(MOSTLY) COMMUTATIVE

ALGEBRA

Antoine Chambert-Loir
Antoine Chambert-Loir
E-mail : [email protected]

English translation in progress - Version July 9, 2020

Mais je ne m’arrête point à expliquer ceci plus en détail, à cause
que je vous ôterais le plaisir de l’apprendre par vous-même, et
l’utilité de cultiver votre esprit en vous exerçant...
René Descartes (1596-1659), La géométrie (1638)
CONTENTS

Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ix
1. Rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1. Definitions. First examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
1.2. Nilpotent elements; regular and invertible elements; division
rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
1.3. Algebras, polynomials . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
1.4. Ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
1.5. Quotient rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 33
1.6. Fraction rings of commutative rings . . . . . . . . . . . . . . . . . . . . . . . . 39
1.7. Relations between quotient rings and fraction rings . . . . . . . . . 47
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
2. Ideals and divisibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
2.1. Maximal ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.2. Maximal and prime ideals in a commutative ring . . . . . . . . . . . 67
2.3. Hilbert’s Nullstellensatz . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
2.4. Principal ideal domains, euclidean rings . . . . . . . . . . . . . . . . . . . 82
2.5. Unique factorization domains . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
2.6. Polynomial rings are unique factorization domains . . . . . . . . . 94
2.7. Resultants and another theorem of Bézout . . . . . . . . . . . . . . . . . . 98
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
3. Modules. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 115
3.1. Definition of a module . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 116
3.2. Morphisms of modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 122
3.3. Operations on modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125
3.4. Quotients of modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
vi CONTENTS

3.5. Generating sets, free sets; bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . 134

3.6. Localization of modules (commutative rings) . . . . . . . . . . . . . . . 140
3.7. Vector spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
3.8. Alternate multilinear forms. Determinants . . . . . . . . . . . . . . . . . 152
3.9. Fitting ideals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 164
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
4. Field extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
4.1. Integral elements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176
4.2. Integral extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
4.3. Algebraic extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
4.4. Separability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
4.5. Finite fields. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198
4.6. Galois’s theory of algebraic extensions . . . . . . . . . . . . . . . . . . . . . 200
4.7. Norms and traces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
4.8. Transcendence degree . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
5. Modules over principal ideal rings . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
5.1. Matrix operations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 228
5.2. Application to linear algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 232
5.3. Hermite normal form . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 237
5.4. Finitely generated modules over a principal ideal domain . . . 245
5.5. Application: Finitely generated abelian groups . . . . . . . . . . . . . 253
5.6. Application: Endomorphisms of a finite dimensional vector
space . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 264
6. Further results on modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
6.1. Nakayama’s lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
6.2. Length . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
6.3. The noetherian property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
6.4. The artinian property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
6.5. Support of a module, associated ideals . . . . . . . . . . . . . . . . . . . . . 297
6.6. Primary decomposition . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 306
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
7. First steps in homological algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . 323
 CONTENTS vii

7.1. Diagrams, complexes and exact sequences . . . . . . . . . . . . . . . . . . 324

7.2. The “snake lemma”. Finitely presented modules . . . . . . . . . . . 328
7.3. Projective modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 334
7.4. Injective modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
7.5. Exactness conditions for functors. . . . . . . . . . . . . . . . . . . . . . . . . . . 348
7.6. Adjoint functors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 357
7.7. Differential modules. Homology and cohomology . . . . . . . . . . 363
Exercises. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 371
8. Tensor products and determinants . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
8.1. Tensor product of two modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
8.2. Tensor product of modules over a commutative ring . . . . . . . . 389
8.3. Tensor algebras, symmetric and exterior algebra . . . . . . . . . . . . 394
8.4. The exterior algebra and determinants . . . . . . . . . . . . . . . . . . . . . 401
8.5. Adjunction and exactness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 407
8.6. Flat modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 411
8.7. Faithful flatness . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 421
8.8. Faithfully flat descent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
8.9. Galois descent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 444
9. Algebras of finite type over a field . . . . . . . . . . . . . . . . . . . . . . . . . . . . 451
9.1. Noether’s normalization theorem . . . . . . . . . . . . . . . . . . . . . . . . . . 452
9.2. Dimension and transcendence degree . . . . . . . . . . . . . . . . . . . . . . 458
9.3. Krull’s Hauptidealsatz and applications . . . . . . . . . . . . . . . . . . . . 463
9.4. Heights and dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
9.5. Finiteness of integral closure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 471
9.6. Dedekind rings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 472
Exercises . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480
Appendix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
A.1. Algebra . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 490
A.2. Set Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 493
A.3. Categories . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 509
Index. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
PREFACE

This book stems out lectures on commutative algebra for 4th-year

university students of two French universities (Paris and Rennes). At that
level, students have already followed a basic course in linear algebra and
are essentially fluent with the language of vector spaces over fields. The
new topics to be introduced were arithmetic of rings, modules, especially
principal ideal rings and the classification of modules over such rings,
Galois theory, as well as an introduction to more advanced topics such as
homological algebra, tensor products, and algebraic concepts involved
in algebraic geometry.
Rewriting the text in view of its publication, I have been led to
reorganizing or expanding many sections, for various motivations which
range from giving interesting applications of a given notion, adopting
a more natural definition which would be valid in a broader context,
to calming the anxiety of the author that “this is not yet enough”. For
example, the “(Mostly)” of the title refers to the fact that basic notions
of rings, ideals, modules are equally important in the context of non-
commutative rings — important in the sense that this point of view is
definitely fruitful in topics such as representation theory — but was
initially motivated by the (accordingly naïve) desire of being able to say
that a structure of an A-module on an abelian group M is a morphism
of rings from A to the ring of endomorphisms of M, the latter ring being
non-commutative in general.
This makes the present book quite different from classic textbooks
on commutative algebra like Atiyah/Macdonald’s, Matsumura’s or
Eisenbud’s. I am certainly less terse than the first two, and won’t go as
far as the last two. Doing so, I believe that it will be already accessible
x PREFACE

to younger (undergraduate) students, while giving them an idea of

advanced topics which are usually not discussed at that level.
This book has also not been written as an absolute treatise (such
as Bourbaki’s chapters on the topic). While I have taken the time to
explain basic definitions and elementary examples, its reading probably
presupposes some habit of abstract algebra. Sometimes, it even assumes
some understanding of parts which are discussed in detail only later,
but which certainly were the object of former study by the reader.
Consequently, and this is especially true for the first chapters, the reader
may be willing to skip some sections for a first reading. Its 500 pages of
the book don’t make it fit for a 15-week course anyway. Another hope is
that the reader will enjoy more advanced results of each chapter at the
occasion of another visit.

The main protagonists of the book are rings, even commutative rings,
viewed as generalized numbers in which we have an addition and a
multiplication. The first chapter introduces them, as well as the language
that is needed to understand their properties. Three ways to forge new
rings from old ones are introduced: polynomials, quotients and fractions.
In the second chapter, I study divisibility in general rings. As was
discovered in the xixth century, classical properties of integers such as
unique factorization, may not hold anymore and this leads to different
attitudes. One can either set up new structures — and this leads to the
study of prime or maximal ideals — or restrict oneself to rings that satisfy
that familiar properties: then, principal ideal domains, euclidean rings,
unique factorization domains come in. First applications to algebraic
geometry already appear in this chapter: Hilbert’s Nullstellensatz and
Bézout’s theorem on the number of intersection points of plane curves.
Modules are to rings what vector spaces are to fields. This third
chapter thus revisits classic notions linear algebra in this broader context
— operations (intersection, direct sums, products, quotients), linear
independence, bases, (multi-)linear forms. . . — but also new notions
such as Fitting ideals. Conversely, I have found interesting to retell the
story of vector spaces from this general perspective, establishing the
existence of bases and the well-definition of dimension. Nakayama’s
lemma is also discussed in this chapter, as a tool that sometimes allows
 PREFACE xi

to study of modules over general lings by reduction to the case linear

algebra over fields.
The fourth chapter presents field extensions. While its culminating
point is Galois’s theory, we however do not discuss classical “number-
theoretical” applications of Galois’s theory, such as solvability by radicals
or geometric constructions with compass and straight-edge, and for
which numerous excellent books already exist. The reader will also
observe that the first section of this chapter studies integral dependence
in the context of rings; this is both motivated by the importance of this
notion in later chapter, and by the fact that it emphasizes in a deeper
way how linear algebra is used to study algebraic/integral dependence.
The description of module over fields, that is, of vector spaces, is par-
ticularly easy: they are classified by a single invariant, their dimension,
which in the finitely generated case, is an integer. The next case where
such an explicit description is possible is the object of the fifth chapter,
namely the classification of modules over a principal ideal ring. This classi-
fication can be obtained in various ways and we chose the algorithmic
approach allowed by the Hermite forms of matrices; it also has the
interest to furnish information on the structure of the linear group over
euclidean domains. Of course, we give the two classical applications
of the general theory, to finitely generated abelian groups (when the
principal ideal domain is Z) and to the reduction of endomorphisms
(when it is the ring 𝑘[T] of polynomials in one indeterminate over a
field 𝑘).
Chapter six introduces various tools to study a module over a general
ring. Modules of finite length, or possessing the noetherian, or the artinian
properties, are analogues to finitely dimensional vector spaces over a
field. I then introduce the support of a module, its associated prime
ideals, and establish the existence of a primary decomposition. These
sections have a more geometric flavour which and can be thought as a
first introduction to the algebraic geometry of schemes.
Homological algebra, one of the important concepts invented in the
second half of xxth century, allows to systematically quantify the defect
of injectivity/surjectivity of a linear map, and is now an unavoidable
tool in algebra, geometry, topology,. . . The introduction to this topic I
xii PREFACE

provide here goes in various directions. I define projective and injective

modules, and use them as a pretext for elaborating on the language of
categories, but fall short of defining resolutions and Tor/Ext functors in
general.
The next chapter is devoted to the general study of tensor products. It is
first used to develop the theory of determinants, via the exterior algebra.
The general lack of exactness of this operation leads to the definition of a
flat module and, from that point, to the study of faithfully flat descentand
Galois descent. This theory, invented by Alexander Grothendieck, formal-
izes the following question: imagine you need additional parameters to
define an object, how can you get rid of them?
Algebras of finite type over a field are the algebraic building blocks of
algebraic geometry, and some of the properties studied in this last
chapter, Noether’s normalization theorem, dimension, codimension,
have an obvious geometric content. Algebraically, these rings are also
better behaved than general noetherian rings. A final section is devoted
to Dedekind rings; an algebraic counterparts of curves, they are also very
important in number theory and I conclude the chapter by proving the
finiteness of the class groups of number field.
In the appendix, I first summarize general mathematical conventions.
The second section proves two important results in set theory that are
used all over the text: the Cantor-Bernstein-Schröder theorem and the
Zorn theorem. The main part of the book makes use of the language
of categories and functors, in a hopefully progressive and pedagogical
way, and I have summarized it in a last section.
Every chapter comes with a large supply of exercises of all kind
(around 300 in total), ranging from a simple application of concepts to
additional results that I couldn’t include in the text.
I would like to thank François Loeser, with whom I first taught this
course, for his trust and friendship all over these years, Yuri Tschinkel
who brought it to a publisher, Yves Laszlo and Daniel Ferrand for their
suggestions at various times of the elaboration of the book.
I also thank Colas Bardavid, Pierre Bernard, Samit Dasgupta, Brandon
Gontmacher, Ondine Meyer, Andrey Mokhov for having indicated me a
few unfortunate typos.
CHAPTER 1

RINGS

Rings, the definition of which is the subject of this chapter, are algebraic
objects in which one can compute as in classical contexts, Integers, real numbers
or matrices: one has an addition, a multiplication, two symbols 0 and 1, and
the usual computation rules are satisfied.
The absence of a division, however, gives rise to various subtleties. For
example, the product of two elements can be zero, while none of them is zero, or
the square of a non-zero element can be zero; the reader has probably already
seen such examples exist in matrix rings. Fields, or division rings, in which
every non-zero element is invertible, have more familiar algebraic properties.
Some rings may be given by Nature, but it is the task of a mathematician,
be he/she an apprentice mathematician, to construct new rings from old ones.
To that aim, this chapter proposes three different tools, each of them having a
different goal:
– Polynomial rings allow to add new “indeterminate” elements to a ring,
which are not supposed to satisfy any property but be computed with. We
introduce them and take the opportunity to review a few facts concerning
polynomials in one indeterminate with coefficients in a field and their roots.
– Ideals embody in general rings the classic concepts of divisibility and
congruences that the reader is certainly familiar with when it comes about
integers. Given a (two-sided) ideal I of a ring A, I construct the quotient ring
A/I where the elements of I are formally forced to be “equal to zero”, while
keeping intact the standard computation rules in a ring.
– On the opposite, fraction rings are defined so as forcing given elements to
be invertible. A particular case of this construction is the field of fractions of an
integral domain.
2 CHAPTER 1. RINGS

1.1. Definitions. First examples

Definition (1.1.1). — A ring is a set A endowed with two binary laws, an
addition (𝑎, 𝑏) ↦→ 𝑎 + 𝑏, and a multiplication (𝑎, 𝑏) ↦→ 𝑎𝑏, satisfying the
following axioms.
a) Axioms concerning the addition law and stating that (A, +) is an
abelian group:
(i) For every 𝑎, 𝑏 in A, 𝑎 + 𝑏 = 𝑏 + 𝑎 (commutativity of addition);
(ii) For every 𝑎, 𝑏, 𝑐 in A, (𝑎 + 𝑏) + 𝑐 = 𝑎 + (𝑏 + 𝑐) (associativity of
addition);
(iii) There exists an element 0 ∈ A such that 𝑎 + 0 = 0 + 𝑎 = 𝑎 for every
𝑎 in A (neutral element for the addition);
(iv) For every 𝑎 ∈ A, there exists 𝑏 ∈ A such that 𝑎 + 𝑏 = 𝑏 + 𝑎 = 0
(existence of an opposite for the addition);
b) Axioms concerning the multiplication law and stating that (A, ·) is
a monoid:
(v) There exists an element 1 ∈ A such that 1𝑎 = 𝑎1 = 𝑎 for every 𝑎 ∈ A
(neutral element for the multiplication);
(vi) For every 𝑎, 𝑏, and 𝑐 in A, (𝑎𝑏)𝑐 = 𝑎(𝑏𝑐) (associativity of
multiplication);
(vii) Axiom relating the addition and the multiplication:
(viii) For every 𝑎, 𝑏, and 𝑐 in A, 𝑎(𝑏+𝑐) = 𝑎𝑏+𝑎𝑐 and (𝑏+𝑐)𝑎 = 𝑏𝑎+𝑐𝑎
(multiplication distributes over the addition).
c) One says that the ring A is commutative if, moreover,
(viii) For every 𝑎 and 𝑏 in A, 𝑎𝑏 = 𝑏𝑎 (commutativity).

With their usual addition and multiplication, rational integers, real

numbers, or matrices are fundamental examples of rings. In fact, the
ring axioms specify exactly the relevant computation rules to which one
is accustomed to. We shall give more examples in a moment, but we first
explicit a few computation rules which follow from the stated axioms.

1.1.2. — Let A be a ring.

Endowed with the addition law, A is in particular an abelian group.
As a consequence, it admits exactly one zero element, which is usually
 1.1. DEFINITIONS. FIRST EXAMPLES 3

denoted by 0, or by 0A if it is necessary to specify of what ring it is the

zero element. Moreover, any element 𝑎 ∈ A has exactly one opposite
for the addition, and it is denoted by −𝑎. Indeed, if 𝑏 and 𝑐 are two
opposites, then 𝑏 = 𝑏 + 0 = 𝑏 + (𝑎 + 𝑐) = (𝑏 + 𝑎) + 𝑐 = 𝑐.
For every 𝑎 ∈ A, one has 𝑎 · (0 + 0) = 𝑎 · 0 + 𝑎 · 0, hence 𝑎 · 0 = 0; similarly,
0 · 𝑎 = 0. Then, for any 𝑎, 𝑏 ∈ A, one has 𝑎(−𝑏) + 𝑎𝑏 = 𝑎(−𝑏 + 𝑏) = 𝑎 · 0 = 0,
hence 𝑎(−𝑏) = −(𝑎𝑏); similarly, (−𝑎)𝑏 = −(𝑎𝑏). Consequently, there is no
ambiguity in writing −𝑎𝑏 for either (−𝑎)𝑏, −(𝑎𝑏) or 𝑎(−𝑏).
For any rational integer 𝑛, one defines 𝑛𝑎 by induction, by setting
0𝑎 = 0, and 𝑛𝑎 = 𝑎 + (𝑛 − 1)𝑎 if 𝑛 > 1 and 𝑛𝑎 = −(−𝑛)𝑎 if 𝑛 6 −1.
Observe that for any 𝑎, 𝑏 ∈ A and any rational integer 𝑛, one has
𝑎(𝑛𝑏) = 𝑛(𝑎𝑏) = (𝑛𝑎)𝑏. This is proved by induction on 𝑛: if 𝑛 = 0, then all
three terms are 0; if 𝑛 > 1, then 𝑎(𝑛𝑏) = 𝑎(𝑏 +(𝑛 −1)𝑏) = 𝑎𝑏 + 𝑎((𝑛 −1)𝑏) =
𝑎𝑏 + (𝑛 − 1)(𝑎𝑏) = 𝑛𝑎𝑏 and similarly for the other equality; finally, if
𝑛 6 −1, then 𝑎(𝑛𝑏) = 𝑎(−((−𝑛)𝑏)) = −𝑎((−𝑛)𝑏) = (−𝑛)(−𝑎𝑏) = 𝑛𝑎𝑏.
Similarly, the multiplicative monoid (A, ·) has exactly one neutral
element, usually denoted by 1, or by 1A if it is necessary to specify the
ring, and called the unit element of A.
If 𝑎 belongs to a ring A and 𝑛 is any positive1 rational integer 𝑛 > 0,
one defines 𝑎 𝑛 by induction by setting 𝑎 0 = 1, and, if 𝑛 > 1, 𝑎 𝑛 = 𝑎 · 𝑎 𝑛−1 .
For any integers 𝑚 and 𝑛, one has 𝑎 𝑚+𝑛 = 𝑎 𝑚 𝑎 𝑛 and (𝑎 𝑚 )𝑛 = 𝑎 𝑚𝑛 , as can
be checked by induction.
However, one should take care that 𝑎 𝑛 𝑏 𝑛 and (𝑎𝑏)𝑛 are generally
distinct, unless 𝑎𝑏 = 𝑏𝑎, in which case one says that 𝑎 and 𝑏 commute.

Proposition (1.1.3) (Binomial formula). — Let A be a ring, let 𝑎 and 𝑏 be

elements in A such that 𝑎𝑏 = 𝑏𝑎. Then, for any positive integer 𝑛,
𝑛  
Õ 𝑛
(𝑎 + 𝑏)𝑛 = 𝑎 𝑘 𝑏 𝑛−𝑘 .
𝑘
𝑘=0

Proof. — The proof is the one that is well known when 𝑎, 𝑏 are rational
integers. It runs by induction on 𝑛. When 𝑛 = 0, both sides are equal

1Recall that in this book, positive means greater or equal to 0, and negative means smaller or equal
to 0.
4 CHAPTER 1. RINGS

to 1. Assume the formula holds for 𝑛; then,

𝑛  
Õ 𝑛
(𝑎 + 𝑏)𝑛+1 = (𝑎 + 𝑏)(𝑎 + 𝑏)𝑛 = (𝑎 + 𝑏) 𝑎 𝑘 𝑏 𝑛−𝑘
𝑘
𝑘=0
𝑛   𝑛  
Õ 𝑛 Õ 𝑛
= 𝑎 𝑘+1 𝑏 𝑛−𝑘 + 𝑏𝑎 𝑘 𝑏 𝑛−𝑘
𝑘 𝑘
𝑘=0 𝑘=0
𝑛 𝑛
𝑛 𝑘+1 𝑛−𝑘 𝑛 𝑘 𝑛+1−𝑘
Õ   Õ  
= 𝑎 𝑏 + 𝑎 𝑏
𝑘 𝑘
𝑘=0 𝑘=0
𝑛+1  𝑛
𝑛 Õ 𝑛
Õ   
𝑘 𝑛+1−𝑘
= 𝑎 𝑏 + 𝑎 𝑘 𝑏 𝑛+1−𝑘
𝑘−1 𝑘
𝑘=1 𝑘=0
𝑛+1 
𝑛 𝑛
Õ   
= + 𝑎 𝑘 𝑏 𝑛+1−𝑘
𝑘−1 𝑘
𝑘=0
𝑛+1 
𝑛+1
Õ 
= 𝑎 𝑘 𝑏 𝑛+1−𝑘
𝑘
𝑘=0
𝑛
+ 𝑛𝑘 = 𝑛+1 for any positive integers 𝑛 and 𝑘. This concludes




since 𝑘−1 𝑘
the proof by induction on 𝑛. 

Examples (1.1.4). — a) As well-known basic examples of commutative

rings, let us mention the ring Z of rational integers, the quotient rings
Z/𝑛Z for 𝑛 > 1, the fields Q of rational numbers, R of real numbers, C of
complex numbers, the ring K[X] of polynomials in one indeterminate X
with coefficients in a field (or a commutative ring) K.
b) Let A be a ring and let S be a set. The set AS of functions from S
to A, with pointwise addition and pointwise multiplication, is a ring.
Explicitly, for 𝑓 and 𝑔 ∈ AS , 𝑓 + 𝑔 and 𝑓 𝑔 are the functions such that
( 𝑓 + 𝑔)(𝑠) = 𝑓 (𝑠) + 𝑔(𝑠) and ( 𝑓 𝑔)(𝑠) = 𝑓 (𝑠)𝑔(𝑠). The set 𝒞(X, R) of all
continuous functions from a topological space X into R is a ring, as well
as the sets 𝒞 𝑘 (Ω, R) and 𝒞 𝑘 (Ω, C) of all functions of differentiability
class 𝒞 𝑘 from an open subset Ω of R𝑛 into R or C (here, 𝑘 ∈ N ∪ {∞}).
c) Let S be a set and let (A𝑠 )𝑠∈S be a family of rings indexed by 𝑠. Let
Î
A = 𝑠∈S A𝑠 be the product of this family; this is the set of all families
 1.1. DEFINITIONS. FIRST EXAMPLES 5

(𝑎 𝑠 )𝑠∈S , where 𝑎 𝑠 ∈ A𝑠 for all 𝑠. We endow A with termwise addition

and termwise multiplication: (𝑎 𝑠 ) + (𝑏 𝑠 ) = (𝑎 𝑠 + 𝑏 𝑠 ) and (𝑎 𝑠 ) · (𝑏 𝑠 ) = (𝑎 𝑠 𝑏 𝑠 ).
Then A is a ring; its zero element is the null family (0A𝑠 ), its unit element
is the family (1A𝑠 ).

Let us now give non-commutative examples.

Examples (1.1.5). — a) Let A be a ring, let 𝑛 be a positive integer and

let M𝑛 (A) be the set of 𝑛 × 𝑛 matrices with coefficients in A, endowed
with the usual computation rules: the sum of two matrices P = (𝑝 𝑖,𝑗 )
and Q = (𝑞 𝑖,𝑗 ) is the matrix R = (𝑟 𝑖,𝑗 ) such that 𝑟 𝑖,𝑗 = 𝑝 𝑖,𝑗 + 𝑞 𝑖,𝑗 for
all 𝑖 and 𝑗 in {1, . . . , 𝑛}; the product of these matrices P and Q is the
matrix S = (𝑠 𝑖,𝑗 ) given by
𝑛
Õ
𝑠 𝑖,𝑗 = 𝑝 𝑖,𝑘 𝑞 𝑘,𝑗 .
𝑘=1

Endowed with these laws, M𝑛 (A) is a ring; its zero element is the null
matrix; its unit element is the matrix I𝑛 = (𝛿 𝑖,𝑗 ), where 𝛿 𝑖,𝑗 = 1 if 𝑖 = 𝑗,
and 0 otherwise.
When 𝑛 > 2, or if 𝑛 > 1 and A is not commutative, then the ring
M𝑛 (A) is not commutative.
b) Let G be an abelian group. When 𝜑 are 𝜓 any two endomorphisms
of G, the map 𝑔 ↦→ 𝜑(𝑔) + 𝜓(𝑔) is again an endomorphism of G, written
𝜑 + 𝜓; this endowes the set End(G) of endomorphisms of G with the
structure of an abelian group, whose zero element is the application
𝑔 ↦→ 0. Composition of endomorphisms (𝜑, 𝜓) ↦→ 𝜑 ◦ 𝜓 is an associative
law, and distributes with respect to the addition; the identity map of G,
𝑔 ↦→ 𝑔, is the unit element. Consequently, these laws endow the set
End(G) of all endomorphisms of the group G with the structure of a
ring.
c) Let K be a field (commutative, say). The set EndK (V) of all en-
domorphisms of a K-vector space V is a ring, non-commutative as
soon as dim(V) > 2. Here, the addition law is the pointwise addi-
tion, (𝜑 + 𝜓)(𝑣) = 𝜑(𝑣) + 𝜓(𝑣) for 𝜑, 𝜓 ∈ EndK (V) and 𝑣 ∈ V, while
the multiplication law is the composition of endomorphisms, given by
6 CHAPTER 1. RINGS

𝜑 ◦ 𝜓(𝑣) = 𝜑(𝜓(𝑣)). In fact, EndK (V) is also a K-vector space and the
multiplication is K-linear. One says that EndK (V) is a K-algebra.
If V has finite dimension, say 𝑛, we may choose a basis of V and
identify V with K𝑛 ; then endomorphisms of V identify with matrices of
size 𝑛, and this identifies the rings EndK (V) and M𝑛 (K).

Here is a probably less-known example.

Example (1.1.6). — Let A be a ring and let M be any monoid (a group,

for example).
Inside the abelian group AM of all functions from M to A, let us
consider the subgroup A(M) of functions 𝑓 with finite support, namely
such that 𝑓 (𝑔) ≠ 0 for only finitely many 𝑔 ∈ M.
It also possesses a convolution product, defined by the formula:
Õ
(𝜑 ∗ 𝜓)(𝑔) = 𝜑(ℎ)𝜓(𝑘).
ℎ,𝑘∈M
𝑔=ℎ 𝑘

This product is well-defined: only finitely many terms in the sum are
nonzero, and the convolution of two functions with finite support still
has finite support. Moreover, the convolution product is associative:
indeed, for 𝜑, 𝜓, 𝜃 ∈ A(M) , one has
Õ
(𝜑 ∗ 𝜓) ∗ 𝜃(𝑔) = (𝜑 ∗ 𝜓)(ℎ)𝜃(𝑘)
ℎ,𝑘∈M
𝑔=ℎ 𝑘
Õ Õ
= 𝜑(𝑖)𝜓(𝑗)𝜃(𝑘)
ℎ,𝑘∈M 𝑖,𝑗∈M
𝑔=ℎ 𝑘 ℎ=𝑖𝑗
Õ
= 𝜑(𝑖)𝜓(𝑗)𝜃(𝑘),
𝑖,𝑗,𝑘∈M
𝑔=𝑖𝑗 𝑘

and a similar computation shows that this is as well equal to 𝜑 ∗ (𝜓 ∗ 𝜃)(𝑔).

There is a neutral element for convolution, given by the “Dirac func-
tion” 𝛿 such that 𝛿(𝑔) = 1 for 𝑔 = 𝑒, the neutral element of M, and
 1.1. DEFINITIONS. FIRST EXAMPLES 7

𝛿(𝑔) = 0 otherwise. Indeed, for 𝜑 ∈ A(M) and 𝑔 ∈ M, one has

Õ
𝜑 ∗ 𝛿(𝑔) = 𝜑(ℎ)𝛿(𝑘) = 𝜑(𝑔),
ℎ,𝑘∈M
𝑔=ℎ 𝑘

and similarly, 𝛿 ∗ 𝜑(𝑔) = 𝜑(𝑔).

These laws endow A(M) with the structure of a ring, When A is a
commutative ring, the ring A(M) is also called the monoid algebra of M
(with coefficients in A) (or the group algebra, when M is a group).
For 𝑘 ∈ M, let 𝛿 𝑘 ∈ A(M) be the function such that 𝛿 𝑘 (𝑔) = 1 if 𝑔 = 𝑘,
and 𝛿 𝑘 (𝑔) = 0 otherwise. One has 𝛿 ℎ ∗ 𝛿 𝑘 = 𝛿 ℎ 𝑘 .
Definition (1.1.7). — A subring of a ring A is a subgroup B of A for the
addition, which contains 1, and is stable under the multiplication, so that the
laws of A endow B with the structure of a ring, admitting the same neutral
elements 0 and 1.
Examples (1.1.8). — a) Let A be a ring and let G = A𝑛 , viewed as an
abelian group for addition. With the usual action of matrices on vectors,
M𝑛 (A) is a subset of End(A𝑛 ), with compatible addition, multiplication
and unit element, so that M𝑛 (A) is a subring of End(A𝑛 ).
b) Let A be a ring.
The set Z(A) of all elements 𝑎 ∈ A such that 𝑎𝑥 = 𝑥𝑎 for every 𝑥 ∈ A is
a commutative subring of A, called its center.
More generally, let S be a subset of A and let CS (A) be the set of all
elements 𝑎 ∈ A such that 𝑎𝑥 = 𝑥𝑎 for all 𝑥 ∈ S. This is a subring of A,
called the centralizer of S in A. By definition, S is contained in the center
of CS (A), and CS (A) is the largest subring of A satisfying this property.
c) The intersection of any family of subrings of a ring A is a subring
of A.

1.1.9. — Let A be a ring and let S be a subset of A. Let B be the

intersection of all subrings of A containing S. It is a subring of A; one
calls it the subring of A generated by S.
Lemma (1.1.10). — Let A be a ring and let S be a subset of A; assume that
any two elements of S commute. Then the subring of A generated by S is
commutative.
8 CHAPTER 1. RINGS

Proof. — Let B be this subring. First of all, S is contained in the cen-

tralizer CS of S in A, by assumption, which is a subring of A, hence
B ⊂ CS (A). Then S is contained in the center of CS (A), which is a
subring of CS (A), hence of A, so that B ⊂ Z(CS (A)). Since Z(CS (A)) is a
commutative ring, B is commutative. 

Definition (1.1.11). — Let A and B be two rings. A morphism of rings

𝑓 : A → B is a map that satisfies the following properties:
(i) One has 𝑓 (0) = 0 and 𝑓 (1) = 1;
(ii) For any 𝑎 and 𝑏 in A, one has 𝑓 (𝑎 + 𝑏) = 𝑓 (𝑎) + 𝑓 (𝑏) and 𝑓 (𝑎𝑏) =
𝑓 (𝑎) 𝑓 (𝑏).

Synonyms are homomorphism of rings and ring morphism. An endomor-

phism of the ring A is a ring morphism from A to A.
If A is a ring, the identity map idA : A → A, 𝑎 ↦→ 𝑎, is a ring morphism.
The composition of two ring morphisms is again a ring morphism, and
composition is associative.
Consequently, rings and their morphisms form a category, denoted
by Rings.2
As in general category theory, we say that a morphism of rings
𝑓 : A → B is an isomorphism if there exists a ring morphism 𝑔 : B → A
such that 𝑓 ◦ 𝑔 = idB and 𝑔 ◦ 𝑓 = idA . In that case, there exists exactly
one such morphism 𝑔, and we call it the inverse of 𝑓 ; we also say that the
∼
rings A and B are isomorphic and write A ' B. We also write 𝑓 : A → − B
to say that 𝑓 is an isomorphism of rings from A to B.
If A is a ring, an automorphism of A is an isomorphism from A to A.
The set of all automorphisms of a ring A is a group for composition.

1.1.12. — Let 𝑓 : A → B be a morphism of rings. The image 𝑓 (A) of A

by 𝑓 is a subring of B. The inverse image 𝑓 −1 (C) of a subring C of B is a
subring of A.
Let 𝑓 , 𝑔 : A → B be morphisms of rings. The set C of all 𝑎 ∈ A such
that 𝑓 (𝑎) = 𝑔(𝑎) is a subring of A.

2The theory of categories furnishes a language that is extremly useful to formulate structural
properties of mathematics. We have included a basic summary in appendix A.3.
 1.1. DEFINITIONS. FIRST EXAMPLES 9

Proposition (1.1.13). — A ring morphism is an isomorphism if and only if it

is bijective.

Proof. — Let 𝑓 : A → B be a morphism of rings.

Assume that 𝑓 is an isomorphism, and let 𝑔 be its inverse. The relations
𝑔 ◦ 𝑓 = idA and 𝑓 ◦ 𝑔 = idB respectively imply that 𝑓 is injective and
surjective, so that 𝑓 is bijective.
Conversely, let us assume that 𝑓 is bijective and let 𝑔 be its inverse. We
have 𝑔 ◦ 𝑓 = idA and 𝑓 ◦ 𝑔 = idB , so we just need to show that 𝑔 is a ring
morphism from B to A. Since 𝑓 (0) = 0 and 𝑓 (1) = 1, one has 𝑔(0) = 0
and 𝑔(1) = 1. For any 𝑎 and 𝑏 ∈ B,

𝑓 (𝑔(𝑎 + 𝑏)) = 𝑎 + 𝑏 = 𝑓 (𝑔(𝑎)) + 𝑓 (𝑔(𝑏)) = 𝑓 (𝑔(𝑎) + 𝑔(𝑏))

and
𝑓 (𝑔(𝑎𝑏)) = 𝑎𝑏 = 𝑓 (𝑔(𝑎)) 𝑓 (𝑔(𝑏)) = 𝑓 (𝑔(𝑎)𝑔(𝑏)).
Since 𝑓 is a bijection, 𝑔(𝑎 + 𝑏) = 𝑔(𝑎) + 𝑔(𝑏) and 𝑔(𝑎𝑏) = 𝑔(𝑎)𝑔(𝑏). 

Example (1.1.14). — Let A be a commutative ring and let 𝑝 be a prime number

such that 𝑝1A = 0. The map 𝜑 : A → A defined by 𝜑(𝑥) = 𝑥 𝑝 is a ring
homomorphism, called the Frobenius homomorphism.
Observe that 𝜑(0) = 0 and 𝜑(1) = 1. Moreover, since A is commutative,
we have 𝜑(𝑥𝑦) = (𝑥𝑦)𝑝 = 𝑥 𝑝 𝑦 𝑝 = 𝜑(𝑥)𝜑(𝑦) for every 𝑥, 𝑦 ∈ A. It remains
to show that 𝜑 is additive. Let 𝑥, 𝑦 ∈ A; by the binomial formula, one
has
𝑝  
Õ 𝑝
𝜑(𝑥 + 𝑦) = (𝑥 + 𝑦)𝑝 = 𝑥 𝑘 𝑦 𝑝−𝑘 .
𝑘
𝑘=0
𝑝

But 𝑘 = 𝑝!/𝑘!(𝑝 − 𝑘)!, a fraction the numerator of which is a multiple
of 𝑝. On the other hand, since 𝑝 is prime, 𝑝 divides neither 𝑘! nor (𝑝 − 𝑘)!,
if 1 6 𝑘 6
𝑝 − 1, hence 𝑝 does not divide 𝑘!(𝑝 − 𝑘)!. Consequently, 𝑝
𝑝 𝑝

divides 𝑘 for 1 6 𝑘 6 𝑝 −1. Since 𝑝1A = 0, this implies 𝑘 1A = 0 for any
𝑝

𝑘 ∈ {1, . . . , 𝑝 − 1} and all these terms 𝑘 𝑥 𝑘 𝑦 𝑝−𝑘 vanish. Consequently,

𝜑(𝑥 + 𝑦) = 𝑥 𝑝 + 𝑦 𝑝 = 𝜑(𝑥) + 𝜑(𝑦),

as was to be shown.
10 CHAPTER 1. RINGS

Example (1.1.15). — Let A be a ring and let 𝑎, 𝑏 be elements of A such

that 𝑎𝑏 = 𝑏𝑎 = 1. (We shall soon say that 𝑎 is invertible and that 𝑏 is its
inverse.) Then, the map 𝑥 ↦→ 𝑎𝑥𝑏 is an automorphism of A, called an
interior automorphism. Any C-linear automorphism of M𝑛 (C) is interior
(exercise 1/11).

Example (1.1.16). — Let A = C[X1 , . . . , X𝑛 ] be the ring of polynomials in 𝑛

variables with coefficients in C. (They will be defined in example 1.3.9.)
Let Q1 , . . . , Q𝑛 be elements in A. Consider the map 𝜑 from A to A which
associates to a polynomial P the polynomial P(Q1 , . . . , Q𝑛 ) obtained
from P by substituting the polynomial Q𝑖 to the indeterminate X𝑖 ; it is
an endomorphism of A. It is the unique endomorphism of A such that
𝜑(X𝑖 ) = Q𝑖 for all 𝑖, and 𝜑(𝑐) = 𝑐 for all 𝑐 ∈ C.
Similarly, for any permutation 𝜎 of {1, . . . , 𝑛}, let Φ𝜎 be the unique
endomorphism of A such that Φ𝜎 (X𝑖 ) = X𝜎(𝑖) for all 𝑖 and Φ𝜎 (𝑐) = 𝑐
for all 𝑐 ∈ C. Observe that Φ𝜎𝜏 (X𝑖 ) = X𝜎(𝜏(𝑖)) = Φ𝜎 (X𝜏(𝑖) ) = Φ𝜎 (Φ𝜏 (X𝑖 ));
therefore, the endomorphisms Φ𝜎𝜏 and Φ𝜎 ◦ Φ𝜏 are equal. The map
𝜎 ↦→ Φ𝜎 is a group morphism from the symmetric group 𝔖𝑛 to the
group Aut(C[X1 , . . . , X𝑛 ]).

1.2. Nilpotent elements; regular and invertible elements; division

rings
Some elements of a ring have nice properties with respect to the
multiplication. This justifies a few more definitions.

Definition (1.2.1). — Let A be a ring. One says that 𝑎 is nilpotent if there

exists an integer 𝑛 > 1 such that 𝑎 𝑛 = 0.

As an example, nilpotent elements of the matrix ring M𝑛 (C) are exactly

the nilpotent 𝑛 × 𝑛-matrices.

Definition (1.2.2). — Let A be a ring. We say that an element 𝑎 ∈ A is left

regular if the relation 𝑎𝑏 = 0 in A implies 𝑏 = 0; otherwise, we say that 𝑎
is a left zero divisor Similarly, we say that 𝑎 is right regular if the relation
𝑏𝑎 = 0 in A only holds for 𝑏 = 0, and that it is a right zero divisor otherwise
An element is said to be regular if it is both left and right regular
 1.2. NILPOTENT ELEMENTS; REGULAR AND INVERTIBLE ELEMENTS; DIVISION RINGS 11

We say that a ring A is a domain if it is commutative, nonzero, and if every

nonzero element of A is regular.

In a commutative ring, an element which is not regular is simply called

a zero divisor.

Definition (1.2.3). — Let A be a ring and let 𝑎 be any element of A.

One says that 𝑎 is right invertible if there exists 𝑏 ∈ A such that 𝑎𝑏 = 1; we
then say that 𝑏 is a right inverse of 𝑎. Similarly, we say that 𝑎 is left invertible
if there exists 𝑏 ∈ A such that 𝑏𝑎 = 1; such an element 𝑏 is called a left inverse
of 𝑎. Finally, we say that 𝑎 is invertible, or a unit, if it is both left and right
invertible.

Assume that 𝑎 is right invertible and let 𝑏 be a right inverse for 𝑎, so

that 𝑎𝑏 = 1. If 𝑐𝑎 = 0, then 𝑐𝑎𝑏 = 0, hence 𝑐 = 0. This shows that 𝑎 is
right regular. Similarly, an element which is left invertible is also left
regular.
Assume that 𝑎 be invertible; let 𝑏 be a right inverse and 𝑐 be a left
inverse of 𝑎. One has 𝑏 = 1𝑏 = (𝑐𝑎)𝑏 = 𝑐(𝑎𝑏) = 𝑐1 = 𝑐. Consequently,
the right and left inverses of 𝑎 are equal. In particular, 𝑎 has exactly one
left inverse and one right inverse, and they are equal. This element is
called the inverse of 𝑎, and usually denoted by 𝑎 −1 .
If the ring is commutative, the notions of left and right regular coincide;
an element which is left invertible is also right invertible, and conversely
Let A× be the set of invertible elements in a ring A.

Proposition (1.2.4). — The set of all invertible elements in a ring A is a group

for multiplication. It is called the group of units of A.
Any ring morphism 𝑓 : A → B induces by restriction a morphism of groups
from A× to B× .

Proof. — Let 𝑎 and 𝑏 be two invertible elements of A, with inverses 𝑎 −1

and 𝑏 −1 . Then, (𝑎𝑏)(𝑏 −1 𝑎 −1 ) = 𝑎(𝑏𝑏 −1 )𝑎−1 = 𝑎𝑎 −1 = 1, so that 𝑎𝑏 is right
invertible, with right inverse 𝑏 −1 𝑎 −1 . Similarly, (𝑏 −1 𝑎 −1 )(𝑎𝑏) = 1, so that
𝑎𝑏 is left invertible too. The multiplication of A induces an associative
law on A× . Moreover, 1 is invertible and is the neutral element for that
law. Finally, the inverse of 𝑎 ∈ A× is nothing but 𝑎 −1 . This shows that
A× is a group for the multiplication
12 CHAPTER 1. RINGS

Let 𝑓 : A → B be a ring morphism. Let 𝑎 ∈ A× and let 𝑏 be its inverse.

Since 𝑎𝑏 = 𝑏𝑎 = 1, one has 1 = 𝑓 (1) = 𝑓 (𝑎) 𝑓 (𝑏) = 𝑓 (𝑏) 𝑓 (𝑎). This shows
that 𝑓 (𝑎) is invertible, with inverse 𝑓 (𝑏). Consequently, the map 𝑓
induces by restriction an application from A× to B× . It maps the product
of two invertible elements to the product of their images, hence is a
group morphism. 
Let A be a commutative ring. Say that two elements 𝑎 and 𝑏 ∈ A are
associated if there exists an invertible element 𝑢 ∈ A× such that 𝑎 = 𝑏𝑢.
The relation of “being associated” is an equivalence relation.

Definition (1.2.5). — One says that a ring A is a division ring (or a division
algebra) if it is not the nullring, and if any nonzero element of A is invertible.
A field is a commutative division ring.3 A subfield of a field is a subring of a
field which is still a field.

A few examples of fields are certainly well-known to the reader, but it

is not obvious that there are at all any noncommutative division rings.
Let us begin by quoting the theorem of Wedderburn according to which
any finite division ring is commutative, i.e., is a field; see exercise 4/19
for its proof, borrowed from the Book Aigner & Ziegler (2014). The ring
of quaternions is probably the most renowned of all noncommutative
division rings; it is in fact the first one to have been discovered, by
Hamilton in 1843.

Example (1.2.6). — The underlying abelian group of the quaternions

is H = R4 ; we write (1, 𝑖, 𝑗, 𝑘) for its canonical basis. Besides the
properties of being associative, having 1 as a neutral element, distributing
addition, and being R-bilinear (that is, 𝑡(𝑎𝑏) = (𝑡𝑎)𝑏 = 𝑎(𝑡𝑏) for any two
quaternions 𝑎, 𝑏 and any real number 𝑡), the multiplication H × H → H
is characterized by the following relations: 𝑖 2 = 𝑗 2 = 𝑘 2 = −1 and 𝑖𝑗 = 𝑘.
Provided these laws give H the structure of a ring, other relations follow
quite easily. Indeed, 𝑖 2 = −1 = 𝑘 2 = (𝑖𝑗)𝑘 = 𝑖(𝑗𝑘), hence 𝑖 = 𝑗𝑘 after

3Terminology is not always consistent among mathematicians, and is sometimes chosen in

order to avoid the repetition of adjectives. Some books call fields what we call division rings;
some other use the awkward expression “skew field” to talk about division rings, but then a
skew field is not a field. Since fields, ie, commutative division rings, will play a larger role in
this book than general division rings, it is useful to have a shorter name for them.
 1.2. NILPOTENT ELEMENTS; REGULAR AND INVERTIBLE ELEMENTS; DIVISION RINGS 13

multiplying both sides by −𝑖; the equality 𝑗 = 𝑘𝑖 is proved similarly; then,

𝑘 𝑗 = 𝑘(𝑘𝑖) = 𝑘 2 𝑖 = −𝑘, 𝑖𝑘 = 𝑖(𝑖𝑗) = 𝑖 2 𝑗 = −𝑗, and 𝑗𝑖 = 𝑗(𝑗𝑘) = 𝑗 2 𝑘 = −𝑘.
It is thus a remarkable discovery that “Hamilton’s multiplication table”

1 𝑖 𝑗 𝑘
1 1 𝑖 𝑗 𝑘
𝑖 𝑖 −1 𝑘 −𝑗
𝑗 𝑗 −𝑘 −1 𝑖
𝑘 𝑘 𝑗 −𝑖 −1

gives rise to a ring structure on H. Properties of the addition follow

from the fact that H = R4 is a vector space. Only the products of the
basic vectors are defined, and distributivity is basically a built-in feature
of multiplication. The crucial point is associativity; using distributivity,
it is enough to check the relation 𝑎(𝑏𝑐) = (𝑎𝑏)𝑐 when 𝑎, 𝑏, 𝑐 belong
to {1, 𝑖, 𝑗, 𝑘}. This is obvious if 𝑎 = 1 (1(𝑏𝑐) = 𝑏𝑐 = (1𝑏)𝑐), if 𝑏 = 1
(𝑎(1𝑐) = 𝑎𝑐 = (𝑎1)𝑐), or if 𝑐 = 1 (𝑎(𝑏1) = 𝑎𝑏 = (𝑎𝑏)1). We may thus
assume that 𝑎, 𝑏, 𝑐 belong to {𝑖, 𝑗, 𝑘} but leave the reader to check these
twenty-seven remaining cases!
Let 𝑞 = 𝑎1+𝑏𝑖+𝑐 𝑗+𝑑𝑘 be any quaternion; set 𝑞 = 𝑎1−𝑏𝑖−𝑐 𝑗−𝑑𝑘. For any
two quaternions 𝑞 and 𝑞 0, one has 𝑞𝑞 0 = 𝑞 0 𝑞 and 𝑞𝑞 = (𝑎 2 + 𝑏 2 + 𝑐 2 + 𝑑 2 )1;
in particular, 𝑞𝑞 is a positive real number (multiplied by 1), and vanishes
only for 𝑞 = 0.
Any nonzero quaternion 𝑞 is invertible, with inverse the quaternion
(𝑞 𝑞¯ )−1 𝑞¯ .
Let 𝑞 = 𝑎1 + 𝑏𝑖 + 𝑐 𝑗 + 𝑑𝑘 be a quaternion. Observe that 𝑖 −1 𝑞𝑖 = −𝑖𝑞𝑖 =
𝑎1 + 𝑏𝑖 − 𝑐 𝑗 − 𝑑𝑘, 𝑗 −1 𝑞 𝑗 = 𝑎1 − 𝑏𝑖 + 𝑐 𝑗 − 𝑑𝑘 and 𝑘 −1 𝑞𝑘 = 𝑎1 − 𝑏𝑖 − 𝑐 𝑗 + 𝑑𝑘.
It follows that the center of H consists of the elements 𝑎1, for 𝑎 ∈ R; it is
a subfield of H, isomorphic to the field of real numbers.
Observe that the set of all quaternions of the form 𝑎 + 𝑏𝑖, for 𝑎, 𝑏 ∈ R,
is a subfield of H isomorphis to C. More generally, for any unit vector
(𝑢1 , 𝑢2 , 𝑢3 ) ∈ R3 , the quaternion 𝑢 = 𝑢1 𝑖 + 𝑢2 𝑗 + 𝑢3 𝑘 satisfies 𝑢 2 = −1;
moreover, the set of all quaternions of the form 𝑎 + 𝑏𝑢, for 𝑎 and 𝑏 ∈ R,
is a subfield of H which is also isomorphic to C.
In particular, we see that the equation X2 + 1 = 0 has infinitely many
solutions in H, in marked contrast to what happens in a field, where a
14 CHAPTER 1. RINGS

polynomial equation has at most as many solutions as its degree (see

§1.3.19).

Theorem (1.2.7) (Frobenius, 1878). — Let A be a finite dimensional R-vector

space, endowed with an R-bilinear multiplication law which endowes it with
the structure of a division algebra. Then, A is isomorphic to R, C, or H.

Observe that this statement does not hold if one removes the hypothesis
that A is finitely dimensional (as shown by the field R(X) of rational
functions with real coefficients), or the hypothesis that the multiplication
turns A into a division ring (consider the product ring R × R). It is
also false without the assumption that the multiplication is R-bilinear,
see Deschamps (2001). The proof below follows quite faithfully the paper
by Palais (1968). It makes use of basic results about polynomials in one
variable that will be reviewed in later sections of the book, hence may
be skipped for a first reading.
Proof. — Let 1A be the neutral element of A for the multiplication. Let
us identify R with the subring of A consisting of the elements 𝑥1A , for
𝑥 ∈ R.
We first make a few observations that will be useful in the proof.
First of all, any subring B of A which is an R-vector space is a division
algebra. Indeed, for any nonzero 𝑏 ∈ B, the map 𝑥 ↦→ 𝑏𝑥 from B to itself
is R-linear and injective; since B is a finite dimensional vector space, it
is also surjective so that the inverse of 𝑏 (the preimage of 1 under this
map) belongs to B.
Let also 𝛼 be any element in A R. The ring R[𝛼] generated by R
and 𝛼 is commutative. It is a vector-subspace of A, hence is a field. The
minimal polynomial P of 𝛼 in R[𝛼] is irreducible, hence has degree 6 2.
Since 𝛼 ∉ R, this degree is exactly 2 and there are real numbers 𝑢, 𝑣
such that P =√X2 + 2𝑢X + 𝑣. Then, (𝛼 + 𝑢)2 = 𝑢 2 − 𝑣 so that the element
𝑖 = (𝛼 + 𝑢)/ 𝑣 − 𝑢 2 of R[𝛼] satisfies 𝑖 2 = −1. Let us observe that
R[𝛼] = R[𝑖] is isomorphic to C. In particular, if 𝛼2 ∈ R, then 𝛼2 < 0.
Let us now assume that A ≠ R and let us choose 𝛼 ∈ A R. As we
just saw, there exists 𝑖 ∈ R[𝛼] such that 𝑖 2 = −1 and R[𝛼] = R[𝑖] ' C.
We may identify the subfield R[𝑖] with the field of complex numbers
 1.2. NILPOTENT ELEMENTS; REGULAR AND INVERTIBLE ELEMENTS; DIVISION RINGS 15

and view A as a C-vector space, complex scalars acting on A by left

multiplication.
Let us now assume that A is commutative; we shall show that in this
case, A = C. More generally, we shall show that there is no element
𝛽 ∈ A C which commutes with 𝑖. By contradiction, let 𝛽 ∈ A C such
that 𝛽𝑖 = 𝑖𝛽. The subring C[𝛽] is commutative, and is a field. Moreover,
since 𝛽 ∉ R, the p same argument as above furnishes an element 𝑗 of the
form (𝛽 + 𝑢 )/ 𝑣 0 − (𝑢 0)2 of A such that 𝑗 2 = −1 and R[𝑗] = R[𝛽]. Since
0

𝛽 ∉ C, we see that 𝑗 ≠ ±𝑖. It follows that the polynomial X2 + 1 has at

least four roots (namely, 𝑖, −𝑖, 𝑗 and −𝑗) in the field C[𝛽], which is absurd.
Let 𝜑 : A → A be the map given by 𝜑(𝑥) = 𝑥𝑖. It is C-linear and
satisfies 𝜑 2 = − idA . Since the polynomial X2 + 1 is split in C, with
simple roots, 𝑖 and −𝑖, the space A is the direct sum of the eigenspaces
for the eigenvalues 𝑖 and −𝑖. In other words, A is equal to the direct sum
A+ ⊕ A− , where A+ is the set of all 𝑥 ∈ A such that 𝜑(𝑥) = 𝑥𝑖 = 𝑖𝑥 and
A− is the space of 𝑥 ∈ A such that 𝜑(𝑥) = 𝑥𝑖 = −𝑖𝑥.
Let us remark that A+ is stable under multiplication; indeed, if 𝑥𝑖 = 𝑖𝑥
and 𝑦𝑖 = 𝑖𝑦, then (𝑥𝑦)𝑖 = 𝑥𝑖𝑦 = 𝑖𝑥𝑦. It also contains the inverse of any
of its nonzero elements since, if 𝑥 ∈ A+ is not equal to 0, then 𝑥𝑖 = 𝑖𝑥, so
that 𝑥 −1 𝑖 = 𝑖𝑥 −1 . Hence A+ is a subfield of A which contains the field C.
Since no element of A C commutes with 𝑖, we obtain A+ = C.
Assume finally that A ≠ A+ and let us consider any nonzero element
𝛽 ∈ A− . The map 𝑥 ↦→ 𝑥𝛽 is C-linear, and injective, hence bijective.
If 𝑥𝑖 = 𝑖𝑥, then 𝑥𝛽𝑖 = 𝑥(−𝑖𝛽) = −𝑥𝑖𝛽 = −𝑖𝑥𝛽, so that A+ 𝛽 ⊂ A− .
Conversely, if 𝑦𝑖 = −𝑖𝑦, let us choose 𝑥 ∈ A such that 𝑦 = 𝑥𝛽; then
𝑥𝑖𝛽 = −𝑥𝛽𝑖 = −𝑦𝑖 = 𝑖𝑦 = 𝑖𝑥𝛽, hence 𝑥𝑖 = 𝑖𝑥 since 𝛽 ≠ 0, hence the other
inclusion and A+ 𝛽 = A− .
The same argument shows that A− 𝛽 = A+ . In particular, 𝛽 2 ∈ C ∩ R[𝛽].
These two vector spaces, C and R[𝛽], are distinct, have dimension 2, and
contain R. Consequently, their intersection is equal to R. Since 𝛽 2 ∈ R
and 𝛽 ∉ R, we have 𝛽 2 < 0. Then 𝑗 = (−𝛽 2 )−1/2 𝛽 is an element of A− such
that 𝑗 2 = −1. Moreover, A− = A+ 𝛽 = A+ 𝑗 and A+ = A− 𝑗.
Set 𝑘 = 𝑖𝑗. One has A+ = R1 ⊕ R𝑖, A− = A+ 𝑗 = R𝑗 ⊕ R𝑘, and
A = A+ ⊕ A− . The R-vector space A has dimension 4, and (1, 𝑖, 𝑗, 𝑘) is
a basis. One has 𝑘 2 = 𝑖𝑗𝑖𝑗 = 𝑖(−𝑖𝑗)𝑗 = −𝑖 2 𝑗 2 = −1; more generally, the
16 CHAPTER 1. RINGS

multiplication table of A coincides with that of H. This shows that A

and H are isomorphic. 

1.3. Algebras, polynomials

Definition (1.3.1). — Let 𝑘 be a commutative ring. A 𝑘-algebra is a ring A
together with a morphism of rings 𝑖 : 𝑘 → A whose image is contained in the
center of A.

Formally, a 𝑘-algebra is defined as the ordered pair (A, 𝑖 : 𝑘 → A).

However, we will mostly say “Let A be a 𝑘-algebra”, therefore understat-
ing the morphism 𝑖. For 𝑥 ∈ 𝑘 and 𝑎 ∈ A, we will also commit the abuse
of writing 𝑥𝑎 for 𝑖(𝑥)𝑎, even when 𝑖 is not injective. Observe also that a
𝑘-algebra may be non-commutative.
A subalgebra of a 𝑘-algebra (A, 𝑖) is a subring B containing the image
of 𝑖, so that (B, 𝑖) is a 𝑘-algebra.

Definition (1.3.2). — Let 𝑘 be a commutative ring and let (A, 𝑖) and (B, 𝑗) be
𝑘-algebras. A morphism of 𝑘-algebras 𝑓 : A → B is a ring morphism 𝑓 such
that 𝑓 (𝑖(𝑥)𝑎) = 𝑗(𝑥) 𝑓 (𝑎) for every 𝑥 ∈ 𝑘 and every 𝑎 ∈ A.

A composition of morphisms of 𝑘-algebras is a morphism of 𝑘-algebras;

the inverse of a bijective morphism of 𝑘-algebras is a morphism of 𝑘-
algebras.

Remark (1.3.3). — There is a general notion of 𝑘-algebras based on the

language of modules, that we develop later in this book (from chapter 3
on). Namely, a (general) 𝑘-algebra is a 𝑘-module M endowed with a
𝑘-bilinear “multiplication” 𝜇M : M × M → M, a morphism of 𝑘-algebras
is a 𝑘-linear map 𝑓 : M → N which is compatible with the multiplication:
for 𝑎, 𝑏 ∈ M, 𝑓 (𝜇M (𝑎, 𝑏)) = 𝜇N ( 𝑓 (𝑎), 𝑓 (𝑏)).
Additional properties of the multiplication can of course be required,
giving rise to the notions of unitary 𝑘-algebras (if there is a unit for
multiplication, and morphisms are also required to map the unit to the
unit), commutative 𝑘-algebras (if the multiplication is commutative),
associative 𝑘-algebras (if it is associative), but also Lie 𝑘-algebras (if it
satisfies the Jacobi triple identity), etc.
 1.3. ALGEBRAS, POLYNOMIALS 17

If the multiplication is unitary and associative, then we get a 𝑘-algebra

in our sense; if it is moreover commutative, we get a commutative
𝑘-algebra. The terminology we chose reflects the fact that, at least in this
book, we are essentially interested in unitary and associative algebras.

Examples (1.3.4). — a) Let 𝑘 be a subring of a commutative ring A.

The inclusion 𝑘 ↩→ A endowes A with the structure of a 𝑘-algebra.
b) Any ring is, in a unique way, a Z-algebra. Indeed, for any ring A,
the map defined by 𝑖(𝑛) = 𝑛1A for 𝑛 ∈ Z is the unique morphism of
rings 𝑖 : Z → A.
c) Let K be a division ring and let 𝑖 : Z → K the unique morphism of
rings.
Assume that 𝑖 is injective. Then its image is a subring of K which is
isomorphic to Z. Moreover, the ring K being a division ring, it contains
the set K0 of all fractions 𝑎/𝑏, for 𝑎, 𝑏 ∈ 𝑖(Z) and 𝑏 ≠ 0. This set K0 is a
subfield of K, isomorphic to the field Q of rational numbers; one says
that K has characteristic 0.
Assume now that 𝑖 is not injective and let 𝑝 be the smallest integer
such that 𝑝 > 0 and 𝑖(𝑝) = 0. Let us prove that 𝑝 is a prime number.
Since 𝑖(1) = 1K ≠ 0, one has 𝑝 > 1. Let 𝑚 and 𝑛 be positive integers such
that 𝑝 = 𝑚𝑛; then 𝑖(𝑚)𝑖(𝑛) = 𝑖(𝑝) = 0, hence 𝑖(𝑚) = 0 or 𝑖(𝑛) = 0; by
minimality, one has 𝑚 > 𝑝 or 𝑛 > 𝑝; this proves that 𝑝 is a prime number,
as claimed. The image of 𝑖 is a then subfield K0 of K of cardinality 𝑝 and
one says that K has characteristic 𝑝.
The field K0 is sometimes called the prime field of K.
d) Let 𝑘 be a commutative ring. The ring 𝑘[X] of polynomials with
coefficients in 𝑘 in one indeterminate X is naturally a 𝑘-algebra. We will
define below algebras of polynomials in any number of indeterminates.
e) Let 𝑘 be a commutative ring. The ring M𝑛 (𝑘) of 𝑛 × 𝑛 matrices
with coefficients in 𝑘 is a 𝑘-algebra, where the canonical morphism
𝑖 : 𝑘 → M𝑛 (𝑘) associates with 𝑎 ∈ 𝑘 the matrix 𝑎I𝑛 .

1.3.5. Algebra generated by a set. — Let 𝑘 be a commutative ring, let

A be a 𝑘-algebra and let S be a subset of A. By definition, the 𝑘-algebra
𝑘[S] is the smallest subalgebra of A which contains S. It is the set of
18 CHAPTER 1. RINGS

all sums of elements of the form 𝜆𝑠1 . . . 𝑠 𝑛 , where 𝜆 ∈ 𝑘, 𝑛 ∈ N, and

𝑠1 , . . . , 𝑠 𝑛 ∈ S. If S = {𝑎1 , . . . , 𝑎 𝑛 }, one writes also 𝑘[𝑎1 , . . . , 𝑎 𝑛 ] for 𝑘[S].

Proposition (1.3.6). — Let 𝑘 be a commutative ring, let A be a commutative

𝑘-algebra and let B be an A-algebra. Assume that A is generated as a 𝑘-algebra
by elements (𝑎 𝑖 )𝑖∈I and that B is generated as an A-algebra by elements (𝑏 𝑗 ) 𝑗∈J .
Then B is generated as a 𝑘-algebra by the elements (𝑏 𝑗 ) 𝑗∈J and the elements
(𝑎 𝑖 )𝑖∈I .

Proof. — Let B0 be the 𝑘-subalgebra of B generated by the 𝑎 𝑖 and the 𝑏 𝑗 ,

and let A0 be its inverse image in A. It is a 𝑘-subalgebra of A that contains
the 𝑎 𝑖 , hence is equal to A, so that B0 contains the image of A in B. Since
it also contains the 𝑏 𝑗 , one has B0 = B, as was to be shown. 

Corollary (1.3.7). — Let 𝑘 be a commutative ring. If A is a finitely generated

commutative 𝑘-algebra and B is a finitely generated A-algebra, then B is finitely
generated as a 𝑘-algebra

1.3.8. — Let 𝑘 be a commutative ring, let A be a commutative 𝑘-algebra

which is a field, and let S be a subset of A. Then one also considers the
subfield of A generated by S over 𝑘, denoted by 𝑘(S): this is the smallest
subfield of A that contains 𝑘 and S. It is the set of all fractions 𝑎/𝑏, where
𝑎, 𝑏 ∈ 𝑘[S], the 𝑘-subalgebra of A generated by S, and 𝑏 ≠ 0.

Example (1.3.9) (General polynomial rings). — Let A be a ring and let I

be any set. The set N(I) is the set of all multi-indices indexed by I: its
elements are families (𝑛 𝑖 )𝑖∈I consisting of positive integers, almost all of
which are zero. When I is the finite set {1, . . . , 𝑑}, then N(I) is naturally
identified with the set N𝑑 of 𝑑-tuples of positive integers. The termwise
addition of N(I) endowes it with the structure of a monoid.
The ring 𝒫I is defined as the monoid ring associated with this monoid,
and coefficients in A. Let us review its definition in this particular case.
By definition, 𝒫I is the set A(N ) of families (𝑎 𝑚 )𝑚∈NI of elements of A,
I

indexed by N(I) , with finite support, that is of which all but finitely many
terms are zero (we also that such a family is almost null). Endowed with
term-by-term addition, it is an abelian group. Let P = (𝑝 𝑚 ) and Q = (𝑞 𝑚 )
 1.3. ALGEBRAS, POLYNOMIALS 19

be elements of 𝒫I . For any 𝑚 ∈ N(I) , one may set

Õ
𝑟𝑚 = 𝑝 𝑚0 𝑞 𝑚00 ,
𝑚 0+𝑚 00=𝑚

because the sum is finite; the family (𝑟𝑚 ) is an almost null family of
elements of A indexed by N(I) , hence defines an element R of 𝒫I .
One checks that this law (P, Q) ↦→ R is associative: for P = (𝑝 𝑚 ),
Q = (𝑞 𝑚 ) and R = (𝑟𝑚 ), one has (PQ)R = P(QR) = S, where S = (𝑠 𝑚 ) is
given by
Õ
𝑠𝑚 = 𝑝 𝑚0 𝑞 𝑚00 𝑟𝑚000 .
𝑚 0+𝑚 00+𝑚 000=𝑚

The unit element of 𝒫I is the family (𝜀𝑚 ) such that 𝜀0 = 1 (for the
multi-index 0 = (0, . . . )) and 𝜀𝑚 = 0 for any 𝑚 ∈ N(I) such that 𝑚 ≠ 0.
The map A → A[(X𝑖 )] given by 𝑎 ↦→ 𝑎1 is a morphism of rings.
For any 𝑚 ∈ N(I) , write X𝑚 for the element of 𝒫I whose only nonzero
term is at 𝑚, and equals 1. One has X𝑚 X𝑚 = X𝑚+𝑚 . In particular, the
0 0

elements of the form X𝑚 pairwise commute.

For 𝑖 ∈ I, let 𝛿 𝑖 ∈ N(I) be the multi-index that equals 1 at 𝑖 and 0
otherwise; one writes X𝑖 = X𝛿 𝑖 . For any multi-index 𝑚, one has 𝑚 =
𝑖∈I 𝑚 𝑖 𝛿 𝑖 (a finite sum, since all but finitely many coefficients 𝑚 𝑖 are
Í
zero), so that X = 𝑖∈I X𝑚
𝑚 𝑖
Î
𝑖 . Consequently, for any P = (𝑝 𝑚 ) ∈ 𝒫I , one
has
X𝑚
Õ Õ Ö
𝑚
P= 𝑝𝑚 X = 𝑝𝑚 𝑖 .
𝑖

𝑚∈N(I) 𝑚∈N(I) 𝑖∈I

The ring 𝒫I is called the ring of polynomials with coefficients in A in the

family of indeterminates (X𝑖 )𝑖∈I ; it is denoted by A[(X𝑖 )𝑖∈I ].
The map 𝑎 ↦→ 𝑎X0 is a ring morphism from A to A[(X𝑖 )𝑖∈I ]; the elements
of its image are called constant polynomials.
The choice of the letter X for the denoting the indeterminates is totally
arbitrary; 𝑥 𝑖 , Y𝑖 , T𝑖 , are other common notation. When I = {1, . . . , 𝑛},
one rather writes A[X1 , . . . , X𝑛 ], or A[𝑥 1 , . . . , 𝑥 𝑛 ], or A[Y1 , . . . , Y𝑛 ], etc.
when I has only one element, the indeterminate is often denoted by 𝑥,
T, X, etc. so that this ring is denoted by A[𝑥], A[T], A[X], accordingly.
When I has few elements, the indeterminates are also denoted by distinct
letters, as in A[X, T], or A[X, Y, Z], etc.
20 CHAPTER 1. RINGS

When A is a commutative ring, the ring of polynomials is a commuta-

tive ring, hence an A-algebra via the morphism A → A[(X𝑖 )] given by
𝑎 ↦→ 𝑎1.

Remark (1.3.10). — Let A be a ring, let I be a set, let J be a subset of I and

let K = I J. Extending families (𝑎 𝑗 ) 𝑗∈J by zero, let us identify N(J) with
a subset of N(I) , and similarly for N(K) . Then every element 𝑚 ∈ N(I)
can be uniquely written as a sum 𝑚 = 𝑚 0 + 𝑚 00, where 𝑚 0 ∈ N(J) and
𝑚 00 ∈ N(K) ; in fact, 𝑚 0𝑗 = 𝑚 𝑗 for 𝑗 ∈ J, and 𝑚 00𝑘 = 𝑚 𝑘 for 𝑘 ∈ K.
Then we can write any polynomial P = 𝑚∈N(I) 𝑝 𝑚 X𝑚 in 𝒫I under the
Í
form !
Õ Õ Õ
𝑚 𝑛
P= 𝑝 𝑚+𝑛 X X = P𝑛 X𝑛 ,
𝑛∈N(K) 𝑚∈N(J) 𝑛∈N(K)

where P𝑛 = 𝑚∈N(J) 𝑝 𝑚+𝑛 X𝑚 ∈ 𝒫J . This furnishes a map P ↦→ (P𝑛 )𝑛 from

Í
𝒫I to the ring (A[(X 𝑗 ) 𝑗∈J )[(X 𝑘 ) 𝑘∈K ]. One checks readily that this map is
bijective and is a ring morphism; it is thus an isomorphism of rings.

1.3.11. Monomials and degrees. — For any 𝑚 ∈ N(I) , the element

X𝑚 = 𝑖∈I X𝑚 𝑖 is called the monomial of exponent 𝑚. One
𝑖
Î
defines its
degree with respect to X𝑖 to be 𝑚 𝑖 , and its total degree as 𝑖∈I 𝑚 𝑖 .
Í
Let P be a polynomial in A[(X𝑖 )]. Write P = 𝑎 𝑚 X𝑚 , the monomials
Í
of P are the polynomials 𝑎 𝑚 X𝑚 for 𝑎 𝑚 ≠ 0. For any 𝑖 ∈ I, the degree of P
with respect to X𝑖 , denoted by degX𝑖 (P), is the least upper bound of all
degrees with respect to X𝑖 of all monomials of P. Similarly, the total
degree deg(P) of P is the least upper bound of the total degrees of all
monomials of P. These least upper bounds are computed in N ∪ {−∞},
so that the degrees of the zero polynomial are equal to −∞; otherwise,
they are positive.
The constant term of a polynomial P is the coefficient of the monomial
of exponent 0 ∈ N(I) . A polynomial P is constant if it is reduced to this
constant term, equivalently if its total degree is 0, equivalently if its
degree with respect to every indeterminate is 0.
When there is only one indeterminate T, the degree degT and the total
degree deg of a polynomial P ∈ A[T] coincide. Moreover, if deg(P) = 𝑛,
then T𝑛 is the unique monomial of degree 𝑛, and its coefficient is called
 1.3. ALGEBRAS, POLYNOMIALS 21

the leading coefficient of P. The polynomial P is said to be monic if its

leading coefficient is equal to 1.
Let 𝑑 ∈ N. One says that a polynomial P ∈ 𝒫I is homogeneous of
degree 𝑑 if all of its monomials have total degree 𝑑. One can write
uniquely any polynomial P ∈ 𝒫I has a sum P = 𝑑𝑘=0 P𝑑 , where P𝑑 ∈ 𝒫I
Í
is homogeneous of degree 𝑑. In fact, if P = 𝑚∈N(I) 𝑝 𝑚 X𝑚 , then P𝑑 =
Í
𝑚
deg(X𝑚 )=𝑑 𝑝 𝑚 X .
Í
Let 𝑖 ∈ I. For any polynomials P and Q, one has
degX𝑖 (P + Q) 6 sup(degX𝑖 (P), degX𝑖 (Q)),
with equality if degX𝑖 (P) ≠ degX𝑖 (Q). Moreover,
deg(P + Q) 6 sup(deg(P), deg(Q)).
Regarding multiplication, we have
degX𝑖 (PQ) 6 degX𝑖 (P) + degX𝑖 (Q),
deg(PQ) 6 deg(P) + deg(Q),
and we shall see below that these inequalities are equalities if A is a
domain.

Proposition (1.3.12). — Let P and Q ∈ A[T] be nonzero polynomials in one

indeterminate T. Assume that the leading coefficient of P is left regular, or that
the leading coefficient of Q is right regular. Then, deg(PQ) = deg(P)+deg(Q).
In particular, PQ ≠ 0.

Proof. — Write P = 𝑝0 + 𝑝 1 T + · · · + 𝑝 𝑚 T𝑚 and Q = 𝑞 0 + · · · + 𝑞 𝑛 T𝑛 , with

𝑚 = deg P and 𝑛 = deg Q, so that 𝑝 𝑚 ≠ 0 and 𝑞 𝑛 ≠ 0. Then,
PQ = 𝑝0 𝑞0 + (𝑝0 𝑞1 + 𝑝 1 𝑞 0 )T + · · · + (𝑝 𝑚−1 𝑞 𝑛 + 𝑝 𝑚 𝑞 𝑛−1 )T𝑚+𝑛−1 + 𝑝 𝑚 𝑞 𝑛 T𝑚+𝑛 .
By assumption, 𝑝 𝑚 𝑞 𝑛 ≠ 0 and the degree of PQ is equal to 𝑚 + 𝑛, as
claimed. 

Corollary (1.3.13). — Let A be a domain, and let I be a set. Then, for any
polynomials P, Q ∈ A[(X𝑖 )𝑖∈I ], and any 𝑖 ∈ I, one has
degX𝑖 (PQ) = degX𝑖 (P) + degX𝑖 (Q) and deg(PQ) = deg(P) + deg(Q).
In particular, the polynomial ring A[(X𝑖 )] is a domain.
22 CHAPTER 1. RINGS

Proof. — We first prove that the product of two nonzero polynomials is

not equal to 0, and that its degree with respect to the indeterminate X𝑖 is
the sum of their degrees.
Since P and Q have only finitely many monomials, we may assume
that there are only finitely many indeterminates. We can then argue
by induction on the number of indeterminates that appear in P and Q.
If this number is 0, then there P = 𝑝 and Q = 𝑞, for 𝑝, 𝑞 ∈ A, and
degX𝑖 (P) = degX𝑖 (𝑞) = 0. Since A is a domain, one has 𝑝𝑞 ≠ 0, hence
PQ = 𝑝𝑞 ≠ 0, and degX𝑖 (PQ) = 0.
Let 𝑚 = degX𝑖 (P) and 𝑛 = degX𝑖 (Q). First assume that the indetermi-
nate X𝑖 appears in P or Q, so that 𝑚 > 0 or 𝑛 > 0. Let J = I {𝑖}. There
are polynomials (P𝑚 ) and (Q𝑛 ) in the ring 𝒫J of polynomials with coeffi-
cients in A in the indeterminates X 𝑗 , for 𝑗 ∈ J, such that P = 𝑚 P 𝑘 X𝑖𝑘
Í
𝑘=0
and Q = 𝑛𝑘=0 Q 𝑘 X𝑖𝑘 . Then one can write PQ = 𝑚+𝑛 𝑘
Í Í
𝑘=0 R 𝑘 X 𝑖 for some poly-
nomials R 𝑘 ∈ 𝒫J , and one has R𝑚+𝑛 = P𝑚 Q𝑛 . Since X𝑖 appears in P or Q,
the number of indeterminates that appears in P𝑚 or Q𝑛 has decreased;
by induction, one has P𝑚 Q𝑛 ≠ 0. This proves that degX𝑖 (PQ) = 𝑚 + 𝑛; in
particular, PQ ≠ 0.
If the indeterminate X𝑖 appear neither in P, nor in Q, we can redo this
argument with some other indeterminate 𝑗 that appears in P or in Q; this
shows that PQ ≠ 0. Since the indeterminate X𝑖 does not appear in PQ,
this shows that degX𝑖 (PQ) = 0.
To prove the assertion about the total degrees, let 𝑚 = deg(P), 𝑛 =
deg(Q), and let us write P = 𝑚
Í Í𝑛
𝑖=0 P 𝑖 and Q = P 𝑗 as the sums of
Í𝑗=0
𝑚+𝑛
their homogeneous components. Then PQ = 𝑘=0 R 𝑘 , where R 𝑘 =
𝑖+𝑗=𝑘 P 𝑖 Q 𝑗 . Expanding the products P𝑖 Q 𝑗 , we observe that for each 𝑘,
Í
the polynomial R 𝑘 is a sum of monomials of degree 𝑘, hence R 𝑘 is
homogeneous of degree 𝑘. By assumption, P𝑚 ≠ 0 and Q𝑛 ≠ 0, so that
R𝑚+𝑛 = P𝑚 Q𝑛 ≠ 0. This proves that deg(PQ) = 𝑚 + 𝑛, as was to be
shown. 

Corollary (1.3.14). — Let A be a domain and let I be a set. The units of the
polynomial ring A[(X𝑖 )𝑖∈I ] are the constant polynomials equal to a unit of A.

(For the case of a general commutative ring A, see exercise 1/20.)

 1.3. ALGEBRAS, POLYNOMIALS 23

Proof. — Since the map 𝑎 ↦→ 𝑎1 is a ring morphism from A to A[(X𝑖 )], it

maps units to units, and the constant polynomials equal to a unit of A
are units in A[(X𝑖 )].
Conversely, let P be a unit in A[(X𝑖 )] and let Q ∈ A[(X𝑖 )] be such
that PQ = 1. Then deg(PQ) = deg(P) + deg(Q) = deg(1) = 0, hence
deg(P) = deg(Q) = 0. This proves that P and Q are constant polynomials.
Let 𝑎, 𝑏 ∈ A be such that P = 𝑎 and Q = 𝑏; then 𝑎𝑏 = 1, so that 𝑎, 𝑏
are units in A. In particular, P is constant polynomial equal to a unit
of A. 

Theorem (1.3.15). — Let A be a ring and let P, Q be two polynomials with

coefficients in A in one indeterminate X. One assumes that Q ≠ 0 and that
the coefficient of its monomial of highest degree is invertible. Then, there
exists a unique pair (R, S) of polynomials in A[X] such that P = RQ + S and
deg(S) < deg(Q).

Proof. — Let us begin with uniqueness. If P = RQ + S = R0Q + S0 for

polynomials R, S, R0 , S0 such that deg(S) < deg(Q) and deg(S0) < deg(Q),
then the degree of (R0 − R)Q = S − S0 is at most sup(deg(S), deg(S0)) <
deg Q. Assume R ≠ R0, that is, R0 − R ≠ 0, then deg((R0 − R)Q) =
deg(R0 − R) + deg(Q) > deg(Q), a contradiction. Consequently, R = R0
and S = P − RQ = P − R0Q = S0.
Let us now prove the existence of a pair (R, S) as in the statement of
the theorem. Let 𝑑 = deg(Q) and let 𝑢 be the leading coefficient of Q; by
assumption, it is a unit. We argue by induction on 𝑛 = deg(P). If 𝑛 < 𝑑,
it suffices to set R = 0 and S = P. Otherwise, let 𝑎X𝑛 be the monomial of
highest degree of P. Then, P0 = P − 𝑎𝑢 −1 X𝑛−𝑑 Q if a polynomial of degree
at most 𝑛. However, by construction, the coefficient of X𝑛 is equal to
𝑎 − 𝑎𝑢 −1 𝑢 = 0, so that deg(P0) < 𝑛 = deg P. By induction, there exist
polynomials R0 and S0 in A[X] such that
P0 = R0Q + S0 and deg(S0) < deg(Q).
Then,
P = P0 + 𝑎𝑢 −1 X𝑛−𝑑 Q = (R0 + 𝑎𝑢 −1 X𝑛−𝑑 )Q + S0.
It now suffices to set R = R0 + 𝑎𝑢 −1 X𝑛−𝑑 and S0 = S. This concludes the
proof. 
24 CHAPTER 1. RINGS

Polynomial algebras obey an important universal property.

Proposition (1.3.16). — Let 𝑘 be a commutative ring. Let A be any 𝑘-algebra

and let I be a set. For any family (𝑎 𝑖 )𝑖∈I of elements of A which commute
pairwise, there exists a unique morphism 𝑓 : 𝑘[(X𝑖 )𝑖∈I ] → A of 𝑘-algebras such
that 𝑓 (X𝑖 ) = 𝑎 𝑖 for every 𝑖 ∈ I.

Proof. — If there is such a morphism 𝑓 , it must satisfy

X𝑚 𝑎 𝑖𝑚𝑖
Ö Ö Ö
𝑚𝑖
𝑓 (𝜆 𝑖
𝑖 ) =𝜆 𝑓 (X𝑖 ) =𝜆
𝑖∈I 𝑖∈I 𝑖∈I

for any multi-index (𝑚 𝑖 ) ∈ N(I) and any 𝜆 ∈ 𝑘. (In the previous formulae,
all products are essentially finite, and the order of factors is not relevant,
because the 𝑎 𝑖 commute.) Consequently, for any polynomial P = 𝑝 𝑚 X𝑚 ,
Í
one must have
𝑎 𝑖𝑚𝑖 ,
Õ Ö
𝑓 (P) = 𝑝𝑚
𝑚∈N(I) 𝑖∈I
which proves first that there exists at most one such morphism 𝑓 , and
second, that if it exists, it has to be defined by this formula.
Conversely, it is straightforward to prove, using that the 𝑎 𝑖 commute
pairwise, that this formula defines indeed a morphism of 𝑘-algebras. 
Especially when A = 𝑘, this morphism is sometimes called the
evaluation morphism at the point (𝑎 𝑖 )𝑖∈I , and the image of the poly-
nomial P is denoted by P((𝑎 𝑖 )). When I = {1, . . . , 𝑛}, one writes sim-
ply P(𝑎 1 , . . . , 𝑎 𝑛 ). This gives, for example, a morphism of 𝑘-algebras
𝑘[X1 , . . . , X𝑛 ] → ℱ(𝑘 𝑛 , 𝑘) from the algebra of polynomials in 𝑛 indeter-
minates to the 𝑘-algebra of functions from 𝑘 𝑛 to 𝑘. The functions which
belong to the image of this morphism are naturally called polynomial
functions.
Here is an important example. Let 𝑘 be a field, let V be a 𝑘-vector
space and let A be the 𝑘-algebra of endomorphisms of V. One can also
take A to be the 𝑘-algebra M𝑛 (𝑘) of all 𝑛 × 𝑛 matrices with coefficients
in 𝑘. Then, for any element 𝑎 ∈ A and any polynomial P ∈ 𝑘[X], one may
compute P(𝑎). For P, Q ∈ 𝑘[X], one has P(𝑎) + Q(𝑎) = (P + Q)(𝑎), and
P(𝑎)Q(𝑎) = (PQ)(𝑎). These formulae just express that the map from 𝑘[X]
to A given by P ↦→ P(𝑎) is a morphism of 𝑘-algebras.
 1.3. ALGEBRAS, POLYNOMIALS 25

Lemma (1.3.17). — Let 𝑘 be a commutative ring, let A be a 𝑘-algebra and

let 𝑎1 , . . . , 𝑎 𝑛 be elements of A that commute pairwise. Then, the subalgebra
of A generated by 𝑎1 , . . . , 𝑎 𝑛 , 𝑘[𝑎1 , . . . , 𝑎 𝑛 ], is the image of the morphism from
𝑘[X1 , . . . , X𝑛 ] to A given by evaluation at at (𝑎1 , . . . , 𝑎 𝑛 ).
Proof. — Indeed, let 𝜑 be this evaluation morphism. Since 𝜑(X𝑖 ) = 𝑎 𝑖 ,
the image Im(𝜑) of 𝜑 is a subalgebra of A which contains 𝑎1 , . . . , 𝑎 𝑛 .
Therefore, Im(𝜑) contains 𝑘[𝑎1 , . . . , 𝑎 𝑛 ]. Conversely, any subalgebra of A
which contains 𝑎1 , . . . , 𝑎 𝑛 contains also all elements of A of the form
𝜆𝑎1𝑚1 . . . 𝑎 𝑛𝑚𝑛 , as well as their sums. This shows that 𝑘[𝑎1 , . . . , 𝑎 𝑛 ] contains
Im(𝜑), hence the equality. 

1.3.18. Formal derivatives. — Let A be a ring and let P ∈ A[X] be

a polynomial; writing P = 𝑛𝑚=0 𝑎 𝑚 X𝑚 , its (formal) derivative is the
Í
polynomial P0 = 𝑛𝑚=1 𝑚𝑎 𝑚 X𝑚−1 .
Í
We observe the following rules: (𝑎P)0 = 𝑎P0, (P + Q)0 = P0 + Q0 and
(PQ)0 = P0Q + PQ0 (Leibniz rule). Leaving to the reader the proofs of
the first two, let us prove the third one. By additivity, we may assume
that P = X𝑚 and Q = X𝑛 ; then PQ = X𝑚+𝑛 , P0 = 𝑚X𝑚−1 and Q0 = 𝑛X𝑛−1 ,
hence
(PQ)0 = (𝑚 + 𝑛)X𝑚+𝑛−1 = 𝑚X𝑚+𝑛−1 + 𝑛X𝑚+𝑛−1 = P0Q + PQ0 ,
as claimed.

1.3.19. Roots and their multiplicities. — Let K be a field and let P ∈

K[X] be a non-constant polynomial. A root of P in K is an element 𝑎 ∈ K
such that P(𝑎) = 0.
If 𝑎 is a root of P, then the euclidean division of P by X− 𝑎 takes the form
P = (X − 𝑎)Q + R, where R ∈ K[X] is a polynomial of degree < 1, hence a
constant, and by evaluation, one has 0 = P(𝑎) = (𝑎 − 𝑎)Q(𝑎) + R(𝑎) = R,
hence P = (X − 𝑎)Q and X − 𝑎 divides P. Conversely, if X − 𝑎 divides Q,
then P(𝑎) = 0.
By successive euclidean divisions, one may write P = Q 𝑛𝑖=1 (X − 𝑎 𝑖 )𝑚𝑖
Î
where Q ∈ K[X] is a polynomial without root in K, 𝑎1 , . . . , 𝑎 𝑛 are distinct
elements of K and 𝑚1 , . . . , 𝑚𝑛 are strictly positive integers. (The process
must stop since the degree of Q decreases by 1 at each step, and is
positive.)
26 CHAPTER 1. RINGS

Let us prove that 𝑚 𝑖 is the largest integer 𝑚 such that (X − 𝑎 𝑖 )𝑚 divides P.

This integer is called the multiplicity of the root 𝑎 𝑖 . By definition,
(X − 𝑎 𝑖 )𝑚𝑖 divides P. Conversely, assume by contradiction that there
exists 𝑚 > 𝑚 𝑖 such that (X − 𝑎 𝑖 )𝑚 divides P; let us write P = Q1 (X − 𝑎 𝑖 )𝑚 .
From the equality P = Q1 (X − 𝑎 𝑖 )𝑚 = Q 𝑛𝑗=1 (X − 𝑎 𝑗 )𝑚 𝑗 , we then get
Î
Q1 (X − 𝑎 𝑖 )𝑚−𝑚𝑖 = Q 𝑗≠𝑖 (X − 𝑎 𝑗 )𝑚 𝑗 ; evaluating both sides of this equality
Î
at X = 𝑎 𝑖 , we obtain 0 = Q(𝑎 𝑖 ) 𝑗≠𝑖 (𝑎 𝑖 − 𝑎 𝑗 )𝑚 𝑗 , a contradiction.
Î
Let 𝑎 ∈ K be such that P(𝑎) = 0. Then there exists 𝑖 ∈ {1, . . . , 𝑛} such
that 𝑎 = 𝑎 𝑖 : the elements 𝑎1 , . . . , 𝑎 𝑛 are the roots of P in K.
One has deg(P) = deg(Q) + 𝑛𝑖=1 𝑚 𝑖 , hence 𝑛𝑖=1 𝑚 𝑖 6 deg(P). The
Í Í
number of roots of a polynomial, counted with their multiplicities, is at most its
degree.
Let us give an application of this fact.

Proposition (1.3.20). — If K is a field, then every finite subgroup of the

multiplicative group of K is cyclic.

Proof. — Let G be a finite subgroup of K× and let 𝑛 be its cardinality.

For every prime number 𝑝 that divides 𝑛, there exists 𝑥 𝑝 ∈ G such that
(𝑎 𝑝 )𝑛/𝑝 ≠ 1, because the polynomial T𝑛/𝑝 − 1 has at most 𝑛/𝑝 roots and
𝑛 > 𝑛/𝑝. Let 𝑚 𝑝 be the exponent of 𝑝 in the decomposition of 𝑛 as
𝑛/𝑝 𝑚 𝑝 𝑚𝑝
a product of prime factors; set 𝑎 𝑝 = 𝑥 𝑝 . One has (𝑎 𝑝 )𝑝 = 1 and
𝑚 𝑝 −1 𝑛/𝑝
(𝑎 𝑝 )𝑝 = 𝑥 𝑝 ≠ 1, so that the order of 𝑎 𝑝 in G is equal to 𝑝 𝑚𝑝 .
Set 𝑎 = 𝑝 𝑎 𝑝 and let 𝑚 be its order; let us prove that the 𝑚 = 𝑛.
Î
Otherwise, there exists a prime number ℓ that divides 𝑛/𝑚. In particular,
ℓ divides 𝑛, 𝑚 divides 𝑛/ℓ and 𝑎 𝑛/ℓ = 1. If 𝑝 is a prime number dividing 𝑛
𝑛/ℓ
which is distinct from ℓ , then the integer 𝑝 𝑚𝑝 divides 𝑛/ℓ , hence 𝑎 𝑝 = 1.
𝑛/ℓ 𝑛/ℓ
Consequently, 𝑎 𝑛/ℓ = 𝑎ℓ ≠ 1. On the other hand, 𝑎ℓ ≠ 1 because the
order of 𝑎ℓ , equal to ℓ 𝑚ℓ , does not divide 𝑛/ℓ . We thus have 𝑎 𝑛/ℓ ≠ 1, a
contradiction.
The subgroup G1 of G generated by 𝑎 is cyclic, of order 𝑛, and its
elements are the 𝑛th roots of the polynomial T𝑛 − 1 in K. If 𝑔 ∈ G, then
𝑔 𝑛 = 1, by definition of 𝑛, hence 𝑔 ∈ G1 , so that G = G1 . 
The following lemma characterizes the multiplicity of a root using
derivatives.
 1.4. IDEALS 27

Lemma (1.3.21). — Let P ∈ K[X] be a non-constant polynomial and let 𝑎 ∈ K

be a root of P; let 𝑚 be its multiplicity.
a) For every integer 𝑘 ∈ {0, . . . , 𝑚 − 1}, 𝑎 is a root of P(𝑘) of multiplicity at
least 𝑚 − 𝑘.
b) If the characteristic of K is zero, or if it is strictly larger than 𝑚, then 𝑎 is
not a root of P(𝑚) .

Proof. — Assume that 𝑎 is a root of P with multiplicity at least 𝑚 and let

us write P = (X − 𝑎)𝑚 Q, where Q ∈ K[X] is a polynomial. Taking formal
derivatives, we obtain
P0 = 𝑚(X − 𝑎)𝑚−1 Q + (X − 𝑎)𝑚 Q0 = (X − 𝑎)𝑚−1 (𝑚Q + (X − 𝑎)Q0).
In particular, 𝑎 is a root of P0 with multiplicity > 𝑚 − 1. By induction,
we conclude that 𝑎 is a root of P(𝑘) of multiplicity at least 𝑚 − 𝑘, for every
𝑘 ∈ {1, . . . , 𝑚 − 1}.
If the characteristic of K is zero, or if it is a prime number that does not
divide 𝑚, then the polynomial 𝑚Q + (X − 𝑎)Q0 takes the value 𝑚Q(𝑎)
at 𝑎. In this case, we see in particular that 𝑎 is a root of multiplicity 𝑚 − 1
of P0 if and only if 𝑎 is a root of multiplicity 𝑚 of P.
By induction, we also have P(𝑚) (𝑎) = 𝑚!Q(𝑎).
Assume that 𝑚 is the multiplicity of 𝑎, so that Q(𝑎) ≠ 0. If the
characteristic of K is zero, or if it is strictly greater than 𝑚, then 𝑚! ≠ 0
in K, so that 𝑎 is not a root of P(𝑚) . 
Finally, recall that a field K is said to be algebraically closed if every non
constant polynomial P ∈ K[T] has a root in K. By what precedes, this
is equivalent to saying that every nonzero polynomial is a product of
factors of degree 1; such a polynomial is said to be split. We will return
to this topic in more detail in §4.3.

1.4. Ideals
Definition (1.4.1). — Let A be a ring and let I be a subgroup of A (for the
addition). One says that I is a left ideal if for any 𝑎 ∈ I and any 𝑏 ∈ A, one has
𝑏𝑎 ∈ I. One says that I is a right ideal if for any 𝑎 ∈ I and any 𝑏 ∈ A, one has
𝑎𝑏 ∈ I. One says that I is a two-sided ideal if it is both a left and right ideal.
28 CHAPTER 1. RINGS

In a commutative ring, it is equivalent for a subgroup to be a left ideal,

a right ideal or a two-sided ideal; one then just says that it is an ideal. Let
us observe that in any ring A, the subsets 0 and A are two-sided ideals.
Moreover, an ideal I is equal to A, if and only if it contains some unit, if
and only if it contains 1
For any 𝑎 ∈ A, the set A𝑎 consisting of all elements of A of the form 𝑥𝑎,
for 𝑥 ∈ A, is a left ideal; the set 𝑎A consisting of all elements of the
form 𝑎𝑥, for 𝑥 ∈ A, is a right ideal. When A is commutative, this ideal is
denoted by (𝑎)
To show that a subset I of A is a left ideal, it suffices to prove the
following properties:
(i) 0 ∈ I;
(ii) for any 𝑎 ∈ I and any 𝑏 ∈ I, 𝑎 + 𝑏 ∈ I;
(iii) for any 𝑎 ∈ I and any 𝑏 ∈ A, 𝑏𝑎 ∈ I.
Indeed, since −1 ∈ A and (−1)𝑎 = −𝑎 for any 𝑎 ∈ A, these properties
imply that I is a subgroup of A; the third one then shows that it is a left
ideal.
The similar characterization of right ideals is left to the reader.

Example (1.4.2). — If K is a division ring, the only left ideals (or right ideals)
of K are (0) and K. Indeed, let I be a left ideal of K such that I ≠ 0; let 𝑎
be any nonzero element of I. Let 𝑏 be any element of K. Since 𝑎 ≠ 0, it
is invertible in K and, by definition of a left ideal, 𝑏 = (𝑏𝑎 −1 )𝑎 ∈ I. This
shows that I = K.

Example (1.4.3). — For any ideal I of Z, there exists a unique integer 𝑛 > 0
such that I = (𝑛). More precisely, if I ≠ (0), then 𝑛 is the smallest strictly
positive element of I.
If I = (0), then 𝑛 = 0 is the only integer such that I = (𝑛). Assume now
that I ≠ (0).
If I = (𝑛), with 𝑛 > 0, one observes that the strictly positive elements
of I are {𝑛; 2𝑛; 3𝑛; . . .}, and 𝑛 is the smallest of them. This implies that
the uniqueness of such an integer 𝑛. So let 𝑛 be the smallest strictly
positive element of I. Since 𝑛 ∈ I, 𝑎𝑛 ∈ I for any 𝑎 ∈ Z, so that (𝑛) ⊂ I.
Conversely, let 𝑎 be any element of I. Let 𝑎 = 𝑞𝑛 + 𝑟 be the euclidean
division of 𝑎 by 𝑛, with 𝑞 ∈ Z and 0 6 𝑟 6 𝑛 − 1. Since 𝑎 and 𝑞𝑛 ∈ I,
 1.4. IDEALS 29

𝑟 = 𝑎 − 𝑞𝑛 belongs to I. Since 𝑛 is the smallest strictly positive element

of I and 0 6 𝑟 < 𝑛, we necessarily have 𝑟 = 0 and 𝑎 = 𝑞𝑛 ∈ (𝑛). This
shows that I = (𝑛).

Example (1.4.4). — For any nonzero ideal I of K[X], there exists a unique
monic polynomial P ∈ K[X] such that I = (P); moreover, P is the unique monic
polynomial of minimal degree which belongs to I.
The proof is analogous to that of the previous example. If I = (P), then
the degree of any nonzero polynomial in I is at least deg(P), and P is
the unique monic polynomial in I of minimal degree. This shows the
uniqueness part of the assertion.
Let P be a monic polynomial in I, of minimal degree. Any multiple
of P belongs to I, so (P) ⊂ I. Conversely, let A ∈ I, and let A = PQ + R be
the euclidean division of A by P, with deg(R) < deg(P). Since A and P
belong to I, R = A − PQ ∈ I. If R ≠ 0, the quotient of R by its leading
coefficient is a monic polynomial of degree < deg(P), which contradicts
the definition of P. Consequently, R = 0, A ∈ (P). Hence I = (P).

There are various useful operations on ideals.

1.4.5. Intersection. — Let I and J be left ideals of A; their intersection

I ∩ J is again a left ideal. More generally, the intersection of any family of
left ideals of A is a left ideal of A.
Ñ
Proof. — Let (I𝑠 )𝑠∈S be a family of ideals of A and let I = 𝑠 I𝑠 . (If
S = ∅, then I = A.) The intersection of any family of subgroups being a
subgroup, I is a subgroup of A. Let now 𝑥 ∈ I and 𝑎 ∈ A and let us show
that 𝑎𝑥 ∈ I. For any 𝑠 ∈ S, 𝑥 ∈ I𝑠 , hence 𝑎𝑥 ∈ I𝑠 since I𝑠 is a left ideal. It
follows that 𝑎𝑥 belongs to every ideal I𝑠 , hence 𝑎𝑥 ∈ I. 
We let to the reader to state and prove the analogous statements for
right and two-sided ideals.

1.4.6. Ideal generated by a subset. — Let S be any subset of A; there

exists a smallest left ideal of A containing S and it is called the left ideal
generated by S. Indeed, this ideal is nothing but the intersection of the
family of all left ideals containing S. Equivalently, the left ideal generated
by S contains S, and is contained in any left ideal which contains S.
30 CHAPTER 1. RINGS

Moreover, it is equal to the set of all linear combinations 𝑛𝑖=1 𝑎 𝑖 𝑠 𝑖 , for

Í
𝑛 ∈ N, 𝑎1 , . . . , 𝑎 𝑛 ∈ A and 𝑠1 , . . . , 𝑠 𝑛 ∈ S. Indexing these elements by the
elements of S rather than a set of integers, let us prove that this ideal is
the set of all linear combinations 𝑠∈S 𝑎 𝑠 𝑠, where (𝑎 𝑠 ) ∈ A(S) ranges over
Í
all almost null families of elements of A.
Proof. — By the preceding proposition, the intersection of the family
of all left ideals containing S is a left ideal of A, hence is the smallest
left ideal of A that contains S. Let I be the set of all linear combinations
𝑠∈S 𝑎 𝑠 𝑠, for (𝑎 𝑠 ) ∈ A . Moreover,
(S)
Í
we see by induction on the number
of nonzero terms in this sum that 𝑠∈S 𝑎 𝑠 𝑠 belongs to any left ideal of A
Í
which contains S, so belongs to the left ideal generated by S. This proves
that I is contained in the left ideal generated by S. To conclude, it suffices
to show that I is a left ideal containing S. Taking a family (𝑎 𝑠 ) where
all members are zero except for one of them equal to 1, we have S ⊂ I.
Í
We then check the axioms for a left ideal: we have 0 = 𝑠∈S 0𝑠 ∈ I; if
𝑎 𝑠 𝑠 and 𝑏 𝑠 𝑠 belong to I, then almost all of the terms of the family
Í Í
(𝑎 𝑠 + 𝑏 𝑠 )𝑠∈S are equal to 0 and (𝑎 𝑠 + 𝑏 𝑠 )𝑠 = 𝑎 𝑠 𝑠 + 𝑏 𝑠 𝑠; moreover, if
Í Í Í
𝑎 = 𝑎 𝑠 𝑠 ∈ I and 𝑏 ∈ A, then 𝑏𝑎 = (𝑏𝑎 𝑠 )𝑠 belongs to I.
Í Í

Similar arguments show that there exists a smallest right ideal (resp. a
smallest two-sided ideal) of A containing S; it is the intersection of
the family of all right ideals (resp. of all two-sided ideals) of A which
contain S. They are respectively equal to the set of all linear combinations
𝑠 𝑎 𝑠∈S 𝑎 𝑠 𝑠𝑏 𝑠 , where (𝑎 𝑠 ) and (𝑏 𝑠 ) run along A .
(S)
Í Í
𝑠∈S 𝑠 and
Definition (1.4.7). — The kernel Ker( 𝑓 ) of a ring morphism 𝑓 : A → B is the
set of all elements 𝑎 ∈ A such that 𝑓 (𝑎) = 0.
Since a ring morphism is in particular a morphism of the underlying
additive groups, a ring morphism 𝑓 is injective if and only if Ker( 𝑓 ) = 0.
Proposition (1.4.8). — The kernel of a ring morphism is a two-sided ideal.
Proof. — Let 𝑓 : A → B be a ring morphism. Since a morphism of rings
is a morphism of abelian groups, Ker( 𝑓 ) is a subgroup of A. Moreover,
if 𝑥 ∈ Ker( 𝑓 ) and 𝑎 ∈ A, then 𝑓 (𝑎𝑥) = 𝑓 (𝑎) 𝑓 (𝑥) = 𝑓 (𝑎)0 = 0 so that 𝑎𝑥 ∈
Ker( 𝑓 ). Similarly, if 𝑥 ∈ Ker( 𝑓 ) and 𝑎 ∈ A, then 𝑓 (𝑥𝑎) = 𝑓 (𝑥) 𝑓 (𝑎) = 0, so
that 𝑥𝑎 ∈ Ker( 𝑓 ). This shows that Ker( 𝑓 ) is a two-sided ideal of A. 
 1.4. IDEALS 31

1.4.9. Image and inverse image of ideals. — Let 𝑓 : A → B be a mor-

phism of rings and let J be a left ideal of B; the inverse image of J
by 𝑓 ,
𝑓 −1 (J) = {𝑎 ∈ A ; 𝑓 (𝑎) ∈ J}
is a left ideal of A.
Proof. — Since 𝑓 (0) = 0 ∈ J, we have 0 ∈ 𝑓 −1 (J). For any 𝑎 and 𝑏 ∈ 𝑓 −1 (J),
𝑓 (𝑎 + 𝑏) = 𝑓 (𝑎) + 𝑓 (𝑏) ∈ J since 𝑓 (𝑎) and 𝑓 (𝑏) ∈ J and J is an ideal
of B. Finally, for any 𝑎 ∈ A and 𝑏 ∈ 𝑓 −1 (J), we have 𝑓 (𝑏) ∈ J, hence
𝑓 (𝑎𝑏) = 𝑓 (𝑎) 𝑓 (𝑏) ∈ J. 
Similarly, the inverse image of a right ideal (resp. of a two-sided ideal)
by a morphism of rings is a right ideal (resp. a two-sided ideal).
However, the image of a left ideal by a ring morphism is not necessarily
a left ideal. Let 𝑓 : A → B be a morphism of rings and I be a left ideal
of A. We shall write B 𝑓 (I), or even BI, the left ideal of B generated by 𝑓 (I).

1.4.10. Sum of ideals. — Let I and J be left ideals of a ring A. The set
I + J of all sums 𝑎 + 𝑏, for 𝑎 ∈ I and 𝑏 ∈ J, is a left ideal of A. It is also
the left ideal of A generated by the subset I ∪ J. More generally, for any
family (I𝑠 )𝑠∈S of left ideals of A, the set of sums 𝑠 𝑎 𝑠 of all almost null
Í
families (𝑎 𝑠 )𝑠∈S , where, for any 𝑠, 𝑎 𝑠 ∈ I𝑠 , is a left ideal of A, denoted by
Í Ð
𝑠 I 𝑠 . It is also the left ideal of A generated by 𝑠 I𝑠 .
The similar assertions for right and two-sided ideals hold.
Í
Proof. — Let us prove the result for left ideals. Since 0 = 𝑠 0 and 0 ∈ I𝑠
for every 𝑠, one has 0 ∈ 𝑠 I𝑠 . Then, if 𝑎 = 𝑠 𝑎 𝑠 and 𝑏 = 𝑠 𝑏 𝑠 are any
Í Í Í
two elements of 𝑠 I𝑠 , then 𝑎 + 𝑏 = 𝑠 (𝑎 𝑠 + 𝑏 𝑠 ), where, for every 𝑠 ∈ S,
Í Í
𝑎 𝑠 + 𝑏 𝑠 ∈ I𝑠 , almost all terms of this sum being null; hence 𝑎 + 𝑏 ∈ 𝑠 I𝑠 .
Í
Finally, if 𝑎 = 𝑠 𝑎 𝑠 belongs to I𝑠 and 𝑏 ∈ A, then 𝑏𝑎 = 𝑠 (𝑏𝑎 𝑠 ). For
Í Í Í
every 𝑠, 𝑏𝑎 𝑠 ∈ I𝑠 , so that 𝑏𝑎 ∈ 𝑠 I𝑠 . We have shown that 𝑠 I𝑠 is a left
Í Í
ideal of A.
Í Ð
To prove that I𝑠 is the left ideal of A generated by the subset 𝑠 I𝑠 ,
we must establish two inclusions. Let I denote the latter ideal. First of
all, for any 𝑡 ∈ S and any 𝑎 ∈ I𝑡 , we have 𝑎 = 𝑠 𝑎 𝑠 , where 𝑎 𝑠 = 0 for 𝑠 ≠ 𝑡
Í
and 𝑎 𝑡 = 𝑎. This shows that 𝑎 ∈ 𝑠 I𝑠 and the ideal 𝑠 I𝑠 contains I𝑡 . By
Í Í
Í
definition of the left ideal I, we thus have I ⊂ 𝑠 I𝑠 . On the other hand,
let 𝑎 = 𝑠 𝑎 𝑠 be any element of 𝑠 I𝑠 , where 𝑎 𝑠 ∈ I𝑠 for all 𝑠. All terms of
Í Í
32 CHAPTER 1. RINGS

this sum belong to I, by definition of I. By definition of a left ideal, this

implies 𝑎 ∈ I, hence I ⊃ 𝑠 I𝑠 .
Í


1.4.11. Product of two-sided ideals. — Let A be a ring and let I, J be

two-sided ideals of A. The set of all products 𝑎𝑏, for 𝑎 ∈ I and 𝑏 ∈ J, is
not necessarily a two-sided ideal of A. We define IJ as the two-sided
ideal generated by these products.
Let K be the set of all linear combinations 𝑛𝑠=1 𝑎 𝑠 𝑏 𝑠 , with 𝑛 ∈ N,
Í
𝑎 𝑠 ∈ I and 𝑏 𝑠 ∈ J for every 𝑖. It is contained in IJ. Let us show that K
is a two-sided ideal of A. The relation 0 = 0 · 0 shows that 0 ∈ IJ. If
𝑐 = 𝑛𝑖=1 𝑎 𝑖 𝑏 𝑖 and 𝑐 0 = 𝑛𝑖=1 𝑎 0𝑖 𝑏 0𝑖 are elements of IJ, then 𝑐 + 𝑐 0 = 𝑛+𝑛
Í 0 Í 0
𝑖=1 𝑎 𝑖 𝑏 𝑖 ,
Í
where, for 𝑖 ∈ {𝑛 + 1, . . . , 𝑛 + 𝑛 }, we have set 𝑎 𝑖 = 𝑎 𝑖−𝑛 and 𝑏 𝑖 = 𝑏 0𝑖−𝑛 ;
0 0

hence 𝑐 + 𝑐 0 ∈ IJ. Let moreover 𝑐 = 𝑛𝑖=1 𝑎 𝑖 𝑏 𝑖 ∈ K and let 𝑎 ∈ A. One has

Í
Õ Õ
𝑎𝑐 = 𝑎 𝑎𝑖 𝑏𝑖 =


(𝑎𝑎 𝑖 )𝑏 𝑖 ;

because I is a left ideal, one has 𝑎𝑎 𝑖 ∈ I for every 𝑖, hence 𝑎𝑐 ∈ K. Similarly,

the equalities
Õ Õ
𝑐𝑎 = 𝑎𝑖 𝑏𝑖 𝑎 = 𝑎 𝑖 (𝑏 𝑖 𝑎)

show that 𝑐𝑎 ∈ K since, J being a right ideal, 𝑏 𝑖 𝑎 ∈ J for every 𝑖. Since K

contains all products 𝑎𝑏, with 𝑎 ∈ I and 𝑏 ∈ J, we have IJ ⊂ K and, finally,
IJ = K.
Since I and J are two-sided ideals, for any 𝑎 ∈ I and any 𝑏 ∈ J, the
product 𝑎𝑏 belongs to I and to J. It follows that IJ ⊂ I ∩ J.

1.4.12. Nilradical, radical of an ideal. — Let A be a commutative ring.

The nilradical of A is the set of its nilpotent elements. It is an ideal of A.
One says that the ring A is reduced if its nilradical is (0), that is, if 0 is its
only nilpotent element.
More generally, the radical of an ideal I of the commutative ring A is
defined by the formula
√
I = {𝑎 ∈ A ; there exists 𝑛 > 1 such that 𝑎 𝑛 ∈ I}.

It is an ideal
p of A which contains I. By definition, the nilradical of A is the
radical (0) of the null ideal (0).
 1.5. QUOTIENT RINGS 33
√ √ √
Proof. — Since 01 = 0 ∈ I, one has 0 ∈ I. For 𝑎 ∈ I and 𝑏 ∈ I, let
us choose integers 𝑛 and 𝑚 > 1 such that 𝑎 𝑛 ∈ I and 𝑏 𝑚 ∈ I. By the
binomial formula
𝑛+𝑚−1
𝑛 + 𝑚 − 1 𝑘 𝑛+𝑚−1−𝑘
Õ  
(𝑎 + 𝑏)𝑛+𝑚−1 = 𝑎 𝑏 .
𝑘
𝑘=0
In this sum, all terms belong to I: this holds for those corresponding
to 𝑘 > 𝑛, since, then, 𝑎 𝑘 = 𝑎 𝑛 𝑎 𝑛−𝑘 and 𝑎 𝑛 ∈ I; similarly, if 𝑘 < 𝑛, then
𝑛 + 𝑚 − 1 − 𝑘 > 𝑚, hence 𝑏 𝑛+𝑚−1−𝑘
√ = 𝑏 𝑚 𝑏 𝑛−1−𝑘 belongs
√ to I. Therefore,
(𝑎 + 𝑏) 𝑛+𝑚−1 ∈ I, and 𝑎 + 𝑏 ∈ I. Finally, for any 𝑎 ∈ I and 𝑏 ∈ A,√let us
choose 𝑛 > 1 such that 𝑎 𝑛 ∈ I. Then, (𝑏𝑎)𝑛 = 𝑏 𝑛 𝑎 𝑛 ∈ I, hence 𝑏𝑎 ∈ I. 

1.5. Quotient rings

Given a ring and an adequate equivalence relation on that ring, the
goal of this section is to endow the set of equivalence classes with the
structure of a ring. This will allow to formally make all elements of an
ideal equal to zero without modifying the other rules for computation.

1.5.1. Construction. — Let ℛ be binary relation on a set X. Recall that

one says that ℛ is an equivalence relation if it is reflexive (for any 𝑥, 𝑥 ℛ 𝑥),
symmmetric (if 𝑥 ℛ 𝑦, then 𝑦 ℛ 𝑥) and transitive (if 𝑥 ℛ 𝑦 and 𝑦 ℛ 𝑧,
then 𝑥 ℛ 𝑧). The set of all equivalence classes of X for the relation ℛ is
denoted by X/ℛ; the map clℛ : X → X/ℛ (such that, for every 𝑥 ∈ X,
clℛ (𝑥) is the equivalence class of 𝑥) is a surjection; its important property
is that any two elements have the same image if and only if they are in
relation by ℛ.
Let A be a ring. Let us search all equivalence relations which are
compatible with the ring structure, namely
if 𝑥 ℛ 𝑦 and 𝑥 0 ℛ 𝑦 0, then 𝑥 + 𝑥 0 ℛ 𝑦 + 𝑦 0 and 𝑥𝑥 0 ℛ 𝑦𝑦 0.
Let I be the equivalence class of 0. If 𝑥 ℛ 𝑦, since (−𝑦) ℛ (−𝑦), one gets
𝑥 − 𝑦 ℛ 0, hence 𝑥 − 𝑦 ∈ I, and conversely. Therefore, the relation ℛ can
be recovered from I by the property: 𝑥 ℛ 𝑦 if and only if 𝑥 − 𝑦 ∈ I.
On the other hand, let us show that I is a two-sided ideal of A. Of
course, one has 0 ∈ I. Moreover, for any 𝑥 ∈ I and 𝑦 ∈ I, one has 𝑥 ℛ 0
and 𝑦 ℛ 0, so that (𝑥 + 𝑦) ℛ 0, which shows that 𝑥 + 𝑦 ∈ I. Finally, for
34 CHAPTER 1. RINGS

𝑥 ∈ I and 𝑎 ∈ A, one has 𝑥 ℛ 0, hence 𝑎𝑥 ℛ 𝑎0 and 𝑥𝑎 ℛ 0𝑎; since

𝑎0 = 0𝑎 = 0, we get that 𝑎𝑥 ∈ I and 𝑥𝑎 ∈ I.
In the opposite direction, the preceding computations show that the
following theorem holds.

Theorem (1.5.2). — Let A be a ring and let I be a two-sided ideal of A. The

binary relation ℛI on A given by 𝑥 ℛI 𝑦 if and only if 𝑥 − 𝑦 ∈ I is an
equivalence relation on A which is compatible with its ring structure. There
exists a unique structure of a ring on the quotient set A/ℛI for which the
canonical surjection clℛI : A → A/ℛI be a morphism of rings. This morphism
is surjective and its kernel is equal to I.

The quotient ring A/ℛI is rather denoted by A/I; the morphism clℛI is
called the canonical surjection from A to A/I and is often denoted by clI
Let 𝑎 be any element of the center of A; observe that clI (𝑎) belongs to the
center of A/I. Let indeed 𝑥 ∈ A/I; there exists 𝑏 ∈ A such that 𝑥 = cl(𝑏);
then, cl(𝑎)𝑥 = cl(𝑎) cl(𝑏) = cl(𝑎𝑏) = cl(𝑏𝑎) since 𝑎 is central, hence
cl(𝑎)𝑥 = cl(𝑏) cl(𝑎) = 𝑥 cl(𝑎). Consequently, for any commutative ring 𝑘
and any ring morphism 𝑖 : 𝑘 → A whose image is contained in the center
of A, so that (A, 𝑖) is a 𝑘-algebra, the composition clI ◦𝑖 : 𝑘 → A → A/I
endowes A/I of a structure of a 𝑘-algebra (in fact, the unique one!) for
which the canonical surjection clI is a morphism of 𝑘-algebras.
Quotient rings are most often used through their universal property
embodied in the following factorization theorem.

Theorem (1.5.3). — Let A and B be rings and let 𝑓 : A → B be a morphism of

rings. For any two-sided ideal I of A which is contained in Ker( 𝑓 ), there exists
a unique ring morphism 𝜑 : A/I → B such that 𝑓 = 𝜑 ◦ clI .
The morphism 𝜑 is surjective if and only if 𝑓 is surjective; the morphism 𝜑
is injective if and only if Ker( 𝑓 ) = I.

It is useful to understand this last equality with the help of the following
diagram:
𝑓
→ B
←
A
←

→
→

clI 𝜑
←
A/I
 1.5. QUOTIENT RINGS 35

in which the two paths that go from A to B (either the arrow 𝑓 , or the
composition 𝜑 ◦ clI of the two arrows 𝜑 and clI ) coincide. For that reason,
one says that this diagram is commutative.
In the context of theorem 1.5.3, the morphisms 𝑓 and clI are given,
while this theorem asserts existence and uniqueness of the morphism 𝜑
that completes the diagram and makes it commutative.
Proof. — Necessarily, 𝜑 has to satisfy 𝜑(clI (𝑎)) = 𝑓 (𝑎) for every 𝑎 ∈ A.
Since every element of A/I is of the form clI (𝑎), for some 𝑎 ∈ A, this
shows that there exists at most one ring morphism 𝜑 : A/I → B such
that 𝑓 = 𝜑 ◦ clI .
Let us now show its existence. Let 𝑥 ∈ A/I and let 𝑎 ∈ A such that
𝑥 = clI (𝑎). Let 𝑎 0 be any other element of A such that 𝑥 = clI (𝑎 0); by
definition, 𝑎 0 − 𝑎 ∈ I, hence 𝑓 (𝑎 0 − 𝑎) = 0 since I ⊂ Ker( 𝑓 ); this shows that
𝑓 (𝑎) = 𝑓 (𝑎 0). Thus one can set 𝜑(𝑥) = 𝑓 (𝑎), the result is independent on
the chosen element 𝑎 such that 𝑥 = clI (𝑎). It remains to show that the
map 𝜑 just defined is a morphism of rings.
Since clI (0A ) = 0A/I and clI (1A ) = 1A/I , we have 𝑓 (0A/I ) = 0B and
𝑓 (1A/I ) = 1B . Moreover, let 𝑥 and 𝑦 be elements of A/I, let 𝑎 and 𝑏 ∈ A
be such that 𝑥 = clI (𝑎) and 𝑦 = clI (𝑏). One has 𝑥 + 𝑦 = clI (𝑎 + 𝑏) and
𝜑(𝑥 + 𝑦) = 𝜑(clI (𝑎 + 𝑏)) = 𝑓 (𝑎 + 𝑏) = 𝑓 (𝑎) + 𝑓 (𝑏)
= 𝜑(clI (𝑎)) + 𝜑(clI (𝑏)) = 𝜑(𝑥) + 𝜑(𝑦).
Similarly,
𝜑(𝑥𝑦) = 𝑓 (𝑎𝑏) = 𝑓 (𝑎) 𝑓 (𝑏) = 𝜑(𝑥)𝜑(𝑦).
Therefore, 𝜑 is a morphism of rings, as claimed, and this completes the
proof of the existence and uniqueness of the morphism 𝜑.
By the surjectivity of clI , the relation 𝑓 = 𝜑 ◦ clI implies that 𝑓 is
surjective if and only if 𝜑 is surjective.
Let us assume that 𝜑 is injective. Let 𝑎 ∈ Ker( 𝑓 ); one has 0 = 𝑓 (𝑎) =
𝜑(clI (𝑎)). Consequently, clI (𝑎) ∈ Ker(𝜑), hence clI (𝑎) = 0 since 𝜑 is
injective; this proves that 𝑎 ∈ I, hence Ker( 𝑓 ) ⊂ I. On the other hand,
I ⊂ Ker( 𝑓 ) by assumption, hence I = Ker( 𝑓 ).
Finally, let us assume that I = Ker( 𝑓 ). Let 𝑥 ∈ A/I be such that 𝜑(𝑥) = 0;
let 𝑎 ∈ A be such that 𝑥 = clI (𝑎); then 𝑓 (𝑎) = 𝜑(𝑥) = 0, hence 𝑎 ∈ I and
𝑥 = 0. This proves that 𝜑 is injective. 
36 CHAPTER 1. RINGS

Let 𝑓 : A → B be a morphism of rings. We saw (page 8) that 𝑓 (A) is a

subring of B. Consequently, the factorization theorem allows us to write
the morphism 𝑓 as
clKer( 𝑓 ) 𝜑
A −−−−→ A/Ker( 𝑓 ) −
→ 𝑓 (A) ↩→ B

that is, as the composition of a surjective morphism, an isomorphism,

and an injective morphism of rings.

Let A be a ring and let I be a two-sided ideal of A. We want to describe

the ideals of the quotient ring A/I. So let 𝒥 be a left ideal of A/I. We
know that cl−1 (𝒥) is a left ideal of A. By construction, it contains the
ideal I, for cl(𝑎) = 0 belongs to 𝒥 for any 𝑎 ∈ I. The important property
is the following proposition.

Proposition (1.5.4). — Let A be a ring and let I be a two-sided ideal of A. The

map cl−1 :

left ideals of A/I → left ideals of A containing I

𝒥 ↦→ cl−1 (𝒥)
is a bijection. The same result holds for right ideals and for two-sided ideals.

In other words, for every left ideal J of A containing I, there exists a

unique ideal 𝒥 of A/I such that J = cl−1 (𝒥). Moreover, 𝒥 = cl(J) (so that,
in this case, the image of the ideal J by the canonical surjection cl still is
a left ideal).
Proof. — Let us first give the inverse map. Let J be a left ideal of A and
let us show that cl(J) is a left ideal of A/I. Obviously, 0 = cl(0) ∈ cl(J).
Moreover, let 𝑥 and 𝑦 belong to cl(J), let 𝑎 and 𝑏 be elements of J such
that 𝑥 = cl(𝑎) and 𝑦 = cl(𝑏). Then, 𝑥 + 𝑦 = cl(𝑎) + cl(𝑏) = cl(𝑎 + 𝑏);
since J is a left ideal of A, 𝑎 + 𝑏 belongs to J and 𝑥 + 𝑦 is an element
of cl(J). Finally, let 𝑥 be an element of cl(J) and let 𝑦 be an element of A/I.
Choose again 𝑎 ∈ J and 𝑏 ∈ A such that 𝑥 = cl(𝑎) and 𝑦 = cl(𝑏). Then
𝑦𝑥 = cl(𝑏) cl(𝑎) = cl(𝑏𝑎) ∈ cl(J) since, J being a left ideal of A, 𝑏𝑎 ∈ J.
If 𝒥 is an ideal of A/I, I claim that

cl(cl−1 (𝒥)) = 𝒥.
 1.5. QUOTIENT RINGS 37

We show the two inclusions separately. Any element 𝑥 of cl(cl−1 (𝒥)) is

of the form 𝑥 = cl(𝑎) for some 𝑎 ∈ cl−1 (𝒥), hence 𝑥 ∈ 𝒥. Conversely, let
𝑥 ∈ 𝒥 and let 𝑎 ∈ A be such that 𝑥 = cl(𝑎); Then cl(𝑎) = 𝑥 ∈ 𝒥, hence 𝑎
belongs to cl−1 (𝒥) and 𝑥 is an element of cl(cl−1 (𝒥)), as claimed.
Let us show that for any left ideal of A,

cl−1 (cl(J)) = I + J.

Again, we show the two inclusions. Let 𝑥 ∈ I + J and let us write 𝑥 = 𝑎 + 𝑏

with 𝑎 ∈ I and 𝑏 ∈ J. Consequently, cl(𝑥) = cl(𝑎) + cl(𝑏) = cl(𝑏) ∈ cl(J),
hence 𝑥 ∈ cl−1 (cl(J)). In the other way, let 𝑥 ∈ cl−1 (cl(J)); by definition,
cl(𝑥) ∈ cl(J) and there exists 𝑎 ∈ J such that cl(𝑥) = cl(𝑎). We get
cl(𝑥 − 𝑎) = 0, which means that 𝑥 − 𝑎 ∈ I. Finally, 𝑥 = (𝑥 − 𝑎) + 𝑎 belongs
to I + J, as was to be shown.
If, moreover, J contains I, then I + J = J and the two boxed formulae
show that the map cl−1 induces an bijection from the set of left ideals
of A/I onto the set of left ideals of A which contain I, whose inverse map
is given by cl. 
When J is a left ideal of A which contains I, the ideal cl(J) of A/I is
also denoted by J/I. This notation is precisely useful when the map cl is
omitted from the notation. In particular, when we write “let J/I be an
ideal of A/I. . . ”, we will always mean that J is an ideal of A containing I.

Proposition (1.5.5). — Let A be a ring, let I be a two-sided ideal of A and

let J be a two-sided ideal of A containing I. The composition of the canonical
surjections A → A/I → (A/I)/(J/I) has kernel J. This gives a canonical
isomorphism
A/J ' (A/I)/(J/I).

In short, a quotient of a quotient is again a quotient.

Proof. — The composition of two surjective morphisms is still a surjective
morphism, hence the morphism A → (A/I)/(J/I) is surjective. An
element 𝑎 ∈ A belongs to its kernel if and only if the element cl(𝑎) ∈ A/I
belongs to the kernel of the morphism clJ/I : A/I → (A/I)/(J/I), which
means cl(𝑎) ∈ (J/I). Since J/I = cl(J) and J contains I, this is equivalent to
the condition 𝑎 ∈ cl−1 (cl(J)) = J.
38 CHAPTER 1. RINGS

The factorization theorem (theorem 1.5.3) then asserts the existence of

a unique morphism 𝜑 : A/J → (A/I)/(J/I) which makes the diagram

→ A/I → (A/I)/(J/I)
← ←
A

←
→

→
← 𝜑
A/J
commutative. This map 𝜑 is surjective. Let us show that is is injective.
Let 𝑥 ∈ A/J be such that 𝜑(𝑥) = 0. Let 𝑎 ∈ A be such that 𝑥 = clJ (𝑎). By
definition, 𝜑(𝑥) = clJ/I ◦ clI (𝑎) = 0, that is, 𝑎 ∈ J. Consequently, 𝑥 = 0 and
the map 𝜑 is injective. This proves that 𝜑 is an isomorphism. 
The last part of this proof can be generalized and gives an important
complement to the factorization theorem.
Proposition (1.5.6). — Let 𝑓 : A → B be a morphism of rings and let I be a
two-sided ideal of A contained in Ker( 𝑓 ). Let 𝜑 : A/I → B be the morphism
given by the factorization theorem. Then the kernel of 𝜑 is given by Ker( 𝑓 )/I.
Proof. — Indeed, let 𝑥 ∈ A/I be such that 𝜑(𝑥) = 0 and let 𝑎 ∈ A be
any element with 𝑥 = clI (𝑎). Then 𝑓 (𝑎) = 0, hence 𝑎 ∈ Ker( 𝑓 ) and
𝑥 = cl(𝑎) ∈ clI (Ker( 𝑓 )) = Ker( 𝑓 )/I. Conversely, if 𝑥 ∈ Ker( 𝑓 )/I, there
exists 𝑎 ∈ Ker( 𝑓 ) such that 𝑥 = clI (𝑎). It follows that 𝜑(𝑥) = 𝑓 (𝑎) = 0 and
𝑥 ∈ Ker(𝜑). 

1.5.7. Chinese remainder theorem. — We say that two two-sided ide-

als I and J of a ring A are comaximal if I + J = A. This notion gives rise to
the general formulation of the Chinese remainder theorem.
Theorem (1.5.8). — Let A be a ring, let I and J be two-sided ideals of A.
Let us assume that I and J are comaximal. Then the canonical morphism
A → (A/I) × (A/J) given by 𝑎 ↦→ (clI (𝑎), clJ (𝑎)) is surjective; its kernel is the
two-sided ideal I ∩ J of A. Passing to the quotient, we thus have an isomorphism
A/(I ∩ J) ' A/I × A/J.
Corollary (1.5.9). — Let I and J be comaximal two-sided ideals of a ring A.
For any pair (𝑥, 𝑦) of elements of A, there exists 𝑎 ∈ A such that 𝑎 ∈ 𝑥 + I and
𝑎 ∈ 𝑦 + J.
 1.6. FRACTION RINGS OF COMMUTATIVE RINGS 39

Proof. — In the diagram of rings

A
(clI ,clJ )
clI∩J
← ←
→ →
𝜑
A/(I ∩ J) → A/I × A/J
←

we have to show the existence of a unique morphism 𝜑, drawn as a dashed

arrow, so that this diagram is commutative, and that 𝜑 is an isomorphism.
But the morphism A → A/I × A/J maps 𝑎 ∈ A to (clI (𝑎), clJ (𝑎)). Its
kernel is thus I ∩ J. By the factorization theorem, there exists a unique
morphism 𝜑 that makes the diagram commutative, and this morphism
is injective. For any 𝑎 ∈ A, one has 𝜑(clI∩J (𝑎)) = (clI (𝑎), clJ (𝑎)).
It remains to show that 𝜑 is surjective. Since I + J = A, there are 𝑥 ∈ I
and 𝑦 ∈ J such that 𝑥 + 𝑦 = 1. Then, we have 1 = clI (𝑥 + 𝑦) = clI (𝑦) in
A/I and 1 = clJ (𝑥 + 𝑦) = clJ (𝑥) in A/J. Let 𝜉 = clI∩J (𝑥) and 𝜂 = clI∩J (𝑦).
For any 𝑎, 𝑏 ∈ A, we get
𝜑(𝑏𝜉 + 𝑎𝜂) = 𝑓 (𝑏𝑥 + 𝑎𝑦) = (clI (𝑎), clJ (𝑏)).
Since any element of (A/I) × (A/J) is of the form (clI (𝑎), clJ (𝑏)), the
morphism 𝜑 is surjective. 

Remark (1.5.10). — Let I and J be ideals of a commutative ring such that

I + J = A; then I ∩ J = IJ.
We had already noticed the inclusion IJ ⊂ I ∩ J, which does not require
the ring A to be commutative, nor the ideals I and J to be comaximal.
Conversely, let 𝑎 ∈ I ∩ J. Since I + J = A, there are 𝑥 ∈ I and 𝑦 ∈ J such
that 𝑥 + 𝑦 = 1. Then 𝑎 = (𝑥 + 𝑦)𝑎 = 𝑥𝑎 + 𝑦𝑎. Since 𝑥 ∈ I and 𝑎 ∈ J, 𝑥𝑎 ∈ IJ;
since 𝑦𝑎 = 𝑎𝑦, 𝑦 ∈ J, and 𝑎 ∈ I, we get 𝑦𝑎 = 𝑎𝑦 ∈ IJ. Finally, 𝑎 ∈ IJ.

1.6. Fraction rings of commutative rings

In the preceding section devoted to quotients, we somewhat “forced”
some elements of a ring to vanish. We now want to perform a quite
opposite operation: making invertible all the elements of an adequate
subset. In all of this section, we restrict to the case of commutative rings (for a
preview of the general situation, see remark 1.6.8).
40 CHAPTER 1. RINGS

Definition (1.6.1). — Let A be a commutative ring. A subset S of A is said to

be multiplicative if the following properties hold:
(i) 1 ∈ S;
(ii) For any 𝑎 and 𝑏 in S, 𝑎𝑏 ∈ S.

Given a commutative ring A and a multiplicative subset S of A, our

goal is to construct a ring S−1 A together with a morphism 𝑖 : A → S−1 A
such that 𝑖(S) consists of invertible elements of S−1 A, and which is
“universal” for this property.
Let us first give a few examples.

Examples (1.6.2). — a) Let A = Z and S = Z {0}, the ring S−1 A will

be equal to Q and 𝑖 : Z → Q the usual injection. More generally, if A is a
domain, then S = A {0} is a multiplicative subset of A and the ring
S−1 A will be nothing else but the field of fractions of A.
b) If all elements of S are invertible in A, then S−1 A = A.
c) Let A = Z and S = {1; 10; 100; . . .} be the set of all powers of 10 in Z.
Then S−1 A is the set of decimal numbers, those rational numbers which
can be written as 𝑎/10𝑛 , for some 𝑎 ∈ Z and some positive integer 𝑛.

As these examples show, we are simply going to mimick middle school

calculus of fractions.

1.6.3. Construction. — On the set A × S, let us define an equivalence

relation ∼ by:
(𝑎, 𝑠) ∼ (𝑏, 𝑡) if and only if there exists 𝑢 ∈ S such that (𝑎𝑡−𝑏𝑠)𝑢 = 0.
It is indeed an equivalence relation:
– For any (𝑎, 𝑠) ∈ A × S, (𝑎, 𝑠) ∼ (𝑎, 𝑠), since 1 ∈ S and (𝑎𝑠 − 𝑎𝑠)1 = 0:
the relation is reflexive;
– If (𝑎, 𝑠) ∼ (𝑏, 𝑡), let us choose 𝑢 ∈ S such that (𝑎𝑡 − 𝑏𝑠)𝑢 = 0. Then
(𝑏𝑠 − 𝑎𝑡)𝑢 = 0, hence (𝑏, 𝑡) ∼ (𝑎, 𝑠): the relation is symmetric;
– Finally, if (𝑎, 𝑠) ∼ (𝑏, 𝑡) and (𝑏, 𝑡) ∼ (𝑐, 𝑢), let 𝑣 and 𝑤 ∈ S satisfy
(𝑎𝑡 − 𝑏𝑠)𝑣 = (𝑏𝑢 − 𝑐𝑡)𝑤 = 0. Since
(𝑎𝑢 − 𝑐𝑠)𝑡 = (𝑎𝑡 − 𝑏𝑠)𝑢 + (𝑏𝑢 − 𝑐𝑡)𝑠,
 1.6. FRACTION RINGS OF COMMUTATIVE RINGS 41

we have (𝑎𝑢 − 𝑐𝑠)𝑣𝑤𝑡 = 0 (here we use that the ring A is commutative).

Since 𝑣, 𝑤 and 𝑡 belong to S and S is a multiplicative subset, one has
𝑣𝑤𝑡 ∈ S, hence (𝑎, 𝑠) ∼ (𝑐, 𝑢): the relation is transitive.
We write S−1 A for the set of equivalence classes; the class of a pair
(𝑎, 𝑠) is denoted by 𝑎/𝑠; let also 𝑖 : A → S−1 A be the map that sends
𝑎 ∈ A to the class 𝑎/1. The set A × S is not a ring, in any reasonable way.
However we are going to endow the quotient set S−1 A with the structure
of a ring in such a way that the map 𝑖 is a morphism of rings.
The definition comes from the well known formulae for the sum or
product of fractions: we define

(𝑎/𝑠) + (𝑏/𝑡) = (𝑎𝑡 + 𝑏𝑠)/𝑠𝑡, (𝑎/𝑠) · (𝑏/𝑡) = (𝑎𝑏/𝑠𝑡).

Let us first check that these formulae make sense: if (𝑎, 𝑠) ∼ (𝑎 0 , 𝑠 0) and
(𝑏, 𝑡) ∼ (𝑏 0 , 𝑡 0), we have to show that

(𝑎𝑡 + 𝑏𝑠, 𝑠𝑡) ∼ (𝑎 0𝑡 0 + 𝑏 0 𝑠 0 , 𝑠 0𝑡 0) and (𝑎𝑏, 𝑠𝑡) ∼ (𝑎 0𝑏 0 , 𝑠 0𝑡 0).

Observe that

(𝑎𝑡 + 𝑏𝑠)𝑠 0𝑡 0 − (𝑎 0𝑡 + 𝑏 0 𝑠 0)𝑠𝑡 = (𝑎𝑠 0 − 𝑎 0 𝑠)𝑡𝑡 0 + (𝑏𝑡 0 − 𝑏 0𝑡)𝑠𝑠 0.

Let us choose 𝑢, 𝑣 ∈ S such that (𝑎𝑠 0 − 𝑎 0 𝑠)𝑢 = (𝑏𝑡 0 − 𝑏 0𝑡)𝑣 = 0; we get

((𝑎𝑡 + 𝑏𝑠)𝑠 0𝑡 0 − (𝑎 0𝑡 + 𝑏 0 𝑠 0)𝑠𝑡) 𝑢𝑣 = (𝑎𝑠 0−𝑎 0 𝑠)𝑢(𝑡𝑡 0𝑣)+(𝑏𝑡 0−𝑏 0𝑡)𝑣(𝑢𝑠𝑠 0) = 0,

hence (𝑎𝑡 + 𝑏𝑠, 𝑠𝑡) ∼ (𝑎 0𝑡 + 𝑏 0 𝑠 0 , 𝑠 0𝑡 0). Similarly,

(𝑎𝑏𝑠 0𝑡 0 − 𝑎 0𝑏 0 𝑠𝑡)𝑢𝑣 = (𝑎𝑠 0 − 𝑎 0 𝑠)𝑢(𝑏𝑡 0𝑣) + (𝑏𝑠 0 − 𝑏 0 𝑠)𝑣(𝑢𝑎 0 𝑠𝑣) = 0,

so that (𝑎𝑏, 𝑠𝑡) ∼ (𝑎 0𝑏 0 , 𝑠 0𝑡 0).

Checking that these laws give S−1 A the structure of a ring is a bit
long, but without surprise, and we shall not do it here. For example,
distributivity of addition on multiplication can be proven in the following
way: let 𝑎/𝑠, 𝑏/𝑡 and 𝑐/𝑢 be elements of S−1 A, then

𝑎 𝑏 𝑐 𝑎(𝑏𝑢 + 𝑐𝑡) 𝑎𝑏𝑢 𝑎𝑐𝑡 𝑎𝑏 𝑎𝑐 𝑎𝑏 𝑎 𝑐

 
+ = = + = + = + .
𝑠 𝑡 𝑢 𝑠𝑡𝑢 𝑠𝑡𝑢 𝑠𝑡𝑢 𝑠𝑡 𝑠𝑢 𝑠 𝑡 𝑠 𝑢

The unit element of S−1 A is 1/1, the zero element is 0/1.

42 CHAPTER 1. RINGS

The map 𝑖 : A → S−1 A given, for any 𝑎 ∈ A, by 𝑖(𝑎) = 𝑎/1 is a

morphism of rings. Indeed, 𝑖(0) = 0/1 = 0, 𝑖(1) = 1/1 = 1, and, for any 𝑎
and 𝑏 ∈ A,
𝑖(𝑎 + 𝑏) = (𝑎 + 𝑏)/1 = 𝑎/1 + 𝑏/1 = 𝑖(𝑎) + 𝑖(𝑏)
and
𝑖(𝑎𝑏) = (𝑎𝑏)/1 = (𝑎/1)(𝑏/1) = 𝑖(𝑎)𝑖(𝑏).
Finally, for 𝑠 ∈ S, 𝑖(𝑠) = 𝑠/1 and 𝑖(𝑠)(1/𝑠) = 𝑠/𝑠 = 1, so that 𝑖(𝑠) is
invertible in S−1 A for any 𝑠 ∈ S.
Remarks (1.6.4). — a) A fraction 𝑎/𝑠 is zero if and only if (𝑎, 𝑠) ∼ (0, 1).
By the definition of the relation ∼, this holds if and only if there exists
𝑡 ∈ S such that 𝑡𝑎 = 0. In particular, the canonical morphism 𝑖 : A → S−1 A
is injective if and only if every element of S is regular.
b) In particular, a ring of fractions S−1 A is null if and only if 0 belongs
to S. Indeed, S−1 A = 0 means that 1/1 = 1 = 0, hence that there exists
𝑡 ∈ S such that 𝑡 · 1 = 𝑡 = 0, in other words, that 0 ∈ S. Informally, if you
want to divide by 0 and still pretend that all classical rules of algebra
apply, then everything becomes 0 (and algebra is not very interesting
anymore).
A posteriori, this explains the middle school rule that thou shalt never
divide by zero! if you would, the rules of calculus of fractions (imposed
by algebra) would make any fraction equal to 0!
c) The definition of the equivalence relation used in the construction
of the ring of fractions is at first surprising. It is in any case more
complicated that the middle school rule that would claim that 𝑎/𝑠 = 𝑏/𝑡
if and only if 𝑎𝑡 = 𝑏𝑠. If the ring A is a domain and 0 ∉ S, more generally
when all elements of S are not zero divisors, this simpler definition is
sufficient. However, in the general case, the simpler definition does not
give rise to an equivalence relation.
The importance of this construction is embodied in its universal property,
another factorization theorem.
Theorem (1.6.5). — Let A be a commutative ring and let S be a multiplicative
subset of A. Let 𝑖 : A → S−1 A be the ring morphism that has been constructed
above. Then, for any (possibly not commutative) ring B and any morphism
 1.6. FRACTION RINGS OF COMMUTATIVE RINGS 43

𝑓 : A → B such that 𝑓 (S) ⊂ B× , there exists a unique ring morphism

𝑓 : S−1 A → B such that 𝑓 = 𝜑 ◦ 𝑖.

One can sum up this last formula by saying that the diagram
𝑓
→ B
←
A

←
𝑖
→

→
𝜑
←

S−1 A
is commutative.
Proof. — If such a morphism 𝜑 exists, it has to satisfy
𝜑(𝑎/𝑠) 𝑓 (𝑠) = 𝜑(𝑎/𝑠)𝜑(𝑖(𝑠)) = 𝜑(𝑎/𝑠)𝜑(𝑠/1) = 𝜑(𝑎/1) = 𝜑(𝑖(𝑎)) = 𝑓 (𝑎)
hence
𝜑(𝑎/𝑠) = 𝑓 (𝑎) 𝑓 (𝑠)−1 ,
where 𝑓 (𝑠)−1 is the inverse of 𝑓 (𝑠) in B. Similarly, it also has to satisfy
𝑓 (𝑠)𝜑(𝑎/𝑠) = 𝜑(𝑖(𝑠))𝜑(𝑎/𝑠) = 𝜑(𝑠/1)𝜑(𝑎/𝑠) = 𝜑(𝑎/1) = 𝑓 (𝑎),
hence 𝜑(𝑎/𝑠) = 𝑓 (𝑠)−1 ( 𝑓 (𝑎). This shows that there can exist at most one
such morphism 𝜑.
To establish its existence, it now suffices to check that these two
formulae are compatible and define a ring morphism 𝜑 : S−1 A → B such
that 𝜑 ◦ 𝑖 = 𝑓 .
Let us check that the formula is well posed. So let 𝑎, 𝑏 ∈ A, 𝑠, 𝑡 ∈ S
such that 𝑎/𝑠 = 𝑏/𝑡, let 𝑢 ∈ S be such that (𝑎𝑡 − 𝑏𝑠)𝑢 = 0. Then,
𝑓 (𝑎𝑡) 𝑓 (𝑢) = 𝑓 (𝑏𝑠) 𝑓 (𝑢), hence 𝑓 (𝑎𝑡) = 𝑓 (𝑏𝑠) since 𝑓 (𝑢) is invertible in B.
Since A is commutative, it follows that 𝑓 (𝑡) 𝑓 (𝑎) = 𝑓 (𝑎) 𝑓 (𝑡) = 𝑓 (𝑏) 𝑓 (𝑠),
hence 𝑓 (𝑎) 𝑓 (𝑠)−1 = 𝑓 (𝑡)−1 𝑓 (𝑏).
Applying this equality to (𝑏, 𝑡) = (𝑎, 𝑠), we also obtain the relation
𝑓 (𝑏) 𝑓 (𝑡)−1 = 𝑓 (𝑡)−1 𝑓 (𝑏)
This proves that there is a map 𝜑 : S−1 A → B such that 𝑓 (𝑎/𝑠) =
𝑓 (𝑎) 𝑓 (𝑠)−1 = 𝑓 (𝑠)−1 𝑓 (𝑎) for any 𝑎 ∈ A and any 𝑠 ∈ S.
By construction, for any 𝑎 ∈ A, one has 𝜑◦𝑖(𝑎) = 𝜑(𝑎/1) = 𝑓 (𝑎) 𝑓 (1)−1 =
𝑓 (𝑎), hence 𝜑 ◦ 𝑖 = 𝑓 .
Let us now verify that 𝜑 is a ring morphism. We have
𝜑(0) = 𝑓 (0/1) = 𝑓 (1)−1 𝑓 (0) = 0 and 𝜑(1) = 𝑓 (1/1) = 𝑓 (1)−1 𝑓 (1) = 1.
44 CHAPTER 1. RINGS

Then,
𝜑(𝑎/𝑠) + 𝜑(𝑏/𝑡) = 𝑓 (𝑎) 𝑓 (𝑠)−1 + 𝑓 (𝑏) 𝑓 (𝑡)−1 = 𝑓 (𝑠𝑡)−1 𝑓 (𝑎𝑡) + 𝑓 (𝑏𝑠)

= 𝑓 (𝑠𝑡)−1 𝑓 (𝑎𝑡 + 𝑏𝑠) = 𝜑((𝑎𝑡 + 𝑏𝑠)/𝑠𝑡) = 𝜑((𝑎/𝑠) + (𝑏/𝑡)).

Finally,
𝜑(𝑎/𝑠)𝜑(𝑏/𝑡) = 𝑓 (𝑠)−1 𝑓 (𝑎) 𝑓 (𝑡)−1 𝑓 (𝑏) = 𝑓 (𝑠𝑡)−1 𝑓 (𝑎𝑏)
= 𝜑(𝑎𝑏/𝑠𝑡) = 𝜑((𝑎/𝑠)(𝑏/𝑡)).
The map 𝜑 is thus a ring morphism. That concludes the proof of the
theorem. 
It is also possible to construct the fraction ring as a quotient. I content
myself with the particular case of a multiplicative subset consisting of
the powers of a single element; the generalization to any multiplicative
subset is left as an exercise (exercice 1/48).

Proposition (1.6.6). — Let A be a commutative ring, let 𝑠 be any element of A

and let S = {1; 𝑠; 𝑠 2 ; . . .} be the multiplicative subset of A consisting of powers
of 𝑠. The canonical morphism
𝑓 : A[X] → S−1 A, P ↦→ P(1/𝑠)
is surjective, its kernel is the ideal (1 − 𝑠X). Passing to the quotient, one deduces
an isomorphism
𝜑 : A[X]/(1 − 𝑠X) ' S−1 A.

Proof. — Any element of S−1 A can be written 𝑎/𝑠 𝑛 for some 𝑎 ∈ A and
some integer 𝑛 > 0. Then, 𝑎/𝑠 𝑛 = 𝜑(𝑎X𝑛 ) so that 𝑓 is surjective. Its
kernel contains the polynomial 1 − 𝑠X since 𝑓 (1 − 𝑎X) = 1 − 𝑎/𝑎 = 0.
Therefore, it contains the ideal (1 − 𝑠X) generated by this polynomial.
By the universal property of quotient rings, we get a well-defined
ring morphism 𝜑 : A[X]/(1 − 𝑠X) → S−1 A which maps the class cl(P)
modulo (1 − 𝑠X) of a polynomial P to 𝑓 (P) = P(1/𝑠). Let us show that
𝜑 is an isomorphism. By proposition 1.5.6, it will then follow that
Ker( 𝑓 ) = (1 − 𝑠X).
To show that 𝜑 is an isomorphism, we shall construct its inverse. This
inverse 𝜑 should satisfy 𝜓(𝑎/𝑠 𝑛 ) = cl(𝑎X𝑛 ) for all 𝑎 ∈ A and all 𝑛 ∈ N; we
could check this by a direct computation, but let us rather choose a more
 1.6. FRACTION RINGS OF COMMUTATIVE RINGS 45

abstract way. Let 𝑔 : A → A[X]/(1 − 𝑠X) be the canonical morphism such

that 𝑔(𝑎) = cl(𝑎), the class of the constant polynomial 𝑎 modulo 1− 𝑠X. In
the ring A[X]/(1 − 𝑠X), we have cl(𝑠X) = 1, so that cl(𝑠) is invertible, with
inverse cl(X). By the universal property of fraction rings, there exists a
unique morphism 𝜓 : S−1 A → A[X]/(1 − 𝑠X) such that 𝜓(𝑎/1) = 𝑔(𝑎)
for any 𝑎 ∈ A. By construction, for any 𝑎 ∈ A and any 𝑛 > 0, one has
𝜓(𝑎/𝑠 𝑛 ) = 𝑎 cl(X𝑛 ) = cl(𝑎X𝑛 ).
Finallly, let us show that 𝜓 and 𝜑 are inverses one of the other. For
P ∈ A[X], one has 𝜓(𝜑(cl(P))) = 𝜓(P(1/𝑠)). Consequently, if P = 𝑎 𝑛 X𝑛 ,
Í
we get
𝜓(𝜑(cl(P))) = 𝜓( 𝑓 (P)) = 𝜓(P(1/𝑠))
Õ Õ
𝑛
= 𝜓( (𝑎 𝑛 /𝑠 )) = 𝜓(𝑎 𝑛 /𝑠 𝑛 )
Õ Õ
𝑛
𝑎 𝑛 X𝑛 = cl(P)


= cl(𝑎 𝑛 X ) = cl
and 𝜓 ◦ 𝜑 = id. Finally,
𝜑(𝜓(𝑎/𝑠 𝑛 )) = 𝜑(cl(𝑎X𝑛 )) = 𝑓 (𝑎X𝑛 ) = 𝑎/𝑠 𝑛
and 𝜑 ◦ 𝜓 = id. The ring morphism 𝜑 is therefore an isomorphism, with
inverse 𝜓, as was to be shown. 

Examples (1.6.7). — a) Let A be a domain. The set S = A {0} is a

multiplicative subset of A. The morphism 𝑖 : A → S−1 is injective and
the ring S−1 A is a field, called the field of fractions of A.
Proof. — Since A is a domain, 1 ≠ 0 and 1 ∈ S. On the other hand, for
any two nonzero elements 𝑎 and 𝑏 ∈ A, their product 𝑎𝑏 is nonzero, for
A is a domain. This shows that S is a multiplicative subset of A. Since
no element of S is a zero divisor, the morphism 𝑖 : A → S−1 A is injective.
Any element 𝑥 of S−1 A can be written 𝑎/𝑠, for some 𝑎 ∈ A and some
𝑠 ≠ 0. Assume 𝑥 = 0; then, there exists 𝑏 ∈ S such that 𝑎𝑏 = 0. Then
𝑏 ≠ 0 and, since A is a domain, 𝑎 = 0. Reversing the argument, any
element 𝑎/𝑠 of S−1 A, with 𝑎 ≠ 0, is distinct from 0. In particular, 1/1 ≠ 0
and the ring S−1 A is nonnull. Let again 𝑥 = 𝑎/𝑠 be any nonzero element
of S−1 A. Then 𝑎 ≠ 0, so that we can consider the element 𝑦 = 𝑠/𝑎 of S−1 A.
Obviously, 𝑥𝑦 = (𝑎/𝑠)(𝑠/𝑎) = 𝑎𝑠/𝑎𝑠 = 1, hence 𝑥 is invertible.
We thus have shown that S−1 A is a field. 
46 CHAPTER 1. RINGS

b) Let A be a commutative ring and let 𝑠 ∈ A by any element of A which

is not nilpotent. Then, the subset S = {1; 𝑠; 𝑠 2 ; . . .} of A is a multiplicative
subset (obviously) which does not contain 0. By remark 1.6.4, b) above,
the ring S−1 A is nonnull; it is usually denoted by A𝑠 .
c) Let 𝑓 : A → B be a morphism of commutative rings. For any
multiplicative subset S of A, T = 𝑓 (S) is a multiplicative subset of B.
Moreover, there is a unique morphism F : S−1 A → T−1 B extending 𝑓 ,
in the sense that F(𝑎/1) = 𝑓 (𝑎)/1 for any 𝑎 ∈ A. This follows from the
universal property explained below applied to the morphism F1 from A
to T−1 B given by 𝑎 ↦→ 𝑓 (𝑎)/1; one just needs to check that for any 𝑠 ∈ A,
F1 (𝑠) = 𝑠/1 ∈ T, which is the very definition of T. This can also be shown
by hand, copying the proof of the universal property to this particular
case.
If the morphism 𝑓 is implicit, for example if B is given as an A-algebra,
we will commit the abuse of writing S−1 B instead of T−1 B.
d) Let 𝑓 : A → B be a morphism of commutative rings and let T be a
multiplicative subset of B. Then, S = 𝑓 −1 (T) is a multiplicative subset
of A such that 𝑓 (S) ⊂ T. There is a unique morphism F : S−1 A → T−1 B
extending the morphism 𝑓 .
e) Let I be an ideal of a commutative ring A. The set S = 1 + I of
elements 𝑎 ∈ A such that 𝑎 − 1 ∈ I is a multiplicative subset. Indeed,
this is the inverse image of the multiplicative subset {1} of A/I by the
canonical morphism A → A/I.
f ) Let A be a commutative ring and let P be an ideal of A such that
P ≠ A and such that the condition the ideal I is distinct from A and the
condition 𝑎𝑏 ∈ P implies 𝑎 ∈ P or 𝑏 ∈ P (we shall say that P is a prime
ideal of A, see definition 2.2.2). Then A P is a multiplicative subset of A
(proposition 2.2.3) and the fraction ring (A P)−1 A is generally denoted
by AP .

Remark (1.6.8). — Let A be a (non necessarily commutative) ring and

let S be a multiplicative subset of A. There are several purely formal
ways of proving existence and uniqueness of a ring AS , endowed with
a ring morphism 𝑖 : A → AS , that satisfies the universal property: for
any ring morphism 𝑓 : A → B such that 𝑓 (S) ⊂ B× , there exists a unique
 1.7. RELATIONS BETWEEN QUOTIENT RINGS AND FRACTION RINGS 47

ring morphism 𝑓˜ : AS → B such that 𝑓 = 𝑓˜ ◦ 𝑖. One of them consists

in defining rings of polynomials in noncommuting indeterminated
(replacing the commutative monoid N(I) on a set I by the word monoid I∗
on I), adding such indeterminates T𝑠 for all 𝑠 ∈ S, and quotienting this
ring by the two-sided ideal generated by the elements 𝑠T𝑠 − 1 and T𝑠 𝑠 − 1,
for 𝑠 ∈ S.
However, almost nothing can be said about it. For example, in 1937,
Malcev constructed a ring A without any nonzero zero divisor that
admits no morphism to a field (1 should map to 0!). (See A. Malcev
(1937), “On the immersion of an algebraic ring into a field”, Math. Ann.,
vol. 113, p. 686–691.)
For rings satisfying the so-called Ore condition, one can check that
the construction using fractions furnishes a ring for which the universal
property holds; see exercise 1/45.

1.7. Relations between quotient rings and fraction rings

Finally, I want to study briefly the ideals of a fraction ring S−1 A. Here
is a first result.
Proposition (1.7.1). — Let A be a commutative ring and let S be a multiplicative
subset of A; let 𝑖 : A → S−1 A be the canonical morphism. For any ideal ℐ of
S−1 A, there exists an ideal I of A such that ℐ = 𝑖(I)(S−1 A). In fact, the ideal
I = 𝑖 −1 (ℐ) is the largest such ideal.
Proof. — If ℐ = 𝑖(I)(S−1 A), we have in particular 𝑖(I) ⊂ ℐ, that is
I ⊂ 𝑖 −1 (ℐ). Provided that the ideal 𝑖 −1 (ℐ) satisfies the given assertion,
this shows that it is the largest such ideal. We thus need to show that
ℐ = 𝑖(𝑖 −1 (ℐ))(S−1 A).
Since 𝑖(𝑖 −1 (ℐ)) ⊂ ℐ, the ideal generated by 𝑖(𝑖 −1 (ℐ)) is contained in ℐ,
hence the inclusion
𝑖(𝑖 −1 (ℐ))(S−1 A) ⊂ ℐ.
Conversely, let 𝑥 ∈ ℐ, let us choose 𝑎 ∈ A and 𝑠 ∈ S such that 𝑥 = 𝑎/𝑠.
Then, 𝑠𝑥 = 𝑖(𝑎) ∈ ℐ so that 𝑎 ∈ 𝑖 −1 (ℐ). It follows that 𝑠𝑥 ∈ 𝑖(𝑖 −1 (ℐ)),
hence 𝑥 = (𝑠𝑥)(1/𝑠) belongs to 𝑖(𝑖 −1 (ℐ))(S−1 A), the other required
inclusion. 
48 CHAPTER 1. RINGS

To shorten the notation, we can omit the morphism 𝑖 and write IS−1 A
instead of 𝑖(I)S−1 A. In fact, we shall rather denote this ideal by S−1 I,
this last notation being the one used in the general context of fraction
modules (§3.6).

Proposition (1.7.2). — Let A be a commutative ring, let S be a multiplicative

subset of A and let I be an ideal of A. Let T = clI (S) ⊂ A/I be the image of S by
the canonical surjection clI : A → A/I. There exists a unique ring morphism
𝜑 : S−1 A/S−1 I → T−1 (A/I)
such that for any 𝑎 ∈ A, 𝜑(clI (𝑎/1)) = clI (𝑎)/1. Moreover, 𝜑 is an isomor-
phism.

In a more abstract way, the two rings S−1 A/S−1 I and T−1 (A/I) are A-
algebras (a quotient, or a fraction ring, of an A-algebra is an A-algebra).
The proposition states that there exists a unique morphism of A-algebras
between this two rings, and that this morphism is an isomorphism.
Proof. — It is possible to give an explicit proof, but it is more elegant (say,
more abstract and less computational) to rely on the universal properties
of quotients and fraction rings. Let us consider the composition
𝑓 : A → A/I → T−1 (A/I), 𝑎 ↦→ cl(𝑎)/1.
By this morphism, an element 𝑠 ∈ S is sent to cl(𝑠)/1, an invertible
element of T−1 (A/I) whose inverse is 1/cl(𝑠). By the universal property
of fraction rings, there exists a unique ring morphism
𝜑1 : S−1 A → T−1 (A/I)
such that 𝜑1 (𝑎/1) = cl(𝑎)/1.
If, moreover, 𝑎 ∈ I, then
𝜑1 (𝑎/1) = 𝜑1 (𝑎)/1 = cl(𝑎)/1 = 0
since cl(𝑎) = 0 in A/I. It follows that Ker(𝜑 𝑖 ) contains the image of I
in S−1 A; it thus contains the ideal S−1 I generated by I in S−1 A. By the
universal property of quotient rings, there exists a unique ring morphism
𝜑 : S−1 A/S−1 I → T−1 (A/I)
such that 𝜑(cl(𝑎/𝑠)) = cl(𝑎)/cl(𝑠) for any 𝑎/𝑠 ∈ S−1 A.
 1.7. RELATIONS BETWEEN QUOTIENT RINGS AND FRACTION RINGS 49

We have shown that there exists a morphism 𝜑 : S−1 A/S−1 I → T−1 (A/I)
of A-algebras. We can sum up the construction by the commutative
diagram
clS−1 I
S−1 A → S1 A/S−1 I
←
𝑖 →

←
← ←
𝜑
A 𝜑1
← clI →

→
←
→
→ T−1 (A/I).
←
A/I 𝑗
𝑓
→

We also see that 𝜑 is the only possible morphism of A-algebras. Indeed,

by the universal property of quotient rings, the morphism 𝜑 is charac-
terized by its restriction to S−1 A which, by the universal property of
fraction rings, is itself characterized by its restriction to A.
It remains to prove that 𝜑 is an isomorphism. For that, let us now
consider this diagram in the other direction. The kernel of the morphism
𝑔 = clS−1 I ◦𝑖 contains I, hence there is a unique morphism of A-algebras

𝜓1 : A/I → S−1 A/S−1 I

satisfying 𝜓1 (cl(𝑎)) = cl(𝑎/1) for any 𝑎 ∈ A. For every 𝑠 ∈ S, 𝜓1 (cl(𝑠)) =

cl(𝑠/1) is invertible, with inverse cl(1/𝑠). Therefore, the image of T by 𝜓1
consists of invertible elements in S−1 A/S−1 I. There thus exists a unique
morphism of A-algebras,

𝜓 : T−1 (A/I) → S−1 A/S−1 I

(that is, such that 𝜓(cl(𝑎)/1) = cl(𝑎/1) for any 𝑎 ∈ A). Again, these
constructions are summarized by the commutative diagram:
𝑔
→
S−1 A → S1 A/S−1 I
←
→
←

←
𝑖 𝜓1 →
A 𝜓
←

← ←
clI →
→ T−1 (A/I).
←
A/I
50 CHAPTER 1. RINGS

Finally, for 𝑎 ∈ A and 𝑠 ∈ S, we have 𝜑(cl(𝑎/𝑠)) = cl(𝑎)/cl(𝑠) in T−1 (A/I)

and 𝜓(cl(𝑎)/cl(𝑠)) = cl(𝑎/𝑠) in S−1 A/S−1 I. Necessarily, 𝜑 ◦ 𝜓 and 𝜓 ◦ 𝜑
are the identical maps, so that 𝜑 is an isomorphism. 
This property will appear later under the fancy denomination of
exactness of localization.

Exercises
1-exo503) a) Let A be a ring. Let us write Ao for the abelian group A endowed with the
multiplication • defined by 𝑎 • 𝑏 = 𝑏𝑎. Then Ao is a ring, called the opposite ring of A.
b) Let A be the ring of 𝑛×𝑛 matrices with coefficients in C. Show that the transposition
map, M ↦→ Mt , is an isomorphism from A to Ao .
c) Show that the conjugation 𝑞 ↦→ 𝑞¯ of quaternions induces an isomorphism from H
to Ho .
2-exo515) a) Let A be a ring and let (B𝑖 ) be a family of subrings of A. Show that the
intersection of all B𝑖 is a subring of A.
b) Let A and B be rings, let 𝑓 , 𝑔 : A → B be morphisms of rings, and let C be the set
of all 𝑎 ∈ A such that 𝑓 (𝑎) = 𝑔(𝑎). Prove that C is a subring of A.
c) Let A be a ring, let B be a subring of A and let I be a two-sided ideal of A. Let R be
the set of all sums 𝑎 + 𝑏, for 𝑎 ∈ B and 𝑏 ∈ I. Show that R is a subring of A.
3-exo513) Let A be a ring and let S be a subset of A. In the following two cases,
determine explicitly the centralizer CS (A).
a) When A = EndK (V) is the ring of endomorphisms of a finite dimensional K-vector
space V and S = {𝑢} consists in a diagonalizable endomorphism 𝑢;
b) When A = M𝑛 (C) is the ring of complex 𝑛 × 𝑛 matrices and S = {M} consists in a
matrix with minimal polynomial X𝑛 .
4-exo520) Let K be a field and let V be a finite dimensional K-vector space. Show that
the center of the ring EndK (V) consists in the homotheties 𝑥 ↦→ 𝑎𝑥, for 𝑎 ∈ K.
5-exo525) Let A be a ring and let M be a monoid. Let Z be the center of A.
a) For 𝑚 ∈ M, let 𝛿 𝑚 be the function from M to A with value 1 at 𝑚, and 0 elsewhere.
Compute the convolution product 𝛿 𝑚 ∗ 𝛿 𝑚0 in the monoid ring A(M) .
b) For 𝑚 ∈ M, compute the centralizer of the element 𝛿 𝑚 in A(M) .
c) Assume that M is a group G. Show that the center of the ring A(G) consists in all
functions 𝑓 : G → Z with finite support which are constant on every conjugacy class
in G.
6-exo527) a) Let 𝜕 : C[X] → C[X] be the map given by 𝜕(P) = P0, for P ∈ C[X]. A
differential operator on C[X] is a C-linear map of C[X] to itself of the form P ↦→
Í𝑛 𝑖
𝑖=0 𝑎 𝑖 (X)𝜕 (P), where the 𝑎 𝑖 are polynomials.
 EXERCISES 51

b) Show that the set of all differential operators on C[X], endowed with addition and
composition, is a ring.
c) Determine its center.
7-exo529) Let 𝑓 : A → B be a ring morphism.
a) Let R be the set of all ordered pairs (𝑎, 𝑏) ∈ A × A such that 𝑓 (𝑎) = 𝑓 (𝑏). Show
that R, with termwise addition and multiplication, is a ring.
b) One says that 𝑓 is a monomorphism if 𝑔 = 𝑔 0 for any ring C and any pair (𝑔, 𝑔 0) of
ring morphisms from C to A such that 𝑓 ◦ 𝑔 = 𝑓 ◦ 𝑔 0.
Show that a ring morphism is a monomorphism if and only if it is injective.
c) One says that 𝑓 is an epimorphism if 𝑔 = 𝑔 0 for any ring C and any pair (𝑔, 𝑔 0) of
ring morphisms from B to C such that 𝑔 ◦ 𝑓 = 𝑔 0 ◦ 𝑓 .
Show that a surjective ring morphism is an epimorphism. Show also that the
inclusion morphism from Z into Q is an epimorphism.
√ √ √ √
8-exo005) Let Z[ 2] and Z[ 3] be the subrings of C generated by Z, and by 2 and 3
respectively.
√ √ √ √
a) Show that Z[ 2] = {𝑎 + 𝑏 2 ; 𝑎, 𝑏 ∈ Z} and Z[ 3] = {𝑎 + 𝑏 3 ; 𝑎, 𝑏 ∈ Z}.
√
b) Show
√ that besides
√ the identity, there is only one automorphism of Z[ 2]; it maps
𝑎 + 𝑏 2 to 𝑎 − 𝑏 2 for any 𝑎, 𝑏 ∈ Z.
√ √
c) Show that there does not exist a ring morphism from Z[ 2] to Z[ 3].
√3
d) More generally, what are the automorphisms of Z[𝑖] ? of Z[ 2] ?
9-exo514) Let 𝛼 be a complex number. Assume that there exists a monic polynomial P
with integral coefficients, say P = X𝑑 + 𝑎 𝑑−1 X𝑑−1 + · · · + 𝑎 0 , such that P(𝛼) = 0. Show that
the set A of all complex numbers of the form 𝑐0 +𝑐 1 𝛼+· · ·+𝑐 𝑑−1 𝛼 𝑑−1 , for 𝑐0 , . . . , 𝑐 𝑑−1 ∈ Z,
is a subring of C. Give an example that shows that the result does not hold when P is
not monic.
10-exo539) Let K be a field and V be a K-vector space.
V𝑖 . For 𝑥 = 𝑥 𝑖 , with
É Í
a) Let (V𝑖 ) be any family of subspaces of V such that V =
𝑥 𝑖 ∈ V𝑖 , set 𝑝 𝑗 (𝑥) = 𝑥 𝑗 . Show
É that for any 𝑗, 𝑝 𝑗 is a projector in V (namely, 𝑝 𝑗 ◦ 𝑝 𝑗 = 𝑝)
Í 𝑗
with image V 𝑗 and kernel 𝑖≠𝑗 𝑖
V . Show that 𝑝 𝑗 ◦ 𝑝 𝑖 = 0 for 𝑖 ≠ 𝑗, and that id V = 𝑝𝑖 .
b) Conversely, let (𝑝 𝑖 ) be a family of projectors in V such that 𝑝 𝑖 ◦ 𝑝 𝑗 = 0 for 𝑖 ≠ 𝑗, and
such that idV = 𝑝 𝑖 . Let V𝑖 be the image of 𝑝 𝑖 . Show that V is the direct sum of the
Í
subspaces V𝑖 , and that 𝑝 𝑖 is the projector onto V𝑖 whose kernel is the sum of all V 𝑗 , for
𝑗 ≠ 𝑖.
11-exo540) Automorphisms of M𝑛 (C). Let A = M𝑛 (C) be the ring of 𝑛 × 𝑛 matrices with
complex coefficients and let 𝜑 be any automorphism of A. Let Z be the center of A; it
consists in all scalar matrices 𝑎I𝑛 , for 𝑎 ∈ C.
a) Show that 𝜑 induces, by restriction, an automorphism of Z.
In the sequel, we assume that 𝜑| Z = idZ . For 1 6 𝑖, 𝑗 6 𝑛, let E𝑖,𝑗 be the elementary
matrix whose only nonzero coefficient is a 1 at place (𝑖, 𝑗); let B𝑖,𝑗 = 𝜑(E𝑖,𝑗 ).
52 CHAPTER 1. RINGS

b) Show that B𝑖,𝑖 is the matrix of a projector 𝑝 𝑖 of C𝑛 . Show that 𝑝 𝑖 ◦ 𝑝 𝑗 = 0 for 𝑖 ≠ 𝑗,

and that idC𝑛 = 𝑝 𝑖 .
Í

c) Using exercise 1/10, show that there exists a basis ( 𝑓1 , . . . , 𝑓𝑛 ) of C𝑛 such that for
any 𝑖, 𝑝 𝑖 be the projector onto C 𝑓𝑖 with kernel the sum 𝑗≠𝑖 C 𝑓 𝑗 .
Í

d) Show that there exist elements 𝜆 𝑖 ∈ C∗ such that, denoting 𝑒 𝑖 = 𝜆 𝑖 𝑓𝑖 , one has
B𝑖𝑗 (𝑒 𝑘 ) = 0 for 𝑘 ≠ 𝑗, and B𝑖𝑗 (𝑒 𝑗 ) = 𝑒 𝑖 . Deduce the existence of a matrix B ∈ GL𝑛 (C) such
that 𝜑(M) = BMB−1 for any matrix M in M𝑛 (C).
e) What happens when one does not assume anymore that 𝜑 restricts to the identity
on Z?
12-exo512) Let 𝑛 be an integer such that 𝑛 > 1.
a) What are the invertible elements of Z/𝑛Z? For which integers 𝑛 is that ring a
domain? a field?
b) Let 𝑚 be an integer such that 𝑚 > 1. Show that the canonical map from Z/𝑛𝑚Z
to Z/𝑛Z is a ring morphism. Show that it induces a surjection from (Z/𝑚𝑛Z)× onto
(Z/𝑛Z)× .
c) Determine the nilpotent elements of the ring Z/𝑛Z.
13-exo541) Give an example of a ring morphism 𝑓 : A → B which is surjective but such
that the associated group morphism from A× to B× is not surjective.
14-exo504) a) Let K be a field, let V be a K-vector space and let A = EndK (V) be the
ring of endomorphisms of V.
Show that the elements of A which are right invertible are the surjective endomor-
phisms, those who are left invertible are the injective endomorphisms.
b) Give an example of a noncommutative ring and of an element which possesses
infinitely many right inverses.
15-exo508) Product ring. Let A and B be two rings. The set A × B is endowed with the
addition defined by (𝑎, 𝑏) + (𝑎 0 , 𝑏 0) = (𝑎 + 𝑎 0 , 𝑏 + 𝑏 0) and the multiplication defined by
(𝑎, 𝑏) · (𝑎 0 , 𝑏 0) = (𝑎𝑎 0 , 𝑏𝑏 0), for 𝑎 and 𝑎 0 ∈ A, 𝑏 and 𝑏 0 ∈ B,
a) Show that A × B is a ring. What is the neutral element for multiplication?
b) Determine the elements of A × B which are respectively right regular, left regular,
right invertible, left invertible, nilpotent.
c) Show that the elements 𝑒 = (1, 0) and 𝑓 = (0, 1) of A × B satisfy 𝑒 2 = 𝑒 and 𝑓 2 = 𝑓 .
One says that they are idempotents.
16-exo506) Let A be a ring such that A ≠ 0.
a) Let 𝑎 be an element of A which has exactly one right inverse. Show that 𝑎 is
invertible. (First prove that 𝑎 is regular.)
b) If every nonzero element of A is left invertible, then A is a division ring.
c) One assumes that A is finite. Show that any left regular element is right invertible.
If any element of A is left regular, then A is a division ring. When A is commutative,
any prime ideal of A is a maximal ideal.
 EXERCISES 53

d) Same question when K is a field and A is a K-algebra which is finite-dimensional

as a K-vector space. (In particular, we impose that the multiplication of A is K-bilinear.)
e) Assume that any nonzero element of A is left regular and that the set of all right
ideals of A is finite. Show that A is a division ring. (To show that a nonzero element 𝑥 is
rignt inversible, introduce the right ideals 𝑥 𝑛 A, for 𝑛 > 1.)
17-exo509) Let A be a ring and let 𝑒 ∈ A be any idempotent element of A (this means,
we recall, that 𝑒 2 = 𝑒).
a) Show that 1 − 𝑒 is an idempotent element of A.
b) Show that 𝑒A𝑒 = {𝑒 𝑎𝑒 ; 𝑎 ∈ A} is an abelian subgroup of A, and that the
multiplication of A endowes it with the structure of a ring. (However, it is not a subring
of A, in general.)
c) Explicit the particular case where A = M𝑛 (𝑘), for some field 𝑘, the element 𝑒 being
a diagonal matrix of rank 𝑟.
18-exo505) Let A be a ring.
a) Let 𝑎 ∈ A be a nilpotent element. Let 𝑛 > 0 be such that 𝑎 𝑛+1 = 0; compute
(1 + 𝑎)(1 − 𝑎 + 𝑎 2 − · · · + (−1)𝑛 𝑎 𝑛 ). Deduce that 1 + 𝑎 is invertible in A.
b) Let 𝑥, 𝑦 be two elements of A; one assumes that 𝑥 is invertible, 𝑦 is nilpotent, and
𝑥 𝑦 = 𝑦𝑥. Show that 𝑥 + 𝑦 is invertible.
c) Let 𝑥 and 𝑦 be nilpotent elements of A which commute. Show that 𝑥 + 𝑦 is
nilpotent. (Let 𝑛 and 𝑚 be two integers > 1 such that 𝑥 𝑛 = 𝑦 𝑚 = 0; use the binomial
formula to compute (𝑥 + 𝑦)𝑛+𝑚−1 .)
19-exo511) Let A be a ring, let 𝑎 and 𝑏 be elements of A such that 1 − 𝑎𝑏 is invertible
in A.
a) Show that 1 − 𝑏𝑎 is invertible in A and give an explicit formula for its inverse. (One
may begin by assuming that 𝑎𝑏 is nilpotent and give a formula for the inverse of 1 − 𝑏𝑎
in terms of (1 − 𝑎𝑏)−1 , 𝑎 and 𝑏. Then show that this formula works in the general case.)
b) One assumes that 𝑘 is a field and A = M𝑛 (𝑘). Show that 𝑎𝑏 and 𝑏𝑎 have the same
characteristic polynomial. What is the relation with the preceding question?
20-exo033) Let A be a commutative ring, 𝑎 0 , . . . , 𝑎 𝑛 ∈ A and let 𝑓 be the polynomial
𝑓 = 𝑎 0 + 𝑎 1 X + · · · + 𝑎 𝑛 X𝑛 ∈ A[X].
a) Show that 𝑓 is nilpotent if and only if all 𝑎 𝑖 are nilpotent.
b) Show that 𝑓 is invertible in A[X] if and only if 𝑎 0 is invertible in A and 𝑎 1 , . . . , 𝑎 𝑛 are
nilpotent. (If 𝑔 = 𝑓 −1 = 𝑏0 + 𝑏1 X + · · · + 𝑏 𝑚 X𝑚 , show by induction on 𝑘 that 𝑎 𝑛𝑘+1 𝑏 𝑚−𝑘 = 0.)
c) Show that 𝑓 is a zero-divisor if and only if there exists a nonzero element 𝑎 ∈ A
such that 𝑎 𝑓 = 0. (Let 𝑔 be a nonzero polynomial of minimal degree such that 𝑓 𝑔 = 0; show
that 𝑎 𝑘 𝑔 = 0 for every 𝑘.)
21-exo521) Let A be a commutative ring.
a) Use the universal property of polynomial rings to show that there exists a unique
morphism of A-algebras 𝜑 : A[X, Y] → (A[X])[Y] such that 𝜑(X) = X and 𝜑(Y) = Y.
Prove that 𝜑 is an isomorphism.
54 CHAPTER 1. RINGS

b) More generally, let I be a set, let J be a subset of I and let K = I J. Show that
there exists a unique morphism of A-algebras 𝜑 : A[(X𝑖 )𝑖∈I ] → (A[(X 𝑗 ) 𝑗∈J ])[(X 𝑘 ) 𝑘∈K ],
and that this morphism is an isomorphism (see remark 1.3.10).
22-exo526) Let A be a commutative ring, let P and Q be polynomials with coefficients
in A in one indeterminate X. Let 𝑚 = deg(P) and 𝑛 = deg(Q), let 𝑎 be the leading
coefficient of Q and 𝜇 = sup(1 + 𝑚 − 𝑛, 0). Show that there exists a pair (R, S) of
polynomials such that 𝑎 𝜇 P = QR + S and deg S < 𝑛. Show that this pair is unique if A
is a domain, or if 𝑎 is regular.
23-exo052) Let 𝑘 be a field and let A = 𝑘[X1 , . . . , X𝑛 ] be the ring of polynomials with
coefficients in 𝑘 in 𝑛 indeterminates. One says that an ideal of A is monomial if it is
generated by monomials.
a) Let (M𝛼 )𝛼∈E be a family of monomials and let I be the ideal they generate. Show
that a monomial M belongs to I if and only if it is a multiple of one of the monomials M𝛼 .
b) Let I be an ideal of A. Show that I is a monomial ideal if and only if, for any
polynomial P ∈ I, each monomial of P belongs to I.
c) Let I and J be monomial√ ideals of A. Show that the ideals I + J, IJ, I ∩ J,
I : J = {𝑎 ∈ A ; 𝑎J ⊂ I} and I are again monomial ideals. Given monomials which
generate I and J, explicit monomials which generate those ideals.
24-exo071) Let 𝑘 be a field, let (M𝛼 )𝛼∈E be a family of monomials of A = 𝑘[X1 , . . . , X𝑛 ],
and let I be the ideal they generate. The goal of this exercise is to show that there
exists a finite subset F ⊂ E such that I = (M𝛼 )𝛼∈F . The proof runs by induction on the
number 𝑛 of indeterminates.
a) Treat the case 𝑛 = 1.
b) In the sequel, one assumes that 𝑛 > 2 and that the property holds if there are
strictly less than 𝑛 indeterminates.
For 𝑖 ∈ {1, . . . , 𝑛}, one defines a morphism of rings 𝜑 𝑖 by
𝜑 𝑖 (P) = P(X1 , . . . , X𝑖−1 , 1, X𝑖+1 , . . . , X𝑛 ).
Using the induction hypothesis, observe that there exists a finite subset F𝑖 ⊂ E such
that for any 𝛼 ∈ E the monomial 𝜑 𝑖 (M𝛼 ) can be written 𝜑 𝑖 (M𝛼 ) = M0𝛼 × 𝜑 𝑖 (M𝛽 ) for
some 𝛽 ∈ F𝑖 .
c) Let F0 be the set of all 𝛼 ∈ E such that for all 𝑖 ∈ {1, . . . , 𝑛}, one has
degX𝑖 (M𝛼 ) < sup{degX𝑖 (M𝛽 ) ; 𝛽 ∈ F𝑖 } .
Ð𝑛
Set F = 𝑖=0 F𝑖 . Show that I = (M𝛼 )𝛼∈F .
25-exo522) Let A be a ring, let I be a set and let M be the set N(I) of all multi-indices
on I. Let ℱI = AM be the set of all families of elements of A indexed by M. It is an
abelian group for termwise addition.
a) Show that the formulae that define the multiplication of polynomials make sense
on ℱI and endow it with the structure of a ring, of which the ring of polynomials 𝒫I is
a subring.
 EXERCISES 55

Let X𝑖 be the indeterminate with index 𝑖. Any element of ℱI is formally written as

an infinite sum 𝑚 𝑎 𝑚 X1𝑚1 . . . X𝑚 𝑛
Í
𝑛 . The ring ℱI is called the ring of power series in the
indeterminates (X𝑖 )𝑖∈I . When I = {1, . . . , 𝑛}, it is denoted A[[X1 , . . . , X𝑛 ]], and A[[X]]
when I has a single element, the indeterminate being written X.
Let 𝑘 be a commutative ring.
b) For any 𝑘-algebra A and any family (𝑎 1 , . . . , 𝑎 𝑛 ) of nilpotent elements which com-
mute pairwise, show that there is a unique morphism of 𝑘-algebras 𝜑 : 𝑘[[X1 , . . . , X𝑛 ]] →
A such that 𝜑(X𝑖 ) = 𝑎 𝑖 for all 𝑖.
𝑎 𝑛 X𝑛 of 𝑘[[X]] is invertible if and only if 𝑎 0 is invertible
Í
c) Show that an element
in 𝑘.
26-exo619) Let 𝑘 be a field. Let Φ be the morphism from 𝑘[X1 , . . . , X𝑛 ] to ℱ(𝑘 𝑛 , 𝑘) that
maps a polynomial to the corresponding polynomial function.
a) Let A1 , . . . , A𝑛 be subsets of 𝑘. Let P ∈ 𝑘[X1 , . . . , X𝑛 ] be a polynomial in 𝑛 indeter-
minates such that degX𝑖 (P) < Card(A𝑖 ) for every 𝑖. One assumes that P(𝑎 1 , . . . , 𝑎 𝑛 ) = 0
for every (𝑎 1 , . . . , 𝑎 𝑛 ) ∈ A1 × · · · × A𝑛 . Show that P = 0.
b) If 𝑘 is an infinite field, show that Φ is injective but not surjective.
c) One assumes that 𝑘 is a finite field. Show that Φ is surjective. For any function
𝑓 ∈ ℱ(𝑘 𝑛 ; 𝑘), explicit a polynomial P such that Φ(P) = 𝑓 (think of the Lagrange
interpolation polynomials). Show that Φ is not injective. More precisely, if 𝑞 = Card(𝑘),
𝑞
show that Ker(Φ) is generated by the polynomials X𝑖 − X𝑖 , for 1 6 𝑖 6 𝑛.
27-exo867) Let A be a ring.
a) Endow B = A(N) with termwise addition and with multiplication given by
(𝑎 𝑛 ) · (𝑏 𝑛 ) = (𝑐 𝑛 ), with 𝑐 𝑛 = 𝑛𝑘=0 𝑎 𝑘 𝑏 𝑛−𝑘 . Prove that B is a ring, with unit element
Í
(1, 0, . . . ). Let T = (0, 1, 0, . . . ); prove that for every integer 𝑛, T𝑛 is the sequence whose
only nonzero term is equal to 1 and has index 𝑛. Prove also that (𝑎 𝑛 ) = ∞ 𝑛
𝑛=0 𝑎 𝑛 T for
Í
every (𝑎 𝑛 ) ∈ B.
Elements of B are called noncommutative polynomials with coefficients in A and denoted
by A[T]. The degree deg(P) of an element P = 𝑎 𝑛 T𝑛 ∈ A[T] is the supremum of
Í
all 𝑛 such that 𝑎 𝑛 ≠ 0; if P ≠ 0, then deg(P) ∈ N and the coefficient 𝑎 deg(P) is called the
leading coefficient of P.
b) Let P, Q ∈ A[T]; assume that Q ≠ 0 and that the leading coefficient of Q is
equal to 1. Prove that there exists a unique pair (U, V) of elements of A[T] such that
P = QU + V and deg(V) < deg(Q).
c) Assume that Q = T− 𝑎, for some 𝑎 ∈ A. Prove that there exists a unique polynomial
U ∈ A[T] and a unique element 𝑏 ∈ A such that P = U · (T − 𝑎) + 𝑏. If P = 𝑐 𝑛 T𝑛 ,
Í
prove that 𝑏 = 𝑐 𝑛 𝑎 𝑛 .
Í

d) Let K be a commutative ring and let A = M𝑛 (K) be the ring of 𝑛 × 𝑛 matrices

with coefficients in K. Let U ∈ A and let V ∈ M𝑛 (K[T]) be the adjugate the comatrix of
U − TI𝑛 . Using the formula (U − TI𝑛 ) · V = PU (T), where PU ∈ K[T] is the characteristic
polynomial of U, prove that PU (U) = 0 (theorem of Cayley–Hamilton).
56 CHAPTER 1. RINGS

28-exo524) One says that a ring A admits a right euclidean division if there exists a
map 𝜑 : A {0} → N such that for any pair (𝑎, 𝑏) of elements of A, with 𝑏 ≠ 0, there
exists a pair (𝑞, 𝑟) of elements of A such that 𝑎 = 𝑞𝑏 + 𝑟 and such that either 𝑟 = 0 or
𝜑(𝑟) < 𝜑(𝑏).
a) Assume that A admits a right euclidean division. Show that any left ideal of A is
of the form A𝑎, for some 𝑎 ∈ A.
b) Let K be a division ring. Show that the polynomial ring K[X] admits a right
euclidean division.
29-exo528) Let A be a ring and let I be a right ideal of A.
a) Show that the left ideal generated by I in A is a two-sided ideal.
b) Show that the set J of all elements 𝑎 ∈ A such that 𝑥𝑎 = 0 for any 𝑥 ∈ I (the right
annihilator of I) is a two-sided ideal of A.
30-exo001) Let A be a commutative ring, and let I, J, L be ideals of A. Show the
following properties:
a) I · J ⊂ I ∩ J;
b) (I · J) + (I · L) = I · (J + L);
c) (I ∩ J) + (I ∩ L) ⊂ I ∩ (J + L);
d) If J ⊂ I, then J + (I ∩ L) = I ∩ (J + L);
e) Let K be a field and assume that A = K[X, Y]. Set I = (X), J = (Y) and L = (X + Y).
Compute (I ∩ J) + (I ∩ L) and I ∩ (J + L); compare these two ideals.
31-exo535) Let A be a ring.
a) Give an example where the set of all nilpotents elements of A is not an additive
subgroup of A. (One may check that A = M2 (C) works.)
b) Let N be the set of all elements 𝑎 ∈ A such that 𝑎𝑥 is nilpotent, for every 𝑥 ∈ A.
Show that N is a two-sided ideal of A, every element of which is nilpotent.
c) Let I be a two-sided ideal of A such that every element of I is nilpotent. Show that
I ⊂ N.
32-exo516) Let K be a field, let V be a K-vector space and let A be the ring of
endomorphisms of V.
a) For any subspace W of V, show that the set NW of all endomorphisms whose
kernel contains W is a left ideal of A, and that set set IW of all endomorphisms whose
image is contained in W is a right ideal of A.
b) If V has finite dimension, then all left ideals (resp. right ideals) are of this form.
c) If V has finite dimension, the only two-sided ideals of A are (0) and A.
d) The set of all endomorphisms of finite rank of V (that is, those whose image
has finite dimension) is a two-sided ideal of A. It is distinct from A if V has infinite
dimension.
33-exo519) Let A be a commutative ring and let 𝑎, 𝑏 be two elements of A.
 EXERCISES 57

a) If there exists a unit 𝑢 in A such that 𝑎 = 𝑏𝑢 (in which case one says that 𝑎 and 𝑏
are associates), show that the ideals (𝑎) = 𝑎A and (𝑏) = 𝑏A coincide.
b) Conversely, assuming that A is a domain and that (𝑎) = (𝑏), show that 𝑎 and 𝑏 are
associates.
c) Assume that A is the quotient of the ring Z[𝑎, 𝑏, 𝑥, 𝑦] of polynomials in four
indeterminates by the ideal generated by (𝑎 − 𝑏𝑥, 𝑏 − 𝑎𝑦). Prove that the ideals (𝑎)
and (𝑏) coincide but that 𝑎 and 𝑏 are not associate.
34-exo006) Let A, B be commutative rings and let 𝑓 : A → B be a ring morphism. For
any ideal I of A, let 𝑓∗ (I) be the ideal of B generated by 𝑓 (I); we say it is the extension of I
to B. For any ideal J of B, we call the ideal 𝑓 −1 (J) the contraction of J in A.
Let I be an ideal of A and let J be an ideal of B; show the following properties.
a) I ⊂ 𝑓 −1 ( 𝑓∗ (I)) and J ⊃ 𝑓∗ ( 𝑓 −1 (J));
b) 𝑓 −1 (J) = 𝑓 −1 𝑓∗ ( 𝑓 −1 (J) and 𝑓∗ (I) = 𝑓∗ 𝑓 −1 ( 𝑓∗ (I) .

Let 𝒞 be the set of ideals of A which are contractions of ideals of B, and let ℰ be the
set of ideals of B which are extensions of ideals of A.
c) Show that 𝒞 = {I; I = 𝑓 −1 𝑓∗ (I) } and ℰ = {J; J = 𝑓∗ 𝑓 −1 (I) };

d) The map 𝑓∗ defines a bijection from 𝒞 onto ℰ; what is its inverse?

Let I1 and I2 be ideals of A, let J1 and J2 be ideals of B. Show the following properties:
e) 𝑓∗ (I1 + I2 ) = 𝑓∗ (I1 ) + 𝑓∗ (I2 ) and 𝑓 −1 (J1 + J2 ) ⊃ 𝑓 −1 (J1 ) + 𝑓 −1 (J2 );
f ) 𝑓∗ (I1 ∩ I2 ) ⊂ 𝑓∗ (I1 ) ∩ 𝑓∗ (I2 ), and 𝑓 −1 (J1 ∩ J2 ) = 𝑓 −1 (J1 ) ∩ 𝑓 −1 (J2 );
g) 𝑓∗ (I1 · I2 ) = 𝑓∗ (I1 ) · 𝑓∗ (I2 ), and 𝑓 −1 (J1 · J2 ) ⊃ 𝑓 −1 (J1 ) · 𝑓 −1 (J2 );
√ √ p
h) 𝑓∗ ( I) ⊂ 𝑓∗ (I), and 𝑓 −1 ( J) = 𝑓 −1 (J).
p

35-exo184) Let I and J be two ideals of a commutative ring A. Assume that I + J = A.

Then, show that I𝑛 + J𝑛 = A for any positive integer 𝑛.
36-exo517) Let A be a commutative ring, let I be an ideal of A and let S be any subset
of A. The conductor of S in I is defined by the formula
(I : S) = {𝑎 ∈ A ; for all 𝑠 ∈ S, 𝑎𝑠 ∈ I}.
Show that it is an ideal of A, more precisely, the largest ideal J of A such that JS ⊂ I.
37-exo014) Let K be a field and let A = K[X, Y]/(X2 , XY, Y2 ).
a) Compute the invertible elements of A.
b) Determine all principal ideals of A.
c) Determine all ideals of A.
38-exo534) Let A be a ring and let I be the two-sided ideal of A generated by the
elements of the form 𝑥𝑦 − 𝑦𝑥, for 𝑥 and 𝑦 ∈ A.
a) Show that the ring A/I is commutative.
b) Let J be a two-sided ideal of A such that A/J is a commutative ring. Show that
I ⊂ J.
39-exo568) Let A be a ring, let I be a right ideal of A.
58 CHAPTER 1. RINGS

a) Show that the set B of all 𝑎 ∈ A such that 𝑎I ⊂ I is a subring of A. Show that I is a
two-sided ideal of B.
b) Define an isomorphism from the ring EndA (A/I), where A/I is a right A-module,
onto the ring B/I.
c) Let I be a maximal right ideal of A. Show that B/I is a division ring.
40-exo536) Let A be a ring, let I be an ideal of A. We let I[X] be the set of polynomials
P ∈ A[X] all of which coefficients belong to I.
a) Show that I[X] is a left ideal of A[X].
b) If I is a two-sided ideal of A, show that I[X] is a two-sided ideal of A[X] and
construct an isomorphism from the ring A[X]/I[X] to the ring (A/I)[X].
41-exo537) Let A be a ring, let I be a two-sided ideal of A and let M𝑛 (I) be the set of
matrices in M𝑛 (A) all of which coefficients belong to I.
a) Show that M𝑛 (I) is a two-sided ideal of M𝑛 (A) and construct an isomorphism of
rings from M𝑛 (A)/M𝑛 (I) onto M𝑛 (A/I).
b) Conversely, show that any two-sided ideal of M𝑛 (A) is of the form M𝑛 (I), for some
two-sided ideal I of A.
42-exo544) a) What are the invertible elements of the ring of decimal numbers?
Let A be a commutative ring and let S be a multiplicative subset of A.
b) Show that an element 𝑎 ∈ A is invertible in S−1 A if and only if there exists 𝑏 ∈ A
such that 𝑎𝑏 ∈ S.
c) Let T be a multiplicative subset of A which contains S. Construct a ring morphism
from S−1 A to T−1 A.
d) Let S˜ be the set of elements of A whose image in S−1 A is invertible. Show that S˜
contains S and that the ring morphism from S−1 A to S˜ −1 A is an isomorphism.
Give an explicit proof of this fact, as well as a proof relying only on the universal
property.
43-exo559) Let A be a commutative ring.
a) Let 𝑠 be an element of A and let S = {1; 𝑠; 𝑠 2 ; . . .} be the multiplicative subset
generated by 𝑠. We write A𝑠 for the ring of fractions S−1 A. Show that the following
properties are equivalent:
(i) The canonical morphism 𝑖 : A → A𝑠 is surjective;
(ii) The decreasing sequence of ideals (𝑠 𝑛 A)𝑛 is ultimately constant;
(iii) For any large enough integer 𝑛, the ideal 𝑠 𝑛 A is generated by an idempotent
element.
(To show that (ii) implies (iii), show by induction on 𝑘 that a relation of the form 𝑠 𝑛 = 𝑠 𝑛+1 𝑎
implies that 𝑠 𝑛 = 𝑠 𝑛+𝑘 𝑎 𝑘 ; then conclude that 𝑠 𝑛 𝑎 𝑛 is an idepotent element.)
b) Let S be a multiplicative subset of A consisting of elements 𝑠 for which the
morphism A → A𝑠 is surjective. Show that the morphism A → S−1 A is surjective too.
 EXERCISES 59

c) Let A be a ring which is, either finite, or which is a finite dimensional vector
subspace over a subfield (more generally, an artinian ring). Show that condition (ii)
holds for any element 𝑠 ∈ A.
44-exo543) a) Let A be a subring of Q. Show that there exists a multiplicative subset S
of Z such that A = S−1 Z.
b) Let A = C[X, Y] be the ring of polynomials with complex coefficients in two
indeterminates X and Y. Let B = A[Y/X] be the subring generated by A and Y/X in the
field C(X, Y) of rational functions.
Show that the unique ring morphism from C[T, U] to B that maps T to X and U to Y/X
is an isomorphism. Deduce that A× = B× = C× , hence that B is not the localization
of A with respect to some multiplicative subset.
45-exo538) Let A be a (not necessarily commutative) ring and let S be a multiplicative
subset of A.
Let AS be a ring and let 𝑖 : A → AS be a morphism of rings. One says that (AS , 𝑖) is a
ring of right fractions for S if the following properties hold:
(i) For any 𝑠 ∈ S, 𝑖(𝑠) is invertible in AS ;
(ii) Any element of AS is of the form 𝑖(𝑎)𝑖(𝑠)−1 for some 𝑎 ∈ A and 𝑠 ∈ S.
a) Assume that A admits a ring of right fractions for S. Show the following properties
(right Ore conditions):
(i) For any 𝑎 ∈ A and any 𝑠 ∈ S, there exists 𝑏 ∈ A and 𝑡 ∈ S such that 𝑎𝑡 = 𝑠𝑏;
(ii) For any 𝑎 ∈ A and any 𝑠 ∈ S such that 𝑠𝑎 = 0, there exists 𝑡 ∈ S such that
𝑎𝑡 = 0.
(The second condition is obviously satisfied if every element of S is regular.)
b) Conversely, assume that the right Ore conditions hold. Define an equivalence
relation ∼ on the set A × S by “(𝑎, 𝑠) ∼ (𝑏, 𝑡) if and only if there exist 𝑐, 𝑑 ∈ A and
𝑢 ∈ S such that 𝑢 = 𝑠𝑐 = 𝑡𝑑 and 𝑎𝑐 = 𝑏𝑑.” Construct a ring structure on the quotient
set AS such that the map 𝑖 : A → AS sending 𝑎 ∈ A to the equivalence class of (𝑎, 1) is
ring morphism, and such that for any 𝑠 ∈ S, 𝑖(𝑠) be invertible in AS with inverse the
equivalence class of (1, 𝑠). Conclude that AS is a ring of right fractions for S. If every
element of S is regular, prove that 𝑖 is injective.
46-exo560) Let 𝑎 and 𝑏 be positive integers.
a) Show that there exist integers 𝑚 and 𝑛 such that 𝑚 is prime to 𝑏, each prime
divisor of 𝑛 divides 𝑏, and 𝑎 = 𝑚𝑛. (Beware, 𝑛 is not the gcd of 𝑎 and 𝑏.)
b) Show that the ring (Z/𝑎Z)𝑏 is isomorphic to Z/𝑛Z. To that aim, construct explicitly
a ring morphism from Z/𝑛Z to (Z/𝑎Z)𝑏 and show that it is an isomorphism.
47-exo561) a) Show that the ring Z[𝑖] is isomorphic to the ring Z[X]/(X2 + 1).
b) Let 𝑎 be an integer. Considering the ring Z[𝑖]/(𝑎 + 𝑖) as a quotient of the ring of
polynomials Z[X], define an isomorphism
∼
Z[𝑖]/(𝑎 + 𝑖) →
− Z/(𝑎 2 + 1)Z.
60 CHAPTER 1. RINGS

c) More generally, let 𝑎 and 𝑏 be two coprime integers. Show that the image of 𝑏 in
Z[𝑖]/(𝑎 + 𝑖𝑏) is invertible. Write this ring as a quotient of Z𝑏 [X] and then define an
isomorphism
∼
Z[𝑖]/(𝑎 + 𝑖𝑏) →
− Z/(𝑎 2 + 𝑏 2 )Z.
(Observe that if 1 = 𝑎𝑢 + 𝑏𝑣, then 1 = (𝑎 + 𝑏𝑖)𝑢 + 𝑏(𝑣 − 𝑢𝑖).)
48-exo531) Let A be a commutative ring, let S be a multiplicative subset of A.
a) One assumes that there are elements 𝑠 and 𝑡 ∈ S such that S be the set of all 𝑠 𝑛 𝑡 𝑚 ,
for 𝑛 and 𝑚 in N. Show that the morphism A[X, Y] → S−1 A, P(X, Y) ↦→ P(1/𝑠, 1/𝑡)
is surjective and that its kernel contains the ideal (1 − 𝑠X, 1 − 𝑡Y) generated by 1 − 𝑠X
and 1 − 𝑡Y in A[X, Y]. Construct an isomorphism A[X, Y]/(1 − 𝑠X, 1 − 𝑡Y) ' S−1 A.
b) More generally, assume that S is the smallest multiplicative subset of A containing
a subset T, and let h1 − 𝑡X𝑡 i𝑡∈T the ideal of the polynomial ring (with a possibly infinite
set of indeterminates X𝑡 , for 𝑡 ∈ T) generated by the polynomials 1 − 𝑡X𝑡 , for 𝑡 ∈ T.
Then show that the canonical morphism
A[(X𝑡 )𝑡∈T ] → S−1 A, P ↦→ P((1/𝑡)𝑡 )
induces an isomorphism
A[(X𝑡 )𝑡∈T ]/h1 − 𝑡X𝑡 i𝑡∈T ' S−1 A.
49-exo091) Let A be a commutative ring and let S be a multiplicative subset of A such
that 0 ∉ S.
a) If A is a domain, show that S−1 A is a domain.
b) If A is reduced, show that S−1 A is reduced.
c) Let 𝑓 : A → S−1 A be the canonical morphism 𝑎 ↦→ 𝑎/1. Show that the nilradical
of S−1 A is the ideal of S−1 A generated by the image of the nilradical of A.
50-exo532) Let B ⊂ R(X) be the set of all rational functions with real coefficients of the
form P/(X2 + 1)𝑛 , where P ∈ R[X] is a polynomial and 𝑛 is an integer. Let A be the
subset of B consisting of those fractions P/(X2 + 1)𝑛 for which deg(P) 6 2𝑛.
a) Show that A and B are subrings of R(X).
b) Determine their invertible elements.
c) Show that B is a principal ideal domain. Show that the ideal of A generated by
1/(X2 + 1) and X/(X2 + 1) is not principal.
CHAPTER 2

IDEALS AND DIVISIBILITY

In this second chapter, we study deeper aspects of divisibility in rings. This

question appeared historically when mathematicians — notably, in the hope of
proving Fermat’s Last Theorem! — tried to make use of unique factorization in
rings where it didn’t hold. Ideals are one device that was invented then so to
have a better understanding of divisibility. In this context, there are two natural
analogues of prime numbers, namely maximal and prime ideals.
I introduce maximal ideals in the general framework of possibly noncommuta-
tive rings; I then define the Jacobson radical. Matrix rings furnish interesting
examples.
However, from that point on, the chapter is essentially focused on commutative
rings. I define prime ideals and show their relations with nilpotent elements (via
the nilpotent radical), or local rings. The set of all prime ideals of a commutative
ring is called its spectrum, and it is endowed with a natural topology. The
detailed study of spectra of rings belongs to the algebraic geometry of schemes,
as put forward by Alexander Grothendieck. Although that is a more advanced
subject than what is planned for this book, I believe that the early introduction
of the spectrum of a ring allows to add an insightful geometric interpretation to
some constructions of commutative algebra.
Hilbert’s “Nullstellensatz” describes the maximal ideals of the ring of
polynomials in finitely many indeterminates over an algebraically closed field.
After proving it in the particular case of the field of complex numbers (by a short
argument which some colleagues find too special), I show how this theorem
gives rise to a correspondence between geometry and algebra. (The general case
of the theorem will be proved in chapter 9; a few other proofs also appear as
exercises in that chapter, as well as in this one.)
62 CHAPTER 2. IDEALS AND DIVISIBILITY

Another beautiful geometric example is given by Gelfand’s theorem that

describes the maximal ideals of the ring of continuous functions on a compact
metric space.
I then go back to the initial goal of understanding uniqueness of a decom-
position into prime numbers in general rings, and introduce three classes of
commutative rings of increasing generality in which the divisibility relation
is particularly well behaved: euclidean rings, principal ideal domains and
unique factorization domains. In particular, I prove a theorem of Gauss that
asserts that polynomial rings are unique factorization domains.
I conclude the chapter by introducing the resultant of two polynomials and
use it to prove a theorem of Bézout about the number of common roots to two
polynomials in two indeterminates — geometrically, this gives an upper bound
for the number of intersection points of two plane curves.

2.1. Maximal ideals

Definition (2.1.1). — Let A be a ring. One says that a left ideal of A is
maximal if it maximal among the left ideals of A which are distinct from A.
In other words, a left ideal I is maximal if one has I ≠ A and if the only
left ideals of A containing I are A and I. Consequently, to check that a
left ideal I ≠ A is maximal, it suffices to show that for any 𝑎 ∈ A I, the
left ideal I + A𝑎 is equal to A.
There is an analogous definition for the right and two-sided ideals.
These notions coincide when the ring A is commutative, but the reader
shall be cautious in the general case: When the ring A is not commutative,
a two-sided ideal can be a maximal ideal as a two-sided ideal but not as
a left ideal, or be a maximal ideal as a left ideal but not as a right ideal.
Examples (2.1.2). — a) The ideals of Z have the form 𝑛Z, for some
𝑛 ∈ Z (example 1.4.3). If 𝑛 divides 𝑚, then 𝑚Z ⊂ 𝑛Z, and conversely.
It follows that the maximal ideals of Z are the ideals 𝑝Z, where 𝑝 is a
prime number.
Similarly, if K is a field, the maximal ideals of the ring K[X] of polyno-
mials in one indeterminate are the ideals generated by an irreducible
polynomial. When K is algebraically closed, the maximal ideals are the
ideals (X − 𝑎), for some 𝑎 ∈ K.
 2.1. MAXIMAL IDEALS 63

b) Let K be a field and let V be a K-vector space of finite dimension.

In exercise 1/32, we asked to determine the left, right, and two-sided
ideals of EndK (V).
The left ideals of End(V) are the ideals IW , where W is a vector subspace
of V and IW is the set of endomorphisms whose kernel contains W. For
W ⊂ W0, IW0 ⊂ IW . Consequently, the maximal left ideals of End(V) are
those ideals IW , where W is a line in V.
The right ideals of V are the ideals RW where W is a vector subspace
of V and RW is the set of endomorphisms whose image is contained
in W. For W ⊂ W0, IW ⊂ IW0 . Consequently, the maximal right ideals
of End(V) are those ideals RW , where W is a hyperplane in V.
The only two-sided ideals of End(V) are (0) and End(V), so that (0) is
the only maximal two-sided ideal of End(V).
c) A maximal left ideal of a ring is a maximal right ideal of the opposite
ring.
The following general theorem asserts the existence of maximal ideals
of a given ring; it is a consequence of Zorn’s lemma (theorem A.2.11).
Theorem (2.1.3) (Krull). — Let A be a ring and let I be a left ideal of A, distinct
from A. There exists a maximal left ideal of A which contains I.
In particular, any nonzero ring possesses at least one maximal left
ideal (take I = 0).
The analogous statements for right ideals and two-sided ideals are
true, and proven in the same way.
Proof. — Let ℐ be the set of left ideals of A which contain I and are
distinct from A. Let us endow ℐ with the ordering given by inclusion.
Let us show that ℐ is inductive. Let indeed (J𝑖 ) be a totally ordered
family of left ideals of A such that I ⊂ J𝑖 ( A for any 𝑖; let us show that it
has an upper bound J in ℐ.
If this family is empty, we set J = I.
Otherwise, let J be the union of the ideals J𝑖 . Let us show that J ∈ ℐ.
By construction, J contains I, because there is an index 𝑖, and J ⊃ J𝑖 ⊃ I.
Since 1 does not belong to J𝑖 , for any index 𝑖, we deduce that 1 ∉ J and
J ≠ A. Finally, we prove that J is a left ideal of A. Let 𝑥, 𝑦 ∈ J; there are
indices 𝑖 and 𝑗 such that 𝑥 ∈ J𝑖 and 𝑦 ∈ J 𝑗 . Since the family (J𝑖 ) is totally
64 CHAPTER 2. IDEALS AND DIVISIBILITY

ordered, we have J𝑖 ⊂ J 𝑗 or J 𝑗 ⊂ J𝑖 . In the first case, 𝑥 + 𝑦 ∈ J 𝑗 , in the

second, 𝑥 + 𝑦 ∈ J𝑖 ; in any case, 𝑥 + 𝑦 ∈ J. Finally, if 𝑥 ∈ J and 𝑎 ∈ A, let 𝑖
be such that 𝑥 ∈ J𝑖 ; since J𝑖 is a left ideal of A, 𝑎𝑥 ∈ J𝑖 , hence 𝑎𝑥 ∈ J.
By Zorn’s lemma (theorem A.2.11), the set ℐ possesses a maximal
element J. By definition of the ordering on ℐ, J is a left ideal of A,
distinct from A, which contains I, and which is maximal for that property.
Consequently, J is a maximal left ideal of A containing I, hence Krull’s
theorem. 

Corollary (2.1.4). — Let A be a ring. For an element of A to be left invertible

(resp. right invertible), it is necessary and sufficient that it belongs to no
maximal left ideal (resp. to no maximal right ideal) of A.

Proof. — Let 𝑎 be an element of A. That 𝑎 is left invertible means that

the left ideal A𝑎 equals A; then no maximal left ideal of A can contain 𝑎.
Otherwise, if A𝑎 ≠ A, there exists a maximal left ideal of A containing
the left ideal A𝑎, and that ideal contains 𝑎. 

Definition (2.1.5). — Let A be a ring. The Jacobson radical of A is the

intersection of all maximal left ideals of A.

Lemma (2.1.6). — Let A be a ring and let J be its Jacobson radical. Let 𝑎 ∈ A.
The following properties are equivalent:
(i) One has 𝑎 ∈ J;
(ii) For every 𝑥 ∈ A, 1 + 𝑥𝑎 is left invertible;
(iii) For every 𝑥, 𝑦 ∈ A, 1 + 𝑥𝑎𝑦 is invertible.

Proof. — We may assume that A ≠ 0, the lemma being trivial otherwise.

(i)⇒(ii). Let 𝑎 ∈ J, let 𝑥 ∈ A and let us show that 1 + 𝑥𝑎 is left invertible.
Otherwise, by Krull’s theorem 2.1.3, there exists a maximal left ideal I
of A such that 1 + 𝑥𝑎 ∈ I. Since 𝑎 ∈ J, one has 𝑎 ∈ I and 𝑥𝑎 ∈ I; this
implies that 1 ∈ I, a contradiction.
(ii)⇒(i). Let 𝑎 ∈ A be such that 1 + 𝑥𝑎 is left invertible, for every 𝑥 ∈ A.
Let I be a maximal left ideal of A and let us prove that 𝑎 ∈ I. Otherwise,
one has I + A𝑎 = A, by the definition of a maximal left ideal, hence there
exists 𝑏 ∈ I and 𝑥 ∈ A such that 𝑏 + 𝑥𝑎 = 1. Then 𝑏 = 1 − 𝑥𝑎 is left
 2.1. MAXIMAL IDEALS 65

invertible, by assumption, contradicting the fact that 𝑏 ∈ I. This proves

that 𝑎 ∈ J.
(ii)⇒(iii). Let 𝑎 ∈ A be such that 1 + 𝑥𝑎 is left invertible, for every
𝑥 ∈ A.
Let us first prove that 𝑎𝑦 ∈ J, for every 𝑦 ∈ A. Let I be a maximal left
ideal of A; assume that 𝑎𝑦 ∉ I. Then A = I + A𝑎𝑦 by the definition of a
maximal left ideal. In particular, 𝑦 ∈ I + A𝑎𝑦, so that there exists 𝑏 ∈ I
and 𝑥 ∈ A such that 𝑦 = 𝑏 + 𝑥𝑎𝑦, hence (1 − 𝑥𝑎)𝑦 = 𝑏 ∈ I. Since 1 − 𝑥𝑎 is
left invertible, this implies that 𝑦 ∈ I, hence 𝑎𝑦 ∈ I, a contradiction. In
particular, 𝑎𝑦 ∈ I. This proves that 𝑎𝑦 ∈ J.
The implication (iii)⇒(ii) being obvious, this concludes the proof of
the lemma. 

Corollary (2.1.7). — Let A be a ring and let J be its Jacobson radical.

a) The ideal J is the intersection of all right maximal ideals of A, and is a
two-sided ideal.
b) An element 𝑎 ∈ A is invertible (resp. left invertible, right invertible) if
and only if the same holds of its image in A/J.
Proof. — a) Since characterization (iii) of the Jacobson radical in
lemma 2.1.6 is symmetric, it implies that J is equal to the Jacobson
radical of the opposite ring Ao , that is, the intersection of all maximal
right ideals of A. In particular, J is both a left ideal and a right ideal,
hence a two-sided ideal.
b) Let 𝑎 ∈ A. If 𝑎 is left invertible, then so is its image in A/J. So let
us assume that the image of 𝑎 in A/J is left invertible. Let 𝑏 ∈ A be any
element whose image modulo J is a left inverse of 𝑎. One has 𝑏𝑎 ∈ 1 + J.
By lemma 2.1.6, 𝑏𝑎 is left invertible, so that 𝑎 is left invertible as well.
By symmetry, 𝑎 is right invertible if and only if its image in A/J is right
invertible. Consequently, 𝑎 is invertible if and only if its image in A/J is
invertible. 

Definition (2.1.8). — Let A be a ring and let J be its Jacobson radical. One
says that the ring A is local if the quotient ring A/J is a division ring.
Proposition (2.1.9). — Let A be a ring, let J be its Jacobson radical. The
following properties are equivalent:
66 CHAPTER 2. IDEALS AND DIVISIBILITY

(i) The ring A is local;

(ii) The ring A admits a unique maximal left ideal ;
(ii0) The ring A admits a unique maximal right ideal ;
(iii) The ideal J is a maximal left ideal of A;
(iii0) The ideal J is a maximal right ideal of A;
(iv) The set of noninvertible elements of A is an additive subgroup of A.

Proof. — If A = 0, then A is not local, it has no maximal left ideal, no

maximal right ideal, and every element of A is invertible; this proves
the equivalences of the given properties in this case. Let us now assume
that A ≠ 0.
Assume that A is a local ring. Then 0 is a the unique maximal left (resp.
right) ideal of the division ring A/J, so that J is the unique maximal left
(resp. ideal) of A. Moreover, every nonzero element of A/J is invertible,
so that A J is the set of invertible elements of A, by corollary 2.1.7. This
proves that (i) implies all other assertions (ii)–(iv).
Let us now assume (ii) that A admits a unique maximal left ideal; then
it is equal to this left ideal J, so that J is the unique maximal left ideal
of A. This shows the equivalence of (ii) and (iii). Similarly, (ii0) and (iii0)
are equivalent.
Let us assume (iii). No element of J is left invertible; conversely, an
element of A J belongs to no left maximal ideal, hence is left invertible.
Let 𝑎 ∈ A J; let 𝑥 ∈ A be such that 𝑥𝑎 = 1. Since J is a right ideal, one
has 𝑥 ∉ J, hence 𝑥 is left invertible. Then 𝑥 is both left and right invertible,
hence 𝑥 is invertible. This implies that 𝑎 is invertible. We thus have
shown that J is the set of noninvertible elements of A. Since J ≠ A, this
implies that A/J is a division ring, hence A is a local ring. We have shown
the implication (iii)⇒(i), and the proof of the implication (iii0)⇒(i) is
analogous.
Let us now assume (iv), ie, that the set N of noninvertible elements
of A is an additive subgroup of A.
We first prove that an element 𝑎 ∈ A which is left invertible is invertible.
Let 𝑏 ∈ A be such that 𝑏𝑎 = 1. The element 1 − 𝑎𝑏 is not left invertible,
since the equality 𝑥(1 − 𝑎𝑏) = 1 implies 𝑎 = 𝑥(1 − 𝑎𝑏)𝑎 = 𝑥(𝑎 − 𝑎𝑏𝑎) = 0,
 2.2. MAXIMAL AND PRIME IDEALS IN A COMMUTATIVE RING 67

but 𝑎 ≠ 0. Since 1 = 𝑎𝑏 + (1 − 𝑎𝑏) is invertible, this proves that 𝑎𝑏 is

invertible; then 𝑎 is right invertible, hence it is invertible.
Let M be a maximal left ideal of A. No element of M can be invertible,
hence M ⊂ N. Conversely, if 𝑎 ∈ A M, then 𝑎 is left invertible, hence is
invertible, by what precedes, and 𝑎 ∉ N; in other words, N ⊂ M. This
proves that N is the unique maximal left ideal of A, hence (ii). By what
precedes, N = J and A is a local ring. 

2.2. Maximal and prime ideals in a commutative ring

In this section, we restrict ourselves to the case of a commutative ring.

Proposition (2.2.1). — Let A be a commutative ring. An ideal I of A is maximal

if and only if the ring A/I is field.

Proof. — Let us assume that A/I is a field. Its ideals are then (0) and A/I.
Thus, the ideals of A containing I are I and A, which implies that I is a
maximal ideal. Conversely, if I is a maximal ideal, this argument shows
that the ring A/I is nonzero and that its only ideals are 0 and A/I itself.
Let 𝑥 be any nonzero element of A/I. Since the ideal (𝑥) generated by 𝑥
is nonzero, it equals A/I; therefore, there exists 𝑦 ∈ A/I such that 𝑥𝑦 = 1,
so that 𝑥 is invertible. This shows that A/I is a field. 

Definition (2.2.2). — Let A be a commutative ring. An ideal P of A is said to

be prime if the quotient ring A/P is an integral domain.

In particular, any maximal ideal of A is a prime ideal. Together

with Krull’s theorem (theorem 2.1.3), this shows that any nonzero
commutative ring possesses at least one prime ideal.
The set of all prime ideals of a ring A is denoted by Spec(A) and called
the spectrum of A The set of all maximal ideals of A is denoted by Max(A)
and called its maximal spectrum.

Proposition (2.2.3). — Let A be a commutative ring and let P be an ideal of A.

The following properties are equivalent:
(i) The ideal P is a prime ideal;
68 CHAPTER 2. IDEALS AND DIVISIBILITY

(ii) The ideal P is distinct from A and the product of any two elements of A
which do not belong to P does not belong to P;
(iii) The complementary subset A P is a multiplicative subset of A.

Proof. — The equivalence of the last two properties is straightforward.

Moreover, A/P being an integral domain (property (i), by definition)
means that P ≠ 0 (an integral domain is nonnull) and that the product of
any two elements not in P does not belong to P, hence the equivalence
with property (ii). 

Examples (2.2.4). — a) The ideal P = (0) is prime if and only if A is an

integral domain.
b) In the ring Z, the maximal ideals are the ideals (𝑝), for all prime
numbers 𝑝; the only remaining prime ideal is (0).
These ideals are indeed prime. Conversely, let I be a prime ideal of Z
and let us show that I is of the given form. We may assume that I ≠ (0);
let then 𝑝 be the smallest strictly positive element of I, so that I = 𝑝Z.
We need to show that 𝑝 is a prime number. Since Z is not a prime ideal
of Z, one has 𝑝 > 1. If 𝑝 is not prime, there are integers 𝑚, 𝑛 > 1 such
that 𝑝 = 𝑚𝑛. Then, 1 < 𝑚, 𝑛 < 𝑝, so that neither 𝑚 nor 𝑛 belongs to I;
but since 𝑝 = 𝑚𝑛 belongs to I, this contradicts the definition of a prime
ideal.

Proposition (2.2.5). — Let 𝑓 : A → B be a morphism of rings and let Q be a

prime ideal of B. Then P = 𝑓 −1 (Q) is a prime ideal of A.

Proof. — The ideal P is the kernel of the morphism from A to B/Q

given by the composition of 𝑓 and of the canonical morphism B → B/Q.
Consequently, passing to the quotient, 𝑓 defines an injective morphism
from A/P to B/Q. In particular, A/P is a domain and the ideal P is prime.
We could also have proved this fact directly. First of all, P ≠ A since
𝑓 (1) = 1 and 1 ∉ Q. Let then 𝑎 and 𝑏 be elements of A such that 𝑎𝑏 ∈ P.
Then, 𝑓 (𝑎𝑏) = 𝑓 (𝑎) 𝑓 (𝑏) belongs to Q, hence 𝑓 (𝑎) ∈ Q or 𝑓 (𝑏) ∈ Q. In the
first case, 𝑎 ∈ P, in the second, 𝑏 ∈ P. 
Thus a morphism of rings 𝑓 : A → B gives rise to a map 𝑓 ∗ : Spec(B) →
Spec(A), given by 𝑓 ∗ (Q) = 𝑓 −1 (Q) for every Q ∈ Spec(B).
 2.2. MAXIMAL AND PRIME IDEALS IN A COMMUTATIVE RING 69

Remarks (2.2.6). — Let A be a commutative ring.

a) For every ideal I of A, let V(I) be set of prime ideals P of A which
contain I. One has V(0) = Spec(A) and V(A) = ∅.
Í
Let (I 𝑗 ) be a family of ideals of A and let I = I 𝑗 be the ideal it generates.
A prime ideal P contains I if and only if it contains every I 𝑗 ; in other
Ñ
words, 𝑗 V(I 𝑗 ) = V(I).
Let I and J be ideals of A. Let us show that V(I) ∪ V(J) = V(I ∩ J) = V(IJ).
Indeed, if P ⊃ I or P ⊃ J, then P ⊃ I ∩ J ⊃ IJ, so that V(I) ∪ V(J) ⊂ V(I ∩ J) ⊂
V(IJ). On the other hand, let us assume that P ∉ V(I) ∪ V(J). Then, P ⊅ I,
so that there exists 𝑎 ∈ I such that 𝑎 ∉ P; similarly, there is 𝑏 ∈ J such that
𝑏 ∉ P. Then, 𝑎𝑏 ∈ IJ but 𝑎𝑏 ∉ P, hence P ∉ V(IJ). This proves the claim.
These properties show that the subsets of Spec(A) of the form V(I) are
the closed subsets of a topology on Spec(A). This topology is called the
Zariski topology.
b) Let I be an ideal of A. If I = A, then V(I) = ∅. Conversely, let us
assume that I ≠ A. By Krull’s theorem (theorem 2.1.3), there exists a
maximal ideal P of A such that I ⊂ P; then P ∈ V(I), hence V(I) ≠ ∅.
c) Let 𝑓 : A → B be a morphism of commutative rings. Let I be an
ideal of A; a prime ideal Q of B belongs to the inverse image ( 𝑓 ∗ )−1 (V(I))
of V(I) if and only if 𝑓 −1 (Q) contains I; this is equivalent to the fact that Q
contains 𝑓 (I), or, since Q is an ideal, to the inclusion Q ⊃ 𝑓 (I)B. We thus
have shown that ( 𝑓 ∗ )−1 (V(I)) = V( 𝑓 (I)B). This shows that the inverse
image of a closed subset of Spec(A) is closed in Spec(B). In other words,
the map 𝑓 ∗ : Spec(B) → Spec(A) induced by 𝑓 is continous.

Proposition (2.2.7). — Let A be a commutative ring, let S be a multiplicative

subset of A and let 𝑓 : A → S−1 A be the canonical morphism. The map
𝑓 ∗ : Spec(S−1 A) → Spec(A) is injective, and its image is the set of prime
ideals of A which are disjoint from S.

Proof. — For every prime ideal Q of S−1 A, the ideal P = 𝑓 ∗ (Q) of A is a

prime ideal of A. By proposition 1.7.1, it satisfies Q = P · S−1 A. Moreover,
P ∩ S = ∅ since, otherwise, one would have 1 ∈ Q. This implies that
the map 𝑓 ∗ : Spec(S−1 A) → Spec(A) is injective and that its image is
contained in the set of prime ideals of A which are disjoint from S.
70 CHAPTER 2. IDEALS AND DIVISIBILITY

Conversely, let P be a prime ideal of A which is disjoint from S. The

ideal Q = P · S−1 A of S−1 A is the set of all fractions 𝑎/𝑠, with 𝑎 ∈ P and
𝑠 ∈ S.
Let us first show that 𝑓 −1 (Q) = P: let 𝑎 ∈ A be such that 𝑎/1 ∈ Q; then
there exists 𝑏 ∈ P and 𝑠 ∈ S such that 𝑎/1 = 𝑏/𝑠, hence there exists 𝑡 ∈ S
such that 𝑡𝑠𝑎 = 𝑡𝑏. In particular 𝑡𝑠𝑎 ∈ P; one has 𝑡𝑠 ∉ P since P ∩ S = ∅,
hence 𝑎 ∈ P, by definition of a prime ideal.
Let us show that Q is a prime ideal of S−1 A. Since 𝑓 −1 (Q) = P ≠ A, one
has Q ≠ S−1 A. Let 𝑎, 𝑏 ∈ A and let 𝑠, 𝑡 ∈ S be such that (𝑎/𝑠)(𝑏/𝑡) ∈ Q.
Then 𝑎𝑏/𝑠𝑡 ∈ Q, hence 𝑎𝑏/1 ∈ Q. By what precedes, 𝑎𝑏 ∈ P. Since P is a
prime ideal of A, either 𝑎, or 𝑏 belongs to P. This implies that either 𝑎/𝑠
or 𝑏/𝑡 belongs to Q.
We thus have proved that 𝑓 ∗ is a bijection from Spec(S−1 A) to the
subset of Spec(A) consisting of prime ideal P such that P ∩ S = ∅. 

Example (2.2.8). — Let A be a commutative ring and let P be a prime ideal

of A. Then the fraction ring AP is a local ring, and PAP is its unique maximal
ideal. Indeed, the prime ideals of the fraction ring AP are of the form QP ,
where Q is a prime ideal of A such that Q ∩ (A P) = ∅, that is, Q ⊂ P.
This observation leads to an useful general proof technique in com-
mutative algebra, namely to first prove the desired result in the case for
local rings, and to reduce to this case by localization.

Remark (2.2.9). — When Spec(A) and Spec(S−1 A) are endowed with

their Zariski topologies, the map 𝑓 ∗ is continuous, by remark 2.2.6, b).
Let us prove that it is a homeomorphism from Spec(S−1 A) onto its image
in Spec(A).
It suffices to show that it maps closed sets of Spec(S−1 A) to traces of
closed sets of Spec(A). Let thus J be an ideal of S−1 A and let I = 𝑓 −1 (J)
be the set of all 𝑎 ∈ A such that 𝑎/1 ∈ J. Then I is an ideal of A and
I · S−1 A = J. Let us prove that 𝑓 ∗ (V(J)) = V(I) ∩ 𝑓 ∗ (Spec(SA )).
Let Q be a prime ideal of S−1 A and let P = 𝑓 ∗ (Q). Assume that Q
contains J; let 𝑎 ∈ I, then 𝑎/1 ∈ J, so that 𝑎/1 ∈ Q and finally 𝑎 ∈ P, so
that P contains I. Assume conversely that P contains I; let 𝑎 ∈ P and let
𝑠 ∈ S; then 𝑎/1 ∈ J by definition of I, hence 𝑎/𝑠 ∈ J, hence Q contains J.
This concludes the proof.
 2.2. MAXIMAL AND PRIME IDEALS IN A COMMUTATIVE RING 71

Examples (2.2.10). — The next two examples give some geometric content
to the terminology “localization” which is used in the context of fraction
rings.
a) Let 𝑎 ∈ A and let S be the multiplicative subset {1; 𝑎; 𝑎 2 ; . . .}. For
a prime ideal P, being disjoint from S is equivalent to not containing 𝑎;
consequently, 𝑓 ∗ identifies Spec(S−1 A) with the complement of the closed
subset V((𝑎)) of Spec(A).
b) Let P be a prime ideal of A and let S be the multiplicative subset
A P. Being disjoint from S is equivalent to being contained in P.
Let us show that 𝑓 ∗ identifies Spec(S−1 A) with the intersection of all
neighborhoods of P in Spec(A).
Let Q be a prime ideal of A such that Q ⊂ P. Let U be an open
neighborhood of P in Spec(A) and let I be an ideal of A such that
U = Spec(A) V(I). Then P ∈ U, hence P ∉ V(I), hence P does not
contain I. A fortiori, Q does not contain I, so that Q ∉ V(I). This proves
that the image of Spec(S−1 A) is contained in any open neighborhood of P
in Spec(A). Conversely, if Q is a prime ideal of A which is not contained
in P, then P ∉ V(Q), and Spec(A) V(Q) is an open neighborhood of P
that does not contain Q.
Proposition (2.2.11). — Let A be a commutative ring.
a) An element of A is nilpotent if and only if it belongs to every prime ideal
of A.
√
b) For any ideal I of A, its radical I is the intersection of all prime ideals
of A containing I.
Proof. — a) Let 𝑎 ∈ A and let P be a prime ideal of A. If 𝑎 ∉ P, the
definition of a prime ideal implies by induction on 𝑛 that 𝑎 𝑛 ∉ P for any
integer 𝑛 > 0. In particular, 𝑎 𝑛 ≠ 0 and 𝑎 is not nilpotent. Nilpotent
elements belong to every prime ideal.
Conversely, let 𝑎 ∈ A be a non-nilpotent element. The set S =
{1, 𝑎, 𝑎 2 , . . .} of all powers of 𝑎 is a multiplicative subset of A which does
not contain 0. The localization S−1 A is then non null. Let M be a prime
ideal of S−1 A (for example, a maximal ideal of this nonzero ring) and let
P be the set of all elements 𝑥 ∈ A such that 𝑥/1 ∈ M. By definition, P is
the inverse image of M by the canonical morphism from A to S−1 A. It is
72 CHAPTER 2. IDEALS AND DIVISIBILITY

therefore a prime ideal of A. Since 𝑎 ∈ S, 𝑎/1 is invertible in S−1 A, hence

𝑎/1 ∉ M and 𝑎 ∉ P. We thus have found a prime ideal of A which does
not contain 𝑎, and this concludes the proof.
Part b) follows formally. √ Indeed, let B be the quotient ring A/I. An
element 𝑎 ∈ A belongs to I if and only if its class clI (𝑎) is nilpotent
in A/I. By a), this is equivalent to the fact that clI (𝑎) belongs to every
prime ideal of A/I. Recall that every ideal of A/I is of the form P/I,
for a unique ideal P of A which contains I; moreover, (A/I)/(P/I) is
isomorphic to A/P, so that P/I is a prime ideal of A/I if and only √ if P is a
prime ideal of A which contains I. Consequently, 𝑎 belongs to I if and √
only if it belongs to every prime ideal containing I. In other words, I
is the intersection of all prime ideals of A which contain I, as was to be
shown. 

Lemma (2.2.12) (Prime avoidance lemma). — Let A be a commutative ring,

let 𝑛 be an integer > 1 and let P1 , . . . , P𝑛 be prime ideals of A, and let I be an
ideal of A. If I ⊂ 𝑛𝑖=1 P𝑖 , then there exists 𝑖 ∈ {1, . . . , 𝑛} such that I ⊂ P𝑖 .
Ð

Proof. — We prove the result by induction on 𝑛. It is obvious is 𝑛 = 1.

Assume that for every 𝑖, I is not contained in P𝑖 . By induction, for
every 𝑖 ∈ {1, . . . , 𝑛}, the ideal I is not contained in the union of the prime
ideals P 𝑗 , for 𝑗 ≠ 𝑖. Consequently, there exists 𝑎 𝑖 ∈ I such that 𝑎 𝑖 ∉ P 𝑗 if
𝑗 ≠ 𝑖. Since I ⊂ 𝑛𝑗=1 P 𝑗 , we see that 𝑎 𝑖 ∈ P𝑖 . Let 𝑎 = 𝑎1 + 𝑎2 . . . 𝑎 𝑛 ; since
Ð
the elements 𝑎 1 , . . . , 𝑎 𝑛 belong to I, we have 𝑎 ∈ I. On the other hand,
P1 is prime, hence 𝑎2 . . . 𝑎 𝑛 ∉ P1 ; moreover, 𝑎1 ∈ P1 , so that 𝑎 ∉ P1 . For
𝑖 ∈ {2, . . . , 𝑛}, 𝑎2 . . . 𝑎 𝑛 ∈ P𝑖 , but 𝑎1 ∉ P𝑖 , hence 𝑎 ∉ P𝑖 . Consequently,
𝑎 does not belong to any of the ideals P1 , . . . , P𝑛 . This contradicts the
hypothesis that I is contained in their union. 

The following lemma is another application of Zorn’s lemma, but

constructs minimal prime ideals.

Lemma (2.2.13). — Let A be a commutative ring, let I be an ideal of A such

that I ≠ A. Then, for any prime ideal Q containing I, there exists a prime ideal
P of A such that I ⊂ P ⊂ Q and which is minimal among all prime ideals of A
containing I.
 2.2. MAXIMAL AND PRIME IDEALS IN A COMMUTATIVE RING 73

In particular, every prime ideal of A contains a prime ideal which is

minimal for inclusion.
Proof. — Let us order the set 𝒮 of prime ideals of A containing I by
reverse inclusion: we say that P  Q if P contains Q. The lemma will
follow at once from Zorn’s lemma (theorem A.2.11) once we show that
𝒮 is an inductive set.
To that aim, let (P 𝑗 ) 𝑗∈J be a totally ordered family of prime ideals of A. If
this family is empty, the prime ideal Q contains I and is an upper bound.
Let us assume that this family is not empty and let P be its intersection.
It contains I, because I is contained in each P 𝑗 , and is distinct from A
because it is contained in P 𝑗 ≠ A, for any 𝑗 ∈ J. Let us show that P is a
prime ideal of A. Let 𝑎, 𝑏 ∈ A be such that 𝑎𝑏 ∈ P but 𝑏 ∉ P; let us show
that 𝑎 belongs to every ideal P 𝑗 . So let 𝑗 ∈ J. By definition of P, there
exists 𝑖 ∈ J such that 𝑏 ∉ P𝑖 . Since the family (P 𝑗 ) is totally ordered, we
have either P𝑖 ⊂ P 𝑗 , or P 𝑗 ⊂ P𝑖 . In the first case, the relations 𝑎𝑏 ∈ P𝑖 and
𝑏 ∉ P𝑖 imply that 𝑎 ∈ P𝑖 , hence 𝑎 ∈ P 𝑗 . In the latter case, we have 𝑏 ∉ P 𝑗 ;
since 𝑎𝑏 ∈ P 𝑗 and P 𝑗 is prime, this implies 𝑎 ∈ P 𝑗 . Consequently, P ∈ 𝒮,
hence 𝒮 is an inductive set. 

Remark (2.2.14). — Let S be a multiplicative subset of A. The map

P ↦→ S−1 P is a bijection respecting inclusion of the set of prime ideals
of A disjoint from S to the set of prime ideals of S−1 A. In particular, the
inverse image of a minimal prime ideals of S−1 A by this bijection is a
minimal prime ideal of A.
Conversely, if P is a minimal prime ideal of A which is disjoint from S,
then S−1 P is a minimal prime ideal of S−1 A.
Lemma (2.2.15). — Let A be a commutative ring and let P be a minimal prime
ideal of A. For every 𝑎 ∈ P there exists 𝑏 ∈ A P and an integer 𝑛 > 0 such
that 𝑎 𝑛 𝑏 = 0.
Proof. — Let 𝑎 ∈ A and let S be the set of elements of A of the form
𝑎 𝑛 𝑏, for 𝑏 ∈ A P and 𝑛 > 0. It is a multiplicative subset of A that
contains A P. Assume 0 ∉ S. Then, S−1 A is a nonzero ring, hence
contains a prime ideal Q0. The inverse image of Q in A is a prime ideal P0
disjoint from S. In particular, P0 ⊂ P. Moreover, 𝑎 ∉ P0, because 𝑎 ∈ S,
hence P0 ≠ P. This contradicts the hypothesis that P is a minimal prime
74 CHAPTER 2. IDEALS AND DIVISIBILITY

ideal of A. Consequently, 0 ∈ S and there exists 𝑏 ∈ A P and 𝑛 > 0

such that 𝑎 𝑛 𝑏 = 0. 

2.3. Hilbert’s Nullstellensatz

The following theorem of Hilbert gives a precise description of maximal
ideals of polynomial rings over an algebraically closed field.
Theorem (2.3.1) (Hilbert’s Nullstellensatz). — Let K be an algebraically
closed field and let 𝑛 be a positive integer. The maximal ideals of K[X1 , . . . , X𝑛 ]
are the ideals (X1 − 𝑎1 , . . . , X𝑛 − 𝑎 𝑛 ), for (𝑎1 , . . . , 𝑎 𝑛 ) ∈ K𝑛 .
Proof. — Let us first show that for any (𝑎1 , . . . , 𝑎 𝑛 ) ∈ K𝑛 , the ideal
(X1 − 𝑎1 , . . . , X𝑛 − 𝑎 𝑛 ) of K[X1 , . . . , X𝑛 ] is indeed a maximal ideal. Let
𝜑 : K[X1 , . . . , X𝑛 ] → K be the “evaluation at (𝑎1 , . . . , 𝑎 𝑛 )” morphism,
defined by 𝜑(P) = P(𝑎1 , . . . , 𝑎 𝑛 ). It is surjective and induces an iso-
morphism K[X1 , . . . , X𝑛 ]/Ker(𝜑) ' K; since K is a field, Ker(𝜑) is a
maximal ideal of K[X1 , . . . , X𝑛 ]. It thus suffices to show that Ker(𝜑)
coincides with the ideal (X1 − 𝑎1 , . . . , X𝑛 − 𝑎 𝑛 ). One inclusion is obvious:
if P = (X1 − 𝑎 1 )P1 + · · · + (X𝑛 − 𝑎 𝑛 )P𝑛 for some polynomials P1 , . . . , P𝑛 ,
then 𝜑(P) = 0. Conversely, let P ∈ K[X1 , . . . , X𝑛 ] be such that 𝜑(P) = 0.
Let us perform the euclidean division of P by X1 − 𝑎1 with respect to the
variable X1 ; we get a polynomial P1 and a polynomial R1 ∈ K[X2 , . . . , X𝑛 ]
such that
P = (X1 − 𝑎1 )P1 + R1 (X2 , . . . , X𝑛 ).
Let us repeat the process and divide by X2 − 𝑎2 , etc.: we see that there
exist polynomials P1 , . . . , P𝑛 , where P𝑖 ∈ K[X𝑖 , . . . , X𝑛 ] and a constant
polynomial R𝑛 , such that
P = (X1 − 𝑎 1 )P1 + · · · + (X𝑛 − 𝑎 𝑛 )P𝑛 + R𝑛 .
Evaluating at (𝑎1 , . . . , 𝑎 𝑛 ), we get
𝜑(P) = P(𝑎1 , . . . , 𝑎 𝑛 ) = R𝑛 .
Since we assumed 𝜑(P) = 0, we have R𝑛 = 0 and P belongs to the ideal
(X1 − 𝑎1 , . . . , X𝑛 − 𝑎 𝑛 ).
We shall only prove the converse assertion under the supplementary
hypothesis that the field K is uncountable. (Exercise 2/5 explains how to
 2.3. HILBERT’S NULLSTELLENSATZ 75

derive the general case from this particular case; see also corollary 9.1.3
for an alternative proof.) Let M be a maximal ideal of K[X1 , . . . , X𝑛 ] and
let L be the quotient ring K[X1 , . . . , X𝑛 ]/M; it is a field. Let 𝑥 𝑖 denote the
class of X𝑖 in L. The image of K by the canonical homomorphism is a
subfield of L which we identify with K.
The field L admits a natural structure of K-vector space. As such, it
is generated by the countable family of all 𝑥 1𝑖1 . . . 𝑥 𝑛𝑖 𝑛 , for 𝑖 1 , . . . , 𝑖 𝑛 ∈ N.
Indeed, any element of L is the class of some polynomial, hence of a
linear combination of monomials. Moreover, the set N𝑛 of all possible
exponents of monomials is countable.
Let 𝑎 be any element of L and let 𝜑 : K[T] → L the ring morphism
given by 𝜑(P) = P(𝑎). Assume that it is injective. Then, 𝜑 extends as a
(still injective) morphism of fields, still denoted by 𝜑, from K(T) to L. In
particular, the elements 1/(𝑎 − 𝑐) (for 𝑐 ∈ K) are linearly independent
over K since they are the images of the rational functions 1/(T − 𝑐)
which are linearly independent in K(T) in view of the uniqueness of
the decomposition of a rational function in simple terms. However,
this would contradict the lemma 2.3.2 below: any linearly independent
family of L must be countable, while K is not. This implies that the
morphism 𝜑 is not injective.
Let P ∈ K[T] be a nonzero polynomial such that P(𝑎) = 0. The
polynomial P is not constant. Since K is algebraically closed, it has the
form 𝑐 𝑚 𝑖=1 (T − 𝑐 𝑖 ), for some strictly positive integer 𝑚 and elements
Î
𝑐1 , . . . , 𝑐 𝑚 ∈ K and 𝑐 ∈ K∗ . Then, 𝑐 𝑛𝑖=1 ( 𝑓 − 𝑐 𝑖 ) = 0 in L. Since L is a
Î
field, there exists 𝑖 ∈ {1, . . . , 𝑚} such that 𝑎 = 𝑐 𝑖 ; consequently, 𝑎 ∈ K
and L = K.
In particular, there exists, for any 𝑖 ∈ {1, . . . , 𝑛}, some element 𝑎 𝑖 ∈ K
such that 𝑥 𝑖 = 𝑎 𝑖 . This implies the relations X𝑖 − 𝑎 𝑖 ∈ M, hence the
ideal M contains the ideal (X1 − 𝑎1 , . . . , X𝑛 − 𝑎 𝑛 ). Since the latter is a
maximal ideal, we have equality, and this concludes the proof. 

Lemma (2.3.2). — Let K be a field, let V be a K-vector space which is generated

by a countable family. Then every subset of V which is linearly independent is
countable.
76 CHAPTER 2. IDEALS AND DIVISIBILITY

Proof. — This will follow from a general result on dimension of vec-

tor spaces, possibly infinite dimensional. However, one may give an
alternate argument that only makes use of finite dimensional vector
spaces.
Let J be a countable set and (𝑒 𝑗 ) 𝑗∈J be a generating family of V. Let
(𝑣 𝑖 )𝑖∈I be a linearly independent family in V; we have to prove that I is
countable. For every finite subset A of J, the subspace VA of V generated
by the elements 𝑒 𝑗 , for 𝑗 ∈ A, is finite dimensional. In particular, the
subset IA of all 𝑖 ∈ I such that 𝑣 𝑖 ∈ VA is finite. Since the set of all finite
subsets of J is countable, the union I0 of all subsets IA , when A runs
over all finite subsets of J, is countable. For every 𝑖 ∈ I, there is a finite
subset A of J sch that 𝑣 𝑖 is a linear combination of the 𝑒 𝑗 , for 𝑗 ∈ A, hence
𝑖 ∈ I0. Consequently, I = I0 and I is countable. 
The following theorem is a topological analogue.

Theorem (2.3.3) (Gelfand). — Let X be a topological space; let 𝒞(X) be the

ring of real valued, continuous functions on X.
For any 𝑥 ∈ X, the set M𝑥 of all continuous functions on X which vanish at 𝑥
is a maximal ideal of the ring 𝒞(X).
If X is a compact metric space, the map 𝑥 ↦→ M𝑥 from X to the set of maximal
ideals of 𝒞(X) is a bijection.

Proof. — Let 𝜑 𝑥 : 𝒞(X) → R the ring morphism given by 𝑓 ↦→ 𝑓 (𝑥)

(“evaluation at the point 𝑥”). It is surjective, with kernel M𝑥 . This proves
that M𝑥 is a maximal ideal of 𝒞(X).
Let us assume that X is a compact metric space, with distance 𝑑. For
any point 𝑥 ∈ X, the function 𝑦 ↦→ 𝑑(𝑥, 𝑦) is continuous, vanishes at 𝑥
but does not vanish at any other point of X. Consequently, it belongs
to M𝑥 but not to M 𝑦 if 𝑦 ≠ 𝑥. This shows that M𝑥 ≠ M 𝑦 for 𝑥 ≠ 𝑦 and
the mapping 𝑥 ↦→ M𝑥 is injective.
Let I be an ideal of 𝒞(X) which is not contained in any of the maximal
ideals M𝑥 , for 𝑥 ∈ X. For any 𝑥 ∈ X, there thus exists a continuous
function 𝑓𝑥 ∈ I such that 𝑓𝑥 (𝑥) ≠ 0. By continuity, the set U𝑥 of points of X
at which 𝑓𝑥 does not vanish is an open neighborhood of 𝑥. These open
sets U𝑥 cover X. Since X is compact, there exists a finite subset S ⊂ X such
that the open sets U𝑠 , for 𝑠 ∈ S, cover X as well. Let us set 𝑓 = 𝑠∈S ( 𝑓𝑠 )2 .
Í
 2.3. HILBERT’S NULLSTELLENSATZ 77

This is a positive continuous function, and it belongs to I. If 𝑥 ∈ X

is a point such that 𝑓 (𝑥) = 0, then for any 𝑠 ∈ S, 𝑓𝑠 (𝑥) = 0, that is to
say, 𝑥 ∉ U𝑠 . Since the U𝑠 cover X, we have a contradiction and 𝑓 does
not vanish at any point of X. It follows that 𝑓 is invertible in 𝒞(X) (its
inverse is the continuous function 𝑥 ↦→ 1/ 𝑓 (𝑥)). Since 𝑓 ∈ I, we have
shown I = 𝒞(X). Consequently, any ideal of 𝒞(X) distinct from 𝒞(X) is
contained in one of the maximal ideals M𝑥 , so that these ideals constitute
the whole set of maximal ideals of 𝒞(X). 
Let K be an algebraically closed field. Hilbert’s Nullstellensatz is the
basis of an admirable correspondence between algebra (some ideals of
the polynomial ring K[X1 , . . . , X𝑛 ]) and geometry (some subsets of K𝑛 )
which we describe now.

Definition (2.3.4). — An algebraic set is a subset of K𝑛 defined by a family of

polynomial equations.

Specifically, a subset Z of K𝑛 is an algebraic set if and only if there

exists a subset S of K[X1 , . . . , X𝑛 ] such that
Z = 𝒵(S) = {(𝑎1 , . . . , 𝑎 𝑛 ) ∈ K𝑛 ; for every P ∈ S, P(𝑎1 , . . . , 𝑎 𝑛 ) = 0}.

Proposition (2.3.5). — a) If S ⊂ S0, then 𝒵(S0) ⊂ 𝒵(S).

b) The empty set and K𝑛 are algebraic sets.
c) If hSi is the ideal generated by S in K[X1 , . . . , X𝑛 ], then 𝒵(hSi) = 𝒵(S).
d) The intersection of a family of algebraic sets, the union of two algebraic
sets are algebraic sets.
√
e) If I is any ideal of K[X1 , . . . , X𝑛 ], then 𝒵(I) = 𝒵( I).

Proof. — a) Let (𝑎1 , . . . , 𝑎 𝑛 ) ∈ 𝒵(S0) and let us show that (𝑎1 , . . . , 𝑎 𝑛 ) ∈

𝒵(S). If P ∈ S, we have to show that P(𝑎1 , . . . , 𝑎 𝑛 ) = 0, which holds since
P ∈ S0.
b) One has ∅ = 𝒵({1}) (the constant polynomial 1 does not vanish
at any point of K𝑛 ) and K𝑛 = 𝒵({0}) (the zero polynomial vanishes
everywhere; we could also write K𝑛 = 𝒵(∅)).
c) Since S ⊂ hSi, we have 𝒵(hSi) ⊂ 𝒵(S). Conversely, let (𝑎1 , . . . , 𝑎 𝑛 ) ∈
𝒵(S) and let us show that (𝑎1 , . . . , 𝑎 𝑛 ) ∈ 𝒵(hSi). Let P ∈ hSi; by definition,
78 CHAPTER 2. IDEALS AND DIVISIBILITY

there are finite families (P𝑖 )𝑖∈I and (Q𝑖 )𝑖∈I of polynomials in K[X1 , . . . , X𝑛 ]
such that P = P𝑖 Q𝑖 and Q𝑖 ∈ S for every 𝑖 ∈ I. Then,
Í
Õ
P(𝑎1 , . . . , 𝑎 𝑛 ) = P𝑖 (𝑎1 , . . . , 𝑎 𝑛 )Q𝑖 (𝑎1 , . . . , 𝑎 𝑛 ) = 0
𝑖∈I

since Q𝑖 (𝑎1 , . . . , 𝑎 𝑛 ) = 0. Consequently, (𝑎1 , . . . , 𝑎 𝑛 ) ∈ 𝒵(hSi)

d) Let (Z 𝑗 ) be a family of algebraic sets, for every 𝑗, let S 𝑗 be a subset
of K[X1 , . . . , X𝑛 ] such that Z 𝑗 = 𝒵(S 𝑗 ). We shall show that
Ù Ø
𝒵(S 𝑗 ) = 𝒵( S 𝑗 ).
𝑗 𝑗

Indeed, (𝑎1 , . . . , 𝑎 𝑛 ) belongs to 𝑗 𝒵(S 𝑗 ) if and only if P(𝑎 1 , . . . , 𝑎 𝑛 ) for

Ñ
every 𝑗 and every P ∈ S 𝑗 , which means exactly that P(𝑎 1 , . . . , 𝑎 𝑛 ) = 0 for
every P ∈ 𝑗 S 𝑗 , that is, (𝑎1 , . . . , 𝑎 𝑛 ) ∈ 𝒵( 𝑗 S 𝑗 ).
Ð Ð
Let S and S0 be two subsets of K[X1 , . . . , X𝑛 ]. Let T = {PP0 ; P ∈
S, P0 ∈ S0 }. We are going to show that 𝒵(S) ∪ 𝒵(S0) = 𝒵(T). Indeed,
if (𝑎1 , . . . , 𝑎 𝑛 ) ∈ 𝒵(S) and Q ∈ T, we may write Q = PP0 with P ∈ S
and P0 ∈ S0. Then, Q(𝑎1 , . . . , 𝑎 𝑛 ) = P(𝑎1 , . . . , 𝑎 𝑛 )P0(𝑎1 , . . . , 𝑎 𝑛 ) = 0 since
(𝑎1 , . . . , 𝑎 𝑛 ) ∈ 𝒵(S). In other words, 𝒵(S) ⊂ 𝒵(T). Similarly, 𝒵(S0) ⊂
𝒵(T), hence 𝒵(S) ∪ 𝒵(S0) ⊂ 𝒵(T). Conversely, let (𝑎1 , . . . , 𝑎 𝑛 ) ∈ 𝒵(T).
To show that (𝑎1 , . . . , 𝑎 𝑛 ) ∈ 𝒵(S) ∪ 𝒵(S0), it suffices to prove that if
(𝑎1 , . . . , 𝑎 𝑛 ) ∉ 𝒵(S0), then (𝑎1 , . . . , 𝑎 𝑛 ) ∈ 𝒵(S). By definition, there is a
polynomial P0 ∈ S0 such that P0(𝑎 1 , . . . , 𝑎 𝑛 ) ≠ 0. Then, for every P ∈ S,
one has PP0 ∈ T, hence (PP0)(𝑎1 , . . . , 𝑎 𝑛 ) = 0 = P(𝑎1 , . . . , 𝑎 𝑛 )P0(𝑎 1 , . . . , 𝑎 𝑛 )
so that P(𝑎1 , . .√ . , 𝑎 𝑛 ) = 0, as was
√ to be shown.
e) Since I ⊂ √ I, one has 𝒵( I) ⊂ 𝒵(I). Conversely, let (𝑎1 , . . . , 𝑎 𝑛 ) ∈
𝒵(I). Let P ∈ I, let 𝑚 > 1 be such that P𝑚 ∈ √ I. Then P𝑚 (𝑎1 , . . . , 𝑎 𝑛 ) = 0,
hence P(𝑎1 , . . . , 𝑎 𝑛 ) = 0 and (𝑎1 , . . . , 𝑎 𝑛 ) ∈ 𝒵( I). 

Remark (2.3.6). — The preceding proposition can be rephrased by saying

that there exists a topology on K𝑛 for which the closed sets are the
algebraic sets. This topology is called the Zariski topology.

We have constructed one of the two directions of the corespondence:

with any ideal I of K[X1 , . . . , X𝑛 ], we associate the algebraic set 𝒵(I).
 2.3. HILBERT’S NULLSTELLENSATZ 79

From the proof of the previous proposition, we recall the formulae

𝒵(0) = K𝑛
𝒵(K[X1 , . . . , X𝑛 ]) = ∅
Õ Ù
𝒵( I𝑗 ) = 𝒵(I 𝑗 )
𝑗

𝒵(IJ) = 𝒵(I) ∪ 𝒵(J).

The other direction of the correspondence associates an ideal with any
subset of K𝑛 .
Definition (2.3.7). — Let V be a subset of K𝑛 . One defines
ℐ(V) = {P ∈ K[X1 , . . . , X𝑛 ] ; P(𝑎1 , . . . , 𝑎 𝑛 ) = 0 for every (𝑎1 , . . . , 𝑎 𝑛 ) ∈ V}
— a) For every V ⊂ K𝑛 , ℐ(V) is an ideal of K[X1 , . . . , X𝑛 ].
Proposition (2.3.8). p
Moreover, ℐ(V) = ℐ(V).
b) If V ⊂ V0, then ℐ(V0) ⊂ ℐ(V).
c) For any two subsets V and V0 of K𝑛 , one has ℐ(V ∪ V0) = ℐ(V) ∩ ℐ(V0).
Proof. — a) For any (𝑎1 , . . . , 𝑎 𝑛 ) ∈ K𝑛 , the map P ↦→ P(𝑎1 , . . . , 𝑎 𝑛 ) is a
ring morphism from K[X1 , . . . , X𝑛 ] to K, and ℐ(V) is the intersection of
the kernels of those morphisms, for (𝑎1 , . . . , 𝑎 𝑛 ) ∈ V. Consequently, it is
an ideal of K[X1 , . . . , X𝑛 ]. Moreover, if P ∈ K[X1 , . . . , X𝑛 ] and 𝑚 > 1 is
such that P𝑚 ∈ ℐ(V), then P𝑚 (𝑎1 , . . . , 𝑎 𝑛 ) = 0 for every (𝑎1 , . . . , 𝑎 𝑛 ) ∈ V;
p that P(𝑎1 , . . . , 𝑎 𝑛 ) = 0, hence P ∈ ℐ(V). This shows that
this implies
ℐ(V) = ℐ(V).
b) Let P ∈ ℐ(V0). For any (𝑎1 , . . . , 𝑎 𝑛 ) ∈ V, we have P(𝑎1 , . . . , 𝑎 𝑛 ) = 0,
since V ⊂ V0. Consequently, P ∈ ℐ(V).
c) By definition, a polynomial P belongs to ℐ(V ∪ V0) if and only if it
vanishes at any point of V and of V0. 

Proposition (2.3.9). — a) For any ideal I of K[X1 , . . . , X𝑛 ], one has I ⊂

ℐ(𝒵(I)).
b) For any subset V of K𝑛 , one has V ⊂ 𝒵(ℐ(V)).
Proof. — a) Let P ∈ I; let us show that P ∈ ℐ(𝒵(I)). We need to show that
P vanishes at every point of 𝒵(I). Now, for any point (𝑎1 , . . . , 𝑎 𝑛 ) ∈ 𝒵(I),
one has P(𝑎1 , . . . , 𝑎 𝑛 ) = 0, since P ∈ I.
80 CHAPTER 2. IDEALS AND DIVISIBILITY

b) Let (𝑎1 , . . . , 𝑎 𝑛 ) ∈ V and let us show that (𝑎1 , . . . , 𝑎 𝑛 ) belongs to

𝒵(ℐ(V)). We thus need to prove that for any P ∈ ℐ(V), one has
P(𝑎1 , . . . , 𝑎 𝑛 ) = 0. This assertion is clear, since (𝑎1 , . . . , 𝑎 𝑛 ) ∈ V. 
We are going to use Hilbert’s Nullstellensatz to establish the following
theorem.
Theorem (2.3.10). — For any ideal I of K[X1 , . . . , X𝑛 ], one has
√
ℐ(𝒵(I)) = I .
Before we pass to the proof, let us show how this gives rise to a bijection
between algebraic sets — geometry — and ideals equal to their own
radical — algebra.
Corollary (2.3.11). — The maps V ↦→ ℐ(V) and I ↦→ 𝒵(I) induce bijec-
tions, inverse the one of the other, 𝑛
√ between algebraic sets in K and ideals I
of K[X1 , . . . , X𝑛 ] such that I = I.
√
Proof of the corollary. — Let I be an ideal of K[X1 , . . . , X𝑛 ] such that I = I.
By theorem 2.3.10, one has
√
ℐ(𝒵(I)) = I = I.
Conversely, let V be an algebraic set and let S be any subset of
K[X1 , . . . , X𝑛 ] such that V = 𝒵(S). Letting
√ I = hSi, we have V = 𝒵(I). By
proposition 2.3.5, we even have V = 𝒵( I). Then,
√
ℐ(V) = ℐ(𝒵(I)) = I,
hence √
V = 𝒵(I) = 𝒵( I) = 𝒵(ℐ(V)).
This concludes the proof of the corollary. 
√
Proof of theorem 2.3.10. — The inclusion I ⊂ ℐ(𝒵(I)) is easy (and has
been proved incidentally√ in the course of the proof of prop. 2.3.5).
Let indeed P ∈ I, let 𝑚 > 1 be such that P𝑚 ∈ I. Then, for any
(𝑎1 , . . . , 𝑎 𝑛 ) ∈ 𝒵(I), one has P𝑚 (𝑎1 , . . . , 𝑎 𝑛 ) = 0 hence P(𝑎1 , . . . , 𝑎 𝑛 ) = 0.
Consequently, P ∈ ℐ(𝒵(I)).
Conversely, let P be a polynomial belonging to ℐ(𝒵(I)). We want to
show that there exists 𝑚 > 1 such that P𝑚 ∈ I. Let us introduce the
 2.3. HILBERT’S NULLSTELLENSATZ 81

ideal J of K[X1 , . . . , X𝑛 , T] which is generated by I and by the polynomial

1 − TP. We have 𝒵(J) = ∅. Let indeed (𝑎1 , . . . , 𝑎 𝑛 , 𝑡) ∈ K𝑛+1 be a point
belonging to 𝒵(J). Since P ∈ I, it follows from the definition of the ideal J
that P belongs to J, hence P(𝑎 1 , . . . , 𝑎 𝑛 ) = 0. On the other hand, 1 − TP
belongs to J too, so that we have 𝑡P(𝑎1 , . . . , 𝑎 𝑛 ) = 1, a contradiction.
By Hilbert’s Nullstellensatz, the ideal J is contained in no maximal
ideal of K[X1 , . . . , X𝑛 , T], hence J = K[X1 , . . . , X𝑛 , T]. Consequently, there
are polynomials Q𝑖 ∈ I, R𝑖 ∈ K[X1 , . . . , X𝑛 , T] and R ∈ K[X1 , . . . , X𝑛 , T]
such that Õ
1 = (1 − TP)R + Q𝑖 R𝑖 ,
𝑖
and the image of the ideal I in the quotient ring K[X1 , . . . , X𝑛 , T]/(1 − TP)
generates the unit ideal.
Now, recall from proposition 1.6.6 that this quotient ring is iso-
morphic to the localization of K[X1 , . . . , X𝑛 ] by the multiplicative set
S = {1, P, P2 , . . . , } generated by P. Consequently, the image of the
ideal I in the fraction ring S−1 K[X1 , . . . , X𝑛 ] generates the full ideal,
which means that the multiplicative subset S meets the ideal I. In other
words, there is an integer 𝑚 such that P𝑚 ∈ I, as was to be shown. 

Remark (2.3.12). — The introduction of a new variable T in the preced-

ing proof is known as Rabinowitsch trick. However, in most of the
presentations of the proof, it is often followed by the following, more
elementary-looking, argument. We start as above until we reach the
equality 1 = (1 − TP)R + Q𝑖 R𝑖 . of polynomials in K[X1 , . . . , X𝑛 , T]. Our
Í
goal now is to substitute T = 1/P. Formally, we obtain an equality of
rational functions in the field K(X1 , . . . , X𝑛 ):
Õ
1= Q𝑖 (X1 , . . . , X𝑛 )R𝑖 (X1 , . . . , X𝑛 , 1/P(X1 , . . . , X𝑛 )).
𝑖
The denominators are powers of P; chasing them, we shall obtain a
relation of the form
Õ
P M
= Q𝑖 (X1 , . . . , X𝑛 )S𝑖 (X1 , . . . , X𝑛 )
𝑖

which shows that PM belongs to I. To make this argument more precise,

let M be an integer larger than the T-degrees of all of the polynomials R𝑖
82 CHAPTER 2. IDEALS AND DIVISIBILITY

and of the polynomial R. We can then write

M
Õ
R(X1 , . . . , X𝑛 , T) = S𝑖 (X1 , . . . , X𝑛 )T𝑚
𝑚=0
and, for any 𝑖,
M
Õ
R𝑖 (X1 , . . . , X𝑛 , T) = S𝑖,𝑚 (X1 , . . . , X𝑛 )T𝑚 ,
𝑚=0

for some polynomials S𝑖 and S𝑖,𝑚 in K[X1 , . . . , X𝑛 ]. Multiplying both

sides of the initial equality by PM , we obtain a relation
M
Õ M
ÕÕ
𝑚
P M
= (1 − TP) (PT) S𝑚 P M−𝑚
+ (PT)𝑚 Q𝑖 S𝑖,𝑚 PM−𝑚 .
𝑚=0 𝑖 𝑚=0
Collecting the monomials of various T-degrees, we deduce
Õ
P M
= S0 P + M
Q𝑖 S𝑖,0 PM ,
𝑖
Õ
0 = S𝑚 P M−𝑚
− S𝑚−1 P M−𝑚+1
+ Q𝑖 S𝑖,𝑚 PM−𝑚
𝑖
for 𝑚 ∈ {1, . . . , M}, and
0 = −SM .
Summing all of these relations, we get
M
ÕÕ
PM = Q𝑖 S𝑖,𝑚 PM−𝑚 .
𝑖 𝑚=0

Since the polynomials

√ Q𝑖 belong to I, we conclude that PM belongs to I,
hence P ∈ I.

2.4. Principal ideal domains, euclidean rings

2.4.1. — Let A be a commutative ring.
Recall that an ideal of A is said to be principal if it is of the form 𝑎A, for
some 𝑎 ∈ A. One also writes (𝑎) for 𝑎A.
Let 𝑎, 𝑏 ∈ A. The ideal 𝑎A is contained in the ideal 𝑏A if and only if
there exists 𝑐 ∈ A such that 𝑎 = 𝑏𝑐, that is, if and only if 𝑏 divides 𝑎.
 2.4. PRINCIPAL IDEAL DOMAINS, EUCLIDEAN RINGS 83

Assume moreover that A is an integral domain. Let 𝑎, 𝑏 ∈ A be such

that 𝑎A = 𝑏A. Then, there exist 𝑐, 𝑑 ∈ A such that 𝑎 = 𝑏𝑐 and 𝑏 = 𝑎𝑑,
hence 𝑎 = 𝑎(𝑐𝑑) and 𝑏 = (𝑐𝑑). If 𝑎 ≠ 0 then 𝑏 ≠ 0; simplifying by 𝑎, we
get 𝑐𝑑 = 1, hence 𝑐 and 𝑑 are invertible. In other words, two nonzero
elements 𝑎 and 𝑏 of an integral domain A generate the same ideal if and only if
there exists a unit 𝑢 ∈ A such that 𝑏 = 𝑎𝑢.
The units of the ring A = Z are ±1; it is thus customary to choose,
as a generator of a principal ideal, an positive element. Similarly, the
units of the ring K[X] of polynomials in one indeterminate X and with
coefficients in a field K are the nonzero constant polynomials; we then
rather choose for a generator of a principal ideal, a monic polynomial.
Definition (2.4.2). — One says that a (commutative) ring is a principal ideal
domain if it is a domain and if all of its ideals are principal.
Examples (2.4.3). — a) The ring Z is a principal ideal domain (exam-
ple 1.4.3), as well as the ring K[X] of polynomials in one variable with
coefficients in a (commutative) field K (example 1.4.4).
b) In the ring K[X, Y] of polynomials in two variables with coefficients
in a field K, the ideal (X, Y) is not principal. For if it were generated by
a polynomial P, this polynomial would need to divide both X and Y.
Necessarily, P has to be a nonzero constant. It follows that there exist
Q, R ∈ K[X, Y] such that 1 = XQ(X, Y) + YR(X, Y). This, however, is
absurd, since the right hand term of this equality has not constant term.

2.4.4. Greatest common divisor, least common multiple. — Let A be

a principal ideal domain. Let (𝑎 𝑖 ) be a family of elements of A. By the
assumption on A, the ideal I generated by the (𝑎 𝑖 ) is generated by one
element, say 𝑎. It follows that 𝑑 divides 𝑎 𝑖 for any 𝑖: 𝑑 is a common
divisor of all of the 𝑎 𝑖 . Moreover, if 𝑑0 is a common divisor of the 𝑎 𝑖 , then
𝑎 𝑖 ∈ (𝑑0) for every 𝑖, hence I ⊂ (𝑑0) and 𝑑0 divides 𝑑. One says that 𝑑 is a
greatest common divisor (gcd) of the 𝑎 𝑖 . The word “greatest” has to be
understood in the sense of divisibility: the common divisors of the 𝑎 𝑖 are
exactly the divisors of their gcd. There is in general no preferred choice
of a greatest common divisor, all differ by multiplication by a unit in A.
Let J be the intersection of the ideals (𝑎 𝑖 ) and let 𝑚 be a generator of
the ideal J. For any 𝑖, 𝑚 ∈ (𝑎 𝑖 ), that is, 𝑚 is a multiple of 𝑎 𝑖 for every 𝑖.
84 CHAPTER 2. IDEALS AND DIVISIBILITY

Moreover, if 𝑚 0 ∈ A is a multiple of 𝑎 𝑖 for every 𝑖, then 𝑚 0 ∈ (𝑎 𝑖 ) for

every 𝑖, hence 𝑚 0 ∈ (𝑚) and 𝑚 0 is a multiple of 𝑚. One says that 𝑚 is
a least common multiple (lcm) of the 𝑎 𝑖 . Again, the word “least” has
to be understood in the sense of divisibility. As for the gcd, there is no
preferred choice and all least common multiples differ by multiplication
by a unit in A.
As explained above, when A = Z is the ring of integers, one may
choose for the gcd and the lcm the unique positive generator of the ideal
generated by the 𝑎 𝑖 , resp. of the intersection of the (𝑎 𝑖 ). Then, except for
degenerate cases, 𝑑 is the greatest common divisor and 𝑚 is the least
common (nonzero) multiple in the naïve sense too.
Similarly, when A = K[X] is the ring of polynomials in one indeter-
minate X, it is customary to choose of the gcd and the lcm to be monic
polynomials (or the zero polynomial).

Remark (2.4.5). — Let K be a field, let P, Q ∈ K[X] be polynomials (not

both zero) and let D ∈ K[X] be their gcd, chosen to be monic. Let
P1 , Q1 ∈ K[X] be defined by P = DP1 and Q = DQ1 . Since D is a
generator of the ideal (P, Q), there exist polynomials U and V such that
D = UP + VQ, hence 1 = UP1 + VQ1 and (P1 , Q1 ) = K[X].
It follows that for every field L containing K, one has (P1 , Q1 )L[X] = L[X],
and D is still a gcd of P and Q in the ring L[X].
Stated in the opposite direction, this result states that the monic gcd
of P and Q in L[X] belongs to K[X].

Definition (2.4.6). — Let A be a domain. A map 𝛿 : A {0} → N, is called a

euclidean gauge1 on A if it satisfies the following two properties:
– For any 𝑎, 𝑏 ∈ A {0}, 𝛿(𝑎𝑏) > sup(𝛿(𝑎), 𝛿(𝑏));
– For any 𝑎, 𝑏 ∈ A such that 𝑏 ≠ 0, there exists 𝑞 and 𝑟 ∈ A such that
𝑎 = 𝑏𝑞 + 𝑟 and such that either 𝑟 = 0 or 𝛿(𝑟) < 𝛿(𝑏).
If there exists a euclidean gauge on A, then one says that A is a euclidean ring.

1According for example to Bourbaki and Wedderburn, the official word is stathm.
 2.5. UNIQUE FACTORIZATION DOMAINS 85

Examples (2.4.7). — a) The ring of integers and the ring of polynomials

in one variable with coefficients in a field are euclidean rings, with gauges
given by the usual absolute value and the degree.
b) The ring Z[𝑖] of Gaussian integers is an euclidean ring, with gauge 𝛿
defined by 𝛿(𝑧) = 𝑧 𝑧¯ = |𝑧| 2 (exercise 2/15). See also exercises 2/16, 2/17
and 2/14 for other examples of euclidean rings.

Remark (2.4.8). — Property (i) of gauges implies that 𝛿(𝑎) 6 𝛿(𝑏) when
𝑎 divides 𝑏. Consequently, if 𝑢 is a unit, then 𝛿(𝑎) = 𝛿(𝑎𝑢) for any non
zero element of A.
This property (i) is however not crucial for the definition of a euclidean
ring. Indeed, if 𝛿 is any map satisfying property (ii), one may modify it
in order to get a gauge; see exercise 2/18.

Proposition (2.4.9). — Any euclidean ring is a principal ideal domain.

Proof. — Let indeed A be a euclidean ring with gauge 𝛿, and let I be a

non zero ideal of A. Let 𝑎 be a nonzero element of I such that 𝛿(𝑎) is
minimal among the values of 𝛿 on I {0}. Let 𝑥 ∈ I and let us consider
an euclidean division 𝑥 = 𝑎𝑞 + 𝑟 of 𝑥 by 𝑎; one has 𝑟 = 𝑥 − 𝑎𝑞 ∈ I. If
𝑟 ≠ 0, then 𝛿(𝑟) < 𝛿(𝑎), which contradicts the choice of 𝑎. So 𝑟 = 0 and
𝑥 = 𝑎𝑞 ∈ 𝑎A, hence I = (𝑎). 

Remark (2.4.10). — There exist principal ideal domains which are not
euclidean, for any map 𝛿. One
√
such example is the set of all complex
numbers of the form 𝑎 + 𝑏 1+i2 19 , with 𝑎 and 𝑏 ∈ Z. (See D. Perrin, Cours
d’algèbre, Ellipses, p. 53–55; the proof that this ring is not euclidean can
be found in exercise 2/20.)

2.5. Unique factorization domains

Definition (2.5.1). — Let A be a domain. One says that an element 𝑎 ∈ A is
irreducible if it is not a unit and if the relation 𝑎 = 𝑏𝑐, for some 𝑏 and 𝑐 ∈ A
implies that 𝑏 or 𝑐 is a unit.

Examples (2.5.2). — a) The irreducible elements of Z are the prime

numbers and their opposites.
86 CHAPTER 2. IDEALS AND DIVISIBILITY

b) Let 𝑘 be a field; the irreducible elements of 𝑘[X] are the irreducible

polynomials, that is, the polynomials of degree > 1 which cannot be
written as the product of two polynomials of degree > 1.
c) The element 0 is never irreducible: it can be written 0 × 0 and 0 is
not a unit (A being a domain, 1 ≠ 0).
Proposition (2.5.3) (Gauss’s lemma). — Let A be a principal ideal domain.
For a nonzero ideal of A to be prime, it is necessary and sufficient that it be
generated by an irreducible element; it is then a maximal ideal.
Proof. — Let I be a prime ideal of A; assume that I ≠ 0. Since A is a
principal ideal domain, there exists 𝑎 ∈ A such that I = (𝑎). Since I ≠ 0,
we have 𝑎 ≠ 0; Let us show that 𝑎 is irreducible. Since A is not a prime
ideal, 𝑎 is not a unit. Let 𝑏 and 𝑐 be elements of A such that 𝑎 = 𝑏𝑐. Since
I is a prime ideal, 𝑏 or 𝑐 belongs to I. Assume that 𝑏 ∈ I; then, there exists
𝑢 ∈ A such that 𝑏 = 𝑎𝑢, hence 𝑎 = 𝑎𝑢𝑐 and 𝑐𝑢 = 1 after simplifying by 𝑎.
This shows that 𝑐 is a unit. Similarly, if 𝑐 ∈ I then 𝑏 is a unit. It follows
that 𝑎 is irreducible, as claimed.
Conversely, let 𝑎 be an irreducible element of A and let us show that
the ideal I = (𝑎) is a maximal ideal of A. Let 𝑥 ∈ A be any element which
is not a multiple of 𝑎 and let J = I + (𝑥). Let 𝑏 ∈ A be such that J = (𝑏).
Since 𝑎 ∈ J, there exists 𝑐 ∈ A such that 𝑎 = 𝑏𝑐. If 𝑐 were a unit, one
would have (𝑎) = (𝑏) = I + (𝑥), hence 𝑥 ∈ (𝑎), contrary to the assumption.
Since 𝑎 is irreducible, it follows that 𝑏 is a unit, hence J = A. This shows
that I is a maximal ideal of A. 
The first part of the proof shows, more generally, that in a domain, if a
nonzero prime ideal is generated by one element, then this element is
irreducible. The converse does not hold for general rings and leads to
the notion of a unique factorization domain.
Definition (2.5.4). — Let A be a domain. One says that A is a unique
factorization domain if it satisfies the following two properties:
(i) Every increasing sequence of principal ideals of A is stationary;
(ii) Every ideal generated by an irreducible element is a prime ideal.
The first condition will allow us to write any nonzero element as
the product of finitely many irreducible elements; it is automatic if the
 2.5. UNIQUE FACTORIZATION DOMAINS 87

ring A is noetherian (see definition 6.3.5). The second one is the most
important and will guarantee, up to minor tweeks, the uniqueness of
such a decomposition in irreducible factors.
Let us rewrite this condition somewhat. Let 𝑝 be an irreducible element
of A. Since 𝑝 is not a unit, the ideal (𝑝) is a prime ideal if and only if the
product of two elements of A cannot belong to (𝑝) unless one of them
belongs to 𝑝. In other words, 𝑎𝑏 is divisible by 𝑝 if and only if either 𝑎,
or 𝑏 is divisible by 𝑝.
Examples (2.5.5). — a) A field is a unique factorization domain.
b) Proposition 2.5.3 states that every irreducible element of a principal
ideal domain generate a maximal ideal. Given lemma 2.5.6 below, we
conclude that principal ideal domains are unique factorization domains.
c) We will prove (corollary 2.6.7) that for any unique factorization
domain A, the ring A[X1 , . . . , X𝑛 ] of polynomials with coefficients in A
is also a unique factorization domain. In particular, polynomial rings
with coefficients in a field, or in Z, are unique factorization domains.
Lemma (2.5.6). — In a principal ideal domain, any increasing sequence of
ideals is stationary. 2
Proof. — Let A be a principal ideal domain, let (I𝑛 ) be an increasing
sequence of ideals. Let I be the union of all ideals I𝑛 . Since the sequence
is increasing, I is again an ideal of A. Since A is a principal ideal domain,
there exists 𝑎 ∈ I such that I = (𝑎). Let then 𝑚 ∈ N be such that 𝑎 ∈ I𝑚 .
For 𝑛 > 𝑚, we have I = (𝑎) ⊂ I𝑛 ⊂ I, hence the equality I𝑛 = I: the
sequence (I𝑛 ) is stationary. 

Theorem (2.5.7). — Let A be a unique factorization domain and let 𝑎 be any

nonzero element of A.
a) There exist an integer 𝑛 > 0, irreducible elements 𝑝 1 , . . . , 𝑝 𝑛 ∈ A, and
a unit 𝑢 ∈ A, such that 𝑎 = 𝑢𝑝 1 . . . 𝑝 𝑛 (existence of a decomposition in
irreducible factors).
b) Let us consider two such decompositions of 𝑎, say 𝑎 = 𝑢𝑝1 . . . 𝑝 𝑛 =
𝑣𝑞 1 . . . 𝑞 𝑚 . Then 𝑛 = 𝑚 and there exists a permutation 𝜎 of {1, . . . , 𝑛} and
2In other terms, a principal ideal domain is a noetherian ring (definition 6.3.5).
88 CHAPTER 2. IDEALS AND DIVISIBILITY

units 𝑢𝑖 , for 1 6 𝑖 6 𝑛, such that 𝑞 𝑖 = 𝑢𝑖 𝑝 𝜎(𝑖) for every 𝑖 (uniqueness of a

decomposition in irreducible factors).

This is often taken as the definition of a unique factorization domain;

in any case, this explains the chosen denomination!
Let us comment a little bit the use of the word “uniqueness” in
the theorem. Strictly speaking, there is no unique decomposition in
irreducible factors; indeed, it is always possible to change the order
of the factors, or simultaneously multiply some irreducible factor and
dividing the unit in front of the decomposition by the same unit. The
content of the uniqueness property is that these are the only two ways
by which two decompositions may differ.
Proof. — When 𝑎 is invertible, property a) is obvious: take 𝑛 = 0 and
𝑢 = 𝑎. Otherwise, there exists a maximal ideal containing 𝑎, hence
an irreducible element 𝑝1 ∈ A such that 𝑝 1 divides 𝑎. Let 𝑎1 ∈ A be
such that 𝑎 = 𝑝1 𝑎1 ; we then have (𝑎) ( (𝑎1 ). If 𝑎1 is not invertible, we
may redo the argument, obtaining irreducible elements 𝑝1 , . . . , 𝑝 𝑛 of A,
and elements 𝑎1 , . . . , 𝑎 𝑛 ∈ A such that 𝑎 = 𝑝1 . . . 𝑝 𝑛 𝑎 𝑛 . Might we go
on forever, we obtain a strictly increasing sequence of principal ideals
(𝑎) ( (𝑎1 ) ( (𝑎 2 ) ( . . . , which contradicts the first axiom of a unique
factorization domain. So 𝑎 𝑛 is a unit for some 𝑛, and property a) is
proved.
Let now 𝑎 = 𝑢𝑝1 . . . 𝑝 𝑛 = 𝑣𝑞 1 . . . 𝑞 𝑚 be two decompositions of 𝑎 as the
product of a unit and of irreducible elements. Let us prove property b)
by induction on 𝑚. If 𝑚 = 0, then 𝑎 = 𝑣 is a unit; consequently, 𝑢𝑝 1 . . . 𝑝 𝑛
is a unit too, which implies 𝑛 = 0 and 𝑎 = 𝑢 = 𝑣 (indeed, if 𝑛 > 1,
𝑢𝑝1 . . . 𝑝 𝑛 belongs to the maximal ideal (𝑝 1 ) so cannot be a unit). Assume
𝑚 > 1. Since 𝑞 𝑚 divides 𝑢𝑝 1 . . . 𝑝 𝑚 and the ideal (𝑞 𝑚 ) is prime, 𝑞 𝑚
divides one of the factors 𝑢, 𝑝1 , . . . , 𝑝 𝑛 . Since 𝑢 is a unit, there exists
an integer 𝑗 ∈ {1, . . . , 𝑛} such that 𝑞 𝑚 divides 𝑝 𝑗 ; let 𝑠 ∈ A be such that
𝑝 𝑗 = 𝑠𝑞 𝑚 . Since 𝑝 𝑗 is irreducible and 𝑞 𝑚 is not a unit, 𝑠 is necessarily a
unit; we then set 𝑢𝑚 = 𝑠 −1 so that 𝑞 𝑚 = 𝑢𝑚 𝑝 𝑗 . Set 𝑏 = 𝑎/𝑞 𝑚 . It admits two
decompositions as a product of irreducible elements, namely 𝑣𝑞 1 . . . 𝑞 𝑚−1
and (𝑢/𝑢𝑚 )𝑝 1 . . . 𝑝b𝑗 . . . 𝑝 𝑛 , where the hat on 𝑝 𝑗 indicates that this factor is
omitted from the product. By induction, we have 𝑚 − 1 = 𝑛 − 1 and there
exist a bijection 𝜎 from {1, . . . , 𝑚 − 1} to {1, . . . , b 𝑗, . . . , 𝑛}, and units 𝑢𝑖 ,
 2.5. UNIQUE FACTORIZATION DOMAINS 89

such that 𝑞 𝑖 = 𝑢𝑖 𝑝 𝜎(𝑖) for 1 6 𝑖 6 𝑚 − 1. It follows that 𝑚 = 𝑛 and

the mapping (still denoted by 𝜎) from {1, . . . , 𝑚} to {1, . . . , 𝑛} which
extends 𝜎 and maps 𝑚 to 𝑗 is a bijection. For every 𝑖 ∈ {1, . . . , 𝑛}, one has
𝑞 𝑖 = 𝑢𝑖 𝑝 𝜎(𝑖) . This concludes the proof of the uniqueness property. 

Remark (2.5.8). — Conversely, let A be a domain which satisfies the conclusion of the
theorem.
For any non zero 𝑎 ∈ A, let 𝜔(𝑎) be the number of irreducible factors in a decom-
position of 𝑎 as a product of irreducible elements. It does not depend on the chosen
decomposition. Let 𝑎, 𝑏, 𝑐 be elements of A such that 𝑎 = 𝑏𝑐 and 𝑎 ≠ 0. The uniqueness
property implies that 𝜔(𝑎) = 𝜔(𝑏) + 𝜔(𝑐).
Let (𝑎 𝑛 ) be a sequence of elements of A such that the sequence of ideals (𝑎 0 ), (𝑎 1 ), . . .
is increasing. For every integer 𝑛, there thus exists 𝑏 𝑛 ∈ A such that 𝑎 𝑛 = 𝑏 𝑛 𝑎 𝑛+1 .
Consequently, 𝜔(𝑎 𝑛+1 ) 6 𝜔(𝑎 𝑛 ), so that the sequence (𝜔(𝑎 𝑛 ))𝑛 is a decreasing sequence
of positive integers; it is thus stationary. Moreover, if 𝜔(𝑎 𝑛 ) = 𝜔(𝑎 𝑛+1 ), then 𝑏 𝑛 is a unit
and the ideals (𝑎 𝑛 ) and (𝑎 𝑛+1 ) coincide. This implies that the sequence (𝑎 𝑛 ) of ideals is
itself stationary.
Let 𝑝 be any irreducible element of A. Let us show that it generates a prime ideal
of A. Since 𝑝 is irreducible, 𝑝 is not a unit and (𝑝) ≠ A. Let then 𝑎, 𝑏 be elements
of A such that 𝑝 divides 𝑎𝑏; let 𝑐 ∈ A be given by 𝑎𝑏 = 𝑝𝑐. If 𝑐 = 0, then 𝑝𝑐 = 0
and either 𝑎 or 𝑏 is zero. Let us assume that 𝑐 ≠ 0; then 𝑎 ≠ 0 and 𝑏 ≠ 0. Pickup
decompositions of 𝑎, 𝑏 and 𝑐 as products of irreducible factors, say 𝑎 = 𝑢𝑝1 . . . 𝑝 𝑛 ,
𝑏 = 𝑣𝑞 1 . . . 𝑞 𝑚 , and 𝑐 = 𝑤𝑟1 . . . 𝑟 𝑠 . (Here, 𝑢, 𝑣, 𝑤 are units, 𝑛, 𝑚, 𝑠 are positive integers,
while the 𝑝 𝑖 , 𝑞 𝑗 and 𝑟 𝑘 are irreducible elements of A.) Then we have two decompositions
of 𝑎𝑏, namely 𝑢𝑣𝑝 1 . . . 𝑝 𝑛 𝑞 1 . . . 𝑞 𝑚 and 𝑤𝑝𝑟1 . . . 𝑟 𝑠 . By the uniqueness property, the
factor 𝑝 which appears in the second one has to intervene in the first one; precisely,
there exists a unit 𝛼 ∈ A and either an integer 𝑖 ∈ {1, . . . , 𝑛} such that 𝑝 = 𝛼𝑝 𝑖 , or an
integer 𝑗 ∈ {1, . . . , 𝑚} such that 𝑝 = 𝛼𝑞 𝑗 . In the first case, 𝑝 divides 𝑎, in the second
one, 𝑝 divides 𝑏. This shows that (𝑝) is a prime ideal.
These remarks prove that the ring A is a unique factorization domain.

Remark (2.5.9). — One of the reasons of the nonuniqueness of the decom-

position into irreducible factors is that one may multiply those factors
by units.
In certain rings, it is possible to distinguish privileged irreducible
elements so as to remove this source of ambiguity.
For example, in Z, the irreducible elements are the prime numbers
and their opposites, but one may decide to prefer the prime numbers
themselves. Up to the order of the factors, any nonzero integer is then
uniquely written as the product of ±1 (the only units in Z) by a product
of prime numbers.
90 CHAPTER 2. IDEALS AND DIVISIBILITY

Similarly, in the ring K[T] in one indeterminate T over a field K one may
prefer the monic irreducible polynomials. Still up to the order of factors,
any nonzero polynomial can be uniquely written as the product of a
nonzero constant (the units in K[T]) by a product of monic irreducible
polynomials.
In a general unique factorization domain A, let us show how to
normalize the decomposition in irreducible elements so that it only can
be modified by the order of the factors.
Let us choose a family (𝜋 𝑖 ) of irreducible elements of A in such a way
that
– For 𝑖 ≠ 𝑗, 𝜋 𝑖 and 𝜋 𝑗 are not associated;
– Every irreducible element of A is associated to one of the 𝜋 𝑖 .
(To prove the existence of such a family, just choose one element in every
equivalence class of irreducible elements for the equivalence relation of
being associated.) Then, any nonzero element 𝑎 of A can be uniquely
written on the form 𝑎 = 𝑢 𝑖 𝜋𝑟𝑖 𝑖 where 𝑢 is a unit and the 𝑟 𝑖 are positive
Î
integers, all but finitely many of them being equal to zero.
In other words, the map from A× × N(I) to A {0} which maps (𝑢, (𝑟 𝑖 ))
to the product 𝑢 𝑖 𝜋𝑟𝑖 𝑖 is an isomorphism of monoids.
Î
One interesting aspect of this normalization is that it makes the
divisibility relation explicit: an element 𝑎 = 𝑢 𝑖 𝜋𝑟𝑖 𝑖 divides an element
Î
𝑏 = 𝑣 𝑖 𝜋 𝑠𝑖 𝑖 if and only if 𝑟 𝑖 6 𝑠 𝑖 for every 𝑖. Indeed, it is clear that
Î
this condition is sufficient, for it suffices to set 𝑐 = (𝑣𝑢 −1 ) 𝑖 𝜋 𝑠𝑖 𝑖 −𝑟𝑖 to get
Î

𝑎 = 𝑏𝑐. Conversely, if 𝑐 ∈ A is such that 𝑏 = 𝑎𝑐, decompose 𝑐 as 𝑤 𝑖 𝜋𝑡𝑖 𝑖 ;

Î
then,
𝜋 𝑠𝑖 𝑖 = 𝑢𝑤 𝜋𝑟𝑖 𝑖 +𝑡 𝑖 ,
Ö Ö
𝑏=𝑣
𝑖 𝑖
hence 𝑠 𝑖 = 𝑟 𝑖 + 𝑡 𝑖 for every 𝑖. This implies 𝑠 𝑖 > 𝑟 𝑖 .

2.5.10. Greatest common divisor, least common multiple. — Let A

be a unique factorization domain.
Let (𝑎 𝑛 ) be a family of elements of A. We are going to show that it
possesses a greatest common divisor (gcd) and a least common multiple (lcm).
As in the case of principal ideal domains, a gcd of the family (𝑎 𝑛 ) is any
element 𝑑 ∈ A which divides each of the 𝑎 𝑛 and such that every such
 2.5. UNIQUE FACTORIZATION DOMAINS 91

common divisor divides 𝑑, and a lcm of this family is an element 𝑚 ∈ A

which is a multiple of all of the 𝑎 𝑛 and such that every such common
multiple is a multiple of 𝑚.
We first treat particular, essentially trivial, cases. Remark that every
element of A divides 0, but that the only multiple of 0 is 0 itself.
Consequently, if one of the 𝑎 𝑛 is equal to 0, it has only one common
multiple, namely 0, which thus is its lcm. On the other hand, in order to
show that the family (𝑎 𝑛 ) has a gcd, we may remove all the terms equal
to 0.
If the family is empty, then every element is a common divisor, so that
0 is a greatest common divisor of the empty family. Similarly, every
element is a common multiple, so that 1 is least common multiple of the
empty family.
These remarks allow to assume that the family (𝑎 𝑛 ) is nonempty and
consists of nonzero elements. To simplify the construction, assume
also that we have normalized as above the decomposition in irreducible
Î 𝑟
factors. For every 𝑛, let 𝑎 𝑛 = 𝑢𝑛 𝑖 𝜋 𝑖 𝑛,𝑖 be the decomposition of 𝑎 𝑛 into
irreducible factors. For every 𝑖, set 𝑑 𝑖 = inf𝑛 (𝑟𝑛,𝑖 ); this is a positive integer,
and 𝑑 𝑖 = 0 for all but finitely many 𝑖. This allows to set 𝑑 = 𝑖 𝜋 𝑑𝑖 𝑖 .
Î
Let us show that 𝑑 is a gcd of the family (𝑎 𝑛 ). Since 𝑑 𝑖 6 𝑟𝑛,𝑖 for
every 𝑖, 𝑑 divides 𝑎 𝑛 for every 𝑛. Let 𝑏 be a common divisor ot the 𝑎 𝑛 ,
let 𝑏 = 𝑣 𝑖 𝜋 𝑠𝑖 𝑖 be its decomposition in irreducible factors. Since 𝑏
Î
divides 𝑎 𝑛 , 𝑠 𝑖 6 𝑟𝑛,𝑖 for every 𝑖. It follows that 𝑠 𝑖 6 𝑑 𝑖 for every 𝑖, hence
𝑏 divides 𝑑.
For every 𝑖, set 𝑚 𝑖 = sup(𝑟𝑛,𝑖 ); this is an element of N ∪ {+∞}. If
every 𝑚 𝑖 is an integer, and if all but finitely many of them are 0, we may
set 𝑚 = 𝑖 𝜋𝑚 𝑖 ; otherwise, we set 𝑚 = 0. In each case, 𝑚 is a common
𝑖
Î
multiple of all of the 𝑎 𝑛 . Conversely, let 𝑏 be any nonzero element of A;
let 𝑏 = 𝑣 𝑖 𝜋 𝑠𝑖 𝑖 be its decomposition into irreducible elements. For 𝑏
Î
to be a multiple of 𝑎 𝑛 , it is necessary and sufficient that 𝑠 𝑖 > 𝑟𝑛,𝑖 for
every 𝑖; consequently, 𝑏 is a common multiple of the family (𝑎 𝑛 ) if and
only if 𝑠 𝑖 > 𝑚 𝑖 for every 𝑖. If 𝑚 𝑖 is infinite for some 𝑖, of if infinitely many
terms of the family (𝑚 𝑖 ) are nonzero, this never holds, so that 0 is the
only common multiple to the family (𝑎 𝑛 ). Otherwise, we see that 𝑏 is a
multiple of 𝑚.
92 CHAPTER 2. IDEALS AND DIVISIBILITY

Remark (2.5.11). — Unless they are zero, two greatest common divisors
(resp. two least common multiples) of a family (𝑎 𝑛 ), differ by mul-
tiplication by a unit. As we have seen above, choosing a particular
normalization for the decomposition in irreducible factors allows to get
a well-defined representative of the gcd (resp. of the lcm).
When we shall write equalities involving greatest common divisors or
least common multiples, we shall always assume that they are properly
normalized. In any case, it is always possible to read these equalities up
to multiplication by a unit.

Definition (2.5.12). — One says that a family of elements of a unique factor-

ization domain consists of coprime elements if this family has 1 for a greatest
common divisor.

Remark (2.5.13). — Let (𝑎 𝑛 ) be a family of elements of A. For every 𝑥 ∈ A,

gcd((𝑥𝑎 𝑛 )) = 𝑥 gcd(𝑎 𝑛 ). This follows easily from the construction above;
thanks to the formula inf(𝑟𝑛 )+𝑠 = inf(𝑟𝑛 +𝑠) for any family (𝑟𝑛 ) of integers
and any integer 𝑠. Conversely, let 𝑑 = gcd((𝑎 𝑛 )). For every 𝑛, there is
𝑏 𝑛 ∈ A such that 𝑎 𝑛 = 𝑑𝑏 𝑛 . Then 𝑑 gcd((𝑏 𝑛 )) = gcd((𝑎 𝑛 )). Assume that
at least one of the 𝑎 𝑛 is nonzero. Then 𝑑 ≠ 0 hence gcd((𝑏 𝑛 )) = 1, so that
the 𝑏 𝑛 are coprime.

Proposition (2.5.14). — Let A be a unique factorization domain and let (𝑎 𝑛 ) be

a family of elements of A, with gcd 𝑑 and lcm 𝑚. The ideal 𝑑A is the smallest
principal ideal containing the ideal 𝑎 𝑛 A, and the ideal 𝑚A is the largest
Í
principal ideal containing the ideal 𝑛 𝑎 𝑛 A.
Ñ
Assume in particular that A is a principal ideal domain. Then two elements 𝑎
and 𝑏 of A are coprime if and only if the ideals (𝑎) and (𝑏) are comaximal.

Proof. — Since the inclusion of ideals 𝑎A ⊂ 𝑏A is equivalent to the fact

that 𝑏 divides 𝑎, this is but a reformulation of the discussion above. 

Corollary (2.5.15) (Bézout’s theorem). — Let A be a principal ideal domain,

let (𝑎 𝑛 ) be a family of elements of A with gcd 𝑑. There exist elements 𝑢𝑛 ∈ A,
all but finitely many of which are zero, such that 𝑑 = 𝑎 𝑛 𝑢𝑛 .
Í

In a euclidean ring, there is a simple algorithm to compute the gcd of

two elements 𝑎 and 𝑏, as well as a relation 𝑑 = 𝑎𝑢 + 𝑏𝑣.
 2.5. UNIQUE FACTORIZATION DOMAINS 93

Proposition (2.5.16) (Euclid’s algorithm). — Let A be an euclidean ring; let

𝑎, 𝑏 be elements of A. One defines four sequences (𝑑𝑛 ), (𝑢𝑛 ), (𝑣 𝑛 ) and (𝑞 𝑛 ) by
induction on 𝑛 by setting
𝑑0 = 𝑎 𝑢0 = 1 𝑣0 = 0
𝑑1 = 𝑏 𝑢1 = 0 𝑣1 = 1

and then, if 𝑑𝑛 ≠ 0, let 𝑞 𝑛 be the quotient of a euclidean division of 𝑑𝑛−1 by 𝑑𝑛 ,

and

𝑑𝑛+1 = 𝑑𝑛−1 − 𝑞 𝑛 𝑑𝑛 𝑢𝑛+1 = 𝑢𝑛−1 − 𝑞 𝑛 𝑢𝑛 𝑣 𝑛+1 = 𝑣 𝑛−1 − 𝑞 𝑛 𝑣 𝑛 .

These sequences are finite: there exists a smallest integer 𝑛 such that 𝑑𝑛+1 = 0;
then, 𝑑𝑛 = 𝑎𝑢𝑛 + 𝑏𝑣 𝑛 is a greatest common divisor of 𝑎 and 𝑏.

Proof. — Let 𝛿 be the gauge of A. If 𝑑𝑛 ≠ 0, then 𝑑𝑛+1 is the remainder

of an euclidean division by 𝑑𝑛 , so that either 𝛿(𝑑𝑛+1 ) < 𝛿(𝑑𝑛 ), or 𝑑𝑛+1 =
0. Consequently, the sequence of positive integers (𝛿(𝑑𝑛 )) is strictly
decreasing as soon as it is defined. Let us also remark that the pairs (𝑏, 𝑎 −
𝑏𝑞) and (𝑎, 𝑏) have the same common divisors, so that gcd(𝑏, 𝑎 − 𝑏𝑞) =
gcd(𝑎, 𝑏). By induction, gcd(𝑑𝑛−1 , 𝑑𝑛 ) = gcd(𝑑𝑛 , 𝑑𝑛+1 ) for every 𝑛. If 𝑑𝑛 ≠
0 and 𝑑𝑛+1 = 0, we then have gcd(𝑎, 𝑏) = gcd(𝑑0 , 𝑑1 ) = gcd(𝑑𝑛 , 𝑑𝑛+1 ) = 𝑑𝑛 .
By induction, the proof of the relation 𝑑𝑛 = 𝑎𝑢𝑛 + 𝑏𝑣 𝑛 is immediate. 
The following result is very useful.

Proposition (2.5.17) (Gauss). — Let A be a unique factorization domain and

let 𝑎, 𝑏, 𝑐 ∈ A.
a) If 𝑎 is prime to 𝑏 and 𝑐, then 𝑎 is prime to 𝑏𝑐.
b) If 𝑎 is prime to 𝑐, and if 𝑎 divides 𝑏𝑐, then 𝑎 divides 𝑏.

Proof. — a) Let 𝑝 be an irreducible element of A that divides 𝑎. By

assumption, 𝑝 does not divide 𝑏, and 𝑝 does not divide 𝑐. Since A is a
unique factorization domain, 𝑝 does not divide 𝑏𝑐. This proves that 𝑎 is
prime to 𝑏𝑐.
b) The result is obvious if 𝑐 is a unit. We then argue by induction on
the number of irreducible factors of 𝑐. Let 𝑝 be a prime number that
divides 𝑐; write 𝑐 = 𝑝𝑐 0. Then 𝑎 is prime to 𝑐 0 and 𝑎 divides 𝑏𝑝𝑐 0 𝑐, hence
𝑎 divides 𝑏𝑝 by induction. Let 𝑑 ∈ A be such that 𝑏𝑝 = 𝑎𝑑. Since A is a
94 CHAPTER 2. IDEALS AND DIVISIBILITY

unique factorization domain, the ideal (𝑝) is prime. one has 𝑎 ∉ (𝑝) by
assumption, and 𝑎𝑑 ∈ (𝑝); consequently, 𝑑 ∈ (𝑝). Let 𝑑0 ∈ A be such that
𝑑 = 𝑝𝑑0; then 𝑏𝑝 = 𝑝𝑎𝑑0, hence 𝑏 = 𝑎𝑑0: this shows that 𝑎 divides 𝑏. 

2.6. Polynomial rings are unique factorization domains

One of the most important and basic results in the theory of unique
factorization domains is the theorem of Gauss according to which
polynomial rings with coefficients in a unique factorization domain are
themselves unique factorization domains.
So let A be a unique factorization domain. We first recall that the
units of A[T] are the units of A× , viewed as constant polynomials
(corollary 1.3.14).
Definition (2.6.1). — Let A be a unique factorization domain and let P be any
polynomial in A[T]. The content of P, denoted by ct(P), is defined as a greatest
common divisor of the coefficients of P. One says that a polynomial is primitive
if its content is a unit, that is to say, if its coefficients are coprime.
As usual for questions of gcd, the content of a nonzero polynomial is
only well defined if we have normalized the decomposition in irreducible
factors; otherwise, it is defined up to multiplication by a unit. The content
of the zero polynomial is 0.
Lemma (2.6.2). — Let A be a unique factorization domain, let K be its field of
fractions.
a) For any polynomial P ∈ K[T], there exist a primitive polynomial P1 ∈ A[T]
and an element 𝑎 ∈ K such that P = 𝑎P1 .
b) Let P = 𝑎P1 be such a decomposition. Then, P ∈ A[T] if and only if
𝑎 ∈ A; in that case, 𝑎 = ct(P). In particular, P is a primitive polynomial
of A[T] if and only if 𝑎 is a unit in A.
Proof. — a) If P = 0, we set 𝑎 = 0 and P1 = 1. Assume P ≠ 0. Let
𝑑 ∈ A {0} be a common denominator to all of the coefficients of P, so
that 𝑑P ∈ A[T]. Let then 𝑏 ∈ A be the content of the polynomial 𝑑P and
set P1 = (𝑑P)/𝑏 and 𝑎 = 𝑏/𝑑. The polynomial P1 belongs to A[T] and is
primitive; one has P = 𝑎P1 .
 2.6. POLYNOMIAL RINGS ARE UNIQUE FACTORIZATION DOMAINS 95

b) If 𝑎 ∈ A, it is clear that P ∈ A[T]. Conversely, assume that P ∈ A[T]

and let us show that 𝑎 ∈ A. Let 𝑏 and 𝑐 be elements of A such that
𝑎 = 𝑏/𝑐. We write 𝑏 = 𝑎𝑐, hence 𝑏P1 = 𝑎𝑐P1 = 𝑐P, from which we deduce
that 𝑏 = ct(𝑐P) = 𝑐 ct(P). It follows that 𝑐 divides 𝑏 and 𝑎 = ct(P) ∈ A.
If 𝑎 is a unit in A, then P is a primitive polynomial in A[T]. Conversely,
assume that P is a primitive polynomial in A[T]. By the preceding
paragraph, we have 𝑎 ∈ A. Then ct(P) = 𝑎 ct(P1 ) is a unit, hence 𝑎 is a
unit. 

Proposition (2.6.3). — Let A be a unique factorization domain and let P, Q be

two polynomials in A[T]. Then, ct(PQ) = ct(P) ct(Q).

Proof. — We first treat the particular case where P and Q are primitive.
We then need to show that PQ is primitive as well. Let 𝜋 be any
irreducible element of A and let us show that 𝜋 does not divide all of the
coefficients of PQ. Since P is primitive, the reduction cl(P) of P modulo 𝜋
is a nonzero polynomial with coefficients in the ring A/(𝜋). Similarly,
cl(Q) is a nonzero polynomial with coefficients in A/(𝜋). Since 𝜋 is
irreducible and A a unique factorization domain, the quotient ring A/(𝜋)
is a domain, hence the polynomial ring (A/(𝜋))[T] is again a domain
(corollary 1.3.13). It follows that the product cl(P) cl(Q) = cl(PQ) is a
nonzero polynomial in (A/𝜋)[T]. This means exactly that 𝜋 does not
divide all of the coefficients of PQ, as was to be shown.
Let us now treat the general case. We may assume that P and Q
are nonzero, so that ct(P) and ct(Q) are nonzero too. By definition,
we can write P = ct(P)P1 and Q = ct(Q)Q1 , where P1 , Q1 ∈ A[T]
are primitive polynomials. Then, P1 Q1 is primitive and the equality
PQ = ct(P) ct(Q)P1 Q1 shows that, up to a unit, ct(PQ) is thus equal to
ct(P) ct(Q). 

Corollary (2.6.4). — Let A be a unique factorization domain and let K be its

field of fractions. Let P, Q ∈ A[T] be polynomials.
a) Assume that Q is primitive and that Q divides P in K[T]. Then Q
divides P in A[T].
b) Let R = gcd(P, Q) be a greatest common divisor of P, Q. Then R is a gcd
of P and Q in K[T].
96 CHAPTER 2. IDEALS AND DIVISIBILITY

Proof. — a) Let R ∈ K[T] be such that P = RQ. Write R = 𝑎R1 , where

𝑎 ∈ K and R1 ∈ A[T] is a primitive polynomial. Then P = 𝑎R1 Q. Since
R1 Q is primitive, we have 𝑎 ∈ A, hence R ∈ A[T], which shows that Q
divides P in A[T].
b) Let S be a gcd of P and Q in K[T]. Of course, R divides P and Q
in K[T], so that R divides S. Conversely, we need to prove that S divides R.
Writing S = 𝑎S1 , with 𝑎 ∈ K and S1 ∈ A[T] primitive, we assume that
S ∈ A[T] is primitive. By a), S divides P and Q in A[T], so that it divides R.
This concludes the proof. 
Thanks to this fundamental proposition, we are now able to determine
the irreducible elements of A[T].

Proposition (2.6.5). — Let A be a unique factorization domain and let K be its

field of fractions. The irreducible elements of A[T] are the following:
– Irreducible elements of A, considered as constant polynomials;
– Primitive polynomials in A[T] which are irreducible as polynomials in K[T].

Proof. — We shall begin by proving that these elements are indeed

irreducible in A[T], and then show that there are no other.
Let 𝑎 ∈ A be an irreducible element. It is not invertible in A, hence is
not a unit of A[T]. Let P, Q be polynomials in A[T] be such that 𝑎 = PQ.
One has deg(P) + deg(Q) = deg(PQ) = 0, hence deg(P) = deg(Q) = 0. In
other words, P and Q are constant polynomials. Since 𝑎 is irreducible,
either P or Q is a unit in A, hence in A[T]. This shows that 𝑎 is irreducible
in A[T].
Let now P ∈ A[T] be a primitive polynomial which is irreducible
in K[T]. Since constant polynomials are units in K[T], P is not constant;
in particular, it is not a unit in A[T]. Let Q, R be polynomials in A[T]
such that P = QR. A fortiori, they furnish a decomposition of P in K[T],
so that Q or R is a unit in K[T]. In particular, Q or R is constant. To fix
the notation, assume that R is the constant 𝑎. We thus have P = 𝑎Q. It
follows that the content of P satisfies
ct(P) = ct(𝑎Q) = 𝑎 ct(Q).
Since P is primitive, 𝑎 is a unit in A, hence in A[T]. This shows that P is
irreducible in A[T].
 2.6. POLYNOMIAL RINGS ARE UNIQUE FACTORIZATION DOMAINS 97

Conversely, let P be an irreducible element of A[T]. Let P1 be a

primitive polynomial in A[T] such that P = ct(P)P1 . Necessarily, ct(P) is
a unit in A, or P1 is a unit in A[T].
Assume first that ct(P) is not a unit. Then, P1 is a unit in A[T], which
means that P1 is a constant polynomial and a unit in A. In other words, P
is an element 𝑎 of A. Let us show that 𝑎 is irreducible in A. It is not a unit
(otherwise, P would be a unit). And if 𝑎 = 𝑏𝑐, for some elements 𝑏, 𝑐 ∈ A,
we get P = 𝑏𝑐. Since P is irreducible in A[T], 𝑏 or 𝑐 is a unit in A[T], that
is, 𝑏 or 𝑐 is a unit in A.
Assume now that P1 is not a unit in A[T]; then ct(P) is a unit and P
is primitive. Let us prove that P is irreducible in K[T]. , First of all,
deg(P) > 0, for otherwise, P would be a unit in A. In particular, P is a not
a unit in K[T]. Let P = QR be a factorization of P as the product of two
polynomials in K[T]. By the above lemma 2.6.2, we may write Q = 𝑞Q1
and R = 𝑟R1 , where 𝑞, 𝑟 ∈ K and Q1 , R1 are primitive polynomials
in A[T]. Then, P = (𝑞𝑟)Q1 R1 . Since P is a primitive polynomial in A[T],
lemma 2.6.2 implies that 𝑞𝑟 is a unit in A. Since P is irreducible in A[T],
either Q1 or R1 is a unit in A[T], hence in K[T]. It follows that either
Q = 𝑞Q1 or R = 𝑟R1 is a unit in K[T]. 

Theorem (2.6.6) (Gauss). — If A is a unique factorization domain, then so is

A[T].

Proof. — Let us prove that A[T] satisfies the two properties of the
definition 2.5.4 of a unique factorization domain.
1) Let (P𝑛 )𝑛 be a sequence of polynomials in A[T] such that the
sequence ((P𝑛 ))𝑛 of principal ideals is increasing; let us prove that it is
stationary.
If P𝑛 ≠ 0, then P𝑚 ≠ 0 for all 𝑛 > 𝑚. The case of the constant sequence
((0)) being obvious, we may assume that P𝑛 ≠ 0 for all 𝑛.
For 𝑛 > 𝑚, P𝑛 divides P𝑚 , hence ct(P𝑛 ) divides ct(P𝑚 ). Consequently,
the sequence ((ct(P𝑛 )))𝑛 of principal ideals of A is increasing too. Since
A is a unique factorization domain, it is stationary. Moreover, for
𝑛 > 𝑚, deg(P𝑛 ) 6 deg(P𝑚 ), so that the sequence ((deg(P𝑛 ))𝑛 of integers
is decreasing, hence stationary.
98 CHAPTER 2. IDEALS AND DIVISIBILITY

Let N be any integer such that deg(P𝑛 ) = deg(PN ) and ct(P𝑛 ) = ct(PN )
for 𝑛 > N. Let 𝑛 be some integer such that 𝑛 > N. Since P𝑛 divides PN ,
there exists a polynomial Q such that PN = QP𝑛 . Necessarily, deg(Q) =
0 and ct(Q) = 1, so that Q is a constant polynomial, with constant
term ct(Q), hence is invertible. Consequently, the ideals (P𝑛 ) and (PN )
coincide. This shows that the sequence (P𝑛 )𝑛 of principal ideals of A[T]
is stationary.
2) Let us now show that the irreducible elements of A[T] generate
prime ideals. Since irreducible elements are not units, it suffices to
show that if an irreducible element of A[T] divides a product PQ of two
polynomials in A[T], then it divides P or Q.
Let first 𝜋 be an irreducible element of A; assume that 𝜋 divides PQ.
Taking the contents, we see that 𝜋 divides ct(PQ) = ct(P) ct(Q). Since 𝜋
is irreducible in A and A is a unique factorization domain, 𝜋 divides
ct(P) or ct(Q), hence 𝜋 divides P or Q.
Let now Π be a primitive polynomial of A[T], irreducible in K[T], such
that Π divides PQ. Since K[T] is a principal ideal domain, it is a unique
factorization domain and Π divides P or Q in K[T]. By corollary 2.6.4, Π
divides P or Q in A[T]. 

Corollary (2.6.7) (Gauss). — Let A be a unique factorization domain. For any

integer 𝑛, A[X1 , . . . , X𝑛 ] is a unique factorization domain. In particular, if K
is a field, then K[X1 , . . . , X𝑛 ] is a unique factorization domain.
Proof. — This is immediate by induction on 𝑛, using the isomorphisms
(remark 1.3.10):
A[X1 , . . . , X𝑚 ] ' (A[X1 , . . . , X𝑚−1 ])[X𝑚 ].


2.7. Resultants and another theorem of Bézout

In all of this section, A is a commutative ring. We shall make some
use of determinants of matrices.
Definition (2.7.1). — Let 𝑚, 𝑛 be positive integers, let P and Q be two polynomi-
als in A[X] such that deg(P) 6 𝑛 and deg(Q) 6 𝑚. Write P = 𝑎 𝑛 X𝑛 + · · · + 𝑎0
 2.7. RESULTANTS AND ANOTHER THEOREM OF BÉZOUT 99

and Q = 𝑏 𝑚 X𝑚 + · · · + 𝑏 0 , for 𝑎0 , . . . , 𝑎 𝑛 , 𝑏0 , . . . , 𝑏 𝑚 ∈ A. The resultant (in

sizes (𝑛, 𝑚)) of (P, Q) is defined as the determinant

𝑎0 0 𝑏0 0
𝑎1 𝑎0 𝑏1 𝑏0
.. ... .. ...
. .
𝑎 𝑚−1 𝑎0 𝑏 𝑚−1 𝑏0
.. .. 𝑏𝑚 ...
. .
Res𝑛,𝑚 (P, Q) = .. .. ... 𝑏0
. .
𝑎 𝑛 𝑎 𝑛−1 𝑎 𝑛−𝑚+1 ..
𝑏𝑚 .
.. ... ..
𝑎𝑛 . .
... .. . . . ...
.
0 𝑎𝑛 0 𝑏𝑚
| {z }| {z }
𝑚 columns 𝑛 columns

(Precisely, the column vector (𝑎0 , . . . , 𝑎 𝑛 ) is copied 𝑚 times, each time

shifted by one row, then the column vector (𝑏 0 , . . . , 𝑏 𝑚 ) is copied 𝑛 times,
each time shifted by one row.)

Remark (2.7.2). — Let P and Q ∈ A[X] be polynomials as above, and let

𝑓 : A → B be a morphism of rings. Let P 𝑓 and Q 𝑓 be the polynomials
in B[X] deduced from P and Q by applying 𝑓 to their coefficients.
One has deg(P 𝑓 ) 6 𝑛 and deg(Q 𝑓 ) 6 𝑚. The resultant Res𝑛,𝑚 (P, Q)
is the determinant of the matrix of the definition, while the resultant
Res𝑛,𝑚 (P 𝑓 , Q 𝑓 ) is the determinant of the matrix obtained by applying 𝑓
to each entry. Since the determinant is a polynomial expression, we
obtain the equality

Res𝑛,𝑚 (P 𝑓 , Q 𝑓 ) = 𝑓 (Res𝑛,𝑚 (P, Q)).

Proposition (2.7.3). — Let K be a field. Let P, Q be two polynomials in K[X],

let 𝑛, 𝑚 be strictly positive integers such that deg(P) 6 𝑛 and deg(Q) 6 𝑚.
Then, Res𝑛,𝑚 (P, Q) = 0 if and only if
– either P and Q are not coprime;
– or 𝑎 𝑛 = 𝑏 𝑚 = 0.
100 CHAPTER 2. IDEALS AND DIVISIBILITY

Proof. — If 𝑝 is any integer > 0, let K[X]𝑝 be the K-vector space of

polynomials of degree 6 𝑝. The family (1, X, . . . , X𝑝 ) is a basis of K[X]𝑝
hence this space has dimension 𝑝 + 1. Let us observe that Res𝑛,𝑚 (P, Q)
is the determinant of the linear map
𝜌 : K[X]𝑚−1 × K[X]𝑛−1 → K[X]𝑚+𝑛−1 , (U, V) ↦→ UP + VQ
in the bases (1, . . . , X𝑚−1 ; 1, . . . , X𝑛−1 ) of K[X]𝑚−1 × K[X]𝑛−1 and
(1, X, . . . , X𝑚+𝑛−1 ) of K[X]𝑚+𝑛−1 . Consequently, Res𝑛,𝑚 (P, Q) = 0 if and
only if 𝜌 is not invertible. We shall compute the kernel of 𝜌.
Assume first that P = 0 or Q = 0. Then some column of this matrix is
zero, hence Res𝑛,𝑚 (P, Q) = 0.
Let us now assume that P and Q are both nonzero and let D be their
gcd; one has D ≠ 0. We can thus write P = DP1 and Q = DQ1 , where P1
and Q1 are two coprime polynomials in K[X]. Let (U, V) ∈ Ker(𝜌). One
has UP + VQ = 0, hence UP1 + VQ1 = 0 since D ≠ 0. Since P1 and Q1 are
coprime (see §2.5.10), we obtain that Q1 divides U and P1 divides V. We
can thus write U = Q1 S and V = P1 T, for two polynomials S, T ∈ K[X].
Then UP1 + VQ1 = P1 Q1 (S + T), so that T = −S, U = Q1 S and V = −P1 S.
Since U ∈ K[X]𝑚−1 and V ∈ K[X]𝑛−1 , one has deg(S) 6 𝑚 − 1 − deg(Q1 )
and deg(S) 6 𝑛 − 1 − deg(P1 ). Now,
𝑚 − 1 − deg(Q1 ) = 𝑚 − deg(Q) + deg(Q) − deg(Q1 ) − 1
= (𝑚 − deg(Q)) + deg(D) − 1,
and, similarly,
𝑛 − 1 − deg(P1 ) = (𝑛 − deg(P)) + deg(D) − 1.
Set
𝑠 = inf(𝑛 − deg(P), 𝑚 − deg(Q))
so that 𝑠 = 0 unless 𝑎 𝑛 = 𝑏 𝑚 = 0, in which case 𝑠 > 1. Conversely,
every element of K[X]𝑚−1 × K[X]𝑛−1 of the form (Q1 S, −P1 S), for S ∈
K[X]𝑠+deg(D)−1 , belongs to Ker(𝜌).
This shows that Res𝑛,𝑚 (P, Q) = 0 if and only if 𝑠 + deg D > 0, that is, if
and only if either 𝑎 𝑛 = 𝑏 𝑚 = 0 or deg(D) > 0. 

Corollary (2.7.4). — Let K be an algebraically closed field, and let A = K[Y];

let us identify the ring A[X] with K[X, Y] (remark 1.3.10). Let P, Q be
 2.7. RESULTANTS AND ANOTHER THEOREM OF BÉZOUT 101

two polynomials in K[X, Y] = A[X]. Let 𝑚, 𝑛 be positive integers such that

degX (P) 6 𝑛 and degX (Q) 6 𝑚 and let us write
P = P𝑛 (Y)X𝑛 + · · · + P0 (Y) and Q = Q𝑚 (Y)X𝑚 + · · · + Q0 (Y)
where P0 , . . . , P𝑛 , Q0 , . . . , Q𝑚 ∈ K[Y]. Let R = Res𝑛,𝑚 (P, Q) ∈ K[Y] be the
resultant in sizes (𝑛, 𝑚) of the pair (P, Q). Then, an element 𝑦 ∈ K is a root
of R if and only if
– either the polynomials P(X, 𝑦) and Q(X, 𝑦) have a common root in K;
– or P𝑛 (𝑦) = Q𝑚 (𝑦) = 0.
Proof. — By the definition of the resultant (remark 2.7.2), we have
R(𝑦) = Res𝑛,𝑚 (P, Q) (𝑦) = Res𝑛,𝑚 (P(X, 𝑦), Q(X, 𝑦)).

It thus suffices to apply the preceding proposition to the polynomi-

als P(X, 𝑦) and Q(X, 𝑦) of K[X]. 

Theorem (2.7.5) (Bézout). — Let K be an algebraically closed field. Let P, Q

be two coprime polynomials in K[X, Y], with total degrees 𝑝 and 𝑞 respectively.
Then, the set of common roots to P and Q in K2 , that is, the pairs (𝑥, 𝑦) ∈ K2
such that P(𝑥, 𝑦) = Q(𝑥, 𝑦) = 0, has at most 𝑝𝑞 elements; in particular, it is
finite.
Proof. — Since P and Q are coprime in the ring K[X, Y], they remain co-
prime in K(Y)[X] (see corollary 2.6.4). By proposition 2.7.3, when viewed
as polynomials in K(Y)[X], their resultant R is a nonzero polynomial RY
of K[Y]. Consequently, there are only finitely many possibilities for the
ordinates 𝑦 of the common roots (𝑥, 𝑦) to P and Q. Exchanging the roles
of X and Y, we prove similarly that there are only finitely possibilities
of the abscissae 𝑥 of these common roots. It follows that the set Σ of
common roots to P and Q is finite.
We now show that Card(Σ) is less or equal to the product 𝑝𝑞 of the
degrees of P and Q. To that aim, we make a linear change of variables
so that any horizontal line contains at most one point of Σ. Since there
are only finitely directions to avoid, this may be done. This modifies the
polynomials P and Q, but not their degrees 𝑝 and 𝑞.
Let us write
P = P𝑛 (Y)X𝑛 + · · · + P0 (Y) and Q = Q𝑚 (Y)X𝑚 + · · · + Q0 (Y)
102 CHAPTER 2. IDEALS AND DIVISIBILITY

where P𝑛 and Q𝑚 are nonzero polynomials. Let R = Res𝑛,𝑚 (P, Q)

(resultant with respect to X). We know that for any 𝑦 ∈ K, R(𝑦) = 0 if
and only if either P𝑛 (𝑦) = Q𝑚 (𝑦) = 0, or 𝑦 is the ordinate of a point of Σ.
It thus suffices to show that deg(R) 6 𝑝𝑞.
We first observe that for any integer 𝑖, deg(P𝑖 ) 6 𝑝 − 𝑖 and deg(Q𝑖 ) 6
𝑞 − 𝑖. Let us explicit the entry R𝑖𝑗 at row 𝑖 and column 𝑗 of the matrix
whose determinant is R:
– for 1 6 𝑗 6 𝑚, one has R𝑖𝑗 = P𝑖−𝑗 when 0 6 𝑖 − 𝑗 6 𝑛, and R𝑖𝑗 = 0
sinon ;
– for 𝑚 + 1 6 𝑗 6 𝑚 + 𝑛, one has R𝑖𝑗 = Q𝑖−𝑗+𝑚 when 0 6 𝑖 − 𝑗 + 𝑚 6 𝑚,
and R𝑖𝑗 = 0 otherwise.
In particular, deg(R𝑖𝑗 ) is bounded above by
(
𝑝−𝑖+𝑗 if 1 6 𝑗 6 𝑚 ;
deg(R𝑖𝑗 ) 6 .
𝑞−𝑚−𝑖+𝑗 if 𝑚 + 1 6 𝑗 6 𝑚 + 𝑛.
The determinant R is a sum of products of the form 𝑚+𝑛
Î
𝑗=1 R 𝜎(𝑖)𝑖 , for
all permutations 𝜎 of {1; . . . ; 𝑚 + 𝑛}. The degree of such a product is
bounded above by
𝑚+𝑛
Õ 𝑚
Õ 𝑚+𝑛
Õ
deg(R𝜎(𝑗)𝑗 ) 6 (𝑝 − 𝜎(𝑗) + 𝑗) + (𝑞 − 𝑚 − 𝜎(𝑗) + 𝑗)
𝑗=1 𝑗=1 𝑗=𝑚+1
𝑚+𝑛
Õ 𝑚+𝑛
Õ
6 𝑝𝑚 + 𝑛(𝑞 − 𝑚) − 𝜎(𝑗) + 𝑗
𝑗=1 𝑗=1

6 𝑝𝑞 − (𝑝 − 𝑛)(𝑞 − 𝑚) 6 𝑝𝑞.
Consequently, deg(R) 6 𝑝𝑞 and the theorem is proved. 

Remark (2.7.6). — There is a more precise version of this theorem of

Bézout that takes into account the multiplicity of common roots (for
example, if the curves with equations P and Q are tangent at some
intersection point, this point will have multiplicity > 2), as well as the
possible common roots “at infinity”. Then, the number of common roots
is exactly 𝑝𝑞.

Let us now give two additional properties of the resultant.

 2.7. RESULTANTS AND ANOTHER THEOREM OF BÉZOUT 103

Proposition (2.7.7). — Let A be a commutative ring, let P, Q be two polynomials

in A[X], let 𝑛, 𝑚 be integers such that deg(P) 6 𝑛 and deg(Q) 6 𝑚. Then,
the resultant Res𝑛,𝑚 (P, Q) belongs to the ideal
(P, Q)A[X] ∩ A
of A.
Proof. — We may compute the determinant defining the resultant
Res𝑚,𝑛 (P, Q) in any (commutative) overring of A, in particular in A[X].
Let thus add to the first row X times the second one, X2 times the third
one, etc. We obtain that Res𝑛,𝑚 (P, Q) is the determinant of a matrix with
coefficients in A[X] whose first row is
P XP . . . X𝑚−1 P Q XQ . . . X𝑛−1 Q.
If we expand the determinant with respect to this row we see that
Res𝑛,𝑚 (P, Q) has the form UP + VQ, for two polynomials U and V
in A[X]. This shows that Res𝑛,𝑚 (P, Q) belongs to the ideal (P, Q)A[X]
generated by P and Q in A[X]. Since it also belongs to A, this concludes
the proof of the proposition. 

Proposition (2.7.8). — Let A be a commutative ring and let P, Q be two split

polynomials of A[X]: P = 𝑎 𝑛 𝑛𝑖=1 (X − 𝑡 𝑖 ) and Q = 𝑏 𝑚 𝑚𝑗=1 (X − 𝑢 𝑗 ). Then,
Î Î
Ö
Res𝑛,𝑚 (P, Q) = (−1)𝑚𝑛 𝑎 𝑛𝑚 𝑏 𝑚
𝑛
(𝑡 𝑖 − 𝑢 𝑗 )
𝑖,𝑗
𝑚
Ö 𝑛
Ö
𝑛
= 𝑏𝑚 P(𝑢 𝑗 ) = 𝑎 𝑛𝑚 (−1)𝑚𝑛 Q(𝑡 𝑗 ).
𝑗=1 𝑖=1

Proof. — It is obvious that the three written formulae on the right of

the first equal sign are pairwise equal. We shall prove that they are
equal to Res𝑛,𝑚 (P, Q) by induction on 𝑛. If 𝑛 = 0, then P = 𝑎0 hence
Res0,𝑚 (P, Q) = 𝑎0𝑚 , so that the formula holds in this case. Let us now
show by performing linear combinations on the resultant matrix that
Res𝑛+1,𝑚 ((X − 𝑡)P, Q) = (−1)𝑚 Q(𝑡) Res𝑛,𝑚 (P, Q).
Let us indeed write P = 𝑎 𝑛 X𝑛 + · · · + 𝑎0 . Then,
(X − 𝑡)P = 𝑎 𝑛 X𝑛+1 + (𝑎 𝑛−1 − 𝑡𝑎 𝑛 )X𝑛 + · · · + (𝑎0 − 𝑡𝑎1 )X + 𝑎0
104 CHAPTER 2. IDEALS AND DIVISIBILITY

and Res𝑛+1,𝑚 ((X − 𝑡)P, Q) equals

−𝑡𝑎 0 𝑏0
..
𝑎 1 − 𝑡𝑎 0 −𝑡𝑎0 𝑏1 .
.. ..
. 𝑎 0 − 𝑡𝑎1 −𝑡𝑎0 . 𝑏0
... .. ..
. .
..
−𝑡𝑎0 .
.. ..
𝑎 𝑛−1 − 𝑡𝑎 𝑛 . .
.. ..
𝑎𝑛 . .
... ... ..
𝑏 𝑚−1 .
... ..
𝑏𝑚 .
.. ..
. 𝑎 𝑛−1 − 𝑡𝑎 𝑛 . 𝑏 𝑚−1
𝑎𝑛 𝑏𝑚

(There are 𝑚 “𝑎” columns and 𝑛 + 1 “𝑏” columns.) Beginning from the
bottom, let us add to each row 𝑡 times the next one. This does not change
the determinant hence Res𝑛+1,𝑚 ((X − 𝑡)P, Q) equals

0 ... 0 𝑏 0 + 𝑡𝑏 1 + · · · + 𝑡 𝑚 𝑏 𝑚 𝑡(𝑏 0 + 𝑡𝑏 1 + . . . ) . . . 𝑡 𝑛 (𝑏 0 + . . . )
𝑎0 (𝑏 1 + 𝑡𝑏 2 + . . . ) (𝑏 0 + 𝑡𝑏 1 + . . . ) 𝑡 𝑛−1 (𝑏 0 + . . . )
.. . . ..
. . 𝑎0 .
.. .. ..
. . 𝑏𝑚 .
.. ...
𝑎𝑛 . 𝑏 𝑚−1 + 𝑡𝑏 𝑚
𝑎𝑛 𝑏𝑚

We observe that Q(𝑡) is a factor of each entry of the first row, so that
Res𝑛+1,𝑚 ((X − 𝑡)P, Q) is equal to

0 ... 0 1 𝑡 ... 𝑡𝑛
𝑎0 (𝑏 1 + 𝑡𝑏 2 + . . . ) (𝑏 0 + 𝑡𝑏 1 + . . . ) 𝑡 𝑛−1 (𝑏 0 + . . . )
.. . . ..
. . 𝑎0 .
Q(𝑡) .. .. ..
. . 𝑏𝑚 .
.. ..
𝑎𝑛 . . 𝑏 𝑚−1 + 𝑡𝑏 𝑚
𝑎𝑛 𝑏𝑚
 EXERCISES 105

Beginning from the right, we then may subtract from each “𝑏”-column
𝑡 times the preceding one; we then get the determinant
0 ... 0 1 0 ... 0
𝑎0 𝑏 1 + 𝑡𝑏 2 + . . . 𝑏0
.. . . .. .
. . 𝑎0 . 𝑏1 . .
Q(𝑡) .. .. .. ..
. . . .
.. ...
𝑎𝑛 . 𝑏 𝑚−1
..
𝑎𝑛 . 𝑏𝑚
It now suffices to expand the determinant with respect to the first row
and we obtain
(−1)𝑚 Q(𝑡) Res𝑚,𝑛 (P, Q).
This concludes the proof of the proposition by induction. 

Exercises
1-exo599) a) Show that the ideal (2, X) of the ring Z[X] is not principal.
b) Let A be a commutative ring such that the ring A[X] is a principal ideal domain.
Show that A is a field.
2-exo569) a) Show that the set of continuous functions with compact support, or the
set of functions which vanish at any large enough integer, are ideals of the ring 𝒞(R)
of continuous functions on the real line R. Prove that they are not contained in any
ideal M𝑥 , for 𝑥 ∈ R.
b) Let A be the ring of holomorphic functions on a neighborhood of the closed unit
disk. Show that any ideal of A is generated by a polynomial P ∈ C[𝑧] whose roots have
modulus 6 1. Prove that the maximal ideals of A are the ideals (𝑧 − 𝑎), for 𝑎 ∈ C such
that |𝑎| 6 1.
c) (Generalization.) Let K be a compact, connected and nonempty subset of C and
let ℋ be the ring of holomorphic functions an an open neighborhood of K. Show that
the ring ℋ is a principal ideal domain. Show that all maximal ideals of A are the
ideals (𝑧 − 𝑎), for 𝑎 ∈ K.
3-exo571) Let A be a commutative local ring (definition 2.1.8). Let I and J be two ideals
of A, let 𝑎 ∈ A be a regular element such that IJ = (𝑎).
a) Show that there exist 𝑥 ∈ I and 𝑦 ∈ J such that 𝑥𝑦 = 𝑎. Check that 𝑥 and 𝑦 are
regular.
b) Deduce from this that I = (𝑥) and J = (𝑦).
4-exo572) Let A be the product ring of all fields Z/𝑝Z, where 𝑝 runs over the set of all
prime numbers. Let N be the subset of A consisting of all families (𝑎 𝑝 ) such that 𝑎 𝑝 = 0
for every but finitely many prime numbers 𝑝; let B be the quotient ring A/N.
106 CHAPTER 2. IDEALS AND DIVISIBILITY

a) Let M be a maximal ideal of A which does not contain N. Show that thre exists a
prime number 𝑞 such that M be the set of all families (𝑎 𝑝 ), where 𝑎 𝑞 = 0. What is the
quotient ring A/M?
b) Let 𝑝 be a prime number; show that 𝑝B = B.
c) Show that the ring B admits a unique structure of a Q-algebra.
d) Let M be a maximal ideal of A containing N. Show that the field A/M has
characteristic 0.
5-exo620) Let 𝑘 be a field, let A be the ring 𝑘 N (with termwise addition and multiplica-
tion) and let N be the subset 𝑘 (N) of all almost null sequences.
a) Show that N is an ideal of A. Explain why there exists a maximal ideal M of A
which contains N. Then set K = A/M. Show that K is a field extension of 𝑘.
b) Let 𝑎 = (𝑎1 , . . . , 𝑎 𝑚 ) ∈ A𝑚 be an element of N𝑚 . Show that the set of integers 𝑛
such that 𝑎 𝑖,𝑛 ≠ 0 for some 𝑖 ∈ {1, . . . , 𝑛} is finite. (Otherwise, construct 𝑏 1 , . . . , 𝑏 𝑚 ∈ A
such that 𝑚𝑖=1 𝑏 𝑖 𝑎 𝑖 maps to 1 in K.)
Í
c*) If 𝑘 is infinite, show that the cardinality of K is uncountable.
d) Assuming that 𝑘 is algebraically closed, show that K is algebraically closed too.
e) Let I be an ideal of 𝑘[X1 , . . . , X𝑛 ] and let J be the K-vector subspace of K[X1 , . . . , X𝑛 ]
generated by the elements of I. Show that it is an ideal of K[X1 , . . . , X𝑛 ] and that J ≠ (1)
if I ≠ (1).
f ) Combine the previous construction with the special case of Hilbert’s Nullstellensatz
proved in the text to derive the general case. (You may admit that any ideal of 𝑘[X1 , . . . , X𝑛 ]
is finitely generated.)
6-exo256) Let A be a commutative ring.
a) Let I and J be ideals of A such that V(I) ∩ V(J) = ∅ in Spec(A). Show that I + J = A.
b) Let I and J be ideals of A such that V(I) ∪ V(J) = Spec(A). Show that every element
of I ∩ J is nilpotent.
c) Let I and J be ideals of A such that V(I) ∪ V(J) = Spec(A) and V(I) ∩ V(J) = ∅.
Show that there exist an idempotent 𝑒 ∈ A (this means that 𝑒 2 = 𝑒) such that I = (𝑒)
and J = (1 − 𝑒).
d) Show that Spec(A) is connected if and only if the only idempotents of A are 0
and 1.
7-exo856) Let R be an integral domain which is not a field and let 𝑎 ∈ R {0; 1} such
that the fraction ring R𝑎 is a field. Prove that 1 − 𝑎 is a unit.
8-exo259) One says that a non-empty topological space T is irreducible if for any
closed subsets Z and Z0 of T such that T = Z ∪ Z0, either Z = T or Z0 = T. Let A be a
commutative ring. In this exercise, we study the irreducible closed subsets of Spec(A).
a) Let P be a prime ideal of A. Show that V(P) is irreducible.
b) Show that Spec(A) is irreducible if and only if it has exactly one minimal prime
ideal.
 EXERCISES 107

c) Let I be an ideal of A such that V(I) is irreducible. Show that there exists a unique
minimal prime ideal P containing I and that V(I) = V(P).
9-exo261) An irreducible closed subset of a topological space is called an irreducible
component if it is maximal.
a) Let T be a topological space and let A be a closed irreducible subset of T. Show
that the closed subsets of T containing A is inductive; conclude that there exists an
irreducible component of T that contains A.
b) Deduce from the preceding question that every prime ideal P of A contains a
minimal prime ideal of A.
10-exo260) Let K be an algebraically closed field. Let Z be a subset of K𝑛 which is
closed for the Zariski topology.
a) Show that the following properties are equivalent:
(i) Z is irreducible;
(ii) There exists a prime ideal P of K[T1 , . . . , T𝑛 ] such that Z = 𝒱(P);
(iii) The ideal ℐ(Z) is prime.
b) Show that a closed subset Z0 of K𝑛 is an irreducible component of Z if and only if
ℐ(Z0) is a minimal prime ideal contained in ℐ(Z).
11-exo645) Let E be a field and let X be an infinite set. Let A = EX be the ring of
functions X → E. For 𝑓 ∈ A, let 𝒱( 𝑓 ) = {𝑥 ∈ X ; 𝑓 (𝑥) = 0}.
a) Show that a function 𝑓 ∈ A is invertible if and only if it does not vanish.
b) Let I be an ideal of A such that I ≠ A and let ℱI = {𝒱( 𝑓 ) ; 𝑓 ∈ I}. Show that ℱI is
a filter of subsets of X: (i) ∅ ∉ ℱ; (ii) If Y1 , Y2 are subsets of X such that Y1 ⊂ Y2 and
Y1 ∈ ℱ, then Y2 ∈ ℱ; (iii) If Y1 , Y2 are elements of ℱ, then Y1 ∩ Y2 ∈ ℱ.
c) Let I, J be ideals of A such that I ⊂ J ≠ A; prove that ℱI ⊂ ℱJ .
d) Let P be a prime ideal of A. Prove that the filter ℱP is an ultrafilter: for every
subset Y of X, either Y, or X Y belongs to ℱP .
e) Let ℱ be an ultrafilter on X. Prove that the set of all 𝑓 ∈ A such that 𝒱( 𝑓 ) ∈ ℱ is a
maximal ideal of A.
f ) Assume that I is a finitely generated prime ideal. Prove that there exists 𝑥 ∈ X
such that I = { 𝑓 ∈ A ; 𝑓 (𝑥) = 0}, that I is principal, and that ℱI = {Y ⊂ X ; 𝑥 ∈ Y}. (One
says that ℱI is a principal ultrafilter.)
g) Let I be the set of all 𝑓 ∈ A such that X 𝒱( 𝑓 ) is finite. Prove that I is an ideal
of A and I ≠ A. Prove that if P is a prime ideal that contains I, then P is not principal.
12-exo823) Let P, Q, R ∈ C[T1 , . . . , T𝑛 ] be polynomials such that P does not divide R.
a) Assume that P(𝑎) = 0 for every 𝑎 ∈ C𝑛 such that R(𝑎) ≠ 0. Prove that P = 0.
b) Assume that P(𝑎) = 0 for every 𝑎 ∈ C𝑛 such that R(𝑎) ≠ 0 and Q(𝑎) = 0. Prove that
if Q is irreducible, then Q divides P.
c) Under the hypothesis of b), what can be deduced if Q is no more assumed to be
irreducible?
108 CHAPTER 2. IDEALS AND DIVISIBILITY

13-exo646) A ring A is called a von Neumann ring if for every 𝑎 ∈ A, there exists 𝑥 ∈ A
such that 𝑎 = 𝑎𝑥𝑎.
a) Prove that a division ring is a von Neumann ring.
Î
b) Let (A𝑖 )𝑖∈I be a family of von Neumann rings. Then the product ring A = 𝑖∈I A 𝑖
is a von Neumann ring.
c) Prove that a local von Neumann ring is a division ring.
d) Let K be a division ring and let V be a K-vector space. Prove that EndK (V) is a von
Neumann ring.
14-exo012) Let A be the ring C[X, Y]/(XY − 1). Let 𝑥 and 𝑦 be the images of X and Y
in A.
a) Show that 𝑥 is invertible in A. Show that any nonzero element 𝑎 ∈ A can
be written uniquely on the form 𝑎 = 𝑥 𝑚 P(𝑥), for some integer 𝑚 ∈ Z and some
polynomial P ∈ C[T] whose constant term is nonnull.
b) For 𝑎, 𝑚 and P as above, set 𝑒(𝑎) = deg(P). Show that the map 𝑒 : A {0} → N is
a gauge on A, hence that A is a euclidean ring.
c) Conclude that A is a principal ideal domain.
15-exo644) Let A be the set of all complex numbers of the form 𝑎 + 𝑏𝑖, for 𝑎 and 𝑏 ∈ Z.
Let K be the set of all complex numbers of the form 𝑎 + 𝑏𝑖, for 𝑎 and 𝑏 ∈ Q.
a) Show that K is a subfield of C and that A is a subring of K. Show also that any
element of A (resp. of K) can be written in a unique way in the form 𝑎 + 𝑏𝑖 with 𝑎, 𝑏 ∈ Z
(resp. 𝑎, 𝑏 ∈ Q). (One says that (1, 𝑖) is a basis of A as a Z-module, and a basis of K as a
Q-vector space.)
b) For 𝑥 = 𝑎 + 𝑏𝑖 ∈ C, set 𝛿(𝑥) = |𝑥| 2 = 𝑎 2 + 𝑏 2 . Show that 𝛿(𝑥𝑦) = 𝛿(𝑥)𝛿(𝑦) for any
𝑥, 𝑦 ∈ K.
c) For 𝑥 = 𝑎 + 𝑏𝑖 ∈ K, set {𝑥} = {𝑎} + {𝑏}𝑖, where {𝑡} denotes the integer which is
the closest to a real number 𝑡, chosen to be smaller than 𝑡 if there are two such integers.
Show that 𝛿(𝑥 − {𝑥}) 6 12 .
d) Show that 𝛿 is a gauge on A, hence that A is an euclidean ring.
√
16-exo597) Let A be the set of all real numbers of the
√ form 𝑎 + 𝑏 2, for 𝑎 and 𝑏 ∈ Z. Let
K be the set of all real numbers of the form 𝑎 + 𝑏 2, for 𝑎 and 𝑏 ∈ Q.
a) Show that K is a subfield of R and that A is a subring of K. Show also that√ any
element of A (resp. of K) can be written in a√unique way in the form 𝑎 + 𝑏 2 with
𝑎, 𝑏 ∈ Z (resp. 𝑎, 𝑏 ∈ Q). — One says that (1, 2) is a basis of A as a Z-module, and a
basis of K as a Q-vector space.
√
b) For 𝑥 = 𝑎 + 𝑏 2 ∈ K, set 𝛿(𝑥) = 𝑎 2 − 2𝑏 2 . Show that 𝛿(𝑥𝑦) = 𝛿(𝑥)𝛿(𝑦) for any
𝑥, 𝑦 ∈ K.
√ √
c) For 𝑥 = 𝑎 + 𝑏 2 ∈ K, set {𝑥} = {𝑎} + {𝑏} 2, where {𝑡} denotes the integer which is
the closest to a real number 𝑡, chosen to be smaller than 𝑡 if there are two such integers.
Show that 𝛿(𝑥 − {𝑥}) 6 12 .
 EXERCISES 109

d) Show that 𝛿 is a gauge on A, hence that A is an Euclidean ring.

√
17-exo598) Let 𝜔 be the complex number given by 𝜔 = (1 + 𝑖 3)/2; let K be the set of
all complex numbers of the form 𝑎 + 𝑏𝜔, with 𝑎, 𝑏 ∈ Q, and let A be the set of all such
elements of K where 𝑎, 𝑏 ∈ Z.
a) Show that K is a subfield of C and that A is a subring of K.
b) Show that A is an euclidean ring for the gauge given by 𝑧 ↦→ |𝑧| 2 .
18-exo605) Let A be a domain and let 𝛿 : A {0} → N be a map which satisfies the
second property of the definition of a euclidean gauge, namely: for any 𝑎, 𝑏 ∈ A such
that 𝑏 ≠ 0, there exists 𝑞 and 𝑟 ∈ A such that 𝑎 = 𝑏𝑞 + 𝑟 and such that, moreover, either
𝑟 = 0, or 𝛿(𝑟) < 𝛿(𝑏).
For any 𝑎 ∈ A {0}, set 𝛿0(𝑎) = inf𝑏≠0 𝛿(𝑎𝑏).
Show that 𝛿0 is a gauge on A, hence that A is an Euclidean ring.
19-exo611) Let A be an Euclidean ring, with gauge 𝛿.
a) Let 𝑎 ∈ A be a nonzero, and non-unit element, with minimal gauge. Show that for
any 𝑥 ∈ A which is not a multiple of 𝑎, there exists a unit 𝑢 ∈ A such that 1 − 𝑢𝑥 be a
multiple of 𝑎.
b) Let 𝑛 be the cardinality of the set of units in A. Show that there exists a maximal
ideal M ⊂ A such that the cardinality of the quotient field A/M is smaller or equal
to 𝑛 + 1.
√
20-exo612) Let A be the subring of C generated by 𝜀 = (1 + 𝑖 19)/2.
a) Show that 𝜀2 = 𝜀 − 5. Deduce that any element of A can be written uniquely as
𝑎 + 𝜀𝑏, for 𝑎, 𝑏 ∈ Z.
b) Show that for any 𝑎 ∈ A, |𝑎| 2 is an integer. Show that 𝑎 ∈ A is a unit if and only if
|𝑎| 2 = 1. Conclude that A× = {−1, +1}.
c) Let M be any maximal ideal of A. Show that there exists a prime number 𝑝 such
that 𝑝 ∈ M. Show that A/M has cardinality 𝑝 2 if P = X2 − X + 5 is irreducible in Z/𝑝Z,
and cardinality 𝑝 otherwise.
d) Show that the polynomial X2 − X + 5 is irreducible in Z/2Z and Z/3Z. Conclude
that the cardinality of A/M is at least 4.
e) Show that A is not a euclidean ring.
21-exo872) Let X be a compact metric space and let A = 𝒞(X) be the ring of real valued,
continuous functions on X.
a) If I is an ideal of A, let V(I) be the set of all points 𝑥 ∈ X such that 𝑓 (𝑥) = 0 for all
𝑓 ∈ I. Prove that V(I) is a closed subset of X.
b) Prove that the map 𝑓 ↦→ 𝑓 −1 (0) gives a bijection between idempotents of A and
closed open subsets of X.
c) Let P be a prime ideal of A. Prove that there exists a unique point 𝑥 ∈ X such that
V(P) = {𝑥}.
√
d) Let I be a finitely generated ideal of A such that I = I. Prove that V(I) is open
in X. Prove that there exists an idempotent 𝑒 ∈ A such that I = 𝑒A.
110 CHAPTER 2. IDEALS AND DIVISIBILITY

22-exo873) Let X be metric space and let A = 𝒞(X) be the ring of real valued continuous
functions on X. Let P be a prime ideal of X.
a) Let 𝑓 , 𝑔 ∈ A; assume that 0 6 𝑓 6 𝑔 and 𝑔 ∈ P. Prove that there exists ℎ ∈ A such
that 𝑓 2 = ℎ 𝑔. Conclude that 𝑓 ∈ P.
b) Prove that the (partial) order on A induces a total order on the quotient ring A/P.
c) Let Q, Q0 be prime ideals of X containing X. Prove that either Q ⊂ Q0 or Q0 ⊂ Q
(Kohls, 1958).
23-exo888) Let A be a commutative von Neumann ring (see exercise 2/13).
a) Prove that 0 is the only nilpotent element of A.
b) Let S be a multiplicative subset of A. Prove that the fraction ring S−1 A is a von
Neumann ring.
c) Prove that all prime ideals of A are maximal.
√
24-exo600) Let A be the set of all complex numbers of the form 𝑎 + 𝑏𝑖 5, for 𝑎 and
𝑏 ∈ Z.
a) Show that A is a subring of C.
b) Show that the only units of A are 1 and −1.
√ √
c) Show that 2, 3, 1 + 𝑖 5 and 1 − 𝑖 5 are irreducible in A.
√ √
d) Observing that 2 · 3 = (1 + 𝑖 5)(1 − 𝑖 5), prove that A is not a unique factorization
domain; in particular it is not principal ideal ring.
25-exo625) Let 𝑝 be a prime number, let 𝑛 be an integer such that 𝑛 > 2 and let P be
the polynomial P = X𝑛 + X + 𝑝.
a) Assume that 𝑝 ≠ 2. Show that any complex root 𝑧 of P satisfies |𝑧| > 1.
b) Stil assuming 𝑝 ≠ 2, show that P is irreducible in Z[X].
c) Assume now that 𝑝 = 2. If 𝑛 is even, show that P is irreducible in Z[X]. If 𝑛 is odd,
show that X + 1 divides P but that P/(X + 1) is irreducible in Z[X].
d) More generally, let P = 𝑎 𝑛 X𝑛 + · · · + 𝑎 1 X + 𝑎 0 be any polynomial with integer
coefficients such that |𝑎 0 | is a prime number strictly greater than |𝑎1 | + · · · + |𝑎 𝑛 |. Show
that P is irreducible in Z[X].
26-exo626) Let 𝑛 be any integer > 2 and let S be the polynomial X𝑛 − X − 1. The goal of
this exercise is to show, following Selmer (Math. Scand. 4 (1956), p. 287–302) that S is
irreducible in Z[X].
a) Show that S has 𝑛 distinct roots in C.
b) For any polynomial P ∈ C[X] such that P(0) ≠ 0, set
𝑚
Õ 1

𝜑(P) = 𝑧𝑗 − ,
𝑧𝑗
𝑗=1

where 𝑧 1 , . . . , 𝑧 𝑚 are the complex roots of P, repeated according to their multiplicities.

Compute 𝜑(P) in terms of the coefficients of P. In particular, compute 𝜑(S).
 EXERCISES 111

c) If P and Q are two polynomials in C[X] such that P(0)Q(0) ≠ 0, show that
𝜑(PQ) = 𝜑(P) + 𝜑(Q).
d) For any complex root 𝑧 of S, show that
1
1
2< 𝑧 − > − 1.
𝑧 |𝑧| 2
(Set 𝑧 = 𝑟𝑒 𝑖𝜃 and estimate cos(𝜃) in terms of 𝑟.)
Î𝑚
e) For any positive real numbers 𝑥1 , . . . , 𝑥 𝑚 with 𝑗=1 𝑥 𝑗 = 1, establish the inequality
𝑚
Õ
𝑥 𝑗 > 𝑚.
𝑗=1
f ) Let P and Q be polynomials in Z[X] of strictly positive degrees such that S = PQ.
Show that |P(0)| = 1 and that 𝜑(P) is a strictly positive integer. Deduce a contradiction
and conclude that S is irreducible in Z[X].
27-exo017) Eisenstein’s irreducibility criterion. Let A be an integral domain, let K be its
fraction field and let P be a prime ideal of A. Let 𝑓 (X) = 06𝑘6𝑛 𝑎 𝑘 X 𝑘 be a polynomial
Í
of degree 𝑛 > 1 with coefficients in A. We assume that 𝑎 0 , . . . , 𝑎 𝑛−1 ∈ P, 𝑎 𝑛 ∉ P and
𝑎 0 ∉ P2 .
a) Let 𝑔 ∈ A[X] be a non-constant polynomial that divides 𝑓 . Prove that deg(𝑔) = 𝑛.
(Consider a factorization 𝑓 = 𝑔 ℎ and reduce it modulo P.)
b) Show that 𝑓 is irreducible in K[X].
c) Is 𝑓 necessary irreducible in A[X]?
28-exo623) Let P = X𝑛 + 𝑎 𝑛−1 X𝑛−1 + · · · + 𝑎 0 be a monic polynomial in Z[X] such that
𝑎 0 ≠ 0 and
|𝑎 𝑛−1 | > 1 + |𝑎 𝑛−2 | + · · · + |𝑎 0 | .
a) Using Rouché’s theorem in the theory of holomorphic functions, show that P
has exactly one complex root with absolute value > 1, and that all other roots have
absolute value < 1.
b) Show that P is irreducible in Z[X] (Perron’s theorem).
29-exo849) Let K be a field. The Newton polytope N 𝑓 of a polynomial 𝑓 ∈ K[T1 , . . . , T𝑛 ]
in 𝑛 indeterminates is the convex hull in R𝑛 of the set of exponents of all monomials
that appear in 𝑓 . If C is a convex subset of R𝑛 , a point 𝑢 ∈ C is called a vertex if for
every 𝑣, 𝑤 ∈ C such that 𝑢 ∈ [𝑣, 𝑤] (the line segment with endpoints 𝑣 and 𝑤 in R𝑛 ),
one has 𝑣 = 𝑤 = 𝑢.
a) When 𝑛 = 2, describe all possible Newton polytopes of polynomials of degree 6 3.
Let 𝑓 , 𝑔, ℎ ∈ K[T1 , . . . , T𝑛 ] be polynomials such that 𝑓 = 𝑔 ℎ.
b) Let 𝑢 be a vertex of N 𝑔 + N ℎ ; prove that there exists a unique pair (𝑣, 𝑤) where
𝑣 ∈ N 𝑔 and 𝑤 ∈ N ℎ such that 𝑢 = 𝑣 + 𝑤. Prove that 𝑣 is a vertex of N 𝑔 and that 𝑤 is a
vertex of N ℎ ; conclude that 𝑢 ∈ N 𝑓 .
c) Prove that N 𝑓 = N 𝑔 + N ℎ .
112 CHAPTER 2. IDEALS AND DIVISIBILITY

d) We assume that 𝑛 = 2. Let 𝑎, 𝑏, 𝑢, 𝑣 be coprime integers. Describe the set of

polynomials 𝑓 ∈ K[T1 , T2 ] whose Newton polytope is a triangle with vertices (𝑎, 0),
(0, 𝑏) and (𝑢, 𝑣). Prove that these polynomials are irreducible.
30-exo198) One considers the C-algebra C[X, Y] of polynomials with complex coeffi-
cients in two variables X and Y. Let I be the ring of C[X, Y] generated by the polynomial
Y2 − X3 + X and let A be the quotient ring C[X, Y]/I. One writes 𝑥 and 𝑦 for the images
of X and Y in A.
The goal of the exercise is to show that A is not a unique factorization domain.
a) Show that Y2 − X3 + X is a irreducible polynomial and that A is a domain.
b) Show that the canonical morphism from C[T] to A which sends T to 𝑥 is injective.
Deduce from this that the subring C[𝑥] of A generated by C and 𝑥 is isomorphic to the
ring of polynomials C[T]. We define the degree of an element of C[𝑥] to be the degree
of its unique preimage in C[T].
c) Let 𝑎 ∈ A. Show that there exists a unique pair (𝑝, 𝑞) of elements of C[𝑥] such that
𝑎 = 𝑝 + 𝑞𝑦.
d) Show that there exists a unique C-linear endomorphism 𝜎 of A such that 𝜎(𝑥) = 𝑥
and 𝜎(𝑦) = −𝑦. Show that 𝜎 is an automorphism of A. For 𝑎 ∈ A, show that 𝜎(𝑎) = 𝑎 if
and only if 𝑎 ∈ C[𝑥].
e) For any 𝑎 ∈ A, set N(𝑎) = 𝑎𝜎(𝑎). Show that N(𝑎) ∈ C[𝑥], that N(1) = 1, and that
deg(N(𝑎)) ≠ 1. Show also that N(𝑎𝑏) = N(𝑎)N(𝑏) for any 𝑎, 𝑏 ∈ A.
f ) Deduce from questions b), c) and e) that C {0} is the set of units of A.
g) Show that 𝑥, 𝑦, 1 − 𝑥 and 1 + 𝑥 are irreducible in A.
h) Show that 𝑦 does not divide 𝑥, 1 + 𝑥 nor 1 − 𝑥. Conclude that A is not a unique
factorization domain.
31-exo627) Let A be a commutative ring.
a) Let P and Q be polynomials in A[X] Assume that the coefficients of P, resp. those
of Q, generate the unit ideal (1) of A. Show that the coefficients of PQ generate the unit
ideal too. (Modify the proof of proposition 2.6.3: observe that the hypothesis implies that for
any maximal ideal M of A, P and Q are nonzero modulo M.)
b) Show that there are polynomials
W𝑖 ∈ Z[P0 , . . . , P𝑛 , Q0 , . . . , Q𝑛 , U0 , . . . , U𝑛 , V0 , . . . , V𝑛 ]
(for 0 6 𝑖 6 2𝑛) such that if P = 𝑝 𝑖 X𝑖 , Q = 𝑞 𝑖 X𝑖 are polynomials of degree at
Í Í
most 𝑛, 1 = 𝑝 𝑖 𝑢𝑖 and 1 = 𝑞 𝑖 𝑣 𝑖 are Bézout relations for their coefficients, then,
Í Í
writing R = PQ = 𝑟 𝑖 X𝑖 , the coefficients of R satisfy a Bézout relation
Í

2𝑛
Õ
1= 𝑟 𝑖 W𝑖 (𝑝0 , . . . , 𝑝 𝑛 , 𝑞0 , . . . , 𝑞 𝑛 , 𝑢0 , . . . , 𝑢𝑛 , 𝑣0 , . . . , 𝑣 𝑛 ).
𝑖=0
c) For P ∈ A[X], let ict(P) be the ideal of A generated by the coefficients of P. By
the first question, we have ict(PQ) = A if ict(P) = ict(Q) = A. If A is a principal ideal
domain, ict(P) is the ideal generated by ct(P), hence ict(PQ) = ict(P) ict(Q) by Gauss’s
 EXERCISES 113

lemma. Show however that the equality ict(PQ) = ict(P) ict(Q) does not hold in general,
for example if A = C[U, V], P = UX + V and Q = VX + U.
32-exo801) Let A be an integral domain and let S be a multiplicative subset of A such
that 0 ∉ S. This exercise studies the relations between A and S−1 A being unique
factorization domains (theorems of Nagata).
a) Assume that A is a unique factorization domain; prove that S−1 A is a unique
factorization domain.
b) Let 𝑝 ∈ A be such that the ideal (𝑝) is prime. Assume that S = {𝑝 𝑛 ; 𝑛 ∈ N}
and that the ring S−1 A is a unique factorization domain. Prove that A is a unique
factorization domain.
c) Assume that every increasing sequence of principal ideals of A is stationary. Prove
that if S−1 A is a unique factorization domain, then A is a unique factorization domain.
d) We consider the particular case of a polynomial ring A = R[T], where R is a unique
factorization domain. Taking S = R {0}, deduce from the preceding question that A
is a unique factorization domain.
e) Let 𝑘 be a field in which −1 is not a square and let A = 𝑘[X, Y, Z]/(X2 + Y2 + Z2 − 1).
Prove that A is a unique factorization domain. (Take S = {(1 − 𝑧)𝑛 , 𝑛 ∈ N}, where 𝑧 is
the class of Z in A.)
f ) Let 𝑘 be an algebraically closed field of characteristic different from 2, let 𝑛 > 5
and let 𝑓 = X21 + · · · + X2𝑛 . Prove that 𝑘[X1 , . . . , X𝑛 ]/( 𝑓 ) is a unique factorization domain.
(Prove that X23 + · · · + X2𝑛 is irreducible, then take S = {(𝑥1 + 𝑖𝑥2 )𝑚 , 𝑚 ∈ N}, where 𝑖 ∈ 𝑘
satisfies 𝑖 2 = −1.)
33-exo846) Let K be a field.
a) Let 𝑎, 𝑏 ∈ K× and 𝑑 > 1. Prove that every irreducible factor of 𝑎X𝑑 + 𝑏Y𝑑 is
homogeneous.
b) Let 𝑎, 𝑏, 𝑐 ∈ K× and 𝑑 > 1, Prove that 𝑎X𝑑 + 𝑏Y𝑑 + 𝑐 is irreducible.
c) Let P1 , P2 , . . . , P𝑛 be non-constant polynomials in K[T], of degrees 𝑑1 , . . . , 𝑑𝑛 . Let
P = P1 (T1 ) + · · · + P𝑛 (T𝑛 ). Prove that P is irreducible if 𝑛 = 2 and 𝑑1 , 𝑑2 are coprime
integers. (Assign weight 𝑑1 to T1 and 𝑑2 to T2 , hence 𝑚1 𝑑1 + 𝑚2 𝑑2 to a monomial T1𝑚1 T2𝑚2 ,
and consider the monomials of largest weights in a factorization of P.)
d) Prove that P is irreducible if 𝑛 > 3 (a theorem of Tverberg (1964)).
CHAPTER 3

MODULES

Modules are to rings what vector spaces are to fields and this introductory
chapter studies them under the angle of linear algebra. The first sections have
to set a number of definitions up, which are the closed companions to similar
definitions for vector spaces: modules and their submodules; morphisms from
a module to another (the analogues of linear maps), their kernel and image,
isomorphisms; generating families, independent families, bases. A notable
difference with the linear algebra of vector spaces is that modules do not possess
a basis in general.
I also present three general constructions of modules: direct sums and
products; quotient of a module by a submodule; and localization of a module
by a multiplicative subset of the base ring. These last constructions have
universal properties which I will use systematically in the rest of the book.
These universal properties not only characterize the modules they consider;
above all, they furnish an explicit way to work with them — in some sense,
they have to be understood as mere computation rules. They also furnish an
appropriate opportunity to introduce a bit of categorical language.
As I said in the introduction to the book, I presuppose in this book a basic
acquaintance with basic algebra, such as vector spaces. However, in case the
theory of vector spaces would have only been presented to the reader in a
restricted case (such as with a base field equal to R or C), I decided to prove
the main theorems here. One excuse I may present is that I allow for general
division rings and not only (commutative) fields.
Now restricting myself to commutative rings, I then discuss the theory of
determinants, from the point of view of alternate multilinear forms. (I will
also revisit this theory in the chapter devoted to tensor products.) As in linear
116 CHAPTER 3. MODULES

algebra over fields, determinants allow to characterize injective, resp. surjective

endomorphisms of a free module of finite rank, but the proofs are a bit subtler.
Determinants are also used to define the Fitting ideals of a finitely generated
module. The definition of these ideals, will reappear in the classification of
modules over principal ideal domain.

3.1. Definition of a module

Definition (3.1.1). — Let A be a ring. A right A-module is a set M endowed
with two binary laws: an internal addition, M × M → M, (𝑚, 𝑛) ↦→ 𝑚 + 𝑛,
and a scalar multiplication M × A → M, (𝑚, 𝑎) ↦→ 𝑚𝑎, subject to the
following properties:
a) Axioms concerning the addition law and stating that (M, +) is an
abelian group:
(i) For every 𝑚, 𝑛 in M, 𝑚 + 𝑛 = 𝑛 + 𝑚 (commutativity of addition);
(ii) For every 𝑚, 𝑛, 𝑝 in M, (𝑚 + 𝑛) + 𝑝 = 𝑚 + (𝑛 + 𝑝) (associativity
of addition);
(iii) There exists an element 0 ∈ M such that 𝑚 + 0 = 0 + 𝑚 = 𝑚 for
every 𝑚 ∈ M (neutral element for the addition);
(iv) For every 𝑚 ∈ M, there exists 𝑛 ∈ M such that 𝑚 + 𝑛 = 𝑛 + 𝑚 = 0
(existence of an opposite for the addition);
b) Axioms concerning the multiplication law:
(v) For any 𝑚 ∈ M, 𝑚1 = 𝑚 (the unit of A is neutral for the scalar
multiplication);
(vi) For any 𝑎, 𝑏 ∈ A and any 𝑚 ∈ M, one has 𝑚(𝑎𝑏) = (𝑚𝑎)𝑏
(associativity of the scalar multiplication);
c) Axioms relating the ring addition, the module addition and the
scalar multiplication:
(vii) For any 𝑎, 𝑏 ∈ A and any 𝑚 ∈ M, one has 𝑚(𝑎 + 𝑏) = 𝑚𝑎 + 𝑚𝑏
(scalar multiplication distributes over the ring addition);
(viii) For any 𝑎 ∈ A and any 𝑚, 𝑛 ∈ M, one has (𝑚 + 𝑛)𝑎 = 𝑚𝑎 + 𝑛𝑎
(scalar multiplication distributes over the module addition).
 3.1. DEFINITION OF A MODULE 117

Analogously, a left A-module is a set M endowed with two binary laws: an

internal addition, M×M → M, (𝑚, 𝑛) ↦→ 𝑚 +𝑛, and a scalar multiplication
A × M → M, (𝑎, 𝑚) ↦→ 𝑎𝑚, subject to the following properties:

a) Axioms concerning the addition law and stating that (M, +) is an

abelian group:
(i) for every 𝑚, 𝑛 in M, 𝑚 + 𝑛 = 𝑛 + 𝑚 (commutativity of addition);
(ii) for every 𝑚, 𝑛, 𝑝 in M, (𝑚 + 𝑛) + 𝑝 = 𝑚 + (𝑛 + 𝑝) (associativity
of addition);
(iii) there exists an element 0 ∈ M such that 𝑚 + 0 = 0 + 𝑚 = 𝑚 for
every 𝑚 ∈ M (neutral element for the addition);
(iv) for every 𝑚 ∈ M, there exists 𝑛 ∈ M such that 𝑚 + 𝑛 = 𝑛 + 𝑚 = 0
(existence of an opposite for the addition);
b) Axioms concerning the multiplication law:
(v) for any 𝑚 ∈ M, 1𝑚 = 𝑚 (the unit of A is neutral for the scalar
multiplication);
(vi) for any 𝑎, 𝑏 ∈ A and any 𝑚 ∈ M, one has (𝑎𝑏)𝑚 = 𝑎(𝑏𝑚)
(associativity of the scalar multiplication);
c) Axioms relating the ring addition, the module addition and the
scalar multiplication:
(vii) for any 𝑎, 𝑏 ∈ A and any 𝑚 ∈ M, one has (𝑎 + 𝑏)𝑚 = 𝑎𝑚 + 𝑏𝑚
(scalar multiplication distributes over the ring addition);
(viii) for any 𝑎 ∈ A and any 𝑚, 𝑛 ∈ M, one has 𝑎(𝑚 + 𝑛) = 𝑎𝑚 + 𝑎𝑛
(scalar multiplication distributes over the module addition).

Endowed with its internal addition law, any (left or right) module M
is in particular an abelian group. The opposite of an element 𝑚 ∈ M
is written −𝑚. The neutral element of M is written 0, as is the neutral
element of the ring A for the addition. If needed, one may write 0M , 0A
to insist on their differences.
Observe that for any right A-module M and any 𝑚 ∈ M, one has
𝑚0A = 𝑚(0A + 0A ) = 𝑚0A + 𝑚0A , hence 𝑚0A = 0M ; moreover, 0M =
𝑚0A = 𝑚(1 − 1) = 𝑚1 + 𝑚(−1), hence −𝑚 = 𝑚(−1). Similarly, for any
left A-module M and any 𝑚 ∈ M, one has 0A 𝑚 = 0M and (−1)𝑚 = −𝑚.
118 CHAPTER 3. MODULES

3.1.2. — Let M be a right A-module, let I be a set, let (𝑎 𝑖 )𝑖∈I be a family

of elements of A and let (𝑚 𝑖 )𝑖∈I be a family of elements of M. Assume
that the family (𝑎 𝑖 ) has finite support. Then the family (𝑚 𝑖 𝑎 𝑖 ) has finite
support; its sum 𝑖∈I 𝑚 𝑖 𝑎 𝑖 is called a linear combination of the elements 𝑚 𝑖 .
Í
If I0 , I00 are disjoint subsets of I such that 𝑎 𝑖 = 0 for all 𝑖 ∈ I (I0 ∪ I00),
then one has Õ Õ Õ
𝑚𝑖 𝑎 𝑖 = 𝑚𝑖 𝑎 𝑖 + 𝑚𝑖 𝑎 𝑖 .
𝑖∈I 𝑖∈I0 𝑖∈I00
Moreover, for every 𝑏 ∈ A, one has
!
Õ Õ
𝑚𝑖 𝑎 𝑖 𝑏 = 𝑚 𝑖 (𝑎 𝑖 𝑏).
𝑖∈I 𝑖∈I

Finally, if (𝑎 𝑖 )𝑖∈I and (𝑏 𝑖 )𝑖∈I are two families with finite support, then the
family (𝑎 𝑖 + 𝑏 𝑖 )𝑖∈I has finite support as well, and one has
Õ Õ Õ
𝑚 𝑖 (𝑎 𝑖 + 𝑏 𝑖 ) 𝑚𝑖 𝑎 𝑖 + 𝑚𝑖 𝑏 𝑖 .
𝑖∈I 𝑖∈I 𝑖∈I

Examples (3.1.3). — a) Let A be a ring; multiplication A × A → A

endowes the abelian group with two structures of modules, one left,
denoted A𝑠 (sinister, Latin for left), and one right, denoted A𝑑 (dexter,
Latin for right). Let I be a left ideal of A; the multiplication of A, A×I → I,
endowed I with the structure of a left A-module. Similarly, if I is a right
ideal of A, the multiplication I × A → I endowes it with the structure of
a right A-module.
b) Let A be a ring and let M be a left A-module. Let Ao be the opposite
ring to A, with the same addition and multiplication computed in the
opposite order. Endowed with the scalar multiplication M × Ao → M
given by (𝑚, 𝑎) ↦→ 𝑎𝑚, M becomes a right Ao -module.
This correspondence allows to deduce from a statement for right
A-modules an analogous statement for left A-modules.
When A is commutative, Ao is identical to A, hence any left A-module
is also a right A-module, with the same addition and the same scalar
multiplication, only written in the opposite order.
c) An abelian group G can be converted to a Z-module in a unique
way. Indeed, for any integer 𝑛 and any element 𝑔 ∈ G, 𝑛 𝑔 is defined by
 3.1. DEFINITION OF A MODULE 119

induction by the rules 0𝑔 = 0, 𝑛 𝑔 = (𝑛 −1)𝑔 + 𝑔 if 𝑛 > 1 and 𝑛 𝑔 = −(−𝑛)𝑔

if 𝑛 6 −1.
d) When A is a field, in particular commutative, the notion of a A-
module coincides with that of a A-vector space which has presumably
be already studied extensively enough. We keep this terminology of
a vector space when A is a division ring, thus talking of left or right
A-vector space.
e) Let 𝑓 : A → B be a morphism of rings, the scalar multiplication
B × A → B given by (𝑏, 𝑎) ↦→ 𝑏 𝑓 (𝑎) endowes B with the structure of a
right A-module.
More generally, if 𝑓 : A → B is a morphism of rings and if M is a right
B-module, the scalar multiplication M×A → M given by (𝑚, 𝑎) ↦→ 𝑚 𝑓 (𝑎)
endowes M with the structure of a right A-module.
We leave to the reader to explicit the similar statements for left A-
modules.
f) Let A and B be rings. An (A, B)-bimodule is an abelian group M
endowed with a structure of a left A-module and a structure of right
B-module such that 𝑎(𝑚𝑏) = (𝑎𝑚)𝑏 for every 𝑎 ∈ A, 𝑚 ∈ M and 𝑏 ∈ B.
This amounts to the datum of a structure of left A × Bo -module, or to
the datum of a structure of right Ao × B-module.

In the sequel, a A-module (without further precision) shall be under-

stood as a right A-module and most of the stated properties will be
only proved for them. I will often leave to the conscientious reader to
prove (and sometimes even to explicit) the analogous statements for left
A-modules.

Remark (3.1.4). — Let A be a ring and let M be a left A-module. For

𝑎 ∈ A, let 𝜇𝑎 : M → M be the map given by 𝜇𝑎 (𝑚) = 𝑎𝑚. This is an
endomorphism of M as an abelian group. Indeed, 𝜇𝑎 (𝑚 +𝑛) = 𝑎(𝑚 +𝑛) =
𝑎𝑚 + 𝑎𝑛 = 𝜇𝑎 (𝑚) + 𝜇𝑎 (𝑛) for any 𝑚, 𝑛 ∈ M.
Moreover, the map 𝑎 ↦→ 𝜇𝑎 is a morphism of rings from A to the
ring End(M) of endomorphisms of the abelian group M. Indeed, 𝜇𝑎
is the zero map, 𝜇1 = idM , and for any 𝑎, 𝑏 ∈ A and any 𝑚 ∈ M, one
has 𝜇𝑎+𝑏 (𝑚) = (𝑎 + 𝑏)𝑚 = 𝑎𝑚 + 𝑏𝑚 = 𝜇𝑎 (𝑚) + 𝜇𝑏 (𝑚) = (𝜇𝑎 + 𝜇𝑏 )(𝑚) and
120 CHAPTER 3. MODULES

𝜇𝑎𝑏 (𝑚) = (𝑎𝑏)𝑚 = 𝑎(𝑏𝑚) = 𝜇𝑎 (𝑏𝑚) = 𝜇𝑎 ◦ 𝜇𝑏 (𝑚), so that 𝜇𝑎+𝑏 = 𝜇𝑎 + 𝜇𝑏

and 𝜇𝑎𝑏 = 𝜇𝑎 ◦ 𝜇𝑏 .
Conversely, let M be an abelian group and let 𝜇 : A → End(M) be a
morphism of rings. Let us define an scalar multiplication A × M → M
by (𝑎, 𝑚) ↦→ 𝜇(𝑎)(𝑚). One checks that this enriches the abelian group
(M, +) with a structure of a left A-module such that for any 𝑎 ∈ A, 𝜇(𝑎)
be the multiplication by 𝑎.
Consequently, given an abelian group M, it is equivalent to endow
it with a structure of a left A-module and to give oneself a morphism
of rings A → End(M). In particular, every abelian group M becomes
canonically a left EndAb (M)-module.
A structure of right A-module on an abelian group is equivalent to
the datum of a morphism of rings A → End(M)o .

3.1.5. — Let A be a ring and let M be an A-module.

Let 𝑚 ∈ M. The set AnnA (𝑚) of all 𝑎 ∈ A such that 𝑚𝑎 = 0 is a right
ideal of A, called the annihilator of 𝑚. (Indeed, for 𝑎, 𝑏 ∈ A, one has
𝑚0 = 0, 𝑚(𝑎 + 𝑏) = 𝑚𝑎 + 𝑚𝑏 = 0 if 𝑚𝑎 = 𝑚𝑏 = 0, and 𝑚(𝑎𝑏) = (𝑚𝑎)𝑏 = 0
if 𝑚𝑎 = 0.)
The set AnnA (M) of all 𝑎 ∈ A such that 𝑚𝑎 = 0 for all 𝑚 ∈ M is
called the annihilator of M. Since it is the intersection of all AnnA (𝑚), for
𝑚 ∈ M, it is a right ideal of A. On the other hand, if 𝑎 ∈ AnnA (M) and
𝑏 ∈ A, one has 𝑚(𝑏𝑎) = (𝑚𝑏)𝑎 = 0 for all 𝑚 ∈ M, so that 𝑏𝑎 ∈ AnnA (M).
Consequently, AnnA (M) is a two-sided ideal of A.
If AnnA (M) = 0, one says that the A-module M is faithful.
In fact, the ideal AnnA (M) is the kernel of the morphism 𝜇 : A →
End(M), 𝑎 ↦→ 𝜇𝑎 . In this way, M can be viewed as an A/I-module, for
every ideal I of A such that I ⊂ AnnA (M). Taking I = AnnA (M), we
observe that M is a faithful (A/AnnA (M))-module.
Conversely, if M is an (A/AnnA (M))-module, the composition A →
A/AnnA (M) → End(M) endowes M with the structure of an A module
whose annihilator contains AnnA (M).

Definition (3.1.6). — Let A be a ring and let M be a right A-module. A

submodule of M is a subset N of M satisfying the following properties:
(i) N is an abelian subgroup of M;
 3.1. DEFINITION OF A MODULE 121

(ii) for every 𝑎 ∈ A and every 𝑚 ∈ M, then 𝑚𝑎 ∈ N.

There is a similar definition for left A-modules, and for (A, B)-
bimodules.

Examples (3.1.7). — a) Let M be a A-module. The subsets {0} and M

of M are submodules.
b) Let A be a ring. The left ideals of A are exactly the submodules of
the left A-module A𝑠 . The right ideals of A are the submodules of the
right A-module A𝑑 .
c) Let A be a division ring. The submodules of an A-vector space are
its vector subspaces.

Lemma (3.1.8). — Let A be a ring, let M be a right A-module and let N be a

subset of M. To show that N is a submodule of M, it suffices to show that it
satisfies the following properties:
(i) 0 ∈ N;
(ii) if 𝑎 ∈ A and 𝑚 ∈ N, then 𝑚𝑎 ∈ N;
(iii) if 𝑚 ∈ N and 𝑛 ∈ N, then 𝑚 + 𝑛 ∈ N.

Proof. — Indeed, the second property applied to 𝑎 = −1 and 𝑚 ∈ N

implies that −𝑚 ∈ N. Combined with the two other properties, this
shows that N is an subgroup of the abelian group M. Then the second
property shows that it is a submodule of M 

Example (3.1.9). — Let A be a commutative ring and let M be an A-

module. An element 𝑚 ∈ M is said to be torsion if there exists a regular
element 𝑎 ∈ A such that 𝑚𝑎 = 0.
The set T(M) of all torsion elements of M is a submodule of M. Indeed, one
has 0 ∈ T(M), because 0M · 1A = 0. Let 𝑚 ∈ T(M) and let 𝑎 ∈ A; let 𝑏 be
a regular element such that 𝑚𝑏 = 0. Then (𝑚𝑎)𝑏 = 𝑚(𝑎𝑏) = (𝑚𝑏)𝑎 = 0,
hence 𝑚𝑎 ∈ T(M). Finally, let 𝑚, 𝑛 ∈ T(M), let 𝑎, 𝑏 ∈ A be regular
elements such that 𝑚𝑎 = 𝑛𝑏 = 0; then 𝑎𝑏 is a regular element of A and
(𝑚 + 𝑛)𝑎𝑏 = (𝑚𝑎)𝑏 + (𝑛𝑏)𝑎 = 0, hence 𝑚 + 𝑛 ∈ T(M).
One says that M is torsion-free if T(M) = 0, and that M is a torsion A-module
if M = T(M).
122 CHAPTER 3. MODULES

3.2. Morphisms of modules

Definition (3.2.1). — Let A be a ring, let M and N be two A-modules. A
morphism from M to N is a map 𝑓 : M → N such that

𝑓 (𝑚𝑎 + 𝑛𝑏) = 𝑓 (𝑚)𝑎 + 𝑓 (𝑛)𝑏

for every 𝑎, 𝑏 ∈ A and every 𝑚, 𝑛 ∈ M. One writes HomA (M, N) for the set
of morphisms from M to N.
A morphism from M to itself is called an endomorphism of M. The set of
all endomorphisms of an A-module M is denoted EndA (M).

The expressions “A-linear mapping”, or even “linear mapping” are

synonyms for “morphism of A-modules”.
The identity map idM from an A-module M to itself is an endomor-
phism.
Let A be a ring, let M, N, P be three A-modules. If 𝑓 : M → N and
𝑔 : N → P are morphisms, then their composition 𝑔 ◦ 𝑓 : M → P is a
morphism of A-modules.
We thus can say that A-modules, and morphisms of A-modules form a
category Mod A . Moreover, if 𝑓 : A → B is a morphism of rings, viewing
a B-module as an A-module furnishes a functor 𝑓 ∗ : Mod B → Mod A .
One says that a morphism of A-modules 𝑓 : M → N is an isomorphism
if there exists a morphism 𝑔 : N → M such that 𝑓 ◦ 𝑔 = idN and
𝑔 ◦ 𝑓 = idM . Then, there is exactly one such morphism 𝑔, called
the reciprocal (or inverse) of 𝑓 . Indeed, if 𝑓 ◦ 𝑔 = 𝑓 ◦ ℎ = idN and
𝑔 ◦ 𝑓 = ℎ ◦ 𝑓 = idM , then ℎ = ℎ ◦idN = ℎ ◦( 𝑓 ◦ 𝑔) = (ℎ ◦ 𝑓 )◦ 𝑔 = idM ◦𝑔 = 𝑔.
One says that a morphism 𝑓 : M → N is left invertible if there exists a
morphism 𝑔 : N → M such that 𝑔 ◦ 𝑓 = idM , and that it is right invertible
if there exists a morphism 𝑔 : N → M such that 𝑓 ◦ 𝑔 = idN . A left
invertible morphism is injective, a right invertible morphism is surjective,
but the converse assertions are not true in general.

Proposition (3.2.2). — For a morphism of A-modules to be an isomorphism, it

is necessary and sufficient that it be bijective.

Proof. — If 𝑓 : M → N if an isomorphism, with reciprocal 𝑔, then 𝑔 is

also the inverse bijection of 𝑓 .
 3.2. MORPHISMS OF MODULES 123

Conversely, let 𝑓 : M → N be a bijective morphism and let 𝑔 be its

inverse bijection. Then, 𝑔 is a morphism. Indeed, for 𝑛, 𝑛 0 ∈ N and
𝑎, 𝑎 0 ∈ A, we have

𝑓 𝑔(𝑛)𝑎 + 𝑔(𝑛 0)𝑎 0 = 𝑓 (𝑔(𝑛))𝑎 + 𝑓 (𝑔(𝑛 0))𝑎 0 = 𝑛𝑎 + 𝑛 0 𝑎 0

hence 𝑔(𝑛)𝑎 + 𝑔(𝑛 0)𝑎 0 = 𝑔(𝑛𝑎 + 𝑛 0 𝑎 0), so that 𝑔 is linear. 

Definition (3.2.3). — Let 𝑓 : M → N be a morphism of A-modules. The

kernel of 𝑓 , written Ker( 𝑓 ), is the set of all 𝑚 ∈ M such that 𝑓 (𝑚) = 0.

It is the kernel of 𝑓 as a morphism of abelian groups.

Proposition (3.2.4). — Let 𝑓 : M → N be a morphism of A-modules.

For any submodule M0 of M, 𝑓 (M0) is a submodule of N. For any submod-
ule N0 of N, then 𝑓 −1 (N0) is a submodule of M.
In particular, the kernel Ker( 𝑓 ) and the image Im( 𝑓 ) = 𝑓 (M) of 𝑓 are
submodules (respectively of M and N).

Proof. — Let us show that 𝑓 (M0) is a submodule of N. Since 𝑓 (0M ) = 0N

and 0M ∈ M0, we have 0N ∈ 𝑓 (M0). On the other hand, for 𝑛, 𝑛 0 ∈ 𝑓 (M0),
there exist 𝑚, 𝑚 0 ∈ M0 such that 𝑛 = 𝑓 (𝑚) and 𝑛 0 = 𝑓 (𝑚 0). Then,

𝑛 + 𝑛 0 = 𝑓 (𝑚) + 𝑓 (𝑚 0) = 𝑓 (𝑚 + 𝑚 0) ∈ 𝑓 (M0).

Finally, if 𝑛 = 𝑓 (𝑚) belongs to 𝑓 (M0) and if 𝑎 ∈ A, then 𝑛𝑎 = 𝑓 (𝑚)𝑎 =

𝑓 (𝑚𝑎) belongs to 𝑓 (M0) since 𝑚𝑎 ∈ M0.
Let us show that 𝑓 −1 (N0) is a submodule of M. Since 𝑓 (0M ) = 0N ∈ N0,
we have 0M ∈ 𝑓 −1 (N0). Moreover, for 𝑚, 𝑚 0 ∈ 𝑓 −1 (N0) and for 𝑎, 𝑏 ∈ A,
we have
𝑓 (𝑚𝑎 + 𝑚 0𝑏) = 𝑓 (𝑚)𝑎 + 𝑓 (𝑚 0)𝑏 ∈ N0
since 𝑓 (𝑚) and 𝑓 (𝑚 0) both belong to N0 and N0 is a submodule of N.
Hence 𝑚𝑎 + 𝑚 0𝑏 belongs to 𝑓 −1 (N0). 

Example (3.2.5). — Let A be a ring and let M, N be two A-modules. The

set HomAb (M, N) of all morphisms of abelian groups from M to N is
again an abelian group, the sum of to morphisms 𝑓 and 𝑔 being the
124 CHAPTER 3. MODULES

morphism 𝑓 + 𝑔 given by 𝑚 ↦→ 𝑓 (𝑚) + 𝑔(𝑚). If 𝑓 and 𝑔 are A-linear,

then 𝑓 + 𝑔 is A-linear too, because

( 𝑓 + 𝑔)(𝑚𝑎) = 𝑓 (𝑚𝑎) + 𝑔(𝑚𝑎) = 𝑓 (𝑚)𝑎 + 𝑔(𝑚)𝑎

= ( 𝑓 (𝑚)𝑎 + 𝑔(𝑚)𝑎) = ( 𝑓 + 𝑔)(𝑚)𝑎,

for 𝑎 ∈ A and 𝑚, 𝑛 ∈ M. Similarly, the zero map is a morphism, as

well as the map − 𝑓 : 𝑚 ↦→ − 𝑓 (𝑚) for any morphism 𝑓 ∈ HomA (M, N).
Consequently, HomA (M, N) is a subgroup of Hom(M, N).
When M = N, EndA (M) is moreover a ring, whose multiplication is
given by composition of morphisms; it is a subring of End(M).
Assume that A is a commutative ring. Let 𝑓 ∈ HomA (M, N) and let
𝑎 ∈ A. Then, the map 𝑓 𝑎 defined by 𝑚 ↦→ 𝑓 (𝑚)𝑎 is a morphism of
abelian groups, but is also A-linear, since

( 𝑓 𝑎)(𝑚𝑏) = 𝑓 (𝑏𝑚)𝑎 = 𝑓 (𝑚)𝑏𝑎 = 𝑓 (𝑚)𝑎𝑏 = ( 𝑓 𝑎)(𝑚)𝑏,

for any 𝑚 ∈ M and any 𝑏 ∈ A. This endowes the abelian

group HomA (M, N) with the structure of an A-module.
However, when the ring A is not commutative, the map 𝑚 ↦→ 𝑓 (𝑚)𝑎 is
not necessarily A-linear, and the abelian group HomA (M, N) may not
have any natural structure of an A-module.
Assume however that N be a (B, A)-bimodule. One can then endow
HomA (M, N) with the structure of a left B-module by defining a linear
map 𝑏 𝑓 , for 𝑓 ∈ HomA (M, N) and 𝑏 ∈ B, by the formula (𝑏 𝑓 )(𝑚) = 𝑏 𝑓 (𝑚).
(It is clearly a morphism of abelian groups; the equalities 𝑏 𝑓 (𝑚𝑎) =
𝑏( 𝑓 (𝑚𝑎)) = 𝑏 · 𝑓 (𝑚) · 𝑎 = (𝑏 𝑓 (𝑚))𝑎 prove that it is A-linear.) Moreover,
one has (𝑏𝑏 0) 𝑓 = 𝑏(𝑏 0 𝑓 ) for any 𝑏, 𝑏 0 ∈ B and any 𝑓 ∈ HomA (M, N).
Similarly, if M is a (B, A)-bimodule, one can endow HomA (M, N)
with a structure of right B-module by defining, for 𝑓 ∈ HomA (M, N)
and 𝑏 ∈ B, a linear map 𝑓 𝑏 by the formula ( 𝑓 𝑏)(𝑚) = 𝑓 (𝑏𝑚). One has
𝑓 (𝑏𝑏 0) = ( 𝑓 𝑏)𝑏 0 for 𝑏, 𝑏 0 ∈ B and 𝑓 ∈ HomA (M, N). Indeed, for any
𝑚 ∈ M, one has ( 𝑓 𝑏𝑏 0)(𝑚) = 𝑓 (𝑏𝑏 0 𝑚); on the other hand, ( 𝑓 𝑏)𝑏 0 maps 𝑚
to ( 𝑓 𝑏)(𝑏 0 𝑚) = 𝑓 (𝑏𝑏 0 𝑚), hence the relation 𝑓 𝑏𝑏 0 = ( 𝑓 𝑏)𝑏 0.

Remark (3.2.6). — We have explained how an abelian group M has a

natural structure of an End(M)-module. Similarly, if A is a ring, than
 3.3. OPERATIONS ON MODULES 125

any right A-module M is endowed with a canonical structure of a left

EndA (M)-module.
Assume that A is commutative, so that EndA (M) is now naturally an
A-module. Let, moreover, 𝑢 ∈ EndA (M). By the universal property of
polynomial rings (proposition 1.3.16), there is a unique ring morphism
𝜇𝑢 : A[X] → EndA (M) that coincides with 𝜇 on A and maps X to 𝑢. It
maps a polynomial P = 𝑝 𝑛 X𝑛 to the endomorphism P(𝑢) = 𝑝 𝑛 𝑢 𝑛 .
Í Í
This endowes the A-module M with a structure of an A[X]-module.
This construction will be of great use in the case where A is a field,
thanks to the facts that A[X] is a principal ideal domain and to the
classification of modules over such rings.

Definition (3.2.7). — Let A be a ring, let M be a right A-module. The dual

of M, written M∨ , is the abelian group HomA (M, A𝑑 ) endowed with its natural
structure of a left A-module (for which (𝑎𝜑)(𝑚) = 𝑎𝜑(𝑚) for any 𝑎 ∈ A,
𝑚 ∈ M and 𝜑 ∈ M∨ ). Its elements are called linear forms on M.

Similarly, the dual of a left A-module is the abelian group HomA (M, A𝑠 )
endowed with its natural structure of a right A-module for which
(𝜑𝑎)(𝑚) = 𝜑(𝑚)𝑎 for any 𝑎 ∈ A, 𝑚 ∈ M and 𝜑 ∈ M∨ .

3.3. Operations on modules

Proposition (3.3.1). — Let A be a ring, let M be a right A-module and let
Ñ
(N𝑠 )𝑠∈S be a family of submodules of M. Then, its intersection N = 𝑠 N𝑠 is
a submodule of M.

Proof. — Since 0 ∈ N𝑠 for every 𝑠, one has 0 ∈ N. Let 𝑚 and 𝑛 be two

elements of N. For any 𝑠, 𝑚 and 𝑛 belong to N𝑠 , hence so does 𝑚 + 𝑛,
so that 𝑚 + 𝑛 belongs to N. Finally, let 𝑚 ∈ N and 𝑎 ∈ A. For every 𝑠,
𝑚 ∈ N𝑠 , hence 𝑚𝑎 ∈ N𝑠 and finally 𝑚𝑎 ∈ N. Therefore, N is a submodule
of M. 

Proposition (3.3.2). — Let A be a ring, let M be a right A-module and let X be

a subset of M. There exists a smallest submodule hXi of M that contains X: it
is the intersection of the family of all submodules of M which contain X. It is
126 CHAPTER 3. MODULES

also the set of sums 𝑥∈X 𝑥𝑎 𝑥 , where (𝑎 𝑥 )𝑥∈X runs among the set of all almost
Í
null families of elements of A.

One says that hXi is the submodule of M generated by X.

Proof. — The intersection hXi of all of the submodules of M that contain X
is a submodule of M; it contains X. By construction, hXi is contained in
every submodule of M which contains X; it is therefore the smallest of
them all.
Let (𝑎 𝑥 )𝑥 be an almost-null family of elements of A; then, 𝑥∈X 𝑥𝑎 𝑥 is a
Í
linear combination of elements in hXi, hence belongs to hXi. This shows
that the set hXi0 of all such linear combinations is contained in hXi.
To obtain the other inclusion, let us first show that hXi0 is a submodule
of M. First of all, 0 = 𝑥 𝑥0 belongs to hXi0. On the other hand, let 𝑚 and
Í
𝑛 be two elements of hXi0, let (𝑎 𝑥 )𝑥 and (𝑏 𝑥 )𝑥 be two almost-null families
such that 𝑚 = 𝑥 𝑥𝑎 𝑥 and 𝑛 = 𝑥 𝑥𝑏 𝑥 . Then, the family (𝑎 𝑥 + 𝑏 𝑥 )𝑥 is
Í Í
almost-null and one has
Õ Õ Õ
𝑚+𝑛 = 𝑥𝑎 𝑥 + 𝑥𝑏 𝑥 = 𝑥(𝑎 𝑥 + 𝑏 𝑥 )



𝑥 𝑥 𝑥

so that 𝑚 + 𝑛 belongs to hXi0. Finally, let 𝑚 ∈ hXi0 and 𝑎 ∈ A, let (𝑎 𝑥 )𝑥 be

an almost-null family such that 𝑚 = 𝑥 𝑥𝑎 𝑥 . Then, 𝑚𝑎 = 𝑥 𝑥(𝑎 𝑥 𝑎), so
Í Í
that 𝑚𝑎 ∈ hXi0. This concludes the proof that hXi0 is a submodule of M.
Since it contains X, one has hXi ⊂ hXi0, the other desired inclusion. 

Definition (3.3.3). — Let A be a ring, let M be a right A-module and let

Í
(M𝑠 )𝑠∈S be a family of submodules of M. Its sum, written 𝑠∈S M𝑠 , is the
Ð
submodule of M generated by the union 𝑠 M𝑠 of the submodules M𝑠 .

It is also the set of all linear combinations 𝑠 𝑚 𝑠 where (𝑚 𝑠 )𝑠 is an

Í
almost-null family of elements of M such that 𝑚 𝑠 ∈ M𝑠 for every 𝑠 ∈
S. Indeed, this set of linear combinations is a submodule of M, it
Ð
contains 𝑠 M𝑠 , and is contained in every submodule of M which
contains all of the M𝑠 .

Definition (3.3.4). — Let A be a ring, let (M𝑠 )𝑠∈S be a family of right A-modules.
Î
Its direct product is the set 𝑠∈S M𝑠 endowed with the laws:
(𝑚 𝑠 )𝑠 + (𝑛 𝑠 )𝑠 = (𝑚 𝑠 + 𝑛 𝑠 )𝑠 , (𝑚 𝑠 )𝑠 𝑎 = (𝑚 𝑠 𝑎)𝑠
 3.3. OPERATIONS ON MODULES 127

which endow it with the structure of a right A-module. É

Î
The direct sum of the family (M𝑠 ) is the submodule 𝑠∈S M𝑠 of 𝑠∈S M𝑠
consisting of elements (𝑚 𝑠 )𝑠 such that 𝑚 𝑠 = 0 for all but finitely many 𝑠 ∈ S.

Remark (3.3.5). — If all of É

the M𝑠 are equal to a given module M, one
(S)
Î
writes 𝑠∈S M𝑠 = M and
S
𝑠∈S M𝑠 = M .

Lemma (3.3.6). — Let (M𝑠 )𝑠∈S be a family of right A-modules. For every 𝑡 ∈ S,
define maps
Ê Ö
𝑖 𝑡 : M𝑡 → M𝑠 , 𝑝𝑡 : M 𝑠 → M𝑡
𝑠 𝑠

by 𝑖 𝑡 (𝑚) = (𝑚 𝑠 )𝑠∈S where 𝑚𝑡 = 𝑚 and 𝑚 𝑠 = 0 for 𝑠 ≠ 𝑡, and 𝑝 𝑡 ((𝑚 𝑠 )𝑠∈S ) = 𝑚𝑡 .

These are morphisms of right A-modules.

Proof. — Let 𝑡 ∈ S, let 𝑚, 𝑛 ∈ M𝑡 , let 𝑎, 𝑏 ∈ A. Then,

𝑖 𝑡 (𝑚𝑎 + 𝑛𝑏) = (0, . . . , 0, 𝑚𝑎 + 𝑛𝑏, 0, . . . )

(in the right hand side, the term 𝑚𝑎 + 𝑛𝑏 has index 𝑡)

= (0, . . . , 0, 𝑚, 0, . . . )𝑎 + (0, . . . , 0, 𝑛, 0, . . . , 0)𝑏

= 𝑖 𝑡 (𝑚)𝑎 + 𝑖 𝑡 (𝑛)𝑏.

Consequently, 𝑖 𝑡 is a morphism of A-modules. We leave as an exercise

the proof that 𝑝 𝑡 is a morphism. 
The morphism 𝑖 𝑡 is injective, and called the canonical injection of
index 𝑡; the morphism 𝑝 𝑡 is surjective; it is called the canonical surjection
of index 𝑡.
Direct products and direct sums of modules satisfy a universal property
which we now state.

Theorem (3.3.7). — Let A be a ring and let (M𝑠 )𝑠∈S be a family of right
A-modules.
a) For every right A-module N and any family ( 𝑓𝑠 )𝑠∈S , where 𝑓𝑠 : N → M𝑠
is a morphism, there exists a unique morphism 𝑓 : N → 𝑠 M𝑠 such that
Î
𝑝 𝑠 ◦ 𝑓 = 𝑓𝑠 for every 𝑠 ∈ S.
128 CHAPTER 3. MODULES

b) For every right A-module N and any family ( 𝑓𝑠 )É

𝑠∈S , where 𝑓𝑠 : M𝑠 → N
is a morphism, there exists a unique morphism 𝑓 : 𝑠 M 𝑠 → M such that
𝑓 ◦ 𝑖 𝑠 = 𝑓𝑠 for every 𝑠 ∈ S.

Proof. — a) Assume that 𝑓 : N → 𝑠 M𝑠 satisfies 𝑝 𝑠 ◦ 𝑓 = 𝑓𝑠 for every 𝑠 ∈

Î
S. Then, if 𝑛 ∈ N and 𝑓 (𝑛) = (𝑚 𝑠 )𝑠 , we have

𝑚 𝑠 = 𝑝 𝑠 ((𝑚 𝑠 )𝑠 ) = 𝑝 𝑠 ( 𝑓 (𝑛)) = (𝑝 𝑠 ◦ 𝑓 )(𝑛) = 𝑓𝑠 (𝑛),

so that there is at most one morphism 𝑓 satisfying 𝑝 𝑠 ◦ 𝑓 = 𝑓𝑠 for all 𝑠.

Conversely, let us define a map 𝑓 : N → 𝑠 M𝑠 by 𝑓 (𝑛) = ( 𝑓𝑠 (𝑛))𝑠∈S for
Î
𝑛 ∈ N and let us show that it is a morphism. Indeed, for any 𝑎, 𝑏 ∈ A
and any 𝑚, 𝑛 ∈ N,

𝑓 (𝑚𝑎 + 𝑛𝑏) = 𝑓𝑠 (𝑚𝑎 + 𝑛𝑏) = 𝑓𝑠 (𝑚)𝑎 + 𝑓𝑠 (𝑛)𝑏

𝑠 𝑠
= 𝑓𝑠 (𝑚) 𝑠 𝑎 + 𝑓𝑠 (𝑛) 𝑠 𝑏 = 𝑓 (𝑚)𝑎 + 𝑓 (𝑛)𝑏,

so that 𝑓 is linear.
b) Assume that 𝑓 : 𝑠 M 𝑠 → N satisfies 𝑓 ◦ 𝑖 𝑠 = 𝑓𝑠 for every 𝑠. Then,
É

the image by 𝑓 of any element (0, . . . , 0, 𝑚, 0, . . . ) = 𝑖 𝑠 (𝑚) (where

É𝑚 ∈ M𝑠
has index 𝑠) is necessarily equal to 𝑓𝑠 (𝑚). Any element 𝑚 of M𝑠 is a
family (𝑚 𝑠 )𝑠∈S , where 𝑚 𝑠 ∈ M𝑠 for every 𝑠, almost finitely many of them
being nonzero. It follows that 𝑚 = 𝑠 𝑖 𝑠 (𝑚 𝑠 ) (the sum is finite)
Í

Õ Õ Õ
𝑓 (𝑚) = 𝑓 ( 𝑖 𝑠 (𝑚 𝑠 )) = ( 𝑓 ◦ 𝑖 𝑠 )(𝑚 𝑠 ) = 𝑓𝑠 (𝑚 𝑠 ),
𝑠 𝑠 𝑠

Éa map 𝑓 . Conversely, the map 𝑓 :

É
hence the uniqueness of such M𝑠 →
M defined for any (𝑚 𝑠 )𝑠 ∈ M𝑠 by
Õ
𝑓 ((𝑚 𝑠 )𝑠 ) = 𝑓𝑠 (𝑚 𝑠 ) (finite sum)
𝑠

is a morphism of A-modules and satisfies 𝑓 ◦ 𝑖 𝑠 = 𝑓𝑠 for every

É 𝑠 ∈ S. Let
indeed 𝑎, 𝑏 ∈ A, and (𝑚 𝑠 )𝑠∈S , (𝑛 𝑠 )𝑠∈S be two elements of 𝑠 M 𝑠 ; then,
 3.3. OPERATIONS ON MODULES 129

one has
𝑓 ((𝑚 𝑠 )𝑠 𝑎 + (𝑛 𝑠 )𝑠 𝑏) = 𝑓 ((𝑚 𝑠 𝑎 + 𝑛 𝑠 𝑏)𝑠 )
Õ Õ
𝑓𝑠 (𝑚 𝑠 𝑎 + 𝑛 𝑠 𝑏) = 𝑓𝑠 (𝑚 𝑠 )𝑎 + 𝑓𝑠 (𝑛 𝑠 )𝑏


=
𝑠
Õ Õ𝑠
𝑓𝑠 (𝑚 𝑠 ) 𝑎 + 𝑓𝑠 (𝑛 𝑠 ) 𝑏



=
𝑠 𝑠
= 𝑓 ((𝑚 𝑠 )𝑠 )𝑎 + 𝑓 ((𝑛 𝑠 )𝑠 )𝑏.


Remark (3.3.8). — One can reformulate this theorem as follows: for every
right A-module N, the maps
Ê Ö
HomA ( M𝑠 , M) → HomA (M𝑠 , M), 𝑓 ↦→ ( 𝑓 ◦ 𝑖 𝑠 )𝑠
𝑠 𝑠
and
Ö Ö
HomA (M, M𝑠 ) → HomA (M, M𝑠 ), 𝑓 ↦→ (𝑝 𝑠 ◦ 𝑓 )𝑠
𝑠 𝑠
are bijections. In fact, they are isomorphisms of abelian groups (and of
A-modules if A is commutative). É
Î
In the language of category theory, this says that 𝑠 M𝑠 and 𝑠 M𝑠
(endowed with the morphisms 𝑖 𝑠 and 𝑝 𝑠 ) are respectively a product and
a coproduct of the family (M𝑠 ).

3.3.9. Internal direct sums. — Let M be a right A-module and let

(M𝑠 )𝑠∈S be a family of submodules of M. Then, there is a morphism of
𝑚 𝑠 for every almost-
É Í
A-modules, 𝑠 M 𝑠 → M, defined by (𝑚 𝑠 ) →
↦
null family (𝑚 𝑠 )𝑠∈S , where 𝑚 𝑠 ∈ M𝑠 for every 𝑠. The image of this
Í
morphism is 𝑠 M𝑠 . One says that the submodules M𝑠 are in direct sum
if this morphism is an isomorphism; this means that any element of M
can be written in a unique way as a sum 𝑠∈S 𝑚 𝑠 , with 𝑚 𝑠 ∈ M𝑠 for
Í
everyÉ 𝑠, all but finitely many of them being null. In that case, one writes
M= 𝑠∈S M 𝑠 (“internal” direct sum).
When the set S has two elements, say S = {1, 2}, the kernel of this
morphism is the set of all pairs (𝑚, −𝑚), where 𝑚 ∈ M1 ∩ M2 , and its
image is the submodule M1 + M2 . Consequently, M1 and M2 are in direct
130 CHAPTER 3. MODULES

sum if and only if M1 ∩ M2 = 0 and M1 + M2 = M. The picture is slightly

more complicated for families indexed by a set with 3 or more elements;
this is already the case for vector spaces, see exercise 3/9.
Let M be a right A-module, let N be a submodule of M; a direct summand
of N is a submodule P of M such that M = N ⊕ P (N and P are in direct
sum). Contrarily to the case of vector spaces, not any submodule has a
direct summand. If a submodule N of M has a direct summand, one
also says that N is a direct summand.

Definition (3.3.10). — Let A be a ring, let M be right A-module and let I

be a right ideal of A. One defines the submodule MI of M as the submodule
generated by all products 𝑚𝑎, for 𝑚 ∈ M and 𝑎 ∈ I.

𝑚 𝑖 𝑎 𝑖 , where 𝑎 𝑖 ∈ I and
Í
It is the set of all finite linear combinations
𝑚 𝑖 ∈ M for every 𝑖.

3.4. Quotients of modules

Let A be a ring and let M be a right A-module. We are interested in
the equivalence relations ∼ on M which are compatible with its module
structure, namely such that for any 𝑚, 𝑚 0 , 𝑛, 𝑛 0 ∈ M, and any 𝑎, 𝑏 ∈ A,

if 𝑚 ∼ 𝑚 0 and 𝑛 ∼ 𝑛 0, then 𝑚𝑎 + 𝑛𝑏 ∼ 𝑚 0 𝑎 + 𝑛 0𝑏.

Let N be the set of all 𝑚 ∈ M such that 𝑚 ∼ 0. Since an equivalence

relation is reflexive, one has 0 ∈ N. If 𝑚, 𝑛 ∈ N, then 𝑚 ∼ 0 and 𝑛 ∼ 0,
hence 𝑚𝑎 + 𝑛𝑏 ∼ (𝑎0 + 𝑏0) = 0, hence 𝑚𝑎 + 𝑛𝑏 ∈ N. This shows that N
is a submodule of M. Moreover, if 𝑚 and 𝑛 are elements of M which
are equivalent for ∼, then the relations 𝑚 ∼ 𝑛 and (−𝑛) ∼ (−𝑛) imply
that 𝑚 − 𝑛 ∼ 0, that is, 𝑚 − 𝑛 ∈ N. Conversely, if 𝑚, 𝑛 ∈ M are such that
𝑚 − 𝑛 ∈ N, then 𝑚 − 𝑛 ∼ 0, and using that 𝑛 ∼ 𝑛, we obtain 𝑚 ∼ 𝑛.
Conversely, let N be a submodule of M and let ∼ be the relation on M
defined by “𝑚 ∼ 𝑛 if and only if 𝑚 − 𝑛 ∈ N”. This is an equivalence
relation on M. Indeed, since 0 ∈ N, 𝑚 ∼ 𝑚 for any 𝑚 ∈ M; if 𝑚 ∼ 𝑛, then
𝑚 − 𝑛 ∈ N, hence 𝑛 − 𝑚 = −(𝑚 − 𝑛) ∈ N and 𝑛 ∼ 𝑚; finally, if 𝑚 ∼ 𝑛 and
𝑛 ∼ 𝑝, then 𝑚−𝑛 and 𝑛−𝑝 belong to N, so that 𝑚−𝑝 = (𝑚−𝑛)+(𝑛−𝑝) ∈ N
and 𝑚 ∼ 𝑝. Moreover, this equivalence relation is compatible with its
 3.4. QUOTIENTS OF MODULES 131

structure of module: if 𝑚 ∼ 𝑚 0, 𝑛 ∼ 𝑛 0 and 𝑎, 𝑏 ∈ A, then

(𝑚 0 𝑎 + 𝑛 0𝑏) − (𝑚𝑎 + 𝑛𝑏) = (𝑚 − 𝑚 0)𝑎 + (𝑛 − 𝑛 0)𝑏 ∈ N,
since 𝑚 − 𝑚 0 ∈ N and 𝑛 − 𝑛 0 ∈ N, hence 𝑚 0 𝑎 + 𝑛 0𝑏 ∼ 𝑚𝑎 + 𝑛𝑏.
Let M/N be the set of all equivalence classes on M for this relation,
and let clN : M → M/N be the canonical projection. (If not confusion
can arise, this map shall only be written cl.) From the above calculation,
we get the following theorem:

Theorem (3.4.1). — Let A be a ring, let M be an A-module and let N be a

submodule of M. The relation ∼ on M given by 𝑚 ∼ 𝑛 if and only if 𝑚 − 𝑛 ∈ N
is an equivalence relation on M which is compatible with its structure of module.
The quotient set M/N has a unique structure of an A-module for which the
map clN : M → M/N is a morphism of A-modules.
The map clN is surjective; its kernel is N.

We now prove a factorization theorem, the universal property of quotient

modules.

Theorem (3.4.2). — Let A be a ring, let M be an A-module and let N be a

submodule of M. For any A-module P and any morphism 𝑓 : M → P such
that N ⊂ Ker( 𝑓 ), there exists a unique morphism 𝜑 : M/N → P such that
𝑓 = 𝜑 ◦ clN .
Moreover, Im(𝜑) = Im( 𝑓 ) and Ker(𝜑) = clN (Ker( 𝑓 )). In particular, 𝜑 is
injective if and only if Ker( 𝑓 ) = N, and 𝜑 is surjective if and only if 𝑓 is
surjective.

One can represent graphically the equality 𝑓 = 𝑓˜ ◦ cl of the theorem

by saying that the diagram
𝑝
→ P
←
M
→
←

𝜑
→

clN ←
M/N
is commutative. This theorem allows to factor any morphism 𝑓 : M → N
of A-modules as a composition
clN 𝜑
M −−→ M/Ker( 𝑓 ) −
→ Im( 𝑓 ) ↩→ N
132 CHAPTER 3. MODULES

of a surjective morphism, an isomorphism, and an injective morphism.

Proof. — Necessarily, 𝜑(cl(𝑚)) = 𝑓 (𝑚) for any 𝑚 ∈ M. Since every
element of M/N is of the form cl(𝑚) for a certain 𝑚 ∈ M, this shows that
there is at most one morphism 𝜑 : M/N → P such that 𝜑 ◦ cl = 𝑓 .
Let us show its existence. Let 𝑥 ∈ M/N and let 𝑚, 𝑚 0 ∈ M be two
elements such that 𝑥 = cl(𝑚) = cl(𝑚 0). Then 𝑚 − 𝑚 0 ∈ N, hence
𝑓 (𝑚 − 𝑚 0) = 0 since N ⊂ Ker( 𝑓 ). Consequently, 𝑓 (𝑚) = 𝑓 (𝑚 0) and one
may define 𝜑 by setting 𝜑(𝑥) = 𝑓 (𝑚), where 𝑚 is any element of M such
that cl(𝑚) = 𝑥, the result does not depend on the chosen element 𝑚.
It remains to show that 𝜑 is a morphism. So let 𝑥, 𝑦 ∈ M/N and let
𝑎, 𝑏 ∈ A. Let us choose 𝑚, 𝑛 ∈ M such that 𝑥 = cl(𝑚) and 𝑦 = cl(𝑛).
Then 𝑥𝑎 + 𝑦𝑏 = cl(𝑚)𝑎 + cl(𝑛)𝑏 = cl(𝑚𝑎 + 𝑛𝑏) hence

𝜑(𝑥𝑎 + 𝑦𝑏) = 𝜑(cl(𝑚𝑎 + 𝑛𝑏)) = 𝑓 (𝑚𝑎 + 𝑛𝑏)

= 𝑓 (𝑚)𝑎 + 𝑓 (𝑛)𝑏 = 𝜑(𝑥)𝑎 + 𝜑(𝑦)𝑏.

Hence, 𝜑 is linear.
It is obvious that Im( 𝑓 ) ⊂ Im(𝜑). On the other hand, if 𝑝 ∈ Im(𝜑), let
us choose 𝑥 ∈ M/N such that 𝑝 = 𝜑(𝑥), and 𝑚 ∈ M such that 𝑥 = cl(𝑚).
Then, 𝑝 = 𝜑(cl(𝑚)) = 𝑓 (𝑚) ∈ Im( 𝑓 ), so that Im(𝜑) ⊂ Im( 𝑓 ), hence the
equality.
Finally, let 𝑥 ∈ M/N be such that 𝜑(𝑥) = 0; write 𝑥 = cl(𝑚) for some
𝑚 ∈ M. Since 𝑓 (𝑚) = 𝜑 ◦ cl(𝑚) = 𝜑(𝑥) = 0, we have 𝑚 ∈ Ker( 𝑓 ).
Conversely, if 𝑥 = cl(𝑚) for some 𝑚 ∈ Ker( 𝑓 ), then 𝜑(𝑥) = 𝜑(cl(𝑚)) =
𝑓 (𝑚) = 0, hence 𝑥 ∈ Ker 𝜑. This shows that Ker(𝜑) = cl(Ker( 𝑓 )). 

Example (3.4.3). — Let A be a ring and let 𝑓 : M → N be a morphism

of A-modules. The A-module N/ 𝑓 (M) is called the cokernel of 𝑓 and is
denoted by Coker( 𝑓 ).
One has Coker( 𝑓 ) = 0 if and only 𝑓 is surjective. Consequently, a
morphism of A-modules 𝑓 : M → N is an isomorphism if and only if
Ker( 𝑓 ) = 0 and Coker( 𝑓 ) = 0.

The next proposition describes the submodules of a quotient mod-

ule M/N.
 3.4. QUOTIENTS OF MODULES 133

Proposition (3.4.4). — Let A be a ring, let M be an A-module, let N be a

submodule of M. The map cl−1 :

submodules of M/N → submodules of M containing N

𝒫 ↦→ cl−1
N (𝒫)

is a bijection.

So, for any submodule P of M such that N ⊂ P, there is exactly

one submodule 𝒫 of M/N such that P = cl−1 N (𝒫). Moreover, one has
𝒫 = clN (P). The submodule clN (P) of M/N will be written P/N. This is a
coherent notation. Indeed, the restriction of cl to the submodule P is a
morphism clN | P : P → M/N with kernel P∩N = N and image clN (P). By
the factorization theorem, clN | P induces an isomorphism P/N → clN (P).
Proof. — The proof is an immediate consequence of the two following
formulae: if P is a submodule of M, then

cl−1
N (clN (P)) = P + N

while if 𝒫 is a submodule of M/N,

clN (cl−1
N (𝒫)) = 𝒫.

Indeed, if P ⊂ N, P + N = P and these formulae precisely show that the

map cl−1N in the statement of the proposition is bijective, with reciprocal
map clN .
Let us show the first formula. Let 𝑚 ∈ cl−1N (cl(P)); we have clN (𝑚) ∈
clN (P), hence there is 𝑝 ∈ P such that clN (𝑚) = clN (𝑝) and it follows
that clN (𝑚 − 𝑝) = 0, hence 𝑛 = 𝑚 − 𝑝 ∈ N; then, 𝑚 = 𝑝 + 𝑛 belongs
to P + N. Conversely, if 𝑚 ∈ P + N, write 𝑚 = 𝑝 + 𝑛 for some 𝑝 ∈ P and
some 𝑛 ∈ N; then, clN (𝑚) = clN (𝑝 + 𝑛) = clN (𝑝) belongs to clN (P), hence
𝑚 ∈ cl−1N (clN (P)).
Let us now show the second formula. By definition, one has
clN (cl−1
N (𝒫)) ⊂ 𝒫. Conversely, if 𝑥 ∈ 𝒫, let 𝑚 ∈ M be such
𝑥 = clN (𝑚). Then, clN (𝑚) ∈ 𝒫 so that 𝑚 ∈ cl−1 N (𝒫), hence
𝑥 = clN (𝑚) ∈ clN (cl−1
N (𝒫)). 

We now compute quotients of quotients.

134 CHAPTER 3. MODULES

Proposition (3.4.5). — Let A be a ring, let M be an A-module and let N, P be

two submodules of M such that N ⊂ P. Then, there is a unique map
(M/P) → (M/N)/(P/N)
such that clP (𝑚) mapsto clP/N (clN (𝑚)) for every 𝑚 ∈ M, and this map is an
isomorphism.

Proof. — The uniqueness of such a map follows from the fact that every
element of M/P is of the form clP (𝑚), for some 𝑚 ∈ M. Let 𝑓 be the
morphism given by
𝑓 : M → (M/N) → (M/N)/(P/N), 𝑚 ↦→ clP/N (clN (𝑚)).
It is surjective, as the composition of two surjective morphisms. An
element 𝑚 ∈ M belongs to Ker( 𝑓 ) if and only if clN (𝑚) ∈ Ker clP/N =
P/N = clN (P); since P contains N, this is equivalent to 𝑚 ∈ P. Conse-
quently, Ker( 𝑓 ) = P and the factorization theorem (theorem 3.4.2) asserts
that there is a unique morphism 𝜑 : M/P → (M/N)/(P/N) such that
𝜑(clP (𝑚)) = clP/N (clN (𝑚)) for every 𝑚 ∈ M. Since Ker( 𝑓 ) = P, the mor-
phism 𝜑 is injective; since 𝑓 is surjective, the morphism 𝜑 is surjective.
Consequently, 𝜑 is an isomorphism. 

Remark (3.4.6). — Let M be an A-module, let N be a submodule of M. Assume

that N has a direct summand P in M. The morphism 𝑓 from P to M/N given
by the composition of the injection of P into M and of the surjection of M
onto M/N is an isomorphism.
Let indeed 𝑚 ∈ P be such that 𝑓 (𝑚) = 0. Then, 𝑚 ∈ N hence
𝑚 ∈ N ∩ P = 0; this shows that Ker( 𝑓 ) = 0 so that 𝑓 is injective.
Observe now that 𝑓 is surjective: for 𝑚 ∈ M, let 𝑛 ∈ N and 𝑝 ∈ P
be such that 𝑚 = 𝑛 + 𝑝; then clN (𝑚) = clN (𝑛) + clN (𝑝) = 𝑓 (𝑝), hence
Im 𝑓 = clN (M) = M/N.

3.5. Generating sets, free sets; bases

Definition (3.5.1). — Let A be a ring and let M be a right A-module.
One says that a family (𝑚 𝑖 )𝑖∈I of elements of M is:
– generating if the submodule of M generated by the 𝑚 𝑖 is equal to M;
 3.5. GENERATING SETS, FREE SETS; BASES 135

– free if for every almost-null family (𝑎 𝑖 )𝑖∈I of elements of A, the relation

𝑖∈I 𝑚 𝑖 𝑎 𝑖 = 0 implies 𝑎 𝑖 = 0 for every 𝑖 ∈ I;
Í

– bonded if it is not free;

– a basis of M if it is free and generating.

On defines similar notions of generating subset, free subset, bonded subset,

and basis, as subsets S of M for which the family (𝑠)𝑠∈S is respectively
generating, free, bonded, and a basis.
When the ring is non null, the members of a free family are pairwise
distinct (otherwise, the linear combination 𝑥−𝑦 = 0, for 𝑥 = 𝑦, contradicts
the definition of a free family since 1 ≠ 0). Consequently, with regards to
these properties, the only interesting families (𝑚 𝑖 )𝑖∈I are those for which
the map 𝑖 ↦→ 𝑚 𝑖 is injective and the notions for families and subsets
correspond quite faithfully one to another.
A subfamily (subset) of a free family (subset) is again free; a family
(set) possessing a generating subfamily (subset) is generating.

Proposition (3.5.2). — Let A be a ring, let M be a right A-module. Let (𝑚 𝑖 )𝑖∈I

be a family of elements of M and let 𝜑 be the morphism given by
Õ
(I)
𝜑: A𝑑 → M, (𝑎 𝑖 )𝑖∈I ↦→ 𝑚𝑖 𝑎 𝑖 .
𝑖∈I

Then,
– the morphism 𝜑 is injective if and only if the family (𝑚 𝑖 )𝑖∈I is free;
– the morphism 𝜑 is surjective if and only if the family (𝑚 𝑖 )𝑖∈I is generating;
– the morphism 𝜑 is an isomorphism if and only if the family (𝑚 𝑖 )𝑖∈I is a
basis.

Proof. — The kernel of 𝜑 is the set of all almost-null families (𝑎 𝑖 ) such

that 𝑖∈I 𝑚 𝑖 𝑎 𝑖 = 0. So (𝑚 𝑖 )𝑖∈I is free if and only if Ker(𝜑) = 0, that is, if
Í
and only if 𝜑 is injective.
The image of 𝜑 is the set of all linear combinations of terms of (𝑚 𝑖 ). It
follows that Im(𝜑) = h{𝑚 𝑖 }i, so that 𝜑 is surjective if and only if (𝑚 𝑖 )𝑖∈I
is generating.
By what precedes, (𝑚 𝑖 ) is a basis if and only if 𝜑 is bijective, that is, an
isomorphism. 
136 CHAPTER 3. MODULES

(I)
Example (3.5.3). — Let I be a set and let M = A𝑑 . For every 𝑖 ∈ I, let
𝑒 𝑖 ∈ M be the family (𝛿 𝑖,𝑗 ) 𝑗∈I , whose only nonzero term is that of index 𝑖,
equal to 1. For every almost-null family (𝑎 𝑖 )𝑖∈I , one has (𝑎 𝑖 )𝑖∈I = 𝑖∈I 𝑒 𝑖 𝑎 𝑖 .
Í
(I)
Consequently, the family (𝑒 𝑖 )𝑖∈I is a basis of A𝑑 , called its canonical basis.
(I)
The module A𝑑 is often called the free right A-module on I.

Definition (3.5.4). — One says that a module is free if it has a basis. If a

module has a finite generating subset, it is said to be finitely generated.

Proposition (3.5.5). — Let M be an A-module, let N be a submodule of M.

a) If M is finitely generated, then M/N is finitely generated;
b) If N and M/N are finitely generated, then M is finitely generated;
c) If N and M/N are free A-modules, then M is a free A-module.

However, with this notation, it may happen that M be finitely generated

but that N be not, as it may happen that M be free, but not N or M/N.
More precisely, we shall prove the following statements:
a0) If M has a generating subset of cardinality 𝑟, then so does M/N;
b0) If N and M/N have generating subsets of cardinalities respectively 𝑟
and 𝑠, then M has a generating subset of cardinality 𝑟 + 𝑠;
c0) If N and M/N have bases of cardinalities respectively 𝑟 and 𝑠, then M
has a basis of cardinality 𝑟 + 𝑠.
Proof. — a0) The images in M/N of a generating subset of M gener-
ate M/N, since the canonical morphism from M to M/N is surjective. In
particular, M/N is finitely generated if M is.
b0) Let (𝑛1 , . . . , 𝑛 𝑟 ) be a generating family of N, and let (𝑚1 , . . . , 𝑚 𝑠 ) be
a family of elements of M such that (clN (𝑚1 ), . . . , clN (𝑚 𝑠 )) generate M/N.
Let us show that the family (𝑚1 , . . . , 𝑚 𝑠 , 𝑛1 , . . . , 𝑛 𝑟 ) generates M.
Let 𝑚 ∈ M. By hypothesis, clN (𝑚) is a linear combination of
clN (𝑚1 ), . . . , clN (𝑚 𝑠 ). There thus exist elements 𝑎 𝑖 ∈ A such that
clN (𝑚) = 𝑠𝑖=1 clN (𝑚 𝑖 )𝑎 𝑖 . Consequently, 𝑛 = 𝑚 − 𝑠𝑖=1 𝑚 𝑖 𝑎 𝑖 belongs
Í Í
to N and there exist elements 𝑏 𝑗 ∈ A such that 𝑛 = 𝑟𝑗=1 𝑛 𝑗 𝑏 𝑗 . Then
Í
𝑚 = 𝑠𝑖=1 𝑚 𝑖 𝑎 𝑖 + 𝑟𝑗=1 𝑛 𝑗 𝑏 𝑗 is a linear combination of the 𝑚 𝑖 and of the 𝑛 𝑗 .
Í Í
c0) Let us moreover assume that (𝑛1 , . . . , 𝑛 𝑟 ) be a basis of N and
that (clN (𝑚1 ), . . . , clN (𝑚 𝑠 )) be a basis of M/N; let us show that
 3.5. GENERATING SETS, FREE SETS; BASES 137

(𝑚1 , . . . , 𝑚 𝑠 , 𝑛1 , . . . , 𝑛 𝑟 ) is a basis M. Since we already proved that

this family generates M, it remains to show that it is free. So let
0 = 𝑠𝑖=1 𝑚 𝑖 𝑎 𝑖 + 𝑟𝑗=1 𝑛 𝑗 𝑏 𝑗 be a linear dependence relation between
Í Í
these elements. Applying clN , we get a linear dependence relation
0 = 𝑠𝑖=1 clN (𝑚 𝑖 )𝑎 𝑖 for the family clN (𝑚 𝑖 ). Since this family is free, one
Í
has 𝑎 𝑖 = 0 for every 𝑖. It follows that 0 = 𝑟𝑗=1 𝑛 𝑗 𝑏 𝑗 ; since the family
Í
(𝑛1 , . . . , 𝑛 𝑟 ) is free, 𝑏 𝑗 = 0 for every 𝑗. The considered linear dependence
relation is thus trivial, as was to be shown. 

Corollary (3.5.6). — Let M be an A-module, let M1 , . . . , M𝑛 be submodules

of M which are finitely generated. Their sum 𝑛𝑖=1 M𝑖 is finitely generated.
Í
In particular, the direct sum of a finite family of finitely generated modules is
finitely generated.

Proof. — By induction, it suffices to treat the case of two modules, that

is, 𝑛 = 2. The second projection pr2 : M1 × M2 → M2 is a surjective
linear map, and its kernel, equal to M1 × {0}, is isomorphic to M1 , hence
is finitely generated. Since M1 and M2 are finitely generated, their direct
product M1 ×M2 is finitely generated. The morphism (𝑚1 , 𝑚2 ) ↦→ 𝑚1 +𝑚2
from M1 × M2 to M has image M1 + M2 ; consequently, M1 + M2 is finitely
generated. 

Remark (3.5.7). — When the ring A is commutative (and nonzero), or

when A is a division ring, we shall prove in the next section that all bases
of a finitely generated free module have the same cardinality. This is not
true in the general case. However, one can prove that if a module M
over a nonzero ring is finitely generated and free, then all bases of M are
finite (see exercise 3/12).
The case of a zero ring A = 0 is pathological: there is no other A-
module than the zero module M = 0, and every family (𝑚 𝑖 )𝑖∈I is a
basis...

Proposition (3.5.8). — Let A be a ring, let M be a free A-module and let (𝑒 𝑖 )𝑖∈I
be a basis of M. For any A-module N and any family (𝑛 𝑖 )𝑖∈I of elements of N,
there is a unique morphism 𝜑 : M → N such that 𝜑(𝑒 𝑖 ) = 𝑛 𝑖 for every 𝑖 ∈ I.
138 CHAPTER 3. MODULES

Proof. — The map A(I) → M given by (𝑎 𝑖 ) ↦→ 𝑒 𝑖 𝑎 𝑖 is an isomorphism

Í
of modules, so that M inherits from the universal property of direct
sums of modules. To give oneself a linear map from M to N is equivalent
to give oneself the images of the elements of a basis. The morphism 𝜑
such that 𝜑(𝑒 𝑖 ) = 𝑛 𝑖 for every 𝑖 is given by 𝜑( 𝑒 𝑖 𝑎 𝑖 ) = 𝑢(𝑒 𝑖 )𝑎 𝑖 .
Í Í


Corollary (3.5.9). — Let A be a ring, let M be a free A-module and let (𝑒 𝑖 )𝑖∈I
be a basis of M. There exists, for every 𝑖 ∈ I, a unique linear form 𝜑 𝑖 on M such
that 𝜑 𝑖 (𝑒 𝑖 ) = 1 and 𝜑 𝑖 (𝑒 𝑗 ) = 0 for 𝑗 ≠ 𝑖. For any element 𝑚 = 𝑖∈I 𝑒 𝑖 𝑎 𝑖 of M,
Í
one has 𝜑 𝑖 (𝑚) = 𝑎 𝑖 .
The family (𝜑 𝑖 )𝑖∈I is free.
If I is finite, then it is a basis of the dual module M∨ .

Proof. — Existence and uniqueness of the linear forms 𝜑 𝑖 follow from

the proposition.
Let (𝑎 𝑖 )𝑖∈I be an almost-null family such that 𝑎 𝑖 𝜑 𝑖 = 0. Apply this
Í
relation to 𝑒 𝑖 ; one gets 𝑎 𝑖 = 0. This proves that the family (𝜑 𝑖 ) is free.
Assume that I is finite and let us show that (𝜑 𝑖 ) is a generating family
in M∨ . Let 𝜑 be a linear form on M and let 𝑚 = 𝑖 𝑒 𝑖 𝑎 𝑖 be an element
Í
of M. One has 𝜑(𝑚) = 𝑖 𝜑(𝑒 𝑖 )𝑎 𝑖 , hence 𝜑(𝑚) = 𝑖 𝜑(𝑒 𝑖 )𝜑 𝑖 (𝑚). This
Í Í
proves that 𝜑 = 𝑖∈I 𝜑(𝑒 𝑖 )𝜑 𝑖 .
Í


When I is finite, the family (𝜑 𝑖 )𝑖∈I defined above is called the dual basis
of the basis (𝑒 𝑖 )𝑖∈I .

3.5.10. Matrices. — One of the interest of free modules is that they

allow matrix computations. Let Φ = (𝑎 𝑖,𝑗 ) ∈ M𝑚,𝑛 (A) be a matrix with
𝑚 rows and 𝑛 columns and entries in A. One associates to Φ a map
𝜑 : A𝑛 → A𝑚 given by the formula
𝑛
Õ
𝜑(𝑥 1 , . . . , 𝑥 𝑛 ) = ( 𝑎 𝑖,𝑗 𝑥 𝑗 )16𝑖6𝑚 .
𝑗=1

This is a morphism of right A-modules from (A𝑑 )𝑛 to (A𝑑 )𝑚 . Indeed, for

every 𝑥 = (𝑥 1 , . . . , 𝑥 𝑛 ) ∈ A𝑛 , every 𝑦 = (𝑦1 , . . . , 𝑦𝑛 ) ∈ A𝑛 , any 𝑎, 𝑏 ∈ A,
 3.5. GENERATING SETS, FREE SETS; BASES 139

one has

𝜑(𝑥𝑎 + 𝑦𝑏) = 𝜑((𝑥1 𝑎 + 𝑦1 𝑏, 𝑥2 𝑎 + 𝑦2 𝑏, . . . , 𝑥 𝑛 𝑎 + 𝑦𝑛 𝑏))

𝑛
Õ
= 𝑎 𝑖,𝑗 (𝑥 𝑗 𝑎 + 𝑦 𝑗 𝑏)
𝑗=1
𝑛
Õ 𝑛
Õ
𝑎 𝑖,𝑗 𝑥 𝑗 𝑎 + + 𝑎 𝑖,𝑗 𝑦 𝑗 𝑏



=
𝑗=1 𝑗=1

= 𝜑(𝑥)𝑎 + 𝜑(𝑦)𝑏.

Conversely, any morphism 𝜑 : A𝑛𝑑 → A𝑚 𝑑

of right A-modules has this
form.
Let indeed (𝑒1 , . . . , 𝑒 𝑛 ) be the canonical basis of (A𝑑 )𝑛 (recall that
𝑒 𝑗 ∈ A𝑛 has all of its coordinates equal to 0, but for the 𝑗th coordinate,
which is equal to 1). . We represent any element of (A𝑑 )𝑛 as a column
matrix, i.e., with one column and 𝑛 rows. Let Φ = (𝑎 𝑖,𝑗 ) be the matrix
with 𝑚 rows and 𝑛 columns whose 𝑗th column is equal to 𝜑(𝑒 𝑗 ). In
other words, we have 𝜑(𝑒 𝑗 ) = (𝑎1,𝑗 , . . . , 𝑎 𝑚,𝑗 ) for 1 6 𝑗 6 𝑛. For any
𝑥 = (𝑥 1 , . . . , 𝑥 𝑛 ) ∈ A𝑛 , one has 𝑥 = 𝑛𝑗=1 𝑒 𝑗 𝑥 𝑗 , hence
Í

𝑛
Õ 𝑛
Õ
𝜑(𝑥) = 𝜑(𝑒 𝑗 )𝑥 𝑗 = ( 𝑎 𝑖,𝑗 𝑥 𝑗 ) = ΦX,
𝑗=1 𝑗=1

if X is the column matrix (𝑥1 , . . . , 𝑥 𝑛 ). This shows that 𝜑 is given by the

𝑛 × 𝑚 matrix Φ = (𝑎 𝑖,𝑗 ).
The identity matrix I𝑛 is the 𝑛 × 𝑛 matrix representing the identity
endomorphism of A𝑛 ; its diagonal entries are equal to 1, all the other
entries are 0.
Let Φ0 = (𝑏 𝑗,𝑘 ) ∈ M𝑛,𝑝 be a matrix with 𝑛 rows and 𝑝 columns,
let 𝜑0 : (A𝑑 )𝑝 → (A𝑑 )𝑛 be the morphism it represents. The product
matrix Φ00 = ΦΦ0 has 𝑚 rows and 𝑝 columns and its coefficient 𝑐 𝑖,𝑘 with
index (𝑖, 𝑘) is given by the formula
𝑛
Õ
𝑐 𝑖,𝑘 = 𝑎 𝑖,𝑗 𝑏 𝑗,𝑘 .
𝑗=1
140 CHAPTER 3. MODULES

It represents the morphism 𝜑00 = 𝜑 ◦ 𝜑0 : (A𝑑 )𝑝 → (A𝑑 )𝑚 . Indeed one

has by definition
Õ Õ Õ
00
𝜑 (𝑥 1 , . . . , 𝑥 𝑝 ) = ( 𝑐 𝑖,𝑘 𝑥 𝑘 )16𝑖6𝑚 = ( 𝑎 𝑖,𝑗 𝑏 𝑗,𝑘 𝑥 𝑘 )𝑖 = ( 𝑎 𝑖,𝑗 𝑦 𝑗 )𝑖 ,
𝑘 𝑗,𝑘 𝑗

where (𝑦1 , . . . , 𝑦𝑛 ) = 𝜑0(𝑥1 , . . . , 𝑥 𝑝 ). Consequently,

𝜑00(𝑥1 , . . . , 𝑥 𝑝 ) = 𝜑(𝑦1 , . . . , 𝑦𝑛 ) = 𝜑(𝜑0(𝑥1 , . . . , 𝑥 𝑝 )),

so that 𝜑00 = 𝜑 ◦ 𝜑0.

This shows in particular that the map M𝑛 (A) → EndA (A𝑛𝑑 ) that sends
a square 𝑛 × 𝑛 matrix to the corresponding endomorphism of (A𝑑 )𝑛 is
an isomorphism of rings.
Here is the very reason why we preferred right A-modules!

3.6. Localization of modules (commutative rings)

Let A be a commutative ring and let S be a multiplicative subset of A.
Let M be an A-module. Through calculus of fractions we constructed
a localized ring S−1 A as well as a morphism of rings A → S−1 A. We
are now going to construct by a similar process an S−1 A-module S−1 M
together with a localization morphism of A-modules M → S−1 M.
Define a relation ∼ on the set M × S by the formula

(𝑚, 𝑠) ∼ (𝑛, 𝑡) ⇔ there exists 𝑢 ∈ S such that 𝑢(𝑡𝑚 − 𝑠𝑛) = 0.

Let us check that it is an equivalence relation. In fact, the argument runs

exactly as for the case of rings (p. 40). The equality 1 · (𝑠𝑛 − 𝑠𝑛) = 0 shows
that this relation is reflexive; if 𝑢(𝑡𝑚 − 𝑠𝑛) = 0, then 𝑢(𝑠𝑛 −𝑡𝑚) = 0, which
shows that it is symmetric. Finally, let 𝑚, 𝑛, 𝑝 ∈ M and 𝑠, 𝑡, 𝑢 ∈ S be such
that (𝑚, 𝑠) ∼ (𝑛, 𝑡) and (𝑛, 𝑡) ∼ (𝑝, 𝑢); let 𝑣, 𝑤 ∈ S be such 𝑣(𝑡𝑚 − 𝑠𝑛) = 0
and 𝑤(𝑡𝑝 − 𝑢𝑛) = 0; then

𝑡𝑣𝑤(𝑢𝑚 − 𝑠𝑝) = 𝑢𝑤 𝑣(𝑡𝑚 − 𝑠𝑛) + 𝑠𝑣𝑤 (𝑢𝑛 − 𝑡𝑝) = 0

so that (𝑚, 𝑠) ∼ (𝑝, 𝑢), since 𝑢𝑤 ∈ S and 𝑠𝑣 ∈ S: this relation is transitive.

Let S−1 M be the set of equivalence classes; for 𝑚 ∈ M and 𝑠 ∈ S, write
𝑚/𝑠 for the class in S−1 M of the pair (𝑚, 𝑠) ∈ S−1 M. We then define two
 3.6. LOCALIZATION OF MODULES (COMMUTATIVE RINGS) 141

laws on S−1 M: first, an addition, given by

(𝑚/𝑠) + (𝑛/𝑡) = (𝑡𝑚 + 𝑠𝑛)/(𝑠𝑡)
for 𝑚, 𝑛 ∈ M and 𝑠, 𝑡 ∈ S, and then an external multiplication, defined
by
(𝑎/𝑡)(𝑚/𝑠) = (𝑎𝑚)/(𝑡𝑠).
for 𝑚 ∈ M, 𝑎 ∈ A, 𝑠 and 𝑡 ∈ S,
Theorem (3.6.1). — Endowed with these two laws, S−1 M is an S−1 A-module.
If one views S−1 M as an A-module through the canonical morphism of rings
A → S−1 A, then the map 𝑖 : M → S−1 M given by 𝑖(𝑚) = 𝑚/1 is a morphism
of A-modules.
The proof is left as an exercise; anyway, the computations are totally
similar as those done for localization of rings.
Proposition (3.6.2). — Let A be a commutative ring, let S be a multiplicative
subset of A and let M be an A-module. Let 𝑖 : M → S−1 M be the canonical
morphism; for 𝑠 ∈ S, let 𝜇𝑠 : M → M be the morphism given by 𝑚 ↦→ 𝑠𝑚.
a) An element 𝑚 ∈ M belongs to Ker(𝑖) if and only if there exists 𝑠 ∈ S such
that 𝑠𝑚 = 0;
b) In particular, 𝑖 is injective if and only if 𝜇𝑠 is injective, for every 𝑠 ∈ S;
c) The morphism 𝑖 is an isomorphism if and only if 𝜇𝑠 is bijective, for every
𝑠 ∈ S.
Proof. — Let 𝑚 ∈ M. If 𝑖(𝑚) = 0, then 𝑚/1 = 0/1, hence there exists
𝑠 ∈ S such that 𝑠𝑚 = 0, by the definition of S−1 M. Conversely, if 𝑠𝑚 = 0,
then 𝑚/1 = 0/1, hence 𝑖(𝑚) = 0. This proves a), and assertion b) is a
direct consequence.
Let us prove c). Let us assume that 𝑖 is an isomorphism. All mor-
phisms 𝜇𝑠 are then injective by b). Moreover, in the S−1 A-module S−1 M,
multiplication by an element 𝑠 ∈ S is surjective, since 𝑠 · (𝑚/𝑠𝑡) = 𝑚/𝑡 for
every 𝑚 ∈ M and 𝑡 ∈ S. Therefore, 𝜇𝑠 is surjective as well, so that it is an
isomorphism. Conversely, let us assume that all morphisms 𝜇𝑠 : M → M
are isomorphisms. Then 𝑖 is injective, by b). Let moreover 𝑚 ∈ M and
𝑠 ∈ S; let 𝑚 0 ∈ M be such that 𝑠𝑚 0 = 𝑚; then 𝑚/𝑠 = 𝑚 0/1 = 𝑖(𝑚 0); this
proves that 𝑖 is surjective. 
142 CHAPTER 3. MODULES

Remark (3.6.3). — Let us recall a few examples of multiplicative subsets.

a) First of all, for any 𝑠 ∈ A, the set S = {1; 𝑠; 𝑠 2 ; . . .} is multiplicative.
In that case, the localized module S−1 M is denoted via an index 𝑠, so
that M𝑠 = S−1 M is an A𝑠 -module.
b) If P is a prime ideal of A, then S = A P is also a multiplicative
subset of A. Again, the localized module and ring are denoted via an
index P; that is AP = (A P)−1 A. and MP = (A P)−1 M.

Proposition (3.6.4). — Let A be a commutative ring, let S be a multiplicative

subset of A. Let 𝑓 : M → N be a morphism of A-modules. There exists a unique
morphism of S−1 A-modules 𝜑 : S−1 M → S−1 N such that 𝜑(𝑚/1) = 𝑓 (𝑚)/1
for every 𝑚 ∈ M. For every 𝑚 ∈ M, and every 𝑠 ∈ S, one has 𝜑(𝑚/𝑠) =
𝑓 (𝑚)/𝑠.

In other words, the diagram

𝑓
→ N
←
M
←

𝑖 𝑖
→

𝜑
S−1 M → S−1 N
←

is commutative.
Proof. — To check uniqueness, observe that 𝑚/𝑠 = (1/𝑠)(𝑚/1), so that if
such a morphism 𝜑 exists, then 𝜑(𝑚/𝑠) must be equal to (1/𝑠)𝜑(𝑚/1) =
(1/𝑠)( 𝑓 (𝑚)/1) = 𝑓 (𝑚)/𝑠. To establish the existence of 𝜑, we want to use
the formula 𝜑(𝑚/𝑠) = 𝑓 (𝑚)/𝑠 as a definition, but we must verify that it
makes sense, that is, we must prove that if 𝑚/𝑠 = 𝑛/𝑡, for some 𝑚, 𝑛 ∈ M
and 𝑠, 𝑡 ∈ S, then 𝑓 (𝑚)/𝑠 = 𝑓 (𝑛)/𝑡.
So assume that 𝑚/𝑠 = 𝑛/𝑡; by the definition of the equivalence relation,
there exists 𝑢 ∈ S such that 𝑢(𝑡𝑚 − 𝑠𝑛) = 0. Then,

𝑓 (𝑚) 𝑢𝑡 𝑓 (𝑚) 𝑓 (𝑢𝑡𝑚) 𝑓 (𝑢𝑠𝑛) 𝑓 (𝑛)

= = = = ,
𝑠 𝑢𝑡𝑠 𝑢𝑡𝑠 𝑢𝑡𝑠 𝑡

which shows that 𝜑 is well defined.

 3.6. LOCALIZATION OF MODULES (COMMUTATIVE RINGS) 143

Now, for 𝑚, 𝑛 ∈ M, 𝑠, 𝑡 ∈ S, we have

𝑚 𝑛
𝑡𝑚 + 𝑠𝑛
𝑓 (𝑡𝑚 + 𝑠𝑛)
𝜑 + =𝜑 =
𝑠 𝑡 𝑠𝑡 𝑠𝑡
𝑡 𝑓 (𝑚) 𝑠 𝑓 (𝑛) 𝑓 (𝑚) 𝑓 (𝑛) 𝑚 𝑛
= + = + = 𝜑( ) + 𝜑( ),
𝑠𝑡 𝑠𝑡 𝑠 𝑡 𝑠 𝑡
hence 𝜑 is additive Finally, for 𝑚 ∈ M, 𝑎 ∈ A, 𝑠 and 𝑡 ∈ S, we have

𝑎 𝑚
𝑎𝑚
𝑓 (𝑎𝑚) 𝑎 𝑓 (𝑚) 𝑎 𝑓 (𝑚) 𝑎 𝑚

𝜑 =𝜑 = = = = 𝜑
𝑠 𝑡 𝑠𝑡 𝑠𝑡 𝑠𝑡 𝑠 𝑡 𝑠 𝑡
which shows that 𝜑 is S−1 A-linear. 

Localization of modules gives rise to an universal property, analogous

to the one we established for localization of rings.

Corollary (3.6.5). — Let 𝑓 : M → N be a morphism of A-modules; assume

that for every 𝑠 ∈ S, the morphism 𝜇𝑠 : N → N defined by 𝑛 ↦→ 𝑠𝑛 is an
isomorphism. Then, there exists a unique morphism of A-modules 𝑓˜ : S−1 M →
N such that 𝑓˜(𝑚/1) = 𝑓 (𝑚)/1 for every 𝑚 ∈ M.

Proof. — In fact, if 𝜑 : S−1 M → S−1 N, 𝑚/𝑠 ↦→ 𝑓 (𝑚)/𝑠, is the morphism

constructed in the previous proposition, and 𝑖 : N → S−1 N is the
canonical morphism, then the desired property for 𝜑 is equivalent to
the equality 𝑖 ◦ 𝑓˜ = 𝜑. By proposition 3.6.2, 𝑖 is an isomorphism of
A-modules; this implies the existence and uniqueness of an A-linear
map 𝑓˜ : S−1 M → N such that 𝑓˜(𝑚/1) = 𝑓 (𝑚)/1 for every 𝑚 ∈ M. 

As we already observed for localization of rings and ideals, localization

behaves nicely with respect to submodules; this is the second appearance
of exactness of localization.

Proposition (3.6.6). — Let A be a commutative ring, let S be a multiplicative

subset of A. Let M be an A-module and let N be a submodule of M.
Then, the canonical morphism 𝚤˜ : S−1 N → S−1 M deduced from the injection
𝑖 : N → M induces an isomorphism from S−1 N to a submodule of S−1 M.
Moreover, the canonical morphism 𝑝˜ : S−1 M → S−1 (M/N) deduced from the
surjection 𝑝 : M → M/N is surjective and its kernel is 𝚤˜(S−1 N).
144 CHAPTER 3. MODULES

In practice, the morphism 𝚤˜ is removed from the notation, and we

still denote by S−1 N the submodule of S−1 M image of this morphism.
Passing to the quotient, we then obtain from 𝑝˜ an isomorphism
∼
S−1 M/S−1 N →
− S−1 (M/N).
Proof. — Let 𝑛 ∈ N and let 𝑠 ∈ S. The image in S−1 M of the element 𝑛/𝑠 ∈
S−1 N is still 𝑛/𝑠, but 𝑛 is now seen as an element of M. It vanishes if and
only if there exists 𝑡 ∈ S such that 𝑡𝑛 = 0 in M. But then 𝑡𝑛 = 0 in N so
that 𝑛/𝑠 = 0 in S−1 N. This shows that the map S−1 N → S−1 M is injective.
It is thus an isomorphism from S−1 N to its image in S−1 M.
Since 𝑝 is surjective, the morphism 𝑝˜ is surjective too. Indeed, any
element of S−1 (M/N) can be written clN (𝑚)/𝑠, for some 𝑚 ∈ M and
some 𝑠 ∈ S. Then, clN (𝑚)/𝑠 is the image of 𝑚/𝑠 ∈ S−1 M. Obviously, S−1 N
is contained in the kernel of 𝑝; ˜ indeed, 𝑝(cl
˜ N (𝑛)/𝑠) = clN (𝑝(𝑛))/𝑠 = 0.
On the other hand, let 𝑚 ∈ M and let 𝑠 ∈ S be such that 𝑚/𝑠 ∈ Ker(𝑝); ˜
then 0 = clN (𝑚)/𝑠, so that there exists 𝑡 ∈ S such that 𝑡 clN (𝑚) = 0
in M/N. This implies that 𝑡𝑚 ∈ N. Then, 𝑚/𝑠 = 𝑡𝑚/𝑡𝑠 = 𝚤˜(𝑡𝑚/𝑡𝑠)
belongs to 𝚤˜(S−1 N). This concludes the proof that Ker(𝑝)˜ = 𝚤˜(S−1 N). 

Proposition (3.6.7). — Let A be a commutative ring, let S be a multiplicative

subset of A and let M be an A-module. Let 𝑖 : M → S−1 M be the canonical
morphism.
Let 𝒩 be an S−1 A-submodule of S−1 M. Then N = 𝑖 −1 (𝒩) is an A-submodule
of M such that 𝒩 = S−1 N; it is the largest of all such submodules.

Proof. — First of all, S−1 N ⊂ 𝒩. Indeed, if 𝑚 ∈ N, then 𝑚/1 ∈ 𝒩 by

definition of N; it follows that 𝑚/𝑠 ∈ 𝒩 for every 𝑠 ∈ S. Conversely, let
𝑥 ∈ 𝒩. One may write 𝑥 = 𝑚/𝑠 for some 𝑚 ∈ M and 𝑠 ∈ S. This implies
that 𝑠𝑥 = 𝑚/1 belongs to 𝒩, hence 𝑚 ∈ N and 𝑥 ∈ S−1 N. We thus have
shown that 𝒩 = S−1 N.
Finally, let P be a submodule of M such that S−1 P = 𝒩. Let 𝑚 ∈ P; one
has 𝑚/𝑠 ∈ 𝒩, hence 𝑠(𝑚/𝑠) = 𝑚/1 ∈ 𝒩. Then, 𝑚 ∈ N, so that P ⊂ N.
This shows that N is the largest submodule of M such that S−1 N = 𝒩. 

Proposition (3.6.8). — Let A be a commutative ring, let S be a multiplicative

subset of A. Let M be an A-module and let (N𝑖 ) be a family of submodules
 3.7. VECTOR SPACES 145

of M. Then, one has the following equalities of submodules of S−1 M:

Õ Õ
S−1 N𝑖 = S−1 N𝑖 .
𝑖

Proof. — Let N = N𝑖 . For every 𝑖, N𝑖 ⊂ N, hence an inclusion S−1 N𝑖 ⊂

Í
S−1 N. It follows that 𝑖 S−1 N𝑖 ⊂ S−1 N. Conversely, let 𝑛/𝑠 ∈ S−1 N.
Í
There exists an almost null family (𝑛 𝑖 ), where 𝑛 𝑖 ∈ N𝑖 for every 𝑖, such
that 𝑛 = 𝑛 𝑖 . Then, 𝑛 𝑖 /𝑠 belongs to S−1 N𝑖 and 𝑛/𝑠 = 𝑖 (𝑛 𝑖 /𝑠) belongs
Í Í
to S−1 N𝑖 , so that S−1 N ⊂ S−1 N𝑖 .
Í Í


Corollary (3.6.9). — If M is a finitely generated A-module, then S−1 A is a

finitely generated module.
Proof. — Let (𝑚1 , . . . , 𝑚𝑛 ) be a generating family. For 𝑖 ∈ {1, . . . , 𝑛},
let N𝑖 = A𝑚 𝑖 . By construction, S−1 N𝑖 is generated by 𝑚 𝑖 as a S−1 A-
module. By the proposition, S−1 N𝑖 = S−1 M, hence S−1 M is finitely
Í
generated. 

3.7. Vector spaces

Let us recall that a vector space is a module over a division ring.
Proposition (3.7.1). — Let K be a division ring and let M be a right K-vector
space. Let F, B, G be subsets of M such that F ⊂ B ⊂ G. We assume that F is
free and that G is generating.
The following assertions are equivalent:
(i) B is a basis;
(ii) B is maximal among the free subsets of G containing F;
(iii) B is minimal among the generating subsets of G containing F.
Proof. — Let us start with a remark and prove that if G, G0 are subsets
of M such that G ( G0, and G is generating, then G0 is not free. Indeed,
let 𝑥 ∈ G0 G; let us write 𝑥 as a linear combination of elements of G,
𝑥 = 𝑔∈G 𝑔𝑎 𝑔 , where (𝑎 𝑔 ) 𝑔∈G is an almost-null family. Set 𝑎 𝑥 = −1 and
Í
𝑎 𝑦 = 0 for 𝑦 ∈ G0 G such that 𝑦 ≠ 𝑥. The family (𝑎 𝑔 ) 𝑔∈G0 is almost null,
but not identically null, and one has 𝑔∈G0 𝑔𝑎 𝑔 = 0, so that G0 is not free.
Í
We now prove the equivalence of the three given assertions.
146 CHAPTER 3. MODULES

Assume that B is a basis of M. Then, B is free and generating. By the

initial remark, for every subset G of M such that B ⊂ G and B ≠ G, then
G is not free. In other words, B is maximal among all free subsets of M,
which proves (i)⇒(ii).
Let B0 be a subset of M such that F ⊂ B0 ( B; by the initial remark, if B0
is generating, then B is not free. Consequently, B is minimal among the
generating subsets of G containing F, hence (i)⇒(iii).
We now assume that B is a free subset of G, maximal among those
containing F. Let us show that B is generating. This holds if G = B.
Otherwise, let 𝑚 ∈ G B. The subset B ∪ {𝑚} is not free, so that there are
elements 𝑎 and (𝑎 𝑏 )𝑏∈B of K, all but finitely many of them being zero, but
not all of them, such that 𝑏∈B 𝑏𝑎 𝑏 + 𝑚𝑎 = 0. If 𝑎 = 0, this is a non trivial
Í
linear dependence relation among the elements of B, contradicting the
hypothesis that B is free. So 𝑎 ≠ 0 and 𝑚 = − 𝑏∈B 𝑏𝑎 𝑏 𝑎 −1 belongs to the
Í
subspace M0 of M generated by B. Consequently, M0 contains G. Since G
is generating, M0 = M, and B generates M. It is thus a basis of M.
It remains to show that a generating subset of G which is minimal
among those containing F is a basis. If such a subset B were not free,
there would exist a non-trivial linear dependence relation 𝑏∈B 𝑏𝑎 𝑏 = 0
Í
among the elements of B. Let 𝛽 ∈ B be such that 𝑎 𝛽 ≠ 0; we have
𝛽 = − 𝑏≠𝛽 𝑏𝑎 𝑏 𝑎 𝛽−1 . It follows that 𝛽 belongs to the vector subspace M0
Í

generated by the elements of B {𝛽}. Consequently, M0 contains B,

hence M0 = M since B is generating. In particular, B {𝑏} is generating,
contradicting the minimality hypothesis on B. 

Proposition (3.7.2) (Exchange property). — Let K be a division ring, let M

be a right K-vector space. Let (𝑢𝑖 )∈I and (𝑣 𝑗 ) 𝑗∈J be families of elements of M.
One assumes that (𝑢𝑖 ) is generating and that (𝑣 𝑗 ) is free. For every 𝑘 ∈ J, there
exists 𝑖 ∈ I such that the family (𝑤 𝑗 ) 𝑗∈J defined by 𝑤 𝑘 = 𝑢𝑖 and 𝑤 𝑗 = 𝑣 𝑗 for
𝑗 ≠ 𝑘 is free.
In other words, given a free subset and a generating subset of a vector
set, one can replace any chosen element of the free subset by some
element of the generating set without losing its property of being free.
Proof. — Set J0 = J {𝑘} and let M0 be the subspace of M generated
by the family (𝑣 𝑗 ) 𝑗∈J0 . Let 𝑖 ∈ I and assume that we cannot replace 𝑣 𝑘
 3.7. VECTOR SPACES 147

by 𝑢𝑖 . This means that the family (𝑤 𝑗 ) 𝑗∈J defined by 𝑤 𝑘 = 𝑢𝑖 and 𝑤 𝑗 = 𝑣 𝑗

for 𝑗 ∈ J0 is not free; let 𝑗∈J 𝑤 𝑗 𝑎 𝑗 = 0 be a nontrivial linear dependence
Í
relation. Since the family (𝑣 𝑗 ) 𝑗∈J0 is free, one has 𝑎 𝑘 ≠ 0. Then we can
write Õ
−1
𝑢𝑖 = 𝑤 𝑘 = 𝑤 𝑘 𝑎 𝑘 𝑎 𝑘 = − 𝑣 𝑗 𝑎 𝑗 𝑎 −1
𝑘
𝑗∈J0
so that 𝑤 𝑘 ∈ M0. Consequently, if the exchange property does not hold,
then all 𝑢𝑖 belong to M0. Since the family (𝑢𝑖 ) is generating, one has
M0 = M. But then, 𝑣 𝑘 belongs to M0, hence one may write 𝑣 𝑘 = 𝑗∈J0 𝑣 𝑗 𝑏 𝑗 ,
Í
for some almost-null family (𝑏 𝑗 ) 𝑗∈J0 . Passing 𝑣 𝑘 to the right hand side, this
gives a linear dependence relation for the family (𝑣 𝑗 ) 𝑗∈J , contradicting
the hypothesis that it is free. 
One of the fundamental results in the theory of vector spaces is the
following theorem, presumably well known the reader in the case of
(commutative) fields!
Theorem (3.7.3). — Let K be a division ring and let M be a right K-vector
space.
a) M has a basis.
b) More precisely, if F is free subset of M and G is a generating subset of M
such that F ⊂ G, there exists a basis B of M such that F ⊂ B ⊂ G.
c) All bases of M have the same cardinality.
The common cardinality of all bases of M is called the dimension of M
and is denoted dimK (M), or dim(M) if no confusion on K can arise.
Proof. — Assertion a) results from b), applied to the free subset L = ∅
and to the generating subset G = M.
Let us prove b). Let ℱ be the set of all free subsets of G which contain F,
ordered by inclusion. Let (F𝑖 ) be a totally ordered family of free subsets
of G and let F0 be their union if the family is nonempty, and let F0 = F
otherwise. Let us observe that F0 is free. We may assume that the family
is nonempty. Let then (𝑎 𝑚 )𝑚∈F0 be a an almost-null family of elements
of K such that we have a linear dependence relation 0 = 𝑚𝑎 𝑚 between
Í
the members of F0. Let F00 be be the set of all 𝑚 ∈ F0 such that 𝑎 𝑚 ≠ 0; it is
a finite set. By induction on the cardinality on F00, there is an index 𝑖 such
that F00 ⊂ F𝑖 . Set 𝑎 𝑚 = 0 for 𝑚 ∈ F𝑖 F00; we have 𝑚∈F𝑖 𝑚𝑎 𝑚 = 0. Since
Í
148 CHAPTER 3. MODULES

F𝑖 is free, 𝑎 𝑚 = 0 for every 𝑚 ∈ F𝑖 . In particular, 𝑎 𝑚 = 0 for every 𝑚 ∈ F0.

This shows that the set F0 is free. Since it is an upper bound for the
family (F𝑖 ), the ordered set ℱ is inductive.
By Zorn’s lemma (theorem A.2.11), the ordered set ℱ has a maximal
element B. By proposition 3.7.1, the set B is a basis of M and F ⊂ B ⊂ G
by construction.
c) Let B and B0 be two bases of M. Since B is free and B0 is generating,
lemma 3.7.4 asserts the existence of an injection 𝑓 : B ↩→ B0. On the
other hand, since B0 is free and B is generating, there is also an injection
𝑔 : B0 ↩→ B. By Cantor-Bernstein’s theorem (theorem A.2.1), the bases B
and B0 are equipotent. 

Lemma (3.7.4). — Let K be a division ring, let M be a right K-vector space.

Let F and G be subsets of M such that F is free and G is generating. Then there
exists an injection from F into G.

Proof. — Let Φ be the set of all pairs (F0 , 𝜑) where F0 is a subset of F and
𝜑 is an injective map from F0 to G such that the set (F F0) ∪ 𝜑(F0) is free.
We endow this set Φ with the ordering for which (F01 , 𝜑1 ) ≺ (F02 , 𝜑2 ) if
and only if F01 ⊂ F02 and 𝜑2 (𝑚) = 𝜑1 (𝑚) for every 𝑚 ∈ F01 .
This set Φ is nonempty, because the pair (∅, 𝜑) belongs to Φ, where
𝜑 is the unique map from the empty set to G. Let us show that Φ is
inductive. So let (F0𝑖 , 𝜑 𝑖 )𝑖 be a totally ordered family of elements of Φ, let
F0 be the union of all subsets F0𝑖 and let 𝜑 be the map from F0 to G which
coincides with 𝜑 𝑖 on F0𝑖 . The map 𝜑 is well-defined: if an element 𝑚
of F0 belongs both to F0𝑖 and F0𝑗 , we may assume that (F0𝑖 , 𝜑 𝑖 ) ≺ (F0𝑗 , 𝜑 𝑗 ),
and then 𝜑 𝑗 (𝑚) = 𝜑 𝑖 (𝑚).
Let us show that (F0 , 𝜑) belongs to Φ. The map 𝜑 is injective: let 𝑚, 𝑚 0
be distinct elements of F0; since the family (F0𝑖 , 𝜑 𝑖 ) is totally ordered,
there exists an index 𝑖 such that 𝑚 and 𝑚 0 both belong to F0𝑖 ; since 𝜑 𝑖 is
injective, 𝜑 𝑖 (𝑚) ≠ 𝜑 𝑖 (𝑚 0); then 𝜑(𝑚) ≠ 𝜑(𝑚 0).
For 𝑚 ∈ F F0, set 𝜑(𝑚) = 𝑚, and let (𝑎 𝑚 )𝑚∈F be an almost-null family
of elements of K such that 𝑚∈F 𝜑(𝑚)𝑎 𝑚 = 0. There exists an index 𝑖
Í
such that all elements 𝑚 ∈ F0 for which 𝑎 𝑚0 ≠ 0 belong to F0𝑖 . Then, we
get a linear dependence relation among the members of (F F0𝑖 ) ∪ 𝜑(F0𝑖 ),
so that 𝑎 𝑚 = 0 for every 𝑚 ∈ F, as was to be shown.
 3.7. VECTOR SPACES 149

By Zorn’s lemma (theorem A.2.11), Φ has a maximal element (F0 , 𝜑).

Let us show that F0 = F. Otherwise, let 𝜇 ∈ F F0. For every 𝑚 ∈ F F0,
write 𝜑(𝑚) = 𝑚. By the exchange property (proposition 3.7.2) applied
to the free family (𝜑(𝑚))𝑚∈F , to the index 𝜇 and to the generating set G,
there exists an element 𝜈 ∈ G such that the family (𝜑0(𝑚))𝑚∈F is free,
where 𝜑0 : F → M coincides with 𝜑 on F {𝜇}, and 𝜑0(𝜇) = 𝜈. Define
F01 = F0 ∪ {𝜇} and let 𝜑1 ∪ F01 → G be the map given by 𝜑1 (𝜇) = 𝜈
and 𝜑1 (𝑚) = 𝜑(𝑚) for 𝑚 ∈ F0. The map 𝜑1 is injective. Consequently,
(F01 , 𝜑1 ) belongs to Φ and is stricly bigger than (F0 , 𝜑), contradicting the
maximality hypothesis on (F0 , 𝜑).
So 𝜑 : F → G is injective, establishing the lemma. 

Remark (3.7.5). — Let K be a division ring and let M be a right K-vector

space. When proving the preceding theorem, we made use of the axiom
of choice, by way of Zorn’s lemma. This cannot be avoided in general.
However, if M is finitely generated, then usual induction is enough.
The first question of exercise 3/30 suggests a simple direct proof of
lemma 3.7.4 in that case, leading to a simplification of the proof that all
bases of a (finitely generated) vector space have the same cardinality.
Corollary (3.7.6). — Let A be a non null commutative ring and let M be an
A-module. If M is free, then all bases of M have the same cardinality.
This cardinality is called the rank of the free A-module M, and is
denoted by rkA (M).
Proof. — Let J be a maximal ideal of A, so that the quotient ring K = A/J
is a field. Let MJ be the submodule of M generated by all products 𝑚𝑎,
for 𝑎 ∈ J and 𝑚 ∈ M; set V = M/MJ. This is an A-module. However, if
𝑎 ∈ J and 𝑚 ∈ M, then clMJ (𝑚)𝑎 = clMJ (𝑚𝑎) = 0; consequently, V can be
viewed as an A/J-module, that is, a K-vector space. (See also §3.1.5.)
Assume that M is free and let (𝑒 𝑖 )𝑖∈I be a basis of M. Let us show
that the family (clMJ (𝑒 𝑖 ))𝑖∈I is a basis of V. First, since the 𝑒 𝑖 generate M,
this family generates V as an A-module, hence as a K-vector space. It
remains to show that it is free. Let (𝛼 𝑖 )𝑖∈I be an almost-null family of
elements of K such that 𝑖∈I clMJ (𝑒 𝑖 )𝛼 𝑖 . For every 𝑖 ∈ I such that 𝛼 𝑖 ≠ 0,
Í
let us choose 𝑎 𝑖 ∈ A such that clJ (𝑎 𝑖 ) = 𝛼 𝑖 ; if 𝛼 𝑖 = 0, set 𝑎 𝑖 = 0. Then, the
Í
family (𝑎 𝑖 )𝑖∈I is almost-null and 𝑖∈I clMJ (𝑒 𝑖 ) clJ (𝑎 𝑖 ) = 0. In other words,
150 CHAPTER 3. MODULES

𝑚 = 𝑖∈I 𝑒 𝑖 𝑎 𝑖 belongs to MJ. Since MJ is generated by the elements of

Í
the form 𝑒 𝑖 𝑎 for 𝑖 ∈ I and 𝑎 ∈ J, there is a family (𝑏 𝑖 )𝑖∈I of elements of J
such that 𝑚 = 𝑖∈I 𝑒 𝑖 𝑏 𝑖 . The family (𝑒 𝑖 ) being free, one has 𝑎 𝑖 = 𝑏 𝑖 for
Í
every 𝑖 ∈ I. Consequently, 𝛼 𝑖 = clJ (𝑎 𝑖 ) = clJ (𝑏 𝑖 ) = 0 for every 𝑖 ∈ I. The
linear dependence relation that we have considered is thus trivial, so
that the family (𝑒 𝑖 )𝑖∈I is free, as claimed.
Consequently, this family is a basis of the K-vector space V. This
implies that dimK (V) = Card(I) and dimK (V) is the common cardinality
of all bases of M. 

Theorem (3.7.7). — Let K be a division ring and let M be a K-vector space.

Every subspace N of M has a direct summand P and one has
dim(N) + dim(P) = dim(M).
Proof. — Let B1 be a basis of N. By theorem 3.7.3, there exists a basis B
of M which contains B1 . The subset B2 = B B1 of M is then free
and generates a vector subspace P of M such that N + P = M. Let us
show that N ∩ P = 0. So let 𝑚 ∈ N ∩ P; one can write 𝑚 both as a
linear combination of elements of B1 and as a linear combination of
elements of B2 . Subtracting these relations, we get a linear dependence
relation between the members of M. If 𝑚 ≠ 0, this relation is nontrivial,
contradicting the hypothesis that B is free. Hence N ∩ P = 0 and P is a
direct summand of N in M.
By definition, one has dim(N) = Card(B1 ) and dim(M) = Card(B).
Moreover, B2 is free and generates P, so that dim(P) = Card(B2 ). Since
B1 and B2 are disjoint, one has
dim(M) = Card(B) = Card(B1 ∪ B2 )
= Card(B1 ) + Card(B2 ) = dim(N) + dim(P),
as was to be shown. 
Many classical results in the theory of vector spaces over a (commu-
tative) field that remain true, with the same proof, in the more general
context of vector spaces over a division ring. Let us give two examples.
Corollary (3.7.8). — Let K be a division ring. Let V, W be K-vector spaces
and let 𝑓 : V → W be a K-linear map.
 3.7. VECTOR SPACES 151

a) One has dim(Im( 𝑓 )) + dim(Ker( 𝑓 )) = dim(V);

b) If dim(V) > dim(W), then 𝑓 is not injective;
c) Assume that dim(V) = dim(W). Then the following properties are
equivalent: (i) 𝑓 is injective; (ii) 𝑓 is surjective; (iii) 𝑓 is bijective.

Proof. — a) Passing to the quotient, the morphism 𝑓 defines an injective

K-linear map 𝜑 : V/Ker( 𝑓 ) → W, which induces an isomorphism from
V/Ker( 𝑓 ) to Im(𝜑) = Im( 𝑓 ), so that dim(Im( 𝑓 )) = dim(V/Ker( 𝑓 )). Let
V0 be a direct summand of Ker( 𝑓 ) in V; one has dim(V) = dim(Ker( 𝑓 )) +
dim(V0). Finally, the canonical surjection clKer( 𝑓 ) : V → V/Ker( 𝑓 ) in-
duces, by restriction, a bijective morphism from V0 to V/Ker( 𝑓 ), so that
dim(V0) = dim(V/Ker( 𝑓 )). Consequently, we have
dim(Im( 𝑓 )) + dim(Ker( 𝑓 )) = dim(V/Ker( 𝑓 )) + dim(Ker( 𝑓 ))
= dim(V0) + dim(Ker( 𝑓 )) = dim(V).
b) Since Im( 𝑓 ) ⊂ W, one has dim(Im( 𝑓 )) 6 dim(W). If dim(V) >
dim(W), this forces dim(Ker( 𝑓 )) > 0, hence 𝑓 is not injective.
c) If 𝑓 is injective, then dim(Im( 𝑓 )) = dim(V) = dim(W), hence Im( 𝑓 ) =
W and 𝑓 is surjective. If 𝑓 is surjective, then dim(Ker( 𝑓 )) = 0, hence
Ker( 𝑓 ) = 0 and 𝑓 is injective. This implies the equivalence of the three
statements. 

Corollary (3.7.9). — Let K be a division ring. Let V, W be K-vector spaces

and let 𝑓 : V → W be a K-linear map.
a) The map 𝑓 is injective if and only if it has a left inverse;
b) The map 𝑓 is surjective if and only if it has a right inverse.

Proof. — If a map 𝑓 has a left inverse (resp. a right inverse), then it

is injective (resp. surjective). It thus suffices to prove the converse
assertions.
a) Assume that 𝑓 is injective, so that 𝑓 induces an isomorphism 𝑓1
from V to its image 𝑓 (V). Let W1 be a direct summand of 𝑓 (V) in W
(theorem 3.7.7) and let 𝑝 : W → 𝑓 (V) be the projector with image 𝑓 (V)
and kernel W1 . The morphism 𝑔 = 𝑓1−1 ◦ 𝑝 : W → V satisfies 𝑔 ◦ 𝑓 = idV ,
so that 𝑓 is left invertible.
152 CHAPTER 3. MODULES

b) Assume that 𝑓 is surjective. Let V1 be a direct summand of Ker( 𝑓 )

in V (theorem 3.7.7). Then, the morphism 𝑓1 = 𝑓 | V1 : V1 → W is injective
and surjective (for 𝑤 ∈ W, let 𝑣 ∈ V such that 𝑓 (𝑣) = 𝑤, and write
𝑣 = 𝑣1 + 𝑚, where 𝑣 1 ∈ V1 and 𝑚 ∈ Ker( 𝑓 ); one has 𝑓 (𝑣1 ) = 𝑓 (𝑣) = 𝑤),
hence an isomorphism. Let 𝑗 : V1 → V be the injection and ℎ = 𝑗 ◦ 𝑓1−1 ;
this is a morphism from W to V such that 𝑓 ◦ ℎ ◦ 𝑓1 = 𝑓 ◦ 𝑗 = 𝑓1 , hence
𝑓 ◦ ℎ = idW since 𝑓1 is surjective. Consequently, 𝑓 is right invertible. 

3.8. Alternate multilinear forms. Determinants

Definition (3.8.1). — Let A be a commutative ring, let M and N be A-modules,
let 𝑝 be a positive integer. A map 𝑓 : M𝑝 → N is said to be 𝑝-linear map if it
is linear with respect to each variable: for every (𝑚1 , . . . , 𝑚 𝑝 ) ∈ M𝑝 , the map
𝑚 ↦→ 𝑓 (𝑚1 , . . . , 𝑚 𝑖−1 , 𝑚, 𝑚 𝑖+1 , . . . , 𝑚 𝑝 ) is A-linear.
One says that a 𝑝-linear map 𝑓 : M𝑝 → N is alternate if for ev-
ery (𝑚1 , . . . , 𝑚 𝑝 ) ∈ M𝑝 for which there exist 𝑖 ≠ 𝑗 such that 𝑚 𝑖 = 𝑚 𝑗 , one has
𝑓 (𝑚1 , . . . , 𝑚 𝑝 ) = 0.

When N = A, a 𝑝-linear map from M to A is called a 𝑝-linear form.

Let A be a commutative ring, let M and N be A-modules, let 𝑝 be
a positive integer. Let 𝑓 be a 𝑝-linear alternate map from M to N; let
(𝑚1 , . . . , 𝑚 𝑝 ) ∈ M𝑝 , let 𝑖 and 𝑗 be distinct elements in {1, . . . , 𝑝} and let
us apply the alternating property to the family obtained by replacing
both 𝑚 𝑖 and 𝑚 𝑗 by their sum 𝑚 𝑖 + 𝑚 𝑗 . Expanding, one gets (assuming
𝑖 < 𝑗 to fix ideas)
0 = 𝑓 (𝑚1 , . . . , 𝑚 𝑖−1 , 𝑚 𝑖 + 𝑚 𝑗 , . . . , 𝑚 𝑗−1 , 𝑚 𝑖 + 𝑚 𝑗 , . . . , 𝑚 𝑝 )
= 𝑓 (𝑚1 , . . . , 𝑚 𝑖−1 , 𝑚 𝑖 , . . . , 𝑚 𝑗−1 , 𝑚 𝑗 , . . . , 𝑚 𝑝 )
+ 𝑓 (𝑚1 , . . . , 𝑚 𝑖−1 , 𝑚 𝑗 , . . . , 𝑚 𝑗−1 , 𝑚 𝑖 , . . . , 𝑚 𝑝 ),
since the two other terms vanish. In other words, 𝑓 is changed into its
opposite when two variables are exchanged. This property is called
antisymmetry.
Now, any permutation 𝜎 of {1, . . . , 𝑛} can be written as a product
of finitely many permutations of two elements; the number of such
permutations is even or odd, according to the signature 𝜀(𝜎) of 𝜎 being 1
 3.8. ALTERNATE MULTILINEAR FORMS. DETERMINANTS 153

or −1. One thus gets

(3.8.1.1) 𝑓 (𝑚 𝜎(1) , . . . , 𝑚 𝜎(𝑛) ) = 𝜀(𝜎) 𝑓 (𝑚1 , . . . , 𝑚𝑛 ).

Conversely, let 𝑓 be a 𝑝-linear antisymmetric map. If there are

two indices 𝑖 ≠ 𝑗 such that 𝑚 𝑖 = 𝑚 𝑗 , exchanging 𝑚 𝑖 and 𝑚 𝑗 changes
𝑓 (𝑚1 , . . . , 𝑚 𝑝 ) both into itself and its opposite, so that 𝑓 (𝑚1 , . . . , 𝑚 𝑝 ) =
− 𝑓 (𝑚1 , . . . , 𝑚 𝑝 ). Consequently, 2 𝑓 (𝑚1 , . . . , 𝑚 𝑝 ) = 0 hence 2 𝑓 is alternate.
If 2 is regular in A or, more generally, if the multiplication by 2 in the
module N is injective, then 𝑓 is alternate too.
However, if A = Z/2Z, any symmetric 𝑝-linear map is antisymmetric.
For example, the 2-linear form (𝑥1 , 𝑥2 ) ↦→ 𝑥 1 𝑥2 on the free A-module A
is antisymmetric, but it is not alternate.
The set of all alternate 𝑝-linear maps from M to N is denoted Λ𝑝 (M, N);
it is naturally a A-module. By definition, Λ1 (M, N) = HomA (M, N). On
the other hand, the A-module Λ0 (M, N) identifies with N. Indeed, M0 is
a singleton so that a map 𝑓 : M0 → N is just an element of N, and every
such map is linear in each variable (since there is no variable).
Let 𝑢 : M → M0 be a morphism of A-modules. For any 𝑓 ∈ Λ𝑝 (M0 , N),
the map (𝑚1 , . . . , 𝑚 𝑝 ) ↦→ 𝑓 (𝑢(𝑚1 ), . . . , 𝑢(𝑚 𝑝 )) is a 𝑝-linear alternate map
from M to N, denoted Λ𝑝 (𝑢)( 𝑓 ), or simply 𝑢 ∗ ( 𝑓 ). The map 𝑓 ↦→ Λ𝑝 (𝑢)( 𝑓 )
is a morphism of A-modules from Λ𝑝 (M0 , N) to Λ𝑝 (M, N). One has
Λ𝑝 (idM ) = idΛ𝑝 (M,N) , and Λ𝑝 (𝑢 ◦ 𝑣) = Λ𝑝 (𝑣) ◦ Λ𝑝 (𝑢). When N is fixed,
the construction M ↦→ Λ𝑝 (M, N) is thus a contravariant functor in M.
Let 𝑣 : N → N0 be a morphism of A-modules, For any 𝑓 ∈ Λ𝑝 (M, N),
the map 𝑣 ◦ 𝑓 : (𝑚1 , . . . , 𝑚 𝑝 ) ↦→ 𝑣( 𝑓 (𝑚1 , . . . , 𝑚 𝑝 )) is a 𝑝-linear alternate
map from M to N0, also denoted Λ𝑝 (𝑣)( 𝑓 ), or 𝑣∗ ( 𝑓 ). The map 𝑓 ↦→
Λ𝑝 (𝑣)( 𝑓 ) is a morphism of A-modules from Λ𝑝 (M, N) to Λ𝑝 (M, N0). One
has Λ𝑝 (idN ) = idΛ𝑝 (M,N) , and Λ𝑝 (𝑢 ◦ 𝑣) = Λ𝑝 (𝑢) ◦ Λ𝑝 (𝑣). When M is fixed,
the construction N ↦→ Λ𝑝 (M, N) is thus a covariant functor in N.

Proposition (3.8.2). — Let A be a commutative ring. Let M be an A-module

and let (𝑚1 , . . . , 𝑚𝑛 ) be a generating family of M. For any A-module P and
any integer 𝑝 > 𝑛, one has Λ𝑝 (M, P) = 0.

Proof. — Let 𝑓 : M𝑝 → P be an alternate 𝑝-linear map. Let (𝑥1 , . . . , 𝑥 𝑝 ) ∈

M𝑝 . By assumption, there exist elements 𝑎 𝑖,𝑗 (for 1 6 𝑖 6 𝑛 and 1 6 𝑗 6 𝑝)
154 CHAPTER 3. MODULES

such that 𝑥 𝑗 = 𝑎 𝑖,𝑗 𝑚 𝑖 . Since 𝑓 is 𝑝-linear, one has

Í

𝑛
Õ 𝑛
Õ
𝑓 (𝑥 1 , . . . , 𝑥 𝑝 ) = ... 𝑎 𝑖1 ,1 . . . 𝑎 𝑖 𝑝 ,𝑝 𝑓 (𝑚 𝑖1 , . . . , 𝑚 𝑖 𝑝 ).
𝑖 1 =1 𝑖 𝑝 =1

Since 𝑝 > 𝑛, in each 𝑝-tuple (𝑖1 , . . . , 𝑖 𝑝 ) of elements of {1, . . . , 𝑛}, at least

two terms are equal. The map 𝑓 being alternate, one has 𝑓 (𝑚 𝑖1 , . . . , 𝑚 𝑖 𝑝 ) =
0. Consequently, 𝑓 (𝑥1 , . . . , 𝑥 𝑝 ) = 0 and 𝑓 = 0. 
More generally, the above proof shows that an alternate 𝑝-linear map
is determined by its values on the 𝑝-tuples of distinct elements among
(𝑚1 , . . . , 𝑚𝑛 ).

Proposition (3.8.3). — Let M be a free A-module. A family (𝑚1 , . . . , 𝑚 𝑝 ) of

elements of M is free if and only if there exists, for any 𝑎 ∈ A {0}, a 𝑝-linear
alternate form 𝑓 on M such that 𝑎 𝑓 (𝑚1 , . . . , 𝑚 𝑝 ) ≠ 0.

Proof. — Assume that the family (𝑚1 , . . . , 𝑚 𝑝 ) is bonded and let us

consider a non trivial linear dependence relation 𝑎 1 𝑚1 + · · · + 𝑎 𝑝 𝑚 𝑝 = 0.
Up to reordering the family (𝑚1 , . . . , 𝑚 𝑝 ), we assume that 𝑎1 ≠ 0. Then
for any alternate 𝑝-linear form 𝑓 on M, one has

0 = 𝑓 (𝑎1 𝑚1 + · · · + 𝑎 𝑝 𝑚 𝑝 , 𝑚2 , . . . , 𝑚 𝑝 )
𝑝
Õ
= 𝑎 𝑖 𝑓 (𝑚 𝑖 , 𝑚2 , . . . , 𝑚 𝑝 )
𝑖=1
= 𝑎1 𝑓 (𝑚1 , . . . , 𝑚 𝑝 ).

Let us now assume that the family (𝑚1 , . . . , 𝑚 𝑝 ) is free and let us show
the desired property by induction on 𝑝.
The property holds for 𝑝 = 0, for the 0-alternate form equal to the
constant 1 is adequate. Let us assume that the property holds for 𝑝−1 and
let us prove it for 𝑝. Let 𝑎 ∈ A {0}. Since the family (𝑚1 , . . . , 𝑚 𝑝−1 ) is free,
there exists an alternate (𝑝−1)-linear form 𝑓 such that 𝑎 𝑓 (𝑚1 , . . . , 𝑚 𝑝−1 ) ≠
0. Then, the map
𝑝
Õ
𝑓 0 : M𝑝 → M, 𝑓 0(𝑥 1 , . . . , 𝑥 𝑝 ) = (−1) 𝑗 𝑓 (𝑥1 , . . . , 𝑥b𝑗 , . . . , 𝑥 𝑝 )𝑥 𝑗
𝑗=1
 3.8. ALTERNATE MULTILINEAR FORMS. DETERMINANTS 155

from M𝑝 to M is a 𝑝-linear alternate map. Assume that 𝑎 𝑓 0(𝑚1 , . . . , 𝑚 𝑝 ) =

0. This gives a linear dependence relation among the 𝑚 𝑖 . Since the coeffi-
cient of 𝑚 𝑝 is equal to (−1)𝑝 𝑎 𝑓 (𝑚1 , . . . , 𝑚 𝑝−1 ), this relation is non-trivial,
contradicting the assumption that (𝑚1 , . . . , 𝑚 𝑝 ) is free. Consequently,
𝑎 𝑓 0(𝑚1 , . . . , 𝑚 𝑝 ) ≠ 0. In particular, there exists a linear form 𝜑 on M
(for example, a suitable coordinate form in a basis of M) such that
𝜑(𝑎 𝑓 0(𝑚1 , . . . , 𝑚 𝑝 )) ≠ 0. The composite 𝑔 = 𝜑 ◦ 𝑓 0 is an alternate 𝑝-linear
form such that 𝑎 𝑔(𝑚1 , . . . , 𝑚 𝑝 ) ≠ 0. 

Corollary (3.8.4). — Let A be a nonzero commutative ring. Let M be a free

A-module which is generated by 𝑛 elements. Then the cardinality of any free
family in M is at most 𝑛.
Proof. — Let (𝑒1 , . . . , 𝑒 𝑝 ) be a free family in M. By proposition 3.8.3
applied to 𝑎 = 1 and the family (𝑒1 , . . . , 𝑒 𝑝 ), there exists a nonzero
alternate 𝑝-linear form on M. Proposition 3.8.2 then implies that 𝑝 6
𝑛. 

Corollary (3.8.5). — Let A be a nonzero commutative ring. Let M be a finitely

generated free A-module. Then all bases of M are finite and their cardinalities
are equal.
Proof. — Let (𝑚1 , . . . , 𝑚𝑛 ) be a generating family of M. By the preceding
proposition, every free family in M is finite and has cardinality at most 𝑛.
In particular, all bases of M are finite.
Let (𝑒1 , . . . , 𝑒 𝑚 ) and ( 𝑓1 , . . . , 𝑓𝑝 ) be bases of M. Since ( 𝑓1 , . . . , 𝑓𝑝 ) gener-
ates M and since (𝑒1 , . . . , 𝑒 𝑚 ) is free, one has 𝑚 6 𝑝. Since (𝑒1 , . . . , 𝑒 𝑚 )
generates M and since ( 𝑓1 , . . . , 𝑓𝑝 ) is free, one has 𝑝 6 𝑚. Consequently,
𝑚 = 𝑝. 

Definition (3.8.6). — Let A be a nonzero commutative ring. If M is a finitely

generated free A-module, the common cardinality of all of the bases of M is
called its rank and denoted rkA (M).
If the ring A is clear from the context, one also writes rk(M) for rkA (M).
Theorem (3.8.7). — Let A be a nonzero commutative ring and let M be a free
A-module of finite rank 𝑛 > 0. For any basis (𝑒1 , . . . , 𝑒 𝑛 ) of M, there exists
156 CHAPTER 3. MODULES

a unique alternate 𝑛-linear form 𝜑 on M such that 𝜑(𝑒1 , . . . , 𝑒 𝑛 ) = 1. This

form 𝜑 is a basis of the A-module Λ𝑛 (M, A). In particular, Λ𝑛 (M, A) is a
free A-module of rank 1.

Proof. — Let (𝜑1 , . . . , 𝜑𝑛 ) be the basis of M∨ dual to the basis (𝑒1 , . . . , 𝑒 𝑛 ).

It is almost obvious that there is at most one alternate linear form as in
the theorem. Indeed, let 𝜑 be an alternate 𝑛-linear form on M. For any
(𝑚1 , . . . , 𝑚𝑛 ) ∈ M𝑛 and any 𝑗, one has 𝑚 𝑗 = 𝑛𝑖=1 𝜑 𝑖 (𝑚 𝑗 )𝑒 𝑖 . Expanding,
Í
one gets
𝑛
Õ 𝑛
Õ
𝑓 (𝑚1 , . . . , 𝑚𝑛 ) = ... 𝜑 𝑖1 (𝑚1 ) . . . 𝜑 𝑖 𝑛 (𝑚𝑛 ) 𝑓 (𝑒 𝑖1 , . . . , 𝑒 𝑖 𝑛 ).
𝑖 1 =1 𝑖 𝑛 =1

Moreover, 𝑓 (𝑒 𝑖1 , . . . , 𝑒 𝑖 𝑛 ) = 0 if two indices 𝑖 𝑗 are equal; when they are

pairwise distinct, the map 𝑗 ↦→ 𝑖 𝑗 is a permutation 𝜎 of {1, . . . , 𝑛} and
𝑓 (𝑒 𝑖1 , . . . , 𝑒 𝑖 𝑛 ) = 𝜀(𝜎) 𝑓 (𝑒1 , . . . , 𝑒 𝑛 ) = 𝜀(𝜎).
Consequently, if such an alternate 𝑛-linear form exists at all, it is given
by the (well known) formula:
𝑛
!
Õ Ö
(3.8.7.1) 𝜑(𝑚1 , . . . , 𝑚𝑛 ) = 𝜀(𝜎) 𝜑 𝑖 (𝑚 𝜎(𝑖) ) 𝜑(𝑒1 , . . . , 𝑒 𝑛 ).
𝜎∈𝔖𝑛 𝑖=1

Let us now prove the theorem by induction on 𝑛. It suffices to show

that the previous formula defines an alternate 𝑛-linear form. It is clear
that it is 𝑛-linear; the alternating property is not much harder. Assume
indeed that 𝑚 𝑗 = 𝑚 𝑘 ; in formula (3.8.7.1), the terms corresponding to the
permutations 𝜎 and 𝜎𝜏𝑗,𝑘 , where 𝜏𝑗,𝑘 is the transposition exchanging 𝑗
and 𝑘, only differ by their sign. If 𝜎 runs among all even permutations,
𝜎 or 𝜎𝜏𝑗,𝑘 meet every permutation of {1, . . . , 𝑛} exactly once, so that the
final sum is 0.
We just proved that there is an alternate 𝑛-linear form 𝜑 on M which
maps (𝑒1 , . . . , 𝑒 𝑛 ) to 1, and that any alternate 𝑛-linear form 𝑓 on M is of
the form 𝑎𝜑, where 𝑎 = 𝜑(𝑒1 , . . . , 𝑒 𝑛 ). This concludes the proof of the
theorem. 

Definition (3.8.8). — Let A be a commutative ring.

 3.8. ALTERNATE MULTILINEAR FORMS. DETERMINANTS 157

Let M be a free A-module of rank 𝑛 and let ℬ = (𝑒1 , . . . , 𝑒 𝑛 ) be a basis of M.

The unique alternate 𝑛-linear form on M which maps (𝑒1 , . . . , 𝑒 𝑛 ) to 1 is called
the determinant (with respect to the basis ℬ); it is written detℬ .
Let U be an 𝑛 × 𝑛 matrix with entries in A. The determinant det(U) of U is
the determinant of the column vectors of U in the canonical basis of A𝑛 . For
such a matrix U = (𝑎 𝑖𝑗 ), one thus has
Õ 𝑛
Ö
det(U) = 𝜀(𝜎) 𝑎 𝑖𝜎(𝑖) .
𝜎∈𝔖𝑛 𝑖=1

Example (3.8.9). — Let M be a free A-module of rank 𝑛 and let ℬ =

(𝑒1 , . . . , 𝑒 𝑛 ) be a basis of M; write (𝜑1 , . . . , 𝜑𝑛 ) for the dual basis. Let I
be a subset of {1, . . . , 𝑛} with cardinality 𝑝; enumerate its elements in
increasing order, so that I = {𝑖 1 , . . . , 𝑖 𝑝 } with 1 6 𝑖 1 < · · · < 𝑖 𝑝 6 𝑛. Let
MI be the submodule of M generated by the 𝑒 𝑖 , for 𝑖 ∈ I, and let 𝜑I be
the map from M𝑝 to A defined by
Õ 𝑝
Ö
𝜑I (𝑚1 , . . . , 𝑚 𝑝 ) = 𝜀(𝜎) 𝜑 𝑖 𝑗 (𝑚 𝜎(𝑗) ).
𝜎∈𝔖𝑝 𝑗=1

It is 𝑝-linear alternate. This can be shown by a direct computation.

Alternatively, let detℬI be the determinant form on MI with respect to its
basis ℬI = (𝑒 𝑖 )𝑖∈I and let 𝜋I be the projector onto MI , whose kernel is the
submodule generated by the 𝑒 𝑖 for 𝑖 ∉ I. Then
𝜑I (𝑚1 , . . . , 𝑚 𝑝 ) = detℬI (𝜋I (𝑚1 ), . . . , 𝜋I (𝑚 𝑝 )).

Corollary (3.8.10). — Let M be a free A-module of rank 𝑛 and let 𝑝 be any

integer such that 0 6 𝑝 6 𝑛. The family (𝜑I ), where I runs among all subset
of {1, . . . , 𝑛} with cardinality 𝑝, is a basis of the A-module Λ𝑝 (M, A). In
particular, the A-module Λ𝑝 (M, A) is free of rank 𝑛𝑝 .

Proof. — Let I = {𝑖 1 , . . . , 𝑖 𝑝 }, with 𝑖 1 < · · · < 𝑖 𝑝 . It follows from the defi-

nition of 𝜑I that 𝜑I (𝑒 𝑖1 , . . . , 𝑒 𝑖 𝑝 ) = 1. Let J be another subset of {1, . . . , 𝑛}
with 𝑝 elements. Write {𝑗1 , . . . , 𝑗𝑝 }, with 𝑗1 < · · · < 𝑗𝑝 . Since J ≠ I
there exists 𝑘 ∈ {1, . . . , 𝑝} such that 𝑗 𝑘 ∉ I (otherwise, J ⊂ I, hence
J = I since they both have 𝑝 elements); consequently, 𝜋I (𝑒 𝑗𝑘 ) = 0, hence
158 CHAPTER 3. MODULES

𝜑I (𝑒 𝑗1 , . . . , 𝑒 𝑗𝑝 ) = 0. This implies that for any alternate 𝑝-linear form 𝑓

on M, Õ
𝑓 = 𝑓 (𝑒 𝑖1 , . . . , 𝑒 𝑖 𝑝 )𝜑I .
I={𝑖 1 <···<𝑖 𝑝 }
Indeed, the two members of the preceding equality are alternate 𝑝-
linear forms that take the same value at all 𝑝-tuples of elements of the
basis {𝑒1 , . . . , 𝑒 𝑛 }. This shows that the family (𝜑I ) generates Λ𝑝 (M, A).
Moreover, for any family (𝑎I ) indexed by the set of 𝑝-elements subsets
of {1, . . . , 𝑛}, the alternate 𝑝-linear form 𝑎I 𝜑I takes the value 𝑎I at
Í
(𝑒 𝑖1 , . . . , 𝑒 𝑖 𝑝 ) if I = {𝑖 1 < · · · < 𝑖 𝑝 }. In particular, if 𝑎I 𝜑I = 0, then 𝑎 I = 0
Í
for all I. In other words, the family (𝜑I ) is free. 

Theorem and Definition (3.8.11). — Let M be a free A-module of rank 𝑛 and

let 𝑢 be an endomorphism of M. The endomorphism Λ𝑛 (𝑢) of Λ𝑛 (M, A) is a
homothety; its ratio is called the determinant of 𝑢 and denoted by det(𝑢).
Proof. — Recall that A-module Λ𝑛 (M, A) is isomorphic to A: if ℬ =
(𝑒1 , . . . , 𝑒 𝑛 ) is a basis of M, the determinant form detℬ is a basis of
Λ𝑛 (M, A). Since Λ𝑛 (𝑢)(detℬ ) is an alternate 𝑛-linear form on M, there
exists a unique element 𝜆 ∈ A such that Λ𝑛 (𝑢)(detℬ ) = 𝜆 detℬ . Let
now 𝑓 be any alternate 𝑛-linear form on M; there exists 𝑎 ∈ A such that
𝑓 = 𝑎 detℬ . Since Λ𝑛 (𝑢) is linear, one has Λ𝑛 (𝑢)( 𝑓 ) = 𝑎Λ𝑛 (𝑢)(detℬ ) =
𝑎𝜆 detℬ . Since A is commutative, Λ𝑛 (𝑢)( 𝑓 ) = 𝜆 𝑓 . This shows that Λ𝑛 (𝑢)
is a homothety of ratio 𝜆. 

3.8.12. — Let A be a commutative ring, let M be a free A-module of

rank 𝑛 and let 𝑢 be an endomorphism of M. Let ℬ be a basis of M and
let U be the matrix of 𝑢 in the basis ℬ. One has det(𝑢) = det(U).
Indeed, by definition, the alternate 𝑛-linear form Λ𝑛 (𝑢)(detℬ ) maps
(𝑚1 , . . . , 𝑚𝑛 ) to detℬ (𝑢(𝑚1 ), . . . , 𝑢(𝑚𝑛 )). In particular, it maps (𝑒1 , . . . , 𝑒 𝑛 )
to detℬ (𝑢(𝑒1 ), . . . , 𝑢(𝑒 𝑛 )) = det(U). This shows that Λ𝑛 (𝑢)(detℬ ) =
(det U) detℬ , so that det(𝑢) = det(U).
Let 𝑢 and 𝑣 be endomorphisms of M. One has Λ𝑛 (𝑣◦𝑢) = Λ𝑛 (𝑢)◦Λ𝑛 (𝑣),
hence
det(𝑣 ◦ 𝑢) = det(𝑢) det(𝑣) = det(𝑣) det(𝑢).
One also has Λ𝑛 (idM ) = id, so that det(idM ) = 1.
 3.8. ALTERNATE MULTILINEAR FORMS. DETERMINANTS 159

Assume that 𝑢 is invertible, with inverse 𝑣. Then Λ𝑛 (𝑢) is also

invertible with inverse Λ𝑛 (𝑣); this implies that det(𝑢) is invertible with
inverse det(𝑣). In other words,
det(𝑢 −1 ) = det(𝑢)−1 .
Of course, the same formulae hold for matrices: if U and V are two 𝑛 ×𝑛
matrices, then det(VU) = det(V) det(U). One also has det(Ut ) = det(U),
either by a direct computation, or using the fact that the transpose
of a homothety is a homothety with the same ratio, so that for any
endomorphism 𝑢 of M, Λ𝑛 (𝑢 t ) = Λ𝑛 (𝑢)t is the homothety of ratio det(𝑢)
in Λ𝑛 (M∨ , A).
Let U = (𝑢𝑖,𝑗 ) be an 𝑛 × 𝑛 matrix. The cofactor 𝑣 𝑖,𝑗 of the coefficient 𝑢𝑖,𝑗
of index (𝑖, 𝑗) is the determinant of the (𝑛 − 1) × (𝑛 − 1) matrix obtained by
removing from U the 𝑖th row and the 𝑗th column, multiplied by (−1)𝑖+𝑗 .
The adjugate of U is the matrix U ˜ = (𝑣 𝑗,𝑖 )16𝑖,𝑗6𝑛 , transpose of the matrix
with entries the cofactors of U.
Proposition (3.8.13) (Adjugate formula). — The following formula holds:
UU˜ =U ˜ U = (det U)I𝑛 .

Proof. — For 𝑖 ∈ {1, . . . , 𝑛}, let N𝑖 be the set {1, . . . , b 𝑖, . . . , 𝑛} =

{1, . . . , 𝑛} {𝑖}.
Let 𝑖, 𝑗 ∈ {1, . . . , 𝑛}. Consider the (𝑛 − 1) × (𝑛 − 1) matrix U 𝑖,𝑗 obtained
by deleting the 𝑖th row and the 𝑗th column as indexed by N𝑖 for rows
∼
and N 𝑗 for columns. If 𝜎 : N𝑖 → − N 𝑗 is a bijection, let 𝑚(𝜎) be the number
of inversions of 𝜎, that is, the number of pairs (𝑘, ℓ ) of elements in N𝑖 such
that 𝑘 < 𝑙 but 𝜎(𝑘) > 𝜎(ℓ ) and let 𝜀(𝜎) = (−1)𝑚(𝜎) . Then, the expansion
of the determinant of U 𝑖,𝑗 writes
Õ Ö
𝑖,𝑗
det(U ) = 𝜀(𝜎) 𝑢 𝑘,𝜎(𝑘) .
∼ 𝑘≠𝑖
𝜎 : N𝑖 →
− N𝑗
∼
Now, a bijection 𝜎 : N𝑖 →− N 𝑗 can be extended to a bijection 𝜎¯ ∈ 𝔖𝑛 such
that 𝜎(𝑖)
¯ = 𝑗, and conversely, every such bijection 𝜎¯ is obtained exactly
once. Let us compare the number of inversions of 𝜎 with that of 𝜎. ¯ By
definition, 𝑚(𝜎)
¯ − 𝑚(𝜎) is the number of pairs (𝑘, ℓ ) ∈ {1, . . . , 𝑛} such
that 𝑘 < ℓ , 𝜎(𝑘)
¯ > 𝜎(ℓ
¯ ), and either 𝑘 or ℓ is equal to 𝑖. Let 𝑎 and 𝑏 be
the numbers of such pairs with 𝑘 = 𝑖 and ℓ = 𝑖 respectively. A pair
160 CHAPTER 3. MODULES

(𝑖, ℓ ) counts if and only if ℓ > 𝑖 and 𝜎(ℓ ¯ ) < 𝑗, so that 𝑎 values among
𝜎(𝑖
¯ + 1), . . . , 𝜎(𝑛)
¯ are smaller than 𝑗. Similarly, a pair (𝑘, 𝑖) counts if and
only if 𝑘 < 𝑖 and 𝜎(𝑘)¯ > 𝑗, so that 𝑏 values among 𝜎(1),
¯ . . . , 𝜎(𝑖
¯ − 1) are
above 𝑗 ; consequently, 𝑖 − 1 − 𝑏 such values are smaller than 𝑗. This
implies that 𝑎 + (𝑖 − 1 − 𝑏) = 𝑗 − 1, hence 𝑎 − 𝑏 = 𝑗 − 𝑖. Consequently,
𝑎 + 𝑏 ≡ 𝑗 + 𝑖 (mod 2) and

𝑚(𝜎) = 𝑚( 𝜎)
¯ − 𝑎 − 𝑏 ≡ 𝑚( 𝜎)
¯ + (𝑖 + 𝑗) (mod 2),

so that
(−1)𝑚(𝜎) = (−1)𝑚(𝜎)
¯
(−1)𝑖+𝑗 .
In particular, for any 𝑗 ∈ {1, . . . , 𝑛},
𝑛
Õ 𝑛 Õ
Õ Ö
𝑖+𝑗 𝑖,𝑗 𝑚( 𝜎)
¯
𝑢𝑖,𝑗 (−1) 𝑢𝑖,𝑗 𝑢𝑝, 𝜎(𝑝)


det(U ) = (−1) ¯
𝑖=1 𝑖=1 𝜎∈𝔖
¯ 𝑛 𝑝≠𝑖
𝜎(𝑖)=𝑗
Õ 𝑛
Ö
𝑚(𝜎)
¯
= (−1) 𝑢𝑝, 𝜎(𝑝)
¯
𝜎∈𝔖
¯ 𝑛 𝑝=1

= det(U).

On the other hand, for 𝑖 ≠ 𝑘, this formula shows that

𝑛
Õ
(−1)𝑖+𝑗 det(U 𝑖,𝑗 )𝑢 𝑘,𝑗
𝑗=1

is the determinant of the matrix U 𝑘 obtained by replacing its 𝑖th row by

the 𝑘th row of U; indeed, this manipulation does not modify any of the
matrices U 𝑖,𝑗 but replaces 𝑢𝑖,𝑗 by 𝑢 𝑘,𝑗 . Since two rows of U 𝑘 are equal,
det(U 𝑘 ) = 0. (The definition of the determinant implies that it vanishes
when two rows are equal; by transposition, it vanishes whenever two
rows are equal.) Altogher, we have shown that U U ˜ = det(U)I𝑛 . The
other formula follows by transposition: the adjugate of Ut is the transpose
˜ therefore, Ut U
of U; ˜ t = det(Ut )I𝑛 = det(U)I𝑛 , hence UU
˜ = det(U)I𝑛 . 

In the course of the proof of this formula, we established the following

Laplace expansion formula for the determinant, as a linear combination of
smaller determinants:
 3.8. ALTERNATE MULTILINEAR FORMS. DETERMINANTS 161

Corollary (3.8.14). — For every integer 𝑖 ∈ {1, . . . , 𝑛}, one has

𝑛
Õ
det(U) = (−1)𝑖+𝑗 𝑢𝑖,𝑗 det(U 𝑖,𝑗 ).
𝑗=1

Definition (3.8.15). — The characteristic polynomial of a matrix U ∈ M𝑛 (A)

is the determinant of the matrix XI𝑛 − U with coefficients in A[X].
Expansion of this determinant shows that the characteristic polynomial
of U is a monic polynomial of degree 𝑛. In fact, one has
PU (X) = X𝑛 − Tr(U)X𝑛−1 + · · · + (−1)𝑛 det(U),
Í𝑛
where Tr(U) = 𝑖=1 U𝑖𝑖 is the trace of U.
Remark (3.8.16). — Let U ∈ M𝑛 (A) and V ∈ GL𝑛 (A). Then the matrices U
and V −1 UV have the same characteristic polynomial. Indeed,
PV−1 UV (X) = det(XI𝑛 −V −1 UV) = det(V −1 (XI𝑛 −U)V) = det(XI𝑛 −U) = PU (X).
As a consequence, one may define the characteristic polynomial of
an endomorphism 𝑢 of a free finitely generated A-module M: it is the
characteristic polynomial of the matrix U of 𝑢 in any basis of M.
Similarly, the trace Tr(𝑢) of 𝑢 is defined as Tr(U). The map 𝑢 ↦→ Tr(𝑢)
is a linear form on EndA (M). It follows from the properties of traces and
determinants of matrices that Tr(𝑢𝑣) = Tr(𝑣𝑢) and det(𝑢𝑣) = det(𝑣𝑢) =
det(𝑢) det(𝑣) for any 𝑢, 𝑣 ∈ EndA (M).
Corollary (3.8.17) (Cayley-Hamilton). — Let U ∈ M𝑛 (A) be a matrix and
let PU be its characteristic polynomial. One has PU (U) = 0 in M𝑛 (A).
Proof. — Let V ∈ M𝑛 (A[X]) be the matrix defined by V = XI𝑛 − U and
˜ be the adjugate of V. Every entry V
let V ˜ 𝑖𝑗 of V
˜ is the determinant of a
matrix of size 𝑛 − 1 extracted from V = XI𝑛 − U. Consequently, V ˜ 𝑖𝑗 is a
polynomial with coefficients in A of degree 6 𝑛 − 1, so that there exist
matrices V˜ 0, . . . , V
˜ 𝑛−1 ∈ M𝑛 (A) such that

V ˜ 𝑛−1 X𝑛−1 + · · · + V
˜ =V ˜ 0.

By the adjugate formula applied to the matrix V in the ring A[X], one
˜ = det(V)I𝑛 , and det(V) = PU (X). Consequently, we have
has V V
˜ 𝑛−1 X𝑛−1 + · · · + V
(XI𝑛 − U)(V ˜ 0 ) = PU (X)I𝑛 .
162 CHAPTER 3. MODULES

Let 𝑎0 , . . . , 𝑎 𝑛−1 ∈ A be such that

PU (X) = X𝑛 + 𝑎 𝑛−1 X𝑛−1 + · · · + 𝑎0 .
Identifying the coefficients of each monomial, one obtain the relations
˜ 𝑛−1 = I𝑛
V
˜ 𝑛−2 − UV
V ˜ 𝑛−1 = 𝑎 𝑛−1 I𝑛
˜ 0 − UV
...V ˜ 1 = 𝑎 1 I𝑛
˜ 0 = 𝑎 0 I𝑛 .
−UV
If we multiply the first relation by U 𝑛 , the second by U 𝑛−1 , etc., and add
them up, we obtain
0 = U 𝑛 + 𝑎 𝑛−1 U 𝑛−1 + · · · + 𝑎0 I𝑛 ,
that is, PU (U) = 0. 

Corollary (3.8.18). — Let A be a commutative ring, let M be a finitely gen-

erated A-module and let 𝑢 be an endomorphism of M. There exists a monic
polynomial P ∈ A[X] such that P(𝑢) = 0.
Proof. — If M is the free A-module A𝑛 , this follows from Cayley-
Hamilton’s theorem (corollary 3.8.17): one may take for P the char-
acteristic polynomial of 𝑢.
In the general case, let (𝑚1 , . . . , 𝑚𝑛 ) be a finite family generating M.
Let 𝑝 : A𝑛 → M be the morphism given by 𝑝(𝑎1 , . . . , 𝑎 𝑛 ) = 𝑎 𝑖 𝑚 𝑖 . The
Í
morphism 𝑝 is surjective. In particular, for every 𝑖 ∈ {1, . . . , 𝑛}, there
exists 𝑣 𝑖 ∈ A𝑛 such that 𝑝(𝑣 𝑖 ) = 𝑢(𝑚 𝑖 ). Let 𝑣 : A𝑛 → A𝑛 be the morphism
defined by 𝑣(𝑎1 , . . . , 𝑎 𝑛 ) = 𝑎 𝑖 𝑣 𝑖 . One has 𝑝 ◦ 𝑣 = 𝑢 ◦ 𝑝. Let V be
Í
the matrix of 𝑣 and let P be its characteristic polynomial. By Cayley-
Hamilton’s theorem, one has P(V) = 0, hence P(𝑣) = 0. Consequently,
0 = 𝑝 ◦ P(𝑣) = P(𝑢) ◦ 𝑝. Since 𝑝 is surjective, this implies that P(𝑢) = 0
and concludes the proof of the corollary. 

Proposition (3.8.19). — Let A be a commutative ring and let M be a free

A-module of rank 𝑛.
a) An endomorphism of M is surjective if and only if its determinant is
invertible; it is then an isomorphism.
 3.8. ALTERNATE MULTILINEAR FORMS. DETERMINANTS 163

b) An endomorphism of M is injective if and only if its determinant is regular.

Proof. — Let 𝑢 be an endomorphism of M. Let us fix a basis ℬ =

˜ be its
(𝑒1 , . . . , 𝑒 𝑛 ) of M, let U be the matrix of 𝑢 in this basis and let U
adjugate. One has UU ˜ = det(U)I𝑛 . If det(U) is regular, the homothety
of ratio det(U) is injective in M, hence 𝑢 is injective. On the other hand,
if det(U) is invertible, the endomorphism 𝑣 of M with matrix det(U)−1 U ˜
is a left inverse of 𝑢, hence 𝑢 is surjective.
Assume that 𝑢 is surjective. For any 𝑖 ∈ {1, . . . , 𝑛}, let 𝑓𝑖 ∈ M be such
that 𝑢( 𝑓𝑖 ) = 𝑒 𝑖 ; let 𝑤 be the unique endomorphism of M that maps 𝑒 𝑖
to 𝑓𝑖 for every 𝑖. One has 𝑢(𝑤(𝑒 𝑖 )) = 𝑒 𝑖 for every 𝑖, hence 𝑢 ◦ 𝑤 = idM .
Consequently, det(𝑢) det(𝑤) = 1 and det(𝑢) is invertible.
It remains to show that det(𝑢) is regular in A if 𝑢 is injective. By
hypothesis, the family (𝑢(𝑒1 ), . . . , 𝑢(𝑒 𝑛 )) is free. Let 𝑎 be a nonzero
element of A. By proposition 3.8.3, there exists an alternate 𝑛-linear
form 𝑓 on M such that 𝑎 𝑓 (𝑢(𝑒1 ), . . . , 𝑢(𝑒 𝑛 )) ≠ 0. As any alternate 𝑛-linear
form, 𝑓 is a multiple of the determinant form detℬ in the basis (𝑒1 , . . . , 𝑒 𝑛 ),
so that 𝑎 det(𝑢(𝑒1 ), . . . , 𝑢(𝑒 𝑛 )) ≠ 0, which means exactly that 𝑎 det(𝑢) ≠ 0.
Since 𝑎 was chosen arbitrarily, this proves that det(𝑢) is regular in A. 

Corollary (3.8.20). — Let M be a free A-module and let ℬ = (𝑒1 , . . . , 𝑒 𝑛 ) be a

basis of M.
A family (𝑥1 , . . . , 𝑥 𝑛 ) of elements of M is free if and only if its determinant
in the given basis is regular in A.
Such a family generates M if and only if its determinant in the basis ℬ is a
unit of A; then, this family is even a basis of M.

Proof. — Let 𝑢 be the unique endomorphism of M that maps 𝑒 𝑖 to 𝑥 𝑖 .

By definition, one has det(𝑢) = detℬ (𝑥 1 , . . . , 𝑥 𝑛 ). For (𝑥1 , . . . , 𝑥 𝑛 ) to be
free, it is necessary and sufficient that 𝑢 be injective, hence that det(𝑢)
be regular. For (𝑥1 , . . . , 𝑥 𝑛 ) to generate M, it is necessary and sufficient
that 𝑢 be surjective, hence that det(𝑢) be invertible; then, 𝑢 is invertible
and (𝑥 1 , . . . , 𝑥 𝑛 ) is a basis of M. 
164 CHAPTER 3. MODULES

3.9. Fitting ideals

3.9.1. — Let A be a commutative ring. For every matrix U ∈ M𝑛,𝑚 (A)
and every positive integer 𝑝, let us denote by Δ𝑝 (U) the ideal of A
generated by the determinants of all 𝑝 × 𝑝 matrices extracted from U.
We also write this ideal as Δ𝑝 (𝑢1 , . . . , 𝑢𝑚 ), where 𝑢1 , . . . , 𝑢𝑚 ∈ A𝑛 are the
columns of U. It will be convenient to call any determinant of a 𝑝 × 𝑝
matrix extracted from U a minor of size 𝑝 of U. By Laplace expansion
of determinants, a minor of size 𝑝 is a linear combination of minors of
size 𝑝 −1. This implies that Δ𝑝 (U) ⊂ Δ𝑝−1 (U) for every 𝑝. If 𝑝 > inf(𝑚, 𝑛),
then Δ𝑝 (U) = 0. One also has Δ0 (U) = A (the determinant of a 0 × 0
matrix is equal to 1); for simplicity of notation, we also set Δ𝑝 (U) = A
for 𝑝 < 0.

Example (3.9.2). — Let D = diag(𝑑1 , . . . , 𝑑min(𝑛,𝑚) ) ∈ M𝑛,𝑚 (A) be a “di-

agonal matrix”, by which we mean that all of its coefficients are zero,
unless their row and column indices are equal. Assume moreover
that 𝑑 𝑖 divides 𝑑 𝑖+1 for every integer 𝑖 such that 1 6 𝑖 < min(𝑛, 𝑚).
For any integer 𝑝 ∈ {0, . . . , min(𝑛, 𝑚)}, the ideal Δ𝑝 (D) is generated by the
product 𝑑1 . . . 𝑑 𝑝 .
Let I ⊂ {1, . . . , 𝑛} and J ⊂ {1, . . . , 𝑚} be two subsets of cardinality 𝑝,
let DIJ = (𝑑 𝑖,𝑗 ) 𝑖∈I be the matrix obtained by extracting from D the rows
𝑗∈J
whose indices belong to I and the columns whose indices belong to J.
Assume that I ≠ J and let us show that det(DIJ ) = 0. Indeed, det(DIJ ) is
a sum of products 𝑑 𝑖1 ,𝑗1 . . . 𝑑 𝑖 𝑝 ,𝑗𝑝 (with signs), where I = {𝑖1 , . . . , 𝑖 𝑝 } and
J = {𝑗1 , . . . , 𝑗𝑝 }. Since I ≠ J, there must be an index 𝑘 such that 𝑖 𝑘 ≠ 𝑗 𝑘 ,
so that 𝑑 𝑖 𝑘 ,𝑗𝑘 = 0. However, if I = J, one such product is 𝑖∈I 𝑑 𝑖 , and all
Î
other vanish. By the divisibility assumption on the diagonal entries of D,
we see that all minors of size 𝑝 of M are divisible by 𝑑1 . . . 𝑑 𝑝 , and that
𝑑1 . . . 𝑑 𝑝 is one such minor (namely, for I = J = {1, . . . , 𝑝}). Consequently,
Δ𝑝 (D) = (𝑑1 . . . 𝑑 𝑝 ).

Lemma (3.9.3). — Let U ∈ M𝑛,𝑚 (A). For any P ∈ GL𝑛 (A), Q ∈ GL𝑚 (A)
and 𝑝 ∈ {1, . . . , min(𝑛, 𝑚)}, one has

Δ𝑝 (PUQ) = Δ𝑝 (U).
 3.9. FITTING IDEALS 165

Proof. — The columns of the matrix UQ are linear combinations of

columns of U. By multilinearity of the determinant, each minor of
size 𝑝 of UQ is then a linear combination of minors of size 𝑝 of U,
hence Δ𝑝 (UQ) ⊂ Δ𝑝 (U). Similarly, the rows of the matrix PU are
linear combinations of rows of U and Δ𝑝 (PU) ⊂ Δ𝑝 (U). It follows
that Δ𝑝 (PUQ) ⊂ Δ𝑝 (UQ) ⊂ Δ𝑝 (U). Since P and Q are invertible, we
can write U = P−1 (PUQ)Q−1 , hence Δ𝑝 (U) ⊂ Δ𝑝 (PUQ), so that, finally,
Δ𝑝 (U) = Δ𝑝 (PUQ). 

3.9.4. — Let M be a finitely generated A-module and let 𝜑 : A𝑛 →

M be a surjective morphism of A-modules. For every integer 𝑝 ∈
{0, . . . , 𝑛}, let then J𝑝 (𝜑) be the ideal of A generated by the ideals
Δ𝑝 (U) = Δ𝑝 (𝑢1 , . . . , 𝑢𝑝 ), where U = (𝑢1 , . . . , 𝑢𝑝 ) ranges over all subsets
of Ker(𝜑)𝑝 . One has J0 (𝜑) = A. If 𝑝 < 0, then we set J𝑝 (𝜑) = A.

Lemma (3.9.5). — Let 𝑝 be an integer such that 1 6 𝑝 6 𝑛.

a) One has J𝑝 (𝜑) ⊂ J𝑝−1 (𝜑);
b) Let (𝑣 𝑖 )𝑖∈I be a family of elements of Ker(𝜑) which generates Ker(𝜑).
Then J𝑝 (𝜑) is generated by the elements Δ𝑝 (𝑣 𝑖1 , . . . , 𝑣 𝑖 𝑝 ), where 𝑖 1 , . . . , 𝑖 𝑝 ∈ I.
c) One has AnnA (M) · J𝑝−1 (𝜑) ⊂ J𝑝 (𝜑).

Proof. — a) Expanding a minor of size 𝑝 of U along one column, one gets

an inclusion Δ𝑝 (U) ⊂ Δ𝑝−1 (U). This implies the inclusion J𝑝 (𝜑) ⊂ J𝑝−1 (𝜑).
b) One inclusion is obvious. Moreover, multilinearity of the determi-
nants implies that if U = (𝑢1 , . . . , 𝑢𝑝 ) is an 𝑛 × 𝑝-matrix with columns
in Ker(𝜑), then any minor involved in the definition of Δ𝑝 (U) is a linear
combination of elements of A of the form Δ𝑝 (𝑣 𝑖1 , . . . , 𝑣 𝑖 𝑝 ).
c) Let 𝑎 ∈ AnnA (M). Let 𝑢2 , . . . , 𝑢𝑝 ∈ Ker(𝜑); let also 𝑢 ∈ A𝑛 be any
basis element, and let 𝑢1 = 𝑎𝑢. Since 𝑎 ∈ AnnA (M), one has 𝑢1 ∈ Ker(𝜑).
The Laplace expansion along the first column of the determinant of any
𝑝 × 𝑝 submatrix of U is ±𝑎 times a minor of size 𝑝 − 1 of U, and all such
minors appear when 𝑢 varies, so that Δ𝑝 (𝑢1 , . . . , 𝑢𝑝 ) = 𝑎Δ𝑝−1 (𝑢2 , . . . , 𝑢𝑝 ).
In particular, 𝑎J𝑝−1 (𝜑) ⊂ J𝑝 (𝜑). We thus have shown the inclusion
AnnA (M) · J𝑝−1 (𝜑) ⊂ J𝑝 (𝜑). 
166 CHAPTER 3. MODULES

Theorem (3.9.6). — Let A be a commutative ring, let M be a finitely generated

A-module. Let 𝜑 : A𝑚 → M and 𝜓 : A𝑛 → M be surjective morphisms. For
every integer 𝑑 > 0, the ideals J𝑚−𝑑 (𝜑) and J𝑛−𝑑 (𝜓) are equal.

Proof. — We split the proof in 5 steps. 1) We first assume that 𝑚 = 𝑛

and that there exists an automorphism 𝑢 of A𝑛 such that 𝜓 = 𝜑 ◦ 𝑢. Then
J𝑝 (𝜑) = J𝑝 (𝜓) for every 𝑝. Indeed, observe that for X ∈ A𝑛 , one has
X ∈ Ker(𝜓) if and only if 𝑢(X) ∈ Ker(𝜑). Let X1 , . . . , X𝑝 ∈ Ker(𝜓);
write X = (X1 , . . . , X𝑝 ) and 𝑢(X) = (𝑢(X1 ), . . . , 𝑢(X𝑝 )). By multilinearity
of determinant, Δ𝑝 (𝑢(X)) ⊂ Δ𝑝 (X) ⊂ J𝑝 (𝜓). When X runs among all
elements of Ker(𝜓)𝑝 , 𝑢(X) runs among all elements of Ker(𝜑)𝑝 so that
J𝑝 (𝜑) ⊂ J𝑝 (𝜓). Equality follows by symmetry.
2) Let (𝑒1 , . . . , 𝑒 𝑚 ) be the canonical basis of A𝑚 ; for every 𝑖 ∈ {1, . . . , 𝑚},
there exists 𝑢𝑖 ∈ A𝑛 such that 𝜓(𝑢𝑖 ) = 𝜑(𝑒 𝑖 ), because 𝜓 is surjective. Then,
the unique morphism 𝑢 from A𝑚 to A𝑛 such that 𝑢(𝑒 𝑖 ) = 𝑢𝑖 for every 𝑖
satisfies 𝜑 = 𝜓 ◦ 𝑢. One constructs similarly a morphism 𝑣 : A𝑛 → A𝑚
such that 𝜓 = 𝜑 ◦ 𝑣.
Identify A𝑚+𝑛 with A𝑚 × A𝑛 ; we define three morphisms 𝜃, 𝜃0 , 𝜃00 from
A𝑚+𝑛 to M by 𝜃(𝑥, 𝑦) = 𝜑(𝑥) + 𝜓(𝑦), 𝜃0(𝑥, 𝑦) = 𝜑(𝑥) and 𝜃00(𝑥) = 𝜓(𝑦).
They are surjective. Observe that one has
𝜃(𝑥, 𝑦) = 𝜑(𝑥) + 𝜓(𝑦) = 𝜑(𝑥) + 𝜑(𝑣(𝑦)) = 𝜑(𝑥 + 𝑣(𝑦)) = 𝜃0(𝑥 + 𝑣(𝑦), 𝑦);
likewise,
𝜃(𝑥, 𝑦) = 𝜃00(𝑥, 𝑢(𝑥) + 𝑦).
Moreover, the endomorphism (𝑥, 𝑦) ↦→ (𝑥 +𝑣(𝑦), 𝑦) of A𝑚+𝑛 is an isomor-
phism (with inverse (𝑥, 𝑦) ↦→ (𝑥 − 𝑣(𝑦), 𝑦)), and so is the endomorphism
(𝑥, 𝑦) ↦→ (𝑥, 𝑢(𝑥) + 𝑦). By the first case above, it follows that for every
𝑝 6 𝑚 + 𝑛, one has
J𝑝 (𝜃) = J𝑝 (𝜃0) = J𝑝 (𝜃00).
3) By construction, 𝜓 : A𝑛 → M is a surjective morphism and 𝜃00 : A𝑚 ×
A𝑛 → M is the composition of the second projection with 𝜓. We then
prove that J𝑝 (𝜃00) = J𝑝−𝑚 (𝜓) for any 𝑝 6 𝑚 + 𝑛. This is trivial if 𝑝 6 0,
hence we assume 𝑝 > 1. By induction on 𝑚, it suffices to treat the case
𝑚 = 1. Then, Ker(𝜃00) is generated by the vector (1, 0, . . . , 0) together
with the vectors of the form (0, 𝑦) for 𝑦 ∈ Ker(𝜓).
 3.9. FITTING IDEALS 167

Let 𝑧1 , . . . , 𝑧 𝑝 be such vectors and let Z be the matrix with columns

(𝑧1 , . . . , 𝑧 𝑝 ). If the vector (1, 0, . . . ) appears twice among 𝑧1 , . . . , 𝑧 𝑝 ,
then two columns of Z are equal; since the number of columns of Z
is precisely 𝑝, this proves that any minor of size 𝑝 of Z vanishes. If
that vector appears exactly once, then we see by expanding a minor of
size 𝑝 along the corresponding column, that Δ𝑝 (Z) ⊂ J𝑝−1 (𝜓). Finally, if
all 𝑧 𝑗 are of the form (0, 𝑦), then Δ𝑝 (Z) ⊂ J𝑝 (𝜓) ⊂ J𝑝−1 (𝜓). This shows
that Δ𝑝 (Z) ⊂ J𝑝−1 (𝜓) hence J𝑝 (𝜃00) ⊂ J𝑝−1 (𝜓). Observe moreover that all
generators of J𝑝−1 (𝜓) can be obtained as minors of size 𝑝 of a suitable
matrix Z: just take 𝑧1 = (1, 0, . . . , 0) and 𝑧2 , . . . , 𝑧 𝑝 of the form (0, 𝑦) with
𝑦 ∈ Ker(𝜓). (When 𝑝 = 1, this gives Δ𝑝 (Z) = A = J0 (𝜓).) Consequently,
J𝑝 (𝜃00) = J𝑝−1 (𝜓).
4) By symmetry, one has J𝑝 (𝜃0) = J𝑝−𝑛 (𝜑) for every 𝑝 6 𝑚 + 𝑛.
5) The conclusion of the proof is now straightforward. Let 𝑑 be an
integer such that 𝑑 > 0. By step 3, we have J𝑛−𝑑 (𝜓) = J𝑚+𝑛−𝑑 (𝜃00);
by step 4, we have J𝑚−𝑑 (𝜑) = J𝑚+𝑛−𝑑 (𝜃0); by step 2, it follows that
J𝑛−𝑑 (𝜓) = J𝑚−𝑑 (𝜓) = J𝑚+𝑛−𝑑 (𝜃). 

Definition (3.9.7). — Let A be a commutative ring and let M be a finitely

generated A-module. Let 𝑚 be an integer and let 𝜑 : A𝑚 → M be a surjective
morphism of A-modules. For any integer 𝑑 > 0, the ideal J𝑚−𝑑 (𝜑) is called the
𝑑th Fitting ideal of M and denoted Fit𝑑 (M).

By theorem 3.9.6, it is independent of the choice of 𝜑. One has

Fit𝑑 (M) = A for 𝑑 large enough.

Corollary (3.9.8). — Let A be a commutative ring, let M be a finitely generated

A-module and let I = AnnA (M) be its annihilator.
a) For any integer 𝑑 > 0, one has Fit𝑑 (M) ⊂ Fit𝑑+1 (M) and I · Fit𝑑+1 (M) ⊂
Fit𝑑 (M).
b) If M is generated by 𝑑 elements, then Fit𝑑 (M) = A.

Proof. — This is a direct application of lemma 3.9.5. Let 𝜑 : A𝑚 → M

be a surjective morphism.
a) One has Fit𝑑 (M) = J𝑚−𝑑 (𝜑) ⊂ J𝑚−𝑑−1 (𝜑) = Fit𝑑+1 (𝜑) for every 𝑑 > 0.
Moreover, I · Fit𝑑+1 (𝜑) = I · J𝑚−𝑑−1 (𝜑) ⊂ J𝑚−𝑑 (𝜑) = Fit𝑑 (M).
168 CHAPTER 3. MODULES

b) Assume that 𝑚 is generated by 𝑑 elements and let 𝜓 : A𝑑 → M be a

surjective morphism of A-modules. Then Fit𝑑 (M) = J0 (𝜓) = A. 

Exercises
1-exo546) Let A be a ring and let M be a right A-module.
a) Let N be a submodule of M. Show that the set (N : M) of all 𝑎 ∈ A such that
𝑚𝑎 ∈ N for every 𝑚 ∈ M is a two-sided ideal of A.
b) Let 𝑚 ∈ M and let AnnA (𝑚) = {𝑎 ∈ A ; 𝑚𝑎 = 0} (annihilator of 𝑚). Show that it
is a right ideal of A; give an example where it is not a left ideal of A. Show that the
submodule generated by 𝑚 is isomorphic to A𝑑 /AnnA (M).
c) Let I be a right ideal of A, let N be a submodule of M and let (N :M I) be the set of
all 𝑚 ∈ M such that 𝑚𝑎 ∈ N for all 𝑎 ∈ I. Prouve that (N :M I) is a submodule of M.
2-exo027) Let A and B be two rings and let 𝑓 : A → B be a morphism of rings.
a) Let M be a right B-module. Show that one defines a right A-module by endowing
the abelian group M with the multiplication (M, A) → M given by (𝑚, 𝑎) ↦→ 𝑚 𝑓 (𝑎).
This A-module shall be denoted 𝑓 ∗ M.
b) Let 𝑢 : M → N be a morphism of right B-modules; show that 𝑢 defines a morphism
of A-modules 𝑓 ∗ M → 𝑓 ∗ N.
c) Show that the so-defined map HomB (M, N) → HomA ( 𝑓 ∗ M, 𝑓 ∗ N) is an injective
morphism of abelian groups. Give an example where it is not surjective.
d) Let M be a right B-module. Compute the annihilator of the A-module 𝑓 ∗ M in
terms of the annihilator of M.
3-exo553) Let A and B be rings and let 𝑓 : A → B be a morphism of rings. The additive
group of B is endowed with the structure of right A-module deduced from 𝑓 .
a) Assume that the image of 𝑓 be contained in the center of B (so that B is an
A-algebra). Show that the multiplication of B is A-bilinear, namely the maps 𝑥 ↦→ 𝑥𝑏
and 𝑥 ↦→ 𝑏𝑥, for fixed 𝑏 ∈ B, are A-linear.
b) Conversely, if the multiplication of B is A-bilinear, show that the image of 𝑓 is
contained in the center of B.
4-exo028) Let A be a commutative ring, let M and N be two A-modules.
a) Let 𝑢 ∈ EndA M. Show that there is a unique structure of a A[X]-module on M
for which X · 𝑚 = 𝑢(𝑚) and 𝑎 · 𝑚 = 𝑎𝑚 for every 𝑚 ∈ M and every 𝑎 ∈ A. This
A[X]-module will be denoted M𝑢 .
b) Show that the map 𝑢 ↦→ M𝑢 is a bijection from EndA (M) to the set of all structures
of A[X]-module on M such that 𝑎 · 𝑚 = 𝑎𝑚 for every 𝑚 ∈ M and every 𝑎 ∈ A.
c) Let 𝑢 ∈ EndA M and 𝑣 ∈ EndA N. Determine the set of all morphisms from M𝑢
to N𝑣 .
 EXERCISES 169

d) If M = N, under what necessary and sufficient condition on 𝑢 and 𝑣 are the two
modules M𝑢 and M𝑣 isomorphic?
e) Translate the results of the exercise when A = K is a field and M = K𝑛 is the
standard K-vector space of dimension 𝑛.
5-exo547) a) Give an example of two submodules of a module whose union is not a
submodule.
b) Let (M𝑛 )𝑛∈N be an increasing family of submodules of an A-module M, meaning
that M𝑛 ⊂ M𝑝 if 𝑛 6 𝑝. Show that its union 𝑛 M𝑛 is a subbmodule of M.
Ð

c) Let K be a division ring, let V be a K-vector space and let (W𝑖 )16𝑖6𝑛 be a family
Ð
of subspaces of V such that W = 𝑖 W𝑖 is a vector subspace. If K is infinite or, more
generally, if K has at least 𝑛 elements show that there exists 𝑖 ∈ {1, . . . , 𝑛} such that
W = W𝑖 .
d) Let V1 , V2 and V3 be the sets of all pairs (𝑥, 𝑦) ∈ Z2 such that 𝑥, resp. 𝑥 + 𝑦,
resp. 𝑦 be even. Show that these are submodules of Z2 , distinct from Z2 , and that
Z2 = V1 ∪ V2 ∪ V3 .
6-exo039) Let A be a ring, let M be a A-module, let M1 , . . . , M𝑟 be submodules of M
of which M is the direct sum, let I1 = (0 : M1 ), . . . , I𝑟 = (0 : M𝑟 ) be their annihilators.
Assume that the ideals I1 , . . . , I𝑟 are pairwise comaximal.
Set I = 𝑟𝑖=1 I𝛼 and, for 𝑖 ∈ {1, . . . , 𝑟}, set J𝑖 = 𝑗≠𝑖 I 𝑗 .
Ñ Ñ

a) Show that for any 𝑖, I𝑖 and J𝑖 are comaximal two-sided ideals of A.

For every 𝑖, let N𝑖 be the submodule of M generated by the submodules M 𝑗 , for 𝑗 ≠ 𝑖.
For any two-sided ideal J of A, let (0 : J) be the submodule of M defined by
{𝑚 ∈ M ; 𝑚𝑎 = 0 for every 𝑎 ∈ J}. Show the following formulas:
b) J𝑖 = (0 : N𝑖 ) and N𝑖 = (0 : J𝑖 ) ;
Ñ
c) N𝑖 = MI𝑖 and MJ𝑖 = M𝑖 = 𝑗≠𝑖 N𝑗 .
7-exo129) Let M be a A-module and let 𝑚 ∈ M be an element of M. Show that the
following properties are equivalent:
(i) The annihilator of 𝑚A is (0) and 𝑚A has a direct summand in M;
(ii) There exists a linear form 𝑓 on M such that 𝑓 (𝑚) = 1.
When they hold, show that M = 𝑚A ⊕ Ker 𝑓 .
8-exo197) Let 𝑓 : M → N be a morphism of A-modules.
a) Show that there exists a morphism 𝑔 : N → M such that 𝑔 ◦ 𝑓 = idM if and only if
𝑓 is injective and Im( 𝑓 ) has a direct summand in N.
b) Show that there exists a morphism 𝑔 : N → M such that 𝑓 ◦ 𝑔 = idN if and only if
𝑓 is surjective and Ker( 𝑓 ) has a direct summand in M.
9-exo565) Let A be a ring, let M be a right A-module and let (M𝑖 )𝑖∈I be a family of
submodules of M whose sum is equal to M.
a) For the M𝑖 to be in direct sum, it is necessary and sufficient that for every 𝑖 ∈ I, the
Í
intersection of the submodules M𝑖 and 𝑗≠𝑖 M 𝑗 is reduced to {0}.
170 CHAPTER 3. MODULES

b) Give an example of a A-module M, of a family (M1 , M2 , M3 ) of submodules of M

whose sum equals M but which are not in direct sum for which M𝑖 ∩ M 𝑗 = 0 for every
pair (𝑖, 𝑗) with 𝑖 ≠ 𝑗.
10-exo036) Let A be a commutative ring and let M be an A-module.
a) Consider the following definition of a weakly torsion element: an element 𝑚 ∈ M
is weakly torsion if there exists 𝑎 ∈ A {0} such that 𝑚𝑎 = 0. Give an example that
shows that the set of weakly torsion elements of a module may not be a submodule.
Recall that T(M) is the set of all 𝑚 ∈ M for which there exists a regular element 𝑎 ∈ A
such that 𝑎𝑚 = 0.
b) Let S be the set of non-zero divisors of A. Show that T(M) is the kernel of the
canonical morphism from M to S−1 M.
c) Show that the quotient module M/T(M) is torsion free.
d) Let 𝑓 : M → N be a morphism of A-modules. Show that 𝑓 (T(M)) ⊂ T(N) ∩ 𝑓 (M).
Give an example where the inclusion is not an equality.
11-exo549) Let A be a ring, let M be an A-module and let N be a subbmodule of M. A
projector in M is an endomorphism 𝑝 ∈ EndA (M) such that 𝑝 ◦ 𝑝 = 𝑝.
a) If 𝑝 is a projector, prove that M = Ker(𝑝) ⊕ Im(𝑝).
b) Show that the following conditions are equivalent:
(i) N has a direct summand in M;
(ii) N is the kernel of a projector in M;
(iii) N is the image of a projector in M.
c) Let 𝑝 0 and 𝑝 be two projectors in M, with the same image N, let 𝑢 be the map
𝑝 − 𝑝0 . Show that 𝑢 is a morphism such that Im(𝑢) ⊂ N and Ker(𝑢) ⊃ N. Construct, by
quotient, a linear map 𝑢 : M/N → N.
d) Let 𝑝0 be a given projector in M and let N be its image. Show that the map 𝑝 ↦→ 𝑢
defined in the preceding question induces a bijection from the set of of projectors in M
with image N to the abelian group HomA (M/N, N).
12-exo701) Let M be a module which is finitely generated.
a) Show that any generating family of M possesses a finite subfamily which is
generating.
b) If M is free, then any basis of M is finite.
13-exo040) Let A be an integral domain, let K be its field of fractions. One assumes
that K ≠ A. Show that K is not a free A-module.
14-exo196) Bidual. Let A be a ring, let M be a right A-module. One writes M∨ =
HomA (M, A) for its dual (a left A-module) and M∨∨ = HomA (M∨ , A) for its bidual,
that is, the dual of its dual (a right A-module).
a) Let 𝑚 ∈ M. Show that the map
𝜆𝑚 : M∨ → A, 𝜑 ↦→ 𝜑(𝑚)
is A-linear. Prove that the map 𝑚 ↦→ 𝜆𝑚 is a morphism of A-modules 𝜆 : M → M∨∨ .
 EXERCISES 171

b) One assumes that M = A𝑛𝑑 for some integer 𝑛 > 1. Show that 𝜆 is an isomorphism.
One says that M is reflexive if the morphism 𝜆 : M → M∨∨ is an isomorphism.
c) Give an example where 𝜆 is not injective and an example where 𝜆 is not surjective.
15-exo314) Let A = Z(N) and B = ZN be the infinite countable direct sum and the
infinite countable product of the abelian group Z respectively.
a) For 𝑛 ∈ N, let 𝑒 𝑛 be the element of B whose 𝑛th term is 1, all other being 0. Prove
that (𝑒 𝑛 )𝑛∈N is a basis of A, as a Z-module.
b) For every 𝑏 = (𝑏 𝑛 ) ∈ B, prove that there exists a unique linear form 𝜑(𝑏) ∈ A∨ such
that 𝜑(𝑏)(𝑒 𝑛 ) = 𝑏 𝑛 for every 𝑛 ∈ N. Prove that the map 𝑏 ↦→ 𝜑(𝑏) is an isomorphism
from B to A∨ .
c) For 𝑛 ∈ N, let 𝜃 : B∨ 𝑡𝑜B be the map given by 𝜃( 𝑓 ) = ( 𝑓 (𝑒 𝑛 ))𝑛∈Z . Prove that 𝜃 is a
morphism of Z-modules.
d) Let 𝑓 ∈ Ker(𝜃). Let 𝑥 ∈ B; prove that there exist 𝑦, 𝑧 ∈ B such that 𝑥 = 𝑦 + 𝑧, and
such that 2𝑛 | 𝑦𝑛 and 3𝑛 | 𝑧 𝑛 for every 𝑛 ∈ N. Prove that 2𝑛 | 𝑓 (𝑦) and 3𝑛 | 𝑓 (𝑧), for
every 𝑛 ∈ Z. Conclude that 𝑓 (𝑥) = 0, hence 𝑓 = 0. This shows that 𝜃 is injective.
e) Let 𝑓 ∈ B∨ and let 𝑎 = 𝜃( 𝑓 ); assume that 𝑎 ∉ A. Let (𝑠 𝑛 ) be a strictly increasing
sequence of integers and let (𝑏 𝑛 ) ∈ B be such that 2𝑠 𝑛 | 𝑏 𝑛 for all 𝑛. Prove that
𝑓 (𝑏) ≡ 𝑛−1 𝑠𝑛
𝑘=0 𝑎 𝑛 𝑏 𝑛 (mod 2 ). Construct (𝑠 𝑛 ) and (𝑏 𝑛 ) ∈ B so as to prevent any integer
Í
to satisfy all of these congruences. This proves that 𝜃(B∨ ) ⊂ A.
f ) Prove that 𝜃 : B∨ → A is an isomorphism.
g) Prove that B is not a free Z-module.
16-exo030) Let M and N be A-modules. Let 𝑓 : M → N be a morphism of A-modules.
Its transpose 𝑓 ∨ is the map from N∨ to M∨ defined by 𝑓 ∨ (𝜑) = 𝜑 ◦ 𝑓 , for every
𝜑 ∈ N∨ = HomA (N, A).
a) Show that 𝑓 ∨ is a morphism of A-modules. Show that ( 𝑓 + 𝑔)∨ = 𝑓 ∨ + 𝑔 ∨ . Show
that ( 𝑓 ◦ 𝑔)∨ = 𝑔 ∨ ◦ 𝑓 ∨ .
b) From now on, one assumes that M = N and that A is commutative. Let
𝑓 ∈ EndA (M). Show that the set I( 𝑓 ) of all polynomials P ∈ A[X] such that P( 𝑓 ) = 0 is
an ideal of A[X].
c) Show that I( 𝑓 ) ⊂ I( 𝑓 ∨ ).
d) If M is reflexive, show that I( 𝑓 ) = I( 𝑓 ∨ ).
17-exo022) Let A be a commutative ring, let I be a nonzero ideal of A, viewed as a
submodule of A. Show that I is a free module if and only if the ideal I is principal and
is generated by a regular element.
18-exo025) Let A be a ring, let M and N be right A-modules and let 𝜑 : M → N be a
surjective morphism.
a) If Ker(𝜑) are N are finitely generated, show that M is finitely generated.
In the sequel, one assumes that M and N are finitely generated and that N is a free
A-module.
172 CHAPTER 3. MODULES

b) Show that 𝜑 has a right inverse 𝜓, namely a morphism 𝜓 : N → M such that

𝜑 ◦ 𝜓 = idN .
c) Show that M ' Ker(𝜑) ⊕ Im(𝜓).
d) Recover the fact that any vector subspace of a vector space has a direct summand.
e) Show that Ker(𝜑) is finitely generated.
19-exo591) Let A be a ring, let M be a right A-module which is the (internal) direct
sum of an infinite family of nonzero submodules (M𝑖 )𝑖∈I .
For 𝑥 ∈ M, write 𝑥 = 𝑖∈I 𝑥 𝑖 , with 𝑥 𝑖 ∈ M𝑖 , and let I(𝑥) be the set of all 𝑖 ∈ I such that
Í
𝑥 𝑖 ≠ 0. This is a finite subset of I.
a) Let S be a generating subset of M. Show that I is equal to the union, for 𝑥 ∈ S, of
the subsets I(𝑥).
b) Show that S is infinite.
c*) Show that Card(S) > Card(I).
20-exo551) Let A be a commutative ring, let S be a multiplicative subset of A, let M be
an A-module.
a) Let (𝑚 𝑖 )𝑖∈I be a generating family of M; show that the family (𝑚 𝑖 /1) is a generating
family of the S−1 A-module S−1 M.
b) Let (𝑚 𝑖 )𝑖∈I be a linearly independent family in M. Assume that S does not contain
any zero-divisor. Then show that the family (𝑚 𝑖 /1) in S−1 M is still linearly independent.
c) Assume that A is an integral domain and that M is generated by 𝑛 elements, for
some integer 𝑛. Show that the cardinality of any free subset of M is at most 𝑛. (Take
for S the set of all nonzero elements of A.)
21-exo315) Let A be a commutative ring and let S be a multiplicative subset of A. Let
M be an A-module and let 𝑖 : M → S−1 M be the canonical morphism.
a) Let 𝒩 be a submodule of S−1 M and let N = 𝑖 −1 (𝒩). Prove that N satisfies the
following property: for every 𝑚 ∈ M, if there exists 𝑠 ∈ S such that 𝑠𝑚 ∈ N, then
𝑚 ∈ N. (One says that N is saturated with respect to S.)
b) One assumes in this question that S is the set of regular elements of A. Prove that
a submodule N of M is saturated with respect to S if and only if M/N is torsion-free.
c) Let N be a submodule of M which is saturated with respect to S. Prove that
N = 𝑖 −1 (S−1 N).
d) Deduce from the preceding questions that the map 𝒩 ↦→ 𝑖 −1 (𝒩) defines a bijection
from the set of submodules of S−1 M to the set of submodules of M which are saturated
with respect to S.
22-exo865) Let A be a ring and let M be a right A-module. One says that a submodule
N of M is small if for every submodule K of M such that N + K = M, one has K = M;
one says that it is maximal if N ≠ M and if for every submodule K of M such that
N ⊂ K ⊂ M, one has K = N or K = M. Let JM be the intersection of all maximal
submodules of M.
a) Prove that M = JM if and only if M is not finitely generated.
 EXERCISES 173

b) Prove that JM is the sum of all small submodules of M.

23-exo866) Let A be a ring, let J be its Jacobson radical and let M be a right A-module.
Assume that there exists an A-module N such that M ⊕ N is a free A-module (that is,
M is projective).
a) Prove that JM = MJ, where JM is the intersection of all maximal submodules of M
(exercise 3/22).
b) Prove that M ≠ MJ (Bass (1960), proposition 2.7).
24-exo041) Let A be a ring which is not a division ring. Give examples:
a) of non-free modules;
b) of a free family with 𝑛 elements of A𝑛 which is not a basis;
c) of a minimal generating set of a module which is not a basis;
d) of a submodule which has no direct summand;
e) of a free module possessing a non-free submodule.
25-exo555) a) Let M be a nonzero submodule of Q which is finitely generated.
Show that M is free and that any basis of M has cardinality 1. (Show by induction that
there exists 𝑎 ∈ Q such that M = Z𝑎.)
b) Show that the Z-module Q is not finitely generated.
c) What are the maximal free subsets of Q?
d) Does the Z-module Q possess minimal generating subsets?
26-exo554) Let K be a field, let V be a right K-vector space and let A be the ring of
endomorphisms of V. One assumes that V is not finitely generated.
Show that the right A-module (A𝑑 )2 is isomorphic to A𝑑 .
27-exo558) Let A be a ring and let 𝜑 : (A𝑑 )𝑚 → (A𝑑 )𝑛 be a morphism of right A-modules.
Let Φ be the matrix of 𝜑.
For any morphism of rings 𝑓 : A → B, one writes Φ 𝑓 for the matrix whose entries
are the images by 𝑓 of those of Φ.
a) Show that Φ 𝑓 is a the matrix of a morphism 𝜑0 : (B𝑑 )𝑚 → (B𝑑 )𝑛 of right B-modules.
b) Show that if 𝜑 is an isomorphim, then 𝜑0 is an isomorphism too. (Introduce the
matrix Φ0 of the inverse of 𝜑, then the matrix (Φ0) 𝑓 .)
c) Conclude that if A is a ring which possesses a morphism to a division ring K,
then the A-module (A𝑑 )𝑚 can only be isomorphic to the A-module (A𝑑 )𝑛 when 𝑚 = 𝑛.
In other words, in that case, the cardinalities of all bases of a finitely generated free
A-module are equal to a common integer, called the rank of this module.
28-exo562) Let K be a division ring, let V be a right K-vector space and let W be a
subspace of V.
a) Show that V is the intersection of the kernels of all linear forms on V which vanish
on W.
b) More generally, what is the intersection of the kernels of all linear forms on V
which vanish on a given subset S of V?
174 CHAPTER 3. MODULES

c) Give an example of a ring A and of a nonzero A-module M for which every linear
form is null.
d) Give an example of an integral domain A, of a free A-module M, and of a
submodule N of M such that N be not equal to the intersection of the kernels of all
linear forms on M which vanish on N.
29-exo564) Let K be a division ring, let V1 , V2 , V3 , V4 be right K-vector spaces, let
𝑢 : V1 → V2 , 𝑣 : V3 → V4 and 𝑤 : V1 → V4 be linear maps.
a) In order that there is a linear map 𝑓 : V2 → V4 such that 𝑓 ◦ 𝑢 = 𝑤, it is necessary
and sufficient that Ker(𝑢) ⊂ Ker(𝑤).
b) In order that there is a linear map 𝑔 : V1 → V3 such that 𝑣 ◦ 𝑔 = 𝑤, it is necessary
and sufficient that Im(𝑤) ⊂ Im(𝑣).
c) In order that there is a linear map ℎ : V2 → V3 such that 𝑣 ◦ ℎ ◦ 𝑢 = 𝑤, it is
necessary and sufficient that Ker(𝑢) ⊂ Ker(𝑤) and Im(𝑤) ⊂ Im(𝑣).
30-exo702) Let K be a division ring and let M be a right K-vector space. Assume that M
is finitely generated.
a) Let (𝑚1 , . . . , 𝑚𝑛 ) be a generating family in M, let (𝑣1 , . . . , 𝑣 𝑝 ) be a free family in M.
Show by induction on 𝑛 that 𝑝 6 𝑛.
b) Show that all bases of M are finite and have the same cardinality.
31-exo590) a) Let M be a free finitely generated A-module and let 𝑢 be an endomorphism
of M. Show that the A-module M/Im(𝑢) is annihilated by det(𝑢).
b) Let M and N be free finitely generated A-modules, let (𝑒1 , . . . , 𝑒 𝑚 ) be a basis of M,
let ( 𝑓1 , . . . , 𝑓𝑛 ) be a basis of N. Let 𝑢 : M → N be a linear map and let U be its matrix in
the above given bases. Assume that 𝑚 > 𝑛. Then show that N/𝑢(M) is annihilated by
every 𝑛 × 𝑛 minor from U.
CHAPTER 4

FIELD EXTENSIONS

One aspect of commutative algebra is to not only consider modules over a

given fixed ring, but also morphisms of (commutative) rings.
Let A and B be commutative rings. Recall that it is equivalent to give oneself
a morphism of rings 𝑓 : A → B or a structure of an A-algebra on B, related by
the formula 𝑓 (𝑎) = 𝑎 · 1B .
When A and B are fields, any morphism 𝑓 : A → B is injective and the
study of B as an A-algebra often assumes that A is a subfield of B, 𝑓 being the
inclusion. The classical terminology is to say that B is a field extension of A;
one reason for this terminology is that A was often assumed to be given, and B
was constructed by enlarging A with new “imaginary” elements.
I will generalize this terminology here and say that the B is an extension of
rings of A if it is given with the structure of an A-algebra. In fact, one can
essentially reduce the study of a morphism of rings 𝑓 : A → B as the study, on
the one side, of the pair consisting of B and its subring 𝑓 (A), and on the other
side, of the quotient ring 𝑓 (A) = A/I, where I = Ker( 𝑓 ). In particular, if 𝑓 is
injective, it induces an isomorphism from A to 𝑓 (A) and one is in the situation
of a subring.
Often, one understates the underlying morphism 𝑓 : A → B, and write
sentences such as “Let A → B be an extension of rings.”
The main theme of this chapter is to study the ring B in terms of whether or not
its elements satisfy polynomial equations with coefficients in A. If yes, one says
that the extension is algebraic; and one says that the extension is integral if
every element of B satisfies a monic polynomial equation with coefficients in A.
As we will learn in the first sections, this property is witnessed by adequate
finitely generated A-submodules of B.
176 CHAPTER 4. FIELD EXTENSIONS

The importance of integral extensions of commutative rings will be explored

in chapter 9 and I rapidly focus here on algebraic extensions of fields. Because
they allow to make use of linear algebras, finite extensions are essential.
Although the reader has certainly heard about the notion of algebraically
closed fields, I explain it here, and prove Steinitz’s theorem that any field has an
algebraic closure, and essentially only one.
The next section is devoted to separability, a property of algebraic extensions
which makes their study considerably simpler. Hopefully, many fields only have
separable extensions, and they are called perfect.
Building on these results, I can then study finite fields. All in all, the
approach that I followed to describe them is certainly not the most economic one,
and it would have been possible to prove these results earlier, at the only cost of
having to redo some constructions later in a greater generality.
In the next section, I define Galois extensions of fields, their Galois group,
and show the Galois correspondence, namely a bijection between intermediate
extensions of a Galois extension and subgroups of the Galois group. However, I
do not discuss the classical applications of Galois’s theory, such as construction
with ruler and compass or the impossibility of resolutions in radicals.
I then define norms and traces, establish their basic properties; I also show
the how separability is equivalent to the non-degeneracy of the trace map.
In a final section, I discuss general extension of fields and define their
transcendence degree.

4.1. Integral elements

Definition (4.1.1). — Let A → B be an extension of commutative rings. One
says that 𝑏 is integral over A if there exists a monic polynomial P ∈ A[X]
such that P(𝑏) = 0.

A relation of the form 𝑏 𝑛 + 𝑎 𝑛−1 𝑏 𝑛−1 · · ·+ 𝑎1 𝑏 + 𝑎0 = 0, where 𝑎0 , . . . , 𝑎 𝑛−1

are elements of A, is called an integral dependence relation for 𝑏.

Example
√ (4.1.2). — The complex numbers 𝑧 = exp(2𝑖𝜋/𝑛), 𝑢 = (−1 +
5)/2 are integral over Z: they satisfy the relations 𝑧 𝑛 = 1 and 𝑢 2 +𝑢 +1 =
0.
 4.1. INTEGRAL ELEMENTS 177

To establish the most basic properties of integral elements, we will

make use of the following elementary lemma on finitely generated
modules.

Lemma (4.1.3). — Let A → B be an extension of rings and let M be a B-module.

Let (𝑏 𝑖 )𝑖∈I be a family of elements of B which generates B as an A-module and
let (𝑚 𝑗 ) 𝑗∈J be a family of elements of M which generates it as a B-module.
a) The family (𝑏 𝑖 𝑚 𝑗 )(𝑖,𝑗)∈I×J generates M as an A-module.
b) If (𝑏 𝑖 )𝑖∈I is a basis of B as an A-module and (𝑚 𝑗 ) 𝑗∈J is a basis of M as a
B-module, then the family (𝑏 𝑖 𝑚 𝑗 )(𝑖,𝑗) is a basis of M as an A-module.

Proof. — a) Let 𝑚 ∈ M; since the family (𝑚 𝑗 ) generates M as a B-module,

there exists an almost-null family (𝑐 𝑗 ) in B such that 𝑚 = 𝑗∈J 𝑐 𝑗 𝑚 𝑗 . For
Í
every 𝑗 ∈ J such that 𝑐 𝑗 ≠ 0, there exists an almost-null family (𝑎 𝑖𝑗 )𝑖∈I of
elements of A such that 𝑐 𝑗 = 𝑎 𝑖𝑗 𝑏 𝑖 , because (𝑏 𝑖 )𝑖∈I generates B as an
Í
A-module. For 𝑗 ∈ J such that 𝑐 𝑗 = 0, set 𝑎 𝑖𝑗 = 0 for all 𝑖 ∈ I. Then the
family (𝑎 𝑖𝑗 )(𝑖,𝑗)∈I×J is almost-null and one has
Õ ÕÕ
𝑚= 𝑐 𝑗 𝑚𝑗 = 𝑎 𝑖𝑗 𝑏 𝑖 𝑚 𝑗 ,
𝑗∈J 𝑗∈J 𝑖∈I

which proves that the family (𝑏 𝑖 𝑚 𝑗 ) generates M as an A-module.

b) Let (𝑎 𝑖𝑗 ) be an almost-null family in A such that 𝑖,𝑗 𝑎 𝑖𝑗 𝑏 𝑖 𝑚 𝑗 = 0. For
Í
every 𝑗 ∈ J, set 𝑐 𝑗 = 𝑖∈I 𝑎 𝑖𝑗 𝑏 𝑖 . The family (𝑐 𝑗 ) 𝑗∈J in B is almost-null, and
Í
one has 𝑗∈J 𝑐 𝑗 𝑚 𝑗 = 0. Since the family (𝑚 𝑗 ) is free, one has 𝑐 𝑗 = 0 for
Í
all 𝑗 ∈ J. Fix 𝑗 ∈ J; from the relation 𝑖∈I 𝑎 𝑖𝑗 𝑏 𝑖 = 0, the freeness of the
Í
family (𝑏 𝑖 ) implies that 𝑎 𝑖𝑗 = 0 for all 𝑖. This shows that 𝑎 𝑖𝑗 = 0 for all 𝑖
and 𝑗. This proves that the family (𝑏 𝑖 𝑚 𝑗 ) is free, as claimed. 

Corollary (4.1.4). — Let A → B be an extension of rings and let M be a

B-module. If B is finitely generated (resp. free and finitely generated) as an
A-module, and M is finitely generated (resp. free and finitely generated) as a
B-module, then M is finitely generated (resp. free and finitely generated) as
an A-module.

The next theorem gives a very useful characterization of integral

elements.
178 CHAPTER 4. FIELD EXTENSIONS

Theorem (4.1.5). — Let A → B be an extension of rings, let 𝑏 be an element

of B and let A[𝑏] be the A-subalgebra of B generated by 𝑏. The following
assertions are equivalent:
(i) The element 𝑏 is integral over A;
(ii) The A-algebra A[𝑏] is a finitely generated A-module;
(iii) There exists an A[𝑏]-module whose annihilator is zero, and which is
finitely generated as an A-module.
Proof. — (i)⇒(ii). Let P ∈ A[X] be a monic polynomial such that
P(𝑏) = 0. Let us write it as P = X𝑛 + 𝑎1 X𝑛−1 + · · · + 𝑎 𝑛 , for 𝑎1 , . . . , 𝑎 𝑛 ∈ A
and let us show that the family (1, 𝑏, . . . , 𝑏 𝑛−1 ) generates A[𝑏] as an
A-module. By definition, an element of A[𝑏] is of the form Q(𝑏), for
some polynomial Q ∈ A[X]. Since P is monic, we may consider an
euclidean division of Q by P, say Q = PQ1 + R, where deg(R) < 𝑛.
Then Q(𝑏) = P(𝑏)Q1 (𝑏) + R(𝑏) = R(𝑏), hence is a linear combination of
(1, . . . , 𝑏 𝑛−1 ) with coefficients in A, as required.
(ii)⇒(iii). It suffices to set M = A[𝑏]. Indeed, the module M is finitely
generated as an A-module, by assumption. Moreover, if 𝑥 ∈ A[𝑏] is
such that 𝑥M = 0, we obtain 𝑥1B = 𝑥 = 0, so that the annihilator of M as
an A[𝑏]-module is trivial.
(iii)⇒(i). Let 𝜑 : M → M be the A-linear morphism defined by
𝜑(𝑚) = 𝑏𝑚. Since M is finitely generated, it follows from corollary 3.8.18
to Cayley-Hamilton’s theorem that there exist 𝑛 ∈ N and 𝑎0 , . . . , 𝑎 𝑛−1 ∈ A
such that
𝜑 𝑛 + 𝑎 𝑛−1 𝜑 𝑛−1 + · · · + 𝑎0 = 0
in EndA (M). In particular, for every 𝑚 ∈ M, one has
(𝑏 𝑛 + 𝑎 𝑛−1 𝑏 𝑛−1 + · · · + 𝑎0 )𝑚 = 0.
Since the annihilator of M is trivial, this implies that 𝑏 𝑛 + 𝑎 𝑛−1 𝑏 𝑛−1 + · · · +
𝑎0 = 0. Consequently, 𝑏 is integral over A. 

Definition (4.1.6). — Let A → B be an extension of rings. The set of elements

of B which are integral over A is called the integral closure of A in B.
Corollary (4.1.7). — If A → B is an extension of rings, then the integral
closure of A in B is a subalgebra of B.
 4.1. INTEGRAL ELEMENTS 179

Proof. — Let 𝑓 : A → B be the morphism of rings which gives B the

structure of an A-algebra.
Let 𝑎 ∈ A. Its image 𝑓 (𝑎) in B is integral over A, since it is cancelled by
the polynomial X − 𝑎. Consequently, every element of 𝑓 (A) is integral
over A. In particular, the zero and the unit elements of B are integral
over A.
Let then 𝑏, 𝑐 be elements of B which are integral over A. By theo-
rem 4.1.5, the subalgebra A[𝑏] of B is a finitely generated A-module
in B. Since 𝑐 is integral over A, it is integral also integral over A[𝑏]. By
theorem 4.1.5 again, the A-subalgebra A[𝑏, 𝑐] = A[𝑏][𝑐] of B is a finitely
generated A[𝑏]-module. It then follows from corollary 4.1.4 that the
subalgebra A[𝑏, 𝑐] of B is a finitely generated A-module. Corollary 4.2.2
implies that every element of A[𝑏, 𝑐] is integral over A. In particular,
𝑏 + 𝑐 and 𝑏𝑐 are integral over A. 

Definition (4.1.8). — a) Let B be a ring and let A be a subring of B. One

says that A is integrally closed in B if it coincides with its own integral closure
in B.
b) One says that an integral domain is integrally closed if it is integrally
closed in its field of fractions.

Theorem (4.1.9). — A unique factorization domain is integrally closed.

In particular, principal ideal domains and polynomial rings over a

field are integrally closed.
Proof. — Let A be a unique factorization domain and let K be its field of
fractions. Let 𝑥 ∈ K; let us assume that 𝑥 is integral over A and let us
prove that 𝑥 ∈ A. Let P = X𝑛 + 𝑎 1 X𝑛−1 + · · · + 𝑎 𝑛 be a monic polynomial
with coefficients in A such that P(𝑥) = 0. Let 𝑎, 𝑏 be coprime elements
of A such that 𝑥 = 𝑎/𝑏 and let us multiply by 𝑏 𝑛 the relation P(𝑥) = 0;
we obtain the relation
𝑎 𝑛 + 𝑎 1 𝑎 𝑛−1 𝑏 + · · · + 𝑎 𝑛−1 𝑎𝑏 𝑛−1 + 𝑎 𝑛 𝑏 𝑛 = 0.
Consequently, 𝑎 𝑛 = −𝑏(𝑎1 𝑎 𝑛−1 + · · · + 𝑎 𝑛 𝑏 𝑛−1 ) is a multiple of 𝑏. Since
𝑏 is prime to 𝑎, it is prime to 𝑎 𝑛 (proposition 2.5.17). Necessarily, 𝑏 is
invertible and 𝑥 ∈ A. 
180 CHAPTER 4. FIELD EXTENSIONS

Proposition (4.1.10). — Let A be an integral domain which is integrally closed.

For any multiplicative subset S ⊂ A which does not contain 0, the ring S−1 A is
integrally closed.
Proof. — Let K be the field of fractions of A, so that we have natural
injections A ⊂ S−1 A ⊂ K, and K is the field of fractions of S−1 A. Let
𝑥 ∈ K be any element which is integral over S−1 A. Let
𝑥 𝑛 + 𝑎 𝑛−1 𝑥 𝑛−1 + · · · + 𝑎0 = 0
be an integral dependence relation, where 𝑎0 , . . . , 𝑎 𝑛−1 are elements
of S−1 A. Let 𝑠 ∈ S be a common denominator to all the 𝑎 𝑖 , so that
𝑏 𝑖 = 𝑠 𝑎 𝑖 ∈ A for every 𝑖 ∈ {0, . . . , 𝑛 − 1}. Multiplying the preceding
relation by 𝑠 𝑛 , we obtain the relation
(𝑠𝑥)𝑛 + 𝑏 𝑛−1 (𝑠𝑥)𝑛−1 + · · · + 𝑏1 𝑠 𝑛−1 (𝑠𝑥) + 𝑏 0 𝑠 𝑛−1 = 0,
which shows that 𝑠𝑥 is integral over A. Since A is integrally closed in K,
𝑠𝑥 ∈ A. Consequently, 𝑥 = (𝑠𝑥)/𝑠 ∈ S−1 A and S−1 A is integrally closed
in K. 

4.2. Integral extensions

Definition (4.2.1). — Let A → B be an extension of commutative rings. One
says that B is integral over A if every element of B is integral over A.
We start from corollaries of theorem 4.1.5
Corollary (4.2.2). — Let A be a ring and let B be a finitely generated A-algebra.
Then B is integral over A if and only if B is finitely generated as an A-module.
Proof. — First assume that B is finitely generated as an A-module and
let us prove that B is integral over A. Let 𝑏 ∈ B and let M be the A[𝑏]-
module B. Its annihilator is 0 since for any 𝑥 ∈ B, the relation 𝑥M = 0
implies 0 = 𝑥1B = 𝑥. Finally, it is finitely generated as an A-module, by
assumption. Consequently, 𝑏 is integral over A.
Conversely, let us assume that B is integral over A. Let (𝑏 1 , . . . , 𝑏 𝑛 ) be
a finite family of elements of B which generates B as an A-algebra. Let
B0 be the image of A in B; for every 𝑖 ∈ {1, . . . , 𝑛}, let B𝑖 = A[𝑏 1 , . . . , 𝑏 𝑖 ];
observe that B𝑖 = B𝑖−1 [𝑏 𝑖 ]. By assumption, 𝑏 𝑖 is integral over A, hence
 4.2. INTEGRAL EXTENSIONS 181

it is integral over B𝑖−1 . By theorem 4.1.5, B𝑖 is a finitely generated B𝑖−1 -

module. By induction, it then follows from corollary 4.1.4 that B = B𝑛 is
a finitely generated B0 -module, hence a finitely generated A-module. 

Corollary (4.2.3). — Let A → B and B → C be extensions of rings. Assume

that B is integral over A.
If an element 𝑐 of C is integral over B, then 𝑐 is integral over A. In particular,
if C is integral over B, then C is integral over A.
Proof. — Let 𝑐 ∈ C. Let P ∈ B[X] be a monic polynomial such that
P(𝑐) = 0. Let B1 be the A-subalgebra of B generated by the coefficients
of P. It is finitely generated as an A-module, because B is integral over A
(corollary 4.2.2). Moreover, 𝑐 is integral over B1 by construction, so that
the algebra B1 [𝑐] is finitely generated as a B1 -module. By corollary 4.1.4,
B1 [𝑐] is an A-subalgebra which is finitely generated as an A-module. By
theorem 4.1.5, 𝑐 is integral over A. 

Lemma (4.2.4). — Let A → B be an extension of rings.

a) For any multiplicative subset S of A, the ring S−1 B is integral over S−1 A.
b) Let J be an ideal of B and let I be an ideal of A such that 𝑓 (I) ⊂ J. By way
of the canonical morphism 𝜑 : A/I → B/J, the ring B/J is integral over A/I.
Proof. — a) Let 𝑥 ∈ S−1 B; let us show that 𝑥 is integral over S−1 A. Let
𝑏 ∈ B and 𝑠 ∈ S be such that 𝑥 = 𝑏/𝑠. By assumption, there exists an
integral dependence relation over A for 𝑏, say
𝑏 𝑛 + 𝑎 𝑛−1 𝑏 𝑛−1 + · · · + 𝑎0 = 0.
Then
𝑎 𝑛−1 𝑛−1 𝑎0
𝑥𝑛 + 𝑥 +···+ 𝑛 = 0
𝑠 𝑠
is an integral dependence relation for 𝑥 over S−1 A.
b) Let 𝑥 ∈ B/J and let 𝑏 ∈ B be any element such that 𝑥 = clJ (𝑏). Let
𝑏 𝑛 + 𝑎 𝑛−1 𝑏 𝑛−1 + · · · + 𝑎0 = 0
be an integral dependence relation for 𝑏 over A. Then
𝑥 𝑛 + clI (𝑎 𝑛−1 )𝑥 𝑛−1 + · · · + clI (𝑎0 ) = 0
is an integral dependence relation for 𝑥 over A/I. 
182 CHAPTER 4. FIELD EXTENSIONS

Proposition (4.2.5). — Let B be a ring and let A be a subring of B such that B

is integral over A.
a) Assume that A is a field. Then every regular element of B is invertible
in B.
b) Assume that B is an integral domain. Then A is a field if and only if B is
a field.

Proof. — a) Let us assume that A is a field, let 𝑏 ∈ B be a regular element

and let P = X𝑛 + 𝑎 𝑛−1 X𝑛−1 + · · · + 𝑎0 ∈ A[X] be a monic polynomial of
minimal degree such that P(𝑏) = 0. Let Q = X𝑛−1 + 𝑎 𝑛−1 X𝑛−2 + · · · + 𝑎1 , so
that P = XQ + 𝑎0 . One has 𝑏Q(𝑏) = −𝑎0 . By the minimality assumption
on 𝑛, Q(𝑏) ≠ 0. Since 𝑏 is regular, 𝑎 0 = −𝑏Q(𝑏) ≠ 0. Since 𝑎0 ∈ A and A
is a field, this implies that 𝑏 is invertible, with inverse 𝑏 0 = −Q(𝑏)/𝑎0 .
b) Assume that A is a field. Since B is an integral domain, B ≠ 0;
moreover, every non-zero element of B is regular, hence invertible by
part a). This shows that B is a field.
Conversely, let us assume that B is a field. Since the unit elements
of A and B coincide, A ≠ 0. Let 𝑎 be a non-zero element of A. Then 𝑎 is
invertible in B so that there exists 𝑏 ∈ B such that 𝑎𝑏 = 1. Since B is integral
over A, there exists a monic polynomial P = X𝑛 + 𝑎 𝑛−1 X𝑛−1 + · · · + 𝑎0
with coefficients in A such that P(𝑏) = 0. In other words,
𝑏 𝑛 + 𝑎 𝑛−1 𝑏 𝑛−1 + · · · + 𝑎0 = 0.
Multiplying this relation by 𝑎 𝑛−1 , we obtain
𝑏 = 𝑎 𝑛−1 𝑏 𝑛 = −𝑎 𝑛−1 − 𝑎 𝑛−2 𝑎 − · · · − 𝑎0 𝑎 𝑛−1 .
In particular, 𝑏 ∈ A so that 𝑎 is invertible in A. This proves that A is a
field. 

4.3. Algebraic extensions

Definition (4.3.1). — Let A → B be an extension of rings. One says that
an element 𝑏 of B is algebraic over A if there exists a non-zero polynomial
P ∈ A[X] such that P(𝑏) = 0.
If every element of B is algebraic over A, one says that B is algebraic over A.
 4.3. ALGEBRAIC EXTENSIONS 183

A non-trivial relation of the form 𝑎 𝑛 𝑏 𝑛 + 𝑎 𝑛−1 𝑏 𝑛−1 · · · + 𝑎1 𝑏 + 𝑎0 = 0,

where 𝑎0 , . . . , 𝑎 𝑛−1 , 𝑎 𝑛 are elements of A, is called an algebraic dependence
relation for 𝑏.
Remarks (4.3.2). — a) Let 𝑏 ∈ B be algebraic over A, and let P =
𝑎 𝑛 X𝑛 + · · · + 𝑎 0 be a polynomial with coefficients in A such that P(𝑏) = 0.
Then 𝑎 𝑛 𝑏 is integral over A. Indeed, if we multiply the relation P(𝑏) = 0
by 𝑎 𝑛𝑛−1 , we obtain
(𝑎 𝑛 𝑏)𝑛 + 𝑎 𝑛−1 (𝑎 𝑛 𝑏)𝑛−1 + · · · + 𝑎0 𝑎 𝑛𝑛−1 = 0,
an integral dependence relation for 𝑎 𝑛 𝑏.
b) Conversely, let 𝑏 ∈ B and let 𝑎 ∈ A be a non-nilpotent element such
that 𝑎𝑏 is integral over A. Then 𝑏 is algebraic over A. Indeed, an integral
dependence relation
(𝑎𝑏)𝑛 + 𝑐 𝑛−1 (𝑎𝑏)𝑛−1 + · · · + 𝑐 0 = 0
for 𝑎𝑏 gives rise to an algebraic dependence relation
𝑎 𝑛 𝑏 𝑛 + 𝑎 𝑛−1 𝑐 𝑛−1 𝑏 𝑛−1 + · · · + 𝑎𝑐1 𝑏 + 𝑐0 = 0
for 𝑏. Since 𝑎 is not nilpotent, 𝑎 𝑛 ≠ 0, which imply that this relation is
non-trivial.
In particular, if A is a field, then an element 𝑏 ∈ B is algebraic over A if
and only if it is integral over A.
c) The notion of being algebraic over A is not really interesting if A is
not an integral domain or if some non-zero elements of A annihilate B.
In fact, we shall mostly use it when A is a field.
Corollary (4.3.3). — Let K be a field and let A be a K-algebra which is an
integral domain. The set of all elements of A which are algebraic over K is a
field extension of K.
We call it the algebraic closure of K in A. Be cautious to distinguish this
notion from that of an “absolute” algebraic closure of a field which will
be defined below.
Remark (4.3.4). — Let K be a field and let A be a K-algebra. Let 𝑎 ∈ L
be an element which is algebraic over K. The set of all polynomials
P ∈ K[T] such that P(𝑎) = 0 is the kernel of the morphism from K[T]
184 CHAPTER 4. FIELD EXTENSIONS

to A given by P ↦→ P(𝑎). Consequently, the monic generator P of this

ideal is the monic polynomial of minimal degree such that P(𝑎) = 0; one
says that it is the minimal polynomial of 𝑎 over K.
Passing to the quotient, one gets an injective morphism K[T]/(P) → A.
Assume moreover that A is an integral domain. Then K[T]/(P) is an
integral domain as well, so that the polynomial P is irreducible.

Definition (4.3.5). — One says that a field extension K → L is finite if L is a

finite dimensional K-vector space. We then write [L : K] = dimK (L).

Corollary (4.3.6). — Let K be a field, let A be a K-algebra which is an integral

domain and let 𝑎1 , . . . , 𝑎 𝑛 ∈ A be elements of A which are algebraic over K.
Then the sub-algebra K[𝑎1 , . . . , 𝑎 𝑛 ] of A generated by these elements is a subfield
of A, and a finite extension of K.

Proof. — Let us denote this algebra by B. It follows from the previous

corollary that B is a field. Then, corollary 4.2.2 implies that it is a finite
extension of K. 

Proposition (4.3.7). — Let K → L and L → M be two extensions of fields.

Then the composed field extension K → M is finite if and only if both extensions
K → L and L → M are finite; then one has

[M : K] = [M : L] [L : K].

Proof. — Let (𝑦 𝑖 )𝑖∈I be a basis of M as an L-vector space, let (𝑥 𝑗 ) 𝑗∈J be a

basis of L as a K-vector space. Let us show that (𝑥 𝑖 𝑦 𝑗 )(𝑖,𝑗)∈I×J is a basis
of M as a K-vector space.
Let indeed 𝑚 ∈ M; one may write 𝑚 = 𝑖∈I ℓ 𝑖 𝑦 𝑖 , for an almost-null
Í
family (ℓ 𝑖 ) of elements of L. For every 𝑖, let (𝑘 𝑖𝑗 ) 𝑗∈J be an almost null family
of elements of K such that ℓ 𝑖 = 𝑗∈J 𝑘 𝑖𝑗 𝑥 𝑗 . Then the family (𝑘 𝑖𝑗 )(𝑖,𝑗)∈I×J is
Í
almost-null and 𝑚 = (𝑖,𝑗)∈I×J 𝑘 𝑖𝑗 𝑥 𝑗 𝑦 𝑖 . This shows that the family (𝑥 𝑖 𝑦 𝑗 )
Í
generates M. On the other hand, let (𝑘 𝑖𝑗 )(𝑖,𝑗) ) be a family of elements of K
such that (𝑖,𝑗) 𝑘 𝑖𝑗 𝑥 𝑗 𝑦 𝑖 = 0. For every 𝑖, let us set ℓ 𝑖 = 𝑗∈J 𝑘 𝑖𝑗 𝑥 𝑗 ∈ L. The
Í Í
family (ℓ 𝑖 ) is almost null and 𝑖∈I ℓ 𝑖 𝑦 𝑖 = 0. Since (𝑦 𝑖 ) is free over L, one
Í
has ℓ 𝑖 = 0 for every 𝑖. Since (𝑥 𝑗 ) is free over K, one has 𝑘 𝑖𝑗 = 0 for every 𝑖
and every 𝑗. This shows that the family (𝑥 𝑗 𝑦 𝑖 )(𝑖,𝑗)∈I×J is basis of M over K.
 4.3. ALGEBRAIC EXTENSIONS 185

It is finite if and only if I and J are both finite. Then,

[M : K] = Card(I × J) = Card(I) Card(J) = [M : L][L : K]. 

Let K be a field. Recall that a polynomial P ∈ K[X] in one indetermi-

nate X is said to be split if there are elements 𝑐, 𝑎1 , . . . , 𝑎 𝑛 ∈ K such that
P = 𝑐(X − 𝑎1 ) . . . (X − 𝑎 𝑛 ).
Corollary (4.3.8). — Let K be a field. The following assertions are equivalent:
(i) The field K has no algebraic extension K → L of degree > 1;
(ii) Every irreducible polynomial of K[X] has degree 1;
(iii) Every non-constant polynomial with coefficients in K has a root in K;
(iv) Every non-constant polynomial of K[X] is split.
Proof. — (i)⇒(ii). Let P be an irreducible polynomial with coefficients
in K. Then the extension K[X]/(P) is algebraic of degree deg(P). Conse-
quently, deg(P) = 1.
(ii)⇒(iii). Let P be a non-constant polynomial with coefficients in K.
Let Q be an irreducible factor of P; then there exists 𝑐 and 𝑎 in K such
that Q = 𝑐(X − 𝑎); in particular, 𝑎 is a root of P.
(iii)⇒(iv). Let P ∈ K[X] be a non-constant polynomial; let us prove by
induction on deg(P) that P is split. Let 𝑎 be a root of P in K and let Q be
the quotient of the euclidean division of P by X − 𝑎, so that P = (X − 𝑎)Q.
Since deg(Q) < deg(P), the polynomial Q is either constant or split, by
induction. It follows that P is split.
(iv)⇒(i). Let K → L be an algebraic extension; let us prove that it
has degree 1. We may replace K by its image in L and assume that
K is a subfield of L. Let then 𝑎 be an element of L K. Its minimal
polynomial M𝑎 over K is irreducible in K. By assumption, it is split.
This implies that deg(M𝑎 ) = 1, hence 𝑎 ∈ K. Consequently, L = K and
[L : K] = 1. 

Definition (4.3.9). — A field K which satisfies the properties of corollary 4.3.8

is said to be algebraically closed
Theorem (4.3.10) (“Fundamental theorem of Algebra”)
The field C of complex numbers is algebraically closed.
186 CHAPTER 4. FIELD EXTENSIONS

Contrary to its name, this is not a theorem in Algebra, but in Analysis.

Indeed, every proof of this theorem needs to use, at some point, an
analytic or topological property of the field R of real numbers. The short
proof below is a variant of the Maximum Principle in Complex Analysis.
A more algebraic proof is suggested in exercise 4/15.
Proof. — Let P ∈ C[X] be a non-constant polynomial; let us prove
that P has a root in C. We may assume that P is monic. Let us
write P(X) = X𝑛 + 𝑎 𝑛−1 X𝑛−1 + · · · + 𝑎0 , where 𝑎0 , . . . , 𝑎 𝑛−1 ∈ C and let
M = max(1, |𝑎0 | + · · · + |𝑎 𝑛−1 |). Observe for any 𝑧 ∈ C such that |𝑧| > M,
one has

|P(𝑧)| = 𝑧 𝑛 + 𝑎 𝑛−1 𝑧 𝑛−1 + · · · + 𝑎0 > |𝑧 𝑛 | − 𝑎 𝑛−1 𝑧 𝑛−1 − · · · − |𝑎0 |

> |𝑧| 𝑛 − M |𝑧| 𝑛−1 > |𝑧| 𝑛−1 (|𝑧| − M).

Let R = M + |P(0)|. Since R > M > 1, the preceding inequality shows

that |P(𝑧)| > 1 + |P(0)| > |P(0)| if |𝑧| > R.
The disk D of center 0 and radius R is compact, and the function 𝑧 ↦→
|P(𝑧)| is continuous on D. Consequently, there exists a point 𝑎 ∈ D where
|P| achieves its infimum. Necessarily, |P(𝑎)| 6 |P(0)|, hence |𝑧| < R.
Let us now consider the Taylor expansion of P at 𝑎, P(𝑎 + X) =
𝑐0 +𝑐1 X+· · ·+𝑐 𝑛 X𝑛 . One thus has 𝑐0 = P(𝑎). Let us argue by contradiction
and assume that 𝑐 0 ≠ 0. Since P is non-constant, there exists a least
integer 𝑚 > 0 such that 𝑐 𝑚 ≠ 0. Let 𝑢 be an 𝑚th root of −𝑐0 /𝑐 𝑚 . Then,
for any small enough real number 𝑡 > 0, one has |𝑎 + 𝑢𝑡 | 6 R and

P(𝑎+𝑢𝑡) = 𝑐0 +𝑐 𝑚 𝑢 𝑚 𝑡 𝑚 +negligible terms = 𝑐 0 (1−𝑡 𝑚 )+negligible terms.

In particular, |P(𝑎 + 𝑢𝑡)| < |𝑐0 | if 𝑡 > 0 is small enough. We obtain a

contradiction, since inf𝑧∈D |P(𝑧)| = |P(𝑎)| = |𝑐0 |. Consequently, P(𝑎) = 0
and P has a root in C, as was to be shown. 

Definition (4.3.11). — Let K be a field. An algebraic closure of K is an algebraic

extension K → Ω such that Ω is an algebraically closed field.

Knowing one example of an algebraically closed field allows to construct

other ones, as well as algebraic closures of some fields.
 4.3. ALGEBRAIC EXTENSIONS 187

Proposition (4.3.12). — Let K be a field, let E be an extension of K. Assume

that E is an algebraically closed field. Let Ω be the algebraic closure of K in E.
Then, Ω is algebraically closed and K → Ω is an algebraic closure of K.
Proof. — Since Ω is algebraic over K, by construction, we just need to
prove that Ω is an algebraically closed field. Let P be a non-constant
polynomial in Ω[X] and let us prove that P has a root in Ω. By assumption,
P has a root, say 𝑎, in E. Then 𝑎 is algebraic over Ω; since Ω is algebraic
over K, corollary 4.2.3. implies that 𝑎 is algebraic over K. Consequently,
𝑎 ∈ Ω. 

Example (4.3.13). — The set Q of all complex numbers which are algebraic
over Q (called algebraic numbers) is an algebraic closure of Q.
Theorem (4.3.14) (Steinitz). — Every field has an algebraic closure.
Lemma (4.3.15). — Let K be a field and let ℱ be a family of polynomials
in K[X]. There exists an algebraic extension L of K such that every polynomial
of ℱ is split in L.
Proof. — We first show that for any polynomial P ∈ K[X], there exists
a finite extension K → L such that P is split in L. The proof goes by
induction on deg(P). It holds if deg(P) = 0. Now assume that deg(P) > 1
and let Q be an irreducible factor of P. Let L1 be the field K[X]/(Q);
this is a finite extension of K and Q has a root, say 𝑎1 , in L1 . Let P1
be the quotient of the euclidean division of P by X − 𝑎1 in L1 [X]. One
has deg(P1 ) = deg(P) − 1 < deg(P). By induction, there exists a finite
extension L of L1 such that P1 is split in L. Then P = (X − 𝑎 1 )P1 is split
in L.
If ℱ is finite, this particular case suffices to establish the lemma. Indeed,
let P be the product of the members of ℱ and let K → L be an extension
such that P is split in L. Any divisor of P, hence any element of ℱ, is
then split in L, as was to be shown.
Let us now treat the general case. We may assume that every poly-
nomial of ℱ is monic. For P ∈ ℱ, let us set 𝑛P = deg(P) and let
𝑐P,1 , . . . , 𝑐 P,𝑛P be elements of K such that
P(T) = T𝑛P + 𝑐P,1 T𝑛P −1 + · · · + 𝑐 P,𝑛 .
188 CHAPTER 4. FIELD EXTENSIONS

Let us introduce the polynomial algebra A = K[X] in the family X = (XP,𝑗 )

of indeterminates XP,𝑗 , for P ∈ ℱ and 𝑗 ∈ {1, . . . , 𝑛P }. For every
polynomial P ∈ ℱ, write P(T) − 𝑛𝑗=1 (T − XP,𝑗 ) = 𝑛𝑗=1 QP,𝑗 T𝑛P −𝑗 , where
Î P Í P

QP,1 , . . . , QP,𝑛P ∈ A. Let I be the ideal of A generated by the elements QP,𝑗 ,

where P runs among ℱ.
Let us show that I ≠ A. Otherwise, there would exists an almost-null
family (SP,𝑗 ) in A such that
𝑛P
ÕÕ
1= SP,𝑗 QP,𝑗 .
P∈ℱ 𝑗=1

Let ℱ1 be the set of all P ∈ ℱ such that SP,𝑗 ≠ 0 for some 𝑗 ∈ {1, . . . , 𝑛P }.
It is finite. By the first part of the proof, there exists an algebraic
extension K → L1 such that every polynomial P ∈ ℱ1 is split in L1 .
For P ∈ ℱ1 , let 𝑎P,1 , . . . , 𝑎 P,𝑛P be a family of elements of L1 such that
P(T) = 𝑛𝑗=1 (T − 𝑎P,𝑗 ); for P ∉ ℱ1 , set 𝑎P,𝑗 = 0 for all 𝑗. Let 𝑎 = (𝑎P,𝑗 ).
Î P

By construction, QP,𝑗 (𝑎) = 0 if P ∈ ℱ1 , while SP,𝑗 (𝑎) = 0 if P ∉ ℱ1 .

SP,𝑗 QP,𝑗 at 𝑎, one thus obtains
Í
If one evaluates the relation 1 =
Í
1 = SP,𝑗 (𝑎)QP,𝑗 (𝑎) = 0, a contradiction.
By Krull’s theorem (theorem 2.1.3), there exists a maximal ideal M
of A such that I ⊂ M. Let Ω = A/M. For every P ∈ ℱ, the polynomial P
is split in Ω because the classes 𝑎P,1 , . . . , 𝑎 P,𝑛P of XP,1 , . . . , XP,𝑛P satisfy
QS,𝑗 (𝑎P,1 , . . . , 𝑎 P,𝑛P ) = 0, hence
𝑛P
Ö
P(T) = (T − 𝑎P,𝑗 ).
𝑗=1

Moreover, as an A-algebra, Ω is generated by the elements 𝑎P,𝑗 , and

those are algebraic. Consequently, Ω is algebraic over K. 
Proof of Steinitz’s theorem. — Let Ω be an algebraic extension of K such
that every polynomial of K[X] is split in Ω. Such an extension exists
by the preceding lemma. Let us show that Ω is an algebraic closure
of K. Since it is an algebraic extension of K, it suffices to prove that Ω is
algebraically closed. Let Ω → L be an algebraic extension of Ω; let us
show that it has degree 1. Replacing Ω by its image in L, we may assume
that Ω is a subfield of L. Let 𝑥 ∈ L; since it is algebraic over Ω and Ω
is algebraic over K, the element 𝑥 is algebraic over K. Let P ∈ K[X] be
 4.3. ALGEBRAIC EXTENSIONS 189

its minimal polynomial over K. One has P(𝑥) = 0 and P is split in Ω,

by construction of this extension: this implies that there exists 𝑎 ∈ Ω
such 𝑥 = 𝑎1L . Consequently, the morphism Ω → L is surjective and the
extension Ω → L has degree 1, as was to be shown. 

Theorem (4.3.16). — Let K be a field, let 𝑖 : K → L be an algebraic extension

and let 𝑗 : K → Ω be an algebraically closed extension.
The set Φ of morphisms of fields 𝑓 : L → Ω such that 𝑓 ◦ 𝑖 = 𝑗 is non-empty.
If K → L is a finite extension, one has Card(Φ) 6 [L : K].

Proof. — a) We first prove the theorem under the assumption that the
extension K → L is finite. Let (𝑎1 , . . . , 𝑎 𝑟 ) be a finite family of elements
of L such that L = K[𝑎1 , . . . , 𝑎 𝑟 ].
For 𝑟 = 0, one has L = K and the result holds. Let us assume that 𝑟 > 1
and that the result is true for any extension generated by 𝑟 − 1 elements.
Set L1 = K[𝑎1 ]; let P be the minimal polynomial of 𝑎1 over K, so that
[L1 : K] = deg(P). Since L1 ' K[X]/(P), the map 𝑗1 ↦→ 𝑗1 (𝑎1 ), induces a
bijection from the set of K-morphisms 𝑓1 : L1 → Ω to the set of roots
of P in Ω. Observe that P has at least one root in Ω, and at most deg(P),
so that there is at least one, and at most deg(P) morphisms 𝑗1 : L1 → Ω
such that 𝑓1 ◦ 𝑗 = 𝑖. By induction, each of these morphisms can be
extended to L, in at least one, and in at most [L : L1 ] ways. This shows
that 1 6 Card(Φ) 6 [L1 : K][L : L1 ] = [L : K].
b) It remains to prove that Φ is non-empty in the case where the
extension K → L is infinite. Let ℱ be the set of pairs (L0 , 𝑓 ), where L0 is
a subfield of L containing 𝑖(K) and 𝑓 : L → Ω is a field morphism such
that 𝑓 ◦ 𝑖 = 𝑗. We order the set ℱ by the relation (L01 , 𝑓1 ) ≺ (L02 , 𝑓2 ) if
L01 ⊂ L02 and 𝑓2 | L01 = 𝑓1 . Let us prove that the set ℱ is inductive. Let then
((L0𝛼 , 𝑓𝛼 ))𝛼∈A be a totally ordered family of elements of ℱ. If it is empty,
then it is bounded above by the pair (𝑖(K), 𝑓 ) ∈ ℱ, where 𝑓 : 𝑖(K) → Ω
be the unique morphism such that 𝑓 ◦ 𝑖 = 𝑗. Otherwise, let L0 be the
union of the subfields L0𝛼 ; since the family (L0𝛼 ) is non-empty and totally
ordered, L0 is a subfield of L that contains 𝑖(K). Moreover, there exists a
unique map 𝑓 : L0 → Ω such that 𝑓 | L0𝛼 = 𝑓𝛼 for every 𝛼 ∈ A, and 𝑓 is a
morphism of fields such that 𝑓 ◦ 𝑖 = 𝑗.
190 CHAPTER 4. FIELD EXTENSIONS

By Zorn’s theorem, ℱ has a maximal element, say (L0 , 𝑓 ). We now

show that L0 = L. Otherwise, let 𝑎 be any element in L such that 𝑎 ∉ L0.
The extension L0 ⊂ L0[𝑎] is finite. By part a), there exists a morphism
𝑓 0 : L0[𝑎] → Ω such that 𝑓 0 | L0 = 𝑓 . Consequently, (L0[𝑎], 𝑓 0) ∈ ℱ and is
strictly larger than (L0 , 𝑓 ), a contradiction. 

Corollary (4.3.17). — Two algebraic closures of a field are isomorphic.

Proof. — Let 𝑖 : K → Ω and 𝑗 : K → Ω0 be two algebraic closures of a
field K. Let 𝑓 : Ω → Ω0 be any morphism of fields such that 𝑗 ◦ 𝑓 = 𝑖
(theorem 4.3.16). The morphism 𝑓 induces an isomorphism of fields
from Ω to its image L in Ω0. In particular, L is an algebraically closed
field, and it remains to prove that L = Ω0. Let thus 𝑎 ∈ Ω0. Since Ω0 is
algebraic over K, the element 𝑎 is algebraic over K, and, in particular, it
is algebraic over L; since L is algebraically closed, one has 𝑎 ∈ L. This
concludes the proof. 

4.4. Separability
4.4.1. — Let K be a field, let P ∈ K[X] be a non-constant polynomial and
let 𝑎 be a root of P in an extension L of K. Generalizing the definition
from §1.3.19, the largest integer 𝑚 such that (X − 𝑎)𝑚 divides P in L[X] is
called the multiplicity of 𝑎 as a root of P.
If L0 is a further extension of L, then 𝑎 is also a root of P in L0; let us
remark that its multiplicity remains unchanged. Of course, if 𝑚 is the
multiplicity of 𝑎 as a root of P in L, then (X − 𝑎)𝑚 still divides P in L0[X],
so that the multiplicity of 𝑎 as a root of P in L0 is at least 𝑚. Conversely, let
𝑚 be the multiplicity of 𝑎 as root of P in L0; one can write P = (X − 𝑎)𝑚 Q,
where Q ∈ L0[X]. Since the coefficients of P and (X − 𝑎)𝑚 belong to the
subfield L of L0, uniqueness of the euclidean division implies that the
coefficients of Q belong to L; in particular, (X − 𝑎)𝑚 divides P in L[X] and
the multiplicity of 𝑎 as a root of P in L is at least 𝑚.
We also recall from lemma 1.3.21 that 𝑎 is a root of P(𝑘) of multiplicity
at least 𝑚 − 𝑘, for every integer 𝑘 ∈ {0, . . . , 𝑚 − 1}, with equality if the
characteristic of K is zero, or if it is strictly larger than 𝑚 (in particular, 𝑎
is not a root of P(𝑚) ).
 4.4. SEPARABILITY 191

Definition (4.4.2). — Let K be a field. One says that a polynomial P ∈ K[X] is

separable if it has no multiple roots in any field extension of K.

Lemma (4.4.3). — Let K be a field and let P ∈ K[X] be a non-zero polynomial.

The following properties are equivalent:
(1) The polynomial P is separable;
(2) There exists a field extension L of K in which P is split such that P has no
multiple root in L;
(3) The polynomials P and P0 are coprime in K[X].

Proof. — Let 𝑛 = deg(P) and write P = 𝑎 𝑛 X𝑛 + · · · + 𝑎0 for some elements

𝑎 0 , . . . , 𝑎 𝑛 ∈ K; in particular, 𝑎 𝑛 ≠ 0. Let D = gcd(P, P0).
The implication (i)⇒(ii) is obvious, because there are extensions of K
in which P is split, for example, an algebraic closure.
To prove (ii)⇒(iii), let us assume that P and P0 are not coprime, so that
deg(D) > 1. Let L be a field extension of K in which P is split. Since D
divides P, the polynomial D is split in L as well, hence it has a root in L,
say 𝑎. One has P(𝑎) = P0(𝑎) = 0, so that 𝑎 is a multiple root of P in L,
contradicting (ii).
Let 𝑎 be a multiple root of P in some extension L of K. Then 𝑎 is a
root of both P and P0. Considering a Bézout relation D = UP + VP0, we
obtain D(𝑎) = 0. In particular, P and P0 are not coprime. This establishes
the implication (iii)⇒(i). 

Proposition (4.4.4). — Let K be a field.

a) Assume that the characteristic of K is zero. Than any irreducible polynomial
in K[X] is separable.
b) Assume that the characteristic of K is a prime number 𝑝. An irreducible
polynomial P in K[X] is not separable over K if and only if it is a polynomial
in X𝑝 , if and only if its derivative is 0.
c) Assume that the characteristic of K is a prime number 𝑝 and that the
Frobenius homomorphism of K is surjective. Then every irreducible polynomial
in K[X] is separable.

Proof. — Let P ∈ K[X] be an irreducible polynomial which is not sepa-

rable. By lemma 4.4.3, the polynomials P and P0 have a common factor.
192 CHAPTER 4. FIELD EXTENSIONS

Since P is irreducible, then P divides P0. Since deg(P0) < deg(P), one
must have P0 = 0.
Let 𝑛 = deg(P) and write P = 𝑎 𝑛 X𝑛 + · · · + 𝑎 0 , so that 𝑎 𝑛 ≠ 0. One has
P0 = 𝑛𝑎 𝑛 X𝑛−1 + · · · + 2𝑎2 X + 𝑎1 . If K has characteristic 0, then 𝑛𝑎 𝑛 ≠ 0,
so that deg(P0) = 𝑛 − 1, which contradicts the hypothesis P0 = 0. This
proves a).
Assume now that K has characteristic 𝑝 > 0. Since P0 = 0, one has
𝑚𝑎 𝑚 = 0 for every 𝑚 ∈ {0, . . . , 𝑛}. If 𝑚 is prime to 𝑝, then 𝑚 is invertible
in K, hence 𝑎 𝑚 = 0. Consequently, P = 𝑚 𝑎 𝑝𝑚 X𝑝𝑚 = Q(X𝑝 ), where
Í
Q(X) = 𝑚 𝑎 𝑝𝑚 X𝑚 . This proves b).
Í
c) Let us moreover assume that the Frobenius homomorphism 𝜑 of K
is surjective. Let us keep the letter 𝜑 for the Frobenius homomorphism
𝑝
of K[X]. For every 𝑚, let 𝑏 𝑚 ∈ K be such that 𝑏 𝑚 = 𝑎 𝑝𝑚 . Then,
𝑝
Õ Õ Õ
𝑝
P(X) = Q(X ) = 𝑏 𝑚 X𝑝𝑚 = 𝑚
𝜑(𝑏 𝑚 X ) = 𝜑( 𝑏 𝑚 X𝑚 ) = R(X)𝑝 ,
𝑚 𝑚 𝑚

where R(X) = 𝑚 𝑏 𝑚 X𝑚 . Consequently, R divides P, contradicting the

Í
hypothesis that P is irreducible. 

Definition (4.4.5). — Let 𝑝 be a prime number and let K be a field of char-

acteristic 𝑝. If the Frobenius morphism of K is surjective, one says that K is
perfect.

Example (4.4.6). — a) A finite field is perfect. Indeed, the Frobenius

homomorphism 𝜎 : F → F being injective (as is any morphism of fields),
one has Card(𝜎(F)) = Card(F). Consequently, 𝜎 is surjective.
b) Let F be a field of characteristic 𝑝. The field K = F(T) is not
perfect. Indeed, there is no rational function 𝑓 ∈ K such that 𝑓 𝑝 = T.
Suppose by contradiction that there is such an element 𝑓 . The relation
𝑓 𝑝 = T shows that 𝑓 is integral over the subring F[T] of K. Since F[T]
is a unique factorization domain, it is integrally closed in its field of
fractions (theorem 4.1.9), hence 𝑓 ∈ F[T]. Then 1 = deg(T) = 𝑝 deg( 𝑓 ),
contradiction.
The polynomial X𝑝 − T ∈ F[X, T] is irreducible in F(X)[T] since it has
degree 1; since its content is 1, it is irreducible in F[X][T] = F[X, T].
Consequently, it is also irreducible in F(T)[X] = K[X]. However, if
 4.4. SEPARABILITY 193

𝑎 is a root of X𝑝 − T in an extension L of K, one has T = 𝑎 𝑝 , hence

X𝑝 − T = X𝑝 − 𝑎 𝑝 = (X − 𝑎)𝑝 , so that the polynomial X𝑝 − T, viewed as an
element of K[X], is inseparable.
Definition (4.4.7). — Let K be a field and let L be an algebraic extension of K.
One says that an element 𝑎 ∈ L is separable over K if its minimal polynomial
is separable.
One says that L is separable over K if every element of L is separable over K.
Proposition (4.4.8). — Let K be a field. If the characteristic of K is zero, or if K
is a perfect field, then every algebraic extension of K is separable over K.
Proof. — Let L be an algebraic extension of K and let 𝑎 ∈ L. Let P be the
minimal polynomial of 𝑎 over K; it is an irreducible polynomial in K[X].
By proposition 4.4.4, it is separable. By definition, this says that 𝑎 is
separable over K, hence the algebraic extension K → L is separable. 

Lemma (4.4.9). — Let K → L and L → E be two algebraic extensions of

fields. If an element 𝑎 ∈ E is separable over K, then it is separable over L.
Consequently, if E is separable over K, then E is separable over L.
Proof. — It suffices to prove the first statement. Let 𝑎 ∈ E be separable
over K and let P ∈ K[X] be its minimal polynomial. Since P(𝑎) = 0, the
minimal polynomial of 𝑎 over L divides P. In particular, its roots are
simple in any extension of L, so that 𝑎 is separable over L. 

Theorem (4.4.10). — Let 𝑖 : K → L be a finite extension of fields and let

𝑗 : K → Ω be an algebraically closed extension of K. Let Φ be the set of
morphisms of fields 𝑓 : L → Ω such that 𝑗 = 𝑓 ◦ 𝑖. The following conditions
are equivalent:
(i) The field extension L is separable over K;
(ii) The field extension L is generated by elements which are separable over K;
(iii) Card(Φ) = [L : K].
Proof. — The implication (i)⇒(ii) is obvious.
Let 𝑎 ∈ L and let L0 = K[𝑎] be the subextension it generates. We
know that K-morphisms L0 → Ω correspond bijectively to the roots R
in Ω of the minimal polynomial P of 𝑎. Moreover, by theorem 4.3.16, a
194 CHAPTER 4. FIELD EXTENSIONS

K-morphism 𝑓 0 : L0 → Ω can be extended to a K-morphism 𝑓 : L → Ω

in at most [L : L0] ways. This implies that Card(Φ) 6 Card(R)[L : L0]. If
𝑎 is not separable, then Card(R) < deg(P) = [L0 : K], hence Card(Φ) <
[L0 : K][L : L0] = [L : K]. This establishes the implication (iii)⇒(i).
Finally, let (𝑎1 , . . . , 𝑎 𝑟 ) be a finite family of separable elements of L such
that L = K[𝑎 1 , . . . , 𝑎 𝑟 ]. Let us prove by induction on 𝑟 that Card(Φ) =
[L : K].
This is obvious if 𝑟 = 0, hence assume that 𝑟 > 1 and set L1 = K[𝑎 1 ].
If 𝑟 = 1, then L = L1 , and the K-morphisms from L to Ω are in
bijection with the roots in Ω of the minimal polynomial P of 𝑎1 . By
assumption, these roots are simple, so that Card(Φ) = deg(P) = [L1 : K].
By the induction hypothesis, each of these K-morphisms L1 → Ω can be
extended in exactly [L : L1 ] ways to a K-morphism L → Ω. This proves
that Card(Φ) = [L : L1 ][L1 : K] = [L : K], as was to be shown. 

Corollary (4.4.11). — Let 𝑓 : K → L and 𝑔 : L → E be field extensions.

Assume that L is separable over K.
If an element 𝑎 ∈ E is separable over L, then it is separable over K.
If E is separable over L, then E is separable over K.

Proof. — We first treat the case where these extensions are finite. Let
𝑖 : E → Ω be an algebraic closure of K. Since L is separable over K, there
are exactly [L : K] morphisms 𝑗 from L to Ω such that 𝑗 ◦ 𝑓 = 𝑖. Since E is
separable over L, for each of these morphisms 𝑗, there are exactly [E : L]
morphisms 𝑘 such that 𝑘 ◦ 𝑔 = 𝑗. In particular, there are [E : L][L : K]
morphisms ℎ from K to Ω such that ℎ ◦ (𝑔 ◦ 𝑓 ) = 𝑖. Consequently, E is
separable over K.
Let us now treat the general case. Let 𝑎 ∈ E which is separable over L
and let E1 = L[𝑎]. Let P be the minimal polynomial of 𝑎 over L and
let L1 ⊂ L be the extension of K generated by the coefficients of P. By
assumption, the polynomial P is separable. By construction, the minimal
polynomial of 𝑎 over L1 is equal to P, which implies that 𝑎 is separable
over L1 , hence E1 is separable over L1 . Moreover, L1 is separable over K,
by assumption. By the case of finite extensions, the extension K → E1 is
separable. In particular, 𝑎 is separable over K. 
 4.4. SEPARABILITY 195

Proposition (4.4.12) (Dirichlet). — Let G be a monoid, let Ω be a field and let

Φ be a set of morphisms of monoids from G to the multiplicative monoid of Ω.
Then the set Φ is a linearly independent subset of the Ω-vector space of maps
from G to Ω.

Proof. — Let us argue by contradiction and let us consider a minimal

subset of Φ, say {𝜑1 , . . . , 𝜑𝑛 }, which is linearly dependent over Ω. Let
𝑎 1 , . . . , 𝑎 𝑛 ∈ Ω, not all zero, such that
𝑎1 𝜑1 (𝑥) + · · · + 𝑎 𝑛 𝜑𝑛 (𝑥) = 0
for every 𝑥 ∈ G. By minimality, 𝑎 𝑖 ≠ 0 for every 𝑖 ∈ {1, . . . , 𝑛}, for
otherwise, the subset of Φ obtained by taking out 𝜑 𝑖 would be linearly
independent, contradicting the minimality assumption. Moreover, 𝑛 > 2
(otherwise, 𝜑1 would be zero).
Let 𝑦 ∈ G and let us apply the relation to 𝑥𝑦, for 𝑥 ∈ G. Since
𝜑(𝑥𝑦) = 𝜑(𝑥)𝜑(𝑦) for every 𝑥 ∈ G, this gives
𝑎1 𝜑1 (𝑦)𝜑1 (𝑥) + · · · + 𝑎 𝑛 𝜑𝑛 (𝑦)𝜑𝑛 (𝑥) = 0,
for every 𝑥 ∈ G. Subtracting from this relation 𝜑𝑛 (𝑦) times the first one,
we obtain
𝑎1 (𝜑1 (𝑦) − 𝜑𝑛 (𝑦))𝜑1 (𝑥) + · · · + 𝑎 𝑛−1 (𝜑𝑛−1 (𝑦) − 𝜑𝑛 (𝑦))𝜑𝑛−1 (𝑥) = 0.
This holds for every 𝑥 ∈ G, hence we get a linear dependence relation
𝑎1 (𝜑1 (𝑦) − 𝜑𝑛 (𝑦))𝜑1 + · · · + 𝑎 𝑛−1 (𝜑𝑛−1 (𝑦) − 𝜑𝑛 (𝑦))𝜑𝑛−1 = 0
for 𝜑1 , . . . , 𝜑𝑛−1 . By the minimality assumption, 𝑎 𝑖 (𝜑 𝑖 (𝑦) − 𝜑𝑛 (𝑦)) = 0
for every 𝑖, hence 𝜑 𝑖 (𝑦) = 𝜑𝑛 (𝑦) since 𝑎 𝑖 ≠ 0. Since 𝑦 is arbitrary, we
obtain 𝜑 𝑖 = 𝜑𝑛 , a contradiction. 

Remark (4.4.13). — Let K → L be a finite extension of fields, let Ω

be an algebraically closed extension of K. Every K-linear morphism
of fields, 𝜎 : L → Ω, induces a morphism of groups from L× to Ω× ,
hence a morphism of monoids from L× to Ω. Consequently, the set Φ
of all K-linear field morphisms 𝑓 : L → Ω is a linearly independent
subset of the Ω-vector space of maps from L to Ω. Observe that Φ is
contained in the subspace of K-linear maps from L to Ω, which is a
196 CHAPTER 4. FIELD EXTENSIONS

space of dimension [L : K]. Consequently, we obtain another proof of

the inequality Card(Φ) 6 [L : K] of theorem 4.3.16.
Theorem (4.4.14) (Primitive element theorem). — Let E be a field and let
E → F be a finite separable extension. Then there exists 𝑎 ∈ F such that
F = E[𝑎]. More precisely, if E is infinite and if F = E[𝑎 1 , . . . , 𝑎 𝑑 ], there exist
elements 𝑐2 , . . . , 𝑐 𝑑 ∈ E such that 𝑎 = 𝑎 1 + 𝑐2 𝑎2 + · · · + 𝑐 𝑑 𝑎 𝑑 satisfies F = E[𝑎].
Proof. — If E is a finite field, then F is a finite field too. By proposi-
tion 1.3.20, the multiplicative group F× is a finite cyclic group. If 𝑎 is
a generator of this group, then E[𝑎] contains F× , as well as 0, so that
E[𝑎] = F.
We now assume that E is infinite. By induction on 𝑑, it suffices to
prove the following result, which we state as an independent lemma. 

Lemma (4.4.15). — Let E → F be a finite extension of infinite fields. Assume

that there are elements 𝑎, 𝑏 ∈ F such that F = E[𝑎, 𝑏]. If 𝑏 is separable over E,
then for at most finitely elements 𝑐 ∈ E, one has F = E[𝑎 + 𝑐𝑏].
Proof. — Let P and Q be the minimal polynomials of 𝑎 and 𝑏 respectively;
let 𝑝 = deg(P) and 𝑞 = deg(Q). Let Ω be an algebraic extension of F in
which P and Q are split; let 𝑎1 , . . . , 𝑎 𝑝 be the roots of P in Ω, let 𝑏 1 , . . . , 𝑏 𝑞
be the roots of Q in Ω, ordered so that 𝑎1 = 𝑎 and 𝑏 1 = 𝑏. By hypothesis,
the elements 𝑏 1 , . . . , 𝑏 𝑞 are pairwise distinct. Consequently, for any
pair (𝑖, 𝑗) of integers, where 1 6 𝑖 6 𝑝 and 2 6 𝑗 6 𝑞, the equation
𝑎 𝑖 + 𝑐𝑏 𝑗 = 𝑎1 + 𝑐𝑏1 in the unknown 𝑐 has at most one solution, namely
𝑐 = −(𝑎 𝑖 − 𝑎1 )/(𝑏 𝑗 − 𝑏1 ). Let 𝑐 ∈ E be any element which is not equal to
any of those, set 𝑡 = 𝑎 + 𝑐𝑏 and let us show that F = E[𝑡].
The polynomial R(X) = P(𝑡 − 𝑐X) belongs to E[𝑡][X] and 𝑏 is a root of R
since R(𝑏) = P(𝑡 − 𝑐𝑏) = P(𝑎) = 0. For any 𝑗 ∈ {2, . . . , 𝑝}, observe that
𝑡 − 𝑐𝑏 𝑗 = 𝑎1 + 𝑐𝑏1 − 𝑐𝑏 𝑗 ∉ {𝑎1 , . . . , 𝑎 𝑝 } hence P(𝑡 − 𝑐𝑏 𝑗 ) ≠ 0 and R(𝑏 𝑗 ) ≠ 0.
Consequently, 𝑏 = 𝑏1 is the only common root to R and Q in Ω. Since Q
is separable, all of its roots are simple, hence gcd(R, Q) = X − 𝑏. Since R
and Q have their coefficients in E[𝑡], it follows from remark 2.4.5 that
X − 𝑏 ∈ E[𝑧]. In other words, 𝑦 ∈ E[𝑡]. Then 𝑎 = 𝑡 − 𝑐𝑏 ∈ E[𝑡] so that
E[𝑎, 𝑏] ⊂ E[𝑡] ⊂ E[𝑎, 𝑏]. It follows that F = E[𝑡]. 
We end this section by a few remarks about inseparable extensions.
 4.4. SEPARABILITY 197

Lemma (4.4.16). — Let K be a field, let Ω be an algebraic closure of K. Let P be

an irreducible monic polynomial in K[X] which has only one root, say 𝑎, in Ω.
a) If the characteristic of K is 0, then 𝑎 ∈ K and P = X − 𝑎.
b) If the characteristic of K is a prime number 𝑝, there exists a smallest
𝑛 𝑛 𝑛
integer 𝑛 > 0 such that 𝑎 𝑝 ∈ K, and P = X𝑝 − 𝑎 𝑝 .

Such an element 𝑎 is said to be radicial over K.

Proof. — Let 𝑚 be the multiplicity of 𝑎 as a root of P; by assumption,
one has P = (X − 𝑎)𝑚 .
Assume that K has characteristic zero. Being irreducible, the poly-
nomial P is then separable so that its root 𝑎 is simple. In other words,
𝑚 = 1 and P = X − 𝑎.
Now assume that K has characteristic 𝑝. If P is separable, then
P = X − 𝑎 as in the characteristic 0 case. Otherwise, we proved that P is a
polynomial in X𝑝 . Let 𝑛 be the largest integer such that P is polynomial
𝑛
in X𝑝 : it is the largest integer such that 𝑝 𝑛 divides 𝑚. We thus may write
𝑛
P(X) = Q(X𝑝 ), Q(X) = X𝑟 + 𝑏 𝑟−1 X𝑟−1 + · · · + 𝑏 0 .
By assumption, Q is not a polynomial in X𝑝 (otherwise, we could write
𝑛+1
Q(X) = R(X𝑝 ), hence P(X) = R(X𝑝 ), contradicting the maximality of 𝑛).
𝑛
Consequently, Q is separable. Moreover, 𝑏 = 𝑎 𝑝 is the only root of Q
𝑛 𝑛
in Ω. One thus has Q(X) = X − 𝑏, hence P = X𝑝 − 𝑎 𝑝 .
𝑛
It remains to prove that 𝑛 is the smallest integer such that 𝑎 𝑝 ∈ K. Let
𝑑 𝑑 𝑑
𝑑 6 𝑛 be an integer such that 𝑎 𝑝 ∈ K. Then, X𝑝 − 𝑎 𝑝 divides P; since P
is irreducible, we conclude that 𝑝 𝑑 > 𝑝 𝑛 , hence 𝑑 = 𝑛. 

Remark (4.4.17). — Let K → L be a finite field extension and Lsep ⊂ L

be the subfield of L generated by all elements of L which are separable
over K. By corollary 4.4.11, the extension K → Lsep is separable, and it
is the largest subextension of L.
If the characteristic of K is zero, then L = Lsep , hence let us assume
that the characteristic of K is a prime number, 𝑝.
By corollary 4.4.11, every element of L Lsep is inseparable over Lsep .
One says that L is a radicial extension of Lsep . The detailed analysis of such
extensions is slightly more intricate than that of separable extensions.
Let us just show here the following two facts:
198 CHAPTER 4. FIELD EXTENSIONS

a) Every element of L is radicial over Lsep .

Let 𝑎 ∈ L Lsep and let P = X𝑚 + 𝑐 𝑚−1 X𝑚−1 + · · · + 𝑐 0 be its minimal
polynomial over Lsep . Since 𝑎 is not separable over L and Lsep is
separable over L, the element 𝑎 is not separable over Lsep . In particular,
P is a polynomial in X𝑝 . Let 𝑟 be the greatest integer such that there
𝑟
exists Q ∈ Lsep [X] with P = Q(X𝑝 ); by construction, Q is the minimal
𝑟 𝑟
polynomial of 𝑎 𝑝 , and is not a polynomial in X𝑝 . In particular, 𝑎 𝑝 is
𝑟 𝑟
separable over Lsep , hence 𝑎 𝑝 ∈ Lsep . This implies that Q = X − 𝑎 𝑝 ,
𝑟 𝑟 𝑟
hence P = X𝑝 − 𝑎 𝑝 = (X − 𝑎)𝑝 . This proves that 𝑎 is radicial over Lsep .
b) There exists an integer 𝑛 > 0 such that [L : Lsep ] = 𝑝 𝑛 .
Let us consider elements 𝑎1 , . . . , 𝑎 𝑛 of L Lsep such that L =
Lsep [𝑎1 , . . . , 𝑎 𝑛 ] and let us prove by induction on 𝑛 that [L : Lsep ]
is power of 𝑝. This holds if 𝑛 6 1, by the analysis of a). Let
E = Lsep [𝑎1 , . . . , 𝑎 𝑛−1 ], so that [E : Lsep ] is a power of 𝑝 by induction.
To simplify notation, let us pose 𝑎 = 𝑎 𝑛 and 𝑟 = 𝑟𝑛 , so that L = E[𝑎],
𝑟 𝑟 𝑟
the minimal polynomial of 𝑎 over Lsep being P = X𝑝 − 𝑎 𝑝 = (X − 𝑎)𝑝 .
Since the minimal polynomial of 𝑎 over E divides P in E[X], it has only
one root, 𝑎. By lemma 4.4.16, this implies that 𝑎 is radicial over E; in
particular, its degree is a power of 𝑝. Since [L : Lsep ] = [L : E][E : Lsep ],
this implies that [L : Lsep ] is a power of 𝑝.

4.5. Finite fields

A finite field is what it says: a field whose cardinality is finite.

Proposition (4.5.1). — The characteristic of a finite field is a prime number 𝑝;

its cardinality is a power of 𝑝.

Proof. — Let F be a finite field. Since Z is infinite, the canonical mor-

phism from Z to F is not injective. Its kernel is a non-zero prime ideal
of Z, hence is generated by a prime number, say 𝑝, the characteristic of F.
This morphism induces a field extension Z/𝑝Z → F, so that F is a vector
space over the finite field Z/𝑝Z. Its dimension 𝑑 must be a finite integer,
so that Card(F) = 𝑝 𝑑 . 
 4.5. FINITE FIELDS 199

Conversely, let 𝑞 > 1 be a power of a prime number 𝑝. We are going to

show that there exists a finite field of characteristic 𝑝 and cardinality 𝑞,
and that any two such fields are isomorphic.

Theorem (4.5.2). — Let 𝑝 be a prime number, let 𝑓 be an integer such that 𝑓 > 1,
let 𝑞 = 𝑝 𝑓 . Let Ω be a field of characteristic 𝑝 in which the polynomial X𝑞 − X
is split, for example, an algebraically closed field of characteristic 𝑝. There exists
a unique subfield of Ω which has cardinality 𝑞: it is the set F of all roots in Ω of
the polynomial X𝑞 − X. Moreover, any finite field of cardinality 𝑞 is isomorphic
to F.

Proof. — Let 𝜑 : Ω → Ω be the Frobenius homomorphism, defined by

𝑥 ↦→ 𝑥 𝑝 (see example 1.1.14). Let 𝜑 𝑞 = 𝜎 𝑓 be the 𝑓 th power of 𝜑, so
𝑓
that 𝜑 𝑞 (𝑥) = 𝑥 𝑝 = 𝑥 𝑞 for every 𝑥 ∈ Ω. The set F of all 𝑥 ∈ Ω such that
𝑥 𝑞 = 𝜑 𝑞 (𝑥) = 𝑥 is then a subfield of Ω, as claimed. The derivative of the
polynomial P = X𝑞 − X is P0 = 𝑞X𝑞−1 − 1 = −1, since Ω has characteristic 𝑝
and 𝑝 divides 𝑞. Consequently (see §1.3.19), P has no multiple root and
Card(F) = 𝑞.
Let K be a finite field of cardinality 𝑞. The multiplicative group K×
of K has cardinality 𝑞 − 1, hence 𝑥 𝑞−1 = 1 for every 𝑥 ∈ K× , by Lagrange’s
theorem. This implies that 𝑥 𝑞 = 𝑥 for every 𝑥 ∈ K× , and also for 𝑥 = 0.
If K is a subfield of Ω, this shows that K ⊂ F, hence K = F since these
two fields have the same number of elements.
In the general case, we already know that K is an extension of Z/𝑝Z
in which the polynomial P = X𝑞 − X is split, and which is generated by
the roots of P. Indeed, K consists only of roots of P! It follows from
theorem 4.3.16 that there exists a morphism of fields 𝑖 : K → Ω. Then,
𝑖(K) is a finite field of characteristic 𝑞 contained in Ω, hence 𝑖(K) = F and
K ' F. 

Corollary (4.5.3). — Let F be a finite field of cardinality 𝑞, let 𝑑 be an integer > 1.

There exists an irreducible monic polynomial P ∈ F[X] such that deg(P) = 𝑑.
The F-algebra F[X]/(P) is a finite field of cardinality 𝑞 𝑑 .

Proof. — We may fix an extension Ω of Z/𝑝Z in which the polynomial

𝑑 𝑑 𝑑 𝑑−1
X𝑞 − X is split. Since X𝑞 − X = X(X𝑞 −1 − 1) = X(X𝑞−1 − 1)(X𝑞 + · · · + 1),
X𝑞 − 1 is also split in Ω. We may then assume that F ⊂ Ω.
200 CHAPTER 4. FIELD EXTENSIONS

Then Ω has a unique subfield K of cardinality 𝑞 𝑑 . Let 𝑥 ∈ K× be a

generator of the finite cyclic group K× (theorem 5.5.4). The subfield F[𝑥]
of K contains the multiplicative subgroup generated by 𝑥, which is K× ,
so that F[𝑥] = K. Let P be the minimal polynomial of 𝑥 over F. This is an
irreducible monic polynomial with coefficients in F such that P(𝑥) = 0,
and K = F[𝑥] ' F[X]/(P). In particular, deg(P) = [K : F] = 𝑑. 

A defect of the preceding proof is that it does not give an effective

way to get one’s hand on a given finite field with 𝑞 elements and
characteristic 𝑝. However, if 𝑞 = 𝑝 𝑑 , we know that it will be possible to
find, whatever how, an irreducible polynomial of degree 𝑑 over Z/𝑝Z.
Such a polynomial will be a factor of X𝑞 − X over Z/𝑝Z.

Example (4.5.4). — Let us construct a field of cardinality 8. We need to

find an irreducible polynomial of degree 3 with coefficients in Z/2Z.
There are 8 = 23 monic polynomials of degree 3; four of them vanish at 0
and are not irreducible. The remaining four are X3 + 𝑎X2 + 𝑏X + 1 where
(𝑎, 𝑏) ∈ {0, 1}2 ; their value at X = 1 is 𝑎 + 𝑏. Thus two polynomials of
degree 3 with coefficients in Z/2Z have noo root, namely X3 + X + 1 and
X3 + X2 + 1. They are irreducible; indeed, they would otherwise have a
factor of degree 1, hence a root. Consequently, F8 ' F2 [X]/(X3 + X + 1).

4.6. Galois’s theory of algebraic extensions

4.6.1. — Let K → L be an extension of fields. Let Aut(L/K) be the set
of field automorphisms of L which are K-linear. In the particularly
important case where K is a subfield of L and the morphism K → L is
the inclusion, an automorphism 𝜎 of L belongs to Aut(L/K) if and only
if its restriction to K is the identity.
Observe that Aut(L/K) is a group for composition.

Examples (4.6.2). — a) Let 𝑐 : C → C be the complex conjugation.

Then Aut(C/R) = {id, 𝑐} ' Z/2Z.
It is clear that id and 𝑐 are elements of Aut(C/R). Conversely, let 𝜎 be
an automorphism of C such that 𝜎(𝑥) = 𝑥 for every 𝑥 ∈ R. Observe that
𝜎(𝑖)2 = 𝜎(𝑖 2 ) = 𝜎(−1) = −1, hence 𝜎(𝑖) ∈ {±𝑖}. If 𝑧 = 𝑥 + 𝑖𝑦 ∈ C, with
 4.6. GALOIS’S THEORY OF ALGEBRAIC EXTENSIONS 201

𝑥, 𝑦 ∈ R, then 𝜎(𝑧) = 𝜎(𝑥) + 𝜎(𝑖)𝜎(𝑦) = 𝑥 + 𝜎(𝑖)𝑦. Consequently, 𝜎 = id

if 𝜎(𝑖) = 𝑖 and 𝜎 = 𝑐 if 𝜎(𝑖) = −𝑖.
b) Let 𝑞 be a power of a prime number and let 𝑒 be an integer > 2.
Let F be a field of cardinality 𝑞 𝑒 and let E be the subfield of F of
cardinality 𝑞. Then Aut(F/E) is a cyclic group of order 𝑒, generated by
the automorphism 𝜎𝑞 : 𝑥 ↦→ 𝑥 𝑞 .
Indeed, 𝜎𝑞 belongs to Aut(F/E). For 0 6 𝑖 6 𝑒 − 1, its power 𝜎𝑞𝑖 is the
𝑖 𝑖
morphism 𝑥 ↦→ 𝑥 𝑞 . If 0 < 𝑖 < 𝑒, the polynomial X𝑞 − X has at most 𝑞 𝑖
roots in F, so that 𝜎𝑞𝑖 ≠ id. This implies that the elements id, 𝜎𝑞 , . . . , 𝜎𝑞𝑒−1
of Aut(F/E) are distinct. By the following lemma, they fill up all of
Aut(F/E).
√3
c) Let E √be the subfield of C generated by 2. Then Aut(E/Q) = {id}.
Let 𝛼 = 2, so that E = Q[𝛼]; in particular, E is a subfield of R. Any
3

automorphism 𝜎 of E is determined by 𝜎(𝛼). Moreover, 𝜎(𝛼)3 = 𝜎(𝛼3 ) =

𝜎(2) = 2, so that 𝜎(𝛼) is a cube root of 2 contained in E. Observe that 𝛼
is the only cube root of 2 which is a real number, hence the only cube
root of 2 in E. Consequently, 𝜎(𝛼) = 𝛼 and 𝜎 = id.

Definition (4.6.3). — One says that a finite extension of fields K → L is Galois

if Card(Aut(L/K)) = [L : K].

Note that elements of Aut(L/K) coincide with K-morphisms from L to

itself, since the latter are automatically surjective when the extension
K → L is finite. Moreover, theorem 4.3.16 (or remark 4.4.13) asserts that
Card(Aut(L/K)) 6 [L : K]. It also follows from theorem 4.4.10 that a
Galois extension is separable.
When an extension K → L is Galois, its automorphism group is also
denoted by Gal(L/K) and is called its Galois group.
The extensions R ⊂ C, as well as the extensions of finite fields given in
examples 4.6.2 are Galois extensions; the third one is not.
The next lemma furnishes another, fundamental example, of Galois
extension.

Lemma (4.6.4) (Artin). — Let L be a field, let G be a finite group of automor-

phisms of L. Let K = LG be the set of all 𝑥 ∈ L such that 𝜎(𝑥) = 𝑥 for every
202 CHAPTER 4. FIELD EXTENSIONS

𝜎 ∈ G. Then K is a subfield of L, the extension K ⊂ L is finite, Galois, and

Aut(L/K) = G.

Proof. — We leave to the reader to prove that K is a subfield of L. Let

us prove by contradiction that the extension K ⊂ L is finite and that
[L : K] 6 Card(G). Otherwise, let 𝑛 be some integer > Card(G) and let
𝑎1 , . . . , 𝑎 𝑛 be elements of L which are linearly independent over K. Let
us consider the system of linear equations with coefficients in L,
𝑛
Õ
𝑥 𝑖 𝜎(𝑎 𝑖 ) = 0 (𝜎 ∈ G),
𝑖=1

in unknowns 𝑥1 , . . . , 𝑥 𝑛 . By assumption, it has more unknowns than

equations, hence it has a non-zero solution (𝑥 1 , . . . , 𝑥 𝑛 ) ∈ E𝑛 . Let us
consider one such solution for which the number of non-zero coefficients
is minimal. Up to reordering 𝑎1 , . . . , 𝑎 𝑛 and 𝑥1 , . . . , 𝑥 𝑛 , we assume that
𝑥1 , . . . , 𝑥 𝑚 are non-zero, but 𝑥 𝑚+1 = · · · = 𝑥 𝑛 = 0. By linearity, we may
assume 𝑥 𝑚 = 1, so that
𝑚−1
Õ
𝑥 𝑖 𝜎(𝑎 𝑖 ) + 𝜎(𝑎 𝑚 ) = 0 (𝜎 ∈ G).
𝑖=1

Let then 𝜏 ∈ G and apply 𝜏 to the preceding equality. One obtains

𝑚−1
Õ
𝜏(𝑥 𝑖 )(𝜏 ◦ 𝜎)(𝑎 𝑖 ) + (𝜏 ◦ 𝜎)(𝑎 𝑚 ) = 0.
𝑖=1

Subtracting the relation corresponding to the element 𝜏 ◦ 𝜎 of G, one

gets
𝑚−1
Õ
(𝜏(𝑥 𝑖 ) − 𝑥 𝑖 )(𝜏 ◦ 𝜎)(𝑎 𝑖 ) = 0.
𝑖=1

This holds for any 𝜎 ∈ G, and the elements 𝜏 ◦ 𝜎 run among all elements
of G. Consequently, for any 𝜎 ∈ G,
𝑚−1
Õ
(𝜏(𝑥 𝑖 ) − 𝑥 𝑖 )𝜎(𝑎 𝑖 ) = 0.
𝑖=1
 4.6. GALOIS’S THEORY OF ALGEBRAIC EXTENSIONS 203

The 𝑛-uple (𝑦1 , . . . , 𝑦𝑛 ) defined by 𝑦 𝑖 = 𝜏(𝑥 𝑖 ) − 𝑥 𝑖 is a solution of the

initial system, but has < 𝑚 non-zero entries. By the choice of 𝑚, one has
𝑦 𝑖 = 0 for all 𝑖. Consequently, 𝜏(𝑥 𝑖 ) = 𝑥 𝑖 for every 𝑖.
Since 𝜏 is arbitrary, we conclude that 𝑥 𝑖 ∈ K for every 𝑖. In particular,
Í𝑛
𝑖=1 𝑥 𝑖 𝑎 𝑖 = 0 which contradicts the initial hypothesis that these elements
were linearly independent over K.
Consequently, [L : K] 6 Card(G). In particular, the extension K ⊂ L
is finite, and algebraic. Moreover, every element of G is an element
of Aut(L/K), so that Card(Aut(L/K)) > Card(G). Since the opposite
inequality always holds by theorem 4.3.16, one has G = Aut(L/K). 
Let K ⊂ L be a Galois extension, let G = Gal(L/K). Let ℱL be the set
of subfields of L containing K; let 𝒢G be the set of subgroups of G. We
order both sets by inclusion.
Theorem (4.6.5) (Fundamental theorem of Galois Theory)
Let K ⊂ L be a finite Galois extension with Galois group G = Gal(L/K).
a) For every subgroup H ⊂ G, the set
LH = {𝑥 ∈ L ; ∀𝜎 ∈ H, 𝜎(𝑥) = 𝑥}
is a subfield of L containing K.
b) For every subfield E of L which contains K, the extension L/E is Galois.
c) The maps 𝜑 : H ↦→ LH from 𝒢G to ℱK , and 𝛾 : E ↦→ Gal(L/E) from ℱL
to 𝒢G , are decreasing bijections, inverse one of the other.
Proof. — Part a) follow from lemma 4.6.4.
Let us prove b). Elements of Aut(L/E) are automorphisms of E
which restrict to the identity on L. In particular, Aut(L/E) ⊂ Gal(L/K).
The inequality Card(Aut(L/E)) 6 [L : E] holds by theorem 4.3.16 and
we have to prove the opposite one. Let us consider the map 𝑟 from
Gal(L/K) to HomK (E, L) given by 𝑔 ↦→ 𝑔| E . By construction, Aut(L/E) =
𝑟 −1 (𝑟(idL )). Let Ω be an algebraic closure of L. It follows from 4.3.16
that Card(HomK (E, L)) 6 [E : K] and that the fibers of 𝑟 have cardinality
at most [L : E]. Since [E : K][L : E] = Card(Gal(L/K)), we see that
Card(HomK (E, L)) = [E : K] and all fibers of 𝑟 have cardinality exactly [L :
E]. In particular, the fiber Aut(L/E) of 𝑟(idL ) has cardinality [L : E], as
required.
204 CHAPTER 4. FIELD EXTENSIONS

c) Parts a) and b) show that the maps 𝜑 and 𝛾 are well-defined. If

0
H and H0 are two subgroups of G such that H ⊂ H0, then LH ⊂ LH ,
so that 𝜑(H0) ⊂ 𝜑(H); hence 𝜑 is decreasing. Similarly, if E and E0 are
two subfields of L containing K such that E ⊂ E0, then Gal(L/E0) is a
subgroup of Gal(L/E), that is, 𝛾(E0) ⊂ 𝛾(E) and 𝛾 is decreasing.
We proved in lemma 4.6.4 that Gal(L/LH ) = H. In other words,
𝛾 ◦ 𝜑 = id. In particular, the map 𝜑 is injective. Let E be a subfield of L
which contains K; let E0 = 𝜑(𝛾(E)) = LGal(L/E) . By definition of Gal(L/E),
one has E ⊂ E0. Moreover, we know by Part b) that L/E is Galois, and
by Part a) that L/E0 is Galois as well, and that Gal(L/E0) = Gal(L/E).
Consequently, [L : E0] = [L : E]. This implies that [E0 : E] = 1, that is,
E = E0. 
It remains to give an effective construction of Galois extensions.
Proposition (4.6.6). — Let K ⊂ L be a finite extension. Then the following
properties are equivalent:
(i) The extension K ⊂ L is Galois;
(ii) The extension K ⊂ L is separable and every irreducible polynomial
in K[X] which has a root in L is already split in L;
(iii) There exists an element 𝑎 ∈ L such that L = K[𝑎] and the minimal
polynomial of which is separable and split in L;
(iv) There exists a separable polynomial P ∈ K[X] which is split in L, with
roots 𝑎1 , . . . , 𝑎 𝑛 ∈ L, such that L = K[𝑎1 , . . . , 𝑎 𝑛 ].
Proof. — (i)⇒(ii). We already proved that a Galois extension is separable.
Let P ∈ K[X] be an irreducible polynomial, let 𝑎 ∈ L be a root of P in L.
Let E = K[𝑎]. Let Q = 𝑔∈G (X − 𝑔(𝑎)). For every ℎ ∈ G, the polynomial
Î

Q ℎ obtained by applying ℎ to the coefficients of Q is given by

Ö Ö
Qℎ = (X − ℎ(𝑔(𝑎)) = (X − 𝑔(𝑎)) = Q,
𝑔∈G 𝑔∈G

since the map 𝑔 ↦→ ℎ 𝑔 is a bijection of G to itself. Consequently, the

coefficients of Q belong to LG = K. Moreover, Q(𝑎) = 0, hence P divides Q.
Since Q is split in L, it follows that P is split in L as well.
(ii)⇒(iii). Since the extension K ⊂ L is separable, the Primitive element
theorem (theorem 4.4.14) implies that there exists 𝑎 ∈ L such that
 4.6. GALOIS’S THEORY OF ALGEBRAIC EXTENSIONS 205

L = K[𝑎]. The minimal polynomial of 𝑎 is separable and irreducible,

hence is split in L.
(iii)⇒(iv). It suffices to take for P the minimal polynomial of 𝑎.
(iv)⇒(i). Let E0 = K; for 𝑖 ∈ {1, . . . , 𝑛}, let E𝑖 = K[𝑎 1 , . . . , 𝑎 𝑖 ] and let
M𝑖 be the minimal polynomial of 𝑎 𝑖 over E𝑖−1 . It divides P hence is
separable and split in L. For every 𝑖, let Φ𝑖 be the set of morphisms
from E𝑖 to L which restricts to the identity on K; let 𝑟 𝑖 : Φ𝑖 → Φ𝑖−1 be the
restriction map, 𝜎 ↦→ 𝜎| E𝑖−1 .
Given a morphism 𝑓 ∈ Φ𝑖−1 , there are exactly deg(M𝑖 ) = [E𝑖 : E𝑖−1 ]
morphisms from E𝑖 to L which extend 𝑓 . Consequently, the fibers of the
restriction map 𝑟 𝑖 all have cardinality [E𝑖 : E𝑖−1 ]. Since Φ0 has exactly
one element, one has
Card(Φ𝑛 ) = [E𝑛 : E𝑛−1 ] Card(Φ𝑛−1 ) = . . . = [E𝑛 : E𝑛−1 ] . . . [E1 : E0 ] Card(Φ0 )
= [L : K].
Moreover, any morphism 𝑓 ∈ Φ𝑛 is surjective, since [ 𝑓 (L) : K] =
dimK ( 𝑓 (L)) = dimK (L) = [L : K]. Consequently, Aut(L/K) = Φ𝑛 and the
extension L/K is Galois. 
For future use, part of statement (ii) needs to be given a name:
Definition (4.6.7). — One says that a finite extension K ⊂ L is normal if
every irreducible polynomial in K[X] which has a root in L is already split in L.
Lemma (4.6.8). — Let K ⊂ L be a finite extension. The following properties
are equivalent:
(i) The extension K ⊂ L is normal;
(ii) There exists a polynomial P ∈ K[X] which is split in L, with roots
𝑎 1 , . . . , 𝑎 𝑛 , such that L = K[𝑎1 , . . . , 𝑎 𝑛 ];
(iii) For every extension Ω of L and any K-homomorphism 𝜎 : L → Ω, one
has 𝜎(L) = L.
Proof. — (i)⇒(ii). Assume that the extension K ⊂ L is normal and let
𝑎 1 , . . . , 𝑎 𝑛 be elements of L such that L = K[𝑎1 , . . . , 𝑎 𝑛 ]. For every 𝑖, let P𝑖
be the minimal polynomial of 𝑎 𝑖 over K. It is irreducible, hence split in L.
It follows that the product P = P1 . . . P𝑛 is split in L and L is generated
by its roots.
206 CHAPTER 4. FIELD EXTENSIONS

(ii)⇒(iii). Let P ∈ K[X] be a polynomial which is split in L, with

roots 𝑎1 , . . . , 𝑎 𝑛 , such that L = K[𝑎1 , . . . , 𝑎 𝑛 ]. Let 𝜎 : L → Ω be a K-
homomorphism into an extension Ω of L. Since 𝜎(P(𝑎 𝑖 )) = P(𝜎(𝑎 𝑖 )) = 0,
𝜎(𝑎 𝑖 ) is a root of P, hence 𝜎(𝑎 𝑖 ) ∈ {𝑎1 , . . . , 𝑎 𝑛 }. In particular, 𝜎(𝑎 𝑖 ) ∈ L.
Since L = K[𝑎 1 , . . . , 𝑎 𝑛 ], it follows that 𝜎(L) ⊂ L. Since 𝜎 is injective, one
has [𝜎(L) : K] = [L : K], hence 𝜎(L) = L.
(iii)⇒(i). Let Q ∈ K[X] be an irreducible polynomial which has a root 𝑏
in L. We need to show that Q is split in L. Let Ω be an algebraic closure
of L and let 𝑏 0 ∈ Ω be a root of Q. Let 𝜑 : K[𝑏] → Ω be the unique
K-morphism such that 𝜑(𝑏) = 𝑏 0. The morphism 𝜑 can be extended to
a K-morphism 𝜎 from L to Ω. By assumption, 𝜎(L) ⊂ L. In particular,
𝑏 0 = 𝜎(𝑏) ∈ L. 

Corollary (4.6.9). — Let K ⊂ L be a finite normal extension and let G =

Aut(L/K). The extension K → LG is radicial. In particular, if the characteristic
of K is 0 or if K is perfect, then LG = K.

Proof. — Let us prove that every element of LG is inseparable over K.

Let 𝑎 ∈ LG and let P ∈ K[T] be its minimal polynomial. Let 𝑗 : L → Ω
be an algebraic closure of L. For every root 𝑏 of P in Ω, there exists a
unique K-morphism 𝑓 : K[𝑎] → Ω such that 𝑓 (𝑎) = 𝑏. Then 𝑓 extends to
a K-morphism 𝜑 : L → Ω and, by definition of a normal extension, the
image of 𝜑 is equal to 𝑗(L), so that 𝜑 ∈ Aut(L/K). By definition of LG ,
one thus has 𝑎 = 𝜑(𝑎) = 𝑏. This implies that 𝑎 is the only root of P in Ω,
and 𝑎 is radicial over K.
If, moreover, the extension K → L is separable, one then has 𝑎 ∈ K, so
that LG = K in this case. 

Remark (4.6.10). — Let K ⊂ L be a finite extension of fields. There

exists a finite extension L0 of L such that the composed extension
K → L0 is normal. Indeed, let 𝑎1 , . . . , 𝑎 𝑛 be elements of L such that
L = K[𝑎1 , . . . , 𝑎 𝑛 ]. For every 𝑖, let P𝑖 be the minimal polynomial of P𝑖 and
let P be the least common multiple of P1 , . . . , P𝑛 . If Ω is any extension
of L in which P is split (for example, an algebraic closure of L), the
field L0 generated by the roots of P in L0 is normal.
 4.6. GALOIS’S THEORY OF ALGEBRAIC EXTENSIONS 207

Assume moreover that the extension K ⊂ L is separable. By the

Primitive element theorem, there exists 𝑎 ∈ L such that L = K[𝑎]. The
minimal polynomial P of 𝑎 is separable and the field L0 generated by
its roots in some extension Ω in which it is split is a Galois extension
of K containing L. In fact, this field L0 is the smallest Galois extension
of K contained in Ω such that L ⊂ L0. Indeed, every such extension must
contain the roots of P, hence L0.

Proposition (4.6.11). — Let K ⊂ L be a finite Galois extension with group

G = Gal(L/K). Let H be a subgroup of G and let NG (H) = {𝜎 ∈
Gal(L/K) ; 𝜎H𝜎 −1 = H} be the normalizer of H in Gal(L/K).
a) For any 𝜎 ∈ Gal(L/K), one has 𝜎(LH ) = L𝜎H𝜎 .
−1

b) The morphism NG (H) → Aut(LH /K) is surjective and its kernel is H.

In particular, the extension K ⊂ LH is Galois if and only if H is a normal
subgroup of G. In that case, Gal(LH /K) ' G/H

Proof. — a) Let 𝑎 ∈ L. That 𝑎 belongs to LH is equivalent to the relation

ℎ(𝑎) = 𝑎 for every ℎ ∈ H. Consequently, 𝑏 = 𝜎(𝑏) belongs to 𝜎(LH ) if
and only if ℎ ◦ 𝜎 −1 (𝑏) = 𝜎 −1 (𝑏) for every ℎ ∈ H, hence if and only if
(𝜎ℎ𝜎 −1 )(𝑏) = 𝑏 for every ℎ ∈ H, hence if and only if 𝑏 ∈ L𝜎H𝜎 .
−1

b) By what precedes, any element 𝜎 of NG (H) induces by restriction

a K-morphism from LH to itself. The restriction map 𝜎 ↦→ 𝜎| LH is thus
a morphism of groups 𝜌 from NG (H) to Aut(LH /K). The kernel of this
morphism consists in those elements 𝜎 ∈ NG (H) such that 𝜎(𝑎) = 𝑎 for
every 𝑎 ∈ LH , hence Ker(𝜌) = H.
Let us show that 𝜌 is surjective. Let 𝜎 ∈ Aut(LH /K). Let us view 𝜎 as a
K-morphism from LH to L. Let us consider an extension 𝜎1 : L → Ω of 𝜎,
where Ω is an algebraically closed extension of L. Since the extension
K ⊂ L is Galois, it is normal and one has 𝜎1 (L) = L; in other words,
𝜎1 ∈ Aut(L/K) and 𝜎 = 𝜎1 | LH . Since 𝜎1 (LH ) = 𝜎(LH ) = LH , one has
𝜎1 ∈ NG (H) and 𝜎 = 𝜌(𝜎1 ).
Passing to the quotient by H, the morphism 𝜌 induces an isomorphism
NG (H)/H ' Aut(LH /K). In particular, the extension K ⊂ LH is Galois
if and only if [LH : K] = Card(NG (H))/Card(H). Since the extension
LH ⊂ L is Galois of group H, one has [L : LH ] = Card(H); moreover,
[L : K] = Card(G), so that [LH : K] = Card(G)/Card(H). Consequently,
208 CHAPTER 4. FIELD EXTENSIONS

the extension K ⊂ LH is Galois if and only if Card(NG (H)) = Card(G),

that is, if and only if G = NG (H), which means that H is a normal
subgroup of G. 

4.7. Norms and traces

Let A be a ring and let B be an A-algebra; let us assume that B is a free
finitely generated A-module. An important case of this situation arises
when A is a field and B is a finite extension of A.
Since A is commutative, all bases of B have the same cardinality, equal to
𝑛 = rkA (B), by definition of the rank of B. For any 𝑏 ∈ B, multiplication by
𝑏 gives rise to an endomorphism 𝑥 ↦→ 𝑏𝑥 of the A-module B; one defines
the characteristic polynomial of 𝑏 as the characteristic polynomial P𝑏 of
this endomorphism. This polynomial is defined either as det(X − M𝑏 ),
where M𝑏 ∈ M𝑛 (A) is the matrix of the multiplication by 𝑏, written in
any basis of B.
Let us write
P𝑏 (X) = X𝑛 − TrB/A (𝑏)X𝑛−1 + · · · + (−1)𝑛 NB/A (𝑏).

Definition (4.7.1). — The elements TrB/A (𝑏) and NB/A (𝑏) are called the trace
and the norm of 𝑏 with respect to A.

Note that TrB/A (𝑏) and NB/A (𝑏) are the trace and the norm of the
matrix U𝑏 ∈ M𝑛 (A) of multiplication by 𝑏, written in any basis of B.

Remark (4.7.2). — Using notions that we will introduce later, an intrinsic

definition is possible: consider the A[X]-module B[X] = A[X] ⊗A B;
it is free of rank 𝑛 and multiplication by X − 𝑏 in B[X] induces an
endomorphism of the free rank-1 module Λ𝑛 B[X], which is precisely
multiplication by P𝑏 (X). Moreover, Λ𝑛 (B) is a free A-module of rank 1
and multiplication by 𝑏 in B induces multiplication by NB/A (𝑏) on Λ𝑛 (B).

Proposition (4.7.3). — a) One has TrB/A (1B ) = rkA (B)1A and NB/A (1B ) =
1A ;
b) For any 𝑎 ∈ A and any 𝑏 ∈ B, one has TrB/A (𝑎𝑏) = 𝑎 TrB/A (𝑏) and
NB/A (𝑎𝑏) = 𝑎 rkA (B) NB/A (𝑏);
 4.7. NORMS AND TRACES 209

c) For any 𝑏, 𝑏 0 ∈ B, TrB/A (𝑏 + 𝑏 0) = TrB/A (𝑏) + TrB/A (𝑏 0) and NB/A (𝑏𝑏 0) =

NB/A (𝑏)NB/A (𝑏 0);
d) An element 𝑏 ∈ B is invertible if and only if its norm NB/A (𝑏) is invertible
in A.

Proof. — Let us fix a basis of B as an A-module; let 𝑛 = rkA (B).

a) Multiplication by 1B is the identity, hence its matrix is I𝑛 . Conse-
quently, TrB/A (1B ) = 𝑛1A and NB/A (1B ) = 1A .
b) If U is the matrix of multiplication by 𝑏, the matrix of multiplication
by 𝑎𝑏 is 𝑎U. Then, TrB/A (𝑎𝑏) = Tr(𝑎U) = 𝑎 Tr(U) = 𝑎 TrB/A (𝑏) and
NB/A (𝑎𝑏) = det(𝑎U) = 𝑎 𝑛 det(U) = 𝑎 𝑛 NB/A (𝑏).
c) Let U0 be the matrix of multiplication by 𝑏 0. Since (𝑏 + 𝑏 0)𝑥 = 𝑏𝑥 + 𝑏 0 𝑥
for every 𝑥 ∈ B, the matrix of multiplication by 𝑏 + 𝑏 0 is U + U0. Then

TrB/A (𝑏 + 𝑏 0) = Tr(U + U0) = Tr(U) + Tr(U0) = TrB/A (𝑏) + TrB/A (𝑏 0).

Moreover, the formula (𝑏𝑏 0)𝑥 = 𝑏(𝑏 0 𝑥), for 𝑥 ∈ B, implies that the matrix
of multiplication by 𝑏𝑏 0 is UU0. Then

NB/A (𝑏𝑏 0) = det(UU0) = det(U) det(U0) = NB/A (𝑏)NB/A (𝑏 0).

d) Assume that 𝑏 is invertible, with inverse 𝑏 0. Then 𝑏𝑏 0 = 1B hence

multiplication by 𝑏 is an isomorphism of B, whose inverse is multipli-
cation by 𝑏 0. Consequently, its determinant is invertible, that is to say,
NB/A (𝑏) is invertible.
Conversely, if NB/A (𝑏) is invertible, then multiplication by 𝑏 is invertible.
In particular, there exists 𝑏 0 ∈ B such that 𝑏𝑏 0 = 1B and 𝑏 is invertible. 

Proposition (4.7.4). — Let K → L be a finite separable extension, let Ω be an

algebraic closure of L and let Σ be the set of K-morphisms from L to Ω. Then,
for every 𝑥 ∈ L, one has
Õ Ö
TrL/K (𝑥) = 𝜎(𝑥) and NL/K (𝑥) = 𝜎(𝑥).
𝜎∈Σ 𝜎∈Σ

Proof. — Let 𝑥 ∈ L, let P ∈ K[X] be its minimal polynomial, and let

𝑥 = 𝑥1 , . . . , 𝑥 𝑑 be its (pairwise distinct) roots in Ω. By assumption, 𝑥
210 CHAPTER 4. FIELD EXTENSIONS

is separable, hence P(X) = (X − 𝑥1 ) . . . (X − 𝑥 𝑑 ). Let 𝑎1 , . . . , 𝑎 𝑑 ∈ K be

defined by the relation
P(X) = X𝑑 − 𝑎1 X𝑑−1 + · · · + (−1)𝑑 𝑎 𝑑 ,
so that 𝑎1 = 𝑥 1 + · · · + 𝑥 𝑑 and 𝑎 𝑑 = 𝑥 1 . . . 𝑥 𝑑 . In the basis {1, 𝑥, . . . , 𝑥 𝑑−1 }
of K[𝑥], the matrix of multiplication by 𝑥 is given by
0 (−1)𝑑−1 𝑎 𝑑
© .
­1 . . ..
.
ª
®
­ ... 0
­ ®
­ −𝑎2 ®®
« 1 𝑎1 ¬
hence is the companion matrix CP . In particular, its trace is 𝑎 1 and its
determinant is 𝑎 𝑑 . In other words,
𝑑
Õ 𝑑
Ö
TrK[𝑥]/K (𝑥) = 𝑥𝑖 and NK[𝑥]/K (𝑥) = 𝑥𝑖 .
𝑖=1 𝑖=1

Let now {𝑒1 , . . . , 𝑒 𝑠 } be a basis of L over K[𝑥]. The family (𝑒 𝑖 𝑥 𝑗 ), indexed

by pairs (𝑖, 𝑗) of integers such that 1 6 𝑖 6 𝑠 and 0 6 𝑗 < 𝑑, is a basis of L
as a K-vector space. When this basis is written in lexicographic order
(𝑒1 , 𝑒1 𝑥, . . . , 𝑒1 𝑥 𝑑−1 , 𝑒2 , . . . , 𝑒 𝑠 𝑥 𝑑−1 ), the matrix of multiplication by 𝑥 is
block-diagonal, with 𝑠 blocks all equal to the companion matrix CP .
Consequently, TrL/K (𝑥) = 𝑠𝑎1 and NL/K (𝑥) = 𝑎 𝑑𝑠 .
Recall that for every 𝑖 ∈ {1, . . . , 𝑑} there is exactly one K-morphism 𝜎𝑖
from K[𝑥] to Ω such that 𝑥 ↦→ 𝑥 𝑖 , and every K-morphism from K[𝑥]
to Ω is one of them. Moreover, since the extension K[𝑥] ⊂ L is separable
of degree 𝑠, each of these morphisms extends in exactly 𝑠 ways to L.
Therefore,
Õ 𝑑
Õ
𝜎(𝑥) = 𝑠 𝜎𝑖 (𝑥) = 𝑠 TrK[𝑥]/K (𝑥) = 𝑠𝑎1 = TrL/K (𝑥)
𝜎∈Σ 𝑖=1

and
𝑑
!𝑠
Ö Ö
𝑠
𝜎(𝑥) = 𝜎𝑖 (𝑥) = NK[𝑥]/K (𝑥) = 𝑎 𝑑𝑠 = NL/K (𝑥). 
𝜎∈Σ 𝑖=1
 4.7. NORMS AND TRACES 211

Corollary (4.7.5). — If K → L is a Galois extension, then

Ö Õ
NL/K (𝑥) = 𝜎(𝑥) and TrL/K (𝑥) = 𝜎(𝑥)
𝜎∈Gal(L/K) 𝜎∈Gal(L/K)

for every 𝑥 ∈ L.
Proof. — Let Ω be an algebraic closure of L. Since the extension K → L
is Galois, it is normal the K-morphisms from L to Ω coincide with the
elements of Gal(L/K). Since a Galois extension is separable, the result
follows from the preceding proposition. 

Corollary (4.7.6). — Let K ⊂ L be a finite extension of fields, let A be a subring

of K and let B be its integral closure in L. The trace TrL/K (𝑥) and the norm
NL/K (𝑥) of any element 𝑥 ∈ B are integral over A. In particular, if A is
integrally closed, then TrL/K (𝑥) and NL/K (𝑥) belong to A, for every 𝑥 ∈ B.
Proof. — Let Ω be an algebraic closure of L, let 𝑛 = [L : K] and let
𝜎1 , . . . , 𝜎𝑛 : L → Ω be the 𝑛 distinct K-morphisms from L to Ω. If
𝑖 ∈ {1, . . . , 𝑛} and 𝑥 ∈ B, then 𝜎𝑖 (𝑥) is integral over A (it is a root of the
same monic polynomial as 𝑥), so that TrL/K (𝑥) = 𝑛𝑖=1 𝜎𝑖 (𝑥) is integral
Í
over A. Similarly, NL/K (𝑥) = 𝑛𝑖=1 𝜎𝑖 (𝑥) is integral over A.
Î
Moreover, TrL/K and NL/K (𝑥) are elements of K. If A is integrally closed
in K, this implies TrL/K (𝑥) ∈ A and NL/K (𝑥) ∈ A. 

Theorem (4.7.7). — Let K ⊂ L be a finite separable extension. The map

𝑡 : L × L → K defined by 𝑡(𝑥, 𝑦) = TrL/K (𝑥𝑦) is a non-degenerate symmetric
K-bilinear form.
Proof. — The map 𝑡 is symmetric since the multiplication is commutative.
The formulas
𝑡(𝑎𝑥 + 𝑎 0 𝑥 0 , 𝑦) = TrL/K ((𝑎𝑥 + 𝑎 0 𝑥 0)𝑦) = TrL/K (𝑎𝑥𝑦 + 𝑎 0 𝑥 0 𝑦)
= 𝑎 TrL/K (𝑥𝑦) + 𝑎 0 TrL/K (𝑥 0 𝑦) = 𝑎𝑡(𝑥, 𝑦) + 𝑎 0𝑡(𝑥 0 , 𝑦)
show that it is K-linear with respect to the first variable. By symmetry, it
is also K-linear with the second one.
For every 𝑦 ∈ L, let 𝜏(𝑦) be the map defined by 𝜏(𝑦)(𝑥) = 𝑡(𝑥, 𝑦).
Since 𝑡 is K-linear with respect to the second variable, one has 𝜏(𝑦) ∈
HomK (L, K). Moreover, the map 𝜏 : L → HomK (L, K) is K-linear. We
212 CHAPTER 4. FIELD EXTENSIONS

need to prove that 𝜏 is an isomorphism of K-vector spaces. Let 𝑑 = [L : K],

so that 𝑑 = dimK (L) = dimK (HomK (L, K)). Consequently, it suffices to
prove that 𝜏 is injective.
Let Ω be an algebraic closure of L and let 𝜎1 , . . . , 𝜎𝑑 be the 𝑑 distinct
K-morphisms from L to Ω. For every 𝑥 ∈ L, one has
𝑑
Õ
TrL/K (𝑥) = 𝜎 𝑗 (𝑥).
𝑗=1

Let us assume that 𝑦 ∈ Ker(𝜏). Then, for every 𝑥 ∈ L, TrL/K (𝑥𝑦) = 0,

hence
𝑑
Õ 𝑑
Õ
0= 𝜎 𝑗 (𝑥𝑦) = 𝜎 𝑗 (𝑥)𝜎 𝑗 (𝑦).
𝑗=1 𝑗=1
Since 𝜎1 , . . . , 𝜎𝑑 are linearly independent on L (proposition 4.4.12), this
implies 𝜎 𝑗 (𝑦) = 0 for all 𝑗, hence 𝑦 = 0. 

Remark (4.7.8). — Let K → L be a finite extension which is not separable.

Let us prove that TrL/K (𝑥) = 0 for every 𝑥 ∈ L. The characteristic of K is a
prime number 𝑝. Let Lsep be the subfield of L generated by all elements
of L which are separable over K. By remark 4.4.17, the degree [L : Lsep ]
is a multiple of 𝑝.
Let 𝑥 ∈ L. If 𝑥 ∈ Lsep , one has
TrL/K (𝑥) = [L : Lsep ]] TrLsep /K (𝑥) = 0
since 𝑝 divides [L : K[𝑥]]. Otherwise, 𝑥 ∈ L Lsep is inseparable over K;
then its minimal polynomial is a polynomial in X𝑝 , so has no term of
degree one less its degree and TrK[𝑥]/K (𝑥) = 0. Then, TrL/K (𝑥) = [L :
K[𝑥]] TrK[𝑥]/K (𝑥) = 0.

4.8. Transcendence degree

Proposition (4.8.1). — Let K → L be a field extension, let (𝑥 𝑖 )𝑖∈I be a family of
elements of L. The following propositions are equivalent:
(i) For every non-zero polynomial P ∈ K[(X𝑖 )𝑖∈I ], one has P((𝑥 𝑖 )) ≠ 0;
(ii) The canonical morphism of K-algebras from K[(X𝑖 )𝑖 ] to L such that
X𝑖 ↦→ 𝑥 𝑖 is injective;
 4.8. TRANSCENDENCE DEGREE 213

(iii) There exists a K-morphism of fields K((X𝑖 )𝑖 ) → L such that X𝑖 ↦→ 𝑥 𝑖 .

Proof. — Let 𝜑 : K[(X𝑖 )𝑖 ] → L be the unique morphism of K-algebras

such that 𝜑(X𝑖 ) = 𝑥 𝑖 for all 𝑖 ∈ I.
Condition (i) means that 𝜑(P) ≠ 0 if P ≠ 0; it is thus equivalent to the
equality Ker(𝜑) = 0 of (ii).
Assume (ii). Note that K[(X𝑖 )𝑖∈I ] is an integral domain, and its field of
fractions is K((X𝑖 )]. If 𝜑 is injective, then, since L is a field, it extends to
a morphism of fields from K[(X𝑖 )] to L, hence (iii).
Finally, assume that (iii). Let 𝜓 : K((X𝑖 )) → L be a K-morphism of
fields such that 𝜓(X𝑖 ) = 𝑥 𝑖 for all 𝑖. The restriction of 𝜓 to K[(X𝑖 )] is the
morphism of K-algebras which was denoted by 𝜑. Since 𝜓 is injective,
𝜑 is injective too, hence (ii). 

Definition (4.8.2). — If these conditions hold, one says that the family (𝑥 𝑖 )𝑖∈I
is algebraically independent over K.

Note the analogy with the similar proposition 3.5.2 in linear algebra,
where we considered linear relations between elements of a module,
while we know study algebraic relations in a ring, and the theory of
matroids pushes this analogy farther than mere analogous language.
Based on this analogy, a possible definition of “basis” in this context
can be However, in the present context, “generating” families are not
families that generate the extension L, but those for which L is algebraic
over the field it generates.

Definition (4.8.3). — Let K → L be a field extension. One says that a family

(𝑥 𝑖 )𝑖∈I in L is a transcendence basis of L over K if it is algebraically independent
over K and if L is algebraic over the subfield K((𝑥 𝑖 )).

By abuse of language, one says that a subset B of L is algebraically

independent (resp. a transcendence basis of L) over K if the family (𝑏)𝑏∈B
is algebraically independent (resp. a transcendence basis of L) over K.

Theorem (4.8.4). — Let K → L be a field extension.

a) Let B, C be subsets of L. Assume that L is algebraic over the subfield K(C)
generated by C over K and that B is algebraically independent. Then B is a
214 CHAPTER 4. FIELD EXTENSIONS

transcendence basis if and only if, for every 𝑐 ∈ C B, the subset B ∪ {𝑐} is
not algebraically independent.
b) Let A and C be subsets of L such that A ⊂ C. One assumes that A is
algebraically independent over K and that L is algebraic over the subfield K(C)
generated by C over K. Then, there exists a transcendence basis B of L such
that A ⊂ B ⊂ C.
Proof. — a) Assume that B is a transcendence basis of L. Let 𝑐 ∈ C B.
By definition, L is algebraic over K(B); in particular, 𝑐 is algebraic
over K(B). Let P1 ∈ K(B)[T] be a non-zero polynomial in one inde-
terminate T such that P1 (𝑐) = 0. Write P1 = 𝑎 𝑛 T𝑛 + · · · + 𝑎0 , where
𝑎0 , . . . , 𝑎 𝑛 ∈ K(B). For 𝑗 ∈ {0, . . . , 𝑛}, there exist polynomials 𝑓 𝑗 , 𝑔 𝑗 ∈
K[(X𝑏 )𝑏∈B ] such that 𝑎 𝑗 = 𝑓 𝑗 ((𝑏))/𝑔 𝑗 ((𝑏)). Multiplying P1 by the product
of the elements 𝑔 𝑗 ((𝑏)), we reduce to the case where 𝑔 𝑗 = 1 for every 𝑗.
Let P = 𝑛𝑗=0 𝑓 𝑗 ((X𝑏 ))T 𝑗 ; this is a non-zero polynomial in K[(X𝑏 )𝑏∈B , T]. By
Í
construction, P((𝑏), 𝑐) = 0. This proves that the family deduced from B
by adjoining 𝑐 is not algebraically independent. Since 𝑐 ∉ B, B ∪ {𝑐} is
not algebraically independent.
Conversely, assume that for every 𝑐 ∈ C B, B ∪ {𝑐} is not algebraically
independent. Let us prove that every element 𝑐 ∈ C is algebraic over K(B).
This is obvious if 𝑐 ∈ B; otherwise, consider a non-zero polynomial
P ∈ K[(X𝑏 )𝑏∈B , T] such that P((𝑏), 𝑐) = 0. Write P = 𝑓𝑛 T𝑛 + · · · + 𝑓0 , where
𝑓0 , . . . , 𝑓𝑛 ∈ K[(X𝑏 )𝑏∈B ], and 𝑓𝑛 ≠ 0. Since B is algebraically independent,
one has 𝑓𝑛 ((𝑏)) ≠ 0. Consequently, the polynomial P1 = P((𝑏), T) =
𝑓𝑛 ((𝑏))T𝑛 + · · · + 𝑓0 ((𝑏)) is a non-zero element of K(B)[T] and P1 (𝑐) = 0.
This proves that 𝑐 is algebraic over K(B).
Then K(C) is algebraic over K(B) and, since L is algebraic over K(C), L
is algebraic over K(B). This proves that B is a transcendence basis of L.
b) Let ℬ be the set of all subsets B of L such that A ⊂ B ⊂ C and such
that B is algebraically independent over K.
Let us prove that the set ℬ is inductive when ordered by inclusion.
It is non-empty, since A ∈ ℬ. Let then (B𝑖 ) be a non-empty totally
ordered family of elements of ℬ and let B be its union. One has A ⊂ B
because the family is non-empty, and B ⊂ C because every B𝑖 is contained
in C. Moreover, B is algebraically independent over K. Let indeed P
be a polynomial in indeterminates (X𝑏 )𝑏∈B such that P((𝑏)𝑏∈B ) = 0.
 4.8. TRANSCENDENCE DEGREE 215

The polynomial P depends on only finitely many indeterminates, say

X𝑏1 , . . . , X𝑏 𝑛 , where 𝑏 1 , . . . , 𝑏 𝑛 ∈ B. One has P(𝑏 1 , . . . , 𝑏 𝑛 ) = 0 Since
the family (B𝑖 ) is totally ordered, there exists an index 𝑖 such that
{𝑏 1 , . . . , 𝑏 𝑛 } ⊂ B𝑖 . Since B𝑖 is algebraically independent over K, one has
P = 0.
By Zorn’s lemma (theorem A.2.11), the set ℬ has a maximal element,
say B. One has A ⊂ B ⊂ C, by construction, and, by definition of a
maximal element, B ∪ {𝑐} is not algebraically independent, for every
𝑐 ∈ C B. By a), the set B is a transcendence basis of L. 
Applying the theorem to A = ∅ and L = C, we obtain:

Corollary (4.8.5). — Any field extension has a transcendence basis.

Theorem (4.8.6) (Exchange lemma). — Let K → L be a field extension, let B

be a transcendence basis of L over K and let C be a subset of L such that L is
algebraic over K(C). Then, for every 𝛽 ∈ B C, there exists 𝛾 ∈ C B such
that (B {𝛽}) ∪ {𝛾} is a transcendence basis of L over K.

Proof. — Set A = B {𝛽}.

Since B = A ∪ {𝛽} is algebraically independent, the element 𝛽 is not
algebraic over K(A) and, in particular, L is not algebraic over K(A). On
the other hand, L is algebraic over K(C). Consequently, there must exist
𝛾 ∈ C such that 𝛾 is not algebraic over K(A). Let B0 = A ∪ {𝛾} and let us
show that B0 is a transcendence basis of L over K.
I claim that there exists a non-zero polyomial P in indetermi-
nates (X𝑏 )𝑏∈B and T such that P((𝑏), 𝛾) = 0. Indeed, since L is algebraic
over K(B), there exists an irreducible polynomial R1 ∈ K(B)[T] such
that R1 (𝛾) = 0. Since B is algebraically independent over K, there
exists a unique polynomial R ∈ K((X𝑏 ))[T], whose coefficients are
rational functions in indeterminates X𝑏 , such that R1 = R((𝑏))[T]. If we
multiply R be a common denominator of the coefficients of R, we obtain
a non-zero polynomial P ∈ K[(X𝑏 ), T] such that P((𝑏), 𝛾) = 0, as claimed.
Since 𝛾 is not algebraic over K(A), D is algebraically independent
𝑗
over K. We then write the polynomial P as a polynomial 𝑚
Í
𝑗=0 P 𝑗 X𝛽
in X𝛽 with coefficients in the indeterminates X𝑏 , for 𝑏 ∈ A, as well as T,
where 𝑚 = degX𝛽 (P), so that P𝑚 ≠ 0. Since 𝛾 is not algebraic over K(A),
216 CHAPTER 4. FIELD EXTENSIONS

P𝑚 ((𝑏)𝑏∈A , 𝛾) ≠ 0. Consequently, the relation

𝑚
Õ
P 𝑗 ((𝑏)𝑏∈A , 𝛾)𝛽 𝑗 = P((𝑏), 𝛾) = 0
𝑗=0

shows that 𝛽 is algebraic over K(D). Any other element of B belongs

to D, hence is algebraic over K(D) as well. Therefore, K(B) is algebraic
over K(D). Since L is algebraic over K, L is algebraic over K(D). This
concludes the proof that D is a transcendence basis of L over K. 

Corollary (4.8.7). — Let K → L be an extension of fields. All transcendence

bases of L over K have the same cardinality.
Proof. — Let B and B0 be two transcendence bases of L over K. According
to the following lemma, there exists an injection 𝜑 from B to B0, as well
as an injection 𝜑0 from B0 to B. It then follows from the Cantor-Bernstein
theorem that B and B0 are equipotent. 

Lemma (4.8.8). — Let K → L be an extension of fields, let A and B be subsets

of L such that A is algebraically independent over K and L is algebraic over K(B).
Then there exists an injection from A to B.
Proof. — Let ℱ be the set of all pairs (A0 , 𝜑) where A0 is a subset
of A and 𝜑 : A0 → B is an injection such that (A A0) ∪ 𝜑(A0) is
algebraically independent over K. We order the set ℱ by saying that
(A01 , 𝜑1 ) ≺ (A02 , 𝜑2 ) if A01 ⊂ A02 and 𝜑2 (𝑎) = 𝜑1 (𝑎) for every 𝑎 ∈ A01 .
Let us prove that the set ℱ is inductive.
Let (A0𝑖 , 𝜑 𝑖 )𝑖 be a totally ordered family of elements of ℱ. Let A0 be
the union of the A0𝑖 and let 𝜑 : A0 → B be the unique function such
that 𝜑(𝑎) = 𝜑 𝑖 (𝑎) for any index 𝑖 such that 𝑎 ∈ A0𝑖 . The map 𝜑 is
injective. Let us indeed consider two elements 𝑎, 𝑎 0 ∈ A0 such that
𝜑(𝑎) = 𝜑(𝑎 0). Let 𝑖 be any index such that 𝑎 and 𝑎 0 both belong to A0𝑖 .
Then 𝜑 𝑖 (𝑎) = 𝜑(𝑎) = 𝜑(𝑎 0) = 𝜑 𝑖 (𝑎 0). Since 𝜑 𝑖 is injective, 𝑎 = 𝑎 0.
Let us show that (A A0) ∪ 𝜑(A0) is algebraically independent. Let P
be a polynomial in indeterminates (X𝑎 )𝑎∈A A0 and (Y𝑎 )𝑎∈A0 such that
P((𝑎)𝑎∉A0 , (𝜑(𝑎))𝑎∈A0 ) = 0 and let us prove that P = 0. Only finitely many
indeterminates Y𝑎 , for 𝑎 ∈ A0, appear in P; since the family (A0𝑖 ) of
subsets of A is totally ordered, all the corresponding elements 𝑎 belong
 4.8. TRANSCENDENCE DEGREE 217

to some common set A0𝑖 . The polynomial P can then be viewed as an

algebraic dependence relation for (A A0) ∪ 𝜑(A0𝑖 ). Since this set is
contained in the transcendence basis (A A0𝑖 ) ∪ 𝜑(A0𝑖 ), the polynomial P
is zero, as claimed.
The pair (A0 , 𝜑) is an element of ℱ such that (A0𝑖 , 𝜑 𝑖 ) ≺ (A0 , 𝜑) for
every 𝑖. This shows that ℱ is inductive.
By Zorn’s theorem, the set ℱ has a maximal element, say (A0 , 𝜑). Let us
prove that A0 = A. Otherwise, let B0 be a transcendence basis of L over K
which contains (A A0)∪ 𝜑(A0) and let 𝛽 be any element of A A0. By the
exchange lemma (theorem 4.8.6), there exists 𝛾 ∈ B such that the family
obtained by merging (A A0 {𝛽}), 𝜑(A0) and 𝛾 is a transcendence basis
of L over K. Let A00 = A0 ∪ {𝛽} and let 𝜑0 : A00 → B be the map which
coincides with 𝜑 on A0 and such that 𝜑0(𝛽) = 𝛾. By construction, the
family (𝜑0(𝑎))𝑎∈A00 is algebraically independent; in particular, the map 𝜑0
is injective. Moreover, (A A00) ∪ 𝜑0(A00) is algebraically independent.
This shows that (A00 , 𝜑0) belongs to ℱ, which contradicts the hypothesis
that (A0 , 𝜑) were a maximal element. 

Definition (4.8.9). — The cardinality of any transcendence basis of L over K is

called the transcendence degree of L over K and denoted by tr degK (L).

Example (4.8.10). — For any integer 𝑛, the field K(X1 , . . . , X𝑛 ) of rational

functions in 𝑛 indeterminates has transcendence degree 𝑛 over K.
Indeed, it just follows from the definition that {X1 , . . . , X𝑛 } is a tran-
scendence basis.

Example (4.8.11). — Let K be a field, let P ∈ K[X1 , . . . , X𝑛 ] be an irre-

ducible polynomial and let L be the field of fractions of the integral
domain A = K[X1 , . . . , X𝑛 ]/(P). Let us show that tr degK (L) = 𝑛 − 1.
For 𝑗 ∈ {1, . . . , 𝑛}, let 𝑥 𝑗 be the image of X 𝑗 in L.
Since P is irreducible, it is non-constant and its degree in at least
one of the indeterminates is at least 1. For simplicity of notation,
assume that 𝑚 = degX𝑛 (P) > 1 and write P = 𝑓𝑚 X𝑚 𝑛 + · · · + 𝑓0 , where
𝑓0 , . . . , 𝑓𝑚 ∈ K[X1 , . . . , X𝑛−1 ] and 𝑓𝑚 ≠ 0.
Let us first prove that it is algebraically independent. Let Q ∈
K[X1 , . . . , X𝑛−1 ] be a non-zero polynomial such that Q(𝑥1 , . . . , 𝑥 𝑛−1 ) =
218 CHAPTER 4. FIELD EXTENSIONS

0. This implies that Q belongs to the ideal (P) generated by P in

K[X1 , . . . , X𝑛 ]. Since degX𝑛 (Q) = 0 < degX𝑛 (P), we get a contradiction.
In particular, 𝑓𝑚 (𝑥1 , . . . , 𝑥 𝑛−1 ) ≠ 0. Then the expression
0 = P(𝑥 1 , . . . , 𝑥 𝑛 ) = 𝑓𝑚 (𝑥1 , . . . , 𝑥 𝑛−1 )𝑥 𝑛𝑚 + · · · + 𝑓0 (𝑥1 , . . . , 𝑥 𝑛−1 )
proves that 𝑥 𝑛 is algebraic over K(𝑥1 , . . . , 𝑥 𝑛−1 ). Since L = K(𝑥1 , . . . , 𝑥 𝑛 ),
every element of L is then algebraic over K(𝑥 1 , . . . , 𝑥 𝑛−1 ).
This proves that (𝑥1 , . . . , 𝑥 𝑛−1 ) is a transcendence basis of L; in particu-
lar, tr degK (L) = 𝑛 − 1.

Exercises
1-exo1151) Let A be a ring and let P ∈ A[T] be a monic polynomial of degree 𝑛.
a) Consider the quotient ring B = A[T]/(P). Prove that B is integral over A and that
there exists 𝑏 ∈ B such that P(𝑏) = 0.
b) Prove that B is a free A-module of rank 𝑛. In particular, the natural morphism
A → B is injective.
c) In the ring C = A[X1 , . . . , X𝑛 ], let I be the ideal generated by the coefficients of the
polynomial
𝑛
Ö
P− (T − X𝑖 ) ∈ C[T].
𝑖=1
Prove that the ring C/I is a free A-module of rank 𝑛! = 𝑛(𝑛 − 1) . . . 2 · 1. (This ring is
called the splitting algebra of P.)
d) Assume that A is a field and let M be a maximal ideal of C. Prove that the field
C/M is an extension of A in which the polynomial P is split and which is generated by
the roots of P.
2-exo1116) Let B be a ring, let A be a subring of B and let 𝑥 ∈ B be an invertible element
of B.
Let 𝑦 ∈ A[𝑥] ∩ A[𝑥 −1 ].
a) Prove that there exists an integer 𝑛 such that the A-submodule M = A + A𝑥 + · · · +
A𝑥 𝑛 of B satisfies 𝑦M ⊂ M.
b) Prove that 𝑦 is integral over A.
3-exo1173) Let A be an integral domain and let K be its field of fractions.
a) Let 𝑥 ∈ K be integral over A. Prove that there exists 𝑎 ∈ A {0} such that 𝑎𝑥 𝑛 ∈ A
for every 𝑛 ∈ N.
b) Assume that A is a noetherian ring. Let 𝑎 ∈ A {0} and 𝑥 ∈ K be such that
𝑎𝑥 𝑛 ∈ A for every 𝑛 ∈ N. Prove that 𝑥 is integral over A.
 EXERCISES 219

4-exo807) a) Let A be a commutative ring, let B be a commutative A-algebra and

let U ∈ M𝑛 (B) be a matrix. Let Δ = det(U). Let M = B𝑛 be the free B-module, with
canonical basis (𝑒1 , . . . , 𝑒 𝑛 ). The matrix U is viewed as an endomorphism 𝑢 of M.
We assume that U is integral over A.
b) Let V ⊂ M be the submodule generated by the elements 𝑢 𝑘 (𝑒 𝑖 ), for 𝑖 ∈ {1, . . . , 𝑛}
and 𝑘 ∈ N. Prove that V is a finitely generated A-module, stable under 𝑢.
c) Construct a finitely generated submodule of Λ𝑛 (M) that contains 𝑒1 ∧ · · · ∧ 𝑒 𝑛 and
that is stable under Λ𝑛 (𝑢). Conclude that det(U) is integral over A.
d) Introducing the ring A[T] and the matrix TI𝑛 − U, prove that the coefficients of
the characteristic polynomial of U are integral over A. In particular, Tr(U) is integral
over A.
5-exo1231) Let 𝑓 : A → B be a morphism of commutative rings.
a) Assume that 𝑓 is integral. Prove that the continuous map 𝑓 ∗ : Spec(B) → Spec(A)
induced by 𝑓 is closed: the image of a closed subset is closed.
b) More generally, if 𝑓 is integral, prove that for every A-algebra C, the continuous
map ( 𝑓C )∗ : Spec(B ⊗A C) → Spec(C) is closed.
c) Let us consider the particular case C = A[T]. Observe that the map 𝑓C identifies
with the map 𝑔 : A[T] → B[T]. We assume that 𝑔 ∗ : Spec(B[T]) → Spec(A[T]) is
closed.
d) Let 𝑏 ∈ B and let Z = V(𝑏T − 1) in Spec(B[T]). Let I be an ideal of A such that
𝑔 ∗ (Z)
= V(I). Prove that I + (T) = A[T]. (Consider a prime ideal P of A that contains I + (T).)
e) Prove that 𝑏 is integral over A.
6-exo845) Let A be a commutative ring and let I be an ideal such that every element
of I is nilpotent. Let 𝑓 ∈ A[T] be a polynomial, let 𝑎 ∈ A be such that 𝑓 (𝑎) ∈ I and 𝑓 0(𝑎)
is a unit.
We define a sequence (𝑎 𝑛 ) of elements of A by the formula 𝑎 𝑛+1 = 𝑎 𝑛 − 𝑓 (𝑎 𝑛 ) 𝑓 0(𝑎)−1 .
a) By induction on 𝑛, prove that, modulo I𝑛 , 𝑎 𝑛 is the unique element of A such that
𝑎 𝑛 ≡ 𝑎 (mod I) and 𝑓 (𝑎 𝑛 ) ∈ I𝑛 .
b) Prove that 𝑓 (𝑎 𝑛 ) ∈ ( 𝑓 (𝑎)𝑛 ) for all 𝑛.
c) Conclude that there exists a unique element 𝑏 ∈ A such that 𝑓 (𝑏) = 0 and 𝑏 ≡ 𝑎
(mod I).
d) Let K be a field, let U ∈ M𝑛 (K) be a matrix. Let P be the characteristic polynomial
of U and let Q be the product of its irreducible factors. Assume that Q is separable
(this is automatic if K is a perfect field). Prove the Chevalley–Dunford decomposition
of U : there exists a unique pair (D, N) of commuting matrices such that D becomes
diagonalizable over an algebraic closure of K, N is nilpotent and U = D + N. (Solve the
equation Q(D) = 0 in K[U], with unknown D.)

√ Let 𝑑 ∈ Z which is not divisible by the square of an integer and let

7-exo1119)
K = Q[ 𝑑].
√
a) What is the minimal polynomial of an element 𝑥 = 𝑎 + 𝑏 𝑑 ∈ K, where 𝑎, 𝑏 ∈ Q?
220 CHAPTER 4. FIELD EXTENSIONS

√
b)
√ Prove that the integral closure of Z in K is Z[(1 + 𝑑)/2] if 𝑑 ≡ 1 (mod 4), and is
Z[ 𝑑] otherwise.
8-exo1164) a) Let A be an integral domain and let 𝑡 ∈ A be such that the ring A/𝑡A has
no nilpotent element except for 0. Assume also that the ring of fractions A𝑡 is integrally
closed. Prove that A is integrally closed.
b) Prove that the ring A = C[X, Y, Z]/(XZ − Y(Y + 1)) is integrally closed. (Apply the
first question with 𝑡 = X.)
9-exo1060) Let B be a ring and let A be a subring. Assume that A is integrally closed
in B.
a) Let P, Q be monic polynomials of B[T] such that PQ ∈ A[T]. Prove that P, Q ∈
A[T]. (Use exercise 4/1 to introduce a ring C containing B such that P = 𝑚𝑖=1 (T − 𝑎 𝑖 ) et
Î
Î𝑛
Q = 𝑖=1 (T − 𝑏 𝑖 ) where 𝑎 1 , . . . , 𝑎 𝑚 , 𝑏1 , . . . , 𝑏 𝑛 ∈ C. )
b) Prove that A[T] is integrally closed in B[T]. (If P ∈ B[T] is integral over A[T], consider
the monic polynomials Q = T𝑚 + P, for large enough integers 𝑚.)
10-exo1223) Let K → L be a finite algebraic extension and let Ω be an algebraic closure
of K.
a) Prove that the algebra L ⊗K Ω is reduced (0 is the only nilpotent element) if and
only if the extension K → L is separable.
b) In this case, prove that for every field extension M of K, the tensor product L ⊗K M
is reduced.
11-exo802) Let E be a field and let P, Q ∈ E[T] be two irreducible polynomials. Let
Ω be an algebraic closure of E and let 𝑥, 𝑦 ∈ Ω be such that P(𝑥) = Q(𝑦) = 0. Let
P1 , . . . , P𝑟 ∈ E[𝑦][T] be pairwise distinct irreducible polynomials and 𝑚1 , . . . , 𝑚𝑟 be
strictly positive integers such that P = 𝑟𝑖=1 P𝑚 𝑖 . Similarly, let Q1 , . . . , Q𝑠 ∈ E[𝑥][T] be
𝑖
Î
pairwise distinct irreducible polynomials and 𝑛1 , . . . , 𝑛 𝑠 be strictly positive integers
𝑛𝑗
such that Q = 𝑠𝑗=1 Q 𝑗 .
Î

a) Introducing the E-algebra R = E[X, Y]/(P(X), Q(Y)), construct isomorphisms of

E-algebras
R ' E[𝑥][Y]/(Q1 (Y)𝑛1 ) ⊕ · · · ⊕ E[𝑥][Y]/(Q𝑠 (Y)𝑛 𝑠 )
' E[𝑦][X]/(P1 (X)𝑚1 ) ⊕ · · · ⊕ E[𝑦][X]/(P𝑟 (X)𝑚𝑟 ).
b) What are the idempotents of R? (If F is a field, 𝑚 > 1, and P ∈ F[T] is an irreducible
polynomial, show that the only idempotents of F[T]/(P𝑚 ) are 0 and 1.)
c) Prove that 𝑟 = 𝑠 and that there exists a permutation 𝜎 ∈ 𝔖𝑟 such that if 𝑗 = 𝜎(𝑖),
𝑛𝑗
then E[𝑥][Y]/(Q 𝑗 ) and E[𝑦][X]/(P𝑚 𝑖
𝑖 ) are isomorphic as E-algebras.
d) In particular, if 𝑗 = 𝜎(𝑖), then 𝑛 𝑗 deg(Q 𝑗 ) deg(P) = 𝑚 𝑖 deg(P𝑖 ) deg(Q).
12-exo811) Let 𝑝 be a prime number; let E = F𝑝 .
a) Let 𝑛 > 2 be an integer prime to 𝑝 and let F be the extension of E generated by a
primitive 𝑛th root of unity 𝛼. Prove that [F : E] is the order of 𝑝 in Z/𝑛Z.
 EXERCISES 221

b) Describe the action of Gal(F/E) on the set {𝛼, 𝛼 2 , . . . , 𝛼 𝑛−1 }. Give a simple
condition of this action to factor through the alternate group on 𝑛 − 1 elements. (How
does the Frobenius morphism 𝜑 act on this set? What is his signature?)
 
𝑎
c) Assume in the sequel that 𝑝 is odd. For 𝑎 ∈ Z, one sets 𝑝 = 0 if 𝑎 is a multiple
   
𝑎 𝑎
of 𝑝, 𝑝 = 1 if 𝑎 is a non-zero square modulo 𝑝, and 𝑝 = −1 otherwise (Legendre
 
𝑎
symbol). Prove that 𝑝 ≡ 𝑎 (𝑝−1)/2 (mod 𝑝).
d) Let
Ö 𝜑(𝛼 𝑗 ) − 𝜑(𝛼 𝑖 )
𝛿= .
16𝑖<𝑗6𝑛−1
𝛼 𝑗 − 𝛼𝑖

Prove that 𝛿 = ((−1)𝑛(𝑛−1)/2 𝑛 𝑛−2 )(𝑝−1)/2 = ±1.

𝑝 𝑞
  
e) One assumes moreover that 𝑛 = 𝑞 is an odd prime number. Prove that 𝑞 𝑝 =
(−1)(𝑝−1)(𝑞−1)/4 (Gauss’s quadratic reciprocity law).
13-exo1214) Let K be a field and let 𝑛 > 2 be an integer. Let E = K[𝛼] be an extension
of K generated by a primitive 𝑛th root of unity 𝛼. (In particular, the characteristic of K
does not divide 𝑛.)
a) Prove that the set C of 𝑛th roots of unity in E is a cyclic group of order 𝑛, generated
by 𝛼. Prove that E is a Galois extension of K.
b) Let 𝜎 ∈ Gal(E/K). Prove that there exists an integer 𝑑 prime to 𝑛 such that
𝜎(𝛼) = 𝛼 𝑑 . Prove that 𝜎(𝑥) = 𝑥 𝑑 for every 𝑥 ∈ C.
c) Construct that there exists a ring morphism 𝜑 : Gal(E/K) → (Z/𝑛Z)× such that
𝜑(𝜎) is the class of 𝑑 if 𝜎(𝛼) = 𝛼 𝑑 . Prove that 𝜑 is injective; in particular, Gal(E/K) is
an abelian group.
14-exo1215) Let K be a field and let E = K(T) be the field of rational functions in one
indeterminate T with coefficients in K.
a) Prove that there exists a unique K-automorphism of E, 𝛼 (resp. 𝛽), such that
𝛼(T) = 1/T (resp. 𝛽(T) = 1 − T).
b) Let G be the subgroup of Gal(E/K) generated by 𝛼 and 𝛽. Prove that it is isomorphic
to the symmetric group 𝔖3 .
c) Let F = EG be the subfield of E consisting of rational functions P ∈ K(T) such that
𝛼(P) = 𝛽(P) = P. Prove that F contains
(T2 − T + 1)3
𝑓 = .
T2 (T − 1)2
d) Prove that F = K( 𝑓 ).
15-exo1206) In this exercise, we prove that the field C of complex number is algebraically
closed. In fact, the proof will show a bit more. So let R be a field and let C = R(𝑖) be
the field obtained by adjoining an element 𝑖 such that 𝑖 2 = −1. We make the following
assumptions:
(i) The characteristic of R is zero;
222 CHAPTER 4. FIELD EXTENSIONS

(ii) Every polynomial of odd degree in R[T] has a root in R;

(iii) Every element of C is a square.
a) Prove that the field R of real numbers satisfies these assumptions.
b) Let E be a non-trivial finite Galois extension of R containing C. Let G = Gal(E/R)
be its Galois group. Prove that the order of G is a power of 2. (Let P be a 2-Sylow
subgroup of G. Prove that EP is an extension of R of odd degree. Conclude that EP = R.)
c) Let G1 = Gal(E/C). Prove that G1 = {1} by introducing a subgroup H of index 2
in G1 , and considering the quadratic extension EH of C. Conclude that E = C.
d) Prove that C is algebraically closed.
16-exo824) Let F be a finite field. Let 𝑞 = Card(F).
a) For every 𝑎 ∈ F× and 𝑏 ∈ F, prove that there exists a unique automorphism 𝜏 of
F(T) such that 𝜏(T) = 𝑎T + 𝑏.
b) Prove that the automorphisms of this form constitute a finite group G of cardinal-
ity 𝑞(𝑞 − 1). Let A be the subgroup of G of such automorphisms with 𝑏 = 0, and B be
the subgroup of G of such automorphisms with 𝑎 = 1.
c) Show that F(T)A = F(T𝑞−1 ) and that F(T)B = F(T𝑞 − T).
d) Describe the intermediate subfields between F(T)A and F(T).
e) Compute the subfield F(T)G . Are the extensions F(T)G → F(T)A and F(T)G → F(T)B
Galois ?
17-exo828) Let E be a field, let P ∈ E[T] be a monic polynomial, let 𝑛 = deg(P), and let
E → F be a splitting extension of P. Let R be the set of roots of P in F. Let G = Gal(E/F).
a) Prove that the action of G on F induces an action of G on R.
b) What are the orbits of this action? In particular, prove that this action is transitive
if and only if P is irreducible.
c) Let 𝑎 ∈ R, and let H be the set of 𝜎 ∈ G such that 𝜎(𝑎) = 𝑎. Prove that H is a
subgroup of G. What is its index.
d) If 𝑔 ∈ G, what is the fixed subfield of the subgroup 𝑔H𝑔 −1 ? If P is irreducible,
prove that 𝑔∈G 𝑔H𝑔 −1 = {idF }.
Ñ

e) Let E → F be a finite Galois extension and let G = Gal(F/E). Prove that it is the
splitting extension of an irreducible polynomial P ∈ E[T] of degree 𝑛 if and only if there
exists a subgroup H ⊂ G of index 𝑛 such that 𝑔∈G 𝑔H𝑔 −1 = {idF }. What happens in
Ñ
the case where G is abelian?
18-exo1218) For 𝑛 > 1, let Φ𝑛 ∈ C[X] be the unique monic polynomials whose roots
are simple, equal to the primitive 𝑛th roots of unity in C.
Φ𝑑 = X𝑛 − 1. Prove by induction that Φ𝑛 ∈ Z[X] for all 𝑛.
Î
a) Prove that 𝑑|𝑛
b) If 𝑝 is a prime number, compute Φ𝑝 (X). Prove that there are integers 𝑎 1 ,. . . , 𝑎 𝑝−1
such that Φ𝑝 (1 + X) = X𝑝−1 + 𝑝𝑎1 X𝑝−2 + · · · + 𝑝𝑎 𝑝−1 , and 𝑎 𝑝−1 = 1. Using Eisenstein’s
criterion (see exercice 2/27), prove that Φ𝑝 is irreducible in Q[X].
 EXERCISES 223

c) Let 𝑛 > 2. In the rest of the exercise, we will prove that Φ𝑛 is irreducible. Let
𝛼 ∈ C be a primitive 𝑛th root of unity and let P be its minimal polynomial. Prove that
P ∈ Z[X] and that P divides Φ𝑛 in Z[X].
Let 𝑝 be a prime number that does not divide 𝑛. Prove that there exists 𝑏 ∈ Z[𝛼]
such that P(𝛼 𝑝 ) = 𝑝𝑏.
d) Prove that 𝛼 𝑝 is a primitive 𝑛th root of unity. If P(𝛼 𝑝 ) ≠ 0, prove by differentiating
the polynomial X𝑛 − 1 that 𝑛𝛼 𝑝(𝑛−1) ∈ 𝑝Z[𝛼]. Derive a contradiction, hence that
P(𝛼 𝑝 ) = 0 for every prime number 𝑝 which does not divide 𝑛.
e) Prove that Φ𝑛 is irreducible in Q[X].
19-exo618) In this exercise, we give a proof of the following theorem of Wedderburn:
Any finite division ring is a field. Let F be a finite division ring.
a) Show that any commutative subring of F is a field.
b) Let Z be the center of F. Show that Z is a subring of F. Let 𝑞 be its cardinality.
Show that there exists an integer 𝑛 > 1 such that Card F = 𝑞 𝑛 .
c) Let 𝑥 ∈ F. Show that the set C𝑥 of all 𝑎 ∈ F such that 𝑎𝑥 = 𝑥𝑎 is a subring of F.
Show that there exists an integer 𝑛 𝑥 dividing 𝑛 such that Card C𝑥 = 𝑞 𝑛 𝑥 . (Observe
that left multiplication by elements of C𝑥 endowes F with the structure of a C𝑥 -vector
space.)
d) Let 𝑥 ∈ F× ; compute in terms of 𝑛 𝑥 the cardinality of the conjugacy class 𝒞(𝑥) of 𝑥
in F× (by which we mean the set of all elements of F× of the form 𝑎𝑥𝑎 −1 , for 𝑎 ∈ F× ).
e) If 𝑥 ∉ Z, show that the cardinality of 𝒞(𝑥) is a multiple of Φ𝑛 (𝑞). (We write Φ𝑛 for
the 𝑛th cyclotomic polynomial, defined in exercise 4/18.)
f ) Using the class equation, show that Φ𝑛 (𝑞) divides 𝑞 𝑛 − 𝑞. Conclude that 𝑛 = 1,
hence that F is a field.
20-exo834) For every integer 𝑛 > 1, we denote by K𝑛 the subfield of C generated by
the 𝑛th roots of unity. We also write 𝜌𝑛 : Gal(K𝑛 /Q) → (Z/𝑛Z)× for the morphism of
groups characterized by 𝜎(𝜔) = 𝜔 𝜌(𝜎) for every 𝜎 ∈ Gal(K/Q) and every 𝑛th root of
unity 𝜔.
a) Let 𝑚 and 𝑛 be integers > 1 and let 𝑑 = gcd(𝑚, 𝑛). Prove that K𝑚 ∩ K𝑛 = K𝑑 .
b) Let 𝑝 be a prime number and let 𝑒 > 1 be an integer. Compute the cyclotomic
polynomial Φ𝑝 𝑒 . Apply the Eisenstein criterion to prove that it is irreducible in Q[T].
Conclude that [K𝑝 𝑒 : Q] = 𝜑(𝑝 𝑒 ) = 𝑝 𝑒−1 (𝑝 − 1).
c) Prove that [K𝑛 : Q] = 𝜑(𝑛) for every integer 𝑛 > 1. Prove that the cyclotomic
polynomial Φ𝑛 is irreducible in Q[T].
21-exo829) Let E → F be a finite extension of fields.
a) Assume that this extension is separable. Prove that the set of subfields of F that
contain E is finite. (First treat the case where this extension is Galois.)
b) Let 𝑝 be a prime number, let 𝑘 be a field of characteristic 𝑝, let F = 𝑘(X, Y) and let
E = 𝑘(X𝑝 , Y𝑝 ). For every 𝑎 ∈ 𝑘, let F𝑎 = 𝑘(X𝑝 , Y𝑝 , X + 𝑎Y). Prove that F𝑎 is a subfield
of F that contains E. Prove that F𝑎 ∩ F𝑏 = E if 𝑎 ≠ 𝑏.
224 CHAPTER 4. FIELD EXTENSIONS

22-exo1219) Let K be an infinite field and let K → L be a finite Galois extension. We let
𝑑 = [L : K] and write 𝜎1 , . . . , 𝜎𝑑 for the elements of Gal(L/K).
a) Let (𝑒1 , . . . , 𝑒 𝑑 ) be a family of elements in L. Prove that the determinant of the
matrix (𝜎𝑖 (𝑒 𝑗 ))16𝑖,𝑗6𝑑 is non-zero if and only if (𝑒1 , . . . , 𝑒 𝑑 ) is a basis of L as a K-vector
space.
In the following, we fix such a basis (𝑒1 , . . . , 𝑒 𝑑 ). Let P ∈ L[X1 , . . . , X𝑑 ] be such that
P(𝜎1 (𝑥), . . . , 𝜎𝑑 (𝑥)) = 0
for every 𝑥 ∈ L.
b) Let Q ∈ L[X1 , . . . , X𝑑 ] be the polynomial
𝑑
Õ 𝑑
Õ
Q = P( 𝜎1 (𝑒 𝑖 )X𝑖 , . . . , 𝜎𝑑 (𝑒 𝑖 )).
𝑖=1 𝑖=1

Prove that Q(𝑥1 , . . . , 𝑥 𝑑 ) = 0 for every (𝑥1 , . . . , 𝑥 𝑑 ) ∈ K𝑑 . Prove that Q = 0.

c) Prove that P = 0 (algebraic independence of 𝜎1 , . . . , 𝜎𝑑 ).
d) Prove that there exists 𝜃 ∈ L such that (𝜎1 (𝜃), . . . , 𝜎𝑑 (𝜃)) is a K-basis of L. (Such a
basis is called a normal basis of L over K.)
23-exo1217) Let F → E be a Galois extension of fields, with Galois group G = Gal(E/F).
a) Let 𝛼 ∈ E× and let 𝑐 : G → E× be the map defined by 𝑐(𝜎) = 𝛼/𝜎(𝛼) for every
𝜎 ∈ G. Prove that for every 𝜎 and 𝜏 in G, one has
𝑐(𝜎𝜏) = 𝑐(𝜎)𝜎(𝑐(𝜏)).
b) Conversely, let 𝑐 : G → E× by any map satisfying this relation. Using proposi-
tion 4.4.12, prove that there exists 𝑥 ∈ E such that
Õ
𝛼= 𝑐(𝜎)𝜎(𝑥) ≠ 0.
𝜎∈G

Conclude that one has 𝑐(𝜎) = 𝛼/𝜎(𝛼) for every 𝜎 ∈ G.

c) Let 𝜒 : G → F× be a group morphism. Prove that there exists 𝛼 ∈ E× such that
𝜒(𝜎) = 𝛼/𝜎(𝛼) for every 𝜎 ∈ G.
d) We assume that G is a cyclic group; let 𝜎 be a generator of G. Let 𝑥 ∈ F; prove
that NL/K (𝑥) = 1 if and only if there exists 𝛼 ∈ F× such that 𝑥 = 𝛼/𝜎(𝛼). Explicit the
particular cases where K → L is the extension R → C or an extension F𝑞 → F𝑞 𝑛 of
finite fields.
24-exo1220) Let K be a finite field, let 𝑝 be its characteristic and let 𝑞 be its cardinality.
a) For 𝑚 ∈ N, compute S𝑚 = 𝑥𝑚.
Í
𝑥∈K
b) For P ∈ K[X1 , . . . , X𝑛 ], we write S(P) = P(𝑥). If deg P < 𝑛(𝑞 − 1), prove that
Í
𝑥∈K𝑛
S(P) = 0.
Í𝑟
c) Let P1 , . . . , P𝑟 be polynomials in K[X1 , . . . , X𝑛 ] such that 𝑖=1 deg P 𝑖 < 𝑛. Let
𝑞−1 𝑞−1
P = (1 − P1 ) . . . (1 − P𝑟 ); prove that S(P) = 0.
 EXERCISES 225

Let V be the set of all 𝑥 ∈ K𝑛 such that P1 (𝑥) = · · · = P𝑛 (𝑥) = 0}. Prove that Card(V)
is divisible by 𝑝 (Chevalley–Warning’s theorem).
d) Let P ∈ K[X1 , . . . , X𝑛 ] be a homogeneous polynomial of degree 𝑑 > 0. If 𝑑 < 𝑛,
prove that there exists 𝑥 ∈ K𝑛 {0} such that P(𝑥) = 0.
25-exo624) This exercise is the basis of Berlekamp’s algorithm to factor polynomials
over finite field.
Let 𝑝 be a prime number and let P be a nonconstant separable polynomial with
coefficient in a finite field F𝑝 . Let P = 𝑐 𝑟𝑖=1 P𝑖 the factorization of P in irreducible
Î
monic polynomials, with 𝑐 ∈ F×𝑝 . Set 𝑛 𝑖 = deg P𝑖 . Let RP be the ring F𝑝 [X]/(P).
a) Show that the ring RP𝑖 is a finite field with 𝑝 𝑛 𝑖 elements.
b) For A ∈ RP , let 𝜌 𝑖 (A) be the remainder of an Euclidean division of A by P𝑖 . Show
that the map A ↦→ (𝜌1 (A), . . . , 𝜌𝑟 (A)) induces an isomorphism of rings RP ' 𝑟𝑖=1 RP𝑖 .
Î

c) For A ∈ RP , let 𝑡(A) = A𝑝 − A. Show that 𝑡 is a F𝑝 -linear endomorphism of RP

(viewed as a F𝑝 -vector space) which corresponds, by the preceding isomorphism, to
the mapping
𝑟 𝑟
𝑝 𝑝
Ö Ö
F 𝑝 𝑛𝑖 → F𝑝 𝑛𝑖 , (𝑎 1 , . . . , 𝑎 𝑟 ) ↦→ (𝑎 1 − 𝑎 1 , . . . , 𝑎 𝑟 − 𝑎 𝑟 ).
𝑖=1 𝑖=1

d) Show that the kernel of 𝑡 is a vector subspace of RP , of dimension 𝑟.

e) Let 𝑎 be any element in Ker(𝑡). Show that there exists a monic polynomial
Q ∈ F𝑝 [X], of minimal degree, such that Q(𝑎) = 0. Show that the polynomial Q is
separable and split over F𝑝 .
f ) (followed) If 𝑎 ∉ F𝑝 , show that Q is not irreducible. From a partial nontrvial
factorization Q = Q1 Q2 , show how to get a partial nontrivial factorization of P.
26-exo1222) Let K ⊂ L and L ⊂ M two finitely generated field extensions. Prove that
tr degK M = tr degK L + tr degL M.

27-exo1221) Let 𝑘 be a field, let K ⊂ 𝑘(T) be a subfield containing 𝑘, but distinct from 𝑘.
a) Prove that the extension K → 𝑘(T) is finite. Let 𝑛 = [𝑘(T) : K] be its degree.
b) Prove that the minimal polynomial of T over K takes the form
𝑓 (X) = X𝑛 + 𝑎 1 X𝑛−1 + · · · + 𝑎 𝑛 ,
where 𝑎 1 , . . . , 𝑎 𝑛 ∈ K and that there exists 𝑗 ∈ {1; . . . ; 𝑛} with 𝑎 𝑗 ∉ 𝑘.
c) Let 𝑗 be such an integer, and let 𝑢 = 𝑎 𝑗 ; let 𝑔, ℎ ∈ 𝑘[T] be two coprime polynomials
such that 𝑢 = 𝑔/ℎ; let 𝑚 = sup(deg 𝑔, deg ℎ). Prove that 𝑚 > 𝑛 and that there exists
𝑞 ∈ K[T] such that 𝑔(X) − 𝑢 ℎ(X) = 𝑞(X) 𝑓 (X).
d) Prove that there exist polynomials 𝑐0 , . . . , 𝑐 𝑛 ∈ 𝑘[T], such that 𝑐 𝑗 /𝑐 0 = 𝑎 𝑗 for all 𝑗
and gcd(𝑐 0 , . . . , 𝑐 𝑛 ) = 1.
e) One sets 𝑓 (X, T) = 𝑐 0 (T)X𝑛 +· · ·+𝑐 𝑛 (T) = 𝑐 0 (T) 𝑓 (X). Prove that 𝑓 (X, T) is irreducible
in 𝑘[X, T].
226 CHAPTER 4. FIELD EXTENSIONS

f ) Prove that there exists 𝑞 ∈ 𝑘[X, T] such that

𝑔(T)ℎ(X) − 𝑔(X)ℎ(T) = 𝑞(X, T) 𝑓 (X, T).
Conclude that 𝑚 = 𝑛, and K = 𝑘(𝑢) (Lüroth’s theorem).
CHAPTER 5

MODULES OVER PRINCIPAL IDEAL

RINGS

One result that greatly simplifies undergraduate linear algebra is that vector
spaces over a field have a basis. This allows to explicit computations in
coordinates, as well as to a representation of linear maps by matrices. Over
a ring which is not a field, there exist modules which are not free, and the
classification of modules over general rings is much more delicate, if not
impossible.
In contrast, finitely generated modules over principal ideal rings have a nice
concise description using invariant factors. Explaining this description and
some of its applications is the main objective of this chapter.
There are several possibilities to establish this result and the approach I have
chosen combines algorithmic considerations related to matrix operations with
the more abstract theory of Fitting ideals of chapter 3.
Matrix operations are the classical elementary modifications of matrices,
permutation of rows, multiplication of a row by an invertible scalar, addition to
a row of a multiple of another one, and the analogous operations on columns.
Beyond their algorithmic quality, they furnish important results about matrix
groups. Here again, the case of fields gives rise to the easier reduced echelon
forms, and I present it first, as well as its consequences for classical linear
algebra. I then pass to the Hermite and Smith normal forms for matrices
with coefficients first in a euclidean ring, and then in a principal ideal domain,
and explain the translation in terms of of finitely modules over a principal
ideal domain. I end this chapter by explaining the two classical particular
cases: when the principal ideal domain is the rings of integers, we obtain a
classification of finitely generated abelian groups; when it is the ring of
polynomials in one indeterminate over a field, we obtain a classification (up to
228 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

conjugacy) of endomorphisms of a finite dimensional vector space, and

the Jordan decomposition.

5.1. Matrix operations

5.1.1. Elementary matrices. — Let A be a ring. The group GL𝑛 (A) of
𝑛 × 𝑛 invertible matrices with entries in A contains some important
elements.
We write (𝑒1 , . . . , 𝑒 𝑛 ) for the canonical basis of A𝑛 and (𝑒 𝑖,𝑗 ) for the
canonical basis of M𝑛 (A); that is, 𝑒 𝑖,𝑗 is the matrix of whose all entries
are 0, except the one on row 𝑖 and column 𝑗, which equals 1.
For 𝑖, 𝑗 ∈ {1, . . . , 𝑛}, 𝑖 ≠ 𝑗, and 𝑎 ∈ A, set E𝑖𝑗 (𝑎) = I𝑛 + 𝑎𝑒 𝑖,𝑗 , where, we
recall, I𝑛 is the identity matrix. We have the relation
E𝑖𝑗 (𝑎)E𝑖𝑗 (𝑏) = E𝑖𝑗 (𝑎 + 𝑏)
for any 𝑎, 𝑏 ∈ A. Together with the obvious equality E𝑖𝑗 (0) = I𝑛 , it implies
that the matrices E𝑖,𝑗 (𝑎) are invertible and that the map 𝑎 ↦→ E𝑖,𝑗 (𝑎) is a
morphism of groups from the additive group of A to the group GL𝑛 (A).
We say that these matrices E𝑖,𝑗 (𝑎) (for 𝑖 ≠ 𝑗 and 𝑎 ∈ A) are elementary and
write E𝑛 (A) for the subgroup they generate in GL𝑛 (A)
For 𝜎 ∈ 𝔖𝑛 , we let P𝜎 be the matrix of the linear map which sends, for
every 𝑗, the vector 𝑒 𝑗 to the vector 𝑒 𝜎(𝑗) . Explicitly, if P𝜎 = (𝑝 𝑖,𝑗 ), one has
𝑝 𝑖,𝑗 = 1 if 𝑖 = 𝜎(𝑗) and 𝑝 𝑖,𝑗 = 0 otherwise. For all permutations 𝜎, 𝜏 ∈ 𝔖𝑛 ,
one has P𝜎𝜏 = P𝜎 P𝜏 and Pid = I𝑛 . Consequently, the map 𝜎 ↦→ P𝜎 is an
isomorphism of groups from 𝔖𝑛 to a subgroup of GL𝑛 (A) which we
denote by W. The group W is called the Weyl group of GL𝑛 ; its elements
are called permutation matrices.
Finally, for 1 6 𝑗 6 𝑛 and 𝑎 ∈ A, let D 𝑗 (𝑎) be the diagonal matrix
I𝑛 + (𝑎 − 1)𝑒 𝑗,𝑗 . The entries of D 𝑗 (𝑎) off the diagonal are zero, those on
the diagonal are 1 except for the entry on row 𝑗 and column 𝑗 which
equals 𝑎. For any 𝑎, 𝑏 ∈ A, one has D 𝑗 (𝑎)D 𝑗 (𝑏) = D 𝑗 (𝑎𝑏) and D 𝑗 (1) = I𝑛 ;
if 𝑎 ∈ A∗ , then D 𝑗 (𝑎) belongs to GL𝑛 (A).
Let GE𝑛 (A) be the subgroup of GL𝑛 (A) generated by the elementary
matrices E𝑖,𝑗 (𝑎), for 𝑖 ≠ 𝑗 and 𝑎 ∈ A, the permutation matrices P𝜎 , for
𝜎 ∈ 𝔖𝑛 , and the diagonal matrices D 𝑗 (𝑎), for 𝑖 ∈ {1, . . . , 𝑛} and 𝑎 ∈ A∗ .
One has E𝑛 (A) ⊂ GE𝑛 (A), and the inclusion is strict in general.
 5.1. MATRIX OPERATIONS 229

5.1.2. Elementary operations. — Let M ∈ M𝑛,𝑝 (A) be a matrix with 𝑛

rows and 𝑝 columns with entries in A.
Multiplying M on the right by elementary matrices from M𝑝 (A)
corresponds to the classical operations on the columns of M. Indeed,
the matrix ME𝑖,𝑗 (𝑎) is obtained by adding to the 𝑗th column of M its
𝑖th column multiplied by 𝑎, operation that we represent by writing
C 𝑗 ← C 𝑗 + C𝑖 𝑎. The matrix MP𝜎 is obtained by permuting the columns
of M: the 𝑗th column of MP𝜎 is the 𝜎(𝑗)th column of M. Finally, the
matrix MD 𝑗 (𝑎) is obtained by multiplying the 𝑗th column of M by 𝑎 (we
write C 𝑗 ← C 𝑗 𝑎).
Similarly, multiplying M on the left by elementary matrices from M𝑛 (A)
amounts to performing classical operations on the rows of M. The
matrix E𝑖,𝑗 (𝑎)M is obtained by adding 𝑎 times the 𝑗th row of M to its 𝑖th
row (we write R𝑖 ← R𝑖 + 𝑎R 𝑗 ); the 𝑖th row of M is the row of index 𝜎(𝑖)
of the matrix P𝜎 M; the rows of D𝑖 (𝑎) are those of M, the row of index 𝑖
being multiplied by 𝑎 on the left (in symbols, R𝑖 ← 𝑎R𝑖 ).

5.1.3. Row and column equivalences. — One says that two matrices
M, M0 ∈ M𝑛,𝑝 (A) are row equivalent if there exists a matrix P ∈ GE𝑛 (A)
such that M0 = PM. This means precisely that there exists a sequence of
elementary row operations that transforms M into M0. Row equivalence
is an equivalence relation; its equivalence classes are the orbits of the
group GE𝑛 (A), acting on M𝑛,𝑝 (A) by left multiplication.
One says that two matrices M, M0 ∈ M𝑛,𝑝 (A) are column equivalent if
there exists a matrix Q ∈ GE𝑝 (K) such that M0 = MQ. This amounts
to saying that one can pass from M to M0 by a series of elementary
column operations. Column equivalence is an equivalence relation, its
equivalence classes are the orbits of the group GE𝑝 (A) acting by right
multiplication on M𝑛,𝑝 (A).

5.1.4. Reduced row echelon forms. — One says that a matrix M =

(𝑚 𝑖,𝑗 ) ∈ M𝑛,𝑝 (A) is in reduced row echelon form if there exist an inte-
ger 𝑟 ∈ {1, . . . , inf(𝑛, 𝑝)} and integers 𝑗1 , . . . , 𝑗𝑟 such that the following
conditions hold:
(i) One has 1 6 𝑗1 < 𝑗2 < · · · < 𝑗𝑟 6 𝑝;
230 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

(ii) For any 𝑖 ∈ {1, . . . , 𝑟} and any 𝑗 such that 1 6 𝑗 < 𝑗𝑖 , one has
𝑚 𝑖,𝑗 = 0;
(iii) For any 𝑖 ∈ {1, . . . , 𝑟}, one has 𝑚 𝑖,𝑗𝑖 = 1, and 𝑚 𝑘,𝑗𝑖 = 0 for any other
index 𝑘;
(iv) If 𝑖 ∈ {𝑟 + 1, . . . , 𝑛} and 𝑗 ∈ {1, . . . , 𝑝}, then 𝑚 𝑖,𝑗 = 0.
The entries 𝑚 𝑖,𝑗𝑖 = 1 are called the pivots of the matrix M, the inte-
gers 𝑗1 , . . . , 𝑗𝑟 are called the pivot column indices; the integer 𝑟 is called
the row rank of M. In this language, the above conditions thus say that
the pivot column indices are strictly increasing (condition (i)), the first
nonzero entry of each of the first 𝑟 rows is a pivot (condition (ii)), all
entries of a pivot column other than the pivot itself are 0 (condition (iii)),
and the rows of index > 𝑟 are zero (condition (iv)).
We mention the important following observation as a lemma.

Lemma (5.1.5). — Let M ∈ M𝑛,𝑝 (A) be a matrix in reduced row echelon form.
For any integer 𝑘, the matrix deduced from M by taking its first 𝑘 columns is
still in reduced row echelon form.

5.1.6. Reduced column echelon form. — One says that a matrix M ∈

M𝑛,𝑝 (A) is in reduced column echelon form if its transpose matrix Mt is
in reduced row echelon form. Explicitly, this means that there exist
an integer 𝑠 ∈ {1, . . . , inf(𝑛, 𝑝)} and integers 𝑖1 , . . . , 𝑖 𝑠 such that the
following conditions hold:
(i) One has 1 6 𝑖1 < 𝑖2 < · · · < 𝑖 𝑠 6 𝑛;
(ii) For any 𝑗 ∈ {1, . . . , 𝑠} and any integer 𝑖 such that 1 6 𝑖 < 𝑖 𝑗 , one
has 𝑚 𝑖,𝑗 = 0;
(iii) For any 𝑗 such that 1 6 𝑗 6 𝑠, one has 𝑚 𝑖 𝑗 ,𝑗 = 1 and 𝑚 𝑖 𝑗 ,𝑘 = 0 for
any other 𝑘 ≠ 𝑗;
(iv) If 𝑗 ∈ {𝑠 + 1, . . . , 𝑝} and 𝑖 ∈ {1, . . . , 𝑛}, then 𝑚 𝑖,𝑗 = 0.
In English: the first nonzero entry of each of the first 𝑠 columns is equal
to 1, called a pivot, the pivot row indices are increasing, all entries of a
pivot row other than the pivot itself are 0, and the columns of index > 𝑠
are zero. The integer 𝑠 is called the column rank of M.
 5.1. MATRIX OPERATIONS 231

5.1.7. Application to the resolution of linear systems. — Let M ∈

M𝑛,𝑝 (A).
If M is in reduced row echelon form, the analysis of the linear system
MX = 0, with unknown X ∈ A𝑝 is very easy. In fact, this system is
explicitly solved, the variables corresponding to pivot column indices
being expressed in terms of the other variables.
More generally, let us assume that there exists a matrix M0 in reduced
row echelon form which is row-equivalent to M. Since each row operation
can be reversed, it transforms the system MX = 0 into an equivalent
one. Consequently, the systems MX = 0 and M0X = 0 are equivalent.
Alternatively, there exists by assumption a matrix P ∈ GE𝑛 (A) such
that M0 = PM, and since GE𝑛 (A) ⊂ GL𝑛 (A), the condition MX = 0 is
equivalent to the condition M0X = 0. The original system MX = 0 is now
replaced by a system in reduced row echelon form.
This can also be applied to “inhomogeneous” systems of the form
MX = Y, where Y ∈ A𝑛 is given. Indeed, this system is equivalent to the
system MX + 𝑦Y = 0 in the unknown (X, 𝑦) ∈ A𝑝+1 to which we add the
condition 𝑦 = −1.
Let us assume that there exists a matrix [M0 Y0] ∈ M𝑛,𝑝+1 (A) in reduced
row echelon form which is row equivalent to the matrix [M Y]. The
system MX = Y is then equivalent to the system M0X = Y0. Two
possibilities arise. If the last column of [M0 Y0] is not that of a pivot, the
system M0X = Y0, the pivot variables being expressed in terms of the
other variables and the entries of Y0, and 𝑦 = −1 is one of these other
variables. Otherwise, when the last column of [M0Y0] is that of a pivot,
the system M0X = Y0 contains the equation 0 = 𝑦 which is inconsistent
with the condition 𝑦 = −1, so that the system M0X = Y0 has no solution.
In that case, the system MX = Y has no solution neither.
We can also reason directly on M. Let P ∈ GE𝑛 (A) be such that
M0 = PM is in reduced row echelon form, with row rank 𝑟 and pivot
column indices 𝑗1 , . . . , 𝑗𝑟 . The system MX = Y is equivalent to the system
M0X = PY. The last 𝑛 − 𝑟 rows of this system are of the form 0 = 𝑦 𝑖0,
where (𝑦10 , . . . , 𝑦𝑛0 ) are the entries of Y0 = PY. If we think of the entries
(𝑦1 , . . . , 𝑦𝑛 ) of the vector Y to be inderminates, these 𝑛 − 𝑟 equations
are linear conditions on the entries of Y which must be satisfied for the
232 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

system MX = Y to have a solution. When they hold, the first 𝑟 rows of

the system M0X = Y0 express the pivot variables in terms of the other
variables and of the entries of Y0 = PY.
This generalizes directly to conjunctions, that is, systems of the form
MX = Y1 , . . . , MX = Y𝑞 . Assume that there exists a matrix [M0 Y01 . . . Y0𝑞 ]
in reduced row echelon form which is row equivalent to [M Y1 . . . Y𝑞 ].
If all of its pivot indices are 6 𝑝, the system is solved, otherwise, it is
inconsistent.

5.2. Application to linear algebra

Let K be a division ring. In this subsection, I want to show how to
systematically apply row echelon forms to solve standard problems
of linear algebra. In fact, we will even recover all the standard and
fundamental results of linear algebra.
For any integer 𝑛, we consider K𝑛 as a right vector-space and we recall
that any matrix M ∈ M𝑛,𝑝 (K) defines a K-linear map X ↦→ AX from K𝑝
to K𝑛 .
We begin by proving that any matrix is row-equivalent to a unique
matrix in reduced row echelon form.
Proposition (5.2.1). — Let K be a division ring. For any matrix M ∈ M𝑛,𝑝 (K),
there exists exactly one matrix M0 which is in reduced row echelon form and
is row-equivalent to M. Similarly, for any matrix M ∈ M𝑛,𝑝 (K), there exists
one matrix M0, and only one, which is in reduced column echelon form and is
column-equivalent to M.
Proof. — By transposition, it is enough to show the assertion for row-
equivalence. The first part of the proof will show the existence of a matrix
equivalent to M which is in reduced row echelon form by explaining
an explicit method which reduces any matrix to a matrix of this form.
Technically, we show by induction on 𝑘 ∈ {0, . . . , 𝑝} that we can perform
elementary row operations on M so that the matrix M 𝑘 obtained by
extracting the first 𝑘 columns of M can be put in reduced row echelon
form.
There is nothing to show for 𝑘 = 0; so assume that 𝑘 > 1 and that
the assertion holds for 𝑘 − 1. Let A ∈ GE𝑛 (K) be such that AM 𝑘−1 is in
 5.2. APPLICATION TO LINEAR ALGEBRA 233

reduced row echelon form, with row rank 𝑟 and pivot column indices
𝑗1 , . . . , 𝑗𝑟 .
Let us then consider the matrix (𝑚 𝑖,𝑗 ) = AM 𝑘 in M𝑛,𝑘 (K). Its (𝑘 − 1)
first columns are those of AM 𝑘−1 .
If the entries 𝑚 𝑖,𝑘 for 𝑖 > 𝑟 in the last column of AM 𝑘 are all zero,
then AM 𝑘 is in reduced row echelon form, with row rank 𝑟 and pivot
indices 𝑗1 , . . . , 𝑗𝑟 . So the assertion holds for 𝑘 in this case.
Otherwise, let 𝑖 be the smallest integer > 𝑟 such that 𝑚 𝑖,𝑘 ≠ 0 and let us
swap the rows of indices 𝑖 and 𝑟. This allows to assume that 𝑚𝑟+1,𝑘 ≠ 0;
let us then divide the row of index 𝑟 + 1 by 𝑚𝑟+1,𝑘 ; we are now reduced
to the case where 𝑚𝑟+1,𝑘 = 1. Now, if, for every 𝑖 ∈ {1, . . . , 𝑛} such that
𝑖 ≠ 𝑟 + 1, we perform the operation R𝑖 ← R𝑖 − 𝑚 𝑖,𝑘 R𝑟+1 , we obtain a
matrix in reduced row echelon form whose row rank is 𝑟 + 1 and pivot
column indices are 𝑗1 , . . . , 𝑗𝑟 , 𝑘. This proves the assertion for 𝑘.
By induction, the assertion holds for 𝑘 = 𝑝, so that any 𝑛 × 𝑝-matrix is
row equivalent to a matrix in reduced row echelon form.
To establish the uniqueness of a matrix in reduced row echelon form
which is row equivalent to our original matrix, we shall make use of the
interpretation through linear systems.
We want to prove that there exists at most one matrix in reduced row
echelon form which is row equivalent to M. To that aim, it suffices to
show that if M and M0 are two matrices in reduced row echelon form
which are row equivalent, then M = M0.
As for the existence part of the proof, let us show by induction on
𝑘 ∈ {0, . . . , 𝑝} that the matrices M 𝑘 and M0𝑘 obtained by extracting the
𝑘 first columns from M and M0 are equal. This holds for 𝑘 = 0. So
assume that assumption for 𝑘 ∈ {0, . . . , 𝑝 − 1} and let us show it for 𝑘 + 1.
Let 𝑗1 , . . . , 𝑗𝑟 be the pivot column indices of M 𝑘 . Write Y and Y0 for the
(𝑘 + 1)th columns of M and M0, so that M 𝑘+1 = [M 𝑘 Y] and M0𝑘+1 = [M0𝑘 Y0].
We already know that M 𝑘 = M0𝑘 , so it remains to show that Y = Y0.
Since the matrices M 𝑘+1 and M0𝑘+1 are row-equivalent, the systems
M 𝑘 X = Y and M0𝑘 X = Y0 are equivalent. We now analyse these systems
making use of the fact that both matrices M 𝑘+1 and M0𝑘+1 are in reduced
row echelon form.
234 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

The systems M 𝑘 X = Y and M0𝑘 X = Y0 have the same solutions. If they

have one, say X, then Y = M 𝑘 X = M0𝑘 X = Y0 since M 𝑘 = M0𝑘 by the
induction hypothesis. Otherwise, both systems are inconsistent, which
implies that 𝑘 + 1 is a pivot column index both of M 𝑘+1 and of M0𝑘+1 . By
definition of a matrix in reduced row echelon form, Y is the column
vector all of which entries are 0 except for the (𝑟 + 1) which equals 1, as
well as Y0, so that Y = Y0.
We have thus established the induction hypothesis for 𝑘 + 1. By
induction, M𝑝 = M0𝑝 , that is M = M0, as was to be shown. 

Definition (5.2.2). — Let K be a division ring and let M ∈ M𝑛,𝑝 (K). The row
rank of M and the pivot column indices of M are the row rank and the pivot
column indices of the unique matrix in reduced row echelon form which is row
equivalent to M. Similarly, the column rank and the pivot row indices of M
are the column rank and the pivot row indices of the unique matrix in reduced
column echelon form which is column equivalent to M.

Let K be a division ring and let M ∈ M𝑛,𝑝 (K). Let 𝑟 be its row rank and
𝑗1 , . . . , 𝑗𝑟 be its pivot column indices.
The linear system MX = 0 has X = 0 for unique solution if and only
if all variables are pivot variables, that is if and only if 𝑟 = 𝑝 and 𝑗𝑖 = 𝑖
for 1 6 𝑖 6 𝑝. This implies in particular that 𝑛 > 𝑝. We therefore have
proved that a linear system which has (strictly) more unknowns than equations
has a nonzero solution. Equivalently, if 𝑝 > 𝑛, a linear map from K𝑝 to K𝑛
is not injective.
Let now Y ∈ K𝑛 and let [M0 Y0] be the matrix in reduced row echelon
form which is row equivalent to [M Y]. Observe that the matrix M0 is in
reduced row echelon form and is row equivalent to M. Consequently,
the pivot column indices of [M Y] are either 𝑗1 , . . . , 𝑗𝑟 , or 𝑗1 , . . . , 𝑗𝑟 , 𝑝 + 1.
In the first case, the system MX = Y is equivalent to the solved system
M0X = Y0, hence has a solution; in the second case, it has no solution.
Let us show how this implies that for 𝑝 < 𝑛, a linear map from K𝑝
to K𝑛 is not surjective. Indeed, assuming 𝑝 < 𝑛, we have 𝑝 + 1 6 𝑛 and
we may find vectors Y ∈ K𝑝 such that the pivot column indices of the
matrix [M Y] are 𝑗1 , . . . , 𝑗𝑟 , 𝑝 + 1; it suffices to choose A ∈ GE𝑛 (K) such
that M0 = AM and set Y = A−1 Y0, where Y0 = (0, . . . , 0, 1). In fact, one
 5.2. APPLICATION TO LINEAR ALGEBRA 235

may find such vectors as soon as 𝑟 < 𝑛. So what we have proved is: if the
row rank of M is 𝑟 < 𝑛, then the linear map from K𝑝 to K𝑛 is not surjective.
When 𝑛 = 𝑝, the above discussion implies the following proposition.

Proposition (5.2.3). — Let K be a division ring.

For any matrix M ∈ M𝑛 (K), the following seven conditions are equivalent:
(i) The endomorphism of K𝑛 defined by the matrix M is injective;
(ii) The row rank of M is equal to 𝑛;
(iii) The matrix M is row equivalent to the identity matrix I𝑛 ;
(iv) The matrix M belongs to GE𝑛 (K);
(v) The endomorphism of K𝑛 defined by M is bijective;
(vi) The endomorphism of K𝑛 defined by M is surjective;
(vii) The matrix M belongs to GL𝑛 (K).
In particular, GE𝑛 (K) = GL𝑛 (K): the linear group GL𝑛 (K) is generated by
the elementary matrices.

Proof. — Let M ∈ M𝑛 (K). Assuming (i), we have seen that 𝑟 = 𝑛, hence

(ii). If 𝑟 = 𝑛, the pivot column indices are 𝑗𝑖 = 𝑖 for 𝑖 ∈ {1, . . . , 𝑛}, so that
M is row equivalent to I𝑛 , and (iii) is proved. Observe that (iii) and (iv)
are equivalent, for M is row-equivalent to I𝑛 if and only if there exists
P ∈ GE𝑛 (K) such that M = PI𝑛 = P. Let us suppose (iii); the row rank
of M is equal to 𝑛 and the pivot indices of any matrix [M Y] can only be
1, . . . , 𝑛, implying that any system of the form MX = Y has exactly one
solution and (v) holds. Of course, (v) implies both (i) and (vi), and is
equivalent to (vii). Finally, if the endomorphism of K𝑛 defined by M is
surjective, we have seen that 𝑟 > 𝑛; since one always has 𝑟 6 𝑛, we get
𝑟 = 𝑛, so that all other properties hold. 

5.2.4. Computation of a basis. — Let K be a division ring, let Y1 , . . . , Y𝑝

be vectors of K𝑛 (viewed as a right vector space) and let V be the subspace
of K𝑛 generated by Y1 , . . . , Y𝑝 . A standard problem in linear algebra
consists in determining a basis of V, or a system of linear equations
defining V.
Let M be the matrix [Y1 . . . Y𝑝 ]. Observe that the solutions of the
system MX = 0 are exactly the coefficients of linear dependence relations
236 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

among the vectors Y 𝑗 . Imposing additional constraints of the form

𝑥 𝑗 = 0 for 𝑗 in a subset J of {1, . . . , 𝑝}, where X = (𝑥 1 , . . . , 𝑥 𝑝 ), amounts
to searching for the linear dependence relations among the vectors Y 𝑗 ,
for 𝑗 ∉ J. Let 𝑟 be the row rank of M and let 𝑗1 , . . . , 𝑗𝑟 be its pivot column
indices. I claim that the family (Y 𝑗1 , . . . , Y 𝑗𝑟 ) is a basis of V. Let us first
show that it is linearly independent. The system MX = 0 is solved,
expressing 𝑥 𝑗1 , . . . , 𝑥 𝑗𝑟 as a linear combination of the other entries of X.
If we search for relations among Y 𝑗1 , . . . , Y 𝑗𝑟 , this amounts to imposing
𝑥 𝑗 = 0 for 𝑗 ∉ {𝑗1 , . . . , 𝑗𝑟 }, which implies 𝑥 𝑗1 = · · · = 𝑥 𝑗𝑟 = 0. Let us then
observe that every vector Y 𝑘 is a linear combination of Y 𝑗1 , . . . , Y 𝑗𝑟 . To
prove that, we solve the system MX = 0 imposing moreover that 𝑥 𝑗 = 0
if 𝑗 ∉ {𝑗1 , . . . , 𝑗𝑟 , 𝑘} and 𝑥 𝑘 = −1. As above, the solution of the system
furnishes values for 𝑥 𝑗1 , . . . , 𝑥 𝑗𝑟 .
Let Y be a vector (𝑦1 , . . . , 𝑦𝑛 ) whose entries are indeterminates, let M0 be
in reduced row echelon form and row equivalent to M, let A ∈ GE𝑛 (K)
be such that M0 = AM and set Y0 = AY; the entries of Y0 are linear
combinations of the 𝑦 𝑖 . The (𝑛 − 𝑟) last rows of M0 are zero, so that the
conditions 𝑦 𝑖0 = 0, for 𝑟 + 1 6 𝑖 6 𝑛, are exactly the linear dependence
relations among 𝑦1 , . . . , 𝑦𝑛 that are satisfied if and only Y ∈ V. We
have thus obtained a system of (𝑛 − 𝑟) linear equations that defines
the subspace V in K𝑛 . In practice, we obtain [M0 Y0] by applying the
above algorithm for a reduced row echelon form to the matrix [M Y],
but stoping this algorithm once we reach the last column.

5.2.5. Dimension of a vector subspace. — Let V be a vector subspace

of K𝑛 . Let (Y1 , Y2 , . . . , Y𝑝 ) be a family of vectors in V chosen in such a
way that each of them is linearly independent from the preceding ones.
For each integer 𝑚, the vectors Y1 , . . . , Y𝑚 are linearly independent, the
row rank of the matrix [Y1 . . . Y𝑚 ] is equal to 𝑚, hence necessarily 𝑝 6 𝑛.
This allows to assume that 𝑝 is as large as possible. Since (Y1 , . . . , Y𝑝 )
cannot be extended without violating the linear independence condition,
every vector of V is a linear combination of Y1 , . . . , Y𝑝 . Then (Y1 , . . . , Y𝑝 )
is a basis of V.
The choice of such a basis (Y1 , . . . , Y𝑝 ) of V identifies it with K𝑝 . Since
a linear map from K𝑝 to K𝑟 can be bijective only if 𝑝 = 𝑟, this shows that
all bases of V have the same cardinality, which we call the dimension of V.
 5.3. HERMITE NORMAL FORM 237

5.2.6. The row rank equals the column rank. — It is also useful to
put a matrix into a reduced column echelon form. Let indeed M =
[Y1 . . . Y𝑝 ] ∈ M𝑛,𝑝 (K) and let M0 = [Y01 . . . Y0𝑝 ] ∈ M𝑛,𝑝 (K) be the unique
matrix in reduced column echelon form which is column equivalent
to M. In particular, the subspace V of K𝑛 generated by the Y𝑖 coincides
with the subspace V0 generated by the Y0𝑖 . So dim(V) = dim(V0). By
definition, dim(V) = 𝑟, the row rank of M. On the other hand, if 𝑠 is
the column rank of M, we see by direct computation that Y01 , . . . , Y0𝑠 are
linearly independent, while Y0𝑠+1 = · · · = Y0𝑝 = 0, so that (Y01 , . . . , Y0𝑠 )
is a basis of V0. (This is another way to obtain a basis of a subspace.)
In particular, 𝑠 = dim(V) = 𝑟: the row rank and the column rank of a
matrix coincide.
Let also 𝑖 1 , . . . , 𝑖 𝑠 be the pivot column indices of M. For 𝑖 ∈ {0, . . . , 𝑛}
and 𝑡 ∈ {1, . . . , 𝑠}, the intersection V ∩ {0} 𝑖 × K𝑛−𝑖 contains Y0𝑡 if and
only if 𝑖 < 𝑖 𝑡 . More generally, a linear combination of Y01 , . . . , Y0𝑠 belongs
to {0} 𝑖 ∩ K𝑛−𝑖 if and only if the coefficient of Y0𝑡 is 0 when 𝑖 𝑡 6 𝑖. This
shows that dim(V ∩ ({0} 𝑖 × K𝑛−𝑖 )) > 𝑠 − 𝑡 + 1 if and only if 𝑖 < 𝑖 𝑡 .

5.3. Hermite normal form

5.3.1. — Let A be a principal ideal domain. We choose, once for all, a
set 𝒫(A) of representatives of (A {0})/A× in A, that is, a privileged
generator of each nonzero ideal. We also choose, for every nonzero
𝑎 ∈ A, a set 𝒫(A, 𝑎) of representatives of the quotient ring A/(𝑎) in A.
Let us give three important examples:
a) Assume that A is a field. Then we take 𝒫(A) = {1} and 𝒫(A, 𝑎) =
{0} for every 𝑎 ∈ A {0}.
b) Assume that A = Z, so that Z× = {±1}. we take for 𝒫(Z) the
set of positive integers; for each nonzero 𝑛 ∈ Z, we set 𝒫(Z, 𝑛) =
{0, 1, . . . , |𝑛| − 1}.
c) Assume that K is a field and A = K[T]. One has A× = K× . We
set 𝒫(K[T]) to be the set of monic polynomials. Moreover, for every
nonzero polynomial P ∈ K[T], we define 𝒫(K[T], P) to be the set of
polynomials Q such that deg(Q) < deg(P).
238 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

5.3.2. Row Hermite form. — Let M = (𝑚 𝑖,𝑗 ) ∈ M𝑛,𝑝 (A). We say

that the matrix M is in row-Hermite form if there exist an integer
𝑟 ∈ {1, . . . , inf(𝑛, 𝑝)} and integers 𝑗1 , . . . , 𝑗𝑟 , such that the following
conditions hold:
(i) One has 1 6 𝑗1 < 𝑗2 < · · · < 𝑗𝑟 6 𝑝;
(ii) For every 𝑖 ∈ {1, . . . , 𝑟} and any 𝑗 such that 1 6 𝑗 < 𝑗𝑖 , one has
𝑚 𝑖,𝑗 = 0;
(iii) For every 𝑖 ∈ {1, . . . , 𝑟}, one has 𝑚 𝑖,𝑗𝑖 ∈ 𝒫(A), 𝑚 𝑘,𝑗 ∈ 𝒫(A, 𝑚 𝑖,𝑗𝑖 ) if
𝑘 < 𝑖, and 𝑚 𝑘,𝑗 = 0 if 𝑘 > 𝑖;
(iv) If 𝑖 ∈ {𝑟 + 1, . . . , 𝑛} and 𝑗 ∈ {1, . . . , 𝑝}, then 𝑚 𝑖,𝑗 = 0.
When A is a field (and the sets of representatives have been chosen
as above), we recover the definition of a matrix in reduced row echelon
form. The entries 𝑚 𝑖,𝑗𝑖 are thus called the pivots of the matrix M; the
integers 𝑗1 , . . . , 𝑗𝑟 are called the pivot column indices; the integer 𝑟 is
called the row rank of M.

Lemma (5.3.3). — Let M ∈ M𝑛,𝑝 (A) be a matrix in row Hermite form. For
every integer 𝑘, the matrix deduced from M by taking its first rows (resp. its
first columns) is still in row Hermite form.

Proposition (5.3.4). — Let A be a euclidean domain. For every matrix M ∈

M𝑛,𝑝 (A), there exists exactly one matrix M0 which is in row Hermite form and
which is row-equivalent to M.

Proof. — As for the reduced row echelon form (proposition 5.2.1), the
existence part of the proof is an algorithm describing a sequence of
elementary operations that transform a matrix M into a matrix in row
Hermite form. We show by induction on 𝑘 ∈ {0, . . . , 𝑝} that we can
perform elementary row operations on M so that the matrix M 𝑘 obtained
by extracting the first 𝑘 columns of M is in row Hermite form. We fix a
euclidean gauge 𝛿 on A.
There is nothing to prove for 𝑘 = 0; so assume that 𝑘 > 1 and that the
assertion holds for 𝑘 − 1. Let P ∈ GE𝑛 (A) be such that PM 𝑘−1 is in row
Hermite form, with row rank 𝑟 and pivot indices 𝑗1 , . . . , 𝑗𝑟 . Let us then
consider the matrix (𝑚 𝑖,𝑗 ) = PM 𝑘 in M𝑛,𝑘 (A). Its (𝑘 − 1) first columns are
those of PM 𝑘−1 .
 5.3. HERMITE NORMAL FORM 239

We will now perform row operations on the rows of index 𝑖 > 𝑟,

arguing by induction on the infimum of all 𝛿(𝑚 𝑖,𝑘 ), for 𝑖 > 𝑟 such that
𝑚 𝑖,𝑘 ≠ 0.
If the entries 𝑚 𝑖,𝑘 , for 𝑖 > 𝑟, of the last column of PM 𝑘 are all zero, then
PM 𝑘 is in row-Hermite form, with row rank 𝑟 and pivot indices 𝑗1 , . . . , 𝑗𝑟 ,
hence the assertion holds for 𝑘.
Otherwise, let 𝑖 be such that 𝑚 𝑖,𝑘 has the smallest gauge among the
nonzero entries 𝑚𝑟+1,𝑘 , . . . , 𝑚𝑛,𝑘 and let us swap rows with indices 𝑟 + 1
and 𝑖, reducing to the case 𝑖 = 𝑟 + 1.
Then, for every integer 𝑖 such that 𝑟 +1 < 𝑖 ≠ 𝑛, we consider a euclidean
division 𝑚 𝑖,𝑘 = 𝑢𝑚𝑟+1,𝑘 + 𝑣, where either 𝑣 = 0 or 𝛿(𝑣) < 𝛿(𝑚𝑟+1,𝑘 ), and
perform the row operation R𝑖 ← R𝑖 − 𝑢R𝑟+1 . The matrix we obtain is
row equivalent to M 𝑘 , and its entries 𝑚𝑟+2,𝑘 , . . . , 𝑚𝑛,𝑘 are either zero, or
their gauges are strictly smaller than that of 𝑚𝑟+1,𝑘 . By induction, we
may thus assume that 𝑚𝑟+2,𝑘 = · · · = 𝑚𝑛,𝑘 = 0.
We then divide row 𝑟 + 1 by the unique unit 𝑎 ∈ A× such that
𝑚𝑟+1,𝑘 /𝑎 ∈ 𝒫(A); we thus are reduced to the case where 𝑚𝑟+1,𝑘 ∈ 𝒫(A).
Now, for every integer 𝑖 such that 1 6 𝑖 6 𝑟, we consider a congruence
𝑚 𝑖,𝑘 = 𝑢𝑚𝑟+1,𝑘 + 𝑣, where 𝑣 ∈ 𝒫(A, 𝑚𝑟+1,𝑘 ), and perform the row
operation R𝑖 ← R𝑖 − 𝑢R𝑟+1 . The 𝑛 × 𝑘 matrix thus obtained is in row-
Hermite form, with row rank 𝑟 + 1 and pivot indices 𝑗1 , . . . , 𝑗𝑟 , 𝑘. This
proves the assertion for 𝑘.
By induction, the assertion holds for 𝑘 = 𝑝, so that any 𝑛 × 𝑝-matrix is
row equivalent to a matrix in row Hermite form.
To establish the uniqueness assertion, we need to prove the following
assertion: let M, M0 ∈ M𝑛,𝑝 (A) be two matrices in row Hermite form
which are row equivalent; then M = M0. We prove this assertion by
induction on 𝑝, the case 𝑝 = 0 being void, so we assume now that 𝑝 > 1.
Possibly exchanging M and M0, we assume the number 𝑟 of pivot
columns of M is greater or equal than that for M0. Deleting the last
𝑛 − 𝑟 rows of M and M0, we then assume that all rows of M are nonzero,
and write 𝑗1 , . . . , 𝑗𝑛 for the pivot columns.
There are now two cases, depending on whether 𝑗𝑛 = 𝑝 or 𝑗𝑛 < 𝑝.
First assume that 𝑗𝑛 < 𝑝. Then, we write M = [M1 𝑣] and M0 = [M01 𝑣 0],
where M1 , M01 ∈ M𝑛,𝑝−1 (A) and 𝑣, 𝑣 0 ∈ A𝑛 . We observe that M1 and M01
240 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

are in row Hermite form, and are row equivalent. By the induction
hypothesis, one thus has M1 = M01 . Moreover, M1 has 𝑛 pivots as well, so
that it has rank 𝑛, as a matrix with coefficients in the fraction field K of A.
Consequently, there exists 𝜉 ∈ K𝑝−1 such that M1 𝜉 = 𝑣. Let 𝑑 ∈ A {0}
be a common denominator of the entries of 𝜉, and let 𝑥 = 𝑑𝜉, so that
𝑥 ∈ A𝑝−1 ; then M1 𝑥 = 𝑑𝑣. The matrices M = (M1 𝑣) and M0 = (M1 𝑣 0)
being row equivalent, the systems M1 X + 𝑦V = 0 and M1 X + 𝑦V0 = 0 (in
unknowns X ∈ A𝑝−1 and 𝑦 ∈ A) are equivalent so that M1 𝑥 = 𝑑𝑣 0. Since
𝑑 ≠ 0, this implies 𝑣 = 𝑣 0 and M = M0.
Let us know treat the case where 𝑗𝑛 = 𝑝. Now we write M = M01 𝑣𝑑 and


   
M01 𝑣 0 M01
M0 M1


= . Again, the matrices
𝑢 0 𝑑0
and
0 are in row Hermite
𝑢0
form, and are row equivalent. By induction, they are equal, hence
M1 = M01 and 𝑢 0 = 0. Moreover, M1 has 𝑛 − 1 pivots, so that it has
rank 𝑛 − 1 as a matrix with coefficients in the fraction field K of A. As
above, we may find 𝑥 ∈ A𝑝−1 and 𝑎 ∈ A {0} such that M1 𝑥 = 𝑎𝑣. Let
𝑥 be the image of 𝑥 in (A/(𝑎𝑑))𝑝−1 and let 𝑎 be the class of 𝑎 in A/(𝑎𝑑),
so that 𝜉 = (𝑥, −𝑎) is a solution of the system M𝜉 = 0 in the unknown
𝜉 ∈ (A/(𝑎𝑑))𝑝 . By row equivalence, this system is equivalent to the
system M0𝜉 = 0, so that M1 𝑥 = 𝑎𝑣 0 in (A/(𝑎𝑑))𝑛−1 and 𝑑0 𝑎 = 0 in A/(𝑎𝑑).
This last relation implies that 𝑑 divides 𝑑0; by symmetry, 𝑑0 divides 𝑑 as
well, so that 𝑑 and 𝑑0 are associates. By definition of the row Hermite
form, one has 𝑑, 𝑑0 ∈ 𝒫(A), hence 𝑑 = 𝑑0.
The first relation implies that 𝑎(𝑣 − 𝑣 0) ≡ 0 in (A/(𝑎𝑑))𝑛−1 , so that 𝑣 ≡ 𝑣 0
in (A/(𝑑))𝑛−1 . By definition of the row Hermite form again, this implies
that 𝑣 = 𝑣 0 and concludes the proof by induction. 

Theorem (5.3.5). — Let A be a euclidean ring and let M ∈ M𝑛,𝑝 (A). There
exist matrices P ∈ E𝑛 (A), Q ∈ E𝑝 (A) and a “diagonal matrix” D (meaning
𝑑 𝑖𝑗 = 0 for 𝑖 ≠ 𝑗) such that 𝑑 𝑖,𝑖 divides 𝑑 𝑖+1,𝑖+1 for any integer 𝑖 satisfying
1 6 𝑖 < inf(𝑛, 𝑝) so that M = PDQ. Moreover, the diagonal entries 𝑑1,1 , . . .
of D are well defined up to units.

Proof. — Let us write 𝛿 for the gauge of A and let us show the existence of
such a decomposition M = PDQ by a double induction, first on sup(𝑛, 𝑝)
and then on the minimum value of the gauge 𝛿 at nonzero entries of A.
 5.3. HERMITE NORMAL FORM 241

We first observe that each of the following 2 × 2 matrices can be

deduced from the preceding by some elementary row operation:
       
1 0 R1 ←R1 +R2 1 1 R2 ←R2 −R1 1 1 R1 ←R1 +R2 0 1
−−−−−−−−→ −−−−−−−−→ −−−−−−−−→ .
0 1 0 1 −1 0 −1 0

0 1 belongs to E (Z). Let 𝑖 and 𝑗 be distinct

Consequently, the matrix −1 0 2
integers in {1, . . . , 𝑛}. Performing similar operations on the rows of
indices 𝑖 and 𝑗, we see that there exists, for any transposition (𝑖, 𝑗) ∈ 𝔖𝑛 ,
an element of E𝑛 (A) which exchanges up to sign the 𝑖th and the 𝑗th
vector of the canonical basis of A𝑛 and leaves the other fixed.
Now, let (𝑖, 𝑗) be the coordinates of some nonzero coefficient of M with
minimal gauge. By what precedes, we may apply elementary operations
first on rows 1 and 𝑖, then on columns 1 and 𝑗, of M so as to assume
that this coefficient is in position (1, 1). Then let 𝑚1𝑘 = 𝑚11 𝑞 𝑘 + 𝑚1𝑘 0

by the euclidean division of 𝑚1𝑘 by 𝑚11 . The column operation C 𝑘 ←

C 𝑘 − C1 𝑞 𝑘 transforms the matrix M into the matrix M0 = M E1𝑘 (−𝑞 𝑘 )
whose entry 𝑚1𝑘 is now 𝑚1𝑘 0
. If 𝑚1𝑘 0
≠ 0, we have 𝛿(𝑚1𝑘 0
) < 0 and we
conclude by induction, since the minimal value of the gauge function at
nonzero entries of M has decreased. So we may assume that on the first
row, only the first entry is nonzero.
By similar operations on rows, we may also assume by induction that
all entries of the first column, except the first one, are zero.
Assume that some entry 𝑚 𝑖,𝑗 of M, with 𝑖 > 1 and 𝑗 > 1 is not divisible
by 𝑚11 . Then, let us perform the row operation R1 ← R1 + R𝑖 (which
amounts to left multiply M by E1𝑖 (1)); this transforms the first row of M
into the row (𝑚1,1 , 𝑚 𝑖,2 , . . . , 𝑚 𝑖,𝑝 ). Let 𝑚1,𝑗 = 𝑚1,1 𝑞 + 𝑟 be a euclidean
division of 𝑚1,𝑗 by 𝑚1,1 . The column operation C 𝑗 ← C 𝑗 − C1 𝑞 transforms
the matrix M into a new matrix whose (1, 𝑗) entry is 𝑟. By hypothesis,
𝑟 ≠ 0 and 𝛿(𝑟) < 𝛿(𝑚1,1 ). By induction, this matrix may transformed
into a matrix of the form

©𝑚11 0 . . . 0ª
­ 0 ®
­ . ®,
­ .. 𝑚11 M ®®
0
­
« 0 ¬
242 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

where M0 ∈ M𝑛−1,𝑝−1 (A). By induction, there exist matrices P0 ∈ E𝑛−1 (A),

Q0 ∈ E𝑝−1 (A) and a diagonal matrix D0 ∈ M𝑛−1,𝑝−1 (A), each of whose
diagonal coefficient divides the next one, such that M0 = P0D0Q0. Let us
then define the following block matrices:
     
1 0 1 0 1 0
P= , D = 𝑚11 , Q= .
0 P0 0 D0 0 Q0

We have P1 MQ1 = PDQ, hence M = (P1 )−1 PDQ(Q1 )−1 , which shows that
M has a decomposition of the desired form.
The uniqueness assumption follows from the theory of Fitting ideals.
With the notation of section 3.9, one has Δ𝑝 (M) = Δ𝑝 (D) = (𝑑1,1 · · · 𝑑 𝑝,𝑝 ) =
(𝑑 𝑝,𝑝 )Δ𝑝−1 (D) for every integer 𝑝. If Δ𝑝−1 (D) ≠ 0, this characterizes 𝑑 𝑝,𝑝
up to a unit; if Δ𝑝−1 (D) = 0, the divisibility assumption on 𝑑1,1 , . . .
implies that 𝑑 𝑝,𝑝 = 0. 

Corollary (5.3.6). — Let A be a euclidean ring. One has E𝑛 (A) = SL𝑛 (A) and
GE𝑛 (A) = GL𝑛 (A).

Proof. — The two inclusions E𝑛 (A) ⊂ SL𝑛 (A) and GE𝑛 (A) ⊂ GL𝑛 (A) are
obvious (all generators of E𝑛 (A) have determinant 1). We need to prove
that any matrix M ∈ GL𝑛 (A) belongs to GE𝑛 (A), and that any matrix
M ∈ SL𝑛 (A) belongs to E𝑛 (A).
So, let M ∈ GL𝑛 (A); let M = PDQ be some decomposition with
P, Q ∈ E𝑛 (A) and D ∈ M𝑛 (A) a diagonal matrix. Observe that the
diagonal entries 𝑑 𝑖,𝑖 of D are invertible. Consequently,

D = D1 (𝑑1,1 ) . . . D𝑛 (𝑑𝑛,𝑛 )

belongs to GE𝑛 (A) and M ∈ GE𝑛 (A) as well. It now follows from
lemma 5.3.7 below that there exist matrices P0 , Q0 ∈ E𝑛 (A), and a
unit 𝑎 ∈ A× such that M = P0 diag(𝑎, 1, . . . , 1)Q0.
Assume, moreover, that M ∈ SL𝑛 (A), we obtain that 𝑎 = det(M) = 1.
It follows that M = P0Q0, hence M ∈ E𝑛 (A). 
 5.3. HERMITE NORMAL FORM 243

Lemma (5.3.7). — Let A be a ring and let 𝑎1 , . . . , 𝑎 𝑛 be units in A; let

𝑎 0 = 𝑎 𝑛 𝑎 𝑛−1 . . . 𝑎1 . There exist matrices U, V ∈ E𝑛 (A) such that

© 𝑎1 © 𝑎0
𝑎
ª ª
­ ® ­ 1 ®
... ®.
2
U ­­ ... V =
® ­ ®
® ­
­ ® ­ ®
« 𝑎𝑛 ¬ « 1¬

Proof. — Let 𝜆 and 𝜇 be two units in A. Let us observe that each of

the following matrices is deduced from the preceding one through an
elementary row or column operation:

𝜆 0 R1 ←R1 +𝜇−1 R2 𝜆 1 C1 ←C1 −C2𝜆

     
0 1
−−−−−−−−−−→ −−−−−−−−−→
0 𝜇 0 𝜇 −𝜇𝜆 𝜇
𝜇𝜆 1 − 𝜇 R2 ←R2 +R1 𝜇𝜆 1 − 𝜇
   
R1 ←R1 −R2
−−−−−−−−→ −−−−−−−−→
−𝜇𝜆 𝜇 0 1
𝜇𝜆 0
 
C2 ←C2 −C1 (𝜇𝜆)−1 (1−𝜇)
−−−−−−−−−−−−−−−−→ .
0 1
Consequently, there exist matrices U, V ∈ E2 (A) such that

𝜆 𝜇𝜆
   
U V= ,
𝜇 1
which proves the result for 𝑛 = 2.
In the general case, we can perform the corresponding row and column
operations for indices 𝑛 − 1 and 𝑛, replacing 𝑎 𝑛−1 and 𝑎 𝑛 by 𝑎 𝑛 𝑎 𝑛−1 and 1.
We repeat this process for indices 𝑛 − 2 and 𝑛 − 1, etc., until we do it for
indices 2 and 1, which gives the result. 

5.3.8. Principal ideal domains. — We want to generalize theorem 5.3.5

to the case of any principal ideal domain A. However, elementary opera-
tions on rows and columns will not suffice anymore, the group E𝑛 (A) is
not necessarily equal to SL𝑛 (A) and we need to use the full group SL𝑛 (A).

Lemma (5.3.9). — Let A be a principal ideal domain, let 𝑎, 𝑏 be two nonzero

elements of A. Let 𝑑 be a gcd of (𝑎, 𝑏), let 𝑢, 𝑣 ∈ A be such that 𝑑 = 𝑎𝑢 + 𝑏𝑣;
244 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

𝑢 𝑣 ) belongs to
let 𝑟 and 𝑠 ∈ A be such that 𝑎 = 𝑑𝑟 and 𝑏 = 𝑑𝑠. The matrix ( −𝑠 𝑟
SL2 (A) and one has
𝑢 𝑣 𝑎 ∗ 𝑑 ∗
    
= .
−𝑠 𝑟 𝑏 ∗ 0 ∗

Proof. — We have 𝑑 = 𝑎𝑢 + 𝑏𝑣 = 𝑑(𝑢𝑟 + 𝑣𝑠). Since 𝑑 ≠ 0, we obtain

𝑢 𝑣 ) ∈ SL (A). The rest is immediate.
𝑢𝑟 + 𝑣𝑠 = 1 so that ( −𝑠 
𝑟 2

Theorem (5.3.10). — Let A be a principal ideal domain, let M ∈ M𝑛,𝑝 (A).

There exists matrices P ∈ SL𝑛 (A), Q ∈ SL𝑝 (A) and D ∈ M𝑛,𝑝 (A) such that
M = PDQ, D is “diagonal” (meaning 𝑑 𝑖,𝑗 = 0 for 𝑖 ≠ 𝑗) and such that 𝑑 𝑖,𝑖
divides 𝑑 𝑖+1,𝑖+1 for any integer 𝑖 such that 1 6 𝑖 < inf(𝑛, 𝑝). Moreover, the
diagonal entries 𝑑1,1 , . . . of D are well defined up to units.

Proof. — The proof is very close to that of theorem 5.3.5, and we only
indicate the modifications to be done. For any nonzero 𝑎 ∈ A, let 𝜔(𝑎) be
the size of 𝑎, namely the number of irreducible factors of 𝑎, counted with
multiplicities. We now argue by a double induction, first on sup(𝑛, 𝑝)
and then on the minimal size of a nonzero entry of M.
As above, we may assume that 𝑚1,1 is a nonzero element of minimal
size. If all entries of the first column are divisible by 𝑚1,1 , elementary
operations on rows allow us to assume that the entry (1, 1) is the only
nonzero entry of this first column. Otherwise, it follows from lemma 5.3.9
that there exists a matrix P1 ∈ SL𝑛 (A) of the form
𝑢 0 ... 𝑣 0 ...
© ª
­0 1 0 ®
­ . . . . ...
­ ..
®
®
­ ®
­−𝑠
­ 𝑟 ®
®
­ 1 ®
...
­ ®
« ¬
such that the (1, 1) entry of P1 M is a gcd of (𝑚1,1 , 𝑚1,𝑗 ), the integer 𝑗 ∈
{2, . . . , 𝑝} being chosen such that 𝑚1,𝑖 is not a multiple of 𝑚1,1 . The
size of the (1, 1)-entry of P1 M is now strictly smaller than 𝜔(𝑚1,1 ); by
induction, there exist P ∈ SL𝑛 (A) and Q ∈ SL𝑝 (A) such that PP1 MQ is
diagonal, each diagonal entry dividing the next one.
 5.4. FINITELY GENERATED MODULES OVER A PRINCIPAL IDEAL DOMAIN 245

By right multiplication with analogous matrices, we may also assume

that the 𝑚1,1 is the only nonzero entry of the first row. Finally, we are
reduced to the case where the matrix M takes the form

©𝑚11 0 . . . 0ª
­ 0 ®
­ .
­ ..
®
M 0 ®
­ ®
« 0 ¬
with M0 ∈ M𝑛−1,𝑝−1 (A). By the same argument as in the euclidean case,
we may then assume that all entries of M0 are divisible by 𝑚1,1 . We
then conclude in the same way, applying the induction hypothesis to
the matrix M0 whose size is smaller than that of M.
For the proof of the uniqueness assumption, it suffices to copy the one
we have done in the case of euclidean rings, see theorem 5.3.5. 

5.4. Finitely generated modules over a principal ideal domain

The theory of modules over a given ring A strongly depends on
its algebraic properties. The case of division rings gives rise to the
particularly well behaved theory of vector spaces. In order of complexity,
the next case is that of principal ideal domains to which we now turn.
Proposition (5.4.1). — Let A be a principal ideal domain, let M be a free
A-module of rank 𝑛 and let N be a submodule of M. Then, N is a free A-module
and its rank is 6 𝑛.
Proof. — It suffices to show that every submodule N of A𝑛 is free of
rank 6 𝑛; let us prove this by induction on 𝑛.
If 𝑛 = 0, then A𝑛 = 0, hence N = 0 so that N is a free A-module of
rank 0.
Assume that 𝑛 = 1. Then, N is an ideal of A. If N = 0, then N is free
of rank 0. Otherwise, A being a principal ideal domain, there exists a
nonzero element 𝑑 ∈ A such that N = (𝑑). Since A is a domain, the map
𝑎 ↦→ 𝑑𝑎 is an isomorphism from A to N, so that N is free of rank 1.
Let now 𝑛 be an integer > 2 and let us assume that for any integer 𝑟 < 𝑛,
every submodule of A𝑟 is free of rank 6 𝑟. Let N be a submodule of A𝑛 .
Let 𝑓 : A𝑛 → A be the linear form given by (𝑎 1 , . . . , 𝑎 𝑛 ) ↦→ 𝑎 𝑛 ; it is
246 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

surjective and its kernel is the submodule M0 = A𝑛−1 × {0} of A𝑛 . By

induction, the ideal N1 = 𝑓 (N) of A is free of rank 6 1. The submodule
N0 = N ∩ M0 of M0 is isomorphic to a submodule of A𝑛−1 , so is free of
rank 6 𝑛 − 1. It then follows from proposition 3.5.5 that N is free of
rank 6 𝑛. 

Remark (5.4.2). — One shall observe that this proposition recovers the
fact that when K is a field, every subspace of K𝑛 is free of dimension 6 𝑛.
It is important to observe that the hypothesis on the ring A is necessary.
If A is neither a principal ideal domain nor a field, then there exists a
nonzero ideal of A which is not free as a A-module. Moreover, there
are noncommutative rings A possessing a submodule isomorphic to A2 .
Finally, observe that it is possible that a submodule N ⊂ A𝑛 have rank 𝑛
while being distinct from A𝑛 , as is witnessed by the simple example
A = Z, 𝑛 = 1 and N = 2Z ⊂ Z.

The following theorem is more precise: it furnishes a basis of a free

module over a principal ideal domain which is adapted to a given
submodule.

Theorem (5.4.3). — Let A be a principal ideal domain, let M be a free A-module

of rank 𝑛 and let N be a submodule of M. There exists a basis (𝑒1 , . . . , 𝑒 𝑛 )
of M, an integer 𝑟 such that 0 6 𝑟 6 𝑛 and elements 𝑑1 , . . . , 𝑑𝑟 of A such
that 𝑑 𝑖 divides 𝑑 𝑖+1 for every integer 𝑖 such that 1 6 𝑖 < 𝑟 and such that
(𝑑1 𝑒1 , . . . , 𝑑𝑟 𝑒 𝑟 ) is a basis of N.
Moreover, the integer 𝑟 and the ideals (𝑑1 ), . . . , (𝑑𝑟 ) do not depend on the
choice of a particular such basis (𝑒1 , . . . , 𝑒 𝑛 ).

Proof. — We may assume M = A𝑛 . By the preceding proposition, the

submodule N is free of some rank 𝑝 ∈ {0, . . . , 𝑛}. In other words, there
exists a morphism 𝑓1 : A𝑝 → A𝑛 which induces an isomorphism from A𝑝
to N. (In fact, the existence of a morphism 𝑓 : A𝑝 → A𝑛 with image N
is all that we will need, that is, it suffices to know that N is finitely
generated; we will show below how this follows from the fact that A is
a noetherian ring.) By theorem 5.3.10, there exist matrices P ∈ GL𝑛 (A),
Q ∈ GL𝑝 (A) and D ∈ M𝑛,𝑝 (A) such that U = PDQ, the matrix D being
 5.4. FINITELY GENERATED MODULES OVER A PRINCIPAL IDEAL DOMAIN 247

diagonal and each of its diagonal entries 𝑑1 , . . . , 𝑑inf(𝑛,𝑝) dividing the

next one.
If 𝑑1 = 0, set 𝑟 = 0. Otherwise, let 𝑟 be the largest integer
in {1, . . . , inf(𝑛, 𝑝)} such that 𝑑𝑟 ≠ 0; one then has 𝑑𝑟+1 = · · · = 𝑑inf(𝑛,𝑝) =
0.
Let 𝑢 : A𝑝 → A𝑝 be the automorphism with matrix Q, 𝑣 : A𝑛 → A𝑛 be
the automorphism with matrix P, and let 𝑔 : A𝑝 → A𝑛 be the morphism
with matrix D; one has 𝑓 = 𝑣 ◦ 𝑔 ◦ 𝑢. Let (𝜀1 , . . . , 𝜀𝑛 ) for the canonical
basis of A𝑛 . We see that (𝑑1 𝜀1 , . . . , 𝑑𝑟 𝜀𝑟 ) is a basis of Im(𝑔) = Im(𝑔 ◦ 𝑢),
so that (𝑑1 𝑣(𝜀1 ), . . . , 𝑑𝑟 𝑣(𝜀𝑟 )) is a basis of Im(𝑣 ◦ 𝑔 ◦ 𝑢) = Im( 𝑓 ) = N.
On the other hand, (𝑣(𝜀1 ), . . . , 𝑣(𝜀𝑛 )) is a basis of A𝑛 , because 𝑣 is an
automorphism. It thus suffices to set 𝑒 𝑖 = 𝑣(𝜀𝑖 ) for 𝑖 ∈ {1, . . . , 𝑛}.
From this description, it follows that the Fitting ideals of the mod-
ule M/N can be computed from 𝑑1 , . . . , 𝑑𝑛 , namely: Fit𝑛−𝑘 (M/N) =
Δ 𝑘 (D) = (𝑑1 . . . 𝑑 𝑘 ) for 𝑘 6 𝑟 and Fit𝑛−𝑘 (M/N) = (0) for 𝑘 > 𝑟. As in
theorem 5.3.10, this allows to recover the integer 𝑟 as well as the elements
𝑑1 , . . . , 𝑑𝑟 up to units.
We can also argue directly. Let (𝑒10 , . . . , 𝑒 𝑛0 ) be a basis of A𝑛 , 𝑠 be
an integer in {0, . . . , 𝑛}, 𝑑10 , . . . , 𝑑0𝑠 elements of A such that 𝑑0𝑖 |𝑑0𝑖+1 for
every 𝑖 < 𝑠 and such that (𝑑10 𝑒10 , . . . , 𝑑0𝑠 𝑒 𝑠0 ) is a basis of N. Since any two
bases of a free module over a commutative ring have the same cardinality,
we have 𝑠 = 𝑟. The matrix of the canonical injection from N to A𝑛 in the
bases (𝑒1 , . . . , 𝑒 𝑟 ) and (𝑒1 , . . . , 𝑒 𝑛 ) is equal to D = diag(𝑑1 , . . . , 𝑑𝑟 ). Let S
be the matrix expressing the basis (𝑒10 , . . . , 𝑒 𝑛0 ) in the basis (𝑒1 , . . . , 𝑒 𝑛 ),
let T be the matrix expressing the basis (𝑑10 𝑒10 , . . . , 𝑑0𝑟 𝑒 𝑟0 ) of N in the
basis (𝑑1 𝑒1 , . . . , 𝑑𝑟 𝑒 𝑟 ) and let D0 = diag(𝑑10 , . . . , 𝑑0𝑟 ). We have D = S−1 D0T.
Since S ∈ GL𝑛 (A) and T ∈ GL𝑟 (A), it follows from the uniqueness
assertion of theorem 5.3.10 that (𝑑1 ) = (𝑑10 ),. . . , (𝑑𝑟 ) = (𝑑0𝑟 ). 

Corollary (5.4.4). — Let A be a principal ideal domain and let M be a finitely

generated A-module. There exists an integer 𝑛 and elements 𝑑1 , . . . , 𝑑𝑛 of A,
non-units, such that 𝑑 𝑖 divides 𝑑 𝑖+1 for everyÉ integer 𝑖 such that 1 6 𝑖 < 𝑛
𝑛
such that M is isomorphic to the direct sum 𝑖=1 A/(𝑑 𝑖 ); the Fitting ideals
of M satisfy Fit𝑝 (M) = (𝑑1 · · · 𝑑𝑛−𝑝 ) for every integer 𝑝 such that 0 6 𝑝 < 𝑛,
and Fit𝑝 (M) = A for 𝑝 > 𝑛.
248 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

Moreover, if 𝑚 is an integer, 𝛿 1 , . . . , 𝛿 𝑚 are non-units of A such that 𝛿 𝑗

É𝑚 𝛿 𝑗+1 for every 𝑗 satisfying 1 6 𝑗 < 𝑚 and such that M is isomorphic to
divides
𝑗=1 A/(𝛿 𝑗 ), then 𝑚 = 𝑛 and (𝛿 𝑖 ) = (𝑑 𝑖 ) for every 𝑖 ∈ {1, . . . , 𝑛}.

Proof. — Let (𝑚1 , . . . , 𝑚𝑛 ) be a family of elements of M which gener-

ates M. Let 𝑓 : A𝑛 → M be the unique morphism that sends the 𝑖th
vector of the canonical basis of A𝑛 to 𝑚 𝑖 ; it is surjective by construction.
If N denotes its kernel, 𝑓 induces an isomorphism from A𝑛 /N to M.
Let (𝑒1 , . . . , 𝑒 𝑛 ) be a basis of A𝑛 , let 𝑑1 , . . . , 𝑑 𝑝 be elements of A such
that (𝑑1 𝑒1 , . . . , 𝑑 𝑝 𝑒 𝑝 ) is a basis of N and such that 𝑑 𝑖 divides 𝑑 𝑖+1 for
any integer 𝑖 such that 1 6 𝑖 < 𝑝 (theorem 5.4.3). Set 𝑑 𝑖 = 0 for
𝑖 ∈ {𝑝 + 1, . . . , 𝑛}. The morphism 𝜑 from A𝑛 to itself which maps
(𝑎1 , . . . , 𝑎 𝑛 ) to 𝑎1 𝑒1 + · · · + 𝑎 𝑛 𝑒 𝑛 is an isomorphism, because (𝑒1 , . . . , 𝑒 𝑛 )
is a basis of A𝑛 . Moreover, 𝜑(𝑎1 , . . . , 𝑎 𝑛 ) belongs to N if and only if 𝑑 𝑖
divides 𝑎 𝑖 for every 𝑖 ∈ {1, . . . , 𝑛}. É𝑛Passing to the quotient, we see that
M = Im( 𝑓 ◦ 𝜑) is isomorphic to 𝑖=1 A/(𝑑 𝑖 ).
For every 𝑖 ∈ {1, . . . , 𝑛 − 1} such that 𝑑 𝑖+1 is invertible, then so is 𝑑 𝑖 ,
because 𝑑 𝑖 divides 𝑑 𝑖+1 . Consequently, there exists an integer 𝑟 ∈
{0, 1, . . . , 𝑛} such that 𝑑 𝑖 is a unit if and only if 𝑖 6 𝑟. Then, M '
É 𝑛
𝑖=𝑟+1 A/(𝑑 𝑖 ); this proves the first part of the proof.
Resetting notation, we assume that 𝑟 = 0.
By the definition of Fitting ideals, one has Fit 𝑘 (M) = J𝑛−𝑘 ( 𝑓 ◦ 𝜑).
Moreover, the kernel of 𝑓 ◦ 𝜑 is generated by the vectors (𝑑1 𝑒1 , . . . , 𝑑𝑛 𝑒 𝑛 ).
Consequently, Fit𝑝 (M) = (1) if 𝑝 > 𝑛, Fit𝑝 (M) = (𝑑1 . . . 𝑑𝑛−𝑝 ) if 0 6 𝑝 6
𝑛 − 1, and Fit𝑝 (M) = (0) if 𝑝 < 0.
The uniqueness É𝑚property follows from this description of the Fitting
ideals: if M ' 𝑗=1 A/(𝛿 𝑗 ), where 𝑚 is an integer and 𝛿 1 , . . . , 𝛿 𝑚 are
non-units in A such that 𝛿1 |𝛿2 | . . . |𝛿 𝑚 , then (𝑑1 . . . 𝑑𝑛−𝑝 ) = (𝛿 1 . . . 𝛿 𝑚−𝑝 )
for every integer 𝑝. Taking 𝑝 = 𝑛, this implies that 𝑚 6 𝑛, since
otherwise, the ideal (𝛿 1 . . . 𝛿 𝑚−𝑛 ) would be non-trivial; by symmetry,
one has 𝑛 6 𝑚, hence 𝑛 = 𝑚. The equalities (𝑑1 . . . 𝑑𝑛−𝑝 ) = (𝛿 1 . . . 𝛿 𝑛−𝑝 )
for every 𝑝 ∈ {0, . . . , 𝑛} imply that there are units 𝑢1 , . . . , 𝑢𝑛 such that
𝛿1 . . . 𝛿 𝑘 = 𝑢 𝑘 𝑑1 . . . 𝑑 𝑘 for all 𝑘 ∈ {1, . . . , 𝑛}.
Let us show that there are units 𝑣1 , . . . , 𝑣 𝑛 such that 𝛿 𝑘 = 𝑣 𝑘 𝑑 𝑘 for
every 𝑘 ∈ {1, . . . , 𝑛}. This holds for 𝑘 = 1, with 𝑣 1 = 𝑢1 . Assume that
it holds for 𝑘. If 𝑑1 . . . 𝑑 𝑘 = 0, then 𝛿 1 . . . 𝛿 𝑘 = 0; there exists 𝑝, 𝑞 6 𝑘
 5.4. FINITELY GENERATED MODULES OVER A PRINCIPAL IDEAL DOMAIN 249

such that 𝑑 𝑝 = 𝛿 𝑞 = 0, and the divisibilities 𝑑 𝑘 | 𝑑 𝑘+1 and 𝛿 𝑘 | 𝛿 𝑘+1 imply

that 𝛿 𝑘+1 = 𝑑 𝑘+1 = 0; in this case, we set 𝑢 𝑘+1 = 1. Otherwise, one has
𝑑1 . . . 𝑑 𝑘 ≠ 0; writing
𝑢 𝑘+1 𝑑1 . . . 𝑑 𝑘+1 = 𝛿 1 . . . 𝛿 𝑘+1 = 𝑢 𝑘 𝑑1 . . . 𝑑 𝑘 𝛿 𝑘+1 ,
we deduce that 𝛿 𝑘+1 = 𝑢 𝑘+1 𝑢 𝑘−1 𝑑 𝑘+1 . This proves the result by induction.
Here is another approach. É𝑛
We thus assume that an isomorphism M ' 𝑖=1 A/(𝑑 𝑖 ) be given,
where 𝑑1 , . . . , 𝑑𝑛 are non-units in A such that 𝑑 𝑖 divides 𝑑 𝑖+1 for 1 6 𝑖 < 𝑛.
Let 𝑝 be an irreducible element of A; by lemma 5.4.5 applied with 𝑎 = 𝑝 𝑚
and 𝑏 = 𝑝, the module 𝑝 𝑚−1 M/𝑝 𝑚 M is an (A/𝑝A)-vector space and
its dimension is the number of indices 𝑖 ∈ {1, . . . , 𝑛} such that 𝑝 𝑚
divides 𝑑 𝑖 . Using the divisibility 𝑑 𝑖 |𝑑 𝑖+1 for 𝑖 ∈ {1, . . . , 𝑛 − 1}, we deduce
the equivalence: 𝑝 𝑚 divides 𝑑 𝑖 if and only if dimA/𝑝A (𝑝 𝑚−1 M/𝑝 𝑚 M) >
𝑚 + 1 − 𝑖. This determines the irreducible factors of the elements 𝑑 𝑖 , as
well as their exponents, hence the ideals (𝑑 𝑖 ). 

Lemma (5.4.5). — Let A be a principal ideal domain, let 𝑑 be an element of A

and let M = A/(𝑑). Let 𝑝 be an irreducible element of A; for every integer 𝑚
such that 𝑚 > 1, let M𝑚 = 𝑝 𝑚−1 M/𝑝 𝑚 M. Then, M𝑚 is an (A/𝑝)-vector
space; its dimension is 1 if 𝑝 𝑚 divides 𝑑, and is 0 otherwise.
Proof. — Since the multiplication by 𝑝 in the A-module M𝑚 is zero,
the canonical morphism A → End(M𝑚 ) factors through A/𝑝A: this
endowes M𝑚 with the structure of an (A/𝑝A)-vector space. (Recall that
A/𝑝A is a field.)
Let 𝑛 be the exponent of 𝑝 in the decomposition of 𝑑 into prime factors,
that is, the supremum of all integer 𝑛 such that 𝑝 𝑛 divides 𝑑. (If 𝑑 = 0,
then 𝑛 = +∞; otherwise, 𝑛 is finite, 𝑝 𝑛 divides 𝑑 but 𝑝 𝑛+1 doesn’t.) The
canonical bijection between submodules of A/𝑑A and submodules of A
containing (𝑑) maps 𝑝 𝑚 M to the ideal (𝑝 𝑚 , 𝑑) = (𝑝 inf(𝑚,𝑛) ). This furnishes
an isomorphism

𝑚−1 𝑚 𝑝 inf(𝑚−1,𝑛) A/𝑑A 𝑝 inf(𝑚−1,𝑛) A

M𝑚 = 𝑝 M/𝑝 M ' ' .
𝑝 inf(𝑚,𝑛) A/𝑑A 𝑝 inf(𝑚,𝑛) A
Consequently, M𝑚 = 0 if and only if inf(𝑚−1, 𝑛) = inf(𝑚, 𝑛), that is, if and
only if 𝑚 > 𝑛, which amounts to say that 𝑝 𝑚 does not divide 𝑑. Otherwise,
250 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

if 𝑚 6 𝑛, then inf(𝑚, 𝑛) = 𝑚, inf(𝑚−1, 𝑛) = 𝑚−1 and M𝑚 ' 𝑝 𝑚−1 A/𝑝 𝑚 A.

Now, the map A → 𝑝 𝑚−1 A given by 𝑎 ↦→ 𝑝 𝑚−1 A induces an isomorphism
from A/𝑝A to 𝑝 𝑚−1 A/𝑝 𝑚 A, so that dimA/𝑝A (M𝑚 ) = 1 in that case. 

Definition (5.4.6). — Let A be aÉ principal ideal domain and let M be a finitely

𝑛
generated A-module. Let M ' 𝑖=1 A/(𝑑 𝑖 ) be a decomposition of M where
𝑑1 , . . . , 𝑑𝑛 are non-units in A such that 𝑑 𝑖 divides 𝑑 𝑖+1 for any integer 𝑖 such
that 1 6 𝑖 < 𝑛.
The ideals (𝑑1 ),. . . , (𝑑𝑛 ) are called the invariant factors of M. The number
of those ideals which are zero is called the rank of M.
With these notations, we thus have 𝑑 𝑖 = 0 if and only if 𝑛 − 𝑟 +1 6 𝑖 6 𝑛,
so that M can also be written
𝑛−𝑟
Ê
𝑟
M'A ⊕ A/(𝑑 𝑖 ).
𝑖=1

ÉMoreover,
𝑛−𝑟
with the terminology of example 3.1.9, the second module
A/(𝑑 𝑖 ) in the preceding expression corresponds to the torsion
𝑖=1
submodule of M.
Corollary (5.4.7). — A finitely generated module over a principal ideal domain
is free if and only if it is torsion-free.
Corollary (5.4.8). — Let A be a principal ideal domain, let M be a finitely
generated free A-module and let N be a submodule of M. For N to admit a
direct summand in M, it is necessary and sufficient that the quotient M/N be
torsion-free.
Proof. — Assume that N has a direct summand P. Then, the canonical
morphism cl from M to M/N induces an isomorphism from P onto M/N.
Since P is a submodule of the free A-module M, it is torsion-free.
Consequently, M/N is torsion-free.
Conversely, let us assume that M/N is torsion-free. Since it is finitely
generated and A is a principal ideal domain, then M/N is a free A-
module (corollary 5.4.7). Let ( 𝑓1 , . . . , 𝑓𝑟 ) be a basis of M/N; for any
𝑖 ∈ {1, . . . , 𝑟}, let 𝑒 𝑖 be an element of M which is mapped to 𝑓𝑖 by the
canonical surjection. Let then P be the submodule of M generated by
𝑒1 , . . . , 𝑒 𝑟 . Let us show that P is a direct summand of N in M.
 5.4. FINITELY GENERATED MODULES OVER A PRINCIPAL IDEAL DOMAIN 251

Let 𝑚 ∈ N ∩ P and write 𝑚 = 𝑎 𝑖 𝑒 𝑖 for some 𝑎1 , . . . , 𝑎 𝑟 ∈ A. We have

Í
cl(𝑚) = 0, because 𝑚 ∈ N, hence 𝑎 𝑖 𝑓𝑖 = 0. Since the family ( 𝑓𝑖 ) is free,
Í
it follows that 𝑎 𝑖 = 0 for all 𝑖. Consequently, 𝑚 = 0 and N ∩ P = 0.
Conversely, let 𝑚 ∈ M; let (𝑎 1 , . . . , 𝑎 𝑟 ) be elements of A such that
𝑎 𝑖 𝑓𝑖 . Then, 𝑝 = 𝑎 𝑖 𝑒 𝑖 belongs to P and cl(𝑝) = cl(𝑚). It
Í Í
cl(𝑚) =
follows that cl(𝑚 − 𝑝) = 0, hence 𝑚 − 𝑝 ∈ N, which proves that 𝑚 ∈ P + N.
This shows that M = P + N.
We have thus shown that P is a direct summand of N, as claimed. 

Lemma and Definition (5.4.9). — Let A be a principal ideal domain, let M be

an A-module. For every irreducible element 𝑝 ∈ A, the set M(𝑝) of all 𝑚 ∈ M
for which there exists 𝑛 > 0 with 𝑝 𝑛 𝑚 = 0 is a submodule of M, called the
𝑝-primary component of M.
Proof. — One has 0 ∈ M(𝑝). Let 𝑚, 𝑚 0 ∈ M(𝑝); we want to show that
𝑚 + 𝑚 0 ∈ M(𝑝). Let 𝑛, 𝑛 0 be positive integers such that 𝑝 𝑛 𝑚 = 𝑝 𝑛 𝑚 0 = 0;
0

set 𝑘 = sup(𝑛, 𝑛 0). Then, 𝑝 𝑘 (𝑚 + 𝑚 0) = 𝑝 𝑘−𝑛 (𝑝 𝑛 𝑚) + 𝑝 𝑘−𝑛 (𝑝 𝑛 𝑚 0) = 0, so

0 0

that 𝑚 + 𝑚 0 ∈ M(𝑝). Finally, let 𝑚 ∈ M(𝑝) and let 𝑎 ∈ A; if 𝑛 > 0 is such

that 𝑝 𝑛 𝑚 = 0, we see that 𝑝 𝑛 (𝑎𝑚) = 𝑎(𝑝 𝑛 𝑚) = 0, hence 𝑎𝑚 ∈ M(𝑝). This
shows that M(𝑝) is a submodule of M. 
Let 𝑝 and 𝑞 be two irreducible elements of A. If there exists a unit 𝑢 ∈ A
such that 𝑞 = 𝑢𝑝, it is clear from the definition that M(𝑝) ⊂ M(𝑞),
hence M(𝑝) = M(𝑞) by symmetry. Otherwise, it will follow from
proposition 5.4.10 below that M(𝑝) ∩ M(𝑞) = 0.
We fix a set 𝒫 of irreducible elements in A such that any irreducible
element of A is equal to the product of a unit by a unique element
of 𝒫. In particular, when 𝑝 varies in 𝒫, the ideals (𝑝) are maximal and
pairwise distinct.
Proposition (5.4.10) (Primary decomposition). — Let A be a principal ideal
domain, let M be an A-module.
É Then the torsion submodule of M decomposes
as the direct sum T(M) = 𝑝∈𝒫 M(𝑝).

Proof. — Let us first prove these modules M(𝑝) are in direct sum. Let
thus (𝑚 𝑝 )𝑝∈𝒫 be an almost-null family of elements of M, where 𝑚 𝑝 ∈ M(𝑝)
for every 𝑝 ∈ 𝒫, and assume that 𝑚 𝑝 = 0. We want to prove that
Í
𝑚 𝑝 = 0 for every 𝑝 ∈ 𝒫. Let I be the subset of 𝒫 consisting of those 𝑝
252 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

for which 𝑚 𝑝 ≠ 0; by hypothesis, it is a finite subset. For every 𝑝 ∈ I, let

𝑛 𝑝 be a positive integer such that 𝑝 𝑛 𝑝 𝑚 𝑝 = 0. For any 𝑝 ∈ I, let
Ö
𝑎𝑝 = 𝑞 𝑛𝑞 .
𝑞∈I
𝑞≠𝑝

By construction, 𝑎 𝑝 𝑚 𝑞 = 0 for every 𝑞 ∈ I such that 𝑞 ≠ 𝑝, and also for

every 𝑞 ∈ 𝒫 I. Consequently, 𝑎 𝑝 𝑚 𝑝 = 𝑎 𝑝 ( 𝑞 𝑚 𝑞 ) = 0. However, let us
Í
observe that 𝑎 𝑝 and 𝑝 𝑛 𝑝 are coprime. Since A is a principal ideal domain,
there exist 𝑢, 𝑣 ∈ A such that 𝑢𝑎 𝑝 + 𝑣𝑝 𝑛 𝑝 = 1; we get
𝑚 𝑝 = (𝑢𝑎 𝑝 + 𝑣𝑝 𝑛 𝑝 )𝑚 𝑝 = 𝑢(𝑎 𝑝 𝑚 𝑝 ) + 𝑣(𝑝 𝑛 𝑝 𝑚 𝑝 ) = 0.
We thus have shown that 𝑚 𝑝 = 0 for every 𝑝 ∈ 𝒫; in other words, the
modules M(𝑝) are in direct sum, as claimed.
By definition, one has M(𝑝) ⊂ T(M) for every 𝑝, hence
É
𝑝 M(𝑝) ⊂
T(M). Conversely, let 𝑚 ∈ T(M) and let 𝑎 be any nonzero element
of A such that 𝑎𝑚 = 0; let 𝑎 = 𝑢 𝑝∈I 𝑝 𝑛 𝑝 be the decomposition of 𝑎
Î
in irreducible factors, where I is a finite subset of 𝒫 and É 𝑢 is a unit.
Let us prove by induction on Card(I) that 𝑚 belongs to 𝑝 M(𝑝). This
is obvious if Card(I) 6 1. Otherwise, fix 𝑞 ∈ I and let 𝑏 = 𝑎/𝑞 𝑛 𝑞 ; the
elements 𝑏 and 𝑞 𝑛 𝑞 are coprime, hence there exist 𝑐, 𝑑 ∈ A such that
𝑐𝑏 + 𝑑𝑞 𝑛 𝑞 = 1, hence 𝑚 = 𝑐𝑏𝑚 + 𝑑𝑞 𝑛 𝑞 𝑚. Since 𝑞 𝑛 𝑞 (𝑏𝑚) = 𝑎𝑚 = 0, É one has
𝑛𝑞 𝑛𝑞
𝑏𝑚 ∈ M𝑞 . On the other hand, 𝑏𝑞 𝑚 = 𝑎𝑚 = 0, hence 𝑞 𝑚 ∈ 𝑝 M𝑝
because 𝑏 has less irreducibleÉ
factors than 𝑎. By induction, itÉ follows that
𝑚 belongs to the submodule 𝑝 M(𝑝) of M, hence T(M) = 𝑝 M(𝑝), as
claimed. 

Remark (5.4.11). — Let A be a principal ideal domain, let M be a finitely

generated torsion A-module. Let 𝑝 ∈ 𝒫 be an irreducible element of A
and let M(𝑝) be the corresponding 𝑝-primary component of M. By the
primary decomposition, we see that M(𝑝) is isomorphic to a quotient
of M, namely, the quotient by the direct sum of the other primary
components. Consequently, M(𝑝) is finitely generated. Let 𝑚1 , . . . , 𝑚𝑟
be elements of M(𝑝) which generate M(𝑝); for every 𝑖 ∈ {1, . . . , 𝑟}, let
𝑘 𝑖 ∈ N be such that 𝑝 𝑘 𝑖 𝑚 𝑖 = 0 and set 𝑘 = sup(𝑘1 , . . . , 𝑘 𝑟 ); then 𝑝 𝑘 𝑚 𝑖 = 0
for every 𝑖, hence 𝑝 𝑘 𝑚 = 0 for every 𝑚 ∈ M(𝑝).
 5.5. APPLICATION: FINITELY GENERATED ABELIAN GROUPS 253

The invariant factors of M(𝑝) are of the form (𝑝 𝑛1 ), . . . , (𝑝 𝑛 𝑠 ), where

𝑛1 6 · · · 6 𝑛 𝑠 are positive integers. Their knowledge, for every 𝑝 ∈ 𝒫,
determines the invariant factors of M, and conversely. We shall see
explicit examples in the next section.

5.5. Application: Finitely generated abelian groups

The main examples of principal ideal rings are Z and K[X] (where K is
a field). In this section we explicit the case of the ring Z; the case of a
polynomial ring will be the subject of the next section.
Recall that a Z-module is nothing but an abelian group, so that
finitely generated Z-modules are just finitely generated abelian groups.
Moreover, any ideal of Z has a unique positive generator. In the case
A = Z, corollary 5.4.4 gives the following theorem:

Theorem (5.5.1). — Let G be a finitely generated abelian group. There exists

an integer 𝑟 > 0 and a family (𝑑1 , . . . , 𝑑 𝑠 ) of integers at least equal to 2, both
uniquely determined, such that 𝑑 𝑖 divides 𝑑 𝑖+1 for 1 6 𝑖 < 𝑠 and such that

G ' Z𝑟 ⊕ (Z/𝑑1 Z) ⊕ · · · ⊕ (Z/𝑑 𝑠 Z).

The integer 𝑟 is the rank of G, the integers 𝑑1 , . . . , 𝑑 𝑠 are called its

invariant factors.
Theorem 5.5.1 furnishes a “normal form” for any finitely generated
abelian group, which allows, in particular, to decide whether two such
groups are isomorphic or not.
However, one must pay attention that the divisibility condition is
satisfied. To determine the invariant factors of a group, written as a
direct sum of cyclic group, the simplest procedure consists in using
the Chinese remainder theorem twice: first, decompose each cyclic
group as the direct sum of its primary components; then, collect factors
corresponding to distinct prime numbers, beginning with those of
highest exponents.

Example (5.5.2). — Let us compute the invariant factors of the abelian

groups G1 = (Z/3Z) ⊕ (Z/5Z) and G2 = (Z/6Z) ⊕ (Z/4Z).
254 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

Since 3 and 5 are coprime, (Z/3Z) ⊕ (Z/5Z) is isomorphic to Z/15Z,

by the Chinese remainder theorem. Consequently, the group G1 has
exactly one invariant factor, namely 15.
The integers 6 and 4 are not coprime, their gcd being 2. Since 6 = 2 · 3,
and 2 and 3 are coprime, it follows from the Chinese remainder that
G2 '(Z/6Z) ⊕ (Z/4Z) ' (Z/2Z) ⊕ (Z/3Z) ⊕ (Z/4Z)
' (Z/2Z) ⊕ (Z/4Z) ⊕ (Z/3Z) .

Observe that the preceding expression furnishes the primary decom-

position of the group G2 : the factors (Z/2Z) ⊕ (Z/4Z) and (Z/3Z) being
respectively the 2-primary and 3-primary components of G2 . We will
now collect factors corresponding to distinct prime numbers, and we
begin by those whose exponents are maximal, namely Z/3Z and Z/4Z.
We then obtain
G2 ' ((Z/2Z)) ⊕ ((Z/4Z) ⊕ (Z/3Z)) ' (Z/2Z) ⊕ (Z/12Z).
This shows that the invariant factors of G2 are 2 and 12.

Using this result we can also make the list of all finite abelian groups
whose cardinality is a given integer 𝑔. Indeed, it suffices to make the list
of all families of integers (𝑑1 , . . . , 𝑑 𝑠 ) such that 𝑑1 > 2, 𝑑 𝑖 divides 𝑑 𝑖+1 if
1 6 𝑖 < 𝑠, and 𝑔 = 𝑑1 . . . 𝑑 𝑠 . Again, the computation is easier if one first
considers the primary decomposition of the abelian group.

Example (5.5.3). — Let us determine all abelian groups G of cardinality 48.

Since 48 = 24 × 3, such a group will have a primary decomposition of the
form G = G2 ⊕ G3 , where G2 has cardinality 24 and G3 has cardinality 3.
In particular, we have G3 = Z/3Z. Let us now find all possible 2-primary
components. This amounts to finding all families (𝑑1 , . . . , 𝑑 𝑠 ) of integers
such that 𝑑1 > 2, 𝑑 𝑖 divides 𝑑 𝑖+1 if 1 6 𝑖 < 𝑠 and 𝑑1 . . . 𝑑 𝑠 = 24 . All of
the 𝑑 𝑖 must be powers of 2, say of the form 2𝑛 𝑖 , for integers 𝑛1 , . . . , 𝑛 𝑠
such that 1 6 𝑛1 6 𝑛2 6 · · · 6 𝑛 𝑠 , and one has 𝑠𝑖=1 𝑛 𝑖 = 4. The list is the
Í
following:
– 𝑑1 = 2, 𝑑2 = 2, 𝑑3 = 2, 𝑑4 = 2;
– 𝑑1 = 2, 𝑑2 = 2, 𝑑3 = 4;
– 𝑑1 = 2, 𝑑2 = 4, but then 𝑑3 6 2 so that this case does not happen;
 5.5. APPLICATION: FINITELY GENERATED ABELIAN GROUPS 255

– 𝑑1 = 2, 𝑑2 = 8;
– 𝑑1 = 4, 𝑑2 = 4;
– 𝑑1 = 8, but then 𝑑2 6 2, so that this case does not happen neither;
– 𝑑1 = 16.
We thus find that up to isomorphy, there are only five abelian groups of
cardinality 16:

(Z/2Z)4 , (Z/2Z)2 ⊕ (Z/4Z), (Z/2Z) ⊕ (Z/8Z), (Z/4Z)2 , (Z/16Z).

The abelian groups of cardinality 48 are obtained by taking the product

of one of these five groups with Z/3Z. We collect the factor Z/3Z with
the 2-primary factor with highest exponent and obtain the following list:

(Z/2Z)3 ⊕ (Z/6Z),
(Z/2Z)2 ⊕ (Z/12Z),
(Z/2Z) ⊕ (Z/24Z),
(Z/4Z) ⊕ (Z/12Z),
(Z/48Z).

Let us conclude with an application to field theory.

Theorem (5.5.4). — Let K be a (commutative) field and let G be a finite subgroup

of the multiplicative group of K. Then K is a cyclic group. In particular, the
multiplicative group of a finite field is cyclic.

Proof. — Let 𝑛 be the cardinality of G. We may assume that G ≠ {1}.

By theorem 5.5.1, there exist an integer 𝑟 > 1 and integers 𝑑1 , . . . , 𝑑𝑟 > 2
such that 𝑑 𝑖 divides 𝑑 𝑖+1 for every 𝑖 ∈ {1, . . . , 𝑟 − 1} and such that
G ' (Z/𝑑1 Z) × · · · × (Z/𝑑𝑟 Z). In particular, 𝑥 𝑑𝑟 = 1 for every 𝑥 ∈ G.
Since a nonzero polynomial with coefficients in a field has no more
roots than its degree, we conclude that 𝑛 6 𝑑𝑟 . Since 𝑛 = 𝑑1 . . . 𝑑𝑟 , this
implies that 𝑟 = 1 and 𝑑1 = 𝑛. Consequently, G ' Z/𝑛Z is a cyclic group
of order 𝑛. 
256 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

5.6. Application: Endomorphisms of a finite dimensional vector

space
Let K be a field and let A = K[X] be the ring of polynomials in one
indeterminate.

5.6.1. — Recall from remark 3.2.6 (see also exercise 3/4) that an endo-
morphism 𝑢 of a K-vector space V endowes it with the structure of a
K[X]-module, given by P · 𝑥 = P(𝑢)(𝑥) for any polynomial P ∈ K[X] and
Í𝑑
any vector 𝑥 ∈ V. Explicitly, if P = 𝑛=0 𝑎 𝑛 X𝑛 , one has

𝑑
Õ
P·𝑥 = 𝑎 𝑛 𝑢 𝑛 (𝑥).
𝑛=0

Let V𝑢 be the K[X]-module so defined. Conversely, let M be a K[X]-

module. Forgetting the action of X, we get a K-vector space V. The
action of X on V is a K-linear endomorphism 𝑢, and M = V𝑢 .
In order for the theory of K[X]-modules to reflect faithfully that of
K-vector spaces endowed with endomorphisms, we need to understand
this correspondence on a categorical point of view, that is, at the level of
morphisms.
Let W be a K-vector space and let 𝑣 be an endomorphism of W. A
morphism of K[X]-modules from V𝑢 to W𝑣 is an additive map 𝑓 : V → W
such that 𝑓 (P · 𝑥) = P · 𝑓 (𝑥) for every 𝑥 ∈ V and every P ∈ K[X]. Taking
P = 𝜆 ∈ K, this gives 𝑓 (𝜆𝑥) = 𝜆 𝑓 (𝑥): the map 𝑓 is morphism of K-vector
spaces. Taking P = X, this gives 𝑓 (𝑢(𝑥)) = 𝑣( 𝑓 (𝑥)), hence 𝑓 ◦ 𝑢 = 𝑣 ◦ 𝑓 .
Conversely, if 𝑓 : V → W is a morphism of K-vector spaces such that
𝑓 ◦ 𝑢 = 𝑣 ◦ 𝑓 , then 𝑓 ◦ P(𝑢) = P(𝑣) ◦ 𝑓 for every P ∈ K[X]. (Indeed,
the set of polynomials that satisfy this relation is a subring of K[X]
that contains K and X, hence is the whole of K[X].) Consequently,
P · 𝑓 (𝑥) = P(𝑣)( 𝑓 (𝑥)) = 𝑓 (P(𝑢)(𝑥)) = 𝑓 (P · 𝑥) for every P ∈ K[X] and every
𝑥 ∈ V, so that 𝑓 is a morphism of K[X]-modules.
In particular, isomorphisms 𝑓 : V𝑢 → W𝑣 of K[X]-modules are pre-
cisely isomorphisms 𝑓 : V → W of K-vector spaces such that 𝑓 ◦ 𝑢 = 𝑣 ◦ 𝑓 .
In the particular case V = W, we conclude that the K[X]-modules V𝑢
and V𝑣 are isomorphic if and only if there exists 𝑓 ∈ GL(V) such that
 5.6. APPLICATION: ENDOMORPHISMS OF A FINITE DIMENSIONAL VECTOR SPACE 257

𝑢 = 𝑓 −1 ◦ 𝑣 ◦ 𝑓 , that is, if and only if the endomorphisms 𝑢 and 𝑣 are

conjugate under GL(V).

Definition (5.6.2). — One says that a nonzero K[X]-module M is cyclic if

there exists a nonzero polynomial P ∈ K[X] such that M ' K[X]/(P).

We remark that the cyclic K[X]-module M = K[X]/(P) is finitely

dimensional as a K-vector space and its dimension is equal to deg(P).
Indeed, euclidean division by P shows that any polynomial of K[X] is
equal modulo P to a unique polynomial of degree < deg(P). Consequently,
the elements cl(1), cl(X), . . . , cl(Xdeg(P)−1 ) form a basis of M.

Lemma (5.6.3). — Let V be a K-vector space, let 𝑢 be an endomorphism of V.

The K[X]-module V𝑢 is cyclic if and only if there exists a vector 𝑥 ∈ V and an
integer 𝑛 > 1 such that the family (𝑥, 𝑢(𝑥), . . . , 𝑢 𝑛−1 (𝑥)) is a basis of V.

Such a vector 𝑥 ∈ V is called a cyclic vector for the endomorphism 𝑢.

Proof. — Assume that the K[X]-module V𝑢 is cyclic and fix an isomor-
phism 𝜑 of K[X]/(P) with V𝑢 , for some nonzero polynomial P ∈ K[X].
Let 𝑛 = deg(P). The image of cl(1) by 𝜑 is an element 𝑥 of V. Since
(1, cl(X), . . . , cl(X𝑛−1 )) is a basis of K[X]/(P) as a K-vector space, the
family (𝑥, 𝑢(𝑥), . . . , 𝑢 𝑛−1 (𝑥)) is a basis of V.
Conversely, let 𝑥 ∈ V be any vector such that (𝑥, 𝑢(𝑥), . . . , 𝑢 𝑛−1 (𝑥)) is a
basis of V. Write 𝑢 𝑛 (𝑥) = 𝑛−1 𝑝
𝑝=0 𝑎 𝑝 𝑢 (𝑥) in this basis, for some elements
Í

𝑎 0 , . . . , 𝑎 𝑛−1 ∈ K. Let Π = X𝑛 − 𝑛−1 𝑝

𝑝=0 𝑎 𝑝 X ∈ K[X]. Then, the map
Í
𝑓 : K[X] → V, P ↦→ P(𝑢)(𝑥), is a surjective morphism of K[X]-modules.
Let us show that Ker( 𝑓 ) = (Π). Indeed, one has
𝑛−1
Õ
𝑛
𝑓 (Π) = 𝑢 (𝑥) − 𝑎 𝑝 𝑢 𝑝 (𝑥) = 0,
𝑝=0

so that Π ∈ Ker( 𝑓 ); it follows that (Π) ⊂ Ker( 𝑓 ). Let then P ∈ Ker( 𝑓 )

and let R be the euclidean division of P by Π; one has deg(R) < 𝑛
and 𝑓 (R) = 𝑓 (P) = 0. We get a linear dependence relation between
𝑥, 𝑢(𝑥), . . . , 𝑢 𝑛−1 (𝑥); since this family is a basis of V, this dependence
relation must be trivial, which gives R = 0. This shows that Ker( 𝑓 ) =
(Π), so that 𝑓 induces, passing to the quotient, an injective morphism
258 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

𝜑 : K[X]/(P) → V. Since 𝑓 is surjective, 𝜑 is surjective as well, hence is

an isomorphism. This concludes the proof of the lemma. 

Remark (5.6.4). — If V𝑢 is a cyclic module and 𝑥 ∈ V is a cyclic vector,

the matrix of 𝑢 in the basis (𝑥, . . . , 𝑢 𝑛−1 (𝑥)) is of the form

0 𝑎 0
𝑎1 ®
© ª
­1 0
­ .
.. ... .. ®®
­ . ®,
. . . 0 ... ®
­
­
­ ®
« 1 𝑎 𝑛−1 ¬

It is the companion matrix CΠ of the polynomial

Π = X𝑛 − 𝑎 𝑛−1 X𝑛−1 − · · · − 𝑎1 X − 𝑎0 .

Let us prove that Π is both the minimal and characteristic polynomial

of the matrix CΠ (and of the endomorphism 𝑢).
Since 𝑥, . . . , 𝑢 𝑛−1 (𝑥) are linearly independent, one has Q(𝑢)(𝑥) ≠ 0
for every non-zero polynomial Q of degree < 𝑛. Moreover, one has
𝑢 𝑛 (𝑥) = 𝑎0 𝑥 + 𝑎1 𝑢(𝑥) + · · · + 𝑢 𝑛−1 (𝑥), so that Π(𝑢)(𝑥) = 0. Since 𝑢 𝑘 and
Π(𝑢) commute, it follows that Π(𝑢)(𝑢 𝑘 (𝑥)) = 𝑢 𝑘 (Π(𝑢)(𝑥)) = 0 for every
𝑘 ∈ {0, . . . , 𝑛 − 1}. Consequently, Π(𝑢) = 0. Since deg(Π) = 𝑛, this
implies that Π is the minimal polynomial of 𝑢, and of CΠ .
The characteristic polynomial of 𝑢, or of CΠ , is a monic polynomial
of degree 𝑛; by Cayley–Hamilton’s theorem, it is a multiple of Π. It is
therefore equal to Π. One can also compute it explicitly by induction
on 𝑛: expanding the determinant along the first row, one has

X 0 −𝑎
© ª
­−1 X −𝑎1 ®
­ . . . .. ®
PCΠ (X) = det ­
­ −1 . ®
... X
®
­
­ −𝑎 𝑛−2 ®®
« −1 X − 𝑎 𝑛−1 ¬
 5.6. APPLICATION: ENDOMORPHISMS OF A FINITE DIMENSIONAL VECTOR SPACE 259

X 1 −𝑎
© ª
­−1 X −𝑎2 ®
..
−1 . . .
­ ®
= X det ­­ . ®
... X
®
­
­ −𝑎 𝑛−2 ®®
« −1 X − 𝑎 𝑛−1 ¬
−1 X
−1 . . .
© ª
+ (−1)𝑛 𝑎0 det ­­
­ ®
...
®
­ X ®®
« −1¬
= X(X𝑛−1 − 𝑎 𝑛−1 X𝑛−2 − · · · − 𝑎1 ) + (−1)𝑛 𝑎0 (−1)𝑛−1
= Π(X).

Applied to the K[X]-module defined by an endomorphism of a finite

dimensional K-vector space, corollary 5.4.4 thus gives the following
theorem.

Theorem (5.6.5). — Let K be a field. Let V be a finite dimensional K-vector space

and let 𝑢 be an endomorphism of V. There exists a unique family (P1 , . . . , P𝑟 )
of monic, non-constant, polynomials of K[X] such that P𝑖 divides P𝑖+1 for any
integer 𝑖 such that 1 6 𝑖 < 𝑟 and such that, in some basis of V, the matrix of 𝑢
takes the following block-diagonal form

©CP1 ª
­ C P2
®
®.
...
­
­ ®
­ ®
« CP 𝑟 ¬

The polynomials (P1 , . . . , P𝑟 ) are called the invariant factors of the

endomorphism 𝑢. It is apparent from the matrix form above that P𝑟 is
the minimal polynomial of 𝑢, while P1 . . . P𝑟 is its characteristic polynomial.

Corollary (5.6.6). — Two endomorphisms of a finite dimensional vector space

are conjugate if and only if they have the same invariant factors.

Applying the theory of Fitting ideals, we also get the following algo-
rithm to compute the invariant factors of a matrix.
260 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

Proposition (5.6.7). — Let K be a field, let U be a matrix in M𝑛 (K). For any

integer 𝑝 such that 1 6 𝑝 6 𝑛, let Δ𝑝 ∈ K[X] be the (monic) gcd of the 𝑝 × 𝑝
minors of the matrix XI𝑛 − U. One has Δ1 | Δ2 | · · · | Δ𝑛 . Let 𝑚 ∈ {0, . . . , 𝑛}
be such that Δ𝑝 = 1 for 𝑝 6 𝑚 and Δ𝑝 ≠ 1 for 𝑝 ∈ {𝑚 + 1, . . . , 𝑛}. Then,
there exist monic polynomials Q1 , . . . , Q𝑛−𝑚 ∈ K[X] such that
Q1 = Δ𝑚+1 , Q1 Q2 = Δ𝑚+2 , ..., Q1 . . . Q𝑛−𝑚 = Δ𝑛 .
The invariant factors of U are the polynomials Q1 , . . . , Q𝑛−𝑚 .
Proof. — Let A = K[X]. The endomorphism U of K𝑛 endowes K𝑛 with
a structure of an A-module (K𝑛 )U defined by P · 𝑣 = P(𝑢)(𝑣) for P ∈ A
and 𝑣 ∈ K𝑛 . The proof of the proposition consists in computing in two
different ways the Fitting ideals of (K𝑛 )U , once by proving that (K𝑛 )U is
isomorphic to A𝑛 /Im(XI𝑛 −U), and once by the formula of corollary 5.4.4,
Fit 𝑘 ((K𝑛 )U ) = (P1 . . . P𝑟−𝑘 ) for all integers 𝑘 > 0, where P1 , . . . , P𝑟 are the
invariant factors of U.
Let (𝑒1 , . . . , 𝑒 𝑛 ) be the canonical basis of the free A-module A𝑛 ; let also
( 𝑓1 , . . . , 𝑓𝑛 ) be the canonical basis of K𝑛 . The matrix U ∈ M𝑛 (K) induces
an endomorphism of K𝑛 , as well as an endomorphism of A𝑛 ; with denote
both of them by 𝑢. By the universal property of free modules, there is a
unique morphism 𝜓 of A-modules from A𝑛 to (K𝑛 )𝑢 such that 𝜓(𝑒 𝑖 ) is
the 𝑖th vector 𝑓𝑖 of the canonical basis of K𝑛 , and a unique morphism 𝜃
of K-vector spaces from K𝑛 to A𝑛 such that 𝜃( 𝑓𝑖 ) = 𝑒 𝑖 .
By construction, the matrix of the morphism
𝜑 : A𝑛 → A𝑛 , 𝑒 𝑖 ↦→ X𝑒 𝑖 − 𝑢(𝑒 𝑖 )
is equal to XI𝑛 − U. For any 𝑖 ∈ {1, . . . , 𝑛}, one has
𝜓(𝜑(𝑒 𝑖 )) = 𝜓(X𝑒 𝑖 − 𝑢(𝑒 𝑖 )) = 𝑢(𝜓(𝑒 𝑖 )) − 𝜓(𝑢(𝑒 𝑖 )) = 0
since 𝜓 is a morphim of K[X]-modules. Consequently, 𝜓 ◦ 𝜑 = 0, that is,
Im(𝜑) ⊂ Ker(𝜓). Passing to the quotient, 𝜓 induces a morphism 𝜓˜ from
A𝑛 /Im(𝜑) to (K𝑛 )𝑢 . Let 𝜃˜ be the composition of 𝜃 with the projection
to A𝑛 /Im(𝜑). For any 𝑖 ∈ {1, . . . , 𝑛}, one has
˜ 𝑓𝑖 ) = cl(𝜃( 𝑓𝑖 )) = cl(𝑒 𝑖 ),
𝜃(
and
˜
𝜃(𝑢( 𝑓𝑖 )) = cl(𝑢(𝑒 𝑖 )) = cl(X · 𝑒 𝑖 ) = X · cl(𝑒 𝑖 ),
 5.6. APPLICATION: ENDOMORPHISMS OF A FINITE DIMENSIONAL VECTOR SPACE 261

so that 𝜃˜ is a morphism of A-modules. For any 𝑖, one has

𝜃˜ ◦ 𝜓(cl(𝑒
˜ ˜ ˜
𝑖 )) = 𝜃 ◦ 𝜓(𝑒 𝑖 ) = 𝜃( 𝑓𝑖 ) = cl(𝑒 𝑖 );

since the vectors cl(𝑒 𝑖 ) generate the A-module A𝑛 /Im(𝜑), the composition
𝜃˜ ◦ 𝜓˜ is the identity of A𝑛 /Im(𝜑). On the other hand, one has
˜ 𝑓𝑖 ) = 𝜓(cl(𝑒
𝜓˜ ◦ 𝜃( ˜ 𝑖 )) = 𝜓(𝑒 𝑖 ) = 𝑓𝑖

for each 𝑖, so that 𝜓˜ ◦ 𝜃˜ is the identity of (K𝑛 )𝑢 . Finally, 𝜓˜ is an

isomorphism and (K𝑛 )𝑢 is isomorphic to A𝑛 /Im(𝜑).
Since the matrix of 𝜑, as an endomorphism of A𝑛 , is equal to XI𝑛 − U,
we see that for all integers 𝑝, one has
(Δ𝑝 ) = J𝑝 (XI𝑛 − U) = Fit𝑛−𝑝 ((K𝑛 )𝑢 ).
On the other hand, the K[X]-module associated to a companion
matrix CP is isomorphic to K[X]/(P), so that (K𝑛 )𝑢 has an alternative
presentation as the quotient of A𝑟 by the image of the diagonal matrix V
with diagonal entries (P1 , . . . , P𝑟 ), as in theorem 5.6.5. By example 3.9.2,
one then has


 (1) if 𝑟 − 𝑛 + 𝑝 6 0;
Fit𝑛−𝑝 ((K𝑛 )𝑢 ) = J𝑟−𝑛+𝑝 (V) = (P1 . . . P𝑟−𝑛+𝑝 )


if 1 6 𝑟 − 𝑛 + 𝑝 6 𝑟;

 (0) if 𝑟 − 𝑛 + 𝑝 > 𝑟.


We thus obtain


 (1) if 𝑝 6 𝑛 − 𝑟;


(Δ𝑝 ) = (P1 . . . P𝑝+𝑟−𝑛 ) if 𝑛 − 𝑟 + 1 6 𝑝 6 𝑛;

 (0) if 𝑝 > 𝑛.


This implies that Δ1 = · · · = Δ𝑛−𝑟 = 1 and Δ𝑛−𝑟+𝑝 = (P1 . . . P𝑝 ) for
1 6 𝑝 6 𝑟 and concludes the proof of the proposition, with 𝑚 = 𝑛 − 𝑟. 
All the preceding theory has the following application, which is
important in subsequent developments of algebra. Note that it would
not be so obvious to prove it without the theory of invariant factors,
especially if the field K is finite.

Corollary (5.6.8). — Let K be a field and let U and V be two matrices in M𝑛 (K).
Let K → L be an extension of fields; assume that U and V are conjugate as
262 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

matrices of M𝑛 (L), that is to say, there exists a matrix P ∈ GL𝑛 (L) such that
V = P−1 UP. Then, U and V are conjugate over K: there exists an invertible
matrix Q ∈ GL𝑛 (K) such that V = Q−1 UQ.

Proof. — Let (P1 , . . . , P𝑟 ) be the family of invariant factors of U, viewed

as a matrix in M𝑛 (K). Therefore, there exists a basis of W = K𝑛 in
which the matrix of U has a block-diagonal form, the blocks being the
companion matrices with characteristic polynomials P1 , . . . , P𝑟 . The
matrix that expresses this new basis of W gives also a basis of L𝑛 in which
the matrix of U is exactly the same block-diagonal matrix of companion
matrices. In other words, when one “extends the scalars” from K to L,
the invariant factors of U are left unchanged. Since, as matrices with
coefficients in K, the invariant factors of U coincide with those of V, the
matrices U and V have the same invariant factors as matrices in M𝑛 (K),
hence are conjugate. 
1

Theorem (5.6.9) (Jordan decomposition). — Let K be an algebraically closed

field. Let V be a finite dimensional K-vector space and let 𝑢 be an endomorphism
of V. There exists a basis of V in which the matrix of 𝑢 is block-diagonal, each
block being of the form (“Jordan block”)
𝜆 1 0
­ 𝜆 ... ®
© ª
J𝑛 (𝜆) = ­­ . . . 1 ® ∈ M𝑛 (K),
®
­ ®
«0 𝜆¬
where 𝜆 is an eigenvalue of 𝑢. Moreover, two endomorphisms 𝑢 and 𝑢 0 are
conjugate if and only if, for every 𝜆 ∈ K, the lists of sizes of Jordan blocks
(written in increasing order) corresponding to 𝜆 for 𝑢 and 𝑢 0 coincide.

Proof. — Let us first write down the primary decomposition of the

K[X]-module V𝑢 . Since K is algebraically closed, the irreducible monic
polynomials in K[X] are the polynomials X − 𝜆, for 𝜆 ∈ K. For 𝜆 ∈ K, let
V𝜆 be the (X − 𝜆)-primary component of V𝑢 ; it is the set of all 𝑣 ∈ V such
that there exists 𝑛 ∈ N with (X − 𝜆)𝑛 · 𝑣 = 0, that is, (𝑢 − 𝜆 id)𝑛 (𝑣) = 0.

1À revoir à partir d’ici !

 5.6. APPLICATION: ENDOMORPHISMS OF A FINITE DIMENSIONAL VECTOR SPACE 263

In other words, V𝜆 is the subspace of V known as the characteristic

subspace of 𝑢 associated to the eigenvalue 𝜆. (It is indeed nonzero if
and only if 𝜆 is an eigenvalue.) By construction, V𝜆 is a K[X]-submodule
of V𝑢 , which means that this subspace of V is stable under the action
of 𝑢.
The (X − 𝜆)-primary component V𝜆 of V is isomorphic, as a K[X]-
module, to a direct sum
𝑠
Ê
K[X]/((X − 𝜆)𝑛 𝑖 ),
𝑖=1

for some unique increasing family (𝑛1 , . . . , 𝑛 𝑠 ) of integers. Let us observe

the following lemma:

Lemma (5.6.10). — The K[X]-module corresponding to the Jordan matrix

J𝑛 (𝜆) is isomorphic to K[X]/(X − 𝜆)𝑛 .

Proof. — Let W = K𝑛 and let 𝑤 be the endomorphism of W given by

the matrix J𝑛 (𝜆). Let (𝑒1 , . . . , 𝑒 𝑛 ) be the canonical basis of K𝑛 ; one has
𝑤(𝑒 𝑚 ) = 𝜆𝑒 𝑚 + 𝑒 𝑚−1 if 𝑚 ∈ {2, . . . , 𝑛}, and 𝑤(𝑒1 ) = 𝜆𝑒1 . In other words,
if we view W as a K[X]-module W𝑤 via 𝑤, one has (X − 𝜆) · 𝑒 𝑚 = 𝑒 𝑚−1
for 2 6 𝑚 6 𝑛 and (X − 𝜆) · 𝑒1 = 0; in other words, (X − 𝜆)𝑚 · 𝑒 𝑛 = 𝑒 𝑛−𝑚
for 𝑚 ∈ {0, . . . , 𝑛 − 1} and (X − 𝜆)𝑛 · 𝑒 𝑛 = 0.
There is a unique morphism 𝜑 of K[X]-modules from K[X] to W𝑤 such
that 𝜑(1) = 𝑒 𝑛 ; it is given by 𝜑(P) = P(X) · 𝜑(1) = P(𝑤)(𝑒 𝑛 ). The above
formulae show that the vectors 𝑒 𝑚 belong to the image of 𝜑, hence 𝜑 is
surjective. Moreover, these formulae also show that, (X𝜆 )𝑛 ∈ Ker(𝜑), so
that 𝜑 defines a morphism 𝜑˜ from K[X]/(X − 𝜆)𝑛 to W𝑣 . The morphism
𝜑˜ has the same image as 𝜑, hence is surjective. Since, K[X]/(X − 𝜆)𝑛 has
dimension 𝑛 as a K-vector space, as well as W, the morphism 𝜑˜ must be
an isomorphism. 
Let us now return to the proof of the Jordan decomposition. Thanks
to the lemma, we see that the subspace V𝜆 has a basis in which the
matrix of 𝑢 (restricted to V𝜆 ) is block-diagonal, the blocks being Jordan
matrices J𝑛1 (𝜆), . . . , J𝑛 𝑠 (𝜆). Moreover, the polynomials (X − 𝜆)𝑛 𝑖 (ordered
by nondecreasing degrees) are the invariant factors of V𝜆 . This shows
the existence of the Jordan decomposition.
264 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

Conversely, the Jordan blocks determine the invariant factors of each

primary component of V𝑢 , so that two endomorphisms 𝑢 and 𝑢 0 are
conjugate if and only if for each 𝜆 ∈ K, the lists of sizes of Jordan blocks
corresponding 𝜆 for 𝑢 and 𝑢 0 coincide. 

Exercises
1-exo603) a) Show that the permutation matrix P associated to the transposition
(1, 2) ∈ 𝔖2 belongs to the subgroup of GL2 (Z) generated by E2 (Z) together with the
matrices D𝑖 (−1) for 𝑖 ∈ {1, 2}.
b) Does it belong to the subgroup E2 (Z)?
c) Prove that the Weyl group of GL𝑛 (Z) is contained in the subgroup generated
by E𝑛 (Z) together with the matrices D𝑖 (−1).
2-exo614) Let A be a commutative ring.
a) Show that E𝑛 (A) is a normal subgroup of the group GE𝑛 (A); show also that
GE𝑛 (A) = E𝑛 (A) · {D1 (𝑎)} so that GE𝑛 (A) is a semidirect product of E𝑛 (A) by A× .
b) Same question, replacing GE𝑛 (A) by GL𝑛 (A) and E𝑛 (A) by SL𝑛 (A).
3-exo615) Let K be a division ring and let D = D(K× ) be the derived group of the
multiplicative group K× , that is, the quotient of K× by its subgroup generated by all
commutators [𝑎, 𝑏] = 𝑎𝑏𝑎 −1 𝑏 −1 , for 𝑎, 𝑏 ∈ K× . Write 𝜋 for the canonical projection
from K× to D.
Let 𝑛 be an integer and let V be a right K-vector space of dimension 𝑛. Let B(V) be
the set of bases of V and let Ω(V) be the set of all maps 𝜔 : B(V) → D satisfying the
following properties:
𝜔(𝑣1 𝑎 1 , . . . , 𝑣 𝑛 𝑎 𝑛 ) = 𝜔(𝑣1 , . . . , 𝑣 𝑛 )𝜋(𝑎 1 . . . 𝑎 𝑛 )
and
𝜔(𝑣1 , . . . , 𝑣 𝑖−1 , 𝑣 𝑖 + 𝑣 𝑗 , 𝑣 𝑖+1 , . . . , 𝑣 𝑛 ) = 𝜔(𝑣1 , . . . , 𝑣 𝑛 )
for (𝑣 1 , . . . , 𝑣 𝑛 ) ∈ B(V), 𝑎 1 , . . . , 𝑎 𝑛 ∈ K× and 𝑖, 𝑗 ∈ {1, . . . , 𝑛} such that 𝑖 ≠ 𝑗.
a) For 𝜔 ∈ Ω(V), (𝑣1 , . . . , 𝑣 𝑛 ) ∈ B(V) and 𝜎 ∈ 𝔖𝑛 , prove that
𝜔(𝑣 𝜎(1) , . . . , 𝑣 𝜎(𝑛) ) = 𝜋(𝜀(𝜎))𝜔(𝑣1 , . . . , 𝑣 𝑛 ).
b) For 𝛽 = (𝑣1 , . . . , 𝑣 𝑛 ) ∈ B(V), let 𝛽 𝑖 = (𝑣1 , . . . , 𝑣 𝑖−1 , 𝑣 𝑖+1 , . . . , 𝑣 𝑛 ). Let W be a
hyperplane of V and let 𝑣 ∈ V W. Let 𝜑 ∈ Ω(W). Prove that there exists a unique
𝜔 ∈ Ω(V) such that
𝜔(𝑣1 , . . . , 𝑣 𝑛−1 , 𝑣) = 𝜑(𝑣 1 , . . . , 𝑣 𝑛−1 )
for every (𝑣1 , . . . , 𝑣 𝑛−1 ) ∈ B(W).
c) Let (𝑣1 , . . . , 𝑣 𝑛 ) ∈ B(V) and let 𝑎 ∈ D. Prove that there exists a unique 𝜔 ∈ Ω(V)
such that 𝜔(𝑣1 , . . . , 𝑣 𝑛 ) = 𝑎.
 EXERCISES 265

d) Let U ∈ GL(V). Prove that there exists a unique element 𝛿(U) ∈ D such that
𝜔(U(𝑣1 ), . . . , U(𝑣 𝑛 )) = 𝛿(U)𝜔(𝑣1 , . . . , 𝑣 𝑛 ) for every (𝑣1 , . . . , 𝑣 𝑛 ) ∈ B(V). Prove that the
map 𝛿 : GL(V) → D is a morphism of groups.
When K is a commutative, then D = K× , and the morphism 𝛿 : GL𝑛 (K) → K× is the
determinant. For this reason, the map 𝛿 is called the noncommutative determinant. This
construction is due to J. Dieudonné.
𝑎 𝑏 . If 𝑎 ≠ 0, then 𝛿(U) = 𝜋(𝑎𝑑 − 𝑎𝑐𝑎 −1 𝑏); if 𝑎 = 0,


e) Let U ∈ GL2 (K) have matrix 𝑐 𝑑
then 𝛿(U) = 𝜋(−𝑐𝑏).
4-exo604) Let N be a positive integer.
a) Show that the group SL𝑛 (Z/NZ) is generated by its elementary matrices.
b) Deduce that the canonical morphism SL𝑛 (Z) → SL𝑛 (Z/NZ) (reduction mod. N) is
surjective.
c) Is the analogous morphism GL𝑛 (Z) → GL𝑛 (Z/NZ) surjective?
5-exo608) a) Let 𝑢 ∈ Z𝑛 be a vector whose entries are coprime. Show by induction
on the smallest nonzero entry of 𝑢 that there exists a matrix M ∈ SL𝑛 (Z) whose first
column is 𝑢.
b) Let M be a matrix with 𝑛 rows and 𝑝 columns with entries in a principal ideal
domain A. Assume that 𝑝 6 𝑛.
Show that it is possible to complete M into a matrix P ∈ GL𝑛 (A) (that is, there exists
such a matrix P whose first 𝑝 columns give M) if and only if the ideal Δ𝑝 (M) generated
by all minors of size 𝑝 of M is equal to (1).
6-exo607) Let A be a principal ideal domain and let K be its fraction field.
a) Let 𝑢 = (𝑎 1 , . . . , 𝑎 𝑛 ) ∈ A𝑛 . Let 𝑎 ∈ A be a generator of the ideal (𝑎 1 , . . . , 𝑎 𝑛 ). Show
that there exists a matrix M ∈ E𝑛 (A) such that M𝑢 = (𝑎, 0, . . . , 0).
b) Show that the group E𝑛 (A) acts transitively on the set of lines of K𝑛 .
c) Let 𝑢1 = (1, 2) and 𝑢2 = (2, 1) in Q2 , let D1 = Q𝑢1 and D2 = Q𝑢2 be the lines in Q2
generated by 𝑢1 and 𝑢2 . Show that there does not exist a matrix M ∈ GL2 (Z) such that
M(D1 ) and M(D2 ) are the two coordinate axes.
d) Show that there does not exist a matrix P ∈ GL2 (𝑘[X, Y]) which maps the line
generated by the vector (X, Y) ∈ 𝑘(X, Y)2 to the line generated by the vector (1, 0).
7-exo160) Let A be a principal ideal domain, let K be its fraction field.
a) Let X be a nonzero vector in K𝑛 . Show that there exists a matrix M ∈ GL𝑛 (A)
whose first column is proportional to X.
b) Show that each square matrix M ∈ M𝑛 (K) can be written as a product PB where
P ∈ GL𝑛 (A) and B is an upper-triangular matrix with entries in K. (Argue by induction).
c) Numerical example: In the case A = Z, give such an explicit decomposition for the
matrix
1/2 1 −1/4
M = ­2/5 2 2/3 ® .
© ª

«3/4 1/7 −1 ¬
266 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

8-exo133) Let A be a principal ideal domain and let M be a free finitely generated
A-module.
a) For any 𝑚 ∈ M, show that the following properties are equivalent:
(i) The vector 𝑚 belongs to some basis of M;
(ii) There exists 𝑓 ∈ M∨ such that 𝑓 (𝑚) = 1;
(iii) In any basis of M, the coordinates of M are coprime;
(iv) There exists a basis of M in which the coordinates of 𝑚 are coprime;
(v) For any 𝑎 ∈ A and 𝑚 0 ∈ M such that 𝑚 = 𝑎𝑚 0, one has 𝑎 ∈ A× ;
(vi) For any 𝑎, 𝑎 0 ∈ A and 𝑚 0 ∈ M such that 𝑎𝑚 = 𝑎 0 𝑚 0 and 𝑎 ≠ 0, then 𝑎 0
divides 𝑎.
Such a vector is said to be primitive.
b) Show that any nonzero vector is a multiple of a primitive vector.
c) Example : A = Z, M = Z4 , 𝑚 = (126, 210, 168, 504).
9-exo046) Let A be a principal ideal domain, let L, M be two finitely generated A-
modules. Show that HomA (L, M) is a finitely generated A-module.
10-exo617) Le A be a principal ideal domain and let M be a finitely generated A-
module. Let (𝑑1 ), . . . , (𝑑𝑛 ) be its invariant factors, where 𝑑1 , . . . , 𝑑𝑛 are non-units and
𝑑 𝑖 divides 𝑑 𝑖+1 for 1 6 𝑖 < 𝑛.
a) Let 𝑚 be a vector of M whose annihilator equals (𝑑𝑛 ). Show that the submodule A𝑚
of M generated by 𝑚 admits a direct summand in M.
b) One assumes that A = Z and M = (Z/𝑝Z) ⊕ (Z/𝑝 2 Z). Give a necessary and
sufficient condition on a vector 𝑚 ∈ M for the subbmodule A𝑚 to possess a direct
summand.
11-exo601) Let 𝑞(𝑥, 𝑦) = 𝑎𝑥 2 + 2𝑏𝑥𝑦 + 𝑐𝑦 2 be a positive definite quadratic form with
real coefficients. Let 𝑚 = inf(𝑥,𝑦)∈Z2 {0} 𝑞(𝑥, 𝑦).
a) Show that there exists a nonzero vector 𝑒1 ∈ Z2 such that 𝑚 = 𝑞(𝑒1 ).
b) Show that there exists 𝑒2 ∈ Z2 such that (𝑒1 , 𝑒2 ) is a basis of Z2 .
p
c) Deduce from the inequality 𝑞(𝑒2 + 𝑛𝑒1 ) > 𝑞(𝑒1 ) that 𝑚 6 2 (𝑎𝑐 − 𝑏 2 )/3.
12-exo610) Let A be a commutative É𝑛 ring and let I1 , . . . , I𝑛 be ideals of A such that
I1 ⊂ I2 ⊂ · · · ⊂ I𝑛 ( A. Let M = 𝑖=1 A/I 𝑖 .
a) Let J be a maximal ideal of A containing I𝑛 . Endow the A-module M/JM with the
structure of an (A/J)-vector space. Show that its dimension is equal to 𝑛.
b) Show that any generating family of M has at least 𝑛 elements.
c) One assumes that A is a principal ideal domain. Let M be a finitely generated
A-module and let (𝑑1 ), . . . , (𝑑𝑟 ) be its invariant factors. Show that any generating set
in M has at least 𝑟 elements.
13-exo616) The goal of this exercise is to provide another proof of theorem 5.4.3 that
does not use matrix operations.
Let A be a principal ideal domain, let M be a free finitely generated A-module and
let N be a submodule of M.
 EXERCISES 267

a) Show that there exists a linear form 𝑓 on M such that the ideal I = 𝑓 (N) of A is
maximal among the ideals of this form (that is, there does not exist 𝑔 ∈ M∨ such that
𝑓 (N) ( 𝑔(N)).
Let M0 be the kernel of 𝑓 and let N0 = N ∩ M0.
b) Show that for any linear form 𝑓 0 on M0 and any vector 𝑚 ∈ N0, one has 𝑓 0(𝑚) ∈ I.
c) Using the same induction method as the one used for proving proposition 5.4.1,
show that there exists a basis (𝑒1 , . . . , 𝑒 𝑛 ), an integer 𝑟 ∈ {0, . . . , 𝑛} and elements
𝑑1 , . . . , 𝑑𝑟 of A such that 𝑑 𝑖 divides 𝑑 𝑖+1 for any integer 𝑖 < 𝑟 and such that (𝑑1 𝑒1 , . . . , 𝑑𝑟 𝑒 𝑟 )
is a basis of N.
14-exo086) Let A be a principal ideal domain and let M be a finitely generated A-module.
a) Justify the existence of an integer 𝑠, elements 𝑚1 , . . . , 𝑚 𝑠 of M, elements 𝑑1 , . . . , 𝑑 𝑠
of A suchÉ that 𝑑 𝑖 divides 𝑑 𝑖+1 for 1 6 𝑖 < 𝑠 − 1 such that (𝑑 𝑖 ) is the annihilator of 𝑚 𝑖
𝑠
and M = 𝑖=1 A𝑚 𝑖 .
b) Let 𝑖 ∈ {1, . . . , 𝑠}. Show that there exists 𝑢𝑖 ∈ EndA (M) such that
𝑢𝑖 (𝑚1 ) = · · · = 𝑢𝑖 (𝑚 𝑠−1 ) = 0, 𝑢𝑖 (𝑚 𝑠 ) = 𝑚 𝑖 .
c) Let 𝑢 be any element in the center of EndA (M). Show that there exists 𝑎 ∈ A such
that 𝑢(𝑚) = 𝑎𝑚 for every 𝑚 ∈ M.
d) Let 𝑢 : M → M be an additive map such that 𝑢 ◦ 𝑣 = 𝑣 ◦ 𝑢 for every 𝑣 ∈ EndA (M).
Show that there exists 𝑎 ∈ A such that 𝑢(𝑚) = 𝑎𝑚 for every 𝑚 ∈ M.
15-exo159) a) Give the list of all abelian groups of order 16 (up to isomorphism).
b) Give the list of all abelian groups of order 45 (up to isomorphism).
16-exo157) Let M be the set of triples (𝑥, 𝑦, 𝑧) ∈ Z3 such that 𝑥 + 𝑦 + 𝑧 is even.
a) Show that M is a submodule of Z3 ; prove that it is free of rank 3 by exhibiting 3
linearly independent vectors in M.
b) Construct a basis of M.
c) Exhibit a linear map 𝑓 : M → Z/2Z such that M = Ker( 𝑓 ). Conclude that Z3 /M is
isomorphic to Z/2Z.
17-exo137) Let L be the set of vectors (𝑥, 𝑦, 𝑧) ∈ Z3 such that
𝑥 − 3𝑦 + 2𝑧 ≡ 0 (mod 4) et 𝑥+𝑦+𝑧≡0 (mod 6).
a) Show that L is a free Z-submodule of Z3 . What is its rank?
b) Construct a morphism 𝑓 from Z3 /L to (Z/4Z) × (Z/6Z) such that L = Ker( 𝑓 ).
Prove that 𝑓 is surjective; conclude that Z3 /L is isomorphic to (Z/4Z) × (Z/6Z).
c) Compute integers 𝑑1 , 𝑑2 , 𝑑3 and a basis (𝑒1 , 𝑒2 , 𝑒3 ) of Z3 such that 𝑑1 |𝑑2 |𝑑3 and such
that (𝑑1 𝑒1 , 𝑑2 𝑒2 , 𝑑3 𝑒3 ) is a basis of L.
18-exo211) a) Let G be a finite abelian group; let 𝑛 be the smallest positive integer such
that 𝑛G = 0. Show that there exists an element 𝑔 ∈ G of exact order 𝑛.
b) Let K be a field. Recover the fact that a finite subgroup of K∗ is cyclic.
268 CHAPTER 5. MODULES OVER PRINCIPAL IDEAL RINGS

c) Let G be the multiplicative subgroup {±1, ±𝑖, ±𝑗, ±𝑘} of Hamilton’s quaternions H.
Prove that G is not cyclic (it suffices to remark that G is not commutative!). This shows
that in the previous question, the commutativity assumption on K is crucial.
19-exo635) a) Let G, H be finitely generated abelian groups such that G × G ' H × H.
Show that G ' H.
b) Let G, G0 , H be finitely generated abelian groups such that G × H ' G0 × H. Show
that G ' G0.
c) Show that the finite generation hypothesis is necessary by exhibiting abelian
groups for which the two previous results fail.
20-exo632) If G is a finite abelian group, one writes G∗ for the set of all group morphisms
from G to C∗ . An element of G∗ is called a character of G.
a) Show that G∗ is a group for pointwise multiplication. Show also that it is finite.
b) In this question, one assumes that G = Z/𝑛Z, for some positive integer 𝑛. If
𝜒 ∈ G∗ , show that 𝜒(1) is an 𝑛th root of unity Prove that the map 𝜒 ↦→ 𝜒(1) from G∗
to C∗ is an isomorphism from G∗ to the group of 𝑛th roots of unity in C∗ .
Construct an isomorphism from G to G∗ .
c) If G = H × K, show that G∗ is isomorphic to H∗ × K∗ .
d) Show that for any finite abelian group, G∗ is isomorphic to G.
21-exo633) Let G be a finite abelian group. For any function 𝑓 : G → C and any
character 𝜒 ∈ G∗ , Õ
𝑓ˆ(𝜒) = 𝑓 (𝑔)𝜒(𝑔).
𝑔∈G

This defines a function 𝑓ˆ on G∗ , called the Fourier transform of 𝑓 .

a) Compute 𝑓ˆ when 𝑓 = 1 is the constant function 𝑔 ↦→ 1 on G.
b) For any 𝑔 ∈ G, show that
(
Õ Card(G) if 𝑔 = 0;
𝜒(𝑔) =
𝜒∈G∗
0 otherwise.

c) Let 𝑓 : G → C be a function. Show that for any 𝑔 ∈ G, one has the following
Fourier inversion formula:
1 Õ
𝑓 (𝑔) = 𝑓ˆ(𝜒)𝜒(𝑔)−1 .
Card(G) ∗ 𝜒∈G

d) Show also the Plancherel formula:

Õ 2 Õ
𝑓ˆ(𝜒) = Card(G) 𝑓 (𝑔) .
2

𝜒∈G∗ 𝑔∈G

22-exo636) Let A be a matrix in M𝑛 (Z), viewed as an endomorphism of Z𝑛 . Let

M = Z𝑛 /A(Z𝑛 ). Show that M is a finite abelian group if and only if det(A) ≠ 0. In that
case, show also that Card(M) = |det(A)|.
 EXERCISES 269

23-exo634) Let G be an abelian group and let H be a subgroup of G. One assumes that
H is divisible: for any ℎ ∈ H and any strictly positive integer 𝑛, there exists ℎ 0 ∈ H that
𝑛 ℎ 0 = ℎ.
a) Let K be a subgroup of G such that K ∩ H = {0} and K + H ≠ G. Let 𝑔 ∈ G (K + H)
and let K0 = K + Z𝑔. Show that K0 ∩ H = {0} and K0 ) K.
b) Using Zorn’s lemma, show that there is a maximal subgroup K ⊂ G such that
K ∩ H = {0}. Using the previous question, show that G = K + H.
c) Show that G ' H × (G/H).
24-exo641) Let K be a field, let V be a finite dimensional K-vector space and let
𝑢 ∈ EndK (V). Show that an endomorphism of V which commutes with every
endomorphism which commutes with 𝑢 is a polynomial in 𝑢. (Introduce the K[X]-
module structure on V which is defined by 𝑢.)
25-exo613) Let 𝑘 be a field. Determine all conjugacy classes of matrices M ∈ M𝑛 (𝑘)
satisfying the indicated properties.
a) The characteristic polynomial of M is X3 (X − 1).
b) The minimal polynomial of M is X(X − 1).
c) One has (M − I𝑛 )2 = 0.
CHAPTER 6

FURTHER RESULTS ON MODULES

This chapter explores more advanced results in the theory of modules.

Despite its apparent simplicity, Nakayama’s lemma allows to efficiently
transfer results which are initially proved by reasoning modulo a maximal ideal,
or modulo the Jacobson radical. It is a very important tool in commutative
algebra.
The next three sections are devoted to three different finiteness conditions
for modules, all of them phrased in terms of chains of submodules, that is,
totally ordered sets of submodules: we say that a module has finite length if the
cardinality of all of these chains is bounded above, that it is noetherian if any
chain has a maximal element, and artinian if any chain has a minimal element.
Modules of finite length are probably the closest to finite dimensional vector
spaces. The length function is an integer-valued function that behaves similarly
to the dimension, and it is as useful.
The noetherian property, named in honor of Emmy Noether, is of a particular
importance in commutative algebra, as it is often guarantees that given modules
are finitely generated. As applications of this notion, I present two fundamental
finiteness theorems of Hilbert. First, the finite basis theorem asserts that
polynomial rings over a noetherian ring are polynomial rings. Using a method
due to Emil Artin and John Tate, I also prove that when a finite group acts on a
finitely generated algebra, the subalgebra of invariant elements is itself finitely
generated; this theorem is one of the starting results in the theory of invariants.
The role of the artinian property, named in honor of Emil Artin, is more
subtle and will be less obvious in this book. (Its strength would require the
introduction of other techniques, such as completion, which are not studied
here.) Nevertheless, I prove Akizuki’s theorem that asserts that a ring is artinian
if and only if it has finite length; in the case of commutative rings, artinian
272 CHAPTER 6. FURTHER RESULTS ON MODULES

rings can be also be characterized by the fact that they are noetherian and that
all of its prime ideals are maximal.
The last two sections consider commutative rings only and have an implicit
geometric content in algebraic geometry. I define the support of a module, and
its associated ideals. My exposition of associated ideals is slightly different
from the classical one, which is limited to modules over noetherian rings;
contrary to a plausible expectation, I feel that it makes the first results easier,
and their proofs more natural.
I then discuss the existence of primary decomposition, a theorem due
to Emmy Noether and Emmanuel Lasker that gives an approximation of the
decomposition into powers of primes. I conclude the chapter by proving Krull’s
intersection theorem.

6.1. Nakayama’s lemma

Recall (definition 2.1.5) that the Jacobson radical J of a ring A is the
intersection of all maximal left ideals of A; it is a two sided ideal of A
and 1 + 𝑎 is invertible for every element 𝑎 ∈ J (lemma 2.1.6).

Theorem (6.1.1) (Nakayama’s lemma). — Let A be a ring, let J be the Jacobson

radical of A. Let M be a finitely generated A-module such that M = MJ. Then
M = 0.

Proof. — We prove the theorem by induction on the number of elements

of a generating set of M.
If M is generated by 0 element, then M = 0.
Let then 𝑛 > 1 and assume that the theorem holds for any A-module
which is generated by (𝑛 − 1) elements. Let M be an A-module which
is generated by 𝑛 elements, say 𝑒1 , . . . , 𝑒 𝑛 , and such that M = MJ. Let
N = M/𝑒1 A. Then, N is generated by the classes of 𝑒2 , . . . , 𝑒 𝑛 and one
has N = NJ. Consequently, N = 0. In other words, M = 𝑒1 A.
Since M = MJ, there is 𝑎 ∈ J such that 𝑒1 = 𝑒1 𝑎, hence 𝑒1 (1 − 𝑎) = 0.
Since 𝑎 ∈ J, 1 − 𝑎 is right invertible in A and 𝑒1 = 0. This implies that
M = 0. 
 6.1. NAKAYAMA’S LEMMA 273

Corollary (6.1.2). — Let A be a ring, let J be its Jacobson radical. Let M be a

finitely generated A-module, let N be a submodule of M such that M = N + MJ.
Then, M = N.

Proof. — Setting P = M/N, the hypothesis M = N + MJ implies that

P = PJ. Consequently, P = 0 and M = N. 

Corollary (6.1.3). — Let A be a ring and let J be its Jacobson radical. Let
U ∈ M𝑛 (A) be a matrix and let U be its image in M𝑛 (A/J). If U is right
invertible (resp. left invertible, resp. invertible), then U is right invertible (resp.
left invertible, resp. invertible).

Proof. — Let M = A𝑛𝑑 and let 𝑢 be the endomorphism of M represented

by the matrix U in the canonical basis (𝑒1 , . . . , 𝑒 𝑛 ) of A𝑛𝑑 . If U is right
invertible, the endomorphism 𝑢 of (A/J)𝑛𝑑 it defines is surjective, so that,
for every 𝑚 ∈ M, there exists 𝑚 0 and 𝑚 00 ∈ M such that 𝑚 = 𝑢(𝑚 0) + 𝑚 00
and all coordinates of 𝑚 00 belong to J. In other words, M = 𝑢(M) + MJ.
By the preceding corollary, M = 𝑢(M), so that 𝑢 is surjective. For every
𝑖 ∈ {1, . . . , 𝑛}, choose 𝑚 𝑖 ∈ M such that 𝑢(𝑚 𝑖 ) = 𝑒 𝑖 and let 𝑣 be the
endomorphism of M such that 𝑣(𝑒 𝑖 ) = 𝑚 𝑖 . Then, 𝑢 ◦ 𝑣(𝑒 𝑖 ) = 𝑢(𝑚 𝑖 ) = 𝑒 𝑖
for every 𝑖 so that 𝑢 ◦ 𝑣 = idM . In particular, 𝑢 is right invertible. The
matrix V of 𝑣 is then a right inverse to U.
The case of left invertible matrices is proven analogously, or by consid-
ering the transposed matrices, or by working in the opposite ring; the
case of invertible matrices follows. 

Corollary (6.1.4). — Let A be a commutative ring, let I be an ideal of A. Let

M be a finitely generated A-module, let N be a submodule of M such that
M = N + IM. Then, there exists 𝑎 ∈ I such that (1 + 𝑎)M ⊂ N.

Proof. — Let S = 1 + I = {1 + 𝑎 ; 𝑎 ∈ I}; it is a multiplicative subset

of A. The equation M = N + MI implies the equality S−1 M = S−1 N +
S−1 M · S−1 I of S−1 A-modules. For every 𝑎 ∈ S−1 I, 1 + 𝑎 is invertible
in S−1 A. Indeed, write 𝑎 = 𝑢/(1 + 𝑣), for some 𝑢, 𝑣 ∈ I; then one has
1 + 𝑎 = (1 + (1 + 𝑣)𝑢)/(1 + 𝑣), which is invertible since 1 + (1 + 𝑣)𝑢 ∈ 1 + I.
Consequently, S−1 I is contained in the Jacobson radical of S−1 A. Since
274 CHAPTER 6. FURTHER RESULTS ON MODULES

S−1 M is finitely generated (corollary 3.6.9), the preceding corollary

implies that S−1 M = S−1 N.
It follows that for every 𝑚 ∈ M, there exists 𝑎 ∈ S such that 𝑚𝑎 ∈ N.
More precisely, let (𝑚1 , . . . , 𝑚𝑛 ) be a generating family of M; for every
𝑖 ∈ {1, . . . , 𝑛}, let 𝑎 𝑖 ∈ S be such that 𝑚 𝑖 𝑎 𝑖 ∈ N, and set 𝑎 = 𝑎1 . . . 𝑎 𝑛 .
Then 𝑚 𝑖 𝑎 ∈ N for every 𝑖 ∈ {1, . . . , 𝑛}, hence M𝑎 ⊂ N. 

Remark (6.1.5). — One can also prove the corollary by a similar induction
to the one of Nakayama’s theorem.

Corollary (6.1.6). — Let A be a commutative ring, let M be a finitely generated

A-module and let 𝑢 ∈ EndA (M) be an endomorphism of M. If 𝑢 is surjective,
then 𝑢 is an isomorphism.

Proof. — Let us use 𝑢 to view M as an A[X]-module, setting P · 𝑚 =

P(𝑢)(𝑚) for every P ∈ A[X] and every 𝑚 ∈ M (remark 3.2.6). Then M
is finitely generated as an A[X]-mdule. Since 𝑢 is surjective, one has
M = 𝑢(M) = X · M = (X)M. By the preceding corollary, there exists
a polynomial P ∈ A[X] such that (1 − XP(X)) · M = 0. In other words,
0 = 𝑚 − 𝑢(P(𝑢)(𝑚)) for every 𝑚 ∈ M, so that idM = 𝑢 ◦ P(𝑢). Since
P(𝑢) ◦ 𝑢 = 𝑢 ◦ P(𝑢) = (XP)(𝑢), this shows that P(𝑢) is the inverse of 𝑢,
hence 𝑢 is an isomorphism. 

6.2. Length
Definition (6.2.1). — Let A be a ring. One says that an A-module is simple if
it is nonzero and if its only submodules are 0 and itself.

Examples (6.2.2). — a) The module 0 is not simple.

b) Let A be a ring and let I be a right ideal of A. The bijection
V ↦→ cl−1
I (V) between A-submodules of A𝑑 /I and submodules of A 𝑑
containing I preserves the containment relation. Consequently, the
A-module A𝑑 /I is simple if and only if I ≠ A and the only right ideals
of A which contain I are I and A. In other words, A𝑑 /I is a simple right
A-module if and only if I is a maximal right ideal of A.
 6.2. LENGTH 275

c) Let M be a simple A-module and let 𝑚 be a nonzero element of M.

The set 𝑚A of all multiples of M is a nonzero submodule of M. Since M
is simple, one has 𝑚A = M.
The set I of all 𝑎 ∈ A such that 𝑚𝑎 = 0 is is a right ideal of A and the
map 𝑎 ↦→ 𝑚𝑎 induces an isomorphism from A𝑑 /I to M. Consequently, I
is a maximal right ideal of A.
d) Assume that A is a division ring. Then, a simple A-module is
nothing but a 1-dimensional A-vector space.
e) Let A be a ring and let I be a two-sided ideal of A. In the identification
between A-modules annihilated by I and (A/I)-modules, A-submodules
correspond to (A/I)-submodules. Consequently, an A-module which
is annihilated by I is a simple A-module if and only if it is simple as an
(A/I)-module.
f) Let M be an A-module and let N be a submodule of M. The
submodules of M/N correspond to the submodules of M containing N.
Consequently, M/N is a simple A-module if and only if N is a maximal
submodule of M, in the sense that M ≠ N and the only submodules of M
containing N are N and M.

Proposition (6.2.3) (Schur lemma). — Let A be a ring and let 𝑢 : M → N a

nonzero morphism of A-modules.
a) If M is simple, then 𝑢 is injective.
b) If N is simple, then 𝑢 is surjective.
c) If M and N are both simple, then 𝑢 is an isomorphism.

Proof. — The image of 𝑢 is a nonzero submodule of N; if N is simple,

then Im(𝑢) = N and 𝑢 is surjective. The kernel of 𝑢 is a submodule of M,
distinct from M; if M is simple, then Ker(𝑢) = 0 and 𝑢 is injective. If both
M and N are simple, then 𝑢 is bijective, hence an isomorphism. 

Corollary (6.2.4). — The ring of endomorphisms of a simple A-module is a

division ring.

Proof. — Let M be a simple A-module. By the proposition, any nonzero

element of EndA (M) is an isomorphism, hence is invertible in EndA (M).
276 CHAPTER 6. FURTHER RESULTS ON MODULES

Moreover, M ≠ 0, so that idM ≠ 0 and EndA (M) is not the zero ring. This
proves that EndA (M) is a division ring. 

Definition (6.2.5). — Let A be a ring and let M be an A-module. A chain of

submodules of M is a sequence (M0 , . . . , M𝑛 ) of submodules of M such that
M0 ( M1 · · · ( M𝑛 ; its length is 𝑛.
The length of M is the supremum of the lengths of all chains of submodules
of M; it is denoted by ℓA (M) or ℓ (M).

Examples (6.2.6). — a) The null A-module has length 0, and con-

versely.
b) Simple A-modules are exactly the A-modules of length 1.
Let indeed M be an A-module. If M is simple, then the only chains of
submodules of M are (0), (M) (both of length 0) and (0, M) (of length 1),
so that M has length 1. Conversely, if ℓ A (M) = 1, then M ≠ 0 (otherwise,
the only chain of submodules of M would be (0), of length 0) and if there
existed a submodule N such that N ≠ 0 and N ≠ M, the chain (0, N, M)
would imply that ℓ A (M) > 2.
c) Assume that A is a division ring. The expression “chain of sub-
modules” is equivalent to “increasing sequence of vector subspaces”. In
such a sequence, the dimension increases at least by 1 at each inclusion,
and exactly by 1 if the sequence cannot be enlarged. This implies that
ℓA (M) = dimA (M): the length of M is its dimension as a vector space.
d) The length of the Z-module Z is infinite, as witnessed by the
arbitrarily long chains of ideals: (2𝑛 Z, 2𝑛−1 Z, . . . , Z).
e) Let I be a two-sided ideal and let M be an A-module annihilated
by I. Then every submodule of M is also annihilated by I, hence can
be viewed as an A/I-module, and conversely. Consequently, the length
of M as an A-module is equal to its length as an A/I-module.
f) A chain (M0 , M1 , . . . , M𝑛 ) of submodules is maximal if it cannot
be made longer by inserting a module at the beginning, between two
elements, or at the end of the sequence without destroying its increasing
character. Therefore, such a chain is maximal if and only if M0 = 0,
M𝑖 /M𝑖−1 is a simple A-module for 1 6 𝑖 6 𝑛, and M𝑛 = M.
 6.2. LENGTH 277

If an A-module has finite length, any chain can be extended to a

maximal one. A chain of length ℓ A (M) must be maximal; and we shall
prove below (theorem 6.2.10) that all maximal chains have length ℓ A (M),
but this is not a priori obvious.

Lemma (6.2.7). — Let A be a ring, let M be a right A-module, let M0 , M00 , N be

submodules of M such that M0 ⊂ M00, M0 ∩N = M00 ∩N and M0 +N = M00 +N.
Then M0 = M00.

Proof. — Let 𝑚 ∈ M00. Since M00 ⊂ M00 + N and M00 + N = M0 + N, there

exist 𝑚 0 ∈ M0 and 𝑛 ∈ N such that 𝑚 = 𝑚 0 +𝑛. Then, 𝑛 = 𝑚−𝑚 0 ∈ M00 ∩N,
hence 𝑛 ∈ M0 ∩ N. This implies that 𝑚 = 𝑚 0 + 𝑛 belongs to M0.
Consequently, M00 ⊂ M0, hence M00 = M0. 

Proposition (6.2.8). — Let A be a ring, let M be an A-module and let N be a

submodule of N. Assume that among the modules M, N and M/N, two have
finite length. Then the length of the other one is finite too and one has
ℓA (M) = ℓA (N) + ℓ A (M/N).

Proof. — Let (N0 , N1 , . . . , N𝑎 ) be a chain of submodules of N of length 𝑎.

Let (P0 , . . . , P𝑏 ) be a chain of submodules of M/N of length 𝑏; for
each 𝑖 ∈ {0, . . . , 𝑏}, there exists a unique submodule M𝑖 ⊂ M contain-
ing N such that P𝑖 = M𝑖 /N and (M0 , . . . , M𝑏 ) is a chain in M. Then,
(N0 , N1 , . . . , N𝑎 , M1 , . . . , M𝑏 ) is a chain of length 𝑎 + 𝑏 of submodules
of M, hence the inequality ℓ (M) > ℓ (N) + ℓ (M/N), with the usual con-
vention ∞ = 𝑛 + ∞. In particular, if M has finite length, then so have N
and M/N.
Conversely, let us assume that N and M/N have finite length; we want
to prove that M has finite length and that ℓ (M) = ℓ (N) + ℓ (M/N). Let
thus (M0 , . . . , M𝑎 ) be a chain of submodules of M. Applying lemma 6.2.7
to M0 = M𝑖 and M00 = M𝑖+𝑖 , we see that at least one of the two inclusions
M𝑖 ∩ N ⊂ M𝑖+1 ∩ N and M𝑖 + N ⊂ M𝑖+1 + N
is strict. Consequently, removing the inclusions which are equalities in
the sequences
M0 ∩ N ⊂ M1 ∩ N ⊂ · · · ⊂ M𝑎 ∩ N
278 CHAPTER 6. FURTHER RESULTS ON MODULES

and

(M0 + N)/N ⊂ (M1 + N)/N ⊂ · · · ⊂ (M𝑎 + N)/N,

we obtain two chains of submodules in N and in M/N whose lengths add

up to 𝑎, at least. In particular, ℓ (N) + ℓ (M/N) > ℓ (M). This concludes
the proof of the proposition. 

É𝑛
Corollary (6.2.9). — A direct sum 𝑖=1 M𝑖 of a family M1 , . . . , M𝑛 of right
A-modules of finite lengths has finite length.

Theorem (6.2.10) (Jordan–Hölder). — Let A be a ring and let M be a right

A-module. Let M0 ( M1 ( . . . M𝑚 and N0 ( N1 ( . . . N𝑛 be maximal chains
of submodules of A. Then 𝑚 = 𝑛 = ℓA (M) and there exists a permutation 𝜎
of {1, . . . , 𝑛} such that for each 𝑖 ∈ {1, . . . , 𝑛}, the module M𝑖 /M𝑖−1 is
isomorphic to N𝜎(𝑖) /N𝜎(𝑖)−1 .

In particular, if M possesses a maximal chain (M0 , M1 , . . . , M𝑛 ) of

submodules of M, then ℓA (M) = 𝑛. Moreover, up to reordering, the
simple A-modules M𝑖 /M𝑖−1 , for 1 6 𝑖 6 𝑛, do not depend on the chosen
maximal chain.
On the other hand, if M does not possess a maximal chain, than it
admits arbitrarily long chains, hence ℓA (M) is infinite.
Proof. — For 1 6 𝑖 6 𝑚 and 0 6 𝑗 6 𝑛, let us set M𝑖,𝑗 = M𝑖−1 + M𝑖 ∩ N 𝑗 .
It is a submodule of M such that M𝑖−1 ⊂ M𝑖,𝑗 ⊂ M𝑖 ; Moreover, M𝑖,0 =
M𝑖−1 and M𝑖,𝑛 = M𝑖 . Consequently, there exists a smallest integer
𝜎(𝑖) ∈ {1, . . . , 𝑛} such that M𝑖,𝜎(𝑖) = M𝑖 ; then, M𝑖−1 + M𝑖 ∩ N𝜎(𝑖) = M𝑖
and M𝑖 ∩ N𝜎(𝑖)−1 ⊂ M𝑖−1 . In fact, 𝜎(𝑖) is the only integer 𝑗 such that
M𝑖,𝑗 /M𝑖,𝑗−1 ≠ 0.
Similarly, let us set N 𝑗,𝑖 = N 𝑗−1 + N 𝑗 ∩ M𝑖 , for 1 6 𝑗 6 𝑛 and 0 6 𝑖 6 𝑚.
For any 𝑗, there exists a smallest integer 𝜏(𝑗) ∈ {1, . . . , 𝑚} such that
N 𝑗,𝑖 = N 𝑗 ; it is the unique integer 𝑖 such that N 𝑗,𝑖 /N 𝑗,𝑖−1 ≠ 0.
 6.2. LENGTH 279

By Zassenhaus’s lemma below, for any 𝑖, 𝑗 such that 1 6 𝑖 6 𝑚 and

1 6 𝑗 6 𝑛, we have isomorphisms
M𝑖,𝑗 M𝑖−1 + M𝑖 ∩ N 𝑗
=
M𝑖,𝑗−1 M𝑖−1 + M𝑖 ∩ N 𝑗−1
M𝑖 ∩ N 𝑗
'
(M𝑖−1 ∩ N 𝑗 ) + (M𝑖 ∩ N 𝑗−1 )
N 𝑗−1 + M𝑖 ∩ N 𝑗
'
N 𝑗−1 + M𝑖−1 ∩ N 𝑗
N 𝑗,𝑖
=
N 𝑗,𝑖−1
of A-modules. In particular, M𝑖,𝑗 /M𝑖,𝑗−1 is nonzero precisely 𝑗 = 𝜎(𝑖),
hence if and only if N 𝑗,𝑖 /N 𝑗,𝑖−1 is nonzero. Considering 𝑗 = 𝜎(𝑖), we
get 𝜏(𝜎(𝑖)) = 𝑖; setting 𝑖 = 𝜏(𝑗), we get 𝜎(𝜏(𝑗)) = 𝑗. This implies that 𝜎
is a bijection from {1, . . . , 𝑚} to {1, . . . , 𝑛}, with reciprocal bijection 𝜏;
in particular, 𝑚 = 𝑛. Moreover, if 𝑗 = 𝜎(𝑖), M𝑖,𝑗 = M𝑖 , M𝑖,𝑗−1 = M𝑖−1 ,
while N 𝑗,𝑖 = N 𝑗 and N 𝑗,𝑖−1 = N𝑖−1 ; the above isomorphisms imply that
the A-modules M𝑖 /M𝑖−1 and N𝜎(𝑖) /N𝜎(𝑖)−1 are isomorphic, hence the
theorem. 

Lemma (6.2.11) (Zassenhaus). — Let M be an A-module, let N0 ⊂ N and

P0 ⊂ P be submodules of M. There exist isomorphisms of A-modules
N0 + (N ∩ P) ∼ N∩P ∼ P0 + (N ∩ P)
←
− →
− 0 ,
N0 + (N ∩ P0) (N0 ∩ P) + (N ∩ P0) P + (N0 ∩ P)
respectively induced by the inclusions N ∩ P → N0 + (N ∩ P) and N ∩ P →
P0 + (N ∩ P).

Proof. — Let 𝑓 : N ∩ P → (N0 + (N ∩ P))/(N0 + (N ∩ P0)) be the linear map

defined as the composition of the injection 𝑗 from N ∩ P into N0 + (N ∩ P)
and of the canonical surjection 𝜋 from N0 + (N ∩ P) to its quotient by
N0 + (N ∩ P0). Let us show that 𝑓 is surjective. Indeed, let 𝑥 ∈ N0 + (N ∩ P);
one may write 𝑥 = 𝑛1 + 𝑛2 , with 𝑛1 ∈ N0 and 𝑛2 ∈ N ∩ P. One has
𝜋(𝑛1 ) = 0, hence 𝜋(𝑥) = 𝜋(𝑛2 ) = 𝑓 (𝑛2 ) and 𝜋(𝑥) belongs to the image
of 𝑓 .
280 CHAPTER 6. FURTHER RESULTS ON MODULES

Let now 𝑥 ∈ N ∩ P be any element such that 𝑓 (𝑥) = 0. By assumption,

one may write 𝑥 = 𝑛1 + 𝑛2 , where 𝑛1 ∈ N0 and 𝑛2 ∈ N ∩ P0. In particular,
𝑛2 ∈ P so that 𝑛1 = 𝑥 − 𝑛2 ∈ P. It follows that 𝑛1 ∈ N0 ∩ P and 𝑥
belongs to the sum of the submodules N0 ∩ P and N ∩ P0 of N ∩ P.
Conversely, these submodules are both contained in Ker( 𝑓 ), hence
Ker( 𝑓 ) = (N0 ∩ P) + (N ∩ P0). Passing to the quotient, 𝑓 induces an
isomorphism
N∩P N0 + (N ∩ P)
' ,
(N0 ∩ P) + (N ∩ P0) N0 + (N ∩ P0)
so that the first two A-modules in the lemma are indeed isomorphic.
Exchanging the roles of N0 , N and of P0 , P, the second and the third
A-modules are isomorphic too, QED. 

Remark (6.2.12). — Let K be a division ring and let M be a K-vector space.

a) Let us explain how the concept of length allows to recover the main
results about the dimension of vector spaces.
If M is a simple K-module, then, for any nonzero element 𝑥 of M,
𝑥K is a nonzero submodule of M, hence 𝑥K = M. Consequently, (𝑥)
is a basis of M. Conversely, assume that M is a K-module generated
by one element 𝑥; then, for any nonzero 𝑦 ∈ M, there exists 𝑎 ∈ K
such that 𝑦 = 𝑥𝑎. Necessarily, 𝑎 ≠ 0, so that 𝑥 = 𝑦𝑎 −1 belongs the
submodule generated by 𝑦, hence M = 𝑦K. The simple K-modules are
the K-modules generated by a single nonzero element.
Assume that M is finitely generated, say by a family (𝑥 1 , . . . , 𝑥 𝑛 ). Let
us also assume that this family is minimal, so that it is a basis of M.
Then, for 𝑖 ∈ {0, . . . , 𝑛}, define M𝑖 = Span(𝑥 1 , . . . , 𝑥 𝑖 ). We thus defined
a family of subspaces of M such that M0 ⊂ M1 ⊂ · · · ⊂ M𝑛 . If M𝑖 = M𝑖−1 ,
then 𝑥 𝑖 ∈ Span(𝑥1 , . . . , 𝑥 𝑖−1 ), contradicting the hypothesis that the family
(𝑥 1 , . . . , 𝑥 𝑛 ) be minimal among those generating M. Therefore, M𝑖 /M𝑖−1
is a nonzero vector space generated by one element (the class of 𝑥 𝑖 ), hence
is a simple K-module. Consequently, the theorem of Jordan-Hölder
implies the two following results:
– ℓK (M) = 𝑛;
– Any minimal generating family of M has exactly 𝑛 elements.
This reproves the fact that any two bases of M have the same cardinality.
 6.2. LENGTH 281

b) Assume that M is finitely dimensional. Let E = (𝑒1 , . . . , 𝑒 𝑛 ) and

F = ( 𝑓1 , . . . , 𝑓𝑛 ) be two bases of M.
For 0 6 𝑖 6 𝑛, define M𝑖 = Span(𝑒1 , . . . , 𝑒 𝑖 ) and N𝑖 = Span( 𝑓1 , . . . , 𝑓𝑖 ).
The proof of the theorem of Jordan-Hölder furnishes a (unique) per-
mutation 𝜎 of {1, . . . , 𝑛} such that M𝑖−1 + M𝑖 ∩ N𝜎(𝑖)−1 = M𝑖−1 and
M𝑖−1 + M𝑖 ∩ N𝜎(𝑖) = M𝑖 , for every 𝑖 ∈ {1, . . . , 𝑛}. For any 𝑖, let 𝑥 𝑖 be
a vector belonging to M𝑖 ∩ N𝜎(𝑖) but not to M𝑖−1 . For every 𝑖, one has
Span(𝑥 1 , . . . , 𝑥 𝑖 ) = M𝑖 ; it follows that X = (𝑥1 , . . . , 𝑥 𝑛 ) is a basis of M;
moreover, there exists a matrix B1 , in upper triangular form, such that
X = EB1 .
Set 𝜏 = 𝜎 −1 . Set 𝑦 𝑗 = 𝑥 𝜏(𝑗) for 𝑗 ∈ {1, . . . , 𝑛}; one has 𝑦 𝑗 ∈ N 𝑗 ∩ M𝜏(𝑖)
and 𝑦 𝑗 ∉ N 𝑗−1 . Similarly, one thus has Span(𝑥 𝜏(1) , . . . , 𝑥 𝜏(𝑗) ) = N 𝑗 for
every 𝑗. Consequently, there exists a matrix B2 , still in upper triangular
form, such that (𝑥 𝜏(1) , . . . , 𝑥 𝜏(𝑛) ) = FB2 . Let P𝜏 be the permutation matrix
associated to 𝜏, we have (𝑥 𝜏(1) , . . . , 𝑥 𝜏(𝑛) ) = (𝑥 1 , . . . , 𝑥 𝑛 )P𝜏 . This implies
that FB2 = EB1 P𝜏 , hence F = EB1 P𝜏 B−1 2
. Therefore, the matrix A =
B1 P𝜏 P−1
2
that expresses the coordinates of the vectors of F in the basis E
is the product of an upper-triangular matrix, a permutation matrix and
another upper-triangular matrix.
In the group GL(𝑛, K), let B be the subgroup consisting of upper-
triangular matrices, and let W be the subgroup consisting of permutation
matrices. We have proved that GL(𝑛, K) = BWB: this is called the Bruhat
decomposition.

Proposition (6.2.13). — Let A be a commutative ring, let S be a multiplicative

subset of A and let M be a A-module of finite length. Then S−1 M is a
S−1 A-module of finite length, and ℓS−1 A (S−1 M) 6 ℓA (M).

Proof. — Let us pose N = S−1 M and let (N0 , N1 , . . . , N𝑛 ) be a chain

of submodules of N. For every 𝑖, let M𝑖 be the inverse image of N𝑖
in M by the canonical morphism M → S−1 M. By construction, we have
inclusions of submodules M0 ⊂ M1 ⊂ · · · ⊂ M𝑛 . More precisely, since
S−1 M𝑖 = N𝑖 for every 𝑖 (see proposition 3.6.7), this is even a chain of
submodules, hence ℓA (M) > 𝑛. By definition, the length ℓS−1 A (S−1 M) is
the supremum of those integers 𝑛, hence ℓ A (M) > ℓS−1 A (S−1 M). 
282 CHAPTER 6. FURTHER RESULTS ON MODULES

6.3. The noetherian property

Definition (6.3.1). — Let A be a ring. One says that an A-module M is
noetherian if every nonempty family of submodules of M possesses a maximal
element.
Proposition (6.3.2). — Let A be a ring and let M be an A-module. The
following properties are equivalent:
(i) The module M is noetherian;
(ii) Every submodule of M is finitely generated;
(iii) Every increasing family of submodules is stationary.
Proof. — (i)⇒(ii) Let us assume that M is noetherian, that is, any
nonempty family of submodules of M admits a maximal element and
let us show that every submodule of M is finitely generated.
Let N be a submodule of M and let us consider the family 𝒮N of all
finitely generated submodules of N. This family is nonempty because
the null module 0 belongs to 𝒮N . By hypothesis, 𝒮N possesses a maximal
element, say, N0. By definition, the A-module N0 is a finitely generated
submodule of N and no submodule P of N such that N0 ( P is finitely
generated.
For every 𝑚 ∈ N, the A-module P = N0 + A𝑚 satisfies N0 ⊂ P ⊂ N and
is finitely generated; by maximality of N0, one has P = N0, hence 𝑚 ∈ N0.
This proves that N0 = N, hence N is finitely generated.
(ii)⇒(iii). Let us assume that every submodule of M is finitely gener-
ated and let us consider an increasing sequence (M𝑛 )𝑛∈N of submodules
Ð
of M. Let N = M𝑛 be the union of these modules M𝑛 . Since the
family is increasing, N is a submodule of M. By hypothesis, N is finitely
generated. Consequently, there exists a finite subset S ⊂ N such that
N = hSi. For every 𝑠 ∈ S, there exists an integer 𝑛 𝑠 ∈ N such that 𝑠 ∈ M𝑛 𝑠 ;
then, 𝑠 ∈ M𝑛 for any integer 𝑛 such that 𝑛 > 𝑛 𝑠 . Let us set 𝜈 = sup(𝑛 𝑠 ),
so that S ⊂ M𝜈 . It follows that N = hSi is contained in M𝜈 . Finally, for
𝑛 > 𝜈, the inclusions M𝜈 ⊂ M𝑛 ⊂ N ⊂ M𝜈 for 𝑛 > 𝜈 show that M𝑛 = M𝜈 .
We have shown that the sequence (M𝑛 ) is stationary.
(iii)⇒(i) Let us assume that any increasing sequence of submodules
of M is stationary and let (M𝑖 )𝑖∈I be a family of submodules of M indexed
by a nonempty set I. Assuming by contradiction that this family has no
 6.3. THE NOETHERIAN PROPERTY 283

maximal element, we are going to construct from the family (M𝑖 ) a strictly
increasing sequence of submodules of M. Fix 𝑖 1 ∈ I. By hypothesis,
M𝑖1 is not a maximal element of the family (M𝑖 ), so that there exists
𝑖 2 ∈ I such that M𝑖1 ( M𝑖2 . Then, M𝑖2 is not maximal neither, hence
the existence of 𝑖 3 ∈ I such that M𝑖2 ( M𝑖3 . Finally, we obtain a strictly
increasing sequence (M𝑖 𝑛 )𝑛∈N of submodules of M, hence the desired
contradiction. (This part of the proof has nothing to do with modules, it is
valid in any ordered set.) 

Proposition (6.3.3). — Let A be a ring, let M be an A-module and let N be

a submodule of M. Then M is noetherian if and only if both N and M/N are
noetherian.

Proof. — Let us assume that M is noetherian. Since every submodule

of N is also a submodule of M, every submodule of N is finitely generated,
and N is noetherian. Let 𝒫 be a submodule of M/N, let P = cl−1 (𝒫)
be its inverse image by the canonical morphism cl : M → M/N. By
hypothesis, P is finitely generated, so that 𝒫 = cl(P) is the image of a
finitely generated module, hence is finitely generated too. This shows
that M/N is noetherian.
Let us assume that N and M/N are noetherian. Let P be a submodule
of M. By assumption, the submodule P ∩ N of N is finitely generated, as
well as the submodule cl(P) of M/N. Moreover, cl(P) ' P/(P ∩ N). By
proposition 3.5.5 applied to the module P and its submodule P ∩ N, P is
finitely generated. 

Corollary (6.3.4). — The direct sum M1 ⊕ · · · ⊕ M𝑛 of a finite family of

noetherian A-modules M1 , . . . , M𝑛 is noetherian.

Definition (6.3.5). — One says that a ring A is a left noetherian ring if the
left A-module A𝑠 is noetherian. One says that A is a right noetherian ring if
the right A-module A𝑑 is noetherian.

When a commutative ring A is left noetherian, it is also right noetherian,

and vice versa; one then simply says that A is noetherian.
284 CHAPTER 6. FURTHER RESULTS ON MODULES

Remark (6.3.6). — Let A be a ring. The submodules of the left A-

module A𝑠 are its left ideals. Consequently, the ring A is left noetherian
if and only if one of the following (equivalent) properties holds:
(i) Every nonempty family of left ideals of A possesses a maximal
element;
(ii) Every left ideal of A is finitely generated;
(iii) Every increasing family of left ideals of A is stationary.
In particular, a principal ideal domain (all of whose ideals are generated
by one element) is a noetherian ring. We had already observed this fact
in lemma 2.5.6 in the course of the proof that a principal ideal domain is
a unique factorization domain.

Corollary (6.3.7). — Assume that the ring A is left noetherian (resp. right
noetherian) and let 𝑛 be an integer. Every submodule of the left A-module A𝑛𝑠
(resp. of the right A-module A𝑛𝑑 ) is finitely generated.

Corollary (6.3.8). — Assume that the ring A be left noetherian (resp. right
noetherian). Then, for any two-sided ideal I of A, the quotient ring A/I is left
noetherian (resp. right noetherian).

Proof. — Indeed, every left ideal of A/I is of the form J/I for some left
ideal J of A containing I. By hypothesis, J is finitely generated, so that
J/I is finitely generated too. The case of right ideals is analogous. 

Remark (6.3.9). — Let A be a ring and let M be a right A-module. Let

I = AnnA (M) be the annihilator of M; it is a two sided ideal of A. If M is
noetherian, then the quotient ring A/I is right noetherian.
Let indeed (𝑚1 , . . . , 𝑚𝑛 ) be a finite generating family of M. The map
𝑓 : A𝑑 → M𝑛 given by 𝑓 (𝑎) = (𝑚1 𝑎, . . . , 𝑚𝑛 𝑎) is a morphism of right A-
modules, and its kernel is equal to I. Consequently, 𝑓 defines an injective
morphism 𝜑 : A𝑑 /I → M𝑛 . Since M is noetherian, M𝑛 is noetherian too
(corollary 6.3.4). Consequently, A𝑑 /I is a noetherian right A-module.
Since its submodules are the right ideals of A/I, this proves that A/I is
right noetherian.
 6.3. THE NOETHERIAN PROPERTY 285

Proposition (6.3.10). — Let A be a commutative ring and let S be a multiplica-

tive subset of A. If M is a noetherian A-module, then S−1 M is a noetherian
S−1 A-module.

Proof. — Let 𝒩 be a submodule of S−1 M. By proposition 3.6.7, there

exists a submodule N of M such that 𝒩 = S−1 N. Since M is a noetherian
A-module, N is finitely generated. Consequently, 𝒩 is a finitely generated
S−1 A-module. This shows that S−1 M is a noetherian S−1 A-module. 

Corollary (6.3.11). — Let A be a commutative noetherian ring and let S be a

multiplicative subset of A. The fraction ring S−1 A is noetherian.

Proof. — By proposition 6.3.10, S−1 A is a noetherian S−1 A-module.

Consequently, S−1 A is a noetherian ring. 

Theorem (6.3.12) (Hilbert). — For any a commutative noetherian ring A, the

ring A[X] is noetherian.

Proof. — Let I be an ideal of A[X]; let us prove that I is finitely generated.

For 𝑛 > 0, let I𝑛 be the set of coefficients of X𝑛 , in all polynomials of
degree 6 𝑛 belonging to I; explicitly, I𝑛 is the set of all 𝑎 ∈ A for which
there exists a polynomial P ∈ I such that P = 𝑎X𝑛 +terms of lower degree
(which we will write P = 𝑎X𝑛 + . . . ). Observe that I𝑛 is an ideal of A.
Indeed, if 𝑎, 𝑏 ∈ I𝑛 , there are polynomials P, Q ∈ I of degree 6 𝑛
such that P = 𝑎X𝑛 + · · · and Q = 𝑏X𝑛 + · · · . Then, for any 𝑢, 𝑣 ∈ A,
𝑢P + 𝑣Q = (𝑢𝑎 + 𝑣𝑏)X𝑛 + · · · ; since I is an ideal of A, 𝑢P + 𝑣Q ∈ I, hence
𝑢𝑎 + 𝑣𝑏 ∈ I𝑛 . Moreover, one may write the polynomial 0 as 0 = 0X𝑛 + · · · ,
so that 0 ∈ I𝑛 .
For any integer 𝑛, one has I𝑛 ⊂ I𝑛+1 ; indeed, if 𝑎 ∈ I𝑛 and P ∈ I is such
that P = 𝑎X𝑛 + · · · , then XP = 𝑎X𝑛+1 + · · · , so that 𝑎 ∈ I𝑛+1 . Since A is a
noetherian ring, the sequence (I𝑛 ) is stationary. Let 𝜈 ∈ N be such that
I𝑛 = I𝜈 for any integer 𝑛 > 𝜈.
The ideals I0 , I1 , . . . , I𝜈 are finitely generated; let us choose, for every
integer 𝑛 6 𝜈, a generating family (𝑎 𝑛,𝑖 )16𝑖6𝑟(𝑛) of I𝑛 , as well as polyno-
mials P𝑛,𝑖 ∈ I𝑛 such that P𝑛,𝑖 = 𝑎 𝑛,𝑖 X𝑛 + · · · . Let J be the ideal of A[X]
generated by the polynomials P𝑛,𝑖 for 0 6 𝑛 6 𝜈 and 1 6 𝑖 6 𝑟(𝑛). The
ideal J is finitely generated and, by construction, one has J ⊂ I. We
286 CHAPTER 6. FURTHER RESULTS ON MODULES

shall now show the converse inclusion by induction on the degree of a

polynomial P ∈ I.
If deg(P) = 0, then P is constant and belongs to I0 , so that P ∈ J. Assume
now that every polynomial in I whose degree is < 𝑛 belongs to J. Let
P ∈ I be a polynomial of degree 𝑛 and let 𝑎 be its leading coefficient. Let
𝑚 = inf(𝑛, 𝜈); if 𝑛 6 𝜈, then 𝑚 = 𝑛, hence 𝑎 ∈ I𝑚 ; otherwise, 𝑚 = 𝜈 and
𝑎 ∈ I𝜈 since I𝑛 = I𝜈 by definition of 𝜈. Consequently, there are elements
𝑐 𝑖 ∈ A, for 1 6 𝑖 6 𝑟(𝑚) such that 𝑎 = 𝑖 𝑐 𝑖 𝑎 𝑚,𝑖 ; let Q be the polynomial
Í
Q = P − X𝑛−𝑚 𝑖 𝑐 𝑖 P𝑚,𝑖 . By construction, deg(Q) 6 𝑛; moreover, the
Í
coefficient of X𝑛 in Q is equal to 𝑎 − 𝑖 𝑐 𝑖 𝑎 𝑚,𝑖 = 0, so that deg(Q) < 𝑛. By
Í
induction, Q ∈ I, so that Q ∈ J. It follows that P ∈ J.
We thus have proved that I = J. In particular, I is finitely generated. 

Corollary (6.3.13). — Let A be a commutative noetherian ring and let 𝑛 be a

positive integer. The ring A[X1 , . . . , X𝑛 ] is noetherian. In particular, for any
field K, the ring K[X1 , . . . , X𝑛 ] is noetherian.

Corollary (6.3.14). — Let A and B be commutative rings. Assume that A is

a noetherian ring and that B is a finitely generated A-algebra. Then, B is a
noetherian ring.

Proof. — By hypothesis, there are elements 𝑏 1 , . . . , 𝑏 𝑛 ∈ B which gen-

erate B as an A-algebra, so that the canonical morphism of A-algebras
from A[X1 , . . . , X𝑛 ] to B which maps X𝑖 to 𝑏 𝑖 is surjective. This shows
that B is a quotient of the noetherian ring A[X1 , . . . , X𝑛 ]. Consequently,
B is noetherian. 

Theorem (6.3.15) (Hilbert, 1893). — Let K be a field, let A be a finitely

generated commutative K-algebra and let G be a finite subgroup of AutK (A).
Then, the set AG of all 𝑎 ∈ A such that 𝑔(𝑎) = 𝑎 for any 𝑔 ∈ G is a finitely
generated K-algebra.

It is precisely to prove this theorem that Hilbert introduced the notion

of a noetherian ring and proved that polynomial rings over a field are
noetherian.1
1This needs a more precise reference!
 6.3. THE NOETHERIAN PROPERTY 287

Example (6.3.16). — Let us assume that A = K[X1 , . . . , X𝑛 ] and G is the

symmetric group 𝔖𝑛 acting on A by permuting the indeterminates X𝑖 .
Then, A𝔖𝑛 is the algebra of symmetric polynomials. In this case, it
is generated by the elementary symmetric polynomials S0 , S1 , . . . , S𝑛
defined, for 1 6 𝑗 6 𝑛, by
Õ
S𝑗 = X𝑖1 . . . X𝑖 𝑗 .
𝑖 1 <···<𝑖 𝑗

For example, S0 = 1, S1 = X1 + · · · + X𝑛 and S𝑛 = X1 . . . X𝑛 .

In fact, we prove a more precise result: Let K be any commutative ring,
and let P ∈ K[X1 , . . . , X𝑛 ] be a symmetric polynomial; there exists a unique
polynomial Q ∈ K[T1 , . . . , T𝑛 ] such that P = Q(S1 , . . . , S𝑛 ); more precisely, if
𝑐T1𝑚1 . . . T𝑚𝑛 is a monomial that appears in Q, then 𝑚1 + 2𝑚2 + · · · + 𝑛𝑚 𝑛 6
𝑛

deg(P). (We say that the “weighted degree” of Q is at most that of P.)
Let us now prove this result by induction, first on 𝑛, then on deg(P).
When 𝑛 = 1, one has S1 = X1 and A𝔖1 = A = K[X1 ]. Assume that the
result holds for (𝑛 − 1). Writing S00 , . . . , S0𝑛−1 for the symmetric polyno-
mials in the variables X1 , . . . , X𝑛−1 , we observe the relations S0 = S00 , S1 =
S01 +X𝑛 S00 , S2 = S02 +X𝑛 S01 , . . . , S𝑛−1 = S0𝑛−1 +X𝑛 S0𝑛−2 , and S𝑛 = X𝑛 S0𝑛−1 . Con-
sider a symmetric polynomial P ∈ A[X1 , . . . , X𝑛 ]. Then P(X1 , . . . , X𝑛−1 , 0)
is a symmetric polynomial in A[X1 , . . . , X𝑛−1 ], hence it can be expressed
as a polynomial in S01 , . . . , S0𝑛−1 . Let thus Q1 ∈ A[T1 , . . . , T𝑛1 ] be such that
P(X1 , . . . , X𝑛−1 , 0) = Q1 (S01 , . . . , S0𝑛−1 ), where the weighted degree of Q1
is at most that of P(X1 , . . . , X𝑛−1 , 0), hence at most that of P. Let us then
consider the polynomial P1 = P − Q1 (S1 , . . . , S𝑛−1 ); this is a symmetric
polynomial of degree 6 deg(P) such that P1 (X1 , . . . , X𝑛−1 , 0) = 0, so that
all monomials apppearing in P1 are multiple of X𝑛 and X𝑛 divides P1 .
By symmetry, X1 , . . . , X𝑛1 divide P1 , which means that every monomial
appearing in P1 is divisible by the product X1 . . . X𝑛 = S𝑛 . There exists
a polynomial P2 ∈ K[X1 , . . . , X𝑛 ] such that P1 = X1 . . . X𝑛 P2 , and P2
is necessarily symmetric (because X1 . . . X𝑛 is symmetric and regular),
and its degree is 6 deg(P) − 𝑛. By induction, there exists a polyno-
mial Q2 ∈ A[T1 , . . . , T𝑛 ] of weighted degree 6 deg(P) − 𝑛 such that
P2 = Q2 (S1 , . . . , S𝑛 ). Let us set Q = Q1 + T𝑛 Q2 ; its weighted degree
is 6 deg(P) and Q(S1 , . . . , S𝑛 ) = Q1 (S1 , . . . , S𝑛−1 ) + S𝑛 Q2 (S1 , . . . , S𝑛 ) =
(P − P1 ) + X1 . . . X𝑛 P2 = P.
288 CHAPTER 6. FURTHER RESULTS ON MODULES

This proves existence; let us not establish the uniqueness of such a

polynomial. Let Q ∈ K[T1 , . . . , T𝑛 ] be such that P = Q(S1 , . . . , S𝑛 ) =
0. Then 0 = P(X1 , . . . , X𝑛−1 , 0) = Q(S01 , . . . , S0𝑛−1 , 0). By induction,
the polynomial Q(T1 , . . . , T𝑛−1 , 0) vanishes, which means that there
exists a polynomial Q1 ∈ K[T1 , . . . , T𝑛 ] such that Q = T𝑛 Q1 . Then
0 = Q(S1 , . . . , S𝑛 ) = S𝑛 Q1 (S1 , . . . , S𝑛 ). Since S𝑛 = X1 . . . X𝑛 is a regular
element of K[X1 , . . . , X𝑛 ], one has Q1 (S1 , . . . , S𝑛 ) = 0. By induction on
the degree of Q, one has Q1 = 0, hence Q = T𝑛 Q1 = 0.

The proof of theorem 6.3.15 requires three steps.

Lemma (6.3.17). — The set AG is a subalgebra of A.

Proof. — We need to prove that

– The elements 0 and 1 belong to AG ;
– If 𝑎 and 𝑏 belong to AG , so do 𝑎 + 𝑏 and 𝑎𝑏;
– If 𝑎 belongs to AG and 𝜆 ∈ K, then 𝜆𝑎 belongs to AG .
Recall that every 𝑔 ∈ G is an automorphism of A as a K-algebra, so that
𝑔(0) = 0, 𝑔(1) = 1, 𝑔(𝑎 + 𝑏) = 𝑔(𝑎) + 𝑔(𝑏), 𝑔(𝑎𝑏) = 𝑔(𝑎)𝑔(𝑏) for 𝑎, 𝑏 ∈ A,
𝑔(𝜆𝑎) = 𝜆𝑔(𝑎) for 𝑎 ∈ A, 𝑔 ∈ G and 𝜆 ∈ K. In particular, we see that
0 ∈ AG and 1 ∈ AG . Moreover, if 𝑎, 𝑏 ∈ AG , then these formulae imply
that 𝑔(𝑎 + 𝑏) = 𝑔(𝑎) + 𝑔(𝑏) = 𝑎 + 𝑏 and 𝑔(𝑎𝑏) = 𝑔(𝑎)𝑔(𝑏) = 𝑎𝑏 for every
𝑔 ∈ G, so that 𝑎 + 𝑏 and 𝑎𝑏 belong to AG . Finally, if 𝑎 ∈ AG and 𝜆 ∈ K,
then 𝑔(𝜆𝑎) = 𝜆𝑔(𝑎) = 𝜆𝑎 for every 𝑔 ∈ G, hence 𝜆𝑎 ∈ AG . 

Lemma (6.3.18). — Under the hypotheses of theorem 6.3.15, the ring A is a

finitely generated AG -module.

Proof. — Since A is finitely generated as a K-algebra, we may choose

elements 𝑎1 , . . . , 𝑎 𝑟 ∈ A such that A = K[𝑎1 , . . . , 𝑎 𝑟 ].
For 𝑖 ∈ {1, . . . , 𝑟}, let us consider the polynomial
Ö
P𝑖 (X) = (X − 𝑔(𝑎 𝑖 ))
𝑔∈G

in A[X]. Let G act on A[X] by 𝑔(P) = 𝑔(𝑐 𝑛 )X𝑛 if P = 𝑐 𝑛 X𝑛 and 𝑔 ∈ G.

Í Í
Let us write
P𝑖 (X) = X𝑛 + 𝑏 1 X𝑛−1 + · · · + 𝑏 𝑛
 6.3. THE NOETHERIAN PROPERTY 289

where 𝑏1 , . . . , 𝑏 𝑛 ∈ A and 𝑛 = Card(G). For any ℎ ∈ G, one then has

Ö Ö Ö
ℎ(P𝑖 (X)) = ℎ( (X − 𝑔(𝑎 𝑖 ))) = (X − ℎ(𝑔(𝑎 𝑖 ))) = (X − 𝑔(𝑎 𝑖 )) = P𝑖 (X)
𝑔∈G 𝑔∈G

so that 𝑏1 , . . . , 𝑏 𝑛 are invariant under ℎ. It follows that P𝑖 ∈ AG [X], hence

𝑏 1 , . . . , 𝑏 𝑛 ∈ AG . By construction, P𝑖 (𝑎 𝑖 ) = 0, a monic polynomial relation
for the 𝑎 𝑖 , with coefficients in AG . In the terminology of definition 4.1.1,
we just proved that the elements 𝑎 𝑖 are integral over AG ; we now proceed
to show that A is a finitely generated AG -module. Indeed, let us prove
that A is generated, as an AG -module, by the 𝑛 𝑟 products of the form
Î𝑟 𝑛 𝑖
𝑖=1 𝑎 𝑖 , where, for every 𝑖, 0 6 𝑛 𝑖 6 𝑛 − 1. Let A be the A -submodule
0 G

of A generated by these elements.

Since A is generated by all products 𝑟𝑖=1 𝑎 𝑖𝑛 𝑖 with 𝑛1 , . . . , 𝑛 𝑟 ∈ N, as a
Î
K-module, henceforth as an AG -module, it suffices to prove that each
such product belongs to A0. Let X𝑛 𝑖 = Q𝑖 (X)P𝑖 (X)+R𝑖 (X) be the euclidean
division of X𝑛 𝑖 by the monic polynomial P𝑖 in AG [X] (theorem 1.3.15).
Since P𝑖 (𝑎 𝑖 ) = 0, we have 𝑎 𝑖𝑛 𝑖 = R𝑖 (𝑎 𝑖 ) for all 𝑖, hence
𝑟 𝑟
𝑎 𝑖𝑛 𝑖
Ö Ö
= R𝑖 (𝑎 𝑖 ).
𝑖=1 𝑖=1

If we expand this last product, we see that it belongs to A0. This shows
that A0 = A, hence that A is finitely generated as an AG -module. 

Lemma (6.3.19) (Artin–Tate). — Let K, B, A be three commutative rings such

that K ⊂ B ⊂ A. Assume that K is a noetherian ring, that A is a finitely
generated K-algebra as well as a finitely generated B-module. Then B is a finitely
generated K-algebra.

Proof. — Let (𝑥 1 , . . . , 𝑥 𝑟 ) be a finite family of elements of A which

generates A as a K-algebra; let (𝑎 1 , . . . , 𝑎 𝑛 ) be a finite family of elements
of A which generates A as a B-module. We assume, as we may, that
𝑥 1 = 1 and 𝑎1 = 1. Then, every element of A can be written both as a
polynomial in 𝑥1 , . . . , 𝑥 𝑟 with coefficients in K and as a linear combination
of 𝑎1 , . . . , 𝑎 𝑛 with coefficients in B. If we apply this remark to 𝑥1 , . . . , 𝑥 𝑟
and to the products 𝑥 𝑖 𝑥 𝑗 (for 1 6 𝑖, 𝑗 6 𝑟), we obtain elements 𝑏 𝑖,ℓ and
290 CHAPTER 6. FURTHER RESULTS ON MODULES

𝑐 𝑖,𝑗,ℓ ∈ B such that

𝑛
Õ
𝑥𝑖 = 𝑏 𝑖ℓ 𝑎ℓ
ℓ =1

for 𝑖 ∈ {1, . . . , 𝑟} and

𝑛
Õ
𝑎𝑖 𝑎 𝑗 = 𝑐 𝑖,𝑗,ℓ 𝑎ℓ
ℓ =1

for 𝑖, 𝑗 ∈ {1, . . . , 𝑛}.

Let B0 be the K-subalgebra of B generated by all the elements 𝑏 𝑖,ℓ and
𝑐 𝑖,𝑗,ℓ ; it is a finitely generated K-algebra, in particular a noetherian ring.
Let A0 be the B0 -submodule of A generated by the elements 𝑎 𝑖 . It
is a K-submodule of A; moreover, all products 𝑎 𝑖 𝑎 𝑗 belong to A0 by
construction, so that A0 is stable by the multiplication law of A; and
1 ∈ A0 . Consequently, A0 is a subalgebra of A. Still by construction, the
elements 𝑥1 , . . . , 𝑥 𝑟 belong to A0 , hence A0 = A. This proves that A is a
finitely generated B0 -module.
Since B0 is a noetherian ring, the ring A is noetherian as a B0 -module,
so that every B0 -submodule of A is finitely generated. Since B is such a
B0 -submodule, we obtain that B is finitely generated as a B0 -module; it
is in particular finitely generated as a B0 -algebra.
Recall that B0 is finitely generated as a K-algebra. By corollary 1.3.7,
this implies that B is also finitely generated as a K-algebra, as was to be
shown. 

Proof of theorem 6.3.15. — Set B = AG , so that K ⊂ B ⊂ A. By the second

lemma, A is finitely generated as a B-module; by hypothesis, it is finitely
generated as a K-algebra. Artin–Tate’s lemma (lemma 6.3.19) now
implies that B is finitely generated as a K-algebra. 

6.4. The artinian property

Definition (6.4.1). — Let A be a ring, let M be a right A-module One says
that M is artinian if every nonempty family of submodules of M possesses a
minimal element.
 6.4. THE ARTINIAN PROPERTY 291

Definition (6.4.2). — Let A be a ring. One says that the ring A is a right
artinian ring if it is artinian as a right A-module, and that it is left artinian if
it is artinian as a left A-module.

If the ring A is commutative, then it is equivalent for A to be left

artinian or to be right artinian; we then say that A is an artinian ring.
In some sense, the artinian property is symmetric to the noetherian
property. However, we shall see in theorem 6.4.11 that for a ring,
the property of being artinian is quite stronger than the one of being
noetherian. But for the moment, let us observe a few consequences which
are quite analogous to those obtained from the noetherian property.

Remark (6.4.3). — An A-module M is artinian if and only if every de-

creasing sequence of submodules of M is stationary. This is a general
result about ordered sets, which is proved essentially in the same way
as the equivalence of (i) and (iii) in proposition 6.3.2.
Let us indeed assume that M is artinian and let (M𝑛 ) be a decreasing
sequence of submodules. By assumption, it has a minimal element,
say M𝑚 , and one has M𝑛 = M𝑚 for 𝑛 > 𝑚.
In the other direction, let (M𝑖 )𝑖∈I be nonempty family of submodules
of M that has no minimal element. Let 𝑖 0 ∈ I; then M𝑖0 is not a minimal
element of (M𝑖 )𝑖∈I , hence there exists 𝑖 1 ∈ I such that M𝑖1 ( M𝑖0 . Similarly,
one constructs by induction a sequence (𝑖 𝑛 )𝑛∈N of elements of I such that
the sequence (M𝑖 𝑛 )𝑛∈N is strictly decreasing.

Proposition (6.4.4). — Let A be a ring. Let M be an A-module and let N be a

submodule of M. Then M is artinian if and only if N and M/N are artinian
modules.

Proof. — a) Let clN : M → M/N be the canonical surjection. Let us

first assume that N and M/N are artinian. Let (M𝑛 ) be a decreasing
sequence of submodules of M. Then (M𝑛 ∩N) is a decreasing sequence of
submodules of N, hence is stationary; similarly, (clN (M𝑛 )) is a decreasing
sequence of submodules of M/N, so is stationary too. Consequently,
there exists 𝑝 ∈ N such that M𝑛 ∩N = M𝑛+1 ∩N and clN (M𝑛 ) = clN (M𝑛+1 )
for 𝑛 > 𝑝. Applying lemma 6.2.7 to the submodules M𝑛 , M𝑛+1 , N of M,
we conclude that M𝑛 = M𝑛+1 for 𝑛 > 𝑝: the sequence (M𝑛 ) is stationary.
292 CHAPTER 6. FURTHER RESULTS ON MODULES

Conversely, let us assume that M is artinian. A decreasing sequence

of submodules of N is also a decreasing sequence of submodules of M,
hence is stationary. Consequently, N is artinian. Let then (P𝑛 ) be a
decreasing sequence of submodules of M/N. The sequence (cl−1 N (P𝑛 )) of
submodules of M is decreasing, hence is stationary since M is artinian.
Since clN is surjective, one has P𝑛 = clN (cl−1
N (P𝑛 )), hence the sequence
(P𝑛 ) is also stationary. 

Corollary (6.4.5). — The direct sum M1 ⊕ · · · ⊕ M𝑛 of a finite family of artinian

A-modules M1 , . . . , M𝑛 is artinian.
Proposition (6.4.6). — Assume that A is commutative and let S be a multi-
plicative subset of A. If M is an artinian A-module, then S−1 A is an artinian
S−1 A-module.
Proof. — Let us assume that M is an artinian A-module and let 𝑖 : M →
S−1 M be the canonical morphism. Let (P𝑛 ) be a decreasing sequence
of submodules of S−1 M. The sequence (𝑖 −1 (P𝑛 )) of submodules of M
is decreasing, hence is stationary. By proposition 3.6.7, one has P𝑛 =
S−1 (𝑖 −1 (P𝑛 )) for any integer 𝑛. This implies that the sequence (P𝑛 ) is
stationary. 

Corollary (6.4.7). — Let A be a commutative ring. If A is artinian, then S−1 A

is artinian for any multiplicative subset S of A.
Theorem (6.4.8). — Let A be a ring. An A-module has finite length if and only
if it is both artinian and noetherian.
Proof. — Let us assume that M has finite length and let us consider
a sequence (M𝑛 ) of submodules of M which is either increasing or
decreasing. The sequence (ℓ A (M𝑛 )) of their lengths is then increasing
or decreasing, according to the case; it is bounded from below by 0,
and bounded from above by ℓA (M). Consequently, it is stationary. This
implies that the sequence (M𝑛 ) is stationary. (Recall that if P ⊂ Q are
two modules of the same finite length, then ℓA (Q/P) = ℓA (Q) − ℓA (P) = 0;
consequently, Q/P = 0 and Q = P.) Considering decreasing sequences,
we obtain that M is artinian; considering increasing sequences, we
conclude that M is noetherian.
 6.4. THE ARTINIAN PROPERTY 293

Let us now assume that M is artinian and that its length is infinite,
and let us prove that M is not noetherian.
Let us set M0 = 0. By assumption, M ≠ 0, so that the set of nonzero
submodules of M is nonempty. Since M is artinian, it admits a minimal
element M1 ; this module M1 is a nonzero submodule of M such that
any strict submodule of M1 is null; in other words, M1 is simple. Since
ℓA (M1 ) = 1 and ℓA (M) = ∞, one has M1 ≠ M and we may construct
similarly a submodule M2 of M containing M1 such that M2 /M1 is simple,
hence ℓA (M2 ) = ℓ A (M1 ) + 1 = 2. We thus construct by induction an in-
creasing sequence (M𝑛 ) of submodules of M such that M𝑛+1 /M𝑛 is simple
and ℓA (M𝑛 ) = 𝑛 for every integer 𝑛. In particular, the sequence (M𝑛 ) is
not stationary, hence M is not noetherian. 

Proposition (6.4.9). — Let A be a right artinian ring and let J be the Jacobson
radical of A. There exists an integer 𝑛 such that J𝑛 = 0.
Proof. — Since A is right artinian, the decreasing sequence (J𝑛 ) of right
ideals of A is stationary. Consequently, there exists an integer 𝑛 > 0
such that J𝑛 = J𝑛+1 . Let us assume by contradiction that J𝑛 ≠ 0.
The set of right ideals I of A such that J𝑛 I ≠ 0 is nonempty (for example,
I = A is such a right ideal). Since A is right artinian, there exists a
minimal such ideal, say I. Let 𝑥 be any element of I such that J𝑛 𝑥 ≠ 0;
then 𝑥A is a right ideal of A such that J𝑛 (𝑥A) ≠ 0; moreover, 𝑥A ⊂ I.
By minimality of I, one has I = 𝑥A. In particular, I is finitely generated.
Moreover, J𝑛 JI = J𝑛+1 I = J𝑛 I ≠ 0. Since JI ⊂ I, the minimality of I implies
that I = JI. It then follows from Nakayama’s lemma (theorem 6.1.1) that
I = 0, in contradiction with the hypothesis that J𝑛 I ≠ 0. 

Lemma (6.4.10). — Let A be a ring, let 𝑛 be a positive integer. Let M be an

artinian A-module. Assume that M is the sum of its submodules of length 6 𝑛.
Then M has finite length.
Proof. — Let (M𝑖 ) be a family of submodules of M such that ℓA (M𝑖 ) 6 𝑛
for every 𝑖, and M = 𝑖 M𝑖 . Let us show that ℓA (M) < ∞ by induction
Í
on 𝑛.
Let us first treat the case 𝑛 = 1. If the length of M is infinite, we
may construct by induction on 𝑚 a sequence 𝑖(𝑚) such that M𝑖(𝑚) is
294 CHAPTER 6. FURTHER RESULTS ON MODULES

not included in the sum of the modules M𝑖(𝑘) for 𝑘 < 𝑚 (otherwise,
one would have M = 𝑘<𝑚 M𝑖(𝑘) and ℓA (M) would be finite). Since all
Í
Í
modules M𝑖 are simple or zero, one has M𝑖(𝑚) ∩ 𝑘<𝑚 M𝑖(𝑘) = 0 and
Í
the modules M𝑖(𝑘) Éare in direct sum in their union N = 𝑘∈N M𝑖(𝑘) . For
every 𝑚, set N𝑚 = 𝑘>𝑚 M 𝑖(𝑘) . The sequence (N𝑚 ) is strictly decreasing,
and this contradicts the hypothesis that M is artinian.
Let us now assume that 𝑛 > 2 and that the result holds for 𝑛 − 1.
For every 𝑖, let M0𝑖 be a submodule of M𝑖 such that ℓA (M0𝑖 ) 6 𝑛 − 1
and ℓA (M𝑖 /M0𝑖 ) 6 1. Set M0 = 𝑖 M0𝑖 ; this is a submodule of M; by
Í
the induction hypothesis, ℓA (M0) < ∞. Moreover, M/M0 is the sum of
the modules M𝑖 /(M0 ∩ M𝑖 ) whose lengths are 6 1. By the case 𝑛 = 1,
ℓA (M/M0) < ∞. By proposition 6.4.4, one has ℓA (M) < ∞, as was to be
shown. 

Theorem (6.4.11) (Akizuki-Hopkins-Levitzki). — Let A be a right artinian

ring. Then the ring A is right noetherian and has finite length as a right
A-module.
Proof2. — It suffices to prove that ℓA (A𝑑 ) is finite.
Let 𝒮 be the set of right ideals I of A such that A/I has finite length.
Since 𝒮 is nonempty (it contains A) and A is right artinian, we may
consider a minimal element I of 𝒮. Let 𝑛 = ℓA (A𝑑 /I).
Let 𝑥 be an element of A such that ℓA (𝑥A) < ∞, and let J = {𝑎 ∈
A ; 𝑥𝑎 = 0} be its right annihilator, so that J is a right ideal of A and
𝑥A ' A𝑑 /J. I claim that A𝑑 /(I ∩ J) has finite length. First of all, (I + J)/I
has finite length, as a submodule of A𝑑 /I. Then, the composition of the
inclusion morphism J → I + J and of the projection I + J → (I + J)/I is
surjective, and its kernel is I ∩ J, so that J/(I ∩ J) has finite length. This
module is a submodule of A𝑑 /(I ∩ J), with quotient A𝑑 /J, which has
finite length, by the assumption on 𝑥. By proposition 6.2.8, we conclude
that ℓ A (A/(I ∩ J)) < ∞. Since I ∩ J ⊂ I, the minimality of I implies that
I ∩ J = I, hence I ⊂ J. In particular, ℓA (𝑥A) = ℓA (A/J) 6 ℓA (A/I) 6 𝑛.
Let K be the sum of all right ideals 𝑥A of A such that ℓ A (𝑥A) < ∞.
By lemma 6.4.10, ℓA (K) < ∞. Moreover, every right ideal K0 such that
ℓA (K0) < ∞ is contained in K. Indeed, if 𝑥 ∈ K0, then ℓA (𝑥A) 6 ℓA (K0) <
2This proof is borrowed from Bourbaki (2012), §1, no 2, théorème 1.
 6.4. THE ARTINIAN PROPERTY 295

∞, hence 𝑥 ∈ K. In other words, K is the largest right ideal of A which is

of finite length.
Let us assume by contradiction that K ≠ A. Let K0 be a right ideal of A
which is minimal among the ideals containing K and distinct from K.
Such an ideal exists because A is right artinian. Necessarily, the right
A-module K0/K is simple, hence has length 1. Consequently, ℓA (K0) < ∞,
which contradicts the definition of K. This shows that K = A. In
particular, ℓA (A𝑑 ) < ∞. 

Corollary (6.4.12). — Let A be a right artinian ring and let M be right

A-module. The following conditions are equivalent:
(i) The module M is finitely generated;
(ii) The module M has finite length;
(iii) The module M is artinian;
(iv) The module M is noetherian.

Proof. — Without any hypothesis on the ring A, (ii) implies (iii) and (iv),
and (iv) implies (i).
Assume that M is finitely generated, so that there exists an integer 𝑛
and a surjective morphism A𝑛𝑑 → M of A-modules. Then ℓA (M) 6
ℓA (A𝑛𝑑 ) = 𝑛 ℓA (A𝑑 ) is finite. This proves the implication (i)⇒(ii).
It remains to show that if M is artinian, then M has finite length. Let 𝑥 ∈
M and let J = {𝑎 ∈ A ; 𝑥𝑎 = 0}, so that 𝑥A ' A/J. Since A is right artinian,
its length is finite (theorem 6.4.11) and ℓA (𝑥A) = ℓ A (A/J) 6 ℓA (A𝑑 ). This
shows that M is generated by its submodules of lengths 6 ℓA (A𝑑 ). By
lemma 6.4.10, ℓ A (M) is finite. 

Lemma (6.4.13). — Let A be a commutative artinian ring.

a) If A is an integral domain, then A is a field.
b) Every prime ideal of A is maximal.
c) The set of maximal ideals of A is finite.

Proof. — a) Let us assume that A is a domain. Let 𝑎 ∈ A {0} and

let us consider the decreasing sequence (𝑎 𝑛 A)𝑛>0 of principal ideals.
Since A is artinian, there exists an integer 𝑛 such that 𝑎 𝑛 A = 𝑎 𝑛+1 A,
hence there exists an element 𝑏 ∈ A such that 𝑎 𝑛 = 𝑎 𝑛+1 𝑏. Consequently,
296 CHAPTER 6. FURTHER RESULTS ON MODULES

𝑎 𝑛 (1 − 𝑎𝑏) = 0. Since A is a domain and 𝑎 ≠ 0, one has 1 − 𝑎𝑏 = 0 hence

𝑎 is invertible.
b) Let P be a prime ideal of A. Then, A/P is an integral domain.
Moreover, it is artinian as an A-module, because it is a quotient of
the A-module A. Since the A-submodules of A/P coincide with its
(A/P)-submodules, we see that the ring A/P is artinian. By a), it is a
field, hence P is a maximal ideal.
c) Let us argue by contradiction and let (M𝑛 ) be a sequence of pairwise
distinct maximal ideals of A. Let us consider the decreasing sequence of
ideals
M1 ⊃ M1 M2 ⊃ · · · ⊃ M1 . . . M𝑛 ⊃ · · ·
For every 𝑖 6 𝑛, M𝑖 is not contained in M𝑛+1 , so there is an element 𝑎 𝑖 ∈ M𝑖
such that 𝑎 𝑖 ∉ M𝑛+1 . The product 𝑎 = 𝑎1 . . . 𝑎 𝑛 belongs to M1 . . . M𝑛 ;
since M𝑛+1 is a prime ideal, 𝑎 does not belong to M𝑛+1 . This shows
that the preceding sequence of ideals is not stationary, contradicting the
hypothesis that A is artinian. 

Theorem (6.4.14) (Akizuki). — Let A be a commutative ring. The following

conditions are equivalent:
(i) The ring A is artinian;
(ii) The ring A is noetherian and all of its prime ideals are maximal ideals;
(iii) The A-module A has finite length.

Proof. — By theorem 6.4.8, condition (iii) is equivalent to the fact that A

is both artinian and noetherian. This shows the implication (iii)⇒(i).
We also have shown in theorem 6.4.11 that an artinian ring is noetherian;
moreover, by lemma 6.4.13, all of its prime ideals are maximal, so that
(i)⇒(ii).
It remains to show that under assumption (ii), the ring A has finite
length.
Let 𝒮 be the set of ideals of A such that the length of A/I is infinite. If
A has not finite length, then I = (0) belongs to 𝒮, hence 𝒮 is nonempty.
Since A is noetherian, the set 𝒮, ordered by inclusion, has a maximal
element I. The ideal I is such that the length of A/I is infinite, while
the length of A/J is finite for any ideal J such that J ) I. The ring A/I
 6.5. SUPPORT OF A MODULE, ASSOCIATED IDEALS 297

is noetherian, and its prime ideals, being in correspondance with the

prime ideals of A containing I, are maximal ideals. We may thus replace
A by A/I and assume that I = (0).
Let us now show that A is an integral domain. Since ℓA (A) = +∞, we
have A ≠ 0. Let then 𝑎, 𝑏 be two nonzero elements of A such that 𝑎𝑏 = 0.
By assumption, A/𝑎A has finite length. Moreover, the map 𝑥 ↦→ 𝑏𝑥
vanishes on 𝑎A, hence induces a surjective morphism from A/𝑎A to 𝑏A;
in particular, 𝑏A has finite length. Since A/𝑏A has finite length, this
implies that A has finite length, a contradiction. We have shown that A
is an integral domain.
In particular, (0) is a prime ideal of A. By assumption, (0) is maximal,
so that A is a field. In particular, A is a simple module. In particular, its
length is finite. This contradiction shows that the hypothesis that ℓA (A)
is infinite is false. 

Remark (6.4.15). — We shall establish in the next section (theorem 6.5.16)

that for any noetherian commutative ring A, there is an integer 𝑛, a
sequence (I0 , . . . , I𝑛 ) of ideals of A and a sequence (P1 , . . . , P𝑛 ) of prime
ideals of A such that
0 = I0 ⊂ I1 ⊂ · · · ⊂ I𝑛−1 ⊂ I𝑛 = A
and I 𝑘 /I 𝑘−1 ' A/P 𝑘 for every integer 𝑘 ∈ {1, . . . , 𝑛}. Under assump-
tion (ii) of theorem 6.4.14, all quotients I 𝑘 /I 𝑘−1 are simple A-modules.
This gives another proof that A has finite length.

6.5. Support of a module, associated ideals

In this section, all rings are tacitly assumed to be commutative.
Definition (6.5.1). — Let A be a commutative ring, let M be an A-module.
The support of M is the subset of Spec(A) consisting of all prime ideals P of A
such that MP ≠ 0. It is denoted by SuppA (M), or Supp(M) if no confusion
can arise regarding the ring A.
Let us expand the definition. Let M be an A-module and let P be a
prime ideal. The condition MP ≠ 0 means that there exists 𝑚 ∈ M such
that 𝑚/1 ≠ 0 in MP , hence 𝑠𝑚 ≠ 0 for any 𝑠 ∈ A P. In other words,
298 CHAPTER 6. FURTHER RESULTS ON MODULES

a prime ideal P of A belongs to SuppA (M) if and only if there exists 𝑚 ∈ M

whose annihilator AnnA (𝑚) is contained in P. This can also be rewritten as
Ø
SuppA (M) = V(AnnA (𝑚)).
𝑚∈A

Example (6.5.2). — Let A be a commutative ring and let I be an ideal of A.

For 𝑚 ∈ A/I, one has I ⊂ AnnA (𝑚), hence V(AnnA (𝑚)) ⊂ V(I), so that
SuppA (M) ⊂ V(I). Conversely, AnnA (clI (1)) = I, hence V(I) ⊂ SuppA (M).
This proves that SuppA (A/I) = V(I), the set of all prime ideals of A which
contain I.
Proposition (6.5.3). — Let A be a commutative ring, let M be an A-module and
let N be a submodule of M. Then SuppA (M) = SuppA (N) ∪ SuppA (M/N).
Proof. — Recall (exactness of localization, proposition 3.6.6) that for any
prime ideal P of A, NP identifies with a submodule of MP and one has
an isomorphism (M/N)P ' MP /NP . Consequently, MP = 0 if and only if
NP = (M/N)P = 0. In other words,
P ∉ SuppA (M) ⇔ P ∉ SuppA (N) ∪ SuppA (M/N),
that is to say, SuppA (M) = SuppA (N) ∪ SuppA (M/N). 

Theorem (6.5.4). — Let A be a commutative ring and let M be an A-module.

a) If M ≠ 0, then SuppA (M) ≠ ∅;
b) Let P be a prime ideal in SuppA (M); then every prime ideal containing P
belongs to SuppA (M);
c) One has SuppA (M) ⊂ V(AnnA (M)): any prime ideal of A belonging
to SuppA (M) contains AnnA (M);
d) If M is finitely generated, then SuppA (M) = V(AnnA (M)) is the set of
all prime ideals of A which contain AnnA (M).
Proof. — a) Let us assume that M ≠ 0 and let 𝑚 ∈ M {0}. In other
words, AnnA (𝑚) ≠ A, hence V(AnnA (𝑚)) ≠ ∅ (this follows from Krull’s
theorem, theorem 2.1.3, see remark 2.2.6, b)). Since SuppA (M) con-
tains V(AnnA (𝑚)), this implies that SuppA (M) ≠ ∅.
b) Indeed, every subset of Spec(A) which is a union of closed subsets
satisfies this property. Let P and Q be prime ideals of Q such that P ⊂ Q
 6.5. SUPPORT OF A MODULE, ASSOCIATED IDEALS 299

and P ∈ SuppA (M). Let 𝑚 ∈ M be such that P ∈ V(AnnA (𝑚)). Then

Q ⊃ P ⊃ AnnA (𝑚)), hence Q ∈ V(AnnA (𝑚)), hence Q ∈ SuppA (M).
c) For every 𝑚 ∈ M, one has AnnA (M) ⊂ AnnA (𝑚), hence
V(AnnA (𝑚)) ⊂ V(AnnA (M)). Consequently, SuppA (M) ⊂ V(AnnA (M)).
d) Let (𝑚1 , . . . , 𝑚𝑛 ) be a finite family of elements of M which gener-
ates M. Let us observe that
AnnA (M) = AnnA (𝑚1 ) ∩ · · · ∩ AnnA (𝑚𝑛 ).
Indeed, the inclusion ⊂ is obvious; conversely, if for every 𝑖, one has
𝑎 ∈ AnnA (𝑚 𝑖 ), then 𝑎𝑚 = 0 for every linear combination 𝑚 of the 𝑚 𝑖 s,
hence 𝑎𝑚 = 0 for every 𝑚 ∈ M. In particular,


V(AnnA (M)) = V AnnA (𝑚1 ) ∩ · · · ∩ AnnA (𝑚𝑛 )
= V(AnnA (𝑚1 )) ∪ · · · ∪ V(AnnA (𝑚𝑛 )),
by remark 2.2.6, a).
Using c), we thus have
SuppA (M) ⊂ V(AnnA (𝑚1 )) ∪ · · · ∪ V(AnnA (𝑚𝑛 )),
and the other inclusion follows from the definition of SuppA (M), hence
the desired equality. 

Remark (6.5.5). — Let M be an A-module. If M is finitely generated,

then assertion d) of the theorem says that its support is the closed
subset V(AnnA (M)) of Spec(A). In general, SuppA (M) is not Énecessarily
closed. For example, let us consider A = Z and M = 𝑛>1 (Z/𝑛Z).
Since every element of M is torsion, one has M(0) = 0, hence (0) ∉
SuppZ (M). On the other hand, for every prime number 𝑝, the module M
contains Z/𝑝Z as a submodule, whose support is {(𝑝)}, so that (𝑝) ∈
SuppZ (M). Consequently, SuppZ (M) = {(2), (3), . . .} = Spec(Z) {(0)}
and this set is not closed in Spec(Z).
By assertion b), the support of an A-module is a union of closed subsets,
hence it contains the closure of any of its points. One says that it is closed
under specialization.

Definition (6.5.6). — Let A be a commutative ring and let M be an A-module.

One says that a prime ideal P of A is associated to M if there exists 𝑚 ∈ M
300 CHAPTER 6. FURTHER RESULTS ON MODULES

such that P is minimal among all prime ideals of A containing AnnA (𝑚). The
set of all associated prime ideals of M is denoted AssA (M).
Example (6.5.7). — Let A be a commutative ring, let P be a prime ideal
of A. Then AssA (A/P) = {P}.
Let clP : A → A/P be the canonical surjection. Let 𝑥 ∈ A/P. If 𝑥 = 0,
then AnnA (𝑥) = A and no prime ideal of A contains AnnA (𝑥). Let us
assume that 𝑥 ≠ 0 and let us prove that AnnA (𝑥) = P. The inclusion
P ⊂ AnnA (𝑥) is obvious. Conversely, let 𝑎 ∈ A be such that 𝑥 = clP (𝑎);
one has 𝑎 ∉ P; consequently, if 𝑏 ∈ AnnA (𝑥), the equality 𝑏𝑥 = 0 means
𝑏𝑎 ∈ P, hence it implies 𝑏 ∈ P by the definition of a prime ideal. In
that case, P is the only minimal prime ideal containing AnnA (𝑥). This
implies the claim.
Theorem (6.5.8). — Let A be a commutative ring and let M be an A-module.
a) Multiplication by an element 𝑎 ∈ A is injective in M if and only if 𝑎 does
not belong to any element of AssA (M).
b) An element 𝑎 ∈ A belongs to every element of AssA (M) if and only if
M𝑎 = 0.
c) In particular, AssA (M) = ∅ if and only if M = 0.
Proof. — a) Let P ∈ AssA (M) and let 𝑎 ∈ P. Let 𝑚 ∈ M be such that P is
minimal among the prime ideals containing AnnA (𝑚). By lemma 2.2.15,
𝑎 is a zero-divisor in the ring A/AnnA (𝑚). In other words, there exists
𝑏 ∈ A AnnA (𝑚) such that 𝑎𝑏 ∈ AnnA (𝑚). Consequently, 𝑏𝑚 ≠ 0 and
𝑎𝑏𝑚 = 0. In particular, multiplication by 𝑎 is not injective in M.
Conversely, let 𝑎 ∈ A and 𝑚 ∈ M be such that 𝑎𝑚 = 0 but 𝑚 ≠ 0. One
thus has AnnA (𝑚) ≠ A, so that there exists a minimal prime ideal P that
contains AnnA (𝑚) (lemma 2.2.13). Since 𝑎 ∈ AnnA (𝑚), one has 𝑎 ∈ P.
b) Let us assume that M𝑎 = 0. Let 𝑚 ∈ M; since M𝑎 = 0, there exists an
integer 𝑛 > 1 such that 𝑎 𝑛 𝑚 = 0. Consequently, 𝑎 𝑛 ∈ AnnA (𝑚) and any
prime ideal containing AnnA (𝑚) contains 𝑎. In particular, 𝑎 belongs to
every element of AssA (M).
Conversely, let us assume that M𝑎 ≠ 0 and let 𝑚 ∈ M be such that
𝑚/1 ≠ 0 in M𝑎 . The ideal AnnA (𝑚) does not contain any power 𝑎 𝑛 ,
hence generates an ideal I of A𝑎 such that I ≠ A𝑎 . By lemma 2.2.13
(applied to the ideal I of the ring A𝑎 ), there exists a prime ideal Q of A𝑎
 6.5. SUPPORT OF A MODULE, ASSOCIATED IDEALS 301

which contains I and is minimal among those prime ideals. Let P be the
inverse image of Q in A, so that Q = PA𝑎 . Then, the ideal P is prime
and does not contain 𝑎; it is also a minimal prime ideal among those
containing AnnA (𝑚); in particular, P ∈ AssA (M) and 𝑎 ∉ P.
c) It suffices to apply b) to 𝑎 = 1. 

Proposition (6.5.9). — Let A be a commutative ring, let M be an A-module

and let N be a submodule of M. Then,
AssA (N) ⊂ AssA (M) ⊂ AssA (N) ∪ AssA (M/N).

Proof. — It follows from the definition that any prime ideal P which
is associated to N is also associated to M. Let now P be a prime ideal
which is associated to M, and let 𝑚 ∈ M be such that P is minimal
among the prime ideals containing AnnA (𝑚). Let clN : M → M/N
be the canonical surjection. One has AnnA (clN (𝑚)) ⊃ AnnA (𝑚). If P
contains AnnA (clN (𝑚)), then P is also minimal among the prime ideals
containing AnnA (clN (𝑚)), hence P ∈ AssA (M/N). Let us thus assume
that P does not contain AnnA (clN (𝑚)), and let 𝑏 ∈ AnnA (clN (𝑚)) such
that 𝑏 ∉ P. Let 𝑎 ∈ AnnA (𝑏𝑚); then 𝑎𝑏𝑚 = 0, hence 𝑎𝑏 ∈ AnnA (𝑚),
hence 𝑎𝑏 ∈ P; since 𝑏 ∉ P, this implies 𝑎 ∈ P, hence the inclusions
AnnA (𝑚) ⊂ AnnA (𝑏𝑚) ⊂ P. It follows that the prime ideal P is minimal
among those containing AnnA (𝑏𝑚). This implies P ∈ AssA (N) since, by
choice of 𝑏, we have 𝑏𝑚 ∈ N. 

Proposition (6.5.10). — Let A be a commutative ring, let S be a multiplicative

subset of A, let M be an A-module. A prime ideal P disjoint from S is associated
to the A-module M if and only if S−1 P is associated to the S−1 A-module S−1 M.

Viewing Spec(S−1 A) as a subset of Spec(A) via proposition 2.2.7, this

can be rephrased as the equality AssS−1 A (S−1 M) = AssA (M) ∩ Spec(S−1 A).
Proof. — Let a prime ideal P, disjoint from S, be associated to M. Let 𝑚 ∈
M be such that P is minimal among the prime ideals containing AnnA (𝑚).
The inverse image in A of AnnS−1 A (𝑚/1) contains AnnA (𝑚). Moreover,
let 𝑎 ∈ A be such that (𝑎/1)(𝑚/1) = 0 in S−1 M; then, there exists 𝑠 ∈ S
such that 𝑠𝑎𝑚 = 0, hence 𝑠𝑎 ∈ AnnA (𝑚) and a fortiori 𝑠𝑎 ∈ P. Since P
is disjoint from S, we get 𝑎 ∈ P and 𝑎/1 ∈ S−1 P. Consequently, S−1 P
302 CHAPTER 6. FURTHER RESULTS ON MODULES

is minimal among the prime ideals of S−1 A containing AnnS−1 A (𝑚/1),

hence, S−1 P ∈ AssS−1 A (S−1 M).
In the other direction, assume that S−1 P is associated to S−1 M. Let
𝑚 ∈ M, 𝑠 ∈ S be such that S−1 P is minimal among the prime ideals
of S−1 A containing AnnS−1 A (𝑚/𝑠). Let 𝑎 ∈ AnnA (𝑚); then 𝑎𝑚/𝑠 = 0,
hence 𝑎/1 ∈ AnnS−1 A (𝑚/𝑠) and in particular, 𝑎/1 ∈ S−1 P. Therefore,
there exists 𝑡 ∈ S such that 𝑡𝑎 ∈ P; since S and P are disjoint, 𝑎 ∈ P. In
other words, AnnA (𝑚) ⊂ P.
Let us show that P is minimal among the prime ideals of A contain-
ing AnnA (𝑚). Let Q be a prime ideal of A such that AnnA (𝑚) ⊂ Q ⊂ P.
Then, AnnS−1 A (𝑚/1) ⊂ S−1 Q ⊂ S−1 P. This implies that S−1 Q = S−1 P.
Since P and Q are disjoint from S and prime, we get Q = P, as was to be
shown. 

Corollary (6.5.11). — Let A be a commutative ring, let M be an A-module.

a) A prime ideal of A belongs to the support SuppA (M) if and only if it
contains an ideal of AssA (M).
b) One has AssA (M) ⊂ SuppA (M), and both sets have the same minimal
elements.

Proof. — a) Let 𝑚 ∈ M. Let P be a prime ideal of A. By definition

of the support of a module, P ∈ SuppA (M) if and only if MP ≠ 0.
By theorem 6.5.8, this is itself equivalent to AssAP (MP ) ≠ ∅ Finally,
proposition 6.5.10 implies that this holds if and only if there exists
Q ∈ AssA (M) such that Q ⊂ P.
b) This is essentially a reformulation. Let P ∈ AssA (M); by a), since
P ⊃ P, one has P ∈ SuppA (M).
Let us assume that P is minimal in AssA (M), and let us prove that P is
also minimal in SuppA (M). Let Q ∈ SuppA (M) be such that Q ⊂ P. By
a), there exists an associated ideal P0 ∈ AssA (M) such that Q ⊃ P0; then
P0 ⊂ P, hence P = P0 by minimality, hence P = Q.
In the other direction, let P ∈ SuppA (M) be a minimal element of
the support of M and let us prove that P belongs to AssA (M). Let
𝑚 ∈ M be such that P ⊃ AnnA (𝑚). If Q is a prime ideal of A such that
AnnA (𝑚) ⊂ Q ⊂ P, then Q ∈ SuppA (M), hence P = Q by minimality
 6.5. SUPPORT OF A MODULE, ASSOCIATED IDEALS 303

of P. In other words, P is minimal among the prime ideals of A that

contain AnnA (𝑚). This proves that P ∈ AssA (M). 

Proposition (6.5.12). — Let A be a commutative ring, let M be an A-module

and let P be a prime ideal of A which is finitely generated. Then P is associated
to M if and only if there exists 𝑚 ∈ M such that P = AnnA (𝑚).

Proof. — If P = AnnA (𝑚), then P is obviously a minimal prime contain-

ing AnnA (𝑚), hence P ∈ AssA (M).
To prove the converse assertion, we first treat the case where A is a
local ring and P is its maximal ideal. Let 𝑚 ∈ M be such that P is the only
minimal prime ideal containing AnnA (𝑚). By proposition 2.2.11 applied
to A/AnnA (𝑚), every element of P is nilpotent modulo AnnA (𝑚): for
every 𝑎 ∈ P, there exists an integer 𝑛 > 1 such that 𝑎 𝑛 ∈ AnnA (𝑚).
Since P is finitely generated, there exists an integer 𝑛 > 1 such that
P𝑛 ⊂ AnnA (𝑚). Let us choose a minimal such integer 𝑛. Then P𝑛−1 ·𝑚 ≠ 0,
hence we may choose 𝑎 ∈ P𝑛−1 such that 𝑎𝑚 ≠ 0; then P ⊂ AnnA (𝑎𝑚) ⊂ P,
hence P = AnnA (𝑎𝑚).
Let us now treat the general case. By proposition 6.5.10, one has
PAP ∈ AssAP (MP ). Moreover, the ideal PAP is finitely generated. By
the case of a module over a local ring, there exists an element of MP ,
say 𝑚/𝑎, with 𝑚 ∈ M and 𝑎 ∈ A P, with annihilator PAP . Then
AnnAP (𝑚/1) = PAP .
Let us now assume that P = (𝑎1 , . . . , 𝑎 𝑛 ) and let us construct a se-
quence (𝑚0 , . . . , 𝑚𝑛 ) of multiples of 𝑚 such that for all 𝑘 ∈ {0, . . . , 𝑛},
PAP = AnnAP (𝑚 𝑘 /1) and 𝑎1 𝑚 𝑘 = · · · = 𝑎 𝑘 𝑚 𝑘 = 0. Let us set 𝑚0 = 𝑚.
Assume 𝑚0 , . . . , 𝑚 𝑘−1 are defined. Since 𝑎 𝑘 (𝑚 𝑘−1 /1) = 0, there exists
𝑏 ∈ A P such that 𝑏𝑎 𝑘 𝑚 𝑘−1 = 0; set 𝑚 𝑘 = 𝑏𝑚 𝑘−1 . Since 𝑏 ∉ P, one
has AnnAP (𝑚 𝑘 /1) = AnnAP (𝑚 𝑘−1 /1) = PAP . This constructs the de-
sired sequence by induction. Finally, 𝑚𝑛 is an element of M such that
PAP = AnnAP (𝑚𝑛 /1) and P ⊂ AnnA (𝑚𝑛 ). It follows that P = AnnA (𝑚𝑛 ):
if 𝑎𝑚𝑛 = 0, then 𝑎(𝑚𝑛 /1) = 0, hence 𝑎 ∈ PAP , hence 𝑎 ∈ P. 

Corollary (6.5.13). — Let A be a noetherian ring and let M be an A-module.

A prime ideal P of A is associated to M if and only if there exists 𝑚 ∈ M such
that P = AnnA (𝑚).
304 CHAPTER 6. FURTHER RESULTS ON MODULES

Corollary (6.5.14). — Let A be a commutative ring and let M be a noetherian

A-module. A prime ideal P of A is associated to M if and only if there exists
𝑚 ∈ M such that P = AnnA (𝑚).

Proof. — Let M be a noetherian A-module and let P be a prime ideal

which is associated with M. Let 𝑚 ∈ M be such that P is a minimal prime
ideal among those containing I = AnnA (𝑚). The morphism 𝑎 ↦→ 𝑎𝑚
from A to M induces an isomorphism from A/I to the submodule A𝑚
of M. Since M is a noetherian A-module, so is its submodule A𝑚. Since
the A-submodules of A/I coincide with the ideals of A/I, the ring A/I
is noetherian (see also remark 6.3.9). Moreover, P/I is a minimal prime
ideal among those containing (0) = AnnA/I (1), so that P/I ∈ AssA/I (A/I)
and there exists 𝑥 ∈ A/I such that P/I = AnnA/I (𝑥) (corollary 6.5.13).
Let 𝑎 ∈ A be such that 𝑥 = clI (𝑎). For every 𝑏 ∈ P, one has 𝑏𝑎 ∈ I, hence
𝑏𝑎𝑚 = 0; conversely, if 𝑏 ∈ A statisfies 𝑏𝑎𝑚 = 0, then 𝑏𝑥 = 0 hence 𝑏 ∈ P.
This proves that P = AnnA (𝑎𝑚). 

Remark (6.5.15). — There is a conflicting terminology in the litterature

between two notions of associated primes. Books where the emphasis
lies on noetherian rings, such as Matsumura (1986) usually prefer the
one given by corollary 6.5.13, and Bourbaki (1989) call weakly associated
primes those of definition 6.5.6. In more advanced works on commutative
algebra, the latter definition is prefered; I also believe that the notion
and the exposition of the main results are more natural. However, the
following results require the hypothesis that M is noetherian.

Theorem (6.5.16). — Let A be a commutative ring and let M be a noetherian

A-module. There exist a finite family (M0 , . . . , M𝑛 ) of submodules of M and a
finite family (P1 , . . . , P𝑛 ) of prime ideals of A such that
0 = M0 ⊂ M1 ⊂ · · · ⊂ M𝑛 = M
and M𝑖 /M𝑖−1 ' A/P𝑖 for every 𝑖 ∈ {1, . . . , 𝑛}.

Proof. — Let us begin by a remark. Let N be a submodule of M such

that N ≠ M. Applying corollary 6.5.14 to the noetherian A-module M/N,
there exists 𝑚 ∈ M and a prime ideal P of A such that (A𝑚 + N)/N is
isomorphic to A/P.
 6.5. SUPPORT OF A MODULE, ASSOCIATED IDEALS 305

We set M0 = 0 and apply this remark inductively, constructing an

increasing sequence (M1 , M2 , . . . ) of submodules of M, and a sequence
(P1 , P2 , . . . ) of prime ideals of A such that M𝑖 /M𝑖−1 is isomorphic to
A/P𝑖 for every 𝑖. Since M is noetherian, this strictly increasing sequence
must be finite and the process must stop at some level 𝑛. Then, M𝑛 = M,
hence the theorem. 

Corollary (6.5.17). — Let A be a commutative ring and let M be a noetherian

A-module. Then AssA (M) is a finite set.

Proof. — Let us consider a composition series

0 = M0 ⊂ M1 ⊂ · · · ⊂ M𝑛 = M,
and a sequence (P1 , . . . , P𝑛 ) of prime ideals of A, as given by theo-
rem 6.5.16, so that A/P𝑖 ' AnnA (M𝑖 /M𝑖−1 ) for 1 6 𝑖 6 𝑛.
By proposition 6.5.9, Ass(M) ⊂ Ass(M𝑛−1 ) ∪ Ass(A/P𝑛 ). We had
also explained in example 6.5.7 that Ass(A/P𝑛 ) = {P𝑛 }. Consequently,
Ass(M) ⊂ Ass(M𝑛−1 ) ∪ {P𝑛 }. By induction on 𝑛, we obtain
AssA (M) ⊂ {P1 ; . . . ; P𝑛 }.
In particular, AssA (M) is a finite set. 

Corollary (6.5.18). — Let A be a commutative ring and let M be a noetherian

A-module. Then, M has finite length if and only if all of its associated ideals are
maximal ideals.

Proof. — Let us assume that M has finite length. Let then

0 = M0 ⊂ M1 ⊂ · · · ⊂ M𝑛 = M
be a Jordan-Hölder composition series, so that for every 𝑖, there is a
maximal ideal P𝑖 of A such that M𝑖 /M𝑖−1 ' A/P𝑖 . By the same proof
as for the preceding corollary, AssA (M) is contained in {P1 ; . . . ; P𝑛 }. In
particular, all associated prime ideals of M are maximal ideals.
Conversely, let us assume that all associated prime ideals of M are
maximal and let us consider a composition series
0 = M0 ⊂ M1 ⊂ · · · ⊂ M𝑛 = M,
306 CHAPTER 6. FURTHER RESULTS ON MODULES

and prime ideals P1 , . . . , P𝑛 , as given by theorem 6.5.16, so that for

every 𝑖 ∈ {1, . . . , 𝑛}, one has M𝑖 /M𝑖−1 ' A/P𝑖 . By proposition 6.5.3,
SuppA (M) = {P1 ; . . . ; P𝑛 }.
By corollary 6.5.11, the minimal elements of SuppA (M) are associated
prime ideals, hence are maximal ideals of A, by assumption. This
implies that P1 , . . . , P𝑛 are maximal ideals of A. In particular, M𝑖 /M𝑖−1
is a simple A-module, for every 𝑖, and M has finite length. 

6.6. Primary decomposition

Definition (6.6.1). — Let A be a commutative ring.
One says that an A-module M is coprimary if there AssA (M) has exactly
one element. If AssA (M) = {P}, one also says that M is P-coprimary.
One says that a submodule N of an A-module is primary if M/N is coprimary.
If M/N is P-coprimary, one says that N is P-primary.
Lemma (6.6.2). — Let A be a commutative ring and let M be a nonzero
A-module.
a) The module M is coprimary if and only if for every 𝑎 ∈ A, either the
multiplication by 𝑎 is injective in M, or the fraction module M𝑎 is 0.
p
b) Assume that M is coprimary. If M is finitely generated, then AnnA (M)
is the unique associated prime ideal of M.
Proof. — a) By theorem 6.5.8, the second condition is equivalent to
saying that the union of the associated prime ideals of M (the set of
elements 𝑎 such that multiplication by 𝑎 is not injective in M) coincides
with the intersection of the associated prime ideals of M (the set of
elements 𝑎 such that M𝑎 = 0). Since M ≠ 0, it possesses associated
ideals. Consequently, the union of the associated prime ideals contains
the intersection of the prime ideals. If there are at least two distinct
associated prime ideals, say P and Q, then P ∪ Q strictly contains P ∩ Q,
so that that containment is strict unless there is exactly one associated
prime ideal, that is to say, unless M is coprimary.
b) Let P be the unique associated prime ideal of M. Let 𝑎 ∈ A. By a),
either 𝑎 ∉ P and the multiplication by 𝑎 is injective on M, or 𝑎 ∈ P and
 6.6. PRIMARY DECOMPOSITION 307

one has M𝑎 = 0. Since M is finitely generated, the condition M𝑎 = 0 is

equivalent to the existence of an integer 𝑛 such that 𝑎 𝑛 M = 0, that is,
by every power 𝑎 𝑛
p
𝑎 ∈ AnnA (M). In the first case, the multiplication
p
of 𝑎 is still injective
p on M, so that 𝑎 ∉ AnnA (M), since M ≠ 0. This
proves that AnnA (M) = P. 

Examples (6.6.3). — Let A be a commutative ring.

a) Let I be an ideal of A. The ideal I is a primary submodule of A if
and only if for any 𝑎, 𝑏 ∈ A such that
√ 𝑎𝑏 ∈ I and 𝑏 ∉ I, there exists 𝑛 > 1
𝑛
such that 𝑎 ∈ I. If this holds, then I is a prime ideal of A.
b) If I is a prime ideal of A, then I is primary. This follows either from
the characterization in a), or from the computation of AssA (A/I) done in
example 6.5.7.
c) Assume that A is a principal ideal domain. Let us show that
primary ideals of A other than (0) are the ideals of the form (𝑝 𝑛 ), for
some irreducible element 𝑝 and some integer 𝑛 > √ 1. Let I be a primary
ideal distinct from (0), let 𝑝 ∈ A be such that I = (𝑝). Necessarily,
𝑝 is the only prime divisor of I, so that there exists 𝑛 > 1 such that
I = (𝑝 𝑛 ). Conversely, let us prove that such ideals (𝑝 𝑛 ) are (𝑝)-primary.
Let 𝑎, 𝑏 ∈ A be such that 𝑎𝑏 ∈ I but 𝑏 ∉ I. Then, 𝑝 𝑛 divides 𝑎𝑏 but 𝑝 𝑛
does not divide 𝑏; necessarily, 𝑝 divides 𝑎, so that 𝑎 𝑛 ∈ I.
In particular, any ideal of A, say (𝑝1𝑛1 . . . 𝑝 𝑟𝑛𝑟 ) can be written as an
intersection of the primary ideals (𝑝 𝑖𝑛 𝑖 ). The primary decomposition, due to
Lasker and Noether, generalizes this fact to all noetherian rings.
√
d) Let I be an ideal of A such that P = I is a maximal ideal of A. Then I
is P-primary. (This generalizes the previous example.) Let 𝑎, 𝑏 ∈ A be
such that 𝑎𝑏 ∈ I but 𝑎 ∉ P. Since P is maximal, the image of 𝑎 in A/P√is
invertible and there exists 𝑐 ∈ A such that 𝑥 = 1 − 𝑎𝑐 ∈ P. Since P = I,
there exists 𝑛 > 1 such that 𝑥 𝑛 = (1−𝑎𝑐)𝑛 ∈ I. Expanding the power, there
is 𝑦 ∈ A such that 𝑥 𝑛 = 1 − 𝑦𝑎𝑐. Then, 𝑏 = (𝑥 𝑛 + 𝑦𝑎𝑐)𝑏 = 𝑥 𝑛 𝑏 + 𝑦𝑎𝑏𝑐 ∈ I.
It is however simpler to apply the general theory. (Is it really?) √
Observe that a prime ideal of A contains I if and only if it contains I.
Consequently, the support of A/I is equal to {P}. Since AssA (A/I) and
SuppA (A/I) have the same minimal elements (corollary 6.5.11), one has
AssA (A/I) = {P} and I is a P-primary ideal.
308 CHAPTER 6. FURTHER RESULTS ON MODULES

Lemma (6.6.4). — Let A be a ring, Let M be an A-module and let N, N0 be

two submodules of M.
a) Then, AssA (M/(N ∩ N0)) ⊂ AssA (M/N) ∪ AssA (M/N0).
b) Let P be a prime ideal of A such that N, N0 are P-primary; then N ∩ N0 is
P-primary.
c) Let P be a prime ideal of A such that N is P-primary. If N ∩ N0 ≠ N0,
then P ∈ AssA (M/(N ∩ N0)).
Proof. — a) Let us set V = M/(N ∩ N0) and let V0 be the submodule
N0/(N ∩ N0) of V, so that V/V0 is isomorphic to M/N0. In particular,
AssA (V/V0) = AssA (M/N0). Then, N0 ∩N is the kernel of the composition
of the inclusion map N0 → M with the canonical surjection M → M/N,
so that V0 = N0/(N ∩ N0) is isomorphic to a submodule of M/N. By
proposition 6.5.9, AssA (V0) ⊂ AssA (M/N) and AssA (V) ⊂ AssA (V0) ∪
AssA (V/V0). In particular, AssA (V) ⊂ AssA (M/N) ∪ AssA (M/N0), as was
to be shown.
b) Let P be a prime ideal of A and assume, moreover, that both N
and N0 are P-primary submodules of M. By a), we have AssA (V) ⊂ {P}.
Since V ≠ 0, AssA (V) ≠ ∅; consequently, AssA (V) = {P}, which means
that N ∩ N0 is a P-primary submodule of M.
c) By assumption, AssA (M/N) = {P}. Moreover, N0 ∩ N ≠ N0, so
that V0 ≠ 0. Being a nonempty subset of AssA (M/N) = {P}, one
has AssA (V0) = {P}. Since AssA (V) contains AssA (V0), the lemma is
proved. 

Theorem (6.6.5) (Lasker–Noether). — Let A be a commutative ring and let

M be a noetherian A-module. Any submodule of M is the intersection of a finite
family of primary submodules of M.
Proof. — Assume that the proposition does not hold. Since the mod-
ule M is noetherian, there is a largest submodule N of M which cannot
be written as the intersection of finitely many primary submodules of M.
Observe that N ≠ M (for M is the intersection of the empty family) and
N is not a primary submodule of M (for one could write N = N). By
lemma 6.6.2, there exists 𝑎 ∈ A such that N𝑎 ≠ M𝑎 and 𝑚 ∈ M N such
that 𝑎𝑚 ∈ N.
 6.6. PRIMARY DECOMPOSITION 309

For every integer 𝑘, let P 𝑘 be the submodule of M consisting of

those 𝑝 ∈ M such that 𝑎 𝑘 𝑝 ∈ N. By assumption, P 𝑘 ≠ M: otherwise, we
would have 𝑎 𝑘 M ⊂ N, hence M𝑎 = N𝑎 .
The family (P 𝑘 ) is an increasing family of submodules of M, hence
it is stationary, because M is noetherian. Let 𝑘 ∈ N such that P 𝑘 =
P 𝑘+1 . It is clear that N ⊂ (N + A𝑚) ∩ (N + 𝑎 𝑘 M). Conversely, let
𝑛 ∈ (N + A𝑚) ∩ (N + 𝑎 𝑘 M); let us write 𝑛 = 𝑛 0 + 𝑏𝑚 = 𝑛 00 + 𝑎 𝑘 𝑝, where
𝑛 0 , 𝑛 00 ∈ N, 𝑏 ∈ A, and 𝑝 ∈ M. Then, 𝑎𝑛 = 𝑎𝑛 0 + 𝑎𝑏𝑚 ∈ N, hence
𝑎 𝑘+1 𝑝 = 𝑎(𝑛 − 𝑛 00) ∈ N. Consequently, 𝑝 ∈ P 𝑘+1 ; by the choice of 𝑘,
it follows that 𝑝 ∈ P 𝑘 , so that 𝑎 𝑘 𝑝 ∈ N and 𝑛 ∈ N. This shows that
N = (N + A𝑚) ∩ (N + 𝑎 𝑘 M).
On the other hand, 𝑚 ∉ N, so that N ( N + A𝑚; moreover, N𝑎 ≠ M𝑎 , so
that 𝑎 𝑘 M ≠ N, and N ( N+ 𝑎 𝑘 M. By the choice of N, the submodules N+
A𝑚 and N + 𝑎 𝑘 M can be written as the intersections of finitely many
primary submodules of M. Then so can N, a contradiction. 

Definition (6.6.6). — Let A be a ring, let M be an A-module and let N be a

submodule of M. A primary decomposition of N in M is a finite sequence
(N1 , . . . , N𝑛 ) of primary submodules of M such that N = N1 ∩ · · · ∩ N𝑛 .
A primary decomposition as above is said to be minimal if the following two
properties also hold:
(i) For every 𝑖 ≠ 𝑗, AssA (M/N𝑖 ) ≠ AssA (M/N 𝑗 );
(ii) For every 𝑗, N ≠
Ñ
𝑖≠𝑗 N𝑖 .
Remark (6.6.7). — Let us restate the definition in the important case of
ideals. Let A be a ring, let I be an ideal of A. A primary decomposition of I
in A is an expression of the form I = Q1 ∩ · · · ∩ Q𝑛 , where Q1 , . . . , Q𝑛 are
primary ideals of A. Such a decomposition is said to be minimal if, moreover,
√
(i) For every 𝑖 ≠ 𝑗, Q𝑖 ≠ Q 𝑗 ;
p

(ii) For every 𝑗, I ≠

Ñ
𝑖≠𝑗 Q𝑖 .
Corollary (6.6.8). — Let A be a ring and let M be a noetherian A-module.
Every submodule of M possesses a minimal primary decomposition.
Proof. — Let N be a submodule of M. By theorem 6.6.5, we know that
N admits a primary decomposition: there is an integer 𝑛 and primary
submodules N1 , . . . , N𝑛 of M such that N = N1 ∩ · · · ∩ N𝑛 . Let us choose
310 CHAPTER 6. FURTHER RESULTS ON MODULES

such a decomposition for which 𝑛 is minimal and let us prove that we

have indeed a minimal primary decomposition.
Let P be a prime ideal of A which is associated to at least two mod-
ules M/N 𝑗 . By lemma 6.6.4, the intersection of all modules N𝑖 such
that AssA (M/N𝑖 ) = {P} is a P-primary submodule of M. This gives
rise to a new primary decomposition of N, with strictly less factors. By
the minimality assumption on 𝑛, this implies that the associated prime
ideals of the N𝑖 are pairwise distinct.
Now, if we can remove one of the modules from the list without
changing the intersection, we get a new primary decomposition with
one less factor. By the minimality assumption on 𝑛, this cannot hold. 

Theorem (6.6.9). — Let A be a ring, let M be an A-module, let N be a submodule

of M. Let (N1 , . . . , N𝑛 ) be a minimal primary decomposition of N; for every 𝑖,
let P𝑖 be the unique prime ideal associated to M/N𝑖 .
a) One has AssA (M/N) = {P1 , . . . , P𝑛 }.
b) If P𝑖 is minimal among the associated primes of M/N, then
N𝑖 = M ∩ NP𝑖 = {𝑚 ∈ M ; ∃𝑎 ∉ P𝑖 , 𝑎𝑚 ∈ N}.

Proof. — a) By induction on 𝑛, it follows from lemma 6.6.4 that

AssA (M/N) ⊂ {P1 , . . . , P𝑛 }. This inclusion does not depend on the
fact that the given primary decomposition is minimal. Let us now
show that for every 𝑖, P𝑖 is an associated prime ideal to M/N. Let
N0 = 𝑗≠𝑖 N 𝑗 , so that N = N𝑖 ∩ N0. Since the given decomposition
Ñ
is minimal, N ≠ N0. Since M/N𝑖 is P𝑖 -primary, it then follows from
lemma 6.6.4 that AssA (M/N) contains P𝑖 , as required.
b) Let 𝑗 ∈ {1, . . . , 𝑛} be such that 𝑗 ≠ 𝑖. By proposition 6.5.10,
AssAP𝑖 ((M/N 𝑗 )P𝑖 ) is the set of prime ideals of the form PAP𝑖 where
P ∈ Ass(M/N 𝑗 ) = {P 𝑗 } is a prime ideal of A contained in P𝑖 . By mini-
mality of P𝑖 , P 𝑗 is not contained in P𝑖 , so that AssAP𝑖 ((M/N 𝑗 )P𝑖 ) = ∅. By
theorem 6.5.8, (M/N 𝑗 )P𝑖 = 0 and (N 𝑗 )P𝑖 = MP𝑖 . Consequently,
NP𝑖 = (N1 )P𝑖 ∩ · · · ∩ (N𝑛 )P𝑖 = (N𝑖 )P𝑖 .
Let now N0𝑖 = M ∩ (N𝑖 )P𝑖 . It is obvious that N𝑖 ⊂ N0𝑖 . Conversely,
let 𝑚 ∈ N0𝑖 , so that there exists 𝑎 ∉ P𝑖 such that 𝑎𝑚 ∈ N𝑖 . Since
 6.6. PRIMARY DECOMPOSITION 311

AssA (M/N𝑖 ) = {P𝑖 }, multiplication by 𝑎 is injective on M/N𝑖 , hence

𝑚 ∈ N𝑖 . 

Theorem (6.6.10) (Krull’s intersection theorem). — Let A be a commutative

ring, let M be a noetherian A-module and let I be a finitely generated ideal
of A. Let N = 𝑛>0 I𝑛 M. Then N = IN and there exists 𝑎 ∈ I such that
Ñ
(1 + 𝑎)N = 0.

Proof. — We first prove that N = IN. To that aim, let us consider

a primary decomposition of the submodule IN of M, namely, IN =
N1 ∩ · · · ∩ N𝑚 ; for 𝑗 ∈ {1, . . . , 𝑚}, let P 𝑗 be the unique prime ideal of A
which is associated to M/N 𝑗 . Let us assume, by contradiction, that
N ≠ IN and let 𝑥 ∈ N IN. Let 𝑗 ∈ {1, . . . , 𝑚} be such that 𝑥 ∉ N 𝑗 .
Let 𝑎 ∈ I; one has 𝑎𝑥 ∈ IN ⊂ N 𝑗 , so that the multiplication by 𝑎
on M/N 𝑗 is not injective. By the definition of a primary submodule,
one has (M/N 𝑗 )𝑎 = 0. Since M is finitely generated, there exists an
integer 𝑛 > 1 such that 𝑎 𝑛 M ⊂ N 𝑗 .
Since the ideal I is finitely generated, there exists an integer 𝑘 > 1
such that I 𝑘 M ⊂ N 𝑗 . Now, using that 𝑥 ∈ N = 𝑛>0 I𝑛 M, we see that
Ñ
𝑥 ∈ I 𝑘 M ⊂ N 𝑗 , a contradiction which establishes that N = IN.
Since M is noetherian, its submodule N is a finitely generated A-
module. By Nakayama’s theorem (corollary 6.1.4), there exists 𝑎 ∈ I such
that (1 + 𝑎)N = 0. 

Corollary (6.6.11). — Let A be a noetherian commutative ring and let I be an

ideal of A. Under each of the following hypotheses, one has 𝑛>0 I𝑛 = 0:
Ñ

(i) Every element of 1 + I is regular;

(ii) The ideal I is contained in the Jacobson radical of A;
(iii) One has I ≠ A and the ring A is an integral domain.

Proof. — Let J = 𝑛>0 I𝑛 . By Krull’s intersection theorem 6.6.10, there

Ñ
exists 𝑎 ∈ I such that (1 + 𝑎)J = 0. Under each of the hypotheses (i), (ii),
(iii), the element 1 + 𝑎 of A is regular: this is obvious in case (i); in case (ii),
1 + 𝑎 is invertible (lemma 2.1.6); in case (iii), 1 + 𝑎 ≠ 0. Consequently,
J = 0. 
312 CHAPTER 6. FURTHER RESULTS ON MODULES

Exercises
1-exo251) Show that in Nakayama’s lemma, one cannot omit the hypothesis that the
module be finitely generated. (Consider the Z-module Q.)
2-exo250) This exercise leads to another proof of Nakayama’s lemma. Let A be a ring
and let J be its Jacobson radical. Let M be a finitely generated right A-module such
that M ≠ 0.
Let N be a maximal submodule of M (exercise 6/9). Show that MJ ⊂ N; conclude
that M ≠ MJ.
3-exo178) Let A be a ring, let I be a finitely generated two sided ideal of A such that
I = I2 . Show that there exists a central element 𝑒 ∈ A such that 𝑒 = 𝑒 2 and I = 𝑒A = A𝑒.
(Apply Nakayama’s lemma to find an element 𝑎 ∈ I such that I(1 + 𝑎) = 0.)
4-exo117) Let A be a commutative ring, let V be a finitely generated A-module.
a) For any maximal ideal M of A, one defines
𝑑V (P) = dimAP /PAP (VP /PVP ).
b) Let M be a maximal ideal of A. Prove that 𝑑V (M) = dimA/M (V/MV).
c) Let P be a prime ideal of A. Show that the AP -module VP is generated by 𝑑V (P)
elements.
d) Show that there exists 𝑎 ∈ A P such that the A𝑎 -module V𝑎 is generated by 𝑑V (P)
elements.
e) Let Q be a prime ideal of A such that 𝑎 ∉ Q. Show that 𝑑V (Q) 6 𝑑V (P). (In other
words, the function P ↦→ 𝑑V (P) from Spec(A) to N is upper-semicontinuous, when
Spec(A) is endowed with the Zariski topology.)
5-exo642) Let A be a ring, let M, N be A-modules and let 𝑓 : M → N be a morphism.
Assume that M has finite length.
a) Show that Ker( 𝑓 ) and Im( 𝑓 ) have finite lengths and that ℓA (Im( 𝑓 )) + ℓ A (Ker( 𝑓 )) =
ℓA (M).
b) Assume that ℓA (N) > ℓA (M); prove that 𝑓 is not injective.
c) Assume that ℓA (N) > ℓA (M); prove that 𝑓 is not surjective.
d) Assume that N = M (so that 𝑓 is an endomorphism). Show that the following
conditions are equivalent: (i) 𝑓 is bijective; (ii) 𝑓 is injective; (iii) 𝑓 is surjective.
6-exo566) Let A be a ring, let M be an A-module of finite length and let 𝑢 be an
endomorphism of M.
a) Show that there exists a smallest integer 𝑝 > 0 such that Ker(𝑢 𝑛 ) = Ker(𝑢 𝑝 ) for
any 𝑛 > 𝑝.
b) Show that there exists a smallest integer 𝑞 > 0 such that Im(𝑢 𝑛 ) = Im(𝑢 𝑞 ) for any
𝑛 > 𝑞.
c) Show that 𝑝 = 𝑞.
d) Show that Ker(𝑢 𝑝 ) and Im(𝑢 𝑝 ) are direct summands in M.
 EXERCISES 313

7-exo567) Let A be a ring, let M be an A-module, let (S𝑖 )𝑖∈I be a family of simple
Í
submodules of M such that M = 𝑖∈I S𝑖 . (One says that M is a semisimple A-module.)
a) Let N be a submodule of M. Using Zorn’s lemma that there exists a maximal
subset J of I such that N and the submodules S 𝑗 , for 𝑗 ∈ J, are in direct sum in their
sum. (You may restrictÉ to the case

where I is finite, if you wish to avoid Zorn’s lemma.)
Show that M = N ⊕ 𝑗∈J S 𝑗 .
b) In particular, any submodule of M has a direct summmand in M.
É
c) Show that there exists J ⊂ I such that M is isomorphic to the module 𝑗∈J S 𝑗 .
8-exo574) Let K be a field.
a) For any non-zero polynomial P ∈ K[X], show that K[X]/(P) is a K[X]-module of
finite length equal to the number of irreducible factors of P (repeated according to
their multiplicities). What happens if K is algebraically closed.
b) Let A = K[X1 , . . . , X𝑛 ]. Show that an A-module M has finite length if and only if
dimK (M) < ∞. (Use Hilbert’s Nullstellensatz.)
c) If, moreover, K is algebraically closed, prove that ℓA (M) = dimK (M).
9-exo584) Let A be a ring and let M be a non-zero right A-module which is finitely
generated.
a) Show that the set of all submodules of M which are distinct from M is inductive.
b) Show that for any submodule N of M such that N ≠ M, there exists a submodule P
of M such that N ⊂ P ⊂ M and M/P is a simple A-module.
c) Assume that A possesses a unique maximal right ideal I. Show that
HomA (M, A/I) ≠ 0.
d) Set A = Z and M = Q. Show that there does not exist any submodule P ⊂ M such
that M/P is simple.
10-exo622) Let A be a principal ideal domain.
a) Let 𝑎 be any nonzero element of A. Prove that the A-module A/(𝑎) has finite length
and compute its length in terms of a decomposition of 𝑎 as a product of irreducible
elements.
b) Using the Jordan-Hölder Theorem, give a second proof of the uniqueness property
of decomposition of 𝑎 as a product of irreducible elements.
11-exo881) Let A be a ring. One says that an A-module M is indecomposable if M ≠ 0
and if there does not exist non-zero submodules N, N0 of M such that M = N ⊕ N0.
a) Let M be an A-module such that EndA (M) is a local ring. Prove that M is
indecomposable.
b) Let M be an indecomposable A-module of finite length. Prove that EndA (M) is a
local ring.
c) Let M be an A-module of finite length. Prove that there exists a finite family
(N1 , . . . , N𝑟 ) of indecomposable submodules of M such that M = N1 ⊕ · · · ⊕ N𝑟 .
314 CHAPTER 6. FURTHER RESULTS ON MODULES

d) Let M be an A-module and let N, N0 be submodules of M such that M = N ⊕ N0.

Assume that N is indecomposable. Prove that for every 𝑢 ∈ EndA (M), either 𝑢 or
idM −𝑢 induces an isomorphism from N to a direct summand of N0 in M.
e) Let M be an A-module of finite length. Let (N1 , . . . , N𝑟 ) and (N01 , . . . , N0𝑠 ) be finite
families of indecomposable submodules of M such that M = N1 ⊕· · ·⊕N𝑟 = N01 ⊕· · ·⊕N0𝑠 .
Prove that 𝑟 = 𝑠 and that there exists a bijection 𝜎 ∈ 𝔖𝑟 such that N0𝑗 ' N𝜎(𝑗) for every
𝑗 ∈ {1, . . . , 𝑟}. (Theorem of Krull–Remak–Schmidt)
12-exo1136) Let A be a local noetherian commutative ring, let M be its maximal ideal.
Let I be an ideal of A. Show that A/I has finite length if and only if there exists 𝑛 > 0
such that M𝑛 ⊂ I.
13-exo066) Let A be a ring, let (I𝑛 ) be an increasing family of left ideals of A. Let
Ð
I = I𝑛 . Show that I is a left ideal of A. If I is finitely generated, show that the
sequence (I𝑛 ) is stationary.
14-exo135) Let A be a ring, let I, J be two-sided ideals of A such that I ∩ J = (0). Show
that A is left noetherian if and only if both A/I and A/J are left noetherian.
15-exo050) Examples of non-noetherian rings. Show that the following commutative rings
are not noetherian.
a) The ring 𝑘[X1 , X2 , . . . , X𝑛 , . . . ] of polynomials in infinitely many indeterminates
with coefficients in a non zero commutative ring 𝑘;
b) The ring 𝒞 0 (R, R) of real valued continuous functions on R;
c) The ring 𝒞 ∞ (R, R) of real valued indefinitely differentiable functions on R. Show
however that the ideal of functions vanishing at 0 is a principal ideal;
d) The subring of C[X, Y] generated as a submodule by C and the ideal (X).
16-exo182) Let ℱ be the set of all polynomials P ∈ Q[X] such that P(𝑛) ∈ Z for every
𝑛 ∈ Z.
a) Show that ℱ is a Z-subalgebra of Q[X].
b) Let 𝑓 : Z → Z be a function. One assumes that 𝑓 (0) ∈ Z and that there exists
P ∈ ℱ such that 𝑓 (𝑛) = P(𝑛 + 1) − P(𝑛) for every 𝑛 ∈ Z. Show that there exists a unique
element Q ∈ ℱ such that 𝑓 (𝑛) = Q(𝑛) for every 𝑛 ∈ Z.



X

c) Show that the polynomials X0 = 1, X1 = X, X2 = X(X − 1)/2, . . . , 𝑝 =
X(X − 1) . . . (X − 𝑝 + 1)/𝑝!, . . . form a basis of ℱ as a Z-module.
d) Show that the ring ℱ is not noetherian.
17-exo064) Let A be a ring, let M be a noetherian right A-module and let 𝜑 : M → M
be an endomorphism of M. Show that there exists an integer 𝑛 > 1 such that
Ker(𝜑 𝑛 ) ∩ Im(𝜑 𝑛 ) = (0).
18-exo648) Let A be a commutative ring which is not noetherian. Let ℱ be the set of all
ideals of A which are not finitely generated, so that ℱ ≠ ∅.
a) Prove that ℱ is an inductive set for the inclusion relation. Let P be a maximal
element of ℱ. In the rest of the exercise, we shall prove that P is a (non-finitely
 EXERCISES 315

generated) prime ideal of A. We argue by contradiction. Let 𝑎, 𝑏 ∈ A be such that

𝑎𝑏 ∈ P while 𝑎 ∉ P and 𝑏 ∉ P.
b) Prove that there exist 𝑢1 , . . . , 𝑢𝑚 ∈ P and 𝑣1 , . . . , 𝑣 𝑛 ∈ A such that P + (𝑎) =
(𝑢1 , . . . , 𝑢𝑚 , 𝑎) and (P : (𝑎)) = (𝑣1 , . . . , 𝑣 𝑛 ).
c) Prove that P = (𝑢1 , . . . , 𝑢𝑚 , 𝑎𝑣1 , . . . , 𝑎𝑣 𝑛 ). Derive from this contradiction that P is
a prime ideal of A.
This proves that a commutative ring is noetherian if and only if every prime ideal is finitely
generated, a theorem of I. S. Cohen.
19-exo808) Let A be a commutative ring. One assumes that for every maximal ideal M
of A, the fraction ring AM is noetherian and that, for every nonzero 𝑎 ∈ A, the set of
maximal ideals of A containing 𝑎 is finite.
Let I be a nonzero ideal of A. Let M1 , . . . , M𝑛 be the maximal ideals of A that
contain I.
a) Prove that there exist elements 𝑎1 , . . . , 𝑎 𝑚 ∈ I such that every maximal ideal of A
containing (𝑎 1 , . . . , 𝑎 𝑚 ) is one of the M 𝑗 .
b) For every 𝑖 ∈ {1, . . . , 𝑛}, prove that there exist an integer 𝑚 𝑖 and elements 𝑏 𝑖,𝑗 (for
𝑗 ∈ {1, . . . , 𝑚 𝑖 }) such that 𝑏 𝑖,𝑗 /1, . . . , 𝑏 𝑖,𝑚𝑖 /1 generate the ideal IAM𝑖 of AM𝑖 .
c) Prove that I is finitely generated.
d) Prove that A is noetherian.
20-exo879) Let R be a commutative noetherian ring and let I be an ideal of R; let
(𝑎 1 , . . . , 𝑎 𝑛 ) be a generating family of I. Let J = 𝑘>1 I 𝑘 and let 𝑏 ∈ J.
Ñ

a) Show that for every integer 𝑘 > 1, there exists a polynomial P 𝑘 ∈ R[T1 , . . . , T𝑛 ],
homogeneous of degree 𝑘, such that 𝑏 = P 𝑘 (𝑎 1 , . . . , 𝑎 𝑛 ).
b) Show that there exist an integer 𝑚 > 1 and, for every 𝑘 ∈ {1, . . . , 𝑚}, a polynomial
Q 𝑘 ∈ R[T1 , . . . , T𝑛 ] which is homogeneous of degree 𝑘, such that P𝑚+1 = Q𝑚 P1 + · · · +
Q1 P𝑚 .
c) Prove that there exists 𝑎 ∈ I such that 𝑏 = 𝑎𝑏. In particular, one has J = IJ. (This
proof of Krull’s intersection theorem is due to Perdry (2004).)
É∞ ring R is graded if there is a family (R𝑛 )𝑛>0
21-exo1126) One says that a commutative
of subgroups of (R, +) such that R = 𝑛=0 R𝑛 and R𝑛 · R𝑚 ⊂ R𝑛+𝑚 for all 𝑛, 𝑚 > 0.
É
a) Show that R0 is a subring of R and that I = 𝑛>1 R𝑛 is an ideal of R.
b) One assumes that R0 is a noetherian ring and that R is finitely generated as an R0
algebra. Show that R is noetherian.
c) Conversely, let us assume that R is a noetherian ring. Show that R0 is noetherian.
Show that there is an integer 𝑟 > 0, elements 𝑥1 , . . . , 𝑥 𝑟 ∈ R, integers 𝑛1 , . . . , 𝑛 𝑟 such
that 𝑥 𝑖 ∈ R𝑛 𝑖 for every 𝑖 and such that I = (𝑥 1 , . . . , 𝑥 𝑟 ). Show by induction on 𝑛 that
R𝑛 ⊂ R0 [𝑥1 , . . . , 𝑥 𝑟 ]. Conclude that R is a finitely generated R0 -algebra.
22-exo1127) Let R = ⊕R𝑛 be a commutative noetherian graded ring. One says that
a R-module M is graded if there is a family (M𝑛 )𝑛>0 of subgroups of M such that
𝑛>0 M𝑛 and such that R𝑛 · M𝑛 ⊂ M𝑛+𝑚 for all 𝑚, 𝑛 > 0.
É
M=
316 CHAPTER 6. FURTHER RESULTS ON MODULES

Let M be a graded R-module.

a) Observe that M𝑛 is a R0 -module. If M is finitely generated, show that for each 𝑛,
M𝑛 is a finitely generated R0 -module.
b) Assume that R0 is an artinian ring. Let PM (𝑡) be the power series with integer
coefficients given by
∞
Õ
PM (𝑡) = ℓ R0 (M𝑛 )𝑡 𝑛 .
𝑛=0
Assume that there is an integer 𝑟, elements 𝑥1 , . . . , 𝑥 𝑟 ∈ R, integers 𝑛1 , . . . , 𝑛 𝑟 such that
𝑥 𝑖 ∈ R𝑛 𝑖 for every 𝑖 and such that R = R0 [𝑥1 , . . . , 𝑥 𝑟 ]. Prove by induction on 𝑟 that there
exists a polynomial 𝑓M ∈ Z[𝑡] such that
𝑟
Ö
𝑓M (𝑡) = PM (𝑡) (1 − 𝑡 𝑛 𝑖 ).
𝑖=1

c) One assumes moreover that 𝑛 𝑖 = 1 for every 𝑖. Show that there is a polynomial 𝜑M ∈
Q[𝑡] such that
ℓR0 (M𝑛 ) = 𝜑M (𝑛)
for every large enough integer 𝑛.
23-exo858) Let A be a noetherian commutative ring and let I be an ideal of A.
a) Let B be the set of all polynomials P ∈ A[T] of the form P = 𝑎 𝑛 T𝑛 , where 𝑎 𝑛 ∈ I𝑛
Í
for every 𝑛. Show that B is a noetherian ring. (Use exercise 6/21.)
b) Let M be a finitely generated A-module and let N be a submodule of M. Let
M[T] = M(N) be considered as an A[T]-module via T · (𝑚0 , 𝑚1 , . . . ) = (0, 𝑚0 , 𝑚1 , . . . ).
Let P (resp. Q) be the subset of M[T] consisting of sequences (𝑚𝑛 ) such that 𝑚𝑛 ∈ I𝑛 M
(resp. such that 𝑚𝑛 ∈ N ∩ I𝑛 M) for every 𝑛 ∈ N. Prove that P is a B-submodule of M[T],
and that Q is B-submodule of P.
c) Prove that there exists 𝑚 ∈ N such that N ∩ I𝑛 M = I𝑛−𝑚 (N ∩ I𝑚 M) for every
integer 𝑛 > 𝑚. (A result known as the Artin–Rees lemma.)
d) Let 𝑓 : M → N be a morphism of finitely generated A-modules. Prove that there
exists an integer 𝑚 such that
𝑓 −1 (I𝑛 N) = Ker( 𝑓 ) + I𝑛−𝑚 · 𝑓 −1 (I𝑚 M) and 𝑓 (M) ∩ I𝑛 M ⊂ 𝑓 (I𝑛−𝑚 M)
for every integer 𝑛 > 𝑚.
24-exo859) Let A be a noetherian commutative ring and let I be an ideal of A. Let M be
a finitely generated A-module and let N = 𝑛∈N I𝑛 M.
Ñ

a) Prove that there exists 𝑎 ∈ I such that (1 + 𝑎)N = 0. (Use exercise 6/23 and Nakayama’s
lemma.)
b) Assuming that I is contained in the Jacobson radical of A, prove that N = 0.
c) Let P be a prime ideal of A containing I. Prove that there exists 𝑎 ∈ A P such
that 𝑎N = 0.
 EXERCISES 317

d) Assume that A is an integral domain, I ≠ A, and M is torsion free. Prove that

N = 0.
25-exo047) In this exercise, we explicit all Z-submodules of Q.
For any nonzero 𝑎 ∈ Q, let 𝑣 𝑝 (𝑎) be the exponent of 𝑝 in the decomposition of 𝑎 as a
product of prime factors; we also set 𝑣 𝑝 (0) = +∞. Let V = 𝑝 prime (Z ∪ {−∞, +∞}) and
Î
𝑣 : Q → V be the map given by 𝑣(𝑎) = (𝑣 𝑝 (𝑎))𝑝 . The set V is endowed with the product
ordering: for two families 𝑎 = (𝑎 𝑝 )𝑝 and 𝑏 = (𝑏 𝑝 )𝑝 in V, say that 𝑎 6 𝑏 if 𝑎 𝑝 6 𝑏 𝑝 for
every prime number 𝑝.
a) Let 𝑥 and 𝑦 be two elements of Q such that 𝑥Z ⊂ 𝑦Z; show that 𝑣(𝑦) 6 𝑣(𝑥).
b) Show that any subset of V admits a least upper bound and a greatest lower bound
in V.
c) Let M be a submodule of Q. Assume that M is finitely generated. Show that there
exists 𝑚 ∈ Q such that M = 𝑚Z. Show that the element 𝑣(𝑚) of V does not depend on
the choice of a generator 𝑚 ∈ M; it will be denoted 𝑣(M). Then, prove that
M = {𝑥 ∈ Q ; 𝑣(𝑥) > 𝑣(M)}.
d) For an arbitrary submodule of Q, set
𝑣(M) = inf 𝑣(𝑥).
𝑥∈Q

Conversely, for any 𝑢 ∈ V, set

M𝑢 = {𝑥 ∈ Q ; 𝑣(𝑥) > 𝑢}.
Show that M𝑢 = 0 if and only if 𝑝 sup(0, 𝑢𝑝 ) = +∞; explicitly, M𝑢 = 0 if and only if
Í
one of the two following conditions holds:
(i) There exists 𝑝 such that 𝑢𝑝 = +∞;
(ii) The set of all prime numbers 𝑝 such that 𝑢𝑝 > 0 is infinite.
e) Show that the map M ↦→ 𝑣(M) induces a bijection from the set of all nonzero
submodules of Q to the subset V0 of V consisting of families 𝑢 ∈ V such that
𝑝 sup(0, 𝑢𝑝 ) < ∞.
Í

f ) For 𝑢 ∈ V0 , show that M𝑢 is finitely generated if and only if 𝑢𝑝 < ∞, that is if

Í
𝑝
and only if:
(i) For every 𝑝, 𝑢𝑝 ≠ −∞;
(ii) The set of all primes 𝑝 such that 𝑢𝑝 < 0 is finite.
If this holds, show that M𝑢 is generated by one element.
g) For 𝑢 ∈ V0 , show that M𝑢 contains Z if and only if 𝑢 6 0, namely 𝑢𝑝 6 0 for
every 𝑝. If this holds, show that M𝑢 /Z is artinian if and only if the set of all primes 𝑝
such that 𝑢𝑝 < 0 is finite.
h) Let 𝑢 ∈ V0 be such that 𝑢 6 0. Show that M𝑢 /Z has finite length if and only if
(i) For every 𝑝, 𝑢𝑝 ≠ −∞;
(ii) The set of all primes 𝑝 such that 𝑢𝑝 < 0 is finite.
More precisely, show that ℓZ (M𝑢 /Z) = 𝑝 𝑢𝑝 .
Í
318 CHAPTER 6. FURTHER RESULTS ON MODULES

26-exo1154) Let A be a ring, let M be an artinian A-module and let 𝑢 be an endomor-

phism of M.
a) Assume that 𝑢 is injective. Show that 𝑢 is bijective.
b) Give an example where 𝑢 is surjective but not bijective.
27-exo1162) Let A be a ring, let M be an artinian A-module and let 𝑢 be an endomor-
phism of M.
a) Show that there exists an integer 𝑛 > 1 such that Ker(𝜑 𝑛 ) + Im(𝜑 𝑛 ) = M.
b) Assume moreover that M has finite length. Using exercise 6/17, show that the
preceding sum is a direct sum.
28-exo1213) Let A be a commutative ring.
a) Let V be an A-module of finite length. Show that the canonical morphism
V → M∈Max(A) VM (sending 𝑥 ∈ M to the family of fractions 𝑥/1, Max(A) being the
Î
set of all maximal ideals of A) is an isomorphism of A-modules.
b) Assume that A is an artinian ring. Show that the canonical morphism A →
Î
M∈Max(A) AM is an isomorphism of rings: Any commutative artinian ring is a product of
local rings.
29-exo868) Let A be a ring and let M be a right A-module. Let A0 = EndA (M) be
the ring of endomorphisms of M. Then M is seen as a left A0-module and we set
A00 = EndA0 (M).
a) For 𝑎 ∈ A, let 𝜌M (𝑎) be the homothety of ratio 𝑎 in M. Prove that 𝜌M (𝑎) ∈ A00.
Prove that the map 𝜌M : A → A00 is a morphism of rings.
One says that the A-module M is balanced if the morphism 𝜌M is an isomorphism.
b) Let 𝑛 > 1 be an integer such that M𝑛 is balanced. Prove that M is balanced.
(Represent endomorphisms of M𝑛 by matrices with entries in EndA (M).)
c) Let M be a right A-module. For every 𝑒 ∈ A ⊕ M, prove that there exists
𝑢 ∈ EndA (A𝑑 ⊕ M) such that 𝑢(1, 0) = 𝑒. Prove that A𝑑 ⊕ M is balanced.
d) Let M be a right A-module such that A is a quotient of M𝑛 , for some 𝑛 > 1 (one
says that M is generator). Prove that M is balanced.
e) Assume that the ring A is simple. Prove that every nonzero right ideal of A is
generator, hence balanced (Rieffel (1965)).
30-exo869) Let A be a simple artinian ring. Following Henderson (1965) and Nicholson
(1993), we prove in this exercise a theorem of Wedderburn according to which A is
isomorphic to a matrix ring over a division ring.
a) Prove that A contains a right ideal I which is a simple right A-module.
b) Prove that I = A · I and deduce from this that I2 ≠ 0.
c) Prove that there exists an idempotent element 𝑒 ∈ A such that I = 𝑒A and such
that D = 𝑒A𝑒 is a division ring.
d) For 𝑎 ∈ A, the homothety 𝜌I (𝑎) of ratio 𝑎 in I is an element of EndD (I); prove that
the map 𝜌I : A → EndD (I) is an isomorphism. (Use that 1 ∈ A𝑒A.)
 EXERCISES 319

e) Prove that I is finitely dimensional as a left D-vector space. (Observe that the
endomorphisms of I of finite rank form a non-trivial two-sided ideal.) Conclude that there
exists an integer 𝑛 > 1 such that A ' M𝑛 (D).
31-exo842) Let A be a commutative ring and let M be a finitely generated A-module.
Let I be an ideal of A. Prove that SuppA (M/IM) = V(I) ∩ V(AnnA (M)).
32-exo844) Let A be a commutative ring and let U ∈ M𝑚,𝑛 (A). Let 𝑢 : A𝑛 → A𝑚 be the
associated morphism of A-modules and let M = Coker(𝑢).
a) Let P be a prime ideal of A. Prove that P ∈ SuppA (M) if and only if det(V) ∈ P for
every 𝑚 × 𝑚 submatrix V of A.
b) Prove that Fit0 (M) is finitely generated and V(Fit0 (M)) = SuppA (M).
c) Prove that Spec(A) SuppA (M) is a quasi-compact subset of Spec(A).
33-exo1061) a) Let A be a ring. Let M and N be A-modules such that M ⊂ N. Show
that AssA (M) ⊂ AssA (N).
b) Give examples that show that, in general, there is no relation between the associated
prime ideals of a module and those of a quotient.
c) Let M be an A-module, let M1 and M2 be submodules of M such that M = M1 ⊕ M2 .
Prove that AssA (M) = AssA (M1 ) ∪ AssA (M2 ).
d) One only assumes that M = M1 + M2 . How can you relate AssA (M) with AssA (M1 )
and AssA (M2 )?
34-exo1065) Let A be a commutative ring, let 𝑥 ∈ A. Assume that 𝑥 is regular but not
a unit. Show that for any integer 𝑛 > 1, A/𝑥A and A/𝑥 𝑛 A have the same associated
prime ideals.
35-exo1304) Let A be a commutative ring and let I be an ideal of A such that I ≠ A.
Show that I is primary if and only if every element of A/I is either nilpotent or regular.
36-exo1301) Let A be a commutative ring, let 𝑓 : M0 → M be a morphism of A-modules
and let N be a submodule of M0. If N is primary, show that 𝑓 −1 (N) is a primary
submodule of M. Give two proofs, one relying on the manipulation of associated
prime ideals, the other on lemma 6.6.2.
37-exo1302) Let A be a commutative ring.
a) Let M be a maximal ideal of A. For any integer 𝑛 > 1, show that M𝑛 is an
M-primary ideal of A.
b) Let P be a prime ideal of A, let 𝑛 > 1 be an integer and let P(𝑛) be the inverse image
in A of the ideal P𝑛 AP of AP . Show that P(𝑛) is a P-primary ideal of A.
c) Prove that P𝑛 is P-primary if and only if P𝑛 = P(𝑛) .
d) Give an example of a prime ideal P such that P𝑛 is not a P-primary ideal.
38-exo302) Let 𝑘 be a field, let A = 𝑘[X1 , X2 , . . . ] be the ring of polynomial in infinitely
many indeterminates X1 , . . . . Let M = (X1 , . . . ) be the ideal generated by the indeter-
minates, I = (X21 , . . . ). be the ideal generated by their squares, J = (X1 , X22 , . . . ) be the
ideal generated by the elements X𝑛𝑛 .
320 CHAPTER 6. FURTHER RESULTS ON MODULES

a) Show that the nilradical of I and J is equal to M and that M is a maximal ideal. In
particular, I and J are primary ideals.
b) Show that there is no integer 𝑛 such that M𝑛 ⊂ J.
c) Show that there is no element 𝑎 ∈ A such that M = {𝑥 ∈ A ; 𝑎𝑥 ∈ I}.
d) Prove that AssA (A/I) = {M} but that there does not exist any 𝑓 ∈ A/I such that
M = AnnA ( 𝑓 ).
39-exo301) Let A be a commutative ring, let M be an A-module.
Î
Show that the canonical morphism from M to P∈AssA (M) MP is injective.
40-exo258) Let A be a commutative ring, let M be an A-module and let S be a
multiplicative subset of A. Let N be the kernel of the canonical morphism from M
to S−1 M.
a) Show that a prime ideal P of A belongs to AssA (N) if and only if it belongs
to AssA (M) and P ∩ S ≠ ∅.
b) Show that a prime ideal P of A belongs to AssA (M/N) if and only if it belongs
to AssA (M) and P ∩ S = ∅.
41-exo822) Let A be a commutative noetherian ring and let M, N be finitely generated
A-modules.
a) Assume that A is local, with maximal ideal P. Prove that M ≠ 0 if and only if there
exists a non-zero morphism 𝑓 : M → A/P.
b) We still assume that A is local, with maximal ideal P. Prove that
P ∈ AssA (Hom(M, N)) if and only if M ≠ 0 and P ∈ AssA (N).
c) Prove that AssA (Hom(M, N)) = Supp(M) ∩ AssA (N).
42-exo885) Let 𝑓 : A → B be a morphism of commutative rings. Let N be a B-module
and let M be an A-submodule of N.
a) Let P ∈ AssA (M); prove that there exists Q ∈ AssB (N) such that P = 𝑓 −1 (Q).
b) Let Q ∈ AssB (N), let N0 be a Q-primary submodule of N and let M0 = N0 ∩ M.
Assume that M0 ≠ M. Prove that M0 is a P-primary submodule of M, where P = 𝑓 −1 (Q).
43-exo647) Let A be a commutative von Neumann ring.
a) Prove that every prime ideal of A is associated to A.
b) Let P be a prime ideal of A which is of the form AnnA (𝑎), for some 𝑎 ∈ A.
Prove that P is finitely generated. (If E is a field, the von Neumann ring EN shows that
proposition 6.5.12 does not hold without the hypothesis that A be noetherian.)
44-exo650) Let A be a commutative ring, let S be a multiplicative subset of A and let M
be an A-module. Let N be the kernel of the canonical morphism 𝑖 : M → S−1 M.
a) Prove that AssA (M/N) is the set of primes P ∈ AssA (M) which are disjoint from S.
b) Prove that AssA (N) is the set of primes P ∈ AssA (M) which meet S.
45-exo651) Let A be a commutative ring.
 EXERCISES 321

a) Let M be an A-module, let P ∈ AssA (M) be an associated ideal of M and let 𝑚 ∈ M

be such that P is minimal among all prime ideals containing AnnA (𝑚). Prove that for
every 𝑎 ∈ P, there exists 𝑏 ∈ A P and 𝑛 ∈ N such that 𝑎 𝑛 𝑏𝑚 = 0. (First treat the case
where A is local with maximal ideal P.)
b) If A is reduced, then AssA (A) is the set of minimal prime ideals of A. (First treat
the case where A is local and and P ∈ AssA (A) is its maximal ideal.)
c) Assume that AP is reduced for every prime ideal P of A. prove that A is reduced.
d) Assume that for every P ∈ AssA (A), the local ring AP is a field. Prove that for every
prime ideal P of A, the local ring AP is reduced. (Considering a minimal counterexample,
prove that there exists 𝑎 ∈ P which is not a zero divisor and introduce the ring AP [1/𝑎].)
Conclude that A is reduced.
46-exo055) Let A be the ring of continuous functions on [−1; 1].
a) Is the ring A an integral domain? Is it reduced?
b) Let I be the ideal of A consisting of functions 𝑓 such that 𝑓 (0) = 0. Show that I is
not finitely generated. Show that I = I2 .
c) Show that the ideal (𝑥) is not primary. (In fact, its radical is not prime.)
𝑛)
Ñ
d) Prove that 𝑛 (𝑥 ≠ 0.
47-exo890) Let A = 𝒞([0; 1], R) be the ring of real valued, continuous functions on [0; 1].
The support S( 𝑓 ) of a function 𝑓 ∈ A is the closure of {𝑥 ; 𝑓 (𝑥) ≠ 0}.
a) Let 𝑓 ∈ A; what is AnnA ( 𝑓 )? Prove in particular it is a radical ideal and that
A/AnnA ( 𝑓 ) ' 𝒞(S( 𝑓 ); R).
b) Prove that AnnA ( 𝑓 ) is not a prime ideal.
c) Prove that the zero ideal (0) has no primary decomposition in A.
48-exo058) Let 𝑘 be a field, let A be the ring A = 𝑘[X, Y, Z]/(XY − Z2 ). One writes
𝑥 = cl(X), etc. Let P be the ideal (𝑥, 𝑧) ⊂ A.
a) Show that P is a prime ideal.
b) Show that P2 is not a primary ideal. (Consider the multiplication by 𝑦 in A/P2 .)
c) Show that (𝑥) ∩ (𝑥 2 , 𝑦, 𝑧) is a minimal primary decomposition of P2 .
49-exo825) Let E be a field, let I be the ideal (X) ∩ (X, Y)2 of the polynomial ring E[X, Y].
a) Prove that I = (X2 , XY).
Let A = E[X, Y]/I be the quotient ring and let 𝑥, 𝑦 be the classes of X, Y modulo I.
Let 𝑛 be an integer such that 𝑛 > 1.
b) Prove that the ideal (𝑦 𝑛 ) of A is (𝑥, 𝑦)-primary and that 0 = (𝑥) ∩ (𝑦 𝑛 ) is a minimal
primary decomposition of the ideal 0 in A.
c) Let 𝑎 ∈ E× . Prove that the ideal (𝑥 + 𝑎𝑦 𝑛 ) is (𝑥, 𝑦)-primary and that 0 = (𝑥 + 𝑎𝑦 𝑛 )
is a (𝑥, 𝑦)-primary ideal and that 0 = (𝑥) ∩ (𝑥 + 𝑎𝑦 𝑛 ) is also a minimal primary
decomposition of 0.
50-exo887) Let A be a commutative noetherian ring.
322 CHAPTER 6. FURTHER RESULTS ON MODULES

a) Let I be an ideal of A and let 𝑎 ∈ I be a regular element Let P be a prime ideal of A

such that P ∈ AssA (A/I). Prove that there exist a P-primary ideal Q and an ideal J of A
such that I = Q ∩ J and P ∉ AssA (A/J).
b) Prove that J contains a regular element 𝑥.
c) Prove that I : (𝑥) = Q : (𝑥) and that this ideal is P-primary.
d) Prove that there exists an integer 𝑛 > 0 such that P𝑛 ⊄ I : (𝑥) and P𝑛+1 ⊂ I : (𝑥).
Prove that P𝑛 contains a regular element 𝑦 such that 𝑦 ∉ I : (𝑥).
e) Prove that P = I : (𝑥𝑦).
f ) Let 𝑎 ∈ A be a regular element, let P be a prime ideal of A such that P ∈ AssA (A/(𝑎)).
Let 𝑏 ∈ A be a regular element such that 𝑏 ∈ P; prove that P ∈ AssA (A/(𝑏)). (Find
regular elements 𝑥, 𝑦 ∈ A such that 𝑏𝑥 = 𝑎𝑦 and P = (𝑎) : (𝑥); then prove that P = (𝑏) : (𝑦).)
51-exo889) Let K be a field and let 𝑛 ∈ N. Let A = K[T1 , . . . , T𝑛 ].
a) Let 𝑎 1 , . . . , 𝑎 𝑛 be integers. Prove that the ideal (T1𝑎1 , . . . , T𝑛𝑎 𝑛 ) of A is primary and
compute its radical.
b) Let J be a monomial ideal of A and let 𝑓 , 𝑔 be two monomials. If 𝑓 and 𝑔 are
coprime, then J + ( 𝑓 𝑔) = (J + ( 𝑓 )) ∩ (J + (𝑔)).
c) Compute a minimal primary decomposition of the ideal (X2 YZ, Y2 Z, YZ3 ) of
K[X, Y, Z].
d) Let J be a monomial ideal of A. Show that it admits a minimal primary decompo-
sition which consists in monomial ideals.
CHAPTER 7

FIRST STEPS IN HOMOLOGICAL

ALGEBRA

In this chapter, we explain a few elementary notions in homological algebra.

At the heart of algebraic topology, where it has striking applications (such
as Brouwer’s theorem, for example), the algebraic formalism of homological
algebra is now spread in many fields of mathematics — algebraic topology,
commutative algebra, algebraic geometry, representation theory for example —
and its applications range as far as robotics!
Algebraic topology and differential calculus naturally furnish sequences
M0 , M1 , . . . of modules linked by morphisms 𝑓0 : M0 → M1 , 𝑓1 : M1 → M2 ,
etc., such that the composites 𝑓1 ◦ 𝑓0 , 𝑓2 ◦ 𝑓1 , etc. of two successive morphisms
vanish. One important example of such a situation, the De Rham complex of
an open subset U of R2 , is given in the last section of this chapter, where elements
of M0 , M1 , M2 respectively are functions, vector fields, and functions on U, and
𝑓0 and 𝑓1 correspond to gradient and curl. The relation 𝑓𝑝 ◦ 𝑓𝑝−1 = 0 means
that Im( 𝑓𝑝−1 ) ⊂ Ker( 𝑓𝑝 ), and a good reason for the inclusion is the equality,
in which case one says that the sequence is exact. The game of homological
algebra starts by systematically associating with such sequences its homology
modules, defined has Ker( 𝑓𝑝 )/Im( 𝑓𝑝−1 ) which quantify in what respect that
inclusion is not an equality.
Two classes of modules have particularly good properties, that of projective
and injective modules. I define them in sections 7.3 and 7.4. While their role
in more advanced homological algebra is very important, it is barely touched
here, and I only prove the theorem of Kaplansky that projective modules over a
local ring are free, and the theorem of Baer that every module is a submodule of
an injective module.
Although one may start with exact sequences, algebraic constructions tend
to lose this property and I initiate the study of the exactness of some functors,
324 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

such as the functor of homomorphisms. This is the occasion of introducing

adjoint functors.

7.1. Diagrams, complexes and exact sequences

The notion of an exact sequence allows to combine many algebraic
properties in a diagram which is quite simple to write down as well as
to read.

7.1.1. — We have already seen a few diagrams in this book, such as the
one on p. 34 that immediately follows and illustrates theorem 1.5.3.
Such a diagram starts from a quiver, that is a collection of vertices and
arrows linking one vertex to another; moreover, the vertices are labeled
by an object from a category, and an arrow linking one vertex to another
is labeled by a morphism from the object that labels its source to the
object that labels its target.
Some arrows might be drawn using a dashed line; this usually reflects
that the existence or the uniqueness of the corresponding morphisms is
the conclusion of a mathematical statement that the diagram illustrates.
A path in a diagram is a finite sequence of consecutive arrows, meaning
that the target of an arrow is the source of the next one. Given such a
path, one may compose the morphisms by which the arrows are labeled.
A diagram is said to be commutative if for any two paths linking one
vertex to another, the two morphisms that they define coincide.
Definition (7.1.2). — Let A be a ring. A complex of A-modules is a diagram
𝑓1 𝑓2 𝑓𝑛−1
M1 −
→ M2 −
→ · · · → M𝑛−1 −−−→ M𝑛
where M1 , . . . , M𝑛 are A-modules and 𝑓1 , . . . , 𝑓𝑛−1 are homomorphisms such
that 𝑓𝑖 ◦ 𝑓𝑖−1 = 0 for 𝑖 ∈ {2, . . . , 𝑛 − 1}.
One says that such a diagram is an exact sequence if Ker( 𝑓𝑖 ) = Im( 𝑓𝑖−1 )
for every 𝑖 ∈ {2, . . . , 𝑛 − 1}.
Observe that the condition 𝑓𝑖 ◦ 𝑓𝑖−1 = 0 means that Im( 𝑓𝑖−1 ) ⊂ Ker( 𝑓𝑖 ).
Consequently, an exact sequence is a complex.
Sometimes, complexes are implicitly extended indefinitely by adding
null modules and null morphisms.
 7.1. DIAGRAMS, COMPLEXES AND EXACT SEQUENCES 325

Definition (7.1.3). — Let

𝑓1 𝑓𝑛−1
M1 −
→ M2 → · · · → M𝑛−1 −−−→ M𝑛
be a complex of A-modules. Its homology modules are the modules H2 , . . . , H𝑛−1
defined by H𝑖 = Ker( 𝑓𝑖 )/Im( 𝑓𝑖−1 ) for 2 6 𝑖 6 𝑛 − 1.
As a consequence of this definition, a complex is an exact sequence if
and only if its homology modules are zero.
If the complex is extended by null modules, we get also H0 = Ker( 𝑓1 )
and H𝑛 = M𝑛 /Im( 𝑓𝑛−1 ).
Example (7.1.4). — Let A be a ring and let 𝑓 : M → N be a morphism of
A-modules.
𝑓
The diagram 0 → M → − N → 0 is a complex, and its homology
modules are Ker( 𝑓 ) and N/Im( 𝑓 ) = Coker( 𝑓 ) (see example 3.4.3). In
particular, this diagram is an exact sequence if and only if 𝑓 is an
isomorphism.
In general, note that Ker( 𝑓 ) and Coker( 𝑓 ) sit in an exact sequence
𝑓
0 → Ker( 𝑓 ) → M →
− N → Coker( 𝑓 ) → 0.
Proposition (7.1.5). — A diagram of A-modules
𝑖 𝑝
0 → N→
− M→
− P→0
is an exact sequence if and only if
– The morphism 𝑖 is injective;
– One has Ker(𝑝) = Im(𝑖) ;
– The morphism 𝑝 is surjective.
Then, 𝑖 induces an isomorphism from N to the submodule 𝑖(N) of M, and 𝑝
induces an isomorphism of M/𝑖(N) with P.
Such diagrams are very important in practice, and called short exact
sequences.
Proof. — It suffices to write down all the conditions of an exact sequence.
The image of the map 0 → N is 0; it has to be the kernel of 𝑖, which means
that 𝑖 is injective. The next condition is Im(𝑖) = Ker(𝑝). Finally, the image
of 𝑝 is equal to the kernel of the morphism P → 0, which means that 𝑝 is
326 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

surjective. The rest of the proof follows from the factorization theorem:
if 𝑝 is surjective, it induces an isomorphism from M/Ker(𝑝) to P; if 𝑖 is
injective, it induces an isomorphism from N to 𝑖(N) = Ker(𝑝). 

Lemma (7.1.6). — Let A be a ring and let

𝑖 𝑝
0 → N→
− M→
− P→0
be a short exact sequence of A-modules.
a) The map 𝑞 ↦→ Im(𝑞) is a bijection from the set of right inverses of 𝑝
(morphisms 𝑞 : P → M such that 𝑝 ◦ 𝑞 = idP ) to the set of direct summands
of 𝑖(N) in M.
b) The map 𝑗 : Ker(𝑗) is a bijection from the set of left inverses of 𝑖 (morphisms
𝑗 : M → N such that 𝑗 ◦ 𝑖 = idN ) to the set of direct summands of 𝑖(N) in M.
c) The morphism 𝑝 has a right inverse if and only if the morphism 𝑖 has
a left inverse, if and only if the submodule 𝑖(N) = Ker(𝑝) of M has a direct
summand.

Definition (7.1.7). — An exact sequence which satisfies the three equivalent

properties of lemma 7.1.6, c), is said to be split

Proof. — a) Let 𝑞 : P → M be a right inverse of 𝑝, let Q = 𝑞(P) be

the image of 𝑞 and let us show that Q is a direct summand of 𝑖(N)
in M. It suffices to prove that for any 𝑚 ∈ M, there is a unique pair
(𝑥, 𝑦) such that 𝑥 ∈ N, 𝑦 ∈ P and 𝑚 = 𝑖(𝑥) + 𝑞(𝑦). If 𝑚 = 𝑖(𝑥) + 𝑞(𝑦),
then 𝑝(𝑚) = 𝑝(𝑖(𝑥)) + 𝑝(𝑞(𝑦)) = 𝑦 since 𝑝 ◦ 𝑖 = 0 and 𝑝 ◦ 𝑞 = idP . Then,
𝑖(𝑥) = 𝑚−𝑞(𝑝(𝑚)). Observe that 𝑝(𝑚−𝑞(𝑝(𝑚))) = 𝑝(𝑚)−(𝑝◦𝑞)(𝑝(𝑚)) = 0;
by the definition of an exact sequence, Im(𝑖) = Ker(𝑝), so that 𝑚 −
𝑞(𝑝(𝑚)) ∈ Im(𝑖). Since 𝑖 is injective, there is a unique 𝑥 ∈ N such that
𝑖(𝑥) = 𝑚 − 𝑞(𝑝(𝑚)). Then, 𝑦 = 𝑝(𝑚) satisfies 𝑚 − 𝑖(𝑥) = 𝑞(𝑝(𝑚)) = 𝑞(𝑦).
Moreover, if 𝑚 = 𝑖(𝑥) + 𝑞(𝑦) = 𝑖(𝑥 0) + 𝑞(𝑦 0), then 𝑖(𝑥 − 𝑥 0) = 𝑞(𝑦 0 − 𝑦); then
0 = 𝑝 ◦ 𝑖(𝑥 − 𝑥 0) = 𝑝 ◦ 𝑞(𝑦 0 − 𝑦) = 𝑦 0 − 𝑦, hence 𝑦 0 = 𝑦; then 𝑖(𝑥) = 𝑖(𝑥 0),
hence 𝑥 = 𝑥 0 since 𝑖 is injective.
Conversely, let Q ⊂ M be a direct summand of 𝑖(N) and let us prove
that the morphism 𝑝| Q : Q → P is an isomorphism. Its kernel is
Q ∩ Ker(𝑝) = Q ∩ 𝑖(N) = 0, so that it is injective. On the other hand, let
𝑥 ∈ P; since 𝑝 is surjective, there exists 𝑚 ∈ M such that 𝑝(𝑚) = 𝑥; let us
 7.1. DIAGRAMS, COMPLEXES AND EXACT SEQUENCES 327

write 𝑚 = 𝑦 + 𝑛, with 𝑛 ∈ 𝑖(N) and 𝑦 ∈ Q; one has 𝑥 = 𝑝(𝑚) = 𝑝(𝑦), so

that 𝑝| Q is surjective. Let 𝑞 : P → Q be the inverse of this isomorphism;
one has 𝑞 ◦ 𝑝| Q = idQ and 𝑝 ◦ 𝑞 = idP ; in particular, 𝑞 is a right inverse
of 𝑝.
These constructions 𝑞 ↦→ Q and Q ↦→ 𝑞 are inverse one to another; this
proves a).
b) Let Q be the kernel of 𝑗. Let us show that Q is a direct summand
of 𝑖(N). If 𝑥 ∈ Q ∩ 𝑖(N), one may write 𝑥 = 𝑖(𝑦) for some 𝑦 ∈ N, so
that 0 = 𝑗(𝑥) = 𝑗(𝑖(𝑦)) = 𝑦. Consequently, 𝑦 = 0 and 𝑥 = 0. Moreover,
let 𝑥 ∈ M and set 𝑦 = 𝑥 − 𝑖(𝑗(𝑥)). Then, 𝑗(𝑦) = 𝑗(𝑥) − 𝑗(𝑖(𝑗(𝑥)) = 0 so
that 𝑦 ∈ Q. Therefore, 𝑥 = 𝑖(𝑗(𝑥)) + 𝑦 belongs to 𝑖(N) + Q. Hence
M = Q ⊕ 𝑖(N).
Conversely, let Q be a direct summand of 𝑖(N) in M. Since 𝑖 is injective,
any element of M can be written uniquely as 𝑥 + 𝑖(𝑦), for some 𝑥 ∈ Q
and some 𝑦 ∈ N. Let 𝑗 : M → N be the map such that 𝑗(𝑚) = 𝑦 if
𝑚 = 𝑥 + 𝑖(𝑦) with 𝑥 ∈ Q and 𝑦 ∈ N. One checks directly that 𝑗 is a
morphism. Finally, for any 𝑦 ∈ N, the decomposition 𝑚 = 𝑖(𝑦) = 0 + 𝑖(𝑦)
shows that 𝑗(𝑚) = 𝑦, so that 𝑗 ◦ 𝑖 = idN .
Again, these two constructions 𝑗 ↦→ Q and Q ↦→ 𝑗 are inverse one to
the other, which proves b).
Finally, assertion c) follows from a) and b). 

Examples (7.1.8). — 1) Any subspace of a vector space has a direct

summand. Consequently, every exact sequence of modules over a
division ring is split.
2) Let A = Z, let M = Z, let N = 2Z and let P = M/N = Z/2Z. The
natural injection 𝑖 of N to M and the canonical surjection from M to P
give rise to an exact sequence

𝑖 𝑝
0 → 2Z →
− Z→
− Z/2Z → 0.

This exact sequence is not split. Indeed, if 𝑞 were a right inverse to 𝑝, its
image would be a submodule of Z isomorphic to Z/2Z. However, there
is no nonzero 𝑥 element of Z such that 2𝑥 = 0 while there is such an
element in Z/2Z.
328 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

7.2. The “snake lemma”. Finitely presented modules

Theorem (7.2.1) (Snake lemma). — Let A be a ring. Let us consider a diagram
of morphisms of A-modules
𝑖 𝑝
→ M → P → 0
← ← ←
N

←
←
←
𝑓 𝑔 ℎ

→
𝑖0 𝑝0
→ N0 → M0 → P0.
← ← ←
0

Let us assume the two rows are exact sequences, and that the two squares are
commutative, meaning that 𝑖 0 ◦ 𝑓 = 𝑔 ◦ 𝑖 and 𝑝 0 ◦ 𝑔 = ℎ ◦ 𝑝.
a) There is a unique morphism of A-modules 𝜕 : Ker(ℎ) → Coker( 𝑓 ) such
that 𝜕(𝑝(𝑦)) = cl(𝑥 0) for 𝑦 ∈ 𝑝 −1 (Ker(ℎ)) and 𝑥 0 ∈ N0 such that 𝑔(𝑦) = 𝑖 0(𝑥 0).
b) These morphisms induce an exact sequence
𝑖∗ 𝑝∗ 𝜕 (𝑖 0)∗ (𝑝 0)∗
Ker( 𝑓 ) →
− Ker(𝑔) − − Coker( 𝑓 ) −−→ Coker(𝑔) −−−→ Coker(ℎ);
→ Ker(ℎ) →

c) If 𝑖 is injective, then 𝑖 ∗ is injective;

d) If 𝑝 0 is surjective, then (𝑝 0)∗ is surjective.

Proof. — The proof needs a number of steps. In particular, assertion a)

is proved in step 7; assertions c) and d) are proved in steps 3 and 5.
1) We first show that 𝑖(Ker( 𝑓 )) ⊂ Ker(𝑔) and 𝑝(Ker(𝑔)) ⊂ Ker(ℎ). Let
indeed 𝑥 ∈ N be such that 𝑓 (𝑥) = 0; then 𝑔(𝑖(𝑥)) = (𝑔 ◦𝑖)(𝑥) = (𝑖 0 ◦ 𝑓 )(𝑥) =
𝑖 0( 𝑓 (𝑥)) = 0, so that 𝑖(𝑥) ∈ Ker(𝑔). Similarly, for any 𝑦 ∈ Ker(𝑔), one has
ℎ(𝑝(𝑦)) = (ℎ ◦ 𝑝)(𝑦) = (𝑝 0 ◦ 𝑔)(𝑦) = 𝑝 0(𝑔(𝑦)) = 0, hence 𝑝(𝑦) ∈ Ker(ℎ).
Consequently, the morphisms 𝑖 and 𝑝 induce by restriction morphisms
𝑖∗ : Ker( 𝑓 ) → Ker(𝑔) and 𝑝 ∗ : Ker(𝑔) → Ker(ℎ).
2) One has 𝑖 0(Im 𝑓 ) ⊂ Im 𝑔 and 𝑝 0(Im 𝑔) ⊂ Im ℎ. Indeed, let 𝑥 0 ∈ Im( 𝑓 )
and let 𝑥 ∈ N be such that 𝑥 0 = 𝑓 (𝑥). Then 𝑖 0(𝑥 0) = (𝑖 0 ◦ 𝑓 )(𝑥) = (𝑔 ◦ 𝑖)(𝑥) =
𝑔(𝑖(𝑥)), which shows that 𝑖 0(𝑥 0) belongs to Im(𝑔). Similarly, let 𝑦 0 ∈ Im(𝑔),
let 𝑦 ∈ M be such 𝑦 0 = 𝑔(𝑦). One has 𝑝 0(𝑦 0) = (𝑝 0 ◦ 𝑔)(𝑦) = (ℎ ◦ 𝑝)(𝑦) =
ℎ(𝑝(𝑦)), hence 𝑝 0(𝑦 0) belongs to Im(ℎ).
Consequently, the kernel of the composition

0 𝑖
0
− M0 → M0/Im 𝑔 = Coker(𝑔)
N →
 7.2. THE “SNAKE LEMMA”. FINITELY PRESENTED MODULES 329

contains Im( 𝑓 ). Passing to the quotient, we obtain a morphism

(𝑖 0)∗ : Coker( 𝑓 ) = N0/Im( 𝑓 ) → Coker(𝑔). In the same way, we deduce
from 𝑝 0 a morphism (𝑝 0)∗ : Coker(𝑔) → Coker(ℎ).
3) Assume that 𝑖 is injective. Then the morphism 𝑖∗ is injective: if
𝑖 ∗ (𝑥) = 0, then 𝑖(𝑥) = 0 hence 𝑥 = 0.
4) Since 𝑝 ∗ is the restriction of 𝑝 to Ker(𝑔), and since 𝑝 ◦ 𝑖 = 0, we
have 𝑝 ∗ ◦ 𝑖 ∗ = 0, hence Im 𝑖∗ ⊂ Ker 𝑝∗ . Conversely, let 𝑦 ∈ Ker 𝑝 ∗ , that is,
𝑦 ∈ Ker(𝑔) and 𝑝(𝑦) = 0. Since the first row of the diagram is an exact
sequence, 𝑦 ∈ Im(𝑖) and there exists 𝑥 ∈ N such that 𝑦 = 𝑖(𝑥). Then
0 = 𝑔(𝑦) = 𝑔(𝑖(𝑥)) = (𝑔 ◦ 𝑖)(𝑥) = (𝑖 0 ◦ 𝑓 )(𝑥) = 𝑖 0( 𝑓 (𝑥)). Since 𝑖 0 is injective,
𝑓 (𝑥) = 0 and 𝑥 ∈ Ker( 𝑓 ). This implies that 𝑦 = 𝑖(𝑥) ∈ 𝑖(Ker( 𝑓 )) =
𝑖 ∗ (Ker( 𝑓 )).
5) Assume that 𝑝 0 is surjective. Then the morphism 𝑝 ∗0 is surjective:
let 𝜁0 ∈ Coker(ℎ), and let 𝑧 0 ∈ P0 be such that 𝜁0 = cl(𝑧 0). Since 𝑝 0 is
surjective, there exists 𝑦 0 ∈ M0 such that 𝑧 0 = 𝑝 0(𝑦 0). By definition of 𝑝 ∗0 ,
𝜁0 = 𝑝 ∗0 (cl(𝑦 0)), hence 𝜁0 ∈ Im(𝑝 ∗0 ).
6) One has 𝑝∗0 ◦ 𝑖 ∗0 = 0. Indeed, the definition of 𝑖 ∗0 implies that for any
𝑥 0 ∈ N0, 𝑖 ∗0 (cl(𝑥 0)) = cl(𝑖 0(𝑥 0)), hence

𝑝∗0 (𝑖 ∗0 (cl(𝑥 0)) = 𝑝∗0 (cl(𝑖 0(𝑥 0))) = cl(𝑝 0(𝑖 0(𝑥 0))) = 0.

Conversely, if 𝑝 ∗0 (cl(𝑦 0)) = 0, then cl(𝑝 0(𝑦 0)) = 0, so that 𝑝 0(𝑦 0) ∈ Im(ℎ).
Let us thus write 𝑝 0(𝑦 0) = ℎ(𝑧) for some 𝑧 ∈ P. Since 𝑝 is surjective, there
exists 𝑦 ∈ M such 𝑧 = 𝑝(𝑦) and 𝑝 0(𝑦 0) = ℎ(𝑝(𝑧)) = 𝑝 0(𝑔(𝑧)). Consequently,
𝑦 0 − 𝑔(𝑧) belongs to Ker(𝑝 0), hence is of the form 𝑖 0(𝑥 0) for some 𝑥 0 ∈ N0.
Finally,
cl(𝑦 0) = cl(𝑔(𝑧) + 𝑖 0(𝑥 0)) = cl(𝑖 0(𝑥 0)) = 𝑖∗0 (𝑥 0),
which proves that Ker(𝑝 ∗0 ) = Im(𝑖∗0 ).
7) Let us now define the morphism 𝜕 : Ker(ℎ) → Coker( 𝑓 ). Let
𝑦 ∈ 𝑝 −1 (Ker(ℎ)); then 𝑝(𝑦) ∈ Ker(ℎ), hence 𝑝 0(𝑔(𝑦)) = ℎ(𝑝(𝑦)) = 0; one
thus has 𝑔(𝑦) ∈ Ker(𝑝 0) = Im(𝑖 0), since the bottom row of the diagram is
exact; since 𝑖 0 is injective, there exists a unique element 𝑥 0 ∈ N0 such that
𝑔(𝑦) = 𝑖 0(𝑥 0). Let then 𝜕1 (𝑦) ∈ Coker( 𝑓 ) be the class of 𝑥 0 modulo Im( 𝑓 ).
This defines a map 𝜕1 : 𝑝 −1 (Ker(ℎ)) → Coker( 𝑓 ). This map 𝜕1 is a
morphism of A-modules. This may be proved by a direct elementary
computation, but we simply observe that 𝜕1 is the composition of the
330 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

morphism 𝑔| 𝑝 −1 (Ker(ℎ)) (which lands into Ker(𝑝 0)) with the inverse of the
isomorphism N0 → Ker(𝑝 0) = Im(𝑖 0) induced by 𝑖 0, with the projection
to Coker( 𝑓 ).
The surjective morphism 𝑝 : M → P induces, by restriction, a surjective
morphism from 𝑝 −1 (Ker(ℎ)) → Ker(ℎ) whose kernel is Ker(𝑝). Let us
show that there exists a unique morphism 𝜕 : Ker(ℎ) → Coker( 𝑓 ) such
that 𝜕 ◦ 𝑝 = 𝜕1 . By the factorization theorem, it suffices to prove that
Ker(𝑝) ⊂ Ker(𝑝1 ). Let 𝑦 ∈ Ker(𝑝) ⊂ 𝑝 −1 (Ker(ℎ)) and 𝑥 0 ∈ N0 be such
that 𝑖 0(𝑥 0) = 𝑔(𝑦). Since the top row of the diagram is exact, there exists
𝑥 ∈ N such that 𝑦 = 𝑖(𝑥); then 𝑔(𝑦) = 𝑔(𝑖(𝑥)) = 𝑖 0( 𝑓 (𝑥)), so that 𝑥 0 = 𝑓 (𝑥)
and 𝜕1 (𝑦) = cl(𝑥 0) = 0. Consequently, Ker(𝜕1 ) contains Ker(𝑝), as was to
be shown.
Conversely, if 𝜕0 : Ker(ℎ) → Coker( 𝑓 ) satisfies the conditions of as-
sertion a), one has 𝜕0 ◦ 𝑝 = 𝜕1 = 𝜕 ◦ 𝑝, hence 𝜕 = 𝜕0 since 𝑝 induces a
surjective morphism from 𝑝 −1 (Ker(ℎ)) to Ker(ℎ).
8) Let us show that Im(𝑝 ∗ ) = Ker(𝜕). Let 𝑧 ∈ Im(𝑝∗ ). There exists
𝑦 ∈ Ker(𝑔) such that 𝑝(𝑦) = 𝑧. In other words, 𝑔(𝑦) = 0; with the
notation of the preceding paragraph, we may take 𝑥 0 = 0, hence 𝜕(𝑧) = 0
and 𝑧 ∈ Ker(𝜕).
Conversely, let 𝑧 ∈ Ker(𝜕). Keeping the same notation, we have
𝑥 ∈ Im( 𝑓 ), so that there is 𝑥 ∈ N such that 𝑥 0 = 𝑓 (𝑥), hence 𝑔(𝑦) =
0

𝑖 0(𝑥 0) = 𝑔(𝑖(𝑥)) and 𝑦 − 𝑖(𝑥) ∈ Ker(𝑔). It follows that 𝑧 = 𝑝(𝑦) =

𝑝(𝑦 − 𝑖(𝑥)) ∈ 𝑝(Ker(𝑔)) = Im(𝑝∗ ), as required.
9) Finally, let us show that Im(𝜕) = Ker(𝑖 ∗0 ). Let 𝑧 ∈ N; with the
same notation, 𝑖 0(𝜕(𝑧)) = 𝑖 0(cl(𝑥 0)) = cl(𝑖 0(𝑥 0)) = cl(𝑔(𝑦)) = 0, hence
𝜕(𝑧) ∈ Ker(𝑖∗0 ) and Im(𝜕) ⊂ Ker(𝑖 ∗0 ).
In the other direction, let 𝜉0 ∈ Ker(𝑖 ∗0 ). There is 𝑥 0 ∈ N0 such that
𝜉0 = cl(𝑥 0), so that 𝑖∗0 (𝜉0) = cl(𝑖 0(𝑥 0)). This shows that 𝑖 0(𝑥 0) ∈ Im(𝑔).
Let 𝑦 ∈ M be such that 𝑖 0(𝑥 0) = 𝑔(𝑦); the definition of 𝜕 shows that
𝜕(𝑝(𝑦)) = cl(𝑥 0) = 𝜉0 so that Ker(𝑖∗0 ) ⊂ Im(𝜕).
This concludes the proof of the theorem. 

Corollary (7.2.2). — Let us retain the hypotheses of theorem 7.2.1.

a) If 𝑓 and ℎ are injective, then 𝑔 is injective. If 𝑓 and ℎ are surjective, then
𝑔 is surjective.
 7.2. THE “SNAKE LEMMA”. FINITELY PRESENTED MODULES 331

b) If 𝑓 is surjective and 𝑔 is injective, then ℎ is injective.

c) If 𝑔 is surjective and ℎ is injective, then 𝑓 is surjective.
Proof. — a) Assume that 𝑓 and ℎ are injective. The exact sequence
𝑖∗ 𝑝∗
given by the snake lemma begins with 0 → − Ker(𝑔) − → 0. Necessarily,
Ker(𝑔) = 0. If 𝑓 and ℎ are surjective, the exact sequence ends with
𝑖 ∗0 𝑝∗0
0→
− Coker(𝑔) − → 0, so that Coker(𝑔) = 0 and 𝑔 is surjective.
b) If 𝑓 is surjective and 𝑔 is injective, one has Ker(𝑔) = 0 and Coker( 𝑓 ) =
𝑝∗
0. The middle of the exact sequence can thus be rewritten as 0 −
→
𝜕
− 0, so that ℎ is injective.
Ker(ℎ) →
c) Finally, if 𝑔 is surjective and ℎ is injective, we have Ker(ℎ) = 0,
𝜕 𝑖 ∗0
Coker(𝑔) = 0, hence an exact sequence 0 → − Coker( 𝑓 ) →
− 0, which
implies that Coker( 𝑓 ) = 0 and 𝑓 is surjective. 

Definition (7.2.3). — Let A be a ring. Let M be a right A-module. A

presentation of M consists in an exact sequence
𝜓 𝜑
F−
→E−
→ M → 0,
where E and F are free A-modules.
One says that this presentation is finite if moreover F and G are finitely
generated.
If an A-module M has a finite presentation, one also says that M is
finitely presented.
𝜓 𝜑
Remark (7.2.4). — Let M be an A-module and let F − → E − → M →
0 be a presentation of M. By definition of an exact sequence, the
morphism 𝜑 is surjective. Let (𝑒 𝑖 )𝑖∈I be a basis of E. Then (𝜑(𝑒 𝑖 ))𝑖∈I is a
generating family in M. Let ( 𝑓 𝑗 ) 𝑗∈J be a basis of F. The family (𝜓( 𝑓𝑖 )) 𝑗∈J
generates Im(𝜓). By the definition of an exact sequence, we see that this
family generates Ker(𝜑).
Reversing the argument, we can construct a presentation of an A-
module M as follows. First choose a generating family (𝑚 𝑖 )𝑖∈I in M. Let
then 𝜑 : A(I) → M be the A-linear map defined by 𝜑((𝑎 𝑖 )) = 𝑖∈I 𝑚 𝑖 𝑎 𝑖 .
Í
Let (𝑛 𝑗 ) 𝑗∈J be a generating family of Ker(𝜑) and let 𝜓 : A(J) → A(I) be the
332 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

𝜓 𝜑
A-linear map defined by 𝜓((𝑎 𝑗 )) = 𝑛 𝑗 𝑎 𝑗 . Then A(J) −
→ A(I) −
Í
→M→0
is a presentation of M.

Lemma (7.2.5). — Let A be a ring and let M be a right A-module. Let

𝑞 𝑝
0 → P→ − N→ − M → 0 be an exact sequence. If M is finitely presented and N
is finitely generated, then P is finitely generated.
𝜓 𝜑
Proof. — Let F − →E− → M → 0 be a finite presentation of M. Let (𝑒 𝑖 )𝑖∈I be
a finite generating family of E and let ( 𝑓 𝑗 ) 𝑗∈J be a finite generating family
of F. For every 𝑖 ∈ I, choose an element 𝑢𝑖 ∈ N such that 𝑝(𝑢𝑖 ) = 𝜑(𝑒 𝑖 );
such an element exists because 𝑝 is surjective. Let 𝑢 : E → N be the
unique morphism of A-modules such that 𝑢(𝑒 𝑖 ) = 𝑢𝑖 for every 𝑖 ∈ I. One
has 𝑝 ◦ 𝑢 = 𝜑.
Let 𝑗 ∈ J. One has 𝑝(𝑢(𝜓( 𝑓 𝑗 )) = 𝜑(𝜓( 𝑓 𝑗 )) = 0. Consequently, 𝑢(𝜓( 𝑓 𝑗 )) ∈
Ker(𝑝). By definition of an exact sequence, the morphism 𝑞 induces
an isomorphism from P to Im(𝑞). Consequently, there exists a unique
element 𝑣 𝑗 ∈ P such that 𝑢(𝜓( 𝑓 𝑗 )) = 𝑞(𝑣 𝑗 ). Let then 𝑣 : A(J) → P be
the unique morphism such that 𝑣( 𝑓 𝑗 ) = 𝑣 𝑗 for every 𝑗 ∈ J. One has
𝑢 ◦ 𝜓 = 𝑞 ◦ 𝑣.
We have constructed a commutative diagram
𝜓 𝜑
→ E → M → 0
← ← ←
F
←

𝑣 𝑢
→

idM
𝑞 𝑝
→ P → N → M → 0
← ← ← ←
0

the two rows of which are exact sequences. By the snake lemma
(theorem 7.2.1), we deduce from this diagram an exact sequence

Ker(idM ) → Coker(𝑣) → Coker(𝑢) → Coker(idM ).

Since idM is an isomorphism, one has Ker(idM ) = Coker(idM ) = 0 so

that Coker(𝑣) and Coker(𝑢) are isomorphic. Since Coker(𝑢) = N/Im(𝑢)
is a quotient of the finitely generated module N, it is finitely generated.
Consequently, Coker(𝑣) = P/Im(𝑣) is finitely generated.
On the other hand, Im(𝑣) is a quotient of the finitely generated A-
module F, hence Im(𝑣) is finitely generated.
 7.2. THE “SNAKE LEMMA”. FINITELY PRESENTED MODULES 333

By proposition 3.5.5, the module P is finitely generated, as was to be

shown. 

7.2.6. — We now let A be a commutative ring and explore the behavior

of homomorphisms with respect to localization. Let S be a multiplicative
subset of A. Let M and N be A-modules. Every morphism 𝑓 : M →
N gives rise to a morphism S−1 𝑓 : S−1 M → S−1 N of S−1 A-modules,
characterized by (S−1 𝑓 )(𝑚/𝑠) = 𝑓 (𝑚)/𝑠 for 𝑚 ∈ M and 𝑠 ∈ S. The map
𝑓 ↦→ S−1 𝑓 from HomA (M, N) to HomS−1 A (S−1 M, S−1 N) is A-linear, and
its target is an S−1 A-module; consequently, it extends to a morphism
𝛿 M,N : S−1 HomA (M, N) → HomS−1 A (S−1 M, S−1 N) of S−1 A-modules.
Proposition (7.2.7). — Let A be a commutative ring, let S be a multiplicative
subset of A, let M and N be A-modules. If M is a finitely presented A-module,
then the morphism 𝛿 M,N is an isomorphism.
Proof. — We first treat the case where M is a free finitely generated
A-module. Let (𝑒1 , . . . , 𝑒 𝑚 ) be a basis of M. Then every morphism
𝑓 : M → N is determined by the image of the 𝑒 𝑖 , and the map
HomA (M, N) → N𝑚 given by 𝑓 ↦→ ( 𝑓 (𝑒1 ), . . . , 𝑓 (𝑒 𝑚 )) is an isomorphism
of A-modules. Moreover, S−1 M is a free S−1 A–module, with basis
(𝑒1 /1, . . . , 𝑒 𝑚 /1), and the map HomS−1 A (S−1 M, S−1 N) → (S−1 N)𝑚 given
by 𝑓 ↦→ ( 𝑓 (𝑒1 /1), . . . , 𝑓 (𝑒 𝑚 /1)) is an isomorphism of S−1 A-modules.
Given these isomorphism, the morphism 𝛿 M,N identifies as the morphism
from S−1 (N𝑚 ) to S−1 N)𝑚 given by (𝑛1 , . . . , 𝑛 𝑚 )/𝑠 ↦→ (𝑛1 /𝑠, . . . , 𝑛 𝑚 /𝑠).
Since this morphism is an isomorphism, this implies that 𝛿 M,N is an
isomorphism if M is a free finitely generated A-module.
𝑢 𝑝
Let now F → − E→ − M → 0 be a finite presentation of M. By left
exactness of the functor HomA (•, N), it induces an exact sequence
𝑝∗ 𝑢∗
0 → HomA (M, N) −
→ HomA (E, N) −→ HomA (F, N)
of A-modules. By exactness of localization, it then induces an exact
sequence
S−1 𝑝 ∗ S−1 𝑢 ∗
0 → S−1 HomA (M, N) −−−→ S−1 HomA (E, N) −−−−→ S−1 HomA (F, N)
of S−1 A-modules.
334 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

S−1 𝑢
The given presentation of M also induces a finite presentation S−1 F −−−→
S−1 𝑝 S −1
S−1 E −−−→→
− M → 0 and, similarly, an exact sequence
𝑝∗ 𝑢∗
0 → HomS−1 A (S−1 M, S−1 N) −→ HomS−1 A (S−1 E, S−1 N) −→ HomS−1 A (S−1 F, S−1 N)

of S−1 A-modules.
Then the morphisms 𝛿 M,N , 𝛿 E,N and 𝛿 F,N sit within a commutative
diagram
S−1 𝑝 ∗ S−1 𝑢 ∗
→ S−1 HomA (M, N) → S−1 HomA (E, N) → S−1 HomA (F, N)
← ← ←
0
←

←
𝛿M,N 𝛿E,N 𝛿F,N
→

→
(S−1 𝑝)∗ (S−1 𝑢)∗
0 → HomS−1 A (S−1 M, S−1 N) → HomS−1 A (S−1 E, S−1 N) → HomS−1 A (S−1 F, S−1 N)
← ← ←

the two rows of which are exact sequences. We also know that 𝛿E,N and
𝛿 F,N are isomorphisms, because E and F are free and finitely generated.
To deduce from this that 𝛿M,N is an isomorphism, we will apply
the snake lemma. However, the morphism S−1 𝑢 ∗ is not surjective a
priori, hence me replace the module S−1 HomA (F, N) by the image of
S−1 𝑢 ∗ and the morphism 𝛿 F,N by its restriction 𝛿0F,N . The snake lemma
(theorem 7.2.1) then furnishes a morphism 𝜕 sitting in an exact sequence

0 → Ker(𝛿 M,N ) → Ker(𝛿 E,N ) → Ker(𝛿0F,N )

𝜕
− Coker(𝛿 M,N ) → Coker(𝛿 E,N ) → Coker(𝛿0F,N ).
→
The morphism 𝛿E,N is an isomorphism, hence Ker(𝛿 E,N ) = Coker(𝛿 E,N ) =
0; the morphism 𝛿0F,N is injective, hence Ker(𝛿0F,N ) = 0. The preceding
exact sequence thus becomes
𝜕
0 → Ker(𝛿 M,N ) → 0 → 0 →
− Coker(𝛿 M,N ) → 0
so that Ker(𝛿 M,N ) = Coker(𝛿 M,N ) = 0. This proves that 𝛿 M,N is an
isomorphism, as was to be shown. 

7.3. Projective modules

Definition (7.3.1). — One says that an A-module P is projective if every short
exact sequence
0→N→M→P→0
 7.3. PROJECTIVE MODULES 335

is split.

Proposition (7.3.2). — Let A be a ring and let P be an A-module. The following

properties are equivalent:
(i) The module P is projective;
(ii) The module P is isomorphic to a direct summand of a free A-module;
(iii) For every A-modules M and N, any surjective morphism 𝑝 : M → N
and any morphism 𝑓 : P → N, there exists a morphism 𝑔 : P → M such that
𝑓 = 𝑝 ◦ 𝑔;
(iv) For every A-module M, every surjective morphism 𝑓 : M → P has a
right inverse.

The third property of the theorem is often taken as the definition of a

projective module. It can be summed up as a diagram

M
𝑔
←

→ 𝑝


←
𝑓
→ N
←
P

where the solid arrows represent the given maps, and the dashed arrow
is the one whose existence is asserted by the property.
Proof. — (i)⇒(ii). Let us assume that P is projective. We need to
construct a free A-module L, submodules Q and Q0 of L such that
L = Q ⊕ Q0 and P ' Q. Let S be a generating family of P and let
L = A(S) be the free A-module on S; let (𝑒 𝑠 )𝑠∈S be the canonical basis
of L (see example 3.5.3). Let 𝑝 : L → P be the unique morphism such
that 𝑝(𝑒 𝑠 ) = 𝑠 for every 𝑠 ∈ S; it is surjective (proposition 3.5.2). Let
𝑝
N = Ker(𝑝); we thus have a short exact sequence 0 → N → L → − P → 0.
Since P is projective, this short exact sequence is split and N has a
direct summand Q in L, namely the image of any right inverse of 𝑝
(lemma 7.1.6). In particular, P ' Q and L = Q ⊕ N. This shows that P is
(isomorphic) to a direct summand of free module.
(ii)⇒(iii). Let us assume that there exists an A-module Q such that
L = P ⊕ Q is a free A-module. Let S be a basis of L.
336 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

Let M and N be A-modules, let 𝑝 : M → N and 𝑓 : P → N be

morphisms of A-modules, 𝑝 being surjective. We need to show that
there exists a morphism 𝑔 : P → M such that 𝑝 ◦ 𝑔 = 𝑓 .
Let us define a morphism 𝜑 : L → N by 𝜑(𝑥 + 𝑦) = 𝑓 (𝑥) for 𝑥 ∈ P
and 𝑦 ∈ Q. By construction, 𝜑| P = 𝑓 . Since 𝑝 is surjective, there exists
for every 𝑠 ∈ S an element 𝑚 𝑠 ∈ M such that 𝑝(𝑚 𝑠 ) = 𝜑(𝑠). Then the
universal property of free A-modules implies that there exists a unique
morphism 𝛾 : L → M such that 𝛾(𝑠) = 𝑚 𝑠 for every 𝑠 ∈ S, in particular,
𝑝 ◦ 𝛾(𝑠) = 𝑝(𝑚 𝑠 ) = 𝜑(𝑠). Since S is a basis of L, 𝑝 ◦ 𝛾 = 𝜑. The restriction
𝑔 = 𝜑| P of 𝜑 to P satisfies 𝑝 ◦ 𝑔 = 𝜑| P = 𝑓 .
(iii)⇒(iv). Let 𝑝 : M → P be a surjective morphism of A-modules.
Let us apply the hypothesis of (iii) to N = P and 𝑓 = idP . There exists
𝑔 : P → M such that 𝑝 ◦ 𝑔 = idP ; in other words, 𝑝 has a right inverse.
𝑖 𝑝
(iv)⇒(i). Let 0 → N → − M→ − P → 0 be a short exact sequence of
A-modules. By assumption, 𝑝 has a right inverse. It then follows from
lemma 7.1.6 that this exact sequence is split. 

Corollary (7.3.3). — Every free A-module is projective.

Remark (7.3.4). — Let P be a projective A-module. Let (𝑒 𝑖 )𝑖∈I be a

generating family of elements of P and let 𝑓 : A(I) → P be the surjective
morphism given by 𝑓 ((𝑎 𝑖 )) = 𝑎 𝑖 𝑒 𝑖 . Let 𝑔 : P → A(I) be a right inverse
Í
of 𝑓 . For every 𝑖, let 𝜑 𝑖 : P → A be the morphism such that 𝜑 𝑖 (𝑥) is
the 𝑖th coordinate of 𝑔(𝑥), for every 𝑥 ∈ P. Then (𝜑 𝑖 )𝑖∈I is a family of
elements of P∨ , for every 𝑥 ∈ I, the family (𝜑 𝑖 (𝑥)) has finite support and
one has 𝑥 = 𝑓 (𝑔(𝑥)) = 𝑓 ((𝜑 𝑖 (𝑥))𝑖 ) = 𝑖∈I 𝜑 𝑖 (𝑥)𝑒 𝑖 .
Í
In this case, the family (𝜑 𝑖 ) plays more or less the role of a “dual basis”
of the generating family (𝑒 𝑖 ).
Conversely, let P be an A-module and let us assume that there exist a
family (𝑒 𝑖 )𝑖∈I in P and a family (𝜑 𝑖 )𝑖∈I in P∨ such that, for every 𝑥 ∈ I, the
family (𝜑 𝑖 (𝑥)) has finite support and 𝑥 = 𝑖∈I 𝜑 𝑖 (𝑥)𝑒 𝑖 . Let us prove that P
Í
is projective. As above, we consider the morphism 𝑓 : A(I) → P such that
𝑓 ((𝑎 𝑖 )) = 𝑎 𝑖 𝑒 𝑖 and the map 𝑔 : P → A(I) given by 𝑔(𝑥) = (𝜑 𝑖 (𝑥))𝑖 . By
Í
assumption, one has 𝑓 ◦ 𝑔(𝑥) = 𝑥 for every 𝑥 ∈ P, so that 𝑓 is surjective
𝑓
and has a right inverse. This implies that 0 → Ker( 𝑓 ) → A(I) →
− P→0
 7.3. PROJECTIVE MODULES 337

is a split exact sequence and P is isomorphic to a direct summand of a

free A-module. This proves that P is projective.
Example (7.3.5). — Let A be a commutative ring, let S be a multiplicative
subset of A and let P be a projective A-module. Then S−1 P is a projective
S−1 A-module.
Let indeed L be a free A-module admitting submodules L0 and L00 such
that L = L0 ⊕ L00 and L0 ' P. Then the modules S−1 L0 and S−1 L00 identify to
submodules of the free S−1 A-module S−1 L such that S−1 L = S−1 L0 ⊕S−1 L00.
In particular, S−1 L0 is a projective S−1 A-module. Moreover, S−1 L0 is
isomorphic to S−1 P, so that S−1 P is a projective S−1 A-module.
Remark (7.3.6). — Let P be a projective A-module and let S ⊂ P be a
generating set. The proof of implication (i)⇒(ii) in proposition 7.3.2
shows that P is (isomorphic to) a direct summand of the free A-module
A(S) . In particular, if P is a finitely generated projective A-module, then
P is a direct summand of a free finitely generated A-module.
Example (7.3.7). — Let A be a principal ideal domain and let P be a
finitely generated projective A-module. Then, there exists an integer 𝑛
such that P is a direct summand of A𝑛 . In particular, P is isomorphic to
a submodule of A𝑛 . By proposition 5.4.1, P is free.
Theorem (7.3.8) (Kaplansky). — Let A be a local ring and let P be a projective
A-module. Then P is free.
Let A be a ring and let J be its Jacobson radical. We recall from
definition 2.1.8 that A is local if and only if A/J is a division ring; then J
is the unique maximal left ideal of A.
Remark (7.3.9). — Let us first prove this theorem under the additional
assumption that P be finitely generated. Let J be the maximal left ideal
of A. Let (𝑒 𝑖 )16𝑖6𝑛 be a family of minimal cardinality in P such that the 𝑒 𝑖
generate P/PJ; let us prove that it is a basis of P.
We first consider the morphism 𝜑 : A𝑛 → P defined by (𝑎1 , . . . , 𝑎 𝑛 ) ↦→
𝑒 𝑖 𝑎 𝑖 . Let P0 = Im(𝜑); by assumption, one has P = P0 + P0J. By
Í
Nakayama’s lemma (corollary 6.1.2), one has P = P0 and 𝜑 is surjective.
Let us now prove that 𝜑 is an isomorphism. Let Q = Ker(𝜑). Since
P is projective, the morphism 𝜑 has a right inverse 𝜓, and the image
338 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

of 𝜓, which is isomorphic to Q, is a direct summand of Q. We get an

isomorphism P ⊕ Q ' A𝑛 . Taking quotients modulo J, we also have an
isomorphism P/PJ ⊕ Q/QJ ' (A/J)𝑛 . Since 𝑛 = dimA/J (P/PJ), we obtain
the equality Q = QJ. Since Q is a direct summand of A𝑛 , it is finitely
generated. By Nakayama’s lemma again (theorem 6.1.1), one has Q = 0.
Consquently, 𝜑 is an isomorphism and P is free.

The remaining of this section is devoted to the proof of Kaplansky’s

theorem.

Lemma (7.3.10). — Let A be a ring, let P be a projective A-module. There

exists a family (P𝑖 ) of É
submodules of P, each of them generated by a countable
family, such that P = P𝑖 .

Proof. — We may assume that there are a free A-module M containing P

and a submodule Q of M such that M = P ⊕ Q. Let (𝑒 𝑖 )𝑖∈I be a basis
of M, so that we identify M with A(I) . For any subset J of I, let MJ be the
submodule of M generated by the elements 𝑒 𝑗 , for 𝑗 ∈ J. Observe that
MJ has a direct summand in M, namely the submodule of M generated
by the elements 𝑒 𝑖 , for 𝑖 ∈ I J; in particular, a submodule of some MJ
which has a direct summand in MJ also has a direct summand in M.
By transfinite induction,1 we construct an increasing sequence (J𝛼 ) of
subsets of I satisfying the following properties:
(i) For any ordinal 𝛽, J𝛽+1 J𝛽 is at most countable, and is empty if and
only if J𝛽 = I;
(ii) If 𝛼 is a limit ordinal, then J𝛼 is the union of the J𝛽 for 𝛽 < 𝛼;
(iii) For any 𝛼, MJ𝛼 = (P ∩ MJ𝛼 ) ⊕ (Q ∩ MJ𝛼 ).
Let 𝛼 be an ordinal.
Is 𝛼 is a limit ordinal, then the prescription (ii) defines a subset J𝛼 of I.
Let us show that condition (iii) is satisfied. Since P ∩ Q = 0, it suffices to
show that MJ𝛼 is the sum of P ∩ LJ𝛼 and Q ∩ MJ𝛼 . So let 𝑚 ∈ MJ𝛼 . Only
finitely many coordinates of 𝑚 in the basis (𝑒 𝑖 ) are nonzero, so that there
exists 𝛽 < 𝛼 such that 𝑚 ∈ MJ𝛽 ; consequently, 𝑚 ∈ (P ∩ LJ𝛽 ) + (Q ∩ MJ𝛽 )
and a fortiori, 𝑚 ∈ (P ∩ LJ𝛼 ) + (Q ∩ MJ𝛼 ).

1Faut-il zornifier cette démonstration?

 7.3. PROJECTIVE MODULES 339

Let us now assume that 𝛼 is of the form 𝛽 + 1, for some ordinal 𝛽. If

I = J𝛽 , we let J𝛼 = J𝛽 . Otherwise, choose some element 𝑗 ∈ I J𝛽 and
define a sequence (K𝑛 )𝑛∈N of finite subsets of I as follows. One sets
K0 = {𝑗}. Assume K𝑛 has been defined; for every element 𝑘 ∈ K𝑛 , write
𝑒 𝑘 as a sum 𝑝 𝑘 + 𝑞 𝑘 , where 𝑝 𝑘 ∈ P and 𝑞 𝑘 ∈ Q. Then, let K𝑛+1 be the union,
for all 𝑘 ∈ K𝑛 , of the indices 𝑖 ∈ I such that the 𝑖th coordinate of 𝑝 𝑘 or 𝑞 𝑘
is nonzero. For every integer 𝑛, K𝑛 is finite, so that the union K = K𝑛
Ð
is either finite or countable. Then, set J𝛼 = J𝛽 ∪ K; condition (i) is satisfied.
Let us prove that (iii) holds. Again, it suffices to show that any element
𝑚 ∈ MJ𝛼 can be written as the sum of an element of P ∩ MJ𝛼 and of an
element of Q ∩ MJ𝛼 . It suffices to prove this property for every element
of the form 𝑒 𝑘 , for some 𝑘 ∈ J𝛼 . It holds by induction if 𝑘 ∈ J𝛽 , hence we
may assume that 𝑘 ∈ K. By definition, there is an integer 𝑛 such that
𝑘 ∈ K𝑛 . Then 𝑒 𝑘 = 𝑝 𝑘 + 𝑞 𝑘 and by construction of K𝑛+1 , 𝑝 𝑘 ∈ P ∩ MK𝑛+1
and 𝑞 𝑘 ∈ Q ∩ MK𝑛+1 . In particular, 𝑝 𝑘 ∈ P ∩ MJ𝛼 and 𝑞 𝑘 ∈ Q ∩ MJ𝛼 , as was
to be shown.
This concludes the construction of the family (J𝛼 ) by transfinite induc-
tion.
Since not all ordinals can be embedded in a given set, this transfinite
construction is necessarily stationary. Let 𝜆 be an ordinal such that
J𝜆+1 = J𝜆 . By condition (i), one has J𝜆 = I.
ÉWe then define a sequence (S𝛼 ) of submodules of P such that P ∩ MJ𝛼 =
𝛽6𝛼 S𝛽 for every ordinal 𝛼 6 𝜆. If 𝛼 is a limit ordinal, we set S𝛼 = 0.
Otherwise, there exists an ordinal 𝛽 such that 𝛼 = 𝛽 + 1; then, P ∩ MJ𝛽
is a direct summand of MJ𝛽 , hence a direct summand of M. Restricting
to P ∩ MJ𝛼 a left inverse of the injection (P ∩ MJ𝛽 ) → M, we see that
P ∩ MJ𝛽 has a direct summand SÉ 𝛼 in P ∩ MJ𝛼 . Moreover, S𝛼 is countably
generated. Then P = P ∩ MJ𝜆 = 𝛼6𝜆 S𝛼 . 

Say that an A-module M has property (K) if for every element 𝑚 ∈ M, there
exists a direct summand N of M containing 𝑚 which is free.

Lemma (7.3.11). — Let A be a ring, let P be a countably generated A-module.

Assume that any direct summand M of P satisfies property (K). Then P is a free
A-module.
340 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

Proof. — Let (𝑝 𝑛 )𝑛>1 be a generating family of P. We construct by

induction families (P𝑛 )>1 and (Q𝑛 )𝑛>0 of submodules of P such that P0 =
P, P𝑛−1 = P𝑛 ⊕Q𝑛 for every 𝑛 > 1, and 𝑝1 , . . . , 𝑝 𝑛 ∈ Q1 ⊕· · ·⊕Q𝑛 , and such
that for every integer 𝑛, Q𝑛 is free. Assume P0 , . . . , P𝑛−1 , Q1 , . . . , Q𝑛−1
have been defined. Observe that P = P𝑛−1 ⊕ (Q1 ⊕ · · · ⊕ Q𝑛−1 ) and let
𝑓 : P → P𝑛−1 be the projection with kernel Q1 ⊕ · · · ⊕ Q𝑛−1 . Apply
property (K) to the element 𝑓 (𝑝 𝑛 ) of P𝑛−1 : there exists a decomposition
of P𝑛−1 as a direct sum P𝑛 ⊕ Q𝑛 , where Q𝑛 is free and contains 𝑓 (𝑝 𝑛 ).
Consequently, 𝑝 𝑛 ∈ Q1 ⊕ · · · ⊕ Q𝑛 . By induction on 𝑛, this concludes the
asserted construction.
Let us now show that P is the direct sum of the family (Q𝑛 )𝑛>1 . By
Í
construction, this family is in direct sum in 𝑛>1 Q𝑛 . Moreover, for
every 𝑚 > 1, one has 𝑝 𝑚 ∈ Q𝑚 ⊂ 𝑛>1 Q𝑛 . Since the family (𝑝 𝑚 )
Í
Í
generates P, this proves that P = 𝑛>1 Q𝑛 and concludes the proof of
the lemma. 

Lemma (7.3.12). — Let A be a local ring. Any projective A-module has

property (K).
Proof. — Let P be a projective A-module. Let Q be a A-module such
that M = P ⊕ Q is a free A-module. Let 𝑝 ∈ P. Let (𝑒 𝑖 )𝑖∈I be a basis of M
in which the coordinates (𝑎 𝑖 ) of 𝑝 have the least possible nonzero entries.
For simplicity, assume that these coordinates have indices 1, . . . , 𝑛. For
each 𝑖 ∈ {1, . . . , 𝑛}, write 𝑒 𝑖 = 𝑝 𝑖 + 𝑞 𝑖 , where 𝑝 𝑖 ∈ P and 𝑞 𝑖 ∈ Q. Then
𝑛
Õ 𝑛
Õ 𝑛
Õ
𝑝= 𝑒𝑖 𝑎𝑖 = 𝑝𝑖 𝑎𝑖 + 𝑞𝑖 𝑎𝑖 ,
𝑖=1 𝑖=1 𝑖=1
is a decomposition of 𝑝 as the sum of an element of P and an element
of Q, so that
𝑛
Õ 𝑛
Õ
𝑝= 𝑒𝑖 𝑎𝑖 = 𝑝𝑖 𝑎𝑖 .
𝑖=1 𝑖=1
We may then decompose each vector 𝑝 𝑖 along the basis (𝑒 𝑖 ): letting 𝑏 𝑗𝑖
be the 𝑗th coordinate, we write
𝑛
Õ
𝑝𝑖 = 𝑒 𝑗 𝑏 𝑗𝑖 + 𝑝 0𝑖 ,
𝑗=1
 7.3. PROJECTIVE MODULES 341

where 𝑝 0𝑖 = 𝑗∈J {1,...,𝑛} 𝑒 𝑗 𝑏 𝑗𝑖 is the sum of terms corresponding to the

Í
coordinates other than 1, . . . , 𝑛. Combining the last two equations, we
obtain
𝑛
Õ 𝑛 Õ
Õ 𝑛 𝑛
Õ Õ
𝑒𝑖 𝑎𝑖 = 𝑒 𝑗 𝑏 𝑗𝑖 𝑎 𝑖 + 𝑒 𝑗 𝑏 𝑗𝑖 𝑎 𝑖 .
𝑖=1 𝑖=1 𝑗=1 𝑖=1 𝑗∈J {1,...,𝑛}

Equating the coefficients of 𝑒1 , . . . , 𝑒 𝑛 , we get

𝑛
Õ
𝑎𝑗 = 𝑏 𝑗𝑖 𝑎 𝑖 , (1 6 𝑗 6 𝑛).
𝑖=1

If 𝑗 ∈ J {1, . . . , 𝑛}, we also get

𝑛
Õ
𝑏 𝑗𝑖 𝑎 𝑖 = 0,
𝑖=1
hence
𝑛
Õ 𝑛 Õ
Õ 𝑛
𝑝= 𝑒𝑗 𝑎𝑗 = 𝑒 𝑗 𝑏 𝑗𝑖 𝑎 𝑖 .
𝑗=1 𝑖=1 𝑗=1
Let us assume that there exists 𝑖, 𝑗 such that 1 6 𝑖, 𝑗 6 𝑛 such that
1 − 𝑏 𝑖𝑖 (if 𝑗 = 𝑖) or 𝑏 𝑗𝑖 (if 𝑗 ≠ 𝑖) is invertible in A. Then we can then
rewrite 𝑎 𝑗 as a left linear combination
𝑛
Õ
𝑎𝑗 = 𝑐𝑖 𝑎𝑖
𝑖=1
𝑖≠𝑗

of the 𝑎 𝑖 , for 𝑖 ≠ 𝑗. Let then write

𝑛
Õ 𝑛
Õ 𝑛
Õ
𝑝= 𝑒𝑖 𝑎𝑖 = (𝑒 𝑖 𝑎 𝑖 + 𝑒 𝑗 𝑐 𝑖 𝑎 𝑖 ) = (𝑒 𝑖 + 𝑒 𝑗 𝑐 𝑖 )𝑎 𝑖 .
𝑖=1 𝑖=1 𝑖=1
𝑖≠𝑗 𝑖≠𝑗

Let us then define a family (𝑒 𝑖0) by 𝑒 𝑖0 = 𝑒 𝑖 if 𝑖 ∉ {1, . . . , 𝑛}, 𝑒 0𝑗 = 𝑒 𝑗 , and

𝑒 𝑖0 = 𝑒 𝑖 + 𝑒 𝑗 𝑐 𝑖 if 𝑖 ∈ {1, . . . , 𝑛} and 𝑖 ≠ 𝑗. Observe that it is a basis of M.
However, the vector 𝑝 has 𝑛 − 1 nonzero coordinates in this new basis,
which contradicts the assumption made at the beginning of the proof.
This shows that 1 − 𝑏 𝑖𝑖 (for 1 6 𝑖 6 𝑛) and 𝑏 𝑖𝑗 (for 1 6 𝑖 ≠ 𝑗 6 𝑛) belong
to the Jacobson radical J of the local ring A. In other words, the image of
the matrix B = (𝑏 𝑗𝑖 ) in the quotient ring A/J is the identity matrix. By
342 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

corollary 6.1.3, the matrix B = (𝑏 𝑗𝑖 ) is invertible in M𝑛 (A). Consequently,

the family ( 𝑓𝑖 ) defined by 𝑓 𝑗 = 𝑛𝑖=1 𝑒 𝑗 𝑏 𝑗𝑖 (for 𝑗 ∈ {1, . . . , 𝑛}) and 𝑓 𝑗 = 𝑒 𝑗
Í
(otherwise) is a basis of M.
Since 𝑝 𝑖 = 𝑓𝑖 + 𝑝 0𝑖 and 𝑝 0𝑖 belongs to the submodule spanned by the 𝑒 𝑗 ,
for 𝑗 ∉ {1, . . . , 𝑛}, we see that the family consisting of 𝑝1 , . . . , 𝑝 𝑛 and
the 𝑒 𝑗 , for 𝑗 ∈ J {1, . . . , 𝑛} is a basis of M.
By construction, the submodule F of P generated by 𝑝 1 , . . . , 𝑝 𝑛 ; con-
tains 𝑝. It is also free and it admits a direct summand in M, because
it is generated by a subset of a basis of M. Consequently, the injection
F → M admits a left inverse, so that the injection F → P has a left inverse
too; this implies that F has a direct summand in P (lemma 7.1.6). This
concludes the proof of the lemma. 

Proof of Kaplansky’s theorem. — Let A be a local ring and let P be a projec-

tive A-module. By lemma 7.3.10, there exists a family (P𝑖 ) of countable
generated submodules of P of which P is the direct sum. Since P𝑖 is a
direct summand of a projective A-module, it is itself projective. The
ring A being local, P𝑖 satisfies property (K), by lemma 7.3.12. Moreover,
each direct summand of P𝑖 is projective, hence satisfies property (K), so
that P𝑖 satisfies the the hypotheses of lemma 7.3.11. Consequently, P𝑖
is a free A-module. It follows that P is a direct sum of free A-modules,
hence is free. 

Proposition (7.3.13). — Let A be a commutative ring and let M be a finitely

presented A-module. The following properties are equivalent:
(i) The A-module M is projective;
(ii) For every prime ideal P of A, the AP -module MP is free;
(iii) For every maximal ideal P of A, the AP -module MP is free.

Proof. — The implication (i)⇒(ii) follows from Kaplansky’s theorem

(theorem 7.3.8) while the implication (ii)⇒(iii) is obvious. Let us thus
assume that the AP -module MP is free, for every maximal ideal P of A.
Let 𝑢 : N → N0 be a surjective morphism of A-modules and let us show
that the map 𝑢∗ : HomA (M, N) → HomA (M, N0) is surjective. It is a
morphism of A-modules; let Q be its cokernel. Let P be a maximal ideal
 7.4. INJECTIVE MODULES 343

of A. By exactness of localization, the exact sequence

𝑢∗
→ HomA (M, N0) → Q → 0
HomA (M, N) −
gives rise to an exact sequence of AP -modules
(𝑢∗ )P
HomA (M, N)P −−−→ HomA (M, N0)P → QP → 0.
On the other hand, the hypothesis that M is a finitely presented A-
module allows to identify HomA (M, N)P with HomAP (MP , NP ), and
HomA (M, N0)P with HomAP (MP , N0P ) (proposition 7.2.7). The preceding
exact sequence then becomes
(𝑢P )∗
HomAP (MP , NP ) −−−→ HomAP (MP , N0P ) → QP → 0.
Since MP is a projective AP -module, the morphism (𝑢P )∗ is surjective,
hence QP = 0.
This implies that Supp(Q) contains no maximal ideal of A. On the other
hand, if Supp(Q) contains a prime ideal P, then it contains all maximal
ideals that contain P (theorem 6.5.4, b) and Krull’s theorem 2.1.3), so that
Supp(Q) = ∅. By theorem 6.5.4, a), this shows that Q = 0, so that 𝑢∗ is
surjective. Consequently, M is a projective A-module. 

7.4. Injective modules

Definition (7.4.1). — Let A be a ring. One says that an A-module M is
injective if every short exact sequence
0→M→N→P→0
is split.

The notion of an injective module is kind of dual of that of a projective

module. However, the category of modules is quite different from its
opposite category. As is pointed out by Matsumura in (Matsumura,
1986, p. 277), there does not exist a notion that would be the dual to that
of a free module. Consequently, proposition 7.3.2 has no analogue for
injective modules and the existence of injective modules is not obvious.
We start with a partial counterpart to proposition 7.3.2.
344 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

Proposition (7.4.2). — Let A be a ring, let M be a right A-module. The

following conditions are equivalent.
(i) The module M is injective;
(ii) For every right A-modules N and P, every injective morphism 𝑖 : N → P
and any morphism 𝑓 : N → M, there exists a morphism 𝑓 : P → M such that
𝑓 = 𝑔 ◦ 𝑖;
(iii) For every right ideal I of A and any morphism 𝑓 : I → M, there exists a
morphism 𝑔 : A → M such that 𝑓 = 𝑔| I ;
(iv) Every injective morphism from M to any A-module has a left inverse.
The second condition is often taken as the definition of an injective
module.
Proof. — The equivalence of (i) and (iv) follows from lemma 7.1.6.
(iv)⇒(ii). Let 𝑖 : N → P be an injective morphism and let 𝑓 : N → M be
a morphism. We want to prove that there exists a morphism 𝑔 : P → M
such that 𝑓 = 𝑔 ◦ 𝑖.
Let Q be the image of the morphism 𝑥 ↦→ ( 𝑓 (𝑥), −𝑖(𝑥)) from N to M × P.
let 𝑝 : M × P → (M × P)/Q be the canonical projection and let 𝑗 : M →
(M × P)/Q be the map given by 𝑗(𝑥) = 𝑝(𝑥, 0) for 𝑥 ∈ M. Let us show
that 𝑗 is injective; let 𝑥 ∈ M be such that (𝑥, 0) ∈ Q. Then there exists
𝑦 ∈ N such that 𝑓 (𝑦) = 𝑥 and −𝑖(𝑦) = 0. Since 𝑖 is injective, 𝑦 = 0, hence
𝑥 = 0, as was to be shown.
By (iv), there exists a morphism 𝛾 : (M × P)/Q → M such that 𝛾 ◦ 𝑗 =
idM , that is, 𝛾(𝑝(𝑥, 0)) = 𝑥 for every 𝑥 ∈ M. Let 𝑔 : P → M be the
morphism given by 𝑦 ↦→ 𝛾(𝑝(0, 𝑦)). For every 𝑥 ∈ N, one has
𝑔(𝑖(𝑥)) = 𝛾(𝑝(0, 𝑖(𝑥))) = 𝛾(𝑝(− 𝑓 (𝑥), 𝑖(𝑥))) + 𝛾(𝑝( 𝑓 (𝑥), 0)) = 𝑓 (𝑥),
so that 𝑓 = 𝑔 ◦ 𝑖, as required.
(ii)⇒(iii). It suffices to take for modules N = I and P = A, the injective
morphism 𝑖 being the injection from I into A.
(iii)⇒(iv). Let 𝑖 : M → N be an injective morphism. We want to prove
that there exists a morphism 𝑓 : N → M which is a left inverse of 𝑖, that
is, 𝑓 ◦ 𝑖 = idM .
Let ℱ be the set of all pairs (N0 , 𝑓 ) where N0 is a submodule of N
containing 𝑖(M) and 𝑓 : N0 → M is a morphism of A-modules such that
𝑓 ◦ 𝑖 = idM .
 7.4. INJECTIVE MODULES 345

Let us define an ordering ≺ on ℱ by setting

(N01 , 𝑓1 ) ≺ (N02 , 𝑓2 ) ⇔ N01 ⊂ N02 and 𝑓2 | N01 = 𝑓1 .
Let us show that the ordered set ℱ is inductive. Let indeed (N𝛼 , 𝑓𝛼 )
be a totally ordered family of elements of ℱ. If it is empty, we take
for its upper bound the pair (N0 = 𝑖(M), 𝑓 ), where 𝑓 : 𝑖(M) → M is the
inverse of the injective map 𝑖. Otherwise, since the nonempty family
(N𝛼 ) of submodules of N is totally ordered, its union N0 = N𝛼 is a
Ð
submodule of N. We may then define a morphism 𝑓 : N0 → M by setting
𝑓 (𝑦) = 𝑓𝛼 (𝑦) for any 𝛼 such that 𝑦 ∈ N𝛼 . Indeed, it does not depend on
the choice of 𝛼.
By Zorn’s lemma (theorem A.2.11), the ordered set ℱ has a maximal
element (N0 , 𝑓 ). Let us prove by contradiction that N0 = N. Otherwise,
let 𝑦 ∈ N N0 and let I = {𝑎 ∈ A ; 𝑦𝑎 ∈ N0 }.
Let 𝜑 : I → N be the morphism given by 𝑎 ↦→ 𝑓 (𝑦𝑎), for 𝑎 ∈ I. By
assumption, there exists a morphism 𝜓 : A → N such that 𝜓| I = 𝜑. Let
then set N01 = N0 + 𝑦A and define a morphism 𝑓10 : N01 → M by setting
𝑓1 (𝑥 + 𝑦𝑎) = 𝑓 (𝑥) + 𝜓(𝑎) for 𝑥 ∈ N0 and 𝑎 ∈ A. It is well-defined; indeed,
if 𝑥 + 𝑦𝑎 = 𝑥 0 + 𝑦𝑎 0, with 𝑥, 𝑥 0 ∈ N0 and 𝑎, 𝑎 0 ∈ A, then 𝑥 − 𝑥 0 = 𝑦(𝑎 0 − 𝑎),
so that 𝑎 0 − 𝑎 ∈ I and
𝜓(𝑎 0) − 𝜓(𝑎) = 𝜑(𝑎 0 − 𝑎) = 𝑓 (𝑦(𝑎 0 − 𝑎)) = 𝑓 (𝑥 − 𝑥 0) = 𝑓 (𝑥) − 𝑓 (𝑥 0),
so that 𝑓 (𝑥 0)+𝜓(𝑎 0) = 𝑓 (𝑥)+𝜓(𝑎). It is easy to show that 𝑓1 is a morphism
of A-modules. Moreover, the construction of 𝑓1 shows that 𝑓1 | N0 = 𝑓 . In
particular, (N01 , 𝑓1 ) is an element of ℱ, but this contradicts the hypothesis
that (N0 , 𝑓 ) is a maximal element of ℱ, because N01 ) N0.
It follows that any maximal element of ℱ is of the form (N, 𝑓 ), where
𝑓 : N → M is a left inverse to 𝑖. This concludes the proof of (iv). 

Corollary (7.4.3). — Products of injective A-modules are injective.

Proof. — Let (M𝑠 )𝑠∈S be a family of injective right A-modules and let
M = 𝑠∈S M𝑠 . For every 𝑠 ∈ S, let 𝑝 𝑠 be the canonical projection from M
Î
to M𝑠 .
Let I be a right ideal of A, let 𝑓 : I → M be a morphism. For every 𝑠 ∈ S,
let 𝑓𝑠 = 𝑝 𝑠 ◦ 𝑓 . Since M𝑠 is injective, there exists a morphism 𝑔𝑠 : A → M𝑠
such that 𝑔𝑠 | I = 𝑓𝑠 . Let 𝑔 = (𝑔𝑠 ) : A → M be the unique morphism such
346 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

that 𝑝 𝑠 ◦ 𝑔 = 𝑔𝑠 for every 𝑠 ∈ S. One has 𝑔| I = (𝑔𝑠 | I ) = ( 𝑓𝑠 ) = 𝑓 . This

shows that M is an injective A-module. 

Corollary (7.4.4). — Let A be a principal ideal domain and let M be an A-

module. Then M is an injective module if and only if for any nonzero 𝑎 ∈ A,
the morphism 𝜇𝑎 : M → M, 𝑥 ↦→ 𝑎𝑥, is surjective.
The condition that the morphisms 𝜇𝑎 be bijective, for all 𝑎 ≠ 0, is
rephrased as saying that M is a divisible A-module.
Proof. — Let us assume that M is an injective module; let 𝑎 ∈ A {0}.
Let 𝑥 ∈ M. Let I be the ideal (𝑎) in A. Since 𝑎 ≠ 0 and A is a domain, the
map from A to I given by 𝑡 ↦→ 𝑎𝑡 is an isomorphism; consequently, there
exists a unique morphism 𝑓 : I → M such that 𝑓 (𝑎𝑡) = 𝑡𝑥 for every 𝑡 ∈ A.
By property (iii) in proposition 7.4.2, there exists a morphism 𝑔 : A → M
such that 𝑔(𝑎𝑡) = 𝑓 (𝑎𝑡) = 𝑡𝑥 for any 𝑡 ∈ A. Then 𝑎 𝑔(1) = 𝑔(𝑎) = 𝑥. This
implies that 𝜇𝑎 is surjective.
Conversely, let us assume that 𝜇𝑎 be surjective for every nonzero 𝑎 ∈ A.
Let I be an ideal of A, let 𝑓 : I → M be a morphism of A-modules; we need
to show that there exists a morphism 𝑔 : A → M such that 𝑔| I = 𝑓 . If
I = 0, we may take 𝑔 = 0. Otherwise, since A is a principal ideal domain,
there exists a nonzero element 𝑎 ∈ A such that I = (𝑎). By hypothesis,
there exists 𝑥 ∈ M such that 𝑎𝑥 = 𝑓 (𝑎). Let 𝑔 : A → M be the morphism
given by 𝑔(𝑡) = 𝑡𝑥. For any 𝑡 ∈ A, one has 𝑔(𝑎𝑡) = 𝑎𝑡𝑥 = 𝑡 𝑓 (𝑎) = 𝑓 (𝑎𝑡),
so that 𝑔| I = 𝑓 . By proposition 7.4.2, the A-module M is injective. 

Example (7.4.5). — The Z-module Q/Z is injective.

Recall that the existence of projective A-modules is no great mistery,
since free modules are projective. More generally, every A-module is the
quotient of a free A-module, a fortiori a projective A-module. The dual
property asserts that every A-module is a submodule of an injective
A-module and is more difficult. It is the object of the following theorem.
Theorem (7.4.6) (Baer). — Let A be a ring, let M be an A-module. There exists
an injective A-module N and an injective morphism 𝑖 : M → N.
Let A be a ring, let M be a right A-module (resp. a left A-module).
Then M∗ = HomZ (M, Q/Z) is an abelian group. We endow it with
 7.4. INJECTIVE MODULES 347

the structure of a left A-module (resp. of a right A-module) given by

(𝑎 · 𝑓 )(𝑥) = 𝑓 (𝑥𝑎) (resp. ( 𝑓 · 𝑎)(𝑥) = 𝑓 (𝑎𝑥)) for every 𝑎 ∈ A, 𝑓 ∈ M∗ and
𝑥 ∈ M.

Lemma (7.4.7). — Let A be a ring, let M be a right A-module. For any 𝑥 ∈ M,

let 𝑖(𝑥) be the map from M∗ to Q/Z given by 𝑓 ↦→ 𝑓 (𝑥). One has 𝑖(𝑥) ∈ M∗∗
and the map 𝑖 : M → M∗∗ so defined is an injective morphism of A-modules.

Proof. — It is clear that 𝑖(𝑥) is a morphism of abelian groups, hence

𝑖(𝑥) ∈ M∗∗ . Moreover, for every 𝑥 ∈ M and every 𝑎 ∈ A, one has
𝑖(𝑥𝑎)( 𝑓 ) = 𝑓 (𝑥𝑎) = (𝑎 · 𝑓 )(𝑥) = 𝑖(𝑥)(𝑎 · 𝑓 ) = (𝑖(𝑥) · 𝑎)( 𝑓 ).
Finally,
𝑖(𝑥 + 𝑦)( 𝑓 ) = 𝑓 (𝑥 + 𝑦) = 𝑓 (𝑥) + 𝑓 (𝑦) = 𝑖(𝑥)( 𝑓 ) + 𝑖(𝑦)( 𝑓 ),
for every 𝑥, 𝑦 ∈ M, which shows that 𝑖 : M → M∗∗ is a morphism of right
A-modules.
Let us finally prove that 𝑖 is injective. Let 𝑥 ∈ M be a nonzero
element. We need to show that there exists a morphism of abelian
groups, 𝑔 : M → Q/Z, such that 𝑔(𝑥) ≠ 0. The abelian group 𝑥Z
generated by 𝑥 in M is either isomorphic to Z, or to Z/𝑛Z, for some
integer 𝑛 > 2. We define 𝑓 : 𝑥Z → Q/Z by 𝑓 (𝑥𝑚) = 𝑚/2 (mod Z) in
the former case, and 𝑓 (𝑥𝑚) = 𝑚/𝑛 (mod Z) in the latter. Observe that
𝑓 (𝑥) ≠ 0.
Since Q/Z is an injective Z-module, there exists a morphism 𝑔 : M →
Q/Z of Z-modules such that 𝑔| 𝑥Z = 𝑓 (proposition 7.4.2). The mor-
phism 𝑔 belongs to M∗ and satisfies 𝑔(𝑥) ≠ 0. 

Lemma (7.4.8). — Let A be a ring.

a) The right A-module (A𝑠 )∗ is injective.
b) For every free left A-module M, the right A-module M∗ is injective.

Proof. — a) Let I be a right ideal of A and let 𝑓 : I → (A𝑠 )∗ be a morphism

of A-modules. Let 𝜑 : I → Q/Z be the map given by 𝜑(𝑥) = 𝑓 (𝑥)(1),
for 𝑥 ∈ I. It is a morphism of abelian groups. Since Q/Z is an injective
Z-module, there exists a morphism 𝛾 : A → Q/Z of abelian groups such
that 𝛾| I = 𝜑.
348 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

Let 𝑔 : A → (A𝑠 )∗ be the map defined by 𝑔(𝑥)(𝑦) = 𝛾(𝑥𝑦), for 𝑥, 𝑦 ∈ A.

It is additive. Let 𝑎, 𝑥 ∈ A; then 𝑔(𝑥𝑎) is the element 𝑦 ↦→ 𝛾(𝑥𝑎𝑦)
of (A𝑠 )∗ , while 𝑔(𝑥) · 𝑎 is the map 𝑦 ↦→ 𝑔(𝑥)(𝑎𝑦) = 𝛾(𝑥𝑎𝑦). Consequently,
𝑔(𝑥𝑎) = 𝑔(𝑥) · 𝑎 and 𝑔 is A-linear.
Let 𝑥 ∈ I. For any 𝑦 ∈ A, one has 𝑔(𝑥)(𝑦) = 𝛾(𝑥𝑦) = 𝜑(𝑥𝑦) since
𝑥𝑦 ∈ I and 𝛾 extends 𝜑. Consequently, 𝑔(𝑥)(𝑦) = 𝑓 (𝑥𝑦)(1). Since 𝑓 is
A-linear, 𝑓 (𝑥𝑦) = 𝑓 (𝑥) · 𝑦, so that 𝑓 (𝑥𝑦)(1) = 𝑓 (𝑥)(𝑦). This shows that
𝑔(𝑥)(𝑦) = 𝑓 (𝑥)(𝑦), hence 𝑔(𝑥) = 𝑓 (𝑥). We thus have shown that 𝑔| I = 𝑓 .
By proposition 7.4.2, the right A-module (A𝑠 )∗ is injective.
b) Let (𝑒 𝑖 )𝑖∈I be a basis of M, so that M is identified with (A𝑠 )(I) . Then, the
morphism 𝑓 ↦→ ( 𝑓 (𝑒 𝑖 )) is an isomorphism of right A-modules from M∗
to (A∗𝑠 )I . Consequently, M∗ is a product of injective A-modules. It then
follows from corollary 7.4.3 that M∗ is an injective A-module. 

Proof of theorem 7.4.6. — We are now able to prove Baer’s theorem that
every A-module is a submodule of an injective A-module.
Let M be a right A-module. Let F be a free left A-module together with
a surjective A-linear morphism 𝑝 : F → M∗ . For example, one may take
∗
for F the free-A-module A(M ) with basis indexed by M∗ , and for 𝑝 the map
that sends the basis vector 𝑒 𝑓 indexed by 𝑓 ∈ M∗ to 𝑓 , for every 𝑓 ∈ M∗ .
Since 𝑝 is surjective and A-linear the map 𝑝 ∗ from M∗∗ to F∗ given by
𝜑 ↦→ 𝜑 ◦ 𝑝 is injective and A-linear. Let 𝑖 be the morphism from M to M∗∗
defined in lemma 7.4.7; it is injective. Then, the map 𝑝 ∗ ◦ 𝑖 : M → F∗ is
A-linear and injective. By lemma 7.4.8, the right A-module F∗ is injective.
This concludes the proof. 

7.5. Exactness conditions for functors

7.5.1. Definitions. — Let A and B be two rings. Let us recall that a
functor F from the category Mod A of A-modules to the category Mod B of
B-modules associates to each A-module M a B-module F(M) and to each
morphism 𝑓 : M → N of A-modules a morphism F( 𝑓 ) : F(M) → F(N)
subject to the following constraints:
– For every A-module M, one has F(idM ) = idF(M) ;
 7.5. EXACTNESS CONDITIONS FOR FUNCTORS 349

– For all A-modules M, N, P and all morphisms 𝑓 : M → N and

𝑔 : N → P, one has F(𝑔 ◦ 𝑓 ) = F(𝑔) ◦ F( 𝑓 ).
Such a functor is said to be covariant by opposition to contravariant functors
which reverse the direction of maps.
Indeed, a contravariant functor F from the category of A-modules
to the category of B-modules associates to each A-module M a B-
module F(M) and to each morphism 𝑓 : M → N of A-modules a mor-
phism F( 𝑓 ) : F(N) → F(N) subject to the analogous requirements:
– For every A-module M, one has F(idM ) = idF(M) ;
– For all A-modules M, N, P and all morphisms 𝑓 : M → N and
𝑔 : N → P, one has F(𝑔 ◦ 𝑓 ) = F( 𝑓 ) ◦ F(𝑔).

Example (7.5.2) (Forgetful functors). — Let A and B be rings, let 𝑢 : B → A

be a morphism of rings. Then, every A-module M may be viewed as a B-
module 𝑢 ∗ (M), with external multiplication given by the composition of
the morphism 𝑢 and of the external multiplication of M. Let 𝑓 : M → N
be a morphism of A-modules, the map 𝑓 can be viewed as a map
𝑢 ∗ ( 𝑓 ) : 𝑢 ∗ (M) → 𝑢 ∗ (N) which is obviously B-linear. So 𝑢 ∗ (M) and 𝑢 ∗ ( 𝑓 )
are just M and 𝑓 with different names to indicate that we no more
consider them as A-modules but only as B-modules through 𝑢. In
particular, 𝑢 ∗ (𝑔 ◦ 𝑓 ) = 𝑢 ∗ (𝑔) ◦ 𝑢 ∗ ( 𝑓 ) if 𝑔 : N → P is a second morphism
of A-modules.
We thus have defined a covariant functor from the category of A-
modules to the category of B-modules. In particular, if B = Z, we
obtain a functor from the category Mod A of A-modules to the category
AbGr = Mod Z of abelian groups. Since these functors forget the initial
A-module structure (at least, part of it), they are called forgetful functors.

7.5.3. — One says that a functor F (covariant or contravariant) from the

category Mod A of A-modules to the category Mod B of B-modules is
additive if the maps 𝜑 ↦→ F(𝜑) from HomA (M, N) to HomB (F(M), F(N))
(in the covariant case, and to HomB (F(N), F(M)) in the contravariant case)
are morphisms of abelian groups: F(0) = 0 and F( 𝑓 + 𝑔) = F( 𝑓 ) + F(𝑔)
for every 𝑓 , 𝑔 ∈ HomA (M, N).
350 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

Remark (7.5.4). — Let 𝑓 : M → N and 𝑔 : N → P two morphisms of

𝑓 𝑔
A-modules such that 𝑔 ◦ 𝑓 = 0, so that the diagram M → − N→ − P is a
complex. Then, for any covariant additive functor F : Mod A → Mod B ,
F( 𝑓 ) F(𝑔)
the diagram F(M) −−−→ F(N) −−−→ F(P) is a complex of B-modules.
Indeed, one has F(𝑔) ◦ F( 𝑓 ) = F(𝑔 ◦ 𝑓 ) = F(0) = 0.
Of course, an analogous remark holds for contravariant functors.
Definition (7.5.5). — Let F be an additive (covariant) functor from the category
of A-modules to the category of B-modules.
One says that F is left exact if for every exact sequence
0 → M → N → P,
the complex
0 → F(M) → F(N) → F(P)
is exact.
One says that F is right exact if for every exact sequence
M → N → P → 0,
the complex
F(M) → F(N) → F(P) → 0
is exact.
One says that F is exact if it is both left and right exact.
A left exact functor preserves injective morphisms; a right exact functor
preserves surjective morphisms. Observe that the left-exactness of F
means that the morphism F(M) → F(N) is an isomorphism from F(M) to
the kernel of the morphism F(N) → F(P). Similarly, the right-exactness
of F means that the morphism F(N) → F(P) is surjective and its kernel
is the image of the morphism F(M) → F(N). Saying that a functor F is
exact means that for any module M and any submodule N of M, F(N) is
identified (via F(𝑖)) a submodule of F(N) and F(M)/F(N) is identified (via
F(𝑝)) F(M/N), where 𝑖 : N → M and 𝑝 : M → M/N are the canonical
injection and surjection.
There are analogous definitions for contravariant additive functors.
Such a functor F is right exact if for every exact sequence of the form
0 → M → N → P, the complex F(P) → F(N) → F(M) → 0 is exact. One
 7.5. EXACTNESS CONDITIONS FOR FUNCTORS 351

says that F is left exact if for every exact sequence M → N → P → 0, the

complex 0 → F(P) → F(N) → F(M) is an exact sequence. One says that
F is exact if it is both left and right exact.
Forgetful functors associated to a morphism of rings transform any
diagram of modules into the same diagram, where only part of the
linearity of the maps has been forgotten. In particular, they transform
any exact sequence into an exact sequence: these forgetful functors are
exact.

Remark (7.5.6). — a) If a functor F is exact, then for every short exact

𝑢 𝑣 F(𝑢) F(𝑣)
sequence 0 → M → − N→− P → 0, the complex 0 → F(M) −−−→ F(N) −−−→
F(P) → 0 is exact, since the two complexes obtained by removing an
extremity are exact, by assumption.
Conversely, let us show that the exactness of such complexes is
sufficient. If 𝑢 : M → N is an injective morphism, considering the
𝑢
exact sequence 0 → M → − N → N/𝑢(M) → 0 and its image by
the functor F shows that F(𝑢) is injective. Similarly, if 𝑣 : N → P
is a surjective morphism, we see, considering the exact sequence
𝑣
0 → Ker(𝑣) → N → − P → 0 and its image by F, that F(𝑣) is surjec-
tive.
𝑢 𝑣
Let now 0 → M → − N→ − P be an exact sequence. Let 𝑣 0 : N → Im(𝑣)
be the morphism deduced from 𝑣 and let 𝑗 : Im(𝑣) → P be the inclusion,
𝑢 𝑣0
so that 𝑣 = 𝑗 ◦ 𝑣 0. Then 0 → M →
− N−
→ Im(𝑣) → 0 is a short exact
F(𝑢) F(𝑣 0)
sequence, so that its image by F, 0 → F(M) −−−→ F(N) −−−→ F(Im(𝑣)) → 0,
is an exact sequence too. In particular, Ker(F(𝑣 0)) = Im(F(𝑢)). On the
other hand, 𝑗 is injective, so that F(𝑗) is injective as well, and the relation
F(𝑣) = F(𝑗) ◦ F(𝑣 0) implies that Ker(F(𝑣 0)) = Ker(F(𝑣)). This proves that
F(𝑢) F(𝑣)
the complex 0 → F(M) −−−→ F(N) −−−→ F(P) is exact. In particular, F is
left exact.
𝑢 𝑣
We prove analogously that F is right exact. Let indeed M →
− N→
− P→0
be an exact sequence. Let 𝑢 0 : M/Ker(𝑢) → N be the morphism deduced
from 𝑢, and let 𝑝 : M → M/Ker(𝑢) be the canonical projection so that
𝑢0 𝑣
𝑢= 𝑢0 ◦ 𝑝. Then 0 → M/Ker(𝑢) −
→ N→
− P → 0 is an exact sequence,
F(𝑢 0) F(𝑣)
hence 0 → F(M/Ker(𝑢)) −−−→ F(N) −−−→ F(P) → 0 is exact as well,
352 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

hence that Ker(F(𝑣)) = Im(F(𝑢 0)). Since 𝑝 is surjective, F(𝑝) is surjective;

the relation F(𝑢) = F(𝑢 0) ◦ F(𝑝) then implies that Im(F(𝑢)) = Im(F(𝑢 0)).
F(𝑢) F(𝑣)
Consequently, F(M) −−−→ F(N) −−−→ F(P) → 0 is an exact sequence, as
was to be shown.
b) Assume that F is an exact functor. Let us prove that for every exact
𝑢 𝑣 F(𝑢) F(𝑣)
sequence M → − N→ − P, the complex F(M) −−−→ F(N) −−−→ F(P) is exact.
Let 𝑝 : M → M/Ker(𝑢) be the projection, let 𝑢 0 : M/Ker(𝑢) → N be
the unique morphism such that 𝑢 0 ◦ 𝑝 = 𝑢. Let 𝑗 : Im(𝑣) → P and let
𝑣 0 : N → Im(𝑣) be the morphism déduced from 𝑣 so that 𝑣 = 𝑗 ◦ 𝑣 0. The
exactness of the initial diagram furnishes an exact sequence
𝑢0 𝑣0
0 → M/Ker(𝑢) −
→N−
→ Im(𝑣) → 0.
Since F is exact, the diagram
F(𝑢 0) F(𝑣 0)
0 → F(M/Ker(𝑢)) −−−→ F(N) −−−→ F(Im(𝑣)) → 0
is thus an exact sequence, which means that the image of F(𝑢 0) is the
kernel of F(𝑣 0).
On the other hand, the relation 𝑢 0 ◦𝑝 = 𝑢 implies that F(𝑢 0)◦F(𝑝) = F(𝑢);
since F is rightjexact and 𝑝 is surjective, F(𝑝) is surjective as well, so
that Im(F(𝑢 0)) = Im(F(𝑢)). Similarly, the relation 𝑣 = 𝑗 ◦ 𝑣 0 implies that
F(𝑣) = F(𝑗) ◦ F(𝑣 0); since F is left exact and 𝑗 is injective, F(𝑗) is injective
and Ker(F(𝑣)) = Ker(F(𝑣 0)).
This proves that Ker(F(𝑣)) = Im(F(𝑢)), as claimed.

7.5.7. The Hom functors. — Let A be a ring, let Q be a fixed (A, B)-
bimodule. For every left A-module M, let us consider the abelian group
HomA (Q, M), endowed with its natural structure of a left B-module. (For
𝑢 ∈ HomA (Q, M) and 𝑏 ∈ B, let 𝑏𝑢 ∈ HomA (Q, M) be the map 𝑥 ↦→ 𝑢(𝑥𝑏).
For 𝑏, 𝑏 0 ∈ B and 𝑥 ∈ Q, 𝑏(𝑏 0𝑢) is the morphism given by 𝑥 ↦→ (𝑏 0𝑢)(𝑥𝑏) =
𝑢(𝑥𝑏𝑏 0), so that 𝑏(𝑏 0𝑢) = (𝑏𝑏 0)𝑢.) If 𝑓 : M → N is a morphism of left
A-modules, let us consider the map 𝑓∗ : HomA (Q, M) → HomA (Q, N)
given by 𝜑 ↦→ 𝑓 ◦ 𝜑. For 𝑓 : M → N and 𝑔 : N → P, one has (𝑔 ◦ 𝑓 )∗ (𝜑) =
𝑔 ◦ 𝑓 ◦ 𝜑 = 𝑔 ◦( 𝑓 ◦ 𝜑) = 𝑔∗ ( 𝑓∗ (𝜑)) = (𝑔∗ ◦ 𝑓∗ )(𝜑) for every 𝜑 ∈ HomA (Q, M).
Consequently, we have defined a covariant functor HomA (Q, •) from
the category of left A-modules to the category of left B-modules.
 7.5. EXACTNESS CONDITIONS FOR FUNCTORS 353

We can also consider the abelian group HomA (M, Q) endowed with
its natural structure of a right B-module defined by (𝑏𝑢)(𝑥) = (𝑏(𝑥))𝑢 for
𝑢 ∈ HomA (M, Q), 𝑏 ∈ B and 𝑥 ∈ M. For two left A-modules M and N, and
any morphism 𝑓 : M → N, let us consider the map 𝑓 ∗ : HomA (N, Q) →
HomA (M, Q) defined by 𝜑 ↦→ 𝜑 ◦ 𝑓 . For 𝑓 : M → N and 𝑔 : N → P, one
has (𝑔 ◦ 𝑓 )∗ (𝜑) = 𝜑◦ 𝑔 ◦ 𝑓 = (𝜑◦ 𝑔)◦ 𝑓 = 𝑓 ∗ (𝑔 ∗ (𝜑)), so that (𝑔 ◦ 𝑓 )∗ = 𝑓 ∗ ◦ 𝑔 ∗ .
We thus have defined a contravariant functor HomA (•, Q) from the
category of left A-modules to the category of right B-modules.

Proposition (7.5.8). — Let A, B be rings and Q be a (A, B)-bimodule.

a) The functor HomA (Q, •) is left exact. It is exact if and only if the
A-module Q is projective.
b) The functor HomA (•, Q) is left exact. It is exact if and only if the
A-module Q is injective.

Proof. — a) Let us consider an exact sequence of A-modules

𝑓 𝑔
0→M→
− N→
− P

and let
𝑓∗ 𝑔∗
0 → HomA (Q, M) −
→ HomA (Q, N) −
→ HomA (Q, P)

be the complex obtained by applying the functor HomA (Q, •). We need
to show that this complex is exact.
Exactness at HomA (Q, M). Let 𝜑 ∈ HomA (Q, M) be such that 𝑓∗ (𝜑) =
𝑓 ◦ 𝜑 = 0. This means that for every 𝑥 ∈ Q, 𝑓 (𝜑(𝑥)) = 0; since 𝑓 is
injective, 𝜑(𝑥) = 0. Consequently, 𝜑 = 0 and 𝑓∗ is injective.
Exactness at HomA (L, N). We want to show that Ker(𝑔∗ ) = Im( 𝑓∗ );
the inclusion Im( 𝑓∗ ) ⊂ Ker(𝑔∗ ) holds since we have a complex. So let
𝜑 ∈ HomA (Q, N) be such that 𝑔∗ (𝜑) = 0 and let us show that ther exists
𝜓 ∈ HomA (Q, M) such that 𝜑 = 𝑓∗ (𝜓). Let 𝑥 ∈ Q; then 𝑔(𝜑(𝑥)) = 0, hence
𝜑(𝑥) ∈ Ker(𝑔). Since Ker(𝑔) = Im( 𝑓 ) by assumption, Im(𝜑) ⊂ 𝑓 (M). In
other words, 𝜑 is really a morphism from Q to 𝑓 (M). Since 𝑓 is injective,
it induces an isomorphism from M to 𝑓 (M) and there exists a morphism
𝜓 : Q → M such that 𝜑 = 𝑓 ◦ 𝜓 = 𝑓∗ (𝜓), as desired.
354 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

Exactness. The functor HomA (Q, •) is right exact if and only if for every
exact sequence
𝑓 𝑔
0→M→
− N→
− P → 0,
the complex
𝑓∗ 𝑔∗
0 → HomA (Q, M) −
→ HomA (Q, N) −
→ HomA (Q, P) → 0

is exact. The exactness at HomA (Q, P) is the only missing point, that is,
the surjectivity of 𝑔∗ assuming that of 𝑔. Precisely, 𝑔∗ is surjective if and
only if, for every morphism 𝜑 : Q → P, there exists a morphism 𝜓 : Q →
N such that 𝜑 = 𝑔∗ (𝜓) = 𝑔 ◦ 𝜓. Since 𝑔 is surjective, proposition 7.3.2
asserts precisely that 𝑔∗ is surjective if and only if Q is a projective
A-module.
b) Let us now consider an exact sequence
𝑓 𝑔
M→
− N→
− P→0

and let
𝑔∗ 𝑓∗
0 → HomA (P, Q) −→ HomA (N, Q) −→ HomA (M, Q)

be the complex obtained by applying the contravariant func-

tor HomA (•, Q). We need to show that this complex is exact.
Exactness at HomA (P, Q). Let us show that 𝑔 ∗ is injective. Let 𝜑 ∈
HomA (P, Q) be such that 𝑔 ∗ (𝜑) = 𝜑 ◦ 𝑔 = 0. This means that Ker(𝜑)
contains 𝑔(N). Since 𝑔 is surjective, 𝑔(N) = P and 𝜑 = 0.
Exactness at HomA (N, Q). We have to show that Ker( 𝑓 ∗ ) = Im(𝑔 ∗ ). Let
𝜑 ∈ HomA (N, Q) be such that 𝜑 ◦ 𝑓 = 0 and let us show that there exists
𝜓 ∈ HomA (P, Q) such that 𝜑 = 𝑔 ∗ (𝜓) = 𝜓 ◦ 𝑔. Since 𝜑 ◦ 𝑓 = 0, Im( 𝑓 ) is
contained in Ker(𝜑). Passing to the quotient, we deduce a morphism
𝜓0 : N/Im( 𝑓 ) → Q such that 𝜑(𝑥) = 𝜓0 (cl(𝑥)) for every 𝑥 ∈ N, where
cl : N → N/Im( 𝑓 ) is the canonical projection. Since 𝑔 is surjective,
it induces an isomorphism 𝑔 0 from N/Ker(𝑔) to P. By assumption,
Ker(𝑔) = Im( 𝑓 ); let then 𝜓 : P → Q be the composition 𝜓 = 𝜓0 ◦ (𝑔 0)−1 By
construction, for every 𝑥 ∈ P, 𝑔 ∗ (𝜓)(𝑥) = 𝜓0 ◦ (𝑔 0)−1 ◦ 𝑔(𝑥) = 𝜓0 (cl(𝑥)) =
𝜑(𝑥). In other words, 𝑔 ∗ (𝜓) = 𝜑, as requested.
 7.5. EXACTNESS CONDITIONS FOR FUNCTORS 355

Exactness. Let us now consider an exact sequence

𝑓 𝑔
0→M→
− N→
− P→0
and let
𝑔∗ 𝑓∗
0 → HomA (P, Q) −→ HomA (N, Q) −→ HomA (M, Q) → 0
be the complex obtained by applying the contravariant func-
tor HomA (•, Q). We need to show that this complex is exact if
and only if Q is an injective A-module.
Only the exactness at HomA (M, Q) has not been shown, that is, assum-
ing the injectivity of 𝑓 , the surjectivity of 𝑓 ∗ . Surjectivity of 𝑓 ∗ means
that for every 𝜑 ∈ HomA (M, Q), there exists 𝜓 ∈ HomA (N, Q) such that
𝜓 ◦ 𝑔 = 𝜑. By proposition 7.4.2, this condition holds if and only if Q is
an injective A-module. 
In fact, the Hom functor can be used to detect exactness of complexes.

Lemma (7.5.9). — Let A be a ring. A complex

𝑓 𝑔
M→
− N→
− P→0
of A-modules is exact if and only if the complex
𝑔∗ 𝑓∗
0 → HomA (P, Q) −→ HomA (N, Q) −→ HomA (M, Q)
is exact for every A-module Q.

Proof. — One direction of the assertion is given by the left-exactness of

the functor Hom(•, Q). Conversely, let us assume that
𝑔∗ 𝑓∗
0 → HomA (P, Q) −→ HomA (N, Q) −→ HomA (M, Q)
is exact for every A-module Q and let us show that the diagram
𝑓 𝑔
M→
− N→
− P→0
is exact.
We first prove that 𝑔 is surjective. Let Q = Coker(𝑔) = P/𝑔(N) and let
𝜑 : P → Q be the canonical surjection. By construction, 𝑔 ∗ (𝜑) = 𝜑◦ 𝑔 = 0.
Since 𝑔 ∗ is injective, 𝜑 = 0. This means that Q = 0, hence 𝑔 is surjective.
356 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

By assumption, 𝑔 ◦ 𝑓 = 0, hence Im( 𝑓 ) ⊂ Ker(𝑔). Let Q = N/Im( 𝑓 )

and let 𝜑 : N → Q be the canonical surjection. By construction, 𝑓 ◦ 𝜑 =
0, hence 𝜑 belongs to the kernel of the map 𝑓 ∗ from HomA (N, Q)
to HomA (M, Q). By assumption, there exists 𝜓 ∈ HomA (P, Q) such that
𝜑 = 𝑔 ∗ (𝜓) = 𝜓 ◦ 𝑔. In particular, Ker(𝑔) is contained in Ker(𝜑) = Im( 𝑓 ),
as was to be shown. 

Remark (7.5.10). — It is tempting to state a dual version, that asserts that

a complex
𝑓 𝑔
0→M→
− N→
− P
of A-modules is exact if and only if the complex
𝑓∗ 𝑔∗
0 → HomA (Q, M) −
→ HomA (Q, N) −
→ HomA (Q, P)
is exact for every A-module Q. Of course, one direction of this statement
follows from the left-exactness of the functor HomA (Q, •). However, the
other is trivial, since setting Q = A recovers the initial complex, which is
thus an exact sequence.

7.5.11. Localization functors. — Let A be a commutative ring and let

S be a multiplicative subset of A. For any A-module M, we have defined
in section 3.6 an S−1 A-module S−1 M. Moreover, if 𝑓 : M → N is any
morphism of A-modules, we have constructed (cf. proposition 3.6.4)
a morphism S−1 𝑓 : S−1 M → S−1 N characterized by the property that
(S−1 𝑓 )(𝑚/𝑠) = 𝑓 (𝑚)/𝑠 for every 𝑚 ∈ M and every 𝑠 ∈ S. Moreover,
S−1 ( 𝑓 ◦ 𝑔) = S−1 𝑓 ◦ S−1 𝑔.
In other words, localization induces a functor from the category of A-
modules to that of S−1 A-modules. It is crucial that these localization
functors are exact; this is in fact a reformulation of proposition 3.6.6.

Proposition (7.5.12) (Exactness of localization). — Let A be a ring and let S

𝑓 𝑔
be a multiplicative subset of A. For any exact sequence 0 → M →
− N→
− P→0
S−1 𝑓 S−1 𝑔
of A-modules, the complex 0 → S−1 M
−−−→ S−1 N
−−−→ S−1 P → 0 is an
exact sequence. In other words, the localization functor from A-modules to
S−1 A-modules is an exact functor.
 7.6. ADJOINT FUNCTORS 357

Proof. — We can use the injective morphism 𝑓 to identify M with the

submodule 𝑓 (M) of N, and the surjective morphism 𝑔 to identify P with
the quotient N/ 𝑓 (M) of N. By proposition 3.6.6, the morphism S−1 𝑓 is
injective; moreover, the map S−1 𝑝 : S−1 N → S−1 P identifies the quotient
S−1 N/S−1 M with S−1 (N/M) = S−1 P. In other words, the morphism S−1 𝑔
is surjective and its kernel is the image of S−1 𝑓 . This means that the
diagram
S−1 𝑓 S−1 𝑔
0 → S M −−−→ S N −−−→ S−1 P → 0
−1 −1

is an exact sequence. 

7.6. Adjoint functors

Let A and B be rings. Let F be a functor from the category of A-modules
to the category of B-modules, and let G be a functor from the category
of B-modules to the category of A-modules.
Definition (7.6.1). — An adjunction for the pair (F, G) is the datum, for any
A-module M and any B-module N, of a bijection
ΦM,N : HomB (F(M), N) → HomA (M, G(N))
such that for any morphism 𝑓 : M → M0 of A-modules and any morphism
𝑔 : N → N0 of B-modules, one has
𝑓 ∗ ◦ G(𝑔)∗ ◦ ΦM0 ,N = ΦM,N0 ◦ F( 𝑓 )∗ ◦ 𝑔∗ .
Explicitly, this condition means that for every 𝜑 ∈ HomB (F(M0), N),
the two elements
G(𝑔) ◦ ΦM0 ,N (𝜑) ◦ 𝑓 and ΦM,N0 (𝑔 ◦ 𝜑 ◦ F( 𝑓 ))
of HomA (M, G(N0)) are equal. It is often visually represented by saying
that the diagram
ΦM0 ,N
HomB (F(M0), N) → HomA (M0 , G(N))
←
←

F( 𝑓 )∗ ◦𝑔∗ 𝑓 ∗ ◦G(𝑔)∗
→

ΦM,N0
HomB (F(M), N0) → HomA (M, G(N0))
←

is commutative.
358 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

If there exists such an adjunction, then one says that F and G form a pair of
adjoint functor, or, simply, are adjoint; one also says that G is a right adjoint
of F and that F is a left adjoint of G.
Lemma (7.6.2). — Assume that F and G are adjoint and let Φ = (ΦM,N ) be an
adjunction for the pair (F, G).
a) For any A-module M, let 𝛼M : M → G(F(M)) be the morphism
ΦM,F(M) (idF(M) ). For any B-module N and any morphism 𝑔 : F(M) → N, one
has ΦM,N (𝑔) = G(𝑔) ◦ 𝛼 M .
b) For any B-module N, let 𝛽 N : F(G(N)) → N be the unique morphism
such that ΦG(N),N (𝛽 N ) = idG(N) . Then, for any A-module M and any mor-
phism 𝑓 : M → G(N), one has Φ−1 M,N
( 𝑓 ) = 𝛽 N ◦ F( 𝑓 ).
Proof. — For a), one applies the definition with M0 = M, 𝑓 = idM , and
𝜑 = idF(M) . This gives
ΦM,N (𝑔) = ΦM,N (𝑔 ◦ idF(M) ◦F(idM ))
= G(𝑔) ◦ ΦM,F(M) (idF(M) ) ◦ idM
= G(𝑔) ◦ 𝛼 M .
For b), one considers M0 = G(N), N0 = N and 𝑔 idN . Then, one gets
ΦM,N (𝛽 N ◦ F( 𝑓 )) = ΦM,N0 (idN ◦𝛽 N ◦ F( 𝑓 ))
= G(idN ) ◦ ΦG(N),N (𝛽 N ) ◦ 𝑓
= idG(N) ◦ idG(N) ◦ 𝑓
= 𝑓. 

Example (7.6.3). — Let A and B be rings, let 𝑓 : B → A be a morphism

of rings and let F be the corresponding forgetful functor from right
A-modules to right B-modules. Let P be the ring A viewed as an (A, B)-
bimodule via the laws 𝑎 · 𝑥 · 𝑏 = 𝑎𝑥 𝑓 (𝑏) for any 𝑎 ∈ A, 𝑥 ∈ P and 𝑏 ∈ B.
Let then G be the functor that associates to any right B-module N the
right A-module HomB (P, N).
Let M be a right A-module, let N be an abelian group; let us define
a bijection ΦM,N from HomB (F(M), N) to HomA (M, G(N)). Let 𝜑 ∈
HomB (F(M), N); in other words, 𝜑 is an additive map from M to N such
that 𝜑(𝑚 𝑓 (𝑏)) = 𝜑(𝑚)𝑏 for any 𝑚 ∈ M and any 𝑏 ∈ B. For any 𝑚 ∈ M, let
 7.6. ADJOINT FUNCTORS 359

𝜓 𝑚 : P → N be the map given by 𝑎 ↦→ 𝜑(𝑚𝑎); it is additive; for any 𝑎 ∈ A

and any 𝑏 ∈ B, one has 𝜓 𝑚 (𝑎)𝑏 = 𝜑(𝑚𝑎)𝑏 = 𝜑(𝑚𝑎 𝑓 (𝑏)) = 𝜓 𝑚 (𝑎 𝑓 (𝑏)).
Consequently, 𝜓 𝑚 belongs to HomB (P, N) = G(N). The map 𝜓 : 𝑚 ↦→ 𝜓 𝑚
from M to G(N) is additive. It is even A-linear. Indeed, let 𝑎 ∈ A and let
𝑚 ∈ M; the map 𝜓 𝑚𝑎 sends 𝑎 0 ∈ A to 𝜑(𝑚𝑎𝑎 0); on the other hand, 𝜓 𝑚 · 𝑎
is the map 𝑎 0 ↦→ 𝜓 𝑚 (𝑎𝑎 0) = 𝜑(𝑚𝑎𝑎 0). Consequently, 𝜓 𝑚𝑎 = 𝜓 𝑚 · 𝑎, as was
to be shown. Let us then set ΦM,N (𝜑) = 𝜓.
The map ΦM,N : HomB (F(M), N) → HomA (M, G(N)) is bijective. In-
deed, if 𝜓 : M → HomB (P, N) is a morphism of A-modules, then the mor-
phism 𝜑 : F(M) → N given by 𝜑(𝑚) = 𝜓(𝑚)(1) satisfies ΦM,N (𝜑) = 𝜓,
since
ΦM,N (𝜑)(𝑚)(𝑎) = 𝜑(𝑚𝑎) = 𝜓(𝑚𝑎)(1) = 𝜓(𝑚)(𝑎),

and it is the only one.

We now check that the maps ΦM,N satisfy the condition given in the
definition of adjoint functors. Let 𝑓 : M → M0 be a morphism of right
A-modules, let 𝑔 : N → N0 be a morphism of right B-modules. Let
𝜑 ∈ HomA (F(M0), N) and let 𝑚 ∈ M. Then,

𝑓 ∗ ◦ G(𝑔)∗ ◦ ΦM0 ,N (𝜑)(𝑚) and ΦM,N0 ◦ F( 𝑓 )∗ ◦ 𝑔∗ (𝜑)(𝑚)

are both equal to the map 𝑎 ↦→ 𝑔(𝜑( 𝑓 (𝑚)𝑎)).

Consequently, F and G form a pair of adjoint functors.
Let us also explicit the maps 𝛼 M and 𝛽 N . Let M be an A-module,
so that G(F(M)) is the right A-module HomB (A, M). The morphism
𝛼M : M → G(F(M)) is defined as ΦM,F(M) (idF(M) ). Consequently, for any
𝑚 ∈ M, 𝛼M (𝑚) is the element 𝑎 ↦→ 𝑚𝑎 of HomB (A, M). In particular, the
morphism 𝛼 M is injective.
Let N be a B-module; F(G(N)) is the right B-module HomB (A, N).
Moreover, 𝛽 N : F(G(N)) → N maps an element 𝑢 ∈ HomB (A, N) to
𝑢(1) ∈ N. Indeed, let us write 𝛽0N for this morphism and let us compute
ΦG(N),N (𝛽0N ). With the previous notation, we set M = G(N), 𝜑 = 𝛽0N and
𝜓 = ΦG(N),N (𝛽0N ). For 𝑢 ∈ G(N), 𝜓𝑢 : P → N is the map 𝑎 ↦→ 𝛽0N (𝑢 · 𝑎).
By definition, 𝛽0N (𝑢 · 𝑎) = (𝑢 · 𝑎)(1) = 𝑢(𝑎), so that 𝜓𝑢 = 𝑢. Consequently,
𝜓 = idG(N) and 𝛽0N = 𝛽 N , as was to be shown. We observe in particular
that the morphism 𝛽 N is surjective.
360 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

In the preceding example, the functor G, being of the form HomB (P, •),
is left exact. The next proposition shows that this property is shared by
all functors which are right adjoints.

Proposition (7.6.4). — Assume that F and G are adjoint. Then F is right exact
and G is left exact.
𝑓 𝑔
Proof. — a) Let M →− N→ − P → 0 be an exact sequence of A-modules
and let us prove that the complex
F( 𝑓 ) F(𝑔)
F(M) −−−→ F(N) −−−→ F(P) → 0
is exact. By lemma 7.5.9, it suffices to show that for every B-module Q,
the complex
F(𝑔)∗ F( 𝑓 )∗
0 → HomA (F(P), Q) −−−→ HomA (F(N), Q) −−−→ HomA (F(M), Q)
is exact. Applying the adjunction property, this complex identifies with
the complex
𝑔∗ 𝑓∗
0 → HomA (P, G(Q)) −→ HomA (N, G(Q)) −→ HomA (M, G(Q))
which is exact, because the functor HomA (•, G(Q)) is left exact.
𝑓 𝑔
b) Let 0 → M →− N→− P be an exact sequence of B-modules and let us
prove that the complex
G( 𝑓 ) G(𝑔)
0 → G(M) −−−→ G(N) −−−→ G(P)
of A-modules is exact. Let Q be an A-module and let us apply the
functor HomA (Q, •) to the preceding complex: we get
G( 𝑓 )∗ G(𝑔)∗
0 → HomA (Q, G(M)) −−−−→ HomA (Q, G(N)) −−−−→ HomA (Q, G(P)).
Using the adjunction property of the functors F and G, we can rewrite
this complex as
𝑓∗ 𝑔∗
0 → HomA (F(Q), M) −
→ HomA (F(Q), N) −
→ HomA (F(Q), P),
which is an exact sequence since the functor HomA (F(Q), •) is left
exact. 
 7.6. ADJOINT FUNCTORS 361

Proposition (7.6.5). — Let A and B be rings. Let F be a functor from the

category of A-modules to the category of B-modules, let G be a functor from
the category of B-modules to the category of A-modules. Assume that F and G
form an adjoint pair.
a) If F is exact, then G(N) is an injective A-module for any injective B-
module N.
b) If G is exact, then F(M) is a projective B-module for any projective
A-module M.

Proof. — Let (ΦM,N ) be an adjunction for the pair (F, G). For any A-
module M, let 𝛼M = ΦM,F(M) (idF(M) ). For any B-module N, let 𝛽 N =
Φ−1
G(N),N
(idG(N) ).
a) Let M be an A-module and let 𝑓 : G(N) → M be an injective
morphism; let us show that 𝑓 has a left inverse. Let P = Coker( 𝑓 ) so that
we have an exact sequence
𝑓
0 → G(N) →
− M → P → 0.

Applying the exact functor F, we obtain an exact sequence

F( 𝑓 )
0 → F(G(N)) −−−→ F(M) → F(P) → 0.

In particular, F( 𝑓 ) is injective. Since N is an injective B-module and

F( 𝑓 ) is injective, proposition 7.4.2 implies that there exists a morphism
𝑣 : F(M) → N such that 𝑣 ◦ F( 𝑓 ) = 𝛽 N (apply assertion (ii) with 𝑖 = F( 𝑓 )
and 𝑓 = 𝛽 N ). Set 𝑔 = ΦM,N (𝑣). By the definition of adjoint functors, one
has

𝑔 ◦ 𝑓 = ΦM,N (𝑣) ◦ 𝑓 = ΦG(N),N (𝑣 ◦ F( 𝑓 )) = ΦG(N),N (𝛽 N ) = idG(N) .

This shows that 𝑔 is a left inverse of 𝑓 and concludes the proof that G(N)
is an injective A-module.
b) This is analogous. Let N be a B-module and let 𝑓 : N → F(M) be a
surjective morphism of B-modules. Let P = Ker( 𝑓 ), so that we have an
exact sequence of B-modules
𝑓
0→P→N→
− F(M) → 0.
362 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

Applying the exact functor G, we obtain an exact sequence

G( 𝑓 )
0 → G(P) → G(N) −−−→ G(F(M)) → 0

of A-modules; in particular, the morphism G(F( 𝑓 )) is surjective. Since M

is a projective A-module, there exists a morphism 𝑢 : M → G(N) such
that G( 𝑓 ) ◦ 𝑢 = 𝛼 M . Let 𝑔 = Φ−1
N,M
(𝑢). By definition of an adjoint pair,
one has

ΦM,F(M) ( 𝑓 ◦ 𝑔) = G( 𝑓 ) ◦ ΦN,M (𝑔) = G( 𝑓 ) ◦ 𝑢 = 𝛼M .

Consequently, 𝑓 ◦ 𝑔 = idF(M) . In particular, 𝑓 has a right inverse. This

shows that F(M) is a projective B-module. 

Remark (7.6.6). — As an application of the fact that right adjoint of exact

functors preserve injectives, let us give another (easier) proof that every
module can be embedded in an injective module.
We first treat the case of the ring Z. So let M be a Z-module. Write M
as the quotient of a free Z-module: let (𝑒 𝑖 )𝑖∈I be a generating family in M,
let 𝑓 : Z(I) → M be the unique morphism given by (𝑎 𝑖 )𝑖 ↦→ 𝑚 𝑖 𝑎 𝑖 and
Í
let N = Ker( 𝑓 ), so that M ' Z(I) /N. View Z(I) as a submodule of Q(I)
and define M0 = Q(I) /N. The injection Z(I) → Q(I) induces an injection
from M to M0. Since Q(I) is divisible, so is M0. By corollary 7.4.4, M0 is
an injective Z-module.
Let now A be a ring, let F be the forgetful functor from right A-modules
to Z-modules. By example 7.6.3, the functor G = HomZ (A, •) is a right
adjoint to F. Moreover, F is exact and G is left exact (proposition 7.5.8),
so that G preserves injective modules. Let M be a right A-module Let
𝑓 : F(M) → M0 be an injection from the Z-module F(M) to an injective
Z-module M0. Then, G(M0) is an injective A-module and the morphism
G( 𝑓 ) is injective.
Moreover, we have seen that in this case, the morphism 𝛼 M : M →
G(F(M)) is injective. The composition G( 𝑓 ) ◦ 𝛼M is thus an injective
morphism from M to an injective A-module.
 7.7. DIFFERENTIAL MODULES. HOMOLOGY AND COHOMOLOGY 363

7.7. Differential modules. Homology and cohomology

Definition (7.7.1). — Let A be a ring. A differential A-module is a pair
(M, 𝑑) where M is an A-module and 𝑑 is an endomorphism of M such that
𝑑2 = 0, called the differential of M.
Let (M, 𝑑M ) and (N, 𝑑N ) be differential A-modules. One says that a morphism
𝑓 : M → N of A-modules is a morphism of differential modules if 𝑑N ◦ 𝑓 =
𝑓 ◦ 𝑑M .
Definition (7.7.2). — To any differential A-module (M, 𝑑), one associates the
following A-modules:
– the module Z(M) = Ker 𝑑 of cycles;
– the module B(M) = Im 𝑑 of boundaries;
– the module H(M) = Z(M)/B(M) = Ker(𝑑)/Im(𝑑) of homologies.
Since 𝑑 2 = 0, observe that Im(𝑑) ⊂ Ker(𝑑), so that the definition of the
module of homologies makes sense.
Lemma (7.7.3). — Let 𝑓 : (M, 𝑑M ) → (N, 𝑑N ) be a morphism of differential
modules. One has 𝑓 (Z(M)) ⊂ Z(N) and 𝑓 (B(M)) ⊂ B(N). Consequently, 𝑓
induces a morphism H( 𝑓 ) : H(M) → H(N) of A-modules.
Proof. — By definition 𝑓 ◦ 𝑑M = 𝑑N ◦ 𝑓 . Consequently, if 𝑥 ∈ M is such
that 𝑑M (𝑥) = 0, one has 𝑑N ( 𝑓 (𝑥)) = 𝑓 (𝑑M (𝑥)) = 0, so that 𝑑M (𝑥) ∈ Z(N).
Similarly, if 𝑥 ∈ M, one has 𝑓 (𝑑M (𝑥)) = 𝑑N ( 𝑓 (𝑥)) so that 𝑓 (𝑑M (𝑥)) ∈ B(N)
and 𝑓 (B(M)) ⊂ B(N). The definition of H( 𝑓 ) follows by passing to the
quotient. 

Remark (7.7.4). — Morphisms of differential A-modules can be com-

posed, and the identity morphism is a morphism of differential A-
modules, so that differential A-modules form a category. We have associ-
ated to every differential A-module (M, 𝑑M ) its homology module H(M),
which is an A-module, and to every morphism 𝑓 : (M, 𝑑M ) → (N, 𝑑N ) of
differential A-modules a morphism H( 𝑓 ) : H(M) → H(N) of A-modules.
Observe that if 𝑓 = idM , then H( 𝑓 ) = idH(M) . Moreover, if
𝑓 : (M, 𝑑M ) → (N, 𝑑N ) and 𝑔 : (N, 𝑑N ) → (P, 𝑑P ) are morphisms of
differential modules, then H(𝑔 ◦ 𝑓 ) = H(𝑔) ◦ H( 𝑓 ), since both of these
morphisms map the class of 𝑥 ∈ Z(M) modulo B(M) to the class
364 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

of 𝑔( 𝑓 (𝑥)) ∈ Z(P) modulo B(P). Moreover, if 𝑓 , 𝑔 : (M, 𝑑M ) → (N, 𝑑N )

are morphisms of differential modules, one has H( 𝑓 + 𝑔) = H( 𝑓 ) + H(𝑔)
In other words, H is an additive functor from the category of differential
A-modules to the category of A-modules.

Many important differential A-modules are graded differential A-

modules.

Definition (7.7.5). — A graded differential A-module is a differential module

(M, 𝑑) such that
– The module M is the direct sum of a family (M𝑛 )𝑛∈Z (one says that M is
Z-graded, or simply, graded);
– There exists an integer 𝑟 such that 𝑑(M𝑛 ) ⊂ M𝑛+𝑟 for every 𝑛 ∈ Z (one
says that 𝑑 has degree 𝑟).
In this case, one defines Z𝑛 (M) = Z(M) ∩ M𝑛 , B𝑛 (M) = B(M) ∩ M𝑛 and
H𝑛 (M) = Z𝑛 (M)/B𝑛 (M).

Lemma (7.7.6). — The modules of cycles, boundaries and homologies of a graded

É if (M, 𝑑) is a graded É
differential A-module are graded. Explicitly, differential
A-module, one has equalities Z(M) =É 𝑛 Z𝑛 (M) and B(M) = B𝑛 (M);
they induce an isomorphism H(M) ' 𝑛∈Z H𝑛 (M).

Proof. — Observe that Z𝑛 (M) is a submodule of M𝑛 ; since the mod-

ules M𝑛 are in direct sum, so are the Z𝑛 (M). Moreover, Z𝑛 (M) ⊂ Z(M)
for every 𝑛. On the other hand, let 𝑥 ∈ Z(M); one can write 𝑥 as
a sum 𝑥 = 𝑛∈Z 𝑥 𝑛 of an almost-null family, where 𝑥 𝑛 ∈ M𝑛 for ev-
Í
ery 𝑛 ∈ Z. Then 𝑑(𝑥) = 𝑛∈Z 𝑑(𝑥 𝑛 ). For every 𝑛, 𝑑(𝑥 𝑛 ) ∈ M𝑛+𝑟 ; since
Í
the modules M𝑛 are in direct sum, 𝑑(𝑥É𝑛 ) = 0 for every 𝑛. Consequently,
𝑥 𝑛 ∈ Z(M𝑛 ) for every 𝑛 and Z(M) = 𝑛 Z𝑛 (M).
The modules B𝑛 (M) are in direct sum, and are submodules of B(M).
Conversely, let 𝑥 ∈ B(M); let 𝑦 ∈ M be such that 𝑥 = 𝑑(𝑦). One may write
𝑦 = 𝑛 𝑦𝑛 , where 𝑦𝑛 ∈ M𝑛 for every 𝑛. It follows that 𝑥 = 𝑛 𝑑(𝑦𝑛 ). For
Í Í
every 𝑛, 𝑑(𝑦𝑛 ) ∈ B𝑛+𝑟 , hence 𝑥 ∈ 𝑛 B(M𝑛 ).
Í
 7.7. DIFFERENTIAL MODULES. HOMOLOGY AND COHOMOLOGY 365

Finally, one has isomorphisms

! !
Ê Ê
H(M) = Z(M)/B(M) = Z𝑛 (M) / B𝑛 (M)
𝑛 𝑛
Ê Ê
= (Z𝑛 (M)/B𝑛 (M)) = H𝑛 (M).
𝑛 𝑛

This proves the lemma. 

Example (7.7.7). — Let us consider a complex

𝑓𝑛−1 𝑓𝑛 𝑓𝑛+1
. . . −−−→ M𝑛−1 −→ M𝑛 −−−→ M𝑛+1 → . . .

𝑛∈Z M𝑛 and let 𝑑 ∈ End(M) be the

É
of A-modules. Let us define M =
unique endomorphism of M such that 𝑑(𝑥) = 𝑓𝑛 (𝑥) for every 𝑥 ∈ M𝑛
and every 𝑛 ∈ Z.
Then (M, 𝑑) is a graded differential A-module whose differential
has degree 1. Moreover, H𝑛 (M) = Ker( 𝑓𝑛+1 )/Im( 𝑓𝑛 ). In other words,
the vanishing of H𝑛 (M) witnesses the exactness of the complex at the
module M𝑛 . Tradition denotes H𝑛 (M) these modules and calls them
cohomology modules of the complex.

Example (7.7.8) (De Rham complex of a manifold)

Let U be an open subset of R2 . For 𝑝 ∈ {0, 1, 2}, let Ω𝑝 (U) be the
R-vector space of differential forms of degree 𝑝 on U. These spaces
can be described explicitly as follows. For 𝑝 = 0, Ω0 (U) = 𝒞 ∞ (U) is
the real vector space of 𝒞 ∞ -functions on U. For 𝑝 = 1, Ω1 (U) is a
free 𝒞 ∞ (U)-modules of rank 2, with basis (d𝑥, d𝑦). In other words, a
differential form 𝜔 of degree 1 on U can be uniquely written as

𝜔 = A(𝑥, 𝑦) d𝑥 + B(𝑥, 𝑦) d𝑦
366 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

Finally, Ω2 (U) is a free 𝒞 ∞ (U)-module of rank 1, with basis denoted

d𝑥 ∧ d𝑦. The “exterior differential” consists of the maps
𝑑 : Ω0 (U) → Ω1 (U)
𝜕𝑓 𝜕𝑓
𝑓 ↦→ d𝑥 + d𝑦
𝜕𝑥 𝜕𝑦
𝑑 : Ω1 (U) → Ω2 (U)
𝜕B(𝑥, 𝑦) 𝜕A(𝑥, 𝑦)
 
A(𝑥, 𝑦) d𝑥 + B(𝑥, 𝑦) d𝑦 ↦→ − d𝑥 ∧ d𝑦
𝜕𝑥 𝜕𝑦
and 𝑑 vanishes identically on Ω2 (U).
Observe that 𝑑 ◦ 𝑑 = 0. The only necessary computation is that of 𝑑 2 ( 𝑓 )
for 𝑓 ∈ Ω0 (U). Then,
𝜕𝑓 𝜕𝑓 𝜕2 𝑓 𝜕2 𝑓
   
𝑑 (𝑓) = 𝑑
2
d𝑥 + d𝑦 = − d𝑥 ∧ d𝑦
𝜕𝑥 𝜕𝑦 𝜕𝑥𝜕𝑦 𝜕𝑦𝜕𝑥
hence 𝑑 2 ( 𝑓 ) = 0 by Schwarz’s theorem.
Consequently, we have defined a complex Ω• :
𝑑 𝑑
0 → Ω0 (U) →
− Ω1 (U) →
− Ω2 (U) → 0,
𝑖
the de Rham complex of U. Its cohomology groups HdR (U) are called the
de Rham cohomology groups of U. These real vector spaces of fundamental
interest for topology.
Let us compute H0dR (U). Let 𝑓 ∈ Z0 (Ω• ); this means that 𝑓 ∈ 𝒞 ∞ (U)
𝜕𝑓 𝜕𝑓
and that 𝑑𝑓 = 0, that is, 𝜕𝑥
= 𝜕𝑦
= 0. Consequently, 𝑓 is constant on every
connected component of U. Since B0 (Ω• ) = 0, we obtain an isomorphism
H0dR (U) = R𝜋0 (U) , where 𝜋0 (U) is the set of connected components of U.
Let us assume that U is simply connected; for example, U could be R2 ,
or contractible, or star-shaped. Then, Poincaré’s lemma asserts that any
differential form 𝜔 on U of degree > 0 which is a cycle (one says that 𝜔
is closed) is a boundary (one says that 𝜔 is exact). In Physics or Vector
calculus, this lemma appears under a more elementary formulation: a
vector field whose rotational is zero is a gradient.
Let us prove it when U is star-shaped with respect to the origin 0
of R2 . Let 𝜔 = A(𝑥, 𝑦) d𝑥 + B(𝑥, 𝑦) d𝑦 be any closed differential form
of degree 1. Since U is star-shaped with respect to the origin, for every
 7.7. DIFFERENTIAL MODULES. HOMOLOGY AND COHOMOLOGY 367

point (𝑥, 𝑦) ∈ U, and every 𝑡 ∈ [0, 1], one has (𝑡𝑥, 𝑡 𝑦) ∈ U. We may thus
set
∫ 1
𝑓 (𝑥, 𝑦) = 𝑥A(𝑡𝑥, 𝑡 𝑦) + 𝑦B(𝑡𝑥, 𝑡 𝑦) d𝑡.


0
Then 𝑓 is 𝒞∞(see a course in Calculus) and its partial derivatives can be
computed by differentiating under the integral-sign. Thus one obtains
𝜕𝑓 𝜕A 𝜕B
∫ 1  
(𝑥, 𝑦) = A(𝑡𝑥, 𝑡 𝑦) + 𝑡𝑥 (𝑡𝑥, 𝑡 𝑦) + 𝑡 𝑦 (𝑡𝑥, 𝑡 𝑦) d𝑡.
𝜕𝑥 0 𝜕𝑥 𝜕𝑥
𝜕B 𝜕A
Since 𝑑𝜔 = 0, 𝜕𝑥
= 𝜕𝑦
so that

𝜕𝑓 𝜕A 𝜕B
∫ 1  
(𝑥, 𝑦) = A(𝑡𝑥, 𝑡 𝑦) + 𝑡𝑥 (𝑡𝑥, 𝑡 𝑦) + 𝑡 𝑦 (𝑡𝑥, 𝑡 𝑦) d𝑡
𝜕𝑥 0 𝜕𝑦 𝜕𝑥
𝑑
∫ 1  
= A(𝑡𝑥, 𝑡 𝑦) + 𝑡 (A(𝑡𝑥, 𝑡 𝑦)) d𝑡
0 𝑑𝑡
𝑑
∫ 1  1
= (𝑡A(𝑡𝑥, 𝑡 𝑦)) = 𝑡A(𝑡𝑥, 𝑡 𝑦)
0 𝑑𝑡 0
= A(𝑥, 𝑦).
𝜕𝑓
One proves similarly that 𝜕𝑦
(𝑥, 𝑦) = B(𝑥, 𝑦), so that 𝑑𝑓 = 𝜔.
On the other hand, if {0}, one can prove that H1dR (U) has
U = R2
dimension 1 and is generated by the class of the differential form
𝑦 𝑥
− 𝑑𝑥 + 𝑑𝑦.
𝑥2 + 𝑦2 𝑥2 + 𝑦2
This construction can be generalized to any open subset of R𝑛 , and
even to any differentiable manifold.

Theorem (7.7.9). — Let A be a ring, let M, N, P be differential modules, let

𝑓 : M → N and 𝑔 : N → P be morphisms of differential modules. Assume
𝑓 𝑔
that one has an exact sequence 0 → M →
− N→ − P → 0. (These conditions can
be summed up by saying that we have an exact sequence of differential modules.)
Then, there exists a morphism of A-modules 𝜕 : H(P) → H(M) such that

Ker(𝜕) = Im(H(𝑔)), Ker(H(𝑔)) = Im(H( 𝑓 )), Ker(H( 𝑓 )) = Im(𝜕).

368 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

Consequently, the morphism 𝜕 sits in an “exact triangle” :

H(P)
𝜕 ← → H(𝑔)
→ ←
H( 𝑓 )
→ H(N)
←
H(M)

Proof. — a) Let us show that Ker(H(𝑔)) = Im(H( 𝑓 )). Since 𝑔 ◦ 𝑓 = 0, one

has H(𝑔) ◦ H( 𝑓 ) = H(𝑔 ◦ 𝑓 ) = 0 and Im(H( 𝑓 )) ⊂ Ker(H(𝑔)). Conversely,
let 𝜉 ∈ Ker(H(𝑔)). By definition, 𝜉 is the class of an element 𝑥 ∈ Ker(𝑑N ).
Then, H(𝑔)(𝜉) = cl(𝑔(𝑥)) = 0, so that there exists 𝑦 ∈ P such that
𝑔(𝑥) = 𝑑P (𝑦). Since the morphism 𝑔 is surjective, there exists 𝑧 ∈ N such
that 𝑦 = 𝑔(𝑧). Then 𝑔(𝑥) = 𝑑P (𝑦) = 𝑑P (𝑔(𝑧)) = 𝑔(𝑑N (𝑧)) hence 𝑥 − 𝑑N (𝑧)
belongs to Ker(𝑔). Since Ker(𝑔) = Im( 𝑓 ), there exists 𝑡 ∈ M such that
𝑥 − 𝑑N (𝑧) = 𝑓 (𝑡). It follows that 𝜉 = cl(𝑥) = cl(𝑑N (𝑧) + 𝑓 (𝑡)) = cl( 𝑓 (𝑡)) =
H( 𝑓 )(cl(𝑡)). Consequently, Ker(H(𝑔)) ⊂ Im(H( 𝑓 )), as was to be shown.
b) Let us now define a morphism 𝜕 : H(P) → H(M). Let 𝜉 ∈ H(P).
Write 𝜉 = cl(𝑥) for some 𝑥 ∈ Ker 𝑑P . Since 𝑔 is surjective, there exists
𝑦 ∈ N such that 𝑥 = 𝑔(𝑦). Then 0 = 𝑑P (𝑥) = 𝑑P (𝑔(𝑦)) = 𝑔(𝑑N (𝑦))
hence 𝑑N (𝑦) ∈ Ker(𝑔). Since Ker(𝑔) = Im( 𝑓 ), there exists 𝑧 ∈ M such
that 𝑑N (𝑦) = 𝑓 (𝑧). One has 𝑓 (𝑑M (𝑧)) = 𝑑N ( 𝑓 (𝑧)) = 𝑑N ◦ 𝑑N (𝑦) = 0
since 𝑑N 2
= 0. Since 𝑓 is injective, 𝑑M (𝑧) = 0. We are going to set
𝜕(𝜉) = cl(𝑧) ∈ H(M).
Let us first check that cl(𝑧) is independent of any choice. Recall
that 𝑥, 𝑦, 𝑧 have been chosen in such a way that one has 𝑓 (𝑧) = 𝑑N (𝑦)
and 𝑥 = 𝑔(𝑦). Let 𝑥 0 , 𝑦 0 , 𝑧 0 be such that 𝑓 (𝑧 0) = 𝑑N (𝑦 0) and 𝑥 0 = 𝑔(𝑦 0).
Assume that cl(𝑥) = cl(𝑥 0); then there exists 𝑥 00 ∈ P such that 𝑥 =
𝑥 0 + 𝑑P (𝑥 00). Let us also choose 𝑦 00 ∈ N such that 𝑥 00 = 𝑔(𝑦 00). One has
𝑔(𝑦 0 − 𝑦) = 𝑥 0 − 𝑥 = 𝑑P (𝑥 00) = 𝑑P (𝑔(𝑦 00)) = 𝑔(𝑑N (𝑦 00)), hence there exists
𝑧 00 ∈ M such that 𝑦 0 − 𝑦 − 𝑑N (𝑦 00) = 𝑓 (𝑧 00). Consequently, 𝑓 (𝑧 0 − 𝑧) =
𝑑N (𝑦 0) − 𝑑N (𝑦) = 𝑑N (𝑑N (𝑦 00) + 𝑓 (𝑧 00)) = 𝑑N ( 𝑓 (𝑧 00)) = 𝑓 (𝑑M (𝑧 00)). Since 𝑓
is injective, 𝑧 0 − 𝑧 = 𝑑M (𝑧 00) and cl(𝑧 0) = cl(𝑧) in H(M).
Moreover, 𝜕 is a morphism of A-modules. Indeed, having chosen the
triple (𝑥 1 , 𝑦1 , 𝑧1 ) for 𝜉1 and the triple (𝑥 2 , 𝑦2 , 𝑧2 ) for 𝜉2 , we may choose
the triple (𝑥 1 𝑎1 + 𝑥2 𝑎2 , 𝑦1 𝑎1 + 𝑦2 𝑎 2 , 𝑧1 𝑎1 + 𝑧2 𝑎2 ) for 𝜉1 𝑎1 + 𝜉2 𝑎 2 . Then
𝜕(𝜉1 𝑎1 + 𝜉2 𝑎 2 ) = 𝜕(𝜉1 )𝑎1 + 𝜕(𝜉2 )𝑎2 .
 7.7. DIFFERENTIAL MODULES. HOMOLOGY AND COHOMOLOGY 369

c) Let us show that Ker(H( 𝑓 )) = Im(𝜕). Let 𝜉 ∈ H(M) be such

that H( 𝑓 )(𝜉) = 0 and let 𝑥 ∈ Z(M) be such that 𝜉 = cl(𝑥). Then
H( 𝑓 )(𝜉) = cl( 𝑓 (𝑥)), hence 𝑓 (𝑥) ∈ Im(𝑑N ) and there exists 𝑦 ∈ N such that
𝑓 (𝑥) = 𝑑N (𝑦). By definition of the morphism 𝜕, one has 𝜕(cl(𝑔(𝑦))) =
cl(𝑥) = 𝜉, hence Ker(H( 𝑓 )) ⊂ Im(𝜕).
Conversely, let 𝜂 ∈ Im(𝜕), let 𝜉 ∈ H(P) such that 𝜂 = 𝜕(𝜉) and let 𝑥 ∈
Z(P) be such that 𝜉 = cl(𝑥). Let 𝑦 ∈ N be such that 𝑥 = 𝑔(𝑦) and let 𝑧 ∈ M
be such that 𝑑N (𝑦) = 𝑓 (𝑧); then 𝜂 = 𝜕(𝜉) = cl(𝑧), by the construction of
the morphism 𝜕 in b). We then have H( 𝑓 )(𝜂) = cl( 𝑓 (𝑧)) = cl(𝑑N (𝑦)) = 0,
so that 𝜂 ∈ Ker(H( 𝑓 )), as was to be shown.

d) Let us show that Im(H(𝑔)) = Ker(𝜕).

Let 𝜉 ∈ H(P) be such that 𝜕(𝜉) = 0; let 𝑥 ∈ Z(P) be such that 𝜉 = cl(𝑥),
let 𝑦 ∈ N be such that 𝑥 = 𝑔(𝑦) and let 𝑧 ∈ M be such that 𝑑N (𝑦) = 𝑓 (𝑧),
so that 𝜕(𝜉) = cl(𝑧). By assumption, 𝑧 ∈ Im(𝑑M ); let then 𝑧 0 ∈ M be
such that 𝑧 = 𝑑M (𝑧 0). We have 𝑓 (𝑧) = 𝑓 (𝑑M (𝑧 0)) = 𝑑N ( 𝑓 (𝑧 0)) = 𝑑N
2
(𝑦) = 0.
Since 𝑓 is injective, 𝑧 = 0 and 𝑑N (𝑦) = 𝑓 (𝑧) = 0. Moreover, the class of 𝑦
in H(N) satisfies H(𝑔)(cl(𝑦)) = cl(𝑔(𝑦)) = cl(𝑥) = 𝜉, which proves that
𝜉 ∈ Im H(𝑔).
Conversely, let 𝜉 ∈ Im(H(𝑔)); there exists 𝑦 ∈ Z(N) such that 𝜉 =
H(𝑔)(cl(𝑦)), that is, 𝜉 = cl(𝑔(𝑦)). Set 𝑥 = 𝑔(𝑦); one has 𝑥 ∈ Z(P); let then
𝑧 ∈ M be such that 𝑑N (𝑦) = 𝑓 (𝑧). Then 𝑓 (𝑧) = 0, hence 𝑧 = 0 since 𝑓 is
injective, and the definition of 𝜕 implies that 𝜕(𝜉) = cl(𝑧) = 0. 

Corollary (7.7.10). — Let us assume moreover that M, N, P are graded

differential modules and that the morphisms 𝑓 and 𝑔 are of degree 0 (meaning
𝑓 (M𝑛 ) ⊂ N𝑛 and 𝑓 (N𝑛 ) ⊂ P𝑛 for every 𝑛). Then, the morphism 𝜕 constructed
in theorem 7.7.9 has degree 1 and its restriction 𝜕𝑛 to H𝑛 (P) gives rise to an
exact sequence

𝜕𝑛
· · · → H𝑛 (M) → H𝑛 (N) → H𝑛 (P) −→ H𝑛+1 (M) → . . .
370 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

The conditions of the corollary are often used by saying that one has
an exact sequence of complexes, namely a diagram
0 0 0

←
→

→
→ M𝑛−1 → M𝑛 → M𝑛+1 →
← ← ← ←

←
→

→
→ N𝑛−1 → N𝑛 → N𝑛+1 →
← ← ← ←
←

←
→

→
→ P𝑛−1 → P𝑛 → P𝑛+1 →
← ← ← ←
←

←
→

→
0 0 0
where the rows are complexes and the columns are exact sequences.
Proof. — Recall that for any 𝜉 ∈ H(P), 𝜕(𝜉) has been defined as cl(𝑧),
where 𝑥 ∈ Z(P), 𝑦 ∈ N and 𝑧 ∈ M are such that 𝜉 = cl(𝑥), 𝑥 = 𝑔(𝑦)
and 𝑑N (𝑦) = 𝑓 (𝑧). Assume that 𝜉 ∈ H𝑛 (P). Then, one may replace
𝑥, 𝑦, 𝑧 by their components in P𝑛 , N𝑛 and M𝑛+1 respectively, so that
𝜕(𝜉) ∈ H𝑛+1 (M). The rest of the corollary follows from lemma 7.7.11. 

Lemma (7.7.11). — Let I be a set, let (M𝑖 )𝑖∈I and (N𝑖 )𝑖∈I be families of A-
Éfor every 𝑖, let
module; É𝑓𝑖 : M𝑖 → N𝑖 be a morphism of A-modules. Set
M = 𝑖∈I M 𝑖 , N = 𝑖∈I N 𝑖 and let 𝑓 : M → N be the morphism of A-
modules given by 𝑓 ((𝑚 𝑖 )) = ( 𝑓𝑖 (𝑚 𝑖 )) for (𝑚 𝑖 ) ∈ M. Then for every 𝑖 ∈ I, one
has Ker( 𝑓𝑖 ) = Ker( 𝑓 ) ∩ M𝑖 , Im( 𝑓𝑖 ) = Im( 𝑓 ) ∩ N𝑖 and
Ê Ê
Ker( 𝑓 ) = Ker( 𝑓𝑖 ) and Im( 𝑓 ) = Im( 𝑓𝑖 ).
𝑖∈I 𝑖

Proof. — For 𝑚 = (𝑚 𝑖 ) ∈ M, one has 𝑓 (𝑚) = ( 𝑓𝑖 (𝑚 𝑖 )) = 0, so 𝑚 ∈ Ker( 𝑓 ) if

and only if 𝑚 𝑖 ∈ Ker( 𝑓𝑖 ) for every 𝑖. This shows that Ker( 𝑓 )∩M𝑖É= Ker( 𝑓𝑖 ).
Moreover, for 𝑚 ∈ Ker( 𝑓 ), one has 𝑚 = 𝑚 𝑖 , hence Ker( 𝑓 ) = Ker( 𝑓𝑖 ).
Í
Let also 𝑛 = (𝑛 𝑖 ) ∈ N. If 𝑛 = 𝑓 (𝑚), for 𝑚 ∈ M, then 𝑓 (𝑚 𝑖 ) = 𝑛 𝑖
for every 𝑖, so that 𝑛 𝑖 ∈ Im( 𝑓𝑖 ). Conversely, assume that 𝑛 𝑖 ∈ Im( 𝑓𝑖 )
for every 𝑖; let J be the finite set of 𝑖 ∈ I such that 𝑛 𝑖 ≠ 0. For 𝑖 ∈ J,
choose 𝑚 𝑖 ∈ M𝑖 such that 𝑛 𝑖 = 𝑓𝑖 (𝑚 𝑖 ); for 𝑖 ∈ I J, set 𝑚 𝑖 = 0. Then
 EXERCISES 371

𝑖 )) = 𝑛, hence 𝑛 ∈ Im( 𝑓 ). The relation 𝑛 =

𝑓 ((𝑚É 𝑛 𝑖 also shows that
Í
𝑛∈ Im( 𝑓𝑖 ). 

Exercises
1-exo800) Let A be a ring, let M1 , M2 , M3 , M4 be A-modules and let 𝑤 : M1 → M2 ,
𝑣 : M2 → M3 and 𝑢 : M3 → M4 be morphisms. If 𝑓 is a morphism of A-modules, we
denote by 𝜌( 𝑓 ) the length of Im( 𝑓 ), or +∞ if this length is infinite.
a) Deduce from 𝑤 an A-morphism
𝑓 : Ker(𝑢 ◦ 𝑣 ◦ 𝑤)/Ker(𝑢 ◦ 𝑣) → Ker(𝑢 ◦ 𝑣)/Ker(𝑣).
b) Deduce from 𝑣 an A-morphism
𝑔 : Ker(𝑢 ◦ 𝑣)/Ker(𝑣) → Im(𝑣)/Im(𝑣 ◦ 𝑤).
c) Deduce from 𝑢 an A-morphism
ℎ : Im(𝑣)/Im(𝑣 ◦ 𝑤) → Im(𝑢 ◦ 𝑣)/Im(𝑢 ◦ 𝑣 ◦ 𝑤).
d) Show that the diagram
𝑓
→ Ker(𝑢 ◦ 𝑣 ◦ 𝑤)/Ker(𝑢 ◦ 𝑣) → Ker(𝑢 ◦ 𝑣)/Ker(𝑣)
← ←
0
𝑔 ℎ
→ Im(𝑣)/Im(𝑣 ◦ 𝑤) → Im(𝑢 ◦ 𝑣)/Im(𝑢 ◦ 𝑣 ◦ 𝑤) → 0
← ← ←

is an exact sequence.
e) Prove the inequality (due to Frobenius):
𝜌(𝑣 ◦ 𝑤) + 𝜌(𝑢 ◦ 𝑣) 6 𝜌(𝑣) + 𝜌(𝑢 ◦ 𝑣 ◦ 𝑤).

2-exo1035) Let A be a ring.

𝑓0 𝑓1 𝑓𝑛−1
a) Let M0 − → . . . −−−→ M𝑛 be a complex of A-modules. Show that this complex
→ M1 −
𝑓𝑖
is an exact sequence if and only if, for every 𝑖, the diagram (0) → Ker( 𝑓𝑖 ) → M𝑖 −
→
Ker( 𝑓𝑖+1 ) → (0) is a short exact sequence.
𝑓1 𝑓2 𝑓𝑛−1
b) Let 0 → M1 −
→ M2 − → . . . −−−→ M𝑛 → 0 be an exact sequence. Assuming that the
modules M𝑖 have finite length, show that 𝑛𝑖=1 (−1)𝑖 ℓA (M𝑖 ) = 0.
Í
𝑓1 𝑓2 𝑓𝑛−1
c) More generally, let 0 → M1 − → M2 − → . . . −−−→ M𝑛 → 0 be a complex. Assume
that the modules M𝑖 have finite length. For every 𝑖, let H𝑖 = Ker( 𝑓𝑖 )/Im( 𝑓𝑖−1 ); prove
that H1 , . . . , H𝑛 have finite length and that
𝑛
Õ 𝑛
Õ
(−1)𝑖 ℓA (M𝑖 ) = (−1)𝑖 ℓA (H𝑖 ).
𝑖=1 𝑖=1

3-exo1207) Let A be a commutative ring, let M be an A-module and let (𝑎, 𝑏) be two
elements of A.
372 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

a) Show that the map 𝑑1 : M → M × M and 𝑑2 : M × M → M defined by

𝑑1 (𝑥) = (𝑎𝑥, 𝑏𝑥) et 𝑑2 (𝑥, 𝑦) = 𝑏𝑥 − 𝑎𝑦
give rise to a complex M• :
𝑑1 𝑑2
0 → M −→ M × M −→ 0.
b) Show that H0 (M• ) = {𝑥 ∈ M ; 𝑎𝑥 = 𝑏𝑦 = 0} and that H2 (M• ) = M/(𝑎M + 𝑏M).
c) Assume that the multiplication by 𝑎 is injective in M. Then show that multiplication
by 𝑏 in M/𝑎M is injective if and only if H1 (M• ) = 0.
4-exo1210) Let A be a commutative ring and let I be an ideal of A.
a) Assume that the canonical morphism cl : A → A/I admits a right inverse 𝑓 . Show
that there exists 𝑎 ∈ I such that 𝑎 = 𝑎 2 and I = (𝑎).
b) Assume that A is an integral domain. Then show that A/I is a projective A-module
if and only if I = 0 or I = A.
5-exo257) Let 𝑘 be a field, let A be the subring 𝑘[X2 , XY, Y2 ] of 𝑘[X, Y]. Show that the
ideal I = (X2 , XY) of A is not a projective A-module.
6-exo253) Let A = 𝒞([0, 1], R) be the ring of continuous functions on [0, 1] with real
values, let I be the set of functions 𝑓 ∈ A which are identically 0 in some neighborhood
of 0.
a) Show that I is not finitely generated.
b) Show that I = I2 .
c) Show that there exists a family (𝜃𝑛 )𝑛>1 of continuous functions on [0, 1] such
that 𝜃𝑛 (𝑥) = 0 if 𝑥 ∉ [ 𝑛1 , 𝑛2 ], and that 𝑛>1 𝜃𝑛 (𝑥) > 0 for every 𝑥 ∈ (0, 1]. (For example,
Í
set 𝜃𝑛 (𝑥) = (𝑛𝑥 − 1)(2 − 𝑛𝑥) for 𝑥 ∈ [ 𝑛1 , 𝑛2 ].)
d) Show that the ideal I is a projective A-module.
7-exo254) Let 𝑘 be a ring, let A = M𝑛 (𝑘) be the ring of matrices of size 𝑛 with entries
in 𝑘.
a) Let M = 𝑘 𝑛 , viewed as a left A-module. Show that A ' M𝑛 .
b) Assume that 𝑘 is commutative or that it is a division ring. Show that M is not free.
8-exo255) Let A be a ring, let P be a finitely generated projective A-module. Show that
EndA (P) is again a finitely generated and projective A-module.
𝑓 𝑔
9-exo1208) Let A be a ring and let 0 → M →
− N→
− P → 0 be an exact sequence of
A-modules.
a) Assume that this exact sequence is split. Then show that for every A-module X,
the diagrams
𝑓∗ 𝑔∗
0 → Hom(X, M) −
→ Hom(X, N) −
→ Hom(X, P) → 0
and
𝑔∗ 𝑓∗
0 → Hom(P, X) −→ Hom(N, X) −→ Hom(M, X) → 0
 EXERCISES 373

are split exact sequences.

b) Conversely, assume that for every A-module X, the diagram
𝑓∗ 𝑔∗
0 → Hom(X, M) −
→ Hom(X, N) −
→ Hom(X, P) → 0
is an exact sequence. Show that the initial exact sequence is split.
c) Solve the same question for the second sort of exact sequences.
10-exo882) a) Prove that Z/3Z is a projective (Z/6Z)-module which is not free.
b) Let A be a commutative ring. Let 𝑒 ∈ A {0, 1} be a non-trivial idempotent
(𝑒 2= 𝑒). Prove that 𝑒A is a projective submodule of A which is not free.
11-exo852) Let A be a ring and let
𝑓
→ N
←
M
←

←
𝑢 𝑣
→

→
𝑓0
M0 → N0
←

be a commutative diagram of A-modules.

a) Let 𝜓 : M → M0 ⊕ N and 𝜑 : M0 ⊕ N → N0 be the maps given by by 𝜓(𝑥) =
(𝑢(𝑥), 𝑓 (𝑦)) and 𝜑(𝑥, 𝑦) = 𝑓 0(𝑥) − 𝑣(𝑦). Prove that 𝜑 ◦ 𝜓 = 0.
b) Assume that 𝜓 is injective and Im(𝜓) = Ker(𝜑). (One says that this diagram
is a pull-back.) Prove that for every A-module P, every morphism 𝑝 : P → N and
every morphism 𝑞 : P → M0 such that 𝑓 0 ◦ 𝑞 = 𝑣 ◦ 𝑝, there exists a unique morphism
𝑟 : P → M such that 𝑝 = 𝑓 ◦ 𝑟 and 𝑞 = 𝑢 ◦ 𝑟.
c) Assume that 𝜑 is surjective and Im(𝜓) = Ker(𝜑). (One says that this diagram
is a push-out.) Prove that for every A-module Q, every morphism 𝑝 : N → Q and
every morphism 𝑞 : M0 → Q such that 𝑝 ◦ 𝑓 = 𝑞 ◦ 𝑢, there exists a unique morphism
𝑟 : N0 → Q such that 𝑝 = 𝑟 ◦ 𝑣 and 𝑞 = 𝑟 ◦ 𝑓 0.
d) Let A be a ring and let I, J be two-sided ideals of A. Prove that the diagram

A/(I ∩ J) → A/I
←
←

←
→

→ A/(I + J)
←
A/J

(whose maps are the obvious ones) is both a pull-back and a push-out.
12-exo851) Let A be a ring and let
𝑓
→ P
←
N
←
←

𝑢 𝑣
→

𝑓0
N0 → P0
←

be a commutative diagram of A-modules.

374 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

a) Assume that this diagram is a pull-back (see exercise 7/11). Let 𝑔 0 : M → N0 be a

morphism of A-modules such that Im(𝑔 0) = Ker( 𝑓 0). Prove that there exists a unique
morphism 𝑔 : P → M such that 𝑔 0 = 𝑢 ◦ 𝑔 and the sequence
𝑔 𝑓
M→
− N→
− P
is exact.
b) Assume that this diagram is a push-out (see exercise 7/11). Let 𝑔 : P → Q0 be a
morphism of A-modules such that Ker(𝑔) = Im( 𝑓 ). Prove that there exists a unique
morphism 𝑔 0 : P0 → Q0 such that 𝑔 = 𝑔 0 ◦ 𝑣 and the sequence
𝑓0 𝑔0
N0 −
→ P0 −
→ Q0
is exact.
13-exo850) Prove this generalization of the snake lemma. Let A be a ring and let
𝑓 𝑔
→ N → P
← ←
M

←
←

←
𝑢 𝑣 𝑤
→

→
𝑓0 𝑔0
M0 → N0 → P0
← ←

be a commutative diagram of A-modules with exact rows, namely Ker(𝑔) = Im( 𝑓 ) and
Ker(𝑔 0) = Im( 𝑓 0).
a) Prove that the given morphisms induce exact sequences
𝑢 𝑓 𝑔
− Ker( 𝑓 0 ◦ 𝑢) →
0 → Ker( 𝑓 ) → − Ker(𝑣) →
− Ker(𝑤) ∩ Im(𝑔)
and
𝑓0 𝑔0 idP0
M0/(Im(𝑢) + Ker( 𝑓 0)) − → Coker(𝑤 ◦ 𝑔) −−→ Coker(𝑔 0) → 0.
→ Coker(𝑣) −
b) Construct a morphism of modules
𝛿 : Ker(𝑤) ∩ Im(𝑔) → M0/(Im(𝑢) + Ker( 𝑓 0))
that connects the preceding two exact sequences and gives rise to an exact sequence.
(In other words, Im(𝑔) = Ker(𝛿) and Im(𝛿) = Ker( 𝑓 0).)
c) Discuss the particular cases : (i) 𝑓 0 is injective; (ii) 𝑔 is surjective; (iii) 𝑓 0 is injective
and 𝑔 is surjective.
d) Show that the given morphisms induce an exact sequence
idM 𝑓
→ Ker( 𝑓 ) → Ker(𝑔 ◦ 𝑓 ) → Ker(𝑔)
← ← ←
0
←

𝛿
→

← Coker(𝑔) ← Coker(𝑔 ◦ 𝑓 ) 𝑔← Coker( 𝑓 )

→ → →
0
idP

in which the morphism 𝛿 is defined as in b).

 EXERCISES 375

14-exo1239) Give examples of commutative rings A, multiplicative subsets S of A, and A-

modules M, N such that the natural map of proposition 7.2.7 𝛿 M,N : S−1 HomA (M, N) →
HomS−1 A (S−1 M, S−1 N) is not injective, resp. not surjective.
15-exo853) Let A be a ring, let M be a A-module and let N be a submodule of M.
a) If N is injective, then it is a direct summand.
b) If M/N is projective, then N is a direct summand.
16-exo308) Let A be a ring, let P be an A-module and let M be a direct summand of P.
If P is an injective A-module, show that M is injective.
17-exo309) Let A be a ring.
a) Assume that A is right noetherian. Show that any direct sum of injective right
A-modules is injective. (Apply criterion (ii) of proposition 7.4.2.)
b) Let (I𝑛 )𝑛∈N be an increasing sequence of right ideals of A and let I be its union. For
every 𝑛, let M𝑛 be a right A-module containing A/I𝑛 , let 𝑓𝑛 : A → M𝑛 be the canonical
surjection from A to A/I𝑛 composed with the injection to M𝑛 . Let M be the direct sum
of the modules M𝑛 .
Show that there exists a unique map 𝑓 : I → M such that 𝑓 (𝑎) = ( 𝑓𝑛 (𝑎))𝑛 for every
𝑎 ∈ I.
c) (followed) Assuming that M is injective, prove that there exists 𝑛 ∈ N such that
I𝑛 = I.
d) (followed) Assume that every direct sum of injective right A-modules is injective.
Show that A is right noetherian (theorem of Bass-Papp).
18-exo307) Let A be a ring, let P be an A-module and let M be a submodule of P. One
says that a submodule N of P essential for M if it contains M and if N1 ∩ M ≠ 0 for every
non-zero submodule N1 of N.
a) Show that there exists a maximal submodule N of P containing M which is
essential for M. (Let 𝒮 be the set of submodules of P containing M which are essential. Show
that 𝒮 is inductive and apply Zorn’s lemma.)
b) Show that there exists a maximal submodule N0 of P such that M ∩ N0 = 0. (Show
that the set of such submodules is inductive.)
c) Assume that P is an injective A-module. Show that P = N⊕N0. Using exercise 7/16,
conclude that N is an injective A-module.
19-exo310) Let A be a ring and let M be a right A-module. Let 𝑗 : M → N and
𝑗 0 : M → N0 be two morphisms of M into right A-modules N and N0.
a) Assume that N is essential with respect to 𝑗(M) and that N0 is injective. Prove that
there exists an injective morphism 𝑓 : N → N0 such that 𝑗 0 = 𝑓 ◦ 𝑗.
b) Assume furthermore that N is injective and that N0 is essential with respect
to 𝑗(M0). Prove that 𝑓 is an isomorphism.
If M is an A-module, theorem 7.4.6 shows that there exists an injective A-module N
containing M. By exercise 7/18, there exists such an injective module N which is essential with
respect to M. The result of the present exercise shows that N is well-defined.
376 CHAPTER 7. FIRST STEPS IN HOMOLOGICAL ALGEBRA

20-exo311) Let A be a ring, let M, N be A-modules and let 𝑓 : M → N be a morphism.

a) Prove that 𝑓 is an epimorphism in the category of A-modules if and only 𝑓 is
surjective.
b) Prove that 𝑓 is a monomorphism in the category of A-modules if and only if 𝑓 is
injective.
21-exo312) Let A, B be rings. Let F be a functor from the category of A-modules to that
of B-modules, let G be a right adjoint to F. For any A-module M and any B-module N,
let ΦM,N : HomB (F(M), N) → HomA (M, G(N)) be the adjunction map.
a) Assume that the functor F is faithful. Show that for any injective morphism 𝑔 of
B-modules, the morphism G(𝑔) is injective. (Use the characterization of exercise 7/20.)
b) Assume that the functor G is faithful. Show that for any surjective morphism 𝑓 of
A-modules, the morphism F( 𝑓 ) is surjective.
22-exo316) Let A and B be rings, let F be a functor from the category of A-modules to
the category of B-modules, let G be a functor from the category of B-modules to the
category of A-modules. Assume that these functors form a pair of adjoint functors
and let (ΦM,N ) be an adjonction.
a) If F and G are additive, prove that the maps ΦM,N are additive.
b) Assume that A = B is a commutative ring and that the functors F, G are A-linear,
that is they are additive and map a homothety of ratio 𝑎 to a homothety of ratio 𝑎.
Prove that the maps ΦM,N are morphisms of A-modules.
23-exo895) Let A be a ring, let P be an A-module. Assume that for every injective
A-module Q, every surjective morphism 𝑝 : Q → Q00 and every morphism 𝑓 : P → Q00,
there exists a morphism 𝑔 : P → Q such that 𝑝 ◦ 𝑔 = 𝑓 . The goal of the first two
questions is to prove that P is projective.
𝑗 𝑝
a) Let 0 → M0 → − M → − M00 → 0 be an exact sequence of A-modules and let
𝑓 : P → M00 be a morphism of A-modules. Let Q be an injective A-module and
𝑖 : M → Q an injective morphism (Baer). Let Q0 = 𝑖(𝑗(M0)), let Q00 = Q/Q0 and let
𝑞 : Q → Q00 be the canonical surjection Prove that there exists a morphism 𝑘 : M00 → Q00
such that 𝑘 ◦ 𝑝 = 𝑞 ◦ 𝑖.
b) Let 𝑔 : P → Q be a morphism such that 𝑞 ◦ 𝑔 = 𝑘 ◦ 𝑓 . Prove that Im(𝑔) ⊂ Im(𝑖).
Conclude.
c) Let Q be an A-module. Show that Q is injective if and only if, for every projective
A-module P, every injective morphism 𝑖 : P0 → P and every morphism 𝑓 : P0 → Q,
there exists a morphism 𝑔 : P → Q such that 𝑔 ◦ 𝑖 = 𝑓 .
24-exo894) One says that a ring A is right hereditary if every right ideal of A is a
projective A-module.
a) Let A be a right hereditary ring and let 𝑛 be an integer. Prove that every
submodule M of A𝑛 is a projective A-module. (Argue by induction on 𝑛; consider the
morphism from M to A given by the last coordinate.)
 EXERCISES 377

b) More generally, if A is right hereditary, then every submodule of a free right

A-module is a direct sum of ideals. (Let L be a free A-module and let M be a submodule
of L. Choose a well-ordering of a basis of L, and argue similarly.)
c) Using exercise 7/24, prove that A is right hereditary if and only if every quotient
of an injective A-module is injective.
(Commutative) integral domains which are right hereditary are called Dedekind rings. They
will be studied in §9.6.
25-exo896) Let A = Z[T] and let K = Q(T) be its fraction field.
a) Let I = (2, T). Prove that there exists a morphism 𝑓 : I → K/A such that 𝑓 (2) = 0
and 𝑓 (T) is the class of 1/2. Prove that 𝑓 does not extend to A.
b) Prove that K/A is divisible but is not an injective A-module.
26-exo898) Schanuel’s lemma. Let A be a ring.
a) Let 0 → P0 → P → M → 0 and 0 → Q0 → Q → M → 0 be exact sequences of
right A-modules.
If P and Q are projective, then P ⊕ Q0 ' P0 ⊕ Q.
b) Give an example that shows that the projectivity assumption is necessary.
c) Let 0 → M → P → P0 → 0 and 0 → M → Q → Q0 → 0 be exact sequences of
right A-modules.
If P and Q are injective, then P ⊕ Q0 ' P0 ⊕ Q.
CHAPTER 8

TENSOR PRODUCTS AND

DETERMINANTS

A companion chapter to linear algebra, multilinear algebra studies situations

where quantities depend on many variables and vary linearly in each of them.
The tensor product construction takes two modules M and N and makes a
module M⊗N out of it, which reflects bilinear maps from M×N. When M and N
are free, then so is M ⊗ N, and the tensors of old-style differential/riemannian
geometry, a bunch of real numbers with indices or exponents, are nothing but
the coordinates of such algebraic objects. Tensor products are also important in
other fields of mathematics, such as representation theory where they are used
to define induced representations.
This construction can also serve to associate various algebras with a module
over a commutative ring. Despite its commutative character, the symmetric
algebra will not be studied thoroughly here; in the simplest case, when the
given module is free, it furnishes another construction of rings of polynomials.
However, because of its relation with determinants and, henceforth, with linear
algebra, I give a more detailed presentation of the exterior algebra.
In the next section, I prove that tensoring with a given module preserves some
exact sequences. This is an extremly important property which I first prove by
relatively explicit computations. I then give another, more abstract, proof, based
on the fact that this functor is right exact, a property that itself derives from
the universal property of the tensor product construction.
However, tensoring with a given module is not left exact in general, and
modules which preserve all exact sequences are called flat. Free modules, more
generally projective modules, are flat, and it is useful to characterize flat modules.
Over principal ideal domains, flat modules are just torsion-free modules, but
this condition is not sufficient in general. I then prove a theorem of Lazard that
shows that a finitely presented module is flat if and only if it is projective.
380 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Faithful flatness is a reinforcement of flatness. Its importance is revealed in

the next section where I explain Grothendieck’s idea of faithfully flat descent. If
𝑓 : A → B is a morphism of commutative rings, the tensor product construction
furnishes a functor M ↦→ B ⊗A M from A-modules to B-modules. When the
morphism 𝑓 is faithfully flat, the module B ⊗A M reflects many properties of M,
and I will show some of them. Even more remarkably, Grothendieck proved that
the module B ⊗A M carries an additional structure, called a descent datum
which allows to recover the initial A-module M.
In a final section, I show how when 𝑓 is a finite Galois extension of fields, this
general theorem translates to Galois descent via a too lengthy computation.

8.1. Tensor product of two modules

Let A be a ring, let M be a right A-module and N be a left A-module.
We are interested here in the balanced bi-additive maps from M × N into an
abelian group P, that is to say, to the maps 𝑓 : M × N → P which satisfy
the following properties:
(i) 𝑓 (𝑚 + 𝑚 0 , 𝑛) = 𝑓 (𝑚, 𝑛) + 𝑓 (𝑚 0 , 𝑛) (left additivity);
(ii) 𝑓 (𝑚, 𝑛 + 𝑛 0) = 𝑓 (𝑚, 𝑛) + 𝑓 (𝑚, 𝑛 0) (right additivity);
(iii) 𝑓 (𝑚𝑎, 𝑛) = 𝑓 (𝑚, 𝑎𝑛) (balancing condition).
Our first goal is to linearize the map 𝑓 , that is, to replace it by a linear
map 𝜑 between appropriate modules so that the study of 𝜑 would be
more or less equivalent to that of 𝑓 .

8.1.1. Construction. — Let T1 = Z(M×N) be the free Z-module with

basis M × N; for 𝑚 ∈ M and 𝑛 ∈ N, let 𝑒 𝑚,𝑛 be the element (𝑚, 𝑛) of the
basis of T1 . Let T2 be the submodule of T1 generated by the following
elements:
𝑒 𝑚+𝑚0 ,𝑛 − 𝑒 𝑚,𝑛 − 𝑒 𝑚0 ,𝑛 , 𝑒 𝑚,𝑛+𝑛0 − 𝑒 𝑚,𝑛 − 𝑒 𝑚,𝑛0 , 𝑒 𝑚𝑎,𝑛 − 𝑒 𝑚,𝑎𝑛 ,
where 𝑚, 𝑚 0 run along M, 𝑛, 𝑛 0 run along N and 𝑎 runs along A.
Finally, let T = T1 /T2 and let 𝑚 ⊗ 𝑛 denote the class of 𝑒 𝑚,𝑛 in the
quotient T. Let 𝜃 be the map from M × N to M ⊗ N given by 𝜃(𝑚, 𝑛) =
𝑚 ⊗ 𝑛. By construction, the map 𝜃 is balanced and bi-additive. Indeed,
𝜃(𝑚+𝑚 0 , 𝑛)−𝜃(𝑚, 𝑛)−𝜃(𝑚 0 , 𝑛) is the class in T1 /T2 of 𝑒 𝑚+𝑚0 ,𝑛 −𝑒 𝑚,𝑛 −𝑒 𝑚0 ,𝑛 ,
hence is 0; the two other axioms are verified in the same way.
 8.1. TENSOR PRODUCT OF TWO MODULES 381

By the universal property of free Z-modules, there is a unique

morphism 𝜑1 from T1 to P such that 𝑓 (𝑚, 𝑛) = 𝜑1 (𝑒(𝑚,𝑛) ) for ev-
ery (𝑚, 𝑛) ∈ M × N. Then, the kernel of 𝜑1 contains T2 if and only
if 𝜑1 vanishes on each of the given generators of T2 , in other words, if
and only if 𝑓 is A-balanced and bi-additive. Consequently, there exists a
morphism of abelian groups 𝜑 : T → P such that 𝜑(𝑚 ⊗ 𝑛) = 𝑓 (𝑚, 𝑛).

Definition (8.1.2). — The abelian group T defined above is called the tensor
product of the A-modules M and N; it is denoted M ⊗A N. Its elements are
called tensor The map 𝜃 : M × N → M ⊗A N defined by 𝜃(𝑚, 𝑛) = 𝑚 ⊗ 𝑛
is called the canonical A-balanced bi-additive map. The elements of M ⊗A N
of the form 𝑚 ⊗ 𝑛, for some (𝑚, 𝑛) ∈ M × N, (the image of 𝜃) are called split
tensors.

Since the split tensors are the image in T = T1 /T2 of a basis of T1 , they
form a generating family of T.

Proposition (8.1.3) (Universal property of tensor products)

For any abelian group P and any A-balanced and bi-additive map 𝑓 : M ×
N → P, there exists a unique morphism of abelian groups 𝜑 : M ⊗A N → P
such that 𝜑 ◦𝜃 = 𝑓 (explicitly: 𝑓 (𝑚, 𝑛) = 𝜑(𝑚 ⊗ 𝑛) for every (𝑚, 𝑛) ∈ M×N).

Proof. — We just showed the existence of a morphism 𝜑 such that

𝜑 ◦ 𝜃 = 𝑓 . Let 𝜑, 𝜑0 be two such morphisms, and let us prove that
𝜑 = 𝜑0. Let 𝛿 = 𝜑0 −𝜑; it is a morphism of an abelian groups from M⊗A N
to P. By assumption,

𝛿(𝑚 ⊗ 𝑛) = 𝜑0(𝑚 ⊗ 𝑛) − 𝜑(𝑚 ⊗ 𝑛) = 𝑓 (𝑚, 𝑛) − 𝑓 (𝑚, 𝑛) = 0

for every (𝑚, 𝑛) ∈ M ⊗ N. Consequently, the kernel of 𝛿 contains every

split tensor. Since they generate M ⊗A N, one has Ker(𝛿) = M ⊗A N and
𝛿 = 0. 

Remark (8.1.4). — It is probably useful to repeat things once more.

382 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

The property for the map 𝜃 : (𝑚, 𝑛) ↦→ 𝑚 ⊗ 𝑛 to be A-balanced and

bi-additive is equivalent to the following relations

(𝑚 + 𝑚 0) ⊗ 𝑛 = 𝑚 ⊗ 𝑛 + 𝑚 0 ⊗ 𝑛,
𝑚 ⊗ (𝑛 + 𝑛 0) = 𝑚 ⊗ 𝑛 + 𝑚 ⊗ 𝑛 0 ,
𝑚𝑎⊗𝑛 = 𝑚 ⊗ 𝑎𝑛

for 𝑚, 𝑚 0 ∈ M, 𝑛, 𝑛 0 ∈ N and 𝑎 ∈ A.
Moreover, both the universal property of the tensor product M ⊗A N
expressed by proposition 8.1.3 and its construction above show that
to define a morphism 𝜑 from the abelian group M ⊗A N to an abelian
group P, we just need to construct a A-balanced and bi-additive map 𝑓
from M × N to P, the relation between 𝜑 and 𝑓 being the equality
𝑓 (𝑚, 𝑛) = 𝜑(𝑚 ⊗ 𝑛).
Finally, to check that a morphism from a tensor product M ⊗A N to
an abelian group P vanish, it suffices to check that it maps all split
tensors to 0. Similarly, to check that two morphisms from a tensor
product M ⊗A N to P coincide, it suffices to check that their difference
vanishes, hence that these two morphisms coincide on split tensors.

8.1.5. Functoriality. — Let M0 be another right A-module, let N0 be

another left A-module. Let 𝑢 : M → M0 and 𝑣 : N → N0 be morphisms
of A-modules. The map from M × N to M0 ⊗A N0 that sends (𝑚, 𝑛) to
𝑢(𝑚) ⊗ 𝑣(𝑛) is A-balanced and bi-additive, as the following computations
show:

𝑢(𝑚 + 𝑚 0) ⊗ 𝑣(𝑛) = (𝑢(𝑚) + 𝑢(𝑚 0)) ⊗ 𝑣(𝑛) = 𝑢(𝑚) ⊗ 𝑣(𝑛) + 𝑢(𝑚 0) ⊗ 𝑣(𝑛),
𝑢(𝑚) ⊗ 𝑣(𝑛 + 𝑛 0) = 𝑢(𝑚) ⊗ 𝑣(𝑛) + 𝑢(𝑚) ⊗ 𝑣(𝑛 0),
𝑢(𝑚𝑎) ⊗ 𝑣(𝑛) = 𝑢(𝑚)𝑎 ⊗ 𝑣(𝑛) = 𝑢(𝑚) ⊗ 𝑎𝑣(𝑛) = 𝑢(𝑚) ⊗ 𝑣(𝑎𝑛).

By the universal property of the tensor product, there exists a unique

morphism of abelian groups from M ⊗A N to M0 ⊗A N0, written 𝑢 ⊗ 𝑣,
such that (𝑢 ⊗ 𝑣)(𝑚 ⊗ 𝑛) = 𝑢(𝑚) ⊗ 𝑣(𝑛) for every (𝑚, 𝑛) ∈ M × N.
Let M00 and N00 be yet other respectively right and left A-modules, let
𝑢 0 : M0 → M00 and 𝑣 0 : N0 → N00 be morphisms of A-modules. Applied
to a split tensor 𝑚 ⊗ 𝑛, the morphisms (𝑢 0 ◦𝑢)⊗(𝑣 0 ◦𝑣) and (𝑢 0 ⊗𝑣 0)◦(𝑢 ⊗𝑣),
 8.1. TENSOR PRODUCT OF TWO MODULES 383

both furnish the split tensor (𝑢 0 ◦ 𝑢(𝑚)) ⊗ (𝑣 0 ◦ 𝑣(𝑛)). Consequently, these

two morphisms are equal.
Let us now assume that M = M0, N = N0 and let us take for morphisms
the identity maps of M and N. One has idM ⊗ idN = idM⊗N , since these
two morphisms send a split tensor 𝑚 ⊗ 𝑛 to itself.
Let us still assume that M0 = M and N0 = N, but let 𝑢, 𝑣 be arbitrary
endormorphisms of M and N, respectively. Then, 𝑢 ⊗ 𝑣 is an endo-
morphism of M ⊗A N. Moreover, the maps EndA (M) → End(M ⊗A N)
and EndA (N) → End(M ⊗A N) respectively given by 𝑢 ↦→ 𝑢 ⊗ idN and
𝑣 ↦→ idM ⊗𝑣 are morphisms of rings.

Remark (8.1.6). — Let A be a ring, let M be a right A-module and let N

be a left A-module. Let 𝜋 be an element of the tensor product M ⊗A N.
By definition, it can be written 𝑖∈I 𝑚 𝑖 ⊗ 𝑛 𝑖 for a finite set I and finite
Í
families (𝑚 𝑖 )𝑖∈I in M and (𝑛 𝑖 )𝑖∈I in N. Let M0 be the submodule of M
generated by the family (𝑚 𝑖 ), and let N0 be the submodule of N generated
by the family (𝑛 𝑖 ); they are finitely generated, by construction. Moreover,
the element 𝜋 is the image of an element 𝜋0 ∈ M0 ⊗A N0 by the canonical
morphism 𝑓0 ⊗ 𝑔0 : M0 ⊗A N0 → M ⊗A N deduced from the inclusions
𝑓0 : M0 → M and 𝑔0 : N0 → N: it suffices to set 𝜋0 = 𝑖∈I 𝑚 𝑖 ⊗ 𝑛 𝑖 , where
Í
these split tensors are viewed in M0 ⊗A N0 .
However, a representation of a tensor 𝜋 = 𝑚 𝑖 ⊗ 𝑛 𝑖 is by no means
Í
unique, and we may well have 𝜋 = 0 and 𝜋0 ≠ 0. If 𝜋 = 0, the
construction of the tensor product module M ⊗A N shows that the
element 𝑒 𝑚𝑖 ,𝑛 𝑖 of T1 = Z(M×N) , in the notation of 8.1.1, belongs to T2 ,
Í
hence is itself some finite linear combination of basic elements of the
form 𝑒 𝑚+𝑚0 ,𝑛 − 𝑒 𝑚,𝑛 − 𝑒 𝑚0 ,𝑛 , 𝑒 𝑚,𝑛+𝑛0 − 𝑒 𝑚,𝑛 − 𝑒 𝑚,𝑛0 and 𝑒 𝑚𝑎,𝑛 − 𝑒 𝑚,𝑎𝑛 . Let M1
be the submodule of M generated by M0 and by all the elements 𝑚, 𝑚 0
that appear; similarly, let N1 be the submodule of N generated N0 and
by all the elements 𝑛, 𝑛 0 that appear in such a linear combination. Let
𝑓1 : M1 → M and 𝑔1 : N1 → N be the inclusions. Under the morphism
𝑓1 ⊗ 𝑔1 , the tensor 𝜋1 = 𝑚 𝑖 ⊗ 𝑛 𝑖 of M1 ⊗A N1 maps to 𝜋. Moreover, one
Í
has 𝜋1 = 0.
In other words, the nullity of an explicitly written tensor 𝜋 of M ⊗A N
can be witnessed in the tensor product M1 ⊗A N1 , where M1 and N1 are
finitely generated submodules of M and N respectively.
384 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

8.1.7. Bimodules. — Let us assume in particular that M be endowed

with the structure of a (B, A)-bimodule and that N be endowed with
the structure of an (A, C)-bimodule. If we have no such structure at our
disposal, we can always take B = Z or C = Z. If A is a comutative ring,
we can also choose B = C = A.
Anyway, we have a morphism of rings B → EndA (M) (which sends
𝑏 ∈ B to the left-multiplication 𝜆𝑏 by 𝑏 in M) and a morphism of rings
Co → EndB (N) (which sends 𝑐 ∈ C to the right-multiplication 𝜌 𝑐 by 𝑐
in N). This shows that we would essentially consider the most general
situation by assuming that B = EndA (M) and C = EndA (M)o , the above
morphism of rings would then be identities.
By composition, we get two ring morphisms B → End(M ⊗A N) and
C → End(M ⊗A N), given by 𝑏 ↦→ 𝜆𝑏 ⊗ idN and 𝑐 ↦→ idM ⊗𝜌 𝑐 . They
o

endow the abelian group M ⊗A N with a structure of a left B-module

and of a right C-module. Moreover, for 𝑏 ∈ B, 𝑐 ∈ C, 𝑚 ∈ M and 𝑛 ∈ N,
we have
𝑏 · ((𝑚 ⊗ 𝑛) · 𝑐) = 𝑏 · (𝑚 ⊗ 𝑛𝑐) = 𝑏𝑚 ⊗ 𝑛𝑐 = (𝑏 · (𝑚 ⊗ 𝑛)) · 𝑐,
so that these two structures are compatible. This shows that M ⊗A N
has a natural structure of a (B, C)-bimodule.
Let P be a (B, C)-bimodule. For a morphism of abelian groups 𝜑 : M ⊗A
N → P to be (B, C)-linear, it is necessary and sufficient that the A-
balanced and bi-additive map 𝑓 = 𝜑◦𝜃 : M×N → P satisfies 𝑓 (𝑏𝑚, 𝑛𝑐) =
𝑏 𝑓 (𝑚, 𝑛)𝑐 for every 𝑚 ∈ M, every 𝑛 ∈ N, every 𝑏 ∈ B and every 𝑐 ∈ C.
In particular, if 𝑢 : M → M0 is a morphism of (B, A)-bimodules and
𝑣 : N → N0 is a morphism of (A, C)-bimodules, then the morphism 𝑢 ⊗ 𝑣
from M ⊗A N to M0 ⊗A N0 is a morphism of (B, C)-modules.
If A is a commutative ring and B = C = A, we then have 𝑎(𝑚 ⊗ 𝑛) =
(𝑚𝑎) ⊗ 𝑛 = 𝑚 ⊗ (𝑎𝑛) = (𝑚 ⊗ 𝑛)𝑎, so that the two given structures of an
A-module on M ⊗A N coincide.

Proposition (8.1.8) (Compatibility with direct sums)

Let (M𝑖 )𝑖∈I be a family of right A-modules and (N 𝑗 ) 𝑗∈J be a family É of left
É For everyÉ
A-modules. (𝑖, 𝑗) ∈ I × J, let P𝑖,𝑗 = M𝑖 ⊗A N 𝑗 . Set M = 𝑖∈I M 𝑖 ,
N= 𝑗∈J N 𝑗 , P = (𝑖,𝑗)∈I×J P𝑖,𝑗 . For 𝑖 ∈ I and 𝑗 ∈ J, let 𝛼 𝑖 : M 𝑖 → M,
𝛽 𝑗 : N 𝑗 → N, 𝛾𝑖,𝑗 : P𝑖,𝑗 → P be the canonical injections. There is a unique
 8.1. TENSOR PRODUCT OF TWO MODULES 385

morphism 𝜆 : P → M ⊗A N such that 𝜆 ◦ 𝛾𝑖,𝑗 = 𝛼 𝑖 ⊗ 𝛽 𝑗 for every (𝑖, 𝑗) ∈ I × J.

Moreover, 𝜆 is an isomorphism.

Proof. — Existence and uniqueness of 𝜆 follows from the universal

property of the direct sum. To prove that 𝜆 is an isomorphism, we shall
construct its inverse explicitly. For 𝑖 ∈ I and 𝑗 ∈ J, let 𝑝 𝑖 : M → M𝑖 ,
𝑞 𝑗 : N → N 𝑗 and 𝑟 𝑖,𝑗 : P → M𝑖 ⊗ N 𝑗 be the canonical
É projections.
É
By construction of the direct sums M = M𝑖 and N = N 𝑗 , the
families (𝑝 𝑖 (𝑚))𝑖∈I and (𝑞 𝑗 (𝑛)) 𝑗∈J are almost null, for any (𝑚, 𝑛) ∈ M × N.
Consequently, in the family (𝑝 𝑖 (𝑚) ⊗ 𝑞 𝑗 (𝑛))(𝑖,𝑗)∈I×J , only finitely many
terms are nonzero, so that this family is not only an element É of the
Î
product module (𝑖,𝑗)∈I×J P𝑖,𝑗 but one of the direct sum P = P𝑖,𝑗 . This
defines a map 𝑓 : M × N → P. Let us show that this map is A-balanced
and bi-additive. Indeed, for any 𝑚, 𝑚 0 ∈ M, 𝑛, 𝑛 0 ∈ N and any 𝑎 ∈ A,

𝑓 (𝑚 + 𝑚 0 , 𝑛) = 𝑝 𝑖 (𝑚 + 𝑚 0) ⊗ 𝑞 𝑗 (𝑛)


𝑖,𝑗
0
= 𝑝 𝑖 (𝑚) ⊗ 𝑞 𝑗 (𝑛) + 𝑝 𝑖 (𝑚 ) ⊗ 𝑞 𝑗 (𝑛)


𝑖,𝑗

= 𝑝 𝑖 (𝑚) ⊗ 𝑞 𝑗 (𝑛) + 𝑝 𝑖 (𝑚 0) ⊗ 𝑞 𝑗 (𝑛)

𝑖,𝑗
= 𝑓 (𝑚, 𝑛) + 𝑓 (𝑚 0 , 𝑛)

and a similar proof shows that

𝑓 (𝑚, 𝑛 + 𝑛 0) = 𝑓 (𝑚, 𝑛) + 𝑓 (𝑚, 𝑛 0).

Moreover, for every 𝑎 ∈ A, one has

𝑓 (𝑚𝑎, 𝑛) = 𝑝 𝑖 (𝑚𝑎) ⊗ 𝑞 𝑗 (𝑛)

𝑖,𝑗

= 𝑝 𝑖 (𝑚)𝑎 ⊗ 𝑞 𝑗 (𝑛)


𝑖,𝑗

= 𝑝 𝑖 (𝑚) ⊗ 𝑎𝑞 𝑗 (𝑛)


𝑖,𝑗

= 𝑝 𝑖 (𝑚) ⊗ 𝑞 𝑗 (𝑎𝑛)


𝑖,𝑗
= 𝑓 (𝑚, 𝑎𝑛),

since the maps 𝑝 𝑖 and 𝑞 𝑗 are A-linear. Consequently, there exists

a unique morphism of abelian groups 𝜑 : M ⊗A N → P such that
𝜑(𝑚 ⊗ 𝑛) = 𝑓 (𝑚, 𝑛) for every (𝑚, 𝑛) ∈ M × N.
386 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

For every (𝑚, 𝑛) ∈ M × N, one has

Õ
𝜆(𝜑(𝑚 ⊗ 𝑛))𝜆( 𝑓 (𝑚, 𝑛)) = 𝜆 𝑝 𝑖 (𝑚) ⊗ 𝑞 𝑗 (𝑚)


𝑖,𝑗
Õ Õ
= 𝜆(𝑝 𝑖 (𝑚) ⊗ 𝑞 𝑗 (𝑛)) = 𝛼 𝑖 (𝑝 𝑖 (𝑚)) ⊗ 𝛽 𝑗 (𝑞 𝑗 (𝑛))
𝑖,𝑗 𝑖,𝑗
!
Õ ©Õ
= 𝛼 𝑖 (𝑝 𝑖 (𝑚)) ⊗ ­ 𝛽 𝑗 (𝑞 𝑗 (𝑛))® = 𝑚 ⊗ 𝑛.
ª
𝑖 « 𝑗 ¬
Since the split tensors generate M ⊗A N, one has 𝜆(𝜑(𝜉)) = 𝜉 for every
𝜉 ∈ M ⊗A N, so that 𝜆 ◦ 𝜑 is the identity on M ⊗A N. On the other hand,
for any (𝑢, 𝑣) ∈ I × J, any 𝑚 ∈ M𝑢 and any 𝑛 ∈ N𝑣 , one has
Õ
𝜑(𝜆(𝛾𝑢,𝑣 (𝑚 ⊗ 𝑛))) = 𝜑(𝛼 𝑢 (𝑚) ⊗ 𝛽 𝑣 (𝑛)) = 𝑝 𝑖 (𝛼 𝑢 (𝑚)) ⊗ 𝑞 𝑗 (𝛽 𝑣 (𝑛)) .


𝑖,𝑗

By definition of the projections 𝑝 𝑖 , one has 𝑝 𝑖 (𝛼 𝑢 (𝑚)) = 𝑚 if 𝑖 = 𝑢, and

𝑝 𝑖 (𝛼 𝑢 (𝑚)) = 0 if 𝑖 ≠ 𝑢; similarly, 𝑞 𝑗 (𝛽 𝑣 (𝑛)) = 𝑛 if 𝑗 = 𝑣 and is zero
otherwise. This implies that
𝜑(𝜆(𝛾𝑢,𝑣 (𝑚 ⊗ 𝑛))) = 𝛾𝑢,𝑣 (𝑚 ⊗ 𝑛).
Since the split tensors 𝑚 ⊗𝑛, for 𝑚 ∈ M𝑢 and 𝑛 ∈ N𝑣 , generate M𝑢 ⊗A N𝑣 =
P𝑢,𝑣 , one has 𝜑(𝜆(𝛾𝑢,𝑣 (𝜉))) = 𝛾𝑢,𝑣 (𝜉) for every 𝜉 ∈ P𝑢,𝑣 . Since the
submodules 𝛾𝑢,𝑣 (P𝑢,𝑣 ) of P generate P, it follows that 𝜑 ◦ 𝜆 = idP .
This implies that 𝜆 is an isomorphism and that 𝜑 is its inverse. 

Remark (8.1.9). — Let us keep the hypotheses of the proposition. Let

B, C be rings and let us assume moreover that the modules M𝑖 are
(B, A)-bimodules, and the modules N 𝑗 are (A, C)-bimodules. Then, M
is a (B, A)-bimodule, N is an (A, C)-bimodule and 𝜆 is an isomorphism
of (B, C)-bimodules.
We need to show that 𝜆(𝑏𝑡𝑐) = 𝑏𝜆(𝑡)𝑐 for every 𝑏 ∈ B, every 𝑐 ∈ C and
every 𝑡 ∈ M ⊗A N. It follows from the definition of 𝜆 that
𝜆(𝑏(𝑚 ⊗ 𝑛)𝑐) = 𝜆((𝑏𝑚) ⊗ (𝑛𝑐)) = 𝑏𝜆(𝑚 ⊗ 𝑛)𝑐
for every 𝑚 ∈ M, 𝑛 ∈ N, 𝑏 ∈ B and 𝑐 ∈ C. Since the split tensors 𝑚 ⊗ 𝑛
generate M ⊗A N, the result follows.
 8.1. TENSOR PRODUCT OF TWO MODULES 387

Proposition (8.1.10). — Let A be a ring, let M be a right A-module, let I be

a left ideal of A. Then, the map 𝑓 from M to M ⊗A (A/I) which maps 𝑚 to
the tensor 𝑚 ⊗ cl(1) is a surjective morphism of abelian groups. Its kernel
is the subgroup MI of M, and 𝑓 induces an isomorphism of abelian groups
𝜑 : M/MI → M ⊗A (A/I).
Moreover, if B is a ring and M is a (B, A)-bimodule, then 𝑓 is a morphism of
left B-modules.
Proof. — It is obvious that 𝑓 is additive (and B-linear if M is a (B, A)-
bimodule). Let 𝑚 ∈ M and let 𝑎 ∈ I; we have
𝑓 (𝑚𝑎) = 𝑚𝑎 ⊗ cl(1) = 𝑚 ⊗ 𝑎 cl(1) = 𝑚 ⊗ 0 = 0,
so that the kernel of 𝑓 contains the abelian subgroup MI of M. Conse-
quently, 𝑓 defines, by quotient, a morphism 𝜑 : M/MI → M ⊗A (A/I).
To prove that 𝜑 is an isomorphism, we will construct its inverse.
The map 𝑔 : M × (A/I) → M/MI sending (𝑚, cl(𝑎)) to the class of 𝑚𝑎
is well-defined: if cl(𝑎) = cl(𝑏), there exists 𝑥 ∈ I such that 𝑏 = 𝑎 + 𝑥;
then, 𝑚𝑏 = 𝑚𝑎 + 𝑚𝑥 = 𝑚𝑎 (mod MI). It is obviously bi-additive, as
well as A-balanced (for 𝑚 ∈ M and 𝑎, 𝑏 ∈ A, 𝑔(𝑚𝑏, cl(𝑎)) = cl(𝑚𝑏𝑎) =
𝑔(𝑚, cl(𝑏𝑎))). Therefore, there exists a unique morphism of abelian
groups 𝛾 : M ⊗A (A/I) → M/MI such that 𝛾(𝑚 ⊗ cl(𝑎)) = cl(𝑚𝑎) for
every 𝑚 ∈ M and every 𝑎 ∈ A.
If 𝑚 ∈ M, one has 𝛾(𝜑(clMI (𝑚))) = 𝛾(𝑚 ⊗ 1) = clMI (𝑚), hence 𝛾 ◦ 𝜑 is
the identity of M/MI. On the other hand, for any 𝑚 ∈ M and any 𝑎 ∈ A,
one has
𝜑(𝛾(𝑚 ⊗ cl(𝑎))) = 𝜑(cl(𝑚𝑎)) = 𝑚𝑎 ⊗ cl(1) = 𝑚 ⊗ 𝑎 cl(1) = 𝑚 ⊗ cl(𝑎).
Since the split tensors generate M ⊗ (A/I), 𝜑 ◦ 𝛾 is the identity morphism
of M ⊗A (A/I). This concludes the proof of the proposition. 

Remark (8.1.11). — Let M be a right A-module, N be a left A-module,

P be an abelian group, and let 𝑓 : M ⊗A N → P be a morphism of
abelian groups. Suppose we need to prove that 𝑓 is an isomorphism.
Surjectivity of 𝑓 is often easy, because in most cases it is quite obvious
to spot preimages of suitable generators of P. On the other hand,
the injectivity is more difficult. A direct approach would begin by
considering a tensor 𝑚 𝑖 ⊗ 𝑛 𝑖 with image 0, and then trying to prove
Í
388 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

that this tensor is zero. By the given definition of the tensor product,
this requires to prove that the element 𝑒(𝑚𝑖 ,𝑛 𝑖 ) in the free abelian group
Í
Z(M×N) is a linear combination of the elementary relations. But how?
In all important cases, an efficient way of proving that 𝑓 is an isomor-
phism consists in defining its inverse 𝑔. To that aim, the proof of the
surjectivity of 𝑓 is useful. Indeed, one has often written a formula for
the inverse 𝑔(𝑧) of some elements 𝑧 which generate P. Then try to prove
that there exists a unique morphism 𝑔 from P to M ⊗A N; in general,
this amounts to checking that some formula does not depend on choices
needed to write it down. To conclude the proof, show that 𝑓 and 𝑔 are
inverses one of the other.

8.1.12. Change of base ring. — Let A, C be rings, let 𝛼 : A → C be

a morphism of rings. Let M be a right A-module. The ring C can be
seen as a left A-module via the morphism 𝛼, and as a right C-module;
consequently, the abelian group M ⊗A C has a natural structure of right
C-module.
In this setting, the universal property of the tensor product furnishes
the following result: For every right C-module P and any A-linear morphism
𝑓 : M → P, there exists a unique C-linear morphism 𝜑 : M ⊗A C → P such
that 𝜑(𝑚 ⊗ 1) = 𝑓 (𝑚) for every 𝑚 ∈ M.
Let indeed 𝑓 : M → P be an A-linear map, that is, an additive map
such that 𝑓 (𝑚𝑎) = 𝑓 (𝑚)𝛼(𝑎) for every 𝑎 ∈ A. Let us first show that there
is at most one C-linear map 𝜑 as required. Indeed, if 𝜑 and 𝜑0 are two
such maps, one has
𝜑(𝑚 ⊗ 𝑐) = 𝜑((𝑚 ⊗ 1)𝑐) = 𝜑(𝑚 ⊗ 1)𝑐 = 𝑓 (𝑚)𝑐
for every pair (𝑚, 𝑐) ∈ M × C, and 𝜑0(𝑚 ⊗ 𝑐) = 𝑓 (𝑚)𝑐 likewise. Conse-
quently, the C-linear map 𝜑0 − 𝜑 vanishes on all split tensors of M ⊗A C,
hence is zero.
To prove the existence of map 𝜑 such that 𝜑(𝑚 ⊗ 1) = 𝑓 (𝑚) for every
𝑚 ∈ M, we first consider the map 𝑓1 : (𝑚, 𝑐) ↦→ 𝑓 (𝑚)𝑐 from M×C to P. It is
obviously bi-additive, as well as A-balanced since 𝑓1 (𝑚𝑎, 𝑐) = 𝑓 (𝑚𝑎)𝑐 =
𝑓 (𝑚)𝑎𝑐 = 𝑓1 (𝑚, 𝑎𝑐). Consequently, there exists a unique morphism of
abelian groups 𝜑 : M ⊗A C → P such that 𝜑(𝑚 ⊗ 𝑐) = 𝑓 (𝑚)𝑐 for every
𝑚 ∈ M and every 𝑐 ∈ C. In particuliar, 𝜑(𝑚 ⊗ 1) = 𝑓 (𝑚).
 8.2. TENSOR PRODUCT OF MODULES OVER A COMMUTATIVE RING 389

Let us show that 𝜑 is C-linear. For any 𝑐 0 ∈ C, one has 𝜑((𝑚 ⊗ 𝑐)𝑐 0) =
𝜑(𝑚⊗𝑐𝑐 0) = 𝑓 (𝑚)𝑐𝑐 0 = 𝜑(𝑚⊗𝑐)𝑐 0. Since the split tensors generate M⊗A C,
one has 𝜑(𝜉𝑐 0) = 𝜑(𝜉)𝑐 0 for every 𝜉 ∈ M ⊗A C and every 𝑐 0 ∈ C. In other
words, 𝜑 is C-linear.

8.2. Tensor product of modules over a commutative ring

This is probably the most important case and we first sum-up the
situation.
Let A be a commutative ring, let M, N be A-modules. We constructed
an abelian group M ⊗A N and a A-balanced bi-additive map 𝜃 : M × N →
M ⊗A N. We also endowed the group M ⊗A N with the unique structure
of an A-module for which

𝑎(𝑚 ⊗ 𝑛) = (𝑎𝑚) ⊗ 𝑛 = 𝑚 ⊗ (𝑎𝑛).

Consequently, the map 𝜃 satisfies the relations

(i) 𝜃(𝑚 + 𝑚 0 , 𝑛) = 𝜃(𝑚, 𝑛) + 𝜃(𝑚 0 , 𝑛),
(ii) 𝜃(𝑚, 𝑛 + 𝑛 0) = 𝜃(𝑚, 𝑛) + 𝜃(𝑚, 𝑛 0),
(iii) 𝜃(𝑎𝑚, 𝑛) = 𝜃(𝑚, 𝑎𝑛) = 𝑎𝜃(𝑚, 𝑛),
for 𝑚 ∈ M, 𝑛 ∈ N and 𝑎 ∈ A. In other words, 𝜃 is A-bilinear.
Let 𝑓 be an A-bilinear map from M × N to an A-module P. There exists
a unique map which is bi-additive and A-balanced 𝜑 : M ⊗A N → P
such that 𝜑(𝑚 ⊗ 𝑛) = 𝑓 (𝑚, 𝑛) for every (𝑚, 𝑛) ∈ M × N. Moreover, if
𝑎 ∈ A, then

𝜑(𝑎(𝑚 ⊗ 𝑛)) = 𝜑(𝑎𝑚 ⊗ 𝑛) = 𝑓 (𝑎𝑚, 𝑛) = 𝑎 𝑓 (𝑚, 𝑛) = 𝑎𝜑(𝑚 ⊗ 𝑛).

Since the split tensors generate M ⊗A N, we have 𝜑(𝑎𝜉) = 𝑎𝜑(𝜉) for every
𝜉 ∈ M ⊗A N. This means that 𝜑 is A-linear and is the unique A-linear
map from M ⊗A N to P such that 𝜑 ◦ 𝜃 = 𝑓 .

Proposition (8.2.1). — Let A be a commutative ring, let M and N be A-modules.

Let (𝑒 𝑖 )𝑖∈I and ( 𝑓 𝑗 ) 𝑗∈J be families of elements respectively in M and N.
If these families (𝑒 𝑖 ) and ( 𝑓 𝑗 ) are generating families (resp. are bases), then
the family (𝑒 𝑖 ⊗ 𝑓 𝑗 )(𝑖,𝑗)∈I×J generates (resp. is a basis of) M ⊗A N.
390 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Proof. — Let us assume that both families (𝑒 𝑖 ) and ( 𝑓 𝑗 ) are generating

families and let us show that the family (𝑒 𝑖 ⊗ 𝑓 𝑗 ) generates M ⊗A N. Since
the split tensors of M ⊗A N generate M ⊗A N, it suffices to show that every
split tensor is a linear combination of elements of the form 𝑒 𝑖 ⊗ 𝑓 𝑗 . Let
thus 𝑚 ∈ M and 𝑛 ∈ N. By assumption, there exists a family (𝑎 𝑖 ) ∈ A(I)
such that 𝑚 = 𝑎 𝑖 𝑒 𝑖 , as well as a family (𝑏 𝑗 ) ∈ A(J) such that 𝑛 = 𝑏 𝑗 𝑓 𝑗 .
Í Í
Then,
Õ Õ Õ Õ
𝑚⊗𝑛= 𝑎𝑖 𝑒𝑖 ⊗ 𝑏 𝑗 𝑓𝑗 = (𝑎 𝑖 𝑒 𝑖 ) ⊗ (𝑏 𝑗 𝑓 𝑗 ) = 𝑎𝑖 𝑏 𝑗 𝑒𝑖 ⊗ 𝑓𝑗 ,



𝑖 𝑗 𝑖,𝑗 𝑖,𝑗

hence the claim.

Let us now assume that the families (𝑒 𝑖 ) and ( 𝑓 𝑗 ) are bases. Since we
already shew that the family (𝑒 𝑖 ⊗ 𝑓 𝑗 ) generates M ⊗A N, we just need
to show that it is free. To that aim, let (𝑎 𝑖,𝑗 ) be an almost-null family of
elements of A, indexed by I × J, such that 𝑖,𝑗 𝑎 𝑖,𝑗 𝑒 𝑖 ⊗ 𝑓 𝑗 = 0. This implies
Í

Õ ©Õ
𝑒𝑖 ⊗ ­ 𝑎 𝑖,𝑗 𝑓 𝑗 ® = 0.
ª
𝑖∈I « 𝑗∈I ¬
Let (𝑒 𝑖∗ ) be the “dual basis” of the basis (𝑒 𝑖 ) and let ( 𝑓 𝑗∗ ) be the “dual basis”
of the basis ( 𝑓 𝑗 ) (they do not form bases unless I is finite and J is finite).
Let (𝑝, 𝑞) ∈ I × J; the map from M × N to A given by (𝑚, 𝑛) ↦→ 𝑒 𝑝∗ (𝑚)𝑒 𝑞∗ (𝑛)
is A-bilinear, hence there exists a unique morphism of A-modules from
M ⊗A N to A which maps 𝑚 ⊗ 𝑛 to 𝑒 𝑝∗ (𝑚)𝑒 𝑞∗ (𝑛) for every (𝑚, 𝑛) ∈ M × N.
Let us apply this morphism to our linear combination 𝑖,𝑗 𝑎 𝑖,𝑗 𝑒 𝑖 ⊗ 𝑓 𝑗 . We
Í
get
Õ
0= 𝑎 𝑖,𝑗 𝑒 𝑝∗ (𝑒 𝑖 )𝑒 𝑞∗ (𝑒 𝑗 ) = 𝑎 𝑝,𝑞 .
𝑖∈I
𝑗∈J

Consequently, 𝑎 𝑖,𝑗 = 0 for every (𝑖, 𝑗) ∈ I × J, so that the family (𝑒 𝑖 ⊗ 𝑓 𝑗 ) is

free, as we needed to show. 

Example (8.2.2). — Let A be a commutative ring, let S be a multiplicative

subset of A and let M be an A-module.
 8.2. TENSOR PRODUCT OF MODULES OVER A COMMUTATIVE RING 391

The map 𝑚 ↦→ 𝑚 ⊗ 1 from M to M ⊗A S−1 A is a morphism of A-modules.

Since M ⊗A S−1 A is an S−1 A-module, it extends to a morphism of S−1 A-
modules, 𝜑 : S−1 M → M ⊗A S−1 A given, explicitly, by 𝑚/𝑠 ↦→ 𝑚 ⊗ (1/𝑠)
for 𝑚 ∈ M and 𝑠 ∈ S. Let us show that 𝜑 is an isomorphism.
Let 𝑔 : M ×A S−1 A → S−1 M be the map given by 𝑔(𝑚, 𝑎/𝑠) = 𝑎𝑚/𝑠.
One checks that is well defined, that is, if 𝑎/𝑠 = 𝑏/𝑡, then 𝑎𝑚/𝑠 = 𝑏𝑚/𝑡,
and A-bilinear. Consequently, there exists a unique morphism of A-
modules, 𝜓 : M ⊗A S−1 A → S−1 M, such that 𝜓(𝑚 ⊗ 𝑎/𝑠) = 𝑎𝑚/𝑠 for all
𝑎 ∈ A, 𝑠 ∈ S and 𝑚 ∈ M.
One has
𝜑 ◦ 𝜓(𝑚 ⊗ 𝑎/𝑠) = 𝜑(𝑎𝑚/𝑠) = 𝑎𝑚 ⊗ (1/𝑠) = 𝑚 ⊗ 𝑎/𝑠,
so that 𝜑 ◦ 𝜓 is the identity morphism. Similarly,
𝜓 ◦ 𝜑(𝑚/𝑠) = 𝜓(𝑚 ⊗ (1/𝑠)) = 𝑚/𝑠,
hence 𝜓 ◦ 𝜑 is the identity morphism as well. Consequently, 𝜑 and 𝜓
are isomorphisms of A-modules, inverse one of the other.

8.2.3. — Let M be an A-module, let M∨ = HomA (M, A) be its dual. Let

N be an A-module.
Let 𝑑 : M∨ × N → HomA (M, N) be the map such for every 𝜑 ∈ M∨
and every 𝑛 ∈ N, 𝑑(𝜑, 𝑛) is the morphism 𝑥 ↦→ 𝜑(𝑥)𝑛 from M to N. It is
A-bilinear. By the universal property of the tensor product, there exists a
unique morphism of A-modules 𝛿 M,N : M∨ ⊗A N → HomA (M, N) such
that 𝛿 M,N (𝜑 ⊗ 𝑛) = 𝑑(𝜑, 𝑛) for every 𝜑 ∈ M∨ and every 𝑛 ∈ N.

Proposition (8.2.4). — Let M be an A-module.

a) If M is finitely generated and projective, then the morphism 𝛿 M,N : M∨ ⊗A
N → HomA (M, N) is an isomorphism for every A-module N.
b) If the morphism 𝛿 M,M : M∨ ⊗A M → EndA (M) is surjective, then M is
finitely generated and projective.

Proof. — a) Let (𝑒 𝑖 )16𝑖6𝑟 be a finite generating family of M, let 𝑓 : A𝑟 → M

be the surjective morphism given by 𝑓 (𝑎1 , . . . , 𝑎 𝑛 )) = 𝑟𝑖=1 𝑎 𝑖 𝑒 𝑖 . Since M
Í
is projective, there exists a morphism 𝑔 : M → A𝑟 such that 𝑓 ◦ 𝑔 = idM .
For 𝑖 ∈ {1, . . . , 𝑟}, let 𝜑 𝑖 ∈ M∨ be the composition of 𝑔 with the 𝑖th
392 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

projection A𝑟 → A. By construction, one has 𝑚 = 𝑟𝑖=1 𝜑 𝑖 (𝑚)𝑒 𝑖 for every

Í
𝑚 ∈ M. In other words, idM = 𝛿 M,M ( 𝑟𝑖=1 𝜑 𝑖 ⊗ 𝑒 𝑖 ),
Í
Let N be an A-module, let 𝑢 : M → N be a morphism of A-modules.
Let 𝜉 = 𝑟𝑖=1 𝜑 𝑖 ⊗ 𝑢(𝑒 𝑖 ). By definition, the morphism 𝛿 M,N (𝜉) sends
Í
an element 𝑚 ∈ M to 𝑟𝑖=1 𝜑 𝑖 (𝑚)𝑢(𝑒 𝑖 ) = 𝑢( 𝑟𝑖=1 𝜑 𝑖 (𝑚)𝑒 𝑖 ) = 𝑢(𝑚), hence
Í Í
𝑢 = 𝛿 M,N (𝜉). This proves that 𝛿 M,N is surjective.
Let then 𝜆 : HomA (M, N) → M∨ ⊗ N be the map given by 𝜆(𝑢) =
Í𝑟
𝑖=1 𝜑 𝑖 ⊗ 𝑢(𝑒 𝑖 ). It is A-linear and the previous computations show
that 𝛿 M,N ◦ 𝜆 = idHomA (M,N) . Conversely, let 𝜑 ∈ M∨ and 𝑛 ∈ N; since
𝛿M,N (𝜑 ⊗ 𝑛)(𝑒 𝑖 ) = 𝜑(𝑒 𝑖 )𝑛, we have
𝑟
Õ 𝑟
Õ
𝜆 ◦ 𝛿 M,N (𝜑 ⊗ 𝑛) = 𝜑 𝑖 ⊗ 𝜑(𝑒 𝑖 )𝑛 = 𝜑(𝑒 𝑖 )𝜑 𝑖 ⊗ 𝑛 = 𝜑0 ⊗ 𝑛,
𝑖=1 𝑖=1
Í𝑟
where 𝜑0 = 𝑖=1 𝜑(𝑒 𝑖 )𝜑 𝑖 . For every 𝑖 ∈ {1, . . . , 𝑟}, one has
𝑟
Õ 𝑟
Õ
𝜑0(𝑒 𝑖 ) = 𝜑(𝑒 𝑖 )𝜑 𝑖 (𝑒 𝑖 ) = 𝜑( 𝜑 𝑖 (𝑒 𝑖 )𝑒 𝑖 ) = 𝜑(𝑒 𝑖 ),
𝑖=1 𝑖=1

so that 𝜑0 = 𝜑. This proves that 𝜆 ◦ 𝛿 M,N (𝜑 ⊗ 𝑛) = 𝜑 ⊗ 𝑛. Consequently,

𝜆 ◦ 𝛿 M,N coincides with the identity morphism on split tensors, hence
on the whole of M∨ ⊗A N. In particular, the morphism 𝛿 M,N is injective.
We thus have shown that 𝛿 M,N is an isomorphism.
b) Assume that 𝛿 M,M is surjective and let 𝜉 ∈ M∨ ⊗ M be such that
𝛿M,M (𝜉) = idM ; write 𝜉 = 𝑟𝑖=1 𝜑 𝑖 ⊗ 𝑒 𝑖 , with 𝜑1 , . . . , 𝜑𝑟 ∈ M∨ and
Í
𝑒1 , . . . , 𝑒 𝑟 ∈ M.
Let 𝑚 ∈ M. The equality 𝑚 = idM (𝑚) = 𝛿 M,M (𝜉) rewrites as
𝑟
Õ
𝑚= 𝜑 𝑖 (𝑚)𝑒 𝑖 .
𝑖=1

This proves that (𝑒1 , . . . , 𝑒 𝑟 ) generates M, which is thus finitely generated.

Let 𝑓 : A𝑟 → M be the morphism given by 𝑓 (𝑎1 , . . . , 𝑎 𝑟 ) = 𝑟𝑖=1 𝑎 𝑖 𝑒 𝑖 . Let
Í
𝑔 : M → A𝑟 be the morphism given by 𝑔(𝑚) = (𝜑1 (𝑚), . . . , 𝜑𝑟 (𝑚)). One
has 𝑓 ◦ 𝑔 = idM . This implies that M is isomorphic to the submodule 𝑔(M)
of A𝑟 , and 𝑔(M) has a direct summand. In particular, M is projective. 
 8.2. TENSOR PRODUCT OF MODULES OVER A COMMUTATIVE RING 393

8.2.5. — Let M be an A-module, let M∨ = HomA (M, A). Consider the

morphism 𝛿 M : M∨ ⊗A M → EndA (M) of §8.2.3.
Let 𝑡 : M∨ × M → A be the map given by 𝑡(𝜑, 𝑚) = 𝜑(𝑚) for 𝜑 ∈ M∨
and 𝑚 ∈ M. Again, it is A-bilinear. Consequently, there exists a unique
linear form 𝜏M on M∨ ⊗A M such that 𝑡(𝜑 ⊗ 𝑚) = 𝜑(𝑚) for every 𝜑 ∈ M∨
and every 𝑚 ∈ M.

Example (8.2.6). — Let us assume that M is free and finitely generated. In

this case, proposition 8.2.4 asserts that 𝛿 M,N : M∨ ⊗A N → HomA (M, N)
is an isomorphism for every A-module N.
Let (𝑒1 , . . . , 𝑒 𝑟 ) be a basis of M and let (𝑒1∗ , . . . , 𝑒 𝑟∗ ) be its dual basis. Take
N to be free and finitely generated and let ( 𝑓 𝑗 )16𝑗6𝑠 be a basis of N. Then
the family (𝑒 𝑖∗ ⊗ 𝑓 𝑗 )16𝑖6𝑟 is a basis if M∨ ⊗A N. Let 𝜉 = 𝑟𝑖=1 𝑠𝑗=1 𝑎 𝑖,𝑗 𝑒 𝑖∗ ⊗ 𝑓 𝑗
Í Í
16𝑗6𝑠
be any element of M∨ ⊗A N and let 𝑢 = 𝛿M (𝜉). By definition, 𝑢(𝑒 𝑖 ) =
Í𝑠
𝑗=1 𝑎 𝑖,𝑗 𝑓 𝑗 , so that U = (𝑎 𝑖,𝑗 ) ∈ M 𝑠,𝑟 (A) is the matrix of 𝑢 in the bases
(𝑒1 , . . . , 𝑒 𝑟 ) of M and ( 𝑓1 , . . . , 𝑓𝑠 ) of N.
Now assume that M = N. Then
𝑛
Õ 𝑛
Õ
𝜏M (𝑢) = 𝑎 𝑖,𝑗 𝑒 𝑖∗ (𝑒 𝑗 ) = 𝑎 𝑖,𝑖 = Tr(U),
𝑖,𝑗=1 𝑖=1

and the linear form 𝜏M : M∨ ⊗A M → A corresponds to the trace of an

endomorphism of M.

8.2.7. Tensor product of algebras. — Let 𝑘 be a commutative ring, let

A and B be (non necessarily commutative) 𝑘-algebras. Let us show
how to endow the 𝑘-module A ⊗ 𝑘 B with the structure of a 𝑘-algebra.
For 𝑥 ∈ A (or B), write 𝜆 𝑥 for the endomorphisms of A (or B) given
by left-multiplication by 𝑥. Let (𝑎, 𝑏) ∈ A × B. The endomorphism
𝜆 𝑎 ⊗ 𝜆𝑏 of A ⊗ 𝑘 B is 𝑘-linear. Moreover, the map (𝑎, 𝑏) ↦→ End 𝑘 (A ⊗ 𝑘 B) is
𝑘-bilinear since the images of 𝑘 in A and B are central. By the universal
property of the tensor product, this gives a canonical 𝑘-linear morphism
from A ⊗ 𝑘 B to End 𝑘 (A ⊗ 𝑘 B), denoted by 𝑡 ↦→ 𝜆𝑡 , which maps a split
tensor 𝑎 ⊗ 𝑏 to 𝜆 𝑎 ⊗ 𝑘 𝜆𝑏 . This morphism then furnishes a 𝑘-bilinear
morphism 𝜇 from (A ⊗ 𝑘 B) × (A ⊗ 𝑘 B) to A ⊗ 𝑘 B given by (𝜉, 𝜂) ↦→ 𝜆𝜉 (𝜂).
394 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Observe that

M(𝑎 ⊗ 𝑏, 𝑎 0 ⊗ 𝑏 0) = 𝜆 𝑎⊗𝑏 (𝑎 0 ⊗ 𝑏 0) = (𝜆 𝑎 ⊗ 𝜆𝑏 )(𝑎 0 ⊗ 𝑏 0) = (𝑎𝑎 0) ⊗ 𝑏𝑏 0.

From that point on, it is straightforward (but a bit tedious) to prove that
the composition law defined by M is associative, that 1 ⊗ 1 is its unit
element, and that it is commutative if A and B are commutative. This
endowes A ⊗ 𝑘 B with the structure of a 𝑘-algebra.

8.3. Tensor algebras, symmetric and exterior algebra

In this section, A is a commutative ring.
The tensor product of two modules represents bilinear maps; we first
generalize the construction for multilinear maps on products of more
than two modules.

Lemma (8.3.1). — Let (M𝑖 )𝑖∈I be a family of A-modules. There exists an

A-module T and a multilinear map 𝜃 :
Î
𝑖 M 𝑖 → T that possesses the fol-
lowing universal property: for every A-module N and every multilinear map
𝑓 : M𝑖 → P, there exists a unique morphism of A-modules 𝜑 : T → P such
Î
that 𝜑 ◦ 𝜃 = 𝑓 .
Î
Proof. — Let F be the free A-module with basis 𝑖 M𝑖 . For any element
𝑚 = (𝑚 𝑖 )𝑖∈I of 𝑖 M𝑖 , one writes 𝑒 𝑚 for the corresponding basis element
Î
of F. One then defines T as the quotient of the free A-module F by the
submodule R generated by the following elements:
– the elements 𝑒 𝑚 − 𝑒 𝑚0 − 𝑒 𝑚00 whenever 𝑚 = (𝑚 𝑖 ), 𝑚 0 = (𝑚 𝑖0), 𝑚 00 = (𝑚 𝑖00)
are three elements of 𝑖 M𝑖 for which there exists 𝑗 ∈ I such that
Î
𝑚 𝑖 = 𝑚 𝑖0 = 𝑚 𝑖00 if 𝑖 ≠ 𝑗, and 𝑚 𝑗 = 𝑚 0𝑗 + 𝑚 00𝑗 ;
– the elements 𝑎𝑒 𝑚 − 𝑒 𝑚0 whenever 𝑎 ∈ A and 𝑚 = (𝑚 𝑖 ), 𝑚 0 = (𝑚 𝑖0) are
two elements of 𝑖 M𝑖 for which there exists 𝑗 ∈ I such that 𝑚 𝑖 = 𝑚 𝑖0 if
Î
𝑖 ≠ 𝑗, and 𝑚 0𝑗 = 𝑎𝑚 𝑗 .
For 𝑚 ∈ 𝑖 M𝑖 , let 𝜃(𝑚) be the class of 𝑒 𝑚 in T. The map 𝜃 is multilinear,
Î
because we precisely modded out by all the necessary elements. Observe
that the elements of the form 𝑒 𝑚 generate F, hence the image of 𝜃
generates T.
 8.3. TENSOR ALGEBRAS, SYMMETRIC AND EXTERIOR ALGEBRA 395

Let 𝑓 : 𝑖 M 𝑖 → N be any multilinear map, let 𝑓0 : F Î

Î
→ N be the
unique morphism such that 𝑓0 (𝑒 𝑚 ) = 𝑓 (𝑚), for every 𝑚 ∈ 𝑖 M𝑖 . Since
𝑓 is multilinear, the submodule R is contained in Ker( 𝑓0 ); consequently,
there exists a morphism 𝜑 : T → N such that 𝜑(𝜃(𝑚)) = 𝑓0 (𝑒 𝑚 ) = 𝑓 (𝑚)
for every 𝑚 ∈ 𝑖 M𝑖 . In other words, 𝜑 ◦ 𝜃 = 𝑓 .
Î
If 𝜑 and 𝜑0 are two morphisms such that 𝑓 = 𝜑 ◦ 𝜃 = 𝜑0 ◦ 𝜃, then
𝜑 − 𝜑0 vanishes on the image of 𝜃, hence on the submodule generated
by 𝜃 which is T. Consequently, 𝜑 = 𝜑0. 
The A-module T isË called the tensor product of the family (M𝑖 ) of A-
modules; it is denoted 𝑖∈I M 𝑖 . When I = {1, . . . , 𝑛}, one writes it rather
M1 ⊗ · · · ⊗ M𝑛 . For any element 𝑚 = (𝑚 𝑖 ) ∈ M𝑖 , the element 𝜃(𝑚) of T
Î
is written ⊗𝑖 𝑚 𝑖 , or 𝑚1 ⊗ · · · ⊗ 𝑚𝑛 if I = {1, . . . , 𝑛}.

Remark (8.3.2). — As any solution of a universal problem, the pair

(𝜃, T) is uniquely characterized by this universal property. Let indeed
(𝜃0 , T0) be a pair consisting of an A-module T0 and of a multilinear
map 𝜃0 : M𝑖 → T0 satisfying the universal property. First of all, by
Î
the universal property applied to (𝜃, T) and to the multilinear map 𝜃0,
there exists a morphism of A-modules 𝑓 : T → T0 such that 𝜃0 = 𝑓 ◦ 𝜃.
Similarly, there exists a morphism of A-modules 𝑓 0 : T0 → T such that
𝜃 = 𝑓 0 ◦ 𝜃0. Then, one has 𝜃 = ( 𝑓 0 ◦ 𝑓 ) ◦ 𝜃; by the universal property
of (𝜃, T), idT is the unique morphism 𝑔 : T → T such that 𝜃 = 𝑔 ◦ 𝜃;
consequently, 𝑓 0 ◦ 𝑓 = idT . Reversing the roles of T and T0, we also have
𝑓 ◦ 𝑓 0 = idT0 . This implies that 𝑓 is an isomorphism.

Remark (8.3.3). — When I = {1, . . . , 𝑛}, one can also construct the tensor
product M1 ⊗ · · · ⊗ M𝑛 by induction. More precisely, let M1 , . . . , M𝑛 be
A-modules. Let P be a way of parenthesizing the product 𝑥1 . . . 𝑥 𝑛 . They
are defined by induction using the following rules:
– If 𝑛 = 1, then 𝑥 1 has a unique parenthesizing, 𝑥 1 ;
– If 𝑛 > 2, one obtains the parenthesizings of 𝑥1 . . . 𝑥 𝑛 by choosing
𝑗 ∈ {1, . . . , 𝑛−1}, and P = (P0)(P00) where P0 is a parenthesizing of 𝑥 1 . . . 𝑥 𝑗
and P00 is a parenthesizing of 𝑥 𝑗+1 . . . 𝑥 𝑛 .
For example, 𝑥 1 𝑥2 has a unique parenthesizing, namely (𝑥 1 )(𝑥 2 ); for
𝑛 = 5, ((𝑥 1 )(𝑥2 ))(((𝑥 3 )(𝑥 4 ))(𝑥5 )) is a parenthesizing of 𝑥 1 𝑥2 𝑥 3 𝑥4 𝑥5 .
396 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

To any such parenthesizing, one can define by induction a module MP

and a multilinear map 𝜃P : M𝑖 → MP . When 𝑛 = 1, one sets MP = M1
Î
and 𝜃P = id. When 𝑛 > 2 and P = (P0)(P00), one defines MP = MP0 ⊗A MP00
and 𝜃P = 𝜃P0 ⊗ 𝜃P00 . For exemple, the previous parenthesizing P of
𝑥1 𝑥 2 𝑥 3 𝑥4 𝑥 5 gives rise to
MP = (M1 ⊗ M2 ) ⊗ ((M3 ⊗ M4 ) ⊗ M5 )
and
𝜃P (𝑚1 , 𝑚2 , 𝑚3 , 𝑚4 , 𝑚5 ) = (𝑚1 ⊗ 𝑚2 ) ⊗ ((𝑚3 ⊗ 𝑚4 ) ⊗ 𝑚5 ).
However, a map on M1 × · · · × M𝑛 is multilinear if and only if it is mul-
tilinear with respect to both the first group of variables (𝑚1 , . . . , 𝑚 𝑗 ) and
the second group (𝑚 𝑗+1 , . . . , 𝑚𝑛 ). By induction, we see that the map 𝜃P is
multilinear hence that the module MP is also a solution of the universal
property.
Ë𝑛 Consequently, there exists a unique morphism of A-modules
from 𝑖=1 M𝑖 to MP which maps 𝜃(𝑚1 , . . . , 𝑚 𝑛 ) to 𝜃P (𝑚1 , . . . , 𝑚 𝑛 ).
In the sequel, we shall neglect these various “associativity” isomor-
phisms.

Proposition (8.3.4). — Let M1 , . . . , M𝑛 be A-modules. For every

𝑖
𝑖 ∈ {1, . . . , 𝑛}, let (𝑒 𝑗 ) 𝑗∈J𝑖 be a family of elements of M𝑖 . Let J = J1 × · · · × J𝑛 ;
for every 𝑗 = (𝑗1 , . . . , 𝑗𝑛 ) ∈ J, let 𝑒 𝑗 = 𝑒 1𝑗1 ⊗ · · · ⊗ 𝑒 𝑗𝑛𝑛 .
a) If for each 𝑖, (𝑒 𝑗𝑖 ) 𝑗∈J𝑖 generates M𝑖 , then the family (𝑒 𝑗 ) 𝑗∈J generates the
tensor product M1 ⊗ · · · ⊗ M𝑛 .
b) If for each 𝑖, (𝑒 𝑗𝑖 ) 𝑗∈J𝑖 is a basis of M𝑖 , then the family (𝑒 𝑗 ) 𝑗∈J is a basis of the
tensor product M1 ⊗ · · · ⊗ M𝑛 .

Proof. — The case 𝑛 = 1 is trivial, the case 𝑛 = 2 is proposition 8.2.1. The

general case follows by induction thanks to the step by step construction
of M1 ⊗ · · · ⊗ M𝑛 . 

8.3.5. The tensor algebra. — Let M be an A-module. Set M0 = A and,

for any positive integer 𝑛, set M𝑛 be the tensor product of 𝑛 copies of the
A-module M. Modulo the associativity isomorphisms, we have M𝑛 =
M𝑝 ⊗A M𝑞 whenever 𝑝, 𝑞, 𝑛 are nonnegative integers such that 𝑝 + 𝑞 = 𝑛.
The map (𝑚1 , . . . , 𝑚𝑛 ) ↦→ 𝑚1 ⊗ · · · ⊗ 𝑚𝑛 from M𝑛 to M𝑛 is 𝑛-linear and the
 8.3. TENSOR ALGEBRAS, SYMMETRIC AND EXTERIOR ALGEBRA 397

module M𝑛 satisfies the following universal property: for every 𝑛-linear

map 𝑓 from M𝑛 to an A-module P, there exists a unique morphism of
A-modules, 𝜑 : M𝑛 → P, such that 𝜑(𝑚1 ⊗ · · · ⊗ 𝑚𝑛 ) = 𝑓 (𝑚1 , . . . , 𝑚𝑛 ) for
every (𝑚1 , . . . , 𝑚𝑛 ) ∈ M𝑛 .
If M is free and (𝑒 𝑖 )𝑖∈I is a basis of M, then for every 𝑛 ∈ N, the
A-module M𝑛 is free with basis the family (𝑒 𝑖1 ⊗ · · · ⊗ 𝑒 𝑖 𝑛 ) indexed by
(𝑖1 , . . . , 𝑖 𝑛 ) ∈ I𝑛 . This follows indeed from proposition 8.3.4.
Let T(M) be the direct sum of all A-modules M𝑛 , for 𝑛 ∈ N. The
submodule M𝑛 of T(M) will be written T𝑛 (M). An element of T𝑛 (M) will
be said to be of degree 𝑛.
The associativity isomorphisms M𝑝 ⊗ M𝑞 ' M𝑝+𝑞 furnish bilinear maps
T𝑝 (M) × T𝑞 (M) → T𝑝+𝑞 (M), for 𝑝, 𝑞 > 0. There is a unique structure
of an A-algebra on the A-module T(M) of which multiplication law is
given by these maps. The resulting algebra is called the tensor algebra of
the A-module M. It contains M = T1 (M) as a submodule. The elements 𝑚,
for 𝑚 ∈ M = T1 (M), generate T(M) as an A-algebra.
This algebra satisfies the following universal property: for every A-
algebra (assocative, with unit) B, and every morphism of A-modules 𝑓 : M → B,
there exists a unique morphism of A-algebras, 𝜑 : T(M) → B, such that
𝜑(𝑚) = 𝑓 (𝑚) for every 𝑚 ∈ M.
If M is free with basis (𝑒 𝑖 )𝑖∈I , then T(M) is a free A-module with
basis the disjoint union of the bases 1 ∈ T0 (M) and 𝑒 𝑖1 ⊗ · · · ⊗ 𝑒 𝑖 𝑛 for
(𝑖1 , . . . , 𝑖 𝑛 ) ∈ I𝑛 and 𝑛 > 1.

8.3.6. The symmetric algebra. — The tensor algebra T(M) is not com-
mutative in general. By definition, the symmetric algebra of the A-
module M is the quotient of T(M) by the two-sided ideal I generated
by elements of the form 𝑚1 ⊗ 𝑚2 − 𝑚2 ⊗ 𝑚1 , for 𝑚1 , 𝑚2 ∈ M. It is
denoted S(M).
For 𝑛 ∈ N, let I𝑛 = I ∩ T𝑛 (M); this is a subbmodule 𝑛
Éof T (M). The ideal I
is homogeneous in the sense that one has I = 𝑛 I𝑛 . The inclusion
Í (𝑛)
𝑚 , 𝑚 𝑚 𝑚1
É
𝑛 𝑛
I ⊂ I is obvious. Conversely, let 1 2 ∈ M and write 1 =
(𝑛) (𝑛)
and 𝑚2 = 𝑚2 , where 𝑚 𝑖 ∈ T𝑛 (M) for all 𝑛. Then
Í
Õ Õ  (𝑝) (𝑞) (𝑞) (𝑝)

𝑚1 ⊗ 𝑚2 − 𝑚2 ⊗ 𝑚1 = 𝑚1 ⊗ 𝑚2 − 𝑚1 ⊗ 𝑚2
𝑛∈N 𝑝+𝑞=𝑛
398 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS
É
belongs to 𝑛 I𝑛 ; this
É implies that the generators
É of I are contained in
the two sided ideal I𝑛 , hence that I = I𝑛 .
These computations also show that I0 = 0 and I1 = 0.
Write S𝑛 (M) Éfor𝑛the image of T (M)
𝑛 in S(M); one has S𝑛 (M) = T𝑛 (M)/I𝑛
and S(M) = S (M); moreover, S0 (M) = A and S1 (M) = M.
Observe that, by construction, all elements of S1 (M) pairwise commute,
and that S(M) is generated by S1 (M). Consequently, the algebra S(M) is
commutative.
The algebra S(M) is a commutative A-algebra together with a morphism
M → S(M) which satisfies the following universal property: For every
commutative A-algebra B and every morphism 𝑓 : M → B of A-modules,
there exists a unique morphism of A-algebras, 𝜑 : S(M) → B, such that
𝜑(𝑚) = 𝑓 (𝑚) for every 𝑚 ∈ M = S1 (M).
For every integer 𝑛, the A-module S𝑛 (M) is endowed with an 𝑛-
linear symmetric map M𝑛 → S𝑛 (M). It satisfies the following universal
property: For every A-module P and any 𝑛-linear symmetric map 𝑓 : M𝑛 → P,
there exists a unique morphism of A-modules 𝜑 : S𝑛 (M) → P such that
𝜑(𝑚1 . . . 𝑚𝑛 ) = 𝑓 (𝑚1 , . . . , 𝑚𝑛 ) for every 𝑚1 , . . . , 𝑚𝑛 ∈ M.
1

8.3.7. The exterior algebra. — As we have observed in section 3.8,

alternate multilinear maps are a very important object. Let us understand
them with the tool of the tensor product, analogously to the description
of symmetric multilinear maps from the the symmetric algebra.
Let J be the two-sided ideal of T(M) generated by all elements of the
form 𝑚 ⊗ É𝑚, for 𝑚 ∈ M. It is an “homogeneous ideal” of T(M), that is to
∞ 𝑛
É
say, J = 𝑛=0 J𝑛 , where J𝑛 = J ∩ T (M). Indeed, 𝑛 J𝑛 is a two-sided
ideal which contains all elements of the form 𝑚 ⊗ 𝑚, for 𝑚 ∈ M, hence
is equal to J. Moreover, J0 = 0 and J1 = 0.
Let Λ(M) be the quotient of the algebra É∞ T(M) by this two-sided
𝑛 𝑛
ideal J. As an A-module, one has Λ(M) = 𝑛=0 Λ (M), where Λ (M) =
T𝑛 (M)/J𝑛 . In particular, Λ0 (M) = A and Λ1 (M) = M. The multiplication
of Λ(M) is denoted with a symbol ∧. For example, one has 𝑚 ∧ 𝑚 = 0
for every 𝑚 ∈ M.

1Faire les démonstrations ? – cf le cas alterné et l’exercice 8/20. . .

 8.3. TENSOR ALGEBRAS, SYMMETRIC AND EXTERIOR ALGEBRA 399

For any 𝑚1 , 𝑚2 ∈ M, one has

𝑚1 ⊗ 𝑚2 + 𝑚2 ⊗ 𝑚1 = (𝑚1 + 𝑚2 ) ⊗ (𝑚1 + 𝑚2 ) − 𝑚1 ⊗ 𝑚1 − 𝑚2 ⊗ 𝑚2 ∈ J,

so that 𝑚1 ∧ 𝑚2 = −𝑚2 ∧ 𝑚1 . More generally, let 𝑚1 , . . . , 𝑚𝑛 ∈ M and let

𝜎 ∈ 𝔖𝑛 be any permutation of {1, . . . , 𝑛}. Then,

𝑚 𝜎(1) ∧ · · · ∧ 𝑚 𝜎(𝑛)) = 𝜀(𝜎)𝑚1 ∧ · · · ∧ 𝑚𝑛 ,

where 𝜀(𝜎) is the signature of 𝜎.

The A-algebra Λ(M) satisfies the following universal property: For
every A-algebra B and any morphism 𝑓 : M → B such that 𝑓 (𝑚) · 𝑓 (𝑚) =
0 for every 𝑚 ∈ M, there exists a unique morphism of A-algebras,
𝜑 : Λ(M) → B, such that 𝜑(𝑚) = 𝑓 (𝑚) for every 𝑚 ∈ M = Λ1 (M).
An element of Λ𝑝 (M) is called a 𝑝-vector. Any 𝑝-vector can be written
as a sum of split 𝑝-vectors, namely 𝑝-vectors of the form 𝑚1 ∧ · · · ∧ 𝑚 𝑝 ,
for 𝑚1 , . . . , 𝑚 𝑝 ∈ M. By 𝑝-linearity, we see that we can restrict ourselves
to vectors 𝑚 𝑖 in some generating family of M.
More precisely, let us assume that (𝑚 𝑖 )16𝑖6𝑛 generates M. Then
any 𝑝-vector in Λ𝑝 (M) is a linear combination of vectors of the form
𝑚 𝑖1 ∧ · · · ∧ 𝑚 𝑖 𝑝 , for 𝑖1 , . . . , 𝑖 𝑝 ∈ {1, . . . , 𝑛}. By the alternate property, we
can even assume that 𝑖1 6 · · · 6 𝑖 𝑝 . Then, if 𝑝 > 𝑛, we see that two
consecutive indices are equal hence 𝑚 𝑖1 ∧ · · · ∧ 𝑚 𝑖 𝑝 = 0 by the alternate
property. This shows that Λ𝑝 (M) = 0 for 𝑝 > 𝑛.

Proposition (8.3.8). — Let 𝑛 ∈ N. The A-module Λ𝑛 (M) is endowed with

an alternate 𝑛-linear map M𝑛 → Λ𝑛 (M), (𝑚1 , . . . , 𝑚𝑛 ) ↦→ 𝑚1 ∧ · · · ∧ 𝑚𝑛 .
It satisfied the following universal property: For every A-module P and
any 𝑛-linear alternate map 𝑓 : M𝑛 → P, there exists a unique morphism
𝜑 : Λ𝑛 (M) → P of A-modules such that 𝜑(𝑚1 ∧ · · · ∧ 𝑚𝑛 ) = 𝑓 (𝑚1 , . . . , 𝑚𝑛 ).

Proof. — The uniqueness of the morphism 𝜑 follows from the fact that
the products 𝑚1 ∧ · · · ∧ 𝑚𝑛 , for 𝑚1 , . . . , 𝑚𝑛 ∈ M, generate Λ𝑛 (M). Let
us show its existence. Let 𝜑1 : T𝑛 (M) → P the canonical morphism
of A-modules which is deduced from the multilinear map 𝑓 ; one has
𝜑1 (𝑚1 ⊗ · · · ⊗ 𝑚𝑛 ) = 𝑓 (𝑚1 , . . . , 𝑚𝑛 ) for every (𝑚1 , . . . , 𝑚𝑛 ) ∈ M𝑛 . Let
us show that 𝜑1 vanishes on the kernel J𝑛 of the canonical morphism
from T𝑛 (M) to Λ𝑛 (M).
400 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Let 𝑚1 , . . . , 𝑚𝑛 ∈ M. If there exists 𝑖 ∈ {1, . . . , 𝑛 − 1} such that

𝑚 𝑖 = 𝑚 𝑖+1 , then the tensor
𝑚1 ⊗ · · · ⊗ 𝑚𝑛 = (𝑚1 ⊗ · · · ⊗ 𝑚 𝑖−1 ) ⊗ (𝑚 𝑖 ⊗ 𝑚 𝑖 ) ⊗ (𝑚 𝑖+2 ⊗ . . . 𝑚𝑛 )
belongs to the ideal J, kernel of the canonical morphism of algebras
from T(M) to Λ(M). Moreover, when 𝑛 varies, as well as 𝑚1 , . . . , 𝑚𝑛 ∈ M,
we see that these tensors generate a two-sided ideal of T(M) containing
all elements of the form 𝑚 ⊗ 𝑚, hence is equal to J. In particular, any
element of J𝑛 = J ∩ T𝑛 (M) is a linear combination of such tensors.
Since 𝑓 is alternate, one has
𝜑1 (𝑚1 ⊗ · · · ⊗ 𝑚𝑛 ) = 𝑓 (𝑚1 , . . . , 𝑚𝑛 ) = 0
as soon as there exists 𝑖 ∈ {1, . . . , 𝑛 − 1} such that 𝑚 𝑖 = 𝑚 𝑖+1 . Conse-
quently, 𝜑1 vanishes on J𝑛 . This implies that there exists a morphism
of A-modules 𝜑 : Λ𝑛 (M) → P such that 𝜑(𝑚1 ∧· · ·∧𝑚𝑛 ) = 𝜑1 (𝑚1 ⊗. . . 𝑚𝑛 )
for every (𝑚1 , . . . , 𝑚𝑛 ) ∈ M𝑛 , that is, 𝜑(𝑚1 ∧ · · · ∧ 𝑚𝑛 ) = 𝑓 (𝑚1 , . . . , 𝑚𝑛 ).
This concludes the proof. 

Proposition (8.3.9). — Let M and N be A-modules, let 𝑢 : M → N be a

morphism. There exists a unique morphism of A-algebras, Λ(𝑢) : Λ(M) →
Λ(N), which coincides with 𝑢 on Λ1 (M). One has Λ(idM ) = idΛ(M) . If P is
another A-module and 𝑣 : N → P is a morphism, then one has Λ(𝑣 ◦ 𝑢) =
Λ(𝑣) ◦ Λ(𝑢).
Proof. — This follows at once from the universal property. Indeed, let
𝜃 : N → Λ(N) be the canonical injection; then, the map 𝜃 ◦ 𝑢 maps every
element 𝑚 of M to the element 𝑢(𝑚) of Λ1 (N), hence its square is zero.
Consequently, there exists a morphism of algebras from Λ(M) to Λ(N)
which coincides with 𝑢 on M ' Λ1 (M).
Morover, Λ(𝑣) ◦ Λ(𝑢) is a morphism of algebras from Λ(M) to Λ(P)
which coincides with 𝑣 ◦ 𝑢 on M, hence is equal to Λ(𝑣 ◦ 𝑢). 
Let 𝑛 ∈ N. By restriction to Λ𝑛 (M), the morphism of algebras Λ(𝑢)
defines a morphism Λ𝑛 (𝑢) : Λ𝑛 (M) → Λ𝑛 (M) of A-modules. It is
characterized by the formula
Λ𝑛 (𝑢)(𝑚1 ∧ . . . 𝑚𝑛 ) = 𝑢(𝑚1 ) ∧ · · · ∧ 𝑢(𝑚𝑛 )
for every (𝑚1 , . . . , 𝑚𝑛 ) ∈ M𝑛 .
 8.4. THE EXTERIOR ALGEBRA AND DETERMINANTS 401

8.4. The exterior algebra and determinants

Proposition (8.4.1). — Let M be a free A-module, let (𝑒 𝑖 )𝑖∈I be a basis of M and
let  be a total ordering on I. Then, the family of 𝑝-vectors 𝑒 𝑖1 ∧ . . . 𝑒 𝑖 𝑝 , where
(𝑖1 , . . . , 𝑖 𝑝 ) runs along the set of stricly increasing sequences of indices, is a
basis of Λ𝑝 (M). In particular, if M is finitely generated and 𝑛 is its rank, then
𝑛
Λ𝑝 (M) is free of rank 𝑝 .

Let J be a subset of I with 𝑝 elements, and let (𝑖 1 , . . . , 𝑖 𝑝 ) be the unique

strictly increasing tuple such that J = {𝑖1 , . . . , 𝑖 𝑝 }; we set 𝑒J = 𝑒 𝑖1 ∧ . . . 𝑒 𝑖 𝑝 .
Proof. — Any 𝑝-vector in M is a linear combination of 𝑝-vectors of the
form 𝑒 𝑖1 ∧· · ·∧ 𝑒 𝑖 𝑝 , where 𝑖1 , . . . , 𝑖 𝑝 belong to I. Using the relations 𝑒 𝑖 ∧ 𝑒 𝑗 =
−𝑒 𝑗 ∧ 𝑒 𝑖 for 𝑖 ≠ 𝑗, as well as 𝑒 𝑖 ∧ 𝑒 𝑖 = 0, we may only consider families
(𝑖1 , . . . , 𝑖 𝑝 ) such that 𝑖1 < · · · < 𝑖 𝑝 . This shows that the elements 𝑒J , for
all 𝑝-subsets J of I, generate Λ𝑝 (M).
Let us show that these elements actually form a basis of Λ𝑝 (M) and let
us consider a linear dependence relation J 𝑎J 𝑒J = 0 among them. The
Í
A-module T𝑝 (M) is free and the family (𝑒 𝑖1 ⊗ · · · ⊗ 𝑒 𝑖 𝑝 ), for (𝑖1 , . . . , 𝑖 𝑝 ) ∈ I𝑝 ,
is a basis. Let us fix a subset J ⊂ {1, . . . , 𝑛} with cardinality 𝑝; let us
write J = {𝑗1 , . . . , 𝑗𝑝 } where 𝑗1 < · · · < 𝑗𝑝 . Let 𝑓J : M𝑛 be the unique
𝑝-linear form on M which maps (𝑒 𝑖1 , . . . , 𝑒 𝑖 𝑝 ) to 0 if {𝑖 1 , . . . , 𝑖 𝑝 } ≠ J and
(𝑒 𝑖1 , . . . , 𝑒 𝑖 𝑝 ) to the signature of the permutation which maps 𝑖 𝑘 to 𝑗 𝑘 , for
any 𝑘 ∈ {1, . . . , 𝑝}. It is alternate, hence there exists a unique morphism
of A-modules 𝜑J : Λ𝑝 (M) → A which maps 𝑒I to 0 if I ≠ J and 𝑒J to 1.
Let us apply 𝜑J to the dependence relation J 𝑎J 𝑒J = 0; we get 𝑎J = 0.
Í
Consequently, the given linear dependence relation is trivial, hence the
family (𝑒J ) is free. It is thus a basis of Λ𝑝 (M). 

Lemma (8.4.2). — Let M be an A-module and let 𝑓 : M𝑛−1 → A be an

alternate (𝑛 − 1)-linear form. The map 𝑓 0 : M𝑛 → M given by
𝑛
Õ
0
𝑓 (𝑥1 , . . . , 𝑥 𝑛 ) = (−1)𝑝 𝑓 (𝑥 1 , . . . , 𝑥b𝑝 , . . . , 𝑥 𝑛 )𝑥 𝑝 ,
𝑝=1

where the 𝑥b𝑝 means that this element is omitted from the tuple, is 𝑛-linear and
alternate.
402 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Proof. — It is obvious that 𝑓 0 is linear with respect to each variable.

Moreover, if 𝑥 𝑖 = 𝑥 𝑖+1 , the terms 𝑓 (𝑥 1 , . . . , 𝑥b𝑝 , . . . , 𝑥 𝑛 )𝑥 𝑝 are 0 whenever
𝑝 ≠ 𝑖 and 𝑝 ≠ 𝑖 + 1, and the terms with indices 𝑖 and 𝑖 + 1 are opposite
one of the other. This proves the lemma. 

Proposition (8.4.3). — Let M be a free finitely generated A-module and let

𝑥1 , . . . , 𝑥 𝑛 be elements of M. For the elements 𝑥1 , . . . , 𝑥 𝑛 to be linearly
dependent, it is necessary and sufficient that there exists 𝜆 ∈ A {0} such that
𝜆𝑥 1 ∧ · · · ∧ 𝑥 𝑛 = 0.
Proof. — Assume that 𝑥1 , . . . , 𝑥 𝑛 are linearly dependent and let 𝑎1 𝑥1 +
· · · + 𝑎 𝑛 𝑥 𝑛 = 0 be a nontrivial linear dependence relation. To fix the ideas,
assume that 𝑎1 ≠ 0. Then, 𝑎1 𝑥 1 is a linear combination of 𝑥2 , . . . , 𝑥 𝑛 , so
that (𝑎1 𝑥1 ) ∧ 𝑥 2 ∧ · · · ∧ 𝑥 𝑛 = 0; this implies 𝑎1 𝑥1 ∧ · · · ∧ 𝑥 𝑛 = 0.
Conversely, let us show by induction on 𝑛 that if 𝜆𝑥 1 ∧ · · · ∧ 𝑥 𝑛 = 0,
for some nonzero 𝜆 ∈ A, then 𝑥 1 , . . . , 𝑥 𝑛 are linearly dependent.
This holds when 𝑛 = 0 (there is nothing to prove) and when 𝑛 = 1
(by definition). So, let us assume that this assertion holds up to 𝑛 − 1
and assume that 𝜆𝑥1 ∧ · · · ∧ 𝑥 𝑛 = 0, where 𝜆 ≠ 0. If 𝑥2 , . . . , 𝑥 𝑛 are
linearly dependent, then 𝑥 1 , . . . , 𝑥 𝑛 are also linearly dependent too, so
that we may assume that 𝑥 2 , . . . , 𝑥 𝑛 are linearly independent and that
𝑥2 ∧ · · · ∧ 𝑥 𝑛 ≠ 0. Since Λ𝑛−1 (M) is a free A-module, there exists a
linear form 𝜑 on Λ𝑛−1 (M) such that 𝜇 = 𝜑(𝑥 2 ∧ · · · ∧ 𝑥 𝑛 ) ≠ 0. In other
words, the map 𝑓 : M𝑛−1 → A given by (𝑣2 , . . . , 𝑣 𝑛 ) ↦→ 𝜑(𝑣2 ∧ · · · ∧ 𝑣 𝑛 )
is an alternate (𝑛 − 1)-linear form on M and 𝑓 (𝑥 2 , . . . , 𝑥 𝑛 ) = 𝜇 ≠ 0. Let
𝑓 0 : M𝑛 → M be the alternate 𝑛-linear map defined in lemma 8.4.2 above.
Since 𝜆𝑥1 ∧ · · · ∧ 𝑥 𝑛 = 0, one has 𝑓 0(𝜆𝑥1 , 𝑥2 , . . . , 𝑥 𝑛 ) = 0, hence a relation
𝑛
Õ
−𝜆 𝑓 (𝑥2 , . . . , 𝑥 𝑛 )𝑥1 + 𝜆 𝑓 (𝑥 1 , . . . , 𝑥b𝑝 , . . . , 𝑥 𝑛 )𝑥 𝑝 = 0.
𝑝=2

Since 𝜆 𝑓 (𝑥 2 , . . . , 𝑥 𝑛 ) ≠ 0, the vectors 𝑥 1 , . . . , 𝑥 𝑛 are linearly dependent,

hence the proposition. 

Corollary (8.4.4). — Let A be a nonzero commutative ring, let M be a free

A-module, let 𝑛 > 1 be an integer, and let 𝑒1 , . . . , 𝑒 𝑛 be elements of M. Then
(𝑒1 , . . . , 𝑒 𝑛 ) is a basis of M if and only if 𝑒1 ∧ · · · ∧ 𝑒 𝑛 is a basis of Λ𝑛 (M).
 8.4. THE EXTERIOR ALGEBRA AND DETERMINANTS 403

Proof. — If 𝑒1 , . . . , 𝑒 𝑛 is a basis of M, we proved in proposition 8.4.1 that

𝑒1 ∧ · · · ∧ 𝑒 𝑛 is a basis of Λ𝑛 (M).
Conversely, let us assume that 𝑒1 ∧ · · · ∧ 𝑒 𝑛 be a basis of Λ𝑛 (M). By
the preceding proposition, the family (𝑒1 , . . . , 𝑒 𝑛 ) is free. It then follows
from proposition 8.4.1 that M is finitely generated, that its rank is equal
to 𝑛, that Λ𝑛−1 (M) ≠ 0 and that Λ𝑛+1 (M) = 0. Let 𝑓 be the map from M𝑛
to A such that 𝑓 (𝑥 1 , . . . , 𝑥 𝑛 ) is the unique element 𝜆 ∈ A such that
𝑥 1 ∧ · · · ∧ 𝑥 𝑛 = 𝜆𝑒1 ∧ · · · ∧ 𝑒 𝑛 , for any 𝑥1 , . . . , 𝑥 𝑛 ∈ M. The map 𝑓 is
𝑛-linear and alternate (it is the composition of the 𝑛-linear alternate map
M𝑛 → Λ𝑛 (M) given by (𝑥1 , . . . , 𝑥 𝑛 ) ↦→ 𝑥 1 ∧ · · · ∧ 𝑥 𝑛 , and of the inverse
of the isomorphism A → Λ𝑛 (M), given by 𝑎 ↦→ 𝑎𝑒1 ∧ · · · ∧ 𝑒 𝑛 ). Since
Λ𝑛+1 (M) = 0, the map 𝑓 0 from M𝑛+1 to M defined in lemma 8.4.2 is zero,
because it is (𝑛 + 1)-linear and alternate. In particular, for every 𝑥 ∈ M,
𝑛
Õ
0
𝑓 (𝑥, 𝑒1 , . . . , 𝑒 𝑛 ) = − 𝑓 (𝑒1 , . . . , 𝑒 𝑛 )𝑥+ (−1)𝑝+1 𝑓 (𝑥, 𝑒1 , . . . , 𝑒b𝑝 , . . . , 𝑒 𝑛 )𝑒 𝑝 = 0.
𝑝=1

Since 𝑓 (𝑒1 , . . . , 𝑒 𝑛 ) = 1, one gets

𝑛
Õ
𝑥= (−1)𝑝+1 𝑓 (𝑥, 𝑒1 , . . . , 𝑒b𝑝 , . . . , 𝑒 𝑛 )𝑒 𝑝 = 0,
𝑝=1

which proves that the family (𝑒1 , . . . , 𝑒 𝑛 ) generates M. It is thus a basis

of M. 

Corollary (8.4.5). — Let A be a a nonzero commutative ring and let M be an

A-module which possesses a basis of cardinality 𝑛.
a) Every free family in M has at most 𝑛 elements.
b) Every generating family in M has at least 𝑛 elements.

Proof. — Let 𝑥 1 , . . . , 𝑥 𝑝 be elements of M. If (𝑥 1 , . . . , 𝑥 𝑝 ) is free, then

𝑥 1 ∧ · · · ∧ 𝑥 𝑝 ≠ 0, hence Λ𝑝 (M) ≠ 0 and 𝑝 6 𝑛. If (𝑥1 , . . . , 𝑥 𝑝 ) is generating,
then Λ 𝑘 (M) = 0 for 𝑘 > 𝑝, hence 𝑛 6 𝑝. 
Consequently, when A is a nonzero commutative ring, it is unam-
biguous to define the rank of a free finitely generated A-module as the
cardinality of any basis of M.
404 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Definition (8.4.6). — Let M be a free A-module of rank 𝑛 and let 𝑢 be an

endomorphism of M. The endomorphism Λ𝑛 (𝑢) of Λ𝑛 (M) is an homothety. Its
ratio is called the determinant of 𝑢 and written det(𝑢).

By the preceding proposition, Λ𝑛 (M) is isomorphic to A. Let 𝜀 be a basis

of Λ𝑛 (M) and let 𝛿 ∈ A by such that Λ𝑛 (𝑢)(𝜀) = 𝛿𝜀. For any 𝑥 ∈ Λ𝑛 (M),
let 𝑎 ∈ A be such that 𝑥 = 𝑎𝜀; then, Λ𝑛 (𝑥) = Λ𝑛 (𝑎𝜀) = 𝑎Λ𝑛 (𝜀) = 𝑎𝛿𝜀 = 𝛿𝑥.
This shows that Λ𝑛 (𝑢) is an homothety.
Let 𝑢 and 𝑣 be endomorphism of a free A-module of rank 𝑛. One has
Λ𝑛 (𝑣 ◦ 𝑢) = Λ𝑛 (𝑣) ◦ Λ𝑛 (𝑢), hence det(𝑣 ◦ 𝑢) = det(𝑣) det(𝑢). One also
has Λ𝑛 (idM ) = id, so that det(idM ) = 1. If 𝑢 is invertible, with inverse 𝑣,
then Λ𝑛 (𝑢) is invertible too, with inverse Λ𝑛 (𝑣). It follows that det(𝑢) is
an invertible element of A, with inverse det(𝑣). In other words,
det(𝑢 −1 ) = det(𝑢)−1 .

Proposition (8.4.7). — Let A be a nonzero commutative ring, let M be a free

A-module of rank 𝑛 and let 𝑢 be an endomorphism of M.
a) The morphism 𝑢 is an isomorphism if and only if 𝑢 is surjective, if and
only if det(𝑢) is invertible.
b) The morphism 𝑢 is injective if and only if det(𝑢) is regular in A.

Proof. — Let (𝑒1 , . . . , 𝑒 𝑛 ) be a basis of M.

If 𝑢 is an automorphism of M, we just proved that det(𝑢) is invertible
in A. More generally, let us assume that 𝑢 is surjective. Since M is
free, the morphism 𝑢 has a right inverse 𝑣. (For every 𝑖 ∈ {1, . . . , 𝑛},
let 𝑓𝑖 ∈ M be any element such that 𝑢( 𝑓𝑖 ) = 𝑒 𝑖 , and let 𝑣 be the unique
endomorphism of M such that 𝑣(𝑒 𝑖 ) = 𝑓𝑖 for every 𝑖.) One has 𝑢 ◦ 𝑣 = idM ,
hence det(𝑢) det(𝑣) = det(idM ) = 1. This shows that det(𝑢) is invertible.
Let us now prove that if det(𝑢) is invertible, then 𝑢 is an automorphism
of M. By definition, one has 𝑢(𝑒1 )∧· · ·∧𝑢(𝑒 𝑛 ) = det(𝑢)𝑒1 ∧· · ·∧𝑒 𝑛 . If det(𝑢)
is invertible, then 𝑢(𝑒1 ) ∧ · · · ∧ 𝑢(𝑒 𝑛 ) is a basis of Λ𝑛 (M). By corollary 8.4.4,
(𝑢(𝑒1 ), . . . , 𝑢(𝑒 𝑛 )) is a basis of M. Let 𝑣 be the unique endomorphism
of M such 𝑣(𝑢(𝑒 𝑖 )) = 𝑒 𝑖 for every 𝑖 ∈ {1, . . . , 𝑛}. One has 𝑣 ◦ 𝑢(𝑒 𝑖 ) = 𝑒 𝑖 for
every 𝑖, hence 𝑣 ◦ 𝑢 = id. Consequently, 𝑢 ◦ 𝑣(𝑢(𝑒 𝑖 )) = 𝑢(𝑒 𝑖 ) for every 𝑖;
since (𝑢(𝑒1 ), . . . , 𝑢(𝑒 𝑛 )) is a basis of M, this implies that 𝑢 ◦ 𝑣 = id. In
other words, 𝑢 is invertible, and 𝑣 is its inverse.
 8.4. THE EXTERIOR ALGEBRA AND DETERMINANTS 405

For the morphism 𝑢 to be injective, it is necessary and sufficient that the

family (𝑢(𝑒1 ), . . . , 𝑢(𝑒 𝑛 )) be free. If this holds, proposition 8.4.3 asserts
that for every 𝜆 ∈ A {0}, 𝜆𝑢(𝑒1 ) ∧ · · · ∧ 𝑢(𝑒 𝑛 ) ≠ 0. Since 𝑒1 ∧ · · · ∧ 𝑒 𝑛 is
a basis of Λ𝑛 (M), we see that 𝑢 is injective if and only if 𝜆 det(𝑢) ≠ 0 for
𝜆 ≠ 0, in other words, if and only if det(𝑢) is regular in A. 

8.4.8. Determinant of a matrix. — One defines the determinant of a

matrix U ∈ M𝑛 (A) as the determinant of the endomorphism 𝑢 of A𝑛 it
represents in the canonical basis (𝑒1 , . . . , 𝑒 𝑛 ) of A𝑛 . One has
Λ𝑛 (𝑢)(𝑒1 ∧ · · · ∧ 𝑒 𝑛 ) = 𝑢(𝑒1 ) ∧ · · · ∧ 𝑢(𝑒 𝑛 )
𝑛
Õ 𝑛
Õ
= ... U𝑖1 ,1 . . . U𝑖 𝑛 ,𝑛 𝑒 𝑖1 ∧ 𝑒 𝑖2 ∧ · · · ∧ 𝑒 𝑖 𝑛
𝑖 1 =1 𝑖 𝑛 =1
Õ
= U𝜎(1),1 . . . U𝜎(𝑛),𝑛 𝜀(𝜎)𝑒1 ∧ · · · ∧ 𝑒 𝑛 .
𝜎∈𝔖𝑛

Consequently,
Õ
det(U) = 𝜀(𝜎)U𝜎(1),1 . . . U𝜎(𝑛),𝑛 .
𝜎∈𝔖𝑛

Since a permutation and its inverse have the same signature, we also
obtain Õ
det(U) = 𝜀(𝜎)U1,𝜎(1) . . . U𝑛,𝜎(𝑛) .
𝜎∈𝔖𝑛
In particular, a matrix and its transpose have the same determinant.

8.4.9. Laplace expansion of a determinant. — Let 𝑛 be an integer and

let U be an 𝑛 × 𝑛-matrix with coefficients in a commutative ring A.
Let 𝑢 be the endomorphism of A𝑛 whose matrix in the canonical basis
(𝑒1 , . . . , 𝑒 𝑛 ) is equal to U. For any subset I of {1, . . . , 𝑛}, let ¯I be the
complementary subset. If I = {𝑖 1 , . . . , 𝑖 𝑝 }, with 𝑖 1 < · · · < 𝑖 𝑝 , we denote
𝑒I = 𝑒 𝑖1 ∧ · · · ∧ 𝑒 𝑖 𝑝 . If I and J are any two subsets of {1, . . . , 𝑛}, one has
𝑒I ∧ 𝑒J = 0 if I ∩ J ≠ ∅. Assume that I ∩ J = ∅, and let 𝑚 be the number of
pairs (𝑖, 𝑗) ∈ I × J such that 𝑖 > 𝑗 and set 𝜀I,J = (−1)𝑚 ; then, 𝑒I ∧ 𝑒J = 𝜀I,J 𝑒I∪J .
Let I and J be subsets of {1, . . . , 𝑛}; let UI,J be the submatrix (U𝑖,𝑗 ) 𝑖∈I
𝑗∈J
of U obtained by selecting the rows with indices in I and the columns
406 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

with indices in J. When I and J have the same cardinality, the determinant
of UI,J is called the (I, J)-minor of U. With these notations, we can state
the following proposition.

Proposition (8.4.10). — a) Let I be any subset of {1, . . . , 𝑛}, let 𝑝 =

Card(I). Then, Λ𝑝 (𝑢)(𝑒I ) = J det(UJ,I )𝑒J , where J runs among all subsets of
Í
cardinality 𝑝 of {1, . . . , 𝑛}.
b) Let J and K be subsets of {1, . . . , 𝑛} with the same cardinality. Then
(
Õ det(U) if J = K;
𝜀J,¯J 𝜀K,K¯ det(UI,J ) det(U¯I,K¯ ) =
I
0 otherwise,

the sum being over all subsets I of {1, . . . , 𝑛} with cardinality 𝑝.

Proof. — a) Let 𝑖1 , . . . , 𝑖 𝑝 be the elements of I, written in increasing order.

One has

Λ𝑝 (𝑢)(𝑒I ) = 𝑢(𝑒 𝑖1 ) ∧ · · · ∧ 𝑢(𝑒 𝑖 𝑝 )

𝑛
Õ 𝑛
Õ
= ... U 𝑗1 ,𝑖1 . . . U 𝑗𝑝 ,𝑖 𝑝 𝑒 𝑗1 ∧ · · · ∧ 𝑒 𝑗𝑝
𝑗1 =1 𝑗𝑝 =1

Õ ©Õ
= 𝜀(𝜎)U 𝑗𝜎(1) ,𝑖1 . . . U 𝑗𝜎(𝑝) ,𝑖 𝑝 ® 𝑒 𝑗1 ∧ · · · ∧ 𝑒 𝑗𝑝
ª
­
𝑗1 <···<𝑗𝑝 𝜎∈𝔖𝑝
Õ « ¬
= det(UJ,I )𝑒J ,
J

as was to be shown.
b) By definition,

Λ𝑝 (𝑢)(𝑒I ) ∧ Λ𝑛−𝑝 (𝑢)(𝑒K¯ ) = Λ𝑛 (𝑢)(𝑒I ∧ 𝑒K¯ )

¯ have a common element, that is if I ≠ K, since they

vanishes if I and K
have the same cardinality. Otherwise, if I = K, it is equal to

Λ𝑛 (𝑢)(𝜀I,¯I 𝑒1 ∧ · · · ∧ 𝑒 𝑛 ) = 𝜀I,¯I det(U)𝑒1 ∧ · · · ∧ 𝑒 𝑛 .

 8.5. ADJUNCTION AND EXACTNESS 407

On the other hand, it follows from a) that

Õ Õ
𝑝 𝑛−𝑝
Λ (𝑢)(𝑒I ) ∧ Λ (𝑢)(𝑒K¯ ) = det(UJ,I )𝑒J det(UL,
¯ K¯ )𝑒L¯
Card(J)=𝑝 Card(L)=𝑝
Õ Õ
= det(UJ,I ) det(UL,
¯ K¯ )𝑒J ∧ 𝑒L¯
Card(J)=𝑝 Card(L)=𝑝
Õ
= det(UJ,I ) det(U¯J,K¯ )𝑒J ∧ 𝑒¯J
Card(J)=𝑝
L=J

© Õ
=­ det(UJ,I ) det(U¯J,K¯ )𝜀J,¯J ® 𝑒1 ∧ · · · ∧ 𝑒 𝑛 .
ª

«Card(J)=𝑝 ¬
Comparing these two formulae, we obtain that
Õ
𝜀I,¯I 𝜀J,K¯ det(UJ,I ) det(U¯J,K¯ )
Card(J)=𝑝

equals 0 if I ≠ K, and det(U) otherwise. This concludes the proof of the

proposition. 
Applied with 𝑝 = 1, the formula of b) recovers the well-known formula
for the expansion of a determinant along a column.

Corollary (8.4.11). — Let U be any 𝑛 × 𝑛 matrix with coefficients in A. Let C

be the adjugate matrix of U: for every pair (𝑖, 𝑗) the coefficient C𝑖,𝑗 is equal to
(−1)𝑖+𝑗 times the determinant of the matrix obtained from U by deleting line 𝑖
and column 𝑗. One has Ct · U = U · Ct = det(U)I𝑛 .

Proof. — The formula Ct · U = det(U)I𝑛 is nothing but the Laplace

expansion formula for 𝑝 = 1. The other then follows by transposition.
Indeed, the adjugate matrix of Ut is the transpose of C. Consequently,
Ct · U = det(Ut )I𝑛 , hence Ut · C = det(U)I𝑛 . 

8.5. Adjunction and exactness

𝑓
Theorem (8.5.1). — Let A be a ring, let M be a right A-module, let N1 →
−
𝑔
N2 →
− N3 → 0 be an exact sequence of left A-modules. Then, the following
408 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

diagram

idM ⊗ 𝑓 idM ⊗𝑔
M ⊗A N1 −−−−−→ M ⊗A N2 −−−−−→ M ⊗A N3 → 0

is exact.

Proof. — Exactness at M⊗A N3 . We need to prove that idM ⊗𝑔 is surjective.

Since every element of M ⊗A N3 is a sum of split tensors, it suffices
to show that for any 𝑚 ∈ M and any 𝑧 ∈ N3 , the tensor 𝑚 ⊗ 𝑧 has a
preimage in M ⊗A N2 . Since 𝑔 is surjective, there exists 𝑦 ∈ N2 such that
𝑔(𝑦) = 𝑧. Then 𝑚 ⊗ 𝑧 = (idM ⊗ 𝑔)(𝑚 ⊗ 𝑦), hence the claim.
Exactness at M⊗A N2 . We need to prove that Ker(idM ⊗𝑔) = Im(idM ⊗ 𝑓 ).
The inclusion Im(idM ⊗ 𝑓 ) ⊂ Ker(idM ⊗𝑔) is easy, since (idM ⊗𝑔) ◦
(idM ⊗ 𝑓 ) = idM ⊗(𝑔 ◦ 𝑓 ) = 0. The other inclusion is the difficult one.
Indeed, it is quite illusory to prove it directly, each element at a time.
The incredulous reader should try by himself, and, after some effort,
read again remark 8.1.11.
Since Im(idM ⊗ 𝑓 ) ⊂ Ker(idM ⊗𝑔), the morphism idM ⊗𝑔 induces a
morphism of abelian groups 𝑔 0 : (M ⊗A N2 )/Im(idM ⊗ 𝑓 ) → M ⊗A N3 .
Since idM ⊗𝑔 is surjective, 𝑔 0 is surjective. In fact, we need to prove that
𝑔 0 is an isomorphism. Let us construct an inverse ℎ 0 of 𝑔 0. To shorten the
notation, let P = (M ⊗A N2 )/Im(idM ⊗ 𝑓 ) and let 𝑝 : M ⊗A N2 → P be the
canonical surjection. The sought-for morphism ℎ 0 goes from a tensor
product M ⊗A N3 to P. By the universal property of the tensor product,
its definition amounts to the definition of a biadditive and A-balanced
map ℎ : M × N3 → P such that ℎ(𝑚, 𝑧) = ℎ 0(𝑚 ⊗ 𝑧). Moreover, ℎ 0(𝑚 ⊗ 𝑧)
is supposed to be a preimage of 𝑚 ⊗ 𝑧 by idM ⊗𝑔. After this paragraph
of explanation, let us pass to the construction.
Let 𝑚 ∈ M, let 𝑧 ∈ N3 . For any 𝑦 and 𝑦 0 ∈ N2 such that 𝑔(𝑦) = 𝑔(𝑦 0),
one has 𝑦 0 − 𝑦 ∈ Ker(𝑔) = Im( 𝑓 ), so there exists 𝑥 ∈ N1 such that
𝑦 0 − 𝑦 = 𝑓 (𝑥); consequently, 𝑚 ⊗ 𝑦 0 − 𝑚 ⊗ 𝑦 belongs to Im(idM ⊗ 𝑓 ) and
𝑝(𝑚 ⊗ 𝑦 0) = 𝑝(𝑚 ⊗ 𝑦). Therefore, one defines a map ℎ : M × N3 → P
by setting ℎ(𝑚, 𝑧) = 𝑝(𝑚 ⊗ 𝑦) where 𝑦 is any element of N2 such that
𝑔(𝑦) = 𝑧.
 8.5. ADJUNCTION AND EXACTNESS 409

The map ℎ is biadditive. Indeed, for 𝑧, 𝑧 0 ∈ N3 , 𝑦, 𝑦 0 ∈ N2 such that

𝑔(𝑦) = 𝑧 and 𝑔(𝑦 0) = 𝑧 0, one has 𝑔(𝑦 + 𝑦 0) = 𝑧 + 𝑧 0, so that
ℎ(𝑚, 𝑧 + 𝑧 0) = 𝑝(𝑚 ⊗ (𝑦 + 𝑦 0)) = 𝑝(𝑚 ⊗ 𝑦) + 𝑝(𝑚 ⊗ 𝑦 0) = ℎ(𝑚, 𝑧) + ℎ(𝑚, 𝑧 0).
Similarly, for 𝑚, 𝑚 0 ∈ M, 𝑧 ∈ N3 and 𝑦 ∈ N2 such that 𝑔(𝑦) = 𝑧, one has
ℎ(𝑚 + 𝑚 0 , 𝑧) = 𝑝((𝑚 + 𝑚 0) ⊗ 𝑦) = 𝑝(𝑚 ⊗ 𝑦)+ 𝑝(𝑚 ⊗ 𝑦 0) = ℎ(𝑚, 𝑧)+ ℎ(𝑚 0 , 𝑧).
Finally, the map ℎ is A-balanced: let 𝑚 ∈ M, 𝑧 ∈ N3 , 𝑦 ∈ N2 such that
𝑔(𝑦) = 𝑧, and 𝑎 ∈ A. Then, 𝑔(𝑎𝑦) = 𝑎 𝑔(𝑦) = 𝑎𝑧, since 𝑔 is A-linear, so
that
ℎ(𝑚𝑎, 𝑧) = 𝑝(𝑚𝑎 ⊗ 𝑦) = 𝑝(𝑚 ⊗ 𝑎𝑦) = ℎ(𝑚, 𝑎𝑧).
Consequently, there exists a unique morphism of abelian groups
ℎ 0 : M ⊗A N3 → P such that ℎ 0(𝑚 ⊗ 𝑧) = ℎ(𝑚, 𝑧) for any (𝑚, 𝑧) ∈ M × N3 .
It remains to show that ℎ 0 is the inverse of 𝑔 0. Let 𝑚 ∈ M and 𝑦 ∈ N2 , let
𝑡 = 𝑝(𝑚 ⊗ 𝑦); one has 𝑔 0(𝑡) = 𝑔 0(𝑝(𝑚 ⊗ 𝑦)) = (idM ⊗ 𝑔)(𝑚 ⊗ 𝑦) = 𝑚 ⊗ 𝑔(𝑦),
so that ℎ 0(𝑔 0(𝑡)) = 𝑝(𝑚 ⊗ 𝑦) = 𝑡. Since the tensors of the form 𝑚 ⊗ 𝑦
generate M ⊗A N2 and 𝑝 is surjective, this implies that ℎ 0 ◦ 𝑔 0 = idP . In
particular, 𝑔 0 is injective. Since it is surjective, this is an isomorphism.
Had we wished so, we could also have computed 𝑔 0 ◦ ℎ 0: let 𝑚 ∈ M
and let 𝑧 ∈ N3 , let 𝑦 ∈ N2 be such that 𝑔(𝑦) = 𝑧; then 𝑔 0(ℎ 0(𝑚 ⊗ 𝑧)) =
𝑔 0(ℎ(𝑚, 𝑧)) = 𝑔 0(𝑝(𝑚 ⊗ 𝑦)) = 𝑚 ⊗ 𝑔(𝑦) = 𝑚 ⊗ 𝑧. Since the tensors of the
form 𝑚 ⊗ 𝑧 in M ⊗ N3 generate this module, it follows that 𝑔 0(ℎ 0(𝑡)) = 𝑡
for every 𝑡 ∈ M ⊗ N3 , hence 𝑔 0 ◦ ℎ 0 = idM⊗N3 . 

8.5.2. — This important result gains in being rephrased in the categorical

language. Let A be a ring, let P be a right A-module. Observe that
one defines a functor TP by associating to any left A-module M the
abelian group TP (M) = P ⊗A M, and to any morphism 𝑓 : M → N of left
A-modules the morphism TP ( 𝑓 ) = idM ⊗A 𝑓 . Indeed, TP ( 𝑓 ) goes from
P ⊗A M = TP (M) to P ⊗A N = TP (N), TP (idM ) = idP ⊗A idM = idP⊗A M =
idTP (M) , and
TP (𝑔 ◦ 𝑓 ) = idP ⊗(𝑔 ◦ 𝑓 ) = (idP ⊗𝑔) ⊗ (idP ⊗ 𝑓 ) = TP (𝑔) ◦ TP ( 𝑓 )
for all morphisms 𝑓 : M1 → M2 and 𝑔 : M2 → M3 of left A-modules.
This “tensor product by P” functor is additive and we may rephrase
theorem 8.5.1 as follows:
410 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Corollary (8.5.3). — Let A be a ring, let M be a right A-module. The functor

“tensor product by M” is right exact.

In fact, once stated in this language, theorem 8.5.1 is justiciable of

another, more categorical, proof. Thanks to the following proposition,
it will be a consequence of the general result that left adjoints are right
exact (proposition 7.6.5).
Let us generalize this functor a little bit. Let B be a second ring and let
us assume that P is a (B, A)-bimodule. Then, for any left A-module M,
P ⊗A M is naturally a left B-module and for any morphism 𝑓 : M → N
of left A-modules, idP ⊗ 𝑓 is B-linear. Consequently, the functor TP goes
from the category of left A-modules to the category of left B-modules.

Proposition (8.5.4). — Let A and B be rings, let P be a (B, A)-bimodule. The

“tensor product by P” functor TP defined above is a left adjoint of the functor
HomA (P, •).

Proof. — For any left A-module M and any left B-module N, we need to
define a bijection ΦM,N : HomB (P ⊗A M, N) → HomA (M, HomB (P, N))
satisfying the relations required by the definition of a pair of adjoint
functors.
Before we give such a formula, let us prove it with words. By the
universal property of the tensor product, HomB (P ⊗A M, N) is the set of
biadditive, A-balanced maps 𝑢 from P × M to N which are also B-linear
with respect to P. Such a map is additive in each of its variables, B-linear
in the P-variable, and A-balanced. In particular, it induces, for every
𝑚 ∈ M, a B-linear map 𝑢(·, 𝑚) from P to N; moreover, 𝑢(𝑝𝑎, 𝑚) = 𝑢(𝑝, 𝑎𝑚)
for every 𝑎 ∈ A, 𝑝 ∈ P and 𝑚 ∈ M. Conversely, a family (𝑢𝑚 )𝑚∈M of
B-linear maps from P to N such that the map 𝑚 ↦→ 𝑢𝑚 is A-linear gives
exactly a map 𝑢 as required.
For the sake of the reader, let us put this in formulae. Let 𝑢 : P ⊗A M →
N be a morphism of B-modules. Let 𝑚 ∈ M. Then the map 𝑝 ↦→ 𝑢(𝑝 ⊗ 𝑚)
is B-linear, hence is an element ΦM,N (𝑢)(𝑚). For 𝑎, 𝑎 0 ∈ A, 𝑚, 𝑚 0 ∈ M,
and 𝑝 ∈ P, one has
ΦM,N (𝑢)(𝑎𝑚 + 𝑎 0 𝑚 0)(𝑝) = 𝑢(𝑝 ⊗ (𝑎𝑚 + 𝑎 0 𝑚 0)) = 𝑢(𝑝𝑎 ⊗ 𝑚) + 𝑢(𝑝𝑎 0 ⊗ 𝑚 0)
= ΦM,N (𝑢)(𝑚)(𝑝𝑎) + ΦM,N (𝑢)(𝑚)(𝑝𝑎 0),
 8.6. FLAT MODULES 411

so that
ΦM,N (𝑢)(𝑎𝑚 + 𝑎 0 𝑚 0) = 𝑎 · ΦM,N (𝑢)(𝑚) + 𝑎 0 · ΦM,N (𝑢)(𝑚 0).
Consequently, the map 𝑚 ↦→ ΦM,N (𝑢)(𝑚) is A-linear, so that 𝜑M,N (𝑢)
belongs to HomA (M, HomB (P, N)). We thus have defined a map ΦM,N .
Let us prove that ΦM,N is bijective. Let 𝑣 ∈ HomA (M, HomB (P, N)). For
𝑢 ∈ HomB (P ⊗A M, N), the relations ΦM,N (𝑢) = 𝑣 and 𝑢(𝑝 ⊗ 𝑚) = 𝑣(𝑚)(𝑝)
are equivalent. Since the tensors of the form 𝑝 ⊗ 𝑚 generate P ⊗A M,
this implies that ΦM,N is injective. Moreover, the map from P × M
to N given by ℎ(𝑝, 𝑚) = 𝑣(𝑚)(𝑝) is biadditive. Since 𝑚 ↦→ 𝑣(𝑚) is
A-linear, one has 𝑣(𝑎𝑚)(𝑝) = (𝑎 · 𝑣(𝑚))(𝑝) = 𝑣(𝑚)(𝑝𝑎), so that ℎ is A-
balanced. Consequently, there exists a unique morphism of abelian
groups 𝑢 : P ⊗A M → N such that 𝑢(𝑝 ⊗ 𝑚) = ℎ(𝑝, 𝑚) for 𝑝 ∈ P and
𝑚 ∈ M. Since 𝑣(𝑚) is B-linear for every 𝑚 ∈ M, one has
𝑢(𝑏 · (𝑝 ⊗ 𝑚)) = 𝑢(𝑏𝑝 ⊗ 𝑚) = 𝑣(𝑚)(𝑏𝑝) = 𝑏(𝑣(𝑚)(𝑝)) = 𝑏𝑢(𝑝 ⊗ 𝑚)
and the map 𝑢 is B-linear. It is the unique preimage of 𝑣 in HomB (P ⊗A
M, N). This shows that ΦM,N is bijective.
Let M, M0 be left A-modules, N, N0 be left B-modules, and let 𝑓 : M →
M0 and 𝑔 : N → N0 be morphisms of modules. By construction, for any
morphism 𝑢 : P ⊗A M0 → N of B-modules, any 𝑝 ∈ P and any 𝑚 ∈ M,
one has
(𝑔 ◦ ΦM0 ,N (𝑢) ◦ 𝑓 )(𝑚)(𝑝) = 𝑔(ΦM0 ,N (𝑢)( 𝑓 (𝑚))(𝑝)) = 𝑔(𝑢(𝑝 ⊗ 𝑓 (𝑚))),
while
ΦM,N0 (𝑔 ◦ 𝑢 ◦ (idP × 𝑓 ))(𝑚)(𝑝) = (𝑔 ◦ 𝑢 ◦ idP × 𝑓 )(𝑝 × 𝑚)
= (𝑔 ◦ 𝑢)(𝑝 ⊗ 𝑓 (𝑚))
= 𝑔(𝑢(𝑝 ⊗ 𝑓 (𝑚))).
This shows that the maps ΦM,N satisfy the adjunction relations and
concludes the proof of the proposition. 

8.6. Flat modules

Definition (8.6.1). — Let A be a ring and let M be a right A-module. One
says that M is flat if the functor “tensor product by M”, TM , M ⊗A •, is exact.
412 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Since this functor is always right exact (corollary 8.5.3), a module M

is flat if and only if, for any injective morphism 𝑓 : N → N0 of left
A-modules, idM ⊗ 𝑓 is injective.
One defines similarly the notion of a flat left A-module M, by the
exactness of the right exact functor • ⊗A M.
Example (8.6.2). — The map N → A𝑑 ⊗A N given by 𝑛 ↦→ 1 ⊗ 𝑛 is an
isomorphism of A-modules, and a morphism 𝑢 is changed into idA ⊗𝑢
under this isomorphism. Consequently, A𝑑 is a flat A-module.
Example (8.6.3). — Let A be a commutative ring and let S be a multiplica-
tive subset of A. According to example 8.2.2, tensoring a module M with
S−1 A corresponds to taking the fraction module S−1 M. If 𝑢 : M → N is a
morphism of A-modules, then under the isomorphisms of that example,
the morphisms S−1 𝑢 : S−1 M → S−1 N and 𝑢 ⊗id : M⊗A S−1 A → N⊗A S−1 A
coincide. This identifies the functor •⊗A S−1 A with the functor M ↦→ S−1 A
which is exact, by proposition 7.5.12.
Consequently, S−1 A is a flat A-module.
Remark (8.6.4). — It is possible that the module M be non flat, but
that the functor TM preserves the exactness of some exact sequences
0 → N → N0 → N00 → 0.
For example, if 𝑓 : N → N0 is a split injection of A-modules, with
left inverse 𝑔 : N0 → N, then idM ⊗ 𝑔 is a left inverse of idM ⊗ 𝑓 , so that
idM ⊗ 𝑓 is again a split injection.
A more advanced discussion of homological algebra will construct
A-modules TorA 𝑝 (M, N), for 𝑝 > 1, and a long exact sequence,

00 0
· · · → TorA
2 (M, N ) → Tor1 (M, N) → Tor1 (M, N ) →
A A

00 0 00
→ TorA
1 (M, N ) → M ⊗ N → M ⊗ N → M ⊗ N → 0.

Proposition (8.6.5). — Let A be a ring and let M be a direct sum of a family

(M𝑖 )𝑖∈I of A-modules. Then M is flat if and only if M𝑖 is flat for every 𝑖.

𝑖 M 𝑖 be a direct sum of submodules; for 𝑖 ∈ I, let 𝑗𝑖 be

É
Proof. — Let M =
the injection of M𝑖 into M and let 𝑝 𝑖 : M → M𝑖 be the projection. Then, for
𝜃 𝑝 𝑖 ⊗ idN : M ⊗A N '
Í
any left A-module N, one has an isomorphism =
𝑖 M 𝑖 ⊗A N whose inverse is induced by the natural maps 𝑗𝑖 ⊗ idN from
É
 8.6. FLAT MODULES 413

M𝑖 ⊗A N to M ⊗A N. Let 𝑢 : N → N0 be a morphism of left A-modules.

In the isomorphism 𝜃, the morphism idM ⊗𝑢 is transformed into the
direct sum of the morphisms idM𝑖 ⊗𝑢. In particular, idM ⊗𝑢 is injective
if and only if idM𝑖 ⊗𝑢 is injective for every 𝑖.
Let 𝑢 : N → N0 is an injective morphism of A-modules. If all A-
modules M𝑖 are flat, then idM𝑖 ⊗𝑢 is injective for every 𝑖, hence idM ⊗𝑢
is injective. This proves that M is flat.
Conversely, if M is flat, then idM𝑖 ⊗𝑢 is injective for every injective
morphism 𝑢 : N → N0 and every 𝑖, so that M𝑖 is flat for every 𝑖. 

Corollary (8.6.6). — Any projective A-module, in particular any free A-module,

is flat.

Proof. — Since the right A-module A𝑑 is flat, every free right A-module is
flat. Any projective module, being a direct summand of a free A-module,
is flat. 

Example (8.6.7). — Any module over a division ring is flat.

Theorem (8.6.8). — Let A be a ring, let M be a right A-module. Then M is flat

if and only if for every left ideal J of A, the canonical morphism from M ⊗A J
to M (sending 𝑚 ⊗ 𝑎 to 𝑚𝑎, for every 𝑚 ∈ M and every 𝑎 ∈ J) is injective.

Proof. — Assume that M is flat. Let J be a left ideal of A and let us

consider the exact sequence 0 → J → A → A/J → 0 of left A-modules.
Since M is flat, tensoring it with M, we obtain an exact sequence
0 → M ⊗ J → M → M ⊗ A/J → 0 of abelian groups, where the injection
M ⊗ J → M is the one of the lemma.
Conversely, let us now assume that the natural map M ⊗A J → M is
injective, for every left ideal J, and let us prove that M is a flat A-module.
Let 𝑓 : N → N0 be an injective morphism of left A-modules and let us
show that idM ⊗ 𝑓 is injective.
Step 1. We first prove this in the case N0 = A𝑛 , by induction on 𝑛,
𝑓 being the inclusion of the submodule N. The case 𝑛 = 0 is obvious
(N = N0 = 0), and the case 𝑛 = 1 is exactly the assumption of the lemma
(identifying N0 with A, then N is a left ideal). Let 𝑞 0 : N0 → A𝑛−1 = N01
be the projection to the (𝑛 − 1) first coordinates and let N02 = {0} 𝑛−1 × A
414 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

be its kernel; let 𝑗 0 : N02 → N0 be the canonical inclusion. Let N1 = 𝑞(N)

and N2 = N ∩ N02 ; this gives rise to a commutative diagram with exact
rows
𝑗 𝑞
→ N2 → N → N1 → 0
← ← ← ←
0

←
←

←
𝑓2 𝑓 𝑓1

→
→
𝑗0 𝑞0
→ N02 → N0 → N01 → 0,
← ← ← ←
0

where 𝑗 and 𝑞 are the restrictions of 𝑗 0 and 𝑞 0 to N2 and N1 , and the

morphisms 𝑓1 : N1 → N01 and 𝑓2 : N2 → N02 are deduced from 𝑓 . Let us
apply the functor “tensor product with M” to this diagram. We obtain a
commutative diagram

idM ⊗𝑗 idM ⊗𝑞
M ⊗A N2 → M ⊗A N → M ⊗A N1 → 0
← ← ←
←

←
idM ⊗ 𝑓2 idM ⊗ 𝑓 idM ⊗ 𝑓1
→

→
→

idM ⊗𝑗 0 idM ⊗𝑞 0
→ M ⊗A N02 → M ⊗A N0 → M ⊗A N01 → 0.
← ← ← ←
0

The upper row is an exact sequence, because the functor TM is right

exact. The lower row is exact as well, because the initial sequence is
split (remark 8.6.4). By the snake lemma (theorem 7.2.1), the morphism
idM ⊗ 𝑓 is injective, as was to be shown.
Step 2. Let us now prove the result under the assumption that N0 is
finitely generated. Let 𝑠 0 : P0 = A𝑛 → N0 be a surjective morphism, let
K = Ker(𝑠 0) and let P = (𝑠 0)−1 (N). Write 𝑗 : P → P0 and 𝑘 : K → P for
the canonical inclusions; let also 𝑘 0 = 𝑗 ◦ 𝑘 : K → P0 and 𝑠 = 𝑠 0 | P , so
that K = Ker(𝑠). We sum up these constructions by the commutative
diagram with exact rows:

𝑘 𝑠
→ K → P → N → 0
← ← ← ←
0
⇐

𝑗 𝑓
→

→
⇐

𝑘0 𝑠0
→ K → P0 → N0 → 0.
← ← ← ←
0

Let us tensor apply the functor “tensor product with M”. By right-
exactness of the tensor product, we obtain a commutative diagram with
 8.6. FLAT MODULES 415

exact rows
idM ⊗𝑘 idM ⊗𝑠
M ⊗A K → M ⊗A P → M ⊗A N → 0
← ← ←

←
idM ⊗ 𝑗 idM ⊗ 𝑓

→
⇐
idM ⊗𝑘 0 idM ⊗𝑠 0
M ⊗A K → M ⊗A P0 → M ⊗A N0 → 0.
← ← ←

By the first step, we also know that the morphisms idM ⊗𝑗 and idM ⊗𝑘 0
are injective, so that idM ⊗𝑘 is injective as well. Let us prove that idM ⊗ 𝑓
is injective; the argument is analogous to that of theorem 7.2.1. Let then
𝜇 ∈ M ⊗A N be an element of Ker(idM ⊗ 𝑓 ). Let 𝜋 ∈ M ⊗A P be an element
such that idM ⊗𝑠(𝜋) = 𝜇. One has
(idM ⊗𝑠 0) ◦ (idM ⊗ 𝑗)(𝜋) = (idM ⊗ 𝑓 ) ◦ (idM ⊗𝑠)(𝜋) = (idM ⊗ 𝑓 )(𝜇) = 0.
Consequently, there exists an element 𝜋0 ∈ M ⊗A K such that
(idM ⊗𝑗)(𝜋) = (idM ⊗𝑘 0)(𝜋0), and 𝜋 = (idM ⊗𝑘)(𝜋0). Then
𝜇 = (idM ⊗𝑠)(𝜋) = (idM ⊗𝑠) ◦ (idM ⊗𝑘)(𝜋0) = 0,
as was to be shown.
Step 3. We finally prove the assertion in general. Let 𝜇 ∈ Ker(idM ⊗ 𝑓 )
be given as a finite linear combination of split tensors 𝜇 = 𝑚 𝑖 ⊗ 𝑛 𝑖 ,
Í
with 𝑚 𝑖 ∈ M and 𝑛 𝑖 ∈ N. By assumption, one has 𝑚 𝑖 ⊗ 𝑓 (𝑛 𝑖 ) = 0 in
Í
M ⊗A N0.
By remark 8.1.6, there exists a finitely generated submodule N00 of N,
containing the elements 𝑓 (𝑛 𝑖 ), such that 𝑚 𝑖 ⊗ 𝑓 (𝑛 𝑖 ) = 0 in M ⊗A N00 .
Í
Let N0 be the submodule of N generated by the 𝑛 𝑖 , let 𝑗 : N0 → N be
the inclusion, and let 𝑓0 : N0 → N00 be the map induced by 𝑓 ; let 𝜇0 =
𝑚 𝑖 ⊗ 𝑛 𝑖 viewed in M ⊗A N0 . By construction, one has (idM ⊗ 𝑓0 )(𝜇0 ) = 0,
Í
that is, 𝜇0 ∈ Ker(idM ⊗ 𝑓0 ). Since N00 is finitely generated, step 2 proves
that 𝜇0 = 0. Then 𝜇 = (idM ⊗𝑗)(𝜇0 ) = 0. This proves that idM ⊗ 𝑓 is
injective. 

Corollary (8.6.9). — Let A be an integral domain and let M be an A-module.

If M is flat, then M is torsion-free. The converse holds if A is a principal ideal
domain.

Proof. — Let 𝑎 be a nonzero element of A. The ideal (𝑎) is a free A-

module, with basis 𝑎, so that the morphism M → (𝑎) ⊗A M given by
416 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

𝑚 ↦→ 𝑎 ⊗ 𝑚 is an isomorphism. Under this identification, the morphism

(𝑎) ⊗A M → M given by 𝑏 ⊗ 𝑚 ↦→ 𝑏𝑚 identifies with the morphism
𝑚 ↦→ 𝑎𝑚. If M is flat, then this morphism is injective (theorem 8.6.8)
so that no nonzero element 𝑚 ∈ M satisfies 𝑎𝑚 = 0. This proves that
a flat A-module is torsion-free. Conversely, let us assume that A is a
principal ideal domain and M is torsion-free. In this case, this argument
shows that the canonical morphism J ⊗A M → M given by 𝑏 ⊗ 𝑚 ↦→ 𝑏𝑚
is injective for every nonzero ideal J of A. On the other hand, if J = 0, one
has J ⊗A M = 0 and this morphism is injective as well. By theorem 8.6.8,
this shows that M is flat. 

8.6.10. — Let M and P be right A-modules. Let M∨ = HomA (M, A) be

the dual of M, viewed as a left A-module. For 𝜑 ∈ M∨ and 𝑝 ∈ P, the
map 𝑝𝜑 : 𝑥 ↦→ 𝑝𝜑(𝑥) from M to P is a morphism of A-modules, and
the map (𝑝, 𝜑) ↦→ 𝑝𝜑 is biadditive and A-balanced, since (𝑝𝑎)𝜑(𝑥) =
𝑝𝑎𝜑(𝑥) = 𝑝(𝑎𝜑)(𝑥) for every 𝑥 ∈ M, every 𝑝 ∈ P and every 𝜑 ∈ M∨ .
Consequently, there exists a unique morphism

𝛿 M : P ⊗A M∨ → HomA (M, P)

such that 𝛿 M (𝑝 ⊗ 𝜑) = 𝑝𝜑 for every 𝜑 ∈ M∨ and every 𝑝 ∈ P. (See also

proposition 8.2.4.)

Proposition (8.6.11). — Let A be a ring and let P be a flat right A-module. Then,
for every finitely presented right A-module M, the morphism 𝛿 M : P ⊗A M∨ →
HomA (M, P) is an isomorphism.

Proof. — Since M is finitely presented, there exists an exact sequence of

A-modules.
𝑢 𝑣
L0 →
− L→
− M → 0,

where L and L0 are free and finitely generated. Let us apply the right
exact functor •∨ = HomA (•, A) to this exact sequence; this furnishes an
exact sequence
𝑣∗ 𝑢∗
0 → M∨ −
→ L∨ −→ (L0)∨ .
 8.6. FLAT MODULES 417

We now tensor this exact sequence with P and obtain, since P is flat, an
exact sequence
idP ⊗𝑣 ∗ idP ⊗𝑢 ∗
0 → P ⊗A M∨ −−−−−→ P ⊗ L∨ −−−−−→ P ⊗ (L0)∨ .
Similarly, we apply the right exact functor HomA (•, P) and obtain an
exact sequence
𝑣∗ 𝑢∗
→ HomA (L, P) −→ HomA (L0 , P).
0 → HomA (M, P) −
Via the various morphisms 𝛿, these last two exact sequences can be
combined into the following diagram
idP ⊗𝑣 ∗ idP ⊗𝑢 ∗
→ P ⊗A M∨ → P ⊗A L∨ → P ⊗A (L0)∨
← ← ←
0
←

←
𝛿M 𝛿L 𝛿L0
→

→
𝑣∗ 𝑢∗
→ HomA (M, P) → HomA (L, P) → HomA (L0 , P).
← ← ←
0

This left square is commutative: if 𝜑 ∈ M∨ and 𝑝 ∈ P, since 𝛿 L ◦

(idP ⊗𝑣 ∗ )(𝑝 ⊗ 𝜑) and (𝑣 ∗ ◦𝛿 M )(𝑝 ⊗ 𝜑) both are the morphism 𝑦 ↦→ 𝜑(𝑣(𝑦))𝑝
from L to P. Consequently, the morphisms 𝛿 L ◦ (idP ⊗𝑣 ∗ ) and 𝑣 ∗ ◦ 𝛿M
coincide on split tensors, hence are equal. This proves that the left hand
square is commutative. The commutativity of the right hand one is
analogous: the maps𝛿 L0 ◦ (idP ⊗𝑢 ∗ ) and 𝑢 ∗ ◦ 𝛿 L both map a split tensor
𝑝 ⊗ 𝜑 (with 𝑝 ∈ P and 𝜑 ∈ L∨ ) to the morphism 𝑦 ↦→ 𝑝𝜑(𝑢(𝑦)), hence
they coincide.
The module L is free and finitely generated. As in the proof of
proposition 8.2.4 (which concerns the case where A is commutative),
we prove that the morphism 𝛿L is an isomorphism. Indeed, if (𝑒 𝑖 )𝑖∈I
is a finite basis of L, with dual basis (𝑒 𝑖∗ ), and if 𝑢 ∈ HomA (L, P), then
𝜇 = 𝑖 𝑢(𝑒 𝑖 ) ⊗ 𝑒 𝑖∗ is the unique element of P ⊗A M∨ such that 𝛿 L (𝜇) = 𝑢.
Í
Similarly, 𝛿 L0 is an isomorphism.
Let us deduce that 𝛿 M is an isomorphism. One has 𝛿 L ◦ (idP ⊗𝑣 ∗ ) =
𝑣 ∗ ◦ 𝛿 M , and 𝛿 L ◦ (idP ⊗𝑣 ∗ ) is injective, so that 𝛿 M is injective. Let
𝑓 ∈ HomA (M, P). Consider the morphism 𝑣 ∗ ( 𝑓 ) = 𝑓 ◦ 𝑣 ∈ HomA (L, P).
Since 𝛿L is an isomorphism, there exists a unique element 𝜇 ∈ P ⊗A L∨
such that 𝛿 L (𝜇) = 𝑓 ◦ 𝑣. Then
𝛿L0 ◦ (idP ⊗𝑢 ∗ )(𝜇) = 𝑢 ∗ ◦ (idP ⊗𝛿 L )(𝜇) = 𝑢 ∗ (𝑣 ∗ ( 𝑓 )) = 𝑓 ◦ 𝑣 ◦ 𝑢 = 0.
418 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Since 𝛿 L0 is an isomorphism, one has idP ⊗𝑢 ∗ (𝜇) = 0. By the exactness

of the first row of the diagram, there exists 𝜈 ∈ P ⊗A M∨ such that
𝜇 = (idP ⊗𝑣 ∗ )(𝜈). Then,
𝛿 M (𝜈) ◦ 𝑣 = 𝛿 L ◦ (idP ⊗𝑣 ∗ )(𝜈) = 𝛿 L (𝜇) = 𝑓 ◦ 𝑣.
Since 𝑣 is surjective, 𝑓 = 𝛿M (𝜈). This concludes the proof that 𝛿 M is an
isomorphism. 

Theorem (8.6.12) (Lazard). — Let A be a ring and let P be a right A-module.

The module P is flat if and only if, for every finitely presented right A-module M
and every morphism 𝑓 : M → P, there exists a free finitely generated right
A-module L and morphisms 𝑔 : M → L and ℎ : L → P such that 𝑓 = ℎ ◦ 𝑔.

Proof. — Assume that P is flat, let M be a finitely presented right A-

module and let 𝑓 : M → P be a morphism of A-modules. Let us choose a
tensor 𝜇 ∈ P ⊗A M∨ such that 𝛿 M (𝜇) = 𝑓 and let us write 𝜇 = 𝑛𝑖=1 𝑝 𝑖 ⊗ 𝜑 𝑖 ,
Í
with 𝜑1 , . . . , 𝜑𝑛 ∈ M∨ and 𝑝1 , . . . , 𝑝 𝑛 ∈ P. Let 𝜑 = (𝜑1 , . . . , 𝜑𝑛 ) : M →
A𝑛 and 𝑝 : A𝑛 → P be given by 𝑝(𝑎1 , . . . , 𝑎 𝑛 ) = 𝑝1 𝑎1 + · · · + 𝑝 𝑛 𝑎 𝑛 ; these
are morphisms of A-modules. Moreover, for every 𝑚 ∈ M, one has
𝑛
Õ
𝑝 ◦ 𝜑(𝑚) = 𝑝 𝑖 𝜑 𝑖 (𝑚) = 𝛿 M (𝜇)(𝑚) = 𝑓 (𝑚),
𝑖=1

so that 𝑚 = 𝑝 ◦ 𝜑. This proves the desired property.

Conversely, let us assume that this property holds and let us prove
that P is flat. Let J be a left ideal of A and let us show that the canonical
morphism 𝑓 : P ⊗A J → P is injective. Let 𝑥 be any element of Ker( 𝑓 ),
written as 𝑥 = 𝑛𝑖=1 𝑥 𝑖 ⊗ 𝑎 𝑖 , where 𝑎1 , . . . , 𝑎 𝑛 ∈ A and 𝑥 1 , . . . , 𝑥 𝑛 ∈ P.
Í
Let M be the quotient of the free A-module A𝑛𝑑 , with canonical basis
(𝑒1 , . . . , 𝑒 𝑛 ), by its submodule generated by 𝑒1 𝑎1 + · · · + 𝑒 𝑛 𝑎 𝑛 ; let 𝑝 : A𝑛𝑑 →
M be the canonical surjection. The module M is finitely presented. By
construction, the morphism from A𝑛𝑑 to P such that 𝑒 𝑖 ↦→ 𝑥 𝑖 for each 𝑖
vanishes on (𝑎1 , . . . , 𝑎 𝑛 )A. Consequently, there exists a unique morphism
𝑓 : M → P such that 𝑓 (𝑝(𝑒 𝑖 )) = 𝑥 𝑖 for every 𝑖. Let 𝑚 be an integer, let
𝑔 : M → A𝑚 𝑑
and ℎ : A𝑚
𝑑
→ P be morphisms such that 𝑓 = ℎ ◦ 𝑔.
Let ( 𝑓1 , . . . , 𝑓𝑚 ) be the canonical basis of A𝑚𝑑
; For every 𝑗 ∈ {1, . . . , 𝑚},
let 𝑦 𝑗 = ℎ( 𝑓 𝑗 ). For each 𝑖 ∈ {1, . . . , 𝑛}, let (𝑏 𝑖,𝑗 ) be the coordinates
 8.6. FLAT MODULES 419

Í𝑚
of 𝑔(𝑝(𝑒 𝑖 )) in the basis ( 𝑓1 , . . . , 𝑓𝑚 ); one has 𝑔(𝑝(𝑒 𝑖 )) = 𝑗=1 𝑓 𝑗 𝑏 𝑖,𝑗 . Since
𝑖 𝑝(𝑒 𝑖 )𝑎 𝑖 = 𝑝( 𝑒 𝑖 𝑎 𝑖 ) = 0, one has
Í Í

𝑛 𝑚 𝑛
!
Õ Õ Õ
0= 𝑔(𝑝(𝑒 𝑖 )𝑎 𝑖 ) = 𝑓𝑗 𝑏 𝑖,𝑗 𝑎 𝑖 ,
𝑖=1 𝑗=1 𝑖=1
Í𝑛
hence 𝑖=1 𝑏 𝑖,𝑗 𝑎 𝑖 = 0 for every 𝑗. Moreover,
𝑚
Õ
𝑥 𝑖 = 𝑓 (𝑝(𝑒 𝑖 )) = ℎ(𝑔(𝑝(𝑒 𝑖 ))) = 𝑦 𝑗 𝑏 𝑖,𝑗
𝑗=1

for every 𝑖. Then one has, in P ⊗A J, the relations

𝑛
Õ Õ𝑛 𝑚
©Õ
𝑥𝑖 ⊗ 𝑎𝑖 = 𝑦 𝑗 𝑏 𝑖,𝑗 ® ⊗ 𝑎 𝑖
ª
­
𝑖=1 𝑖=1 « 𝑗=1 ¬ !
Õ𝑚 Õ𝑛
= 𝑦 𝑗 𝑏 𝑖,𝑗 ⊗ 𝑎 𝑖
𝑗=1 𝑖=1
𝑚 𝑛
!
Õ Õ
= 𝑦𝑗 ⊗ 𝑏 𝑖,𝑗 𝑎 𝑖
𝑗=1 𝑖=1

= 0.
This proves that the canonical morphism P ⊗A J → P is injective and
concludes the proof that P is a flat A-module. 

Corollary (8.6.13). — Let A be a commutative ring and let M be a finitely

presented A-module. Then M is flat if and only if M is projective.
Proof. — A projective module is flat (corollary 8.6.6), hence it suffices
to prove that M is projective if it is flat. By the preceding proposition
applied to the identity morphism idM : M → M, there exist a free finitely
generated A-module L and morphisms 𝑓 : M → L and 𝑔 : L → M such
that idM = 𝑔 ◦ 𝑓 . The morphism 𝑓 is injective and 𝑔 is a left inverse to 𝑓 .
𝑓
By lemma 7.1.6, applied to the exact sequence 0 → M →− L → L/M → 0,
the submodule 𝑓 (M) of L has a direct summand, hence is projective
(proposition 7.3.2). Since 𝑓 induces an isomorphism from M to its
image 𝑓 (M), the module M is projective, as was to be shown. 
420 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Remark (8.6.14). — By definition, a finitely presented right A-module M

is the cokernel of an A-morphism 𝑢 : A𝑛𝑑 → A𝑚 𝑑
. Such a morphism
can be represented by a matrix U ∈ M𝑚,𝑛 (A). Moreover, a morphism
𝑝 𝑝
from M to a free A-module A𝑠 corresponds to a morphism 𝑣 : A𝑛𝑑 → A𝑑
that vanishes on the image of 𝑢. The morphism 𝑣 is represented by a
matrix V ∈ M𝑝,𝑚 (A), and the condition that 𝑣(Im(𝑢)) = 0 translates as
the relation V · U = 0.
Thanks to these remarks, theorem 8.6.12 can be rewritten as the
following equational criterion of flatness. Namely, the following properties
are equivalent:
(i) The A-module M is flat;
(ii) For every family (𝑥1 , . . . , 𝑥 𝑚 ) of elements of M, every matrix (𝑎 𝑖,𝑗 ) ∈
M𝑚,𝑛 (A) such that 𝑚 𝑖=1 𝑥 𝑖 𝑎 𝑖,𝑗 = 0 for every 𝑗 ∈ {1, . . . , 𝑛}, there exist an
Í
integer 𝑝, a family (𝑦1 , . . . , 𝑦 𝑝 ) in M and a matrix (𝑏 𝑘,𝑖 ) ∈ M𝑝,𝑚 (A) such
that
𝑝
Õ
𝑥𝑖 = 𝑦 𝑘 𝑏 𝑘,𝑖
𝑘=1
for every 𝑖 ∈ {1, . . . , 𝑚}, and
𝑚
Õ
𝑏 𝑘,𝑖 𝑎 𝑖,𝑗 = 0
𝑖=1

for every 𝑘 and 𝑗.

(iii) Same assertion as (ii), with 𝑛 = 1.
The equivalence of (i) and (ii) is the content of theorem 8.6.12, and in
the proof, we have seen that it is sufficient to consider A-modules M
which are cokernels of a morphism 𝑢 : A𝑑 → A𝑚 𝑑
, ie, with 𝑛 = 1.

Proposition (8.6.15). — Let A be a commutative ring and let M be an A-module.

a) Let S be a multiplicative subset of A. If M is flat, then the S−1 A-module
S−1 M is flat.
b) If MP is a flat AP -module for every maximal ideal P of A, then M is flat.

Proof. — a) For any S−1 A-module N, there is a morphism of S−1 A-

modules M ⊗A N ' S−1 M ⊗S−1 A N sending 𝑚 ⊗ 𝑛 to (𝑚/1) ⊗ 𝑛. This
 8.7. FAITHFUL FLATNESS 421

morphism is an isomorphism. Moreover, if 𝑢 : N → N0 is a morphism

of S−1 A-modules, this morphism transforms idM ⊗𝑢 into idS−1 M ⊗𝑢.
If 𝑢 is injective and M is flat, it follows that idS−1 M ⊗𝑢 is injective. Since
𝑢 is arbitrary, S−1 M is a flat S−1 A-module.
b) Let 𝑢 : N → N0 be an injective morphism of A-modules. The image
of idM ⊗𝑢 is M ⊗ 𝑢(N); let Q be the kernel of idM ⊗𝑢 so that we have an
exact sequence
idM ⊗𝑢
0 → Q → M ⊗A N −−−−−→ M ⊗A 𝑢(N) → 0.
Let P be a maximal ideal of A and let us tensor this exact sequence
of A-modules by the flat A-module AP ; we obtain an exact sequence
idMP ⊗𝑢P
0 → QP → MP ⊗A NP −−−−−−→ MP ⊗AP 𝑢(N)P → 0.
Since AP is a flat A-module (example 8.6.3), the morphism 𝑢P is injective.
Since MP is assumed to be flat, the morphism idMP ⊗𝑢P is injective.
Consequently, QP = 0 for every P ∈ Spec(A). One thus has Supp(Q) = 0,
hence Q = 0 by theorem 6.5.4.
This implies that the map idM ⊗𝑢 is injective. Since 𝑢 is arbitrary, M is
a flat A-module. 

8.7. Faithful flatness

Definition (8.7.1). — Let A be a commutative ring and let M be a A-module.
𝑢 𝑣
One says that M is faithfully flat if a complex N1 →
− N2 →
− N3 of A-modules
idM ⊗𝑢 idM ⊗𝑣
is exact if and only if the complex M ⊗A N1 −−−−−→ M ⊗A N2 −−−−−→ M ⊗A N3
is exact.
The “only if” direction of the definition shows that a faithfully flat
A-module is flat.
Proposition (8.7.2). — Let M be a flat A-module. The following properties are
equivalent:
(i) The A-module M is faithfully flat;
(ii) For every morphism 𝑢 : N → N0 of A-modules such that idM ⊗𝑢 = 0,
one has 𝑢 = 0;
422 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

(iii) For every nonzero A-module N, one has M ⊗A N ≠ 0;

(iv) For every prime ideal P of A, the A-module M ⊗A KP is nonzero, where
KP is the field of fractions of the integral domain A/P;
(v) For every maximal ideal P of A, one has M ≠ PM.

Proof. — (i)⇒(ii). Assume that idM ⊗𝑢 = 0. Let us consider the complex

𝑢 idN0
− N0 −−−→ N0 and let us tensor it with M. We obtain the complex
N→
idM ⊗𝑢 id
M ⊗A N −−−−−→ M ⊗A N0 −
→ M ⊗A N0 ,
which is exact since idM ⊗𝑢 = 0, so that its image, 0, equals the kernel of
the identity isomorphism. Since M is faithfully flat the initial complex
was exact and Im(𝑢) = Ker(idN0 ) = 0, hence 𝑢 = 0.
(ii)⇒(iii). One has M ⊗ N = 0 if and only if idM ⊗ idN = 0; if (ii) holds,
then idN = 0, hence N = 0.
The implications (iii)⇒(iv) and (iv)⇒(v) are obvious.
𝑢 𝑣
Let us prove the implication (iii)⇒(i). Let N1 → − N2 → − N3 be a
complex of A-modules which becomes exact after tensorization by M.
Let K = Ker(𝑣)/Im(𝑢).
Let 𝑗 : 𝑢(N1 ) → N2 be the inclusion; since M is flat, the morphism
idM ⊗𝑗 is injective and its image is a submodule of M ⊗A N2 isomorphic
to M ⊗A Im(𝑢). By right exactness, this submodule is also Im(idM ⊗𝑢).
𝑘 𝑣
Then consider the exact sequence 0 → Ker(𝑣) → − N2 , where 𝑘 is
− N2 →
the inclusion morphism, and apply the tensor product by M; since M is
flat, this gives an exact sequence
idM ⊗𝑘 idM ⊗𝑣
0 → M ⊗A Ker(𝑣) −−−−−→ M ⊗A N2 −−−−−→,
so that idM ⊗𝑘 is an isomorphism from M ⊗A Ker(𝑣) to Ker(idM ⊗𝑣).
By right exactness of the tensor product, one has
M ⊗A K = M ⊗A (Ker(𝑣)/Im(𝑢)) ' (M ⊗A Ker(𝑣))/(M ⊗A Im(𝑢)).
Consequently, M ⊗A K = 0, by assumption. Since M is faithfully flat, this
shows that K = 0 and the initial complex is exact.
We finally prove that (v)⇒(iii). Let N be an A-module such that
M ⊗A N = 0 and let us prove that N = 0. Let 𝑥 ∈ N and let I = {𝑎 ∈
A ; 𝑎𝑥 = 0}, so that the morphism 𝑎 ↦→ 𝑎𝑥 induces an injection A/I ↩→ N.
 8.7. FAITHFUL FLATNESS 423

By flatness of M, we thus have an injection M ⊗A (A/I) ↩→ M ⊗A N, so

that M ⊗A (A/I) = 0 and M = IM. If I ≠ A, there exists a maximal ideal P
of A that contains I; then M = IM ⊂ PM, which contradicts (v). Hence
I = A and 𝑥 = 0, so that N = 0. 

Definition (8.7.3). — One says that a morphism 𝑓 : A → B of commutative

rings is flat (resp. faithfully flat) if B is a flat (resp. faithfully flat) A-module.

Proposition (8.7.4). — Let 𝑓 : A → B be a flat morphism of commutative

rings. The following properties are equivalent:
(i) The morphism 𝑓 is faithfully flat;
(ii) The continuous map 𝑓 ∗ : Spec(B) → Spec(A) is surjective;
(iii) For every maximal ideal M of A, there exists a prime ideal P of B such
that M = 𝑓 −1 (P).

Proof. — Let P be a prime ideal of A and let KP be the fraction field of

the residue ring A/P. Note that KP = AP /PAP . Let T = 𝑓 (A P); it is a
multiplicative subset of B; by abuse of language, we write BP for T−1 B.
Let us consider the following commutative diagram of ring morphisms
𝑓
→ B → B/ 𝑓 (P)B
← ←
A
←

←
→
→

𝑓P
→ B 𝑓 (P) → BP / 𝑓 (P)BP ,
← ←
AP

where the unlabeled vertical arrows represent the obvious localization

morphisms, and the unlabeled horizontal arrows represent the obvi-
ous quotient maps. Moreover, one has an isomorphism BP / 𝑓 (P)BP '
T−1 (B/ 𝑓 (P)B) ' B ⊗A KP .
Under these morphisms, a prime ideal Q of BP / 𝑓 (P)BP correspond
to a prime ideal of BP that contains 𝑓 (P)BP , or to a prime ideal of B
containing 𝑓 (P) and disjoint from T, and its preimage 𝑓 −1 (Q) is a
prime ideal of A containing P and disjoint from A P, hence 𝑓 −1 (Q) = P.
Conversely, such an ideal furnishes a prime ideal of BP / 𝑓 (P)BP ' B ⊗A KP .
If 𝑓 is faithfully flat, then B ⊗A KP is nonzero, so that it admits a prime
ideal Q, and 𝑓 −1 (Q) = P belongs to the image of 𝑓 ∗ . This proves the
implication (i)⇒(ii).
424 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

If M is a maximal ideal, then KM = A/M and the above argument shows

that (iii) implies that B ⊗A (A/M) ≠ 0, that is, B ≠ MB. Consequently, B
is faithfully flat.
Finally, the implication (ii)⇒(iii) is obvious. 

Examples (8.7.5). — Let A be a commutative ring.

a) For every multiplicative subset S of A, the morphism 𝑓 : A → S−1 A
given by 𝑓 (𝑎) = 𝑎/1 is flat (example 8.6.3). It is faithfully flat if and only
if S consists of invertible elements, in which case it is an isomorphism.
Indeed, a prime ideal P of A is of the form 𝑓 −1 (Q) for a prime ideal Q
of S−1 A if and only if P ∩ S = ∅. If S contains a noninvertible element 𝑎,
then there is a prime ideal of A containing 𝑎, and this prime ideal does
not belong to the image of 𝑓 ∗ .
b) Let 𝑎1 , . . . , 𝑎 𝑛 be a family of elements of A, let B = 𝑛𝑖=1 A𝑎 𝑖 and
Î
let 𝑓 : A → B be the morphism given by 𝑓 (𝑎) = (𝑎/1, . . . , 𝑎/1) for every
𝑎 ∈ A. The morphism 𝑓 is flat.
It is faithfully flat if and only if, for every prime ideal P of A there
exists 𝑖 ∈ {1, . . . , 𝑛} such that 𝑎 𝑖 ∉ P, in other words, if and only if
Ñ𝑛
𝑖=1 V((𝑎 𝑖 )) = ∅ in Spec(A). Indeed, prime ideals of B are of the form
Q1 × · · · × Q𝑛 , where, for some 𝑖 ∈ {1, . . . , 𝑛}, Q𝑖 is a prime ideal of A𝑎 𝑖 ,
and Q 𝑗 = A𝑎 𝑗 otherwise. Its preimage by 𝑓 is then 𝑓 −1 (Q𝑖 ), a prime
ideal that does not contain 𝑎 𝑖 , and every such prime ideal of A can be
obtained.
Example (8.7.6). — Let 𝑓 : A → B be a flat morphism of commutative
rings. Assume that 𝑓 has a left inverse, that is, that there exists a
morphism 𝑓 0 : B → A such that 𝑓 0 ◦ 𝑓 = idA . Then 𝑓 is faithfully flat.
Indeed, one has 𝑓 ∗ ◦ ( 𝑓 0)∗ = idSpec(A) , hence 𝑓 ∗ is surjective.

8.7.7. — Let 𝑓 : A → B be a morphism of commutative rings. We

consider the diagram 𝒟 𝑓 of A-modules:
𝑓 𝑔
0→A→
− B→
− B ⊗A B,
where 𝑔(𝑏) = 1 ⊗ 𝑏 − 𝑏 ⊗ 1, for every 𝑏 ∈ B. For 𝑎 ∈ A, one has
1 ⊗ 𝑓 (𝑎) = 𝑓 (𝑎) ⊗ 1 = 𝑎 · (1 ⊗ 1) in B ⊗A B, so that 𝑔 ◦ 𝑓 = 0: the diagram 𝒟 𝑓
is a complex.
 8.7. FAITHFUL FLATNESS 425

More generally, for every A-module M, let 𝑓M = 𝑓 ⊗ idM : M → B ⊗A M

be the map given by 𝑚 ↦→ 1 ⊗ 𝑚, and let 𝑔M = 𝑔 ⊗ idM : B ⊗A M →
B ⊗A B ⊗A M be given by 𝑔M (𝑏 ⊗ 𝑚) = 1 ⊗ 𝑏 ⊗ 𝑚 − 𝑏 ⊗ 1 ⊗ 𝑚, and we
consider the diagram 𝒟M 𝑓
:
𝑓M 𝑔M
0 → M −−→ B ⊗A M −−→ B ⊗A B ⊗A M.
Again, 𝑔M ◦ 𝑓M = 0, so that 𝒟M
𝑓
is a complex of A-modules.

Proposition (8.7.8). — Assume that there exists a morphism of rings 𝑓 0 : B →

A such that 𝑓 0 ◦ 𝑓 = idA . Then the diagram 𝒟 𝑓 is an exact sequence. More
generally, for every A-module M, the complex 𝒟M 𝑓
is exact.

We will generalize this proposition below, in theorem 8.7.10, an prove

that its conclusion holds as soon as 𝑓 is faithfully flat.
Proof. — Since 𝑓 0 ◦ 𝑓 = idA , the morphism 𝑓 is injective.
One has 𝑔 ◦ 𝑓 = 0, hence Im( 𝑓 ) ⊂ Ker(𝑔). Conversely, the map
from B × B to B given by (𝑏, 𝑏 0) ↦→ 𝑓 0(𝑏)𝑏 0 is A-bilinear, hence there
exists a unique morphism of A-modules 𝑔 0 : B ⊗A B → B such that
𝑔 0(𝑏 ⊗ 𝑏 0) = 𝑓 0(𝑏)𝑏 0. One has 𝑔 0 ◦ 𝑔(𝑏) = 𝑔 0(1 ⊗ 𝑏) − 𝑔 0(𝑏 ⊗ 1) = 𝑏 − 𝑓 ( 𝑓 0(𝑏)),
hence 𝑔 0 ◦ 𝑔 = idB − 𝑓 ◦ 𝑓 0. Assume that 𝑔(𝑏) = 0. Then 𝑔 0(𝑔(𝑏)) = 0,
hence 𝑏 = 𝑓 ( 𝑓 0(𝑏)) and 𝑏 ∈ Im( 𝑓 ), hence Ker(𝑔) ⊂ Im( 𝑓 ). This proves
that Ker(𝑔) = Im( 𝑓 ).
Let us prove the more general result. Since the map from B × M to M
given by (𝑏, 𝑚) ↦→ 𝑓 0(𝑏)𝑚 is A-bilinear, there exists a unique morphism
of A-modules 𝑓M 0
: B ⊗A M → M such that 𝑓M 0
(𝑏 ⊗ 𝑚) = 𝑓 0(𝑏)𝑚. One has
𝑓M0
◦ 𝑓M (𝑚) = 𝑓M 0
(1 ⊗ 𝑚) = 𝑚 for every 𝑚 ∈ M, hence 𝑓M 0
◦ 𝑓M = idM . This
proves that 𝑓M is injective.
Similarly, there exists a unique morphism 𝑔M 0
: B ⊗A B ⊗A M → B ⊗A M
such that 𝑔M 0
(𝑏 ⊗ 𝑏 0 ⊗ 𝑚) = 𝑔 0(𝑏 ⊗ 𝑏 0) ⊗ 𝑚 = 𝑓 0(𝑏)𝑏 0 ⊗ 𝑚. One has
0 0
𝑔M ◦ 𝑔M (𝑏 ⊗ 𝑚) = 𝑔M (1 ⊗ 𝑏 ⊗ 𝑚 − 𝑏 ⊗ 1 ⊗ 𝑚)
= 𝑏 ⊗ 𝑚 − 𝑓 0(𝑏)1B0 ⊗ 𝑚
= (𝑏 − 𝑓 ( 𝑓 0(𝑏)) ⊗ 𝑚,
for every 𝑏, 𝑏 0 ∈ B and every 𝑚 ∈ M. Consequently, 𝑔M0
◦ 𝑔M = (idB − 𝑓 ◦
𝑓 )M . Let us now prove that Ker(𝑔M ) = Im( 𝑓M ). Since 𝑔M ◦ 𝑓M = 0, one
0

has Im( 𝑓M ) ⊂ Ker(𝑔M ).

426 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Conversely, let 𝜉 ∈ Ker(𝑔M ). Then 𝑔M

0
(𝑔M (𝜉)) = 0, that is 𝜉 = 𝑓M ◦ 𝑓M
0
(𝜉):
we have proved that 𝜉 ∈ Im( 𝑓M ). 

Corollary (8.7.9). — For every A-module M, the diagram of B-modules

idB ⊗ 𝑓M idB ⊗𝑔M
0 → B ⊗A M −−−−−→ B ⊗A M −−−−−−→ B ⊗A B ⊗A M
is exact.
Proof. — The map 𝑓B : B → B⊗A B is given by 𝑓B (𝑏) = 1⊗ 𝑏 is a morphism
of rings. There is a unique morphism of A-modules 𝑔 : B ⊗A B → B such
that 𝑔(𝑏 ⊗ 𝑏 0) = 𝑏𝑏 0 for every 𝑏, 𝑏 0 ∈ A; this is even a morphism of rings,
and one has 𝑔 ◦ 𝑓B = idB .
Moreover, the morphism 𝑔B : B ⊗A B → B ⊗A B ⊗A B is given by
𝑔B (𝑏 ⊗ 𝑏 0) = 1 ⊗ 𝑏 ⊗ 𝑏 0 − 𝑏 ⊗ 1 ⊗ 𝑏 0.
Under the isomorphism
∼
− B ⊗A B ⊗A B, (𝑏 ⊗ 𝑏 0) ⊗ (𝑐 ⊗ 𝑐 0) ↦→ 𝑏 ⊗ 𝑏 0 𝑐 ⊗ 𝑐,
(B ⊗A B) ⊗B (B ⊗A B) →
it identifies with the morphism 𝑔 associated with the morphism 𝑓B .
Consequently, the diagram of the question is an exact sequence. 

Theorem (8.7.10). — Let 𝑓 : A → B be a morphism of commutative rings. If

𝑓 is faithfully flat, then for every A-module M, the complex 𝒟M
𝑓
is exact.
Proof. — Let M be an A-module. The preceding corollary shows that
the complex 𝒟N𝑓
is exact, where N = B ⊗A M. On the other hand, the
complex 𝒟N
𝑓
identifies with the complex deduced from 𝒟M
𝑓
by tensoring
with B. Since B is a faithfully flat A-module, the complex 𝒟M
𝑓
is exact. 

8.8. Faithfully flat descent

8.8.1. — Let 𝑓 : A → B be a faithfully flat morphism of commutative
rings. Set C = B ⊗A B; there are two ring morphisms 𝑔1 , 𝑔2 : B → C,
respectively given by 𝑔1 (𝑏) = 𝑏 ⊗ 1 and 𝑔2 (𝑏) = 1 ⊗ 𝑏, so that 𝑔 = 𝑔2 − 𝑔1 .
Consequently, the exactness of the complex 𝒟 𝑓 asserted by theorem 8.7.10
says that the ring A can be recovered from B as the subring where these
two ring morphisms coincide.
 8.8. FAITHFULLY FLAT DESCENT 427

8.8.2. — Let M be an A-module, write MB = B ⊗A M and MC = C ⊗A M =

B ⊗A MB . By base change, these are a B-module and a C-module
respectively, and we may view MC as a B-module in two ways, via 𝑔1 or 𝑔2 .
The map 𝑔1M = 𝑔1 ⊗ idM : MB → MC maps 𝑏 ⊗ 𝑏 to 𝑏 ⊗ 1 ⊗ 𝑚 = 𝑔1 (𝑏) ⊗ 𝑚,
hence is a morphism of B-modules if C is viewed as a B-module via 𝑔1 ;
similarly, the map 𝑔2M = 𝑔2 ⊗ idM is a morphism of B-modules if C is
viewed as a B-module via 𝑔2 . Moreover, theorem 8.7.10 asserts that
M can be recovered from MB as the A-submodule where these two
morphisms coincide.
It is therefore natural to ask which properties of the A-module M can
be witnessed on the B-module MB . Similarly, a morphism 𝑢 : M → M0
of A-modules gives rise to a morphism of B-modules 𝑢B = idB ⊗𝑢, and
one may wonder what properties of 𝑢 are witnessed by 𝑢B .

Proposition (8.8.3). — Let 𝑓 : A → B be a faithfully flat morphism of commu-

tative rings. Let 𝑢 : M → N be a morphism of A-modules.
a) If 𝑢B = 0, then 𝑢 = 0;
b) If 𝑢B is surjective, then 𝑢 is surjective;
c) If 𝑢B is injective, then 𝑢 is injective.

Proof. — Assertion a) holds, by proposition 8.7.2.

Let us prove b) and c). Let K = Ker(𝑢) and let 𝑗 : K → M be the
inclusion; let P = Coker(𝑢) and let 𝑣 : N → P be the canonical projection.
𝑗 𝑢 𝑣
By construction, the diagram 0 → K →
− M→ − P → 0 is exact. Since
− N→
𝑗B 𝑢B 𝑣B
𝑓 is flat, the diagram 0 → KB − → MB −→ NB −→ PB → 0 is exact as well.
If 𝑢B is surjective, then PB = 0, hence P = 0 since 𝑓 is faithfully flat; this
shows that 𝑢 is surjective.
If 𝑢B is injective, then KB = 0, hence K = 0 since 𝑓 is faithfully flat, and
𝑢 is injective. 

Proposition (8.8.4). — Let 𝑓 : A → B be a faithfully flat morphism of commu-

tative rings and let M be an A-module.
a) If MB is a finitely generated B-module, then M is a finitely generated
A-module;
428 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

b) If MB is a finitely presented B-module, then M is a finitely presented

A-module;
c) If MB is a flat (resp. faithfully flat) B-module, then M is a flat (resp.
faithfully flat) A-module;
d) If MB is a finitely generated projective B-module, then B is a finitely
generated projective A-module;
e) If MB is a noetherian (resp. artinian) B-module, then M is a noetherian
(resp. artinian) A-module.

Proof. — a) Let (𝑚 𝑖 )𝑖∈I be a generating family of M. The family (1 ⊗ 𝑚 𝑖 )𝑖∈I

generates MB . Since MB is finitely generated, there exists a finite subset J
of I such that the family (1 ⊗ 𝑚 𝑖 )𝑖∈J generates MB . Let 𝑢 : AJ → M be the
morphism that maps (𝑎 𝑗 ) 𝑗∈J to 𝑎 𝑗 𝑚 𝑗 . Then 𝑢B : BJ → MB is surjective,
Í
hence 𝑢 is surjective. This proves that M is finitely generated.
b) By a), M is finitely generated. Let 𝑢 : A𝑛 → M be a surjective
morphism and let K = Ker(𝑢); we want to show that Ker(𝑢) is finitely
generated. The morphism 𝑢B : B𝑛 → MB is surjective, and its kernel is
finitely generated, because MB is finitely presented. Since 𝑓 is flat, one
has KB = Ker(𝑢B ). By a), K is finitely generated, as was to be shown.
c) We first show that M is flat if MB is flat. Let us consider a complex 𝒞,
𝑢 𝑣
N1 →− N2 → − N3 , of A-modules, and let us consider the complex 𝒞M :
idM ⊗𝑢 idM ⊗𝑣
M ⊗A N1 −−−−−→ M ⊗A N2 −−−−−→ M ⊗A N3 .
If we tensor it by B, we obtain the complex 𝒞M,B
idB ⊗ idM ⊗𝑢 idB ⊗ idM ⊗𝑣
B ⊗A M ⊗A N1 −−−−−−−−−→ B ⊗A M ⊗A N2 −−−−−−−−−→ B ⊗ M ⊗A N3 ,
which identifies with the complex
idMB ⊗𝑢B idMB ⊗𝑣B
MB ⊗B N1,B −−−−−−→ MB ⊗B N2,B −−−−−−→ MB ⊗B N3,B ,
which is deduced from the complex 𝒞B of B-modules
𝑢B 𝑣B
N1,B −→ N2,B −→ N3,B
by tensor product with MB . Assume that the complex 𝒞 is exact. Since B
is a flat A-module, the complex 𝒞B is exact; since MB is a flat B-module,
 8.8. FAITHFULLY FLAT DESCENT 429

the complex 𝒞M,B is exact as well. Since B is a faithfully flat A-module,

the complex 𝒞M is exact. This shows that M is a flat A-module.
Let us now assume that MB is faithfully flat. By what precedes, M is flat.
Assume that the complex 𝒞M is exact. Since B is flat, the complex 𝒞M,B
is flat. Since MB is faithfully flat, the complex 𝒞 is exact. This shows
that M is faithfully flat.
d) Since MB is finitely generated and projective, it is a finitely presented
and flat B-module. By b) and c), M is a finitely presented and flat
A-module. It then follows from corollary 8.6.13 that M is a finitely
generated projective A-module.
e) Let us assume that MB is noetherian Let (M𝑛 ) be an increasing
sequence of submodules of M. Since B is flat, the modules M𝑛,B identify
as submodules of MB and the sequence (M𝑛,B ) is increasing Consequently,
it is stationary and there exists 𝑚 ∈ N such that M𝑛,B = M𝑚,B for 𝑛 > 𝑚.
Since (M𝑛 /M𝑚 )B ' M𝑛,B /M𝑚,B , one has (M𝑛 /M𝑚 )B = 0 for 𝑛 > 𝑚; since
B is a faithfully flat A-module, one has M𝑛 /M𝑚 = 0, hence (M𝑛 ) is
stationary. This shows that M is a noetherian A-module.
The proof that M is artinian if MB is artinian is analogous; just replace
“increasing” by “decreasing” and the module M𝑛 /M𝑚 by the module
M𝑚 /M𝑛 in the preceding proof. 

Proposition (8.8.5). — Let 𝑓 : A → B be a faithfully flat morphism of commu-

tative rings. Then B/A is a flat A-module.

Proof. — Since 𝑓 is faithfully flat, it is injective. Let us consider the

𝑓 𝑔
exact sequence 0 → A → − B→− B/A → 0, where 𝑔 is the canonical
projection. By tensor product by B, we obtain the exact sequence
𝑓B 𝑔B
0 → B −→ B ⊗A B −→ B ⊗A (B/A) → 0, where 𝑓B (𝑏) = 𝑏 ⊗A 1 and
𝑔B (𝑏) = 𝑏 ⊗ 𝑔(𝑏). On the other hand, the morphism of rings ℎ : B⊗A B → B
such that ℎ(𝑏 ⊗ 𝑏 0) = 𝑏𝑏 0 satisfies ℎ ◦ 𝑓B = idB and is a morphism of B-
modules. Consequently, the latter exact sequence is split and B ⊗A (B/A)
is a direct summand of B ⊗A B. Since B is a flat A-module, B ⊗A B is a
flat B-module, and B ⊗A (B/A) is a flat B-module. By proposition 8.8.4,
B/A is a flat A-module. 
430 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Corollary (8.8.6). — Let 𝑓 : A → B be a morphism of integral domains that

induces an isomorphism on their fields of fractions. If 𝑓 is faithfully flat, then 𝑓
is an isomorphism.

Proof. — Since 𝑓 is faithfully flat, it is injective and we consider A as a

subring of B. Every element 𝑏 ∈ B can be written as a fraction 𝑎 0/𝑎 00, for
𝑎 0 , 𝑎 00 ∈ A and 𝑎 00 ≠ 0, so that the image of 𝑏 in B/A is torsion. Since B/A
is a flat A-module, it is torsion-free (corollary 8.6.9), so that 𝑏 ∈ A. This
proves that B = A. 

8.8.7. — Let 𝑓 : A → B be a faithfully flat morphism of commutative

rings. Following the impetus of Grothendieck (1960), we can go further
and try to recover:
a) A morphism 𝑢 : M → M0 of A-modules from the B-morphism
𝑢 ⊗ idB : M ⊗A B → M0 ⊗A B, in particular, describe which morphisms
𝑣 : M ⊗A B → M0 ⊗A B are of the form 𝑢 ⊗ idB ;
b) An A-module M from the B-module M ⊗A B and an additional
datum.
This process has been coined faithfully flat descent by Grothendieck.

8.8.8. — The map B × B → B ⊗A B given by (𝑏, 𝑏 0) ↦→ 𝑏 0 ⊗ 𝑏 is A-bilinear,

hence there exists a morphism 𝜏 : B ⊗A B → B ⊗A B of A-modules such
that 𝜏(𝑏 ⊗ 𝑏 0) = 𝑏 0 ⊗ 𝑏 for all 𝑏, 𝑏 0 ∈ B. This is a morphism of A-algebras.
For 𝑏 ∈ B, notice the relation

𝑔(𝑏) = 1 ⊗ 𝑏 − 𝑏 ⊗ 1 = (id −𝜏)(1 ⊗ 𝑏).

Let M be an A-module. We let 𝜏M = 𝜏 ⊗ idM be the automorphism of

B ⊗A B ⊗A M. Observe that although the A-module B ⊗A B ⊗A M can be
viewed as a B ⊗A B-module, the morphism 𝜏M is not B ⊗A B-linear, but
satisfies the relation

𝜏M (𝛽 · 𝜇) = 𝜏(𝛽) · 𝜏M (𝜇)

for 𝛽 ∈ B ⊗A B and 𝜇 ∈ B ⊗A B ⊗A M. We say that 𝜏M is 𝜏-linear. Indeed,

if 𝛽 = 𝑏 1 ⊗ 𝑏 10 and 𝜇 = 𝑏 ⊗ 𝑏 0 ⊗ 𝑚 are split tensors, this just follows from
 8.8. FAITHFULLY FLAT DESCENT 431

the equalities
𝜏M (𝛽 · 𝜇) = 𝜏M ((𝑏 1 ⊗ 𝑏 10 ) · (𝑏 ⊗ 𝑏 0 ⊗ 𝑚)) = 𝜏M (𝑏 1 𝑏 ⊗ 𝑏 10 𝑏 0 ⊗ 𝑚)
= 𝑏 10 𝑏 0 ⊗ 𝑏 1 𝑏 ⊗ 𝑚 = (𝑏 10 ⊗ 𝑏 1 ) · (𝑏 0 ⊗ 𝑏 ⊗ 𝑚)
= 𝜏(𝑏1 ⊗ 𝑏10 ) · 𝜏M (𝑏 ⊗ 𝑏 0 ⊗ 𝑚) = 𝜏(𝛽) · 𝜏M (𝜇).
For 𝑏 ∈ B and 𝑚 ∈ M, notice the relation
𝑔M (𝑏 ⊗ 𝑚) = 1 ⊗ 𝑏 ⊗ 𝑚 − 𝑏 ⊗ 1 ⊗ 𝑚 = (id −𝜏M )(1 ⊗ 𝑏 ⊗ 𝑚).
Consequently, for 𝜇 ∈ B ⊗A M, one has 𝑔M (𝜇) = (id −𝜏M )(1 ⊗ 𝜇).

Proposition (8.8.9). — Let 𝑓 : A → B be a faithfully flat morphism of com-

mutative rings and let M, M0 be A-modules. The map 𝑢 ↦→ idB ⊗𝑢 from
HomA (M, M0) to HomB (M ⊗A B, M0 ⊗A B) is injective; its image is the the
set of all B-morphisms 𝑣 : B ⊗A M → B ⊗A M0 such that (idB ⊗𝑣) ◦ 𝜏M =
𝜏M0 ◦ (idB ⊗𝑣).

Proof. — The injectivity of this map follows from proposition 8.8.3, a).
Let 𝑣 ∈ HomB (B ⊗A M, B ⊗A M0). If the morphism 𝑣 is of the form idB ⊗𝑢,
for 𝑢 ∈ HomA (M, M0), then for all 𝑏, 𝑏 0 ∈ B and 𝑚 ∈ M, one has
𝜏M0 ◦ (idB ⊗𝑣)(𝑏 ⊗ 𝑏 0 ⊗ 𝑚) = 𝜏M0 (𝑏 ⊗ 𝑏 0 ⊗ 𝑢(𝑚))
= 𝑏 0 ⊗ 𝑏 ⊗ 𝑢(𝑚)
= (idB ⊗𝑣)(𝑏 0 ⊗ 𝑏 ⊗ 𝑚)
= (idB ⊗𝑣) ◦ 𝜏M (𝑏 ⊗ 𝑏 0 ⊗ 𝑚).
This proves that the two morphisms (idB ⊗𝑣) ◦ 𝜏M and 𝜏M0 ◦ (idB ⊗𝑣)
coincide on split tensors, hence are equal, as was to be shown.
Conversely, let us assume that (idB ⊗𝑣) ◦ 𝜏M = 𝜏M0 ◦ (idB ⊗𝑣). Then,
for every 𝑚 ∈ M, one has
𝜏M0 (1 ⊗ 𝑣( 𝑓M (𝑚))) = 𝜏M0 ◦ (idB ⊗𝑣)(1 ⊗ 1 ⊗ 𝑚)
= (idB ⊗𝑣) ◦ 𝜏M (1 ⊗ 1 ⊗ 𝑚)
= 1 ⊗ 𝑣(1 ⊗ 𝑚) = 1 ⊗ 𝑣( 𝑓M (𝑚)),
and 𝑔M0 (𝑣( 𝑓 (𝑚))) = 0. Consequently, the image of the morphism 𝑣 ◦
𝑓M : M → B ⊗A M0 is contained in Ker(𝑔M0 ). Since 𝑓M0 is an isomorphism
from M0 to Ker(𝑔M0 ), by theorem 8.7.10, this implies that there exists a
432 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

morphism 𝑢 : M → M0 such that 𝑣 ◦ 𝑓M = 𝑓M0 ◦ 𝑢. Then, for 𝑏 ∈ B and

𝑚 ∈ M, one has
𝑣(𝑏 ⊗ 𝑚) = 𝑣(𝑏 · (1 ⊗ 𝑚)) = 𝑏 · 𝑣(1 ⊗ 𝑚) = 𝑏 · (1 ⊗ 𝑢(𝑚)) = 𝑏 ⊗ 𝑢(𝑚),
so that 𝑣 and idB ⊗𝑢 coincide on split tensors, hence are equal. 

8.8.10. — If a B-module N is of the form B ⊗A M, then 𝜏M is a 𝜏-linear

automorphism of the B ⊗A B-module B ⊗A N = B ⊗A B ⊗A M.
In fact, this automorphism 𝜏M satisfies an additional relation that we
have to make explicit.
To every 𝜏-linear endomorphism 𝜃 of a B ⊗A B-module B ⊗A N, we
associate two endomorphisms of the A-module B ⊗A B ⊗A N, respectively
given by
𝜃1 = idB ⊗𝜃 : 𝑏 ⊗ 𝑏 0 ⊗ 𝑛 ↦→ 𝑏 ⊗ 𝜃(𝑏 0 ⊗ 𝑛)
and
𝜏1 = 𝜏 ⊗ idN : 𝑏 ⊗ 𝑏 0 ⊗ 𝑛 ↦→ 𝑏 0 ⊗ 𝑏 ⊗ 𝑛.
The maps 𝜎 = idB ⊗𝜏 and 𝜎0 = 𝜏 ⊗ idB are automorphisms of the A-
algebra B⊗A B⊗A B. For 𝑏, 𝑏 0 and 𝑏 00 ∈ B, one has 𝜎(𝑏 ⊗𝑏 0 ⊗𝑏 00) = 𝑏 ⊗𝑏 00 ⊗𝑏 0
and 𝜎0(𝑏 ⊗ 𝑏 0 ⊗ 𝑏 00) = 𝑏 0 ⊗ 𝑏 ⊗ 𝑏 00: the automorphism 𝜎 swaps the last
two components, while the automorphism 𝜎0 swaps the first two ones.
A reflexion of the braid identity
(2 3)(1 2)(2 3) = (1 3) = (1 2)(2 3)(1 2)
in the symmetric group 𝔖3 , one has
𝜎 ◦ 𝜎0 ◦ 𝜎 = 𝜎0 ◦ 𝜎 ◦ 𝜎0 .
If N = B ⊗A M and 𝜃 = 𝜏M = 𝜏 ⊗ idM , then 𝜃1 = 𝜎 ⊗ idM and
𝜏1 = 𝜎0 ⊗ idM , so that
𝜃1 ◦ 𝜏1 ◦ 𝜃1 = 𝜏1 ◦ 𝜃1 ◦ 𝜏1 .

Definition (8.8.11). — Let 𝑓 : A → B be a morphism of rings and let 𝜏 : B ⊗A

B → B ⊗A B be the automorphism of A-algebras such that 𝜏(𝑏 ⊗ 𝑏 0) = 𝑏 0 ⊗ 𝑏
for 𝑏, 𝑏 0 ∈ B. A descent datum of a B-module N with respect to the morphism
𝑓 : A → B is a 𝜏-linear automorphism 𝜃 of B ⊗A N such that
𝜃1 ◦ 𝜏1 ◦ 𝜃1 = 𝜏1 ◦ 𝜃1 ◦ 𝜏1 .
 8.8. FAITHFULLY FLAT DESCENT 433

If N, N0 are B-modules endowed with descent data 𝜃, 𝜃0 with respect to 𝑓 , a

B-morphism 𝑢 : N → N0 is compatible with the given descent data if one has
(idB ⊗𝑢) ◦ 𝜃 = 𝜃0 ◦ (idB ⊗𝑢).

With this definition, we can reformulate the preceding results as

follows, for a ring morphism 𝑓 : A → B:
a) If M is an A-module, the map 𝜏M = 𝜏 ⊗ idM is a descent datum of
the B-module B ⊗A M with respect to 𝑓 ;
b) If M, M0 are A-modules, and if 𝑓 is faithfully flat, then the map
𝑢 ↦→ idB ⊗𝑢 induces a bijection from HomA (M, M0) to the set of all
B-morphisms HomB (B ⊗A M, B ⊗A M0) which are compatible with the
descent data 𝜏M and 𝜏M0 .
One says that 𝜏M is the canonical descent datum of the B-module B ⊗A M.
Observe that B-modules N endowed with a descent datum 𝜃 form a
category: objects are pairs (N, 𝜃), and a morphism from (N, 𝜃) to (N0 , 𝜃0)
is a B-morphism 𝑓 : N → N0 which is compatible with the desent datas.
Let us denote this category by Mod B/A . By what precedes, the tensor
product functor M ↦→ (B ⊗A M, 𝜏M ) is a functor from Mod A to Mod B/A .
In this language, proposition 8.8.9 says that this functor is fully faithful;
the following theorem of Grothendieck implies that it is an equivalence
of categories.

Theorem (8.8.12) (Faithfully flat descent; Grothendieck)

Let 𝑓 : A → B be a faithfully flat morphism of rings. Let N be a B-module,
let 𝜃 be a descent datum of N with respect to 𝑓 and let M be the subset of all
𝑥 ∈ N such that 𝜃(1 ⊗ 𝑥) = 1 ⊗ 𝑥. Then the following properties hold:
a) M is an A-submodule of N;
b) There exists a unique morphism 𝜓 : B⊗A M → N such that 𝜓(𝑏 ⊗ 𝑥) = 𝑏𝑥
for all 𝑏 ∈ B and all 𝑥 ∈ M;
c) The morphism 𝜓 is compatible with the descent datum 𝜏M of B ⊗A M and
the given descent datum 𝜃 of N;
d) The morphism 𝜓 is an isomorphism of B-modules.

Proof. — a) Since the morphism idB ⊗𝜃 is A-linear, the set M is an

A-submodule of N.
434 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

b) The map from B ×A M to N given by (𝑏, 𝑥) ↦→ 𝑏𝑥 is A-multilinear,

hence there exists a unique morphism 𝜓 : B ⊗A M → N such that
𝜓(𝑏 ⊗ 𝑚) = 𝑏𝑚 for all 𝑏 ∈ B and all 𝑚 ∈ M.
c) Let us show that 𝜓 is compatible with the given descent data. Let
𝑏, 𝑏 0 ∈ B and 𝑚 ∈ M; then
𝜃 ◦ (idB ⊗𝜓)(𝑏 ⊗ 𝑏 0 ⊗ 𝑚) = 𝜃(𝑏 ⊗ 𝜓(𝑏 0 ⊗ 𝑚)) = 𝜃(𝑏 ⊗ 𝑏 0 𝑚)
= 𝜃((𝑏 ⊗ 𝑏 0)(1 ⊗ 𝑚)) = 𝜏(𝑏 ⊗ 𝑏 0) · 𝜃(1 ⊗ 𝑚)
= (𝑏 0 ⊗ 𝑏) · (1 ⊗ 𝑚) = 𝑏 0 ⊗ 𝑏𝑚.
On the other hand,
(idB ⊗𝜓) ◦ 𝜏M (𝑏 ⊗ 𝑏 0 ⊗ 𝑚) = (idB ⊗𝜓)(𝑏 0 ⊗ 𝑏 ⊗ 𝑚) = 𝑏 0 ⊗ 𝑏𝑚,
so that 𝜃 ◦ (idB ⊗𝜓) = (idB ⊗𝜓) ◦ 𝜏M , since these two A-morphisms
coincide on split tensors. This proves that 𝜓 is compatible with the given
descent data.
d) By definition, M is the kernel of the morphism 𝑓N − 𝜃 ◦ 𝑓N , so that
we have an exact sequence
𝑗 𝑓N
→ M → N → B ⊗A N
← ← ←
0
−𝜃◦ 𝑓N

of A-modules, where 𝑗 is the inclusion of M into N. Since B is a faithfully

flat A-module, tensoring this exact sequence by B furnishes again an
exact sequence, which is the upper row of the following diagram, the
second row being the exact sequence 𝒟N 𝑓
:

idB ⊗ 𝑗 idB ⊗ 𝑓N
→ B ⊗A M → B ⊗A N → B ⊗A B ⊗A N
← ← ←
0
− idB ⊗𝜃◦ 𝑓N
←

𝜓 𝜃 𝜃1
→

𝑓N 𝑔N
→ N → B ⊗A N → B ⊗A B ⊗A N.
← ← ←
0
Let us check that the diagram is commutative. For 𝑏 ∈ B and 𝑚 ∈ M,
one has
𝜃 ◦ (idB ⊗𝑗)(𝑏 ⊗ 𝑚) = 𝜃(𝑏 ⊗ 𝑚) = 𝜃 (𝑏 ⊗ 1) · (1 ⊗ 𝑚)

= (1 ⊗ 𝑏)𝜃(1 ⊗ 𝑚) = (1 ⊗ 𝑏) · (1 ⊗ 𝑚)
= 1 ⊗ 𝑏𝑚 = 𝑓N ◦ 𝜓(𝑏 ⊗ 𝑚),
 8.8. FAITHFULLY FLAT DESCENT 435

so that the morphisms 𝜃 ◦ (idB ⊗𝑗) and 𝑓N ◦ 𝜓 coincide on split tensors;

they are thus equal, hence the commutativity of the left square.
Recall that one has 𝑔N = 𝑔N 1
− 𝑔N
2
, where 𝑔N
1
(𝑏 ⊗ 𝑛) = 1 ⊗ 𝑏 ⊗ 𝑛 and
𝑔N (𝑏 ⊗ 𝑛) = 𝑏 ⊗ 1 ⊗ 𝑛, for 𝑏 ∈ B and 𝑛 ∈ N. We first prove that
2

𝜃1 ◦ idB ⊗ 𝑓N = 𝑔B1 ◦ 𝜃. Indeed, for 𝑏 ∈ B and 𝑛 ∈ N, one has

𝜃1 ◦ 𝜏1 ◦ (idB ⊗ 𝑓N )(𝑏 ⊗ 𝑛) = 𝜃1 ◦ 𝜏1 (𝑏 ⊗ 1 ⊗ 𝑛)
= 𝜃1 (1 ⊗ 𝑏 ⊗ 𝑛)
= 1 ⊗ 𝜃(𝑏 ⊗ 𝑛)
= 𝑔N
1
◦ 𝜃(𝑏 ⊗ 𝑛).
This proves that 𝜃1 ◦ 𝜏1 ◦ (idB ⊗ 𝑓N ) = 𝑔N
1
◦ 𝜃.
Let 𝑏 ∈ B and 𝑛 ∈ N; let (𝑏 𝑖 ) and (𝑛 𝑖 ) be finite families such that
𝜃(1 ⊗ 𝑛) = 𝑏 𝑖 ⊗ 𝑛 𝑖 ; then
Í
Õ
𝜃(𝑏 ⊗ 𝑛) = 𝜃 (𝑏 ⊗ 1) · (1 ⊗ 𝑛) = (1 ⊗ 𝑏) · 𝜃(1 ⊗ 𝑛) = 𝑏 𝑖 ⊗ 𝑏𝑛 𝑖 .

Moreover,
𝜏1 ◦ 𝜃1 ◦ 𝜏1 (𝑏 ⊗ 1 ⊗ 𝑛) = 𝜏1 ◦ 𝜃1 (1 ⊗ 𝑏 ⊗ 𝑛)
= 𝜏1 (1 ⊗ 𝜃(𝑏 ⊗ 𝑛))
Õ
= 𝜏1 ( 1 ⊗ 𝑏 𝑖 ⊗ 𝑏𝑛 𝑖 )
Õ
= 𝑏 𝑖 ⊗ 1 ⊗ 𝑏𝑛 𝑖
Õ
𝑔N 𝑏 𝑖 ⊗ 𝑏𝑛 𝑖
2


=
= 𝑔N
2
◦ 𝜃(𝑏 ⊗ 𝑛).
In other words,
𝑔N
2
◦ 𝜃 = 𝜏1 ◦ 𝜃1 ◦ 𝜏1 ◦ 𝑔N
2
.
On the other hand,
(idB ⊗𝜃 ◦ 𝑓N )(𝑏 ⊗ 𝑛) = 𝑏 ⊗ 𝜃(1 ⊗ 𝑛)
= 𝜃1 (𝑏 ⊗ 1 ⊗ 𝑛)
= 𝜃1 ◦ 𝑔N
2
(𝑏 ⊗ 𝑛),
so that
𝜃1 ◦ 𝜏1 ◦ (idB ⊗𝜃 ◦ 𝑓N ) = 𝜃1 ◦ 𝜏1 ◦ 𝜃1 ◦ 𝑔N
2
.
436 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Thus the relation

𝜃1 ◦ 𝜏1 ◦ (idB ⊗(𝜃 ◦ 𝑓N )) = 𝑔N
2
◦𝜃
follows from the definition of a descent datum, and the right hand
square is commutative as well.
We now deduce from the snake lemma (theorem 7.2.1) that 𝜓 is
an isomorphism. In the above diagram, replace the source of the
morphism 𝜃1 with the image of the morphism idB ⊗( 𝑓N − 𝜃 ◦ 𝑓N ), and
the morphism 𝜃1 with its restriction 𝜃10 , which is then injective. Since
idB ⊗𝑗 is injective, the exact sequence of theorem 7.2.1 then writes
0 → Ker(𝜓) → Ker(𝜃) → Ker(𝜃10 ) → Coker(𝜓) → Coker(𝜃).
Since Ker(𝜃) = Ker(𝜃10 ) = Coker(𝜃) = 0, we conclude that Ker(𝜓) =
Coker(𝜓) = 0, hence 𝜓 is an isomorphism, as was to be shown. 

8.9. Galois descent

Proposition (8.9.1). — 2 Let K → L be a finite extension of fields. The following
properties are equivalent:
a) The extension is Galois;
b) There exists a finite set I and an isomorphism of L-algebras L ⊗K L ' LI .
Assume that they hold and let G = Gal(L/K). There exists a unique morphism
of K-algebras from L ⊗K L to LG such that 𝑎 ⊗ 𝑏 ↦→ (𝑎𝜎(𝑏))𝜎∈G , and this
morphism is an isomorphism of L-algebras.

Proof. — Assume that the extension K → L is Galois and let G =

Gal(L/K). The map from L × L to LG given by (𝑎, 𝑏) ↦→ (𝑎𝜎(𝑏))𝜎∈G is K-
bilinear, hence there exists a unique morphism of K-algebras 𝜑 : L⊗K L →
LG such that 𝜑(𝑎 ⊗ 𝑏) = (𝑎𝜎(𝑏))𝜎∈G for all 𝑎, 𝑏 ∈ L. It is also L-linear
when L ⊗K L is viewed as an L-algebra via the morphism 𝑎 ↦→ 𝑎 ⊗ 1.
Let us prove that 𝜑 is injective. Let (𝑏1 , . . . , 𝑏 𝑑 ) be a basis of L as
a K-vector space; then (1 ⊗ 𝑏 1 , . . . , 1 ⊗ 𝑏 𝑑 ) is a basis of L ⊗K L as an
L-vector space. Let (𝑎1 , . . . , 𝑎 𝑑 ) ∈ L𝑑 be such that 𝜑( 𝑎 𝑖 (1 ⊗ 𝑏 𝑖 )) = 0.
Í
By assumption, this implies that 𝑑𝑖=1 𝑎 𝑖 𝜎(𝑏 𝑖 ) = 0 for every 𝜎 ∈ G. As
Í

2Mettre la réciproque en exercice ?

 8.9. GALOIS DESCENT 437

a consequence, 𝑑𝑖=1 𝑎 𝑖 𝜎(𝑦) = 0 for every 𝑦 ∈ L, so that 𝑑𝑖=1 𝑎 𝑖 𝜎 = 0 in

Í Í
the set of K-linear morphism from L to L. By linear independence of
characters, 𝑎 1 = · · · = 𝑎 𝑑 = 0.
Since the extension K → L is Galois, one has Card(G) = [L : K] = 𝑑,
hence dimL (L ⊗K L) = dimL (LG ) so that the injective L-linear map 𝜑 is
an isomorphism.
Let us now assume that there exists a finite set I and an isomorphism
of L-algebras 𝜑 : L ⊗K L ' LI . Let Ω be an algebraic closure of L. Using
the associativity isomorphism Ω ⊗L L ⊗K L ' Ω ⊗K L, the morphism
idΩ ⊗𝜑 induces an isomorphism of Ω-algebras Ω ⊗K L ' ΩI . Let 𝑏 ∈ L
and let P ∈ K[T] be its minimal polynomial; then K[𝑏] is a sub-algebra
of L isomorphic to K[T]/(P), and Ω[T]/(P) identifies with a sub-algebra
of ΩI . Since ΩI is reduced, this implies that P has only simple roots
in Ω: it is a separable polynomial, and 𝑏 is separable. Consequently,
L is a separable extension of K. Let us now use the primitive element
and choose 𝑏 ∈ L such that L = K[𝑏]; let P be its minimal polynomial.
Then LI ' L ⊗K L ' L[T]/(P). On the other hand, if P = 𝑚
Î
𝑖=1 P𝑖 is the
decomposition of P as a product of irreducible polynomials in L[T],
we get an isomorphism LI ' 𝑚
Î
𝑖=1 L[T]/(P𝑖 ). The maximal ideals of
the left hand side are of the kernels of the projections from LI to L
and their residue fields are L; the maximal ideals of the right hand
side are the kernels of the projections to L[T]/(P𝑖 ), for 𝑖 ∈ {1, . . . , 𝑚},
and their residue fields are L[T]/(P𝑖 ). This implies that deg(P𝑖 ) = 1
for every 𝑖 ∈ {1, . . . , 𝑚}: the polynomial P is split in L[T]. Let us
write P = 𝑚 𝑖=1 (T − 𝑏 𝑖 ). For every 𝑖, there exists a unique morphism
Î
of K-algebras 𝜎𝑖 from L = K[𝑏] to itself such that 𝜎𝑖 (𝑏) = 𝑏 𝑖 . Since the
extension K → L is finite, these are isomorphisms, hence are elements of
Gal(L/K). This shows that Card(Gal(L/K)) = 𝑚 = [L : K]; the extension
K → L is then Galois. 

8.9.2. — Let 𝑓 : K → L be a Galois extension and let G be its Galois

group. Since every module over a field is flat, this morphism 𝑓 is flat,
and it is even faithfully flat since Spec(K) and Spec(K) are both reduced
to one element. Faithfully flat descent asserts that the tensor product
functor from K-vector spaces to L-vector spaces, V ↦→ L ⊗K V, induces
438 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

an equivalence of categories from Mod K to the category Mod L/K of

L-vector spaces W endowed with descent data 𝜃.
The definition of a descent datum involves L ⊗K L-modules and
L ⊗K L ⊗K L-modules, and the goal here is to reformulate it in terms of
L-vector spaces plus various maps indexed by G, using proposition 8.9.1
to replace the algebra L ⊗K L with the isomorphic algebra LG . The result
of this translation will be called Galois descent.
In fact, while the proposition describes an isomorphism L ⊗K L ' LG
of L-algebras, when L ⊗K L is viewed as an L-algebra by the morphism
𝑎 ↦→ 𝑎 ⊗ 1, it will be more convenient in what follows to view L ⊗K L as
an L-algebra by the morphism 𝑏 ↦→ 1 ⊗ 𝑏. It is also isomorphic to LG ,
but via the morphism 𝜑 : L ⊗K L → LG such that 𝜑(𝑎 ⊗ 𝑏) = (𝜎(𝑎)𝑏) for
every 𝑎, 𝑏 ∈ L.

8.9.3. — For every 𝜎 ∈ G, we let 𝛿 𝜎 = (𝛿 𝜎,𝜏 )𝜏∈G be the element of LG

such that 𝛿 𝜎,𝜏 = 1 if 𝜏 = 𝜎 and 𝛿 𝜎,𝜏 = 0 otherwise. One has 𝛿 𝜎 𝛿 𝜏 = 0 if
𝜎 ≠ 𝜏, and 𝛿 2𝜎 = 𝛿 𝜎 ; moreover, 1 = 𝜎∈G 𝛿 𝜎 .
Í
Let 𝜀𝜎 ∈ L ⊗K L be the unique element such that 𝜑(𝜀𝜎 ) = 𝛿 𝜎 . In
particular, for all 𝜎, 𝜏 ∈ G, one has 𝜀𝜎 𝜀𝜏 = 0 if 𝜎 ≠ 𝜏, and 𝜀2𝜎 = 𝜀𝜎 ;
moreover, 1 = 𝜎∈G 𝜀𝜎 .
Í

Lemma (8.9.4). — Let 𝜃 : L ⊗K L → L ⊗K L be the automorphism of K-

algebras such that 𝜃(𝑎 ⊗ 𝑏) = 𝑏 ⊗ 𝑎 for every 𝑎, 𝑏 ∈ L. Then 𝜑 ◦ 𝜃 ◦ 𝜑 −1 is
the automorphism (𝑎 𝜎 ) ↦→ (𝜎(𝑎 𝜎−1 )).

Proof. — Let (𝑥 𝜎 ) ∈ LG . Then (𝑥 𝜎 ) = 𝜎∈G 𝛿 𝜎 𝑥 𝜎 , so that (𝑥 𝜎 ) = 𝜑(𝑥),

Í
where 𝑥 = 𝜎∈G 𝜀𝜎 (1 ⊗ 𝑥 𝜎 ). For every 𝜎 ∈ G, fix a representation 𝜀𝜎 =
Í
Í 𝜎 𝜎
𝑎 𝑖 ⊗ 𝑏 𝑖 . Then 𝑥 = 𝜎∈G 𝑖 𝑎 𝑖𝜎 ⊗ 𝑏 𝑖𝜎 𝑥 𝜎 , so that 𝜃(𝑥) = 𝜎∈G 𝑖 𝑏 𝑖𝜎 𝑥 𝜎 ⊗ 𝑎 𝑖𝜎 .
Í Í Í Í
Let 𝑦 = (𝑦𝜏 )𝜏∈G = 𝜑(𝜃(𝑥)) = 𝜑(𝜃(𝜑 −1 (𝑥 𝜎 ))). By definition, one has
ÕÕ ÕÕ
𝑦𝜏 = 𝜏(𝑏 𝑖𝜎 𝑥 𝜎 )𝑎 𝑖𝜎 =𝜏 𝜏−1 (𝑎 𝑖𝜎 )𝑏 𝑖𝜎 𝑥 𝜎


𝜎∈G 𝑖 𝜎∈G 𝑖
Õ
=𝜏 𝜑(𝜀𝜎 )𝜏−1 𝑥 𝜎 = 𝜏(𝑥 𝜏−1 ),


𝜎∈G

as was to be shown. 

Proposition (8.9.5). — Let W be an L-vector space.

 8.9. GALOIS DESCENT 439

a) There exists a unique morphism of K-vector spaces 𝜑W : L ⊗K W → WG

such that 𝜑W (𝑎 ⊗ 𝑣) = (𝜎(𝑎)𝑣)𝜎∈G for all 𝑎 ∈ L and all 𝑣 ∈ W. It is bijective.
b) A map 𝜆 : L ⊗K W → L ⊗K W is a descent datum if and only if there
exists a family (𝜆 𝜎 )𝜎∈G satisfying the properties:
(i) For every 𝜎 ∈ G, 𝜆 𝜎 is a bijective K-linear map W → W such that
𝜆 𝜎 (𝑎𝑣) = 𝜎(𝑎)𝜆 𝜎 (𝑣) for every 𝑎 ∈ L and every 𝑣 ∈ W;
(ii) One has 𝜑W ◦ 𝜆 ◦ 𝜑W
−1
((𝑣 𝜎 )) = (𝜆 𝜎 (𝑣 𝜎−1 )) for every (𝑣 𝜎 ) ∈ WG ;
(iii) For every 𝜎, 𝜏 ∈ G, one has 𝜆 𝜎𝜏 = 𝜆 𝜎 ◦ 𝜆𝜏 .

Proof. — a) The map L × W → WG given by (𝑎, 𝑣) ↦→ (𝜎(𝑎)𝑣)𝜎∈G is

K-bilinear; consequently, there exists a unique morphism 𝜑W : L ⊗K W →
WG such that 𝜑W (𝑎 ⊗ 𝑣) = (𝜎(𝑎)𝑣).
By definition, one has 𝜑W ((𝑎 ⊗ 𝑏)·(𝑐 ⊗ 𝑣)) = 𝜑W (𝑎𝑐 ⊗ 𝑏𝑣) = (𝜎(𝑎𝑐)𝑏𝑣)𝜎 =
𝜑(𝑎 ⊗ 𝑏)𝜑W (𝑐 ⊗ 𝑣). This proves that 𝜑W (𝛼 · 𝜉) = 𝜑(𝛼)𝜑W (𝜉) for every
𝛼 ∈ L ⊗K L and every 𝜉 ∈ L ⊗K W.
In particular, 𝜑W (𝜀𝜎 (1⊗𝑣)) = 𝛿 𝜎 (𝑣, . . . , 𝑣). Since (𝑣 𝜎 ) = 𝜎∈G 𝛿 𝜎 (𝑣 𝜎 , . . . ),
Í
this implies that (𝑣 𝜎 ) = 𝜎∈G 𝜑W (𝜀𝜎 (1 ⊗ 𝑣 𝜎 )). Therefore, the K-linear
Í
map 𝜓W : WG → L ⊗K W given by 𝜓W ((𝑣 𝜎 )) = 𝜎∈G 𝜀𝜎 (1 ⊗ 𝑣 𝜎 ) satisfies
Í
𝜑W ◦ 𝜓W = id; in particular, 𝜑W is surjective. If dimK (W) is finite,
then dimK (L ⊗K W) = [L : K] dimK (W) = dimK (WG ) so that 𝜑W is an
isomorphism. Let 𝜉 be an element of Ker(𝜑W ). Let us write 𝜉 = 𝑎 𝑖 ⊗ 𝑣 𝑖
Í
and let W1 be the subspace of W generated by the family (𝑣 𝑖 ); it is finitely
dimensional. One has 𝜉 ∈ L ⊗K W1 , hence 𝜑W1 (𝜉) = 0, hence 𝜉 = 0. We
thus have proved that 𝜑W is bijective.
b) Let 𝜆 : L ⊗K W → L ⊗K W be a K-linear map. We express in terms
of the K-linear map 𝜆0 = 𝜑W ◦ 𝜆 ◦ 𝜑W −1
the conditions for 𝜆 to be a descent
datum.
Let 𝜃0 = 𝜑 ◦ 𝜃 ◦ 𝜑−1 ; by lemma 8.9.4, one has 𝜃0((𝑎 𝜎 )) = (𝜎(𝑎 𝜎−1 )) for
every (𝑎 𝜎 ) ∈ LG . In particular, 𝜃0(𝛿 𝜎 ) = 𝛿 𝜎−1 .
First of all, 𝜆 has to be bijective and 𝜃-linear, that is, 𝜆(𝛼𝜉) = 𝜏(𝛼)𝜆(𝜉)
for every 𝛼 ∈ L ⊗K L and every 𝜉 ∈ L ⊗K W. Since 𝜑W is an isomorphism,
this means 𝜆0(𝑎𝑣) = 𝜃0(𝑎)𝜆0(𝑣) for every 𝑎 ∈ LG and every 𝑣 ∈ WG .
Let us assume that this relation holds. Let 𝑣 ∈ WG ; since 1 = 𝜎∈G 𝛿 𝜎 ,
Í
one has 𝑣 = 𝜎∈G 𝛿 𝜎 𝑣 = 𝜎∈G 𝛿 𝜎 𝑣 𝜎 , where, by abuse, we write 𝛿 𝜎 𝑣 𝜎
Í Í
for the element of WG whose 𝜎-coordinate is equal to 𝑣 𝜎 , and all other
440 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

coordinates are 0. We obtain

Õ Õ Õ
0 0 0 0
𝜆 (𝑣) = 𝜆 (𝛿 𝜎 𝑣) = 𝜃 (𝛿 𝜎 )𝜆 (𝑣) = 𝛿 𝜎−1 𝜆0(𝑣).
𝜎∈G 𝜎∈G 𝜎∈G

Moreover, using that 𝛿 2𝜎 = 1, we have

𝛿 𝜎−1 𝜆0(𝑣) = 𝜆0(𝛿 𝜎 𝑣) = 𝜆0(𝛿 2𝜎 𝑣) = 𝛿 𝜎−1 𝜆(𝛿 𝜎 𝑣).

In other words, the 𝜎 −1 -component of 𝜆0(𝑣), being equal to 𝜆(𝛿 𝜎 𝑣), only
depends of 𝑣 𝜎 . This proves the existence of maps 𝜆 𝜎 : W → W such
that 𝜆0((𝑣 𝜎 )) = (𝜆 𝜎 (𝑣 𝜎−1 )); these maps 𝜆 𝜎 are necessarily K-linear. Let
𝑣 = (𝑣 𝜎 ) ∈ WG and 𝑎 = (𝑎 𝜎 ) ∈ LG ; one has

𝜆0(𝑎𝑣) = 𝜆0((𝑎 𝜎 𝑣 𝜎 )) = (𝜆 𝜎 (𝑎 𝜎−1 𝑣 𝜎−1 ))

and
𝜃0(𝑎)𝜆0(𝑣) = (𝜎(𝑎 𝜎−1 ) (𝜆 𝜎 (𝑣 𝜎−1 ) = (𝜎(𝑎 𝜎−1 )𝜆 𝜎 (𝑣 𝜎−1 )),
so that the relation 𝜆0(𝑎𝑣) = 𝜃0(𝑎)𝜆0(𝑣) is equivalent to the relations
𝜆 𝜎 (𝑎𝑣) = 𝜎(𝑎)𝜆 𝜎 (𝑣) for all 𝑎 ∈ L, 𝑣 ∈ W and 𝜎 ∈ G.
The map 𝜆 is bijective if and only if 𝜆0 is bijective, which is equivalent
to the maps 𝜆 𝜎 to be bijective. We thus see that this first condition for a
descent datum is equivalent to properties (i) and (ii).
Property (iii) will be a reformulation of the equality “𝜃1 ◦ 𝜏1 ◦ 𝜃1 =
𝜏1 ◦ 𝜃1 ◦ 𝜏1 ” of maps from L ⊗K L ⊗K W to itself.
With the present notation, “𝜏1 ” is the map 𝜃˜ = 𝜃 ⊗ idW and “𝜃1 ” is the
map 𝜆˜ = idL ⊗𝜆. The map

𝜑˜ W = 𝜑WG ◦ (idL ⊗𝜑W ) : L ⊗K ⊗K W → WG×G

satisfies

𝜑˜ W (𝑎 ⊗ 𝑏 ⊗ 𝑣) = 𝜑WG (𝑎 ⊗ (𝜎(𝑏)𝑣)) = (𝜏(𝑎)𝜎(𝑏)𝑣)𝜎,𝜏∈G ,

for every 𝑎, 𝑏 ∈ L and every 𝑣 ∈ W; it is a K-linear isomorphism.

Let 𝜃˜ 0 = 𝜑˜ W ◦ 𝜃˜ ◦ 𝜑˜ W
−1
and 𝜆˜ 0 = 𝜑˜ W ◦ 𝜆˜ ◦ 𝜑˜ W
−1
. The second condition
for 𝜆 to be a descent datum writes 𝜆˜ 0 ◦ 𝜃˜ 0 ◦ 𝜆˜ 0 = 𝜃˜ 0 ◦ 𝜆˜ 0 ◦ 𝜃˜ 0.
Let us first compute 𝜃˜ 0. For 𝑎, 𝑏 ∈ L and 𝑣 ∈ W, one has 𝜃(𝑎 ˜ ⊗ 𝑏 ⊗ 𝑣) =
𝑏 ⊗ 𝑎 ⊗ 𝑣, so that 𝜃˜ 0((𝜏(𝑎)𝜎(𝑏)𝑣)) = (𝜎(𝑎)𝜏(𝑏)𝑣). Consequently, 𝜃˜ 0((𝑣 𝜎,𝜏 )) =
(𝑣 𝜏,𝜎 ) for every element (𝑣 𝜎,𝜏 ) of WG×G of the form 𝜑˜ W (𝑎 ⊗ 𝑏 ⊗ 𝑣), with
 8.9. GALOIS DESCENT 441

𝑎, 𝑏 ∈ L and 𝑣 ∈ W. Since they generate WG×G as a K-vector space, this

proves that
𝜃˜ 0((𝑣 𝜎,𝜏 )) = (𝑣 𝜏,𝜎 )𝜎,𝜏∈G
for every (𝑣 𝜎,𝜏 ) ∈ WG×G .
Let us now compute 𝜆˜ 0. To that aim, we will decompose an element
(𝑣 𝜎,𝜏 ) as a sum
Õ
𝛿 (𝜎,𝜏) 𝛿(𝑣 𝜎,𝜏 ),
𝜎,𝜏∈G

where 𝛿 (𝜎,𝜏) is the element of LG×G all of whose coordinates are zero,
but for the coordinate (𝜎, 𝜏) which is equal to 1, and 𝑣 𝜎,𝜏 is an abuse of
language for the element of WG×G all of whose coordinates are equal
to 𝑣 𝜎,𝜏 .
Fix 𝜎0 , 𝜏0 ∈ G, as well as decompositions 𝜀𝜎0 = 𝑎 𝑖 ⊗ 𝑏 𝑖 and 𝜀𝜏0 =
Í
𝑐 𝑗 ⊗ 𝑑 𝑗 in L ⊗K L — in other words, 𝜎(𝑎 𝑖 )𝑏 𝑖 = 1 if 𝜎 = 𝜎0 , and 0
Í Í
otherwise, and 𝜏(𝑐 𝑗 )𝑑 𝑗 = 1 if 𝜏 = 𝜏0 and 0 otherwise. Let 𝑣 ∈ W and let
Í
Õ Õ
𝜉= (𝑐 𝑗 ⊗ 1 ⊗ 𝑑 𝑗 )(1 ⊗ 𝑎 𝑖 ⊗ 𝑏 𝑖 )(1 ⊗ 1 ⊗ 𝑣) = 𝑐 𝑗 ⊗ 𝑎 𝑖 ⊗ 𝑏 𝑖 𝑑 𝑗 𝑣.
𝑖,𝑗 𝑖,𝑗

For 𝜎, 𝜏 ∈ G, the (𝜎, 𝜏)-component of 𝜑˜ W (𝜉) is equal to

Õ Õ Õ
𝜏(𝑐 𝑗 )𝜎(𝑎 𝑖 )(𝑏 𝑖 𝑑 𝑗 𝑣) = 𝜏(𝑐 𝑗 )𝜎(𝑑 𝑗 ) 𝜎(𝑎 𝑖 )𝑏 𝑖 𝑣
𝑖,𝑗 𝑗 𝑖
Õ
= 𝜏(𝑐 𝑗 )𝜎(𝑑 𝑗 )𝛿 𝜎0 𝑣
𝑗
(
𝑣 if 𝜏 = 𝜏0 and 𝜎 = 𝜎0
=
0 otherwise.

Moreover,
Õ
𝜉= (𝑐 𝑗 ⊗ 𝑎 𝑖 ⊗ 𝑏 𝑖 𝑑 𝑗 )(1 ⊗ 1 ⊗ 𝑣),
𝑖,𝑗

so that
Õ
idL ⊗𝜆(𝜉) = (𝑐 𝑗 ⊗ 𝑏 𝑖 𝑑 𝑗 ⊗ 𝑎 𝑖 )(1 ⊗ 𝜆(1 ⊗ 𝑣)),
𝑖,𝑗
442 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

since 𝜆 is 𝜃-linear. Let us then choose a decomposition 𝜆(1 ⊗ 𝑣) =

𝑢 𝑘 ⊗ 𝑣 𝑘 , for some elements 𝑢 𝑘 ∈ L and 𝑣 𝑘 ∈ W. One thus has
Í
Õ Õ
idL ⊗𝜆(𝜉) = (𝑐 𝑗 ⊗ 𝑏 𝑖 𝑑 𝑗 ⊗ 𝑎 𝑖 )(1 ⊗ 𝑢 𝑘 ⊗ 𝑣 𝑘 ) = 𝑐 𝑗 ⊗ 𝑏 𝑖 𝑑 𝑗 𝑢𝑘 ⊗ 𝑎 𝑖 𝑣 𝑘 .
𝑖,𝑗,𝑘 𝑖,𝑗,𝑘

It thus follows that for every 𝜎, 𝜏 ∈ G, one has

𝜆0(𝛿(𝜎0 ,𝜏0 ) 𝑣) = 𝜑˜ W (id ⊗𝜆(𝜉))𝜎,𝜏
Õ
= 𝜏(𝑐 𝑗 )𝜎(𝑏 𝑖 𝑑 𝑗 𝑢 𝑘 )𝑎 𝑖 𝑣 𝑘
𝑖,𝑗,𝑘
Õ Õ Õ
= 𝜎(𝑏 𝑖 )𝑎 𝑖 𝜏(𝑐 𝑗 )𝜎(𝑑 𝑗 ) 𝜎(𝑢 𝑘 )𝑣 𝑘 .
𝑖 𝑗 𝑘

One has
Õ Õ
𝜎(𝑏 𝑖 )𝑎 𝑖 = 𝜎( 𝜎 −1 (𝑎 𝑖 )𝑏 𝑖 ) = 𝜎(𝛿 𝜎,𝜎−1 ) = 𝛿 𝜎,𝜎−1
0 0
𝑖 𝑖

and
Õ Õ
𝜏(𝑐 𝑗 )𝜎(𝑑 𝑗 ) = 𝜎( 𝜎 −1 𝜏(𝑐 𝑗 )𝑑 𝑗 ) = 𝜎(𝛿 𝜎−1 𝜏,𝜏0 ) = 𝛿 𝜎−1 𝜏,𝜏0 .
𝑗 𝑗

Moreover,
Õ
( 𝜎(𝑢 𝑘 )𝑣 𝑘 )𝜎 = 𝜑W (𝜆(1 ⊗ 𝑣)) == 𝜆0(𝜑W (1 ⊗ 𝑣)) = 𝜆0((𝑣)𝜎 ) = (𝜆 𝜎 (𝑣)).
Consequently,
𝑣 𝜎,𝜏 = 𝛿 𝜎,𝜎−1 𝛿 𝜎−1 𝜏,𝜏0 𝜆 𝜎 (𝑣).
0

This is 0 unless 𝜎 = 𝜎0−1 and 𝜎 −1 𝜏 = 𝜏0 , that is 𝜏 = 𝜎0−1 𝜏0 ; in this case, one

has
𝑣 𝜎−1 ,𝜎−1 𝜏0 = 𝜆 𝜎−1 (𝑣).
0 0 0

We thus obtain that

𝜆0(𝛿 (𝜎0 ,𝜏0 ) 𝑣) = 𝛿(𝜎−1 ,𝜎−1 𝜏0 )𝜆 𝜎−1 (𝑣).
0 0 0

The relation (𝜎, 𝜏) = (𝜎0−1 , 𝜎0−1 𝜏0 ) is equivalent to (𝜎0 , 𝜏0 ) = (𝜎 −1 , 𝜎0 𝜏) =

(𝜎 −1 , 𝜎 −1 𝜏), so that, for every element 𝑣 = (𝑣 𝜎,𝜏 ) ∈ WG×G , we have
𝜆0(𝑣) = (𝜆 𝜎 (𝑣 𝜎−1 ,𝜎−1 𝜏 )).
 8.9. GALOIS DESCENT 443

At this point, we can finally express the relation 𝜆0 ◦ 𝜃0 ◦ 𝜆0 = 𝜃0 ◦ 𝜆0 ◦ 𝜃0.

For 𝑣 = (𝑣 𝜎,𝜏 ) ∈ WG×G , one has
(𝜆0 ◦ 𝜃0 ◦ 𝜆0(𝑣))𝜎,𝜏 = 𝜆 𝜎 ((𝜃0 ◦ 𝜆0(𝑣))𝜎−1 ,𝜎−1 𝜏 )
= 𝜆 𝜎 ((𝜆0(𝑣))𝜎−1 𝜏,𝜎−1 )
= 𝜆 𝜎 (𝜆 𝜎−1 𝜏 (𝑣 𝜏−1 𝜎,𝜏−1 )).
On the other hand, one has
(𝜃0 ◦ 𝜆0 ◦ 𝜃0(𝑣))𝜎,𝜏 = (𝜆0 ◦ 𝜃0(𝑣))𝜏,𝜎
= 𝜆𝜏 ((𝜃0(𝑣))𝜏−1 ,𝜏−1 𝜎 ) = 𝜆𝜏 (𝑣 𝜏−1 𝜎,𝜏−1 ).
Consequently, the relation 𝜆0 ◦ 𝜃0 ◦ 𝜆0 = 𝜃0 ◦ 𝜆0 ◦ 𝜃0 is equivalent to the
equalities 𝜆 𝜎 ◦ 𝜆 𝜎−1 𝜏 = 𝜆𝜏 for all 𝜎, 𝜏 ∈ G. This concludes the proof of
the proposition.


Theorem (8.9.6) (Galois descent). — Let W be an L-vector space and let (𝜆 𝜎 )

be a family of K-linear maps, 𝜆 𝜎 : W → W, satisfying the properties:
(i) One has 𝜆 𝜎 (𝑎𝑣) = 𝜎(𝑎)𝜆 𝜎 (𝑣) for all 𝑎 ∈ L and all 𝑣 ∈ W;
(ii) One has 𝜆 𝜎𝜏 = 𝜆 𝜎 ◦ 𝜆𝜏 for all 𝜎, 𝜏 ∈ G.
Let V be the set of all elements 𝑣 ∈ W such that 𝜆 𝜎 (𝑣) = 𝑣 for all 𝜎 ∈ G. Then
the following properties hold:
a) V is a K-subvector space of W;
b) There exists a unique morphism 𝜓 : L ⊗K V → W such that 𝜓(𝑎 ⊗ 𝑣) = 𝑎𝑣
for all 𝑎 ∈ L and all 𝑣 ∈ V;
c) The morphism 𝜓 is an isomorphism.

Proof. — This is just a reformulation of theorem 8.8.12 in the case of

the Galois extension K → L. Let 𝜑W : L ⊗K W → WG be the K-linear
isomorphism such that 𝜑W (𝑎 ⊗ 𝑣) = (𝜎(𝑎)𝑣)𝜎 for all 𝑎 ∈ L and all 𝑣 ∈ W.
The given conditions on the family (𝜆 𝜎 ) furnish a descent datum 𝜆 on W,
given by 𝜑W ◦ 𝜆 ◦ 𝜑W −1
((𝑣 𝜎 )) = (𝜆 𝜎 (𝑣 𝜎−1 )) for all (𝑣 𝜎 ) ∈ WG . The condition
𝜆(1 ⊗ 𝑣) = 1 ⊗ 𝑣 defining the submodule denoted by M in theorem 8.8.12
rewrites as 𝜆 𝜎 (𝑣) = 𝑣 for all 𝜎 ∈ G, hence is equal to V. This concludes
the proof. 
444 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

Exercises
1-exo074) Let 𝑚 and 𝑛 be two coprime integers. Show that
(Z/𝑚Z) ⊗Z (Z/𝑛Z) = 0.

2-exo076) Let I and J be ideals of a commutative ring A. Construct an isomorphism of

A-algebras
(A/I) ⊗A (A/J) ' A/(I + J).

3-exo094) Let X be a topological space. Show that

𝒞(X, R) ⊗R C ' 𝒞(X, C).

4-exo077) Let M be a Z-module.

a) Show that M ⊗Z Q is torsion-free.
b) Let S = Z {0}. Construct an isomorphism between M ⊗Z Q and S−1 M.
c) Show that Mtor is the kernel of the canonical morphism from M to M ⊗Z Q.
5-exo585) Let A, B, C, D be rings. Let M be an (A, B)-bimodule, N be a (B, C)-bimodule
and P be a (C, D)-bimodule. Show that there is a unique morphism of abelian groups
M ⊗B (N ⊗C P) → (M ⊗B N) ⊗C P
which maps 𝑚 ⊗ (𝑛 ⊗ 𝑝) to 𝑚 ⊗ (𝑛 ⊗ 𝑝). Show that it is an isomorphism of (A, D)-
bimodules.
6-exo814) Let A be a ring, let M be a right A-module and let N be a left A-module. Let
(𝑥 𝑖 )𝑖∈I be a generating family in M, let (𝑦 𝑗 ) 𝑗∈J be a generating family in N and let (𝑚 𝑗 ) 𝑗∈J
be an almost-null family of elements of M; let 𝜇 = 𝑚 𝑗 ⊗ 𝑦 𝑗 ∈ M ⊗A N.
Í

a) Assume that there exists an almost-null family (𝑎 𝑖,𝑗 ) in A, indexed by I × J, such

that 𝑚 𝑗 = 𝑖∈I 𝑥 𝑖 𝑎 𝑖,𝑗 for all 𝑗 ∈ J, and 0 = 𝑗∈J 𝑎 𝑖,𝑗 𝑦 𝑗 for all 𝑖 ∈ I. Prove that 𝜇 = 0.
Í Í

b) Conversely, one assumes that 𝜇 = 0. Introducing the kernel K of the unique

morphism 𝑢 : A(J) → N that maps the basis element of index 𝑗 to 𝑦 𝑗 , for every 𝑗 ∈ J,
prove that there exists an almost-null family (𝑎 𝑖,𝑗 ) as above.
7-exo813) Let M be a right A-module and let N be a submodule of M; let us denote
by 𝑗 : N → M the inclusion morphism. One says that 𝑗 is pure, or that N is a pure
submodule of M if, for every left A-module X, the morphism 𝑗 ⊗ idX : N ⊗A X → M ⊗A X
is injective.
a) Assume that N is pure. Let 𝑚, 𝑛 be positive integers, let U = (𝑎 𝑖,𝑗 ) ∈ M𝑚,𝑛 (A)
and let (𝑦1 , . . . , 𝑦𝑛 ) ∈ N𝑛 . Assume that there exists (𝑥1 , . . . , 𝑥 𝑚 ) ∈ M𝑚 such that
𝑦𝑗 = 𝑚 𝑖=1 𝑥 𝑗 𝑎 𝑗,𝑖 for every 𝑗. Prove that there exists such a family in N .
𝑚
Í

b) Conversely, prove that this property implies that for every finitely presented left
A-module X, the map 𝑗 ⊗ idX is injective.
c) Under this assumption, prove that N is a pure submodule of M.
 EXERCISES 445

8-exo826) Let 𝑘 be a field, let A be the polynomial ring 𝑘[X, Y] and let M = (X, Y) be the
ideal of A consisting of polynomials without constant term. Let 𝑔 : A → A ⊕ A be the
morphism given by 𝑔(P) = (YP, −XP) and 𝑓 : A ⊕ A → A be the morphism defined by
𝑓 (P, Q) = XP + YQ.
a) Show that one has an exact sequence
𝑔 𝑓
0→A→
− A⊕A→
− M → 0.
b) By restriction and tensor product, deduce the following exact sequences:
𝑔 𝑓
0→A→
− M⊕M→
− M2 → 0,
and
𝑔⊗idM 𝑓 ⊗idM
0 → M −−−−−→ M ⊕ M −−−−−→ M ⊗A M → 0.
(One write M2 for the square of the ideal M.)
c) Define a commutative diagram of A-modules with exact rows:

→ A → M⊕M → M2 → 0
← ← ← ←
0
←

⇐
←

⇐
→

→ 𝑘 → M ⊗A M → M2 → 0.
← ← ← ←
0
d) Let 𝑡 = X ⊗ Y − Y ⊗ Y ∈ M ⊗A M. Prove that 𝑡 ≠ 0 and that it generates the torsion
submodule of M ⊗A M.
e) Prove that the submodule of M ⊗A M generated by X ⊗ X, X ⊗ Y and Y ⊗ Y is
isomorphic to M2 .
f ) Construct an isomorphism M ⊗A M ' M2 ⊕ 𝑘.
9-exo081) Let 𝑘 be a field, let P ∈ 𝑘[X] be an irreducible polynomial of degree > 1. Let
K = 𝑘[X]/(P).
a) Show that K is a field.
b) Show that there exists a morphism 𝜑 of 𝑘-algebras from K ⊗ 𝑘 K to K which maps
𝑎 ⊗ 𝑏 to 𝑎𝑏 for every 𝑎, 𝑏 ∈ K.
c) Show that 𝜑 is not injective and conclude that K ⊗ 𝑘 K is not a field.
10-exo147) Let A be a local ring, let J be its maximal ideal and K = A/J its residue field.
Let M and N be two finitely generated A-modules, one also assumes that N is free.
Let 𝑓 : M → N be a morphism of A-modules such that the morphism
𝑓 ⊗ idK : M ⊗A K → N ⊗A K
is an isomorphism.
a) Using Nakayama’s lemma, show that 𝑓 is surjective.
b) Show that 𝑓 has right inverse 𝑔, that is, show that there exists a morphism
𝑔 : N → M such that 𝑓 ◦ 𝑔 = idN .
c) Show that 𝑔 is surjective and conclude that 𝑓 is an isomorphism.
11-exo583) Let A be a commutative ring, let M and N be A-modules.
446 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

a) One assumes that M and N are finitely generated; show that M ⊗A N is finitely
generated.
b) One assumes that M and N are simple. Show that M ⊗A N is isomorphic to M if
M and N are isomorphic, and is 0 otherwise.
c) One assumes that M and N are A-modules of finite length. Show that M ⊗A N has
finite length, and that
ℓA (M ⊗A N) 6 ℓA (M) ℓA (N).
12-exo087) Let A be a local commutative ring, let J be its maximal ideal and let K = A/J
be its residue field, let M and N be two A-modules.
a) Recall the structures of K-modules on M/JM and N/JN. Construct a surjective
homomorphism from M ⊗A N to (M/JM) ⊗K (N/JN).
b) One assumes that M ⊗A N = 0. Show that M/JM = 0 or N/JN = 0.
c) If, moreover, M and N are finitely generated, show using Nakayama’s lemma that
M = 0 or N = 0.
d) Give an example of a commutative local ring A and of non-zero A-modules M
and N such that M ⊗A N = 0.
13-exo864) Let A be a commutative ring, let M and N be finitely generated A-modules.
Using exercise 8/12, prove that Supp(M ⊗A N) = Supp(M) ∩ Supp(N).
14-exo171) Let A be a local commutative ring, let J be its maximal ideal. One assumes
that A is an integral domain, let K be its field of fractions.
Let M be a finitely generated A-module such that
dimA/J M/JM = dimK M ⊗A K.
Prove that M is a free A-module.
15-exo582) Let 𝑘 be a field, let K and L be 𝑘-algebras which are (commutative) fields.
a) Observe that K ⊗ 𝑘 L is a commutative 𝑘-algebra.
b) Show that there are canonical morphisms from K to K ⊗ 𝑘 L and from L to K ⊗ 𝑘 L
such that 𝑎 ↦→ 𝑎 ⊗ 1 and 𝑏 ↦→ 1 ⊗ 𝑏 respectively. Observe that they are injective.
c) Show that there exists a field E together with field morphisms 𝑖 : K → E and
𝑗 : L → E. (In other words, there exists a field E which contains simultaneously
isomorphic copies of K and L.)
16-exo817) Let 𝑓 : A → B be a morphism of commutative rings such that for every
A-module M, the map 𝑓M : M → M ⊗A B (given by 𝑚 ↦→ 𝑚 ⊗ 1) is injective. (One says
that 𝑓 is pure.)
a) Prove that 𝑓 is injective.
b) Prove that for every ideal I of A, one has I = 𝑓 −1 ( 𝑓 (I) · B).
c) Assume that A is a domain. Prove that B is a domain.
d) Assume that A is a domain and is integrally closed; prove that the same holds
for B.
 EXERCISES 447

17-exo818) Let 𝑓 : A → B be a morphism of commutative rings such that I = 𝑓 −1 ( 𝑓 (I)·B)

for every ideal I of A. For every A-module M, one denotes by 𝑓M the canonical morphism
M → M ⊗A B.
a) Let I be an ideal of A. Prove that 𝑓A/I is injective.
b) Let M be a A-module and let N be a submodule of M. Assume that 𝑓N and 𝑓M/N
are injective; prove that 𝑓M is injective.
c) Let M be a finitely generated A-module. Prove that 𝑓M is injective.
d) Prove that 𝑓 is pure, that is, 𝑓M is injective for every A-module M.
18-exo587) Let M be a free A-module of rank 𝑛. Show that the symmetric algebra of M
is isomorphic to the ring A[T1 , . . . , T𝑛 ] of polynomials in 𝑛 indeterminates.
19-exo588) Let M be an A-module. Let 𝑛 be a nonnegative integer.
a) Define an action of the symmetric group 𝔖𝑛 on T𝑛 (M) in such a way that
𝜎(𝑚1 ⊗ · · · ⊗ 𝑚𝑛 ) = 𝑚 𝜎−1 (1) ⊗ · · · ⊗ 𝑚 𝜎−1 (𝑛) . for every 𝜎 ∈ 𝔖𝑛 and any 𝑚1 , . . . , 𝑚𝑛 ∈ M.
b) One says that a tensor 𝑥 in T𝑛 (M) is symmetric 𝜎(𝑥) = 𝑥 for every 𝜎 ∈ 𝔖𝑛 . Let
T𝑛 (M)sym be the set of all symmetric tensors in T𝑛 (M). Show that it is a submodule
of T𝑛 (M).
c) Let then T(M)sym be the direct sum of the submodules T𝑛 (M)sym in T(M). Show
that it is a commutative sub-algebra of T(M).
d) Assume that 𝑛! be invertible in A. Show that one defines en endomorphism 𝑠
of T𝑛 (M) by defininig 𝑠(𝑥) = 𝑛!1 Í 𝑛
𝜎∈𝔖𝑛 𝜎(𝑥), for 𝑥 ∈ T (M) (symmetrization of the
tensor 𝑥). Show that 𝑠(𝑥) ∈ T𝑛 (M)sym for every 𝑥 ∈ T𝑛 (M). Deduce that the canonical
map from T𝑛 (M) to S𝑛 (M) induces an isomorphism from T𝑛 (M)sym to S𝑛 (M).
e) In particular, if A contains Q, the A-module T(M)sym is canonically isomorphic
to S(M). This gives another structure of an A-algebra on T(M)sym , which is the one
deduced from that of S(M) by this canonical isomorphism. Explicitly, if 𝑥 and 𝑦 are
symmetric tensors, show that their new product is the symmetrization of the tensor
𝑥 ⊗ 𝑦.
f ) Assume that A = Z and that M = Z2 . Let (𝑒, 𝑓 ) be the canonical basis of M. Show
that (𝑒 ⊗ 𝑒 , 𝑒 ⊗ 𝑓 , 𝑓 ⊗ 𝑒, 𝑓 ⊗ 𝑓 ) is a basis of T2 (M). Show that (𝑒 ⊗ 𝑒, 𝑒 ⊗ 𝑓 + 𝑓 ⊗ 𝑒, 𝑓 ⊗ 𝑓 )
is a basis of T2 (M)sym . Show that the canonical morphism from T2 (M)sym to S2 (M)
is not surjective. More precisely, show that the quotient of S2 (M) by the submodule
generated by T2 (M) is isomorphic to Z/2Z.
20-exo589) Let M and P be A-modules, let 𝑓 : M𝑛 → P be a symmetric 𝑛-linear map
from M𝑛 to P. Show that there exists a unique morphism of A-modules, 𝜑 : S𝑛 (M) → P,
such that 𝜑(𝑚1 . . . 𝑚𝑛 ) = 𝑓 (𝑚1 , . . . , 𝑚𝑛 ) for every (𝑚1 , . . . , 𝑚𝑛 ) ∈ M𝑛 .
21-exo593) Let M be a free A-module of rank 𝑛, let 𝑢 be an endomorphism of M. Let
M[X] be the A[X]-module M ⊗A A[X].
É
a) Identify M[X] with the abelian group 𝑛∈N M endowed with the external
𝑛
multiplication ( 𝑎 𝑛 X ) · (𝑚𝑛 ) = ( 𝑘 𝑎 𝑘 𝑚𝑛−𝑘 )𝑛∈N .
Í Í

b) Show that (𝑚𝑛 ) ↦→ (𝑢(𝑚𝑛 )) is an endomorphism of M[X], still denoted 𝑢.

448 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

c) Show that the determinant of X idM[X] −𝑢 is the characteristic polynomial P𝑢 of 𝑢.

d) Identify the quotient of M[X] by the image of the endomorphism X − 𝑢 : 𝑚 ↦→
X𝑚 − 𝑢(𝑚) to the A[X]-module M𝑢 . (Recall that M𝑢 is the abelian group M with
external multiplication given by the morphism P ↦→ P(𝑢), from A[X] to End(M)).
e) Using exercise 3/31, show that P𝑢 (𝑢) = 0 (theorem of Hamilton-Cayley).
22-exo592) Let M be a finitely generated A-module, let 𝑢 be an endomorphism of M.
Let 𝑓 : A𝑛 → M be any surjective morphism of A-modules.
a) Show that there exists an endomorphism 𝑣 of A𝑛 such that 𝑓 (𝑣(𝑥)) = 𝑢( 𝑓 (𝑥)) for
every 𝑥 ∈ A𝑛 .
b) Show that there exist elements 𝑎 1 , . . . , 𝑎 𝑛 ∈ A such that
𝑢 𝑛 (𝑥) + 𝑎1 𝑢 𝑛−1 (𝑥) + · · · + 𝑎 𝑛−1 𝑢(𝑥) + 𝑎 𝑛 𝑥 = 0
for every 𝑥 ∈ M.
23-exo595) Let K be a field, let V be a K-vector space. Let V0 = Span(𝑥1 , . . . , 𝑥 𝑛 ) and
V00 = Span(𝑦1 , . . . , 𝑦𝑛 ) be subspaces of V with the same dimension 𝑛.
a) If V0 = V00, show that the vectors 𝑥1 ∧ · · · ∧ 𝑥 𝑛 and 𝑦1 ∧ · · · ∧ 𝑦𝑛 in Λ𝑛 (V) are
collinear.
b) Let 𝜑1 , . . . , 𝜑𝑛 be linear forms on V. Show that there exists a unique linear form Φ
on Λ𝑛 (V) such that Φ(𝑒1 ∧ · · · ∧ 𝑒 𝑛 ) = det(𝜑 𝑖 (𝑒 𝑗 )) for every 𝑒1 , . . . , 𝑒 𝑛 ∈ V.
c) Show that two vectors of a vector space are collinear if and only if any linear form
which vanishes on one vanishes on the other.
d) If 𝑥1 ∧ · · · ∧ 𝑥 𝑛 and 𝑦1 ∧ · · · ∧ 𝑦𝑛 are collinear, show that V0 = V00.
24-exo594) Let M be an A-module, let M0 and M00 be submodules of M such that
M = M0 ⊕ M00.
a) Show that there exists a unique morphism of A-modules 𝜃 : Λ(M0) ⊗ Λ(M00) →
Λ(M) which maps (𝑒1 ∧ · · · ∧ 𝑒 𝑝 ) ⊗ ( 𝑓1 ∧ · · · ∧ 𝑓𝑞 ) to 𝑒1 ∧ · · · ∧ 𝑒 𝑝 ∧ 𝑓1 ∧ · · · ∧ 𝑓𝑞 for every
𝑒1 , . . . , 𝑒 𝑝 ∈ M0 and 𝑓1 , . . . , 𝑓𝑞 ∈ M.
b) If M0 and M00 are free, show that 𝜃 is an isomorphism.
c) Show that 𝜃 is not a morphism of algebras (unless M0 = 0 or M00 = 0).
d) Show that there exists a unique structure of an A-algebra on Λ(M0) ⊗ Λ(M00) such
that (𝜉0 ⊗ 𝜉00)(𝜂0 ⊗ 𝜂00) = (−1)𝑞 𝑝 (𝜉0 ∧ 𝜂0) ⊗ (𝜉00 ∧ 𝜂00), if 𝜉0 ∈ Λ𝑝 (M0), 𝜉00 ∈ Λ𝑝 (M00),
0 00 0 00

𝜂0 ∈ Λ𝑞 (M0), 𝜂00 ∈ Λ𝑞 (M00).

0 00

e) Show that 𝜃 is an isomorphism for this structure.

25-exo804) Let A and B be noetherian local rings, let MA and MB denote their maximal
ideals and let 𝑓 : A → B be a morphism of rings such that MA = 𝑓 −1 (MB ).
a) Prove that dim(B) 6 dim(A) + dim(B/MA B). (Use theorem 9.3.3.)
b) Assume, moreover, that 𝑓 is flat. Let 𝑛 = dim(A) and let (P0 , . . . , P𝑛 = MA ) be a
chain of prime ideals of A. Prove that there exists a chain (Q0 , . . . , Q𝑛 ) of prime ideals
of B such that P 𝑗 = 𝑓 −1 (Q 𝑗 ) for all 𝑗. (This is a Going-Down theorem for flat morphisms.)
c) Still assuming that 𝑓 is flat, prove that dim(B) = dim(A) + dim(B/MA B).
 EXERCISES 449

26-exo815) Let A be a principal ideal domain and let 𝑎 ∈ A. Let M be an A/(𝑎)-module.

a) Let 𝑏, 𝑐 ∈ A be such that 𝑎 = 𝑏𝑐, let I = 𝑏A/𝑎A. Show that the morphism 𝑥 ↦→ 𝑏𝑥
induces an isomorphism from A/𝑐A to I, and an isomorphism from M/𝑐M to I ⊗A/(𝑎) M.
b) Prove that M is flat if and only if the following condition holds: for every pair
(𝑏, 𝑐) in A such that 𝑎 = 𝑏𝑐, and for every 𝑥 ∈ M such that 𝑏𝑥 = 0, there exists 𝑦 ∈ M
such that 𝑥 = 𝑐𝑦.
27-exo816) Let 𝑓 : A → B be a flat morphism of commutative rings. Let 𝑎 ∈ A be a
regular element. Prove that 𝑓 (𝑎) is a regular element of B.
28-exo839) Let A be a commutative ring, let 𝑎 1 , . . . , 𝑎 𝑛 ∈ A and let B =
A[T1 , . . . , T𝑛 ]/(𝑎0 + 𝑎1 T1 + · · · + 𝑎 𝑛 T𝑛 ).
a) Assume that there exists 𝑖 ∈ {1, . . . , 𝑛} such that 𝑎 𝑖 is a unit. Prove that R is a free
A-module.
b) Assume that (𝑎 1 , . . . , 𝑎 𝑛 ) generate the unit ideal. Prove that B is a faithfully flat
A-module.
c) Assume that (𝑎0 , . . . , 𝑎 𝑛 ) generate the unit ideal. Prove that B is a flat A-module,
and that it is faithfully flat if and only if (𝑎 1 , . . . , 𝑎 𝑛 ) generate the unit ideal. (Prove
that a prime ideal P ∈ Spec(A) is of the form A ∩ Q, for Q ∈ Spec(B) if and only if
P ∉ V((𝑎1 , . . . , 𝑎 𝑛 )).)
(The more general exercise 6.4 of Eisenbud (1995) treats the case where 𝑎 0 + 𝑎 1 T1 +
· · · + 𝑎 𝑛 T𝑛 is replaced by any polynomial 𝑓 ∈ A[T1 , . . . , T𝑛 ], but its proof goes beyond
the techniques of this book.)
29-exo838) Let 𝑓 : A → B be a flat morphism of commutative local rings. Let JA be
the maximal ideal of A and let JB be the maximal ideal of B; assume that 𝑓 −1 (JB ) = JA .
Prove that for every A-module M, one has

ℓB (M ⊗A B) = ℓA (M) ℓB (B/JA B).

30-exo862) Let A be a commutative ring.

a) Let 𝑎 ∈ A be such that the A-module A/(𝑎) is flat. Prove that there exists 𝑏 ∈ A
such that 𝑎 = 𝑎 2 𝑏.
b) If every A-module is flat, then A is a von Neumann ring.
c) Convesely, if A is a von Neumann ring, then every A-module is flat.
31-exo861) Let A be a (possibly noncommutative) local ring and let J be its maximal
ideal. Let M be a finitely presented right A-module.
a) Construct a morphism 𝑓 : A𝑛 → M that induces an isomorphism (A/J)𝑛 → M/MJ.
b) Let P = Ker( 𝑓 ). Construct a morphism 𝑔 : A𝑚 → P that induces an isomorphism
(A/J)𝑚 → P/PJ.
c) Prove that the canonical morphism P ⊗A J → P is an isomorphism.
d) Conclude that P = 0 and that M is a free A-module.
450 CHAPTER 8. TENSOR PRODUCTS AND DETERMINANTS

e) Let A be any commutative ring (not necessary local), and let M be a a finitely
presented flat A-module. Prove that for every prime ideal P of A, the AP -module MP
is free. Conclude that M is projective (cf. corollary 8.6.13).
CHAPTER 9

ALGEBRAS OF FINITE TYPE OVER A FIELD

In this final chapter, where all rings are assumed to be commutative, I initiate
the study of general noetherian rings, with a special emphasis on finitely
generated algebras over a field. Once translated into geometric properties of
algebraic subsets, these results form the foundations of algebraic geometry, but
this interpretation will only be slightly evoked here.
Noether’s normalization theorem is a tool to reduce the study of a general
finitely generated algebra over a field to the study of a polynomial ring. The
proof I explain is valid over any field, and the classical proof over infinite fields
is given as an exercise. I then explore a few of its consequences: a general proof
of Hilbert’s Nullstellensatz and Zariski’s version (which, however, can be proved
more directly, see the exercises), the fact that every prime ideal of such an algebra
is an intersection of maximal ideals, a property which gives rise to the notion of
a Jacobson ring, etc.
The next three sections are devoted to the notions of dimension and codi-
mension of rings. Based on lengths of chains of prime ideals, they embody the
idea that a point is contained in a curve, which is contained in a surface, etc.
For integral finitely generated algebras over a field, I show that the dimension
equals the transcendence degree of the field of fractions. The theory of codimen-
sion requires Krull’s Hauptidealsatz that embodies another intuition, that 𝑝
equations, if well chosen, define a subset of codimension 𝑝. I then establish the
expected relation between dimension and codimension.
I give another application of Noether’s normalization theorem to prove that if
an integral domain A is a finitely generated algebra over a field, then the integral
closure of A is a finitely generated A-module.
A last section studies Dedekind rings, the algebraic analogues of curves.
Their importance had first been revealed in algebraic number theory, for rings of
452 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

integers in number fields furnish fundamental examples. Although Dedekind

rings are not unique factorization domain in general, their (nonzero) ideals
have similar factorization as product of maximal ideals. Their defect of unique
factorization is encoded in their class groups; in the case of rings of integers of
number fields, I prove that Minkowski’s theorem that this class group is finite.

9.1. Noether’s normalization theorem

Combined with the properties of integral morphisms, Noether’s nor-
malization theorem is a powerful tool for the study of finitely generated
algebras over a field. This section and the following illustrate this fact.

Theorem (9.1.1) (Normalization theorem). — Let K be a field, let A be

a finitely generated K-algebra. Then there exist an integer 𝑛 > 0 and
elements 𝑎1 , . . . , 𝑎 𝑛 ∈ A such that the unique morphism of K-algebras
from K[X1 , . . . , X𝑛 ] to A such that X𝑖 ↦→ 𝑎 𝑖 is injective and integral.

Proof. — Let (𝑥 1 , . . . , 𝑥 𝑚 ) be a family of elements of A such that A =

K[𝑥 1 , . . . , 𝑥 𝑚 ]. Let us prove the theorem by induction on 𝑚. If 𝑚 = 0,
then A = K and the result holds with 𝑛 = 0. We thus assume that 𝑚 > 1
and that the result for any K-algebra which is generated by at most 𝑚 − 1
elements.
Let 𝜑 : K[X1 , . . . , X𝑚 ] → A be the unique morphism of K-algebras
such that 𝜑(X𝑖 ) = 𝑥 𝑖 . If 𝜑 is injective, the result holds, taking 𝑛 = 𝑚 and
𝑎 𝑖 = 𝑥 𝑖 for every 𝑖.
We may thus assume that there is a non-zero polynomial
P ∈ K[X1 , . . . , X𝑚 ] such that P(𝑥1 , . . . , 𝑥 𝑚 ) = 0. Let (𝑐 n ) be the
coefficients of P, so that
𝑚
X𝑛𝑖 𝑖 .
Õ Ö
P= 𝑐n
n∈N𝑚 𝑖=1

Let 𝑟 be an integer strictly greater than the degree of P in each variable;

in other words, for every ∈N𝑚 , one has 𝑐n = 0 unless 𝑛 𝑖 < 𝑟 for all 𝑖;
𝑖−1
then set 𝑦 𝑖 = 𝑥 𝑖 − 𝑥 1𝑟 for 𝑖 ∈ {2, . . . , 𝑚}. Let B = K[𝑦2 , . . . , 𝑦𝑚 ] be the
subalgebra of A generated by 𝑦2 , . . . , 𝑦𝑚 ; we are going to show that A is
integral over B.
 9.1. NOETHER’S NORMALIZATION THEOREM 453

We define a polynomial Q ∈ B[T] by

𝑚−1
Q(T) = P(T, 𝑦2 + T𝑟 , . . . , 𝑦𝑚 + T𝑟 )
𝑚−1
Õ
= 𝑐n T𝑛1 (𝑦2 + T𝑟 )𝑛2 . . . (𝑦𝑚 + T𝑟 )𝑛 𝑚
n∈N𝑚
𝑛2 𝑛𝑚  
𝑛2 𝑛𝑚
 
𝑛 −𝑗 𝑛 −𝑗
Õ Õ Õ Í𝑚 𝑖−1
= ... ... 𝑐n 𝑦2 2 2 . . . 𝑦𝑚𝑚 𝑚 T𝑛1 + 𝑖=2 𝑗𝑖 𝑟
𝑗2 𝑗𝑚
n∈N𝑚 𝑗2 =0 𝑗𝑚 =0

and observe that

𝑚−1
Q(𝑥 1 ) = P(𝑥1 , 𝑦2 + 𝑥1𝑟 , . . . , 𝑦𝑚 + 𝑥 1𝑟 ) = P(𝑥 1 , 𝑥2 , . . . , 𝑥 𝑚 ) = 0.
Order N𝑚 with the “reverse lexicographic order”: (𝑛10 , . . . , 𝑛 𝑚 0 ) <

(𝑛1 , . . . , 𝑛 𝑚 ) if and only if 𝑛 𝑚 0 < 𝑛 , or 𝑛 0 = 𝑛 and 𝑛 0

𝑚 𝑚 𝑚 𝑚−1
< 𝑛 𝑚−1 , etc.
If one only considers sequences n such that sup(𝑛 𝑗 ) < 𝑟, in particular
for elements n ∈ N𝑚 such that 𝑐n ≠ 0, this ordering corresponds with
the one given by the expansion in base 𝑟 (written in the reverse order):
(𝑛10 , . . . , 𝑛 𝑚
0 ) < (𝑛 , . . . , 𝑛 ) if and only if
1 𝑚
0 𝑚−1
𝑛𝑚 𝑟 0
+ 𝑛 𝑚−1 𝑟 𝑚−2 + · · · + 𝑛20 𝑟 + 𝑛10 < 𝑛 𝑚 𝑟 𝑚−1 + 𝑛 𝑚−1 𝑟 𝑚−2 + · · · + 𝑛2 𝑟 + 𝑛1 .
Let n be the largest multi-index in N𝑚 , for that ordering, such that
𝑐n ≠ 0. For any n0 ∈ N𝑚 such that 𝑐n0 ≠ 0, one has 𝑛 0𝑖 < 𝑟 for every 𝑖, by
definition of 𝑟, so that for any 𝑗2 ∈ {0, . . . , 𝑛20 }, . . . , 𝑗𝑚 ∈ {0, . . . , 𝑛 𝑚
0 },

𝑛10 + 𝑗2 𝑟 + · · · + 𝑗𝑚 𝑟 𝑚−1 6 𝑛10 + 𝑛20 𝑟 + · · · + 𝑛 𝑚

0 𝑚−1
𝑟 6 𝑛1 + 𝑛2 𝑟 + · · · + 𝑛 𝑚 𝑟 𝑚−1 ,
and equalities are only possible if n0 = n and 𝑗2 = 𝑛20 , . . . , 𝑗𝑚 = 𝑛 𝑚
0 . This

implies that the degree of Q is equal to 𝑛1 + 𝑛2 𝑟 + · · · + 𝑛 𝑚 𝑟 𝑚−1 and that

only the term with 𝑗 𝑘 = 𝑛 𝑘 for 𝑘 ∈ {2, . . . , 𝑚} contributes the leading
coefficient, which thus equals 𝑐n , a non-zero element of K. In particular,
Q is a polynomial in B[T] whose leading coefficient is a unit, so that 𝑥 1
is integral over B.
Consequently, B[𝑥 1 ] is integral over B. For every 𝑖 ∈ {2, . . . , 𝑚}, one
𝑖−1
has 𝑥 𝑖 = 𝑦 𝑖 − 𝑥1𝑟 ∈ B[𝑥1 ]. Since A = K[𝑥1 , . . . , 𝑥 𝑚 ], we conclude that
A = B[𝑥 1 ] and A is integral over B.
By induction, there exist an integer 𝑛 6 𝑚 −1 and elements 𝑎1 , . . . , 𝑎 𝑛 ∈
B such that the unique morphism 𝑓 : K[T1 , . . . , T𝑛 ] → B of K-algebras
such that 𝑓 (T𝑖 ) = 𝑎 𝑖 for all 𝑖 is injective and such B is integral over
454 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

K[𝑎1 , . . . , 𝑎 𝑛 ]. Then A is integral over K[𝑎1 , . . . , 𝑎 𝑛 ] as well, and this

concludes the proof of the theorem. 
As a first application, let us prove the general case of Hilbert’s Nullstel-
lensatz (theorem 2.3.1) which we had only proved under the assumption
that the field was uncountable.
The most fundamental form of this theorem is due to Zariski and
claims:

Theorem (9.1.2) (Zariski). — Let K be a field and let A be a finitely generated

K-algebra. One assumes that A is field. Then A is a finite extension of K.

Proof. — By Noether’s normalization theorem, there exist an integer 𝑛 >

0 and an injective morphism of K-algebras 𝑓 : K[X1 , . . . , X𝑛 ] → A such
that A is integral over the image B of 𝑓 . Since A is a field, proposition 4.2.5
implies that B is a field as well. However, B, being the image of
the injective morphism 𝑓 , is isomorphic to the ring K[X1 , . . . , X𝑛 ] of
polynomials in 𝑛 variables with coefficients in K. For 𝑛 > 1, this ring
is not a field. Consequently 𝑛 = 0 and A is algebraic over K. Being
moreover finitely generated as an algebra, corollary 4.2.2 implies that it
is a finite extension of K. 
As a corollary, we can also give a complete proof of theorem 2.3.1. Let
us first recall its statement.

Corollary (9.1.3). — Let 𝑛 be a positive integer, let K be an algebraically closed

field and let M be a maximal ideal of the ring K[X1 , . . . , X𝑛 ]. There exists a
unique element (𝑎1 , . . . , 𝑎 𝑛 ) ∈ K𝑛 such that M = (X1 − 𝑎1 , . . . , X𝑛 − 𝑎 𝑛 ).

Proof. — Let L be the residue fiekd K[X1 , . . . , X𝑛 ]/M and let

𝜃 : K[X1 , . . . , X𝑛 ] → L be the canonical surjection. By construc-
tion, L is a finitely generated K-algebra, and it is a field. By Zariski’s
theorem (theorem 9.1.2), L is an algebraic extension of K. Since K
is algebraically closed, L = K. For 𝑖 ∈ {1, . . . , 𝑛}, let 𝑎 𝑖 = 𝜃(X𝑖 ); by
definition of the quotient ring, it is the only element of K such that
X𝑖 − 𝑎 𝑖 ∈ M. Then M contains the ideal (X1 − 𝑎1 , . . . , X𝑛 − 𝑎 𝑛 ). On the
other hand, the same proof as for theorem 2.3.1 implies that this ideal is
maximal, so that M = (X1 − 𝑎1 , . . . , X𝑛 − 𝑎 𝑛 ). 
 9.1. NOETHER’S NORMALIZATION THEOREM 455

At this point, we advise the reader to read again about the correspon-
dence between ideals and algebraic sets discussed in section 2.3.
We now pass to other important consequences of theorem 9.1.2.
Corollary (9.1.4). — Let K be a field, let A be a finitely generated K-algebra
and let M be a prime ideal of A. Then M is maximal if and only if A/M is
finitely dimensional over K; it is then a finite extension of K.
Proof. — Assume that M is maximal; then A/M is a finitely generated
K-algebra which is a field. By theorem 9.1.2, it is a finite extension of K
and, in particular, its dimension as a K-vector space is finite.
Conversely, assume that the quotient K-algebra L = A/M is finitely
dimensional. It is also an integral domain. For every non-zero 𝑎 ∈ L, the
map 𝑏 ↦→ 𝑎𝑏 from L to itself is K-linear, and injective, hence it is bijective;
in particular, every non-zero element of L is invertible and L is a field.
Consequently, M is a maximal ideal of A. 

Corollary (9.1.5). — Let K be a field and let 𝑓 : A → B be a morphism of

finitely generated K-algebras. For every maximal ideal M of B, 𝑓 −1 (M) is a
maximal ideal of A.
Proof. — We know that the operation 𝑓 −1 induces a map from Spec(B)
to Spec(A), so that 𝑓 −1 (M) is a prime ideal; we need to prove that it is
even maximal. Passing to the quotient, 𝑓 induces an injective morphism
𝜑 : A/ 𝑓 −1 (M) → B/M of finitely generated K-algebras. Moreover, B/M
is a field. By theorem 9.1.2, it is thus a finite extension of K, hence 𝑓 −1 (M)
is a prime ideal of A such that A/ 𝑓 −1 (M) is finite dimensional. By the
preceding corollary, 𝑓 −1 (M) is a maximal ideal. 
The following corollary strengthens proposition 2.2.11.
Corollary (9.1.6). — Let K be a field and let A be a finitely generated K-algebra.
a) The nilradical and the Jacobson
√ radical of A coincide.
b) For every ideal I of A, I is the intersection of all maximal ideals of A
which contain I.
c) In particular, every prime ideal P of A is the intersection of the maximal
ideals of P which contain P.
One sums up assertions b) and c) by saying that A is a Jacobson ring.
456 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

Proof. — a) We need to prove that an element 𝑎 ∈ A is nilpotent if and

only if it belongs to every maximal ideal of I. One direction is clear: if 𝑎
is nilpotent, it belongs to every prime ideal of I, hence to every maximal
ideal of I. Conversely, let us assume that 𝑎 is not nilpotent and let us
show that there exists a maximal ideal M of A such that 𝑎 ∉ M. Since
𝑎 is not nilpotent, the K-algebra A𝑎 = A[1/𝑎] is not null; it is also a
finitely generated K-algebra. By corollary 9.1.5 the inverse image in A of
a maximal ideal of A𝑎 is a maximal ideal of A which does not contain 𝑎.
This concludes the proof of assertion a).
b) Let B = A/I; it is a finitely generated K-algebra and its maximal
ideals are of the form M/I, where M is a maximal ideal of A containing I.
By part a), the nilradical of B is the intersection of the maximal ideals
of B.√Since the class in B√of an element 𝑎 ∈ A is nilpotent if and only if
𝑎 ∈ I, this implies that I is the intersection of all maximal ideals of A
which contain I.
c) follows from b) applied to P = I. 
Let us give another important application.
Theorem (9.1.7). — Let K be an algebraically closed field and let A, B be two
finitely generated K-algebras.
a) If A and B are integral domains, then A ⊗K B is also an integral domain.
b) If A and B are reduced, then A ⊗K B is also reduced.
Proof. — We first prove the theorem under the assumption that A and
B are finitely generated.
a) Assume that A and B are integral domains.
The tensor product of two non-zero K-vector spaces is a non-zero
K-vector space; consequently, A ⊗K B ≠ 0.
Let then 𝑓 and 𝑔 be two elements of A ⊗K B such that 𝑓 𝑔 = 0. We may
decompose 𝑓 as a sum 𝑟𝑖=1 𝑎 𝑖 ⊗ 𝑏 𝑖 of split tensors, where 𝑏 1 , . . . , 𝑏 𝑟 are
Í
linearly independent over K. Similarly, we write 𝑔 = 𝑠𝑗=1 𝑎 0𝑗 ⊗ 𝑏 0𝑗 , where
Í
𝑏10 , . . . , 𝑏 0𝑠 are linearly independent over K.
Let M be a maximal ideal of A. The quotient ring A/M is a finitely
generated K-algebra, and is a field; consequently, it is an algebraic
extension of K, hence is isomorphic to K since K is algebraically closed.
Let clM : A → K be the corresponding morphism of K-algebras with
 9.1. NOETHER’S NORMALIZATION THEOREM 457

kernel M. Let also 𝜃M : A ⊗K B → B be the morphism clM ⊗ idB ; it is a

morphism of K-algebras.
Since
𝜃M ( 𝑓 )𝜃M (𝑔) = 𝜃M ( 𝑓 𝑔) = 0
and B is an integral domain, either 𝜃M ( 𝑓 ) = 0 or 𝜃M (𝑔) = 0. Moreover,
one has
𝑟
Õ 𝑠
Õ
𝜃M ( 𝑓 ) = clM (𝑎 𝑖 )𝑏 𝑖 and 𝜃M (𝑔) = clM (𝑎 0𝑗 )𝑏 0𝑗 .
𝑖=1 𝑗=1

Assume that 𝜃M (𝑔) = 0. Since 𝑏 1 , . . . , 𝑏 𝑟 are linearly independent

over K, we conclude that clM (𝑎 𝑖 ) = 0 for every 𝑖 ∈ {1, . . . , 𝑟}; in other
words, the ideal I = (𝑎1 , . . . , 𝑎 𝑟 ) is contained in M.
Similarly, if 𝜃M ( 𝑓 ) = 0, we obtain that the ideal J = (𝑎10 , . . . , 𝑎 0𝑠 ) is
contained in M.
In any case, I ∩ J ⊂ M.
This is valid for any maximal ideal M of A. By corollary 9.1.6, every
element of I ∩ J is nilpotent. Since A is an integral domain, I ∩ J = 0.
Assume that 𝑓 ≠ 0. Then I ≠ 0; let thus 𝑥 be a non-zero element of I.
For every 𝑦 ∈ J, one has 𝑥𝑦 ∈ I ∩ J, hence 𝑥𝑦 = 0. Since A is an integral
domain, this implies 𝑦 = 0, hence J = 0, hence 𝑎10 = · · · = 𝑎 0𝑠 = 0 and
𝑔 = 0. This concludes the proof that A ⊗K B is an integral domain.
b) The proof is analogous. Let 𝑓 ∈ A ⊗K B be a nilpotent element. We
write 𝑓 = 𝑟𝑖=1 𝑎 𝑖 ⊗ 𝑏 𝑖 , where 𝑎1 , . . . , 𝑎 𝑟 ∈ A and 𝑏 1 , . . . , 𝑏 𝑟 ∈ B, chosen
Í
such that 𝑏 1 , . . . , 𝑏 𝑟 are linearly independent over K.
Let M be a maximal ideal of A, let clM : A → A/M ' K be the quotient
morphism and let 𝜃M = clM ⊗ idB : : A ⊗K B → B. Since 𝑓 is nilpotent,
𝜃M ( 𝑓 ) is nilpotent in B, hence 𝜃M ( 𝑓 ) = 0, because B is reduced. Since
𝜃M ( 𝑓 ) = 𝑟𝑖=1 clM (𝑎 𝑖 )𝑏 𝑖 , this implies that the ideal I = (𝑎1 , . . . , 𝑎 𝑟 ) is
Í
contained in M.
By corollary 9.1.6, every element of I is nilpotent. Since A is reduced,
I = (0) and 𝑎1 = · · · = 𝑎 𝑟 = 0. This proves 𝑓 = 0.
We now remove the assumption that A and B are finitely generated.
For a), we consider 𝑓 , 𝑔 ∈ A ⊗K B such that 𝑓 𝑔 = 0; for b), we
consider 𝑓 ∈ A ⊗K B such that 𝑓 is nilpotent. In both cases, there
exist finitely generated subalgebras A0 ⊂ A and B0 ⊂ B such that
458 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

𝑓 ∈ A0 and 𝑔 ∈ B0 (with the notation introduced above, just take

A0 = K[𝑎1 , . . . , 𝑎 𝑟 , 𝑎10 , . . . , 𝑎 0𝑠 ] and B0 = K[𝑏 1 , . . . , 𝑏 𝑟 , 𝑏10 , . . . , 𝑏 0𝑠 ] in case a),
and A0 = K[𝑎1 , . . . , 𝑎 𝑟 ] and B0 = K[𝑏 1 , . . . , 𝑏 𝑟 ] in case b)). These algebras
A0 and B0 are integral domains (resp. reduced), and A0 ⊗K B0 is a
subalgebra of A ⊗K B. One thus has 𝑓 𝑔 = 0 in A0 ⊗K B0 (resp. 𝑓 is
nilpotent in A0 ⊗K B0). By the finitely generated case, 𝑓 = 0 or 𝑔 = 0 (resp.
𝑓 = 0), as was to be shown. 

9.2. Dimension and transcendence degree

Definition (9.2.1). — Let A be a ring.
Let P be a prime ideal of A. The height of P is the supremum, denoted
by htA (P), of the lengths, 𝑛, of chains of prime ideals (P0 , P1 , . . . , P𝑛 ) where
P𝑛 = P.
The dimension of A, denoted dim(A), is the supremum of the heights of its
prime ideals.

Recall that a chain is a strictly increasing finite sequence. We say

that such a chain (P0 , . . . , P𝑛 ) is maximal if it cannot be extended by
inserting another prime ideal, either before P0 , between P 𝑘−1 and P 𝑘 for
any 𝑘 ∈ {1, . . . , 𝑛}, nor after P𝑛 .

Example (9.2.2). — One has htA (P) = 0 if and only if P is a minimal prime
ideal of A. If A is a field or, more generally, if A is an artinian ring, then
dim(A) = 0.

Example (9.2.3) (Principal ideal domains). — If A is a principal ideal

domain which is not a field, then dim(A) = 1. In particular, dim(Z) = 1 and
dim(K[X]) = 1 for any field K.
The prime ideals of A are (0), and the maximal ideals (𝑝), where
𝑝 ranges among all irreducible elements of A. One has htA ((0)) = 0.
Moreover, the chains of prime ideals beginning with (0) are (0) and
((0), (𝑝)), where 𝑝 is an irreducible element of A, so that htA ((𝑝)) = 1.
This shows that dim(A) = 1.

Example (9.2.4). — Recall that the map Q ↦→ QAP induces a bijection

between the prime ideals of A contained in P and the prime ideals
 9.2. DIMENSION AND TRANSCENDENCE DEGREE 459

of AP , and this bijection respects the inclusion relation. Consequently,

htA (P) = dim(AP ).

Example (9.2.5). — Let A be a unique factorization domain. A prime ideal

of A has height 1 if and only if it is generated by an irreducible element of A.
Let P be a prime ideal of A such that htA (P) = 1. In particular, P ≠ 0;
let thus 𝑓 be any non-zero element of P. Since P is prime, 𝑓 is not a unit
in A and one of its irreducible factors, say 𝑎, must belong to P. Since
A is a unique factorization domain, the ideal (𝑎) is prime and one has
inclusions (0) ( (𝑎) ⊂ P. The hypothesis that ht(P) = 1 implies the
equality P = (𝑎).
Conversely, let 𝑎 be an irreducible element of A, so that (𝑎) is a prime
ideal of A. The strict inclusion (0) ( (𝑎) implies that ht((𝑎)) > 1. Let P be
a non-zero ideal of A contained in (𝑎). Let 𝑥 ∈ P be a non-zero element
whose number of irreducible factors is minimal. Since P ⊂ (𝑎), there
exists 𝑦 ∈ A such that 𝑥 = 𝑎𝑦; then the number of irreducible factors
of 𝑦 is one less than the number of irreducible factors of 𝑥, so that 𝑦 ∉ P.
Since P is prime, this implies that 𝑎 ∈ P, hence P = (𝑎). This shows that
ht((𝑎)) = 1.

Remark (9.2.6) (Geometrical interpretations). — a) Let K be an alge-

braically closed field, let I be an ideal of K[X1 , . . . , X𝑛 ] and let A =
K[X1 , . . . , X𝑛 ]/I. Any prime ideal of A defines an irreducible algebraic
set contained in 𝒵(I), and conversely (see exercise 2/10). Consequently,
the dimension of A is the supremum of the lengths of chains of irre-
ducible algebraic sets contained in 𝒵(I).
b) Let A be an arbitrary ring. Irreducible closed subsets of Spec(A) are
subsets of the form V(P), for some prime ideal P of A, and conversely
(see exercise 2/8).Consequently, dim(A) is the supremum of the lengths
of chains of irreducible closed subsets of Spec(A). For any prime
ideal P of A, ht(P) is the supremum of the lengths of those chains
of closed subsets which contain V(P) or, equivalently, which contain
the point P ∈ Spec(A). This is interpreted as the codimension of V(P)
in Spec(A).
c) Let K be a field. Generalizing the equality dim(K[X]) = 1 from exam-
ple 9.2.3, we will prove in theorem 9.2.10 below that dim(K[X1 , . . . , X𝑛 ]) =
460 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

𝑛. The chain of prime ideals

(0) ( (X1 ) ( (X1 , X2 ) ( · · · ( (X1 , . . . , X𝑛 )

shows that dim(K[X1 , . . . , X𝑛 ]) > 𝑛, but the opposite inequality is more

difficult.

Remark (9.2.7). — Without any hypotheses, the dimension of a ring may

be infinite. We shall show that the dimension of a finitely generated
algebra over a field is finite. We will also prove in theorem 9.3.3 that if
the ring A is noetherian, then the height of any prime ideal is finite, and
that the dimension of a noetherian local ring is finite. However, there
exist noetherian rings of infinite dimension (exercise 9/12).

In order to relate dimension and transcendence degrees of K-algebras,

we first study the behavior of dimension with respect to integral exten-
sions of rings.

Theorem (9.2.8) (Going-up theorem of Cohen–Seidenberg)

Let B be a ring and let A be a subring of B. Assume that B is integral over A.
a) Let Q be a prime ideal of B and let P = Q ∩ A. Then P is a maximal ideal
of A if and only if Q is a maximal ideal of B.
b) Let Q ⊂ Q0 be prime ideals of B such that Q ∩ A = Q0 ∩ A. Then Q = Q0.
c) The canonical map from Spec(B) to Spec(A) is surjective: for every prime
ideal P of A, there exists a prime ideal Q of B such that Q ∩ A = P.

Proof. — a) Passing to the quotients, one gets an integral extension of

integral domains A/P ⊂ B/Q. By proposition 4.2.5, A/P is a field if and
only if B/Q is a field; in other words, P is maximal in A if and only if Q
is maximal in B.
b) Let P = Q ∩ A and let us consider the integral extension of rings
AP ⊂ BP induced by localization by the multiplicative subset A P (it is
indeed injective by proposition 3.6.6, and integral by lemma 4.2.4). The
ideal AP ∩ QBP of AP contains the maximal ideal PAP of AP , but does
not contain 1, hence it is equal to PAP . Then, a) implies that QBP is a
maximal ideal of BP .
 9.2. DIMENSION AND TRANSCENDENCE DEGREE 461

Since Q ∩ A = P = Q0 ∩ A, the same arguments imply that Q0BP is a

maximal ideal of BP . However, the inclusion Q ⊂ Q0 implies an inclusion
QBP ⊂ Q0BP . One thus has QBP = Q0BP .
Since localization induces a bijection from the set of prime ideals of B
disjoint from A P to the set of prime ideals of BP , one gets Q = Q0.
c) Let P be a prime ideal of A and let us consider the integral extension
AP ⊂ BP of localized rings. Let M be a maximal ideal of BP . By a),
M ∩ AP is a maximal ideal of AP , hence M ∩ AP = PAP , and P ⊂ M ∩ A.
There exists a unique prime ideal Q of B such that Q ∩ (A P) = ∅ and
M = QBP . The intersection Q ∩ A is a prime ideal of A, contained in P
by construction; by what precedes, it contains P, hence Q ∩ A = P. This
concludes the proof of the theorem of Cohen–Seidenberg. 

Corollary (9.2.9). — Let B be a ring, let A be a subring of B. If B is integral

over A, then dim(A) = dim(B).

Proof. — Let (Q0 , . . . , Q𝑛 ) be a chain of prime ideals of B. Let us intersect

these ideals with A; this gives an increasing family (Q0 ∩ A, . . . , Q𝑛 ∩ A)
of prime ideals of A. By part b) of theorem 9.2.8, this is even a chain of
prime ideals, so that dim(A) > dim(B).
Conversely, let P0 ( · · · ( P𝑛 be a chain of prime ideals of A. For
each 𝑚 ∈ {0, . . . , 𝑛}, let us construct by induction a prime ideal Q𝑚 of B
such that Q𝑚 ∩ A = P𝑚 and such that Q0 ⊂ · · · ⊂ Q𝑛 . This will imply
that dim(B) > dim(A), hence the corollary.
By part c) of theorem 9.2.8, there exists a prime ideal Q0 of B such
that Q0 ∩ A = Q0 . Assume Q0 , . . . , Q𝑚 are defined. Let us consider the
integral extension A/P𝑚 ⊂ B/Q𝑚 of integral domains. By theorem 9.2.8,
c), applied to the prime ideal P𝑚+1 /P𝑚 of A/P𝑚 , there exists a prime
ideal Q of the ring B/Q𝑚 such that Q ∩ (A/P𝑚 ) = P𝑚+1 /P𝑚 . Then,
there exists a prime ideal Q𝑚+1 containing Q𝑚 such that Q = Q𝑚+1 /Q𝑚 .
Moreover, Q𝑚+1 ∩ A = P𝑚+1 . 

The next theorem is at the ground of a dimension theory in Algebraic

geometry.
462 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

Theorem (9.2.10). — Let K be a field, let A be a finitely generated K-algebra.

Assume that A is an integral domain and let F be its field of fractions. Then
dim(A) = tr degK (F).

Proof. — We prove the theorem by induction on the transcendence

degree of F.
If tr degK (F) = 0, then A is integral over K; being a finitely generated K-
algebra, it is finitely dimensional, hence a field. One thus has dim(A) = 0.
Now assume that the theorem holds for finitely generated K-
algebras whose field of fractions has transcendence degree strictly less
than tr degK (F).
By Noether’s normalization theorem (theorem 9.1.1), there exists
an integer 𝑛 > 0, elements 𝑎1 , . . . , 𝑎 𝑛 of A such that the morphism
𝑓 : K[X1 , . . . , X𝑛 ] → A such that 𝑓 (X𝑖 ) = 𝑎 𝑖 is injective, and such that
A is integral over the subring B = K[𝑎1 , . . . , 𝑎 𝑛 ] = 𝑓 (K[X1 , . . . , X𝑛 ]).
Moreover, 𝑛 = tr degK (F). By corollary 9.2.9, it suffices to prove that the
dimension of the polynomial ring K[X1 , . . . , X𝑛 ] is equal to 𝑛.
We already stated in remark 9.2.6 that dim(K[X1 , . . . , X𝑛 ]) > 𝑛, in
view of the chain ((0), (X1 ), (X1 , X2 ), . . . , (X1 , . . . , X𝑛 )) of prime ideals of
K[X1 , . . . , X𝑛 ].
Conversely, let ((0), P1 , . . . , P𝑚 ) be a chain of prime ideals of
K[X1 , . . . , X𝑛 ] and let us set A0 = K[X1 , . . . , X𝑛 ]/P1 . Since P1 is a prime
ideal, the ring A0 is an integral domain; let F0 be its field of fractions.
Moreover, A0 is a finitely generated K-algebra and dim(A0) > 𝑚 − 1,
because ((0), P2 /P1 , . . . , P𝑚 /P1 ) is a chain of prime ideals of A0 of
length 𝑚 − 1. On the other hand, any non-zero polynomial 𝑓 ∈ P1
furnishes gives a non-trivial algebraic dependence relation between the
classes 𝑥 1 , . . . , 𝑥 𝑛 of X1 , . . . , X𝑛 in A0. Consequently, tr degK (F0) 6 𝑛 − 1.
(See also example 4.8.11.) By induction, one has tr degK (F0) = dim(A0),
hence 𝑚 − 1 6 𝑛 − 1, and 𝑚 6 𝑛. This concludes the proof. 

In the course of the proof of the theorem, we established the following

particular case.

Corollary (9.2.11). — For any field K, one has dim(K[X1 , . . . , X𝑛 ]) = 𝑛.

 9.3. KRULL’S HAUPTIDEALSATZ AND APPLICATIONS 463

9.3. Krull’s Hauptidealsatz and applications

By theorem 9.2.10, the dimension of a finitely generated algebra over
a field is finite. One could think that the same property holds for
noetherian rings: after all, the dimension of a ring involves strictly
increasing sequences of prime ideals, and the noetherian property tells
us that every such sequence is finite. However, it does not: although
there are no infinite strictly increasing sequences of prime ideals, there
may be such sequences of arbitrarily large length.
One of the main consequences of this section is that local noetherian
rings are finite dimensional. Stated differently, the height of any prime
ideal of a noetherian ring is finite.
Since any such prime ideal is generated by a finite set (this is the
definition of a noetherian ring!), it is natural to investigate the behavior of
dimension when one quotients a ring by a principal ideal. Geometrically,
this will amount to understand the dimension of a hypersurface.
Theorem (9.3.1) (Krull’s Hauptidealsatz). — Let A be a noetherian ring, let
𝑎 ∈ A and let P be a prime ideal of A, minimal among those containing 𝑎.
a) One has ht(P) 6 1.
b) If, moreover, 𝑎 is regular, then ht(P) = 1.
Proof. — a) We need to prove that that there is no chain (Q0 , Q, P)
of prime ideals in A. So let Q0 , Q, P be prime ideals of A such that
Q0 ⊂ Q ( P and let us prove that Q = Q0.
We first pass to the quotient by Q0: in the noetherian ring A/Q0, we
have the prime ideals 0 ⊂ Q/Q0 ( P/Q0. Moreover, the prime ideal P/Q0
of A/Q0 is minimal among the prime ideals of A/Q0 containing the class
of 𝑎. This allows to assume that A is an integral domain and Q0 = 0.
We then replace A by its localization at the prime ideal P. The ring
AP is a noetherian local ring, with inclusions 0 ⊂ QAP ( PAP of prime
ideals, and the maximal ideal PAP is minimal among the prime ideals
of AP containing 𝑎/1.
We are thus reduced to the following particular case: the ring A is an
integral domain, local, noetherian, its maximal ideal P is the only prime
ideal containing the element 𝑎, the ideal Q of A is prime and distinct
from P, and we need to prove that Q = 0.
464 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

For every integer 𝑛 > 1, let Q𝑛 = Q𝑛 AQ ∩ A. By definition of

localization, Q𝑛 is the set of elements 𝑥 ∈ A such that there exists 𝑦 ∉ Q
such that 𝑥𝑦 ∈ Q𝑛 .
The sequence (Q𝑛 )𝑛 is decreasing; let us prove that it is stationary.
The ring A/𝑎A is noetherian, and since its prime ideals are in bijection
with the prime ideals of A containing 𝑎, P(A/𝑎A) is its only prime
ideal. Consequently (see theorem 6.4.14), A/𝑎A is an artinian ring. In
particular, the decreasing sequence (Q𝑛 (A/𝑎A))𝑛 of ideals of A/𝑎A is
stationary and there exists an integer 𝑛0 such that

Q𝑛 + 𝑎A = Q𝑛+1 + 𝑎A

for every integer 𝑛 > 𝑛0 .

Let 𝑛 be an integer > 𝑛0 and let 𝑥 ∈ Q𝑛 . There exists 𝑦 ∈ A such
that 𝑥 + 𝑎𝑦 ∈ Q𝑛+1 ; since Q𝑛+1 ⊂ Q𝑛 , we have 𝑎𝑦 ∈ Q𝑛 , hence there
exists 𝑧 ∉ Q such that 𝑎𝑦𝑧 ∈ Q𝑛 . Since 𝑎 ∉ Q, the product 𝑎𝑧 does
not belong to Q, so that 𝑦 belongs to Q𝑛 . (We could have used the
result of exercise 6/37 that Q𝑛 is a Q-primary ideal.) Consequently,
𝑥 ∈ 𝑎Q𝑛 + Q𝑛+1 , hence 𝑥 ∈ Q𝑛+1 + PQ𝑛 . This implies the equality of
ideals
Q𝑛 = Q𝑛+1 + PQ𝑛 .

Since the ring A is noetherian, Q𝑛 is finitely generated and Nakayama’s

lemma (theorem 6.1.1, observe that P is the Jacobson radical of A) implies
that Q𝑛 = Q𝑛+1 for all 𝑛 > 𝑛0 .
Then,

Q𝑛 AQ = Q𝑛 AQ = Q𝑛+1 AQ = Q𝑛+1 AQ = QQ𝑛 AQ .

Again, AQ is noetherian and Q𝑛 AQ is finitely generated; consequently,

Nakayama’s lemma implies that Q𝑛 AQ = 0. Since A is an integral
domain, this implies that Q = 0, as was to be shown.
b) If 𝑎 is regular, then it does not belong to any associated prime ideal
of A (theorem 6.5.8, a)), hence no minimal prime ideal of A contains 𝑎.
Since 𝑎 ∈ P, the prime ideal P is not minimal and htA (P) > 0. By part a),
the height of P is equal to 1. 
 9.3. KRULL’S HAUPTIDEALSATZ AND APPLICATIONS 465

Corollary (9.3.2). — Let A be a noetherian integral domain. Then A is a

unique factorization domain if and only if every prime ideal of height 1 is a
principal ideal.

Proof. — We have already proved in example 9.2.5 that if A is a unique

factorisation domain, then its prime ideals of height 1 are principal
ideals (generated by irreducible elements).
Let us now assume that every prime ideal of A which is of height 1 is a
principal ideal and let us prove that A is a unique factorization domain.
Since A is noetherian, every increasing sequence of principal ideals
of A is stationary, so that it suffices to prove that irreducible elements
generate prime ideals. Let 𝑝 ∈ A be an irreducible element and let P
be a prime ideal of A, minimal among those containing 𝑝. Since A is a
domain and 𝑝 ≠ 0, theorem 9.3.1 implies that htA (P) = 1. By assumption,
the ideal P is principal, hence there exists 𝑎 ∈ A such that P = (𝑎). The
inclusion (𝑝) ⊂ P = (𝑎) implies that 𝑎 divides 𝑝; let 𝑏 ∈ A be such that
𝑝 = 𝑎𝑏. Since P is prime, 𝑎 is not a unit. By definition of an irreducible
element, 𝑏 is a unit and P = (𝑎) = (𝑝). This concludes the proof of the
corollary. 

Theorem (9.3.3). — Let A be a noetherian ring and let P be a prime ideal of A.

a) The height htA (P) of P is finite.
b) More precisely, htA (P) is the the least integer 𝑛 such that there exist
𝑎 1 , . . . , 𝑎 𝑛 ∈ P such that the prime ideal P is minimal among those contain-
ing 𝑎1 , . . . , 𝑎 𝑛 .

Proof. — a) We first prove the following statement by induction on 𝑛:

Let A be a noetherian ring, let 𝑎1 , . . . , 𝑎 𝑛 ∈ A and let P be a prime ideal which
is minimal among those containing 𝑎1 , . . . , 𝑎 𝑛 , then ht(P) 6 𝑛. (Observe
that for 𝑛 = 1, this is exactly Krull’s Hauptidealsatz, theorem 9.3.1.)
The assertion holds if ht(P) = 0 (trivially), or if 𝑛 = 0 (because P is then
a minimal prime ideal of A, hence ht(P) = 0). Let us thus assume that
ht(P) > 0 and 𝑛 > 0. In the fraction ring AP , the ideal PAP is the maximal
ideal, and is minimal among those containing 𝑎1 /1, . . . , 𝑎 𝑛 /1; moreover,
htA (P) = htAP (PAP ). We may thus assume that A is a local ring with
maximal ideal P. By the definition of the height of a prime ideal, to
466 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

establish that htA (P) 6 𝑛, we need to prove that for any prime ideal Q
of A such that Q ( P, one has htA (Q) 6 𝑛 − 1. Since A is noetherian,
there exists a prime ideal Q0 such that Q ⊂ Q0 ( P, and which is maximal
among those ideals. Since one has htA (Q) 6 htA (Q0), it suffices to prove
that htA (Q0) 6 𝑛 − 1. We may thus assume that there is no prime ideal Q0
in A such that Q ( Q0 ( P.
Since Q ( P, the definition of P implies that there exists an integer 𝑖
such that 𝑎 𝑖 ∉ Q. To fix the notation, let us assume that 𝑎 1 ∉ Q.
Then Q ( Q + (𝑎1 ) ⊂ P, so that P is a minimal prime ideal which
contains Q + (𝑎1 ), hence is the unique prime ideal containing Q + (𝑎 1 ).
Consequently, every element of P is nilpotent modulo Q + (𝑎1 ). In
particular, there exist an integer 𝑚 > 1 and elements 𝑥2 , . . . , 𝑥 𝑛 ∈ A,
𝑦2 , . . . , 𝑦𝑛 ∈ Q, such that

𝑎2𝑚 = 𝑎1 𝑥 2 + 𝑦2 , . . . , 𝑎 𝑛𝑚 = 𝑎1 𝑥 𝑛 + 𝑦𝑛 .

These relations show that any prime ideal of A which contains 𝑎1 and
𝑦2 , . . . , 𝑦𝑛 also contains 𝑎2 , . . . , 𝑎 𝑛 , hence contains P. In other words,
P/(𝑦2 , . . . , 𝑦𝑛 ) is a prime ideal of the quotient ring A/(𝑦2 , . . . , 𝑦𝑛 ) which
is minimal among those containing the class of 𝑎1 . By Krull’s Haup-
tidealsatz (theorem 9.3.1), the height of P/(𝑦2 , . . . , 𝑦𝑛 ) in A/(𝑦2 , . . . , 𝑦𝑛 )
is 6 1. Since (𝑦2 , . . . , 𝑦𝑛 ) ⊂ Q ( P, the ideal Q/(𝑦2 , . . . , 𝑦𝑛 ) is then a
minimal prime ideal of A/(𝑦2 , . . . , 𝑦𝑛 ). Equivalently, Q is a minimal
prime ideal of A among those containing (𝑦2 , . . . , 𝑦𝑛 ). By induction,
htA (Q) 6 𝑛 − 1, as was to be shown.
b) It follows that the height of any prime ideal P of A is finite. Indeed,
since A is noetherian, the ideal P is finitely generated and there exists
𝑎1 , . . . , 𝑎 𝑛 ∈ A such that P = (𝑎1 , . . . , 𝑎 𝑛 ). By the part of the proof already
shown, one has htA (P) 6 𝑛.
c) We now prove by induction on 𝑛 the following statement: Let A be
noetherian ring and let P be a prime ideal of A, let 𝑛 = htA (P). There exist
elements 𝑎 1 , . . . , 𝑎 𝑛 ∈ P such that P is the minimal prime ideal containing
(𝑎1 , . . . , 𝑎 𝑛 ).
This holds if 𝑛 = 0, for P is then a minimal prime ideal.
Assume that 𝑛 = htA (P) > 0 and that the result holds for every prime
ideal of A of height < 𝑛. Since 𝑛 > 0, P is not contained in any minimal
 9.4. HEIGHTS AND DIMENSION 467

prime ideal of A. Since the set of minimal prime ideals of A is finite, the
Prime avoidance lemma 2.2.12 implies that there exists an element 𝑎1 ∈ P
which does not contain any minimal prime ideal of A. Consequently,
the height of any prime ideal of A which is minimal among those
containing 𝑎1 is strictly positive. Thus, any chain of prime ideals of A
containing 𝑎1 and contained in P can be extended by one of the minimal
prime ideals of A, so that the height of the prime ideal P/(𝑎1 ) of the
ring A/(𝑎1 ) is at most htA (P) − 1 = 𝑛 − 1. By induction, there exist
𝑎 2 , . . . , 𝑎 𝑛 ∈ P such that the height of the prime ideal P/(𝑎1 , . . . , 𝑎 𝑛 ) of
the ring A/(𝑎1 , . . . , 𝑎 𝑛 ) is zero. This proves that P is a minimal prime
ideal among those containing (𝑎1 , . . . , 𝑎 𝑛 ). This concludes the proof of
the theorem. 

Corollary (9.3.4). — Let A be a noetherian local ring. Then dim(A) is finite.

More prexisely,
p dim(A) is the least integer 𝑛 such that there exist 𝑎1 , . . . , 𝑎 𝑛
such that (𝑎1 , . . . , 𝑎 𝑛 ) is the maximal ideal of A.

Proof. — If A is local, every maximal chain of prime ideal ends at its

maximal ideal M, hence dim(A) = htA (M). The corollary thus follows
from theorem 9.3.3 applied to M. 

9.4. Heights and dimension

Proposition (9.4.1). — Let A be an integral domain, let E be its field of fractions
and let E → F be a finite normal extension of fields. Assume that A is integrally
closed in E and let B the integral closure of A in F.
For every prime ideal P of A, the group Aut(F/E) acts transitively on the set
of prime ideals Q of B such that Q ∩ A = P.

Proof. — Let G = Aut(F/E). It is a finite group of automorphisms of F;

the subfield FG of F is a radicial extension of E (corollary 4.6.9) and
FG ⊂ F is a Galois extension of group G (Artin’s lemma 4.6.4).
First of all, according to the Going-up theorem (theorem 9.2.8), the set
of prime ideals Q of B such that Q ∩ A = P is nonempty.
Let Q and Q0 be two prime ideals of B such that Q ∩ A = Q0 ∩ A = P.
468 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

Let 𝑥 ∈ Q0, let 𝑦 = 𝜎∈G 𝜎(𝑥); since 𝑦 ∈ FG and since the extension
Î
E → FG is radicial, there exists an integer 𝑞 > 1 such that 𝑦 𝑞 ∈ E
(lemma 4.4.16; one has 𝑞 = 1 if the characteristic of E is zero).
By definition of B, 𝑥 is integral over A, as well as its images 𝜎(𝑥), for
𝜎 ∈ G. Consequently, 𝑦 is integral over A, hence 𝑦 𝑞 is integral over A.
Since 𝑦 𝑞 ∈ E and A is integrally closed in E, one has 𝑦 𝑞 ∈ A. Since 𝑥 ∈ Q0,
it follows that 𝑦 𝑞 ∈ Q0 ∩ A = P. Using that P = Q ∩ A, we deduce that
𝑦 𝑞 ∈ Q. Finally, since Q is a prime ideal, we conclude that 𝑦 ∈ Q.
Moreover, since 𝑦 = 𝜎∈G 𝜎(𝑥), there exists an element 𝜎 ∈ G such that
Î
𝜎(𝑥) ∈ Q. Since G is a group, this gives the inclusion Q0 ⊂ 𝜎∈G 𝜎(Q).
Ð
Let 𝜎 ∈ G. Let us show that 𝜎(Q) is a prime ideal of B. Indeed, if 𝑏 ∈ F,
then 𝑏 and 𝜎(𝑏) satisfy the same polynomial relations with coefficients
in E. In particular, 𝑏 is integral over A if and only if 𝜎(𝑏) is integral
over A. This implies that 𝜎(B) = B, hence 𝜎(Q) is a prime ideal of B.
By the Prime avoidance lemma (lemma 2.2.12), there exists 𝜎 ∈ G
such that Q0 ⊂ 𝜎(Q). Then P = Q0 ∩ A ⊂ 𝜎(Q) ∩ A = Q ∩ A = P,
hence Q0 ∩ A = 𝜎(Q) ∩ A. By the first Cohen-Seidenberg theorem
(theorem 9.2.8), we thus have Q0 = 𝜎(Q). 

Theorem (9.4.2) (Going-down theorem of Cohen–Seidenberg)

Let B be an integral domain, let A be a subring of B. One assumes that A is
integrally closed in its field of fractions and that B is integral over A.
Let P0 , . . . , P𝑛 be prime ideals of A such that P0 ⊂ . . . P𝑛 and let Q𝑛 be
a prime ideal of B such that Q𝑛 ∩ A = P𝑛 . Then, there exist prime ideals
Q0 , . . . , Q𝑛−1 in B such that Q0 ⊂ · · · ⊂ Q𝑛 and Q𝑚 ∩ A = P𝑚 for every
𝑚 ∈ {0, . . . , 𝑛}.

Proof. — Let E be the field of fractions of A, let F be the field of fractions

of B and let F0 be a finite normal extension of E containing F. Let B0 be
the integral closure of A in F0. According to theorem 9.2.8, there exist
prime ideals Q0𝑚 in B0 (for 0 6 𝑚 6 𝑛), such that Q00 ⊂ · · · ⊂ Q0𝑛 and
P𝑚 = Q0𝑚 ∩ A for every 𝑚 ∈ {0, . . . , 𝑛}.
Let Q0 be a prime ideal of B0 such that Q0 ∩ B = Q𝑛 . Since A is
integrally closed in its field of fractions, proposition 9.4.1 asserts the
existence of an automorphism 𝜎 ∈ Aut(F0/E) such that 𝜎(Q0𝑛 ) = Q0. For
every integer 𝑚 ∈ {0, . . . , 𝑛 − 1}, let Q𝑚 = 𝜎(Q0𝑚 ) ∩ B. By construction,
 9.4. HEIGHTS AND DIMENSION 469

(Q0 , . . . , Q𝑛−1 ) is an increasing family of prime ideals of B contained

in Q𝑛 . Moreover, for every 𝑚 ∈ {0, . . . , 𝑛 − 1}, one has
Q𝑚 ∩ A = 𝜎(Q0𝑚 ) ∩ B ∩ A = 𝜎(Q0𝑚 ∩ A) = 𝜎(P𝑚 ) = P𝑚 .
The theorem is proved. 

Corollary (9.4.3). — Let B be an integral domain, let A be a subring of B. One

assumes that A is integrally closed in its field of fractions and that B is integral
over A. Then for every prime ideal Q of B, one has htB (Q) = htA (Q ∩ A).
Proof. — Let P = Q ∩ A. Let (Q0 , . . . , Q𝑛 ) be a chain of prime ideals
of B, contained in Q. By theorem 9.2.8, their intersections with A,
(Q0 ∩ A, . . . , Q𝑛 ∩ A), form a chain of prime ideals of A contained in P.
This implies that htA (P) > htB (Q).
Conversely, let (P0 , . . . , P𝑛 ) be a chain of prime ideals of A contained
in P. By theorem 9.4.2, there exists a family (Q0 , . . . , Q𝑛 ) of prime
ideals of B such that Q0 ⊂ · · · ⊂ Q𝑛 ⊂ Q and Q𝑚 ∩ A = P𝑚 for every
𝑚 ∈ {0, . . . , 𝑛}. By theorem 9.2.8, (Q0 , . . . , Q𝑛 ) is a chain of prime ideals
in B. Consequently, htB (Q) > htA (P).
This concludes the proof of the corollary. 

Theorem (9.4.4). — Let K be a field and let A be a finitely generated K-algebra.

Assume that A is an integral domain. Then, for every prime ideal P of A, one
has
htA (P) = dim(A) − dim(A/P).
In particular, for every maximal ideal M of A, one has htA (M) = dim(A).
Proof. — Let (P0 , . . . , P𝑛 ) be a chain of prime ideals of A contained in P.
Let also (Q0 , . . . , Q𝑚 ) be a chain of prime ideals of A/P. For every
integer 𝑖 ∈ {0, . . . , 𝑚}, let Q0𝑖 be the preimage of Q𝑖 in A by the canonical
surjection from A to A/P. Then (Q00 , . . . , Q0𝑚 ) is a chain of prime ideals
of A containing P. Now observe that (P0 , . . . , P𝑛 , Q01 , . . . , Q0𝑚 ) is chain of
prime ideals of A — the inclusions P𝑛 ⊂ P ⊂ Q00 ( Q01 establish the strict
inclusion P𝑛 ( Q01 , and the other are obvious. We thus have shown the
inequality dim(A) > htA (P) + dim(A/P).
The converse inequality is more delicate and we prove it by induction
on dim(A). It clearly holds if dim(A) = 0, since then A is a field and
470 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

P = (0). Let us assume that it holds for all prime ideals of a finitely
generated K-algebra of dimension < dim(A) which is an integral domain.
Let us first apply Noether’s normalization theorem: there exist an
integer 𝑛 and elements 𝑎1 , . . . , 𝑎 𝑛 ∈ A, algebraically independent over K,
such that A is integral over the subring B = K[𝑎1 , . . . , 𝑎 𝑛 ], and dim(A) =
dim(B) = 𝑛. Let Q = P ∩ B. Since the natural morphism from B/Q
to A/P is injective and integral, one also has dim(A/P) = dim(B/Q).
Moreover, theorem 9.4.2 asserts that htA (P) = htB (Q). Consequently, we
may assume that A = B, hence that A = K[X1 , . . . , X𝑛 ].
If P = (0), then htA (P) = 0, A ' A/P, so that the equality htA (P) +
dimA (A/P) = dim(A) holds.
Let us assume that P ≠ (0). Then, since it is a prime ideal, P contains an
irreducible element, say 𝑓 . By example 9.2.5, the ideal ( 𝑓 ) has height 1.
Then set A0 = A/( 𝑓 ) and P0 = P/( 𝑓 ). The K-algebra A0 is finitely
generated; it is an integral domain and, by example 4.8.11, its field of
fractions has transcendence degree 𝑛 − 1 over K. By theorem 9.2.10,
dim(A0) = 𝑛 − 1 = dim(A) − 1.
Moreover, A0/P0 is isomorphic to A/P, so that
dim(A/P) = dim(A0/P0).
On the other hand, a chain of length 𝑚 of prime ideals of A0 contained
in P0 corresponds to a chain of prime ideals of A contained in P and
containing ( 𝑓 ); adjoining (0), this gives a chain of length 𝑚 + 1 of prime
ideals of A contained in P. Consequently,
htA0 (P0) + 1 6 htA (P).
Finally, the induction hypothesis asserts that
htA0 (P0) + dim(A0/P0) = dim(A0).
Combining these four relations, we obtain
htA (P) + dim(A/P) > htA0 (P0) + 1 + dim(A0/P0)
> dim(A0) + 1 = dim(A).
This concludes the proof of the theorem. 
 9.5. FINITENESS OF INTEGRAL CLOSURE 471

Corollary (9.4.5). — Let K be a field and let A be a finitely generated K-algebra

which is an integral domain. Every maximal chain of prime ideals of A has
length dim(A).
Proof. — Let (P0 , P1 , . . . , P𝑛 ) be a maximal chain of prime ideals of A.
This means that this chain cannot be extended by inserting prime ideals
(which is a different assertion than asserting that this sequence has
maximal length). In particular, one has P0 = (0) because A is an integral
domain, and P𝑛 is a maximal ideal.
Let us now argue by induction on 𝑛. If 𝑛 = 0, then P0 is a maximal
ideal, hence A is a field and dim(A) = 0.
Let us now assume that 𝑛 > 1. Since the chain (P0 , P1 ) is maximal
among all chains terminating at P1 , we have htA (P1 ) = 1. The K-algebra
A/P1 is finitely generated and an integral domain, and (P1 /P1 , . . . , P𝑛 ⊂
P1 ) is a maximal chain of prime ideals of A/P1 . By induction, one has
𝑛 − 1 = dim(A/P1 ). Consequently,
𝑛 = (𝑛 − 1) + 1 = dim(A/P1 ) + htA (P1 ) = dim(A),
by theorem 9.4.4. 

9.5. Finiteness of integral closure

Proposition (9.5.1). — Let A be a noetherian integral domain and let E be its
field of fractions. Let F be a separable finite algebraic extension of F and let B
be the integral closure of A in F. If A is integrally closed, then B is a finitely
generated A-module, in particular, a noetherian ring.
Proof. — Let (𝑒1 , . . . , 𝑒 𝑛 ) be a basis of F as an E-vector space. Up to
multiplying them by a non-zero element of A, we may assume that
they belong to B. Since the extension E ⊂ F is separable, the symmetric
bilinear form defined by the trace is non-degenerate (theorem 4.7.7).
Consequently, there exists a basis ( 𝑓1 , . . . , 𝑓𝑛 ) of F as an E-evector space
such that for every 𝑖 and 𝑗 ∈ {1; . . . ; 𝑛}, TrE/F (𝑒 𝑖 𝑓 𝑗 ) = 0 if 𝑖 ≠ 𝑗, and
TrE/F (𝑒 𝑖 𝑓𝑖 ) = 1. Let D be a non-zero element of A such that D 𝑓𝑖 ∈ B for
every 𝑖.
Let then 𝑥 be an element of B, and let 𝑥1 , . . . , 𝑥 𝑛 ∈ E be such that
𝑥 = 𝑛𝑖=1 𝑥 𝑖 𝑒 𝑖 . For every 𝑖 ∈ {1; . . . ; 𝑛}, one has (D 𝑓𝑖 )𝑥 ∈ B, hence
Í
472 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

TrF/E (D 𝑓𝑖 𝑥) ∈ A, by corollary 4.7.6, hence D𝑥 𝑖 ∈ A. Consequently,

B ⊂ D−1 𝑛𝑖=1 A𝑒 𝑖 . In other words, B is an A-submodule of a free A-
Í
module of rank 𝑛. Since A is a noetherian ring, this implies that B is a
finitely generated A-module.
Since ideals of B are A-submodules, this also implies that B is noethe-
rian. 

Theorem (9.5.2). — Let 𝑘 be a field and let A be a finitely generated 𝑘-algebra

which is an integral domain. Let E be the field of fractions of A and let E ⊂ F
be a finite algebraic extension. Finally, let B be the integral closure of A in F.
Then, B is a finitely generated A-module.

To simplify the proof, we assume that the characteristic of 𝑘 is zero. The

general case can be proved along the same lines but requires an additional
study of inseparable extensions in the style of remark 4.4.17.
Proof. — Let us apply to A the Noether normalization lemma (theo-
rem 9.1.1). Let 𝑥 1 , . . . , 𝑥 𝑛 be elements of A, algebraically independent
over 𝑘, such that A is integral over A0 = 𝑘[𝑥1 , . . . , 𝑥 𝑛 ].
Let us remark that an element of F is integral over A if and only if it
is integral over A0 . Consequently, B is the integral closure of A0 in F.
Since the extension E0 = 𝑘(𝑥 1 , . . . , 𝑥 𝑛 ) ⊂ E is finite, the extension E0 ⊂ F
is finite too.
Let us also remark that A0 is a unique factorization domain (theo-
rem 6.3.15) hence is integrally closed (theorem 4.1.9). Since we assumed
that the characteristic of 𝑘 is zero, the extension E0 ⊂ F is separable.
Consequently, proposition 9.5.1 implies that B is a finitely generated
A0 -module, hence a finitely generated A-module. 

9.6. Dedekind rings

Definition (9.6.1). — Let A be an integral domain. One says that A is a
Dedekind ring if every ideal of A is a projective A-module.

Example (9.6.2). — a) A field is a Dedekind ring.

b) Since non-zero principal ideals of an integral domain A are free
A-modules of rank 1, any principal ideal domain is a Dedekind ring.
 9.6. DEDEKIND RINGS 473

c) Let A be a Dedekind ring and let S be a multiplicative subset of A.

Then S−1 A is a Dedekind ring.
Let indeed J be an ideal of S−1 A. There exists an ideal I of A such
that J = S−1 I. Since A is a Dedekind ring, the A-module I is projective.
Consequently, the S−1 A-module S−1 I is projective (example 7.3.5).

9.6.3. — Let A be an integral domain, let K be its field of fractions.

The set ℐ(A) of non-zero ideals of A is a commutative monoid for the
multiplication of ideals. The subset 𝒫(A) of ℐ(A) consisting of principal
ideals is a submonoid. The quotient monoid 𝒞(A) = ℐ(A)/𝒫(A) is
called the class monoid of A. This monoid is trivial if and only if every
ideal of A is principal. As we will show, Dedekind rings are actually
characterized by the property that 𝒞(A) is a group.
To study 𝒞(A), it is convenient to consider the more general setup
of fractional ideals. By definition, a fractional ideal of A is a non-zero
submodule of K of the form 𝑎I, where I is an ideal of A and 𝑎 ∈ K× .
Fractional ideals can be multiplied as ideals can be, and the set ℐ0 (A) of
fractional ideals is a monoid, with the set 𝒫0 (A) of principal fractional
ideals as a submonoid.
The inclusion ℐ(A) → ℐ0 (A) induces a morphism of monoids
from 𝒞(A) to ℐ0 (A)/𝒫0 (A). Since the class of a fractional ideal 𝑎I,
where I is an ideal of A, is equal to the class of I, this morphism is
surjective. Let then I and J be non-zero ideals of A which have the
same class in ℐ0 (A)/𝒫0 (A); let 𝑎, 𝑏 ∈ K× be such that 𝑎I = 𝑏J; write
𝑎 = 𝑎1 /𝑎2 and 𝑏 = 𝑏 1 /𝑏 2 , with 𝑎1 , 𝑎2 , 𝑏1 , 𝑏2 ∈ A; then 𝑎1 𝑏 2 I = 𝑏 1 𝑎2 J and
[I] = [𝑎1 𝑏 2 I] = [𝑏1 𝑎2 J] = [J] in 𝒞(A).
Finally, a fractional ideal of A is isomorphic, as an A-module to an
ideal of A, so that a ring A is a Dedekind ring if and only if fractional
ideal of A is projective.

9.6.4. — Let A be an integral domain and let K be its field of fractions.

Let I be a fractional ideal of A. One write I−1 for the set of elements
𝑎 ∈ K such that 𝑎I ∈ A. One has I · I−1 ⊂ A.
The map I−1 → I∨ given by 𝑎 ↦→ (𝑥 ↦→ 𝑎𝑥) is a morphism of A-modules.
It is injective, because I ≠ 0. Let us show that it is surjective; let 𝜑 ∈ I∨ .
474 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

Let 𝑥 be a non-zero element of I and let 𝑎 = 𝜑(𝑥)𝑥 −1 . Let 𝑦 be a non-

zero element of I; there exist 𝑢, 𝑣 ∈ A {0} such that 𝑢𝑥 = 𝑣𝑦; then
𝑢𝜑(𝑥) = 𝑣𝜑(𝑦), so that 𝜑(𝑦) = 𝑢𝑣 −1 𝜑(𝑥) = 𝑎𝑦. In particular, 𝑎𝑦 ∈ A for
every 𝑦 ∈ I, and 𝜑 is the image of 𝑎.

Lemma (9.6.5). — Let I be a fractional ideal of A. The following properties are

equivalent:
(i) One has I · I−1 = A;
(ii) The A-module I is projective;
(iii) The fractional ideal I is invertible in the monoid ℐ0 (A);
(iv) The class of I is invertible in 𝒞(A).
If they hold, I is finitely generated.

If these properties hold, one simply says that the fractional ideal I is
invertible.
Proof. — (i)⇒(ii). Since I · I−1 = A, there exist elements 𝑎 1 , . . . , 𝑎 𝑛 ∈ I
and 𝑏 1 , . . . , 𝑏 𝑛 ∈ I−1 such that 1 = 𝑛𝑖=1 𝑎 𝑖 𝑏 𝑖 . Then the morphism
Í
𝛿I,I : I∨ ⊗ I → EndA (I) maps 𝑏 𝑖 ⊗ 𝑎 𝑖 to idI , so that I is finitely generated
Í
and projective.
(ii)⇒(i). Assume that I is projective. Let (𝜑 𝑖 )𝑖 and (𝑎 𝑖 ) be families in I∨
and I respectively such that 𝑥 = 𝜑 𝑖 (𝑥)𝑎 𝑖 for every 𝑥 ∈ I (remark 7.3.4).
Í
For every 𝑖, let 𝑏 𝑖 ∈ I−1 be such that 𝜑 𝑖 (𝑥) = 𝑏 𝑖 𝑥 for every 𝑥 ∈ I. Since
I ≠ 0, this implies 𝑎 𝑖 𝑏 𝑖 = 1. Then 1 ∈ I · I−1 , so that I · I−1 = A.
Í
(i)⇒(iii). Assume that I · I−1 = A. By definition, this says that I is
invertible in ℐ0 (A), with inverse I−1 .
(iii)⇒(iv). If I is invertible in the monoid ℐ0 (A), then its class [I] is
invertible in the class monoid 𝒞(A).
(iv)⇒(i). Finally, assume that the class [I] of I is invertible in 𝒞(A)
and let J be a fractional ideal of A such that I · J is a principal fractional
ideal 𝑎A. Replacing J by 𝑎 −1 J, we may assume that I · J = A. Then, J ⊂ I−1 ,
hence A = I · J ⊂ I · I−1 ⊂ A, so that I · I−1 = A. This proves (i). 

Corollary (9.6.6). — An integral domain A is a Dedekind domain if and only

if the monoid 𝒞(A) is a group.
 9.6. DEDEKIND RINGS 475

When I is an invertible ideal, one can define I𝑛 for every integer 𝑛

by the formula I𝑛 = (I−1 )−𝑛 if 𝑛 < 0. From the relation I · I−1 = A, one
deduces that I𝑛+𝑚 = I𝑛 · I𝑚 for every 𝑚, 𝑛 ∈ Z.
Proposition (9.6.7). — Let A be a local integral domain. Then A is a Dedekind
ring if and only if it is a principal ideal domain.
Proof. — By example 9.6.2, principal ideal domains are Dedekind rings;
it thus suffices to prove that a Dedekind ring A which is a local ring is
principal. Let P be the maximal ideal of A.
Let I be an ideal of A; it is finitely generated because A is noetherian.
If I = PI, then Nakayama’s lemma (theorem 6.1.1) implies that I = 0. In
particular, I is principal.
Otherwise, let 𝑎 ∈ I PI and let us prove that I = (𝑎). Since 𝑎A ⊂ I, one
has 𝑎I−1 ⊂ A. If 𝑎I−1 ≠ A, then 𝑎I−1 is contained in the maximal ideal P
of A, that is, 𝑎I−1 ⊂ P; this implies 𝑎A = 𝑎I−1 · I ⊂ PI, a contradiction.
Consequently, the ideal I is principal. 

Theorem (9.6.8). — Let A be an integral domain and let S be the set of its
maximal ideals. Assume that every maximal ideal of A is invertible.
a) The ring A is a Dedekind ring.
b) For every fractional ideal I of A, there exists a unique element (𝑎M )M∈S
in Z(S) such that I = M∈S M𝑎M .
Î

c) If I = M∈S M𝑎M and J = M∈S M𝑏M , then I ⊂ J is equivalent to the

Î Î
inequalities 𝑏M 6 𝑎M for every M ∈ S.
Recall that we write M𝑛 for (M−1 )−𝑛 when 𝑛 < 0.
This result serves as a partial replacement, in Dedekind rings, to the
possible non-uniqueness of decomposition in irreducible elements.
Proof. — We first prove that for every non-zero ideal I of A, there exists
a family (𝑎M ) ∈ N(S) such that I = M∈S M𝑎M . This will also prove that I
Î
is invertible, hence A is a Dedekind ring.
Let us argue by contradiction; since A is noetherian, there exists a
Î
maximal counterexample I. Since A itself can be written A = M∈S M0 ,
one has I ≠ A, hence there exists a maximal ideal P of A such that
I ⊂ P. Set J = IP−1 ; one has J ⊂ A and I = IP−1 · P = PJ. In particular,
I ≠ J, since otherwise, Nakayama’s lemma (corollary 6.1.4) implies
476 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

that I = 0. Consequently, there exists a family (𝑏 M )M∈S ∈ N(S) such

that J = M∈S M𝑏M . Set 𝑎M = 𝑏 M for M ≠ P and 𝑎P = 𝑏P + 1; then
Î
(𝑎M )M∈S ∈ N(S) and I = PJ = M∈S M𝑎M , a contradiction. In particular,
Î
the result holds for every non-zero ideal of A.
Let then I be an arbitrary fractional ideal. There exists 𝑎 ∈ A {0} such
that 𝑎I ⊂ A. Let (𝑏 M ) ∈ N(S) be such that 𝑎I = M∈S M𝑏M . Let also (𝑎M ) ∈
Î
N(S) be such that 𝑎A = M∈S M𝑎M . Then the ideal J = M∈S M𝑏M −𝑎M
Î Î
satisfies 𝑎J = 𝑎I, hence J = I.
The uniqueness of a family (𝑎M )M∈S ∈ Z(S) such that I = M∈S M𝑎M
Î
will follow from the final assertion.
Let thus I = M∈S M𝑎M and J = M∈S M𝑏M be two fractional ideals. If
Î Î
𝑎M 6 𝑏 M for all M ∈ S, then J = I · M∈S M𝑏M −𝑎M , hence J ⊂ I. Conversely,
Î
assume that J ⊂ I and let us prove that 𝑎 M 6 𝑏 M for all M. Multiplying
both sides of this inclusion by M∈S M− inf(𝑎M ,𝑏M ) , we may assume that
Î
inf(𝑎M , 𝑏M ) = 0 for every M ∈ S; we now need to prove that 𝑎 M = 0
for every M. Assume that there exists P ∈ S with 𝑎P > 0. Then
I = M∈S M𝑎M ⊂ P. on the other hand, 𝑏 P = 0 hence J = M∈S M𝑏M ⊄ P
Î Î
(choose an element 𝑥M ∈ M P for every M ∈ S such that 𝑎M > 0; their
product belongs to J P). This contradicts the assumption that J ⊂ I and
concludes the proof of the theorem. 

Theorem (9.6.9). — Let A be an integral domain. The following conditions are

equivalent:
(i) The ring A is a Dedekind ring;
(ii) The ring A is noetherian and for every maximal ideal M of A, the ring AM
is principal;
(iii) The ring A is noetherian, integrally closed and dim(A) 6 1.

Proof. — (i)⇒(ii). Assume that A is a Dedekind ring. We have already

seen that every fractional ideal, a fortiori, every ideal of A is finitely
generated; in particular, A is noetherian.
Let P be a maximal ideal of A. The local ring AP is a Dedekind ring;
by proposition 9.6.7, it is a principal ideal domain.
(ii)⇒(iii). By assumption, the ring A is noetherian and that AP is a
principal ideal domain, for every maximal ideal P of A. In particular,
 9.6. DEDEKIND RINGS 477

dim(AP ) 6 1, that is, htA (P) 6 1. Since P is arbitrary, this implies

dim(A) 6 1.
Let us prove that A is integrally closed. Let 𝑥 ∈ K be integral over A.
Let I be the set of 𝑎 ∈ A such that 𝑎𝑥 ∈ A; this is an ideal of A. Let P
be a maximal ideal of A. The element 𝑥 of K is integral over AP , hence
belongs to AP because a principal ideal domain is integrally closed. This
proves that there exists 𝑎 ∈ A P such that 𝑎𝑥 ∈ A; in other words, I ⊄ P.
Since P is arbitrary, it follows from Krull’s theorem (theorem 2.1.3) that
I = A. In particular, 𝑥 ∈ A.

(iii)⇒(i) Let A be an integral domain which is noetherian, integrally

closed and such that dim(A) 6 1. If dim(A) = 0, then A is a field, hence
a Dedekind ring. We now assume that A is not a field; then dim(A) = 1
and every non-zero prime ideal of A is maximal.
We first prove that every maximal ideal of A is invertible. Let M be a
maximal ideal of A; in particular, M ≠ 0. Then M−1 is a fractional ideal
of A; let us prove that M · M−1 = A.
One has M ⊂ A, hence A ⊂ M−1 . We first prove that M−1 ≠ A.
Let 𝑎 be a non-zero element of M. The ring A/𝑎A has dimension 0
and is noetherian. By Akizuki’s theorem (theorem 6.4.14), it has finite
length. Then there exists a finite sequence (M1 , . . . , M𝑛 ) of maximal
ideals of A such that M1 . . . M𝑛 ⊂ 𝑎A, and we may choose this sequence
so that 𝑛 is minimal. Then M1 . . . M𝑛 ⊂ 𝑎A ⊂ M. Since M is a
prime ideal, there exists 𝑖 ∈ {1, . . . , 𝑛} such that M𝑖 ⊂ M (otherwise,
choose elements 𝑎 𝑖 ∈ M𝑖 M; the product 𝑎1 . . . 𝑎 𝑛 does not belong
to M). Since M𝑖 is maximal, this implies M = M𝑖 . Up to renumbering
the M𝑖 , we thus assume that M = M1 . By the minimality assumption
on 𝑛, one has M2 . . . M𝑛 ⊄ 𝑎A. Let 𝑏 ∈ M2 . . . M𝑛 𝑎A. One has
𝑏M ∈ MM2 . . . M𝑛 ⊂ 𝑎A, hence (𝑏𝑎 −1 )M ⊂ A. By the definition of M−1 ,
this implies 𝑏𝑎 −1 ∈ M−1 . Since 𝑏𝑎 −1 ∉ A, we thus have shown that
M−1 ≠ A.
From the inclusion A ⊂ M−1 , we deduce that M ⊂ M · M−1 . Since M is
a maximal ideal of A, it follows that either M · M−1 = A, or M · M−1 = M.
Arguing by contradiction, let us assume that M · M−1 = M.
Let 𝑥 ∈ M−1 ; one has 𝑥M ⊂ M. Since A is noetherian, M is a finitely
generated A-module and theorem 4.1.5 implies that 𝑥 is integral over A.
478 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

Since A is integrally closed, this shows that 𝑥 ∈ A. Consequently,

M−1 ⊂ A, hence M−1 = A, a contradiction.
By theorem 9.6.8, the ring A is then a Dedekind ring. 
The following corollary will be a plethoric source of examples of
Dedekind rings. We refer the reader to exercise 9/16 for a stronger
result that does not require the separable hypothesis (theorem of Krull–
Akizuki).

Corollary (9.6.10). — Let A be a Dedekind ring, let K be its field of fractions.

Let L be a separable extension of K and let B be the integral closure of A in L.
Then B is a finitely generated A-module and a Dedekind ring.

Proof. — By theorem 9.6.9, the ring A is noetherian, integrally closed

and its dimension is at most 1. By proposition 9.5.1, B is a finitely
generated A-module and a noetherian ring. Since B is integral over A,
one has dim(B) = dim(A) 6 1 (corollary 9.2.9). Finally, B is integrally
closed in its field of fractions, by construction. Theorem 9.6.9 then
implies that B is a Dedekind ring. 

Example (9.6.11). — Let K be a number field, that is, a finite extension

of Q, and let A be the integral closure of Z in K: it is called the ring of
integers of K. Many exercises of this book consider particular cases of
this situation.
Let 𝑛 = [K : Q]. By corollary 9.6.10, A is a finitely generated Z-module
and a Dedekind ring. Since A ⊂ K, it is torsion free, hence is a free
Z-module. For every 𝛼 ∈ K, there exists an integer 𝑐 > 1 such that
𝑐𝛼 ∈ A, for example the leading coefficient of the minimal polynomial
of 𝛼. This implies that A contains a basis of K as a Q-vector space, hence
A ' Z𝑛 .
The study of these rings, in particular the discovery that they are
not necessarily unique factorization domains, is at the heart of the
developement of algebraic number theory since the xviiith century. By
corollary 9.6.10, A is a Dedekind ring, and an important result of
Minkowski asserts that its class group 𝒞(A) is finite. However, many
mysteries remain about these groups.
 9.6. DEDEKIND RINGS 479

Theorem (9.6.12) (Minkowski). — Let K be a finite extension of Q and let A

be its ring of integers of K. The class group 𝒞(A) of A is finite.

The proof that we give is due to Hurwitz; it follows the book of Ireland
& Rosen (1990).
Proof. — Let 𝑛 = [K : Q]. Let M be the integer provided by lemma 9.6.13
and let 𝒥 be the set of ideals of A that contain M!. The elements of 𝒥
are in bijection with the ideals of the quotient ring A/M!A. Since A is
isomorphic to Z𝑛 as an abelian group, this quotient ring is finite and the
set 𝒥 is finite. To prove that 𝒞(A) is finite, we will now prove that every
non-zero ideal of A has the same class that an element of 𝒥.
Let I be a non-zero ideal of A. Choose a non-zero element 𝑎 ∈ I
such that NK/Q (𝑎) is minimal (recall from corollary 4.7.6 that it is an
integer). Let 𝑏 ∈ I. By the following lemma there exists an integer M
(depending on K only), an integer 𝑢 ∈ {1, . . . , M} and an element 𝑐 ∈ A
such that NK/Q (𝑢𝑏 − 𝑎𝑐) < NK/Q (𝑎) . Since 𝑢𝑏 − 𝑎𝑐 ∈ A, the definition
of 𝑎 implies that 𝑢𝑏 − 𝑎𝑐 = 0. In particular 𝑢𝑏 ∈ 𝑎A, hence M! I ⊂ 𝑎A, so
that J = M!𝑎 −1 I is an ideal of A. By construction, I and J have the same
class in 𝒞(A). Since 𝑎 ∈ I, one has M! ∈ J, hence J ∈ 𝒥. This concludes
the proof. 

Lemma (9.6.13) (Hurwitz). — There exists an integer M such that the follow-
ing property holds: for every 𝑎, 𝑏 ∈ A with 𝑎 ≠ 0, there exists 𝑢 ∈ {1, . . . , M}
and 𝑐 ∈ A such that NK/Q (𝑢𝑏 − 𝑎𝑐) 6 NK/Q (𝑎) .

Proof. — Let 𝑛 = [K : Q] and let 𝛼1 , . . . , 𝛼 𝑛 be a basis of A as a Z-module;

it is a basis of K as a Q-vector space. Let 𝜎1 , . . . , 𝜎𝑛 : K → C be the
𝑛 distinct embeddings of K into C. Set
𝑛 𝑛
!
Ö Õ
C= 𝜎 𝑗 (𝛼 𝑖 )
𝑗=1 𝑖=1

and let 𝑚 = bC1/𝑛 c + 1.

For 𝑘 = (𝑘1 , . . . , 𝑘 𝑛 ) ∈ {1, . . . , 𝑚} 𝑛 , let P 𝑘 be the cube [(𝑘1 −
1)/𝑚, 𝑘1 /𝑚] × · · · × [(𝑘 𝑛 − 1)/𝑚, 𝑘 𝑛 /𝑚] of side 1/𝑚 in R𝑛 . These 𝑚 𝑛
cubes cover the unit cube P = [0, 1]𝑛 . For 𝑥 ∈ R𝑛 , we write
b𝑥c = (b𝑥 1 c, . . . , b𝑥 𝑛 c) ∈ Z𝑛 and {𝑥} = 𝑥 − b𝑥c ∈ P.
480 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

Let 𝑥 = (𝑥 1 , . . . , 𝑥 𝑛 ) ∈ Q𝑛 be such that 𝑏/𝑎 = 𝑛𝑗=1 𝑥 𝑗 𝛼 𝑗 . By the pigeon-

Í
hole principle, there exists 𝑢 ∈ {1, . . . , 𝑚} 𝑛 such that P𝑢 contains at least
two terms of the sequence (0, {𝑥}, . . . , {𝑚 𝑛 𝑥}) of length 𝑚 𝑛 + 1; let 𝑠 < 𝑡
be their indices and let 𝑢 = 𝑡 − 𝑠; observe that 1 6 𝑢 6 𝑚 𝑛 = M. Let
𝑦 = b𝑡𝑥c − b𝑠𝑥c and 𝑧 = {𝑡𝑥} − {𝑠𝑥}; by construction, one has 𝑦 ∈ Z𝑛
and the coordinates of 𝑧 satisfy 𝑧 𝑗 6 1/𝑚 for 𝑗 ∈ {1, . . . , 𝑛}. Moreover
𝑢𝑥 = 𝑡𝑥 − 𝑠𝑥 = 𝑦 + 𝑧.
Set 𝑐 = 𝑦1 𝛼1 + · · · + 𝑦𝑛 𝛼 𝑛 and 𝑑 = 𝑧1 𝛼1 + · · · + 𝑧 𝑛 𝛼 𝑛 ; one has 𝑐 ∈ A,
𝑟 ∈ K, and 𝑢𝑏 = 𝑎𝑐 + 𝑎𝑟.
Moreover,
𝑛 𝑛
!
Ö Õ
NK/Q (𝑟) = 𝑧 𝑖 𝜎 𝑗 (𝛼 𝑖 )
𝑗=1 𝑖=1
𝑛 𝑛
!
Ö Õ
6 sup(|𝑧1 | , . . . , |𝑧 𝑛 |)𝑛 𝜎 𝑗 (𝛼 𝑖 )
𝑗=1 𝑖=1

6 𝑚 −𝑛 C < 1,
so that
NK/Q (𝑢𝑏 − 𝑎𝑐) = NK/Q (𝑎𝑟) = NK/Q (𝑎) NK/Q (𝑟) < NK/Q (𝑎) .
This concludes the proof of the lemma. 

Exercises
1-exo649) Let 𝑘 be a field and let A be a finitely generated 𝑘-algebra. Assume that A is
a field.
a) Prove that there exists an integer 𝑛 > 0, elements 𝑥1 , . . . , 𝑥 𝑛 ∈ A which are
algebraically independent such that A is algebraic over 𝑘[𝑥1 , . . . , 𝑥 𝑛 ].
In the rest of the exercice, we will prove that 𝑛 = 0: A is algebraic over 𝑘 (this
furnishes a rather direct proof of theorem 9.1.2). We argue by contradiction, assuming
that 𝑛 > 1.
b) Prove that there exists 𝑓 ∈ 𝑘[𝑥 1 , . . . , 𝑥 𝑛 ] {0} such that A is integral over
𝑘[𝑥1 , . . . , 𝑥 𝑛 ][1/ 𝑓 ]. Conclude that 𝑘[𝑥 1 , . . . , 𝑥 𝑛 ][1/ 𝑓 ] is a field.
c) Prove that 𝑓 belongs to every maximal ideal of 𝑘[𝑥 1 , . . . , 𝑥 𝑛 ].
d) Observing that 1 + 𝑓 is a unit of 𝑘[𝑥1 , . . . , 𝑥 𝑛 ], deduce from this that 𝑓 ∈ 𝑘 and
derive a contradiction.
 EXERCISES 481

2-exo871) Let A be a noetherian integral domain which is not a field and let E be its
field of fractions.
Prove the equivalence of the following assertions:
a) The A-algebra E is finitely generated;
b) There exists a non-zero element of A which belongs to every non-zero prime ideal;
c) The ring A has only finitely many non-zero prime ideals of height 1;
d) One has dim(A) = 1 and the ring A has only finitely many maximal ideals.
(For c)⇒d), use Krull’s Hauptidealsatz. For b)⇒a), take an element 𝑎 ∈ A and consider a
primary decomposition of the ideal 𝑎A.)
3-exo841) Let 𝑗 : A → B be an injective morphism of integral domains; we assume
that B is finitely generated over A. The goal of the exercise is to prove that for every
non-zero 𝑏 ∈ B, there exists 𝑎 ∈ A such that for every algebraically closed field K and
every morphism 𝑓 : A → K such that 𝑓 (𝑎) ≠ 0, there exists a morphism 𝑔 : B → K
such that 𝑓 = 𝑔 ◦ 𝑗 and 𝑔(𝑏) ≠ 0.
a) Treat the case where B = A[𝑥], for some 𝑥 ∈ B which is transcendental over A.
b) Treat the case where B = A[𝑥] for some 𝑥 ∈ B which is algebraic over A.
c) Treat the general case.
d) Let V be a non-empty open subset of Spec(B); prove that its image 𝑗 ∗ (V) in Spec(A)
contains a non-empty open subset.
4-exo840) Recall that a Jacobson ring is a commutative ring in which every prime ideal
is the intersection of the maximal ideals that contain it (see p. 9.1).
a) Prove that Z is a Jacobson ring. What principal ideal domains are Jacobson rings?
b) Let A be a Jacobson ring and let I be an ideal of A. Prove that A/I is a Jacobson
ring.
Let A be a Jacobson ring and let B be an A-algebra.
c) If B is integral over A, then B is a Jacobson ring.
d) Assume that B is finitely generated over A. Prove that for every maximal ideal M
of B, the intersection M ∩ A is a maximal ideal of A, and the residue field B/M is a
finite extension of A/(M ∩ A). Prove that B is a Jacobson ring.
e) Let M be a maximal ideal of Z[T1 , . . . , T𝑛 ]. Prove that the residue field
Z[T1 , . . . , T𝑛 ]/M is a finite field.
f ) Let K be a field and let A be a K-algebra. Assume that there exist a set I such
that Card(I) < Card(K), and a family (𝑎 𝑖 )𝑖∈I such that A = K[(𝑎 𝑖 )]. Prove that for every
maximal ideal M of A, the residue field A/M is an algebraic extension of K, and that A
is a Jacobson ring. (Follow the arguments in the proof of theorem 2.3.1.)
5-exo1227) Let 𝑘 be an infinite field and let A be a 𝑘-algebra.
a) Let P ∈ 𝑘[T1 , . . . , T𝑛 ] be a non-zero homogeneous polynomial. Show that there
exists 𝑐 ∈ 𝑘 𝑛 such that P(𝑐) ≠ 0 and 𝑐 𝑛 = 1.
482 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

b) Let 𝑎 1 , . . . , 𝑎 𝑛 ∈ A and let P ∈ 𝑘[T1 , . . . , T𝑛 ] be a non-zero polynomial such that

P(𝑎 1 , . . . , 𝑎 𝑛 ) = 0. Let 𝑑 = deg(P) and let P𝑑 be the homogeneous component of
degree 𝑑 of P. Let 𝑐 ∈ 𝑘 𝑛 be such that P𝑑 (𝑐) ≠ 0 and 𝑐 𝑛 = 1; for 𝑚 ∈ {1, . . . , 𝑛 − 1}, let
𝑏 𝑚 = 𝑎 𝑚 + 𝑐 𝑚 𝑎 𝑛 . Prove that 𝑎1 , . . . , 𝑎 𝑛 are integral over 𝑘[𝑏 1 , . . . , 𝑏 𝑛−1 ].
c) Let 𝑎 1 , . . . , 𝑎 N ∈ A be elements such that A = 𝑘[𝑎1 , . . . , 𝑎 N ]. Prove that there exist
an integer 𝑛 > 0 and algebraically independent elements 𝑏1 , . . . , 𝑏 𝑛 which are 𝑘-linear
combinations of 𝑎1 , . . . , 𝑎 N such that A is integral over its subring 𝑘[𝑏 1 , . . . , 𝑏 𝑛 ]. (A
version of Noether’s normalization lemma.)
6-exo1226) Let 𝑘 be a field and let E, F be two finitely generated field extensions of 𝑘. Let
R = E ⊗ 𝑘 F. The goal of the exercise is to prove that dim(R) = inf(tr deg 𝑘 (E), tr deg 𝑘 (F)),
a formula due to Grothendieck.
a) Let 𝑛 = tr deg 𝑘 (E) and let (𝑎 1 , . . . , 𝑎 𝑛 ) be a transcendence basis of E; let E1 =
𝑘(𝑎 1 , . . . , 𝑎 𝑛 ) and R1 = E1 ⊗ 𝑘 F. Prove that R1 is a subring of R and that R is integral
over R1 . Conclude that dim(R) = dim(R1 ).
b) Show that A1 = 𝑘[𝑎 1 , . . . , 𝑎 𝑛 ] ⊗ 𝑘 F is a subring of R of which R1 is a fraction ring.
Prove that A1 is isomorphic to F[T1 , . . . , T𝑛 ] and that dim(R) 6 𝑛.
c) Assume, moreover, that tr deg 𝑘 (F) > 𝑛. Prove that there exists a surjective
morphism of F-algebras 𝜓 : R1 → F.
d) Using the fact that Ker(𝜓) ∩ A1 is a maximal ideal of A1 , prove that dim(R1 ) > 𝑛.
7-exo810) Let 𝑓 : A → B be a morphism of commutative rings.
a) Show that 𝑓 is an epimorphism in the category of rings if and only if the unique
morphism B ⊗A B → B such that 𝑏 ⊗ 𝑏 0 ↦→ 𝑏𝑏 0 for all 𝑏, 𝑏 0 ∈ B is an isomorphism.
b) Let M be a finitely generated non-zero A-module. Using a surjective morphism of
the form 𝑔 : M → A/I, prove that M ⊗A M ≠ 0.
c) Assume that 𝑓 is an epimorphism and that B is finitely generated as an A-module;
prove that 𝑓 is surjective. (First prove that (B/A) ⊗A B = 0 and that (B/A) ⊗A (B/A) = 0.)
8-exo880) Let A be a ring. For every 𝑎 ∈ A, let S𝑎 be the set of elements of A of the
form 𝑎 𝑛 (1 + 𝑎𝑏), for 𝑛 ∈ N and 𝑏 ∈ A.
a) Let 𝑎 ∈ A. Prove that S𝑎 is a multiplicative subset of A. Prove that S𝑎 contains 0 if
and only if 𝑎 is invertible or 𝑎 is nilpotent.
b) Let 𝑎 ∈ A. Prove that M ∩ S𝑎 ≠ ∅ for every maximal ideal M of A. Conclude that
dim(S−1
𝑎 A) + 1 6 dim(A).
c) Let M be a maximal ideal of A, let P be a prime ideal of A such that P ⊂ M. Prove
that P ∩ S𝑎 = ∅ for every 𝑎 ∈ M P. Conclude that there exists 𝑎 ∈ A such that
dim(S−1
𝑎 A) + 1 = dim(A).
d) Let 𝑛 ∈ N. Prove that dim(A) 6 𝑛 if and only if, for every 𝑎 0 , . . . , 𝑎 𝑛 ∈ A, there
exist 𝑏 0 , . . . , 𝑏 𝑛 ∈ A and 𝑚0 , . . . , 𝑚𝑛 ∈ N such that

𝑎0𝑛0 (𝑎 0 𝑏 0 + 𝑎 1𝑛1 (𝑎 1 𝑏 1 + · · · + 𝑎 𝑛𝑚𝑛 (1 + 𝑎 𝑛 𝑏 𝑛 ))) = 0.

 EXERCISES 483

e) Let K be a field. Prove that dim(K[T1 , . . . , T𝑛 ]) = 𝑛. (This characterization of Krull’s

dimension is due to Coquand & Lombardi (2005).)
9-exo843) Let A be a ring and let B = A[X1 , . . . , X𝑛 ].
a) Prove that dim(B) > dim(A) + 𝑛.
b) Let P be a prime ideal of A, let (Q0 , . . . , Q𝑚 ) be a chain of prime ideals of B such
that Q 𝑗 ∩ A = P for every 𝑗. Prove that 𝑚 > 𝑛. (Passing to the quotient by P and Q0 , reduce
to the case where P = 0; by localization, then reduce to the case where A is a field.)
c) Prove that dim(B) 6 dim(A) + 𝑛(1 + dim(A)).
In the case 𝑛 = 1, i.e., B = A[X], this exercise says that dim(A) + 1 6 dim(A[X]) 6
2 dim(A) = 1, and Seidenberg (1954) has shown that all possibilities actually appear. When
A is noetherian, exercise 9/14 shows that one has dim(B) = dim(A) + 𝑛.
10-exo847) Let A be a noetherian ring and let I be an ideal of A. Let B be the subalgebra
of A[T] consisting of polynomials P = 𝑎 𝑛 T𝑛 such that 𝑎 𝑛 ∈ I𝑛 for all 𝑛.
Í

a) Using exercise 6/21, prove that B is a noetherian ring.

b) Assume that there exists a prime ideal P of A containing I such that dim(A/P) =
dim(A); prove that dim(B) = dim(A) + 1.
c) Otherwise, prove that dim(B) = dim(A).
11-exo848) Let X be a topological space and let A = 𝒞(X; R) be the ring of real valued
continuous functions on X.
a) Assume that every function 𝑓 ∈ A is locally constant. Prove that A is a von
Neumann ring and that dim(A) = 0.
Let 𝜌 ∈ A be a continuous function which is not locally constant in any neighborhood
of point 𝜉 ∈ X; we assume 𝜌(𝜉) = 0.
b) Let P be the set of functions 𝑓 ∈ A such that
lim 𝑓 (𝑥)/𝜌(𝑥)𝑡 = 0
𝑥→𝜉

for every 𝑡 ∈ N. Prove that P is an ideal of A, contained in the maximal ideal M𝜉 of

functions vanishing at 𝜉. Prove that P is a radical ideal and that P ≠ M𝜉 .
c) Let 𝔘 be an ultrafilter (see exercise 2/11) which converges to 𝜉, in the sense that all
neighborhoods of 𝜉 belong to 𝔘. For every 𝑐 ∈ R+ , let P𝑐 be the set of functions 𝑓 ∈ A
such that
lim 𝑓 (𝑥) exp(𝑡𝜌(𝑥)−𝑐 ) = 0
𝔘
for all 𝑡 ∈ R.
d) Prove that P𝑐 is a prime idal of A.
e) Let 𝑐, 𝑐 0 ∈ R+ be such that 𝑐 < 𝑐 0. Prove that P𝑐0 ( P𝑐 .
f ) Prove that dim(A) = +∞.
12-exo805) Let K be a field and let A = K[T1 , T2 , . . . ] be the ring of polynomials in
(countably) infinitely many indeterminates.
a) Prove that A is not noetherian and that dim(A) is infinite.
484 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

b) Let (𝑚𝑛 )𝑛>1 be a strictly increasing sequence of strictly positive integers that that
the sequence (𝑚𝑛+1 − 𝑚𝑛 ) is unbounded. For every 𝑛 > 1, let P𝑛 be the ideal of A
generated by the indeterminates T𝑖 , for 𝑚𝑛 6 𝑖 < 𝑚𝑛+1 . Prove that P𝑛 is a prime ideal
of A.
c) Let S be the intersection of the multiplicative subsets S𝑛 = A P𝑛 , for all 𝑛 > 1.
Prove that dim(S−1 A) is infinite.
d) Prove that S−1 A is a noetherian ring.
13-exo806) Let A be a local noetherian ring and let M be its maximal ideal, and let
K = A/M be its residue field .
a) Explain why the A-module structure of M/M2 endows it with the structure of a
K-vector space.
b) Let 𝑎 1 , . . . , 𝑎 𝑛 be elements of M which generate M/M2 . Prove that M = (𝑎 1 , . . . , 𝑎 𝑛 ).
Deduce from this that dim(A) 6 dimK (M/M2 ).
In the rest of the exercise, one assumes that dim(A) = dimK (M/M2 ) (one says that A
is a regular ring) and the goal is to prove that A is an integral domain. We argue by
induction on dim(A).
c) Treat the case dim(A) = 0.
d) Assume that dim(A) > 0. Let P1 , . . . , P𝑛 be the minimal prime ideals of A. Prove
that there exists 𝑎 ∈ M such that 𝑎 ∉ M2 ∪ 𝑛𝑖=1 P𝑖 .
Ð

e) Let B = A/(𝑎) and let N = MB be its maximal ideal. Using the induction hypothesis,
prove that (𝑎) is a prime ideal of A.
f ) Let 𝑖 ∈ {1, . . . , 𝑛} be such that P𝑖 ⊂ (𝑎). Using Nakayama’s lemma, prove that
P𝑖 = (0). Conclude.
14-exo870) Let A be a noetherian ring and let B = A[X].
a) Let P be a prime ideal of A, let Q and Q0 be two prime ideals of B such that Q ( Q0
and P = A ∩ A = Q0 ∩ B. Prove that Q = PB. (Argue as in question b) of exercise 9/9.)
b) Let I be an ideal of A and let P be a minimal prime ideal containing I. Prove that
PB is a minimal prime ideal of B containing IB.
c) Let P be a prime ideal of A. Prove that htA (P) = htB (PB).
d) Prove that dim(B) = dim(A) + 1.
e) More generally, prove that dim(A[X1 , . . . , X𝑛 ]) = dim(A) + 𝑛.
15-exo830) Let A be a noetherian integral domain such that dim(A) = 1. Let K be its
fraction field. Let M be a finitely generated A-module.
a) Let S = A {0}. Prove that S−1 M is a finite dimensional K-vector space. Let 𝑟 be
its dimension. Prove that 𝑟 is the maximal number of A-linearly independent subsets
of M.
b) Prove the equivalence of the following assertions: (i) Every element of M is torsion;
(ii) ℓA (M) is finite; (iii) 𝑟 = 0.
 EXERCISES 485

c) From now on, we assume that M is torsion-free. Show that M contains a free
submodule L of rank 𝑟 such that ℓA (M/L) is finite.
d) Prove that for every integer 𝑛 > 1, one has
ℓA (M/𝑎 𝑛 M) 6 ℓA (L/𝑎 𝑛 L) + ℓA (M/L).
Also prove that
ℓA (M/𝑎 𝑛 M) = 𝑛 ℓA (M/𝑎M).
e) Prove that ℓA (M/𝑎M) 6 𝑟 ℓA (A/𝑎A).
f ) Generalize the previous inequality to any torsion-free A-module. Give an example
where this inequality is strict.
16-exo831) Let A be a integral domain, let K be its field of fractions, let L be finite
extension of K and let B be a subring of L containing A.
a) Let I be a non-zero ideal of B. Prove that I ∩ A ≠ 0.
In the sequel, one assumes that A is noetherian and dim(A) = 1.
b) Let I be an ideal of B and let 𝑎 be a non-zero element of I ∩ A. Prove that B/𝑎B
and I/𝑎B are finite length B-modules. (Use exercise 9/15.)
c) Prove that B is noetherian and that dim(B) 6 1. (Theorem of Krull–Akizuki.)
17-exo819) Let A be a noetherian integral domain such that every non-zero prime ideal
of A is maximal.
a) Let 𝑎 ∈ A {0}. Prove that the A-module A/𝑎A has finite length.
b) Let 𝑎, 𝑏 ∈ A {0}. Prove that ℓ A (A/(𝑎𝑏)) = ℓ A (A/(𝑎)) + ℓA (A/(𝑏)).
c) Prove that there exists a unique morphism of groups 𝜔A : K× → Z such that
𝜔A (𝑎) = ℓA (A/(𝑎)) for every 𝑎 ∈ A {0}.
d) Assume that A is a principal ideal domain. Compute 𝜔A (𝑎) in terms of the
decomposition of 𝑎 as a product of irreducible elements.
18-exo820) Let A be a ring, let 𝜑 : M → N be a morphism of A-modules. One says
that 𝜑 is admissible if Ker(𝜑) and Coker(𝜑) have finite length; one then defines
𝑒A (𝜑) = ℓA (Coker(𝜑)) − ℓA (Ker(𝜑)).
a) If M and N have finite length, then 𝜑 is admissible and 𝑒A (𝜑) = ℓA (N) − ℓA (M).
b) Let 𝜑 : M → N and 𝜓 : N → P be morphisms of A-modules. If two among 𝜑, 𝜓,
𝜓 ◦ 𝜑 are admissible, then so is the third and
𝑒A (𝜓 ◦ 𝜑) = 𝑒A (𝜓) + 𝑒A (𝜑).
c) One assumes that A is an integral domain and that every non-zero prime ideal
of A is maximal. Let M be a free and finitely generated A-module and let 𝜑 be an
injective endomorphism of M. Show that 𝜑 is admissible and that ℓA (Coker(𝜑)) =
ℓA (A/(det(𝜑))). (First treat the case where the matrix of 𝜑 in some basis of M is an elementary
matrix.)
d) In the case where A is a principal ideal domain, deduce the result of the previous
question from theorem 5.4.3.
486 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

19-exo809) Let A be an integral domain and let K be its field of fractions.

a) Assume that for every 𝑎 ∈ A {0} and every prime ideal P which is associated
with A/(𝑎), the fraction ring AP is a principal ideal domain. Prove that A is integrally
closed.
Conversely, one assumes that A is noetherian and integrally closed. Let 𝑎 ∈ A {0}
and let P be a prime ideal of A which is associated with A/(𝑎).
b) Prove that there exists 𝑏 ∈ A such that P = {𝑥 ∈ A ; 𝑥𝑏 ∈ (𝑎)}.
c) Prove that the element 𝑎/𝑏 of K belongs to AP and that PAP = (𝑎/𝑏)AP . (If
(𝑏/𝑎)PAP ≠ AP , prove that 𝑏/𝑎 is integral over A.)
d) Prove that AP is a principal ideal domain and that every prime ideal associated
with A/(𝑎) is minimal.
20-exo877) Let 𝑘 be a field and let A = 𝑘[X, Y] be the ring of polynomials in two
indeterminates.
a) Let I = (X, Y). Prove that I−1 = A. Conclude that I is not invertible.
b) Prove that a fractional ideal of A is invertible if and only if it is principal.
21-exo876) Let A be a Dedekind ring.
a) Assume that Spec(A) is finite. Prove that A is a principal ideal domain.
b) Let I be an ideal of A and let 𝑎 ∈ I {0}. Prove that there exists 𝑏 ∈ I such that
I = 𝑎A + 𝑏A. (Start by proving that the ring A/I is artinian.)
22-exo832) Let A be a principal ideal domain which is a local ring. Let 𝑝 be a generator
of its maximal ideal and let K be its field of fractions. Let P ∈ A[T] be an Eisenstein
polynomial (see exercise 2/27 where it is proved that P is irreducible in K[T]), that is,
P = T𝑛 + 𝑎 𝑛−1 T𝑛−1 + · · · + 𝑎 0 , where 𝑎 0 , . . . , 𝑎 𝑛−1 ∈ (𝑝) and 𝑎 𝑛 ∉ (𝑝 2 ).
Let L = K[T]/(P), let 𝜔 be the class of T in L, and let B = A[𝜔].
a) Prove that the ideal (𝜔) of B is maximal.
b) Prove that (𝜔) is the unique maximal ideal of B. (Use Nakayama’s lemma to prove
that every maximal ideal of B contains 𝑝.)
c) Prove that for every 𝑏 ∈ B {0}, there exists a unit 𝑢 ∈ B× and an integer 𝑛 ∈ N
such that 𝑏 = 𝑢𝜔 𝑛 .
d) Prove that B is a principal ideal domain and is the integral closure of A in L.
23-exo833) Let A be a local ring, let K be its field of fractions, let M be its maximal ideal,
and let 𝑘 = A/M be its residue field. Let P ∈ A[T] be a monic irreducible polynomial
whose image P ∈ 𝑘[T] is separable; let P = 𝑚
Î
𝑗=1 P 𝑗 be the decomposition as a product
of monic irreducible polynomials; for every 𝑗, fix a monic polynomial P 𝑗 ∈ A[T] with
reduction P 𝑗 modulo M. Let L = K[T]/(P) and let B = A[T]/(P).
a) Let N be a maximal ideal of B. Prove that there exists a unique element 𝑗 ∈
{1, . . . , 𝑚} such that N = MB + (P 𝑗 ).
 EXERCISES 487

b) Assume that A is integrally closed in its field of fractions. Prove that B is the
integral closure of A in L. (Using the separability of the polynomial P, prove that the
determinant of the 𝑛 × 𝑛 matrix (TrL/K (𝜔 𝑖+𝑗 ))06𝑖,𝑗<𝑛 , the discriminant of P, is a unit of A.)
c) Assume that A is a principal ideal domain. Prove that B is a principal ideal domain.
(Deduce from theorem 9.6.9that for every 𝑗, the local ring BN 𝑗 is a principal ideal domain.)
24-exo878) Let 𝑘 be a field and let A = 𝑘[X2 , X3 ] be the subalgebra of 𝑘[X] generated
by X2 and X3 .
a) Prove that A is a noetherian integral domain and that dim(A) = 1.
b) Prove that A is not integrally closed. (Prove that X is integral over A.)
c) Prove that the ideal (X2 , X3 ) of A is not invertible.
25-exo891) Let A = R[X, Y]/(X2 + Y2 − 1).
a) Prove that A is a Dedekind ring.
b) Let M be a maximal ideal of A. Prove that A/M ' R or A/M ' C. Give examples
of both cases.
c) Let M be a maximal ideal of A such that A/M ' C. Prove that M is principal.
d) Let M be a maximal ideal of A such that A/M ' R; prove that there exists a
unique pair (𝑎, 𝑏) ∈ R2 such that 𝑎 2 + 𝑏 2 = 1 and A = (X − 𝑎, Y − 𝑏). Prove that M is not
principal but that M2 is principal. Prove also that all of these maximal ideals have the
same class in the class group 𝒞(A) of A.
e) What is the class group of A?
√
26-exo893) Let 𝑑 be an integer such that 𝑑 > 2 and 𝑑 is squarefree, let K = Q(𝑖 𝑑) and
let A be the integral closure of Z in K; it is a Dedekind ring.
√ √
a) Prove that A = Z[(−1 + 𝑖 𝑑)/2] is 𝑑 ≡ −1 (mod 4), and that A = Z[𝑖 𝑑] otherwise.
b) Let I be a non-zero ideal of A. Let (𝑒, 𝑓 ) be a basis of I as a Z-module. Applying
exercise 5/11 to the quadratic form (𝑥, 𝑦) ↦→ NK/Q (𝑥𝑒 + 𝑦 𝑓 ), prove that there exists
p
a non-zero element 𝑎 ∈ I such that NK/Q (𝑎) 6 Card(A/I) 𝛿/3, where 𝛿 = 𝑑 if 𝑑 ≡ −1
(mod 4), and 𝛿 = 4𝑑 otherwise.
c) (followed)pLet J = (𝑎) : I. Prove that J is non-zero ideal of A such that IJ = (𝑎) and
Card(A/J) 6 𝛿/3.

p d) Prove that 𝒞(A) is generated by the maximal ideals P of A such that Card(A/P) 6
𝛿/3. This bound is better than the one implicitly given in the proof of theorem 9.6.12; it is
also slightly better than the general Minkowski bound.
√
27-exo892) Let A = Z[𝑖 5] (see exercise 2/24).
a) Prove that A is a Dedekind ring.
√
b) Prove that M = (2, 1 + 𝑖 5) is the unique maximal ideal of A containing 2. Prove
that M2 = (2) and that M is not principal.
√ √
c) Prove that the ideals P = (3, 1 + 𝑖 5) and P0 = (3, 1 − 𝑖 5) are the unique maximal
ideals of A containing 3. Prove that P · P0 = (3) and that neither P nor P0 is principal.
488 CHAPTER 9. ALGEBRAS OF FINITE TYPE OVER A FIELD

d) Prove that the classes of P, P0 , M in the class group 𝒞(A) are equal.
e) Using exercise 9/26, prove that 𝒞(A) ' Z/2Z.
28-exo929) Let 𝑘 be a finite field, let K = 𝑘(T) be the field of rational functions in one
indeterminate and let L be a finite separable extension of K. Let A = 𝑘[T] and let B be
the integral closure of A in L.
a) Show that B is a Dedekind ring and a free A-module of rank 𝑛, where 𝑛 = [L : K].
b) Let 𝑎 ∈ 𝑘(T); prove that there exists a unique pair (𝑏, 𝑐) where 𝑏 ∈ 𝑘[T] and
𝑐 ∈ 𝑘(T) are such that 𝑎 = 𝑏 + 𝑐 and deg(𝑐) < 0. (The degree deg(𝑎) of an element
𝑎 ∈ 𝑘(T) is defined as deg( 𝑓 ) − deg(𝑔), where 𝑓 , 𝑔 ∈ 𝑘[T] are polynomials such that
𝑔 ≠ 0 and 𝑎 = 𝑓 /𝑔.)
c) Let (𝑒1 , . . . , 𝑒 𝑛 ) be a basis of L as an K-vector space. Prove that there exists an
integer M such that deg(NL/K ( 𝑎 𝑖 𝑒 𝑖 ) 6 M + 𝑛 sup𝑖 deg(𝑎 𝑖 ) for every (𝑎 1 , . . . , 𝑎 𝑛 ) ∈ K𝑛 .
Í

d) Prove that there exists an integer M such that the following property holds: for
every 𝑎, 𝑏 ∈ B with 𝑎 ≠ 0, there exists a polynomial 𝑓 ∈ A such that deg( 𝑓 ) 6 M and
𝑔 ∈ B such that deg(NL/K ( 𝑓 𝑏 − 𝑎𝑐)) 6 deg(NL/K (𝑎)).
e) Prove that the class group 𝒞(B) is finite. (Adapt the proof of theorem 9.6.12, using the
preceding question to replace Hurwitz’s lemma.)
29-exo897) Let A be a Dedekind ring. Prove that every divisible A-module is injective.
(Conversely, it follows from exercise 7/24 that if A is a ring such that every divisible right
A-module is injective, then A is right hereditary. If A is an integral domain, it is then a
Dedekind ring.)
30-exo899) Let A be a Dedekind ring and let M be a finitely generated A-module.
a) If M is torsion-free, then M is projective.
b) Prove that the torsion submodule T(M) of M has a direct summand.
c) Let I, J be fractional ideals of A and let 𝑓 : J → I be a morphism of A-modules.
Prove that there exists a unique element 𝑎 ∈ A such that 𝑓 (𝑥) = 𝑎𝑥 for every 𝑥 ∈ J.
Prove that 𝑎 ∈ IJ−1 .
d) Let I1 , . . . , I𝑚 and J1 , . . . , J𝑛 be fractional ideals of A, let M = I1 ⊕ · · · ⊕ I𝑚 and
N = J1 ⊕ · · · ⊕ J𝑛 , and let 𝑓 : N → M be an isomorphism of A-modules. Prove that
𝑚 = 𝑛 and I1 . . . I𝑚 ' J1 . . . J𝑛 . (Represent 𝑓 by a matrix U = (𝑎 𝑖𝑗 ) ∈ M𝑚,𝑛 (K), where
𝑎 𝑖𝑗 ∈ I𝑖 J−1
𝑗 . Prove that I = (det(U))J.)
e) Let I, J be fractional ideals of A. Prove that there exist non-zero elements 𝑎 ∈ I and
𝑏 ∈ J such that (I ⊕ J)/(𝑎, 𝑏)A is torsion-free. Conclude that I ⊕ J ' A ⊕ IJ.
f ) Let M be a finitely generated A-module. Assume that M is torsion free and
non-zero. Prove that there exists an integer 𝑛 > 1 and a fractional ideal I such that
M ' A𝑛−1 ⊕ I.
APPENDIX

This appendix consists in three different sections.

The first one, called algebra, collects a few definitions and facts which are
essentially prerequisites for the reading of this book. Their role is also to precise
a few conventions on which there is no unanimity, such as the precise meaning
of a “positive” number or an “increasing” sequence.
The second section is devoted to results of set theory which are used at a few
places of the book to manage infinite cardinals. The Cantor-Bernstein theorem
will imply that the dimension of a vector space, or the transcendence degree
of a field extensions, are well defined. On the other hand, Zorn’s theorem
asserts the existence of maximal elements in adequate (“inductive”) ordered
sets; it is used to prove the existence of bases of a vector space, of maximal
ideals, of minimal prime ideals, characterize injective modules, etc. The proof of
Kaplansky’s theorem, however, relies on a transfinite inductive construction,
and I couldn’t “zornify” it fully. For this reason, rather than giving a direct (but
maybe unnatural proof) of Zorn’s theorem, I chose to start from the inductive
principle for well-ordered sets, and Hartogs’s lemma. To make this appendix
reasonably complete, I also give a proof of the theorems of Cantor and Zermelo.
Finally, I give a brief introduction to category theory. Mathematical practice
progressively showed the importance of considering not only mathematical
objects but morphisms between them, and a category is the abstraction of this
idea. It then appeared that categories can themselves be treated as mathematical
objects, their morphisms are called functors, and that this point of view is
extremly fruitful in many fields of mathematics. While categories are only used
as a language in the first chapters of this book, they really are a useful tool in the
chapters devoted to homological algebra and tensor products.
490 APPENDIX

A.1. Algebra
A.1.1. Numbers. —
A.1.1.1. — The notation N, Z, Q, R and C are respectively used
for the sets positive integers (0, 1, 2, . . . ), the rational integers
(0, 1, 2, . . . , −1, −2, −3, . . . ), the rational numbers (fractions of rational
integers), the real numbers and the complex numbers.
A.1.1.2. — Between real numbers, the ordering relation 6 is pronounced
as “less than”, or sometimes “less than or equal to” if we feel it necessary
to insist on the possibility of equality. The expression “𝑥 < 𝑦” reads “𝑥 is
strictly smaller than 𝑦”. A similar convention for the opposite ordering
relation > (“greater than”, or “greater than or equal to”), and to the
relation > (“strictly greater than”).
In particular, a real number 𝑥 is said to be positive if 𝑥 > 0, strictly
positive if 𝑥 > 0, negative if 𝑥 6 0, and strictly negative if 𝑥 < 0.
A.1.1.3. Integers. — An integer 𝑛 is said to be a prime number if it is
strictly greater than 1 and if its only positive divisors are 1 and itself.
The first prime numbers are 2, 3, 5, 7, 11, . . .
A.1.1.4. Binomial coefficients. — Let 𝑛 be a positive integer. One sets 𝑛!
(“factorial 𝑛”) to be the product
𝑛! = 1 · 2 . . . 𝑛.
One has 0! = 1.
Let 𝑚, 𝑛 be positive integers such that 𝑚 6 𝑛. The binomial coefficient
𝑛

𝑚 is defined by
𝑛 𝑛!
 
= .
𝑚 𝑚!(𝑛 − 𝑚)!
It
is always an integer. This follows by induction from the equalities
𝑛
0 = 1 and from the recurrence relation

𝑛+1 𝑛 𝑛
     
= + ,
𝑚+1 𝑚 𝑚+1
whenever 0 6 𝑚 6 𝑛.

A.1.2. Sets, subsets, maps. —

 A.1. ALGEBRA 491

A.1.2.1. — As usual, I write 𝑥 ∈ S to mean that 𝑥 is an element of the

set S.
If S and T are two sets, I write S ⊂ T or T ⊃ S to mean the equivalent
formulations S is a subset of T, S is contained in T, or T contains S, that
is: every element of S belongs to T. The symbols S ( T mean that S ⊂ T
but S ≠ T.
The set of all subsets of a set S is denoted by 𝔓(S).
A.1.2.2. — A map 𝑓 : S → T determines, for every element 𝑠 ∈ T, an
element 𝑓 (𝑠) ∈ T; the element 𝑓 (𝑠) of T is called the image of 𝑠 by 𝑓 . The
map 𝑓 has a graph Γ 𝑓 which is the subset of S × T consisting in all pairs
of the form (𝑠, 𝑓 (𝑠)), for 𝑠 ∈ S.
The identity map, 𝑖𝑑S : S → S, is such that every element of S is its
own image.
A.1.2.3. — If A is a subset of S, one writes 𝑓 (A) for the set of all elements
𝑓 (𝑎), for 𝑎 ∈ A.
If B is a subset of T, one writes 𝑓 −1 (T) for the set of all elements 𝑠 ∈ S
such that 𝑓 (𝑠) ∈ T.
A.1.2.4. — One says that the map 𝑓 is injective if distinct elements of S
have distinct images. One says that the map 𝑓 is bijective if every element
of T is the image of some element of S.
One says that the map 𝑓 is bijective if it is both injective and surjective.
Then there exists a unique map 𝑔 : T → S such that 𝑔 ◦ 𝑓 = idS and
𝑓 ◦ 𝑔 = idT ; the map 𝑔 is called the inverse of 𝑓 and is often denoted
by 𝑓 −1 .
Two sets S and T are said to be equipotent if there exists a bijection
from S to T.

A.1.3. Ordered sets. —

A.1.3.1. — Let us recall that an order ≺ on a set S is a relation satisfying
the following axioms:

– The assertions 𝑥 ≺ 𝑦 and 𝑦 ≺ 𝑥 are incompatible;

– If 𝑥 ≺ 𝑦 and 𝑦 ≺ 𝑧, then 𝑥 ≺ 𝑧.
492 APPENDIX

An ordered set is a set together with an order on it. In an ordered set,

one writes 𝑥  𝑦 as a shorthand to 𝑥 ≺ 𝑦 or 𝑥 = 𝑦; one also defines
𝑥  𝑦 and 𝑥  𝑦 as synonyms for 𝑦 ≺ 𝑥 and 𝑦  𝑥.
A.1.3.2. — Let S be an ordered set and (𝑎 𝑛 )𝑛∈N be a sequence of elements
of S. One says that it is increasing if 𝑎 𝑛+1  𝑎 𝑛 for every 𝑛 ∈ N, and that
it is decreasing if 𝑎 𝑛+1  𝑎 𝑛 for every 𝑛 ∈ N.
The usual US terminology would be “non-decreasing” and “non-
increasing”. However, this terminology is terribly misleading, since
the two sentences “the sequence is non-decreasing” and “the sequence
is not decreasing” have totally different meanings: while the first one
means that 𝑎 𝑛+1  𝑎 𝑛 for every 𝑛 ∈ N, the second one holds if and only
if there exists 𝑛 ∈ N such that 𝑎 𝑛+1  𝑎 𝑛 is false.
If one wants to insist that 𝑎 𝑛+1  𝑎 𝑛 for every 𝑛, one says that it is
strictly increasing. One defines similarly strictly decreasing sequences.
Finally, a sequence (𝑎 𝑛 ) is stationary if there exists 𝑚 ∈ N such that
𝑎 𝑛 = 𝑎 𝑚 for every integer 𝑛 > 𝑚.
A.1.3.3. — Let S be an ordered set. If, for every pair (𝑥, 𝑦) of elements
of S, one has 𝑥 ≺ 𝑦, or 𝑦 ≺ 𝑥, or 𝑥 = 𝑦, one says that S is totally ordererd.
An initial segment of an ordered set S is a subset I such that for
every 𝑥 ∈ I and any 𝑦 ∈ S such that 𝑦 ≺ 𝑥, then 𝑦 ∈ I.
A.1.3.4. — Let A be a subset of a totally ordered set S.
An upper bound of A is an element 𝑢 ∈ S such that 𝑎  𝑢 for every
𝑎 ∈ A; if A has an upper bound, one says that A is bounded above. A
maximal element of A is an element 𝑎 of A such that there is no 𝑥 ∈ A such
that 𝑥  𝑎. which is an upper bound of A. Observe that A may have an
upper bound but no maximal element, and that a maximal element of A
may not be an upper bound of A, unless A is totally ordered. A largest
element of A is an element 𝑎 of A such that 𝑥  𝑎 for every 𝑥 ∈ A.
A lower bound of A is an element 𝑙 ∈ S such that 𝑙  𝑎 for every 𝑎 ∈ A;
if A has a lower bound, one says that it is bounded below. A minimal
element of A is an element 𝑎 of A such that there is no 𝑥 ∈ A such that
𝑥 ≺ 𝑎. A smallest element of A is an element 𝑎 of A such that 𝑎  𝑥 for
every 𝑥 ∈ A.
 A.2. SET THEORY 493

A least upper bound is a minimal element of the set of all upper bounds
for A. A largest lower bound is a maximal element of the set of all lower
bounds for A.

A.2. Set Theory

Theorem (A.2.1) (Cantor-Bernstein theorem). — Let A and B be sets. As-
sume that there exists an injection from A into B, as well as an injection from B
into A. Then, the sets A and B are equipotent.

Proof. — Let 𝑓 0 : 𝑓 (A) → A be the map such that for any 𝑏 ∈ 𝑓 (A), 𝑓 0(𝑏)
is the unique preimage of 𝑏 by 𝑓 , so that 𝑓 ◦ 𝑓 0 = id 𝑓 (A) and 𝑓 0 ◦ 𝑓 = idA .
Define similarly 𝑔 0 : 𝑔(B) → B. The idea of the proof consists in iterating
successively 𝑓 0 and 𝑔 0 as much as possible. Let A𝑒 be the set of elements
of A of the form 𝑔 𝑓 𝑔 𝑓 . . . 𝑓 (𝑏), where 𝑏 ∈ B 𝑓 (A); let A𝑜 be the set of
elements of A of the form 𝑔 𝑓 𝑔 𝑓 𝑔 𝑓 . . . 𝑓 𝑔(𝑎), where 𝑎 ∈ A 𝑔(B). The
set A𝑒 consists on all elements of A where you can apply an even number
of times 𝑓 0 and 𝑔 0, so that one ends up by 𝑔 0; the set A𝑜 consists on all
elements of A where you can apply an odd number of times 𝑓 0 and 𝑔 0,
ending up by 𝑓 0. Let A∞ be the complementary subset of A𝑜 ∪ A𝑒 in A.
Define analogously B𝑒 , B𝑜 and B∞ .
By construction, the maps 𝑓 induces bijections from A𝑒 to B𝑜 and
from A∞ to B∞ . The map 𝑔 induces a bijection from B𝑒 to A𝑜 .
Let ℎ : A → B be the map that coincides with 𝑓 on A𝑒 ∪ A∞ and with 𝑔 0
on A𝑜 . It is a bijection. This concludes the proof of the theorem. 

A.2.2. Cardinals. — Set theory defines the cardinal Card(A) of an arbi-

trary set in such a way that Card(A) = Card(B) if and only if A and B are
equipotent. (The precise construction often uses von Neumann ordinals
but is irrelevant in this book.)
The Cantor–Bernstein theorem implies that one defines an ordering
relation on cardinals by writing Card(A) 6 Card(B) if there exists an
injection from A to B.
Cantor’s theorem (corollary A.2.16 below) will imply that this ordering
relation is total.
494 APPENDIX

A.2.3. Well-orderings. — Let S be an ordered set. One says that S

is well-ordered (or that the given order on S is a well-ordering) if every
non-empty subset of S possesses a smallest element.
Let S be a non-empty well-ordered set. Applying the definition to S
itself, we see that S has a smallest element. In particular, S has a lower
bound.
Applying the definition to a pair (𝑥, 𝑦) of elements of S, we obtain that
S is totally ordered: the smallest element of {𝑥, 𝑦} is smaller than the
other.
The set of all natural integers with the usual order is well-ordered. On
the contrary, the set of all real numbers which are positive or zero is not
well-ordered: the set of all real numbers 𝑥 > 1 has no smallest element.
A subset of a well-ordered set is itself well-ordered.
Properties depending of an element of a well-ordered set can be proved
by induction, in a similar way to the induction principle for properties
depending of an integer.

Lemma (A.2.4) (Transfinite induction principle). — Let S be a well-ordered

set and let A be a subset of S. Assume that for every 𝑎 ∈ S such that A contains
{𝑥 ∈ S ; 𝑥 ≺ 𝑎}, one has 𝑎 ∈ A (“induction hypothesis”). Then A = S.

Proof. — Arguing by contradiction, assume that A ≠ S and let B = S A.

Then B is a non-empty subset of S; let 𝑎 be the smallest element of B.
By assumption, A contains every element 𝑥 ∈ S such that 𝑥 ≺ 𝑎; by the
induction hypothesis, one has 𝑎 ∈ A, but this contradicts the definition
of 𝑎. Consequently, A = S. 

A.2.5. — The induction principle can be also reformulated by consider-

ing the nature of the elements of a well-ordered set S. If S has a largest
element, we set S∗ = S {sup(S)}; otherwise, we set S∗ = S.
Let 𝑎 ∈ S∗ ; since 𝑎 is not the largest element of S, the set {𝑥 ∈ S ; 𝑥  𝑎}
is non-empty; its smallest element is called the successor of 𝑎; let us
denote it by 𝜎(𝑎). One has 𝑎 ≺ 𝜎(𝑎).
Let 𝑎, 𝑏 ∈ S∗ be such that 𝑎 ≺ 𝑏. Then 𝜎(𝑎)  𝑏, by definition of 𝜎(𝑎),
hence 𝜎(𝑎) ≺ 𝜎(𝑏). This proves that the map 𝜎 : S∗ → S is strictly
increasing; in particular, it is injective.
 A.2. SET THEORY 495

An element 𝑎 ∈ S is a successor if and only if {𝑥 ∈ S ; 𝑥 ≺ 𝑎} has a

largest element 𝑎 0; then 𝑎 = 𝜎(𝑎 0). Otherwise, one says that 𝑎 is limit.

A.2.6. Transfinite constructions. — It is as well possible to make trans-

finite constructions indexed by a well-ordered set S. For 𝑎 ∈ S, let us
write S𝑎 = {𝑥 ∈ S ; 𝑥 ≺ 𝑎}, so that 𝑎 is the smallest element of S S𝑎 .

Proposition (A.2.7) (Transfinite construction principle)

Let X be a set. Let S be a well-ordered set and let 𝒟X be the set of pairs (𝑎, 𝑓 ),
where 𝑎 ∈ S and 𝑓 : S𝑎 → X is a function. For every Φ ∈ 𝒟X , there exists a
unique function F : S → X such that F(𝑎) = Φ(𝑎, F| S𝑎 ) for every 𝑎 ∈ S.

Proof. — Let F, F0 be two such functions. Let us prove that F = F0 by

transfinite induction: let A be the set of 𝑎 ∈ S such that F(𝑎) = F0(𝑎)
and let us prove that A satisfies the transfinite induction hypothesis.
Let 𝑎 ∈ A be such that S𝑎 ⊂ A; by definition, one has F(𝑥) = F0(𝑥) for
every 𝑥 ∈ S𝑎 , so that F| S𝑎 = F0 | S𝑎 ; by definition of F and F0, one has
F(𝑎) = Φ(F| S𝑎 , 𝑎) = Φ(F0 | S𝑎 , 𝑎) = F0(𝑎); this proves that 𝑎 ∈ A.
The existence of F is proved by transfinite induction as well. Let S∗ be
the well-ordered set obtained by adding to S a largest element 𝜔 and
let A be the set of 𝑎 ∈ S∗ such that there exists a function F : S𝑎 → X
satisfying F(𝑏) = Φ(F| S𝑏 , 𝑏) for every 𝑏 ∈ S𝑎 , and let us prove that A
satisfies the transfinite induction hypothesis. Let 𝑎 ∈ S∗ be such that
S𝑎 ⊂ A. We distinguish two cases:
– If 𝑎 is a successor, let 𝑏 ∈ S𝑎 be such that 𝑎 = 𝜎(𝑏) and let F : S𝑏 → X be
a function such that F(𝑥) = Φ(F| S𝑥 , 𝑥) for every 𝑥 ∈ S𝑏 . Then S𝑎 = S𝑏 ∪ {𝑏};
let F0 : S𝑎 → X be defined by F0(𝑥) = F(𝑥) if 𝑥 ∈ S𝑏 and F0(𝑏) = Φ(F| S𝑏 , 𝑏).
Then F0 satisfies the required hypothesis, hence 𝑎 ∈ A.
– Otherwise, 𝑎 is a limit; for every 𝑏 ∈ S𝑎 , there exists a function
F𝑏 : S𝑏 → X satisfying the relation F(𝑥) = Φ(F| S𝑥 , 𝑥) for every 𝑥 ∈ S𝑏 .
By uniqueness, if 𝑏 ≺ 𝑐, then F𝑏 and F𝑐 coincide on S𝑏 . Since 𝑎 is
a limit, the union of all S𝑏 , for 𝑏 ∈ S𝑎 , is equal to S𝑎 , so that there
exists a unique function F : S𝑎 → X which coincides with F𝑏 on S𝑏 .
This function F satisfies the desired relation F(𝑥) = Φ(F| S𝑥 , 𝑥) for all
𝑥 ∈ S𝑎 since for such 𝑥, there exists 𝑏 ∈ S𝑎 such that 𝑥 ∈ S𝑏 , and
F(𝑥) = F𝑏 (𝑥) = Φ(F𝑏 | S𝑥 , 𝑥) = Φ(F| S𝑥 , 𝑥).
496 APPENDIX

By transfinite induction, one has 𝜔 ∈ A. Since S𝜔 = S, this proves that

there exists a function F : S → X satisfying the relation F(𝑎) = Φ(F| S𝑎 , 𝑎)
for every 𝑎 ∈ S. 

Corollary (A.2.8). — Let S, T be two well-ordered sets. One and only one of
the following assertions holds:
(i) There exists an increasing bijection 𝑓 : S → T;
(ii) There exists 𝑎 ∈ S and an increasing bijection 𝑓 : S𝑎 → T;
(iii) There exists 𝑏 ∈ T and an increasing bijection 𝑓 : S → T𝑏 .
Moreover, in each case, the indicated bijection (and the element 𝑎, resp. 𝑏) is
uniquely determined.

This is a very strong rigidity property of well-ordered sets.

Proof. — a) Let us add to T an element 𝜔 and endow the set T∗ = T ∪ {𝜔}
with the order such that 𝜔 > 𝑏 for all 𝑏 ∈ T, so that 𝜔 is the largest
element of T∗ . Then T∗ is well-ordered. By transfinite induction, there
exists a unique map 𝑓 : S → T∗ such that 𝑓 (𝑎) = inf((T∗ 𝑓 (S𝑎 )) ∪ {𝜔})
for every 𝑎 ∈ S. Explicitly, one has 𝑓 (𝑎) = inf(T 𝑓 (S𝑎 )) if 𝑓 (S𝑎 ) ≠ T, and
𝑓 (𝑎) = 𝜔 if 𝑓 (S𝑎 ) ⊃ T.
It follows from this definition that if 𝜔 ∈ 𝑓 (S), then 𝑓 (S) = T∗ . Let
moreover 𝑥, 𝑦 ∈ S such that 𝑥 < 𝑦 and 𝑓 (𝑥) = 𝑓 (𝑦); by the definition
of 𝑓 , one has 𝑓 (𝑦) = 𝜔. Consequently, the map 𝑓 induces a bijective
increasing map from 𝑓 −1 (T) to 𝑓 ( 𝑓 −1 (T)).
We will show that the three cases of the corollary correspond to
the three mutually incompatible possibilities: 𝑓 (S) = T, 𝜔 ∈ 𝑓 (S) and
𝑓 (S) ( T.
First assume that 𝑓 (S) = T. Then 𝑓 is a bijection.
Assume that 𝜔 ∈ 𝑓 (S) and let 𝑎 = inf( 𝑓 −1 (𝜔)). Then 𝑓 (S𝑎 ) ⊂ T and
𝑓 (𝑎) = 𝜔 = inf(T∗ 𝑓 (S𝑎 )), so that 𝑓 (S𝑎 ) = T. Consequently, 𝑓 induces
an increasing bijection from S𝑎 to T.
Assume finally that 𝑓 (S) ( T and let 𝑏 = inf(T 𝑓 (S)), so that T𝑏 ⊂ 𝑓 (S)
and 𝑏 ∉ 𝑓 (S). Conversely, if 𝑥 ∈ S, then 𝑓 (𝑥) = inf(T 𝑓 (S𝑥 )) 6
inf(T 𝑓 (S)), since S𝑥 ⊂ S. This implies that 𝑓 (S) = T𝑏 , and 𝑓 is an
increasing bijection from S to T𝑏 .
 A.2. SET THEORY 497

b) To establish the uniqueness assertion, we first prove that if 𝑓 : S → T

is a strictly increasing map, then 𝑓 (𝑥) > 𝑥 for all 𝑥 ∈ S. By transfinite
induction, it suffices to prove that for any 𝑎 ∈ S such that 𝑓 (𝑥) > 𝑥 for
all 𝑥 < 𝑎, then 𝑓 (𝑎) > 𝑎. If this does not hold, that is, if 𝑓 (𝑎) < 𝑎, then
one has 𝑓 ( 𝑓 (𝑎)) < 𝑓 (𝑎) because 𝑓 is stricly increasing, which contradicts
the induction hypothesis.
Let then 𝑓 , 𝑔 : S → T be two increasing bijective maps and let us prove
that 𝑓 = 𝑔. Applying the initial result for 𝑔 −1 ◦ 𝑓 , we have 𝑔 −1 ◦ 𝑓 (𝑥) > 𝑥
for all 𝑥 ∈ S, hence 𝑓 (𝑥) > 𝑔(𝑥) because 𝑔 is increasing. By symmetry,
one has 𝑔(𝑥) > 𝑓 (𝑥) for all 𝑥 ∈ S, hence 𝑓 = 𝑔.
Let 𝑎 ∈ S and let 𝑓 : S → S𝑎 be an increasing bijection. Applying the
initial result, one has 𝑎 6 𝑓 (𝑎); by assumption, 𝑓 (𝑎) ∈ S𝑎 , hence 𝑓 (𝑎) < 𝑎,
a contradiction.
This implies that case (i) of a) is incompatible with the cases (ii) or (iii),
the uniqueness of a bijection 𝑓 as indicated in each case, as well as the
fact that the elements 𝑎 (in (ii)) or 𝑏 (in (iii)) are well defined. Cases (ii)
and (iii) are themselves incompatible: if 𝑓 : S𝑎 → T and 𝑔 : S → T𝑏 are
increasing bijections, let 𝑎 0 ∈ S be such that 𝑓 (𝑎 0) = 𝑏; then 𝑓 induces an
increasing bijection from S𝑎0 to T𝑏 , hence 𝑔 −1 ◦ 𝑓 induces an increasing
bijection from S𝑎0 to S, a contradiction. 

Proposition (A.2.9) (Hartogs). — Let X be a set. There exists a well-ordered

set S such that no map 𝑓 : S → X is injective.

Proof. — Let 𝒮 be the set of well-orderings on a subset A of X. Let us

define two relations ' and  on 𝒮. First, we say that (A, 6A ) ' (B, 6B )
if there exists an increasing bijection from A to B; it is an equivalence
relation. We then say that that (A, 6A )  (B, 6B ) if there exists an
increasing bijection from A to B, or an increasing bijection from A to B𝑏 ,
for some 𝑏 ∈ B. The relation  is compatible with the equivalence
relation ' and corollary A.2.8 implies that on the quotient set S = 𝒮/',
the relation  induces a total order.
In fact, S is even well-ordered. Let indeed 𝒜 be a non-empty subset of S
and let 𝒜0 be its preimage in 𝒮. It suffices to find an element (B, 6B ) ∈ 𝒜0
such that (B, 6B )  (C, 6C ) for every (C, 6C ) ∈ 𝒜0.
498 APPENDIX

Fix an element (A, 6A ) in 𝒜0. Let then ℬ0 be the set of all elements
(B, 6B ) in 𝒜0 such that (B, 6B ) ≺ (A, 6A ). If ℬ0 is empty, then the class
of (A, 6A ) modulo ' is the smallest element of 𝒜. Otherwise, for every
B ∈ ℬ0, let 𝜑(B) be the unique element of A such that there exists an
increasing bijection 𝑓B from B to A𝜑(B) ; let then 𝑎 be the smallest element
of A which is of this form; fix B ∈ ℬ0 such that 𝑎 = 𝜑(B).
Let (C, 6C ) ∈ 𝒜0. If (C, 6C ) ∉ 𝒜0, then (A, 6A )  (C, 6C ), hence
(B, 6B )  (C, 6C ). Otherwise, one has (C, 6C ) ∈ 𝒜0 and 𝜑(C) > 𝜑(B).
Let then 𝑓B : B → A𝜑(B) and 𝑓C : C → A𝜑(C) be increasing bijections. If
𝜑(B) = 𝜑(C), then (B, 6B ) ' (C, 6C ); in particular, (B, 6B )  (C, 6C ).
Otherwise, one has 𝜑(B) < 𝜑(C), so that there exists 𝑐 ∈ C such that
𝑓C (𝑐) = 𝜑(B); then 𝑓C induces an increasing bijection from C𝑐 to A𝜑(B) ,
so that C𝑐 ' B and (B, 6B ) ≺ (C, 6C ). This proves that the class of
(B, 6B ) is the smallest element of 𝒜, and concludes the proof that S is
well-ordered.
To conclude the proof of the proposition, it remains to establish that
there does not exist an injection from S to X. Let us argue by contradiction
and consider such a map ℎ : S → X. Let A = ℎ(S). Transferring the
ordering of S to A by ℎ, one gets a well-ordering 6A on A, so that (A, 6A )
is an element of 𝒮; let 𝛼 be its class in S. For every 𝑥 ∈ S, A 𝑓 (𝑥) is a
well-ordered subset of A; let 𝑔(𝑥) be its class in S; one has 𝑔(𝑥) < 𝛼
by corollary A.2.8. The map 𝑔 : S → S is strictly increasing, so that
𝑔(𝑥) > 𝑥 for every 𝑥 ∈ S. Taking 𝑥 = 𝛼, we get the desired contradiction
𝛼 6 𝑔(𝛼) < 𝛼. 

A.2.10. — From now on, we will make use of the axiom of choice: the
Î
product 𝑖∈I S𝑖 of any family (S𝑖 )𝑖∈I of non-empty sets is non-empty.

Theorem (A.2.11) (Zorn’s lemma). — Let S be an ordered set. Assume that

every well-ordered subset of S has an upper bound. Then S has a maximal
element.

Proof. — Let us argue by contradiction, assuming that S doesn’t have a

maximal element. Let S∗ be the ordered set obtained by adjoining to S a
largest element 𝜔.
 A.2. SET THEORY 499

Let A be a well-ordered subset of S. let 𝑎 be an upper bound of A;

since 𝑎 is not a maximal element of S, there exists 𝑎 0 ∈ S such that 𝑎 < 𝑎 0.
Using the axiom of choice, there exists a function 𝑓 that assigns, to any
well-ordered subset A of S, an element 𝑓 (A) ∈ S such that 𝑥 < 𝑓 (A) for
every 𝑥 ∈ A. We extend 𝑓 to 𝔓(S∗ ) by setting 𝑓 (A) = 𝜔 if A is not a
well-ordered subset of S.
Let T be a well-ordered set that admits no injection to S. By the
transfinite construction principle, there exists a unique map ℎ : T → S∗
such that ℎ(𝑡) = 𝑓 (T𝑡 ) for every 𝑡 ∈ T. Let us prove that ℎ is strictly
increasing and that ℎ(T) ⊂ S. By the transfinite induction principle,
it suffices to prove that if ℎ| T𝑡 is strictly increasing and its image is
contained in S, then ℎ(𝑡) ∈ S and ℎ(𝑥) < ℎ(𝑡) for every 𝑥 ∈ T𝑡 . Then
ℎ(T𝑡 ) is a well-ordered subset of S, hence ℎ(𝑡) = 𝑓 (ℎ(T𝑡 )) ∈ S and satisfies
ℎ(𝑡) > ℎ(𝑥) for every 𝑥 ∈ T𝑡 , hence the claim.
In particular, ℎ is an injective map from T to S, and this contradicts the
choice of T. 

A.2.12. — Classical applications of Zorn’s lemma (theorem A.2.11) hap-

pen in situations where the ordered set S satisfies stronger assumptions.
One says that an ordered set S is inductive if every totally ordered
subset of S has an upper bound. Applying the definition to the empty
subset, we observe that an inductive set is not empty. Moreover, if S
is an inductive set, then for every 𝑎 ∈ S, the set {𝑥 ∈ S ; 𝑎  𝑥} is also
inductive. The following result is then an immediate corollary of Zorn’s
lemma.

Corollary (A.2.13). — Every inductive set has a maximal element. More

precisely, if S is an inductive set and 𝑎 is an element of S, then S has a maximal
element 𝑏 such that 𝑎  𝑏.

A.2.14. — Let X be a set and let S be a subset of 𝔓(X), ordered by

inclusion. One says that S is of finite character if it contains the empty set
and if a subset A of X belongs to S if and only if every finite subset of A
belongs to S. (An important example is given by the set of free subsets
of a vector space, or of a module.)
500 APPENDIX

Lemma (A.2.15). — Let X be a set and let S be a subset of 𝔓(X), ordered by

inclusion. If S is of finite character, then S is inductive.

Proof. — Let T be a totally ordered subset of S; let us prove that T has

an upper bound in S. If T is empty, then ∅ is an upper bound for T,
and ∅ ∈ S, by definition. Otherwise, let V be the union of all elements
of T. By definition, one has A ⊂ V for every A ∈ T, hence V is an upper
bound for T. To conclude the proof of the lemma, it suffices to prove
that V belongs to S.
By the definition of a set of finite character, in order to prove that V ∈ S,
it suffices to prove that every finite subset A of V belongs to S.
In fact, let us prove by induction on the cardinality of A that there
exists a subset B ∈ T such that A ⊂ B. This holds if A = ∅, since ∅ ∈ S,
by assumption. Otherwise, let 𝑎 ∈ A and let A0 = A {𝑎}; by induction,
there exists B0 ∈ T such that A0 ⊂ B0. Since 𝑎 ∈ A ⊂ V, there exists a
set B00 ∈ T such that 𝑎 ∈ B00. Since T is totally ordered, one has either
B0 ⊂ B00, or B00 ⊂ B0; since A = A0 ∪ {𝑎}, we get A ⊂ B00 or A ⊂ B0, which
proves the claim by induction.
Since A is finite, the definition of a set of finite character implies that
A belongs to S.
Applying once more the definition of a set of finite character, this
proves that V ∈ S. 

Corollary (A.2.16) (Cantor). — Let X, Y be sets. At least one of the two

possibilities hold:
(i) There exists an injective map 𝑓 : X → Y;
(ii) There exists an injective map 𝑓 : Y → X;

Proof. — Let 𝒞 be the set of all pairs (A, 𝑓 ), where A is a subset of X and
𝑓 : A → Y is an injective map. We order 𝒞 as follows: (A, 𝑓 ) 6 (B, 𝑔) if
A ⊂ B and 𝑔| A = 𝑓 . Let us prove that the set 𝒞 is inductive.
Let ((A𝑖 , 𝑓𝑖 ))𝑖∈I be a totally ordered family of elements of 𝒞; let A =
𝑖 A 𝑖 . I claim that there exists a unique map 𝑓 : A → Y such that
Ð
𝑓 (𝑥) = 𝑓𝑖 (𝑥) for every 𝑖 ∈ I and every 𝑥 ∈ A𝑖 . Let indeed 𝑖, 𝑗 ∈ I be
such that 𝑥 ∈ A𝑖 and 𝑥 ∈ A 𝑗 ; if (A𝑖 , 𝑓𝑖 ) 6 (A 𝑗 , 𝑓 𝑗 ), then A𝑖 ⊂ A 𝑗 and
 A.3. CATEGORIES 501

𝑓𝑖 = 𝑓 𝑗 | A𝑖 , so that 𝑓𝑖 (𝑥) = 𝑓 𝑗 (𝑥); by symmetry, the same equality holds if

(A 𝑗 , 𝑓 𝑗 ) 6 (A𝑖 , 𝑓𝑖 ).
Let 𝑥, 𝑦 ∈ A be such that 𝑓 (𝑥) = 𝑓 (𝑦). Let 𝑖, 𝑗 ∈ I be such that 𝑥 ∈ A𝑖
and 𝑦 ∈ A 𝑗 . If (A𝑖 , 𝑓𝑖 ) 6 (A 𝑗 , 𝑓 𝑗 ), then 𝑥, 𝑦 ∈ A 𝑗 , hence 𝑓 (𝑥) = 𝑓 𝑗 (𝑥) and
𝑓 (𝑥) = 𝑓 𝑗 (𝑦), hence 𝑓 𝑗 (𝑥) = 𝑓 𝑗 (𝑦); since 𝑓 𝑗 is injective, this implies 𝑥 = 𝑦.
By symmetry, the same property holds if (A 𝑗 , 𝑓 𝑗 ) 6 (A𝑖 , 𝑓𝑖 ). This proves
that 𝑓 is injective.
By Zorn’s theorem, the set 𝒞 admits a maximal element, say (A, 𝑓 ).
If A = X, then 𝑓 is an injective map from X to Y. Assume that A ≠ X; I
then claim that 𝑓 is surjective. Otherwise, there exists 𝑎 ∈ X A and
𝑏 ∈ Y 𝑓 (A); the map 𝑔 : A ∪ {𝑎} → Y such that 𝑔| A = 𝑓 and 𝑔(𝑎) = 𝑏 is
injective, and one has (A, 𝑓 ) 6 (A ∪ {𝑎}, 𝑔), contradicting the hypothesis
that (A, 𝑓 ) is a maximal element of 𝒞. The map 𝑓 : A → Y is thus
bijective; its inverse 𝑓 −1 : Y → A is an injective map from Y to X. This
concludes the proof of Cantor’s theorem. 

Corollary (A.2.17) (Zermelo, 1904). — Every set can be well-ordered.

Proof. — Let X be a set. Let S be a set such that there is no injection
from S to X. By Cantor’s theorem (corollary A.2.16), there exists an
injection 𝑓 from X to S. Then there exists a unique ordering relation
on X such that 𝑓 is increasing; this ordering is a well-ordering on X,
establishing the corollary. 

A.3. Categories
It is a very useful and common vocabulary to describe certain algebraic
structures (the so-called categories) and the way by which one can
connect them (functors).

A.3.1. Categories. — A category C consists in the following data:

– A collection ob C of objects;
– For any two objects M, N, a set C (M, N) called morphisms from M
to N;
– For any three objects M, N, P, a composition map C (M, N) × C (N, P),
( 𝑓 , 𝑔) ↦→ 𝑔 ◦ 𝑓 ,
502 APPENDIX

so that the following axioms are satisfied:

(i) For any object M, there is a distinguished morphism idM ∈ C (M, M),
called the identity;
(ii) For
(iii) One has idN ◦ 𝑓 = 𝑓 for any 𝑓 ∈ C (M, N);
(iv) One has 𝑔 ◦ idN = 𝑔 for any 𝑔 ∈ C (N, P);
(v) For any four objects M, N, P, Q, and any three morphisms 𝑓 ∈
C (M, N), 𝑔 ∈ C (N, P), ℎ ∈ C (P, Q), the two morphisms ℎ ◦ (𝑔 ◦ 𝑓 ) and
(ℎ ◦ 𝑔) ◦ 𝑓 in C (M, Q) are equal (associativity of composition).
A common notation for C (M, N) is also HomC (M, N). Finally, instead
of 𝑓 ∈ C (M, N), one often writes 𝑓 : M → N.
Let 𝑓 : M → N be a morphism in a category C . One says that 𝑓 is
left-invertible, resp. right-invertible, resp. invertible, if there exists a
morphism 𝑔 : N → M such that 𝑔 ◦ 𝑓 = idM , resp. 𝑓 ◦ 𝑔 = idN , resp.
𝑔 ◦ 𝑓 = idM and 𝑓 ◦ 𝑔 = idN .
One proves in the usual way (see p. ) that if 𝑓 is both left- and right-
invertible, then it is invertible. An invertible morphism is also called an
isomorphism.
Let us now give examples of categories. It will appear clearly that all
basic structures studied of algebras fall within the categorical framework.

Examples (A.3.2). — The category Set of sets has for objects the sets,
and for morphisms the usual maps between sets.
The category Gr of groups has for objects the groups and for mor-
phisms the morphisms of groups. The category AbGr of Abelian groups
has for objects the Abelian groups and for morphisms the morphisms
of groups. Observe that objects of AbGr are objects of Gr , and that
morphisms in AbGr coincide with those in Gr ; one says that AbGr is
a full subcategory of Gr .
The category Ring of rings has for objects the rings and for morphisms
the morphisms of rings.
Similarly, there is the category Field of fields and, if 𝑘 is a field, the
category Ev 𝑘 of 𝑘-vector spaces. More generally, for any ring A, there
is a category Mod A of right A-modules, and a category Mod A of left
A-modules.
 A.3. CATEGORIES 503

Example (A.3.3). — Let C be a category; its opposite category C o has

the same objects than C , but the morphisms of C o are defined by
C o (M, N) = C (N, M) and composed in the opposite direction.
It resembles the definition of an opposite group. However, a category
is usually different from its opposite category.

Remark (A.3.4). — Since there is no set containing all sets, nor a set
containing all vector spaces, the word collection in the above definition
cannot be replaced by the word set (in the sense of Zermelo-Fraenkel’s
theory of sets). In fact, a proper treatment of categories involves set
theoretical issues. There are at least three ways to solve them:
– The easiest one is to treat a category as a formula (in the sense of
first order logic). For example Ring is a formula 𝜑Ring with one free
variable A that expresses that A is a ring. This requires to encode a ring
A and all its laws as a tuple: for example, one may consider a ring to
be a tuple (A, S, P) where A is the ring, S is the graph of the addition
law and P is the graph of the multiplication law. The formula 𝜑Ring (𝑥)
then checks that 𝑥 is a triplet of the form (A, S, P), where S ⊂ A3 and
P ⊂ A3 , that S is the graph of a map A × A → A which is associative,
commutative, has a neutral element, and for which every element has
an opposite, etc.
Within such a framework, one can also consider functors (defined
below), but only those which can be defined by a formula.
This treatment would be sufficient at the level of this book.
– One can also use another theory of sets, such as the one of Bernays-
Gödel-von Neumann, which allows for two kinds of collections: sets
and classes. Sets, obey to the classical formalism of sets, but classes are
more general, so that one can consider the class of all sets (but not the
class of all classes). Functors are defined as classes.
This is a very convenient possibility at the level of this book. However,
at a more advanced development of algebra, one is lead to consider the
category of categories, or categories of functors. Then, this approach
becomes unsufficient as well.
– Within the classical theory of sets, Grothendieck introduced universes
which are very large sets, so large than every usual construction of sets
504 APPENDIX

does not leave a given universe. One needs to add the axiom that there
is an universe, or, more generally, that any set belongs to some universe.
This axiom is equivalent to the existence of inaccessible cardinals, an axiom
which is well studied and often used in advanced set theory.
In this book, categories is mostly a language to state algebraic results of
quite a formal nature.
Definition (A.3.5). — Let C be a category, let M, N be objects of C and let
𝑓 ∈ C (M, N).
One says that 𝑓 is a monomorphism if for any object P of C and any
morphisms 𝑔1 , 𝑔2 ∈ C (N, P) such that 𝑔1 ◦ 𝑓 = 𝑔2 ◦ 𝑓 , one has 𝑔1 = 𝑔2 .
One says that 𝑓 is an epimorphism if for any object L of C and any
morphisms 𝑔1 , 𝑔2 ∈ C (P, M) such that 𝑓 ◦ 𝑔1 = 𝑓 ◦ 𝑔2 , one has 𝑔1 = 𝑔2 .
Example (A.3.6). — Monomorphisms and epimorphisms in Set or in
categories of modules are respectively injections and surjections (see
exercise 7/20).

A.3.7. Functors. — Functors are to categories what maps are to sets.

Let C and D be two categories.
A functor F from C to D consists in the following data:
– an object F(M) of D for any object M of C ;
– a morphism F( 𝑓 ) ∈ D(F(M), F(N)) for any objets M, N of C and any
morphism 𝑓 ∈ C (M, N),
subject to the two following requirements:
(i) For any object M of C , F(idM ) = idF(M) ;
(ii) For any objects M, N, P of C and any morphisms 𝑓 ∈ C (M, N)
and 𝑔 ∈ C (N, P), one has
F(𝑔 ◦ 𝑓 ) = F(𝑔) ◦ F( 𝑓 ).
A contravariant functor F from C to D is a functor from C o to D.
Explicitly, it consists in the following data
– an object F(M) of D for any object M of C ;
– a morphism F( 𝑓 ) ∈ D(F(N), F(M)) for any objets M, N of C and any
morphism 𝑓 ∈ C (M, N),
subject to the two following requirements:
 A.3. CATEGORIES 505

(i) For any object M of C , F(idM ) = idF(M) ;

(ii) For any objects M, N, P of C and any morphisms 𝑓 ∈ C (M, N)
and 𝑔 ∈ C (N, P), one has
F(𝑔 ◦ 𝑓 ) = F( 𝑓 ) ◦ F(𝑔).
One says that such a functor F is faithful, resp. full, resp. fully faithful
if for any objects M, N of C , the map 𝑓 ↦→ F( 𝑓 ) from C (M, N) to
C (F(M), F(N)) is injective, resp. surjective, resp. bijective. A similar
definition applies for contravariant functors.
Example (A.3.8) (Forgetful functors). — Many algebraic structures are
defined by enriching other structures. Often, forgetting this enrichment
gives rise to a functor, called a forgetful functor.
For example, a group is already a set, and a morphism of groups
is a map. There is thus a functor that associates to every group its
underlying set, thus forgetting the group structure. One gets a forgetful
functor from Gr to Set. It is faithful, because a group morphism is
determined by the map between the underlying sets. It is however not
full because there are maps between two (non-trivial) groups which are
not morphism of groups.
Example (A.3.9). — Let C be a category and let P be an object of C .
One defines a functor F from the category C to the category of sets as
follows:
– For any object M of C , one sets F(M) = C (P, M);
– For any morphism 𝑓 : M → N in C , F( 𝑓 ) is the map 𝑢 ↦→ 𝑓 ◦ 𝑢 from
C (P, M) to C (P, N).
This functor is often denoted HomC (P, •). Such a functor is also called
a representable functor.
One can also define a contravariant functor G, denoted HomC (•, P) as
follows:
– For any object M of C , one sets F(M) = C (M, P);
– For any morphism 𝑓 : M → N in C , F( 𝑓 ) is the map 𝑢 ↦→ 𝑢 ◦ 𝑓 from
C (N, P) to C (M, P).
In other words, G is the functor HomC o (P, •). One says that it is
corepresentable.
506 APPENDIX

Let F and G be two functors from a category C to a category D.

A morphism of functors 𝛼 from F to G consists in the datum, for
every object M of C , of a morphism 𝛼 M : F(M) → G(M) such that the
following condition holds: For any morphism 𝑓 : M → N in C , one has
𝛼N ◦ F( 𝑓 ) = G( 𝑓 ) ◦ 𝛼 M .
Morphisms of functors can be composed, and for any functor F, one
has an identity morphism from F to itself. Consequently, functors from
C to D form themselves a category, denotes F (C , D).

A.3.10. Universal properties and representable functors. — This

book contains many universal properties: the free module on a given basis,
quotient ring, quotient module, direct sum and product of modules,
localization, algebra of polynomials on a given set of indeterminates.
They are all of the following form: “in such algebraic situation, there
exists an object and a morphism satisfying such property and such that
any other morphism which satisfies this property factors through it”.
The prototype of an universal property is the following.

Definition (A.3.11). — Let C be a category.

One says that an object I of C is an initial object if for every object M of C ,
(I, M) has exactly one element.
One says that an object T of C is a terminal object if for every object M
of C , C (M, T) has exactly one element.

Observe that an initial object of a category is a terminal object of the

opposite category, and vice versa.

Examples (A.3.12). — In the category Set of sets, the empty set is the
only initial object, and any set of cardinality one is a terminal object.
In the category Ring of rings, the ring Z is an initial object (for any
ring A, there is exactly one morphism from Z to A, see Example 1.3.4).
Moreover, the ring 0 is a terminal object.
In the category of modules over a ring A, the null module is both an
initial and a terminal object.

The property for an object I to be an initial object can be rephrased as

a property of the representable functor Hom C (I, •), namely that this
 A.3. CATEGORIES 507

functor coincides with (or, rather, is isomorphic to) the functor F that
sends any object of C to a fixed set with one element.
This allows to rephrase the definition of an initial object as follows: an
object I is an initial object if it represents the functor F defined above.

Definition (A.3.13). — Let C be a category and let F be a functor from C to

the category Set of sets. Let P be an object of C . One says that P represents
the functor F is the functor HomC (P, •) is isomorphic to F.
Similarly, if G is a contravariant functor from C to Set, one says that an
object P of C co-represents G is the functors HomC (•, P) and G are isomorphic.

Objects that represent a given functor are unique up to an isomorphism:

Proposition (A.3.14) (Yoneda’s Lemma). — Let C be a category, let A and

B be two objects of C .
For any morphism of functors 𝜑 from HomC (A, •) to HomC (B, •), there
is a unique morphism 𝑓 : B → A such that 𝛼 M (𝑢) = 𝑢 ◦ 𝑓 for any object M
of C and any morphism 𝑢 ∈ C (A, M).
In particular, 𝜑 is an isomorphism if and only if 𝑓 is an isomorphism.

Proof. — Let us write F = HomC (A, •) and G = HomC (B, •). Recall the
definition of a morphism of functors: for every object M of C , one has a
map 𝜑M : F(M) → G(M) such that G(𝑢) ◦ 𝜑M = 𝜑N ◦ F(𝑢) for every two
objects M, N of C and every morphism 𝑢 : M → N. In the present case,
this means that for every object M of C , 𝜑M is a map from C (A, M)
to C (B, M) and that 𝑢 ◦ 𝜑M ( 𝑓 ) = 𝜑N (𝑢 ◦ 𝑓 ) for every 𝑓 ∈ C (A, M) and
every 𝑢 ∈ (M, N). In particular, taking M = A and 𝑓 = idA , one obtains
𝜑N (𝑢) = 𝑢 ◦ 𝜑A (idA ). This proves the existence of a morphism 𝑓 as
required by the lemma, namely 𝑓 = 𝜑A (idA ). The uniqueness of 𝑓
follows also from this formula, applied to N = A and 𝑢 = idA : if
𝑓 0 : N → M is any morphism possessing the required property„ one has
𝑓 = 𝜑A (idA ) = idA ◦ 𝑓 0 = 𝑓 0. 

A.3.15. Adjunction. — Let C and D be two categories, let F be a

functor from C to D and G be a functor from D to C .
508 APPENDIX

An adjunction for the pair (F, G) is the datum, for any objects M of C
and N of D, of a bijection
∼
ΦM,N : D(F(M), N) →
− C (M, G(N))
such that the following holds: for any objects M, M0 of C , any morphism
𝑓 ∈ C (M, M0), any objects N, N0 of D, any morphism 𝑔 ∈ D(M, M0),
and any morphism 𝑢 ∈ D(F(M0), N),
𝑔 ◦ ΦM0 ,N (𝑢) ◦ F( 𝑓 ) = ΦM,N0 (G(𝑔) ◦ 𝑢 ◦ 𝑓 ).
If there exists an adjunction for the pair (F, G), one says that F is a left
adjoint to G, or that G is a right adjoint to F.
Section 7.6 of the book offers a first study of adjoint morphisms in
the framework of categories of modules. The reader is invited to try to
generalize what is explained there to other categories.
Remark (A.3.16). — Assume that F and G are adjoint functors. Then for
any object M of C , F(M) represents the functor HomC (M, G(•)) from D
to Set.
Conversely, let G be a functor from D to D. One proves that if the
functor HomC (M, G(•)) is representable for any object M of C , then
G has a left adjoint. The proof consists in choosing, for any object M
of C , an object F(M) which represents the given functor. Moreover,
any morphism 𝑓 : M → M0 in C gives rise to a morphism from the
functor HomC (M0 , G(•)) to the functor HomC (M, G(•)), hence, by the
contravariant version of Yoneda’s Lemma, to a morphism F( 𝑓 ) : F(M) →
F(M0).
BIBLIOGRAPHY

M. Aigner & G. M. Ziegler (2014), Proofs from The Book, Springer-Verlag,

Berlin, fifth edition. URL http://dx.doi.org/10.1007/978-3-662-44205-0,
including illustrations by Karl H. Hofmann.
H. Bass (1960), “Finitistic dimension and a homological generalization
of semi-primary rings”. Transactions of the American Mathematical
Society, 95 (3), pp. 466–488.
N. Bourbaki (1989), Elements of Mathematics. Commutative Algebra. Chap-
ters 1-7. Transl. from the French. 2nd Printing., Berlin etc.: Springer-
Verlag, 2nd printing edition.
N. Bourbaki (2012), Éléments de mathématique. Algèbre. Chapitre 8. Modules
et anneaux semi-simples, Springer, Berlin. Second revised edition of
the 1958 edition [MR0098114].
T. Coquand & H. Lombardi (2005), “A Short Proof for the Krull Di-
mension of a Polynomial Ring”. The American Mathematical Monthly,
112 (9), p. 826.
B. Deschamps (2001), “À propos d’un théorème de Frobenius”. Annales
mathématiques Blaise Pascal, 8 (2), pp. 61–66.
D. Eisenbud (1995), Commutative algebra with a view towards algebraic
geometry, Graduate Texts in Math. 150, Springer-Verlag.
A. Grothendieck (1960), “Technique de descente et théorèmes
d’existence en géométrie algébrique. I. Généralités. Descente par
morphismes fidèlement plats”. Séminaire Bourbaki : années 1958/59 -
1959/60, exposés 169-204, Séminaire Bourbaki 5, pp. 299–327, Société
mathématique de France. URL http://www.numdam.org/item/SB_
1958-1960__5__299_0, talk:190.
D. W. Henderson (1965), “A Short Proof of Wedderburn’s Theorem”.
The American Mathematical Monthly, 72 (4), p. 385.
K. Ireland & M. Rosen (1990), A Classical Introduction to Modern Number
Theory, Graduate Texts in Mathematics 84, Springer New York, New
510 BIBLIOGRAPHY

York, NY.
C. W. Kohls (1958), “Prime ideals in rings of continuous functions”.
Illinois Journal of Mathematics, 2 (4A), pp. 505–536.
H. Matsumura (1986), Commutative ring theory, Cambridge studies in
advanced mathematics, Cambridge Univ. Press.
W. K. Nicholson (1993), “A short proof of the Wedderburn-Artin
theorem”. New Zealand Journal of Mathematics, 22 (1), pp. 83–86.
R. S. Palais (1968), “The classification of real division algebras”. Amer.
Math. Monthly, 75, pp. 366–368.
H. Perdry (2004), “An Elementary Proof of Krull’s Intersection Theo-
rem”. The American Mathematical Monthly, 111 (4), p. 356.
M. A. Rieffel (1965), “A General Wedderburn Theorem”. Proceedings of
the National Academy of Sciences, 54 (6), pp. 1513–1513.
A. Seidenberg (1954), “On the dimension theory of rings. II.” Pacific
Journal of Mathematics, 4 (4), pp. 603–614.
H. Tverberg (1964), “A remark on Ehrenfeucht’s criterion for irreducibil-
ity of polynomials”. Commentationes Mathematicae, 8 (2).
INDEX

A associated
additive — elements, 12
— functor, 349 associated prime ideal, 300, 305
adjugate, 407 automorphism
adjunction, 508 — of ring, 8
Akizuki’s theorem, 296 interior —, 10
algebra, 6, 16 axiom of choice, 498
group —, 7
integral, 180 B
monoid —, 7 Baer’s theorem, 346
algebraic closure, 186 balanced bi-additive map, 380
base normale, 224
— of a field in an algebra, 183
basis, 135
existence of an —, 187
canonical — of (A𝑑 )𝑛 , 139
algebraic set, 77
Bass-Papp
algebraically closed field, 185
theorem of —, 375
algebraically independent family,
Berlekamp’s algorithm, 225
213
bimodule, 119
alternate binomial formula, 33
see𝑝-linear map, 152 Bézout— theorem, 101
annihilator Bézout’s theorem, 92
of a module, 120 bonded subset, 135
of an element, 120 Bruhat decomposition, 281
artinian
— module, 290 C
— ring, 291 canonical basis, 136
Artin–Rees lemma, 316 Cantor-Bernstein, 493
Artin–Tate lemma, 289 category, 501

511
512 INDEX

Cayley–Hamilton’s theorem, 55 with respect to an

center of a ring, 7 indeterminate, 20
centralizer of a subset, 7 derivative of a polynomial, 25
character of a finite abelian determinant
group, 268 — of a square matrix, 157
characteristic of a field, 17, 198 — of an endomorphism of a
characteristic polynomial free module, 158
— of an endomorphism, 259 (alternate linear form on a free
Chevalley–Dunford module), 157
decomposition of a matrix, diagram, 324
219 commutative, 324
Chinese remainder theorem, 38 differential, 363
Cohen’s theorem, 315 differential module, 363
cokernel, 132 morphism of —s, 363
column equivalence of matrices, dimension
229 — of a vector space, 147
column rank, 230, 234 direct summand, 130, 335
comaximal ideals, 92 divisible, 346
Commuting elements in a ring, 3 division algebra, 12
companion matrix, 258 division ring, 12
complex, 324 domain, 11
conductor, 57 unique factorization —, 86
content, 94 dual
contraction, 57 — of a module, 125
coprimary dual basis, 138
— module, 306
coprime elements, 92 E
coproduct elementary operations on rows
in a category, 129 and columns of a matrix,
corps 229
fini, 225 endomorphism
cyclic module, 257 — of module, 122
cyclic vector, 257 euclidean
gauge, 84
D ring, 84
decreasing, 492 Euclidean division
strictly —, 492 for polynomials, 23
degree of a polynomial exact sequence, 324
total —, 20 short —, 325
 INDEX 513

split —, 326 right exact —, 350

exactness Fundamental theorem of
— of localization, 50, 143 Algebra, 185
exactness of localization, 356
Exchange lemma, 215 G
exponent Galois extension, 201, 203
of a monomial, 20 Galois group, 201
extension, 57 Galois’s theorem, 203
of rings, 175 gaugesee also eclidean, 84
generating subset, 134
F greatest common divisor, 90
factorization theorem, 34, 38 H
field, 12 height of a prime ideal, 458
— of fractions, 45 Hilbert’s Nullstellensatz, 80
algebraically closed —, 27 Hilbert’s theorem, 286
finite field, 198, 201
finitely generated I
— algebra, 286, 289 ideal, 28
— module, 136, 289 comaximal —s, 38
Fitting ideal left —, 27
of a module, 167 right —, 27
flat module, 411 two-sided —, 27
fractional ideal, 473 idempotent
free — element, 52
— module, 136 increasing, 492
free module, 335 strictly —, 492
free subset, 135 indecomposable module, 313
Frobenius, 191 inductive ordered set, 499
Frobenius homomorphism, 9, injective module, 343, 353
199, 201 integral
functor, 348, 504 — element, 176
— contravariant, 349 integral closure, 178
— covariant, 349 integrally closed, 179
contravariant —, 504 invariant factors, 250
exact —, 350 inverse, 11
faithful —, 505 invertible element, 11
full —, 505 invertible fractional ideal, 474
fully faithful —, 505 irreducible element, 85
left exact —, 350 isomorphism
514 INDEX

— of modules, 122 dual —, 125

faithful, 120
J finitely generated, 288, 305
Jacobson ring, 455 finitely generated —, 282, 304
Jordan block, 262 finitely presented —, 331
Jordan decomposition, 262 free — on a set, 136
Jordan–Hölder theorem, 278 graded —, 364
left —, 117
K
of finite length, 305
kernel
right —, 116
— of a morphism of modules,
semisimple, 313
123
monic polynomial, 21
— of a ring morphism, 30
Krull’s theorem, 188 monomial, 20
morphism
L — of modules, 122
Laplace expansion, 407 — of rings, 8
Laplace expansion of the (in a category), 501
determinant, 160 multiplicative
largest lower bound, 493 — subset, 40
leading coefficient, 21 multiplicity
least common multiple, 90 of a root of a polynomial, 26
least upper bound, 493
linear combination, 118
linear form, 125 N
Lüroth’s theorem, 226 Nakayama’s lemma, 312
nilpotent, 10
M — element, 319
matrix nilradical, 32
diagonal —, 164 Noether normalization theorem,
elementary —, 228 472
permutation —, 228 noetherian
maximal spectrum, 67 — module, 282, 283
minimal polynomial — ring, 87, 285
— of an endomorphism, 259 ring, 304, 305
minor Noether’s normalization
— of a matrix, 164 theorem, 452
module, 117 norm, 208
direct product of —s, 126 normal extension, 467
direct sum of —s, 127 normalizer, 207
 INDEX 515

O projective
object — module, 334
(of a category), 501 projective module, 335, 353
initial —, 506
terminal —, 506
Q
Ore condition, 47, 59
quaternion, 12
quotient ring, 34
P
perfect field, 192
permutation R
— matrix, 228 Rabinowitsch trick, 81
Perron’s theorem, 111 radical
pivot column indices, 234 — of an ideal, 32
pivot row indices, 234 Jacobson —, 64
𝑝-linear form, 152 nilpotent —, 32
alternate —, 152 rank
𝑝-linear map, 152 — of a module, 250
alternate —, 152 of a free module, 149
antisymmetric —, 152 reduced
Poincaré’s lemma, 366 ring, 32
polynomial reduced column echelon form,
split —, 27 230
presentation, 331 reduced row echelon form, 229
finite —, 331 regular
primary — element, 10, 319
— submodule, 306 resultant, 99, 103
primary component, 251 — and roots, 103
primary decomposition, 309 degree of the —, 102
minimal —, 309 ring, 2
prime field commutative, 2
of a division ring, 17 noetherian ring, 283
prime ideal, 67 ring morphism
Primitive element theorem, 196 pure —, 446
primitive polynomial, 94 ring of polynomials, 19
principal ideal, 82 ring of power series, 55
principal ideal domain, 83, 346 root, 25
product Rouché’s theorem, 111
in a category, 129 row equivalence of matrices, 229
product of ideals, 32 row rank, 230, 234
516 INDEX

S torsion module, 121

separable torsion submodule, 121, 250
— element, 193 torsion-free module, 121
— extension, 193 totally ordered set, 492
— polynomial, 191 trace, 208
set of finite character, 499 transcendence degree, 217
signature of a permutation, 152
simple U
— module, 274 unit, 11
snake lemma, 328 unit element of a ring, 3
spectrum universal property, 34, 394
— of a ring, 67 — of direct products of
split polynomial, 185 modules, 127
stathmsee also eclidean gauge, 84 — of direct sums of modules,
stationary, 492 127
Steinitz’s theorem, 187 — of free A-modules, 336
submodule, 120 — of polynomial algebras, 24
— generated, 126 — of quotient modules, 131
intersection of —s, 125 — of quotient rings, 34
pure —, 444 — of tensor products, 381
sum of —s, 126 of fraction rings, 42
subring, 7
support, 297 V
von Neumann ring, 108
T
tensor, 381 W
split —, 381 Wedderburn’s theorem, 318
theorem well-ordered set, 494
— of Cohen–Seidenberg, 460 Weyl group, 228
Theorem of Krull–Akizuki, 485
théorème Z
— de Chevalley–Warning, 225 Zariski topology, 69, 78
— de d’Alembert–Gauß, 222 zero divisor, 11
— de Lüroth, 226 zero element of a ring, 3
théorème de Wedderburn, 223 Zorn’ lemma, 345
torsion element, 121 Zorn’s lemma, 498