank less than n.) □

Example 12.6. Let A be the matrix

 
 0 1 0 
A =  0 0 1
 
 
−1 −2 −3

Then we have
3+t
     
 t −1 0   1 2   1 0 0 
tI3 − A =  0 t −1  ∼  0 t −1 ∼ 0 t −1
     
  
1 2 3+t 0 −1 − 2t −3t − t2
  
t −1 0
   
 1 0 0   1 0 0 
∼  0 −1 t  ∼  0 −1 0
   

 2
0 −3t − t −1 − 2t 0 0 −t3 − 3t2 − 2t − 1

so that (Q3 , A) is isomorphic as a Q[t]-module to Q[t]/⟨g⟩ where g = t3 + 3t2 + 2t + 1.

49
13 Appendix A: Polynomial rings and convolution.
In this appendix we discuss in somewhat more detail the construction of polynomials rings
with coefficients in an arbitrary ring, and point out how the construction generalizes in a num-
ber of interesting ways.
Consider the set C[t] of polynomials with complex coefficients. This is a ring with the “obvi-
ous” addition and multiplication: if p, q are polynomials, then p+q and p.q are the polynomials
given by pointwise addition and multiplication – that is, p.q(z) = p(z).q(z) for all z ∈ C and simi-
larly for addition. In other words, we realise the ring of polynomials with complex coefficients
as a subring of the ring of all functions from C to itself. To check that polynomials do indeed
form a subring, we need to check (amongst other things55 ) that p.q is a polynomial if p and q
are. But let p = k=0 ak tk and q = k=0
PN PM
bk tk . Then
 N  M 
X  X  X
p.q =  ak tk   bl tl  = ak bk tk+l
k=0 l=0 k,l
(13.1)
N+M
X X
= ( ak bk )t ,
n

n=0 k+l=n

where we take ak = 0 for k > N and bl = 0 for l > M, and the second line is evidently a
polynomial function as required.
However, if we want to consider polynomials with coefficients in an arbitrary ring, we en-
counter the problem that a polynomial will not determined by its values on elements of the
ring: for example if R = Z/2Z, then since R has two elements there are only four functions from
R to itself in total, but we want two polynomials to be equal only if all their coefficients are
the same and so we want infinitely many polynomials even when our coefficient ring is finite.
Indeed, for example, we want 1, t, t2 , . . . to all be distinct as polynomials, but as functions on
Z/2Z they are all equal!
The solution is much like what we do when we construct complex numbers – there we
simply define a new multiplication on R2 and check it, along with vector addition, satisfy the
axioms for a field. We start by viewing a polynomial as its sequence of coefficients, and define
what we want the addition and multiplication to be, and again just check that the ring axioms
are satisfied. This approach will also give us a new ring, called the ring of formal power series,
simply by allowing all sequences in R, not just ones which are zero for large enough n ∈ N.

Definition 13.1. Let R be a ring, and define R[[t]] = {a : N → R} the set of sequences taking
values in R. We define binary operations as follows: for sequences56 (an ), (bn ) ∈ R[[t]] let
X
(an ) + (bn ) = (an + bn ); (an ) ⋆ (bn ) = (cn ), where cn = ak bl
k+l=n

Thus the addition just comes from pointwise addition of functions from N to R, but the multi-
plication comes from formula we got in Equation (13.1).

It is immediate that the sequence with all terms equal to 0 ∈ R is an additive identity in
R[[t]], while the identity for our multiplication operation ⋆ is the sequence 1 = (1, 0, . . .), that is
1n = 1 if n = 0 and 1n = 0 if n > 0. The fact that ⋆ distributes over addition is also straightforward
to check, while the associativity of ⋆ follows because
X X
(an ) ⋆ (bn ) ⋆ (cn ) = (ak bl )c p = ak (bl c p ) = (an ) ⋆ (bn ) ⋆ (cn )





k+l+p=n k+l+p=n
55
Though closure under product is probably the most substantial thing to check.
56
As is standard enough for sequences, we write an rather than a(n) for the values of the sequence a : N → R.

50
Definition 13.2. Now let R[t] be the subset of R[[t]] consisting of sequences (an ) such that there
is some N ∈ N for which an = 0 for all n > N. To check this is a subring, notice that if (an ), (bn ) ∈
R[t] and an = 0 for all n > N and bn = 0 for all n > M, then the sequence (cn ) = (an ) ⋆ (bn ) is zero
for all n > N + M: indeed if k + l = n > N + M we cannot have both k ≤ N and l ≤ M and so the
product ak bl will be zero, and hence cn = k+l=n ak bl = 0 for all n ≥ N + M.
P

Finally we want to relate our construction to the notation we are used to for polynomials.
Let t ∈ R[t] be the sequence (0, 1, 0, . . .), that is tn = 1 for n = 1 and tn = 0 for all other n ∈ N.
Then it is easy to check by induction that tk = t ⋆ t ⋆ . . . ⋆ t (k times) has tnk = 1 if n = k and
tnk = 0 for all other n ∈ N. It follows that if (an ) is a sequence in R[t] for which an = 0 for all n > N
then (an ) = k=0 ak tk . Note that if (an ) is any element of R[[t]] it is the case that (an ) = k∈N ak tk ,
PN P
where the right-hand side gives a well-defined sequence in R despite the infinite sum, because
for any integer k only finitely many (in fact exactly one) of the terms in the infinite sum are
non-zero. This is why the ring R[[t]] is known as the ring of formal power series.

Remark 13.3. This definition also allows us to define polynomial rings with many variables:
given a ring R let R[t1 , . . . , tk ] be defined inductively by R[t1 , . . . , tk+1 ] = (R[t1 , . . . , tk ])[tk+1 ]. Thus
for example, R[t1 , t2 ] is the ring of polynomials in t2 with coefficients in R[t1 ].

Remark 13.4. The problem that a polynomial p ∈ R[t] is not determined by the function it gives
on the ring R can be resolved: recall the Evaluation Lemma which says that if S is a ring and
we are given a ring homomorphism i : R → S and an element s ∈ S , then there is a unique
ring homomorphism θ s : R[t] → S which restricts to i on R and has θ s (t) = s. This allows
us to produce, for every ring homomorphism i : R → S and polynomial p ∈ R[t] a function
pS : S → S : simply take pS (s) = θ s (p). In other words, given any homomorphism of rings
i : R → S we can evaluate a polynomial in R[t] on the elements of S , so the polynomial gives
not just a function on R but a function on any ring we can relate R to. In particular, if R is a
subring of S , the it makes sense to evaluate p ∈ R[t] on every element of S . The collection of
all the functions we can associate to a polynomial in this way does completely determine the
polynomial.

13.1 Convolution algebras.

The key to defining polynomial rings was the multiplication formula given by (13.1). This uses
the fact that N has an addition operation. We can generalise this to give another interesting and
important construction of a ring.

Definition 13.5. Let G be a group and let R be a ring. If f : G → R is a function, we let supp( f ) =
{x ∈ G : f (x) , 0}. Let R[G] be the set of R-valued functions on G which have finite support,
that is functions f for which supp( f ) is a finite set. We define ⋆ to be the binary operation
X
( f ⋆ g)(x) = f (y1 )g(y2 ). (13.2)
y1 y2 =x

If S 1 = supp( f ) and S 2 = supp(g), then the terms in the sum on the right-hand side are
zero unless (y1 , y2 ) ∈ S 1 × S 2 , hence this sum is finite since S 1 and S 2 are. Moreover, supp( f ⋆ g)
is a subset of {xy ∈ G : x ∈ S 1 , y ∈ S 2 }, which is also clearly finite, thus ⋆ is indeed a binary
operation. It is associative because
X
(( f ⋆ g) ⋆ h)(x) = ( f ⋆ (g ⋆ h))(x) = f (y1 )g(y2 )h(y3 ).
y1 y2 y3 =x

Just as for polynomials, it is straight-forward to check that (R[G], +, ⋆, 0, δe ) is a ring, where

+ is pointwise addition, 0 is the function on G which is 0 on every element of G, and δe is the

51
indication fucntion of the identity element e of G, that is δe (x) = 1 if x = e, δe (x) = 0 otherwise.
This ring is known as the group algebra of G with coefficients in R. This ring (when R = C) will
be very important in the study of representations of finite groups in the Part B representation
theory course.

Example 13.6. Let G = Z. Show that the ring R[Z] is just the ring of Laurent polynomials
R[t, t−1 ].

Remark 13.7. i) The natural number N are of course not a group under addition, but the
convolution product still makes sense. This is because we only ever use the associativity
of the product and the existence of an identity element in constructing the ring R[G]. Thus
the construction actually works for any monoid (Γ, ×, e), that is, a set Γ with an associative
binary operation × and an identity element e ∈ Γ for the binary operation.

ii) Some books, especially those on representations of finite groups, present the group alge-
bra slightly differently: they define R[G] to be the set of formal R-linear combinations on
the elements of the group, and then define multiplication by extending the group prod-
uct “linearly”, that is the elements of R[G] are of the form g∈G ag .g where ag ∈ R and the
P
product is given by
X X X X X
( ag .g)( bh .h) = ag bh (g.h) = ( ag bh ).x
g∈G h∈G g,h∈G x∈G gh=x

This is readily seen to be isomorphic to the definition above via the map which sends
a group element g to the function eg which takes the value 1 on g and 0 on all other el-
ements of G. The function-based approach is more important when studying infinite
groups with additional structure (such as being a metric space say) when you can con-
sider subrings of the ring of all functions, such as continuous functions.

Remark 13.8. The restriction on the support of the functions in R[G] ensures that the formula
(13.2) gives a well-defined operation. Products given by this formula are called convolution
products and come up in many parts of mathematics. Note that the formula is sometimes writ-
ten less symmetrically as: X
( f ⋆ g)(x) = f (xy−1 )g(y),
y∈G

If the group G is infinite, for example if G = R, then instead of summing over elements of the
group one can integrate (imposing some condition on functions which ensures the integral
makes sense and is finite) and the convolution formula becomes:
Z
( f ⋆ g)(x) = f (x − y)g(y)dy,
R

which may be familiar from the study of integral transforms.

52
14 Appendix B: Unique Factorization for Z.
In this Appendix we establish unique factorization for the ring Z. The strategy of proof will be
what motivates the definition of a Euclidean Domain, so if you have a good understanding of
the material in this note, it should make the part of the course on EDs, PIDs and UFDs easier
to grasp.
Theorem 14.1. If n ∈ Z\{0, ±1}, then we may write n = p1 . . . pk where the pi are primes and the
factorization is unique up to sign and reordering of the factors.
Remark 14.2. i) Most of the work we will need to do to prove the theorem will be to under-
stand what the right notion of a “prime” is. Once we establish that, the proof of unique
factorization will be quite straight forward.

ii) We work with all integers, not just positive ones, so we will have positive and negative
primes numbers.

iii) The uniqueness statement is slightly cumbersome to say, but in essence it says the factor-
ization is as unique as it can possibly be: for example if n = 6 then we can write

n = 2.3 = 3.2 = (−2).(−3) = (−3).(−2),

and while each of these are prime factorizations they are clearly all “essentially the same”.
The ambiguity of signs would be removed if we insisted that n and all the primes were
positive, but it is more natural to ask for a statement which holds for any element of Z.

14.1 Highest common factors.

Let’s begin with some (hopefully familiar enough) terminology:
Definition 14.3. If n ∈ Z we say that a divides n, or a is a factor of n is there is an integer b such
that n = a.b. We will use the notation a | n to denote the fact that a divides n.
Remark 14.4. Notice that if a | b then any multiple of b is also a multiple of a and so bZ ⊆ aZ.
Thus divisibility corresponds to containment of ideals. In particular, since {0} ⊆ nZ, every
integer divides 0 while 0 only divides itself.
The most basic observation about the integers and division is that, given any pair of integers
a, b, provided b is not zero, we can do “division by b with remainders”. The next lemma makes
this precise. Since we are working with all integers rather than positive integers, it is convenient
to use the absolute value function |.| : Z → Z≥0 .
Lemma 14.5. (Division algorithm.) For any a, b ∈ Z such that b , 0, there exist integers q and r such
that
i) a = q.b + r.

ii) |r| < |b|;

Proof. Note that if a = q.b + r then −a = (−q).b + (−r), and |r| = | − r| so that the pair (−q, −r)
satisfy the conditions we require for the integers (−a, b). Thus it is enough to prove the Lemma
for a ≥ 0. Similarly if a = q.b + r then a = (−q).(−b) + r, so it is enough to prove the Lemma in
the case b > 0 also. We will prove this by induction on a. If a < b then the result is clear as we
may take q = 0 and r = a. If a ≥ b then consider the set S = {q.b : q ∈ N, q.b ≤ a}. Since S is
clearly finite (it is for example contained in the set {1, 2, . . . , a}) it has a maximal element. Set q
to be this element. Then qb ≤ a < (q + 1)b, and so if r = a − q.b it follows 0 ≤ r < b, and we are
done. □

53
Remark 14.6. Note that if we work with nonnegative integers a, b and insist that q and r are
nonnegative also then, for given a, b the integers q and r are unique, but if we work with Z then
they are not: Indeed qb + r = (q + 1).b + (r − b), and if 0 < r < b then −b < b − r < 0, so (q + 1, r − b)
is an alternative solution. Concretely, if (a, b) = (10, 7) say, then 10 = 1.7 + 3 = 2.7 − 4.
Notice also that while it makes sense to say m divides n for m, n elements of any ring R,
condition ii) of the Division Algorithm uses the absolute value function and the ordering on
positive integers, thus it won’t make sense for an arbitrary ring. (It will, however, motivate the
definition of a class of rings called “Euclidean Domains”.)

The first step in understanding how factorization works in Z is to understand the notion of
a highest common factor. The crucial point is that the right condition for a common factor to
be the “highest” is not just to ask for the largest of the common factors in the usual sense:

Definition 14.7. Let n, m ∈ Z. We say that d is a common factor of m, n if d | m and d | n. We

say that c is the highest common factor if it is a common factor and whenever d ∈ Z is any other
commmon factor then d|c. We will write h.c.f(m, n) for a highest common factor of m and n.

The only downside of this definition is that it is not immediately clear that h.c.fs always
exist! On the other hand, it does follow from the definition that the highest common factor,
if it exists, is unique up to sign: If c1 and c2 are highest common factors, then because c1 is
a common factor and c2 is a highest common factor we must have c1 |c2 , but symmetrically
we also see that c2 |c1 . It is easy to see from this that c1 = ±c2 , and so if we require highest
common factors to be non-negative, they are unique. (Indeed the argument essentially repeats
the proof we saw in lectures that in an integral domain, the generators of a principal ideal are
all associates.)
The existence of highest common factors relies on the division algorithm, as we now show.
The argument also proves that the ideals in Z are exactly the principal ideals nZ, so since that
is of independent interest, we establish this first.

Lemma 14.8. i) Let I be an ideal of Z. Then there is an n ∈ Z such that I = nZ, that is, I is principal.

ii) Let m, n ∈ Z. The highest common factor h.c.f(m, n) exists and moreover there are integers r, s ∈ Z
such that h.c.f(m, n) = am + bn.

Proof. For the first part, if I = {0} then clearly I is the ideal generated by 0 ∈ Z and we are done.
If I , {0}, then the set {|k| : k ∈ I\{0}} is nonempty, and so we may take n ∈ I with |n| minimal
among nonzero elements of I. But now if a ∈ I is any element, we may write a = qn + r for some
q, r ∈ Z with |r| < |n|. But r = a − q.n ∈ I, so by the minimality of |n| we must have r = 0 and so
a = qn. It follows that I ⊆ nZ. But since n ∈ I we have by definition that nZ (the ideal generated
by n) must lie in I, hence I = nZ as required.
For the second part, note that if m, n are integers then

I = nZ + mZ = {r.n + s.m : r, s ∈ Z},

is the ideal generated by {m, n}. By the first part, I must be principal, and hence there is some
k ∈ Z such that I = kZ. But then since n, m ∈ I we must have k | n and k | m so that k is a common
factor. On the other hand, since k ∈ I, it follows immediately from the definition of I that there
are integers a, b such that k = am + sb. Now if d is any common factor of m and n, then it is clear
that d divides any integer of the form r.m + s.n, and so d divides every element of I and hence
d | k. The second part of the Lemma follows immediately. □

Remark 14.9. The second part of the above Lemma is usually known as Bézout’s Lemma. Its
proof has the advantage that it can actually be made constructive. (This is not needed for the
rest of this note, but is something you saw before in Constructive Mathematics.)

54
 Suppose that m, n are integers and 0 < n < m. Euclid’s Algorithm gives a way to compute
h.c.f(n, m). Let n0 = m, n1 = n, and if n0 > n1 > . . . > nk > 0 have been defined, define nk+1 by
setting nk−1 = qk nk + nk+1 where 0 ≤ nk+1 < nk (since we are insisting everything is positive, the
division algorithm ensures this uniquely defines the integers qk and nk+1 ). Clearly this process
must terminate with nl−1 > nl = 0 for some l > 0.

Lemma 14.10. The integer nl−1 is the highest common factor of the pair (m, n)

Proof. The equation nk−1 = qk nk + nk+1 shows that any common factor of the pair (nk−1 , nk ) is
also a common factor of the pair (nk , nk+1 ). Thus

h.c.f(m, n) = h.c.f(n0 , n1 ) = h.c.f(n1 , n2 ) = . . . = h.c.f(nl−1 , 0) = nl−1 .

14.2 Characterising prime numbers.

We are almost ready to prove unique factorization now. The last ingredient that we need is a
better understanding of the properties of prime numbers.

Definition 14.11. An integer n ∈ Z is said to be irreducible if n < {±1} and its only factors are
{±1, ±n}, that is, if n = a.b for some a, b ∈ Z then either a = ±1 or b = ±1.

The notion of an irreducible integer is what people normally call a “prime”, but there is
another characterization of prime integers which is the key to the proof of unique factorization,
and we reserve the term “prime” for this characterization. (For rings other than Z, the two
notions are not necesarily the same.)

Definition 14.12. Let n ∈ Z. Then we say n is prime if n < {±1} and whenever n | a.b, either
n | a or n | b (or both). (Using the terminology for ideals which we have developed, n is prime
whenever nZ is a prime ideal in Z).

Remark 14.13. Note that it follows easily by induction that if p is a prime number and p |
a1 . . . ak for ai ∈ Z, (1 ≤ i ≤ k), then there is some i with p | ai .

We now want to show that the irreducible and prime integers are the same thing. This is a
consequence of Bézout’s Lemma, as we now show:

Lemma 14.14. If n ∈ Z\{0} then n is irreducible if and only if n is prime.

Proof. Suppose that n is a nonzero prime and write n = a.b for integers a, b. Then clearly n | a.b
so by definition we must have n | a or n | b. By symmetry we may assume that n | a. Then
a = n.p for some integer p and so n = (np)b = n(pb) and hence n(1 − pb) = 0. Since n is nonzero,
it follows that p.b = 1 so that b = ±1, and thus n is irreducible.
Conversely, suppose that n is irreducible. Then if n | a.b, suppose that n does not divide a.
Then by irreducibility, we must have h.c.f(a, n) = 1, and so by part ii) of Lemma 14.8 (Bézout’s
Lemma) we may write 1 = ra + sn for some r, s ∈ Z. But then b = r(a.b) + n(sb), and hence n
divides b as required. □

14.3 Unique factorization.

We are now ready to prove unique factorization for Z.

Theorem 14.15. Any integer n ∈ Z\{0, ±1} can be written as a product n = p1 p2 . . . pk of primes,
uniquely up to reordering and sign.

55
Proof. We first show that any such integer n is a product of primes using induction on |n|. Since
n < {0, ±1}, the smallest value for |n| is 2, and in that case n = ±2 and so n itself is prime. If
|n| > 2, then there are two cases: either n is prime, in which we are done, or it can be written
as a product n = a.b where |a| > 0 and |b| > 0. But then certainly |a|, |b| < |n|, so by induction we
can write a = r1 . . . r p and b = s1 . . . sq where the ri and s j are primes. Thus we see that

n = a.b = r1 . . . r p s1 . . . sq ,

showing that n is a product of primes.

Next we must show the product is unique. For this we again use induction, but now on
the number of prime factors of n. Since we don’t know uniqueness yet, we have to be slightly
careful here: it might be that n can be written a product of primes in two different ways where
the number of factors is different in the two factorizations. Thus we define, for n ∈ Z\{0, ±1},
the number P(n) to be the minimum number of prime factors occuring in a prime factorization
of n, and use induction on P(n). If P(n) = 1 then n is prime. But then the only factors of n are
{±1, ±n} and so n can be written uniquely as a product of one prime (itself) (or as (−1).(−n)).
Now if P(n) = k > 1, assume uniqueness holds for any m ∈ Z with P(m) < k. Write n =
p1 . . . pk and suppose that we have another prime factorization n = q1 . . . qt (where by definition
t ≥ k). Now p1 is prime and p1 | n = q1 . . . qt , so as above we must have p1 | qi for some
i (1 ≤ i ≤ t) and reordering the q j s we can assume that i = 1. Then since q1 is prime we
find p1 = ±q1 . Now if m = p2 . . . pk we must have P(m) ≤ k − 1, and so by induction unique
factorization holds for m = p2 . . . pk = (±q2 ) . . . ql , and so it follows that the pi s and q j s for i, j > 1
are also equal up to signs and reordering. □

Remark 14.16. As mentioned before, the ambiguity about signs disappears if we only consider
factorization for positive integers. If we were only interested in Z that might well be the best
thing to do, but since we are interested in generalising to other rings where there may not be
a convenient analogue of the notion of a positive integer, it is better to find a statement valid
for all integers. Note that the reason signs appear in the uniqueness statement is because {±1}
is the group of units in the ring Z. Thus one could rephrase the statement by saying that the
factorization is unique “up to reordering and units”.

We finish with a brief discussion of a ring only slightly bigger than Z where the notion of
an irreducible element and a prime element are different, and where unique factorization fails.
√ √
Example 14.17. Let R = Z[ −17] = {a + b −17 : a, b ∈ Z}. √ It is straight-forward to check that R
is a subring of C. The function N : R → Z given by N(a + b −17) = a2 + 17b2 is multiplicative, in
that N(z.w) = N(z).N(w) (indeed it is just the√restriction to R of the function z 7→ |z|2 = zz̄ on C).
Using this, one can show that 2, 3, and 1 ± −17 are all irreducible element of R (in the same
sense as we used for ordinary integers). It then follows that none of these elements are prime –
indeed it is easy to see that there is an element of R which is a product of both two irreducibles
and three irreducibles.

56
15 Appendix C: A PID which is not a ED
In this appendix, following the article57 of √ Cámploi which you can read through JSTOR, we
outline a proof that the ring R = Z[ 2 (1 + −19)] is a PID but not a Euclidean domain. (This is
1

for curiosity only, it is not examinable.) √

For convenience we write θ = 21 (1 + −19). We need to do two things, first to show that
R is not a Euclidean domain, and then to show that nevertheless it is a PID. We will need the
restriction of the square of the modulus function on complex numbers, which we will write as
N, that is N(a + ib) = a2 + b2 (and crucially N(z.w) = N(z).N(w)). It is easy to check that θ2 = θ − 5,
so that in particular R = {a + bθ : a, b ∈ Z}. Moreover, we have N(a + bθ) = a2 + ab + 5b2 , and so
if z ∈ R\{0} then a2 + ab + 5b2 = N(z) is a positive integer.

15.1 R is not a Euclidean Domain

Using the function N it is not too hard to see that the units in R are exactly {±1}, because if r ∈ R
is a unit N(r) must be 1. But if a2 + ab + 5b2 = 1 and a.b ≥ 0 then clearly we must have b = 0 and
a = ±1. If a.b ≤ 0 then writing N(a + bθ) = (a + b)2 − ab + 4b2 we again see that b = 0, and so the
only units are ±1. Similar considerations using N allow you to show that 2, 3 are irreducible
elements in R. Using these facts we can show that R is not a Euclidean domain: Suppose for
the sake of a contradiction that d : R\{0} → N is a Euclidean function on R. Then consider the
minimal value k of d on the elements of R\{0} which are not units, and pick some m ∈ R with
d(m) = k. If we divide m into 2 we see that we have 2 = q.m + r, where r = 0 or d(r) < d(m), and
hence either m divides 2 or r is a unit. Since 2 is irreducible this would imply either m = ±2
(since the only units in R are ±1) or 2 = q.m ± 1, which since m is not a unit, forces q.m = 3 and
hence again since 3 is irreducible m = ±3. Thus we see that m is ±2 or ±3. But now consider
dividing m into θ, say θ = q.m + r. Certainly θ cannot be divisible by 2 or 3, but then as before
we must have r = ±1, which would imply θ ± 1 = ±2q or ±3q for some q ∈ R, which is clearly
impossible. It follows that R is not a Euclidean Domain.

15.2 R is a PID
To see that R is a PID the key is to show that it is not too far from being a Euclidean Domain.
Recall that the proof that a Euclidean Domain is a PID takes a element n of minimal Euclidean
norm in an ideal I and shows this must be a generator because if m ∈ I then m = q.n + r where
d(r) < d(n) or r = 0, and r = m − q.n ∈ I forces r = 0. This argument works for r any linear
combination of m and n, so we can prove a ring is a PID if we can find a function d which satisfies
the following weaker version of the condition for a Euclidean norm: Say that d : R\{0} → N is a
weak norm if given any n ∈ R\{0} and m ∈ R there exist α, β ∈ R such that d(α.m + β.n) < d(n). If a
ring has a weak norm then the above argument shows it is a PID. Hence to see that R = Z[θ] is
a PID, it is enough to check that N(a + bθ) = a2 + ab + 5b2 , the squared modulus, is a weak norm.
The proof is similar but somewhat more involved to how one shows that Z[i] are a Euclidean
Domain with the same function. Let m, n ∈ R, and consider m/n ∈ C. We want to find α, β ∈ R
so that N(β(m/n) − α) < 1, so that N(β.n − α.n) < N(n). But it is easy to check that any ratio of
elements of R lies in {a + bθ : a, b ∈ Q} (just clear denominators using the complex conjugate).
Hence for any such m/n we can subtract
√ from it an element of R to ensure that the imaginary
part of the result lies between ± 19/4. Now 58 if the imaginary part of the result, q say, is less
√
that 3/2√in modulus, then √ q will be within 1 or an element of Z ⊂ R, so it suffices to consider
the case 3/2 < Im(q) < 19/4 (the case when Im(q) < 0 being similar). But then 2q − θ has
57
“A Principal Ideal Domain that is not a Euclidean Domain”, Oscar A. Cámpoli, American Mathematical
Monthly, vol. 95, no. 9, 868-871.
58
This analysis follows Rob Wilson’s note which you can read at www.maths.qmul.ac.uk/ raw/MTH5100/PIDnotED.pdf.

57
 √ √ √
imaginary part between 3 − 19/2 and 0, which you can check is less that 3/2, and so we
are done.

58