GEIR ELLINGSRUD

M A 4 2 0 0 — C O M M U TAT I V E
ALGEBRA
 1

These are informal notes for the course MAT4200 autumn 2018. As you
will see they are "under construction", so they highly unfinished and errors
abound. Some chapters, the first two or three, are not that preliminary and
should be readable. Hopefully the rest will improve during the semester. Each
chapter has its own version number. If the number is positive, the chapter
should be readable(and as usual the higher version number the more ready
the chapter is), but if negative (like ´8) it is sketchy if existing at all. I shall
indicate major changes, so you can follow the development.
Geir
 Contents

1 Rings 7
Rings 8
Polynomials 15
Direct products and idempotents 18

2 Ideals 23
Ideals 23
Quotient rings and kernels 28
Prime ideals and maximal ideals 31
Principal ideals 34
Existence theorems 39
Local rings 44
Direct products and the Chinese Remainder Theorem 46
Graded rings and homogenous ideals 48
The prime spectrum and the Zariski topology 52

3 Unique factorization domains I 57

Unique factorization domains 57

4 Modules 65
The axioms 65
Direct sums and direct products 72
Finitely generated modules 77
Bases and free modules 79
Exact sequences 85
Snakes and alike 95
 4

5 Tensor products 101

Introducing the tensor product 101
Basic working formulas 103
Functorial properties 109
Change of rings 114
Tensor products of algebras 117

6 Localization 123
Localization of rings 124
Localization of modules 135
Nakayama’s lemma 139
The support of a module 143

7 Chain conditions 151

Noetherian modules 151
Noetherian rings 155
Hilbert’s Basis Theorem and two other results 159
Modules of finite length 164
Artinian ring 170

8 Primary decomposition 173

Primary ideals 174
The Lasker-Noether theorem 178
The homogeneous case 186
Primary decomposition of modules 189

9 Integral extensions 191

Definition and basic properties 191
Examples 197
The Cohen–Seidenberg Theorems 200
Noether’s normalization lemma 204
The Nullstellensatz 206
Discrete valuation rings 208
Appendix—skirmishes 209
 5

10 Krull dimension 213

Krull’s Principal Ideal Theorem 218
System of parameters 220

11 Principal ideal domains 225

Elementary properties. 226
Some general facts 231
Finitely generated modules 232
Lecture 1

Rings

Preliminary version 1.1 as of 2018-08-22 at 08:11 (typeset 3rd December 2018 at 10:03am)—Prone
to misprints and errors and will change.
2018-08-21 Added a few sentences in paragraph 1.2.
2018-08-22 Added some words about isomorphisms, paragraph 1.6.
2018-08-23 Added exersices 1.13 and 1.20. Section 1.3 rewritten in a simpler manner.
30/08/2018 Added a hint to exercise 1.15.
06/09/2018 Added exercise 1.16.

The starring role in commutative algebra is played by the commutative rings

and their ideals— they are even the main targets of the investigations. In
this chapter we become acquainted with the rings, and the ideals will be
introduced in the next chapter.
Commutative rings come in various flavours and arise from different
sources. Some are best though of as “number systems” like the ring Z of
integers and its well-known larger siblings the field of rationals Q, the field
of real numbers R, and the field of complex numbers C (this suite continues,
but in a non-commutative way; the next two members of the family being
the quaternions and the octonians), there are also some little brothers; the
fields Z{pZ of integral residue classes modulo primes p and the the other
finite fields Fq . The earliest systematic study of commutative rings was of var-
ious “generalized number systems”; certain subrings of the ring of algebraic
numbers. Already Gauss undertook such studies, but it really took off in the
nineteenth century with the work of Kronecker’s and Dedekind’s.
Other commutative rings resemble rings of functions on different kinds of
spaces, like continuous functions on topological spaces (with real or complex
values) or holomorphic functions in open domains in the complex plain, but
the rings most relevant in our context arise in algebraic geometry. These are
rings of polynomial functions on so-called algebraic varieties.
The development took a new direction around the middle of the twentieth
century, when mathematicians like Zariski and Weil strived for establishing a
sound foundation of algebraic geometry, and the recognition of the power of
algebraic geometric methods in number theory eventually lead to the happy
marriage of algebraic geometry and number theory — consummated by
Groethendieck and his invention of schemes.
8 ma4200—commutative algebra

1.1 Rings
1.1 Recall that a ring is an algebraic structure consisting of a set endowed with
two binary operations; an addition which makes A an abelian group, and
a multiplication. The multiplication is assumed to be distributive over the
addition, and in this course it will always be associative and commutative
(or at least almost always). There are of course many extremely interesting
non-commutative rings and non-associative rings, but this course is dedicated
to rings that are associative and commutative.
The sum of two elements will of course be denoted as a ` b, and the product
will we indicated in the traditional way by a dot or simply by juxtaposition;
that is, as a ¨ b or just by ab. The left distributive law asserts that apb ` cq “ ab ` bc,
and since rings for us are commutative, it follows that the right distributive law
pb ` cqa “ ba ` ca holds as well.
We shall also assume that all rings have a unit element; that is, an element
1 A such that 1 A ¨ a “ a for all members a of the ring. At most occasions the
reference to A will be dropped, and we shall write 1 for the unit element
whatever the ring is.
Example 1.1 The simplest of all rings are the ring Z of integers and the
rings Z{nZ of residue classes of integers modulo n. The traditional numbers
systems of real numbers R and complex numbers C are well-known rings. K

Zero divisors and nilpotents

1.2 Generally, ring-elements can behave quite differently than we are used to in
a classical setting. It might very well happen that ab “ 0 without neither a nor
b being zero. Such elements are called zero divisors. Be aware that the familiar Zero divisors (nulldivi-
cancellation law does not hold in a ring with zero divisors in that ab “ ac not sorer)

necessarily implies that b “ c. Rings without zero divisors are called integral
domains or, for short, domains. Integral domains (In-
Obviously, elements that are not zero divisors are called non-zero divisors, tegritetsområder)
Non-zero divisors (Ikke-
another name being regular elements. A regular element a has the virtue that nulldivisorer)
xa “ 0 implies that x “ 0 and can therefore be cancelled from equalities like Regular elements (Reg-
ab “ ac (the difference b ´ c being killed by a vanishes). ulære elementer)
For instance, in case n is a composite number, say n “ pq, the ring Z{nZ
has zero divisors; it holds true that pq “ 0, and p and q are both different from
zero (neither having n as a factor).
A more geometric example could be the ring of continuous functions on the
space X which is the union of the x-axis and the y-axis in the plane. On X the
function xy vanishes identically, but neither x nor y does (x does not vanish on
the y-axis and y not on the x-axis).
1.3 It might also happen that a powers of a non-zero elements vanish, i.e. one
has an “ 0 for some natural number n, but a ‰ 0. For instance, in the ring
 rings 9

Z{p2 Z one has p2 “ 0, but p ‰ 0. Such elements are called nilpotent. Rings Nilpotent elements
deprived of nilpotent elements, are said to be reduced. (nilpotente elementer)
Reduced rings (reduserte
ringer)
Units and fields

1.4 Division by non-zero elements is generally not possible in rings. For

instance, if p and q are two different primes in Z, the fraction p{q is not an
integer and does not lie in Z. Elements in a ring A that are invertible, i.e. ring-
elements a for which there is an element a´1 in A with aa´1 “ 1, are called
units. They form an abelian group under multiplication which we shall denote Units (enheter)
by A˚ .
Rings A all whose non-zero members are invertible; that is, which satisfy
˚
A “ Azt0u, are called fields. In fields division by non-zero elements can be Fields (kropper)
performed unconditionally.
Example 1.2 Well-known fields are the fields of rational numbers Q, of real
numbers R and of complex numbers C. If p is a prime number, the ring Z{pZ
of integers modulo p is a field usually denoted by F p . It is a finite field having
p elements. K

Examples

1.3 We do not assume that 1 ‰ 0 although it holds in all but one ring. The
exception is the so-called null-ring. If in a ring it holds that 0 “ 1, one has The null-ring (nullrin-
a “ a ¨ 1 “ a ¨ 0 “ 0, so zero will be the sole element. The only role the gen)

null-ring plays and the only reason not to throw it over board, is that it allows
significantly simpler formulations of a few results. It does not merit a proper
notation (well. . . , one alway has the alternative 0).

1.4 The set of polynomials Qrx1 , . . . , xr s in r variables x1 , . . . , xr with ratio-

nal coefficients is clearly a ring with the usual sum and product, as are the
set of real polynomials Rrx1 , . . . , xr s and the set of complex polynomials
Crx1 , . . . , xr s.

1.5 The complex rational functions in a variable x form a field Cpxq. The
elements are meromorphic functions in C expressible as the quotient ppxq{qpxq
of two polynomials p and q and q not being identically zero.

1.6 For any set X Ď Cr one may consider the set of polynomial functions on X; Polynomial functions
that is, the functions on X that are restrictions of polynomials in r variables. (polynomiale funskjoner)

They form a ring AC pXq under point-wise addition and multiplication.

1.7 Associated with any topological space X are the sets CR pXq and CC pXq of
of continuous functions on X assuming respectively real or complex values.
Point-wise addition and multiplication make them (commutative) rings. When
X has more structure than just a topology, there are further possibilities. Two
10 ma4200—commutative algebra

instances being the ring of smooth functions on a smooth manifold, and the
ring O pΩq of holomorphic functions in an open domain Ω of the complex
plane.

1.8 An example of a class of rings, important in algebraic number theory, is

? ?
the class of the quadratic extensions Zr ns “ t a ` b n | a, b P Z u, where n quadratic extensions
is any integer (positive or negative). These rings are contained in the field of (kvadratiske utvidelser)

complex numbers C and inherit their ring structure from C; they are closed
under addition (which is obvious) and multiplication:
? ? ?
pa ` b nqpa1 ` b1 nq “ paa1 ` nbb1 q ` pab1 ` a1 bq n.

Homomorphisms
When studying mathematical objects endowed with a certain structure—like
rings for instance—maps preserving the structure are fundamental tools.
Working with topological spaces one uses continuous maps all the time, and
linear algebra is really about linear maps between vector spaces. And of
course, the theory of groups is inconceivable without group homomorphisms;
that is, maps respecting the group laws. A new class of objects in mathemat-
ics is always accompanied by a new class of maps. This observation can be
formalized and leads to the definition of categories—see xxx. categories (kategorier)

1.5 In our present context the relevant maps are the so-called ring homomor-
phism, which also will be referred to as maps of rings or ring-maps. These are Ring homomorphisms
maps φ : A Ñ B between two rings A and B preserving all the structures (ringhomomorfier)

around; that is, the additive group structure, the multiplication and the unit
element 1. In other words, they comply to the rules

o φpa ` bq “ φpaq ` φpbq and φp0q “ 0;

o φpabq “ φpaqφpbq and φp1q “ 1.

The sum of two maps of rings is in general not a map of rings (it is additive,
but does not respect the multiplication) neither is their product (it respects
multiplication, but not addition), but of course, the composition of two com-
posable ring-maps is a ring-map.
1.6 A homomorphism φ : A Ñ B is an isomorphism if there is a ring homomor-
phism ψ : B Ñ A such that the two relations ψ ˝ φ “ id A and φ ˝ ψ “ idB Isomorphisms of rings
(isomorfier av ringer)
hold true. One most often writes φ´1 for the inverse map, and it is common
usage to call isomorphisms invertible maps. For φ to be invertible it suffices it
be bijective. Multiplication will then automatically be respected since when φ is
injective, φ´1 pabq “ φ´1 paqφ´1 pbq is equivalent to ab “ φpφ´1 paqφ´1 paqq , and
 rings 11

the latter relations is a consequence of φ respecting multiplication. Applying

φ´1 to φp1 A q “ 1B one sees that φ´1 p1B q “ 1 A , so the inverse map sends the
unit element to the unit element as well.

Examples

1.9 So-called evaluation maps are omnipresent examples of ring homomor-

phisms. To illustrate this concept, we pick a point a P Cn . Sending a polyno-
mial f to the value it assumes at a, gives a map Crx1 , . . . , xn s Ñ C, and by the
very definition of the ring structure of the polynomial ring this is a map of
rings.
Any ring A of functions—say with complex values—on any topological
space X possesses analogue evaluation maps. The operations in A being
defined point-wise the map f ÞÑ f pxq is a ring-map from A to C for any point
x P X.
1.10 Another series of well-known examples of ring-maps are the maps Z Ñ
Z{nZ that send an integer a to its residue class ras modulo the integer n.

Subrings and polynomial expressions

1.7 We begin by recalling the notion of a polynomial expression. Assume given

a ring A and sequence a “ pa1 , . . . , ar q of elements from A. For any multi-index
α “ pα1 , . . . , αn q; that is, a sequence on non-negative integers, one has the
monomial expression Monomial expression
α
aα “ a1 1 ¨ . . . ¨ aαnn . (monomiale uttrykk)

These expressions show an exponential behavior in that aα ¨ a β “ aα`β . A

polynomial expression in the ai ’s is just a finite linear combination of such Polynomial expressions
monomials. Frequently one wants to confine the coefficients to lie in a specific (polynomiale uttrykk)

subset S of A, and then one speaks about polynomial expressions with coefficients
in S. They are thus elements of A shaped like
ÿ ÿ α
sα ¨ aα “ sα ¨ a1 1 ¨ . . . ¨ aαnn ,
α α

where the summation extends over all multi-indices, and where the non-zero
coefficients are finite in number and confined to lie in S.
A successive application of the distributive law gives the classical formula
for the product of two polynomial expressions:
ÿ ÿ ÿ ÿ
p sα ¨ aα q ¨ p t β ¨ a β q “ p sα t β q ¨ aγ . (1.1)
α β γ α` β“γ
12 ma4200—commutative algebra

1.8 A subring B of A is a ring contained in A whose ring operations are in- Subrings (Underringer)
duced from those of A. Phrased differently, it is an additive subgroup con-
taining the unit element which is closed under multiplication; to be specific,
it holds that 0 P B and 1 P B, and for any two elements a and b belonging to
B, both the sum a ` b and the product ab belong to B. The intersection of any
number of subrings of A is a clearly subring.
Example 1.11 The integers Z is a subring of the rationals Q. K
1.9 Given a ring A and a subring B and a set of elements a1 , . . . , ar from A.
One constructs a subring Bra1 , . . . , ar s of A as the set of all polynomial expres-
sions ÿ α
bα ¨ a1 1 ¨ . . . ¨ arαr

where α “ pα1 , . . . , αr q runs through the multi-indices and the bα ’s are elements
from B, only finitely many of which are different from zero. It is straightfor-
ward to check, using the classical formula (1.1) above, that this subset is closed
under multiplication and hence is a subring of A (it is obviously closed under
addition). It is called the subring generated by the ai ’s over B, and is the small- Subrings generated by
est subring of A containing the ring B and all the elements ai . Common usage elements (underringer
generert av elementer)
is also to say that Bra1 , . . . , ar s is obtained by adjoining the ai ’s to B.
This construction works fine even for infinitely many ai ’s since each polyno-
mial expression merely involves finitely many of them. Thus there is a subring
Brai |i P Is for any subset tai uiP I of A.

Examples

1.12 Let n be an integer. The ring Zr1{ns “ t m{ni | i P N0 , m P Z u is a

subring of Q. The elements are the rational numbers whose denominator
is a power of n. More generally, if S is any set of integers, one may form
Zrn´1 | n P Ss, which is the subring of Q consisting of the rational numbers
whose denominator is a product of numbers from S.
Be aware that quite different sets S can give rise to the same subring. For
instance, when p1 , . . . , pr are the primes occurring in the prime factorization of
1
the integer n, it holds true that Zr1{ns “ Zrp´ ´1
1 , . . . , pr s.

1.13 The subring Crt2 , t3 s of Crts is a ubiquitous example in algebraic geome-

try; it is the coordinate ring of a so-called cusp and consists of all polynomials
whose first derivative vanishes at the origin; of phrased differently, the polyno-
mials without a linear term.
1.14 The subring Crx, 1{xs of the rational function field Cpxq consists of ele-
ments of the form ppx´1 q ` c ` qpxq where p and q are polynomials vanishing at
the origin and c a complex constant.

K
 rings 13

The prime ring and the characteristic

1.10 Every ring has a canonical subring called the prime ring. The unit element
1 in A generates an additive cyclic subgroup of A whose elements are just
sums of 1 or ´1 a certain number of times; that is, they are shaped like n “
1 ` . . . ` 1 or n “ ´1 ´ . . . ´ 1. This subgroup is obviously closed under
multiplication and is hence a subring. It is called the prime ring of A. The prime ring (primrin-
As is well known, a cyclic group is either finite and isomorphic to Z{nZ for gen)

some positive integer n, or it is infinite and isomorphic to Z. The prime ring

is therefore either isomorphic to one of the rings Z{nZ or to Z. In the former
case the integer n is called the characteristic of A, in the latter case one says The characteristic of a
that A is of characteristic zero. So, in any case, the characteristic of a ring A is a ring (karakteristikken til
en ring)
non-negative integer attached to A.
1.11 Any field contained in A contains the prime ring. Hence, if A contains a
field, the characteristic is either prime or zero. In case it is prime, the prime
ring equals the field F p , and in case the characteristic is zero, the ring A
contains Q as well. We say that F p respectively Q is the prime field of A. The prime field (primkrop-
pen)

Algebras
Frequently when working in commutative algebra there are “coefficients”
around; that is, one is working over a “ground ring”. So the most natural
objects to work with are perhaps not rings, but the so-called algebras.
1.12 The notion of an algebra is a relative notion involving two rings A and B.
To give a B-algebra structure on A is just to give a map of rings φ : B Ñ A. One Algebras (Algebraer)
may then form products φpbq ¨ a of elements a from A with elements of the form
φpbq. The map φ (even though it is an essential part of the B-algebra structure
of A) is often tacitly understood and suppressed from the notation; one simply
writes b ¨ a for φpbq ¨ a. Later on, when we have introduced modules, a B-algebra
structure on a ring A will be the same as a B-module structure on A.
Example 1.15 Every ring has a canonical structure as a Z-algebra (defined as
in Paragraph 1.10 above). The class of algebras is therefore a strict extension of
the class of rings. Since a ring is an algebra over any subring, over-rings give a
large number of examples of algebras. K
1.13 Faithful to the principle that any new type of objects is accompanied by
a corresponding new type of maps; one says that a map of rings φ : A Ñ A1
between two B-algebras is an B-algebra homomorphism1 if it respects the action Algebra homomorphisms
of B; in other words, it holds true that φpb ¨ aq “ b ¨ φpaq for all elements a P A (algebrahomomorfismer)
1
Or any morphological
and b P B. Composition of two composable B-algebra homomorphisms is derivative thereof, like
a B-algebra homomorphism so that the B-algebras form a category denoted map of B-algebras or
Alg B . B-algebra-map etc.

1.14 One says that A is finitely generated over B, or is of finite type over B, if Finitely generated
A “ Bra1 , . . . , ar s for elements a1 , . . . , ar from A. algebras (endeliggenererte
algebraer)
Finite type algebras
(algebraer av endelig
type)
14 ma4200—commutative algebra

Example 1.16 A note of warning might be appropriate, algebra structures

can be delusive. Every ring is of course an algebra over itself in a canonical
way (the algebra structure is given by the identity map), but there can be other
unorthodox ways A can be an A-algebra. A simple example to have in mind
is the field C of complex numbers which has an alternative algebra structure
induced by complex conjugation. In this structure z acts on w as z̄ ¨ w.
The two structures are not isomorphic as C-algebras although the underly-
ing rings are the same. A good try for an isomorphism would be the identity
map, but it does not respect the two algebra-structures2 . 2
Of course, it holds
Another examples of unorthodoxy are funished by the Frobenius homomor- true that idC pzwq
equals z idC w and not
phism of rings of positive characteristic (see Exercsise 1.7 below). K z̄ idC w

Problems

1.1 Find all nilpotent and all zero-divisors in Z{72Z. What are the units? ˇ

1.2 Generalize the previous exercise: Let n be a natural number. Determine ˇ

nilpotents, zero-divisors and units in Z{nZ.

1.3 Assume that A is finite integral domain. Show that A is a field. ˇ

1.4 Show that prime ring is the smallest subring; i.e. it is contained in all other
subrings of the given ring.

1.5 Convince yourself that the binomial theorem persists being true in any
commutative ring; that is, check that your favourite proof still holds water.

1.6 Show that the sum of two nilpotent elements is nilpotent. Hint: You can
rely on the binomial theorem.

1.7 (The Frobenius homomorphism.) Let A be a ring of positive characteristic p.

Show that the relation
pa ` bq p “ a p ` b p
holds true for all a, b P A. Hence there is a ring homomorphism A Ñ A
sending a to a p . It is called the Frobenius homomorphism. Hint: The binomial
coefficients pr have p as factor when 1 ă r ă p.
` ˘

1.8 Show that any intermediate ring Z Ď A Ď Q is of the form A “ Zrp´1 |p P

Ss for some set S of primes.

1.9 Let φ : A Ñ B be a map of rings. Show that φ induces a group homomor-

phism mapping A˚ into B˚ .
?
1.10 Let n be a natural number. Show that an element x P Zr ´ns is a unit ˇ
if and only if }x} “ 1 (where }x} denotes the ordinary absolute value of the
?
complex number x), and use this to determine the units in Zr ´ns.
 rings 15

1.11 (The Eisenstein integers.) Let ω be the cube root ω “ e2πi{3 of unity. Show ˇ
that the subring Zrωs of C is given as

Zrωs “ t n ` mω | n, m P Z u.

Determine the group of units Zrωs˚ . Hint: It holds true that ω 2 ` ω ` 1 “ 0.

1.12 Assume that a is a nilpotent element of the ring A. Show that 1 ` a ˇ

is invertible. More precisely: If an “ 0, the inverse is given as p1 ` aq´1 “
1 ´ a ` a2 ´ . . . ` p´1qn´1 an´1 . Conclude that if u is a unit and a nilpotent, then
u ` a is invertible. Hint: Use the good old formula for the sum of a geometric
series.

1.2 Polynomials
We are well acquainted with polynomials with real or complex coefficients;
we met them already during the happy days at school. They were then intro-
duced as functions depending on a of real (or a complex one if you went to a
French school) variable whose values were given by a polynomial expressions.
We shall in this section introduce polynomials with coefficients in any (com-
mutative) ring A. The point of view will necessarily be formal and without
reference to functions, and there will be more than just one variable.
1.15 In an earlier paragraph we met polynomial expressions in a set of ring el-
ements. In the present situation where there is no surrounding ring, we must,
as signalled above, proceed in a formal way. A polynomial in the variables Polynomials (Polynomer)
x1 , . . . , xr is defined as a formal sum
ÿ α
f px1 , . . . , xn q “ aα x1 1 ¨ . . . xnαn , (1.2)
α

where the summation extends over all multi-indices α “ pα1 , . . . , αn q with the
αi ’s being non-negative integers, and where the coefficients aα are elements
from the ground ring A, only finitely many of which are non-zero. Do not
speculate much3 about what the term “formal sum” means, the essential 3
Or do Exercise 1.19
point is that two such “formal sums” are equal exactly when corresponding below where the more
general construction of
coefficients agree. the so-called monoidal
α algebras is described in
1.16 The “pure” terms aα x1 1 ¨ . . . xnαn occurring in (1.2) are called monomials. The
α a precise manner.
abbreviated notation x α “ x1 1 ¨ . . . xnαn is convenient and practical. The degree of
α
ř
a non-zero monomial aα ¨ x is the sum i αi of the exponents, and the highest Monomials (Monomer)
degree of a non-zero monomial term in a polynomial, is the degree of the The degree of a poly-
polynomial. Non-zero constants are of degree zero, but the zero polynomial nomial (Graden til et
polynom)
is not attributed a well-defined degree—it is rather considered to be of any
degree (it equals 0 ¨ x α for any α).
16 ma4200—commutative algebra

A polynomial is said to be homogenous if all its monomial terms are of the Homogenous polynomials
same degree. For example the polynomial x2 y ` z3 is homogeneous of degree (Homogene polynomer)

three whereas x2 y ` z2 is not; it is still of degree three, but not homogeneous.

Every polynomial may be expressed as a sum of homogenous polynomi-
als of different degrees—just recollect the homogenous terms with the same
degree—and these are called the homogenous components of f . They are unam- Homogenous components
biguously associated with f . of a polynomials (Homo-
gene komponenter til et
1.17 Adding two polynomials is simply done term by term, and neither is polynom)
there any hocus-pocus about multiplying them. The good old pattern is
followed where
ÿ ÿ ÿ ÿ
aα x α ¨ bβ x β “ p aα bβ qx γ . (1.3)
α β γ α` β“γ

In particular, the product of monomials comply to the exponential law x α x β “

x α`β ; with this in mind, the content of formula (1.3) is that the product is
bilinear over A.
Equipped with the operations just described the set Arx1 , . . . , xr s of polyno-
mials in the variables x1 , . . . , xr becomes a ring. Of course, there are axioms to
be verified; a tedious and uninteresting process without even small obstacles,
so we voluntarily skip it (if you suffer from mathematical paranoia, feel free to
do the checking).
ř ř
Problem 1.13 Let f “ α aα x α and g “ α bα x α be two polynomials with
coefficients from the rig A. Show that deg f g ď deg f ` deg g (recall the
convention that 0 has all degrees). Show that equality holds when A is an
integral domain. Give examples that strict inequality holds M

1.18 There is a slight difference between a polynomial and a polynomial func-

tion. Over finite rings, like Z{nZ for instance, different polynomials can give
rise to identical polynomial functions. Simple examples being polynomials in
F2 rts; that is, polynomials in one variable over the field F2 with two elements.
For instance, such polynomials without constant term and with an even num-
ber of non-zero terms will vanish identically as a function on F2 . Over infinite
fields however, the two notions coincide.

The universal mapping property

1.19 The polynomial ring Arx1 , . . . , xr s has a so-called universal mapping prop-
erty; one may freely assign values to the variables to obtain homomorphisms. A universal mapping
property (En universell
avbildningsegenskap)

Proposition 1.20 (The Universal Mapping Property) Let A be a ring.

Assume given a sequence b1 , . . . , br of elements from an A-algebra B. Then there
is a uniquely determined algebra homomorphism φ : Arx1 , . . . , xr s Ñ B such that
φpxi q “ bi for 1 ď i ď r.
 rings 17

ř
Proof: A polynomial p is given as p “ α aα x α . Since the coefficients aα are
ř α
unambiguously determined by p, setting φppq “ α aα β 11 ¨ . . . ¨ brαr gives a
well-defined map which easily is seen to be additive. Since a relation like the
one in (1.3) is universally valid in commutative rings, φ respects multiplication
as well, and we have an algebra homomorphism. o

Example 1.17 The universal mapping property is a very particular property

most algebras do not have. For instance, the algebra Crx2 , x3 s from exam-
ple 1.13 on page 12 does not have it. That algebra has the generators x2 and
x3 , and the equality px2 q3 “ px3 q2 imposes a constraint on the values homo-
morphisms can assume on the two generators; it must hold true that φpx2 q3
coincides with φpx3 q2 (even though there is no such thing as φpxq). K

Two further constructions

There are two further constructions closely related to the construction of the
polynomial rings.
1.21 One may consider polynomial expressions over A in infinite number
of variables x1 , x2 , . . . , xn , . . ., although each polynomial merely involves
finitely many of the variables. For every n the polynomial ring Arx1 , . . . , xn s
is obviously contained in Arx1 , . . . , xn`1 s and the polynomial rings thus form
a nested sequence of rings. The polynomial in countably many variables
Arx1 , x2 , . . .s is just the union of all these.

Problem 1.14 Convince yourself that the universal mapping property holds ˇ
even for polynomial rings in infinitely many variables. M

1.22 The second type of ring we have in mind, are the rings of formal power
series. A formal power series is an expression like in (1.2) Rings of formal power
series (Ringen av formelle
ÿ potensrekker)
f px1 , . . . , xn q “ aα x α1 ¨ . . . xnαn ,
α

except that the sum is not require be to finite. Addition is done term by term,
and the multiplication is defined by formula (1.3), which is legitimate since the
expression for each coefficient involves only finitely many terms. The formal
power series ring is denoted AJx1 , . . . , xn K.

Problems

1.15 Show that a polynomial f pxq “ i ai xi in Arxs is invertible if and only if

ř
ˇ
a0 is invertible and all the other coefficients are nilpotent. Hint: Assume that
f pxq “ 1 ` a1 x ` . . . ` an x n is invertibel with inverse f pxq´1 “ 1 ` b1 ` . . . `
bm x m . Show that ain`1 bm´i “ 0 for 0 ď i ă n ` m. Conclude that an is nilpotent.
18 ma4200—commutative algebra

1.16 Let A be a reduced ring. Show that group of units in the polynomial ring
Arxs equals A˚ .

1.17 Assume that k is a field. Show that krts and krt, 1{ts are not isomorphic as ˇ
rings.

1.18 Assume that k is field. Show that a the number of zeros of a non-zero
polynomial in krts is less than the degree. Show that if two polynomials in
krts define the same function on k and k is infinite, then they coincide as
polynomials. If k is finite, exhibit a polynomial that vanish identically on k.

1.19 (Monoidal algebras.) In this exercise the definition of polynomial rings is ˇ

made precise and generalized.
Let G be commutative monoid4 written additively. As an abelian group 4
A monoid is a set
À
ArGs is the direct sum of copies of A indexed by G; that is, ArGs “ αPG A. endowed with an asso-
ciative binary operation
The elements are sequences p “ ppα qαPG with finite support, and addition is that has a neutral el-
defined component-wise. Introduce a product on ArGs by the formula ement, in which the
cancellation law holds.
ÿ The set N0 of non-
pp ¨ qqα “ p β ¨ q α´ β negative integers and
βP G its cartesian products
N0r are arch-examples
Let x α denote the sequence all whose components are zero apart from the one of monoids.
in the slot with index α which equals one.
a) Show that x α form an additive basis for ArGs.
b) Show that x α ¨ x β “ x α`β .
c) Show that p α pα x α qx β “ α pα x α`β . Verify that ArGs is a ring.
ř ř

d) Show that ArN0r s » Arx1 , . . . , xr s.

1.20 Let tAi u be a collection of subrings of the ring A. Prove that the intersec-
Ş
tion iP I Ai is a subring.
Give examples of two subrings A1 and A2 of A such that their union is not
a subring. Assume that the collection has the property that any two rings from
Ť
the collection are contained in a third. Prove that in that case the union iP I Ai
is a subring.

1.3 Direct products and idempotents

As a motivation consider the disjoint union X Y Y of two topological spaces
X and Y. Giving a continuous function on X Y Y is of course the same as
giving one continuous function on X and one on Y. Therefore the ring of
continuous functions CR pX Y Yq decomposes as the product CR pX Y Yq “
CR pXq ˆ CR pYq and both addition and multiplication are given component-wise.
 rings 19

This indicates that in the interplay between geometry and rings, direct product
of rings correspond to disconnected spaces.
Below we define the direct product of a collection of rings regardless of the
cardinality of the collection and introduce the notion of idempotent elements
(elements e such that e2 “ e). Multiplication by idempotents are projection
operators (equal to their squares) and serve to decompose rings (and later on
modules) into direct products.
The archetype of an idempotent function is the characteristic function
eX of a connected component, say X, of a topological space Z; that is, the
function that assumes the value one on X and zero on the rest of Z. Since
X is a connected component of Z, this function is continuous and of course,
e2X “ eX . Moreover, the restriction f |X of any function f equals eX f , or put
more precisely, eX f is the restriction f |X extended by zero to the entire space
Z. Anyhow, in this way the set eX CR pZq is a ring naturally identified with
CR pXq with the idempotent eX corresponding to unit element in CR pXq. The
lesson learned is that idempotents are algebraic counterpart to the geometric
connected components.

Direct products of rings

We start out by considering two rings A1 and A2 . The cartesian product
A “ A1 ˆ A2 consisting of the pairs pa1 , a2 q is a ring when equipped with the
pairwise operations. The underlying additive group is the direct product of
the underlying groups of the two rings, and the product is given as

pa1 , a2 q ¨ pa11 , a12 q “ pa1 ¨ a11 , a2 ¨ a12 q.

The unit element is the pair p1, 1q, and the two projections πi : A Ñ Ai are ring
homomorphisms. Moreover, the direct product possesses two special elements
e1 “ p1, 0q and e2 “ p0, 1q, which satisfy ei2 “ ei and e1 e2 “ 0. The sets ei A
equal respectively A1 ˆ t0u or t0u ˆ A2 , and are, with a liberal interpretation5 , 5
Since the unit element
subrings of A isomorphic to respectively A1 and A2 . p1, 1q does not lie in
either Ai , they are
1.23 To generalize what we just did for a pair of rings, let tAi uiP I be any collec- properly speaking not
subrings even though
tion of rings, which can be of any cardinality. In our context it will mostly be
ś they are closed under
finite, but occasionally will be countable. The direct product iP I Ai has as un- both addition and
derlying additive group the direct product of the underlying additive groups multiplication.

of the Ai ’s. The elements are “tuples” or strings pai qiP I indexed6 by I whose 6
The reference to
i-th component ai belongs to Ai , and the addition of two such is performed the index set I will
frequently be dropped
component-wise. The same is true of the multiplication, also performed com- and the strings written
ponent for component; that is, it holds true that pai q ¨ pbi q “ pai ¨ bi q. The ring pai q.
axioms can be checked component-wise and thus come for free.
Ť
Interpreting tuples a “ pai q as maps a : I Ñ iP I Ai , the ring operations
of the direct product are just the point-wise operations. The unit element, for
instance, is the “constant7 ” function that sends each index i to 1. 7
Why the quotation
marks?
20 ma4200—commutative algebra

ś
1.24 The projections πi : iP I Ai Ñ Ai are ring homomorphisms (this is
just another way of saying that the ring operations are defined component-
wise) and enjoy the following universal property. Given any ring B and any B
φ
/ śi A i
collection φi : B Ñ Ai of ring homomorphisms, there is an unambiguously
ś πi
defined map of rings φ : B Ñ iP I Ai such that φi “ πi ˝ φ for all i P I: $ 
φi

Indeed, this amounts to the map given by φpxq “ pφi pxqqiP I being a ring Ai
homomorphism.

Idempotents

1.25 In any ring A an element e satisfying e2 “ e is said to be idempotent, and if Idempotent elements
f is another idempotent, one says that f and g are orthogonal when f g “ 0. The (idempotente elementer)
Orthogonal idempotents
element 1 ´ e is always idempotent when e is and is orthogonal to e as shown (ortogonale idempotenter)
by the little calculations

p1 ´ eq2 “ 1 ´ 2e ` e2 “ 1 ´ 2e ` e “ 1 ´ e,
ep1 ´ eq “ e ´ e2 “ e ´ e “ 0.

The subset eA “ t ea | a P A u is a ring with e as a unit element. Indeed

ea ¨ eb “ e2 ab “ eab,

so eA is closed under multiplication and trivially it is closed under addition as

well; finally
e ¨ ea “ e2 a “ e,
and e serves as the unit element.
It is common usage to count the unit element and zero among the idempo-
tents; they are called the trivial idempotents. The trivial idempotents
(De trivielle idempoten-
1.26 We saw above that in direct product A1 ˆ A2 there appears two natural tene)
defined idempotents. Conversely, let A a ring. To any family tei u1ďiďr family
of mutually orthogonal idempotents that add up to 1 there corresponds a
decomposition of A as direct product:

Proposition 1.27 Let e1 , . . . , er be pairwise orthogonal idempotents in a ring A

ř
and assume that i ei “ 1. Then each set ei A is a subring in the restricted sense, and
the association x Ñ pe1 x, . . . , er xq gives an isomorphism of rings

A
» / ś ei A.
i

The projection onto ei A is realized as multiplication by ei .

Proof: To begin with, we verify that the map in the proposition, call it φ, is a
ring homomorphisms. So let x and y be two elements from A. The ei ’s being
idempotents we find

φpxqφpyq “ pei xqpei yq “ pei ei xyq “ pei xyq “ φpxyq.

 rings 21

Thus φ respects the multiplication, and moreover φ is clearly additive. The

unit element 1 maps to the string pei q which is the unit element in the product
since each ei serves as the unit element in ei A.
ř
Now, we have supposed that the e’s add up to one; that is 1 “ i ei . Hence
ř
x “ i ei x, and it ensues that φ injective; indeed, that φpxq “ pei xq “ 0 means
that each ei x “ 0.
Finally, let us check that φ is surjective. Given an element pei xi q in the
ř
product, and put x “ i xi . Using that the ei ’s are mutually orthogonal, we
ř
find e j x “ i e j ei x “ e j x, and we see that x maps to the given element pei xi q. o

Problems

1.21 Determine the idempotents in Z{12Z and in Z{36Z.

1.22 (The p-adic integers.) Let p be a prime number and let ρn : Z{pn`1 Z Ñ
Z{pn Z be the reduction map that sends the residue class ras pn`1 of an integer
a modulo pn`1 to the residue class ras pn of a modulo pn . Let Z p be the subset
of the direct product nPN Z{pn Z whose members are strings pan qnPN such
ś

that ρn pan`1 q “ pan q.

a) Show that Z p is a subring of the product.
b) Show that the map Z Ñ Z p sending an integer a to the string pras pn q is an
injective ring homomorphism.
c) Show that Z p is an integral domain,
d) Let π : Z p Ñ Z{pZ be induced by the projection from nPN Z{pn Z onto
ś

Z{pZ. Show that an element a in Z p is invertible if and only if πpaq ‰ 0.

M
Lecture 2

Ideals

The ideals were first defined by Richard Dedekind in 1876, but the name comes
from the so called “ideal numbers” of Ernst Kummer which he introduced in a
series of papers around 1847.
Working with rings of integers in algebraic number fields, the algebraists of
that period realized that analogues of the Fundamental Theorem of Arithmetic
does not always hold in such rings. Recall that this theorem asserts that any
integer is a product n “ p1 ¨ . . . ¨ pr of signed primes and the factors are
Richard Dedekind
unique up to order and sign—changing the order of the factors does not (1831–1916)
affect the product, and changing the sign of one factor can be compensated by German mathematician
simultaneously changing the sign of another.
It is not too complicated to show that in a vast class of rings, including
the rings of algebraic numbers above, any element can be expressed as a
product of irreducible elements; that is, elements that can not be factored
further (they can of course always be altered by a unit, but that is not an
honest factorization). However the point is, that these factors are not unique
(up to order and units) in general.
The classical example omnipresent in text books, is the factorisazion 2 ¨ 3 “ Ernst Eduard Kummer
? ? ?
p1 ` i 5qp1 ´ i 5q in the ring Zri 5s. The four involved numbers are all (1810–1893)
irreducible and no two of them related by units. German mathematician

The ideals came about to remedy this fault and, in fact, in certain rings
called Dedekind rings, the situation can be salvaged; there is a factorization
theorem of ideals replacing the Fundamental Theorem of Arithmetic. Hence
the name ideals, they were “the ideal numbers”
Dedekind rings are however a very restricted class of rings, and today ide-
als play an infinitely wider role than just being “ideal numbers”. In algebraic
geometry for instance, they appear as the sets of polynomials in krx1 , . . . , xr s
vanishing along a subset of kr , and this is the clue to the coupling between
algebra and geometry.

2.1 Ideals
24 ma4200—commutative algebra

2.1 Let A be a ring. An additive subgroup a of A is called an ideal if it is closed Ideals (Idealer)
under multiplication by elements from A. That is, a satisfies the two following
requirements; the first merely being a rephrasing that a is a subgroup.

o If a P a and b P a, then a ` b P A and 0 P a;

o If a P A and b P a, then ab P a.

Both the trivial additive subgroup 0 and the entire ring satisfy these require-
ments and are ideals, although special ideals. In many texts the ring when
considered an ideal, is denoted by p1q.
2.2 An ideal a is said to be a proper ideal if it is not equal to the entire ring. This Proper ideals (Ekte
is equivalent to no member of a being invertible. Indeed, if a P a is invertible idealer)

one has b “ ba´1 ¨ a P a for any b P A; and if a “ A, of course, 1 P a.

From this observation ensues the following characterization of fields in
terms of of ideals:

Proposition 2.3 A ring A is a field if and only if its only ideals are the zero ideal
and A itsef.

Proof: We just saw that an ideal a equals A precisely when a X A˚ ‰ H. If

A is a field, A˚ “ Azt0u, and any ideal, apart from the zero ideal, meets A˚ .
The other way round, any non-zero and proper ideal must contain a non-zero
element, which cannot be invertible, and consequently A is a not a field. o

Examples

2.1 The subset nZ of Z consisting of all multiples of the integer n is an ideal; a

so-called principal ideal. The ideal nZ is frequently written pnq.

2.2 For any subset S Ď Cr be a subset the polynomials in Crx1 , . . . , xr s vanish-

ing on S form an ideal.

Operations on ideals—the lattice of ideals

2.4 The set I pAq of ideals in the ring A has—in addition to being partially
ordered under inclusion—a lot of structure. One may form the intersection
Ş
iP I ai of any family tai uiP I of ideals. It is easily seen to be an ideal, and is the
largest ideal contained in all the ai ’s. Likewise, one has the notion of the sum of
a family of ideals. It is the ideal consisting of all finite sums of elements from
the ai ’s:
ÿ
ai “ t a1 ` . . . ` ar | ai P ai , r P N u,
iP I
 ideals 25

and is the smallest ideal containing all the ai ’s. So I pAq is what one technically
calls a complete lattice; every subset of I pAq has a greatest lower bound (the
sum) and a smallest upper bound (the intsersection). It is the lattice of idealsin The lattice of ideals
A. (Ideallattiset(???))

2.5 A construct similar to the sum of a family of ideals is the ideal generated
by a set of elements tai uiP I from A. It will be denoted pai |i P Iq, or in case the
set S “ ta1 , . . . , ar u is finite, the alternative notation pa1 , . . . , ar q is common us-
age. Its members are all finite linear combinations of the ai ’s with coefficients
from the ring A; that is, it holds that
ř
pai |i P Iq “ t iP J ci ai | ci P A, J Ď I finite u.

The elements ai are called generators. Ideals which are generated by finitely Generators (Generatorer)
many elements are naturally called finitely generated. An ideal generated by Finitely generated ideals
a single element, is called a principal ideal; it consists of all multiples of the (Endeliggenererte idealer)
Principal ideals (Hoved-
generator. If a is the generator, it is denoted by paq or by aA, and it holds that idealer)
paq “ t c ¨ a | c P A u.
2.6 The product of two ideals a and b is the ideal generated by all products of The product of ideals
one element from a and one from b; that is, the product ab is formed of all (Produktet av idealer)

finite sums of such products:

ab “ t a1 b1 ` . . . ` ar br | ai P a, bi P b, r P N u.

2.7 The last operation we offer is the formation of the transporter between two The transporter (Trans-
ideals. Some texts call it the quotient of the two deals—however, this term portøren)

should be reserved for another construction we shortly come to, and hence
should be avoided. So let a and b be two ideals in A. We define the transporter
pa : bq to be set of elements which on multiplication send b into a; that is

pa : bq “ t x P A | xb Ď a u.

It is easily seen to be an ideal. In the particular case that a “ p0q and b is a

principal ideal, say b “ paq, the transporter p0 : aq (an immediate simplification
of the overloaded expression pp0q : paqq) coincides with the annihilator of a; that The annihilator of an
is element (Annihilatoren til
et element)
p0 : aq “ Ann a “ t x | xa “ 0 u.

Examples

2.3 In Z it holds that 100 : 10 “ p10q. More generally if a and b are elements
` ˘

from the ring A and b is a non-zero divisor, one has pab : bq “ paq. Indeed,
xb “ yab is equivalent to x “ ya since cancellation by b is allowed b being
a non-zero divisor. If b is a zero-divisor it anyhow holds that pab, bq “ paq `
Ann b.
26 ma4200—commutative algebra

2.4 In the polynomial ring Crx, ys it holds that pxy, y2 q : px, yq “ pyq.
` ˘
` ˘
Clearly pyq is contained in pxy, y2 q : px, yq . For the converse inclusion
assume that f x “ gxy ` hy2 where f , g and h are polynomials in Crx, ys. Since
x divides the terms f x and gxy, it divides hy2 as well, and by cancelling x, we
infere that f “ gy ` h1 y with h1 P Crx, ys; that is, f P pyq.

2.5 In Z{40Z one has Ann 2 “ p20q, that Ann 4 “ p10q and that Ann 20 “ p5q.

Functorially
A map of rings φ : A Ñ B induces two maps between the ideal-lattices I pAq
and I pBq, one in a covariant and one in a contravariant way. One can move
ideals forward with the help of φ and the the usual inverse image construct is
a way move ideals backwards along φ. The new ideals are in some texts either
called extensions or contractions in the respective case. Extensions of ideals
(Utvidlese av ideal)
2.8 We start with the contravariant one. The inverse image φ´1 pbq
of an ideal Contractions of ideals
b in B is an ideal in A; indeed, φpabq “ φpaqφpbq belongs to b whenever φpbq (Tilbaketrekning)
does, and this gives rise to a map φ´1 : I pBq Ñ I pAq. Obviously it preserves
inclusions and takes intersections to intersections since the pullback of sets
respects intersections in general. Sums and products of ideals however, are not
generally preserved. One has

o φ´1 paq X φ´1 pbq “ φ´1 pa X bq,

o φ´1 paq ` φ´1 pbq Ď φ´1 pa ` bq,

o φ´1 paq ¨ φ´1 pbq Ď φ´1 pa ¨ bq,

but equality in the last two does not hold in general.

In the frequently occurring case that A is a subring of B one usually sup-
presses the reference to the inclusion map and uses the natural notation a X A
for the “pullback” of an ideal a. We shall here give an example showing that
the inclusion in the second relation can be strict, but postpone giving an
example for the last inclusion (see Example 2.8 below on page 30).
Example 2.6 A simple example of strict inclusion in the second relation
above is the diagonal map δ : A Ñ A ˆ A sending a to pa, aq. The two ideals
b “ t p0, aq | a P A u and b1 “ t pa, 0q | a P A u are both pulled back to the zero
ideal, but since b ` b1 “ A ˆ A, their sum is pulled back to the entire ring A. K

2.9 We then come to the covariant construction. If a is an ideal in A, the image

φpaq is not necessarily an ideal in B unless φ is surjective. A stupid example
can be the image of any ideal in Z under the inclusion Z Ď Q. The ideal
generated by φpaq however is, and we shall usually denote this ideal by φpaqB
 ideals 27

or simply by aB. This induces a map I pAq Ñ I pBq. Inclusions are preserved,
and one easily verifies the following relations
o φpa ¨ bqB “ pφpaqBq ¨ pφpbqBq,

o φpa ` bqB “ φpaqB ` φpbqB,

o φpa X bqB Ď φpaqB X φpbqB.

The inclusion in the last line may be strict, see Example 2.9 on page 30 below.

Problems

2.1 Let a and b be ideals in a ring A. Show that the relations a ¨ b Ď a X b and ˇ
pa X bq2 Ď a ¨ b hold. Show by giving examples that there might be a strict
inclusion in both cases.

2.2 Assume that a and b are ideals in a ring A satisfying a ` b “ p1q. Show that ˇ
a ¨ b “ a X b.

2.3 Let a,b and c be ideals in the ring A. Show that apb ` cq “ ab ` ac. Show
that a X pb ` cq Ď a X b ` a X c, and by exhibiting an example, show that the
inclusion can be strict.

2.4 Verify the equalities in paragraphs 2.8 and 2.9.

2.5 Let tai u be a collection of ideals in the ring A. Show that for any ideal b it ˇ
Ş Ş ř Ş
holds true that p iP I ai : bq “ iP I pai : bq and that pb : iP I ai q “ iP I pb : ai q.

2.6 Show that any ideal in the ring Z of integers is generated by one element,
which is unique up to sign.
` ˘
2.7 Let m and n be two integers. Show that pnq ¨ pmq “ pnq X pmq ¨ pn, mq.

2.8 Consider the two ideals a “ p144q and b “ p24q in Z . Describe pa : bq.
In general, describe pn : mq in terms of the prime factorizations of the two
integers n and m.

2.9 Let krx, ys be the polynomial ring in the variables x and y over the field k, ˇ
and let m be the ideal generated by x and y; that is m “ px, yq. Let n denote a
natural number.
a) Exhibit a set of generators for the power mn .
b) Let µ and ν be two natural numbers. Show that mn Ď px µ , yν q for n suffi-
ciently large. What is the smallest n for which this holds?
? ?
2.10 Let A “ Zr 2, 3s. Show that as an abelian group A free of rank four ˇ
and exhibit a basis. Show that the underlying abelian groups of the principal
? ?
ideals p 2q and p 3q both are of rank four. Exhibit additive bases for both.

M
28 ma4200—commutative algebra

2.2 Quotient rings and kernels

In one way ideals play the same role in the category of rings as normal sub-
groups do in the category of groups. They are precisely the subobjects that
appear as kernels of homomorphisms, and consequently, the ones that can be
factored out.
2.10 By definition the kernel of a ring homomorphism φ : A Ñ B is the kernel Kernels of ring homomor-
of φ considered a mapping between the underlying additive groups; that phisms (Kjernen til en
ringavbildning)
is, the subset of elements mapping to zero, or written in symbols one has
ker φ “ t a P A | φpaq “ 0 u. If a P ker φ and b P A, we find

φpa ¨ bq “ φpaq ¨ φpbq “ 0 ¨ φpbq “ 0,

and we can conclude that ab P ker φ. Hence the kernel ker φ is an ideal.
2.11 To see that any ideal is a kernel, one introduces the concept of quotient
rings. An ideal a in A being an additive subgroup, there is a quotient group Quotient rings (Kvotient-
A{a. It consists of the residue classes ras “ a ` a of elements in A, and the sum ringer)

of two such, say ras and rbs, equals ra ` bs. To put a ring structure on A{a we
simply define the product of two classes ras and rbs as

ras ¨ rbs “ ra ¨ bs “ a ¨ b ` a.

Some checking is needed; the most urgent one being that the product only
depends on the residue classes ras and rbs and not on the representatives a and
b. This is encapsulated in the formula

pa ` aq ¨ pb ` aq “ a ¨ b ` a ¨ a ` b ¨ a ` a ¨ a Ď a ¨ b ` a.

It is left to the students to verify that this product comply with the associative
and the distributive laws. Finally, by definition of the ring operations in A{a,
the quotient map π : A Ñ A{a is a map of rings whose kernel equals the given
ideal a.
Example 2.7 It is appropriate to mention what quotients by the two “extreme”
ideals are. The quotient A{a equals A if and only if a is the zero-ideal, and it
equals1 the null-ring if and only if a “ A. K 1
This examplifies
what purpose the null-
2.12 The quotient ring A{a together with the quotient map π : A Ñ A{a enjoys ring serves; it allows
a so-called universal property—the rather pretentious notion “solves a universal a general existence
theorem (avoiding the
problem” is also common usage—which is convenient way of characterizing hypothesis a ‰ A).
many types of mathematical objects. The origin of the technique is found in A universal property (En
category theory where objects not always have “elements” and one relies on universell egenskap)

“arrows” to express properties.

Any map of rings φ : A Ñ B that vanishes on a; that is, which satisfies
a Ď ker φ, factors in a unique way through the quotient A{a. In other words,
there is a unique ring-map ψ : A{a Ñ B such that φ “ ψ ˝ π. Indeed, since
 A
π
/ A{a
ideals
ψ
29
φ

B

φpaq “ 0, the map φ is constant on every residue class ras “ a ` a, and we put
ψprasq equal to that constant value. This value is forced upon ψ, so ψ is unique,
and it is a ring-map since φ is. We have proven:

Proposition 2.13 (The Factorization Theorem) Given an ideal a in the

ring A. A map of rings A Ñ B vanishes on a if and only if it factors through the
quotient map A Ñ A{a. The factorization is unique.
A
If it happens that ker φ “ a, the induce map ψ will be injective, and hence, φ1 φ

a priori being surjective, is an isomorphism. The images of all ring-maps  

im φ1
θ
/ im φ
with the same kernel are therefore isomorphic, in the strong sense that the »

isomorphisms fit into diagrams like the one in the margin.

Ideals in quotients

2.14 There is a natural one-to-one correspondence between ideals in A{a and

ideals in A containing the ideal a. Indeed, if b Ď A is an ideal with a Ď b, the
image πpbq equals the additive subgroup b{a Ď A{a and since π is surjective,
this is an ideal in A{a. Moreover, if c Ď A{a is an ideal, the inverse image
π ´1 pcq is an ideal in A satisfying πpπ ´1 pcqq “ c (again because π is surjective);
or in other words, π ´1 pcq contains a and π ´1 pcq{a “ c.
Ideals b in A
Proposition 2.15 Let a be an ideal in the ring A and π : A Ñ A{a the quotient
map. The following three statements hold true: Ideals with a Ď b Ideals in A{a
π ´1
o For every ideal b in A containing a, the quotient b{a is an ideal in A{a. Every
ideal in c Ď A{a is of this form for a unique ideal b; indeed, one has c “ π ´1 pcq{a.

o For every ideal b in A it holds true that π ´1 pπpbqq “ b ` a.

o An ideal is mapped to the zero ideal in A{a if and only if it is contained in a.

Proof: We already saw that πpbq “ b{a is an ideal. That b is unique follows
from the second assertion since if πpbq “ πpb1 q it ensues that b ` a “ b1 ` a, and
hence b “ b1 when both contain a. The second assertion is clear since a “ ker π.
Likewise is the third for the same reason. o

2.16 Proposition 2.15 above may be rephrased in terms of the lattices of ideals
I pAq and I pA{aq. The map π ´1 : I pA{aq Ñ I pAq sending an ideal c to the
inverse image π ´1 pcq is injective and the image consists of the ideals in A con-
taining a. The map the other way round, which sends an ideal b in I pAq to its
image πpbq, serves as a section to π ´1 over the sublattice of ideals containing
a. The map π ´1 preserves inclusions and respects intersections, products and
sums.
2.17 The image in A{a of an ideal b Ď A, which not necessarily contains a,
is the ideal pb ` aq{a. This holds since obviously πpb ` aq “ πpbq. Now,
30 ma4200—commutative algebra

ker π|b “ a X b from which ensues the following isomorphism

b{b X a » pa ` bq{a. (2.1)

Finally, we mention that when a and b are two ideals with a Ď b, there is a
natural isomorphism
pA{aq{pb{aq » A{b. (2.2)
Indeed, in the diagramme in the margin, the composition πb{a ˝ πa has the A
πa
/ A{a
ideal b as kernel, and therefore factors through πb by say θ. The map θ is
πb πb{a
surjective since the composition is and injective since the composition has  
b “ πa´1 pb{aq as kernel. These two formulas are often referred to as the A{b / pA{aq{pb{aq
θ
Isomorphisms theorems.
Theorem 2.18 (The Isomorphism Theorem) Let a and b be ideals in A.
Then the following two equalities hold, where in the second is assumed that a Ď b;
o b{b X a » pa ` bq{a;

o pA{aq{pb{aq » A{b.

Problem 2.11 Let k be a field and φ : k Ñ A a ring homomorphism. Show

that φ is injective unless A is the null ring. M

Examples

2.8 We promised to give examples that strict inclusion can hold in the last
statement of in Paragraph 2.8 (on page 26). Here comes one. Let B “ krX, Y, Zs{pZ ´
XYq, and as usual the lower case version of a variable will stand for its class in
B. Then B “ krx, y, zs with z “ xy.
We let the map φ : A Ñ B be the natural inclusion of A “ krzs in B “
krx, y, zs; for any ideal c in B the inverse image φ´1 pcq is then nothing but the
intersection c X B.
Consider the two principal ideals a “ pxqB and b “ pyqB in B. Since z “ xy,
it holds that z P pxqB and z P pyqB, and since pzqA is a maximal ideal in
A “ krzs , we see that a X A “ b X A “ pzqA, and hence pa X Aq ¨ pb X Aq “
pz2 qA. On the other hand ab “ pxyqB “ pzqB so that a ¨ b X A “ pzqA; hence
φ´1 paq ¨ φ´1 pbq Ř φ´1 pa ¨ bq.
2.9 For an example of strict inclusion in the last statement Paragraph 2.9 (on
page 26) let A “ krX, Y, Zs and B “ krX, Y, Zs{pZX ´ ZYq and consider the
projection map φ : A Ñ B; then φpcq “ cB for any ideal c in A.
Let a “ pXq and b “ pYq. Then a X b “ pXYq. With the usual lower case
convention, since zx “ zy, it holds that and zx P aB X bB, but zx R pxyqB; and
hence φpaq X φpbq Ř φpa X bq.

K
 ideals 31

2.3 Prime ideals and maximal ideals

Two classes of ideals are infinitely more important than others. We are speak-
ing about the prime ideals and the maximal ideals. The prime ideals are defined
in terms of multiplicative properties of the ring, and are generalizations of
prime numbers. They took the place of the primes in Kummer and Dedekind’s
world of “ideal numbers”. Maximal ideals are defined in terms of inclusions.
They are, as the name indicates, maximal among the proper ideals; that is, they
are maximal elements in the partially ordered set I pAqztAu.
2.19 An ideal a in a ring A is a prime ideal if it is proper and satisfies the follow-
ing requirement: Prime ideals (Prim-
idealer)
o If ab P a, then either a P a or b P a.

An ideal a is said to be maximal if it is proper and satisfies the following Maximal ideals (Maksi-
requirement: male idealer)

o If b is an ideal and a Ď b, then either a “ b or b “ A.

In other words, a is maximal among the proper ideals. Notice that both prime
ideals and maximal ideals are proper by definition.
2.20 One has the following characterization of the two classes of ideals in
terms of properties of quotients.:

Proposition 2.21 An ideal a in A is a prime ideal if and only if the quotient A{a
is an integral domain. The ideal a is maximal if and only if A{a is a field.

Proof: It holds true that A{a is an integral domain if and only if rasrbs “ 0
implies that either ras “ 0 or rbs “ 0; that is, if and only if ab P a implies that
either a P a or b P a, which proves the first assertion.
Bearing in mind the relation between ideals in A{a and ideals in A con-
taining a (as in Proposition 2.15 on page 29), the second assertion is pretty
obvious. There is no ideal strictly between a and A if and only if A{a has no
non-trivial proper ideal; that is, if and only if A{a is a field (Proposition 2.3 on
page 24). o
Notice that the zero ideal p0q is a prime ideal if and only if A is an integral
domain, and it is maximal if and only if A is a field. When m is a maximal
ideal, the field A{m is called the residue class field of A at m and now and then Residue class fields
denoted by kpmq. Since fields are integral domains, we see immediately that (Restklassekropper)

maximal ideals are prime. The converse does not hold as we shortly shall see
examples of (Example 2.11 below).

Proposition 2.22 A maximal ideal m is prime.

2.23 Not only for elements is it true that a product lies in a prime ideal only
when one of the factors does, the same applies to products of ideals as well:
32 ma4200—commutative algebra

Proposition 2.24 Let a and b be two ideals in A such that ab is contained in the
prime ideal p. Then either a or b is contained in p.

Proof: Assume that neither a nor b lies in p and pick elements a P a and b P b
not being members of p. Since ab is contained in p, the product ab belongs to p,
and since p is prime, it either holds that a P p or that b P p. Contradiction. o

2.25 In the correspondence between ideals in A and A{a described in Propo-

sition 2.15 on page 29 prime ideals correspond to prime ideals (containing
a) and maximal ideals to maximal ideal (containing a). The last statement is
clear since with the notation as in Proposition 2.15 the inverse image map π ´1
is an isomorphisms of the lattice I pA{aq with the sublattice of I pAq whose
members contain a, and hence maximal elements correspond to maximal el-
ements. The first ensues from the general truth that prime ideals pull back
along ring maps to prime ideals; indeed, assume that p is a prime ideal in B
and φ : A Ñ B a ring map. If the product ab lies in the inverse image φ´1 ppq, it
follows that φpabq P p; but φpabq “ φpaqφpbq, and hence either φpaq lies in p or
φpbq lies there; that is, either a P φ´1 ppq or b P φ´1 ppq.

Proposition 2.26 Let A be a ring and a an ideal. The prime ideals in the quotient
A{a are precisely those of the form p{a with p a prime ideal in A containing a, and
the maximal ideals are those shaped like m{α with m a maximal ideal in A likewise
containing a.

Examples

2.10 The archetype of maximal ideals are the kernels of evaluation maps. For
instance, let a “ pa1 , . . . , ar q be a point in kr where k is any field, and consider
the map krx1 , . . . , xr s Ñ k sending a polynomial f to its value f paq at a. The
kernel m is a maximal ideal since krx1 , . . . , xr s{m is the field k. The kernel may
be described as m “ px1 ´ a1 , . . . , xr ´ ar q. This is obvious when a is the origin,
and introducing fresh coordinates xi1 “ xi ´ ai , one reduces the general case to
that case.

2.11 A simple example of a prime ideal not being maximal can be the princi-
pal ideal pyq in the the polynomial ring krx, ys in the two variables x and y over
a field k. One has krx, ys{pyq » krxs, for instance, since the partial evaluation
map krx, ys Ñ krxs sending f px, yq to f px, 0q has kernel pyq, and pyq is prime be-
cause krxs is an integral domain. Moreover pyq is not maximal being contained
in px, yq (or if you prefer, because krxs is not a field).

K
 ideals 33

Prime avoidance and a pair of twin lemmas

A lemma about prime ideals that will be useful now and then, is the so-called
Prime Avoidance Lemma. It asserts that an ideal contained in a finite union
of prime ideals must lie entirely in one of them. The name stems from the
equivalent statement that if an ideal a is not contained in any of the prime
ideals p1 , . . . , pr , it has an element not lying in any of the pi ’s.
2.27 As a warm up, let us do the case of two prime ideals, in which case
the statement is simply a statement about abelian groups: If a subgroup B
of an abelian group is contained in the union of two others, A1 and A2 , it
is contained in one of them; indeed, pick sn xi P Ai X B but xi R A j . Then
x1 ` x2 P B but x1 ` x2 R A1 Y A2 , for were it in A1 , it would follow that x2 P A1
which is not the case. For three groups the corresponding statement is faulty
as shows the vector space F23 : It is the union of the three one-dimensional
subspaces it has; so in general some multiplicative structure is required.
Lemma 2.28 (Prime avoidance lemma) Let p1 , . . . , pr be prime ideals in the
Ť
ring A. If a is an ideal contained in the union i pi , then a is contained in at least one
of the p j ’s.

Proof: We can certainly assume that the union is irredundant; that is, none of
the prime ideals are contained in the union of the others. Then, for each j one
may pick an element y j lying in the prime ideal p j but not in any of the others.
Assume for a contrapositive argument that a is not contained in any of the
pi ’s, and chose elements x j from a not in p j . Consider for each j the element
ś
z j “ x j i‰ j yi . It lies in pi when i ‰ j but not in p j , and of course it belongs to
a since x j does. It ensues that the sum z1 ` . . . ` zr is a member of a, but not of
any of the pi ’s (each term lies in a and for a given j all but one lie in p j ). o
Notice that the proof merely requires a be closed under addition and multi-
plication, so the ideal a may be replaced with a “weak subring” (a subring
without a unit element).
2.29 At several later occasions we shall meet diverse unions and intersections
of prime ideals. In case there are no non-trivial inclusions among the involved
prime ideals, they enjoy strong uniqueness properties; in fact, the primes
involved are determined by their intersection or their union, as expressed in
the following twin lemmas.
Lemma 2.30 Let tp1 , . . . , pr u and tq1 , . . . , qs u be two families of prime ideals having
the same union; that is, p1 Y ¨ ¨ ¨ Y pr “ q1 Y ¨ ¨ ¨ Y qs . Assume that there are no
non-trivial inclusion relations in either family. Then the two families coincide.
Ť
Proof: For each index ν one has pν Ď j q j and the Prime avoidance lemma
gives that there is an index αpνq so that the relation pν Ď qαpνq holds. By sym-
metry, for each µ there is a βpµq such that qµ Ď pβpµq . Now

pν Ď qαpνq Ď pβpαpνqq
34 ma4200—commutative algebra

and there being no non-trivial inclusion relations among the pi ’s we infer that
βpαpνqq “ ν. In a symmetric manner one shows that αpβpµqq “ µ and we can
conclude that α is a bijection from t1, . . . , ru to t1, . . . , su with pν “ qα pνq, and
we are happy. o

Lemma 2.31 Let tp1 , . . . , pr u and tq1 , . . . , qs u be two families of prime ideals having
the same intersection; that is, p1 X ¨ ¨ ¨ X pr “ q1 X ¨ ¨ ¨ X qs . Assume that there are no
non-trivial inclusion relations in either family. Then the two families coincide.
Ş
Proof: For each index ν one has p1 . . . pr Ď q j Ď qν and therefore at least for
one index, say αpνq, the relation pαpνq Ď qv holds. By symmetry, for each µ there
is a βpµq such that qβpµq Ď pµ . Now

pαpβpµqq Ď qβpµq Ď pµ

and there being no non-trivial inclusion relations among the pi ’s we infer that
αpβpµqq “ µ. In a symmetric manner one shows that βpαpvqq “ ν and we can
conclude that α is a bijection from t1, . . . , ru to t1, . . . , su with with φαpvq “ qv o

Problems

2.12 Let p be a prime ideal in a ring A. Show that pArTs is prime

2.4 Principal ideals

Principal ideals are among the simplest type of ideals, and being omnipresent
in the theory they deserve special attention. So we offer a few basic results
about them.
2.32 Different elements can generate the same principal ideal, but they will, at
least in domains, be closely related, as described in the next lemma.

Lemma 2.33 Two non-zero divisors a and b in a ring A generate the same principal
ideal precisely when they are related by a unit; that is, when a “ ub where u P A˚ .

Proof: When paq Ď pbq there is a ring element u so that a “ ub, and when
pbq Ď paq it holds that b “ va, Hence a “ vua. And a not being a zero-divisor,
we conclude that vu “ 1. o
In the case of the ring A possessing zero-divisors, things are more complicated.
For instance, in the ring krx, y, zs{pzp1 ´ xyqq it holds true that pzq “ pxzq, but
the two generators z and xz are not related by a unit (see Exercise 2.19 below).
 ideals 35

Primes and irreducibles

In general rings there are two twin notions each arising from a different aspect
of prime numbers.
2.34 The first of the twins is the notion of prime elements in a ring. The defini-
tion is verbatim the same as one of the characterizations of prime numbers:
A prime element in a ring A is an element a which is neither zero nor a unit Prime elements (primele-
and is such that if a divides a product, it divides one of the factors; in other menter)

words, a relation like bc “ ya for some y implies c “ xa or b “ xb for some x.

The property is not restricted to products with two factors, a straightforward
induction proves that if a is a prime element and divides the product b1 ¨ ¨ ¨ br ,
it divides one of the factors.
The concept of a prime ideal is also inspired by that of prime numbers, and
for principal ideals the two coincide; an element a being prime is equivalent to
the principal ideal paq being a prime ideal.

Proposition 2.35 A principal ideal paq is a prime ideal exactly when a is a prime
element.

Proof: Recall what a being prime means: If a|bc then either a|b or a|c. Trans-
lated into a statement about ideals x|y means that y P pxq. Hence, bc P paq is
equivalent to a|bc, and b P paq or c P paq to a|b or a|c . o

2.36 An other aspect of prime numbers (which in fact is the usual definition) is
they can not be further factored; that is, their sole factors are 1 and the prime
itself. Irreducible polynomials in krxs share this quality except they can be
changed by non vanishing constant factors (of course f “ c´1 ¨ c f for any
non-zero constant c). Generalizing these two notions, one says that a non-
zero element a from the ring A is irreducible if it is not a unit, and if a relation irreducible elements
a “ bc implies that either b or c is a unit. (irreducible elementer)

In a domain, this can phrased in terms of ideals and a certain maximality

condition.

Proposition 2.37 An element a in the domain A is irreducible if and only if paq is

maximal among the proper principal ideal.

Proof: A relation a “ bc is equivalent to an inclusion paq Ď pbq, and when paq

enjoys the maximality property it ensues that either paq “ pbq and c is a unit, or
pbq “ A and b is a unit. o

Proposition 2.38 Every prime element in a domain A is irreducible.

Proof: Assume that a is prime element in A and that a “ bc. Since a is prime
it holds true that b “ xa or c “ xa for some x P A, say b “ xa. Substituting
back yields a “ xca and cancelling a, which is legal since A is supposed to be a
domain, we arrive at 1 “ xc which shows that c is a unit. o
36 ma4200—commutative algebra

The converse of this proposition is not generally valid, in fact one is tempted
to say that in most rings it does not hold, but we shall shortly meet classes
(see Proposition 2.41 on page 36) of rings where it is true. There are simple
examples of irreducibles not being prime in quadatic extensions of Z. We give
?
one in the ring Zr ´5s below ( Example 2.14 on page 37).

Principal ideal domains

Rings in which all ideals are principal, are among the easiest rings to under-
stand. They are called principal ideal domains, and this being a long name the Principal ideal domains
acronym pid emerges expeditiously. In a later chapter, we shall come back to (Hovedidealområder)

these rings together with three of their cousins, but because fundamental rings
like Z and the polynomial ring krxs are pid’s, the pid’s merit an entrance early
in the play.
2.39 One of the particular properties enjoyed by a principal ideal domain is
there is no distinction between maximal and non-zero prime ideals.
Proposition 2.40 In a principal ideal domain A, any non-zero prime ideal is
maximal.

Proof: A non-zero prime ideal is generated by a prime element a, and as any

other prime element, a is irreducible. From Proposition 2.37 above ensues that
paq then is maximal among the proper principal ideals, but all ideals being
principal, paq is maximal. o
Neither is there any distinction between prime and irreducible elements:
Proposition 2.41 In a principal ideal domain A an element is prime if and only if
it is irreducible.

Proof: An irreducible element a generates according to Proposition 2.37

above an ideal maximal among the proper principal ideals, but because all
ideals are principal, paq is a maximal ideal. Hence it is a prime ideal, and a is a
prime element. o

Euclidean rings
The classical division algorithm for integers, which also goes under the name
of Euclid’s algorithm (you certainly learned it in school), ensures that the ring
of integers Z is a pid:
Proposition 2.42 The ring Z of integers is a principal ideal domain.
Proof: Let a be a non-zero ideal in Z, and let n be the smallest positive
number in a. Any other element a from a may be divided by n to give a
relation a “ q ¨ n ` r where the remainder r is of absolute value less than
n. Now, r “ a ´ q ¨ n is an element of a, and this contradicts the minimality of n
unless r “ 0; therefore a “ q ¨ n, and the ideal a equals the principal ideal pnq. o
 ideals 37

2.43 The argument above is a classic but not limited to the ring of integers.
It goes through once there is a function on A for which there is a Euclidian
algorithm; that is, a function δ on A assuming values in N or N0 with the
following property2 : 2
One is not confined
to use functions with
o For any pair x and y of elements in A there are elements q and r in A so values in N; functions
with values in any
that x “ yq ` r and δprq ă δpyq. well-ordered set will
do.
Such a function is called a Euclidian function, and a domain A is said to be a Euclidian functions
Euclidean domain if it possesses one. (Euklidske funksjoner)
The absolute value used in the proof of Proposition 2.42 above is one exam- Euclidean domain
(Euklidske områder)
ple, but there are many others. For instance, on the polynomial ring krxs over
a fields, putting δp f q “ deg f ` 1 if f ‰ 0 and δp0q “ 0 we get one; indeed, the
Euclidean condition is fulfilled by the procedure of long division. The point
with these Euclidian functions is that they forces a domain to be a pid as in
the following proposition; whose proof is mutatis mutandis the same as that of
Proposition 2.42.

Proposition 2.44 A domain A that possesses a Euclidean function is a pid.

Proof: Let a be a non-zero proper ideal in A, and let a P a minimize δpxq

as x runs through the non-zero elements in a. We contend that a generates
a. If f P a, there are elements q and r with f “ aq ` r and δprq ă δpaq. Now,
r “ f ´ aq lies in a, and consequently r “ 0 since δpaq is minimal among values
of δ assumed on non-zero elements in a. o

Corollary 2.45 Let k be a field. Then the polynomial ring krxs is a pid.

Examples

2.12 In the polynomial krxs over a field k principal ideals p f pxqq with f irre-
ducible, are maximal ideals. The quotient K “ krxs{p f pxqq is a field which is
obtained from k by adjoining a root of f . If you wonder what that root is, it is
just the residue class rxs of the variable x. This illustrates the devise that what
matters in modern mathematics is “what objects do, not what they are”.
2.13 The quotient Rrxs{px2 ` 1q is isomorphic to C as one sees mapping
x to i. In a similar vein, if p is a prime number, the polynomial Φ p pxq “
x p´1 ` x p´2 ` . . . ` x ` 1 is irreducible over Q, so that Qrxs{Φ p pxq is a field.
Sending x to a primitive p-root of unity ξ , gives an isomorphism with Qpξq.
2.14 Simple and concrete examples of irreducible elements that are not prime
?
are found in the ring Zr ´5s where among others the relation
? ?
2 ¨ 3 “ p1 ` i 5q ¨ p1 ´ i 5q (2.3)
?
holds. So, for instance, 2 is not a prime element since neither 1 ` i 5 nor
38 ma4200—commutative algebra

?
1 ´ i 5 is a unit (units have absolute value 1, cfr. Exercise 1.10 on page 14). The
?
element 2 is however irreducible in Zr ´5s, for if 2 “ zw, one has 2 “ }z}}w},
?
and as the square of the absolute value of members of Zr ´5s are natural
numbers, this entails that either }z} “ 1 or }w} “ 1; hence, again in view
of Exercise 1.10, either z or w is a unit. Of course, the three other numbers
appearing in (2.3) are irreducible as well, and Exercise 2.20 below asks you to
check this. For a generic example of irreducible elements not being prime, see
Exercises 2.43 and 2.44 on page 52.

2.15 (The Gaussian integers) The absolute value works as a Euclidean function
on the ring Zris. Indeed, geometrically the Gaussian integers form the integral
lattice in the complex plane; that is, the set of points both whose coordinates
are integers. Given two Gaussian integers a and q with q ‰ 0, the distance
from a to the nearest point in the integral lattice is obviously less then half
the diagonal of a lattice square; that is, there is an element c P Zris so that
?
}a{b ´ c} ď 2{2 ă 1. Putting r “ qpa{q ´ cq, we have a “ cq ` r with }r} ă }q}.

Problems

2.13 Let p1 , . . . , pr be prime numbers and let A “ Z{pp1 ¨ . . . ¨ pr q. Show that

the prime ideals in A are precisely the principal ideals ppi q. Prove that A{pi A
is the field F pi with pi elements. How many elements does ppi q have? An how
many are there in the principal ideal pp1 ¨ . . . ¨ p̂i ¨ . . . ¨ pr q (the "hat" indicates that
pi is not included in the product).

2.14 Given two ideals pnq and pmq in Z. Show that pnq Ď pmq if and only if
m|n. Conclude that the partially ordered set I pZqztp0qu of non-zero ideals in
Z is isomorphic to the the set of natural numbers ordered by reverse divisibil-
ity.

2.15 Let n and m be two natural numbers. Describe pn, mq and pnq X pmq.

2.16 Find an explicit isomorphism between Rrxs{px2 ` x ` 1q and C. ˇ

2.17 Let p be a prime not congruent one modulo four. Show that the polyno- ˇ
mial x2 ` 1 is irreducible over the field F p , and that F p rxs{px2 ` 1q is a field
?
isomorphic to F p p ´1q. How many elements does it have?

2.18 Into which of the fields F3 , F5 and F7 is there a map of rings from Zris?
If there is one, describe the kernel.

2.19 Show that the units in krx, y, zs{pzp1 ´ xyqq are the constants k˚ . Show ˇ
that there is no unit u ‰ 1 so that uz “ xz. Conclude that z and xz are not
 ideals 39

related by a unit even though pzq “ pxzq. Hint: Killing z gives a ring-map
krx, y, zs{pzp1 ´ xyqq Ñ krx, ys. Setting x “ 1, gives a ring map krx, y, zs{pzp1 ´
xyqq Ñ krz, ys{pzp1 ´ yqq.

2.20 Referring to Example 2.14 show that the three other involved numbers 3,
? ?
1 ` i 5 and 1 ´ i 5 are irreducible.
?
2.21 Show that Zr ´2s is principal ideal domain. Hint: Prove that A is
Euclidean with the absolute value as a Euclidean function.
?
2.22 Let A “ t n{2 ` im{2 3 | n, m P Z u. Prove that A is a subring of C. Prove
that A is Euclidean

2.5 Existence theorems

A useful technique for showing that ideals (and in later chapters submodules)
of various kinds exist relies on the so-called Zorn’s lemma. The lemma is a
general result about existence of maximal elements in partially ordered sets,
which for us will be the lattice I pAq of ideals ordered by inclusion, and it
turns out to be very useful when studying rings and it will be used several
times in the sequel.

Zorn’s lemma
Zorn’s lemma which is one of a few theorems that for some reason keep being
called lemmas, is usually attributed to Max Zorn, but as often happens, it can
be traced further back. At least Felix Hausdorff published versions of it some
ten years before Zorn. Any how, "Zorns lemma" is good name (so good that
an experimental and non-narrative film made by Hollis Frampton in 1970 was Max August Zorn
called "Zorns lemma"). (1906–1993)
German mathematician
2.46 A maximal element x in the partially ordered set Σ is one for which there
is no strictly larger element; that is, if y ě x then y “ x. One should not Maximal elements
(Maksimale elementer)
confuse "maximal elements" with "largest elements" the latter being elements
larger than all other elements in Σ. A partially ordered set can have several
maximal elements whereas a largest element, if there is one, is unique. There
is of course, analogous notions of minimal elements and least elements.
A partially ordered set is said to be linearly ordered or totally ordered if any Linearly ordered sets
two of its elements can be compared. Phrased differently, for any pair x, y of (Lineært ordnede
mengder)
elements either x ď y or y ď x should hold. A chain in Σ is a linearly ordered Totally ordered sets
subset of Σ. The chain is bounded above if for some element x P Σ it holds true (Totalt ordnede mengder)
that y ď x for all elements y in the chain, and then of course, x is called an Chains (Kjeder)
upper bound for the chain. Similarly, the chain is said to be bounded below when Bounded above (Opptil
begrenset)
Upper bounds (Øvre
skranker)
Bounded below (Nedtil
begrenset)
40 ma4200—commutative algebra

having a lower bound in Σ; that is, an element x P Σ satisfying x ď y for all Lower bound (Nedre
members y of the chain. skranke)

We are now prepared to formulate Zorn’s lemma, however we shall not

prove it, only mention that it is equivalent to the axiom of choice (If you are
interested in reading more about this, consult xxx)

Theorem 2.47 (Zorn’s lemma) Let Σ be a partially ordered set in which every
chain is bounded above. Then Σ possesses a maximal element.

2.48 A chain C in Σ is called saturated or maximal if it is not properly contained Saturated or maximal
in any larger chain, or phrased differently, C is a maximal chain in Σ; that is, chains (Mettede eller
maksimale kjeder)
if C1 is another chain with C Ď C1 , then C “ C1 . A chain is saturated precisely
when it impossible to insert any new element in-between two members of
C. As an illustration of the mechanism of Zorn’s lemma, let us prove the
following

Proposition 2.49 Let C be a chain in the partially ordered set Σ. Then there is a
saturated chain containing C.

Proof: The set of chains in Σ is partially ordered by inclusion, and we intend

to apply Zorn’s lemma to that set.
Ť
If C is a chain of chains (!!) clearly the union CPC C is anew a chain; in-
deed, suppose that x and y belong to the union so that there are chains Cx
and Cy with and x P Cx and y P Cy . By assumption C is a chain, and either
Cx Ď Cy or Cy Ď Cx holds. In either case x and y lie in a common chain and are
comparable. o

A basic existence result

A frequent application of Zorn’s lemma in commutative algebra is to prove
existence of ideals that are maximal subjected to a given condition, in many
situations such maximizing ideals turn out to be prime ideals.
In this section we shall establish a basic existence result with several impor-
tant applications, one being that every ring has at least one maximal ideal. We
shall be interested in ideals maximal among those containing a fixed ideal and
being disjoint from a fixed set S. These maximizing ideals turn out to be prime
when S is multiplicatively closed; that is, if the product of any two elements
from S lie in S.

Theorem 2.50 (The Basic Existence Theorem for Ideals) Assume given
a ring A, an ideal a in A and a subset S not meeting a. Then there exists an ideal b
maximal subjected to the two following conditions

o S X b “ H;

o a Ď b.
 ideals 41

If S is multiplicatively closed, the ideal b will be a prime ideal.

Proof: Consider the set Σ of ideals in A satisfying the two requirements in

the proposition. It is non-empty because a is supposed not to meet S. Ob-
viously, the union of the ideals belonging to a chain in Σ, will lie in Σ, and
thus will be an upper bound for the chain. Zorn’s lemma applies, and we can
conclude that there is a maximal element in Σ.
Assume then that the set S is closed under multiplication and let a and b be
elements in A such that a ¨ b P b. If neither belongs to b, the ideals b ` paq and
b ` pbq both meet S, being strictly larger than b. Hence we can find elements
x ` αa and y ` βb in S with x, y P b and α, β P A, and multiplying, out we find

px ` αaqpy ` βbq “ xy ` αax ` βbx ` αβab.

The left side belongs to S as S is supposed to be multiplicatively closed, and

since x, y and ab all lie in b, the right side belongs to b, which contradicts the
fact that S X b “ H. o

Theorem 2.51 (Existence of maximal ideals) Let A be a ring different

from the null-ring. Every proper ideal a in a ring A is contained in a maximal ideal.
In particular, there is at least one maximal ideal in every ring.

Proof: We apply the proposition with S merely consisting of the unit element,
that is S “ t1u. The maximizing ideal is proper and not contained in any other
proper ideal. Hence it is maximal. To prove the second statement, apply the
first to the zero ideal. o

The radical of an ideal

Primes frequently occur with higher multiplicities in a factorization of an
integer n. It is of course interesting to get hold of the primes involved; that
is, the primes p so that some power of p divides n. In the transcription of
Kummer and Dedekind into the language of ideals, this leads to the notion of
the radical of a given ideal.
‘
2.52 The radical a of a given ideal a in A consists of the elements a power of The radical of an ideal
which lies in a; that is, (radikalet til et ideal)

a “ t a P A | an P a for some n P N u.
‘

‘
The elements of a can also be characterized as the elements in A whose
residue class in A{a is nilpotent. Along the same line, taking a to be the zero
‘
ideal, we see that p0q is the set of nilpotent elements in A.
‘
2.53 The first thing to establish is that the radical a in fact is an ideal.
‘
Lemma 2.54 Let a be an ideal in the ring A. Then the radical a is an ideal.
42 ma4200—commutative algebra

Proof: The radical is obviously closed under multiplication by ring elements,

and we merely have to check it is closed under addition. So assume that a
and b are two elements in the ring such that an P a and bm P a. The binomial
theorem gives
ÿ ˆN˙
N
pa ` bq “ a N ´i b i .
i
0ď i ď N

Choosing N “ n ` m ´ 1, we see that when i ă m it holds that N ´ i ě n, so

either a N ´i or bi lies in a. Every term of the sum therefore lies in a, and by that
the sum itself. o
Specializing a to be the zero ideal yields the following.

Corollary 2.55 The set of nilpotent elements in A form an ideal.

2.56 An ideal a in A is a radical if it equals is own radical; i.e. it holds true that
‘
a “ a. One easily verifies that the radical of an ideal is a radical ideal: that Radical ideals (Radikale
‘‘ ‘ idealer)
is, one has a “ a. In the same manner with prime ideals and maximal
ideals, radical ideals can be characterized in terms of quotients:

Proposition 2.57 An ideal a in the ring A is a radical ideal if and only if the
quotient A{a is reduced.

Proof: The residue class ras in A{a of an element a is nilpotent precisely

when a power an lies in a , which in its turn is equivalent to a P a since a “ a
‘

by hypothesis. o
2.58 The radical of an ideal a must be contained in any prime ideal containing
it because if an P a and a Ď p with p prime, it holds that a P p. The converse is
also true and hinges on the basic existence theorem above.
‘
Proposition 2.59 Assume that a is a proper ideal in the ring A. The radical a
equals the intersection of the prime ideals containing a; that is,
‘ č
a“ p.
a Ď p, p prime
‘ Ş
Proof: We already saw that a Ď a Ď p p, so assume that a is an element
‘
not lying in the radical a. We shall apply Theorem 2.50 on page 40 with S
being the set t an | n P N u of powers of a, which obviously is closed under
‘
multiplication. Since a R a, it holds that S X a “ H, and by the theorem we
conclude that there is prime ideal p containing a disjoint from S; that is, a R p.
o
The special case that a “ p0q merits to be pointed out:

Corollary 2.60 The set of nilpotent elements in A equals the intersection of all
prime ideals in A; that is
‘ č
p0q “ p.
p prime
 ideals 43

2.61 One might be tempted to discard the prime ideals not being minimal
among those containing a from the intersection in Proposition 2.59, and thus
write č
‘
a“ p. (2.4)
p minimal prime over a

Such a representation is certainly valid, but the argument is more complicated

since a priori there could be infinitely descending chains of distinct prime
Ş
ideals. However, if tpi uiP I is a chain of prime ideals, the intersection iP I pi is
a prime ideal, and so by Zorn’s lemma, every prime containing a contains a
prime ideal minimal among those containing a; which is exactly what is need
to have a representation as in (2.4). (See and do Exercise 2.23 below.)
2.62 The operation of taking the radical commutes with taking finite intersec-
tions; one has

Lemma 2.63 For every finite collection tai u of ideals in A it holds true that
Ş ? aŞ
i ai “ i ai .
‘
Proof: When an element a from A belongs to each of the radicals ai , there
are integers ni so that ani P ai . With n “ max ni , it then holds true that
an P ai for each i, and thus a P
‘Ş
Ş ‘ i ai . This shows that one has the inclusion
‘Ş
i a i Ď a
i i . The other inclusion is straightforward. o

Examples
‘ ‘
2.16 Even if a power of every element in a lies in a, no power of a will in
general be contained in a. A simple but typical example being the ideal a “
px1 , x22 , x33 , . . .q generated by the powers xii in the polynomial ring krx1 , x2 , x3 , . . .s
in countably many variables. The radical of a is equal to the maximal ideal
m “ pxi |i P N0 q generated by the variables, but no power of m is contained in a.
Indeed, the exponent needed to force a power of xi to lie in a, tends to infinity
with i.
2.17 The operation of taking radicals respects finite intersections, but this is no
more true when it comes to infinite intersections. For instance, if p is a prime
number, one has pr Z “ pZ and therefore r pr Z “ pZ, but evidently it
‘ Ş ‘

holds true that r pr Z “ 0.

Ş

Problems

2.23 That minimal and maximal prime ideals exist ensues directly from ˇ
Zorn’s lemma as this exercise shows. Let tpi uiP I be a chain of prime ideals.
Ť Ş
Show that both the union iP I pi and the intersection iP I pi are prime ideals.
44 ma4200—commutative algebra

Show that every prime ideal containing a given ideal a contains a a prime
ideal minimal over a.

2.24 A multiplicatively closed set S in the ring A is said to be saturated if with ˇ

x it contains every factor of x; that is, if x P S and x “ yz, then y P S (and by Saturated multiplicative
sets (Meted multiplicative
symmetry z P S). Show that S is a saturated multiplicative set if and only if the mengder)
complement AzS is the union of prime ideals. Hint: Assume that a R S and
apply Theorem 2.50 with the ideal a being the principal ideal paq.

2.25 Let A be any ring. Show that the set of non-zero divisors in A form a ˇ
saturated multiplicative set. Conclude that the set of zero divisors is the union
of prime ideals.

2.6 Local rings

Rings having one single maximal ideal are called local rings. They occupy Local rings (Lokale
a central place in the theory being simpler than many other rings, and a ringer)

frequently applied strategy of proof is to reduce an issue to a statement about

local rings. In the analogy with rings of functions, the local rings correspond
to rings of germs of functions near a point—hence the name. There is also the
notion of a semi-local ring, which is a ring with finitely many maximal ideals. Semi-local rings
(Semilokale ringer)
2.64 In a local ring A with maximal ideal m the complement of m coincides
with the group of the units; that is, every element a P A not lying in m is
invertible. Indeed, if it were not, the principal ideal paq would be a proper
ideal and by Theorem 2.51 on page 41 would be contained in a maximal ideal,
obviously different from m, which would contradict that m is the sole maximal
ideal in A. This proves that the first statement in the following proposition
implies the second.

Proposition 2.65 Let A be a ring and m a proper ideal in A. The following three
statements are equivalent.

o A is a local ring with maximal ideal m;

o The group of units and the complement of m coincide; that is, A˚ “ Azm ;

o The ideal m is maximal and consists of the elements a such that 1 ` a is invertible.

Proof: To see the last statement ensues from the second, let a be a member of
m. Then of course 1 ` a is not in m and hence is invertible.
Finally, let m be maximal and assume that any element shaped like 1 ` a is
invertible. Let x be an element not in m. Since m is maximal, it holds true that
m ` pxq “ A; hence x “ 1 ` a for some a P m, and x is invertible. o
 ideals 45

The assumption in the last statement that m be maximal, is necessary; for an

example see Exercise 2.28 below.
2.66 The argument in previous paragraph partially goes through in a slightly
more general situation involving the so-called Jacobson radical JpAq of a ring Jacobson radical
A, which is the intersection of all the maximal ideals in A. In other words, (Jacobson-radikalet)
Ş
JpAq “ m Ď A maximal m.

Proposition 2.67 Let A be a ring. The Jacobson radical of A consists of the ring
elements a so that 1 ` xa is invertible for all x P A.

Proof: Fix an element a in A. Firstly, assume that elements of shape 1 ` xa

all are invertible. If there is a maximal ideal m so that a R m, it holds true that
m ` paq “ A. Then there is a relation 1 “ y ` ax with y P m. It ensues that 1 ´ xa
lies in m, but on the other hand, 1 ´ ax is invertible by assumption; and we have
the contradiction that m “ A. Remember, maximal
Assume then that a lies in all maximal ideals. If p1 ` axq is a proper ideal, it ideals are proper ideals

is by the fundamental existence theorem (Theorem 2.50 on page 40) contained

in a maximal ideal n. Since a P n, it follows that 1 P n, contradicting n being
proper. Hence p1 ` axq is not proper, and 1 ` ax is invertible. o

2.68 Assume that A and B are two local rings whose maximal ideals are m A
and mB respectively. A map of rings φ : A Ñ B is said to be a local homomor-
phism, or a map of local rings, φpm A q Ď mB . Equivalently, one may request that Local homomorphism
(Lokal homomorfi)
φ´1 pmB q “ m A .

Examples

2.18 The set of rational functions over a field k that may be expressed as
Ppxq{Qpxq with Ppxq and Qpxq polynomials and Qp0q ‰ 0, is a local ring whose
maximal ideal equals the set of the functions vanishing at the origin. The
evaluation map Ppxq{Qpxq ÞÑ Pp0q{Qp0q identifies the residue field with the
field of complex numbers C.

2.19 Let p be a prime number and let Zp pq be the ring of rational numbers
expressible as n{m where the denominator m is relatively prime to p. Then
Zp pq is a local ring whose maximal ideal is generated by p.
More is true, the only ideals in Zp pq are the principal ideals ppv q; indeed,
every rational number lying in Zp pq may be written as pν n{m with ν ě 0 and
neither n nor m having p as factor. And among these ideals ppq contains all the
others.

2.20 In a polynomial ring Crx1 , . . . , xr s for all points a P Cr the ideal of polyno-
mials vanishing at a is a maximal ideal. It follows that the Jacobson radical f
Crx1 , . . . , xr s equals p0q; i
46 ma4200—commutative algebra

2.21 Assume that p and q are two prime numbers. Let A be the ring of ratio-
nal numbers with denominator relatively prime to pq. That is A “ t n{m |
n, m P Z, pm, pqq “ 1 u. The principal ideals by ppq and pqq are the only two
maximal ideals in A, and JpAq “ ppq X pqq “ ppqq.

Problems

2.26 Show that a ring has just one prime ideal if and only if its elements are ˇ
‘
either invertible or nilpotent. Prove that this is the case if and only if A{ p0q is
a field.

2.27 With reference to Exercise 1.22 on page 21, show that the ring Z p of
p-adic integers is a local ring whose sole maximal ideal is generated by p.

2.28 Let A be the subring of Q whose elements are the rational numbers a
expressible as a “ m{n where n does have neither 2 nor 3 as factor. Show that
A has two maximal ideals p2q and p3q whose intersection equals p6q. What are
the two residue fields? Show that 1 ` a is invertible in A for all members a P p6q.

2.29 Let p1 , . . . , pr be distinct prime numbers and let A be the subset of

Q whose members can be written as m{n with n relatively prime to pi for
1 ď i ď r. Show that A is a semi-local ring. Describe the maximal ideals and
the residue fields. What is the Jacobson radical?
ś
2.30 Let k1 , . . . , kr be fields. Show that the product ring i k i is a semi-local ˇ
ring. What are the maximal ideals?

2.31 Let f pxq be any polynomial in krxs where k is a field. Show that krxs{p f pxqq ˇ
is semi-local.

2.7 Direct products and the Chinese Remainder Theorem

Ideals in a direct product
ś
Let A “ 1ďiďr Ai be the direct product of rings Ai . There is a simple descrip-
tion of ideals in A in terms of ideals in the Ai ’s. One produces an ideal a in
ś
A from a sequence of ideals ai in the Ai ’s simply by putting a “ i ai . And,
indeed, all ideals in A are of this shape.
To see this, let tei u1ďiďr be the orthogonal idempotents corresponding to the
decomposition of A as a direct product. Then Ai “ ei A and each ei a is an ideal
ř ř
in Ai contained in a, and because i ei “ 1, it holds true that a “ i ei a.
 ideals 47

ś ś
Proposition 2.69 The ideals of A “ Ai are all of the form 1ďiďr ai
1ď i ď r ś
where each ai is ideal in Ai . It holds true that A{a » 1ďiďr Ai {ai . The ideal p is a
prime ideal if and only if pi “ Ai for all but one index i.

Proof: The projection mappings A Ñ Ai , coinciding with multiplication

by ei , send a to ai and induce maps A{a Ñ Ai {ai , which in their turn give a
ś ś
ring-map A{a Ñ i Ai {ai . The map is surjective since A Ñ i Ai is, and if x
in A such that ei x P ai for each i, the element x belongs to a, and the map is
injective.
What remains to be seen is the statement about the prime ideals, which
ś
follows since the principal idempotents in i Ai {ai are orthogonal and so
ś
when at least two of them that are non-trivial, the product i Ai {ai is not
integral domain. o

2.70 It is appropriate to give a comment about the zero ring at this stage. In
Proposition 1.27 the idempotents ei ’s are not required to be different from
zero, but if ei “ 0, of course ei A is the zero ring, and does not contribute in a
significant way to the product (it holds true that 0 ˆ A “ A). This is particularly
pertinent for the formulation of Proposition 2.69; it might happen that ai “ Ai
so that A{ai is the zero ring.

The Chinese remainder theorem

A classical result that at least goes back to third century AD is the so called
Chinese Remainder Theorem, a more informative name would be the Theorem
of Simultaneous Congruences: As long as the moduli n1 and n2 are relatively
prime, two congruences x ” y1 mod n1 and x ” y2 mod n2 have a common
solution. It seems that the first written account of this result is found in the
book Sunzi Suanjing by a Chinese mathematician “Master Sun”— hence the
Chinese theorem.
This can of course be generalized to any number of congruences as long
as the moduli are pairwise prime, and there is a formulation for general ring
Some old Chinese
with the moduli replaced by ideals. To formulate the appropriate condition on mathematics.
the ideals, the notion of comaximal ideals is introduced. Two ideals a and b
are said to be comaximal is a ` b “ A, equivalently, one may write 1 “ a ` b with Comaximal ideals
a P a and b P b. (Komaksimale idealer)

Given a finite collection tai u1ďiďr ideals in the ring A. There is an obvious
map
ź
AÑ A{ai
i

sending a ring-element a to the tuple whose i-th component is the residue

class of a modulo ai . Its kernel consists of the elements in A lying in all the
48 ma4200—commutative algebra

ai ’s, and hence there is induced an injective map

ź
ψ : A{a1 X . . . X ar ãÑ A{ai .
i

The Chinese statement is that, under certain circumstances, this map is an

isomorphism.

Theorem 2.71 (The Chinese Remainder Theorem) Let A be a ring and let
ta u be a finite collection of pairwise comaximal ideals. Then A{a1 X . . . X ar »
śi 1ďiďr
1ďiďr A{ai .

Proof: It suffices to find elements ai in A which are congruent one modulo

ai and congruent zero modulo all the other ideals in the collection. Indeed,
ř
the sum i yi ai , with the yi ’s being arbitrary ring-elements, will then have the
same residue class as yi modulo ai .
For each pair of indices i and j with i ‰ j we may write 1 “ cij ` c ji with
cij P a j . Then cij is congruent 1 modulo ai and congruent zero modulo a j .
ś
Hence the product ai “ j‰i cij is congruent 1 modulo ai and congruent zero
modulo a j for j ‰ i; and we are done. o

Problems

2.32 Let a and b be two comaximal ideals such that a X b “ 0. If a ` b “ 1 with

a P a and b P b, show that a and b are idempotents.

2.33 Show that 28y ´ 27z solves the simultaneous congruences x ” y mod 9
and x ” z mod 4.

2.34 Let A be a semi-local ring. Show that A{JpAq is a product of fields. ˇ

2.35 Assume that a1 , . . . , ar are pair-wise comaximal ideals. Show that one
has a1 ¨ . . . ¨ ar “ a1 X . . . X ar .

2.8 Graded rings and homogenous ideals

Recall that any polynomial can be written as the sum of its homogenous
components. Several techniques, useful when working with polynomials,
involve this decomposition; just two mention two, induction on the degree
of the lowest or the highest term are powerful tools. A class of rings sharing
some of these properties are the so-called graded rings whose elements possess
a decomposition mimicking the one of polynomials into homogeneous terms.
 ideals 49

Even more forceful techniques are available to handle graded rings which
satisfy appropriate finiteness conditions. For instance, when all the homoge-
nous components Rn are finite dimensional vector spaces over some field
k Ď R0 , the so called Hilbert function h R pνq “ dimk Rn is a very strong invari-
ant of R.
A the present stage of the course we merely scratch the surface of the
theory of the graded rings, but they will reappear at several later occasions.
2.72 A graded ring R is a ring together with a decomposition of the underlying Graded rings (graderte
abelian group as a direct sum ringer)

à
R“ Rν (2.5)
νPZ

of additive subgroups Rν subjected to the rule Rν ¨ Rµ Ď Rν`µ for any pair of

indices ν, µ.
2.73 Elements from the subgroup Rν are said to be homogenous of degree ν. Homogenous elements
Notice that the zero element 0 lies in every one of the subgroups Rν and (homogene elementer)

one can not attribute a well-defined degree to it, but it will be considered to
be homogenous of any degree. From a decomposition as in (2.5) it ensues
ř
that elements a in R can be expressed as sums a “ ν aν whose terms av
are homogenous of degree ν with merely finitely many being different from
zero. The aν ’s are uniquely determined by a and go under the name of the
homogenous components of a. Homogenous components
Notice that R0 ¨ R0 Ď R0 , so R0 is a subring of R. Similarly, for any ν it holds (homogene komponenter)

true that R0 ¨ Rν Ď Rν , and the ring R0 of elements homogenous of degree zero

acts on the group of those homogenous of degree ν. In particular, if k Ď R0 is a
field, the additive subgroups Rν will all be vector spaces over k.
2.74 If a is an ideal in the graded ring R, we denote by aν the subgroup aν “
a X Rν consisting of the homogenous elements of degree ν that lie in a. One
ř
says that a is a homogenous ideal whenever a “ ν aν ; in other words, whenever Homogenous ideals
all the homogeneous components of elements from a belong to a as well. (homogene idealer)

Since homogenous components are unambiguously defined (or if you prefer,

because aν X aµ “ p0q whenever ν ‰ µ), the sum is a direct sum, and we are
À
entitled to write a “ ν aν .

Proposition 2.75 Let a be an ideal in the graded ring R. The following three
statements are equivalent.

o The ideal a is homogenous;

o All homogenous components of elements in a belong to a;

o The ideal a may be generated by homogenous elements.

Proof: That the first two statements are equivalent, is just a rephrasing of
how homogenous ideals were defined. So assume that a is a homogenous
50 ma4200—commutative algebra

ideal. The homogeneous components of any set of generators of a then belong

to a, and obviously they generate a (the original generators are sums of them).
The other implication is also straightforward. Let tai uiP I be a set of homoge-
neous generators for a, and say ai is of degree di . Any element a P a can then
ř
be expressed as f “ i f i ¨ ai with f i P R, and expanding the sum into a sum of
homogenous term we find
ÿ ÿ`ÿ ˘ ÿ` ÿ ˘
f “ f i ¨ ai “ f i,ν ¨ ai “ f i,ν ¨ ai ,
i i ν d ν`d i “d

where f i,ν denotes the homogeneous component of f i of degree ν (most of

these will vanish), and where we in the last sum have recollected all terms
ř
f i,ν ¨ ai of degree d. Hence ν`di “d f i,ν ¨ ai is the homogeneous component of f
of degree d, and it belongs to a since the ai ’s lie there. o

2.76 A rich source of graded rings are the quotients of polynomial rings by
homogenous ideals, or more generally the quotient of any graded ring by a
homogenous ideal.

Proposition 2.77 Let R be a graded ring and a Ď R a homogeneous ideal. Then

the quotient R{a is a graded ring whose homogeneous components are given as
pR{aqν “ Rv {aν .

Proof: This follows immediately from the the direct sum decompositions R “
À À
ν Rν and aν “ ν aν . Notice first Ri X a “ ai and it holds
ř
that Ri {ai Ď R{a.
Hence any class ras P R{a with a decomposing as a “ i ai in homogeneous
ř
terms of degree i decomposes as ras “ i rai s, where we can consider the
rai s’s to be elements in R{a or in P Ri {ai . Moreover, the classes rai s are unique
ř ř
because is i ai and i bi were two such decompositions inducing the same
ř
element in R{a, it would hold true that i pai ´ bi q P a. The ideal a being
homogeneous and each term ai ´ bi being homogeneous of degree i, it would
follow that ai ´ bi P a, and hence rai s “ rbi s. o

Example 2.22 (A weighted grading) There is a way of giving polynomial rings

another grading than the traditional one, which sometimes turns out to be
useful. We shall illustrate this in the case two variables R “ krx, ys. The idea is
to give each variables x and y a weight, that is putting deg x “ α and deg y “ β
where α and β can be any pair of integers. The degree of the monomial xi y j
is then defined as deg xi y j “ iα ` jβ. This defines a graded structure on the
polynomial ring with
k ¨ xi y j .
à
Rv “
iα` jβ“ν

Since already R is the direct sum R “ i,j k ¨ xi y j , one arrives at the direct sum
À

R “ ν Rν by just recollecting terms k ¨ xi x j with the same degree.

À
K
 ideals 51

Example 2.23 A natural and useful condition on a graded ring is that

Rn “ p0q for n ă 0; that is, the degrees of any non-zero element is non-
negative (it opens up for induction arguments). However, several graded
rings occurring naturally are not like that. One example is the subring R of
kpx1 , . . . , xn q consisting of rational functions shaped like f {gν where g is a
fixed homogenous polynomial, and f P krx1 , . . . , xn s and ν P N0 . Putting
deg f {gν “ deg f ´ ν ¨ deg g makes R a graded ring (check that!), and then
deg 1{gν “ ´ν ¨ deg g. K

Problems

2.36 Generalise Example 2.22 above to polynomials in any number of vari-

ables by giving each variable xi a weight αi .

2.37 With reference to the Example 2.22 above , show that the subring R0
of elements of degree zero in the case α “ 1, β “ ´1 is isomorphic to the
polynomial ring over k in one variable. Describe Ri for all i.

2.38 (Monomial ideals.) An ideal a in the polynomial ring krx1 , . . . , xr s is said

to be a monomial if it holds true that a polynomial f belongs to a if and only Monomial ideals (Mono-
if every monomial occurring in f lies there. Show that this is equivalent to a miale idealer)

being generated by monomials.

2.39 Assume that k is an infinite field. The multiplicative group k˚ acts on

the polynomial ring krx1 , . . . , xr s in a natural way, the result of t P k˚ acting on
f px1 , . . . , xr q being f t px1 , . . . , xr q “ f ptx1 , . . . , txr q. Show that the polynomial f
is homogeneous of degree d if and only if f t “ td ¨ f for all t. Show that an ideal
a is homogeneous if and only if a is invariant under this action.

2.40 Assume that k is an algebraically closed field and that f px, yq is a homo-
geneous polynomial in krx, ys. Show that f px, yq splits as a product of linear
factors. Hint: If f is of degree d, it holds that f px, yq “ yd f px{y, 1q. Consider
f px{y, 1q as a polynomial in t “ x{y.

2.41 (Homogenization of polynomials.) Let k be a field and let f P krx1 , . . . , xr s

be any non-zero polynomial. Let d be the degree of f . Define a fresh poly-
nomial f H P krx0 , . . . , xr s in one more variable by putting f H px0 , . . . , xr q “
x0d f px1 {x0 , . . . , xr {x0 q. Show that f H is homogenous of degree d. Show that Homogenization (ho-
f H p1, x1 , . . . , xr q “ f px1 , . . . , xr q. mogenisering)

2.42 (Dehomogenization of polynomials.) The homogenization process described

in the previous exercise has a natural reverse prossess called dehomogenization Dehomogenization
(with respect to one of the variables, x0 in this exercise). When g P krx0 , . . . , xr s (dehomogenisering)

is homogenous of degree d, one puts g D px1 , . . . , xr q “ gp1, x1 , . . . , xr ). Show

that pg D q H “ x0s g for some non-negative integer s ď d. Give examples to see
that s actually can have any value between 0 and s.
52 ma4200—commutative algebra

À
2.43 Assume that A “ iě0 Ai is a graded integral domain with A0 being
a field. Show that any element homogenous of degree one is irreducible.
Hint: Work with components of highest degree.

2.44 Let A “ Zrx, y, z, ws{pxy ´ wzq show that the class of x is irreducible but
not prime.

2.9 The prime spectrum and the Zariski topology

Every ring has a geometric incarnation, called the prime spectrum Spec A of Prime spectrum of a ring
the ring A whose points are the prime ideals in A. It carries a topology called (Primspekteret til en ring)

the Zariski topology after Oscar Zariski, one of the great men of algebraic
geometry. This topological space depends functorially on the A, a ring map
φ : A Ñ B induces a map φ̃ : Spec B Ñ Spec A simply by sending a prime φ in
B to the inverse image φ´1 ppq (which is a prime ideal in A).
The spectra of rings enter as building blocks in Alexander Grothedieck’s
scheme theory , which is an infinitely larger ocean with as yet huge unex-
plored region, and the spectra form merely the shore.
In happy marriages the spouses exert a strong mutual influence, so also in
the relationship between algebra and algebraic geometry. Several geometric
features of Spec A are paramount to understanding algebraic properties of the
ring A, and, of course, vice versa. Even modern number theory and arithmetic
are inconceivable without the geometric language along with the geometric
intuition. However, in these matters we shall only superficially scratch the
surface; giveing the basic definitions and a few examples.
There is another geometric construct antecedent of the schemes by about a
century. Basically it goes back to René Descartes’s idea of using coordinates
and equations to describe geometric objects. We have all experienced parts of
the menagerie of plane curves and surfaces in the space. In general, subsets of
Cr being the common zeros of a set of polynomials, are called varieties and are Varieties (varieteter)
the subjects of interest for many algebraic geometers.

Prime spectra

2.78 Being a geometric gadget, the prime spectrum is endowed with a topol-
ogy which is called the Zariski topology. This topology is best defined by giving The Zariski topology
the closed subsets. These are denoted Vpaq where a is any ideal A, and given (Zariski topologien)

as the set of prime ideals containing a; that is, one has

Vpaq “ t p Ď A | p a prime ideal pĚ a u.

 ideals 53

There are some axioms to be verified. First of all, Vp0q “ Spec A and VpAq “
H, so the empty set and the entire space are both closed. Secondly, we must
check that unions of finitely many closed subsets are closed (it suffices to
check it for unions of two) and that the intersection of any family of closed
sets is closed. To the former, observe that Vpaq Y Vpbq Ď Vpabq since both
ab Ď a and ab Ď b hold, and the other inclusion Vpabq Ď Vpaq Y Vpbq follows
since ab Ď p implies that either a Ď p or b Ď p according to Proposition 2.24 on
page 2.24 above. Hence Vpaq Y Vpbq “ Vpabq.
ř
To the latter, notice the trivial fact that i ai lies in p if and only if each of
the ai ’s lies in p. We have shown
Lemma 2.79 Let A be a ring.
o Vp0q “ Spec A and Vp1q “ H

o For any ideals a and b in A it holds true that Vpaq Y Vpbq “ Vpabq; Oscar Zariski
(1899–1986)
Russian-born American
ř Ş
o For any family tai uiP I of ideals one has Vp iP I ai q “ iP I Vpai q;
mathematician
o If a Ď b, then Vpbq Ď VpFaq
‘
o Vpaq “ Vp aq.

2.80 The Zariski topology has certain peculiar features never met when work-
ing with mundane topologies like the ones of manifolds. For instance, there
are lots of points in Spec A that are not closed; so in particular the prime
spectra tend to be seriously non-Hausdorff. One has:
Lemma 2.81 The closed points in Spec A are the maximal ideals.
Proof: We saw that every proper ideal is contained in a maximal ideal (The-
orem 2.51 on page 2.51); hence Vpaq will always have a maximal ideal as
member. So if tpu is closed; that is, equal to Vpaq for some a, the prime ideal p
must be maximal.
On the other hand whenever m is maximal, obviously Vpmq “ tmu since no
prime ideal strictly contains m. o

Examples
We do not intend to dive deeply into a study of prime spectra, but to give an
idea of what can happen, let us figure out which of the topological spaces with
only two points can be a prime spectrum.
There are three non-homeomorphic topologies on a two-point set; the
discrete topology with all points being close (and hence all being open as
well), the trivial topology whose sole closed sets are the empty set and the
entire space, and finally the so-called Sierpiński space, a two-point space with
just one of the points being closed (and consequently the other being open).
And two of these occur as Zariski topologies.
54 ma4200—commutative algebra

2.24 The direct product of two fields A “ k ˆ k1 has merely the two prime ideals
p0q ˆ k1 and k ˆ p0q which both are maximal. Hence Spec k ˆ k1 consists of two
points and is equipped with the discrete topology.
2.25 The ring Zp pq of rational numbers expressible as fractions with a de-
nominator prime to p has just two prime ideals, namely p0q and the principal
ideal ppq (Example 2.19 on page 45). Hence tp0qu is an open set being the
complement of the closed point ppq. Hence Spec A is the Sierpiński space.
2.26 Finally, the trivial topology having no closed point, can not be the Zariski
topology of any non-empty prime spectrum; in every ring different from the
null-ring there are maximal ideals, and the spectrum of the null-ring is empty.

K
2.82 The spectrum Spec A depends functorially on the ring A. If φ : A Ñ B
is a map of rings, pulling back ideals along φ takes prime ideals to prime
ideals; indeed, let p Ď B be prime and assume that ab P φ´1 ppq. This means
that φpaqφpbq P p, and so either φpaq P p or φpbq P p. In other words, either a lies
in P φ´1 ppq or b lies there. We thus obtain a map φ̃ : Spec B Ñ Spec A.
Proposition 2.83 The map φ̃ is continuos.
Proof: Let a Ď A be an ideal. The one has φ̃´1 Vpaq “ VpaBq; indeed, tautolog-
ically it holds true that a Ď φ´1 p if and only if φpaq Ď p. o
A byproduct of the proof is that the inverse image under φ̃ of the closed set
Vpaq is homeomorphic to Spec B{aB. Indeed, the prime ideals in B{aB are
in a one-to-one correspondence with the prime ideals in B containing aB
(Theorem 2.15 on page 29) , and these are, as we saw in the proof, precisely
the points in Spec B mapping to points in Vpaq. Moreover, the whole lattice of
ideals I pB{aBq is isomorphic to the lattice of ideals in B containing aB. That
takes care of the topology; closed sets correspond to closed sets.
Proposition 2.84 Let φ̃ : Spec B Ñ Spec A be induced by φ : A Ñ B. Then the
inverse image φ̃´1 Vpaq is homeomorphic to Spec B{aB. In particular, for any point
p P Spec A the fibre over p equals Spec B{pB.

2.85 The Zariski topology has a particular basis of open sets called the
distinguished open sets. For each element f P A there is one such open The distinguished open
set Dp f q whose members are the prime ideals not containing f ; that is, sets (Særskilte åpne
mengder)
Dp f q “ t p | f R p u. This is an open set since the complement is the closed set
Vp f q.
Lemma 2.86 The distinguished open sets form a basis for the topology on Spec A.
Proof: A typical open subset U is the complement of a set shaped like Vpaq.
Now, a prime p does not contain a if and only if there is an element f lying in
Ť
a but not in p. Hence U “ f Pa Dp f q, and we are through. o
 ideals 55

Problem 2.45 Let A and B be two rings. Show that Spec pA ˆ Bq is the disjoint
union of Spec A and Spec B. M
‘
Problem 2.46 Show that Vp aq “ Vpaq. M

Complex Varieties
For any S Ď Crx1 , . . . , xr s of polynomials, the zero set VpSq is defined as the set
in Cr of simultaneous zeros of the polynomials in S; that is,

VpSq “ t x P Cr | f pxq “ 0 for all f P S u.

Curves in plane and surfaces in the space, as we know them fro earlier course
are most often of this type with S just containing one element, the equation of
the gadget under consideration. There is one difference however, we consider
points with complex coordinates not only reals.
Example 2.27 The good old parabola equals Vpy ´ x2 q in C2 , or more precisely
the old chap consists of the real points (those with both coordinates being real)
of Vpy ´ x2 q. K
2.87 One easily convinces oneself that the ideal a generated by S has the same
zero-set as S (sums of functions vanishing at point vanish there too, and even
more, multiples of one vanishing vanishes), so without loss one may confine
the study to sets shaped like Vpaq. One has the following formulas, where a
and b are ideals in Crx1 , . . . , xr s:

o If a Ď b then Vpbq Ď Vpaq;

o Vpa ` bq “ Vpaq X Vpbq;

o Vpa X bq “ Vpabq “ Vpaq Y Vpbq

‘
o Vp aq “ Vpaq

o Vpp1qq “ H and Vpp0qqq “ Cr

We have seen that to any point in Cr there corresponds a maximal ideal

ma in Crx1 , . . . , xr s, namely the one whose members are the polynomials
vanishing at a. It is generated by the linear polynomials xi ´ ai . There is a
famous theorem of David Hilbert’s called the Hilbert Nullstellensatz asserting,
in one of its guises, that all maximal ideal in Crx1 , . . . , xr s are of this shape.
So anticipating that result we conclude that the points in Vpaq are in natural
one-one correspondence with the maximal ideals in Crx1 , . . . , xr s containing a;
hence with the closed points in Spec Crx1 , . . . , xr s{a.
History:
2018-08-21: Added examples 2.1, 2.2, 2.3 and 2.4. Corrected misprints and improved the
language a few places.
23/08/2018 Added example 2.4;
56 ma4200—commutative algebra

24/08/2018 Corrected many misspellings in section 2.9

27/08/2018 Moved examples in paragraphs 2.8 and 2.9 to after quotients have been treated.
Expanded paragraph 2.56. Added a small paragraph 2.25 about prime- and max- ideals in
quotients.
30/08/2018 Expanded example 2.4. Added exercise 2.13.
06/09/2018 Corrected the definition of irreducible and prime elements. Added example 2.20;
07/09/2018 Have rewritten the section about principal ideals. Taken out the section abpout
ufd’s which has become a chapter of its own.
17/09/2018 Minor changes; discarded exercise 2.25
21/10/2018 Added Lemma 2.28 on page 33
26/10/2018 Have two more lemmas. Needed in chapter 8, but Their natural place is here
Lecture 3

Unique factorization domains I

Preliminary version 0.1 as of 2018-09-13 at 14:13 (typeset 3rd December 2018 at 10:03am)—Prone
to misprints and errors and will change. 2018-09-13 Corrected proof of Prop 3.11.
13/09/2018 Added hypothesis that Rbe a doamain in Proposition 3.20.

When working with integers the Fundamental Theorem of Arithmetic

is a most valuable tool used all the time. In general is does not hold and
we have to do without it, and the birth of algebraic number theory and by
that beginning of commutative algebra, came as a response to this “defect”.
However in certain nice rings the fundamental theorem persists, and these
rings are called unique factorization domains or factorial ring, and out of the Unique factorization
inherent human laziness comes the acronym ufd which is in widespread use. domains (Entydig
faktoriseringsområder)
Factorial rings
3.1 Unique factorization domains ufd’s

3.1 To be precise a ufd is a domain where every non-zero element which is not
a unit, can be expressed in an essentially unique way as a product

a “ p1 ¨ . . . ¨ pr

of irreducible elements pi . The qualifyer "essentially unique" must be under-

stood in the large sense; the order of the factors can of course be changed at
will, and replacing a factor pi with upi , where u is a unit, can be compensated
by multiplying another factor by the inverse u´1 . So "essentially unique"
means that the factors are unique up to order and multiplication by units.
In the definition there are two separate conditions which are of quite differ-
ent flavour, and we shall take a closer look at each.
3.2 The first is the requirement that any element can be expressed as a finite
product of irreducibles. This is in essence a finiteness condition on A and
is fulfilled e.g. in Noetherian rings, but it does certainly not hold in general.
An example can be the ring of entire functions in C. The irreducibles in this
ring are of the form upzqpz ´ aq where upzq is a unit (i.e. a non-vanishing
entire function) and a P C any point, so any entire function with infinitely
many zeros—like our good old friend sin z—can not be a finite product of
irreducibles.
58 ma4200—commutative algebra

One can always attempt a recursive attack. Any ring element a which is not
irreducible, is a product a “ a1 a2 of non-units. These being irreducible makes
us happy, but if they are not, they are in their turn products a1 “ a11 a12 and
a2 “ a21 a22 . If some of the fresh factors are not irreducible, we split them into
products of non-units. Continuing like this we establish a recursive process
which, if it terminates, yields a finite factorization of a into irreducibles.
However, the process may go onfor every—like it will for e.g. a “ sin z—but
in many cases there are limiting condition making it end. For instance, in the
case of Z, the number of steps is limited by the absolute value }a}, and in the
case of polynomials in krxs by the degree of a.
3.3 The second condition is the uniqueness requirement which is much more
of an algebraic nature, and hinges on the condition that irreducible elements
be prime. In fact, in any ring the uniqueness condition holds automatically for
finite factorizations into prime elements:

Lemma 3.4 Let A be a ring. Assume that tpi u1ďiďr and tqi u1ďiďs are two collec-
tion of prime elements from A whose products agree; that is it holds true that

p1 ¨ ¨ ¨ pr “ q1 ¨ ¨ ¨ q s .

Then the pi ’s and the qi ’s coincide up to order and unit factors.

Proof: The proof goes by induction on r. Since p1 is prime, it divides one

of the qi ’s, and after reordering the qi ’s and adjusting q1 by a unit, we may
assume that p1 “ q1 . Cancelling p1 gives p2 ¨ ¨ ¨ pr “ q2 ¨ ¨ ¨ qs , and we finish the
proof by induction. o
Since irreducibles are prime in Z and krxs (both are principal ideal do-
mains), we immediately conclude that Z and krxs are factorial rings.

Examples
The main examples of factorial rings are principal ideal domains and polyno-
mial rings over those, as we shortly shall see, and producing other examples
demands some technology not available to us for the moment. So we confine
ourselves to examples of non-factorial rings.
3.1 The classical example of a ring which is not factorial, which is ubiqui-
tous in texts, is the ring Zri 5s. In it 6 has two distinct factorizations in irre-
‘

ducibles (see Example2.14 on page 37)

‘ ?
6 “ 2 ¨ 3 “ p1 ` i 5qp1 ´ i 5q.

3.2 The standard example from algebraic geometry is the quotient ring1 1
A geometer would
A “ krX, Y, Z, Ws{pXY ´ ZWq. Indicating the class of a variable by a lowercase call it "the cone over
a quadratic surface in
version of the name; the relation projective 3-space

xy “ zw (3.1)
 unique factorization domains i 59

holds in A. We saw in Example 3.5 on page 64 that A is a domain and its

graded since the polynomial XY ´ ZW is homogenous. It is not to challeng-
ing to see that the class of any non-zero linear form is irreducible in A (see
Exercise 2.43 on page 52). Hence the relation (3.1) shows that A is not factorial.
This also gives an easy example of the intersection of two principal ideals
being non-principal; i.e. distinct from their product, in that pxq X pzq “ pxy, xzq.
3.3 Another famous example from algebraic geometry can be B “ CrX, Ys{pY 2 ´
XpX ´ aqpX ´ aqq where a and b are elements in C. The relation Figure 3.1: The real
points of a cubic curve
y2 “ xpx ´ aqpx ´ bq, y2 “ xpx ´ aqpx ´ bq
with a ‰ b and ab ‰ 0.
where x and y denotes the classes of X and Y in B, prevails in B and gives two
different decompositions of an element into irreducibles; of course one must
verify that the involved linear factors are irreducible (see exercise xxx). Plane
curves given by equations like

y2 ´ xpx ´ aqpx ´ bq

with a, b P C are called elliptic curves and have been at the centre-stage of
algebraic geometry since its beginning. They are closely related to the so-
called elliptic functions, and in fact, they were the very starting point for the
development of algebraic geometry.
3.4 The polynomial y2 ´ xpx ´ aqpx ´ bq is irreducible in krx, ys for any field k.
Indeed, if it were not, it would have a linear factor, and one could write

y2 ´ xpx ´ aqpx ´ bq “ pαx ` βy ` γqQpx, yq (3.2)

with α, β and γ constants from k and not both α and β being zero.
If β ‰ 0, one may substitute y “ ´β´1 pαx ` γq into this equation. The right
side then vanishes, and we obtain the polynomial identity

β´2 pαx ` γq2 ´ xpx ´ aqpx ´ bq “ 0,

which is impossible since the left side is a monic cubic polynomial.

If β “ 0, it holds true that α ‰ 0, and we may substitute x “ ´α´1 γ
which makes the right side of p3.2q vanish. The leftside will then be a monic
quadratic polynomial in y which is identically zero; absurd!

Irreducibles in a ufd

3.5 As we saw (Proposition 2.38 on page 35), in a domain prime elements are
always irreducible, but if the domain is factorial, the converse holds as well.
Among domains with an appropriate finiteness condition—like the Noetherian
domains we shall come to—this even characterises the factorial domains
(Lemma 3.4 above).
60 ma4200—commutative algebra

Proposition 3.6 For members of a ufd being prime is equivalent to being irre-
ducible.

Proof: Assume that a is irreducible, and that a|xy. Let x “ p1 ¨ ¨ ¨ ps and

y “ q1 ¨ ¨ ¨ qt be decompositions into irreduicbles. Then of course

xy “ p1 ¨ ¨ ¨ ps ¨ q1 ¨ ¨ ¨ qr

is a factorization into irreducibles as well. On the other hand, xy{a is an

element in A and has a factorization xy{a “ r1 ¨ ¨ ¨ rm into irreducibles, so that

a ¨ r1 ¨ ¨ ¨ r m “ p1 ¨ ¨ ¨ p s ¨ q1 ¨ ¨ ¨ qr

are two equal products of irreducibles. Irreducible factors coinciding up to

order and units, this means that, up to a unit, a is either one of the pi ’s or one
of the qi ’s, or phrased differently, a divides either x or y. o

Greatest common divisors and least common multiples

3.7 In a ufd any two elements have a greatest common divisor; that is an element The greatest common
d so that the principal ideal pdq is minimal among the principal ideals con- divisor (Største felles
divisor)
taining both principal ideals paq and pbq where a and b are the two elements
under consideration. Expressed in terms of divisibility, there is a d so that d|a
and d|b and any other x dividing both a and b divides d; that is x|a and x|b
implies that x|d. The greatest common divisor of a and b is not unique, but
only determined up to an invertible factor; however the principal ideal pdq is
unambiguously defined.
3.8 The two elements also have a least common multiple. This is an element m The least common
in A so that the principal ideal pmq is maximal among the principal ideals multiple (Minste felles
multiplum)
contained in paq and pbq; or phrased in terms of divisibility, it holds that a|m
and b|m, and for any other member x of A it ensues from a|x and b|x that m|x.
Again, merely the principal ideal pmq is unambiguously defined.

Proposition 3.9 In a udf any two elements have a greatest common divisor and a
least common multiple.

Proof: Let a and b be the elements. Proceed to write down factorizations say
a “ p1 ¨ ¨ ¨ pr and b “ q1 ¨ ¨ ¨ qs into irreducibles, and pick up the “common
factors”: Reordering the factors, we find a non-negative integer t so that
ppi q “ pqi q for i ď t and ppi q ‰ pqi q for t ą i. Then d “ p1 ¨ ¨ ¨ pt is a greatest
common divisor. It might of course happen that no ppi q equals any pq j q, in
which case t “ 0, and the greatest common divisor equals one.
To lay hands on a least common multiple, make the product of all the
ideals ppi q and all the ideals pq j q not found among the ppi q’s. This product is a
principal ideal, and any generator serves as a least common multiple. o
 unique factorization domains i 61

Kaplansky’s criterion

3.10 I am especially found of the formulation2 in the following criterion due to 2

The slogan is: "A is
Irving Kaplansky whose proof is an elegant application of the Basic Existence a ufd if and only if
every prime contains a
Theorem. prime".

Proposition 3.11 A domain A is a ufd if and only if every non-zero prime ideal
contains a prime element.

Proof: The implication one way is clear. Let p be a non-zero prime ideal and
consider any of its non-zero elements. It factors as a product of primes, and
one of the factors must lie in p.
To prove the other implication it suffices, in view of Lemma 3.4 on page 58,
to show that any non-zero member of A is either a unit or a finite product of
prime elements, so let Σ be the set of elements in A that can be expressed as
a finite product of one or more prime elements. It is certainly multiplicative
closed, and A having at least one maximal ideal it is not empty (maximal
ideals are prime and contain prime elements by assumption). We contend that
Σ coincides with the set of non-zero non-units of A.
Assume this is not true; that is, there is a member x of A, neither zero nor a
unit, that does not lie in Σ. Then pxq X S “ H since if xy “ p1 ¨ ¨ ¨ pr with the pi ’s Irving Kaplansky
(19171–2006)
being prime elements and y not being a unit, it follows by an easy induction Canadian mathematician
on r that x belongs to S. By the Basic Existence Theorem (Theorem 2.50 on
page 40), there is a prime ideal in A maximal subjected to not meeting S
and containing x. That prime ideal is not the zero ideal, and by assumption
there is therefore a prime element lying in p, which is a contradiction since all
primes lie in Σ. Hence Σ fills up the entire set of non-zero non-units in A. o
An immediate corollary is the following (which also can be proved in several
other ways):

Corollary 3.12 Every pid is a ufd.

Proof: Prime ideals are generated by prime elements. o

Gauss’ lemma and polynomials over udf’s

Every domain A is contained in a canonically determined field called the frac-
tion field of A. The elements are fractions of shape a{b with a and b elements
from A and, of course, b ‰ 0. The arithmetic of these fractions is governed by
the usual rules for rational fractions. For the moment we have not shown such
field exist, but shall assume it. Later on they will be constructed as particular
cases of a general “localization process”.
Several of the domains we have seen are a priori contained in a field, like
?
Z and Zr ds which are subrings of C, and a polynomial ring Crx1 , . . . , xr s
62 ma4200—commutative algebra

of over the complex numbers is contained in the rational function field

Cpx1 , . . . , xr q. And these fields obviously have a fraction field a priori.
The objective of the current section is to establish the result that the poly-
nomial rings over ufd’s are udf’s. That result hinges on a key concept of a
primitive polynomial and a key lemma, the so-called Gauss lemma. We begin
with that before attacking the main assertion.
Johann Carl Friedrich
3.13 For a moment image a polynomial f pxq “ a0 ` a1 x ` . . . ` ar xr with
Gauss
coefficient from the udf A. Extracting the greatest common divisor d of the (1777–1855)
coefficient ai one may write write f “ d ¨ g where g P Arxs is a polynomial German mathematician

whose coefficients are without a common divisor. This naturally leads to

the concept of a primitive polynomial: A polynomial is called primitive if the Primitive polynomials
greatest common divisor of the coefficients equals 1. (Primitive polynomer)

With a splitting f pxq “ c f ¨ gpxq like above with g primitive and c f P

A, it is widespread usage to call the element c f the content of f . It is not Content of a polynomial
unambiguously associated with f , but like a greatest common divisor, unique (innehold)

up to an invertible factor.

Lemma 3.14 (Gauss’ lemma) Assume that A is a ufd. Let f and g be primitive
polynomials in Arxs. Then the product f g is primitive.

It ensues immediately from the lemma that for any two polynomials f and g it
holds true that c f g “ uc f c g with u P A˚ ; in other words, the content depends
up to units on the polynomial in a multiplicative manner
Proof: Write f “ 0ďiďn ai xi and g “ 0ďiďm bi xi . Let d be a non-unit in A.
ř ř

The polynomial f being primitive, there is a least i0 so that d does not divide
ai0 and ditto a least j0 so that d does not divide b j0 . Consider the coefficient of
xi0 ` j0 in the product f g; that is, the sum
ÿ
ai b j .
i` j“i0 ` j0

If i ‰ i0 and j ‰ j0 , either i ă i0 or j ă j0 ; in the former case d|ai and in the

latter d|b j , so in both cases d|ai b j . Hence all terms of the sum are divisible by d
except ai0 b j0 and the sum is not divisible by d. o

3.15 So to our main objective

Theorem 3.16 If A is a ufd, then the polynomial ring Arxs is a ufd. The ir-
reducible elements in Arxs are the irreducibles in A and the primitive, irreducible
polynomials.

Induction and the fact that krxs is a ufd immediately give the corollary that
polynomial rings over fields are ufd’s. Ot is also worth while mentioning that
the same applies to polynomials rings over the integers.

Corollary 3.17 If k is a field krx1 , . . . , xr s is a ufd.

 unique factorization domains i 63

Corollary 3.18 The ring Zrx1 , . . . , xr s are factorial.

Proof of 3.16: We let K be the fraction field of A. Given a polynomial p P Arxs

which we to begin with assume is primitive. It can be factored as p “ f 1 ¨ ¨ ¨ f r
with the f i ’s being irreducible polynomials in Krxs. Multiplying through with
an element from A on gets rid of the denominators of the coefficients of the
f i ’s and subsequently recollecting in one product the common divisors of the
coefficients of each the f i ’s, one arrives at a relation

a ¨ p “ b ¨ g1 ¨ ¨ ¨ g r

where a and b are members of A and the gi ’s are primitive polynomials in

Arxs. By Gauss’ lemma, the product g1 ¨ ¨ ¨ gr is primitive, and since the content
is unique up to a unit, it follows that b “ ua with u P A˚ . This takes care of the
existence.
To the unicity, assume that f 1 ¨ ¨ ¨ f r “ g1 ¨ ¨ ¨ gs in Arxs. Then f i “ ui gi with
ui P K, so clearing denominators we arrive at

a i f i “ bi g i

hence ai “ vi bi with vi P A˚ o
3.19 Factoring a polynomial into a product of irreducibles, or for that matter,
showing a polynomial is irreducible, is often an unpleasant task. That the
polynomial is homogenous might sometimes be helpful because then one
knows a priori that the irreducible components are homogeneous; indeed, one
has the following proposition.

Proposition 3.20 Let R be a graded factorial domain satisfying Rn “ 0 for n ă 0.

Then the irreducible factors of a homogeneous element a are homogeneous.

Proof: Let ai be the irreducible factors of a and develop each ai as the sum
ř
ai “ ν ai, ν of the homogeneous components. Denote by ai, νi the non-zero
component of ai of lowest degree. Since the degree of every element is non-
negative, it holds true that
ź ź
a“ ai “ ai, νi ` terms of higher degree,
i i
ś
and i ai is non-zero as R is assumed to be a domain. But now, homogeneous
components are unambiguously defined, and a is homogenous. Hence the
sum of the high degree terms vanishes, and we have expressed a as a product
of homogeneous elements
ź
a“ ai, νi .
i
By induction on the degree, each ai, νi has only homogeneous irreducible
components, and the same applies therefore to a. o
64 ma4200—commutative algebra

Example 3.5 The polynomial f “ xy ´ wz is irreducible. If f were the product

of two linear terms, each variable would occur in precisely one of them since
no term of f is a square, hence cross-terms like xw or xz would appear in f . K

Example 3.6 The polynomial y p ´ x q are irreducible when p and q are rel-
atively prime. To see this give krx, ys the grading for which deg x “ p and
deg y “ q. An irreducible polynomial has both non-zero terms of the form α ¨ yn
and of the form β ¨ x m , unless it equals x or y. If it additionally it is homoge-
neous, it holds true that nq “ mp and n “ ap and m “ bq for some natural
numbers a and b. It follows that y p ´ x q is irreducible since any irreducible
factor is not reduced to x or y. K
âĂć

Problem 3.1 Assume that ppxq is a monic polynomial in Zrxs wich factors
as ppxq “ rpxqspxq in Qrxs. Show that rpxq and spxq both lie in Zrxs and are
monic. Hint: multiply by the least common multiples of denominators of
coefficents and appeal to Gauss’s lemma. M
?
Problem 3.2 Let d “ 2k be an even integer and consider the ring Zri 2ks ˇ
?
with k an odd natural number. Show that p “ p2, i 2kq is not a principal ideal
?
but that p2 “ p2q. Prove that Zri 2ks is not a udf M
?
Problem 3.3 Consider the ring Zri 14s. Prove that ˇ
? ?
34 “ p5 ` 2i 14qp5 ´ 2i 14q
?
and show that the involved elements all are irreducible elements of Zri 14s.
M
Lecture 4

Modules

Preliminary version 1.05 as of 2018-10-02 at 14:37 (typeset 3rd December 2018 at 10:03am)—To be
completed. Prone to misprints and errors and will change.
2018-08-21: Rewritten paragraph 4.23 Proposition 4.26 corrected; added exercise 4.20
27/08/2018 Finished the paragraph 4.48 on page 87 about split exact sequences;
29/08/2018 Reworked section 4.5 about exact sequences.
15/09/2018 Reworked the snake section. Added exercise 4.15. Added examples 4.12, 4.13 and
4.14.
17/09/2018 Added example 4.18 exercises 4.25 and 4.26.
18/09/2018 Added example 4.24 on page 93; Added exercise 4.38 on page 95;
25/09/2018 Rewritten proof of Splitting Criterion, Prop 4.51; Corrected misprints; changed
exercise 4.38 slightly. Added a new exercise 4.16.
26/09/2018 Added an exercise about elliptic curves and free modules, ex 4.27 on page 84.
Added exercise 4.34 on page 89.
27/09/2018 Corrected several minors misprints.
Extended problem 4.34 about additive functors. Several minor improvements.
02/10/2018 Added exercise 4.5. Added example 4.17 Added a simple lemma snake-section
(lemma 4.65).

Along with every ring comes a swarm of objects called modules; they
are the additive groups on which the ring acts. The axioms for modules
resemble the axioms for a vector spaces, and modules over fields are in fact
just vector spaces. Over general rings however, they are much more diverse
and seriously more complicated than vector spaces. Ideals for instance, are
modules, and any over-ring is a module over any subring to mention two
instances. An abelian group is nothing but a Z-module and a module over the
polynomial ring krts over a field k is just a vector space over k endowed with
an endomorphism; so the module theory encompasses all abelian groups and
the entire linear algebra!

4.1 The axioms

4.1 A module M over the ring A, or an A-module as one also says, has two
layers of structures. It is endowed with an underlying structure as an abelian Modules (Moduler)

group, which will be written additively, on top of which lies a linear action of
the ring A. Such an action is specified by a map A ˆ M Ñ M, whose value at
pa, mq will be denoted by a ¨ m or simply by am. It is subjected to the following
66 ma4200—commutative algebra

four conditions:

o apm ` m1 q “ am ` am1 ;

o pa ` a1 qm “ am ` a1 m;

o 1 ¨ m “ m;

o a ¨ pa1 ¨ mq “ paa1 q ¨ m.

where a, a1 P A and m, m1 P M are arbitrary elements. The first condition

expresses that the action is A-linear; that is, the map A Ñ HomSets pM, Mq
sending a to the “multiplication-by-a-map” m ÞÑ am takes values in the ring
HomAb pM, Mq of group-homomorphism. The last three requirements express
that this map is a ring homomorphism; in other words, it is the ring A that
acts.
4.2 One recognizes these conditions from linear algebra; they are word for
word the vector space axioms, the sole difference being that A is not required
to be a field but can be any ring. In case A is a field k, there is nothing new; a
k-module is just a vector space over k. One should not draw this analogy to
far however, general modules are creatures that behave very differently than
vector spaces.

Examples

4.1 The primordial examples of modules over a ring A are the ideals a in A
and the quotients A{a. Already here, the difference from the case of vector
spaces surfaces; fields have no non-zero and proper ideals. There are also the
"subquotients" b{a of two nested ideals. Of course, these examples include the
ring itself; every ring is a module over itself.

4.2 Another examples more in the flavour of vector spaces are the direct
sums of copies of A. The underlying additive group is just the direct sum
A ‘ A ‘ . . . ‘ A of a certain finite number, say r, copies of A. The elements are
r-tuples pa1 , . . . , ar q, and addition is performed component-wise. The action of
A is also defined component-wise: a ¨ pa1 , . . . , ar q “ pa ¨ a1 , . . . , a ¨ ar q. We insist
on this being an additive1 construction and shall write rA for this module, not 1
The direct sum plays
Ar as is common usage in linear algebra. an additive role in the
category of A-modules,
the multiplicative role
4.3 Another familiar class of modules are the abelian groups. They are noth-
is taken by another
ing but modules over the ring Z of integers. An integer n acts on an element construct, the tensor
from the module by just adding up the appropriate number of copies of the product. But that’s for
a later chapter.
element and then correcting the sign.

4.4 Over-rings form an abundant source of examples—when A is a subring

of B, multiplication by elements from A makes B into an A-module. So for
 modules 67

instance, krx, ys is a krxs-module as is krx, x´1 s. And krxs will be a module over
the subring krx2 , x3 s.
? ?
If m and n are two natural numbers the ring Zr m{ns is a Zr ms-module
(and, of course, it is a Z-module, and also Zr1{ns-module for that matter).
More generally, any ring homomorphism φ : A Ñ B induces an A-module
structure on B through the action a ¨ b “ φpaqb of an element a P A on b P B.
One says that B is an A-algebra. Algebras (Algebraer)

4.5 Suppose that k is a field. Giving a krts-module is the same as giving a

k-vector space M and an endomorphism of M; that is, a linear map τ : M Ñ M.
A polynomial pptq in krts acts on M as pptq ¨ m “ ppτqpmq.

Problem 4.1 Let A and B be two rings and assume that B has an A-module ˇ
structure compatible with the ring structure; i.e. a ¨ bb1 “ bpa ¨ b1 q. Show that
there is ring homomorphism A Ñ B inducing the module structure. M

Homomorphisms between modules

A new concept in mathematics is always followed by a fresh class of relevant
maps; so also in our present case of modules. An A-module homomorphism Module homomorphisms
φ : M Ñ N between two A-modules M and N is a homomorphism of the (Modulhomomorfier)

underlying abelian groups that respects the action of A; that is, φpamq “ aφpmq
for all a’s in A and all m’s in M. Simply said, a module homomorphism is just
an A-linear map from M to N. With this notion of morphisms, the A-modules
form a category Mod A .
4.3 The set Hom A pM, Nq is naturally contained in the set HomZ pM, Nq of
group homomorphism from M to N (which are just the additive maps) and
consists of those commuting with the action of A on both M and N. It is well-
known that the sum of two additive maps is additive (left for the zealous
students to check ), and when both commute with the actions of A, the sum
does so as well. Therefore Hom A pM, Nq is an abelian group, and defining a ¨ φ
as the map sending m to aφpmq gives it an A-module structure. One must of
course verify that a ¨ φ is A-linear, but this is easy: If b P A is another element, 2
For non-commutative
one finds rings A the set
a ¨ φpbmq “ apbφpmqq “ bpaφpmqq, Hom A pM, Nq is merely
an abelian group in
where the first equality holds since φ is A-linear and the second because A is general. It does not
carry an A-module
commutative2 . structure unless fur-
ther hypotheses are
4.4 The module Hom A pM, Nq depends functorially on both the variables M
imposed on A.
and N, and historically, it was one of the very first functors to be studied. The
dependence on the first variable is contravariant—the direction of arrows are
reversed— whereas the dependence on the second is covariant—directions
68 ma4200—commutative algebra

are kept. The induced maps are just given by composition. Of course, such
constructions are feasible in all categories, what is special in Mod A is that
Hom A pM, Nq is an A-module and the induced maps are A-linear. The techni- M
φ
/N
cal name is that category Mod A has internal homs—the set of maps stay within
ψ
the family! ψ˚ pφq

To be presice let M, N and L be three A-modules and ψ : N Ñ L an A-linear L

map. Sending φ to ψ ˝ φ yields an associated map

ψ˚ : Hom A pM, Nq Ñ Hom A pM, Lq.

It is A-linear, and if ψ : L Ñ L1 is another A-linear map, one has pψ1 ˝ ψq˚ “

ψ˚1 ˝ ψ˚ . M
φ
/N
In a similar fashion, the contravariant upper-star version
ψ
˚
φ˚ pψq 
φ : Hom A pN, Lq Ñ Hom A pM, Lq, L

which sends ψ to ψ ˝ φ, is A-linear as well, and it is functorial; i.e. pφ1 ˝ φq˚ “

φ˚ ˝ φ1˚ for composable maps φ and φ1 .
4.5 It follows readily from the involved maps being A-linear that the composi-
tion of composable maps is an A-bilinear operation. That is, one has

φ ˝ paψ ` a1 ψ1 q “ aφ ˝ ψ ` a1 φ ˝ ψ1 and paφ ` a1 φ1 q ˝ ψ “ aφ ˝ ψ ` a1 φ1 ˝ ψ

where the maps are composable3 and a and a1 denote ring elements. 3
That is, φ and φ1 are
maps from M to N and
ψ and ψ1 from N to L.
Problems

4.2 Show that there is a canonical isomorphism Hom A pA, Mq » M. ˇ

4.3 Let A be an integral domain and a Ď A a non-zero ideal. Show that ˇ

Hom A pA{a, Aq “ 0.

4.4 Let p and q be two prime numbers. Show that HomZ pZ{pZ, Z{qZq “ 0 if ˇ
p ‰ q, and that HomZ pZ{pZ, Z{pZq » Z{pZ.

4.5 Let a and b be two ideals in the ring A. Show that there is a canonical
isomorphism Hom A pA{a, A{bq » pa : bq{b.

Submodules

4.6 A submodule N of an A-module M is a subgroup closed under the action Submodules (Undermod-
of A; in other words, for arbitrary elements a P A and n P N it holds true that uler)

an P N, and of course, N being a subgroup the sum and the difference of two
elements from N belong to N.
 modules 69

Examples

4.6 Ideals in the ring A are good examples of submodules, and by definition
they constitute all submodules of A.
4.7 If a Ď A is an ideal and M an A-module, the subset aM of M formed by all
multiples am with a P a and m P M is a submodule.
4.8 Given an ideal a in A. The set p0 : aq M of elements in M annihilated by all
members of a form a submodule. It holds true that Hom A pA{a, Mq “ p0 : aq M .

The lattice of submodules

4.7 Just like the ideals in A the submodules of a given A-module M form a
partially ordered set4 denoted I pMq under inclusion. 4
The set I pMq forms
Ş
The intersection iP I Ni of a collection tNi uiP I of submodules of M is a what is called a com-
plete lattice. A partially
submodule. It is the largest submodule of M contained in all the submodules ordered set is called a
from the collection. In the similar way, the smallest submodule containing lattice when every pair
ř of elements possesses
all the modules in the collection is the sum iP I Ni whose elements are finite a least upper bound
A-linear combinations of elements from the Ni ’s; that is, elements shaped like and a greatest lower
bound, and it is said
ÿ to be complete if the
ai mi , (4.1) same is true for any
subset. In our case the
greatest lower bound is
where mi P Ni , and the ai ’s are elements from A only finitely many of which the intersection and the
are non-zero. More generally, for any set S Ď M there is a smallest submodule least upper bound the
of M containing S; it is called the submodule generated by S and consists of sum.

elements like in (4.1) but with the mi ’s from S. Submodules generated by

elements (Undermoduler
generert av elementer)
Kernels and images

4.8 An A-module homomorphism φ : M Ñ N is in particular a group ho-

momorphism and as such has a kernel and an image. Both these subgroups
are submodules as well; this ensues from the equality aφpmq “ φpamq satis-
fied by A-linear maps. Indeed, one immediately sees that the image is closed
under multiplication by elements from A, and if φpmq “ 0, it follows that
φpamq “ aφpmq “ 0 as well.
Problem 4.6 Let M and N be A-modules and φ : M Ñ N an A-linear map.
Show that for every submodule L Ď N the inverse image φ´1 pLq is a submod-
ules of M. M

Quotients
70 ma4200—commutative algebra

4.9 Just as with ideals in a ring, one can form quotient of a module by a
submodule.
Let M be the module and N the submodule. From the theory of abelian
groups we already know that the two underlying additive groups have a
quotient group M{N, which is formed by the cosets rms “ m ` N. Endowing
M{N with an A-module structure amounts to telling how elements a P A act
on M{N, and one does this simply by putting a ¨ rms “ rams. Of course, some
verifications are needed. Routinely one checks that the class rams does not
depend on the representative m, which is the case since apm ` Nq “ am ` aN “
am ` N. Secondly, the module axioms in paragraph 4.1 must be verified; this is
straightforward and left to the zealous students.
4.10 By definition the canonical map π : M Ñ M{N that sends m to the class
rms is A-linear, and characterized by the universal property that any A-linear
map vanishing on N factors through it:

Proposition 4.11 (Universal propety of quotients) Let N be a sub-

module of the A-module M. The quotient map π : M Ñ M{N enjoys the following
universal property. For every A-module homomorphism φ : M Ñ L with N Ď ker φ
there exists a unique A-linear map ψ : M{N Ñ L so that φ “ ψ ˝ π.

Proof: The proof is mutatis mutandis the same as for (abelian) groups. The M
π
/ M{N
map φ is constant on the residue classes rms “ m ` N, and ψprmsq is defined
as (and compelled to be) that constant value. Since φ is A-linear, the constant φ
ψ

value on rm ` m1 s “ m ` m1 ` N equals φpmq ` φpm1 q, and on rams “ am ` N it  

L
is aφpmq. Hence ψ is A-linear. o

Corollary 4.12 (The first ismorphism theorem) A surjective A-linear

map φ : M Ñ N induces an isomorphism M{ ker φ » N.

Proof: By the universal property φ factors through a map ψ : M{ ker φ Ñ N.

This is surjective since φ is, and injective since it kills the kernel. (That a class
rxs goes to zero, implies that φpxq “ 0, hence x P ker φ). o
M
π
/ M{ ker φ
One easily establishes the two following results. The analogue assertions for
abelian groups are well known, and the proofs persist being valid for modules »
φ
as well.  
N
Proposition 4.13 Let π : M Ñ M{N be the quotient map. The “inverse-image-
map” π ´1 : I pM{Nq Ñ I pMq is a one-to-one correspondence between submodules
of M containing N and submodules of M{N. It respect inclusions, arbitrary intersec-
tions and arbitrary sums.

Proof: Routine. o

Proposition 4.14 (The second ismorphism theorem) If N and N 1 are

two submodules of M, there are canonical isomorphisms where in the second one
assumes that N 1 Ď N:
 modules 71

o pN ` N 1 q{N 1 » N{N X N 1 ;

o pM{N 1 q{pN{N 1 q » M{N.

Proof: Routine. o

4.15 In a famous paper Grothendieck introduced axiomatically the notion

of an abelian category. The axioms reflect the main categorical properties of Abelian category (Abelske
kategorier)
the module category Mod A . Among the requirements is that all hom-sets
be abelian groups and all compositions be bilinear (like we discussed in
paragraph 4.4) and all maps have kernels and a cokernels. Moreover, there is
an axiom which fabulously can be formulated as “the kernel of the cokernel
equals the cokernel of the kernel”.
Let us also mention that a category whose hom-sets are abelian groups and
whose compositions are bilinear is called an additive category. When, as is true Additive categories
of Mod A , the hom-sets are A-modules and the compositions A-bilinear, it is (Additive kategorier)

said to an A-linear category. A-linear category (A-

lineære kategorier)
4.16 The definition of the cokernel in a categorical vernacular must be for-
The cokernel (Kokjernen)
mulated exclusively in terms of arrows and therefore given as a universal
property. The cokernel of an A-linear map φ : N Ñ M is an A-linear map
π : M Ñ coker φ such that π ˝ φ “ 0, and which is universal with respect to
that property. By The First Isomorphism Theorem (Theorem 4.12) the quotient
M{ im φ fulfils that requirement, and hence serves as the cokernel of φ. N
φ
/M π
/ coker φ
The two stablemates kernel and cokernel are from a categorical viewpoint
ψ
dual concepts, and a definition of the kernel just in terms of arrows is indi-
  |
0

cated with a diagram in the margin. L

/N φ
/M
Proposition 4.17 Every A-module homomorphism φ : M Ñ N has a kernel, an
ker φ
a O ?
image and a cokernel. ψ
0

L
Proof: As mentioned above, the quotient N{ im φ serves as the cokernel. The
kernel is the usual tangible subset of M consisting of the elements sent to zero.
o

Example 4.9 In Mod A the fabulous axiom cited above boils down to the
obvious: The kernel of the cokernel and the cokernel of the kernel both equal
the image. (If you find this rather cryptical than obvious, think twice!!) K

Example 4.10 The submodules aM where a is an ideal in A form a particular

important class of submodules of M. A quotient M{aM inherits a natural
structure of module over the quotient ring A{a; indeed, the product x ¨ m
between elements x P A and m P M{aM only depends on the residue class rxs
of x modulo a since px ` aqm “ xm ` am “ xm for any a P a.
So for instance, the module M itself is in a canonical way a module over
A{ Ann M; or for that matter, over A{a for any a that kills M. K
72 ma4200—commutative algebra

Example 4.11 Given a ring map φ : A Ñ B between two rings A and B.

This allows one to consider any B-module M as an A-module just by letting
members a of A act on elements m P M as φpaq ¨ m, and in this ways one obtains
a natural functor from ModB to Mod A . Sometimes one sees the notation Mφ
or M A for this module, but to avoid symbols looking like overdecorated
Christmas trees, letting the A-module structure be tacitly understood and
simply writing M is to prefer in most instances. K

4.2 Direct sums and direct products

There are two important and closely related constructions one can make in the
category Mod A of A-modules namely the direct product and the direct sum.
There is no restriction on the cardinality of the involved families, but during
practical work in algebraic geometry or number theory one mostly meets finite
families, and in that case the two constructs agree.

Direct products

4.18 In this section we work with a collection tMi uiP I of modules over the
ring A. From earlier courses direct product of a collection of groups is well
ś
known, and here we consider the direct product iP I Mi of the underlying Direct products of mod-
additive groups of the Mi ’s. The elements are strings or tuples pmi qiP I indexed ules (Direkte produkter
av moduler)
by the set I, and the addition is performed component-wise; i.e. pmi q ` pm1i q “
pmi ` m1i q. In case I is finite, say I “ t1, . . . , ru, an alternative notation for a
tuple is pm1 , . . . , mr q. The actions of A on the different Mi ’s induce an action
ś
on iP I Mi , likewise defined component for component; a ring element a acts
like a ¨ pmi q “ pa ¨ mi q. The module axioms in paragraph (4.1) are cheap to verify,
ś
and we have an A-module structure on iP I Mi .
ś
The projections πi : iP I Mi Ñ Mi are A-linear since the module operations
of the product are performed component-wise.

Direct sums
À
4.19 The direct sum of the module collection tMi u is denoted iP I Mi , and Direct sums of modules
is defined as the submodule of the direct product consisting of strings m “ (Direkte summer av
moduler)
pmi qiP I with all but a finite number of the mi ’s vanishing—or so to say, there
is an i0 (that depends on m) such that mi “ 0 for i ě i0 . Obviously this is a
submodule, and again, the projections are A-linear.
When the index set I is finite requiring strings to merely have finitely many
non-zero components imposes no constraint, so in that case the direct sum
and the direct product coincide. However, when the index set I is infinite, they
are certainly not isomorphic; they are not even of the same cardinality. For
instance, the direct sum of countably many copies of Z{2Z is countable (being
 modules 73

the set of finite sequences of zeros and ones) whereas the direct product of
countably many copies of Z{2Z has the cardinality of the continuum (every
real number has a 2-adic expansion).

Universal properties

4.20 Both the product and the direct sum are characterised by universal prop-
erties. It is noticeable that these properties are dual to each other; reversing all
arrows in one, yields the other. For this reason the direct sum is frequently
called the co-product in the parlance of category theory.
We proceed to describe the two universal properties and begin with the N
φ
/ śiPI Mi
direct product. In that case, the set-up is formed by an A-module N and a
πi
collection of A-linear maps φi : N Ñ Mi . The conclusion is there exists a
& 
φi
ś
unique A-linear map N : N Ñ iP I Mi such that πi ˝ φ “ φi . Indeed, this Mi
amounts to the map φpnq “ pφi pnqqiP I being A-linear.
In the case of the direct sum the universal property does not involve the
À
projections, but rather the natural inclusions ι j : M j Ñ iP I Mi that send an
m P M j to the string having all entries equal to zero but the one in slot j which
equals m. The given maps are maps φi : Mi Ñ N, and the conclusion is that
N of φ À
Mi
iPI
O
À
there exists a unique map iP I Mi Ñ N so that φ ˝ ι j “ φj . The map φ is
compelled to be defined as φi
ιi

` ˘ ÿ Mi
φ pmi qiP I “ φi pmi q,
iP I

and this is a legitimate definition since merely finitely many of the mi ’s are
non-zero.

Problem 4.7 Work out all the details in the above reasoning. M

4.21 With the stage rigged as in the previous paragraphs we round off the
discussion of the universal properties of direct products and direct sums by
offering equivalent formulations in terms of the hom-modules:

Proposition 4.22 There are canonical isomorphisms

À ś
o Hom A p iP I Mi , Nq » iP I Hom A pMi , Nq;
ś ś
o Hom A pN, iP I Mi q » iP I Hom A pN, Mi q.

Notice that in the first isomorphism, which involves the contravariant slot, the
direct sum is transformed into a direct product. It further warrants a special
comment that when the index set is finite, the direct product coincides with
the direct sum, and the proposition may be summarized by saying that the
hom-functor commutes with finite direct sums. In the vernacular of category
theory one says that it is additive in both variables.
74 ma4200—commutative algebra

Problem 4.8 Figure out the precise definitions of the isomorphisms in

Proposition 4.22 above. Hint: The key word is universal properties. M
À
4.23 We shall identify each module M j with the image ι j pM j q in iP I Mi under
the natural inclusion ι j ; that is, with the submodule of elements having all
entries zero except at slot j.
Fix one the indices say ν. Forgetting the ν-th entry in string pmi qiP I gives a
string pmi qiP I ztνu indexed by the subset Iztνu of indices different from ν. The
operations in direct sums being performed component-wise, this is clearly an
A-linear assignment; hence gives an A-linear map
À
Mi /À Mi
iP I iP I ztνu

The kernel is obviously equal to Mν (identified with the submodule of the

direct sum wher merely the ν-th entry is non-zero), and the Isomorphism
theorem (Theorem 4.12 on page 70) yields an isomorphism
`à ˘ à
Mi {Mν » Mi
iP I iP I ztνu

The slogan is: Killing one addend of a direct sum yields the sum of the others.

Problem 4.9 Generalize the slogan above to any sub-collection: Let J Ď I be a

subset. Prove that there is a canonical isomorphism
à à à
pMi q{ MJ » Mi
iP I jP J iP I z J

Split submodules, direct sums and idempotent maps

It is of interest to have criteria for a submodule to be a direct summand of the
surrounding module. We know every subvector space is a direct summand
(bases can be extended to the containing space), but for general rings most
submodules are not.
4.24 A synonym for a submodule N Ď M to be a direct summand, is that N lies
split in M—this of course means that there is another submodule N 1 so that Split submodules (Splitt
M “ N ‘ N 1 — and just as in linear algebra, the submodule N 1 is called a com- undermoduler)

plement to N. Equivalently, every element m from M can be unambiguously Complementary submod-

expressed as a sum m “ n ` n1 with n P N and n1 P N; or phrased differently, ules (Komplementære
undermoduler)
the two conditions N X N 1 “ 0 and N ` N 1 “ M are satisfied.
Example 4.12 A principal ideal pnq i Z, with n neither being zero nor plus-
minus one, is not a direct summand of Z since every other ideal contains
multiples of n. K
 modules 75

Example 4.13 Cheap but omnipresent examples of non-split submodules are

ideals a in domains A. By paragraph 4.23 , any complement of a would be
isomorphic to A{a. If a is a non-zero proper ideal, the quotient A{α contains
non-zero elements killed by a which is absurd since A was assumed to be a
domain. K
Example 4.14 A ring that is not a domain, may possess non-zero proper
ideals lying split. The simples example is the direct product A “ k ˆ k of two
fields. The subspaces k ˆ p0q and p0q ˆ k are both ideals. K
4.25 When we treated rings that were direct products of other rings, the no-
tion of idempotent elements turned out to be quite useful. This notion can
be generalizations in several directions and in various contexts, the virtue
of idempotents always being that they express some kind of “direct decom-
position”. In our present context of modules over a ring A, an A-linear map
e : M Ñ M is said to be idempotent if e2 “ id M . Idempotent map of
modules (Idempotent
Proposition 4.26 Let M be an A-module. If e is an idempotent endomorphism of modulavbildning)
M, the map M Ñ ker e ‘ im e sending m to pm ´ epmq, epmqq is an isomorphism.
A submodule N of M lies split if and only if there is an idempotent endomorphism
e : M Ñ M with im e “ N.

Proof: Suppose to begin with that e is an idempotent endomorphism of M.

We contend that M “ im e ‘ ker e. Indeed, it holds true that m “ pm ´ epmqq `
epmq. Obviously epmq lies in im e and m ´ epmq lies in the kernel ker e because e
is idempotent:

epm ´ epmqq “ epmq ´ e2 pmq “ epmq ´ epmq “ 0.

On the other hand, im e X ker e “ 0 since if x “ epyq lies in ker e, it holds that
epxq “ e2 pyq “ epyq “ x, and hence x “ 0. This takes care of one implication.
Suppose then that M “ N ‘ N 1 and let π : M Ñ N and ι : N Ñ M respec-
tively be the projection and the inclusion map, then ι ˝ π “ id N . Put e “ π ˝ ι.
Then
e2 “ pι ˝ πq ˝ pι ˝ πq “ ι ˝ pπ ˝ ιq ˝ π “ ι ˝ π “ e.
o

Problems

4.10 Assume that for each i P I there is given a submodule Ni Ď Mi . Prove

À À
that iP I Ni is a submodule of iP I Mi in a natural way and that there is a
À À À
natural isomorphism iP I Mi {Ni » p iP I Mi q{p iP I Ni q.

4.11 Let M1 and M2 be two submodules of the A-module M whose intersec- ˇ

tion vanishes; that is, M1 X M2 “ p0q. Prove that M1 ` M2 is naturally isomor-
phic with the direct sum M1 ‘ M2 . Hint: Establish that any m P M1 ` M2 can
76 ma4200—commutative algebra

be expressed as m “ m1 ` m2 with m1 and m2 unambiguously defined elements

in respectively M1 and M2 .

4.12 Let tMi uiP I be a family of submodules of the A-module N. Assume that
they comply to the following rule: For any index ν P I and any finite subset
ř
J Ď I not containing ν, the intersection of Mν and jP J M j vanishes; that is,
ř ř
M X iP J M j “ p0q. Prove that iP I Mi is isomorphic with the direct sum
Ài
i P I Mi .

4.13 Generalise Proposition 4.26 in the following way. Assume that e1 , . . . , er

are mutually orthogonal idempotent endomorphisms of the A-module M.
ř
Suppose they satisfy ei “ 1. Show that putting Mi “ ei pMq one obtains a
À
decomposition M “ i Mi . Prove the converse: If such a decomposition exists,
exhibit a collection of idempotents inducing it.
ś
4.14 Extend Proposition 4.27 to an arbitrary direct product A “ iP I Ai :
ś
Prove that any A module M is isomorphic to a product iP I Mi where Mi is
an Ai -module unambiguously associated with M.

Modules over direct products

Consider a direct product A “ A1 ˆ . . . ˆ Ar of a collection tAi u1ďiďr of rings.
We aim at describing all modules over A in terms of modules over the factors
Ai . The crucial construction being the following. Suppose given an A-module
À
Mi for each i. The additive group i Mi has a natural A-module structure and
a string a “ pai qiP I of ring elements acts on a string m “ pmi qiP I of module
elements according to the rule a ¨ m “ pai ¨ mi q (once more the axioms come for
free the action being defined component-wise). We contend that all A-modules
are shaped like this.
Proposition 4.27 Let A1 , . . . , Ar be rings and put A “ A1 ˆ . . . ˆ Ar . Assume
that M is an A-module. Then there are canonically defined submodules Mi of M that
À
are Ai -modules and are such that M » i Mi .
Proof: Let e1 , . . . , er be the idempotents arising from the decomposition of A
as a product; in other words they are the “basis elements” ei “ p0, . . . , 0, 1, 0, . . . , 0q
with the 1 located in slot i. Let ai be the kernel of the projection A Ñ Ai ;
that is, ai is the ideal generated by the idempotents e j other than ei . The set
Mi “ ei M is an A-submodule of M killed by ai ; hence it is an Ai -module.
ř À
Since i ei “ 1, it follows that M “ i Mi . o
4.28 Let us take closer look at the case when A is a direct product of finitely
many fields; say A “ k1 ˆ . . . ˆ kr . Then Proposition 4.27 above tells us that all
modules over A are shaped like direct sums V1 ‘ . . . ‘ Vr with each Vi a vector
space over k i .
 modules 77

Unions of submodules
There is no reason that the union of a collection of submodules in general
should be a submodule; no more than the union of two lines through the
origin in space is a plane! This an additive issue (unions of submodules are
obvious closed under multiplication by ring elements) and takes place in
most abelian groups: If H1 and H2 are two subgroups of an abelian group H,
neither containing the other, their union cannot be a subgroup. Indeed, if x1
lies in H1 but not in H2 and x2 in H2 but not in H1 , the sum x1 ` x2 cannot
belong to H1 Y H2 . Suppose for instance it belonged to H1 , then x2 would lie
there too which it was chosen not to.
4.29 However, there is one natural condition that ensures the union to be a
submodule. One says that the collection is directed if for any two modules Directed collection of sub-
modules (Rettet samling
in the collection there is a third containing both; that is—if tMi uiP I is the
av undermoduler)
collection—for any pair Mi and M j there should be an index k so that Mi Ď Mk
Ť
and M j Ď Mk . The union iP I Mi will then be closed under addition (and as
multiplication poses no problem, will thence be a submodule); indeed, let x
and y be two members of the union. This entails there are indices i and j so
that x P Mi and y P M j , and the collection being directed, one may find an
index k so that Mi Y M j Ď Mk . Both elements x and y then lies in Mk , and hence
their sum does as well. So the sum belongs to the union. We have proven:

Proposition 4.30 Let tMi uiP I be a directed collection of submodules of the A-

Ť
module M. Then the union iP I Mi is a submodule.

4.3 Finitely generated modules

A particular class of modules are the finitely generated ones for which he the-
ory is riche. But be aware that many modules arising for instance in algebraic
geometry are not finitely generated.
4.31 A particularly class of modules are the finitely generated modules as the Finitely generated
name indicates, they are modules M for which there is a finite set of elements modules (Endelig
genererte moduler)
that generate M. In other words, one may find elements m1 , . . . , mr from M
ř
such that any m P M can be expressed as a linear combination m “ i ai mi
with the ai ’s being elements from A.
Example 4.15 Several important and natural modules are non finitely gen-
erated. One example can be the Z-module B “ Zrp´1 s where p is a nat-
ural number. For each non-negative integer i we consider the submodule
Bi “ Z ¨ p´i . Because p´i “ p ¨ p´i´1 , it holds true that Bi Ď Bi`1 , and the Bi ’s
form an ascending chain of submodules. Now, every element in B is of the
form a ¨ p´n for some n, in other words one has B “ i Bi . Any finite set of
Ť

elements from B is contained in some BN for N sufficiently large (just take N

larger than all exponents of p appearing in the denominators), so if if finitely
78 ma4200—commutative algebra

many elements generate B it would hold true that BN “ B for some N. This is
obviously absurd, as p can appear in a denominator with any exponent. K

Cyclic modules

4.32 Modules requiring only a single generator are said to be cyclic or mono- Cyclic modules (Sykliske
genic, and they are omnipresent. Among the ideals, the principal ideals are moduler)
Monogenic modules
precisely the cyclic ones, and more generally, if M is any module and m P M (Monogene moduler)
an element5 , the submodule Am “ t a ¨ m | a P A u is cyclic. 5
The zero module is
Now, assume that M is a cyclic A-module and let m P M be a generator. counted among the
cyclic ones, so m “ 0 is
Multiplication induces an A-linear map φ : A Ñ M that sends a to am, and admitted.
this map is surjective since m was chosen as a generator. The kernel consists
by definition of those a’s that kill m, or which amounts to the same, that kill
M. Hence ker φ “ Ann M, and by Corollary 4.12 on page 70, we arrive at an
isomorphism M » A{ Ann M.

Lemma 4.33 Every cyclic A-module M is isomorphic to A{ Ann M.

So the cyclic modules are up to isomorphism precisely the quotient A{a of A

by an ideal a; notice that A itself is cyclic, corresponding to a “ 0. The name
cyclic is inherited from the theory of groups; the cyclic groups being those
generated by a single element, in other words, those shaped like Z{nZ or Z.
The ideal a is of course uniquely determined by the isomorphism class of M
as an A-module (it equals the annihilator Ann M of M), but different ideals
may give rise to quotient that are isomorphic as rings (but of course not as
modules). For instance, the quotients Crxs{px ´ aq with a P C are all isomorphic
to C.

Simple modules

4.34 The simplest modules one can envisage are the ones without other sub-
modules than the two all modules have, and they are simply called simple: A
non-zero A-module M is said to be simple if it has no non-zero proper submod- Simple modules (Simple
ule. Simple modules are cyclic and any non-zero element generates; indeed, moduler)

if m ‰ 0, the submodule Am is non-zero (as m lies in it) and hence equals M

since M has no proper non-zero submodules. Lemma 4.33 above gives that
M is of the form A{ Ann M. Moreover, the annihilator ideal Ann M must be a
maximal ideal since if Ann M Ď a, the quotient a{ Ann M is a submodule of M
which either equals 0 or M. In the former case Ann M “ a, and in the latter it
holds that a “ A. Thus all simple A-modules are characterized:

Proposition 4.35 An A-module M is simple if and only if it is cyclic and its

annihilator Ann M is a maximal ideal; i.e. M is simple if and only if M is isomorphic
to A{m for some maximal ideal m.
 modules 79

Problems

4.15 Let N be a submodule of M. Show that if N and M{N both are finitely ˇ
generated, then M is finitely generated as well. Give an example of modules
M and N so that M and M{N are finitely generated but N is not.

4.16 Let N and L be submodules of the A-module M. If N X L and N ` L are ˇ

finitely generated, show that both N and L are finitely generated.

4.17 Show that krx, x´1 s is not a finitely generated module over krxs. ˇ

4.18 Show that krxs{krx2 , x3 s is a cyclic module over krx2 , x3 s. What is the ˇ
annihilator? What can you say about krxs{krx2 , x p s where p is an odd prime?

4.19 Assume that k is a field. Consider the polynomial ring krxs as a module ˇ
over the subring krx3 , x7 s. Prove it is finitely generated by exhibiting a set of
generators. Determine the annihilator of the quotient krxs{krx3 , x5 s.

4.20 (Schur’s lemma.) Assume that if M and N are two simple A-module that ˇ
are not isomorphic. Prove that Hom A pM, Nq “ 0. Prove that Hom A pM, Mq “
A{ Ann M.

4.4 Bases and free modules

Just like for vector spaces one says that a set of generators tmi uiP I (not neces-
sarily finite) is a basis for M if every element from M can be written as linear Bases for modules (Basis
combination of the mi ’s in only one way; that is, the coefficients ai in an ex- for en modul)
ř
pression m “ i ai mi are unambiguously determined by m. But be aware that
unlike what is the case for vector spaces, most modules do not have a basis.
Example 4.16 The two elements x and y generate the ideal px, yq in the
polynomial ring krx, ys, but do not form a basis since the element xy can be
expressed as two different linear combinations, namely6 one has xy “ x ¨ y “ 6
In the first expression
y ¨ x. And of course, x and y form a minimal set of generators in the sense x is the coefficient and
y the generator, where
that one can not do without either, so even minimal generator sets are not as in the second it is
necessarily bases. The natural question then arises: Can the ideal px, yq be the other way around;
y is the coefficient and
generated by one element? The answer is no! A generator would divide both x x the generator.
and y which is absurd.

Problem 4.21 Show that the property, familiar from the theory of vector
ř
spaces, that i ai mi “ 0 implies that ai “ 0 is sufficient for a generating set
m1 , . . . , mr to be a basis. Hint: Consider the difference of two equal linear
combinations of the mi ’s. M
80 ma4200—commutative algebra

The gist of this example is that x and y commute, and this indicates that the
phenomenon is inherent in commutative rings. Any set of generators for an
ideal consisting of at least two elements can never be a basis simply because
the generators commute. K

Free modules

4.36 The lack of bases for most modules, leads to a special status of those
that have one. One says that an A-module F is free if it has a basis. The reason Free modules (Fri
behind the suggestive name “free” is that one may freely prescribe the values moduler)

a linear map takes on the basis elements—a principle that goes under the
name of the Universal Mapping Principle:

Proposition 4.37 (The Universal Mapping Principle) Suppose that F is

a free A-module with a basis t f i uiP I , and let M be another A-module. For any subset
tmi uiP I of M indexed by I, there is a unique A-linear map φ : F Ñ M such that
φp f i q “ mi .
ř
Proof: Every element x P F is expressible as x “ iP I ai f i with the coefficients
ai ’s from A, merely finitely many of which are non-zero, and most importantly,
ř
they are uniquely determined by x. Hence sending x to iP I ai mi gives a well
defined map φ : F Ñ M. We leave the task of checking it is A-linear to the
zealous student. (This amounts to checking that the coefficients of a linear
combination is the corresponding linear combination of the coefficients which
ensues from the coefficients being uniquely determined.) o
4.38 We retrun to Example 4.16 about when ideals are free, and give a precise
statement:

Proposition 4.39 An ideal a in the ring A is a free A-module if and only if it is

principal and generated by a non-zero divisor.

Proof: We saw in Example 4.16 above that when a requires at least two
generators, it has no basis and therefore is not free. Nor can principal ideals
generated by a zero divisor be free since if a f “ 0 with a ‰ 0, the relation
a f “ 0 f gives two representations of 0. The other way around, if the non-zero
divisor f is a generator for a, it is basis; indeed f being a non-zero divisor it
can be cancelled from an equality like a f “ b f . o
4.40 Archetypes of free modules are the direct sums nA “ A ‘ . . . ‘ A of n
copies of the ring A which we already met in Example 4.2 on page 4.2. They
come equipped with the so-called standard basis familiar from courses in linear
algebra. The basis elements ei are given as ei “ p0, . . . , 0, 1, 0, . . . 0q with the one
sitting in slot number i.
There is no reason to confine these considerations to direct sums of finitely
À
many copies of A. For any set I, the direct sum iP I A has a standard basis
 modules 81

tei uiP I and is a free module; the basis element ei is the string with a one in slot
i and zeros everywhere else.

Proposition 4.41 Assume that F is a free A-module with basis t f i uiP I . Then there
À
is an isomorphism between F and the direct sum iP I A that sends each basis vector
f i to the standard basis vector ei .
À
Proof: By Proposition 4.37 above, we may define a map φ : F Ñ iP I A
À
by sending f i to the standard basis vector ei ; conversely, since iP I A is free,
À
sending ei to f i sets up a map ψ : iP I A Ñ F. These two maps are obviously
mutual inverses. o

Corollary 4.42 Any two bases of a free module have the same cardinality. Two
free modules are isomorphic if and only if they possess bases of the same cardinality.

The common cardinality of the bases for a free module is called the rank of the Rank of a free module
module, and the rank is the sole invariant of free modules; up to isomorphism (Rangen til en fri modul)

it determines the module. When the module is a vector space over a field and
the rank is finite, the rank is just the dimension of the vector space.
Proof: After Proposition 4.41 above we need merely to verify that when two
À À
direct sums iP I A and jP J A are isomorphic as A-module, the index sets I
and J are of the same cardinality. This is well known from the theory of vector
spaces when A is a field, so take any maximal ideal in A and consider the
À À
isomorphic vector spaces iP I A{m and jP J A{m over A{m (isomorphic in
view of exercise 4.10 on page 75). One has a basis of the same cardinality as I,
the other one of cardinality that of J; hence I and J are equipotent. o
Example 4.17 (Free modules with given basis) From time to time it will be
convenient to operate with free A-modules with a given set S as basis. There
is no constraint on the set S, it can be whatever one finds useful. The formal
way to construct AS is as the set of maps α : S Ñ A of finite support; that is the
maps such that αpsq ‰ 0 for at most finitely many members s of S; in symbols

AS “ t α : S Ñ A | α of finite support u.

The module structure of AS is given point-wise: pα ` α1 qpsq “ αpsq ` α1 psq and

pa ¨ αqpsq “ aαpsq.
There is a collection of function termed generalized Kronecker-δ’s that
constitute a natural basis for AS . These are in a one-to-one correspondence
with the set S: There is one such function δs for each member s P S and it is
defined as $
&0 when s ‰ t
δs ptq “
%1 when t “ s .

It is a trivial matter to verify they form a basis: They generate AS because

ř ř
any α can be expressed as α “ sPS αpsqδs , and if as δs “ 0 is a dependence
relation, one just plugs in any t from S to find that at “ 0.
82 ma4200—commutative algebra

A suggestive way of denoting elements from the module AS is as linear

ř
combinations s as ¨ s of elements from S, which merely amounts to writing s
for the function δs .
The module AS depends functorially on S. Indeed, given any map φ : S Ñ
S1 . Because the δs ’s form a basis for AS , we obtain according to the Universal
1
Mapping Principle for free modules, a map φ˚ : AS Ñ AS by sending each
1
basis element δs to the element δφpsq of AS . In the alternative notation, the
ř ř
map φ˚ takes the form φ˚ p s as ¨ sq “ s as ¨ φpsq, and it is pretty obvious that
pψ ˝ φq˚ “ ψ˚ ˝ φ˚ when ψ is another map which is composable with φ. So, AS
is a covariant functor from the category Sets of sets to the category Mod A of
A-modules. K FM
φ˚
/ FN

Problem 4.22 Let M be an A-module and let FM “ A M be the free module θM θN

 
with elements from M as a basis. Then M is a quotient of FM in a canonical M /N
φ
way: Define a map θ M : FM Ñ M by sending δm to m (in the alternative nota-
tion it takes the hypertautological form m ÞÑ m). Show that θ M is surjective
and that φ ˝ θ M “ θ N ˝ φ˚ whenever φ : M Ñ N is an A-linear map. M

Matrices and maps of free modules

Just like linear maps between two vector spaces of finite dimension, A-linear
maps between two free A-modules can be described by matrices, and the
mechanism works exactly in the same way. Be aware however, that the task
of describing maps between modules that are not free, is substantially more
complicated, if a good description at all is accessible.
4.43 The representation of a map by a matrix depends on the choice of bases
for each module. So let E and F two finitely generated and free A-module and
let te j u and t f i u be the two bases. When each φpe j q is expressed in terms of
the basis t f i u, the coefficients in the expressions make up the matrix. If E is of
rank n and F of rank M the matrix is the m ˆ n-matrix given7 as Mpφq “ paij qij 7
The notation MF E pφq
ř
where φpe j q “ i aij f i . i cumbersome, but
sometimes helpful.
4.44 The familiar property that compositions of maps correspond to products Then the rule for
composition takes the
of matrices still holds true, and the verification is mutatis mutandis the same as E pφq ¨ M F pψq “
form MF G
for linear maps between vector spaces (and we leave it to students needing to E
MG pφ ¨ ψq
fresh up their knowledge of linear algebra); that is, if ψ is A-linear map from F
to a third free module G, one has

Mpψ ˝ φq “ Mpψq ¨ Mpφq.

Likewise, associating a matrix to a map persists being a linear operation in

that
Mpαφ ` βφ1 q “ αMpφq ` βMpφ1 q,
whenever φ1 : E Ñ F is A-linear and α and β are two elements from the ring
A.
 modules 83

4.45 The all important construction of determinants of linear maps generalizes

word for word to A-linear maps bewteen free modules, and all the basic
properis of determinants continue to hold; like multiplicativity, alteration in
rows and columns, expantions along rows or columns.
In particular, we would like to point out the formula

A ¨ A: “ det A ¨ I (4.2)

valid for a square matrix A, where A: is the so-called cofactor-matrix of A

and I is the identity matrix. The ij-th entry of A: is the sub-determinant of A The cofactor-matrix
(Kofaktormatrisen)
with the j-th row and i-column struck out adjusted with the sign p´1qi` j . The
formula (4.2) follows from the rules for expanding a determinant along a row.
Contrary to the case of vector spaces, it does not suffice that det A be non-zero
for A to be invertible, the determinant must be invertible in A, but in that case i

it ensues from 4.2 that the inverse is given as

j a ji
A´1 “ pdet Aq´1 ¨ A: .

4.46 Observer that an endomorphism φ of a free module E has a canonically

defined determinant. Indeed, let M be a matrix for φ in some basis for E and
M1 in another. If N is the base-change matrix, it holds true that M1 “ N MN ´1
and consequently

det M1 “ det N ¨ det M ¨ det N ´1 “ det M.

This makes det φ “ det M a legitimate definition as the determinant det M of φ

does not depend on which basis is used.
This allows the definition of the characterstic polynomial of an endomorphism The characterstic polyno-
φ of a free module E of finite rank, namely as Pφ ptq “ detpt ¨ idE ´φq. It is an mial of an endomorphism
(Det karakteristiske poly-
element of the polynomial ring Arts. nomet til en endomorfi)
Ours is a pedestrian approach to determinants—there is also a Formula 1
way based on so-called exterior powers. It has the disadvantage of requiring
more advanced machinery and being rather opaque for beginners, but has the
great advantage of being completely functorial. It also open up for defining
determinants of endomorphisms of a wider class of modules than the free
ones.
Example 4.18 (The norm) An indispensable tool in algebtaic number theory is
the norm. We have already come across it at occations, e.g. when showing that
?
2 and 3 were irreducible eements in Zri 5s, butin the guise of the ordinary
absolute value of compelx numbers.
The setting is as follows. We are given an algebra B over a ring A which
is free of finite rank n as A-module. Every element b in B defines an A-linear
"multiplication-by-b" map8 rbs : B Ñ B that sends x to b ¨ x. Its determinant 8
Not to be confused
det rbs is called the norm of b and denoted NB{ A pbq. Clearly rbb1 s “ rbs ˝ rb1 s with the residue class
notation.
The norm of an element
(Normen til et element)
84 ma4200—commutative algebra

and the determinant of a composition being equal to the product of the deter-
minants, we infer that NB{ A pbb1 q “ NB{ A pbqNB{ A pb1 q. The norm is therefore a
multiplicative functions on B with values in A. Moreover, for elements a P A
one has NB{ A paq “ an . This follows because the map ras is just a ¨ idB and
detpa ¨ idB q “ an (the map ras is represented as a scalar matrix in any basis). K

Problems

4.23 Prove that any finitely generated A-module is the quotient of a free ˇ
module of finite rank.

4.24 Show that any A-module is the quotient of a free module. Hint: let ˇ
À
FM “ mP M A be the direct sum of copies of A, one for each element m P M.
? ?
4.25 Let A “ Zr ds. Show that A is a free Z-module with basis 1, d. Use ˇ
?
this to show that NA{Z px ` y dq “ x2 ´ dy2 .
?
4.26 Assume that d is an innteger such d ” 1 mod 4. Let α “ p1 ` dq{2. ˇ
Show that α2 “ α ` pd ´ 1q{4. Prove that A “ Zrαs is free Z-module of rank 2
with 1, α as a basis. Determine the matrix of the map x ÞÑ αx in this basis and
compute the characteristic polynomial.

4.27 Let A “ krX, Ys{pY 2 ´ XpX ´ aqpX ´ bqq be the coordinate ring of an ˇ
elliptic curve and as usual let x be the class of X and y that of Y.
a) Show that krxs is isomorphic with the polynomial ring krXs and that A is a
free krxs-module of rank 2.
b) Show that krys is isomorphic to the polynomial ring krYs and that A is a
free krys module of rank 3.
c) Show that x and y are irreducible elements in A and conclude that A is not
a udf. M
The next series of exercises, which culiminate with problem 4.30, are aimed at
giving an example that countable products of free modules are not neceras-
saryli free
4.28 Show that in a free Z-module every element is divisible by at most
finitely many integers.

4.29 Show that the direct sum of countably many copies of Z is countable,
whereas the direct product of countably many copies is not (it has the cardinal-
ity of the continuum).

4.30 (Infinite products are not always free.) The task is to show that the direct
product P “ iPN Z of a countable number of copies of Z is not a free Z-
ś

module. Aiming for a contradiction, suppose that the product has a basis
iPN Z lies in P and has the standard basis
À
t f i uiP I . The direct sum Q “
 modules 85

elements ei (with a one in slot i as sole non-zero component). Each ei can be

developed as ei “ j aij f j with aij P Z in terms of the basis elements f j .
ř

a) Prove that I cannot be countable.

b) Prove that there is a countable subset J Ď I so that the module Q generated
by the f j ’s with j P J contains the direct sum iPN Z. Hint: Let j be in J
À

when the coefficient aij ‰ 0 for at least one i.

c) For any element x “ pn1 , n2 , . . .q in P and any i P N prove that the element
y “ p0, 0, . . . , 0, ni , ni`1 , . . .q has the same image in P{Q as x.
d) Show that there are strictly increasing sequences tnk u of natural numbers
with nk |nk`1 and so that a “ pn1 , n2 , . . .q does not lie in Q. Hint: Q is count-
able.
e) Show that the image of a in P{Q is divisible by infinitely many numbers
and hence P{Q cannot be free.

4.5 Exact sequences

Let φ and φ be two composable A-linear maps and write them they down in a
sequence
/N / L.
φ ψ
M

This sequence is said two be exact if ker ψ “ im ψ. It frequently happens that Exact sequences (Eksakte
such a sequence is part of a longer sequence of maps, extending to the left or følger)

to the right, and then the extended sequence is said to be exact at N as well. A
sequence exact at all places, is simply said to be exact.
4.47 Two special cases warrant mentioning; the first being when M “ 0:

/N / L.
ψ
0

The image of the zero map being the zero submodule p0q, exactness boils
down in this case to ψ being injective. Similarily, when L “ 0, the sequence is
shaped like
/N / 0,
φ ψ
M

and it being exact is equivalent to φ being surjective.

Example 4.19 Any map α : M1 Ñ M lives in the exact sequnece

0 / ker α / M1 /M / coker α / 0.

K
86 ma4200—commutative algebra

Short exact sequences

By far the most often met exact sequences, are the so-called short exact se-
quences; and they are a valuable tool when one tries to study a module by
breaking it down into smaller (and presumptive simpler) pieces.
4.48 A three-term sequence (or a five-term sequence if you count the zeros)

/ M1 /M / M2 /0
α β
0 (4.3)

is called a short exact sequence when it is exact. This means that α is injective,
that β surjective and that im α “ ker β. Of course, the term “short” in the name Short exact sequence
(Korteksakte følger)
implies there are long exact sequence as well, and indeed there are, as we shall
see later on.
It ensues from the First Isomorphism Theorem (Corollary 4.12 on
page 70) that there is a unique isomorphism θ : M2 » M{αpM1 q shaped in a
way that β corresponds to the quotient map. In other words, θ enters in the
following commutative diagram

/ M1 /M / M2 /0
α β
0 (4.4)
α| M1 θ
 
0 / αpM1 q /M / M{αpM1 q /0

where the maps in the bottom row are respectively the inclusion of im M1
into M and the quotient map. In short, up to isomorphisms all short exact
sequence appear as

0 /N /M / M{N / 0,

where N Ď M is a submodule, and the two maps are respectively the inclusion
and the quotient map.
Example 4.20 (Direct sums) The direct sum M ‘ N of two A-modules fits
naturally into the short exact sequence

0 /N / N‘M /M /0

where the left-hand map is the natural inclusion sending x to px, 0q and the
one to the right is the projection onto M, which maps px, yq to y. If N and N 1
are two submodules of the A-module M, then there is a short exact sequence

0 / N X N1 ι / N ‘ N1 σ / N ` N1 /0

where ι is the inclusion map and σpx, yq “ x ´ y. K

Example 4.21 (A Chinese squence) The Chinese remainder theorem (Theo-

 modules 87

rem 2.71 on page 48) for two ideals may be generalized by saying that the
sequence

0 / aXb /A / A{a ‘ A{bb / A{a ` b / 0, (4.5)

where the two maps in the middle are given by the following two assignments
x ÞÑ prxsa , rxsb q and prxsa , rysb q ÞÑ rxsa`b ´ rysa`b . Having four non-zero terms
it is to long to be called short exact, but it may be obtained by splicing the two
short exact sequences

0 / aXb /A / A{a X b /0

and
0 / A{a X b / A{a ‘ A{b / A{a ` b /0

Problem 4.31 Verify that the two maps defined in the example above are
well defined and that the sequence is exact. Deduce the Chinese Remainder
Theorem from it. M

Problem 4.32 Write down a “Chinese sequence” involving three ideals that
generalizes the sequence (4.5) above. Prove it is exact and deduce the Chinese
Remainder Theorem for three ideals from it. Hint: It will have six non-zero
terms. M

Split exact sequences

Some short exact sequence stand out from all the crowd, to wit, the so-called
split exact sequences. A short exact sequence

/ M1 /M / M2 / 0.
α β
0 (4.6)

is split exact when being isomorphic to the standard sequence of Example 4.20
on page 86. This not only requires that M be isomorphic to the direct sum Split exact sequences
(Splitteksakte følger)
M1 ‘ M2 , but the somehow stronger requirement that there be an isomorphism
inducing the identity on M1 and pairing β with the projection, must be met:
that is, the isomorphism must fit into the following commutative diagram

/ M1 /M / M2 /0
α β
0
 »
0 / M1 / M1 ‘ M2 / M2 / 0.

where the maps in the bottom sequence are the projection and the inclusion.
4.49 Of course all sequences are not split exact, and even if the two extreme
modules are the same the middle modules need not be isomorphic. The
88 ma4200—commutative algebra

easiest example is found among abelian group: Both Z{p2 Z and Z{pZ ‘ Z{pZ
appear in the midst of short exact sequences with both extreme modules being
Z{pZ. In general, it is an unsurmountable challenge to classify all possible
middle modules given the two extreme ones.
4.50 There is nice criterion for a short exact sequence to be split involving
only one of the maps α or β; to formulate it we need two new concepts. A
right section for β is an A-linear map σ : N Ñ M such that β ˝ σ “ id N , and a left Right sections (Høyresek-
section for α is one τ : M Ñ M1 with the property that τ ˝ α “ id M1 . sjoner)
Left sections (Venstresek-
Proposition 4.51 (Splitting criterion) Let the short exact sequence sjoner)

/ M1 /M / M2 /0
α β
0

of A-modules be given. Then the following three statements are equivalent

o The sequence is split;

o The map α has a left section;

o The map β has a right section.

Before starting with the proof we observe that a map β : M Ñ M2 that pos-
sesses a right section σ is automatically surjective since if x P M2 it holds true
that βpσpxqq “ x. Thus, once the map β has a right section its source M will be
isomorphic to M2 ‘ ker β. Moreover, the restriction of β to the image σpM2 q
is easily seen to be an isomorphism so that M is the direct sum of the two
submodules ker β and σpM2 q.
Proof: We start out by assuming that the sequence is split and that θ : M Ñ
M1 ‘ M2 is an isomorphism of the right type. Then one easily verifies that
the maps given by m2 ÞÑ θ ´1 p0, m2 q and m1 ÞÑ θ ´1 pm1 , 0q are sections for
respectively β and α of the right variance and hence the first assertion implies
each of the two others.
Assume then that the map β has a right section σ and define a map Φ : M1 ‘
M2 Ñ M simply by the assignment px, yq ÞÑ αpxq ` σpyq. This map is clearly
A-linear and it lives in the diagram

/ M1 /M / M2 /0
α β
0 O
Φ

0 / M1 / M1 ‘ M2 / M2 /0
ι π

whose rows are exact, and where ι and π are respectively the canonical inclu-
sion and projection. The left square commutes since Φpιpxqq “ αpxq ` σp0q “
αpxq, and that the right does, ensues from the following little calulation

βpΦpx, yqq “ βpαpxq ` σpyqq “ βpσpyqq “ y “ πpx, yq.

 modules 89

where we use that σ is a right section for β. Now, there is a general fact (called
The Five Lemma, see Exercise ?? on page ?? below) that a map in the midst
of two isomorphisms as in the diagram above is an isomorphism; but it is
also easily checked ad hoc: If 0 “ Φpx, yq “ αpxq ` σpyq, it follows that y “ 0
since the right square above is commutative, hence αpxq “ 0, and consequently
x “ 0 because α is injective. To prove surjectivity of Φ, observe that for z P M,
the difference z ´ σpβpzqq is killed by β, and because ker β “ im α, one has
z ´ σpβpzqq “ αpyq for some y so that z “ αpyq ` σpφpzqq.
Finally, let us treat the case that α has a left section. The map Ψ : M Ñ
M1 ‘ M2 defined by Ψpzq “ pτpzq, βpzqq fits in the following diagram analogue
to the one above

/ M1 /M / M2 /0
α β
0
Ψ

0 / M1 / M1 ‘ M2 / M2 /0
ι π

The right square commutes trivially, and the left coomutes since one has

Ψpαpzqq “ pτpαpzqq, βpαpzqqq “ pz, 0q “ ιpzq.

From there on one proceedes in an ad hoc manner as above, or resort to the

general principle of the five lemma. o
When σ is a right section for β, the isomorphism Φ gives a decomposition
M “ αpM1 q ‘ σpM2 q “ ker β ‘ σpMq into a direct sum of two submodules
(hence the equalities are genuine), and in a similar fashion, when α has a left
section τ, the module M decomposes as M “ αpM1 q ‘ ker τ, also a sum of two
submodules.

Problem 4.33 In this exercise the setting is as in the proposition, and the
point is to connect up with the splitting criterion formulated in terms of idem-
potents we gave when discussing direct sums and products (Proposition 4.26
on page 75).
a) Assume that σ is a right section for β. Prove that σ ˝ β is an idempotent
endomorphism of M, and that the corresponding decomposition in a direct
sum is the one in remark abovef; that is M “ ker β ‘ σpM2 q.
b) Assume that τ is a left section for α. Prove that α ˝ τ is idempotent, and that
the corresponding decomposition is M “ αpM2 q ‘ ker τ. M

Problem 4.34 (Aditive functors.) A functor F between two module categories9

Mod A Ñ ModB is said to be additive if it takes sums of maps to sums of maps;
9
To make things
simple, we contend
that is Fpφ ` ψq “ Fpφq ` Fpψq whenever φ and ψ are A-linear maps between ourselves to module
two A-modules. The functors Hom A p´, Nq and Hom A pN, ´q are examples of categories. There is
however a notion
such animals. of additive categories
(where hom-sets are
abelian groups and
compositions bilinear),
and between such
categories the term
additive functor is
meaningful.
Additive functors
(Additive funktorer)
90 ma4200—commutative algebra

a) Let F be any covariant functor F : Mod A Ñ ModB and let β : M Ñ N be

an A-linear map having a right section σ. Prove that Fpσq is a right section of
Fpβq.
b) Assume additionally that F is additive. Show that if the sequence

/ M1 /M / M2 /0
α β
0

is split exact, then the sequence

F pαq F p βq
0 / FpM1 q / FpMq / FpM2 q /0

is split exact as well. In particular, the functor F transforms finite direct sums
into finite direct sums. Hint: If σ and τ are sections of respectively β and α,
then Fpσq and Fpτq will be sections of respectively Fpβq and Fpαq. M

Left exactness of hom-functors

4.52 Suppose given a short exact sequence like (4.3) above and let N be an
A-module. Applying the covariant hom-functor Hom A pN, ´q to the sequence,
we obtain an induced sequence which is shaped like

/ Hom A pN, M1 q / Hom A pN, Mq / Hom A pN, M2 q .

α˚ β˚
0 (4.7)

The maps are simply given by composition; i.e. α˚ φ “ α ˝ φ and β ˚ φ “ β ˝ φ,

and since β ˚ ˝ α˚ “ pβ ˝ αq˚ and β ˝ α “ 0, it holds true that β ˚ ˝ α˚ “ 0. More
is true, the induced sequence will in fact be exact. / M1 α
/M / M2
0 O @
To understand why this is true, there are two spots where exactness is to
φ
0
be checked; the first being that α˚ is injective. But α˚ pφq “ φ ˝ α, and since α
N
is assumed to be injective, α˚ is injective as well; indeed α ˝ φ “ 0 means that
the image im φ lies in the kernel of α. Secondly, to verify the sequence (4.7) is
/ M1 α
/M / M2
exact at the middle spot, assume that β ˝ φ “ 0 for a map φ : N Ñ M. Then φ 0 ^ O @
factors through the image αpM1 q, and α being injective, φ can be represented as φ
0
α ˝ φ1 for a map φ1 : N Ñ M1 which is precisely what we desire.
N
4.53 In a similar vein, the contravariant version Hom A p´, Nq applied to (4.3)
yields the sequence where the arrows are reversed

β˚ α˚
0 / Hom A pM2 , Nq / Hom A pM, Nq / Hom A pM1 , Nq , (4.8)

and an argument like above shows that this also is an exact sequence.
Problem 4.35 Give the argument referred to in the previous paragraph in
detail. M
4.54 It is common usage to refer to the phenomena described above as saying
that Hom A pN, ´q and Hom A p´, Nq are left exact functors. There are crowds of Left exact functors
(Venstre-eksakte funk-
torer)
 modules 91

examples that β ˚ and α˚ are not surjective, so Hom A pN, ´q and Hom A p´, Nq
are seldom exact functors in the sense that they take short exact sequences to Exact functors (Eksakte
short exact sequences. funktorer)
A large part of homological algebra was developed just to describe the
"missing cokernels" coker β ˚ and coker α˚ . In general the only answer to this
challenge is that the two sequences can be extended ad infinitum to the right to
yield long exact sequences that involve so-called Ext-modules. These modules
do not depend on the original short exact sequence only on the modules
involved and N of course (but the maps in the long exact sequence do), and in
some cases these long exact sequences can be controlled.
Two classes of modules stand out namely the ones with the property that
either the functor Hom A pN, ´q or the functor Hom A p´, Nq exact. The former
are called projective modules (they are ubiquitous in commutative algebra, and Projective modules
we shall come back to them) and the latter are the so-called injective modules. (Projektive moduler)
Injective modules
4.55 In paragraph 4.52 above we proved the "only-if-part" of the following (Injektive moduler)
proposition (although we worked with short exact sequences like in (4.3) we
never used that β was surjective).
Proposition 4.56 (Left exactness I) Let the sequence

/ M1 /M / M2
α β
0 (4.9)

be given and assume that β ˝ α “ 0. The sequence is exact if and only if for all
A-modules N the sequence

0 / Hom A pN, M1 q / Hom A pN, Mq / Hom A pN, M2 q (4.10)

is exact.

Proof: To attack the "if-part" assume that (4.10) is exact for all A-modules N.
If α is not injective, take N “ ker α, which is non-zero, and let ι be the
inclusion of ker α in M1 . Then α ˝ ι “ 0, but ι is non-zero so α˚ is not injective.
In a similar vein, if the image im α is strictly smaller than the kernel ker β,
take N “ ker β and consider the inclusion map ι of N in M. By choice it holds
that β ˝ ι “ 0, but ι cannot factor though α since im α is strictly contained in im ι.
o
4.57 As alluded to in Paragraph 4.54 above, even if the map β is surjective, the
induced map β ˚ will in most cases not be surjective. An easy example is given
below. That β ˚ surjective means that any A-linear map φ : N Ñ M2 can be
lifted to an A-linear map into M, as illustrated in the following diagram.

/ M1 /M / M2 /0
α β
0 a O
φ

N
92 ma4200—commutative algebra

Proposition 4.58 If F is a free A-module, the functor Hom A pF, ´q is exact; in

other words, free modules are projective.

Proof: Let t f i u be a basis for F. Since β is surjective, there are elements

tmi uiP I such that βpmi q “ φp f i q. By the Universal Mapping Property of free
modules (Proposition 4.37 on page 80), there is a map Φ : F Ñ M with Φp f i q “
mi . Then β ˝ Φ “ φ since βpΦp f i qq “ βpmi q “ φp f i q, and two maps agreeing on a
basis are equal. o

Proposition 4.59 An A-module is projective if and only if it is a direct summand

in a free module.

Proof: Let P be the module. And assume first that P is projective. Let tpi uiP I
be a generating set for P (finite or not, we do not care about the size. One
could e.g. use the canonical free module FP from Example 4.17 on page 81)
and consider the exact sequence

/K /À /P / 0,
φ
0 iP I A
ř
where a string pai qiP I is sent to i ai pi (the string lives in the direct sum so
merely finitely many of the ai ’s are non-zero). Since the pi generate P, this
map is surjective. Now, since P is projective, the identity map idP : P Ñ P
À
can be lifted to a σ : P Ñ iP I A. This means that φ ˝ σ “ idP , and the lifting P
 ι
/ P ‘ P1
À
σ is a section of φ . Hence P lies split in iP I A by the Splitting Criterion π
(Proposition 4.51 on page 88). ψ

ψ˝ι P
To prove the other implication, let φ : P Ñ M2 be given assume that P ‘ P1
is free. Consider the map P ‘ P1 Ñ M2 sending px, yq to φpxq. This map can be φ
  
lifted as P ‘ P1 is free, and restricting the lifted map to P yields a lifting of φ. o M / M2 /0
β
4.60 There is also an assertion dual to the one of Proposition 4.56 above. The Diagram illustrating the last
part of the proof; π is the
proof of the two being quit similar, we leave all the checking to the zealous projection and ι the inclusion.
and not so zealous students; it is a good training for these diagram-arguments. Notice that π ˝ ι “ idP .
The assertion reads as follows:

Proposition 4.61 (Left exactness II) Let the sequence

/M / M2 /0
α β
M1 (4.11)

be given and assume that β ˝ α “ 0. The sequence is exact if and only if for all
A-modules N the sequence

0 / Hom A pM1 , Nq / Hom A pM, Nq / Hom A pM2 , Nq (4.12)

is exact.

4.62 In most cases the map α˚ will not be surjective even if α is. Surjectivity
means that any map φ : M Ñ N can be extended to a map M Ñ N, like in the
 modules 93

diagram below
/ M1 /M / M1 /0
α β
0
φ
 }
N

Examples

4.22 Consider the short exact sequence

p
0 /Z /Z / Z{pZ /0 (4.13)

where p is a prime number. No non-zero map from Z{pZ to Z exists (all

elements in Z{pZ are killed by p, but none in Z apart from 0), so the iden-
tity idZ{ pZ can not be lifted to a map Z{pZ Ñ Z. Since it holds true that
HomZ pZ{pZ, Zq “ 0 and HomZ pZ{pZ, Z{pZq “ Z{pZ, applying HomZ pZ{pZ, ´q
yields the sequence

0 /0 /0 / Z{pZ,

and of course a map 0 Ñ Z{pZ cannot be surjective!

4.23 We observe that HomZ pZ, Z{pZq “ HomZ pZ{pZ, Z{pZq “ Z{pZ, so the
functor HomZ p´, Z{pZq when applied to the sequence (4.13) in example 4.22
above, yields the exact sequence

p˚
0 / Z{pZ / Z{pZ / Z{pZ.

The map p˚ is just multiplication by p which on Z{pZ is the zero map; hence
the rightmost arrow is zero. The left arrow is therefore an isomorphism; this
ensues from exactness of the sequence, but neither is it hard to verify ad hoc
that it equals the identity map.

Example 4.24 (A projective module that is not free) The prime ideals one meets
in number theory which are not principal are in most cases projective, and
later on we shall prove this for ideals in the so-called Dedekind rings. For
the moment we content ourself with giving just one example illustrating this
phenomenon.
? ?
The example will be the ideal a “ p2, 1 ` i 5q in the ring A “ Zri 5s, which
turns out to be a projective but not a free A-module. Indeed, we shall with an
explicit construction see that a is a direct summand in the free module A ‘ A
and hence it will be projective (Proposition 4.59 on page 92), and since it is not
principal, it is not free.
94 ma4200—commutative algebra

Showing that a is a summand in A ‘ A amounts to finding an isomorphism

A ‘ A » a ‘ M, for some module M. In fact, one may show that the comple-
ment M is forced to be isomorphic with a, so that A ‘ A » a ‘ a, but that will
be for some other time. This also provides an example of two direct sums that
are isomorphic without the summands being isomorphic.
The free module A ‘ A has the usual basis e1 “ p1, 0q and e2 “ p0, 1q The gist
of the construction is the map

φ: A ‘ A Ñ aĎ A
?
defined by the assignments e1 ÞÑ 2 and e2 ÞÑ 1 ` i 5. We shall identify a
submodule M inside A ‘ A that φ maps isomorphically onto a and thereby
proving that a lies split in A ‘ A. The submodule M in question is generated by
? ?
the two elements a1 “ ´2e1 ` p1 ´ i 5qe2 and a2 “ ´p1 ` i 5qet1 ` 3e2
We begin with checking that the restriction φ| M is surjective. This ensues
from the very definition of φ as the following little calculations show:
? ?
φpa1 q “ ´2 ¨ 2 ` p1 ´ i 5q ¨ p1 ` i 5q “ ´4 ` 6 “ 2,
? ? ?
φpa2 q “ ´p1 ` i 5q ¨ 2 ` 3 ¨ p1 ` i 5q “ p1 ` i 5q.

To prove that φ is injective on M, pick an element a “ xa1 ` ya2 in M that maps

?
to zero. This means that 2x ` p1 ` i 5qy “ 0 in A. Now, one has
` ? ˘ ? ` ? ˘
xa1 ` ya2 “ ´ 2x ` p1 ` i 5qy e1 ` pp1 ´ i 5qx ` 3yqe2 “ p1 ´ i 5qx ` 3y e2 ,

and since
` ? ˘ ` ? ? ˘
2 p1 ´ i 5qx ` 3y “ ´ p1 ` i 5qp1 ´ i 5q ` 6 y “ 0,
?
it follows that p1 ´ i 5qx ` 3y “ 0 as well. Hence a “ 0, and we are through. K

Problems

4.36 Let A be a ring and suppose that B is an A-algebra. Assume given an ˇ

exact sequence
0 /E /F /G /0

of B-modules that regarded as A-modules are free of finite rank. Prove that
rk A F “ rkk E ` rk A G.
In the following exercise we let Pφ ptq “ detpt ¨ idE ´φq be the characteristic
polynomial of an endomorphism φ : E Ñ E of a free A-module of finite rank.
4.37 (Multiplicativity of the characteristic polynomial.) Assume given a commuta- ˇ
tive diagram
0 /E /F /G /0

ψ φ θ
  
0 /E /F /G /0
 ψ
modules
‹‹
95

where the rows are exact and the involved modules are free of finite rank. 0 θ

Prove that det φ “ det ψ ¨ det θ. Show that Pφ ptq “ Pψ ptq ¨ Pθ ptq. Hint: Exhibit a
basis for F in which the matrix of φ has an appropriate block decomposition
(as in the margin). Conclude that det φ “ det θ ¨ det ψ and that tr φ “ tr θ ` tr ψ.
?
4.38 With reference to Example 4.24 above, let A “ Zri 5s and a “ p2, 1 ` i5q.
Let Ψ : 2A Ñ 2A be the A-linear map given by the matrix
˜ ? ¸
´3 1`i 5
Ψ“ ?
1´i 5 ´2

a) Show that there is an exact sequence

Ψ / 2A
φ
/a / 0,
2A

where φ is the map from Example 4.24, and hence that coker Ψ » a.
b) Show that Hom A pa, Aq » a.
c) Let Φ : 2A Ñ 2A be the map given the matrix
˜ ? ¸
2 1`i 5
Φ“ ?
1´i 5 3

Show that Φ ˝ Ψ “ Ψ ˝ Φ “ 0, and that there is an exact sequence

Φ / 2A Ψ / 2A Φ / 2A Ψ / 2A Φ / ...
...

that extends infinitely in both directions.

d) Show that coker Φ » a, and conclude that 2A » 2a.

4.6 Snakes and alike

4.63 An all important feature in homological algebra are the so-called con-
necting maps which relate homology groups of complexes in various ways.
A simple but very useful instance of this feature is described in the Snake
Lemma. The name is of “bourbakistic origin”, and mnemotechnical efficient.
See the next section below for a reason for the name.
96 ma4200—commutative algebra

Lemma 4.64 (The snake lemma) Given a diagram

/ M2 / M3 /0
φ1 φ2
M1 (4.14)
α1 α2 α3
  
0 / N1 / N2 / N3
ψ1 ψ2

where the rows are exact and the squares are commutative. There exists a map
δ : ker α3 Ñ coker α1 rendering the following sequence exact

ker α2 / ker α3 δ / coker α1 / coker α2 , (4.15)

where the two extremal maps from left to right are induced by and φ2 and ψ1 .
Proof: The proof of the Snake Lemma is an example of a sport called diagram
chasing which was extensively practised among homological algebraists. We
have two missions to complete, firstly the map δ must be constructed and
secondly, we must verify that the sequence (4.15) is exact.
We begin with the first and most interesting one: A short and dirty mnemotech-
nical definition of δ is ψ1´1 ˝ α2 ˝ φ2´1 which of course is meaningless as it stands
since neither φ2 nor ψ1 is invertible, but it gives a hint of how to construct δ.
Now the fun is starting: Pick an element x P M3 so that α3 pxq “ 0 and lift
it to an element y in M2 ; that is, pick an element y P M2 with φ2 pyq “ x. The
rightmost square of (4.14) being commutative, we infer that ψ2 pα2 pyqq “ 0; the
bottom line of 4.14 being exact, there is thence a z in N1 with ψ1 pzq “ α2 pyq.
And that is it; the image of z in coker α1 is the wanted guy δpxq.  /x
_
y _
We have done a choice on the way—the choice of a lift of x to M2 —and for
the definition of δ to be legitimate the trapped z must be independent of that   
/ α2 pyq  /0
choice. So assume that y1 is another element of M3 that maps to x; then we _z
may write y1 “ y ` w with φ2 pwq “ 0. Since the line on top of 4.14 is exact it
holds that w “ φ1 puq for some u, and we find 
δpxq
α2 py1 q “ α2 pyq ` α2 pφ1 pwqq “ α2 pyq ` ψ1 pα1 puqq
Luckily, ψ1 is injective (the bottom line of 4.14 is exact), so if z1 P N1 is such
that ψ1 pz1 q “ y1 one has
z1 “ z ` α1 puq
and finally, this means the images of z and z1 in coker α1 agree, and δ is well
defined!
The big game has been snared, and it remains only to check exactness
of 4.15: We shall do half of the job and check exactness at ker α3 , letting the
zealous students have the fun of checking the other half. So assume that
δpxq “ 0. This means that z “ α1 pvq for some v P M1 ; hence α2 pyq “ ψ1 pzq “
ψ1 pα1 pvqq “ α2 pφ1 pvqq. It follows that y “ φ1 pvq ` t with t P ker α2 and
consequently x “ φ2 pyq “ φ2 ptq, which is what we need. o
 modules 97

Why snake?
The reason for the name “Snake Lemma” is apparent when one considers the
diagram below.
There the map δ connecting ker α3 to coker α1 we constructed in the Snake
Lemma, zig-zags like a green snake through the diagram.

ker α1 ker α2 ker α3

M1 M2 M3 0
α1 α2 α3

0 N1 N2 N3

coker α1 coker α2 coker α3

In applications of the Snake Lemma one frequently happens that the maps
φ1 : M1 Ñ M2 is injective and one ψ2 : N2 Ñ N3 is surjective. The diagram we
start with thus is shaped like

/ M1 / M2 / M3 /0
φ1 φ2
0 (4.16)
α1 α2 α3
  
0 / N1 / N2 / N3 /0
ψ1 ψ2

Such a diagram induces two three term exact sequences, one formed by the
kernels of the αi ’s and one by their cokernels, and point is that the snake map
δ connects these two sequences. In other words, we have a six term exact
sequence

0 ker α1 ker α2 ker α3

δ

coker α1 coker α2 coker α3 0.

(4.17)

Lemma 4.65 (Snake lemma II) Assume given a commutative diagram with
exact row as in (4.16). Then the six term sequence (4.17) above is exact.

Proof: The sequence is trivially exact at the two extreme slots ker α1 and
coker α3 , and that the snake-part is exact, is just the Snake Lemma, so what
remains to be done is checking exactness at ker α2 and coker α2 , and this
98 ma4200—commutative algebra

follows by two simple hunts in the diagram. We shall check exactness at

ker α2 , but shall leave exactness at coker α2 for the students to practise diagram
chasing. So assume that x P ker α2 is such that φ2 pxq “ 0. Then x “ φ1 pyq for
some y P M1 , and ψ1 pα1 pyqq “ α2 pφ1 pyqq “ α2 pxq “ 0. Since ψ1 is assumed to be
injective, it follows that y P ker α1 and we are through. o

Problems

4.39 (The five lemma I.) Use the Snake Lemma to prove the following abbrevi- ˇ
ated and preliminary version of the five lemma in the next exercise. Assume
given a commutative diagram
0 / M1 / M2 / M3 /0

α1 α2 α3
  
0 / N1 / N2 / N3 /0

with exact rows. If two of the αi ’s are isomorphisms, then the third one is as
well.

4.40 (The five lemma II.) Given a commutative diagram ˇ

M0 / M1 / M2 / M3 / M4

α0 α1 α2 α3 α4
    
M0 / N1 / N2 / N3 / N4

of A-modules with exact rows. Show that α2 is an isomorphism whenever the

four other αi ’s are.
There is a slightly stronger assertion namely that if α1 and α3 are isomor-
phisms, α0 surjective and α4 injective, then α2 is an isomorphism. Prove this.
M
There is a plethora of small results like this involving diagrams of different
geometric shapes and with suggestive names like the star lemma and the
diamond lemma. Once you have grasped the essence of diagram chasing and
remember the Snake Lemma you should be safe in that corner of the territory
of homological algebra. The most important use of connecting homomor-
phisms is when constructing long exact sequences of homology associated
with a complexes; but that will be for a later occasion.
In the next two exercises one take the following statement for granted:
Proposition 4.66 Assume that A is a pid and that φ : E Ñ E is an endomorphism
of a finitely generated free A-module E. Then φ lives in a commutative diagram
0 /F /E /A /0

ψ φ
  
0 /F /E /A /0
 modules 99

where the rows are exact and where F is free of rank n ´ 1.

4.41 Infer from the previous exercise that if A “ Z and φ has non-vanishing ˇ
determinant, it holds true that the cokernel coker φ is finite and that } det φ} “
# coker φ. Hint: Use the Snake Lemma.

4.42 In the same vein as in exercise 4.41 above, assume that A “ krts is ˇ
the polynomial ring over the field k and that φ has non-vanishing determi-
nant. Prove that coker φ is of finite dimension over k and that deg det φ “
dimk coker φ Hint: Again, the snake is the solution.

4.43 (Modules of finite presentation.) One says that a module M is of finite ˇ

presentation if it sits in an exact sequence Modules of finite pre-
sentation (Moduler av
An / Am /M /0 endelig presentasjon)

where An and Am are finitely generated free modules. Let N Ď M be a sub-

module. In general this is a more restrictive condition on a module than being
finitely generated, however over Noetherian rings the two are equivalent.
Prove that if both N and M{N are of finite presentation, then the same holds
for M. Hint: Establish a diagram

0 /N /M / M{N /0
O O O

0 / Ar / A r `s / As /0

with exact rows and all three vertical map being surjective. Then apply the
Snake Lemma and Proposition 4.15 on page 79.

M
Lecture 5

Tensor products

Version 2.1 as of 2018-10-18 at 09:53 (typeset 3rd December 2018 at 10:03am). Prone to misprints
and errors and will change.

The term “tensor” appeared for the first time with a meaning resembling
the current one in 1898. The German physicists Woldemar Voigt used the
word in a paper about crystals. Tensors are these days extensively used in
physics, and may be the most prominent example is the so-called “stress-
energy-tensor” of Einstein. It governs the general theory of relativity and
thereby our lives in the (very) large!
A slightly less influential occurrence took place in 1938 when the American
mathematician Hassler Whitney when working on the universal coefficient
theorem in algebraic topology introduced the tensor product of two abelian
groups. Certain isolated cases had been known prior to Whitney’s work, but
Whitneys construct was general, and it is the one we shall give (although sub-
sequently polished by several mathematicians, in particular Nicolas Bourbaki,
and generalized to modules).
How far apart stress in crystals and the universal coefficient theorem may
appear, the concept of tensors is basically the same—the key word being Woldemar Kummer
(1850–1919)
bilinearity. German physicist

5.1 Introducing the tensor product

First of all, let us recall what a bilinear map M ˆ N Ñ L is where M, N and L are
three modules over a ring A.
It is simply what the name says, a map β which is linear in each of the two
variables; that is, when one of the variables is kept fixed, it dependents linearly
on the other. For instance, when the second variable is kept constant, it holds
true that
Hassler Whitney
βpax ` by, zq “ aβpx, zq ` bβpy, zq, (1907–1989)
American mathematician
where a and b belong to the ring A and x and y are elements in M (and ditto
when the first variable is fixed). Frequently, when several rings are around, Bilinear maps (Bilineære
avbildninger)
one says A-bilinear to be reminded which ring is considered the base ring.
102 ma4200—commutative algebra

A typical example from the world of vector spaces over a field k, would be
a scalar product on a vector space V, and within the realm of commutative
algebra, the products of an A-algebra B is a good example; the multiplication
map pa, bq ÞÑ ab is an A-bilinear map B ˆ B Ñ B.
5.1 There is naturally also the notion of multilinear maps, which involves more Multilinear maps (Multi-
ś
than two modules. In that case, the source of the map is a product i Mi lineære avbildninger)

of finitely many A-modules Mi and its target is another A-module L. The

constituting property is mutatis mutandis the same as for bilinear ones: When
all but one of the variables are kept constant, the resulting map is A-linear.

The universal property

5.2 The tensor product captures in some sense all possible bilinear maps
defined on the product of two A-modules M and N, or at least makes them
linear. This rather vague formulation becomes precise when phrased as a
universal property.
5.3 The tensor product is a pair consisting of an A-module Mb A N together
with an A-bilinear map τ : M ˆ N Ñ Mb A N that abide by the following rule: The tensor product
(Tensorproduktet)

o For any bilinear map β : M ˆ N Ñ L, there is a unique A-linear map

γ : Nb M Ñ L suct that β “ γ ˝ τ.

In other words, every A-bilinear β factors linearly via τ, as expressed by the

commutative diagramme in the margin. And as ususal with objects satis- MˆN
τ
/ Mb A N
fying a universal property, the pair τ and Mb A N is unique up to a unique
isomorphism. β
γ

! 
L
Exsistence
The construction of the tensor product is rather abstract and serves the sole
purpose of establishing the existence. It will seldom be referred to in the
sequel, if at all. To ease getting a grasp on the tensor product remember the
mantra, so true in moderen mathematics: “Judge things by what they do, not
by what they are”.
5.4 The construction starts out with the free A module F “ A Mˆ N on the set
ř
M ˆ N. The elements of F are finite, formal linear combinations i ai ¨ pxi , yi q
with xi P M, yi P M and ai P A. In particular, every pair px, yq is an element of
F, and by definition these pairs form a basis for F. We proceede by letting G
be the submodule of F generated by all expressions either of the form

pax ` a1 x1 , yq ´ apx, yq ´ a1 px1 , yq, (5.1)

or of the form
px, ay ` a1 y1 q ´ apx, yq ´ a1 px, y1 q, (5.2)
 tensor products 103

where a and a1 are elements from A while x and x1 lie in M and y and y1 in N.

The tensor product Nb A M is defined as the quotient F{G, and the residue
class of a pair px, yq will be denoted by xby. Having forced the two expres-
sions (5.1) and (5.2) above to be zero by factoring out the submodule G, we
have made xby a bilinear function of x and y; that is, the following two rela-
tions hold true in Mb A N:

pax ` a1 x1 qby “ apxbyq ` a1 px1 byq,

(5.3)
xbpay ` ay1 q “ apxbyq ` a1 pxby1 q.

In other words, the map τ : M ˆ N Ñ Mb A N sending px, yq to xby is A-

bilinear.

Proposition 5.5 The pair τ and Mb A N as constructed above satisfy the universal
property in paragraph 5.3; in other words, they are the tensor product of M and N.

Proof: We already saw that τ is bilinear, so we merely have to check the

factorization property. To that end, let β : M ˆ N Ñ L be bilinear. Since
F “ A Mˆ N is a free module on M ˆ N, we may , according to the Universal
Mapping Principle for free modules (Proposition 4.37 on page 80), define
an A-linear map βs : F Ñ L by sending a the basis-elements px, yq to the MˆN / Mb A N
values βpx, yq. Since β is bilinear, this map vanishes on the submodule G.
γ
Consequently it factors through the quotient F{G “ Mb A N and thus gives the
% 
β

wanted map γ : Mb A N Ñ L. L

Elements shaped like xby generate the tensor product, and because the
value at xby of any factorization of β is compelled to be βpx, yq, the unique-
ness of γ comes for free. o

5.2 Basic working formulas

In this section we present a few principles and properties of the tensor prod-
ucts which together with some basic formulas hopefully should help students
grasp "the spirit of the tensor product" and make it easier to work with it. We
also discuss some particular classes of modules, like cyclic modules and free
modules, which behave particularly well when exposed to a tensor product.

Decomposable tensors.

5.6 For several reasons, tensors of the form xby deserve a special name; they
are dubbed decomposable tensors. Only in a very few highly special cases all Decomposable tensors
elements in a tensor product will be decomposable; the usual situation is that (Dekomponerbare
tensorer)
most are not (A simple example is discussed in Problem 5.13 on page 113
below. See also Example 5.6 on page 120). A general element in Mb A N may
ř
however, be expressed as a finite linear combination i ai ¨ xi byi of decompos-
able tensors since this is already true in the free module F “ A Mˆ N .
104 ma4200—commutative algebra

Consequently, if txi u is a set of generators of M and ty j u one for N, the de-

composable tensors txi by j u form a set of generators for Mb A N; in particular,
if both factors are finitely generated, the same holds for the tensor product
Mb A N.
5.7 To define a map φ from Mb A N into any module, it suffices to give the
values of φ on decomposable tensors xby, provided these values depend
bilinearily on x and y. This is an informal and convenient reformulation of
the universal property from Paragraph 5.3, certainly more suggestive than
working with pairs px, yq.
5.8 Another useful property of decomposable tensors is subsumed in the
slogan “scalars can be moved past the tensor product”; or in precise terms, for
every element a P A it holds true that

paxqby “ xbpayq.

This is a simple consquence of the fundamental bilinear relations (5.3) on

page 103; with the notation of (5.3), just set x1 “ y1 “ 0.

Functoriality
Linear maps between A-modules are fundamental tools in algebra, and and
it comes as no surprise that exploring how maps behave when exposed to
tensor products occupies a large of the theory. As a modest start we shall
observe that the tensor product construct is functorial, in the precise meaning
that when considered a function of either variable, it gives a functor Mod A Ñ
Mod A ; so we have to tell how to tensorize maps.
5.9 Any A-linear map φ : M Ñ M1 gives rise to an A-linear map Mb A N Ñ
M1 b A N that on decomposable tensors acts as xby ÞÑ φpxqby: The expression
φpxqby depending bilinearily on x and y, this is a viable definition, and the
resulting map is naturally babtized φbid N .
Obviously, it holds true that ψbid N ˝ φbid N “ pψ ˝ φqbid N when ψ and φ
are two composable maps (the identity holds for decomposable tensors), and
clearly id M bid N “ id Mb A N . This means that the pair of assignments

M ÞÑ Mb A N and φ ÞÑ φbid N

define a functor ´b A N : Mod A Ñ Mod A .

Proposition 5.10 The functor ´b A N is an A-linear functor. In particular, it

transforms direct sums into direct sums.

A formal consequence of a functor being additive is that it preserves direct

sums (as we established in Problem 4.34 on page 89), however we shall come
back to the matter below and sketch an ad hoc proof when discussing Proposi-
tion 5.13 below.
 tensor products 105

Proof: Recall that in Exercise 4.34 on page 89 we introduced the notion of

additive functors: Saying the functor is additive is saying it transforms sums additive functors (Addi-
of maps to sums of maps, and it is A-linear if it additionally respects products tive funktorer)
A-linear functors
with scalars; expressed in symbols this reads (Lineære fuktorer)

paφ ` bψqbid N “ a ¨ φbid N ` b ¨ ψbid N . (5.4)

This follows easily from how ´b A N acts on maps together with the basic
bilinear relations in (5.3) on page 103. Indeed, one finds

paφ ` bψqbid N pxbyq “ppaφpxq ` bψpxqqby “

` ˘
“ aφpxqby ` bψpxqby “ a ¨ φbid N ` b ¨ ψbid N xby,

and the two sides of (5.4) agree on decomposable tensors. Hence they are
equal since the decomposable tensors generate Mb A N.
5.11 The situation is completely symmetric in the two variables, so if ψ : N Ñ
N 1 is a map, there is a map id M bψ from MbN to MbN 1 that sends xby to
xbψpyq, and naturally, on sets φbψ “ pφbid N q ˝ pid M1 bψq.

Some formulas

5.12 When working with tensor products a series of formulas are invaluable.
Here we given the most basic ones revealing the multiplicative nature of the
tensor product; together with the direct sum it behaves in a way resembling
the product in a ring.

Proposition 5.13 Suppose that M, N and L are modules over the ring A. Then
we have the following four canonical isomorphisms.

o Neutrality: Mb A A » M;

o Symmetry: Mb A N » Nb A M;

o Associativity: pMb A Nqb A L » Mb A pNb A Lq;

o Distributivity: pM ‘ Nqb A L » pMb A Lq ‘ pNb A Lq.

There are some comments to be made. Firstly, these isomorphisms are so

natural that for all practical purposes they may be consider as identities.
Secondly, the general mechanism that extends associativity from products
with three factors to products with arbitrary many factors applies to tensor
products, so that any number of parentheses placed in any way in a tensor
product with any number of factors can be resolved. And finally, an easy
induction establishes the fourth property for any number of summands; with
a somehow subtler argument, one may even show it holds for infinitely many.
Proof: In each case we indicate how a pair of mutual inverses A-linear maps
acts on decomposable tensors; this will basically suffice in the two first cases,
but in particular the case of associativity, requires some more work.
106 ma4200—commutative algebra

Neutrality: For the first formula the actions on decomposables are xba Ñ xa
and x ÞÑ xb1. The product xa is bilinear in x and a and therefore the map
xba Ñ xa descends to an A-linear map Mb A A Ñ M which obviously have
x ÞÑ xb1 as inverse.
Symmetry: In this case the short hand assignments are xby Ø ybx. In both
directions the assignments are bilinear in view of the fundamental relations
(5.3), and obviously they define inverse maps.
Associativity: This case is more subtle than one should believe at first sight;
the (very short) shorthand definition would be pxbyqbz Ø xbpybzq, but
this is not viable since xby is not a general member of Mb A N. To salvage the
situation one introduces some auxiliary maps, one for each z P L.
So, fix an element z from L and define an A-linear map ηz : Mb A N Ñ
Mb A pNb A Lq by the assignment xby ÞÑ xbpybzq, which is legitimate since
the expression xbpybzq is bilinear in x and y (the third variable z is kept
fixed).
Obviously the map ηz ptq is linear in z and a priori being linear in t, it de-
pends bilinearily on t and z. We infer that sending tbz to ηt pzq induces a map
pMb A Nqb A L Ñ Mb A pNb A Lq. On decomposable tensors this map behaves as
wanted; that is, it sends pxbyqbz to xbpybzq.
A symmetric construction yields a map the other way which sends a de-
composable tensor xbpybzq to pxbyqbz. Finally, these two maps are mutually
inverses since they act as inverse maps on the decomposables, and the decom-
posables generate the tensor products.
Distributivity: Another way of phrasing this is to say at the tensor product
respects directs sums, and this we already established in Proposition 5.10
above. However to at least halfway fulfil our promise, we vaguely indicate the
beginning of an ad hoc proof in the flavour of the three preceding parts of the
proof. It is based on the short hand version px, yqbz Ø pxbz, ybzq, and the
salient point is to extend these to maps defined on the entire tensor products
using bilinearity; but, of course, details are left to the benefit of the zealous
students. o
5.14 It is worth while to dwell a little on the associativity. Since the paren-
theses are irrelevant, we may as well skip them and write Mb A Nb A L for
MbpNb A Lq (or for that matter for pMb A Nqb A L). According to the universal
property of the tensor product bilinearity is the clue for defining maps having
source Mb A N, and there is a similar trilinearity principle for defining maps
sourced in a triple tensor product Mb A Nb A L. It suffices to specify φ on de-
composable tensors xbybz as long as the specifying expression is trilinear in
x, y and z; the precise statement is a as follows:

Lemma 5.15 (Trilinearity principle) Let A be a ring and M N, K and L

four A-modules. Assume given a map φ : M ˆ N ˆ K Ñ L such that φpx, y, zq
depends in a trilinear manner on the variables. Then there is unique A-linear map
 tensor products 107

φ : Mb A Nb A K Ñ L such that φpxbybzq “ φpx, y, zq.

Proof: The argument is mutatis mutandis the same as we gave in the proof
of Proposition 5.13 concerning the associative law: Fix an element z P K and
consider φpx, y, zq; it depends in a bilinear manner on x and y and hence gives
rise to a linear map ηz : Mb A N Ñ L. The dependence of ηz on z obviously
being linear, assigning ηz ptq to tbz for t P Mb A N and z P K is a bilinear in z
and t and therefore yields the desired map pMb A Nqb A K Ñ L.
And again, the map is unambiguously determined since its values are
prearranged on the decomposable tensors which generate pMb A Nqb A K. o
As a final comment, there is nothing special about the number three in this
context. A similar statement—that is, a principle of multi-linearity—holds
true for tensor products with any number of factors, but we leave that to the
imagination of the reader.
Problem 5.1 Consider the four isomorphisms in Proposition 5.13 on page 105.
Be explicit about what it means that they are functorial (in every variable in-
volved) and prove your assertions. M

The case of cyclic modules

We now turn to discuss two situation which are frequently met when working
with tensor products. Hopefully they will illuminate the working mechanism
of the tensor product, but anyhow, they are illustrate some of the different
phenomena that can occur.
5.16 Our first example is about the tensor product of two cyclic module and
reads as follows:

Proposition 5.17 Let A be a ring. For any two ideals a and b in A it holds true
that A{a b A A{b » A{pa ` bq. In particular, one has A{a b A A{b “ 0 if and only if
the two ideals a and b are comaximal.

Proof: Sending a pair prxsa , rysb q from A{a ˆ A{b to the element rxysa`b in
A{a ` b is well defined and bilinear; indeed, changing x (resp. y) by a member
of a (resp. b) changes xy by a member of a (resp. b) as well, so the map is well
defined, and a product clearly depends bilinearily on the factors. This induces
a map A{ab A A{b Ñ A{a ` b.
One the other hand, the tensor product A{ab A A{b is a cyclic A-module
generated by 1b1 because rasa brbsb “ ab ¨ 1b1, and clearly elements from both
the ideals a and b kill it since

x ¨ 1b1 “ px ¨ 1qb1 “ 1bpx ¨ 1q.

So A{ab A A{b is a quotient of A{a ` b and consequently it is squeezed between

two copies of A{a ` b, by two maps both sendig 1 to 1, and hence all three must
coincide. o
108 ma4200—commutative algebra

The proposition shows that the tensor product of two non-zero modules
very well may vanish, and for cyclic modules this happens precisely when
the respective annihilators are comaximal. We also observe that taking b “ a
yields an isomorphism A{ab A A{a » A{a.

Finite abelian groups

5.18 A modest instance of the tensor product being zero, is found among
finite abelian groups. Powers of two relatively prime integers p and q are
comaximal—for natural numbers µ and ν it holds that ppµ , qν q “ Z—and we
infer that
Z{pµ ZbZ Z{qν Z “ 0.

We also infer that if the two natural nu,bers satify µ ď ν it holds that true
that
Z{pµ ZbZ Z{pν Z » Z{pµ Z

for any prime number p since ppµ , pν q “ ppminpν,µq q. Together with the formu-
las from Proposition 5.13 and the Fundamental Theorem for Finitely Abelian
Groups, these two formulas make it clear how to compute the tensor product
of any pair of finite abelian groups.

The case of free modules

5.19 In the second example we show that the tensor product of two free
modules is free.

Proposition 5.20 Assume that E and F are free A-modules. Then the tensor
product Eb A F is free. More precisely, if tei uiP I and t f j u jP J are bases for respectively E
and F, the tensors ei b f j with pi, jq P I ˆ J form a basis for Eb A F.

This proposition holds true regardless of the cardinalities of I and J, but the
case when E and F are of finite rank, warrants to be mentioned specially. One
may deduce the finite tank case from Proposition 5.13 by a sta esforward
induction, however we offer another and simple proof that is generally valid.

Corollary 5.21 If E and F are free A-modules of finite ranks n and m respectively,
the tensor product Eb A F is free of rank nm. In particular, for vector spaces V and W
of finite dimension over a field k it holds true that dimk Vbk W “ dimk V ¨ dimk W.

Proof of Proposition 5.20: We contend that the set tei b f j upi,jqP I ˆ J is a

basis for the tensor product Eb A F. As already observed, the elements ei b f j
form a generating set forEb A F so we merely have to verify they are linearly
independent.
Denote by ai pxq the i-th coordinate of an element x P E relative to the basis
ř
tei u; that is, one has x “ i ai pxqei . Similarily, let β j pyq be the j-th coordinate
 tensor products 109

of an element y P F. All the ai pxq’s and all the b j pyq’s depend linearly on their
arguments.
For each pair of indices µ P I and ν P J the expression aµ pxqbν pyq depends
bilinearly on x and y and therefore xby Ñ aµ pxqbν pyq gives a map δµν : EbF Ñ
A. This map vanishes on ei b f j unless i “ µ and j “ ν, and takes the value one
on eµ b f ν .
With the δµν ’s up our sleeve, the rest is a piece of cake: Just apply δµν to a
potential dependence relation
ÿ
cij ¨ ei b f j “ 0

to obtain cµν “ 0 for every pair µ and ν of indices. o

5.3 Functorial properties

5.22 The tensor product functor ´b A N and the functor Hom A pN, ´q live
in a close partnership; they form what is called a pair of adjoint functors in
the vernacular of homological algebra; that is, there is an identity as in the
following proposition:

Proposition 5.23 (Adjointness) There is a functorial isomorphism

Hom A pMb A N, Lq » Hom A pM, Hom A pN, Lqq.

The word “functorial” refers to all three variable. The dependence is covariant
in L and contravariant in M and N (A sanity check is that the variances are the
same on both sides!). We refrain from diving into the general details afraid of
laying a notational smokescreen over the matter which is quit simpe, but in
Example 5.1 below we discuss the situation with M being the variabale.
Proof: The salient point is that Hom A pM, Hom A pN, Lqq is canonically isomor-
phic to the space of bilinear maps M ˆ N Ñ L; and once one realizes that, the
proposition becomes just another reformulation of the Universal Property of
the tensor product.
One may think about the members of Hom A pM, Hom A pN, Lqq as being
maps Φpx, yq defined on M ˆ N: When x is a specified member of M, the
corresponding map Φpx, ´q from N to L is given as y ÞÑ Φpx, yq. That Φpx, ´q
is A-linear, means that Φpx, yq is linear in y, and that Φpx, ´q depends linerly
on x, means that Φpx, yq is linear in x: Hence at the end of the day Φpx, yq is
bilinear!
And that’s it: By the Universal Property of the tensor product any such
map can be written unambiguously as Φpx, yq “ φpxbyq with a linear map
φ : Mb A N Ñ L. o

Example 5.1 As promised we shall check functoriallity with respect to maps

110 ma4200—commutative algebra

β : M Ñ M1 ; that is, we shall prove the formula

θ M ˝ pβbid N q˚ “ β˚ ˝ θ M1 , (5.5)

or in other words verify the diagram below commutes, whose vertical maps
are the isomorphisms θ M and θ M1 furnished by Proposition 5.23:

p βb id N q˚
Hom A pM1 b A N, Lq / Hom A pMb A N, Lq

θ M1 θM
 
Hom A pM1 , Hom A pN, Lqq / Hom A pM, Hom A pN, Lqq
β˚

Now, β˚ acts on maps by composition from the right so that β˚ pΦq “ Φ ˝ β.

That is, β˚ pΦqpx, yq “ Φpβpxq, yq, and hence

β˚ θ M1 pφqpx, yq “ φpβpxqbyq.

The same applies to pβbid N q˚ so that pβbid N q˚ pφqpxbyq “ φpβpxqbyq. And

voilà: We find
θ M pβ˚ bid N qφpx, yq “ φpβpxqbyq,

and the two sides of equation (5.5) agree. K

Problem 5.2 Convince yourself (along the lines of this example) that the
isomorphisms in Proposition (5.23) are functorial in N and L as well. M

Right exactness
In analogy with the notion of left exactness, which we discussed in connec-
tion with the hom-functors, a covariant and additive functor F between two
module-categories1 Mod A and ModB is said to be right exact if it transforms 1
Or more generally,
exact sequences shaped like between two abelian
categories

M0 / M1 / M2 /0 Right exact functors

(Høyre eksakte funktorer)

into exact sequences shaped like

FpM0 q / FpM1 q / FpM2 q / 0.

A fundamental and most useful property of the tensor product is that it is

right exact. This section is devoted to giving a proof this, with some easy
consequences included atthe end.
5.24 Here it comes:

Proposition 5.25 (Right exactness) Given a ring A and an A-module N.

The functor ´b A N is a right exact functor.
 tensor products 111

Our approach relies on Proposition 5.23 above and illustrates the general fact
that adjoint functors tend to share exactness properties; if one is exact in some
sense, the other tends to be exact in a related sense. It is possible to give a
proof of right exactness based on the construction of the tensor product. This
is however tedious, cumbersome and not very enlightening, and according to
our mantra should be avoided.
It is common usage to call N a flat A-module if the functor p´qb A N is exact; Flat modules (Flate
i.e. when it transforms injective maps into injective maps. moduler)

Proof: Let the exact sequence2 2

A notation like M‚
is commonplace in
M‚ : M0
α / M1 β
/ M2 /0 homological algebra for
so-called complexes; the
bullet indicates a slot
be given, and the task is to show that the sequence where to place indices.
A complex is a linear
αb id N βb id N
M‚ b A N : M0 b A N / M1 b A N / M2 b A N /0 sequence of A-linear
maps the composition
of two consecutive once
is exact. Our tactics will be to apply the principle of left exactness of the
being zero.
hom-functor as expressed in the proposition called Left Exactness II (Propo-
sition 4.61 on page 92), and in fact, we shall do this twice. With that principle
in mind, we start out by observing that the sequence Hom A pM‚ b A N, Lq being
exact for every A-module L will be sufficent, and this sequence appears as the
upper line in the following grand diagram.

p βb id N q˚ pαb id N q˚
0 / Hom A pM2 b A N, Lq / Hom A pM1 b A N, Lq / Hom A pM0 b A N, Lq
»

»
θ M2
  
0 / Hom A pM2 , Hom A pN, Lqq / Hom A pM1 , Hom A pN, Lqq / Hom A pM0 , Hom A pN, Lqq
β˚ α˚

The next step is to evoke Proposition 5.23 above and replace the complex
Hom A pM‚ b A N, Lq with Hom A pM‚ , Hom A pN, Lqq; the latter is displayed
as the bottom line of the grand diagram. The crux of the proof is that this
latter sequence is exact, again by Left Exactness II, so once we know
that the two rows in the grand diagram are isomorphic (as sequences) we
are through. But indeed, they are since with the vertical maps being the
canonical isomorphisms from Proposition 5.23 (Adjointness), the two
squares commute according to Example 5.1 above. o
5.26 Proposition 5.17 on page 107 describes the tensor product of two cyclic
module. An analogous result holds true with just one of the modules being
cyclic while the other can be arbitrary:
Proposition 5.27 Let a Ď A be a an ideal and M and A-module. Then one has a
canonical isomorphism Mb A A{a » M{aM that sends mbras to rams.

Proof: The starting point is the exact sequence

0 /a /A / A{a / 0,
112 ma4200—commutative algebra

which when tensorized by M, yields the exact sequence

ab A M
α /M / Mb A A{a / 0,

because the tensor product is right exact. The map α sends abx to ax, hence
its image is equal to aM, and we are done. o

Example 5.2 Be aware that the tensor product can be a bloodthirsty killer.
Injective maps may cease being injective when tensorized, and they can even
become zero. The simplest example is multiplication by an integer n, that is;
the map Z Ñ Z that sends x to nx. It vanishes when tensorized by Z{nZ.
This also illustrates the fact that the functor ´b A N is not always exact, even
though always being right exact. In this example the exact sequence

0 /Z n /Z / Z{nZ / 0,

is transformed into the exact sequence (right exactness of b)

/ Z{nZ / Z{nZbZ Z{nZ / 0,

0 β
Z{nZ

where β must be an isomorphism—its kernel is zero since the sequence is

exact. So part of the conclusion is that Z{nZbZ Z{nZ “ Z{nZ which as well
ensues from Proposition 5.17 on page 107. K
5.28 Even whole modules may succumb under the action of tge tensor prod-
uct; for instance, we saw that Z{pZbZ Z{qZ “ 0 when p and q are relatively
prime integers, which illustrates a general fact. Recall that an A-module M is
divisible by an element a P A if the multiplication map M Ñ M is surjective; in
other words every x P M may be written as ax1 for some x1 P M.

Proposition 5.29 Let a P A and assume that M and N are A-modules such that
M is divisible by a and a P Ann N, then Mb A N “ 0.

Proof: The short argument goes like this. Every x P M is of the form ax1 for
some x1 P M, so that xby “ ax1 by “ x1 bay “ 0, and as the decomposable
tensors generate Mb A N we are through. o
 tensor products 113

Problems

5.3 Compute Z{16ZbZ Z{36Z and pZ{6Z ‘ Z{15ZqbZ pZ{21Z ‘ Z{14Zq. ˇ

5.4 Let A be a ring and M an A-module. Show that Mb A A{ Ann M “ M. ˇ

Show that if N is a second A-module and the annihilators Ann M and Ann N
are comaximal ideals, then Mb A N “ 0.

5.5 Show that Q{ZbZ Q “ 0. Show that one has QbZ Q » Q, but that ˇ
Q{ZbZ Q{Z “ 0. Hint: Proposition 5.29.

5.6 Let a be a proper ideal in the ring A. Assume that M is an A-module for
which there is surjection M Ñ A{a. Show that aM ‰ M.

5.7 Let M be finitely generated module over the ring A: Show that Mb n ‰ 0 ˇ
for any natural number n. Hint: Exhibit a surjective map M Ñ A{p. Proceed
by induction on n and right exactness of the tensor product.

5.8 (The Kronecker product.) Let φ : E Ñ F and ψ : G Ñ H be two A-linear

maps between free A-modules of finite rank. Let tei uiP I , t f j u jP J , tgk ukPK and
thl ul P L be bases of E, F, G and H respectively, and let the matrices of φ and
ψ in the appropriate bases be Φ and Ψ. Show that the matrix of φbψ in the
bases tei b f j upi,jqP I ˆ J and tgk bhl upk,l qPKˆ L is given as the matrix pΦij Ψkl q with
rows indexed by pairs pi, jq P I ˆ J and colums by pk, lq P K ˆ L. This matrix is
called the Kronecker product of Ψ and Φ.

5.9 Assume that G is a finite abelian group. Show that GbZ Q “ 0. Leopold Kronecker
(1823–1891)
5.10 Let A be a domain contained in a field K and let M be an A-module. German mathematician
Assume that Ann M ‰ p0q. Prove that Kb A M “ 0. The Kronecker product
(Kronecker-produktet)
5.11 Let A Ñ B be a surjective ring homomorphism. Show that for any two ˇ
B-modules M and N (which automatically are A-modules) it holds true that ˇ
ˇ
Mb A N “ MbB N.

5.12 Show that ZrisbZ Zris is a free abelian group of rank 4 while ZrisbZri sZris “ ˇ
Zris and is of rank two as an abelian group.

5.13 Merely in a very few highly special cases will all elements in a tensor
product be decomposable; the commonplace situation is that most are not. A
simple example is W “ Vbk V where V is a two-dimensional vector space over
k. Let te1 , e2 u be a basis for V. The tensor product W is of dimension four with
a basis tei be j u where 1 ď i, j ď 2. Let xi be coordinates relative to this basis;
that is, any vector v is expressed as v “ x1 ¨ e1 be1 ` x2 ¨ e1 be2 ` x3 ¨ e2 be1 ` x4 ¨
e2 be2 ,
a) Establish that he decomposable tensors are shaped like

pue1 ` ve2 qbpse1 ` te2 q “ us ¨ e1 be1 ` ut ¨ e1 be2 ` vs ¨ e2 be1 ` vt ¨ e2 be2 ,

114 ma4200—commutative algebra

with u, v, s and t being scalars.

b) Show that the decomposable tensors are precisely those lying on the subset
x1 x4 ´ x2 x3 “ 0.
c) In the real case, that is when k “ R, convince youself that this locus is the
cone in R4 with apex the origin over a saddle-surface in R3 ; i.e. one given as
z “ xy (or in our coordinates x3 “ x1 x2 ).

5.14 (The rank stratification.) Let V be a vector space over a field k of finite
dimension. The dual space V ˚ “ Homk pV, kq consists of linear functionals on
V and is a vector space of the same dimension as V. Chose a basis tei u for V
and let tφi u be the dual basis for V ˚ . That is, the functionals φi are defined as
φi pe j q “ 0 when i ‰ j and φi pei q “ 1.
a) Let W be a second vector space of finite dimension over k with basis t f j u.
Prove that the assignment φbw to v ÞÑ φpvqw induces an isomorphism
»
Γ : V ˚ bk W Ñ Homk pV, Wq.
b) Given a linear map θ : V Ñ W whose matrix relative to the bases tei u
and t f j u is paij q. Show that the element in V ˚ bW corresponding to θ equals
ř ř ř
i φi bθpei q and that i φi bθpei q “ ij aij φi b f j .
c) Show that the non-zero decomposable tensors in V ˚ bk W under the map Γ
correspond to the linera maps of rank one. Hint: Chose an appropriate basis
for V.
d) Show that a linear map in Homk pV, Wq is of at most rank r if and only if the
corresponding tensor in V ˚ bW is the sum of at most r decomposable tensors.
Hint: Chose an appropriate basis for V

5.4 Change of rings

A commonplace application of the tensor product is to change the ground
ring.
Suppose that B is an A-algebra. In particular B has an A-module sructure,
and consequently we may form the tensor product Mb A B of B with any A-
module M. Now, the point is that one may multiply elements in Mb A B by
elements from B "in the second variable", and in that way produce a B-module
structure on the tensor product Mb A B. Indeed, if b P B the "multiplication by
b map" x ÞÑ bx is an A-linear map rbs : B Ñ B, and therefore induces a map
id M brbs on the tensor product. Its action on a tensor t will we denoted b ¨ t or
simply bt, and on decomposable tensors it acts as b ¨ xbc “ xbbc. Checking the
module axioms is straightforward (for free from functoriality of ´b A B), and
in view of the tensor product being additive (Proposition 5.10 on page 104)
and right exact (Proposition 5.25 on page 110) we arrive at
 tensor products 115

Proposition 5.30 Given an A-algebra B. The the change-of-rings-functor, which

sends M to Mb A B, is a right exact additive functor ´b A B : Mod A Ñ ModB .

5.31 Notice, that elements in B coming from A may be moved past the b-sign;
i.e. if the structure map is u : A Ñ B one has xbab “ axbb. Be aware however,
there is a hidden pitfall. For any b P B the notation a ¨ b is a sloppy version
of the correct notation upaq ¨ b, where the product is the product in the ring B.
Hence axbb “ xbupaqb would be the correct way of writing.
This leads to equalities like apxbbq “ upaqpxbbq where on the left side
Mb A B is considered an A module (through the left factor) and on the right a
B-module (through the right factor).

Transitivity and adjointness

5.32 Sometimes one wants to perform consecutive base changes, and the
tensor product behaves well in such a situation. It is transitive in the following
sense.

Proposition 5.33 (Transitivity) Assume that B is an A-algebra and C is

a B-algebra. Then there is a canonical isomorphism pMb A BqbB C » Mb A C as
C-modules.

Proof: The short description of the maps are mbxby ÞÑ mbxy and mbz ÞÑ
mb1bz. By the principles of bi- and tri-linearity both extend to maps between
the tensor products, and they act as mutually inverses on the decomposable
tensors. Hence the are mutually inverse maps. o

5.34 Changing the base ring preserves tensor products, but not hom-modules
in general. One has the following:

Proposition 5.35 Let B be an A algebra and M and N two A-modules. Then

there is a canonical isomorphism of B-modules

pMb A Nqb A B » pMb A BqbB pNb A Bq.

In other words, the functor p´qb A B takes tensor products into tensor products.

Proof: Two mutually inverse B-module homomorphisms are defined by the

assignments
xbybb ÞÑ xb1bybb
xbybbb1 Ðß xbbbybb1
of decomposable tensors. The obey respectively a tri-linear and a quadri-linear
requirement and thereby define genuine maps between the modules. o

5.36 In a base change situation there are two natural functors between the
two module categories. In addition to the base change functor, given as the
116 ma4200—commutative algebra

tensor product p´qb A B : Mod A Ñ ModB , there is a functor going the other
way: p´q A : ModB Ñ Mod A . The action of this functor is kind of trivial. If N is
a B-module, NA is equal to N but regarded as an A-module—one forgets the
B-module structure. With maps the same happens; they are kept intact, but
merely regarded as being A-linear. Such functors that “throws away” part of
the structure, are called forgetful functors in the parlance of the category theory.
The point is that the tensor product p´qb A B and the forgetful functor p´q A
are adjoint functors:
Proposition 5.37 (Adjointness) Given an A-algebra B. Then there is a canon-
ical isomorphism
HomB pMb A B, Nq » Hom A pM, NA q.

Proof: The from left two right is simply a “restriction” map. It sends a given
B-linear map φ : Mb A B Ñ N to φpxb1q which is A-linear in x. To define
a map from the right side to the left side, let φ : M Ñ NA be A-linear. The
expression ψpxbbq “ bφpxq is A-bilinear and by the universal property enjoyed
by the tensor product it extends to an A-linear map ψ : Mb A B Ñ N which
turns out to be B-linear (remember, multiplication by elements from B is
performed in the right factor):

ψpb1 ¨ xbbq “ ψpxbbb1 q “ bb1 φpxq “ b1 ¨ ψpxbbq.

At last and as usual, the two maps are inverses to each other, since they are
when acting on decomposable tensors. o

Maps between free modules and base change

The tensor product being an additive functor, it is clear that the change-of-ring-
functors transforms free modules into free modules; indeed, if E » nA with
a basis tei u corresponding to the standard basis tei ui of nA, it holds true that
Eb A B » nB with a basis tei b1ui corresponding to the standard basis tei ui (as a
basis for nB this time).
5.38 It is of interest to know how changing the base ring affects maps between
free modules and how their associated matrices change. So let F be a second
free A-module of rank m with a basis t f j u and suppose further that φ : E Ñ F
is an A-linear map. Recall that the entires of the matrix Φ “ paij q associated
with φ are the coefficients in the developments
ÿ
φpei q “ a ji f j
j

of φpei q in terms of the basis elements f j . Applying φbidB to the basis element
ei b1, yields
ÿ ÿ
φbidB pei b1q “ φpei qb1 “ p a ji f j qb1 “ upa ji q ¨ f j b1
j
 tensor products 117

where u : A Ñ B denotes the structure maps as in paragraph 5.31 above.

Hence the matrix of φbidB is the matrix pupa ji qq obtained by applying the
structure map u to the entries of pa ji q.
Example 5.3 As an example we consider the polynomial ring A “ Crxs and
let a P C be a complex number. Furthermore, we let B “ Crxs{px ´ aq » C,
so that the structure map u is the evaluation at a; i.e. upPpxqq “ Ppaq. If φ is
a map between two free Crxs-modules with matrix P “ pPij pxqq relative to
some bases, the matrix of φbidC is just the matrix P evaluated at a; that is, the
matrix pPij paqq. K
Example 5.4 For a second example, take A “ Z and B “ Z{pZ “ F p for
some prime number p. If φ is a map between two free abelian groups whose
matrix relative to some bases is pnij q, the matrix of φbidF p relative to the
corresponding bases will be prnij sq (where rns as usual denotes the congruence
class of an integer n modulo p).
On the other hand, changing the ring from Z to Q; that is, passing to the
map φbidQ , does not change the matrix. We merely consider the integers nij
as being rational numbers! K
Example 5.5 Our third example is of a rather different flavour than the two
previous ones. This time we let A be a ring of characteristic p; that is, we
assume it contains the prime field F p . The Frobenius map a Ñ a p is then a
ring homomorphism A Ñ A and gives A an alternative A-algebra structure
in which a ¨ x “ a p x (the product to the left is the new product, whereas the
one to the right is the original product in A). In view of the considerations
above, a ring-change via the Frobenius map, changes matrices of A-linear
maps between free A- modules by rising their entries to the p-th power. K

5.5 Tensor products of algebras

Setting the stage of this section we let C and introduce two C-algebras A and
B. The star of the show will be the tensor product AbC B and the objective of
the play is to give a ring structure compatible with its underlying C-module
structure, thus making it a C-algebra.
5.39 On decomposable tensors the product ought to abide by the rule

abb ¨ xby “ axbby, (5.6)

and indeed, this extends to a product on AbC B:

Proposition 5.40 Given a ring C and two C-algebras A and B. Then there is a
unique C-algebra structure on AbC B whose product on decomposable tensors satisfies
abb ¨ a1 bb1 “ aa1 bbb1 .

Notice that the C-module structure on AbC B is in pace a priori being the in-
duced structure on the tensor product, so merely the ring structure is lacking.
118 ma4200—commutative algebra

Proof: To argue that the assignment of (5.6) can be extended to give a prod-
uct of arbitrary tensors, we once more we appeal to the principle of bilinearity
(at the end of paragraph 5.2 on page 103). In fact, we shall apply it twice—
once for each factor—the basic observation being that the right side of (5.6) is
C-bilinear both in pa, bq and px, yq
The first application of the principle shows that multiplication by a fixed
pair pa, bq extends to a C-linear map AbC B Ñ AbC B. This yields a map

η0 : A ˆ B Ñ HomC pAbC B, AbC Bq

that sends pa, bq to the multiplication-by-abb-map; that is, it holds that

ř ř
η0 pa, bqp i xi byi q “ i axi bbyi . In its turn, this map depends bilinearily
on the pair pa, bq, and by a second application of the principle we arrive at a
C-linear map
η : AbC B Ñ HomC pAbC B, AbC Bq,
which on decomposable tensors behave as desired; i.e. ηpabbqpxbyq “ axbby.
Subsequently we define the product of two arbitrary tensors s and t as s ¨ t “
ηptqpsq. This establishes the product in Cb A B, but there are verifications to be
done.
That the ring axioms hold, is a matter of straightforward verifications
—they follow by the uniqueness parts of the principles of bilinearity and
trilinearity. For example, both expressions ηptqpsq and ηpsqptq are bilinear in s
and t and since they agree on decomposable tensors, they are equal, so that
the product is commutative. One checks associativity in a similar manner,
but by using the Trilinearity Principle (Lemma 5.15 on page 106). The two
expressions ηptuqpsq and ηptqpusq are both linear in each of the variables s, t
and u, and agreeing on decomposable tensors, they coincide; that is, pt ¨ uq ¨ s “
t ¨ pu ¨ sq. o

The universal property

The tensor product AbC B enjoys a universal property that plays a paramount
role in algebraic geometry. It reflects the geometric construction of so called
fibre products; the simple variant being the product X ˆ Y of two schemes. This
is a foundational construct on which the whole theory rest.
5.41 With the setting as in the previous section, there are two canonical C-
algebra homomorphisms having target AbC B; one with A as source and the
other sourced at B.
The first, call it ι A , is given as a ÞÑ ab1 and the second, call it ι B , as b ÞÑ
1bb. Elements from C may be moved past the tensor product sign so that
cb1 “ 1bc for c P C; or expressed in terms of a diagrams: The upper diagram
in the margin commutes. (Where the maps C Ñ A and C Ñ B are the
structure maps defining the C-algebra structures.)
 tensor products 119

The tensor product is universal among C-algebras living in such diagrams:

AbC B
? _
Proposition 5.42 (The universal property) With the notation as just ιA ιB

introduced, assume given a C algebra D and two C-algebra homomorphisms η A : A Ñ

D and ηB : B Ñ D. Then there is a unique map of C-algebras AbC B Ñ D such that A_ ?B
η ˝ ι A “ η A and η ˝ ι B “ ηB .

Proof: Indeed, the expression η A paqηB pbq is C-bilinear, and according to

C
the universal property of the tensor product it gives rise to a C-linear map
AbC B Ñ D satisfying ηpabbq “ η A paqηB pbq. This is our desired map, but
some checking remains to be done. Let us begin with verifying that η respects
products. Since e know η is linear, it will suffice to do this for decomposable B DO \
tensors: η

ηpaa1 bbb1 q “ η A paa1 qηB pbb1 q “ η A paqη A pa1 qηB pbqηB pb1 q “ ηpabbqηpa1 bb1 q ηA
AbC B
ηB
? _
where the two extreme equalities hold true by the very definition of η, and the
ιA ιB
middle one because both ηa and ηB are ring maps.
A_ ?B
Next, one has η ˝ ι A “ η A and η ˝ ι B “ ηB since η A p1q “ ηB p1q “ 1, and finally,
that η is unique follows, since it is determined by the values on decomposable
tensors, and these satisfy
C
ηpabbq “ ηppab1qp1bbqq “ ηpab1qηp1bbq “ η A paqηB pbq

Polynomial rings
We continue with the stage set as above, with B being an A-algebra through
the structure map u : A Ñ B. In the name of the inherent human laziness, we
shall write Arx‚ s for a polynomial ring Arx1 , . . . , xr s.
5.43 A natural question is how polynomial rings behave under base change,
and the answer is they do in the obvious and simplest way.
There is a map of A-algebras Arx1 , . . . , xr s Ñ Brx1 , . . . , xr s sending xi to xi
ř ř
(hence a polynomial α aα x α maps to α upaα qx α ). Together with the inclu-
sion of B as the constants in Brx1 , . . . , xr s it induces, in view of the universal
property of the tensor product, a map of A-algebras Arx1 , . . . , xr sb A B Ñ
Recall the notation
Brx1 , . . . , xr s. It sends the variable to the variables, and gives an isomorphism: from Paragraph 1.16
on page 15 where the
Lemma 5.44 Let A be a ring and B be an A-algebra. Then Arx1 , . . . , xr sb A B “ α
monomial x1 1 ¨ . . . ¨ xrαr
was denoted by x α ,
Brx1 , . . . , xr s
with α being the multi-
index pα1 , . . . , αr q.
Proof: Considered as A-module, the polynomial ring Arx1 , . . . , xr s is free, and
the monomials x α , with α running through all multi-indices, form a basis. The
same holds for Brx1 , . . . , xr s; the monomials x α form a B-basis.
120 ma4200—commutative algebra

According to what we figured out in Paragraph 5.36 on page 116 the ele-
ments x α b1 form a B basis for Arx1 , . . . , xr sb A B; but our map sends these to
x α which form a basis for Brx1 , . . . , xr s. o

Lemma 5.45 Let a Ď Arx1 , . . . , xr s be an ideal. Then Arx1 , . . . , xr s{ab A B “

Brx1 , . . . , xr s{aBrx1 , . . . , xr s

Suppose that a is generated by a collection t f i u of polynomials. Then

aBrx1 , . . . , xr s will be generated by the images of f i ’s under the natural map
Arx1 , . . . , xr s Ñ Brx1 , . . . , xr s that is the polynomials obtained by applying the
structure map u to coefficients.
In several context, when e.g. u is a canonical inclusion; like Z Ď Q or
Q Ď C, the effect is nil, one just considers the generators as being members
of Brx1 , . . . , xr s. However, in other situations the effect on f i ’s can be dramatic,
in the worst case they can even vanish! For examples this occurs if the struc-
ture map is the reduction mod p-map u : Z Ñ Z{pZ “ F p and the coefficients
of f i P Zrxs are all divisible by p.
Example 5.6 A particular application of the previous lemma is that the
tensor product of two polynomial rings (over the base ring) again is a poly-
nomial ring. If case both rings are polynomial rings in one variable, one has
Arxsb A Arys “ Arx, ys, and in general it holds true that

Arx1 , . . . , xr sb A Ary1 , . . . , ys s “ Arx1 , . . . , xr , y1 , . . . , ys s.

Polynomials give a striking example that the decomposable tensors are

scarce. In Arx, ys for instance, the decomposable tensors are the polynomials
that factor as ppxqqpyq of which there are few. K
Example 5.7 Be aware that the tensor product of two integral domains need
not be an integral domain. Even if both factors are fields, the tensor prod-
uct might acquire zero-divisors. A simple example is CbR C. The complex
numbers C can be described as the quotient Rrxs{px2 ` 1q so by Lemma 5.45
above it holds that CbR C “ Crxs{px2 ` 1q. This latter ring is isomorphic to
the direct product C ˆ C, sending the residue class of a polynomial ppxq to the
pair pppiq, pp´iqq yields an isomorphism, and the direct product has many zero
divisors. K

Problems

5.15 Let k be a field and let f pxq be an irreducible polynomial in krxs so that ˇ
K “ krxs{p f pxqq is field. Moreover let L be field extension of k over which f
splits as a product f pxq “ f 1 pxq . . . f r pxq of irreducible polynomials. Use the
ś
Chinese remainder theorem to prove that Kbk L » Li where Li is the field
Li “ Lrxs{p f i pxqq.
 tensor products 121

5.16 Assume that K is an algebraic number field; i.e. a finite extension of the ˇ
field Q of rational numbers. Show that KbQ R is isomorphic to a product of
fields each being isomorphic either to R or C. Show that rK : Qs “ r1 ` 2r2
where r1 denotes the number of factors isomorphic to R and r2 the number of
factors isomorphic to C. Hint: By the Primitive Element Theorem one may
assume that K “ Qrxs{p f pxqq where f pxq P Qrxs is irreducible polynomial.

5.17 Show that A “ Rrx, ys{px2 ` y2 q is an integral domain, but that AbR C is ˇ
not.

5.18 Let φ : A Ñ B be a ring homomorphism and p Ď A a prime ideal. Show

that the fibre of φ˚ : Spec B Ñ Spec A over p is naturally homeomorphic to
Spec A{pb A B.

5.19 Let K be a field of positive characteristic p and let t P K be an element

which is not a p-th power. Show that L “ Krxs{px p ´ tq is a field and that LbK L
has non-trivial nilpotent elements.

5.20 Let k be a field of positive characteristic p, and let a be an ideal in A “

ř
krx1 , . . . , xr s generated by polynomials f i pxq “ α aiα x α . Let F : k Ñ k be
the Frobenius map a ÞÑ a p ; and let k F denote the field k endowed with the
k-algebra structure induced by the Frobenius map; that is, members a of k act
on k F as a ¨ x “ a p x, Show that pkrx1 , . . . , xr s{aqbk k F » krx1 , . . . , xr s{aF where aF
ř p
is the ideal generated by the polynomials f i “ α aiα x α

5.21 Let C s be C equipped with the alternative algebra structure induced by

complex conjugation; i.e. a ¨ z “ āz. Let f pxq P Crxs be a polynomial. Describe
Crxs{p f pxqqbC C.
s When are the C-algebras Crxs{p f pxqqbC C s and Crxs{p f pxqq
isomorphic?

M
Changes:
2018-09-23 Have reworked almost everything until section 5.4; not yet finished.
05/10/2018 Reworked the last part about tensor products of algebras.
07/10/2018 Added an exercise, corrected lots of misprints and minor errors.
08/10/2018 Have rewritten the proof of Proposition 5.23 about Adjointness on page 109 and
the subsequent example with substantial simplifications.
10/10/2018 Have added a prop about tensor prods and bases change; prop 5.35 on page 115.
11/10/2018 Added Exercise 5.6 on page 113
18/10/2018 Corrected misprints, minor changes in the text.
22/10/2018 Minor changes of the language.
Lecture 6

Localization

Preliminary version 2.0 as of 2018-10-22 at 09:43 (typeset 3rd December 2018 at 10:03am)—Prone
to misprints and errors and will change.
28/10/2018 Added new exercise 6.6

Very early in our mathematical carrier, if not in our lives, we were intro-
duced to fractions, and we should be familiar with their construction. Anyhow,
recall that to every pair of integers m and n with n ‰ 0, one forms the “frac-
tion” m{n with m being the enumerator and n the denominator. Two such
fractions m1 {n1 and m{n are considered equal—that is, have the same numeri-
cal value—precisely when nm1 “ n1 m. The fractions—or the rational numbers
as we call them—are entities per se and not only results of division. They obey
the rules for adding and multiplying we learned in school—that is, the field
axioms—and they thus form a field Q.
There is a simple and very general version of this construction. It gives us
the freedom to pass to rings were a priori specified elements become invertible.
Virtually any set of elements can be inverted; there is merely one natural
constraint. If s and t occur as denominators, their product st will as well;
indeed, one has s´1 t´1 “ pstq´1 . Hence the natural notion to work with is the
concept of multiplicatively closed sets.
The process is indeed very general. It even accepts zero-divisors as denom-
inators, but can then be "murderous". If a is a zero-divizor, say a ¨ b “ 0, and
a becomes inverted, b gets killed; indeed, it will follow that b “ a´1 ¨ a ¨ b “
a´1 ¨ 0 “ 0. In principle, one can even push this so far that 0 is inverted; but
this will be devastating and the resulting ring not very interesting, it will be
the null-ring.
There are several ways of defining the localized rings. We shall follow most
text books and mimic the way one constructed the rational numbers. This is a
direct and intuitive construction not requiring much machinery.
The name “localization” has its origin in geometry and rings of functions,
say on a topological space X. If U Ď X is an open set, every function whose
zeros all lie in the complement XzU of U, become invertible when restricted to
U; hence one obtains many functions on U by inverting certain functions on X.
In general far from all are shaped like that, but in special situations, important
124 ma4200—commutative algebra

in algebraic geometry, all algebraic functions on U arise in this way.

6.1 Localization of rings

We start out with introducing the notion of multiplicatively closed sets, and
proceed to the construction of the localized rings. They will be characterized
by a universal property. Core examples will be given, and the basic properties
of localized rings will be established, notably the relation between ideals in the
ring A and ideals in the localizations AS .
6.1 Rrecall that a subset S Ď A is multiplicatively closed, or for short a multiplica- Multiplicatively closed
tive set, if it contains the unit element and the product of any two elements sets (Multiplicativt
lukkede mengder)
from S belongs to S. That is, the following two conditions are satisfied

o 1 P S;

o If s, t P S, then st P S.

6.2 We mention two important examples. If a is any element in A the set

S “ t an | n P N0 u of all the powers of a is obviously multiplicatively closed.
Secondly, the complement S “ Azp of any prime ideal p in A is multiplicatively
closed as well; indeed, from st R S ensues that st P p and at least one of the
factors s or t belongs to p; that is, it does not lie in S.
6.3 It is fairly clear that the intersection of two multiplicatively closed sets is
multiplicatively closed, and one can speak about the the multiplicative set gener-
ated by a subset T of A. It equals the intersection of all multiplicatively closed
sets containing T, and one convinces oneself on the spot that the elements
are all finite products of elements from T. So for example, the multiplicative
set in Z generated by 2 and 5 consists of all numbers of the form 2a 5b , with
a, b P N0 .

The universal property

6.4 The localized ring AS comes together with a ring homomorphism ι S : A Ñ

AS such that all the images ι S psq of elements s from S are invertible, and
moreover, the map ι S is universal in this respect. This means the following: φ
/B
A O
o Any map of rings φ : A Ñ B sending members of S to invertible elements, ψ
#
ιS
factors through ι S in a unique manner; i.e. one has φ “ ψ ˝ ι S for an unam-
AS
biguously defined map of rings ψ : AS Ñ B.

In the sequel we shall mostly suppress the reference to the map ι S and just
write a for ι S paq. This calls however for a word of warning: The map ι S is not
always injective! Membres of A killed by elements from S become zero in AS ;
indeed, from as “ 0 follows ι S paqι S psq “ 0, and ι S psq being invertible forces
ι S paq “ 0.
 localization 125

As any other objects characterised by a universal property, the pair ι S and

AS is unique up to an unambiguous isomorphism; if ι1S : A Ñ A1S solves the
same universal problem, one has ι1S “ ψ ˝ ι S for a unique ψ : AS Ñ A1S . A
ιS ι1S

The construction of the localization AS  

AS / A1S
»
6.5 The construction of the localized ring AS follows grosso modo the same
lines as the construction of the rational numbers, but there is a necessary twist
to it due to the possible presence of zero divisors in S with the consequence
that the cancellation law does not necessarily hold.
6.6 A fraction has an enumerator and a denominator, and the latter is con-
fined to S. A natural starting point is therefore the product A ˆ S with the first
factor representing the enumerators and the second the denominators. The
next step is to introduce an equivalence relation on A ˆ S telling when two frac-
tions are to be considered equal, and inspired by the case of classical fractions,
we declare the pairs pa, sq and pb, tq to be equivalent when for some u P S, it
holds true that upat ´ bsq “ 0. In that case we temporarily write pa, sq „ pb, tq.
This relation is obviously reflexive and a symmetric. To see it is transitive,
assume given three pairs pa, sq, pb, tq and pc, uq such that pa, sq „ pb, tq and
pb, tq „ pc, uq; transcribing the equivalences into equalities in A, we find that
vpat ´ bsq “ 0 and wpbu ´ ctq “ 0 for some elements v, w P S. Since

tpau ´ csq “ upat ´ bsq ` spbu ´ ctq,

we infer that

vwtpau ´ csq “ wupvpat ´ bsqq ` svpwpbu ´ ctqq “ 0.

From S being multiplicatively closed it ensues that vwt P S, and so pa, sq „

pc, uq.
6.7 Now, we let AS be the set of equivalence classes A ˆ S{ „, and we denote
by a{s or as´1 the class of the pair pa, sq. The next task is to give a ring struc-
ture to AS , and there is no hocus-pocus about that; it is done by the familiar
formulas for adding and multiplying fractions:

a{s ` b{t “ pat ` bsq{st a{s ¨ b{t “ ab{st. (6.1)

However, some checking is necessary. First of all, the definitions are expressed
in terms of representatives of the equivalence classes, and it is paramount
they do not dependent on which representatives are used. Secondly, the
ring axioms must be verified; once the definitions are in place, this is just
straightforward high school algebra safely left to volunteering students.
6.8 As an illustration, let us check that the sum is well defined. Notice that
it suffices to vary the representatives of one of the addends at the time; so
assume that pa, sq „ pa1 , s1 q; i.e. upas1 ´ a1 sq “ 0 for some u P S. We find

s1 tpat ` bsq ´ stpa1 t ` bs1 q “ t2 ps1 a ´ a1 sq

126 ma4200—commutative algebra

which is killed by u. Therefore the sum does not depend on the representative
of the first addend used, and by symmetry, neither on the representative of the
second, and the sum is well defined.
Problem 6.1 Show that the product is well defined. On a rainy day when all
your friends are away, verify the ring axioms for AS . M
6.9 The localisation map ι S is nothing but the canonical map ι S : A Ñ AS that
sends an element a in A to the class of the pair pa, 1q; that is, a is mapped to
the fraction a{1. By the very definition in (6.1) of the sum and the product
in the localization AS this a ring homomorphism. (Seemingly, this map does
nothing to a, but kill it if necessary).
6.10 Our next proposition gives a more practical description of the ring AS :
Proposition 6.11 All elements in AS are of the form a{s. It holds true that
ι S paq “ 0 if and only if a is killed by some element from S; i.e. if and only if there
is an s P S such that sa “ 0.

Proof: By definition every element in AS is an equivalence class a{s. The

zero element in AS is represented by the pair p0, 1q and ι S paq by the pair pa, 1q.
Hence ι S paq “ 0 if and only if s ¨ pa ¨ 1 ´ 0 ¨ 1q “ 0 for an s P S; that is, if and only
if s ¨ a “ 0 for an s P S. o
6.12 The map ι S will often be injective; for instance, when A is a domain this
is always the case. This allows us to identify A with its image in AS which
significantly simplifies the notation. We can safely write a instead of ι S paq, and
1
the inverse image ι´S b of an ideal (or any subset for that matter) will simply
be the intersection A X b.

The universal property

We shall show that the pair AS and ι S solves the universal problem formulated
at the top of paragraph 6.4.
6.13 So assume we are given another ring B and a map of rings φ : A Ñ B
such that φpsq is invertible for all s P S. The sole way to realize a ring map
ψ : AS Ñ B extending φ is to put

ψpa{sq “ φpaq ¨ φpsq´1 ,

and the salient point is that this is a legitimate definition, i.e. φpaq ¨ φpsq´1 is
independent on the chosen representative pa, sq. But from a{s “ a1 {s1 ensues
that t ¨ pas1 ´ sa1 q “ 0 for an element t P S, and hence, since φ is map of rings,
that
` ˘
φptq ¨ φpaq ¨ φps1 q ´ φpa1 q ¨ φpsq “ 0.
The element φptq is invertible by assumption, and we conclude that φpaq ¨
φpsq´1 “ φpa1 q ¨ φps1 q´1 .
This proves most of the following proposition:
 localization 127

Proposition 6.14 Let A be a ring and S Ď A a multiplicative subset. Then there

is a ring AS and a ring homomorphism ι S : A Ñ AS solving the universal problem in φ
/B
paragraph 6.4. The pair is unique up to an unambiguous isomorphism.
A O
ψ
#
ιS
Proof: What remains to be seen is that ψ is a ring homomorphism, but this AS
follows directly from the rules in (6.1) once ψ is well defined. o

Examples

6.1 We have not excluded that 0 lies in S. In this case however, the localized
ring will be the null-ring since 0 becomes invertible. This situation occurs e.g.
when S has nilpotent members.

6.2 An simple situation to have in mind is when A is contained in a field K.

The localized ring AS is then just the subring Ars´1 |s P Ss of K generated by
the inverses of members of S. The elements of this ring are all shaped like
1
as´1 ; indeed, every sum i ai s´
ř
i can be rendered on this form with s the com-
mon denominator of the terms. The universal property of the localized ring
AS then immediately gives a map of rings AS Ñ Ars´1 |s P Ss which one easily
checks is an isomorphism using the description of AS in Proposition 6.11 on
page 126.

6.3 (The field of fraction of a domain) Every domain is contained in a smallest

field KpAq called the field of fractions of A. The set S of non-zero elements in The field of fractions of a
A is multiplicativily closed when A is a domain (by definition, a domain has domain (Kvotientkroppen
til et område)
no zero-divisors), and KpAq is the localization AS . Every non-zero element
becomes invertible in KpAq, hence KpAq is a field. The elements are of the
form ab´1 with b ‰ 0, and since zero divisors are absent, it holds true that
a{b “ a1 {b1 if and only if ab1 “ a1 b. In the sequel we shall tacitly identify a and
a{1, and A will be considered a subring of KpAq.
The field KpAq is the smallest field containing A in the sense that if A Ď L
with L a field, there is a canonical copy of KpAq lying between A and L; i.e.
one has A Ď KpAq Ď L.
Examples of fraction fields are the field Q of rational numbers and the
field Cpx1 , . . . , xr q of rational functions in the variable s x1 , . . . , xr , which is the
fraction field of the polynomial ring Crx1 , . . . , xr s.

6.4 The ring of integers Z within the field rationals Q is a particular instance
of the situation in the previous example. When S is the set of all powers of a
given number p, that is, S “ tpn u, the ring ZS “ Zr1{ps “ t a{pn | a P Z, n P
N0 u will be the ring of rational numbers whose denominators are powers of p.
(see also Example 1.12 on page 12).
In a similar vein, when p is a prime and S is the complement of the prin-
cipal ideal pZ, the localization ZS will be the ring Zp pq “ t a{b | a, b P
128 ma4200—commutative algebra

Z, pp, bq “ 1 u of rational numbers whose denominator is prime to p. We have

already met these ring in Example 2.19 on page 45.
6.5 Consider the polynomial ring krx, y, zs in three variable over the field k. We
shall describe the localisation of krx, y, zs in the prime ideaal pzq and show that
krx, y, zspzq “ kpx, yqrzspzq ; that is, the polynomial ring over the fraction ield
kpx, yq localised at the prime pzq.
In process we introduce the subset of krx, ys whose members are the poly-
nomials involving only the variables x and y. It is obviously multiplicatively
closed, and as obvious it holds that krx, y, zsS “ kpx, yqrzs. Localising both
rings in the ideals generated by z we obtain the desired equality, noticing
that S Ď krx, yszpzq, so elements in S are already invertible in krx, y, zspzq and
pkrx, y, zsS qpzq “ krx, y, zspzq .

Problems

6.2 Show that Zr1{10s “ Zr1{2, 1{5s. Generalize.

6.3 Prove that any intermediate ring Z Ď A Ď Q is a localization of Z in a ˇ

multiplicative set S.

6.4 Prove that the group of units A˚ in A is a multiplicative set. Show that the
localization maps ι S is an isomorphism if and only if S is a subset of A˚ .

6.5 When L is a set of primes in Z (finite or infinite), let Zp Lq denote the ˇ

localization in all primes outside L; that is, we put Zp Lq “ pP L Zp pq . Show
Ş

that there is a natural isomorphism Zp Lq bZ Zp L1 q » Zp LX L1 q .

6.6 Consider be the polynomial ring A “ krx1 , . . . , xn s over the field k. Let
S be the set polynomials in A that depend only on the first r variables; i.e.
those on the form ppx1 , . . . , xr q. Show that S is multiplicatively closed and that
AS “ Krxr`1 , . . . , xn s where K is the field kpx1 , . . . , xr q of rational functions.

Ideals and localization

6.15 Extension of ideals from A to AS yields a map from the set I pAq of ideals
in A to the set of ideals I pAS q in the fraction ring AS . The map operates by
sending an ideal a to the ideal aAS generated by the image ι S paq, and one
verifies that members of aAS are all shaped like as´1 for some a P a and some
s P S. The map is inclusion preserving and preserves products and sums of
ideals as extension maps always do (see Paragraph 2.9 on page 26).
 localization 129

6.16 In general the extension map from I pAq to I pAS q is not injective. For
instance, it may happen that S X a ‰ H, in which case the extension aAS will
contain an element invertible in AS and consequently be equal to AS ; and
of course, this may be the case for several different ideals. In quite another
corner, ideals a contained in the kernel of ι S reduce to the zero ideal in AS . So,
some ideals are blown up to AS (those meeting S) and some collapsed to zero
(those contained in ker ι S ).
Example 6.6 A simple instance of the extension map not being injective is
the case when A “ Z and S “ t pn | n P Z u for some prime p. All the
ideals a “ ppm q extend to the entire ring ZS . This also illustrates that forming
extensions does not commute with forming infinite intersections; indeed, one
has m ppm q “ 0 whereas m pm ZS “ ZS .
Ş Ş
K
6.17 The extension map is however surjective. Any ideal b Ď AS equals
1
ι´
S pbqAS ; that is, when pulling an ideal back to A and subsequently extending
the result, one recovers the original ideal. To see this, notice that if b “ a{s
1
belongs to b, the element a belongs to ι´
S pbq as ι S paq “ b ¨ s, and therefore b lies
1
in the extension ι´ S pbqAS .

Proposition 6.18 (Ideals in localizations) The extension map from I pAq

to I pAS q given by a Ñ aAS is surjective. It preserves inclusions, products, sums and
1
finite intersections. One has ι´
S paAS q “ t a P A | sa P a for some s P S u, and for
1
ideals b Ď AS it holds true that b “ ι´
S pbqAS .

Proof: We have already proved most of the proposition, only the assertions
about sums, products and intersections remain unproven. It is a general
feature of extension of ideals that products and sums are preserved, and we
concentrate on the finite intersections; and of course, the case of two ideals
will suffice.
Clearly pa X a1 qAS Ď aAS X a1 AS . So assume that b P aAS X a1 AS . One may
then express b as b “ a{s “ a1 {s1 with elements a and a1 from respectively a
and a1 . This yields tpa ¨ s1 ´ a1 ¨ sq “ 0 for some t P S. But then tsa1 “ ts1 a P a X a1
and consequently b “ tsa1 {tss1 lies in pa X a1 qAS . o
6.19 Prime ideals behave more lucidly under localization than general ideals.
Either they blow up and become equal to the entire localized ring AS , or they
persist being prime. Moreover, every prime ideal in AS is of the shape pAS for
an unambiguous prime ideal p of A. One has:

Proposition 6.20 (Prime ideals in localizations) Assume that p is a

prime ideal in the ring A and S is a multiplicative subset of A. Extending p to the
localization AS has two possible outcomes. Either pAS “ AS , and this occurs if and
1
only if p X S ‰ H, or otherwise pAS is a prime ideal and ι´S ppAq “ p.

Proof: If S X p ‰ H the ideal p blows up to the entire ring in AS ; that is,

it holds that pAS “ AS . If not, pAS is a prime ideal; indeed, suppose that
130 ma4200—commutative algebra

bb1 P pAS and that b “ a{s and b1 “ a1 {s1 with a, a1 P A and s, s1 P S. We infer
that tss1 bb1 “ taa1 P p for some t P S, and hence either a or a1 lies in p since t
does not. Moreover, if ι S paq “ a1 s´1 for some a1 P p, if follows that sta “ ta1 P p,
hence a P p since p is a prime ideal. o

Proposition 6.21 The prime ideals of AS are precisely the ideals of the form pAS
for p a prime ideal in A not meeting S. The prime ideal p is uniquely defined.

Proof: By the previous proposition, the ideals pAS are all prime, so let q
1
be a prime ideal in AS . Then p “ ι´S q is prime, and by the last sentence in
proposition 6.18 above it holds that q “ pAS . o
6.22 The localization processe commutes with the formation of radicals:

Proposition 6.23 Let a be an ideal in A and S Ď a multiplicative set. The it holds

true that
? ?
aS “ p aqS .
?
Proof: If xs´1 P p aqS with x n P a and s P S, it holds that pxs´1 qn “ x n s´n P
? ?
aS , and so xs´1 lies in aS . If xs´1 P aS , it holds that pxs´1 qn “ at´1 with
?
t P S and a P a. Hence pxtqn “ sn atn´1 P a, and consequently x P p aqS . o

Problems

6.7 Show that if aAS “ AS , then the same holds for all powers am of a.

6.8 Let p and q be different prime numbers and let S be the multiplicative set ˇ
S “ t pn | n P N0 u. Describe Z X ppqqZS .

6.9 Let S be multiplicatively closed in the ring A and let a Ď A be an ideal. ˇ

Show that the ideals pa : sq when s runs through S form a directed family of
1
ideals; hence their union is an ideal. Show that sPS pa : sq “ ι´
Ť
S paAS q.

6.10 Suppose that p is a prime ideal and that a an ideal contained in p. Show
1
that ι´
S paAs q Ď p.

6.11 Let j : Spec AS Ñ Spec A be the map induced from the localization map
ι S : A Ñ AS . Show that j is a homeomorphism onto its image (when the image
is endowed with induced topology). Show that the image jpSpec Aq equals the
intersection of all the open sets containing it.

6.12 Show by an example that j is not necessarily an open embedding. Hint:

Let e.g. A “ Z and S the multiplicative subset generated by every second
prime.

6.13 Given an example of a ring A and a non-zero prime ideal p such that
Ap “ A{p. Hint: Let A be the product of two fields.
 localization 131

6.14 Let a and b be two ideals in A, Prove that pa : bqS “ paS : bS q. ˇ

Functoriality

6.24 The ring AS depends of course on both A and S, so functoriality is

naturally formulated in terms of the pair pA, Sq. Given another pair T and B
and a map of rings φ : A Ñ B taking elements of S into T, there is an induced A
φ
/B
map of rings φS,T : AS Ñ BT satisfying φS,T ˝ ι S “ ι T . Since φ takes S into T,
ιS ιT
the elements φpsq become invertible in BT and the universal property of AS  
guarantees that ι T ˝ φ extends to a map AS Ñ A T . This map simply sends a{s to AS / BT
φS,T
φpaq{φpsq.
6.25 A particular case to notice is when A “ B and φ “ id A . If S Ď T, there is a
canonical map AS Ñ A T which just interprets fractions a{s in AS as a fractions
in A T . This map might appear very much like doing nothing; but be aware, it
can have a non-trivial kernel. When some member of T kills elements in A not
killed by anyone in S there will non-zero members of the kernel.
However, when S contains no zero-divisors, the localization map A Ñ AS
is injective and we shall identify A as a subring of AS . In particular, when A
is an integral domain all the localization maps are injective, and we may—and
tacitly will do—identify the localized rings as subrings of the field of fractions
KpAq. After this identification, the ring AS is a subring of A T , when S Ď T.

The local ring at a prime ideal.

A few ways of forming rings of fraction are omnipresent in algebra and alge-
braic geometry and are used over and over again. The most prominent one is
may be the localization Ap at a prime ideal p.
6.26 The complement S “ Azp of a prime ideal p is multiplicatively closed,
and the corresponding localized ring is written as Ap . The elements are
fractions a{b with b R p. It is a local ring whose only maximal ideal is pAp :

Proposition 6.27 The localisation Ap is a local ring with maximal ideal pAp . The
assignment q ÞÑ qAp is a one-to-one correspondence bewteen prime ideals in Ap and
prime ideals q in A contained in p.

Proof: This is nothing but Proposition 6.21 above on page 6.21, according
to which the prime ideals in Ap are precisely the ideals in Ap of the form
qAp where q is a prime ideal in A contained in p, and it holds true that q “
ι´1 pqAp q. o
Notice that the kernel of the localization map ι : A Ñ Ap is contained in any
prime ideal q Ď p; indeed, if sa “ 0 with s R p, a fortiori s R q and hence a P q.
132 ma4200—commutative algebra

The residue field Ap {pAp is frequently written kppq. Since ι´1 ppAp q “ p
it holds true that A{p maps invectively into Ap {pAp . And in fact, Ap {pAp
equals the fraction field KpA{pq of A{p.
Example 6.7 If p is a prime number, the localized ring Zp pq at the maximal
ideal generated by p consists of rational numbers whose reduced form is a{b
where b is relatively prime to p. The maximal ideal is generated by p and the
residue field is the field F p with p elements. K
6.28 Clearly an arbitrary intersection of multiplicatively closed sets is mul-
tiplicatively closed, therefore complements of unions of prime ideals are
multiplicatively closed. If the prime ideals involved are tpi uiP I ; and hence
Ť
that S “ Az iP I pi , the maximal ideals of AS will be the extension pi AS . In
particular, if the pi ’s are merely finite in numbers, AS will be a semi-local ring.

Inverting powers of a single element.

Given an element f P A. The set S “ t f n | n P N0 u of all powers of f is
obviously multiplicatively closed, and the corresponding ring fractions AS is
denoted A f . The prime ideals in A f are exactly those on the form pA f for p a
prime ideal in A with f R p; that is, for the members of the distinguished open
subset Dp f q of Spec A.
There is a natural isomorphism between Arxs{px f ´ 1q and A f that sends
x to f ´1 . By the universal mapping property of the polynomial ring the map
is well defined, and f being invertibel in Arxs{px f ´ 1q, the unversal property
of A f furnishes an inverse. This makes the notation Ar f ´1 s for A f legitimate;
the usage is however poisonous when f is a zero-divisor. Adding f ´1 kills,
and in case f is nilpotent, the intoxication is lethal; everything is killed and
Ar f ´1 s “ 0.
Example 6.8 It is worthwhile mentioning a concrete example. Consider the
ring A “ Crzs of complex polynomials in the variable z and let f “ z ´ a. The
localized ring A f consists of those rational functions that are regular away
from a; that is, they have at most a pole at a. This generalizes to the ring O pΩq
of functions holomorphic in any domain Ω of the complex plane containing a.
The localized ring O pΩqz´a has the functions meromorphic in Ω with at most
a pole at a as elements. K

The total ring of fractions

The set S of non-zero divisors in the ring A is closed under multiplication;
indeed, if s and t are non-zero divisors and sta “ 0 with a ‰ 0 it would follow
that ta ‰ 0 contradicting that s is a non-zero divisor. The corresponding
ring of fractions is denoted by KpAq and called the total ring of fractions of A. The total ring of fractions
When A is an integral domain, S “ Azt0u, and all non-zero elements become (Kvotientringen)

invertible in KpAq. Consequently KpAq is a field; it is called the field of fractions The field of fractions
(Kvotientkroppen)
 localization 133

of A, which we met already in Example 6.3 on page 6.3. The ring KpAq is in
general not a field, but by definition has the property that all non-zero divisors
are invertible.
In any case, the canonical map A Ñ KpAq is injective; indeed, if s is a
non-zero divisor, by definition sa “ 0 implies that a “ 0.

Proposition 6.29 The total ring of fractions KpAq of a ring A has the property
that every non-zero divisor is invertible. The natural map A to KpAq is injective.
Moreover, KpAq is a field if and only if A is an integral domain.

6.30 We give two examples. The first is a simple but typical situation in
algebraic geometry. The ring A is the ring of polynomial functions on a variety
X with two components; in our specific example the two components are the
coordinate axes in the plane. The total ring of fractions of A turns out to be
the product of two fields, the elements are pairs of rational functions, one on
each of the axes. Contrary to what is requested of polynomial functions on X,
there is no continuity condition at the origin to be fulfilled; the two rational
functions may take different values, or for that matter, may have poles at the
origin.
This behavior is fairly general; for any ring A without nilpotent elements
and which merely has finitely many minimal prime ideals (that is, there is a
representation p0q “ p1 X . . . X pr of the zero ideal as a finite intersection of
prime ideals) it holds that KpAq is a product of fields. The only moral of the
second example, is that if A has nilpotents, the ring of fractions KpAq can be
rather involved; and far from being a product of fields.
Example 6.9 Let A “ krX, Ys{pXYq and let x and y be the classes of X and
Y respectively. Since pXYq X krXs “ p0q it holds that krXs » krxs and ditto
krYs » krys, thus we may talk about the rational function fields kpXq and KpYq
as kpxq and kpyq.
We shall see the total quotient ring KpAq is isomorphic to the product
kpxq ˆ kpyq of the rational functions fields in kpxq and kpyq. In geometric terms,
A is the ring of polynomial functions on the union Z “ Vpxq Y Vpyq of the
y-axis and the x-axis in the plane, and the elements of KpAq are just a pair of a
rational function, one on the x-axis and one on the y-axis.
All cross-terms are killed in A, and the elements are all shaped like a `
ppxq ` qpyq where ppxq and qpyq are polynomials vanishing at zero and a is
scalar in k. The zero-divisors in A is the union1 pxq Y pyq, so that the non-zero- 1
If Ppx, yq is a zero di-
divisors of A are of the form a ` ppxq ` qpyq with either a ‰ 0 or neither ppxq visor, there is a relation
like PpX, YqQpY, Yq “
nor qpyq identically equal to zero. ApX, YqXY, so either X
There is a map of rings or Y divides PpX, Yq

A Ñ kpxq ˆ kpyq

sending the class of a polynomial a ` ppxq ` qpyq to the pair pa ` ppxq, a ` ppyqq,
134 ma4200—commutative algebra

these are invertibel in kpxq ˆ kpyq and the there is induced a ring map

KpAq Ñ kpxq ˆ kpyq,

which clearly is injective; but it is slightly more subtle that it is surjective. Let
ξ “ pQpxq{Rpxq, Spyq{Tpyqq be an element in kpxq ˆ kpyq; then Rpxq and Tpyq are
non-vanishing polynomial. Then

ξ “ pQpxqx{Rpxqx, Spyqy{Tpyqyq,

and xRpxq ` yTpyq is a non-zero divisor in A mapping to pxRpxq, yTpyqq K

Problem 6.15 Why is it necessary to introduce the seemingly pointless

factors x and y in the fractions? M

Example 6.10 Let B “ krX, Ys{pX 2 , XYq and let as usual x and y be the classes
of X and Y in B. Then m “ px, yq is a maximal ideal (one has B{m “ k). Let
A “ Bm be the ring B localized at m. As in every local ring the units of A are
precisely the members not lying in the maximal ideal m. We contend that all
elements in m are zero-divisors, and hence A is its own total ring of fraction.
Indeed, x kills both x and y, hence the whole ideal m “ px, yq.
It worth while noticing that pxq is a prime ideal, since it holds true that
B{pxq “ krX, Ys{pX 2 , XY, Xq “ krYs is an integral domain, and pxq is the only
a
other prime ideal in B and is equal to the radical p0q of B; indeed, if p is
prime one has x2 P p, hence x P p.
Thus B has two prime ideals, the maximal ideal px, yq whose elements
constitute all zero-divisors and a sole minimal ideal pxq whose elements are all
the nilpotents of B. K

Problem 6.16 Show that Apxq equals the rational function field kpYq. M

Problems

6.17 Let n be a natural number. Determine the total quotient ring of Z{nZ. ˇ

6.18 Let A be any ring. Show that the nil-radical of KpAq is equal to the ˇ
extension of the nil-radical of A.

6.19 Let A be a ring. ˇ

a) Show that if the elements of A are either zero divisors or invertible, then
A “ KpAq.
b) If A has only one prime ideal, prove that KpAq “ A.
c) Let A be a direct product (of any cardinality) of any number of rings each
having only one prime ideals. Prove that KpAq “ A.
 localization 135

6.20 Let k be a field and consider the polynomial ring A “ krx1 , . . . , xn s. ˇ

Let r ă n be a natural number. Let S be the subset of A of polynomials
in the variable xr`1 , . . . , xn . Show that S is multiplicatively closed and that
krx1 , . . . , xn sS “ Krx1 , . . . , xr s where K “ kpxr`1 , . . . , xn q is the field of rational
functions in the variables xr`1 , . . . , xn .

6.21 Let A be a domain with quotient field K. Denote by S the multiplicative

set Az0 of non-zero elements in A. Show that ArTsS “ KrTs.

A last example

6.31 There is multiplicatively closed set associated with any ideal a in A which
is not that frequently met. It equals the set S “ 1 ` a; that is, the set consisting
of elements shaped like 1 ` a with a P a. one has the subset S of elements of
the form 1 ` a with a P a which is multiplicatively closed. In this case the the
maximal ideals in AS are those of the form mAS ; hence aAS is contained in the
Jacobson radical of AS .

6.2 Localization of modules

There is also a procedure to localize any A-module M in S closely resembling
the way the fraction ring AS was constructed. The localized module will be
denoted by MS . The construction of MS is functorial in M and gives a functor
Mod A Ñ Mod AS with the important property of being additive and exact.
6.32 To construction the localized module MS we mimick the way AS was
fabricated. Details will be skipped; they may mutatis mutandis be verified as in
the case of ring.
To begin with one introduces an equivalence relation on the Cartesian
product M ˆ A by declaring two pairs pm, sq and pm1 , s1 q to be equivalent if

tpms1 ´ m1 sq “ 0 (6.2)

for some t P S. The equivalence class of pm, sq will be designated by either

m{s or ms´1 . The module structure on MS consists of an additive group
structure and an action of AS . The additive structure is defined in analogy
with the usual way of adding fraction, namely as m{s ` n{t “ pmt ` snq{st.
And the action of an element a{s P AS is given in the staightforward way:
a{s ¨ m{t “ am{st. There is natural map ι S : M Ñ MS sending m to the class of
pm, 1q, and as in the case of rings, we shall often merely write m for the image.
Naturally, there is a lot of checking to be done. Every single step is straight-
forward, and we leave these soporific verifications to the students for a rainy
day. Summing up, one has:
136 ma4200—commutative algebra

Lemma 6.33 Every element of MS is of the form m{s. The map ιS : M Ñ MS is

A-linear, and its kernel consists of the elements m in M killed by some member of S;
that is, ker ι “ t m P M | tm “ 0 for some t P S u.

6.34 The same applies to the maps ι S when it comes to modules as with rings.
To simplify the notion, one soon drops the reference to the map and writes x
or x{1 for ι S pxq, but with some cautiousness since the image very well can be
zero.
When the module M is finitely generated, say by members m1 , . . . , mr , the
images ι S pmi q of the mi ’s will obviously generate MS . Indeed, pick a member
xs´1 from MS and write x “ ai mi then of course xs´1 “ ai s´1 mi .
ř ř

Functoriality

6.35 Given two A-modules M and N and an A-linear map φ : M Ñ N. Send-

ing m{s to φpmq{s gives an AS -linear map between the localized modules MS
and NS ; that is, a map φS : MS Ñ NS .
A formal definition starts with the map pm, sq Ñ pφpmq, sq between the
Cartesian product, and the salient point is that this respects the equivalence
relations as in (6.2). Indeed, a relation like tpms1 ´ m1 sq “ 0 leads to the relation
tpφpmqs1 ´ φpm1 qsq “ 0 because φ is A-linear. Therefore there is induced a map
between the equivalence classes.
6.36 From the definition of φS we infer immediately that a linear combination
of maps between N and M localizes to the corresponding linear combination;
that is, one has
paφ ` bψqS “ aφS ` bψS ,

where φ and ψ are A-linear maps from M to N and a and b ring elements.
And it is equally clear that the association is functorial; it holds true that

pψ ˝ φqS “ ψS ˝ φS

whenever φ and ψ are composable since it already holds at the level of the
Cartesian products—and of course, pid M qS “ id MS .

Proposition 6.37 Let A be a ring and S a multiplicative subset of A. The localiza-

tion functor Mod A Ñ Mod AS is additive and exact.

Proof: The only subtle point is that the functor is exact. In other words that it
brings an exact sequence

/M /L
ψ φ
N (6.3)

to an exact sequence. So our task is to verify that the sequence

 localization 137

/ MS / LS
ψS φS
NS

is exact, which amounts to checking that ker φS “ im ψS . To that end, pick an

element m{s in the kernel of φS . This means that φpmq{s “ 0, hence tφpmq “ 0
for some t P S. But then tm P ker φ, and since the sequence (6.3) is exact, there
is an element n in N such that ψpnq “ m. But then we have ψS pn{sq “ ψpnq{s “
m{s, and we are through. o

Submodules
Given a submodule N Ď M. The localized module NS can be considered to be
submodule of MS . The inclusion map localizes to an injection whose image
consists of elements shaped like fractions ns´1 with n P N and s P S, and thus
it can naturally be identified with NS .
6.38 Localization behaves nicely with respect to sums and finite intersections
of submodules; the following two assertions hold:
ř ř
o p i Ni qS “ i pNi qS

o pN X N 1 qS “ NS X NS1

where N, N 1 and the Ni ’s are submodules of N. However, localization does

not commute with infinite intersections as we saw in example 6.6 on page 129.
In particular, formation of arbitrary direct sums commute with localizations.
À À
o p iP I Mi q S » iP I pMi qS

To establish the first equality observe that it ensues from the inclusion
ř ř ř ř
Ni Ď i Ni that pNi qS Ď p i Ni qS , hence that i pNi qS Ď p i Ni qS . Any element
in p i Ni qS is of the form p i xi qs´1 with merely finitely many of the xi ’s being
ř ř
ř
non-zero, which obviously lies in i pNi qS .
The second follows because y “ n{s “ n1 {s1 means ts1 n “ tsn1 for some t P S,
and putting x “ ts1 n we then infer that x P N X N 1 and y “ x{tss1 .

Problem 6.22 Locization does not commute with infinite direct products. Let
p P Z be a number and denote by S the multiplicative set S “ tpn u in Z. Show
that there is a natural inclusion
ź ź
p ZqS Ď ZS ,
iPN i PN

but that the inclusion is strict. M

Relation with the tensor product

138 ma4200—commutative algebra

6.39 The action of AS on M is expressed by the bilinear map M ˆ AS Ñ MS

that sends pm, as´1 q to am ¨ s´1 , and in view of the univeral property enjoyed
by the tensor product, it induces an A-linear map Ψ : Mb A AS Ñ MS which on
decomposable tensors acts by sending mbas´1 to ams´1 . This map turns out
to be an isomorphism:

Proposition 6.40 The map Ψ defined above is an isomorphism MS » Mb A AS .

Proof: The crux of the poof is that all the elements in Mb A AS are decompos-
able; that is, they are of the form mbs´1 with m P M and s P S. Granted this,
if mbs´1 is mapped to zero; that is, if ms´1 “ 0, the element m is annihilated
by some t from S. But then mbs´1 “ tmbs´1 t´1 “ 0, so that the map Ψ is
injective, and Ψ is obviously surjective as well.
1
A priori an element from Mb AS is of the shape i mi bai s´
ř
i with ai P A
and si P S. Moving the ai through the tensor product, we we may bring it on
1 ´1
the form i mi bs´
ř
i . The trick is now to let s “ s1 ¨ ¨ ¨ sr and ti “ ssi , and we
arrive at ÿ ÿ
x“ mi ti bs´1 “ p mi ti qbs´1 “ mbs´1
i i
ř
with m “ i mi ti . o
6.41 We saw that base change preserves tensor products (Proposition 5.35
on page 115) and combining that with Proposition 6.40 above, we see that
localization process preserves tensor product as well:

Proposition 6.42 Let M and N be two A-modules and S a multiplicative set in A.

Then there is a canonical isomorphism

pMb A NqS » MS b AS NS .

6.43 When it comes to hom-sets, the behaviour is rather nice, at least for
modules of finite presentation. In general, sending an A-linear map φ be-
tween two A-modules M and N to the localized map φS is an A-linear map
Hom A pM, Nq Ñ Hom AS pMS , NS q. By the universal property of localization it
extends to a map Hom A pM, NqS Ñ Hom AS pMS , NS q; and in case M is of finite
presentation, this map is an isomorphism:

Proposition 6.44 Let M and N be two A modules and S a multiplicative set in A.

Assume that M is of finite presentation. Then the canonical map
»
Hom A pM, NqS ÝÑ Hom AS pMS , NS q

induced by sending φ to φS is an isomorphism.

Proof: Recall that both localization and the hom-functors are additive func-
tors, hence the proposition holds true whenever M is a free A module of finite
rank n; indeed, one finds

Hom A pnA, NqS » pnNqS » npNS q » Hom AS pnAS , NS q

 localization 139

where the isomorphisms are the natural ones (the one in the middle is an
isomorphism since localization is additive, and the two others are because
hom-functors are additive). Since M is assumed to be of finite presentation, it
lives in an exact sequence

/ nA /M / 0,
φ π
mA (6.4)

with m, n P N and where φ and π are A-linear maps. Consider the diagram

0 / Hom A pM, NqS / Hom A pnA, NqS / Hom A pmA, NqS

  
0 / Hom A pMS , NS q / Hom A pnAS , NS q / Hom A pmAS , NS q
S S S

The upper sequence is obtain from (6.4) by applying Hom A p´, NqS to it and is
therefore exact by left exactness of hom-functors. The bottom sequence is the
localization in S of the upper one, and since localization is an exact functor, it
is exact. The vertical maps are the canonical maps induced by sending φ to φs ,
and it is a matter of simple verification to check that the squares commute.
Now, the final point is that the two rightmost maps are isomorphisms by
the beginning of the proof, and then the Five Lemma tells us that the third
map is an isomorphisms as well, which is precisely what we aim at proving! o

6.3 Nakayama’s lemma

Nakayama’s lemma is a workhorse in commutative algebra and is applied over
and over again. As often is the case with popular courses, it comes in quit a
lot of different flavours. One way of viewing this famous result—which we
shall adopt as our point of departure—is as an extension to finitely generated
modules of the fundamental existence result for maximal ideals in rings
(Theorem 2.50 on page 40): Maximal proper submodules of finitely generated
modules always exist; or what amounts to the same in view of the character of
simple modules (Proposition 4.35 on page 78), simple quotients exist.
This is may be an unorthodox place to treat Nakayama’s lemma, but in fact
Nakayama’s lemma is about whether modules are zero or not, and we find its
appropriate placement to be in together with the treatment of the support of
modules.
6.45 The notion of the Jacobson radical will frequently appear in the hypothe-
ses of subsequent results, so for the convenience of the students we recall the
definition. The Jacobson radical JpAq of a ring A is the intersection of all max-
Ş
imal ideals in A; that is, JpAq “ m Ď A m where m runs through the maximal
ideals in A. The elements in JpAq are precisely those a P A so that 1 ` ax is
invertible for all x P A (Proposition 2.67 on page 45).
140 ma4200—commutative algebra

Nakayama’s lemma and simple quotients

6.46 It is not true that every A-module has a simple quotient, an elementary
example being the Z-module Q; every ideal a in Z satisfies aQ “ Q and hence
Q has no simple quotient. However finitely generated modules always have,
and this is our first version of Nakayama’s lemma:

Proposition 6.47 (Nakayama’s lemma I) Let M be a non-trivial finitely

generated A-module. Then M has a non-zero simple quotient. In other words, there
exists maximal ideal m with an A-linear surjection M Ñ A{m; or equivalently, mM
is a proper submodule of M.

Proof: If M is cyclic, it is of the form A{a for some proper ideal and has A{m
as a quotient for any maximal ideal m containing a. Assume next that M is not
cyclic, and let n be the the least number such that M can be generated by n
elements. Then n ě 2. Let x1 , . . . , xn generate M. The submodule N generated
by x2 , . . . , xn is a proper submodule and M{N is cyclic (generated by the class
of x1 ), and has a simple quotient by the first part of the proof. o

6.48 An alternative proof of Nakayama’s lemma (tailored to the same pattern

as the proof of the Basic Existence Theorem for ideals ( Theorem 2.50 on
page 40)) is to apply Zorn’s to establish that any finitely generated A-module
M has a maximal proper submodule. If N is one, M{N will be a nontrivial
simple quotient. To checks that chains have upper bounds, observe that the
union of any chain tMi u of proper submodules of M will be proper; were it
not, the elements of a finite generating set would eventually all lie in one of
the submodules from the chain, and that submodules would not be proper.
In the case of M being Noetherian, any collection of submodule contains
maximal members, and existence of maximal proper submodules, comes for
free!

Nakayama classic

6.49 To assure anyone (hopefully there are none) that finds our approach a
blasphemous assault on their most cherished tradition, we surely shall include
Nakayama classic; here it comes:

Proposition 6.50 (Nakayama classic) Let a be an ideal in A contained in

the Jacobson radical of A. Let M be a finitely generated A-module and assume that
aM “ M. Then M “ 0.

Proof: Recall that the Jacobson radical of A equals the intersection of all the
maximal ideals in A. Assume M ‰ 0. By Nakayama I (Proposition 6.47 above)
there is a maximal ideal m such that mM is a proper submodule, which is
impossible since a Ď m and aM “ M by assumption. o
 localization 141

With the isomorphism Mb A A{a » M{aM in mind, one may rephrase

Nakayama’s lemma as follows.

Proposition 6.51 (Nakayama’s lemma III) Let M be a finitely generated

A-module and a an ideal contained in the Jacobson-radical of A. If Mb A A{a “ 0, it
holds true that M “ 0.

6.52 The by far most common situation when Nakayama’s lemma is applied,
is when A is a local ring and a “ m is the maximal ideal, and this situation
merits to be mentioned specially:

Proposition 6.53 (Nakayama’s lemma for local rings) Let A be a local

ring with maximal ideal m. Assume that M is finitely generated A module such that
mM “ M. Then M “ 0.

Proof: The Jacobson radical of A equals m. o

Other formulations

6.54 There are several other reformulations of Nakayama’s lemma, and here
we offer a few of the most frequently applied version.

Proposition 6.55 Assume that φ : N Ñ M is A-linear and M is finitely gener-

ated. Moreover, let a be an ideal contained in the Jacobson radical of A: If φbid A{a is
surjective, it holds that φ is surjective.

Proof: The tensor product is right exact so cokerpφbid A{a q “ pcoker φqb A{a.
The cokernel coker φ is finitely generated since M is, and cokerpφbid A{a q “
pcoker φqb A{a “ 0 by hypothesis. Hence coker φ “ 0 by Nakayama III
(Proposition 6.51 above). o

Proposition 6.56 Let M is be a finitely generated A module. Assume that a is an

ideal contained in the Jacobson radical of A and that N a submodule of M such that
N ` aM “ M. Then N “ M.

Proof: The quotient M{N is finitely generated since M is, and it holds true
that aM{N “ M{N because any m from M lies in aM modulo elements in N;
or if you prefer, use the previous proposition with φ being the inclusion map
N ãÑ M. o

Proposition 6.57 Assume that a Ď A is an ideal contained in the Jacobson-radical

of A. Let M be a finitely generated A-module and assume that tmi uiP I are elements in
M whose residue classes generate M{aM. Then the mi ’s generate M.

Proof: Let N be the submodule of M generated by the mi ’s. The hypothesis

that the residue classes generate M{aM translates into the statement that
M “ N ` aM, and the proposition follows from Proposition 6.56. o
142 ma4200—commutative algebra

An extended version of Nakayama’s lemma

6.58 There is a version of Nakayama valid for all ideals not only those lying in
the Jacobson radical; but of course, when weakening the hypothesis you get a
weaker conclusion. The proof relies on a localization technique.

Proposition 6.59 (Nakayama extended) Let a be an ideal in the ring A, and

assume that M is a finitely generated A-module satisfying aM “ M. Then M is
killed by an element of the form 1 ` a with a P a; that is, there is an a P a so that
p1 ` aqM “ 0.

Proof: Let S be the multiplicative set t 1 ` a | a P a u. Then aAS is contained in

the Jacobson radical of AS (by Propostion 2.67 on page 45), and by Nakayama
Classic we may conclude that MS “ 0. Thence there is for each generator xi of
M, an element si P S killing xi . The xi ’s being finite in number we may form
the product of the si ’s, which obviously kills M and is of the required form. o

Problems

6.23 Let Φ be an n ˆ n-matrix with coefficients in a local ring A and denote ˇ

by Φs the matrix whose entries are the classes of the entries of Φ in the residue
class field k of A. Show that if the determinant det Φ
s does not vanish, then Φ
is invertible.

6.24 (Demystifying Nakayama’s lemma.) Let A be a local ring with residue class ˇ
field k. Assume that φ : E Ñ F is an A-linear map between free module of
finite rank, and let Φ be the matrix of φ in some bases.
a) Show that if one of the maximal minors of Φ
s does not vanish, one of the
maximal minors of Φ is invertible in A. Conclude φ is surjective when φbidk
is.
b) Show the classical Nakayama’s lemma for finitely presented modules over a
local ring by using the previous subproblem.
c) (Mystifying the demystification.) Show Nakayma’s lemma for finitely gener-
ated modules over a local ring by using subproblem a). Hint: The key word
is "right sections" of linear maps, if you don’, prefer coping with maximal
minors of n ˆ 8-matrices!!

6.25 Let A be a local ring with residue class field k. Let φ : E Ñ F be a ˇ

map between finitely generated free A-modules, and suppose that φbidk
is injective. Prove that φ is a split injection. Hint: Prove that at least one
maximal minor of the matrix of φ in some bases is invertible in A. Then the
projection π : mA Ñ nA corresponding to that minor furnishes a section.
 localization 143

6.26 Let M an A-module such that mM “ M for any maximal ideal m. Show
that M has the property that if one discards any finite part from a generating
set one still has a generating set.

6.27 Let M be a finitely generated A-module and let φ : M Ñ M be a surjective ˇ

A-linear map. Show that φ is injective. Show by exhibiting examples that this
is no longer true if M is not finitely generated. Hint: Regard M as a module
over the polynomial ring Arts with t acting on x P M as t ¨ x “ φpxq. Use the
extended version of Nakayama’s lemma with a “ ptqArts.

6.28 (Nilpotent Nakayama.) This exercise is about a result related to Nakayama’s

lemma, but of a much more trivial nature. Let A be a ring M an A-module.
Assume that a is a nilpotent ideal in A. Show that if aM “ M, then M “ 0.
À
6.29 (Graded Nakayama.) Let M “ i Mi be a graded module over the graded
À
ring R “ i Ri . Assume that M´i “ 0 for i sufficiently big; that is, the degrees
of the elements from M are bounded below. Let a be a homogenous ideal
whose generators are of positive degree. Assume that aM “ M and show that
M “ 0. Hint: Consider the largest n so that M´n ‰ 0.

6.30 Let A be ring and P a finitely generated projective module. Show that
there is a set of elements t f i u in A such that the distinguished open subsets
Dp f i q cover Spec A, and such that each localized module Pf i is a free module
over A f i .

6.31 Let A a ring and let e be a non-trivial idempotent element. Show that
À
the principal ideal I “ peqA is projective, and that a direct sum i I of a any
number, finite or not, of copies of I never can be free. Hint: Such sums are
killed by 1 ´ e.

6.4 The support of a module

The spectrum Spec A of a ring a geometric structure associated to A. The
points are the prime ideals in A and the topology is the Zariski topology
whose closed sets are those of the form Vpaq “ t p P Spec A | a Ď p u, where a is
any ideal in A.
Based on the belief that modules over local rings are simpler than other,
a general technique is to try to pass from local data—that is properties of
the localized modules Mp —to global data; that is, to infer properties of the
module M itself from properties of the localized modules Mp .
Since prime ideals p such that Mp “ 0 are insignificant in this process, it
is very natural to introduce the subset Supp M of Spec A consisting of the
144 ma4200—commutative algebra

prime ideals p so that Mp ‰ 0. This subset is called the support of M and is a The support of a module
geometric structure associated with M. In many case (e.g. when M is finitely (støtten til en modul)

generated) it is a closed subset.

The localness of being zero

We shall see several applications of the local to global principles, but begin
with the simplest of all properties, namely that of being zero! Applied to
kernels and cokernels this leads to a local criterion for a homomorphism to be
injective or surjective.
6.60 The point of departure is the following easy lemma that describes when
elements remain non-zero in localizations.

Lemma 6.61 Let M be an A-module and x and element in M. Assume that p is a

prime ideal in A. Then x does not map to zero in Mp if and only if Ann x Ď p.

Proof: The module Mp is M localized in the multiplicative set S “ Azp.

Recall from Lemma 6.33 on page 136 that the image of x in Mp being zero is
equivalent to there being an element in S “ Azp killing x; that is an element
belonging to Ann x but not to p. o
This lemma immediately yields the following principle:

Proposition 6.62 An A-module M equals zero if and only if Mm “ 0 for all

maximal ideals m in A.

That the ideals be maximal in the second condition can (of course) be replaced
by they being prime.
Proof: One way is obvious. So assume that M is non-zero and let x be a non-
zero element in M. Then the annihilator Ann x of x is a proper ideal as x ‰ 0
and hence contained in a maximal ideal m. By the simple lemma above, the
image of x in Mm is non-zero and a fortiori Mm is non-zero. o

Corollary 6.63 A map M Ñ N is injective (respectively surjective) if and only if

φm : Mm Ñ Nm is injective (respectively surjective) for all maximal ideals m in A.

Proof: Localization is exact, so pker φqm “ kerpφm q and Proposition 6.62

above tells us that ker φ “ 0 if and only if ker φm “ 0 for all m. o

Corollary 6.64 A map φ : M Ñ M is an isomorphism if and only if the localized

map φm : Mm Ñ Nm is an isomorphism for all maximal ideals m.

The support of a module

When we introduced the spectrum Spec A of a ring and gave it the Zariski
topology, we associated a closed set Vpaq with every ideal a in A. It con-
sisted of all the prime ideals p in A such that a Ď p. One may interpreted this
 localization 145

in terms of localizations. A prime ideal p belongs to Vpaq precisely when

pA{aqp ‰ 0; just apply the simple lemma 6.61 to the element 1 in A{a.
This can be generalized to finitely generated modules; for any such module
there is associated a closed subset of Spec A where the module is non-zero.
6.65 Given a module M, finitely generated or not. Recall that the support of M The support of a module
is denote by Supp M and consists of the prime ideals p such that Mp ‰ 0; that (Sstøtten til en modul)

is,
Supp M “ t p P Spec A | Mp ‰ 0 u.

Remember that elements of Mp are all shaped like xs´1 , and such an element
is zero precisely when tx “ 0 for some t R p.

Proposition 6.66 If M is finitely generated A-module, the support Supp M

equals the closed subset VpAnn Mq of Spec A; that is, it consists of the prime ideals p
containing Ann M.

Proof: Our task is to show that Mp ‰ 0 if and only if Ann M Ď p, or equiva-

lently, that Mp “ 0 if and only if Ann M Ę p. In case an element a P A kills
M and does not belong to p, it holds true that Mp “ 0 because a becomes
invertible in Ap ; this takes care of the if part of the proof. To attack the only if
part, assume that Mp “ 0, and let x1 , . . . , xr be generators of M. By the simple
lemma 6.61 above, there is for each of the xi ’s an element si killing xi , but not
lying in p. The product s of the si ’s clearly kills M, and it does not belong to
the prime ideal p since none of the si ’s does. Therefore s is an element with the
desired property. o

6.67 The hypothesis that M be finitely generated was used only in the last
part of the proof, and it holds true for a general module M that Supp M Ď VpAnn Mq.
The other inclusion may however fail when M is not finitely generated.

Examples

6.11 Clearly the support of a cyclic module A{a equals the closed set Vpaq. In-
deed, an element s lying in a, but not in p would kill A{a and hnece pA{aqp “
0.

6.12 One has Supp Q “ Spec Z since QS “ Q for any multiplicative set S
in Z. More generally, for the fraction field K of any domain A it holds that
Supp K “ Spec A.

6.13 An example of this failure for “large modules”, can be the Z-module
Z p8 “ Zrp´1 s{Z where p is a prime.
Every element of Z p8 is killed by a power of p; indeed, an element x is the
class of a rational number shaped Every element of Z p8 is of the form x “
a{pr with a prime to p. Since ax P Z is and only if a is divisible by pr , one has
Ann x “ ppr q and from Lemma 6.61 above it follows that Supp Z p8 “ ppqZ.
146 ma4200—commutative algebra

Even though every element of Z p8 is killed by a power of p, the annihilator

of Z p8 reduces to the zero ideal because no power of p kills the entire module
Z p8 (the power pr kills p´n only if n ď r).
This shows that even though Supp Z p8 is closed, it differs from VpAnn Z p8 q.
6.14 The support is not always a closed subset of Spec A. Take any infinite
sequence of primes pi not including all primes—for instance, every second
i Z{pi Z. The support of M is the
À
prime—and consider the module M “
infinite subset tppi qu. The only infinite closed subset of Spec Z being the entire
spectrum, this set is not closed.

The support of extensions

6.68 Since localisation is an additive functor so that pM ‘ Nqp » Mp ‘ Np it

is obvious that the support of a direct sum of two A-modules is the union of
their supports. This generalises to so-called extensions; that is, modules in the
midst of an exact sequence which is not necessarily split exact.

Proposition 6.69 Assume that

0 /N /M /L /0

is an exact sequence of A-modules. Then Supp M “ Supp N Y Supp L.

Proof: This follows immediately from localization functor being exact. For
each prime p the localized sequence

0 / Np / Mp / Lp /0

is exact, and the middle module vanishes if and only if the two extreme do. o

The support of a tensor product

6.70 The aim of this paragraph is the prove that the support of a tensor prod-
uct is the intersection of the supports of the two factors, at least when the
involved modules are finitely generated. The result hinges on Nakayama’s
lemma.

Proposition 6.71 Let M and N be two finitely generated A-modules. Then

Supp Mb A N “ Supp N X Supp M.

We begin with a local lemma:

Lemma 6.72 Let A be a local ring with maximal idea m. Let M and N be two
finitely generated A-modules. Then Mb A N “ 0 if and only if either N “ 0 or
M “ 0.
 localization 147

Proof: The proof is an application of Nakayama’s lemma. Let k “ A{m be the

residue class field of A. Assume that both N and M are non-zero. Nakayama’s
lemma then ensures that both Nb A k and Mb A k are non-zero, and since base
change respects tensor products (Proposition 5.35 on page 115), one has

pMb A Nqb A k “ pMb A kqbk pNb A kq.

The tensor product of two non-zero vector spaces being non-zero (e.g. Proposi-
tion 5.20 on page 108), we infer that pMb A Nqb A k ‰ 0, and hence Nb A M ‰ 0
a fortiori. o
Proof of Proposition 6.71: Since localization is an exact operation, the
localized modules Np and Mp are finitely generated over Ap whenever M and
N are finitely generated over A, and in view of the isomorphism

pMb A Nqp » Mp b Ap Np ,

the proposition then follows from the lemma. o

Example 6.15 Proposition Proposition 6.71 may fail when one of the factors
is not finitely generated. For instance, if one factor equals the fraction field
K of a domain A and the other is of the form A{a where a is a non-trivial
proper ideal, it holds true that A{ab A K “ 0; hence Supp A{ab A K “ H, but the
fraction field K is of global support (one has Kp “ K for all p P Spec A) so that
Supp K X sup A{a “ Vpaq, which is non-empty. K

Problem 6.32 Let p be a prime number and let M be the abelian group M “
iPN0 Z{p Z. Determine the annihilator Ann M and the support Supp M.
i
À
M

Problem 6.33 An abelian group M is said to be of bounded exponent if

some power pn of a prime p kills every element of M. Give an example of a
group of bounded exponent that is not finitely generated. Prove that if M is of
bounded exponent, then Supp M “ VpAnn Mq. M

Problem 6.34 Let M be a finitely generated A-module. Prove that if Mb A A{m “

0 for all maximal ideals m in A, then M “ 0. Hint: Combine Nakayama’s
lemma with Proposition 6.62. M
 localization 149

09/10/2018 Still Much work remaining; but the first section should be readable now. More
follows!
10/10/2018 Added a exercise 6.8; Minor changes in the first section; brushed up the second
section.
11/10/2018 Added prop 6.23 on page 130 and a new exercise 6.14 on page 131.
15/10/2018 Have moved the section about Nakayama’s lemma to this chapter.
Lecture 7

Chain conditions

Preliminary version 1.2 as of 2018-11-01 at 10:16 (typeset 3rd December 2018 at 10:03am)—Much
work remaining. Prone to misprints and errors and will change.
20/10/2018 Almost every thing has been redone!!
29/10/2018 Corrected a few misprints in lemma ?? in section about Krull’s intersection
theorem.
29/10/2018 Have added exercise 7.13 in connection with Krulls’ intersection theorem. Several
minor cahnges
01/11/2018 Added two exercises on page 172. Rewritten prop 7.50 on page 169. Added
theorem 7.58 on page 172.

One of the great moments of mathematics was the appearance of Emmy

Noether’s paper Idealtheorie in Ringbereichen in 1921 where she introduced
the ascending chain condition on ideals and proved the general version of
the Primary Decomposition Theorem. The chain conditions turned out to
be extremely useful, and today they permeate both commutative and non-
commutative algebra.

7.1 Noetherian modules

We introduced the concept of chains in partially ordered sets already when
discussing Zorn’s lemma (in Paragraph 2.46 on page 39). Recall that a chain C
in a partially ordered set Σ is just a linearly ordered subset; that is, a set such
that any two members of the subset C are comparable. Emmy Noether
(1882–1935)
7.1 In the present setting, when studying modules over a ring A, we give the German mathematician
term chain a more restrictive meaning. The chains we shall consider will all be
countable and well ordered. For practical reasons two sorts of chains will be
distinguished, ascending and descending ones. An ascending chain in M will be Ascending chains
a sequence of submodules tMi uiPN0 such that every term Mi is contained in in (oppstigend kjeder)

the successor Mi`1 ; or written out in a display, it is a chain of inclusions like

M0 Ď M1 Ď . . . Ď Mi Ď Mi`1 Ď . . . .

Similarly, a descending chain is a sequence tMi uiPN0 of submodules fitting into a Descending chain
chain of inclusions shaped like (nedstigende kjeder)

. . . Ď Mi`1 Ď Mi Ď . . . Ď M1 Ď M0 .
152 ma4200—commutative algebra

Such chains are said to be eventually constant or eventually terminating if the Eventually constant
submodules become equal from a certain point on; that is, for some index i0 it chains (terminerende
kjeder)
holds that Mi “ M j whenever i, j ě i0 . Common usage is also to say the chain
stabilizes at i0 .
7.2 An A-module M is said to be Noetherian if every ascending chain in M is Noetherian modules
eventually constant. This condition is frequently referred to as the Ascending (noetherske moduler)

Chain Condition abbreviated to acc . The module is Artinian if every descending The Ascending Chain
chain terminates, a condition also called the Descending Chain Condition with Condition (Den oppsti-
gende kjedebetingelsen)
the acronym dcc . Artinian modules
A ring A is called Noetherian if it is Noetherian as module over itself, and (artinske moduler)
of course, it is Artinian if it is Artinian as module over itself. The submodules The Descending Chain
Condition (Den nedsti-
of A are precisely the ideals, so A being Noetherian amounts to ideals of A gende kjedebetingelser)
satisfying the acc, and similarly, A is Artinian precisely when the ideals Noetherian and Artinian
comply with the dcc. rings (noetherske og
artinske ringer)
7.3 The two conditions, being Noetherian and Artinian, might look similar,
but in fact there is a huge difference between the two. Noetherian and Ar-
tinian modules belong in some sense to opposite corners of the category Mod A .
In what follows we shall treat Noetherian modules and Noetherian rings and
establish their basic properties, but will lack time to discuss the Artinian ones
in any depth, although Artinian rings will be discussed (in section 7.5 below).
In fact, according to a result of Akizuki, they turn out to be Noetherian as well.

7.4 The constituting property of Noetherian modules is the following theorem.

It is due to Emmy Noether and appears as one of the main theorems in her
famous paper from 1921.

Proposition 7.5 (The main theorem for Noetherian modules) Let A

be a ring and let M be a module over A. The following three conditions are equivalent:
Emil Artin (1898–1962)
o M is Noetherian; that is, it satisfies the ascending chain condition; Austrian mathematician

o Every non-empty family of submodules has a maximal element;

o Every submodule of M is finitely generated.

Proof: Assume first that M is Noetherian and let Σ be a non-empty set of

submodules. We must prove that Σ has a maximal element. Assuming the
contrary—that there is no maximal elements in Σ—one proves by an easy
induction on the length that every finite chain in Σ can be strictly lengthened
upwards. The resulting chain does not terminate, and the acc is violated.
Next, suppose that every non-empty set of submodules in N possesses
maximal elements. Our mission is to prove1 that every submodule N is finitely 1
Appealing to Zorn’s
generated. To that end, let Σ denote the set of finitely generated submodules. lemma, from which
the implication follows
It is clearly non-empty (the zero modules is finitely generated) and conse- directly, would be
quently has a maximal element N0 . Let x P N be any element. The module using a sledgehammer
to crack a nut. Apart
from avoiding unduly
use of brute force,
it is of interest that
the proposition is
independent of the
Axiom of Choice.
 chain conditions 153

Ax ` N0 is finitely generated and contains N0 , so from the maximality of N0 it

ensues that x P N0 . Hence N “ N0 , and N is finitely generated.
For the third and last implication, assume that all submodules of N are
finitely generated, and let an ascending chain

M0 Ď M1 Ď . . . . Ď Mi Ď Mi`1 Ď . . .
Ť
be given. The union N “ i Mi is by assumption finitely generated and have
say x1 , . . . , xr as generators. Each x j lies in some Mνj , and the chain being
ascending, they all lie in Mν with v “ max j νj . Therefore N “ Mν , and the
chain stabilizes at ν. o
7.6 It warrants a comment that Zorn’s lemma is not used in the proof of
families of submodules having maximal elements. As chains are eventually
constant, we avoid problems with limit ordinals, and ordinary inductions
works perfectly. As a consequence a large portion of the theory of Noethe-
rian modules does not depend on the Axiom of Choice. Notably existence
theorems like Theorem 2.50 on page 40 come for free for Noetherian rings.
7.7 The Noetherian modules, as do the Artinian modules, form a subcategory
of Mod A which enjoys a strong closedness property. They are what in category
theory are called thick subcategories. Submodules and quotients of Noetherian
modules are Noetherian as is an extension of two, and the same is true for
Artinian modules.

Proposition 7.8 Let M1 , M and M2 be three A-modules fitting in the short exact
sequence
0 / M1 /M / M2 / 0.

Then the middle module M is Noetherian (respectively Artinian) if and only if the two
extremal modules M1 and M2 are.

In particular—as may be proven by a straightforward induction—finite direct

sums of Noetherian (or Artinian) modules will be Noetherian (respectively
Artinian), and vice versa: If a direct sum is Noetherian (or Artinian) it is finite
and all the summands are Noetherian (or Artinian).
Proof: We may without loss of generality identify M1 with its image. Every
chain in M1 is of course a chain in M, so if M is Noetherian (or Artinian), the
same is true for M1 . In the same vein, if β : M Ñ M2 denotes the quotient
map, a chain tNi u in M2 lifts to the chain tβ´1 Ni u in M. Since β is surjective,
it holds that ββ´1 Mi “ Mi , and the former stabilizes whenever the latter does,
so if M is Noetherian (or Artinian), M2 is as well.
To prove the remaining half of the proposition assume that the two extreme
modules M1 and M2 are Noetherian (or Artinian) and let tNi u be a chain in M.
The chain tNi X M1 u stabilizes at some ν, hence Ni X M1 “ Nj X M1 for i, j ě ν.
Mapping the Ni ’s into M2 one obtains the chain tβpNi qu in M2 , and since
2
M by assumption is Noetherian (or Artinian), it stabilizes at some µ. Hence
154 ma4200—commutative algebra

βpNi q “ βpNj q for i, j ě µ. For i, j ě max µ, ν this gives

Ni {Ni X M1 “ βpNi q “ βpNj q “ Nj {Ni X M1 ,

and hence Ni “ Nj . o

7.9 The properties of being Noetherian or Artinian are retained when a

module is localized. The setting is as follows: A is a ring with a multiplicative
set S and M is an A-module. We let ι : M Ñ MS denote the localization map.

Proposition 7.10 Let S be a multiplicative set in the ring A and let M be an

A-module, if M is Noetherian or Artinian the localized module MS is Noetherian or
Artinian as well.

Proof: The proof is based on the simple observation that for any submodule
N Ď MS one has pι´1 NqS “ N, which is clear since elements in N are of the
form ιpyqs´1 . Now, any chain tNi u in MS , whether ascending or descending,
induces a chain tι´1 Ni u in M, and if this chain stabilizes, say ι´1 Ni “ ι´1 Nj
for i, j ě i0 , it holds true that Ni “ pι´1 Ni qS “ pι´1 Nj qS “ Nj , and the original
chain stabilizes at i0 as well. o

Example 7.1 (Vector spaces) A vector space V over a field k is Noetherian if

and only if it is of finite dimension. Indeed, if V is of finite dimension it is the
direct sum of finitely many copies of k, hence Noetherian.
If V is not of finite dimension one may find an infinite set v1 , . . . , vr , . . . of
linearly independent vectors, and the subspaces Vi “ă v1 , . . . , vi ą form a
strictly ascending chain of subspaces; hence V is not Noetherian. A similar
argument shows that neither is V Artinian: The spaces Wj “ă v j , v j`1 ¨ ¨ ¨ ą
form a strictly decreasing chain of subspaces. K

Example 7.2 (Finite product of fields) The conclusions of the preceding example
ś
extend to rings that are finite products of fields; say A “ 1ďiďr k i . Modules
À
over such rings are direct sums V “ 1ďiďr Vi where each Vi is a vector
space over k i with the A-module structure induced by the projection A Ñ k i .
From Proposition 7.8, or rather the succeeding comment, ensues that V is
Noetherian (or Artinian) if and only if each Vi is of finite dimension over k i . K

Problems

7.1 Prove the assertion just after Proposition 7.8 that if a direct sum of Noethe-
rian modules is Noetherian, the sum is finite and all the summands are
Noetherian.

7.2 Show that Z is a Noetherian Z-module, but that Z p8 “ Zrp´1 s{Z is not.
Show that Z p8 is an Artinian Z-module, but that Z is not.
 chain conditions 155

7.3 let φ : A Ñ B be a map of rings and let M be a B-module. Prove that if M

is Noetherian as a A-module, it is Noetherian as an B-module as well. Show
by exhibiting examples that the converse is not true in general, but prove that
the converse holds true when φ is surjective.

7.4 Show that a direct sum of finitely many simple modules is both Noethe-
rian and Artinian.

7.2 Noetherian rings

Recall that a ring A is called Noetherian if it is Noetherian as a module over
itself. The Noetherian rings form a large natural class of rings with a very rich
theory. The lion’s share of the rings appearing in classical algebraic geometry
are of so-called essential finite type over a field k (or over a Noetherian ground
ring); that is, they are localizations of finitely generated k-algebras. All these
rings are Noetherian. Hilbert’s basis theorem ensures that algebras finitely
generated over a Noetherian base are Noetherian, and by Proposition 7.10
above localizing a ring preserves the property of being Noetherian.
Be aware that even having lots nice properties, Noetherian rings can be
treacherous and show an unexpectedly bad behaviour. Local Noetherian rings,
however, are rather tame and well-behaved animals.
7.11 The ring A being Noetherian means that any ascending chain of ideals
eventually terminates. Applying Proposition 7.5 on page 152 to the ring A
itself while remembering that the submodules of A are precisely the ideals, we
arrive at the following:

Proposition 7.12 (The main theorem for Noetherian rings) Let A be

a ring. The following three conditions are equivalent:

o A is Noetherian, that is, the ideals in A comply to the ascending chain condition;

o Every non-empty family of ideals in A has a maximal element;

o Every ideal in A is finitely generated.

It is trivial that fields are Noetherian and we shall shortly see that polynomial
rings over fields are Noetherian; this is the celebrated Hilbert’s Basis Theorem.
Other examples are the principal ideal domains (see Exercise 7.6 below).
7.13 Quotients of Noetherian rings are Noetherian (Proposition 7.12), but
subrings are not necessarily Noetherian. A stupid example being the fraction
field of a non-noetherian domain, a more subtle example will be given below
(Example 7.4).
156 ma4200—commutative algebra

Proposition 7.14 Let A be a noetherian ring and M an A-module. Then M is

Noetherian if and only if M is finitely generated.

Proof: A finitely generated A-module M can be realized as the quotient of a

finite direct sum nA of n copies of A. When A is Noetherian, it follows from
Proposition 7.8 on page 153 that nA is Noetherian; indeed, one obtains nA by
successive extensions of A by itself. By Proposition 7.8 again, quotients of nA
are Noetherian as well.
Noetherian modules are finitely generated since all their submodules are
(Proposition 7.5 on page 152). o
7.15 A converse to this proposition does not hold in the sense that rings may
have Noetherian modules without being Noetherian; in fact, this applies to all
non-Noetherian rings. Simple modules are Noetherian (all submodules are
finitely generated!), and every ring possesses non-trivial simple modules by
The Basic Existence Theorem (Theorem 2.50 on page 40). These examples
are in some sense illustrative; any Noetherian module over a non-Noetherian
ring must have a non-trivial annihilator ideal; or phrased in another way,
they can not be what are called faithful modules; that is, modules M with Faithful modules (trofaste
Ann M “ p0q. moduler)

Proposition 7.16 Assume that M is a module over A. If M is Noetherian, then

A{ Ann M is Noetherian as well.

Proof: Let x1 , . . . , xr be generators for M, and consider the map φ : A Ñ

rM sending x to the tuple px ¨ x1 , . . . , x ¨ xr q. If x kills all the xi ’s, it kills the
entire module M, the xi ’s being generators, and we infer that the kernel of φ
equals the annihilator Ann M. This means that A{ Ann M is isomorphic to a
submodule of rM, hence it is Noetherian by Proposition 7.8 above. o

Examples

7.3 The obvious example of a non-Noetherian ring is the ring Arx1 , x2 , . . .s of

polynomials in infinitely many variables over any ring A. The chain of ideals

px1 qĂ px1 , x2 qĂ . . . Ă px1 , x2 , . . . , xi qĂ . . .

does obviously not stabilize.

7.4 One might be mislead by the previous example to believe that non-
noetherian rings are monstrously big. There are, however, non-noetherian
rings contained in the polynomial ring Qrxs. The simplest example is even a
subring of the ring Zrp´1 srxs where p is a natural number greater than one. It
is formed by those polynomials in Zrp´1 srxs that assume an integral value at
zero; that is, the polynomials Ppxq such that Pp0q P Z. In this ring A one finds
the following ascending chain of principal ideals

pp´1 xqĂ pp´2 xqĂ . . . Ă pp´i xqĂ . . . ,

 chain conditions 157

which does not stabilize. Indeed, if p´pi`1q x P pp´i xq, one would have
p´pi`1q x “ Ppxqp´i x for some polynomial Ppxq P A. Cancelling p´i x would
give p´1 “ Ppxq, which contradicts that Pp0q P Z.

7.5 A large class of important non-Noetherian rings are formed by the rings
HpΩq of holomorphic functions in an open domain Ω in the complex plane.
Chains that do not terminate arise from sequences of distinct points in Ω
that do not accumulate in Ω. If tzi u is such a sequence, let an be the ideal of
functions in HpΩq vanishing in the set Zn “ tzn`1 , zn`2 , . . .u. These ideals
clearly form an ascending chain, and from Weierstrass’ Existence Theorem
ensues that there are functions f n holomorphic in Ω whose zeros are exactly
the points in Zn . Then f n P an , but f n R an´1 , and the chain can not stabilize at
any stage.

A structure theorem
As an illustration of the strength and elegance the Noetherian method can
show, we offer a structure theorem for finitely generated modules over Noethe-
rian rings- it does not reveal the fine features of a module , but rather de-
scribes the overall structure. Every such module is obtained by successive
extensions of cyclic modules shaped like A{p with p a prime ideal.
7.17 The structure theorem builds on the following result which is of indepen-
dent importance and will be use later.

Proposition 7.18 Assume that A is a Noetherian ring and M an A-module. Let

Ann x be maximal among the annihilators of non-zero elements in M. Then Ann x is
a prime ideal.

Proof: To begin with, observe that Ann x is a proper ideal as x is non-zero.

Let then a and b be ring elements such that ab P Ann x and assume that
a R Ann x. Then ax ‰ 0. It is generally true that Ann x Ď Ann ax, but since
ax ‰ 0 it holds that Ann x “ Ann ax because Ann x is maximal among
annihilators of non-zero elements. Now, bax “ 0, so b P Ann ax “ Ann x. o

Corollary 7.19 Any non-zero module over a Noetherian ring contains a module
isomorphic to A{p for a prime ideal p.

Proof: The set of annihilators of non-zero elements is non empty and has a
maximal element since A is Noetherian. Then we cite Proposition 7.18 above.
o

7.20 The ground is now prepared for the announced structure theorem; here it
comes:
158 ma4200—commutative algebra

Proposition 7.21 Let A a Noetherian ring and let M be a non-zero A module.

Then M finitely generated if and only if it possesses a finite chain of submodules
tMi u1ďiďν whose subquotients are shaped like cyclic modules A{pi with the pi ’s being
prime; that is, there are short exact sequences

0 / M i ´1 / Mi / A{pi /0

for 1 ď i ď ν.

Proof: Let M be finitely generated A module. The set of submodules of M

for which the theorem is true is non-empty by Corollary 7.19 and has thus a
maximal element, say N. If N were a proper submodule, the quotient M{N
would be non-zero and hence contain a submodule isomorphic to A{p for
some prime p. The inverse image N 1 of A{p in M would be a submodule
containing N and satisfying N 1 {N » A{p, so the theorem would also hold for
N 1 violating the maximality of N. o

Problems

7.5 Let A Ď Q be any proper subring. Show that the polynomials in Qrts ˇ
assuming values in A at the origin, is not Noetherian.

7.6 Let A be a principal ideal domain. Show that A is Noetherian. Hint: If ˇ

Ť
ai “ pai q is an ascending chain, the union i ai is also principal.

7.7 Let tA1 , . . . , Ar u be a finite family of Noetherian rings. Show that the ˇ
ś
product i Ai is Noetherian.
ś
7.8 Let k be a field. Show that the product i PN k of a countable numbers of ˇ
copies of k is not Noetherian.

7.9 Show that the ring of numerical polynomials in Qrxs is not Noetherian.

7.10 Let A be the subring of the complex numbers C whose elements are
algebraic integers; that is, they are solutions of equations of the type

zn ` an´1 zn´1 ` . . . ` a1 z ` a0 “ 0

where the coefficients ai are integers. Show that A is not Noetherian. Hint:
?
For instance, the principal ideals p 2n 2q form an ascending sequence that does
not terminate.

M
 chain conditions 159

7.3 Hilbert’s Basis Theorem and two other results

There is almost an infinity of strong results about Noetherian rings, unfortu-
nately we have time to treat too few of them. As a beginning, in this section
we shall discuss three. In addition to Hilbert’s basis theorem, we treat a cri-
terion for rings being Noetherian due to I.S. Cohen and finally give one of
Wolfgang Krull’s many important results, his intersection theorem.

Hilbert’s Basis Theorem

As one might think the name indicates, Hilbert’s Basis Theorem lies at the
basis for the theory of commutative rings, and thereby is paramount for the
development of algebraic geometry. It guarantees that most rings appearing in
those parts of mathematics are Noetherian. However the name originate from
the content of the theorem, that any ideal in a polynomial ring over a field has
a finite basis—the modern version is that polynomial rings over Noetherian
rings are Noetherian.
Hilbert proved this theorem as early as in 1890. The proof was published
in the paper Über die Theorie der algebraischen Formen. Naturally the formula-
tion was slightly different from the modern one (the term Noetherian was
David Hilbert
of course not in use; Emmy Noether was only eight years old at the time), (1862–1943)
and the context was confined to polynomial rings over fields or specific rings German mathematician

like the integers, but the spirit was entirely the same. The abstract and non-
constructive proof was revolting at a time when that part of mathematics was
ruled by long and soporific computations, making it extremely difficult to
obtain general results, and it opened up the path to modern algebra. Part of
the mythology surrounding the theorem is the exclamation by the “König der
Invariant Theorie” Paul Gordan: “Das ist nicht Mathematik, das ist Theolo-
gie!”. The truth is that Hilbert had proved in a few pages what Gordan and Paul Albert Gordan
his school had not proved in twenty years. (1837–1912)
German mathematician
7.22 There are several different proofs in circulation, and we shall give one
of the shortest. These days many constructive proofs are known and good
algorithms exist for exhibiting explicit generators for ideals in polynomial
rings; however, we shall present a non-constructive proof in the spirit of
Hilbert’s.

Theorem 7.23 (Hilbert’s Basis Theorem) Assume that A is a Noetherian

ring. Then the polynomial ring Arxs is Noetherian.

Before giving the proof of Hilbert’s basis theorem we state three important
corollaries. A straightforward induction on the number variables immediately
yields the following:

Corollary 7.24 Assume that A is a Noetherian ring. Then the polynomial ring
Arx1 , . . . , xn s is Noetherian.
160 ma4200—commutative algebra

An important special case is when the ground ring A is a field. Since fields are
Noetherian, the Basis Theorem tells us that polynomial rings over fields are
Noetherian. Moreover, quotients of Noetherian rings are Noetherian, and we
obtain directly the next corollary. In particular it says that algebras of finite
type over a field, a class of rings that include the coordinate rings of affine
varieties, are Noetherian.

Corollary 7.25 Any algebra finitely generated over a Noetherian ring is Noethe-
rian.

Finally, the last corollary we offer before giving the proof of Hilbert’s Basis
Theorem, combines that theorem with Proposition 7.10 on pag 154 which
states that localization preserves Noetherianess. Recall that and A-algebra is
said to be essentially of finite type if it is the localization of a finitely generated
A-algebra.

Corollary 7.26 Any ring essential of finite type over a Noetherian ring is Noethe-
rian.

Proof of Hilbert’s basis theorem: Let a be an ideal in Arxs and for each n
let an be the set of leading coefficients of elements from a of degree at most n.
Each an is an ideal in A, and they form an ascending chain which eventually
stabilizes, say for n “ N. Each an is finitely generated, so for each n ď N
we may chose a finite set of polynomials of degree at most n whose leading
coefficients generate an . Let f 1 , . . . , f r be all these polynomials in some order
and let a1 , . . . , ar be their leading coefficients.
We contend that the f i ’s generate a. So assume not, and let f be of minimal
degree n among those polynomials in a that do not belong to the ideal gen-
erated by the f i ’s. If the leading coefficient of f is a, it holds that a P an and
ř
we may write a “ j b j ai j with the f i j ’s corresponding to ai j of degree at most
ř pdeg f ´deg f i j q
the degree of f . Then f ´ j b j x f i j is of degree less than deg f and
does not lie in the ideal generated by the f i ’s since f does not contradicting the
minimality of deg f . o

Cohen’s criterion
One may wonder if there are conditions only involving prime ideals that en-
sure a ring be Noetherian. An acc-condition on prime ideals is far from
sufficient; there are non-Noetherian rings with merely one prime ideal.
For instance, let m be the ideal generated by all the variables xi in the ring
krx1 , x2 , . . .s of polynomials in infinitely many variables. The quotient krx1 , x2 , . . .s{m2
has only one prime ideal, namely the one generated by the images of all the
variables, but is not Noetherian since that prime ideal is not finitely generated.
However, a result of Irvin Cohen’s tells us that for a ring to be Noetherian it
suffices that the prime ideals are finitely generated.
 chain conditions 161

7.27 We begin with stating a lemma about maximal ideals that are not finitely
generated; it joins the line of results of type ideals maximal subjected to some
condition being prime:

Lemma 7.28 Let a be maximal among the ideals in A that are not finitely generated.
Then a is a prime ideal.

Cohen’s criterion ensues easily from this lemma:

Proposition 7.29 Assume that all prime ideals in the ring A are finitely gener-
ated. Then A is Noetherian.

Proof: Assume that A is not Noetherian. The set of ideals that are not finitely
generated is then non-empty and according to Zorn’s lemma has a maximal
Ť
element, say a; indeed, if the union i ai of an ascending chain of ideals were
finitely generated, the chain would stabilize (argue as in the last part of the
proof of Proposition 7.12 on page 155) and a member of the chain would
be finitely generated. From the lemma we infer that a is prime, which is a
flagrant contradiction. o
Proof of Lemma 7.28: The ring A{a is Noetherian since all its ideals are
finitely generated. Let a and a1 be two members of A and assume that the
product aa1 lies in a, but that neither a nor a1 lies there. Let c “ a ` paq and
c1 “ a ` pa1 q. These ideals both contain a properly and are therefore finitely
generated by the maximality of a, moreover it holds that cc1 Ď a because aa1 P c.
The quotient c{cc1 is a finitely generated module over A{a, and contains a{cc1 .
Hence a{cc1 is finitely generated, and by consequence a since cc1 is finitely
generated. o

Krull’s intersection theorem

The German mathematician Wolfgang Krull was one of the greatest contrib-
utor to the development of algebra in the years between the two World Wars,
and in this section we shall discuss one of his most famous results, the so-
called the “Krull’s intersection theorem”. In its simplest form, the theorem
asserts that all the powers aν of a proper ideal a in a local Noetherian ring do
Ş
not have common elements apart from 0; that is, it holds true that ν aν “ 0.
There are several proofs of Krull’s intersection theorem, and the one we
give is among the shortest possible with the means at hand at the present Wolfgang Krull
(1899–19715)
stage of the course. There is a really short an elegant proof for the case of the
German mathematician
ring itself due to Hervé Perdry which we present as Exercise 7.13.
Example 7.6 To motivate and illustrate the reasons behind the result, let
us consider the ring of complex polynomials Crx1 , . . . , xr s and a point a “
pa1 , . . . , ar q in Cr . The ideal m “ px1 ´ a1 , . . . , xr ´ ar q consists precisely of the
polynomials that vanish at a, and the members of the powers mν are those that
162 ma4200—commutative algebra

vanish to the ν-th order. In this simple situation Krull’s theorem expresses
the well-known and obvious fact that no non-zero polynomial vanishes to
all orders. Of course, Krull’s result is a vast generalization of this prosaic
example; ideals in local Noetherian rings are infinitely more intricate than
maximal ideals in a ring of complex polynomials. K

7.30 The show begins with a technical lemma, and again submodules maxi-
mal subjected to a specific condition enter the scene:

Lemma 7.31 Let a Ď A be a finitely generated ideal. Let M be a Noetherian A-

module and N a submodule. If K is submodule of M maximal subjected to the condi-
tion K X N “ aN, then aν M Ď K for some ν P N.

Proof: Since a is finitely generated, it suffices to show that x ν M Ď K for every

x P a and ν sufficiently big. By the maximality of K, it suffices to prove that
px ν M ` Kq X N “ aN. The crucial inclusion is px ν M ` Kq X N Ď aN, the other
being clear as aN “ K X N.
Now, the submodules pK : xi q form an ascending chain which since M is
Noetherian stabilizes at say ν; so that pK : x ν q “ pK : x ν`1 q. If y “ x ν m ` k with
m P M and k P K, is a member of px ν M ` Kq X N it holds that xy P xN Ď K X N
from which ensues that m P pK : x ν`1 q since xy “ x v`1 m ` xk. Hence
m P pK : x ν q, and y P K X N “ aN. o

Proposition 7.32 Suppose that A is a ring, that a is a finitely generated ideal

ai M, one has
Ş
in A and that M is a Noetherian module over A. Putting N “ i
aN “ N.

Proof: Because M is Noetherian, there is a maximal submodule K of M such

that K X N “ aN. By the lemma it holds that aν M Ď K for some natural number
ν. Because N Ď aν M, i holds that N Ď K, and consequently N “ N X K “ aN. o

7.33 Combining Proposition 7.32 above with Nakayama’s lemma, we obtain

the classical version of Krull’s intersection theorem:

Theorem 7.34 (Krull’s intersection theorem) Let A be ring and a an

ideal contained in the Jacobson radical of A. Assume that a is finitely generated. If M
is a Noetherian A-module, it holds true that i ai M “ 0.
Ş

Proof: Let N “ i ai M. Then aN “ N after Proposition 7.32, and we

Ş

may finish by applying Nakayama’s lemma (Proposition 6.50 on page 140)

since N is a submodule of the Noetherian module M and therefore is finitely
generated. o

Corollary 7.35 Let A be a Noetherian ring and a an ideal contained in the

Jacobson radical of A. Then i ai “ 0. In particular, if A is a Noetherian local ring
Ş

whose maximal ideal is m, one has i mi “ 0.

Ş
 chain conditions 163

7.36 In general it is not true that the intersection of successive powers of

an ideal vanishes even when the ring is Noetherian and the ideal is proper.
Principal ideals generated by non-zero idempotents e furnish simple counter
examples. If a “ peq with e idempotent, one has a2 “ a and a straightforward
induction shows that ai “ a for all i. Hence i ai “ a. However, the powers of
Ş

proper ideals in Noetherian integral domains have vanishing intersections:

Corollary 7.37 Assume that a is a proper ideal in the Noetherian integral domain
ai “ 0.
Ş
A, then i

Proof: We combine Proposition 7.32 above by the extended Nakayama’s

lemma (Proposition 6.59 on page 142) and exhibit an element a P a so that
p1 ` aqN “ 0 where N “ i ai . But a being proper, 1 ` a is non-zero, and
Ş

consequently N “ 0 since A is an integral domain. o

Problems

7.11 The aim in this exercise is to exhibits a domain A with a maximal ideal
m, which is principal, such the intersection i mi is non-zero. It is in some a
Ş

minimal example of this behaviour, and illustrates how Krull’s intersection

theorem (may) fail in non-noetherian rings.
Let krT, X1 , X2 . . .s be the ring of polynomials in infinite many variables
having coefficients in the field k. Consider the quotient A “ krT, X1 , X2 , . . .s{a
by the ideal a generated by the polynomials Xi ´ TXi`1 for i P N. That is, the
ideal a is given as a “ pX1 ´ TX2 , X2 ´ TX3 , . . .q. As usual, letting lower case
letters, in the present case t and the xi ’s, stand for the classes of their upper
case counterparts, the equalities xi “ txi`1 hold in A.
a) Show that m “ ptqA is a maximal ideal and that the ideal p “ px1 , x2 , . . .q
generated by all the xi ’s is a prime ideal contained in m.
b) Prove that i mi “ p.
Ş

c) Let B be a domain containing k in which there is a principal ideal a “ p f q

such that i ai ‰ p0q. Show that there is a map of k-algebras A Ñ B.
Ş

7.12 Let A be a local ring with maximal ideal m. Assume that m is a principal
ideal. Prove that if i mi “ p0q, then the powers mi are the only ideals in A.
Ş

Prove that A is Noetherian if and only if i mi “ p0q.

Ş

7.13 (Perdry’s proof of Krull’s intersection theorem.) Let a be an ideal in a

Noetherian ring A and assume that x P i ai . The aim of the exercise is to
Ş

prove that x P x ¨ a. Assume that a1 , . . . , ar are generators for a.

164 ma4200—commutative algebra

a) Let ν P N be a natural number. Use that x P aν to prove there is a

homogenous polynomial Pν px1 , . . . , xr q of degree v in Arx1 , . . . , xr s so that
x “ Ppa1 , . . . , ar q.
b) Let cn be the ideal in Arx1 , . . . , xr s generated by P1 , . . . , Pn . Show there is an
N so that c N `1 “ c N .
ř
c) Show that one has a relation PN `1 “ 1ďiď N Qi ¨ Pi where the Qi ’s are
homogenous polynomials of positive degree.
d) Conclude that x P xa.
e) Deduce hat i ai “ 0 if a is contained in the Jacobson radical of A.
Ş

7.14 Let p be a prime number and let A0 “ Zp pq and m0 “ ppqZp pq . Let

An “ A0 rp1{n s be the ring obtained by adjoining a n-th root of p to A0 .
a) Show that An is local wih mn as sole macximal ideal. Hint: show that
m X A0 “ µ0

7.4 Modules of finite length

Finite dimensional vector space are civilized creatures having several features
that make them pleasant to work with, one being that they have a dimension.
Over any ring there is a class of modules with a numerical invariant attached
resembling the dimension of a vector space. This invariant is called the length,
and the modules in question are said to be of finite length. Modules do not
possess bases in general, so it is a lot more involved to define the length than
the dimension. The trick is to use maximal chains of submodules, maximal in Maximal chains (maksi-
the sense that there is no room for inserting new modules in the chain. male kjeder)

7.38 An ascending chain tMi u in an A-module M is called a composition series Composition series
if all of its subquotients Mi`1 {Mi are simple modules. Even though most (komposisjonsserier)

composition series we shall meet are finite, we do not exclude they being
infinite. By convention simple modules are non-zero, so in particular the
inclusions Mi Ĺ Mi`1 are strict. A finite composition series when displayed
appears like
0 “ M0 Ď M1 Ď . . . Ď Mn´1 Ď Mn “ M,

where each subquotient Mi`1 {Mi is shaped like A{mi for some maximal ideal
mi in A. The number n is called the the length of the series; it is the number of The length of a composi-
non-zero constituencies of the series. In case the series is infinite, the length is tion series (lengden av en
komposisjonskjede)
of course said to be infinite. More generally, the length of any finite chain will The length of a chain
be the number of inclusions; that is, one less than the number of modules. (lengden av en kjede)
Being a composition series is equivalent to being a maximal chain as de-
scribed in the introduction to this section; that is, no module is lying strictly
 chain conditions 165

between two consecutive terms of the chain. Would-be submodules in-between

Mi and Mi`1 would we in a one-to-one correspondence with submodules of
Mi`1 {Mi with a submodule N corresponding to the quotient N{Mi , but as
each Mi`1 {Mi is simple, there are none. The term saturated is also a common Saturated chains (mettede
usage for suchlike chains. kjeder)

7.39 The main result of this section is that once a module has one finite
composition series, they are all finite and have the same length. This is a result
of Jordan-Hölder type, but one on the weak side—the true Jordan-Hölder
theorem states that the subquotients of any two composition series are the
same up to a permutation. The original Jodan-Hölder theorem is about (finite)
groups, but one finds analogues in many categories, so also in the subcategory
of Mod A of finite lengths modules.

Theorem 7.40 (Weak Jordan-Hölder) Assume that M has a finite composi-

tion series. Then all composition series in M are finite and they have the same length.

The common length of the composition series is called the length of the mod- The length of modules
ule and denoted ` A pMq; for modules not being of finite length this means that (lengden til moduler)

` A pMq “ 8. As a matter of pedantry, the zero module2 is considered to be of 2

This is just a formal-
finite length and its length is zero (what else?). ity: By convention, the
zero module is not
Proof: The proof goes by induction on the length of the shortest composition included among the
series in a module; this is well defined and finite for all modules concerning simple modules and
does therefore not have
us. A module having a composition series of length one is simple, and for a composition series!
those the theorem is obviously true. The induction can begin and the fun can
start.
Let M “ tMi u be a composition series of minimal length n in M which
displayed is shaped like

0 “ M0 Ĺ M1 Ĺ . . . Ĺ Mn “ M.

The image of M in the quotient M{M1 is one shorter than M, hence all
composition series in M{M1 are of length n ´ 1 by induction. Denote by
β : M Ñ M{M1 the quotient map.
Given another another composition series N “ tNj u in M. Its length r is at
least n, and by induction its image in M{M1 is of length n ´ 1. Consequently
at least one of the inclusions in N becomes an equality in M{M1 ; that is, for
some index ν it holds that βpNµ q “ βpNµ`1 q, and we pick ν to be the least such
index. Then βpN q displays as

0 “ βpN0 q Ĺ βpN1 q Ĺ . . . Ĺ βpNν q “ βpNν`1 q Ď . . . Ď βpNr q “ M. (7.1)

We contend that there is just one index µ for which equality occurs— this is
the fulcrum of the proof from which it clearly ensues that r “ n; indeed, on
the one hand, the length of βpN q is one less than that of N and on the other, it
equals n ´ 1 by induction .
166 ma4200—commutative algebra

From the equality βpNν q “ βpNν`1 q comes the equality Nν ` M1 X Nv`1 “

Nv`1 , and M1 being simple, either M1 X Nv`1 vanishes or equals M1 . It cannot
vanish, however, because Nν ‰ Nv`1 , and thus we infer that M1 Ď Nν`1 . It
follows that there are strict inclusions βpNj q Ĺ βpNj`1 q for j ě ν ` 1, and hence
all inclusions in βpN q are strict except the one at stage ν. o
7.41 A closer look at the proof above reveals that it as well gives the full
Jordan-Hölder theorem:

Theorem 7.42 (True Jordan-Hölder) Any two composition series of a mod-

ule of finite length have the same subquotients up to order.

Proof: We continue with the above proof and go on with induction on the
length; hence the two series βpMq and βpN q have the same subquotients up to
order. Now, the subquotients of M and βpMq differ only at the bottom stage
M1 , so M has the subquotient M1 in addition to those shared with βpMq. On
the other hand, the subquotients of N and βpN q coincide except at stage ν,
but in the proof above, we showed that Nv ` M1 “ Nv`1 , and the additional
subquotient of N is M1 as well. o
7.43 Just like the dimension of vector spaces the length is additive along
short exact sequence, and this is an indispensable property making it possible
to compute the length in many cases. Observe also that a submodule (or a
quotient) of M having the same length as M must be equal to M.

Proposition 7.44 (Additivity of length) Given a short exact sequence of

A-modules
/ M1 /M / M2 /0
α β
0
Then M is of finite length if and only if the two others are, and it holds true that
` A pMq “ ` A pM1 q ` ` A pM2 q.

Proof: Firstly, assume M is of finite length. Pushing a finite composition

series forward3 along β gives one in M2 and pulling it back along α gives one 3
Of course, not repeat-
in M1 . Assume next that the two modules on the side are of finite length. It ing terms that become
equal in 2
suffices to exhibit one composition series of M with the additive property.
To this end, we begin with one in M2 , say tMi2 u, and pull it back to M along
β. The smallest module in the pulled back chain is β´1 M02 “ β´1 p0q, which
equals M1 , so we may splice tβ´1 Mi2 u with any composition series of M1 to
obtain one in M, and obviously, the length of the spliced series equals the sum
of lengths of the two being spliced. o
7.45 An immediate corollary of Proposition 7.44 is that modules of finite
length are both Noetherian and Artinian. Obviously this is true for sim-
ple modules (no submodules, no chains) and hence follows in general by
a straightforward induction on the length using Proposition 7.8 on page 153.
The converse holds as well:
 chain conditions 167

Proposition 7.46 An A-module M is of finite length if and only if it is both

Noetherian and Artinian.

Proof: Assume M both Noetherian and Artinian. Since M is Artinian every

non-empty set of submodules has a minimal element, so if M is not of finite
length, there is a submodule, N say, minimal subjected to being non-zero
and not of finite length. It is finitely generated because M is Noetherian and
hence Nakayama’s lemma applies: There is surjection φ : M Ñ k onto a simple
module k. The kernel of φ is of finite length by the minimality of N, and hence
N itself is of finite length by Proposition 7.44 above. o
7.47 Be aware that the base ring A is a serious part of the game and can
have a decisive effect on the length of a module. If A Ñ B is a map of rings
and M a B-module which is of finite length over both A and B, there is in
general no reason that ` A pMq and ` B pMq should agree. Already when k Ď K is
a finite non-trivial extension of fields the two lengths differ in that dimK V “
rK; ks ¨ dimk V for vector spaces over K. You will find a simple but slightly
more subtle example in Example 7.10 below. And of course there are stupid
examples like Q Ď R with R being of length over itself, but as a module over Q
its length is infinite (even uncountable!)
However, when the map A Ñ B is surjective, the two lengths agree since
then the B-submodules and the A-submodules of M coincide.
7.48 Unlike what is true for vector spaces, module of the same length need
not be isomorphic. A simple example are the Z-modules F p “ Z{pZ. They
are all of length one but not two are isomorphic!

Examples

7.7 (Vector spaces) Over fields, of course the length of modules coincide with
their vector space dimensions. In a similar fashion, if the A-module M is killed
by a maximal ideal m in A, and therefore is a vector space over the field A{m,
one has ` A pMq “ dim A{m M.
7.8 (Finite groups) The only abelian groups that are of finite length are the
finite groups. They are all direct sums of cyclic groups of shape Z{pν Z where
p is a prime and ν a natural number; that is such a group M enjoys a direct
sum decomposition
Z{pi i Z.
à ν
M“
i
ř
We contend that `Z pMq “ i νi . By additivity of the length it suffices to show
that for a each prime p the length of Z{pν Z is given as ` A pZ{pv Zq “ v, and
this one does by an induction argument based on the standard4 mod pν . short 4
The map p is close to
exact sequences being a “multiplication-
by- p-map”; it sends a
p class rxs pν´1 mod pν´1
0 / Z{pν´1 Z / Z{pv Z / Z{pZ / 0. to the class rpxs pν
168 ma4200—commutative algebra

Since `Z pZ{pZq “ 1, the length ` A pZ{pν Zq increases by one when ν increases

by one, and we are done.
7.9 Let A “ krx, ys and Mn “ krx, ys{mn . Then there are exact sequences like

0 / mn´1 {mn / Mn / M n ´1 /0

so that ` A pMn q “ ` A pMn´1 q ` dimk mn´1 {mn “ ` A pMn q ` n and by induction

one finds
n ˆ ˙
ÿ n`1
` A pMn q “ i“
2
i “1

7.10 We let A “ Z and B “ Zris “ Zrxs{px2 ` 1q and let M “ Zris{p105qZris.

The prime factorization of 105 is 105 “ 3 ¨ 5 ¨ 7, and one checks easily that
x2 ` 1 is irreducible mod 3 and 7 but decomposes over F5 ; hence the Chinese
remainder theorem gives a decomposition

M “ F3 piq ‘ F5 ‘ F5 ‘ F7 piq,

and we conclude that `Z pMq “ 6 but `Zris pMq “ 4. (The primes 3 and 7 persist
being primes in Zris, but 5 splits up in the product 5 “ p2 ` iqp2 ´ iq.)

Problems

7.15 A frequently met situation in algebraic geometry is that a ring A is ˇ

a k-algebra; that is, it contains a ground field k (for instance, algebras like
krx1 , . . . , xr s{a are of this type). Then any A-module is a vector space over
k. Assume that A in addition to being a k-algebra is local ring. Denote the
maximal ideal by m and let kpmq “ A{m be the residue class field.
a) Assume that k maps isomorphically onto kpmq. Prove that a module M is
of finite length over A if and only if is of finite dimension over k and in case it
holds true that dimk M “ ` A pMq.
b) Assume merely that kpmq is finite extension of the image of k. Prove that
dimk M “ rkpmq : ks ¨ ` A pMq.
ś
7.16 (Modules of finite length over finite product of fields.) Let A “ 1ďiďr k i be ˇ
À
a finite product of fields. Show that an A-module V “ 1ďiďr Vi , where Vi is
a vector space over k i , is of finite length if and only if Vi is of finite dimension
over k i , and in case it holds true that
à ÿ
` A p Vi q “ dimki Vi .
i i

7.17 Show that the the length `Z pZ{nZq equals the number of prime factors ˇ
of n (counted with multiplicity).
 chain conditions 169

7.18 (Modules of finite length over pid’s.) Let A be pid and let f P A be an ˇ
element. Show that ` A pA{p f qAq is the number of prime factors in f (counted
with multiplicity).

7.19 Assume that M is an A-module of finite length and that a is an ideal

contained in the Jacobson radical of A. Show that for some integer n it holds
true that an M “ 0. Hint: Consider the descending chain ai M and remember
Nakayama’s lemma.

Finite length and support

7.49 We finish of the story about modules of finite length with a criterion for
a module being of finite length in terms of the support of the module, and a
structure theorem, essentially saying that a module is of finite length is just a
finite direct sum of "local contributions"—but of course, it says nothing about
how the local contributions are shaped.

Proposition 7.50 A finitely generated module M over a Noetherian ring is of

finite length if and only if its support Supp M is a finite union of closed points.
Recall that the closed
Proof: Assume to begin with that M is of finite length and let tMi u be a points in Spec A are
precisely the maximal
composition series. Then citing Proposition 6.69 on page 146 we infer that ideals. It may well
Ť Ť
Supp M “ i Supp Mi {Mi`1 “ i tmi u where Mi {Mi`1 » A{mi are the happen that a subset
subquotients of the composition series tMi u. So the support is a finite union of of Spec A is closed
and finite without all
closed points. points being closed; for
For the other implication we resort to the Structure Theorem 7.21 on instance, Spec Zppq is
finite.
page 158 assuring there is a descending chain tMi u of submodules in M
whose subquotients are shaped like A{pi with pi being prime. Again by Propo-
Ť
sition 6.69 on page 146 it holds that Supp M “ i Vppi q. Now, if M is of finite
length, the the subquotients A{pi ’s will be as well, and by the lemma below, all
the pi ’s must be maximal. Hence Supp M is finite. o

Lemma 7.51 Assume that an integral domain A is Artinian, then A is a field.

Proof: Let f P A be a non-zero member of A. The principal ideals p f i q

form a descending chain and is ultimately constant since A is artinian; i.e.
p f ν`1 q “ p f ν q for some ν. Then f ν “ a f ν`1 for some a P A, and cancelling f ν ,
which is permissible as A is a domain, we find 1 “ a f ; i.e. f is invertible. o

Corollary 7.52 (Structure of finite length modules) Assume that M

is of finite length over the ring A. Then there is a canonical isomorphism
à
M» Mm .
mPSupp M
170 ma4200—commutative algebra

Proof: The localization maps M Ñ Mm combine to a map Φ : M Ñ

À
mPSupp M Mm , which fits in the exact sequence

/ ker Φ /M / / coker Φ / 0.
À
0 mPSupp M Mm

Given a prime ideal p there are two outcomes when the sequence is local-
ized at p. Either p does not contain any of the mi ’s, and then all terms of the
sequence become zero when localized, or p is one of the maximal ideals in
Supp M; say p “ m. If m1 ‰ m is another member of Supp M, there are ele-
ments in m1 not belonging to m, and hence pMm1 qm “ 0. Therefore the only
term in the sum that survives being localized at m is Mm , and the sum lo-
calises to Mm . But of course M localises to Mm as well, and the localization
map localizes to the identity! It follows that ker Φ and coker Φ localizes to
zero everywhere, and hence they are zero by the"The Localness of Being Zero"
principle (Proposition 6.62 on page 144). o

7.5 Artinian ring

We now turn to the rings whose ideals satisfy the descending chain condition;
that is, rings being Artinian modules over themselves. Even though the def-
initions may appear symmetric, the class of Artinian rings is astonishingly
different from the class of Noetherian rings. The latter is a large class encom-
passing almost all rings one meets in algebraic geometry, whereas the Artinian
ones merely serve special (but important) purposes. They are the tiny, little
brothers among the Noetherian rings—but they carry a great lot of subtleties.
7.53 It turns out that, as we shorty shall see, that Artinian rings are Noethe-
Yasuo Akizuki
rian. This is specific for rings, but far from true for modules. The Artinian
(1902–1984)
rings are characterized among the Noetherian ones by the property that all Japanese mathematician
their prime ideals are maximal and that the maximal ideals are finite in num-
ber. In geometric terms, their spectra are finite sets and the Zariski topology is
discrete.
The theorem we are about to prove is due to the Japanese mathematician
Yasuo Akizuki. There is an analogue version valid for non-commutative rings,
proved about at the same time as Akizuki proved his theorem, which usually
is contributed to Charles Hopkins and Jacob Levitzki, but the the commutative
version is Akizuki’s:

Theorem 7.54 (Akizuki’s Theorem) An Artinian ring is Noetherian and all

its prime deals are maximal and they are finite in number.

7.55 Recall that a module which is both Noetherian and Artinian is of finite
length. Hence Artinian rings are of finite length (regarded as modules over
themselves) and hence come with a natural numerical invariant, the length
lpAq; that is, the number of (simple) subquotients in a composition series.
 chain conditions 171

The proof of Akizuki’s theoremùñ

The proof of Akizuki’s theorem will be done in series of two lemmas. We
begin with proving the second statement in the theorem, the one about prime
ideals being maximal and finite in number. This is the easy piece, that A is
noetherian, is deeper.
Lemma 7.56 Every prime ideal in an Artinian ring A is maximal and they are finite
in number. Hence if J is the radical of A, the quotient A{J is a finite product of fields.

Proof: We have already established the first assertion, if p is a prime in A, the

quotient A{p is an Artinian domain, hence a field by Lemma 7.51 above.
As to the second statement, assume that tmi uiPN is a countable set of dif-
ferent maximal ideals in A. For each natural number r consider the ideal
Ş
Nr “ iďr mi . They form a descending chain, and A being Artinian it holds
Ş
true that Nn “ Nn`1 for some n. This means that iďr mi Ď mn`1 , and by the
Prime Avoidance Lemma (Lemma 2.28 on page 33), one of the mi ’s must lie
within mn`1 , contradicting the assumption that the mi ’s are different.
The last assertion ensues from the Chinese remainder theorem. If m1 , . . . , mr
Ş
are the prime ideals in A, the radical J equals J “ i mi , and since all the mi ’s
are maximal, they are pair-wise comaximal. Hence the Chinese Remainder
ś
Theorem gives an isomorphism A{J » i A{mi . o
The preceding lemma implies that the radical J of A coincides with inter-
section of the maximal ideals in A. The elements are nilpotent, but A is not
a priori Noetherian, so J is not a priori a nilpotent ideal. However, our next
lemma says it is.
Lemma 7.57 The radical J of an Artinian ring A is nilpotent; that is, J n “ 0 for
some n. Moreover, J is Noetherian.

This lemma concludes the proof of Akizuki’s theorem. Since A{J, being the
product of finite number of fields is Noetherian, we infer that A is Noetherian.
Proof: Consider the set of powers J r that are not finitely generated. If J is
not finitely generated, that set is non-empty and since A is Artinian it has
a a minimal element, say J m . In any case, there is an m so that J v is finitely
generated for v ě m.
The descending chain of powers J r becomes stationary at a certain stage as
well, that is, there is an m such that J ν`1 “ J ν whenever ν ě m.
Altogether, we infer there is an r so that J r is finitely generated and satisfies
J 1 “ J r and we are thence in the position to apply Nakayama’s lemma to the
r `

J r , and may conclude that J r “ 0. o

The final step of the proof of Akizuki’s theorem is an induction argument
to show that J is Noetherian. For ν sufficiently big, we saw in the previous
lemma that J ν “ 0, and for each ν there is a short sequence:

0 / J v `1 / Jν / J ν {J ν`1 / 0.
172 ma4200—commutative algebra

Submodules and quotients of Artinian modules are Artinian (Proposition 7.8

on page 153), so it follows that J ν , and therefore also J ν {jν`1 , is Artinian. But
J ν {J ν`1 is a module over A{J which we just proved is a finite product of fields,
and over such rings any Artinian module is Noetherian (Example 7.2 on
page 154); and we are through by descending induction on ν.

The structure of Artinian rings

Since Artinian rings are of finite length over themselves, we may apply the
Structure Theorems (Proposition 7.50 and Theorem 7.52 both on page 169) to
obtain the following description

Theorem 7.58 Let A be an Artinian ring. Then Spec A is finite and discrete, and
the localisation maps A Ñ Am induce an isomorphism
ź
A» Am .
mPSpec A

If A is Noetherian and Spec A is finite and discrete, then A is Artinian.

Saying Spec A is finite and discrete is just another way of saying that all prime
ideals in A are maximal and finite in number. Anticipating the notion of Krull
dimension, a ring all whose prime ideals are maximal is said to be of Krull
dimension zero. Hence a Noetherien ring A is Atinian if and only if its Krull
dimension equals zero.
The theorem says nothing about local Artinian rings, even if they might
appear small and innocuous, they can be extremely intricate creatures.

Problems

7.20 Let n and m be two natural numbers and let a be the ideal a “ px m , yn q in ˇ
krx, ys. Show that A “ krx, ys{a is Artinian and compute its length.

7.21 Let n, m and r be three natural numbers and let A “ krx, y, zs{px n , ym , zr q.
Prove that A is Artinian and compute its length.

M
Lecture 8

Primary decomposition

Preliminary version 1.0 as of 2018-11-15 at 11:29:54 (typeset 3rd December 2018 at 10:03am)—
Much work remaining. Prone to misprints and errors and will change.
2018-11-15 Added exercise 6.20, corrected exercises 8.9 and ??

The beginning of the story about primary decomposition is the Fundamen-

tal Theorem of Arithmetic which states that any integer has a representation
as a product of prime numbers unambiguous up to order. The early 19th cen-
tury mathematicians discovered however that algebraic integers do not share
this property unconditionally, there are rings of algebraic numbers for which
the analogue of the Fundamental Theorem does not hold; where irreducible
factors are not always unique.
However, for a large class rings appearing in algebraic number theory—the
so-called Dedekind domains—the situation could be saved by using prime
ideals instead of prime numbers; any non-zero and proper ideal in a Dedekind
domain is a product of powers of prime ideals in an unambiguous ways.
Dedekind domains are special, and the question arose quickly what is
generally true. Emanuel Lasker1 was one of the first to give a partial answer, 1
In addition to be an
he established the primary decomposition for ideals in rings finitely generated eminent mathematician,
Lasker was World
over fields. The final breakthrough came in Emmy Noether’s famous 1921- Chess Champion for 27
paper. Her results were profoundly more general and her proofs enormously years
easier and more translucent than the previous.
In a general decomposition products are replaced by intersections and pow-
ers of prime ideals by so-called primary ideals. Every ideal a in a Noetherian
ring has a such a primary decomposition. One may write a as an intersection
a “ q1 X . . . X qr of primary ideals, but the uniqueness of the intervening ideals
are only partially true.
There is also a geometric interpretation of this decomposition. The closed
subset X “ Vpaq of the spectrum Spec A—or the variety X “ Vpaq in An if
one prefer working with ideals in a polynomial ring—can be decomposed in a
union of closed, irreducible2 subsets called the irreducible components of X. For Emanuel Lasker
instance, the subset given by xyz “ 0 in C3 has the thee coordinate planes as (1868–1941)
German mathematician
components.
The road map of this chapter is as follows: We begin with introducing the 2
A topological space is
irreducible if it is not the
union of two proper,
closed subsets.
174 ma4200—commutative algebra

new important players, the primary ideals, and establish their basic properties.
Next follows the announcement of the main theorem and a discussion around
it, and finally the main theorem is proven through a series of lemmas.

8.1 Primary ideals

As alluded to in the introduction to this chapter, one needs a notion of primary
ideals which generalizes the notion of “prime powers” we have for integers.
The naive try would be to just use powers of prime ideals, but this is in fact
too naive and doesn’t work—the issue is certainly of a more subtle character.
8.1 The property of an ideal q being being primary is best introduced as
a property of the quotient A{q. To motivate the definition, consider the
multiplication-by-n-map Z{pr Z Ñ Z{pr Z. It is either bijective or nilpotent3 , 3
When n is prime to p
and the salient point is that this characterises prime powers. If an integer m it is bijective, and when
n has p as factor, it will
has different primes p and q as factors, multiplication by p (or by q) in Z{mZ be nilpotent.
is neither bijective nor nilpotent.
Inspired by this exquisite property of prime powers, we call a proper ideal q As for prime ideals, we
a primary ideal if the following condition is satisfied: insist on primary ideals
being proper.
o For every element x P A the multiplication map A{q Ñ A{q that sends an Primary ideals
(primæridealer)
element y to x ¨ y is either injective or nilpotent.
The "multiplication-by-x-map" is frequently called by a more scientific name,
‘
namely the homothety by x Since the radical q consists of ring elements with Homotheties (homotetier)
a power lying in q, the condition may be reformulated as
‘
o If xy P q, then either y P q or x P q.

Basic properties of primary ideals

In this section a series of four small propositions we shall establish four basic
properties of primary ideal. We shall see that their radicals are prime ideals
(as would be expected from offshoots of prime powers), that they behave well
with respect to intersections, localizations and quotient formation.
8.2 The first property of primary ideals we discuss is that their radicals are
‘
prime ideals. Once this is established, it follows that q is minimal among
the prime prime ideals containing q. Indeed, as for any ideal, the radical of q
equals the intersection of the prime ideals containing q (see Proposition 2.59
on page 42). In other words, the ring A{q has just one minimal prime ideal,
which is formed by the nilpotent elements.
‘
Proposition 8.3 If q is a primary ideal in the ring A, the radical q is a prime
ideal, and it is the smallest prime ideal containing q.

Proof: Assume that xy P q, but y R q; then x n yn lies in q for some n, but

‘ ‘

yn R q, so some power of x n lies there. Hence x P q.

‘
o
 primary decomposition 175

‘
It is customary to say that a primary ideal q is p-primary when p “ q, or one p-primary ideals (p-
also says that p belongs to q. The converse of Proposition 8.3 does not hold in primære idealer)

general; the radical being prime is not sufficient for an ideal to be primary.
Example 8.1 below is an easy and concrete instance of this, and more elaborate
example in a polynomial ring is found in Exercise 8.4 below. However, if the
radical of q is maximal, q is primary:

Proposition 8.4 An ideal q whose radical is maximal, is primary.

‘ ‘
Proof: Assume that the radical q is maximal and write m “ q. Because m
is both maximal and the smallest prime containing q, the ring A{q is a local
ring with maximal ideal m{q as the only prime ideal. Therefore the elements
of m{q are nilpotent while those not in m{q are invertible. o

Corollary 8.5 The powers mn of a maximal ideal m are m-primary.

Proof: The radical of mn equals m. o

Examples

8.1 The ideal a “ px2 , xyq in the polynomial ring krx, ys has a radical that is
prime, but a is not primary. Clearly the radical of px2 , xyq equals pxq which is
prime, but in the quotient krx, ys{px2 , xyq multiplication by y is neither injective
nor nilpotent (y kills the class of x, but no power of y lies in px2 , xyq). One
decomposition of px2 , xyq as an intersection of primary ideals is

px2 , xyq “ pxq X px2 , yq.

Checking the equality is a nice little exercise: A relation z “ ax “ bx2 ` cy

entails that x|c (the polynomial ring is ufd), and hence z P px2 , xyq. Notice
that both ideals in the intersections are primary; pxq since it is prime and
px2 , yq because the radical equals px, yq which is maximal. One also has the
decomposition
px2 , xyq “ pxq X px, yq2 ,

indeed, the polynomials in px2 , xyq are those with x as factor that vanish at
least to the second order at the origin. This gives an example of the primary
decomposition not being unique.

8.2 A standard example of a prime ideal whose square is not primary is as

follows. Let A “ krX, Y, Zs{pX 2 ´ YZq and, by the usual convention, x, y and z
are the classes of the variables in A. The ideal p “ px, yq is prime, but p2 is not
primary; indeed, yz lies there, but neither does y lie in p2 nor does z lie in p. A
decomposition into primary ideals of p2 is shaped like

px, yq2 “ px2 , xy, y2 q “ pyz, yx, y2 q “ px, y, zq X pyq.

176 ma4200—commutative algebra

Obviously the ideal px, y, zq being maximal is primary. The ideal pyq is more
interesting. Killing y, we obtain the ring A{pyq “ krX, Zs{pX 2 q, whose elements
are either non-zero divisors or nilpotent4 and pyq is a primary ideal. It’s 4
The elements are of
radical equals px, yq. the form apzq ` bpzqx
with a, b P krzs, and
one easily sees that this
is a non-zero divisor
K unless a “ 0, but then
the square is zero.
8.6 The intersection of finitely many p-primary ideals persist being p-primary.
In the analogy with the integers, this reflects the simple fact that the greatest
common divisor of some powers of the same prime number is a power of that
prime.

Proposition 8.7 If tqi u is a finite collection of p-primary ideals, then the intersec-
Ş
tion i qi is p-primary.

Proof: Taking radicals commutes with taking finite intersection (Lemma 2.63
‘Ş Ş ‘
on page 43), and therefore one has i qi “ i qi “ p. Assume next that
Ş Ş
xy P i qi , but y R i qi ; that is, xy P qi for each i, but y R qν for some ν. Since
‘
qν is p-primary x lies in the radical qν of qν , which equals p, but as we just
Ş
checked, p is as well the radical of the intersection i qi . o
The hypothesis that the intersection be finite cannot be discarded. Powers
mi of a maximal ideal are all primary and have the same radical, namley m,
but but at least when A is Noetherian, their intersection equals the zero ideal
p0q by Krull’s intersection theorem— which might be primary, but certainly
not m-primary (in most cases). There are however instances when infinite
intersection of primary ideals are primary.

Problems

8.1 With notation as in Example 8.2 on the previous page, show that pn is not ˇ
primary for any n ě 2. Hint: Show that zyn´1 P pn but yn´1 R pn .

8.2 Let tqi uiP I be a collection of primary ideals (of any cardinality) all with the ˇ
same radical p. Assume that there is a natural number n so that pn Ď qi for all i.
Ş
Prove that the intersection iP I qi is primary with p as radical.

8.3 Let the group G act5 on the Noetherian ring A and let q be a p-primary ˇ
ideal. Assume that p is invariant under G. Prove that gPG qg is p-primary and The action induces
Ş 5

an action on the
invariant under G. ideals by setting
ag “ t ga | g P G, a P a u

8.8 The property of being primary is compatible with localizations, at least

when these are performed with respect to multiplicative sets disjoint from the
radical.
 primary decomposition 177

Proposition 8.9 Let S a multiplicative set in the ring A and let q be a p-primary
ideal. Assume that S X p “ H. Then q AS is p AS -primary and it holds true that
1
ι´
S pq AS q “ q.

Proof: Localizing commutes with forming radicals (Proposition 6.23 on

page 130) so the radical of q AS equals p AS . Assume that x{s ¨ y{s1 P q AS , but
that y{s1 R q AS . Then txy P q for some t P S, and obviously it holds that y R q.
Hence tx lies in the radical p of q, and since t R p, we conclude that x P p; in
other words x{s lies in p AS .
1
To verify that ι´S pq AS q “ q let x P A be such that ι S pxq P q AS . This means
that sx P q for some s P S, but by hypothesis p X S “ H so that s R p; thence
x P q because q is primary. o
8.10 The fourth property permits us to replace A by A{q and q by the zero
ideal, which makes some arguments cleaner and notationally simpler.

Proposition 8.11 Let φ : A Ñ B be a surjective map of rings with kernel a.

Assume that q is an ideal in A containing the kernel a. Then the image qB “ q{a of
q in B is primary if and only if q is. The radical of the image equals the image of the
‘ ‘
radical; or in symbols, pq{aq “ p qq{a.

Proof: This is pretty obvious because by the Isomorphism Theorem (Theo-

rem 2.18 on page 30) it holds that A{q “ B{qB, so the multiplication-by-what-
ever-maps are the same. o
In particular, we observe that the ideal q is primary if and only if the zero
ideal p0q is a primary ideal in the quotient A{q.
Problem 8.4 This problem is meant to illustrate that even in a polynomial
ring powers of prime ideals are not always primary. The polynomial ring
R “ krx, y, zs is given a grading by assigning the following degrees to the
variables: deg x “ 3, deg y “ 4 and deg z “ 5.
Consider the map φ : krx, y, zs Ñ krts defined by x ÞÑ t3 , y ÞÑ t4 and z ÞÑ t5 .
It preserves the degrees (when krts is given the usual grading with the degree
of t being one) hence the kernel p is a homogeneous prime ideal. The image of
φ is the ring krt3 , t4 , t5 s.
a) Show that the polynomials f “ xz ´ y2 , g “ x3 ´ yz and h “ yx2 ´ z2 all lie in
p.
b) Verify that the homogeneous polynomials of degree less than 8 in R are x,
x2 and xy, and prove that no element of p is of degree less than 8.
c) Show that g2 ´ f h is divisible by x and that u “ pg2 ´ f hqx´1 is homogeneous
of degree 15.
d) Conclude that p2 is not a primary ideal. Hint: xu P p2 but neither u nor
any power of x lies in p2 . M
178 ma4200—commutative algebra

8.2 The Lasker-Noether theorem

Minimal primary decompositions

8.12 Given a collection tSi u of set. It might very well happen that the inter-
Ş
section i Si does not change if one throws away one or more of the Si ’s (for
instance, if S1 Ď S2 , one stupidly has S1 X S2 “ S1 ) and in that case one says that
intersection is redundant. In the opposite case, that all the Si contribute to the Redundant intersections
Ş Ş
intersection, or in other words, when i Si Ř i‰ j Si for all j, the intersection (redundant snitt)

is called irredundant. Irredundant intersections

The superfluous sets of an intersection are precisely those S j such that (irreduntante snitt)
Ş
i‰ j Si Ď S j , and one may render the intersection irredundant by just discard-
ing those sets.
8.13 Now, let a be an ideal in the ring A, a primary decomposition is an ex- Primary decomposition
pression of a as a finite intersection of primary ideals; that is an equality like (primær dekomposisjon)

a “ q1 X . . . q r (8.1)

where qi ’s are primary ideals. We have already seen several examples (exam-
ples 8.1 and 8.2 above).
Without further constraints there are several trivial6 ways such a decom- 6
And it can be in non-
trivial ways too; we
position can be ambiguous. First of all, it can be an irredundant intersection.
shortly return to those.
Secondly a p-primary ideal can be the intersections of other p-primary ideals
in infinitely many different ways (see the upcoming example 8.3). The first
type of ambiguity is coped with by just discarding superfluous ideals, and
Proposition 8.7 above helps us coping with the second. We just group those
qi ’s with the same radical together, and replace them by their intersection,
which will be primary with the same radical.
The primary decomposition (8.1) is called minimal or reduced if all the Minimal or reduced
‘
radicals qi are different and the intersection is irredundant. primary decompositions
(minimale eller reduserte
primærdekomposisjoner)
Lemma 8.14 Any primary decomposition a “ q1 X . . . X qr can be rendered a
‘
minimal one; that is, an irredundant intersection with the radicals qi being distinct.

Example 8.3 Consider m2 “ px2 , xy, y2 q in krx, ys (where m “ px, yq). For all
scalars α and β with α ‰ 0 one has the equality

px2 , xy, y2 q “ px2 , yq X py2 , αx ` βyq.

Indeed, this amounts to the two lines generated by the class of y and the class
of αx ` βy in the two dimensional vector space m{m2 intersecting in 0. K

Example 8.4 If the ring A is a pid, there is nothing much new. The prime ide-
als are the principal ideals ppq generated by an irreducible p. The ppq-primary
ideals are those generated by powers of p; that is, those on the form ppv q. In
 primary decomposition 179

ν
general, if f “ p11 . . . prν is a factorisation of f into a product of irreducible
elements, the primary decomposition of p f q is unambiguous and it is given as

p f q “ pp1 qν1 X . . . X ppr qνr .

The same applies to principal ideals in any ufd; but be warned that not all ideals
are principal! K
Finally in this paragraph, we notice that by Proposition 8.9 a primary
decomposition localises well:
Proposition 8.15 Assume that S is a multiplicatively closed subset of A and that
a “ q1 X . . . X qr is a primary decomposition of a with radicals pi . Then a AS “
q1 AS X . . . X qr As , and either qi AS is primary with radical pi AS or qi AS “ AS

The resulting decomposition of a AS is by no means always irredundant

even if the one of a one starts with is. The primes pi meeting S blow up to
the entire ring AS . They will not contribute to the intersection and can be
discarded, and thus one may write
č
a AS “ qi A S .
SXpi “H

A particularily interesting case is to take S to be the complement of one of the

pi ’s, say pν . Then a AS “ qv AS , and a AS is primary in AS !

Existence of Primary Decompositions

In rings that are not Noetherian, ideals may or may not have a finite primary
decomposition; but in Noetherian rings they always have. The proof is appli-
cation of Noetherian induction (the principle of attacking a maximal crook!)

Proposition 8.16 In a Noetherian ring A any ideal a is the intersection of finitely

many primary ideals.

Proof: Since the ring A is assumed to be Noetherian, the set of ideals for
which the conclusion fails, if non-empty, has a maximal element a. Replacing
A by A{a we may assume that the zero ideal is not the intersection of finitely
many primary ideals (in particular it is not primary), but that all non-zero
ideals in A are.
In A there will be two elements x and y such that xy “ 0, but with x ‰ 0
and y not nilpotent. The different annihilators Ann yi form an ascending
chain of ideals, hence Ann yν “ Ann yν`1 for some ν. We contend that p0q “
Ann y X pyν q. Indeed, if a “ byν is an element in pyν q that lies in Ann y, one
has ay “ byν`1 “ 0 and therefore b P Ann yν`1 “ Ann yν , and it follows that
a “ byν “ 0. Now, x P Ann y is a non-zero element, and since y is not nilpotent,
both ideals pyν q and Ann y are non-zero and are therefore finite intersections
of primary ideals; the same is thus true for p0q. o
180 ma4200—commutative algebra

Theorem 8.17 (The Lasker-Noether theorem) Every ideal in a Noetherian

ring has a minimal primary decomposition.

Proof: Start with any decomposition of an ideal a into primary ideals (there
is at least one according to the proposition above). By Lemma 8.14 on page 178
it can be made minimal by regrouping ideals with the same radical and
discarding redundant ones. o

The First Uniqueness Theorem

There are two main uniqueness issues concerning primary decompositions.
One may ask if the radicals of the components are unique, or ask if the pri-
mary components themselves are unique. The first question is answered by
an unconditionally yes (well, the ring must be Noetherian), but to the second
the answer is no in general (also in Noetherian rings), although being a partial
yes.
8.18 A way to show that the radicals of the primary components of an ideal a
are invariants of the ideal, is to characterize them without referring to any of
the decompositions. The idea is to consider the collection of transporter ideals
pa : xq when x varies in A, and it turns out that the radicals of the primary
components of a are precisely the prime ideals among these.

Proposition 8.19 Let a be an ideal in a Noetherian ring A. The radicals that

occur in an irredundant primary decomposition of a, are precisely the prime ideals
among the transporter ideals pa : xq with x in A.

Passing to the quotient A{a and observing that pa : xq{a equals the annihilator
p0 : rxsq of the class rxs in A{a, the theorem has the equivalent formulation
(remember Proposition 8.11 on page 177) which is the one we shall prove:

Proposition 8.20 (Principle of annihilators) The radicals occurring in

an irredundant primary decomposition of the zero ideal p0q in a Noetherian ring A,
are those ideals among the annihilators Ann x that are prime.

Proof: Fix an irredundant primary decomposition of the zero ideal p0q.

There are two implications to prove. We begin with letting q be one of the
‘
components and letting p “ q denote the radical, and we aim at exhibiting
an x such that p “ Ann x. Denote by c the intersection of the other primary
components in the decomposition. Then c X q “ 0, but c ‰ 0 since the
decomposition is irredundant.
Let x P c be a non-zero element such Ann x is maximal among the annihi-
lators of non-zero elements of c. We contend that Ann x “ p, and begin with
showing the inclusion Ann x Ď p. Because x ‰ 0, it holds that x R q, and hence
xy “ 0 implies that y P p as q is p-primary.
 primary decomposition 181

In order to show the other inclusion pick an y P p and assume that xy ‰

0. Some power of y lies in q and therefore kills x. Hence there is a natural
number n so that yn x “ 0, but yn´1 x ‰ 0. By the maximality of Ann x it holds
true that Ann x “ Ann yn´1 x, and consequently y P Ann x, which contradicts
the assumption that xy ‰ 0.
For the reverse implication, assume that Ann x is a prime ideal. Let I be the
Ş
set of indices such that qi does not contain x. Then iP I qi Ď Ann x, because
č č č č č
x¨ qi Ď qi ¨ qi Ď qi X qi “ p0q.
iP I iR I iP I iR I iP I

Consequently it holds true that the product of appropriate powers of the

corresponding radicals pi is contained in Ann x. Since Ann x is supposed to
be prime, it follows that pν Ď Ann x for one ν P I. On the other hand, it holds
true that p0q “ x ¨ Ann x Ď qv from which ensues that Ann x Ď pv because qv is
pν -primary and x R pν . o

Theorem 8.21 (The First Uniqueness Theorem) The radicals occurring in

an irredundant primary decomposition of an ideal in a Noetherian ring are unambigu-
ously determined by the ideal.

8.22 The radicals of the primary components are of course tightly related to
the ideal, vaguely analogous to the prime factors of an integer, and they merit
a proper name. They are called the associated prime ideals of a, and the set they Associated prime ideals
constitute is denoted by Ass A{a. In particular, Ass A will be the set of prime (assosierte primidealer)

ideals associated to zero.

8.23 There are no inclusion relations between the components of an irredun-
dant primary decomposition (irredundant means precisely this), but that does
not exclude inclusion relations between the associated primes. In Example 8.1,
for instance, we found that px2 , xyq “ pxqXpx, yq2 with the associated primes
being pxq and px, yq. One distinguish between isolated and embedded associated Isolated associated
primes. The former are those being minimal in Ass A; that is, they do not primes (isolerte assosierte
primidealer)
contain any other associated prime, whereas the latter are those that do. In the Embedded associated
example above, pxq is an isolated prime whilst px, yq is embedded7 . primes (embeddede
Primary components with an isolated radical are called isolated components assosierte primidealer)
7
You might wonder
and those with an embedded radical are called embedded components. why t embedded com-
8.24 Early in the course, when discussing the radical of an ideal, we proved ponents since they are
‘ not contained in, but
that the radical 0 of A equals the intersection of all minimal primes in A on the contrary contain
‘ Ş
(Paragraph 2.61 on page 2.61); that is, 0 “ p, the intersection extending other associated primes.
The usage stems from
over the minimal elements of Spec A.
‘ geometry since inclu-
On the other hand, we just expressed the radical 0 as the intersection of sions between varieties
‘ Ş are the reversed of
the prime ideals minimal in Ass A so that 0 “ p where the intersection
those between ideals.
extends over the minimal elements in Ass A. When the intersections of two Isolated components
families of prime ideals are equal and there are no (non-trivial) inclusion (isolerte komponenter)
Embedded components
(embeddede komponenter)
182 ma4200—commutative algebra

relations between members of either family, the families coincide (Lemma 2.31
on page 34). Hence the sets Spec A and Ass A have the same minimal primes.
We have proved:

Proposition 8.25 In a Noetherian ring A the sets Spec A and Ass A have the
same minimal elements; in other words, the minimal primes of A are precisely the
isolated associated primes. In particular, there are finitely many minimal primes.

8.26 We have seen the intersection of the associated primes of A is the set of
nilpotent elements, and their union turns out to be the set of zero divisors:

Proposition 8.27 The set of zero-divisors in a Noetherian ring A equals the union
Ť
pPAss A p of the associated primes.

Proof: Let Ann z be maximal among the annihilators of non-zero elements in

A. Then Ann z is prime and hence an associated prime of A. Indeed, if xyz “ 0
and xz ‰ 0, it it ensues from the maximality of Ann z that Ann z “ Ann xz
because obviously Ann z Ď Ann xz. Hence y P Ann z, and as any annihilator
ideal is contained in a maximal one, we are through. o
Example 8.5 We offer one more example and consider the ideal

a “ px2 y, y2 z, z2 xq.

in the polynomial ring Crx, y, zs and aim at determining a primary decomposi-

tion.
To get an idea of where to start we resort to geometry, and take a look at
the zero-locus Vpaq inside C3 . It is given by x2 y “ y2 z “ z2 x “ 0, and is
easily seen to be the union of the three coordinate-axes. This means that there
must be some components supported along each axis, and no component
can be supported elsewhere. So let us consider the x-axis. Localizing at x, i.e.
passing to A x “ Crx, x´1 , y, zs and eliminating the two other axes, we see
that aA x “ py, y2 z, z2 q; so we suspect this to be one of the components; by
symmetry we find two more suspects, and in fact, it holds true that

px2 y, y2 z, z2 xq Ď py, y2 z, z2 q X px, x2 y, y2 q X pz, z2 x, x2 q.

This is however not the whole story. The element xyz lies in the intersection
to the right, but not in a. Now, clearly px, y, zq Ď pa : xyzq and px, y, zq being
maximal, it holds that pa : xyzq “ px, y, zq, so there must be an px, y, zq-primary
component. After a few tries (and a more failures) one finds the equality

px2 y, y2 z, z2 xq “ py, y2 z, z2 q X px, x2 y, y2 q X pz, z2 x, x2 q X px2 , y2 , z2 q.

The associated primes are pz, yq, px, yq, pz, xq and px, y, zq. Once one has
guessed correctly, it is relatively easy to check the answer. All involved ideals
are generated by monomials, and such ideals have the nice property that they
 primary decomposition 183

have a polynomial as member if and only if all the monomial terms of the
polynomial are members. Hence it suffices to check that every monomial lying
in the ideal to the right lies in the one to the left as well. But monomials in
px2 , y2 , z2 q have either x2 , y2 or z2 as a factor, and by symmetry we may as-
sume it is y2 . Lying in pz, z2 x, x2 q too, our monomial must have either z or x2
as a factor and thereby also zy2 or x2 y2 ; but these two both lie in a, and we are
done. K

The second uniqueness theorem

We have now to come the uniqueness issue for the primary component. Al-
ready our first example (Exampe 8.1 on page 175) showed that they are not
unique. We found that

px2 , xyq “ pxq X px2 , yq “ pxq X px, yq2 .

and both px2 , yq and px, yq2 are primary components. Notice that they have the
same radical (they must!) and that they are embedded components. Now, the
bad news are that px2 , xyq, have infinitely many different primary decompo-
sition (Example 8.6 below), but the good news are that merely the embedded
components differ. This is generally true. The Second Uniqueness Theorem
we are about to prove, states that isolated components are unique. The point
is that the isolated components may be expressed in terms of the correspond-
ing associated primes which are independent of the decomposition. We shall
see that if q is the component and q “ p it holds that q “ ι´1 pa Ap q where
‘

ι : A Ñ Ap Ñ is the localization map, or simply that q “ A X a Ap when ι is

injective and suppressed from the notation.

Theorem 8.28 (The second uniqueness theorem) The isolated compo-

nents of an ideal a in a Noetherian ring A are unambiguously defined by a.

Proof: We shall concentrate on one of the isolated associated prime ideals

p of a, but the main player will be a component q that belongs to one of the
minimal primary decompositions of a and has radical p.
The salient point is the equality

q “ ι´1 pa Ap q, (8.2)

from which the theorem ensues since the isolated components are invariants
of a.
Ş
To establish (8.2) express the decomposition of a as a “ q X i qi where the
intersection extends over the primary components different from q. Localizing
at p one finds
č
a A p “ q A p X qi A p “ q A q (8.3)
i
since the qi ’s blow up when localized; that is, qi Ap “ Ap . Indeed, since p is
184 ma4200—commutative algebra

isolated, pi Ę p holds for all i. In each pi we may therefore find elements s

not belonging to p, and as a power of s lies in qi , we infer qi blows up. Taking
inverse images on both sides of (8.3) and citing Proposition 8.9 on page 177 we
conclude that ι´1 pa Ap q “ q.
Example 8.6 The equality

px2 , xyq “ pxq X px2 , xy, yn q

holds true in the polynomial ring krx, ys for any natural number n, and is an
example of infinitely many different minimal primary decompositions. Indeed,
the Ď -inclusion is obvious, and to check the other, assume that a belongs to
the right side. Then
a “ p ¨ x “ q ¨ x2 ` r ¨ yn ` sxy

with p, q r and s polynomials in krx, ys. It follows that x divides r, and hence
that f P px2 , xyq. K

Example 8.7 So far all our examples have merely involved monomial ideals,
but of course most idelas are not shaped like that. Primary decompositions are
notoriously strenuous to lay hands on, and the monomial ideals are among
the easiest to handle, hence their tendency to appear in texts. However, we
are obliged to give at least one example of a more mainstream situation. It
illustrates as well that the decomposition is largely of a geometric nature; that
is, at least the isolated associated prime ideals are; the primary components
may conceal subtler structures.
We shall analyse the familiar case of the intersection of two quadratic
curves; the unit circle centred at p0, 1q and a standard parabola. So let a “
px2 ` py ´ 1q2 ´ 1, y ´ x2 q in krx, ys, where k is any field of characteristic
different from 2. A standard manipulation shows that the common zeros
of the two polynomials are the points p1, 1q, p´1, 1q and p0, 0q, and the same
manipulations give

a “ px2 ` py ´ 1q2 ´ 1, y ´ x2 q “ px2 px2 ´ 1q, y ´ x2 q.

Any prime ideal p containing a must contain either x, x ´ 1 or x ` 1. It contains

y if x lies in it, and because y ´ x2 “ y ´ px ` 1qpx ´ 1q ´ 1, one has y ´ 1 P p in
the two other cases. We conclude that the px, yq, px ´ 1, y ´ 1q and px ` 1, y ´ 1q
are the only prime ideals containing a; and since they all three are maximal,
the associated primes are found among them, and there can be no embedded
component.
To determine the primary components of a, we localize (as in Theorem 8.28
on the previous page). In the local ring A “ krx, yspx`1,y´1q where both x and
x ´ 1 are invertible, we obtain the equality

a A “ px2 ´ 1, y ´ x2 q “ px ` 1, y ´ x2 q “ px ` 1, y ´ 1q.
 primary decomposition 185

In similar fashion, in B “ Crx, yspx´1,y´1q both x and x ` 1 are invertible and

one has
aB “ px2 px2 ´ 1q, y ´ x2 q “ px ´ 1, y ´ 1q.

Finally, in C “ krx, ysp x, yq both x ` 1 and x ´ 1 have inverses, and we see that

aC “ px2 , yq.

Since there are no embedded components, we conclude that

a “ px ´ 1, y ´ 1q X px ` 1, y ´ 1q X px2 , yq.

When the characteristic of k equals two, there things evolve in a slighty

different manner. In that case, then two ideals px ´ 1, y ´ 1q and px ` 1, u ´ 1q
conicide and x2 ` 1 “ px ` 1q2 . We find

a “ ppx ` 1q2 , y ´ 1q X px2 , yq.

Problems

8.5 Show that for any scalar a P k it holds true that ˇ

px2 , xyq “ pxq X px2 , y ` axq

and that this is a minimal primary decomposition. Show that different scalars
a give different decompositions.

8.6 Determine a minimal primary decomposition of px3 , y2 x2 , y3 xq. ˇ

8.7 Let a be the ideal in the polynomial ring krx, y, zs given as a “ pyz, xz, xyq. ˇ
Show that the minimal primary decomposition of a is shaped like

pyz, xz, xyq “ py, zq X px, zq X px, yq.

Show that the maximal ideal px, y, zq is associated to a2 and determine a

minimal primary decomposition of a2 . Hint: Consider pa : xyzq.

8.8 Let p be the ideal in the polynomial ring krx1 , . . . , xn s over a field k gener- ˇ
ated by the r first variables; that is, p “ px1 , . . . , xr q. Show that every power pm
is p-primary. Hint: Consider pm Krx1 , . . . , xr s where K “ kpxr`1 , . . . , xn q and
show that pm Krx1 , . . . , xr s X krx1 , . . . , xn s “ pm .

8.9 Let k be a field. Let ps “ px1 , . . . , xs q in krx1 , . . . , xn s be the ideal generated ˇ

by the s first variables. Consider

a “ p1 p2 p3 . . . pr “ px1 qpx1 , x2 qpx1 , x3 , x3 q . . . px1 , . . . , xr q.

186 ma4200—commutative algebra

Prove that the following equality holds true

a “ p1 X p22 X . . . X prr´ 1 r
´1 X p r

and is a minimal primary decomposition of a. Hint: Show that if c is any

ideal generated by monomials of degree s ´ 1 in x1 , . . . , xs , then cps “ c X pss ;
then use induction on r.

8.10 (Symbolic powers.) We have seen that the powers pn of a prime ideal p
in A are not necessarily p-primary (unless p is maximal). But there is an in
some sense canonical primary ideal associated to the powers pn ; the so-called
symbolic power ppnq . They arise in the following way. The ideal p Ap is maximal Symbolic powers
in the local ring Ap and its powers are therefore primary by Proposition 8.4 (symbolske potenser)

on page 175. Pulling primary ideals back along the localization map ι results
in primary ideals (Proposition 8.9 on page 177), the ideal ppnq “ ι´1 pn Ap
(or when ι is injective ppnq “ A X pn Ap ) will be primary, and this is the n-th
symbolic power of p.
n`m
a) Show that if n and m are natural numbers it holds that ppnq ¨ ppmq Ď p
b) Show that pn “ ppnq if and only if pn has no embedded component.
c) With the notation as in Example 8.2 on page 175, let p “ px, yq and
determine the symbolic square pp2q .

8.3 The homogeneous case

It is of great interest to know that the subsidiary ideals of a homogeneous
ideal arising from a minimal primary decomposition are homogeneous. This
not so much because of most ideals one meets in examples or exercises be-
ing homogeneous, but because the so-called projective varieties, which are
ubiquitous in algebraic geometry, are defined by homogeneous ideals.
In what follows, we shall se that the associated primes and the isolated
component of homogeneous ideals are homogeneous. When it comes to the
more elusive embedded components, all we can hope for is that they may be
chosen to be homogeneous, and fortunatly this is the case as well, and the
Lasker–Noether theorem is fully valid in a grade context.

Preparitions

8.29 We start with a little lemma (in fact, we already met a version of it in
Exercise 1.15 as early as on page 17).
Lemma 8.30 Let R be a graded ring and assume that x and y are two elements such
that x ¨ y “ 0. Let xe be the lowest homogeneous term of x. Then xeν y “ 0 for some ν.
In particular xeν yi “ 0 for each graded part yi of y.
 primary decomposition 187

Proof: Let x “ xe ` . . . ` xn and y “ y f ` . . . ` ym respectively be the

expansions of x and y in homogeneous components. We shall show that
xeν`1 ¨ y f `ν “ 0, and letting ν “ m ´ f we see that the proposition ensues
from this. As xe y f is the term of xy of lowest degree, it holds that xe y f “ 0,
and the induction can start. Expanding the product xy one finds that the
homogeneous component of degree e ` f ` ν is given as

xe ¨ y f `ν ` xe`1 ¨ y f `ν´1 ` . . . ` xe` j ¨ y f `ν´ j ` . . . ` xe`ν ¨ y f “ 0.

Multiplying through by xeν`1 gives xeν`1 ¨ y f `ν “ 0 because by induction it holds

true that
j ν ´ j `1
xeν`1 ¨ y f `ν´ j “ xe ¨ xe ¨ y f `ν´ j “ 0.
o
8.31 The first step in establishing the graded Noether–Lasker Theorem is to
see that associated primes to homogeneous ideals are homogeneous.

Proposition 8.32 Let R be a graded ring and let p be a prime ideal associated to
the homogeneous ideal a. Then p is homogeneous.

Proof: Replacing R by R{a, we may assume that a “ 0. Let p “ p0 : aq and

assume that x P p0 : aq. If xe is the homogeneous part of x of lowest degree, the
lemma tells us that xeν kills a. Hence xev P p0 : aq and by consequence xe P p0 : aq
as p0 : aq is a prime ideal. Noticing that x ´ xe also lies in p0 : aq, we may finish
off the proof by induction on the degree of the lowest non-vanishing term of x.
o
8.33 Next step is to treat the primary components, and as alluded to, a em-
bedded components are not unique they are therefore not forced to be homo-
geneous, but they can be chosen to be homogeneous.For any ideal a, we shall
denote by a7 the largest homogeneous ideal contained in a; that is, the one
generated by all homogeneous elements belonging to a.

Lemma 8.34 Let q be primary ideal in the graded ring R whose radical is homoge-
neous. Then q7 is primary and
‘ 7 ‘
q “ q.

Proof: We start out by proving that q7 has q as its radical. Obviously

‘
‘ 7 ‘ ‘
q Ď q. Attacking the other inclusion, we pick a member x P q. Since
‘ ‘
q is homogeneous all the homogeneous parts of x lie in q, and we can
safely assume x to be homogeneous. Since x ν P q for some ν, and any homo-
geneous member of q belongs to q7 , it follows for free that x ν P q7 and we are
through.
Next, suppose that xy P q7 but that y R q7 . The task is to show that x P q7 ;
‘
‘
that is, x P q since the two radicals coincide. If y f is the homogeneous term
in y of lowest degree, we may assume that y f R q7 , since if y f lay in q7 , we
might replace y by y ´ y f , and repeating this procedure if need was, we would
188 ma4200—commutative algebra

finally end up with an element whose term of lowest degree does not belong
to q7 . With y f in place outside q7 , it follows that y f does not belong to q, hence
x P q “ q7 .
‘ ‘
o

8.35 Finally we have come to the graded version of the Lasker–Noether

theorem; needless to say, the two uniqueness theorem persist unchanged, they
can of course be applied to any minimal primary decompositions.

Proposition 8.36 (Graded Lasker–Noether Theorem) A homogeneous

ideal in a Noetherian graded ring has a minimal primary decomposition with all
components being homogeneous, and all its associated primes are homogeneous.

Proof: Observe first that all prime ideals associated to a homogeneous ideals
are homogeneous. (Proposition 8.32 above).
It is fairly clear that pa X bq7 “ a7 X b7 (the homogeneous elements in a X b
are the honogenous elements the lie in both a and b!!) so starting out with a
Ş
minimal primary decomposition a “ i qi and applying the 7-construction to
it, one arrives at a decomposition
č 7
a “ a7 “ qi , (8.4)
i

and according to Lemma 8.34 on the preceding page, this is a primary decom-
position. Moreover, the radicals of the sharpened ideals q7i are the same as
the radicals of the qi ’s, and we can conclude that (8.4) is a minimal primary
decomposition. o

Arriving by group actions

The stage in this paragraph is confined to ideals in the polynomial ring over
the complex numbers C. The r-dimensional torus C˚ qr acts on x “ px1 , . . . , xr q
by homothety in the variables; that is x t “ pt1 x1 , . . . , tr xr q and this action
induces an action on the polynomial ring f t pxq “ f ptxq.
Recall we saw in xxx that an ideal a was homogenous if and only if it is
invariant under this action. Obviously qt is pt -primary if and only if q is p-
primary. Hence
č
a “ at “ qit
i

is a primary decomposition, and is obviously the radical of qt . We conclude

pt
that the associated primes are permuted by G, bu they are finite in number
and as G has no finite quotient, each is is invariant under G and hence is
homogeneous. Since the isolated components are unique, each of those is
invariant as well and is homogenous. As to the embedded components, let q
be one. After exercise xxx , the intersection tPG qt is p-primary, and hence a
Ť

component of a.
 primary decomposition 189

8.4 Primary decomposition of modules

The primary decomposition of ideals is easily generalised to modules. In
a Noetherian module, any submodule has a primary decomposition, with
analogous uniqueness properties as in the ideal case. The proofs are basically
the same but with the necessary changes. In this we sketch the development
but leave most of the details to the students.

Associated prime ideals

Generalizing the concept of associated primes for rings, we say that a prime
ideal p is associated to the module M if it is the annihilator of an element x in Prime ideals associated
the M; that is p “ p0 : xq, and in concordance with the notation for rings we to modules (primidealer
assosiert til moduler)
denote the the set of associated primes of M by Ass M . We proved in chapter
7 that the ideals maximal among the annihilators of non-zero elements from M
are prime ideals (Proposition 7.18 on page 157), so for Noetherian modules we
infer that Ass M is non empty.
Proposition 8.37 Let A be a ring and M a Noetherian A-module. Then Ass M is
finite and non-empty. The minimal primes in Ass M and Supp M coincide.

Proof: This follows from The structure theorem on page 158. o

‘
The radical p0q “ t x P A | x acts nilpotently on M u is an ideal, and is a
prime ideal if M is primary. Indeed, if pxyqν kills M and no power of y does,
there is a z P M so that yn z ‰ 0; and since x n pyn zq “ 0, it follows that x is
nilpotent as M is primary
Proposition 8.38 Every Noetherian module has a primary decomposition
Proof: Let N be the maximal rascal, and replace M by M{N. That is M is
not primary. Hence xz “ 0 but the homothety by x is not nilpotent. Let
kerrxi s Ď kerrxi`1 s is an ascending chain, hence stabilizes at certain point, say
ν. We contend that p0q “ kerrxs X x ν M. Indeed, if xz “ 0 and z “ x v w it follows
that w P kerrxsv`1 , hence in kerrxsv . By consequence z “ x ν`1 w “ 0. We o

Primary modules
Given a submodule N of the module M, one has the transporter ideal pN : Mq
consisting of ring elements that multiply M into N; that is

pN : Mq “ t x P A | xM Ď N u,
‘
and the radical pN : Mq is called the radical of N relative to M. The elements
are the ring elements such a power multiplies M into M. If N “ 0, they consist
of the ring elements inducing a nilpotent homothety on M.
and one extends the notion of primary ideals to modules in the following
fashion:
190 ma4200—commutative algebra

o The homothety by x on M{N is either injective or nilpotent.

Or in other words

o if xz P N then either z P N or xN P pN : Mq for some n

Proposition 8.39 Let N Ď M be a module and a submodule.

o If N is primary, the radical pN : Mq is a primary ideal and the consequently the

‘
radical p “ pN : Mq is a prime ideal. One says that N is p-primary.

o Finite intersections of p-primary submodules are p-primary.

o If N is p-primary and S X p “ H, then NS is pAS -primary submodule of MS .

o Assume that L Ď N. Then N is p-primary, then N{L is a p-primary submodule of

M{L.
Lecture 9

Integral extensions

Preliminary version 1.1 as of 2018-11-19 at 16:45 (typeset 3rd December 2018 at 10:03am)—For the
moment nothing here!! Prone to misprints and errors and will change.
Completely rewritten and extended a lot. For the moment not many exercises or examples—
hopefully this will follow soon. Also probably peppered with misprints and impressions (sorry,
improvements will shortly follow) and organisation will also be better soon.
19/11/2018 Corrected som mistakes and imoroved the text many places.

From an algebraic point of view there is a huge difference between the ring
of integers Z and the field of rationals Q; we need only mention the primes.
They are visible in Z as generators of the prime ideals, but in Q they are, at
least from an algebraic point of view, on an equal footing with the other units.
When the exploration of number fields1 begun, an immediate want arose 1
That is, finite field
to have subrings playing the role of the integers and where the deep secrets extensions of the
rationals
of the field could be revealed. These rings was made up of the “integral
elements” in the field, or more precisely those being integral over Z.
In algebraic geometry these rings give rise to what is called normal varieties
where the geometry of the codimension one subvarieties strongly influence the
geometry of the entire space.

9.1 Definition and basic properties

Throughout this section the setting will be an extension of rings A Ď B; that is,
a pair of a ring and subring. An element x P B is said to integral over A if it Integral elements (hele
satisfies a monic relation as elementer)

x n ` an´1 x n´1 ` . . . ` a1 x ` a0 “ 0, (9.1)

where the coefficients ai are members of A. It is all-important that the leading

coefficient be one, so there is heavy stress on the word monic. This distin-
guishes integral elements from their cousins the algebraic elements, which
satisfy similar equations, but without constraints on the leadings coefficient;
of course, if the leading coefficient is invertible in A it goes for the same. A
relation like (9.1) is called an integral dependence relation for x over A. Integral dependence
relation (helavhenighetsre-
lasjon)
192 ma4200—commutative algebra

9.1 If all the elements in B are integral over A one says that B is integral over
A or that B is an integral extension of A. The subset of B consisting of the Integral extension (hel
elements integral over A is called the integral closure of A in B and is denotes utvidlese)
Integral closure (helavs-
by A.
s Of course it depends on B, but to keep notation simple, we do not
lutningen)
include a reference to B in the notation—the context will make it clear where
the integral closure is taken.
It is a basic, but subtle fact that the integral closure is ring, which we shortly
shall prove. Finally, one says that A is integrally closed in B if A “ A; s that is, Integrally closed (helavs-
luttet)
every element which is integral over A belongs to A.
9.2 Both in algebraic geometry and algebraic number theory the integral clo-
sure of a domain A in its field of fractions is a frequently used construction
and we shall use the notation A r for it to distinguish it from all the crowd. Do-
mains being integrally closed in their field of fractions; that is, those satisfying
r are called normal, and for general domians A
A “ A, r is sometimes called the Normal rings (normale
normalization of A. ringer)
Normalization (normalis-
9.3 To illustrate the difference between algebraic an integral dependence ering)
relation consider the two simple equations

y2 ´ z “ 0
pz ´ 1qy2 ´ z “ 0
‘
over the complex numbers. The first one “has z as a solution”, but due to the
ambiguity of the square root, it is impossible to find a continuous (yet alone
analytic) solution in the entire plane Only in simply connected domain not
containing the origin can a continuous solution be found. The solutions of the
second equation suffer the same defect, but additionally they acquire a pole at
z “ 1. The difference between solutions of algebraic and integral relations is
precisely the occurrence of poles in the former.
Example 9.1 The golden section p1 ` 5q{2 is integral over Z as it satisfies the
‘

equation
x2 ´ x ´ 1 “ 0.
The integral closure of Z in Qp 5q equals Zrp1 ` 5q{2s. Indeed, a number
‘ ‘
‘
a ` b 5 has the minimal equation

x2 ´ 2ax ` pa2 ´ 5b2 q “ 0,

and if it is integral over Z, the coefficients are integral by Gauss lemma (Exer-
cise 3.1 on page 64). Then n “ 2a P Z and 20b2 P Z, and hence m “ 2b P Z as
well. Substituting back, gives 0 ” pn2 ´ 5m2 q ” pn2 ´ m2 q mod 4, which holds
if and only if m and n have the same parity. K
Example 9.2 The ring of integers in the number field Qpi 5q equals Zri 5s.
‘ ‘

Indeed, the minimal equation of an element a ` ib 5 P Qpi 5q is

‘ ‘

x2 ´ 2ax ` a2 ` 5b2 “ 0.
 integral extensions 193

If x is integral over Z, the coefficients are integral, and as in the previous

example, this entails that n “ 2a and m “ 2b are integral. Substituting back
gives 0 ” n2 ` 5m2 ” m2 ` n2 mod 4, which occurs only if both n and m are
even (squares are either equal to 1 or 0 mod 4). Hence a and b are integers.
The difference between the ring of integers in these two example illustrates
a general phenomenon. If d square free, the ring of integers in Qp dq equals
‘

Zrp1 ` dq{2s when d ” 1 mod 4 and equals Zr ds else.

‘ ‘
K

The basic properties

9.4 If x is an element of B which is integral over A, the subring Arxs of B

obtained by adjoining the element x to A is a finitely generated A-module. An
integral dependence relation as (9.1) above, entails that

x n “ ´pan´1 x n´1 ` . . . ` a1 x ` a0 q,

and a straightforward induction yields that Arxs is generated by the n first

powers of x. The converse is also true, as shown in the next proposition.
When the A-module Arxs is Noetherian, this comes almost for free; one just
considers the ascending chain Mi “ p1, x, x2 , . . . , xi q of submodules of Arxs,
and at the point where it stabilises; that is, when Mν`1 “ Mν , one obtains an
integral dependence relation for x since x ν`1 P Mν . For the general proof a
little twist is needed.

Proposition 9.5 (Basic characterisation) Let A Ď B be an extension of

rings and x P B an element. The following three statements are equivalent:

o The element x is integral over A;

o Arxs is a finite A-module;

o There is faithful Arxs-module which is finite over A.

Proof: The only implication that shows any substantial resistance is that the
last statement entails the first. So let M be a module as in the last assertion
and let m1 , . . . , mn be generators for M over A. One may express each element
x ¨ mi in terms of the m j ’s, and this gives relations

x ¨ mi “ f ij m1 ` . . . f in mn

for 1 ď i ď n where each f ij P A. We introduce an n ˆ n-matrix Φ by letting

Φ “ x ¨ In ´ p f ij qij where In is the identity matrix2 . The equations above 2
For n “ 3 the matrix Φ is
then translate into Φ ¨ m “ 0 with m “ pm1 , . . . , mn q. Hence det Φ “ 0 shaped like
¨ ˛
x ´ f 11 f 12 f 13
by the determinantal trick (Lemma 9.47 on page 209). But developing the ˝ f 21 x ´ f 22 f 23 ‚
determinant shows that detpx ¨ In ´ p f ij qq is a monic polynomial in x whose f 31 f 32 x ´ f 33
coefficients lie in A; that is, det Φ “ 0 is an integral dependence relation for x
over A. o
194 ma4200—commutative algebra

We notice the immediate corollary—which may also easily be proven ad hoc—

that all elements in Arxs are integral over A when x is. An Arxs module M
which is finite over A, will be faithful over any subring of Arxs, in particular
over Arzs for any z P Arxs.

Corollary 9.6 If x is integral over A, all elements in Arxs are integral over A.

9.7 There is a close relationship between integral and finite extensions as was
unveiled in the previous proposition. Finite extensions are integral, but in gen-
eral the converse is not true. There are even examples of Noetherian domains
whose normalizasion A r is not a finite module over A; they are however, rather
exotic creatures, and the lions share of the rings appearing in mainstream
algebraic geometry—that is, domains finitely generated over field and their
localizations—have normalizations which are finitely generated as modules.
9.8 The first conclusion to be drawn from the Basic Characterization is that
finitely generated algebras which are integral, are finitely generated modules;
an important observations since the integral closure being finite over A or not,
is an issue. One has

Proposition 9.9 Let B be an integral ring-extension of A. Any subalgebra C of B

which is a finitely generated algebra over A, is a finite A-module.

Proof: The poof goes by induction on the number of generators of C. So let

C1 Ď C be a subalgebra generated over A by the same elements as C but one.
By induction C1 is finite over A, and C “ C1 rxs with x P C the last generator.
The element x being integral over A is even so more over C1 . Hence C is finite
over C1 by Proposition 9.5, and because the property of being a finite extension
is transitive in towers (Lemma 9.48 on page 209 in the appendix), it holds that
C is finite over A. o

Corollary 9.10 (Transitivity) Assume that A Ď B Ď C are ring-extension and

that B is integral over A. Then every element in C which is integral over B is integral
over A.

Proof: Let x be an element in C which is integral over B and satisfies the

dependence relation

x n ` b1 x n´1 ` . . . ` bn´1 x ` bn “ 0. (9.2)

with the coefficients bi lying in B. We let D be the sub A-algebra of B the bi ’s

generate. Then x is integral over D (the relation (9.2) has coefficients in D) and
consequently Drxs is a finite module over D. Now, D is a finite module over A
after the Proposition and therefore Drxs is finite over A as well. Hence we can
conclude by the Basic Characterization of integral elements. o
 integral extensions 195

9.11 It is by no means obvious how to deduce a dependence relation for

a product (or for a sum) from a dependence relation for the factors (or the
addends); that the integral closure is a ring, is a slightly subtle property.
However, once the Basic Characterisation (Proposition 9.5 on page 193) is
in place, it follows readily. For a different approach, see Problem 9.4 on the
following page.
Proposition 9.12 (The integral closure is a ring) If x and y are ele-
ments in B both integral over A, then their sum and product are integral over A as
well. The integral closure A
s of A in B is a subring of B. The integral closure A
s is
integrally closed in B.

Proof: This is just a combination of Corollary 9.6 and the transitivity property
(Corollary 9.10). Indeed, let x and y be integral over A. The ring A[x] is an
integral extension of A and y being integral over A, the extension Arx, ys
is integral over Arxs. Hence Arx, ys is integral over A by the transitivity. In
particular, this applies to both the product x ¨ y and the sum x ` y.
The last statement of the proposition might appear as a tautology, but an
argument is in fact needed. We have to see that elements integral over A s are
integral over A, which is exactly what Corollary 9.10 above tells us since A s is
integral over A. o

Integral extensions, localization and quotients

Integral extensions are well behaved and compatible with the formation of
localizations and quotients.
9.13 We treat the loacalizations first:
Proposition 9.14 Let S Ď A be a multiplicatively closed subset and assume that
B is an integral extension of A. Then BS is an integral extension of AS . Forming the
integral closure commutes with localization; i.e. holds that pA
Ęq “ p Aq
S
s S.

Proof: Let xs´1 be an element in B with x P A and s P S. All elements of B

are integral over A, so x satisfies an integral dependence relation shaped like

x n ` a n ´1 x n ´1 ` . . . ` a i x i ` . . . ` a 0 “ 0

with the ai ’s lying in A. Multiplying through by s´n we find the relation

pxs´1 qn ` an´1 s´1 x n´1 s´pn´1q ` . . . ` ai s´n`i pxs´1 qi ` . . . ` a0 s´n “ 0, (9.3)

which is a monic equation whose coefficients lie in AS and hence is an integral

dependence relation for xs´1 over AS .
For the second statement, the inclusion p Aq Ę ensues from the first
s S Ď pAq S
claim, and it suffices to prove that pAqS Ď p AqS . To that end, assume that
Ę s
xs´1 P pAq
Ę . It satisfies an integral dependence relation as
S

pxs´1 qn ` bn´1 pxs´1 qn´1 ` . . . ` bi pxs´1 qi ` . . . ` b0 “ 0, (9.4)

196 ma4200—commutative algebra

where each bi lies in AS and hence may be written as bi “ ai t´1 with ai P A

and t P S (extending the fractions, we may use a common denominator for all
the bi ’s). Multiplying (9.5) through by tn yields the relation

ptxs´1 qn ` tan´1 ptxs´1 q´pn´1q ` . . . ` tn´i´1 ai ptxs´1 qi ` . . . ` tn´1 a0 “ 0, (9.5)

and it follows that tx is integral over A. We conclude that xs´1 “ ptxqs´1 t´1 P
p Aq
s S. o

9.15 Next comes the quotients, and in addition to the usual the staging of this
chapter with an integral extension A Ď B, an ideal b in B is given. We let a be
the ideal b induces in A; that is, a “ A X b. Then A{a Ď B{b is an extension, and
it persits being integral.

Proposition 9.16 Let b Ď B be an ideal and let a “ b X A. If B is integral over A,

then B{b is integral over A{a.

Proof: Let x P B{b and let y P B be an element that maps to x. By assumption

it is integral over A and there is therefore a relation

yn ` an´1 yn´1 ` . . . ` a1 y ` a0 “ 0

with the ai ’s from A. Reducing that relation modulo b gives the relation

x n ` ran´1 sx n´1 ` . . . ra1 sx ` ra0 s “ 0,

where rai s as usual denotes the classes of ai in A{a. Hence x is integral over
A{a. o

Problems

9.1 Let B be a ring and tBi uiP I a family of subrings of B. If each Bi is inte-
Ş
grally closed in B, then the intersection iP I Bi is integrally closed as well.

9.2 Assume that tBi uiP I is a family of subrings of a field K. Prove that if each
Ş
Bi is integrally closed in K, then the intersection iP I Bi is as well.

9.3 Show that x is integral over A if and only if there is a square matrix with
coefficients in A having x as an eigenvalue.

9.4 Show that if x and y are eigenvalues for Φ and Ψ then x ¨ y is and eigen-
value for the Kronecker product ΦbΨ and that x ` y is one for the matrix
Φb Im ` In bΨ where the In are Im are identity matrices of appropriate size.
Conclude that the integral closure is a ring.

M
 integral extensions 197

Being normal is a local property

As mention in the introduction to this section, a particular important situation
is when A is a domain and B “ KpAq is the field of fractions of A. Recall that
the integral closure of A in KpAq is called normalization of A, and in case A is Normalizations (normalis-
integrally closed in KpAq, one says that A is a normal domain. eringer)
Normal domains (nor-
9.17 Being normal is a local property. male områder)

Proposition 9.18 (Being normal is local) Assume that A is a domain.

Then A is normal if and only if Am is normal for all maximal ideals m in A. It is also
equivalent to Ap being normal for all prime ideals p in A.

Proof:
Notice first, that all the localization Am have K as fraction field as well.
Consider the inclusion A ãÑ A, r which fits into the short exact sequence

0 /A /A
r / A{A
r /0

of A-modules. Because localization is an exact functor, it gives rise to the

following short exact sequence

0 / Am / p Aq
rm / p A{Aq
r m
/0

when localized at a maximal ideal m. According to Proposition 9.14, forming

integral closures commute with localization, so it holds that p Aq
r m “ pA
rm q.
Thence p A{Aqm “ p Am q{Am , and the claim follows by the Localness of Being
r r
Zero (Proposition 6.62 on page 144). o

9.19 In the case Ar is finite over A, the quotient A{A

r is finitely generated over
A as well, and it ensues that Supp A{ A is a closed subset of Spec A equal
r
to VpAnn A{ Aq.
r And localizing at p0q, or equivalently tensorizing by K, we
see that A{Ab A K “ 0 because A and A
r r both have K as fraction field. Hence
A{A
r is not of global support. So in that good case, when A r is finite over A,
for "most" primes p the local ring Ap is normal;that is, for primes in an open
non-empty subset of Spec A.

9.2 Examples
We indulge ourselves in two examples of rather large classes of rings that
are normal. Firstly, the unique factorization domains are always normal, and
secondly, there is a principle that rings of invariant of the actions of a finite
groups are normal, at least when the ring upon which the group act is normal.
This class includes the so-call "quotient singularities". We shall treate a few
simple examples in detail but leave the general case to the zealous students in
the form of a guided exercise.
198 ma4200—commutative algebra

9.20 We start out with the ufd-case.

Proposition 9.21 If the domain A is a ufd, then A is normal.

Proof: Let K be the fraction field of A, and let z “ x{y P K be an element

which is integral over A, and we may assume that x and y are without com-
mon factors. The element z being integral means there is a relation

z n ` a n ´1 z n ´1 ` . . . ` a i z i ` . . . ` a 0 “ 0

with the ai ’s lying in A. Multiplying through by yn and moving the leading

term to the left, gives

´x n “ an´1 x n´1 y ` . . . ` ai xi yn´i ` . . . ` a0 yn .

Every irreducible factor of y divides the right side, hence it divides the left
side and consequently also x. Contradiction. o

Example 9.3 Integral closures and normalizations play an important role in

algebraic geometry which is particularly accentuated in the theory of curves.
To illustrate this, we let C Ď C2 be the plane curve parametrized by x “ t2 ´ 1
and y “ tpt2 ´ 1q. It is easily seen to satisfy the equation y2 “ x2 px ` 1q; indeed,
py{xq2 “ t2 “ x ` 1.
The real points of the curve is depicted in the margin. For points on C
with x ‰ 0 the corresponding parameter value is uniquely defined, and
the parametrization is one-to-one away from the origin. However, the two
parameter values t “ ˘1 both give the origin, and the curve has a double point
there.
The parametrization may be though of as the map Spec Crts Ñ Spec Crx, ys
induced by the ring-map Crx, ys Ñ Crts that sends x Ñ t2 ´ 1 and y Ñ tpt2 ´ 1q;
or in the language of varieties, it is the the map C Ñ C “ Vpy2 ´ x2 px ´ 1qq Ď C2
sending t to the point pt2 ´ 1, tpt2 ´ 1qq.
This leads to considering the subring A “ Crt2 ´ 1, tpt2 ´ 1qs Ď Crts. The
point of the example is that t is integral over A; indeed, almost tautologically it
satisfies the equation
X 2 ´ t2 “ 0,

and t2 “ pt2 ´ 1q ` 1 P A.
Moreover, the ratio between the two generators of A equals t, so that its
fraction field is Cptq. Now, the polynomial ring Crts being a ufd is normal,
and we can conclude that Crts equals the normalization of A. It is typical for
curves that their normalisation “resolves the singularities”; that is, it separates
the different branches of the curve passing through the double points (or
points of of higher multiplicity). K

Abraham Seidenberg
(1916–1988)
American mathematician
 integral extensions 199

Rings of invariants

9.22 Now comes the promised result on rings of invariants, and as promised,
we shall proceed rather in relaxed way merely treating the simples possible
case. That is, the case of a cyclic group of order two acting on a normal do-
main B. Such an action is just given by an involution on B; in other words, a Involutions (invo-
ring map σ : B Ñ B with σ2 “ idB . The map σ extends to an involution of lusjoner)

the fraction field K of B by the obvious assignment σpxy´1 q “ σpxqσpyq´1 .

Furthermore, we let A “ Bσ “ t x P B | σpxq “ x u be the ring of invariants and
L its field of fractions. In this setting we have

Proposition 9.23 It holds true that L “ K σ “ t z P K | σpzq “ z u. Moreover B is

integral over A and if B is normal, A will be normal.

Proof: Clearly L Ď K σ . If σpxq{σpyq “ x{y it holds that yσpxq “ xσpyq, and

we may write x{y “ σpxqx{σpxqy with both σpxqx and σpxqy being invariant.
Hence L “ K σ .
Any x P B satisfies the relation

Kσ “ L Ď K
x2 ´ pσpxq ` xqx ` σpxqx “ 0. (9.6)

Ď
Bσ “ A Ď B
Both σpxq ` x and xσpxq are invariant under σ and belong therefore to A, hence
(9.6) is an integral dependence relation for x over A.
Finally, as B is integral over A, the integral closure of A in K equals B
r by
transitivity, and hence A “ B X L. From this ensues that A “ B X L “ A in case
r r r
B “ B.r o

9.3 The Cohen–Seidenberg Theorems

There is cluster of results proven by I. S. Cohen and A. Seidenberg dubbed
“going–up”, “going–down” and “lying–over” by Cohen and Seidenberg. They
all relate prime ideals in one ring to prime ideals in another ring which is
integral over the first. So let A Ď B be the two rings. Every prime ideal q Ď B
intersects A in a prime ideal φ “ A X q, and this is sets up the canonical map
Spec B Ñ Spec A. We call π. The results are basically about this map and he
proofs hinge on a basic lemma which is first we prove.

Basic lemma—the case of fields

9.24 The basic lemma treats the special case of fields:

Lemma 9.25 Let A Ď B be an integral extension of domains. If one of the rings is a

field the other one is as well.
200 ma4200—commutative algebra

Proof: Assume first that B is a field. If y P A is an element, the inverse y´1 is

integral over A and satisfies a dependence relation

y´n ` an´1 y´pn´1q ` . . . ` a1 y´1 ` a0 “ 0.

Multiplying through by yn gives

1 ` ypan´1 ` . . . ` a1 yn´2 ` a0 yn´2 q “ 0

which shows that y is invertible in A. Next assume that A is a field, and let
x P B be given. It satisfies a relation

x n ` a1 x n´1 ` . . . ` a1 x ` a0 “ 0,

and assuming that the degree n is minimal, it holds that a0 ‰ 0. Then a0 is

invertible and we have

1 n´1
x ¨ a´
0 px ` a1 x n´2 ` . . . ` a1 q ` 1 “ 0.

Corollary 9.26 Assume that A Ď B is an integral extension. A prime ideal n in B

is maximal if and only if n X A is maximal.

Proof: The extension

A{n X A Ď B{n

is integral by Proposition 9.16 on page 196, and the corollary ensues since the
quotient by an ideal is a field if and only if the ideal is maximal. o

The Lying–Over Theorem

The first theorem we shall treat—the Lying-over Theorem—gives the structure of the Lying-over Theorem
the fibres of the map π; that is, it gives a qualitative description of the set of (Lying–over-teoremet)

prime ideals in B intersecting A in a given fixed prime ideal. The fibres are not
empty, or in other words, all prime ideals p in A are of the form p “ q X A, and
moreover, there are no inclusion relations between the members of the fibres.
In topological terms this means that the fibres of π are discrete topological
spaces.
9.27 Here it comes:

Proposition 9.28 (Lying-over) The map π is surjective, and the fibres are
discrete. In other words, for each p Ď A there is at least one q Ď B such that q X A “ p.
Moreover, if q and q1 are prime ideals in B with q X A “ q1 X A and q Ď q1 , then
q “ q1 .
 integral extensions 201

Proof: We begin with treating the local case; the rest of the proof is a reduc-
tion to that case.
Assume then that A is local with maximal ideal m. Let n be any maxi-
mal ideal in B which exists according to the Basic Existence Theorem (Theo-
rem 2.50 on page 40). By Corollary 9.26 n X A is maximal, hence equal to m
since m is the only maximal ideal in A.
To see that no ideal in the fibre is contained in another, assume that q Ď q1
and that both intersect A in m (we are still in the local situation). Again by
Corollary 9.26 both q and q1 are maximal and must be equal since q Ď q1 .
Let p be a prime ideal in A. In order to reduce the general case to a local
situation we pass to the localized Ap Ď Bp extension, which persists being
integral in view of Proposition 9.14. According to what we established in the
local case, there is an ideal in Bp , which as all ideals in Bp is of the form qBp ,
such that qBp X Ap “ pAp . But then q X A “ p, this follows e. g. as
pAp X qBq q X A “ A X qAp “ a X pB X qBp q “ A X q
.
If q X A “ q1 X A “ p for two prime ideals in B, one included in the other, it
holds by the local case that qBp “ q1 Bp , and hence q and q1 are equal. o
Example 9.4 It might well happen that π is surjective and has finite fibres y
without B being integral over A. An example can be the extension
A “ krx2 s Ď krx, px ´ 1q´1 s “ B
where we assume that k is algebraically closed and not of characteristic two.
The geometric interpretation of the example is the parabola X given as y “ x2
with a hole punched in it; that is, the point p1, 1q is removed. The map π is just
projection Xztp1, 1qu to the y-axis.
The ring krx2 s is isomorphic to the polynomial ring krys (rebaptize x2 to y). x
Every maximal ideal m in A is of therefore the form x2 ´ a2 (all elements in k
have a square-root) with a ‰ 0, and it holds true that px ´ aqB X A “ px2 ´ a2 qA
since x2 ´ a2 “ px ` aqpx ´ aq. However px ´ 1q´1 is not integral over A; indeed,
if it were, multiplying an integral dependence relation of degree n by px ´ 1qn ,
would yield a relation
1 ` pn´1 px ´ 1q ` . . . ` p1 px ´ 1qn´1 ` p0 px ´ 1qn “ 0,
where the coefficients pi ’s are elements on krx2 s; that is, they are polynomials
pi px2 q in x2 . Putting x “ 1 gives an obvious contradiction.
As of the fibres, it is a nice exercise to check that if a ‰ 1, the two prime
ideals (x-a)B and px ` aqB are the ones lying over px2 ´ a2 qA, but the sole prime
ideal lying over px ´ 1qA is px ` 1qB. K

Problem 9.5 Let A Ď B be an integral extension. Show that the map π : Spec B Ñ
Spec A is a closed map, by showing that πpVpaqq “ Vpa X Aq for any ideal a in
A. Hint: Apply Lying–Over to the extension A{a X A Ď B{a. M
202 ma4200—commutative algebra

Going–Up
The Going–Up Theorem is about extending, or lifting as one also says, ascend- The Going–Up Theorem
ing chains of prime ideals in A to chains in B by climbing them; contrary to (Going–Up teoremet)

the Going–Down Theorem where chains are lifted by a downwards “move

ment”. If a one-step chain in A can be lifted, an easy induction ensures that
any finite ascending chain tpi u of prime ideals in A can be lifted ; that is, there
q0 Ď q1 Ď B
is an ascending chain tqi u of prime ideals in B so that qi X A “ pi .

Ď
9.29 The one step case is what is usually called the Going–Up Theorem p0 Ď p1 Ď A

Theorem 9.30 (Going–Up) Let A Ď B is an integral extension of rings, and let

p0 Ď p1 be two prime ideals in A. Furthermore assume that q0 is a prime ideal in B
lying over p0 . Then there is a prime ideal q1 in B containing q0 and lying over p1 .

Proof: Consider the extension A{p0 Ď B{q0 which is integral by Proposi-

tion 9.16. By the Lying-Over Theorem, there is a prime ideal in B{q0 lying over
p1 {p0 . As all prime ideals in B{q0 are, it is shaped like q1 {q0 for some q1 in B.
Then q1 X A “ p1 . o

Corollary 9.31 (Going–Up II) Any finite chain tpi u of prime ideals in A has a
chain tqi u of prime ideals in B lying over it.

Proof: The proof goes by induction on the number of prime ideals in the
chain A, and one should find it completely transparent pondering the follow-
ing display:
q 0 Ă q 1 Ă . . . Ă q n´ 1
Ď

p0 Ă p1 Ă ... Ă pn´1 Ă pn
The upper chain exists by induction, an one just fils in the upper right corner
citing the Going–Up Theorem. o
9.32 A chain tqi u in B that lifts the chain tpi u, will be saturated whenever tpi u
is; indeed, any prime strictly in between qi and qi`1 would either meet A in
pi or pi`1 since tpi u is saturated, but this can not happen since Lying–Over
guarantees there are no inclusions among primes in the fibres. We conclude
that the suprema of the lengths of chains in the two rings coincide, and hence
we have

Corollary 9.33 If A Ď B is an integral extension, then A and B have the same

Krull-dimension, or in symboles dim B “ dim A.

Going–Down
The Going–Down Theorem the version for descending along chains. One says
that Going–Down property holds for an extension A Ď B if for any pair p1 Ď p0 The Going–Down
of prime ideals in A and any prime ideal q0 in B lying over p0 there is a prime property (Going–Down-
egenskapen)
 integral extensions 203

ideal q1 contained in q0 and lying over p1 ; in other words lifted chains may be
continued downwards. When A and B domains and A integrally closed in its
fraction field the Going-Down property holds for te extension A Ď B, but it is
significantly more subtle to establish than Going-Up.
9.34 As indicated we content ourself by formulating the Going–Down theo-
rem.

Theorem 9.35 Let A Ď B be an extension of interal domains and assume that A

is normal. Given two prime ideals p1 Ď p0 in A and one q0 in B luing over p0 . Then
there exists a prime ideal B contained in q0 and lying over p1 .

Proof: Not yet here! o

9.4 Noether’s normalization lemma

Once more a lemma that has become a theorem and once more an important
y
result due to Emmy Noether.

The Normalization lemma

The Normalization Lemma is a result about domains that are finitely gener-
ated algebras over fields; typical examples being coordinate rings of affine x
varieties, which are shaped like krx1 , . . . , xr s{a. It states that such a domain A
can be realized as a finite module over a polynomial ring. That is, one may
find elements w1 , . . . , wn in A such that krw1 , . . . , wn s is isomorphic to a poly-
nomials ring (the wi ’s corresponding to the variables) and such that A a finite
module over krw1 , . . . , wn s.
The wi ’s will be algebraically independent, and since A is a finite module
over krw1 , . . . , wn s, they form a transcendence basis for the fraction field K of
A. So the number n will be the transcendence degree trdegk K.
The standard proof, which is close to Noether’s original and is the one we
offer, works over any infinite field, but a slightly less translucent proof, due to
Nagata, works over finite field as well.
9.36 Starting out with k Ď A and adjoining elements, one will sooner or later
exhaust the entire ring A since A is finitely generated over k. In the beginning
one may add new elements which are algebraically independent of those al-
ready added, but at a certain point, when the maximal number of algebraically
independent elements is reached, new elements are forced to be algebraically
dependent on the previous. If a new element is integrally dependent on the
old, we are happy, if not, we have to go back and perturb the already added
elements to make the new-comer integral.
Example 9.5 A non-integral new-comer will typically have a pole, and to
illustrate the perturbation process, we consider the simplest way of adding
204 ma4200—commutative algebra

a function with a pole, namely the extension krx, 1{xs of krxs. The geometric
counterpart is the projection of the classical hyperbola, xy “ 1 onto the x-axis,
the hyperbola just being the graph of the function 1{x.
The ring krx, 1{xs is not finite over krxs, but perturbing x slightly, we obtain
a subring over which krx, 1{xs is finite. The subring krx ` 1{xs will do the job;
indeed, krx, 1{xs “ krx, x ` 1{xs is generated by x over krx ` 1{xs and one has
the integral dependence relation

x2 ´ xpx ` 1{xq ` 1 “ 0.

It is remarkable that almost any perturbation of x will work; that is, krx, 1{xs
is finite over krax ` b{xs as long as both the scalars a and b are non-zero. K

Problem 9.6 Show that krx, 1{xs is a finite module over krax ` b{xs for any
scalars a b both being different from zero. M
9.37 The proof of Noether’s Normalization Lemma goes by induction of the
number of generators A requires as an algebra over k, and the basic ingredient
in the induction step is the following lemma:

Lemma 9.38 Let k be an infinite field and let A “ krX1 , . . . , Xm s{a. Assume that
A is a domain whose fraction field K has transcendence degree at most m ´ 1 over
k. Then there are elements y1 , . . . , ym´1 in A such that A is a finite module over
kry1 , . . . , ym´1 s.

Proof: Following our usual convention, we let xi denote the image of the
variable Xi in A. Since the transcendence degree of K over k is less than m, the
m elements x1 , . . . , xm can not be algebraically independent and must satisfy
and equation
ppx1 , . . . , xm q “ 0,
where p is a non-zero polynomial with coefficients in k. Let d be the degree of
p and let pd be the homogenous component of degree d. Now, perturbe the
xi ’s by putting yi “ xi ´ αi xm for i ď m ´ 1 where the αi ’s are scalars to be
chosen. This gives3
3
Recall that for any
0 “ ppx1 , . . . , xm q “ ppy1 ` α1 xm , . . . , ym´1 ` αm´1 xm , xm q “ polynomial ppxq
it holds true that
“ pd pα1 xm , α2 xm , . . . , αm´1 xm´1 , xm q ` qpxm , y1 , . . . , . . . , ym´1 q “ ppx ` yq “ ppxq `
d yqpx, yq where q is a
“ pd pα1 , . . . , αm´1 , 1qxm ` qpxm , y1 , . . . , . . . , ym´1 q polynomial of total
degree less than the
where q is a polynomial of degree less that d in the last variable xm . Now, degree of p. Apply this
since the ground field is infinite, for a generic choice of the scalars αi it holds to pd .
true that pd pα1 , . . . , αm´1 , 1q ‰ 0 (see Exercise 9.9 below). Hence the element xm
is integral over kry1 , . . . , ym´1 s and by consequence, A is a finite module over
the algebra kry1 , . . . , ym´1 s. o
9.39 By induction on m one obtains the full version of the normalization
lemma:
 integral extensions 205

Theorem 9.40 (Noether’s Normalization Lemma) Let k be an infinite

field and let A “ krX1 , . . . , Xm s{a. Assume that A is a domain whose fraction field K
has transcendence degree n over k. Then there are algebraically independent elements
w1 , . . . , wn in A such that A is a finite module over krw1 , . . . , wn s.

Theorem 9.41 (Noether’s Normalization Lemma) Let k be an infinite

field and let A “ krX1 , . . . , Xm s{a. Assume that A is a domain whose fraction field K
has transcendence degree n over k. Then there are algebraically independent elements
w1 , . . . , wn in A such that A is a finite module over krw1 , . . . , wn s.

Proof: We fix the transcendence degree n and proceed by induction on m. If

m ď n, the elements x1 , . . . , xm must be algebraically independent since they
generate the field K over k. But any non-zero polynomial in a would give a
dependence relation between them, so we infer that a “ 0, and hence that
A “ krX1 , . . . , Xm s.
Suppose then that m ą n. By Lemma 9.38 above, there are elements
y1 , . . . , ym´1 such that A is finite over kry1 , . . . , ym´1 s. Now, K is algebraic
over the fraction field of B “ kry1 , . . . , ym´1 s since A is finite over B, and the
latter is therefore of transcendence degree n over k. By induction there are
algebraically independent elements w1 , . . . , wn in B so that B is finite over
krw1 , . . . , wn s. But then also A is finite over krw1 , . . . , wn s, and we are through.
o

Problems

9.7 Let k be an infinite field and f px1 , . . . , xm q a non-zero polynomial with

coefficients from k. Show that f pa1 , . . . , am q ‰ 0 for infinitely many choices
of ai from k. Hint: Use induction on n and expand f as f px1 , . . . , xm q “
i
ř
i gi px1 , . . . , xm´i qxm .

9.8 Prove the formula in Footnote number 3. Hint: Use the binomiall theo-
rem to check it for the polynomials px ` yqν .

9.9 In stead of the linear change of variables yi “ xi ´ αi x1 , show that the proof
n
of Lemma 9.38 goes through over any field with the change yi “ xi ` x1 i when
ni are chosen sufficiently large.

M
206 ma4200—commutative algebra

Corollaries

9.5 The Nullstellensatz

Let a Ď krX1 , . . . , Xn s be an ideal. Remember the the zero locus of Vpaq of a. It
is the set a “ pa1 , . . . , an q so that f paq “ 0 for all members f of a. For every
subset S Ď kn one may als consider the ideal IpSq consisting of polynomials
in krX1 , . . . , Xn s that vanish on S. In general there is no reason that VpIpSqq
should be eqaue to S, but when S a priori is known to be an algebraically
closed subset one has equality. That is VpIpVpaqqq “ Vpaq.

The Nullstellensatz
Hilbert’s Nullstellensatz is about the composition of I and V the other way
?
around, namely about IpVpaqq. Polynomials in the radical a vanish along
?
Vpaq and therefore a Ď IpVpaqq, and the Nullstellensatz tells us that this
inclusion is an equality. We formulate the Nullstellensatz here, together with
two of its weak avatars, but shall come back with a thorough discussion of the
proof(s) a little later.

Theorem 9.42 (Hilbert’s Nullstellensatz) Assume that k is an alge-

braically closed field, and that a is an ideal in krx1 , . . . , xn s. Then one has IpVpaqq “
?
a.

Notice that the ground field must be algebraically closed. Without this as-
sumption the result is not true. The simplest example of an ideal in a polyno-
mial ring with empty zero locus is the ideal px2 ` 1q in Rrxs.

Theorem 9.43 (Weak Nullstellensatz I) Let A be a finitely generated

algebra over the field k and let m be a maximal ideal in A. Then A{m is a finite field
extension of k.

Proof: The field K “ A{m is finitely generated as an k-algebra since A is.

If it not algebraic, it has transcendence degree at least one oveer k and by
Noether’s Normalization Lemma it is a finite module over a polynomial ring
krw1 , . . . , wr s. By Lemma 9.25 it ensues that krw1 , . . . , wn s is field which is a
contradiction since polynomial ring are not fields (if w is a variable 1{w is
certainly not a polynomial). Hence K is finite over the ground field k. o

Theorem 9.44 (Weak Nullstellensatz II) Let k be an algebraically closed

and m a maximal ideal in the polynomial ring krx1 , . . . , xn s. Then m is of the form
a “ px1 ´ a1 , . . . , xn ´ an q.

Proof: By version I, the field krx1 , . . . , xr s{m is algebraic over k hence equa
to since k is assumed to algebraically closed. The ensuing homomorphism
krx1 , . . . , xn s Ñ k has xi ´ ai in its kernel m, and since we already know that
 integral extensions 207

px1 ´ a1 , . . . , xn ´ an q is a maximal ideal it must be equal to m, and we are

through. o

The Rabinowitsch trick

9.45 We proceed to present the J.L. Rabinowitsch trick proving that the weak
version of the Nullstellensatz (Theorem ?? on page ??) implies the strong.
?
That is, we need to demonstrate that IpZpaqq Ď a for any proper ideal a in
krx1 , . . . , xn s.
The crux of the trick is to introduce a new auxiliary variable xn`1 and for
each g P IpZpaqq to consider the ideal b in the polynomial ring krx1 , . . . , xn`1 s
given by
b “ a ¨ krx1 , . . . , xn`1 s ` p1 ´ xn`1 ¨ gq.

In geometric terms Zpbq Ď An`1 is the intersection of the the subset Z “

Zp1 ´ xn`1 ¨ gq and the inverse image π ´1 Zpaq of Zpaq under the projection
π : An`1 Ñ An that forgets the last and auxiliary coordinate. This intersection
is empty, since obviously g does not vanish along Z, but vanishes identically
on π ´1 Zpaq.
The weak Nullstellensatz therefore gives that 1 P b, and hence there are
polynomials f i in a and hi and h in krx1 , . . . , xn`1 s satisfying a relation like
ÿ
1“ f i px1 , . . . , xn qhi px1 , . . . , xn`1 q ` h ¨ p1 ´ xn`1 ¨ gq.

We substitute xn`1 “ 1{g and multiply through by a sufficiently4 high power

g N of g to obtain
4
For instance the
ÿ highest power of xn`1
gN “ f px1 , . . . , xn qHi px1 , . . . , xn q, that occurs in any of
the hi ’s.
?
where Hi px1 , . . . , xn q “ g N ¨ hi px1 , . . . , xn , g´1 q. Hence g P a.

9.6 Appendix—skirmishes
We have relegated a few simple results of preparatory character to this ap-
pendix. They do not take part in the main battle, but are merely skirmishes
on the flanks, though of significant importance for the progress. At least the
two firsts are easy and elementary. The concept of transcendence degree might
be a little more involved, but should be known to mosts students from earlier
courses.

The determinantal trick

The determinantal trick (so named by Miles Reid), is a little lemma from linear
algebra adapted to modules. It is just a reformulation of the arch-classical fact
208 ma4200—commutative algebra

that a square matrix with a non-trivial kernel has a vanishing determinant—

which is the raison d’être for determinants and the reason for their name.
We start by recalling that the good old matrix multiplication may be ex-
tended to modules. Indeed, if Φ “ p f ij q is an n ˆ n-matrix with entries from
a ring A and m “ pm1 , . . . , mn q a string of elements from an A-module M, the
expression Φ ¨ m is meaningful. It is the string of elements from M given as
p f ij m j q1ďiďn ; or expressed in a fancy functorial fashion, the map m ÞÑ Φ ¨ m is
ř

just the map id M bΦ : Mb A nA Ñ Mb A nA.

Lemma 9.46 (The determinental trick) Let Φ be an n ˆ n-matrix with

entries in the ring A, and let M be an A-module. Assume that M has generators A
ι
/ nM
m1 , . . . , mn such that Φ ¨ pm1 , . . . , mn q “ 0. Then the determinant det Φ kills M. Φ

det Φ¨id A nM
Proof: Consider the A-linear map ι : A Ñ nM sending x to px ¨ m1 , . . . , x ¨ mn q.
Then the hypothesis of the lemma translates into the relation Φ ˝ ι “ 0. Citing Φ:

the adjunction formula from linear that states Φ: ¨ Φ “ det Φ ¨ I, we find  

A / nM
ι

det φ ¨ ι “ Φ ˝ Φ ˝ ι “ 0.
:

This means that pdet Φ ¨ m1 , . . . , det Φ ¨ mn q “ det Φ ¨ ιp1q “ 0; hence det Φ kills
M as the mi ’s generate M. o

Finite generation in towers

The next preparation is an utterly simple lemma which we give for the benefit
of the students: Two successive finite ring extension gives a finite extension
having generators the products of the generators of the two.

Lemma 9.47 Let A Ď B Ď C be a tower of rings and assume that C i finite over B
and B is finite over A, then C is a finite over A.

Proof: Let x1 , . . . , xr be a generating set for B as an A-module and y1 , . . . , ys

one for C over B. Then the products xi y j will generated C over A. This is
ř
elementary indeed, if z “ b j y j with b j P B, write each coefficient b j as
ř
b j “ j aij xi to obtain
ř ř ř ř
z “ j b j y j “ j p i aij xi qy j “ i,j aij xi y j .

Transcendence degree

9.48 Let k Ď K be a field extension and let x P K be an element not lying in k.

The elements in K can be parted in to classes, the algebraic elements and the
transcendental ones. The algebraic ones are those x that satisfies a relation like

a n x n ` a n ´1 x n ´1 ` . . . ` a 1 x ` a 0 “ 0 (9.7)
 integral extensions 209

where the ai ’s are elements form k and an ‰ 05 , and the transcendental ones
are the rest; that is, those for which ppxq ‰ 0 for any non-zero-polynomial in 5
Since K is field we
krXs. may as well assume
More generally a collection x1 , . . . , xr of element from the bigger field K is that an “ 1
said to be algebraically dependent over k if for some non-zero polynomial p in Algebraically dependent
r variables and coefficients from k it holds true that ppx1 , . . . , xr q “ 0, and of elements (algebraisk
avhengige elementer)
course, if no such polynomial can be found, the collection is called algebraically
independent over k. Algebraically independent
elements (algebraisk
A collection of algebraically independent elements x1 , . . . , xr is a tran-
uavhengige elementer)
scendence basis of K over k if it is maximal; that is firstly, it is algebraically transcendence basis
independent and secondly, adding any new element to it will make it alge- (transcendence basis)
braically dependent. In terms of the field K that x1 , . . . , xr is a transcendence
basis means that the field K is an algebraic extension of kpx1 , . . . , xr q, and that
kpx1 , . . . , xr q is isomorphic to the rational function field over k in r variables..
We cite the following result but refrain from giving a proof:

Proposition 9.49 Every field extension K of k that is not algebraic, has a transcen-
dence basis. All transcendent bases for K over k have the same number elements.

That common number is called the transcendence degree of K over k and de- Transcendence degree
noted by trdegk pKq. (transcendensgrad)

9.50 In linear algebra The Tietze’s Extension Lemma is used to show that
bases of vector spaces have the same number of elements. An analoguous
result will give that different transcendence bases of fiels extension have same
cardinality.

Lemma 9.51 (Exchange emma) Let k Ď L Ď K be a tower of fields and let a and
b be two elements of K. Assume that b is transcendent over L but algebraic over Lpaq,
then a is algebraic over Lpbq.

Proof: Since b is algebraic over Lpaq, there is a polynomial f px, yq with coeffi-
cients from L such that f pa, yq ‰ 0 but f pa, bq “ 0. Expandingh f in powers of
x yields
ÿ
0 “ f pa, bq “ qi pbqai ,

wehere qi pyq P Lrys. This looks very much like a dependence relation for a over
Lpbq, it only remains to see that f px, bq is not identically zero, but since b is
trancsendent over L and at least one of the polynomials qi pyq’s is non-zero, this
holds true. o
Proof of Trancendence bases: We only treate the case of finite bases. Let
A “ ta1 , . . . , an u and B “ tb1 , . . . , bm u be two transcendence bases for K over k.
o
Lecture 10

Krull dimension

Very preliminary version 0.01 as of 2018-12-03 at 10:01 (typeset 3rd December 2018 at 10:03am)—
Prone to misprints and errors and will change.
03/12/2018 Several changes, still unfinished.

Dimension is in general a complicated and subtle notion. In a few good

cases it is well defined. Vector spaces have a dimension as do manifolds of
course (or at least each connected component has). They are locally isomor-
phic to open sets in some euclidean space, and the dimension of that euclidean
space is constant along connected components, and is the dimension of the
component.
There is another and naive approach to the conception of dimension. For
example, in three dimensional geometric gadgets, called threefolds, we may
imagine increasing chains of subgadgets of length three; points in curves,
curves in surfaces and surfaces in the threefold. This may be formalized by
using closed and irreducible subsets as “subgadgets”, and the dimension
will be the maximal of length of chains such, or rather the supremum of
the lengths as the lengths might be bounded. This definition works for any
topological space, but the ensuing dimension does not carry much information
unless the topology is “Zariski-like”. Translated into algebra, where prime
ideals correspond to closed irreducible subsets, this leads to the consept of
Krull dimension of a ring; the supremum of the length of chains of prime
ideals.
For varieties there is another good candidate for the dimension, namely the
transcendence degree of the fraction field KpXq over the ground field. This
may be motivated by the fact the Krull dimension of the polynomial ring
krx1 , . . . , xn s equals n (which is not obvious at all but follows from the Princi-
pal Ideal Theorem), and obviously the transcendence degree of kpx1 , . . . , xn q is
n. That the two coincide, follows from the Normalization lemma which states
that every variety is a finite cover of an affine space.

The definiton
212 ma4200—commutative algebra

10.1 Let A be a ring. The point is to consider strictly ascending and finite
chains tpi u of prime ideals

p0 Ă p1 Ă . . . Ă pν .

Recall that ν is called the length of the chain; it is one less than the number of
prime ideals, or if you want, the number of inclusions. The Krull dimension Krull dimension (Krull
of A be the supremum of the length of all such chains in A. It is denoted by dimensjon)

dim A. We shall say that a chain is saturated if there are no prime ideals of saturated chains (mettede
A lying strictly between two of the terms, and it is maximal if additionally it kjeder)
Maximal chains (maksi-
neither can be lengthened upwards nor downwards. male kjeder)
10.2 Even if each chain is finite, there might be arbitrary long chain, and the
Krull dimension will in that case be infinite. It is easy to find examples among
non-noetherian rings whose Krull dimension is infinite; Example10.1 below is
an obvious example of one with an infinite ascending chain. Even Noetherian
rings might have infinite Krull dimension. However, these examples live on
the fringe of the Noetherian society, and rings met in mainstream algebraic
geometry will all have finite dimension. When dim A ă 8, there are saturated
chains of maximal length, that is of length dim A. We shall also see that local
Noetherian rings have finite Krull dimension.
Example 10.1 The polynomial ring in A “ krx1 , . . . , xr , . . .s in infinite many
variables is of inifinite Krulll dimension. Each of the ideals pr “ px1 , . . . , xr q is
a prime ideal, and they form an infinite ascending chain.
There is also an infinite descending chain of prime ideals in A whose
members are the ideals qr “ pxr , xr`1 , . . .q. K

Example 10.2 (Rings of dimension zero) A ring A is zero-dimensional when

there are no chains of prime ideals with two or more terms; or in other words,
when all its prime ideals are both minimal and maximal. If A in addition is
Noetherian, there are according to the Lasker-Noether Theorem (Teorem 8.17
on page 180) only finitely many minimal prime ideals, and we may char-
acterise Artinian rings as those having finitely many prime ideals all being
maximal (Theorem 7.58 on page 172). Hence a Noetherian ring is of dimension
zero if and only if it is Artinian. In particular, fields are of dimension zero. K

Example 10.3 (Domains of dimension one) It is worth while contemplate one-

dimensional rings as well. In a one-dimensional domain the zero ideal is a
prime ideal so the saturated chains are all of the form p0q Ă p. All non-zero
prime ideal are therefore maximal, and they are as well minimal over the zero
ideal. Examples are pid’s; in particular polynomial rings krXs in one variable
over fields and, of course, the integers Z will all be one-dimensional.
If A is not a domain, there may as well be saturated chains of length one;
in other words, some of the minimal prime ideals might also be maximal (see
Example 10.4 below). K
 krull dimension 213

10.3 If A has several minimal prime ideals, the space Spec A will have several
irreducible components, as in the example above with p0, 1q and the x-axis be-
ing the components. The dimension of A will we the largest of the dimensions
of the components, or if there are infinitely many, the supremum. In algebraic
terms this translates as:

Proposition 10.4 If tpi u are the minimal primes of A, then dim A “ supi dim A{pi .

Proof: The intersection of the prime ideals of a chain being prime, any max-
imal chain starts at minimal prime (see also Paragraph 2.61 on page 43 and
Exercise 2.23 on page 43). o
10.5 It is common usage to call dim Ap the the height of p, or more generally Height of ideals (høyden
til idealer)
for any ideal a in A the heigh is the least height of any prime ideal containing
a; that is
ht a “ min ht p.
aĎp

A useful inequality

10.6 Any chain tpi ui of prime ideals in A may be broken in two at any stage,
say at term p “ pν . It is the concatenation of a lower chain, formed by the
members of the chain contained in p, and an upper chain, the one formed by
those containing p. And of course, one may as well splice two chains provided
one ends at the prime where the other one begins.

p0 Ă p1 Ă . . . Ă pν´1 Ă pν Ă pν`1 Ă . . . Ă pn

Now, the primes contained in p are in a one-to-one correspondence with

the prime ideals in Ap , hence the lower chains correspond to chains in the
localization Ap . This means that dim Ap .
In similar way, prime ideals containing p correspond to prime ideals in
the quotient A{p and hence the upper chains correspond to chains in A{p.
When considering the suprema of the lengths of such splices, we arrive at the
following formula:

Proposition 10.7 Let A be a ring and p a prime ideal. Then

dim Ap ` dim A{p ď dim A.

Notice that the proposition is still valid if one or more of the dimensions are
infinite, with the usual intepretation that n ` 8 equals 8.
In some rings there are maximal chains—that is, saturated chains which
cannot be lengthened—of different lengths, and the Krull dimension is of
course the length of the longest. For any prime ideal in a shorter chain, the
inequality as in the proposition will be strict. It is easy to give such examples
when the ring A is not a domain, one simply takes an A with components
214 ma4200—commutative algebra

of different dimension (see Example 10.4) below). However, there are also
examples when A is a Noetherian domain, but these are again rather exotic
constructs you don’t tumble over in algebraic geometry.
Example 10.4 Let A “ krX, Ys{a where a “ pXY, YpY ´ 1qq, and as usual, we y

let lower case letters denote the classes of their upper case versions. Geomet- p0,1q
rically Vpaq is the subset of k2 being the union of X-axis and the point p0, 1q.
x
The primary decomposition of p0q in A is p0q “ pyq X px, y ´ 1q. Hence pyq and
px, y ´ 1q are the minimal prime ideals in A. Now, px ´ a, yq is a maximal ideal
for any a P k, so A possesses saturated chains

pyq Ă px ´ αq,

and therefore dim A “ 1. On other hand px, y ´ 1q is clearly a maximal ideal,

and hence is both maximal and minimal. K

Example 10.5 If you want an example that is a local ring consider the ideal
a “ pZq X pX, Yq “ pZX, ZYq in krX, Y, Zs and let A be the localization at
px, y, zq of the quotient krX, Y, Zs{a. Then p0q “ pzq X px, yq is the primary
decomposition of p0q and pzq and py, xq are minimal prime ideals in A; but
there maximal chains pzq Ă px, zq Ă px, y, zq and px, yq Ă px, y, zq in A. K

Going–Up again

A most useful consequence of the Going–Up Theorem is that any chain in an

integral extension B of a ring A, when intersected with A becomes a chain in
A; this ensures that the dimension is preserved in integral extensions. We have

Proposition 10.8 Let A Ď B be an integral extension of domains. Then dim A “

dim B.

Proof: This is a direct consequence of the Going-Up theorem as formulated

in Going–Up IIS. Indeed, let tqui be a chain of length d in B and consider the SCorollary 9.31 on page 203

prime ideals pi “ qi X A. They form a chain in A whose length is d since no

two qi ’s intersect A in the same ideal according to Lying–OverT. This shows TProposition 9.28 on page 201

that dim B ď dim A. On the other hand by Going–Up II, every chain in A has
a chain in B lyinig over it which means that dim B ď dim A. o

10.9 For example if Q Ď K is any field extension (finite or not) the ring of
integers in K, that is the integral closure A of Z in K, is of course integral
over Z and consequently is of dimension one. In particular this applies to the
quadratic extensions K “ Qp dq we have seen, but also to the more impressive
‘

extension K “ Q,s the field of algebraic numbers. The ring of algebraic integers
Z
s is therefore of dimension one, but recall, it is not a Noetherian ring.
 krull dimension 215

Cutting out a hypersurface

Cutting a varity X with a hypersurface is a rather common technique in
geometry, which on the level of algebras corresponds to passing to a quotient
A{p f qA of A by a principal ideal. One suspects the dimension to go down,
but it might happen that the hypersurface contains one of the components of
X, and in that case the dimension might stay the same (if the component is
one of maximal dimension). To avoid such an accidental behaviour, one must
assumes that f does not lie in any of the minimal prime ideals of A; then one
has:

Proposition 10.10 Let A be ring of finite Krull dimension and let f P A is an

element not belonging to any minimal prime ideal in A, then dim A{p f qA ă dim A.

Proof: Chains of prime ideals in A{p f qA are in one-to-one correspondence

with chains in A containing f . Moreover, a prime ideal p that is minimal over
p f q, properly contains a minimal prime q of A, and consequently any chain in
A starting upwards from p can be lengthened downwards by joining q to it. o

The dimension of polynomial rings

There is an obvious chain of prime ideals in the polynomial ring krx1 , . . . , xn s
over a field k corresponding to a chain of linear subspaces in kn with the
dimension jumping by one between two consecutive. The terms of the ideal
chain are the ideals generated by the variables, and the length is n:

p0q Ă px1 q Ă px1 , x2 q Ă . . . Ă px1 , . . . , xn q. (10.1)

This shows that the Krull dimension of krx1 , . . . , xn s is at least n. To see it

equals n, is a subtler matter which asks for some bigger artillery; it hinges on
the Normalization LemmaU. UTh. 9.40 on page 205

Proposition 10.11 (Dimesion of Polynomial Ring) Let k be a field and

let krx1 , . . . , xn s be the ring of polynomials in the variables x1 , . . . , xn . Then one has
dim krx1 , . . . , xn s “ n.

Proof: Assume that

p0q Ă p1 Ă p1 Ă . . . Ă pd (10.2)

is a chain with p1 minimal over p0q. Pick a non-zero irreducible element f

in p1 ; then p f q is a prime ideal and p f q “ p1 . Consider the quotient algebra
A “ krx1 , . . . , xn s{p f q. It is a domain, and since we have imposed a non-trival
condition on the variables, its fraction field is of transcendence degree at most
n ´ 1. By Noether’s Normalization Lemma it follows that A is finite over
a polynomial ring in at most n ´ 1 variables, and hence by the Going–Up
Theorem the dimension of A is at most n ´ 1 as well.
216 ma4200—commutative algebra

Now, the chain (10.2) induces a chain in A shaped like

p0q Ă p2 {p1 Ă . . . Ă pd {p1 ,

and it follows that n ´ 1 ě dim A ě d ´ 1, and consequently it holds true that

dim krx1 , . . . , xn s ď n. o

Corollary 10.12 Let A be a domain finitely generated over the field k whose field
of fractions is K. Then dim A “ trdegk K

Proof: Let n “ trdegk K. By Noteher’s Normalization LemmaV there is V 9.40 on page 205

a polynomial ring krx1 , . . . , xn s Ď A over which A is finite, hence dim A “

dim krx1 , . . . , xn s “ n. o

10.1 Krull’s Principal Ideal Theorem

This important theorem is also known under its German name Krull’s Haup-
tidealsatz. It lies at the bottom of the dimension theory in commutative alge-
bra and algebraic geometry, and in his book1 Irving Kaplansky refers to it as 1

“may be the most important single theorem in the theory of Noetherian rings”.
There is however a rather simple underlying intuition that the dimension of a
solution space goes down by at most one when an additional equation is intro-
duced. We recongnize this from theory of linear equations, but the principle
has a much wider scope (as shows e. g. the Hauptidealsatz).
10.13 The scene is a ring A with the two main players, a prime ideal p and
an element x over which p is minimal; in other words, there is no prime ideal
properly lying between pxq and p. And the conclusion is that ht p is at most
one. Of course, the prime ideal p might be a minimal one and then the height
would be equal to zero, but if A is a domain and x is non-zero, the height will
be one.

Theorem 10.14 (Krull’s Prinicipal Ideal Theorem) Let A be a Noethe-

rian ring and x an element from A. Assume that p is a prime ideal in A which is
minimal over pxq. Then ht p ď 1.

Proof: We are to show that there are no chain of prime ideals of length two pxq Ď p
Ă

as q1 Ă q Ă p. By passing to the quotient A{q1 and subsequently localizing in q

q{q1 , we may assume that A is a local domain with maximal ideal p, and we
Ă

q1
our task is to prove that if q Ă p, then q “ 0.
The first observation is that, since p is minimal over pxq, the ring A{xA has
only one prime ideal and being Noetherian, it is Artinian, and we shall have
the opportunity to activate the descending chain condition. The chain we shall
exploit, is the descending chain tpxq ` qpnq un , where qpnq is the n-th symbolic2 2
Don’t panic! We
power of q, that is, qpnq “ A X qn Aq . The chain tpxq ` qpnq u corresponds to the would rather have used
the powers qn . But
powers of prime ideals
can be unruly, and at a
later point in the proof
they will not serve our
purpose.
 krull dimension 217

descending chain t pxq ` qpnq {pxqu in A{xA and must eventually be stable, as
` ˘

A{xA is Artinian. Hence there is an n so that

pxq ` qpn`1q “ pxq ` qpnq .

This entails that if a P qpnq , one may write a “ b ` cx with b P qpn`1q , so that cx P
qpnq . This entails3 that c P qpnq , and consequently that qpnq Ď qpn`1q ` xqpnq Ď qpnq 3
This is where we
Nakayama’s lemma yields that qpnq “ qpn`1q ; this gives qn Aq “ qn`1 Aq , and use the symbolic
power; since x R q, it
appealing once more to Nakayama’s lemma, we may conclude that qAq “ 0; is invertible in Aq , and
that is, q “ 0. o c P A X qn A q

The general version of The Principal Ideal Theorem

This result, ofen called "The height theorem", is the natural generalisation
of the Principal Ideal theorem which applies to minimal primes over ideals
with more than one generator. The natural guess that the height is at most
the number of generators is actually true, and in agreement with the naive
intuition that imposing r constraints on a system should at most lower the
dimension of the solution space by r, reasoning that it diminishes at most with
one for each new condition imposed. This points to an induction argument,
but one which will be slightly more subtle than the naive intuitive version.

Theorem 10.15 (The Height Theorem) Let A be a Noetherian ring and let p
be a prime ideal minial over an ideal a generated by r elements. Then ht p ď r.

Proof: Let a “ pa1 , . . . , ar q. As indicated above the proof goes by inductuion

on r, and heading for a contradiction, we assume that there is a chain tpi u in A
of length d ą r ending at p. We may assume that a1 is not in pd´1 , and so there
is no prime lying properly bewteen pa1 q ` pd´1 and p. The radical of pa1 q ` pd´1
A
therefore equals p, and a power of p is contained in pa1 q ` p1 . Write

ait “ ci a1 ` bi p

pa1 q pd´1
with bi P pd´1 , and let b “ pb2 , . . . , br q. Then b is contained in pd´1 . We contend
that there is prime ideal q lying between b and pd´1 , properly contained in
q
pd´1 ; indeed, if pd´1 were minimal over b, the height of pd´1 would be at most pd´2
r ´ 1 by induction, but being next to the top in a chain of length d, the ideal
pd´1 is of height at least d ´ 1, and r ´ 1 ă d ´ 1. b

Now, the idea is to pass to the ring A{q. The ideal q ` pa1 q contains a power
of a, hence there is no prime ideal between q ` pa1 q and p, which means that the
p{q is minimal over the principal ideal q ` pa1 q{q, and therefore of height one
after the Principal Ideal Theorem, but there is also the chain 0 Ă pd´1 {q Ă p{q.
Contradiction. o
10.16 It ensues from the Height Theorem that any local Noetherian ring A
has a finite Krull dimension. Indeed, the maximal ideal m is finitely generated,
218 ma4200—commutative algebra

and by the Height Theorem the height of m, which is the same as dim A, is
bounded by the number of generators. Similarily, Noetherian rings enjoy a
descending chain condition for the prime ideals. Any term in a chain is finitely
generated and hence is of finite height.

Proposition 10.17 A local Noetherian ring is of finite Krull dimension. Noethe-

rian rings satisfy the ddc for prime ideals.

10.18 Among the different numbers associated with a ring having some
flavour of a dimension, is the so-called embedding dimension of a local ring A. Embedding dimension
If m denotes maximal ideal of A, embedding dimension of A is defined as the (embeddingsdimensjon)

vector space dimension dim A{m m{m2 (the module m{m2 is killed by m and
therefore is a vector space over A{m). Any vector space basis of m{m2 is of the
form rx1 s, . . . , rxr s for members xi of m. Nakayama’s lemma implies that the
xi ’s generate m, and in its turn The Height Theorem yields that dim A ď r. We
have thus proved

Proposition 10.19 Assume that A is a Noetherian ring with maximal ideal m.

Then dim A ď dim m{m2 .

UFD’s once more

In Lecture 3 we showed a criterion of Kaplansky’s (Proposition 3.11 on
page 61) which tells us that a domain A is a udf if and only if "every prime
contains a prime". When A is a Noetherian domain, this criterion can be im-
proved. Citing Proposition 10.17 above which asserts that prime ideals in
a Noetherian ring satisfy the descending chain condition, we infer that any
prime ideal in A contains a prime ideal of height one (if this is not true, one
easily constructs a descending chain that does not terminate). To ensure that A
is a ufd it therefore suffices that prime ideals of height one contain prime ele-
ments, but of course in that case, since prime elements generate prime ideals,
the height one ideal is itself generted by the prime element. This leads to

Theorem 10.20 A Noetherian domain A is a ufd if and only if every prime ideal of
height one is principal.

10.2 System of parameters

10.21 Let A be a local ring with maximal ideal m whose Krull dimension is
n. A sequence x1 , . . . , xn of n elements in A is called a system of parameters if System of parameters
the ideal a they generate is m-primary. Since m is maximal, this amount to the (parmetersystemer)

radical being equal to m (Proposition 8.4 on page 175), or if A is Noetherian

that a contains some power of m.

Proposition 10.22 Every local Noetherian ring has a system of parameters.

 krull dimension 219

Proof: Let A be the ring, let m the maximal ideal, and let n “ dim A. We
shall, by a recursive construction, exhibit a sequence x1 , . . . , xn of elements in
m so that the ideal ai “ px1 , . . . , xi q generated by the i first has all its minimal
primes of height i. Assume that aν has been constructed and consider the
prime ideals tp j u minimal over aν . They all have height ν so if ν ă n, none
of them equals m. Hence their union is not equal to m by Prime Avoidance
(Lemma 2.28 on page 33), and we may pick an element xν`1 from A so that
Ť
xν`1 P m, but xν`1 R i pi . Then any prime ideal minimal over aν`1 “
px1 , . . . , xν`1 q has height ν ` 1; indeed, let p be one of them. It is not among
the minimal prime ideals pi of aν , and therefore must contain one of pi ’s
properly, say p j , and we infer that ht p ą ht p j “ ν. The other inequality; that is
ht p ď ν ` 1, ensues from the Height Theorem. o

10.23 The geometric counterpart of a system of parameters is, given a variety

X and point P on X, a sequence of hypersurfaces that intersect the given vari-
ety in just the point P, or more precisely since we talk about a local concept,
intersect a neighbourhood of P just in P.
10.24 One cannot in general hope that the maximal ideal itself is generated by
dim A elements. The plane cusp A “ krX, Ys{pY 2 ´ X 3 q gives a simple example.
The maximal ideal m “ px, yq requires both x and y as generators since no
linear combination αX ` βY with α, β P k for degree reasons can lie in pY 2 ´ X 3 q,
and therefore x and y are linearly independent modulo m2 “ px2 , xy, y2 q.
The corresponding geometric situation is like this. Both the x-axis and the
y-axis intersect the curve C “ VpY 2 ´ X 3 q only at the origin, but because C
has a double point there, there will always be an intersection multiplicity;
indeed, any line through the origin intersects C only at the origin, but with a
multiplicity.
Noetherian local rings whose maximal ideal needs no more generators than
the Krull dimension are said to regular. A general Noetherian ring is regular if Regular local rings
the local rings Ap are regular for all prime ideals p in A. (regulære lokale ringer)

Problem 10.1 With notations as above, show that ` Ap pAp {pX ´ αYqAp q “ 2
when α is a scalar and that ` Ap pAp {pYqAp q “ 3. The excess length in the latter
case is explain by the x-axis, that is the line Y “ 0, being tangent C at the
origin in some sense. Show that ` A pA{pβX ´ αYqq “ 3 for all linear forms
βX ´ αY. M

Dimension and fibres

10.25 From linear algebra we know that for a given linear map φ : V Ñ W one
has the formula

dim im φ ` dim ker φ “ dim V ` dim W,

220 ma4200—commutative algebra

or using that dim im φ ď dim V, it yields the inequality dim V ď dim ` dim φ´1 p0q.
For a smooth map φ : X Ñ Y between manifolds there is a similar inequality

dim Y ď dim X ` dim φ´1 pyq,

for y P φpXq, which in fact is just the inequality from linear algebra applied
to the derivative of φ. If now, φ : X Ñ Y is a map of varieties, or between
spectra of rings, there is a similar formula, however we confine ourselves to an
algebraic version valid for maps of local rings.
10.26 Recall that map of local rings is map of rings between two local rings Maps of local rings
which sends the maximal ideal into the maximal ideal. (lokale ringavbildning)

Proposition 10.27 Let A and B be the two local rings having maximal ideals m
and n respectively, and assume that φ : A Ñ B is a map of local rings. Then it holds
true that
dim B ď dim A ` dim B{mB.

Proof: We begin with choosing two systems of parameters. One is a system

of parameters x1 , . . . , xr for the maximal ideal m, and the other is one for the
ideal n{mB in the ring B{mB, and let y1 , . . . , ys be liftings to B of the members
of the latter. We contend that the ideal a “ pφpx1 q, . . . , φpxr q, y1 , . . . , ys q the two
systems generate in B is n-primary; indeed, some power nν is contained in
py1 , . . . , ys q ` φpmq, and consequently some higher power lies in a since high
powers of m lie in px1 , . . . , xr q. o

Example 10.6 (Strict inequality may occur) Consider the map ψ : C2 Ñ C2 send-
ing a point pu, vq to pu, uvq. The fibre over p0, 0q is the entire line u “ 0, and
thus of dimension strictly larger than the difference between the dimensions of
the source and the target.
Transcribing this into local algebra we consider the map of rings φ : kru, vs Ñ
krx, ys that sends u ÞÑ x and v ÞÑ xy . The appropriates localizations are
A “ kru, vspu,vq with m “ pu, vqA and B “ krx, yspx,yq with n “ px, yqB. Then
mB “ px, xyq and the “fibre” B{px, yqB “ pkrx, ys{px, xyqqpx,yq “ kryspyq is of
dimension one, whereas of course, dim B ´ dim A “ 0. K

10.28
10.29 As an example connecting up with the geometric situation, assume that
A Ñ B is a mp of rings, not necessarily local, and that m is a maximal ideal in
A. Furthermore, assume that n is maximal ideal so that φ´1 n “ m. Then the
proposition yields
dim Bn ď dim Am ` dim B{mB.

The number dim Am is the dimension of the component of Y “ Spec X that

passes through m and dim Bn that of the component passing through n, and
finally, Spec B{mB is identified with the fibre of φ.
 krull dimension 221

Examples

10.7 (A polynomial ring of excess dimension) The following example of Krull’s

illustrates that non-Noetherian rings may show a pathological behaviour when
it comes to forming polynomial rings. We shall exhibit a one-dimensional local
ring R such that the Krull dimension of the polynomial ring RrTs equals three;
that is dim RrTs “ dim R ` 2.
The ring R is no more exotic than the ring of rational functions f px, yq in
two variables over a field k which are defined and constant on the y-axis. The
elements of R when written in lowest terms have thus a denominator not
divisible by x, and f p0, yq, which then is meaningful, lies in k.
The ring R is not Noetherian, for instance the ideals pxy´i q with i P N
form an ascending chain which does not stabilize (very much like in Ex-
ample 7.4 on page 156). The elements f P R such that f p0, yq “ 0 clearly
form a maximal ideal m, and in fact, it is the only prime ideal in R. Indeed, if
f px, yq “ ppx, yq{qpx, yq and f p0, yq ‰ 0, the numerator ppx, yq cannot have x as
a factor and qpx, yq{ppx, yq lies in R as well. Hence R is a local one-dimensional
ring.
Consider now the polynomial ring RrTs. It has the prime ideals 0, mRrTs
and pTq ` mRrTs, but there is forth one, namely the ideal q of polynomial
FpTq such that Fpyq “ 0. This in not equal to zero as xT ´ xy lies there, and
it is contained in mRrTs. Assume namely that FpTq “ i ri px, yqT i P q. Then
ř

substituting y for T, gives 0 “ Fpyq “ i ri px, yqyi “ 0, which with x “ 0 yields

ř
i
ř
i ri p0, yqy . By definition of R the functions ri p0, yq belongs to k, and since
the different powers yi are linearly independent, it ensues that ri P m; that is,
FpTq P mRrTs.

10.8 (A Noetherian ring of infinite dimension) This examples was found by

Masayoshi Nagata. In his famous book “Local ring” ends with a series of
peculiar rings having unexpected properties. The first is a Noetherian ring of
infinite Krull dimension. Each maximal is of course of finite height, but there
are maximal ideals of arbitrary high height.
The construction has the ring krx1 , x2 , . . .s of polynomials in countably many
variables over a field k as starting point, and depends on a decomposition
N “ i Ii of the natural numbers as a disjoint union of finite subsets Ii so
Ť

that #Ii tends to infinity when i does. It can as simples as I1 “ t1u, I2 “ t2, 3u,
I3 “ t4, 5, 6u an so forth with Ii consisting of i consecutive numbers.
Let pi be the prime ideal generated by the variables x j for which the index
j belongs to Ii , moreover S will be the set of elements in krx1 , . . . , xr . . .s not
belonging to any of the prime ideals pi . Then S is multiplicatively closed and
we let A “ krx1 , x2 , . . .sS and qi “ pi A.
The main observation is that local rings Api are equal to polynomial rings
over a certain field in the variables x j with index j P Ii localized at the prime
ideal those variables generate; that is, Aqi “ Ki rx j |j P Ii spx j | jP Ij q , and where
222 ma4200—commutative algebra

Ki is the field of rational functions over k in the variables x j for j R Ii . This

shows that Aqi is Noetherian and of dimension #Ii (See problem 6.20 on
page 135). It follows that A is infinite Krull dimension, and it remains to see it
it Noetherian, which follows from the following lemma

Lemma 10.30 Assume that A is a ring such that all localisations Am at maximal
ideals are Noetherian and that any element i A is contained in finitely many maximal
ideals. Then A is Noetherian.

Proof: Let a be an ideal. Each ideal aAm is generated by finitely many ele-
ments, as Am is Noetherian), and they may be chosen to lie in a. Recollecting
all these generators for all the finitely many maximal ideals containing a, one
obtains a finite set of generators for a. o

Problem 10.2 Let A be a Noetherian ring and let p Ă q be two prime ideals.
Prove that if there is a prime ideal properly contained between p and q then
there are infinitely many. Hint: Assume that r1 , . . . , rr are the primes lying
Ť
properly between p and q. Then i ri is not equal to q according to the princi-
Ť
ple of Prime Avoidance. Pick an element x P pz i ri and apply the Principal
Ideal Theorem. M

Problem 10.3 Show that among the Noetherian rings only the Artinian ones
and the semi local one-dimensional ones have finitely many prime ideals. M
Lecture 11

Principal ideal domains

Preliminary version ´8 as of 2018-08-19 at 07:30:08 (typeset 3rd December 2018 at 10:03am)—

Much work remaining. Prone to misprints and errors and will change.

A class of rings which are omnipresent in mathematics are the principal ideal
domains. As the name indicates, they are integral domains all whose ideals Principal ideal domains
are principal; that is, every ideal is generated by a single element. They are (Hovedidealområder)

simple kind of rings with a lot of nice properties. As we shortly shall see they
are automatically Noetherian and unique factorization domains.
The most prominent members of this club are the ring of integers Z and the
ring of polynomials krts over a field k. Modules over Z are just abelian groups,
incontestably met everywhere in mathematics. Modules over krts appear, if
possible, even more frequently but often disguished. Such a module is just a
vector space V together with a k-linear the action of t; that is, together with a
linear operator.
The of club of principal ideal domains is a subclub of the even more fab-
shionalble club of Dedekind domains, and these are as frequently met as the
pid. The ring of integers in an algebraic function field is a Dedekind domain,
so they are everywhere in algebraiv number theory. An in algebraic geomtery
they are the coordianet sings of non-singualr affine curves. So teh outspring of
both number theory and algebaric geoemtry is found in that club.
They are integrally closed domains of Krull dimesion one: So the ring of
regular affine curves!
The principal ideal domains live in a family of four. The closest relativ be-
ing the big brothers, rthe Bezout domains. They are not necessarly Noetehrein,
but have proerty that any fintely generated ideal is principal.
The two other family members are the so called Dede kind rings and the
Prufer rings. They There is a genaralization of pid’s called Bezout domains.
These rings are not requiered to be Noetherian, but they have the property
that any finitely generated ideal is principal. A theorem in the theory of
functions states that the ring of holomorphic functions i a domain Ω of C are
Bezout domians which makes the Bezout rings important. The Bezout rings
share several nice properties with the pid’s and we shall prove some whose
proof does not have addirional difficulties.
224 ma4200—commutative algebra

The finitely generated modules over a pid are one of the very few classes of
modules which are completely classified. The well-known “Main theorem for
finitely generated abelian groups” states that such a group M decomposes as a
direct sum of cyclic groups; that is, one has

M » Zν ‘ Z{pi i Z,
à ν

where ν is non-negative integer, the νi ’s are positive integers and the pi are
prime numbers; of course the sum is finite. Over any pid finitely generated
modules decomposes in total analogy to this; it holds true that
à ν
M » Aν ‘ A{pi i A (11.1)
i

where now the pi ’s are irreducible members of A, or what amounts to the

same, prime elements. As to uniqueness, the integers ν and the sequence pνi qi
of the νi ’s appearing is unique. Moreover the sequence ppi qi of the irreducible
elements pi ’s is unique up units; in other words, the sequence of the ideals
ppi qA of unique.
v
The isomorphism classes of the modules A{pi i A, as is the number of times
each occur in (11.1), are thus invariants of M, but the direct summands, re-
garded as submodules of M, are no more unique than bases are for vector
spaces.
These classification results hinge on two facts, that any submodule of a free
module is free, and that any matrix a with entries from A can be diagonalized;
to be precise it can be expresses as a matrix product a “ bdc where d is
diagonal and b and c invertible.
The decomposition theorem persists being true over quotint rings R “
A{pqqA of a pid A. A finite generated module over R is just an A-module that
is killed by q. Remarkably the decomposition theorem prevails for any module
over R; finitely generated or not. Frequrntly the teroem is atted for so called
A-module being of of bounded exponent which is synomymus with being killed
by an lement q.
n
q as q “ p1 1 . . . prnr with the pi ’different this amounts to M.

11.1 Elementary properties.

As indicated in the introduction a domain A is said to be a Bezout domain if Bezout domains (Bezout-
every finitely generated ideal is principal. Clearly a pid is a Bezout domain, and områder)

the converse holds when the Bezout domain is Noetherian.

11.1 A notion often met in algebraic number theory is that of associated
elements. Two elements a and b are associate when they are related by an in- Associate elements
vertible factor; in other words, when a “ ηb for a unit η in A. Of course, being (Assosierte elementer)

associate is equivalent to generating the same principal ideal. For integers this
 principal ideal domains 225

is just the good old ˘-thing, and polynomials over a field are associate when
they differ by a constant factor; but of course, units in most rings are not that
simple to describe.
Being associate is evidently an equivalence relation, and the set of principal
ideals in A incarnates the set of equivalence classes. Two elements a|b if and
only if pbq Ď paq; so the divisibility lattice mod assocition equals the opposit of
the lattice of principal ideals.
11.2 Recall that a greates common divisor of a finite collection a1 , . . . , ar of Greates common divisor
elements from a ring A is an element d so that d|ai for all i, and if b is another (Største felles divisor)

ring element dividing all the ai ’s, then b|d. If a greatest common divisor exists,
it is unique up to multiplication by a unit.
Greatest common divisors do not necessarily exist in general, but in a
Bezout domain any finite set of elements has one. The ideal pa1 , . . . , ar q is in
that case principal, and any generator serves as a greatest common divisor of
the ai ’s. Indeed, if d is generator, it holds that d|ai since ai P pdq, and to see that
ř ř
b divides d when b|ai for all i, just write d “ i ci ai . Then d “ p i ci di qb where
the di ’s are elements with ai “ di b.
Ş
In a similar vein, a generator d of the intersection i pai q is a least common
multiple of the ai ’s. It is clear that ai |d. Moreover, if ai |b for all indices i, the Least common multiple
Ş
element b lies in the intersection i pai q, and is therefore divisible by d. In a (Minste felles multiplum)

Bezout domain the intersection of finitely many principal ideals is principal.

When the ring is a pid, this is automatic, but for general Bezout domains an
argument is needed. In the next lemma we prove it for the intersection of two
ideals; the general case follows by an easy induction.

Lemma 11.3 Let a and b be elements from a domain A. If pa, bq is principal, then
the intersection paq X pbq is principal.

Proof: Assume that pa, bq “ pdq and let x and y be ring elements such that
a “ xd and b “ yd. Moreover there are ring elements u and v such that
ux ` vy “ 1. An element s from the intersection paq X pbq is shaped like
s “ za “ wb, and then zx “ wy. Multiplying the relation ux ` vy “ 1 by z,
one obtains z “ zux ` zvy “ ypwu ` zyq, or z “ γy with γ “ wu ` zy. Hence
za “ γya and ya (which equals xb) generates the intersection. o
The end of this story is that we have established the following:

Proposition 11.4 In a Bezout domain every finite set of elements has as greatest
common divisor and a least common multiple.

11.5 As signalled in the introduction, the principal ideal domains are the
Noetherian members of the Bezout club.

Proposition 11.6 A pid is Noetherian. Consequently, a Bezout ring is a pid if and

only if it is Noetherian.
226 ma4200—commutative algebra

Proof: Assume that A is a pid. Let tpai qu be an ascending chain of ideals

in A and let a be a generator of their union. Then a lies in one of the pai q’s,
Ť
and at that point the chain stabilizes; indeed, paq Ď pai q Ď j pa j q “ paq so that
Ť
pai q “ j pa j q and consequently pai q “ pa j q for j ě i.
As to second statement, if a Bezout domain is Noetherian, every ideal is
finitely generated and hence principal. o

pid s and ufd s.

11.7 The notion of prime numbers can be generalized to elements in any

ring. A prime element in a ring A is an element a such if a divides a product, Prime elements (Primele-
it divides one of the factors; in other words, a relation like bc “ ya for some menter)

y implies c “ xa or b “ xb for some x. This is obviously equivalent to the

principal ideal paq being a prime ideal.
In a polynomial ring one has the notion of irreducible polynomials which
generalizes in the following way. An irreducible element a in a ring A is essen- irreducible elements
tially an element without other factors than itself. One can of course modify a (irreducible elementer)

by multiplying it with a unit, so the natural requirement for a to be irreducible,

is that a relation a “ bc implies that either b or c is a unit. In terms of ideals,
this is equivalent to the principal ideal paq being maximal among the proper
principal ideals.
11.8 In general being prime implies being irreducible; indeed, if a is prime
and a “ bc, it holds true that b “ xa or c “ xa, hence a “ xca and 1 “ xc
or a “ xba and 1 “ xb. The converse does not hold in general; but in Bezout
domains it holds:
Proposition 11.9 In Bezout ring being prime is equivalent to being irreducible.
Proof: Asumme that a is irreducible and that xa “ bc, we have to see that
a|b or a|c. The ring A being Bezout, it the ideal pa, bq is principal and obvi-
ously paq Ď pa, bq. By assumption a is irreducible so that paq is maximal among
the principal ideals. It follows that either paq “ pa, bq or pa, bq “ 1. In the
former case a|b and in the latter there is a relation ya ` zb “ 1, which upon
multiplication by c transfomez into pyc ` xqa “ c; that is, a|c. o

Corollary 11.10 In a pid every prime ideal is maximal.

Proof: Assume that ppq is a prime ideal. Let m be a maximal ideal containing
ppq; it is principal since the ring is pid. By Proposition 11.9 above p is irre-
ducible and hence ppq is maximal among the principal ideals; thus ppq “ m.
o
Example 11.1 A kind of generic example of irreducibles elements not being
prime is manifest in the ring krx, y, z, ws{pxy ´ zwq. Clearly, the class of x is not
a prime element, but for degree reasons, it must be irreducible. Of course, the
same holds for any of the other variables. See Problem 11.2 below. K
 principal ideal domains 227

Example 11.2 The simplest example of irreducibles not being prime is found
?
in the ring Zr ´5s. There the relation
? ?
2 ¨ 3 “ p1 ` i 5q ¨ p1 ´ i 5q

holds, so for instance, 2 is not a prime element. It is however irreducible, for

if 2 “ zw, one has 4 “ }2}2 “ }z}2 }w}2 . But being shaped like a2 ` 5b2 with
a, b P Z, both }z}2 and }w}2 are integers, and we infer that either }w} “ 1 or
}z} “ 1. Finally, a relation like a2 ` 5b2 “ 1 between integers entails that a “ 1
and b “ 0, and we may conclude that either z or w equals ˘1. K

Proposition 11.11 In a Noetherian ring any element is a product of finitely many

irreducible elements.

Proof: We consider the set Σ of “bad guys”: The set of principal ideals pxq so
that x is a counterexample to the assertion; or in other words, so that x is not
a product of finitely many irreducibles. If non-empty, the set Σ has a maximal
member, say paq. Then a is not irreducible, and we may write a “ bc with
neither factor being a unit. This means that paq Ĺ pbq and paq Ĺ pcq, and by
consequence, neither pbq nor pcq belongs to Σ as paq was maximal there. Hence
both b and c can be expressed as finite products of irreducibles, and the same
is true for a. Contradiction. o

Proposition 11.12 In a Bezout domain the factors in a finite product of irre-

ducibles are unique up to order and multiplication by a unit.

Proof: The proof proceeds by induction on the number irreducible factors.

ś ś
Assume that i pi “ j q j both products being finite and all the factors being
irreducible. By Proposition 11.9 above the factors are all prime elements, and
it follows that one of the pi ’s divide one of the q j ’s, and after reordering the
factors we may assume that p1 “ aq1 . Now p1 is irreducible, so a is a unit,
and we may as well assume that p1 “ q1 . Cancelling p1 we have reduced the
number of factors, and we are done by induction. o
11.13 Combining the two previous propositions, we see that in principal ideal
domain one has unique factorization:

Theorem 11.14 Every pid is a ufd.

Proof: A pid is Noetherian so every element factors as a product of finitely

many irreducibles (Proposition 11.11). A pid being a Bezout domain, the factors
are unique up yo order and units (Proposition 11.12). o
Example 11.3 The theorem is not true for Bezout domains that are not
Noetherian; For example, the irreducible elements in the ring of holomor-
phic functions in a domain are (up to units) the simple linear functions z ´ a;
and finite products of such of course merely have finitely many zeros. In
228 ma4200—commutative algebra

any domain Ω Ď C there are functions with infinitely many zeros which can
not be products of finitely many irreducibles. In fact, a famous theorem of
Weierstrass’ asserts that for any sequence tan u of distinct points in Ω not accu-
mulating in Ω, there is a function vanishing at the an ’s and nowhere else; one
may even prescribe the order of vanishing at each an . K

Problems

11.1 Referring to Example 11.2 show that the three other involved numbers 3,
? ?
1 ` i 5 and 1 ´ i 5 are irreducible.

11.2 This exercise gives a general version of Example 11.1. Assume that
À
A “ iě0 Ai is a graded integral domain with A0 being a field. Assume
there are four homogeneous elements of degree one x, y, z and w that satisfy
xy “ zw. Show that x is irreducible but not prime. Hint: Assume that
there is a factorization x “ pq. Work with the non-vanishing homogeneous
components of p and q of highest degree.

11.3 Let A “ Zrx, y, z, ws{pxy ´ wzq show that the class of x is irreducible but
not prime.

11.4 (Euclidean domains.) A domain A is said to be Euclidean if there is func- Euclidean domains
tion δ on A assuming values in the set N0 of non-negative integers that has (Euklidske områder)

the following property:

o For any pair x and y with y ‰ 0 there are elements q and r in A so that
x “ yq ` r and either δprq ă δpyq.

Show that a Euclidean domain is a pid. Hint: An ideal a will be generated by

an x minimizing δ over the non-zero members of a.

11.5 Show that the conclusion that A is a pid in the previous exercise persist
when in the definiton of δ the set N0 is replaced with any well-ordered set W .

11.6 Let n P Z and put δpnq to be the number binary digit of }n}. Show that
δ is the smallest Eucliden function on Z. The article1 is a nice paper about 1

Eucliden algorithms.

11.7 Show that Zris is Euclidean.

?
11.8 Show that Zr ´ds is Euclidean for d “ 1, 3, 7, 11.

11.9 Let r ą 0 be a real number. Show that set of power series that converge
for }z} ą r is an Euclidean ring. Hint: The number of zeros in the disk
t z | }z} ď r u is a Euclidean function.

M
 principal ideal domains 229

11.2 Some general facts

Torsion modules
Let A be a domain and let M by any module over A. Recall that a torsion
element in A is an element killed by some non-zero member of the ring. Torsion elements (Torsjon-
Cleary the sum of two torsion elements is a torsion element (if a kills x and selementer)

b kills y, the product ab kills x ` y) as is cx for any c P A. In other words, the

torsion elements form a submodule TpMq of M:

TpMq “ t x P M | am “ 0 for a non-zero a P A u.

The torsion module depends functorially on M, and it sits in the short exact
sequence
0 / TpMq /M / M{TpMq / 0.

The quotient M{TpMq is clearly torsion free; if ax is torsion, x will be. A

module M is called a torsion module if T “ TpMq, and M is said to be primary Torsion modules (Torsjon-
if x P M ar killed by powers of a single prime p; that is, for some integer smoduler)

pν x “ 0 for every x P M. The actual power need for the killing depends on x,
and need not be bounded. For intstance when p is a prime number, the group
Z p8 has elements of every odre a power of p. The annihilator of such modules
is zero eventhough the annihilator over single element has ppqA as radical.

Lemma 11.15 Assume that A is a pid. A torsion module T over A decomposes as

À
the direct sum T “ p Tp of the primary modules Tp , where the sum extends over all
prime ideals ppq in A.

Proof: Obviously Tp X Tq “ 0 when p and q are realtively prime; Writing

1 “ αp ` βq gives x “ αpx ` βqx “ 0 whenever x belongs to the intersection.
Let x be any element in T and assume that ax “ 0 with a ‰ 0. As A is a
ś ν
ufd, there is a factorization a “ i pi i of a into a product of powers of distinct
ś ν
irreducible elements. The crucial fact is that the products q j “ i‰ j pi i do not
have common factors for different j’s; hence there is relation

1 “ c1 q1 ` . . . ` cr qr .

And from ensues upon multiplication by x the relation

x “ c1 q1 x ` . . . ` cr qr x.
vj νj
Now a “ q j p j , so that each summand c j q j x is annihilated by p j and therefore
belongs to Tp j . o

Example 11.4 The abelian group Q{Z is a typical infinite torsion group, and
the p-torsion part equals the group Z p8 “ Zrp´1 s{Z. Indeed, that a rational
number x satisfies pν x P Z, means that x “ a{pv for some a P Z. It follows
230 ma4200—commutative algebra

that there is a decomposition Q{Z “ p Z p8 where the sum extends over all
À

primes.
Verbatim, this reasoning works for any principal ideal domain A, and one
has a decomposition
à
K{A “ A p8
p

where K is the fraction field of A and A p8 “ Arp´1 s, and where the sum is
taken over all prime ideals ppq. K

11.3 Finitely generated modules

Principal ideal domains enjoy the property that all projective modules are free.
It holds unconditionally, but we shall prove it merely for projective modules
of finite rank to avoid diving into the deep waters of transfinite induction
arguments. Moreover, any submodule of a free module is free which is a
rather rare property for a ring to have. Consequently any module has a free
resolution of length two; that is, it sits in an exact sequence of shape

0 / Am / An /M / 0,

where in general n and m designate possibly infinite cardinal numbers. When

M is finitely generated however, both are finite, and M can be described as the
cokernel of a matrix Φ with entries in A. Subsequently we shall establish that
Φ may even be taken to be diagonal, and the decomposition theorem ensues
directly from this.
Again we shall work with Bezout rings. All the above results follow, but for
modules of finite presentation. An A-module M is of finite presentation if it
sits in an exact sequence

Am / An /M / 0.

In other words it is finitely generated and one may chose the generators in
way that the relations between them are finitely generated as well. Over a
Noetherian ring every submodule of An is finitely generated, hence being
finitely generated is equivalent to being finitely presented.

Free and projective modules

Recall that the rank of a module M over a domain A is the dimension dimK Mb A K
of Mb A K as a vector space over the function field K of A. Since localization
is an exact operation, the rank is additive over exact sequences, and a free
module is of rank n precisely when having a basis with n elements.
Recall also that if A is a domain and M is a torsion free A-module, the
canonical map M Ñ Mb A K is injective; indeed, the kernel consists of those
 principal ideal domains 231

elements m in M that are killed by a non-zero divisor in A, and as M is torsion

free, there are no such a part from zero. This leads to the following

Lemma 11.16 Assume that M is a finitely generated torsion free module over the
domain A. Then there is non-zero A-linear map φ : M Ñ A.

Proof: Start by choosing a non-zero K-linear map φ̃ : Mb A K Ñ K. The

module M is contained in and spans the K-vector space Mb A K. Therefore φ̃
does not vanish on M, but of course it does not necessarily assume values in
A. To achieve this let m1 , . . . , mr be generators for M and let a be a common
denominator for the images φ̃pmi q. Then φ “ a ¨ φ̃ does the job. o

Theorem 11.17 Assume that A is a Bezout domain. Then every finitely generated
submodule of a finitely generated torsion free A-module E is free. In particular, every
finitely generated torsion free A-module is free.

Torsion free modules over a pid which are not finitely generated is a devilish
class, and even for abelian groups there are large white areas on the map.
For instance, torsion free subgroup contained in Q2 seem to be impossible to
classify. However those contained in Q are rather easy to lay hands on (see
exercise xxx).
The case of principal ideal domains merits is own announcement. The
finitely general case i cover by the theorem above, but it holds without hy-
potheses on the projective modules—in fact almost Noetherian domains
projective modules that are not finitely generated are free.

Corollary 11.18 Assume that A is a pid. Then every projective module is free and
every submodule of a free module is free.

Proof of Theorem 11.17: Since E is a submodule of itself, the second asser-

tion ensues from the first. The proof of the first proceeds by induction on the
rank of the projective module E. Let F Ď E be a submodule.
Choose any non-zero A-linear map π : E Ñ A that does not vanish on F. Its
image is a finitely generated ideal which is principal because A is Bezout. But
principal ideals are isomorphic to A as A-modules, and consequently we have
a surjective A-linear map π : E Ñ A. This settles the rank one case, as π will be 0 0

an isomorphism in that case.

 
In a similar vein, the image of F under π is an ideal which is finitely gen- 0 / ker ρ / ker π
erated and thus principal, say generated by a. Then we may write π|F “ a ¨ ρ
 
where ρ is a surjective map F Ñ A. Lemma xxx ensures that both π and 0 /F /E
π|F are split surjections, and there are decompositions E » A ‘ ker π and ρ π
F » A ‘ ker ρ.  
0 /A /A
a
Now, ker π being a direct summand in a finitely generated projective mod-
ule, is finitely generated and projective; moreover, its rank is obviously one  
0 0
less than that of E. By induction we infer that ker ρ is free, and consequently F
is free in view of the isomorphism F » A ‘ ker ρ. o
232 ma4200—commutative algebra

11.19 Just being slightly more careful in the proof the lemma can be general-
ized to
Lemma 11.20 Let A be Prufer domain, then every finitely generated torsion free
A-module is isomorphic to a direct sum of projective modules of rank one.

Recall that a Prufer domain is an integral domain with all finitely generated
ideals being projective. Dedekind domains are of this sort, and of course
principal ideal domains as well, their ideals are even free modules. In the
dictionary between algebra and geometry, projective modules correspond to
vector bundles. Hence on the affine curve X “ Spec A, any vector bundle
decomposes as the sum of line bundles. This is far from being the case on
projective curves, the projective line being the sole exception.
Proof: Induction on the rank will do. The rank one case being the hypothesis
that A be Prufer. Once we have non-zero map

EÑA

the kernel decomposes as a direct sum of rank one projectives, and since E is
finitely generated the image is a projective ideal. Hence E » I ‘ ker π, and
ker π is finitely generated projective. The ranks has dropped by one, and the
induction hypothesis applies to ker π.
So have to infer that E˚ has a non zero element, but this is just xxxx, any
element E that maps to Eb A K is torsion. o

Problem 11.10 Assume that M is a finitely generated graded A-module that

is torsion free. Show that M is free; i.e. M is isomorphic to a finite sum M »
À
i Apni q Hint: Copy the proof of lemma xxx and notice that homogeneous
ideals in krx, ys are principal. M

The fundamental theorem

With basis in the lemma we esatbilished in the previous paragraph, we shall
establish that any A-lineaer map between two free A-modules can with ap-
propiate choices of bases in two module be represented by a diagonal matrix.
any matrix with enries in F can be diagonalise. The proof we present is to-
tally elemnetary and goes back to Wedderburn who showed the teorem for
matrices of holomorphicx function.
The assumpition that A is Bezout does no at complications,
But in the wek sens
he classification ensues from the a priori stronger result that matrices with
entries in a pid can be diagonalized. Indeed, as mentioned in the beginning of
the section every finitely presented module is the cokernel of an A-linear map
between free modules.
11.21 Our first lemma is a criterion for when an element in a free module is
part of a basis:
 principal ideal domains 233

Lemma 11.22 Assume that A is a Bezout domain. Let E be a free A-module of finite
rank and let e P E be an element. Assume there is an A-linear map π : E Ñ A with
πpeq “ 1. Then e is part of a basis for E.

Proof: The kernel of π is free by Proposition 11.17, so pick a basis tei u1ďin´1
for it. We contend that joining e to that basis we obtain a basis for E. Applying
ř
π to a linear dependence relation i ai ei ` ae “ 0 immediately gives a “ 0, so e
together with the ei ’s constitute a linearly independent set. They also generate
E since for any x P E the element x ´ πpxqe lies in ker π, and hence is a linear
combination of the ei ’s. o

Proposition 11.23 Let E and F be free A-modules of the same rank. Let φ : F Ñ
E be an injective linear map. Then there are bases t f i u for F and tei u for E and ring
elements λi so that φp f i q “ λi ei .

In the case A is a field and E and F are vector spaces of the same dimension,
the proposition just says that the image of basis under an injection is a basis;
in fact, in that case, the matrix may be taken to be the identity matrix. The
result is far less subtle than any existence theporem for eigenvalues.
Proof: Pick an element f P F that is part of a basis t fˆi u for F, and chose a
basis têi u for E. These two bases are auxiliary and not the ones we shall end
up with—hence the hat. Anyhow, φp f q may be expressed in terms of the êi ’s;
ř
that is, it can be written as a linear combination φp f q “ j ai êi with the ai ’s
from A. The ideal pa1 , . . . , an q generated by the ai ’s is principal since A is a
Bezout domain, and let us say it is generated by the element d.

/ ker π
φ|ker ρ
o The case d “ 1. ker ρ
ř
This means that there are ring elements ci so that i ci ai “ 1. Define an A-
 
φ
/E
ř ř
linear map π : E Ñ A by sending an element i xi êi to i ci xi . Put ρ “ π ˝ φ; F
then by the choice of the ci ’s, it holds true that ρp f q “ 1, and the maps ρ and ρ π
π are both split surjections. Their kernels are therefore free (Proposition 11.17)  
A
id A
/A
and ranks have dropped by one. By induction the two kernels have bases
t f i u1ďiďn´1 and tei u1ďiďn´1 respectively so that

φp f i q “ λi ei (11.2)

for 1 ď i ď n ´ 1 and for approriate ring elements λi . By Lemma 11.22 above

adjoining e “ φp f q and f respectively to the bases tei u and t f i u we obtain
bases for E and F. Evidently the relations in (11.2) persist, and by construction
φp f q “ e.

o The genral case.

The idea is to factor φ as a map falling under the first case and one which
apriori has diagonal martxi. Assume then that d ‰ 1 then there are ring
234 ma4200—commutative algebra

elements gi so that ai “ gi d and pg1 , . . . , gn q “ A. Define a new map ψ : F Ñ E

by ψp f q “ i gi êi and ψp fˆi q “ φp fˆi q. Then ψ satisfy the hypothesis of the first
ř

case, and there are bases tei u and t f i u like in the first case. Now define a map
σ : F Ñ F by putting σp f q “ d f and σp f i q “ f i for the rest of the basis f i . Then
it holds true that φ “ ψ ¨ σ and we are trhough. o

Theorem 11.24 Let E and F be two free A-module of finite rank, and φ : F Ñ F an
A-linear map. The there are bases for F and E so that the matrix is semi-diagonal.

Semi-diagonal means that the only non-zero elements of the matrix are situ-
ated on the diagonal that emanates from the upper left corner. Of course when
matrix is squar, this is just being diagonal.
Proof: The image of φ is a free module after Ppropostion ?? as is the kernel,
and hence F »P φ ‘ ker φ. The map φ factors trhroug an injective φ̃ : im φ Ñ E,
and by Proposition xxx there are bases for P φ and E relative to which φ̃ has a
diagonal matrix. Lift these basis elements to F and complement them with a
basis for ker φ to obtain a basis for F of the right sort. o

Corollary 11.25 Let A be a Bezout domian. Then everyr finitely presented A-

module has a decompositiob in cyclicv modules

à ν
M » Aν ‘ A{pi i A
i

where ν is a non-negative integer and where the pi ’s are irreducidble elements in A

and the νi ’s natural numbers.

Proof of Proposition ??: For simplicity we assume that F is of finite

rank; say n, and proceed by induction on n. The assumption that A is a pid
immediatly solves the case that n “ 1; indeed, the image of φ is an ideal in A
which is principal as A is a pid, hence it is isomorphic to A as an A-module.
As to the induction step, the trick is choose an A-linear map π : F Ñ A in
the following optimal way. The images im π ˝ φ where π runs through the
surjective maps π : F Ñ A, form a non-empty set of ideals which, since A
is noetherian, contains a maximal2 element. We let π0 be a map realizing a 2
It is not necessarily
maximal image and let a be a generator for im φ ˝ π0 , or in other words, φ ˝ π0 a maximal ideal, but
merely maximal
factors as aρ for some surjective A-linear map ρ : E Ñ A. among the ideals under
These maps enter into a selfexplanatory commutative diagamme3 with exact consideration.
3
If you don’t find it self
explanatory: E1 and
F1 are the kernels of ρ
and π0 ; the column to
the right is made up of
cokernels and is exact
by the snake lemma
 principal ideal domains 235

rows and column

0 0 0
 φ1  
0 / E1 / F1 / M2 /0
  
/E /F /M /0
φ
0
ρ
  π0 
0 /A /A / M1 /0
a
  
0 0 0

By lemma xxx on page xxx we infer that the middle column is split so that
E » F1 ‘ A, and it follows that F1 is projective. By lemma xxx from this ensues
that F1 is free, and obviously, its rank is one less than F. Hence the induction
hypothesis applies, and we can conclude that E1 is free. Now, the leftmost
column is also split and E “ F1 ‘ A. Thus E is free and the first assertion is
established.
Then we attack the assertion about the bases, and we denote the rank of E
by M and that of F by n. By induction, there are bases e1 , . . . , em´1 for E1 and
f 1 , . . . , f n´1 for F1 such that φ|E1 sends ei to ai f i with ai P A We extend these
bases to bases for E and F by respectively adding an element em of E such that
ρpem q “ 1 and and element f n of F with π0 p f n q “ 1.
The crucial observation is that for any projection π : F Ñ A, it holds true
that πpρpem qq P paq. Indeed, assume not, and let b “ πpρpem qq. Then pa, bq “ 1,
and one may write αa ` βb “ 1 for appropriate elements α and β in A. Then
απ0 ` βπ sends em to 1, contradicting the maximality of paq.
Applying this observation to the projections onto the different basis ele-
ř
ments f i , leads us to an expression φpem q “ a f m ` i ci a f i for ci ’s in A, and
ř
simply replacing f m by f m ` i ci f i , we have the requested basis element. o
In fact observing that if E “ A ‘ M1 and M is finitely generated, if follows
that M1 is finitely generated. This follows for instance, by decomposing id M “
η1 ` η2 into a sum of two oprthogonal idempotents, such that η2 pMq “ M1 .
Å lese: Lam, T. Y. (2006), Serre’s Problem on Projective Modules, Springer Monographs in
Mathematics, Berlin, Heidelber
Ât’