Geir Ellingsrud

16 lectures on
COMMUTATIVE ALGEBRA
An introduction

Fell Forlag
Contents

Lecture 0: Intro 9

Lecture 1: Rings 13
Rings 14
Polynomials 21
Direct products and idempotents 25

Lecture 2: Ideals 29
Ideals 30
Kernels and quotients 35
Prime ideals and maximal ideals 39
Primes and irreducibles 44
Existence theorems 48
Local rings 53
Direct products and the Chinese Remainder Theorem 56
Graded rings and homogenous ideals 58
The prime spectrum and the Zariski topology 62

Lecture 3: Unique factorization domains 67

Being a unique factorization domain 67
Common divisors and multiples 72
A criterion of Kaplansky’s 73
Gauss’ lemma and polynomials over factorial rings 73
Example: Quadratic extensions and the norm 80

Lecture 4: Modules 85
The axioms 85
 Direct sums and direct products 92
Finitely generated modules 97
Bases and free modules 99
Graded modules 104
Nakayama’s lemma 105
Appendix: The determinant and the characteristic polynomial 108
Appendix: Unions of submodules and direct limits 113

Lecture 5: Exact sequences 121

Exact sequences 121
Left exactness of hom-functors 129
Snakes and alike 135
Complexes 139

Lecture 6: Tensor products 147

Introducing the tensor product 147
Basic working formulas 149
Functorial properties 156
Change of rings 161
Tensor products of algebras 164
Appendix: Flatness 168

Lecture 7: Localization 173

Localization of rings 174
Localization of modules 190
Local properties 196
An extended version of Nakayama’s lemma 198
The support of a module 199
The rank of a module 202

Lecture 8: Projective modules 203

Projective and locally free modules 204
Working formulas 207
The Picard group 210
Examples 215

Lecture 9: Chain conditions 229

 Noetherian modules 229
Noetherian rings 233
A structure theorem for modules 237
Hilbert’s Basis Theorem 239
Krull’s intersection theorem 242
Modules of finite length 245
Artinian rings 253

Lecture 10:Primary decomposition 259

Primary ideals 260
The Lasker-Noether theorem 265
The Uniqueness Theorems 268
The homogeneous case 276
The case of monomial ideals 279
Primary decomposition of modules 282
Appendix: Primary decomposition and group actions 284

Lecture 11:Krull dimension 287

Definition and basic properties 288
Krull’s Principal Ideal Theorem 293
UFD’s once more 297
System of parameters 299
Dimension of polynomial rings 302

Lecture 12:Integral extensions 315

Definition and basic properties 315
Examples 324
The Cohen–Krull–Seidenberg Theorems 327
Then finiteness issue 334
Appendix: Trace and separability 336

Lecture 13:Algebras of finite type over fields 341

Noether’s normalization lemma 341
Hilbert’s Nullstellensatz 345
Consequences 350

Lecture 14:Examples of unexpected rings 357

8 CONTENTS

A Noetherian ring of infinite dimension 357

A polynomial ring of excess dimension 359
A Noetherian ring that is not catenary 361
A Noetherian bubble space 364

Lecture 15:Dedekind rings 371

Discrete valuation rings 372
Normal domains and discrete valuation rings 380
Dedekind rings 383
Finitely generated modules over Dedekind rings 392
Appendix: General valuations 395

Lecture 16:Hilbert functions 403

Numerical polynomials 404
Multiplicities and Hilbert–Samuel functions 421
Graded rings and modules 421
Filtrations, the Artin–Rees lemme and Samuel functions 424

Lecture 17:Regular sequences 427

Depth, regular sequences and unmixedness 427

Lecture 18:Categories 435

Lecture 19:Solutions 441

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 0

Intro

Warning: This is a preliminary version. I am working on them and new (and hopefully)
better versions will surface from time to time. They still suffer from several shortcomings
and are prone to errors (not so many and not so serious, I hope), but they will (hopefully)
improve! Some sections are thoroughly checked while others are raw and under
construction and they are marked with a warning sign: “careful—construction!”.
Version 4.0 (run 30)—10th January 2021 at 11:32am
These notes grew out my giving the introductory course in commutative algebra at
UiO at several occasions during the last ten years. This is course where the students
meet serious commutative algebra for the first time. Their backgrounds are diverse.
They know some linear algebra, but mostly not from a theoretical standpoint, and very
few have come as far as the Cayley–Hamilton Theorem. They have had a rudimentary
experience in commutative algebra, and have heard about rings and ideals and have
seen some examples, but to indicate their level, most do not know Gauss’ lemma. Most
have followed a course in group and Galois theory, which occasionally goes as far as
the Sylow’s theorems, and which include basic Galois theory. Given these conditions,
the notes starts at the very beginning with the very basic properties of rings and ideals
With that starting point the theory is developed introducing the fundamental con-
cepts and techniques; in short, a guide to a beginners tools necessary to start off
practising commutative algebra; And of course, subsequently this leads to the usual
collection of the “great theorems” of David Hilbert, Emmy Noether and Wolfgang Krull;
the corner stones of the whole theory. A primary function of the course is to prepare
the ground for studies in algebraic geometry and number theory,
Being a preparatory course, there is a risk it leaves you with a lurking melancholy
as expresses in the lyrics from Leonard Bernstein’s song Some Other Time: “just when
the fun is starting comes the time for parting”( so beautifully performed by Monica
Zetterlund and Bill Evans). But there is a cure: Do more mathematics!
The notes are written for the students. The style is rather ample with detailed
10 intro

explanations, which makes the text rather long. But redundancy of the language is an
important factor in making a text accessible and easy to read, however redundancy
without variation is futile if not contra productive. Remember the french saying: when
you complain you don’t understand what the British say, they just repeat the phrase but
louder. There is also a gradient in the redundancy—as the course evolves more details
are left to the students.
Categories and functors entered mathematics in the 1940’s work of Eilenberg and
Maclane, and as Peter Freud states “in a fairly explosively manner functors and natural
transformations permeated a wide variety of subjects”. To give a master course in
algebra today—about eighty years later—in an attitude that the word category is a slip
of the tongue, would be close to a heresy. But categories and functors do not enter the
presentation in a substantial way, they only appear as notational devices, except at a few
placed a mild use of easy categorical techniques will be convenient and clarifying. And
for the benefit of the students a very short appendix is included with the rudimental
definitions.
Mathematical theories are not linearly ordered, but rather constitute some kind of
intricate graph with nodes being statements and edges being implications. But time is
linear, and a challenge for a lecturer is to find a path through this mathematical skein
valid both scientifically and pedagogically. And speaking about time, to reach through
the curriculum before the term ends, can be sever. The notes suffer from the common
syndrome of a never ending expansion, and threaten to end up obese, so any lecturer
that might chose to them is obliged to make a reasonable choice among the chapters.
Giving examples is an important part of the teaching, establishing a broad back-
ground for the students intuition. So examples abound, some are mainstream situations,
but others function as eyeopeners: they are meant to illustrate what delicate situation
one risks finding oneself in and what denizens one risks meeting when venturing the
stormy waters where the standard hypotheses of the theory no more comply. It is also
important to understand why the specific hypotheses of a result are required, and often
this is best illustrated through examples.
Doing exercise are as well a fundamental when learning mathematics; so we include
almost four hundred (397 to be exact)—some are easy and some more demanding.
Solutions are provided for many (for the moment only 99 exercises from the first
chapters are solved; the solution part is still a construction cite), and they are indicated
ˇ by a golden star in the margin. To jump forth and back between an exercise and its
solution just click on the numbers. A habit of many authors is to bundle up parts of the
theory with the exercises, not (always) out of laziness, but most often as an attempt to
limit the number of pages. Anyhow, from the pedagogical angle it is sound practice to
force students to participate actively in developing the matter. So, also in these notes
some exercises are part of the theory. Ideally, solutions should always be provided for

10th January 2021 at 11:32am

Version 4.0 run 30
 11

these exercises (and eventually will).

Definitions are not, as in many texts, formatted in special typography to stand out, Definition (definisjon)
but are indicated in the margin in blue with the Norwegian version of the name in red.
An insurmountable means of keeping track of maps and equalities between com-
positions, is to draw diagrams; a simple one is shown in the margin . Diagrams come ?B
all kinds of shapes and can be utterly intricate, but for us they will always be simple β

and mostly triangular or square. One says that a diagram, or a part of a diagram, is
commutative if possible equalities hidden in the geometry of the diagram, in fact are A γ

equalities; so for example α = γ ˝ β in the marginal example. Almost every diagram we

 
α
shall draw will be commutative, and we shall tacitly assumed they are.
C
So a few words about the gothic alphabet, which poperly should be called the
blackletter typefaces. Mathematicians are always in shortage of alphabets and letters, and
tend to use all kinds of creative decorations to have enough glyphs. To overcome this
typographical shortfall we have decided to designate ideals by a blackletter typeface.
There is certainly a very strong historical evidence—a reminder of the deep German
roots of algebra—and it is also convenient in many ways: like a team in uniforms you
recognize ideals immediately. However, the inconvenience is that blackletter letters can
be difficult to write by hand, and handwritten ones difficult to read, especially the p and
the q stand out in this respect. We confined the use to the lower case letters a, b c, m, n,
p, q, and may be occasionally an r. Letters like I and J mostly denote sets of indices, A,
B and C and from time to time R and S, are rings and while M and N will be modules.

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 1

Rings

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 rings, and ideals will be introduced in the next chapter.
Commutative rings come in a great variety of flavours, and the sources where they
arise are as diverse. Some rings are best though of as “number systems” as 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 may be brought at least
two steps further, but in a non-commutative way; the next two members are called
the quaternions and the octonions). There are also some ubiquitous “little brothers”;
the rings Z/nZ of integral residue classes modulo a natural number n. Among them
we find the finite fields F p = Z/pZ with p elements, p being a prime, and there are
also the other finite fields Fq with q elements, q being a prime power q = pr . .. And
there are naturally also some “big brothers”; for instance, the field Q
s consisting of the
complex numbers that are roots of polynomials with rational coefficients; and inside Q̄
we find the the ring Zs of algebraic integers; the complex numbers being roots of monic
polynomials with integral coefficients.
The earliest systematic study of commutative rings was of various “generalized
number systems”; certain subrings of the ring of algebraic numbers. Already Gauss
undertook such studies, but they really sparked off in the nineteenth century with the
work of Kronecker and Dedekind.
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 with values
in a field k defined on so-called varieties, which are vanishing loci in kn for sets of
polynomials.
The development took a new direction around the middle of the twentieth century,
 14 rings

when mathematicians like Zariski and Weil strived for establishing a sound foundation
of algebraic geometry, and the recognition of the power algebraic geometric methods
have in number theory, eventually lead to the happy marriage of algebraic geometry
and number theory — consummated by Grothendieck and his invention of schemes.

1.1 Rings
(1.1) Recall that a ring A 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 both many non-commutative rings and non-associative rings that are extremely
interesting, but this course is dedicated to rings that are associative and commutative.
The sum of two elements will naturally 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 as ab. The left distributive law asserts that a(b + c) = ab + bc, and since rings for
us are commutative, it follows that the right distributive law (b + c) a = 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 ¨ 1 A = a for all members a of the ring. At most occasions the reference to
A will be dropped and the unit element written as 1 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 rational
numbers Q, of real numbers R and of complex numbers C are well-known rings. K

Zero divisors and nilpotents

(1.2) Elements in general rings can behave quite differently from what we are used
to in a classical setting of real and complex numbers. It might very well happen that
Zero divisors ab = 0 without neither a nor b being zero. Such elements are called zero divisors. Be
(nulldivisorer) aware that the familiar cancellation law does not hold in a ring with zero divisors in
Integral domains that ab = ac not necessarily implies that b = c. Rings without zero divisors are called
(integritetsområder) integral domains or, for short, domains. Obviously, elements that are not zero divisors are
Non-zero divisors called non-zero divisors, another name being regular elements. A regular element a has
(ikke-nulldivisorer) the virtue that xa = 0 implies that x = 0 and can therefore be cancelled from equalities
Regular elements like ab = ac (the difference b ´ c, being killed by a, vanishes). So in an integral domain,
(regulære elementer) the cancellation law is in force.
For instance, the rings Z/nZ may have zero divisors. This happens whenever n is a
composite number. If n factors as n = pq with p and q natural numbers both different

10th January 2021 at 11:32am

Version 4.0 run 30
 rings 15

from n, it holds true that pq = 0 in Z/nZ, and p and q are both non-zero in Z/nZ
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 powers of non-zero elements vanish, i. e. one has an = 0 Nilpotent elements
for some natural number n with a ‰ 0. For instance, in the ring Z/p2 Z it holds true (nilpotente elementer)

that p2 = 0, but p ‰ 0. Such elements are called nilpotent. Rings deprived of nilpotent Reduced rings
elements are said to be reduced. (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 Units (enheter)
multiplication, which we shall denote 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 performed Fields (kropper)
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 exceptional
ring is the so-called null-ring. If it holds that 0 = 1, one has a = a ¨ 1 = a ¨ 0 = 0, so The null-ring
zero will be the sole element. The only role the null-ring plays and the only reason not (nullringen)

to throw it over board, is that it allows significantly simpler formulations of a many

results, and it does not merit a proper notation (well, one always has the alternative 0).
(1.4) The set of polynomials Q[ x1 , . . . , xr ] in r variables x1 , . . . , xr with rational coeffi-
cients is a ring when equipped with the usual sum and product, as are the set of real
polynomials R[ x1 , . . . , xr ] and the set of complex polynomials C[ x1 , . . . , xr ].
(1.5) The complex rational functions in a variable x form a field C( x ). The elements are
meromorphic functions in C expressible as the quotient p( x )/q( x ) of two polynomials
p and q with q not being identically zero. One is not confined to just one variable; the
field C( x1 , . . . , xr ) of rational functions in the variables x1 , . . . , xr consists of fractions
p( x1 , . . . , xr )/q( x1 , . . . , xr ) where p and q are polynomials and where q is not the zero
polynomial.
(1.6) For any set X Ď Cr one may consider the set of polynomial functions on X; that is, Polynomial functions
the functions on X that are restrictions of polynomials in r variables. They form a ring (polynomiale
funskjoner)
AC ( X ) under point-wise addition and multiplication.
10th January 2021 at 11:32am
Version 4.0 run 30
 16 rings

(1.7) Associated with any topological space X are the sets CR ( X ) and CC ( X ) of continu-
ous 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 instances are the ring of smooth
functions on a smooth manifold, and the ring O(Ω) of holomorphic functions in an
open domain Ω of the complex plane.
(1.8) Quadratic extensions: An example of a class of rings, important in algebraic number
? ?
Quadratic extensions theory, is the class of the quadratic extensions Z[ n] = t a + b n | a, b P Z u, where n is
(kvadratiske utvidelser) any integer (positive or negative). These rings are contained in the field of complex
numbers C and inherit their ring structure from C; to verify that they are rings is suffices
to see they are closed under addition—which is obvious—and multiplication, which
ensues from the little calculation
? ? ?
( a + b n)( a1 + b1 n) = ( aa1 + nbb1 ) + ( ab1 + a1 b) n,

the point being that ( aa1 + nbb1 ) and ( ab1 + a1 b) are integers when a, a1 , b, b1 and n are.
Some special names are in use; for instance, elements of Z[i ] which are called Gaussian
integers.
K
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 mathematics is always accompanied by a new class of maps.
Categories (kategorier) This observation can be formalized and leads to the definition of categories.
Ring homomorphisms (1.5) In our present context the relevant maps are the so-called ring homomorphism, which
(ringhomomorfier) also will be referred to as maps of rings or ring-maps. These are maps φ : A Ñ B between
two rings A and B preserving all the structures around; that is, the additive group
structure, the multiplication and the unit element 1. In other words, they comply with
the two rules

o φ ( a + b ) = φ ( a ) + φ ( b );

o φ( ab) = φ( a)φ(b) and φ(1) = 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 composable ring-maps is a ring-map.

10th January 2021 at 11:32am

Version 4.0 run 30
 rings 17

The rings (commutative with unit) together with their homomorphisms form a category
denoted Rings.
(1.6) A homomorphism φ : A Ñ B is an isomorphism if there is a ring homomorphism Isomorphisms of rings
ψ : B Ñ A such that the two relations ψ ˝ φ = id A and φ ˝ ψ = idB hold true. One most (isomorfier av ringer)

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 φ´1 ( ab) = φ´1 ( a)φ´1 (b) is equivalent to ab =
φ(φ´1 ( a)φ´1 (b)), and the latter equality is a consequence of φ respecting multiplication.
Applying φ´1 to φ(1 A ) = 1B one sees that φ´1 (1B ) = 1 A , so the inverse map sends the
unit element to the unit element as well. An analogous argument shows that φ´1 also
is additive.
Examples
(1.9) So-called evaluation maps are omnipresent examples of ring homomorphisms. To
illustrate this concept, we pick a point a P Cn . Sending a polynomial f to the value
it assumes at a, gives a map C[ x1 , . . . , xn ] Ñ C, and by the very definition of the ring
structure of the polynomial ring (addition and multiplication are point-wise operations)
this is a map of rings.
Any ring A of functions—say with complex values—on any space X possesses
analogue evaluation maps. The operations in A being defined point-wise the map
f ÞÑ f ( x ) 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 [ a] modulo the integer n.
K
Subrings and polynomial expressions
(1.7) We begin by recalling the notion of a polynomial expression. Assume given a ring A
and sequence a = ( a1 , . . . , ar ) of elements from A. For any multi-index α = (α1 , . . . , αn );
that is, a sequence on non-negative integers, one has the monomial expression Monomial expression
(monomiale uttrykk)
α1
aα = a1 ¨ . . . ¨ aαnn .

These expressions show an exponential behavior in that aα ¨ a β = aα+ β . A polynomial Polynomial expressions
expression in the ai ’s is just a finite linear combination of such monomials. Frequently (polynomiale uttrykk)

one wants to confine the coefficients to a specific 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 S.

10th January 2021 at 11:32am

Version 4.0 run 30
 18 rings

A successive application of the distributive law and the exponential behaviour of

monomials gives the classical formula for the product of two polynomial expressions:
ÿ ÿ ÿ ÿ
( sα ¨ aα ) ¨ ( t β ¨ a β ) = ( sα t β ) ¨ aγ . (1.1)
α β γ α+ β=γ

Subrings (underringer) (1.8) A subring B of A is a ring contained in A whose ring operations are induced from
those of A. Phrased differently, it is an additive subgroup containing 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 family of subrings of A clearly is a 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 B[ a1 , . . . , ar ] of A as the set of all polynomial expressions
ÿ α
bα ¨ a1 1 ¨ . . . ¨ arαr

where α = (α1 , . . . , αr ) runs through the multi-indices and the bα ’s are elements from
B, only finitely many of which are different from zero. It is straightforward 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
Subrings generated by subring generated by the ai ’s over B, and is the smallest subring of A containing the ring
elements (underringer B and all the elements ai . Common usage is also to say that B[ a1 , . . . , ar ] is obtained by
generert av elementer)
adjoining the ai ’s to B.
This construction works fine even for infinitely many ai ’s since each polynomial
expression merely involves finitely many of them. Thus there is a subring B[ ai |i P I ] for
any subset tai uiP I of A. It equals the intersection of all sub rings of A containing B all
the ai ’s.
Examples
(1.12) Let n be an integer. The ring Z[1/n] = 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 Z[n´1 | n P S], 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 the integer n, it
1
holds true that Z[1/n] = Z[ p´ ´1
1 , . . . , pr ].
(1.13) The subring C[t2 , t3 ] of C[t] is a ubiquitous example in algebraic geometry; it
is the coordinate ring of a so-called cusp and consists of all polynomials whose first

10th January 2021 at 11:32am

Version 4.0 run 30
 rings 19

derivative vanishes at the origin; or phrased differently, the polynomials without a

linear term.
(1.14) The subring C[ x, 1/x ] of the rational function field C( x ) consists of elements of
the form p( x´1 ) + c + q( x ) where p and q are polynomials vanishing at the origin and
c a complex constant.
K
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
As is well known, a cyclic group is either finite and isomorphic to Z/nZ for some (primringen)

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 that A is of characteristic zero. So, The characteristic of a
in any case, the characteristic of a ring A is a non-negative integer attached to A. ring (karakteristikken
til en ring)
(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 The prime field
respectively Q is the prime field of A. (primkroppen)

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 may then form Algebras (algebraer)
products φ(b) ¨ a of elements a from A with elements of the form φ(b). 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 φ(b) ¨ 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 with the extra condition that multiplcation in A is B-linear.
Example 1.15 Every ring has a canonical structure as a Z-algebra (defined as in Para-
graph 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

10th January 2021 at 11:32am

Version 4.0 run 30
 20 rings

˚ Or any morphological
(1.13) Faithful to the principle that any new type of objects is accompanied by a
derivative thereof, like corresponding new type of maps; one says that a map of rings φ : A Ñ A1 between two
map of B-algebras or
B-algebra-map etc. B-algebras is an B-algebra homomorphism* if it respects the action of B; in other words,
it holds true that φ(b ¨ a) = b ¨ φ( a) for all elements a P A and b P B. Composition of
Algebra two composable B-algebra homomorphisms is a B-algebra homomorphism so that the
homomorphisms (alge- B-algebras form a category denoted Alg B .
brahomomorfismer)
(1.14) One says that A is finitely generated over B, or is of finite type over B, if A =
B[ a1 , . . . , ar ] for elements a1 , . . . , ar from A.
Finitely generated Example 1.16 A note of warning might be appropriate, algebra structures can be
algebras
deceptive. Every ring is of course an algebra over itself in a canonical way (the algebra
(endeliggenererte
algebraer) structure is given by the identity map), but there can be other unorthodox ways A can
Finite type algebras be an A-algebra. A simple example to have in mind is the field C of complex numbers,
(algebraer av endelig which has an alternative algebra structure induced by complex conjugation. In this
type)
structure a complex number z acts on another complex number w as z̄ ¨ w.
The two structures are not isomorphic as C-algebras although the underlying rings
are the same. A good try for an isomorphism would be the identity map, but it does
˚ Of course, it holds not respect the two algebra-structures*.
true that idC (zw) Similar unorthodoxy will arise from any ring endomorphism A Ñ A. Examples of
equals z idC w and not
z̄ idC w such are furnished by the Frobenius homomorphisms of rings of positive characteristic
(see Exercsise 1.7 below). K
Exercises
ˇ (1.1) Assume that A is a finite ring. Show that the units are precisely the elements that
are not zero divisors. Conclide that if A is an integral domain, it is a field.
ˇ (1.2) Find all nilpotents and all zero divisors in Z/72Z. What are the units?
ˇ (1.3) Generalize the previous exercise: Let n be a natural number. Determine nilpotents,
zero divisors and units in Z/nZ.
(1.4) Show that the prime ring is the smallest subring of a ring; i. e. it is contained in
all other subrings of the given ring.
(1.5) The Binomial Theorem. 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 prime characteristic p.
Show that the relation
( a + b) p = a p + b p
holds true for all a, b P A. Hence the map A Ñ A that sends a to the pth power a p is
a ring homomorphism. It is called the Frobenius homomorphism. Hint: The binomial
coefficients ( pr) have p as factor when 1 ă r ă p.

10th January 2021 at 11:32am

Version 4.0 run 30
 polynomials 21

ˇ (1.8) Show that any intermediate ring Z Ď A Ď Q is of the form A = Z[ p´1 |p P S] for
some set S of primes.
(1.9) Let φ : A Ñ B be a map of rings. Show that φ induces a group homomorphism
mapping A˚ into B˚ .
ˇ (1.10) Units in imaginary quadratic extensions. Let n be a natural number. Show that an
?
element x P Z[ ´n] is a unit 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
?
Z[ ´n].
ˇ (1.11) Assume that a is a nilpotent element of the ring A. Show that 1 + a is invert-
ible. More precisely: If an = 0, the inverse is given as (1 + a)´1 = 1 ´ a + a2 ´ . . . +
(´1)n´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.
M

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 introduced as functions
depending on a real (or complex) variable whose values were given by a polynomial
expressions. In this section we shall introduce polynomials with coefficients in any
(commutative) ring A. The point of view will necessarily be formal and without
reference to functions, and there is no reason to confine oneself to just one variable.
(1.15) In an earlier paragraph we met polynomial expressions in a set of ring elements.
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 x1 , . . . , xr is defined as a formal Polynomials
sum (Polynomer)
ÿ α
f ( x1 , . . . , x n ) = aα x1 1 ¨ . . . xnαn , (1.2)
α

where the summation extends over all multi-indices α = (α1 , . . . , αn ) 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 much* about what the ˚ Or do Exercise 1.19

term “formal sum” means, the essential point is that two such “formal sums” are equal below where a more
general construction of
exactly when corresponding coefficients agree. so-called monoidal
α
(1.16) The “pure” terms aα x1 1 ¨ . . . xnαn occurring in (1.2) are called monomials, and the algebras is described
α in a precise manner.
abbreviated notation x α = x1 1 ¨ . . . xnαn is convenient and practical. The degree of a
ř Monomials (monomer)
non-zero monomial aα ¨ x α is the sum i αi of the exponents, and the highest degree of
a non-zero monomial term in a polynomial is the degree of the polynomial. Non-zero The degree of a
constants are of degree zero, but the zero polynomial is not attributed a well-defined polynomial (graden til
et polynom)
degree—it is rather considered to be of any degree (it equals 0 ¨ x α for any α).

10th January 2021 at 11:32am

Version 4.0 run 30
 22 rings

Homogenous A polynomial is said to be homogenous if all its monomial terms are of the same
polynomials degree. For example, the polynomial x2 y + z3 is homogeneous of degree three whereas
(Homogene polynomer)
x2 y + z2 is not; it is still of degree three, but not homogeneous.
Every polynomial may be expressed as a sum of homogeneous polynomials of
different degrees—just recollect the homogenous terms with the same degree—and
Homogenous these are called the homogenous components of f . They are unambiguously associated
components of a with f .
polynomials (homogene
komponenter til et (1.17) Adding two polynomials is simply done term by term, and neither is there any
polynom) hocus-pocus about multiplying them. The good old pattern is followed where
ÿ ÿ ÿ ÿ
aα x α ¨ bβ x β = ( aα bβ ) x 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 A[ x1 , . . . , xr ] of polynomials in
the variables x1 , . . . , xr becomes a ring. Of course, there are axioms to be verified; a
tedious and uninteresting process without obstacles, so we voluntarily skip it (such an
indolence being reserved for professors, students are urged to do the checking).
ř ř
Exercise 1.12 Let f = α aα x α and g = α bα x α be two non-zero polynomials with
coefficients from the ring A such that f g is non-zero. Show that deg f g ď deg f + deg g.
Show that equality holds when A is an integral domain. Give examples where strict
inequality holds. M
(1.18) There is a notable difference between a polynomial and a polynomial function.
Over finite rings, like Z/nZ for instance, different polynomials can give rise to identical
polynomial functions. Simple examples being polynomials in F2 [t]; that is, polynomials
in one variable over the field F2 with two elements. For instance, such polynomials
without constant term and with an even number of non-zero terms will vanish identically
as a function on F2 . Over infinite fields however, the two notions coincide.

The universal mapping property

A universal mapping (1.19) The polynomial ring A[ x1 , . . . , xr ] has a so-called universal mapping property; one
property (en universell may freely assign values to the variables to obtain homomorphisms.
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 : A[ x1 , . . . , xr ] Ñ B such that φ( xi ) = bi for 1 ď i ď r.
ř
Proof: A polynomial p is given as p = α aα x α . Since the coefficients aα are unam-
ř α
biguously determined by p, setting φ( p) = α aα b1 1 ¨ . . . ¨ brαr gives a well-defined map
which is easily seen to be additive. Since a relation like the one in (1.3) is universally

10th January 2021 at 11:32am

Version 4.0 run 30
 polynomials 23

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 rather special property most algebras
do not have. For instance, the algebra C[t2 , t3 ] from Example 1.13 on page 18 does not
have it. That algebra has the generators t2 and t3 , and the equality (t2 )3 = (t3 )2 imposes
a constraint on the values a homomorphism φ can assume on the two generators: it
must hold true that φ(t2 )3 coincides with φ(t3 )2 (note that there is no such thing as
φ(t)). 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 an infinite number of
variables x1 , x2 , . . . , xn , . . ., although each polynomial merely involves finitely many of
the variables. For every n the polynomial ring A[ x1 , . . . , xn ] is obviously contained in
A[ x1 , . . . , xn+1 ], and these polynomial rings thus form a nested sequence of rings. The
polynomial ring in countably many variables A[ x1 , x2 , . . .] is just the union of all these.
It will also be denotede A[ xi |i P N].
Exercise 1.13 Convince yourself that the universal mapping property holds even for
polynomial rings in infinitely many variables. M
(1.22) The second type of rings we have in mind, are the rings of formal power series. A Rings of formal power
formal power series is an expression as in (1.2) series (ringen av
formelle potensrekker)
ÿ
f ( x1 , . . . , x n ) = aα x α1 ¨ . . . xnαn ,
α

except that the sum is not require to be finite (but the summation still extends over
multi-indices with the αi ’s being non-negative). 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.
The case of power series in one single variabel with coefficients in a field k, merits a
few comments. The units in the ring kJxK are precisly the series with a non-zero constant
term; i. e. those shaped like f ( x ) = a0 + a1 x + a2 + . . . with a0 ‰ 0. A potential inverse
series g( x ) = b0 + b1 x + b2 + . . . must in addition to a0 b0 = 1 satisfy the relations

a0 bn + a1 bn´1 + . . . + an b0 = 0 (1.4)

for n ě 1, simply because (1.4) expresses that the terms of degree n ě 1 of f g vanish.
But since a0 is invertible, bn is readily solved from (1.4) in terms of the ai ’s and the bi ’s
for i ă n, thus allowing the inverse g to be constructed by recursion.

10th January 2021 at 11:32am

Version 4.0 run 30
 24 rings

In courses about complex function theory it is taught that meromorphic functions

Formal Laurent series have Laurent series near a pole, and there is a formal analogue of those. So a formal
(formelle Laurent Laurent series with coefficients in k is a formal sum F ( x ) = iě´n ai x´i . It looks pretty
ř
rekker)
much like a power series, but the difference is that finitely many terms with negative
exponents are allowed. They form a ring k(( x )). Sums of two Laurent series are done
term-wise, and the coefficients of the product of iě´n ai xi and iě´m bi xi are defined
ř ř

by the formula
ÿ
cr = ai br´i ,
i

the sum extends a priori over all integers i, but in reality it is finite since br´i = 0 for
large i and ai = 0 for i small—some checking of axioms is of course necessary, but we
leave that to the industrious students.
Every element g( x ) is thus on the form x n f ( x ) where n P Z and f ( x ) is an invertible
power series, and it ensues that k(( x )) is a field. Clearly kJxK is contained in k(( x )), and
every element in k (( x )) is the quotient between two elements from kJxK. We say that
k(( x )) is the fraction field of kJxK.

Exercises
ˇ (1.14) Units in polynomial rings. Show that a polynomial f ( x ) = i ai xi in A[ x ] is
ř

invertible if and only if a0 is invertible and all the other coefficients are nilpotent.
Hint: Assume that f ( x ) = 1 + a1 x + . . . + an x n is invertible with inverse f ( x )´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.
(1.15) Let A be a reduced ring. Show that group of units in the polynomial ring A[ x ]
equals A˚ .
ˇ (1.16) Assume that k is a field and let t and u be variables. Show that there is no
injective ring homomorphism from k[t, 1/t] to k[u].
(1.17) Let k be a (finite) field. Prove that there is infinitely many monic irreducible
polynomials with coefficients in k. Hint: Mimic Euclid’s proof of the prime numbers
being infinitely many.
ˇ (1.18) Long division. Let A be a ring and b(t) a polynomial in A[t] whose leading
coefficient is invertible.
a) Show that long division by b(t) works in A[t]; that is, show that for any polyno-
mial a(t) P A[t] there are polynomials q(t) and r (t) with deg r ă deg b such that
a ( t ) = b ( t ) q ( t ) + r ( t ).
b) Assume that A is a domain. Conclude that the number of different zeros of a non-
zero polynomial in A[t] is less than the degree. In a later example (Example 2.11
on page 38) we shall be exhibit a counterexample when A is not domain.

10th January 2021 at 11:32am

Version 4.0 run 30
 direct products and idempotents 25

(1.19) Monoidal algebras. In this exercise the definition of polynomial rings is made
precise and generalized.
Let G be commutative monoid* written additively. As an abelian group A[ G ] is the ˚ A monoid is a set
endowed with an
À
direct sum of copies of A indexed by G; that is, A[ G ] = αPG A. The elements are
associative binary
sequences p = ( pα )αPG with finite support, and addition is defined component-wise. operation that has a
Introduce a product on A[ G ] by the formula neutral element. It is
like a group but with
ÿ elements lacking an
( p ¨ q)α = p β ¨ qγ . inverse. The set N0 of
β,γPG, β+γ=α non-negative integers
and its Cartesian
Let x α denote the sequence all whose components are zero apart from the one in the products N0r are
arch-examples of
slot with index α which equals one. monoids.
a)Show that the x α form an additive basis for A[ G ].
b)Show that x α ¨ x β = x α+ β .
Show that ( α pα x α ) x β = α pα x α+ β . Verify that A[ G ] is a ring.
ř ř
c)
d)Show that A[N0r ] » A[ x1 , . . . , xr ].
(1.20) The formal derivative. Let f ( x ) = 0ďiďd ai xi be a polynomial with coefficients in
ř

a field k. Copying the classical formula for the derivative of a polynomial, one defines
the formal derivative f 1 ( x ) of f as f 1 ( x ) = 1ďiďd i ¨ ai xi´1 .
ř

a) Show that derivation is a linear operation and that both the Leibnitz’ rule
and the chain rule hold true; that is ( f ( x ) g( x ))1 = f ( x ) g1 ( x ) + f 1 ( x ) g( x ) and

1
f ( g( x )) = f 1 ( g( x )) g1 ( x ).
b) Show that if k is of charactristic zero, then f 1 = 0 if and only if f is constant; i. e.
f if and only if is of degree zero.
c) Show that if f 1 = 0 and f is not of degree zero, then k has positive characteristic,
r
say p, and f ( x ) = g( x p ) for some r where g is a polynomial with g1 ‰ 0.
(1.21) Let tAi uiP I be a collection of subrings of the ring A. Prove that the intersection
Ş
iP I Ai is a subring.
(1.22) Give examples of two subrings A1 and A2 of a ring A such that their union is
not a subring. Assume that the collection tAi uiP I of subrings of A has the property that
Ť
any two rings from it are contained in a third. Prove that in that case the union iP I Ai
is a subring.
M

1.3 Direct products and idempotents

As an introductory motivation to this section, consider the disjoint union X Y Y of two
topological spaces X and Y. Giving a continuous function on X Y Y is the same as
giving one continuous function on X and one on Y. Therefore the ring of continuous

10th January 2021 at 11:32am

Version 4.0 run 30
 26 rings

functions CR ( X Y Y ) decomposes as the direct product CR ( X Y Y ) = CR ( X ) ˆ CR (Y ),

in which both addition and multiplication are given component-wise. This indicates
that in the interplay between geometry and rings, direct product of rings correspond to
disconnected spaces.
Below we shall define the direct product of a collection of rings regardless of its
cardinality and introduce the notion of idempotent elements (elements e such that e2 = e).
Multiplication by idempotents are projection operators (they are 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 to
X of any function f on Z equals f ¨ eX , or put more precisely, f ¨ eX is the restriction
f |X extended by zero to the entire space Z. Anyhow, in this way the set CR ( Z ) ¨ eX is a
ring naturally identified with CR ( X ) with the idempotent eX corresponding to the unit
element in CR ( X ). The lesson learned is that idempotents are algebraic counterparts to
the geometric notion of connected components (at least when the components are finite
in number).

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 ( a1 , a2 ) becomes a ring when equipped with the componentwise
operations. The underlying additive group is the direct product of the underlying
groups of the two rings, and the product is given as

( a1 , a2 ) ¨ ( a11 , a12 ) = ( a1 ¨ a11 , a2 ¨ a12 ).

The unit element is the pair (1, 1), and the two projections πi : A Ñ Ai are ring
˚ Since the unit
homomorphisms. Moreover, the direct product possesses two special elements e1 =
element (1, 1) does not (1, 0) and e2 = (0, 1), which satisfy ei2 = ei and e1 e2 = 0. The sets e1 A and e2 A equal
lie in either Ai , they
are properly speaking respectively A1 ˆ t0u or t0u ˆ A2 , and are, with a liberal interpretation*, subrings of A
not subrings even isomorphic to respectively A1 and A2 .
though they are closed
under both addition (1.23) To generalize what we just did for a pair of rings, let tAi uiP I be any collection
and multiplication. of rings, which can be of any cardinality. In our context it will mostly be finite, but
ś
occasionally will be countable. The direct product iP I Ai has as underlying additive
˚ The reference to the group the direct product of the underlying additive groups of the Ai ’s. The elements
index set I will are “tuples” or strings ( ai )iP I indexed* by I whose i-th component ai belongs to Ai ,
frequently be dropped
and strings written as and the addition of two such is performed component-wise. The same is true of the
( a i ). multiplication, also performed component for component; that is, it holds true that

10th January 2021 at 11:32am

Version 4.0 run 30
 direct products and idempotents 27

( ai ) ¨ (bi ) = ( ai ¨ bi ). The ring axioms can be checked component-wise and thus come
for free.
Ť
Interpreting tuples a = ( ai ) 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
“constant*” function that sends each index i to 1. ˚ Why the quotation
marks?
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 collection φi : B Ñ Ai of ring φ
/
ś
ś B i Ai
homomorphisms, there is an unambiguously defined map of rings φ : B Ñ iP I Ai such
that φi = πi ˝ φ for all i P I. Indeed, this amounts to the map given by φ( x ) = (φi ( x ))iP I πi
$ 
φi
being a ring homomorphism. Ai
Idempotents
(1.25) In any ring A an element e satisfying e2 = e is said to be idempotent, and if f is Idempotent elements
another idempotent, one says that f and g are orthogonal when f g = 0. The element (idempotente
elementer)
1 ´ e is always idempotent when e is and is orthogonal to e as shown by the little
calculations Orthogonal
idempotents
(1 ´ e)2 = 1 ´ 2e + e2 = 1 ´ 2e + e = 1 ´ e, (ortogonale
idempotenter)
e(1 ´ e) = e ´ e2 = e ´ e = 0.

The subset Ae = t ae | a P A u is a ring with e as a unit element. Indeed

ae ¨ be = abe2 = abe,

so Ae is closed under multiplication and trivially it is closed under addition as well;

finally
e ¨ ae = ae2 = ae,
so that e serves as the unit element.
It is common usage to count the unit element and zero among the idempotents; they
are called the the trivial idempotents. The trivial idempotents
(1.26) We saw above that in the direct product A1 ˆ A2 there appears two natural (de trivielle
idempotentene)
defined idempotents. Conversely, let A a ring. To any family tei u1ďiďr of mutually
orthogonal idempotents that add up to 1, there corresponds a decomposition of A as a
direct product:

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

ř
that i ei = 1. Then each set Aei is a subring in the restricted sense, and the association
x ÞÑ ( xe1 , . . . , xer ) gives an isomorphism of rings
» /
ś
A i Aei .

The projection onto Aei is realized as multiplication by ei .

10th January 2021 at 11:32am

Version 4.0 run 30
28 rings

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. Clearly φ is additive, moreover,
the ei ’s being idempotents, we find

φ( x )φ(y) = ( xei )i ¨ (yei )i = ( xyei ei )i = ( xyei )i = φ( xy),

and thus φ also respects the multiplication. The unit element 1 maps to the string (ei )i
which is the unit element in the product since each ei serves as unit element in Aei .
ř
Now, we have supposed that the ei ’s add up to one; that is, 1 = i ei . Hence
ř
x = i xei , from which ensues that φ injective; indeed, that φ( x ) = ( xei )i = 0, means
that each xei = 0.
Finally, let us check that φ is surjective. Given an element ( xi ei )i in the product, we set
ř ř
x = i xi ei . Using that the ei ’s are mutually orthogonal, we find xe j = i xi ei e j = x j e j ,
and x maps to the given element ( xi ei )i . o

Exercises
ˇ (1.23) Determine the idempotents in Z/12Z and in Z/36Z.
(1.24) What is the prime ring of Q ˆ F p ?
M

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 2

Ideals

Ideals were first defined by Richard Dedekind in 1876, but the name comes from the so
called “ideal numbers” of Ernst Eduard Kummer which he introduced in a series of
papers around 1847.
Working with rings of integers in algebraic number fields, the algebraists of the
period realized that analogues of the Fundamental Theorem of Arithmetic do not always
hold in such rings. Recall that the Fundamental Theorem asserts that any integer is a
product n = p1 ¨ . . . ¨ pr of signed primes, and that the factors are unique up to order and
sign—changing the order of the factors does not affect the product, and changing the
sign of one factor can be compensated by 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, as a product of elements which may not be factored further (they can Richard Dedekind
of course always be altered by a unit, but that is not considered an honest factorization). (1831–1916)

The point is however, that these factors are not always unique (apart from the innocuous German mathematician

ambiguities caused by unit factors and change of order).

The classical example, which is omnipresent in text books, is the factorization
? ? ?
2 ¨ 3 = (1 + i 5)(1 ´ i 5) in the ring Z[i 5]. The four involved numbers are all
irreducible, and no two of them are related by units.
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 for ideals
replacing the Fundamental Theorem of Arithmetic. Hence the name ideals, they were
“the ideal numbers” Ernst Eduard Kummer
(1810–1893)
Dedekind rings are however a very restricted class of rings, and today ideals play German mathematician
an infinitely wider role than just being “ideal numbers”. In algebraic geometry for
instance, they appear as the sets of polynomials in k[ x1 , . . . , xr ] vanishing along a subset
of kr , and this is the clue to the coupling between algebra and geometry.
 30 ideals

2.1 Ideals
Ideals (idealer) (2.1) Let A be a ring. An additive subgroup a of A is called an ideal if it is closed 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 requirements
and are ideals, although special ideals. In many texts the ring itself when considered an
ideal, is denoted by (1).
Proper ideals (ekte (2.2) An ideal a is said to be a proper ideal if it is not equal to the entire ring. This is
idealer) equivalent to no member of a being invertible. Indeed, if a P a is invertible, 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 ideals:

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

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

then 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 (n) or (n)Z.
(2.2) For any subset S Ď Cr the polynomials in C[ x1 , . . . , xr ] vanishing on S form an
ideal.
K

Operations on ideals—the lattice of ideals

(2.4) The set I( A) of ideals in the ring A has—in addition to being partially ordered
Ş
under inclusion—a lot of structure. One may form the intersection iP I ai of any family
tai uiP I of ideals. It is easily seen to be an ideal, and it 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 a j1 + . . . + a jr | a ji P ai , r P N u,
iP I

10th January 2021 at 11:32am

Version 4.0 run 30
 ideals 31

and it is the smallest ideal containing all the ai ’s. So I( A) is what one technically calls
a complete lattice; every subset of I( A) has a greatest lower bound (the intersection) and
a smallest upper bound (the sum). It is the lattice of ideals in A. The lattice of ideals
(2.5) A construct similar to the sum of a family of ideals is the ideal generated by a set (Ideallattiset(???))

of elements tai uiP I from A. It will be denoted ( ai |i P I ), or in case the set is finite, say
equal to ta1 , . . . , ar u, the alternative notations ( a1 , . . . , ar ) or ( a1 , . . . , ar ) A are common
usage. Its members are the finite linear combinations of the ai ’s with coefficients from
the ring A; that is, it holds that
ř Generators
( ai |i P I ) = t iP J ci ai | ci P A, J Ď I finite u.
(generatorer)

The elements ai are called generators. Ideals which are generated by finitely many
elements are naturally called finitely generated. Finitely generated
(2.6) An ideal generated by a single element is called a principal ideal and is denoted by ideals (endeliggenererte
idealer)
( a) or by aA. It consists of all multiples of the generator; i. e. ( a) = t c ¨ a | c P A u.
In some rings all ideals are principal as is the case for the integers Z and the Principal ideals
polynomial ring k[t] over a field. These rings, if also being domains, are called Principal (hovedidealer)
Principal Ideal
Ideal Domains, frequently referred to by the acronym pid. Domains
(hovedidealområder)
Different elements can of course generate the same principal ideal, but they will, at
least in domains, be closely related, as described in the next lemma.

Lemma 2.7 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 ( a) Ď (b) there is a ring element u so that a = ub, and when (b) Ď ( a) it
holds that b = va, Hence a = vua. And a not being a zero-divisor, we conclude that
vu = 1. o
One often says that a and b are associates if one is a unit times the other; i. e. if a = ub Associates (assosierte
with u P A˚ . The lemma then says that two non-zero divisors a and b generate the same elementer)

principal ideal if and only if they are associates. So in a domain the set of principal
ideals is naturally identified with A modulo association (which easily is seen to be an
equivalence relation ).
In the case when the ring A possesses zero-divisors, things are more complicated,
see Example 2.10 on page 38 below.
(2.8) The product of two ideals a and b is the ideal generated by all products of one The product of ideals
element from a and one from b; that is, the product ab is formed of all finite sums of (produktet av idealer)

such products:
ab = t a1 b1 + . . . + ar br | ai P a, bi P b, r P N u.
(2.9) The last operation we offer is the formation of the transporter between two ideals. The transporter
Some texts call it the quotient of the two deals—however, that term should be reserved (transportøren)

10th January 2021 at 11:32am

Version 4.0 run 30
 32 ideals

for another construction we shortly come to. So let a and b be two ideals in A. We
define the transporter (a : b) to be set of elements which on multiplication send b into a;
that is
(a : b) = t x P A | xb Ď a u.
It is easily seen to be an ideal. In the particular case that a = (0) and b is a principal
ideal, say b = ( a), the transporter (0 : a) (an immediate simplification of the notational
The annihilator of an overloaded expression ((0) : ( a))) coincides with the annihilator of a; that is
element (annihilatoren
til et element) (0 : a) = Ann a = t x P A | xa = 0 u.

Similarily, any ideal a has an annihilator ideal which is defined as Ann a = (0 : a) =

t x P A | xa = 0 for all a P a u.
Examples
(2.3) In Z it holds that 100 : 10 = (10). More generally if a and b are elements from

the ring A and b is not a zero divisor, one has ( ab : b) = ( a). 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 ( ab, b) = ( a) + Ann b.
(2.4) In Z/40Z one has Ann 2 = (20), that Ann 4 = (10) and that Ann 20 = (5).
(2.5) In the polynomial ring C[ x, y] it holds that ( xy, y2 ) : ( x, y) = (y). Clearly (y) is

contained in ( xy, y2 ) : ( x, y) . For the converse inclusion assume that f x = gxy + hy2

where f , g and h are polynomials in C[ x, y]. 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 C[ x, y];
that is, f P (y).
K
Functorially
A map of rings φ : A Ñ B induces two maps between the ideal lattices I( A) and I( B),
one in a covariant and one in a contravariant way. One can move ideals forward with
Extensions of ideals the help of φ, and the the usual inverse image construct gives a way to move ideals
(utvidlese av ideal) backwards along φ. The new ideals are in some texts respectively called extensions or
contractions of the old.
Contractions of ideals (2.10) We begin with the contravariant way. The inverse image φ´1 (b) of an ideal b in B,
(tilbaketrekning) also called the pullback, is evidently an ideal in A—indeed, φ( ab) = φ( a)φ(b) belongs
Pullback to b whenever φ(b) does—and this gives rise to a map φ´1 : I( B) Ñ I( A). In the
(tilbaketrekning) frequently reccurring case when A is a subring of B, the reference to the inclusion map
is most often suppressed, and one uses the natural notation b X A for the “pullback” of
the ideal b.
Obviously the inverse image preserves inclusions, and it takes intersections to
intersections ( pullbacks of sets respect intersections in general). Sums and products

10th January 2021 at 11:32am

Version 4.0 run 30
 ideals 33

of ideals however, are not generally preserved, but the inclusions in the upcoming
proposition are easily verified. One has:

Proposition 2.11 (Pullbacks) Let φ : A Ñ B be a ring homomorphism and let a and b be

two ideals in B. The pullback map φ´1 : I( B) Ñ I( A) preserves inclusions, and the following
claims hold true:
i) φ´1 (a) X φ´1 (b) = φ´1 (a X b);
ii) φ´1 (a) + φ´1 (b) Ď φ´1 (a + b);
iii) φ´1 (a) ¨ φ´1 (b) Ď φ´1 (a ¨ b).

Equality does not hold in general in the two last statements, but notice that both
inclusions will be equalities when φ is a surjective map. We postpone giving examples
to the end of the paragraph (Examples 2.6 and 2.7 below).
(2.12) Next we come to the covariant construction. If a is an ideal in A, the image φ(a)
is not necessarily an ideal in B unless φ is surjective. A stupid example can be the
image of any non-zero ideal in Z under the inclusion Z Ď Q. The ideal generated by
Extension of an ideal
φ(a) however is, and we shall usually denote this ideal by φ(a) B or simply by aB; as (utvidelse av et ideal)
mentioned above it is called the extension of a, but is also frequently referred to as the
pushout of a. This induces a map I( A) Ñ I( B). Inclusions are obviously preserved, Pushout of an ideal
and one leisurely verifies the other relations in the following proposition. (pushout, frempuff,
fremskudd)
Proposition 2.13 (Pushouts) Let φ : A Ñ B be a map of rings and let a and b be two ideals
in A. Then the map a ÞÑ aB preserves inclusions. Moreover, the following hold true:
i) φ(a ¨ b) B = φ(a) B ¨ φ(b) B;
ii) φ(a + b) B = φ(a) B + φ(b) B;
iii) φ(a X b) B Ď φ(a) B X φ(b) B.

The inclusion in the last statement may be strict (see Example 2.9 on page 38), but just
like with the previous proposition, equality holds in the third statement whenever φ is
surjective.

Examples
(2.6) A simple example of strict inclusion in statement ii) in Proposition 2.11 above is the
diagonal map δ : A Ñ A ˆ A that sends a to ( a, a). The two ideals b = t (0, a) | a P A u
and b1 = t ( a, 0) | 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.
(2.7) We intend to give an example of strict inclusion in iii) in Proposition 2.11. Consider
the subring k [ xy] of the polynomial ring k[ x, y] and let φ be the inclusion. Let a and
b be the two principal ideals ( x ) and (y) in k [ x, y]. We contend that a X k[ xy] = b X
k [ xy] = ( xy); indeed, clearly ( xy) Ď ( x ) X k [ xy], and equality holds since an identity like

10th January 2021 at 11:32am

Version 4.0 run 30
34 ideals

x f ( x, y) = g( xy) between polynomials forces g to be without a constant term. Similarily,

(y) X k[ xy] = ( xy). So (a X k[ xy])(b X k[ xy]) = (( xy)2 ), but ( x ) X (y) = ( xy)k[ x, y],
which intersects k[ xy] in the ideal ( xy).
K

Exercises
ˇ (2.1) Let a, b and c be ideals in a ring A.
a) Show that the two relations a ¨ b Ď a X b and (a X b)2 Ď a ¨ b hold. Show by giving
examples that there might be a strict inclusion in both cases.
b) Assume that a + b = (1). Show that a ¨ b = a X b.
c) Show that a(b + c) = ab + ac. Show that a X b + a X c Ď a X (b + c), and by
exhibiting an example, show that the inclusion can be strict.
(2.2) Let tai u be a collection of ideals in the ring A. Show that for any ideal b it holds
Ş Ş ř Ş
true that ( iP I ai : b) = iP I (ai : b) and that (b : iP I ai ) = iP I (b : ai ).
(2.3) Show that any non-zero ideal in the ring Z of integers is principal, generated by
any of the two members of smallest absolute value. Show that each non-zero ideal in
the polynomial ring k[ x ] over a field k is principal, generated by any member of smallest
degree.
(2.4) Given two ideals (n) and (m) in the ring of integers Z.
a) Show that (n) Ď (m) if and only if m|n. Conclude that the partially ordered set
I(Z)zt(0)u of non-zero ideals in Z is lattice isomorphic to the the set of natural
numbers ordered by reverse divisibility;


b) Describe the ideals (n, m) and (n) X (m). Show that (n) ¨ (m) = (n) X (m) ¨
(n, m);
c) Describe ((144) : (24));
d) Describe the transporter (n : m) in terms of the prime factorizations of the two
integers n and m.
(2.5) Let k [ x, y] be the polynomial ring in the variables x and y over the field k, and let
m be the ideal generated by x and y; that is m = ( x, y). 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 ( x µ , yν ) for n sufficiently
large. What is the smallest n for which this holds?
? ?
ˇ (2.6) Let A = Z[ 2, 3]. Show that as an abelian group A is free of rank four and
?
exhibit a basis. Show that the underlying abelian groups of the principal ideals ( 2)
?
and ( 3) both are of rank four. Exhibit additive bases for both.
M

10th January 2021 at 11:32am

Version 4.0 run 30
 kernels and quotients 35

2.2 Kernels and quotients

In one way ideals play the same role in the category of rings as normal subgroups 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.
(2.14) By definition the kernel of a ring homomorphism φ : A Ñ B is the kernel of φ Kernels of ring
considered a map between the underlying additive groups; that is, it equals the subset homomorphisms
(kjernen til en
of elements mapping to zero, or written in symbols ker φ = t a P A | φ( a) = 0 u. If ringavbildning)
a P ker φ and b P A, we find

φ( a ¨ b) = φ( a) ¨ φ(b) = 0 ¨ φ(b) = 0,

and we can conclude that ab P ker φ. Hence the kernel ker φ is an ideal.
(2.15) To see that any ideal is a kernel, one introduces the concept of quotient rings. An Quotient rings
ideal a in A being an additive subgroup, there is a quotient group A/a which consists (kvotientringer)

of the residue classes [ a] = a + a of elements in A. The sum of two such, say [ a] and [b],
equals [ a + b]. To put a ring structure on A/a we simply define the product of two
classes [ a] and [b] as
[ a] ¨ [b] = [ a ¨ b] = a ¨ b + a.
Some checking is needed; the most urgent one being that the product only depends on
the residue classes [ a] and [b] and not on the choice of representatives a and b. This is
encapsulated in the formula where x and y are arbitrary elements from a

( a + x ) ¨ (b + y) + a = a ¨ b + a ¨ y + b ¨ x + x ¨ y + a = a ¨ b + a.

It is left to the students to verify that this product complies with the associative,
commutative and distributive laws. Finally, by definition of the ring operations in A/a,
the quotient map π : A Ñ A/a that sends a to the residue class [ a], is a map of rings
whose kernel equals the given ideal a.
˚ This examplifies what
Example 2.8 It is appropriate to mention what quotients by the two “extreme” ideals purpose the null-ring
are. The quotient A/a equals A if and only if a is the zero-ideal, and it equals* the serves; it allows a
general existence
null-ring if and only if a = A. K
theorem (avoiding the
(2.16) The quotient ring A/a together with the quotient map π : A Ñ A/a enjoys a hypothesis a ‰ A).
so-called universal property—the rather pretentious notion “solves a universal problem”
is also common usage—which is a convenient way of characterizing many types of
A universal property
mathematical objects. The origin of the technique is found in category theory where (en universell
objects not always have “elements” and one must rely on “arrows” to express properties. egenskap)
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 φ(a) = 0, the map φ is

10th January 2021 at 11:32am

Version 4.0 run 30
 36 ideals

constant on every residue class [ a] = a + a, and we put ψ([ a]) 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 2.17 (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.

The statement may be illustrated by the first commutative diagram in the margin. The
A
π / A/a solid arrows are the given ones, and the dashed arrow is the one claimed to exist. If
it happens that ker φ = a, the induced map ψ will be injective, and hence, if a priori
ψ
!  surjective, an isomorphism. The images of all ring-maps with the same kernel are
φ
B therefore isomorphic, in the strong sense that the isomorphisms fit into diagrams like
A the second one in the margin.
φ1 φ

 Ideals in quotients
im φ1
θ / im φ (2.18) 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 π (b) 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 (c) is an ideal in A satisfying
π (π ´1 (c)) = c (again because π is surjective); or in other words, π ´1 (c) contains a and
π ´1 (c)/a = c.

Proposition 2.19 (Ideals in Quotients) Let a be an ideal in the ring A and let π : A Ñ
A/a the quotient map. The following three statements hold true:
i) For every ideal b in A it holds true that π ´1 (π (b)) = b + a. Each ideal c in A/a is
of the form π (b) = b/a; indeed, c = π (π ´1 (c));
ii) The lattice of ideals in A/a and the lattice of ideals A containing a are isomorphic
lattices, with c ÞÑ π (c) and c ÞÑ π ´1 (c) as mutually inverse maps;
iii) 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 π (b) = b/a is an ideal and that c = π (π ´1 (c)), so the last
part of i) is clear. For the first claim, if π ( x ) = π (b) for some element b P b it evidently
holds true that x ´ b P ker π = a and hence x P b + a so that π ´1 (π (b)) Ď b + a. The
reverse inclusion follows immediately since π (b + a) = π (b), again because ker π = a.
To show ii) observe that the equality π ´1 (π (b)) = b + a entails that π ´1 takes
values in the sublattice of I( A) whose members contain a. It equally implies that
π ´1 (π (b)) = b whenever a Ď b, and we conclude that the two maps are mutual inverses.
They both respect inclusions and are thus lattice isomorphisms.
The third claim iii) is just a rephrasing of a being the kernel of π. o

10th January 2021 at 11:32am

Version 4.0 run 30
 kernels and quotients 37

Note that both maps π and π ´1 respect intersections, sums and products of ideals. The
picture below is an illustration of the situation:

Ideals b in A

Ideals with a Ď b Ideals in A/a

π

(2.20) The image in A/a of an ideal b Ď A, which not necessarily contains a, is the ideal
(b + a)/a. This holds since obviously π (b + a) = π (b). Now, ker π|b = a X b from
which ensues the following isomorphism

b/b X a » (a + b)/a. (2.1)

The two members of (2.1) are ideals in different rings, so we must be cautious about
what isomorphic means (it does not mean equal even though it might seem so the
map sending a class [b] to the class [b], but those classes are mod different ideals). The
isomorphism is certainly an isomorphism of abelian groups, but it preserves a lot more
structure. The two sides are what we later shall call A-modules: Elements from A
operate by multiplication on both sides (this evidently holds for ideals in any quotient
ring of A), and the isomorphism respects these operations.
Finally, we mention that when a and b are two ideals with a Ď b, there is a natural
isomorphism
( A/a)/(b/a) » A/b. (2.2)
A
πa
/ A/a
Indeed, in the diagramme in the margin, the composition πb/a ˝ πa has the ideal b as
πb πb/a
kernel, and therefore factors through πb by say θ. The map θ is surjective since the  
composition is and injective since the composition has b = πa´1 (b/a) as kernel. The A/b / ( A/a)/(b/a)
θ
two formulas (2.1) and (2.2) are often referred to as the Isomorphism Theorems.

Theorem 2.21 (The Isomorphism Theorem) Let a and b be ideals in A. Then the following
two isomorphism relations hold, where in the second is assumed that a Ď b:
i) b/b X a » (a + b)/a;
ii) ( A/a)/(b/a) » A/b.

As remarked above, the isomorphisms are module isomorphisms; they are isomor-
phisms of abelian groups which respect multiplication by elements from A.

A convenient convention
(2.22) Ever recurring ingredients of a set-up in commutative algebra are rings shaped
like quotients k[ X1 , . . . , Xr ]/a of a polynomial ring. When working with such rings,

10th January 2021 at 11:32am

Version 4.0 run 30
 38 ideals

it is very natural and suggestive to denote the class of a variable by the lower case
variant of the uppercase letter used for the variables. To avoid repeating this formula ad
infinum like a yogi’s mantra, we adhere to the following convention: we say that the
Constituting relations ring A = k[ x1 , . . . , xr ] has constituting relations f 1 ( x1 , . . . , xr ) = . . . = f s ( x1 , . . . , xr ) = 0,
(konstituerende if A = k[ X1 , . . . , Xr ]/a with a = ( f 1 , . . . , f s ) and the upper case Xi ’s correspond to the
relasjoner)
lower case ones.
A convenient way of defining ring maps with source A and target a k-algebra B, is
to assign values bi P B to the generators xi . This can of course not be done freely, but
when the constituting relations persist holding in B when the bi ’s are substituted for the
xi ’s, there is a well-defined and unique ring map A Ñ B such that xi ÞÑ bi . This ensues
from the Factorization Theorem (Theorem 2.17 on page 36); indeed, sending Xi to bi
defines a map k[ X1 , . . . , Xr ] Ñ B which factors via A since it vanishes on the ideal a.

Examples

(2.9) This example aim at illustrating that strict inclusion in the last inequality of
Proposition 2.13 on page 33 may occur. So let A = k[ X, Y, Z ] and let B = k[ x, y, z] with
constituting relation zx = zy, and consider the natural map φ : A Ñ B that sends upper
case letters to their lower case versions. Let a and b be the principal ideals ( X ) and (Y )
in A. We conted that φ(a X b) Ă φ(a) X φ(b).
It holds that a X b = ( XY ), so (a X b) ¨ B = ( xy). Since zx = zy, we see that and
zx P ( x ) X (y) = aB X bB, but zx R ( xy) B; indeed, one way of seeing this is to observe
that sending z to 1 and both x and y to t gives a well-defined ring map k[ x, y, z] Ñ k[t]
(since the relation zx = zy persists as t = t), which maps zx to t and xy to t2 . We deduce
that φ(a X b) Ř φ(a) X φ(b).

(2.10) Let A = k [ x, y, z] with constituting relation z = zxy. In A the two principal ideals
(z) and ( xz) coincide, but there is no unit u ‰ 1 in A so that uz = xz; hence z and xz
are not associates even though (z) = ( xz).
The salient point is that A˚ = k˚ . One way of seeing this, is to observe that killing
z gives a well-defined ring map A Ñ k[ x, y]. It takes units to units, and the group of
units in the polynomial ring equals k˚ . So u would be a scalar. Setting z equal to 1 and
x = y´1 gives a ring map A Ñ k [y, y´1 ], and in the latter ring obviously x is not a scalar
being equal to y´1 .

(2.11) Let A = k[ xi |i P N] with constituting relations xi2 = ´1 for i P N. Then the

polynomial t2 + 1 has infinitely many roots in A. The ring A is not an integral domain
in that xi2 = x2j fot all i and j; so that ( xi + x j )( xi ´ x j ) = 0.
K

10th January 2021 at 11:32am

Version 4.0 run 30
 prime ideals and maximal ideals 39

Exercises
(2.7) Let k be a field and φ : k Ñ A a ring homomorphism. Show that φ is injective
unless A is the null ring.
(2.8) Let A be a ring and let a and b be two ideals in A. Show that there is a natural
equality b( A/a) = (a + b)/a. Use the Isomorphism Theorem to show that there is a
canonical isomorphism A/a + b » ( A/a)/b( A/a), so that you can divide out by a sum
of two ideals by successively dividing out by one at a time. Of course, the order doesn’t
matter: swapping the two ideals yields an isomorphism A/a + b » ( A/b)/a( A/b).
M

2.3 Prime ideals and maximal ideals

Two classes of ideals are infinitely more important than others. We are speaking
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
played the role of the primes in Kummer and Dedekind’s world of “ideal numbers”.
If your ambitions are high and you try to understand all ideals in a ring, you have to
begin with understanding the prime ideals, and then accomplish the draconian task to
explain how other ideals are built from the prime ideals.
(2.23) An ideal a in a ring A is a prime ideal if it is proper and satisfies the following Prime ideals
requirement: (primidealer)

o If ab P a, then either a P a or b P a.

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( A)ztAu. So an ideal a is maximal if it is proper and satisfies the following Maximal ideals
requirement: (maksimale idealer)

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

Notice that both prime ideals and maximal ideals are proper by definition.
(2.24) One has the following characterization of the two classes of ideals in terms of
properties of quotients.

Proposition 2.25 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: The quotient A/a is an integral domain if and only if [ a][b] = 0 implies that
either [ a] = 0 or [b] = 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.

10th January 2021 at 11:32am

Version 4.0 run 30
 40 ideals

Bearing in mind the relation between ideals in A/a and ideals in A containing a (as
in Proposition 2.19 on page 36), 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 30). o
Notice that the zero ideal (0) 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
Residue class fields called the residue class field of A at m and now and then denoted by k(m).
(restklassekropper) Since fields are integral domains, we see immediately that maximal ideals are prime.
The converse does not hold as we shortly shall see examples of (Example 2.14 below).

Proposition 2.26 A maximal ideal m is prime.

(2.27) 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:

Proposition 2.28 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: If neither a nor b lies in p, one may find 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
The claim is not restricted to products of only two ideals. With an easy induction one
proves that if a finite product a1 ¨ . . . ¨ ar of ideals is contained in a prime ideal p, one of
the factors ai lies in p.

Prime ideals in quotients

(2.29) In the correspondence between ideals in A and A/a described in Proposition 2.19
on page 36 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 2.19 the inverse image map π ´1 is an isomorphisms of the lattice I( A/a)
with the sublattice of I( A) whose members contain a, and hence maximal elements
correspond to maximal elements. The first ensues from the general truth that prime
ideals pull back to prime ideals along ring maps ; indeed, assume that p is a prime
ideal in B and that φ : A Ñ B a ring map. That the product ab lies in the inverse image
φ´1 (p), means that φ( ab) P p, but φ( ab) = φ( a)φ(b), and hence either φ( a) lies in p or
φ(b) lies there; that is, either a P φ´1 (p) or b P φ´1 (p). We have thus established:

Proposition 2.30 (Prime and maximal ideals in quotients) Assume A is 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.

10th January 2021 at 11:32am

Version 4.0 run 30
 prime ideals and maximal ideals 41

Examples
(2.12) The archetype of maximal ideals are the kernels of evaluation maps. For in-
stance, let a = ( a1 , . . . , ar ) be a point in kr where k is any field, and consider the
map k [ x1 , . . . , xr ] Ñ k sending a polynomial f to its value f ( a) at a. The kernel m is
a maximal ideal since k[ x1 , . . . , xr ]/m is the field k. The kernel may be described as
m = ( x1 ´ a1 , . . . , xr ´ ar ). This is obvious when a is the origin, and introducing fresh
coordinates xi1 = xi ´ ai , one reduces the general case to that case.
(2.13) The zero ideal in A is prime if and only if A is a domain, and it is maximal if and
only if A is a field.
(2.14) There are plenty of prime ideals that are not maximal. Continuing the pre-
vious example the ideal p generated by a proper subset of the variables is prime
but not maximal; that is, after eventually renumbering the variables, p = ( x1 , . . . , xs )
with s ă r is prime but not maximal. This is best seen by considering the partial
evaluation map k[ x1 , . . . , xr ] Ñ k[ xs+1 , . . . , xr ] that sends a polynomial f ( x1 , . . . , xr ) to
f (0, .., 0, xs+1 , . . . , xr ), whose kernel is p. Since the polynomial ring k[ xs+1 , . . . , xr ] is a
domain, it ensues that p is prime, and p is obviously not maximal as s ă r. By a linear
change of variable one also shows that the ideals ( x1 ´ a1 , . . . , xs ´ as ) are all prime.
(2.15) Consider the ring of Gaussian integers Z[i ]. It is isomorphic to the quotient
Z[t]/(t2 + 1) of the polynomial ring Z[t], the isomorphism sends t to i. Let p P Z
be a prime number and consider the ideal pZ[i ]. Citing Exercise 2.8 on page 39 we
infer that Z[t]/( p, t2 + 1) on the one hand will be isomorphic to Z[i ]/pZ[i ] and on
the other to the quotient F p [t]/(t2 + 1), and we conclude that there is an isomorphism
Z[i ]/pZ[i ] » F p [t]/(t2 + 1) that swaps t and i.
K

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.
(2.31) 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,
assume not and pick elements xi P Ai X B but xi R A j for ti, ju = t1, 2u. Then x1 + x2 P B
but x1 + x2 R A1 Y A2 , for were it in A j , it would follow that xi P A j , again with
ti, ju = t1, 2u, and this is not the case.

10th January 2021 at 11:32am

Version 4.0 run 30
 42 ideals

The corresponding statement for three subgroups is faulty as shows the vector space
F2 ‘ F2 with four elements. It is the union of the three one-dimensional subspaces it has;
so for a general claim involving more than two ideals to be true, some multiplicative
structure is required, but the case of two groups is reflected in the assertion in that two
of the ideals need not be prime:

Lemma 2.32 (Prime avoidance lemma) Let a1 , . . . , ar be ideals in the ring A of which all
Ť
but at most two are prime. If a is an ideal contained in the union i ai , then a is contained in at
least one of the a j ’s.

Proof: We shall assume that a is not contained in any of the ai ’s and search for an
Ť
element in a not lying in the union i ai ; that is, not belonging to any of the ai ’s. The
proof proceedes by induction on r, the case r = 2 already being settled. So we assume
r ą 2 and that the lemma holds true for r ´ 1. Hence a is not contained in the union
Ť Ť
j‰i a j for any i. We can therefore for each i pick an element xi in a not in j‰i a j , and
we may safely assume that xi P ai (were it not, we would be through). Since r ě 2 at
least one of the ai ’s is a prime ideal, and we may as well assume that it is the case for ar .
With these assumptions, we contend that the element

x = x 1 ¨ . . . ¨ x r ´1 + x r ,

which clearly lies in a, does not belong to any ai . If i ď r ´ 1 this holds because xi lies
in pi , but xr does not. For i = r we know that xr lies in ar , but x1 ¨ . . . ¨ xr´1 does not
since none of the factors lie there and ar is prime, so x R ar as well. o
Notice that the proof merely requires a to be closed under addition and multiplication,
so the ideal a may be replaced with a “weak subring” of A; that is, a subset closed
˚ Lacking a unit under addition and multiplication*. The second remark to make is that if A is an
element it is not algebra over an infinite field, one may even skip the requirement that the ideals be
genuine ring according
to our conventions prime (Exercise 2.12 below).
(2.33) At several later occasions unions and intersections of prime ideals will play an
important role, and we use the occasion to introduce some terminology.
Ť
Redundant (redundant) A union i Si of sets is said to be redundant if one of the sets can be discarded
Ť
without the union changing. This means that for some index ν it holds that Sν Ď i‰r Si .
Irredundant unions If the union is not redundant, naturally one calls it irredundant. For finite unions of
(irredundante unioner)
Ť
prime ideals the Prime Avoidance Lemma entails that the union i pi is irredundant if
and only there is no inclusion relation among the pi ’s. Indeed, if there is such a relation,
Ť
the union is obviously redundant, and if say pν Ď i‰ν pi , the lemma gives that pν is
contain in one of the other pi ’s.
Ş
Similarly, an intersection i Si is irredundant if one cannot discard one of sets with-
out changing the intersection. For a finite intersection of prime ideals Proposition 2.28

10th January 2021 at 11:32am

Version 4.0 run 30
 prime ideals and maximal ideals 43

on page 40 implies that the intersection being irredundant is equivalent to there being
no inclusions among the prime ideals.
(2.34) Irredundant unions and intersections of prime ideals enjoy strong uniqueness
properties; in fact, the prime ideals involved are determined by their intersection or
their union, as expressed in the following twin lemmas.

Lemma 2.35 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 α(ν) so that pν Ď qα(ν) . By symmetry, for each µ there is a β(µ) such
that qµ Ď pβ(µ) . Now
pν Ď qα(ν) Ď pβ(α(ν)) ,
and since there are no non-trivial inclusion relations among the pi ’s, we infer that
β(α(ν)) = ν. In a symmetric manner one shows that α( β(µ)) = µ; so α is a bijection
from t1, . . . , ru to t1, . . . , su with pν = qα(ν) , and we are through. o

Lemma 2.36 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 α(ν), the relation pα(ν) Ď qv holds. By symmetry, for each µ there is a β(µ) such that
qβ(µ) Ď pµ . Now
pα( β(µ)) Ď qβ(µ) Ď pµ ,
and there being no non-trivial inclusion relations among the pi ’s we may conclude that
α( β(µ)) = µ. In a symmetric manner one shows that β(α(v)) = ν and we can conclude
that α is a bijection from t1, . . . , ru to t1, . . . , su with with pα(v) = qv o
Exercises
ˇ (2.9) Let p be a prime ideal in a ring A. Show that pA[t] is prime.
ˇ (2.10) Prove that pullback of prime ideals are prime, but show by examples that
pullbacks of maximal ideals need not be maximal. Show by giving examples that the
extension of a prime ideal is not necessarily prime.
ˇ (2.11) Let a and b be two ideals in a ring A, furthermore let p1 , . . . , pr be prime ideals
in A. Show that if azb is not contained in any of the pi ’s, then a is not contained in the
Ť
union i pi .
ˇ (2.12) Assume that A is an algebra over an infinite field; show that the Prime Avoidance
Lemma persists being true without any of the pi ’s being prime. Hint: Prove a “vector
subspace avoiding lemma” over infinite fields.

10th January 2021 at 11:32am

Version 4.0 run 30
 44 ideals

2.4 Primes and irreducibles

There are two familiar characterizations of prime numbers. One says a number is prime
precisely when the only factors are plus or minus one and the number itself, and the
other asserts that a number is prime precisely when it divides one of the factors when
dividing a product. In general rings these two aspects of prime numbers generalize to
separated notions which are no more equivalent.
(2.37) The first of the twin notions is that of prime elements. The definition is verbatim the
Prime elements same as the second characterizations above: A prime element in a ring A is an element a
(primelementer) which is neither zero nor a unit and is such that if a divides a product, it divides one of
the factors; in other words, a relation like bc = ya for some y implies c = xa or b = xa
for some x; or expressed in symbols, a|bc implies that a|b or a|c. The property is not
restricted to products with two factors, a straightforward induction proves that if a is a
prime element that 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 ( a) being a prime ideal.

Proposition 2.38 A principal ideal ( a) 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. Translated into a
statement about ideals the divisibility relation x|y means that y P ( x ). Hence, bc P ( a) is
equivalent to a|bc, and b P ( a) or c P ( a) to respectively a|b or a|c . o
(2.39) The other aspect of prime numbers is that they can not be further factored; that is,
their sole factors are ˘1 and the prime itself. Irreducible polynomials in k[ x ] share this
quality except that they can be changed by non-vanishing constant factors (of course,
f = c´1 ¨ c f for any non-zero constant c). Generalizing this, one says that a non-zero
Irreducible elements element a from a ring A is irreducible if it is not a unit, and if a relation a = bc implies
(irreduktible elementer) that either b or c is a unit. This can be phrased in terms of a certain maximality condition
for ideals.

Proposition 2.40 An element a in the ring A is irreducible if and only if ( a) is maximal

among the proper principal ideal.

Proof: A relation a = bc is equivalent to an inclusion ( a) Ď (b), and when ( a) enjoys

the maximality property it ensues that either ( a) = (b) and c is a unit, or (b) = A and b
is a unit and a is irreducible. Assume then that a is irreducible and let (b) be a proper
principal ideal containing ( a), which means that a = cb. So c is a unit since b is not, the
ideal (b) being proper, and we deduce that ( a) = (b). o

10th January 2021 at 11:32am

Version 4.0 run 30
 primes and irreducibles 45

Proposition 2.41 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
The converse of this proposition is not generally valid, in fact one is tempted to say
that in most rings it does not hold. There are simple examples of irreducibles not
?
being prime in quadratic extensions of Z. We give one in the ring Z[ ´5] below
(Example 2.18 on page 46). We shall, however, shortly meet classes of rings where it is
true (see Proposition 2.44).
(2.42) Rings in which all ideals are principal, the pid’s, are among the easiest rings to
understand. One of the particular properties they enjoy is that there is no distinction
between maximal ideals and non-zero prime ideals.

Proposition 2.43 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 2.40 above ensues that ( a) then
is maximal among the proper principal ideals, but all ideals being principal, ( a) is
maximal. o
Neither is there any distinction between prime and irreducible elements:

Proposition 2.44 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 2.40 above an ideal

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

Examples

(2.16) In the polynomial ring k[ x ] over a field k, principal ideals ( f ( x )) with f irreducible,
are maximal ideals. The quotient K = k[ x ]/( f ( x )) 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 [ x ] of
the variable x. This illustrates the saying that what matters in modern mathematics is
“what objects do, not what they are”, or as Obi-Wan Kenobi in Star Wars teaches Luke
Skywalker: “Do not equate ability with appearance”.

10th January 2021 at 11:32am

Version 4.0 run 30
46 ideals

(2.17) The quotient R[ x ]/( x2 + 1) is isomorphic to C as one sees by mapping x to i. In a

similar vein, if p is a prime number, the polynomial Φ p ( x ) = x p´1 + x p´2 + . . . + x + 1
is irreducible over Q, so that Q[ x ]/(Φ p ( x )) is a field. Sending x to a primitive pth -root
of unity ξ , gives an isomorphism with Q(ξ ).
(2.18) Simple and concrete examples of irreducible elements that are not prime are easily
found in the ring Z[i 5] where among others the relation
‘

? ‘
2 ¨ 3 = (1 + i 5) ¨ (1 ´ i 5) (2.3)
?
holds. So, for instance, 2 is not a prime element since it neither divides (1 + i 5) nor
?
(1 ´ i 5). This follows since squares of absolute values of members of Z[i 5] are
‘
?
natural numbers so that if relation 2 = x ¨ (1 + i 5) held, we would find
?
4 = |2|2 = | x |2 |(1 + i 5)|2 = | x |2 6

with | x | P N, which absurd. The element 2 is however irreducible. Indeed, If 2 = zw,

one would have 2 = |z||w|, which entails that either |z| = 1 or |w| = 1, and in view of
the units precisely being the elements of norm one (Exercise 1.10 on page 21), either z or
w would be a unit. Of course, the three other numbers appearing in (2.3) are irreducible
as well, and Exercise 2.17 below asks you to check this. For a generic example of
irreducible elements not being prime, see Exercises 2.53 and 2.54 on page 62.
(2.19) The ring Z[i ] of Gaussian integers is a pid: The absolute value works as a so-called
Euclidean function on the ring Z[i ], which means there is division algorithm valid in Z[i ]
similar to the Euclidean algorithm for integers. For any two given Gaussian integers a
and b with b ‰ 0, one may find two others, the quotient q and the remainder r, so that
a = qb + r, and most importantly the remainder satisfies |r | ă |b|.
To establish this, observe that geometrically the Gaussian integers form the integral
lattice in the complex plane; that is, the set of points whose coordinates both are integers.
Given two Gaussian integers a and b with b ‰ 0, the distance from ab´1 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 q P Z[i ] so that | ab´1 ´ q| ď 2/2 ă 1. Putting r = b( ab´1 ´ q),
we have a = qb + r with |r | ă |b|.
Now, any ideal in Z[i ] is generated by a shortest non-zero member a0 . Indeed, if
a P a divide a by a0 to obtain a = qa0 + r with |r | ă | a0 |. But r = a ´ qa0 lies in a and
since a0 is the shortest non-zero member of a, it ensues that r = 0.
K

Exercises
ˇ (2.13) Let p1 , . . . , pr be prime numbers and let A = Z/( p1 ¨ . . . ¨ pr ). Show that the
prime ideals in A are precisely the principal ideals ( pi ). Prove that A/pi A is the field

10th January 2021 at 11:32am

Version 4.0 run 30
 primes and irreducibles 47

F pi with pi elements. How many elements does ( pi ) have? An how many are there
in the principal ideal ( p1 ¨ . . . ¨ ppi ¨ . . . ¨ pr ) (the "hat" indicating that pi is not part of the
product).
(2.14) Find an explicit isomorphism between R[ x ]/( x2 + x + 1) and C.
ˇ (2.15) Primes in the Gaussian integers. The aim of this exercise is to analyse the primes in
the ring of Gaussian integers Z[i ]. Let p be a prime number.
a) Assume that p is odd. Show that there is an exact sequence of multiplicative
groups
φ ψ
1 µ2 F˚p F˚p µ2 1

where µ2 = t˘1u, and the maps φ and ψ are given as φ( x ) = x2 and ψ( x ) =

x ( p´1)/2 ;
b) Conclude that ´1 has a square root in F p is and only if p ” 1 mod 4;
c) Show that x2 + 1 is irreducible over the field F p if and only if p is odd and p ı 1
mod 4;
?
d) Show that the ring F p [ x ]/( x2 + 1) is isomorphic to the field F p ( ´1) if p is
odd and p ı 1 mod 4 and isomorphic to F p ˆ F p if p ” 1 mod 4. Show that if
p = 2, it is isomorphic to F2 [ x ]/( x2 );
e) Consider the ring of Gaussian integer as an extension Z Ď Z[i ]. Discuss the
possible shapes of the quotient Z[i ]/pZ[i ] where p P Z is a prime. When is
pZ[i ] a prime ideal? Euclidean function
(2.16) Into which of the fields F3 , F5 and F7 is there a map of rings from Z[i ]? If there (euklidsk funksjon)

is one, describe the kernel.

?
(2.17) Referring to Example 2.18 show that the three other involved numbers 3, 1 + i 5
?
and 1 ´ i 5 are irreducible.
ˇ (2.18) Euclidian functions and pid’s. Let W be a well-ordered set (for instance N0 ). A
Euclidean function with values in W on a ring A is mapping δ : A Ñ W such that for any
pair a and b of elements from A there are elements q and r in A with a = bq + r and
δ(r ) ă δ(b). Show that a domain A which possesses a Euclidian function, is a pid.
Hint: Minimize δ over non-zero elements in ideals.
?
(2.19) Prove that Z[ ´2] is a principal ideal domain by showing it has a division
algorithm with the absolute value as a Euclidean function. Hint: The convex hull of
‘ ‘ Gotthold Eisenstein
i 2, 1 and 0 is a rectangle whose longest diagonal has length 3. (1823–1852)
ˇ (2.20) The Eisenstein integers. Gotthold Eisenstein is among the young geniuses who German mathematician
died early. He succumbed to tuberculosis in 1852 at the age of 29. The numbers in the
ring Z[η ] with η the cube root of unity η = e2πi/3 = (´1 + i 3)/2 are named after him.
‘

a) Verify that Z[η ] = t n + mη | n, m P Z u is a subring of C;

Hint: η 2 + η + 1 = 0

10th January 2021 at 11:32am

Version 4.0 run 30
 48 ideals

b) Determine the units in Z[η ];

c) Prove that Z[η ] has a division algorithm with the absolute value as a Euclidean
function. Conclude that Z[η ] is a pid with every ideal generated by a shortest
member. Hint: Compute the diameter of the convex set spanned by η, 1 + η, 1
and 0:
d) How many different generators does an ideal have?
M

η 1+ η

0 1

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, sets which for us mostly will
be subsets of the lattices I( A) of ideals in a ring A ordered by inclusion, or later on
subsets of the lattice of submodules of a module. The lemma turns out to be utterly
useful when studying rings, and in the sequel it will be crucial at several occasions.

Zorn’s lemma
Zorn’s lemma is one of quite a few theorems that for some reason keep being called
Max August Zorn (1906–1993)
lemmas. It is usually attributed to Max Zorn, but as often happens, its history can
German mathematician
be traced further back; Felix Hausdorff published versions of it some ten years before
Zorn. Anyhow, "Zorn’s lemma" is a good name (so good that an experimental and
non-narrative film made by Hollis Frampton in 1970 was called "Zorns lemma").
(2.45) We begin with introducing some terminology. 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
Maximal elements
(maksimale elementer) y = x. One should not 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.

10th January 2021 at 11:32am

Version 4.0 run 30
 existence theorems 49

A partially ordered set is said to be linearly ordered or totally ordered if any two of Linearly ordered sets
its elements can be compared. Phrased differently, for any pair x, y of elements either (lineært ordnede
mengder)
x ď y or y ď x should hold. A chain in Σ is a linearly ordered subset of Σ. The chain is
Totally ordered sets
bounded above if for some element x P Σ it holds true that y ď x for all elements y in the (totalt ordnede
chain, and then of course, x is called an upper bound for the chain. Similarly, the chain mengder)
is said to be bounded below when having a lower bound in Σ; that is, an element x P Σ Chains (kjeder)

satisfying x ď y for all members y of the chain. Bounded above (opptil

We are now prepared to formulate Zorn’s lemma, however we shall not prove it, begrenset)
Upper bound (øvre
only mention that it is equivalent to the axiom of choice (If you are interested in reading
skranke)
more about this, consult [[?]]) Bounded below (nedtil
begrenset)
Lower bound (nedre
Theorem 2.46 (Zorn’s lemma) Let Σ be a partially ordered set in which every chain is
skranke)
bounded above. Then Σ possesses a maximal element.

(2.47) A chain C in Σ is called saturated or maximal if it is not properly contained in any Saturated or maximal
larger chain; that is, if C1 is another chain with C Ď C1 , then C = C1 . A chain is saturated chains (mettede eller
maksimale kjeder)
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 2.48 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 (!!) the union CPC C is anew a chain; indeed, suppose that
x and y belong to the union so that there are chains Cx and Cy with x P Cx and y P Cy .
By assumption C is a chain, and so 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 fundamental 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, and in surprisingly many
situations such maximizing ideals turn out to be prime ideals.
In this section we shall precisely do this and establish a basic existence result with
several important applications, one being that every ring has at least one maximal ideal.
Our interest will be 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.

10th January 2021 at 11:32am

Version 4.0 run 30
 50 ideals

Theorem 2.49 (The Fundamental Existence Theorem for Ideals) A ring A, an ideal
a in A and a subset S not meeting a are given. Then there exists an ideal b maximal subjected to
the two following conditions
i) S X b = H;
ii) a Ď b.
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 theorem.
It is non-empty because a is supposed not to meet S and is a member of Σ. Obviously,
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 may 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 ab P b. If neither belongs to b, the ideals b + ( a) and b + (b) 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
( x + αa)(y + βb) = xy + αay + 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 2.50 (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
Prime factors frequently occur with higher multiplicities in a factorization of an integer
n, and it is of course interesting to get hold of the primes involved. In the transcription
of Kummer and Dedekind into the language of ideals, this leads to the notion of the
radical of a given ideal.
‘
The radical of an ideal (2.51) The radical a of a given ideal a in A consists of the elements a power of which
(radikalet til et ideal) lies in a; that is,
a = t a P A | an P a for some n P N u.
‘
‘
The nil radical The elements of a are also characterized as the elements in A whose residue class in
(nilradikalet) ‘
A/a is nilpotent. Along the same line, taking a to be the zero ideal, we see that (0) is
the set of nilpotent elements in A; it is called the nil radical of A.

10th January 2021 at 11:32am

Version 4.0 run 30
 existence theorems 51

‘
(2.52) The first thing to establish is that the radicala in fact is an ideal.
‘
Lemma 2.53 Let a be an ideal in the ring A. Then the radical a is an ideal.

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 N
N
( a + b) = 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 2.54 The set of nilpotent elements in A form an ideal.

(2.55) An ideal a in A is said to be radical if it equals its own radical; i. e. it holds true Radical ideals (radikale
idealer)
‘
that a = a. One easily verifies that the radical of an ideal is a radical ideal so that
‘ ‘ ‘
the equality ( a) = a holds true. In a similar manner as prime ideals and maximal
ideals radical ideals may be characterized in terms of quotients:

Proposition 2.56 An ideal a in the ring A is radical if and only if the quotient A/a is reduced.

Proof: The residue class [ a] in A/a of an element a is nilpotent precisely when a power
an lies in a, so that A is reduced precisely when a = a.
‘
o
‘
(2.57) The radical a must be contained in any prime ideal containing a because if
an P a and a Ď a with a prime, it holds that a P a, so a lies within the intersection of the
prime ideals containing it. The converse inclusion also holds and hinges on the basic
existence result above.

Proposition 2.58 (The radical as intersection of primes) Let A be a ring and assume
‘
that a is a proper ideal in 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 observed that a Ď a Ď p p, so assume that a is an element not
‘
lying in the radical a. We shall apply The Basic Existence Theorem (Theorem 2.49 on
page 49) 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 true that S X a = H, and by the theorem
we conclude that there is prime ideal a containing a disjoint from S; that is, a R a. o
(2.59) The special case that a = (0) merits to be pointed out:

10th January 2021 at 11:32am

Version 4.0 run 30
52 ideals

Corollary 2.60 The set of nilpotent elements in A equals the intersection of all prime ideals
in A; that is,
‘ č
(0) = p.
p prime

Of course the larger of two ideals, one containing the other, is not needed in an intersec-
tion, and one might be tempted to discard from the intersection in Proposition 2.58 the
prime ideals not being minimal among those containing a and thus write

‘ č
a= p, (2.4)
p minimal

where the intersection extends over all prime ideals minimal over a. Such a represen-
tation is certainly valid, but the argument is more complicated than indicated since
a priori there could be infinitely descending chains of distinct prime ideals. However,
Ş
if tpi uiP I is a chain of prime ideals, the intersection iP I pi is a prime ideal (you are
asked to check this in Exercise 2.21 on the next page), and so by Zorn’s lemma, every
prime containing a contains a prime ideal minimal among those containing a; and this
is exactly what we need to have a representation as in (2.4).
(2.61) The operation of forming the radical commutes with forming finite intersections;
one has:
Ş ‘ ‘Ş
Lemma 2.62 For every finite collection tai u of ideals in A the equality i ai = i ai holds
true.
‘
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 (here we use that the ni ’s are finite in number), 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
‘Ş
inclusion i ai Ď i ai . The other inclusion is straightforward. o

Examples
‘ ‘
(2.20) 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 = ( x1 , x22 , x33 , . . .)
generated by the powers xii in the polynomial ring k[ x1 , x2 , x3 , . . .] in countably many
variables. The radical of a equals the maximal ideal m = ( xi |i P N0 ) 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.
(2.21) The operation of forming radicals does not necessarily respect infinite intersections.
For instance, if p is a prime number, one has pr Z = pZ and therefore r pr Z = pZ.
‘ Ş ‘

But evidently it holds true that r p Z = 0.

Ş r

10th January 2021 at 11:32am

Version 4.0 run 30
 local rings 53

Exercises
ˇ (2.21) Unions and intersections of chains of primes. Let tpi uiP I be a chain of prime ideals
Ť Ş
in the ring A. Show that both the union iP I pi and the intersection iP I pi are prime
ideals. Show that every prime ideal containing a given ideal a contains a a prime ideal
minimal over a. Show that any prime ideal contained in a, is contained in a prime
maximal among those contained in a.
ˇ (2.22) Assume that p Ă q are two distinct prime ideals in the ring A. Show that there
are prime ideals p1 and q1 with p Ď p1 Ă q1 Ď q and so that there is no prime ideal lying
strictly between p1 and q1 .
ˇ (2.23) Saturated multiplicative sets and zero divisors. A multiplicatively closed set S in the
ring A is said to be saturated if with x it contains every factor of x; that is, if x P S and Saturated
x = yz, then y P S (and by symmetry z P S). multiplicative sets
(mettede multiplikative
a) Show that S is a saturated multiplicative set if and only if the complement AzS mengder)
is a union of prime ideals.
b) Show that the set of non-zero divisors in A form a saturated multiplicative set.
c) Conclude that the set of zero divisors in A is a union of prime ideals.
(2.24) Show that the group of units A˚ is contained in any saturated multiplicative
set.
ˇ (2.25) The prime ideals that appear as maximizing in the proof of Proposition 2.58 are
of special kind. Let S = tan u be the set of powers of an element from the ring A, and
let p be maximal among the ideals not meeting S. Show that the class [ a] in A/p is
contained in every non-zero prime ideal of A/p.
(2.26) Let a be a finitely generated ideal. Show that a sufficiently high power of a is
‘
contained in the radical a.
ˇ (2.27) Let A be a pid. Show that every ascending chain of ideals in A is eventually
constant.
M

2.6 Local rings

Rings having only a single maximal ideal are called local rings. They occupy a central Local rings (lokale
place in the theory being of a simpler kind of rings, and a frequently applied strategy of ringer)

proof is to reduce an issue to a statement about local rings. In the analogy with rings of ˚ Recall that a germ of
functions, the local rings correspond to rings of germs* of functions near a point—hence functions at a point is
the name—and the maximal ideal consists of the germs vanishing at the point. There is a class of function
coinciding on
also the notion of a semi-local ring, which is a ring with merely finitely many maximal neighbourhoods of the
ideals. point
(2.63) In a local ring A the complement of the maximal ideal m coincides with the group Semi-local rings
(semilokale ringer)
of units; that is, every element a P A not lying in m is invertible. Indeed, if it were not,

10th January 2021 at 11:32am

Version 4.0 run 30
 54 ideals

the principal ideal ( a) would be a proper ideal, and by existence of maximal ideals
(Theorem 2.50 on page 50) it would be contained in a maximal ideal, obviously different
from m, which is incompatible with m being the sole maximal ideal in A. This proves
that the first statement in the following proposition implies the second.

Proposition 2.64 Let A be a ring and m a proper ideal in A. The following three statements
are equivalent.
i) A is a local ring with maximal ideal m;
ii) The group of units and the complement of m coincide; that is, A˚ = Azm ;
iii) The ideal m is maximal and consists of elements a such that 1 + a is invertible.

Proof: To see that the last statement ensues from the second, observe that if a is a
member of m, then 1 + a is not in m and hence is invertible.
Finally, assume that m be maximal and that all elements be shaped like 1 + a with
a P m are invertible. Let x be an element not in m. Since m is maximal, it holds true that
m + ( x ) = A; hence x = 1 + a for some a P m, and x is invertible. This shows that iii)
implies i). o
The assumption in the last statement that m be maximal, is necessary; for an example
see Exercise 2.30 below.
(2.65) The argument in the previous paragraph partially goes through in a slightly more
Jacobson radical general staging involving the so-called Jacobson radical J ( A) of a ring A—the intersection
(Jacobson-radikalet)
Ş
of all the maximal ideals in A—that is, J ( A) = m Ď A maximal m.

Proposition 2.66 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 all elements of shape 1 + xa are
invertible. If there is a maximal ideal m so that a R m, it holds true that m + ( a) = A,
and there is a relation 1 = y + ax with y P m. It ensues that 1 ´ xa lies in m, but on
˚ Remember, maximal the other hand, 1 ´ ax is invertible by assumption; and we have the contradiction* that
ideals are proper ideals m = A.
Assume then that a lies in all maximal ideals. If (1 + ax ) were a proper ideal, it
would by the Fundamental Existence Theorem (Theorem 2.49 on page 49) be contained
in a maximal ideal n. Since a P n, it would follow that 1 P n, contradicting that n is
proper. Hence (1 + ax ) is not proper, and 1 + ax is invertible. o

The category of local rings

(2.67) Assume that A and B are two local rings whose maximal ideals are m A and
Local homomorphisms mB respectively. A map of rings φ : A Ñ B is said to be a local homomorphism, or a
(lokal homomorfier) map of local rings, if it holds true that φ(m A ) Ď mB . Equivalently, one may request that
φ´1 (mB ) = m A ( the inclusion φ´1 (mB ) Ď m A always holds since φ´1 (mB ) is a proper

10th January 2021 at 11:32am

Version 4.0 run 30
 local rings 55

ideal). The field k = A/m A is called the residue class field of A. Together with these
homomorphisms the local rings form a category LocRings. Ring maps between local
rings that are not local abound, a stupid example being the inclusion of a local domain
in a field; e. g. the inclusion of the ring Z( p) in Q (cfr. Example 2.23 below)

Examples
(2.22) The set of rational functions over a field k that may be expressed as P( x )/Q( x )
with P( x ) and Q( x ) polynomials and Q(0) ‰ 0, is a local ring whose maximal ideal
equals the set of the functions vanishing at the origin. The evaluation map given by
P( x )/Q( x ) ÞÑ P(0)/Q(0) identifies the residue field with the ground field k.

(2.23) Let p be a prime number and let Z( p) be the ring of rational numbers expressible
as n/m where the denominator m is relatively prime to p. Then Z( p) is a local ring
whose maximal ideal is generated by p. Even more is true, the only ideals in Z( p) are
the principal ideals ( pv ); indeed, every rational number lying in Z( p) may be written
as pν n/m with ν ě 0 and neither n nor m having p as factor. And among these ideals
( p) contains all the others. The residue class field of Z p is the finite field F p with p
elements.

(2.24) In a polynomial ring C[ x1 , . . . , xr ] for all points a P Cr the ideal of polynomials

vanishing at a is a maximal ideal. It follows that the Jacobson radical of C[ x1 , . . . , xr ]
equals (0).

(2.25) Assume that p and q are two prime numbers. Let A be the ring of rational numbers
with denominator relatively prime to pq. That is A = t n/m | n, m P Z, (m, pq) = 1 u.
The principal ideals by ( p) and (q) are the only two maximal ideals in A, and J ( A) =
( p) X (q) = ( pq).
K

Exercises
ˇ (2.28) Show that a ring has just one prime ideal if and only if its elements are either
‘
invertible or nilpotent. Prove that this is the case if and only if A/ (0) is a field.
(2.29) Let k be a field. Show that power series ring kJtK is a local ring with maximal
ideal (t)kJtK.
(2.30) 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 (2) and (3) whose intersection equals (6). What are the two residue fields? Show
that 1 + a is invertible in A for all members a P (6).
ˇ (2.31) 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

10th January 2021 at 11:32am

Version 4.0 run 30
56 ideals

A is a semi-local ring. Describe the maximal ideals and the residue fields. What is the
Jacobson radical?
ś
(2.32) Let k1 , . . . , kr be fields. Show that the product ring i k i is a semi-local ring.
What are the maximal ideals?
(2.33) Let f ( x ) be any polynomial in k [ x ] where k is a field. Show that k [ x ]/( f ( x )) is
semi-local.
ˇ (2.34) Let A be a principal ideal domain with infinitely many maximal ideals. Show
that J ( A) = (0).
ˇ (2.35) Show that the polynomial ring C[ x1 , . . . , xn ] has a vanishing Jacobson radical.
M

2.7 Direct products and the Chinese Remainder Theorem

Ideals in a direct product
ś
Let A = 1ďiďr Ai be a direct product of rings Ai . There is a simple description of
all the ideals in A in terms of ideals in the Ai ’s. One produces an ideal a in A from a
ś
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.
ś ś
Proposition 2.68 The ideals of A = 1ďiďr Ai are all of the form 1ďiďr ai where each ai
ś
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 i0 and pi0 is a prime ideal.
Proof: The first claim is already dealt with. To the second: the projection mappings
A Ñ Ai , coinciding with multiplication by ei , send an ideal a to ai = ei a, and hence
ś
induce maps A/a Ñ Ai /ai . In their turn these give rise to a ring-map A/a Ñ i Ai /ai ,
which evidently is surjective. It x is an element in A such that ei x P ai for each i, the
ř ř
element x belongs to a since x = i ei x and a = i ai . and the map is injective.
What remains to be verified is the statement about the prime ideals: it follows since
ś
the principal idempotents in i Ai /ai are orthogonal, and so when at least two of them
ś
are non-trivial, the product i Ai /ai is not an integral domain. o
(2.69) It is appropriate to give a comment about the zero ring at this stage. In Proposi-
tion 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 2.68; it might happen that ai = Ai so that A/ai is the zero ring.
Example 2.26 The description of the ideals in Proposition 2.68 is not valid for infinite
ś
products. For instance, the ideals in an infinite product iP I k i of fields are described

10th January 2021 at 11:32am

Version 4.0 run 30
 direct products and the chinese remainder theorem 57

by so-called filters and ultrafilters on the index set I. One example can be the set a of
strings ( ai ) with ai ‰ 0 only for finitely many i (which by the way equals the direct
sum of the k i ’s). This is easily checked to be an ideal; it is certainly not prime, but it is
contained in at least one maximal ideal as every proper ideal is. Ideals containing a can
not be of the form described in the theorem since a contains each k i . K

The Chinese Remainder Theorem

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

(2.71) Given a finite collection tai u1ďiďr of 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 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 2.72 (The Chinese Remainder Theorem) Let A be a ring and assume we are
given a finite collection of pairwise comaximal ideals tai u1ďiďr . Then the canonical reduction
ś
maps induce an isomorphism A/a1 X . . . X ar » 1ďiďr A/ai .

Proof: It suffices to find elements ai in A which are congruent one modulo ai and
ř
congruent zero modulo all the other ideals in the collection. Indeed, the sum i yi ai ,
with the yi ’s being arbitrary ring-elements, will then have the same residue class as yi
modulo ai .

10th January 2021 at 11:32am

Version 4.0 run 30
 58 ideals

For each pair of indices i and j with i ‰ j we may write 1 = cij + c ji with cij P a j and
c ji P ai . Then cij is congruent one modulo ai and congruent zero modulo a j . Hence the
ś
product ai = j‰i cij is congruent one modulo ai and congruent zero modulo a j for
j ‰ i; and we are done. o

Exercises
(2.36) 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.
(2.37) Show that two ideals whose radicals are comaximal, are comaximal.
(2.38) Show that 28y ´ 27z solves the simultaneous congruences x ” y mod 9 and
x ” z mod 4.
(2.39) Let A be a semi-local ring. Show that A/J ( A) is a product of fields.
ˇ (2.40) Assume that a1 , . . . , ar are pairwise comaximal ideals in the ring A.
a) Show that a1 and a2 ¨ . . . ¨ ar are comaximal;
b) Show that one has a1 ¨ . . . ¨ ar = a1 X . . . X ar ;
ś
c) For each i with 1 ď i ď r let bi = j‰i a j . Prove that the b1 , . . . , br are comaximal
ideals; i. e. that b1 + . . . + br = A.
ˇ (2.41) Determine integers representing the idempotents in Z/30Z and Z/105Z.
(2.42) Prove that a reduced ring decomposes as A » A1 ˆ . . . ˆ Ar where each Ai is an
algebra over a finite field.
ˇ (2.43) Locally nilpotent ideals. One says that an ideal n in a ring A is locally nilpotent if
? ?
each element in A is nilpotent. Show that for each ideal a in A it holds that a + n = a
whenever n is locally nilpotent. Let A Ñ B be a surjection of commutative rings whose
kernel is locally nilpotent. Show that the map Spec B Ñ Spec A is a homeomorphism.
(2.44) Lifting of idempotents. Let A Ñ B be a surjective map of (not necessarily
˚ For once, we work commutative*) rings whose kernel a is locally nilpotent; that is, every element of a is
with non-commutative nilpotent. Let e be an idempotent in B. The aim of the exercise is to show that there is
rings; it will useful to
lift idempotent an idempotent e in A mapping to e. Choose any element x in A that maps to e and let
endomorphisms of y = 1 ´ x.
modules over
commutative rings a) Show that xy P a.
b) Let n be such that ( xy)n = 0 and define the two elements e = iąn (2n i 2n´i
ř
i )x y
and γ = iďn (2n i 2n´i . Show that 1 = e + γ and that eγ = 0.
ř
i )x y
c) Conclude that e is an idempotent in A that maps to e.
M

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 such a decomposi-

10th January 2021 at 11:32am

Version 4.0 run 30
 graded rings and homogenous ideals 59

tion; just to mention two powerful tools: induction on the degree of the lowest or the
highest term. A class of rings sharing some of these nice properties polynomials have,
are the so-called graded rings whose elements possess a decomposition mimicking the
one of polynomials into homogeneous terms.
Even more forceful techniques are available to handle graded rings which satisfy
appropriate finiteness conditions. For instance, when all the homogenous components
Rν are finite dimensional vector spaces over some field k Ď R0 , the so called Hilbert
function h R (ν) = dimk Rν is a very strong invariant of R.
A the present stage of the course we merely scratch the surface of the theory of
graded rings, but they will reappear later at several occasions.
(2.73) A graded ring R is a ring together with a decomposition of the underlying abelian Graded rings (graderte
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.
(2.74) Elements from the subgroup Rν are said to be homogenous of degree ν. Notice that Homogenous elements
the zero element 0 lies in every one of the subgroups Rν . One can therefore not attribute (homogene elementer)

a well-defined degree to it, but it will consider it to be homogenous of any degree. From
a decomposition as in (2.5) it ensues that each non-zero element a in R can be expressed
ř
as a sum a = ν aν whose terms av are homogenous of degree ν merely finitely many
of which are different from zero. The aν ’s are uniquely determined by a and go under
the name of the homogenous components of a. Homogenous
Notice that R0 ¨ R0 Ď R0 , so R0 is a subring of R. Similarly, for every ν it holds true components (homogene
komponenter)
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.
(2.75) 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 says that a is
ř
a homogenous ideal whenever a = ν aν ; in other words: if a belongs to a then all the Homogenous ideals
homogeneous components of a belong to a as well. Since homogenous components are (homogene idealer)

unambiguously defined (or if you prefer, because aν X aµ = (0) whenever ν ‰ µ), the
À
sum is a direct sum, and we are entitled to write a = ν aν .

Proposition 2.76 Let a be an ideal in the graded ring R. The following three statements are
equivalent.
i) The ideal a is homogenous;
ii) All homogenous components of elements in a belong to a;
iii) The ideal a may be generated by homogenous elements.

10th January 2021 at 11:32am

Version 4.0 run 30
60 ideals

Proof: That the first two statements are equivalent, is just a rephrasing of how homoge-
nous ideals were defined. Let us then prove that i) and iii) are equivalent; so assume
that a is a homogenous ideal. The homogeneous components of all members 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 homogeneous
generators for a, and say ai is of degree di . Any element a P a can then be expressed as
ř
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 + di = d

where f i,ν denotes the homogeneous component of f i of degree ν, and where we in the
ř
last sum have recollected all terms f i,ν ¨ ai of the same 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
(2.77) 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 2.78 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 ( R/a)ν = Rv /aν .
À
Proof: This follows without great effort from the direct sum decompositions R = ν Rν
À
and a = ν aν . Notice first that as Rν X a = aν , there are natural inclusions Rν /aν Ď R/a.
ř
Hence any class [ a] P R/a with a decomposing as a = i ai in homogeneous terms
ř
decomposes as [ a] = i [ ai ], where we at will can consider the [ ai ]’s to be elements
ř ř
in R/a or in Ri /ai . Moreover, the classes [ ai ] are unique because if i ai and i bi
were two such decompositions inducing the same element in R/a, it would hold true
ř
that i ( ai ´ bi ) 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 [ ai ] = [bi ]. o

Example 2.27 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 of case two variables R = k [ x, y]. 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
Rν =
iα+ jβ=ν

10th January 2021 at 11:32am

Version 4.0 run 30
 graded rings and homogenous ideals 61

Since already R is the direct sum R = i,j k ¨ xi y j , one arrives at the direct sum R =
À
i j
À
ν Rν by just recollecting terms k ¨ x y with the same degree; that is, the polynomials
in Rν satisfy iα + jβ = ν. Some call these polynomials isobaric. K

Example 2.28 A natural and useful condition on a graded ring is that Rn = (0) 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 k( x1 , . . . , xn ) consisting of rational functions
shaped like f /gν where g is a fixed homogenous polynomial, and f P k[ x1 , . . . , xn ] 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

Exercises
(2.45) Generalize Example 2.27 above to polynomials in any number of variables by
giving each variable xi a weight αi .
ˇ (2.46) With reference to the Example 2.27 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.
ˇ (2.47) Homogeneous prime ideals. Let p be a homogenous ideal in a graded ring R. Show
that p is a prime ideal if and only of x ¨ y P p implies that either x P p or y P p for
homogenous elements x and y.
(2.48) Monomial ideals. An ideal a in the polynomial ring k[ x1 , . . . , xr ] is said to
be a monomial if it holds true that a polynomial f belongs to a if and only if every Monomial ideals
monomial occurring in f lies there. Show that this is equivalent to a being generated by (monomiale idealer)

monomials.
ˇ (2.49) Assume that k is an infinite field. The multiplicative group k˚ acts on the
polynomial ring k[ x1 , . . . , xr ] in a natural way; the result of the action of α P k˚ on the
polynomial f ( x1 , . . . , xr ) being f α ( x1 , . . . , xr ) = f (αx1 , . . . , αxr ).
a) Show that the polynomial f is homogeneous of degree d if and only if f α = αd ¨ f
for all α.
b) Show that an ideal a is homogeneous if and only if a is invariant under this
action.
(2.50) Assume that k is an algebraically closed field and that f ( x, y) is a homo-
geneous polynomial in k[ x, y]. Show that f ( x, y) splits as a product of linear factors.
Hint: If f is of degree d, it holds that f ( x, y) = yd f ( x/y, 1). Consider f ( x/y, 1) as a
polynomial in t = x/y.
ˇ (2.51) Homogenization of polynomials. Let k be a field and let f P k [ x1 , . . . , xr ] be any non-
zero polynomial. Let d be the degree of f . Define a fresh polynomial f H P k [ x0 , . . . , xr ]

10th January 2021 at 11:32am

Version 4.0 run 30
 62 ideals

Homogenization in one more variable by putting f H ( x0 , . . . , xr ) = x0d f ( x1 /x0 , . . . , xr /x0 ). Show that f H
(homogenisering) is homogenous of degree d. Show the quality f H (1, x1 , . . . , xr ) = f ( x1 , . . . , xr ).
ˇ (2.52) Dehomogenization of polynomials. The homogenization process described in the
Dehomogenization previous exercise has a natural reverse process called dehomogenization It is not canonical,
(avhomogenisering) but depends on the choice of a variables, which will be x0 in this exercise. When
g P k[ x0 , . . . , xr ] is homogenous of degree d, one puts g D ( x1 , . . . , xr ) = g(1, x1 , . . . , xr ).
Show that g = x0s ( g D ) H for some non-negative integer s ď d. Give examples to see that
s actually can have any value between 0 and d.
À
ˇ (2.53) 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. Conclude that
k[ x, y, z, w]/( xy ´ zw) is not a ufd. Hint: Work with components of highest degree.
(2.54) Let A = Z[ x, y, z, w]/( xy ´ wz) show that the class of x is irreducible but not
prime.
M

2.9 The prime spectrum and the Zariski topology

Prime spectrum of a Every ring has a geometric incarnation called the prime spectrum. It is denoted Spec A if
ring (Primspekteret til the ring is A, and its points are the prime ideals in A. The spectrum carries a topology
en ring)
called the Zariski topology after Oscar Zariski. The topological space Spec A depends
functorially on the ring 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 (a) (which is a prime ideal
in A).
The spectra of rings are the building blocks for the schemes as constructed in the
ingenious scheme theory of Alexander Grothendieck; they form an infinitely larger
ocean with as yet huge unexplored regions, where the spectra merely constitute 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 vice
versa. Both modern number theory and modern arithmetic are inconceivable without
the geometric language and the geometric intuition of spectra and schemes. However, in
these matters, we shall only superficially scratch the surface; giving 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 (or subsets of kr for any
algebraically closed field k) being the common zeros of a set of polynomials, are called

10th January 2021 at 11:32am

Version 4.0 run 30
 the prime spectrum and the zariski topology 63

closed algebraic sets. When they satisfy certain additional conditions, they are called
varieties, and varieties are the main objects of interest for many algebraic geometers. Varieties (varieteter)

Prime spectra
(2.79) In order to make it a genuine geometric object, the prime spectrum will be
endowed with a topology, which is called the Zariski topology, named after one of the The Zariski topology
fathers of modern algebraic geometry, Oscar Zariski. This topology is best defined by (Zariski topologien)

giving the closed subsets. With any ideal a in A is associated a closed subset denoted
V (a) whose members are the prime ideals containing a; that is, one has
V (a) = t p Ď A | p a prime ideal pĚ a u.
There are some axioms to be verified. First of all, V (0) = Spec A and V ( A) = H (recall
that prime ideals by definition are proper ideals), so the empty set and the entire space
are both closed. The two other axioms for a topology require that the union of finitely
many closed subsets is closed (it suffices to check it for the union of two) and that the
intersection of any family of closed sets is closed. To the former, observe that V (a) Y Oscar Zariski (1899–1986)
V (b) Ď V (ab) since both ab Ď a and ab Ď b hold. The inclusion V (ab) Ď V (a) Y V (b) Russian-born American
follows since if ab Ď p, either a Ď p or b Ď p according to Proposition 2.28 on page 40. mathematician
ř
Hence V (a) Y V (b) = V (ab). To verify the latter axiom, notice the trivial fact that i ai
lies in a if and only if each summand ai ’s lies in a. Summing up, we have shown most
of the following proposition.

Proposition 2.80 (The Zariski topology is a topology) Let A be a ring.

i) V (0) = Spec A and V (1) = H;
ii) For any ideals a and b in A it holds true that V (a) Y V (b) = V (ab);
ř Ş
iii) For any family tai uiP I of ideals one has V ( iP I ai ) = iP I V (ai );
iv) If a Ď b, then V (b) Ď V (a)
‘
v) V (a) = V ( a).
Proof: What remains to be observed are the two last assertions. It is trivial that
‘
a Ď b implies that V (b) Ď V (a), and the last assertion ensues from the radical a being
contained in any prime ideal containing a. o
(2.81) The Zariski topology has certain peculiar features never met when working 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 2.82 The closed points in Spec A are the maximal ideals.
Proof: We saw that every proper ideal is contained in a maximal ideal (Theorem 2.50
on page 2.50); hence V (a) will always have a maximal ideal as member. So if tpu is
closed; that is, equal to V (a) for some a, the prime ideal p must be maximal.

10th January 2021 at 11:32am

Version 4.0 run 30
 64 ideals

On the other hand whenever m is maximal, obviously V (m) = tmu since no prime
ideal strictly contains m. o

Examples
We do not intend to dive deeply into a study of prime spectra, but only to give a faint
idea of what might 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 closed (and hence all being open as well), the trivial
topology whose sole closed sets are the empty set and the entire space, and finally, we
Sierpiński space have the so-called Sierpiński space, a two-point space with just one of the points being
(Sierpiński-rommet) closed (and consequently the other being open). And two of these occur as Zariski
topologies.
(2.29) The direct product of two fields A = k ˆ k1 has merely the two prime ideals
(0) ˆ k1 and k ˆ (0) which both are maximal. Hence Spec k ˆ k1 consists of two points
and is equipped with the discrete topology.
(2.30) The ring Z( p) of rational numbers expressible as fractions with a denominator
prime to p has just two prime ideals, namely (0) and the principal ideal ( p) (Exam-
ple 2.23 on page 55). Hence t(0)u is an open set being the complement of the closed
point ( p). Hence Spec A is the Sierpiński space.
(2.31) 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.
(2.32) The spectrum of a polynomial ring: As a counterweight to the peculiarity of the
previous examples, let us consider a more mainstream situation, namely the spectrum
Spec k[t] of the polynomial ring over an algebraically closed field k (let it be C, if you
want). Ideals in k [t] are all principal and are prime when generated by irreducible
polynomials. But k being algebraically closed, the only irreducible polynomials are the
linear ones, and so all non-zero prime ideals are maximal and of the shape (t ´ a) for
a P k. Thus the closed points of Spec k[t] are in a one-to-one-correspondence with k.
Additionally, Spec k [t] contains one point (0) (the zero-ideal is prime). It is neither open
nor closed, and its closure is the entire spectrum Spec k[t], and it goes under the name
The finite complement of the generic point. This looks familiar when k = C, we just get C adjoined one generic
topology point, but be aware that the topology is far from being the usual one. The only closed
(endelig-komplement-
topologien) sets are the finite unions of closed points. This topology is frequently called the finite
complement topology since the non-empty open sets are precisely those with a finite and
closed complement (see also Exercise 2.56 below).
K

10th January 2021 at 11:32am

Version 4.0 run 30
 the prime spectrum and the zariski topology 65

Functoriallity
(2.83) The spectrum Spec A depends functorially on the ring A; it is a functor from
the category of (commutative) rings to the category of topological spaces. To justify
this assertion, we have got to tell how maps between rings are affected. If φ : A Ñ B
is a map of rings, pulling back ideals along φ takes prime ideals to prime ideals;
indeed, that ab P φ´1 (p) means that φ( a)φ(b) P p, and so either φ( a) P p or φ(b) P p
whenever p is prime; hence φ´1 (p) is prime. This allows the definition of the map
r : Spec B Ñ Spec A simply as the inverse image map, the important observation being
φ
that φ
r is continuous:

Proposition 2.84 The map φ̃ is continuous.

Proof: Let a Ď A be an ideal. The one has φ r´1 (V (a)) = V (aB); indeed, tautologically it
´ 1
holds true that a Ď φ a if and only if φ(a) Ď a. o
It is clear that when φ and ψ are composable maps between rings, it holds true that
φĆ˝ψ = ψ r and it is totally trivial that id
r ˝ φ, Ą A = idSpec A . So sending A’s to Spec A and
φ’s to φ indeed yields a functor.
r

Inverse images
(2.85) A byproduct of the proof above is that the inverse image under φ r of the closed
set V (a) 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 2.19
on page 36), and these are, as we saw in the proof, precisely the points in Spec B
mapping to points in V (a). Moreover, the whole lattice of ideals I( B/aB) is isomorphic
to the lattice of ideals in B containing aB. This takes care of the topology; closed sets
correspond to closed sets, and we have established the following:

Proposition 2.86 Let φ r : Spec B Ñ Spec A be induced by φ : A Ñ B. Then the inverse

´ 1
image φ (V (a)) is homeomorphic to Spec B/aB. In particular, for any point a P Spec A the
r
˚ The name finite
fibre over a is naturally homeomorphic with Spec B/aB.
complement
topology; a slightly
Exercises misleading name since
(2.55) Let A and B be two rings. Show that Spec ( A ˆ B) is the disjoint union of Spec A Y is not open, but has a
finite complement. The
and Spec B. non-empty opens in the
(2.56) Finite complement topology. Let Y be an infinite set and let η be a point not in Y induced topology on Y,
however, are those with
(whatever you want but a point in Y). The union X = tηu Y Y has a topology* whose a finite complement.
closed sets, apart from tηu, H and X itself, are the finite subsets of Y.
a) Show that this is a topology.
b) Let k be a field and let Y be the set of monic irreducible polynomials with
coefficients from k. Show that X is homeomorphic to Spec k[t].

10th January 2021 at 11:32am

Version 4.0 run 30
 66 ideals

c) Show that if k and k1 are two fields of the same cardinality, then the spectra
Spec k[t] and Spec k1 [t] are homeomorphic.
(2.57) Let c : C Ñ C for a moment denote complex conjugation. Describe the action of
cr on the spectrum Spec C[t]. What are the fixed points? Describe the map Spec C[t] Ñ
Spec R[t] induced by the inclusion R[t] Ď C[t].
ˇ (2.58) Distinguished open sets. The Zariski topology has a particular basis of open sets
The distinguished open called the distinguished open sets. For each element f P A there is one such open set D ( f )
sets (særskilte åpne whose members are the prime ideals not containing f ; that is, D ( f ) = t p | f R p u.
mengder)
a) Show that D ( f ) is open.
b) Show that the distinguished open sets form a basis for the Zariski topology on
Spec A.
c) Let a be an ideal in A and let tD ( f i )u be a family of distinguished open sets.
Show that they form a cover of Spec AzV (a) if the f i ’s generate a.
d) Show that Spec A has the compactness property: any covering by distinguished
open sets can be reduced to a finite covering. Hint: Spec A is the complement
of V (1).
M

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 3

Unique factorization domains

When working with integers the Fundamental Theorem of Arithmetic is a most valuable
tool used all the time, consciously or unconsciously. In a general ring however, a
corresponding theorem does not hold, and one has to do without it. The birth of Unique factorization
algebraic number theory, and by that, the beginning of commutative algebra, came as domains (entydig-
faktoriseringsområder)
a response to this “defect”. Luckily, in certain nice rings the Fundamental Theorem
persists. These rings are called unique factorization domains or factorial rings, and are the Factorial rings
objects we shall study in this chapter. Out of the inherent human laziness springs the (faktorielle ringer)

acronym ufd, which is in widespread use.

ufd’s (ufd)
3.1 Being a unique factorization domain
(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 (3.1)

of irreducible elements pi . The qualifyer "essentially unique" must be understood 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. Or in the terminology introduced in Paragraph 2.6
on page 31, the factors are unique up to order and association.
The definition has two separate conditions—a stipulation of existence and a unique-
ness requirement—and the two are of a quite different flavour. Before proceeding with
the theory, we shall take a closer look at each separately.

Existence
(3.2) The first conditions stipulates that any element can be expressed as a finite product
of irreducibles. This is in essence a finiteness condition on A, which is fulfilled e. g. in
68 unique factorization domains

the so-called Noetherian (important rings to be introduced later). It does certainly not
hold in general rings; an example can be the ring of entire functions in the complex
plane C. The irreducibles in this ring are of the form u(z)(z ´ a) where u(z) is a unit
(i. e. a non-vanishing entire function) and a P C is any point, so any entire function
with infinitely many zeros—like our good old friend sin z—can not be expressed a finite
product of irreducibles.
One can always attempt a recursive attack in the search for a factorization. Any
ring element a which is not irreducible, is a product a = a1 a2 of two non-units. These
being irreducible, makes us happy—we have a factorization—but if one or both are not,
they are in their turn products of non-units. If all the fresh factors are irreducible, we
are again happy; if not, some split into products of non-units. Continuing like this we
establish a recursive process which, if terminating, yields a finite factorization of a into
irreducibles.
In general the process may go on for ever—like it will for e. g. sin z—but in many
cases there are limiting condition making it end. For instance, in the case of the ring
of integers Z, the number of steps is limited by the absolute value | a|, and in the case
of polynomials in k[ x ] by the degree of a. There is a general finiteness condition that
guarantees the process to stop as expressed in the following lemma. It comes in the
disguise of a condition of the partial ordered set of principal ideals and is an anticipation
of the notion of Noetherian rings.

Lemma 3.3 Let A be a domain such that any non-empty collection of principal ideals has a
maximal element. Then every element in A can be expressed as a finite product of irreducibles.

Proof: Let Σ be the set of principal ideals ( a) where a runs through all counterexamples
to the lemma; that is, all elements not expressible as a finite product of irreducibles.
If the lemma were false, the set Σ would be non-empty, and by assumption it would
have a maximal member, say (b). By construction b can not be irreducible and may be
factored as b = b1 b2 with neither b1 nor b2 being a unit; hence (b) is strictly contained
in both (b2 ) and (b2 ). On the other hand, b is not a finite product of irreducibles, and
a fortiori the same holds for either b1 or b2 . Therefore either (b1 ) or (b2 ) belongs to Σ,
which is impossible since both strictly contain the maximal member (b). o
Exercise 3.1 Apply Zorn’s lemma to prove that any principal ideal domain satisfies the
condition in Lemma 3.3. Hint: Any ascending chain of principal ideals must terminate.
M

Uniqueness
(3.4) The second condition a ufd must abide to is the uniqueness requirement, which is
of a more algebraic nature. It generally holds true that prime elements are irreducible,
and the uniqueness requirement essentially boils down to the converse holding true; i. e.

10th January 2021 at 11:32am

Version 4.0 run 30
 being a unique factorization domain 69

that irreducible elements be prime. In fact, in any domain the uniqueness requirement
holds automatically for finite factorizations into prime elements:

Lemma 3.5 Let A be a domain. Assume that tpi u1ďiďr and tqi u1ďiďs are two collections 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 renumbering the qi ’s and adjusting q1 by a unit, we may assume that p1 = q1 .
Cancelling p1 gives p2 ¨ ¨ ¨ pr = q2 ¨ ¨ ¨ qs , and induction finishes the proof. o
(3.6) Since irreducibles are prime in both the rings of integers Z and of polynomials
k [ x ] over a field k (both are principal ideal domains), we immediately conclude that Z
and k[ x ] are factorial rings.

Examples
The main examples of factorial rings are principal ideal domains and polynomial
rings over those—as we shortly shall see—but producing other examples demands
some technology not available to us for the moment. So we confine ourselves to give
some classical examples of non-factorial rings, one from number theory and two from
algebraic geometry.
(3.1) Our first example, the ring Z[i 5], is ubiquitous in number theory texts, and we
‘

already met it on page 46. In Z[i 5] the number 6 has two distinct factorizations in
‘

irreducibles (see Example2.18 on page 46):

‘ ?
6 = 2 ¨ 3 = (1 + i 5)(1 ´ i 5).

(3.2) One of the standard example from algebraic geometry, which geometers would call
the coordinate ring of "the cone over a quadratic surface in projective 3-space", is the
quotient ring A = k[ X, Y, Z, W ]/( XY ´ ZW ). Indicating the classes in A of the variables
by lower case versions of their name, we have A = k[ x, y, z, w] with constituting relation

xy = zw. (3.2)

In Example 3.4 on page 79 we saw that A is a domain, and the polynomial XY ´ ZW

being homogenous, A is a graded ring. It is not too challenging to see that the class of
any non-zero linear form is irreducible in A (see Exercise 2.53 on page 62). Hence the
relation (3.2) shows that A is not factorial.

10th January 2021 at 11:32am

Version 4.0 run 30
 70 unique factorization domains

This also gives an easy example of the intersection of two principal ideals being non-
principal; i. e. their intersection is distinct from their product, in that ( x ) X (z) = ( xy, xz).
(3.3) Elliptic curves I: Another famous example from algebraic geometry can be
A = k[ X, Y ]/(Y 2 ´ X ( X ´ a)( X ´ b)) where a and b are elements in a field k whose
characteristic does not equal two. In Exercise 3.2 below you are asked to show that A is
an integral domain. The relation

y2 = x ( x ´ a)( x ´ b), (3.3)

where x and y denotes the classes of X and Y in A, holds in A and gives two different
decompositions of an element into irreducibles; of course one must verify that the
involved linear factors are irreducible (see Exercise 3.6 on page 81 below). Plane curves
given by equations like
y2 ´ x ( x ´ a)( x ´ b)

with a, b P k are called elliptic curves when a and b are different and non-zero; to be
˚ Complete elliptic precise one should say affine elliptic curves on Weierstrass form*. Elliptic curves have
curves come in quite a always been at the centre-stage of algebraic geometry and are closely related to the
lot of disguises, but can
all be brought on a so-called elliptic functions—in fact, they were the very starting point for the development
normal form called the of modern algebraic geometry.
Weierstrass form. If the
characteristic is Below we have included the sketch of the real points of such a curve with a and
different from two or b real. Pictures can be beautiful and instructive, but should be taken with a grain of
three, their affine
salt; if the ground field for instance, is the algebraic closure of the field F3 with three
incarnations are as
described here. elements or for that matter, the closure of the rational function field Q(t), the picture is
of no relevance.
K

Exercises
(3.2) The aim of the exercise is to prove that for any field k the cubic polynomial
Karl Theodor Wilhelm
Weierstrass (1815–1897) y2 ´ x ( x ´ a)( x ´ b) is irreducible in the polynomial ring k [ x, y]. Were it not, it would
German mathematician

10th January 2021 at 11:32am

Version 4.0 run 30
 being a unique factorization domain 71

have a linear factor, and one could write

y2 ´ x ( x ´ a)( x ´ b) = (αx + βy + γ) Q( x, y) (3.4)

with α, β and γ constants from k and not both α and β being zero.
a) If β ‰ 0, substitute y = ´β´1 (αx + γ) in (3.4) to obtain the impossible polyno-
mial identity
β´2 (αx + γ)2 ´ x ( x ´ a)( x ´ b) = 0.
b) If β = 0, substitute x = ´α´1 γ to make the right side of (3.4) vanish. Conclude
that the left side will be a monic quadratic polynomial in y which is identically
zero; which is absurd!
(3.3) Let the k-algebra A = k[ x, y, z] have the constituting relation

y2 z ´ x ( x ´ az)( x ´ bz)

where a are b are scalars.

a) Show that A is a graded domain Hint: Cast a glance at Exercise 3.2.
b) By evoking Exercise 2.53 on page 62, prove that y and x ´ zc for any c P k are
irreducible elements in A.
M
Irreducibles and primes in a ufd
(3.7) As we saw (Proposition 2.41 on page 45), in a domain prime elements are always
irreducible, and as we are going to see, in factorial domains the converse holds as well.
Among domains satisfying an appropriate finiteness condition that guarantees existence
of a factorization—as for instance the Noetherian domains we shall come to—this even
characterizes the factorial domains (Lemma 3.5 above).

Proposition 3.8 For members of a ufd being prime is equivalent to being irreducible.
Proof: As primes always are irreducible, we merely need to see that irreducibles are
prime. So assume that a is irreducible, and that a|xy. Let x = p1 ¨ ¨ ¨ ps and y = q1 ¨ ¨ ¨ qt
be decompositions into irreducibles. Then of course

xy = p1 ¨ ¨ ¨ ps ¨ q1 ¨ ¨ ¨ qr

is a factorization into irreducibles as well. On the other hand, xya´1 is an element in A

and has a factorization xya´1 = r1 ¨ ¨ ¨ rm into irreducibles; hence

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

are two equal products of irreducibles. The ring we work in being a ufd, irreducible
factors coincide up to order and units, and this means that, up to a unit, a is either one
of the pi ’s or one of the qi ’s. Phrased differently, a divides either x or y. o

10th January 2021 at 11:32am

Version 4.0 run 30
 72 unique factorization domains

3.2 Common divisors and multiples

The greatest common (3.9) In a ufd any two elements a and b have a greatest common divisor; we shall denote
divisor (største felles it by gcd( a, b). Recall that this is an element d such that d|a and d|b and such that any
divisor)
other element x dividing both a and b divides d; or phrased differently, x|a and x|b
implies that x|d. Expressed in terms of ideals, d is an element so that the principal ideal
(d) is the least among the principal ideals containing both the principal ideals ( a) and
(b). The greatest common divisor of a and b is only determined up to an invertible
factor; however, the principal ideal (d) is unambiguously defined.
The notion of a greatest common divisor of two elements is meaningful in any
domain, but in most domains not every pair has one. However, as mentioned above, in
a factorial ring, any two elements do. The ufd’s share this property with several other
classes of rings, for instance, the rings of holomorphic functions in open domains in
the complex plane C. These rings are not ufd’s, but have the property that any finitely
generated ideal is principal.
The least common (3.10) Two elements may also have a least common multiple: An element m in A so that
multiple (minste felles the principal ideal (m) is greatest among the principal ideals contained in ( a) and (b);
multiplum)
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. We shall denote the least common multiple
of a and b by lcm( a, b) Again, merely the principal ideal (m) is unambiguously defined.

Proposition 3.11 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 of a and b
into irreducibles, say a = p1 ¨ ¨ ¨ pr and b = q1 ¨ ¨ ¨ qs , and pick up the “common factors”:
Reordering the factors, we find a non-negative integer t so that ( pi ) = (qi ) for i ď t and
( pi ) ‰ (qi ) for t ą i. Then d = p1 ¨ ¨ ¨ pt is a greatest common divisor. It might of course
happen that no ( pi ) equals any (q j ), in which case t = 0, and the greatest common
divisor equals one.
To lay hands on a least common multiple of a and b mimic what we just did, or
verify that a ¨ b/ gcd( a, b) is a least common multiple of a and b. o
Exercises
(3.4) Show that two elements a and b from a domain A have a least common multiple
if and only if the intersection ( a) X (b) is a principal ideal.
(3.5) Let a and b be two elements having a gcd( a, b) and a lcm( a, b). Show that
gcd( a, b) ¨ lcm( a, b) = ab up to a unit.
(3.6) Prove that in a ring where all finitely generated ideals are principal, all pairs of
elements have a gcd and a lcm.
M

10th January 2021 at 11:32am

Version 4.0 run 30
 a criterion of kaplansky’s 73

3.3 A criterion of Kaplansky’s

(3.12) The author is especially fond of the formulation* in the following criterion found ˚ The slogan is: "A

in Irving Kaplansky’s book [?] and whose proof is an elegant application of the Basic domain A is factorial if
and only if every prime
Existence Theorem. contains a prime".

Proposition 3.13 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 3.5 on page 69, 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 prime
Irving Kaplansky (1917–2006)
elements. It is certainly multiplicative closed, and A having at least one maximal ideal
Canadian mathematician
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, which does not lie in Σ. Then ( x ) X Σ = H; indeed, if there was an expression
xy = p1 ¨ ¨ ¨ pr with the pi ’s being prime elements, we could chose one with r minimal,
and this would force all the pi ’s to divide x. Hence y would be a unit, and consequently
x P Σ. By the Basic Existence Theorem (Theorem 2.49 on page 49), there is a prime
ideal in A maximal subjected to not meeting Σ and containing x. That prime ideal is
not the zero ideal and by assumption therefore has a prime element as member, 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 and
more elementary ways):

Corollary 3.14 Every pid is a ufd.

Proof: Prime ideals are generated by prime elements. o

There is an important corollary of Kaplansky’s criterion valid for domains for which
every non-zero ideal contains a prime ideal minimal among the non-zero prime ideals.

3.4 Gauss’ lemma and polynomials over factorial rings

Every domain A is contained in a canonically determined field called the fraction field of Fraction field
A. The elements are fractions of shape a/b with a and b elements from A and, of course, (kvotientkropp)

with b ‰ 0. The arithmetic of these fractions is governed by the usual rules for rational

10th January 2021 at 11:32am

Version 4.0 run 30
 74 unique factorization domains

fractions. For the moment we have not shown that such fields 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 met so far are a priori contained in a field, like Z
?
and Z[ d] which are subrings of C, and a polynomial ring k[ x1 , . . . , xr ] over a field k
is contained in the rational function field k ( x1 , . . . , xr ). So a priori these rings have a
fraction field.
(3.15) The objective of the current section is to establish the important result that
polynomial rings over ufd’s are udf’s; a result that hinges on the key concepts of a
primitive polynomial and the content of a polynomial, and a key lemma, the so-called
Gauss’ lemma.
Gauss’ lemma was initially an approach to the issue of comparing factorizations
of a polynomial in Z[ x ] and Q[ x ], but has of course a broader horizon nowadays. As
an illustration of the general mechanism consider, for instance, the simple polynomial
12x + 57 which has the factorisation 3 ¨ (4x + 19). When viewed as a polynomial in Z[ x ],
it is not irreducible—neither 3 nor (4x + 19) is a unit—but considered an element in
Q[ x ], it is. Indeed, 3 becomes invertible in Q[ x ].

Content and primitive polynomials

(3.16) Proceeding along the lines of the example in the previous paragraph, imagine
a general polynomial f ( x ) = a0 + a1 x + . . . + ar xr with coefficients from a factorial
Primitive polynomials ring A. Extracting the greatest common divisor c f of the coefficients ai one may write
(primitive polynomer) f = c f ¨ f 7 where f 7 P A[ x ] is a polynomial whose coefficients have no common divisor.
Content of a This naturally leads to the concept of a primitive polynomial: a polynomial over a ufd
polynomial (inneholdet is called primitive if the greatest common divisor of the coefficients equals 1.
til et polynom)
With a splitting f = c f ¨ f 7 as above, where f 7 primitive and c f P A, it is widespread
usage to call the element c f the content of f . It is not unambiguously determined by
f ; being a greatest common divisor, it is merely unique up to an invertible factor, and
of course, f 7 suffers the same ambiguity. Strictly speaking one should consider c f an
element in the quotient group K˚ /A˚ or equivalently, as the ideal (c f ) in the set of
principal ideals.
(3.17) The notion of content may be extended to polynomials from K [ x ] as well: Any
such polynomial may be expressed as f = c f ¨ f 7 with c f P K˚ and f 7 a primitive
polynomial in A[ x ]; just multiply f by an element d P A so that d ¨ f has coefficients in A
(for instance, use the least common multiple of the coefficients), and put c f = d´1 ¨ cd f
and f 7 = (d f )7 .
Both c f and f 7 will automatically be unique up to units in A. Indeed, if a relation
cg = c1 g1 with g, g1 primitive polynomials in A[ x ] and c, c1 P K˚ holds true, for some
a P A one has acg = ac1 g1 with ac, ac1 P A. The content of polynomials in A[ x ] being

10th January 2021 at 11:32am

Version 4.0 run 30
 gauss’ lemma and polynomials over factorial rings 75

unique up to a unit, it ensues that ac = uac1 for some unit u P A˚ . Cancelling a shows
that c = uc1 .
Notice that f has coefficients in A if and only if the content c f belongs to A, and
Johann Carl Friedrich Gauss
that f is primitive if and only if c f is a unit in A. (1777–1855)

Gauss’s lemma German mathematician

(3.18) This is one of many basic results to be found in Gauss’ immortal Disquisitiones
Arithmeticae. He wrote it in 1798 when he was 21 years old, and it was published in 1801.
The Disquisitiones Arithmeticae is one of the most influential mathematical publications
ever written; certainly among the all time top ten.

Lemma 3.19 (Gauss’s lemma) Assume that A is a ufd. Let f and g be primitive polynomials
in A[ x ]. Then the product f g is primitive.

Proof: Write f = 0ďiďn ai xi and g = 0ďiďm bi xi . Let d be a non-unit in A. The

ř ř

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
consequently the sum is not divisible by d. o
(3.20) The next lemma is an equivalent version of Gauss’ lemma formulated in terms
of the content of polynomials. It clearly implies Gauss’ lemma, just apply it to two
primitive polynomials, but as noted, the lemmas turn out to be equivalent.

Lemma 3.21 Assume that A is a ufd with fraction field K. For non-zero polynomials f and g
in K [ x ] it holds true that c f g = c f c g up to a unit factor.

Proof: One has f = c f ¨ f 7 and g = c g ¨ g7 which gives f g = c f c g f 7 g7 , but according to

Gauss’ lemma f 7 g7 is a primitive polynomial, and from the unicity of the content we
deduce that c f g = c f c g up to units. o
(3.22) To facilitate future reference (and hopefully to make things clearer for the
students) we sum up what we so far have done in this section:

Proposition 3.23 Let A a ufd with fraction field K. With very non-zero polynomial f P K [ x ]
is associate an element c f P K˚ called the content of f , unique up to a unit factor. It holds true
that f = c f ¨ f 7 where f 7 is a primitive polynomial in A[ x ], and this characterizes c f and f 7 up
to unit factors.
a) The content depends multiplicatively on f ; that is, c f g = c f c g up to units;

10th January 2021 at 11:32am

Version 4.0 run 30
76 unique factorization domains

b) f lies in A[ x ] if and only if c f P A;

c) f is primitive if and only if c f P A˚ .

Factoring polynomials over A and over K

We return to the issue mentioned in the beginning of the section: how does the
factorizations of polynomials in A[ x ] relate to its factorization in K [ x ]? In the general
setting with A any ufd and K its field of fractions.
(3.24) Let f be a primitive polynomial in A[ x ] and assume that f splits as a product
f = gh in K [ x ]. It follows that f = c g ch g7 h7 , and from the content being unique up
to units it ensues that c g ch being an associate to c f is a unit in A. Incorporating c g ch
in either g7 or h7 , we arrive at a factorization f = g1 h1 with f ,1 g1 P A[ x ]. Notice that
the new factors g1 and h1 are obtained from f and g merely by multiplying them by
elements from A, hence their degrees are preserved. We have thus established the
following lemma:

Lemma 3.25 Assume that A is a ufd with quotient field K. A primitive polynomial f P A[ x ]
that f splits as product in K [ x ], splits as a product in A[ x ] with factors of the same degree. In
particular, if f is irreducible in A[ x ], it remains irreducible in K [ x ].
(3.26) We then come to one of our main objectives in this chapter:

Theorem 3.27 If A is a ufd, then the polynomial ring A[ x ] is a ufd. The irreducible elements
in A[ x ] are the irreducible elements in A and the primitive, irreducible polynomials in A[ x ].
Induction on the number of variables and the fact that k[ x ] is a ufd immediately give the
corollary that polynomial rings over fields are ufd’s. It is also worthwhile mentioning
that the same applies to polynomials rings over the integers.

Corollary 3.28 If A is a ufd, the polynomial ring A[ x1 , . . . , xr ] is a ufd. In particular the

rings Z[ x1 , . . . , xr ] and k [ x1 , . . . , xr ] where k is a field, are ufd’s.
Proof of 3.27: Invoking Kaplansky’s criterion (Proposition 3.13 above) it suffices to
see that any prime ideal p in A[ x ] contains a prime element. If p X A is non-zero, this
follows from A being a ufd. If not, let f P p be a primitive polynomial of minimal
degree. We contend that f is prime; so assume that f |gh. Certainly f will be irreducible
(it is of minimal degree), it persists being irreducible in K [ x ] and is therefore prime in
K [ x ] (the ring K [ x ] is factorial). Consequently f divides either g or h, say g = p f with
p P K [ x ]. But p is forced to lie in A[ x ]; indeed, g = c p p7 f , and c p equals c g up to a unit
in A, hence belongs to A. o

Ideals in polynomial rings over pid’s.

Another consequence of Gauss’ lemma is that it helps us better understand the prime
ideals in the polynomial rings A[ x ] over a pid A—this includes the polynomial rings

10th January 2021 at 11:32am

Version 4.0 run 30
 gauss’ lemma and polynomials over factorial rings 77

k [ x, y] in two variables over a field k and the ring Z[ x ]—and hopefully it will make you
better appreciate David Mumford’s drawing of Spec Z[ x ] in his famous red book [?], a
copy is shown below.
(3.29) The description of Spec A[ x ] we are about to give, requires the principal ideal
domain A to have infinitely many maximal ideals. When Spec A is finite, some of the
principal ideals ( g( x )) might be maximal as illustrated in Exercise 3.8 below which
treats the case when A has just one maximal ideal.

Proposition 3.30 Let A be a pid with infinitely many maximal ideals. Then the non-zero
prime ideals p in A[ x ] are of two kinds.
i) If p is maximal, then p = ( g( x ), p) where p P A is a prime element and g( x ) is a
polynomial in A[ x ] which is irreducible mod p.
ii) If p is not maximal, it is principal and generated either by an irreducible and primitive
polynomial or by a prime element in A.

Proof: The trick is to consider the intersection q = p X A and separately treat the two
cases according to q being zero or not. Let K be the fraction field of A.
Assume first that q = 0. Then pK [ x ] is a proper ideal: Assume there is a relation
ř
1 = i ai f i with ai P p and f i P K [ x ] and multiply through by a common denominator d
of all coefficients of the f i ’s, and thus obtain d P p. Now d P A and p X A = 0, and we
have a contradiction.
It ensues that pK [ x ] is principal, say generated by f . Replacing f by its primitive
avatar f 7 , we may assume that f belongs to A[ x ] and is primitive. We contend that
p = ( f ( x )). Indeed, if g P p we may write g = h f were h a priori belongs to K [ x ].
However, since f is primitive, we find that ch = ch c f = c g lies in A, and hence h lies in
A [ x ].
Now, an ideal ( g( x )) generated by an irreducible polynomial g( x ) is not a maximal:
if it were, it would hold that ( g( x ), p) = A[ x ] for all primes p P A; in other words,
g( x ) would be a unit in the polynomial ring A/( p)[ x ] for all primes p. But the leading
coefficient of g( x ) has only finitely many prime factors (up to units), and the leading
term survives in the reduction of g( x ) modulo any prime not among those. Our ring A
is assumed to have infinitely many primes, so the reduction of g( x ) mod most of the
primes in A (in fact, ifinitely many) is not a unit. Contradiction.
Next, if q ‰ 0, it holds that q = ( p) for some prime element p P A. Consider
A[ x ]/pA[ x ] = k [ x ] were k is the residue class field k = A/( p). The ideal p/( p) is
either zero, in which case p = ( p) A[ x ], or it is generated by an irreducible polynomial
a( x ), and any lift g( x ) of a( x ) to A[ x ] will then generate p together with p; that is,
p = ( p, g( x )). o

10th January 2021 at 11:32am

Version 4.0 run 30
 78 unique factorization domains

(3.31) Specializing A to be the polynomial ring k [y] with k being an algebraically closed
field, Proposition 3.30 yields the following description of the maximal ideals in k[ x, y],
which is a precursor to the fabulous and all important Nullstellensatz of David Hilbert:

Theorem 3.32 (Nullstellensatz in dimension two) Let k denote a field that is algebraic-
ally closed. Then every maximal ideal m in the polynomial ring k[ x, y] is of the form m =
( x ´ a, y ´ b) with a, b in k.
Proof: Indeed, the only irreducible polynomials in k [ x ] are the linear ones. o

Exercises
(3.7) Assume that p( x ) is a monic polynomial in Z[ x ] which factors as p( x ) = r ( x )s( x )
in Q[ x ]. Show that r ( x ) and s( x ) both lie in Z[ x ] and are monic. Hint: Multiply through
by the least common multiple of the coefficients’ denominators and appeal to Gauss’s
lemma
ˇ (3.8) Polynomial rings over dvr’s. Let A be a local pid (such rings are called discrete
valuation rings abbreviated with the initialism dvr; they will be treated extensively in a
later chapter) and let π be a generator for the maximal ideal; for instance, localizations
like Z( p) or C[t](t´a) are shaped like that.
a) Show that the principal ideal a = (πx ´ 1) in the polynomial ring A[ x ] is
maximal. Hint: Let K be the fraction field of A and show that a equals the
kernel of the map A[t] Ñ K that sends f ( x ) to f (1/π ).
b) Show that any maximal ideal m in A[ x ] either is shaped like ( g( x ), π ) with
g( x ) a polynomial in A[ x ] which is irreducible modulo π, or like ( g( x )) with
g( x ) being an irreducible polynomial in A[ x ] that is invertible modulo π.
Hint: Adapt part of the proof of Proposition 3.30.
(3.9) Separable polynomials and derivatives. Let A be ufd with fraction field K.
Separable polynomials Polynomials without multiple roots in any field extension of K are said to be separable.
(separable polynomer) Show that a polynomial p(t) is not separable; that is, it has a multiple root in some

10th January 2021 at 11:32am

Version 4.0 run 30
 gauss’ lemma and polynomials over factorial rings 79

field extension K1 of K, if and only if there is a polynomial q(t) P A[t] of positive degree
which divides both p(t) and its derivative p1 (t). Hint: Let α be the multiple root. Show
that p(α) = p1 (α) = 0. Consider the minimal polynomial of α over K and make it
primitive.
M

Factoring homogenous polynomials

(3.33) 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 homoge-
nous might sometimes be helpful because one then knows a priori that any irreducible
factor is homogeneous; indeed, one has the following proposition.

Proposition 3.34 Let R be a graded factorial domain satisfying Rn = 0 for n ă 0. Then the
irreducible factors of a homogeneous element are homogeneous.

Proof: Let ai be the irreducible factors of a homogeneous element a and develop each
ř
ai as a sum ai = ν ai, ν of 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, νi 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
Example 3.4 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 3.5 The polynomials y p ´ x q are irreducible when p and q are relatively prime.
To see this give k[ x, y] the grading for which deg x = p and deg y = q. An irreducible
polynomial without constant term, unless it equals α ¨ x or β ¨ y, necessarily contains
non-zero terms both of the form α ¨ yn and of the form β ¨ x m for natural numbers n and
m and scalars α and β. If it additionally is homogeneous, it holds true that nq = mp
and hence n = ap and m = bq for some non-negative integer a. It follows that if f is an
irreducible factor of y p ´ x q it is of degree at most pq, so either it reduces to x or y, or

10th January 2021 at 11:32am

Version 4.0 run 30
 80 unique factorization domains

its degree is pq; the former case does obviously not occur, so f has degree pq, which is
the same as the degree of y p ´ x q , and the two are equal up to a scalar. K
Exercises
ˇ (3.10) Show that if n ě 3, it holds that q( x ) = 1ďiďn xi2 is irreducible in k[ x1 , . . . , xn ]
ř

unless that characteristic of k equals two.

(3.11) Let p1 , p2 and p3 be pairwise coprime integers. Consider the polynomial
p p p
f = x1 1 + x2 2 + x3 3 in the polynomial ring k[ x1 , x2 , x3 ] over the field k. Show that f is
irreducible.
˚ This includes all
(3.12) Let R be a graded ring satisfying Rn = 0 for n ă 0. Show that only homogeneous
connected Lie groups. units in R are those in R˚0
For students initiated
in Lie–theory, it is a ˇ (3.13) Let G be a group without finite quotients* and A a factorial ring. Assume
nice exercise to verify that G acts on A in away that the units are invariant. Show the irreducible factors
this (What is the
derivative of the
of any polynomial semi-invariant under G are semi-invariant. (Recall that f is called
n-power map?). semi-invariant if there is a group homomorphism χ f : G Ñ A˚ so that f g = χ f ( g) f .)
M

3.5 Example: Quadratic extensions and the norm

An indispensable tool in algebraic number theory is the norm, but it certainly plays a
role in algebraic geometry as well. After having treated so-called integral extensions we
shall expound on the norm and its stablemate the trace—but the norm is indispensable
for establishing certain important good examples. hence this ad hoc exclusive treatment
for the class of quadratic extensions.
We have already have come across the norm at occasions, e. g. when showing that 2
?
and 3 were irreducible elements in the quadratic extension Z[i 5], but in the guise of
the square of the ordinary absolute value of complex numbers.
We have seen several examples where adjoining a square root to a ring is a central
feature; the quadratic extensions Z[ n] where n is an integer are shaped like that, as is
‘

the coordinate ring of an affine elliptic curve: it equals k[ x, y] with constituting relation
y2 = x ( x ´ a)( x ´ b) so obtained form k[ x ] by adjoining x ( x ´ a)( x ´ b).
‘
‘
And of course, one is not confined to use cubic polynomial, but can adjoin p( x )
where p( x ) is any polynomial (well, it must have certain good properties as being
without multiple roots). The rings one obtains in this way are coordinate rings of the
so-called affine hyperellipitic curves (they will be more closely discussed in Exercise 3.20).
These ring extensions have many features in common, and here we shall explore
some. So we set the staging by assuming that A is a domain, and pick an element d
from A which is not a square; furthermore we let B = A[t]/(t2 ´ d). Denoting the class
‘ ‘ ‘
of t by d, one may write B = A[ d] and think about B as A with the square root d
adjoined.

10th January 2021 at 11:32am

Version 4.0 run 30
 example: quadratic extensions and the norm 81

‘
Every b P B can be written as b = x + y d with x and y unique elements from a.
indeed, no polynomial in A(t) of degree one lies in (t2 ´ d), just consider the top term.
The multiplication in B is given by the formula

dy)( x1 + dy1 ) = ( xx1 + yy1 ) + ( xy1 + x1 y) d

‘ ‘ ‘
(x + (3.5)

The extension B comes equipped with a conjugation σ : B Ñ B map defined by

‘ ‘
σ ( x + y b) = x ´ y b. The multiplication formula above immediately gives the
relation σ ( ab) = σ ( a)σ (b), and as evidently the conjugation is additive, it is a ring
homomorphism. It is obviously an involution (applied twice it gives the identity), and
if the characteristic of A is not two, the invariant elements are precisely the members of
‘ ‘
A; indeed x + y = x ´ y d if and only iff y = 0 since we are in characteristic different
from two (in characetristic two σ degenerates into he identity). We have shown:

Lemma 3.35 The map σ is an involutive ring homomorphisms whose ring of invariants equals
A.

Next we introduce the norm N (b) of elements from B as the product N (b) = bσ (b).

Proposition 3.36 Let A, B and d be as above.

a) The norm is multiplicative; i. e. N ( ab) = N ( a) N (b);
b) N ( x + y d) = x2 ´ dy2 ;
‘

c) An element b P B is invertible in B if and only if the norm N (b) is invertible in A.

Proof: The statement a) follows as the conjugation is a ring map: N ( ab) = abσ ( ab) =
abσ ( a)σ (b) = aσ ( a)bσ (b) = N ( a) N (b). Next b) enuses directly from the multiplication
formula (3.5). Finally we prove c): if b is invertible it holds that N (b) N (b´1 ) =
N (bb´1 ) = N (1) = 1; and if N (b) is invertible, N (b)´1 σ (b) serves as the inverse of b. o
Example 3.6 Elliptic curves II: Let k be field and a and b two elements from k. Let A be
the coordinate ring of an affine elliptic curve on Weierstrass form; that is, A = k[ x, y]
with constituting relation y2 = x ( x ´ a)( x ´ b). We let p( x ) = x ( x ´ a)( x ´ b). The aim
of this example is to complete Example 3.3 on page 70 by showing that A is not factorial.
It remains to see that the factors in (3.3) are irreducible.
‘
We consider the extension k[ x ] Ď k[ x, y] = A where y = p. Elements are of the
form f + yg with f , g P k [ x ] whose norm equals N ( f + yg) = f 2 ´ pg2 . This is clearly
a scalar if and only if g = 0 and f is constant; and citing Proposition 3.36 above we
conclude that the non-zero constants are the only units in A.
A salient points is the observation that the norm N ( f + yg) never can be of degree
one; indeed, if g ‰ 0 its degree is at least three, and if g = 0, its degree is even.
With these observation up in our sleeve, we can check that any linear expression
z = αy + β( x ´ γ) in x and y is irreducible (where α, β and γ lie in k). Assume that

10th January 2021 at 11:32am

Version 4.0 run 30
 82 unique factorization domains

there is a factorization z = uv and apply the norm to obtain

N (u) N (v) = β2 ( x ´ γ)2 ´ α2 p.

If α ‰ 0, the degree of the right hand side is three, and one of N (u) and N (v) must be
of degree three and the other constant (degree one is forbidden) ; or if α = 0, the right
hand side is of degree two, so that either N (u) or N (v) must be of degree two and the
other constant. In both cases, either u or v is invertible, and that is the end of the affair.
K
Example 3.7 Units in real quadratic extensions: Contrary to the imaginary quadratic
extensions Z[i n] that have a finite unit group, the real quadratic extensions Z[ n] (in
‘ ‘

both cases n is a natural number that is not a square) have infinitely many units, which
constitute a group isomorphic to Z ˆ µ2 . We shall not prove this, but indicate why it
holds true.
So let a = x + y n with x, y P Z be an element in Z[ n]. Its norm is given as
‘ ‘

N ( a) = x2 ´ my2 and according to statement c) of Proposition 3.36 above, a is a unit if

and only if N ( a) is a unit; that is if and only if x and y satisfies the equation

x2 ´ ny2 = ˘1.

Pell’s equations (Pell’s This equation is called Pell’s equations; it is loaded with anecdotes and history and can
ligning) be traced far back. It seems that the Indian mathematician Brahmagupta treated it
extensively as early as in 628. It is not very deep, but requires a certain amount of work,
to see that
x2 ´ ny2 = 1
‘

has a solution for any n.

Let us take for granted that there is a nontrivial unit v in Z[ n], and deduce that
‘

there are infinitely many. We begin with showing there is a smallest unit larger than
one. Indeed, if u = x + y n, it holds that u´1 = ˘1( x ´ y n) (according to N (u) = 1
‘ ‘

or N (u) = ´1) so that u ´ u´1 = 2y n or u ´ u´1 = 2x. In both cases u ´ u´1 will
‘

be bounded away from zero since x and y are integers; so no sequence of units can
approach 1 from above, and hence there must be a smallest one larger than one. Denote
that smallest unit by u0 ; it is called the fundamental unit, and we contend that every
other unit is up to sign a power of u0 : Let u be any non-trivial unit. By exchanging u
with ´u or u´1 or ´u´1 if necessary, we may assume that u ą 1, and thus u ě u0 . Let r
be the smallest natural number so that u0r ě u. Then 1 ď u0r u´1 ă u0 ; and in view of
the minimality of u0 we conclude that u0r u´1 = 1.
Apart from the existence of nontrivial units, we have thus shown that Z[ n]˚ »
‘

µ2 ˆ Z. K

10th January 2021 at 11:32am

Version 4.0 run 30
 example: quadratic extensions and the norm 83

Exercises
ˇ (3.14) Consider the ring Z[i 2k] with k ě 3 an odd natural number. Show that
‘

p = (2, i 2k) is not a principal ideal, but that p2 = (2). Prove that the ring Z[i 2k ] is
‘ ‘

not factorial.
(3.15) Show that 3 and 5 are irreducible members of Z[i 14] that are not prime.
‘
?
(3.16) Consider the ring Z[i 14]. Prove that
? ?
34 = (5 + 2i 14)(5 ´ 2i 14)
?
and show that all the involved elements are irreducible elements of Z[i 14].
?
(3.17) Assume that d is an integer such d ” 1 mod 4. Let α = (1 + d)/2. Show that
α2 = α + (d ´ 1)/4. Prove that A = Z[α] 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. Describe the norm-map.
ˇ (3.18) The ring of real trigonometric polynomials. Let A = R[ x, y] with constituting
relation x2 + y2 = 1. This is the ring of real polynomial functions on the unit circle in
R2 , or if you wish, it may viewed as the ring of trigonometric polynomials; just put
x = sin t and y = cos t.
a) Show that R[ x ] is a polynomial ring and that A is a free module of rank two
over R[ x ] with 1 and y as a basis.
b) Let N denote the norm defined by the extension R[ x ] Ď A. Show that the norm
is given as N ( f ( x ) + g( x )y) = f 2 ( x ) ´ g2 ( x )(1 ´ x2 ), and that the non-zero
constants are the only units in A.
c) Show that y, (1 ´ x ) and (1 + x ) are irreducible elements in A and conclude that
A is not a ufd. Hint: y2 = (1 ´ x )(1 + x ).
(3.19) The ring of complex trigonometric polynomials. Contrary to the ring A from the
previous exercise, the ring B = C[ x, y] with constituting relation x2 + y2 = 1 is a ufd, it
is even a pid.
a) Show that all non-zero and proper ideals in B are of the form ( x ´ a, y ´ b) with
a, b complex numbers such that a2 + b2 = 1. Hint: Theorem 3.32 on page 78.
b) Show that B = C[u, us] with u = x + iy and us = x ´ iy and that uus = 1.
c) Show that ( x ´ a, y ´ b) = (u ´ c) where u = x + iy and c = a + ib.
(3.20) Hyperelliptic curves. Let k be a field. Let p be a polynomial in k [ x ] of degee n
without multiple roots, and let A = k[ x, y] with constituting relation y2 = p. This ring
is the coordinate ring of a so-called hyperelliptic curve Hyperelliptic curves
a) Show that A is an integral domain which is a free k[ x ]-module of rank two. (hyperelliptiske kurver)

b) Compute the norm map A Ñ k[ x ].

c) Assume that the degree of p is odd. Show that the units in A coincide with the
non-zero constants and that A is not factorial.

10th January 2021 at 11:32am

Version 4.0 run 30
84 unique factorization domains

d) Assume the characteristic of k is not two. Show by way of examples that for
each even n there are hyperelliptic curves with non-constant units. Hint: Let q
be a polynomial of degree ν assuming each of the values 1 and ´1 in ν distinct
points and consider p = (1 + q)(1 ´ q).
ˇ (3.21) Determine the fundamental unit in Z[2], Z[ 3] and Z[ 5].
‘ ‘

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 4

Modules

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. Ideals for instance, are
modules, and any over-ring is a module over the subring, to mention two instances.
An abelian group is nothing but a Z-module, and a module over the polynomial ring
k [t] over a field k is just a vector space over k endowed with an endomorphism; so the
module theory encompasses the theory of 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 Modules (moduler)
structures. It is endowed with an underlying structure as an abelian 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 ( a, m) will be denoted by a ¨ m or
simply by am. It is subjected to the following four conditions:
i) a(m + m1 ) = am + am1 ;
ii) ( a + a1 )m = am + a1 m;
iii) 1 ¨ m = m;
iv) a ¨ ( a1 m) = ( aa1 ) ¨ m.
where a, a1 P A and m, m1 P M are arbitrary elements. The first condition requires the
action to be A-linear; or in other words, the map A Ñ HomSets ( M, M ) that sends a
to the “multiplication-by-a-map” m ÞÑ am must take values in the ring HomAb ( M, M)
of group-homomorphism. The last three requirements ensure that this map is a ring
homomorphism; it is the ring A that acts.
(4.2) One recognizes these conditions from linear algebra; they are word for word the
same as the vector space axioms, the sole difference being that A is not required to be a
 86 modules

field, but can be any ring. So in case A is a field k, there is nothing new; a k-module is
just a vector space over k. However, one should not draw this analogy to far; general
modules are creatures that behave very differently from 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 finite number, say r, copies of A. The elements are r-tuples ( a1 , . . . , ar ), and
addition is performed componentwise. The action of A is also defined componentwise:
˚ The direct sum plays a ¨ ( a1 , . . . , ar ) = ( a ¨ a1 , . . . , a ¨ ar ). We insist on this being an additive* construction and
an additive role in the shall write rA for this module, not Ar as is common usage in linear algebra.
category of A-modules,
the multiplicative role (4.3) Another familiar class of modules are the abelian groups. They are nothing but
is taken by another
modules over the ring Z of integers. An integer n acts on an element from the abelian
construct, the tensor
product. But that’s for group by just adding up the appropriate number of copies of the element and then
a later chapter. 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 instance, k[ x, y] is
a k[ x ]-module as is k[ x, x´1 ]. And k[ x ] will be a module over the subring k[ x2 , x3 ]. If
η P C is a root of unity, Z[η ] is a Z-module .
More generally, any ring homomorphism φ : A Ñ B induces an A-module structure
on B through the action a ¨ b = φ( a)b of an element a P A on b P B. This gives B the
Algebras (algebraer) structure of an A-algebra as defined in Paragraph 1.12 on page 19.
(4.5) Suppose that k is a field. Giving a k[t]-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
p(t) in k [t] acts on M as p(t) ¨ m = p(τ )(m).
K
ˇ Exercise 4.1 Let A and B be two rings and assume that B has an A-module structure
compatible with the ring structure; i. e. a ¨ bb1 = b( a ¨ b1 ). 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
Module also in our present case of modules. An A-module homomorphism φ : M Ñ N between
homomorphisms
(modulhomomorfier)

10th January 2021 at 11:32am

Version 4.0 run 30
 the axioms 87

two A-modules M and N is a homomorphism of the underlying abelian groups that

respects the action of A; that is, φ( am) = aφ(m) 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 . (It is easily checked that the
composition of two A-linear maps is A-linear and, of course, identity maps are A-linear
as well.) An A-module homomorphism φ : M Ñ N is said to be an isomorphism if there Isomorphisms of
is another one ψ : N Ñ M which is a two-sided inverse to φ; i. e. it holds true that modules
(modulisomorfier)
φ ˝ ψ = id N and ψ ˝ φ = id M . One easily verifies that it suffices to be bijective for φ to
be an isomorphism.
(4.3) The set Hom A ( M, N ) of A-linear maps from M to N is naturally contained in the
set HomZ ( M, N ) of group homomorphism from M to N (which are just the additive
maps), consisting of those commuting with the actions of A on M and N. It is well-
known that the sum of two additive maps is additive, and when both commute with the
actions of A, the sum does so as well. Therefore Hom A ( M, N ) is an abelian group, and
defining a ¨ φ as the map that sends m to aφ(m), 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,
one finds
a ¨ φ(bm) = a(bφ(m)) = b( aφ(m)),
where the first equality holds since φ is A-linear and the second because A is commuta-
tive*. The module axioms follow readily. ˚ For non-commutative

(4.4) The module Hom A ( M, N ) depends functorially on both variables M and N, and rings A the set
Hom A ( M, N ) is in
historically, it was one of the very first functors to be studied. The dependence on general merely an
the first variable is contravariant—the direction of arrows are reversed— whereas the abelian group; it does
not carry an A-module
dependence on the second is covariant—directions are kept. The induced maps are
structure unless
just given by composition. Of course, such constructions are feasible in all categories, further hypotheses are
what is special in Mod A is that Hom A ( M, N ) is an A-module and the induced maps are imposed on A or the
involved modules.
A-linear. The technical name is that category Mod A has internal homs—the set of maps
stay within the family! To be precise let M, N and L be three A-modules and ψ : N Ñ L φ
/N
M
an A-linear map. Sending φ to ψ ˝ φ yields an associated map
ψ
ψ˚ : Hom A ( M, N ) Ñ Hom A ( M, L).
φ˚ (ψ) 
L
It is A-linear, and if ψ : L Ñ L1 is another A-linear map, one has (ψ1 ˝ ψ)˚ = ψ˚1 ˝ ψ˚ . In φ
/N
M
a similar fashion, the contravariant upper-star version
ψ
φ˚ : Hom A ( N, L) Ñ Hom A ( M, L),
φ˚ ( ψ ) 
L
which sends ψ to ψ ˝ φ, is A-linear as well, and it is functorial; i. e. (φ1 ˝ φ)˚ = φ˚ ˝ φ1˚
for composable maps φ and φ1 .
(4.5) It follows readily from the involved maps being A-linear that composition of

10th January 2021 at 11:32am

Version 4.0 run 30
 88 modules
˚ That is, φ and φ1 are
maps from M to N and
composable maps is an A-bilinear operation. That is, one has
ψ and ψ1 from N to L.
φ ˝ ( aψ + a1 ψ1 ) = aφ ˝ ψ + a1 φ ˝ ψ1 and ( aφ + a1 φ1 ) ˝ ψ = aφ ˝ ψ + a1 φ1 ˝ ψ,

where the maps are composable*, and a and a1 denote ring elements.
Submodules
Submodules (4.6) A submodule N of an A-module M is a subgroup closed under the action of A; in
(Undermoduler) other words, for arbitrary elements a P A and n P N it holds true that an P N, and of
course, N being a subgroup the sum and the difference of two elements from N belong
to N.
Examples
(4.6) Ideals in the ring A are good examples of submodules, and in fact, by definition,
they are all the 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 finite
ř
linear combinations i ai mi with ai P a and mi P M is a submodule.
(4.8) Given an ideal a in A. The set (0 : a) M of elements in M annihilated by all
members of a form a submodule. It holds true that φ ÞÑ φ(1) gives an isomorphism
Hom A ( A/a, M ) » (0 : a) M . Indeed, the value φ(1) of a map φ : A/a Ñ M is killed by a
since for a P a it holds true that a ¨ φ(1) = φ( a ¨ 1) = φ([ a]) = 0. To obtain a resiproque
map, assume that m P (0 : a) M is given. The product x ¨ m does only depend on the
class [ x ] of x as ( x + a) ¨ m = x ¨ m for elements a that kill m. Hence [ x ] ÞÑ x ¨ m is a
legitimate definition of a map A/a Ñ M, and it takes the value m at 1.
K
Exercises
ˇ (4.2) Show that there is a canonical isomorphism Hom A ( A, M ) » M. Hint: The
correspondence is φ Ø φ(1).
(4.3) Show that Hom A ( A/a, A) = 0 whenever a is a non-zero ideal in a domain A.
Show, e. g. by giving examples, that the equality does not necessarily hold true when A
is not a domain. Find an example with A having just four elements!
(4.4) Let p and q be two prime numbers. Show that HomZ (Z/pZ, Z/qZ) = 0 if p ‰ q,
and that HomZ (Z/pZ, Z/pZ) » Z/pZ.
(4.5) Let a and b be two ideals in the ring A. Show that there is a canonical isomorphism
Hom A ( A/a, A/b) » (b : a)/b. Hint: The correspondence is φ Ø φ(1).
(4.6) Assume that M is an A-module. For each element a P A let [ a] denote the
"multiplication–by–a” map in M. Let N be a second A-module. Show that [ a]˚ = [ a]
and [ a]˚ = [ a] (for each occurrence of [ a], it should self-explanatory in which of the
modules M, N or Hom A ( M, N ) multiplication by a takes place).
M

10th January 2021 at 11:32am

Version 4.0 run 30
 ˚ The set I( M ) forms
what is called a
complete lattice. A
partially ordered set is
called a lattice when
every pair of elements
thepossesses
axiomsa least 89upper
bound and a greatest
lower bound, and it is
The lattice of submodules
said to be complete if
(4.7) Just like the ideals in A the submodules of a given A-module M form a partially the same is true for any
ordered set under inclusion* which we shall denoted I( M). subset. In our case the
Ş greatest lower bound is
The intersection iP I Ni of a collection tNi uiP I of submodules of M is a submodule. the intersection and the
It is the largest submodule of M contained in all the submodules from the collection. In least upper bound the
sum.
the similar way, the smallest submodule containing all the modules in the collection is
ř
the sum iP I Ni , whose elements are finite A-linear combinations of elements from the
Ni ’s; that is, the elements are shaped like
ÿ
ai mi , (4.1)

where mi P Ni , and the ai ’s are elements from A only finitely many of which are non-
zero. More generally, for any set S Ď M there is a smallest submodule of M containing
S; it is called the submodule generated by S and consists of elements as in (4.1) but with Submodules generated
the mi ’s confined to S. by elements
(undermoduler
Exercise 4.7 This exercise parallels the list of properties of direct and inverse images generert av elementer)
of ideals in Propositions 2.11 and 2.13 on pages 33 and 33. Let φ : M Ñ N be an A-linea
map between A-modules. Show that for any submodule L Ď N the inverse image φ´1 ( L)
is a submodule of M, and that the induced map φ´1 : I( N ) Ñ I( M ) respects inclusions,
takes arbitrary intersections to intersection. What about sums of ideals? And if a is an
ideal show that Show that aφ´1 ( L) Ď φ´1 (aL), but give examples that strict inclusion
occurs (one can even find examples with φ an inclusion).
In the same vein, show that φ( L) is a submodule of N for each submodue L Ď M,
and that the induced map I( N ) Ñ I( M ) respects inclusions. What happens with
intersections and sums? And what about φ(aL)? M
Kernels and images
(4.8) An A-module homomorphism φ : M Ñ N is in particular a group homomorphism
and as such has a kernel and an image. Both these subgroups are submodules as well;
this ensues from the equality aφ(m) = φ( am) satisfied by A-linear maps. Indeed, one
immediately sees that the image is closed under multiplication by elements from A,
and if φ(m) = 0, it follows that φ( am) = aφ(m) = 0 as well.
Quotients
(4.9) Just as with ideals in a ring, one can form quotient of a module by a submodule
and the construction is word for word the same. Let M be the module and N the
submodule. From the theory of abelian groups we know that the two underlying
additive groups have a quotient group M/N, which is formed by the cosets [m] = m + N
for m P M. 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 ¨ [m] = [ am]. Of

10th January 2021 at 11:32am

Version 4.0 run 30
 90 modules

course, some verifications are needed. The first is to check that the class [ am] only
depends on the class [m] and not on the representative m, which is the case since
a(m + N ) = am + aN Ď am + N. Secondly, the module axioms in Paragraph 4.1 must
be verified; this is however straightforward and left to the zealous students.
(4.10) The quotient group M/N comes together with the canonical defined additive
map π : M Ñ M/N that sends m to the class [m]. Moreover, by the very definition
of the module structure on M/N, this map is A-linear, and it enjoys the important
universal property that any A-linear map that vanishes on N, factors through it:

π / M/N Proposition 4.11 (Universal property of quotients) Let N be a submodule of the A-

M
module M. The quotient map π : M Ñ M/N enjoys the following universal property. For
ψ every A-module homomorphism φ : M Ñ L with N Ď ker φ there exists a unique A-linear map
φ
  ψ : M/N Ñ L so that φ = ψ ˝ π.
L
Proof: The proof is mutatis mutandis the same as for (abelian) groups. The map φ
vanishes on N and is therefore constant on the residue classes [m] = m + N, and ψ([m])
is defined as (and compelled to be) that constant value. Since φ is A-linear and vanishes
on N, the constant value on [m + m1 ] = m + m1 + N equals φ(m) + φ(m1 ), and on
[ am] = am + N it is aφ(m). Hence ψ is A-linear. o

Corollary 4.12 (The first ismorphism theorem) An A-linear map φ : M Ñ N which

is surjective, induces an isomorphism M/ ker φ » N.
M
π / M/ ker φ
Proof: By the universal property φ factors through a map ψ : M/ ker φ Ñ N. This is
φ
» surjective since φ is, and injective since it kills the kernel. (That a class [ x ] goes to zero,
 implies that φ( x ) = 0, hence x P ker φ). o
N (4.13) One easily establishes the two following results. The analogue assertions for
abelian groups are well known, and the proofs persist being valid for modules as
well. The proofs are left as exercises; for inspiration either recall the proofs for abelian
groups (these proofs go through mutatis mutandis) or take a look at the corresponding
statements for ideals.

Proposition 4.14 Let π : M Ñ M/N be the quotient map. The “inverse-image-map”

π ´1 : I( M/N ) Ñ I(M)

that sends and ideal a to π ´1 (a), is a one-to-one correspondence between submodules of M

containing N and submodules of M/N. It respect inclusions, arbitrary intersections and
arbitrary sums.

Proof: The proof is similar to the proof of Proposition 2.19 on page 36 and is left to the
students as an exercise. o

10th January 2021 at 11:32am

Version 4.0 run 30
 the axioms 91

Proposition 4.15 (The second isomorphism theorem) Assume N and N 1 are two sub-
modules of M. There are then canonical isomorphisms where in the second one assumes that
N 1 Ď N:
i) ( N + N 1 )/N 1 » N/N X N 1 ;
ii) ( M/N 1 )/( N/N 1 ) » M/N.

Proof: The proof is similar to the proof of the isomorphism theorem for ideals (Theo-
rem 2.21 on page 37) and is left as a DIY-proof. o

Cokernels
(4.16) In a famous paper Alexander Grothendieck introduced axiomatically the notion
of an abelian category. The axioms reflect the main categorical properties of the module Abelian category
category Mod A . Among the requirements is that there is a zero-objects and all hom-sets (Abelske kategorier)

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”. Additive categories
Let us also mention that a category with a zero object whose hom-sets are abelian (Additive kategorier)

groups and whose compositions are bilinear is called an additive category. When, as is
true of Mod A , the hom-sets are A-modules and the compositions A-bilinear, it is said to
an A-linear category. And not to forget, a zero-object in a category is an objcet 0 so that A-linear category
for each object C in the category, there is precisely one arrow from 0 to C (so that 0 is (A-lineære kategorier)

an initial object) and dually, for each C there is precisely one arrow from C to 0 (and 0 Zero object (nullobjekt)
is thus a final or a terminal object).
(4.17) The definition of the cokernel in a categorical vernacular must be formulated The cokernel
exclusively in terms of arrows and therefore given as a universal property. The cokernel (Kokjernen)

of an A-linear map φ : M Ñ N is an A-linear map π : N 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 N/ im φ fulfils that requirement, and hence serves as the
cokernel of φ. The two stablemates kernel and cokernel are from a categorical viewpoint /N / coker φ
φ π
M
dual concepts, and a definition of the kernel just in terms of arrows is as indicated with
ψ
a diagram in the margin: The kernel is an arrow ι : ker φ Ñ M satisfiying φ ˝ ι = 0 with 0
  ~
the universal property that any A linear map ψ : L Ñ N such that φ ˝ ψ = 0, factors L
through it.
M[ o NO o
φ ι
ker φ
@
Proposition 4.18 Every A-module homomorphism φ : M Ñ N has a kernel, an image and a 0
ψ

cokernel. 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

10th January 2021 at 11:32am

Version 4.0 run 30
 92 modules

Examples
(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 more cryptical than obvious, think twice.)
(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 [ x ] of x modulo a since ( x + a)m = 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 ideal a that kills M.
(4.11) Given a ring map φ : A Ñ B between two rings A and B, 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 φ( a) ¨ m. 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 overdecorated symbols
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
(4.19) In this section we work with a collection tMi uiP I of modules over the ring A. The
ś
Direct products of underlying abelian group of the the direct product iP I Mi will be the direct product
modules (direkte of the abelian groups underlying the Mi ’s, which should be well known from earlier
produkter av moduler)
courses. The elements are strings or tuples (mi )iP I indexed by the set I, and the addition
is performed componentwise; i. e. (mi ) + (m1i ) = (mi + m1i ). In case I is finite, say
I = t1, . . . , ru, an alternative notation for a tuple is (m1 , . . . , mr ). The actions of A on
the different Mi ’s induce an action on the direct product, likewise defined component
for component: a ring element a acts like a ¨ (mi ) = ( a ¨ mi ). The module axioms in
paragraph (4.1) are easily verified component by component, and we have an A-module
ś
structure on iP I Mi .
ś
The projections πi : iP I Mi Ñ Mi are A-linear simply because the module opera-
tions in the product are performed componentwise.

10th January 2021 at 11:32am

Version 4.0 run 30
 direct sums and direct products 93

Direct sums
À
(4.20) The direct sum of the module collection tMi uiP I is denoted by iP I Mi and is Direct sums of modules
defined as the submodule of the direct product consisting of strings m = (mi )iP I with (direkte summer av
moduler)
all but a finite number of the mi ’s vanishing.
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 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 (the elements my be considered to be 2-adic expansions of real numbers).

Universal properties φ
/ś
N iP I Mi
(4.21) Both the product and the direct sum are characterised by a universal properties. It
πi
is noticeable that these properties are dual to each other; reversing all arrows in one,
& 
φi
yields the other. For this reason the direct sum is frequently called the co-product in the Mi
parlance of category theory.
N of
φ À
We first describe the universal property the direct product has. The set-up is an iP O I Mi
A-module N and a collection of A-linear maps φi : N Ñ Mi , and the outcome is that ιi
ś φi
there exists a unique A-linear map φ : N Ñ iP I Mi such that πi ˝ φ = φi . Indeed, this
Mi
amounts to the map φ(n) = (φi (n))iP I being A-linear.
In the case of the direct sum the universal property does not involve the projections,
À
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
À
maps φi : Mi Ñ N, and the conclusion is that there exists a unique map φ : i P I Mi Ñ N
so that φ ˝ ι j = φj . The map φ is compelled to be defined as

ÿ
φ ( mi ) = φi (mi ),
iP I

and this is a legitimate definition since merely finitely many of the mi ’s are non-zero.
Exercise 4.8 Work out all the details in the above reasoning. M
(4.22) 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 4.23 There are canonical isomorphisms

À ś
i) Hom A ( iP I Mi , N ) » iP I Hom A ( Mi , N );
ś ś
ii) Hom A ( N, iP I Mi ) » iP I Hom A ( N, Mi ).

10th January 2021 at 11:32am

Version 4.0 run 30
94 modules

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.
Exercise 4.9 Figure out the precise definitions of the isomorphisms in Proposition 4.23
above. Hint: The key word is universal properties. M
À
(4.24) We shall identify each module M j with the image ι j ( M j ) in iP I Mi under the
natural inclusion ι j ; that is, with the submodule of elements having all entries zero
except in slot j.
Fix one of the indices, say ν. Forgetting the ν-th entry in string (mi )iP I gives a string
(mi )iP 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 it 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
where merely the ν-th entry is non-zero), and the Isomorphism Theorem (Theorem 4.12
on page 90) 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.
Exercise 4.10 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
Mi / MJ » Mi ,
iP I jP J iP I z J

and that there is a corresponding isomorphism for direct products:

ź ź ź
Mi / MJ » Mi .
iP I jP J iP I z J

Split submodules, direct sums and idempotent maps

It is well known from linear algebra that every sub-vector space is a direct summand in
the surrounding space; bases of the subspace can be extended to the entire containing

10th January 2021 at 11:32am

Version 4.0 run 30
 direct sums and direct products 95

space. This stands in contrast to submodules of modules over general rings, most
of which are not direct summands. It is therefore of interest to have criteria for a
submodule to be a direct summand of the surrounding module.
(4.25) A synonym for a submodule N Ď M to be a direct summand, is that N lies split in Split submodules
M—this of course means that there is another submodule N 1 so that M » N ‘ N 1 — and (splitt undermoduler)

just as in linear algebra, the submodule N 1 is called a complement to N. Equivalently, Complementary

every element m from M can be unambiguously expressed as a sum m = n + n1 with submodules
(komplementære
n P N and n1 P N; or phrased differently, the two conditions N X N 1 = 0 and N + N 1 = undermoduler)
M are fulfilled. With a slightly sloppy notation, one usually writes M = N ‘ N 1 .
(4.26) When we treated direct products of other rings, the notion 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 decomposition”. 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
endomorphisms
Proposition 4.27 (Idempotents and direct sums) Let M be an A-module. (idempotente
endomorfier)
i) If e is an idempotent endomorphism of M, then M decomposes as the direct sum
M = ker e ‘ im e;
ii) When e is an idempotent endomorphism of M, one has im e = eM and ker e =
(id[ M] ´ e) M;
iii) A submodule N of M lies split if and only if there is an idempotent endomorphism
e : M Ñ M with im e = N.

Note that id[ M] ´ e is idempotent precisely when e is, so the two appear in a completely
symmetric way in the proposition.
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 x = ( x ´ e( x )) + e( x ). Obviously e( x )
lies in im e and x ´ e( x ) lies in the kernel ker e because e is idempotent:

e( x ´ e( x )) = e( x ) ´ e2 ( x ) = e( x ) ´ e( x ) = 0.

On the other hand, im e X ker e = 0 since if x = e(y) lies in ker e, it holds that
0 = e( x ) = e2 (y) = e(y) = x. This takes care of i). As to ii), we only need to add that if
e( x ) = 0, clearly x ´ e( x ) = x so that ker e Ď (id[ M] ´ e) M, the other inclusion already
being taken care of.
Finally we attack iii) and suppose that M = N ‘ N 1 . Let π : M Ñ N and ι : N Ñ M
be respectively the projection and the inclusion map. Then π ˝ ι = id[ N ]. Putting
e = ι ˝ π we find

e2 = (ι ˝ π ) ˝ (ι ˝ π ) = ι ˝ (π ˝ ι) ˝ π = ι ˝ π = e,

10th January 2021 at 11:32am

Version 4.0 run 30
 e
#
M
π /N ι /M

96 modules

and it follows readily that ker e = N and im e = M. o

Examples
(4.12) A principal ideal (n) 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.
(4.13) Cheap but omnipresent examples of non-split submodules are ideals a in domains
A. By Paragraph 4.24 above, any complement of a would be isomorphic to A/a. If a
is a non-zero proper ideal, the quotient A/a contains non-zero elements killed by a,
which is absurd since A was assumed to be a domain.
(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 (0)
and (0) ˆ k are both ideals.
K

Exercises
(4.11) Let M1 and M2 be two submodules of the A-module M whose intersection
vanishes; that is, M1 X M2 = (0). Prove that M1 + M2 is naturally isomorphic with
the direct sum M1 ‘ M2 . Hint: Establish that any m P M1 + M2 can 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 J Ď I not
ř ř
containing ν, the intersection of Mν and jP J M j vanishes; that is, Mν X iP J M j = (0).
ř À
Prove that iP I Mi is isomorphic with the direct sum iP I Mi .
À
(4.13) Assume that for each i P I there is given a submodule Ni Ď Mi . Prove that iP I Ni
À
is a submodule of iP I Mi in a natural way and that there is a natural isomorphism
À À À
iP I Mi /Ni » ( iP I Mi ) / ( iP I Ni ).
(4.14) Generalize Proposition 4.27 in the following way. Assume that e1 , . . . , er are
mutually orthogonal idempotent endomorphisms of the A-module M. Suppose they
ř
satisfy e = id[ M ]. Show that putting Mi = ei ( M ) one obtains a decomposition
À i
M = i Mi . Prove the converse: If such a decomposition exists, exhibit a collection of
idempotents inducing it.
M

Modules over direct products

We aim at describing all modules over a direct product A = A1 ˆ . . . ˆ Ar of a finite
collection tAi u1ďiďr of rings in terms of modules over the factors Ai . The description
is based on a very natural construction: Suppose given an Ai -module Mi for each i.

10th January 2021 at 11:32am

Version 4.0 run 30
 finitely generated modules 97

À
The additive group i Mi has a natural A-module structure; a string a = ( ai )iP I of
ring elements acts on a string m = (mi )iP I of module elements according to the rule
a ¨ m = ( ai ¨ mi ), and once more the axioms come for free, the action being defined
component-wise. We contend that all A-modules are shaped like this.

Proposition 4.28 Let A1 , . . . , Ar be rings and put A = A1 ˆ . . . ˆ Ar . Assume that M is

an A-module. Then there are canonically defined A-submodules Mi of M that are Ai -modules,
À
and are such that M » i Mi .

Proof: The point is that a decomposition of 1 as a sum of orthogonal idempotents

in A induces a decomposition of M. To be precise, let e1 , . . . , er be the idempotents
ei = (0, . . . , 0, 1, 0, . . . , 0) 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. From
ř ř ř
i ei =ř1 we infer that M = i Mi : It holds that x = i xei , and the sum is direct since
if x = i mi ei , it follows readily from the ei ’s being orthogonal that mi ei = xei ; thence
the terms mi ei depend unambiguously on x. o
(4.29) 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 4.28 above tells us that all modules over
A are shaped like direct sums V1 ‘ . . . ‘ Vr with each Vi a vector space over k i .
ś
Exercise 4.15 Extend Proposition 4.28 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 each Mi is an
Ai -module unambiguously associated with M. M

4.3 Finitely generated modules

(4.30) The finitely generated modules form a particularly important class of modules. Finitely generated
For short, they are also called finite modules. As the name indicates, these modules modules
(endeliggenererte
have finite sets of generators; that is, sets of elements m1 , . . . , mr such that each m P M moduler)
ř
can be expressed as a linear combination m = i ai mi of the mi ’s with coefficients ai ’s Finite modules
from A. The generators are by no means unique, and in general there might very well (endelige moduler)

be non-trivial linear relations among them. The subcategory of the category Mod A of
A-modules whose objects are the finitely generated A-modules and morphisms are the
A-linear maps* will be denoted by mod A . ˚ The maps in mod
A
bewteen two objects in
Example 4.15 Several important and natural occurring modules are not finitely gen- mod A are thus the
erated. One example can be the Z-module B = Z[ p´1 ] where p is a natural num- same as those in
ber. For each non-negative integer i consider the submodule Bi = Z ¨ p´i . Because Mod A . One says that
mod A is a full
p´i = p ¨ p´i´1 , it holds true that Bi Ď Bi+1 , and the Bi ’s form an ascending chain subcategory of Mod A
of submodules. Now, every element in B is of the form a ¨ p´n for some n, in other
Ť
words one has B = i Bi . Any finite set of elements from B is contained in BN for

10th January 2021 at 11:32am

Version 4.0 run 30
 98 modules

some N sufficiently large (just take N larger than all exponents of p appearing in the
denominators), so if finitely many elements generated B, it would hold true that BN = B
for some N. This is obviously absurd, as p can appear to any power in the denominator.
K

Cyclic modules
Cyclic modules (4.31) Modules requiring only a single generator are said to be cyclic or monogenic.
(sykliske moduler) Among the ideals the principal ideals are precisely the cyclic ones, and more generally,
Monogenic modules if M is any module and m P M an element*, the submodule A ¨ m = t a ¨ m | a P A u is
(monogene moduler) cyclic.
˚ The zero module is
Now, assume that M is a cyclic A-module and let m P M be a generator. Multiplica-
counted among the
tion induces an A-linear map φ : A Ñ M that sends a to am, and this map is surjective
cyclic ones, so m = 0
is admitted. since m was chosen to be a generator. The kernel of φ 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 90, we arrive at an isomorphism M » A/ Ann M.

Lemma 4.32 A cyclic A-module M is isomorphic to A/ Ann M.

So the cyclic modules are up to isomorphism precisely the quotients A/a of A by ideals
a. Notice that A itself is cyclic corresponding to a = 0. 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 C[ x ]/( x ´ a) with a P C are
all isomorphic to C.
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.

Simple modules
(4.33) The simplest modules one can envisage are the ones without other submodules
Simple modules than the two all modules have—the zero submodule and the module itself—and they
(Simple moduler) are simply called simple: A non-zero* A-module M is said to be simple if it has no
˚ It might appear non-zero proper submodules. Simple modules are cyclic, and any non-zero element
paradoxical that the generates; indeed, if m ‰ 0, the cyclic submodule A ¨ m is non-zero (as m lies in it)
simplest of all modules
is not counted among and hence equals M since M has no proper non-zero submodules. Lemma 4.32 above
the simple ones; the gives that M is of the form A/ Ann M. Moreover, the annihilator ideal Ann M must be
reason is found in the
maximal since if Ann M Ď a, the quotient a/ Ann M is a submodule of M which either
upcoming
Proposition 4.34. equals 0 or M. In the former case Ann M = a, and in the latter it holds that a = A.
Thus simple A-modules are characterized as follows:

10th January 2021 at 11:32am

Version 4.0 run 30
 bases and free modules 99

Proposition 4.34 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.

Exercises
(4.16) Let N be a submodule of M. Show that if N and M/N both are finitely 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.
(4.17) Let N and L be submodules of the A-module M. If N X L and N + L are finitely
generated, show that both N and L are finitely generated.
(4.18) Show that k[ x, x´1 ] is not a finitely generated module over k[ x ].
(4.19) Assume that k is a field. Consider the polynomial ring k[ x ] as a module over
the subring k[ x2 , x3 ]. Prove it is finitely generated by exhibiting a set of generators.
Determine the annihilator of the quotient k[ x ]/k[ x2 , x3 ]. What can you say about
k [ x ]/k[ x2 , x p ] where p is an odd prime?
(4.20) Assume that k is a field. Consider the polynomial ring k[ x ] as a module over
the subring k[ x3 , x7 ]. Prove it is finitely generated by exhibiting a set of generators.
Determine the annihilator of the quotient k [ x ]/k[ x3 , x5 ].
(4.21) Schur’s lemma. Assume that M and N are two simple A-module that are not
isomorphic. Prove that Hom A ( M, N ) = 0. Prove that Hom A ( M, M ) = A/ Ann M.
M

4.4 Bases and free modules

Just like for vector spaces one says that a set of generators tmi uiP I (not necessarily finite)
is a basis for M if every element from M can be written as a (finite) linear combination Bases for modules
(basis for en modul)
ř
of the mi ’s in only one way; that is, the coefficients ai in an expression m = i ai mi are
unambiguously determined by m. Be aware, however, 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 ( x, y) in the polynomial ring
k [ x, y], but do not form a basis since the element xy can be expressed as two different
linear combinations, namely* one has xy = x ¨ y = y ¨ x. And of course, x and y form a ˚ In the first expression

minimal set of generators, one can not do without either, so even minimal generator x is the coefficient and
y the generator, where
sets are not necessarily bases. The natural question then arises: Can the ideal ( x, y) be as in the second it is
generated by one element? The answer is no! A generator would divide both x and y the other way around;
y is the coefficient and
which is absurd.
x the generator.
The gist of this example is that x and y commute, which indicates that the phe-
nomenon is inherent in commutative rings. Any set of generators for an ideal consisting

10th January 2021 at 11:32am

Version 4.0 run 30
100 modules

of at least two elements can never be a basis simply because the generators commute. K
Exercise 4.22 Show that the property, familiar from the theory of vector spaces, that
ř
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

Free modules
(4.35) The lack of bases for most modules leads to a special status of those that have
Free modules (fri one. One says that an A-module F is free if it has a basis. The reason behind the
moduler) suggestive name “free” is that one may freely prescribe values to linear map on the
basis elements—a principle that goes under the name of the Universal Mapping Principle:

Proposition 4.36 (The Universal Mapping Principle) Suppose that a given A-module
F is free and has 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 φ( f i ) = mi .
ř
Proof: Every element x P F is expressible as x = iP I ai f i with coefficients ai from A,
merely finitely many of which are non-zero, and most importantly, the ai ’s are uniquely
ř
determined by x. Hence sending x to iP I ai mi gives a well defined map φ : F Ñ M.
That this yields an A-linear map amounts to the coefficients of a linear combination
being the corresponding linear combination of the coefficients, which ensues from
coefficients being unique. o
(4.37) We return to Example 4.16 on the preceding page about the question when ideals
are free, to give a precise statement:

Proposition 4.38 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 on the previous page 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 relations a ¨ f = 0 ¨ f = 0 give
two representations of 0. The other way around, if the non-zero divisor f is a generator
for a, it is a basis; indeed f being a non-zero divisor it can be cancelled from an equality
like a f = b f . o
Example 4.17 Another kind of non-free modules ubiquitously present in algebra are the
the torsion modules; among them we find the cyclic modules of the form A/a where a is
a non-zero ideal. Since a ¨ 1 = 0 ¨ 1 for any a P a, such a module can not be free: Any
map φ : A/a Ñ M must have image in the submodule (0 : a) M consisting of elements
killed by a which violates the Universal Mapping Principle (Proposition 4.36 above).
If the ring A is a pid, all non-zero ideals will be free modules, but of course there
will still be torsion modules. However, these are, at least among the finitely generated
modules, the ones that prohibits modules from being free, since any finitely generated

10th January 2021 at 11:32am

Version 4.0 run 30
 bases and free modules 101

À
A-module is isomorphic to a finite direct sum i Mi with the Mi ’s either being A or
A/( ai ) for some ai P A. This nice behaviour does not persist for modules that are not
finitely generated, as Exercise 4.27 on page 103 below shows. K
(4.39) 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 = (0, . . . , 0, 1, 0, . . . 0) with the one sitting in slot number i.
There is no reason to confine these considerations to direct sums of finitely many
À
copies of A. For any set I, the direct sum iP I A has a standard basis 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 4.40 Assume that F is a free A-module with basis t f i uiP I . Then there is an
À
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 4.36 above, we may define a map φ : F Ñ iP I A by sending f i
À
to the standard basis vector ei ; conversely, since iP I A is free, sending ei to f i sets up a
À
map ψ : iP I A Ñ F. These two maps are obviously mutual inverses. o

Corollary 4.41 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 module. Rank of a free module
The rank is the sole invariant of free modules; up to isomorphism it determines the (Rangen til en fri
modul)
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 4.40 above we need merely to verify that when two direct
À À
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, and we reduce
the proof to the case that 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 4.13 on page 96). 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 4.18 Free modules with given basis: From time to time itis 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 such a module, denoted
AS , is as the set of maps α : S Ñ A with finite support; that is, the maps such that
α(s) ‰ 0 for at most finitely many members s of S; in symbols

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

10th January 2021 at 11:32am

Version 4.0 run 30
 102 modules

The module structure of AS is given point-wise: (α + α1 )(s) = α(s) + α1 (s) and ( a ¨

α)(s) = aα(s).
In AS there is a collection of function termed generalized Kronecker-δ’s that constitute
a natural basis and which are in a one-to-one correspondence with the set S. For each
member s P S there is one such function δs defined as
$
&0 when s ‰ t
δs (t) =
%1 when t = s .

It is a trivial matter to verify they form a basis. They generate AS because any α can be
ř ř
expressed as α = sPS α(s)δs , and if as δs = 0 is a dependence relation, one just plugs
in any t from S to find that at = 0.
A suggestive way of denoting elements from AS is as linear combinations s as ¨ s of
ř

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 Mapping
1
Principle for free modules, a map φ˚ : AS Ñ AS by sending each basis element δs
1
to the element δφ(s) of AS . In the alternative notation, the map φ˚ takes the form
ř ř
φ˚ ( s as ¨ s) = s as ¨ φ(s). It is pretty obvious that (ψ ˝ φ)˚ = ψ˚ ˝ φ˚ when ψ is
another map composable with φ, so that AS is a covariant functor from the category
Sets to the category Mod A . K
M
Exercise 4.23 Let M be an A-module and let FM = A be the free module with
elements from M as a basis. Then M is a quotient of FM in a canonical way: Define
φ˚
/ FN a map θ M : FM Ñ M by sending δm to m (in the alternative notation it takes the
FM
hypertautological form m ÞÑ m). Show that θ M is surjective and that φ ˝ θ M = θ N ˝ φ˚
θM θN
whenever φ : M Ñ N is an A-linear map. M
 
M /N Exercise 4.24 Show that a finitely generated A-module is the quotient of a finite free
φ
module. 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 describing maps between modules that
are not free, is substantially more complicated, if a good description at all is possible.
(4.42) 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 bases for E and F respectively. When each φ(e j ) 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
˚ The precise notation ř
rank m the matrix is the m ˆ n-matrix given* as M(φ) = ( aij ) where φ(e j ) = i aij f i .
MFE ( φ ), where E and
(4.43) The familiar property that compositions of maps correspond to products of
F designate the two
bases, is cumbersome,
but sometimes helpful
in that the rule for
composition takes the
form 10th January 2021 at 11:32am
MF E (φ) ¨ MF (ψ) =
G Version 4.0 run 30
MGE (φ ˝ ψ)
 bases and free modules 103

matrices still holds true, and the verification is mutatis mutandis the same as for linear
maps between vector spaces (we leave it to students needing to fresh up their knowledge
of linear algebra); that is, if ψ is A-linear map from F to a third free module G (equipped
with a basis), one has
M ( ψ ˝ φ ) = M ( ψ ) ¨ M ( φ ).
Likewise, associating a matrix to a map persists being a linear operation in that

M (αφ + βφ1 ) = αM (φ) + βM(φ1 ),

whenever φ and φ1 are A-linear maps from E to F and α and β are two elements from
the ring A.
Exercises
The next series of exercises, which culiminates with problem 4.27, is aimed at giving an
example that countable products of free modules are not necessarily free.
(4.25) Show that in a free Z-module every element is divisible* by at most finitely ˚ One says that an

many integers. element x of a module

M is divisible by n if
(4.26) Show that the direct sum of countably many copies of Z is countable, whereas the there is a y P M so
direct product of countably many copies is not (it has the cardinality of the continuum). that ny = x.
(4.27) Infinite products are not 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-module. The crucial
ś

point is to show that if P were free, the direct sum iPN Z would be contained in a
À

proper direct summand Q of P. The quotiont P/Q would then be free which is absurd
since it has infinitely divisible elements.
Aiming for a contradiction, we suppose that the product has a basis t f i uiP I . The
direct sum iPN Z lies in P and has the standard basis elements ei (with a one in slot i
À
ř
as sole non-zero component). Each ei can be developed as a finite sum ei = j aij f j in
terms of the basis elements f j with coefficients aij P Z.
a) Prove that I cannot be countable (cfr. Exercise 4.26).
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. Conclude that Q is a proper
À

direct summand in P. Hint: Let j be in J when the coefficient aij ‰ 0 for at least
one i. Observe that for each i it holds that aij ‰ 0 only for finitely many j.
c) For any element x = (n1 , n2 , . . .) in P and any i P N prove that the element
y = (0, 0, . . . , 0, ni , ni+1 , . . .) 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 = (n1 , n2 , . . .) does not lie in Q. Hint: Q is countable.
e) Show that the image of a in P/Q is divisible by infinitely many numbers and
hence P/Q cannot be free.
M

10th January 2021 at 11:32am

Version 4.0 run 30
 104 modules

4.5 Graded modules

À
Graded modules (4.44) Let A = iPZ Ai be a graded ring. A graded module over A is a module whose
(graderte moduler)
À
underlying additive group decomposes as M = iPZ Mi in way compatible with the
action of A on M; that is, the following condition is satisfied

A i M j Ď Mi + j

Homogeneous maps of for all i and j. Note that each homogenous component M j will be a module over A0 .
degree zero (homogene
It turns out to be important to allow elements of negative degree, and as long as the
avbildninger av grad
null) degrees are bounded away from ´8, this does not pose serious problem; we say that
M isbounded from below if Mi = 0 for i ăă 0.
(4.45) As always, a new concept is followed by the concept of the corresponding
“morphisms” preserving the new structure. In the present case a "morphism” between
two graded A-modules M and M1 is an A-homomorphism φ : M Ñ M1 that respects
Isomorphic graded the grading; it sends homogeneous elements to homogenous elements of the same
modules (isomorfe degree. It is common usage to say that such a homomorphism φ is a homogeneous of
graderte moduler)
Homomorphisms of degree zero, or a homomorphism of graded modules, or just a map of graded modules. It may
graded modules be decomposed as a sum φ = i φi where each φi : Mi Ñ Mi1 is an A0 -linear map. And
ř
(homomorfier av
graderte moduler) as usual, two graded modules are isomorphic if there is a homomorphism of graded
modules φ : M Ñ M1 that has an inverse.
(4.46) The composition of two maps of degree zero is obviously of degree zero, as is any
À À
linear combination of two. The three identities ker φ = i ker φi , coker φ = i coker φi
À
and im φ = i im φi are close to trivial to verify, and hence the kernels, cokernels and
imegas of maps of graded modules are graded in a canonical manner. One leisurely
verifies that these are kernels, cokernel and images also in the category of graded A-
modules; moreover the requirement that "the kernel of the cokernel equals the cokernel
of the kernel" is fulfilled (maps have images) so that GrModA is an abelian category. A
sequence in GrModA is exact if and only if it is exact in Mod A.
À
Exercise 4.28 Show that if tMi uiP I is a collection of graded modules, the sum iP I Mi
ś
and the product iP I Mi are graded in a natural way. Show that with this grading they
are the sum and the product also in the graded category GrModA . M

Shift operators
(4.47) There is a collection of shift operators acting on the category of graded S-modules.
For each graded module M and each integer m P Z there is graded module M(m)
associated to a graded module M. The shift do not alter the module structure of M, not
even the sets of homogeneous elements is affected, but they are given new degrees. The
new degrees are defined by setting

M ( m ) d = Mm + d .

10th January 2021 at 11:32am

Version 4.0 run 30
 nakayama’s lemma 105

In other words, one declares the degree of the elements in Mm to be equal to d ´ m.

Any map φ : M Ñ N between two graded modules which is homogeneous of degree
zero, persists being so when the degrees are shifted uniformly in M and N, hence it
induces a map φ[m] : M (m) Ñ N (m). This means that the shifts (m) are functors from
GrModA to itself. Obviously they are exact and of course (m) ˝ (m1 ) = (m + m1 ).
Example 4.19 For instance, when m ą 0, the shifted polynomial ring A(´m) has no
elements of degree d when d ă m, indeed, A(m)d = Ad´m , and the ground field k
sits as the graded piece of degree m. Whereas the twisted algebra A(m) has non-zero
homogeneous elements of degrees down to ´m with the ground field sitting as the
piece of degree ´m. K

R0 R1 R2 R3 R4 R5 R6 R7 R8

4.6 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, and we shall present the ones most frequently met.
One way of viewing this famous result—which is the most basic and in our view
the best, and which we shall adopt as our point of departure—is as an extension of
the fundamental existence result for maximal ideals (Theorem 2.49 on page 49) to
finitely generated modules: Every non-zero finitely generated module has a maximal
proper submodule, or what amounts to the same, every non-zero finitely generated
module has a simple quotient. Indeed, if N is a submodule of M, the quotient M/N is
simple if and only if N is a maximal proper submodule (simply because submodules of
M/N correspond to submodules of M containing N). Notice, that formulated in this
way, Nakayama’s lemma comes for free for the large an important class of Noetherian
modules (which not yet have been defined, but will be in Chapter 9).
One version of Nakayama’s lemma is best proved using a localization technique,
and is therefore postponed until after localization has been treated.
Tadashi Nakayama
Nakayama’s lemma and simple quotients
(1912–1964)
(4.48) Here comes our first version of Nakayama’s lemma: Japanese
mathematician
Proposition 4.49 (Nakayama’s lemma I) Every finitely generated A-module M which is
non-zero, has a non-zero simple quotient. In other words, there exists a maximal ideal m and an
A-linear surjection M Ñ A/m. Equivalently, mM is a proper submodule of M.

10th January 2021 at 11:32am

Version 4.0 run 30
106 modules

Not all modules have simple quotients; to find an example we need look no further than
to the rationals Q considered a Z-module. For any ideal a in Z It holds true that aQ = Q,
and hence there are no non-zero maps Q Ñ Z/a. A constituting property of Noetherian
modules (which we soon come to) is that every non-empty set of submodules has a
maximal member, so in a Noetherian module maximal proper submodules exist almost
by definition.
Proof: Assume first that M is cyclic. It is then of the form A/a for some proper ideal
and thus has A/m as a quotient for any maximal ideal m containing a. If M is not
cyclic, chose generators x1 , . . . , xn for M with n minimal and n ě 2. The submodule N
generated by x2 , . . . , xn is a proper submodule of M. Consequently M/N is non-zero
and cyclic and has a simple quotient by the first part of the proof. o
(4.50) We can not resist giving another argument for M having a maximal proper
submodule tailored to the same pattern as the proof of the Basic Existence Theorem for
ideals (Theorem 2.49 on page 49). If tMi u is an ascending chain of proper submodules
Ť
and M is finitely generated, the union i Mi is a proper submodule; indeed, the finite
number of generators of M would all be contained in an Mi for i large enough and
thence Mi = M, which is not the case. Zorn’s lemma then ensures there is a maximal
proper submodule.

Nakayama classic
(4.51) To assure anyone (hopefully there are none) that finds our approach a blasphe-
mous assault on their most cherished tradition, we surely shall include Nakayama
classic; and here it comes. Recall that the Jacobson radical of A equals the intersection
of all the maximal ideals in A.

Proposition 4.52 (Nakayama classic) Let a be an ideal in A contained in the Jacobson rad-
ical of A. Let M be a finitely generated A-module and assume that aM = M. Then M = 0.

Proof: Assume M ‰ 0. By Nakayama I (Proposition 4.49 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
The by far most common situation when Nakayama’s lemma is applied is when A is
a local ring. The Jacobson radical then equals the maximal ideal m, and, when M is
finitely generated, an equality mM = M implies that M = 0 .
One may rephrase Nakayama’s lemma as follows.

Proposition 4.53 (Nakayama’s lemma II) Assume that M is a finitely generated A-module
and that a is an ideal contained in the Jacobson-radical of A. If M/aM = 0, then M = 0.

10th January 2021 at 11:32am

Version 4.0 run 30
 nakayama’s lemma 107

Other formulations
(4.54) There are several other reformulations of Nakayama’s lemma, and here we offer
a few of the most frequently applied ones.

Proposition 4.55 (Nakayama’s lemma III) Let M is be a finitely generated A-module. As-
sume 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
a ¨ M/N = M/N because any m from M lies in aM modulo elements in N. o

Proposition 4.56 Assume that φ : N Ñ M is A-linear between two A-modules and that M
is finitely generated. Moreover, let a be an ideal contained in the Jacobson radical of A. If the
induced map φ̄ : N/aN Ñ M/aM is surjective, then φ is surjective.

Proof: That φ̄ is surjective means that x P M there is a y P N such that x = φ(y) + z

with z P aM. Hence M = φ( N ) + aM, and we conclude that M = φ( N ) by the previous
proposition. o

Proposition 4.57 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 4.55. o

Exercises
ˇ (4.29) Let Φ be an n ˆ n-matrix with coefficients in a local ring A and denote by Φ
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 does not vanish, then Φ is invertible.
Hint: Φ ¨ Φ: = det Φ ¨ I and det Φ does not belong to the maximal ideal.
ˇ (4.30) Demystifying Nakayama’s lemma. Let A be a local ring with residue class field k.
Assume that φ : E Ñ F is an A-linear map between free modules 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 that Φ is surjective when Φ̄ 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 Nakayama’s lemma for finitely generated
modules over a local ring by using subproblem a). Hint: The key word is

10th January 2021 at 11:32am

Version 4.0 run 30
108 modules

"right sections" of linear maps, if you don’t prefer juggling maximal minors of
n ˆ 8-matrices!
(4.31) Let A be a local ring with residue class field k. Let φ : E Ñ F be a map between
finitely generated free A-modules, and suppose that the induced map φ̄ : E/mE Ñ
F/mF 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 : F Ñ E
corresponding to that minor furnishes a section.
(4.32) Let M an A-module such that mM = M for every 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.
(4.33) Let M be a finitely generated A-module and let φ : M Ñ M be a surjective
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 A[t] with t acting on x P M as t ¨ x = φ( x ). Use the extended version
of Nakayama’s lemma with a = (t) A[t].
(4.34) 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
(4.35) Graded Nakayama. Let M = i Mi be a graded module over the graded ring
À
R = i Ri . Assume that M´i = 0 for i sufficiently big; that is, the degrees of the non-
zero homogeneous 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.
(4.36) 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 D ( f i ) cover Spec A,
and such that each localized module Pf i is a free module over A f i .
(4.37) Let A a ring and let e be a non-trivial idempotent element. Show that the
À
principal ideal I = (e) A 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.
M

4.7 Appendix: The determinant and the characteristic polynomial

Several notions from classical linear algebra, (and from other branches of mathematics,
for that matter) are of a universal nature. They have a meaning whatever the ground
ring is. Their basic properties are universal too; they persist being true over all rings.
The all important determinant is an examples of such a universal creature.
The classical construction of determinants generalizes word for word to A-linear
maps between free modules over any ring A, and all the fundamental properties

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: the determinant and the characteristic polynomial 109

continue to hold; like multiplicativity, alternation in rows and columns and the rules
for expansion along rows or columns. Moreover, the classical proofs still hold water
over general (commutative*) rings. ˚ Except over a few

The classical approach to determinants is a pedestrian’s—there is also a Formula 1 mildly

non-commutative rings
way based on the so-called exterior powers. It has the disadvantage of requiring a where there is a kind of
more advanced machinery and being rather opaque for beginners, but has the great substitute,
determinants need
advantage of being completely functorial. It also opens up for defining determinants of commutative rings
endomorphisms of a wider class of modules than the free ones.
(4.58) Let C = (cij ) be an n ˆ n-matrix with entries from a ring A. Recall that the
determinant det C is defined as the sum
ÿ
det C = sgn(σ )c1 σ(1) ¨ . . . ¨ cn σ(n) (4.2)
σ PSn

where Sn denotes the symmetric groups on n letters and sgn(σ ) is the sign of the
permutation σ.

The adjunction formula and the determinant trick

(4.59) As mentioned at the top of the section, the basic properties of the determinant
hold true over a general ring, and their classical proofs go through mutatis mutandis (the
students are encouraged to brush up their knowledge of linear algebra by rereading or,
better, reconstruction the proofs). In particular, we would like to point out the adjunction Adjunction formula
formula (adjungsjonsformelen)

C ¨ C: = det C ¨ I, (4.3)

valid for a square n ˆ n-matrix C, where C: is the so-called cofactor-matrix of C and I The cofactor-matrix
denotes the n ˆ n identity matrix. The ij-th entry of C: is the sub-determinant of C with (kofaktormatrisen)

the j-th row and i-th column struck out adjusted with the sign (´1)i+ j . The formula
(4.3) follows from (and is in fact equivalent to) the rules for expanding a determinant
along a row.
Contrary to the case of vector spaces it does not suffice that the determinant det C
be non-zero for C to be invertible, the determinant must be invertible in A. In that case i
it ensues from (4.3) that the inverse is given as
c ji
C´1 = (det C )´1 ¨ C: . j

(4.60) The adjunction formula immediately gives that a complex square matrix with
a non-trivial kernel has a vanishing determinant. There is a reformulation adapted to
modules of this arch-classical fact called the determinant trick (so named by Miles Reid
in his book [?]).

10th January 2021 at 11:32am

Version 4.0 run 30
 110 modules

Let M be an A-module and C = (cij ) an n ˆ n-matrix with entries from A. In

the same way as C induces an endomorphism of the free module nA, it induces an
endomorphism of the iterated direct sum nM; if m = (mi ) P nM, we just let C ¨ m be
ř
the n-tuple ( j cij m j )i .

Lemma 4.61 (The determinant trick) Let C be an n ˆ n-matrix with entries in the ring
A, and let M be an A-module. Assume that the module M has generators m1 , . . . , mn such that
C ¨ (m1 , . . . , mn ) = 0. Then the determinant det C kills M.

Proof: Consider the A-linear map ι : A Ñ nM that sends x to ( x ¨ m1 , . . . , x ¨ mn ). The

ι / nM hypothesis of the lemma translates into the relation C ˝ ι = 0 (where we confuse the
A
matrix C with the map it induces), and citing the adjunction formula (4.3) we find
C

det C¨id[ A] nM det C ¨ ι = C: ˝ C ˝ ι = 0.

C: This means that (det C ¨ m1 , . . . , det C ¨ mn ) = det C ¨ ι(1) = 0, and hence the determinant
 
A / nM det C kills M as the mi ’s generate M. o
ι

The characteristic polynomial

(4.62) An endomorphism γ of a finitely generated free A-module E has a canonically
defined determinant. Indeed, let C and C1 be the matrices of γ in two bases for E. If D
denotes the base-change matrix, it holds true that C1 = DCD´1 , and consequently

det C1 = det D ¨ det C ¨ det D´1 = det C,

which makes det γ = det C a legitimate definition; the determinant of a matrix repre-
senting γ is independent of which basis is used.
The characterstic This opens the way for the definition of the characteristic polynomial of an endomor-
polynomial of an phism γ, namely as Pγ (t) = det(t ¨ id[ E] ´ γ). It is an element of the polynomial ring
endomorphism (det
karakteristiske A [ t ].
polynomet til en For any matrix C = (cij ) with entries in a ring A and any ring homomorphism
endomorfi)

φ : A Ñ B we let φ(C ) = φ(cij ) accepting a slight ambiguity in the notation. The
canonical extension of φ to a map A[t] Ñ B[t] between the polynomial rings will as
well be denoted by φ; that is, φ( ai ti ) = φ( ai )ti . The following lemma is an almost
ř ř

trivial observation:


Lemma 4.63 It holds true that φ PC (t) = Pφ(C) (t) where φ : A Ñ B is any ring homomor-
phism and C is any square matrix with entries from A.

Proof: The determinant is a polynomial in the entries of the matrix, hence it holds true
that det φ( D ) = φ(det D ) for all ring homomorphisms φ and all square matrices D with

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: the determinant and the characteristic polynomial 111

entries in the source of φ. We infer that

φ PC (t) = φ(det(t ¨ I ´ C ) = det(t ¨ I ´ φ C ) = Pφ(C) (t).

The Cayley–Hamilton theorem and the generic matrix

The Cayley-Hamilton theorem is one of the subtler results from elementary linear
algebra. It seems that Frobenius was the first to give a proof in some generality, but
much earlier Cayley and Hamilton did the 2 ˆ 2- and 3 ˆ 3-cases, which turned out to
be sufficient for the theorem to be named after them.
(4.64) The statement involves the characteristic polynomial of a matrix square matrix
C = (cij ) is any with entries cij from any (commutative) ring A. Recall that it is given
as PC (t) = det(t ¨ I ´ C ) where t is a variable and I the identity matrix of the same size
as C. The general Cayley–Hamilton theorem reads as follows.

Theorem 4.65 (General Cayley–Hamilton) Let A be any (commutative) ring and C a Ferdinand Georg
square matrix with entries from A. Then C satisfies its characteristic polynomial; in precise Frobenius (1849–1917)

terms, if PC (t) = det(t ¨ I ´ C ) denotes the characteristic polynomial of C, then PC (C ) = 0. German mathematician

The characteristic polynomial has the virtue of being canonically associated with the
matrix C, and thus it depends functorially on C, in contrast to any arbitrary polynomial
equation satisfied by C.
Example 4.20 It might be instructive to consider the special case when A = k is a field.
If v P kn is an eigenvector of C corresponding to the eigenvalue λ in k, then obviously
PC (C )v = 0 since t ´ λ is a factor of PC (t). Therefore the Cayley–Hamilton theorem
follows whenever kn has a basis of eigenvectors of C. In particular it holds true if C has
n distinct eigenvalues; e. g. if PC (t) has n distinct roots.
It will even suffice that for some field extension K of k, the space K n has a basis of
eigenvector for C (for instance, this will be the case if PC (t) has distinct roots in a field
extension k). Matrices for which this holds, are said to be semi-simple. K Semi-simple matrices
(4.66) There are of course mountains of proofs for such a central result in elementary (semisimple matriser)

linear algebra as is the Cayley–Hamilton theorem. The proof we shall offer is very
simple. With some knowledge of rudimentary field theory and of polynomials over
ufd’s, one will find that it almost reduces to bare common sense*. Moreover, it is a low ˚ But remember

hanging illustration of techniques—specialisation and generalisation— frequently used Einstein’s words:

According to common
in modern mathematics. sense, the earth is flat.

The universal n ˆ n-matrix

(4.67) The salient point of the proof we present is that there is a universal n ˆ n-matrix The universal matrix
Cn = ( xij ) with entries in a ring Rn , and it is universal in the sense that every other (den universelle
matrisen)

10th January 2021 at 11:32am

Version 4.0 run 30
 112 modules

n ˆ n-matrix C = (cij ) with entries in any ring A is obtained as C = (φ( xij )) for an
unambiguously defined ring homomorphism φ : Rn Ñ A. There is no hocus-pocus
about this; the ring Rn will simply be the polynomial ring Rn = Z[ xij |1 ď i, j ď n]
where the xij ’s are variables double indexed as the entries in a matrix are, and of course,
the universal matrix will be Cn = ( xij ). Clearly any n ˆ n-matrix C = (cij ) with entries
in any ring A is of the announced form (φ( xij )); just let φ : Rn Ñ A be defined by
the assignments xij ÞÑ cij . The characteristic polynomial depends functorially on the
matrix (Lemma 4.63 on page 110) and we infer that PC (t) = Pφ(Cn ) (t) = φ( PCn (t)), and
consequently it suffices to verify the Cayley–Hamilton theorem for the single matrix Cn :

Lemma 4.68 If PCn (Cn ) = 0, then PC (C ) = 0 for all n ˆ n-matrices C.

(4.69) One may consider the following theorem as the ultimate formulation the Cayley
Hamilton theorem; anyhow, in view of the above, it implies Cayley–Hamilton as
formulated in Theorem 4.65:

Theorem 4.70 The universal matrix Cn is semi-simple. Hence PCn (Cn ) = 0.

Proof: Let K = Q( xij |1 ď i, j ď n) be the fraction field of Rn . Aiming at an absurdity,

we assume that the characteristic polynomial P(t) = PCn (t) has a multiple root in
some extension L of K. The ring Rn is a udf, and Exercise 3.9 on page 78 yields that
P(t) and its derivative P1 (t) have a common factor Q(t) in Rn [t], which is a monic
polynomial since P(t) is (the units in Rn are just ˘1). Hence for any φ : Rn Ñ A, the




two polynomials φ P(t) and φ P1 (t) have the nontrivial common factor φ Q(t) ,


but φ( P(t)) being the characteristic polynomial of φ(C ) and φ P(t)1 its derivative, it
follows that φ( P(t)) has a multiple root. Consequently, no matrix at all can have distinct
eigenvalues, which is utterly absurd (matrices with distinct eigenvalues exist!). Hence
Cn is semi-simple. o

An epilogue
A technique that permeates modern algebraic geometry is to try to represent functors;
i. e. to find a universal object of the kind one is interested in, which dictates the behavior
of all the crowd. As illustrated above, knowing properties of a universal object can have
strong implications. That universal objects exist, is however by no means always true,
and lot of activity has gone into trying to give criteria fo existence and into coping with
situations where substitutes, almost universal object, may be found. What we did in
this sections, is very simple case indeed, but can serve as an leisurely introduction to
the formalism.
˚ This denotes the (4.71) Consider the functor Mn : Rings Ñ NCRings* that send a (commutative) ring A to
category of not the ring of n ˆ n-matrices with entries in A, and whose action on a ring homomorphism
necessarily

commoutative rings φ is the map that sends M = ( aij ) to φ( M ) = φ( aij ) . Since matrix products and sums

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: unions of submodules and direct limits 113

are preserved, Mn takes values in NCRings. This functor is as one says, representable Representable functors
and it is represented by the universal matrix, which means there is an isomorphism of (representerbare
funktorer)
functors
»
HomRings ( Rn , ´) ÝÑ Mn (´).

Well, it assigns φ( Mn ) to the homomorphism φ : Rn Ñ A; and we checked it is bijective

for each A. It being functorial boils down to the obvious formula ψ(φ( M )) = (ψ ˝
φ)( M ); or more precisely, to the following diagram being commutative for every ring
homomorphism ψ : A Ñ B:

HomRings ( Rn , A) / Mn ( A)

ψ˚ ψ (´)
 
HomRings ( Rn , B) / Mn ( B)

The horizontal arrows are the ones above, and the two vertical ones acts respectively as Functorial maps
φ ÞÑ ψ ˝ φ and M ÞÑ ψ( M ). (funktorielle
avbildninger)
(4.72) Natural transformations In a more general setting, if F, G : C Ñ D are two
functors, a functorial map or a map of functors, also called a natural transformation, from F Natural
to G is just a collection of maps θ A (i. e. arrows in D)—one for each object A in C—such transformations
(naturlige
that the diagram transformasjoner)

F ( A)
θA
/ G ( A)

F (φ) G (φ)
 
F ( B) / G ( B)
θB

commutes for each arrow φ : A Ñ B in C. And of course, a natural transformation is

called an isomorphism, or a natural equivalence as one prefers to say, if there is another Natural equivalences
from G to F such that the two compositions equal the identity. (naturlige ekvivalenser)

4.8 Appendix: Unions of submodules and direct limits

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 is an additive issue (unions of submodules are obvious closed under multiplication
by ring elements) and concerns most abelian groups. In Paragraph 2.31 on page 41 we
argued that the union of two subgroups of an abelian group, neither contained in the
other, is not a subgroup.
(4.73) There is however a natural condition that ensures the union to be a submodule.
One says that the collection is directed if for any two members there is a third containing Directed collection of
submodules (rettede
samlinger av
undermoduler)

10th January 2021 at 11:32am

Version 4.0 run 30
 114 modules

both; that is, for any pair Mi and M j from the collection tMi uiP I there should be an
Ť
index k so that Mi Ď Mk and M j Ď Mk . The union iP I Mi will then be closed under
addition (and as multiplication poses no problem will be a submodule); indeed, let x
and y be two members of the union. This means that there are indices i and j so that
x P Mi and y P M j , and since the collection is 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 their sum does as well. So
the sum belongs to the union. We have proven:

Proposition 4.74 Let tMi uiP I be a directed collection of submodules of the A-module M. Then
Ť
the union iP I Mi is a submodule.

Direct limits
Direct limits (direkte (4.75) There is a general construct called the direct limit inspired by the above reasoning
grenser) It permits us to form “the limit” of certain “directed systems” of modules. The direct
limit is a vast generalization, but under certain circumstances it resembles the "union".
As an illustration, imagine a chain of sets Si indexed by the natural numbers N such
that each Si is a subset of the succeeding set Si+1 . They form an ascending chain which
may be displayed as
S1 Ď . . . Ď S i Ď S i + 1 Ď . . . .

In the traditional set theory there is no way of defining the union of the Si ’s unless they
all are subsets of given set. The introduction of the direct limit of the Si ’s remedies this,
and the direct limit fills the role as their union. But remember, this is just a motivating
example; the direct limit is a much more general construct and can be quite subtle. The
index set can be any ordered set (with some conditions, though) and the inclusions21
may be replaced by any maps (with some compatibility conditions).
One may state the definition of the direct limit in any category, and of course, it
is expressed by way of a universal property. To fix the ideas we shall only work with
modules over a ring A. However, what we shall do is easily translated into several other
categories including Sets, Rings and Alg A .
(4.76) The key notion is that of a directed systems of modules. Such a system has two
ingredients. The first is a collection tMi uiP I of modules over A. The index set I is
Directed orderings supposed to be a partially ordered set whose ordering is directed: for any two indices i
(direkte ordninger) and j there is a third larger than both; i. e. there is a k P I such that k ě i and k ě j. The
second ingredient is a collection of A-linear maps φij : M j Ñ Mi , one for each pair (i, j)
φik of elements from I so that i ě j, which are subjected to the following two conditions:
"
Mk / Mj / Mi o φij ˝ φjk = φik whenever k ď j ď i;
φjk φij

o φii = id Mi .

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: unions of submodules and direct limits 115

The system will be denoted ( Mi , φij ) I . If you would prefer working in a general category
C, just replace the words “A-module” with “object” and A-linear by “arrow”. We
Direct limits (direkte
proceed to describe the universal property that constitutes the direct limit. The direct grenser)
limit of the system ( Mi , φij ) is an A-module lim Mi together with a collection of A-linear
ÝÑ
maps Mj φj
φi : Mi Ñ lim Mi )
ÝÑ
φij
5 lim M
that satisfy φi ˝ φij = φj , and which are universal with respect to this. In other words, ÝÑ i

for any given collection tNi uiP I of A-modules and any given system of A-linear maps Mi
φi

ψi : Mi Ñ N

such that ψi ¨ φij = ψj , there is a unique map η : lim Mi Ñ N such that ψi = φi ˝ η. φi

/ lim Mi
ÝÑ Mi
(4.77) Even though the definition of the direct limit may be formulated in any category, ÝÑ
η
whether it exists or not is quit another question. Every direct system possessing a limit,
# 
ψi
is an exclusive quality most categories do not enjoy. However, the category of modules N
is among the privileged ones.
The main ideas of the construction are quit transparent, but as most proofs of this
type it includes a tiresome list of more or less trivial verifications, whose details we
gladly skip. However, to get the mechanism of the limit under the skin, the students
are urged to do that work.

Proposition 4.78 Let A be any ring. Every directed system ( Mi , φij ) I of modules over A has
a direct limit, which is unique up to a unique isomorphism.
Ť
Proof: We begin with introducing an equivalence relation on the disjoint union i Mi .
Loosely phrased, two elements are to be equivalent if they become equal somewhere M
H nV
out in the hierarchy of the Mi ’s. In precise terms, x P Mi and y P M j are defined to be
equivalent when there is an index k dominating both i and j such that x and y map to
the same element in Mk ; that is, φki ( x ) = φkj (y). We shall write x „ y to indicate that x Ml MH mV
I U
and y are equivalent.
Obviously this relation is symmetric, since φii = id Mi it is reflexive, and it being
transitive ensues from the system being directed: Assume that x „ y and y „ z, with Mi Mj Mk
x, y and z sitting in respectively Mi , M j and Mk . This means that there are indices l
dominating i and j, and m dominating j and k so that the two equalities φli ( x ) = φlj (y)
and φmj (y) = φmk (z) hold true. Because the system is directed, there is an index n
larger than both l and m, and by the first requirement above, we find

φni ( x ) = φnl (φli ( x )) = φnl (φlj (y)) = φnm (φmj (y)) = φnm (φmk (z)) = φnk (z)
Ť
and so x „ z. The underlying set of the A-module lim Mi is the quotient i Mi / „ and
ÝÑ
the maps φi are the ones induced by the inclusions of the Mi ’s in the disjoint union.

10th January 2021 at 11:32am

Version 4.0 run 30
116 modules

The rest of the proof consists of putting an A-module structure on lim Mi and
ÝÑ
checking the universal property. To this end, the salient observation is that any two
elements [ x ] and [y] in the limit may be represented by elements x and y from the same
Mk ; Indeed, if x P Mi and y in M j , chose a k that dominates both i and j and replace
x and y by their images in Mk . Thence forming linear combinations is possible by the
formula a[ x ] + b[y] = [ ax + by] where the last combination is formed in any Mk where
both x and y live; this is independent of the particular k used (the system is directed,
and the φij ’s are A-linear). The module axioms follow suit since any equality involving
a finite number of elements from the limit may be checked in an Mk where all involved
elements have representatives.
Finally, checking the universal property is straightforward: The obvious map from
Ť
the disjoint union i Mi into N induced by the ψi ’s is compatible with the equiva-
lence relation and hence passes to the quotient; that is, it gives the searched for map
η : lim Mi Ñ N. o
ÝÑ
(4.79) Apart from the universal property—which should be the favoured tool for
anybody working with direct limits—there are two principles one should have in
mind. Firstly, every element in lim Mi is induced from an element x P M j for some
ÝÑ
index j; that is, it is of the form φj ( x )—in fact every finite collection of elements may be
represented by elements from a common M j — and secondly an element x P M j maps
to zero in the limit if and only if it maps to zero in an Mi for some i ě j.

Proposition 4.80 (Working principles) With the notation above, the following two state-
ments hold true:
i) Every element in lim Mi is of the form φj ( x ) for some j and some x P M j .
ÝÑ
ii) An element x P M j maps to zero in lim Mi if and only if φij ( x ) = 0 for som i ě j.
ÝÑ
Proof: Clear. o
Example 4.21 Let p be any non-zero integer. Consider the directed system indexed by
the natural numbers N for which Mi = Z and φji ( x ) = p j´i x when j ě i; to give it a
name, let us denote it by (Z, p j´i ). It may depicted as
p p p p p
Z Z ...... Z Z ...

where the drawn maps are the φi+1,i ’s and each of them is multiplication by p. The maps
φji are just the appropriate compositions of j ´ i consecutive maps from the sequence.
We contend that there is a natural isomorphism lim(Z, p j´i ) » Z[1/p]. Indeed,
ÝÑ
define ψi : Z Ñ Z[1/p] by ψi ( x ) = p´i x. Then clearly φij ˝ ψj = ψi and so by the
universal property of direct limits there is induced a map ψ : lim(Z, p j´i ) Ñ Z[1/p]
ÝÑ
that satisfies ψ ˝ φi = ψi for all i. This map is surjective: each element in Z[1/p] is
shaped like ap´i with a P Z and hence ψ([ a]) = ψ(φi ( a)) = ψi ( a) = ap´i . And it is

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: unions of submodules and direct limits 117

injective: assume that ψ([ a]) = 0 and chose an i so that [ a] = φi ( a). It follows that
0 = ψ(φi ( a)) = ψi ( a), but then a = 0 as ψi is injective. K

Exercises
(4.38) Verify the two principles in Pargraph 4.79
(4.39) Let A be a ring. Convince yourself that direct limits exist unconditionally in the
category Alg A of A-algebras.
(4.40) Show that any finite directed set has a largest element. What will a direct limit
indexed by such an ordered set be?
(4.41) Let M = ( Mi , φij ) and N = ( Ni , ψij ) be two directed systems indexed by the same
partial ordered set I. A morphism α from M to N is a sequence of maps αi : Mi Ñ Ni
commuting with the system maps; that is, αi ˝ φij = ψij ˝ α j for all pairs i and j with
i ě j.
a) Show that this makes the set of directed systems of A-modules indexed by I into
a category DirMod A,I .
b) Show that α induces a map r α : lim M Ñ lim Ni so that α ˝ φi = ψi ˝ αi , and this
ÝÑ ÝÑ
construction is fuctorial.
c) Show that (ker αi , φij |ker αi ) is directed system and that the ψij induce maps
ψ̄ij : coker α j Ñ coker αi that makes (coker αi , ψ̄ij ) a directed system Show that
(ker αi , φij |ker αi ) is directed system and that the maps ψij pass to the quotients
and induce maps ψ̄ij : coker α j Ñ coker αi that makes (coker αi , ψ̄ij ) a directed
system
d) Show that the category of directed system DirMod A,I is an abelian category.
e) Show that the direct limit of an exact sequence in DirMod A,I remains exact.
(4.42) Let p be a prime. For i, j P N let Mi = F p [ x ] and let φij be the map given
by φij ( a) = a(i´ j) p . Show that this will be a directed system, and that the limit is
isomorphic to F p [ x1/p ].
(4.43) Consider the set N of natural numbers equipped with the divisibility order; that
is i ě j if and only if j|i. Prove that is order is directed. Consider the system ( Mi , φij )
indexed by N where Mi = Z for all i and φij ( a) = ij´1 a. Check that this is a direct
system and show that its limit is isomorphic to Q.
(4.44) Let A is pid with fraction field K and let p P A be an irreducible element.
Consider the directed system ( A, p j´i ) indexed by N. Show that lim( A, p j´i ) = A[1/p].
ÝÑ
Consider the system ( A/pi A, ψij ) where ψij is the natural inclusion ψij : A/pi A ãÑ
A/pi+1 A that sends x to px. Show that lim( A/pi A, ψij ) is isomorphic to A[1/p]/A.
ÝÑ
M

10th January 2021 at 11:32am

Version 4.0 run 30
 118 modules

Inverse limits
Most concept, if not all, in category theory has a dual counterpart, so also with the
Inverse limit (invers direct limits.The dual notion in this case, is the notion of an inverse limit. The staging
grense) is as for direct limits, but arrows are reversed. We are given an ordered set I whose
ordering being directed, and for each i P I an A-module Mi . Moreover, for every pair i,
j from I with i ď jwe are given φij : M j Ñ Mi which comply with the conditions:

o φij ˝ φjk = φik ;

o φii = id Mi .

The inverse limit lim Mi is a module together with maps φi : lim Mi Ñ Mi that satisfy
ÐÝ ÐÝ
the constraints φj = φi ˝ φij and which is universal in this regard. That is, for any other
module N together with maps ψi : N Ñ Mi such that ψi = φij ˝ ψi there is a unique
η : N Ñ lim Mi such that ψi = φi ˝ η.
ÐÝ
Proposition 4.81 Every directed inverse system has a limit.
ś
Proof: Consider the product i Mi and define a submodule by

L = t ( xi ) | xi = φij ( x j ) for all pairs i, j with i ď j u,

The projections induce maps to φi L : Ñ Mi , and we claim that N together with these
maps constitute the inverse image of the system. A family of maps ψi : N Ñ Mi defines
ś
map η : N Ñ i Mi by x ÞÑ (ψi ( x )) which takes values in L when the ψi ’s satisfying
the compatibility constraints ψi = φij ˝ ψj . It is clearly unique, and that is o
Example 4.22 p-adic integers: One of the most famuos inverse systems is the one of the
p-adic integers. So let p be a prime number. The modules Mi = Z/pi Z and the map
φij is just the canonical reduction map Z/p j Z Ñ Z/pi Z that send a class [ x ] p j mod p j
to the class [ x ] pi mod pi . It may be illustrated by the sequence of reduction maps

... Z/pi+1 Z Z/pi Z ... Z/p2 Z Z/pZ

where each map is the canonical reduction; i. e. φi+1,i and the other maps φij from the
system are just compositions of j ´ i consecutive such.
The inverse limit is denoted by Z p and is called the ring of p-adic integers. They are
extraordinary important rings, almost as fundamental as Z itself. K
(4.82) Let S be the set S = t0, 1, . . . , p ´ 1u of representatives of Z/pZ. Recall that each
element x in Z/pi+1 Z can be written as x = a0 + a1 p + . . . + ai pi where the ai P S and,
strictly speaking, p stands for the class of p. The ai ’s are uniquely determined, and the
pi
image in Z/pi Z is found just by chopping odd the last term ai .

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: unions of submodules and direct limits 119

Each element x of Z p is a string x = ( xi ) P i Z/pi Z that satisfies xi+1 ” xi

ś

mod pi . Writing xi+1 = a0 + a1 p + . . . + ai pi + ai+1 pi this means that xi = a0 + a1 p +

. . . + ai´1 pi´1 , And an infinite string ( a0 , . . . , ai , . . . . . .) of elements from S appears. It is
unambiguously determined by x and there are no restrictions ob the ai ’s save belongig
to S. A suggestive notation is to write x as the power series

x = a0 + a1 p + . . . + a i p i + . . .

Proposition 4.83 The elements from Z p are of the form z = a0 + a1 p + a2 p2 + . . . . . .

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 5

Exact sequences

One of the founders of homological algebra, Saunders Mac Lane, once referred to the
subject as "General abstract nonsense", a term that many may find offensive. However, it
has no pejorative connotation, but is rather a light-hearted way to warn the readers that
arguments are of a very abstract nature far from the specific context—often formulated
in the vernacular of homological algebra or category theory. Notions or arguments
deserving this honorary title are ubiquitous; they are found all over mathematics—hence
their general nature and importance.
We shall in the present chapter lightly touch upon the important notion of a complex,
but most of the chapter will be about the special case of short exact sequences—playing
with short exact sequences is a formal pathfinders game, and before seeing it applied
and experiencing the force of the method, one may find the nickname "General abstract
nonsense" appropriate.

5.1 Exact sequences

labelsection ::moduler::EksakteSekvenserLet φ and φ be two composable A-linear maps
and display them as a sequence

/N / L.
φ ψ
M

This sequence is said two be exact if ker ψ = im ψ, which in particular implies that Exact sequences
ψ ˝ φ = 0. It frequently happens that such a sequence is part of a longer sequence of (eksakte følger)

maps, extending to the left or to the right, and the extended sequence is then said to
be exact at N as well. If the composition of any two consecutive maps in the extended
sequence equals zero, the sequence is called a complex; we shall come back to those later Complex (kompleks)
 122 exact sequences

in the chapter. A sequence exact at all places, is simply said to be exact.

(5.1) 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 (0), exactness boils down to ψ
being injective. Similarly, when L = 0, the sequence is shaped like

/N / 0,
φ ψ
M

and it is exact if and only if φ being surjective.

Example 5.1 Every A-linear map α : M Ñ N lives in the exact sequence

0 / ker α /M /N / coker α / 0.

K
Short exact sequences
By far the most often met exact sequences, are the so-called short exact sequences; they
are the easiest to handle and a long exact sequences can be split into a sequence of short
ones.
They are a valuable tool for several purposes; for instance, when one tries to study a
module by breaking it down into smaller (and presumptive simpler) pieces.
(5.2) A three-term sequence (or a five-term sequence if you count the zeros)

/ M1 /M / M2 /0
α β
0 (5.1)

Short exact sequences is called a short exact sequence when it is exact. This means that α is injective, that β
(korteksakte følger) surjective and that im α = ker β. Of course, the term “short” in the name 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 90) that there
is a unique isomorphism θ : M2 » M/α( M1 ) shaped in a way that β corresponds to the
quotient map. In other words, θ enters into the following commutative diagram

α β
0 M1 M M2 0
α| M1 θ (5.2)
0 α ( M1 ) M M/α( M1 ) 0

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

0 N M M/N 0,

10th January 2021 at 11:32am

Version 4.0 run 30
 exact sequences 123

where N Ď M is a submodule, and the two maps are respectively the inclusion and the
quotient map.

Examples
(5.2) Direct sums: The direct sum M ‘ N of two A-modules fits naturally into the short
exact sequence
0 N N‘M M 0 (5.3)
where the left-hand map is the natural inclusion sending x to ( x, 0) and the one to the
right is the projection onto M, which maps ( x, y) to y. In particular, if N and N 1 are two
submodules of a module M, then there is a short exact sequence
ι σ
0 N X N1 N ‘ N1 N + N1 0

where ι is the “diagonal inclusion map” ι( x ) = ( x, x ) and σ ( x, y) = x ´ y.

(5.3) A Chinese squence: The Chinese Remainder Theorem (Theorem 2.72 on page 57) for
two ideals may be generalized by saying that the sequence

0 aXb A A/a ‘ A/b A/a + b 0 (5.4)

is exact where the two maps in the middle are given by the assignments x ÞÑ ([ x ]a , [ x ]b )
and ([ x ]a , [y]b ) ÞÑ [ x ]a+b ´ [y]a+b . Having four non-zero terms it is to long to be called
short exact, but it may be obtained by splicing together 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.
K

Exercises
(5.1) Let A be a ufd and x and y two elements. Let a be the ideal a = ( x, y). Show
that the sequence
ι π
0 A A‘A a 0

where ι( a) = (ya, ´xa) and π ( a, b) = ax + by, is exact if and only if x and y are without
common factors.
(5.2) Let p be prime. Show that for every pair of natural numbers n and m there is a
short exact sequence of abelian groups

0 / Z/pm Z / Z/pn+m Z / Z/pn Z / 0.

10th January 2021 at 11:32am

Version 4.0 run 30
 124 exact sequences

(5.3) Verify that the two maps defined in the Example 5.3 above are well defined and
that the sequence is exact. Deduce the Chinese Remainder Theorem from it.
(5.4) Write down a “Chinese sequence” involving three ideals that generalizes the
sequence (5.4) above. Prove it is exact and deduce the Chinese Remainder Theorem for
three ideals. Hint: The sequence 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 (5.5)

Split exact sequences is split exact when being isomorphic to the standard sequence (5.3) above. This not only
(Splitteksakte følger) requires that M be isomorphic to the direct sum 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 (5.6)
»
0 / M1 / M1 ‘ M2 / M2 / 0,

where the maps in the bottom sequence are the projection and the inclusion.
(5.3) Of course, all sequences are not split exact, and even if two short exact sequences
have the same two extreme modules, the middle modules need not be isomorphic. The
easiest example is found among finite abelian groups: Both Z/p2 Z and Z/pZ ‘ Z/pZ
appear in the midst of short exact sequences with both extreme modules being Z/pZ.
Right amd left sections
(høyre- og In general, it is an unsurmountable challenge to classify all possible middle modules
venstreseksjoner) given the two extreme ones.
N (5.4) There is a nice criterion for a short exact sequence to be split involving only one of
σ the maps α or β; to formulate it we need two new concepts. Let γ : M Ñ N be A-linear.
~ An A-linear map σ : N Ñ M is said to be a right section for γ if γ ˝ σ = id N , and it is
M /N
γ called a left section if σ ˝ γ = id M . A map having a right section will be surjective and
N` one with a left section will be injective.
σ
Proposition 5.5 (Splitting criterion) Let the short exact sequence
N /M
/ M1 /M / M2 /0
α β
γ 0

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

10th January 2021 at 11:32am

Version 4.0 run 30
 exact sequences 125

i) The sequence is split;

ii) The map α has a left section;
iii) The map β has a right section.

Proof: i) ùñ ii) and i) ùñ iii): If M = M1 ‘ M2 , the canonical inclusion of M2 into M

is a right section for the projection onto M2 , and dually, the projection onto M1 is a left
section for the inclusion of M1 into M.
iii) ùñ i): Let σ be a right section of β so that β ˝ σ = id M2 . We have a very good
candidate for a submodule of M which is a complement of α( M1 ), namely the image
σ ( M2 ). Our task is thus to verify that their intersection vanishes and that their sum
equals M. Since α( M1 ) = ker β, it holds that α( M1 ) X σ ( M2 ) = 0; indeed, if x P M2 and
σ ( x )belongs to the intersection, x = β(σ ( x )) = 0. To see that σ( M) and α( M1 ) span M,
consider the obvious equality

x = ( x ´ σ ( β( x ))) + σ( β( x )).

Clearly the last term in the sum σ ( β( x )) belongs to σ ( M2 ); moreover because β ˝ σ =

id M2 , it holds true that β( x ´ σ ( β( x ))) = β( x ) ´ β( x ) = 0, so that the first term
x ´ σ( β( x )) lies in α( M1 ) (which equals ker β).
The association ( x, y) ÞÑ α( x ) + σ (y) yields an isomorphism γ : M1 ‘ M2 Ñ M
entering into the commutative diagram:

0 M1 M1 ‘ M2 M2 0
γ
α β
0 M1 M M2 0.

Before proceeding with the rest of the proof, we mention that the idempotent endomor-
phism of M responsible for the decomposition, is the map e = σ ˝ β (which we in fact
encountered in the proof).
ii) ùñ i): Let τ be a left section of α so that τ ˝ α = id M1 . The idempotent endomorphism
of M giving rise to the decompostion is in this case e = α ˝ τ; indeed, e is idempotent:
one finds
e2 = (α ˝ τ ) ˝ (α ˝ τ ) = α ˝ (τ ˝ α) ˝ τ = α ˝ id M1 ˝τ = e.

It follows that M decomposes as M = eM ‘ (id M ´e) M, and one verifies that α( M1 ) =

eM (obvious) and that β maps (id M ´e) M isomorphically onto M2 . Tis ensues from
the equality β ˝ (id M ´α ˝ τ ) = β. Indeed, it holds that

ker β X (id M ´e) M = α( M1 ) X (id M ´e) M = e( M) X (id M ´e) M = 0

and obviously β maps (id M ´e) M surjectively onto M2 .

10th January 2021 at 11:32am

Version 4.0 run 30
 126 exact sequences

If γ : M Ñ M1 ‘ M2 is defined by the assignment x ÞÑ α´1 (e( x )), x ´ e( x ) , it

˚ To be entirely correct,
one should write enters into the digram (5.6), and that ends the affair. o
σ + ι ˝ γ where ι
denotes the inclusion of Exercise 5.5 Ambiguity of summands. The two submodules M1 and M2 of M occurring
ker β in M in a decomposition M = M1 ‘ M2 are seldom unique, e. g. just remember that a vector
spaces has many different bases. And in fact, not even their isomorphism classes are
determined; but examples of that will be for later (see Examples 5.8 on page 133 and
8.19 on page 227). Neither are sections unique, and this exercise examines this issue.
˚ To make things
Let β : M Ñ N be A-linear and let σ be a right section.
simple, we contend
ourselves to module a) Show that for any A-linear γ : N Ñ ker β the map σ + γ is another section*;
categories. There is b) Show that σ is unique if and only if Hom A ( N, ker β) = 0;
however a notion of
c) Assume that M1 and M2 are two complementary split submodules of M. Show
additive categories
(where hom-sets are that if they are non-isomorphic simple modules, they are unique as submodules.
abelian groups and
M
compositions bilinear),
and between such
categories the term Additive functors and direct sums
additive functor is An additive functor F : Mod A Ñ ModB between the categories of modules* over two rings
meaningful.
A and B is one that takes sums of maps to sums of maps; that is, F (φ + ψ) = F (φ) + Φ(ψ)
Additive functors for every pair of A-linear maps between to A-modules. Additive functors can be either
(additive funktorer) co- or contra-variant. The functors Hom A (´, N ) and Hom A ( N, ´) are prototypical
examples of such animals.
(5.6) A basic property of an additive functor is that it preserves direct sums, a property of
diverse functors that will be fundamental at several later occasions In stead of repeating
multiple ad hoc proofs, we prefer giving one general version. Additionally, the functors
will often depend on other parameters, as will do the ones in our applications, and the
notation can easily appear rather decorated; we believe it easier to grasp the salient
points when both notation and context are stripped down to bear essentials.

Proposition 5.7 (Additive functors preserve direct sums) Given rings A and B. Let
F : Mod A Ñ ModB be an additive functor. Assume that M1 and M2 are submodules of an
A-module of an A-module M and that α : M Ñ N is an A-linaere map.
i) If M decomposes as a direct sum M = M1 ‘ M2 , then F ( M) decomposes as F ( M) =
F ( M 1 ) ‘ F ( M 2 );
ii) The decomposition of F ( M) is functorial in that F (α) = F (α)|F( M1 ) + F (α)|F( M2 ) .

Proof: A natural way to approach direct sum decompositions is through idempotent

endomorphism (asin Proposition 4.27 on page 95). If M is an A-module and e an
idempotent endomorphism of M—as holds for any functor—F (e) is an idempotent
endomorphism of F ( M ); indeed, e being idempotent means that e ˝ e = e and applying
F to that equality yields F (e) ˝ F (e) = F (e ˝ e) = F (e).

10th January 2021 at 11:32am

Version 4.0 run 30
 exact sequences 127

Now, the equality id M = e + (id M ´e) incarnates the decomposition M = eM ‘

(id M ´e) M, and when F is as applied to it, it becomes transformed into


idF( M) = F (id M ) = F e + (id M ´e) = F (e) + (idF( M) ´F (e)).

because F is additive. Consequently F(M) decomposes into the direct sum F ( M) =

F (e) F ( M ) ‘ idF( M) ´F (e) F ( M).
We proceed to verify that F (eM) = F (e) F ( M ), which is an essential part of the
statement that F respects direct sums. To that end consider the factorisation
e

M π eM ι M (5.7)

where π is just the map e but considered to take values in its image eM, and ι denotes
the inclusion. Obviously π ˝ ι = ideM . Applying F to (5.7) yields
F (e)

F( M) eF ( M ) F( M) (F (5.7))
F (π ) F (ι)

From this ensues that F (ι) maps F (eM ) onto F (e) F ( M), but since π ˝ ι = ideM , it
holds true that F (π ) ˝ F (ι) = idF(eM) so that F (ι) is an injection, and hence F (eM ) =
F (e) F ( M). We have proven the covariant version of the first statement in next proposi-
tion:
What is left to verify is the assertion about functoriallity. Now, α decomposes as
α = α| M1 + α| M2 and since F is additive we will be through, once we have proven that
F (α)|F( M1 ) = F (α| M1 ) (by symmetry, the same will then hold for α| M2 ), but checking
that is effortless: Applying the functor F to the following commutative diagram where
notation is as above,
α| M1

M1 = eM ι M α N
we obtain the commutative diagram

α| F ( M1 )

F ( M1 ) = eF ( M ) ι F( M) F ( N ),
F (α)

and that’s it. o

(5.8) We round off this little excursion into category theory with a result on natural trans-
formations between additive functors, which basically asserts that such transformations

10th January 2021 at 11:32am

Version 4.0 run 30
 128 exact sequences
/ G( M)
ηM
F( M)
preserves direct sums. Assume that F, G : Mod A Ñ ModB are two additive functors and
F (φ) G (φ)
that η : F Ñ G is a natural transformation, which is just a collection of B-linear maps
 
F( N ) / G( N ) η M : F ( M ) Ñ G ( M ), one for each A-module M, such that η N ˝ F (φ) = G (φ) ˝ η M for
ηN
each A-linear φ : M Ñ N.

Proposition 5.9 Assume that we are given two additive functors F, G : Mod A Ñ ModB and
a natural transformation η : F Ñ G. If an A-module M decomposes as M = M1 ‘ M”, then
η M = η M1 + η M2 .

Proof: For any direct summand L1 in an A-module L and any additive functor H we
shall denote by ι L1 the inclusion of L1 in L; then H (ι L1 ) = ι H ( L1 ) , which is the essential
content of Proposition 5.7 above.
With this in mind, we may decompose each x P F ( M ) = F ( M1 ) ‘ F ( M2 ) as

x = ι F( M1 ) y + ι F( M2 ) z = F (ι M1 )y + F (ι M2 )z,

and applying η to that relation, we obtain since η by definition is a module homomor-

phism



η M ( x ) = η M F ( ι M1 ) y + η M F ( ι M2 ) z = G ( ι M1 ) η M1 ( y ) + G ( ι M2 ) η M2 ( z ),

which is precisely the assertion we want. o

Note that the assertions remains true if one e. g. replaces one or both of the categories
by the categories moda and modB of finitely generated modules (or for that matter by
any additive categories; our proofs are entirely arrow based).

Exercises
(5.6) Let F be a covariant functor Mod A Ñ ModB . Let φ : M Ñ N be linear and assume
it has a left (respectively right) section. Show that F (φ) has a left (respectively right
section). What if F is contravariant? Give an example of a functor F and a module with
a decomposition M = N ‘ N 1 such that F ( M ) is not isomorphic to F ( N ) ‘ F ( N 1 ).
(5.7) Additive functors do not necessarily commute with infinite direct sums. Prove
that HomZ ( iPN Z, Z) = iPN HomZ (Z, Z) = iPN Z. Prove that iPN Z is not
À ś ś ś

isomorphic to iPN Z. Hint: For the first question go back up to Proposition 4.23 on
À

page 93; for the second verify that the two abelian groups are of different cardinalities,
or resort to Exercise 4.27 on page 103.
(5.8) Short exact sequences which are not split, may cease bing exact when exposed
to an additive functor. Describe the resulting sequence when one applies the functor
HomZ (Z/pZ, ´) to the standard short exact sequence

0 / Z/pZ / Z/p2 Z / Z/pZ / 0.

10th January 2021 at 11:32am

Version 4.0 run 30
 left exactness of hom-functors 129

What happens when HomZ (Z/p2 Z, ´) is applied?

M

5.2 Left exactness of hom-functors

That hom-functors are left exact, is not a very deep result—one checks it by just doing
what is needed—but is a fundamental property of hom’s.
(5.10) Suppose given a short exact sequence like (5.1) above and let N be any A-module.
/ M1 /M / M2
α β
Applying the covariant hom-functor Hom A ( N, ´) to the sequence we obtain an induced 0 O C
sequence shaped like φ
0
N
/ Hom A ( N, M1 ) / Hom A ( N, M ) / Hom A ( N, M2 ) .
α˚ β˚
0 (5.8)
/ M1 /M / M2
α β
The maps are simply given by composition; i. e. α˚ φ = α ˝ φ and β ˚ φ = β ˝ φ, and 0 [ O C
since β ˚ ˝ α˚ = ( β ˝ α)˚ and β ˝ α = 0, it holds true that β ˚ ˝ α˚ = 0. More is true, the φ
0
sequence (5.8) will in fact be exact. There are two spots where exactness needs to be
N
checked. The first point is that α˚ is injective. But α˚ (φ) = φ ˝ α, and since α 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 that the sequence (5.8) is exact at the middle spot,
assume that β ˝ φ = 0 for a map φ : N Ñ M. Then φ factors through the image α( M1 ),
and α being injective, φ can be represented as α ˝ φ1 for a map φ1 : N Ñ M1 , which is
precisely what we desire.
(5.11) In a similar vein, the contravariant version Hom A (´, N ) applied to (5.1) yields
the sequence

β˚ α˚
0 / Hom A ( M2 , N ) / Hom A ( M, N ) / Hom A ( M1 , N ) , (5.9)

where the arrows are reversed, and repeating mutatis mutandis the argument above one
shows that this also is an exact sequence.
(5.12) It is common usage to refer to the phenomena described above as saying that Left exact functors
(venstre-eksakte
Hom A ( N, ´) and Hom A (´, N ) are left exact functors. The two functors Hom A ( N, ´) funktorer)
and Hom A (´, N ) are however, seldom exact functors in the sense that they take short Exact functors (eksakte
exact sequences to short exact sequences. There are crowds of examples that β ˚ and α˚ funktorer)

are not surjective.

A large part of homological algebra was developed just to describe the "missing
cokernels" coker β ˚ and coker α˚ . In general, the answer to this challenge is that the two
sequences can be extended ad infinitum to the right to yield long exact sequences which
involve so-called Ext-modules. These modules depend only on the modules involved,

10th January 2021 at 11:32am

Version 4.0 run 30
 130 exact sequences

not on the maps in the original short exact sequence, and of course, they depend on N
as well. However, the maps in the long exact sequence depend on the entire short exact
sequence. In some good cases the Ext-modules can be computed and the long exact
sequences be controlled.
Right exact functors (5.13) There is of course the symmetric notion of right exact functors, with the lack of
(høyreeksakte exactness appearing at the left end of a sequence. The tensor product, which shortly
funktorer)
will be introduced, will be of this kind.
One meets these semi exact functors in a variety of contexts and defined in different
abelian categories. The modules—or one should rather say objects— involved in long
exact sequences associated with short exact ones, depend functorially on the objects
and are the famous derived functors. Most cohomology theories in the universe can be
constructed like this.
(5.14) In paragraph 5.10 above we proved the "only-if-part" of the following proposition
(although we worked with short exact sequences like in (5.1), we never used that β was
surjective).

Proposition 5.15 (Left exactness I) Let the sequence

/ M1 /M / M2
α β
0 (5.10)

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 ( N, M1 ) / Hom A ( N, M ) / Hom A ( N, M2 ) (5.11)

is exact.

Proof: To attack the remaining "if-part" assume that (5.11) 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
(5.16) There is also an assertion dual to the one of Proposition 5.15 above. The proofs
of the two being quit similar, we leave all the checking to the zealous students; it is a
good training for these diagram-arguments. The assertion reads as follows:

Proposition 5.17 (Left exactness II) Let the sequence

α β
M1 M M2 0 (5.12)

10th January 2021 at 11:32am

Version 4.0 run 30
 left exactness of hom-functors 131

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 ( M1 , N ) Hom A ( M, N ) Hom A ( M2 , N ) (5.13)

is exact.

Example 5.4 As alluded to in Paragraph 5.12 above, even if the map β is surjective, the
induced map β ˚ will in most cases not be surjective. The simples examples are the
short exact sequences of abelian groups
[n] β
0 Z Z Z/nZ 0, (5.14)

where n is an integer and β the canonical projection. It holds obviously true that
HomZ (Z/nZ, Z) = 0 and HomZ (Z/nZ, Z/nZ) = Z/nZ, so the induced sequence
becomes
˚ Grown up algebraists
0 0 0 Z/nZ,
do not write [n] for the
and, of course, a map 0 Ñ Z/nZ cannot be surjective! We may as well apply the functor multiplication map,
only n. So this is the
HomZ (´, Z/nZ) to 5.14 and obtain the sequence
last instance of the
[n]
notation [n].
0 Z/nZ Z/nZ Z/nZ.

Of course, multiplication* by n (which is an upper-star-map in this case; [n]˚ = [n].) is

the zero map on Z/nZ and is not surjective. K
Exercises / M1 α /M
β
/ M2 /0
0 Z O
(5.9) Convince yourself that β ˚ surjective means that any A-linear map φ : N Ñ M2
ν
can be lifted to an A-linear map into M, as the diagram in the margin illustrates.
(5.10) Give the argument referred to in the previous paragraph in detail. N
(5.11) In most cases the map α˚ will not be surjective even if α is. Convince yourself
that α˚ being surjective means that any map φ : M1 Ñ N can be extended to a map
M Ñ N, as in the marginal diagram / M1 α /M
β
/ M1 /0
0
M
φ
Projective and injective modules  
(5.18) In view of the two left exactness theorems two classes of modules stand out, N
namely the ones such that the functor Hom A ( N, ´) is exact and those such that
Hom A (´, N ) exact. The former are called projective modules (they are ubiquitous in Projective modules
commutative algebra, and we shall come back to them) and the latter are the so-called (projektive moduler)

injective modules.
(5.19) The universal mapping property that free modules enjoy, entails that they are Injective modules
projective, but they are not alone. When the base ring has non-trivial idempotents, (injektive moduler)

10th January 2021 at 11:32am

Version 4.0 run 30
 132 exact sequences

there are cheap examples, but finding examples over say integral domains requires
some effort. Some classes of rings, as the local rings or the polynomial rings, enjoy the
property that all projective modules are free. Over local rings this is an easy consequence
of Nakayma’s lemma when the modules are finite and a result of Kaplansky’s in general,
but over polynomial rings it is a deep theorem, first conjectured by Jean Pierre Serre in
/ M2 /0
β
MO = 1955 and proved by Daniel Quillen and Andrei Suslin about twenty years later.
ψ
φ Proposition 5.20 If F is a free A-module, the functor Hom A ( F, ´) is exact; in other words,
F free modules are projective.

Proof: It suffices to prove that β ˚ is surjective when β : M Ñ M2 is surjective. Suppose

that φ : F Ñ M2 is given, and let t f i uiP I be a basis for F. Since β is surjective, there are
elements tmi uiP I such that β(mi ) = φ( f i ). By the Universal Mapping Property of free
modules (on page 100), there is a map ψ : F Ñ M with ψ( f i ) = mi . Then β ˝ ψ = φ since
β(ψ( f i )) = β(mi ) = φ( f i ), and two maps agreeing on a basis are equal. o
(5.21) Although projective modules are not necessarily free, they are closely related to
free modules; they will always be a direct summand in a free module:

Proposition 5.22 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) and consider the exact
sequence
À φ
0 K iP I A P 0,
ř
where a string ( ai )iP I is sent to i ai pi (which is meaningful since merely finitely many
of the ai ’s are non-zero). Because the pi ’s generate P, this map is surjective. The point
is that because P is projective, the identity map idP : P Ñ P can be lifted to a map
À
σ : P Ñ iP I A. This means that φ ˝ σ = id[ P], and the lifting σ is a right section of
 ι / P ‘ P1
À
φ. Hence P lies split in iP I A by the Splitting Criterion on page 124. To prove
P
2
the converse implication let β : M Ñ M be a surjection and let φ : P Ñ M be given.2
π
 Assume further that P1 is a complement to P in a free module; that is, P ‘ P1 is free.
P Consider the map P ‘ P1 Ñ M2 sending a pair ( x, y) to φ( x ). This map can be lifted as
φ P ‘ P1 is free, and restricting the lifted map to P yields a lifting of φ. o
  
M / / M2 Examples
β

(5.5) A projective module which is not free: Let A = Z/2Z ˆ Z/2Z and consider M =
Z/2Z ˆ (0) which has a natural structure as an A-module. Then M is projective, since

10th January 2021 at 11:32am

Version 4.0 run 30
 left exactness of hom-functors 133

if N = (0) ‘ Z/2Z, we have M ‘ N » A (as A-modules!). However, M is clearly not

free, since any free A-module M » A I must have at least four elements!
(5.6) Another projective module which is not free: A similar example can be constructed
over the ring A = 8 i =0 Z. ś
We may regard M = Z as an A-module embedding it
ś

as the 0-th component Z Ď 8 i =0 Z. Then Z is projective, since Z ‘ i =1 Z » A.

ś8

However, note that the ring A is uncountable. In particular, this means that Z certainly
is not isomorphic to any module of the form A I . The same argument shows that the
A-module M = 8 i =0 Z is projective, but not free.
À

(5.7) More non-free projective modules: The modules emerging in the previous examples
are of a similar character; they are instances of modules arising in general setting where
the ring A is a non-trivial direct product A = A1 ˆ A2 . The factor A1 ˆ t0u (or t0u ˆ A2)
lies split in A as an A-module, and consequently it is projective. However, is not free
as the product is non-trivial. One way to see this is to consider the annihilator ideal
(0 : A1 ˆ t0u). For free modules the annihilator ideal is the zero ideal (indeed, if e is a
basis element, xe = 0 implies x = 0.), where as for A1 ˆ t0u it equals t0u ˆ A2 which is
non-zero since the product is non-trivial.
(5.8) More sophisticated examples: There is a large and all important class of rings called
Dedekind rings in which all ideals are projective. A rich source of Dedekinds rings are
the coordinate rings of the so-called affine regular curves in algebraic geometry, and
the core activity of algebraic number theory is the study of Dedekind rings which are
finitely generated Z-modules. The The quadratic extensions Z[ n] and Z[(1 + n)/2]
‘ ‘

according to n being congruent to one modulo 4 or not, are examples of such*. And ˚ They are the integral

in most of these rings you will find ideals that are not principal; that is, ideals that are closures of Z in
Q( n )
‘
not free modules. We might as well have given examples from geometry, and for the
geometers we offer a treatment of the ideals in the coordinate ring of an affine elliptic
curve at a later occasion (Exercise 8.19 on page 227). The two cases are strikingly similar,
and of course, there is a common theory behind—but that will also be for later.
For the moment we content ourself with giving just one illustrating example, the
? ?
ideal a = (2, 1 + i 5) in the ring A = Z[i 5]. We shall give an explicit construction
of a as a direct summand in the free module A ‘ A. Hence it will be projective, but
not being principal it is not free. In fact, we shall prove that a ‘ a » A ‘ A, which
also serves as an example of two isomorphic direct sums whose summands are not
‘ ‘
isomorphic. To ease the notation, we let z = 1 + i 5; then z̄ = 1 ´ i 5 and zz̄ = 6.
The free module A ‘ A has the usual basis e1 = (1, 0) and e2 = (0, 1). The gist of
the construction is the map
α: A ‘ A Ñ aĎ A
defined by the assignments e1 ÞÑ 2 and e2 ÞÑ z. We shall identify a submodule M inside
A ‘ A that α maps isomorphically onto a and thereby proving that a lies split in A ‘ A.

10th January 2021 at 11:32am

Version 4.0 run 30
 134 exact sequences

The submodule M in question is generated by the two elements a1 = ´2e1 + z̄e2 and
a2 = ´ze1 + 3e2 .
We begin with checking that the restriction α| M is surjective. This ensues from the
very definition of α as the following little calculations show:

α( a1 ) = ´2 ¨ 2 + zz̄ = ´4 + 6 = 2 α( a2 ) = ´z ¨ 2 + 3 ¨ z = z.

To prove that α is injective on M, pick an element a = xa1 + ya2 in M that maps to

zero, which means that 2x + zy = 0 in A. Now, one has

xa1 + ya2 = ´(2x + zy)e1 + (z̄x + 3y)e2 = (z̄x + 3y)e2 ,

and since 0 = z̄(2x + zy) = 2z̄x + z̄zy = 2(z̄x + 3y), it follows that a = 0.
This shows that A ‘ A » a ‘ ker α, but to see that A ‘ A » a ‘ a some further effort
is required. Since a = (2, z) = (2, z̄), the twins z and z̄ enter symmetrically into the
picture, so swapping z and z̄ and e1 and e2 , we may as well apply what we just did to
the submodule N Ď A ‘ A generated by ´ze1 + 2e2 and ´3e1 + z̄e2 , and conclude that
N lies split and is isomorphic to a. But one directly verifies that N Ď ker α, and since
both lie split, they must coincide.
K

Exercises
(5.12) Let n be a natural number. Decide for which natural numbers m the resulting
sequence is exact when the functor HomZ (´, Z/mZ) is applied to the short exact
sequence
0 /Z n /Z / Z/nZ /0.

(5.13) Let A be a ring and suppose that B is an A-algebra. Assume given an 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.
(5.14) Multiplicativity of the characteristic polynomial. Assume given a commutative
ψ ‹‹
diagram of A-modules

0 /E /F /G /0

0 θ ψ φ θ
  
0 /E /F /G /0

10th January 2021 at 11:32am

Version 4.0 run 30
 snakes and alike 135

where the rows are exact and the involved modules are free of finite rank. Show
the equality Pφ (t) = Pψ (t) ¨ Pθ (t). Conclude that det φ = det θ ¨ det ψ and that tr φ =
tr θ + tr ψ. Hint: Exhibit a basis for F in which the matrix of φ has an appropriate block
decomposition (as in the margin).
(5.15) Consider the exact sequence of finite abelian groups

0 A B C 0.

Show that #B = #A ¨ #C.

M

5.3 Snakes and alike

(5.23) An all important feature in homological algebra are the so-called connecting maps
which relate homology modules 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 the reason behind
the name.

Lemma 5.24 (The snake lemma) Given a diagram

/ M2 / M3 /0
φ1 φ2
M1 (5.15)
α1 α2 α3
  
0 / N1 / N2 / N3
ψ1 ψ2

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

ker α2 / ker α3 δ / coker α1 / coker α2 , (5.16)

where the two unmarked maps are the ones induced respectively by and φ2 and ψ1 .

Proof: The proof of the Snake Lemma is an example of a sport called diagram chasing,
which when homological algebra arose, was extensively practised among homological

10th January 2021 at 11:32am

Version 4.0 run 30
 136 exact sequences

algebraists. We have two missions to complete; firstly, the map δ must be constructed,
y /x
_ and secondly, we must verify that the sequence (5.16) is exact.
_
We begin with the first and most interesting task. A short and dirty mnemotechnical
  definition of δ is ψ1´1 ˝ α2 ˝ φ2´1 which of course is meaningless as it stands since neither
 / α2 ( y )  /0
_z φ2 nor ψ1 is invertible, but it gives a hint of how to construct δ. For each x P ker α3 it
holds true, with a liberal interpretation of the inverses, that δ( x ) = [ψ1´1 (α2 (φ2´1 ( x )))],

δ( x ) where [y] designates the class in coker α1 of an element y P N1 .
After this heuristics, the fun is starting: Pick an element x P M3 so that α3 ( x ) = 0
and lift it to an element y in M2 ; that is, pick an element y P M2 with φ2 (y) = x. The
rightmost square of (5.15) being commutative, we infer that ψ2 (α2 (y)) = 0; the bottom
line of (5.15) being exact, there is thence a z in N1 with ψ1 (z) = α2 (y). And that is it;
the image of z in coker α1 is the wanted guy δ( x ).
We made a choice on the way—the choice of a lift of x to M2 —and for the definition
of δ to be legitimate, the image of the trapped z in coker α1 must be independent of that
choice. So assume that y1 is another element of M2 that maps to x; then we may write
y1 = y + w with φ2 (w) = 0. Since the top line of (5.15) is exact, it holds that w = φ1 (u)
for some u, and we find

α2 (y1 ) = α2 (y) + α2 (φ1 (w)) = α2 (y) + ψ1 (α1 (u))

Luckily, ψ1 is injective (the bottom line of (5.15) is exact), so if z1 P N1 is such that

ψ1 (z1 ) = y1 one has

z1 = z + α1 ( u ),

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 (5.16):
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 δ( x ) = 0. This means that
z = α1 (v) for some v P M1 ; hence α2 (y) = ψ1 (z) = ψ1 (α1 (v)) = α2 (φ1 (v)). It follows
that y = φ1 (v) + t with t P ker α2 and consequently x = φ2 (y) = φ2 (t), which is what
we need. o

Why snake?
(5.25) 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

10th January 2021 at 11:32am

Version 4.0 run 30
 snakes and alike 137

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

The Snake Lemma is frequently applied in situations where the map φ1 : M1 Ñ M2

is injective, and the map ψ2 : N2 Ñ N3 is surjective so that the diagram we depart from
is shaped like
/ M1 / M2 / M3 /0
φ1 φ2
0 (5.17)
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 the 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.

(5.18)

Lemma 5.26 (Snake lemma II) Assume given a commutative diagram with exact row as in
(5.17). Then the six term sequence (5.18) 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. What remains to be done
is checking exactness at ker α2 and coker α2 , and this follows by two simple hunts in
the diagram. We shall check exactness at ker α2 , but leave exactness at coker α2 for the
students to practise diagram chasing. So assume that x P ker α2 is such that φ2 ( x ) = 0.
Then x = φ1 (y) for some y P M1 , and ψ1 (α1 (y)) = α2 (φ1 (y)) = α2 ( x ) = 0. Since ψ1 is
assumed to be injective, it follows that y P ker α1 , and we are through. o

10th January 2021 at 11:32am

Version 4.0 run 30
138 exact sequences

Exercises
(5.16) The five lemma I. Use the Snake Lemma to prove the following abbreviated 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.
(5.17) The five lemma II. Given a commutative diagram
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 isomorphisms, 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 homomorphisms 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 5.27 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
where the rows are exact and where F is free of rank n ´ 1.
(5.18) Infer from the previous exercise that if A = Z and φ has non-vanishing
determinant, it holds true that the cokernel coker φ is finite and that |det φ| = # coker φ.
Hint: Use the Snake Lemma.
(5.19) In the same vein as in Exercise 5.18 above, assume that A = k[t] is the polynomial
ring over the field k and that φ has non-vanishing determinant. Prove that coker φ is

10th January 2021 at 11:32am

Version 4.0 run 30
 complexes 139

of finite dimension over k and that deg det φ = dimk coker φ Hint: Again, the snake is
the solution.
(5.20) Modules of finite presentation. One says that an A-module M is of finite presentation Modules of finite
if it sits in an exact sequence presentation (moduler
av endelig
An / Am /M /0 presentasjon)

where An and Am are finitely generated free modules. In general this is a more
restrictive condition on a module than being finitely generated. Over Noetherian rings
however, the two are equivalent. Let N Ď M be a submodule. 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 / Ar + s / As /0
with exact rows and all three vertical map being surjective. Then apply the Snake
Lemma and Exercise 4.16 on page 99.
M

5.4 Complexes
The consept of a complex originated in topology around the end of the end of the 19th
century, when the topologists begun to apply them in the study of so-called triangulated
spaces. In is simplest form such a space is represented as union (with some conditons)
of oriented simplexes, which are continuous images of certain standard simplices. A
standard 1-simplex is just a oriented closed interval, a 2-simplex a triangle, a 3-simplex
a tetrahedron etc.
Long before complexes appeared in analysis, but without playing such a centre stage
rôle as they did in topology. In courses of calculus of several variable we we learned
that relations like curl grad f = 0 and div curl t = 0 for a function f and vector field
T, both defined and twice continuously differentiable in an open subset of R3 , and
the operators grad, curl and div combine to make up a complex, one of those called
deRham complexes.
The definition
In traditionally topology it is customary to distinguish between chain complexes and
cochain complexes, but they are just two ways of representing the same thing, although
the roles they play in topology are different. In modern presentations of homological
algebra, this practice has for the most ceased, and one just speaks about complexes.
(5.28) A complex (Ci , di ) of A-module is a sequence of A-modules tCi uiPZ indexed by Complexes
Z together with a sequence tdi uiPZ of A-linear maps di : Ci Ñ Ci+1 subjected to the (komplekser)

10th January 2021 at 11:32am

Version 4.0 run 30
 140 exact sequences

condition that the composition of two consecutive ones be zero; that is, di+1 ˝ di = 0 for
all i. A complex may be displayed as

... / Ci / Ci+1 / Ci+2 / ...

di d i +1

˚ The names stem from The maps di are called differentials, but they also obey the names boundary or coboundary*
topology where one has maps. The notation C‚ (pronounced C-dot) for a complex is standard, the differential
chains and chain
complexes that give being tacitly understood. The symbolic function of the dot is to indicate a placeholder
homology, cochains for the index.
and cochains complexes
There is another and more compact way of denoting a complex; one simply sums
that give cohomology, À
and this dichotomy up the modules and introduces the module C‚ = i Ci . It is a graded module, and
persists for the the elements of each summand Ci are said to be homogenous of degree i. Summing
diffenertials; hence,
both boundary maps up the di ’s gives an A-linear map d : C‚ Ñ C‚ whose square is zero; i. e. d2 = 0.
and coboundary maps. It is a homogeneous map of degree one; that is, it sends homogeneous elements to
homogeneous elements, but raises their degree by one. So to give a complex of A-
modules is equivalent to giving such a graded A-module together with an A-linear
endomorphism which is homogeneous of degree one and is of square zero.
Morphisms of (5.29) A morphism, or simply a map, ψ from one complex C‚ = (Ci , di ) to another
complexes (morfier D‚ = ( Di , d1i ) is a sequence tψi uiPZ . Each ψi is an A-linear map ψi : Ci Ñ Di , and one
mellom komplekser)
imposes the constituting condition that the ψi ’s commute with the differentials. In other
words, ψi+1 ˝ di = d1i+1 ˝ ψi for all i. Displayed, the map presents itself as a diagram

di d i +1
... / Ci / Ci+1 / Ci+2 / ...

ψi ψi+1 ψi+2
  
... / Di / Di + 1 / Di + 2 / ...
d1i d1i+1

where all squares are commutative. Two composable maps of complexes, i. e. maps so
that the source of one equals the target of the other, are composed level by level, and
with this composition the complexes of A-modules form a category CplxA (which in
fact, turns out to be abelian).
As common usage is, one says that a map ψ between two complexes is an isomorphism
of complexes if it has an inverse. This amounts to each ψi being an isomorphism since
then the inverses automatically commute with differentials.
(5.30) In the compact notation, a complex is a graded module equipped with a differen-
tial of degree one, and a map ψ : C‚ Ñ D‚ between two complexes is just an A-linear
map respecting the grading and commuting with the differentials. The kernel ker ψ,
the image im φ and the cokernel coker φ of φ are all graded modules in a natural way
as described in Paragraph 4.46 on page 104. The kernel and the image are invariant

10th January 2021 at 11:32am

Version 4.0 run 30
 complexes 141

under the differential; indeed, if φ( x ) = 0 one has ψ(dC ( x )) = d D (ψ( x )) = 0 and if

y = φ( x ), it holds d D (y) = d D (ψ(y)) = φ(dC ( x )) =. Thus they are both complexes.
It follows that the differential d D passes to cokernel coker φ and consequently also
coker ψ is a complex. One verifies easily that ker φ (respectively im φ and coker φ) is the
kernel (respectively the image and the cokernel) of φ in the category Cplx A of complexes.
Hence one has the notion of exact sequences of complexes; and in particular short exact
sequences, these are tailored like

/ C‚ / D‚ / E‚ / 0.
α β
0

In the compact notation they are just usual short exact sequences of graded modules,
but with the maps α and β homogenous of degree zero and commuting with the
differentials—as one says, they are chain-wise exact; in each degree i, it holds that.
ker β i = im αi .
(5.31) The direct sum C‚ ‘ D‚ of the two complexes has the obvious underlying
À
graded module C‚ ‘ D‚ = i (Ci ‘ Di ) and is equipped with the differential defined
by d( x, y) = (dC ( x ), d D ( x )). Both the two projections and the two canonical inclusions
are maps of complexes.
(5.32) There is a shift operator acting on complexes which lowers the degrees by one The shift operator
and changes the sign of the differential; that is, the shifted complex C‚ (1) satisfies (skiftoperatoren)

(C‚ (1))i = Ci+1 for all i, and the differential is given as dC(1) = ´dC . The shift also
operates on homomorphisms by the the natural rule φ(1)i = φi+1 , and thus (1) is an
endofunctor (1) : Cplx A Ñ Cplx A . The d-fold iterate of (1) is naturally denoted by (d).
(5.33) There is a variant of the definition of a complex with the differential decreasing
the degree by one; it displays as

di di´1 di´2
... / Ci / C i ´1 / C i ´2 / ... .

There is of course only a notational difference between the two definitions, and one my
pass from one to the other by the convention (C‚ )i = (C‚ )´i —rising (or lowering) the
indices is accompanied with a change of sign.

Exercises
(5.21) Let ψ : C‚ Ñ D‚ be a map of complexes. Assume that each ψi is a bijection and
call the inverse φi . Show that φi ’s commute with the differentials.
(5.22) Let C‚ = (Ci , di ) be a complex of A-modules. Let furthermore tαi uiPZ be a
sequence of units in A Show that C‚1 = (Ci , αi di ) is a complex isomorphic to C‚ . In
particular if tei uiPZ is a sequence of signs; that is, each ei P t1, ´1u, the complexes
(Ci , di ) and (Ci , ei di ) are isomorphic.
j
(5.23) If tC‚ u jP J is a family of complexes of A-modules define their direct sum

10th January 2021 at 11:32am

Version 4.0 run 30
 142 exact sequences

À j ś j
C‚ and a direct product jP J C‚ and verify that they have the approriate universal
jP J
properties in the category Cplx A of complexes.
(5.24) Show that the functor sending an A-module M to the complex M‚ whose only
non-zero term is the module M in degree zero defines an exact functor ModA Ñ Cplx A .
M
The homology of a complex
(5.34) Since di+1 ˝ di = 0, it holds true that im di Ď ker di+1 and one may form the i-th
Homology of a complex homology module Hi (C‚ ) = ker di / im di´1 of the complex. This can of course be done at
(homologien til et each level i, and a fundamental invariant of a complex is the so-called total homology
kompleks) À
H‚ (C‚ ), the direct sum of all the homology modules H‚ (C‚ ) = i Hi (C‚ ).
In the compact notation with C‚ being a graded module equipped with an en-
domorphism dC of square zero, the total homology is just given as the quotient
H‚ (C‚ ) = ker dC / im dC .
Exact complexes A complex is said to be exact or acyclic if all the homology modules vanish; that is, it
(ekasakte komplekser) is exact at every stage.
The homology is a functorial construction. A map of complexes φ : C‚ Ñ D‚ is
Ascyclic complexes required to commute with the differentials and therefore it maps ker dC into ker d D ;
(asykliske komplekser) indeed, d D (φ( x )) = φ(dC ( x )) = 0 whenever x P ker dC . Similarly φ sends im d D into
im d D because φ(dC ( x )) = d D (φ( x )). Thus φ induces an A-linear map H‚ φ : H‚ C‚ Ñ
H‚ D‚ . It as a matter of easy checking that H‚ (φ ˝ ψ) = H‚ φ ˝ H‚ ψ, so that the homology
is a functor.
Long exact sequences and exact triangles
Where there are short exact sequences there must also be long exact sequences, and
we have now come to the point when we shall— hopefully comforting any doubters—
establish this fundamental dogma on which the whole homological algebra rests:
With any short exact sequence of complexes is associated a long exact sequence in a
functorially way. The main players in this performance are the so-called connecting
homomorphisms.
(5.35) A short exact sequence of complexes is just a short exact sequence of the
underlying graded modules; that is, an exact sequence shaped like

/ B‚ / C‚ / D‚ /0
α β
0 (5.19)
where of course α and β are morphisms belonging to the category Cplx A ; that is, they are
A-linear maps homogeneous of degree zero (i. e. respect the grading) and commuting
with the differentials. Saying (5.19) being exact is to say it is exact in each degree; in
other words, each of he sequences

/ Bi / Ci / Di /0
αi βi
0

10th January 2021 at 11:32am

Version 4.0 run 30
 complexes 143

is exact. The long exact sequence associated with (5.19) is, well, a long sequence that is
exact at each term, and it is shaped as follows
H‚ α H‚ β
... Hi B‚ Hi C‚ Hi D‚
δ

H‚ α H‚ β
Hi+1 B‚ Hi+1 C‚ Hi+1 D‚ ...
(5.20)

where the newcomer δ—the maps Hi α and Hi β are old-timers define above—is the
famous connecting homomorphism which we are about to construct. Trying to keep the The connecting
notation as simple and practical as possible we have stripped δ for all sub’s and super’s, homomorphism
(sambandsmorfien)
and the dependence on the sequence (5.19) is tacitly understood, as is the degree i.
The long sequence extends infinitely in both directions, but of course, if the involved
complexes are concentrated in a certain region; for instance, in positive or negative
degrees, the long exact sequence will be confined to the same region.
Long sequences like (5.20) are cumbersome to work with, and a more compact
notation has been devised based on so-called exact triangles*, which compress the long Exact triangles (eksakte
sequence into the compact form triangler)

H‚ α H‚ β
H‚ B‚ / H‚ C‚ / H‚ D‚ δ / H‚ B‚ (1). H‚ B] ‚
H‚ α
/ H‚ C‚

As usual, exactness means that at each module the kernel of the outbound map equals H‚ β
δ


the image of the inbound one, for instance will im δ = ker H‚ (α). Rolling out the H‚ D‚
triangle, degree by degree, one gets back the long sequence (5.20).

Proposition 5.36 With every short exact sequence of complexes,

˚ One may draw the
/ B‚ / C‚ / D‚ /0
α β
0 (5.21) sequence as a triangle
like the one above
is associated a connecting homomorphism δ : H‚ D‚ Ñ H‚ B‚ (1) giving rise to an exact triangle which justifies the
name triangle, but
(and thereby to a long exact sequence) observe that it does not
show that the degree of
H‚ α
H‚ B‚d / H‚ C‚ δ equals one.

δ z H‚ β
H‚ D‚

The connecting homomorphism and the triangle depend functorially on the exact sequence.
Proof: One may well attack the proof with a direct assault chasing in the diagram, but
as the chase already has been done by the snake lemma, and we rely on that result. The
relevant snake diagram is the following one:

10th January 2021 at 11:32am

Version 4.0 run 30
144 exact sequences

ker dC ker d D

α β
B‚ C‚ D‚ 0
dB dC dD

0 ker d B ker dC D‚

H‚ B‚ H‚ C‚
H‚ α

0 0

where the green snake δ1 is a precursor for the connecting map δ. Recall the expression
δ1 ( x ) = [α´1 (dC ( β´1 ( x )))] for the snake, where β´1 is not a genuine map, but β´1 ( x )
denotes any preimage for x, and [y] denotes the homology class of an element y from
B‚ . Homogeneous elements of D‚ may be lifted to homogeneous elements of C‚ and as
the differential dC is homogeneous of degree one, the snake δ1 will be homogeneous of
degree one too.
To see that the precursor δ1 passes to the quotient H‚ D‚ = ker d D / im d D and yields
the desired map δ, we must verify it δ1 vanishes on im d D ; but if x = d D (y), it holds that
dC (z) lifts x if z lifts y, and consequently δ1 ( x ) = [α´1 (dC ( β´1 ( x )))] = [α´1 dC dC (z)] =
0. Hence the snake lemma yields the exact sequence

H‚ β H‚ α
H‚ C‚ / H‚ D‚ δ / H‚ B‚ / H‚ C‚ .

This settles the subtler portion of the exactness statement, and the missing piece, that
the part
H‚ α H‚ β
H‚ B‚ / H‚ C‚ / H‚ D‚

is exact, is not hard and is left to the zealous students.

Finally, we have come to the assertion that the connecting homomorphism δ be
functorial. A homomorphism between two exact sequences in Cplx A is best digested by
drawing the diagram

/ B‚ / C‚ / D‚ /0
α β
0
φB φC φD
  β1 
/ B‚1 α1 / C‚1 / D‚1 /0
0

10th January 2021 at 11:32am

Version 4.0 run 30
 complexes 145

where all maps are maps of complexes, and the two squares are commutative; the
connecting homomorphism being a functorial construct means that the three squares in

H‚ α H‚ β
H‚ B‚ / H‚ C‚ / H‚ D‚ δ / H‚ B‚ (1)

H‚ φB H‚ φC H‚ φD H‚ φB (1)
 α1
 H‚ β1  
H‚
/ H‚ C‚1 / H‚ D‚1 δ1 / H‚ B‚1 (1)
H‚ B‚1

commute. The only challenges establishing this is to verify that the right square
commute, the two others commute by the functoriallity of H‚ α and H‚ β. By the
quasi-formula δ( x ) = [α´1 (dC ( β´1 ( x )))], where x represents a class in H‚ B‚ , we find

φB δ( x ) = [φB α´1 (dC ( β´1 ( x )))] = [α1´1 φC (dC (( β´1 ( x )))] =

= [α1´1 dC (φC ( β´1 ( x )))] = [α1´1 dC ( β1´1 (φD ( x )))] = δ1 (φD ( x )), Jean-Louis Koszul
(1921–2018)
and we are through. o French mathematician
Example 5.9 Koszul complexes: The so-called Koszul complexes* form a large collection ˚ The Koszul complexes

of complexes. A Koszul complex depends on finite sequences f 1 , . . . , f n of ring elements. are named after
Jean-Louis Koszul who
The simplest ones involve merely one element f from the ring and is denoted K ( f ). introduced them in the
It is a two term complex with K1 = K0 = A, (and Ki ( f ) = 0 for all other i’s), and the study of Lie algebras;
and it seems, however,
differential is just multiplication by f :
that Adolf Hurwitz
about 50 years before
f
... /0 /A /A /0 /0 / ... applied several
particular cases in
algebraic geometry, and
The homology modules of K ( f ) that are not automatically equal to zero, are the one in some cases appeared
already in in Hilbert’s
degree zero H0 K ( f ) = A/( f ) A and the one in degree one H1 K ( f ) = (0 : f ), and they
famous paper xxxx.
enter into the exact sequence

f
0 (0 : f ) A A A/( f ) A 0.

When f is a regular element; i. e. a non-zero divisor, the Koszul complex K ( f ) provides

a free resolution of A/( f ) A.
The Koszul complex K ( f 1 , f 2 ) on two elements f 1 and f 2 has three non-zero terms;
so, displayed it appears as

d2 d1
0 A 2A A 0,
Adolf Hurwitz
(1859–1919)
where the repeating zeros are not shown. The first differential is given by the formula
German mathematician
d1 ( a1 , a2 ) = f 1 a1 + f 2 a2 and the second by d2 ( a) = ( f 2 a, ´ f 1 a). The homology module

10th January 2021 at 11:32am

Version 4.0 run 30
146 exact sequences

H2 (K ( f 1 , f 2 )) equals the annihilator of the ideal ( f 1 , f 2 ), while H0 (K ( f 1 , f 2 )) is the quo-

tient A/( f , g). The homology module H1 (K ( f 1 , f 2 )) equals the submodule consisting
of pairs ( a1 , a2 ) such that f 1 a1 = f 2 a2 module those of the form ( f 2 a, ´ f 1 a). In the case
when A is a factorial domain, H2 (K ( f 1 , f 2 )) = 0 unless f 1 = f 2 = 0, and H1 (K ( f 1 , f 2 ))
vanishes if and only if f 1 and f 2 are without common factors, and in that case, the
Koszul complex provides a free resolution of A/( f 1 , f 2 ).
The Koszul complex on three elements is slightly more involved and has four
non-zero terms
0 / A d2 / 3A d1 / 3A d0 / A /0

where d1 with respect to the standard basis is given by the matrix

 
0 f3 ´ f2
´ f 3 0 f1 
 
f2 ´ f1 0

where as the two other maps, d2 and d0 , have matrices

 
f1
( f 1 , f 2 , f 3 ) and  f 2 
 
f3

respectively. For general n the Koszul complexes become much more involved and are
best described by the use of exterior powers of maps and modules. K
Exercise 5.25 Assume that A is a factorial domain and that f and g are two non-zero
elements. Show that homology H1 K ( f , g) of the Kozul complex is monic submodule of
2A generated by the element ( gc´1 , ´ f c´1 ) where c = gcd( f , g). M

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 6

Tensor products

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 math-
ematician Hassler Whitney when working on the universal coefficient theorem in
algebraic topology introduced the tensor product of two abelian groups. Certain iso-
lated cases had been known prior to Whitney’s work, but Whitneys construct was
general, and it is the one we shall give (although subsequently 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, Woldemar Voigt
the concept of tensors is basically the same—the key word being bilinearity. (1850–1919)
German physicist
Bilinear maps
6.1 Introducing the tensor product (bilineære
avbildninger)
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
β( ax + by, z) = aβ( x, z) + bβ(y, z),
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, one says A-bilinear to
be reminded which ring is considered the base ring.
Hassler Whitney
A typical example from the world of vector spaces over a field k, would be a scalar (1907–1989)
product on a vector space V, and within the realm of commutative algebra, the products American
mathematician
 148 tensor products

of an A-algebra B is a good example; the multiplication map ( a, b) ÞÑ ab is an A-bilinear

map B ˆ B Ñ B.
Multilinear maps (6.1) There is naturally also the notion of multilinear maps, which involves more than
(multilineære
ś
two modules. In that case, the source of the map is a product i Mi of finitely many
avbildninger)
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

(6.2) 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.
The tensor product (6.3) The tensor product is a pair consisting of an A-module Mb A N together with an
(tensorproduktet) A-bilinear map τ : M ˆ N Ñ Mb A N that abide by the following rule:

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

γ : Nb M Ñ L suct that β = γ ˝ τ.

MˆN
τ / Mb A N In other words, every A-bilinear β factors linearly via τ, as expressed by the commutative
diagramme in the margin. And as usual with objects satisfying a universal property,
γ
% 
β the pair τ and Mb A N is unique up to a unique isomorphism.
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”.
(6.4) The construction starts out with the free A module F = A Mˆ N on the set M ˆ N.
ř
The elements of F are finite, formal linear combinations i ai ¨ ( xi , yi ) with xi P M,
yi P M and ai P A. In particular, every pair ( x, y) is an element of F, and by definition
these pairs form a basis for F. We proceed by letting G be the submodule of F generated
by all expressions either of the form

( ax + a1 x1 , y) ´ a( x, y) ´ a1 ( x1 , y), (6.1)

or of the form
( x, ay + a1 y1 ) ´ a( x, y) ´ a1 ( x, y1 ), (6.2)

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 ( x, y) will be denoted by xby. Having forced the two expressions (6.1) and

10th January 2021 at 11:32am

Version 4.0 run 30
 basic working formulas 149

(6.2) 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 relations hold true in Mb A N:

( ax + a1 x1 )by = a( xby) + a1 ( x1 by),

(6.3)
xb ( ay + ay1 ) = a( xby) + a1 ( xby1 ).

In other words, the map τ : M ˆ N Ñ Mb A N sending ( x, y) to xby is A-bilinear.

Proposition 6.5 The pair τ and Mb A N as constructed above satisfy the universal property
in paragraph 6.3; 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 4.36 on page 100), define an A-linear map βs : F Ñ L by sending
the basis-elements ( x, y) to the values β( x, y). Since β is bilinear, this map vanishes on
the submodule G. Consequently it factors through the quotient F/G = Mb A N and
thus gives the wanted map γ : Mb A N Ñ L.
Elements shaped like xby generate the tensor product, and because the value at
xby of any factorization of β is compelled to be β( x, y), the uniqueness of γ comes for
free. o
(6.6) Before leaving the details of the tensor product construction, there is one observa- MˆN / Mb A N
tion to be made. It will useful to reduce certain questions to questions about finitely γ
generated modules. % 
β

Assume given two A-modules N and M a sequence of elements x1 , . . . , xr from N L

ř
and a sequence of elements y1 , . . . , yr from M. Assume further that 1ďiďr xi byi = 0
in Nb A M. We contend that one may then find finitely generated modules N0 Ď N
ř
containing the xi ’s and M0 Ď M containing the yi ’s, so that the relation i xi byi = 0
ř
already holds in N0 b A M0 . Indeed, saying that the relation i xi byi = 0 holds in
Nb A M is to say that the element i ( xi , yi ) in A N ˆ M belongs to the submodule G, and
ř

as such it can be expanded as a finite combination of the generators of G; that is, of

elements described in (6.1) and (6.2). Letting N0 (resp. M0 ) be the submodule of N
(resp. M) generated by the xi ’s (resp. the yi ’s) and the finite number of x’s (resp. y’s)
ř
that appear in such an expansion, it follows that i xi byi = 0 in N0 b A M0 .

6.2 Basic working formulas

In this section we present a few principles and properties of the tensor products 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

10th January 2021 at 11:32am

Version 4.0 run 30
 150 tensor products

classes of modules, like cyclic modules and free modules, which behave particularly
well when exposed to a tensor product.
Decomposable tensors.
(6.7) For several reasons, tensors of the form xby deserve a special name; they are
Decomposable tensors dubbed decomposable tensors. Only in a very few highly special cases all elements in a
(dekomponerbare tensor product will be decomposable; the usual situation is that most are not (A simple
tensorer)
example is discussed in Problem 6.13 on page 160 below. See also Example 6.5 on
page 167). A general element in Mb A N may however, be expressed as a finite linear
ř
combination i ai ¨ xi byi of decomposable tensors since this is already true in the free
module F = A Mˆ N .
Consequently, if txi u is a set of generators of M and ty j u one for N, the decomposable
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.
(6.8) 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 6.3, certainly more suggestive than working with pairs ( x, y).
(6.9) 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
( ax )by = xb ( ay).
This is a simple consquence of the fundamental bilinear relations (6.3) on page 149; with
the notation of (6.3), just set x1 = y1 = 0.
Functoriality
Linear maps between A-modules are fundamental tools in algebra, and it comes as no
surprise that exploring how maps behave when exposed to tensor products occupies
a large part 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.
(6.10) 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 φ( x )by. Since the expression φ( x )by
depends bilinearily on x and y, this is a viable definition, and the resulting map is
naturally babtized φbid N .
It holds true that ψbid N ˝ φbid N = (ψ ˝ φ)bid N when ψ and φ are two composable
maps (the identity is obvious for decomposable tensors and thus holds true), and clearly
id M bid N = id Mb A N . Therefore the pair of assignments

M ÞÑ Mb A N and φ ÞÑ φbid N

10th January 2021 at 11:32am

Version 4.0 run 30
 basic working formulas 151

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

Proposition 6.11 The functor ´b A N is an A-linear functor. It transforms direct sums into
direct sums.

A formal consequence of a functor being additive is that it preserves finite direct sums
(as we established in Proposition 5.7 on page 126), but in general additive functors do
not commute with infinite direct sums, so a proof is needed for that case. One is given
in Exercise 6.1. Additive functors
Proof: Recall that in Section 5.4 on page 126 we introduced the notion of additive (additive funktorer)

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

( aφ + bψ)bid N = a ¨ φbid N + b ¨ ψbid N . (6.4)

This follows easily from how ´b A N acts on maps together with the basic bilinear
relations in (6.3) on page 149. Indeed, one finds

( aφ + bψ)bid N ( xby) =(( aφ( x ) + bψ( x ))by =

= aφ( x )by + bψ( x )by = a ¨ φbid N + b ¨ ψbid N xby,

and the two sides of (6.4) agree on decomposable tensors. Hence they are equal since
the decomposable tensors generate Mb A N. o
1
(6.12) The situation is completely symmetric in the two variables, so if ψ : N Ñ N is
a map, there is a map id M bψ from MbN to MbN 1 that sends xby to xbψ(y), and
naturally, one sets φbψ = (φbid N ) ˝ (id M1 bψ).

Some formulas
(6.13) 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 6.14 Suppose that M, N and L are modules over the ring A. Then we have the
following four canonical isomorphisms.
i) Neutrality: Mb A A » M;
ii) Symmetry: Mb A N » Nb A M;
iii) Associativity: ( Mb A N )b A L » Mb A ( Nb A L);
iv) Distributivity: ( M ‘ N )b A L » ( Mb A L) ‘ ( Nb A L).

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

10th January 2021 at 11:32am

Version 4.0 run 30
152 tensor products

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.
Neutrality: For the first formula notice that the product xa is bilinear in x and a and
therefore the map xba Ñ xa extends to an A-linear map Mb A A Ñ M. The map
M Ñ Mb A A sending x to xb1 is obviously an inverse.
Symmetry: In this case the short hand assignments are xby ÞÑ ybx and ybx ÞÑ xby.
They are bilinear in view of the fundamental relations (6.3), and hence yield maps
between Mb A N and Nb A M which obviously are mutually inverse.
Associativity: This case is more subtle than one should believe at first sight; the (very
short) shorthand definition of a map

( Mb A N )b A L Ñ Mb A ( Nb A L)

would be ( xby)bz ÞÑ xb (ybz), 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, for each element z from L, which we keep fixed, we define an A-linear map

ηz : Mb A N Ñ Mb A ( Nb A L)

by the assignment xby ÞÑ xb (ybz); this is legitimate since the expression xb (ybz) is
bilinear in x and y (the third variable z is kept fixed).
Obviously the map ηz (t) is linear in z and a priori being linear in t, it depends
bilinearily on t and z. We infer that sending tbz to ηt (z) induces a map ( Mb A N )b A L Ñ
Mb A ( Nb A L). On decomposable tensors this map behaves as wanted; that is, it sends
( xby)bz to xb (ybz).
A symmetric construction yields a map the other way which sends a decomposable
tensor xb (ybz) to ( xby)bz. Finally, these two maps are mutually inverses since they
act as inverse maps on the decomposables, and the decomposables 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 6.11 above. A vague
indication of an ad hoc proof (in the flavour of the preceding cases) is the short hand

10th January 2021 at 11:32am

Version 4.0 run 30
 basic working formulas 153

definition ( x, y)bz ÞÑ ( xbz, ybz). The salient point is to extend this to an isomorphism.
The detailed proof, formulated for general direct sums, is given in Exercise 6.1 below.
Be aware, that contrary to the tensor product ´b A N, the hom–functors Hom A (´, N ),
even though being additive, do not commute with infinite direct sums; see Exercise 5.7
on page 128.

Exercises
À
(6.1) Tensor products preserve arbitrary direct sums. Let A be a ring and iP I Mi a
direct sum of A-modules where the index set I is of any cardinality. Let N be another
A-module.
À À
a) Define the map τ : ( iP I Mi ) ˆ N Ñ iP I ( Mi b A N ) by sending (( xi )iP I , y) to
ř
i xi by. Show that τ is a well-defined bilinear map. Hint: Merely finitely many
of the xi ’s are non-zero.
À À
b) Show that τ induces an isomorphism ( i Mi )b A N » iP I ( Mi b A N ).
(6.2) Consider the four isomorphisms in Proposition 6.14 on page 151. Be explicit
about what it means that they are functorial (in every variable involved) and prove your
assertions.
M
(6.15) It is worth while to dwell a little on the associativity. Since the parentheses
are irrelevant, we may as well skip them and write Mb A Nb A L for Mb ( Nb A L) (or
for that matter for ( Mb A N )b 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 decomposable tensors xbybz as long as the
specifying expression is trilinear in x, y and z; the precise statement is a as follows:

Lemma 6.16 (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 φ( x, y, z) depends in a trilinear manner on
the variables. Then there is unique A-linear map φ : Mb A Nb A K Ñ L such that φ( xbybz) =
φ( x, y, z).

Proof: The argument is mutatis mutandis the same as we gave in the proof of Proposi-
tion 6.14 concerning the associative law: Fix an element z P K and consider φ( x, y, z);
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 (t) 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 ( Mb A N )b A K Ñ L.

10th January 2021 at 11:32am

Version 4.0 run 30
154 tensor products

And again, the map is unambiguously determined since its values are prearranged
on the decomposable tensors which generate ( Mb A N )b 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.

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 illustrate some of the different phenomena that can occur.
(6.17) Our first example is about the tensor product of two cyclic module and reads as
follows:

Proposition 6.18 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/(a + b). 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 ([ x ]a , [y]b ) from A/a ˆ A/b to the element [ xy]a+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 as 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 [ a]a b [b]b = ab ¨ 1b1, and clearly elements from both the ideals a and b
kill it; indeed
x ¨ 1b1 = ( x ¨ 1)b1 = 1b ( x ¨ 1).

So A/ab A A/b is a quotient of A/a + b and is thus squeezed between two copies of
A/a + b, by two maps one sending 1 to 1b1 and one 1b1 to 1. Hence all three must
coincide. o
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 annihi-
lators are comaximal. We also observe that an inclusion b Ď a yields an isomorphism
A/ab A A/b » A/a, in particular it holds true that A/ab A A/a » A/a.

Finite abelian groups

(6.19) 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 ( pµ , qν ) = Z—and we infer that

Z/pµ ZbZ Z/qν Z = 0.

10th January 2021 at 11:32am

Version 4.0 run 30
 basic working formulas 155

We also infer that if the two natural numbers satify µ ď ν it holds that true that

Z/pµ ZbZ Z/pν Z » Z/pµ Z

for any prime number p since ( pµ , pν ) = ( pmin(ν,µ) ). Together with the formulas from
Proposition 6.14 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

(6.20) In the second example we show that the tensor product of two free modules is
free.

Proposition 6.21 (Tensor product of free modules) 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 (i, j) 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 6.14 by a straightforward induction, however we offer
another simple generally valid proof.

Corollary 6.22 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 6.21: We contend that the set tei b f j u(i,j)P 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 so
we merely have to verify they are linearly independent.
Denote by ai ( x ) the i-th coordinate of an element x P E relative to the basis tei u; that
ř
is, one has x = i ai ( x )ei . Similarly, let b j (y) be the j-th coordinate of an element y P F.
All the ai ( x )’s and all the b j (y)’s depend linearly on their arguments.
For each pair of indices µ P I and ν P J the expression aµ ( x )bν (y) depends bilinearly
on x and y and therefore xby Ñ aµ ( x )bν (y) 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

10th January 2021 at 11:32am

Version 4.0 run 30
156 tensor products

6.3 Functorial properties

(6.23) The tensor product functor ´b A N and the functor Hom A ( N, ´) 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 6.24 (Adjointness) There is a functorial isomorphism

Hom A ( Mb A N, L) » Hom A ( M, Hom A ( N, L)).

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). The full name of the isomorphism in the theorem would be

θ M,N,L : Hom A ( Mb A N, L) Ñ Hom A ( M, Hom A ( N, L)),

but to save the presentation from notational obesity, we shall systematically abbreviate
it to θ (think of the parameters always being present but as darkened lights one can
turn on, if more clarity is needed).
Proof: The salient point is that Hom A ( M, Hom A ( N, L)) is canonically isomorphic to
the space of bilinear maps M ˆ N Ñ L; and once that is realized, the proposition
becomes just another reformulation of the universal property of the tensor product.
One may consider members of Hom A ( M, Hom A ( N, L)) as being maps Φ( x, y) de-
fined on M ˆ N: Assume Φ( x, y) is bilinear; when x is a specified member of M, the
corresponding map Φ( x, ´) from N to L is given as y ÞÑ Φ( x, y). The other way around,
when φ : M Ñ Hom A ( N, l ) is given, we put Φ( x, y) = φ( x )(y). The required linearities
and bilineraties of the involved maps are immediate to check.
And that’s it: according to the universal property the tensor product enjoys, any
such bilinear map Φ can be written unambiguously as Φ( x, y) = φ( xby) with a linear
map φ : Mb A N Ñ L.
Then to the functoriality; the hart of the matter is quit prosaic once one sees through
the formal underwood. Given γ : L Ñ L1 . Functoriality in this case means that

θ ˝ γ˚ = γ˚ ˝ θ

This equality boils down to the trivial and tautological observation that when evaluated
at a pair ( x, y), both sides equal γ(Φ( x, y)); indeed, the sole difference is that on the left
hand side we first apply γ to Φ and then consider the result first a function in x and
then in y; whereas on the right hand side order is opposite: we begin with considering
Φ first as a function in x and then in y for then to apply γ; of course, they are the same.
Next, suppose that α : M Ñ M1 is given; we are then to establish the equality:

θ ˝ (αbid N )˚ = α˚ ˝ θ. (6.5)

10th January 2021 at 11:32am

Version 4.0 run 30
 functorial properties 157

Indeed, both sides equal Φ(α( x ), y), and again this is just an expression for trivial fact
that the order of applying α and considering x and y as a first and a second variable,
does not make a difference. Functoriality in N is quite symmetric to functoriality in M;
and one has
θ ˝ (id M bβ)˚ = β˚ ˝ θ. (6.6)

when β : N Ñ N 1 an A-linear map. o

Right exactness
In analogy with the notion of left exactness, which we discussed in connection with
the hom-functors, a covariant and additive functor F between two module-categories* ˚ Or more generally,

Mod A and ModB is said to be right exact if it transforms exact sequences shaped like between twofunctors
Right exact abelian
categories
(høyre eksakte
M0 / M1 / M2 /0 funktorer)

into exact sequences shaped like

F ( M0 ) / F ( M1 ) / F ( M2 ) / 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 at
he end.
(6.25) Here it comes:

Proposition 6.26 (Right exactness) Given a ring A and an A-module N. The functor
´b A N is a right exact functor.

Our approach relies on Proposition 6.24 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. Let us also
mention that it is common usage to call N a flat A-module if the functor (´)b A N is Flat modules (Flate
exact; i. e. when it transforms injective maps into injective maps. moduler)

Proof: Let the exact sequence

M‚ : M0
α / M1 β
/ M2 /0

be given; the task is then to show that the sequence

αb id N βb id N
M‚ b A N : M0 b A N / M1 b A N / M2 b A N /0

10th January 2021 at 11:32am

Version 4.0 run 30
158 tensor products

is exact. Our tactics will be to apply the principle of left exactness of the hom-functor as
expressed in the proposition Left Exactness II (Proposition 5.17 on page 130), and in
fact, we shall do this twice. With that principle in mind, we start out by observing that
the sequence Hom A ( M‚ b A N, L) being exact for every A-module L will be sufficent,
and this sequence appears as the upper line in the following grand diagram.
( βb id N )˚ (αb id N )˚
0 / Hom A ( M2 b A N, L) / Hom A ( M1 b A N, L) / Hom A ( M0 b A N, L)

»
θ θ θ
  
0 / Hom A ( M2 , Hom A ( N, L)) / Hom A ( M1 , Hom A ( N, L)) / Hom A ( M0 , Hom A ( N, L))
β˚ α˚

The next step is to evoke Proposition 6.24 above and replace Hom A ( M‚ b A N, L) with
the complex Hom A ( M‚ , Hom A ( N, L)); 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 6.24 (Adjointness), the two squares
commute according to the functoriality properties stated in Proposition 6.24. o
(6.27) Proposition 6.18 on page 154 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 6.28 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 mb [ a] to [ am].
Proof: The starting point is the exact sequence

0 /a /A / A/a / 0,

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 6.1 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,

10th January 2021 at 11:32am

Version 4.0 run 30
 functorial properties 159

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 6.18 on page 154. K
(6.29) Even whole modules may succumb under the action of the tensor product; 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 6.30 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

Exercises
(6.3) Decompose Z/16ZbZ Z/36Z and (Z/6Z ‘ Z/15Z)bZ (Z/21Z ‘ Z/14Z) as
direct sums of cyclic groups.
(6.4) Let A be a ring and M an A-module. Show that Mb A ( A/ Ann M ) = M. 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.
(6.5) Show that Q/ZbZ Q = 0. Show further that it holds true that QbZ Q » Q, but
that one has Q/ZbZ Q/Z = 0. Hint: Proposition 6.30.
(6.6) Let a be a proper ideal in the ring A. Assume that M is an A-module for which
there is surjection φ : M Ñ A/a. Show that aM ‰ M. Hint: Tensorize φ by A/a.
(6.7) Let M be finitely generated non-zero module over the ring A.
a) Show that there is a prime ideal p in A and a surjective map M Ñ A. Hint:
Consider a generating set tx1 , . . . , xr u with r minimal and use that the quotient
module M/( x1 , . . . , xr ) is cyclic.
b) Show that it holds true that Mb n ‰ 0 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.

10th January 2021 at 11:32am

Version 4.0 run 30
160 tensor products

(6.8) Assume that G is a finite abelian group. Show that GbZ Q = 0.

(6.9) Let A be a domain contained in a field K and let M be an A-module. Assume
that Ann M ‰ (0). Prove that Kb A M = 0.
(6.10) Let A Ñ B be a surjective ring homomorphism. Show that for any two B-modules
M and N (which automatically are A-modules) it holds true that Mb A N = MbB N.
(6.11) Show that Z[i ]bZ Z[i ] is a free abelian group of rank 4 while Z[i ]bZ[i ]Z[i ] = Z[i ]
and is of rank two as an abelian group.
ˇ (6.12) Let E and F be vector spaces over the field k, and let τ P Ebk F be a tensor.
ř
Show tha one may express τ as a finite sum τ = i ei b f i where the ei ’s are linearly
independent vectors from E.
(6.13) Decomposable tensors and a saddle surface. Only in a very few highly special cases
will all elements in a tensor product be decomposable; the usual 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. This example also illustrate that beautiful geometry
can be unveiled by tensor product considerations.
The tensor product W is of dimension four with a basis tei be j u where 1 ď i, j ď 2.
Let xij be coordinates relative to this basis; that is, any vector v is expressed as v =
x11 ¨ e1 be1 + x12 ¨ e1 be2 + x21 ¨ e2 be1 + x22 ¨ e2 be2 ,
a) Establish that he decomposable tensors are shaped like

(ue1 + ve2 )b (se1 + te2 ) = us ¨ e1 be1 + ut ¨ e1 be2 + vs ¨ e2 be1 + vt ¨ e2 be2 ,

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 ).
(6.14) The rank stratification. Let V be a vector space over a field k of finite dimension.
The dual space V ˚ = Homk (V, k) 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 (e j ) = 0 when i ‰ j and φi (ei ) = 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 φ(v)w induces an isomorphism Γ : V ˚ bk W Ñ
Homk (V, W ).
b) Given a linear map θ : V Ñ W whose matrix relative to the bases tei u and t f j u
is ( aij ). Show that the element in V ˚ bW corresponding to θ equals i φi bθ (ei )
ř
ř ř
and that i φi bθ (ei ) = ij aij φi b f j .

10th January 2021 at 11:32am

Version 4.0 run 30
 change of rings 161

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 (V, W ) 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
(6.15) 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 u(i,j)P I ˆ J and tgk bhl u(k,l )PKˆ L
is given as the matrix (Φij Ψkl ) with rows indexed by pairs (i, j) P I ˆ J and colums by
(k, l ) P K ˆ L. This matrix is called the Kronecker product of Ψ and Φ. The Kronecker product
M (Kronecker-produktet)

6.4 Change of rings

Working in algebra one frequently finds it necessary, or at least highly desirable, to
change the ground ring. For instance, one extends the ground field to have sufficiently
many roots of polynomial at hand, or one reduces the ground ring modulo an ideal to
make arguments simpler. The tensor product is the perfect tool to carry the modules
one studies on to the new ground ring.
(6.31) Assume we are given a new ground ring B which is an algebra over the old one
A. In particular, B is an A-module. and we may form the tensor product Mb A B of
Leopold Kronecker
B with any A-module M. This tensor product will be a B-module in a canonical way: (1823–1891)
multiplication in "in the second variable" of Mb A B produces a B-module structure German mathematician
on Mb A B. Indeed, if b P B, the multiplication-by-b-map x ÞÑ bx is an A-linear map
[b] : B Ñ B, and therefore induces a map id M b [b]. On decomposable tensors it acts as
b ¨ xbc = xbbc. For a general tensor t we shall, of course, denote the action by b ¨ t or
simply by bt.
The module axioms come for free, it ensues directly from functoriality of ´b A B,
and in view of the tensor product being additive (Proposition 6.11 on page 151) and
right exact (Proposition 6.26 on page 157), we arrive at

Proposition 6.32 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 .

(6.33) Notice, that elements in B coming from A may be moved past the b-sign; i. e. if
a P A, 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 u( a) ¨ b, where u : A Ñ B is
the structure map and the product is the product in the ring B. Hence axbb = xbu( a)b

10th January 2021 at 11:32am

Version 4.0 run 30
162 tensor products

would be the correct way of writing. For instance, when A is of positive characteristic p,
the map u could be the Frobenius map a ÞÑ a p . Then a ¨ xbb = xba p b

Transitivity and adjointness

(6.34) Sometimes one wants to perform consecutive base changes, and in that respect
the tensor product behaves well. It is transitive in the following sense.

Proposition 6.35 (Transitivity) Assume that B is an A-algebra and C is a B-algebra. Then

there is a canonical isomorphism ( Mb A B)bB C » Mb A C as C-modules.

Proof: The short descriptions of the pair of inverse 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 acting as mutually inverses on decomposable tensors, they are
mutually inverse. o
(6.36) Changing the base ring preserves tensor products, but not hom-modules in
general. One has the following:

Proposition 6.37 Let B be an A algebra and M and N two A-modules. Then there is a
canonical isomorphism of B-modules

( Mb A N )b A B » ( Mb A B)bB ( Nb A B).

In other words, the functor (´)b A B takes tensor products into tensor products.

Proof: Two mutually inverse B-module homomorphisms are defined by the assign-
ments
xbybb ÞÑ xb1bybb
xbybbb1 Ð[ xbbbybb1
of decomposable tensors. The obey respectively a tri-linear and a quadri-linear require-
ment and thereby define genuine maps between the modules. o
(6.38) In addition to the base change functor given by the tensor product

(´)b A B : Mod A Ñ ModB ,

there is a functor going the other way, (´) A : ModB Ñ Mod A , whose action is kind of
trivial. If N is a B-module, NA is equal to N but regarded as an A-module—one just
forgets the B-module structure. The same happens for maps; they are kept intact, but
regarded as being merely A-linear. Such functors that throws away part of a structure,
are called forgetful functors in the parlance of category theory.
The point is that the tensor product (´)b A B and the forgetful functor (´) A are
so-called adjoint functors:

10th January 2021 at 11:32am

Version 4.0 run 30
 change of rings 163

Proposition 6.39 (Adjointness) Given an A-algebra B. Then there is a canonical isomor-

phism
HomB ( Mb A B, N ) » Hom A ( M, NA ).
functorial in both M and N.
Proof: The map from the left hand side to the right hand side is simply a “restriction”
map. It sends a given B-linear map φ : Mb A B Ñ N to φ( xb1), which cearly is A-linear
in x. To define a map the other way, let φ : M Ñ NA be A-linear. The expression
ψ( xbb) = bφ( x ) is A-bilinear in b and x 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):

ψ(b1 ¨ xbb) = ψ( xbbb1 ) = bb1 φ( x ) = b1 ¨ ψ( xbb).

Finally, and as usual, the two maps are mutially inverses, agreeing 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
transform free modules to free modules; indeed, if E » nA with a basis tei u correspond-
ing to the standard basis tei u of nA, it holds true that Eb A B » nB with a basis tei b1u
corresponding to the standard basis tei u (as a basis for nB this time).
(6.40) 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 that φ : E Ñ F is an A-linear map. Recall
that the entires of the matrix Φ = ( aij ) associated with φ are the coefficients in the
developments
ÿ
φ ( ei ) = a ji f j
j

of φ(ei ) in terms of the basis elements f j . Applying φbidB to the basis element ei b1,
yields
ÿ ÿ
φbidB (ei b1) = φ(ei )b1 = ( a ji f j )b1 = u( a ji ) ¨ f j b1
j

where u : A Ñ B denotes the structure maps as in paragraph 6.33 above. Hence the
matrix of φbidB is the matrix (u( a ji )) obtained by applying the structure map u to the
entries of ( a ji ).

Examples
(6.2) As an example consider the polynomial ring A = C[ x ] and let a P C be a complex
number. Furthermore, let B = C[ x ]/( x ´ a) » C, so that the structure map u is the

10th January 2021 at 11:32am

Version 4.0 run 30
164 tensor products

evaluation at a; i. e. u( P( x )) = P( a). If φ is a map between two free C[ x ]-modules with

matrix P = ( Pij ( x )) relative to some bases, the matrix of φbid[C] is just the matrix P
evaluated at a; that is, the matrix ( Pij ( a)).
(6.3) 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
(nij ), the matrix of φbid[F p ] relative to the corresponding bases will be ([nij ]) (where
[n] 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
φbid[Q], does not change the matrix. We merely consider the integers nij as being
rational numbers!
(6.4) Our third example will be of a rather different flavour than the two previous ones.
This time we let A be a ring of characteristic 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

6.5 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
it a ring structure compatible with the underlying C-module structure, thus making it a
C-algebra.
(6.41) On decomposable tensors the product ought to abide by the rule

abb ¨ xby = axbby, (6.7)

and indeed, this extends to a product on AbC B:

Proposition 6.42 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, a priori, the C-module structure on AbC B is in place, so only the ring
structure is lacking.
Proof: To give an argument that the assignment in (6.7) can be extended to give a
product of arbitrary tensors, we appeal once more to the principle of bilinearity (at the
end of paragraph 6.2 on page 150). In fact, we shall apply it twice, once for each factor,

10th January 2021 at 11:32am

Version 4.0 run 30
 tensor products of algebras 165

the basic observation being that the right hand side of (6.7) is C-bilinear both in ( a, b)
and ( x, y)
The first application of the principle shows that multiplication by abb for a fixed
pair ( a, b) extends to a C-linear map AbC B Ñ AbC B. This yields a map

η0 : A ˆ B Ñ HomC ( AbC B, AbC B)

that sends ( a, b) to the multiplication-by-abb-map; in other words, it holds true that

ř ř
η0 ( a, b)( i xi byi ) = i axi bbyi . In its turn, this map depends bilinearily on the pair
( a, b), and by a second application of the principle we arrive at a C-linear map

η : AbC B Ñ HomC ( AbC B, AbC B),

which on decomposable tensors behave as desired; i. e. η ( abb)( xby) = axbby. Subse-

quently the product of two arbitrary tensors s and t is defined as s ¨ t = η (t)(s). This
establishes the product in Cb A B, but there are of course, 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 η (t)(s) and η (s)(t) are bilinear in s and t and since they agree on de-
composable tensors, they are equal, and hence the product is commutative. One checks
associativity in a similar manner, but by using the Trilinearity Principle (Lemma 6.16
on page 153). The two expressions η (tu)(s) and η (t)(us) are both linear in each of
the variables s, t and u, and agreeing on decomposable tensors, they coincide; that is,
( t ¨ u ) ¨ s = t ¨ ( u ¨ s ). o
Exercise 6.16 Check that the distributive law holds in AbC B. M

The universal property

The tensor product AbC B enjoys a universal property that plays a paramount role in
algebraic geometry. In algebraic terms it reflects the geometric construction of so called
fibre products; a simple variant of which are the products X ˆ Y of two schemes. This is
a foundational construct on which the whole theory of algebraic geometry rests.
AbC B
(6.43) With the setting as in the previous section, there are two canonical C-algebra > `
ιA ιB
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. Aa =B
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 diagram in the margin commutes.
(Where the maps C Ñ A and C Ñ B are the structure maps defining the C-algebra C
structures.) Moreover, the tensor product is universal among C-algebras living in such
diagrams:

10th January 2021 at 11:32am

Version 4.0 run 30
 166 tensor products

Proposition 6.44 (The universal property) With the notation as just 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 η ˝ ι A = η A and η ˝ ι B = ηB .
A DO ] Proof: Indeed, the expression η A ( a)ηB (b) is C-bilinear, and according to the universal
η property of the tensor product it gives rise to a C-linear map AbC B Ñ D satisfying
ηA
AbC B
ηB η ( abb) = η A ( a)ηB (b). This is our desired map, but some checking remains to be done.
> ` Let us begin with verifying that η respects products. Since we know that η is linear, it
ιA ιB will suffice to do this for decomposable tensors:
Aa =B
η ( aa1 bbb1 ) = η A ( aa1 )ηB (bb1 ) = η A ( a)η A ( a1 )ηB (b)ηB (b1 ) = η ( abb)η ( a1 bb1 ),

where the two extreme equalities hold true by the very definition of η and the middle
C
one because both η A and ηB are ring maps.
Next, one has η ˝ ι A = η A and η ˝ ι B = ηB since η A (1) = ηB (1) = 1, and finally, that
η is unique follows, since it is determined by the values on decomposable tensors, and
these satisfy

η ( abb) = η (( ab1)(1bb)) = η ( ab1)η (1bb) = η A ( a)ηB (b).

Base change of polynomial rings and algebras of finite type

We continue with the stage set as above, with B being an A-algebra through the structure
map u : A Ñ B.
(6.45) 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 A[ x1 , . . . , xr ] Ñ B[ x1 , . . . , xr ] sending xi to xi and hence
ř ř
a polynomial α aα x α is mapped to α u( aα ) x α . Together with the inclusion of B as
the constants in B[ x1 , . . . , xr ], it induces, in view of the universal property of the tensor
product, a map of A-algebras A[ x1 , . . . , xr ]b A B Ñ B[ x1 , . . . , xr ].

Lemma 6.46 Let A be a ring and B be an A-algebra. Then the following equality holds true
A [ x1 , . . . , xr ]b A B = B [ x1 , . . . , xr ]

Recall the notation Proof: Considered as A-module, the polynomial ring A[ x1 , . . . , xr ] is free, and the
from monomials x α , with α running through all multi-indices, form a basis. The same holds
Paragraph 1.16 on for B[ x1 , . . . , xr ]; the monomials x α form a B-basis. The lemma then follows since these
page 21 where the monomials correspond. o
monomial
α Lemma 6.47 Let a Ď A[ x1 , . . . , xr ] be an ideal. Then the following equality holds true
x1 1 ¨ . . . ¨ xrαr was
denoted by x α , A[ x1 , . . . , xr ]/ab A B = B[ x1 , . . . , xr ]/aB[ x1 , . . . , xr ].
with α being the
multi-index
( α1 , . . . , αr ).
10th January 2021 at 11:32am
Version 4.0 run 30
 tensor products of algebras 167

Proof: Suppose that a is generated by a collection t f i u of polynomials. Then the

ideal aB[ x1 , . . . , xr ] will be generated by the images of f i ’s under the natural map
A[ x1 , . . . , xr ] Ñ B[ x1 , . . . , xr ] that is the polynomials obtained by applying the structure
map u to the coefficients. Since the tensor product is right exact, the lemma follows. o
In several contexts, when e. g. u is a canonical inclusion, as Z Ď Q or Q Ď C, the effect on
the generators is nil, one just considers the generators as being members of B[ x1 , . . . , xr ].
However, in other situations the effect on the f i ’s can be quite dramatic, in the worst
case they can even vanish. For examples this occurs if the structure map is the reduction
mod p-map u : Z Ñ Z/pZ = F p , and the coefficients of f i P Z[ x ] are all divisible by p.
Examples
(6.5) A particular application of the previous lemma is that the tensor product of two
polynomial rings (over the base ring) again is a polynomial ring, that is, one has
A [ x1 , . . . , xr ]b A A [ y1 , . . . , y s ] = A [ x1 , . . . , xr , y1 , . . . , y s ].
Polynomials give a striking example that the decomposable tensors are scarce. In A[ x, y]
for instance, the decomposable tensors are the polynomials that factor as a product
p( x )q(y) of which there are few compared to the total collection of polynomials.
(6.6) Be aware that the tensor product of two integral domains need not be an integral
domain. Even if both factors are fields, the tensor product might acquire zero-divisors.
A simple example is CbR C. The complex numbers C can be described as the quotient
R[ x ]/( x2 + 1) so by Lemma 6.47 above it holds that CbR C = C[ x ]/( x2 + 1). This latter
ring is isomorphic to the direct product C ˆ C, sending the residue class of a polynomial
p( x ) to the pair ( p(i ), p(´i )) yields an isomorphism, and the direct product has many
zero divisors.
K
Exercises
(6.17) Let k be a field and let f ( x ) be an irreducible polynomial in k[ x ] so that
K = k[ x ]/( f ( x )) is field. Moreover let L be field extension of k in which f splits as a
product f ( x ) = f 1 ( x ) . . . f r ( x ) of irreducible polynomials. Use the Chinese Remainder
ś
Theorem to prove that Kbk L » Li where Li is the field Li = L[ x ]/( f i ( x )).
(6.18) Assume that K is an algebraic number field; i. e. a finite extension of the 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 [K : Q] = 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 = Q[ x ]/( f ( x )) where
f ( x ) P Q[ x ] is irreducible polynomial.
(6.19) Show that A = R[ x, y]/( x2 + y2 ) is an integral domain, but that AbR C is not.
(6.20) Let φ : A Ñ B be a ring homomorphism and p Ď A a prime ideal. Show that

10th January 2021 at 11:32am

Version 4.0 run 30
168 tensor products

the fibre of φ˚ : Spec B Ñ Spec A over p is naturally homeomorphic to the spectrum

Spec k(p)b A B where k (p) is the fraction field of A/p; that is, k(p) = Ap /pA p
(6.21) 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 = K [ x ]/( x p ´ t) is a field and that LbK L has non-trivial
nilpotent elements.
(6.22) Let k be a field of positive characteristic p, and let a be an ideal in A =
ř
k[ x1 , . . . , xr ] generated by polynomials f i ( x ) = 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 (k[ x1 , . . . , xr ]/a)bk k F » k[ x1 , . . . , xr ]/aF where aF is the ideal generated by the
ř p
polynomials f i = α aiα x α
(6.23) Let C s be C equipped with the alternative algebra structure induced by complex
conjugation; i. e. a ¨ z = āz. Let f ( x ) P C[ x ] be a polynomial. Describe C[ x ]/( f ( x ))bC C.
s
When are the C-algebras C[ x ]/( f ( x ))bC C s and C[ x ]/( f ( x )) isomorphic?
M

6.6 Appendix: Flatness

Recall that an A-module M is flat if the functor (´)b A M is exact. This functor is right
exact, as we established in Proposition 6.26 on page 157, so it being exact amounts to
it sending injections to injections. There is however a small, but important, point: it
suffices to consider injections between finitely generated modules.

Proposition 6.48 Let A be a ring and M an A-module. Then the following three statements
are equivalent
i) M is flat;
ii) For all injective maps φ : N Ñ N 1 between two A-modules, the induced map
φbid M : Nb A M Ñ Nb A M is injective;
iii) For all injective maps φ : N Ñ N 1 where N and N 1 are finitely generated A-modules,
the map φbid M : Nb A M Ñ Nb A M is injective.

Proof: We already remarked that i) and ii) are equivalent, and trivially ii) implies iii),
and we only need to show that iii) implies ii).
So assume that φ : N Ñ N 1 is injective and that x = 1ďiďr xi byi P Nb A M maps
ř

to zero in N 1 b A M; that is, the relation i φ( xi )byi = 0 holds in N 1 b A M. To be able to

ř

apply iii) we’ll replace N and N 1 with appropriate finitely generated modules.
The substitute for N is easy, the submodule generated by the xi ’s (which are finite
in number) will do, so henceforth we may assume that N is finitely generated. To find
a replacement of N 1 we appeal to Paragraph 6.6 on page 149, which furnishes finitely
generated submodules N01 Ď N 1 and M0 Ď M such that i φ( xi )byi = 0 in N01 b A M0 .
ř

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: flatness 169

Now, clearly N11 = N01 + φ( N ) is a finitely generated submodule of N 1 in which φ takes

values, and moreover, by construction, i φ( xi )byi = 0 in N11 b A M. By assumption
ř
ř
assertion iii) holds, and we may infer that i xi byi = 0, and we are done. o

(6.49) When using Proposition 6.48 above to check that a module is flat, one may restrict
oneself to consider injections of ideals into the ring. Clearly if a Ď A, tensorizing the
sequence

0 /a ι /A / A/a /0

by a module M and remembering that A/ab A M = M/aM and Ab A M = M, one

obtains the exact sequence

ιb id M
ab A M /M / M/aM / 0,

where ιbid M has the effect abx ÞÑ ax. It ensues that when M is flat, the map ιbid M
will give an isomorphism ab A M » aM, and the theme of this paragraph is that the
converse also holds:

Proposition 6.50 The A-module M is flat if and only if ab A M » aM for all ideals a.
Moreover, it suffices to consider ideals that are finitely generated.

The modern standard proof of this result uses the derived functors of the tensor product
(i. e. the functors ToriA (´, ´)). However a direct proof it is not difficult, it involves
merely a diagram-hunt in a grand diagram; and of course, is nothing but the relevant
tiny part extracted from the big machine that makes the Tor-functors function. The
reduction to the case of finitely generated ideals follows mutatis mutandis the proof of
Proposition 6.48 above, and we’ll not repeat it.

Proof: Given an injection ψ : N 1 Ñ N between two finitely generated modules. We

may assume that N requires just one more generator than N 1 ; indeed, if this is not the
case, we may factor ψ as the composition of two injections N 1 Ñ N 2 Ñ N where the
numbers of generators of N 2 lies strictly between those of N 1 and N. Obviously ψbid M
is the composition of the tensorized injections N 1 b A M Ñ N 2 b A M Ñ Nb A M which
both may be assumed injective by induction on the difference of numbers of generators.
Hence we may concentrate on injections whose cokernel is cyclic, i. e. is shaped like
R/a.

10th January 2021 at 11:32am

Version 4.0 run 30
 170 tensor products

Consider the grand diagram

0O 0O 0O

p1
/ N1 /N / A/a /0
ψ
0 O O O
π1 p2

0 / L1 /Ld π2
/A /0
O O O

0 / D1 /D /a /0
O O O

0 0 0,
where the upper sequence is the one we started with and the rightmost one is the
canonical one. The hub of the diagram is the central module L, which is the so-called
Fibered products fibered product of N and A. It is the submodule of the product N ˆ A where the two
(fibrerte produkter) maps p1 and p2 coincide; that is, it is given as

L = t ( x, y) P N ˆ A | p1 ( x ) = p2 (y) u.

Filling out the rest of the diagram is easy, we just put L1 = ker π2 , D = ker π1 and
D1 = D X L1 . Notice that, and this is the salient point that makes the proof work,
because A is free, the middle horizontal sequence is split exact, and therefore remains
exact after being tensorized by M. The tensorized diagram appears as

0O 0O 0O

N 1 bO A M / Nb A M / M/aM /0
O O

0 / L1 b A M / Lb A M /M /0
O O O

D1 b A M / Db A M / ab A M /0
O

0
where the rightmost sequence is exact by the assumption that ab A M = aM and, as
observed above, the middle horizontal one is exact. The finish of the proof is either

10th January 2021 at 11:32am

Version 4.0 run 30
 _ _ _
 /® /¯
°

appendix: flatness 171

an adaption of the snake lemma, or a direct diagram-hunt, staring with an element

x P N 1 b A M mapping to zero in Nb A M and following the path indicated in the margin.

o
Example 6.7 Flat modules over pid’s : The following criterion for when modules over
a pid are flat is an illustrating example of the use of the criterion in Proposition 6.50.
It applies e. g. to Z-modules and modules over the ring k[ x ] of polynomials with
coefficients in field. Recall that module M over a ring A is torsion free if ax = 0 implies
that either x = 0 or a = 0 where x P M and a P A

Proposition 6.51 A module M over a principal ideal domain A is flat if and only if it is
torsion free.

Proof: An arbitrary ideal a in A will be principal, say generated by t. All elements

of the tensor product ab A M are then of the form tbx, and the map ab A M Ñ M acts
by sending tbx to tx. This map is obviously injective if and only if t does not kill any
non-zero element from M, and this holds for all ideals (t) if and only if M is torsion
free. o
K

Exercises
À
(6.24) Show that the direct sum iP I Mi of a familliy tMi uiP I of A modules is flat if and
only if all the Mi ’s are flat. Conclude that free modules are flat, and hence projective
modules will be flat as well.
(6.25) Show that polynomial ring A[t] is a flat algebra over A.
(6.26) Let A be a ring and assume given a short exact sequence

0 / M1 /M / M2 /0

of A-modules where M2 is flat. Prove that if one of M or M1 is flat, then the other one
is also flat. Give an example where M1 and M are flat, but M2 is not.
(6.27) Show that any direct product iP I Z of copies of Z, is flat over Z.
ś

(6.28) Let a Ď A be an ideal. Show that A/a is flat over A if and only if a2 = a.
M

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 7

Localization

Very early in our mathematical carrier, if not in our lives, we were introduced to fractions,
so we should be well familiar with their construction and have their properties in the
backbone. Anyhow, recall that to every pair of integers m and n with n ‰ 0, one
forms the “fraction” m/nTwo such fractions m1 /n1 and m/n are considered equal—that
is, have the same numerical 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:
formally, they are equivalence classes of pairs (m, n) with respect to the equivalence
relation above. The fractions obey the familiar rules for adding and multiplying we
learned in school, and they form a field, the field Q of rational numbers.
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 = (st)´1 .
Hence the natural notion is the concept of multiplicatively closed sets.
The process is indeed very general. It even accepts zero divisors as denominators,
but this makes it "murderous": if a is a zero divisor, say a ¨ b = 0 with 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 as far as inverting 0, which however will be devastating:
the entire ring collapses to the null-ring.
There are several ways of defining these localized rings. We shall follow most text
books and mimic the way one constructs the rational numbers. This is a direct and
intuitive construction which does not require much machinery.
The name “localization” has its origin in geometry where one considers rings of
functions, say continuous functions 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, becomes 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,
 174 localization

important in algebraic geometry, all algebraic functions on U arise in this way.

7.1 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 S´1 A.

Multiplicatively closed sets

(7.1) The notion of a multiplicatively closed set were already introduced in the formula-
tion of the Fundamental Existence Theorem (Theorem 2.49 on page 49), but we remind
Multiplicatively closed you that a subset S of a ring A is multiplicatively closed set, or for short also called a
sets (multiplikativt multiplicative set, if it contains the unit element, and the product of any two elements
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.

Examples
Examples of multiplicative sets abound, but for the moment we only mention a few of
the more important examples.
(7.1) The set of all powers a of an element in A; that is, the set S = t an | n P N0 u,
obviously is multiplicatively closed.
(7.2) The complement S = Azp of any prime ideal p in A is multiplicatively closed;
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. In fact an ideal a is prime if and only if the complement
Aza is multiplicatively closed. This argument generalizes immediately and shows
Ť
that the complement S = Az iP I pi of a union of prime ideals (finite or not) will be
a multiplicative set. Indeed, if st R S, there is at least one index i so that st P pi ;
consequently either s or t belongs to pi and hence not to S.
(7.3) It is fairly clear that the intersection of any family of multiplicatively closed sets
is multiplicatively closed, and one may therefore speak about the the multiplicative set
generated 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 its elements are all finite

10th January 2021 at 11:32am

Version 4.0 run 30
 localization of rings 175

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 .
(7.4) An occasionally useful multiplicatively set is the set S = 1 + a where a is an ideal
in A.
K
The construction of the localization S´1 A
The construction of the localized ring S´1 A follows grosso modo the same lines as the
construction of the rational numbers from the integers, but with a necessary twist due
to the possible presence of zero divisors in S. This has the serious consequence that the
cancellation law does not hold, which complicates the matter.
(7.2) A fraction has an enumerator and a denominator, and in our context the latter
will be confined to S. A natural starting point is therefore the Cartesian product A ˆ S
with the first factor representing all possible enumerators and the second all possible
denominators. The next step is to introduce an equivalence relation on A ˆ S telling
when two fractions are to be considered equal, and inspired by the case of rational
numbers, we declare the pairs ( a, s) and (b, t) to be equivalent when for some u P S, it
holds true that u( at ´ bs) = 0; the factor u is necessary to resolve problems possible
zero divisors would cause.
The salient point of the construction is that this is an equivalence relation. We shall
(temporarily) write ( a, s) „ (b, t) when ( a, s) and (b, t) are equivalent. This relation
is obviously reflexive and symmetric. To see it that is transitive, assume given three
pairs ( a, s), (b, t) and (c, u) such that ( a, s) „ (b, t) and (b, t) „ (c, u); transcribing the
equivalences into equalities in A, we find that v( at ´ bs) = 0 and w(bu ´ ct) = 0 for
some elements v, w P S. Since

t( au ´ cs) = u( at ´ bs) + s(bu ´ ct),

we infer that

vwt( au ´ cs) = wu(v( at ´ bs)) + sv(w(bu ´ ct)) = 0.

From S being multiplicatively closed it ensues that vwt P S, and so ( a, s) „ (c, u). We let
the localization S´1 A of A in S be the set of equivalence classes A ˆ S/ „, and denote by Localization
a/s or as´1 the class of the pair ( a, s). (lokalisering)

The next task is to give a ring structure to S´1 A, and there is no hocus-pocus about
that, it is done by the familiar formulas for adding and multiplying fractions we know
from school:
a/s + b/t = ( at + bs)/st a/s ¨ b/t = ab/st. (7.1)
However, some checking is necessary. First of all, the definitions in (7.1) are expressed
in terms of representatives of equivalence classes, and it is paramount that they do

10th January 2021 at 11:32am

Version 4.0 run 30
 176 localization

not dependent on which representatives are used. Secondly, the ring axioms must be
verified. Once we know the definitions are legitimate, this is just straightforward high
school algebra, safely left to volunteering students.
Let us check that the sum is well defined, leaving the product to the eager students.
Notice that it will suffices to vary the representatives of one of the addends at the time;
so assume that ( a, s) „ ( a1 , s1 ); i. e. it holds that u( as1 ´ a1 s) = 0 for some u P S. We find

s1 t( at + bs) ´ st( a1 t + bs1 ) = t2 ( as1 ´ a1 s)

which is killed by u. Therefore the sum does not depend on the representative of the first
addend, and by symmetry, neither on the representative of the second. Consequently
the sum is well defined.
Exercise 7.1 Show that the product is well defined. On a rainy day when all your
friends are away, verify the ring axioms for S´1 A. M
localization map (7.3) There is a canonical map ι S : A Ñ S´1 A, which is called the localization map. It is
(lokaliseringsavbild- nothing but the map that sends an element a in A to the class of the pair ( a, 1); that
ning)
is, a is mapped to the fraction a/1. By the very definition in (7.1) of the sum and the
product in the localization S´1 A, this a ring homomorphism. Seemingly, this map does
nothing to a, but kill it if necessary, and our next proposition details this murderous
behaviour of ι S .

Proposition 7.4 All elements in S´1 A are of the form a/s. It holds true that ι S ( a) = 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 S´1 A is an equivalence class a/s. The zero
element in S´1 A is represented by the pair (0, 1) and ι S ( a) by the pair ( a, 1). Hence
ι S ( a) = 0 if and only if s ¨ ( a ¨ 1 ´ 0 ¨ 1) = 0 for an s P S; that is, if and only if s ¨ a = 0 for
an s P S. o
(7.5) When S contains no zero divisors, the map ι S will be injective, and we shall
identify A with its image in S´1 A. This simplifies the notation significantly. We may
1
safely write a instead of ι S ( a) or a/1, and 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

(7.6) The localized ring S´1 A together with the localization map ι s are characterized by
˚ The verb extend is a universal property, which loosely may by phrased by saying that any ring map with
suggestive, but not source A that maps elements from S to units, extends* to a ring map from S´1 A.
entirely pertinent when
ι S is not injective.
Proposition 7.7 Let A be a ring and S Ď A a multiplicative subset. Assume given a ring
/B
φ
A O map φ : A Ñ B that sends S into the group of units in B. Then there is a unique map of rings
ψ ψ : S´1 A Ñ B so that φ = ψ ˝ ι S .
ιS %
S ´1 A

10th January 2021 at 11:32am

Version 4.0 run 30
 localization of rings 177

Proof: The sole way of realizing a ring map ψ : S´1 A Ñ B extending φ is to put

ψ( a/s) = φ( a) ¨ φ(s)´1 , (7.2)

and the gist of the proof is that this is a legitimate definition, i. e. that φ( a) ¨ φ(s)´1
is independent of the chosen representative ( a, s). But from a/s = a1 /s1 ensues that
t ¨ ( as1 ´ sa1 ) = 0 for an element t P S, and hence, since φ is map of rings, that

φ(t) ¨ φ( a) ¨ φ(s1 ) ´ φ( a1 ) ¨ φ(s) = 0.

The element φ(t) is invertible by assumption, and we conclude that φ( a) ¨ φ(s)´1 = A

φ( a1 ) ¨ φ(s1 )´1 . That φ = ψ ˝ ι S , is trivial, and that ψ is a homomorphism follows directly ιS ι1S

from (7.2) and the usual formulas for products and sums of fractions (the formulas in

S ´1 A / S ´1 A 1
(7.1)). o »
As any other object characterized by a universal property, the pair ι S and S´1 A
is unique up to an unambiguous isomorphism: if ι1S : A Ñ S´1 A1 solves the same
universal problem, one has ι1S = ψ ˝ ι S for a unique isomorphism ψ : S´1 A Ñ S´1 A1 .

Examples
(7.5) 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.
(7.6) An simple situation to have in mind is when A is contained in a field K. The
localized ring S´1 A is then just the subring A[s´1 |s P S] of K generated by the inverses
of members of S. The elements of this ring are all shaped like as´1 ; indeed, every sum
ř ´1
i ai si can be rendered on this form with s a common denominator of the terms. The
universal property of the localized ring S´1 A then immediately gives a map of rings
S´1 A Ñ A[s´1 |s P S] which one easily checks is an isomorphism using the description
of S´1 A in Proposition 7.4 on page 176.
(7.7) 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 = t pn | n P N0 u, the ring S´1 Z = Z[1/p] = 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 18).
In a similar vein, when p is a prime and S is the complement of the principal
ideal pZ, the localization S´1 Z will be the ring Z( p) = t a/b | a, b P Z, ( p, b) = 1 u of
rational numbers whose denominator is prime to p. We have already met these rings in
Example 2.23 on page 55.
(7.8) Consider the polynomial ring k[ x, y, z] in three variables over the field k. Here
we aim at describing the localization of k [ x, y, z] in the prime ideal (z) and showing

10th January 2021 at 11:32am

Version 4.0 run 30
 178 localization

that k [ x, y, z](z) = k ( x, y)[z](z) ; that is, the polynomial ring over the fraction field k ( x, y)
localized at the prime ideal (z). The principle from Example 7.6 takes effect, and we
can work with subrings of the rational function field k( x, y, z).
Let S be the subset of k[ x, y, z] whose members are the non-zero polynomials
involving only the variables x and y. It is obviously multiplicatively closed, and just
as obviously, it holds that S´1 k[ x, y, z] = k( x, y)[z]. Localizing both rings in the ideals
generated by z we obtain the desired equality, noticing that S Ď k[ x, y]z(z), so elements
in S are already invertible in k[ x, y, z](z) and thus (S´1 k [ x, y, z])(z) = k [ x, y, z](z) .
K
Functoriality
(7.8) The ring S´1 A depends of course on both A and S, so functoriality is naturally
formulated in terms of the pair ( A, S). Assume given another pair ( B, T ) and a map
A
φ
/B of rings φ : A Ñ B such that φ sends elements of S into T. Then there is induced a
map of rings φS,T : S´1 A Ñ T ´1 B satisfying φS,T ˝ ι S = ι T ˝ φ: since φ takes S into
ιS ιT
  T, the elements φ(s) become invertible in T ´1 B, and the universal property of S´1 A
S´1 A / T ´1 B guarantees that ι T ˝ φ extends to a uniquely defined map φS,T : S´1 A Ñ T ´1 A. This
φS,T
map simply sends a/s to φ( a)/φ(s).
(7.9) A particular case to notice is when A = B and φ = id A . If S Ď T, there is a
canonical map S´1 A Ñ T ´1 A which just interprets fractions a/s in S´1 A as a fractions
in T ´1 A. 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.
Exercise 7.2 Let S and T be two multiplicatively closed subsets of A. Let T 1 = ι S ( T ).
Prove that ι S ( T ) is a multiplicatively closed subset of S´1 A and that ι S ( T )´1 S´1 A is
canonically isomorphic with (ST )´1 A, where ST is the multiplicatively closed set whose
elements are products of elements from S and T. M
The field of fraction of a domain
(7.10) Every domain is contained in a field K ( A) canonically attached to A, which in
The field of fractions of some sense is the smallest field containing A. It is called the field of fractions of A and
a domain is constructed as the localisation Σ´1 A of A in the set Σ of non-zero elements from A;
(kvotientkroppen til et
område) the set Σ is multiplicatively closed since A is a domain. Elements of K ( A) are all of the
form ab´1 with a and b elements from A and b ‰ 0, and since zero divisors are absent
from A, it holds true that a/b = a1 /b1 if and only if ab1 = a1 b. In particular, we see that
K ( A) is a field: that a/b ‰ 0, means that a ‰ 0 and then b/a is defined and serves as
the inverse of a/b.
The field K ( A) is the smallest field containing A in the sense that if A Ď L with L a
field, the universal property of a localization furnishes an injectiv map from K ( A) into
L whose image is a copy of K ( A) lying between A and L.

10th January 2021 at 11:32am

Version 4.0 run 30
 localization of rings 179

Example 7.9 A familiar example of fraction fields is the field Q of rational numbers,
the fraction field of Z. Another is the field C( x1 , . . . , xr ) of rational functions in the
variables x1 , . . . , xr , which is the fraction field of the polynomial ring C[ x1 , . . . , xr ]. K
(7.11) Every multiplicatively closed set S not containing 0 is contained in Σ. Hence
there is a canonical map S´1 A Ñ K ( A), and since there are no zero divisors around, it
is an embedding. This map is as canonical as can be, simply, it sending as´1 to as´1 ,
but there is in the outset a subtle distinction between the two localizations as´1 and
as´1 ; they live formally in the distinct rings S´1 A and K ( A) = Σ´1 A. However, in the
sequel we gladly ignore these subtleties and consider the two rings to be equal: we
shall (when A is a domain) tacitly identify S´1 A with its alter ego in K ( A). Notice that
the maps φS,T from Paragraph 7.8, where T is another multiplicatively closed subset T
containing S, then become inclusions.
Exercises
ˇ (7.3) Let A be a domain with fraction field K, and let Σ = Azt0u. Let t1 , . . . , tn be
variables. Show that Σ´1 A[t1 , . . . , tn ] = K [t1 , . . . , tn ].
(7.4) Show that the field of fractions of the formal power series ring kJtK is the ring
k ((t)) of formal Taylor series in t. That is, the ring whose elements are series of the form
i
ř
iě´n ai t with ai P k and where addition and multiplication are the natural ones. The
addition is performed termwise and the multiplcation is the usual Cauchy product:
i
ř µ
ř ν
ř ř
µě´m aµ t ¨ νě´n bν t = iě´ max (n,m) µ+ν=i aµ bν t .
M
Saturation and equality
A very natural question is when will two multiplicative sets S and T give rise to the
same localization? As usual, care must be taken when saying that things are equal. In
the present context, the precise meaning of S´1 A and T ´1 A being the same, is that
there is an isomorphism θ : S´1 A Ñ T ´1 A compatible with the localization maps; i. e.
it satisfies θ ˝ ι S = ι T . However, when A is a domain and all localizations are considered
to be subrings of the fraction field K ( A), there is no abracadabra, and equal means
equal.
(7.12) Recall that a multiplicatively closed set is said to be saturated if with s every factor
of s belongs to S; that is, if s = uv lies in S, so does u (and hence by symmetry v). Every
multiplicative S has a saturation, a smallest saturated multiplicative set Sp containing S. The saturation of a
It consists of all factors of elements from S, or when described in symbols, it appears as multiplicative set
(metningen til en
Sp = t u P A | uv P S for some v P A u. multiplikativ mengde)

This set is multiplicatively closed: uv P S and u1 v1 P S implies uu1 ¨ vv1 P S since S is

multiplicatively closed, and it is evidently saturated since a factor of a factor is a factor!
Example 7.10 Let a be a natural number and let S be the multiplicatively closed set

10th January 2021 at 11:32am

Version 4.0 run 30
180 localization

S = t an | n P N0 u of powers of a. Moreover, let p1 , . . . , pr be the different prime factors

ν
of a. Then the elements of the saturation Sp are all integers of the form ˘p11 ¨ . . . ¨ prνr
with the exponents νi being arbitrary non-negative integers. K
In the context of the Fundamental Existence Theorem (Theorem 2.49 on page 49) you
were asked (Problem 2.23 on page 53) to show that a multiplicative set S is saturated if
and only if the complement is a union of prime ideals. Connecting up with that, we
have the following description of the saturation S,p or rather of its complement:

Proposition 7.13 Let S be a multiplicatively closed set in A. The complement of the saturation
Sp is the union of the prime ideals maximal subjected to not meeting S.

Proof: A member x of A does not belong to Sp if and only if the the principal ideal ( x )
does not meet S. Hence, according to the Fundamental Existence Theorem, x P AzSp if
and only if there is a prime ideal maximal subjected containing ( x ) and to not meeting
S. o
(7.14) The next lemma answers the retoric question at the top of this paragraph:

Lemma 7.15 The three following assertions hold true for multiplicatively closed subsets S and
T of a ring A:
i) The canonical map S´1 A Ñ Sp´1 A is an isomorphism;
ii) If S Ď T and the canonical map S´1 A Ñ T ´1 A is an isomorphism, then T Ď S; p
iii) There is an isomorphism between S´1 A and T ´1 A compatible with the localization
maps if and only if Sp = T.
p

Proof: We begin with proving i): Take an element from Sp´1 A. It is shaped like au´1
with u P S,p so that uv P S for some v P A. Hence au´1 = av(uv)´1 , and the map is
surjective. That an element as´1 P S´1 A maps to zero, means that a is killed by some u
p but u P Sp means that uv P S for some v. Hence (uv) a = 0 and a = 0 in S´1 A.
in S,
To establish ii) we first observe that the canonical map being injective means that
an element a P A satisfying ta = 0 for some t P T, also satisfies sa = 0 for some s P S.
Now, let t P T. The canonical map being surjective entails that t´1 lies in its image, i. e.
for an s P S it holds that t´1 = as´1 in T ´1 A. This means that u( at ´ s) = 0 for some
u P T. But by the initial observation, it ensues that v( at ´ s) = 0 for some v P S, and t is
a factor of the element vs which lies in S.
The last assertion iii) is a direct consequence of the two others. o

Exercises
ˇ (7.5) Show that Z[1/10] = Z[1/2, 1/5]. Generalize.
ˇ (7.6) Prove that any intermediate ring Z Ď A Ď Q is a localization of Z in a multiplicative
set S.
ˇ (7.7) Prove that the group of units A˚ in A is a multiplicative set. Show that the

10th January 2021 at 11:32am

Version 4.0 run 30
 localization of rings 181

localization maps ι S is an isomorphism if and only if S is a subset of A˚ .

ˇ (7.8) Consider be the polynomial ring A = k[ x1 , . . . , xn ] over the field k. Let S be the
set of non-zero polynomials in A that depend only on the first r variables; i. e. those
on the form p( x1 , . . . , xr ). Show that S is multiplicatively closed and that S´1 A =
K [ xr+1 , . . . , xn ] where K is the field k( x1 , . . . , xr ) of rational functions.
ˇ (7.9) Let A be any ring. Describe the saturation of the set t1u.
ˇ (7.10) Both the set of even and the set of odd (non-zero) numbers are multiplicatively
closed . What are their saturations? Let p be a prime. Verify that the set of integers
congruent one mod p constitute a multiplicative set. What is the saturation?
ˇ (7.11) Describe all the saturated multiplicative sets in Z. Generalise to any factorial
domain A.
ˇ (7.12) Given an ideal a in A. Describe the saturation of the multiplicative set 1 + a.
(7.13) Let M be an A-module. A ring element x P A is called a zero divisor on M if
xm = 0 for some non-zero m P M. Show that the set S of non-zero divisors on M form
a saturated multiplicatively closed set. Hence the set of zero divisors Z ( M ) on M is the
union of prime ideals. Show that Z ( M ) is the union of the ideals maximal among the
prime ideals not meeting S. These are called the maximal associated ideals to M.
(7.14) When L is a set of primes in Z (finite or infinite), let Z( L) denote the localization
in all primes outside L; that is, we put Z( L) = pP L Z( p) . Show that there is a natural
Ş

isomorphism Z( L) bZ Z( L1 ) » Z( LX L1 ) .
M

Ideals and localization

There is a strong relationship between ideals in A and ideals in S´1 A. It relies on the
two functorial ways of transporting ideals to and fro along ring maps (as explained in
Section 2.9 on page 32). On the one hand any ideal b in S´1 A may be pulled back to
1
give an ideal ι´ S (b) in A (when ι S is injective and A is considered to be contained in
S´1 A, this is just the intersection A X b). On the other hand, one may extend ideals in A
to ideals in S´1 A: the extension of a Ď A to S´1 A is the ideal aS´1 A in S´1 A generated
by ι S (a). To simplify the notation we shall write S´1 a for it.
(7.16) The extension map I( A) Ñ I(S´1 A) preserves inclusions (but not strict inclu-
sions) and as extension maps always do, it preserves products and sums of ideals (see
Proposition 2.13 on page 33).
In general the extension map from I( A) to I(S´1 A) is not injective. For instance, it
may happen that S X a ‰ H, in which case the extension S´1 a will contain an element
invertible in S´1 A and consequently be equal to S´1 A; 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 S´1 A. So, some ideals are blown up to S´1 A (those

10th January 2021 at 11:32am

Version 4.0 run 30
182 localization

meeting S) and some collapsed to zero (those contained in ker ι S ). See Exercises 7.17
and 7.18 below for a discussion of when ideals have coinciding localizations.
Example 7.11 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 = ( pm )
extend to the entire ring S´1 Z. This also illustrates that forming extensions does not
commute with forming infinite intersections; indeed, one has m ( pm ) = 0 whereas
Ş

m p S Z = S Z.
Ş m ´1 ´1 K
´ 1 ´ 1 ´1
(7.17) The extension map is however surjective. Any ideal b Ď S A equals S ι S b;
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 belongs to b, the element a
1 ´1 ´1
belongs to ι´
S (b) as ι S ( a ) = b ¨ s, and therefore b lies in the extension S ι S (b).

Proposition 7.18 (Ideals in localizations) The extension map from the lattice I( A) to
the lattice I(S´1 A) given by a Ñ S´1 a is surjective. It preserves inclusions, products, sums
1 ´1
and finite intersections. One has ι´
S ( S a) = t a P A | sa P a for some s P S u, and for ideals
1
b Ď S´1 A it holds true that b = S´1 (ι´S b).

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, so we concentrate on the finite intersections; and
of course, the case of two ideals will suffice.
Clearly S´1 (a X a1 ) Ď S´1 a X S´1 a1 . So assume that b P S´1 a X S´1 a1 . One may then
express b as b = a/s = a1 /s1 with elements a and a1 from respectively a and a1 . This
yields t( a ¨ s1 ´ a1 ¨ s) = 0 for some t P S. But then tsa1 = ts1 a P a X a1 and consequently
b = tsa1 /tss1 lies in S´1 (a X a1 ). o
(7.19) Prime ideals behave more lucidly under localization than general ideals. Either
they blow up and become equal to the entire localized ring S´1 A, or they persist being
prime. Moreover, every prime ideal in S´1 A is of the shape pS´1 A for an unambiguous
prime ideal p of A, so that two different prime ideals persist being different unless both
blow up. One has:

Proposition 7.20 (Prime ideals in localizations) Assume p is a prime ideal in the ring
A and S is a multiplicative subset of A. Extending p to the localization S´1 A has two possible
outcomes. Either S´1 p = S´1 A, and this occurs if and only if p X S ‰ H, or otherwise S´1 p is
1 ´1
a prime ideal and ι´
S ( S p) = p.

Proof: If S X p ‰ H the ideal p blows up to the entire ring in S´1 A; that is, it holds that
S´1 p = S´1 A. If not, S´1 p is a prime ideal; indeed, suppose that bb1 P S´1 p 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 ( a) = 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

10th January 2021 at 11:32am

Version 4.0 run 30
 localization of rings 183

Proposition 7.21 (Prime ideals in localizations II) The prime ideals in the localization
S´1 A are precisely the ideals of the form S´1 p for p a prime ideal in A not meeting S. The
prime ideal p is uniquely defined.
In other words, extension and contraction of ideals are mutually inverse maps between
the sets of prime ideals in S´1 A and of prime ideals in A not meeting S.
Proof: By the previous proposition, the ideals S´1 p are all prime, so let q be a prime
1
ideal in S´1 A. Then p = ι´ S q is prime, and by the last sentence in proposition 7.18
above it holds that q = S´1 p. o
(7.22) Localization commutes as we have seen, with several processes ideals can be
exposed to, like forming products, intersections and sums. In this paragraph we treat
the case of radicals, and as a further example, transporters are covered in Problem 7.23
below.

Proposition 7.23 (Radicals localize) Let a be an ideal in A and S Ď A a multiplicative

set. Then it holds true that
‘ ´1
S a = S´1 a.
‘

Proof: If xs´1 P S´1 a with x ν P a and s P S, it holds that ( xs´1 )ν = x ν s´ν P S´1 a,
‘

and so xs´1 lies in S´1 a. For the other inclusion, assume that xs´1 P S´1 a, which
‘ ‘

means that for some natural number ν it holds that ( xs´1 )ν = at´1 with t P S and a P a.
Hence ( xt)ν = sν atν´1 P a, and consequently x P a and xs´1 P S´1 a.
‘ ‘
o
Exercises
(7.15) Show that if S´1 a = S´1 A, then the same holds for all powers aν of a.
(7.16) Let p and q be different prime numbers and let S be the multiplicative set
S = t pn | n P N0 u. Describe Z X ( pq)S´1 Z.
(7.17) Let S be multiplicatively closed in the ring A and let a Ď A be an ideal.
a) Show that the ideals (a : s) when s runs through S form a directed family of
ideals; hence their union is an ideal;
1 ´1
b) Show that sPS (a : s) = ι´
Ť
S ( S a);
c) Show that sPS (a : s) is maximal among the idelas b such that S´1 b = S´1 a.
Ť

(7.18) Let a and b be two ideals and S a multiplicative set in the ring A. Show that
S´1 a = S´1 b if and only if for each pair of elements a P a and b P b there are elements s
and t in S such that sa = tb.
(7.19) Suppose that p is a prime ideal and that a an ideal contained in p. Show that
1 ´1
ι´S ( S a) Ď p.
(7.20) Let j : Spec S´1 A Ñ Spec A be the map induced from the localization map
ι S : A Ñ S´1 A.
a) 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.

10th January 2021 at 11:32am

Version 4.0 run 30
184 localization

b) Show that j is a homeomorphism onto its image (when the image is endowed
with induced topology). Show that the image j(Spec A) equals the intersection
of all the open sets containing it.
(7.22) 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.
(7.23) Let a and b be two ideals in A, Prove that S´1 (a : b) = (S´1 a : S´1 b).
M

The local ring at a prime ideal.

A few ways of forming rings of fraction are omnipresent in algebra and algebraic
geometry and they are used over and over again. The most prominent one is the
localization Ap of A at a prime ideal p.
The complement S = Azp of a prime ideal p is as we have seen multiplicatively
closed, and the corresponding localized ring is written as Ap . The elements are fractions
a/b with b R p. The ring Ap will be a local ring whose only maximal ideal is the
expansion of p; that is, the ideal pAp :

Proposition 7.24 The localisation Ap is a local ring with maximal ideal pAp . The assignment
q ÞÑ qAp is a one-to-one correspondence between prime ideals in Ap and prime ideals q in A
1
contained in p. The inverse corresprondence is the pull-back p ÞÑ ι´
S (p).

Notice, that when all zero-divisors of A lie in p, so that the localization map is injective,
and we identify A with its image in Ap , the inverse correspondence will just be
p ÞÑ p X A.
Proof: This is nothing but Proposition 7.21 on the previous page, 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 not meeting S, that is, contained in p. Moreover 7.21 also tells us that
q = ι´1 (qAp ). o

Examples
(7.12) If p is a prime number, the localized ring Z( p) at the maximal ideal generated
by p consists of the rational numbers which when written in lowest terms, have a
denominator relatively prime to p. The maximal ideal in Z( p) is generated by p, and
the residue field is the field F p with p elements.
(7.13) If p( x ) is an irreducible polynomial in the polynomial ring k[ x ] over a field k,
the ring k [ x ]( p( x)) is the subring of k( x ) consisting of the rational functions whose
denominator when written in lowest terms does not have p( x ) as factor. The maximal
ideal is generated by p( x ), and the residue field will be the field obtained by adjoining
a root of p( x ) to k. In particular, if p( x ) is linear, say p( x ) = x ´ a, the elements of

10th January 2021 at 11:32am

Version 4.0 run 30
 localization of rings 185

k [ x ]( p( x)) are the rational functions whose denominator does not vanish at a. The residue
field will be k itself.
(7.14) We continue working with a prime ideal p in a ring A. The quotient A/p is
naturally contained in the residue field k(p) = Ap /pAp , as the inverse image of pAp
equals p, and since elements in the latter are all classes of the form [ as´1 ] with a P A
and s P Azp, it ensues that the residue field k(p) equals the fraction field of A/p.
K
(7.25) Although the prime ideal pAp pulls back to the prime ideal p, powers of pAp do
not always pull back to powers of p. However, there is always a ring map

A/pr Ñ Ap /pr Ap (7.3)

for the simple reason that pr maps into pr Ap , but it might very well fail both to be
injective and surjective. That surjectivity may fail is not unexpected, since, for instance,
Ap /pAp will be a field whereas A/p is not unless p is maximal. That injectivity may
fail, is certainly more subtle. It leads to the introduction of the so called symbolic Symbolic powers
powers p(r) = pr Ap X A of p (they will be treated more thoroughly in Exercise 10.11 on (symbolske potenser)

page 275). The kernel of the map in (7.3) equals the quotient p(r) /pr , so the map is not
injective precisely when the two differ. Below (Example 7.15) we shall give an example
of a symbolic square p(2) being different from the plain square p2 . When p is a maximal
ideal, however, the map in (7.3) will always be an isomorphism.

Lemma 7.26 Let m be a maximal ideal in A. Then mr Am X A = mr . Moreover, the canonical

map A/mr Ñ Am /mr Am is an isomorphism.

Proof: This hinges on a classic from algebra, namely the formula

(1 ´ x )(1 + x + . . . + xr´1 ) + xr = 1,

valid for r ě 1 and in any ring. It implies that elements in A/mr not lying in the
maximal ideal are invertible. Indeed, if s P Azm, we may find an element t P Azm such
that x = 1 ´ st P m (simply because A/m is a field: lift the inverse of [s] in A/m to A).
With that x the formula above gives
ÿ
st ¨ (1 ´ st)i ´ 1 P mr .
0 ď i ď r ´1

Consequently the class of st in A/mr , and hence a fortiori the class of s, is invertible.
Now, take any element a P mr Am X A; it may be written as a = xs´1 with x P mr
and s R m. Hence sa = x P mr , but since s is invertible mod mr , it ensues that a P mr .
Indeed, a = (1 ´ ts) a + tx P mr where t is an element in A such that 1 ´ ts P mr .

10th January 2021 at 11:32am

Version 4.0 run 30
186 localization

In the same vein, the map in the lemma is surjective because with t as above, we
find as´1 ´ at P mr Am , and the element at of A maps to as´1 mod mr . o
Example 7.15 A symbolic square that differs from the plain square: Lemma 7.26 does not
always hold for prime ideals that are not maximal; the ideal of a line in the cone over
a plane quadric is among the simplest examples: Let k be a field and let A = k [ x, y, z]
with constituting relation z2 = xy. The ideal p = (z, x ) is a prime ideal, since putting x
and z to zero induces a map A Ñ k[y] whose kernel is p. Neither x R p nor y R p, since
in the polynomial ring k[ X, Y, Z ] no non-zero linear form lies in ( XY, XZ, X 2 , Z2 ´ XY ),
simply for degree reasons, so in particular the map just defined is surjective.
In the local ring Ap the element y is invertible and we therefore have
p2 Ap = (z2 , zx, x2 ) Ap = ( xy, zx, x2 ) Ap = ( x ) Ap ,
whereas in A we have p2 = ( xy, zx, x2 ). We already observed that x R ( xy, zx, x2 ) so
p2 Ĺ p2 Ap X A. K
Exercises
(7.24) Let m be a maximal ideal in the ring A and let r be a natural number. Show
that the localization map A Ñ Am induces an isomorphism between mr /mr+1 and
mr Am /mr+1 Am as vector spaces over A/m.
Ş
(7.25) Let A be a domain with fraction field K. Show that A = m Am . Hint: For any
x P K not in A prove that the ideal t y P A | yx P A u can not be a proper ideal.
M
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 of fractions S´1 A 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 D ( f ) of Spec A.
There is a natural isomorphism between A[ x ]/( x f ´ 1) 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 A[ x ]/( x f ´ 1), the unversal property of A f furnishes an inverse.
This makes the notation A[ f ´1 ] 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 A[ f ´1 ] = 0.
Example 7.16 It is worthwhile mentioning a concrete example. Consider the ring
A = C[z] 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(Ω) of functions holomorphic in
any domain Ω of the complex plane containing a. The localized ring O(Ω)z´a has the
functions meromorphic in Ω with at most a pole at a as elements. K

10th January 2021 at 11:32am

Version 4.0 run 30
 localization of rings 187

The total ring of fractions

The set S of non-zero divisors in ring A is closed under multiplication; indeed, if s and t
are non-zero divisors and st ¨ a = 0 with a ‰ 0, it would follow that ta ‰ 0 which would
contradict that s is a non-zero divisor. The set S is even a saturated multiplicative set The total ring of
since a zero-divisor can not be factor of a non-zero-divisor. fractions of a ring
(kvotientringen til en
The localization of A in S is denoted by K ( A) and is called the total ring of fractions ring)
of A. When A is an integral domain, S = Azt0u, and all non-zero elements become
invertible in K ( A). Consequently K ( A) is a field; it is called the field of fractions of A, a The field of fractions
construct we already met in Paragraph 7.10 on page 178. (kvotientkroppen)

The ring K ( A) 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 K ( A) is injective
since by their very nature non-zero divisors do not kill non-zero elements.

Proposition 7.27 The total ring of fractions K ( A) of a ring A has the property that every
non-zero divisor is invertible. The natural map A to K ( A) is injective. Moreover, K ( A) is a
field if and only if A is an integral domain.

(7.28) The total rings of fractions of a certain class of reduced rings—recall that A being
reduced means it is without non-zero nilpotent elements—has a closer description. The
class of rings we have in mind are the reduced rings with a finite number of minimal
‘
primes. Since the radical (0) equals the intersection of the minimal prime ideals, in
these rings the zero ideal is the intersection of finitely many prime ideals; that is, one
has
(0) = p1 X . . . X pr , (7.4)

where the pi are distinct prime ideals. This is a large class of rings encompassing all
reduced Noetherian rings (a class of rings soon to be introduced). The pi ’s occurring in
(7.4) being minimal the intersection is irredundant; i. e. the intersection of all but one of
the pi ’s is never zero. An important observation is that the set of zero-divisors in A is
precisely the union p1 Y . . . Y pr of the pi ’s (see Exercise 7.30 below).

Proposition 7.29 (Total ring of fractions of reduced rings) Let A be a ring without
non-zero nilpotent elements.
i) The local ring Ap at a minimal prime p is a field;
ii) When A has only finitely many minimal prime ideals p1 , . . . , pr , the total ring of
fraction is a product of fields; i. e. K ( A) = Ap1 ˆ . . . ˆ Apr .

Let us first remark that the statements are not generally true for non-reduces rings see
Example 7.17 below of what may happen if A is not reduced.
The closed subsets Spec A/pi are the irreducible components of Spec A; that is,
Spec A = Spec A/p1 Y . . . Y Spec A/pr , and the proposition says that the total ring of

10th January 2021 at 11:32am

Version 4.0 run 30
188 localization

fractions of Spec A equals the product of the fraction fields of the irreducible compo-
nents of Spec A.
Proof: We begin with proving i) which is the easier: If p is a minimal prime, the local
ring Ap has p as its sole prime ideal simply because prime ideals in Ap correspond to
prime ideals A contained in p. Radicals localize well (7.23 on page 183), and A being
reduced, we infer that Ap is reduced as well. Because the radical of Ap equals pAp , it
follows that pAp = 0, and consequently, Ap is a field.
Statement ii) is a little more elaborate. For each index i the localization map A Ñ Api
extends to a map K ( A) Ñ Api because no non-zero divisor lies in pi , and recollecting
these maps we obtain a map

θ : K ( A) Ñ Ap1 ˆ . . . ˆ Apr .

It sends as´1 to the string ( as´1 , . . . , as´1 ). An element a from A maps to zero in Api
precisely when a is killed by an element not in pi , and thence a P pi . If this occurs for all
indices i, the element a lies in the intersection of the pi ’s and is therefore equal to zero
(the intersection of the pi ’s vanishes as A is reduced). This proves that θ is injective.
To see that θ is surjective requires some further effort. We begin with choosing an
element si for each i such that si R pi , but si P p j when j ‰ i. Then si becomes invertible
in Api , but maps to zero in Apj when j ‰ i; indeed, each si is killed by any non-zero
Ş
element in j‰i p j , and there are such since the intersection in (7.4) is irredundant.
Now, we come to the salient point of the proof: For any choice of elements c1 , . . . , cr
ř
with each c j not in p j , the combination x = j c j s j is a non-zero divisor. Indeed, if x
ř
belonged to pi , it would ensue from s j P pi when j ‰ i that ci si = x ´ j‰i c j s j belonged
ř
to pi , which is absurd since neither ci nor si lies there. Consequently j c j s j is invertible
in K ( A), and in view of si mapping to zero in Apj when j ‰ i, one finds
ÿ
1 ´1
θ ( c j s j ) ´1 = ( c ´ ´1 ´1


1 s1 , . . . , cr sr ).
j

Finally, if a1 , . . . , ar are are arbitrary elements A one arrive at

ÿ ÿ
1
θ ( si ai )( c j s j )´1 = ( a1 c´ ´1


1 , . . . , ar cr ),
j j

showing that θ is surjective. o

To describe the total quotient rings of of rings that are not reduced is in general
much more involved, as indicated by the following example. For a certain class of
rings—those “without embedded components”—a description similar to the one in the
proposition holds. An indication is given in Exercise 7.26 (7.31) below.
Example 7.17 Let B = k[ X, Y ]/( X 2 , XY ), and as usual, let the lower case versions x and
y denote the classes of X and Y in B. We contend that m = ( x, y) consists of all the

10th January 2021 at 11:32am

Version 4.0 run 30
 localization of rings 189

zero-divisors in B, and the total quotient ring K ( B) is therefore given as the local ring
K ( B) = Bm .
Clearly both x and y are zero divisors, so all members in m will be too. Assume then
that ab = 0 with neither a nor b being zero. Consider the classes [ a] and [b] in B/m = k.
Their product is zero, hence at least one of them vanishes, say [ a], which means that
a P m. Assuming that b R m, after possibly rescaling b, one may write a = cx + dy
and b = 1 + ex + f y, where d and f do not belong to the ideal ( x ) B. Then, using the
relations x2 = xy = 0 which hold in B, one finds

0 = ab = (cx + dy) ¨ (1 + ex + f y) = cx + dy + f dy2 = cx + yd(1 + y f ),

and hence yd P ( x ) B. But because B/( x ) B = k[ X, Y ]/( X 2 , XY, X ) = k[Y ], the ideal ( x ) B
is a prime ideal, and one infers that either y P ( x ) or d P ( x ), which is a contradiction.
The ring Bm has two prime ideals, the maximal ideal ( x, y) whose elements consti-
tute all zero-divisors, and a sole minimal prime ideal ( x ) whose elements are all the
nilpotents of B. K
Exercise 7.26 With reference to the example, show that B( x) equals the rational function
field k(Y ). M

Exercises
(7.27) Let n be a natural number. Determine the total quotient ring of Z/nZ.
(7.28) Let A be any ring. Show that the nil-radical of K ( A) is equal to the extension of
the nil-radical of A.
(7.29) Let A be a ring.
a) Show that if the elements of A are either zero divisors or invertible, then
A = K ( A ).
b) If A has only one prime ideal, prove that K ( A) = A.
c) Let A be a direct product (indexed by a set of any cardinality) of rings each
having only one prime ideals. Prove that K ( A) = A.
Ş
(7.30) Assume that A is a reduced ring so that (0) = iP I pi where the intersection
Ť
extends over the minimal prime ideals of a ring A. Show that the union iP I pi equals
Ş
the set of zero divisors in A. Hint: Observe that pi kills j‰i p j .
‘
(7.31) Let t A be a ring. Assume that (0) = p1 X . . . X pr is an irredundant intersection
of prime ideals. Assume further that the set of non-zero divisors of A equals the union
p1 Y . . . Y pr . Show that the total ring of fractions K ( A) decomposes as the direct product
K ( A) = Ap1 ˆ . . . ˆ Apr . Hint: Be inspired by the proof of the second assertion in
Proposition 7.29.
(7.32) Let k be a field and consider the polynomial ring A = k [ x1 , . . . , xn ]. Let r ă n be
a natural number. Let S be the subset of A of polynomials in the variable xr+1 , . . . , xn .

10th January 2021 at 11:32am

Version 4.0 run 30
190 localization

Show that S is multiplicatively closed and that S´1 k[ x1 , . . . , xn ] = K [ x1 , . . . , xr ] where

K = k( xr+1 , . . . , xn ) is the field of rational functions in the variables xr+1 , . . . , xn .
(7.33) Let A be a domain with quotient field K. Denote by S the multiplicative set Az0
of non-zero elements in A. Show that S´1 A[ T ] = K [ T ].
M

7.2 Localization of modules

There is also a procedure to localize an A-module M in a multiplicatively closed set S
closely resembling the way the fraction ring S´1 A was constructed, and the localized
module will be denoted by S´1 M. The construction of S´1 M is functorial in M and
gives a functor Mod A Ñ ModS´1 A with the important properties of being additive and
exact. Moreover, it preserves tensor products and hom-sets between finitely presented
modules. It takes submodules to submodules and respects most of the standard
operations on submodules. The localisation functor turns out to coincide with the base
change functor M ÞÑ Mb A S´1 A.
Just as with rings, one writes Mp and M f for S´1 M when S is respectively the
complement of a prime ideal p i. e. S = Azp and the set S = t f n u of non-negtative
powers of an element.
(7.30) To construction the localized module S´1 M we mimick the way S´1 A was
fabricated. Details will be skipped, but they may be verified mutatis mutandis as in the
case of ring.
To begin with one introduces an equivalence relation on the Cartesian product M ˆ S
by declaring two pairs (m, s) and (m1 , s1 ) to be equivalent if

t(ms1 ´ m1 s) = 0 (7.5)

for some t P S. One checks that this is an equivalence relation (transitivity is the only
challenge) and defines S´1 M to be the set of equivalence classes S´1 M = M ˆ S/ „.
The equivalence class of a pair (m, s) will be designated either by m/s or by ms´1 .
The additive group structure of S´1 M is introduced in analogy with the usual way of
adding fraction, namely as m/s + n/t = (mt + sn)/st. And the action of an element
a/s P S´1 A is given in the straightforward way: a/s ¨ m/t = am/st. There is canonical
map ι S : M Ñ S´1 M that sends m to the class of (m, 1).
Naturally, there is a lot of checking to be done; that definitions are legitimate and
that axioms are satisfied. Every single step is straightforward, and we leave these
soporific verifications to the students for a misty day. Summing up, one has:

Proposition 7.31 The localization S´1 M is an S´1 A-module, and the canonical localization
map ι S : M Ñ S´1 M is A linear. Every element of S´1 M is of the form m/s and two such, m/s

10th January 2021 at 11:32am

Version 4.0 run 30
 localization of modules 191

and m1 /s1 , are equal precisely when ts1 m = tsm1 for some t P S. The kernel of ι S consists of the
elements m in M killed by some member of S; that is, ker ι S = t m P M | tm = 0 for some t P
S u.
To simplify the notion, just as with rings, one soon drops the reference to the map ι S
and writes x or x/1 for ι S ( x ), but with some cautiousness since the image very well can
be zero.
(7.32) When the module M is finitely generated, say by members m1 , . . . , mr , the images
ι S (mi ) of the mi ’s will obviously generate S´1 M. Indeed, pick a member xs´1 from
S´1 M and write x = ai mi then of course xs´1 = ai s´1 mi .
ř ř

Functoriality
(7.33) Given two A-modules M and N and an A-linear map φ : M Ñ N. Sending xs´1
to φ( x )s´1 gives an S´1 A-linear map between the localized modules S´1 M and S´1 N;
that is, a map S´1 φ : S´1 M Ñ S´1 N.
A formal definition starts with the map ( x, s) Ñ (φ( x ), s) between the Cartesian
products M ˆ S and N ˆ S, and the salient point is that this respects the equivalence
relations from (7.5). Indeed, a relation like t( xs1 ´ x1 s) = 0 leads to the relation
t(φ( x )s1 ´ φ( x1 )s) = 0 because φ is A-linear. Thus φ( x )s´1 does not depend on the
choice of representatives, and xs´1 ÞÑ φ( x )s´1 is a legitimate definition.
(7.34) From the definition of S´1 φ we infer immediately that a linear combination of
maps between N and M localizes to the corresponding linear combination; that is, one
has
S´1 ( aφ + bψ) = aS´1 φ + bS´1 ψ,
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

S ´1 ( ψ ˝ φ ) = S ´1 ψ ˝ S ´1 φ

whenever φ and ψ are composable, since it holds true already at the level of the Cartesian
products—and of course, S´1 (id M ) = idS´1 M .

Proposition 7.35 Let A be a ring and S a multiplicative subset of A. The localization functor
Mod A Ñ ModS´1 A 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 (7.6)
to an exact sequence. So our sole task is to verify that the sequence

S´1 ψ S´1 φ
S ´1 N / S ´1 M / S ´1 L

10th January 2021 at 11:32am

Version 4.0 run 30
192 localization

is exact, which amounts to checking that ker S´1 φ = im S´1 ψ. To that end, pick an
element xs´1 in the kernel of S´1 φ. This means that φ( x )s´1 = 0, hence tφ( x ) = 0 for
some t P S. But then tx P ker φ, and since the sequence (7.6) is exact, there is an element
y in N such that ψ(y) = tx. But then we have S´1 ψ(ys´1 t´1 ) = ψ(y)s´1 t´1 = xs´1 ,
and we are through. o

Submodules
Given a submodule N Ď M. The localized module S´1 N can be considered to be a
submodule of S´1 M. 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 S´1 N. Notice that since localization is an exact operation, there is a
canonical isomorphism S´1 ( M/N ) » S´1 M/S´1 N sending a class [m]s´1 to the class
[ms´1 ], and certainly, we shall not refrain from the slight abuse of language it is to
consider the two to be equal.
(7.36) Localization behaves nicely with respect to many of the standard operations one
may perform on submodules like taking sums and finite intersections and forming
annihilators and transporters. However, localization does not commute with infinite
intersections as we saw in example 7.11 on page 182, nor does it commute with forming
annihilators of infinitely generated modules (Exercises 7.34 and 7.35 below), but the
formation of arbitrary direct sums commute with localizations. We summarize some of
these properties in the following proposition:

Proposition 7.37 Let A be a ring and S a multiplicative set in A. Let N, N 1 and tNi u be
submodules of the A-module M. Then the following four properties hold true:
i) S´1 i Ni = i S´1 Ni ;
ř ř

ii) S´1 ( N X N 1 ) = S´1 N X S´1 N 1 ;

iii) S´1 iP I Mi » iP I S´1 Mi ;
À À

iv) Assume that N is finitely generated, then ( N 1 : N )S´1 A = (S´1 N 1 : S´1 N ).

ř
Proof: To establish the first equality, observe that from the inclusion Ni Ď i Ni ensues
that S´1 Ni Ď S´1 i Ni , hence one has i S´1 Ni Ď S´1 i Ni . Any element in S´1 i Ni
ř ř ř ř

is of the form ( i xi )s´1 with merely finitely many of the xi ’s being non-zero, and
ř

therefore lies in i S´1 Ni .

ř

It holds that S´1 ( N X N 1 ) Ď S´1 N X S´1 N 1 as localization respects inclusions, and

the second assertion follows because if y P S´1 N X S´1 N 1 we have y = n/s = n1 /s1
with n P N and n1 P N 1 and s, s1 P S, which means that ts1 n = tsn1 for some t P S.
Putting x = ts1 n = tsn1 , we infer that x P N X N 1 and y = x/tss1 P S´1 ( N X N 1 ).
The third assertion is a direct consequence of the two first for a direct sum of two
modules; hence it holds for a finite sum by an obvious induction argument. Finally, the
À
case of an infinite sum follows from the finite case since each element in iP I Mi and

10th January 2021 at 11:32am

Version 4.0 run 30
 localization of modules 193

each in i S´1 Mi lies in a submodule which is a direct sum of a finite number of the
À

Mi ’s.
In the fourth and last assertion the module N is assumed to be finitely generated.
Let m1 , . . . , mr be generators. These also generate the localized module S´1 N over S´1 A.
The trick is to consider the A-linear mapping

/
µ
M/N 1
À
µ: A i

that sends a ring element a to the sequence ([ am1 ], . . . , [ amr ]) where classes are taken
mod N 1 . The transporter ( N 1 : N ) = t a | aN Ď N 1 u satisfies ( N 1 : N ) = i ( N 1 : Ami )
Ş

and appears as the kernel of this map and thus lives in the exact sequence

/ ( N1 : N) /A /
µ
M/N 1 .
À
0 1ď i ď r

Because localization is an exact operation which commutes with direct sums, when
localized in S, this sequence becomes
S´1 µ
/ S ´1 ( N 1 : N ) / S´1 A / S´1 M/S´1 N 1 ,
À
0 1ď i ď r

where S´1 µ( a) = ([ am1 ], . . . , [ amr ]) with a P S´1 A and the classes being taken mod
S´1 N 1 . Since the mi ’s generate S´1 N, it holds that ker µ = (S´1 N 1 : S´1 N ), and the
equality S´1 ( N 1 : N ) = (S´1 N 1 : S´1 N ) follows. o
It is worthwhile mentioning two particular cases of the fourth assertion, namely
when N = M and N 1 = (0), in which case ( N 1 : N ) = Ann M, and the case when
N 1 = 0 and N is generated by a single element m. Then ( N 1 : N ) = (0 : Am) = Ann m.
In short, for finitely generated modules forming annihilators commute with localization.

Corollary 7.38 Assume that A is a ring with a multiplicative set S and that M is an
A-module. Then
i) For any element m P M it holds true that S´1 (0 : Am) = (0 : S´1 Am);
ii) If M is finitely generated, one has S´1 Ann M = Ann S´1 M.
Exercises
(7.34) Localization does not commute with infinite direct products in general. Let
p P Z be a number and denote by S the multiplicative set S = t pn | n P N0 u in Z.
Show that there is a natural inclusion
ź ź
S ´1 ZĎ S´1 Z,
i PN i PN

but that the inclusion is strict. Hint: Strings shaped like ( ai p´ni )iPN will not lie in the
image when ni tends to infinity with i.
(7.35) Let p be a prime and let S be the multiplicative set t pn | n P N0 u in Z
of all non-negative powers of p. Consider the abelian group i Z/pi Z. Show that
À

10th January 2021 at 11:32am

Version 4.0 run 30
194 localization

S´1 ( i Z/pi Z) = 0. Show that Ann( i Z/pi Z) = (0). But of course it holds true that
À À

Ann S´1 ( i Z/pi Z) = S´1 Z = Z[ p´1 ], hence localization and forming annihilators
À

do not always commute.

M

Relation with the tensor product

(7.39) Given an A-module M, the “action” of S´1 A on M is expressed by the map
M ˆ S´1 A Ñ S´1 M that sends (m, as´1 ) to am ¨ s´1 . This is obviously A-bilinear, and
in view of the universal property enjoyed by the tensor product, induces an A-linear
map Ψ : Mb A S´1 A Ñ S´1 M, which on decomposable tensors acts by sending mbas´1
to ams´1 . This map turns out to be an isomorphism:

Proposition 7.40 The map Ψ is an Mb A S´1 A » S´1 M of A-modules.

Proof: The crux of the poof is that the tensor product Mb A S´1 A is one of the rare
instances that all elements are decomposable; that is, they are all of the form mbs´1
with m P M and s P S. Granted this, if mbs´1 is mapped to zero, the element m is
annihilated by some t from S. But then mbs´1 = tmbs´1 t´1 = 0. So the map Ψ is
injective, and it obviously is also surjective.
1
A priori an element from MbS´1 A is of the shape 1ďiďr 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 the form
ř ´1 ´1
i mi bsi . The trick is now to let s = s1 ¨ ¨ ¨ sr and ti = ssi = s1 . . . spi . . . sr , and with
this we find ÿ ÿ
x= mi ti bs´1 = ( mi ti )bs´1 = mbs´1 ,
i i
ř
with m = i mi ti . o
(7.41) Base change functors preserve tensor products (Proposition 6.37 on page 162)
which combined with Proposition 7.40 above, yields that the localization process pre-
serves tensor product:

Proposition 7.42 Let M and N be two A-modules and S a multiplicative set in A. Then there
is a canonical isomorphism

S´1 ( Mb A N ) » S´1 MbS´1 A S´1 N.

(7.43) When it comes to hom-sets, the behaviour is rather nice, at least for mod-
ules of finite presentation. In general, sending an A-linear map φ between two A-
modules M and N to the localized map S´1 φ is an A-linear map Hom A ( M, N ) Ñ
HomS´1 A (S´1 M, S´1 N ). By the universal property of localization it extends to a map
S´1 Hom A ( M, N ) Ñ HomS´1 A (S´1 M, N ); and in case M is of finite presentation, this
map is an isomorphism:

10th January 2021 at 11:32am

Version 4.0 run 30
 localization of modules 195

Proposition 7.44 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
»
S´1 Hom A ( M, N ) ÝÑ HomS´1 A (S´1 M, S´1 N )

induced by sending φ to S´1 φ is an isomorphism.

Proof: Recall that both localization and the hom-functors are additive functors, hence
the proposition holds true whenever M is a free A module of finite rank n; indeed, one
finds
S´1 Hom A (nA, N ) » S´1 nN » nS´1 N » HomS´1 A (nS´1 A, S´1 N ), (7.7)
where the isomorphisms are the natural ones (the one in the middle is an isomorphism
since localization is additive, and the two others because hom-functors are additive).
Since M is assumed to be of finite presentation, it lives in an exact sequence

/ nA /M / 0,
ψ π
mA (7.8)

with m, n P N and where ψ and π are A-linear maps. Consider the diagram

0 / S´1 Hom A ( M, N ) / S´1 Hom A (nA, N ) / S´1 Hom A (mA, N )

  
0 / Hom ´1 (S´1 M, S´1 N ) / Hom ( nS ´1 A, S´1 N ) / Hom ´1 (mS´1 A, S´1 N ).
S A S´1 A S A

The upper sequence is obtain from (7.8) by applying the functor S´1 Hom A (´, N )
to it, and it is therefore exact by left exactness of hom-functors and exactness of
the localization functor. The bottom sequence comes from (7.8) with the functor
HomS´1 A (S´1 (´), S´1 N ) applied to it, and is exact for the same reasons. The vertical
maps are the canonical maps induced by sending maps φ to S´1 φ, and it is a matter of
simple verification 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
The first part of the proof fails when M is not of finite presentation; for instance,
À
when M = iPN A, one has
à ź ź
Hom A ( A, A) = Hom A ( A, A) = A,
iPN iPN iPN

and infinite products do not in general commute with localization as we saw in Exer-
cise 7.34. The proof only relies on the two facts that the hom-functor’s are left exact and

10th January 2021 at 11:32am

Version 4.0 run 30
 196 localization

that localization is exact, and thus remains valid for any flat base change; or for that
matter, the hom’s may be replaced by any additive left exact functor that sends free
modules to free modules of the same rank.

7.3 Local properties

Based on the belief that modules over local rings are simpler than others, a general
technique is to try to pass from local knowledge—that is, knowledge of the localized
modules Mp —to global knowledge. One envisages to infer properties of the module M
itself from properties of the localized modules Mp .
In this context it is quite natural to introduce the notion of a local property. A property
Local properties (lokale of modules, call it P, is said to be a local property if all the localizations Mp at prime
egenskaper) ideals have P if and only if that the module M itself has P. Equally well, one may speak
about local properties of homomorphisms of modules: Such a property P is local, if a
map φ : M Ñ N has P whenever all localizations φp have P.

The localness of being zero

We shall see several instances of the local to global principle, but begin with the simplest
of all properties, namely that of being zero! Applied to kernels and cokernels this leads
to local criteria for homomorphisms to be injective or surjective.
(7.45) The point of departure is the following easy lemma which describes when
elements remain non-zero in a localization.

Lemma 7.46 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 7.31 on page 190 that the image of x in Mp being zero is equivalent to x being
killed by an element in S; that is, by an element belonging to Ann x but not to p. o
This lemma immediately translates into the following fundamental principle:

Proposition 7.47 (Being zero is a local property) An A-module M is equal to zero if

and only if either of the two following assertions holds.
i) Mm = 0 for all maximal ideals m in A;
ii) Mp = 0 for all prime ideals p in A.

Proof: Two of the implications are obvious; localizing the zero module yields the zero
module. For the rest, it suffices to show that the weaker condition in assertion i) implies
that M = 0. To this end, assume that M is non-zero and let x be a non-zero element
in M. The annihilator Ann x of x is then a proper ideal and is contained in a maximal

10th January 2021 at 11:32am

Version 4.0 run 30
 local properties 197

ideal m. By the simple lemma above, the image of x in Mm is non-zero and a fortiori
Mm is non-zero. o
We have seen that localization is an exact operation and therefore it commutes with the
formation kernels and cokernels of homomorphisms. In combination with the localness
of being zero, this yields the following important local criterion for a map to be injective
or surjective.

Corollary 7.48 (Being injective or surjective is a local property) Assume M and

N be two A-modules. Any A-linear map φ : M Ñ N is injective (respectively surjective) if and
only if either of the two following equivalent conditions is satisfied.
i) The localization φm : Mm Ñ Nm is injective (respectively surjective) for all maximal
ideals m in A;
ii) The localization φp : Mp Ñ Np is injective (respectively surjective) for all prime
ideals p in A.

Proof: Localization is an exact functor, so (ker φ)m = ker(φm ) for every maximal
(respectively prime) ideal m, and Proposition 7.47 above tells us that ker φ = 0 if
and only if ker φm = 0 for all m. This takes care of the part about injectivity, for the
surjectivity part, one replaces ker φ by coker φ. o

Corollary 7.49 (Being an isomorphism is a local property) Let M and N be two A-

modules. An A-linear mapping φ : M Ñ N is an isomorphism if and only if the localized map
φm : Mm Ñ Nm is an isomorphism for all maximal ideals m, or equivalently, if and only if
φp : Mp Ñ Np is an isomorphism for all prime ideals p.

(7.50) There are many other instances of local properties, but let us mention two, namely
flatness and projectivity.

Proposition 7.51 (Flatness is a local property) An A-module M is flat over A if and

only if Mp is flat over Ap for every prime ideal p P Spec A.

Proof: We are to check that an A-module M is flat over A if Mp is flat over Ap for all
p P Spec A. So let
/N / N1
φ
0 (7.9)

be an injection. We are to prove that φb A id M is injective. But for any A-module

L and any prime ideal p it holds ( Lb M)p = Lp b Ap Mp and ditto, for any module
homomorphism ψ one has (φbid M )p = φp b Ap id Mp . Hence the maps in the two
sequences
(φb id M )p
0 / ( Nb A M )p / ( N 1 b A M )p (7.10)

10th January 2021 at 11:32am

Version 4.0 run 30
198 localization

and
φp b Ap id Mp
0 / Np b A Mp / Np1 b A Mp
p p

coincide. The latter is obtained from (7.9) by the two step process of first localizing in
p, which is exact, and subsequently tensorizing by Mp which also is exact since Mp is
assumed to be flat. The map in (7.10) is therefore injective, and citing Corollary 7.48
that injectivity is a local property of homomorphisms, we are through. o
Along the same lines one may show that being projective is a local property for
finitely generated modules; it is true without the finiteness limitation, but we content
ourselves with the case of finitely generated modules, and we also content ourselves
with announcing the result, leaving the proof as an exercise.

Proposition 7.52 (Localness of being projective) Let M be a finitely generated module.

Then M is a projective A-module if and only if Mp is a projective Ap -module for all prime ideals
p.

Exercises
(7.36) Prove Proposition 7.52. Hint: Follow the lines of the proof of Proposition 7.51.
The isomorphism in Proposition 7.44 might be useful.
(7.37) Let C‚ be a complex of A-modules. Show that localization commutes with
taking homology; that is, show that for every prime ideal p P Spec A one has canonical
isomorphisms Hi (C‚ )b A Ap » Hi (C‚ b A Ap ). Conclude that being exact is a local
property of complexes.
M

7.4 An extended version of Nakayama’s lemma

We are prepared to revisit Nakayama’s lemma as announced, and give a version whose
proof relies on a localization technique. The extended version is valid for all ideals not
only those lying in the Jacobson radical; but of course, when weakening the hypothesis,
you get a weaker conclusion.

Proposition 7.53 (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 (1 + a) M = 0.

Proof: Let S be the multiplicative set t 1 + a | a P a u. The ideal S´1 a is contained in

the Jacobson radical of S´1 A (by Proposition 2.66 on page 54): indeed, we are to check
that for each a P a and each x P A the element 1 + s´1 xa is invertible in S´1 A whatever
s P S is. But 1 + s´1 xa = s´1 (s + xa) = s´1 (1 + x1 a1 + xa) with a1 P a and x1 P A, and
both factors are invertible in S´1 A. By Nakayama classic (Proposition 4.52 on page 106)

10th January 2021 at 11:32am

Version 4.0 run 30
 the support of a module 199

we may then conclude that S´1 M = 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 their product,
which obviously kills M and is of the required form. o

7.5 The support of a module

Since prime ideals p such that Mp = 0 are insignificant in the local–to–global process
we just described, it is very natural to introduce the subset Supp M of Spec A consisting
of the prime ideals p so that Mp ‰ 0; that is, we define

Supp M = t p P Spec A | Mp ‰ 0 u.

It is called the support of M. In many cases, e. g. when M is finitely generated, the The support of a
support of M is a closed subset of Spec A, but whether M is finitely generated or not, it module (støtten til en
modul)
has the weaker property of being closed under specialization; that is, with every p lying in
Supp M the closed set V (p) lies there.
When M and N are finitely generated, the support of the direct sum* M ‘ N equals ˚ More generally this is

the union Supp M Y Supp N and the support of the tensor product Nb A M equals the the case for any
extension of M by N.
intersection Supp N X Supp M. It is worth while observing that the support takes the
two “ring-like-operations” direct sum and tensor product in the category of finitely
generated A-modules into the two operations union and intersection of the boolean
ring of closed subsets of Spec A.
Closure properties of the support
(7.54) For a cyclic module M = A/a the support coincides with the closed set V (a)
associated with the ideal a since a prime ideal p belongs to V (a) precisely when
( A/a)p ‰ 0, just apply the simple lemma 7.46 to the element 1 in A/a. This observation
may be generalized to finitely generated modules. Any such module has a support
which is a closed subset of Spec A:

Proposition 7.55 If M is finitely generated A-module, the support Supp M equals the closed
subset V (Ann M) 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 equivalently, that
Mp = 0 if and only if Ann M Ę p. In case an element a P A not belonging to p kills
M, it holds that Mp = 0; indeed, 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 Lemma 7.46 above, there is for each of the xi ’s an element si not in
p killing xi . The product of the si ’s clearly kills M and does not belong to the prime
ideal p since none of the si ’s does; hence Ann M is not contained in p. o
(7.56) The hypothesis that M be finitely generated was used only in the last part of
the proof above, and for a general module M it holds true that Supp M Ď V (Ann M ):

10th January 2021 at 11:32am

Version 4.0 run 30
 200 localization

from Mp ‰ 0 follows that Mq ‰ 0 whenever q Ě p. Indeed, if Ann x Ď p, it obviously

holds that Ann x Ď q, and an element x mapping to a non-zero element in Mp maps to
a non-zero element in Mq . We have thus established

Proposition 7.57 Let M be an A-module. The support of M is closed under specialization;

that is, for each prime ideal p P Supp M it holds that V (p) Ď Supp M.

Examples
(7.18) As already observed, the support of a cyclic module A/a equals the closed set
V (a).
(7.19) One has Supp Q = Spec Z since S´1 Q = Q for any multiplicative set S in Z.
More generally, for the fraction field K of any domain A is of global support; that is,
Supp K = Spec A.
(7.20) An example of the failure for “large modules” of the support being the closed set
defined by the annihilator, can be the Z-module Z p8 = Z[ p´1 ]/Z, where p is a prime.
Each element of Z p8 is the class of a rational number of the form x = a/pr with a
prime to p. Since yx P Z if and only if y is divisible by pr , one has Ann x = ( pr ), and
from Lemma 7.46 above it follows that Supp Z p8 = t( p)u.
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 (a power pr
kills the class of p´n only if n ď r). This shows that Supp Z p8 , although being closed,
differs from V (Ann Z p8 ).
(7.21) The support of a module is not always a closed subset of Spec A. Take any infinite
sequence of primes pi which does not including all primes—for instance, every second
i Z/pi Z. The support of M is the infinite
À
prime—and consider the module M =
subset t( pi )u. The only infinite closed subset of Spec Z being the entire spectrum, this
set is not closed.
K

The support of extensions

(7.58) Since localization is an additive functor so that ( M ‘ N )p » Mp ‘ Np it is obvious
that the support of a direct sum of two A-modules is the union of their supports. This
Extensions of modules generalizes to so-called extensions; that is, modules in the midst of an exact sequence
(utvidelser av moduler) (which is not necessarily split exact).

Proposition 7.59 Assume that

0 /N /M /L /0

is an exact sequence of A-modules. Then Supp M = Supp N Y Supp L.

10th January 2021 at 11:32am

Version 4.0 run 30
 the support of a module 201

Proof: The proposition follows immediately from the 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 ones do. o

The support of a tensor product

(7.60) The aim of this paragraph is the prove that the support of a tensor product is the
intersection of the supports of the two factors, at least when the involved modules are
finitely generated.

Proposition 7.61 Let M and N be two finitely generated A-modules. Then the following
equality holds true Supp Mb A N = Supp N X Supp M.

The basic argument takes place over a local ring:

Lemma 7.62 Let A be a local ring with maximal ideal m and 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.

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 6.37 on page 162), one has

( Mb A N )b A k = ( Mb A k)bk ( Nb A k).

The tensor product of two non-zero vector spaces being non-zero (e. g. Proposition 6.21
on page 155), we infer that ( Mb A N )b A k ‰ 0, and hence Nb A M ‰ 0 a fortiori. o
Proof of Proposition 7.61: 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 isomor-
phism
( Mb A N )p » Mp b Ap Np ,

the proposition then follows from the lemma. o

Example 7.22 Proposition Proposition 7.61 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 = V (a), which
is non-empty. K

10th January 2021 at 11:32am

Version 4.0 run 30
 202 localization

Exercises
(7.38) 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 7.47 on page 196.
Groups of bounded (7.39) An abelian group M is said to be of bounded exponent if some power pn of a prime
exponent (grupper med p kills every element of M. Give an example of a group of bounded exponent that is not
begenset eksponent)
finitely generated. Prove that if M is of bounded exponent, then Supp M = V (Ann M).
(7.40) Let p be a prime number and let M be the abelian group M = iPN0 Z/pi Z.
À

Determine the annihilator Ann M and the support Supp M.

M

7.6 The rank of a module

In this section M is a finitely presented module over a ring A. Recall that this means
that M lives in a short exact sequence

/E /M /0
φ
F

where E and F are free A-modules of finite rank. Let furthermore p be a prime ideal
in A and k(p) is the fraction field of A/pA; i. e. k(p) = Ap /pAp . One defines the local
rank rkp M of M at p as the dimension rkp M = dimk(p) Mb A k(p).
We conclude this chapter by taking a closer look at the special case when A is a
domain and the prime ideal is the zero ideal. That is Ap is the fraction field K of A.
Then Mb A K will be a vector space over K, and the dimension dimK Mb A K is called
The rank of a module the rank of M. The properties of the localization functor translates into properties of the
(rangen til en modul) rank, it will have the nice properties of being both additive and multiplicative:

Theorem 7.63 Let A be a domain with fraction field K and let M and N be two A-modules. It
then holds true that
i) rk M = 0 if and only if M is a torsion module;
ii) rk M = rk N + rk M/N when N Ď M;
iii) rk Mb A N = rk M ¨ rk N.
iv) If M is finitely presented then rk Hom A ( M, N ) = rk M ¨ rk N.

Proof: As already said, this follows from the properties of the localization functor
combined with the appropriate properties of the dimension of a vector space. o

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 8

Projective modules

In paragraph 5.18 we introduced the projective modules as those A-modules P such

that Hom A ( P, ´) is an exact functor, and we showed that P is projective if and only
if it is the direct summand in a free module. The projective modules play a special
and important role both in number theory and algebraic geometry. Certainly more
involved than free modules, but still to a great extent maniable, they enjoy a series of
good properties. In many cases they furnish important invariants of rings, and they are
the supermen of homological algebra, where they among other things serve to define
the so-called Ext- and Tor-groups, which describe respectively the ‘missing cokernels”
and the “missing kernels” we mentioned in Paragraph 5.12 on page 129.
˚ The exceptions will
Contrary to what usually is the case, infinitely generated projective modules turn
out to be much simpler than the finitely generated ones. A famous result of Hyman be free over each
connected component
Bass’s ([?]) states that over Noetherian rings with connected* spectra projective modules of Spec A, but can of
requiring infinitely many generators are in fact free. We shall not treat Bass’s theorem, course have bases with
different cardinality on
but it justifies largely that we mostly work with finitely generated projectives. different components
There is also a result of Kaplansky’s pointing in the same direction as Bass’s, but (and can even be zero
on some of them).
without the Noetherian hypothesis. It asserts that over a local rings all projective
modules are free. A projective module P thus has the virtue of being locally free; that Locally free modules
is, the localization Pp is a free Ap -module for all primes p P Spec A. We shall give a (lokalt frie moduler)

proof of this when P is finitely generated, which is a classical application of Nakayama’s

lemma.
The geometrical counterpart to the locally free modules are the so-called vector
bundles. In topology these are continuous maps E Ñ X with all fibres being vector
spaces (either real or complex) which are locally trivial; that is, over a suitable open cover
tUi u of X it appears (in the complex case) as the projection Cn ˆ Ui Ñ Ui (additionally,
there is also an important requirement that the transition functions arising over the
intersections Ui X Uj all be linear).
204 projective modules

8.1 Projective and locally free modules

Being projective is a local property of a module as we established in Proposition 7.52
on page 198, but being free is not local—in Examples 5.5, 5.6 and 5.8 on page 133
we exhibited projective modules that are not free, but they are easily checked to be
locally free. This illustrates the general fact that the locally free modules, at least
among those that are finitely generated are exactly the projective ones. To establish this
as a generally valid principle one merely needs to show that over local rings finitely
generated projective modules are free:

Proposition 8.1 Let A be a local ring and P a finitely generated projective module. Then P is
free.
Proof: This is a classical application of Nakayama’s lemma. Let k be the residue field
of A and consider Pb A k. It is a finite vector space over k and has a basis, say with r
elements. Lifting the basis elements to elements in A we obtain a map φ : rA Ñ P such
that φbidk is an isomorphism, and Nakayama’s lemma yields that φ is surjective. The
kernel of φ lives in the short exact sequence

/ ker φ / rA /P / 0,
φ
0

and the module P being projective the sequence is split and hence stays exact when
tensorized by k. Again since φbidk is an isomorphism, it follows that ker φb A idk = 0.
Now, any direct summand in a finitely generated module is finitely generated. Therefore
Nakayma’s lemma applies to ker φ, and we may infer that ker φ = 0, which is exactly
what we need to conclude that P » rA; hence P is free. o
As announced, the proposition gives the following corollary:

Corollary 8.2 Let A be a ring and P a finitely generated A-module. Then P is projective if
and only it is locally free; that is, if and only if Pp is a free Ap -module for all p P Spec A.
(8.3) A local basis for P at a prime ideal p can be extended to a basis for P over an open
and distinguished neighbourhood of p in Spec A, and we have the following slightly
stronger result than Proposition 8.1.

Proposition 8.4 Let P be a finitely generated projective module over the ring A. Then there
exist a finite set t f i uiP I of elements from A so that the distinguished open sets D ( f i ) cover
Spec A and such that each localization Pf i is a free module over A f i .
Proof: Since Spec A is quasi-compact (see Exercise 2.58 on page 66) it suffices to
find a distinguished neighbourhood round each point p over which P is free; in clear
text, given a prime ideal p we search for an element f R p such that Pf is free as an
A f -module.

10th January 2021 at 11:32am

Version 4.0 run 30
 projective and locally free modules 205

To find such an element, begin with a basis tai u, with say r elements, for the localized
module Pp over Ap whose elements belong to P. Such a basis defines a map φ : rA Ñ P,
and it lives in the exact sequence

/ ker φ / rA /P / coker φ / 0.
φ
0
Now, coker φ is finitely generated (since P is); its support equals V (Ann coker φ) and
does not contain p. Hence the annihilator ideal Ann coker φ is non-zero, and we let g be
any one of its non-zero elements. Over the localization A g the map φg is surjective so
that the kernel ker φg is a direct summand in rA g and thus it is finitely generated. The
support equals V (Ann ker φg ), which is not the entire Spec A g (it does not contain p),
and we may find a non-zero element h P Ann coker φg X A. Then f = gh is your man.
o
The rank of projective modules
(8.5) Suppose we are given a finitely generated projective module P. At any point p
in the spectrum Spec A the module P being locally free has a local rank rp ( P), namely The local rank (den
the non-negative integer r so that Pp » rAp . This local rank may vary, it can assume lokale rangen)

different values along different connected components of the spectrum Spec A (see
Example 8.1), but when it is constant, it is simply called the rank of P. This is e. g. the The rank of projective
case whenever Spec A is connected. In fact, Proposition 8.4 above, about extension of modules (rangen til
projektive moduler)
local bases, yields that the rank is a locally constant function on Spec A:

Proposition 8.6 (The rank is locally constant) Assume that P is a finitely generated
projective module over A. Then the rank rkp ( P) is locally constant; that is, for each r the set
Ur = t p P Spec A | rkp P = r u is both open and closed. In particular, if Spec A is connected,
the local rank is constant.
Proof: That Ur is open for all r ensues from 8.4, and hence the complement Urc =
Ť
s‰r Us is open as well. o
Example 8.1 When the spectrum is not connected, it is easy to find projective modules
whose local rank takes on different values on different connected components. These
will also be examples of projective modules that are not free. The simplest example is a
direct product A ˆ B of two non-null rings A and B (the most minimalistic example was
already given in Example 5.5). Then Spec A ˆ B is the disjoint union Spec A Y Spec B.
Both A and B are natural A ˆ B-modules—realized as A ˆ (0) and (0) ˆ B—and as
such are direct summands in R, thus they are projective. But for instance, rp ( A) = 1 for
p P Spec A and rp ( A) = 0 for p P Spec B. K
Modules of constant rank
(8.7) A projective module P is, as we just saw, free over each of the local rings Ap , and
of course, it stays free when tensored with the quotient Ap /pAp : When P is of rank

10th January 2021 at 11:32am

Version 4.0 run 30
206 projective modules

r, it holds true that Pb A Ap /pAp » rAp /pAp for each prime ideal p; hence "the fibre
dimensions" dimk(p) Pb A k (p) will all be the same and equal to r (where, we remind
you, k (p) denotes the fraction field of Ap /pAp ). When A is reduced, the converse holds
true too; that is, if a finitely generated module has constant "fibre dimenison", it will be
projective. Actually, only the maximal and the minimal primes come into play. Recall
that when A is reduced, the local rings Ap at the minimal primes p of A are fields
(Proposition 7.29 on page 187), and moreover he set of zero divisors in A is precisely
the union of the minimal prime ideals.

Proposition 8.8 Assume that A is a reduced ring with finitely many minimal prime ide-
als and let P be a finitely generated A-module. Let r be a natural number. Assume that
dim A/m Pb A A/m = dim Ap Pb A Ap = r for all maximal ideals m and all minimal prime
ideals p in A. Then P is projective of rank r.

Proof: Since being projective is a local property, it suffices to show the proposition
when A is local. Let m be the maximal ideal of A and k = A/m the residue field. The
k-vector space Pb A k has a basis of r element, which may be lifted to elements in P. By
Nakayama’s lemma these elements generate P, so that P lives in a short exact sequence
shape liked
0 /M / rA /P / 0,

which when localized at a minimal prime p, becomes the short exact sequence

0 / Mp / rAp / Pp / 0.

Now, Ap is a field over which Pp is assumed to be a vector space of dimension r, and

therefore Mp = 0. We contend that this implis that M = 0; aiming at a contradiction,
assume the contrary and pick a non-zero element x P M.
Let p1 , . . . , pt be the minimal prime ideals in A. Since Mpi = 0, it holds that
Ann x Ę pi and from Prime Avoidance it ensues that Ann x Ę p1 Y . . . Y pr , so there is
a non-zero divisor killing x. But M is contained in rA and having non-zero element
killed by a non-zero divisor, would be absurd, consequently M = 0. o

Example 8.2 The proposition does not hold for modules over the simplest non-reduced
ring Z/4Z. The quotient F2 = Z/2Z is not projective (no submodule of a free
Z/4Z-module is killed by 2), but the rank is one everywhere (well, everywhere is not
very widespread; the ideal (2) is the only prime ideal). A similar example would be
A = k [ x ]/( x2 ) where k is a field. The residue field k = A/( x ) is not projective, but of
rank one everywhere. K

10th January 2021 at 11:32am

Version 4.0 run 30
 working formulas 207

Exercises
ś
(8.1) Let A = i k i be a finite product of fields. Show that any A-module is projective.
Single out those which are free.
(8.2) Let I be a finite set and for each i P I let Ai be a local ring with residue field k i .
ś
Describe the projective modules over i Ai .
M

8.2 Working formulas

The day-to-day working formulas for projective modules basically say that tensor
products and direct sums of projective modules are projective. One might be tempted
to say that the category of projective A-modules is “ring–like” ans this point of view s
pursued in Exercise 8.6 below. Since probably most projective modules one would meet
practising algebraic geometry or commutative algebra, are finitely generated (and in
fact, in view of Bass’ result, the only non-trivial case), we just treat those.
For free modules the formulas are already established (trivial for sums, see Para-
graph 6.20 for products), and bearing in mind that tensor products and hom’s in good
cases commute with localization, the formulas follow easily from the fact that being
locally free is equivalent to being projective, when the involved modules are finitely
generated.
(8.9) Before proving the working formulas, we need two new notions. The dual of a The dual module (den
module M is denoted M˚ and is defined as M˚ = Hom A ( M, A). Sending M to M˚ is dual modulen)

an additive and contravariant functor from Mod A to itself, and the double dual M˚˚
will be a covariant and additive endofunctor of Mod A . For each M there is a canonical
evaluation map γ M : M Ñ M˚˚ = Hom A (Hom A ( M, A), A) defined by the assignment
x ÞÑ (φ ÞÑ φ( x )).
If E is a finitely generated free module with basis tei u1ďiďr , the dual module E˚ is
free with basis the so-called dual basis têi u. It is defined by êi (e j ) = δij , and the easily Dual basis (dual basis)
ř
verified formula φ = i φ(ei )êi , shows that it indeed is a basis.
Furthermore, for every pair of A-modules M and N, there is a canonical map
ρ M,N : M˚ b A N Ñ Hom A ( M, N ), which on decomposable tensors is defined by the
assignment φbx ÞÑ (y ÞÑ φ(y) x ).

Lemma 8.10 (Dual of a free module) For a finitely generated free module E, the map γE
is an isomorphism. Moreover, if F is another finitely generated free module, the map ρ E,F is an
isomorphism.

Proof: The map γE sends the basis element ei to the map φ ÞÑ φ(ei ). Hence, one has
ř
the formula φ = i γE (ei )(φ)eˆi , which shows that γE is an isomorphism.

10th January 2021 at 11:32am

Version 4.0 run 30
208 projective modules

Letting t f j u be a basis for F, it holds that ρ E,F sends êi b f j to the map eν ÞÑ δν,i ¨ f j .
ř
Hence, if φ : E Ñ F has matrix ( aij ), it holds that φ = i,j aij ¨ ρ E,F (êi b f j ), from which
we deduce that ρ E,F is an isomorphism. o
(8.11) We are now well prepared for the working formulas for finitely generated
projective modules:

Proposition 8.16 Let P and Q be two finitely generated projective modules over the ring A.
Moreover, let p denote a prime ideal in A.
i) The direct sum P ‘ Q is projective and rp ( P ‘ Q) = rp ( P) + rp ( Q);
ii) The tensor product Pb A Q is projective, and rp ( Pb A Q) = rp ( P)rp ( Q);
iii) The dual module P˚ = Hom A ( P, A) is projective and rp ( P) = rp ( P˚ ). The
canonical evaluation map γP yields an isomorphism P » P˚˚ ;
iv) The module Hom A ( P, Q) is projective, and the canonical map ρ P,Q is an isomorphism
P˚ b A Q » Hom A ( P, Q).
Proof: As noted above, these statements follows from the facts that a module is
projective if and only if it is locally free (i. e. Pp is free over Ap for all p P Spec A), that
the corresponding statements hold for free modules. together with the good behaviour
of tensor products and hom-modules with respect to localization.
The first statement i) is clear.
To prove statement ii) recall that base change respects tensor productucts (Proposi-
tion 6.37 on page 162), so it holds that ( Pb A Q)p = Pp b Ap Qp . And when Pp and Qp
both are free, it follows from Corollary 6.22 on page 155 that ( Pb A Q)p is free of rank
rp ( P )rp ( Q )
Statement iii) follows since by Proposition 7.44 on page 195 it holds that ( P˚ )p =
( Pp )˚ and the latter is free of rank rp ( P). Moreover, Lemma 8.10 gives that (γP )p =
γPp is an isomorphism for all p, and hence γp is an isomorphism because being an
isomorphism is a local property.
Finally, statement iv) is a consequence of Lemma 8.10 and that forming tensor
products and homomorphism modules commute with localization. o
Example 8.3 Recall Exercise 4.27 on page 103 where you were asked to prove that
the direct product iPN Z of countably many copies of Z is not a free Z-module,
ś

thus giving an example of an infinite product of free modules which is not free. A
slight extension of the proof indicated there shows that neither is iPN Z a projective
ś

Z-module (check it!), so infinite direct products of projective modules are not always
projective. K

Exercises
(8.3) Show that P Ñ P˚˚ is always injective, but that it is not necessarily an isomorphism
when P is not finitely generated, even when the base ring is a field. Hint: Consider the

10th January 2021 at 11:32am

Version 4.0 run 30
 working formulas 209

˚˚
Q-vector space P = iPN Q and show that P and P are not of the same cardinality.
À

(8.4) Show that the tensor product Pb A Q of two projective modules is projective
wether they are finitely generated or not. Hint: Use adjointness, Proposition 6.24 on
page 156.
(8.5) The Eilenberg swindle. The simplistic behaviour of infinitely generated projectives,
is to a large extent rooted in the so called Eilenberg swindle. Let F be a free A-module
that is not finitely generated and assume that P is a direct summand in F. The swindle
is the assertion that P ‘ F » F. Let Q be a complement of P in F so that F = Q ‘ P.
a) Show that F is isomorphic to the direct sum of countably many copies of itself;
that is, F » F ‘ F ‘ . . . ;
b) Show that P ‘ F » P ‘ Q ‘ P ‘ Q ‘ P ‘ . . . ;
c) Conclude that P ‘ F » F. Hint: Swap parantheses.
(8.6) K0 of a ring. At the top of the present subsection we alluded to the category of
projective A-modules being “ring-like”. It possesses a sum operation and a product
operation and these satisfy formulas closely resembling the ring axioms, but merely up
to isomorphism and not up to equality; and of course, there is no subtraction. Passing
to isomorphism classes solves the first fault, and there is general technique to introduce
a subtraction. One passes to so-called virtual projective modules, and formally introduces Virtual projective
the ring K0 ( A) of such whose elements are finite linear combinations i ai [ Pi ] of modules (øyensynlige
ř
projektive moduler)
isomorphism classes of finitely generated projective modules with the ai ’s being integers
(allowed to be negative). The ring operations comply to the rules [ P ‘ Q] = [ P] + [ Q]
and [ Pb A Q] = [ P] ¨ [ Q].
The construction goes as follows: One begins with the free abelian group G with a
basis the set of all isomorphism classes of finitely generated projective A-modules. Next,
one considers the subgroup H of G generated by expressions PĞ ‘ Q ´ Ps ´ Q
s where a
bar indicates an isomorphism class, and introduces the underlying abelian group of
K0 ( A) as the quotient K0 ( A) = G/H. The class of a module P in K0 ( A) is designated
by [ P].
a) Show that the asigment [ P] ¨ [ Q] = [ Pb A Q] extends bilinearly to the entiere
K0 ( A) and makes K0 ( A) a commutative ring with [ A] as unit element;
b) If A Ñ B is a ring homomorphism, show that base change functor P ÞÑ Pb A B
induces a ring homomorphism K0 ( A) Ñ K0 ( B). Show that this makes K0 a
functor;
c) Show that the local rank at a prime ideal p, is a ring homomorphism rp :
K0 ( A) Ñ Z.
(8.7) If one in the definition of K0 ( A) had allowed all projective modules, and not
only the finitely generated ones, show that K0 ( A) would have been the zero ring.
À
Hint: Consider the countable direct sum P = iPN A and show that A ‘ P » P.

10th January 2021 at 11:32am

Version 4.0 run 30
 210 projective modules

8.3 The Picard group

One of the most basic invariant of a variety is the so-called Picard group. The elements
are isomorphism classes of creatures called invertible sheaves and whose algebraic
avatars are the invertible modules which we are about to define. Among other things—
and which might be their most important role—they govern maps from the variety
to projective spaces. The analogue of the Picard groups in algebraic number theory
Ideal class groups are the so-called ideal class groups, a notion which actually predates the notion of the
(idealklassegrupper) Picard group by about a century. It is of fundamental importance for the study of
number fields and measures how far the ring of integers in a number field is from being
factorial.
Invertible modules (8.17) The tensor product Mb A N induces a binary operation on the set of isomorphism
(invertible moduler) classes of A-modules, and the basic working formulas in Proposition 6.14 show it is
associative and commutative and has the class of A as a neutral element. Modules do
not in general have inverses, but there are good hopes that the tensor product will give
a group law on the set of those that have. This motivates the notion of invertible modules:
an A-module M is invertible if there is an A-module N such that Nb A M » A.
The invertible modules turn out to coincide with the finitely generated projective
modules of rank one, or in view of Corollary 8.2 on page 204 the finitely generated
modules which are locally free of rank one.

Proposition 8.18 Let A be a ring and M an A-module. The following three statements are
equivalent:
i) M is an invertible module;
Émil Picard
(1856–1941)
ii) M is finitely generated and projective of rank one;
French mathematician
iii) M is finitely generated and locally free of rank one.
Moreover, if M is invertible, it holds true that the evaluation map gives an isomorphism
Mb A M˚ » A so that the dual M˚ serves as an inverse for M.
Proof: The third assertion iii) is included only for completeness; its equivalence with ii)
was established already in Corollary 8.2 on page 204.
We begin with proving that i) implies ii); so assume that M is invertible and let N
be such that Mb A N » A. Let us first establish that M is finitely generated. To that end,
ř
identify Mb A N and A and write 1 = xi byi with xi P M and yi P N. We contend
that the xi ’s generate M. Indeed, let M1 be the submodule of M generated by the xi ’s
and consider the quotient M/M1 . Since the inclusion M1 Ñ M induces a surjection
M1 b A N Ñ Mb A N, it holds that M/M1 b A N = 0, and consequently
M/M1 » M/M1 b A A » M/M1 b A ( Nb A M) » ( M/M1 b A N )b A M = 0.

10th January 2021 at 11:32am

Version 4.0 run 30
 the picard group 211

We proceed to show that M is projective. Since being projective is a local affair, we

may as well suppose that A is a local ring. Let m be the maximal ideal and k = A/m
the residue field. Using that base change preserves tensor products (Proposition 6.37 on
page 162) we find

( Mb A N )b A k » ( Mb A k)bk ( Nb A k) » kbk k » k.

We conclude that dimk Mb A k = 1, and thus Mb A k » k. Nakayama’s lemma then gives

that M is monogenic so that we may write M » A/a for some ideal a. Thena kills M,
and therefore also A as A » Mb A N, which is absurd unless a = (0).
We assume next that M is finitely generated and projective of rank one and aim at
showing the isomorphism Mb A M˚ » A. Being an isomorphism is a local property
(Corollary 7.49 on page 197) and localization commutes with forming hom’s (Propo-
sition 7.44 on page 195) so we may certainly assume that A is local and that M = A
(M is locally free of rank one). In that setting the evaluation map appears as a map
Ab A Hom A ( A, A) Ñ A, which in view of the harmless identity A = Hom A ( A, A), is
nothing but the map Ab A A Ñ A that sends abb to ab; and that is surely an isomor-
phism. This also proves the final statement in the proposition. o
(8.19) According to the proposition the inverse of an invertible module is well defined
and hence the set Pic A of their isomorphism classes is an abelian group when equipped
with the tensor product as a group law. It is called the Picard group. Summed up we The Picard group
have: (Picard-gruppen)

Proposition 8.20 The set Pic A formed by the isomorphism classes of invertible modules is
an abelian group. The product of the classes of P and Q equals the class of Pb A Q. The neutral
element is the class of A, and the inverse of the class of P is the class of P˚ = Hom A ( P, A).
Whenever A Ñ B is a ring-homomorphism, the base change functor (´)b A B takes Pic A
into Pic B making Pic : Rings Ñ Ab a functor.

Proof: Merely the last statement remains to be commented, and it hinges on the base
change functor respecting the tensor product. If Pb A Q » A, we find

( Pb A B)bB ( Qb A B) » ( Pb A Q)b A B » Ab A B = B.

Fractional and invertible ideals

Over an integral domain A, there is a large class of projective modules of rank one,
which we are about to introduce, formed by the so-called invertible ideals. Up to a certain
equivalence, the invertible ideals constitute a group—the ideal class group—which turns
out to be isomorphic of the Picard group. Contrary to the invertible modules which

10th January 2021 at 11:32am

Version 4.0 run 30
 212 projective modules

suffer from a certain elusiveness, invertible ideals are concrete, being submodules of the
fraction field, and in many cases are a lot easier to lay hands on. The Picard group, on
the other hand, generalizes well: many geometric objects like varieties and schemes, are
inhabited by creatures called invertible sheaves whose isomorphism classes constitute
their Picard group.
In Kummer’s set-up, with his ideal numbers in centre stage and where the numbers
are represented by the principal ideals, fractions will most naturally be represented by
principal submodules of the fraction field K of A; in other words, A-submodules of
K requiring a single generator. The obvious “idealizations” are the A-submodules of
K, which when complying to a minor condition, are called fractional ideals. Fractional
ideals can in a natural way be added and multiplied, and the invertible ideals are those
that possess an inverse.
Fractional ideals (8.21) The precise definition is as follows: An A-submodule a Ď K is called a fractional
(brudne idealer) ideal if there is some non-zero x P A such that xa Ď A; one may think about x as a
common denominator for the elements in a. Just as for ideals, the fractional ideal a
Principal fractional is said to be principal if it generated by a single element. That is, if it is shaped like
ideals (brudne t xa | x P A u (which is fractional since the denominator of a serves as a common
hovedidealer)
denominator), and naturally, it will be denote ( a).
Two fractional ideals a and b can be multiplied; exactly as for ideals one defines
ÿ
a¨b = t ai bi | I finite, ai P a and bi P b u,
˚ Noetherian rings pop iP I

up again; we’ll come to which obviously is an A-submodule of K, and it is a fractional ideal since if x ¨ a Ď A
them! They have the
property that all ideals and y ¨ b Ď B, it surely holds that xy ¨ ab Ď A.
are finitely generated. Every finitely generated submodule a of K is fractional; the product of the denom-
inators of members of a generating set multiplies a into A. Over Noetherian rings*
˚ This transporter has the reciprocal holds true since each fractional ideal is isomorphic to a genuine ideal.
elements from the However, submodules of K requiring infinitely many generators need not be fractional;
fraction field K; do not
confuse it with the an example can be subgroup Z[ p´1 ] of Q where one finds elements with any power of
transporter introduced p as denominator.
in Paragraph 2.9 which
has elements merely (8.22) The inverse of a fractional ideal is the fractional ideal a´1 = ( A : a)K = t x P K |
from A. xa Ď A u*. Obviously a´1 a Ď A, and when equality occurs, one says that a is invertible.
Invertible fractional The invertible ideals form an abelian group J ( A) with a ¨ b as product, a´1 as inverse
ideals (invertible and A as neutral element.
brudne idealer)
If a is principal, say a = ( a/b), it holds true that a´1 = (b/a); indeed, in a domain
The ideal class group the equality x ¨ a/b = d is equivalent to x = d ¨ b/a; so the name inverse is merited.
(idealklassegruppen) The principal fractional ideals form a subgroup P( A) of the ideal groupe J ( A), and
the quotient Cl ( A) = J ( A)/P( A) is the ideal class group. Two invertible ideals a and

10th January 2021 at 11:32am

Version 4.0 run 30
 the picard group 213

b belong to the same class in Cl ( A) if and only if there is an element f P K so that

a = f ¨ b; or equivalently, there are non-zero elements a, b P A such that b ¨ a = a ¨ b.
(8.23) Early in the course (Proposition 4.38 on page 100) we saw that in a domain the
free ideals are precisely the principal ones, so just as projective modules form a larger
class than free modules, the invertible ideals form a larger class than the principal ones.
In view of every (finitely generated) projective modules over a local ring being free, the
invertible ideals may be described as the ideals that are locally principal.

Proposition 8.24 (Characterization of invertible ideals) Let A be a domain with

fraction field K and let a be a fractional ideal over a. Then the following statements are
equivalent:
i) a is invertible;
ii) a is finitely generated and projective as an A-module;
iii) aAm is principal for every maximal ideal m in A.
ř
Proof: i) ñ ii): There is by hypothesis a relation 1 = 1ďiďr ai bi with the ai ’s from a
and the bi ’s from a´1 , and multiplying through by any x P a one finds
ÿ
x= ( bi x ) a i , (8.1)
i

where we notice that bi x P A since bi P a´1 . This shows that the ai ’s generate a. To prove
that a is projective, we let E be a free A-module with a basis e1 , . . . , er . The assignments
α(ei ) = ai define an A-linear map α : E Ñ a, and we contend that α is a split surjection.
It is surjective by the observation above that the ai ’s generate a, and in view of (8.1),
ř
the map σ : a Ñ E given by σ ( x ) = i (bi x )ei serves as a right section for α (note that
bi x P A since bi P a´1 and x P a).
ii) ñ iii): This is just the facts that every finitely generated projective module over a
local ring is free (Proposition 8.1 above), and that an ideal in a domain being free means
it is principal.
iii) ñ i): When iii) holds, the inclusion a´1 a Ď A becomes an equality when localized at
each maximal ideal because principal fractional ideals are invertible, and we conclude
by the local nature of being equal (Corollary 7.49 on page 197). o
(8.25) Among invertible ideals we now have two group operations; when considered
to be invertible modules, the product is given as ab A b and the inverse as Hom A (a, A)
whereas when viewed as invertible ideals, the two operations are ab and a´1 . Of course,
the two ways coincide, which is the principal step towards showing that the Picard
group and the ideal class group are isomorphic. Note that for any two ideals a and
b there is an almost tautological map ι : (b : a)K Ñ Hom A (a, b) that sends an element
x P (b : a)K in the transporter to the transporting “multiplication-by-x-map”.

10th January 2021 at 11:32am

Version 4.0 run 30
214 projective modules

Lemma 8.26 If a and b are two invertible ideals in the domain A, the multiplication map
µ : ab A b Ñ ab is an isomorphism. Each A-linear map from a to b is a homothety; that is, the
canonical inclusion ι is an isomorphism (b : a)K » Hom A (a, b). In particular, it holds true
that a´1 = ( A : a)K » Hom A (a, A).

Proof: The multiplication map µ is surjective by definition of products of ideals. Now,

localizing in the multiplicative set S = Azt0u, we find S´1 a = S´1 b = K, the fraction
field of A, and hence S´1 (ab A b) » S´1 abK S´1 b = K. Thus S´1 µ is an isomorphism.
It follows that S´1 ker µ = 0, and the kernel ker µ is a torsion module. Both a and b are
projective, and by ii) of Proposition 8.16 on page 208 their tensor product ab A b is also
projective. It is therefore a direct summand in a free module and consequently torsion
free since A is an integral domain by assumption. We conclude that ker µ = 0.
Each element x from (b : a)K gives by multiplication a map a Ñ b, and the assertion
in the second claim is that every map a Ñ b is shaped like this. This hinges on the
simple fact that two elements a and b from a commute so that x = φ( a) a´1 will not
dependent on the element a P a, and hence φ will be the homothety by x. Indeed,
because φ is A-linear, it follows that

aφ(b) = φ( ab) = φ(ba) = bφ( a).

Thus φ( a) a´1 = φ(b)b´1 . o

(8.27) The invertible fractional ideals are not a sparsely populated outskirt in the land
of rank one projectives; on the contrary, each isomorphism class of projective rank one
modules contains invertible ideals. Any non-zero element in the dual P˚ is a map
P Ñ A whose localization S´1 φ in the multiplicative set S = Azt0u is an isomorphism.
It ensues that ker φ is killed by a non-zero element, and this is only possible if ker φ = 0
(P is contained in a free A-module and A is a domain). Hence P is isomorphic to its
image φ( A), which is an ideal.
The invertible ideals are invertible modules, and if a and b belong to the same class
in Cl ( A), they are isomorphic A-modules; indeed a = f ¨ b for some f P K. Hence there
is a natural map Cl ( A) Ñ Pic A that sends the class of an ideal to its isomorphism class.
By what we just did it is surjective, and from Lemma 8.26 one deduces painlessly that it
is an injective group homomorphism. Hence:

Proposition 8.28 If A is a domain, the Picard group Pic A and the ideal class group Cl ( A)
are isomorphic.

Exercise 8.8 Let a be an invertible ideal which is generated by two elements. Show
that A ‘ A » a ‘ a´1 . Hint: Copy the staging in the proof of Proposition 8.24 that i) ñ
ii). M
Exercise 8.9 A sequence x, y of two elements in a domain A is said to be regular if

10th January 2021 at 11:32am

Version 4.0 run 30
 examples 215

ax = by implies that a = αy and b = αx for some α P A. For instance, two elements

in a udf without common factors form a regular sequence. Let a Ď A be an ideal in a
domain that contains a regular sequence of two elements, show that Hom A (a, A) = A.
M

8.4 Examples
Modules over principal ideal domains
Any submodule of a free module over a pid is free, which is a rather rare property for
a ring to have—ideals in a domain, for instance, are free if and only they are principal—
it holds unconditionally, but we shall prove it merely for modules of finite rank to avoid
diving into the deep waters of transfinite induction. A simple proof for the non-finite
case may be found in Kaplansky’s book ([?]) (which of course later was superseded
Bass’s general result), and for those who would appreciate a transfinite swim, we have
included an exercise with hints. It follows that a principal ideal domains enjoy the
property that all projective modules are free; among the finitily generated modules even
the torsion free ones will be free.
The class of finitely generated modules over a pid is one of the very rare classes of
modules which are completely classified up to isomorphism. This includes the classical
“Main theorem for finitely generated abelian groups”, which states that such a group M,
up to isomorphism, decomposes as a direct sum of cyclic groups; that is, one has

M » Zν ‘ Z/pi i Z,
à ν

ν
where ν is non-negative integer (the rank of M), the pi i ’s are prime powers; and of
course, the sum is finite. Abelian groups that are not finitely generated can be extremely
complicate and are a largely unexplored part of the mathematical world—even the
apparently simplest cases; i. e. subgroups of Q ‘ Q, seem to form an impenetrable
jungle.
(8.29) We are mainly concerned with rings which are principal ideal domains. However,
the case of their big brothers, the Bézout rings, are of considerable interest—if for nothing Bézout rings (Bézout
else, functions holomorphic in an open domain form a Bézout ring—and as working ringer)

with Bézout rings adds no complications, we shall do that. A Bézout ring is a ring all
whose finitely generated ideals are principal.

Theorem 8.30 Let A be a Bézout rings; that is, a ring where each finitely generated ideal is
principal, and let M be a finitely generated A-module.
i) If M is torsion free, then M is free;
ii) If M is projective it is free, in particular it holds that Pic A = 0.

10th January 2021 at 11:32am

Version 4.0 run 30
 216 projective modules

Since the property of being torsion free obviously is passed to submodules, one has
the corollary that every finitely generated submodule of a free module over a Bézout
ring is free; note however, that when A is a pid, the requirement that the submodule be
finitely generated is automatically fulfilled.
We we shall need the following general lemma.

Lemma 8.31 Assume that M is a finitely generated non-zero torsion free module over a domain
A. Then there are non-zero A-linear maps φ : M Ñ A.

Proof: As usual K denotes the fraction field of A. The module M being torsion free
will be a submodule of the non-zero K-vector space Mb A K which spans Mb A K as a
K-vector space, and thus we may find a non-zero K-linear map ψ : Mb A K Ñ K. As M
spans Mb A K, the map ψ 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 ψ(mi ). Then φ = a ¨ ψ does the job. o
The lemma is not generally valid for modules requiring infinitely many generators; for
instance, it holds true that HomZ (Q, Z) = 0; indeed, the image of an element from Q
will be an integer divisible by any other integer, and zero is the only such integer.
Proof of Theorem 8.30: We proceed by induction on the rank of M. Any non-zero
A-linear map φ : M Ñ A has an image which is a principal ideal since A is Bézout and
M finitely generated, and therefore the image is isomorphic to A. We have thus the
split exact sequence:

/N /M /A / 0.
φ
0

The kernel N is torision free and obviously of rank one less than the rank of M.
Induction applies, and N is free. But M being isomorphic to N ‘ A is therefore free as
well. o
(8.32) The classification result of finitely generated modules over a Bézout ring A is
Henry John Stephen established by showing the apparently stronger result that every map between free
Smith (1826–1983) modules of finite rank over A, or equivalently any matrix with elements from A, can
Irish mathematician be diagonalized. The arche-typical case of the all-important matrices with integral
coefficients, were treated back in 1861 by the Irish mathematician Henry John Stephen
Smith. The case of matrices of holomorphic functions was solved by Joseph Wedderburn
in 1915. His proof relies only on the base ring being Bézout and is the one we shall
present.

Theorem 8.33 Let E and F be free modules of finite rank over a Bézout ring A. Any map
φ : F Ñ E can be represented by a diagonal matrix.

Joseph Wedderburn
(1882–1948)
Scottish mathematician
10th January 2021 at 11:32am
Version 4.0 run 30
 examples 217

Proof: The main observation is a continuation of the stage set in the previous proof.
For each surjective map π : E Ñ A we may form the commutative diagram

/ F1 /F /A /0
ρ
0
φ| F 1 φ
  
0 / E1 /E /A /0
π

where E1 and F1 denote the kernels of π and φ ˝ π respectively, and where the rightmost
square is constructed as follows: The image of π ˝ φ is a principal ideal, and if a
generator g for it, put ρ = g´1 ¨ (π ˝ φ). The rank of E1 is one less than that of E so by
induction, we may find bases for E1 and F1 in which φ|F1 is represented as a diagonal
matrix.
Chose a basis tei u for E, and let f be an element of F which is part of a basis and
which does not map to zero in E (the zero map is trivial to treat). The expansion of
ř ř
φ( f ) in the basis takes the form φ( f ) = ci ei = d i bi ei with d being the greatest
ř
common divisor of the ci ’s. Then i ai bi = 1 for appropriate ring elements ai . Introduce
ř ř
the projection π : E Ñ A by the formula π ( xi ei ) = i xi ai . If d = 1, it holds that
π (φ( f )) = 1, and φ( f ) (respectively f ) forms a basis for E (respectively F) together
with any basis for E1 (respectively F1 ); thus in that case φ is represented by a diagonal
matrix (by induction on the rank of E).
In case d ‰ 1, we extend f to a basis f , f 2 , . . . , f s for F with f 2 , . . . , f s being one for
1
F . The trick is to factor φ as a product φ = ψ ˝ τ with both ψ and τ having diagonal
ř
matrices. To that end, let ψ : F Ñ E be defined by ψ( f ) = i bi ei and ψ( f i ) = φ( f i ) for
i ě 2, and τ : F Ñ F by τ ( f ) = d f and τ ( f i ) = f i for i ě 2. Obviously τ has a diagonal
matrix in any basis for F, and ψ has one by the first part of the proof. o
(8.34) As a corollary of the diagonalization theorem one deduces the classification of
finitely presented modules over principal ideal domains.

Theorem 8.35 (Main theorem for modules over pid’s) Every finitely generated module
M over a principal ideal domain A is isomorphic to a direct sum of cyclic modules. More precisely,
it holds true that
à ν
M » νA ‘ A/pi i A (8.2)
i

where the pi ’s are irreducible elements in A. The integer v and the integers νi ’s are unambigu-
ously determined by the isomorphism class of M and the irreducible elements pi ’s are unique up
to association.

Since A is a pid, the hypothesis that M is finitely presented in Theorem 8.33, may be
weakened to M being finitely generated; indeed, in that case every submodule of a

10th January 2021 at 11:32am

Version 4.0 run 30
 218 projective modules

free module of finite rank is of finite rank. This is ensues from the general theory of
modules over a Noetherian rings (which we soon shall develop), but an ad hoc proof is
offered in Exercise 8.10 below.
The first part of the theorem, that M is a direct sum of cyclic modules persists being
true for modules of finite presentation over Bézout rings, but the summands are not
of the form described in the second statement. The ring Ω of entire functions in the
complex plain C is a Bézout ring, but there are entire functions divisible by infinitely
many irreducibles—our old friend sin πz is one example— and if f is one, Ω/( f )Ω is
cyclic, but not of the prescribed kind.
Proof: It should intuitively clear that the cokernel of a diagonal matrix is the direct sum
of cyclic modules, but we anyhow present a formal proof. Chose a finite presentation
for M
φ π
F E M 0. (8.3)
Furthermore, chose bases tei u and t f j u for F and E in which the matrix of φ is diagonal
and let φ( f i ) = ai ei , where ai might be zero (by adding or discarding summands in F
that map to zero, we may arrange it so that E and F are of the same rank). We claim
that the submodules Mi = Aπ (ei ) generated by the images of ei ’s in M lie split in M;
indeed, fix and index j and define ψ : E Ñ A/( a j ) by φ(ei ) = δij . Then ψ ˝ φ = 0, and
hence ψ passes to M and gives a section to the inclusion of M j in M.
This shows that M is a direct sum of cyclic modules of the form A/( f ) A. If
f = pq with p and q elements from A without common factors, one may write 1 =
ap + bq, and one easily verifies that ap and bq act as orthogonal idempotents in A/( f ) A.
˚ This is the only place Consequently A/( f ) A decomposes as A/( f ) A » A/( p) A ‘ A/(q) A. Induction* on
where we use that A is the number of irreducible factors finishes the proof.
a pid and not merely a
Bézout rings Finally we attack the uniqueness issue. The number ν equals the rank of M and is
of course unambiguously determined. Localizing at a maximal ideal ( p) A throws away
factors not involving p, so we may assume that A is local with maximal ideal ( p) and
that the matrix φ is diagonal with all entries lying in p. We contend that two resolutions
as in 8.3 are isomorphic in the sense that they enter in a commutative diagram
φ1 π1
0 F1 E1 M 0
id M (8.4)
=
»

β α

0 F2 φ2
E2 π2 M 0.

One then easily finds pairs of bases diagonalizing the two matrices so that the diagonal
elements of the two matrices coincide. Once α is in place, β will be the restriction of to
the kernels of the πi ’s. To obtain the isomorphism α, note that πi bidk are isomorphism
as φi bidk = 0. By a now standard Nakayama-argument, any map α : E1 Ñ E2 lifting

10th January 2021 at 11:32am

Version 4.0 run 30
 examples 219

id M (that is, rendering the right hand square in (8.4) commutative) since αbidK equals
the isomorphism (π2 bidk )´1 ˝ (π1 bidk ). o

Finitely generated modules over k [t]

(8.36) The polynomial ring over a field merits to be mentioned specially:

Theorem 8.37 If k is field any finitely generated module M over the polynomial ring k[t] is of
the form
à ν
M » νk [t] ‘ k [ t ] / ( pi i )
iP I
where I is a finite set. Moreover, the non-negative integers ν and νi are unambiguously defined
by M as is the the sequence ( pi )iP I of irreducible monic polynomials. In particular, if k is
algebraically closed it holds true that
à
M » νk[t] ‘ k[t]/(t ´ ai )νi
iP I

where the ai ’s are elements in k.

Exercises
(8.10) Let A be a pid. Show that any submodule of a free module over A of finite rank
is finitely generated. Hint: Induction on the rank.
(8.11) Assume that B is an integral domain which is a finite algebra over a pid A.
Show that B is a free A-module of finite rank. If n is the rank, show that every ideal in
B can be generated by n elements.
(8.12) Jordan–Chevalley decompositon. Let k be an algebraically closed field and V a
vector space of finite dimension over k. Show that any φ : V Ñ V is a sum φ = φs + φn
with φs and φn commuting and where φs is diagonalizable and φn nilpotent. Show that
φs and φn are uniquely defined by φ. Hint: Consider V to be an k[t]-module with t
acting via φ. On a summand of the type k [t]/(t ´ a)ν put φs = a id and φn = t ´ a.
(8.13) Let m be a maximal ideal in the ring A and let ν be a natural number. Show that
any projective module over A/mν is free. Hint: Nilpotent Nakayama.
(8.14) Grothendieck’s theorem. Let k be as field and let A be the polynomial ring k [ x, y].
The aim of this exercise is to prove that every finitely generated torsion free graded
À
A-module M is of the form i A(di ). This is the algebraic version of a theorem in
algebraic geometry which sometimes is called Grothendieck’s theorem.
i) Show that every homogeneous ideal in k [ x, y] is principal. Hint: If f ( x, y) is
homogeneous, work with f ( xy´1 , 1) in the polynomial ring k[ xy´1 ].
ii) Refine the proof of Lemma 8.31 to see that there is a non-zero homogeneous
map from M to A. Hint: Start with a subset of a set homogeneous generators
of M that form a basis for Mb A K.

10th January 2021 at 11:32am

Version 4.0 run 30
220 projective modules

iii) Conclude by copying the proof of Theorem 8.30.

(8.15) Show that any finitely generated projective graded module over the polynomial
À
ring A = k[ x1 , . . . , xn ] is free; that is, it is isomorphic to a finite direct sum i A(di ). It is
true that every finitely generated projective A-module free, but this is a big theorem. It
was conjectured by Jean Pierre Serre and proved independently by Suslin and Quillen.
Hint: Graded Nakayama.
(8.16) Assume that A is a domain such that all non-zero ideals are projective. Show
À
that each finitely generated projective module P is of the form P » i ai where the sum
is finite and the ai ’s are ideals.
(8.17) Assume that A is a pid. An A-module M is a torsion module if any element m
is killed by a non-zero ring element a. If p P A is an irreducible, we let M p be the set of
elements in M killed by some power of p.
a) Show that M p is a submodule of M, and that M p X Mq = 0 if p and q are
irreducibles that are not associate.
À
b) Show that whether M is finitely generated or not, M decomposes at M = p M p
where the summation extends over a set of representatives for the irreducible
elements up to association.
c) Show that the abelian group Q/Z is a torsion group whose p-torsion part
equals the group Z p8 = Z[ p´1 ]/Z. Conclude that there is a decomposition
Q/Z = p Z p8 where the sum extends over all primes.
À

(8.18) For transfinite swimmers. Submodules of free modules over pid’s are free
regardless of the free module being finitely generated or not. In this exercise you are
guided to give a proof of this, but you are warned that it requires you be initiated in
the witchcraft of transfinite induction.
So let E be a free module over the pid A and let tei uiPτ be a basis indexed by a
ř
well-ordered set of ordinal type τ. For each ordinal σ ă τ we let Eσ = iPσ Aeι and put
Fσ = F X Eσ .
i) If σ ă τ show that the quotient Fσ+1 /Fσ will be contained in Eσ+1 /Eσ and
hence it is either zero or isomorphic to A. Conclude that Fσ lies split in Fσ+1 .
Hint: Prove that Fσ = Fσ+1 X Eσ .
ii) If τ has an immediate predecessor, show that F free.
Ť
iii) If τ is a limit ordinal, prove that F = σăτ Fσ with the union extending over
σ’s that are not limit ordinals. Conclude that F is free. Hint: Each Fσ has a
basis and lies split in Fσ+1
M
Elliptic curves III
We have already met elliptic curves at several occasions, or to be precise, one should
rather say affine elliptic curves on Weierstrass form (there are other standard forms like

10th January 2021 at 11:32am

Version 4.0 run 30
 examples 221

for instance Tate’s normal form; one example: the curve with equation y2 + 2xy ´ x3 ´
1/2 = 0, whose real points are depicted below). The coordinate ring A of such a curve
equals A = k[ x, y] with constituting relation y2 = p( x ) where p is a monic polynomial
of degree three with distinct roots, and we must assume that k is an algebraic closed
field whose characteristic is not equal to two. This example is about computing the
Picard group Pic A, which we shall describe in an ad hoc manner. There are general
theories and tools in algebraic geometry that make such an exercise easier and place it
as a small part in a wider general picture, however, we find an abecedarian approach
instructive. It gives use the opportunity to do some amusing and concrete algebra, and
one may view it as a motivation for the more advanced xyzetarian technologies.
(8.38) According to the Nullstellensatz in dimension two (Theorem 3.32 on page 78)
all maximal ideals in A are of the form I ( P) = ( x ´ a, y ´ b) where P = ( a, b) is a point
in k2 lying on the curve C; that is, b2 = p( a). This allows us to introduce an auxiliary
quadratic polynomial q( x ) by the relation

y2 ´ b2 = p ( x ) ´ b2 = ( x ´ a ) q ( x ), (8.5)

of course it depends on the point, but for simplicity this is not reflected in the notation.
(8.39) Our first ad hoc observation is that all ideals in A and consequently all ideals in
the local rings Ap , are finitely generated*; actually they are generated by at most two ˚ That is, A is

elements. Since A is a free module of rank two over the pid k[ x ], this follows directly Noetherian as we will
say later.
from Exercise 8.11.

Lemma 8.40 For each maximal ideal I ( P) the local ring A I ( P) is a pid. In particular, every
non-zero ideal in A is invertible. At the point P = ( a, b) the maximal ideal I ( P) A I ( P) is
generated by x ´ a if b ‰ 0 and by y if b = 0.

There is a simple heuristics behind this lemma. If you take a look at the real curve
depicted below, you will se it has vertical tangents near the intersection points with the

10th January 2021 at 11:32am

Version 4.0 run 30
222 projective modules

x-axis, and the projection to the y-axis is locally bijective there. This indicates that one
may use the y-coordinate as a parameter in a vicinity of such points. All other points
have neighbourhoods where the projection onto the x-axis is one-to-one and where one
thus may use the x-coordinate as a parameter, or x ´ a if you insist on the parameter
being zero at the point under consideration.
Proof: The proof is divided in two part according to b being zero or not. We first
treat the case b ‰ 0. Then y + b R I ( P) and is invertible in A I ( P) , and since y ´ b =
(y + b)´1 ( x ´ a)q( x ), we see that I ( P) A I ( P) = ( x ´ a). If b = 0. Differentiating (8.5)
shows that q( a) = p1 ( a) ‰ 0 (the polynomial p( x ) is assumed not to have multiple
roots). Therefore q( x ) R I ( p) and is invertible in A I ( P) . Hence x ´ a = y2 q( x )´1 and
I ( P) A I ( P) = (y). The maximal ideals I ( P) A I ( P) are therefore all principal, and the claim
follows from the next lemma (which is a precursor for Krull’s intersection theorem).
Finally, after having established that each local ring A I ( P) is a pid, we know that
each non-zero ideal in A is locally free of rank one; hence it is projective of rank one
and thus invertible by Proposition 8.24 on page 213. o

Lemma 8.41 Assume that B is a local ring where all ideals are finitely generated and whose
maximal ideal m is principal. Then B is a pid.

Proof: Let t be a generator for m. We contend that a = i mi = 0; indeed, assume

Ş

a ‰ 0 and let c1 , . . . , cr be a generator set for it with r minimal. It holds true that
c1 = tx for some x P a. One may write x = a1 c1 + . . . + ar cr with ai P B, and thence
c1 = a1 tc1 + . . . + tar cr . Now 1 ´ a1 t R m and is therefore invertible in B. It ensues that
c1 = t(1 ´ a1 t)´1 a2 c2 + . . . + t(1 ´ a1 )´1 ar cr , which is in flagrant contradiction with the
ci ’s forming a minimal generator set.
This done, let a be a non-zero ideal in B. Since i mi = 0, there is a largest natural
Ş

number v so that a Ď mv , and we contend that a = mv = (tν ). Indeed, let x P a

but x R mv+1 . Then x = atv , but a R m. The ring B being local, a is invertible and
consequently a = (tv ). o

10th January 2021 at 11:32am

Version 4.0 run 30
 examples 223

Proposition 8.42 Each non-zero and proper ideal a in A is a product of finitely many maximal
ideals. In other words, the Picard group Pic A is generated by the maximal deals.

Proof: Let f P a be non-zero. Then the norm N ( f ) is a non-zero polynomial which

belongs to a. If the maximal ideal I ( P) contains a, it contains N ( f ) so that the x-
coordinate of P is among the finitely many roots of N ( f ). We conclude that only finitely
many maximal ideals contain a.
For each point P P C, there is a maximal νP so that a Ď I ( P)νP (for most P it will be
zero). Localizing in I ( P), we find aA I ( P) = I ( P)νP by Lemma 8.41 above, and we may
conclude since equality is a local property. o
(8.43) Actually, much more is true. Not only is the group Pic A generated by the
maximal ideals, but as we shall see, the association P ÞÑ I ( P) is a bijection between C
and Pic Azt0u. This a very specific property of elliptic curves; most other curves only
share the property that the Picard group is generated by maximal ideals. The natural
question then arises: How is the group law in Pic A expressed in terms of points on C?
Or phrased differently: To which maximal ideal is the product of two maximal ideals
isomorphic?
Before answering that question we introduce a natural involution σ on A. It arises
from the equation y2 = p( x ) being invariant under the change of the sign of y. We shall
use an exponential notation, and denote the action of σ on element f ( x, y) from A as
f σ ( x, y) = f ( x, ´y), and for each ideal a in A the image t f σ | f P a u will be denoted
by aσ . The geometric incarnation of σ is just the reflection about the x-axis. Its action
on points P P C will be denoted by σ ( P), and σ ( a, b) = ( a, ´b). It has the three points
where C meets the x-axis as fixed points; that is, the three roots of p( x ).
We shall answer the rhetoric question above by proving:

Theorem 8.44 (The group law on an elliptic curve) Associating each point P on the
curve C with the maximals ideal in I ( P) in A yields a bijection between C and Pic Azt0u. The
group structure on C Y t0u induced from that on Pic A has the two properties:
i) ´P = σ ( P);
ii) When P ‰ ´P, it holds true that P + Q + R = 0 if and only if P, Q and R are ˚ When the ground
collinear. field is C, one may give
C the topology
Even though the Picard group has a group law induced by the tensor product and so inherited from the
has a multiplicative touch, it is customary to use an additive notation for the induced standard topology on
C2 , and thenC Y t0u
group structure on C. Notice also that the neutral element 0 does not correspond to a will be a
point on C; this reflects the fact that C is an affine curve. Adding 0 to C as “the point compactification of C.
It is even a Lie group
at infinity” (or closing C up in the projective plane) yields a so-called complete* curve
which turns out to be
which is in bijection with the entire Pic A. The group law is very geometric. To add two isomorphic to the
points P and Q on C, draw the line through them (which means the tangent to C at P if product S1 ˆ S1 of two
circles

10th January 2021 at 11:32am

Version 4.0 run 30
224 projective modules

Q = P) and determine the third intersection point it has with C; then P + Q will be the
reflection of that point through the x-axis.
(8.45) The following lemma reflects the substance of the theorem. We shall explain the
lemma, but leave it to the students to carefully carry out the deduction of the theorem.

Lemma 8.46 Let P, Q and R be three points on the elliptic curve C. The following three
statements hold true:
i) Each ideal I ( P) is projective of rank one and has I ( P)σ as inverse;
ii) The ideal I ( P) is never principal; i. e. it is never a free module;
iii) If no two of the points are conjugate, the three points are collinear if and only if the
ideal I ( P) ¨ I ( Q) ¨ I ( R) is principal;
iv) I ( P) » I ( Q) if and only if P = Q.

Proof: Proof of i): From Lemma 8.26 above we know that I ( P)b A I ( Q) = I ( P) ¨ I ( Q)
for any two points P and Q on C. So our task is to prove that I ( P) ¨ I ( P)σ is principal,
but that I ( P) never is. Let the point be P = ( a, b). We naively compute the product
I ( P) ¨ I ( P)σ :

I ( P) ¨ I ( P)σ = ( x ´ a, y ´ b)( x ´ a, y + b) =
= ( x ´ a)2 , ( x ´ a)(y ´ b), ( x ´ a)(y + b), (y2 ´ b2 ) =

= ( x ´ a)2 , 2b( x ´ a), 2y( x ´ a), ( x ´ a)q( x ) .

This shows that I ( P) ¨ I ( P)σ Ď ( x ´ a) and to arrive at an equality we have to get rid of
either of the factors 2b, 2y or q( x ). When b ‰ 0, we quickly discard the factor 2b and
are happy (this is where we use that k is of characteristic different from 2).
In case b = 0, note that q( a) ‰ 0, so q( x ) and x ´ a do not have common factors, and
we may write 1 = f ( x )q( x ) + g( x )( x ´ a) with f and g from k[ x ]. This gives the identity
( x ´ a) = f ( x )( x ´ a)q( x ) + ( x ´ a)2 g( x ), and we conclude that ( x ´ a) P I ( P) ¨ I ( P)σ .

10th January 2021 at 11:32am

Version 4.0 run 30
 examples 225

Proof of ii): Assume that I ( P) is a principal ideal, say generated by f , from which ensues
that ( x ´ a) = I ( P) ¨ I σ ( P) = ( f ¨ f σ ). Hence x ´ a = g ¨ f ¨ f σ , which is impossible
unless f is a unit because x ´ a is irreducible (Example 3.6 on page 81).
Proof of the “if implication” of iii): The geometric reason behind the third assertion,
when the three points are aligned, say they are lying on the line y = αx + β, is simply
that the line through them intersects the curve C precisely in the three points (when
possible multiplicities are take into account): the x-coordinates of these points are
determined from the cubic equation

p( x ) ´ (αx + β)2 = 0. (8.6)

Denote the three roots by a1 , a2 and a2 , but be aware that two or all three may coincide.
The corresponding y-coordinates will be αai + β. Rebabtising the points as P1 , P2 and
P3 , we contend that
(y ´ (αx + β)) A = I ( P1 ) ¨ I ( P2 ) ¨ I ( P3 ),
which is the proper algebraic way of asserting that the line intersects C exactly in
the three points. By Proposition 8.42 above the ideal (y ´ (αx + β)) is a product
ν
m1 1 ¨ . . . ¨ mrνr of maximal ideals and the exponents νi can be found by localization. We
examine (y ´ (αx + β)) A I ( P) for a general point P in C. There are three cases to handle.
i) P is not among the three Pi ’s. Then y ´ (αx + β) does not vanish at P; it does
not belong to I ( P), and (y ´ (αx + β)) A I ( P) = A I ( P) . The ideal I ( P) does not
occur as factor in (y ´ (αx + β)).
ii) P = Pi = ( ai , b) and b ‰ 0. Then y + αx + β does not vanish in Pi since no two
of the Pj ’s are conjugate. It follows that (y ´ (αx + β)) A I ( Pi ) = (y2 ´ (αx +
β)2 ) A Pi = ( x ´ ai )νi where νi is the multiplicity of the root ai .
iii) If Pi = ( a, 0). Then αa + β = 0 and y = y ´ (αx + β) + α( x ´ a). It follows
that ((y ´ (αx + β))) A I ( P) = (y), and the the corresponding multiplicity is
one. Differentiating 8.6 one sees that a is simple root of 8.6, since it is a simple
zero of p( x ).
Proof of iv): Assume that I ( P1 ) » I ( P2 ). There are three cases: If the two points are
different and non-conjugate points, and consider the line L through P1 and σ ( P2 ). If
P2 = σ ( P1 ), we note that b ‰ 0 since the two points are different, and we let L be the
tangent to C at P1 . It has the explicit equation y ´ y1 ( a)( x ´ a) ´ b where y1 ( a) = q( a)/2b.
In both cases L intersects C in a third point P3 .
In view of i) and the if part of iii) , we find

A » I ( P3 )b A I ( P1 )b A I ( P2 )σ » I ( P3 ),

which is a flagrant contradiction of assertion ii).

10th January 2021 at 11:32am

Version 4.0 run 30
226 projective modules

Proof of the “only if implication” of iii): Assume finally that I ( P1 ) ¨ I ( P2 ) ¨ I ( P3 ) is

principal, and let R be the third intersection point the line through P1 and P2 has
with C. According to what we just did, I ( P3 ) and I ( R) are then both isomorphic to
I ( P1 )´1 b A I ( P2 )´1 , and by iv) we may conclude that I ( P3 ) = I ( R); that is, R = P3 . o
Example 8.4 The Picard group of pid vanishes since by definition all ideals are principal,
and one may show that the same holds for Noetherian factorial rings. K

The algebraic Möbius band

In Exercise 3.18 on page 83 we examined the ring A = R[ x, y] with constituting relation
x2 + y2 = 1; which is the ring of real polynomial functions on the unite circle, or if one
wants, one may interpret A as the ring of trigonometric polynomials by letting x = sin θ
and y = cos θ. Exercise 3.18 was about showing that A is not a ufd, and we are about
to prove that Pic A » Z/2. There is in other words up to isomorphism just one rank
one projective module which is not free and its square is trivial. In the analogy between
projective modules and vector bundles the module P is the algebraic incarnation of the
only nontrivial real line bundle on the circle, the Möbius band.
For each point P = ( a, b) on the unit circle S, the ideal I ( P) = ( x ´ a, y ´ b) is a
maximal ideal in A being the kernel of the evaluation map at P. However, there are
other maximal ideals as described in the next lemma, they are however all principal,
generated by equations of lines not meeting S.

Lemma 8.47 The maximal ideals m in A are of the following two types:
i) Either m = I ( P) for a point P = ( a, b) on the unit circle S,
ii) or m is principal and generated by a linear form αx + βy + ρ where α, β and ρ are
real constants such that α2 + β2 = 1 and ρ ą 1.

Proof: Let m be a maximal ideal in A, and consider the extension C[ x, y] of A; if n is

a maximal ideal in C[ x, y] that contains m ¨ C[ x, y], it holds that n X A = m; likewise,
the conjugate ideal s n contains m ¨ C[ x, y], and s
n X A = m. Now, according to the
Nullstellensatz in dimension two (Theorem 3.32 on page 78) it holds true that n =
( x ´ a, y ´ b) with a and b being complex numbers such that a2 + b2 = 1. If a and b are
both real, we are in case i). If not, we evoke Exercise 3.19 on page 83 and conclude that
n = (u ´ c) with u = x + iy and c = a + ib. Obviously (u ´ c) ¨ (us ´ cs) = 1 + ccs´ ucs´ usc,
and writing u = eit and c = reθt we the find

2´1 r´1 (1 + ccs + ucs ´ usc) = cos(t ´ θ ) + (r + r´1 )/2,

which is of the form required in the lemma since r + r´1 ą 2 for all r. But n X s
n X A = m;
hence m is generated by the real linear form (u ´ c) ¨ (us ´ cs). o

Lemma 8.48 Let P and Q be two real points on the unit circle S.

10th January 2021 at 11:32am

Version 4.0 run 30
 examples 227

i) The maximal ideal I ( P) A I ( P) in the localized ring A I ( P) is a principal ideal, but I ( P)

itself is not principal;
ii) The product I ( P) ¨ I ( Q) is principal; in particular, I ( P)2 » A;
iii) It holds that I ( P) » I ( Q).

Proof: Choosing appropriate coordinates, we may assume that P = (1, 0), and since
x + 1 does not vanish at P it is invertible in the local ring A I ( P) . Hence x ´ 1 =
y2 ( x + 1)´1 in A I ( P) and ( x ´ 1, y) A I ( P) = (y). However, I ( P) is not principal since
both y and x ´ 1 are irreducible (Exercise 3.18) and neither is a factor of the other.
For the second statement, we choose coordinates so that two points are ( x ´ a, y ´ b)
and ( x + a, y ´ b) (just see to the x-axis being parallel to the line joining P and Q or
to the tangent to S at P if P = Q). We find by an abecedarian manipulation, stupidly
multiplying out and using the identity x2 + y2 = a2 + b2 , that

( x ´ a, y ´ b) ¨ ( x + a, y ´ b) = (2b(y ´ b), 2x (y ´ b), 2a(y ´ b), (y ´ b)2 ),

and this ideal is equal to (y ´ b); indeed, a and b are never simultaneously zero. o
We conclude

Theorem 8.49 One has Pic A » Z/2Z generated by the class of the maximal ideal I ( P) for
any point P on the unit circle S.

Proof: The proof of Proposition 8.42 on page 222 goes word by word through in
the present case so that Pic A is generated by the classes of the maximal ideals. By
Lemma 8.48 above it holds that 2[ I ( P)] = 0, and thus [ I ( Q)] = ´[ I ( P)] = [ I ( P)]. o
(8.50) There are clear and simple heuristic geometric explanations of these results. Any
two points on the circle are connected by a real line intersecting the circle precisely
in the points, so the product of the corresponding maximal ideals is generated by the
linear form defining the line. In a similar fashion, the tangent to the circle at a point
does not intersect the circle elsewhere, hence the maximal ideals are two-torsion. Lines
that do not intersect S intersect the complex curve x2 + y2 = 1 in C2 in two conjugate
points, whose maximal ideal therefore have a product generated by the corresponding
linear form, and these are the "other" maximal ideals from case ii) in Lemma 8.47.

Exercises
(8.19) Let C be an elliptic curve as in Subsection 8.36 above . Let A be the coordinate
ring and K its fraction field. Let furthermore let P and Q be points on C.
a) Show that I ( P) ‘ I (´P) » A ‘ A;
b) More general show there is an isomorphism I ( P) ‘ I ( Q) » A ‘ I ( P + Q).
Hint: There is a natural map I ( P) ‘ I ( Q) Ñ I ( P) + I ( Q); examine its kernel.

10th January 2021 at 11:32am

Version 4.0 run 30
228 projective modules

(8.20) Let A be a ring. A finite free resolution of an A-module M is an exact sequence

0 / Fr / Fr´1 / ... / F1 / F0 /M /0

where each Fi is a free module. An A-module M is said to be stably free if M ‘ νA is

free for some non-negative number ν. Show a stably free projective A-module having a
finite free resolution is free.
(8.21) With the notation from subsection above about elliptic curves (Paragraph 8.45
etc), consider the two matrices
! !
y ´ b ´q( x ) y + b q( x )
φ= and ψ =
´( x ´ a ) y + b x´a y´b

with coefficients from A. Show that φψ = ψφ = 0 and that the complex

/ A‘A / A‘A / A‘A / ...

φ ψ φ ψ
...

which extends infinitely in both directions, is exact. Show that coker φ » I ( P) and use
(part of) the complex above to exhibit an infinite free resolution of I ( P). Show that I ( P)
is not stably free.
M

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 9

Chain conditions

One of the great moments of mathematics was the appearance of Emmy Noether’s
revolutionary 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 have turned out to be extremely useful,
and today they permeate both commutative and non-commutative algebra.

9.1 Noetherian modules

We introduced the concept of chains in partially ordered sets already when discussing Emmy Noether
(1882–1935)
Zorn’s lemma (in Theorem 2.46 on page 49). Recall that a chain C in a partially ordered
German mathematician
set Σ is just a linearly ordered subset; that is, a set such that any two members of the
subset C are comparable.
(9.1) In the present setting, when studying modules over a ring A, we give the term
chain a more restrictive meaning. The chains we shall consider will all be countable and
well ordered. Two sorts of chains will be distinguished, ascending and descending ones. Ascending chains
An ascending chain in M will be a sequence of submodules tMi uiPN0 such that every (oppstigende kjeder)

term Mi is contained in in the successor Mi+1 ; or written out in a display, it is a chain Descending chains
of inclusions like (nedstigende kjeder)
Eventually constant
M0 Ď M1 Ď . . . Ď Mi Ď Mi+1 Ď . . . .
chains (terminerende
Similarly, a descending chain is a sequence tMi uiPN0 of submodules fitting into a chain of kjeder)
Noetherian modules
inclusions shaped like (noetherske moduler)

. . . Ď Mi+1 Ď Mi Ď . . . Ď M1 Ď M0 .

Such chains are said to be eventually constant or eventually terminating if the submodules
become equal from a certain point on; that is, for some index i0 it holds that Mi = M j
whenever i, j ě i0 . Common usage is also to say the chain stabilizes at i0 .
(9.2) An A-module M is said to be Noetherian if every ascending chain in M is eventually
 230 chain conditions

The Ascending Chain constant. This condition is frequently referred to as the Ascending Chain Condition
Condition (den abbreviated to acc . The module is Artinian if every descending chain terminates, a
oppstigende
kjedebetingelsen) condition also called the Descending Chain Condition with the acronym dcc.
Artinian modules A ring A is called Noetherian if it is Noetherian as module over itself, and of course,
(artinske moduler) it is Artinian if it is Artinian as module over itself. The submodules of A are precisely the
ideals, so A being Noetherian amounts to ideals of A satisfying the acc, and similarly,
Noetherian and
Artinian rings A is Artinian precisely when the ideals comply with the dcc.
(noetherske og artinske (9.3) The two conditions, being Noetherian and Artinian, might look similar, but there
ringer)
is a huge difference between the two. Noetherian and Artinian 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 modules in any depth, although Artinian rings will be
discussed (in Section 9.7 below). In fact, according to a result of Yasuo Akizuki they
turn out to be Noetherian as well, they form the class of so-called finite length, and are
important both in geometry and number theory.
Emil Artin
(1898–1962) (9.4) The constituting properties of Noetherian modules is asserted in the following
Austrian theorem. It is due to Emmy Noether and appears as one of the main theorems in her
mathematician famous paper from 1921.

Proposition 9.5 (Main theorem for Noetherian modules) Let A be a ring and let M
be a module over A. The following three conditions are equivalent:
i) M is Noetherian; that is, it satisfies the ascending chain condition;
ii) Every non-empty family of submodules has a maximal element;
iii) 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 element1 . 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 extended 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 prove that every submodule N is finitely generated. To
that end, let Σ denote the set of finitely generated submodules. It is clearly non-empty
(the zero modules is finitely generated) and consequently has a maximal element N0 .
Let x P N be any element. The module Ax + N0 is finitely generated and contains N0 ,

1 Eventhough resembling Zorn’s lemma this is quite different. The acc assumption is stronger than what
Zorn’s lemma asks in that chains are required to be eventually constant not only bounded above; on the other
hand the acc places restrictions only on countable and well ordered chains. Anyhow, it is of interest that the
proposition is independent of the Axiom of Choice

10th January 2021 at 11:32am

Version 4.0 run 30
 noetherian modules 231

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
There are statements for Artinian modules that correspond to the two first claims
in the theorem, which are of a kind of order-theoretical nature (you are asked to give
a proof in Exercise 9.5 below). However, there is no substitute for the third, about
submodules being finitely generated, which draws module theory into the business.
(9.6) The Noetherian modules, as do the Artinian modules, form a subcategory of Mod A
which enjoys a strong closure property. They are what in category theory are called
thick subcategories. Submodules and quotients of Noetherian modules are Noetherian as Thick subcategories
is an extension of two, and for of Artinian modules the same holds true. (tykke underkategorier)

Proposition 9.7 Let M1 , M and M2 be three A-modules fitting in a 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 follows by a straightforward induction—finite direct sums of Noe-

therian (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 in M. Every chain
in M1 is then 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 β β´1 ( Ni ) = Ni , so that

tNi u stabilizes whenever tβ´1 ( Ni )u does. Hence, if M is Noetherian (or Artinian), so

will M2 .
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 tβ( Ni )u in M2 , and since M2 by
assumption is Noetherian (or Artinian), it stabilizes at some µ. Hence β( Ni ) = β( Nj )

10th January 2021 at 11:32am

Version 4.0 run 30
232 chain conditions

for i, j ě µ. For i, j ě max(µ, ν) this gives

Ni /Ni X M1 = β( Ni ) = β( Nj ) = Nj /Nj X M1 = Nj /Ni X M1 ,

and hence Ni = Nj . o
(9.8) The properties of being Noetherian or Artinian are retained when a module is
localized.

Proposition 9.9 Let S be a multiplicative set in the ring A and let M be an A-module. If
M is Noetherian (respectively Artinian the localized module MS is Noetherian (respectively
Artinian).

Proof: The proof is based on the simple observation that for any submodule N of
S´1 M one has S´1 (ι´1 N ) = N, where ι : M Ñ S´1 M denotes the localization map:
indeed, if an element ι(y)s´1 in S´1 M belongs to N, so does ι(y).
Now, any chain tNi u in S´1 M, 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 = S´1 (ι´1 Ni ) = S´1 (ι´1 Nj ) = Nj , and the original chain stabilizes at i0 as
well. o

Examples
(9.1) Vector spaces: A vector space V over a field k is Noetherian (or Artinian) 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 (and Artinian).
If V is not of finite dimension there will be infinite sets v1 , . . . , vi , . . . of linearly
independent vectors, and for such the subspaces Vi =ă v1 , . . . , vi ą will form a strictly
ascending chain of subspaces; hence V is not Noetherian. A similar argument shows
that neither is V Artinian: the spaces Wi =ă vi , vi+1 ¨ ¨ ¨ ą form a strictly decreasing
chain of subspaces.
(9.2) Finite product of fields: The conclusions of the preceding example extend to rings
ś
that are finite products of fields; say A = 1ďiďr k i . Modules over such rings are
À
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 9.7, 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

Exercises
ˇ (9.1) Prove the assertion just after Proposition 9.7 that if a direct sum of Noetherian
modules is Noetherian, the sum is finite and all the summands are Noetherian.
ˇ (9.2) Show that Z is a Noetherian Z-module, but that Z p8 = Z[ p´1 ]/Z is not. Show

10th January 2021 at 11:32am

Version 4.0 run 30
 noetherian rings 233

that Z p8 is an Artinian Z-module, but that Z is not. Hint: The only submodules of
Z p8 are the cyclic ones generated by the images of p´i for different i’s.
ˇ (9.3) Let φ : A Ñ B be a map of rings and let M be a B-module. Prove that if M is
Noetherian as an A-module, it is Noetherian as an B-module as well. Show by exhibiting
examples that the converse is not true in general, but holds true when φ is surjective.
ˇ (9.4) Show that a direct sum of finitely many simple modules is both Noetherian and
Artinian.
ˇ (9.5) Let M be an A-module. Show that the following two claims are equivalent:
i) M is an Artinian module;
ii) Every non-empty family Σ of submodules of M has a minimal element.
M

9.2 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 Essential finite type
finite type over a field k (or over a Noetherian ground ring); that is, they are localizations (essensielt av endelig
type)
of finitely generated k-algebras (or algebras over the ground ring). All these rings are
Noetherian. Hilbert’s basis theorem ensures that algebras finitely generated over a
Noetherian base are Noetherian, and by Proposition 9.9 above localizing a ring preserves
the property of being Noetherian.
Be aware that although having lots of nice properties, Noetherian rings can be
treacherous and show an unexpectedly bad behaviour. Even among local Noetherian
rings, which usually are rather tame and well-behaved animals, one finds example with
strange properties.
(9.10) The ring A being Noetherian means that any ascending chain of ideals eventually
terminates. Applying Proposition 9.5 on page 230 to the ring A itself while remembering
that the submodules of A are precisely the ideals, we arrive at the following:

Proposition 9.11 (The main theorem for Noetherian rings) For a ring A the follow-
ing three conditions are equivalent:
i) A is Noetherian; that is, the ideals in A comply with the ascending chain condition;
ii) Every non-empty family of ideals in A has a maximal element;
iii) Every ideal in A is finitely generated.

It is trivial that fields are Noetherian, and shortly we shall see that polynomial rings
over fields are Noetherian too; this is a special case of the celebrated Hilbert’s Basis

10th January 2021 at 11:32am

Version 4.0 run 30
 234 chain conditions

Theorem. Other examples of Noetherian rings are the principal ideal domains, where
ideals are not only finitely generated, but generated by a single element.
(9.12) Quotients of Noetherian rings are Noetherian (Proposition 9.11 on the previous
page), but not necessarily subrings. Non-Noetherian domain are obvious examples:
they are contained in their fraction fields, which are Noetherian. A subtler example will
be given below (Example 9.4).

Proposition 9.13 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 9.7 on
page 231 that nA is Noetherian; indeed, one obtains nA by successive extensions of A by
itself. By Proposition 9.7 again, all quotients of nA, in particular M, will be Noetherian.
Finally, Noetherian modules are finitely generated since all their submodules are. o
(9.14) A converse to Proposition 9.13 does not hold in the sense that rings may have
non-zero Noetherian modules without being Noetherian themselves; 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 Fundamental
Existence Theorem for Ideals (Theorem 2.49 on page 49). These examples are in some
way illustrative; any Noetherian module over a non-Noetherian ring must have a non-
trivial annihilator ideal; or phrased in another way, if A has a Noetherian module with
Faithful modules global support— what is also called a faithful module—it is a Noetherian ring.
(trofaste moduler)
Proposition 9.15 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 that

sends x to the tuple ( x ¨ x1 , . . . , x ¨ xr ). If x kills all the xi ’s, it kills the entire module M,
since the xi ’s form a generating set, and we may 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 9.7 above. o

Minimal primes in Noetherian rings

(9.16) The minimal prime ideals of an ideal a in a ring A (that is, the prime ideals
minimal among those containing a) are in the front line when one starts examining a.
Geometrically, their corresponding closed subsets are the irreducible components of of
Noetherian induction Spec A. And an important feature—in fact a basic finiteness property—of Noetherian
(noethersk induksjon) rings is that the set of minimal primes of any ideal is finite. For this reason it is
appropriate and natural to include a proof at this stage, which also has the bonus of
furnishing a nice and simple illustration of an ever recurring technique called Noetherian

10th January 2021 at 11:32am

Version 4.0 run 30
 noetherian rings 235

induction: If a statement about ideals is not true for all ideals, the set ideals for which it
fails—the gang of bad guys, so to say—is then non-empty and consequently will have a
maximal member, and working with this maximal scoundrel, one tries to establish a
contradiction.
Remembering that the radical of an ideal is the intersection of its minimal primes, one
obtains—as a prelude to the general theory of primary decompositions—the corollary
that radical ideals in Noetherian rings equal the intersections of their finitely many
minimal prime ideals; that is, one has
‘
a = p1 X . . . X pr (9.1)

with the pi ’s being the minimal primes of a. And, of course, the minimal primes are
unambiguously determined by the ideal a, and moreover, no inclusion relation among
them persists.

Proposition 9.17 Each ideal a in a Noetherian ring has only finitely many minimal prime
ideals.
Proof: Let Σ be the set of proper ideals in A having infinitely many minimal prime
ideals. If Σ is non-empty, it has maximal member, say a. Obviously a is not a prime
ideal, so there are elements x and y neither lying in a, but whose product xy belongs to
a. Then a + ( x ) and a + (y) are proper ideals (their product is contained in a) strictly
larger a, and consequently each has merely finitely many minimal primes. Any prime
ideal containing a contains either x or y, hence each minimal prime ideal of a is either
among the finitely many minimal primes of a + ( x ) or the finitely many of a + (y). o

Corollary 9.18 A radical ideal in a Noetherian ring is an irredundant intersection of finitely

many prime ideals. In particular, the nil-radical is the intersection of finitely many prime ideals.
The involved prime ideals are unique.

Examples
Examples of Noetherian rings will soon abound, so here we merely give a few examples
of non-Noetherian rings, noting the words of wisdom of Sun Tzu: “If you know the
enemy and know yourself”.
(9.3) The obvious example of a non-Noetherian ring is the ring A[ x1 , x2 , . . .] of polyno-
mials in infinitely many variables over any ring A. Clearly, the chain of ideals

( x1 )Ă ( x1 , x2 )Ă . . . Ă ( x1 , x2 , . . . , x i )Ă . . .

does not stabilize.

(9.4) 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

10th January 2021 at 11:32am

Version 4.0 run 30
236 chain conditions

polynomial ring Q[ x ]. The simplest example is even a subring of the ring Z[ p´1 ][ x ]
where p is a natural number greater than one. It is formed by those polynomials in
Z[ p´1 ][ x ] that assume an integral value at zero; that is, the polynomials P( x ) such that
P(0) P Z. In this ring A one finds the following ascending chain of principal ideals

( p´1 x )Ă ( p´2 x )Ă . . . Ă ( p´i x )Ă . . . ,

which does not stabilize. Indeed, if p´(i+1) x P ( p´i x ), one would have p´(i+1) x =
P( x ) p´i x for some polynomial P( x ) P A. Cancelling p´i x would give p´1 = P( x ),
which contradicts that P(0) P Z.
(9.5) A large class of important non-Noetherian rings are formed by the rings H (Ω)
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 H (Ω) 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.
K

Exercises
ˇ (9.6) Let A Ď Q be any proper subring. Show that the polynomials in Q[t] assuming
values in A at the origin, is not Noetherian.
(9.7) Let tAi uiP I be a family of Noetherian rings all different from the null ring.
ś
a) Show that the product i Ai is Noetherian when I is finite.
ś
b) Show that the product i Ai is not Noetherian when I is infinite.
(9.8) The ring of numerical polynomials. A polynomial p( x ) in Q[ x ] is called a numerical or
integral polynomial if it assumes integral values on the integers. Every such polynomial
has a unique expansion
ÿ x
p( x ) = cν
ν

where ( νx) = x ( x ´ 1) . . . ( x ´ ν)/ν! and where the cν ’s are integers.

Show that the ring Int(Z) of numerical polynomials in Q[ x ] is not Noetherian.
Hint: Show for instance that the ideal m = t f P Int(Z) | f (0) is even u is not finitely
generated.
(9.9) The ring of algebraic integers. 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

10th January 2021 at 11:32am

Version 4.0 run 30
 a structure theorem for modules 237

where the coefficients ai are integers. Show that A is not Noetherian. Hint: For instance,
?
the principal ideals ( 2n 2) form an ascending sequence that does not terminate.
M

9.3 A structure theorem for modules

As an illustration of the strength and elegance the Noetherian method can show, we
offer a structure theorem for finitely generated modules over Noetherian rings—it does
not reveal the finer features of a module, but rather describes the overall structure.
Every such module is obtained by a series of successive extensions of cyclic modules
shaped like A/p with the p’s being prime ideals. It has the important consequence that
all additive invariants2 of Mod A are determined by their values on the quotients A/p
with p prime.
(9.19) The proof of the structure theorem is built on the following result which is of
independent (and fundamental) importance and will be use later.

Proposition 9.20 (Maximal annihilators) Assume that A is a 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.
The prime ideals that are annihilators of an element of M are said to be associated to M, Associated prime ideals
and the set they form is denoted Ass M. (assosierte primidealer)

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 9.21 Any non-zero module over a Noetherian ring contains a module isomorphic
to A/p for some prime ideal p.
Proof: Observe that the set of annihilators of non-zero elements is non-empty and has
a maximal element since A is Noetherian. Then cite Proposition 9.20 above. o
Exercise 9.10 Show the slight extension of Proposition 9.20 that if Ann x is maximal
among annihilators of non-zero elements from M that are contained in a fixed prime
ideal p of A, then Ann x is prime. M
(9.22) The ground is now well prepared for the promised structure theorem; here it
comes:
2 An additive invariant is a map χ : Mod Ñ G where G is a commutative monoid, such that χ ( M ) =
A
χ( M1 ) + χ( M2 ) each time
0 / M1 /M / M2 /0
is an exact sequence.

10th January 2021 at 11:32am

Version 4.0 run 30
238 chain conditions

Theorem 9.23 (Structure of modules) Let A a Noetherian ring and let M be a non-zero
A module. Then M is finitely generated if and only if it possesses a finite ascending chain of
submodules tMi u0ďiďr with M0 = 0 and Mr = M 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 r.

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 9.21 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

Associated prime ideals and the support

In Section 7.5 of Chapter 7 we introduced the notion of support Supp M of an A-module
M as the subset of Spec A formed by the prime ideals p such that Mp is non-zero. Over
Noetherian rings there is an intimate relation between the support of a module and the
set Ass M of associated primes. The latter is always a subset of the former, and they
have the same minimal elements, so if Supp M is closed, for instance, if M is finitely
generated, it equals the closure of the set Ass M.
(9.24) The result follows here; note that we do not require M to be finitely generated,
but A needs to be Noetherian.

Proposition 9.25 Let A be a Noetherian ring and M an A-module.

i) Then Ass M is non-empty and Ass M Ď Supp M.
ii) The sets Ass M and Supp M have the same minimal elements.

Proof: That Ass M is non-empty is just a restatement of Corollary 9.21 (the combination
of maximal annihilators being prime and that A being Noetherian ensures that they
exists). If p = Ann x is a prime annihilator, the element x survives in Mp according to
Lemma 7.46 on page 196, and so p belongs to Supp M.
We proceed proving the second statement and let p be minimal in the support
Supp M. Consider the Ap -module Mp . It is a non-zero module whose support is
reduced to the maximal ideal pAp since p is minimal in Supp M, so in view of statement
i) the maximal ideal pAp must be associated to Mp . The maximal ideal is therefore
an annihilator, say Ann Ap x, and moreover, the element x may be chosen to lie in M.
It holds that p = pAp X A = Ann Ap x X A. We contend that there is an s R p so that
p = Ann A sx. Indeed, let a1 , . . . , ar be generators for p; for each there is an si with

10th January 2021 at 11:32am

Version 4.0 run 30
 hilbert’s basis theorem 239

si ai x = 0. If s = s1 ¨ . . . ¨ sr , it holds that p Ď Ann sx, and consequently p = Ann sx

because evidently Ann sx Ď p. We conclude p is a prime annihilator, hence it belongs to
Ass M.
For the reverse inclusion, pick a minimal element p from Ass M and let q Ă p be
a prime ideal. If Mq ‰ 0, there must be an element y P M so that Ann y Ď q, but as
A is Noetherian, there is a maximal such annihilator ideal, which according to the
Principle of Maximal Annihilators (9.20 on page 237), or rather the extended version in
Exercise 9.10, is prime; and this contradicts that p is minimal in Ass M. o
(9.26) The union and the intersection of the associated prime ideals have a special
significance for the module M, quite parellel to what is the case for ideals. One has:

Lemma 9.27 Let A be a Noetherian ring and M a finitely generated A-module. Then
Ş ‘
i) pPassM p = Ann M;
Ť
ii) pPAss M p is the set of zero-divisors in M.

Proof: The first follows since Supp M and Ass M have the same minimal prime ideals,
Ť ‘
and since pPSupp M p = Ann M when M is finitely generated (Proposition 7.55 on
page 199). The second follows by the observation that a zero-divisor x lies in an
annihilator, hence in a maximal annihilator, which is prime and belongs to Ass M. o

Exercises
(9.11) Let R = k[t, x1 , x2 , . . .] with constituting relations xi = txi+1 for i ě 1 and let
A = Rm where m is the maximal ideal m = (t). Consider the module M = A/( x1 ) A.
Show that Ass M = H, but that M is of global support; i. e. Supp M = Spec A.
Hint: A has two non-zero prime ideals mA = (t) A and p = iě1 mi .
Ş

9.4 Hilbert’s Basis Theorem

There is almost an infinity of strong results about Noetherian rings, unfortunately we
have time to treat too few of them. As a beginning, in this section we shall discuss two.
In addition to Hilbert’s basis theorem, we treat a criterion for rings being Noetherian
due to I.S. Cohen. and in the next section we follow up and 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 David Hilbert
geometry. It guarantees that most rings appearing in those parts of mathematics are (1862–1943)
Noetherian. However the name originate from the content of the theorem, that any German mathematician

10th January 2021 at 11:32am

Version 4.0 run 30
 240 chain conditions

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 formulation was slightly different
from the modern one (the term Noetherian was of course not in use; Emmy Noether was
only eight years old at the time), and the context was confined to polynomial rings over
fields or specific rings 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
Paul Albert Gordan general results, and it opened up the path to modern algebra. Part of the mythology
(1837–1912) surrounding the theorem is the exclamation by the “König der Invariant Theorie” Paul
German mathematician Gordan: “Das ist nicht Mathematik, das ist Theologie!”. The truth is that Hilbert had
proved in a few pages what Gordan and his school had not proved in twenty years.
(9.28) 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 9.29 (Hilbert’s Basis Theorem) If A is a Noetherian ring, then so i the polyno-
mial ring A[ x ].

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 9.30 The polynomial ring A[ x1 , . . . , xn ] over a Noetherian ring A is Noetherian.

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 9.31 Any algebra finitely generated over a Noetherian ring is Noetherian.

Finally, the last corollary we offer before giving the proof of Hilbert’s Basis Theorem,
combines that theorem with Proposition 9.9 on page 232 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 9.32 Any ring essential of finite type over a Noetherian ring is Noetherian.

Proof of Hilbert’s basis theorem: Let a be an ideal in A[ x ] 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

10th January 2021 at 11:32am

Version 4.0 run 30
 hilbert’s basis theorem 241

in A, and they form an ascending chain, which since A is assumed to be Noetherian,

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 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 ν
among the bad guys; that is, among those polynomials in a that do not belong to the
ideal generated by the f i ’s. If the leading coefficient of f is a, it holds that a P aν and we
ř
may write a = j b j ai j with the polynomials f i j whose leading coefficient is ai j , all are of
degree at most the degree of f . The numbers d j = (deg f ´ deg f i j ) are all non-negative,
and we may thus form the polynomial f ´ j b j x d j f i j . It is of degree less than deg f : the
ř

term of degree deg f vanishes by the very choice of the b j . It does not lie in the ideal
generated by the f i ’s since f does not, and that contradicts the minimality of deg f . o

Cohen’s criterion
One may wonder if there are conditions only involving prime ideals that ensure a ring
being 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 xi ’s in the ring k[ x1 , x2 , . . .] of polynomials in countably many
variables. The quotient k[ x1 , x2 , . . .]/m2 has only one prime ideal, namely the one
generated by the images of all the variables, but is not Noetherian since that 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.
(9.33) We begin with stating a lemma about maximal ideals that are not finitely
generated; it joins the line of results of the type asserting that ideals maximal subjected
to some specific condition are prime:

Lemma 9.34 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 9.35 (Cohen’s criterion) Assume that all prime ideals in the ring A are
finitely generated. 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 element, say a; indeed,
Ť
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 9.11 on page 233)

10th January 2021 at 11:32am

Version 4.0 run 30
 242 chain conditions

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 9.34: 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 + ( a) and c1 = a + ( a1 ). These ideals both
contain a properly and are therefore finitely generated by the maximality of a, hence
the product cc1 is finitely generated. Moreover, it holds that cc1 Ď a, because aa1 P c. The
quotient c/cc1 is a finitely generated module over the Noetherian ring A/a and contains
a/cc1 . Hence a/cc1 is finitely generated, and by consequence a is finitely generated as
well since cc1 is finitely generated. o

Exercises
(9.12) Hilbert Basis Theorem for Power Series. Let A be a ring. The purpose of this
exercise is to prove that if A is Noetherian, so is the power series ring AJxK.
a) Let tgi uiPN0 be a sequence of power series in AJxK. Show that i xi gi is a well
ř

defined power series in AJxK.

Let φ : AJxK Ñ A be the map that sends x to zero; i. e. a power series f ( x ) is sent to the
constant term f (0). Let p be a prime ideal in AJxK and assume that φ(p) = ( a1 , . . . , ar ).
Chose elements f i from p so that f i (0) = ai .
b) If x P p, show that p = ( x, f 1 , . . . , f r ).
c) If x R p, show that p = ( f 1 , . . . , f r ). Hint: Given f P p; for each i recursively
construct a power series hi so that f = h1 f 1 + . . . + hr f r .
d) Conclude by Cohen’s criterion that AJxK is Noetherian whenever A is.
M

9.5 Krull’s intersection theorem

The German mathematician Wolfgang Krull was one of the greatest contributor to the
development of algebra in the years between the two World Wars, and in this section
we shall discuss one of his more famous results, the so-called “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 not have common elements apart from 0; that is, it
Ş
holds true that ν aν = 0.
Wolfgang Krull
There are several proofs of Krull’s intersection theorem, and the one we give is
(1899–1975)
among the shortest possible with the means at hand at the present stage of the course.
German mathematician
˚ The general version
There is another really simple and elementary proof for the case of the ring* itself due
to Hervé Perdry which we present as Exercise 9.19.
concerns an A-module.
Example 9.6 To motivate and illustrate the reasons behind Krull’s theorem, let us
consider the ring of complex polynomials C[ x1 , . . . , xr ] and a point a = ( a1 , . . . , ar ) in

10th January 2021 at 11:32am

Version 4.0 run 30
 krull’s intersection theorem 243

Cr . The ideal m = ( x1 ´ a1 , . . . , xr ´ ar ) consists precisely of the polynomials that vanish

at a, and the members of the powers mν are those that 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. Another obvious example is found in the
ring Z of integers where of course it holds that ν ( pν ) = 0 for any integer p, for the
Ş

simple reason that no integer has infinitely many factors. Of course, Krull’s result is
a vast generalization of these prosaic examples; ideals in local Noetherian rings are
infinitely more intricate than maximal ideals in a ring of complex polynomials or than
principal ideals in Z. K
(9.36) The show begins with a technical lemma, and again submodules maximal
subjected to a specific condition enter the scene, but this time chiefly as catalysts. In
general if N Ď M is a pair of a module and a submodule and a an ideal in A, the
intersection aM X N is not always contained in aN: elements in N might be divisible in
M by elements from a, but not in N. However under appropriate finiteness conditions,
if an element in N is “sufficiently divisible” in M, it will be divisible in N as well; one
has the important:

Lemma 9.37 Let a Ď A be a finitely generated ideal and M be a Noetherian A-module with a
submodule N. If K is a submodule of M maximal subjected to the condition that K X N = aN,
then aν M Ď K for a sufficiently large ν P N. In particular it holds true that aν M X N Ď aN.

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 ( x ν M + K ) X N = aN.
The crucial inclusion is ( x ν M + K ) X N Ď aN, the other being clear as aN = K X N.
Now, the transporter submodules (K : xi ) form an ascending chain, which since M
is Noetherian, stabilizes at say ν; so that (K : x ν ) = (K : x ν+1 ). If y = x ν m + k with
m P M and k P K, is a member of ( x ν M + K ) X N it holds that xy P xN Ď K X N from
which ensues that m P (K : x ν+1 ) since xy = x v+1 m + xk. Hence m P (K : x ν ), and
y P K X N = aN. o

Proposition 9.38 Suppose that A is a ring, that a is a finitely generated ideal in A and that
M is a Noetherian module over A. Putting N = i ai M, one has aN = N.
Ş

Proof: By the lemma there is a ν so that aν M X N Ď aN. But by construction N Ď av M

and we conclude that aN = N. o
(9.39) Combining Proposition 9.38 above with Nakayama classic, we obtain the classical
version of Krull’s intersection theorem:

Theorem 9.40 (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

10th January 2021 at 11:32am

Version 4.0 run 30
244 chain conditions

Proof: Let N = i ai M. Then aN = N after Proposition 9.38, and we may finish by

Ş

applying Nakayama’s lemma (Proposition 4.52 on page 106) since N is a submodule of

the Noetherian module M and therefore is finitely generated. o

Corollary 9.41 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 whose maximal ideal is m,
Ş

one has i mi = 0.
Ş

(9.42) In general it is not true that the intersection of successive powers of a proper ideal
vanishes even when the ring is Noetherian. Principal ideals generated by non-trivial
idempotents furnish simple counterexamples: if a = (e) with e idempotent, one has
a2 = a, and a straightforward induction shows that ai = a for all i. Hence i ai = a.
Ş

However, the powers of proper ideals in Noetherian integral domains have vanishing
intersections:

Corollary 9.43 (Krull’s intersection theorem II) Assume that a is a proper ideal in
the Noetherian integral domain A, then i ai = 0.
Ş

Proof: We combine Proposition 9.38 above by Nakayama Extended (Proposition 7.53

on page 198) and exhibit an element a P a so that (1 + a) N = 0 where N = i ai . But a
Ş

being proper, 1 + a is non-zero, and consequently N = 0 since A is an integral domain.

o

Exercises
(9.13) Let N Ď M be a pair of an A-module and a submodule. Let x P A be an element.
Prove that xM X N = xN if and only if x acts as non-zero divisor on M/N.
(9.14) Let ν be a natural number and p a prime. Describe the submodules N of Z so
that pν+1 Z X N Ď pN but pν Z X N Ę pN.
(9.15) Let A be a domain which is contained in the Noetherian domain B. Let a be an
ideal in A. Show that it either holds true that aB = B or that i ai = 0.
Ş

(9.16) Let R be the ring of real functions that are defined and C8 on an interval
ă ´e, e ą round 0. Let m be the ideal of those functions in R that vanish at the origin.
Show that m is a maximal ideal, and that i mi consists of the functions in R all whose
Ş

derivatives vanish at the origin. Give examples of such functions.

ˇ (9.17) Let A be a local ring with maximal ideal m and assume that m is a principal ideal
generated by a non-nilpotent element.
a) Prove that p = i mi is a prime ideal and that all prime ideals in A other than m
Ş

are contained in p.
b) Prove that if i mi = (0), then the powers mi are the only non-zero ideals in A;
Ş

c) Prove that A is Noetherian if and only if i mi = (0).

Ş

10th January 2021 at 11:32am

Version 4.0 run 30
 modules of finite length 245

(9.18) The aim of this exercise is to exhibit a domain A with a principal maximal ideal
m such the intersection i mi is non-zero. It is in some sense a minimal example of this
Ş

behaviour, and illustrates how Krull’s intersection theorem may fail in non-Noetherian
rings.
Let k be a field and let A be the ring k[t, x1 , x2 . . .] with constituting relations
xi = txi+1 for i P N.
a) Show that m = (t) A is a maximal ideal and that the ideal p = ( x1 , x2 , . . .)
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 = ( f ) such

that i ai ‰ (0). Show that there is a map of k-algebras A Ñ B.
Ş

(9.19) 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 prove that x P x ¨ a. Assume
Ş

that a1 , . . . , ar are generators for a.

a) Let ν P N be a natural number. Use that x P aν to prove there is a homogenous
polynomial Pν ( x1 , . . . , xr ) of degree v in A[ x1 , . . . , xr ] so that x = P( a1 , . . . , ar ).
b) Let cn be the ideal in A[ x1 , . . . , xr ] 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 homoge-
nous 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

9.6 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, and it will turn out that this is equivalent to the modules being both
Noetherian and Artinian. 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 certain maximal Maximal chains
chains of submodules—maximal in the sense that there is no room for inserting new (maksimale kjeder)

modules in the chain—the so-called composition series.

(9.44) A finite ascending chain tMi u in an A-module M, which begins at the zero
module 0 and ends at M, is called a composition series if all its subquotients Mi+1 /Mi Composition series
are simple modules. By convention simple modules are non-zero, so in particular all (komposisjonsserier)

10th January 2021 at 11:32am

Version 4.0 run 30
 246 chain conditions

the inclusions Mi Ď Mi+1 are strict. The series when displayed appears like

0 = M0 Ă M1 Ă . . . Ă Mn´1 Ă Mn = M,
The length of a
composition series
where each subquotient Mi+1 /Mi is shaped like A/mi for some maximal ideal mi in
(lengden av en
komposisjonskjede) A. The number n is called the the length of the series; it is the number of non-zero
The length of a chain constituencies. More generally, the length of any finite chain will be the number of
(lengden av en kjede) inclusions; that is, one less than the number of modules.
For a finite chain to be a composition series is equivalent to being a maximal chain;
that is, no module can be inserted to make it longer. Such a chain must start at zero
and end at M (if not, 0 or M could be adjoined), and there can be no submodules
lying strictly between two consecutive terms. Would-be insertions are submodules in
between Mi and Mi+1 , which are in a one-to-one correspondence with submodules of
Mi+1 /Mi —a submodule N corresponds to the quotient N/Mi —so each Mi+1 /Mi is
Saturated chains simple precisely when there are no intermediate submodules. The term saturated is also
(mettede kjeder) common usage for suchlike chains, but these are neither required to start at the zero
module nor to end at M.
(9.45) At several occasions in the subsequent paragraphs we shall push composition
series forwards along A-linear maps, and it is a crucial fact that they stay composition
series, though with a slight modification. To be precise, let β : M Ñ N be an A-linear
map and M = tMi u a composition series in M. The set tβ( Mi )u of images is obviously
a chain in M1 , but inclusions do not necessarily persist being strict, so there may be
repetitions in tβ( Mi )u. A part from that, tβ( Mi )u will be a composition series:

Lemma 9.46 Let β : M Ñ N and let tMi u be a composition series in M. Disregarding possible
repetitions, the chain tβ( Mi )u will be a composition series.
Proof: A simple diagram-hunt (or a snake argument) yields that γ in the diagram
below is surjective. Hence, as Mi /Mi´1 is simple, the module β( Mi )/β( Mi´1 ) is either
zero or isomorphic to Mi /Mi´1 . Consequently, when repeating terms are discarded,
the subquotient of tβ( Mi )u are all simple.

0 β ( M i ´1 ) β ( Mi ) β( Mi )/β( Mi´1 ) 0
γ

0 M i ´1 Mi Mi /Mi´1 0.

o
ˇ Exercise 9.20 Assume that α : N Ñ M is an injective A-linear map. If M = tMi u is a
composition series in M, show that the chain tα´1 ( Mi )u will be a composition series in
N after possible repeating terms have been discarded. M
(9.47) The main result of this section is that once a module has a composition series,

10th January 2021 at 11:32am

Version 4.0 run 30
 modules of finite length 247

all composition series will have the same length, and the length of any chain will be
bounded by that common number. 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 isomorphic 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 9.48 (Weak Jordan-Hölder) Assume that M has a composition series. Then all
composition series in M have the same length and any chain may be completed to a composition
series.

The common length of the composition series is called the length of the module and The length of a module
denoted ` A ( M). For modules not of finite length, that is those having no composition (lengden til en modul)

series, one naturally writes ` A ( M ) = 8. As a matter of pedantry, the zero module* ˚ This is a pure

is considered to be of finite length and its length is zero (what else?). Note that the formality: by
convention the zero
zero module is the only module of length zero, and that the simple ones are the only module is not included
ones of length one, indeed, having length one, means that (0) Ă M is the one and only among the simple
modules and does
composition series.
therefore not have a
Proof: The proof goes by induction on the length of the shortest composition series in a composition series!
module; this is well defined and finite for all modules concerning us. A module having
a composition series of length one is simple, and for 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 Ă M1 Ă . . . Ă Mn = M.
The image of M in the quotient M/M1 is a composition series of length 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 composition series N = tNj u in M. Its length r is at least n, and by
induction its image in M/M1 is a composition series 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 β( Nµ ) = β( Nµ+1 ), and we may pick ν to be the least such index. Then
β(N ) displays as

0 = β( N0 ) Ă β( N1 ) Ă . . . Ă β( Nν ) = β( Nν+1 ) Ď . . . Ď β( Nr ) = M. (9.2)

We contend that ν is the only index 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 β(N ) will then be one less than that of N , and on the other hand, it equals n ´ 1 by
induction.

10th January 2021 at 11:32am

Version 4.0 run 30
248 chain conditions

From the equality β( Nν ) = β( Nν+1 ) ensues the equality Nν + M1 X Nv+1 = Nv+1 .

Now, M1 is simple, so either M1 X Nv+1 vanishes or equals M1 . However, it cannot
vanish because Nν ‰ Nv+1 , and we infer that M1 Ď Nν+1 . It follows that there are strict
inclusions β( Nj ) Ă β( Nj+1 ) for j ě ν + 1, and hence all inclusions in β(N ) are strict
except the one at stage ν.
The last statement in the theorem, that every chain M can be completed to a
composition series, is easily proved by induction on the length. We may assume that M
has more than two terms. Let Mν any non-zero term of M which is different from M,
and consider the quotient M/Mν . Both Mν and M/Mν are of length less than M, and
the induction hypothesis applies to both. The chain M1 = tMi uiďν in Mν can therefore
be completed to a composition series, and similarly, if β : M Ñ M/Mν is the quotient
map, the chain tβ( Mi )uiąν can be completed to one in M/Mν . Pulling the latter back
to M and splicing it with the former yields a composition series in M extending M. o
(9.49) A closer look at the proof above reveals that it in fact gives the full Jordan-Hölder
theorem:

Theorem 9.50 (True Jordan-Hölder) Any two composition series of a module of finite
length have up to order the same subquotients.

Proof: We resume the proof of the previous theorem, keep the notation and carry on
with induction on the length: By induction the two series β(M) and β(N ) have the
same subquotients up to order. Now, the subquotients of M and β(M) differ only
at the bottom stage M1 , so M has the subquotient M1 in addition to those shared
with β(M). On the other hand, the subquotients of N and β(N ) coincide except at
a certain stage ν, but in the proof above we showed that Nv + M1 = Nv+1 , and since
( Nv + M1 )/Nν » M1 , the additional subquotient of N is isomorphic to M1 as well. o
(9.51) Just like the dimension of vector spaces the length is additive along short exact
sequence, which is an indispensable property that makes 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 9.52 (Additivity of length) Assume given a short exact sequence of A-

modules.
α β
0 M1 M M2 0.

Then M is of finite length if and only if the two others are, and it holds true that ` A ( M) =
` A ( M 1 ) + ` A ( M 2 ).

Proof: Assume first that M is of finite length. Pushing a finite composition series
forward along β gives one in M2 and pulling it back along α gives one in M1 , so M1
and M2 are both of finite length.

10th January 2021 at 11:32am

Version 4.0 run 30
 modules of finite length 249

Assume next that the two modules M1 and M2 are of finite length. It suffices to
exhibit one composition series of M with the additive property. To this end, we begin
with a composition series 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 (0), 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
(9.53) An immediate corollary of Proposition 9.52 is that modules of finite length are
both Noetherian and Artinian. Obviously this is true for simple modules (no non-trivial
submodules, no non-trivial chains) and hence follows in general by a straightforward
induction on the length using Proposition 9.7 on page 231. The converse holds as well:

Proposition 9.54 An A-module M is of finite length if and only if it is both Noetherian and
Artinian.

Proof: Assume that M is 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 9.52
above. o
(9.55) 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 ( M)
and ` B ( M ) should agree. Already when k Ď K is a finite non-trivial extension of fields
the two lengths differ in that dimK V = [K; k] dimk V for a vector space V over K. You
will find a simple but slightly more subtle example in Example 9.10 below. And of
course there are stupid examples like Q Ď R with R being of length one over itself, but
as a module over Q its length is infinite (the dimension is 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.
(9.56) Unlike what is true for vector spaces, module of the same length need not be
isomorphic. Simple examples are the Z-modules F p = Z/pZ for different primes p.
They are all of length one but no two are isomorphic.

Examples
(9.7) Vector spaces: Over fields modules are just vector spaces, and as we verified in
Example 9.1 on page 232, for vector spaces it holds that being Noetherian or Artinian is
equivalent to being of finite dimension; hence also being of finite length is equivalent to
being of finite dimension. And the length of a vector space coincides with its vector

10th January 2021 at 11:32am

Version 4.0 run 30
 250 chain conditions

space dimension: indeed, given a basis v1 , . . . , vr , the subspaces Vi =ă v1 , . . . , vi ą form

a composition series, and any composition series is readily seen to be of this form as
the field itself is the only simple module.
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 ( M ) = dim A/m M. And M is
of finite length over A if and only if it is finitely generated
(9.8) Finite abelian groups: The only abelian groups that are of finite length are the finite
ones. 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 finite direct sum decomposition

Z/pi i Z.
à ν
M=
˚ The map [ p] is close i
to being a ř
We contend that `Z ( M) = i νi . In other words, the length `Z ( M) equals the sum of
“multiplication-by-
p-map”: it sends a the multiplicities of the different primes in the prime factorization of the order | M | of
class [ x ] mod pν´1 to M.
the class [ px ] mod pν .
By additivity of length it suffices to show that for a each prime p the length of the
cyclic group Z/pν Z is given as `Z (Z/pν Z) = ν, and this one does by an inductive
argument based on the standard* short exact sequences

[ p]
0 Z/pν´1 Z Z/pv Z Z/pZ 0.

Since `Z (Z/pZ) = 1, the length ` A (Z/pν Z) increases by one when ν is increased by

one, and we are done.
(9.9) Let A = k [ x, y] and m = ( x, y). For each non-negative integer n let Mn = k[ x, y]/mn .
Then there are short exact sequences

0 mn´1 /mn Mn M n ´1 0,

so that ` A ( Mn ) = ` A ( Mn´1 ) + ` A (mn´1 /mn ). The module mn´1 /mn is a vector space
over the field A/m = k having the classes of the monomials xi yn´1´i for 0 ď i ď n ´ 1
as a basis, and hence ` A (mn´1 /mn ) = dimk mn´1 /mn = n. We conclude that ` A ( Mn ) =
` A ( Mn´1 ) + n and induction on n yields that
n  
ÿ n+1
` A ( Mn ) = i= .
2
i =1

(9.10) We let A = Z and B = Z[i ] = Z[ x ]/( x2 + 1) and let M = Z[i ]/(105)Z[i ]. 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 ; the primes 3 and 7 persist being primes in Z[i ],

10th January 2021 at 11:32am

Version 4.0 run 30
 modules of finite length 251

but 5 splits up in the product 5 = (2 + i )(2 ´ i ). The Chinese Remainder Theorem gives
a decomposition as B-modules

M = F3 ( i ) ‘ F5 ‘ F5 ‘ F7 ( i ) ,

where F3 (i ) and F7 (i ) are fields and the two F5 ’s are Z[i ] modules with i acting as
multiplication by 2 on one and by ´2 on the other. We conclude that `Z ( M ) = 6 but
`Z[i] ( M) = 4.
K

Exercises
(9.21) Compute the length of Z[i ]/525 both as a Z-module and as a Z[i ]-module.
(9.22) Show that the length `Z (Z/nZ) equals the total number of prime factors of n
(counted with multiplicity).
(9.23) Let n = p1 ¨ . . . ¨ pr be the prime factorization of a square free natural number
and consider a quadratic extension Z Ă A. Show that `Z ( A/(n) A) = ` A ( A/(n) A) if
and only if each prime factor pi persists being a prime in A.
(9.24) Let M be a module of finite length over A and let f P A. Show that the quotient
M/ f M and the annihilator (0 : f ) M = t x P M | f x = 0 u are of the same finite length;
that is, ` A ( M/ f M) = ` A ((0 : f ) M )
(9.25) Modules of finite length over pid’s. Let A be pid and let f P A be an element. Show
that ` A ( A/( f ) A) is the number of prime factors in f (counted with multiplicity).
(9.26) 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.
(9.27) A frequently met situation in algebraic geometry is that a ring A is a k-algebra;
that is, it contains a ground field k (for instance, algebras like k [ x1 , . . . , xr ]/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 k(m) = A/m be the
residue class field.
a) Assume that k maps isomorphically onto k(m). 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 ( M).
b) Assume merely that k(m) is finite extension of the image of k. Prove that
dimk M = [k (m) : k] ¨ ` A ( M ).
ś
(9.28) Modules of finite length over finite product of fields. Let A = 1ďiďr k i be a finite
À
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 each Vi is of finite dimension over k i , and in that

10th January 2021 at 11:32am

Version 4.0 run 30
 252 chain conditions

case it holds true that ÿ

à
`A( Vi ) = dimki Vi .
i i

Finite length and support

(9.57) We finish of the story about modules of finite length with a criterion for a module
to be of finite length in terms of the support, and a structure theorem, which essentially
says that modules of finite length are direct sums of "local contributions"—but of course,
it says nothing about how the local contributions are shaped.

Proposition 9.58 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.

Proof: Assume to begin with that M is of finite length and let tMi u be a composition
series. Citing Proposition 7.59 on page 200 we infer that the support of M satisfies
Ť Ť
Supp M = i Supp Mi /Mi+1 = i tmi u where Mi /Mi+1 » A/mi are the subquotients
˚ Recall that the closed of the composition series tMi u. So the support is a finite union of closed points*.
points in Spec A are
For the other implication we resort to the Structure Theorem (Theorem 9.23 on
precisely the maximal
ideals. It may well page 238) which assures that there is a descending chain tMi u of submodules in M
happen that a subset of whose subquotients are shaped like A/pi with pi being prime. Again by Proposition 7.59
Spec A is closed and Ť
finite without all points it holds that Supp M = i V (pi ). Now, if Supp M consists of finitely many closed points,
being closed; for all the prime ideals pi ’s will be maximal and consequently all the subquotients Mi /Mi´1
instance, Spec Z( p) is
will be fields. Hence M is of finite length (in fact, the chain tMi u will be a composition
finite.
series). o
One part of the proposition, that the support of a module of finite length is finite and
discrete, holds true without hypotheses on A, however the other implication depends
on A being Noetherian. An example can be A = B/m2 where B = k[ xi |i P N] is the
polynomial ring in countably many variables and m = ( xi |i P N). Then A is neither
Artinian nor Noetherian, but has only one prime ideal m/m2 .
There are many rings that are not Artinian, but whose spectrum is finite. For instance
the localization Z p of the integers at the prime ( p) is not Artinian, but Spec Z( p) has
only two points namely (0) and pZ( p) . All members of the support being maximal, is
therefore a crucial part of the hypothesis.

The structure of modules of finite length

(9.59) Here comes the promised result about the structure of modules of finite length
over a ring A, they compose as the direct sum of their localizations. Given an A-module
M. For each maximal ideal m Ď A there is a localization maps ιm : M Ñ Mm , and we
À
may combine them into a map φ : M Ñ mPSupp M Mm .

10th January 2021 at 11:32am

Version 4.0 run 30
 artinian rings 253

Proposition 9.60 (Structure of finite length modules) Assume that M is a module

of finite length over the ring A. Then the a canonical φ is an isomorphism
à
M» Mm .
mPSupp M

Proof: The proof will be an application of the local to global principle, more precisely
the one asserting that being an isomorphism is a local property of A-linear maps
(Proposition 7.49 on page 197). Our task is therefore to establish that the localizations
φm of φ are isomorphisms for all m. To cope with the double localizations that appear,
we notice the following lemma:

Lemma 9.61 Let M be a module of finite length over a ring A and let the maximal ideal m
belong to Supp M. Then Mm is of finite length and has support tmu.

Proof of the lemma: Let tMi u be a composition series in M with Mi /Mi´1 » A/mi .
Now we contend that
$
&0 when m ‰ mi ;
( A/mi )m =
% A/m when m = mi .

Indeed, m ‰ mi is equivalent to mi Ę m (both are maximal ideals), and hence to there

being elements in mi not in m. So some element killing A/mi gets inverted in the
localization, and it holds that ( A/mi )m = 0 when m ‰ mi . Furthermore, it obviously
holds that ( A/m)m = A/m (elements not in m act as inversions on the field A/m). We
infer that after possible repetitions are discarded, the chain t( Mi )m u is a composition
series in Mm with all subquotients equal to A/m. o
With this lemma up our sleeve, it follows painlessly that φm is an isomorphism for all
m. To fix the ideas let Supp M = tm1 , . . . , mr u. If m R Supp M, we have Mm = 0 and
À
( Mmi )m = 0, so φm is the zero map (which is an isomorphism in this case). If m is
one of the mi ’s, say m = m j , the lemma gives
à
à
à

Mmi m
= Mmi mj
» Mmi mj
= Mmj ,
i i i

and φm : Mm Ñ Mm is the identity map. o

9.7 Artinian rings

We now turn to the rings whose ideals satisfy the descending chain condition; that
is, rings that are Artinian modules over themselves. Even though the definitions may
appear symmetric, the class of Artinian rings is astonishingly different from the class of

10th January 2021 at 11:32am

Version 4.0 run 30
 254 chain conditions

Noetherian rings. The latter is a large class encompassing 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 even
though being “small”, they are far from being insignificant, and they can indeed carry a
great lot of subtleties.
(9.62) It turns out, as we shorty shall see, that Artinian rings are Noetherian. This is
specific for Artinian rings, but far from being true for Artinian modules. The Artinian
rings are characterized among the Noetherian ones by the property that all their prime
ideals are maximal and that the maximal ideals are finite in number. In geometric terms,
their spectra are finite sets and the Zariski topology is discrete. For a finite spectrum,
the Zariski topology being discrete is equivalent to all points being closed, but this is
no more true for infinite spectra (see Exercise 9.33 for an example).
The theorem we are about to prove is due to the Japanese mathematician Yasuo
Yasuo Akizuki Akizuki. There is an analogue version valid for non-commutative rings, proved about at
(1902–1984)
the same time as Akizuki proved his theorem, which usually is contributed to Charles
Japanese
Hopkins and Jacob Levitzki, but the commutative version is Akizuki’s:
mathematician
Theorem 9.63 (Akizuki’s Theorem) An Artinian ring is Noetherian. Hence A has only
finitely many prime ideals which all are maximal.

(9.64) Recall that a module which is both Noetherian and Artinian is of finite length.
This shows that Artinian rings are of finite length (regarded as modules over themselves)
and hence come with a natural numerical invariant, the length l A ( A); that is, the number
of (simple) subquotients in a composition series. The second statement in the Theorem
is a consequence of A being of finite length as proven in Proposition 9.54, but we have
stated it like that since the proof follows the reverse path: one first proves that Spec A
is discrete and finite and subsequently that A is Noetherian.

The proof of Akizuki’s theoremùñ

The proof of Akizuki’s theorem is organized as a sequence of three lemmas. The first is
about Artinian domains:

Lemma 9.65 An Artinian domain A is a field.

Proof: Let f P A be a non-zero member of A. The principal ideals ( f i ) form a

descending chain which must ultimately be constant; that is, ( f ν+1 ) = ( f ν ) 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
We proceed 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.

10th January 2021 at 11:32am

Version 4.0 run 30
 artinian rings 255

Lemma 9.66 An Artinian ring A has only finitely many prime ideals, and they are all maximal.
Hence, if J denotes 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 9.65 above.
As to the second statement, assume that tmi uiPN is a countable set of different
Ş
maximal ideals in A. For each natural number r consider the ideal Nr = iďr mi . They
form a descending chain, and A being Artinian it holds true that Nν = Nν+1 for some
Ş
ν. This means that iďν mi Ď mν+1 , and by Proposition 2.28 on page 40, one of the mi ’s
must lie in mν+1 , contradicting the assumption that the mi ’s are different.
The last assertion ensues from the 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 Theorem gives
ś
an isomorphism A/J » i A/mi . o
The preceding lemma implies that the radical J of A coincides with intersection of
the maximal ideals in A; that is, J is also the Jacobson radical of 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 9.67 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 citing Proposition 9.7
on page 231.
Proof: The descending chain of powers tJ ν u becomes stationary at a certain stage; that
is, there is an r such that J r+1 = J r . Putting a = Ann J r one finds

(a : J ) = t x | xJ Ď Ann J r u = Ann J r+1 = a.

If a = A, then J r = 0 and we are happy, so assume that a is a proper ideal, and let b be
a minimal ideal strictly containing a; such exist since A is Artinian. Let x P b but x R a,
then b = a + Ax. If a + Jx = b, it follows that Ax/Ax X a = b/Ax X a = J ¨ Ax/Ax X a,
and Nakayama’s lemma* yields that Ax/Ax X a = 0; that is, Ax Ď a, which is not ˚ Ax/Ax X a is

the case. Hence a + Jx is strictly contain in b, and by minimality a + Jx = a. Hence finitely generated over
A and J is the Jacobson
x P (a : J ) = a, which is a contradiction. radical of A.
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 above that J ν = 0, and for each ν
there is a short exact sequence:

0 J v +1 Jν J ν /J ν+1 0.

10th January 2021 at 11:32am

Version 4.0 run 30
 256 chain conditions

Submodules and quotients of Artinian modules are Artinian (as explained in Proposi-
tion 9.7 on page 231), 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 9.2 on page 232); and we
are through by descending induction on ν. o

The structure of Artinian rings

Since Artinian rings are of finite length over themselves, we may apply the Structure
Theorems (Proposition 9.58 and Corollary 9.60 both on page 252) to obtain the following
description:

Theorem 9.68 Let A be an Artinian ring. Then Spec A is finite and discrete, and the localisa-
tion 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
Noetherian ring A is Artinian 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.

Examples
(9.11) Local rings at minimal primes, multiplicities of components: A Noetherian ring with
merely one prime ideal is necessarily Artinian. The prime ideal must be maximal,
Supp A = Spec A has one single point, and we may hence resort to Proposition 9.58. A
particular case of this situation arises when one localizes a ring at a minimal prime: If
p is a minimal prime in the (Noetherian) ring A, the local ring Ap has the sole prime
ideal pAp , since its prime ideals correspond to those in A contained in p, and so Ap is
Artinian. The closed subset V (p) is one of the irreducible components of Spec A, and
Multiplicity of a the length ` A ( A)p is called the multiplicity of the component V (p).
component
(multiplisitet til en (9.12) Multiplicities of components: Consider the ring A = k [ x, y] with constituting relation
komponent) x2 y2 = 0. One checks that both k[ x ] and k[y] are polynomial rings and that in A it
holds true that 0 = ( x2 ) X (y2 ). So the only minimal primes of A are the principal
ideals ( x ) and (y). We contend that A( x) = K [ x ]/( x2 ) where K is the rational function
field K = k (y); it is easily seen that this ring is of length two; indeed, Exercise 7.32
on page 189 yields the equality k[ X, Y ](X ) = K [ X ](X ) , and we may conclude that

10th January 2021 at 11:32am

Version 4.0 run 30
 artinian rings 257

A( x) = k[ X, Y ]/( X 2 Y 2 ) = K [ X ]/X 2 , and the claim follows. The geometric picture

(X)
of V ( X 2 Y 2 ) is the union of the X-axis and the Y-axis in the plane Spec k[ X, Y ], but with
the multiplicities two attached, and one may think about them as being "doubled". The
multiplicities reappears as the lengths of the localized rings Ap at minimal primes.
(9.13) Intersection multiplicities: In the theory of plane curves the so-called Bézout’s
theorem is central; it predicts the number of intersection points between two plane
curves. Important ingredients of the formulation are the “local intersection multiplici-
ties”, which manifests themselves algebraically as the lengths of certain Artinian rings.
The equations of the curves are polynomials f and g in A = k[ x, y], and we assume that
f and g are without common factors.
In view of Proposition 3.30 on page 77 any prime ideal containing ( f , g) is then
maximal and by 9.17 on page 235 there are only finitely many such, say m1 , . . . , mr
(they are the minimal primes of ( f , g)); hence Supp A/( f , g) A is finite and discrete
ś
and A/( f , g) A is an Artinian ring. It decomposes as A/( f , g) A = i Ami /( f , g) Ami .
The local intersection multiplicity at mi is ( f , g)mi = `( A)mi /( f , g) Ami , and the total
number of intersection points is
ÿ
`( A/( f , g) A) = `( Ami /( f , g) Ami ).
i

In Bézout’s theorem there are also local contributions at infinity occurring, and
including those, the theorem states that the total count will be deg f ¨ deg g. For
instance, two lines can be parallel, and their only intersection point then lies at infinity.

(9.14) Consider the intersection of the line y = αx and the curve y2 = x3 ; so that we put
f = y ´ αx and g = y2 ´ x3 . Assume first α ‰ 0, we then

(y ´ αx, y2 ´ x3 ) = (y ´ αx, x2 (α2 ´ x )) = (y ´ αx, x2 ) X (y ´ α3 , x ´ α2 )

So A/( f , g) » k [ x ]/( x2 ) ˆ k. The support has two points, the origin where the multi-
plicity is two and the point (α2 , α3 ) where the multiplicity is one. However if α = 0, we
find
(y ´ αx, y2 ´ x3 = (y, x2 ´ x3 ) = (y, x3 ),
so that in this case A/( f , g) = k[ x ]/( x3 ) whose support is concentrated in the origin,
and its multiplicity there is three.
K

Exercises
(9.29) Let A = k[ X, Y ]/( X a Y b ) where a and b are natural numbers, and let x and y
stand for the images of X and Y in A. Show that the equality (0) = ( x a ) X (yb ) holds

10th January 2021 at 11:32am

Version 4.0 run 30
258 chain conditions

in A and that ( x ) and (y) are the minimal primes of A. Show that A( x) = K [ X ]/( X a )
where K = k(Y ) and that this ring is of length a.
(9.30) Let n and m be two natural numbers and let a be the ideal a = ( x m , yn ) in k [ x, y].
Show that A = k[ x, y]/a is Artinian and compute its length.
(9.31) Let n, m and r be three natural numbers and let A = k [ x, y, z]/( x n , ym , zr ). Prove
that A is Artinian and compute its length.
(9.32) Show that if A = k [ x, y]/a is of length two, then after a linear change of coordi-
nates, A = k[ x, y]/( x, y2 ). If A is of length three, show that either A = k [ x, y]/( x, y)2 or
there is a linear change of coordinates such that A = k[ x, y]/( x, y3 ).
(9.33) Consider the direct product R = i Z/2Z of countably many copies of Z/2Z.
ś

a) Prove that R is not Noetherian by showing that the sets ar = t ( xi ) | xi =

0 for i ě r u (i. e. the sets of strings with a zero tail of given length) form an
ascending chain of ideals which is not stationary.
b) Show that every element in R is idempotent.
c) Let m be a maximal ideal in R. Show that the localization Rm satisfies Rm »
Z/2Z. Conclude that all prime ideals in R are maximal. Hint: The only
idempotents in a local ring are 0 and 1.
d) Show that Spec R is a compact Hausdorff space. Conclude that the Zariski
topology is not discrete.
e) Sow that the direct sum D = i Z/2Z is an ideal in R and that Spec R is the
À

disjoint union of Supp D and Supp R/D.

M

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 10

Primary decomposition

The story about primary decomposition originates in the the Fundamental Theorem of
Arithmetic. Recall that this primordial theorem states that any natural number can be
expressed as a product of prime numbers whose factors are unambiguously determined
up to order. The early 19th century mathematicians when beginning to explore the
algebraic number fields discovered before long that the integers in such fields do not
share this property unconditionally; there are rings of algebraic integers for which the
analogue of the Fundamental Theorem does not hold; the factors are not always unique.
We already saw examples of this phenomenon in Chapter 3.
However, for a large class of rings ubiquitous 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 way (up to order as usual). ˚ In addition to be an
eminent
Dedekind domains, even though they are ubiquitous in number theory, are rather
mathematician, Lasker
special rings, and the question arose quickly what is generally true. Emanuel Lasker* was World Chess
was one of the first to give a partial answer; he established primary decomposition for Champion for 27 years.

ideals in rings finitely generated over fields. The final breakthrough came with Emmy
Noether’s famous 1921-paper [?]. Her results were profoundly more general and her
proofs enormously easier and more translucent than the previous. In the formulation of
the general decomposition theorem—valid for Noetherian rings—products are replaced
by intersections and powers of prime ideals by so-called primary ideals. Every ideal a in
a Noetherian ring has a such a primary decomposition: one may express a as a finite
intersection a = q1 X . . . X qr of primary ideals. The uniqueness of the intervening ideals
however, is only partially true—there are so-called embedded components around that
disturb the picture.
We have already seen an important instance of the theorem. With Corollary 9.18 Emanuel Lasker
on page 235 we established that radical ideals in Noetherian rings are intersections (1868–1941)

of finitely many prime ideals, and the involved prime ideals are unique when the German mathematician
 260 primary decomposition

intersection is irredundant. The case of ideals not being radical is rather more complex,
prime ideals will not be sufficient, and the primary ideals enter the story.
Primary decompositions also have a highly significant geometric aspect. In a geo-
metric language a primary decomposition of an ideal a corresponds to a decomposition
˚ A topological space is of V (a) into a union of closed, irreducible* subsets called the irreducible components of
irreducible if it is not X. For instance, the subset given by xyz = 0 in C3 has the three coordinate planes as
the union of two proper, ‘
closed subsets.
irreducible components. Not that since V (a) = V ( a), the topology does not capture
‘
Irreducible spaces
the full primary decomposition, but only the representation of a as an intersection of
(irreduktibele rom) the minimal primes.
Irreducible components
(irreduktible
The road map of this chapter is as follows: we begin with introducing the new
komponenter ) important players, the primary ideals, and establish their basic properties. Next follows
the announcement of the main existence theorem—the Lasker-Noether theorem— a
discussion around it and its proof, and the two uniqueness theorems are proven.
Homogeneous ideals are in widespread use in algebraic geometry, and an important
fact is that their primary decompositions may be chosen to stay within the realm
of homogeneous ideals, and we have included a proof of that. The same holds for
monomial ideals, and as well for ideals invariant under certain other groups, which we
treat in an appendix.

10.1 Primary ideals

As alluded to in the introduction to this chapter, one needs a notion of primary ideals
that generalizes the notion of “prime powers” for integers. The naive try would be just
to use powers of prime ideals, but this turns out to be far too simple—the issue is of a
much subtler character.
(10.1) The property of an ideal q being primary is best introduced as a property of the
quotient A/q. To motivate the definition, let p be a prime number and n any integer,
˚ When n is prime to p and consider the multiplication-by-n-map Z/pr Z Ñ Z/pr Z. It is either bijective or
it will be bijective, and nilpotent*, and the important point here is that this characterizes prime powers; indeed,
when n has p as factor,
it will be nilpotent. if m = rps with p relatively prime to r and r ą 1, it holds that Z/mZ » Z/ps Z ˆ Z/rZ
and multiplication by p will be injective on one factor and nilpotent on the other, so it is
neither injetive nor nilpotent.
˚ As is the case for Inspired by this exquisite property of prime powers, we call a proper* ideal q a
prime ideals, we insist primary ideal if the following condition is satisfied:
on primary ideals being
proper.
Primary ideals o For every element x P A the multiplication map A/q Ñ A/q that sends an element
(primæridealer)
y to x ¨ y is either injective or nilpotent.

10th January 2021 at 11:32am

Version 4.0 run 30
 primary ideals 261

The "multiplication-by-x-map" is frequently called by the more scientific name “the

‘
homothety by x”. Since the radical q consists of ring elements with a power lying in q; Homotheties
that is, those inducing nilpotent homotheties, the condition may be reformulated as: (homotetier)

‘
o If xy P q, then either y P q or x P q.

Example 10.1 If f is an irreducible element in a unique factorization domain, then the

principal ideals ( f n ) generated by a power of f are primary: if xy = z f n and f n does
not divide y, it follows that f divides x, and so x P ( f n ) = ( f ).
‘
K

Basic properties of primary ideals

Here we turn to some of the basic properties of primary ideals. Their radicals are prime
ideals (as would be expected from offshoots of prime powers), and they behave well
with respect to intersections, localizations and the formation of quotient. Even though
having a prime radical is a primordial property of primary ideals, the notions are not
equivalent—the analogy with prime powers must not be pushed too far.
(10.2) The first property we shall discuss is that primary ideals have radicals which are
‘
prime. Once this is established, it follows that the radical q is the smallest among the
prime ideals containing q. Indeed, as for any ideal, the radical of q is contained in each
prime ideal that contains q In other words, the ring A/q has just one minimal prime
ideal.
‘
Proposition 10.3 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
‘
It is customary to say that a primary ideal q is p-primary when p = q, which also is p-primary ideals
phrased as p belongs to q. The converse of Proposition 10.3 does not hold in general; the (p-primære idealer)

radical being prime is not sufficient for an ideal to be primary. Example 10.4 below is
an easy and concrete instance of this. Even powers of prime ideals need not be primary.
Example 10.5 below is the standard example of this phenomenon, and a more elaborate
example in a polynomial ring is found in Exercise 10.3 below. However, if the radical of
q is maximal, q will be primary:

Proposition 10.4 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

10th January 2021 at 11:32am

Version 4.0 run 30
262 primary decomposition

Corollary 10.5 The powers mn of a maximal ideal m are m-primary.

Proof: The radical of mn equals m. o

Lemma 10.6 Assume that B Ď A is an extension of rings and that q is a p-primary ideal in A.
Then q X B is a p X B-primary ideal in B.
Proof: Since B/q X B Ď A/q, the multiplication by x in B/q X B is either injective or
‘ ‘
nilpotent since this holds for multiplication by x in A/qA. That ( q) X B = (q X B)
is trivial. o

Examples
Many examples in this section will be monomial ideals, and later on we shall dedicate a
special subsection to them (Subsection 10.5 on page 279). The method described there,
may be applied to several of the precent examples, but at this stage of the course we are
confined to using ad hoc methods.
(10.2) The ideal (y, x )2 is a primary ideal in the polynomials ring k[ x, y, z]. To see this,
make k [ x, y, z] a graded ring by assigning the weights deg x = deg y = 1 and deg z = 0
to the variables. The ideal ( x, y)2 = ( x2 , xy, y2 ) is formed by the polynomials all whose
homogenous components are of degree at least two. Consider now a polynomial f ,
with expansion f = f 0 + f 1 + . . . in homogeneous components, and assume it does
not belong to ( x, y)2 = ( x, y). This means that f 0 ‰ 0. Let g be another polynomial
‘

whose expansion in homogeneous components is g = gs + gs+1 + . . ., and assume that

f g P ( x, y)2 . We find that f g = f 0 gs + heigher terms, and so as f 0 gs ‰ 0, we infer that
s ě 2; that is, g P ( x, y)2 .
(10.3) Let k be a field and r ď n two natural numbers. Each of the monomial ideals
( x1α1 , x2α2 , . . . , xrαr ) in the polynomial ring k[ x1 , . . . , xn ] is primary. A detailed proof will
be given in Lemma 10.43 below. The powers ( x1 , . . . , xr )n are primary as well as you
are asked to prove in Exercise 10.2 on page 264.
(10.4) The ideal a = ( x2 , xy) in the polynomial ring k[ x, y] has a radical that is prime,
but a is not primary. The radical of ( x2 , xy) equals ( x ), which is prime, but in the
quotient k[ x, y]/( x2 , xy) multiplication by y is neither injective nor nilpotent (y kills
the class of x, but no power of y lies in ( x2 , xy)). One decomposition of ( x2 , xy) as an
intersection of primary ideals is

( x2 , xy) = ( x ) X ( x2 , y).

Checking the equality is not hard. One inclusion ( Ď ) is trivial, and the other holds
since a relation z = ax = bx2 + cy implies that x divides c (the polynomial ring is ufd),
and hence z P ( x2 , xy). Notice that both ideals in the intersections are primary: ( x )
since it is prime and ( x2 , y) because the radical equals ( x, y) which is maximal. There

10th January 2021 at 11:32am

Version 4.0 run 30
 primary ideals 263

are also other decompositions of a into intersections of primary ideals; for instance, it
holds true that
( x2 , xy) = ( x ) X ( x, y)2 .
Indeed, ( x2 , xy) consists of the polynomials with x as factor that vanish at least to the
second order at the origin. This exemplifies that primary decompositions are not unique
in general.
(10.5) The quadratic cone and a power of a prime that is not primary: This is the standard
example of a prime ideal whose square (or any of its powers, for that matter) is
not primary; we already discussed it in a slightly different context (Example 7.15 on
page 186).
It goes as follows: let A = k[ x, y, z] with constituting relation z2 ´ xy. This is a
graded ring (the relation is homogeneous) and the elements x, y and z form a basis for
the degree one part (the relation is of degree two). The ideal p = (z, x ) is prime (kill
it, and you get the polynomial ring k[y]), but p2 is not primary; indeed, yx lies there
being equal to z2 , but neither does x lie in p2 (for degree reasons) nor does y lie in p
(the degree one part of p has basis x and z).
One decomposition of p2 into primary ideals is shaped like

(z, x )2 = (z2 , zx, x2 ) = (yx, zx, x2 ) = ( x, y, z)2 X ( x ).

The ideal ( x, y, z)2 has the maximal ideal ( x, y, z) as radical and is therefore primary. ˚ The elements are of

The ideal ( x ) is more interesting. Killing x, we obtain the ring A/( x ) = k[y, z] with the form a(y) + b(y)z
with a, b P k [y], and
constituting relation z2 = 0, whose elements are either non-zero divisors or nilpotent*, one easily sees that this
and so ( x ) is a primary ideal. Its radical equals (z, x ). is a non-zero divisor
unless a = 0, but then
The origin of the name the quadratic cone lies in the fact that the geometric locus
the square is zero.
C in C3 where z2 ´ xy = 0 is a cone, which means that the line connecting the ori-
gin to any point on C lies in C: indeed, such a line is parameterized as (ta, tb, tc) The quadratic cone
where ( a, b, c) P C is the point and t the running parameter, and since obviously (den kvadratiske
kjeglen)
(tc)2 ´ (ta)(tb) = t2 (c2 ´ ab) = 0, the line is contained in C. And the reason for quadratic
in the name is simply because the equation of C is of the degree two.
K
(10.7) The intersection of finitely many p-primary ideals persist being p-primary. In
the analogy with the integers this reflects the simplistic fact that the greatest common
divisor of some powers of the same prime number is a power of that prime.
Ş
Proposition 10.8 If tqi u is a finite collection of p-primary ideals, then the intersection i qi
is p-primary.
Proof: Recall that the formation of radicals commutes with taking finite intersection
‘Ş Ş ‘
(Lemma 2.62 on page 52), and therefore one has i qi = i qi = p. Assume next

10th January 2021 at 11:32am

Version 4.0 run 30
264 primary decomposition

Ş Ş
that 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 checked, p is as
Ş
well the radical of the intersection i qi . o
The hypothesis that the intersection be finite cannot be ignored. Powers mi of a maximal
ideal are all primary and have the same radical, namley m, but at least when A is a
Noetherian domain, their intersection equals (0) by Krull’s Prinicipal Ideal Theorem;
the zero ideal might be primary, but certainly not m-primary (in most cases). There are
however instances when infinite intersections of p-primary ideals are p-primary (one is
described in Exercise 10.14 below).
(10.9) The property of being primary is compatible with localizations, at least when
these are performed with respect to multiplicative sets disjoint from the radical.

Proposition 10.10 Let S a multiplicative set in the ring A and let q be a p-primary ideal.
1 ´1
Assume that S X p = H. Then S´1 q is S´1 p-primary, and it holds true that ι´
S ( S q) = q.

Proof: Localizing commutes with forming radicals (Proposition 7.23 on page 183) so
the radical of S´1 q equals S´1 p. Assume that x/s ¨ y/s1 P S´1 q, but that y/s1 R S´1 q.
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 S´1 p.
1 ´1 ´1
To verify that ι´S ( S q) = q let x P A be such that ι S ( x ) P S q. 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
(10.11) The third property we shall discuss permits us, when studying a given primary
ideal q, to replace A by A/q and q by the zero ideal, which occasionally makes arguments
cleaner and notationally simpler.

Proposition 10.12 Let A be a ring and B = A/a where a is an ideal in A. Assume that q is
an ideal in A containing 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, (q/a) = ( q)/a.

Proof: This is pretty obvious because by the Isomorphism Theorem (Theorem 2.21 on
page 37) 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 (0) is a
primary ideal in the quotient A/q.

Exercises
(10.1) With notation as in Example 10.5 on the previous page, show that pn is not
primary for any n ě 2. Hint: Show that xyn´1 P pn but yn´1 R pn .
ˇ (10.2) Let p be the ideal in the polynomial ring k [ x1 , . . . , xn ] over a field k generated by
the r first variables; that is, p = ( x1 , . . . , xr ). Show that every power pm is p-primary.

10th January 2021 at 11:32am

Version 4.0 run 30
 the lasker-noether theorem 265

Hint: Consider k[ x1 , . . . , xn ] to be a polynomial ring over A = k[ xr+1 , . . . , xn ]; it is

naturally graded with A being the part of degree zero, and pm wil be the ideal of
elements whose homogeneous components all are of degree at least equal to m.
(10.3) The example in 10.14 of a prime ideals whose powers are not primary, was an
ideal in the coordinate ring of a cone. Such a phenomenon can take place even for ideals
in a polynomial ring, and the present exercise (which we have borrowed from Melvin
Hochster) illustrates this.
To facilitate the computations we give the polynomial ring R = k[ x, y, z] a grading
by assigning degrees to the variables: deg x = 3, deg y = 4 and deg z = 5.
Consider the map φ : k[ x, y, z] Ñ k [t] defined by x ÞÑ t3 , y ÞÑ t4 and z ÞÑ t5 . It
preserves the degrees (when k[t] 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
k [ t3 , t4 , t5 ].
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 = ( g2 ´ f h) x´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

10.2 The Lasker-Noether theorem

The road towards the final result about primary decomposition has two stages. The
existence of a minimal decomposition is one, and the uniqueness, or partly uniqueness
as one rather should say, is another. This bifurcation was already apparent when we
factored elements in ufd’s; every element in a Noetherian ring is a finite product of
irreducible elements, but only factorizations into prime elements are unique (up to
order and units). In this section we shall accomplish the first stage and establish the
existence of minimal primary decompositions of ideals in Noetherian rings. The result
is named after Emanuel Lasker and Emmy Noether.

Minimal primary decompositions

(10.13) Given a collection tSi u of set. It might very well happen that the intersection
Ş
i Si does not change if one throws away one or more of the Si ’s (for instance, if S1 Ď S2 ,
Redundant
one stupidly has S1 X S2 = S1 ), and in that case one says that intersection is redundant. intersections
In the opposite case, that all the Si contribute to the intersection, or in other words, (redundant snitt)
Ş Ş
when i Si Ę i‰ j Si for all j, the intersection is called irredundant.
Irredundant
intersections
(irreduntante snitt)

10th January 2021 at 11:32am

Version 4.0 run 30
 266 primary decomposition

Ş
The superfluous sets of an intersection are precisely those S j such that i‰ j Si Ď S j ,
and one may render the intersection irredundant by just discarding them.
Primary decomposition (10.14) Now, let a be an ideal in the ring A. A primary decomposition of a is an expression
(primær of a as a finite intersection of primary ideals; that is, an equality like
dekomposisjon)

a = q1 X . . . X qr (10.1)

where the qi ’s are primary ideals. We have already seen a few examples of such
decompositions (Examples 10.4 and 10.5 above).
˚ And it can be in Without further constraints there are several trivial* ways such a decomposition can
non-trivial ways too; be ambiguous. First of all, it can be a redundant intersection. Secondly, a p-primary
we shortly return to
those. ideal can be the intersections of other p-primary ideals, sometimes even in infinitely
many different ways (there is an upcoming Example 10.6). The first type of ambiguity
is coped with by just discarding superfluous ideals, and Proposition 10.8 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 and will have the same
radical.
‘
Minimal or reduced The primary decomposition (10.1) is called minimal or reduced if all the radicals qi
primary are different and the intersection is irredundant. We have proven:
decompositions
(minimale eller
reduserte Lemma 10.15 Any primary decomposition a = q1 X . . . X qr can be rendered a minimal one;
primærdekompo-
‘
that is, an irredundant intersection with the radicals qi being distinct.
sisjoner)

Example 10.6 Consider m2 = ( x2 , xy, y2 ) in the polynomial ring k[ x, y] (where m =

( x, y)). For all scalars α and β with α ‰ 0 one has the equality

m2 = ( x2 , xy, y2 ) = ( x2 , y) X (y2 , αx + βy).

Indeed, one easily checks that m2 Ď ( x2 , y) X (y2 , αx + βy) (since α ‰ 0), and the other
inclusion 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 being distinct so that their intersection is
reduced to the origin. K
Example 10.7 If the ring A is a pid, there is nothing much new. The prime ideals are
the principal ideals ( p) generated by an irreducible p. The ( p)-primary ideals are those
generated by powers of p; that is, those on the form ( pv ). In general, if f = p11 . . . prνr is
ν

a factorisation of f into a product of irreducible elements, the primary decomposition

of ( f ) is unambiguous and it is given as

( f ) = ( p1 )ν1 X . . . X ( pr )νr .

10th January 2021 at 11:32am

Version 4.0 run 30
 the lasker-noether theorem 267

The same applies to principal ideals in any ufd, where irreducible elements are prime.
K
(10.16) Finally in this paragraph, we notice that as a direct consequence of Proposi-
tion 10.10 on page 264 primary decompositions localize well:

Proposition 10.17 Assume that S is a multiplicatively closed subset of the ring A and that
a = q1 X . . . X qr is a primary decomposition of an ideal a and denote the radical of qi by pi .
Then it holds true that S´1 a = S´1 q1 X . . . X S´1 qr . Moreover, either S´1 qi is primary with
radical S´1 pi or S´1 qi = S´1 A.

The resulting decomposition of S´1 a is not always irredundant even if the one one
starts with is. The primes pi meeting S blow up to the entire ring S´1 A and will not
contribute to the intersection; they can thus be discarded, and one may write
č
S ´1 a = S ´1 q i .
S Xp i = H

A particularly interesting case arises when one takes S to be the complement of one of
the pi ’s, say pν . Then* a Apν = qν Apν , and a Apν is a primary ideal in Apν . ˚ Recall the notation
Ap for A localised at
Existence of Primary Decompositions the prime ideal p, that
is in the multiplicative
In rings that are not Noetherian, ideals may or may not have a finite primary decompo- set Azp.
sition, but in Noetherian rings they always have one. The proof is an application of a
technique called Noetherian induction (the principle of assailing a “maximal crook”).

Proposition 10.18 In a Noetherian ring A each 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 (the “maximal crook”).
Replacing A by A/a we may assume that the zero ideal is the only crook, and aim for a
contradiction. So we assume (0) is not the intersection of finitely many primary ideals
in A (in particular it is not primary), but that all non-zero ideals are.
Because (0) is not primary, there will be two elements x and y in A with xy = 0, but
with x ‰ 0 and y not nilpotent. The different annihilators Ann yi form an ascending
chain of ideals, and hence Ann yν = Ann yν+1 for some ν. We contend that (0) =
Ann y X (yν ). Indeed, if a = byν is an element in (yν ) that lies in Ann y, one has
ay = byν+1 = 0, 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 (yν )
and Ann y are non-zero, and both are therefore finite intersections of primary ideals.
The same then obviously holds true for (0), and thus the zero ideal (0) is not crooked,
contradicting the assumption it were. o

10th January 2021 at 11:32am

Version 4.0 run 30
268 primary decomposition

Theorem 10.19 (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 10.15 on page 266 it can be made
minimal by regrouping ideals with the same radical and discarding redundant ones. o

10.3 The Uniqueness Theorems

There are two main uniqueness issues concerning primary decompositions. One may ask
if the radicals of the components are unique, or one may ask if the primary components
themselves are unique. The first question is answered by an unconditionally yes (even
if the ring is not Noetherian), but to the second the answer is no in general (also in
Noetherian rings), although partially being a yes. We shall exclusively work with
Noetherian rings. The proofs basically go through also in the non-Noetherian case, but
they are not as clean cut as for Noetherian rings, and we refrain from that exercise (see
chapter four in Atiyah and Macdonald’s book [?] if you are interested). But of course,
the statements apply only to ideals that have a finite primary decomposition.

The First Uniqueness Theorem

(10.20) The way to show that the radicals of the primary components are invariants of
an ideal, is to characterize them without referring to the decomposition. The idea is to
consider the collection of transporter ideals (a : x ) when x varies in A, and it turns out
that in the Noetherian case the radicals of the primary components of a are precisely
the prime ideals among these.

Proposition 10.21 Let a be an ideal in a Noetherian ring A. The radicals that occur in an
minimal primary decomposition of a, are precisely those transporter ideals (a : x ) with x in A
that are prime.

Passing to the quotient A/a and observing that the quotient (a : x )/a equals the
annihilator (0 : [ x ]) of the class [ x ] in A/a, one may give the theorem the equivalent
formulation (remember Proposition 10.12 on page 264), which is the one we shall prove:

Proposition 10.22 (Principle of annihilators) The radicals arising from a minmal pri-
mary decomposition of the zero ideal in a Noetherian ring are precisely those ideals among the
annihilators Ann x of elements x from the ring that are prime.

Proof: Fix a minimal primary decomposition of the zero ideal (0). There are two
implications to prove. We begin with letting q be one of the components, and letting
‘
p = q denote the radical we aim at exhibiting an element x such that p = Ann x.

10th January 2021 at 11:32am

Version 4.0 run 30
 the uniqueness theorems 269

Denote by c the intersection of the components in the decomposition other than q. Then
c X q = 0, and c ‰ 0 since the decomposition is irredundant.
Let x P c be a non-zero element such Ann x is maximal among the annihilators
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.
In order to show the other inclusion pick a 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 set of
Ş
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 = ( 0 ) .
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 at least one ν P I. On the other hand, it holds true that (0) =
x ¨ Ann x Ď qv from which ensues that Ann x Ď pv because qv is pν -primary and x R pν .
o
As a corollary we arrive at the first uniqueness theorem:

Theorem 10.23 (The First Uniqueness Theorem) The radicals occurring in an minimal
primary decomposition of an ideal in a Noetherian ring are unambiguously determined by the
ideal.

Isolated and embedded components

(10.24) 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 constitute is denoted
by Ass A/a. In particular, Ass A will be the set of prime ideals associated to zero. The
term is certainly ambiguous, having a different meaning when used for modules, but it
is practical, and the two interpretations are closely related: the associate primes of a are
the associate primes of the A-module A/a (and the notation Ass A/a is consistent). Associated prime ideals
(10.25) There are no inclusion relations between the components of an irredundant (assosierte primidealer)

primary decomposition (irredundance means precisely this), but that does not exclude
inclusion relations between the associated primes. In Example 10.4, for instance, we
found that ( x2 , xy) = ( x )X( x, y)2 with the associated primes being ( x ) and ( x, y). This
leads us to distinguish between isolated and embedded associated primes. The former are Embedded associated
prime (embeddede
assosierte primidealer)

Isolated associated
10th January 2021 at 11:32am primes (isolerte
assosierte primidealer)
Version 4.0 run 30
 270 primary decomposition

those being minimal in Ass A; that is, they do not contain any other associated prime,
whereas the latter are those that do. In the example above, ( x ) is an isolated prime
whilst ( x, y) is embedded*.
Primary components with an isolated radical are called isolated components and those
with an embedded radical are called embedded components.
(10.26) Early in the course, when discussing the radical of an ideal, we proved that the
‘
Isolated components radical 0 of A equals the intersection of all minimal primes in A (Paragraph 2.59 on
Ş
(isolerte komponenter)
‘
page 2.59); that is, 0 = p, the intersection extending over the minimal elements of
Embedded components Spec A.
(embeddede
‘
On the other hand, we just expressed the radical 0 as the intersection of the prime
komponenter) ‘ Ş
ideals minimal in Ass A so that 0 = p where the intersection extends over the
minimal elements in Ass A. When the intersections of two finite families of prime ideals
are equal and there are no inclusion relations between members of either family, the
families coincide (Lemma 2.36 on page 43). Hence the sets Spec A and Ass A have the
same minimal primes. We have proved:

Proposition 10.27 In a Noetherian ring A the sets Spec A and Ass A have the same minimal
˚ You might be puzzled
elements; in other words, the minimal primes of A are precisely the isolated associated primes.
by the notion
embedded In particular, there are finitely many minimal primes.
components since they
are not contained in, (10.28) We have seen the intersection of the the associated primes of A is the set of
but on the contrary nilpotent elements in A, and their union turns out to play a particular role at least in
contain other
associated primes. The Noetherian rings, it equals the set of zero divisors:
usage comes from Ť
geometry since Proposition 10.29 The set of zero-divisors in a Noetherian ring A equals pPAss A p, the
inclusions between union of the associated primes.
varieties are the reverse
of those between ideals. 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, it is contained in a
maximal one, we are through. o
Example 10.8 We offer one more example and consider the ideal

a = ( x2 y, y2 z, z2 x ).

in the polynomial ring C[ x, y, z] and aim at determining a primary decomposition.

To get an idea of where to start we resort to geometry, and take a look at the
zero-locus V (a) 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

10th January 2021 at 11:32am

Version 4.0 run 30
 the uniqueness theorems 271

consider the x-axis. Localizing at x, i. e. passing to A x = C[ x, x´1 , y, z] (thus eliminating

the two other axes) we see that aA x = (y, y2 z, z2 ) = (y, z2 ); so we suspect this to be one
of the components; by symmetry we find two more suspects, and in fact, it holds true
that
( x2 y, y2 z, z2 x ) Ď (y, z2 ) X ( x, y2 ) X (z, x2 ).
This is however not the whole story. The element xyz lies in the intersection to the right,
but not in a. Now, clearly ( x, y, z) Ď (a : xyz) and ( x, y, z) being maximal, it holds that
(a : xyz) = ( x, y, z), so there must be an ( x, y, z)-primary component. After a few tries
(and a some failures) one finds the equality

( x2 y, y2 z, z2 x ) = (y, z2 ) X ( x, y2 ) X (z, x2 ) X ( x2 , y2 , z2 ).

The associated primes are (z, y), ( x, y), (z, x ) and ( x, y, z).
Once one has a good guess, it is relatively easy to check if it is correct. All involved
ideals are generated by monomials, and monomial ideals have the nice property that a
polynomial is a member if and only if all the monomial terms of the polynomial are.
Hence it suffices to check that each monomial in the ideal to the right also lies in the
one to the left. But monomials in ( x2 , y2 , z2 ) have either x2 , y2 or z2 as a factor, and by
symmetry we may assume it is y2 . Lying in (z, x2 ) too, our monomial must have either
z or x2 as a factor and thereby also zy2 or x2 y2 ; but both these lie in a, and we are done.
K

The second uniqueness theorem

We now come to the uniqueness issue for the primary components. Already our first
example (Exampe 10.4 on page 262) showed that they are not unique. We found that

( x2 , xy) = ( x ) X ( x2 , y) = ( x ) X ( x, y)2 ,

and both ( x2 , y) and ( x, y)2 are minimal primary components. Notice that they have the
same radical ( x, y) (they must!), and so they are embedded components. The bad news
is that ( x2 , xy) have infinitely many minimal primary decomposition with different
embedded components (Example 10.9 below), but there is good news too: the other
component is unique. This is generally true: the Second Uniqueness Theorem states that
isolated components are unique. The reason for this is that isolated components may
be retrieved from the ideal by localizing at the corresponding associated prime ideals,
and according to the First Uniqueness Theorem these primes are independent of the
decomposition; if q is the component and q = p, it holds true that q = ι´1 (a Ap ) where
‘

ι : A Ñ Ap is the localization map (or simply that q = A X a Ap when ι is injective).

Theorem 10.30 (The Second Uniqueness Theorem) The isolated primary components of
an ideal a in a Noetherian ring A are unambiguously defined by the ideal.

10th January 2021 at 11:32am

Version 4.0 run 30
272 primary decomposition

Proof: We shall concentrate on one of the isolated associated prime ideals p of a, but
the main player will be a p-primary component q from one of the minimal primary
decompositions of a. The salient point is, as already announced, the equality

q = ι ´1 ( a A p ) , (10.2)

from which the theorem ensues as isolated prime ideals are invariants of a.
Ş
To establish (10.2) one writes 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 (10.3)
i

since the qi ’s blow up when localized; that is, qi Ap = Ap . Indeed, since p is isolated, pi Ę
p holds for all i. Taking inverse images of both sides of (10.3) and citing Proposition 10.10
on page 264 we conclude that ι´1 (a Ap ) = q. o

Examples
(10.9) For any natural number n the equality

( x2 , xy) = ( x ) X ( x2 , xy, yn ) (10.4)

holds true in the polynomial ring k[ x, y], and this is an example of infinitely many
different minimal primary decompositions of the same ideal. That the left side of (10.4)
is included in the right is obvious; to check the other inclusion, let a belong to the right
side. Then
a = p ¨ x = q ¨ x2 + r ¨ yn + sxy,

with p, q r and s polynomials in k[ x, y]. It follows that x divides r, and hence that
a P ( x2 , xy).

(10.10) Intersection of two conics: So far all our examples have merely involved monomial
ideals, but of course most ideals are not monomial. 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 also 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—in this case, they conceal the
tangency of two intersecting curves.

10th January 2021 at 11:32am

Version 4.0 run 30
 the uniqueness theorems 273

We shall analyse the familiar case of the intersection of two quadratic curves; the
unit circle centred at (0, 1) and the standard parabola given by y = x2 . So let a =
( x2 + (y ´ 1)2 ´ 1, y ´ x2 ) in k[ x, y], 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 (1, 1), (´1, 1) and (0, 0), and the same manipulations give

a = ( x2 + (y ´ 1)2 ´ 1, y ´ x2 ) = ( x2 ( x2 ´ 1), y ´ x2 ).

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 ´ ( x + 1)( x ´ 1) ´ 1, one has y ´ 1 P p in the two
other cases. We conclude that the ( x, y), ( x ´ 1, y ´ 1) and ( x + 1, y ´ 1) 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 10.30 on
page 271). In the local ring A = k [ x, y]( x+1,y´1) , where both x and x ´ 1 are invertible,
we obtain the equality

a A = ( x2 ( x2 ´ 1), y ´ x2 ) = ( x + 1, y ´ x2 ) = ( x + 1, y ´ 1).

In similar fashion, in B = C[ x, y]( x´1,y´1) both x and x + 1 are invertible, and one has

aB = ( x2 ( x2 ´ 1), y ´ x2 ) = ( x ´ 1, y ´ 1).

Finally, in C = k [ x, y]( x,y) both x + 1 and x ´ 1 have inverses, and we see that

aC = ( x2 , y).

Since there are no embedded components, we conclude that

a = ( x ´ 1, y ´ 1) X ( x + 1, y ´ 1) X ( x2 , y).

When the characteristic of k equals two, things evolve in a slightly different manner.
In that case, the two ideals ( x ´ 1, y ´ 1) and ( x + 1, u ´ 1) conicide and x2 + 1 = ( x + 1)2 .
We find
a = (( x + 1)2 , y ´ 1) X ( x2 , y).

10th January 2021 at 11:32am

Version 4.0 run 30
274 primary decomposition

(10.11) A saddle surface and the union of two planes: Consider the intersection of the
“saddle surface” S given in C3 by z = xy and the union of the xy-plane and the xz-plane,
which has the equation yz = 0.
The plane z = 0 intersects S in the union of the x-axis and the y-axis, and the
plane y = 0 in the x-axis. The x-axis thus appears twice in the intersection, which
algebraically is manifested by the occurrence of a non-prime primary component in the
decomposition of the ideal a = (z ´ xy, zy).
Because zy = (z ´ xy)y + xy2 , it holds that a = ( xy2 , z ´ xy), and consequently one
finds
a = ( xy2 , z ´ xy) = ( x, z ´ xy) X (y2 , z ´ xy);

by observing that any prime containing a contains either ( x, z) or (y, z) so that (10.11)
holds true when localized at any prime ideal, and hence equality persists since equality
is a local property. Now, ( x, z ´ xy) = ( x, z) is a prime ideal, and (y2 , z ´ xy) is (y, z)-
primary (for instance, since k [ x, y, z]/(y2 , z ´ xy) » k[ x, y]/(y2 )), so we have found a
primary decomposition of a.
K

Exercises
ˇ (10.4) Show that for any scalar a P k it holds true that

( x2 , xy) = ( x ) X ( x2 , y + ax )

and that this is a minimal primary decomposition. Show that different scalars a give
different decompositions.
ˇ (10.5) Determine a minimal primary decomposition of a = ( x3 , y2 x2 , y3 x ) in the poly-
nomial ring A = k[ x, y].
ˇ (10.6) Let a be the ideal in the polynomial ring A = k[ x, y, z] given as a = (yz, xz, xy).
Show that the minimal primary decomposition of a is shaped like

(yz, xz, xy) = (y, z) X ( x, z) X ( x, y).

Show that the maximal ideal ( x, y, z) is associated to the square a2 and determine a
minimal primary decomposition of a2 . Hint: Consider (a : xyz).
(10.7) Determine the primary decomposition of the ideal (13) in the ring Z[i ] of
Gaussian integers.
(10.8) With the notation of Section 8.36 on page 220 about elliptic curves, consider
the ring A = k [ x, y] with constituting relation y2 = p( x ) where p( x ) is a monic cubic

10th January 2021 at 11:32am

Version 4.0 run 30
 the uniqueness theorems 275

polynomial with distinct roots. Determine the primary decomposition of the principal
ideals ( x ´ a) and (y ´ b) where a, b P k.
(10.9) Let k be a field. Let ps = ( x1 , . . . , xs ) in k[ x1 , . . . , xn ] be the ideal generated by
the s first variables. Consider

a = p1 p2 p3 . . . pr = ( x1 )( x1 , x2 ( x1 , x3 , x3 ) . . . ( x1 , . . . , xr ).

Prove that the following equality holds true

a = p1 X p22 X . . . X prr´ 1 r
´1 X p r ,

and that this 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.
(10.10) Concider the map k[ x, y, z] Ñ k [t, z] such that x ÞÑ t2 ´ 1 and y ÞÑ t(t2 ´ 1) and
z ÞÑ z. Let p = (z2 ´ ( x + 1), y ´ zx ). Determine the minimal primary decomposition of
pk[t, z].
(10.11) Symbolic powers. We have seen that the powers pn of a prime ideal p in
A not necessarily are p-primary, unless p is maximal. But there are primary ideals
canonical associated to the powers pn ; the so-called symbolic powers p(n) . These arise Symbolic powers
in the following way: The ideal p Ap is maximal in the local ring Ap and its powers (symbolske potenser)

are therefore primary by Proposition 10.4 on page 261. Pulling primary ideals back
along the localization map ι results in primary ideals (Proposition 10.10 on page 264).
The ideal p(n) = ι´1 (pn Ap ) (or when ι is injective, p(n) = A X pn Ap ) will therefore be
primary, and this pullback is the n-th symbolic power of p.
a) Show that if n and m are natural numbers, it holds that p(n) ¨ p(m) Ď pn+m ;
b) Show that pn = p(n) if and only if pn has no embedded components;
c) With the notation as in Example 10.5 on page 263 determine the symbolic square
p(2) of the ideal p = ( x, y).
M
Example: Reduced rings and a criterion of Serre
As an example of how a primary decomposition can be used, we give a criterion due to
Serre for a Noetherian ring to be reduced which goes under name of the R0 –S1 –criterion.
‘
The nil-radical (0) of a Noetherian ring A consists of the nilpotent elements in A,
so one expects to be able to read off from the primary decomposition of the zero ideal
whether A is reduced or not. The radical is expressed as the intersection
‘
(0) = p1 X . . . X pr

of the finitely many minimal primes p1 , . . . , pr of A. Now, A is reduced precisely when

‘
(0) = (0), and we may infer that A reduced if and only if the intersection of the

10th January 2021 at 11:32am

Version 4.0 run 30
276 primary decomposition

minimal primes equals (0):

(0) = p1 X . . . X pr ; (10.5)

The equation (10.5) is a minimal primary decomposition of (0), so we conclude that A

is reduced precisely when the primary components and the minimal primes of A are
the same. In particular, A has no embedded components.

Proposition 10.31 A ring A is reduced if and only if it abides to the two requests that follows.
i) For each minimal prime Ap is a field;
ii) For each non-minimal prime ideal p, the local ring Ap has no zero divisor.

The first requirement is usually is called R0 –condition and the second the S1 -condition
(notions which will be perfectly logical when you have heard about regular rings, depth
and height), hence the name the R0 –S1 –criterion.
Proof: The first resquest is equivalent to all isolated primary components being prime
‘
since if q is one of the sort and p = q, the quotient p/q equals the nil-radical of Aq ,
i. e. it is equivalent to Ap being a field (see also proposition 7.29 on page 187 about the
total ring of fractions of reduced rings). The second statement is equivalent to A being
without embedded components; indeed, it may be reformulated as each annihilator
(0 : x ) being contained in a minimal prime ideal. o
Exercise 10.12 Let A be a Noetherian ring without embedded components and let
q1 , . . . , qr be the isolated components of (0). Show that the total ring of fractions K ( A)
of A is the product
K ( A) = Ap1 ˆ . . . ˆ Apr ,
‘
where pi = qi and where each Aqi is an Artinian local ring. M

10.4 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
most ideals one meets in examples or exercises are homogeneous, but because of the
ubiquity of the so-called projective varieties in algebraic geometry—these are subvarieties
of projective space defined by homogeneous ideals.
In what follows, we shall prove that associated primes and isolated components 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.
Fortunately this is the case, and the Lasker-Noether Theorem is fully valid in a graded
context.
(10.32) We start with a little lemma (in fact, we already met a version of it in Exercise 1.14
as early as on page 24).

10th January 2021 at 11:32am

Version 4.0 run 30
 the homogeneous case 277

Lemma 10.33 Let R be a graded ring and assume that x and y are two elements such that
xy = 0. Let xe be the homogeneous term of x of lowest degree. Then xer y = 0 for some natural
number r. In particular, xer yi = 0 for each homogenous component yi of y.

Proof: Let x = xe + . . . + xn and y = y f + . . . + ym respectively be the expansions of

x and y in homogeneous components. We shall show by induction that xeν+1 y f +ν = 0
for all ν, and any r with r ě m ´ f + 1 will then do. 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 for the homogeneous component of degree e + f + ν that

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 gives xeν+1 ¨ y f +ν = 0 because if j ě 1, it holds true by

induction that
j ´1 ν´ j +1
xeν ¨ y f +ν´ j = xe ¨ xe ¨ y f +ν´ j = 0.

o
(10.34) The first step in establishing the graded Noether–Lasker Theorem is to see that
primes associated to homogeneous ideals are homogeneous.

Proposition 10.35 Let R be a graded ring and let p be a prime ideal associated to the homoge-
neous ideal a. Then p is homogeneous.

Proof: Replacing R by R/a, we may assume that a = 0. Let p be a prime associated

to R. Any associated prime is the annihilator of some element, and there is an a P R
so that p = (0 : a). Assuming that x P (0 : a) we have to show that each homogeneous
components of x also lies in (0 : a). If xe is the homogeneous component of x of lowest
degree, the lemma tells us that for some ν the power xeν kills a . Hence xeν P (0 : a) and
by consequence xe P (0 : a) as (0 : a) is a prime ideal. Noticing that x ´ xe also lies in
(0 : a), we finish the proof by the number of non-vanishing homogeneous terms in x. o
(10.36) The next step is to treat the primary components, and as alluded to above,
embedded components are not unique and therefore not forced to be homogeneous,
but they may be chosen homogeneous. For any ideal a we shall denote by a7 the largest
homogeneous ideal contained in a; that is, the ideal generated by all homogeneous
elements belonging to a.

Lemma 10.37 Let q be a primary ideal in the graded ring R whose radical is homogeneous.
Then q7 is primary and q7 = q.
‘ ‘

Proof: We start out by proving that q7 has q as radical. Obviously q7 Ď q. At-

‘ ‘ ‘
‘ ‘
tacking the other inclusion, we pick a member x P q. Since q is homogeneous
‘
all the homogeneous components of x lie in q, and we may safely assume x to be

10th January 2021 at 11:32am

Version 4.0 run 30
278 primary decomposition

homogeneous. Now, x ν P q for some ν, and as any homogeneous 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 and that y R q7 . The task is to show that x P q7 , or in
‘
‘
other words that x P q, since the two radicals coincide. If y f is the homogeneous term
of y of lowest degree, we may assume that y f R q7 , since if y f lie in q7 , we may replace y
by y ´ y f , and after repeating this procedure if needed, we shall end up with an element
whose term of lowest degree do not belong to q7 . With y f well placed outside of q7 , it
follows that y f is not a member of q (all homogeneous elements of q belongs to q7 by
definition). Now, xy f lies in q7 since all the homogeneous components of xy do, and
‘
therefore it lies in q. Hence x P q. o
(10.38) Finally we are prepared for the graded version of the Lasker–Noether theorem;
needless to say, the two uniqueness theorem persist, they can of course be applied to
any minimal primary decompositions.

Proposition 10.39 (Graded Lasker–Noether Theorem) A homogeneous ideal in a Noe-

therian graded ring has a minimal primary decomposition with all components being homoge-
neous, and all its associated primes are homogeneous.

Proof: Observe first that all the prime ideals associated to a homogeneous ideals are
homogeneous (Proposition 10.35 above).
It is fairly clear that (a X b)7 = a7 X b7 (the homogeneous elements in a X b are the
homogenous elements the lie in both a and b) so starting out with a minimal primary
Ş
decomposition a = i qi of a homogeneous ideal a and applying the 7-construction to
it, one arrives at a decomposition
č 7
a = a7 = qi , (10.6)
i

and according to Lemma 10.37, this is a primary decomposition. Moreover, the radicals
of the ideals q7i being the same as the radicals of the qi ’s, we can conclude that (10.6) is
a minimal primary decomposition. o

A structure theorem for graded modules

This is a natural place to include a graded version of the structure theorem for fintiely
generated modules ( Theorem 9.23) we proved in Chapter 9, and the proof goes through
with minor modifications.
(10.40) The changes are of two types. We want all intervening submodules to ge
graded subsmodules and all prime ideals to be homogenoeus, and this ensured by
Proposition 10.35 above. Secondly we want the chain and the ensuing maps in the
ensuing short exact sequences to be maps in the category of graded modules, that is,
we want the to preserve homogeneous elements and degrees. For this the shift operator

10th January 2021 at 11:32am

Version 4.0 run 30
 the case of monomial ideals 279

is needed. Recall that M(d) has the same homogeneous elements as the graded module
À
M = n Mn , but the degrees are shifted: ( M ( d ))n = Mn+d . In short, we insist the
claim being stated in the context of the category GrModR .

Theorem 10.41 (Structure of graded modules) Let R a Noetherian graded ring and let
M be a non-zero graded A module. Then M is finitely generated if and only if it possesses a
finite ascending chain of graded submodules tMi u0ďiďr with M0 = 0 and Mr = M, and such
that all subquotients are shaped like R/pi (mi ) with pi a homogeneous prime ideal.
In other words, there are short exact sequences in GrModR

0 / Mi´1 / Mi / A/pi (mi ) /0

for 1 ď i ď r.
Proof: Assume M is finitely generated (the other implication is straightforward and
left to the zealous students). Just as in the non-graded case, the proof is by Noetherian
induction. The set of graded submodules for which the theorem is true is non-empty
(zero submodule is there) and has a maximal element N since M is Noetherian. Assume
that N is proper. The quotient M/N is then a non-zero graded module and has an
associated homogeneous prime ideal p = (0 : x ) where x is homogenous. Shifting
degrees by m = ´ deg x, we obtain a map R/p(m) Ñ M/N of graded modules, which
induces the short exact sequence

0 N N1 R/p(m) 0.

Here N 1 is the inverse image in M of Rx » R/p, and the sequence shows that N 1 is
strictly larger than N and has a chain of the required kind. o

10.5 The case of monomial ideals

Just as in the case of homogenous ideals, primary decompositions of monomial ideals
stay within the family, so to say: associated primes and isolated components will
automatically be monomial, and the embedded components can be chosen to be. We
shall give a simple and constructive proof of this useful fact which also furnishes an
easy algorithm that computes the decompositions. It is a sure but laborious way to
a decomposition, but since there are relatively few and easily understood monomial
prime ideals, there will often be smarter ad hoc approaches.
(10.42) The crux of the proof is Lemma 10.43 below, but before stating it we need a
new notion: a monomial f P a is called primitive (relative to a) if no monomial that Primitive monomials
strictly divides f , belongs to a; in other words, if one lowers the power of a variable (primitive monomer)

occurring in f , the resulting monomial does not belong to a. Every monomial ideals
has a generating set consisting of primitive monomials.

10th January 2021 at 11:32am

Version 4.0 run 30
280 primary decomposition

Note that a monomial belongs to a monomial ideal if and only if it is a multiple of

one of the primitive generators: indeed, monomials with different multi-exponents are
linearly independent over k, and a monomial do not get lower multi-exponents when
multiplied by a polynomial.

Lemma 10.43 (The jco–algorithm) Let a be a monomial ideal in the polynomial ring A =
k[ x1 , . . . , xn ] over the field k.
i) If a is generated by pure powers; that is, if for some subset I Ď t1, . . . , nu it holds true
α
that a = ( xi i |i P I ), then a is primary;
ii) Let f be a primitive generator for a and assume that f = f 1 f 2 where f 1 and f 2 are
relatively prime monomials. Then it holds true that:



a = a + ( f1 ) A X a + ( f2 ) A .

Proof: Proof of i): assuming that xy P a and x R a we are to prove that a power of y
belongs to a. From x R a, we infer that at least one of the monomial terms of x does not
belong to a, and because a is monomial, all the monomial terms of xy lies in a since xy
lies there. So when the claim is shown for monomials, we may conclude that a power
of every monomials term of y lies in a, and a standard argument with the multinomial
formula then yields that a high power of y lies there.
Assume then that both x and y are monomial. As x is a monomial not in a, at each
one of the variables xν appears in x with an exponent less than αν , hence as y is a
multiple of one of the generators, xν divides y for at least one index ν. Consequently
y αν belongs to a.
Proof of ii): The ideals a + ( f 1 ) A and a + ( f 2 ) A are both monomial, so it suffices
to prove that each monomial in the intersection belongs to a (the other inclusion is
obvious).
For a monomial x it holds for each i that x P a + ( f i ) A if and only if x either lies in
a or is divisible by f i . Consequently x lies in both the ideals a + ( f i ) A if and only if it
either lies in a or is divisible by both f 1 and f 2 . Since f 1 and f 2 are relatively prime, the
latter condition means that r is divisible by f 1 f 2 , hence also in that case it belongs to a.
o

Proposition 10.44 (Monomial Lasker–Noether) Let A = k[ x1 , . . . , xn ] and let a be a

monomial ideal. Then the associated primes and the isolated primary components of a are
monomial, and additionally, the embedded components may be chosen to be monomial as well.

Proof: We begin by determining a generating sets for a consisting of primitive gener-

ators. The algorithm lies open in the day: if the generators of a are pure powers, a is
primary. If not, at least one variable has a factorization f = f 1 f 2 with the f i ’s involving

10th January 2021 at 11:32am

Version 4.0 run 30
 the case of monomial ideals 281

different variables, and they are therefore relatively prime. The lemma gives

a = (a + ( f 1 )) X (a + ( f 2 ) A),

and the chosen generator set for a induces generator sets for each of the ideals (a + ( f i )):
just adjoint f i to the old set and discard those generators that cease to be primitive. We
iterate and feed each (a + ( f i )) into the algorithm unless it is primary in which case we
leave it as it is. The process must eventually terminate since e. g. the sum of the degrees
of the generators goes strictly down in each step.
And at the end, we have written a as the intersection of monomial ideals that are
primary. It will probably not be minimal, so some tiding up might be necessary. o

Examples
(10.12) An embedded component of a monomial ideal need not be monomial Exercise
10.4 on page 274 exhibits one: There you proved (or should have) that in the polynomial
ring k[ x, y] it holds that
( x2 , xy) = ( x ) X ( x2 , y + ax )

for all a P k.

(10.13) According to the lemma ( x n , xy, ym ) = ( x, ym ) X (y, x n ) in k[ x, y], and all three
ideals have ( x, y) as radical and thus are ( x, y)-primary. This illustrates that a risk with
the algorithm is that primary ideals might be split up into intersections of ideals that at
the end have to be coalesced if one goes for a minimal primary decomposition.

(10.14) Consider a = ( x3 y, xy3 , xz3 , zx3 , yz3 , zy3 ) in the polynomial ring A = k[ x, y, z].
Inverting z, we see that aAz = ( x, y) Az due to occurrence of the generators z3 y and
z3 x. Similarly, aA x = (y, z) A x and aAy = ( x, z) Ay . All these ideals are primary, in fact
prime, and the equalities exclude the prime ideals ( x ), (y) and (z) from being associate,
and we deduce that
a = ( x, y) X ( x, z) X (y, z) X q (10.7)

where q is ( x, y, z)-primary. We claim that one may take q = ( x3 , y3 , z3 ). Indeed, the

inclusion Ď is clear, and the proposition allows us merely to work with monomials
when checking the other. So let the monomial f = x α y β zγ belong to the right side of
(10.7). Since f P q, at least one of the exponents, say α, must be three or more, and as f
lies in (z, y), either β or γ must be one or more, and hence f lies in a.
To illustrate the algorithmic approach we run the algorithm for our ideal a. At each
stage the branching monomials are underlined in red, and all simplifications are tacitly

10th January 2021 at 11:32am

Version 4.0 run 30
 282 primary decomposition

performed:

a =( x3 y, xy3 , xz3 , zx3 , yz3 , zy3 )

=( x3 , xy3 , xz3 , yz3 , zy3 ) X (y, xz3 , zx3 )
=( x, yz3 , zy3 ) X ( x3 , y3 , xz3 , yz3 ) X (y, z) X (y, xz3 , x3 )
=( x, y) X ( x, z3 , zy3 ) X ( x, y3 , yz3 ) X ( x3 , y3 , z3 ) X (y, z) X (y, z3 , x3 )
=( x, y) X ( x, z) X ( x, z3 , y3 ) X ( x3 , y3 , z3 ) X (y, z) X (y, z3 , x3 )
=( x, y) X ( x, z) X (y, z) X ( x3 , y3 , z3 )

The last equality just tidies things up by making the intersection irredundant.
K
ˇ Exercise 10.13 Exhibit, by way of the algorithm in Lemma 10.43, a minimal primary
decomposition of a = ( x α y, z β x, yγ z) in the polynomial ring k[ x, y, z] where α, β and γ
are natural numbers. M

10.6 Primary decomposition of modules

The primary decomposition of ideals is easily generalised to submodules. In a Noethe-
rian module, each submodule has a primary decomposition, which enjoy properties
entirely analogous to the uniqueness properties in the ideal case. The proofs follow the
same patterns and are built around the same ideas, but obviously there are necessary
changes in the terminology. In this section we shall sketch the development, but shall
leave most of the details to the students.

Primary modules
Primary ideals play a essential role in the theory of primary decomposition of ideals,
and when extending the theory to modules, our first task is to establish a corresponding
concept in the category of modules, and it is a carbon copy: a given a submodule N of
Primary submodul the module M, is said to be primary if it fulfils the following condition:
(primære
undermoduler) o the homotheties in M/N induced by ring elements are either injective or nilpotent.

There is, as in the ideal case, an equivalent formulation in terms of the transporter ideal
( N : M) = t x P A | xM Ď N u:

o if xz P N then either z P N or x n P ( N : M ) for some natural number n.

Some of the most basic properties of primary modules are summarized in the following
proposition. They are natural generalizations to modules of well-known properties of
primary ideals. The proofs are basically the same as the proofs of the parallel properties
radical of a submodule in the ideal case, and, as we warned, they are left to the students. Note that the radical
(radikalet til en
undermodul)

10th January 2021 at 11:32am

Version 4.0 run 30
 primary decomposition of modules 283

of a submodule is defined as the radical of the relevant transporter ideal; that is, as
‘
( N : M ).

Proposition 10.45 Let A be ring and M an A-module.

i) The radical of a primary submodule is a prime ideal;
ii) Finite intersections of p-primary submodules are p-primary;
Let N Ď M and L Ď N be submodules and S Ď A a multiplicative set.
iii) If N is a p-primary submodule and S X p = H, then S´1 N is S´1 p-primary
submodule of S´1 M;
iv) If N is p-primary in M, then N/L is a p-primary submodule of M/L.

Primary decomposition
Just as for ideals, a primary decomposition of a submodule is a presentation

N = N1 X . . . X Nr ,

where each Ni is a primary submodule of M. It is minimal if it is irredundant and the minimal primary
radicals are different, and every primary decomposition may be render a minimal one decompositions
(minimale primær
by discarding redundant components and coalescing components with identical radicals. dekomposisjoner)
The components of a minimal primary decomposition are said to primary components of Primary components
N. The Lasker–Noether theorem has a counterpart in the module category: (primærkomponenter)

Theorem 10.46 Every Noetherian module has a minimal primary decomposition

Proof: We attack by Noetherian induction. Consider the set of submodules of M for

which the conclusion does not hold true. If the theorem fails for M, this set is non empty
and has a maximal element N. We may replace M by M/N, so that every non-zero
submodule has a primary decomposition but the zero-submodule (0) does not have
one.
Then (0) is obviously not primary, and there are elements x P A and z P M so that
xz = 0 but the homothety by x is not nilpotent. The kernels of the homotheties [ xi ] by
the powers xi form an ascending sequence of submodules and because M is Noetherian,
must stabilize at a certain point, say at ν. Then ker[ x v ] = ker[ x ν+1 ]. We contend that
(0) = ker[ x ] X x ν M. Indeed, if xz = 0 and z = x v w it follows that w P ker[ x ]v+1 , hence
in ker[ x ]v . By consequence z = x ν w = 0. Both ker[ x ] and x ν M are non-zero submodule,
and hence both are expressed as finite intersections of primary submodules, Evidently
the same then holds for their intersection, which contradicts that no such expression
exists for (0). o
Just as with ideals, one says that a primary components is isolated if its radical is
minimal among the radicals of the components, and it is called embedded in the opposite embedded components
case. (embedded komponent)

10th January 2021 at 11:32am

Version 4.0 run 30
 284 primary decomposition

Primary decompositions of submodules enjoy uniquiness properties in complete

parallel to ideals; we state the two uniqueness theorems, but leave it to the interested
student to work out the proofs (highly recommended) which mutatis mutandis aer the
same as for ideals:

Theorem 10.47 (The two uniqueness theorems) Let M a finitely generated module over
a Noetherian ring A and let N be a submodule.
i) The radicals of the primary components of N coincide with its associated prime ideals,
and unambiguously defined by the submodule N;
ii) The isolated primary components of N are unique.

10.7 Appendix: Primary decomposition and group actions

In many situations there is group G acting on the ring A, and the ideal a we study is
invariant under G. Many rings arising from linear algebra or representation theory
(for example coordinate rings of generic determinantal varieties) come endowed with
natural actions of Lie groups. This is also the case when a is a homogeneous ideals in the
polynomial ring R = k[ x1 , . . . , xn ] over a field k. Indeed, the multiplicative group k˚ acts
on R through homotheties; i. e. the action is given as t ¨ f ( x1 , . . . , xn ) = f (tx1 , . . . , txn ),
and f is homogenous of degree d it holds true that f (tx ) = td f ( x ), hence homogeneous
ideals are invariant. Similarly, the group (k˚ )n acts on k[ x1 , . . . , xr ] with the action of
λ = (λ1 , . . . , λn ) on xi being λ ¨ xi = λi xi . Then, in multi-index notation, it holds for a
monomial xα that λ ¨ x α = λα x α , and so monomial ideals will be invariant.
(10.48) A relevant question is if a G-invariant ideal a has a primary decomposition that
˚ The action induces an is invariant* under G. If G is finite, this is certainly not true; for instance, in Z[i ] the
action on the ideals by primary decomposition of (5) is (5) = (2 + i ) X (2 ´ i ), but complex conjugation swaps
setting ag = t ga | g P
G, a P a u the two prime ideals (2 + i ) and (2 ´ i ). However, if the group G has no non-trivial
finite quotient it holds true.

Proposition 10.49 Let G be a group without non-trivial finite quotients. Let A be a Noetherian
ring on which the group G acts and let a be a G-invariant ideal. Then a has a G-invariant
primary decomposition.

Proof: Let a = q1 X . . . X qr be a minimal primary decomposition of a with Ass A =

tp1 , . . . , pr u being the set of associated primes. For each g P G one has
g g g
a = a = p1 X . . . X pr , (10.8)
g
and since g acts through an isomorphism of A, each ideal qi persists being primary,
g
but will have pi as a possible new radical. Anyhow, (10.8) gives a minimal primary
decomposition of a. The associated primes being unique, the group G permutes them,

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: primary decomposition and group actions 285

and this action is manifested through a group homomorphism from G to the symmetric
group Sym Ass A. By hypothesis, G has no non-trivial finite quotient so the image of
G in the finite group Sym Ass A must be trivial, and consequently each pi is invariant
under G.
A similar argument shows that isolated components of a also are invariant under G.
As to the embedded components, which are not unique, things are slightly more
complicated. They will not all be invariant, but we will be content when just finding
one that is. So let q be one having radical p. Since p is invariant and p N Ď q implies that
(p g ) N Ď q, we may apply Exercise 10.15, and conclude that the intersection tPG q g is
Ş

p-primary. We contend it is a component of a; indeed:

g
č č č č
a= ag = ( q1 X . . . X q r ) = q1 X . . . X q s X qsg+1 X . . . X qr ,
gP G gP G gP G gP G

where the s first components of a are the isolated ones. o

Exercises
(10.14) Let tqi uiP I be a collection of primary ideals (of any cardinality) all with the
same radical p. Assume that there is a natural number n so that pn Ď qi for all i. Prove
Ş
that the intersection iP I qi is primary with p as radical.
(10.15) Let the group G act on the Noetherian ring A and let q be a p-primary ideal.
Assume that p is invariant under G. Prove that gPG qg is p-primary and invariant
Ş

under G.
(10.16) Show that if k is an algebraically closed field, the group (k˚ )n has no finite
quotient.
(10.17) Show that a connected Lie-group has no finite quotient.
M

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 11

Krull dimension

Dimension is generally a complicated and subtle notion, and only in some good cases
is there a satisfactory definition. Vector spaces have a dimension as do manifolds (or at
least each connected component has). Manifolds are locally isomorphic to open sets in
some euclidean space, and the dimension of that euclidean space is constant along each
connected component, and is the dimension of the component.

There is another and naive approach to the concept of dimension. For example, in
three dimensional geometric gadgets—which are called threefolds—one might heuristi-
cally 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 of a topological space as “subgadgets”, and the dimension will be
the maximal length of such a chain, or rather the supremum of the lengths as they
might not be bounded. This definition works for any topological space, but the ensuing
dimension does not carry much information unless the topology is “Zariski-like” (the
only irreducible subspaces in a Hausdorff space are the points, so with this definition
of dimension all Hausdorff spaces will be of dimension zero ). Translated into algebra,
where prime ideals correspond to closed irreducible subsets, this leads to the concept of
Krull dimension of a ring; the supremum of the lengths of chains of prime ideals.

For varieties, or equivalently for algebras finitely generated over fields, there is
another good candidate for the dimension, namely the transcendence degree over the
ground field of the function field; that is the fraction field K ( A) of the coordinate ring
in case the variety is affine . This may be motivated by the fact the Krull dimension of
the polynomial ring k[ x1 , . . . , xn ] equals n (which is not obvious, but follows from the
Prinicipal Ideal Theorem), and obviously the transcendence degree of k( x1 , . . . , xn ) is n.
That the two coincide, is one of the important consequences of Noether’s Normalization
Lemma.
 288 krull dimension

11.1 Definition and basic properties

(11.1) Let A be a ring. We shall consider strictly ascending and finite chains tpi u of
prime ideals
p0 Ă p1 Ă . . . Ă pν .

Recall that the integer ν is called the length of the chain; it is one less than the number of
Krull dimension prime ideals, or if you want, it equals the number of inclusions. The Krull dimension of
(Krull-dimensjon) A will be the supremum of the lengths of all such chains. It is denoted by dim A. A
Saturated chains chain is said to be saturated if there are no prime ideals in A lying strictly between two
(mettede kjeder) of the terms, and it is maximal if additionally the chain cannot be lengthened, neither
Maximal chains upwards nor downwards; so the smallest term of a maximal chain is a minimal prime
(maksimale kjeder) and the larges a maximal ideal.
(11.2) Even if it happens that each chain of prime ideals in A is finite, there might
be arbitrary long chains, and the Krull dimension will be infinite. It is easy to find
examples among non-Noetherian rings whose Krull dimension is infinite—the ring in
Example 11.1 below is an obvious example of one with an infinite ascending chain—but
even Noetherian rings might have infinite Krull dimension; (the first example, which
we shall discuss in Example 14.1 on page 357, was discovered by Masayoshi Nagata).
However, these examples live on the fringe of the Noetherian society, and rings met in
mainstream algebraic geometry will all have finite dimension, and notably, every local
Noetherian ring has finite Krull dimension, as we shall see.
Note that in a ring whose Krull dimension is finite, the supremum is achieved, and
A will have saturated chains of maximal length; that is, of length equal to dim A.

Examples
(11.1) The polynomial ring in A = k[ x1 , . . . , xr , . . .] in infinite many variables is of infinite
Krull dimension. Each of the ideals pr = ( x1 , . . . , xr ) is prime, 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 = ( xr , xr+1 , . . .).

(11.2) Rings of dimension zero: A ring A is zero-dimensional when no chain of prime

ideals has more than one term; or in other words, when all prime ideals are both
minimal and maximal. If A in addition is Noetherian, there are according to the Lasker-
Noether Theorem (Teorem 10.19 on page 268) only finitely many minimal prime ideals.
Hence A has only finitely many prime ideals and they are all maximal, and as this is a
property that characterise Artinian rings among the Noetherian rings (Theorem 9.68
on page 256), a Noetherian ring is of dimension zero if and only if it is Artinian. In
particular, fields and finite products of fields are of dimension zero. But also all rings

10th January 2021 at 11:32am

Version 4.0 run 30
 definition and basic properties 289

shaped like quotients A/a of Noetherian rings by an ideal a that is m-primary for a
maximal ideal m, are of dimension zero.
(11.3) Domains of dimension one: It is worthwhile casting a glance on one-dimensional
rings as well. In a one-dimensional domain the zero ideal is a prime ideal, and saturated
chains are all of the form (0) Ă p. All non-zero prime ideals are therefore maximal, and
they are also minimal over the zero ideal (they are what we later on will be calling the
height one primes, one might also call them subminimal). Examples can be the principal Height one primes
ideal domains; in particular polynomial rings k[ x ] in one variable over fields and, of (høyde-en primidealer)

course, the integers Z will all be one-dimensional.

If A is not a domain, there may as well be saturated chains of length zero; in other
words, some of the minimal prime ideals might also be maximal (see Example 11.5
below).
(11.4) Some domains of dimension two: The polynomial ring A[ x ] over a pid A is of
dimension two. Back in Chapter 3 ( Proposition 3.30 on page 77) we showed, at least
when A has infinitely many prime ideals, that the maximal ideals of A[ x ] are of the type


g( x ), p where p is a prime element in A, and g( x ) a polynomial that is irreducible
modulo p. Furthermore, we saw that the non-maximal non-zero prime ideals all are


principal, they are either equal to ( p) for a prime element p in A or to g( x ) for
an irreducible and primitive polynomial g( x ). From this one easily deduces that all
maximal chains in A[ x ] are of length two. Among the rings of this kind we find the
polynomial rings k[ x, y] and Z[ x ].
The case when the pid A merely has finitely many maximal ideals, is more involved.
The dimension of A[ x ] will still be two, but the description of the ideals is different:
there may be maximal ideals that are principal (see Example 11.8 below).
K
ˇ Exercise 11.1 Let A be a Noetherian domain and let x P A a prime element so that
the principal ideal ( x ) is prime. Show that dim A( x) = 1. Hint: Show that the powers
( x n ) A(x) are all the ideals in A(x) . M
ˇ Exercise 11.2 Let p = (y, z) in the polynomial ring R = k[ x, y, z]. Show that dim Rp = 2
Hint: : Excerise 7.8 on page 181 may be useful. M

A useful inequality
(11.3) Any chain tpi u of prime ideals in A may be broken in two at any stage, say at
p = pν , and thus be presented as the concatenation of a lower chain, formed by the
members of the chain contained in p, and an upper chain, 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. A chain thus split, appears as:

p0 Ă p1 Ă . . . Ă pν´1 Ă pν = p Ă pν+1 Ă . . . Ă pn .

10th January 2021 at 11:32am

Version 4.0 run 30
 290 krull dimension

The primes contained in p are in a one-to-one correspondence with the prime ideals in
the localization Ap , hence the lower chains correspond to chains in Ap . Moreover, the
prime ideals containing p correspond to prime ideals in the quotient A/p, hence the
upper chains correspond to chains in A/p, and considering the suprema of the lengths
of such splices, we arrive at the following formula:

Proposition 11.4 Let A be a ring and p a prime ideal. Then

dim Ap + dim A/p ď dim A.

Note that the proposition is still valid if one or more of the dimensions are infinite,
provided the usual convention that n + 8 equals 8 is in force.
In some rings there are maximal chains—that is, saturated chains which cannot be
lengthened either way—of different lengths, and in that case the Krull dimension is the
length of the longest. For any prime ideal in a shorter chain, the inequality in (11.4)
will be strict. It is easy to give such examples when the ring A is not a domain (but
still is finitely generated over a field). Quite simply, coordinate rings of algebraic sets
with irreducible components of different dimension will do (see Examples 11.5 and 11.6
below).
(11.5) More generally, if the ring A has several minimal prime ideals, the space Spec A
will have several irreducible components, as in Example 11.5 below where the point
(0, 1) and the x-axis are the components. The dimension of A will be the largest of
the dimensions of the components, or if there are infinitely many, the supremum. In
algebraic terms this translates into:

Proposition 11.6 If tpi u are the minimal primes of A, then dim A = supi dim A/pi .
Proof: The intersection of the prime ideals in a chain being prime, any maximal chain
starts at minimal prime (Exercise 2.21 on page 53); furthermore chains of prime ideals
beginning at a prime p are in a one-to-one correspondence with chains of prime ideals
in A/p. o
Height of ideals (11.7) It is common usage to call dim Ap the the height of p, or more generally for any
(høyden til idealer) ideal a in A the heigh is the least height of any prime ideal containing a; that is

ht a = min ht p.
aĎp

Codimension of an One speaks also about the height of p over q when p Ą q are two primes. It equals the
ideal (kodimensjonen supremum of lengths of chains connecting q to p; or equivalently to dim( A/q)p .
til et ideal)
The height ht p is also called the codimension of V (p), since when the inequality in
Proposition 11.4 is an equality, it holds that ht p = dim Ap = dim A ´ dim A/p.
Catenary rings
(11.8) Rings with the property that for each pair of nested prime ideals all saturated
(katenære ringer)
chains connecting the two have the same length, are called catenary. The rings in the

10th January 2021 at 11:32am

Version 4.0 run 30
 definition and basic properties 291

examples just mentioned are both catenary despite having maximal chains of different
length, so being catenary is a weaker property than having maximal chains of uniform
length. There are many examples of Noetherian domains that are not catenary, but
these are rather exotic constructs you wouldn’t tumble over when practicing algebraic
geometry.
ˇ Exercise 11.3 Show that for a ring to be catenary it suffices that the defining property
holds for the nested pairs consisting of a minimal and a maximal ideal. Show that if
A/p is catenary for each minimal prime p, then A is catenary. M
Exercise 11.4 Show that k [ x, y] and Z[ x ] are catenary. Hint: Proposition 3.30 on
page 77. M
In several large and important classes of rings all members are catenary. For instance,
as we shall see later, all algebras which are finitely generated over a field are catenary as
are their localizations. For domains from this class of rings, the stronger property that
all maximal chains are of equal length also holds true, we shall say they are of uniform Uniform altitude
altitude. However this does not necessarily persist being true for localizations of such as (uniform høyde)

Examples 11.7 and 11.8 below shows.

One easily convinces oneself that equality holds in Proposition 11.4 for all primes p
in a local catenary domain, and in fact, Louis Ratliff has shown that the converse also is
true when A is Noetherian—a rather deep result with a rather involved proof.
Examples
(11.5) Let A = k [ x, y] with constituting relations xy = y(y ´ 1) = 0. It is the coordinate
ring of the algebraic subset V of A2k equal to the union of the x-axis and the point (0, 1).
The primary decomposition of the zero ideal in A is given as (0) = (y) X ( x, y ´ 1). y

Hence (y) and ( x, y ´ 1) are the minimal prime ideals in A. Now, ( x ´ a, y) is a maximal (0,1)
ideal for any a P k, so A possesses saturated chains
x
(y) Ă ( x ´ a, y),
and therefore dim A = 1. On other hand ( x, y ´ 1) is clearly a maximal ideal, and
hence it is both maximal and minimal. So V, even though it is one-dimensional, has a
component of dimension zero.
(11.6) If you want an example that is a local ring, refine the previous example and
consider A = k[ x, y, z] with constituting relations zx = zy = 0. Then A is the coordinate
ring of the union V of the xy-plane and the z-axis in A3k . Let further A be the local
ring of V at the origin; that is, the localization of A at ( x, y, z). In A the decomposition
(0) = (z) X ( x, y) is the minimal primary decomposition of (0) so that both (z) and (y, x )
are minimal prime ideals in A; and there are two maximal chains (z) Ă ( x, z) Ă ( x, y, z)
and ( x, y) Ă ( x, y, z) in A of different length.
(11.7) A semi-local ring: Consider the polynomial ring k[ x, y, z] and let p = ( x ) and

10th January 2021 at 11:32am

Version 4.0 run 30
 292 krull dimension

q = (y, z). Let S denote multiplicatively closed subset consisting of elements not in
the union p Y q. Furthermore, let A = S´1 k [ x, y, z]. We contend that A is a semi-local
ring with maximal ideals m = pA and n = qA whose heights are respectively one and
˚ In fact, the height at least* two, and there are maximal chains of unequal length in A. That m and n are
equals two. Later, this the only two maximal ideals is clear: indeed, every prime in A is of the form aA for a
will be crystal clear,
but for the moment we prime ideal a Ď k[ x, y, z] not meeting S; that is, a is contained in the union p Y q. Prime
do not spend energy on Avoidance then ensures that a either lies in p or in q. That dim Ap = 1 and dim Aq ě 2
it
are just Exercises 11.1 and 11.2 on page 289.
(11.8) Chains in the polynomial ring over a dvr: Recall that in Exercise 3.8 on page 78 you
were asked to describe the maximal ideals in the polynomial ring A[ x ] over a discrete
valuation ring A. As in that exercise, we let π be a generator for the unique maximal
ideal m of A. There are two types of maximal ideals in A[ x ]: first we have those of
shape ( g( x ), π ) with g( x ) being irreducible modulo π. These live in saturated chains of
length two:

(0) Ă ( π ) Ă ( g ( x ), π ) or (0) Ă ( g( x )) Ă ( g( x ), π ).

Secondly, we have those which are principal; i. e. those shaped like ( g( x )) where g( x )
is an irreducible polynomial which is constant modulo π (a typical example would be
(πx ´ 1)). These live in chains of length one:

(0) Ă ( g( x )).

So we see that dim A[ x ] = 2, but that there are saturated chains of different lengths.
But A[ x ] is catenary, chains connecting the minimal prime (0) to a fixed maximal ideal
are all of the same length.
It is noteworthy that a prime ideal ( g( x )) with g irreducible mod π lives in a unique
maximal chain: if ( g( x )) Ă (h( x ), π ) with h( x ) irreducible mod π, it is straightforward
to verify that ( g( x ), π ) = (h( x ), π ). So the ideal ( g( x )) is not the intersection of maximal
ideals, and the closed points of closed subsets of Spec A[ x ] are not always dense in the
subset (e. g. the closed points are not dense in V ( g( x )), as there is merely one).
K
Exercise 11.5 Let A be a dvr whose maximal ideal is generated by π. Describe all
prime ideals in A[ x ] lying between (0) and a maximal ideal of shape ( g( x ), π ) (where
g( x ) is a polynomial which is irreducible modulo π). M

Cutting out a hypersurface

Cutting a variety X with a hypersurface is a rather common technique in algebraic
geometry, which on the level of algebras corresponds to passing to the quotient A/( f ) A
of A by a principal ideal. One expects the dimension to go down, however, it might

10th January 2021 at 11:32am

Version 4.0 run 30
 krull’s principal ideal theorem 293

happen that the cutting hypersurface contains one of the components of X, and in case
that component is one of maximal dimension, the dimension stays the same. To avoid
such an accidental behaviour, one must assume that f does not lie in any of the minimal
prime ideals of A; then one has:

Proposition 11.9 Let A be ring of finite Krull dimension and let f P A be an element not
belonging to any minimal prime ideal in A, then dim A/( f ) A ă dim A.

Proof: Chains of prime ideals in A/( f ) A are in one-to-one correspondence with chains
in A all whose members contain f . Moreover, a prime ideal p that is minimal over ( f ),
is by hypothesis not minimal in A, and therefore properly contains a minimal prime
q of A. Consequently any ascending chain in A emanating from p can be lengthened
downwards by appending q. o

11.2 Krull’s Principal Ideal Theorem

This important theorem is also known under its German name Krull’s Hauptidealsatz.
It lies at the bottom of the dimension theory in commutative algebra and algebraic
geometry, and in his book [?] Irving Kaplansky refers to it as “may be the most important
single theorem in the theory of Noetherian rings”. There is however the rather simple
underlying intuition that the dimension of a solution space goes down by at most one
when an additional equation is introduced. We recognize this from the theory of linear
equations, but the principle has a much wider scope (as shows e. g. the Principal Ideal
Theorem).
(11.10) The scene is set with a ring A and the two main players, a prime ideal p and an
element x of p over which p is minimal. That is to say, there is no prime ideal properly
lying between ( x ) and p. And then the conclusion is that the height 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 x is non-zero-divisor, the height will be one.
Example 11.9 As a prelude to the theorem, it might be instructive to consider the special
and simple case that ( x ) is a prime ideal in a Noetherian domain A. Then there are
no non-zero prime ideals properly contained in ( x ): if p Ď ( x ) were one of the kind, it
would follow that xp = p; indeed, each member a P p would be on the form a = bx, and
because x R p, one would have b P p. In view of Nakayama Extended, the ideal p would
be killed by an element shaped like 1 + cx, which is absurd. K
(11.11) In the course of the proof we are obliged to use the symbolic powers of a prime
ideal q in A as a substitute for the actual powers. Recall (see Exercise 10.11 on page 275)
that the symbolic power q(n) is defined as q(n) = qn Aq X A. It has the virtue of being
q-primary, which the power qn might not be in general. But don’t be panic-stricken by
the appearance of these gizmos: A part from q(n+1) being contained in q(n) , the only

10th January 2021 at 11:32am

Version 4.0 run 30
 294 krull dimension

property we shall need, is that from cx P q(n) , but x R q, follows that c P q(n) . Indeed,
x is invertible in the localization Aq as it does not lie in q, so c P qn Aq follows from
cx P qn Aq .

(x) Ď p Theorem 11.12 (Krull’s Principal Ideal Theorem) Let A be a Noetherian ring and x an
element from A. Assume that p is a prime ideal in A which is minimal over ( x ). Then ht p ď 1.
Ă

q Proof: We are to show that there are no chain of prime ideals of length two of shape
Ă

q1 Ă q Ă p. By passing to the quotient A/q1 and subsequently localizing in p/q1 , we

q1 may assume that A is a local domain with maximal ideal p, and our task is to prove
that if q Ă p, then q = 0.
The first observation is that, since p is minimal over ( x ), the ring A/xA has only one
prime ideal. Being Noetherian, it is Artinian, and we have the opportunity to activate the
descending chain condition. The chain to exploit, is the descending chain t( x ) + q(n) un ,
where q(n) is the n-th symbolic power of q. The chain t( x ) + q(n) u corresponds to the
descending chain t ( x ) + q(n) /( x )u in A/xA and must eventually be stable as A/xA

is Artinian. Hence there is an n so that

( x ) + q( n +1) = ( x ) + q( n ) .

This entails that if a P q(n) , one may write a = b + cx with b P q(n+1) , so that cx P
q(n) . From this follows that c P q(n) since x R q, and consequently it holds true that
q(n) Ď q(n+1) + xq(n) Ď q(n) . Nakayama’s lemma yields that q(n) = q(n+1) . In its turn,
this yields that qn Aq = qn+1 Aq , and appealing once more to Nakayama’s lemma, we
may conclude that qAq = 0; that is, q = 0. o
Example 11.10 The non-Noetherian ring A = k[t, x1 , x2 , . . .] with constituting relations
xi = txi+1 for i P N, which was the subject of Exercise 9.18, gives an example underlining
that the Noetherian hypothesis is necessary. The principal ideal (t) A is maximal, but
contains the non-zero prime ideal p = ( x1 , x2 , . . .), and consequently it is of height (at
least) two.
Note that dim A = 2, but when t is killed, all the xi ’s are killed as well, so that
A/(t) A » k. Hence dim A/(t) A = 0, and killing t thus diminishes the dimension by
two. Notice that this may also happens for Noetherian rings. K

The general version of The Principal Ideal Theorem

This result, often called "The height theorem", is the natural generalization of the
Principal Ideal theorem to minimal primes over ideals with more than one generator.
The natural guess that the height will be at most the number of generators, is actually
true, and agrees with the naive intuition that imposing r constraints on a system should
at most lower the dimension of the solution space by r: it diminishes at most with one

10th January 2021 at 11:32am

Version 4.0 run 30
 krull’s principal ideal theorem 295

for each new condition imposed. This points to an induction argument, but a slightly
more subtle one than the naive intuition suggests.

Theorem 11.13 (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 = ( a1 , . . . , ar ). As indicated above the proof goes by induction 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. Because a is not contained in pd´1 , we may assume that a1 does not lie in
pd´1 , and so there is no prime lying properly between ( a1 ) + pd´1 and p. The radical of
( a1 ) + pd´1 therefore equals p, and a power of p is contained in ( a1 ) + pd´1 . Since a Ď p A
we may for ν sufficiently big, write
p
aiν = ci a1 + bi ( a1 ) pd´1

with bi P pd´1 , and we let b = (b2 , . . . , br ). Then b is contained in pd´1 . We contend that q
there is prime ideal q lying between b and pd´1 , properly contained in pd´1 ; indeed, if pd´2

pd´1 were minimal over b, the height of pd´1 would be at most r ´ 1 by induction, but b
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.
Now, the idea is to pass to the ring A/q. The ideal q + ( a1 ) contains a power of a,
hence there is no prime ideal between q + ( a1 ) and p, which means that the ideal p/q
is minimal over the principal ideal q + ( a1 )/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

Some consequences
(11.14) 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, and by the Height
Theorem the height of m, which is the same as dim A, is bounded by the number of
generators. Similarly, Noetherian rings enjoy a descending chain condition for prime
ideals. Any term in a chain is finitely generated and hence is of finite height, and the
length of the chain is bounded by the height of the top member. We have proved the
first part of the following:

Proposition 11.15 A local Noetherian ring is of finite Krull dimension bounded by the number
of generators of the maximal ideal. Noetherian rings satisfy the dcc for prime ideals in the
strong sense that there is a uniform bound on the length of descending chains.

Note that in any ring, Noetherian or not, the weaker property that any non-empty set
of prime ideals has a minimal and a maximal member holds true. A Zorn’s-lemma-
argument gives this, the intersection and the union of the ideals in a chain of primes
being prime.

10th January 2021 at 11:32am

Version 4.0 run 30
 296 krull dimension

Despite every chain of prime ideals being of finite length, Noetherian rings may be
of infinite dimension; the lengths of the different chains could be unbounded. The first
example of such a ring was given by Nagata and will be discussed in Example 14.1 on
page 357.
(11.16) Another special feature the lattice of prime ideals in Noetherian rings has, is that
in between two prime ideals, one strictly contained in the other, there will be infinitely
many prime ideals if any at all.

Proposition 11.17 Let A be a Noetherian ring and let q Ă p be two prime ideals. If there is a
prime ideal lying strictly between p and q, there will be infinitely many.

Proof: Assume there is a prime ideal strictly between p and q which means that the p is
of height at least two over q. If there were only finitely many prime ideals, say p1 , . . . , pr ,
lying strictly between p and q, there would by prime avoidance be an element x P p not
lying in any of the pi since by assumption p is not contained in any of the pi ’s. Then
p would be minimal over q + ( x ), and by the Principal Ideal Theorem p would be of
height one over q; contradiction. o
Curiously enough, the proposition fails spectacularly for non-Noetherian rings. One of
the simplest non-Noetherian valuation rings is of dimension two and has only three
prime ideals—which of course form a chain—and more generally, for every dimension n
there are similar examples of valuation rings having only n + 1 prime ideals. Describing
these example requires some theory about general valuations and will be done later in
the course (Proposition 15.63 and Example 15.9 on page 400).
These examples also show that the Principal Ideal Theorem may fail without the
Noetherian hypothesis; e. g. in the two-dimesional example just mentioned the maximal
ideal is of height two, but it is minimal over any one of its elements not in the other
non-zero prime ideal.

Example 11.11 Here is another example of the principal ideal theorem failing for non-
Noetherian rings. Back in Lecture 9, in Exercise 9.18, you were asked to studied the
non-Noetherian ring A = k[t, x1 , x2 , . . .] with constituting relations xi = txi+1 for i ě 1.
In this ring m = (t) A is a maximal ideal and p = ( x1 , x2 , . . .) is a prime ideal contained
in m (actually equal to i mi ). So in this ring the height of the principal ideal (t) A
Ş

equals two. K
(11.18) Among the different numbers associated with a ring which is reminiscent of
Embedding dimension being a dimension, is the so-called embedding dimension of a local ring A. If m denotes
(embeddingsdimen- maximal ideal of A, the embedding dimension of A is defined as the vector space
sjon)
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 [ x1 ], . . . , [ xr ] for

10th January 2021 at 11:32am

Version 4.0 run 30
 ufd’s once more 297

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 11.19 Assume that A is a local Noetherian ring with maximal ideal m. Then
dim A ď dim m/m2 .
Exercise 11.6 Show that among the Noetherian rings only the Artinian ones and the
semi local one-dimensional ones have finitely many prime ideals. M
Exercise 11.7 Assume that A is a quotient of te polynomial ring k[ x1 , . . . , xn ] over the
field k. Let m be a maximal ideal in A. Show that the embedding dimension of Am is at
most n. M

11.3 UFD’s once more

In Lecture 3 we showed a criterion of Kaplansky’s (Proposition 3.13 on page 73) 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 improved. Citing Proposition 11.15
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
elements, 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 11.20 A Noetherian domain A is a ufd if and only if every prime ideal of height one
is principal.
(11.21) A direct consequence of the theorem is that if A is Noetherian and factorial,
any localization S´1 A of A is factorial; indeed, the height one primes in S´1 A are
precisely the ideals shaped like S´1 p for height one primes p in A not meeting S, and
localizations of principal ideals persist being principal.
The converse is however not true, as shows the example of the simple cusp A =
k [ x, y] with constituting relation y2 = x3 . Inverting x gives A x = k[t, t´1 ], where
t = xy´1 ; this is factorial, but A is not. The element x is irreducible but not prime, and
this is loosely speaking, what prohibits A from being a ufd. Inverting x makes it unit,
and the problem disappears.
However, if one merely inverts prime elements, the ring will be a ufd when the
localization is. The following proposition is due to Nagata:

Proposition 11.22 Let A be a domain with ascending chain condition on principal ideals (e. g.
Noetherian) and S a multiplicative subset generated by prime elements. If S´1 A is factorial,
then A is factorial.

10th January 2021 at 11:32am

Version 4.0 run 30
298 krull dimension

Note that the elements from S are non-empty products on primes so that S does not
contain any unit.
Proof: By Kaplansky’s criterion it suffices to exhibit a prime element in every prime
ideal p of A. If p X S ‰ H, this is immediate, so we may well assume that p X S = H.
Let x be prime element in S´1 p. We may assume that x lies in A, and by the lemma
below, we may also assume that x has no factor from S. That x is a prime in S´1 A
means that if x|ab, it follows that x divides either a or b in S´1 A, say a; that is, it holds
true that sa = a1 x for some s P S and some a1 P A. Now, s is a finite product of primes
not dividing x, consequently s may be cancelled, and a = a2 x for some a2 P A. o
Here comes the required lemma:

Lemma 11.23 Let A be a domain with ascending chain condition on principal ideals (e. g.
Noetherian) and let S be a multiplicative system in A generated by primes. If x P A does not
belong to S, either x has no factor from S, or one may write x = sy where s P S and y has no
factor from S.

Proof: Consider the set Σ of principal ideals (z) where z runs through elements such
that z´1 x P S. If non-empty the set Σ has a maximal element say (y). Then x = sy
with s = xy´1 and s P S. Assume the that y = tz with t P S. Then (z) P Σ because
xz´1 = st P S and it follows that t is a unit, a contradiction. o
Example 11.12 Introduce a grading of the polynomial algebra B = k[ x1 , . . . , xr ] by
assigning positive weights wi to the variables xi ’s. Moreover, let f be a polynomial,
irreducible and homogeneous of degree w. Let A = k[ x1 , . . . , xr , z] with constituting
relation zc = F ( x1 , . . . , xr ). Assume that c ” ˘1 mod w. Then A is factorial.
The trick is to consider the localization A[z´1 ]. The first step is to observe that we
have A/(z) A » k[ x1 , . . . , xr ]/( f ), which is an integral domain because f is assumed to
be irreducible. Hence z is a prime element of A, and it suffices to see that A[z´1 ] is
factorial. Now, in A[z´1 ] one has the elements yi = z´dwi where d is the natural number
so that c = ˘1 + dw. One finds

zc = f ( x1 , . . . , xr ) = f (zdw1 y1 , . . . , zdwr yr ) = zdw f (y1 , . . . , yr ) = zc˘1 f (y1 , . . . , yr ),

and it therefore holds true that

z = f (y1 , . . . , yr ) or z´1 = f (y1 , . . . , yr )

according to c ” 1 or c ” ´1 mod w. But in both cases we conclude that A[z´1 ] =

k[y1 , . . . , yr , f ´1 ], which is a factorial ring. Hence A itself is factorial by Proposition 11.22.
In particular the so-called Du Val E8 -singularity k[ x, y, z]/( x2 + y3 + z5 ) is factorial.
K

10th January 2021 at 11:32am

Version 4.0 run 30
 system of parameters 299

11.4 System of parameters

(11.24) 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 the ideal a they System of parameters
generate is m-primary. Since m is maximal, this amount to the radical of ( x1 , . . . , xn ) (parmetersystemer)

being equal to m (Proposition 10.4 on page 261), or if A is Noetherian that a contains

some power of m.

Proposition 11.25 Every local Noetherian ring has a system of parameters.

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 each ideal
ai = ( x1 , . . . , xi ) generated by the i first elements in the sequence 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 2.32 on page 42), and we
Ť
may pick an element xν+1 from A so that xν+1 P m, but xν+1 R i pi . Then any prime
ideal minimal over aν+1 = ( x1 , . . . , xν+1 ) 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
(11.26) The geometric counterpart of a system of parameters is, given a variety X and
point P on X, a sequence of hyper-surfaces that locally intersect the given variety in just
the point P, or expressed more precisely, that P is an isolated point in the intersection
of X and the hyper-surfaces.
(11.27) One cannot in general hope that the maximal ideal itself is generated by
dim A elements. The simple double point A = k [ X, Y ]/(Y 2 ´ X 2 ( X + 1)) gives an
easy example. The maximal ideal m = ( x, y) requires both x and y as generators;
indeed, no non-trivial linear combination αX + βY with α, β P k can for degree reasons
belong to (Y 2 ´ X 2 ( X + 1)), and therefore x and y are linearly independent modulo
m2 = ( x2 , xy, y2 ).
The geometric situation is as follows. Both the X-axis and the Y-axis (and in fact,
any line through the origin) intersect the curve C = V (Y 2 ´ X 2 ( X + 1)) only at the
origin, but because C has a double point there, there will always be an intersection
multiplicity. Heuristically, the curve C has two branches through the origin, and each
one contributes to the intersection with the line.
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 the local Regular local rings
rings Ap are regular for all prime ideals p in A. (regulære lokale ringer)

10th January 2021 at 11:32am

Version 4.0 run 30
 300 krull dimension

Exercise 11.8 With notations as above, show that ` Ap ( Ap /(Y ´ αX ) Ap ) = 2 when α

is a scalar and α ‰ ˘1, but that ` Ap ( Ap /(Y + X ) Ap ) = ` Ap ( Ap /(Y ´ X ) Ap ) = 3. The
excess length in the latter case is heuristically explain by the lines Y = ˘X being tangent
to C at the origin). M
Exercise 11.9 A du Val singularity. Consider the ring k [ x, y, z] with the constituting
relation xy + zn+1 = 0 (that is, the ring k [ X, Y, Z ]/( XY + Z n+1 )) and let m = ( x, y, z).
a) Show that x, αy + βz is a system of parameters for Am whenever (α, β) ‰ (0, 0).
Compute ` Am ( Am /( x, αy + βz))
b) Show that any pair of linearly independent linear forms w1 and w2 in x, y and z
form a system of parameters for Am . Discuss ` Am ( Am /(w1 , w2 )).
M

Dimension and fibres

(11.28) One of the more important formulas taught in courses in linear algebra relates
the dimension of the kernel and the cokernel of a linear map. Recall that for a given
linear map φ : V Ñ W it reads as

dim im φ + dim ker φ = dim V + dim W,

which, since dim im φ ď dim V, yields the inequality dim V ď dim + dim Wφ´1 (0). For
a smooth map φ : X Ñ Y between manifolds there is a similar inequality

dim X ď dim Y + dim φ´1 (y),

for y when y belongs to the image φ( X ) of φ; in fact, this is just the inequality from
linear algebra above applied to the derivative of φ. When φ : X Ñ Y is a map of varieties,
or between spectra of rings, there is a similar formula. We do not intend to enter any
geometric discussion, but shall give a a closely related algebraic version for maps of
local rings.
Maps of local rings (11.29) Recall that a map of local rings is ribetween two local rings which sends the
(lokale ringavbildning) maximal ideal into the maximal ideal.

Proposition 11.30 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 Noetherian rings. Then it holds true
that
dim B ď dim A + dim B/mB.

Proof: We begin with choosing two systems of parameters. The first will be a system
of parameters x1 , . . . , xr for the maximal ideal m, and the second one for the ideal n/mB
in the ring B/mB. Let y1 , . . . , ys be a lifting to B of the latter. We contend that the
ideal a = (φ( x1 ), . . . , φ( xr ), y1 , . . . , ys ) generated in B by the two systems is n-primary;

10th January 2021 at 11:32am

Version 4.0 run 30
 system of parameters 301

indeed, some power nν is contained in (y1 , . . . , ys ) + mB, and consequently some higher
power lies in a since high powers of m lie in ( x1 , . . . , xr ). o
Example 11.13 Strict inequality may occur—Affine blow up: Consider the map ψ : C2 Ñ C2
sending a point (u, v) to (u, uv). The fibre over (0, 0) 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 the local algebra in the context of Proposition 11.30 we consider
the map of rings φ : k[ x, y] Ñ k[u, v] that sends x ÞÑ u and y ÞÑ uv (think of x and y
as coordinates in the target C2 ). To set up the appropriate localizations a we let
n = (u, v) be the ideal of the origin in the source C2 and B = k [u, v]n the local
ring there; similarly, we put m = ( x, y), the ideal of the origin in target C2 , and let
A = k [ x, y]m . One verifies that mB = (u, uv) = (u), and that the “fibre” B/mB is given
as B/mB = (k [u, v]/(u, uv))(u,v) = k [v](v) . It is of dimension one, whereas, of course,
dim B = dim A.
The map ψ restricts to a bijection between C2 zV (u) and C2 zV ( x ), the complements
of respectively the v-axis and he y-axis: indeed, v may be recovered from uv whenever
u ‰ 0. The most dramatic effect of ψ is to collapse the v-axis to one point, the origin in
the target C2 . For this reason ψ is called a “blow–down", or when seen in the perspective
from the target C2 , a “blow–up”. K
Exercise 11.10 With the notation of Example 11.13 above, show that the map φ induces
an isomorphism between k[ x, x´1 , y] and k [u, u´1 , v]. Show that the set-theoretical image
of ψ equals C2 zV ( x ). Translate this into a statement about prime ideals and the map φ,
and verify it. Let a P k be a scalar. Give a description the inverse image φ´1 (v ´ a)k[u, v]
of the principal ideal (v ´ a)k[u, v] under φ (corresponding to the image of the line
v = a under ψ), and interpret this geometrically. M

v y

u x

E =V ( u )

Exercise 11.11 Assume that A and B are domains that are finitely generated over a
field k and let φ : A Ñ B be an injective k-algebra homomorphism.

10th January 2021 at 11:32am

Version 4.0 run 30
302 krull dimension

a) Show that dim B ď dim A + dim B/mB for any maximal ideal m in A such that
mB is a proper ideal.
b) Show that there is an element f P A so that dim B = dim A + dim B/mB for all
maximal ideals m P D ( f ) Ď Spec A.
Interpretation this in a geometric language. M

11.5 Dimension of polynomial rings

We turn to the question about what the relation between the dimension of a ring A
and of the polynomial ring A[t] is. The intuitive, and naive, guess would be that
the polynomial ring is of dimension one more than A; this is certainly supported by
polynomial rings k[ x1 , . . . , xn ] being of dimension n, and more generally, by being true
whenever A is finitely generated over a field. But nature is not that simple, and thist
does not hold true universally; some non-Noetherian rings show their treacherous
character in this respect.The dimension of A[t] can in fact be as large as 2 dim A + 1.
However, Noetherian rings behave well, and for those dim A[t] = dim A + 1, as we
soon shall see. The proof will be a direct application of the dimension-inequality from
Proposition 11.30 on page 300.
Of course, for algebras finite generated over a field we have already shown theses
results, but we find this "doubling" justified since many important classes of Noetherian
rings, for instance those being finitely generated over Z and the complete Noetherian
rings, are not of finite type over a field.
We begin by studying how the prime ideals in A[t] are related to the prime ideals
in A, or in other words, we shall take a closer look at the fibres of the projection map
π : Spec A[t] Ñ Spec A induced by the inclusion A Ď A[t].
The fibres of the projection map
(11.31) Each prime ideal p in A gives rise to an ideal p+ = pA[t], which is formed by
the polynomials i ai ti having all coefficients in p. Clearly p+ X A = p.
ř

Our next observation is that the ideal p+ is a prime ideal. Since the polynomial ring
A/p[t] is an integral domain, this follows e. g. from the isomorphism

A[t]/pA[t] » A/p[t], (11.1)

which is induced by the map A[t] Ñ A/p[t] that sends ai t to [ ai ]ti . This is obviously
ř i ř

a surjective homomorphism whose kernel is easily seen to be pA[t]: a polynomial ai ti

ř

is mapped to zero if and only if [ ai ] = 0 for all i; that is, if and only all ai lie in p.
(11.32) The prime ideal p+ is one of the ideals that contract to p, or in geometric terms,
one which belongs to the fibre over p of the map π : Spec A[t] Ñ Spec A. But there are
others; in fact, we shall see there are infinitely many. To begin with we consider the case
that A is a domain and p = 0, to which the general case subsequently will be reduced.

10th January 2021 at 11:32am

Version 4.0 run 30
 dimension of polynomial rings 303

Lemma 11.33 Let A be a domain with fraction field K. Sending p to pK [t] sets up a one a one
to one correspondence between non-zero prime ideals p in A[t] such that p X A = 0 and the
non-zero proper ideals in K [t].
Proof: Observe that K [t] equals the localisation S´1 A[t] where S is the multiplicative
set S = Azt0u. The lemma is then just Proposition 7.20 on page 182 on primes in
localizations. o
In general, if q X A = p it holds that p+ Ď q so prime ideals in A[t] that intersects A in p
are in one-to-one correspondence with prime ideals in A[t]/pA[t] whose intersection
with A/p in the zero-ideal, or in other words prime ideals in the polynomial ring A/p[t]
that intersects A/p in zero. Hence by replacing A by A/p we place ourselves in the
situation of the previous lemma, and with the ad hoc notation Kp for the fraction field of
A/p, we arrive at the following description of the fibre of π over p:

Lemma 11.34 Let A be a ring and let p Ď A be a prime ideal. There is a one-to-one inclusion
preserving correspondence between the lattice of primes in A[t] contracting to p and the lattice
of prime ideals in the polynomial ring Kp [t].

Examples
(11.14) One of the simplest examples of the set-up in the lemma above, is the inclusion
C[ x ] Ď C[ x, y], whose geometric incarnation we denote by π : Spec C[ x, y] Ñ Spec C[ x ].
The maximal ideals in C[ x, y] are of the form ( x ´ a, y ´ b) and constitute a C2 , while
those in C[ x ] are shaped like ( x ´ a) and constitute a C. Restricted to the set of
maximal ideals the map π is, of course, just the first projection C2 Ñ C; indeed,
( x ´ a, y ´ b) X C[ x ] = ( x ´ a) (trivial since ( x ´ a) is maximal in C[ x ]).
The fibre over ( x ´ a)C[ x ] consists of the point* ( x ´ a)+ = ( x ´ a)C[ x, y], whose ˚ This is often called

vanishing locus is the line x = a, and the maximal ideals ( x ´ a, y ´ b) corresponding to generic point of the
fibre
points on that line—these are all, since the irreducible polynomials in C[y, x ]/( x ´ a) »
C[y] are linear. The fibre of π over the zero ideal (0) comprises all the principal prime
ideals, i. e. those shaped like ( f ( x, y)) with f irreducible, and additionally the zero
ideal.
It is noteworthy that the set Spec C[ x, y] is not equal to the Cartesian product of
the sets Spec C[ x ] and Spec C[y]— there is no room for the principal primes in the
latter—whereas the set of maximal ideals equals the product of the two sets of maximal
ideals.
(11.15) Back in Paragraph 3.29 on page 77 we took a look at Spec Z[t]. If ( p) P Spec Z
is a prime ideal generated by a prime p, then ( p)+ is just the principal ideal ( p)Z[t].
The fibre of π over ( p) is the closed set V (( p)) consisting of prime ideals containing p.
A part from ( p)Z[t] itself, it comprises all maximal ideals of shape ( p, f (t)) where f (t)
is a polynomial in Z[t] which is irreducible mod p; that is, its class in F p [t] generates

10th January 2021 at 11:32am

Version 4.0 run 30
 304 krull dimension

a maximal ideal. Hence the fibre over ( p) is homeomorphic to Spec F p [t]. The fibre
of π over the zero ideal (0) in Z consists of the principal ideals ( f (t))Z[t] generated
by irreducible polynomials. The polynomials persist being irreducible in Q[t] (e. g. by
Gauss’s Lemma) so that fibre is homeomorphic to Spec Q[t].
Discrete valuation (11.16) Of quite another flavour is the ring A[t] when A is a so-called discrete valuation
rings (diskrete ring; that is, a local Noetherian domain of dimension one whose maximal ideal m is
valuasjonsringer)
principal, say generated by π. The localizations k[t]p and Z( p) are instances of such.
The space Spec A has two points m and (0). The fibre over m is Spec k[t] where k
denotes the residue field k = A/m, and the fibre over (0) equals Spec K [t] with K being
the fraction field of A.
Notice that, unlike in the two preceding examples, the "generic fibre", i. e. the fibre
over (0), contains maximal ideals, namely the principal ideals ( f (t)) that are generated
by polynomials f (t) which are invertible modulo π; that is, they can be brought on of
the form πg(t) + 1.
See also Proposition 3.30 on page 77 and Exercise 3.8.
K

Exercises
(11.12) Let A be a discrete valuation ring (as defined in example 11.16 above). Show
that a non-zero prime ideal in A[t] is of one of the four types:
i) If p is maximal and belongs to the fibre over m; that is, p X A = m, then
p = ( x, f ) with f P A[t] being irreducible mod x.
ii) If p is prime and belongs to the fibre over m, then p = m+ = mA[t].
iii) If p is maximal lies in the generic fibre; i. e. p X A = (0), then p = ( f ) with
f P A[t] primitive and irreducible and being a unit mod x.
iv) If p is prime and in the generic fibre but not maximal, then p = ( f ) with f
primitive and irreducible but not a unit mod x.
(11.13) Describe the fibres of the canonical map π : Spec CJxK[t] Ñ Spec CJxK. If Ctxu
denotes the ring of convergent power series with complex coeffcients, describe the fibres
of π : Spec Ctxu[t] Ñ Spec Ctxu.
(11.14) Describe the fibres of the canonical map π : Spec Z2 [t] Ñ Spec Z2 .
M

Bounds on the dimension of the polynomial ring

(11.35) The fibre over p of the projection map Spec A[t] Ñ Spec A is thus canonically
homeomorphic to Spec Kp [t]. So in addition to p+ , we find infinitely1 many other prime
1 Non-zero prime ideals in K [ t ] are generated by irreducible polynomials of which there are infinitely
p
many

10th January 2021 at 11:32am

Version 4.0 run 30
 dimension of polynomial rings 305

ideals in A[t] contracting to p. They all contain p, but there is no inclusion relation
between any of them.

Proposition 11.36 The inequalities dim A + 1 ď dim A[t] ď 2 dim A + 1 hold true for any
ring A.

Proof: In any chain tpi u in A[t] contracting to a chain tpi u in A at most two members
contract to each pi . Hence the length of tpi u is at most 2 dim A + 1. On the other hand, if
tpi u is a chain in A of length r, the chain pi+ will be of length r since each pi + intersects
back to pi . This gives dim A[t] ě r, but there are prime ideals in A[t] strictly greater
than pr+ which may be joined to the chain, and hence dim A[t] ě r + 1. o
Abraham Seidenberg gave examples of non-Noetherian rings showing that dim A[t]
may take any value between dim A and 2 dim A + 1. Below we reproduce one of these
examples (Example 14.2 on page 359), which in fact goes back to Krull, describing a
ring with dim A = 1, but such that dim A[t] = 3.

The case of Noetherian ring

Noetherian rings are in this context well behaved. One has

Theorem 11.37 Let A be a Noetherian ring, and p Ď A a prime ideal. Then ht p+ = ht p. In

particular, it holds true that dim A[t] = dim A + 1.

Proof: We merely need to prove that the inequality ht p+ ď ht p holds for all prime
ideals p in A; this implies the equalities ht p+= ht p and dim A[t] ď dim A + 1. Indeed,
if m is a maximal ideal in A[t] and p Ď m X A, it holds that ht m = 1 + ht p+ = 1 + ht p ď
dim A + 1. And, as shown in the proof of Proposition 11.36 above, any chain in A
induces a chain in A[t] of the same length, we have ht p+ ě ht p.
Replacing A by Ap we may assume that A is local, and designating the maximal
ideal by m, we are to show
dim A[t]m+ ď dim A.

Bearing the inequality (11.30) in mind (with B = A[t]m+ ) we will be trough once
we prove that dim A[t]m+/mA[t]m+ = 0. But this holds since A[t]m+ /mA[t]m+ is
isomorphic to the rational function field k (t) over the residue field k = A/m: reducing
coefficients modulo m gives an isomorphism A[t]/mA[t] » k[t], as in (11.1), under
which the multiplicative system S = A[t]zm+ maps to the multiplicative set k [t]zt0u in
k [t]. Hence the induced map between the localizations yields the desired isomorphism.
o

Algebras finitely generated over fields again

(11.38) From Theorm 11.37 we deduce by an obvious induction that the polynomial
ring k[ x1 , . . . , xn ] is of dimension n, and gives another proof of this important fact. And

10th January 2021 at 11:32am

Version 4.0 run 30
306 krull dimension

similarily, we see that dim Z[ x1 , . . . , xn ] = n + 1. More over, that ht p+ = ht p yields the

stronger result, that maximal ideals are all of the same height, equal to the dimension.

Theorem 11.39 If A is a domain finitely generated over a field, or over a one dimensional
Noetherian domain R with infinitely many prime ideals. Then all maximal ideals in A are of
height equal to dim A.

The proof of the theorem relies on the following nice proposion from Kaplansky’s book
about elements generating the fraction field of a domain; we already encountered it
hastily in Exercise ?? back in Chapter 12.

Proposition 11.40 Let A be a domain with fraction field K, and u an element in A. Then the
following three statements are equivalent:
i) The element u is contained in all non-zero prime ideals in A;
ii) Every non-zero ideal in A contains a power of u;
iii) K = A[u´1 ].
The only Noetherian rings for which the three statements hold, are the semi–local rings of Krull
dimension at most one.

Proof: That two first statements are equivalent, is clear as the radical of any ideal is
the intersection of the prime ideals containing it, so let us see that ii) implies iii). To
that end, pick any non-zero element x P A. By ii) we may find a natural number r so
that ur P ( x ); in other words, it holds true that x´1 = yu´r for some y P A, and hence
x´1 P A[u´1 ]. To establish the converse implication assume that K = A[u´1 ], and let
a be a non-zero ideal. For every non-zero element x P a it holds that x´1 = yu´r for
some u and some r; that is, ur P ( x ) Ď a.
To prove the last statement, note that any prime ideal in A of height one is minimal
over u, and there is only finitely many such when A is Noetherian (Proposition 9.17 on
page 235). Furthermore if a prime ideal in A were of height two, it would according to
Proposition 11.17 on page 296 contain infinitely many prime ideals of height one, so
there are not any. Hence A is of dimension one. Then all non-zero prime ideals are
maximal and of height one, and as we have seen, there are only finitely many. o

Lemma 11.41 Assume that A is a Noetherian domain and that B is a domain containing A
which is finitely generated as an A-algebra. If the spectrum Spec B is finite, the spectrum
Spec A is also finite.

Proof: By descending induction on the number of generators B requires, it suffices to

do the case that B = A[u]. The ring A[u] is Noetherian (Hilbert’s Basis Theorem) and is
of dimension at most one. Here the proof bifurcates:

10th January 2021 at 11:32am

Version 4.0 run 30
 dimension of polynomial rings 307

1) The case that A[u] is field. Then u is invertible in A[u], and one easily sees that
u´1 is integral over A: indeed, there is an expression

u ´1 = ( a n u n + . . . + a 1 u + a 0 )

with ai P A, which gives an integral dependence relation for u´1 when multipled with
u´n . Applying Proposition 11.40 above to R = A[u´1 ], which is Noetehrian by Hilbert’s
Basis Theorem, we deduce that Rhas finitely many prime ideals. Now R is integral over
A, so by Lying–Over also A has only finitely many prime ideals.
2) The case that A[u] is not a field. Being Noetherian and having just a finite number
of prime ideals, it is of dimension one. The element u cannot be transcendental, for
then A would be a field and in the polynomial ring over a field there are manifestly an
infinite number of prime ideals. Hence dim A = 1 and u is algebraic. Thus there is a
relation
a n u n + . . . + a1 u + a0 = 0
with ai P A. The coeficient an is contained in finitely many prime ideals, which are
exactly those that disappear in the localized ring A an . Moreover, A[u] an is integral over
A an . Then again by Lying–Over, the ring A an has only finitely many prime ideals since
this is the case for A[u] an , and consequently A has only finitely many prime ideals. o
(11.42) Fiannly we are prepared for the proof of theorem 11.39:
Proof of Theorem 11.39: It suffices to see that conclusion holds for polynomial rings
over R. And by induction on the number of variables, it will be enough to prove that if
the theorem holds for A, it holds for the polynomial ring A[t].
We saw above that ht p+ = ht p, and it therefore suffices to show that for all maximal
ideals m in A[t] the intersection p = m X A is a maximal ideal in A; indeed, there is no
other prime in the fibre of π over p than m which contains p+ . Replacing A with A/p
we may assume that A is a domain and m X A = 0, and the task is then to show that A
is a field.
Denote the field A[t]/m by K and let u P K be the inverse of the class [t]; that is,
u = [t]´1 (we may assume that t does not belong to m, for if it did, we would have
A = A[t]/m). There is an inclusion A Ď K and K = A[u´1 ].
The next observation is that u integral over A; indeed, since u P K and K is generated
by u´1 over A, there must be a relation like u = a0 + . . . + ar u´r with ai P A and ar ‰ 0,
and it yields an integral relation for u when multiplied by ur . By Lemma 12.28 on
page 327, it suffices therefore to show that A[u] is a field.
Certainly as A[u´1 ] is a field, it equals the fraction field of A[u], so when we let
Lemma 11.40 above come into play, we may conclude that A[u] semi-local ring of
diemension at most one. If R is a field, the intersection of all maximal ideals in A[u] is
zero, so A[u] is a field. Otherwise R, if R does not map injectively into A[u] the image

10th January 2021 at 11:32am

Version 4.0 run 30
 308 krull dimension

is a field, A[u] is a field by what just did.Finally, if R is contained in A[u] Lemma 11.41
yeids that R is semi-local and it isn’t. o

Appendix: The trace and the norm

In this appendix the stage is set by a finite extension of domains A Ď B with A being
integrally closed in its fraction field K. The fraction field L of B is then a finite extension
of K whose degree we shall denote by n. In particular, L is a vector space over K of
dimension n.
(11.43) With every element a in L we may associate the multiplication map [ a] : L Ñ L
defined by the assignment x ÞÑ ax. It certainly is K-linear, and we may apply linear
algebra when studying it. As any linear endomorphism, the multiplication map [ a] has
The characteristic a characteristic polynomial, which is given as
polynomial (det
karakteristiske Pa (t) = det(t ¨ I ´ [ a]) = tn + an´1 tn´1 + . . . + a1 t + a0 .
polynomet)

The coefficients ai are interesting invariants of the element a, in particular the constant
The trace of elements term and the subleading term which up to sign are the trace and the norm of a;
(sporet
The normtil
of et
anelement)
element more precisely, the trace is defined as tr L/K ( a) = ´an´1 and the norm is given as
(normen til et element) NL/K ( a) = (´1)n an . Observe that the trace is K-linear in a, and the norm, which in fact
is nothing but the determinant det[ a], depends multiplicatively on a.
With a there is also associated another polynomial, the minimal polynomial m a (t)
over K, which is the monic polynomial of lowest degree such that m a ( a) = 0; or, if you
want, the monic generator of the ideal in K [t] consisting of polynomials having a as root.
From elementary field theory we know that K ( x ) = K [t]/(m a ), and hence the degree of
m a coincides with [K ( a) : K ].
(11.44) From the Cayley-Hamilton theorem ensues that a is a root of Pa , and therefore
Pa will always have the minimal polynomial m a as factor. In most cases the two differ,
the exception being when L is a primitive extension of K with a as a generator; that is,
when L = K ( a). Both polynomials will then be of degree n and must agree. However, in
all cases the two are closely related; the characteristic polynomial Pa is always a power
of the minimal one m a :

Lemma 11.45 (Characteristic and minimal polynomials) The characteristic polyno-

mial of a is a power of the minimal polynomial; more precisely, Pa = mra where r = [ L : K ( a)].

Proof: Let x1 , . . . , xr be a basis for K ( a) over K and y1 , . . . , ys one of L over K ( a) chosen

so that x1 = y1 = 1. The products xi y j constitute a basis for L over K, and this leads to
a decomposition of L as a K-vector space shaped like:

L = K ( a ) ‘ K ( a ) y2 ‘ . . . ‘ K ( a ) y s ,

10th January 2021 at 11:32am

Version 4.0 run 30
 dimension of polynomial rings 309

Clearly each of the subspaces K ( a)yi is invariant under multiplication by a, Even more
is true: multiplication by yi is an isomorphism from K ( a) to K ( a)yi which commutes
with the multiplication map [ a]; hence the characteristic polynomial of [ a] acting on
K ( a) and the one of [ a] acting on K ( a)yi are equal. But as we observed just before the
lemma, the characteristic polynomial of [ a] acting on K ( a) coincides with the minimal
polynomial m a .
Finally, it is a general fact that if θ is an endomorphism of a vector space having a
À
direct sum decomposition Vi with each summand invariant under θ, the characteristic
polynomial of θ equals the product of the characteristic polynomials of θ acting on each
direct summand; i. e. of the restrictions θ|Vi . In our case this gives the desired equality
Pa = mra , which closes the proof. o

Corollary 11.46 Assume thatK Ď L is a field extension and a P K an element. Then it holds
that
i) tr L/K a = [ L : K ( a)] trK (a)/K a;

[ L:K (a)]
ii) NL/K ( a) = NK (a)/K ( a) .

(11.47) One of main properties of trace and norm (or any of the other symmetric
functions of the eigenvalues, for that matter) is that they take integral elements to
integral elements:

Proposition 11.48 Let A be a domain with fraction fied K and let K Ď L be an extension of
fields. Assume that a P L is an element integral over A, then tr L/K ( a) and NL/K ( a) are integral
over A as well.

Proof: In an algebraic closure of K, the polynomial Pa splits into a product of linear

factors, the roots being the eigenvalues λi of [ a]. If the element a P L is integral over A,
it satisfies an integral dependence relation

f (t) = td + ad´1 td´1 + . . . + a1 T + a0 = 0,

where the coefficients ai belongs to A. Assuming d is the minimal degree of such a

relation, we infer that p(t) equals the minimal polynomial of a; that is f = m a .
By Lemma 11.45 above, the characteristic polynomial Pa is a power of m a and
therefore has coefficients in A. Consequently, the eigenvalues λi of [ a] being roots in Pa
are integral over A as well. Now, the trace is the sum of the eigenvalues, and the norm
equals their product, so these two are both integral over A (and belong to K). o

The trace
Proposition 11.49 Let K Ď L be an extension of fields. Then the three properties following
hold true that:

10th January 2021 at 11:32am

Version 4.0 run 30
310 krull dimension

i) The trace is tr L/K is K-linear, and tr(1) = [ L : K ];

ii) If a P L is an element one has tr L/K = [ L, K ( a)] trK (a)/K ;
iii) If K Ď L1 Ď L is an intermediate field, then tr L1 /K ˝ tr L/L1 = trl/K ;

(11.50) Well-known properties of linear maps translate into basic properties of the the
norm and the trace:

Lemma 11.51 Let A Ď B be an extension of domains such that the extension of their fraction
fields K Ď L is finite. For elements f , g P L, it holds true that
i) The norm is multiplicative; that is, N ( f ¨ g) = N ( f ) ¨ N ( g) and N (1) = 1. If f P K,
N ( f ) = f [ L:K ] ;
ii) The trace is K linear, and tr(1) = [ L : K ];
iii) When A is normal and f P B, the element f is a factor of N ( f ); that is, N ( f ) = f b
for some b P B

Be aware that the equality in ii) takes place in K; so that when K is a field of characteristic
p and p divides the degree [ L : K ], it holds that tr(1) = 0.
Proof: Since obviously [ f g] = [ f ] ˝ [ g] the norm is multiplicative as the determinant is,
and if f P K, the multiplication map [ f ] is just f times the identity id[ L]. The trace is
the sum the diagonal elements of the matrix of [ f ] in any K-basis for L, from which the
additivity ensues, and it also ensues that the trace of the identity equals the dimension
of L over K.
For the third feature observe that if Pf (t) = tn + an´1 tn´1 + . . . + a1 t + (´1)n N ( f )
is the characteristic polynomial of [ f ], the Cayley–Hamilton theorem yields

N ( f ) = (´1)n+1 ¨ f ¨ ( f n´1 + an´1 f n´2 + . . . + a1 ). (11.2)

and because the ai ’s lie in A and f lie in B, the expression in parenthesis lies in B. o

Separable and unseparable

In the literature a polynomial is said to be separable if it has distinct roots in an algebraic
closure K̄. In a rather vague etymology the usage is some times explained by that the
roots being distinct they can be separated.
There is standard way of detecting multiple roots of a polynomial is by use of the
derivative. A root a of f (t) is not simple precisely when the derivative f 1 (t) vanishes
at a as well. Indeed, one may write f (t) = (t ´ a)m g(t) where g( a) ‰ 0. Leibnitz’ rule
then yields
f 1 ( t ) = m ( t ´ a ) m ´1 + ( t ´ a ) m g 1 ( t ) ,
and we see that f 1 ( a) = 0 if and only if m ě 2.
Obviously, a non-constant irreducible polynomial f (t) cannot share a common factor
with a (non-zero) polynomial of lower degree, so in particular if a were a multiple

10th January 2021 at 11:32am

Version 4.0 run 30
 dimension of polynomial rings 311

root, ( x ´ a) would be a common factor of f and f 1 , which would force f 1 to vanish

identically.
Over fields of characteristic zero the only polynomials with vanishing derivative are
the constants, so in that case an irreducible f has distinct roots and is hence separable.
However, when K has characteristic p, this is not true any more. For instance, the power
t p has derivative pt p´1 which vanishes since p = 0 in K. More generally one finds
applying the chain rule that
ν ν v ´1
( f (t p ))1 = f 1 (t p ) ¨ pν t p = 0.
m
Hence polynomials shaped like f (t p ) have vanishing derivatives. And in fact these are
all.

Lemma 11.52 Let f (t) be a polynomial in K [t] and assume that f 1 (t) is the zero polynomial.
ν
Then K is of positive characteristic, say p, and f (t) = g(t p ) for some g P K [t] whose derivative
is not identical zero, and ν P N0 . If f is irreducible, g will be separable.

Proof: Write f (t) = iP I ai ti where I Ď N are the indices with ai ‰ 0. Then f 1 (t) =
ř
i´1 , and since powers of t are linearly independent, it follows that ia = 0
ř
1ď i ď n i ¨ a i t i
for all i P I. Hence i = 0, that is i is divisible by p and we can write i = pνi mi . Letting ν
ν ´ν v
be the smaller of the νi ’s and g(t) = ai t p i mi , we find f (t) = g(t p ). The exponent
ř

of the term in g(t) with index i so that νi = ν, is not divisible by p, and hence g1 is not
identically zero. Finally, if g has a double root, it can not be irreducible, since g1 ‰ 0,
hence neither f can be irreducible. o

Proposition 11.53 The trace form tr L/K xy is non-degenerate if and only if all elements in L
are separable over K.

Proof: By the Primitive Element Theorem there is an element a such that L = K ( a). Let
Q(t) be the minimal polynomial so that L = K [t]/( Q(t)). Separability means that the
roots of Q(t) in an algebraic closure K̄ are distinct (they can be “separated”). It follows
that ( Q(t)) = (t ´ β 1 ) X . . . X (t ´ bn ) and The Chinese Remainder theorem gives an
isomorphism a
ź
LbK K̄ = K̄ [t]/( Q(t)) » K̄.
i

Now, the multiplication map [ f ] is a K̄ map of LbK K̄, and a basis tei u induces a basis
ei b1 of LbK K̄, so and the matrix of [ f ] in the two are of course equal, and the trace
tr L/K bid = tr LbK K̄/K̄ is same whether f is considered a map of K or of LbK K̄.
But the Chinese basis shows that blabla. . . . . . . o

10th January 2021 at 11:32am

Version 4.0 run 30
312 krull dimension

Appendix (temporary): A geometric version of the Principal Ideal Theorem

We proceed to offer the promised alternative proof of the Hauptidealsatz valid for
algebras finite type over fields, which David Mumford in his red book[?] on varieties
contributes to John Tate. It relies on the Normalization lemma and uses the norm, so
we begin the section with recalling a few properties of the norm (for details see the
Appendix) together with an easy lemma about the transcendence degree.
(11.54) For any finite extension A Ď B of domains with A integrally closed in its fraction
field, one has the multiplicative norm map N : B˚ Ñ A˚ . If K and L designate the
fraction fields of respectively A and B, the norm N ( x ) of an element x P B is the
determinant of the K-linear map [ x ] : L Ñ L that is just the multiplication by x; i. e. it
is the map that sends y to xy. We shall need three of basic properties of the norm all
proved in detail in the Appendix.
i) The norm is multiplicative: N ( xy) = N ( x ) N (y);
ii) For elements x P A it holds that N ( x ) = x [ L:K ] ;
iii) The element x is a factor of N ( x ); that is, N ( x ) = yx for some y P B.
The two first are well known properties of the determinant, and the last is a consequence
of the Cayley-Hamilton theorem asserting that a linear map satisfies its characteristic
polynomial.
(11.55) The crux of Tate’s proof of the Principal Ideal Theorem is the following lemma:
‘ ‘

Lemma 11.56 In the setting just described, it holds true that (x) X A = N ( x ) for any
element x P B.

Proof: In view of x being a factor of N ( x ), one inclusion is obvious, namely that

‘
‘ ‘
N ( x ) Ď ( x ) X A. For the reverse inclusion, assume that y P ( x ) X A; that is,
y P A is an element on the form yr = bx for some b P B and some number r. It then
holds true that ydr = N (yr ) = N (b) N ( x ), where d = [ L : K ], and hence y belongs to
‘

N (x) . o
With theses preparations in place, the proof of the Principal Ideal Theorem is, after an
initial reduction, reduced to a few lines, but of course, the burden of the proof is bore
by the Normalization Lemma.

Theorem 11.57 (Geometric Principal Ideal Theorem) Let k be a field and A a domain
finitely generated k. If f P A is non-zero element and p is a minimal prime ideal over ( f ), then
trdegk A/p = trdegk A ´ 1.

To ease the notation we have written trdegk A for the transcendence degree of the
fraction field of a domain A. In view of Corollary 13.20 above, the conclusion might
as well have been formulated as dim A/p = dim A ´ 1. Notice, that the conclusion

10th January 2021 at 11:32am

Version 4.0 run 30
 dimension of polynomial rings 313

is significantly stronger than just saying ht p = 1, and as we shortly shall see, almost
effortlessly leads to A being catenary.
‘
Proof: The first part of the proof is a standard reduction to the case that ( f ) is a
‘
prime ideal. We let p1 , . . . , ps be the minimal primes of ( f ), so that ( f ) = p1 X . . . X ps ,
and we may as well assume that p1 is the particular one we study. Pick an element h
‘
from A that lies in p1 , but not in any of the others pi ’s. Then ( f ) Ah = p1 Ah is a prime


ideal. Now, Ah /p1 Ah = A/p1 h . Hence A/p1 and Ah /p1 Ah have the same fraction
field, and of course, Ah is finitely generated over k. We may thus replace A by Ah and
‘
assume that p = ( f ).
Let n = trdegk A. By Noether’s Normalization Lemma, there are algebraically
independent elements x1 , . . . , xn from A such that A is a finite module over the poly-
nomial ring R = k[ x1 , . . . , xn ], and then the norm map N : A˚ Ñ R˚ is available. By
‘ ‘
‘

Lemma 11.56 it holds true that p X R = ( f ) X R = N ( f ) . Now, N ( f ) is a
height one prime ideal in a polynomial ring and is therefore principal, say generated
by g. It follows that A/p is finite over R/( g) and hence has the same transcendence
degree, and by Lemma 13.24 above the latter has transcendence degree n ´ 1, o

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 12

Integral extensions

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 all the other non-zero elements, they are all units. When
the exploration of number fields* began in the early 19th century, an immediate want ˚ That is, finite field

arose of subrings playing the role of the integers, and in which the deeper secrets of extensions of the
rationals.
the field could be revealed. These rings were made up of the “integral elements” in the
field, or more precisely what we soon shall be calling the elements integral over Z.
Integral extension are ubiquitous; they are found not only in number theory, but
where ever commutative algebra is seriously used. In algebraic geometry, for instance,
integrally closed (or normal) 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.
In topology one has the notion of “branched covering spaces”; that is, continuous
maps X Ñ Y between two topological spaces which are proper* and have discrete ˚ That a map is proper

fibres (in particular they could be finite). The integral extensions are in some sense means that inverse
images of compact sets
the algebraic counterpart of these, in that the map they induce between the spectra are compact
will have closely resembling properties, as expounded in the circle of ideas round the
Cohen–Krull–Seidenberg theorems.

12.1 Definition and basic properties

Throughout this section we shall work with an extension of rings A Ď B. An element
x P B is said to integral over A if it satisfies a monic relation Integral elements (hele
elementer)
x n + an´1 x n´1 + . . . + a1 x + a0 = 0, (12.1)
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, but of course, the leading coefficient
 316 integral extensions

Integral dependence being invertible in A goes for the same. This distinguishes integral elements from their
relation cousins, the algebraic elements, which satisfy similar equations, but with no constraint
(helhetsrelasjon)
on the leadings coefficient. A relation like (12.1) is called an integral dependence relation
for x over A.
(12.1) If all elements in B are integral over A, one says that B is integral over A, or that
Integral extension (hel
utvidelse) B is an integral extension of A. The subset of B consisting of the elements integral over A
Integral closure is called the integral closure of A in B and is denotes by A. s Of course it depends on B,
(helavslutningen) but to keep notation simple, we do not 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 slightly
subtle fact shortly to be proven, that the integral closure is a ring. Finally, one says that
Integrally closed A is integrally closed in B if A = A;s that is, if every element in B which is integral over
(helavsluttet) A, belongs to A.
(12.2) Both in algebraic geometry and algebraic number theory the integral closure of
a domain A in its field of fractions is an important associate to the domain, and we’ll
Normal rings (normale denote it by A r to distinguish it from all the crowd of integral closures. Domains being
ringer) integrally closed in their field of fractions; that is, those satisfying A = A, r are called
Normalization normal, and for a general domain A is sometimes called the normalization of A.
r
(normalisering)
Examples
Integral closures and normalizations play an important role in algebraic geometry,
which is particularly accentuated in the theory of curves. We’ll illustrate this with the
two simplest examples of curve singularities, or in algebraic parlance, two non-normal
one-dimensional rings, an ordinary double point and a simple cusp. Typically for curves
their normalizations “resolve singularities”; that is, it separates different branches of
the curve passing through the multiple points (as in the case of the double point) or it
resolves the vanishing of a derivative (as for the cusp).
Integral closures are of equally great significance in number theory where the ring
of integers in number fields are the legendary stars of the theory. Our steadfast friends
the quadratic extensions will serve as examples.
(12.1) To illustrate the difference between algebraic an integral dependence relation
consider the two simple equations

y2 ´ z = 0
( z ´ 1) y2 ´ 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 domains not containing the
origin can a continuous solution be found. The solutions of the second equation suffer

10th January 2021 at 11:32am

Version 4.0 run 30
 definition and basic properties 317

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.
(12.2) An ordinary double point: We let C Ď C2 be the plane curve parameterized by
x = t2 ´ 1 and y = t(t2 ´ 1). It is easily seen to satisfy the equation y2 = x2 ( x + 1);
indeed, (y/x )2 = t = 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 de-
fined, and the parameterization 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 parameterization may be though of as the map Spec C[t] Ñ Spec C[ x, y] induced
by the ring-map C[ x, y] Ñ C[t] that sends x Ñ t2 ´ 1 and y Ñ t(t2 ´ 1); or in the
language of varieties, it is the the map C Ñ C = V (y2 ´ x2 ( x ´ 1)) Ď C2 sending t to
the point (t2 ´ 1, t(t2 ´ 1)).
This leads to considering the subring A = C[t2 ´ 1, t(t2 ´ 1)] Ď C[t]. 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 = (t2 ´ 1) + 1 P A.
Moreover, the ratio between the two generators of A equals t, so that the fraction
field of A equals C(t). And anticipating that the polynomial ring C[t] is normal (either
Proposition 12.21 or 12.23 on page 324), we can conclude that C[t] is the normalization
of A.
(12.3) The simple cusp: We also want to give an example from geometry, and the simplest Simple cusp (enkel
example of a variety with a non-normal coordinate ring is the so-called simple cusp. It is cusp eller enkel spiss)

the curve C Ď C2 whose equation is y2 = x3 .

One may parameterize C by using y/x as parameter; the parameterization being
t ÞÑ (t2 , t3 ). On the level of coordinate rings, the parameterization is reflected in
the map C[ x, y] Ñ C[t] such that x ÞÑ t2 and y ÞÑ t3 . The image is clearly the ring
A = C[t2 , t3 ] Ď C[t]. The fraction field of A equals the rational function field C(t) since
t = t2 t´3 lies there. Now, A is not integrally close in C(t); indeed, t is integral over A
being a root of the equation
T 2 ´ t2 = 0,

where we note that t2 P A.

5)/2 is integral over Z since it satisfies
‘
(12.4) The golden section: The golden section (1 + The golden section (det
the equation gylne snitt)

x2 ´ x ´ 1 = 0.

10th January 2021 at 11:32am

Version 4.0 run 30
318 integral extensions

The integral closure of the integers Z in Q( 5) equals Z[(1 + 5)/2]. Indeed, a number
‘ ‘
‘
η = a + b 5 has the minimal equation

x2 ´ 2ax + ( a2 ´ 5b2 ) = 0, (12.2)

and if η is integral over Z, the coefficients of (12.2) are integral by Gauss’s Lemma
(Exercise 3.7 on page 78). Then n = 2a P Z and 20b2 P Z, and hence m = 2b P Z as well.
Substituting back, gives 0 ” (n2 ´ 5m2 ) ” (n2 ´ m2 ) mod 4, which holds if and only if
m and n have the same parity.
(12.5) The ring of integers in the number field Q(i 5) equals Z[i 5]. Indeed, the
‘ ‘

minimal equation of an element a + ib 5 P Q(i 5) is

‘ ‘

x2 ´ 2ax + ( a2 + 5b2 ) = 0.

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.
K

Exercises
ˇ (12.1) Integers in quadratic number fields. This is a classic from elementary number theory.
The difference between the ring of integers in the two previous examples 12.4 and 12.5
illustrates a general phenomenon. Prove that if d is a square free integer, the ring of
integers in Q( d) equals Z[(1 + d)/2] when d ” 1 mod 4, and Z[ d] else.
‘ ‘ ‘

(12.2) A not so simple cusp. Let p be a natural number and consider the plane curve
y2 = x2p+1 . It may be parameterized by t ÞÑ (t2 , t2p+1 ); so the parameter is yx´ p . Let
A = C[t2 , t2p+1 ], let m be the ideal (t2 , t2p+1 ) in A, and let A
r = C[ t ].
a) Show that C[ x, y]/(y2 ´ x2p+1 ) » C[t2 , t2p+1 ] and that m is a maximal ideal;
b) Show that the fraction field of A equals the rational function field C(t) and that
A is not normal;
c) Show that A/A
r is a cyclic A module of length p supported at the maximal ideal
m.
M

The basic properties

(12.3) If x is an element of B which is integral over A, the subring A[ x ] of B obtained
by adjoining the element x to A is a finitely generated A-module. From an integral
dependence relation as in (12.1) above, ensues the identity

x n = ´ ( a n ´1 x n ´1 + . . . + a 1 x + a 0 ) ,

10th January 2021 at 11:32am

Version 4.0 run 30
 definition and basic properties 319

and a straightforward induction yields that A[ x ] is generated by the n first powers of

x. The converse of this is also true, as will be shown in the next proposition. When
the A-module A[ x ] is Noetherian, this comes almost for free; one just considers the
ascending chain Mi = A + xA + x2 A + . . . + xi A of submodules of A[ x ], 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 . The general proof requires however a little twist. Recall
that a faithful module is one whose annihilator is the zero ideal.

Proposition 12.4 (Basic characterisation) Let A Ď B be an extension of rings and let

x P B be an element. The following three statements are equivalent:
i) The element x is integral over A;
ii) A[ x ] is a finitely generated A-module;
iii) There is a faithful A[ x ]-module which is finitely generated over A.

Proof: The only implication showing any substantial resistance is that i) follows from
iii). So let M be a module as in iii), and let m1 , . . . , mn be generators for M over A. Each
element x ¨ mi may be expressed in terms of the m j ’s, and this gives relations

x ¨ mi = f i1 m1 + . . . f in mn , (12.3)

for 1 ď i ď n, where all coefficient f ij belong to A. We introduce an n ˆ n-matrix Φ by

letting Φ = x ¨ I ´ ( f ij )ij where I is the n ˆ n-identity matrix (the matrix Φ for n = 3 is
shown in a footnote)1 . Equation (12.3) above then translate into the equality Φ ¨ m = 0
where m = (m1 , . . . , mn ). Hence the determinant det Φ kills M by the determinant trick
(Lemma 4.61 on page 110), and we deduce that det Φ = 0 as M is a faithful A-module.
But developing the determinant shows that det( x ¨ I ´ ( f ij )) is a monic polynomial in x
whose coefficients lie in A; that is, det Φ = 0 is an integral dependence relation for x
over A. o
We notice the immediate corollary—which may also easily be proven ad hoc—that all
elements in A[ x ] are integral over A when x is. A faithful A[ x ] module M which is
finitely generated over A, will be faithful over any subring of A[ x ], in particular over
A[z] for any z P A[ x ].

Corollary 12.5 If x is integral over A, all elements in A[ x ] are integral over A.

(12.6) There is a close relationship between integral and finite extensions as unveiled in
the previous proposition, but there are also significant differences. Finite extensions are
integral, but in general the converse is not true. There are even examples of Noetherian
 
x ´ f 11 f 12 f 13
1 For n = 3 the matrix Φ is shaped like  f x ´ f 22 f 23 
21
f 31 f 32 x ´ f 33

10th January 2021 at 11:32am

Version 4.0 run 30
320 integral extensions

domains whose normalization A r is not a finite module over A (we shall reproduce one
in Paragraph 14.26 on page 369); 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.
(12.7) The first conclusion to be drawn from the basic characterization of integral
elements 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 12.8 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 proof 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, say x. By induction
C1 is a finitely generated module over A, and we have C = C1 [ x ]. The element x being
integral over A is even so more over C1 . Hence C is finitely generated over C1 by
Proposition 12.4, and because being a finite extension is a property transitive in towers
(Lemma 12.13 at the end of this subsection), it holds that C is finitely generated over A.
o

Corollary 12.9 (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, (12.4)

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 (12.4) has coefficients in D) and consequently
D [ x ] is a finite module over D. Now, D is a finite module over A after Proposition 12.8
above, and therefore D [ x ] is finite over A as well. Hence we can conclude by the Basic
Characterization 12.4 of integral elements that x is integral over A. o
(12.10) It is by no means obvious how to deduce a dependence relation for a product (or
for a sum) from dependence relations for the factors (or the addends); that the integral
closure is a ring, is a slightly subtle property. However, once the Basic Characterization
(Proposition 12.4 on the preceding page) is in place, it follows readily. For a different
approach, see Problem 12.7 on page 322.

Proposition 12.11 (The integral closure is a ring) Assume that A Ď B is an extension

of rings. The sum and the product of two elements from B which are integral over A, are integral

10th January 2021 at 11:32am

Version 4.0 run 30
 definition and basic properties 321

over A. 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 12.5 and the transitivity property
(Corollary 12.9). 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 A[ x, y] is integral over A[ x ].
Hence A[ x, y] is integral over A by transitivity. In particular, both the product x ¨ y and
the sum x + y being members of A[ x, y] are integral over A, and A s is a ring.
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 12.9 above tells us since A is integral over A.
s o
(12.12) We end this subsection by proving the following lemma used in proof of
Proposition 12.8 above:

Lemma 12.13 (Finite generation in towers) Let A Ď B Ď C be a tower of rings and as-
sume that C is a finite module over B and B is finite module over A, then C is a finite module
over A.

Proof: Let x1 , . . . , xr be a generating set for B as an A-module and y1 , . . . , ys one for C

ř
over B. Then the products xi y j will generated C over A. This is elementary: if z = b j y j
ř
with b j P B, write each coefficient b j as b j = j aij xi to obtain
ř ř ř ř
z= j bj y j = j ( i aij xi ) y j = i,j aij xi y j .

Integral extensions, localization and quotients

Integral extensions are well behaved in that they are compatible with the formation of
localizations and quotients.
(12.14) We treat the loacalizations first:

Proposition 12.15 Let S Ď A be a multiplicative subset and assume that B is an integral

extension of A. Then S´1 B is an integral extension of S´1 A. Forming the integral closure
commutes with localization; i. e. it holds that S
Ğ ´1 A = S´1 A.
s

Proof: Let xs´1 be an element in B with x P A and s P S. All elements of B are assumed
integral over A, so x satisfies an integral dependence relation shaped like

x n + an´1 x n´1 + . . . + a0 = 0,

with the ai ’s lying in A. Multiplying through by s´n we find the relation

( xs´1 )n + an´1 s´1 ( xs´1 )n´1 + . . . + a0 s´n = 0, (12.5)

10th January 2021 at 11:32am

Version 4.0 run 30
322 integral extensions

which is a monic equation whose coefficients lie in S´1 A and hence is an integral
dependence relation for xs´1 over S´1 A.
For the second statement, the inclusion S´1 A sĎS Ğ´1 A ensues from the first claim,

and it suffices to prove the converse. To that end, assume that xs´1 P S
Ğ ´1 A. It satisfies

an integral dependence relation as

( xs´1 )n + bn´1 ( xs´1 )n´1 + . . . + b0 = 0, (12.6)

where each bi lies in S´1 A 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 (12.6) through by sn tn yields the relation

(tx )n + an´1 s(txs´1 )n´1 + . . . + sn tn´1 a0 = 0, (12.7)

and it follows that tx is integral over A. We conclude that xs´1 = (tx )s´1 t´1 P S´1 A.
s o
(12.16) Next comes the quotients, and in addition to the usual 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, which persists being integral:

Proposition 12.17 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 chose an element y P B that maps to x. By assumption y 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 we obtain the relation

x n + [ an´1 ] x n´1 + . . . [ a1 ] x + [ a0 ] = 0,

where [ ai ] as usual denotes the classe of ai in A/a. Hence x is integral over A/a. o
Exercises
(12.3) Let B be a ring and tBi uiP I a family of subrings of B. If each Bi is integrally
Ş
closed in B, then the intersection iP I Bi is integrally closed as well.
ˇ (12.4) New. Let A Ď B be an integral extension and let x be a variable. Show that the
extension A[ x ] Ď B[ x ] is integral.
(12.5) Let A be a normal domain and L an extension of the fraction field K of A. An
element x P L is integral over A if and only if the minimal polynomial m x of x has
coefficients from A.
(12.6) Let A Ď B be two domains. Show that x P B is integral over A if and only if there
is a square matrix with coefficients in A having x as an eigenvalue.
(12.7) Show that if x and y are eigenvalues for Φ and Ψ then x ¨ y is and eigenvalue for
the Kronecker product ΦbΨ and that x + y is one for the matrix Φb Im + In bΨ where

10th January 2021 at 11:32am

Version 4.0 run 30
 definition and basic properties 323

the In are Im are identity matrices of appropriate size. Conclude that the integral closure
is a ring.
M

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 = K ( A) is the field of fractions of A. Recall that the integral
Normalizations
closure of A in K ( A) is called the normalization of A, and in case A is integrally closed (normaliseringer)
in K ( A), one says that A is a normal domain. Normal domain
(12.18) In the previous section we showed that the two operations normalization and (normalt område)

localization commute, and it ensues that A being normal implies that all localizations
Ap at prime ideals are normal. The converse also holds, so being normal is a local
property:

Proposition 12.19 (Being normal is local) Assume that A is a domain. Then A is nor-
mal if and only if Am is normal for all maximal ideals m 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, when localized at a maximal

ideal m, the sequence gives rise to the short exact sequence

0 / Am / (A
r)m / ( A/A
r )m / 0.

According to Proposition 12.15, forming integral closures commute with localization, so

it holds that ( A
r)m = ( Arm ). Thence ( A/A
r )m = ( A rm )/Am , and the claim follows by the
Localness of Being Zero (Proposition 7.47 on page 196). o
(12.20) In the friendly case that A is a finitely generated module over A, the quotient
r
A/A
r is a finitely generated module over A as well, and it ensues that Supp A/ A r is a
closed subset of Spec A; in fact, it equals V (Ann A/ A).
r
Localizing at (0) (remember, A is a domain), or equivalently, tensorizing by K, we
see that A/Ab A K = 0 because A and A both have K as fraction field. Hence A/A is
r r r
not of global support. So in that benign 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 dense subset of Spec A the
local rings Ap are normal.
Example 12.6 For example the normalization of Z[i 5] eqauls Z[(1 + i 5)/2], so in
‘ ‘

Spec Z[i 5] the set of “non-normality” is the closed set V ((2)).

‘
K

10th January 2021 at 11:32am

Version 4.0 run 30
324 integral extensions

12.2 Examples
We indulge ourselves in a few examples of rather large classes of rings that are normal.
The unique factorization domains are always normal, and the polynomial rings over
normal domains are normal. The third class we shall investigate are the rings of
invariants of actions of finite groups. It is a general principle that these rings are normal,
at least when the ring acted upon is normal. This class includes all the so-call "quotient
singularities". We shall treat a few simple examples in detail but leave the general case
to the zealous students in the form of a guided exercise.

Factorial domains
A large and versatile class of domains that are normal are the ufd’s:

Proposition 12.21 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 going for a contradiction, we assume that z R A. Reducing the
fraction if need be, we may assume that x and y are without common factors. The
element z being integral means that there is a relation

zn + an´1 zn´1 + . . . + ai zi + . . . + a0 = 0,

with the ai ’s lying in A. Multiplying through by yn and rearranging the equation, 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

Polynomial rings
Proposition 12.22 If A Ď B is an intergral extension, then the extension A[t] Ď B[t] of poly-
nomial rings is integral.
Proof: Let p P B[t], and let C Ď B be the A-subalgebra generated by the coefficients
of p; it is integral over A and hence is a finite A-module. Obviously C [t] is a finite
A[t]-module having the same generators over A[t] as C has over A, and by the Basic
Characterization (Proposition 12.4 on page 319) p is integral over A[t]. o

Proposition 12.23 The polynomial ring A[t] over a normal domain A is normal.
Proof: We let K be the fraction field of A. The polynomial ring K [t] is normal with the
same fraction field as A[t], and therefore it suffices to see that A[t] is integrally closed
in K [t]. To that end, let p(t) be a polynomial in K [t] integral over A[t], and assume that

pn + f 1 pn´1 + . . . + f n´1 p + f n = 0,

10th January 2021 at 11:32am

Version 4.0 run 30
 examples 325

where the f i ’s belong to A[t]. We shall need a root field L of p containing K. Clearly
each root of p in L is a root of f n , so if f n were monic, the roots would all be integral
over A. The coefficients of f n being polynomials in the roots would as well belong to A
since A is normal by hypothesis; and consequently if also p were monic, we would have
p P A[t]. One can achieve this favourable situation simply by replacing p by q = p ´ tr
for r ąą 0. Indeed, a simple computation gives that q = p ´ tr satisfies a relation

q n + g 1 q n ´1 + . . . + g n ´1 q + g n ,

with “constant term”

gn = tnr + q1 tr(n´1) + . . . + q1 tr + qn = 0

which is monic when r ąą 0, and hence q and thereby p, lies in A[t]. o

Rings of invariants
(12.24) Now comes the promised result on rings of invariants, and as promised, we shall
proceed in a rather relaxed way merely treating the simples possible case. That is, the
case of a cyclic group of order two acting on a normal domain B. Such an action is given
by an involution on B; in other words, by a ring map σ : B Ñ B satisfying σ2 = id[ B]. The Involutions
map σ extends to an involution of the fraction field K of B by the obvious assignment (involusjoner)

σ ( xy´1 ) = σ ( x )σ (y)´1 . Furthermore, we let A = Bσ = t x P B | σ ( x ) = x u be the ring

of invariants and L its field of fractions. In this setting we have

Proposition 12.25 It holds true that L = K σ = t z P K | σ (z) = z u. Moreover B is integral

over A and if B is normal, A will be normal.

Proof: Clearly L Ď K σ . If σ ( x )/σ (y) = x/y it holds that yσ ( x ) = xσ(y), and we may
write x/y = σ ( x ) x/σ ( x )y with both σ ( x ) x and σ ( x )y being invariant. Hence L = K σ .
Any element x P B satisfies the relation Kσ = L Ď K

Ď
x2 ´ (σ ( x ) + x ) x + σ ( x ) x = 0. (12.8) Bσ = A Ď B

Both σ ( x ) + x and xσ ( x ) are invariant under σ and therefore belong to A, hence (12.8)
is an integral dependence relation for x over A.
Finally, as B is integral over A, the integral closure of A in K equals Br by transitivity,
and hence A r=B r X L, from which ensues that A = B X L = A r in the case that B = B.
r o

Example 12.7 The quadratic cone again: The coordinate ring A = k[ x, y, z]/( xy ´ z2 ) of
the quadratic cone is not factorial, as we have seen, but it is normal. It thus gives an
example that the converse of Proposition 12.21 on the facing page is not valid. We shall
exhibit A as the ring of invariants of the Z/2Z-action on k[u, v] given by sign change of

10th January 2021 at 11:32am

Version 4.0 run 30
326 integral extensions

both variables; a method that only works when the characteristic of k is different from
two (even though the result remains true).
Sending x ÞÑ u2 , y ÞÑ v2 and z ÞÑ uv induces a ring homomorphism from A onto
the subring k[u2 , uv, v2 ] of the polynomial ring k[u, v], and this map is an isomorphism:
When x, y and z are given the weight two, A will be a graded algebra, and the map will
be homogenous of degree zero. The kernel is therefore a homogeneous ideal. Assume
that P is a homogeneous element in the kernel. Replacing z2 by xy, one may write
P = Q( x, y) + R( x, y)z, and since powers of u and v occurring in Q(u2 , v2 ) are even
whereas those occurring in R(u2 , v2 )uv are odd, it follows that both Q( x, y) and R( x, y)z
lie in the kernel. Now, any homogeneous form in two variables splits as a product
of linear forms, so if the kernel were non-zero, it would contain a linear form. But
since u2 , uv and v2 are linearly independent, this is not the case, and we can conclude
that the map is injective. Obviously, it is surjective, hence it is an isomorphism. So
define an action of the group Z/2Z on k[u, v] by letting the generator σ act by u ÞÑ ´u
and v ÞÑ ´v. Then k[u2 , uv, v2 ] will be the ring of invariants: Indeed, for a polynomial
p(u, v) = aij ui v j one has
ř

ÿ ÿ
p(´u, ´v) = aij ui v j ´ aij ui v j ,
i + j even i + j odd

and this equals p(u, v) precisely when all terms with i + j odd vanish. Hence if aij ‰ 0,
either both i and j are even, say i = 2ν and j = 2µ, and ui v j = (u2 )ν (v2 )µ , or both are
odd, in which case ui v j = (u2 )ν (v2 )µ uv with i = 2ν + 1 and j = 2µ + 1. K

Exercises
ˇ (12.8) Invariants under finite groups. This exercise is a continuation of Proposition 12.25
above. Let the finite group G act on the domain B and let A denote the ring of invariants;
that is, A = BG = t x P B | g( x ) = g for all g P G u. Let K be the fraction field of B and
L that of A.
a) Show that action of G on B extends in a unique way to an action on K.
b) Assume x and y are two elements from B with y ‰ 0 and such that g( xy´1 ) =
xy´1 for all g P G. Show that y g‰e g( x ) is invariant under G. Conclude
ś

that L is the field of invariants in K; that is, L = K G = t xy´1 | g( xy´1 ) =

´1
xy´1 for all g P G u. Hint: x/y = x g‰e g( x ) y g‰e g( x )
ś
ś
.
c) Show that B is integral over A. Hint: Show that any symmetric polynomial in
ś
the g( x )’s is invariant and use that gPG ( x ´ g( x )) = 0.
d) Show that A is normal whenever B is.
ˇ (12.9) New:Transitive action on fibres. Keeping the notation from the previous exercise,
show that G acts transitively on the fibre over a prime ideal p P Spec A; that is, it acts

10th January 2021 at 11:32am

Version 4.0 run 30
 the cohen–krull–seidenberg theorems 327

transitively on the set t q P Spec B | q X A = p u.

(12.10) The cone over rational normal curves. Let ui vd´i with 0 ď i ď d be a basis for the
monomials of degree d in C[u, v]. Show that the ring C[ui vd´i |0 ď i ď d] is normal.
Hint: Let µd be the group of d-th roots of unity and let ζ P µd act on C[u, v] by u ÞÑ ζu
and v ÞÑ ζv.
M

12.3 The Cohen–Krull–Seidenberg Theorems

There is cluster of results proven in full generality by I. S. Cohen and A. Seidenberg and
dubbed “Going–Up”, “Going–Down” and “Lying–Over” by them. These results were
all first found by Krull, whose proofs, however, were valid only for integral domains.
The results 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 p = A X q (one says q lies over p or contracts to p) and as we know,
this sets up the corresponding map π : Spec B Ñ Spec A between the spectra. The
Cohen–Seidenberg theorems are basically results about this map, approaching questions
like when is it surjective, and what are the fibres? What about chains of prime ideals, Abraham Seidenberg
can they extended? Is it a close map? Or an open map? (1916–1988)

(12.26) Notice that B is just assumed to be integral over A and is not necessarily a American

finitely generated A-module. So for example, the highly infinite extension Z Ď Z where mathematician

Z denotes the integral closure of Z in the field of algebraic integers Q will satisfies the
hypothesis.

A basic lemma—the case of fields

(12.27) This pivotal lemma treats the special case of fields:

Lemma 12.28 Let A Ď B be an integral extension of domains. If one of the rings is a field the
other one is a field as well.
Proof: Assume first that B is a field. If y P A is a non-zero element, the inverse y´1 is
integral over A and satisfies a dependence relation

y´n + an´1 y´(n´1) + . . . + a1 y´1 + a0 = 0,

with the ai ’s being elements from A. Multiplying through by yn gives

1 + y ( a n ´1 + . . . + a 1 y n ´2 + a 0 y n ´2 ) = 0

which shows that y is invertible in A. Next assume that A is a field, and let x P B be a
given non-zero element. It satisfies a relation

x n + an´1 x n´1 + . . . + a1 x + a0 = 0,

10th January 2021 at 11:32am

Version 4.0 run 30
 328 integral extensions

with coefficients ai from A, and assuming that the degree n is minimal, it holds that
a0 ‰ 0. Then a0 will be invertible, and we have

1 n ´1
x ¨ a´
0 (x + a1 x n´2 + . . . + a1 ) + 1 = 0.

Corollary 12.29 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 12.17 on page 322, 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 Lying-over The first theorem of the cluster—the Lying-Over Theorem—describes the structure of the
Theorem fibres of the map π; that is, a qualitative description of the set of prime ideals in B
(Lying–over-teoremet)
intersecting A in a given fixed prime ideal. The fibres are non-empty; in other words,
all prime ideals p in A are of the form p = q X A, and there are no inclusion relations
between members of a fibre.
(12.30) Here it comes:

Proposition 12.31 (Lying–Over) Assume that A Ď B is an integral extension. For each

prime ideal p Ď A there is at least one prime ideal 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 .

In geometric terms the Lying–Over theorem asserts that the induced map π : Spec A Ñ
Spec B between the spectra is surjective and has discrete fibres, additionally it will
also be a closed map; i. e. images of closed sets are closed, as Proposition 12.11 below
shows.
Proof: We begin with treating the local case, and subsequently we’ll reduce the general
situation to that case by localizing.
Assume then that A is local with maximal ideal m. Let n be any maximal ideal
in B; there are such according to the The Fundamental Existence Theorem for Ideals
(Theorem 2.49 on page 49). By Corollary 12.29 above the contraction n X A is maximal,
hence equal to m since m is the only maximal ideal in A.
To see that no inclusion relations holds among ideals in a fibre of π, assume that
q Ď q1 are two primes in B both intersecting A in m (we are still in the local situation).
Again by Corollary 12.29 both q and q1 are maximal and must consequently be equal as
one is contained in the other.

10th January 2021 at 11:32am

Version 4.0 run 30
 the cohen–krull–seidenberg theorems 329

Let then p be a prime ideal in A. To reduce to a local situation we pass to the

localized extension Ap Ď Bp , which persists being integral in view of Proposition 12.15.
Lying–Over being true in the local case, implies there is a prime ideal in Bp , which as all
prime ideals in Bp is of the form qBp with q a prime ideal in B, such that qBp X Ap = pAp .
But then the equality q X A = p follows e. g. because

q X A = qB X A = ( Ap X qBq ) X A = pAp X A = p.

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

Proposition 12.32 If A Ď B is an integral extension, the map π : Spec B Ñ Spec A induced

between the spectra will be a closed map

Proof: Let a Ď B be an ideal. We shall show that π (V (a)) = V (a X A). To that end, we
apply Lying–Over to the inclusion A/a X A Ď B/a and conclude that every p P V (a X A)
is of the form q X A with q P V (a). The other inclusion being trivial, we are through. o
Example 12.8 It might well happen that π is surjective and has finite fibres without B
being integral over A. A cheap example can be the extension

A = k[ x2 ] Ď k [ x, ( x ´ 1)´1 ] = B

where we assume that k is algebraically closed and not of characteristic two. The
geometric interpretation is the parabola C given as y = x2 with a hole punched in it:
the point (1, 1) is removed. The map π is just projection Czt(1, 1)u onto the y-axis.
The ring k[ x2 ] is isomorphic to the polynomial ring k[y] (rebaptize x2 to y). Every y
maximal ideal m in A is of therefore the form x2 ´ a2 (all elements in k have a square-
root), and it holds true that ( x ´ a) B X A = ( x2 ´ a2 ) A since x2 ´ a2 = ( x + a)( x ´ a) in
B. However, ( x ´ 1)´1 is not integral over A; indeed, if it were, multiplying an integral
dependence relation of degree n by ( x ´ 1)n , would have given a relation

1 + pn´1 ( x ´ 1) + . . . + p1 ( x ´ 1)n´1 + p0 ( x ´ 1)n = 0,

x
where the coefficients pi ’s are elements on k[ x2 ]. Putting x = 1 gives an obvious
contradiction. As of the fibres of p, a nice exercise would be to check that if a ‰ 1, the
two prime ideals ( x ´ a) B and ( x + a) B are the ones lying over ( x2 ´ a2 ) A, but the sole
prime ideal lying over ( x ´ 1) A is ( x + 1) B. Figuring out what the fibres are when k is
of characteristic two would as well be instructive. K

Going–Up
The Going–Up Theorem is about extending, or lifting as one also says, ascending chains The Going–Up
of prime deals in A to chains in B by climbing them—the lifted chain ascends from a Theorem (Going–Up
teoremet)

10th January 2021 at 11:32am

Version 4.0 run 30
 330 integral extensions

given extension of the smallest prime ideal in the chain in A—which is contrary to the
Going–Down Theorem where chains are lifted by a downwards “movement”. If every
one-step chain in A may be lifted, an easy induction ensures that every finite ascending
q0 Ď q1 Ď B chain may be lifted.
Ď

(12.33) The one step case is what is usually called the Going–Up Theorem:
p0 Ď p1 Ď A
Theorem 12.34 (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 Proposition 12.17. 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 12.35 (Going–Up II) Assume that A Ď B is an integral extension and that q0 Ď B
is a prime ideal. Let p0 = q0 X A. Any saturated chain tpi u of prime ideals in A ascending from
p0 lifts to a saturated chain tqi uof prime ideals in B ascending from q0 .
Proof: The proof goes by induction on the number of prime ideals in the chain in A,
and one should find the proof completely transparent pondering the following display:

q0 Ă q1 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.
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 same the fibre. o
Exercises
ˇ (12.11) New. Assume that A Ď B is an integral ring extension. Show that J ( B) X A =
J ( A ).
ˇ (12.12) Let A Ď B be an extension of rings which is purely inseparable; that is, A is of
ν
characteristic p and for each element x P B there is an exponent pν so that x p P A.
Show that the induced map π : Spec B Ñ Spec A is a homeomorphism.
M
Going–Down
The Going–Up Theorem asserts that the larger of two prime-ideals has an extension
when the smaller one has one as well, in the Going–Down Theorem the order is reversed,

10th January 2021 at 11:32am

Version 4.0 run 30
 the cohen–krull–seidenberg theorems 331

if the larger can be lifted the smaller can be lifted and in such a way that lifted ideal
is contained in the lifting of the larger. Contrary to Going–Up Going–Down does
not hold for general integral extensions, the smaller ring A must be normal, and the
larger B must be a faithful A-module; most frequently one finds formulations with the
weaker requirement that B be a domain. We shall not go into the proof of the general
Going–Down, but confine ourself to state the theorem and give a proof in the special
case that the extension is finite and separable (the inseparable case is treated in an
exercise). However, there is a simple and nice example explaining why the smaller ring
must be normal, which we can not resist giving.
(12.36) As indicated we content ourself to formulate the general Going–Down theorem,
here is the core statement:
q1 Ď q0 Ď B

Ď
Theorem 12.37 (Going–Down) Let A Ď B be an integral extension of integral domains and
assume that A is normal. Given prime ideals p1 Ď p0 in A and q0 in B lying over p0 . Then there p1 Ď p0 Ď A
exists a prime ideal q1 in B contained in q0 and lying over p1 .

By a repeated application of the theorem one readily shows that every descending chain
of prime ideals in A can be lifted to one in B which extends downwards from a given
lifting of the top member of the chain in A.

Corollary 12.38 Let A Ď B be an integral extension of integral domains and assume that A
is normal. Let pn Ď . . . Ď p0 be a chain of prime ideals in A and let q0 be a prime ideal in B
lying over p0 . Then there exists a chain qn Ď . . . Ď q0 in B such that qi X A = pi .

(12.39) As promised, we shall provide a simple proof when B is the integral closure of
A in a finite and separable extension of the fraction field of A. Notably, in characteristic
zero any finite extension is separable so many cases met in practical work is covered
by this version (if you wonder about the inseparable case, do Exercise 12.13 below ).
The proof relies on Exercises 12.8 and 12.9 about group actions (for which solutions are
provided).
Proof of Going-Down in the finite and separable case: We assume first that we are
in the “Galois-situation” where there is a finite group acting on B such that A = BG .
By the Lying–Over Theorem there is a prime q11 lying over p1 (which not necessarily is
contained in q0 ). However, by Going–Up it is contained in a prime ideal q10 that lies over
p0 . Now, by Exercise 12.9 the group G acts transitively on the fibres, so there is a g P G
such that q0 = g(q10 ). Then g(q11 ) is our man.
In the general situation when B is the integral closure of A in a finite and separable
extension L of the fraction field K of A, there is an extension E of L which is Galois over
K, say with Galois group G. Then if C denotes the integral closure of A in E, it holds

10th January 2021 at 11:32am

Version 4.0 run 30
332 integral extensions

that A = C G . Thus by the beginning of the proof Going–Down holds for the extension
A Ď C, an a fortriori for A Ď B since A Ď B Ď C. o
Exercise 12.13 The unseparable case. Extend the above proof to the case where A Ď B is
merely assumed to be finite; that is, it is not necessarily separable (but still A and B are
domains, and A is still integrally closed in its fraction field). Hint: With the notation
of the proof: let E be a finite extension of L normal over K that contains E. If G is the
Galois group of L over K, then C G is a purely inseparable extension of A. Then use
Exercise 12.12 on page 330. M

z z=t

(´1,1) (1,1)

t=´1 t =1
Spec k[z,t]
’

Example 12.9 The example is built on one of the simplest non-normal rings, namely the
coordinate ring of the so-called ordinary double point; the plane curve C with equation
y2 = x2 ( x + 1). The curve C was already studied in Example 12.2 on page 317 where it
was parameterized by the assignments x = (t2 ´ 1) and y = t(t2 ´ 1). To fix the ideas,
we shall work over the complex numbers although things go through over any field
whose characteristic is not two.
The variety we have in mind is the cylinder D over C with the z-axis as generator; so
it is given by the same equation y2 = x2 ( x + 1), but in the three-dimensional space C3 .
The coordinate ring A of D equals C[ x, y, z] with constituting relation y2 ´ x2 ( x + 1) = 0.
This ring also equals the subring C[(t2 ´ 1), t(t2 ´ 1), z] of the polynomial ring C[t, z];
the parameterization tells us that.
A heuristic description of the geometry is as follows: The map ψ : C2 Ñ C3 that
sends (t, z) to (t2 ´ 1, t(t2 ´ 1), z) is bijective onto D except that both (1, z) and (´1, z)
are mapped to (0, 0, z); so the two lines in C2 defined respectively by t = 1 and t = ´1,
are both sent to the z-axis.
Consider the line L in the parameter plane C2 whose equation is z = t. It passes by
P1 = (1, 1), but not by P2 = (´1, 1). The crucial point is that an irreducible curve in C2
with the same image as L must coincide with L where ψ is injective; that is, off the two
lines t = ˘1, and hence is must equal L. Now Q = (0, 0, 1) P ψ( L), and P2 maps to Q,

10th January 2021 at 11:32am

Version 4.0 run 30
 the cohen–krull–seidenberg theorems 333

but no irreducible curve mapping to ψ( L) passes by P2 . It is straightforward to check

that the image ψ( L) is given in C3 by the two equations z2 = x + 1 and y2 = xz.
The algebraic (and precise) picture is as follows: There are two maximal ideals in
C[t, z] lying over m = ( x, y, z ´ 1) (which is the ideal of Q), namely (t ´ 1, z ´ 1) and
(t + 1, z ´ 1) (which are the ideals of P1 and P2 respectively); indeed, one easily finds

m ¨ C[t, z] = (t2 ´ 1, t(t2 ´ 1), z ´ 1) = ((t ´ 1)(t + 1), z ´ 1) = (t ´ 1, z ´ 1) X (t + 1, z ´ 1).

The ideal p = (z2 ´ ( x + 1), y ´ zx ) (which is the ideal in A of ψ( L)) is prime , since the
map C[ x, y, z] Ñ C[ x, z] that sends y to xz and lets x and z be untouched, transforms
it to the prime ideal (z2 ´ x ´ 1). One has p Ď m. We contend that (z ´ t) X A = p and
that z ´ t is the only prime ideal in C[t, z] extending p. Indeed, in C[t, z] one finds the
primary decomposition

p ¨ C[t, z] = (z2 ´ t2 , (t2 ´ 1)(z ´ t)) = (z ´ t) X (z ´ 1, t + 1) X (z + 1, t ´ 1),

from which the claim follows. (The curve ψ( L) hits the z-axis in the two points (0, 0, 1)
and (0, 0, ´1), which explains the occurrence of the component (z + 1, t ´ 1).)
Now the point is that the extension (t + 1, z ´ 1) of m does not contain the only
extension (z ´ t) of p.
K

Consequences for dimension and height

A most useful consequence of the Going–Up Theorem is that any chain in an integral ˚ Remember that

extension B of a ring A, when intersected with A becomes a chain* in A; this ensures inclusions in chains are
supposed to be strict.
that the dimension is preserved in integral extensions: We have

Proposition 12.40 (Going–Up III) If A Ď B is an integral extension of rings, then dim A =

dim B.

Proof: This is a direct consequence of the Going-Up theorems. As formulated in

Going–Up II (Corollary 12.35 on page 330) every chain in A has a chain in B lying over
it, which means that dim A ď dim B. On the other hand, Lying–Over (Proposition 12.31
on page 328) entails that each saturated chain in B remains saturated when intersected
with A. Hence dim B ď dim A. o
(12.41) 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 = Q( d)
‘

we have seen, but also to the more impressive extension K = Q, the field of algebraic
numbers. The ring of algebraic integers Z is therefore of dimension one, but recall, it is
not a Noetherian ring.

10th January 2021 at 11:32am

Version 4.0 run 30
334 integral extensions

(12.42) An easy corollary of the Going–Down Theorem may be expressed by the slogan:
“height is preserved in integral extensions where Going–Down holds”. The precise
statement is as follows:

Corollary 12.43 Assume that A Ď B is an integral extension of integral domains with A

normal. If q is a prime in B and p = q X A, then ht A p = htB q.

Proof: Each strictly decreasing chain tqi u descending from q intersects A in a chain
descending from p which is strictly decreasing by Lying-Over. Hence ht p ě ht q. Each
strictly decreasing chain of prime ideals tpi u descending from p lifts according to Going-
Down to a strictly decreasing chain of prime ideals descending from q, so the inequality
ht p ď ht q holds as well. o

12.4 Then finiteness issue

Domains that are integrally closed in their fraction fields, have several very good
properties which makes them easier to work with, and a very natural technique is to
relate a domain, or a variety, to its normalization.
It suffices to mention the particular case of one-dimensional domains finitely gener-
ated over field; they are precisely the coordinate rings of the affine non-singular curves,
so normalizing a curve is a way to desingularize it; that is, to exhibit a non-singular
curve with the same function field—a so-called non-singular model of the curve. In the
general case, for varieties of higher dimension, the normalizations are not necessarily
non-singular anymore, but still they have lots of good properties.
One dimensional normal domains finite over the integers Z, are the classical Dede-
kind rings for which the extended fundamental theorem of arithmetic is valid (and
which we shall come back to). One desires that any “number domain” can be embedded
in a Dedekind ring A, with the two differing only at finite many places, and moreover,
finite extensions of the number field gives rise, passing to the integral closure, to
extensions of the rings of integers.
It is thus desirable that the domains one works with have an integral closure in their
fraction field, or in a finite extension of the fraction field, that is a finite module. But,
alas, this is not even true for Noetherian domains, (the first examples were found by
Nagata). For domains finitely generated over fields, however, it holds true, and this will
be sufficient for many geometric applications.
We intend to give a very short account of this story—without complete proofs—but
hopefully indicating enough to give the reader a feeling of the inherent mechanisms of
the stuff. The trace will be a valuable tool, and we have included a short (and incomplete)
description of its most important properties in an appendix

10th January 2021 at 11:32am

Version 4.0 run 30
 then finiteness issue 335

The first finiteness theorem

(12.44) We attack the first finiteness theorem:

Theorem 12.45 Let A be a normal Noetherian domain with fraction field K and let L be a finite
separable field extension of K. Then the integral closure A
s of A in L is a finite A-module.

Proof: Let x1 , . . . , xn be elements in A s which form a K-basis for L. After Proposi-

tion 12.50 above we know that the trace gives a non-degenerate quadratic form tr L/K xy
on the field L, and the basis x1 , . . . , xn has a dual basis y1 , . . . , yn with respect to this
form; that is, the yi ’s satisfy the relations tr xi y j = δij . We contend that

s Ď Ay1 + . . . + Ayn ,
A (12.9)

and consequently it will hold that A s is finite over A: indeed, the right hand sum in (12.9)
is Noetherian being finitely generated over the Noetherian ring A, so any submodule
will be finitely generated over A.
To establish the inclusion in (12.9) we express any member a of A s in the dual basis
as a = a1 y1 + . . . + an yn . Multiplying by xi and taking traces we find that ai = tr L/K xi a .
But xi a belongs to A,s and as the trace of an integral element is integral, tr L/K xi a belongs
to A as A is normal. o

The second finiteness theorem

(12.46) Combined with Emmy Noether’s normalization lemma, the First Finiteness
Theorem gives as a corollary the result—important for algebraic geometers—that
normalizations of domains finitely generated over a field are finite modules over the
domain. When the ground field is of positive characteristic there is a separability issue
which we’ll unscrupulously sweep under the carpet. This shouldn’t bother you if you
primary are interested in algebraic geometry over fields of characteristic zero; e. g. over
the complex numbers, but if unresolved issues in positive characteristic disturb your
sleep, you should complete exercise xxx ( not yet written, sorry).

Theorem 12.47 Let A be a domain which is a finitely generated k-algebra and let K be its field
of fractions. Then the normalization A,
r that is integral closure of A in K, is a finite A-module.

Proof: According to the normalization lemma we may find algebraically independent

elements w1 , . . . , wn in A so that A is a finite module over the polynomial ring B =
k [w1 , . . . , wn ]. Thence the fraction field K is finite over L = k (w1 , . . . , wn ), being a
polynomial ring B is integrally closed in L, and A r equals the integral closure of B
in K. In view of Theorem 12.45 above, this finishes the proof in the case that k is of
characteristic zero, since then the extension K of L is automatically separable. And as
warned, the inseparable case is done by under-the-carpet-sweeping. o

10th January 2021 at 11:32am

Version 4.0 run 30
 336 integral extensions

Exercises
(12.14) Let A be a normal domain containing the rationals Q. Assume that A Ď B is an
extension of domains and that B is finite over A. Show that A is a direct summand in B.
Hint: Use the trace tr L/K with L and K the fraction fields of respectively A and B .
(12.15) Let A = k[ x0 , x1 , x2 , . . .] = k[ xi |i P N] where k is is a field whose characteristic is
not 2, and let B = k[ xi x j |i, j P N]. Let further K and L be the fraction fields respectively
of A and B.
a) Show that A is not a finite B-module.
b) Show that L is finite and separable over K, and that [ L : K ] = 2. In fact, the
extension is Galois with group Z/2Z.
c) Show that A is normal, and conclude that A is the integral closure of B in K.
M

12.5 Appendix: Trace and separability

(12.48) Recall that any endomorphism θ of a finite dimensional vector space over a field
The trace of an k has a trace, denoted tr θ. It is defined as the the negative of the subleading coefficient
endomorphism (sporet of the characteristic polynomial of θ; that is, by the formula
til en endomorfi)

det(t ¨ id ´θ ) = tn ´ (tr θ ) ¨ tn´1 + . . . + (´1)n det θ.

The characteristic polynomial splits into linear factors in the algebraic closure of k—the
roots are the eigenvalues or characteristic roots of θ—and the trace equals the sum of these
roots. By developing the determinant one sees that the trace equals the sum of the
diagonal elements in any matrix representing θ. This shows that the dependence on θ is
linear; that is, tr( aθ + a1 θ 1 ) = a tr θ + a1 tr θ 1 .
(12.49) In the staging of this section, where K Ď L is a finite field extension, each element
x P L has a trace tr L/K x, which equals the trace of the endomorphism [ x ] : L Ñ L that
sends y to xy.
The trace gives rise to a K-bilinear form on L, namely the form tr L/K xy, which is
The trace form called the trace form. It is obviously symmetric and K-linear since the trace is. In the
(sporformen) present context the following property of the trace form is all important, but the proof
relies on properties which are relegated to an appendix.

Proposition 12.50 (Trace and separability) A finite extension L of the field K is separable
if and only if the trace form is non-degenerate.

Proof: The main observation is that the trace form is either identically equal to zero or
non-degenerate since if tr L/K x0 ‰ 0, it follows that tr L/K y(y´1 x0 ) ‰ 0 for any non-zero
y P L.

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: trace and separability 337

To begin with suppose that the extension K Ď L is inseparable so there is an element

x P L not in K with x p = a P K. One of the properties of the trace not yet established,
but which we anyhow shall use, is the functoriallity in towers: it holds that tr L/K =
trK ( x)/K ˝ tr L/K ( x) . It thus suffices to see that trK ( x)/K vanishes identically. The minimal
irreducible polynomial of each power xi with 1 ď i ă p is given as

T ip ´ ai ,

hence trK ( x)/K xi = 0 when 0 ă i, and trK ( x)/K 1 = [K ( x ) : K ] = p = 0 as the characteris-

tic K equals p.
Let us prove that a separable extension has a non-degenerated trace form (which
is the implication we shall need); this follows from the primitive element theorem,
which says that L = K ( x ) for some x P L: The minimal polynomial of x is of degree
n = [ L : K ], and it is separable* (by the definition of the extension being separable). The ˚ Recall that a

Cayley–Hamilton theorem tells us that x is a root of the characteristic polynomial Px , polynomial is

separable if all the
hence Px has the minimal polynomial as a factor, and both being of degree n, they are roots in an algebraic
equal. Consequently Px is separable, and Lemma 12.51 below produces an element of closure of the ground
field are simple roots.
non-zero trace. o

Lemma 12.51 Let θ be an endomorphism of an n-dimensional vector space over the field k whose
characteristic polynomial is separable. Then tr θ r ‰ 0 for some r with 0 ď r ă n.

Proof: This gives us the opportunity to retrieve our good old acquaintance the famous
Van der Monde determinant down from the loft. It is the determinant of the n ˆ n-matrix
 
1 1 ... 1
. . . λn 
 
 λ1 λ2
D= .. .. .. .. ,
. . . .


λ1n´1 λ2n´1 ... λnn´1

and it has the virtue of being non-zero when the λi ’s are distinct, as they are in our case.
Recall that tr θ r is the power sum tr θ r = i λri of the eigenvalues λi , so that
ř

(tr θ 0 , tr θ, tr θ 2 , . . . , tr θ n´1 ) = D ¨ (1, . . . , 1),

and since D is invertible, the left hand vector is non-zero, which means that for at least
one exponent r it holds that tr θ r ‰ 0. o

Appendix: transcendence degree

We have relegated a few simple results of preparatory character to this appendix. They
do not take part in the main battle, but are merely skirmishes on the flanks, though of

10th January 2021 at 11:32am

Version 4.0 run 30
 338 integral extensions

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. It should be
known to mosts students from earlier courses, but we include it for the benefit of the
others.

Transcendence degree
(12.52) 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 two 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 (12.10)

where the ai ’s are elements from k and an ‰ 0, and the transcendental ones are the rest;
that is, those for which p( x ) ‰ 0 for any non-zero-polynomial in k[ X ]. Since K is field,
we may as well assume that an = 1 in (12.10) and being algebraic over k is the same as
being integral. The elements in K that are algebraic over k, thus form a subfield s k, the
The algebraic closure algebraic closure of k in K.
(den algebraiske More generally a collection x1 , . . . , xr of element from the bigger field K is said
Algebraically
tillukningen)
dependent elements to be algebraically dependent over k if for some non-zero polynomial p in r variables
(algebraisk avhengige with coefficients from k it holds true that p( x1 , . . . , xr ) = 0, and of course, if no such
elementer)
Algebraically polynomial can be found, the collection is said to be algebraically independent over k. A
independent elements collection of algebraically independent elements x1 , . . . , xn is a transcendence basis of K
(algebraisk uavhengige
elementer)
transcendence basis over k if it is maximal. In other words, the xi ’s are algebraically independent and K is
(transcendence basis) algebraic over k( x1 , . . . , xn ).

Proposition 12.53 Every finitely generated field extension K of k that has a finite transcen-
dence basis.

Proof: Let x1 , . . . , xn generate K over k. If K is not algebraic over k, at least one of

the xi ’s is transcendental, and we may well assume it is x1 . Then K is generated by
x2 , . . . , xn over k ( x1 ) and by induction on n this field extension has a transcendence
basis, which together with x1 yields a transcendence basis for K over k. o
(12.54)

Proposition 12.55 If K is a finitely generated field extension of k, then all transcendent bases
for K over k have the same number elements.

Transcendence degree The common number is called the transcendence degree of K over k and denoted by
(transcendensgrad) trdegk (K ). In case K is not finitely generated, a similar statement holds true; transcen-
dence bases are then no longer necessarily finite, but they the will still be of the same
cardinality.

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: trace and separability 339

Lemma 12.56 (Exchange lemma) 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 L( a). Then a is algebraic
over L(b).

Proof: Since b is algebraic over L( a), there is a polynomial f ( x, y) with coefficients

from L such that f ( a, y) ‰ 0, but f ( a, b) = 0. Expanding f in powers of x we obtain
ÿ
0 = f ( a, b) = qi ( b ) ai ,

where qi (y) P L[y]. This looks very much like a dependence relation for a over L(b); it
only remains to see that f ( x, b) is not identically zero, but since b is transcendent over
L, and at least one of the polynomials qi (y)’s is non-zero, this holds true. o
Proof of Proposition 12.55: Let a1 , . . . , an be a transcendence basis for K over k of
shortest length, and let b1 , . . . , bm be another one.
If the two bases have a common element, induction on n will finish off the proof;
indeed, if a1 = b1 , replacing k by k ( a1 ) = k(b1 ), we may conclude that b2 , . . . , bm and
a2 , . . . , an have the same number of elements.
Now, the role of the Exchange Lemma is to permit us to exchange a1 with one
of the bi ’s: In order to do that, introduce the auxiliary field L = k( a2 , . . . , an ). Not
all the bi ’s cannot be algebraic over L, and we may assume that b1 is transcendental
over L. Since b1 is algebraic over L( a1 ), by the Exchange Lemma a1 will be algebraic
over L(b1 ) = k (b1 , a2 , . . . , an ). As being an algebraic extension is a property transitive
in towers, we may conclude that K, being algebraic over L( a1 , b1 ), is algebraic over
L(b1 ) = k(b1 , a2 , . . . , an ), and so b1 , a2 , . . . , an is a transcendence basis for K over k. o

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 13

Algebras of finite type over fields

The algebras of finite type over a field form the bedrock of the theory of varieties,
and therefore the whole algebraic geometry depends on their properties. This chapter
is devoted to the two most basic results about such algebras and some the central
corollaries.
First comes Noether’s Normalization lemma, a most valuable technically flavoured
result linking algebras of finite type over fields to polynomial rings, and on which rests,
in our approach, the proof of the next basic result, the Nullstellensatz.
Hilbert’s Nullstellensatz is the bridge between algebra and geometry, to which
algebraic geometry owes its very existence. We already met the Nullstellensatz in low
dimensions. In dimension one it is just the definition of a field being algebraically closed
(or, in the complex case one would rather say, the good old Fundamental Theorem of
Algebra), and in dimension two it boils down to a corollary of the classical Gauss’s
lemma (Theorem 3.32 on page 78).
Finally, we close the chapter by giving some consequences of the two big results.
Domains of finite type over a field has the agreeable property that all maximal chains
are of the same length; a property we have termed to be of uniform altitude. Moreover
their Krull dimension equals the transcendence degree of the their fraction field; in
particular, a polynomial ring in n variables is of the highly expected dimension n. And
at the very end of the chapter we skirts the question whether the tensor product of two
domains over a field is a domain, gives a few examples and prove it holds true when
the ground field is algebraically closed.

13.1 Noether’s normalization lemma

Once more a lemma that has become a theorem and once more an important result due
to Emmy Noether. The lemma, or should one say the theorem, states that a domain A
of finite type over k can be realized as a finite algebra over a polynomial ring. In other
342 algebras of finite type over fields

words, one may find elements w1 , . . . , wn in A which are algebraically independent over
k—and hence k[w1 , . . . , wn ] is isomorphic to a polynomials ring —such that A is a finite
module over k[w1 , . . . , wn ].
Since the wi ’s are algebraically independent, and since A is a finite module over
k[w1 , . . . , wn ], they form a transcendence basis over k for the fraction field K of A. The
number n will therefore be equal to the transcendence degree trdegk K.
A heuristic sketch of the proof
(13.1) We find it worthwhile to give a preliminary sketch of the proof, so let A be an
algebra of finite type over k. Starting out with the subfield k Ď A and adjoining elements,
we will sooner or later exhaust the entire ring A since A is finitely generated over k.
In the beginning we may add new elements which are algebraically independent of
those already 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 ones, we are happy, if
not, we have to go back and perturb the already added elements to make the new-comer
integral; and the crux of the proof is to see that this perturbation is possible.
Example 13.1 A non-integral new-comer will typically have a pole, and to illustrate the
perturbation process, we consider the simplest way of adding a function with a pole,
namely the extension k[ x, 1/x ] of k [ x ]. 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.
y

y = 1/x
x

The ring k [ x, 1/x ] is not finite over k[ x ], but perturbing x slightly, we obtain a subring
over which k[ x, 1/x ] is finite. The subring k[ x + 1/x ] will do the job; indeed, k[ x, 1/x ] =
k[ x, x + 1/x ] is generated by x as an algebra over k[ x + 1/x ], and one has the integral
dependence relation
x2 ´ x ( x + 1/x ) + 1 = 0.

10th January 2021 at 11:32am

Version 4.0 run 30
 noether’s normalization lemma 343

Geometrically, the inclusion of k [ x + 1/x ] into k[ x, 1/x ] corresponds to the projection

sending ( x, y) to y + x, which serves a coordinate on the line y = x.
It is worth noticing that almost any perturbation of x will work; that is, k[ x, 1/x ] is
finite over k[ ax + b/x ] as long as both the scalars a and b are non-zero; of course, the
reason lies in the two asymptotes of the hyperbola. K
Exercise 13.1 Show that k[ x, 1/x ] is a finite module over k[ ax + b/x ] for any scalars a
b both being different from zero. M
Exercise 13.2 (Exam mat4200 2020). Let B = k[ x, y, z, w] with constituting relation
zw ´ xy = 0, and let A be the polynomial ring A = k[ x, y, z]. Show that A Ď B and that
the extension is not integral. Describe perturbations x1 , y1 and z1 of x, y and z that make
A integral over k [ x1 , y1 , z1 ]. M

Statement and proof

(13.2) As indicated in the preliminary sketch, the proof of Noether’s Normalization
Lemma goes by induction on the number of generators A requires as an algebra over k,
and the basic ingredient in the induction step is the following lemma:

Lemma 13.3 Let k be a field and let A = k[ x1 , . . . , xm ] be an algebra over k generated by

algebraically dependent elements x1 , . . . , xm . Then there are elements y1 , . . . , ym´1 in A such
that A is a finite module over k[y1 , . . . , ym´1 ].

In the case that m = 1 the conclusion should be understood as x1 being algebraic over k;
that is, it satisfies a polynomial equation p( x1 ) = 0 with coefficients from k.
Proof: By assumption the m elements x1 , . . . , xm are algebraically dependent and thus
satisfy an equation
p( x1 , . . . , xm ) = 0,

where p is a non-zero polynomial in m variables with coefficients in k. The trick is

i
to perturb the variables Xi ’s by putting Yi = Xi+1 ´ X1s where s is a sufficiently large
natural number. Then letting q be the polynomial given by
2 m´1
q( X1 , Y1 , . . . , Ym´1 ) = p( X1 , Y1 + X1s , Y2 + X1s , . . . , Ym´1 + X1s ) (13.1)

we observe that
q( x1 , y1 , . . . , ym´1 ) = 0, (13.2)
i
where yi is the element in A given as xi+1 ´ x1s . Moreover, and this is the important
point, the coefficient of the highest power of X1 occurring in q is a non-zero scalar; that
is,
2 m´1
q( X1 , Y1 , . . . , Ym´1 ) = p( X1 , Y1 + X1s , Y2 + X1s , . . . , Ym´1 + X1s ) = αX1d + lower terms,

10th January 2021 at 11:32am

Version 4.0 run 30
344 algebras of finite type over fields

with α P k and α ‰ 0, and (13.2) is therefore an integral dependence relation for x1 over
k [ y 1 , . . . , y m ´1 ] .
α αm
Indeed, the substitutions above transform a monomial X1 1 ¨ . . . ¨ Xm into a polyno-
mial whose term of highest weight in X1 is X1d where d = α1 + α2 s + . . . + αm sm´1 , and
whose leading coefficient is one. The crux is that all these combinations of αi ’s and
powers of s are different when s is large enough: only finitely many can arise from
(non-zero) terms in q, and equating two can only be done in finitely many ways; hence
there are only finitely many values of s for which two combinations coincide. o
(13.4) We are now well-prepared to attack the full version of the normalization lemma:

Theorem 13.5 (Noether’s Normalization Lemma) Let k be a field and suppose given an
algebra A = k[ x1 , . . . , xm ] generated over k by elements x1 , . . . , xm . Then there are algebraically
independent elements w1 , . . . , wn in A so that A is a finite module over the polynomial ring
k [ w1 , . . . , w n ] .

Proof: As announced, we proceed by induction on m. If the elements x1 , . . . , xm are

algebraically independent, we are through because then A itself will be a polynomial
ring. If not, there is a non-zero polynomial p such that p( x1 , . . . , xm ) = 0, and by
Lemma 13.3 above, there are elements y1 , . . . , ym´1 such that A is finite over B =
k[y1 , . . . , ym´1 ]. By induction there are algebraically independent elements w1 , . . . , wn
in B so that B is finite over k[w1 , . . . , wn ]. But then A will be finite over k [w1 , . . . , wn ] as
well, and we are through. o

Corollary 13.6 Assume that A is a domain which is finitely generated over the field k,
and whose fraction field is of transcendence degree n over k. Then there are n algebraically
independent elements w1 , . . . , wn in A such that A is a finite module over k[w1 , . . . , wn ].

Proof: By the Normalization Lemma we may find algebraically independent elements

w1 , . . . , wn in A such that A is finite over k[w1 , . . . , wn ]. Then K will be a finite extension
of the rational function field k(w1 , . . . , wn ), and hence n = trdegk K. o
Exercise 13.3 The proof of Lemma 13.3 given above is a slightly simplified version of
the proof Nagata gives in his book. In her original proof Emmy Noether used linear
substitutions. They are of a simpler nature than those above, which involve high powers,
but require that the ground field be infinite.
a) Show that setting yi = xi+1 ´ αi x1 with αi P k for each index i, transforms the
relation p( x1 , . . . , xm ) = 0 into an equality

p( x1 , y1 + α1 x1 , . . . , ym´1 + αm´1 x1 ) = pd (1, α1 , . . . , αm´1 ) x1d + lower terms,

where d is the degree of p and pd the homogenous component of p of degree d.

Hint: Taylor expansion.

10th January 2021 at 11:32am

Version 4.0 run 30
 hilbert’s nullstellensatz 345

b) Show that for any non-zero polynomial f (t1 , . . . , tm ) with coefficients in an

infinite field k, there are infinitely many choices of elements αi ’s from k so that
f (α1 , . . . , αm ) ‰ 0. Hint: Induction on the number of variables.
c) Conclude that if A = k [ x1 , . . . , xm ] with x1 , . . . , xm algebraically dependent, there
are elements y1 , . . . , ym´1 in A such that A is finite module over the k-algebra
k [ y 1 , . . . , y m ´1 ] .
M

13.2 Hilbert’s Nullstellensatz

This key result in the fruitful synthesis of algebra and geometry that algebraic geometry
is, comes in about half a dozen different formulations placing different weight on the
geometric and the algebraic aspects. Some are called “weak” and are apparently weaker
than those called “strong”, but of course, in the end they will all be equivalent.
The version of the Nullstellensatz that best elucidates the bridging role between
algebra and geometry, asserts when the ground field k is algebraically closed, that all
maximal ideals in k[ x1 , . . . , xn ] are of the form ( x1 ´ a1 , . . . , xn ´ an ) where ( a1 , . . . , an )
is a point in kn . It thus establishes a one-to-one-correspondence between points in kn ;
that is, geometry, and maximal ideals in k[ x1 , . . . , xn ]; that is, algebra.
And there are scores of different proofs. The one we offer relies on Noether’s
Normalization Lemma and, when the Normalization Lemma is available, is the shortest
path to the Nullstellensatz. However, there are simpler proofs which only use elementary
algebra, and one is offered as an exercise (Exercise 13.7 on page 349 below).
(13.7) After a short paragraph about affine varieties, we begin with we stating the
Nullstellensatz in its original form—as formulated by Hilbert, but in the modern
style of contemporary mathematics—which also goes under the name of the Strong
Nullstellensatz. Subsequently we formulate and prove some of the weaker avatars,
and at the end of the section, we present a proof—the so–called and now classical,
Rabinowitsch trick—that the Strong Nullstellensatz follows from the weak ones.

Varieties
Before the wonderful world of schemes was discovered, the basic geometric objects
in algebraic geometry were the varieties. A drawback varieties suffer compared to
schemes, is that their theory is fully developed only over algebraically closed fields, but
being true to the original Cartesian idea they certainly appeal to the geometric intuition.
(13.8) The building blocks in scheme theory are the spectra of rings, and the similar role Affine varieties (affine
in the theory of varieties is played by the so-called affine varieties. These are zero loci in varieteter)

kn (where k may be any field, but for the most it will be algebraically closed) of prime
ideals in the polynomial ring k[ x1 , . . . , xn ]. Slightly more general, for any ideal a in

10th January 2021 at 11:32am

Version 4.0 run 30
 346 algebras of finite type over fields

k[ x1 , . . . , xn ] the zero locus of a—or the closed algebraic subset defined by a—is the subset
Closed algebraic subset Z (a) Ď kn of points where all the polynomials from a vanish, or expressed in formulae:
(lukket algebraisk
mengde) Z (a) = t ( a1 , . . . , an ) P kn | f ( a1 , . . . , an ) = 0 for all f P a u.

The subsets Z (a) of kn and V (a) in Spec k [ x1 , . . . , xn ] are closely related, but be careful
not to confuse them: a point ( a1 , . . . , an ) P k belongs to Z (a) if and only if the maximal
ideal ( x1 ´ a1 , . . . , xr ´ ar ) lies in V (a), however V (a) is a considerable larger set with
all the prime ideals containing a as members. (The admittedly bad ad hoc notation Z (a)
is not very common, and is chosen just to avoid confusion with V (a). When fluent
in the language of schemes, one would use V for both and say that Z (a) is the set of
k-points in V (a)) .
Another source of notational confusion is the different incarnations of the space kn :
it is a vector space and as such is denoted kn ; it has an incarnation as the spectrum of
the polynomial ring, that is Ank = Spec k[ x1 , . . . , xn ] (which is a different set from kn );
and it is the variety of k-points in Ank and then is denoted by An (k) or simply by kn as
the two are equal.

The Strong Nullstellensatz

Recall the converse of the Z-construction: for any subset S Ď kn the polynomials that
vanish along S, form an ideal I (S) in the polynomial ring, and the Nullstellensatz
describes the relation between these to constructs. A simple but basic observation is
?
that polynomials belonging to the radical a of a all vanish along Z (a), and therefore
?
one has a Ď I ( Z (a)). The Nullstellensatz tells us that this inclusion is an equality. This
is also called the Strong Nullstellensatz since it is easily seen to imply the other versions.

Theorem 13.9 (Hilbert’s Nullstellensatz) Let k be an algebraically closed field and a an

?
ideal in k[ x1 , . . . , xn ]. Then one has I ( Z (a)) = a.

(13.10) The Nullstellensatz has the following consequence, which in fact, is the wording
of one of the weak versions: The locus Z (a) is empty if and only if the ideal a is not a
proper ideal; that is, if and only if a = k[ x1 , . . . , xn ] or equivalently, if and only if 1 P a.
Indeed, a non-zero constant never vanishes, and for the other implication, requiring a
function to vanish at all points in the empty set imposes no restriction, so 1 P I (H).

Weak versions
(13.11) The first version we shall be discussing is slightly out of the line with the others.
It has the virtue of being valid over any field k—also fields which are not algebraically
closed—and is well adapted to Grothendieck’s marvelous world of schemes. It is
formulated purely in algebraic terms, and is readily deduced from the Normalization
Lemma.

10th January 2021 at 11:32am

Version 4.0 run 30
 hilbert’s nullstellensatz 347

(13.12) As a motivating example consider an irreducible polynomial f ( x ) in k [ x ]. It

generates a maximal ideal, and it is well-known that the quotient k[ x ]/( f ( x )) is a finite
field extension of k; powers of the class of x with exponent less than the degree of f
form a basis. This property, that quotients by maximal ideals are finite extensions of the
ground field, is the one that generalizes to the case of several variables.

Theorem 13.13 (General Nullstellensatz) Let A be a finitely generated algebra over a

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 is
not algebraic, it has transcendence degree at least one over k, say r, and by Noether’s
Normalization Lemma it is a finite module over a polynomial ring k[w1 , . . . , wr ] Ď K. By
Lemma 12.28 on page 327 it ensues that k[w1 , . . . , wr ] is a field, which is impossible since
polynomial rings 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
(13.14) We then turn our attention to the two other weak versions of the Nullstellensatz
and return to a situation where the ground field k is algebraically closed. The first
version to be treated, is the one alluded to in the introduction; it describes the maximal
ideals in the polynomial ring k[ x1 , . . . , xn ].
Theorem 13.15 (Weak Nullstellensatz ii) Let k be an algebraically closed field and let
m be a maximal ideal in the polynomial ring k [ x1 , . . . , xn ]. Then m is of the form m =
( x1 ´ a1 , . . . , xn ´ an ) for a point a = ( a1 , . . . , an ) in kn .
Proof: By the general version of the Nullstellensatz above, the field k[ x1 , . . . , xr ]/m is a
finite extension of k, and is therefore equal to k since k is assumed to be algebraically
closed. Let π : k[ x1 , . . . , xn ] Ñ k be the ensuing quotient homomorphism. To retrieve
the point a let ai = π ( xi ). Then obviously all the polynomials xi ´ ai lie in the kernel m
of π, and since a priori ( x1 ´ a1 , . . . , xn ´ an ) is a maximal ideal, it must be equal to m,
and we are through. o
The second weak version asserts that the zero-locus Z (a) is non-empty whenever a is
a proper ideal. Note that it is crucial that the ground field be algebraically closed; if
not, one easily finds examples that Z (a) is empty. For instance, just take a = ( x2 + 1) in
R[ x ].
Theorem 13.16 (Weak Nullstellensatz iii) Let k be algebraically closed and let a be an
ideal in the polynomial ring k [ x1 , . . . , xn ]. Then Z (a) is non-empty if and only if the ideal a is a
proper ideal.
Proof: Since a is a proper ideal, there is a maximal ideal in k[ x1 , . . . , xn ] containing a,
which by the weak version ii is of the form ( x1 ´ a1 , . . . , xn ´ an ). And consequently we
deduce that ( a1 , . . . , an ) P Z (a). o

10th January 2021 at 11:32am

Version 4.0 run 30
 348 algebras of finite type over fields

It is also quite easy to see that this last version implies the previous one: if m is a
maximal deal in k [ x1 , . . . , xn ], the last version implies that Z (m) is non-empty, say
( a1 , . . . , an ) P Z (m). The maximal ideal ( x1 ´ a1 , . . . , xn ´ an ) will then contain m, but m
being maximal, the two must be equal.

The Rabinowitsch trick

We proceed to present the trick found by J. L. Rabinowitsch and published in a thirteen
lines long paper in 1929 (shown in extento at the end of the chapter!) which proves that
the third weak version of the Nullstellensatz (Theorem 13.16 above) implies the full
version. Thus it accomplishes the proof of the Nullstellensatz.
‘
Lemma 13.17 The Weak Nullstellensatz iii implies that I ( Z (a)) Ď a for all ideals in polyno-
mial rings k[ x1 , . . . , xn ].
?
Proof: The task is to demonstrate that I ( Z (a)) Ď a for any proper ideal a in the
George Yuri Rainich
polynomial ring k[ x1 , . . . , xn ], under the assumption that the zero-locus Z (b) is non-
alias J.L. Rabinowitsch
empty whenever b is a proper ideal in a polynomial ring.
(1886–1968)
The crux of the trick is to introduce an auxiliary variable xn+1 and for each element
Ukrainian–American
g P I ( Z (a)) to consider the ideal b in the polynomial ring k [ x1 , . . . , xn+1 ] given by
mathematical physicist

b = a ¨ k [ x 1 , . . . , x n +1 ] + (1 ´ x n +1 ¨ g ).

In geometric terms, the zero-locus Z (b) Ď An+1 (k) equals the intersection of the the
subset Z = Z ((1 ´ xn+1 ¨ g)) and the inverse image π ´1 Z (a) of Z (a) under the

projection π : An+1 (k) Ñ An (k) that forgets the auxiliary coordinate. This intersection is
empty since obviously g does not vanish along Z, but vanishes identically on π ´1 Z (a) .

According to the third version of the Weak Nullstellensatz the ideal b is therefore
not proper, so it holds that 1 P b, and there are polynomials f i in a and hi and h in
k[ x1 , . . . , xn+1 ] satisfying a relation like
ÿ
1= f i ( x 1 , . . . , x n ) h i ( x 1 , . . . , x n +1 ) + h ¨ (1 ´ x n +1 ¨ g ).

Substituting xn+1 = 1/g in this relation and multiplying through by a sufficiently high
power g N of g (for instance, the highest power of xn+1 that occurs in any of the hi ’s will
suffice) we obtain
ÿ
gN = f ( x1 , . . . , xn ) Hi ( x1 , . . . , xn ),

where each Hi ( x1 , . . . , xn ) = g N ¨ hi ( x1 , . . . , xn , g´1 ) is an element in k[ x1 , . . . , xn ]. Hence

?
g P a. o

10th January 2021 at 11:32am

Version 4.0 run 30
 hilbert’s nullstellensatz 349

Exercises
ˇ (13.4) Let k be a field not necessarily algebraic closed. Show that each maximal
ideal in the polynomial ring k [ x, y] is of the form ( f ( x ), g(y)) where f and g are
irreducible polynomials whose stem fields* E f and Eg are linearly disjoint Hint: Recall ˚ Recall that the stem

Proposition 3.30 about polynomial rings over a pid. The concept of linearly disjoint field of an irreducible
polynomial f in k[t] is
extensions is discussed in Exercise 13.13 on page 355. the quotient
(13.5) Let k be a field not necessarily algebraically closed, and let m be a maximal ideal E f = k[t]/( f )
in k[ x1 , . . . , xn ]. Furthermore let K = A/m. Show that there are elements b1 , . . . , bn in K
so that m = ( x1 ´ b1 , . . . , xn ´ bn ) X k [ x1 , . . . , xn ]. M
There are other proofs of the Nullstellensatz much simpler than the one we have
presented, simpler in the sense that they do not rely on substantial results in algebra,
but uses only elementary algebra and standard field theory. It is worthwhile to ponder
over some these proofs, and the two subsequent exercises guide you through two. The
first is a classical proof valid over the complex numbers, and it is based on the complex
field C being of infinite transcendence degree over Q. It exhibit so called "generic
points" in Z (p). The second is one of the simplest proofs around and uses barely more
than there being infinitely many irreducible polynomials over any field.
(13.6) A classic proof over the complex numbers C. This exercise is a guide through a
classical proof of the Nullstellensatz in the weak form iii, which basically is only valid
for algebras over C. It relies on the fact that C is of infinite transcendence degree over
Q (you can take this for granted). For simplicity the exercise is confined to showing
that prime ideals have non-empty zero loci, which is not a severe restriction as any
ideal is contained in a maximal ideal. So let p be a prime ideal in the polynomial ring
C[ x1 , . . . , x n ].
a) Prove that every field K of finite transcendence degree over Q can be embedded
in C. Hint: If y1 , . . . , yr is a transcendence basis for K over Q, the field K will
be algebraic over Q(y1 , . . . , yr ). Use that C is algebraically closed and of infinite
transcendence degree over Q.
b) Assume a finite set of generators t f i u for p is given. Show that there is a finitely
generated field extension k of Q such that each f i lies in R = k[ x1 , . . . , xn ].
c) Let p1 = p X R and let K be the fraction field of R/p1 . Show that K is of finite
transcendence degree over Q, and that there are embeddings of K in C.
d) Conclude that there is a point in Z (p).
ˇ (13.7) A most elementary proof. This exercise describes the steps in one of the simples
proof of the Nullstellensatz. It is completely elementary and relies only on elementary
algebra and rudimentary field theory.
a) Let k any field, finite or not. Prove there are infinitely many irreducible polyno-
mials in k[ x ]. Hint: Revive Euklid’s good old proof that there are infinitely many

10th January 2021 at 11:32am

Version 4.0 run 30
350 algebras of finite type over fields

primes: assume f 1 , . . . , f r are the only irreducibles and consider f 1 ¨ . . . ¨ f r + 1.

b) Let k be a field. Show that the rational function field k (t1 , . . . , tn ) is not a finitely
generated algebra over k. Hint: If it were, only finitely many factors would
appear in denominators.
c) Assume that k Ď K Ď L is a tower of field extension. Assume that L is a finite
algebra over k, and at the same time is a finite dimensional vector space over K.
Prove that K is finitely generated as an algebra over k. Hint: Let t f j u be algebra
generators for L over k and tei u a basis for L over K, both finite. Expand the f j ’s
ř ř
and all the products ei e j as combinations f j = i aij ei and ei e j = l bijl el with
coefficients from K. Then the aij ’s and the bijl ’s will generate K as a k-algebra.
d) Deduce the General Nullstellensatz (Theorem 13.13).
M

13.3 Consequences
The dimension of polynomial rings
(13.18) One might be tempted to consider it intuitively evident that the Krull dimension
of a polynomial ring in n variables is of equal to n. This is true, although astonishingly
subtle to establish. However, mobilizing some of the heavier artillery in the arsenal, the
proof will be straightforward.

Proposition 13.19 (Dimesion of Polynomial Rings) Let k be a field. The Krull dimen-
sion of the polynomial ring k[ x1 , . . . , xn ] in n variables equals n.

Proof: We are going to prove that each maximal ideal m in k[ x1 , . . . , xn ] has height n.
When the ground field is algebraically closed, the Weak Nullstellensatz II (Theorem 13.15
on page 347) says that m is generated by n elements; indeed, m = ( x1 ´ a1 , . . . , xn ´ an ),
and Krull’s Height Theorem (Theorem 11.13) then yields that ht m ď n. On the other
hand, there is the obvious chain of prime ideals

(0) Ă ( x1 ´ a1 ) Ă ( x1 ´ a2 , x2 ´ a2 ) Ă . . . Ă ( x1 ´ a1 , . . . , x n ´ a n ), (13.3)

whose length is n, so that ht m ě n, and we may conclude that ht m = n.

In the general case, let K be an algebraic closure of k and consider the extension
k[ x, . . . , xn ] Ď K [ x1 , . . . , xn ], which is integral by Proposition 12.22 on page 324 . By the
Lying–Over Theorem each maximal ideal in k[ x1 , . . . , xn ] is the contraction of one in
K [ x1 , . . . , xn ] and hence of height n by Corollary 12.43 on page 334. o

Corollary 13.20 Let A be a domain finitely generated over the field k whose field of fractions
is K. Then dim A = trdegk K

10th January 2021 at 11:32am

Version 4.0 run 30
 consequences 351

Proof: If n = trdegk K, there is by Noether’s Normalization Lemma a polynomial

ring k[ x1 , . . . , xn ] Ď A over which A is finite, hence dim A = dim k[ x1 , . . . , xn ] = n by
Going–Up III on page 333. o
(13.21) The Krull dimension of a ring cannot increase when the ring is localized—terms
of chains can only disappear—and frequently it will drop. For example, the fraction
field of a domain is of dimension zero regardless of what the dimension of the domain
is. However, in some distinguished cases it stays the same:

Corollary 13.22 Let A be a domain finitely generated over the field k and let f P A be a
non-zero element. Then dim A f = dim A.

Proof: The algebra A f is finitely generated over k and has the same fraction field as A.
o
Exercise 13.8 Let A be finitely generated domain over k and let S Ď A be a multi-
plicatively closed set. Assume that Spec S´1 A (identified with the subset of Spec A of
primes not meeting S) is dense in Spec A. Show that dim S´1 A = dim A. M
ˇ Exercise 13.9 Given an example of an algebra A of finite type over a field k with a
non-zero elements f such that dim A f ă dim A. Hint: Consider A’s such that Spec A
has components of different dimensions. M

Uniform altitude
Not being catenary is a kind of pathology for, but fortunately most of the rings one
meets when practicing algebraic geometry do not suffer from that shortcoming. In this
section we shall show that the favourites of this chapter, the domains of finite type over
a field, are of uniform altitude; that is, all maximal chains of prime ideals have the
same length. The proof is a reduction to the case of polynomial rings, using Noether’s
Normalization Lemma, combined with an induction argument on the dimension.
(13.23) The induction step requires a little lemma about the dimension of hypersurfaces
in affine space. Examples show that in general dim A/( f ) may drop by two or more
compared to dim A even when A is noetherian. However, the transcendence degree
behaves better, as the following lemma indicates, and combined with Corollary 13.20,
this tells us that the dimension of a hypersurface in affine space is what it should be:

Lemma 13.24 Let k be a field and f P k[ x1 , . . . , xn ] an irreducible polynomial. It then holds

true that dim k[ x1 , . . . , xn ]/( f ) = n ´ 1.

Proof: If K denotes the fraction field of A = k[ x1 , . . . , xn ]/( f ), it will, according to

Corollary 13.20, be sufficient to see that trdegk K = n ´ 1. After renaming the variables,
if need be, we may assume that f is of positive degree in x1 . We claim that the classes
of the remaining variables x2 , . . . , xn then will be algebraically independent over k. If
not, there would be a polynomial g( x2 , . . . , xn ) with coefficients in k belonging to the

10th January 2021 at 11:32am

Version 4.0 run 30
 352 algebras of finite type over fields

ideal ( f ), so that we might write g = h f for some polynomial h. But this is absurd since
g is of degree zero in x1 whereas f is not. It follows that K is algebraic over the rational
function field k ( x2 , . . . , xn ), and we are through. o

Theorem 13.25 Let A be an algebra which is finitely generated over a field k. Then all maximal
chains of prime ideals ascending from a given minimal prime are of the same length. Moreover, if
A is a domain, it is of uniform altitude.

Proof: If q is the given minimal prime ideal in A, there is a one-to-one correspondence

between chains in A with smallest member q and chains in A/q; hence replacing A by
A/q, we may assume that A is a domain, and the minimal prime is then of course the
zero ideal.
The proof goes by induction on the dimension of A, so let n = dim A. By the
Normalization Lemma we my find a polynomial ring B = k[ x1 , . . . , xn ] contained in A
and such that A is finite over B. Consider a maximal chain

0 Ă p1 Ă . . . Ă pr (13.4)

of prime ideals in A. By the Going–Down Theorem (or more precisely its Corol-
lary 12.43), heights are preserved in the extension B Ď A (B is normal being a polynomial
ring, and both rings are domains), and the ideal p1 X B is therefore a height one ideal.
Thus, since B is a ufd, the ideal p1 X B is principal; say p1 X B = ( f ). The extension
B/( f ) Ď A/p1 is integral (Proposition 12.17 on page 322) so that dim A/p1 = dim B/( f )
by Going–Up III. Now, the lemma above tells us that dim B/( f ) = n ´ 1, and hence
dim A/p1 = n ´ 1. By induction dim A/p1 is of uniform altitude, and since the chain

0 Ă p2 /p1 Ă . . . Ă pr /p1

induced by the chain (13.4) is a maximal chain, it follows that r ´ 1 = n ´ 1; in other

words r = n. o
Exercise 13.10 Show that if a ring is catenary so is any localization. Show that if a
ring is of uniform altitude so is any localization M

Density and Jacobson rings

An all important consequence of the Nullstellensatz is that when A is an algebra of
finite type over a field, the maximal ideals in Spec A form a dense subset; and the
even (apparently) stronger property holds true: for any closed subset Z Ď Spec A the
˚ with the
maximal ideals that belong to Z, form a dense subset of Z. This turns out be extremely
understanding that useful and accounts for a lot of the good properties of algebras of finite type over a field.
values at different
points are taken in Heuristically, one is tempted by the explanation that two “functions” (i. e. elements in
varying fields A) assuming the same “values*“ at all closed points are equal.

10th January 2021 at 11:32am

Version 4.0 run 30
 consequences 353

This density property is an important geometric feature shared by many rings not
of finite type over a field. The rings for which each closed set V (a) is the closure its
closed point;t hat is, of set of the maximal ideals containing a, are termed Hilbert rings
by some and Jacobson rings by others, but to day Jacobson rings seems to be the most Hilbert or Jacobson
commonplace usage. rings (Hilbert eller
Jacobson ringer)
Lemma 13.26 A ring A is a Jacobson ring if and only one of the two following equivalent
conditions is satisfied:
i) Every radical ideal is the intersection of the maximal ideals which contain it;
ii) Every prime ideal is the intersection of the maximal ideals which contain it.

Proof: Every radical ideal equals the intersection of the prime ideals containing it
(Proposition 2.58 on page 51) so the two conditions are equivalent.
Ş
For any subset S Ď Spec A it holds true that the closure of S equals V ( pPS p);
Ş
indeed, saying that a Ď p for all p P S, is equivalent to saying that a Ď pPS p. Hence
the closure of the set of maximal ideals containing a equals V (a) if and only if V (a) =
Ş Ş ‘
V ( a Ď m m); that is, if and only if a Ď m m = a. o
(13.27) Of course, rings in general are not Jacobson; an obvious counter example would
be any non-Artinian local ring, it has just one maximal ideal, which it is not the only
prime ideals, so that the spectrum is not reduced to the sole closed point. Algebras
that are finitely generated over a field, however, are all Jacobson. At the bottom this is
a corollary of the Nullstellensatz, but some technical help of the Going-Up results is
needed in the reduction to the case of algebraically closed ground fields.

Proposition 13.28 Let A be an algebra finitely generated over the field k and let a be an ideal
‘
in A. Then the radical a equals the intersection of the maximal ideals containing a; in other
words, A is Jacobson.

Proof: The algebra A is by assumption a quotient of a polynomial ring k[ x1 , . . . , xn ],

and replacing a by the inverse image in k[ x1 , . . . , xn ], we may well assume that A is a
polynomial ring.
We begin with doing the case when k is algebraically closed. A maximal ideal,
which is shaped like ( x1 ´ a1 , . . . , xn ´ an ), contains a if and only if the point ( a1 , . . . , an )
belongs to Z (a). So if f lies in all these maximal ideals, it vanishes along Z (a), and by
‘
the Nullstellensatz, it lies in a.
Proceeding with the general case, we let K denote an algebraic closure of k, and
we may well assume that a is a prime ideal. The extension of polynomial rings
A = k[ x1 , . . . , xn ] Ď K [ x1 , . . . , xn ] = B is integral because k Ď K is (Proposition 12.22 on
page 324), and citing the Lying–Over Theorem (Proposition 12.31 on page 328) we may
chose a prime ideal q lifting p. By the case when the ground field is algebraically closed,

10th January 2021 at 11:32am

Version 4.0 run 30
354 algebras of finite type over fields

q equals the intersection of the maximal ideals containing it; i. e. it holds true that
Ş
q = q Ď m m. Consequently
č č
p = qX A = ( m) X A = (m X A ),
qĎm qĎm

and we are through since by Corollary 12.29 on page 328 each m X A is a maximal ideal
in A. o
Let us remark by the way that the last part of the proof shows that integral extensions
of Jacobson rings are Jacobson.

Tensor products of domains

A most useful feature varieties over algebraically closed fields have is the unrestricted
possibility to form cartesian products. Schemes share this property, and when both
factors are spectra, say Spec A and Spec B, the product in the category of schemes turns
out to be Spec Abk B. Affine varieties (when considered as schemes) are spectra of
integral domains finitely generated over a field k, and so if Spec B and Spec A are two
of the kind, the pertinent question arises wether the tensor product Abk B is a domain
or not (it is clearly finitely generated over k). The answer is the subject of this section;
when k is algebraically closed, the answer is yes, but for general k’s the matter is subtle.
(13.29) The simplest example one can imagine, shows that the answer is no in general.
Just consider CbR C, which is easily shown to be isomorphic as an R-algebra to the
product C ˆ C. More generally, any separable field extension K = k[t]/ f (t) of k suffers
the same fate: Kbk K is not a domain. Indeed, for each root α of f (t) in a root field E
of f , there is an evaluation map K Ñ k (α), and these together induce an isomorphism
Kbk K » K [t]/ f (t) » K ˆ . . . ˆ K with as many factors as f has roots.
Even nilpotents may appear, and this typically happens when the extension k Ď K is
inseparable. For instance, if the characteristic of k equals p and K = k(b) where b is a
pth -root of some a P k (which has no pth -root in k), one finds

Kbk K » K [t]/(t p ´ a) = K [t]/(t p ´ b p ) = K [t]/(t ´ b) p .

(13.30) The cases above are all built on a polynomial getting new roots in an extension,
and indeed, this is at the root of the phenomenon, and is hindered when the ground
field is algebraically closed:

Proposition 13.31 Let k be an algebraically closed field and A and B two finitely generated
k-algebras.
i) If both A and B are reduced, the tensor product Abk B is reduced as well;
ii) If both A and B are domains, then Abk B is a domain as well.

10th January 2021 at 11:32am

Version 4.0 run 30
 consequences 355

Proof: By the Nullstellensatz every quotient B/m of B by a maximal ideal m is equal

to k. We shall temporarily adopt the following suggestive notation and write b( x ) for
the class in B/m of an element b P B, where x stands for the point m in Spec A. Note
ř
further that every element f P Abk B may be expressed as f = 1ďiďt ai bbi with ai P A
and bi P B and with the ai ’s being linearly independent over k (see Exercise 6.12 on
page 160).
Proof of i): Assume that f is a nilpotent element in Abk B, say f n = 0, and write
ř
f = ai bbi as above. Let m be a maximal ideal in B and consider the image fs =
ř sn
i ai bi ( x ) in Abk B/m = A. Then f = 0, and since A is reduced and the ai ’s are
linearly independent, it ensues that all bi ( x ) = 0. In other words, each bi P m, and since
this holds for all m, it follows that each bi = 0; indeed, according to Proposition 13.28
Ş ‘ ‘
above m Ď A m = (0), and (0) = 0 as B is assumed to be reduced.
Proof of ii): We keep our element f ; it will not be nilpotent anymore, but we assume
ř
that f g = 0 for another element g and write g = 1ďiďs ci bdi with the ci ’s linearly
independent over k. For every maximal ideal m in B it then holds that 0 = fsgs in
Abk B/m = A. Hence, either fs = 0 and all bi ( x ) vanish, or gs = 0 and all the di ( x )’s
do; that is, either b = (b1 , . . . , bt ) Ď m or d = (d1 , . . . , ds ) Ď m. It follows that b X d lies in
every maximal of B, and because B is Jacobson, we infer that b X d = (0). Thus either
b = (0) and f = 0, or d = (0) and g = 0 because (0) is a prime ideal in B. o

Exercises
(13.11) This exercises fills in the details of the first example in Paragraph 13.29. Let
k be a field and f (t) a separable and irreducible polynomial in k[t]. Denote by K the
extension K = k[t]/( f (t)) of k. Let E be a root field of f and let α1 , . . . , αn be the roots
of f in E. Define φi : K [t]/( f (t)) Ñ k(αi ) by sending a polynomial p(t) to p(αi ).
a) Show there is an isomorphism Kbk K » K [t]/( f (t));
b) Show that the maps φi are well defined and together yield an isomorphism
K [t]/ f (t) Ñ k(α1 ) ˆ . . . ˆ k(αn ), which sends a polynomial p to the tuple
(φ1 ( p), . . . , φn ( p)).
ˇ (13.12) The hypothesis in Proposition 13.31 that A and B be finitely generated over
k is not necessary. Hint: Show first that if A1 Ď A and B1 Ď B are subalgebras, then
A1 bk B1 Ď Abk B, then reduce to the finite type case.
ˇ (13.13) Linearly disjoint field extensions. If you wonder when the tensor product of two
fields is a field, you probably will appreciate this exercise. Let k Ď E and k Ď F be two
algebraic field extension and assume they appear in towers k Ď E Ď K and k Ď F Ď K in Compositum
a field K. The compositum EF of E and F is the smallest subfield of K containing both. (kompositum)

One says E and F are linearly disjoint if every set of elements from E which are linearly Linearly disjoint fields
independent over k, stay linearly independent over F when considered elements in EF. (lineært ukoblede
kropper)
The condition is a priori asymmetric in the two fields, but will in fact turn out to be

10th January 2021 at 11:32am

Version 4.0 run 30
356 algebras of finite type over fields

symmetric, and it will even be independent of the embeddings in K, as follows from

subproblem c).
a) Show that the assignment eb f ÞÑ e f extends to a k-algebra homomorphism
φ : Ebk F Ñ EF Ď K.
b) Show that φ is always surjective, and is an isomorphism if and only if E and F
are linearly disjoint.
c) Show that Ebk F is a field if and only if E and F are linearly disjoint.
d) Assume E and F to be finite extensions of k. Show that E and F are linearly
disjoint if and only if [ EF : F ] = [ E : k][ F : k ].
M

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 14

Examples of unexpected rings

This chapter is devoted to a series of examples—more of less of a pathological kind—that

show what peculiar phenomena even Noetherian rings can experience. All these rings
are denizens dwelling in the underworld of non-geometric rings, and you wouldn’t
meet them a bright day doing algebraic geometry, but be aware, they lurk near the
boarder, so take care. To be serious: we think it is important and that it improves the
understanding to work out some of the borderline cases in detail; remember the words
of Sun Tzu about knowing your adversary. Moreover constructing these examples gives
us the opportunity to practise many of the techniques taught in the course, and in the
end, they are beautiful pieces of algebra.
The first of the examples we are about to give, is a classic: Nagata’s example that
Noetherian rings may be of infinite dimension. His famous book “Local rings” [?] ends
with the description of a series of peculiar rings having unexpected properties, and this
is the first one. The second example is due to Krull who in [?] gave it as an example of
a normal domain with just one non-zero prime ideal that is not a valuation ring, and it
also appears as the first in a series of examples constructed by Seidenberg ([?, ?] that
the dimension of A[t] can take any value in the admissible range. Our third example
is a domain that is not catenary, and finally, we construct a peculiar Noetherian ring
as an infinite limit of affine blowups; in characteristic p it yields an example of the
Finiteness Theorems (Theorems 12.45 and 12.47 on page 335) not being valid without
the separability condition.

14.1 A Noetherian ring of infinite dimension

(14.1) We proceed to describe Nagata’s example of a Noetherian ring A of infinite Krull
dimension. Each maximal ideal is of course of finite height, but there are maximal
ideals of arbitrary high height.
 358 examples of unexpected rings

The starting point is the ring k[ x1 , x2 , . . .] of polynomials in countably many variables

over a field k, and the construction depends on a decomposition N = i Ii of the natural
Ť

numbers into a disjoint union of finite subsets Ii so that the cardinality #Ii tends to
infinity when i does. For instance, the decomposition can be as simples as the one
with I1 = t1u, I2 = t2, 3u, I3 = t4, 5, 6u an so forth, with Ii consisting of i consecutive
numbers.
For each natural number i we let qi be the prime ideal generated by the variables x j
with j P Ii ; so in the example just mentioned q1 = ( x1 ), q2 = ( x2 , x3 ) etc. Moreover, S
will be the set of elements in k[ x1 , x2 . . .] not belonging to any of the pi ’s; in other words,
it is the set of polynomials with a non-zero constant term. Clearly S is multiplicatively
closed, and we let A be the localization of k[ x1 , x2 , . . .] in S; that is, A = S´1 k[ x1 , x2 , . . .].
Furthermore, we put pi = qi A; they constitute all the maximal ideals in A.
The crucial observation is that when fixing an index i and localizing k [ x1 , x2 , . . .]
˚ which obviously is in the set* T of all none-zero polynomials in the variables x j with j R Ii , we obtain a
multiplicatively closed polynomial ring over a certain field* Ki , namely the rational function field Ki = K ( x j |j R
˚ see also Problem 7.32 Ii ) in the (infinite many) variables x j with j R Ii . And the variables of the polynomial
on page 189 ring are of course the x j ’s for which j P Ii . We further observe that among all the pj ’s
the ideal pi is the only one that survives as a proper ideal in T ´1 k[ x1 , x2 . . .] since all
the others meet T. A subsequent localization in S yields

Api = S´1 T ´1 k[ x1 , x2 , . . .] = Ki [ x j |j P Ii ]mi ,

where mi is the maximal ideal mi = ( x j |j P Ii ) in Ki [ x j |j P Ii ]. This shows that Api , being

the localization of a polynomial ring at a maximal ideal, is Noetherian, and moreover,
its dimension equals the cardinality #Ii . It follows that A has infinite Krull dimension
since #Ii tends to infinity with i.
It remains to see that A is Noetherian, which ensues from the following lemma:

Lemma 14.2 Assume that A is a ring such that all localizations Am at maximal ideals are
Noetherian and that any non-zero element in A is contained in only finitely many maximal
ideals. Then A is Noetherian.

Proof: Let a be an ideal in A. For each maximal ideal m containing a, the ideal aAm
is finitely generated (because Am is Noetherian), and the generators may be chosen to
lie in A. Since a is only contain in finitely many maximal ideals, recollecting all such
generators we get a finite set t f i u. Now, consider a non-zero element a P a. It lies in
finitely many maximal ideals, some of which contain a and some of which do not. Let
tm j u be the ones that do not contain a, and for each j, chose an element a j P a not in m j .
We contend that the f i ’s together with a and the a j ’s generated a. Indeed, these elements

10th January 2021 at 11:32am

Version 4.0 run 30
 a polynomial ring of excess dimension 359

were picked so that they generated aAm for every maximal ideal m: if a Ď m, the f i ’s
will generate, if a Ę m and a R m, the element a will generate, and finally, if a Ę m and
a P m, the a j ’s will do. We may thus conclude since surjectivity is a local property. o
ˇ Exercise 14.1 This exercise is a generalization of Lemma 14.2 above due to William
Heinzer and Jack Ohm. We are given a ring A and a family tAi uiP I of flat A-algebras.
a) Show that if B as flat A-algebra, then (a : x ) B = (aB : x ) for all ideals a and all
elements x in A .
b) Assume that for each maximal m in A, the ideal mAi is proper for at least one i.
Show that if b is an ideal in A such that bAi Ď aAi for all i, then b Ď a.
c) Assume in addition to the assumption in b) that there is a finitely generated ideal
b Ď a such that bAi ‰ aAi for at most finitely many i, and that aAi is finitely
generated for all i. Show that a is finitely generated.
d) Assume that each Ai is Noetherian. Assume further that for each proper ideal
a the ideal aAi is proper for at least one and for at most finitely many i. Show
that A is Noetherian.
M

14.2 A polynomial ring of excess dimension

The following construction of Krull’s furnishes an example of (a necessarily non-
Noetherian) domain having a polynomial ring of pathological large dimension: we
exhibit a one-dimensional local ring A such that the dimension of A[t] equals three;
that is dim A[t] = dim A + 2. This also is an example of a normal domain with just one
non-zero prime ideal that is not a valuation ring; that is, it is not Noetherian.
(14.3) The ring A is no more exotic than the ring of rational functions f ( x, y) in two
variables over a field k which are defined and constant on the y-axis. The elements of A,
when written in lowest terms, have a denominator not divisible by x, and f (0, y), which
then is meaningful, lies in k.
A crucial observation when trying to understand A, is that no rational function g(y)
depending only on y, belongs to A unless it is constant. On the other hand, for every
rational function g(y) the product xg(y) lies in A since it vanishes on the y-axis.
The ring A is not Noetherian; for instance, the principal ideals ( xy´i ) with i P N form
an ascending chain which does not stabilize; indeed, a relation xy´(i+1) = f ( x, y) ¨ xy´i
would give f ( x, y) = y´1 , which is not constant along the y-axis (neither is it well
defined) and hence does not belong to A.
The elements of A are described in the following lemma:

10th January 2021 at 11:32am

Version 4.0 run 30
360 examples of unexpected rings

Lemma 14.4 Every f P A may be written as x ν g(y)α where ν is a non-negative integer, α P A

is a unit and g(y) is a rational function in y alone which is constant if ν = 0.

Proof: One has the equality

a(y) + xb( x, y) a 1 + xba´1

f = xν ¨ = xν ¨ ¨ , (14.1)
c(y) + xd( x, y) c 1 + xdc´1

where a, b, c and d are polynomials with a and c being non-zero and depending on y
alone, and where ν is a non-negative integer. Such a representation is indeed possible:
when f is written in lowest terms, the denominator cannot have x as factor because f is
well defined on the y-axis; this accounts for the exponent ν being non-negative and the
function c(y) being non-zero. Furthermore, a(y) is non-zero when the maximal power
of x is extracted from the enumerator, and the right most fraction in (14.1) will then be
a unit in A. o
Evaluating functions in A on the y-axis (they are by definition well-defined and constant
there) we obtain a ring homomorphism A Ñ k whose kernel is a maximal ideal m in A,
and it turns out that this is the sole non-zero prime ideal in A:

Lemma 14.5 The maximal ideal m is generated by the elements xg(y) with g(y) a rational
function, and it is the only non-zero prime ideal in A. Consequently, A is of dimension one.

Proof: That m is generated by the elements shaped like xg(y) follows immediately
from the previous lemma. Assume then that p is a non-zero prime ideal, and let xi h(y)
be a non-zero element in p with i ě 0 (there are such according to the previous lemma).
Since x ¨ h(y)´1 belongs to A, we infer that xi+1 = ( xi h(y)) ¨ ( xh(y)´1 ) P p, and p being
prime, it ensues that x P p. It follows that ( xg(y))2 = x ¨ xg(y)2 P p for each g(y), and
again as p prime, we infer that xg(y) P p, and by consequence it holds that m = p. o
(14.6) The polynomial ring A[t] has dimension three. It has the prime ideals 0, mA[t]
and (t) + mA[t], but there is also a forth one, namely the ideal p consisting of the
polynomials F (t) such that F (y) = 0. This ideal is not the zero ideal as xt ´ xy lies
there, and it is contained in mA[t]: assume namely that F (t) = i ri ( x, y)ti P p. Then
ř

substituting y for t, gives 0 = F (y) = i ri ( x, y)yi = 0, which with x = 0 yields

ř
i
ř
i ri (0, y ) y = 0. By definition of A the functions ri (0, y ) are constants belonging to k,
and since the different powers yi are linearly independent over k, it ensues that ri P m;
that is, F (t) P mA[t].
(14.7) Let us close the subsection by showing that A is integrally closed in the fraction
field k( x, y), and this is where the hypothesis that k be algebraically closed comes into
play.

Lemma 14.8 The ring A is integrally closed in k( x, y) when k is algebraically closed.

10th January 2021 at 11:32am

Version 4.0 run 30
 a noetherian ring that is not catenary 361

Proof: Let f P k( x, y) be an element integral over A which satisfies the dependence

relation
f n + an´1 f n´1 + . . . + a1 f + a0 = 0. (14.2)
Assume first that f is well defined on the y-axis. With x = 0 the relation (14.2)
becomes a dependence relation for f (0, y) over the field of constants k, but from k being
algebraically closed it then follows that f (0, y) P k. Hence f P A.
It remains to see that f is well defined on the y-axis. So let f = x´i g with g well
defined and not identically zero on the y-axis; multiply through by x ni , the relation
(14.2) becomes transformed into

gn + xi an´1 gn´1 + . . . + x ni a0 = 0

which shows that g(0, y) = 0 if i ą 0, contradiction. Hence f = g, and it is well defined

on the y-axis. o

14.3 A Noetherian ring that is not catenary

There is an easy way to construct such rings, which we shall follow. It is related to
Masayoshi Nagata’s original approach when he gave the first example of Noetherian
non-catenary domains in [?], but with a simplifying twist. The price to pay for simplicity,
however, is that the resulting rings will be pretty weird. There are several other examples
of a much more geometric flavour albeit none of finite type over fields, but they are
Jacobson rings— Tetsushi Ogoma in [?] even gave examples of normal such rings—but
constructing these examples is a much more involved business than we are about to
undertake.
(14.9) The crucial part of the construction is a kind of “pincer movement”. One starts
with a semi local domain R with two maximal ideals m1 and m2 whose residue fields are
isomorphic, say to k, and constructs a subring A of R such that m1 X A = m2 X A = n,
and such that A will be local with maximal ideal n. The construction depends (seriously)
on the isomorphisms R/mi » k, so once and for all we fix these, and we let πi be the
composition of the canonical reduction map and the fixed isomorphism: that is, the
»
composition πi : R Ñ R/mi ÝÑ k.
The induced map Spec R Ñ Spec A identifies the two points corresponding to the
maximal ideals, both are mapped to n, and importantly, nothing much more happens:
the map will be bijective away from the maximal ideals. The staging is shown by the
diagram
R kˆk

A k

10th January 2021 at 11:32am

Version 4.0 run 30
362 examples of unexpected rings

where the uppermost map is the sum π1 + π2 , and where k Ñ k ˆ k is the diagonal.
The ring A with the two remaining maps are defined by the diagram being Cartesian,
which means A is the subring of R given as A = t x P R | π1 ( x ) = π2 ( x ) u. The main
properties of A is described in the next lemma:

Lemma 14.10 In the situation above the following two statements hold true
i) The ring A is local with maximal ideal n = m1 X m2 ;
ii) The map Spec R Ñ Spec A takes each mi to n, and for each finite set of prime ideals
tpi u in A, all different from n, there is an f P n avoiding all the pi ’s, so that A f = R f ;

Proof: The key observation is that the intersection n = m1 X m2 is entirely contained in

A since both maps πi vanish there. It also is the kernel of each restriction πi | A (which
coincide) and is therefore a maximal ideal. One thus has

m1 X m2 = m1 X A = m2 X A.

And n is the only maximal ideal in A since if a is invertible in R and π1 ( a) = π2 ( a)

obviously π1 ( a´1 ) = π2 ( a´1 ) and a´1 P A. This proves i).
We proceed with the proof of ii): by Prime Avoidance there is an element f in n not
belonging to any of the pi ’s. It lies in both the maximal ideals mi , and therefore it kills
each quotient R/mi . Now, one has the exact sequence of A-modules

ρ
0 A R k 0. (14.3)

Here ρ equals π1 ´ π2 ; that is, the composition R Ñ k ‘ k Ñ k of the map π1 + π2 and

the map that sends a pair to the difference. Note that this map is A-linear, but not
R-linear. Since each R/mi is annihilated by f , the sequence (14.3) when localized shows
that A f = R f . The rest follows immediately: there are one-to-one correspondences
between prime ideals in A (respectively in R) not containing f and prime ideals in A f
(respectively in R f ). o
(14.11) The statement ii) is particularly pertinent in our context. It entails that maximal
chains in R ascending to one of the maximal ideals survive unaltered in A, but in A
they all abut at the sole maximal ideal. So if the two maximal ideals in R are of different
heights (both superior to one), the ring A will not be catenary. Remains the question of
Noetherianess of A (as we shall see, it is easy to find appropriate semi local rings).
Of course one must start out with a Noetherian R, subrings however do not always
inherit Noetherianess, but R is a finite module over A. Once this is established, it ensues
from a general theorem independently found by Paul Eakin and Masayoshi Nagata that
A is Noetherian. This result is not covered by this course, so we shall give a simple ad
hoc proof.

10th January 2021 at 11:32am

Version 4.0 run 30
 a noetherian ring that is not catenary 363

Lemma 14.12 In the setting of this section, R is a finitely generated A module, and when R is
Noetherian, A will be Noetherian.

Proof: Let us prove that R is a finite A-module: any two elements e1 and e2 in R with
ei ” δij mod m j generate R over A together with the unity; indeed, their classes in
R/m1 ˆ R/m2 » R/n » k ‘ k form a basis over k » A/n, hence given a P R, there are
elements ai P A so that a ´ a1 e1 ´ a2 e2 P n.
Let now p be a prime ideal in R, it by Cohen’s Criterion it suffices to prove that
p X A is finitely generated; indeed, every prime ideal in A is shaped like that by the
Lying–Over Theorem.
There are two cases: firstly, if a is any ideal in R such that a Ď m1 X m2 , then a is
entirely contained in A. By assumption R is Noetherian so that a is finitely generated
over R, and since R is finitely generated over A, the ideal a is finitely generated over A
as well.
Secondly, assume that p Ď m1 but not contained in m2 (by symmetry this case will
suffice). There is an inclusion

p X A/p X m2 Ď p/p X m2 ,

and p/p X m2 is a finitely generated as a module over R/n » k ˆ k and therefore also as
a vector space over A/n; hence p X A/p X m2 is finitely generated over A. Now p X m2
is contained in m1 X m2 , which we already know is a finitely generated ideal in A; so
we are through. o
(14.13) With these two lemmas up in the sleeve, we summarize and give the example.
It is easy to find semi-local rings of the sought kind: the simplest example is the local-
ization R = S´1 k[ x1 , x2 , y1 , y2 , y3 ] where the multiplicative system S is the complement
of the union ( x1 , x2 ) Y (y1 , y1 , y3 ) of the two prime ideals ( x1 , x2 ) and (y1 , y1 , y3 ). This
ring is semi-local with the two maximal ideals m1 = ( x1 , x2 ) R and m2 = (y1 , y2 , y3 ) R.
One easily verifies that ht m1 = 2 and ht m2 = 3, and that the residue fields are
the function fields respectively given as R/m1 = C(y1 , y2 , y3 ) and R/m2 = C( x1 , x2 ).
These residue fields are of course not isomorphic as C-algebras, but luckily they are
as fields! This is rooted in the fact that C is of infinite transcendence degree over the
field Q of algebraic integers. So in the end, all function fields over C are isomorphic
to C. Choosing a fixed field k in the isomorphism class as well as isomorphisms of
C( x1 , x2 ) and C(y1 , y2 , y3 ) with k, our construction results in a domain A which will
be Noetherian but not catenary. The isomorphisms between function fields and C are
really weird, they mix up infinitely many complex numbers (so for instance, they might
interchange e and π), and the resulting ring we construct will be quirky as well.

10th January 2021 at 11:32am

Version 4.0 run 30
 364 examples of unexpected rings

14.4 A Noetherian bubble space

Many a young geometer has struggled with the subtleties in the examples of Nagata
and Akizuki and finally written them off as belonging to the deepest darkness in the
kingdom of non-geometry. But indeed, many examples are as close to being of finite
type over fields as they can be without being, and have a pronounced geometric flavour.
˚ A more official and The term* “bubble spaces” is used by some mathematicians about some horrendously
serious name is large spaces constructed for studying sequences of blow ups of projective surfaces. Being
“Riemann–Zariski”
spaces. the limit of (infinitely many) affine blow-ups of the simplest sort our space is just a tiny
string of bubbles rising to the surface, but it lies at the heart of several of the Japanese
examples, it is geometric and gives an easy instance of the hypothesis of separability in
the First Finiteness Theorem (Theorem 12.45 on page 335) being needed. In fact, the
amazing thing is that these infinite constructions may yield Noetherian rings—which
also is of independent interest.
In view of the First Finiteness Theorem, an integral closure that is not finite must
involve Frobenius maps in some way, and given the infinite nature of our construct, it is
not surprising that the Frobenius map is not finite.
(14.14) The example A can be constructed as an algebra over any field k. It is a
Noetherian normal domain of dimension two all whose local rings are regular and it is
even a Jacobson ring, but it has a maximal principal ideal m = ( x ) A. The corresponding
point in Spec A is given by one equation, and there is no curve passing by it! The
Krull-dimension dim Am does not equal the height ht m, and there is strict inequality in
the dimension formula in Proposition 11.4 on page 290.
(14.15) We shall be working with a sequence of polynomial rings Ai = k[ x, zi ] indexed
by non-negative integers N0 . The elements zi will all be members of a rational function
field k ( x, z) in two variables, where for the moment k is any field, and they will depend
on a given sequence tai uiPN0 of elements from the ground field k. The definition is
recursive: when i = 0, we set z0 = z, and when i ą 0, the element zi+1 is given by the
relation
z i = x ( z i +1 + a i ). (14.4)

Thus each ring Ai = k[ x, zi ] is identified with the subring k[ x, x (zi+1 + ai ))] of the next
ring Ai+1 = k [ x, zi+1 ], and in this way the Ai ’s form an ascending chain of subrings of
k( x, z):
k[ x, z0 ] Ă k[ x, z1 ] Ă . . . Ă k[ x, zi ] Ă k[ x, zi+1 ] Ă . . . Ă k( x, z). (14.5)

Note that zi+1 = x´1 zi ´ ai for each i P N0 , so for every r it holds that k[ x, x´1 , zi+r ] =
k[ x, x´1 , zi ]; indeed, it readily follows by induction that zi+r P k[ x, x´1 , zi ] for all r P N0 .

10th January 2021 at 11:32am

Version 4.0 run 30
 a noetherian bubble space 365

(14.16) To describe the geometric counterpart of this picture, we introduce the notation
A2i = Spec Ai . These spectra are of course all isomorphic, but they enter the process
as different members of the chain

. . . . . . ÝÑ A2i+1 ÝÑ A2i ÝÑ A2i´1 ÝÑ . . . ÝÑ A21 ÝÑ A20 .

All the maps are so-called “affine blow-ups” which we met in Example 11.13 on page 301.
Heuristically, over the complex numbers each A2i may be thought of as a C2 equipped
with coordinates ( x, zi ), and the maps in the chain send ( x, zi ) Ñ ( x, x (zi + ai )). Thus
the zi -axis collapses to the origin (hence the name “blow up”, as they mostly are seen
from the target), but due to the parameters ai the inverse image of the x-axis is not the
x-axis (as it is in Example 11.13), but a translate, namely the line zi = ´ai .
(14.17) The star of the game will be the union A = k [ x, z0 , z1 , . . .] of all these subrings.
It does not look very Noetherian, and certainly is not in many instances. If all the ai ’s
vanish, for instance, it is not Noetherian (as you were asked to show in Exercise 9.18
on page 245), but amazingly enough it will be when the power series ζ = iě1 ai xi
ř

is transcendental over k; that is, ζ is not algebraic over the polynomial ring k [ x ]. This
condition may also be expressed as all the “tails” ζ r = iě1 ai+r xi with r ě 0 being
ř

algebraically independent:

Lemma 14.18 If ζ is transcendental over k, all the tails will be algebraically independent.

Proof: It holds for any r ě 0 that ζ r = (ζ ´ pr ) x´r where pr is the polynomial pr =

ř
iďr ai xi . Hence, if F ( x, ζ 0 , . . . , ζ r ) is a polynomial, for some sufficiently large natural
number N one has x N F ( x, ζ 0 , . . . , ζ r ) = G ( x, ζ ) where G is a polynomial. Therefore an
algebraic relation between the ζ i gives one for ζ. o
The condition that ζ be transcendental enters the scene by way of a ring homomor-
phism A Ñ kJxK. The power series ζ i = jě1 a j+i x j fulfills the same relations as the zi ’s
ř

do; indeed, one easily verifies that

ÿ
x (ζ i+1 + ai+1 ) = xai+1 + a i +1+ j x j +1 = ζ i ,
jě1

and so zi ÞÑ ζ i is a legitimate definition of a homomorphism. The lemma tells us that it

will be injective if (and only if) the power series ζ is transcendental.
(14.19) Our first observation when starting to investigate the algebra A, is that

( x ) A X k[ x, zi ] = ( x, zi ).

The element zi belongs to ( x ) A because zi = x (zi+1 + ai+1 ), which shows the inclusion
( x, zi ) Ď ( x ) A X k[ x, zi ], and this suffices since ( x, zi ) is a maximal ideal. A consequence

10th January 2021 at 11:32am

Version 4.0 run 30
366 examples of unexpected rings

is that the principal ideal ( x ) A is a maximal ideal: indeed, if a was an ideal strictly larger
than ( x ) A any element in f P a z ( x ) A would lie in some k [ x, zi ], but as ( x ) A X k [ x, zi ] is
maximal, this is impossible.
(14.20) Here comes the salient point of the construction:

Proposition 14.21 If the power series iě1 ai xi is transcendental over k, then A is Noethe-
ř

rian.
Proof: We intend to show that any prime ideal p is finitely generated. Cohen’s criterion
(Proposition 9.35 on page 241) will then imply the lemma. There are two cases to
consider according to wether p Ď ( x ) A or p Ę ( x ) A.
We begin with the latter which is done by a standard argument. Notice, that since
( x ) A is a maximal ideal, it ensues that x R p. For any i it holds that A x = k[ x, x´1 , zi ],
so A x is Noetherian, and we may find f 1 , . . . , f r in p such that ( f 1 , . . . , f r ) A x = pA x .
Furthermore, there is an element g P p not in ( x ) A. We contend that p = ( g, f 1 , . . . , f r ):
let m be any maximal ideal in A. When x R m, the f i ’s will generate pAm , and if
m = ( x ) A, the element g will. Since being surjective is a local property, it follows that
p = ( g, f 1 , . . . , f r ).
The more serious part is the case when p Ă ( x ), which requires that ζ is transcen-
dental. In fact, we contend that this case does not materializes unless p = 0. Now, if p is
non-zero, for each i the intersection p X k[ x, zi ] will a prime ideal properly contained in
( x, zi ), and it is therefore a principal ideal (Proposition 3.30 on page 77), say generated
by an irreducible polynomial f i ( x, zi ), different from x since x R p. Substituting for zi
we find

f i ( x, zi ) = f i ( x, x (zi+1 + ai )) = xg( x, zi+1 ) = h( x, zi+1 ) f i+1 ( x, zi+1 ),

for polynomials g and h in k[ x, zi+1 ]. Since f i+1 is irreducible and different from x, it
ensues that x divides h and consequently that f i P ( x f i+1 ). By induction we infer that
f i P ( xr f i+r ) for all r so that f i P rě1 ( xr ) A. The salient point is now that A Ă kJxK, and
Ş

since the ring kJxK of power series is Noetherian, we can appeal to Krull’s Intersection
Theorem and conclude that rě1 ( xr ) A Ď rě1 ( xr )kJxK = 0.
Ş Ş
o

Exercises
(14.2) Show that the reverse implication in Proposition 14.21 holds: If ζ is not transcen-
dental, then A is not Noetherian. Hint: Show that kernel of A Ñ kJxK is contained in
( xi ) for all i.
ˇ (14.3) Let k be a field of characteristic zero and let d ě 2 be an integer. Show that
i
the power series f (z) = iě0 zd is transcendental. This is an example of a so-called
ř

lacunary series; most of the terms are zero, and the gaps between the non-zero terms
tend (rapidly) to infinity. It persists being transcendental in positive characteristic p

10th January 2021 at 11:32am

Version 4.0 run 30
 a noetherian bubble space 367

(for most p when d is fixed), but the argument is involved. Hint: Use the relation
f (zd ) = f (z) ´ z.
(14.4) The notation is as above. Let P denote the closed point in Spec A corresponding
to the maximal ideal ( x ) A.
a) Show that the distinguished open subset D ( x ) of Spec A is isomorphic to
Spec A x and therefore to A2k zV ( x ).
b) Show that the Zariski closed sets in Spec A are those of the form Z or Z Y tPu
when Z runs though the closed sets Z Ă D ( x ); so D ( x ) is not closed, but of
course, the entire Spec A = D ( x ) Y tPu is.
M

Integral closures in characteristic p

We have come to the example of a domain with an integral closure not being a finite
module; to be precise: we shall construct a Noetherian domain A, in fact a normal one,
whose integral closure in a finite extension of the fraction field is not a finite module
over A. Naturally, the extension must be inseparable, and we shall enter the realm
of rings of positive characteristic. Finally, a slight twist of the construction gives a
Noetherian domain whose integral closure in the fraction field is not finite. ˚ In plain language the

(14.22) The field k will be of characteristic p ą 0, and in the construction we need only p-th power map
p x ÞÑ x p which in
a sequence tai u of elements from k so that the power series associated to tai u and tai u characteristic p is a
both are transcendental over k. One way to achieve this is to assume that k is perfect so ring homomorphism.
ř p
that the Frobenius map* on k is bijective; if i ai xi is transcendental, i ai xi will then
ř

be as well. There are amazingly many amazing transcendental power series over the
finite field F p ; for instance i χP (i ) xi where χP is the characteristic function of the set
ř

of primes*, and another good example is i µ(i ) xi where µ is the Moebius function (see ˚ Geometrically: you
ř

e. g. [?] for proofs and many other examples). So, in fact, there is no restriction on the start blowing up and
each time the number
field since these series coincide with their pth -power series their coefficients being either of blowing ups
0 or ˘1. executed is prime you
move centre a meter;
(14.23) The tactics of the construction are to introduce the Frobenius map in the above then you get a
“bubble-space-setting”, and the salient point is that this Frobenius map is not a finite Noetherian limit-ring!
map.
p
Consider for each i the subring Bi = k[ x p , zi ] of Ai = k[ x, zi ]. The gist of the
example is that the inclusion map Ai Ď Ai+1 takes Bi into Bi+1 ; indeed, since we are in
characteristic p it holds true that
p p p
zi = ( x (zi+1 + ai )) p = x p (zi+1 + ai ).

An even more holds true, the resulting chain of the Bi ’s is of the same shape as the
p
chain (14.5) save being built with the sequence tai u in stead of tai u and the variables x p

10th January 2021 at 11:32am

Version 4.0 run 30
368 examples of unexpected rings

p
and zi . The situation is summarized by the diagram:

p p
... k [ x p , zi ] k [ x p , z i +1 ] ...

... k [ x, zi ] k[ x, zi+1 ] ...

ř p
Note that ζ 1 = i ai xip is transcendental when ζ = i ai xi is. Indeed, assume that
ř

G ( x p , ζ 1 ) = 0. This implies that G ( x p , ζ p ) = 0 and since k is perfect every coefficient of

G has a pth root and there is polynomial H such that H ( x, ζ ) p = G ( x p , ζ p ). The latter
vanishes, and we conclude that H ( x, ζ ) = 0. The games played in the upper chain is
thus exactly the same as the one played in the bottom one, just with slightly altered
p
players; in particular, B = k[ x p , z0 , . . .] is normal and most importantly in our context, it
will be Noetherian.
p p p
In the end, the unions B = i k[ x p , zi ] = k [ x p , z0 , zi , . . .] and A = i k[ x, zi ] =
Ť Ť

k[ x, z0 , z1 , . . .] are normal Noetherian rings and B Ď A. The fraction field of B equals

k( x p , z p ) and that of A is k( x, z). Since A is normal, it is contained in the integral closure
p
B of B in the field k( x, z) in fact, they coincide: each extension Bi = k[ x p , zi ] Ď k [ x, zi ] =
j
Ai is integral (Ai is generated by the monomials x µ zν with 0 ď µ ă p and 0 ă ν ă p
and so is finite over Ai ) and every element in A belongs to some Ai .

Proposition 14.24 The ring B is Noetherian with quotient field k( x p , z p ). The integral closure
B = A in k( x, z) is not a finite module over B.
Proof: We assume for simplicity that p = 2. Aiming for a contradiction, let us assume
that A is finitely generated over B. It will then be a Noetherian B-module since B
is Noetherian, and the ascending chain An = B[ x, z0 , . . . , zn ] of B-modules will be
stationary. Hence for some n ąą 0 it holds that zn+1 P An . Now, An is generated as
a ring over k[ x2 , z2n ]. by the polynomials 1, x, zn and xzn . The coefficients in a relation
expressing zn+1 as a combination of these generators, involve only finitely many of the
zi ’s, and they will all lie in k [ x2 , z2r ] for r ąą 0; hence we may write

zn+1 = p1 ( x2 , z2r ) + xp2 ( x2 , z2r ) + zn p3 ( x2 , z2r ) + xzn p4 ( x2 , z2r )

where the coefficients belong to k[ x2 , z2r ]. Now, zn+1 = xr´n´1 zr + s( x ) and zn =

xr´n zr + t( x ) where s( x ) and t( x ) are polynomials. Merging p1 and xp2 into f , p3 and
xp4 into g and incorporating s and t, we arrive at an identity

xr´n´1 zr = f ( x, z2r ) + xr´n zr g( x, z2r )

between polynomials in k[ x, zr ]. But such a relation is impossible: since the powers of

zr appearing in f ( x, z2r ) all are even and those in all the other terms are odd, we infer

10th January 2021 at 11:32am

Version 4.0 run 30
 a noetherian bubble space 369

that f = 0, thus arriving at the identity:

xr´n´1 zr = xr´n zr g( x, z2r ).

Cancelling xr´n´1 zr gives 1 = xzr g( x, z2r ), which is absurd. o

(14.25) If one insists on having an example where the rings involved are dvr’s just
observe that ( x, z) X B = ( x p , z p ) so denoting this maximal ideal by m, we see that the
discrete valuation ring Bm does not have a finite integral closure in the field k( x, z).

A ring with non-finite normalization

In the previous paragraphs we constructed rings B whose integral closure A in a finite
extension of the fraction field is not a finite module. Almost for free, they give rise to
rings with a normalization that is not a finitely generated module; just take any ring C
lying properly between B and A which is a finite B module and which have the same
fraction field as A (of course, it is not totally for free that such creatures are about).
Indeed, the normalization of C equals A, and A cannot be finite over C as that would
entail that A was finite over B. Moreover, C is Noetherian being a finite module over
the Noetherian ring B.
(14.26) To construct the ring C as described above, we elaborating the example slightly
p
and adjoin x and z0 to B. In that way we obtain the ring C = k[ x, z0 , z1 , . . .]: the fraction
field is k ( x, z) and it is a finitely generated module over B—it is generated by the
monomials x µ z0ν with 0 ď µ ă p and 0 ď ν ă p.
(14.27) Warning: This paragraph presupposes some knowledge of the completion of
a ring. Krull showed that if A is a local Noetherian domain withdim A = 1, then the
normalization A r is finite over A if and only if the completion Ap is without nilpotent
elements; and there are generalisation to other dimension of the same flavour although
not as clean cut. As an illustration of Krull’s result, let us point to a nilpotent element
p the completion of C with respect to the maximal ideal m = ( x, z0 ). Not to create
in C,
p
unnecessary confusion, and to underline that zi is not a pth -power in C when i ą 0, we
p
introduce the notation wi = zi . In C the relation
p p p
z 0 = x ( w1 + a 0 )

holds. Note that if w1 were a pth -power—which it is not—say w1 = ω p , we would have

a nilpotent element since then (z0 ´ x (ω + a1 )) p = 0. However in the completion C
p the
th
element w1 have a p -root. In view of the relations (14.23) a straightforward induction
yields the following equality which is valid for all r ě 1:
r
p
ÿ
w1 = x pr wr+1 + x jp a j .
j =1

10th January 2021 at 11:32am

Version 4.0 run 30
370 examples of unexpected rings

ř8 p
Now x pr wi+r P m pr , so the right hand side converges to j =1 x jp a j , and allows us to
exhibit w1 as a pth -power:
ÿ
w1 = ( xj aj )p.

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 15

Dedekind rings

The work of Richard Dedekind on ideals was the beginning of abstract algebra. And
the rings named after him are central both in algebraic number theory and in algebraic
geometry. In number theory they are the foremost players being the rings of integers in
algebraic number fields, and in algebraic geometry they appear as the coordinate rings
of affine regular curves.
In this chapter we shall first treat the local variant of the Dedekind rings, which
are the so called discrete valuation rings, and as the name indicates, they are members
of the larger family of valuation rings. Valuation rings are closely related to certain
functions on fields with values in ordered groups (which will be the integers Z for
the discrete ones) called valuations. The discrete valuation rings are omnipresent in
both algebraic geometry and algebraic number theory, but the others are seldom met in
contemporary algebraic geometry, and we relegated them to a rudimentary discussion in
an appendix. They are however important in other approach to algebraic geometry than
Grothendieck’s schemey way; the hub in Zariski’s development of algebraic geometry
is the valuation rings in function fields and the so-called places.
The prominent place discrete valuation rings have in mathematics relies on they
being the local one-dimensional integrally closed Noetherian rings. One meets them, for
instance, as the local rings of rational functions regular at a point on any non-singular
curve (e. g. an open subsets of a compact Riemann surface when the ground field is C),
with the valuation being the order of the function at the point. In the study of varieties
of higher dimension, they play a central role in the description of so-called divisors,
which are gadgets built from codimension one subvarieties (note that points on curves
are of codimension one), and in number theory they are omnipresent; rings of integers
in algebraic number fields have local rings all being discrete valuation rings.
A common technique in number theory, which also from time to time is seen
used in algebraic geometry, is to pass back and forth between characteristic zero and
characteristic p, and this is most often done by working with algebras (or schemes) over
a discrete valuation ring whose residue field is of characteristic p while the fraction
 372 dedekind rings

field ha characteristic zero.

15.1 Discrete valuation rings

There are two traditional examples of valuations found in elementary mathematics.
One is the order of vanishing, ordz ( f ), of a holomorphic function f at a point z in the
complex plane (or a rational function to make it more elementary), and the other is
the so-called p-adic valuation v p ( x ) defined on the integers. The latter is the exponent
to which a fixed prime p occurs in the prime factorization of the integer x; in other
words, one has x = pv p ( x) y where the integer y is relatively prime to p. Both functions
are easily extended to fractions—that is, respectively to function meromorphic at the
point and to rational numbers—simply by taking the difference of the values of the
enumerator and the denominator.
The notion of discrete valuations, which we are about to introduce, is built on features
these two examples have in common; how they behave with respect to products and
sums. The order of vanishing, for instance, obey the two well-known rules

ordz ( f ¨ g) = ordz ( f ) + ordz ( g) ordz ( f + g) ě min(ordz ( f ), ordz ( g)),

and the p-valuation shows exactly the same behaviour in that

v p ( xy) = v p ( x ) + v p y v p ( x + y) ě min(v p ( x ), v p (y)).

With every discrete valuation is associated a discrete valuation ring. In the function case
it will be the ring of functions holomorphic at z and in the p-adic case the ring Z( p) of
rational numbers which when written in lowest terms, do not have p as factor in the
denominator. We recognize this ring as the ring of integers Z localized at the prime
ideal ( p).

Discrete valuations
Discrete valuations (15.1) In what follows we shall be working with a field K. A discrete valuation on K is a
(diskrete valuasjoner) non-zero function v : K˚ Ñ Z that obeys the two following rules:

o v( xy) = v( x ) + v(y);

o v( x + y) ě min(v( x ), v(y)),

where x, y P K, and where we must assume that x + y ‰ 0. The fist requirement may be
rephrased as v being a group homomorphism from the multiplicative group K˚ to the
additive group Z. Moreover, since 12 = 1, it immediately follows that v(1) = 2v(1), so
that v(1) = 0, and consequently it holds that v( x´1 ) = ´v( x ).

10th January 2021 at 11:32am

Version 4.0 run 30
 discrete valuation rings 373

If v is surjective, the valuation said to be normalized. The image v(K˚ ) is a subgroup Normalized valuations
of Z and has, as every subgroup of Z, a unique positive generator e. Thus all the values (normaliserte
valuasjoner)
v( x ) have e as factor, and e´1 v will be a normalized valuation which is canonically
associated with v.
It is convenient to introduce a symbol 8 and extend the addition of integers to
Z Y t8u by the rules α + 8 = 8 + α = 8, and of course, we also impose the inequality
α ď 8 for all α P Z. Extending v to K by putting v(0) = 8 one obtains a map
v : K Ñ Z Y t8u still abiding by the two rules above, but without the limitation of only
being defined for non-zero elements.
The two first statement of the following lemma are almost for free, and the third
asserts that when v( x ) and v(y) are different, the inequality in the second rule in fact is
an equality:

Lemma 15.2 Let v be a valuation on the field K and x and y two non-zero elements from K.
Then
i) v( x n ) = nv( x );
ii) v(´x ) = v( x );
iii) If v( x ) ą v(y), it holds that v( x + y) = v(y).

Proof: All is clear when x = 0, so we may assume that x ‰ 0, and then y ‰ 0

as v( x ) ą v(y). The first assertion follows just from v being group homomorphism
and the second from the equality (´1)2 = 1. As to the third, one has v( x + y) =
v(y( xy´1 + 1)) = v(y) + v( xy´1 + 1), so it suffices to show that v(1 ´ t) = 0 whenever
v(t) ą 0 (indeed, set t = ´xy´1 and note that v(t) = v( x ) ´ v(y) ą 0), and since
v(1 ´ t) ě min(0, v(t)), showing that v(1 ´ t) ď 0 will be enough. To that end, consider
the equality
( 1 ´ t ) ´1 = 1 + t + . . . + t n ´1 + t n ( 1 ´ t ) ´1 .

Since v(ti ) = iv(t) ą 0 for i ą 1, it yields

´v(1 ´ t) = v(1 ´ t)´1 ě min(v(1), v(tn (1 ´ t)n )) = min(0, nv(t) ´ v(1 ´ t)) = 0,

when n is sufficiently big, and that’s it. o

ˇ Exercise 15.1 Let v0 : Azt0u Ñ Z be a function on a domain A satisfying the two
axioms for a valuation. Prove that there is a unique discrete valuation v on the fraction
field K extending v0 . Hint: Show that the function v( ab´1 ) = v( a) ´ v(b) is well
defined and complies with the rules. M
ˇ Exercise 15.2 Let p be an irreducible element in the ufd A. Show there is a unique
discrete valuation v p on K with v p ( p) = 1 taking non-negative values on A. M

10th January 2021 at 11:32am

Version 4.0 run 30
374 dedekind rings

Valuation rings
The current section is about discrete valuations and discrete valuation rings, but a few
of their properties have proofs valid for general valuation rings, so for a while, before
coming back to the discrete ones, we shall work with general valuation rings.
Valuation rings (15.3) A domain A with field of fractions K is called a valuation ring if for each element
(valuasjons ring) from K either the element itself or its inverse belongs to A; that is, for each x P K either
x P A or x´1 P A. Note the field K it self will be a valuation ring. Equivalently, one
may require that for any two elements x and y from A either x|y or y|x; indeed, x|y
is equivalent to xy´1 P A and y|x to yx´1 P A. Yet another variant is to ask that the
lattice of principal ideals in A be totally ordered (the condition x|y translates into the
inclusion (y) Ď ( x )), and even more holds true:

Proposition 15.4 A domain A is a valuation ring if and only if the lattice I( A) of ideals in
A is totally ordered.
Proof: Suppose first that A is a valuation ring and assume that two ideals satisfy a Ę b.
We are to show that b Ď a. This is trivial when b = (0), and at the outset a ‰ (0), so
we may assume that both are non-zero. To proceed, pick an element x P a such that
x R b, and let y P b be any non-zero element. Since ( x ) Ę (y), it holds that (y) Ď ( x ),
and consequently y P a.
For the converse implication, let x P K be an element in the fraction field of A and
write x = yz´1 with y and z from A. By hypothesis I( A) is totally ordered, and thus
either (z) Ď (y), in which case x´1 = zy´1 belongs to A, or (y) Ď (z), and x = yz´1 lies
in A. o

Proposition 15.5 Valuation rings are local rings integrally closed in their fraction fields.
Proof: That a valuation ring merely has one maximal ideal, ensues from the lattice of
ideals being totally ordered; indeed, different maximal ideals are not comparable.
Let us then prove that a valuation ring A is integrally closed in its fraction field K.
So assume that x P K is a non-zero element not lying in A, but satisfying an integral
dependence relation
x n + a1 x n´1 + . . . + an´1 x + an = 0,
where the ai ’s are elements from A. Since A is a valuation ring, it holds that x´1 P A,
and in fact, x´1 even lies in the maximal ideal m of A since it is not a unit as x R A. A
simple manipulation gives
1 = ´ a 1 + . . . + a n ´1 x ´ ( n ´2 ) + a n x ´ ( n ´1 ) x ´1 ,

from which ensues the absurdity that 1 P m. o

(15.6) Most valuation rings turn out not to be Noetherian—the exception being the
discrete valuation rings—but they are what is commonly known as Bézout rings, which

10th January 2021 at 11:32am

Version 4.0 run 30
 discrete valuation rings 375

we also met earlier (in Theorem 8.30 on page 215). These resemble the pid’s in that Bézout rings
every finitely generated ideal is principal. (Bézout-ringer)

Proposition 15.7 In a valuation ring each finitely generated ideal is principal.

Proof: By induction on the number of generators it suffices to prove the proposition

for ideals with two generators. So assume that a = ( x, y). Now, since A is a valuation
ring, either x|y and a = (y), or y|x in which case a = ( x ). o
(15.8) Valuation rings are among the larger rings in their function fields. The class of
valuation rings contained in a given field K is closed from above in the sense that any
proper subring of K containing a valuation ring is itself a valuation ring. It ensues that
localizations of a valuation ring is a valuation ring, and in fact, the only proper subrings
of K larger than a valuation ring are the localizations at its prime ideals. In particular,
valuation rings of Krull dimension one, i. e. those having just one non-zero prime ideal,
will be maximal proper subrings of K.

Proposition 15.9 Let A be a valuation ring with fraction field K and let B be an overring of
A different from K; that is, A Ď B Ř K. Then the following holds true:
i) The ring B is a valuation ring;
ii) The maximal ideal of B satisfies m Ď A, and B = Am .

Proof: That B is a valuation ring comes for free since if x R B, a fortiori x R A, and
hence x´1 P A since A is a valuation ring. The maximal ideal m is described as
m = t x P B | x´1 R B u, so if x P m, we may conclude that x´1 R A, and therefore x P A
since A is valuation ring. To see that B = Am , note that if x P A, but x R m, it holds that
x´1 P B. o

Exercises
(15.3) Let A be a domain with fraction field K. Prove that A is a valuation ring if and
only if the set of fractional ideals is totally ordered under inclusion.
ˇ (15.4) dvr’s are maximal subrings. Let A Ď B be two proper subrings of their common
fraction field K. Prove that if A is a dvr, then A = B; that is, dvr’s are maximal proper
subrings of their fraction fields.
ˇ (15.5) Show that the intersection of any collection of valuation rings in a field K is a
valuation ring. Show that the union of an ascending chain of valuation rings in K is a
valuation ring.
ˇ (15.6) Chevalley’s lemma. Prove the so-called Chevalley’s lemma: Assume that A is a
domain and let a be a proper non-zero ideal in A. Let x P K be an element not in
A. Then either aA[ x ] or aA[ x´1 ] is a proper ideal respectively in A[ x ] and A[ x´1 ].

10th January 2021 at 11:32am

Version 4.0 run 30
 376 dedekind rings

Hint: Assume not, and express 1 both as a polynomial in x and as one in x´1 . Use
polynomials of minimal degree, and deduce a contradiction.
ˇ (15.7) Existence of places. Let A be a domain and p a prime ideal in A. Prove that there
is a valuation ring V in the fraction field K of A with maximal ideal mV such that A Ď V
and mV X A = p (such a valuation ring V is traditionally called a place centred at p).
Hint: Consider the set of local rings B containing A whose maximal ideal mB satisfies
mB X A = p. Use Zorn’s lemma and the lemma of Chevalley from the previous exercise
(Exercise 15.6).
M

Discrete valuation rings

With any discrete valuation v on a field K is associated a valuation ring. The underlying
set consists of the elements where v assumes non-negative values; that is, the set

A = t x P K | v( x ) ě 0 u.

That A really is a ring, ensues from the two axioms for valuations. Since v( xy) =
v( x ) + v(y), it holds that v( xy) is non-negative whenever v( x ) and v(y) are, so A
is closed under multiplication. And v( x + y) is non-negative as well, being larger
that both v( x ) and v(y), which shows that A is additively closed. Furthermore, since
v( x´1 ) = ´v( x ), either x P A or x´1 P A; hence A is a valuation ring.
Using that v( x´1 ) = ´v( x ) once more, we see that the group A˚ of units in A
precisely is formed by the elements with v( x ) = 0; indeed, both v( x ) ě 0 and ´v( x ) ě 0
can merely be true for v( x ) = 0. We infer that the maximal ideal in A consists of the
elements in K where v is positive. Thus we have proven:

Proposition 15.10 Let v be a discrete valuation on K and let A be the corresponding valuation
ring. Then the group of units in A is given as A˚ = t x P K | v( x ) = 0 u, and the maximal
ideal as m = t x P K | v( x ) ą 0 u.

Domains that arise as just described—i. e. as subsets of fields where a discrete valuation
Discrete valuation assumes non-negative values—are the famous discrete valuation rings. The initialism dvr
rings (diskrete is widely used.
valuasjonsringer)
(15.11) The descriptions of A, A˚ and m in the proposition are insensitive to changes
of v by positive factors. In particular, they do not change when v is replaced by its
normalization e´1 v (the natural number e being the positive generator of v(K˚ )), and
one may as well restrict one’s attention to the normalized valuations.
(15.12) Ring elements t such that v(t) = e (or v(t) = 1 in case v is normalized) are
Uniformizing called uniformizing parameters for A. The terminology stems from the theory of Riemann
parameters surfaces where a holomorphic function that vanishes simply at a point, may serve as a
(uniformiserende
parametre) local coordinate near the point; it defines an analytic isomorphism between an open

10th January 2021 at 11:32am

Version 4.0 run 30
 discrete valuation rings 377

neighbourhood of the point and an open neighbourhood of the origin in the complex
plane C.
As we are about to explain, in general uniformizing parameters share a few prop-
erties with these local coordinates. In what follows, we assume for simplicity that the
valuation is normalized.

Proposition 15.13 Let v be a normalized discrete valuation on the field K and A the corre-
sponding valuation ring. Then the following hold true:
i) The maximal ideal m of A is principal, generated by any uniformizing parameter;
that is, any element t with v(t) = 1;
ii) If t is a uniformizing parameter, each non-zero element in K may unambiguously be
written as x = a ¨ tv( x) with a being a unit in A;
iii) The non-negative powers mi of the maximal ideal are the only non-zero ideals in A.
In particular, the ring A is pid, and it is Noetherian and of Krull dimension one.

Proof: We begin by attacking the second assertion, and to that end we choose an
element such that v(t) = 1. That an element x P K may be written as x = a ¨ tv( x) with a
a unit in A, amounts to x ¨ t´v( x) being a unit in A, or in terms of the valuation, that
v( x ¨ t´v( x) ) = 0. But v( x ¨ t´v( x) ) = v( x ) ´ v( x )v(t) = 0 as v(t) = 1. In particular, the
element t generates the maximal ideal, and i) comes for free.
It remains to prove that all non-zero ideals are of shape mi ; so let a be one. The
image v(a) is a non-empty subset of the set of non-negative integers and thus has a
least element, say i. Pick an x P a with v( x ) = i. Thence x = a ¨ ti with a P A˚ , and
mi = (ti ) Ď a. By the minimality of i, it holds true that v(yx´1 ) = v(y) ´ v( x ) ě 0 for
any y P a. Hence yx´1 P A, and y P mi . o
A few comments are in place. If the valuation is not normalized, the first assertion
still holds true, but the condition v(t) = 1 must be replaced by v(t) = e with e being
the positive generator for image v(K˚ ).
Secondly, the proposition reveals that the lattice I( A) of ideals in A, ordered with
reverse inclusion, is just the well ordered lattice of non-negative integers N0 . Albeit all
dvr’s have this same ideal structure, they can be dramatically different; their innermost
secrets are hidden in the group of units.
Exercise 15.8 Let A be a dvr with maximal ideal m and residue class field k = A/m.
Show that mn /mn+1 is a one dimensional vector space over k for each natural number
n ě 0. M

dvr’s among local rings

It is of course important to be able to recognise the dvr’s among all the crowd of
local rings. Having a uniformizing parameter is a strong requirement, and one may
wonder if this caracterizes discrete valuation rings. There are, however, local domains

10th January 2021 at 11:32am

Version 4.0 run 30
378 dedekind rings

whose maximal ideal is principal which are not dvr’s (we saw one in Example 11.10
on page 294), but they are not Noetherian—in fact one even finds such rings of any
dimension superior to one—but among local Noetherian domains to have a principal
maximal ideal will suffice for being a dvr. Even the a priori weaker condition that
Ş i
i m = 0 will do.
Above, in Proposition 15.13, the ideal structure of dvr was revealed; the only non-
zero ideals were the powers mi of the maximal ideal, and so the ordered monoid of
ideals is isomorphic to N0 . It turns out that this property is charactertistic for a dvr, at
least among reduced rings; but even the sole order-type of I( A) being N0 implies that
A is a dvr (Exercise 15.11 below).

Proposition 15.14 Let A be a local ring without nilpotent elements that is not a field. Then
the following four assertions are equivalent:
i) A is a dvr;
ii) The powers mi of the maximal ideal are the only non-zero ideals in A;
iii) A is Noetherian and m is principal;
The maximal ideal m is principal and i mi = 0.
Ş
iv)

Proof: Observe that the maximal ideal m is not the zero ideal since A is assumed not
to be a field.
i) ñ ii): This is just statement iii) of Proposition 15.13 above.
ii) ñ iii): Since the powers mi of the maxial ideal are the only non-zero ideals,
every ascending chain must terminate (it is in fact finite) so A is Noetherian, and by
Nakayama’s lemma we may conclude that m2 Ĺ m. Pick an element x P m, but not
in m2 . Then the principal ideal ( x ) can not be equal to any power mi with i ě 2, and
neither can it be zero, hence ( x ) = m.
iii) ñ iv): This is just Krull’s Intersection Theorem.
iv) ñ i): This is close to a repetition of Exercise 9.17 on page 244: Let x generate m
and let y P A be any non-zero element. Since i mi = 0, there is a largest integer ν so
Ş

that y P mν . We may thus write y = ax ν , and since y R mν+1 , the coefficient a is not a
member of m and is a unit. We conclude that A is a domain (indeed, if ax ν ¨ bx µ = 0
with a and b units, x will be nilpotent), and putting v(y) = ν gives a function on A˚
that extends to a valuation on the fraction field in the usual ways (Exercise 15.1). The
salient point is that v is unambiguously defined; indeed, an equality ax ν = bx µ with
µ ą ν and a, b P A˚ entails that x ν ( a ´ bx µ´ν ) = 0 leading to x ν = 0 since a ´ bx µ´ν is
a unit. This in place, the axioms for a valuation ensue painlessly. o

Exercises
(15.9) Assume that A is a local ring whose maximal ideal m is principal, say m = (t).
Prove that b = r mr is a prime ideal and t ¨ b = b. Conclude that if A is a local
Ş

10th January 2021 at 11:32am

Version 4.0 run 30
 discrete valuation rings 379

one-dimensional ring without nilpotents whose maximal ideal is principal, then A is

a dvr.
(15.10) Show that a local ring without nilpotents all whose ideals are principal is
a dvr.
(15.11) Let A be a domain and assume that the lattice of ideals I( A) ordered by reverse
inclusion is order-isomorphic to N0 . Show that A is a dvr.
M

Characterization of dvr’s among Noetherian domains

(15.15) We proceeding to single out the dvr’s in the class of local Noetherian rings.
The two first assertions in the proposition below are equivalent by what we already
have done, but the last statement is deeper and is a cardinal characterization of dvr’s
among Noetherian local rings.

Theorem 15.16 (dvr’s among Noetherian domains) Assume that A is a local Noethe-
rian domain with maximal ideal m. The following statements are equivalent.
i) A is a dvr;
ii) The maximal ideal is principal;
iii) A is normal and of Krull dimension one.

Proof: We already have observed that i) and ii) are equivalent, and that i) implies iii), is
clear: All valuation rings are normal (Proposition 15.5) and the dvr’s are of dimension
one (statement iii) of Proposition 15.13). The only juicy part of the proof is that the two
first assertions follow from the last; we shall show that iii) implies ii). The proof leans
heavily on the theory of primary decompositions, and we have formulated a separate
lemma which also will be useful at a later occasion. From the lemma follows promtly
that m is principal: indeed, if A is of Krull dimension one, the maximal ideal m is the
only non-zero prime. o

Lemma 15.17 Let A be a Noetherian local normal domain and assume that the maximal ideal
m is associated to a principal ideal. Then m is principal.

Proof: Let x P A be such that m is associated to ( x ). Recall that by definition associate

primes are transporters, so for some y P A with y R ( x ) it holds that ym Ď ( x ). Thence
myx´1 Ď A, but yx´1 R A. If myx´1 Ď m the element yx´1 would be integral over A
by the third criterion of Proposition 12.4 on page 319 (because A is Noetherian, m is
finitely generated, and it is faithful as all ideals are). But this is impossible because A is
normal and yx´1 R A. We deduce that myx´1 = A, and consequently there is a relation
zyx´1 = 1 with z P m. Then w = (wyx´1 )z for all w P m (note that wyx´1 lies in A),
and hence m = (z). o
(15.18) The Noetherian hypothesis in Proposition 15.16 is essential. There are local rings

10th January 2021 at 11:32am

Version 4.0 run 30
 380 dedekind rings

with principal maximal ideal of any Krull dimension, but the they will not be Noetherian
when the dimension exceeds one (cfr Subsection 15.62), but curiously enough, those of
dimension one are. There are also examples of normal domains of Krull dimension one
that are not even valuation rings; Krull’s example (in Subsection 14.2 on page 359) is
one.

15.2 Normal domains and discrete valuation rings

A key property of normal Noetherian domains is that their local rings at height one
prime ideals all are discrete valuation rings: they are all one dimensional—this is just
the definition of being of height one—and being localizations of a normal Noetherian
domain they are normal and Noetherian, so Theorem 15.16 above applies. The natural
question then arises whether the converse holds true. The answer is no in general;
normal rings have an additional and more delicate algebraic trait. A rich source of
examples are the cones over space curves in projective three space which are projections
from higher projective spaces; the simplest ones are the cones over quartic space curves,
and we shall examine one in Example 15.1 below.
A recurring annoyance when practising commutative algebra or algebraic geometry
is the occurrence of embedded components, so any result that describes a class of ideals
guaranteed to be without embedded components will be welcome. Among such result
are the so-called “unmixed theorems” (among which Macaulay’s is the most famous);
they predict that the height of the intervening associated prime ideals are uniform, and
this guarantees that there are no embedded components. In this light, the main result
in this section may be seen as a statement that among Noetherian domains the normal
ones are precisely those being regular in codimension one and whose principal ideals
all are unmixed.
A criterion for being normal
(15.19) The criterion we are about to establish is closely related to Serre’s famous R1 –S2
criterion. In the course we do not develop the notion of “the depth” of rings, needed
for stating Serre’s criterion, so we must be content with a minor perturbation of it:

Theorem 15.20 Let A be a Noetherian domain. Then A is normal if and only if the following
two conditions are fulfilled:
˚ This means that there
i) The local rings Ap at each height one prime ideal p is a dvr;
are elements x, y in p
so that x is not a zero ii) Each principal ideal is “unmixed”; i. e. it has no embedded components.
divisor in A and y not
one in A/( x ) A, and it The first condition in the theorem is usually referred to as R1 , or expressed in words,
is equivalent to the that A is “regular in codimension one”, and the second condition bears the label S2 ;
second condition in the
indicating that every ideal in A of height at least two is of depth* at least two. The proof
theorem.
of the theorem is based on the following lemma:

10th January 2021 at 11:32am

Version 4.0 run 30
 normal domains and discrete valuation rings 381

Ş
Lemma 15.21 A domain A is the intersection p Ap where p runs through the prime ideals
associated to principal ideals.

Proof: Seeking a contradiction, we assume there is an element ab´1 that lies in p Ap ,

Ş

but not in A. Consider the ideal a = t y P A | ya P (b) u, which is a proper ideal since
ab´1 R A. Let p be an associated prime to (b) A. Then ab´1 P Ap by assumption, and
we may write ab´1 = cd´1 with c, d P A but d R p. Hence ad = bc, and d P a Ď p, which
is absurd. o
Proof of Theorem 15.20: We start by observing that A is normal when the two con-
ditions are fulfilled: indeed, dvr’s are normal and intersections of normal rings are
normal, and moreover, the second condition combined with Krull’s Principal Ideal
Theorem ensures that all primes associated to a principal ideal are of height one.
Let us then show the other implication: that the two conditions are fulfilled when
A is normal. We have already observed that in that case the local rings at height one
primes are dvr’s, so let p be a prime in A associated to a principal ideal ( x ). Consider
the local ring Ap . Its maximal ideal m = pAp persists being associated to ( x ) Ap , and
citing Lemma 15.17 on page 379 we conclude that the maximal ideal m is principal.
Then Ap is a discrete valuations ring; consequently p is of height one, and therefore it
can not be embedded. o

Corollaries
(15.22) The theorem has a corollary important in geometry, which in a geometric
parlance loosely says that rational functions on a normal varieties can be extended over
codimension two subsets; or equivalently, that the loci where they are not defined, are of
codimension one. It is commonly referred to as Hartogs’ Extension Theorem, even though
it merely is an algebraic reflection of a much deeper result from complex function
theory, proved by Friedrich Hartogs.

Corollary 15.23 (Hartogs’ extension theorem) A normal Noetherian domain A satis-

Ş
fies A = p Ap where the intersection extends over all prime ideals p of height one.

(15.24) Another by-product of Theorem 15.16 is that the only height one primary ideals
in a normal Noetherian domain are the symbolic* powers of the height one primes. And ˚ Recall that the

because it also ensures that principal ideals are without embedded components, the symbolic power of a
prime p is
principal ideals decompose as intersections of symbolic powers. Such decompositions p(ν) = pν Ap X A; see
are entirely described by the occurring “exponents”. Each local ring pAp is a dvr and Exercise 10.11 on
page 275
corresponds to a valuation, vp say, on the fraction field K of A. The exponent of a
“factor” p(ν) in the decomposition of a principal ideal ( f ) is given as ν = vp ( f ), and may
be thought of as the generic order of vanishing of f along the subset V (p) of Spec A.

10th January 2021 at 11:32am

Version 4.0 run 30
382 dedekind rings

Proposition 15.25 Let A be a Noetherian domain. Then A is normal if and only if for every
f P A the minimal primary decomposition of the principal ideal ( f ) is of shape
(ν1 ) (νr )
( f ) = p1 X . . . X pr .

Proof: Assume first that A is normal. By Krull’s Principal Ideal Theorem every minimal
prime of ( f ) is of height one, so by statement ii) of Theorem 15.20 all associated primes
are minimal. If p is a height one prime, the local ring Ap is a dvr with maximal
ideal pAp . Hence if q is p-primary, it holds that qAq is power of pAp , say qAp = pν Ap .
Consequently, by definition of a symbolic power, we find q = qAp X A = pν Ap X A =
p( ν ) .
Attacking the converse implication, we assume that all principal ideals have a
primary decomposition as described. The second condition of 15.20 is then automatic,
so we need only verify the first. To that end, let p be a height one prime, say minimal
over x. Then ( x ) = p(ν) X a, where a is the intersection of primary ideals not belonging
to p. We deduce that ( x ) Ap = p(ν) Ap = pν Ap which is pAp -primary (because pAp
is maximal). Lemma 15.17 then gives that pAp is principal, and Ap is a dvr after
Proposition 15.14 on page 378. o
Exercises
ˇ (15.12) Let p be a principal prime ideal in the ring A which is generated by a non-zero
divisor x. Show that p(v) = pν . Hint: Induction on ν.
(15.13) Let A be a local Noetherian ring A with m and residue class field k.
a) Prove that if some power mi is principal , then A is a dvr;
b) Prove that if dimk mi /mi+1 = 1 for one i, then A is a dvr.
M
Example 15.1 The cone over a quartic space curve: The subring R = k[ x4 , x3 y, x2 y2 , xy3 , y4 ]
of k [ x, y] is called the “cone over a quartic rational normal curve”. In exercise 12.10 on
page 327 you were in particular (with d = 4) asked to prove that R is normal ring.
The ring R is a quotient of the polynomial ring k[t40 , t31 , t22 , t13 , t04 ] with tij being
variables, just send tij to xi y j . So Spec R is a closed subset of A5k . It is a cone with apex
at the origin, in the sense that any line joining a point on Spec R to the origin is entirely
contained in Spec R; and it is the cone over a curve; the rational normal quartic curve.
The idea of the example is to project the curve into a hyperplane by forgetting one of
the coordinates (e. g. the middle one), which on the level of cones corresponds to the
inclusion of tings

A = k[ x4 , x3 y, xy3 , y4 ] Ď k [ x4 , x3 y, x2 y2 , xy3 , y4 ] = R. (15.1)

Such projections of curves tend to be isomorphisms, and in our case it certainly will be,
so on the level of cones the projections only affect the apices. I our case, these apices;

10th January 2021 at 11:32am

Version 4.0 run 30
 dedekind rings 383

that is, the points lying at the origin of A5k in the two cones, correspond to the ideals
m = A X ( x, y) and n = R X ( x, y) respectively.
The extension (15.1) is an integral extension: indeed, R is generated over A by the
element x2 y2 which is integral since it satisfies the equation t2 ´ x4 ¨ y4 = 0.
We contend that for all primes p in A different from m it holds that Ap = Rp . This
hinges on the two facts that a = ( x4 , y4 ) is an m-primary ideal (one easily checks that
m4 Ď a), from which follows that any prime different from m does not contain both the
elements x4 and y4 , and secondly, that x2 y2 = ( x3 y)2 x´4 = ( xy3 )2 y´4 . A consequence
is that Ap is a dvr for all height one primes p; indeed, as ht m = 2, we know that
Ap = Rp when ht p = 1, and by what you did (or should have done) in Exercise 12.10
the ring R is normal, and hence Rp is a dvr.
However, A is not normal as we saw. What goes wrong is that the second condition
in Theorem 15.20, the depth-condition, is not fulfilled at the apex: For instance, m is
associated to the principal ideal ( x4 ): the element x6 y2 does not belong to ( x4 ) (precisely
because x2 y2 is not an element in A), and it holds true that (( x4 ) : x6 y2 ) = m; indeed,
one easily establishes the equalities

y4 ¨ x 6 y2 = x 4 ¨ ( x 3 y )2 xy3 ¨ x6 y2 = x4 ¨ x3 y ¨ y4 x3 y ¨ x6 y2 = ( x4 )2 ¨ xy3

Exercises
In both the following exercises the notation is as in the example above.
(15.14) Show that the minimal primary decomposition of the ideal ( x4 ) in A is given
as ( x4 ) = ( x4 , x3 y, y3 x ) X ( x4 , y4 ).
(15.15) Find an element e P A such that (( x3 y) : e) = m.
M

15.3 Dedekind rings

The original Dedekind rings, the rings that Dedekind studied and for which he proved
his legendary factorisation theorem, were the ring of integers in algebraic function
fields, that is, the integral closures of Z in finite extensions of the rationals Q. In the
introduction to this chapter we referred to the discrete valuation rings as local variants
of the Dedekind rings, and faithful to this point of view, we say that a Noetherian Dedekind ring
domain A is a Dedekind ring or a Dedekind domain if all the local rings Am at maximal (Dedekind-ring)

ideals m are discrete valuation rings. Dedekind domain

(15.26) There are serveral characetrisations of Dedekind rings, and we sum of the most (Dedekind-område)

important ones in the following proposition:

10th January 2021 at 11:32am

Version 4.0 run 30
384 dedekind rings

Proposition 15.27 Assume that A is a Noetherian domain. The following four statements are
equivalent
i) A is normal of Krull dimension one;
ii) Each maximal ideal in A is projective;
iii) Each maximal ideal is invertible;
iv) The ring A is a Dedekind ring; i. e. each localization Am in a maximal ideal is a dvr.
Moreover, in a Dedekind ring every non-zero ideal is projective and invertible.

Proof: We begin with showing the equivalence of i) and iv), and observe that being
normal is a local property (Proposition 12.19 on page 323) so that A is normal if and
only if all localizations Am are. Since dim Am ď dim A, it follows that if dim A = 1, the
same holds for all localizations; and if dim Am = 1 for all m, clearly no chain of prime
ideals can have more than two members. Citing Proposition 15.16 we have shown the
equivalence.
The equivalence of ii) and iv) ensues from finitely generated modules being projective
is equivalent to they being locally free (Corollary 8.2 on page 204). A maximal ideal m
therefore is projective if and only if mAm is free (indeed, mAp = Ap for all p ‰ m); that
is, if and only if mAm is principal. And we are through citing 15.16 again.
Finally, the equivalence of ii) and iii) was already established in Proposition 8.24
on page 213 where invertible ideals were characterized; note that with the Noetherian
hypothesis, all ideals are finitely generated.
Since all non-zero ideals in a dvr are invertible, the final assertion follows by
the Localness of being equal, since the inclusion aa´1 Ď A becomes an equality when
localized at any maximal ideal where Am is a dvr (forming products and transportes
of ideals commute with localization). o

Examples
(15.2) Rings of integers: Due to the first criterion, the original Dedekind rings; that is,
the integral closures of Z in any finite field extension of Q, will be Dedekind rings.
By Going–Up (Corollary 12.35 on page 330) they will be of dimension one, and being
finitely generated (Theorem 12.45 on page 335) modules over Z they are Noetherian,
and of course, they will be integrally closed in K (Proposition 12.11 on page 320).
The same reasoning applies to every one dimensional domain provided one can
prove that the integral closure is Noetherian. We established this for all algebras of finite
type over a field (Theorem 12.47); but in fact, it holds true whenever the field extension
is finite (the integral closure of a one dimensional domain in a finite extension of its
fraction field will always be Noetherian, even though it not necessarily is finite over A).
(15.3) Regular plane curves—The Jacobi criterion: Another inexhaustible source of Dede-
kind rings are the coordinate rings of affine non-singular curves, and among those the

10th January 2021 at 11:32am

Version 4.0 run 30
 dedekind rings 385

plane curves are the more accessible. So let us consider an irreducible curve C Ď A2k
where we for simplicity assume that k is an algebraically closed field. This means that
C = V (( f )) where f is an irreducible polynomial in the polynomial ring k [ x, y], and
C = Spec A with A = k[ x, y]/( f ).
Pick a closed point P P C; it corresponds to a maximal ideal m = ( x ´ a, y ´ b)
in A. We intend to examine the question whether the local ring Am is a dvr or not.
It is certainly a Noetherian domain of dimension one, so the issue is whether m is
principal or not. Now, Taylor expansion gives f ( x, y) = α( x ´ a) + β( x ´ b) + g( x, y)
where g( x, y) P m2 , where α = f x ( a, b) and β = f y ( a, b). And since f = 0 in A this
translates into the identity

α( x ´ a) + β(y ´ b) + g = 0

in A. So if one of the partials does not vanishes at P, say that α ‰ 0, it follows that
( x ´ a) = ´α´1 β( x ´ b) ´ α´1 g, and since m is generated by x ´ a and y ´ b, it ensues
that m is principal.
Now, if the two partials f x and f y have no common zeros along C, which translates
into the algebraic condition that the ideal ( f , f x , f y ) generated by the three is not proper,
all the local rings Am will be dvr s, and we arrive at the following:

Proposition 15.28 Let k be an algebraically closed field. Let f be an irreducible polynomial

in the polynomial ring k[ x, y] and assume that ( f , f x , f y ) = k [ x, y]. Then k[ x, y]/( f ) is a
Dedekind ring.

Exercises
(15.16) Hyperelliptic curves. A particular class of plain regular curves are the so-called
affine hyperelliptic curves1 . They are given by an equation y2 ´ p( x ) = 0 where p( x ) is a
polynomial in k[ x ] without multiple roots. We assume that k is algebraically closed and
not of characteristic two.
a) Show that the ring k[ x, y]/(y2 ´ p( x )) is a Dedekind ring.
b) Show that if a is not a root of p, then x ´ a may serve as a uniformizing parameter
‘
at the point ( a, p( a)), and that y will be one when p( a) = 0.
ˇ (15.17) Let d be a square free integer with d ” 1 mod 4, and consider the quadratic
extension A = Z[ d].
‘

a) Show that the ideal m = ( d + 1, d ´ 1) is maximal in Z[ d] with residue

‘ ‘ ‘

class field equal to F2 ;

1 There is a different and intrinsic definition of projective hyperelliptic curves in algebraic geometry, but

many affine ones may be put on this form.

10th January 2021 at 11:32am

Version 4.0 run 30
386 dedekind rings

b) Show that mAm requires two generators and conclude that Am is not a dvr;
c) Show that m is the only maximal ideal in A such that Am is not a dvr.
(15.18) Let A = k[ x, y]/(y2 ´ x3 ), and we persist with writing x and y for the images
of x and y in A; so that y2 = x3 . We are interested in ideals of A that are contained in
the maximal ideal m = ( x, y).
We contend they are all of the form ( xr , xr´2 ( a + bx )).
M
(15.29) One of the consequences of Proposition 15.27 is that on a Dedekind ring A, or
rather on its fraction field K, there is for each maximal ideal m a valuation vm whose
valuation ring equals Am . There may however be several more valuations on K, but
among those being non-negative on A they are all.

Proposition 15.30 Assume that A is Dedekind ring with fraction field K and let v be a
valuation on K which is non-negative on A. Then there is a maximal ideal m in A such that
vm = v up to normalization.

Proof: Denote by B the valuation ring associated to ν and by n its maximal ideal. By
assumption A Ď B.
It suffices to see that n X A ‰ 0. Indeed, since A is of dimension one, m = n X A
being prime will be maximal, and hence Am Ď B. Now, Am is a dvr as A is Dedekind,
and by Exercise 15.4 on page 375 dvr’s are maximal subrings, and we infer that Am = B.
Hence v and vm are equal up to normalization.
Let x P n and write x = ab´1 with a, b P A. Then 0 ă v( x ) = v( a) ´ v(b) ď v( a),
hence v( a) ą 0 and a belongs to n. o
Example 15.4 The hypothesis in the Lemma above that ν be non-negative on A is
evidently necessary, a stupid example being the localization A x in an element x P A.
Each maximal ideal m containing x induces a valuation vm on K that is not of the
required form relative to A x .
A more interesting example is derived from the degree of a rational function.
Every polynomial in k[t] has a degree deg f , and the negative ´ deg f of the degree
satisfies the two valuation axioms. Indeed, we have deg f g = deg f + deg g and
deg( f + g) ď max(deg f , deg g), and changing sign in the latter we obtain the inequality
´ deg( f + g) ě ´ max(deg f , deg g) = min(´ deg f , ´ deg g).
Consequently, it induces a valuation on the field k(t) of rational functions. The
function t´1 will be a uniformizing parameter and of course, it does not belong to any
maximal ideal in k[t] (not even to k[t] itself). This valuation is just the order of vanishing
at infinity of the rational functions. K
Exercise 15.19 The aim of this exercise is to construct another example of a valuation
on a Dedekind ring A that is not induced by a maximal ideal. Students acquainted

10th January 2021 at 11:32am

Version 4.0 run 30
 dedekind rings 387

with projective geometry will at once see through the example and recognize the extra
valuation as the order of vanishing of functions at the point of infinity of an elliptic
curve.
a) Consider the algebra A = k[ x, y] with constituting relation y2 = x ( x2 ´ 1). The
fraction field of A is a quadratic extension of k( x ) with y2 = x ( x2 ´ 1). Show
that there is a valuation v on K with v( x ) = 2 and v(y) = 1. Hint: If m = ( x, y),
show that Am is a dvr.
b) Consider the two elements z = y´1 and t = xy´1 in K, and show that K = k(z, t),
Show that this a quadratic extension of k (t) with tz2 ´ z + t3 = 0. Show that
there is a valuation w on K with w(t) = 1 and w(z) = 3. Hint: let B = k[z, t]
with constituting relation z ´ t(t2 ´ z2 ) and show that K is the fraction field of B;
and then that Bn with n = (z, t) is a dvr.
c) Conclude that w( x ) = ´2 and w(y) = ´3.
M

Main Theorem on ideals in Dedekind rings

We have now come Richard Dedekind’s honoured generalization of The Fundamental
Theorem of Arithmetic, his theorem about factorization of ideals in the rings named
after him. After the groundbreaking contributions of Noether and Lasker, this is not
very difficult, but is merely an analysis of how primary decompositions are shaped in
Dedekind rings.
(15.31) So let a be an ideal in the Dedekind ring A. That the local ring Am at a maximal
ideal m is a dvr, has the effect that aAm = mν Am for an unambiguously determined
non-negative integer ν, simply because the powers of mAm are all ideals in Am . And
in that way we may associate a number vm (a) with a for each maximal ideal m, it is
called the order of order of a at m, and these numbers will be the exponents appearing in The order of ideals at a
Dedekind’s theorem. Observe that an ideal a is contained in a maximal ideal m precisely maximal ideals
(ordenen til idealer i et
when aAm is proper; that is, when vm (a) ą 0. maksimalt ideal)

Theorem 15.32 (Main Theorem of Dedekind rings) Let A be a Dedekind ring. Then
every non-zero ideal a equals a product of maximal ideals. That is

ν
a = m11 ¨ . . . ¨ mrνr , (15.2)

where the mi ’s are different maximal ideals, and the νi ’s are natural numbers. The ideals mi and
the exponents νi are unambiguously determined by a.

Note that since the mi ’s are maximal, their powers are primary, and since they are
comaximal, their product equals their intersection. Thus the mi ’s appearing in (15.2) are
the minimal primes of a, and the exponents νi are the orders νi = vmi (a) of a at mi ; or if

10th January 2021 at 11:32am

Version 4.0 run 30
388 dedekind rings

ν
one prefers, the νi ’s are the unique numbers such that aAmi = mi i Am . This takes care
of the unicity. In a compact notation the equality (15.2) takes the form
ź
a= mvm (a) ,

(where it is understood that the infinitely many factors equal to A do not contribute to
the product).
Proof: Only existence needs an argument. Let m1 , . . . , mr be the minimal primes of a. As
always in Noetherian rings they are finite in number, and because A is one-dimensional,
they will all be maximal. Letting νi = vm (a) we have the inclusion

a Ď m11 X . . . X mrvr = m11 ¨ . . . ¨ mrvr

ν ν
(15.3)

where the intersection equals the product as the different powers mi νi are comaximal.
Finally, by the very definition of the integers νi , the inclusion in (15.3) becomes an
equality when localised at any maximal ideal and therefore is an equality. o
(15.33) The orders of results of the usual operations on ideals are easily expressibel in
terms of the orders of the invoved ideals:

Proposition 15.34 Let A be a Dedekind ring and a and b two ideals in A. Then the following
statements hold true for all prime ideals m:
i) vm (a X b) = max(vm (a), vm (b));
ii) vm (a + b) = min(vm (a), vm (b));
iii) vm (ab) = vm (a) + vm (b);
Moreover, it holds true that
iv) a Ď b if and only if vm (a) ď vm (b) for all m.

Proof: Let m be a maximal ideal in A. When localized at m, the ideals a and b are
transformed into aAm = mν Am and bAm = mµ Am , and consequently we find:

(a X b) Am = aAm X bAm = mν Am X mµ Am = mmax(ν,µ) Am

(a + b) Am = aAm + bAm = mν Am + mµ Am = mmin(ν,µ) Am .

This takes care of i) and ii), and statement iii) is trivial. One of the implications in iv)
is obvious, so assume that vm (a) ď vm (b) for all maximal ideals m. This means that
aAm Ď bAm for all m; and the inclusion b Ď a + b will when localized in each m, be an
equality, hence it is an equality by the localness of being equal, and we can conclude
that a Ď b. o
Example 15.5 There is a clear qualitative explanation of the ambiguous factorization
2 ¨ 3 = (1 + i 5) ¨ (1 ´ i 5) in the ring Z[i 5], which involves the following three
‘ ‘ ‘

10th January 2021 at 11:32am

Version 4.0 run 30
 dedekind rings 389

‘ ‘ ‘
ideals q = (2, 1 + i 5), p1 = (3, 1 + i 5) and p2 = (3, 1 ´ i 5). They are all prime
ideals, and an easy computation shows that q2 = (2) and p1 p2 = (3). It follows that
the factorization of (6) into a product of prime ideals is (6) = p1 p2 q2 . One checks that
‘ ‘ ‘ ‘
p1 q = (1 + i 5) and that p2 q = (1 ´ i 5). The factorization 2 ¨ 3 = (1 + i 5)(1 ´ i 5)
thus arise from grouping the four factors of p1 p2 q2 into pairs in two different ways. K

Exercises
(15.20) Let k be a field and A = k [ x, y] with constituting relation (y2 ´ x ( x2 ´ 1)).
Determine the factorization of (y ´ ax ) in A.
(15.21) Let A be a Dedekind ring and let a P A be an element. Assume that ( a) =
ś
1ďiďr pi with the pi ’s being prime ideals. Show that a is irreducible if and only if for
ś
no proper subset J Ĺ t1, . . . , ru the ideal iP J pi is principal.
(15.22) Let A be a Noetherian ring and S a multiplicative set in A. Assume that S is
generated by prime elements; that is, every element s P S is a product s = p1 ¨ . . . ¨ pr
with pi a prime element and pi P S.
a) Show that if p Ď A is a prime ideal and pAS is principal, then p is principal.
b) Assume further that A be a Dedekind ring that is not a pid, and let S be the
multiplicative set generated by all prime elements in A. Prove that S´1 A is a
Dedekind ring where no prime ideal is principal.
M

Corollaries
Combining the Main Theorem with the Chinese Remainder Theorem we deduce three
corollaries.
(15.35) Given r points x1 , . . . , xr in the complex line C with r multiplicities ν1 , . . . , νr
associated to the points, one easily finds a polynomial having a zero of order exactly νi
at each xi and having no other zeros. Just take the product of the ( x ´ xi )νi as i varies
from 1 to r.
A similar, but weaker, statement holds true in any Dedekind ring, but with two
important modifications. The obvious first twist is to replace the r points by r maximal
ideals m1 , . . . , mr and ask for an element a with vmi ( a) = νi . Secondly, and much
more substantially, one can no longer guarantee to find elements vanishing only at the
specified prime ideals mi . They might, and often will, have other zeros away from the
given primes.

Proposition 15.36 Let A be a Dedekind ring. Let m1 , . . . , mr be different maximal ideals in A

and denote by v1 . . . , vr the corresponding valuations. Given non-negative integers n1 , . . . , nr .
Then there exist elements a P A so that vi ( a) = ni for 1 ď i ď r; or phrased differently, so that
n
a P mi i zmi ni +1 for 1 ď i ď r.

10th January 2021 at 11:32am

Version 4.0 run 30
390 dedekind rings

n
Proof: In view of the powers mi i being pairwise comaximal this is just the Chinese
Remainder Theorem: the map
n +1
A Ñ A/m1 1 ˆ . . . ˆ A/mrnr +1

that sends a to ([ a]1 , . . . , [ a]r ), where [ a]i denotes the class of a modulo mi ni +1 , is
surjective. If for each index i one lets ti be a uniformizing parameter at mi , any element
n
a that maps to the sequence ([t1 ]1 1 , . . . , [tr ]rnr ), is as wanted. o
(15.37) Ideals in a Dedekind ring require at most two generators, and this is a particular
property even among one-dimensional domains. Heuristically, the coordinate ring of a
curve which near the origin approximates the union of the coordinate axes in Cn , will
have a maximal ideal that requires n generators.

Corollary 15.38 Any ideal a in a Dedekind ring A is generated by at most two elements.

Proof: Let m1 , . . . , mr be the minimal primes of a and chose an element a P A such

that vmi ( a) = vmi (a). The ideals m1 , . . . , mr are certainly among the minimal primes
of ( a), but there might be others, say n1 , . . . , ns . Chose an element b P A such that
vmi (b) = vmi (a) for 1 ď i ď r and vnj (b) = 0 for 1 ď j ď s; i. e. such that b P a, but b R n j
for 1 ď j ď s. Then ( a, b) Ď a, and this inclusion becomes an equality when localized at
each maximal ideal; hence it is an equality. o
Exercise 15.23 Consider the polynomial ring k[ x, y, z] over the field k. Let a =
( x, y) X ( x, z) X (y, z). Show that A = k[ x, y, z]/a is of dimension one, but that ( x, y, z) A
cannot be generated by two elements. Can you find a domain with the same properties?
M
(15.39) The third corollary comes here:

Corollary 15.40 A Dedekind ring with finitely many maximal ideals is a pid,

Proof: Let m1 , . . . , mr be the maximal ideals in A. Each ideal a can according to the
ν
Main Theorem be expressed as a = m11 ¨ . . . ¨ mrνr . In view of Proposition 15.36 on the
previous page there is an element a P A with vmi ( a) = νi for each i. Then ( a) Ď a and
the inclusion becomes an equality when localized at each mi ; that is, at every maximal
ideal of A. Hence it is an equality. o

The ideal class group and the Picard group

For Dedekind rings the ideal class group is, together with the group of units A˚ , the
most important invariant, and it is fair to say that in traditional algebraic number theory
they even were the main objects of study, and the reason is as follows (remember that the
theory arose out of the failure of the factorization theorem). As all ideals in a Dedekind
ring A are invertible, A is a pid precisely when the Ideal class group vanishes, and

10th January 2021 at 11:32am

Version 4.0 run 30
 dedekind rings 391

since Dedekind domains are factorial if and only if they are principal ideal domains,
one has the important:

Proposition 15.41 A Dedekind domain A is factorial if and only if Cl( A) is trivial.

(15.42) In the “Kummer-set-up”, when element (i. e. numbers) are replaced by ideals,
one is tempted to state that the group of units A˚ is the measure of contraction while
Cl( A) measures the expansion, as illustrated by the exact sequence

t1u / A˚ / K˚ / I ( A) / Cl( A) / t1u.

KLADD

The Krull–Akizuki Theorem

Back in chapter 12 we experienced the “finiteness issue”; that the integral closure of a
domain A in a finite extension L of the fraction field K is not always a finite A-module.
However, in the special case of closing up a Dedekind rings, which is important in
number theory, the closure will at least be Dedekind. The pertinent issue is wheteher
the integral closure is Noetherian, all the other requests for being Dedekind, come
easily. There are several more general versions of this result about integral closures of
one-dimensional rings, but the “soul” of all other proofs appears in this simples version
(which also is the most important one in number theory).

Theorem 15.43 (Krull–Akizuki) Let A be a Dedekind ring with fraction field K and let L
be a finite extension of K. Then the integral closure of A in L is a Dedekind ring
Proof: By Going–Up, B is one-dimensional, and being the integral closure it is integrally
closed, and remains to show it is Noetherian, and that Corollary xxx below. o
(15.44) The proof of the integral closure being Noetherian hinges on the following
lemma:

Lemma 15.45 Let A be a Dedekind ring and let M be a torsion free A-module of finite rank.
Then for each maximal ideal m of A the dimension of M/mM as a vector space over A/m is
bounded by rk M.
Proof: Let n = rk M and let K denote the fraction field of A. If m1 , . . . , mn+1 be
arbitrary elements in M our task is to show that the classes [mi ] in M/mM are linearly
dependent. To that end, observe that they are dependent in Mb A K, so there is a relation
ř
1ďiďn+1 ai mi = 0 with coefficients ai P A. The ideal a = ( a1 , . . . , an+1 ) is invertible,
and because a´1 a = A, there is an x P a´1 so that xa Ę m. Then each xai lie in A, and
at least one does not lie in m. Reducing the relation above mod mM then yields a linear
dependence relation among the classes [mi ]. o

Corollary 15.46 If B is a the closure of A in L, then B is Noetherian.

10th January 2021 at 11:32am

Version 4.0 run 30
392 dedekind rings

Proof: By Lying–Over, each non-zero prime p in B intersects A in maximal ideal m;

hence mB Ď p. By the lemma B/mB is of finite dimension over A/m, and thus it is
Noetherian. It follows that each p/mB is finitely generated, which certainly implies that
p is finitely generated, and we are through citing Cohen’s criterion. o
Note that the fibres over m are finite; # bounded by [ L : K ] SLUTT PÅ KLADD

15.4 Finitely generated modules over Dedekind rings

(15.47) Just as is the case for principal ideal domains, it turns out that finitely generated
torsion free modules over Dedekind rings are projective, and as a the corollary submod-
ules of finitely generated projective modules will be projective. Over Dedekind rings
there is also a quite satisfactory classification of projective modules up to isomorphism:
they are all direct sums of invertible ideals (i. e. projective modules of rank one)—in a
geometric language this says that vector bundles over regular affine curves decompose
as direct sums of line bundles—and there is a workable criterion for such sums to be
isomorphic.

Theorem 15.48 (Torsion free modules over Dedekind Rings) Every finitely generated
torsion free module over a Dedekind ring A is isomorphic to a direct sum of (invertible) ideals;
or in other words, to a direct sum of projective rank one modules.

We remind you of Theorem 8.30 about torsion free modules over pid’s, and the proof of
the precent theorem is mutatis mutandis the same as of that result, except that ideals will
be projective and not free.
Proof: The proof goes by induction on the rank of the module, which we call E. Citing
Lemma 8.31 on page 216 we may find a non-zero A-linear map π : E Ñ A. Its image
is an ideal a, which is projective because A is Dedekind, and consequently π is a split
surjection onto it image. Hence E » F ‘ a where F is torsion free and rk F = rk E ´ 1.
If E is of rank one, the rank of F will be zero, and being torsion free it must reduce to
zero. Hence E is isomorphic to an ideal. If rk E ą 1, we invoke the induction hypothesis
to infer that F is a direct sum of ideals. Obviously the same is then true for E. o
We immediately obtain the following corollary:

Corollary 15.49 Every finitely generated torsion free module over a Dedekind ring is projec-
tive. A submodule of a finitely generated projective module over a Dedekind ring is projective.

(15.50) Like the Bézout rings being non-Noetherian analogues of the pid’s, the De-
dekind rings have non-Noetherian cousins called Prüfer rings, in which all finitely
generated ideals are projective. The theorem holds true for these, and the proof goes
through without modifications, merely supplemented with the observation that direct
summands of finitely generated modules are finitely generated. The second assertion in

10th January 2021 at 11:32am

Version 4.0 run 30
 finitely generated modules over dedekind rings 393

the corollary is not true as stated, but persists being true with the additional assumption
that the submodule be finitely generated.
(15.51) From Theorem 15.48 above arises the natural question when two direct sums of
ideals are isomorphic, and it seems that Steinitz (at least Kaplansky attributes the result
to him) was the first to give an adequate answer. The easier part is that the number of
summands—i. e. the rank of the two modules—must be equal, the subtler part that the
products of the ideals involved must be isomorphic. This gives a complete classification
up to isomorphism of finitely generated torsion free modules over Dedekind rings
(provided the Picard group is known). To a finitely generated projective module over
any domain, there is associated an invertible module called the determinant (for those
who know those creatures, it is the highest non-vanishing exterior power). In the present
case the notion boils down to the product of the involved invertible summands, and The determinant of a
in order to be in sync with general accepted usage, we shall write det M = a1 ¨ . . . ¨ ar projective module
À (determinanten til en
when M = i ai and name it the determinant of M (note that this determinant is not a projective modul)
priori an invariant of M, but depends on the specific decomposition as a direct sum).

Theorem 15.52 (Projective modules over Dedekind rings) Any finitely generated tor-
sion free module M over a Dedekind ring is isomorphic to a direct sum of ideals. Two such direct
sums are isomorphic if and only they have the same rank, and their determinants are isomorphic.

The first step in the proof is to treat the special case that all save one of the summands
in one of the sums are trivial:
À
Lemma 15.53 There are isomorphisms 1ďiďr ai » (r ´ 1) A ‘ a1 ¨ . . . ¨ ar .

Proof: We begin with reducing the general case to the case of two summands by
À
induction on r. Let M= 1ďiďr ai be a given direct sum. By induction there is an
À
isomorphism 2ďiďr ai » (r ´ 2) A ‘ a2 ¨ . . . ¨ ar , but by the case of two summands there
is an isomorphism a1 ‘ a2 ¨ . . . ¨ ar » A ‘ a1 ¨ . . . ¨ ar , and consequently we obtain an
isomorphism a1 ‘ (r ´ 2) A ‘ a2 ¨ . . . ¨ ar » (r ´ 1) A ‘ a1 ¨ . . . ¨ ar .
When a1 and a2 are two comaximal ideals, it holds true that a1 X a2 = a1 a2 and
a1 + a2 = A, so that the Chinese exact sequence takes the form

0 / a1 a2 / a1 ‘ a2 /A / 0.

It splits and gives the desired isomorphism. The final step is to see that for every two
non-zero ideals a1 and a2 one may find elements x and y in A so that xa1 and ya2 are
comaximal: two ideals are comaximal precisely when they are not contained in the same
maximal ideal, so let tmi ui and tn j u j be the maximal ideals containing a1 and a2 . The
submodules Hom A (mi , n j ) = (n j : mi ) of K are finitely generated and finite in number,
and hence their sum S does not equal K (which is not a finitely generated module over

10th January 2021 at 11:32am

Version 4.0 run 30
394 dedekind rings

A unless A is semi-local in which case A is a pid and the theorem is trivial). So pick an
element xy´1 from K but not in S. Then xmi Ę yn j for each i and j, and hence xa1 and
ya2 are comaximal. o
The next lemma finishes the proof of Theorem 15.52:

Lemma 15.54 Let A be Dedekind ring, a and b two ideals and r a natural number. If there is
an isomorphism rA ‘ a » rA ‘ b, then a » b.

Proof: The political correct—and admittedly the best—proof uses exterior powers and
determinants of vector bundles. However, we do not have tools of that sophistication
at our disposal and shall present a completely elementary ad hoc proof. Once the
appropriate diagram is displayed, it is an entirely self-propelled induction on r. We may
certainly assume that rA ‘ a = rA ‘ b and shall denote this module by E. Furthermore,
we pick two surjections π and π 1 from E onto A having kernels (r ´ 1) A ‘ a and
(r ´ 1) A ‘ b respectively (we may surely assume that r ě 1 and that the submodules a
and b of E are different). They enter in the diagram

0 0

0 a1 A
π

π1
0 (r ´ 1) A ‘ a E A 0

0 F (r ´ 1) A ‘ b b1 0

0 0 0

with exact columns and rows, where F is the intersection of (r ´ 1) A ‘ a and (r ´ 1) A ‘ a

inside E, and a1 and b1 denote their images in A. With attention on the missing upper
right corner, and activating the snake lemma, we infer that A/a1 » A/b1 . This implies
that a1 = b1 , and because all ideals are projective modules, we may conclude that there
are isomorphisms (r ´ 1) A ‘ a » F ‘ a1 » F ‘ b1 » (r ´ 1) A ‘ b. By induction it follows
that a » b. o

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: general valuations 395

Exercise 15.24 The approximation theorem. Assumethat A is a Dedekind ring with

fraction fiels K. Let ν1 , . . . , νr be distinct discrete valuations on K all being positive on A,
and let n1 , . . . , nr be integers. Show that there are elements x P K such that νi ( x ) = ni .
Hint: Appeal to the Chinese. M
Exercise 15.25 Let a and b be any two ideals in A. Show that a ‘ a » b ‘ b ´ 1 ´ 1 M
Exercise 15.26 Let k be an algebraically closed field whose carachteristic is not two,
and let A = k[ x, y] with constituting relation y2 = x ( x ´ a)( x ´ b); that is, the coordinate
ring of an affine elliptic curve on Weierstrass form. Show that any finitely generated
projective module over A is isomorphic to nA ‘ m for a uniquely defined maximal ideal
m. M

15.5 Appendix: General valuations

The origin of valuation theory, and the notions of valuations and a valuation rings, is
found far back in the history of mathematics, but its systematic study and axiomatic
development started in 1912 by the work of the Hungarian mathematician József
Kürschák whose aim was to bring Kurt Hensel’s theory of p-adic numbers on a solid
footing. The theory was soon substantially extended and improved by several people,
until it reached the form in which we know it to day, which mainly is due to Wolfgang
Krull. General valuation theory was an important part in the formulation of algebraic
geometry a la Zariski. In the book “Commutative Algebra” by Oscar Zariski and Pierre
Samuel, which was one of the bibles of the field in the 1960’s, general valuation theory
takes up about 125 pages. In the post-Grothendieck area general valuation rings went
József Kürschák
out of fashion, even though they are indispensable a few places and seem to have
(1864–1933)
gain popularity the latest years. The Noetherian ones, however, the discrete valuation
Hungarian
ring, persisted and are still flourishing. They are indispensable in both contemporary
mathematician
algebraic geometry and number theory.
Like most text on commutative algebra we concentrate on the dvr’s, but shall
use some space on the general ones here. If only for they being good examples of
non-noetherian rings that well illustrates the importance of the Noetherian hypothesis.
To put it like this: You best appreciate your own way of life by casting a glance on the
struggles of your neighbour.
(15.55) The axioms for general valuations are the same as those for discrete valuations,
the only difference being the group where the valuation assumes values, the so-called
˚ For any subset
value group. It will be any totally ordered group; that is, an additively written abelian
group G equipped with a total order of the elements. And, of course, the two structures S Ď G, the negative of
S is the subset
must be compatible so that α + γ ă β + γ whenever α ă β. ´S = t ´x | x P S u
The set of positive elements of G; i. e. those with α ą 0, form a submonoid
G + of G, called The positive cone, and which is such that* G + X ´G + = t0u and

10th January 2021 at 11:32am

Version 4.0 run 30
 396 dedekind rings

G + Y ´G + = G. In fact, giving such a submonoid is the same thing as equipping G

with a total order; indeed, declaring α ě β to mean that α ´ β P G + , we obtain a total
order compatible with the group law: symmetry holds as G + X ´G + = t0u, transitivity
because (α ´ β) + ( β ´ γ) P G + and G + is closed under addition, and compatibility
ensues from (α + γ) ´ ( β + γ) = α ´ β. Finally, either α ´ β or β ´ α belongs to G +
since G + Y ´G + = G, so the order is total.
Totally ordered groups are torsion free; indeed, from g ą 0 ensues ng ą 0 for all
natural numbers n by a straightforward induction argument.
Example 15.6 Every subgroup of a totally ordered group inherits a total order from
the surrounding group; so for instance, Z, Q and R and all their subgroups are totally
ordered. K
Example 15.7 Recall that if T is any totally ordered set the cartesian product T n is
The lexicographical endowed with the total order called the lexicographical order: an element x = ( x1 , . . . , xn )
order (den precedes an element y = (y1 , . . . , yn ) (that is, x ą y) if at the first place i where x and y
leksikografiske
ordningen) differ, xi is greater than yi . When T is the ordinary alphabet, this order is the one we
know well from dictionaries, hence the name. K
General valuations (15.56) This paves the way for the definition of a general valuation ν on a field K with
(generelle valuasjoner) value-group G. It is a surjective map ν : K˚ Ñ G that abide by the two rules

o ν( xy) = ν( x ) + ν(y);

o ν( x + y) ě min(v( x ), ν(y)),

whenever the sum x + y is non-zero. The first requirement can of course be phrased as
ν being a group homomorphism. Notice that the minimum in the second requirement
is meaningful since G is totally ordered, and just as for discrete valuation, one finds
that ν(1) = 0 because ν(1) = ν(12 ) = 2 ¨ ν(1), and consequently it holds true that
ν( x´1 ) = ´ν( x ) for all x P K˚ .
Not all texts require the map ν to be surjective, and in fact ν’s that are not surjective
arise naturally in several contexts; however, replacing G by the image of ν brings such a
situation in accordance with our convention.
Equivalent valuations Two valuations ν and ν1 on a field K whose respective value groups are G and G1 ,
(ekvivalente valusjoner) are said to be equivalent if there is an isomorphism θ : G Ñ G1 of ordered groups such
ν 4G that ν1 = θ ˝ ν.
K˚
»

θ
 Valuations and valuation rings
ν1 *
G1 Recall that a domain A with fraction field K is called a valuation ring if it has the property
that for any member x of K either x or its inverse x´1 belongs to A. Mimicking what
we did for discrete valuations, we associate a valuation ring to any valuation v on K

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: general valuations 397

and through the definition

A = t x P K | v( x ) ě x u.
The two axioms yields that A is a ring, it will be closed under multiplication by the first
and under addition by the second, and since G is totally ordered, it will be a valuation
ring (either v( x ) or 0 is the smaller, and in the former case, ´v( x ) ě 0).
Notice that the group A˚ of units in A is given as A˚ = t x P A | v( x ) = 0 u; indeed,
an element x P A is precisely a unit when x´1 P A; in other words, precisely when
v( x ) ě 0 and v( x´1 ) = ´v( x ) ě 0. The group of units is thus the kernel of v, and
being surjective by assumption, v induced a group isomorphism K˚ /A˚ » G. The
elements of K˚ are ordered by divisibility; that is x dominates y when x|y, or in other
words when xy´1 P A. The valuation respects this order in that v( x ) ě v(y) precisely
when xy´1 P A, and the isomorphism v induces between K˚ /A˚ and G will be an
isomorphism of ordered groups if K˚ /A˚ is equipped with the induced order. Note as
well that the the quotient K˚ /A˚ is an abelian group whose elements are the principal
fractional ideals—the group operation corresponds to multiplication of ideals
(15.57) Every valuation gives rise to a valuation ring, and the converse is true as well,
Indeed, the the quotient K˚ /A˚ is an abelian group whose elements are the principal
fractional ideals—the group operation corresponds to multiplication of ideals—and the
set of fractional ideals are totally ordered precisely when A is a valuation ring: the set
of principal ideals is totally order under inclusion according to Proposition 15.4, and
this implies that the set of fractional ideals is totally ordered as well (see Exercise 15.3).
So if v is a given valuation, the isomorphism K˚ /A˚ » G respects the orders and is
an isomorphism of ordered groups. The canonical projection π : K˚ Ñ K˚ /A˚ will be
a valuation (the group K˚ /A˚ is written multiplicatively and one has to switch to an
additive notation). So, we have proven most of the following:

Proposition 15.58 (Valuations vs valuation rings) Let K denote a field. Every val-
uation ring in K arises as associated to a valuation. The valuation is, up to equivalence,
unambiguously determined by the subring.
Proof: What remains to do, is to verify that the valuation is determined up to equiv-
alence by the valuation ring it determines. But given a valuation ν with ring A, the
kernel ker ν equals A˚ , and by the Isomorphism Theorem for abelian groups it factors
through the canonical map π : K˚ Ñ K˚ /A˚ showing that π and ν are equivalent. o
K˚
ν π
Ideals in valuation rings
 
(15.59) One may wonder what kind of subsets of G the images v(a) of the ideals G
» / K˚ /A˚
are. For instance, the image v( A) is G + , and the image v(m) of the maximal ideal is θ

G + zt0u = t α P G | α ą 0 u. In general the answer is that they are the so-called final Final segments
(sluttlige segmenter)

10th January 2021 at 11:32am

Version 4.0 run 30
 398 dedekind rings

segments contained in G + : subsets Γ Ď G + that are closed from above; i. e. if α P Γ and

β ě α, then β P Γ. There is no hocus pocus about this: if x P A, it holds that v( x ) ě 0,
so that v( xy) = v( x ) + v(y) ě v(y) for all y; hence if S is a final segment and v(y) P S,
it ensues that v( xy) P S. Moreover, one verifies painlessly that complements in G + of
final segments are characterized by the property that α P T implies γ P T for all γ ď α.
Some of the final segments have a least element and are shaped like ∆α = t β | β ě
α u, but many do not—just think of open intervals in the rationals Q. We shall call
Principal segments segments with a least element principal. Obviously principal segments correspond to
(hovedsegmenter) principal ideals: if x P A is an element with v( x ) = α, then v(( x )) = ∆α .
(15.60) The next pertinent question is “what are the prime ideals?” The complement of
a prime ideal p is closed under multiplication, which means that the complement T of
ν(p) in G + is closed under addition; hence the complement is a monoid, and in fact, it
is the positive part of a subgroup in that N = T Y t0u Y ´T is a subgroup (the point is
that G is totally ordered so that if α P T and ´β P ´T either α ě β and α + (´β) P T or
β ě α and α + (´β) = ´( β ´ α) P ´T) Additionally T has the property that γ ď α lies
in T when α does. This inspires the following definition: a proper subgroup N Ď G is
Isolated subgroups called isolated if elements that are squeezed between two elements from N, belong to N;
(isolerte undergrupper) i. e. if α ď γ ď β and α, β P N, then γ P N.

Proposition 15.61 Let ν be a valuation on K with value group G and let A be the correspond-
ing valuation ring. For the correspondence S ÞÑ ν´1 (S) between subsets of G + and A the
following hold:
i) Ideals in A correspond to final segments in G + ;
ii) Principal ideals correspond to principal final segments in G + ;
iii) Prime ideals correspond to complements in G + of isolated subgroups.

Proof: Let α = ν( x ) and β = ν(y). The crucial observation is that ν(yx´1 ) = β ´ α so

that yx´1 lies in A if and only if β ě α. We infer that, for a subset S Ď G + , it holds that
S is a final segment if and only if ν´1 (S) is closed under multiplication with elements
from A; that is, ν´1 (S) is an ideal. o
Exercise 15.27 Show that the prime ideals correspond to final segments whose
complement in G + are submonoids; that is, the complements are closed under addition.
M
Exercise 15.28 Let G be a totally ordered group. Recall that a subgroup N Ď G is
Isolated subgroups called isolated if elements that are squeezed between two elements from N, belong to N;
(isolerte undergrupper) i. e. if α ď γ ď β and α, β P N, then γ P N.
a) Prove that kernels of homomorphisms of ordered groups are isolated.
b) Prove that If N is an isolated subgroup, then N X G + is a monoid whose
complement in G + is a final segment.

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: general valuations 399

c) The ordered group G is called Archimedean if for any two elements α and β
there is an integer so that nα ě β. Prove that an Archimedean group has no
non-trivial isolated subgroups.
d) Assume that A is the valuation ring associated with a valuation whose value
group is a subgroup of the reals R. Show that dim A = 1.
M

Every totally ordered group appears as a value group

(15.62) With the construction of the monoidal algebras associated to a monoid in
Exercise 1.19 on page 25 is rather straight forward to exhibit examples of valuation
rings with any given totally ordered group as value group. This shows the extreme
diversity of valuations rings.
Let G be a totally ordered group and let k be a field, let k( G ) denote the fraction field
of the monoidal algebra k[ G ]. The monoidal algebra k[ G ], consists of elements which are
ř
finite k-linear combination f = γPG aγ x γ . For such f define ν( f ) = mint γ | aγ ‰ 0 u,
which is meaningful as G is totally ordered and every finite subset has a least element.
We may expressed f as
ÿ
f = αν x ν + aγ x γ
γąν

with ν)ν f and with aν ‰ 0. It is fairly easy to verify that this is a valuation: let g be
another element and write g as
ÿ
g = aµ x µ + bγ x γ
γąµ

Then
ÿ 1
f g = a ν bµ x ν + µ + a γ bγ 1 x γ + γ

where both γ ą ν and γ1 ą µ so that γ + γ1 ą ν + µ. and hence ν( f g) = ν( f ) + ν( g).

As to the second axiom, expand the sum f + g i terms of the x γ to get
ÿ
f + g = a ν x ν + bµ x µ + cγ x γ

where the sum extends over γ with γ ą min(ν, µ). If ν and µ are different, say ν ă µ,
if follows tat v( f + g) = ν0 min(v( f ), v( g)), and in case ν = µ the two first terms
might cancel, but in any case v( f + g) ě ν = min(v( f ), v( g)). The final step is to
extend v to the fraction field of K [ G ] by the usual procedure from Exercise 15.1 setting
v ( f g ´1 ) = v ( f ) ´ v ( g ) .

10th January 2021 at 11:32am

Version 4.0 run 30
400 dedekind rings

Valuation rings with value group Zn

For each integer i with 1 ď i ď n we let Ni be the subgroup of Zn of elements whose
i first coordinate vanish; that is, the set of elements of the form (0, . . . , 0, xi+1 , . . . , xn )
and for consistency, we let N0 = Zn .

Proposition 15.63 The Ni ’s are the only isolated subgroups of Zn . A valuation ring having
Zn as value group is of Krull dimension n.

Proof: The second statement follows because the subgroups Ni form a chain of length
n and because the prime ideals in a valuation ring are in a one-to-one order reversing
correspondence with the isolated subgroups of the value group.
To prove that the Ni ’s are the only isolated subgroups, it suffices, by induction on n,
to show that the proper ones are all contained in N1 . But this is quite clear: if N is an
isolated subgroup, the smallest non-negative first coordinate x1 of any element in N
must be zero since any element in Zn having a positive first coordinated less than x1 ,
would lie in N as N is isolated. o
Example 15.8 Concrete examples of valuations with value group Zn can be the following
valuations on the rational function field k( x1 , . . . , xn ). It is is just a specific instance
of the monoidal construction in Paragraph 15.62 above: As above, the construction
is most smoothly performed on the monoidal algebra An = k[ x1 , x1´1 , . . . , xn , xn´1 ] for
subsequently to be extended to the fraction field k( x1 , . . . , xn ). Elements f in An are
presented as finite sums f = α aα x α , with aα P k and α P Zn , and the valuation is given
ř

as v( f ) = mint α | aα ‰ 0 u. K
ˇ Exercise 15.29 Let A be a valuation ring with value group Zn . Show that the maximal
ideal is principal. M

α2

α1

Example 15.9 Ideals in a two-dimensional valuation ring: It is illustrative to study the ring
A2 and its ideal structure in more more detail. Exploring the ideal structure of A2
amounts to exploring the segment-structure of the positive cone of Z2 . In the figure we

10th January 2021 at 11:32am

Version 4.0 run 30
 appendix: general valuations 401

have depicted part of Z2 , with the positive cone drawn with red dots; remember that
Z2 is ordered such that elements increase upwards and to the right.
There are two types of proper final segments in Z2 . Apart from the principal ones,
there is one segment ∆n = t α | α1 ě n u for each integers n, having as elements the grid
points in the region of the plane to the right of the vertical line α1 = n. If ∆ is a proper
non-principal segment the first coordinate of elements in ∆ will be bounded below,
and hence there is smallest first coordinate n. Either all integers appear as a second
coordinates of members of ∆ having first coordinate n, and ∆ = ∆n , or there is a least
second coordinate, say m. But then ∆ will be the principal segment defined by (n, m)

α2 α2

α1 α1

A non-principal segment A principal segment

The subset of Z2 with α1 = 0; that is, the α2 -axis, is an isolated subgroup as the
inequality
(0, α2 ) ď (γ1 , γ2 ) ď (0, β 2 )
trivially implies that γ1 = 0. And it is the sole proper and non-trivial isolated subgroup.
Indeed, assume the subgroup N Ď Z2 is isolated and has an element γ P N whose first
coordinates is non-zero. Since the first coordinate dominates the ordering, there is for
any α P Z2 an integer n so that nγ ě α, and it follows that α P N as N is isolated.
In addition to the maximal and the zero-ideal there is one other prime ideal in A,
namely the one corresponding to the segment ∆1 . Hence dim A = 2. The other ideals
α
are the principal ones ( x1 1 x2α2 ), with α = (α1 , α2 ) in the positive cone, and the ones
corresponding to the segments ∆n with n ě 2. These, and ∆1 as well, are generated by
any infinite subset of t x1n x2´m | m P N u. Notice that the maximal ideal is generated by
x2 . K

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 16

Hilbert functions

When investigating a graded algebra R, or a graded module M for that matter, which is
finitely generated over a field k—a frequent activity among algebraic geometers—an
extremely powerful tool is the so-called Hilbert function of R. This is simply the function
defined as h R (n) = dim Rn ; that is the dimension of the grade piece of R consisting
of homogeneous elements of degree n. Or in case of a mosul h M (n) = dim Mn The
homogeneous piece Mn is a vector space over k and turns out to be of finite dimension
whenever R is Noetherian and M is finitely generated over R, which makes the definition
is legitimate. There is also the notion of of the Hilbert series of R, the generating series
of the Hilbert function; that is, PM (t) = i dimk Mi ti called the Hilbert–Poincré series
ř

of M.
The prototypical example would the the polynomial algebra R = k[ x1 , . . . , xr ] in n
variables. The space Rr of hogomeneous forms of degree n has the monomials of degree
n as a basis, and it is is an exercise in elementary combinatorics to show that dim Rn is
the binomial coefficient (n+ r ´1
r ´1 ) .
The Hilbert functions have the virtue of behaving like a polynomial (with rational
coefficients) for large values of the variable, and the coefficients of this polynomial
furnish invariants of the algebra. Of course, by their very definition, these polynomials
takes integral values on integers (at leat for large arguments); such polynomials are
called numerical polynomials.
(which will be integral after a simple change). The dimension of a varietriy, the
degrees of a projective varities, and the famous genus of a curve, are all of this type.
One may also define the nultiplicity of a singular point (+++ other invariants) in this
way.
The Riemann–Roch theorem iis a topological interpretation of some of this invariants.
One comes across graded rings in several corners of algebraic geometry. The most
frequently met are the coordinate rings of cones. These can be of several types, the most
prominent ones being cones over projective varieties, and graded rings and their Hilbert
 404 hilbert functions

polynomials are of paramount importance to the projective geometers

An other place is the local study of a variety and the local multiplicities. Contrary
to manifolds varieties may acquire singularities, and they do not have tangent space
but so-called tangent cones. Heuristically such a cone at a point P is union of all lines
emanating from the point and being tangent to the variety at P. There is an algebraic
intrinsic definition of the tangent cone which is a graded ring associated with any local
ring containing much local information.
Example 16.1 Th conormal cone: To a given ideal a in a ring A one may associate a graded
ring whose underlying abelain groups is

Gra = A/a ‘ a/a2 ‘ . . . = ai /ai+1

à
i ě0

That there is a multiplication needs to be explained, so let a P as and b P at be

elements and consider the classes [ a] and [b] in respectively as /as+1 and at /at+1 . The
product ab lies in as+t and the salient point is that [ ab] does not depend on the choice
of representatives a and b: if a P as+1 and β P at+1 , one finds ( a + α)(b + β) =
ab + aβ + bα + αβ, where the three terms involving α or β belong to as+t+1 . This gives
the product of two homogeneous elements, which is extended by linearity to a product
of arbitrary elements, and the ring axioms follow straightforwardly.
In the particular case of a maximal ideal m, the cone Grm A is called the cotangent
cone of A at m. K

16.1 Numerical polynomials

numerical function The Hilbert functions are a priori so-called numerical function, they accept integral
(nummeriske arguments and return integral values; i. e. so they are function h : Z Ñ Z. However,
funksjoiner)
they are not any such functions, but turn out have the very special property of behaving
as a polynomial for sufficiently large arguments: there is a polynomial P P Q[t] so that
numerical polynomial h(n) = P(n) for n ąą 0. The polynomial P will necessarily* be a so-called numerical
(nummeriske polynomial in that it takes integral values on the integers.
polynomer)
(16.1) A numerical function h has a “discrete derivative” which is defined as

˚ Convince yourself ∆h(t) = h(t) ´ h(t ´ 1).

that this is so
It shares the property with the usual calculus-derivative that ∆h = 0 is equivalent with
h being constant. There is also an analogue to the integral, in that it holds true that
r
ÿ
h (r ) ´ h ( s ´ 1) = ∆h(t)
s

10th January 2021 at 11:32am

Version 4.0 run 30
 numerical polynomials 405

for all integers r and s with r ě s. Primeordial prototypes of numerical polynomials are We already
the binomial coefficients which we remind you are given as encountered the

t+n

(t + n)(t + n ´ 1) . . . (t + 1) numerical
= , (16.1) polynomials in
n n!
Exercise 9.8 which
where n is any non-negative integer. It is classical that they assume integral values
asked you to show
on the integers, and moreover, the binomial coefficient in (16.1) is of degree n, and its
that the subring
leading coefficient equals 1/n!.
Int(Z) of Q[t]
As the usual calculus derivative does, the discrete derivative lowers the degree of a
they form is not
polynomial by one; indeed, the Binomial Theorem yields
Noetherian.
tn ´ (t ´ 1)n = tn ´ (tn ´ ntn´1 + . . .) = ntn´1 + . . . .

We also observe that the leading coefficient picks up a factor n, so writing the leading
coefficient of a degree n polynomial P as a0 /n!, the leading coefficient of ∆P(t) will be
a0 /(n ´ 1)!.

A part of the graph of (t+7 7), slightly scaled to better show the behavior.

(16.2) The binomial coefficients form a basis for the ring Int(Z) of numerical polyno-
mials which is well adapted to the discrete derivative operator ∆ as the well known
identity from Pascal’s triangle holds true:
   
t+n t+n´1
∆ = . (16.2)
n n´1
Moreover, a polynomial assuming integral values at all sufficiently large integers, is a
numerical polynomial; we have:

Proposition 16.3 The following two assertions hold true:

i) A polynomial P(t) is a numerical polynomial if and only if it assumes integral values
for sufficiently large integers; i. e. P(n) P Z when n ąą 0;
ii) A numerical polynomial P of degree n has a development
     
t+n t+n´1 t+1
P ( t ) = a0 + a1 + . . . + a n ´1 + an
n n´1 1
where the coefficients ai are uniquely defined integers and a0 ‰ 0.

10th January 2021 at 11:32am

Version 4.0 run 30
406 hilbert functions

Proof: One of the implications in i) is tautological; attacking the other one we assume
that P(n) P Z for n ě s. It follows that ∆P(t) = P(t) ´ P(t ´ 1) P Z for t ě s + 1, and
by induction on the degree we know that ∆P is a numerical polynomial (the claim
is obviously true for polynomials of degree zero, so that the induction may start).
Reintegrating, we find for any pair of integers with r ě s the equality
ÿ
P ( s ) = P (r ) + ∆P(t),
s + 1ď t ď r

in which the terms in the right hand sum all belong to Z. Moreover P(r ) is integral
when r is chosen sufficiently large, and thus we may conclude that P(s) P Z.
The proof of ii) also relies on induction on the degree, and this time the crucial
observation is Pascal’s identity (16.2). Obviously a polynomial of the form as in ii)
is numerical and of the degree n. To prove the converse, assume h is a numerical
polynomial of degree n; then the derivative ∆h will be one of degree n ´ 1, and by
induction we may express ∆h as
 
ÿ t+i
∆h(t) = ai .
i
0 ď i ď n ´1

Consequently the difference

 
ÿ t+i+1
h(t) ´ ai
i+1
0ďiďn´1

has a vanishing derivative and is therefore constant; with an being this constant the
claim follows. o

Proposition 16.4 A numerical function h(t) equals a numerical polynomial of degree n for
for t ąą 0 if and only if the discrete derivative ∆h(t) equals a numerical polynomial of degree
n ´ 1 for t ąą 0.
Proof: If ∆h(t) equals a numerical polynomial for t ąą 0, this polynomial is of the
form as in ii) in Proposition 16.3, and the argument just given shows that h(t) is of the
same form for t ąą 0. o
Exercises
(16.1) Show that
  
t+n n n+1
= t /n! + /n! ¨ tn´1 + . . .
n 2

(16.2) Show that (t+n n) is the unique polynomial that vanishes at the negative integers
between ´1 and ´n and assumes the value one at zero.
M

10th January 2021 at 11:32am

Version 4.0 run 30
 numerical polynomials 407

Positively graded rings

In the introduction to this chapter we mentioned graded rings with R0 a field, say k,
and graded R-modules with each graded piece Mn of finite dimension over k; the point
was that these modules possess the Hilbert functions h M (n) = dimk Mn . The conditions
may certainly be relaxed: when each homogenous piece Mn is of finite length over R0 ,
the module has a Hilbert function defined as h M (n) = ` R0 ( Mn ). The current subsection
is devoted to describing large workable classes of graded rings R and graded modules
M fulfilling this.
(16.5) Save for a few very special cases, graded rings with non-zero elements both
of positive and of negative degree do not have very informative Hilbert Functions, if
having one at all, and they are not interesting in the present context. So we shall confine
our study to the so-called positively graded rings: a graded ring R is said to be positively Positively graded rings
graded if it has no non-zero homogeneous elements of negative degree; in other words, (positivt graderte
ringer)
we require that Ri = 0 for i ă 0. In that case the additive subgroup
à
R+ = Ri
i ą0

generated by the homogeneous elements of positive degree will be an ideal in R, which

is called the irrelevant ideal*; indeed, homogeneous elements being of positive degree The irrelevant ideal
multiply R+ into R+ since deg x ¨ y = deg x + deg y ě deg y for all x. The graded rings (det irrelevante idealet)

Gra A constructed in the introduction (Example 16.1) are all positively graded; they are ˚ This is another

even generated by elements of degree one; namely by the classes of elements generating instance of highly
delusive naming in
a. mathematics, albeit
ˇ Exercise 16.3 The exercise is an illustration of the restrictions one imposes on graded having irrelevant in
the name, these ideals
domains with non-zero elements of both positive degree and of negative degree when
are seriously relevant.
requiring the degree zero part to be a field. Let R be one of the kind and assume The reason for this
R0 = k is a field. Show that R = k [w, w´1 ] for some homogenous element w P R. apparent malpractice is
found in projective
Hint: Show first that each homogeneous element is invertible. M geometry; apices of
(16.6) Most finiteness results about positively graded rings are rooted in the following cones over projective
varieties are invisible
observation which shows the relevance of the irrelevant ideal:
to the projective
geometers eye.
Lemma 16.7 Let R be a positively graded ring and assume that x1 , . . . ., xr are homogeneous
elements that generate the irrelevant ideal R+ . Then the xi ’s generate R as an algebra over R0 ;
that is, R = R0 [ x1 , . . . , xr ].
Proof: We shall show, by induction on the degree of x, that each homogeneous element
x of R of positive degree belongs to R0 [ x1 , . . . , xr ]; this will suffice since every element is
the finite sum of its homogeneous constituents. So we assume that all elements of degree
ř
lower than deg x lie in R0 [ x1 , . . . , xr ]. Now x P R+ , and hence x = ai xi , for elements
ai P R, and replacing the ai ’s by their homogeneous components, we may assume that

10th January 2021 at 11:32am

Version 4.0 run 30
408 hilbert functions

ai ’s are homogeneous and comply with the constraint deg ai + deg xi = deg x. Thus
deg ai ă deg x; by induction we infer that ai belongs to R0 [ x1 , . . . , xr ], and we are done.
o

Proposition 16.8 Let R be a positively graded ring.

i) Then R is Noetherian if and only if R0 is Noetherian and the irrelevant ideal R+ is
finitely generated.
ii) If R is Noetherian, the homogenous pieces Rn are all finitely generated over R0 .

Note that the first statement appears remarkably strong; just one ideal, although special,
being finitely generated implies that all are. The reason behind, as the proof will show,
is that these rings will turn out to be quotients of polynomial rings over R0 and so
Hilbert’s Basis Theorem takes effect.
Proof: We begin with proving i). If R0 is Noetherian and R+ is finitely generated, in
fact it is generated by finitely many homogeneous elements, as just mentioned, Hilbert’s
Basis Theorem together with Lemma 16.7 closes the case.
So assume that R is Noetherian, the of course R+ is finitely generated as any other
ideal is, and we merely have to show that R0 is Noetherian. The crucial observation
is that for any ideal a in R0 one has aR X R0 = a (from which the claim follows by
ř
extending and recontracting chains). Indeed, let the sum ai xi with xi P a and ai P R
belong to R0 . Replacing the ai ’s by their homogeneous components, we may assume that
the ai ’s are homogeneous; furthermore, the terms whose degree is non-zero, add up to
zero, and can be discarded. Hence for each i it holds that 0 = deg ai + deg xi = deg ai .
The second statement follows readily: Let x1 , . . . , xr be homogeneous generators of
α
R+ . A monomial x1 1 ¨ . . . ¨ xrαr belongs to Rn precisely when α1 deg x1 + . . . + αr deg xr =
n and this equation has only finitely many solutions because deg xi ą 0 (each αi is
bounded by n/ deg xi ). Hence there only finitely many monic monomials in Rn , and
these generate Rn over R0 . o

The Hilbert function

The next proposition is an easy consequence of Proposition 16.8 above and describes a
large class of module having a Hilbert function.

Proposition 16.9 Let R be a positively graded Noetherian ring and M a finitely generated
graded R-module.
i) All the graded pieces Mn are finitely generated over R0 .
ii) If additonally R0 is Artinian, each Mn will be of finite length over R0 .

Proof: The lemma is true for R itself by Proposition 16.8, hence for all shifts R(d) (shift-
ing does not alter any algebraic property merely changes degrees), and consequently
À
for every finite direct sum i R(´di ). And if M is finitely generated, it is a quotient of

10th January 2021 at 11:32am

Version 4.0 run 30
 numerical polynomials 409

À
such a finite sum i R(´di ), and the graded pieces Md of M of degree d are quotients
À
of the graded pieces i R(´di )d . o
(16.10) So when R is Noetherian and R0 Artinian, we my define the important invariant
h M (n), the Hilbert function, of such finitely generated modules by putting Hilbert function
(Hilbert function)
h M ( n ) = ` R 0 ( Mn ) .

It is of course a numerical function of n and it has the all important property of being
an additive function on the category Grmod A , which allows one to calculate it in many
instances. This means that if

0 / M1 /M / M2 /0 (16.3)

is an exact sequence in the category GrmodR of finitely generated graded modules, then
the equality
h M ( n ) = h M1 ( n ) + h M2 ( n )

holds true. It ensues from the additive character of the length. Indeed, the maps in the
sequence 16.3 preserves homogeneous elements and degrees, and for each degree n the
sequence therefore induces the sequence

0 / Mn1 / Mn / Mn2 / 0.

which is exact.
(16.11) As with any additive function, if C‚ is a bounded complex—so that Cn = 0 for
|n| sufficiently large—of finitely generated R-modules it holds that
ÿ ÿ
(´1)i hCi (t) = (´1)i h H i (C‚ ) (t).
i i

Examples
(16.2) Recall that the Hilbert function of the polynomial ring Rr = k[ x0 , . . . , xr ] over a
field k is given as
$
&0 when t ă 0;
h Rr ( t ) =
%(r+t) when t ě 0.
r

The first claim is trivial. In case you don’t know the second formula already: it follows
immediately by induction on r and Pascal’s identity (16.2). Indeed, the short exact
sequence
xr
0 Rr Rr R r ´1 0

10th January 2021 at 11:32am

Version 4.0 run 30
 410 hilbert functions

yields that ∆h Rr (t) = h Rr´1 (t), which shows that the claim is true up to a constant, and
evaluation at t = 0 ensures that the constant is zero.
(16.3) The shifts Rr (´m) of the polynomial ring Rr = k[ x0 , . . . , xr ] has the Hilbert
function $
&0 t ă m;
h Rr (´m ) ( t ) =
%(r+t´m) t ě m,
r

since the graded pieces of Rr (´m) are given as Rr (´m)t = ( Rr )t´m . Note that h Rr (´m) (t)
is not a polynomial, but equals one for t ě m.
(16.4) Consider the principle ideal ( F ) Rr in the polynomial ring Rr = k[ x0 , . . . , xr ]
generated by a homogeneous form F of degree m. Multiplication by F induces a
homogeneous isomorphism between Rr (´m) and ( F ) since for an element a P Rr (´m)d
it holds that deg a = d ´ m, so consequently deg aF = d.
The classical short exact sequence is therefore an exact sequence of graded modules:

/ Rr (´m) / Rr / Rr / ( F ) Rr / 0,
µ
0

where the map µ is multiplication by F. Hence we find that

$
&0 if t<0;
’
’
h Rr / ( F ) Rr ( t ) = (r+r t) if 0 ď t ă m;
’
(r+r t) ´ (r+tr´m)
’
if t ě m.
%

One observes that the function h Rr /( F) Rr is a piecewise polynomial, and that for t ě m it
is equal to the polynomial
   
r+t r+t´m
χ Rr / ( F ) Rr ( t ) = ´ = mtr´1 /r! + . . .
r r

whose degree is one less than the Krull dimension of Rr /( F ) Rr . This illustrates the
˚ As we are soon to see, general feature of many* Hilbert functions: they are polynomials for large values of the
this holds for modules variable of degree one less than then Krull dimension of the module. We also observe
over graded rings
which are Noetherian that the leading coefficient, up to the factor 1/r!, equals the degree of F. In general this
and generated in coefficient will be of the form a/r! with a a natural number which, being a consequence
degree one.
of the standard from Proposition 16.3, holds for any numerical polynomial.
Graded algebras generated in degree one over fields all have a geometric incarnation,
which is a projective varieties, and projective varieties have a degree, which turns out to
be equal to the number a.
(16.5) The next examples is slightly more elaborated. It plays an important role in the
theory of plane curves where the result goes under the name of Bezout’s theorem.

10th January 2021 at 11:32am

Version 4.0 run 30
 numerical polynomials 411

The givens are two homogenous polynomials f and g in R = k[ x, y, z] without

common factors; their degrees are respectively n and m. The subsets V ( f ) and V ( g) of
A3 are cones over projective plane curves, and we intend to investigate their intersection.
It is represented by the cone V ( f , g) over the common zeros of f and g, and whose
algebraic incarnation is the homogeneous ring R/( f , g). The two curves should intersect
in finitely many points, so one suspects V ( f , g) to be finite; that is, the cone R/( f , g)
should be one dimensional. Moreover, one believes that “the number of common zeros”
should be the product nm of the degrees; or translated into properties of the Hilbert
function, it should holds that h R/( f ,g) (t) is constant for t ą 0 (i. e. of degree one less
than dim R/( f , g)) and the constant value should equal nm. And indeed, we shall prove
that the Hilbert function h R/( f ,g) (t) = nm for t ě n + m.
Then crucial point is the Koszul complex which we introduced in Example 5.9 on
page 145, and which under current cicumstances furnishes a free resolution of R/( f , g)
shaped like:
d2 d1
0 R(´n ´ m) R(´n) ‘ R(´m) R R/( f , g) 0 (16.4)

where d2 ( a) = ( ag, ´a f ) and d1 ( a, b) = a f + bg. Applying the additivity of the Hilbert

function, this exact sequenecs followed by a small trivial computation one finds that for
t ě n + m it holds that
       
t+2 t+2´n t+2´m t+2´n´m
h R/( f ,g) (t) = ´ ´ + = nm.
2 2 2 2
For values of t smaller than n + m not all four binomial coefficients do appear in the
expression for h R/( f ,g) ; remember that h R(´n) (t) = 0 for t ă n. Which ones do, depends
on the size of t compared to 0, n, m and n + m. In the figure below the graph of
h R/( f ,g) (t) is sketched for n = 16 and m = 31; It is a piecewise polynomial with two
infinite constant sectors and two quadratic parts separated a linear sector.

10th January 2021 at 11:32am

Version 4.0 run 30
412 hilbert functions

The Hilbert polynomial

The main result in this paragraph is that the Hilbert function associated with a graded
module M finite over a Noetherian ring A generated in degree one and with A0 Artinian,
is equal to a polynomial for large values of the variables. This polynomial is called the
Hilbert polynomial of M. It is a numerical polynomial whose coefficients are important
invariants of the graded module M. We shall not say much about these invariants, but
contend ourself to show that the degree equals the dimension of the support of M.
The most common application of the Hilbert polynomial is in cases when A is a
polynomial ring over a field with all the variables being of degree one. The hypothesis
that A be generated in degree one, is an essential one; a stupid example is obtained by
assigning the degree two to the variable x in A = k [ x ]. Then the Hilbert function h A (n)
vanishes for all odd numbers and therefore cannot be equal to a polynomial for large n.
(16.12) Recall that the dimension dim M of a module M finite over a Noetherian ring A
by definition equals the dimension of its support; that is, it is equal to dim A/ Ann M.
Moreover, dim M = 0 if and only if M is a module of finite length. Here comes the
result:

Theorem 16.13 Let R be a Noetherian graded ring with R0 being Artinian and M a finitely
generated graded A-module Assume that R is generated in degree one. Then the following hold
true:
i) The Hilbert function h M (t) equals a polynomial χ M (t) for t ąą 0;
ii) The degree r of χ M is equal to dim M ´ 1, and its leading coefficient is of shape d/r!
with d P N.

It is crucial that R be generated in degree one; a stupid example with R = k[t] where t is
given the degree two. Then Rn = 0 when n is odd, so the Hilbert function has infinitely
many zeros and can not equal a polynomial for large integers.
Proof: The proof goes by induction on dim M. The induction can begin because
dim M = 0 implies that the module M is of finite length and merely has finitely many
non-vanishing homogeneous parts, so that h M (t) = 0 for t ąą 0.
Each finitely generated module M has, as the Structure theorem for graded mod-
ules tells us (Proposition 10.41 on page 279), a finite ascending chain Mi of graded
submodules whose subquotients are of the form A/pi with the pi ’s being homogeneous
prime ideals. Taking the grading into account we arrive at a series of exact sequences in
GrMod A , one for each 0 ď i ď r, which all are shaped like:

0 M i ´1 Mi A/pi (mi ) 0.

Moreover, it holds that Mr = M and M0 = 0.

10th January 2021 at 11:32am

Version 4.0 run 30
 numerical polynomials 413

A successive application of the additivity of the Hilbert function yields the equality
ÿ
h M (t) = h A/pi (t + mi ).
1ďiďr

Since the the dimension of the support Supp M equals the maximum of the dimensions
dim A/pi of the subquotients, and since shifting a module merely affects the Hilbert
functions by translating the variable, it will suffice to show the proposition for quotients
A/p with p being a homogeneous prime ideal. To that end, pick any x P A1 not lying in
p (we may safely assume that dim A/p ą 0 so that p Ř A+ ) and form the exact sequence

0 A/p(´1) A/p A/p + ( x ) 0.

According to Krull’s Principal Ideal Theorem it holds that dim A/p + ( x ) = dim M ´ 1,
induction applies to A/p + ( x ) and ∆h A/p (t) = h A/p+( x) (t) equals a polynomial of
degree dim M ´ 1 for t ąą 0. Evoking Proposition 16.3 on page 405 we infer that
h A/p (t) is a polynomial of degree dim M for large values of t. o

Lemma 16.14 Let R be a Noetherian graded ring with R0 Artinian which is generated by r
elements of degree one. Then the Hilbert polynomial χ A (t) is of degree less than r unless R is
isomorphic to the polynomial ring in r variables over R0 , in which case deg χ A (t) = r.

Proof: Assume that the elements x1 , . . . , xr are elements od R1 that generates R over
R0 . Let X1 , . . . , Xr be variables and define a map φ : R0 [ X1 , . . . , Xs ] Ñ R by sending Xi
to xi . This is a map of graded rings which at the outset is surjective. It follows that
h R (t) ď (t+r r) for all t.
If the map φ is not injective, we may chose a non-zero homogeneous polynomial
F from its kernel. Then φ factors through B = R0 [ X1 , . . . , Xr ]/( F ), and consequently
h R (t) ď h B (t) for all t. According to Proposition 11.9 on page 293 it holds that
dim B ď r ´ 1, and by Proposition 16.13 above χ B (t) is of degree at most r ´ 1, so that
deg χ R (t) ď r ´ 1. o

This is the so-called Roman sur-

face, a realisation with self inter-
sections of the real avatar of the
Veronese surface in R3 . Mod-
ern technology not only produce
beautiful pictures of surfaces,
but with 3D-printers also in-
triguing models are made. This
one find at Mathematical Art
Galleries

10th January 2021 at 11:32am

Version 4.0 run 30
 414 hilbert functions

Examples
(16.6) Given a natural number d. In this example we consider the subalgebra A =
k[ud , ud v, . . . , uvd´1 , vd ] of the polynomial ring R = k[u, v] generated by all monomials
of degree d. It is a graded subalgebra, and one easily verifies that A decomposes as
À
A = i Rid ; that is, as the sum of the homogeneous pieces of R of degree a multiple
of d. Now, changing the degrees of the elements in A by factoring out d, the algebra
A will be generated in degree one, and by the decomposition of A above we find
χ A (t) = χ R (dt) = dt + 1.
More generally, let R be the polynomial ring R = k [u0 , . . . , un ] and consider the
˚ These algebras are subalgebra* A of R generated by all the monomials of degree d; i. e. those shaped like
called Veronese
ř
uα = uα0 ¨ . . . ¨ urαr with i αi = d. Just as in the previous case, the decomposition of A
algebras, since they are À
the homogeneous into homogeneous pieces appears as A = i Rid , form which ensues the identity
coordinate rings of the  
so-call Veronese dt + n
varieties. χ A (t) = χ R (dt) = = dn tn /n! + . . . .
n

We conclude that dim A = n + 1 and that the degree of the corresponding Veronese
variety is dn .
(16.7) Let us examine the algebra R = k[ x, y, x ] with constituting relations x2 ´ y2 =
x2 ´ z2 = 0 with standard grating. The plolynomial ring k[, Y, Z ] is factorial and the
polynomials X 2 ´ Y 2 and X 2 ´ Z2 are without common factors, so by Example 16.5
the Koszul complex is exact and the Hilbert polynomial is constant and equal to the
product of the two degrees; that is, it is equal to four.
It is worthwhile examining the situation more closely. The relations start to be
visible for the Hilbert functtion In degree two: since all the squares of the variable are
equal, R2 is generated by x2 , xy, xz, yz, and more generally, the same reasoning gives
that x n , x n´1 y, x n´1 z, x n´2 yz generate Rn .
This immediately gives that h R (t) ď 4 for t ě 2, but to prove equality with this
approach, one would also need to show they are linearly independent, which amounts
to seeing that m = ( x, y, z) is not an associated prime. Indeed, if 0 = αx n + βx n´1 y +
γx n´1 z + δx n´2 yz = x n´2 (αx2 + βxy + γxz + δyz), there will be an element killed by a
high power of x and hence by high powers of y and z as well. It is not to hard to give an
ad hoc argument for this not being the case, but with the Koszul complex it comes for free.
In fact, one may turn the argument around, and knowing the Hilbert function h R (t),
conclude that m is not associated. The primary decomposition of ( X 2 ´ Y 2 , X 2 ´ Z2 ) is
thence

( X 2 ´ Y 2 , X 2 ´ Z2 ) =
(16.5)
= ( X ´ Y, X ´ Z ) X ( X ´ Y, X + Z ) X ( X + Y, Y ´ Z ) X ( X + Y, X + Z ).

10th January 2021 at 11:32am

Version 4.0 run 30
 numerical polynomials 415

Localized at a prime ideal p not containing two of the linear factors, the inclusion
(16.5) becomes an equality, and if three of them belongs to p, one easily checks that
p = ( x, y, z).
(16.8) A determinantal variety and the Hilbert–Burch complex: Our next example is about
the ideal generated by the maximal minors of the generic 3 ˆ 2-matrix
 
x00 x10
 x01 x11 
 
x02 x12

where the xij ’s are variables, so that M has entries from the polynomial ring R =
k [ xij |0 ď i, j ď 2]. The maximal minors of M are the 2 ˆ 2-minors, and the ideal we shall
explore will be


a = x01 x12 ´ x02 x11 , x02 x10 ´ x00 x12 , x00 x11 ´ x01 x10 . (16.6)

When k is algebraically closed, the closed points of the space A6k = Spec k[ xij ]
parametrize the 3 ˆ 2-matrices with entries from k, and those lying in V (a) constitue the
locus where the matrix drops rank. For general k the same applies to the set of k-points
A6 (k), and quite generally any 3 ˆ 2-matrix N = (bij ) with coefficients in a k-algebra B
is obtained from M by changing base by the map R Ñ B, that sends xij to bij .
We intend to compute the Hilbert function of A = R/a, which gives us the opportu-
nity to introduce the Hilbert–Burch complex. This is the complex

C‚ : 0 / 2R(´3) M / 3R(´2) D /R

where D is the 3 ˆ 1-matrix

D = x01 x12 ´ x02 x11 , ´( x00 x12 ´ x02 x11 ), x00 x11 ´ x01 x10 .

Obviously the images of D is equal to our ideal a, and that D ¨ M = 0 is seen by

successively enlarging M by the each row and then expanding the determinant. For
instance, expansion along the first row gives
 
x00 x01 x02
0 = det  x00 x01 x02  = x00 ( x01 x12 ´ x02 x11 ) ´ x01 ( x00 x12 ´ x02 x11 )+
 
x10 x11 x12
x02 ( x00 x11 ´ x01 x10 ),

and in a similar way, expanding M by adding t the other row and expanding along it,
gives the second equation required for D ¨ M=0 to hold true. We have by that established

10th January 2021 at 11:32am

Version 4.0 run 30
 416 hilbert functions

that 16.7 is a complex. Most often one the complex is displayed with the cokernel of D,
wich equals A, included:

0 / 2R(´3) M / 3R(´2) D /R /A / 0. (16.7)

The salient point is that the sequence (16.7) is exact. Given this, we easily find the
Hilbert polynomial:
     
t+5 t+3 t+2 1 5
χ A (t) = ´3 +2 = t3 + 2 t2 + t + 1
5 5 5 2 2

or expanded in the basis of the binomial polynomials

   
t+3 t+2
χ A (t) = 3 ´2 .
3 2

Note that this one of rare cases that Hilbert polynomial and the Hilbert function coincide
for all t ě 0.

Lemma 16.15 The Hilbert–Burch complex resolves A; that is, the sequence 16.6 is exact.

Proof: That M is injective follows by passing to the quotient field K of R; Then the
minors of M will be invertible, and ker MbR K = 0. But the kernel of M is torsion free
being contained in a free R-module, hence it vanishes.
So, the only hot spot is the middle homology H 1 (C‚ ) = ker D/ im M. The proof it
˚ The use of x is
00 vanishes has two steps. Firstly, we contend that the support of H 1 C‚ is contained* in
´1
dictated by pure V ( x00 ); that is, H 1 (C‚ )bR R[ x00 ] = 0. So assume that ( a, b, c) P 3R satisfies the relation
convenience. By t
elementary row and D ¨ ( a, b, c) = 0; a totally elementary manipulation gives
column operations each
of the variables, in fact ( x00 c ´ x02 a)( x11 x00 ´ x10 x01 ) + ( x00 b + x01 a)( x12 x00 ´ x10 x02 ) = 0
any linear combination
of the variables, may be
brought to the upper and hence since the two involved minors are irreducible and the polynomial ring is a
left corner of M, so ufd, there is a polynomial d P R with
they are all on equal
footing.
x00 b = ´ d( x11 x00 ´ x10 x01 ) + x01 a
x00 c =d( x12 x00 ´ x10 x02 ) + x02 a

Then one finds that    

x00 x10 ! x00 a
 ´dx10 + a
 x01 x11  =  x00 b
  
dx00
x02 x1 x00 c
and since x00 is inveritble on R x00 , we are done.

10th January 2021 at 11:32am

Version 4.0 run 30
 numerical polynomials 417

´1
This shows that H 1 (C‚ )bR R[ x00 ] = 0, and consequently H 1 (C‚ ) is killed by a high
power of x00 . The sequence
x00
0 / C‚ / C‚ / C‚ /x00 C‚ /0

induces a long exact sequence of homology modules, the part of which that interests us
is
H 2 (C‚ /x00 C‚ ) / H 1 (C‚ ) x00 / H 1 (C‚ )

The point is that H 2 (C‚ /x00 C‚ ) = 0, so that multiplication by x00 is injective But we
already observed that a power of x00 kills H 1 (C‚ ), and consequently H 1 (C‚ ) = 0. To
see that H 2 (C‚ /x00 C‚ ) = 0 observe that killing x00 gives us a the matrix
 
0 x01
M1 =  x01 x11 
 
x02 x12
with coefficients in the polynomial ring R1 = R/x00 R = k [ x01 , x02 , x10 , x11 , x12 ], and M1
is is generically of rank 2, hence the map M1 : 2R1 (´3) Ñ 3R1 (´2) is injective. o
K
ˇ Exercise 16.4 The Hilbert–Burch complex C‚ is a meaningful construct for any 3 ˆ 2-
matrix M = ( aij ) with coefficient from any ring R. As in the example, let a be the ideal
generated by the 2 ˆ 2-minors of M and let A = R/a. Mimicking relevant parts of the
approach in the example, show the following claims:
a) If a = R, then the Hilbert–Burch complex is exact.
b) If a contains a non-zero divisor x, then H 2 (C‚ ) = 0.
c) If a contains a regular sequence of length two, then the Hilbert–Burch complex
is a resolution of A = R/a.
M
Exercise 16.5 Let a be the ideal a = ( x1 x3 ´ x22 , x0 x3 ´ x1 x2 , x0 x2 ´ x12 )
in the polynomial
ring k[ x0 , x1 , x2 , x3 ] and let A = R/a equipped with the standard grading. Determine
the Hilbert polynomial of A. The ring A is the coordinate ring of the cone over the
so-called twisted cubic curve. M
The Hilbert-Poincaré series
There is another way of encoding the sizes of the graded pieces of graded modules
by forming their generating series, the so called Hilbert–Poincaré series Contrary to the
approach relying on numerical polynomials, which requires the graded ring to be
generated in degree one, the Hilbert–Poincaré series is usefull for (Noetherian) postively
graded rings without further limitations on the degree of teh genators.
(16.16) So let R a positively graded Noetherian ring with R0 artinian and let M be a
finitely generated R-module. The graded pieces Mn are the of finite length over R0 , and

10th January 2021 at 11:32am

Version 4.0 run 30
 418 hilbert functions

we may form the formal Laurent series

ÿ
P( M, t) = ` R 0 ( Mn ) t d .
n PZ

As M finiteenerted over R, which is positively graded, M will be bounded below, and

the series has only finitely many non-zero terms with negative exponents, in other
Hilbert–Poincaré series words, it is a Laurent series; the Hilbert–Poincaré series of M.
(Hilbert–Poincaré (16.17) The Hilbert–Poincaré series are clearly additive invariants; indeed, an exact
rekken)
sequence
0 / M1 /M / M2 /0

of graded M-modules is exact degree by degree, and hence P( M, t) = P( M1 , t) +

P( M2 , t). Moreover, they are well behaved with respect to the shift operators:

Lemma 16.18 P( M(m), t) = t´m P( M, t)

` R0 ( M ( m ) n ) t n =
ř
Proof: It holds true that M (m)n = Mn+m so that P( M(m), t) =
n 1
ř
n ` R0 ( M n+m ) t , and changing the summation variable by putting n = n + m one
1
obtains ` R0 ( Mn1 )tn ´m = P( M, t)t´m .
ř
o
(16.19) The next theorem describes the over all structure of the Hilbert–Poincaré series
of a finitely generated graded module over Noetherian and positively graded ring. They
turn out to be rational function with poles at certain roots of unity determined by the
generators of the graded ring.

Theorem 16.20 Let R be a graded ring with R0 being Artinian. Assume that R is generaed
over R0 by elements x1 , . . . , xr whose degrees are d1 , . . . , dr , all being positive. Let M be a finite
R-module. Then the Hilbert–Poincare series P( M, t) of M is a rational function of the type
ź
P( M, t) = f ( M, t)/tm (1 ´ t di ) (16.8)
i

where f ( M, t) is a polynomial with integral coefficients. If M is positively generated as well,

then m = 0.

Proof: The proof goes by induction on the number of generaors of R+ and relies on
Structure of Graded Modules (Theorem 10.41 on page 279), which allows us to reduce
to the case that M = R/p for a homogenous prime ideal p. Indeed, the class of rational
functions appearing on the right hand side of in (16.8) is closed under addition and
invariant under multiplication by powers of t (both positive and negative) and the
Hilbert–Poincaré series P( M, t) is an additive invariant.
If R+ Ď p, we are done, since then R/p is Artinian being contained in R0 and the
Hilbert–Poincaré series is in fact a polynomial. If p is not contained i R+ , one of the

10th January 2021 at 11:32am

Version 4.0 run 30
 numerical polynomials 419

generators, say xi , does not belong to p, and one may form the following sequence
which is exact in GrModR :
xi
0 / R/p(´di ) / R/p / R/p + ( xi ) / 0. (16.9)

Now, by Lemma 16.18 above P( R/p(´di )) = tdi P( R/p), and hence it ensues form (16.9)
that

P( R/p + ( xi ), t) = P( R/q, t) ´ P( R/p(´si ), t) = P( R/q, t) ´ tdi P( R/p, t).

Consequently we find that

P( R/p, t) = P( R/p + ( xi ), t)/(1 ´ tdi ),

and we are done since the right hand side by induction on the number of generators is
of the desired shape. o
As a corollaries of the proof we have

Corollary 16.21 Let k[ x1 , . . . , xr ] be a polynomial ring given a grading by letting xi be of

degree di . Then P( R, t) = (1 ´ tdi )´1 .
ś

Proof: Induction on r, successively killing each xi . o

Theorem 16.22 Let R be positively graded Noetherian ring with R0 Artinian and M a graded
module finite over R0 . Then the pole order of P( M, t) at t = 1 equals the dimension dim M,
and the residue of P( M, t) at t = 1 is positive rational number unless M is the zero module.
If P( M, t) happens to be a polynomial, in other words if M is Artinian, then the residue
should be interpreted as the value P( M, 1), that is the total length of M. It is never zero
unless M is the zero module.
Proof: The proof follows the same pattern as the proof of Theorem 16.20. By the
Structure Theorem 10.41 on page 279 it suffices to verify the claims for the quotients
R/p where p is a homogeneous prime; indeed, if R/pi (mi ) with 1 ď i ď r are the arising
subquotients, it holds that
ÿ
P( M, t) = P( R/pi , t)t´mi .
1ď i ď r

The residues at 1 of all the P( R/pi , t) being positive, no cancellation takes place, and
the pole order of the sum equals the maximum of the pole orders of the summands.
By the quotient case this equals max dim R/pi ; but this maximum is also equal to the
dimension dim M.
As to the case of the quotients R/p, one infers from (16.9) that the pole order
of the Hilbert–Poincare series goes up by one when one passes from R/p + ( xi ) to

10th January 2021 at 11:32am

Version 4.0 run 30
420 hilbert functions

R/p, as does the dimension by Krull’s Principal Ideal Theorem. Hence by induction
the dimension and the pole order agree. Keeping the classical identity (1 ´ td ) =
(1 ´ t)(1 + t + . . . + td´1 ) in mind one also infers from (16.9) that
1
Rest=1 P( R/p, t) = Rest=1 P( R/p + ( xi ))d´
i ,

and the claim about the residues being positive rational numbers ensues. o
The residue of P( M, t) at 1 is not always an integer, but it follows from the proof
1
that it is a rational number belonging to Z[d´ ´1
1 , . . . , dr ]; that is, only primes diving one
of the degrees appear in its denominator. When R is generated in degree one so that
all di ’s equal one, the residue will be an integer, and coincides with the degree of V (p)
in case M = R/p. In the general case some would call the residue the orbifold degree of
V (p).
Examples
Example 16.9 Let R = k [u, v] and A = k [uv, un , vn ]. Then A = k[ x, y, z]/(zn ´ xy). K
Example 16.10 Consider the surface with equation x2 + y3 + z5
= 0 in A3 .
It is another
of the so-called du Val singularities which goes under the name of the E8 -singularity
(there is an E and an E7 singularities as well, but no E9 or Eν for ν ě 9). Giving the
degrees to the variables in t deg x = 15, deg y = 10 and deg z = 6 the polynomial
x2 + y3 + z5 becomes homogeneous of degree 30; and from the exact sequence

0 / R[´30] /R /A /0

we find P( A, t) = (1 ´ t15 )´1 (1 ´ t10 )´1 (1 ´ t6 )´1 (1 ´ t30 )

The Laurent expantion of P( A, t) to the third round t = 1 looks like
1 1 269 269
( x ´ 1 ) ´2 + ( x ´ 1 ) ´1 + 3 2 ´ 4 2 ( x ´ 1 )
30 60 2 ¨3 ¨5 2 ¨3 ¨5
611 231
+ 5 3 2 ( x ´ 1)2 + 6 2 ( x ´ 1)3 + . . .
2 ¨3 ¨5 2 ¨5
and we notice that the pole order is two corresponding to the dimension of A being
two and the orbifold degree is 1/30. With modern computational software such series
might feed the numerological inclinations you might have; for instance, the 50th terms
of the series is
¨383 ¨ 6173996327 ¨ 927594632681286257477
( x ´ 1)50 .
253 ¨ 327 ¨ 514
There are nice primes appearing, and immediately a conjecture about the power of two
in the denominator surfaces. It is of course also interesting to expand P in powers of t:
K
Exercise 16.6 Let R be a Noetherian graded ring. Show that N Ď M is an R0 -submodule
then RM X M = N. Conclude that if R is Noetherian, then the all R0 -submodules

10th January 2021 at 11:32am

Version 4.0 run 30
 multiplicities and hilbert–samuel functions 421

À
Nr,s = rďnďs Rn are Noetherian R0 -modules, i particular they are finitely generated
R0 -modules. M
+ À ´ À
Exercise 16.7 Show that R = ně0 Rn and R = nď0 Rn are graded subrings of
R. Show that R is Noetherian if ad only if R0 is Noetherian and R is a finitely generted
R0 -algebra. M

16.2 Multiplicities and Hilbert–Samuel functions

Filtrations
A filtration of a ring A is a descending chain tAi u of ideals filtration of rings
(filtrasjon av ringer)
. . . Ď A i +1 Ď A i Ď . . . Ď A 0 = A

with A0 = A such that Ai A j Ď Ai+ j for all i and j. The only filtrations on rings we shall
meet in this course are the so-called a-adic ones; they are shaped like tai u for an ideal a
in A. These adic filtrations include the two trivial filtrations, one with Ai = A for all i
and one with A0 = A and Ai = 0 for i ą 0.
A filtration of an A-module M compatible with a given filtration of A is just a Filtration (filtrasjoner)
descending chain M = tMi u in M

. . . Ď Mi+1 Ď Mi Ď . . . Ď M0 = M

satisfying the compatibility requirement Ai M j Ď Mi+ j . If M is compatible with the

a-adic filtration of A; that is, if ai M j Ď Mi+ j for all i and j, it is said to be an a-filtration. a-filtrations
A self-propelled induction yields that for M to be a-filtration it suffices that aMi Ď Mi+1 (a-filtrasjoner)

for all i. In case equality eventually reigns; that is, when it holds that ai M j = Mi+ j for
all i and j ąą 0, the filtration is said to be a-stable. Again by a straightforward induction, a-stable filtrations
this is equivalent to aMi = Mi+1 for i ąą 0. (a-stabile filtrasjoner)

(16.23) A filtration tMi u on an A-module M induces one on every submodule N which

simply is given as Ni = Mi X N. There is also induced a filtration on the quotient
module M/N whose terms are the images of the Mi ’s, in other words, Mi + N/N. In
both cases the compatibility with the filtration on A is obvious. If tMi u is a-stable, the
induced filtration Mi X N is a-stable as well. Indeed, ai+r Mi X N = ai (ar Mr X N ).
À
(16.24) To any filtration M is associated the module GrM iě0 Mi /Mi+1 .
If a is finitely generated, say by a1 , . . . , as , the ring Gra ( A) will generated by the
classes [ ai ]1 of the ai ’s in a/a2 , so Gra is finitely generated in degree one over A/a.
The arche-type of an a-adic filtration is the one whose terms are an M.

16.3 Graded rings and modules

Recall from Section 2.8 on page 58 that a graded ring A is a ring whose underlying
À
abelian group decomposes as a direct sum A = i Ai of subgroups where the sum

10th January 2021 at 11:32am

Version 4.0 run 30
422 hilbert functions

extends over all integers. Every element x P A can thus be decomposed as a sum
ř
x = i xi with each xi being homogeneous of degree i; that is, belonging to Ai , and only
finitely many of them being non-zero. The non-zero xi ’s are called the homogenous
components of x of degree i, and the degree of x is the degree of the homogeneous
component of x of highest degree.
À
The decomposition A = i Ai must be compatible with the multiplication in A in
sense that
Ai A j Ď Ai + j

for all i and j; in other words, if x and y are homogeneous of degree i and j respectively,
their product is homogeneous of degree i + j. In particular, the graded piece A0 of
degree zero will be a subring of A, and each Ai is a module over A0 .
Example 16.11 The archetype of a graded ring is of course the polynomial ring R =
R0 [ x0 , . . . , xn ] over some ring R0 with the standard grading, the one giving all the
variable xi the degree one. The homogenous part Ri degree i has an R0 -basis which
consists of the monomials of degree i, and hence is a free R0 -module of rank (n+ i
i ). In
most examples occurring in algebraic geometry the ring R0 will be a field.
Other examples, also omnipresent in algebraic geometry, are the quotients R/a
where a is a homogeneous ideal. It holds true that ai = a X Ri and hence the induced
À
decomposition of R/a into homogeneous pieces is R/a = i ( Ri /ai ). K

Associated graded rings

With any ideal in a ring A one associates a graded ring Gra ( A) by the construction

ai /ai+1 = A/a ‘ a/a2 ‘ . . ..

à
Gra A =
iě0

The ring structure is defined in the straightforward manner: If [ a]i and [b] j are classes in
ai /ai+1 and a j /a j+1 of elements a P ai and b P a j respectively, their product equals the
class [ ab]i+ j in ai+ j /ai+ j+1 , which is a legitimate definition since obviously ab P ai+ j , and
altering a or b by an element from respectively ai+1 or a j+1 changes ab by an element
in ai+ j+1 so that the class [ ab]i+ j is well defined. The axioms for a graded ring are
Normal cones gotten almost for free, and any serious student should check them. The ring Gra A is
(normalkjegler) sometimes called the normal cone of Spec A along V (a), or if m is maximal, the tangent
Tangent cones cone of Spec A at m.
(tangentkjegler)
Example 16.12 To justify the name tangent cone, let us consider the example of a
simple double point in the plane located at the origin. It has two tangents, and the
only reasonable interpretation of the tangent cone is the union of the two lines. To
see this is the case, let A = k[ x, y] with constituting relation y2 = x2 ( x + 1), and let m
be the maximal ideal ( x, y). We contend that the associated graded algebra Grm A is

10th January 2021 at 11:32am

Version 4.0 run 30
 graded rings and modules 423

isomorphic to k[u, v] with constituting relation u2 ´ v2 = 0, where u denotes the class

of x in m/m2 and v that of y. Geometrically, the spectrum of Grm A equals the union
of the two lines y = ˘x in the plane which is what we wanted; at least this holds true
when k is not of characteristic two. In case it is, u + v is a nilpotent element of Grm A,
the square equals zero, and the spectrum represents a double line.
Let X and Y be variables. The assignments X ÞÑ u and U ÞÑ v induce a surjection
k [ X, Y ] onto Grm A. To figure out what the kernel is, observe that a homogeneous form
Fd ( X, Y ) of degree d maps to elements Fd ( x, y) in A and Fd (u, v) in Grm A. The latter
vanishes precisely when the former; that is Fd ( x, y), belongs to md+1 , and then there is
a polynomial G ( X, Y ) of degree exceeding d so that Fd ( x, y) ´ G ( x, y) = 0. Hence

Fd ( X, Y ) ´ G ( X, Y ) = H ( X, Y )(Y 2 ´ X 2 + X 3 )

for some polynomial H ( X, Y ). Separating out the initial homogeneous parts from both
sides, which both are of degree d, yields the equality

Fd ( X, Y ) = Hd´2 ( X, Y )(Y 2 ´ X 2 )

where Hd´2 ( X, Y ) is the homogeneous component of H of degree d ´ 2, and this is

precisely to say that Grm A » k[ X, Y ]/(Y 2 ´ X 2 ). K
(16.25) This example is at its base generic; the argument persists holding water when
the double point is replaced by any hypersurface in An ; that is, by any polynomial
f P k [ X1 , . . . , Xn ] vanishing at the origin (which at least when k is algebraically closed,
imposes no restriction as every point then is the origin in an appropriate coordinate
system). We let A = k[ X1 , . . . , Xn ]/( f ) and m the maximal ideal at the origin. Moreover,
we let f µ denote the initial form of f ; that is, the homogeneous component of f of
lowest degree. Then one verifies mutatis mutandis as for the simple double point, that
Grm A = k [ X1 , . . . , Xn ]/( f µ ).
For brevity we write R = k[ X1 , . . . , Xn ]. Then A = R/( f ) is the coordinate ring of
the hypersurface C = V ( f ) Ď A N , and as usual the lower case letters xi will denote the
the classes of the upper case letters Xi in A. In the polynomial ring R we may write
f = f µ + h where h is a polynomial of higher degree than µ. Denote by ui the classes of
xi in Grm A.
We contend that the assignments Xi ÞÑ ui define an isomorphism between the
tangent cone Grm A and R/( f µ ). Indeed, any form Fd ( X1 , . . . , Xn ) yields elements
Fd ( x1 , . . . , xn ) P A and Fd (u1 , . . . , un ) P Grm A, and the latter being zero means that
Fd ( x1 , . . . , xn ) = G ( x1 , . . . , xn ) with G ( x1 , . . . , xn ) P mi+1 ; hence an equality Fd ´ G =
A ¨ ( f µ + h), holds in k [ X1 , . . . , Xn ], and when we isolate the homogeneous parts of
degree d, we find Fd = Ad´µ f µ where Ad´µ is the homogeneous component of A of
degree d ´ µ; but this means that Fd P ( f µ ) R, and we are happy.

10th January 2021 at 11:32am

Version 4.0 run 30
 424 hilbert functions

Exercises
(16.8) Describe the the tangent cone at the origin of the super-cusp y p = x q , the trefoil
in A2 with equation ( x2 + y2 )2 + 3x2 y ´ y3 = 0 and the du Val singularity xy + zn+1 = 0
in A3 .
(16.9) Consider the algebra A = k[ x, y] with constituting relation y2 + x2 ´ x3 = 0.
Name three different fields k so that Grm A respectively is a domain, is reduced but not
a domain, and is not reduced.
(16.10) Let a Ď k[ x1 , . . . , xn ] be a homogeneous ideal and let A = k [ x1 , . . . , xn ]/a. Show
that A » Grm A where m = ( x1 , . . . , xn ). This is as should be, a cone is its proper
tangent cone
(16.11) Recall that any f P k[ x1 , . . . , xn ] has at initial term Init( f ) which is the homoge-
neous component of lowest degree. Let a Ď k[ x1 , . . . , xn ] be an ideal and let Init(a) be the
ideal generated by all initail terms of members of a; that is by the set t Init( f ) | f P a u.
Show that Grm A » k[ x1 , . . . , xn ]/Init(a).
M

16.4 Filtrations, the Artin–Rees lemme and Samuel functions

David Rees The Artin–Rees lemma is one more important result having kept the status of a lemma,
(1918–2013)
at least in the name. It was more or less simultaneously found by Emil Artin and David
British mathematician
Rees in the mid 1950’s. The proof we shall give is the one from Nagata’s book; albeit
simplistic, it is transparent, and one can see what is going on, at least in the case of
ideals. The general lemma is inferred from the ideal case by using an unabashed trick
Nagata calls “The principal of idealization”.

Proposition 16.26 (Artin–Rees lemma) Let a and I be two ideals in the Noetherian ring
A. Then there is an integer r so that

a n ( ar X I ) = a n + r X I

for all non-negative integers n.

Proof: It is clear that an (ar X I ) Ď an+r X I, so we merely have to verify the converse
inclusion. The ideal a is finitely generated, and we may choose generators a1 , . . . , as
for it. Let Σ be the subset of A[ x1 , . . . , xs ] form by the homogeneous polynomials f
such that f ( a1 , . . . , as ) P ar X I. The polynomial ring A[ x1 , . . . , xs ] being Noetherian by
Hilbert’s basis theorem, the ideal that Σ generates has a finite generator set f 1 , . . . , f ν
consisting of homogeneous forms. The integer r in the statement is the maximum of
their degrees.
Let a P an+r X I. Then a P an+r , and a = f ( a1 , . . . , as ) for some homogeneous
polynomial f of degree n + r. As a P an+r X I Ď ar X I, the polynomial belongs to Σ, and

10th January 2021 at 11:32am

Version 4.0 run 30
 filtrations, the artin–rees lemme and samuel functions 425

ř
we may write f = gi f i with the gi ’s are homogeneous and deg gi + deg f i = n + r.
Hence
ÿ
a = f ( a1 , . . . , a s ) = gi ( a 1 , . . . , a s ) f i ( a 1 , . . . , a s ) .

Since deg gi = n + r ´ deg f i ě n, it holds that gi ( a1 , . . . , as ) P an , and we are through.

o

Proposition 16.27 (General Artin–Rees lemma) Let A be a Noetherian ring and a an

ideal in A. Let M a finitely generated A-module and N Ď M a submodule. Then there is a
natural number r so that
a n ( ar M X N ) = a n + r M X N

for all non-negative integers n.

Proof: The simple trick is to replace the ring A by the ring A1 = A ‘ M in which
multiplication is defined as ( a + m)( a1 + m1 ) = aa1 + am1 + a1 m so that M2 = 0. The
module M and all its submodules will be ideals in A1 . Moreover, one replaces the ideal
a in A by the ideal a1 = a ‘ M in A’; then (a1 )m L = am L for all natural numbers m and
all submodules L of M, and the proposition follows immediately from the Artin–Rees
lemma for ideals. o
(16.28) Assume that M/Mi is of finite length over A, then the Samuel-function associ-
ated with M is the numerical function

SM (n) = ` A ( M/Mn ).

Lemma 16.29 If M is finite A-module and the filtration M of M is an a-filtration, the Samuel
function is a numerical polynomial.
À
Proof: To the filtration M one associates the graded module GrM M = i Mi /Mi+1 .
It holds true that ∆SM = hGrM Gra A so that we may awake prop. o

Lemma 16.30 Assume that A is a local ring and q is an m-primary ideal. Assume that
M = tMi u and N = tNi u are two q-stable filtrations of M. The the Samuel functions SM and
SN have the same degree and the same leading coefficient.

Proof: We may certainly assume that one of the filtrations, say N , is the q-adic one.
Now we have the inclusions qi M = qi M0 Ď Mi and Mi+r = qi Mr Ď qi M0 = qi M hence
it holds true that
S M ( i ) ď Sq ( i ) ď S M ( i + r )

for i ąą 0 from which ensues that the two leading terms coincide. o

10th January 2021 at 11:32am

Version 4.0 run 30
426 hilbert functions

Lemma 16.31 Assume given an exact sequence

0 / M1 /M / M2 /0

of graded modules over the local ring A, then d( M) = max d( M1 ), d( M2 ).

Proof: The filtration M = qν M X M1 induced in M1 from the q-adic filtration on M is

according to the Artin–Rees lemma a q-stable filtration. By additivity

S M,q = S M1 ,M + S M2 ,q

and since the leading coefficient of all Samuel polynomials are positive no cancellation
can occur, and the degree of S M,q equals the larger of the degrees of S M1 ,M and S M2 ,q .
o

Theorem 16.32 Let A be a local ring with maximal ideal. Then it holds true that d( M ) =
dim M

Proof: So show that d( M) ď dim M choose a system of parameters x1 , . . . , xd for A;

that is q they generate is m-primary and hm (v) and hq (v) are of the same degree by xxx.
But the associated graded ring Grq A is a quotient of the polynomial ring A/q[t1 , . . . , td ]
where ti is the class of xi in Grq A from which ensues that the degree of hq (ν) is at most
d.
We attack the other inequality, that dim M ď d( M ) by induction on dim M, and to
that end there is a chain tMi u whose subquotients are A/pi with the pi ’s being prime
ideals:
0 / Mi + 1 / Mi / A/pi (mi ) /0

Now o

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 17

Regular sequences

17.1 Depth, regular sequences and unmixedness

An important ingredient in the full proof of Bézout’s theorem is the concept of so-called
unmixed rings. These are Noetherian rings all whose associated prime ideals are of
the same height, or what amounts to the same in our context of algeras of finite type
over a field, that dim A/p is the same for all associated primes p. In particual A has no
embedded components, the height of an embedded prime would of course be larger
than the height of at least on of the others. In geometric terms, if A = k [ x1 , . . . , xn ]/a,
all the components of the closed algtebraic subset X = Z (a) are of the same dimension
and A has no embedded component.
Macaulay showed that if ( F1 , . . . , Fr ) is of height r, then k[ x1 , . . . ,x ]/( f 1 , . . . , f r ) is
unmixed. That the irreducible components of the closed algebraic set Z ( f 1 , . . . , f r ) all are
of codimension r is clear—the height being the smallest codimension of a components,
and Krull’s Hauptidealsatz tells us that every component is of codimension most r— so
the subtle content is that there are no embedded components. This has consequence
that if Fr+1 is a new polynomial not vanishing along any of the components, then Fr+1
is a non-zero divisor in k[ x1 , . . . , xn ]/( f 1 , . . . , f r ). So we see that ( F1 , . . . , Fr ) being of
height r is equivalent to F1 , . . . , Fr being a regular sequence.

Regular sequences
The theory of Cohen–Macaulay rings and more generally of the Cohen–Macaualy
modules, is based on the concept of regular sequences which was introduced by Jean
Pierre Serre in 1955. Their basic properties are described in this paragraph.
(17.1) The stage is set as follows. We are given a ring A together with a proper ideal a
in A and an A-module M. Most of the time An will be local and Noetherian and M
will be finitely generated over A.
A sequence x1 , . . . , xr of elements belonging to the ideal a is said to be regular for Regular sequences
M, or M-regular for short, if the following condition is fulfilled where for notational (regulære følger)
 428 regular sequences

convenience we let x0 = 0.

o For any i with 1 ď i ď r the multiplication-by-xi map

M/( x1 , . . . , xi´1 ) ÝÑ M/( x1 , . . . , xi´1 )

is injective.

In other words, xi is a not a zero-divisor in M/( x1 , . . . , xi´1 ). In particular, x1 is not

a zero-divisor in M, and this has lead to the usage that x1 being regular in M is
synonymous with x1 being a non-zero divisor in M.
Maximal regular (17.2) A regular sequence x1 , . . . , xr is said to be maximal if it is no longer regular
sequences (maksimale when an element is added to it. When M is a Noetherian module, this is equivalent
regulære følger)
to a being contained in one of the associated primes of M/( x1 , . . . , xr ); indeed, the
union of the associated primes of M/( x1 , . . . , xr ) is precisely the set of zero-divisors in
M/( x1 , . . . , xr ).
Exercise 17.1 Show that maximal regular sequences for Noetherian modules are finite.
Exihibit a counterexample when M is not Noetherian. Hint: Consider the ascending
chain ( x1 , . . . , xi ) of ideals. M
(17.3) Be aware that in general the order of the xi ’s is important, permute them and
the sequence may no more be regular. However, regular sequences for modules finitely
generated over local Noetherian rings remain regular after an arbitrary permutation, and
the same holds true for graded rings with appropriate finiteness conditions. Henceforth,
we shall work with rings that are Noetherian and local and with modules finitely
generated over A (but with a sideways glimpse into the graded case, so important for
projective geometry).

Example 17.1 The simplest example of a sequence that ceases being regular when
permuted is as follows. Start with the three coordinate planes in A3 ; they are given

10th January 2021 at 11:32am

Version 4.0 run 30
 depth, regular sequences and unmixedness 429

as the zero loci of x, y, z. Add a plane disjoint from one of them to the two others; e. g.
consider the zero loci of the three polynomials x (y ´ 1), y and z(y ´ 1).
Clearly x (y ´ 1), z(y ´ 1), y is not a regular sequence in k[ x, y, x ]. The point is that
z(y ´ 1) kills any function on Z ( x (y ´ 1)) that vanishes on the component Z ( x ) (for
example x) and is thus not a zero-divisor in k[ x, y, z]/( x (y ´ 1)).
On the other hand, the sequence x (y ´ 1), y, z(y ´ 1) is regular. Indeed, it holds
that k[ x, y, z]/( x (y ´ 1), y) = k[z], and in that ring z(y ´ 1) is congruent to z and thus
not a zero-divisor. Geometrically, capping Z ( x (y ´ 1)) with Z (y) makes the villain
component Z (y ´ 1) go away.
This example is in fact arche-typical. The troubles occur when two of the involved
closed algebraic sets have a common component disjoint from one of the components
of a third. If all components of all the closed algebraic subsets involved have a point in
common, one is basically in a local situation, and permutations are permitted. K

Permutation permitted
(17.4) As mention in the previous example, in local Noetherian rings a sequence being
regular is a property insensitive to order. The same holds true in a graded setting, and
in both cases Nakayama’s lemma is the tool that makes it work.

Lemma 17.5 Assume that A is a local Noetherian ring with maximal ideal m and M a finitely
generated A-module. If x1 , x2 is a regular sequence in m for M, then x2 , x1 is one as well.

Proof: There are two things to be checked. Firstly, that x2 is a non-zero divisor in M.
The annihilator (0 : x2 ) M = t a P M | x2 a = 0 u must map to zero in M/x1 M because
multiplication by x2 in M/x1 M is injective. Hence (0 : x2 ) M + x1 M = x1 M, and since
x1 P m and M is finitely generated, Nakayama’s lemma applies and (0 : x2 ) M = 0.
Secondly, we are to see that multiplication by x1 is injective on M/x2 M, so assume
that x1 a = x2 b. But multiplication by x2 is injective on M/x1 M, and it follows that
b = cx1 for some c; that is, x1 a = x1 x2 c. Cancelling x1 , which is legal since x1 is a
non-zero divisor in M, we obtain a = cx2 . o

Proposition 17.6 Let A be a local Noetherian ring with maximal ideal m and M a finitely
generated A-module. Assume that x1 , . . . , xr is a regular sequence in m for M. Then for any
permutation σ the sequence xσ(1) , . . . , xσ(n) is regular.

Proof: It suffices to say that any permutation can be achieved by successively swapping
neighbours . o
(17.7) The graded version reads as follows:

Proposition 17.8 Let A be a graded ring satisfying Ai = 0 when i ă 0, and let M be a

finitely generated graded A-module. If x1 , . . . , xr is a sequence of elements from A, homogeneous

10th January 2021 at 11:32am

Version 4.0 run 30
 430 regular sequences

of positive degree, that form a regular sequence in M, then for any permutation σ the sequence
xσ(1) , . . . , xσ(r) is also a regular in M.

Proof: As above, one may assume that r = 2. The proof of Lemma 17.5 goes through
mutatis mutandis; the sub module (0 : x2 ) M will be a graded submodule because x2 is
homogeneous, and a version of Nakayma’s lemma for graded things is available. o
Exercise 17.2 With assumptions as in 17.6 or 17.8, prove that if x1 , . . . , xr is a regular
ν
sequence for M and ν1 , . . . , νr is a sequence of natural numbers, then x11 , . . . , xrνr will be
a regular sequence as well. Hint: Reduce to the case of x1 , . . . , xr´1 , xrν . M
Exercise 17.3 Jean Dieudonné gave the following example of a regular sequence x1 , x2
in local non-Noetherian ring such that x2 , x1 is not regular. Consider the ring B of
germs of C8 -functions near 0 in R. It is a local ring whose maximal ideal m constists
of the functions vanishing at zero. Let a be the ideal a = i mi of functions all whose
Ş

derivatives vanish at the origin. Let A = B[ T ]/aTB[ T ]. Let I be the function I ( x ) = x.

Show that the sequence I, T is a regular sequence in A whereas T, I is not. M

Enters homolocigal algebra—the depth

(17.9) One of the first appearances of homological methods in commutative algebra
was in the circle of ideas round of regular sequences and Cohen-Macaulay modules.
These methods give a caracterization of the maximal length of M-regular sequences
in terms of certain homologically defined modules. The criterion has the virtue of not
explicitly refering to any sequence, and has the consequence that all maximal sequence
are of the same length.
The homological modules in question are modules ExtiA ( M, N ) associated with a
pair of A-modules M and N. In the lingo of homological algebra they appear as derived
functors of the functor Hom A (´, ´). Students not already acquainted with these useful
creatures should consult a textbook about homological algebra for the few of their very
basic properties we shall need.
(17.10) It is natural to introduce the number deptha M as the length of the longest
The depth of a module (maximal) regular M-sequence in a. It is called the depth of M in a. In the end, it turns
(dybden til en modul) out that all maximal M-sequences in a have the same length, but for the moment we do
not know that, and a priori the number is not even bounded. However, we have:

Lemma 17.11 If A is a local Noetherian ring, a a proper ideal and M a finitely generated
A-module, then deptha M ď dim M. In particular, deptha M is finite.

Proof: Induction on dim M (which is finite!). If dim M = 0, the maximal ideal m is

the only associated prime of M. Therefore every element in m is a zero divisor and
deptha M = 0.

10th January 2021 at 11:32am

Version 4.0 run 30
 depth, regular sequences and unmixedness 431

Next, observe that if x is a non-zero divisor in M, it holds true that dim M/xM ă
dim M, and by induction one may infer that

deptha M/xM ď dim M/xM ă dim M. (17.1)

So if x1 , . . . , xr is a maximal regular sequence in M (they are all finite after Problem 17.1),
the sequence x2 , . . . , xr will be one for M/x1 M, and by (17.1) r ´ 1 ă dim M; that is
r ď dim M. o
(17.12) We have comes to the homological characterization. It is notable since it
determines the depth of a module without referring to any regular sequence. We
introduce a number p( M ) which is the smallest integer i such that ExtiA ( A/a, M ) ‰ 0.

Proposition 17.13 Let A be a local Noetherian ring, a a proper ideal and M a finitely generated
A-module. Then deptha M = p( M ).

Proof: The proof goes by induction on the depth of M (which is finite by Lemma 17.11
above). That deptha M = 0, means that there no element in a is regular in M. In other
words, a is contained in one of the associated primes of M, say p. There is then an
inclusion A/p Ñ M and a surjection A/m Ñ A/p. Consequently Hom A ( A/a, M) ‰ 0,
and p( M) = 0.
Assume next that deptha M ą 0. If x is the first member of an M-regular sequence of
maximal length, the quotient M/xM satisfies deptha M/xM = deptha M ´ 1. Moreover,
since x is regular on M, one has the short exact sequences

0 /M x /M / M/xM / 0,

from which one derives a long exact sequence the relevant part for us being

ExtiA ( A/a, M ) / Exti ( A/a, M/xM ) / Exti+1 ( A/a, M) x =0 / Exti+1 ( A/a, M ).

A A A

Since x P a the multiplication by x on the ext-modules is the zero map. Now, if i + 1 ă

p( M) it ensues that ExtiA ( A/a, M/xM ) = 0, and we may conclude that p( M/xM ) + 1 ď
p( M). And if i + 1 = p( M ) it follows that ExtiA ( A/a, M/xM ) » ExtiA+1 ( A/a, M ) ‰ 0
so equality holds.
So both the quantities deptha M and p( M) drops by one when we mod out by x,
and thence they are equal by induction. o
The proposition has an important corollary, which in fact is the main target of this
paragraph:

Theorem 17.14 Let A be a local Noetherian ring, a an ideal in A and M a finitely generated
A-module. Then all maximal regular M-sequences in a have the same length.

10th January 2021 at 11:32am

Version 4.0 run 30
432 regular sequences

Example 17.2 A Noetherian zero-dimensional local ring has of course depth zero. A
Noetherian one-dimensional local ring A has depth one if and only if the maximal ideal
is not associated; that is, A has no embedded component. K
As usual, we also give a graded version:

Theorem 17.15 Let A be a graded ring satisfying Ai = 0 when i ă 0, and let M be a finitely
generated graded A-module. Then all homogeneous maximal regular M-sequences have the same
length.

The bound
In geometry the dimension of a close algebraic set is the maximum dimension of the
irreducible components, and the algebraic counterpart is that the dimension of a ring is
the maximum of the dimensions dim A/p for p running through the associated prime
ideals of A. This maximum is never assumed at an embedded prime since these by
definition strictly contain another associated prime. For a module M, the same holds
true as dim M = dim / Ann M.
The word depth has the flavour of something down, and indeed, depthm M is smaller
then all the dimensions dim A/p where this time p runs through all of the associated
primes, including the embedded ones. And this is the crucial point.

Proposition 17.16 As usual, let A be local Noetherian ring with maximal ideal m and let M
be a finitely generated A-module. It then holds true that

depthm M ď dim A/p

for all prime ideals associated to M.

Proof: The proof goes by induction on the depth of M. If depthm M = 0 there is

nothing to prove. So assume that depthm M ě 1. Then there is a short exact sequence

0 /M x /M / M/xM / 0. (17.2)

Let p be prime ideal associated to M. The sequence (17.2) above induces an exact
sequence

0 / Hom A ( A/p, M) x / Hom A ( A/p, M ) / Hom A ( A/p, M/xM ),

and by Nakayama’s lemma the cokernel of the multiplication-by-x map is a non-zero

submodule of Hom A ( A/p, M/xM ). Hence Hom A ( A/p, M/xM ) ‰ 0, and the ideal
p + ( x ) is contained in an associated ideal prime ideal q of M/xM. Since x is a non-zero
divisor in M and p is associated to M, we may infer that x R p, and therefore p is

10th January 2021 at 11:32am

Version 4.0 run 30
 depth, regular sequences and unmixedness 433

strictly contained in q. It follows that dim A/p ą dim A/q. Now, depthm M/xM =
depthm M ´ 1, and by induction

depthm M ´ 1 ď dim A/q ă dim A/p.

o
Cohen-Macaulay (17.17) The A-module M is said to be a Cohen-Macaulay module if depthm M = dim M,
modules in particular, the ring A itself is Cohen-Macaulay if depthm A = dim A. If x is a non-zero
Cohen-Macaulay
moduler) divisor in A, both the depth and the dimension of A/xA are one less than of A, and
hence A is Cohen–Macaulay if and only if A/xA is.

Theorem 17.18 Assume that A is a local Noetherian Cohen-Macaulay ring. Then A is

unmixed. That is dim a/p = dim A for all associated primes p of A; in particular, A has no
embedded components.

Proof: In view of Proposition 17.16 this is almost a tautology. The lower and the
upper bound of the dimensions dim A/p for p associated with A coincide, hence these
dimensions all coincide. o
(17.19) To check that a ring is Cohen–Macaulay, it suffices to exhibit one regular sequence
of length the dimension of the ring. For instance, the local rings An = k [ x1 , . . . , xn ]mn
where mn = ( x1 , . . . , xn ) are Cohen–Macaulay since the sequence x1 , . . . , xn is regular.
This follows easily by induction because there are natural isomorphisms An /xn An »
An´1 induced by the maps k[ x1 , . . . , xn ] Ñ k[ x1 , . . . , xn´1 ] that send xn to zero.

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 18

Categories

The concepts of categories, functors and natural transformations were introduced

during World War II, or more precisely at end of it in 1945, in paper by Saunders Mac
Lane and Samuel Eilenberg and was a giant step in the development of homological
algebra. Functors and natural transformations existed in mathematics long time before,
but without being neither explicitly defined nor named. After the birth of algebraic
topology and the subsequent development of homological algebra they surfaced as
natural foundations for the theory, and the concepts in a fairly explosive manner
permeated a wide variety of subjects, to lend Peter Freyd’s words.
Some years later, in then 1950’s the theory category were substantially refined and
developed by the french mathematical milieu with Alexander Grothendieck in the front
Samuel Eilenberg
row. His monumental contribution to mathematics was based on categories; you meet (1919–1998)
them in every thing he did. Hungarian–American
This short chapter does not have an ambitious approach to categories and functors, mathematician
and in very few words, merely aims at giving a very first introduction to the language.
In the course the language is used as no more than a notational devise which one hardly
can do without in a modern introduction to commutative algebra. Students aiming at
a pursuit of studies in algebraic geometry or algebraic topology will certainly need a
deeper understanding and a much broader mastering of the subject, but this text will
hopefully be a good beginning.
(18.1) Anyone studying mathematics at a certain level has experienced that the in-
Saunders Mac Lane
troduction of a new class of mathematical objects is accompanied by the introduction (1909–2005)
of a fresh class of maps; namely the maps that respect the structures one wants to American
investigate. Cases to have in mind can be vector space and linear maps or topological mathematician
spaces together with continuous maps.
A category is an axiomatic construct mimicking this situation. A category C has
two components; the class* of objects and the class of “maps” usually called morphisms. ˚ There is a (morfier)
Morphisms
set-theoretical issue
here with paradoxes
like Russel’s lurking ,
hence the word class
and not set. In our
shallow introduction
for beginners we just
ignore this issue and
think about classes
being sets (as in fact,
most researchers do as
well)
 436 categories

From the concrete model cases one merely carries along two things, the possibility to
compose two composable maps—that is, two maps with the target of one equals the
source of the other—and secondly, that every object has an identity map.
Even though the concept of categories is modelled on concrete situations, as in any
axiomatized theory, objects and morphisms can be any class and the composition any
collection of maps as longs the axioms are obeyed. One may be tempted to call the
morphisms arrows, when wanting to emphasise the axiomatic nature of the matter. So
category theory is a game of arrows!
Categories (kategorier) (18.2) Here comes the formal definition: A category consists of a class Ob C of objects,
Source of morphisms and secondly, for each pair X and Y of objects in Ob C a set HomC ( X, Y ) of arrows from
(kilden til morfier) X to Y. If φ is such an arrow, X is called the source and Y the target of φ. For each triple
target of morphisms
of objects X, Y and Z, there must be given a “composition”; that is, a map
(målet til morfier)
HomC ( X, Y ) ˆ HomC (Y, Z ) Ñ HomC ( X, Z )

which is written as φ ˝ ψ. Moreover, for any object X from the category there is a
Identities (identiteter) special arrow id[ X ] P Hom ( X, X ) called the identity. These givens are subjected to the
following two axioms, the first asserting that the identity arrows are neutral with respect
to composition, and the second that composition is associative:
i) φ ˝ id[ X ] = φ and id[Y ] ˝ φ = φ;
ii) φ ˝ (ψ ˝ ρ) = (φ ˝ ψ) ˝ ρ.
ρ
/Z
ψ
/Y
φ
/X where φ and ψ and ψ and ρ are two pairs of composable morphisms.
W
Example 18.1 There is a long list of categories one could call conservative or traditional,
whose objects are sets equipped with some extra structure and whose arrows are maps
respecting that structure, these are the ones we referred to in the motivating introduction
above. However, there are many, many others, and we shall give a few examples.
Examples of the conservative categories are legio. Just to mention a few which are
central in this text: The category Ab of abelian groups and group homomorphisms, the
category Rings of commutative rings with unit and maps of rings, and the categories
ModA and Alg A of respectively A-modules and A-algebras with corresponding homo-
Alexander morphisms. And of course, one has the category Sets whose objects are the sets and
Grothendieck
(1928–1998) whose arrows are just the ordinary set-theoretical maps. K
Stateless–French Example 18.2 Partial ordered sets: Any partially ordered set P can be interpreted as a
mathematician category P. The objects are just the elements in P; that is, Ob P = P, and the arrows are
as follows. If x, y P Ob P, the set HomP ( x, y) is either empty or a singleton, and it has
an element if and only if x ď y. That x ď x, ensures that identity maps exist, and the
order relation being transitive, ensures that composition is defined and associative. K
Example 18.3 Topological spaces up to homotopy: The category Top of topological space
with continuous maps as morphisms was, together with its satellite category, the

10th January 2021 at 11:32am

Version 4.0 run 30
 437

homotopy category HomTop, one of the first to be studied. The category HomTop plays
an important role in algebraic topology, and was a first and fundamental example of
a category whose morphism are not structure preserving maps between sets. Recall
that two continuous maps φ, ψ : X Ñ Y are homotopic if there is a continuous map
Φ : X ˆ I Ñ Y so that Φ( x, 0) = φ( x ) and Φ( x, 1) = ψ( x ). One checks that this is
an equivalence relation compatible with composition; i. e. φ „ φ1 then ψ ˝ φ „ ψ ˝ φ1
and φ ˝ ψ „ φ1 ˝ ψ. This means that one may “compose” homotopy classes, and thus
topological spaces and homotopy classes of maps, form a category. K
Example 18.4 Local rings–subcategories: In the category Loc of local rings the objects are
local rings A and the arrows ring homomorphisms taking the maximal ideal inttha
maximal ideal; that is, maps φ : A Ñ B such that φ(m A ) Ď mB , where m A and mB denote
the two maximal ideals. This is an example of a subcategory. Clearly the objects Ob Loc Subcategories
from a subclass of Ob Rings, and for each pair A and B of objects, the set HomLoc ( A, B) (underkategorier)

is a subset HomRings ( A, B). Moreover, composition in Loc agrees with composition in

Rings.
Notice, that in many instances HomLoc ( A, B) will be a proper subset of the set of
mappings HomLoc ( A, B) (e. g. if B is the fraction field of the local domain A), and for
most subcategories one expects similar proper inclusions. This leads to the concept of a
full subcategory A subcategory C of D is said to full if for any two objects X and Y from Full subcategories
C, the C-arrows from X to Y coincide with the D-arrows; in other words when it holds (fulle underkategorier)

that HomC ( X, Y ) = HomD ( X, Y ) for all X, Y P C. K

Functors
The principle of introducing maps respecting structure along with a new kind of
structures, applies naturally also to categories. These new “maps” are called functors. Functors (funktorer)
In contrast to ordinary maps, they operate on the two levels of a category, both on the
objects and on the arrows. So from a formal viewpoint they are not maps; though, one
often has a mental picture of them as maps.
Functors come in two variants. The contravariant functors reverse the direction of Contravariant functors
all arrows, whereas they are kept by the covariant ones. The two species are equally (kontravariacon-
Covariant functors
travaraintnet
(kovariante funktorer)
important, but for the sake of a short and simple presentation, in what follows we shall funktorer)
only deal with the covariant ones.
Here comes the formal definition. If A and B are two categories a (covariant) functor
F : A Ñ B is a collection of maps with the following constituents. Firstly, a map
F : Ob A Ñ Ob B that takes objects to objects, and, secondly, for each pair X and Y of
objects from A, a map F : HomA ( X, Y ) Ñ HomB ( F ( X ), F (Y )). The ingredient maps are
subjected to rules

o F (id[ X ]) = id[ F ( X )];

10th January 2021 at 11:32am

Version 4.0 run 30
 438 categories

o F ( φ ˝ ψ ) = F ( φ ) ˝ F ( ψ ).

Example 18.5 We have met Hom A (´, M) and ´b A M, which are functors Mod A to
Mod A ; and we have bases change functor ´b A B : Mod A Ñ ModB whenever B is an
A-algebra. K
Example 18.6 From any conservative category C to Sets there is a so-called forgetful
functor that just throws away the extra structure and sends an object X to the underlying
set and a map to underlying set-theoretical map. A variant are functors forgetting parts
of a structure with several layers; for instance, the functor from Rings Ñ Ab that forgets
the multiplication; i. e. it sends a rings to the underlying abelian group and a map to
the underlying homomorphism between the additive structures. K
Exercise 18.1 Make a list of all categories (up to equivalence) with three objects and
not more than nine arrows. M

Natural transformations
This is the thirds concept mentioned at the top of the chapter are the natural transfor-
mations.They are devises to compare different functors, and in fact, they nicely fit into
picture of “maps” preserving structure; but this time “maps” between functors.
Given two functors F and G from A to B. A natural transformation is a collection
of arrows Φ A : F ( A) Ñ G ( A) in B, one for each object A P Ob A, which are compatible
with the double action of functors, on objects and arrow, That is, they are requested to
render commutative all diagrams

ΦA
F ( A) / G ( A)

F (φ) G (φ)
 
F ( A1 ) / G ( A1 )
Φ A1

where φ : A Ñ A1 is any arrow in A. Two natural transformations, from F to G and one

from G to H can obviously be composed, just compose each Ψ A and Φ A . So the class of
functors from A to B form a category, the functor category Func(A, B).
A natural equivalence between two functors F and G is a natural transformations
Φ : F Ñ G having an inverse Ψ : G Ñ F; that is, one so that Φ ˝ Ψ = id[ G ] and
Ψ ˝ Φ = id[ F ]. When F and G are natural equivalent, one writes F „ G.
Equivalent categories (18.3) Two categories are said to be equivalent if there are functors F : A Ñ B and
(ekvivalente kategorier) G : B Ñ A such that F ˝ G „ id[B] and G ˝ F „ id[A].
Equivalence between categories replaces the notion of isomorphism we know from
other areas of axiomatic mathematics. It reflects that every object in one category is

10th January 2021 at 11:32am

Version 4.0 run 30
 439

isomorphic to one in the other and that these isomorphisms preserve the set homomor-
phism; i. e. given two objects X and Y from A, then F and G induce mutually inverse
maps between the arrow-sets HomA ( X, Y ) and HomB ( F ( X ), F (Y )). Every theorem
valid in one will as well be valid in the other, with the understanding that proofs are
formulated merely in specific categorial terms.
Be aware however, that Ob A and Ob B are often far from being bijective.
Example 18.7 A stupid example, the category whose objects are all singletons (or of any
singleton of your choice) and all hom-sets also being singletons (e. g. Homs (txu, tyu) =
t( x, y)u) is equivalent to the tiny toy category with just one object ˚ and Hom (˚, ˚) =
tid[˚]u. K

10th January 2021 at 11:32am

Version 4.0 run 30
Lecture 19

Solutions

Solutions for exercises in Chapter 1

Exercise 1.1 Units are evidently not zero divisors. Le x be a non-zero element in
A. Multiplication by x is an additive group homomorphism A Ñ A which is injective
when x is not a zero divisor A. Hence it is surjective as A is finite. It follows that 1 lies
in the image; that is, there is a y P A with xy = 1, so x is invertible.
Exercise 1.2 We have 72 = 23 32 . If 72|an , both 2 and 3 must divide a; hence the
nilpotent elements are all multiples of 6. If 72|ab, either 2 or 3 must divide a, so the
zero divisors are the multiples of 2 and the multiples of 3. The units are those elements
not a multiple of 3 or 2 (for instance, by the previous exercise), that is, the classes of
numbers prime to 2 and 3.
Exercise 1.3 Let p1 , . . . , pr be the different primes that occur in a factorization of n
into prime powers. If n|am , all the pi ’s divide a; hence the nilpotens are the multiples
of p1 ¨ . . . ¨ pr . If n|ab, one of the pi ’s must divide a. Conversely elements shaped like
1
a = a1 pi kill np´ i , and the latter being non-zero in Z/nZ, they are zero-divisors; so the
zero divisors are precisely the multiples of the pi ’s. Again by Exercise 1.1, the units are
the classes of elements not divisible by any of the pi ’s.
Exercise 1.8 let a/n be an element in A which is written in lowest terms; that is, a
and n are without common factors. Then there is a relation 1 = xa + yn with x, y P Z,
which gives 1/n = x ¨ a/n + y; so 1/n P A, and consequently 1/p P A for all primes
dividing n. Letting S be the set of all primes dividing denominators of elements from
A, we see that A = Z[ p´1 |p P S].
?
Exercise 1.10 (Units in imaginary quadratic extensions) Write η = i n. The elements in
Z[η ] are all the linear combinations x = a + bη where a and b are integers. The absolute
value of x is given as | x |2 = x xs = a2 + b2 n, and this is always an integer. Now, if xy = 1,
it follows that | x |2 |y|2 = 1, and | x |2 being an integer, it ensues that | x | = 1. The inverse
442 solutions

of x is given as x´1 = xs| x |´2 , and it equals xs and lies in Z[η ] whenever | x | = 1. When
n ą 1, the equation a2 + nb2 = 1 forces b = 0 and a = ˘1, but when n = 1, the equation
admits the solutions a = 0 and b = ˘1 as well. So when n ą 1 the only units are ˘1 and
the group of units is cyclic of order two, but when n = 1, they constitute the four-group
t˘1, ˘iu.
Exercise 1.11 The formula for the sum of a geometric series reads (and is valid in any
ring)
1 ´ x ν +1
1 + x + x2 + . . . + x ν = ,
1´x
so with ν = n ´ 1 and x = ´a it does the job.
Exercise 1.14 (Units in polynomial rings) That f is invertible when a is invertible and
all the other coefficients are nilpotent, follows from Exercise 1.11 on page 21. So assume
that f ( x ) = 1 + a1 x + . . . + an x n is invertible and let g( x ) = 1 + b1 + . . . + bm x m be the
inverse (we can safely assume that a0 = b0 = 1) We aim at showing by induction on i
that ain+1 bm´i = 0 for 0 ď i ă n + m. For i = 0 this holds since am bm is the only term in
f g of degree n + m.
Now, it holds that

an bm´i + an´1 bm´(i´1) + an´2 bm´(i´2) + . . . = 0

which gives

ain+1 bm´i + an´1 ain bm´(i´1) + an´2 ain bm´(i´2) + . . . = 0

By induction, ain bm´(i´1) = ain´1 bm´(i´2) = . . . = 0, and we conclude that ain+1 bm´i = 0.
Hence with i = m, it follows that am n
+1 = 0. To finish off the exercise, observe that
n
f ( x ) ´ an x will be invertible when f ( x ) is (Exercise 1.11), so repeating the procedure
we obtain eventually that all the ai ’s are nilpotent.
Exercise 1.16 A homomorphism k[t, t´1 ] Ñ k[u]. would take the group of units of
k[t, t´1 ] into the group of units of k[u]. The latter equals the invertible scalars k˚ , and
all powers of t being invertible in k [t, t´1 ], would be mapped into k˚ . Consequently the
whole of k[t, t´1 ] would be mapped into k, which is absurd!
Exercise 1.18 (Long division)
a) We proceed by induction of the degree d of f ; Let n = deg g. The claim is
trivial if deg f ă deg g, we just take q = 0 and r = f . So let α be the leading
coefficient of f and β that of g. Then f (t) ´ αβ´1 td´n g(t) is of lower degree than
f and may by induction be written f (t) ´ αβ´1 td´n g(t) = q1 (t) g(t) + r (t) with
deg r ă deg g. Then we find

f (t) = g(t)(q1 (t) + βα´1 td´n g(t)) + r (t)

10th January 2021 at 11:32am

Version 4.0 run 30
 443

and we are through.

b) Induction on the degree, the degree one case being evident. Let a be a root of
f , and divide f by t ´ a to obtain f (t) = q(t)(t ´ a) + r (t) where r is of degree
less than one, hence it must be a constant. Setting t = a gives r = 0, and
consequently f (t) = q(t)(t ´ a). If b is another root of f different from a, this
relation gives 0 = f (b) = q(b)(b ´ a). It ensues q(b) = 0 as A is a domain. The
quotient q being of degree deg f ´ 1 induction applies, and we may infer that q
has less than deg f ´ 1 roots. Consequently f has less than deg f roots.
Exercise 1.23 One has 1 = 4 ´ 3; hence it holds that 4 = 42 ´ 12 ” 42 mod 12,
so the class of 4 is an idempotent in Z/12Z. Similarly, ´3 = 12 + (´3)2 , so ´3 is
an idempotent in Z/12Z. They are together with the trivial idempotents the only
ones. Indeed, the corresponding decomposition of Z/12Z into a direct product is
Z/12Z » Z/3Z ˆ Z/12Z with 4 and ´3 representing the unit elements. Now, both
Z/3Z and Z/12Z only have trivial idempotents, and we are done.
One has 1 = 9 ´ 8 so 9 and ´8 are idempotents in Z/36Z. The corresponding
decomposition of Z/36Z in a product is Z/36Z » Z/9Z ˆ Z/8Z with ´8 being 1 in
Z/9Z and 9 being 1 in Z/8Z. The idempotents in Z/8Z and Z/9Z are the trivial
ones, hence we have found all.

Solutions for exercises in Chapter 2

Exercise 2.1 If a P a and b P b obviously ab P a X b, and sums of elements shaped like
ab belong then to a X b as well. If a, b P a X b obviously ab P ab so that (a X b)2 Ď ab.
As to the first example, let a = b = ( x ) in the polynomial ring k [ x ]. Then ab =
a2 = ( x2 ), but a X b = ( x ). For the second take a Ă b; for instance, let a = ( x ) and
b = ( x, y) in the polynomial ring k[ x, y]. Then a X b = ( x ) and (a X b)2 = ( x2 ), whereas
ab = ( x2 , xy).
Exercise 2.6 We contend that the elements 1, 2, 3, 6 form a basis for the Z-
‘ ‘ ‘

module Z[ 2, 3]. That they generate is clear. To see they are linearly independent
‘ ‘

over Z it suffices to see they are linearly independent over Q; in other words, it suffices
to show that the field Q( 2, 3) has dimension four over Q. Now, Q( 2, 3) is a
‘ ‘ ‘ ‘

quadratic extension of Q( 2) (it is easy to see that 3 does not belong to Q(2)), hence
‘ ‘

dimQ Q( 2, 3) = 2 dimQ Q( 2) = 4.
‘ ‘ ‘
‘ ‘
A basis for the ideal ( 2) is obtained by multiplying the basis above by 2; the
basis elements stay independent as the multiplication map is injective. This gives the
‘ ‘ ‘ ‘ ‘ ‘
basis 2, 2, 2 3, 6. In a similar way one finds the basis 3, 3, 3 2, 6 for the ideal
‘
( 3).
Exercise 2.9 The proof is mutatis mutandis the same as the proof of Gauss’ Lemma:
Let f (t) = i ai ti and g(t) = i bi ti be two polynomials whose product lies in pA[t];
ř ř

10th January 2021 at 11:32am

Version 4.0 run 30
444 solutions

i. e. all the products coefficients belong to p. Aiming for a contradiction assume that
neither f nor g is member of pA[t], and let ai0 and b j0 be coefficients of respectively f
and g lowest degree that do not lie in p. Then the coefficient of degree i0 + j0 of the
product equals
ÿ
ai b j = ai0 b j0 + . . .
i + j=i0 + j0

If i + j = i0 + j0 , either i = i0 and j = j0 or one of the inequalities i ă i0 or j ă j0 holds,

and in the latter case it ensues from the minimality of i0 and j0 that the product ai b j
belongs to p; hence ai0 b j0 lies there too, which is incompatible with ai0 and b j0 not doing
so.
Exercise 2.10 Let φ : A Ñ B be the ring homomorphism and p Ď B a prime ideal: That
ab P φ´1 (p) means that φ( ab) = φ( a)φ(b) P p. Then either φ( a) P p or φ(b) P p; that is,
a P φ´1 (p) or b P φ´1 (p).
The most stupid example of the inverse image of a maximal ideal not being maximal,
is the zero ideal in Q (which is maximal): it pulls back to the zero ideal in Z (which is
not maximal).
Consider e. g. the ideal (2)Z in Z and extend it to the Gaussian integers Z[i ]. One
has (1 + i )2 = 2i, and since i is a unit in Z[i ], it holds that (2)Z[i ] = (1 + i )2 Z[i ].
Exercise 2.11 Observe that a Ę b since the empty set is contained in every set. If azb
were contained in the union p1 Y . . . Y pr , it would hold that a Ď b Y p1 Y . . . Y pr . From
the Prime Avoidance Lemma it would follow that either a Ď b, which is not the case, or
that a Ď pi for one i; contradiction.
Exercise 2.12 The claim about vector spaces is the following: if V and V1 , . . . , Vr
are vector subspaces of the vector space W over k and V Ď V1 Y . . . Y Vr , then for at
least one index i it holds that V Ď Vi . Certainly one may assume that the union is
irredundant. We begin with treating the case V = W, that is W = V1 Y . . . Y Vr , and we
do that by induction on r. So let V 1 be the subspace generated by V2 Y . . . Y Vr Then
W = Ď V1 Y V 1 ; let v P V1 and w P W, and consider the line λv + (1 ´ l )v1 . It meets V1
in v and V 1 in the v1 but has infinitely many points, contradicting that W = Ď V1 Y v1 .
Hence either V1 = W and we are through, or V 1 = W and we are thorugh by induction.
The general case is reduced to this case by assuming that V does not lie in any
Vi and choosing points VzVi and replacing W by the span W 1 of v1 , . . . , vr and Vi by
Vi X W 1 .
Exercise 2.13 According to Proposition 2.19 about ideals in quotients the ideal
generated by pi equals ( pi , p1 ¨ . . . ¨ pr )/( p1 ¨ . . . ¨ pr ). By assertion ii) in the Isomorphism
Theorem (Theorem 2.21 on page 37) it follows that A/( pi ) » Z/( pi , p1 ¨ . . . ¨ pr )Z =
Z/pi Z = F pi . It is generally true that the additive group Z/nZ is of order n so A has

10th January 2021 at 11:32am

Version 4.0 run 30
 445

n = p1 ¨ . . . ¨ pr elements, and hence the ideal ( pi ) being the kernel of the canonical map
1
A Ñ Z/pi Z has np´ i = p1 ¨ . . . ¨ ppi ¨ . . . ¨ pr elements.
Finally, let e be the class of p1 ¨ . . . ¨ ppi ¨ . . . ¨ pr in A. It is killed by pi hence the ideal
e generates (which equals the additive subgroup it generates as all elements in A are
classes of integers) is a non-trivial factor of Z/pi Z, hence it is isomorphic to Z/pi Z
and has pi elements.
Exercise 2.15 (Primes in the Gaussian integers)
a) That ψ ˝ ψ = 1 amounts to x p´1 = 1 which is just Fermat’s little theorem (or
the facts that F˚p has p ´ 1 elements and that every group is killed by its order),
and for the same reason ψ assumes values in µ2 . The kernel of φ obviously
equals µ2 , and hence its image has ( p ´ 1)/2 elements, which all lie ker ψ.
Now, the equation x ( p´1)/2 = 1 has at most ( p ´ 1)/2 roots, and consequently
im φ = ker ψ. Finally, we infer that the image of ψ has ( p ´ 1)/(( p ´ 1)/2) = 2
members, i. e. ψ is surjective.
b) This is now a real sweet piece of cake: it suffices to deside when ´1 P ker ψ; that
is, when (´1)( p´1)/2 = 1. And of course, this occurs precisely when ( p ´ 1)/2
is even; that is, when p ” 1 mod 4.
c) The polynomial x2 + 1 is irreducible over F p precisely when it does not have a
root in F p ; that is, when p ı 1 mod 4.
d) blabla

Exercise 2.18 Let a be a non-zero ideal in A. Following the hint we let a P a be

such that δ( a0 ) is the least element in δ(azt0u), which is a legitimate definition since
W is well ordered so that every non-empty subset has a minimal element. We claim
that a = ( a0 ): indeed, for any a P A, we may write a = a0 q + r with δ(r ) ă δ( a0 ), but
evidently r = a ´ a0 q belongs to a, hence by minimality of a0 , it holds that r = 0 and
thence a = qa0 .
Exercise 2.20 (The Eisenstein integers)
a) The set Z[η ] is obviously closed under addition, and using that η 2 = ´η ´ 1, the
little calculation

(n + mη )(n1 + m1 η ) = nn1 + (nm1 + n1 m)η + mm1 η 2 =

= nn1 ´ mm1 + (nm1 + n1 m ´ mm1 )η

shows it is closed under multiplication as well.

b) Note that η + η̄ = ´1 and η η̄ = 1, so if x = n + mη is a unit, the integers n and
m satisfy the equation
n2 ´ nm + m2 = 1, (19.1)

10th January 2021 at 11:32am

Version 4.0 run 30
446 solutions

which ensues from the identity 1 = |η |2 = (n + mη )(n + mη̄ ). Assume n, m ě 0.

If n ą m it follows that n = 1 and m = 0, and symmetrically, m ą n implies
that m = 1 and n = 0, and if n = m, they are both equal to one. The equality
19.1 also forces n and m to be of the same sign, hence we find the solutions ˘1,
˘η and ˘(1 + η ). The group of units is the group µ6 of sixth roots of unit; it is
cyclic of order 6.
c) Consider a complex number z = x + yη with x and y real and set z0 = x ´ [ x ] +
(y ´ [y])η; then z = w + z0 with w P Z[η ] and z0 in the convex hull of η, 1 + η, 1
?
and 0, which is the depicts quadrilateral. The diameter of which is 3. Hence
?
the distance of any complex number to Z[η ] is less than 3/2 ă 1. Next, assume
given Eisenstein integers a and b and choose q P Z[η ] so that | ab´1 ´ q| ă 1. Put
r = b( ab´1 ´ q). Then a = bq + r and |r | ă |b|.

η 1+ η

0 1

d) Since the group of units is of order six, each non-zero element has six associates;
hence each non-zero ideal has six generators.
Ş Ş
Exercise 2.21 Assume that ab P iP I pi , but a R iP I pi . Then there is an i0 so that a
does not belong to pi0 , and thus a R pi for i ě i0 as the pi form a chain. It follows that
b P pi for i ě i0 , and by consequence b lies in pi for all i the pi ’s forming a chain.
Ť Ť
As to the union, let ab P iP I pi and assume that a R iP I pi . Then a does not belong
to any of the pi ’s and hence b belongs every one of them.
Finally, the set Σ of prime ideals containing a and contained in p is trivially non-emty
and every descending chain in Σ has a lower bound. Zorn then tells us that there is a
minimal element.
Exercise 2.22 Let tpi u be a saturated chain of prime ideals connecting p to q. Pick
an element x in q that does not lie in p and consider the two sets S = t pi | x P pi u
and T = t pi | x R pi u. Every prime ideal from T is strictly contained in every prime
ideal from S. Since the original chain is saturated, and the union p1 of the pi ’s from T
being a prime (Exercise 2.21 above) not containing x, the union p lies in T. Similarly, the
intersection q1 of the pi ’s that belong to S, is an element in S. It holds that p1 Ă q1 , and
because the original chain is saturated, there can be no prime ideal lying between p1
and q1 .
Exercise 2.23

10th January 2021 at 11:32am

Version 4.0 run 30
 447

a) The complement of one prime ideal p is multiplicatively closed since if neither

x nor y lies in p neither their product does; and it is saturated since p is closed
under multilplication. Obviously the intersection of saturated multiplicative sets
is a saturated multiplicative set, so the implication in one direction is proved.
For the reciproque, assume that S is multiplicative and saturated and let x R S.
Since S is saturate, from yx P S it would ensue that x P S, and hence the
principal ideal ( x ) is disjoint from S. By the Fundamental Existence Theorem
(Theorem 2.49 on page 49) there is then a prime ideal p containing x and disjoint
from S.
b) Assume that x and y are non-zero divisor, xya = 0 would imply that ya = 0
since x is a non-zero divisors; hence a = 0 since y also is one, and we conclude
that xy is a non-zero divisor, and so the non-zero divisors form a multiplicative
set. It is saturated because if x is a zero divisor, there is an a ‰ 0 xa = 0, and
then obviously xya = 0 too, so that xy will be a zero divisor.
c) Follows directly from a) and b).

Exercise 2.25 Each prime ideal in A/p is shaped like q/a for a prime ideal q in
A containing p. Each prime ideal q that strictly contains p must meet tan u by the
maximality of p, and consequently, q contains a; in other words, the class [ a] lies in q/a.
Ť
Exercise 2.27 Let tai u be the chain, and let a = i ai . Since A is a pid the ideal A is
principal, say generated by a. Since a belongs to the union of the ai ’s, it belong to one
of them; say aν , thus a = ( a) = aν , which evidently ensures that ai = aν for i ě ν.
Exercise 2.28 If p is the only prime ideal in A, the nil radical, being the intersection of
all prime ideals, equals p . Hence each element in p is nilpotent. The ideal p is also the
sole maximal ideal of A, hence all elements outside p are invertible.
Assume then that all elements are either nilpotent or invertible. If p is a prime ideal,
it consists of nilpotent elements as members are not invertible, and of course, every
nilpotent belongs to p. Hence p equals the radical, and is the only prime.
What we have shown amount to the ring A having just one prime ideal if and only
‘ ‘
if (0) is a maximal ideal; that is, if and only if A/ (0) is a field.
ν
Exercise 2.31 Each number in A can be written as x = n/m ¨ p11 ¨ . . . ¨ prνr where m
and n are integers relatively prime to each pi , and hence is a unit in A, and where the
νi ’s are non-negative integers. It follows that x is invertible if and only if all the νi ’s are
zero; hence each ideal ( pi ) A is maximal, and they are the only maximal ideals. Their
intersection, the Jacobson radical, equals ( p1 ¨ . . . ¨ pr ).
And A/( pi ) equals Z/( pi ) = F pi , since the denominator n in the expression for
x above, being relatively prime to pi , is invertible mod pi ; i. e. there is a relation
1 = sn + rpi with r, s P Z.

10th January 2021 at 11:32am

Version 4.0 run 30
448 solutions

Exercise 2.34 The Jacobson ideal J ( A) is a principal ideal, say ( f ).

Exercise 2.35 The ideals ma = ( x1 ´ a1 , . . . , xn ´ an ) are maximal for every point
( a1 , . . . , an ) in Cn , and a polynomial f that is contained in all the ma ’s, vanishes in the
entire Cn , hence is 0.
Exercise 2.40
a) For each index j ą 1 there is relation a j + b j = 1 with a j P a1 and b j P a j because
ś
a1 and a j are comaximal. Developing the product j ( a j + b j ), which equals one,
ś
one sees that all terms except b = j b j belong to a1 . Denoting their sum by a
ś
one finds 1 = a + b and it holds that a P a1 and b P ją1 b j .
b) First we do the casd r = 2. Write 1 = a1 + a2 with a1 P a1 and a2 P a2 . If
x P a1 X a2 we see that x = xa1 + xa2 lies in a1 a2 since both terms lie there.
Induction on r finishes the point. We may assume a2 ¨ . . . ¨ ar = a2 X . . . X ar . By
a) a1 and a2 ¨ . . . ¨ ar are comaximal and the claim follows by the case r = 2.
c) Induction on r: Let b1i = 1 1
ś
j‰i,iăr ai , then b1 + . . . + br´1 = A. Now after
a) ar and br are comaximale so that ar + br = A. These two relations give
ar = ar b11 + . . . + ar b1r´1 = b1 + . . . + br´1 , hence b1 + . . . + br´1 + br = A.
Exercise 2.41 It holds that 30 = 2 ¨ 3 ¨ 5 and 6 + 10 ´ 15 = 1 hence [6], [10] and [´15]
are orthogonal and equal one mod 5, 3 and 2 respectively. Similarly, 105 = 3 ¨ 5 ¨ 7 and
15 + 21 ´ 35 = 1 so that [15], [21], [´35] will be a complete orthogonal set of idempotens;
they are congruent one mod 7, 5 and 3 respectively.
? ?
Exercise 2.43 It is trivial that a Ď a + n. To prove the reverse inclusion, let x be an
element in A so that x n = 1ďiďr ai yi with ai P a and yi P n. Then each yi is nilpotent,
ř

and they being finite in number, there is an m P N so that yim = 0. Appealing to the
binomial theorem, one may express a power x N as a sum of terms, each having as factor
ν
a monomial y11 . . . yrνr of degree N; so choosing N ě rm each term will vanish.
To the second question, observe that if n is a locally nilpotent ideal in a ring A, then
n is contained in every prime ideal of A.
Exercise 2.46 A monomial x µ yν is of degree µ ´ ν hence of degree zero if and only
if µ = ν; i. e. it equals ( xy)µ . In other words a polynomial is of degree zero precisely
when it is shaped like f ( xy) with f a polynomial . Hence R0 equals the polynomial
ring k[ xy].
If n ą 0, the homogenous piece Rn equals x n k [ x, y] whereas Rn = yn k[ x, y] when
n ă 0.
Exercise 2.47 (Homogeneous prime ideals) Let p be a homogenous prime and x and
y two elements. Let xy P p and assume that x R p; we aim at showing that y lies in
p by contradiction, so assume as well that y R p. Let ν and µ be the highest degrees
of the homogeneous component of x respectively y not belonging to p. Decompose

10th January 2021 at 11:32am

Version 4.0 run 30
 449

x = x2 + xn + x1 and y = y2 + yµ + y1 where x2 and y2 recollect the homogeneous terms

of degree superior to ν respectively µ. So x2 and y2 both belong to p, and by definition
( x ´ x2 )(y ´ y2 ) P p. Replacing x with x ´ x2 and y with y ´ y2 we may assume that xν
and yµ are the highest homogeneous term of x and y.
Now xν yµ is the term of highest degree of the product xy and so lies in p because p
is homogeneous, but by construction neither xν nor yµ lies there; contradiction.
The reverse implication is obvious.

Exercise 2.49 We shall write x for ( x1 , . . . , xr ).

a) If the polynomial f is homogenous of degree d, all monomials occurring in f
ν
are of degree d so they are shaped like M( x ) = x11 ¨ . . . ¨ xrνr with the exponents
satisfying ν1 + . . . + νr = d, and for such obviously M(α ¨ x ) = αd¨ x (regardless
of k being finite or not).
ř
For the reverse implication, write f = i f i with f i homogeneous of degree i.
Then f (α ¨ x ) = i f i (α ¨ x ) = i αi ¨ f i ( x ). On the other hand, f (α ¨ x ) = αd ¨ f ( x )
ř ř

by assumptions. Equating the two expression we infer the relation

ÿ
(αd ´ αi ) f i ( x ) = 0.
i

Now, non-zero homogeneous polynomials of different degrees are linearly

independent, so were f not homogeneous, at least one term of degree less
than d, say i, would not vanish, and we could have deduced that αd ´ αi =
αi (αd´i ´ 1) = 0. Hence αd´i ´ 1 = 0 with d ´ i ą 0 and valid for all α P k˚ . This
equation obviously has only finitely many solutions, so k being infinite would
be absurd. Consequently f is homogenous.
b) Assume that the ideal a is invariant. We shall show that all homogeneous
components of a member f of a lie in a, and we intend to do that by induction
ř
on the number of homogeneous components. So let f = 0ďiďn f i be the
expansion of f into homogeneous components. Now, for all α P k˚ it holds that
f α = iăn αi f i + αn f n so that f α ´ αn f = 0ďiăn (αi ´ αn ) f i belongs to a. By
ř ř

induction it follows that αi (1 ´ αn´i ) f i P a and since k˚ infinite we may choose

α so that αn´i ‰ 1. Hence it holds that f i P a for i ă n, and consequently also
f n P a.
If the ideal a is homogeneous, it is generated by homogeneous elements and
hence is invariant by a).

Exercise 2.51 (Homogenization of polynomials) We rely on Exercise 2.49. Identities

between polynomial being formal, they can be check over any extension of the ground

10th January 2021 at 11:32am

Version 4.0 run 30
450 solutions

field, so we may safely assume that k is infinite: then, for α P k˚ we find

f H (αx0 , . . . , αxr ) = (αx0 )d f (αx1 /αx0 , . . . , αxr /αx0 ) =

= αd x0d f ( x1 /x0 , . . . , xr /x0 ) = αd f H ( x1 , . . . , xr ).

The second claim is trivial.

Exercise 2.52 (Dehomogenization of polynomials) We begin with analyzing the case when
g does not have x0 as factor. At least one of the homogeneous terms in g will then not
have x0 as factor and will survive unchanged when we put x0 = 1 ; so g D will be of the
v
same degree d as g. A homogeneous term x00 . . . ¨ ¨xrνr of g becomes x11 ¨ . . . ¨ xrνr in g D ,
ν

D H
and in ( g ) it reappear transformed into
v
x0d ¨ ( x1 /x0 )ν1 ¨ . . . ¨ ( xr /x0 )νr = x00 ¨ x11 ¨ . . . ¨ xrνr ,
ν

because d = ν0 + ν1 + . . . + νr . So in this particular case, when g does not have x0 as

factor, it holds true that g = ( g D ) H . The final remark is that if g = x0s h with h without
x0 as factor, we find g D = h D , and consequently ( g D ) H = h; so the factor x0ν disappears,
and has to be reintroduced to get the desired equality g = x0s ( g D ) H . It is then obvious
to find examples with s arbitrary between 0 and d; e. g. just take g = x0s x1d´s .
Exercise 2.53 Let a be of degree one and assume that a = bc. Let bµ and cν be the
terms of highest degre in b and c respectively. Then bµ cν ‰ 0 and ν + µ = 1, and both
being non-negative, one equals one and the other zero, say µ = 1 and ν = 0. It follows
that c P Ao which is a field, hence c is invertible. We conlsude that a irreducible.
Exercise 2.58 (Distinguished open sets)
a) The closed set V (( f )) is composed of the prime ideals p so that f P p and clearly
equals the complement of D ( f ); hence D ( f ) is open.
Ş
b) It obviously holds that V (a) = f Pa V (( f )) hence the complement of V (a)
satisfies V (a)c = f Pa D ( f ).
Ť

c) If a is generate by the f i ’s, then V (a) = i V (( f i )) and V (a)c = i D ( f i )

Ş Ť

d) Assume that the family tD ( f i )uiP I cover Spec A. Consider the ideal a = ( f i |i P I )
generated by the f i ’s. Since V (a) = H, it holds that a = A. Write 1 = a1 f 1 + . . . +
ar f r ; and then Spec A = D ( f 1 ) Y . . . Y D ( f r ) since no proper ideal can contain
all the f 1 , . . . , f r .

Solutions for exercises in Chapter 3

Exercise 3.8 Let m = (π ) be the maximal ideal in A and let k = A/m. Choose a prime
ideal p in A[t].
a) We follow the hint and consider the evaluation map A[ x ] Ñ K that sends f ( x )
to f (π ´1 ). Clearly πx ´ 1 lies in its kernel. Next, let f ( x ) belong to the kernel

10th January 2021 at 11:32am

Version 4.0 run 30
 451

and assume it is primitive. By standard long division in K [ x ] one finds a relation

f ( x ) = g( x )(πx ´ 1) + r where g P K [ x ] and r P K. Putting x = π ´1 , yields
r = 0, so that f ( x ) = g( x )(πx ´ 1). The content of πx ´ 1 being one, one finds
1 = c f = c g , and g belongs to A[ x ].
b) Assume that p X A = 0. Then pK [ x ] is proper and generated by an irreducible
g( x ) which we may assume lies in A. The ideal ( g( x ), π ) is not proper, since
( g( x )) is maximal, and it follows that ( g( x )) in k[t] is the entire ring, hence g( x )
is inverible mod π. If p X A = m it holds that π P p, and consequently p/(π ) is
a proper prime ideal in k [ x ]; hence it is generated by an irreducible polynomial
whose lift to A[ x ] will be irreducible and together with π it will generate p.

Exercise 3.10 A factorization of q would persist over any field extension of k, so

we may assume that k is algebraically closed. Suppose then that 1ďiďn xi2 is not
ř

irreducible; since it is homogeneous of degree two, it will have two factors both being
ř ř
linear, say f = i ai xi and g = i bi xi . All the ai ’s (and the bi ’s) must be different from
zero, since if if it happened that ai = 0, there would be no xi2 term in q. Absorbing
‘
ai in xi , we may assume that each ai = 1 at the price of changing the coefficients of q,
but no cross term will be introduced. Each cross term xi x j in the product f g has the
coefficient ai + a j . For any pair i, j of indices there is a third index k different from both.
So we find ai = ´a j , a j = ´ak and ai = ´ak which give ai = ´ai , and hence ai = 0 since
the characteristic is not two.
ν
Exercise 3.13 Let f = f 1 1 ¨ . . . ¨ f rνr we a factorization of f into irreducibles and such
g g
that no two f i ’s are associates. Then f = f g = f 1 ¨ . . . ¨ f r . The ring A is assumed to
g g
be factorial, so each f i is associated to a uniquely defined f j . Hence G permutes the
principal ideals ( f i ), and thus maps into the symmetric group Sr . By assumption, G
has no finite quotients, hence the image in Sr is trivial and each ideal ( f i ) is invariant.
g
It holds that f i = χg f i with χ( g) a unit. When G acts trivially on the units, we find
gh
( fi = (χ( g) f i )h = χ( g)h χ(h) f i ; that is χ( gh) = χ( g)χ(h) because χ( g)h = χ( g).
Exercise 3.14 One finds (2, i 2k )2 = (4, 2i 2k, ´2k ), but since k = 2r + 1 one has
‘ ‘

2 = 2k ´ 4 and 2 P p2 .
Assume then that f generates p; then 2 = a f 2 and it follows that 4 = N (2) =
N ( a) N ( f )2 . There are two possible cases which turn out to be impossible: either
N ( f ) = 2 and N ( a) = 1 which is impossible as N ( f ) = x2 + 2ky2 , or N ( a) = 4 and
N ( f ) = 1; that is, a = ˘2 and f = 1 which is impossible since p is a proper ideal.
In Z[i 2k] one has 2(k + 2) = (2 + i 2k ) ¨ (2 ´ i 2k), and e. g. 2 is irreducible since
‘ ‘ ‘

N (2) = 4 and x2 + 2ky2 is larger then 4 unless x = ˘2 and y = 0. Moreover, from

‘
2a = (2 + i 2k) it ensues that 4N ( a) = 4 + 2k hence 2N ( a) = 2 + k which is not the
case since k is odd. It follows that 2 is not associated to either factor to the right.

10th January 2021 at 11:32am

Version 4.0 run 30
452 solutions

Exercise 3.18 (The ring of real trigonometric polynomials)

a) You cannot get rid of a Y 2 term just by multiplication, so no polynomial p( X )
only depending on X can be a multiple of X 2 + Y 2 ´ 1, hence R[ x ] is polynomial
ring. Clearly 1 and y generate R[ x, y] over R[ x ], and if f ( x ) + g( x )y = 0, it
follows that f = g = 0; indeed, a polynomial f ( X ) + g( X )Y can not be a
multiple of X 2 + Y 2 ´ 1, just look at the highest term in Y of a potensial product.
b) this is just Proposition 3.36 above.
‘ ‘ ‘
Exercise 3.21 Answers: 1 + 2, 2 + 3 and 9 + 4 5.

Solutions for exercises in Chapter 4

Exercise 4.1 Define A Ñ B by sending an element a to a ¨ 1B . The module axioms
ensure this will be a map of rings; e. g. it repects multiplcation since aa1 goes to aa1 ¨ 1B
and aa1 ¨ 1B = a ¨ ( a1 ¨ 1B) = ( a ¨ 1B ) ¨ ( a1 ¨ 1B ).
Exercise 4.2 Each φ : A Ñ M gives an element φ(1) P M. For each m P M one may
define φ : A Ñ M by a ÞÑ am. The two construction are evidenly mutually inverse.
Exercise 4.29 The hint says everything: Since det Φ ‰ 0, the determinant det Φ does
not belong to the maximal ideal of the local ring A; thence det Φ is invertible, and by
the adjunction formula Φ is invertible; indeed, Φ´1 = det Φ´1 Φ: .
In the first calculus courses we learned that a continuous function that does not
vanish at a point, does not vanish in a vicinity. Here an analogy to that is playing
and is applied to the determinant; working in a local ring is in a way working in an
unspecified neigbourhood.
Exercise 4.30 (Demystisufying Nakayama’s lemma) Let e1 , . . . , er and f 1 , . . . , f s be bases
for E and F respectively and let the matrix Φ of φ be the s ˆ r-matrix ( aij ).
a) Since Φ is surjective it holds that r ě s, and one of the maximal minor (that is
an s ˆ s-minor) of ( aij ) is non-zero, After permuting basis elements if required,
we may assume that the minor in question is minor ( aij ) with 1 ď i ď s
and 1 ď j ď s; in other words the matrix to the restriction φ|E1 of φ to the
submodule E1 = Ae1 ‘ . . . Aes . By the previous exercise we conclude that φ|E1 is
an isomorphism; hence φ is surjective.
b) Recall that a module of finite presentation a finite generated module so that the
relations is finitely generated as well; in other words M is a cokernel of a map
between finite free modules; it lives in the sequence

E F M 0

Solutions for exercises in Chapter 5

10th January 2021 at 11:32am

Version 4.0 run 30
 453

Solutions for exercises in Chapter 6

ř
Exercise 6.12 Start with an expression τ = 1ďiďs ei b f i , and proceede by induction
on s. If the ei ’s are linearly independent, there is nothing to prove, if not, one of the ei ’s
ř
is a linear combinations of the others, say es = 1ď jďs´1 α j e j . Substituting this in the
ř ř ř
expression for τ yields τ = iďs´1 ei b f i ´ jďs´1 α j e j b f s = iďs´1 ei b ( f i ´ αi f s ), and
we are through by induction.

Solutions for exercises in Chapter 7

Exercise 7.3 Consider the localization Σ´1 A[t1 , . . . , tn ], which is a subring of the
fraction field of A[t1 , . . . , tn ]. It consists of fractions shaped like b´1 f (t1 , . . . , tn ) where
b P Azt0u and f P A[t1 , . . . , tn ]. All these are elements in K [t1 , . . . , tn ]. On the other
hand, if* f = α aα bα´1 tα P K [t1 , . . . , tn ] and b = α bα , it holds that b f P A[t1 , . . . , tn ] ˚ Here we use
ř ś

and so f = b´1 (b f ) P Σ´1 A[t1 , . . . , tn ]. multi-index notation

Exercise 7.5 As 1/2 = 5 ¨ 1/10 and 1/5 = 2 ¨ 1/10, clearly Z[1/2, 1/5] Ď Z[1/10], and
because 1/10 = 1/2 ¨ 1/5, the reverse inclusion also holds true.
Exercise 7.6 Let S Ď Z be the subset S = t n | 1/n P A u. It is clearly multiplicatively
closed. If a/m P A with ( a, m) = 1, we may find integers x and y such that xa + ym = 1,
which upon divison by m gives m´1 = xam´1 + y; hence m P S and ensues that
A = S´1 Z.
Exercise 7.7 The product of two units is obviously a unit. If S = A˚ , the identity map
id A takes elements in S to units, and induces a map φ : AS Ñ A such that φ ˝ ι S = id A .
It ends ι(s)´1 a to s´1 a so it is clearly an isomorphism.
Exercise 7.8 This is just Exercise 7.3 applied with A = k[ x1 , . . . , xr ] and ti = xi+r .
Exercise 7.9 The group of units A˚ is saturated since if xy is invertibel both x and y
are. And of course, if xy P t1u, by definition x and y are units.
Exercise 7.10 That the both the odd and the even numbers form multiplicative sets is
obvious. In the case of the even integers, the saturation will be the multiplicative set
Zzt0u. Indeed, if n ‰ 0, obviously 2n is even, and hence n belongs to the saturation.
In the case of odd integers the saturation will be generated by ´1 and all odd primes:
any odd numbers is a product of such. If [ a] = 1 and [b] = 1 clearly [ ab] = 1 (where [ x ]
denots the class of x in Z/pZ). The saturation will be the set of integers invertible mod
p; that is, those on the form ap + b where (b, p) = 1. Indeed, [ ab] = 1, obviously [ a] is
invertible in Z/pZ.
Exercise 7.11 (T) he saturated multiplicative sets are those generated by ´1 and an
arbitrary set of primes. Indeed, if S is saturated, and n P S any prime factor p of n
belongs to S. Hence, the all elements of S are products of primes belonging to S, and of
course ´1 P S as well ((´1)(´1) = 1 P S). In a general A, the same argument gives that

10th January 2021 at 11:32am

Version 4.0 run 30
454 solutions

the saturated multiplicative sets are the sets generated by the units A˚ and any set of
irreducible (equivalently prime) elements from A.
Ť
Exercise 7.12 Consider the set S = x + a where the union extends over all x P A
whose class [ x ] P A/a is invertible. It is clearly multiplicative and saturated, it contains
1 + a, and saying ( x + a)(y + a1 ) = 1 + aa1 is precisely to say that x (and y) is invertible
mod a. Hence S is saturation of 1 + a.

Solutions for exercises in Chapter 8

Solutions for exercises in Chapter 9
À
Exercise 9.1 Consider each addend Mi of the sum M = iP I Mi as a submodule of
M. Suppose that the index set I is infinite, equip the index set I with a linear order and
À
let Nj = iă j Mi . Then tNj u jP I is an ascending chain which does not stabilize; hence
M is not Noetherian. Submodules of Noetherian modules are Noetherian, hence if M is
Noetherian, so is each Mi .
Exercise 9.2 The ring of integers Z is a pid and pid’s are Noetherian. In the module
Z p8 one has the sub Z-modules ( p´i ) = t [ ap´1 ] | a P Z u. They form an ascending
chain of submodules, but it does not stabilize: if p´i´1 = ap´i + b with a, b P Z, we
would have p´1 = a + bpi P Z, which is absurd.
Now, Z is not Artinian since e. g. t( pi )uiPN is descending chain that does not stabilize
(p being any integer except ˘1).
Every element x in Z p8 lies in some ( p´ν ) with ν ą 0; indeed, write x as a finite
sum x = i ai p´i = bp´ν where ν = min i. Hence the only submodules of Z p8 are the
ř

cyclic modules ( p´i )Z p8 . The submodule ( p´i )Z p8 is killed by pi but not by pi´1 and
it is therefore isomorphic to Z/pi Z, in which there clearly is no infinite descending
chain.
Exercise 9.3 The first assertion is clear: any B-submodule of M is an A-submodule,
so any ascending chain of B-submodules is a chain of A-submodules and stabilizes as
A is Noetherian. Examples abound. A simple one is furnished by the extension Z Ď Q.
Obviously Q is Noetherian as Q-module (every finite dimensional vector space over
a field is), but clearly Q is not a Noetherian Z-module: if it were, it would be finitely
generated by the Main Theorem, and the common denominator of the elements from a
finite generating set would serve at a denominator for all rational numbers.
Finally, suppose that φ is surjective. Then any A-submodule N is a B-module: if
x P B, is of shape x = φ(y) it holds that xN = yN, and the assertion follows since
chains of B-modules stabelizes.
Exercise 9.4 Simple module are Noetherian (the two sole submodules are both finitely
generated), and the claim in the exercise follows directly from Proposition 9.7 on
page 231.

10th January 2021 at 11:32am

Version 4.0 run 30
 455

Exercise 9.5 The implication ii) ñ i) is obvious. For the reverse, we mimic the
argument in the Main Theorem: if no maximal element is found in Σ, any element has
a strict subset lying in Σ, and with this one recursively constructs a strictly descending
infinite chain of submodules from Σ; by consequence, M is not Artinian.
Exercise 9.6 Let n be an integer from A such that 1/n R A, and consider the principal
ideals (n´i t)Q[t]. They form a strictly increasing chain of ideals showing that A is
not Noetherian: that the chain is increasing is clear, and if n´i+1t P (n´i t)Q[t], one
has n´(i+1) t = f (t)n´i f for some polynomials f (t). Cancelling n´i t gives f (t) = n´1 ,
which contradicts the assumption that f (0) P A. Hence it is also strictly increasing.
Exercise 9.17 Let x be the generator of the maximal ideal.
a) Assume that a and b are elements that do not belong to i mi but whose product
Ş

does. Then we may write a = αxi and b = βx j with i and j maximal and
consequently α and β will be units. It holds that ab = γxi+ j+1 for some γ, hence
αβxi+ j = γxi+ j+1 from which it ensues that (αβ ´ γx ) xi+ j = (0). Now, αβ ´ γx
is a unit (αβ is a unit and γx P m) so that xi+ j = 0 and x is nilpotent.
Assume next that p is a prime ideal contained in i mi , but which does not
Ş

contain x. Any element a P p is shaped like a = γx, and since x R p, it follows

that γ P p. Hence p = xp, and an obvious induction shows that p Ď i mi .
Ş

b) If i m = (0), the set of natural numbers i so that a Ď mi is not empty (1 lies

Ş i

there) and is not the entire N (the ideal a is non-zero); hence there is a maximal
i such that a Ď mi . We claim that a = ( xi ), and it suffices to show that xi P a:
since a Ę mi+1 , there is an element a P a not in mi+1 , which must be shaped like
a = αxi with α not in m; hence α is a unit and xi = α´1 a P a.
c) When A is Noetherian, Krull’s Intersection Theorem yields that i mi = (0) (or
Ş

even Nakayama Classic suffices since x ( i mi ) = i mi by a)). If i mi = 0, all

Ş Ş Ş

ideals are principal by b), in particular they are finitely generated.

Exercise 9.20 To ease the notation, we shall identify N with its image in M. Let
Mν´1 Ă Mν be one of the inclusions in a composition series tMi u for M. Consider the
intersection N X Mν . As there is no submodule lying strictly between Mν´1 and Mν ,
there are two possibilities: Either Mν X N = Mν , and in that case Mν´1 X N = Mν´1 Ă
Mi = Mi X N, and Mν X N/Mν´1 X N is simple. Or Mν X N Ă Mν´1 , in which case
Mν X N = Mν´1 X N and a repetition appears.

Solutions for exercises in Chapter 10

Exercise 10.2 Consider k[ x1 , . . . , xn ] as the graded ring A[ x1 , . . . , xr ] where we let
A = k[ xr+1 , . . . , xn ] be the part of degree zero and each xi with i ď r is of degree one.
Then pm consists of elements whose lowest term is of degree m or more.

10th January 2021 at 11:32am

Version 4.0 run 30
456 solutions

Develop f in homogeneous components f = f 0 + f 1 + . . .. If f 0 = 0, the class of f

mod pm is nilpotent—in fact f m is of degree at least m so that f m P pm . So assume that
f 0 ‰ 0, but that f g = 0 mod pm ; that is, all its homogenous components are of degree
at least m, and that g R pm ; i. e. deg g ă m. Develop g in homogeneous components:
g = gs + gs+1 with gs R pm ; i. e. of degree less than m. We find that

f g = f 0 gs + f 1 gs + f 0 gs+1 + higher terms .

For degree reasons—that is, since s ă m, it follows that f 0 gs = 0, and hence gs = 0;

contradiction.
Exercise 10.4 Note that ( x2 , xy) = ( x2 , x (y + a)), changing coordinates y Ñ y + a
the general case follows from the case a = 0, which is just Example 10.4. The ideals
( x, , y + a) are different for different a.
Exercise 10.5 Localizing in x, we find aA x = A x , due to x3 being a generator, and
inverting y, we obtain aAy = ( x ) Ay . Hence we conclude that

a = (x) X q

where q is ( x, y)-primary. A good guess would be

a = ( x ) X ( y3 , y2 x 2 , x 3 ).

The inclusion Ď is clear, and we know that a is monomial, så it suffices to see that any
monomial from theright hand side belongs to a. It must have x as a factor, hence is
either y3 x, y2 x2 or x3 as a factor; and that’s it!
One could also resort to the JCO-algoritm;

a =( x3 , y2 x2 , y3 x )
=( x ) X ( x3 , y2 x2 , y3 )
=( x ) X ( x3 , y2 ) X ( x2 , y3 )

and coalescing the two last ideals results in ( x3 , x2 y2 , y3 ) (again, enough to check
monomials).
Exercise 10.6 The only challenge is to determine the primary decomposition of the
square
a2 = (y2 z2 , y2 x2 , x2 z2 , xyz2 , xzy2 , yzx2 ).
Localizing in x we find that a2 A x = (y, z)2 A x , and since (y, z)2 is primary and by
symmetry we conclude that

a2 = ( x, y)2 X ( x, z)2 X (y, z)2 X q

10th January 2021 at 11:32am

Version 4.0 run 30
 457

where q is ( x, y, z)-primary. Note that xyz lies in the intersection of the three squares to
the right; it is not a member of a2 , but x2 yz, xy2 z and xyz2 are, so get rid of xyz we try
with q = ( x2 , y2 , z2 ). We claim that

a2 = ( x, y)2 X ( x, z)2 X (y, z)2 X ( x2 , y2 , z2 ),

and indeed, this holds true. The inclusion Ď is easy, for the other, it suffices to verify it
for monomials. A monomial lying in the right hand side must contain a square of one
of the variables, say it contains x2 . Lying in (y, z)2 = (y2 , yz, z2 ) it must contain one of
the monomials y2 x2 , yzx2 or z2 x2 , but all these belong to a2 .
Exercise 10.13

=( x α y, z β x, yγ z)
=( x α , z β x, yγ z) X (z β x, y)
=( x α , z β , yγ z) X ( x, yγ z) X (z β , x ) X ( x, y)
=( x α , z β , yγ ) X ( x α , z) X ( x, yγ ) X ( x, z) X (z β , x ) X ( x, y)
=( x α , z β , yγ ) X ( x α , z) X ( x, yγ ) X (z β , x )

Solutions for exercises in Chapter 11

Exercise 11.1 The maximal ideal of A( x) is the principal ideal ( x ) A( x) . According to
Exercise 9.17 on page 244 the powers ( xi ) A( x) are all non-zero ideals in A( x) , and obvi-
ously, none of these are prime. So there is just one chain in A( x) , namely (0) Ď ( x ) A( x)
and the dimesion is one.
Exercise 11.2 Let K = k(z), and S = t f (z) | f poly u: Then S´1 k[ x, y, z] = K [ x, y].
Hence Rp = K [ x, y]( x,y) . We argued in Paragraph 11.4 on page 289 that dim K [ x, y] = 2,
so dim Rp ď 2. And there is the obvious chain (0) Ď (y) Ď ( x, y), so dim Rp = 2.
Exercise 11.3 Indeed, if q Ă p is a given pair, there is minimal prime q0 contained in q
and a maximal ideal m containing p. Fix a saturated chain ascending from q0 to q and
one ascending fra p to m. Then any saturated chain from q to p may be embedded in
one from q0 to m by concatenating it with the two fixed chains, and all such chains have
the same length.

Solutions for exercises in Chapter 12

An element x = a + b d P Q( d) has a minimal equation
‘ ‘
Exercise 12.1

x2 ´ 2ax + a2 ´ b2 d = 0

and it is integral if and only if the coefficients are integers; that is, n = 2a P Z and
a2 ´ b2 d P Z. It follows that 4b2 d P Z which yields that 2b P Z (write b = zy´1 with

10th January 2021 at 11:32am

Version 4.0 run 30
458 solutions

z, y P Z; then 4z2 d = y2 , and each factor in y must be a factor in 2z as d is square-free).

It follows that (2a)2 ” (2b)2 d mod 4.
If d ” 1 mod 4, it follows that (2a)2 ” (2b)2 and 2a and 2b have the same parity.
If they both are odd, one has a + b d = 2´1 (2a + 2b d) = w + 2´1 (1 + d) with
‘ ‘ ‘

w P Z[ d], and if they are even clearly both a and b are integers.
‘

If d ı 1 mod 4, it ensues, since squares are either 1 or 0 mod 4, that (2a)2 ” (2b)2 ”
0 mod 4 and hence that 2a and 2b both are even, so that a, b P Z.
Exercise 12.4 Let f P B[ x ] be an element and write f = bn x n + . . . + b1 x + b0 . Since
B is supposed to be integral over A, the extension C = A[b0 , . . . , bb ] is a finite faithful
module over A, and it esnues that C [ x ] is a finite faithful module over A[ x ]; indeed, it
has the same genrators as C. And of course f P C [ x ], hence it is integral over A.
Exercise 12.8
a) Define g( xy´1 ) = g( x ) g(y)´1 .
b) The crucial observation is that yh( x ) = h(y) x for all h; hence we find
ź ź ź ź
h(y) hg( x ) = h(y) x g( x ) = yh( x ) g( x ) = y g ( x ).
g‰e g‰h,e g‰h,e g‰e

ś
ś
´1
Then as in the hint x/y = x g‰e g( x ) y g‰e g( x ) , and it follows that
K G = L.
ś
c) The polynomial P(t) = gPG (t ´ g( x )) in B[t] where t is a variable, is clearly
invariant under G since G just permutes the factors, hence each coefficient is
invariant and lies in A (different powers of t are linearly independent over B).
The polynomial P is monic and P( x ) = 0 since e P G.
d) If x P L is integral over A, it is a priori integral over B, so if B is integrally closed
in K it lies in B, and being invariant, it therefore belongs to A.
Exercise 12.9 Let q and q1 be primes both lying over p and assume that q is not
equal to any of of the translates g(q1 ). By prime avoidance there is then an x P q not
lying in any of the g(q)1 . Consider gPG g( x ). It is invariant and lies in A X q, but
ś

A X q = p = A X q1 , so gPG g( x ) also lies in q1 . It follows that g( x ) P q1 for some g,

ś

hence x P g(q1 ). Contradiction.

Exercise 12.11 By 12.29 if m is a maximal ideal in B, the intersection m X A is maximal
Ş Ş
is maximal in A; hence n Ď A n Ď m Ď B n X A blabla
Exercise 12.12 Let q1 and q2 be two prime ideals in B lying over the same prime ideal
ν
p in A. A power x p of each element x in q1 lies in A, hence it belongs to q1 X A, which
equals q2 X A. It follows that x P q2 , and we may conclude that q2 Ď q1 . By symmetry
the reverse inclusion holds as well, and q1 = q2 . It follows that π is injective. Now, B is
integral over A and π is therefore surjective and closed, hence it is a homeomorphism.

10th January 2021 at 11:32am

Version 4.0 run 30
 459

Solutions for exercises in Chapter 13

Exercise 13.4 According to Proposition 3.30 on page 77 the maximal ideal m is
of the form ( f ( x ), g(y)) with f and g irreducible. Now, there is an isomorphism
k [ x, y]/( f ( x ), g(y)) » k[ x ]/( f ( x ))bk k[y]/( g(y)) = E f bk Eg , and the tensor product
E f bk Eg is a field precisely when E f and Eg are linearly disjoint.
Exercise 13.7
a) Assume there only are finitely many irreducibles f 1 , . . . , f r and consider g =
1 + f 1 ¨ . . . ¨ f r . Clearly none of the f i ’s divides g, but because k[t] is a ufd, g is a
product of irreducibles, contradiction.
b) Assume that k(t1 , . . . , tn ) is finitely generated and let f 1 , . . . , f r be generators.
If gi is the denominator of f i , the deominator of every element is of the form
ν
g11 ¨ . . . ¨ gnνn which have only finitely many different irreducible factors, but
according to a) there are infinitely many irreducible elements which of course
can occur as denominator (simply as 1/ f ).
c) We keep the notation in the hint, but choose a basis ei so that e1 = 1. Each
α ř
monomial f α = f 1 1 ¨ . . . ¨ f rνr = cα,i ei where cα,i is a polynomial in the aij ’s and
the bijl ’s. This follows by induction of the degree of f :
ÿ ÿ ÿ
fj ¨ f α = aij cα,s ei es = ( aij cα,s bisl )el
i,s l i,s

Hence every element of L is of the same form, and it belongs to K when all
terms vanish except the the one corresponding to e1 .
d) Let m be a maximal ideal in k [ x1 , . . . , xn ]. Then L = k[ x1 , . . . , xn ]/m is of finite
type over k, and if trdegk L = r, it will be finite over a subfield K shaped like
K = k(t1 , .., tr ). Hence K will be of finite type, which is not the case when r ą 0.
Exercise 13.9 Let A = k [ x, y] with constituting relation xy = y(y ´ 1) = 0 Then
dim A = 1, and in Ay one has x = y ´ 1 = 0; so Ay » k and is of dimension 0.
Exercise 13.12 If A1 Ď A and B1 Ď B are subalgebras, one has the sequence of inclusions
A1 bk B1 Ď A1 bk B Ď Abk B; the first because A1 is flat over k and the second because B
ř
is (remember, all algebras over fields are flat). Assume then that f = iP I ai bbi and
ř
g = jP J c j bd j are two elements in Abk B. The idea is to replace A by the subalgebra
A1 = k [ ai , c j |i P I, j P J ] and B by B1 = k[bi , d j |i P I, j P J ], which both are of finite type
over k and both f and g are contained in their tensor product. The proposition gives
that A1 bk B1 is a domain (resp. reduced) when A and B are. Since f P A1 and g P B1 ,
it follows that f g ‰ 0 (resp. f n ‰ 0) in A1 bk B1 when f and g are non-zero, and hence
f g ‰ 0 (resp. f n ‰ 0) in Abk B in view of the inclusions above.
Exercise 13.13

10th January 2021 at 11:32am

Version 4.0 run 30
460 solutions

a) That the map extends is just saying that the assignment eb f is bilinear and it is
a ring map as e1 b f 1 ¨ eb f = ee1 b f f 1 which is sent to ee1 f f 1 = (e f ) ¨ (e1 f 1 ).
b) Both E and F being algebraic, EF will be algebraic; indeed, if x1 , . . . , xr P E and
y1 , . . . , ys P F, it follows that k[ x1 , . . . xr , y1 . . . ys ] is integral over k (the integral
closure is a ring) and hence it is a field by the basic lemma 12.28 on page 327.
It follows that the union of all such extensions, which is a directed union, is a
field, and it must be equal to the compositum EF. This shows that each element
ř
in EF is shaped like a finite sum i xi yi with xi P E and yi P F (not the same as
the ones above) and consequently φ is surjective.
ř
Each element z in the tensor product may be written as a finite sum z = ei b f i
with ei ’s linear independent. If E an F are linearly disjoint, they stay independent
ř
over F in EF, and φ(z) = f i ei is non-zero unless all the f i ’s vanish, and then
of course z = 0. On the other hand, assume that φ is injective and that the ei ’s
are linearly independent over k; they can then be extended to a k-basis for E. If
ř ř
there is a relation g j ei = 0, it follows as φ is injective, that gi b f i = 0. If t f j u
ř
is a basis for f , each gi is of shape gi = j aij f j so that
ÿ ÿ
0= ei bgi = aij ei b f j
i i,j

but by Proposition 6.21 on page 155, the decomposable tensors f i b f j form a

basis for EbF, and it follows that aij = 0.
c) If EbF is a field the map φ is an isomorphism since algebra homomorphisms
from fields automatically are injective. The implication the other way is trivial.

Solutions for exercises in Chapter 14

Exercise 14.1
a) Consider the short exact sequence

0 (a : x ) A A/a 0 (19.2)

where the map A Ñ A/a sends an element a to the class [ xa] mod a. Flatness of
B over A yields the equality (a : x )b A B = (a : x ) B so that tensorizing sequence
(19.2) by B results in the exact sequence

0 (a : x ) B B B/aB 0,

and hence (a : x ) B = (aB : x ).

b) Assume that x P b, but x R a. Then (a : x ) is a proper ideal in A, and there is a
maximal ideal m containing it. By assumption there is an index i so that mAi

10th January 2021 at 11:32am

Version 4.0 run 30
 461

is proper. Now, we assumed that bAi Ď aAi , and therefore (aAi : x ) = Ai , but
by in vie of a and the flatness of Ai , it holds that (aAi : x ) = (a : x ) Ai Ď mA,
contradiction.
c) For each of the finite number of indices such that bAi differ from aAi chose a
finite set of generators for aAi and one may chose them to lie in a. Let c be the
ideal in A generated all these together with a finite generator set for b. Then
aAi Ď cAi for all i, and by b, it follows that a Ď c. By construction c Ď a, so a = c,
and a is finitely generated.
d) Let a P a be a non-zero element. Then ( a) Ď a with equality at least for all but the
finitely many indices i so that ( a) Ai is proper. Then c ensures that a is finitely
generated.
Exercise 14.3 Assume that there is a relation

an ( x ) f ( x )n + an´1 ( x ) f ( x )n´1 + . . . + a0 ( x ) = 0

with the ai ’s from the field k( x ); we may assume that n is minimal and that the an ( x )
and an´1 ( x ) are polynomials with no common factor. Substituting x d for x and using
that f ( x d ) = f ( x ) ´ x we find

an ( x d ) f ( x )n + (´nan ( x d ) x + an´1 ( x d )) f ( x )n´1 + . . . + a0 x d = 0

Eliminating the dominating term of the two relations, we deduce the equality

an ( x d ) an´1 ( x ) = ´nan ( x d ) an ( x ) x + an´1 ( x d ) an ( x ),

which, since an ( x ) and an´1 ( x ) are relatively prime, implies that an ( x d ) divides an ( x );
but is absurd unless an is constant, say c, which must be non-zero. But then an´1 ( x ) =
´ncx + an´1 ( x d ) which is impossible since k is of characteristic zero.

Solutions for exercises in Chapter 15

Exercise 15.1 The function v is well defines since if x P K has the two representations
x = ab´1 = cd´1 , it holds that ad = cb; thus v0 ( a) + v0 (d) = v0 (c) + v0 (b), or equiva-
lently v0 ( a) ´ v0 (b) = v0 (c) ´ v0 (d). The first axiom follows readily from the definition
of n and that v0 complies to the first axiom.
Assume then we are given two elements x = ab´1 and y = cd´1 from K with
say v(y) ě v( x ); that is, v0 (c) ´ v0 (d) ě v0 ( a) ´ v0 (b); or in other words, v0 (cb) =
v0 (c) + v0 (b) ě v0 ( a) + v0 (d) = v0 ( ad). We have x + y = ( ad + bc)/bd and find

v( x + y) = v0 ( ad + bc) ´ v0 (bd) ě v0 ( ad) ´ v0 (bd) = v0 ( a) ´ v0 (b) = v( x ).

Exercise 15.2 Since A is a ufd, each element x P A can be factorized as x = ps z

with z relatively prime to p in an unambiguous manner, and we may define v p ( x ) = s.

10th January 2021 at 11:32am

Version 4.0 run 30
462 solutions

Clearly v p ( xy) = v p ( x ) + v p (y), and if x = ps z and y = pt w with say s ě t, it holds

that x + y = pt ( ps´t z + w) so that v p ( x + y) ě t and the second requirement is also
fulfilled. To see that v p is unique, it suffices to see that if v is another valuation of our
kind, then v( x ) = 0 whenever x is relatively prime to p; so assume that x P A is such
that ( x, p) = 1. Thence there is a relation 1 = ax + bp; from iii) of Lemma 15.2 we have
v( x ) = v( a´1 (1 ´ bp)) = ´v( a) ď 0, and since v( x ) ě 0, it follows that v( x ) = 0.
Exercise 15.4 (dvr’s are maximal subrings) This is just the fact that an over ring B of A
is of the form Ap where p is a non-zero prime ideal, but when A is dvr, the maximal m
ideal is the only no-zero prime ideal, Am = A.
Ş
Exercise 15.5 Clearly x belongs to the intersection i Ai with each Ai being a valuation
ring, it hods that x lies in each Ai and hence x´1 belongs to each Ai as well. If x lies in
the union i Ai , it lies in one of the Ai ’s and hence x´1 lies in then same Ai (the only
Ť

use of the hypothesis that the union be directed is to ensure that the union is a ring).
Exercise 15.6 (Chevalley’s lemma) Assume that neither aA[ x ] nor aA[ x´1 ] is proper.
Then there are relations of minimal degrees

1 = a0 + a1 x + . . . + a n x n
1 =b0 + b1 x´1 + . . . + bm x´m

where ai ’s and bi ’s belong to A. The situation is symmetric in x and x´1 so we may

certainly assume that n ě m, which enables us to eliminate the x n -term: just multiply
´1 x n to obtain an expression of a x n in terms of lower powers
the last equality by an bm n
of x, which can be substituted into the first equality and gives a relation of degree
less than n. (Note that if n ą m, allready multiplying the last equality by x m , gives a
contradiction).
Exercise 15.7 (Existence of places) Say that a local subring B of K dominates another
local subring B1 if B1 Ď B and mB X B1 = mB1 . This a partial order on the set of local
Ť
subrings of K. If tBi u is an ascending chain in this order, the union i Bi will be a local
Ť
subsring of K. The union is directed and so is a ring, and we assert that m = i mBi is
maximal ideal. Indeed, since the maximal ideals mBi form a chain m is an ideal (if j ď i,
it holds that mBj Ď mBi X Bj Ď mBi ); and it is clearly maximal since if x R m, it holds that
x is invertible in each Bi to which it belongs.
So consider the set Σ of local subrings B such that mB X A = p. It is not empty
having the localization Ap as a member, and by what we just did, any chain in Σ has an
upper bund in Σ (namely its union). Thus, by Zorn’s lemma there is a maximal element
B in Σ. We contend that B is a valuation ring. Indeed, pick a non-zero element x P K.
Citing Chevalley’s lemma from the previous exercise we know that either mB B[ x ] or
mB B[ x´1 ] is a proper ideal; by symmetry we may assume that the former is. Choose a

10th January 2021 at 11:32am

Version 4.0 run 30
 463

maximal ideal n in B[ x ] containing mB B[ x ]. Then B[ x ]n is a local ring whose maximal

ideal nB[ x ]n intersects B in a proper ideal containing m, hence the intersection equals m
since m is maximal. By maximality of B, it follows that B = B[ x ]n and hence x P B.
Exercise 15.12 We proceed by induction on ν; the case ν = 1 is clear since p = pAp X A
for any prime ideal. It will suffice to establish the inclusion p(ν) Ď pν , the other one
is trivial. So let a P pν Ap X A = ( x ν ) Ap X A; this means that a = s´1 bx ν with s R p
and b P A, or equivalently tsa = tbx ν for some t R p. It holds that that a P p, and we
may write a = cx with c P A; hence tscx = tbx ν , and cancelling x (which is a non-zero
divisor) we find tsc = tbx ν´1 . It follows that c P p(ν´1) , which by induction equals pν´1 .
Thus c = dx ν´1 for some d P A, and consequently a = dx ν .
Exercise 15.17
a) One easily sees that m = (2, d ´ 1) so that the quotient map Z[ d] Ñ Z[ d]/m
‘ ‘ ‘

factors by F2 [ d]/( d ´ 1) = F2 [t]/(t2 ´ d, t ´ 1), but d ” 1 mod 2, so that

‘ ‘

(t2 ´ d, t ´ 1) = (t2 ´ 1, t ´ 1) = (t ´ 1) and hence F2 [ d]/( d ´ 1) = F2 .

‘ ‘

b) We relay on the isomorphism m/m2 » mAm /m2 Am from xxxx, and the fact that
dim A/m mAm /m2 Am equals the minimal number of generators for mAm . So
let us compute m2 ; and using that d ” 1 mod 4 we find that (2, d ´ 1)2 =
‘

(4, 2 d ´ 2, d + 1 ´ d) = (4, 2( d ´ 1)); in other words, m/m2 » 2F2 as a

‘ ‘ ‘

vectorspace.
c) If Clearly m X Z = (2), and by Lying–Over any other maximal ideal n intersects
Z in an ideal different from (2), Hence 2 R n and by consquent Z[ d]n =
‘

Z[(1 + d)/2]n and latter is a dvr since Z[(1 + d)/2] is normal.

‘ ‘

Exercise 15.29 We contend that m is generated by any element x so that v( x ) =

(0, . . . , 0, 1). The maximal ideal m corresponds to the minimal isolated subgroup (0),
hence m equals the set of elements y so that the first non-zero coordinate is positive. It
follows that v( xy´1 ) ą 0.

Solutions for exercises in Chapter 16

Exercise 16.3 Let x and y be homogenous elements with deg x = a ą 0 and deg y =
´b ă 0. Then deg x b y a = 0; so x b y a P R0 = k and is thus invertible. It follows that
x is invertible. That every non-zero homogenous element is invertible has the effect
that I = t deg x | x P R homogenous u is an ideal: the only pertinent point is that I is
closed under sign change as ´ deg x = deg x´1 . Let now r be the positive generator
of the I. Then r = deg w for some w. If x P R is any homogeous element it holds that
deg x = a deg w for some a: hence x´1 w a P k and x = αw a for some scalar α. That is
R = k[w, w´1 ].
Exercise 16.4

10th January 2021 at 11:32am

Version 4.0 run 30
464 solutions

a) It suffices to see thatfor all maximal ideals m the localized complex (C‚ )m is
exact. So we may assume that R is local with; let the maximal ideal be m. If all
the xij ’s belonged to m, the three minors would belong to it too, but they do
not since a = R. After as series of elementary row and column operations and
possible a renaming, we may assume that x00 is invertible, and can then resort
to the argument in the Example.
b) Localize in x. Then ax = R x and the localized Hilbert-Burch (C‚ ) x complex is
exact. So ker M is killed by a power of x, and as it is contained in a free module,
it follows that ker M = 0 since x is a non-zero divisor.
c) Let x, y be the regular sequence in a. Since x is non-zero divisor, H 2 C‚ = 0. Now
y is non-zero divisor in R/( x ) R lying in a/( x ) R. Hence H 2 (C‚ /xC‚ ) = 0. The
long exact sequence derived from the short exact sequence of complexes

0 C‚ C‚ C‚ /xC‚ 0,

has the relevant part

x
0 = H 2 (C‚ /xC‚ ) H 1 (C‚ ) H 1 (C‚ ).

Hence multiplication by x is injective, and since a power of x kills H 1 (C‚ ), it

ensues that H 1 (C‚ ) = 0.
Exercise 16.5 Any pair of minors does not have common factors (all three are
irreducible). Hence any two form a regular sequence in a, and the Hilbert–Burch
complex is a resolution of A. It follows that the Hilbert polynomial is given as
     
t+3 t+1 t
χ A (t) = ´3 +2 = 3t + 1.
3 3 3

Solutions for exercises in Chapter 17

Solutions for exercises in Chapter 18

10th January 2021 at 11:32am

Version 4.0 run 30