Bosch - Algebraic Geometry and Commutative Algebra PDF

Universitext
Universitext
Series Editors:
Sheldon Axler
San Francisco State University, San Francisco, CA, USA
Vincenzo Capasso
Universit degli Studi di Milano, Milan, Italy
Carles Casacuberta
Universitat de Barcelona, Barcelona, Spain
Angus MacIntyre
Queen Mary, University of London, London, UK
Kenneth Ribet
University of California, Berkeley, Berkeley, CA, USA
Claude Sabbah
CNRS, cole Polytechnique, Palaiseau, France
Endre Sli
University of Oxford, Oxford, UK
Wojbor A. Woyczynski
Case Western Reserve University, Cleveland, OH, USA
Universitext is a series of textbooks that presents material from a wide variety

of mathematical disciplines at masters level and beyond. The books, often well
class-tested by their author, may have an informal, personal, even experimental
approach to their subject matter. Some of the most successful and established
books in the series have evolved through several editions, always following the
evolution of teaching curricula, into very polished texts.
Thus as research topics trickle down into graduate-level teaching, rst textbooks
written for new, cutting-edge courses may make their way into Universitext.
For further volumes:

www.springer.com/series/223
Siegfried Bosch
Algebraic
Geometry and
Commutative
Algebra
Prof. Dr. Siegfried Bosch
Mathematisches Institut
Westflische Wilhelms-Universitt
Mnster, Germany
ISSN 0172-5939 ISSN 2191-6675 (electronic)

Universitext
ISBN 978-1-4471-4828-9 ISBN 978-1-4471-4829-6 (eBook)
DOI 10.1007/978-1-4471-4829-6
Springer London Heidelberg New York Dordrecht
Library of Congress Control Number: 2012953696
Mathematics Subject Classication: 13-02, 13Axx, 13Bxx, 13Cxx, 13Dxx, 13Exx, 13Hxx, 13Nxx, 14-
02, 14Axx, 14B25, 14C20, 14F05, 14F10, 14K05, 14L15
Springer-Verlag London 2013

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specically the rights of translation, reprinting, reuse of illustrations, recitation,
broadcasting, reproduction on microlms or in any other physical way, and transmission or information
storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar methodology
now known or hereafter developed. Exempted from this legal reservation are brief excerpts in connection
with reviews or scholarly analysis or material supplied specically for the purpose of being entered and
executed on a computer system, for exclusive use by the purchaser of the work. Duplication of this pub-
lication or parts thereof is permitted only under the provisions of the Copyright Law of the Publishers
location, in its current version, and permission for use must always be obtained from Springer. Permis-
sions for use may be obtained through RightsLink at the Copyright Clearance Center. Violations are liable
to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specic statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of publica-
tion, neither the authors nor the editors nor the publisher can accept any legal responsibility for any errors
or omissions that may be made. The publisher makes no warranty, express or implied, with respect to the
material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

Preface
The domain of Algebraic Geometry is a fascinating branch of Mathematics

that combines methods from Algebra and Geometry. In fact, it transcends
the limited scope of pure Algebra, in particular Commutative Algebra, by
means of geometrical construction principles. Looking at its history, the theory
has behaved more like an evolving process than a completed workpiece, as quite
often the challenge of new problems has caused extensions and revisions. For
example, the concept of schemes invented by Grothendieck in the late 1950s
made it possible to introduce geometric methods even into elds that formerly
seemed to be far from Geometry, like algebraic Number Theory. This paved
the way to spectacular new achievements, such as the proof by Wiles and Taylor
of Fermats Last Theorem, a famous problem that was open for more than 350
years.
The purpose of the present book is to explain the basics of modern Al-
gebraic Geometry to non-experts, thereby creating a platform from which
one can take o towards more advanced regions. Several times I have given
courses and seminars on the subject, requiring just two semesters of Linear
Algebra for beginners as a prerequisite. Usually I did one semester of Com-
mutative Algebra and then continued with two semesters of Algebraic
Geometry. Each semester consisted of a combination of traditional lectures
together with an attached seminar where the students presented additional ma-
terial by themselves, extending the theory, supplying proofs that were skipped
in the lectures, or solving exercise problems. The material covered in this way
corresponds roughly to the contents of the present book. Just as for my stu-
dents, the necessary prerequisites are limited to basic knowledge in Linear
Algebra, supplemented by a few facts from classical Galois theory of elds.
Explaining Algebraic Geometry from scratch is not an easy task.
Of course, there are the celebrated Elements de Geometrie Algebrique by
Grothendieck and Dieudonne, four volumes of increasing size that were later
continued by seven volumes of Seminaire de Geometrie Algebrique. The series
is like an extensive encyclopaedia where the basic facts are dealt with in striv-
ing generality, but which is hard work for someone who has not yet acquired a
certain amount of expertise in the eld. To approach Algebraic Geometry
from a more economic point of view, I think it is necessary to learn about its
basic principles. If these are well understood, many results become easier to di-
gest, including proofs, and getting lost in a multitude of details can be avoided.
VI Preface
Therefore it is not my intention to cover as many topics as possible in my book.

Instead I have chosen to concentrate on a certain selection of main themes that
are explained with all their underlying structures and without making use of
any articial shortcuts. In spite of thematic restrictions, I am aiming at a self-
contained exposition up to a level where more specialized literature comes into
reach.
Anyone willing to enter Algebraic Geometry should begin with certain
basic facts in Commutative Algebra. So the rst part of the book is con-
cerned with this subject. It begins with a general chapter on rings and modules
where, among other things, I explain the fundamental process of localization, as
well as certain niteness conditions for modules, like being Noetherian or coher-
ent. Then follows a classical chapter on Noetherian (and Artinian) rings, includ-
ing the discussion of primary decompositions and of Krull dimensions, as well
as a classical chapter on integral ring extensions. In another chapter I explain
the process of coecient extension for modules by means of tensor products,
as well as its reverse, descent. In particular, a complete proof of Grothendiecks
fundamental theorem on faithfully at descent for modules is given. Moreover,
as it is quite useful at this place, I cast a cautious glimpse on categories and their
functors, including functorial morphisms. The rst part of the book ends by a
chapter on Ext and Tor modules where the general machinery of homological
methods is explained.
The second part deals with Algebraic Geometry in the stricter sense of
the word. Here I have limited myself to four general themes, each of them dealt
with in a chapter by itself, namely the construction of ane schemes, techniques
of global schemes, etale and smooth morphisms, and projective and proper
schemes, including the correspondence between ample and very ample invert-
ible sheaves and its application to abelian varieties. There is nothing really new
in these chapters, although the style in which I present the material is dierent
from other treatments. In particular, this concerns the handling of smooth mor-
phisms via the Jacobian Condition, as well as the denition of ample invertible
sheaves via the use of quasi-ane schemes. This is the way in which M. Raynaud
liked to see these things and I am largely indebted to him for these ideas.
Each chapter is preceded by an introductory section where I motivate its
contents and give an overview. As I cannot deliver a comprehensive account
already at this point, I try to spotlight the main aspects, usually illustrating
these by a typical example. It is recommended to resort to the introductory
sections at various times during the study of the corresponding chapter, in or-
der to gradually increase the level of understanding for the strategy and point
of view employed at dierent stages. The latter is an important part of the
learning process, since Mathematics, like Algebraic Geometry, consists of
a well-balanced combination of philosophy on the one hand and detailed argu-
mentation or even hard computation on the other. It is necessary to develop a
reliable feeling for both of these components. The selection of exercise problems
at the end of each section is meant to provide additional assistance for this.
Preface VII
Preliminary versions of my manuscripts on Commutative Algebra and

Algebraic Geometry were made available to several generations of students.
It was a great pleasure for me to see them getting excited about the subject, and
I am very grateful for all their comments and other sort of feedback, including
lists of typos, such as the ones by David Krumm and Claudius Zibrowius. Special
thanks go to Christian Kappen, who worked carefully on earlier versions of the
text, as well as to Martin Brandenburg, who was of invaluable help during
the nal process. Not only did he study the whole manuscript meticulously,
presenting an abundance of suggestions for improvements, he also contributed
to the exercises and acted as a professional coach for the students attending
my seminars on the subject. It is unfortunate that the scope of the book did
not permit me to put all his ingenious ideas into eect. Last, but not least,
let me thank my young colleagues Matthias Strauch and Clara Loh, who run
seminars on the material of the book together with me and who helped me
setting up appropriate themes for the students. In addition, Clara Loh suggested
numerous improvements for the manuscript, including matters of typesetting
and language. Also the gure for gluing schemes in the beginning of Section 7.1
is due to her.
Munster, December 2011 Siegfried Bosch

Contents
Part A. Commutative Algebra . . . . . . . . . . . . . . . . . . . . . 1

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1 Rings and Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

1.1 Rings and Ideals . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.2 Local Rings and Localization of Rings . . . . . . . . . . . . . . 18
1.3 Radicals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 26
1.4 Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
1.5 Finiteness Conditions and the Snake Lemma . . . . . . . . . . . 38
2 The Theory of Noetherian Rings . . . . . . . . . . . . . . . . . . . . 55

2.1 Primary Decomposition of Ideals . . . . . . . . . . . . . . . . . 57
2.2 Artinian Rings and Modules . . . . . . . . . . . . . . . . . . . . 66
2.3 The ArtinRees Lemma . . . . . . . . . . . . . . . . . . . . . . 71
2.4 Krull Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3 Integral Extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
3.1 Integral Dependence . . . . . . . . . . . . . . . . . . . . . . . . 85
3.2 Noether Normalization and Hilberts Nullstellensatz . . . . . . . 91
3.3 The CohenSeidenberg Theorems . . . . . . . . . . . . . . . . . 96
4 Extension of Coecients and Descent . . . . . . . . . . . . . . . . . 103

4.1 Tensor Products . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
4.2 Flat Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113
4.3 Extension of Coecients . . . . . . . . . . . . . . . . . . . . . . 123
4.4 Faithfully Flat Descent of Module Properties . . . . . . . . . . . 131
4.5 Categories and Functors . . . . . . . . . . . . . . . . . . . . . . 138
4.6 Faithfully Flat Descent of Modules and their Morphisms . . . . 143
5 Homological Methods: Ext and Tor . . . . . . . . . . . . . . . . . . . 157

5.1 Complexes, Homology, and Cohomology . . . . . . . . . . . . . 159
5.2 The Tor Modules . . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.3 Injective Resolutions . . . . . . . . . . . . . . . . . . . . . . . . 181
5.4 The Ext Modules . . . . . . . . . . . . . . . . . . . . . . . . . . 187
X Contents
Part B. Algebraic Geometry . . . . . . . . . . . . . . . . . . . . . . 193

Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 193
6 Ane Schemes and Basic Constructions . . . . . . . . . . . . . . . . 201

6.1 The Spectrum of a Ring . . . . . . . . . . . . . . . . . . . . . . 203
6.2 Functorial Properties of Spectra . . . . . . . . . . . . . . . . . . 212
6.3 Presheaves and Sheaves . . . . . . . . . . . . . . . . . . . . . . . 216
6.4 Inductive and Projective Limits . . . . . . . . . . . . . . . . . . 222
6.5 Morphisms of Sheaves and Sheacation . . . . . . . . . . . . . 232
6.6 Construction of Ane Schemes . . . . . . . . . . . . . . . . . . 241
6.7 The Ane n-Space . . . . . . . . . . . . . . . . . . . . . . . . . 255
6.8 Quasi-Coherent Modules . . . . . . . . . . . . . . . . . . . . . . 257
6.9 Direct and Inverse Images of Module Sheaves . . . . . . . . . . 266
7 Techniques of Global Schemes . . . . . . . . . . . . . . . . . . . . . 277

7.1 Construction of Schemes by Gluing . . . . . . . . . . . . . . . . 282
7.2 Fiber Products . . . . . . . . . . . . . . . . . . . . . . . . . . . 294
7.3 Subschemes and Immersions . . . . . . . . . . . . . . . . . . . . 304
7.4 Separated Schemes . . . . . . . . . . . . . . . . . . . . . . . . . 312
7.5 Noetherian Schemes and their Dimension . . . . . . . . . . . . . 318
7.6 Cech Cohomology . . . . . . . . . . . . . . . . . . . . . . . . . . 322
7.7 Grothendieck Cohomology . . . . . . . . . . . . . . . . . . . . . 330
8 Etale and Smooth Morphisms . . . . . . . . . . . . . . . . . . . . . . 341

8.1 Dierential Forms . . . . . . . . . . . . . . . . . . . . . . . . . . 343
8.2 Sheaves of Dierential Forms . . . . . . . . . . . . . . . . . . . . 356
8.3 Morphisms of Finite Type and of Finite Presentation . . . . . . 360
8.4 Unramied Morphisms . . . . . . . . . . . . . . . . . . . . . . . 365
8.5 Smooth Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . 374
9 Projective Schemes and Proper Morphisms . . . . . . . . . . . . . . 399

9.1 Homogeneous Prime Spectra as Schemes . . . . . . . . . . . . . 403
9.2 Invertible Sheaves and Serre Twists . . . . . . . . . . . . . . . . 418
9.3 Divisors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
9.4 Global Sections of Invertible Sheaves . . . . . . . . . . . . . . . 446
9.5 Proper Morphisms . . . . . . . . . . . . . . . . . . . . . . . . . 462
9.6 Abelian Varieties are Projective . . . . . . . . . . . . . . . . . . 477
Literature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 485
Glossary of Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495
1. Rings and Modules
The present chapter is devoted to discussing some basic notions and results
on rings and their modules. Except for a few preliminary considerations, all
rings will be meant to be commutative and to admit a unit element 1. Like
a eld, a ring comes equipped with two laws of composition, namely addition
+ and multiplication , which behave in the same way as is known from
the case of elds. The only dierence is that non-zero elements of a ring R do
not need to admit multiplicative inverses in R, a default that has far-reaching
consequences. A prominent example of such a ring is Z, the ring of integers. But
we can easily construct more intricate types of rings. Let k be a eld and write
R for the cartesian product of k with itself, i.e. R = k k. Dening addition
and multiplication on R componentwise by
(1 , 2 ) + (1 , 2 ) = (1 + 1 , 2 + 2 ),
(1 , 2 ) (1 , 2 ) = (1 1 , 2 2 ),
we see that R becomes a ring. The equation (1, 0) (0, 1) = (0, 0) implies that R
contains non-trivial zero divisors, whereas (1, 0)n = (1, 0) for n > 0 shows that
R contains idempotent elements that are dierent from the unit element (1, 1).
However, in this case there are no non-trivial nilpotent elements, i.e. elements
(1 , 2 ) dierent from (0, 0) such that (1 , 2 )n = (0, 0) for some exponent n.
On the other hand, non-trivial nilpotent elements will occur if we take
(1 , 2 ) (1 , 2 ) = (1 1 , 1 2 + 2 1 )
as multiplication on R.
For rings R of general type the notion of ideals is fundamental. An ideal
in R is just an additive subgroup a R that is stable under multiplication by
elements of R. Historically ideals were motivated by the aim to extend unique
factorization results from the ring of integers Z to more general rings of algebraic
numbers. However, as this did not work out well in the conventional setting,
Kummer invented his concept of ideal numbers, which was then generalized
by Dedekind, who introduced the notion of ideals as known today. A further
natural step is to pass from ideals to modules over rings, thereby arriving at a
simultaneous generalization of ideals in rings and of vector spaces over elds.
A very useful notion is the one of so-called polynomial rings. Let R0 be
a ring, viewed as a coecient ring, and (Xi )iI a family of symbols viewed
S. Bosch, Algebraic Geometry and Commutative Algebra, Universitext, 7

DOI 10.1007/978-1-4471-4829-6_1, Springer-Verlag London 2013
8 1. Rings and Modules
as variables. Then the polynomial ring R0 (Xi )iI consists of all nite formal
sums of formal products of type
aXin11 . . . Xinrr , a R0 , n1 , . . . , nr N,
where the indices i1 , . . . , ir I are pairwise distinct and where addition and
multiplication on this ring are dened in the conventional way. In particular,
given any polynomial f R0 (Xi )iI , the latter involves only nitely many of
the variables Xi , i.e. there are indices i1 , . . . , is I and a nite subset N Ns
such that
f= an1 ...ns Xin11 . . . Xinss , an1 ...ns R0 .
(n1 ,...,ns )N
General rings are not so far away from polynomial rings over the coecient
ring R0 = Z, since any ring R can be viewed as a quotient of a polynomial
ring of type Z(Xi )iI . To explain this fact we need to consider ring homo-
morphisms, i.e. maps between rings that respect addition and multiplication
on the source and the target, and preserve unit elements. Starting out from a
ring homomorphism : R0 - R, one can extend it to a ring homomorphism
: R0 (Xi )iI - R simply by prescribing the images of the variables Xi .
Indeed, given a family (xi )iI of elements in R, the map
aXin11 . . . Xinrr - (a)xni11 . . . xnirr
substituting xi for the variable Xi is dened on a certain part of R0 (Xi )iI and
can be extended additively to yield a well-dened map : R0 (Xi )iI - R;
the latter is a ring homomorphism, due to the commutativity of the multiplica-
tion in R. It is easily seen that any ring homomorphism : R0 (Xi )iI - R
is uniquely characterized by its restriction = |R0 to the coecient ring R0
and the values xi = (Xi ) of the variables Xi , a fact that is sometimes referred
to as the universal property of polynomial rings.
Now observe that there is a unique ring homomorphism : Z - R. The
latter can be enlarged to a surjective ring homomorphism : Z(Xi )iI - R
if we take I large enough. For example, let I = R and consider the family of
variables (Xi )iR . Then the substitution Xi - i yields a ring homomorphism
: Z(Xi )iR - R, Xi - i,
that is surjective. Its kernel is an ideal in Z(Xi )iR by 1.1/4 so that induces
an isomorphism Z(Xi )iR / ker - R by the Fundamental Theorem on
Homomorphisms 1.1/5.
For a given ring R, its so-called prime spectrum Spec R, i.e. the set of prime
ideals in R, will be of particular interest. One likes to view Spec R as a point
set, on which the elements of R live as functions. Indeed, for f R and
x Spec R let f (x) be the residue class of f in R/px , where px is a second
(more ideal-friendly) notation used instead of x. For example, the set of all
functions f R vanishing identically on Spec R equals the intersection of all
1.1 Rings and Ideals 9
prime ideals in R, and this turns out to be the nilradical rad(R), namely the ideal
consisting of all nilpotent elements in R; cf. 1.3/4. In a similar way, we can look
at the maximal spectrum Spm R consisting of all maximal ideals in R. Since any
maximal ideal is prime, this is a subset of Spec R, non-empty if R = 0. The set
of functions f R vanishing identically on Spm R is, by denition, the Jacobson
radical j(R) where, of course, rad(R) j(R). In classical algebraic geometry
one considers rings where both radicals coincide; see 3.2/5. Independently of
this, the Jacobson radical j(R) is meaningful within the context of Nakayamas
Lemma 1.4/10 or 1.4/11, addressing generators of modules over R.
Just as we can pass from the ring of integers Z to its eld of fractions Q, we
can x a multiplicatively closed subset S in a ring R and pass to the associated
ring of fractions RS , where we allow denominators being taken only from S; see
Section 1.2. The process is referred to as localization by S. For example, if p
is a prime ideal in R, its complement S = R p is a multiplicatively closed
subset in R and the localization RS is a local ring, meaning that RS contains
a unique maximal ideal, which in this case is generated by the image of p in
RS ; see 1.2/7. By construction the elements of RS can be understood as local
functions on Spec R living on certain neighborhoods of the point x Spec R
that is represented by the prime ideal p. This interpretation will become more
familiar in Chapter 6 where we start discussing basic concepts of Algebraic
Geometry. Also it explains why the process of passing to rings of fractions
is referred to as localization. As we will see already during the discussion of
radicals in Section 1.3, localization is a very useful tool in the theory of rings
and modules as well.
The chapter ends with a thorough study of niteness conditions for mod-
ules, like being of nite type, of nite presentation, as well as being Noetherian,
or coherent. For this the Snake Lemma 1.5/1 serves as a convenient technical
tool. In the world of vector spaces, all these conditions specify nite dimension
or, in other words, the existence of nite generating systems. However, for more
general modules, ner distinctions are necessary. To be on the safe side for poly-
nomial rings in nitely many variables over elds, we prove that such rings are
Noetherian in the sense that all their ideals are nitely generated; see Hilberts
Basis Theorem 1.5/14. Modules over Noetherian rings enjoy the nice property
that all the above niteness conditions are equivalent; see 1.5/12 and 1.5/13.
1.1 Rings and Ideals
Let us recall the denition of a ring.
Denition 1. A set R together with two laws of composition + (addition)

and (multiplication) is called a ring (with unity) if the following conditions
are satised :
(i) R is an abelian group with respect to addition; the corresponding zero
element is denoted by 0 R.
(ii) The multiplication is associative, i.e.
(a b) c = a (b c) for a, b, c R.
(iii) There exists a unit element in R, which means an element 1 R such

that 1 a = a = a 1 for all a R.
(iv) The multiplication is distributive over the addition, i.e. for a, b, c R
we have
a (b + c) = a b + a c, (a + b) c = a c + b c.
The ring R is called commutative if the multiplication is commutative.
We list some important examples of rings:
(1) elds,
(2) Z, the ring of integers,
(3) RX, the polynomial ring in a variable X over a commutative ring R,
(4) 0, the zero ring, which consists of just one element 1 = 0; it is the only
ring with the latter property.
An element a of a ring R is called invertible or a unit if there exists some

element b R such that ab = 1 = ba. It follows that the set
R = {a R ; a is a unit in R}
is a group with respect to the multiplication given on R.

An element a of a ring R is called a zero divisor if there exists an element
b R {0} such that ab = 0 or ba = 0. Furthermore, a commutative ring R = 0
is called an integral domain if it does not contain (non-trivial) zero divisors, i.e.
if ab = 0 with a, b R implies a = 0 or b = 0. For example, any eld is an
integral domain, as well as any subring of a eld, such as the ring of integers
Z Q. Also one knows that the polynomial ring RX over an integral domain
R is an integral domain again. However, by denition, the zero ring 0 is not an
integral domain.
For a eld K, its group of units is K = K {0}. Furthermore, we have
Z = {1, 1} and (RX) = R for an integral domain R.

Denition 2. A map : R - R between rings is called a ring homomor-

phism or a morphism of rings if for all a, b R the following conditions are
satised :
(i) (a + b) = (a) + (b),
(ii) (a b) = (a) (b),
(iii) (1) = 1.
Furthermore, mono-, epi -, iso-, endo-, and automorphisms of rings are de-
ned in the usual way. Namely, a monomorphism is meant as an injective and
an epimorphism as a surjective homomorphism.1 A subset R of a ring R is

called a subring of R if 1 R and if a, b R implies a b, a b R . Then R
is a ring itself under the addition and multiplication inherited from R.
In the following, we will exclusively consider commutative rings with 1. By
abuse of language, such rings will be referred to as rings again:
Convention. From now on, a ring is meant as a commutative ring with 1.
Denition 3. Let R be a ring. A subset a R is called an ideal in R if :

(i) a is an additive subgroup of R, i.e. a is non-empty and a b a for all
a, b a.
(ii) ra a for all r R and a a.
Any ring R contains the so-called trivial ideals, namely the zero ideal 0
consisting only of the zero element 0, and the unit ideal, which is given by
R itself. For a family (ai )iI of elements in R, we can look at the associated
generated ideal, namely

a= Rai
iI

= ri ai ; ri R, ri = 0 for almost2 all i I .
iI
This is the smallest ideal in R that contains all elements ai , i I. If the index
set I is nite, say I = {1, . . . , n}, we write

n
a= Rai = (a1 , . . . , an ).
i=1
Furthermore, an ideal a R is called nitely generated if, as before, it admits

a nite generating system. An ideal a R is called principal if it is generated
by a single element: thus, if there is some element a a such that a = (a). For
example, the trivial ideals a = 0 and a = R are principal; they are generated by
0 and 1, respectively. Furthermore, an integral domain R is called a principal
ideal domain if every ideal of R is principal. In particular, Z and the polynomial
ring KX over a eld
K are principal ideal domains, just as are the rings of
algebraic integers Z 2 and Zi, since all these rings are Euclidean domains;
see [3], Section 2.4.
There are several ways to build new ideals from known ones. Let (ai )iI be
a family of ideals of a ring R. Then
1
The notion of an epimorphism between rings is ambiguous, as it is also used in the sense
of a so-called categorical epimorphism; see Section 8.1. Such epimorphisms are not necessarily
surjective and we therefore tend to avoid the term epimorphism when dealing with surjective
ring homomorphisms.
2
In this setting, for almost all means for all up to nitely many exceptions.

ai = ai ; ai ai , ai = 0 for almost all i I
iI iI
is again an ideal in R; it is called the sum of the ideals ai . Furthermore, the

intersection
ai
iI
of all ideals ai is an ideal in R. For nitely many ideals a1 , . . . , an R we can

also construct their product
n
ai ,
i=1
which is dened as the ideal generated by all products a1 . . . an where ai ai

for i = 1, . . . , n. Clearly,
n
n
ai ai .
i=1 i=1
Finally, for ideals a, b R their ideal quotient is the ideal given by
(a : b) = {r R ; r b a}.
For the special case where a = 0 we write Ann(b) = (0 : b) and call this the
annihilator of b. If b is a principal ideal, say b = (b), we use the notations (a : b)
in place of (a : (b)) and Ann(b) in place of Ann((b)).
Proposition 4. Let : R - R be a morphism of rings. Then
ker = {r R ; (r) = 0}
is an ideal in R and im = (R) is a subring of R .
Within this context, let us briey recall the denition of residue class rings,
sometimes also referred to as quotient rings. Given an ideal a in a ring R, we
would like to construct a surjective ring homomorphism : R - R such that
ker = a. To do this, consider the set
R/a = {r + a ; r R}, where r + a = {r + a ; a a},
of all (additive) cosets of a in R and dene two laws of composition on it by
(r + a) + (r + a) := (r + r ) + a,
(r + a) (r + a) := (r r ) + a.
Of course, it has to be checked that these laws are well dened. This being done,
the dening properties of a ring follow for R/a from those of R and we see that
: R - R/a, r - r + a,
is a surjective ring homomorphism satisfying ker = a, as desired. The ring

R/a is uniquely characterized up to canonical isomorphism by the following
so-called universal property:
Proposition 5 (Fundamental Theorem on Homomorphisms). Let : R - R

be a morphism of rings and a R an ideal satisfying a ker . Then there is
a unique ring homomorphism : R/a - R such that the diagram

R - R
-

?
R/a
is commutative. Furthermore:
injective a = ker
surjective surjective
For a proof, see for example [3], 2.3/4. Let us note as a consequence:
Corollary 6. Let : R - R be a surjection between rings. Then induces

a unique isomorphism : R/ ker - R such that the diagram

R - R
-

?
R/ ker
is commutative.
Next we want to discuss prime ideals and maximal ideals.
Denition 7. Let R be a ring.

(i) An ideal p R is called prime if ab p implies a p or b p, for any
elements a, b R.
(ii) An ideal m R is called maximal if m a implies m = a, for any
proper ideal a R.
For example, the zero ideal 0 R of some ring R is prime if and only if R
is an integral domain. For a eld K, the zero ideal 0 K is prime and maximal
at the same time.
Proposition 8. Let R be a ring and a R an ideal. Then:

(i) a is prime if and only if R/a is an integral domain.
(ii) a is maximal if and only if R/a is a eld.
In particular, any maximal ideal in R is prime as well.
Proof. First, observe that p is a proper ideal in R if and only if the residue ring
R/p is non-zero, likewise for m. Then assertion (i) is easy to verify. If we write
a, b R/p for the residue classes of elements a, b R, then
abp = a p or b p
is equivalent to
ab=0 = a = 0 or b = 0.
Further, assertion (ii) is a consequence of the following lemmata:
Lemma 9. An ideal m R is maximal if and only if the zero ideal 0 R/m

is maximal.
Lemma 10. The zero ideal 0 R of a ring R is maximal if and only if R is

a eld.
Proof of Lemma 9. Let : R - R/m be the canonical homomorphism. Then

it is easily checked that the maps
R a - (a) R/m,
R 1 (b) b R/m,
dene mutually inverse bijections between the set of all ideals a R such that
m a R, and the set of all ideals b R/m. Since these maps respect
inclusions, the desired equivalence is clear.
Alternatively, the claim can be justied by explicit computation. To do
this, recall that m is a proper ideal in R if and only if the residue ring R/m
is non-zero. Now, if m is a proper ideal in R, then m is maximal if and only if
a R m implies Ra + m = R, in other words, if and only if for any such a
there are elements r R and m m such that ra + m = 1. Using the projection
: R - R/m, we see that the latter condition is satised if and only if for
any a R/m{0} we can nd an element r R/m such that r a = 1, thus, if
and only if the zero ideal in R/m is maximal.
Proof of Lemma 10. Assume that the zero ideal 0 R is maximal and consider
an element a R that is non-zero. Then we have aR = R and there is some
b R such that ab = 1. Thus, we get R = R {0} and R is a eld. Conversely,
it is clear that the zero ideal of a eld is maximal.
In order to illustrate the assertion of Proposition 8, consider a prime p Z,

in the sense that p is irreducible in Z. Then the ideal (p) Z generated by p
is maximal and, hence, also prime. Indeed, if a Z, say a = (a), is some ideal
satisfying (p) a Z, we get a | p. Since a Z, we see that a cannot be a unit
in Z. But then, as p is irreducible, a must be associated to p and we get (p) = a.
Thus, (p) Z is maximal. Furthermore, Proposition 8 says that (p) is prime as
well, and we see that Fp := Z/(p) is a eld, in fact, a eld with p elements.
To give another application of Proposition 8, let us consider the polynomial
ring KX1 , X2 over a eld K and show that the ideal generated by X2 is prime,
but not maximal. To justify this, look at the surjection

: KX1 , X2 - KX1 , aij X1i X2j - ai0 X1i ,
i,jN iN
which substitutes X1 , 0 for the variables X1 , X2 . Since KX1 is an integral

domain, but not a eld, we see from Proposition 8 that the ideal generated
by X2 in KX1 , X2 , which coincides with the kernel of , is prime but not
maximal.
Next we want to show:
Proposition 11. Any ring R = 0 contains a maximal ideal.
Proof. We will apply Zorns Lemma, which says that a non-empty partially
ordered set admits a maximal element, provided every totally ordered subset
of admits an upper bound in .
In our case, we dene as the set of all proper ideals a R, using the set
inclusion as order relation. Then, R being non-zero, it contains the zero ideal as
a proper ideal and, hence, = . Now let be a totally ordered subset,
in the sense that for a, a we have a a or a a. In order to construct
an upper bound of in , we may assume non-empty. Then we claim that

b := a
a
is a proper ideal in R. In order to justify this, observe that b contains at least

one ideal of R and, therefore, is non-empty as was assumed to be non-empty.
Now consider elements a, a b, say a a and a a where a, a . Since
is totally ordered, we may assume a a . Then a a a b and also
ra a b for all r R. Hence, b is an ideal in R. Of course, b cannot coincide
with R, since otherwise we would have 1 b and, thus, 1 a for some a ,
which, however, is excluded. As a consequence, we have b and, by its
construction, b is an upper bound of . Thus, all conditions of Zorns Lemma
are satised and contains a maximal element. In other words, R contains a
maximal ideal.
Corollary 12. Let R be a ring and a R a proper ideal in R. Then there

exists a maximal ideal m R such that a m.
Proof. Apply Proposition 11 to the ring R/a, which is non-zero due to a R.

Then the preimage of any maximal ideal n R/a with respect to the projection
R - R/a is a maximal ideal in R containing a.
Corollary 13. Let R be a ring and a R a non-unit. Then there exists a

maximal ideal m R such that a m.
Proof. Apply Corollary 12 to the principal ideal a = (a).
For a ring R the set
Spec R = {p R ; p prime ideal in R}
is called the spectrum or, in more precise terms, the prime spectrum of R.
Likewise, the subset
Spm R = {m R ; m maximal ideal in R}
is called the maximal spectrum of R. It is via such spectra that geometrical

methods become applicable to ring theory. This is the main theme in Alge-
braic Geometry, as we will see from Section 6.1 on. In this setting, Spec R
and Spm R are interpreted as sets of points on which the elements of R can be
evaluated in the sense of functions.
We want to sketch the basics of this approach in brief terms. For f R and
x Spec R dene f (x) as the residue class of f in R/px . Here px is a second
notation instead of x, just to remember that the point x is, in reality, a prime
ideal in R. This way, elements f R can be interpreted as functions

f : Spec R - R/px .
xSpec R
In particular, assertions like f (x) = 0 or f (x) = 0 make sense and just mean
f px or f px . For any ideal a R we can look at the set
V (a) = {x Spec R ; f (x) = 0 for all f a}
of all common zeros of elements of a.
Proposition 14. For any ring R, there exists a unique topology on Spec R, the
so-called Zariski topology, such that the sets of type V (a) for ideals a R are
precisely the closed subsets of Spec R.
As a consequence, we can see that the sets of type D(f ) = Spec R V (f ),

for elements f R, form a basis of the Zariski open subsets of Spec R. For
proofs of these facts and further details we refer to Section 6.1. Let us add:
Remark 15. Let : R - R be a morphism of rings. Then induces a map

a
: Spec R - Spec R , p - 1 (p),
which is continuous with respect to Zariski topologies on Spec R and Spec R .
For the proof see 6.2/5. At this place, let us just point out that the map a
is well-dened. Indeed, for any ideal p R there is a commutative diagram

R - R
? ?

R /1 (p) - R/p .
Now if p is prime, we see from Proposition 8 that R/p and, hence, R /1 (p) are
integral domains. But then 1 (p) must be prime, again by Proposition 8.
Exercises
1. Let : R - R be a ring homomorphism and m Spm(R ). Can we conclude
1 (m) Spm(R)?
2. Prove that an integral domain is a eld as soon as it contains only nitely many
elements. Deduce that in a nite ring every prime ideal is maximal.
3. Prove the Chinese Remainder Theorem: Let R be a ring with ideals a1 , . . . , an
satisfying ai + aj = R for i = j. Then there is a canonical isomorphism

n
n
R/ ai R/ai
i=1 i=1
where the cartesian product of the rings R/ai is viewed as a ring under compo-
nentwise addition and multiplication.
. . . , ar Rsatisfying ai + aj = R for i = j. Show
4. Consider a ring R and ideals a1 ,
that, in this case, the inclusion ni=1 ai ni=1 ai is an equality.
5. Let R be a principal ideal domain.
(a) Give a characterization of Spec R and of Spm R.
(b) Give a characterization of the ideals in R/(a) for any element a R.
6. Let R1 , . . . , Rn be rings and consider the cartesian product R1 . . . Rn as a
ring under componentwise addition and multiplication. Show:
(a) Given ideals ai Ri for i = 1, . . . , n, the cartesian product a1 . . . an is an
ideal in R1 . . . Rn .
(b) Each ideal in R1 . . . Rn is as specied in (a).
(c) There is a canonical bijection

n
Spec(R1 . . . Rn ) - Spec Ri
i=1
and a similar one for spectra of maximal ideals.

1.2 Local Rings and Localization of Rings
Denition 1. A ring R is called local if it contains precisely one maximal ideal

m. The eld R/m is called the residue eld of the local ring R.
Proposition 2. Let R be a ring and m R a proper ideal. The following

conditions on m are equivalent:
(i) R is a local ring with maximal ideal m.
(ii) Every element of R m is a unit in R.
(iii) m is a maximal ideal and every element of type 1 + m with m m is a
unit in R.
Proof. We start by showing that conditions (i) and (ii) are equivalent. If (i)
holds and a R is not a unit, we can use 1.1/13 to conclude that there is a
maximal ideal n R with a n. Necessarily, n must coincide with m. Therefore
the complement R m consists of units and we get (ii). Conversely, if (ii) holds,
every proper ideal a R will be contained in m, since proper ideals cannot
contain units. In particular, m is a unique maximal ideal in R and we get (i).
Next assume (ii) again. Then m is a maximal ideal by (i), and we see for
every m m that 1 + m cannot be contained in m, since 1 m. Thus, by our
assumption, 1 + m is a unit and we have (iii). Conversely, assume (iii) and let
x R m. Since m is a maximal ideal, x and m will generate the unit ideal in
R. Hence, there exists an equation
1 = ax m
with elements a R and m m. Then ax = 1 + m is a unit by (iii) and the

same is true for x so that (ii) holds.
Every eld K is a local ring with maximal ideal 0 K. Further examples

of local rings are provided by discrete valuation rings, which can be viewed
as principal ideal domains R containing just one prime element p R (up to
multiplication by units). In such a ring, (p) R is the only maximal ideal.
To give an explicit example of a discrete valuation ring R, x a prime p N
and consider m
Z(p) = Q ; m, n Z with p n Q
n
as a subring of Q. Then Z(p) is an integral domain, and we claim that Z(p) is, in
fact, a principal ideal domain. To show that any ideal a Z(p) is principal, look
at its restriction a = a Z, which is an ideal in Z. As Z is principal, there is an
element a Z with a = aZ, and we get a = aZ(p) , because we may interpret a
as the set of numerators of fractions m n
a where p n. Next we want to show
that Z(p) contains precisely one maximal ideal and that the latter is generated
by p. To justify this, observe that p1 does not belong to Z(p) and, hence, that p
is not invertible in Z(p) . Therefore pZ(p) is a proper ideal in Z(p) , and we claim
1.2 Local Rings and Localization of Rings 19
that its complement Z(p) pZ(p) consists of units in Z(p) . Any element in Z(p)
can be written as a fraction m n
with p n, and such a fraction satises p m if it
does not belong to pZ(p) . But then ( mn
)1 = mn
Z(p) and mn
is a unit. Therefore
all elements of Z(p) pZ(p) are invertible, and it follows from Proposition 2 (ii)
that Z(p) is a local ring with maximal ideal pZ(p) . In particular, p is a prime
element in Z(p) , in fact, up to multiplication by a unit the only prime element
existing in Z(p) . Indeed, a prime element of Z(p) cannot be invertible and, hence,
must belong to pZ(p) , which means that it is divisible by p. Looking at prime
decompositions of elements in Z(p) , we see that the ideals in Z(p) are precisely
the ones occurring in the chain
Z(p) pZ(p) p2 Z(p) . . . 0.
In general, local rings can be constructed by a localization process, a method

we want to explain now. Also note that Z(p) as constructed above can be inter-
preted as a localization of Z.
Let R be a ring and S R a multiplicative system, i.e. a subset in R
satisfying 1 S and s, s S = ss S. Then, roughly speaking, we want
to consider the ring of all fractions rs , where r R and s S. The latter is
called the localization of R by S. To make sense of such fractions as elements
of a ring, a bit of care is necessary. Guided by the idea that two fractions as

and as with a, a R and s, s S should specify the same element in such a
localization as soon as as = a s and that, in particular, any two fractions of
type as and at
st
for t S should coincide, we dene a relation on R S by
(a, s) (a , s ) there is some t S such that (as a s)t = 0,
claiming that this is an equivalence relation. Evidently, the relation is reexive

(just choose t = 1) and symmetric. To justify transitivity, assume
(a, s) (a , s ), (a , s ) (a , s )
for pairs (a, s), (a , s ), (a , s ) R S. Then there are t, t S such that
(as a s)t = 0, (a s a s )t = 0,
and it follows that
(as a s)s s tt = (as a s)s2 tt + (a s a s )ss tt = 0.
Hence, (a, s) (a , s ), as required.

Now consider the set (R S)/ of all equivalence classes in R S under
the relation and write as for the class of an element (a, s) R S. We
claim that (R S)/ is a ring, using the standard addition and multiplication
of fractional arithmetic
a b at + bs a b ab
+ = , = .
s t st s t st
To justify this claim, we start by showing that these laws are well-dened. Thus,
consider two elements
a a b b
= , = (R S)/
s s t t
with dierent representatives (a, s), (a , s ) and (b, t), (b , t ) in RS. Then there
are elements u, v S such that
(as a s)u = 0, (bt b t)v = 0,
and it follows

(at + bs)s t (a t + b s )st uv
= (as a s)u vtt + (bt b t)v uss = 0,
as well as
(abs t a b st)uv = (as a s)u bt v + (bt b t)v a su = 0.
This means
at + bs a t + b s ab a b
=
, = ,
st st st st
and we see that addition and multiplication on (R S)/ are well-dened.
Furthermore, it is easily checked that (R S)/ is a ring, with 01 serving as the
zero element and 11 as the unit element. Using a more handy notation, we will
write RS (or sometimes also S 1 R) instead of (R S)/.
Denition 3. Let R be a ring and S R a multiplicative system. Then RS is

called the localization of R by S.
In the situation of the denition, the canonical map
- -
a
: R RS , a ,
1
is a homomorphism of rings and we will often write a instead of (a) = a1 . In
using this notation a bit of care is required, because will not be injective in
general, so that we may not be able to interpret R as a subring of RS . However,
common factors from S in numerators and denominators of fractions can be
canceled as usual; for a R and s, t S we have
a at
= ,
s st
because ast ats = 0.
Remark 4. Let R be a ring with a multiplicative system S R and let

: R - RS be the canonical map. Then:
(i) ker = {a R ; as = 0 for some s S}.
(ii) (s) = 1s is a unit in RS for all s S.

(iii) RS = 0 0 S.
(iv) is bijective if S consists of units in R.
Proof. (i) According to the denition of RS , we have a1 = 0 = 01 for a R if and

only if there is some s S satisfying as = (a 1 0 1)s = 0.
(ii) For s S we can write 1s 1s = ss = 1; hence, ( 1s )1 = 1s .
(iii) RS is the zero ring if and only if 11 = 01 , i.e. if and only if there exists
some s S such that s = (1 1 0 1)s = 0.
(iv) If S contains only units of R, we see ker = 0 by (i). Furthermore
a as1
=
s 1
shows that is surjective then.
We want to look at some examples.
(1) For an integral domain R set S := R {0}. Then the canonical map
R - RS is injective and we may view R as a subring of RS . Since all non-
zero elements of RS are invertible, Q(R) := RS is a eld, the so-called eld of
fractions of R. For example, we have Q(Z) = Q. For the polynomial ring KX
in one variable X over a eld K, we obtain as its eld of fractions the so-called
rational function eld in the variable X over K, which is denoted by K(X).
(2) Consider a ring R = 0 and let S = R Z where Z is the set of all zero
divisors in R. Then RS is called the total quotient ring of R. It is not necessarily
a eld, but contains R as a subring.
(3) Consider a ring R and a prime ideal p R. Then S = R p is a
multiplicative system in R, due to the prime ideal property of p, and we call
RRp the localization of R at p. By abuse of language, one mostly writes Rp
instead of using the more clumsy notation RRp .
(4) For an element f of a ring R, the set S = {1, f, f 2 , f 3 , . . .} denes a
multiplicative system in R. The localization RS is denoted by Rf or Rf 1 .
In the following, let RS be the localization of a ring R by a multiplicative

system S R, and let : R - RS be the canonical map. For any ideal
a R, we can consider its extension to RS , which is denoted by aRS . Thereby
we mean the ideal generated by (a) in RS . A simple verication shows
a
aRS = ; a a, s S .
s
On the other hand, we may consider the restriction of any ideal b RS to R,
which is given by b R = 1 (b).
Proposition 5. Let RS be the localization of a ring R by a multiplicative system

S R. Then:
(i) An ideal a R extends to a proper ideal aRS RS if and only if

S a = .
(ii) For any ideal b RS , its restriction a = b R satises aRS = b.
(iii) If p R is a prime ideal such that p S = , then the extended ideal
pRS is prime in RS and satises pRS R = p.
(iv) For any prime ideal q RS , its restriction p = q R is a prime ideal
in R satisfying pRS = q. In particular, p S = by (i).
Proof. Starting with assertion (i), let a R be an ideal containing some s S.

Then aRS contains 1s , which is a unit, and we have aRS = RS . Conversely,
assume aRS = RS . Then there are elements a a and s S such that as = 11
in RS . Hence, we can nd t S such that (a s)t = 0. But then st = at a,
and a S = .
To show aRS = b in the situation of (ii), consider an element as b. Then
a
1
= 1s as implies a a = b R and, hence, as aRS so that b aRS . The
opposite inclusion is trivial.
Next consider a prime ideal p R such that p S = . Then pRS is a

proper ideal in RS by (i). To show that it is prime, consider elements as , as RS

such that as as = aa
ss
pRS , say aa
ss
= as for some a p and s S. Then there
is an equation
(aa s a ss )t = 0
for some t S and we get aa s t = a ss t p. Since s t p, we have aa p

and, thus, a p or a p. Therefore as pRS or as pRS and we see that
pRS is prime in RS . Clearly p pRS R. To verify the opposite inclusion, let
b pRS R. Then there is an equation 1b = as in RS , for some a p, s S.
Hence, there is t S such that (bs a)t = 0. Thus, bst = at p and, since
st p, we get b p.
Finally, let q be a prime ideal in RS . It follows from 1.1/15 that p = q R
is prime in R. To show pRS = q use the argument given in (ii).
One may ask if assertion (iii) of Proposition 5 still holds for more general
ideals than prime ideals. The answer is yes for so-called primary ideals by 2.1/13,
but no in general, as can be seen using 2.1/14 or Exercise 4 below.
Corollary 6. The canonical homomorphism R - RS from a ring R to its

localization by a multiplicative system S R induces a bijection
Spec RS - {p Spec R ; p S = }, q - q R,
that, together with its inverse, respects inclusions between prime ideals in R and
RS .
Corollary 7. For any prime ideal p of a ring R, the localization Rp is a local

ring with maximal ideal pRp .
Proof. We just have to observe that Rp pRp consists of units in Rp .

It remains to discuss the so-called universal property of localizations, which

characterizes localizations up to canonical isomorphism.
Proposition 8. The canonical homomorphism : R - RS from a ring R to

its localization by a multiplicative system S R satises (S) (RS ) and is
universal in the following sense: Given any ring homomorphism : R - R

such that (S) (R ) , there is a unique ring homomorphism : RS - R
such that the diagram

R - RS

?

R
is commutative.
Furthermore, if : R - R satises the same universal property as

does, then : RS - R is an isomorphism.

Proof. We start by the uniqueness assertion. For a R and s S, we have

a a s a
(a) = = = (s)
1 s1 s
and, hence, a
= (a)(s)1 ,
s
which implies that the map is uniquely determined by .
To show the existence of a homomorphism : RS - R as required, set
a
= (a)(s)1
s
a
for a R and s S. We claim that ( as ) is well-dened. Indeed, for a
s
= s
,
say where (as a s)t = 0 for some t S, we get

(a)(s ) (a )(s) (t) = 0
and, since (t) is a unit in R , also
(a)(s ) (a )(s) = 0.
The latter is equivalent to
(a)(s)1 = (a )(s )1 .
Hence, : RS - R is well-dened, and it is easily checked that is a

homomorphism satisfying = .
Now assume that both and are universal in the sense of the assertion.
Then besides : RS - R satisfying = there is a homomorphism
: R - RS such that = . Applying the uniqueness part of the

universal property to the equations
idR = = ( ) , idRS = = ( ) ,
we conclude = idR as well as = idRS . Consequently, : RS - R

is an isomorphism.
As an example of how to work with the above universal property, let us give
an alternative description of localizations.
Lemma 9. Let R be a ring, F = (fi )iI a family of elements in R, and S R

the multiplicative subset generated by F , i.e. S consists of all (nite) products
of members in F . Then, xing a system of variables T = (ti )iI , there is a
canonical isomorphism
RS - RT /(1 fi ti ; i I).
In particular, for a single element f R, there is a canonical isomorphism

Rf Rt/(1 f t).
Proof. The canonical ring homomorphism : R - RT /(1 fi ti ; i I)

sends all elements of S to units. Thus, it factorizes over a well-dened ring
homomorphism : RS - R T /(1 fi ti ; i I). On the other hand, it is
easily checked that satises the universal property of a localization of R by
S. Therefore must be an isomorphism.
We want to discuss some simple compatibility properties for localizations.
Proposition 10. Consider elements f, g R of some ring R and integers

d, e N where d 1. Then there is a canonical commutative diagram
R - Rf
? ?
-
Rf g (Rf )f e gd
where the lower horizontal map is an isomorphism.
Proof. The canonical map R - Rf - (Rf )f e gd sends f , g and, thus, also f g

to units in (Rf )f e gd and therefore factors through a well-dened homomorphism
Rf g - (Rf )f e gd , which is the lower horizontal map of the diagram. To obtain
an inverse of the latter, observe that R - Rf g maps f to a unit and, hence,
factors through a map Rf - Rf g . Since this map sends f e g d to a unit in
Rf g , it factors through a map (Rf )f e gd - Rf g . The latter is easily seen to
be an inverse of the lower one in the diagram, either by relying on the universal
property of localizations, or by using fractional arithmetic.
Proposition 11. Consider a prime ideal p R of some ring and an element

f Rp. Then pRf is a prime ideal in Rf , and there is a canonical commutative
diagram
R - Rf
? ?
-
Rp (Rf )pRf
where the lower horizontal map is an isomorphism.
Proof. It follows from Proposition 5 that pRf is a prime ideal in Rf . Furthermore,

the same result says that the canonical homomorphism R - Rf maps R p
into Rf pRf . In particular, the composition R - Rf - (Rf )pRf maps Rp
into the multiplicative group of units in (Rf )pRf and therefore factors through
a well-dened homomorphism Rp - (Rf )pR , which is the one considered in
f
the diagram. To obtain an inverse of the latter, observe that R - Rp factors
through Rf , since the image of f is a unit in Rp . Furthermore, the induced map
Rf - Rp sends all elements of Rf pRf to units in Rp and therefore induces
a well-dened homomorphism (Rf )pRf - Rp . The latter is easily seen to be
an inverse of the lower one in the diagram. Similarly, as in the proof above, we
may rely on the universal property of localizations, or simply apply fractional
arithmetic.
Note that the same argument of proof works for any localization RS in place
of Rf if the condition f R p of Proposition 11 is replaced by p S = .
Exercises
1. Let R be a non-zero ring. Show that R is local if and only if 1 + a is a unit
for every non-unit a R. In the latter case the maximal ideal of R is given by
R R .
2. For a ring R and a maximal ideal m R show that R/mn is a local ring for any
exponent n 1.
3. Show that every subring of Q is a localization of Z.
4. Consider the polynomial ring KX, Y in two variables over a eld K and its
ideal a = (X 2 , XY ), as well as its multiplicative subset S = KX, Y (X).
Show that the ideal aKX, Y S KX, Y strictly contains a, in contrast to the
assertion of Proposition 5 (iii) for prime ideals.
5. For a ring R and a prime ideal p R show that there is a canonical isomorphism
Q(R/p) - Rp /pRp where Q(R/p) is the eld of fractions of R/p.
6. Show that any localization of a factorial ring is factorial again; factorial means
that the ring satises the theorem of unique prime factor decomposition.
7. Show for any ring R and a variable X that the localization (RX)S of the
polynomial ring RX by the multiplicative system S = {1, X, X 2 , . . .} is the
so-called ring of Laurent

polynomials over R; the latter consists of all formal
expressions of type iZ ai X i where ai R and ai = 0 for almost all i I,
endowed with conventional addition and multiplication.
8. Consider the polynomial ring RX in a family of variables X over a ring R.
Show for a multiplicative subset S R that there is a canonical isomorphism
(RX)S - RS X.
9. Call a ring homomorphism : R - R a localization morphism if there exists
a multiplicative subset S R with (S) R such that satises the universal
property of a localization of R by S. Show that the composition of two localization
morphisms is a localization morphism again.
1.3 Radicals
Denition 1. Let R be a ring. The intersection

j(R) = m
mSpm R
of all maximal ideals in R is called the Jacobson radical of R.
As an intersection of ideals, the Jacobson radical j(R) is an ideal in R

again. If R is the zero ring, it makes sense to put j(R) = R, since an empty
intersection of ideals in a ring R equals R by convention. Let us consider some
further examples. Clearly, a ring R is local if and only if its Jacobson radical
j(R) is a maximal ideal. Furthermore, we claim that the Jacobson radical of a
polynomial ring in nitely many variables X1 , . . . , Xn over a eld K is trivial,
j(KX1 , . . . , Xn ) = 0.
This is a special case of Hilberts Nullstellensatz; see 3.2/5 or 3.2/6. To give a

simple argument for this at the present stage, let K be an algebraic closure of
K. Then, for a point x = (x1 , . . . , xn ) K n , we may consider the ideal
mx = {f KX1 , . . . , Xn ; f (x) = 0},
which is the kernel of the substitution homomorphism
KX1 , . . . , Xn - K, g - g(x).
As the latter is surjective onto the eld K(x1 , . . . , xn ) generated over K by the
components of x, we see that KX1 , . . . , Xn /mx is a eld. Therefore it follows
from 1.1/8 (ii) that mx is a maximal ideal in KX1 , . . . , Xn .
Now let f j(KX1 , . . . , Xn ). Then f mx and, hence, f (x) = 0 for
all x K n . From this we can conclude by induction on n that f is the zero
polynomial. Indeed, the case n = 1 is clear, since any algebraically closed eld
1.3 Radicals 27
contains innitely many elements and, hence, f is a polynomial in one variable

that has an innite number of zeros. On the other hand, if n > 1 we can write f
as a polynomial in Xn with coecients in KX1 , . . . , Xn1 , say f = di=0 fi Xni .
Fixing an arbitrary point x K n1 , the polynomial

d
f (x , Xn ) = fi (x )Xni KXn
i=0
vanishes at all points xn K and, as before, must have coecients fi (x ) that
are trivial. But then, varying x , the polynomials fi will vanish on all points of
K n1 so that the fi must be trivial by the induction hypothesis. In particular,
f = 0 and this shows that the Jacobson radical of KX1 , . . . , Xn is trivial, as
claimed. For a dierent method of proof see Exercise 5 below.
Remark 2. Let R be a ring. For a R the following are equivalent:

(i) a j(R).
(ii) 1 ab is a unit in R for all b R.
Proof. Let a j(R). Then a is contained in every maximal ideal of R, and it

follows that 1 ab for b R cannot be a member of any of these ideals. Since
every non-unit is contained in a maximal ideal of R by 1.1/13, we see that 1ab
must be a unit.
Conversely, consider an element a R j(R). Then there is a maximal
ideal m R such that a m. Hence, (m, a) = R and there is an equation of
type 1 = m + ab, for suitable elements m m and b R. Since 1 ab = m m,
we have found an element b R such that 1 ab is not a unit in R.
Denition 3. Let R be a ring. An element a R is called nilpotent if as = 0

for some s N. Furthermore,
rad(R) = {a R ; a nilpotent}
is called the nilradical of R. If rad(R) = 0, the ring R is called reduced.
As a, b rad(R) implies a b rad(R) using the binomial theorem, it

becomes clear that rad(R) is an ideal in R. For example, consider the ring
R = KX/(X s ), for a eld or even a ring K and some integer s > 0. Writing
X R for the residue class of the variable X, we get X s = 0 and it follows
that X R is nilpotent. Assuming rad(K) = 0 (in the case of a ring K), one
can show that rad(R) = (X). In general the ideal rad(R) will be generated by
rad(K) and X.
Proposition 4. For any ring R we have

rad(R) = p.
pSpec R

Proof. Let us start by showing rad(R) pSpec R p. To achieve this, choose an
element a rad(R) and assume as = 0 for some s N. Then as is a member
prime ideal p R, and it follows that a itself must belong to p. Thus,
of any
a pSpec R p.
To derive the opposite inclusion, consider an element a R rad(R). We
have to show that there exists a prime ideal p R such that a p. To construct
such a prime ideal, look at the multiplicative system
S = {an ; n N} R.
By its choice, a is not nilpotent. Therefore 0 S and this implies RS = 0 by

1.2/4 (iii) for the localization of R by S. Then RS will contain a maximal ideal
q by 1.1/11, which, in particular, is a prime ideal. Considering the canonical
homomorphism : R - RS , we claim that p = 1 (q), which is a prime ideal
in R, cannot contain a. Indeed, otherwise q would contain (a) as a unit and
therefore the prime ideal q would coincide with the unit ideal in RS , which is
impossible. Alternatively, we can use 1.2/5 (iv).
Comparing Proposition 4 with Denition 1, we see that the Jacobson and

the nilradical are characterized in a similar way. In particular, the nilradical
is always contained in the Jacobson radical. The two radicals will be dierent
in general, which becomes most apparent by looking at local rings admitting a
non-maximal prime ideal. For example, Z(p) for a prime p is such a ring. We
want to extend the notion of radicals to ideals.
Denition 5. Let R be a ring and a R an ideal. Then

j(a) = m
mSpm R
am
is called the Jacobson radical of a and
rad(a) = {a R ; an a for some n N}
is called the nilradical of a. If rad(a) = a, the ideal a is called reduced.
Note that for the unit ideal a = R the radicals j(R) and rad(R) in the
sense of Denition 5 are not really signicant. This is why these notions are
predominantly used for the radicals of the ring R as introduced in Denitions 1
and 3.
Remark 6. Let R be a ring and a R an ideal. Let : R - R/a be the

canonical residue homomorphism.
Then:
(i) j(a) = 1 j(R/a)
.
1
(ii) rad(a) = rad(R/a) .
(iii) rad(a) = pSpec R, ap p.
1.3 Radicals 29
Proof. To justify assertions (i) and (iii) observe that the map n - 1 (n)
denes a bijection between all maximal (resp. prime) ideals of R/a and the
maximal (resp. prime) ideals in R that contain a. Therefore (i) and (iii) follow
from Denition 1 and Proposition 4; just use the fact that the formation of
inverse images with respect to commutes with intersections. For (ii) use the
fact that a power bn of some b R belongs to a if and only if (b)n = 0.
Assertion (iii) of Remark 6 admits a geometric interpretation on the prime

spectrum Spec R, as introduced in Section 1.1. To explain this, associate to any
subset Y Spec R the ideal
I(Y ) = {f R ; f (x) = 0 for all x Y } R
of functions in R that vanish onY . Then we have I(x) = px for any point

x Spec R and therefore I(Y ) = xY px . It follows

I V (a) = px = p = rad(a),
xV (a) pSpec R
ap
namely, that the nilradical rad(a) coincides with the ideal of all functions in R
that vanish on the zero set of a.
Finally, let us add two technical lemmata on prime ideals.
Lemma 7. Let R be a ring, a R an ideal, and p1 , . . . , pn R prime ideals

such that a nj=1 pj . Then there is an index i, 1 i n, such that a pi .
Lemma 8. Let R be a ring, a1 , . . . , an R ideals, and p R a prime ideal

such that nj=1 aj p. Then there is an index i, 1 i n, such that ai p.
If, more specically, nj=1 aj = p, then ai = p for this index i.
Proof of Lemma 7. We conclude by induction on n, the case n = 1 being trivial.

Let n > 1 and suppose we have a nj=1 pj , but not a pi for any index
i = 1, . . . , n. By induction hypothesis we may then assume a j=i pj for
i = 1, . . . , n. Thus, for any i, we can nd some element

ai a pj pi .
j=i
Then consider the elements

bi := aj pj , i = 1, . . . , n,
j=i j=i
where bi belongs to a and for j = i also to pj , but not to pi , as can be seen by

using the prime ideal property of pi . Therefore

n
b := bj
j=1
is anelement of a that cannot belong to any of the p1 , . . . , pn . But this means

a nj=1 pj , contradicting our assumption.
n
Proof of Lemma 8. Proceeding indirectly, as before, let us assume j=1 aj p,
but aj p for j = 1, . . . , n. Then there are elements
aj aj p, j = 1, . . . , n,
and using the prime ideal property of p we get

n
a := a1 . . . an aj p,
j=1
n
which contradicts the inclusion j=1 aj p.

Therefore there is an index i, 1 i n, such that ai p. If nj=1 aj = p,
we get ai = p from this for trivial reasons.
Let us point out that Lemma 8 admits a geometric interpretation on the

prime spectrum of R, as introduced in Section 1.1. Namely, for ideals a1 , . . . , an
and a prime ideal p of R, we have the equivalence

n
n
aj p V aj V (p).
j=1 j=1
Indeed, the implication = is obtained by looking at the zeros of the ideals

involved, whereas the implication = uses the formation of ideals I()
of vanishing functions, as introduced above, in conjunction with rad(p) = p.
Furthermore, we can use the assertion of Lemma 8 to show

n n
V aj = V (aj ).
j=1 j=1

Thus, given an inclusion V (p) nj=1 V (aj ), Lemma 8 yields the existence of
an index i, 1 i n, such that V (p) V (ai ). This property characterizes the
so-called irreducibility of V (p) for prime ideals p.
Exercises
1. Let R be a ring and set Rred = R/ rad(R). Show that Rred is reduced and that
any ring homomorphism R - R to a reduced ring R factors through a unique
ring homomorphism Rred - R .
2. Consider the polynomial ring KX, Y in two variables over a eld K and set
R = KX, Y /(X XY 2 , Y 3 ). Writing X, Y for the residue classes of X, Y , show
that rad(R) = (X, Y ) is the only prime ideal in R and that Rred K.
1.4 Modules 31
3. Let R be a ring containing only nitely many prime ideals and assume that a
certain power of the radical rad(R) is zero. Show that R is a cartesian product
of local rings. Hint: Apply the Chinese Remainder Theorem of Exercise 1.1/3 in
conjunction with Exercise 1.1/4.
4. Show rad(RX) = rad(R) RX for the polynomial ring RX in one variable
over a ring R. Furthermore, prove
RX = R + rad(R) (X)
where
d the iright-hand side is to be interpreted as the set of all polynomials of type

i=0 ai X for variable d where a0 R and ai rad(R) for i > 0. Hint: Think
of the geometric series.
5. Show rad(RX) = j(RX) for the polynomial ring in one variable over any
ring R. Hint: Use Exercise 4 above.
1.4 Modules
Denition 1. Let R be a ring. An R-module consists of a set M together with

an inner composition law M M - M , (a, b) - a + b, called addition,
and an external composition law R M - M , (, a) - a, called scalar
multiplication, such that:
(i) M is abelian group with respect to addition.
(ii) ( + ) a = a + a and (a + b) = a + b for all , R,
a, b M , i.e. addition and scalar multiplication satisfy distributivity.
(iii) ()a = ( a) for all , R, a M , i.e. the scalar multiplication
is associative.
(iv) 1 a = a for the unit element 1 R and all a M .
Modules should be seen as a natural generalization of vector spaces. In

particular, a K-module over a eld K is just a K-vector space. On the other
hand, any ideal a of a ring R can be considered as an R-module. Just view a as a
group with respect to the addition of R and use the multiplication of R in order
to dene a scalar multiplication of R on a. In particular, R is a module over
itself. Also note that there is a (1 : 1)-correspondence between abelian groups
and Z-modules.
Denition 2. A map : M - N between R-modules M and N is called a

morphism of R-modules or an R-module homomorphism (or just an R-homo-
morphism if the context of modules is clear ) if
(i) (x + y) = (x) + (y) for all x, y M ,
(ii) (rx) = r(x) for all r R and x M .
Mono-, epi -, iso-, endo-, and automorphisms of R-modules are dened as

usual.
It is possible to combine the concept of modules with the one of rings; we

thereby arrive at the notion of algebras.
Denition 3. Let R be a ring. An R-algebra (associative, commutative, and

with a unit 1) is a ring A equipped with a structure of an R-module such that
the compatibility rule
r (x y) = (r x) y = x (r y)
holds for all r R and x, y A where denotes both, the ring multiplication
and the scalar multiplication on A.
A morphism of R-algebras A - B is a map that is a homomorphism
with respect to the ring and the module structures on A and B.
Given any ring homomorphism f : R - A, we can easily view A as an

R-algebra via f ; just set r x = f (r) x for r R and x A. Conversely, for
any R-algebra A as in the denition, the map
f: R - A, r - r 1A ,
where 1A is the unit element of A, denes a ring homomorphism such that the
R-algebra structure of A coincides with the one induced from f . In fact, to
equip a ring A with the structure of an R-algebra in the sense of the denition
we just have to specify a ring homomorphism R - A, which then is called
structural. Using this point of view, a homomorphism between two R-algebras
R - A and R - B is a ring homomorphism A - B compatible with
structural homomorphisms in the sense that the diagram
A - B

R
is commutative.
Returning to the notion of R-modules, we need to introduce the concept of
submodules.
Denition 4. Let M be an R-module. A subgroup N M is called a submodule

or, in more precise terms, an R-submodule of M if rx N for all r R and
x N . In particular, N is then an R-module itself, using the addition and scalar
multiplication inherited from M .
For example, the R-submodules of any ring R consist precisely of the ideals
in R.
Remark 5. Let : M - N be a morphism of R-modules. Then
ker := {x M ; (x) = 0},

1.4 Modules 33
the kernel of , is an R-submodule of M , and
im := (M ),
the image of , is an R-submodule of N .
If M is an R-module and N M an R-submodule, then just as in the

setting of vector spaces, one can construct the residue class module M/N , often
referred to as the quotient of M by N . It consists of all cosets x + N of N in
M where x varies over M . Then M/N is an abelian group under the addition
(x + N ) + (y + N ) = (x + y) + N, x, y M,
and even an R-module if we take
r(x + N ) = (rx) + N, r R, x M,
as scalar multiplication. These composition laws are well-dened, as can be

checked easily, and the canonical map
: M - M/N, x - x + N,
is an epimorphism of R-modules, which satises the universal property of the

Fundamental Theorem on Homomorphisms, namely:
Proposition 6. Let M be an R-module and N M an R-submodule. If

: M - M is a morphism of R-modules satisfying N ker , there exists
a unique R-module homomorphism : M/N - M such that the diagram

M - M
-

?
M/N
is commutative.
In addition, let us point out that is injective if and only if N = ker , and
surjective if and only if is surjective. In particular, we can conclude that is
an isomorphism if and only if N = ker and is surjective. As consequences
of this fact, let us mention the so-called Isomorphism Theorems.
Proposition 7. Let N, N be submodules of an R-module M . Then the canonical

homomorphism N - N + N - (N + N )/N admits N N as its kernel
and is surjective. Hence, it induces an isomorphism
N/(N N ) - (N + N )/N .
Proposition 8. Let M be an R-module and N N M submodules.

(i) The canonical homomorphism N - M - M/N admits N as kernel
and induces a monomorphism N /N - M/N so that N /N can be viewed as
a submodule of M/N .
(ii) The canonical epimorphism M - M/N factorizes over M/N , i.e.
f
it can be written as a composition M - M/N - M/N where f is an
R-module homomorphism and the canonical map.
(iii) f admits N /N as kernel and induces an isomorphism
(M/N )/(N /N ) - M/N .
Next, let us discuss some construction methods for modules. We start out
from a given R-module M .
(1) Let (xi )iI be a family of elements in M . Then

Rxi = ri xi ; ri R, ri = 0 for almost all i I
iI iI
is a submodule of M , the so-called submodule generated by the xi , i I. This

is the smallest submodule of M containing all elements xi . An R-module M is
called nitely generated,
or of nite type, if there exist elements x1 , . . . , xn M
such that M = ni=1 Rxi .
(2) Let (Ni )iI be a family of submodules of M . Then

N= Ni = xi ; xi Ni , xi = 0 for almost all i I
iI iI
is a submodule of M , called the sum of the submodules Ni of M . The sum is

called direct and we write
N= Ni
iI

if, for every x N , the representations of type x = iI xi with xi Ni are
unique. Note that suppressing trivial summands, such a representation reduces
to a nite sum, say x = xi1 + . . . + xis with indices i1 , . . . , is I. Sometimes
the notation x = xi1 . . . xis is used in this context in order to stress the fact
that the terms xi are unique.
(3) Let (Mi )iI be a family of R-modules. Then the cartesian product

Mi
iI
is an R-module under componentwise addition and scalar multiplication; it is

called the direct product of the Mi . Identifying Mj for j J with the submodule
1.4 Modules 35

Mj for i = j
Ni Mi , where Ni = ,
iI iI
0 j
for i =

we may view each Mj as a submodule of iI Mi . The sum of these submodules
is direct and we have

Mi = (xi )iI Mi ; xi = 0 for almost all i I .
iI iI
Thereby we can dene the direct sum of any family of R-modules not necessarily
given as submodules of an ambient module M . This construction is referred to
as the
constructed or external direct sum of the Mi . For example, we write
R(I) = iI R for the direct
sum of copies
of R, extending over an index set I.
Furthermore, it follows that iI Mi = iI Mi if I is nite.
(4) Given R-modules M, N , the set HomR (M, N ) of all R-module homomor-
phisms M - N is an R-module again. Indeed, using the R-module structure
of N , the sum of arbitrary maps M - N as well as their scalar products with
constants from R are well-dened. Furthermore, any R-module homomorphism
: M - M gives rise to an R-module homomorphism
HomR (M, N ) - HomR (M , N ), - ,
and any R-module homomorphism : N - N gives rise to an R-module

homomorphism
HomR (M, N ) - HomR (M, N ), - .
As an example, consider a family (Mi )iI of submodules in M . Then the inclu-

sion maps i : Mi - M determine an R-module homomorphism

: HomR (M, N ) - HomR (Mi , N ), - ( i )iI .
iI
It is easily seen that is an isomorphism for all R-modules N if and only if M is

the direct sum of the submodules Mi M . Also note that for this assertion we
can replace the inclusion maps i by arbitrary R-module homomorphisms; they
will automatically be injective if is an isomorphism for all N . This gives us the
opportunity to characterize direct sums in terms of a universal mapping prop-
erty. Namely, a direct sum of a family of R-modules (Mi )iI is an R-module M
together with R-module homomorphisms i : Mi - M (necessarily injective)
such that the induced R-module homomorphism as above is an isomorphism
for every R-module N or, in other words, such that for every family of R-module
homomorphisms i : Mi - N , i I, into an arbitrary R-module N there is
a unique R-module homomorphism : M - N satisfying i = i for all
i I. The standard argument for universal mapping properties (for example, as
used in the proof of 1.2/8) shows that M is unique up to canonical isomorphism.
(5) Let a R be an ideal. Then

<
aM = ai xi ; ai a, xi M
i=0
is a submodule of M . It is the submodule generated by all products ax where

a a and x M .
(6) For a family (Mi )iI of submodules of M , the intersection

Mi
iI
is a submodule of M .
Denition 9. Let M be an R-module. A family (xi )iI of elements in M is

given coecients ai R where ai = 0 for
called free if the following holds:
almost all i I, the equation iI ai xi = 0 implies ai = 0 for all i I.
An R-module M is called free if it admits a free generating system; the
latter amounts to a family of non-zero elements (xi )iI such that every x M
admits a representation
x= ai xi
iI
with unique coecients ai R that are trivial for almost all i I.
Free generating systems of modules are to be seen as a generalization of

vector space bases. For example, Rn for some power n N, or R(I) for some
index set I, is a free R-module where the canonical unit vectors give rise to
a free generating system. Every vector space over a eld K is a free K-module.
However, in most cases modules will not admit free generating systems and,
thus, will not be free. Even submodules of free modules need not be free. As
an example, consider the polynomial ring KX, Y in two variables X, Y over
a eld K. Then KX, Y is a free KX, Y -module, since the unit element
1 KX, Y gives rise to a free generating system. Now consider an ideal
a KX, Y . We claim that a is a free submodule of KX, Y if and only
if a is a principal ideal. Indeed, if a = (a) for some a a, then a denes a
free generating system of a, at least if a = 0. If a = 0, the ideal a is zero,
admitting the empty system as a free generating system. On the other hand,
if a is not principal, then every generating system of a contains at least two
elements f, g = 0, and the equation gf f g = 0 shows that such a system
cannot be free. For example, the ideal (X, Y ) KX, Y denes a submodule
of KX, Y that is not free.
The non-existence of free generating systems for arbitrary R-modules is,
of course, related to the possible lack of suciently many units in R. How-
ever, in certain situations, the desired units do exist so that, as a consequence,
one can construct generating systems enjoying special properties. The key fact
underlying such constructions is given by the following result:
1.4 Modules 37
Lemma 10 (Nakayama). Let M be a nitely generated R-module and a j(R)

an ideal such that aM = M . Then M = 0.
Proof. We proceed indirectly and assume M = 0. Since M is nitely generated,

we may consider a minimal generating system x1 , . . . , xn M . Using aM = M ,
there is an equation
xn = a1 x1 + . . . + an xn ,
for suitable coecients ai a. Then
(1 an )xn = a1 x1 + . . . + an1 xn1 ,
and we see from 1.3/2 that 1 an is a unit, since an a j(R). But
then the above equation yields xn n1
i=1 Rxi and M is already gener-
ated by x1 , . . . , xn1 , contradicting our assumption that the generating system
x1 , . . . , xn is minimal.
Corollary 11. Let M be a nitely generated R-module and a j(R) an ideal.

Then, if N M is a submodule satisfying M = N +aM , we must have M = N .
Proof. M = N + aM implies M/N = a(M/N ). Since M/N , just as M is nitely

generated, Nakayamas Lemma yields M/N = 0. Hence, M = N .
Corollary 12. Let R be a local ring with maximal ideal m R. Then, for
any nitely generated R-module M , the quotient M/mM is canonically a vec-
tor space over the eld R/m. Furthermore, if x1 , . . . , xn M are elements
whoseresidue classes x1 , . . . , xn M/mM generate this vector space, then
M = ni=1 Rxi .
n
Proof. From M/mM = i=1 R/m xi we conclude

n
M= Rxi + mM
i=1
n
and, using Corollary 11, M = i=1 Rxi .
Exercises
1. Consider a family (Mi )iI of submodules of an R-moduleM such that Mi Mj
or Mj Mi for any pair of indices i, j I. Show that iI Mi is a submodule
of M . On the other hand, give an example of a module M where the union of
arbitrary submodules is not necessarily a submodule again.
2. Let A be an integral domain that is an algebra over a eld K. Show that A is a
eld if, as a K-vector space, it is of nite dimension.
3. Let (Mi )iI be a family of R-modules. Establish the universal properties for the
direct sum and the direct product of the Mi . Namely, show for any R-module N
that there are canonical bijections

HomR Mi , N - HomR (Mi , N ),
iI iI

HomR N, Mi - HomR (N, Mi ).
iI iI
4. Give an example of a module M over some ring R such that Nakayamas Lemma
(Lemma 10) loses its validity.
5. Generalization of Nakayamas Lemma: Let M be an R-module and a R an
ideal satisfying M = aM . Show that there is some element a a such that
(1+a)M = 0. Hint: Choose generators x1 , . . . , xn of M and write x for the column
vector of the xi . Consider an equation of type x = A x for some (n n)-matrix
A with coecients in a. This yields (I A) x = 0 for the (n n)-unit matrix I.
Deduce det(I A) M = 0 from Cramers rule; see also the proof of 3.1/4.
6. Let : M - M be a surjective endomorphism of a nitely generated R-module
M . Show that is injective. Hint: View M , together with its endomorphism ,
as a module over the polynomial ring in one variable Rt by setting t x = (x)
for x M . Then apply Exercise 5 above.
1.5 Finiteness Conditions and the Snake Lemma
To start with, let us recall the notion of exact sequences of modules over a ring
R. A sequence of R-modules is a chain of morphisms of R-modules
fn2 fn1 fn fn+1
... - Mn1 - Mn - Mn+1 - ...
where the indices are varying over a nite or an innite part of Z. We say
that the sequence satises the complex property at Mn (more specically, at
position n) if we have fn fn1 = 0 or, in equivalent terms, im fn1 ker fn .
Furthermore, the sequence is said to be exact at Mn if, in fact, im fn1 = ker fn .
If the sequence satises the complex property at all places Mn (of course, except
at those where the sequence might terminate), it is called a complex. Likewise,
the sequence is called exact, if it is exact at all places. For example, a morphism
of R-modules f : M - M is injective if and only if the sequence
f
0 - M - M
is exact; here 0 denotes the zero module and 0 - M the zero mapping, the
only possible R-homomorphism from 0 to M . On the other hand, f is surjective
if and only if the sequence
f
M - M - 0
is exact; M - 0 is the zero mapping, the only possible R-homomorphism

from M to 0. Exact sequences of type
1.5 Finiteness Conditions and the Snake Lemma 39
f g
0 - M - M - M - 0
are referred to as short exact sequences. The exactness of such a sequence means:
(1) f is injective,
(2) im f = ker g,
(3) g is surjective.
Thus, for a short exact sequence as above, we can view M as a submodule of
M via f and we see, using the Fundamental Theorem of Homomorphisms 1.4/6,
that g induces an isomorphism M/M - M . Conversely, every submodule
N M gives rise to the short exact sequence
0 - N - M - M/N - 0.
Another type of short exact sequences can be built from the direct sum of
two R-modules M and M , namely
() 0 - M - M M - M - 0,
where M - M M is the canonical injection and M M - M the

projection onto the second factor. Such sequences are the prototypes of so-called
split exact sequences. In fact, an exact sequence of R-modules
() 0 - M - M - M - 0
is called split if it is isomorphic to one of type () in the sense that there is an

isomorphism M M - M making the diagram

0 - M

- M M - M

- 0

?
0 - M - M - M - 0
commutative.
For a short exact sequence
f g
() 0 - M - M - M - 0
as above to be split, it is enough that g admits a section. Thereby we mean

a morphism of R-modules s : M - M such that g s = idM . Then s is
a monomorphism and we may identify M with its image in M so that M
becomes a submodule of M . Viewing M as a submodule of M as well, we get
M M = 0 and M = M +M , as is easily veried. As a result, M = M M
and the exact sequence () is isomorphic to the canonical one (). In a similar
way, one can show that () is split if f admits a retraction, i.e. a morphism of
R-modules t : M - M such that t f = idM . Then g identies ker t with
M and one can show, as before, that M = M M .

f
For any morphism of R-modules M - M we can consider its cokernel
given by
coker f = M/ im f.
Then any morphism f as above gives rise to the canonical exact sequence
f
0 - ker f - M - M - coker f - 0.
Although the denitions of the kernel and cokernel look quite dierent, both
notions are nevertheless closely related. Namely, ker f is characterized by the
exact sequence
0 - ker f - M f- M,
whereas the same is true for coker f and the exact sequence
f
M - M - coker f - 0.
Note that the type of the latter sequence is related to the former one by reversing
arrows. It is this reverse setting, which usually is referred to by the prex co.
Lemma 1 (Snake Lemma). Let

f1 f2
( ) M1 - M2 - M3
u1 u2 u3
? ? ?
g1 g2

( ) N1 - N2 - N3
be a commutative diagram of R-module homomorphisms with exact rows. Then

the diagram extends uniquely to a commutative diagram
f1 f2
ker u1 - ker u2 - ker u3
? f1
? f2
?
M1 - M2 - M3
u1 u2 u3
? g1
? g2
?
N1 - N2 - N3
? g1
? g2
?
coker u1 - coker u2 - coker u3
where the vertical sequences are just the canonical exact sequences associated to
u1 , u2 , u3 . Furthermore:
(i) f2 f1 = 0 and g 2 g 1 = 0.
(ii) If g1 is injective, the top upper row is exact.
(iii) If f2 is surjective, the bottom lower row is exact.
(iv) Let g1 be injective and f2 surjective. Then there exists an R-module

homomorphism d : ker u3 - coker u1 , the so-called snake morphism, dened
as follows:
Starting with x3 ker u3 M3 , choose an f2 -preimage x2 M2 . Then
u2 (x2 ) ker g2 = im g1 and there is some y1 N1 such that g1 (y1 ) = u2 (x2 ).
Now let d(x3 ) be the residue class y 1 of y1 in coker u1 .
(v) In the setting of (iv), the exact sequences of (ii) and (iii) yield an exact
sequence as follows 3 :
Proof. For any x1 ker u1 we have u2 (f1 (x1 )) = g1 (u1 (x1 )) = 0 and, hence,
f1 (x1 ) ker u2 . Thus, f1 restricts to an R-morphism f1 : ker u1 - ker u2 .
Likewise, f2 restricts to an R-morphism f2 : ker u2 - ker u3 . Further-
more, we have g1 (u1 (M1 )) = u2 (f1 (M1 ) and, hence, g1 (im u1 ) im u2 . In
g1
particular, the R-homomorphism N1 - N2 - coker u2 has a kernel
containing im u1 and therefore factorizes uniquely over an R-homomorphism
g 1 : coker u1 - coker u2 . In the same way, we obtain the existence and unique-
ness of the R-homomorphism g 2 .
Assertion (i) is a direct consequence of the exactness of rows ( ) and ( ),
using the relations f2 f1 = 0 and g2 g1 = 0.
In order to verify (ii), assume that g1 is injective. We only have to show that
ker f2 im f1 . To do this, let x2 ker f2 ker f2 M2 . Using the exactness
of ( ), there is some x1 M1 such that f1 (x1 ) = x2 and we claim that, in fact,
x1 ker u1 . Then f1 (x1 ) = x2 will show x2 im f1 and, thus, ker f2 im f1 .
Now x2 ker u2 implies g1 (u1 (x1 )) = u2 (f1 (x1 )) = u2 (x2 ) = 0. Since g1 is
assumed to be injective, we get u1 (x1 ) = 0 and, hence, x1 ker u1 , as desired.
For assertion (iii) it remains to show ker g 2 im g 1 . Therefore choose an
element y 2 ker g 2 , together with a representative y2 N2 . Then the image
g2 (y2 ) is a representative of g 2 (y 2 ) = 0. Thus, g2 (y2 ) im u3 and we can nd
a u3 -preimage x3 M3 of g2 (y2 ). Since f2 was assumed to be surjective, x3
admits an f2 -preimage x2 M2 so that now y2 = y2 u2 (x2 ) is a representative
of y 2 satisfying g2 (y2 ) = 0. But then, using the exactness of ( ), there is a
g1 -preimage y1 N1 of y2 . Writing y 1 coker u1 for the associated residue
class in coker u1 , we get g 1 (y 1 ) = y 2 . Therefore, y 2 im g 1 and we see that
ker g 2 im g 1 , as desired.
Now assume that g1 is injective and f2 surjective. We want to show that we
can dene an R-homomorphism d : ker u3 - coker u1 , as specied in (iv). To
do this, start out from an element x3 ker u3 and, using the surjectivity of f2 ,
3
If we insert the snake morphism d in the above diagram, the exact sequence takes the
shape of a snake; this is how the lemma got its name.
choose an f2 -preimage x2 M2 ; by the exactness of ( ) the latter is unique up

to an additive contribution from im f1 . Then
g2 (u2 (x2 )) = u3 (f2 (x2 )) = u3 (x3 ) = 0
and, using the exactness of ( ), we get u2 (x2 ) ker g2 = im g1 . Thus, u2 (x2 )

admits a g1 -preimage y1 M1 where the latter depends uniquely on u2 (x2 ),
since g1 is injective. Since x2 , as an f2 -preimage of x3 , is unique up to an ad-
ditive contribution from im f1 , we see that y1 depends uniquely on x3 up to an
additive contribution from im u1 . In any case, the residue class y 1 coker u1 is
uniquely determined by x3 and it follows that the map d : ker u3 - coker u1 ,
x3 - y , is well-dened. It is clear that d satises the properties of an
1
R-homomorphism, since d has been dened in terms of taking preimages and
images with respect to R-homomorphisms.
It remains to show that the sequence in (v) is exact at the places ker u3
f d
and coker u1 . Let us start with the sequence ker u2 2- ker u3 - coker u1 .
Clearly, we have im f2 ker d, since any x3 im f2 admits an f2 -preimage x2
in ker u2 . Then u2 (x2 ) = 0. In particular, 0 N1 is a g1 -preimage of u2 (x2 )
and, hence, is a representative of d(x3 ) so that d(x3 ) = 0. Conversely, sup-
pose x3 ker d. Again, let x2 M2 be an f2 -preimage of x3 and y1 N1 a
g1 -preimage of u2 (x2 ). Then, since x3 ker d, we have y1 im u1 and there
exists a u1 -preimage x1 M1 of y1 . Writing
u2 (f1 (x1 )) = g1 (u1 (x1 )) = g1 (y1 ) = u2 (x2 ),
we see x2 f1 (x1 ) ker u2 . Therefore
x3 = f2 (x2 ) = f2 (x2 ) f2 (f1 (x1 )) = f2 (x2 f1 (x1 )) im f2
and, hence, ker d im f2 .

d g
Finally, let us discuss the sequence ker u3 - coker u1 1-
coker u2 .
First, we show im d ker g 1 . To do this, choose an element x3 ker u3 and let
x2 M2 be an f2 -preimage of x3 , as well as y1 N1 a g1 -preimage of u2 (x2 ).
Then y1 is a representative of d(x3 ) and g1 (y1 ) = u2 (x2 ) im u2 a representative
of g 1 (d(x3 )). But then g 1 (d(x3 )) = 0 and we have im d ker g 1 . Conversely,
consider an element y 1 ker g 1 , together with a representative y1 N1 . Then
we have g1 (y1 ) im u2 and there is a u2 -preimage x2 M2 of g1 (y1 ). Observing
the equation u3 (f2 (x2 )) = g2 (u2 (x2 )) = g2 (g1 (y1 )) = 0, where we use that
g2 g1 = 0 due to the exactness of ( ), we conclude x3 := f2 (x2 ) ker u3 and
see from the construction of x3 that d(x3 ) = y 1 . Therefore y 1 im d and, hence,
ker g 1 im d.
Let us mention a special case of the Snake Lemma, which is quite neat to
state.
Corollary 2. Let
f1 f2
0 - M1 - M2 - M3 - 0
u1 u2 u3
? ? ?
0 - N1 - N2 - N3 - 0
g1 g2
be a commutative diagram of R-module homomorphisms with exact rows. Then,

using the notation of Lemma 1, there is a corresponding exact sequence
f1 f2
0 - ker u1 - ker u2 - ker u3
d g1 g2
- coker u1 - coker u2 - coker u3 - 0.
Proof. Apply the Snake Lemma. The injectivity of f1 follows from the injectivity
of f1 and the surjectivity of g 2 from the surjectivity of g2 .
Next, let us discuss some niteness conditions for modules. As a technical

tool, we will need the Snake Lemma. Given an R-module M together with a
generating system (xi )iI , we can look at the R-module homomorphism

: R(I) - M, (ai )iI - a i xi ,
iI
which maps the canonical free generating system (ei )iI of R(I) onto the gener-
ating system (xi )iI of M . The morphism is surjective. On the other hand,
for any epimorphism of R-modules : R(I) - M , the image ((ei ))iI of the
canonical free generating system of R(I) is a generating system of M . Thus, for
a given index set I there exists a generating system of type (xi )iI for M if and
only if there exists an epimorphism R(I) - M.
Denition 3. Let M be an R-module. M is called of nite type (or nitely

generated) if there exists an exact sequence
Rn - M - 0
for some n N. By abuse of language, M is also said to be a nite R-module

in this case. Furthermore, M is called of nite presentation if there exists an
exact sequence
Rm - Rn -M -0
for some m, n N. The latter sequence is referred to as a nite presentation of

M.
It follows from the above explanation that an R-module M is of nite type

in the sense of Denition 3 if and only if it admits a nite generating system
(a condition already mentioned in Section 1.4). Also recall that, by the Fun-
damental Theorem of Homomorphisms 1.4/6, any epimorphism of R-modules
: Rn - M induces an isomorphism Rn / ker - M . Moreover note
that a given R-module M is of nite presentation if and only if there exists an
epimorphism : Rn - M for some n N such that ker is of nite type.
Before applying the Snake Lemma to the niteness conditions just intro-
duced, let us establish a technical lemma, which will turn out to be quite useful.
Lemma 4. Let
Rm Rn
p p
? ?
f g
0 - M - M - M - 0
be a diagram of R-module homomorphisms where the bottom row is exact. Then
the diagram can be enlarged to a diagram of R-module homomorphisms with
exact rows
f g
0 - Rm - Rm+n - Rn -0
p p p
? ? ?
f g
0 - M - M - M - 0.
Proof. We construct the diagram as follows. Let f be the R-module homomor-

phism mapping the canonical free generating system e1 , . . . , em of Rm to the
rst m elements of the canonical free generating system e1 , . . . , em+n of Rm+n .
Furthermore, let g be the R-module homomorphism mapping e1 , . . . , em to 0
and em+1 , . . . , em+n onto the canonical free generating system of Rn . Then the
sequence
f g
0 - Rm - Rm+n - Rn - 0
is exact. To dene the morphism p : R - M , we require e1 , . . . , em to
m+n
be mapped to f (p (e1 )), . . . , f (p (em )), as well as em+1 , . . . , em+n to arbitrarily
chosen g-preimages of the elements p (g(em+1 )), . . . , p (g(em+n )). The resulting
diagram is commutative.
Proposition 5. Let
f g
0 - M - M - M - 0
be an exact sequence of R-modules.

(i) If M is of nite type, the same holds for M .
(ii) If M and M are of nite type, the same holds for M .
Proof. If Rn - M is an epimorphism of R-modules, its composition with g

will be surjective, since g is surjective. Thus, if M is of nite type, the same is
true for M .
Now let M and M be of nite type and let p : Rm - M as well as

p :R n -
M be epimorphisms of R-modules. Then, using Lemma 4, there
is a commutative diagram
f g
0 - Rm - Rm+n - Rn - 0
p p p
? ? ?
f g
0 - M - M - M - 0
with exact rows and Corollary 2 shows that there is an associated exact sequence
... - coker p - coker p - coker p - 0.
However, coker p and coker p are trivial, since p and p are surjective. Hence,
we must have coker p = 0, too, and it follows that p is surjective. In particular,
M is of nite type.
Of course, alternatively, we could have argued in a more conventional way
by showing that a generating system of M together with a lifting of a generating
system of M yields a generating system of M .
In the setting of Proposition 5 (i) we cannot expect that M of nite type

implies the same for M , since a submodule of a nitely generated R-module
is not necessarily nitely generated again. To give an example, consider the
polynomial ring KX1 , X2 , . . . in innitely many variables over a eld K. As
a module over itself, it is generated by the unit element 1. Hence, it is nitely
generated. However, the submodule (X1 , X2 , . . .) KX1 , X2 , . . . , which is
given by the ideal generated by all variables, is not nitely generated.
We want to carry assertion (ii) of Proposition 5 over to modules of nite
presentation.
Proposition 6. Let
f g
0 - M - M - M - 0
be an exact sequence of R-modules. If M and M are of nite presentation,

the same holds for M .
Proof. We proceed similarly as in the proof of Proposition 5 and choose nite

presentations
q p q p
Rs - Rm - M - 0, Rt - Rn - M - 0.
Being images of nitely generated R-modules, we see that ker p = im q and

ker p = im q are nitely generated. Now use Lemma 4 and enlarge the given
short exact sequence together with the morphisms p , p to a commutative dia-
gram
f g
0 - Rm - Rm+n - Rn - 0
p p p
? ? ?
f g
0 - M - M - M - 0
with exact rows. Then the Snake Lemma yields an exact sequence
0 - ker p - ker p - ker p

- coker p - coker p - coker p - 0.
Since p and p are surjective, we get coker p = 0 and coker p = 0, hence, also
coker p = 0. Furthermore, the remaining short exact sequence
0 - ker p - ker p - ker p - 0
shows via Proposition 5 that ker p is nitely generated, since the same is true
for ker p and ker p . In particular, there exists an epimorphism of R-modules of
type R - ker p Rm+n and the resulting sequence
p
R - Rm+n - M - 0
is exact. Hence, M is of nite presentation.
If M is an R-module of nite presentation, then by denition, M may be

seen as part of an exact sequence of type Rm - Rn -M - 0. In
other words, there exists an epimorphism R n - M , whose kernel is nitely
generated. We want to show that then the kernel of any epimorphism of type
Rn - M is nitely generated.
Proposition 7. For an R-module M the following conditions are equivalent:

(i) M is of nite presentation.
(ii) M is of nite type and for every R-module epimorphism : M -M
where M is of nite type, the kernel ker is of nite type.
Proof. We only have to show the implication from (i) to (ii). Therefore let
Rm - Rn - M - 0
be a nite presentation of M and let : M - M be an epimorphism of

R-modules where M is of nite type. In order to show that ker is of nite
type, look at the short exact sequence

0 - ker - M - M - 0.
We want to combine these two exact sequences to a commutative diagram

f1 f2
Rm - Rn - M - 0

u1 u2
?
? g1
0 - ker - M - M - 0
as follows. Dene u2 by requiring that the canonical free generating system

e1 , . . . , en of Rn be mapped onto certain -preimages of f2 (e1 ), . . . , f2 (en ) in
M , which we can freely choose due to the surjectivity of . Then the resulting
(right-hand) square is commutative. Since u2 f1 = f2 f1 = 0, it follows
that the image of u2 f1 is contained in ker = im g1 . Therefore we can dene
u1 : Rm - ker by restricting the range of u2 f1 . Since g1 is injective and
f2 surjective, the Snake Lemma yields an exact sequence
0 = ker idM - coker u1 - coker u2 - coker idM = 0
and, thus, an isomorphism coker u1 - coker u2 . Now coker u2 , as a quotient

of M , is of nite type and, hence, the same is true for coker u1 . Then we may
look at the exact sequence
0 - im u1 - ker - coker u1 - 0.
Since im u1 and coker u1 are of nite type, Proposition 5 (ii) shows that ker
is of nite type. Thus, we are done.
We have already seen that a submodule N of an R-module M is not neces-

sarily of nite type, even if M is of nite type. The same phenomenon occurs for
modules of nite presentation. Furthermore, the image of an R-module of nite
presentation under a module homomorphism is not necessarily of nite presen-
tation again. As an example, consider a ring R admitting an ideal a R that is
not nitely generated. Then R, as a free module over itself, is of nite presen-
tation, but R/a as an R-module is not. The latter follows from Proposition 7,
since a, as the kernel of the canonical surjection R - R/a, is not nitely
generated. A better behavior can be expected if we put suitable niteness con-
ditions on the ring R. We will study the properties of R being Noetherian or
coherent.
Denition 8. Let R be a ring and M an R-module. M is called Noetherian

if every submodule N M is of nite type. M is called coherent if M is of
nite type and if every submodule of nite type N M is of nite presentation.
R itself is called Noetherian (resp. coherent) if R, as an R-module, enjoys the
corresponding property.
If N is a submodule of a Noetherian R-module M , then N is Noetherian

for trivial reasons. Likewise, if M is just coherent, N will be coherent as soon as
it is of nite type. In the following, we want to look more closely at Noetherian
and coherent modules, starting with the Noetherian case.
Lemma 9. For an R-module M the following conditions are equivalent:

(i) M is Noetherian.
(ii) Every ascending chain M1 M2 . . . M of submodules in M be-
comes stationary, i.e. there exists an index i0 N such that Mi0 = Mi for all
i i0 .
Proof. Assume rst that M is Noetherian, as in (i), and let M1 M2 . . . be an

ascending chain of submodules in M . Then it is easily seen that N := iN Mi
is a submodule of M . The latter is nitely generated, say N = nj=1 Rxj , and
there will exist an index i0 N such that x1 , . . . , xn Mi0 . But then we have
N Mi N and, hence, N = Mi for all i i0 , which means that the chain of
submodules Mi M becomes stationary.
Conversely, assume that condition (ii) is satised. Suppose there is a sub-
module N M that is not nitely generated. Then, choosing an element
x1 N , we get Rx1 N and there is an element x2 N Rx1 . Since N
is not nitely generated, we must have Rx1 + Rx2 N . Continuing this way,
we can construct an innite strictly ascending chain of submodules in M which,
however, contradicts condition (ii).
f g
Lemma 10. Let 0 - M -M - M - 0 be an exact sequence of
R-modules. Then the following conditions are equivalent:
(i) M is Noetherian.
(ii) M and M are Noetherian.
Proof. Assume rst that M is Noetherian. Viewing M as a submodule of M via

f , we see that M is Noetherian for trivial reasons. But also M is Noetherian.
Indeed, the preimage g 1 (N ) M of any submodule N M is nitely gen-
erated since M is Noetherian. Then N = g(g 1 (N )), as the image of a nitely
generated R-module, must be nitely generated itself.
Next, let M and M be Noetherian and let N M be a submodule. Then
N gives rise to the short exact sequence
0 - f 1 (N ) - N - g(N ) - 0
and it follows that f 1 (N ) M as well as g(N ) M are submodules of nite

type. Applying Proposition 5 we see that N is of nite type.
As a direct consequence we can show:
Corollary 11. Let M1 , M2 be Noetherian R-modules. Then:

(i) M1 M2 is Noetherian.
(ii) If M1 , M2 are given as submodules of some R-module M , then M1 + M2
and M1 M2 are Noetherian.
Proof. Using Lemma 10, the exact sequence
0 - M1 - M1 M2 - M2 - 0,
consisting of the canonical embedding M1 - M1 M2 and the natural pro-

jection M1 M2 - M2 , shows that M1 M2 is Noetherian, since M1 and M2
are Noetherian. If M1 , M2 are submodules of some R-module M , consider the
canonical epimorphism
p : M 1 M2 - M1 + M2 , m1 m 2 - m1 + m2 ,
as well as the corresponding exact sequence

p
0 - ker p - M1 M2 - M 1 + M2 - 0,
where ker p is isomorphic to M1 M2 . Because we know already that M1 M2 is

Noetherian, we see from Lemma 10 that M1 + M2 and M1 M2 are Noetherian.

Of course, using an inductive argument, the assertion of the corollary ex-

tends to nite sums and nite intersections.
Proposition 12. For a ring R, the following conditions are equivalent:

(i) R is Noetherian.
(ii) Every R-module M of nite type is Noetherian.
Proof. We have only to show the implication (i) = (ii). Therefore assume
that R is Noetherian and let M be an R-module of nite type. There exists an
epimorphism p : Rn - M for some n N, and we can conclude by induction
from Corollary 11 that Rn is Noetherian, since R is Noetherian. But then, using
Lemma 10, the exact sequence
p
0 - ker p - Rn - M - 0,
shows that M is Noetherian.
Corollary 13. Let R be a Noetherian ring and M an R-module of nite type.

Then M is coherent. In particular, R is coherent.
Proof. Since M is a module of nite type over a Noetherian ring, it is Noetherian

by Proposition 12. Now let N M be a submodule. Then N is of nite type
and there is an epimorphism Rn - N for some n N. Since R is Noetherian,
Rn is Noetherian and it follows that the kernel of Rn - N , as a submodule
of Rn , is nitely generated. In particular, N is of nite presentation.
Next, let us prove a result that is fundamental for the construction of

Noetherian rings.
Proposition 14 (Hilberts Basis Theorem). Let R be a eld or, more generally,

a Noetherian ring. Then:
(i) The polynomial ring RY in one variable Y over R is Noetherian.
(ii) In particular, any quotient RX1 , . . . , Xn /a of a polynomial ring in a
nite set of variables Xi by some ideal a RX1 , . . . , Xn is Noetherian.
Proof. To verify assertion (i), consider an ideal a RY , where R is a Noethe-
rian ring. Dene ai R for i N as the set of all elements a R such that a
contains a polynomial of type
aY i + terms of lower degree in Y.
Then it is easily seen that each ai is an ideal in R. Furthermore, these ideals

form an ascending chain
a0 a1 . . . R;
just use the fact that f a implies Y f a. According to Lemma 9, the chain
becomes stationary at some index i0 , because R is Noetherian. Now, for each
i {0, . . . , i0 }, choose nitely many polynomials fij a of degree deg fij = i
such that the highest coecients aij of the fij generate the ideal ai . This is
possible, since R is Noetherian and, hence, the ideals ai are nitely generated.
We claim that the polynomials fij generate the ideal a. Indeed, let g a and
assume g = 0. Let d = deg g and denote by a R the highest coecient of g.
Writing i = min{d, i0 }, we have a ai and there is an equation

a= cj aij , cj R.
j
Obviously, the polynomial

g1 = g Y di cj fij
j
belongs to a again. Its degree, however, is strictly smaller than the degree d of
g, because, by our construction, the coecient of Y d will vanish. If g1 = 0, we
can proceed in the same way with g1 in place of g and so on. This way, after
nitely many steps, we end up with a polynomial gs where gs = 0. We conclude
that g is a linear combination of the fij with coecients in RY and thereby
see that the fij generate the ideal a. This shows that RY is Noetherian.
It follows by induction that the polynomial ring RX1 , . . . , Xn in a nite set
of variables is Noetherian and the same is true for any quotient RX1 , . . . , Xn /a
by Lemma 10.
Finally, let us study coherent modules. We start with an analogue of

Lemma 10.
f g
Lemma 15. Let 0 - M - M - M - 0 be an exact sequence of
R-modules. Then:
(i) If M is of nite type and M is coherent, M is coherent.

(ii) If M and M are coherent, M is coherent.
(iii) If M and M are coherent, M is coherent.
In particular, if two of the modules M , M, M are coherent, all three of
them will have this property.
Proof. We start with assertion (i). Thus, let M be of nite type and M of nite
presentation. Then M is of nite type by Proposition 5. To show that M is
even of nite presentation, choose an epimorphism : Rn - M and look at
the commutative diagram with exact rows

0 - ker - Rn - M - 0

u1 u2
? ?
f g
0 - M - M - M - 0,
where u2 is dened by mapping the canonical generating system e1 , . . . , en of Rn
onto g-preimages of (e1 ), . . . , (en ) and u1 is induced from u2 . We may assume
that u1 is surjective. Indeed, otherwise we can consider a nite generating system
x1 , . . . , xr of M and replace Rn by Rn+r , extending the morphisms and u2
by mapping the additional generators en+1 , . . . , en+r of Rn+r as follows:
(en+i ) = 0, u2 (en+i ) = f (xi ), i = 1, . . . , r.
Applying Corollary 2, we get the exact sequence
0 - ker u1 - ker u2 - ker idM - ...,
where ker idM = 0 implies that we have an isomorphism ker u1 - ker u2 .

Since M is of nite presentation, ker u2 and, hence, also ker u1 are of nite type
by Proposition 7. Then the exact sequence
u1
0 - ker u1 - ker - M - 0
shows by Proposition 5 that ker is of nite type. In particular, M is of nite

presentation.
Now let N M be a submodule of nite type. In order to verify that
it is of nite presentation, consider the preimage N of N with respect to
g: M - M , as well as the resulting exact sequence
0 - M - N - N - 0.
Then N is of nite type by Proposition 5 and, thus, as a nitely generated sub-

module of M , even of nite presentation. But then we are in the same situation
as considered above and we can conclude that N is of nite presentation, and
hence, that M is coherent.
Next, as in (ii), assume that M and M are coherent. Then M is of nite
presentation by Proposition 6. To show that every nitely generated submodule
N M is of nite presentation, let N = f 1 (N ) and N = g(N ). We get the

following commutative diagram with exact rows
0 - N - N - N - 0
u1 u2 u3
? ? ?
f g
0 - M - M - M - 0
where u1 , u2 , u3 are the canonical inclusion maps. Now N , being the image of
the nitely generated R-module N , is of nite type itself and, thus, of nite
presentation, because it is a submodule of a coherent module. Furthermore,
Proposition 7 shows that N is of nite type and, thus, of nite presentation,
again because it is a submodule of a coherent module. But then Proposition 6
implies that N is of nite presentation and, hence, that M is coherent.
Finally, let us assume as in (iii) that M and M are coherent. Then M
is of nite type by Proposition 7. Viewing M as a submodule of the coherent
module M , we can conclude that M is coherent.
Corollary 16. Let M1 , M2 be coherent R-modules. Then:

(i) M1 M2 is coherent.
(ii) Assume that, in addition, M1 , M2 are submodules of a pseudo-coherent
R-module M , i.e. of an R-module M such that every nitely generated submod-
ule is of nite presentation. Then M1 + M2 and M1 M2 are coherent.
Proof. Similarly as in the proof of Corollary 11, we consider the canonical exact
sequence
0 - M1 - M1 M2 - M2 - 0.
If M1 and M2 are coherent, we see from Lemma 15 that M1 M2 is coherent.

If, in addition, M1 and M2 are submodules of a pseudo-coherent R-module M ,
then M1 + M2 is coherent, as it is a nitely generated submodule of M . Using
Lemma 15 again, the exact sequence
0 - M1 M2 - M1 M2 - M1 + M2 - 0
shows that M1 M2 is coherent.
Proposition 17. For a ring R the following conditions are equivalent:

(i) R is coherent.
(ii) Every R-module M of nite presentation is coherent.
Proof. We only have to show the implication from (i) to (ii). Therefore, let R
be coherent and let M be an R-module of nite presentation. Then there is an
epimorphism p : Rn - M for some n N such that ker p is of nite type.
Looking at the short exact sequence
p
0 - ker p - Rn - M - 0,
we know from Corollary 16 that Rn is coherent if R is coherent. But then, since

ker p is of nite type, M is coherent by Lemma 15.
There are some interesting classes of coherent rings transgressing the

Noetherian case. For example, any polynomial ring KX in an innite family
of variables over a eld K is coherent, but not Noetherian; see Exercise 4.3/2.
Furthermore, algebras of topologically nite presentation over a complete (non-
discrete) valuation ring of height 1 are coherent, but in general not Noetherian;
see [4], Section 1, in particular 1.3, for details. These algebras play a fundamen-
tal role in Rigid Geometry.
Exercises
1. Let p N be prime. Determine the (isomorphism classes of) short exact sequences
of Z-modules that are of type
0 - Z/pZ - M - Z/pZ - 0.
2. Let : M - Rn for some n N be an epimorphism of R-modules where M is

of nite type. Show that ker is of nite type.
3. Five Lemma: Consider a commutative diagram of R-modules
M1 - M2 - M3 - M4 - M5
u1 u2 u3 u4 u5
? ? ? ? ?
N1 - N2 - N3 - N4 - N5
with exact rows, where u1 is an epimorphism, u2 , u4 are isomorphisms, and u5 is

a monomorphism. Apply the Snake Lemma to show that u3 is an isomorphism.
4. Prove:
(1) A sequence of R-modules M - M - M - 0 is exact if and only if
the associated sequence
0 - HomR (M , N ) - HomR (M, N ) - HomR (M , N )
is exact for every R-module N .

(2) A sequence of R-modules 0 - N - N - N is exact if and only if
the associated sequence
0 - HomR (M, N ) - HomR (M, N ) - HomR (M, N )
is exact for every R-module M .

5. Consider an R-module homomorphism : Rm - Rn for certain exponents
m, n N and assume R = 0. Show:
(1) If is an isomorphism, then m = n.
(2) If is an epimorphism, then m n.
(3) If is a monomorphism, then m n.
Hint: For the proof of (1) and (2) choose a maximal ideal m R and consider the
induced homomorphism : Rm /mRm - Rn /mRn of R/m-vector spaces. For
the proof of (3) proceed indirectly. Assume m > n, say m = n+1, and construct a
chain of R-module monomorphisms . . . - Rn+2 - Rn+1 - Rn . Just take
for Rn+1+i - Rn+i the cartesian product of : Rn+1 - Rn with the identity
map on Ri . Then the images of the unit vectors en+1 Rn+1+i , i N, form
a free family of elements in Rn and thereby can be used to construct an innite
ascending chain of submodules in Rn . In particular,
if R is Noetherian, we get a
contradiction. In the general case, write (ej ) = ni=1 aij ei for j = 1, . . . , n + 1,
using canonical unit vectors ei Rn , resp. ej Rn+1 , and suitable coecients
aij R. Then consider the smallest subring R R containing all coecients aij .
Show that R is Noetherian and reduce to the case where R is replaced by R .
6. Let R be a Noetherian ring. Show that any quotient R/a by some ideal a R as
well as any localization RS of R are Noetherian again.
7. Let R be a coherent ring and a R a nitely generated ideal. Show that R/a is
a coherent ring. Furthermore, it can be shown that any localization of a coherent
ring is coherent again; see Exercise 4.3/1.
8.
A formal power series with coecients in a ring R is a formal expression of type
iN ai X where ai R for all i. Similarly as in the case of polynomials, X is
i
referred to as a variable. Show that such formal power series form a ring under
conventional addition and multiplication; the latter is denoted by RX. Prove
that RX is Noetherian if R is Noetherian.
9. Let R = C0, 1 be the ring of all continuous real valued functions on the unit
interval 0, 1 R. Show that R is not Noetherian.
10. Let R be the ring of all polynomials f QX such that f (Z) Z. Consider the
polynomials fn (X) = n!1
X(X 1) . . . (X n + 1) QX, n N, and show:
(a) The system (fn )nN is a Q-vector space basis of QX.
(b) All fn belong to R and (fn )nN is a free generating system of R as a Z-module.
(c) If n is prime, fn does not belong to (f1 , . . . , fn1 ), the ideal generated by
f1 , . . . , fn1 in R. In particular, R cannot be Noetherian.
Hint: The Q-linear map : QX - QX, f (X) - f (X + 1) f (X),
satises the formula (fn ) = fn1 for n N where f1 := 0.
2. The Theory of Noetherian Rings
As we have seen in 1.5/8, a ring is called Noetherian if all its ideals are nitely
generated or, equivalently by 1.5/9, if its ideals satisfy the ascending chain con-
dition. The aim of the present chapter is to show that the Noetherian hypothesis,
as simple as it might look, nevertheless has deep impacts on the structure of
ideals and their inclusions, culminating in the theory of Krull dimension, to be
dealt with in Section 2.4.
To discuss some standard examples of Noetherian and non-Noetherian rings,
recall from Hilberts Basis Theorem 1.5/14 that all polynomial rings of type
RX1 , . . . , Xn in nitely many variables X1 , . . . , Xn over a Noetherian ring R
are Noetherian. The result extends to algebras of nite type over a Noethe-
rian ring R, i.e. R-algebras of type RX1 , . . . , Xn /a where a is an ideal in
RX1 , . . . , Xn . In particular, algebras of nite type over a eld K or over the
ring of integers Z are Noetherian. One also knows that all rings of integral alge-
braic numbers in nite extensions of Q are Noetherian (use AtiyahMacdonald
[2], 5.17), whereas the integral closure of Z in any innite algebraic extension of
Q is not; see Section 3.1 for the notion of integral dependence and in particular
3.1/8 for the one of integral closure. Also note that any polynomial ring RX
in an innite family of variables X over a non-zero ring R will not be Noethe-
rian. Other interesting examples of non-Noetherian rings belong to the class of
(general) valuation rings, as introduced in 9.5/13.
To approach the subject of Krull dimension for Noetherian rings, the tech-
nique of primary decomposition, developed in Section 2.1, is used as a key tool.
We will show in 2.1/6 that such a primary decomposition exists for all ideals a
of a Noetherian ring R. It is of type
() a = q 1 . . . qr
where q1 , . . . , qr are so-called primary ideals in R. Primary ideals generalize

the notion of prime powers in principal ideal domains, whereas the concept of
primary decomposition generalizes the one of prime factorization.
Looking at a primary decomposition (), the nilradicals pi = rad(qi ) are of
particular signicance; they are prime in R and we say that qi is pi -primary.
As any nite intersection of p-primary ideals, for any prime ideal p R, is
p-primary again (see 2.1/4), we may assume that all p1 , . . . , pr belonging to
the primary decomposition () are dierent. In addition, we can require that
the decomposition () is minimal in the sense that it cannot be shortened any

56 2. The Theory of Noetherian Rings
further. In such a situation we will show in 2.1/8 that the set of prime ideals
p1 , . . . , pr is uniquely determined by a; it is denoted by Ass(a), referring to the
members of this set as the prime ideals associated to a. There is a uniqueness
assertion for some of the primary ideals qi as well (see 2.1/15), although not all
of them will be unique in general.
Now look at the primary decomposition () and pass to nilradicals, thereby
obtaining the decomposition

rad(a) = p
pAss(a)
for the nilradical of a. This is again a primary decomposition, but maybe not
a minimal one. Anyway, we know already from 1.3/6 that rad(a) equals the
intersection of all prime ideals in R containing a or even better, of all minimal
prime divisors of a; the latter are the prime ideals p R that are minimal with
respect to the inclusion a p. Using 1.3/8, it follows that the minimal prime
divisors of a all belong to Ass(a). Thus, their number is nite since Ass(a) is
nite; see 2.1/12. It is this niteness assertion, which is of utmost importance in
the discussion of Krull dimensions for Noetherian rings. Translated to the world
of schemes it corresponds to the fact that every Noetherian scheme consists of
only nitely many irreducible components; see 7.5/5.
To give an application of the just explained niteness of sets of associated
prime ideals, consider a Noetherian ring R where every prime ideal is maximal;
we will say that R is of Krull dimension 0. Then all prime ideals of R are
associated to the zero ideal in R and, hence, there can exist only nitely many
of them. Using this fact in conjunction with some standard arguments, we can
show in 2.2/8 that R satises the descending chain condition for ideals and,
thus, is Artinian. Conversely, it is shown that every Artinian ring is Noetherian
of Krull dimension 0.
The Krull dimension of a general ring R is dened as the supremum of all
lengths n of chains of prime ideals p0 . . . pn in R and is denoted by dim R.
Restricting to chains ending at a given prime ideal p R, the corresponding
supremum is called the height of p, denoted by ht p. As a rst major result
in dimension theory we prove Krulls Dimension Theorem 2.4/6, implying that
ht p is nite if R is Noetherian. From this we conclude that the Krull dimension
of any Noetherian local ring is nite; see 2.4/8. On the other hand, it is not too
hard to construct (non-local) Noetherian rings R where dim R = . Namely,
following Nagata [22], Appendix A1, Example 1, we consider a polynomial ring
R = KX1 , X2 , . . . over a eld K where each Xi is a nite system of variables,
say of length ni , such that limi ni = . Let pi be the prime ideal that is
generated by Xi in R and let S R be the multiplicative system given by the
complement of the union i=1 pi . Then we claim that the localization RS is a
Noetherian ring of innite dimension. Indeed, looking at residue rings of type
R/(Xr , Xr+1 , . . .) for suciently large indices r, we can use 1.3/7 in order to
show that the pi are just those ideals in R that are maximal with respect to the
property of being disjoint from S. Therefore the ideals mi = pi RS , i = 1, 2, . . .,
2.1 Primary Decomposition of Ideals 57
represent all maximal ideals of RS . Identifying each localization (RS )mi with
the localization K(Xj ; j = i)Xi (Xi ) , we see with the help of Hilberts Basis
Theorem 1.5/14 that all these rings are Noetherian. It follows, as remarked
after dening Krull dimensions in 2.4/2, that dim(RS )mi = ht(mi ) ni . Hence,
we get dim RS = . To show that RS is Noetherian, indeed, one can use the
facts that all localizations (RS )mi are Noetherian and that any non-zero element
a RS is contained in at most nitely many of the maximal ideals mi RS .
See the reference of Nagata [22] given above in conjunction with Exercise 4.3/5.
Krulls Dimension Theorem 2.4/6 reveals a very basic fact: if R is a Noethe-
rian ring and a R an ideal generated by r elements, then ht p r for every
minimal prime divisor p of a. For the proof we need the niteness of Ass(a) as
discussed in 2.1/12 and, within the context of localizations, the characterization
of Noetherian rings of dimension 0 in terms of Artinian rings. Another techni-
cal ingredient is Krulls Intersection Theorem 2.3/2, which in turn is based on
Nakayamas Lemma 1.4/10 and the ArtinRees Lemma 2.3/1. Both, Krulls
Intersection Theorem and the ArtinRees Lemma allow nice topological inter-
pretations in terms of ideal-adic topologies; see the corresponding discussion in
Section 2.3.
For Noetherian local rings there is a certain converse of Krulls Intersection
Theorem. Consider such a ring R of dimension d, and let m be its maximal
ideal so that ht m = d. Then every m-primary ideal q R satises rad(q) = m
and, thus, by Krulls Dimension Theorem, cannot be generated by less than
d elements. On the other hand, a simple argument shows in 2.4/11 that there
always exist m-primary ideals in R that are generated by a system of d elements.
Alluding to the situation of polynomial rings over elds, such systems are called
systems of parameters of the local ring R. Using parameters, the dimension
theory of Noetherian local rings can be handled quite nicely; see for example
the results 2.4/13 and 2.4/14. Furthermore, we can show that a polynomial ring
RX1 , . . . , Xn in nitely many variables X1 , . . . , Xn over a Noetherian ring R
has dimension dim R + n, which by examples of Seidenberg may fail to be true
if R is not Noetherian any more.
A very particular class of Noetherian local rings is given by the subclass of
regular local rings, where a Noetherian local ring is called regular if its maxi-
mal ideal can be generated by a system of parameters. Such rings are integral
domains, as we show in 2.4/19. They are quite close to (localizations of) poly-
nomial rings over elds and are useful to characterize the geometric notion of
smoothness in Algebraic Geometry; see for example 8.5/15.
2.1 Primary Decomposition of Ideals
Let R be a principal ideal domain. Then R is factorial and any non-zero element
a R admits a factorization a = pn1 1 . . . pnr r with a unit R , pairwise non-
equivalent prime elements pi R, and exponents ni > 0 where these quantities
are essentially unique. Passing to ideals, it follows that every ideal a R admits
a decomposition
a = pn1 1 . . . pnr r
with pairwise dierent prime ideals pi that are unique up to order, and expo-
nents ni > 0 that are unique as well. The purpose of the present section is to
study similar decompositions for more general rings R where the role of the
above prime powers pni i is taken over by the so-called primary ideals. In the
following we start with a general ring R (commutative and with a unit element,
as always). Only later, when we want to show the existence of primary decom-
positions, R will be assumed to be Noetherian. For a generalization of primary
decompositions to the context of modules see Serre [24], I.B.
Denition 1. A proper ideal q R is called a primary ideal if ab q for any

elements a, b R implies a q or, if the latter is not the case, that there is an
exponent n N such that bn q.
Clearly, any prime ideal is primary. Likewise, for a prime element p of a

factorial ring, all powers (p)n , n > 0, are primary. But for general rings we will
see that the higher powers of prime ideals may fail to be primary. Also note
that an ideal q R is primary if and only if the zero ideal in R/q is primary.
The latter amounts to the fact that all zero-divisors in R/q are nilpotent. More
generally, if : R - R/a is the canonical projection from R onto its quotient
by any ideal a R, then an ideal q R/a is primary if and only if its preimage
1 (q) is primary in R.
Remark 2. Let q R be a primary ideal. Then p = rad(q) is a prime ideal in

R. It is the unique smallest prime ideal p containing q.
Proof. First, rad(q) is a proper ideal in R, since the same holds for q. Now
let ab rad(q) for some elements a, b R where a rad(q). Then there is an
exponent n N such that an bn q. Since a rad(q) implies an q, there exists

an exponent n N such that bnn q. However, the latter shows b rad(q) and
we see that rad(q) is prime. Furthermore, if p is a prime ideal in R containing
q, it must contain p = rad(q) as well. Hence, the latter is the unique smallest
prime ideal containing q.
Given a primary ideal q R, we will say more specically that q is

p-primary for the prime ideal p = rad(q).
Remark 3. Let q R be an ideal such that its radical rad(q) coincides with a
maximal ideal m R. Then q is m-primary.
Proof. Replacing R by R/q and m by m/q, we may assume q = 0. Then m is the

only prime ideal in R by 1.3/4 and we see that R is a local ring with maximal
ideal m. Since R m consists of units, all zero divisors of R are contained in m
and therefore nilpotent. This shows that q = 0 is a primary ideal in R.
In particular, the powers mn , n > 0, of any maximal ideal m R are

m-primary. However, there might exist m-primary ideals of more general type,
as can be read from the example of the polynomial ring R = KX, Y over a
eld K. Namely, q = (X, Y 2 ) is m-primary for m = (X, Y ), although it is not a
power of m. On the other hand, there are rings R admitting a (non-maximal)
prime ideal p such that the higher powers of p are not primary. To construct such
a ring, look at the polynomial ring R = KX, Y, Z in three variables over a eld
K. The ideal q = (X 2 , XZ, Z 2 , XY Z 2 ) R has radical p = rad(q) = (X, Z),
which is prime. However, q is not primary, since XY q, although X q
and Y rad(q) = p. Now observe that that q = (X, Z)2 + (XY Z 2 ). Since
(XY Z 2 ) q p, we may pass to the residue ring R = R/(XY Z 2 ),
obtaining q = p2 for the ideals induced from p and q. Then p is prime in R, but
p2 cannot be primary since its preimage q R is not primary.
p-primary ideals in R for some prime ideal p R.

Lemma 4. Let q1 , . . . , qr be
Then the intersection q = ri=1 qi is p-primary.

Proof. First observe that rad(q) = ri=1 rad(qi ) = p, since we have rad(qi ) = p
for all i. Next consider elements a, b R such that ab q or, in other words,
ab qi for all i. Now if b rad(q) = p, we get b rad(qi ) for all i. But then, as
all qi are primary, we must have a qi for all i and, hence, a q. This shows
that q is p-primary.
Denition 5. A primary decomposition of an ideal a R is a decomposition

r
a= qi
i=1
into primary ideals qi R. The decomposition is called minimal if the corre-

sponding prime ideals pi = rad(qi ) are pairwise
dierent and the decomposition
is unshortenable; the latter means that i=j qi qj for all j = 1, . . . , r.
An ideal a R is called decomposable if it admits a primary decomposition.
In particular, it follows from Lemma 4 that any primary decomposition of

an ideal a R can be reduced to a minimal one. On the other hand, to show the
existence of primary decompositions, special assumptions are necessary such as
R being Noetherian.
Theorem 6. Let R be a Noetherian ring. Then any ideal a R is decomposable,

i.e. it admits a primary decomposition and, in particular, a minimal one.
Proof. Let us call a proper ideal q R irreducible if from any relation q = a1 a2

with ideals a1 , a2 R we get q = a1 or q = a2 . In a rst step we show that
every irreducible ideal q R is primary. To achieve this we may pass to the
quotient R/q, thereby assuming that q is the zero ideal in R.
Now consider elements a, b R, a = 0, such that ab = 0 and look at the

chain of ideals
Ann(b) Ann(b2 ) . . . R
where the annihilator Ann(bi ) consists of all elements x R such that xbi = 0.
The chain becomes stationary, since R is Noetherian. Hence, there is an index
n N such that Ann(bi ) = Ann(bn ) for all i n. We claim that (a) (bn ) = 0.
Indeed, for any y (a)(bn ) there are elements c, d R such that y = ca = dbn .
Then yb = 0 and, hence, dbn+1 = 0 so that d Ann(bn+1 ) = Ann(bn ). It follows
y = dbn = 0 and therefore (a) (bn ) = 0, as claimed. Assuming that the
zero ideal of R is irreducible, the latter relation shows (bn ) = 0, since a = 0.
Consequently, b is nilpotent and we see that 0 R is primary.
It remains to show that every ideal a R is a nite intersection of ir-
reducible ideals. To do this, let M be the set of all ideals in R that are not
representable as a nite intersection of irreducible ideals. If M is non-empty,
we can use the fact that R is Noetherian and thereby show that M contains a
maximal element a. Then a cannot be irreducible, due to the denition of M .
Therefore we can write a = a1 a2 with ideals a1 , a2 R that do not belong to
M . But then a1 and a2 are nite intersections of irreducible ideals in R and the
same is true for a. However, this contradicts the denition of a and therefore
yields M = . Thus, every ideal in R is a nite intersection of irreducible and,
hence, primary ideals.
To give an example of a non-trivial primary decomposition, consider again

the polynomial ring R = KX, Y, Z in three variables over a eld K and its
ideals
p = (X, Y ), p = (X, Z), m = (X, Y, Z).
Then a = p p satises
a = p p m2
and, as is easily checked, this is a minimal primary decomposition.
In the following we will discuss uniqueness assertions for primary decom-
positions of decomposable ideals. In particular, this applies to the Noetherian
case, although the uniqueness results themselves do not require the Noetherian
assumption. We will need the notion of ideal quotients, as introduced in Sec-
tion 1.1. Recall that for an ideal q R and an element x R the ideal quotient
(q : x) consists of all elements a R such that ax q.
Lemma 7. Consider a primary ideal q R as well as the corresponding prime

ideal p = rad(q). Then, for any x R,
(i) (q : x) = R if x q,
(ii) (q : x) is p-primary if x q,
(iii) (q : x) = q if x p.
Furthermore, if R is Noetherian, there exists an element x R such that
(q : x) = p.
Proof. The case (i) is trivial, whereas (iii) follows directly from the denition of
primary ideals, since p = rad(q). To show (ii), assume x q and observe that
q (q : x) p = rad(q),
where the rst inclusion is trivial and the second follows fromthe fact that q
is p-primary. In particular, we can conclude that rad (q : x) = p. Now let
a, b R such that ab (q : x), and assume b p. Then abx q implies ax q
as b p and, hence, a (q : x). In particular, (q : x) is p-primary.
Next assume that R is Noetherian. Then p = rad(q) is nitely generated
and there exists an integer n N such that pn q. Assume that n is minimal
so that pn1 q, and choose x pn1 q. Then p (q : x), since p x pn q,
and we see from (ii) that (q : x) p. Hence, (q : x) = p.
Theorem 8. For a decomposable ideal a R, let

r
a= qi
i=1
be a minimal primary decomposition. Then the set of corresponding prime ideals

pi = rad(qi ) is independent of the chosen decomposition and, thus, depends only
on the ideal a. More precisely,
it coincides with the set of all prime ideals in R
that are of type rad (a : x) for x varying over R.
r
Proof. Let x R. The primary decomposition of a yields (a : x) = i=1 (qi : x)
and therefore by Lemma 7 a primary decomposition
r

rad (a : x) = rad (qi : x) = pi .
i=1 xqi

Now assume that the ideal rad (a : x) is prime. Then we see from 1.3/8 that it
must coincide with one of the prime ideals pi . Conversely,
x any of the ideals
pi , say pj where j {1, . . . , r}, and choose x ( i=j qi ) qj . Then, as we have
just seen, we get the primary decomposition

rad (a : x) = pi = p j
xqi
and we are done.
The prime ideals p1 , . . . , pr in the situation of Theorem 8 are uniquely as-

sociated to the ideal a. We use this fact to give them a special name.
Denition and Proposition 9. Let a R be an ideal.

(i) The set of all prime ideals in R that are of type rad (a : x) for some
x R is denoted by Ass(a); its members are called the prime ideals associated
to a.1
1
Let us point out that usually the set Ass(a) is dened by the condition in (ii), which
is equivalent to (i) only in the Noetherian case. However, we prefer to base the denition of
(ii) If R is Noetherian, a prime ideal p R belongs to Ass(a) if and only if

it is of type (a : x) for some x R.
(iii) Ass(a) is nite if R is Noetherian or, more generally, if a is decompos-
able.

Proof. Assume that R is Noetherian and let a = ri=1 q i be a minimal pri-
mary decomposition of a; see Theorem 6. Choosing x ( i=j qi ) qj for any
j {1, . . . , r}, we obtain

(a : x) = (qi : x) = (qj : x)
xqi
where (qj : x) is pj -primary by Lemma 7. Furthermore, by the same result, we

can nd an element x R such that (qj : xx ) = (qj : x) : x = pj . But then,
since xx qi for all i = j, we conclude that

(a : xx ) = (qi : xx ) = (qj : xx ) = pj .
xx qi
This shows that any prime ideal in Ass(a) is as stated in (ii). Since prime ideals
of type (ii) are also of type (i), assertion (ii) is clear. Furthermore, (iii) follows
from Theorem 8.
There is a close relationship between the prime ideals associated to an ideal

a R and the set of zero divisors in R/a. For an arbitrary ideal a R let us
call
Z(a) = (a : x) = {z R ; zx a for some x R a},
xRa
the set of zero divisors modulo a in R. Indeed, an element z R belongs to

Z(a) if and only if its residue class z R/a is a zero divisor. Furthermore, if a
power z n of some element z R belongs to Z(a), then z itself must belong to
Z(a) and therefore
Z(a) = rad (a : x) .
xRa
Corollary 10. If a is a decomposable ideal in R, then

Z(a) = p.
pAss(a)
Proof. Let p Ass(a).

Then, by the denition of Ass(a), there exists some x R
such that p = rad (a : x) . Since p is a proper ideal in R, the element x cannot
belong to a and, hence, p Z(a).
Ass(a) on the condition in (i), since this is more appropriate for handling primary decompo-
sitions in non-Noetherian situations.
Conversely, consider an element z Z(a) and let x R a such that

z (a : x). Furthermore, let a = ri=1 qi be a minimal primary decomposition
of a. Then, writing pi = rad(qi ), we get

z (a : x) = (qi : x) pi ,
xqi xqi
for x qi and coincides with R for x qi ; see

since (qi : x) is pi -primary
Lemma 7. Using x a = ri=1 qi , the set of all i such that x qi is non-empty
and we are done.

Although for a minimal primary decomposition a = ri=1 qi of some ideal
a R, proper inclusions of type qi qj are not allowed, these can nevertheless
occur on the level of prime ideals associated to a. For example, look at the
polynomial ring R = KX, Y in two variables over a eld K. Then the ideal
a = (X 2 , XY ) admits the primary decomposition a = p1 p22 where p1 = (X)
and p2 = (X, Y ). Clearly, this primary decomposition is minimal and, hence,
we get Ass(a) = {p1 , p2 } with p1 p2 .
Denition 11. Given any ideal a R, the subset of all prime ideals that
are minimal in Ass(a) is denoted by Ass (a) and its members are called the
isolated prime ideals associated to a. All other elements of Ass(a) are said to
be embedded prime ideals.
The notion of isolated and embedded prime ideals is inspired from geometry.
Indeed, passing to the spectrum Spec R of all prime ideals in R and looking at
zero sets of type V (E) for subsets E R as done in Section 6.1, a strict inclusion
of prime ideals p p is reected on the level of zero sets as a strict inclusion
V (p) V (p ). In particular, the zero set V (p ) is embedded in the bigger one
V (p) and, likewise, p is said to be an embedded prime ideal associated to a
if we have p, p Ass(a) with p p . Also recall from Theorem 8 that if a is
decomposable with primary decomposition a = ri=1 qi , then Ass(a) consists of
the nitely many radicals pi = rad(qi ). Since V (qi ) = V (pi ), one obtains

V (a) = V (p) = V (p),
pAss(a) pAss (a)
as explained at the end of Section 1.3.
Proposition 12. Let a R be an ideal and assume that it is decomposable;

for example, the latter is the case by Theorem 6 if R is Noetherian. Then:
(i) The set Ass(a) and its subset Ass (a) of isolated prime ideals a are nite.
(ii) Every prime ideal p R satisfying a p contains a member of Ass (a).
Consequently, Ass (a) consists of all prime ideals p R that are minimal among
a p R.
the ones satisfying
(iii) rad(a) = pAss (a) p is the primary decomposition of the nilradical of a.

Proof. Let a = ri=1 qi be a minimal primary decomposition and set pi = rad(qi ).
Then Ass(a) is nite, since it consists of the prime ideals p1 , . . . , pr by Theo-
rem 8. Furthermore, we obtain a p for any p Ass(a). Conversely, consider
a prime ideal p R such that a p. Then we see from a = ri=1 qi p by
1.3/8 that there is an index i satisfying qi p and, hence, pi = rad(qi ) p.
In particular, p contains an element of Ass(a) and, thus, of Ass (a). This settles
(ii), whereas (iii) is derived from the primary decomposition of a by passing to
radicals.

Looking at a primary decomposition a = ri=1 qi of some ideal a R, we
know from Theorem 8 that the associated prime ideals pi = rad(qi ) are unique;
this led us to the denition of the set Ass(a). In the remainder of this section
we want to examine the uniqueness of the primary ideals qi themselves. Not
all of them will be unique, as is demonstrated by the two dierent primary
decompositions
(X 2 , XY ) = (X) (X, Y )2 = (X) (X 2 , Y )
in the polynomial ring R = KX, Y over a eld K. We start with the following
generalization of 1.2/5:
Lemma 13. Let S R be a multiplicative system and RS the localization of R

by S. Furthermore, consider a prime ideal p R and a p-primary ideal q R.
(i) If S p = , then qRS = RS .
(ii) If S p = , then pRS is a prime ideal in RS satisfying pRS R = p
and qRS is a pRS -primary ideal in RS satisfying qRS R = q.
Proof. Since p = rad(q), we see that S p = implies S q = and, hence,

qRS = RS by 1.2/5. This settles (i). To show (ii), assume that S p = . Then
we conclude from 1.2/5 that pRS is a prime ideal in RS containing qRS and
satisfying pRS R = p. In particular, qRS is a proper ideal in RS and we can
even see that pRS = rad(qRS ).
To show qRS R = q, observe rst that q qRS R for trivial reasons.
Conversely, consider an element a qRS R. Then there are elements a q

and s S such that a1 = as in RS . So there exists t S such that (as a )t = 0
in R and, hence, ast q. Since st S, no power of it will be contained in q and
we get a q from the primary ideal condition of q. Therefore qRS R = q, as
claimed. It remains to show that qRS is a pRS -primary ideal. Indeed, look at

elements as , as RS such that aa
ss
qRS , and assume as qRS . Then aa1 qRS

and, hence, aa q, as we just have shown. Furthermore, s qRS implies a q
a

so that a p. But then as pRS and we are done.
Proposition 14. Let S R be a multiplicative system and RS the localization

of R by S. Furthermore, consider a minimal primary decomposition a = ri=1 qi
of some ideal a R. Let pi = rad(qi ) for i = 1, . . . , r and assume that r is an
integer, 0 r r, such that pi S = for i = 1, . . . , r and pi S = for
i = r + 1, . . . , r. Then qi RS is pi RS -primary for i = 1, . . . , r and

r
r
(i) aRS = qi R S as well as (ii) aRS R = qi
i=1 i=1
are minimal primary decompositions in RS and R.
Proof. The equations (i) and (ii) can be deduced from Lemma 13 if we know
that the formation of extended ideals in RS commutes with nite intersections.
However, to verify this is an easy exercise. If b, c are ideals in R, then clearly
(b c)RS bRS cRS . To show the reverse inclusion, consider an element
b
s
= sc bRS cRS where b b, c c, and s, s S. Then there is some
t S such that (bs cs)t = 0 and, hence, bs t = cst b c. But then
bs t
t (b c)RS , which justies our claim.
b
s
= ss
Now we see from Lemma 13 that (i) and (ii) are primary decompositions.
The primary decomposition (ii) is minimal, since the decomposition of a we
started with was assumed to be minimal. But then it is clear that the decompo-
sition (i) is minimal as well. Indeed, the pi RS for i = 1, . . . , r are prime ideals
satisfying pi RS R = pi . So they must be pairwise dierent, since p1 , . . . , pr are
pairwise dierent. Furthermore, the decomposition (i) is unshortenable, since
the same is true for (ii) and since qi RS R = qi for i = 1, . . . , r by Lemma 13.

Now we are able to derive the following strengthening of the uniqueness

assertion in Theorem 8:

Theorem 15. Let a R be a decomposable ideal and a = ri=1 qi a minimal
primary decomposition where qi is pi -primary. Then qi = aRpi R for all indices
i such that pi Ass (a).
In particular, a member qi of the primary decomposition of a is unique,
depending only on a and the prime ideal pi = rad(qi ) Ass(a), if pi belongs to
Ass (a). In the latter case, we call qi the pi -primary part of a.
Proof. Consider an index j {1, . . . , r} such that pj = rad(qj ) Ass (a) and
set S = R pj . Then S pj = , while S pi = for all i = j. Therefore we
can conclude from Proposition 14 that aRS R = qj .
Let R be a Noetherian ring. As an application to the above uniqueness

result we can dene for n 1 the nth symbolic power of any prime ideal p R.
Namely, consider a primary decomposition of pn , which exists by Theorem 6.
Then Ass (a) = {p} by Proposition 12 and the p-primary part of pn is unique,
given by pn Rp R. The latter is denoted by p(n) and called the nth symbolic
power of p. Of course, pn p(n) and we have pn = p(n) if and only if pn is
primary itself.
Exercises
1. Show for a decomposable ideal a R that rad(a) = a implies Ass (a) = Ass(a).
Is the converse of this true as well?
2. Let p R be a prime ideal and assume that R is Noetherian. Show that the nth
symbolic power p(n) is the smallest p-primary ideal containing pn . Can we expect
that there is a smallest primary ideal containing pn ?
3. For a non-zero Noetherian ring R let p1 , . . . , pr be its (pairwise dierent) prime
ideals that are associated to the zero ideal 0 R. Show that there is an integer
(n)
n0 N such that ri=1 pi = 0 for all n n0 and, hence, that the latter are
(n) (n )
minimal primary decompositions of the zero ideal. Conclude that pi = pi 0 for
(n)
all n n0 if pi is an isolated prime ideal and show that the symbolic powers pi
are pairwise dierent if pi is embedded.
4. Consider a decomposable ideal a R and a multiplicative subset S R such
that S a = . Then the map p - p R from prime ideals p RS to prime
ideals in R restricts to an injective map Ass(aRS ) - Ass(a).
5. Let R = KX1 , . . . Xr be the polynomial ring in nitely many variables over
a eld K and let p R be an ideal that is generated by some of the variables
Xi . Show that p is prime and that all its powers pn , n 1, are p-primary. In
particular, p(n) = pn for n 1.
6. Let M be an R-module. A submodule N M is called primary if N = M and if
the multiplication by any element a R is either injective or nilpotent on M/N ;
in other words, if there exists an element x M N such that ax N , then
there is an exponent n 1 satisfying an M N .
(a) Let (N : M ) = {a R ; aM N } and show that p = rad(N : M ) is a prime
ideal in R if N is primary in M ; we say that N is p-primary in M .
(b) Show that an ideal q R, viewed as a submodule of R, is p-primary in the
sense of (a) for some prime ideal p R if and only if it is p-primary in the sense
of Denition 1.
(c) Generalize the theory of primary decompositions from ideals to submodules
of M ; see for example Serre [24], I.B.
2.2 Artinian Rings and Modules
As we have seen in 1.5/9, a module M over a ring R is Noetherian if and only

if it satises the ascending chain condition, i.e. if and only if every ascending
chain of submodules M1 M2 . . . M becomes stationary. Switching to
descending chains we arrive at the notion of Artinian modules.
Denition 1. An R-module M is called Artinian if every descending chain of

submodules M M1 M2 . . . becomes stationary, i.e. if there is an index i0
such that Mi0 = Mi for all i i0 . A ring R is called Artinian if, viewed as an
R-module, it is Artinian.
2.2 Artinian Rings and Modules 67
Clearly, an R-module M is Artinian if and only if every non-empty set of

submodules admits a minimal element. In a similar way, Noetherian modules
are characterized by the existence of maximal elements. This suggests a close
analogy between Noetherian and Artinian modules. However, let us point out
right away that such an analogy does not exist, as is clearly visible on the level
of rings. Namely, the class of Artinian rings is, in fact, a subclass of the one
of Noetherian rings, as we will prove below in Theorem 8. Thus, for rings the
ascending chain condition is, in fact, a consequence of the descending chain
condition.
Looking at some simple examples, observe that any eld is Artinian. Fur-
thermore, a vector space over a eld is Artinian if and only if it is of nite
dimension. Since, for any prime p, the ring of integers Z admits the strictly
descending sequence of ideals Z (p) (p2 ) . . ., we see that Z cannot
be Artinian. However, any quotient Z/nZ for some n = 0 is Artinian, since it
consists of only nitely many elements.
For the discussion of general properties of Artinian modules, the Artinian
analogue of 1.5/10 is basic:
f g
Lemma 2. Let 0 - M -M - M - 0 be an exact sequence of
R-modules. Then the following conditions are equivalent:
(i) M is Artinian.
(ii) M and M are Artinian.
Proof. If M is Artinian, M is Artinian as well, since any chain of submodules in

M can be viewed as a chain of submodules in M . Furthermore, given a chain of
submodules in M , its preimage in M becomes stationary if M is Artinian. But
then the chain we started with in M must become stationary as well. Thus, it
is clear that condition (i) implies (ii).
Conversely, assume that M and M are Artinian. Then, for any descending
chain of submodules M M1 M2 . . ., we know that the chains
M f 1 (M1 ) f 1 (M2 ) . . . , M g(M1 ) g(M2 ) . . .
become stationary, say at some index i0 , and we can look at the commutative
diagrams
0 - f 1 (Mi ) - Mi - g(Mi ) -0

i

?
0 - f 1 (Mi0 ) - Mi0 - g(Mi0 ) - 0
for i i0 . Since the rows are exact, all inclusions i : Mi - Mi must be
0
bijective. Hence, the chain M M1 M2 . . . becomes stationary at i0 as
well and we see that M is Artinian.
Using the fact that Noetherian modules can be characterized by the ascend-
ing chain condition, as shown in 1.5/9, the above line of arguments yields an
alternative method for proving the Noetherian analogue 1.5/10 of Lemma 2.

Just as in the Noetherian case, let us derive some standard consequences from
the above lemma.
Corollary 3. Let M1 , M2 be Artinian R-modules. Then:

(i) M1 M2 is Artinian.
(ii) If M1 , M2 are given as submodules of some R-module M , then M1 + M2
and M1 M2 are Artinian.
Proof. Due to Lemma 2, the proof of 1.5/11 carries over literally, just replacing
Noetherian by Artinian.
Corollary 4. For a ring R, the following conditions are equivalent:

(i) R is Artinian.
(ii) Every R-module M of nite type is Artinian.
Proof. Use the argument given in the proof of 1.5/12.
There is a straightforward generalization of Lemma 2 to chains of modules.
Proposition 5. Let 0 = M0 M1 . . . Mn = M be a chain of submodules

of an R-module M . Then M is Artinian if and only if all quotients Mi /Mi1 ,
i = 1, . . . , n, are Artinian.
The same is true for Artinian replaced by Noetherian.
Proof. The only-if part is trivial. To prove the if part we use induction by n,
observing that the case n = 1 is trivial. Therefore let n > 1 and assume by
induction hypothesis that Mn1 is Artinian (resp. Noetherian). Then the exact
sequence
0 - Mn1 - M - M/Mn1 -0
shows that M is Artinian by Lemma 2 (resp. Noetherian by 1.5/10).
Corollary 6. Let R be a ring with (not necessarily distinct) maximal ideals

m1 , . . . , mn R such that m1 . . . mn = 0. Then R is Artinian if and only if it
is Noetherian.
Proof. Look at the chain of ideals
0 = m1 . . . mn . . . m1 m2 m1 R
and, for i = 1, . . . , n, consider the quotient Vi = (m1 . . . mi1 )/(m1 . . . mi )

as an R/mi -vector space. Being a vector space, Vi is Artinian if and only if it is
of nite dimension and the latter is equivalent to Vi being Noetherian. As there
is no essential dierence between Vi as an R/mi -module or as an R-module,
Proposition 5 shows that R is Artinian if and only if it is Noetherian.
2.2 Artinian Rings and Modules 69
Proposition 7. Let R be an Artinian ring. Then R contains only nitely many

prime ideals. All of them are maximal and, hence, minimal as well. In particular,
the Jacobson radical j(R) coincides with the nilradical rad(R) of R.
Proof. Let p R be a prime ideal. In order to show that p is a maximal ideal, we

have to show that the quotient R/p is a eld. So consider a non-zero element
x R/p. Since R/p is Artinian, the chain (x1 ) (x2 ) . . . will become
stationary, say at some exponent i0 , and there is an element y R/p such that
xi0 = xi0 +1 y. Since R/p is an integral domain, this implies 1 = x y and, hence,
that x is a unit. Therefore R/p is a eld and p a maximal ideal in R. Thus, all
prime ideals of R are maximal and it follows with the help of 1.3/4 that the
Jacobson radical of R coincides with the nilradical.
Finally, assume that there is an innite sequence p1 , p2 , p3 , . . . of pairwise
distinct prime ideals in R. Then
p1 p1 p2 p1 p2 p3 . . .
is a descending sequence of ideals in R and, thus, must become stationary, say at

some index i0 . The latter means p1 . . . pi0 pi0 +1 and we see from 1.3/8 that
one of the prime ideals p1 , . . . , pi0 will be contained in pi0 +1 . However, this is
impossible, since the pi are pairwise distinct maximal ideals in R. Consequently,
there exist only nitely many prime ideals in R.
In Section 2.4 we will introduce the Krull dimension dim R of a ring R. It

is dened as the supremum of all integers n such that there is a chain of prime
ideals p0 . . . pn of length n in R. At the moment we are only interested in
rings of dimension 0, i.e. rings where every prime ideal is maximal. For example,
any non-zero Artinian ring R is of dimension 0 by Proposition 7.
Theorem 8. For a non-zero ring R the following conditions are equivalent:

(i) R is Artinian.
(ii) R is Noetherian and dim R = 0.
Proof. Since Artinian rings are of dimension 0, we must show that, for rings
of Krull dimension 0, Artinian is equivalent to Noetherian. To do this we use
Corollary 6 and show that in a Noetherian or Artinian ring R of dimension 0
the zero ideal is a product of maximal ideals.
Assume rst that R is a Noetherian ring of dimension 0. Then we see from
2.1/12 applied to the ideal a = 0 that there are only nitely many prime ideals
p1 , . . . , pn in R and that the intersection of these is the nilradical rad(R). The
latter is nitely generated and, hence, nilpotent, say (rad(R))r = 0 for some
integer r > 0. Then
(p1 . . . pn )r (p1 . . . pn )r = rad(R)r = 0
shows that the zero ideal in R is a product of maximal ideals.

Now consider the case, where R is Artinian. Then R contains only nitely
many prime ideals p1 , . . . , pn by Proposition 7 and all of these are maximal in
R. In particular, we get p1 . . . pn = rad(R) from 1.3/4. As exercised in the
Noetherian case, it is enough to show that the nilradical rad(R) is nilpotent.
To do this, we proceed indirectly. Writing a = rad(R), we assume that
ai = 0 for all i N. Since R is Artinian, the descending chain a1 a2 . . .
becomes stationary, say at some index i0 . Furthermore, let b ai0 be a (non-
zero) minimal ideal such that ai0 b = 0. We claim that p = (0 : ai0 b) is a prime
ideal in R. Indeed, p = R, since ai0 b = 0. Furthermore, let x, y R such that
xy p. Assuming x p, we get ai0 bxy = 0, but ai0 bx = 0. Then bx = b by
the minimality of b and therefore ai0 by = 0 so that y p. Thus, p is prime and
necessarily belongs to the set of prime ideals p1 , . . . , pn . In particular, we have
ai0 p. But then
ai0 b = ai0 ai0 b pai0 b = 0,
which contradicts the choice of b. Therefore it follows that a = rad(R) is nilpo-

tent and we are done.
Exercises
1. Let R be a Noetherian local ring with maximal ideal m. Show that R/q is Artinian
for every m-primary ideal q R.
n
2. Let R1 , . . . , Rn be Artinian rings. Show that the cartesian product i=1 Ri is
Artinian again.
3. Let R be an Artinian ring and let p1 , . . . , pr be its (pairwise dierent) prime

ideals. Show:
r
(a) The canonical homomorphism R - n
i=1 R/pi is an isomorphism if n is
large enough.
(b) The canonical homomorphisms Rpi - R/pni , i = 1, . . . , r, are isomorphisms
if n is large enough. Consequently,
r the isomorphism of (a) can be viewed as a
canonical isomorphism R - i=1 Rpi .
4. Let A be an algebra of nite type over a eld K, i.e. a quotient of type

KX1 , . . . , Xn /a by some ideal a of the polynomial ring KX1 , . . . , Xn . Show
that A is Artinian if and only if the vector space dimension dimK A is nite.
Hint: Use Exercise 3 in conjunction with the fact to be proved in 3.2/4 that
dimK A/m < for all maximal ideals m A.
5. Call an R-module M simple if it admits only 0 and M as submodules. A

JordanHolder sequence for an R-module M consists of a decreasing sequence
of R-modules M = M0 M1 . . . Mn = 0 such that all quotients Mi1 /Mi ,
i = 1, . . . , n, are simple. Show that an R-module M admits a JordanHolder
sequence if and only if M is Artinian and Noetherian.
2.3 The ArtinRees Lemma 71
2.3 The ArtinRees Lemma
We will prove the ArtinRees Lemma in order to derive Krulls Intersection The-
orem from it. The latter in turn is a basic ingredient needed for characterizing
the Krull dimension of Noetherian rings in Section 2.4.
Lemma 1 (ArtinRees Lemma). Let R be a Noetherian ring, a R an ideal,

M a nitely generated R-module, and M M a submodule. Then there exists
an integer k N such that
(ai M ) M = aik (ak M M )
for all exponents i k.
Postponing the proof for a while, let us give some explanations concerning
this lemma. The descending sequence of ideals a1 a2 . . . denes a topology
on R, the so-called a-adic topology; see 6.1/3 for the denition of a topology.
Indeed, a subset E R is called open if for every element x E there exists
an exponent i N such that x + ai E. Thus, the powers ai for i N are
the basic open neighborhoods of the zero element in R. In a similar way, one
denes the a-adic topology on any R-module M by taking the submodules ai M
for i N as basic open neighborhoods of 0 M . Now if M is a submodule of
M , we may restrict the a-adic topology on M to a topology on M by taking
the intersections ai M M as basic open neighborhoods of 0 M . Thus, a
subset E M is open if and only if for every x E there exists an exponent
i N such that x + (ai M M ) E.
However, on M the a-adic topology exists as well and we may try to com-
pare both topologies. Clearly, since ai M ai M M , any subset E M that
is open with respect to the restriction of the a-adic topology on M to M will be
open with respect to the a-adic topology on M . Moreover, in the situation of
the ArtinRees Lemma, both topologies coincide, as follows from the inclusions
(ai M ) M = aik (ak M M ) aik M
for i k.
A topology on a set X is said to satisfy the Hausdor separation axiom
if any dierent points x, y X admit disjoint open neighborhoods. Since two
cosets with respect to ai in R, or with respect to ai M in M are disjoint as soon
as they do not coincide, it is easily seenthat the a-adic topology
on R (resp.
M ) is Hausdor if and only if we have iN ai = 0 (resp. iN ai M = 0). In
certain cases, the latter relations can be derived from the ArtinRees Lemma:
Theorem 2 (Krulls Intersection Theorem). Let R be a Noetherian ring, a R

an ideal such that a j(R), and M a nitely generated R-module. Then

ai M = 0.
iN

Proof. Applying the ArtinRees Lemma to the submodule M = iN ai M of
M , we obtain an index k N such that M = aik M for i k. Since M as
a nitely generated module over a Noetherian ring is Noetherian by 1.5/12, we
see that the submodule M M is nitely generated. Therefore Nakayamas
Lemma 1.4/10 yields M = 0.
Now in order to prepare the proof of the ArtinRees Lemma, consider a

Noetherian ring R, an ideal a R, and an R-module M of nite type together
with an a-ltration (Mi )iN . By the latter we mean a descending sequence of
submodules M = M0 M1 . . . such that ai Mj Mi+j for all i, j N. The
ltration (Mi )iN is called a-stable if there is an index k such that Mi+1 = a Mi
for all i k or, equivalently, if Mk+i = ai Mk for all i N. Furthermore,
i
let us consider the direct sum R = iN a as a ring by using component-
wise addition and by distributively extending the canonical multiplication maps
ai aj - ai+j for i, j N that are given on components. In a similar way,

we can view the direct sum M = iN Mi as an R -module. Note that R is
a graded ring in the sense of 9.1/1 and M a graded R -module in the sense of
Section 9.2.
Observe that the ring R is Noetherian. Indeed, R is Noetherian and, hence,
the ideal a admits a nite system of generators a1 , . . . , an . Thus, the canonical
injection R - R extends to a surjection RX1 , . . . , Xn - R by sending
the polynomial variable X to a for = 1, . . . , n. At this point a has to be
viewed as an element in the rst power a1 , which is the component of index 1 in
R . Since the polynomial ring RX1 , . . . , Xn is Noetherian by Hilberts Basis
Theorem 1.5/14, we see that R is Noetherian as well.
Lemma 3. For a Noetherian ring R, an ideal a R, and an R-module M

of nite type, let R and M be as before. Then the following conditions are
equivalent:
(i) The ltration (Mi )iN is a-stable.
(ii) M is an R -module of nite type.
Proof. Starting with the implication (i) = (ii), let k be an integer such that
Mk+i = ai M k for all i N. Then M is generated as an R -module by
the subgroup ik Mi M . Since M is a nitely generated R-module over
a Noetherian ring, it is Noetherian by 1.5/12. Hence, all its submodules Mi
are nitely generated and the same is true for the nite direct sum ik Mi .
Choosing a nite system of R-generators for the latter, it generates M as an
R -module.
Conversely, assume that M is a nitely generated R -module. Then we
can choose a nite system of homogeneous generators, namely of generators
x1 , . . . , xn , where x M() for certain integers (1), . . . , (n). It follows that
aMi = Mi+1 for i max{(1), . . . , (n)} and, hence, that the ltration (Mi )iN
of M is a-stable.
2.3 The ArtinRees Lemma 73
Now the proof of the ArtinRees Lemma is easy to achieve. Consider the
ltration (Mi )iN of M given by Mi = ai M ; it is a-stable by its denition.
Furthermore, consider the induced ltration (Mi )iN on M ; the latter is given
by Mi = Mi M . Then M is canonically an R -submodule of M . Since M
is of nite type by the Lemma 3 and R is Noetherian, we see from 1.5/12 that
M is of nite type as well. Thus, by Lemma 3 again, the ltration (Mi )iN is
a-stable, and there exists an integer k such that a(ai M M ) = ai+1 M M
for i k. Iteration yields ai (ak M M ) = ak+i M M for i N and, thus, the
assertion of the ArtinRees Lemma.
Exercises
1. Let M be an R-module. Verify for an ideal a R that the a-adic topology on M ,

as dened above, is a topology in the sense of 6.1/3. Show that
the a-adic topology
on M satises the Hausdor separation axiom if and only if iN ai M = 0.
2. Consider the a-adic topology on an R-module M for some ideal a R. Show

that a submodule N M is closed in M if and only if iN (ai M + N ) = N . For
R Noetherian, M nitely generated and a j(R), deduce that every submodule
of M is closed.
3. Generalization of Krulls Intersection Theorem: Consideran ideal a of a Noethe-

rian ring R and a nitely generated R-module M . Then iN ai M consists of all
M that are annihilated by 1 + a for some element a a. In partic-
elements x
ular, show iN ai = 0 for any proper ideal a of a Noetherian integral domain.
Hint: Use the generalized version of Nakayamas Lemma in Exercise 1.4/5.
4. For a eld K and an integer n > 1 consider the homomorphism of formal power
series rings : KX - KX given by f (X) - f (X n ). Show that is
an injective homomorphism between local rings and satises (X) (X) for the
maximal ideal (X) KX. Construct a ring R as the union of the innite
chain of rings

KX - KX - KX - . . .
and show that R is a local integral domain where Krulls Intersection Theorem
is not valid any more. In particular, R cannot be Noetherian. Hint: For the
denition of formal power series rings see Exercise 1.5/8.

5. Given an ideal a R, show that gra (R) = i i+1 is canonically a ring;
iN a /a
it is called the graded ring associated to a. Show that gra (R) is Noetherian if R
is Noetherian. Furthermore, let M= M0 M1 . . . be an a-ltration on an
R-module M . Show that gr(M ) = iN Mi /Mi+1 is canonically a gra (R)-module
and that the latter is nitely generated if M is a nite R-module and the ltration
on M is a-stable.
2.4 Krull Dimension
In order to dene the dimension of a ring R, we use strictly ascending chains

p0 p1 . . . pn of prime ideals in R where the integer n is referred to as the
length of the chain.
Remark 1. Let R be a ring and p R a prime ideal. Then:

(i) The chains of prime ideals in R starting with p correspond bijectively to
the chains of prime ideals in R/p starting with the zero ideal.
(ii) The chains of prime ideals in R ending with p correspond bijectively to
the chains of prime ideals in the localization Rp ending with pRp .
Proof. Assertion (i) is trivial, whereas (ii) follows from 1.2/5.
Denition 2. For a ring R, the supremum of lengths n of chains
p 0 p1 . . . pn ,
where the pi are prime ideals in R, is denoted by dim R and called the Krull
dimension or simply the dimension of R.
For example, elds are of dimension 0, whereas a principal ideal domain is

of dimension 1, provided it is not a eld. In particular, we have dim Z = 1, as
well as dim KX = 1 for the polynomial ring over a eld K. Also we know that
dim KX1 , . . . , Xn n, since the polynomial ring KX1 , . . . , Xn contains the
chain of prime ideals 0 (X1 ) (X1 , X2 ) . . . (X1 , . . . , Xn ). In fact, we will
show dim KX1 , . . . , Xn = n in Corollary 16 below. Likewise, the polynomial
ring KX1 , X2 , . . . in an innite sequence of variables is of innite dimension,
whereas the zero ring 0 is a ring having dimension , since by convention
the supremum over an empty subset of N is . Any non-zero ring contains
at least one prime ideal and therefore is of dimension 0.
Denition 3. Let R be a ring.

(i) For a prime ideal p R its height ht p is given by ht p = dim Rp .
(ii) For an ideal a R, its height ht a is given by the inmum of all heights
ht p, where p varies over all prime ideals in R containing a.
(iii) For an ideal a R, its coheight coht a is given by coht a = dim R/a.
Thus, for a prime ideal p R, its height ht p (resp. its coheight coht p)
equals the supremum of all lengths of chains of prime ideals in R ending at p
(resp. starting at p). In particular, we see that ht p + coht p dim R.
In order to really work with the notions of dimension and height, it is
necessary to characterize these in terms of so-called parameters. For example,
we will show that an ideal a of a Noetherian ring satises ht a r if it can be
generated by r elements; see Krulls Dimension Theorem 6 below. Furthermore,
we will prove that any Noetherian local ring R of dimension r admits a system of
2.4 Krull Dimension 75
r elements, generating an ideal whose radical coincides with the maximal ideal
m R; see Proposition 11 below. If m itself can be generated by r elements,
we face a special case, namely, where R is a so-called regular local ring.
To prepare the discussion of such results, we need a special case of Krulls
Intersection Theorem 2.3/2.
Lemma 4. Let R be a Noetherian ring and p R a prime ideal. Then, looking

at the canonical map R - Rp from R into its localization by p, we have

ker(R - Rp ) = p(i) ,
iN
where p(i) = pi Rp R is the ith symbolic power of p, as considered at the end

of Section 2.1.
Proof. Since Rp is a local ring with maximal ideal pRp by 1.2/7 and since
we can conclude from 1.2/5 (ii) that Rpis Noetherian, Krulls Intersection
Theorem 2.3/2 is applicable and yields iN pi Rp = 0. Using the fact that
taking preimages with respect to R - Rp is compatible with intersections,
we are done.
Now we can prove a rst basic result on the height of ideals in Noetherian
rings, which is a special version of Krulls Principal Ideal Theorem to be derived
in Corollary 9 below. For any ideal a R we consider the set of minimal prime
divisors of a; thereby we mean the set of prime ideals p R that are minimal
among those satisfying a p. In the setting of 2.1/12, this is the set Ass (a)
of isolated prime ideals associated to a. Also note that, for a prime ideal p
containing a there is always a minimal prime divisor of a contained in p. This
follows from 2.1/12 (ii) or without using the theory of primary decompositions
by applying Zorns Lemma.
Lemma 5. Let R be a Noetherian integral domain and consider a non-zero

element a R that is not a unit. Then ht p = 1 for every minimal prime
divisor p of (a).
Proof. For a minimal prime divisor p of (a), we can pass from R to its local-
ization Rp and thereby assume that R is a Noetherian local integral domain
with maximal ideal p. Then we have to show that any prime ideal p0 R that
is strictly contained in p satises p0 = 0. To do this, look at the descending
sequence of ideals
(i)
ai = p0 + (a), i N,
(i)
where p0 is the ith symbolic power of p0 , and observe that R/(a) is Artinian by
2.2/8. Indeed, R/(a) is Noetherian and satises dim R/(a) = 0, as p is the only
prime ideal containing a. Therefore the sequence of the ai becomes stationary,
say at some index i0 . Hence,
(i ) (i)
p0 0 p0 + (a), i i0 ,
and, in fact
(i ) (i) (i )
p0 0 = p0 + ap0 0 , i i0 ,
(i ) (i )
using the equality (p0 0 : a) = p0 0 ; the latter relation follows from 2.1/7, since
(i )
p0 0 is p0 -primary and since a p0 . Now Nakayamas Lemma in the version of
(i ) (i)
1.4/11 implies p0 0 = p0 for i i0 . Furthermore, since R is an integral domain,
Lemma 4 shows (i)
p0 = ker(R - Rp0 ) = 0.
(i )
p0 0 =
iN
(i )
But then we conclude from pi00 p0 0 that p0 = 0 and, thus, that ht p = 1.
Theorem 6 (Krulls Dimension Theorem). Let R be a Noetherian ring and

a R an ideal generated by elements a1 , . . . , ar . Then ht p r for every
minimal prime divisor p of a.
Proof. We conclude by induction on the number r of generators of a. The case

r = 0 is trivial. Therefore let r > 0 and consider a minimal prime divisor p of a
with a strictly ascending chain of prime ideals p0 . . . pt = p in R. It has to
be shown that t r. To do this, we can assume that R is local with maximal
ideal p and, hence, that p is the only prime ideal in R containing a. Furthermore,
we may suppose t > 0 and, using the Noetherian hypothesis, that there is no
prime ideal strictly between pt1 and pt . Since a pt1 , there exists a generator
of a, say ar , that is not contained in pt1 . Then p = pt is the only prime ideal
containing pt1 + (ar ), and it follows from 1.3/6 (iii) that rad(pt1 + (ar )) = p.
Thus, there are equations of type
ani = ai + ar yi , i = 1, . . . , r 1,
for some exponent n > 0 where ai pt1 and yi R. Now consider the ideal
a = (a1 , . . . , ar1 ) R. Since a pt1 , there is a minimal prime divisor p of
a such that p pt1 . By the above equations we have

p = rad(a) rad a + (ar ) rad p + (ar ) p.
In particular, p is a minimal prime divisor of p + (ar ) or, in other words, p/p is

a minimal prime divisor of ar R/p in R/p . But then we get ht(p/p ) = 1 from
Theorem 5 and we can conclude p = pt1 from p pt1 pt = p. Thus, pt1
turns out to be a minimal prime divisor of a . Since the latter ideal is generated
by r 1 elements, we get t 1 r 1 by induction hypothesis and, thus, are
done.
Let us list some immediate consequences from Krulls Dimension Theorem:
Corollary 7. Let a be an ideal of a Noetherian ring R. Then ht a < .

Corollary 8. Let R be a Noetherian local ring with maximal ideal m. Then

dim R dimR/m m/m2 < .
Proof. First note that dimR/m m/m2 < , since R is Noetherian and, hence, m is
nitely generated. Choosing elements a1 , . . . , ar m giving rise to an R/m-basis
of m/m2 , we conclude from Nakayamas Lemma in the version of 1.4/12 that
the ai generate m. But then dim R = ht m r = dimR/m m/m2 by Krulls
Dimension Theorem.
Corollary 9 (Krulls Principal Ideal Theorem). Let R be a Noetherian ring and

a an element of R that is neither a zero divisor nor a unit. Then every minimal
prime divisor p of (a) satises ht p = 1.
Proof. We get ht p 1 from Krulls Dimension Theorem. Since the minimal

prime ideals in R are just the isolated prime ideals associated to the zero ideal
0 R by 2.1/12, the assertion follows from the characterization of zero divisors
given in 2.1/10.
Next, to approach the subject of parameters, we prove a certain converse of

Krulls Dimension Theorem.
Lemma 10. Let R be a Noetherian ring and a an ideal in R of height ht a = r.

Assume there are elements a1 , . . . , as1 a such that ht(a1 , . . . , as1 ) = s 1
for some s N where 1 s r. Then there exists an element as a such that
ht(a1 , . . . , as ) = s.
In particular, there are elements a1 , . . . , ar a such that ht(a1 , . . . , ar ) = r.
Proof. We have only to justify the rst statement, as the second one follows from
the rst by induction. Therefore consider elements a1 , . . . , as1 a, 1 s r,
such that ht(a1 , . . . , as1 ) = s 1, and let p1 , . . . , pn R be the minimal
prime divisors of (a1 , . . . , as1 ). Then, using Krulls Dimension Theorem, we
have ht pi s 1< r for all i. Since ht a = r, this impliesa pi for all
i and, hence, a ni=1 pi by 1.3/7. Thus, choosing as a ni=1 pi , we get
ht(a1 , . . . , as ) s and, in fact ht(a1 , . . . , as ) = s by Krulls Dimension Theorem.

For the discussion of parameters, which follows below, recall from 2.1/3 that
an ideal a of a local ring R with maximal ideal m is m-primary if and only if
rad(a) = m.
Proposition 11. Let R be a Noetherian local ring with maximal ideal m. Then
there exists an m-primary ideal in R generated by d = dim R elements, but no
such ideal that is generated by less than d elements.
Proof. Combine Lemma 10 with Krulls Dimension Theorem 6.

Denition 12. Let R be a Noetherian local ring with maximal ideal m. A set
of elements x1 , . . . , xd m is called a system of parameters of R if d = dim R
and the ideal (x1 , . . . , xd ) R is m-primary.
In particular, we see from the Proposition 11 that every Noetherian local

ring admits a system of parameters.
Proposition 13. Let R be a Noetherian local ring with maximal ideal m. Then,
for given elements x1 , . . . , xr m, the following conditions are equivalent:
(i) The system x1 , . . . , xr can be enlarged to a system of parameters of R.
(ii) dim R/(x1 , . . . , xr ) = dim R r.
Furthermore, conditions (i) and (ii) are satised if ht(x1 , . . . , xr ) = r.
Proof. First observe that condition (i) follows from ht(x1 , . . . , xr ) = r, using
Lemma 10. Next consider elements y1 , . . . , ys m. Writing R = R/(x1 , . . . , xr )
and m = m/(x1 , . . . , xr ), the residue classes y 1 , . . . , y s generate an m-primary
ideal in R if and only if x1 , . . . , xr , y1 , . . . , ys generate an m-primary ideal in R.
In particular, if x1 , . . . , xr , y1 , . . . , ys is a system of parameters of R, as we may
assume in the situation of (i), we have r + s = dim R and can conclude from
Proposition 11 that dim R dim Rr. On the other hand, if the residue classes
of y1 , . . . , ys form a system of parameters in R, we have s = dim R and see that
(x1 , . . . , xr , y1 , . . . , ys ) is an m-primary ideal in R. Therefore dim R r + dim R
by Proposition 11. Combining both estimates yields dim R dim Rr dim R
and, thus, dim R = dim R r, as needed in (ii).
Conversely, assume (ii) and consider elements y1 , . . . , ys m, whose residue
classes form a system of parameters in R so that s = dim R. Then the system
x1 , . . . , xr , y1 , . . . , ys generates an m-primary ideal in R. Since r + s = dim R by
(ii), the system is, in fact, a system of parameters and (i) follows.
Corollary 14. Let R be a Noetherian local ring with maximal ideal m, and let
a m be an element that is not a zero divisor. Then dim R/(a) = dim R 1.
Proof. Use the fact that ht(a) = 1 by Corollary 9.
As an application of the theory of parameters, let us discuss the dimension

of polynomial rings.
Proposition 15. Consider the polynomial ring RX1 , . . . , Xn in n variables

over a Noetherian ring R. Then dim RX1 , . . . , Xn = dim R + n.
Proof. We will show dim RX = dim R + 1 for one variable X, from which
the general case follows by induction. Let p0 . . . pr be a strictly ascending
chain of prime ideals in R of length r. Then
p0 RX . . . pr RX pr RX + XRX

is a strictly ascending chain of prime ideals of length r + 1. From this we see

that dim RX dim R + 1. To show that this is actually an equality, consider
a maximal ideal m RX and set p = m R. Then p is a prime ideal in R
and it is enough to show that ht m ht p + 1. To do this, we replace R by its
localization Rp and interpret the polynomial ring Rp X as the localization of
RX by the multiplicative system R p. Thereby we have reduced our claim
to the case where R is a Noetherian local ring with maximal ideal p; use 1.2/7.
Then R/p is a eld and RX/pRX R/pX a principal ideal domain.
Hence, there is a polynomial f m such that
m = pRX + f RX.
Now let x1 , . . . , xr be a system of parameters of R where r = dim R. Then

x1 , . . . , xr , f generate an ideal in RX whose radical is m. Therefore Theorem 6
shows ht m r + 1 = ht p + 1, as claimed.
Corollary 16. dim KX1 , . . . , Xn = n for the polynomial ring in n variables

over a eld K.
In particular, we can conclude for nitely generated K-algebras, i.e. quo-

tients of polynomial rings of type KX1 , . . . , Xn , that their dimension is nite.
Making use of the fact to be proved later in 3.2/4 that any maximal ideal
m KX1 , . . . , Xn leads to a residue eld KX1 , . . . , Xn /m that is nite
over K, we can even derive the following more specic version of Corollary 16:
Proposition 17. Consider the polynomial ring KX1 , . . . , Xn in n variables

over a eld K and a maximal ideal m KX1 , . . . , Xn . Then:
(i) m is generated by n elements.
(ii) ht m = n.
(iii) The localization KX1 , . . . , Xn m is a local ring of dimension n, whose
maximal ideal is generated by a system of parameters.
Proof. To establish (i) and (ii), we use induction on n, the case n = 0 being
trivial. So assume n 1. Let n = m KX1 , . . . , Xn1 and consider the
inclusions
K - KX1 , . . . , Xn1 /n - KX1 , . . . , Xn /m.
Then KX1 , . . . , Xn /m is a eld that is nite over K by 3.2/4 and, hence,
nite over KX1 , . . . , Xn1 /n. In particular, using 3.1/2, the latter is a eld as
well. Therefore n is a maximal ideal in KX1 , . . . , Xn1 , and we may assume
by induction hypothesis that n, as an ideal in KX1 , . . . , Xn1 , is generated
by n 1 elements.
Now look at the canonical surjection

KX1 , . . . , Xn1 /n Xn - KX1 , . . . , Xn /m
sending Xn onto its residue class in KX1 , . . . , Xn /m. Since on the left-
hand side we are dealing with a principal ideal domain, there is a polynomial
f KX1 , . . . , Xn such that m is generated by n and f . In particular, m is
generated by n elements and assertion (i) is clear. Furthermore, we see that
ht m n by Krulls Dimension Theorem 6. On the other hand, since

KX1 , . . . , Xn /(n) KX1 , . . . , Xn1 /n Xn ,
it is clear that n generates a prime ideal in KX1 , . . . , Xn dierent from m so

that n m. Using ht n = n 1 from the induction hypothesis, we get ht m n
and, hence, ht m = n, showing (ii).
Finally, (iii) is a consequence of (i) and (ii).
To end this section, we briey want to touch the subject of regular local
rings.
Proposition and Denition 18. For a Noetherian local ring R with maximal
ideal m and dimension d the following conditions are equivalent:
(i) There exists a system of parameters of length d in R generating the
maximal ideal m; in other words, m is generated by d elements.
(ii) dimR/m m/m2 = d.
R is called regular if it satises the equivalent conditions (i) and (ii).
Proof. Assume (i) and let x1 , . . . , xd be a system of parameters generating

m. Then, as an R/m-vector space, m/m2 is generated by d elements and,
hence, dimR/m m/m2 d. Since dim R dimR/m m/m2 by Corollary 8, we get
dimR/m m/m2 = d and therefore (ii).
Conversely, assume (ii). Then m/m2 is generated by d elements and the same
is true for m by Nakayamas Lemma in the version of 1.4/12. In particular, such
a system of generators is a system of parameters then.
For example, we see from Hilberts Basis Theorem 1.5/14 in conjunction

with Proposition 17 that for a polynomial ring KX1 , . . . , Xn over a eld K
its localization KX1 , . . . , Xn m at a maximal ideal m is an example of a regular
Noetherian local ring of dimension n. It is known that any localization Rp of a
regular Noetherian local ring R by a prime ideal p R is regular again; see Serre
[24], Prop. IV.23. Also, let us mention the Theorem of AuslanderBuchsbaum
stating that any regular Noetherian local ring is factorial; see Serre [24], Cor. 4
of Thm. IV.9 for a proof of this fact.
To deal with some more elementary properties of regular local rings, it is
quite convenient to characterize Krull dimensions of rings not only via lengths
of ascending chains of prime ideals, or systems of parameters, as we have done,
but also in terms of the so-called Hilbert polynomial. For example, see Atiyah
Macdonald [2], Chapter 11, for such a treatment. From this it is easily seen that
regular Noetherian local rings are integral domains. Since we need it later on,
we will prove this fact by an ad hoc method.
Proposition 19. Let R be a regular Noetherian local ring. Then R is an integral

domain.
Proof. We argue by induction on d = dim R. Let m be the maximal ideal of R.

If d = 0, then m is generated by 0 elements and, hence, m = 0. Therefore R is a
eld. Now let dim R > 0 and let x1 , . . . , xd be a system of parameters generating
m. Furthermore, let p1 , . . . , pr be the minimal prime ideals in R; their number
is nite by 2.1/12. We claim that we can nd an element of type

d
a = x1 + ci x i m m 2
i=2
for some coecients ci R such that a is not contained in any of the minimal
prime ideals p1 , . . . , pr . Clearly, any element a of this type cannot be contained
in m2 since x1 , . . . , xd give rise to an R/m-vector space basis of m/m2 .
Using a recursive construction for a, assume that a p1 , . . . , pt for some
t < r, but that a pt+1 . Applying 1.3/8, there exists an element

t
c pj pt+1
j=1
and we can nd an index i0 {2, . . . , d} such that cxi0 pt+1 . Otherwise,

since c pt+1 and cxi pt+1 implies xi pt+1 , all elements a, x2 , . . . , xd would
be contained in pt+1 so that m pt+1 . However, this contradicts the fact that
dim R = d > 0. Then we see that a = a + cxi0 is of the desired type: a
is not contained in p1 , . . . , pt , since this is true for a and since c tj=1 pj .
Furthermore, a pt+1 , since a pt+1 and cxi0 pt+1 .
Thus, we have seen that there exists an element a m m2 as specied
above that is not contained in any of the minimal prime ideals p1 , . . . , pr of R.
Now look at the quotient R/(a). Its dimension is d 1 by Proposition 13, since
a, x2 , . . . , xd form a system of parameters of R. Furthermore, x2 , . . . , xd give
rise to a system of parameters of R/(a) generating the maximal ideal of this
ring. Therefore R/(a) is regular and, thus, by induction hypothesis, an integral
domain. In particular, we see that the ideal (a) R is prime. Now consider
a minimal prime ideal of R that is contained in (a), say p1 (a). Then any
element y p1 is of type ab for some b R and, in fact b p1 , since a p1 .
Therefore p1 = (a) p1 and Nakyamas Lemma 1.4/10 shows p1 = 0. Hence, R
is an integral domain.
Exercises
1. Consider the polynomial ring RX in one variable over a not necessarily Noethe-
rian ring. Show dim R + 1 dim RX 2 dim R + 1. Hint: Let p1 p2 RX
be two dierent prime ideals in RX restricting to the same prime ideal p R.
Deduce that p1 = pRX.
2. Let KX, Y be the polynomial ring in two variables over a eld K. Show
dim KX, Y /(f ) = 1 for any non-zero polynomial f KX, Y that is not
constant.
3. Let R be a Noetherian local ring with maximal ideal m. Show for a1 , . . . , ar m
that dim R/(a1 , . . . , ar ) dim R r. Hint: Assume r = 1. Show for any chain of
prime ideals p0 . . . pn where a1 pn that there is a chain of prime ideals
p1 . . . pn satisfying a1 p1 and pn = pn ; use induction on n.
4. Consider the formal power series ring R = KX1 , . . . , Xn in nitely many vari-
ables over a eld K. Show that R is a regular Noetherian local ring of dimension
n. Hint: Use Exercise 1.5/8 for the fact that R is Noetherian.
5. Let R be a regular Noetherian local ring. Show that R is a eld if dim R = 0.
Show that R is a discrete valuation ring, i.e. a local principal ideal domain, if
dim R = 1.
6. Let R be a regular Noetherian local ring of dimension d with maximal ideal
m R. Show for elements a1 , . . . , ar m that the quotient R/(a1 , . . . , ar ) is
regular of dimension d r if and only if the residue classes a1 , . . . , ar m/m2 are
linearly independent over the eld R/m.
7. Let R be a regular Noetherian local ring of dimension d with maximal ideal m.
Let m = (a1 , . . . , ad ) and set k = R/m. Show for polynomial variables X1 , . . . , Xd
that the canonical k-algebra homomorphism

kX1 , . . . , Xd - grm (R) = mi /mi+1 , Xj - aj ,
iN
where aj is the residue class of aj in m/m2 , is an isomorphism of k-algebras.

Hints: See Exercise 2.3/5 for the fact that the graded ring grm (R) associated to
m is Noetherian if R is Noetherian. Proceed by induction,
similarly as in the
proof of Proposition 19, and look at the maximal ideal i>0 mi /mi+1 grm (R).
Show that this maximal ideal cannot be a minimal prime ideal in grm (R) if
d = dim R > 0.
3. Integral Extensions
Recall that an extension of elds K - L is called algebraic if each element

x L satises an algebraic equation over K, i.e. an equation of type
xn + a1 xn1 + . . . + an = 0
for suitable coecients ai K. Replacing K - L by an arbitrary (not

necessarily injective) ring homomorphism : R - R , equations of the just
mentioned type are still meaningful; they are referred to as integral equations.
Furthermore, R is said to be integral over R (via ) if every element x R
satises an integral equation over R.
The fact that any nite extension of elds is algebraic, is fundamental in
eld theory. In 3.1/5 we will generalize this result to ring extensions and show
that any ring homomorphism : R - R , which is nite in the sense that it
equips R with the structure of a nite R-module, is integral. We obtain this
assertion from a quite general characterization of integral dependence in terms
of a module setting; see Lemma 3.1/4. The proof is based on Cramers rule and
is much more laborious than in the eld case.
To give a simple example illustrating a basic application of Lemma 3.1/4,
consider the polynomial ring RX in one variable X over a non-zero ring R
and x a monic polynomial
f = X n + a1 X n1 + . . . + an RX
with coecients ai R. For a second variable Y look at the canonical morphism
: RY - RX, Y - f.
We claim that is nite and, hence, integral. Of course, the equation
X n + a1 X n1 + . . . + (an (Y )) = 0
shows that X is integral over RY . From this we conclude by induction that
the RY -submodule generated by X 0 , . . . , X n1 in RX contains all powers of
X and, hence, coincides with RX. In other words, is nite. Alternatively,
we could have derived this fact directly from 3.1/4 (ii). Furthermore, using
3.1/4 (iii) or 3.1/5, it follows that is integral. The latter is a non-trivial fact,
which cannot be derived by a direct ad hoc computation.

84 3. Integral Extensions
Without doubt, Lemma 3.1/4 is the key to handling integral dependence. We

use it in advanced settings, where we relate normal rings to discrete valuation
rings (see the reference below), but also for deriving various standard facts on
integral extensions such as the transitivity property; see 3.1/7. In particular, for
any ring homomorphism : R - R , we can dene its integral closure in R .
Just consider the set of all elements x R that are integral over R. The latter
is a subring of R that is integrally closed, as is seen in 3.1/8. Furthermore,
an integral domain R is called normal if it is integrally closed in its eld of
fractions Q(R). We show in 3.1/10 that any factorial ring is normal, whereas
the reverse is true for Noetherian local integral domains of Krull dimension 1.
In fact, a normal Noetherian local integral domain of dimension 1 is principal
or, as we will say later, a discrete valuation ring; see 9.3/4. This result will serve
in Section 9.3 as an important ingredient for the discussion of so-called divisors
on schemes.
As a more basic application of integral dependence we study in Sec-
tion 3.2 nitely generated algebras over a eld K, or K-algebras of nite
type as we will say. We thereby mean quotients of polynomial rings of type
KX1 , . . . , Xn by some ideal, the Xi being variables. The key tool for attack-
ing such algebras is Noethers Normalization Lemma 3.2/1, asserting that for
every non-zero K-algebra A of nite type there exists a nite (and, hence, in-
tegral) K-monomorphism KY1 , . . . , Yd - A, for a certain set of variables
Y1 , . . . , Yd . From this we can immediately conclude in 3.2/3 that any extension
of elds K - L equipping L with the structure of a K-algebra of nite type
is, in fact, nite algebraic. For example, such a situation occurs when we con-
sider a maximal ideal m in some K-algebra of nite type A and look at the
extension K - A/m. The latter is nite (3.2/4) and even an isomorphism if
K is algebraically closed. This result is sometimes referred to as the weak form
of Hilberts Nullstellensatz and we can easily derive from it the full version; see
3.2/5 and 3.2/6. In pure terms of commutative algebra Hilberts Nullstellensatz
states that any K-algebra A of nite type is Jacobson, i.e. for any ideal a A the
nilradical rad(a), which consists of all elements a A that are nilpotent modulo
a, coincides with the Jacobson radical j(a), which equals the intersection of all
maximal ideals containing a. On the other hand, Hilberts Nullstellensatz is of
geometric signicance from the viewpoint of classical Algebraic Geometry,
as we explain at the end of Section 3.2; see in particular 3.2/6.
A natural question to study is how the Krull dimension of rings, as discussed
in Section 2.4, behaves with respect to integral ring extensions. We do this in
Section 3.3. As a rst result we prove the Lying-over Theorem 3.3/2. It asserts
for an integral ring extension : R - R and a prime ideal p R satisfying
ker p that there exists at least one prime ideal P R lying over p, namely
satisfying P R = p and, furthermore, that between dierent prime ideals
P1 , P2 R with this property there cannot exist any inclusion relation. The
next objective is to lift ascending chains of prime ideals step by step from R
to R (Going-up Theorem 3.3/3) and then to settle the more delicate question
of lifting descending chains of prime ideals (Going-down Theorem 3.3/4). As a
3.1 Integral Dependence 85
corollary we see that dim R = dim R if is an integral monomorphism. Using

this we can show that the Krull dimension of any integral domain A that is of
nite type over a eld K equals the transcendence degree of the eld of fractions
Q(A) over K; see 3.3/7.
3.1 Integral Dependence
Given any ring homomorphism : R - R , it is quite convenient for our

purposes to view R as an R-algebra with respect to ; see 1.4/3. In particular,
R carries then a structure of an R-module where products of type r r for
r R, r R are dened by r r = (r) r , using the multiplication on R .
Denition 1. Let : R - R be a homomorphism of rings. An element

x R is called integral over R with respect to , or is said to depend integrally
on R with respect to , if it satises a so-called integral equation over R, i.e.
if there are elements a1 , . . . , an R such that
xn + a1 xn1 + . . . + an = 0.
The ring R is called integral over R, or is called integral, if each x R is

integral over R.
Furthermore, is called nite if it equips R with the structure of a nite
R-module.
If R = K and R = K are elds, then is injective and we may view K as

a subeld of K . Recall that the extension of elds K K is called algebraic
if K is integral over K.
Remark 2. Let : R - R be a monomorphism of integral domains such

that R is integral over R. Then R is a eld if and only if R is a eld.
Proof. Assume rst that R is a eld and let x = 0 be an element in R . Then x

satises an integral equation over R,
xn + a1 xn1 + . . . + an = 0, a1 , . . . , an R.
Dividing out a suitable power of x, we may assume an = 0 and, hence, that an

is invertible in R. In the eld of fractions of R , we can multiply the equation
by x1 . This yields
x1 = a1
n (x
n1
+ a1 xn2 + . . . + an1 ) R ,
hence R is a eld.
Conversely, if R is a eld, consider an element x R, x = 0. Then x1 R
satises an integral equation over R,
xn + a1 xn+1 + . . . + an = 0, a1 , . . . , an R.
Multiplying by xn1 , we see
x1 = a1 a2 x . . . an xn1 R,
hence R is a eld.
Remark 3. Let : R - R be a homomorphism of rings, which is integral

(resp. nite).
(i) If a R and a R are ideals satisfying (a) a , then the homo-
morphism R/a - R /a induced from is integral (resp. nite).
(ii) For any multiplicative system S R, the induced homomorphism
RS - R
(S) is integral (resp. nite).
Proof. Assertion (i) is trivially veried using the denition of integral (resp.
nite) maps. To justify (ii), observe rst that (S) is a multiplicative system in
R and that induces a homomorphism RS - R(S)
by the universal property

of localizations 1.2/8. Now consider an element (s) x
R(S) with x R and
s S. Then, from an integral equation
xn + a1 xn1 + . . . + an = 0, a1 , . . . , an R,
of x over R, we derive the integral equation

x n a x n1 an
1
+ + ... + n = 0
(s) s (s) s
x
for (s) over RS . Furthermore, if x1 , . . . , xn generate R as an R-module, the

corresponding fractions x11 , . . . , x1n will generate R(S) as an RS -module.
There is a basic characterization of integral dependence in terms of niteness

conditions, which is presented next.
Lemma 4. Consider a homomorphism of rings : R - R and an element

x R . The following conditions are equivalent:
(i) x is integral over R.
(ii) The subring Rx R generated by (R) and x in R is nitely gen-
erated, when viewed as an R-module.
(iii) There exists a nitely generated R-submodule M R such that 1 M
and xM M .
(iv) There exists an Rx-module M such that M is an R-module of nite
type and aM = 0 for any a Rx implies a = 0.
Proof. Let us start with the implication (i) = (ii). So assume there is an
integral equation
xn + a1 xn1 + . . . + an = 0

with coecients a1 , . . . , an R. Then xn belongs to M = n1 i
i=0 Rx , and we
see by induction that x M for all i N. H ence, Rx M and, in fact,
i
Rx = M . In particular, Rx is a nitely generated R-module, as required in

(ii).
The implications (ii) = (iii) and (iii) = (iv) are trivial. Thus, it remains
to justify (iv) = (i). Let M be an Rx-module and assume there are elements
y1 , . . . , yn M such that M = ni=1 Ryi . Then the inclusion xM M leads to
a system of equations
xy1 = a11 y1 + . . . + a1n yn ,

...
...
...
xyn = an1 y1 + . . . + ann yn ,
with coecients aij R. In terms of matrices we can write

y1
= 0,

yn
where = (ij x aij )i,j=1,...,n (Rx)nn ; here ij is Kroneckers delta, which

is given by ij = 1 for i = j and ij = 0 for i = j. Now consider Cramers rule,
i.e. the relation
() ad = (det ) I
with ad (Rx)nn the adjoint matrix of and I (Rx)nn the unit

matrix, as well as det Rx the determinant of . In Linear Algebra
this equation is established for matrices with coecients in a eld, but it applies
to general rings as well. Indeed, comparing coecients of the matrices on both
sides of the equation (), we get a system of polynomial identities involving the
coecients of . To justify these identities, we can view the coecients cij of
as variables and work over the ring Zcij , a case which can be reduced to
the situation of coecients in a eld if we pass to the eld of fractions Q(cij ).
Using Cramers rule (), we get

y1 y1
(det ) = ad = 0

yn yn
and, thus, (det ) yi = 0 for i = 1, . . . , n. As M is generated by y1 , . . . , yn , we

see that (det ) M = 0 and, hence, by our assumption in (iv), that det = 0.
Therefore
det(ij X aij ) RX
is a monic polynomial vanishing at x, as desired.
Corollary 5. Each nite homomorphism of rings R - R is integral.
Proof. Use condition (iii) of Lemma 4 for M = R in order to show that

R - R is integral.
Corollary 6. Let : R - R be a homomorphism of rings, and assume there

are elements y1 , . . . , yr R integral over R such that R = Ry1 , . . . , yr . Then
: R - R is nite and, hence, integral.
Proof. Consider the chain of successive ring extensions

(R) (R)y1 . . . (R)y1 , . . . , yr = R ,
each of which is nite by Lemma 4. It follows then by induction that R is
nite over R. To carry out the induction step, consider a set of generators for
R as a module over (R)y1 , . . . , yr1 and multiply it with a similar system
generating (R)y1 , . . . , yr1 over R, thereby obtaining a set of generators for
R over R.
Corollary 7. Let R - R and R - R be two homomorphisms of rings

which are nite (resp. integral ). Then their composition R - R is nite
(resp. integral ) as well.
Proof. The case of nite homomorphisms is dealt with in the same way as in
the proof of Corollary 6. If R - R and R - R are integral, consider an
element z R . Then z satises an integral equation over R :
z n + b1 z n1 + . . . + bn = 0, b1 , . . . , bn R .
From this we see that z R is integral over Rb1 , . . . , bn , and it follows
from Lemma 4 that the extension Rb1 , . . . , bn - Rb1 , . . . , bn , z is -
nite. Since R - Rb1 , . . . , bn is nite by Corollary 6, the composition
R - R b1 , . . . , bn , z will be nite. But then this homomorphism is inte-
gral by Corollary 5 and we see that z is integral over R. Letting z vary over R ,
we conclude that R - R is integral.
Corollary 8. Let R be a subring of a ring R and let R be the set of all elements
in R that are integral over R. Then R is a subring of R containing R. It is
called the integral closure of R in R . Furthermore, R is integrally closed in R ,
i.e. the integral closure of R in R coincides with R.
Proof. Let x, y R be integral over R. Then Rx, y is integral over R by

Corollary 6. In particular, from x, y R we get x y, x y R and we see that
R is a subring of R containing R. Furthermore, R is integrally closed in R by
Corollary 7.
As an example, we may consider the integral closure Z of Z in C, which is

called the ring of integral algebraic numbers. The latter is countable and, thus,
much smaller than C itself. Also note that Z is dierent from the algebraic
closure of Q in C, since its restriction Z Q will coincide with Z, due to the fact
that Z is integrally closed in Q; the latter will follow from Remark 10 below.
Denition 9. An integral domain R is called normal if it is integrally closed

in its eld of fractions Q(R).
Remark 10. Any principal ideal domain or, more generally, any factorial ring
R is normal.
Proof. Consider r, s R with s = 0 such that rs Q(R) is integral over R. Then

we may assume that rs is reduced in the sense that gcd(r, s) = 1. Now let
n n1
r r
+ a1 + . . . + an = 0
s s
r
be an integral equation for s
over R. Then
rn + sa1 rn1 + . . . + sn an = 0,
and we see that s divides rn , contradicting the fact that r and s were assumed
to be relatively prime.
In particular, it follows that Z is a normal ring. The same is true for polyno-
mial rings in a nite set of variables over Z or over a eld K, because all these
rings are factorial. Also note that due to the Theorem of AuslanderBuchsbaum
[24], Cor. 4 of Thm. IV.9, any regular Noetherian local ring (2.4/18) is factorial
and, hence, normal.
Proposition 11. For an integral domain R, the following conditions are equiv-
alent:
(i) R is normal.
(ii) Rp is normal for all prime ideals p R.
(iii) Rm is normal for all maximal ideals m R.
Proof. Concerning the implication (i) = (ii), we show more generally that, for
a multiplicative system S R {0}, the localization RS will be normal if R is
normal. To do this, note that R RS Q(R), since R is supposed to be an
integral domain, and that therefore the eld of fractions Q(RS ) coincides with
Q(R). Now let x Q(R) be integral over RS , say
a1 n1 an
xn + x + ... + = 0,
s1 sn
for elements a1 , . . . , an R and s1 , . . . , sn S. Writing s = s1 . . . sn , we see
that sx is integral over R and, hence, contained in R if R is normal. But then
x = s1 sx RS and it follows that RS is normal.
The implication (ii) = (iii) being trivial, it remains to justify (iii) = (i).
Let x Q(R) be integral over R and assume that all localizations Rm at maximal
ideals
m Spm R are normal. Then x mSpm R Rm and it is enough to show
mSpm R Rm = R.
To achieve this, x an element x mSpm R Rm and, for each m Spm R,
choose elements am R and bm R m such that x = abmm . As the set of all bm ,
for m Spm R, cannot be contained in a maximal ideal of R, it must generate
the unit ideal in R. Thus, there is an equation

cm bm = 1
mSpm R
with coecients cm R vanishing for almost all m. But then

am
x= cm b m x = cm bm = cm am R
mSpm R mSpm R
bm mSpm R
and we see that R is normal.
Exercises
1. Consider ring morphisms i : R - Ri , i = 1, . . . , n, starting out from a given
ring R. Show that the ring morphism

n

R - Ri , a - 1 (a), . . . , n (a) ,
i=1
is nite (resp. integral) if the same is true for all i .

2. Let R be a ring and a nite group of automorphisms of R. Show that R is an
integral extension of the xed ring R = {a R ; (a) = a for all }.
3. Let R be a normal integral domain with eld of fractions K and let L/K be an
algebraic extension of elds. Show that an element x L is integral over R if and
only if the minimal polynomial of x over K has coecients in R.
4. Let R be a normal integral domain. Show that the polynomial ring in one variable
RX is normal as well.

5. Show that the integral closure of Z in Q( 5) is Z 12 (1 + 5); it is known that
this ring is factorial.

6. Show that the integral closure of Z in Q( 5) is Z 5. Show that Z 5
is a non-factorial
normal
ring. Hint: Consider the decompositions 21 = 3 7 =
(1 + 2 5) (1 2 5).
7. Let R be a valuation ring, i.e. an integral domain with eld of fractions K such
that for any x K we have x R or, if the latter is not the case, x1 R. Show
that R is normal.
8. Let R be a normal Noetherian integral domain with eld of fractions K and let
L/K be a nite separable extension of elds. Show that the integral closure R of
R in L is a nite R-module. Hint: The trace function TrL/K of L/K gives rise to
a non-degenerate bilinear form L L - L, (x, y) - TrL/K (xy). Show that

there is a K-basis y1 , . . . , yd of L such that R di=1 Ryi .
3.2 Noether Normalization and Hilberts Nullstellensatz 91
3.2 Noether Normalization and Hilberts Nullstellensatz
In this section we want to illustrate the concept of integral dependence by

discussing polynomial rings KX = KX1 , . . . , Xn over a eld K, for a nite
set of variables X = (X1 , . . . , Xn ). As we know already from 1.5/14, KX is
Noetherian. Furthermore, it follows from the Lemma of Gau (see [3], 2.7/1)
that KX is factorial and, hence, normal by 3.1/10.
Let A be a K-algebra and call a set of elements x1 , . . . , xn A algebraically
independent over K if the K-homomorphism KX1 , . . . , Xn - A substitut-
ing xi for the variable Xi is injective, and algebraically dependent otherwise.
Furthermore, as before, let us write Kx1 , . . . , xn A for the image of such a
substitution homomorphism. Recall that A is called a K-algebra of nite type
if there is a surjective K-homomorphism KX1 , . . . , Xn - A or, in other
words, if there exist elements x1 , . . . , xn A such that A = Kx1 , . . . , xn .
Theorem 1 (Noethers Normalization Lemma). Let A be a K-algebra of nite

type. If A = 0, there exists a nite injective K-homomorphism
KY1 , . . . , Yd - A
for a certain set of variables Y1 , . . . , Yd .
The proof of the Normalization Lemma needs a technical recursion step,

which is of interest by itself and which we will prove rst.
Lemma 2. Let Kx1 , . . . , xn be a K-algebra of nite type and consider an

element y Kx1 , . . . , xn given by some expression

() y= a1 ...n x11 . . . xnn
(1 ...n )I
with coecients a1 ...n K where the summation extends over a nite non-
empty index set I Nn ; in particular, n 1. Then there exist elements
y1 , . . . , yn1 Kx1 , . . . , xn such that the canonical monomorphism
Ky1 , . . . , yn1 , y - Kx1 , . . . , xn
is nite.
Proof. Let us set
y1 = x1 xsn1 , ..., yn1 = xn1 xsnn1 ,
where the choice of exponents s1 , . . . , sn1 N{0} still has to be made precise.
Then we have
Kx1 , . . . , xn = Ky1 , . . . , yn1 , xn .
Substituting xi = yi + xsni for i = 1, . . . , n 1 in the relation () and splitting

the powers xi i = (yi + xsni )i into a sum of xsni i and terms of lower degrees in
xn , we get a new relation of type

() y= a1 ...n xsn1 1 +...+sn1 n1 +n + f (y1 , . . . , yn1 , xn ),
(1 ...n )I
where f (y1 , . . . , yn1 , xn ) is a polynomial expression in xn with coecients in

Ky1 , . . . , yn1 whose degree in xn is strictly smaller than the maximum of
all numbers s1 1 + . . . + sn1 n1 + n for (1 , . . . , n ) varying over I. As is
easily seen, the numbers s1 , . . . , sn1 N can be chosen in such a way that the
exponents s1 1 + . . . + sn1 n1 + n occurring in () are dierent for all tuples
(1 , . . . , n ) I. Just take t N larger than the maximum of all 1 , . . . , n for
(1 , . . . , n ) I and set
s1 = tn1 , . . . , sn1 = t1 .
Viewing the right-hand side of () as a polynomial in xn with coecients in

Ky1 , . . . , yn1 , the choice of exponents si leads to precisely one term of type
n with a coecient a K
axN y1 , . . . , yn1 and, in fact, a K , whose degree
N dominates the degrees of all the remaining ones. Therefore multiplication of
() with a1 and subtraction of a1 y yields an integral equation for xn over
Ky1 , . . . , yn1 , y. Thus, we see from 3.1/6 that the map
Ky1 , . . . , yn1 , y - Ky1 , . . . , yn1 , xn = Kx1 , . . . , xn
is nite.
Now the proof of Noethers Normalization Lemma is easy to achieve. Since

A = 0, we may view K as a subring of A. Let A = Kx1 , . . . , xn for suitable
elements xi A. If x1 , . . . , xn are algebraically independent over K, nothing
has to be shown. Hence, we can assume the xi to be algebraically dependent
over K. Then, taking y = 0, there is a non-trivial relation () as specied in
Lemma 2, and there are elements y1 , . . . , yn1 A such that the canonical map
Ky1 , . . . , yn1 = Ky1 , . . . , yn1 , y - Kx1 , . . . , xn = A
is nite. If y1 , . . . , yn1 are algebraically independent over K, we are done. Else

we apply the process just described again, this time for the ring Ky1 , . . . , yn1 .
Proceeding recursively, we arrive after nitely many steps at a system y1 , . . . , yd
that is algebraically independent over K and has the property that A is nite
over Ky1 , . . . , yd .
As we will see later from 3.3/6 in conjunction with 2.4/16, the integer d
in Noethers Normalization Lemma is uniquely determined by the K-algebra
A. In fact, d coincides with the Krull dimension of A, as introduced in 2.4/2.
Alternatively, if A is an integral domain, one can show combining 3.1/3 (ii) with
3.1/2 that the eld of fractions Q(A) is nite over the transcendental function
eld K(Y1 , . . . , Yd ) and it follows that d equals the transcendence degree of Q(A)
over K.
Corollary 3. Let K L be an extension of elds where L = Kx1 , . . . , xn for

some elements x1 , . . . , xn L. In other words, we assume that L is a K-algebra
of nite type. Then the extension K L is nite.
Proof. By Noethers Normalization Lemma, we can nd nitely many elements

y1 , . . . , yd L, algebraically independent over K, such that the extension of
rings Ky1 , . . . , yd - L is nite. As L is a eld, the same must be true for
Ky1 , . . . , yd by 3.1/2. However, a polynomial ring in d variables over a eld K
cannot be a eld, unless d = 0. Therefore Ky1 , . . . , yd = K and the extension
K - L is nite.
A situation as in Corollary 3 occurs naturally, if we divide a polynomial

ring KX1 , . . . , Xn by a maximal ideal.
Corollary 4. Let A be a K-algebra of nite type and m A a maximal ideal.

Then the canonical map K - A/m is nite and, hence, L = A/m is a nite
extension eld over K.
Proof. We know that A/m, just as A, is a K-algebra of nite type. As A/m is

a eld, we can conclude from Corollary 3 that it is nite over K.
Another remarkable consequence of Noethers Normalization Lemma is the

fact that any polynomial ring KX1 , . . . , Xn over a eld K is Jacobson, i.e. we
have rad(a) = j(a) for any ideal a KX1 , . . . , Xn . Using 1.3/6, this claim is
derived from the following assertion:
Corollary 5. Let A be a K-algebra of nite type. Then its nilradical coincides

with its Jacobson radical, i.e. rad(A) = j(A).
Proof. Clearly we have

rad(A) = p m = j(A).
pSpec A mSpm A
To verify the opposite inclusion, consider an element f A rad(A). Let us

show f j(A). As f is not nilpotent, it generates a multiplicative system
S A that does not contain 0. Thus, the localization AS = Af 1 is non-zero
by 1.2/4 (iii) and, hence, there exists a maximal ideal m Af 1 . Let n A
be its inverse image with respect to the canonical map : A - Af 1 . Then
induces a monomorphism
A/n - Af 1 /m

of K-algebras. As A is of nite type over K, say A = Kx1 , . . . , xn , so is its

localization Af 1 = Kx1 , . . . , xn , f 1 . Therefore we see from Corollary 4
that the eld Af 1 /m is nite over K and, in particular, over A/n. However,
then A/n must be a eld by 3.1/2 and we see that n is a maximal ideal in A.
Since (f ) is a unit in Af 1 , it cannot be contained in m. In particular, f n
and, hence, f j(A).
Finally, let us discuss the geometric signicance of the preceding corollary

in terms of Hilberts Nullstellensatz. For a ring R and an ideal a R we have
dened its zero set
V (a) = {x Spec R ; f (x) = 0 for all f a} = {p Spec R ; a p},
and for any subset E Spec R its corresponding vanishing ideal

I(E) = {f R ; f (x) = 0 for all x E} = px ;
xE
see Sections. 1.1 and 1.3. Interpreting the nilradical rad(a) as the intersection
of all prime ideals p R containing a, as shown in 1.3/6 (iii), we have obtained
the equation

I V (a) = p = rad(a)
pSpec R
ap
for ideals a R.
Considering an algebra A of nite type over a eld K in place of R and
an ideal a A, we know from the above corollary that the nilradical rad(a)
coincides with its Jacobson radical j(a). Therefore, in the above equation, we
may replace V (a) by its restriction Vmax (a) = V (a) Spm A, thereby obtaining
the following assertion:
Corollary 6 (Hilberts Nullstellensatz). Let A be a K-algebra of nite type and

a A an ideal. Then:

I Vmax (a) = m = rad(a),
mSpm R
am
We want to make the concept of viewing the elements of a K-algebra A

of nite type as functions on the spectrum of its maximal ideals Spm A a bit
more explicit, at least for polynomial rings A = KX1 , . . . , Xn ; see also the
discussion of the ane n-space over a eld K in Section 6.7. Choose an algebraic
closure K of K and, for each point x K n , consider the evaluation map
x : KX1 , . . . , Xn - K, f - f (x).
Since the image of x is integral over K, it is a eld by 3.1/2. Hence, ker x is

a maximal ideal in KX1 , . . . , Xn and there is a well-dened map
: Kn - Spm KX1 , . . . , Xn , x - ker x ,
which we claim is surjective. Indeed, given an ideal m KX1 , . . . , Xn ,

which is maximal, its residue eld KX1 , . . . , Xn /m is nite over K by
Corollary 4, and we may extend the inclusion K - K to an embedding
KX1 , . . . , Xn /m - K. Writing xi K for the image of the residue class of
Xi , we see that (x1 , . . . , xn ) = m. Of course, the point x = (x1 , . . . , xn ) K n
we have constructed from m, is unique only up to a K-automorphism of the eld
K(x1 , . . . , xn ). Therefore, we may interpret Spm KX1 , . . . , Xn as the quotient
of K n by the group of K-automorphisms AutK (K), i.e.
Spm KX1 , . . . , Xn K n / AutK (K).
In particular, for an algebraically closed eld K, the map considered above

will be bijective and a straightforward argument shows that it is given by
: Kn - Spm KX1 , . . . , Xn , (x1 , . . . , xn ) - (X1 x1 , . . . , Xn xn ).
Going back to the situation of Hilberts Nullstellensatz for an arbitrary eld

K, we may alternatively associate to an ideal a KX1 , . . . , Xn the set

VK (a) = {x K n ; f (x) = 0 for all f a} = 1 Vmax (a) ,
and to a subset E K n the ideal

I(E) = {f KX1 , . . . , Xn ; f (x) = 0 for all x E} = I (E) .
Then, via , Hilberts Nullstellensatz transforms into its classical version

I VK (a) = rad(a).
Exercises
In the following let K be a eld.
1. Let : A - B be a morphism of K-algebras where B is of nite type. Show
1
that (m) is a maximal ideal of A for every maximal ideal m B. In other
words, gives rise to a map Spm B - Spm A.
2. Give an explicit solution of Noethers Normalization Lemma for the K-algebra
KX1 , X2 /(X1 X2 ).
3. Let m be a maximal ideal of the polynomial ring KX1 , . . . , Xn in n variables
over K. Show that there exist polynomials fi KX1 , . . . , Xi1 Xi satisfying
m = (f1 , . . . , fn ) where fi is monic in Xi for i = 1, . . . , n.
4. Consider the polynomial ring KX1 , . . . , Xn in n variables over K for n 1.
Let f KX1 , . . . , Xn be a non-constant polynomial. Show that there ex-
ists a monomorphism of type KY1 , . . . , Yn1 - KX1 , . . . , Xn /(f ) dening
KX1 , . . . , Xn /(f ) as a nite free module over KY1 , . . . , Yn1 .
5. Generalization of Noethers Normalization Lemma: Let : R - A be a

monomorphism of rings equipping A with the structure of an R-algebra of -
nite type. Assume that R is an integral domain and that ra = 0 for r R, a A
implies r = 0 or a = 0. Show there exists a nite set of polynomial variables
X1 , . . . , Xd such that extends to a monomorphism : RX1 , . . . , Xd - A
and that the latter becomes nite when localized by the multiplicative system
generated by a suitable element s R {0}.
6. Another generalization of Noethers Normalization Lemma: Let A be an algebra

of nite type over a eld K and consider an ascending chain a1 . . . ar A of
proper ideals in A. Show that there exist elements x1 , . . . , xd A, algebraically
independent over K, such that A is integral over Kx1 , . . . , xd and such that
ai Kx1 , . . . , xd , as an ideal in Kx1 , . . . , xd , is generated by x1 , . . . , xdi for
i = 1, . . . , r and suitable indices d1 . . . dr d. Hint: Reduce to the case
where A is a free polynomial ring over K and use Lemma 2. Consult Serre [24],
Thm. III.2, if necessary.
3.3 The CohenSeidenberg Theorems
In the present section we x an integral ring homomorphism : R - R and

discuss the relationship between prime ideals in R and R . Given a (prime) ideal
P R , we write P R for the restricted (prime) ideal 1 (P) R.
Proposition 1. Let : R - R be an integral ring homomorphism.

(i) A prime ideal P R is maximal if and only if its restriction p = P R
is maximal in R.
(ii) Let P1 , P2 R be prime ideals satisfying P1 R = P2 R. Then
P1 P2 implies P1 = P2 .
Proof. In the situation of (i), the map induces an integral monomorphism of

integral domains R/p - R /P, and we see from 3.1/2 that R/p is a eld if
and only if R /P is a eld. Hence, p is maximal in R if and only if P is maximal
in R .
Now let P1 , P2 R be two prime ideals as required in (ii), namely such
that the induced ideals p = P1 R and P2 R coincide in R. Let S = R p
and consider the ring homomorphism RS - R
(S) induced from , which
is integral by 3.1/3 (ii) since is integral. Then, by 1.2/7, the ideal S 1 p
generated by p in RS is maximal. Likewise we can consider the ideals S 1 P1
and S 1 P2 generated by P1 and P2 in R(S)
. These are prime by 1.2/5, since
(S) is disjoint from P1 and P2 . Then q1 = S 1 P1 RS and q2 = S 1 P2 RS
are prime ideals in RS that contain S 1 p. Because S 1 p is maximal in RS , we
get S 1 p = q1 = q2 . Thus, S 1 P1 and S 1 P2 are two ideals in R(S)

, whose
restrictions to RS are maximal. But then, by (i), we know that S 1 P1 and
S 1 P2 are maximal in R(S)
and it follows S 1 P1 = S 1 P2 from P1 P2 .
3.3 The CohenSeidenberg Theorems 97
Since P1 = S 1 P1 R and P2 = S 1 P2 R by 1.2/5, we conclude P1 = P2 ,

as desired.
Theorem 2 (Lying-over). Let : R - R be an integral ring homomorphism

and p R a prime ideal such that ker p. Then there exists a prime ideal
P R such that P R = p. Furthermore, according to Proposition 1, there is
no proper inclusion between such prime ideals P R .
Proof. As in the proof of Proposition 1, we let S = R p and consider the

homomorphism RS - R(S)
induced from ; it is integral by 3.1/3 (ii). Since

ker p, we have S ker = and therefore 0 (S) so that R(S) is

non-zero by 1.2/4. Therefore we can nd a maximal ideal m R(S) and it
follows from Proposition 1 (i) that m RS is a maximal ideal in RS . But RS is
a local ring with maximal ideal S 1 p so that m RS = S 1 p. Now consider the
commutative diagram
R - R
? ?
-
RS R(S)
and the restriction P = m R of m to R . Then
P R = (m R ) R = (m RS ) R = S 1 p R = p
by 1.2/5, as desired.
Theorem 3 (Going-up). Let : R - R be an integral ring monomorphism

and
p0 p1 . . . p n R
a chain of prime ideals in R. Then, for any prime ideal P0 R such that
P0 R = p0 , there exists a chain of prime ideals
P0 P1 . . . Pn R
satisfying Pi R = pi for all i.
Proof. We conclude by induction on the length n of such chains of prime ide-

als, the case n = 0 being trivial. Thus, let n > 0. By the induction hypoth-
esis, we can nd a chain of prime ideals P0 . . . Pn1 R satisfying
Pi R = pi for i = 1, . . . , n 1. Then induces an integral monomorphism
R/pn1 - R /Pn1 and, by Theorem 2, there is a prime ideal q R /Pn1
lying over the prime ideal pn /pn1 R/pn1 . Let Pn be the preimage of q with
respect to projection R - R /Pn1 . Then Pn1 Pn , and we conclude
from the commutative digram
R - R
? ?
R/pn1 - R /Pn1
that Pn R = pn , as desired.
In Theorem 3 we have constructed extensions of chains of prime ideals with

respect to integral ring extensions where an extension of the smallest prime
ideal was given. We now will look at the analogous problem where an extension
of the largest prime ideal is given.
Theorem 4 (Going-down). Let : R - R be an integral ring monomorphism

where R and R are integral domains and, in addition, R is normal. Then, given
a chain of prime ideals
R p 0 p1 . . . p n
in R and P0 R a prime ideal satisfying P0 R = p0 , there exists a chain of
prime ideals
R P0 P1 . . . Pn
in R such that Pi R = pi for all i.
To do the proof, we need some preparations. In particular, we will use Galois

theory on the level of elds. Let K K be an algebraic extension of elds
which is quasi Galois or normal in the sense that K is a splitting eld of a set
of polynomials in KX. Then we can consider the group G = AutK (K ) of all
K-automorphisms K - K ; in other words, the group of all automorphisms
of K that leave K xed. We call
K G = {a K ; g(a) = a for all g G}
the xed eld of K with respect to G. It contains K and, thus, is a eld between
K and K . One knows from Galois theory (see [3], 4.1/5) that the extension
K G K is Galois with Galois group G and that K K G is purely inseparable.
In particular, for each element y K G , there is an integer n N {0} such
that y n K.
Let us start now by giving a lemma that will settle the main part of the
proof of Theorem 4.
Lemma 5. Let p R be a prime ideal of a normal integral domain R. Fur-

thermore, let K be a quasi Galois extension of the eld of fractions K = Q(R)
and write R for the integral closure of R in K . Then:
(i) Each g AutK (K ) induces a ring automorphism R - R xing R.
(ii) If P is a prime ideal in R satisfying P R = p, the same if true for
g(P), for any g AutK (K ).
(iii) AutK (K ) acts transitively on the set of all prime ideals in R that lie
over p, i.e. given two prime ideals P1 , P2 R such that P1 R = p = P2 R,
there is an element g AutK (K ) such that g(P1 ) = P2 .
Proof. Write G = AutK (K ) and look at an element x R , thus, an element

x K satisfying an integral equation over R. Then each g G transports
this equation into an integral equation for g(x) over R. This way we see that
g restricts to a ring automorphism R - R xing R. In particular, g maps
prime ideals P R to ideals of the same type. Furthermore, from P R = p,
we can conclude
g(P) R = g(P) g(R) = g(P R) = g(p) = p.
This settles assertions (i) and (ii).

To derive (iii), let us rst assume that G is nite, which is automatically
the case if K is nite over K. Let P1 , P2 R be prime ideals satisfying
P1 R = p = P2 R. We have to nd some g G such that g(P1 ) = P2 . For
the latter equation it is enough to show P2 g(P1 ), if we apply Proposition 1
in conjunction with assertion (ii) which has just been proved. Furthermore, the
existence of such a g will follow from Lemma 1.3/7 and the niteness of G, once
we have shown that
P2 g(P1 ).
gG

To justify the latter inclusion, let x P2 . Then y = gG g(x) is xed by
G. Thus, as explained above in terms of Galois theory, there exists an integer
n N{0} such that y n K. Now y n , being a member of R , is integral over R.
Therefore we must have y n R, since R is normal. As G contains the identity
automorphism, we know y n P2 R = p and, hence, y n p P1 . Using the
fact that P1 is a prime ideal, the denition of y shows that there must exist
some g G satisfying g(x) P1 , and we get x g 1 (P1 ), which justies the
above inclusion.
If G is not necessarily nite, we consider intermediate elds K E K
and dene RE = R E as the integral closure of R in E. Let M be the set of all
pairs (RE , gE ) where E is an intermediate eld of the extension K K , quasi
Galois over K, and where gE AutK (E) satises gE (P1 RE ) = P2 RE . We
write (RE , gE ) (RE , gE ) for elements in M , when E E and gE |E = gE .
Then M is a partially ordered set which allows the application of Zorns Lemma
and, thus, admits a maximal element (RE , gE ).
If E is strictly contained in K , we can choose a non-trivial nite extension
E K of E that is quasi Galois over K (and, thus, over E) and extend gE to

a K-automorphism g of E . Then
g (P1 RE ) RE = g (P1 RE ) = P2 RE = (P2 RE ) RE
and, by the above special case, there exists some element g of the nite group
AutE (E ) such that
g g (P1 RE ) = P2 RE .
Now g g AutK (E ) is an extension of g. Thus,
(RE , g g ) M, (RE , gE ) < (RE , g g ),
which contradicts the maximality of (RE , gE ). Therefore we have E = K , hence

RE = R and gE AutK (K ) satises gE (P1 ) = P2 .
The proof of Theorem 4 is now quite easy. Let K and K be the elds of
fractions of R and R . Then the extension K K is algebraic, and we can nd
a eld K extending K such that the extension K K is quasi Galois. Let
R be the integral closure of R in K . Applying Theorem 3, choose a chain of
prime ideals
R P0 P1 . . . Pn
lying over the chain R p0 . . . pn and, according to Theorem 2, a prime
ideal P0 R over P0 R (and, thus, over p R). By Lemma 5, there exists
some g AutK (K ) such that g(P0 ) = P0 . Then
P0 = P0 R = g(P0 ) R g(P1 ) R . . . g(Pn ) R
is a descending chain of prime ideals in R that lies over R p0 . . . pn and

starts with P0 , as desired.
To end this section, we give some applications of the CohenSeidenberg

Theorems to the theory of Krull dimensions, as developed in Section 2.4.
Proposition 6. Let R - R be an integral ring monomorphism, a R an

ideal, and a = a R its restriction to R. Then:
(i) dim R = dim R.
(ii) ht a ht a and, in fact, ht a = ht a if R and R are integral domains
and R is normal.
(iii) coht a = coht a.
Proof. We start with assertion (i), observing that (iii) is a consequence of (i).
According to the Going-up Theorem 3, each strictly ascending chain of prime
ideals in R may be extended to a (strictly) ascending chain of prime ideals in
R . Conversely, each strictly ascending chain of prime ideals in R restricts to an
ascending chain of prime ideals in R where the latter must be strictly ascending
by Proposition 1. This proves (i).
To settle (ii), consider a strictly ascending chain of prime ideals in R . Re-
stricting it to R yields an ascending chain of prime ideals in R, which is strictly
ascending by Proposition 1. This shows ht a ht a if a is a prime ideal in R . If
a is not necessarily prime, choose a minimal prime divisor p of a. Applying the
Lying-over Theorem 2 to the induced monomorphism R/a - R /a , there is
a minimal prime divisor P of a lying over p. Since ht P ht p, as we just have
seen, we get ht a ht a.
Now assume that R and R are integral domains and that R is normal. Then
the Going-down Theorem 4 becomes applicable and we see that ht a = ht a if
a is a prime ideal in R . Furthermore, for general a we see that the minimal
prime divisors of a restrict to minimal prime divisors of a. Combining this with
the above argument, we get ht a = ht a, as desired.
Fixing a eld K, a bit more can be said about K-algebras of nite type.
Proposition 7. Let A be a K-algebra of nite type that is an integral domain.

Then dim A = transgradK (Q(A)), i.e. the Krull dimension of A equals the
transcendence degree of the eld of fractions Q(A) over K.
Proof. By Noethers Normalization Lemma 3.2/1, there is a nite monomor-

phism KX1 , . . . , Xd - A. Then
dim A = dim KX1 , . . . , Xd = d
by 2.4/16 and Proposition 6 (i). Furthermore,

d = transgradK Q(KX1 , . . . , Xd ) = transgradK Q(A) ,
since the extension of elds Q(A)/Q(KX1 , . . . , Xd ) is algebraic.
Proposition 8. Let A be a K-algebra of nite type that is an integral domain.

Then:
(i) ht p + coht p = dim A for every prime ideal p A.
(ii) ht m = dim A for every maximal ideal m A.
Proof. First observe that (ii) is a special case of (i). To show (i) we use induction
on dim A. Applying Noethers Normalization Lemma 3.2/1, there is a nite
monomorphism of type KX1 , . . . , Xd - A where d = dim A by 2.4/16 and
Proposition 6 (i). The case of dimension d = 0 is trivial, since then A is a eld
by 3.1/2. Also the case p = 0 is trivial, since then ht p = 0 and coht p = dim A.
Therefore we can assume d 1 as well as p = 0.
Now choose a non-zero element y p KX1 , . . . , Xd , which exists since
the restriction of the non-zero prime ideal p to KX1 , . . . , Xd must be non-
zero by Proposition 1. Then we may apply 3.2/2 and thereby obtain elements
y1 , . . . , yd1 KX1 , . . . , Xd such that the canonical monomorphism
: Ky1 , . . . , yd1 , y - KX1 , . . . , Xd
is nite. Comparing transcendence degrees of elds of fractions, we see that

the elements y1 , . . . , yd1 , y are algebraically independent over K and, hence,
that Ky1 , . . . , yd1 , y may be viewed as a free polynomial ring in the variables
y1 , . . . , yd1 , y. Composing with KX1 , . . . , Xd - A, we arrive at a nite
monomorphism of type KY1 , . . . , Yd - A where the Yi are variables and Yd
is mapped into p. Now observe that the latter morphism satises the require-
ments of the Going-down Theorem 4, since KY1 , . . . , Yd , as a factorial ring, is
normal by 3.1/10. Therefore we can nd a prime ideal p0 A that is contained
in p and restricts to the prime ideal (Yd ) KY1 , . . . , Yd . Then, looking at the
nite monomorphism
KY1 , . . . , Yd1 - A/p0
and the ideal p/p0 A/p0 , we can conclude from the induction hypothesis that
ht(p/p0 )+coht(p/p0 ) = d1. Since ht p ht(p/p0 )+1 and coht p = coht(p/p0 ),
we get ht p + coht p d and then, necessarily, ht p + coht p = d.
Exercises
1. Show for rings R, R that dim(R R ) = max(dim R, dim R ). Hint: Use Exer-
cise 1.1/6.
2. Let R - R be an integral morphism of rings. Show dim R dim R .
3. Let R - R be an integral monomorphism of rings. Show that every morphism

R - L into an algebraically closed eld L admits an extension R - L.
4. Let R - R be an integral morphism of rings. Show that R is Jacobson if R
is Jacobson. Hint: Use the fact that a ring is Jacobson if each of its prime ideals
is an intersection of maximal ideals.
5. Let A be an algebra of nite type over a eld K. Assume that A is an integral
domain. Show that every maximal chain of prime ideals in A is of length dim A.
6. Let R be a ring and a nite group of automorphisms of R. Then R is integral
over the xed ring R by Exercise 3.1/2. Show for any prime ideal p R that
there are only nitely many prime ideals P R lying over p and that acts
transitively on these.
7. For an integral morphism of rings : R - R consider the associated map on
prime spectra a : Spec R - Spec R, p - p R. Show:
(a) a (V (a )) = V (a R) for any ideal a R . In particular, a is a closed map
with respect to Zariski topologies on Spec R and Spec R .
(b) The bers of a are nite if is nite.
4. Extension of Coecients and Descent
The main theme of the present chapter is to discuss the process of coecient
extension for modules and its reverse, called descent. For example, imagine a
ring R and an R-module M whose structure seems to be dicult to access.
Then one can try to replace the coecient domain R by a bigger ring R , over
which the situation might become easier to handle. In other words, we would
select a certain extension homomorphism R - R and use it in order to

derive from M a best possible R -module M extending the R-module structure
we are given on M . In particular, M will respect all relations that are already
present in M . The technical frame for such a construction is given by the so-
called tensor product. Passing from M to the tensor product M = M R R
we say that M is obtained from M via coecient extension with respect to
R - R . Of course, the extension homomorphism R - R must be chosen
in an intelligent way so that the results obtained for M can be descended to
meaningful information on M .
Let us discuss an example from Linear Algebra. We consider a quadratic
matrix with coecients from R, say A Rnn where n > 0, and look at the
R-linear map
Rn - Rn , x - A x.
Recall that an element R is called an eigenvalue of A if there exists an
associated eigenvector, i.e. a vector z Rn {0} such that Az = z. Note that
the eigenvalues of A are precisely the zeros of the characteristic polynomial
A (X) = det(X id A). Since the eld R is not algebraically closed, it is
possible that the set of eigenvalues of A is empty.
However, if we assume A to be symmetric, then the characteristic polynomial
A (X) decomposes completely into linear factors over R and, hence, the set of
eigenvalues of A cannot be empty. We want to explain how this result can
be derived by means of coecient extension from R to C. Viewing Rn as an
R-vector space, a canonical candidate for its coecient extension via R - C
is of course the C-vector space Cn . So we look at the C-linear map
Cn - Cn , x - A x.
Furthermore, consider the canonical Hermitian form on Cn given by x, y = xt y

for column vectors x, y Cn where xt means the transpose of x and y the
complex conjugate of y. Then, since A is a symmetric matrix with real entries,
we get

104 4. Extension of Coecients and Descent
A x, y = (A x)t y = xt A y = x, A y
for x, y Cn . Now use the fact that the eld C is algebraically closed. There-
fore the characteristic polynomial A (X) admits a zero C and there is a
corresponding eigenvector z Cn {0}. Since z, z = 0, the equation
z, z = z, z = A z, z = z, A z = z, z = z, z
shows = . Hence, all zeros of the characteristic polynomial A (X) must
be real and we are done. In our argument we can rely on the fact that the
characteristic polynomial A (X) is the same for A as a matrix in Cnn or in
Rnn . This makes the descent from C to R particularly easy.
Having seen that the technique of extending coecients can be quite useful,
let us discuss now how to construct such extensions in more generality. Let M
be a module over a ring R and consider a ring homomorphism R - R . In

particular, the latter equips R with the structure of an R-module. To extend
the coecients of M from R to R we would like to construct an object M
where products of type x r for x M and r R make sense. Of course, this
product should be R-linear in x and R-linear in r , which requires that M must
be imagined to be at least an R-module. Moreover, if M is not exceedingly big,
namely generated over R by all products of type x r , which is enough for our
purposes, we can expect that it is automatically an R -module. Therefore we
look at R-bilinear maps M R - T into R-modules T and try to nd a
universal one among these, i.e. an R-bilinear map : M R - T such that
for every R-bilinear map : M R - E into another R-module E there is a
unique R-linear map : T - E satisfying = . Then, according to 4.1/1,
T is called a tensor product of M and R over R and is denoted by M R R ,
where the latter notation is justied, since such a tensor product is unique up
to canonical isomorphism; see 4.1/2. Furthermore, it follows from 4.1/3 that
the tensor product M R R always exists. Usually the inherent R-bilinear map
: M R - M R R is not mentioned explicitly, since for x M and
r R one writes x r in place of (x, r ), interpreting this as the product

x r in the sense discussed above.

The construction of tensor products works more generally for R-modules
M and N , resulting in an R-module M R N where M is as before and N
replaces the ring homomorphism R - R providing R with the structure of
an R-module. The rst two sections of this chapter are devoted to the study of
such tensor products. We begin with some basic properties and then investigate
how tensor products of modules behave with respect to exact sequences. Fixing
an R-module N , we can associate to each R-module M the R-module M R N
and to each morphism of R-modules f : M - M the morphism of R-modules
f R N : M R N - M R N given by the R-bilinear map
M N - M R N, (x, y) - f (x) y.
As we will say in Section 4.5, the process of taking tensor products with N over
R is a covariant functor on the category of R-modules. Now, what happens
when we tensor a short exact sequence
Outline and Motivation 105
0 - M - M - M - 0
of R-modules with N over R? The answer we will give in 4.2/1 is that the right
part of the resulting sequence
0 - M R N - M R N - M R N - 0,
starting at M R N , is exact. This is the so-called right exactness of tensor

products. It gives us the opportunity to call an R-module N at if taking
the tensor product with N over R always preserves short exact sequences of
R-modules. The latter is equivalent to the fact that for any monomorphism of
R-modules M - M the resulting R-morphism M R N - M R N is
a monomorphism as well. Let us add that the atness of modules is a quite
complicated feature. In part it is characterized by the lack of zero divisors.
Indeed, a module N over a principal ideal domain R is at if and only if each
equation ax = 0 for a R and x N implies a = 0 or x = 0; see 4.2/7. On
the other hand, looking at a module M of nite presentation (or even of nite
type) over a local ring R, we can read from 4.4/3 (resp. Exercise 4.4/4) that M
is at if and only if it is free. Moreover, we show in 4.3/3 that every localization
morphism R - RS is at, i.e. denes the localization RS as a at module
over R.
Let us turn back to the situation where the R-module N is present in the
form of a morphism of rings R - R . Then, as we have indicated above and
will show in more detail in Section 4.3, the extension process
R R : M - M R R
produces from any R-module M an R-module M R R , which is canonically

an R -module via multiplication with R from the right. The process preserves
several interesting module properties. For example, if M is a free R-module,
M R R will be a free R -module, since tensor products are compatible with
direct sums; see 4.1/9. But also more general properties like nite type, nite
presentation, atness, or even faithful atness are preserved, as shown in 4.4/1.
The reverse problem of descending R -module properties from M R R to
R-module properties of M is much more dicult. For this to work out well we
need strong assumptions on the morphism R - R . This is illustrated by the
fact that the zero morphism R - 0, which is at for trivial reasons, turns
any R-module M into the zero module M R 0 = 0, the only possible module
over the zero ring. Of course, from the zero module M R 0 nothing can be
read about M itself. In fact, for the descent to work well, we have to assume
that a tensor product M R R is zero if and only if M is zero. If this is the
case for every R-module M and if, in addition, R - R is at, then the latter
morphism is said to be faithfully at. For example, we show in 4.4/1 that the
above-mentioned module properties, except for the property of being free, all
descend from R to R if R - R is faithfully at. A more delicate situation is
given for an R-module M such that M R R is a free R -module. At the end of
Section 4.4 we give an example of an R-module M and a coecient extension
morphism R - R such that M R R is a free R -module without M being

free. But M will be locally free in the sense of 4.4/2 and we show in 4.4/3
that the property of being locally free of nite rank descends with respect to a
faithfully at extension morphism R - R .
A truly demanding venture is to descend R -modules M and their mor-
phisms with respect to a given extension morphism R - R , even when the
latter is faithfully at. But precisely what is meant by descent in this case? Of
course, the rst naive idea is that, starting with an R -module M , we would
like to construct an R-module M , whose extension via R - R is isomorphic
to M , i.e. satises M R R M . But, what about the uniqueness of M if
M exists at all? In some cases one can easily guess an R-module M extending
to M , for example, if M is free. Then one may x free R -generators of M
and dene M as the free R-module generated by these. This will produce an
isomorphism M R R - M , as desired. However, the example at the end of
Section 4.4 shows that there exist non-free R-modules M nevertheless extend-
ing to a free R -module M R R . From this we conclude that R-modules M
that are naively descended from M cannot be expected to be unique. In other
words, if f : M - N is a morphism of R -modules and even if M , N are
known to descend to R-modules M , N , we cannot expect without any further
assumption that f descends to a morphism of R-modules f : M - N.
The key point for accessing a more elaborate version of module descent
consists in a very careful analysis of morphisms obtained via coecient exten-
sion. To explain this, we use a special notation for extended modules that is
particularly convenient for the purposes of descent. Namely, given a morphism
of rings p : R - R and an R-module M , we write p M for the extended
R -module M R R . Likewise, if f : M
- N is a morphism of R-modules,
we write p f : p M - p N for the R -module morphism obtained from f via
taking tensor products with R over R. Proceeding like this may look a bit arti-
cial, but the notation has a rather plausible geometric background, as we will
explain at the beginning of Section 4.6. Now set R = R R R and consider
-
the canonical morphisms p1 , p2 : R - R = R R R . Then the compositions
of p1 , p2 with p coincide; let q = p1 p = p2 p. Thus, using 4.3/2, there are
canonical isomorphisms
p1 (p M ) q M p2 (p M )
so that we can look at the diagram

p p1
-
HomR (M, N ) - HomR (p M, p N ) - HomR (q M, q N )
p2
of module morphisms over R, R , and R . Assuming p to be faithfully at,

the striking assertion of 4.6/1 says that this diagram is exact, i.e. that p
denes a bijection between HomR (M, N ) and the set of those morphisms in
HomR (p M, p N ) that have the same image with respect to p1 and p2 . This
fully characterizes all R -morphisms on extended R -modules that descend with
respect to p : R - R . In particular, the descent of morphisms is unique.
4.1 Tensor Products 107
Furthermore, looking at the above diagram and the isomorphisms preced-

ing it, the descent of morphisms suggests for the descent of modules that
R -modules M equipped with further data should be considered, such as suit-
able R -isomorphisms : p1 M - p2 M . If an isomorphism of this type
satises the so-called cocycle condition, which is a natural compatibility be-
tween all possible extensions of to the level of R = R R R R R , then
is called a descent datum for M . We will see from Grothendiecks Theorem
to be proved in 4.6/5 that any descent datum on an R -module M provides a
natural way to descend M to an R-module M . The more precise statement of
4.6/5 uses the language of categories and functors, which is quite appropriate
in this context and which will be explained in detail in Section 4.5. Although
we do not touch descent theory in the later part on Algebraic Geometry,
the same methods as presented in Section 4.6 can be applied in the setting of
quasi-coherent modules on schemes; see [5], 6.1/4.
Also note that the phenomenon we have encountered before, namely that an
R -module M may be viewed as the extension via R - R of dierent non-
isomorphic R-modules M , corresponds to the fact that there can live dierent
incompatible descent data on a single R -module M .
4.1 Tensor Products
Let M and N be R-modules. Recall that a map : M N - E to some

R-module E is called R-bilinear if, for all x M and y N , the maps
(x, ) : N - E, z - (x, z),

(, y) : M - E, z - (z, y),
are R-linear, by which we mean that they dene morphisms of R-modules.
Denition 1. A tensor product of M and N over R consists of an R-module

T together with an R-bilinear map : M N - T such that the following
universal property holds:
For each R-bilinear map : M N - E to some R-module E, there is
a unique R-linear map : T - E such that = , i.e. such that the
diagram

M N - T

?
E
is commutative.
Remark 2. Tensor products are uniquely determined by the dening universal

property, up to canonical isomorphism.
The proof consists of a well-known standard argument, which we would like

to repeat once more. Let
: M N - T, : M N - T
be tensor products of M and N over R. Then there is a diagram

M N - T
-

?

T
with R-linear maps , , where the existence of satisfying = follows
from the universal property of : M N - T and, likewise, the existence of
satisfying = from the universal property of : M N - T . Then
we have
idT = = = ( ) .
Thus, considering : M N - T as an R-bilinear map and using the universal
property of , there is a unique R-linear map : T - T such that = .
Because, as we have just seen, the maps = idT as well as = solve this
problem, we get necessarily = idT . Likewise, we can conclude = idT
from
idT = = = ( ) .
Therefore and are mutually inverse isomorphisms between T and T , and
we see that both tensor products are canonically isomorphic.
When considering a tensor product : M N - T of two R-modules

M and N , we will write more specically M R N in place of T and, further-
more, x y in place of (x, y), for x M and y N . We call x y the
tensor constructed from x and y. Using this notation, the R-bilinear map is
characterized by
M N - M R N, (x, y) - x y.
In particular, tensors are R-bilinear in their factors, i.e.
(ax + a x ) (by + b y )
= ab(x y) + ab (x y ) + a b(x y) + a b (x y )
for a, a , b, b R, x, x M, y, y N . In most cases, we will not mention the

dening R-bilinear map : M N - M R N explicitly and just call M R N
the tensor product of M and N over R. Proceeding like this, we assume that all
tensors x y in M R N are known, which allows us to reconstruct the map .
Proposition 3. The tensor product T = M R N exists for arbitrary R-modules

M and N .
Proof. The idea behind the construction of the tensor product M R N is quite
simple. We consider R(M N ) as an R-module,
e(x,y) = (x,x y,y )(x ,y )M N , x M, y N,
being its canonical free generating system. Writing (x, y) instead of e(x,y) , we
may view R(M N ) as the free R-module generated by all pairs (x, y) M N .
Then we divide out the smallest submodule Q R(M N ) such that the residue
classes (x, y) of the elements (x, y) = e(x,y) R(M N ) acquire the property of
tensors. This means, we consider the submodule Q R(M N ) generated by all
elements of type
(x + x , y) (x, y) (x , y),

(x, y + y ) (x, y) (x, y ),
(ax, y) a(x, y),
(x, ay) a(x, y),
where a R, x, x M , y, y N . Setting T = R(M N ) /Q, the canonical

map : M N - T , sending a pair (x, y) to the residue class (x, y) T ,
is then R-bilinear. We claim that satises the universal property of a tensor
product, as mentioned in Denition 1. To justify this, let : M N - E be
an R-bilinear map to some R-module E. There is an associated R-linear map
: R(M N ) - E given as follows: set (x, y) = (x, y) for the canonical free
generators (x, y) R(M N ) and dene on all of R(M N ) by R-linear extension.
Then the R-bilinearity of implies that ker contains all generating elements
of Q, as listed above, so that induces an R-linear map : R(M N ) /Q - E
satisfying = :
: M N - R(M N ) - R(M N ) /Q

?

E
Finally, observe that is uniquely determined by the relation = . Indeed,
the residue classes (x, y) for (x, y) M N generate R(M N ) /Q as an R-module
and we have
(x, y) = (x, y) = (x, y).
Therefore is unique on a set of generators of T = R(M N ) /Q and, thus, unique
on T itself.
To handle tensor products, the explicit construction, as given in the proof

of Proposition 3, is only of minor importance. In almost all cases it is much
more appropriate to derive the needed information from the universal property
of tensor products.
Remark 4. Let M and N be R-modules. Then every element z M R N

can be written as a nite sum of tensors, say z = ri=1 xi yi with elements
xi M and yi N .
Proof. The assertion follows directly from the explicit construction of tensor
products, as given in the proof of Proposition 3, but can also be deduced from
the universal property of tensor products as follows. Consider the submodule
T M R N generated by all tensors x y where x M and y N . Then the
canonical R-bilinear map : M N - M R N restricts to an R-bilinear map
: M N - T , which, just as , satises the universal property of a tensor
product of M and N over R. Indeed, if : M N - E is an R-bilinear map
to some R-module E, there exists an R-linear map : M R N - E such
that = . Then the restriction = |T satises = . Furthermore,
is uniquely determined by this condition, since, on tensors (x, y) for x M ,
y N , it is given by ( (x, y)) = (x, y). Therefore, T is a tensor product of
M and N over R, and the uniqueness assertion in Remark 2 implies that the
inclusion map T - M R N is bijective so that T = M R N .
Corollary 5. Let (xi )iI and (yj )jJ be generating systems of two R-modules
M and N . Then (xi yj )iI,jJ is a generating system of M R N .
Proof. Given x M and y N , there are equations of type

x= ai x i , y= b j yj ,
iI jJ
with coecients ai , bj R, and we get

xy = ai xi b j yj = ai bj xi yj .
iI jJ iI,jJ
Using Remark 4, it follows that (xi yj )iI,jJ is a generating system of M R N .

Working with tensors, a little bit of caution is necessary. For example, we

have 2 1 = 0 in (2Z) Z (Z/2Z), as we will see further below. On the other
hand, the equation 2 1 = 0 holds in Z Z (Z/2Z), because the bilinearity of
tensors shows
2 1 = 2 (1 1) = 1 2 = 1 0 = 0
in Z Z (Z/2Z). Therefore, when considering tensors, the associated tensor
products should always be specied, unless this is clear from the context.
Special care is also necessary when constructing R-linear maps from a tensor
product to another R-module. The following principle, which is a reformulation
of the dening universal property of tensor products, is useful:
Remark 6. Let M, N be R-modules and (zx,y )xM,yN a family of elements of

some R-module E.
(i) If : M R N - E is an R-linear map satisfying (x y) = zx,y for

all x M , y N , then is unique.
(ii) Assume that the map : M N - E given by (x, y) = zx,y is
R-bilinear. Then there exists an R-linear map : M R N - E such that
(x y) = zx,y for all x M , y N , and is unique by (i).
It is quite practical to dene R-linear maps M N - E using the

assertion (ii) above, as we will see below.
Remark 7. Let M, N, P be R-modules and F R a free R-module, generated

by a single element e. Then there exist canonical isomorphisms of R-modules
F R M - M, ae x - ax,
M R N - N R M, xy - y x,
(M R N ) R P - M R (N R P ), (x y) z - x (y z).
Proof. In all three cases the construction of the required isomorphism is accom-
plished in more or less the same way. We keep the third assertion, which is a
bit more laborious than the others, for Exercise 1 below and discuss only the
rst assertion here. The map
F M - M, (ae, x) - ax,
is R-bilinear and, thus, induces a well-dened R-linear map
: F R M - M where ae x - ax.
Moreover, we can consider the R-linear map
: M - F R M, x - e x.
Then, clearly, = idM and also
(ae x) = (ax) = e (ax) = a(e x) = ae x
for a R, x M . Thus, and are mutually inverse to each other and

therefore bijective.
As an application, we see immediately that the tensor 2 1 2Z Z Z/2Z is

non-trivial, as claimed above. Indeed, the isomorphism 2ZZ Z/2Z - Z/2Z
obtained from Remark 7 maps the tensor 2 1 to the residue class 1 = 0 in
Z/2Z.
Proposition 8. Let (Mi )iI be a family of R-modules and N another R-module.

Then there exists a canonical isomorphism

Mi R N - (Mi R N ), (xi )iI y - (xi y)iI .
iI iI
Proof. The map

Mi N - (Mi R N ), (xi )iI , y - (xi y)iI ,
iI iI
is R-bilinear and, thus, induces an R-linear map

: Mi R N - (Mi R N )
iI iI
of the type mentioned in

the assertion. To exhibit an inverse of , look at the
inclusion maps Mj - iI Mi , for j I, and at the induced maps

Mj N - Mi N - Mi R N.
iI iI
These are R-bilinear and therefore give rise to R-linear maps

j : Mj R N - Mi R N, j I,
iI
and, hence, to an R-linear map

: (Mi R N ) - Mi R N, (zi )iI - i (zi ).
iI iI iI
It is easily checked that and are inverse to each other.
Corollary 9. Let M be an R-module with a free generating system (xi )iI .

Then, for any R-module N , there are isomorphisms

M R N (Rxi R N ) N (I) .
iI

Proof. Since (xi )iI is a free generating system of M , we have M = iI Rxi .
Therefore the rst isomorphism follows from Proposition 8 and, since Rxi R,
the second from Remark 7.
Exercises
1. Associativity of tensor products: As claimed in Remark 7, show that there is a
canonical isomorphism of R-modules (M R N ) R P - M R (N R P ) for
given R-modules M, N, P .
2. n-fold tensor products: Let M1 , . . . , Mn be R-modules, n 1. Show that there
exists an R-multilinear map : M1 . . .Mn - T into some R-module T such
that the following universal property is satised: For every R-multilinear map
: M1 . . . Mn - E into some R-module E, there is a unique morphism of
R-modules : T - E such that = . We write T = M1 R . . . R Mn
and call this the tensor product of the R-modules Mi .
4.2 Flat Modules 113
3. For R-modules M, N, E show that there are canonical isomorphisms of R-modules

HomR (M, HomR (N, E)) HomR (M R N, E) HomR (N, HomR (M, E)).
4. Let m, n Z be prime to each other. Show Z/mZ Z Z/nZ = 0.
5. Show dimK (V K W ) = dimK V dimK W for vector spaces V, W over a eld K.
6. Let : R - R be a ring morphism and consider R -modules M , N . Given
any R -module E , write E/R
for the module obtained from E by viewing it
as an R-module via . Show that there is a canonical morphism of R-modules
N
M/R R /R
- (M R N )/R and that the latter is surjective.
7. Compatibility of tensor products with cartesian products: Let (Mi )iI be a family
of R-modules and N another R-module. Show that there is a canonical morphism
of R-modules : ( iI Mi ) R N -
iI (Mi R N ) and that the latter is an
isomorphism if I is nite. Show that does not need to be injective nor surjective
if I is innite. Hint: Take I = N, R = Z, and N = Q. Choose a prime p N
and set Mi = Z/pi Z for i N to show that will not be injective. Furthermore,
taking Mi = Z for all i N, it is seen that is not surjective.
4.2 Flat Modules
Given two morphisms of R-modules : M - M and : N - N , their

tensor product over R is dened as the R-linear map
: M R N - M R N , xy - (x) (y),
which is well-dened, due to the fact that the map M N - M R N ,

(x, y) - (x) (y), is R-bilinear in x and y. In particular, we can consider
the tensor product
idN : M R N - M R N
of an R-linear map : M - M with the identity map idN : N - N on

any R-module N . Thereby we tensor with N over R, as we will say. In the
same way, we can tensor sequences of R-linear maps with N . As is easily seen,
the process of tensoring R-linear maps with an R-module N commutes with the
composition of such maps.
Proposition 1. Let

M - M - M - 0
be an exact sequence of R-modules. Then, for any R-module N , the sequence
idN idN
M R N - M R N - M R N - 0
obtained by tensoring with N is exact. The property is referred to as the right

exactness of tensor products.
Proof. First observe that im( idN ) ker( idN ), since
( idN ) ( idN ) = ( ) idN = 0.
Therefore idN admits the factorization

idN
M R N - M R N
-

?
(M R N )/ im( idN )
and we see:
im( idN ) = ker( idN ) is injective,

idN is surjective is surjective.
Thus, in order to nish the proof of the proposition, it remains to show that
is an isomorphism.
Looking for an inverse of , let us consider the map
: M N - (M R N )/ im( idN ), (x , y) - (x ) y,
where (x ) M denotes an arbitrary -preimage of x M . We claim

that is well-dened in the sense that it is independent of the choices of the
-preimages (x ). Indeed, if x1 , x2 M are two -preimages of some element
x M , we can conclude x1 y = x2 y for all y N as follows. Since
x1 , x2 1 (x ), we get x1 x2 ker = im , and there will exist some
x M such that x1 x2 = (x ). But then it follows
x1 y x2 y = (x1 x2 ) y = (x ) y = ( idN )(x y) = 0.
This being done, it is easy to see that is R-bilinear. Indeed, consider two
elements x1 , x2 M with -preimages (x1 ), (x2 ) M . Then (x1 ) + (x2 ) is
a -preimage of x1 + x2 and we may assume (x1 + x2 ) = (x1 ) + (x2 ), thereby
obtaining
(x1 + x2 , y) = (x1 + x2 ) y = ((x1 ) + (x2 )) y

= (x1 ) y + (x2 ) y = (x1 , y) + (x2 , y).
Likewise, for a R, we may assume (ax1 ) = a(x1 ) and see
(ax1 , y) = (ax1 ) y = (a(x1 )) y = a((x1 ) y) = a(x1 , y).
Therefore is R-linear in the rst argument and, for trivial reasons, also in the
second one.
As a consequence, induces an R-linear map
: M R N - (M R N )/ im( idN ), x y - (x ) y,
which is related to as follows:

x y - (x ) y - x y

xy - (x) y - (x) y = x y
In particular, we get = id as well as = id, and is an isomorphism,

thereby concluding our proof.
We want to sketch a second argument of proof, which is more conceptual,
but a bit more advanced. It uses the functor Hom, briey touched in Section 1.4
and to be explained in more detail in Section 5.3, as well as its left exactness;
for the concept of functors, see Section 4.5. We start out from the commutative
diagram
idN
M
R N
- M
R N
- coker( idN ) - 0

?
idN idN
M R N - M R N - M R N - 0
where the rst row is the exact sequence associated to the morphism idN
and the second one satises at least the complex property. As before, we have
to show that is an isomorphism. To do this x any R-module E and apply
the functor hE = HomR (, E) to the above diagram, thereby obtaining the
commutative diagram
0 - hE (M R N ) - hE (M R N ) - hE (M R N )

hE ( )

?
0 - hE (coker( idN )) - hE (M R N ) - hE (M R N ) .
The second row is exact since the contravariant functor hE = HomR (, E) is

left exact in the sense that it turns right exact sequences into left exact ones;
see Section 5.3. We claim that the rst row is exact as well and, hence, that
hE ( ) will be an isomorphism. Indeed, using the universal property of the tensor
product, we can identify the rst row with the canonical sequence
0 - BilR (M N, E) - BilR (M N, E) - BilR (M N, E)
of R-bilinear maps to E. Since there is a natural identication
BilR (M N, E) = HomR (M, HomR (N, E))
and similarly for M and M (see Exercise 4.1/3), the latter sequence may be
viewed as the sequence
0 - hhE (N ) (M ) - hhE (N ) (M ) - hhE (N ) (M )

derived from M - M - M - 0 by applying the functor hh (N ) . This

E
functor is left exact. Hence, it follows that, indeed, the rst row in the above
diagram is exact so that the map
hE ( ) : hE (M R N ) - hE (coker( idN ))
is an isomorphism. Now observe that this holds for every R-module E and that
then : coker( idN ) - M R N must be an isomorphism as well. Indeed,
taking E = coker( idN ), the map hE (M R N ) corresponding to the
identity map in hE (E) will satisfy = id, whereas the relation = id
follows with the help of the isomorphism hE ( ) for E = M R N .
As an application, let us consider two R-modules M, N and a submodule

M M . In order to give a description of the tensor product (M/M ) R N ,
look at the canonical exact sequence

M - M - M/M - 0
and tensor it over R with N , thereby obtaining the exact sequence

idN idN
M R N - M R N - (M/M ) R N - 0.
The latter induces an isomorphism
(M R N )/ im( idN ) - (M/M ) R N, xy - x y,
where, as we want to point out, the tensor product map idN will not be
injective in general, even if : M - M has this property, as in our case. Thus,
in most cases, it is not permitted to view M R N as a submodule of M R N ,
since we cannot generally identify M R N with its image im(idN ). A typical
example of this kind is given by the canonical inclusion map : 2Z - Z.
Namely, applying the isomorphism 2Z Z in terms of Z-modules, the tensor
product map id : 2Z Z Z/2Z - Z Z Z/2Z corresponds to the zero map
0 : Z/2Z - Z/2Z.
Let us add that the situation can be described in more explicit terms if we
tensor the inclusion map of some ideal a - R with an R-module N :
Corollary 2. Let N be an R-module and a R an ideal. Then there is a

commutative diagram with exact rows
a R N - R R N - R/a R N - 0

? ? ?
0 - aN - N - N/aN - 0,
where is surjective, is the canonical isomorphism of 4.1/7, and is the
isomorphism induced from . In particular, there is a canonical isomorphism
R/a R N - N/aN, rx - rx.

Proof. The upper row of the diagram is obtained by tensoring the exact sequence
a - R - R/a - 0 over R with N . Hence, by Proposition 1, it is exact.
For the remaining part of the assertion, consider the isomorphism
: R R N - N, rx - rx,
of 4.1/7 and observe that it maps the image of a R N - R R N onto the

submodule aN N .
We take the phenomenon that an injective morphism of R-modules might

not remain injective when tensoring it with some R-module N , as reason to
introduce the notion of atness for modules.
Denition 3. An R-module N is called at if for every monomorphism of

R-modules M - M the map M R N - M R N obtained by tensoring
over R with N is injective. A ring homomorphism : R - R is called at

if R , viewed as an R-module via , is at.
For example, it follows from 4.1/7 that any ring viewed as a module over
itself is at. Furthermore, the direct sum of at modules is at by 4.1/8. In
particular, any free module is at.
Proposition 4. For an R-module N the following conditions are equivalent:

(i) N is at.
(ii) If 0 - M -M - M - 0 is a short exact sequence of
R-modules, the sequence
0 - M R N - M R N - M R N - 0
obtained by tensoring over R with N is exact.

(iii) If M - M - M is an exact sequence of R-modules, the sequence
M R N - M R N - M R N
obtained by tensoring over R with N is exact.
Proof. Using the denition of atness, the implication (i) = (ii) follows imme-
diately from Proposition 1.
Next, assume condition (ii). In order to derive (iii), consider an exact se-

quence M - M - M and look at the associated commutative diagram
with exact rows:
M - im
- 0

-
0 - ker - M - im - 0

-
0 - im - M
Since the rst and the third rows can be completed to form short exact se-
quences, it follows from (ii) that the diagram
M R N - R N
im - 0

idN
-
0 - ker R N - M R N - im R N -0

idN
-
0 - im R N - M R N
obtained by tensoring over R with N has exact rows as well. In particular, the
sequence
idN
M R N - M R N idN- M R N
is exact.
Finally, the implication (iii) = (i) is trivial.
If M is an R-module and M - M a submodule, then, for any at

R-module N , the map M R N - M R N obtained by tensoring over R
with N remains injective. Thus, in this case, M R N can canonically be viewed
as a submodule of M R N .
Corollary 5. Let N be a at R-module and : M - M a morphism of

R-modules. Then the tensor product R N commutes with the formation of
ker , coker and im , i.e. there are canonical isomorphisms of R-modules
(ker ) R N - ker( idN ), (coker ) R N - coker( idN ),
(im ) R N - im( idN ).
Proof. The claimed isomorphisms are worked out by tensoring the exact se-
quence

0 - ker - M - M - coker - 0,
as well as the diagram
M - im - 0

-
0 - im - M ,
over R with N , using the fact that exact sequences remain exact when tensoring
them with a at module.

Proposition 6. Let (Ni )iI be a family of R-modules. The direct sum iI Ni
is at if and only if Ni is at for all i I.
Proof. Use the fact (4.1/8) that the tensor product commutes with direct sums.

Applying the canonical isomorphism M R R - M from 4.1/7 to

R-modules M and their submodules, it is easily seen that R, as an R-module,
is at. Therefore the above proposition shows that the direct sum R(I) , for an
arbitrary index set I, is a at R-module. In other words, every free R-module is
at. Likewise, any polynomial ring RX, for a set of variables X, is at over R.
Proposition 7. Let N be an R-module.

(i) If N is at and a R is not a zero divisor in R, then an equation
ax = 0 for some x N implies x = 0.
(ii) If R is a principal ideal domain, N is at if and only if any equation
ax = 0 for a R and x N implies a = 0 or x = 0.
Proof. Starting with assertion (i), assume that a R is not a zero divisor. Then
the map R - R, r - ar, is injective and, for any at R-module N , the
map
R R N - R R N, r x - ar x,
obtained by tensoring over R with N has the same property. Now apply 4.1/7
and identify R R N with N via r x - rx. We thereby see that the map
N - N, x - ax, is injective, as claimed.
Next assume that R is a principal ideal domain. Then no element in R {0}
is a zero divisor and the only-if part of (ii) follows from (i). Thus, it remains to
show that N is at if an equation of type ax = 0 for some elements a R and
x N can only hold if a = 0 or x = 0; the latter corresponds to the condition
of N being torsion-free. Now if N is nitely generated, the structure theorem
for nitely generated modules over principal ideal domains (see [3], Section 2.9)
says that N is free and, hence, at.
Using the lemma below, this argument extends to the case where N is not
necessarily nitely generated. Indeed, consider
a monomorphism of R-modules
: M - M and an element z = x yi belonging to the kernel of
iI i

idN : M R N - M R N . Then iI (xi ) yi , viewed as an element in
M R N , is zero. Therefore Lemma 8 provides a nitely generated submodule
N N such that, looking at the canonical commutative diagram
idN
M R N - M R N
? ?
idN
M R N - M R N ,
the element z ker(idN ) admits a preimage z ker(idN ). But we know

already that idN must be injective, since N is nitely generated. Hence,
z = 0 and therefore also z = 0 so that idN is injective. It follows that N is
a at R-module.
Lemma 8. Let M, N be two R-modules and consider an equation of type

iI xi yi = 0 for some elements xi M , yi N , and a nite index set I.
Then there exist nitely generated submodules M M and
N N such that

xi M , yi N for all i I, and such that the equation iI xi yi = 0 holds
already in M R N .
Proof. Look back at the construction of M R N as a quotient R(M N ) /Q, where

R(M N ) is the free R-module generated by all symbols (x, y) with x M , y N
and where Q R(M N ) is the submodule generated by all relations that are
needed in order to render the symbols
(x, y) R-linear in both arguments. Then
iI x i y i = 0 is equivalent to iI (x i , yi ) Q. Now choose nitely generated
submodules M M and N N that are suciently large in order to satisfy
the following conditions:
(1) xi M , yi N for all i I.

(2) All generators of Q that are needed to write iI (xi , yi ) as an R-linear
combination of these, are contained in Q , where Q stands for module of rela-

tions needed for the construction of the tensor product M R N = R(M N ) /Q .

Then, clearly, iI (xi , yi ) Q and therefore iI xi yi = 0 in M R N .

Regarding the above proof, one may ask if it is really necessary to go back
to the explicit construction of tensor products. Indeed, it must be admitted that
this is only an ad hoc solution at this place. A more rigorous approach would
establish the compatibility of tensor products with so-called inductive limits, as
considered in Section 6.4.
Alternatively, the proof of Proposition 7 (ii) can also be settled by using
the following criterion of atness (for a proof see 5.2/8):
Proposition 9. An R-module N (over an arbitrary ring R) is at if and only

if for every inclusion a - R where a is an ideal in R, the induced map
a R N - R R N is injective.
Indeed, if R is a principal ideal domain, every ideal a R is principal, say

a = (a) with a generator a a. If a is non-trivial, then a, as an R-module,
admits a as a free generating system and, thus, is isomorphic to R. Looking at
the canonical isomorphisms
(a) R N - N, ra x - rx,
R R N - N, rx - rx,
of 4.1/7, it follows that the injectivity of the canonical map aR N - RR N
is equivalent to the injectivity of the multiplication by a on N .
Finally, let us introduce the notion of faithful atness, which is a certain
strengthening of atness.
Denition 10. An R-module N is called faithfully at if the following condi-

tions are satised :
(i) N is at.
(ii) If M is an R-module such that M R N = 0, then M = 0.
A ring homomorphism : R - R is called faithfully at if R , viewed
as an R-module via , is faithfully at.
In particular, condition (ii) implies that a faithfully at module over a ring

R = 0 must be non-zero.
Proposition 11. For an R-module N , the following conditions are equivalent:

(i) N is faithfully at.
(ii) N is at and, given a morphism of R-modules : M - M such that
idN : M R N - M R N is the zero morphism, then = 0.
(iii) A sequence of R-modules M -M - M is exact if and only if
the sequence M R N - M R N - M R N obtained by tensoring over
R with N is exact.
(iv) N is at and, for every maximal ideal m R, we have mN = N .
Proof. We will freely use the fact that, according to Corollary 5, the tensor
product with a at R-module N is compatible with the formation of kernels
and images of R-module homomorphisms. Starting with the implication from
(i) to (ii), assume that N is faithfully at and let : M - M be a morphism
of R-modules. Then = 0 is equivalent to im = 0 and, hence, using (i), to
im( idN ) = (im ) R N = 0, thus, to idN = 0.

A sequence of R-module homomorphisms M - M - M is exact if
and only if we have im( ) = 0 and the canonical map ker - ker / im
is zero. Given (ii), this is equivalent to

im ( idN ) ( idN ) = im ( ) idN = 0
and the condition that
ker R N - (ker / im ) R N = (ker R N )/(im R N )
is the zero map, hence to im R N = ker R N . Therefore (ii) implies

condition (iii).
Next assume (iii). To derive (i), observe that tensoring with N over R
respects exact sequences then. Hence, N is at. In order to show that N is even
faithfully at, consider an R-module M satisfying M R N = 0. This implies
that tensoring the sequence 0 -M - 0 over R with N yields an exact
sequence 0 - M R N - 0. But then, given condition (iii), the sequence
0 -M - 0 must be exact and we get M = 0.
Now, let us assume (i) and derive (iv). If m R is a maximal ideal, the
quotient R/m is non-zero. Hence, by Corollary 2, the faithful atness of N yields
N/mN N R R/m = 0
and therefore mN = N . Thus, (iv) is clear.

It remains to establish the implication (iv) = (i). To do this, assume that
N is at and that N/mN N R R/m = 0 for all maximal ideals m R.
We will show that the tensor product M R N is non-trivial for every non-
trivial R-module M = 0. Therefore, let M be a non-trivial R-module and let
x M {0}. Looking at the submodule M = Rx M and the epimorphism
R - M , a - ax,
as well as the corresponding kernel a R, we get an isomorphism of R-modules

R/a - M . The inclusion M M induces a monomorphism
(R/a) R N M R N - M R N,
since N is at. Thus, to show that M R N is non-trivial, it is enough to show

that (R/a) R N is non-trivial. Now R/a M is non-zero by its denition.
Hence, a R is a proper ideal and there exists a maximal ideal m R such
that a m. Considering the commutative diagram with exact rows
(R/a) R N - (R/m) R N - 0

N/aN - N/mN - 0
where N/mN = 0 by (iv), we get
(R/a) R N N/aN = 0,
as desired.
We can conclude from Proposition 11 that any ring, viewed as a module

over itself, is faithfully at, as it is at anyway. In particular, a free module over
a non-zero ring is faithfully at if and only if it is non-zero.
4.3 Extension of Coecients 123
Exercises

1. Let M - M - M as well as N - N - N be morphisms of
R-modules. Show that the resulting morphisms M R N - M R N given

by ( ) ( ) and ( ) ( ) coincide.
2. Let M be an R-module such that each nitely generated submodule of M is at.
Show that M is at itself.
3. Let M, N be at R-modules. Show that M R N is a at R-module.
4. Let M be a module over an integral domain R such that the annihilator ideal
{a R ; aM = 0} R is non-zero. Show that M cannot be at.
5. Consider the polynomial ring KX, Y in two variables over a eld K. Show that
the ideals (X), (Y ), (XY ) KX, Y are faithfully at KX, Y -modules, but
that the ideal (X, Y ) KX, Y fails to be at. Hint: There is a canonical exact
sequence of KX, Y -modules 0 - (XY ) - (X) (Y ) - (X, Y ) - 0.

6. Let (Mi )iI be a family of R-modules. Show that the direct sum iI Mi is
faithfully at if and only if all Mi are at and at least one of these is faithfully
at.
7. Let : R - R be a at ring morphism. Show that is faithfully at if and
only if the associated map Spec R - Spec R, p - p R, is surjective.
8. Let P be a projective R-module (in the sense that there exists another R-module
P such that P P is free). Show that P is at.
9. Let (Mi )iI be a family of R-modules and
N an R-module-
of nite
presentation.
Show that the canonical morphism : ( iI Mi ) R N iI (M i R N ) of
Exercise 4.1/7 is an isomorphism.
10. Let 0 - M - M - M - 0 be a short exact sequence of R-modules

where M is at. Show that the resulting sequence
0 - M R N - M R N - M R N - 0
is exact for all R-modules N . Hint: View N as a quotient of a free R-module

and use diagram chase. It is not necessary to apply the theory of Tor modules,
as done later in 5.2/9.
4.3 Extension of Coecients
Let : R - R be a ring homomorphism and N an R -module. Then N

can be viewed as an R-module via ; just look at N as an additive group and
dene the scalar multiplication by rx = (r)x for elements r R and x N
where (r)x is the product on N as an R -module. We say that the R-module
structure on N is obtained by restriction of coecients with respect to . In
particular, R itself can be viewed as an R-module via , and we see that the
tensor product M R R , for any R-module M , makes sense as an R-module.
It is easily seen that M R R can even be viewed as an R -module. Indeed, let
a R and consider the R-bilinear map
M R - M R R , (x, b ) - x a b ,
as well as the induced R-linear map
M R R - M R R , x b - x a b .
Taking the latter as multiplication by a R on M R R , it is straightforward

to see that the additive group M R R becomes an R -module this way. We
say that M R R , viewed as an R -module, is obtained from M by extension of
coecients with respect to . More generally, given any R -module N , we can
view the tensor product M R N as an R -module, just by using the map
R (M R N ) - M R N , (a , x y ) - x a y
as scalar multiplication.
Remark 1. Let : R - R be a ring homomorphism and M an R-module.

(i) If (xi )iI is a generating system of M over R, then (xi 1)iI is a
generating system of M R R over R .
(ii) If (xi )iI is a free generating system of M over R, then (xi 1)iI is a
free generating system of M R R over R .
Proof. Assertion (i) is veried by direct computation, whereas assertion (ii) is a

consequence of 4.1/9.
Remark 2. Let R - R - R be ring morphisms and M an R-module.

Then there is a canonical isomorphism of R -modules
(M R R ) R R - M R R , (x a ) a - x a a .
Similarly, for any R -module N there is a canonical isomorphism of R -modules
(M R R ) R N - M R N , (x a ) y - x a y .
Proof. Starting with the second assertion, x y N and observe that the map
M R - M R N , (x, a ) - x a y ,
is R-bilinear. Thus, it induces a map
M R R - M R N , x a - x a y ,
which is R-linear, and even R -linear. Therefore we can consider the R -bilinear
map

(M R R ) N - M R N , (x a ), y - x a y ,
as well as the induced R -linear map

(M R R ) R N - M R N , (x a ) y - x a y .
The latter is an isomorphism, since the map
M R N - (M R R ) R N , x y - (x 1) y ,
serves as an inverse.
Now, replacing N by R , viewed as an R -module via R - R , the above
argument yields an isomorphism of R -modules
(M R R ) R R - M R R , (x a ) a - x a a ,
and it is easily seen that the latter is even R -linear.
Next we want to generalize the process of localization from rings to modules.

To do this, consider an R-module M and a multiplicative system S R. Then
we dene a relation on M S as follows:
(x, s) (x , s ) there is some t S such that (xs x s)t = 0
As in the case of localization of rings, one shows that satises the condi-
tions of an equivalence relation. Writing xs for the equivalence class of an element
(x, s) M S and x
MS = ; x M, s S
s
for the set of all these equivalence classes, it is easily checked that MS is an
RS -module under the rules of fractional arithmetic. We call MS the localization
of M by S. Similarly as in the case of rings, we write Mp instead of MRp if p is
a prime ideal in R, as well as Mf for the localization of M by the multiplicative
system {1, f 1 , f 2 , . . .} generated from a single element f R.
Proposition 3. Let S R be a multiplicative system. Then:

(i) R - RS is at, i.e. RS is a at R-module via R - RS .
(ii) For every R-module M , there is a canonical isomorphism
a - ax
M R RS - MS , x .
s s
Proof. Let us start with assertion (ii). Since we have so far skipped the details of
viewing MS as an RS -module, let us examine this structure more closely now.
The R-bilinear map
a ax
M RS - MS , x, - ,
s s

is well-dened. Indeed, any equation as = as for elements a, a R and s, s S,
where (as a s)t = 0 for some t S, implies (xas xa s)t = 0 and, thus,
ax
s
= asx . Therefore the above map induces an R-linear map
-
a -
ax
: M R RS MS , x ,
s s
which, clearly, is even RS -linear. In order to show that is an isomorphism, we
try to dene an RS -linear inverse by
-
x -
1
: MS M R RS , x .
s s
Obviously, is an inverse of , provided we can show that it is well-dened.

To justify the latter, assume xs = xs for elements x, x M and s, s S, say
(xs x s)t = 0 for some t S. This yields
1 1
x x ss t = (xs 1 x s 1)t = ((xs sx )t) 1 = 0.
s s
Using the fact that M R RS is an RS -module and the product ss t is a unit in
RS , we get
1 1
x x = 0.
s s
Hence, is well-dened.
It remains to justify assertion (i). To do this, let : M - M be a mono-
morphism of R-modules. We have to show that the map obtained by tensoring
with RS over R or, applying (ii), the map
x (x)
S : MS - MS , - ,
s s
is injective. To do this, consider elements x M and s S where S ( xs ) = 0
and, hence, (x)
s
= 0. Then there is some t S such that (tx) = t(x) = 0,
and we can conclude tx = 0 since is injective. But then we must have xs = 0
in MS , as desired.
As an application of the localization of modules, we want to give a local

characterization of atness. Such a characterization will be of special interest
later, when dealing with at morphisms of schemes; see for example 8.5/17. We
begin with a simple auxiliary result.
Lemma 4. For any R-module M the canonical map

M - Mm
mSpm R
into the cartesian product of all localizations of M by maximal ideals m R is

injective. In particular, M = 0 if Mm = 0 for all m Spec R.
Proof. Consider an element x M , whose image in Mm is trivial for all maximal

ideals m R. Then the module analogue of 1.2/4 (i) shows for every such m
that there is an element am R m satisfying xam = 0. Since the elements am
for m Spm R generate the unit ideal in R, we must have x = 0.
Proposition 5. For a ring homomorphism : R - R and an R -module N

the following conditions are equivalent:
(i) N is at when viewed as an R-module via .
(ii) For every maximal ideal m R the localization Nm is a at R-module.
(iii) For every maximal ideal m R the localization Nm is a at Rm -module
where m = 1 (m ).
The same equivalences are valid for prime ideals p R in place of maximal
ideals.
Proof. Let N be a at R-module as in (i) and x a maximal ideal (or a prime

ideal) m R . To show that the localization Nm is a at R-module, look at
a monomorphism of R-modules M - M . Then the derived morphism of
R-modules M R N - M R N is injective, since N is at over R. Further-

more, Rm is at over R by Proposition 3. Consequently, using the associativity
of tensor products, the morphism of R-modules
M R Nm = M R N R Rm

-
M R N R Rm = M R Nm
is injective and we see that Nm is a at R-module. Thus, (i) implies (ii).
Next let us show that (ii) is equivalent to (iii). Assume rst that Nm is a
at R-module for a certain maximal ideal (resp. prime ideal) m R . Since
induces a ring morphism Rm - R for m = 1 (m ), it is clear that Nm
m
can be viewed as an Rm -module. Now if M - M is a monomorphism of
Rm -modules, the atness of Nm over R yields a monomorphism of R-modules
M R Nm - M R Nm .
Using the fact that M and M are already Rm -modules, the canonical mor-
phism M R Rm - M as well as the corresponding one for M in place of

M are isomorphisms. Therefore, in conjunction with Remark 2, the preceding
monomorphism can be written in the form
M Rm Nm = (M R Rm ) Rm Nm - (M R Rm ) Rm Nm = M Rm Nm
and we see that Nm is a at Rm -module. Conversely, if the latter is the case,
choose a monomorphism of R-modules M - M . Localizing it at m yields a
monomorphism of Rm -modules Mm - Mm and tensoring it with Nm over Rm
yields a monomorphism
M R Rm Rm Nm = Mm Rm Nm - Mm Rm Nm = M R Rm Rm Nm
if Nm is at over Rm . But then, applying Remark 2 again, the canonical mor-
phism M R Nm - M R Nm is injective and we see that Nm is at over
R.
It remains to show that (ii) implies (i). So assume that Nm is a at R-module
for all maximal ideals m R . Choosing a monomorphism of R-modules
M - M , there is a canonical commutative diagram
M R N - M R N

? ?

-
(M R N )m (M R N )m
m Spm R m Spm R
where M R N and M R N are viewed as R -modules via N and the vertical

maps are injective by Lemma 4. Using the isomorphisms
(M R N )m M R N R Rm

M R Nm , m Spm R ,
and the same ones with M replaced by M , we conclude that in the above
diagram the lower horizontal map is injective. Therefore the same will be true
for the upper one and we see that N is a at R-module.
Finally, let us briey address the subject of tensor products of R-algebras.

Recall from 1.4/3 and the explanations following it that an R-algebra may be
dened as a ring A together with a ring homomorphism f : R - A. We will
say that A is an R-algebra via f . In particular, there is an underlying structure of
an R-module on A. An R-algebra homomorphism between two such R-algebras
R - A and R - A is dened as a ring homomorphism A - A

that is compatible with the structures of A , A as R-algebras; i.e. such that the
diagram
A - A

R
is commutative. Furthermore, the tensor product A R A of two R-algebras
A and A can be constructed in the setting of R-modules, but is also a ring
under the multiplication
(a b) (c d) = ac bd,
which is easily seen to be well-dened. Therefore, via the canonical map
R - A R A , a - a 1 = 1 a,
the tensor product A R A is an R-algebra again.
Lemma 6. Let A and A be R-algebras. Then the canonical R-algebra homo-
morphisms
: A - A R A , a -a 1,
: A - A R A , a - 1 a ,
enjoy the following universal property:

For any two R-algebra homomorphisms : A - A and : A - A,
there is a unique R-algebra homomorphism : A R A - A such that the
diagram
A

? -

A R A - A
-
6

A
is commutative.
Proof. The uniqueness of follows from the equation

(a a ) = (a 1) (1 a ) = (a 1) (1 a ) = (a ) (a )
for a A , a A . To show the existence of , look at the R-bilinear map
A A - A, (a , a ) - (a ) (a ),
which induces an R-linear map
: A R A - A, a a - (a ) (a ).
It is easily checked that is an R-algebra homomorphism, making the diagram

mentioned in the assertion commutative.
As an application, let us discuss the extension of coecients for polynomial

rings.
Proposition 7. Let R be an R-algebra, X a system of variables and a RX

an ideal. Then there are canonical isomorphisms of R -algebras
RX R R - R X, f a - a f,
(RX/a) R R - R X/aR X, f a - a f .
Proof. We see from Lemma 6 that the canonical maps
: RX - R X, : R - R X
induce an R-algebra homomorphism
: RX R R - R X, f a - a f.
On the other hand, consider the homomorphism
R - RX R R , a - 1 a ,
which we extend to a ring homomorphism

R X - RX R R , X - X 1,
using the universal property of polynomial rings. It is easy to verify that
and are mutually inverse isomorphisms and, in fact, isomorphisms of R- or
R -algebras. Alternatively, we could have based our argument on the fact that
R X satises the universal property of the tensor product RX R R , in the
sense of R-algebras.
For the second isomorphism use 4.2/2.
Exercises
1. If R is a coherent ring (1.5/8), show that any localization RS is coherent as well.
2. Let X = (Xi )iI be an innite family of variables. Show that the polynomial ring
KX over a eld K is coherent (1.5/8), but not Noetherian.
3. Show for ideals a, b R that R/a R R/b R/(a + b). Deduce from this that
Z/mZ Z Z/nZ Z/gcd(m, n)Z for any integers m, n Z.

4. Show that Q( 2) Q Q( 3) is a eld, whereas
Q( 2) Q Q( 2)is not, because

the latter Q-algebra is isomorphic to Q( 2)X/(X 2 2) Q( 2) Q( 2).
5. Show that a morphism of R-modules M - N is injective (resp. surjective,
resp. bijective) if and only if the induced morphism Mm - Nm is injective
(resp. surjective, resp. bijective) for all maximal ideals m R or, alternatively,
for all prime ideals m R.
6. Compatibility of tensor products with extension of coecients: Given R-modules
M, N and a ring morphism R - R , show that there is a canonical isomorphism
(M R N ) R R (M R R ) R (N R R ).
7. For nitely generated modules M, N over a local ring R, show that M R N = 0
implies M = 0 or N = 0. Hint: Use Exercise 6 above.
8. Universal property of coecient extensions: Let R - R be a ring morphism
and consider an R-module M as well as an R -module N . Write N/R for the
R-module obtained from N by restricting coecients to R and show that there
is a canonical bijection
HomR (M R R , N ) - HomR (M, N )./R
9. Let M, N be R-modules and S R a multiplicative system. Show that there is

a canonical morphism of RS -modules

HomR (M, N ) S - HomR (MS , NS )
S
and that the latter is an isomorphism if M is of nite presentation. Give an

example where, indeed, the canonical morphism fails to be an isomorphism.
10. Show for a nitely generated ideal a R that the following conditions are equiv-
alent:
(a) R/a is at over R.
(b) a2 = a
(c) a = (e) for some idempotent element e R.
Hint: Use Nakayamas Lemma in the version of Exercise 1.4/5.
4.4 Faithfully Flat Descent of Module Properties 131
4.4 Faithfully Flat Descent of Module Properties
For any ring homomorphism R - R , the process R R of tensoring with

R over R can be viewed as an assignment that attaches to an R-module M
the R -module M R R and to a morphism of R-modules : M - M the
corresponding morphism of R -modules id : M R R - M R R . We
talk about a so-called functor from the category of R-modules to the category of
R -modules. As will be seen in Section 4.6, the general problem of descent is, in
some sense, to nd an inverse to this process. As a preparation, categories and
their functors will be discussed more extensively in Section 4.5. At this place
we start descent theory by looking at several module properties that behave
well when switching back and forth between R- and R -modules by means of
extension of coecients via tensor products.
Proposition 1. Let M be an R-module and R - R a ring homomorphism.

(i) If M is of nite type over R, then M R R is of nite type over R .
(ii) If M is of nite presentation over R, then M R R is of nite presen-
tation over R .
(iii) If M is a at R-module, then M R R is a at R -module.
(iv) If M is a faithfully at R-module, then M R R is a faithfully at

R -module.
(v) If R - R is faithfully at in the sense that R is a faithfully at
R-module via R - R , then the reversed implications of (i)(iv) hold as well.
Proof. If
Rn - M - 0
or
Rm - Rn - M - 0
are exact sequences of R-modules, then the sequences obtained by tensoring
with R over R are exact by 4.2/1. Using the fact that the isomorphisms
Rm R R (R )m and Rn R R (R )n furnished by 4.1/9 are, in fact, iso-
morphisms of R -modules, assertions (i) and (ii) are clear.
Now let M be a at R-module. To establish (iii), we have to show for every
monomorphism of R -modules E - E that the tensorized map
E R (R R M ) - E R (R R M )
is injective as well. To do this, look at the commutative diagram
E R (R R M ) - E R (R R M )
? ?
E R M - E R M
where the vertical maps are the canonical isomorphisms from 4.3/2. Since M is
a at R-module, the lower horizontal homomorphism is injective and the same
holds for the upper horizontal one.
If E is an R -module such that E R (R R M ) = 0, we get E R M = 0

from 4.3/2. Hence, E as an R-module and even as an R -module will be trivial
if M is faithfully at, which settles assertion (iv).
In order to verify (v), assume that R - R is faithfully at. Then, if

M R R is of nite type over R , there exists a nite generating system of
M R R , which we may assume to be of type xi 1, for i = 1, . . . , n. We claim
that the elements xi , i = 1, . . . , n, generate M . Indeed, let M = ni=1 Rxi M
be the submodule of M generated by these elements and consider the canonical
sequence M -M - 0. The latter is exact if and only if the tensorized

sequence M R R - M R R - 0 is exact, since R - R is faithfully
at. However, the tensorized sequence is exact, due to the denition of M .
Hence, M being the image of the nitely generated R-module M will be nitely
generated, too.
Next assume that M R R is of nite presentation over R and that
R - R is faithfully at. In particular, M R R is then of nite type and the
same is true for M , as we have just seen. Hence, there exists an exact sequence
of type

Rn -M -0
and it is enough to show that ker is nitely generated. Clearly, the tensorized
sequence
idR
(R )n - M R R - 0
is exact and ker( idR ) is of nite type by 1.5/7, the same being true for
(ker ) R R , due to 4.2/5. But then the reversed version of (i) shows that
ker is of nite type, and we see that M is of nite presentation.
For the reversed version of (iii), assume that M R R is at over R , keeping
the assumption that R - R is faithfully at. To see that M is at over R,
consider a monomorphism of R-modules E - E and look at the tensorized
map E R M - E R M , which must be shown to be a monomorphism, too.
Since R is at over R, the map R R E - R R E is injective and, since

M R R is at over R , the same is true for the map
(R R E ) R (R R M ) - (R R E) R (R R M ).
Using 4.3/2 in conjunction with 4.1/7, the latter can canonically be identied
with the map
R R (E R M ) - R R (E R M ),
obtained from E R M - E R M by tensoring with R over R. Since R is
faithfully at over R, we conclude from 4.2/11 that E R M - E R M is
injective, as desired. Consequently, M is a at R-module.
Finally, let M R R be a faithfully at R -module. If R - R is faithfully
at, we know already that M is at. To show that M is even faithfully at,
consider an R-module E such that E R M = 0. Using 4.3/2 we get
(E R R ) R (M R R ) E R (M R R ) (E R M ) R R = 0.
The faithful atness of M R R as an R -module yields E R R = 0 and the

faithful atness of R over R yields E = 0. Thus, E R M = 0 implies E = 0
and we see that M is faithfully at.
We just have seen that the properties for a module to be of nite type, of
nite presentation, at, or faithfully at behave well with respect to arbitrary
extension of coecients, but also with respect to the reverse process of descent,
when the coecient extension is faithfully at. One may ask if the same is true
for the property of a module to be free. Of course, for a free R-module M and
an arbitrary coecient extension R - R , the resulting R -module M R R
is free by 4.1/9. On the other hand, if M R R admits a free generating system,
for example a nite free one, it is not necessarily true that M has the same
property, even if R - R is faithfully at; see the example we give at the
end of this section. A property that is better adapted to descent is the one of
being locally free, which we want to discuss now. For the necessary details on
localizations of rings we refer to Section 1.2.
Denition 2. An R-module M is called locally free of nite rank if, for every
x Spec R, there is some f R such that f (x) = 0 and Mf = M R Rf is a
nite free 1 Rf -module.

(i) M is locally free of nite rank.
(ii) M is at and of nite presentation.
(iii) M is of nite presentation and M R Rm is free for all maximal ideals
m R.
In particular, Proposition 3 shows for a local ring R that a at R-module

is free as soon as it is of nite presentation. Let us point out that this assertion
remains true if we substitute nite type in place of nite presentation; see
Exercise 4 below. However, for the full version of Proposition 3, it is essential
to require the nite presentation in conditions (ii) and (iii).
Corollary 4. Let M be an R-module and : R - R a faithfully at ring

homomorphism. Then M is locally free of nite rank if and only if the same
holds for M R R as an R -module.
Proof of Corollary 4. Use Proposition 3 in conjunction with Proposition 1.
Proof of Proposition 3. We start with the implication (i) = (ii). Therefore let
M be locally free of nite rank. Then, for any x Spec R, there is an element
fx R such that fx (x) = 0 and Mfx is a nite free Rfx -module; as usual,
Mfx and Rfx denote the localizations of M and R by the multiplicative system
1
A nite free module is meant as a free module admitting a nite generating system; the
latter amounts to the fact that it admits a nite free generating system.
generated by fx in R. By construction, the elements fx , x Spec R, cannot

have a common zero on Spec R, which means that they cannot be contained in
a single prime or maximal ideal. As a consequence, they must generate the unit
ideal in R. But then nitely many of these, say f1 , . . . , fr , generate the unit
ideal in R, and it follows that the canonical homomorphism
R - R = R f1 . . . R fr
is faithfully at where R is considered as a ring under componentwise addition
and multiplication. Indeed, Rfi is at over R by 4.3/3 and R = ri=1 Rfi is at
over R by 4.2/6 since, in terms of R-modules, it can be viewed as the direct
sum of the Rfi . Furthermore, R is even faithfully at over R, as follows from
4.2/11. Indeed, if m R is a maximal ideal, there exists an index i such that
fi m, and this implies mRfi Rfi by 1.2/5. Thus, indeed, R - R is
faithfully at, and to show that M is at and of nite presentation over R, it is
enough to show that M R R , viewed as an R -module, has these properties;
cf. Proposition 1.
We write ei for the unit element in Rfi , thereby obtaining the relations
ei ej = ij in R , for i, j = 1, . . . , r, as well
as e1 + . . . + er = 1. Then any
R -module E can be decomposed as E = ri=1 Ei with Ei = ei E E R Rfi ,
using the canonical projection R - Rfi in conjunction with the right exact-
ness of tensor products 4.2/1. This way we obtain a bijective correspondence
between R -modules E and families (Ei )i=1,...,r consisting
of Rfi -modules Ei .
In particular, for E = M R R we have E ri=1 Mfi where in this case
Ei M R Rfi Mfi by 4.3/2. Now Mfi , as a free Rfi -module, is at, and we
claim that then M R R is at over R as well. To verify this, just observe that
the above correspondence extends in a natural way to R -linear maps and that
a tensor product of type E R F corresponds to the family (Ei Rfi Fi )i=1,...,r ;
use the relations ei ej = ij in conjunction with 4.1/8. Now write Mfi Rfnii and
let n = maxi=1,...,r ni . Then it is easy to construct an epimorphism of type

r
(R )n - (Rfi )ni M R R ,
i=1
whose kernel is nitely generated. Thus, M R R is of nite presentation over

R and we see that (i) implies (ii).
Next let us verify the implication (ii) = (iii). Assuming that M is at
and of nite presentation, choose a maximal ideal m R and let us show
that M R Rm is a free Rm -module. We know already from Proposition 1 that
M R Rm is at and of nite presentation over Rm . Therefore it is enough to
show that any at module of nite presentation over a local ring R is free. To do
this, consider such a module M over a local ring R with maximal ideal m, and
let x1 , . . . , xn M give rise to a basis of M/mM
as a vector space over R/m.
Nakayamas Lemma 1.4/10 then yields M = ni=1 Rxi , and we can consider the
canonical exact sequence
0 - K - Rn - M - 0
that is associated to the morphism of R-modules mapping the canonical free

generating system of Rn to the chosen elements x1 , . . . , xn M . Then K, the
kernel of this map, is of nite type by 1.5/7, since M is of nite presentation.
Now, tensoring the inclusion map m - R with the modules of the above
sequence, we get a commutative diagram
m R K - mn - m R M - 0
u1 u2 u3
? ? ?
0 - K - Rn - M - 0
with exact rows. Since M is at, the map u3 is injective, and the Snake
Lemma 1.5/1 yields an exact sequence
0 = ker u3 - coker u1 - (R/m)n - M/mM
where (R/m)n - M/mM is bijective by construction. But this implies

K/mK = coker u1 = 0 and, by Nakayamas Lemma, K = 0. Therefore
Rn - M must be an isomorphism and, hence, M is free.
It remains to consider the implication (iii) = (i). Thus, let M be of nite
presentation and assume that M R Rm is a free Rm -module for each maximal
ideal m R. We want to show that, given a prime ideal p R and a maximal
ideal m R containing it, there is an element f R m R p such
that Mf = M R Rf is free over Rf . To do this, x a free generating system
x1 , . . . , xn of Mm = M R Rm . Then there are elements yi M , si R m
such that xi = ysii , for i = 1, . . . , n. Since each si induces a unit in Rm , we may
assume si = 1 so that xi , for each i, admits yi as a preimage with respect to
the canonical map M - Mm . Then look at the R-linear map : Rn - M,
sending the canonical free generating system of Rn to the elements y1 , . . . , yn ,
and consider the attached exact sequence

() 0 - ker - Rn - M - coker - 0.
Tensoring the latter over R with Rm and using the atness of Rm over R, we
obtain the exact sequence
idRm
0 - (ker ) R Rm - Rm
n - M R Rm - (coker ) R Rm - 0.
By the denition of , the tensorized map idRm is an isomorphism. Therefore

we must have (ker ) R Rm = 0 and (coker ) R Rm = 0.
Next we claim that, for an R-module N of nite type satisfying N R Rm = 0,
there always exists an element f R m such that f N = 0 and, hence,
N R Rf = Nf = 0. Indeed, if z1 , . . . , zr is a generating system of N , we get
zi
1
= 0 in N R Rm = Nm , for all i. By the module analogue of 1.2/4 (i), there
exist elements f1 , . . . , fr R m such that fi zi = 0 in N . Therefore, setting
f = f1 . . . fr gives f R m and f N = 0, as claimed.
Let us apply this fact to the above situation. The R-module coker is nitely
generated and satises (coker ) R Rm = 0. Therefore we can nd an element
f R m such that (coker ) R Rf = 0. Now, using the compatibilities for

localizations 1.2/10 and 1.2/11, we may tensor the exact sequence () over R
with Rf and write R instead of Rf again. Thereby we can assume coker = 0
and, hence, that is surjective. Then ker is of nite type by 1.5/7, since M
is of nite presentation. Using the same argument as before, we conclude from
(ker )R Rm = 0 that there is an element f Rm such that (ker )R Rf = 0.
Then the exact sequence
idRf
0 = (ker ) R Rf - Rfn - M R Rf - (coker ) R Rf = 0
shows that idRf : Rfn - M R Rf is an isomorphism and, hence, that

M R Rf is free.
Finally, we would like to give an example showing that Corollary 4 does

not extend to nite free modules in place of locally free ones of nite rank.
So we will construct a module M over a ring R where M is locally free of
nite rank, but not free, although it becomes globally free after performing a
suitable faithfully at coecient extension R - R . To do this we choose
R as a Noetherian normal integral domain of Krull dimension 1; such rings
are called Dedekind domains. Any principal ideal domain is Dedekind. But the
most prominent examples of Dedekind domains arise from Number Theory.
For a nite algebraic extension K/Q let R be the integral closure of Z in K.
Then R is normal, as we can see from 3.1/7, and of Krull dimension 1 by 3.3/6.
Furthermore, one knows that R is nite over Z and, hence, Noetherian; see
Exercise 3.1/8 or use AtiyahMacdonald [2], 5.17. In particular, such a ring R
will be a Dedekind domain.
There are Dedekind domains admitting ideals that are not principal. For
example, we can take R = Z 5. By elementary
computation one can see
that this is the integral closure of Z in Q( 5) and therefore a Dedekind do-
main; cf. Exercise 3.1/6. But it cannot be factorial and, hence, is not a principal
ideal domain, since there are the following essentially dierent decompositions
into irreducible factors:

6 = 2 3 = (1 + 5) (1 5)

Alternatively, one can verify directly that the ideal a = (2, 1 + 5) R is not
principal.
Let us x a Dedekind ring R containing some ideal a that is not principal.
Then, for any maximal ideal m R, the tensor product aR Rm denes an ideal
in the localization Rm of R, since the localization map R - Rm is at by 4.3/3.
Furthermore, Rm is normal by 3.1/11. Therefore, using the characterization of
normal Noetherian local integral domains to be proved later in 9.3/4, we see
that each localization Rm is a local principal ideal domain. In particular, the
ideal a R Rm Rm is principal and, thus, a free Rm -module of rank 1.
Since R is Noetherian, a is an R-module of nite presentation, and we can
conclude from Proposition 3 that a is a locally free submodule of nite rank in
R. Hence, there are elements f1 , . . . , fr R generating the unit ideal of R such

that aR Rfi is a free Rfi -submodule in Rfi for all i, necessarily a principal ideal
of Rfi . Now, as exercised in the proof of Proposition 3, we use the faithfully at
morphism R - R = Rf . . . Rf to extend the coecients of a. Then
1 r
we can observe that a R R is free, namely a principal ideal in R , although a
itself cannot be free, since it is not principal.
Exercises
1. Let M, N be R-modules where N is faithfully at. Show that M is at (resp.

faithfully at) if and only if the same is true for M R N .
2. For a ring morphism R - R consider a faithfully at R -module M as well

on R, by viewing M as an R-module via R
as its restriction M/R - R . Show
that M/R is a at (resp. faithfully at) R-module if and only if R - R is at

(resp. faithfully at).
3. Consider a ring morphism R - R , an R -module M , as well as its restriction

on R, and assume that the latter is a faithfully at R-module. Show:
M/R
(a) For any R-module M , the canonical morphism of R-modules M - M R R ,
x - x 1, is injective.
(b) Any ideal a R satises aR R = a.
(c) If R is Noetherian (resp. Artinian), the same is true for R.
Hint: Consider the canonical morphism of R-modules M/R - M R R and
/R

its retraction M/R R R - M given by x a
- ax. The composition of
/R
and, thus, leads to a split short exact sequence.
these is the identity on M/R
4. Let M be a at R-module of nite type. Show that all localizations M R Rm

at maximal ideals m R are nite free. Hint: Reduce to the case where R is a
local ring with maximal ideal m. Then proceed as in the proof of Proposition 3 by
considering an exact sequence of type 0 - K - Rn - M - 0 where
K/mK = 0. For any x K let ax R be the ideal generated by all coordinates of
x as a point in Rn . Show max = ax and conclude ax = 0 by Nakayamas Lemma.
5. For a ring R let L(R) denote the set of isomorphism classes of locally free
R-modules of nite rank.
(a) Show that L(R) becomes a monoid, when the multiplication is dened by the
tensor product. For example, L(R) = (N, ) if R is a principal ideal domain or a
local ring.
(b) Let Pic(R) L(R) be the group of invertible elements; it is called the Picard
group of R. Show that Pic(R) consists of all locally free R-modules of rank 1, i.e.
of all M L(R) such that Mm Rm for all maximal ideals m R.
(c) Show that the inverse of any M Pic(R) is given by HomR (M, R).
Hint: For (c) use Exercise 4.3/9.
4.5 Categories and Functors
The language of categories and their functors is an essential tool in advanced

Algebraic Geometry. Implicitly this concept has already appeared in earlier
sections, mostly in the form of universal properties. But we need to make more
intensive use of it, especially for the descent of modules in Section 4.6. Since
pure category theory is not very enlightening by itself, we have chosen to include
only basic material at this place.
Denition 1. A category C consists of a collection 2 Ob(C) of so-called objects

and, for each pair of objects X, Y Ob(C), of a set Hom(X, Y ) of so-called
morphisms (or arrows), together with a law of composition
Hom(Y, Z) Hom(X, Y ) - Hom(X, Z), (g, f ) - g f,
for any objectsX, Y, Z Ob(C). The following conditions are required :

(i) The composition of morphisms is associative.
(ii) For all X Ob(C) there is a morphism idX Hom(X, X) such that
idX f = f for all f Hom(Y, X) and f idX = f for all f Hom(X, Y ).
Note that idX is unique, it is called the identity morphism on X.
Sometimes we write HomC (X, Y ) instead of Hom(X, Y ), in order to specify

the category C whose morphisms are to be considered. In most cases, morphisms
between objects X, Y are indicated by arrows X - Y , thereby appealing to
the concept of a (set theoretical) map. However, since only the above conditions
(i) and (ii) are required, morphisms can be much more general than just maps.
A morphism f : X - Y between two objects of C is called an isomorphism
if there is a morphism g : Y - X such that g f = idX and f g = idY . Let
us consider some examples.
(1) All sets together with maps in the usual sense as morphisms form a
category, denoted by Set.
(2) All groups together with group homomorphisms form a category, de-
noted by Grp. Likewise, all rings together with ring homomorphisms form a
category, denoted by Ring.
(3) For a ring R, all R-modules together with R-module homomorphisms
form a category, denoted by R-Mod. Given two objects M , N in this category,
the set of morphisms M - N is denoted by HomR (M, N ), as before. This
set can be canonically interpreted as an R-module again, using the R-module
structure of N .
(4) If C is a category, we can consider the associated dual category C0 where
Ob(C0 ) = Ob(C) and HomC0 (X, Y ) = HomC (Y, X) for objects X, Y . The com-
position of two morphisms X - Y and Y - Z in C0 is done by composing
2
Note that a collection, like a set, groups together certain mathematical objects. However,
it is not required that a collection respects the axioms of a set, as considered in set theory.
4.5 Categories and Functors 139
the underlying morphisms Z - Y and Y - X in C. Thus, C0 is constructed

from C, as we say, by inverting directions of arrows. Of course, dualizing C0 yields
C again. The dual category C0 is also referred to as the opposite category of C
and denoted by Cop .
(5) Let C be a category and S Ob(C) an object. Then the category CS of
relative objects over S, also called S-objects, is dened as follows. Ob(CS ) is the
collection of all morphisms of type X - S where X varies over the objects in
C. For two S-objects X - S and Y - S, let Hom(X - S, Y - S)
be the set of all morphisms X - Y in HomC (X, Y ) such that the diagram
X - Y

S
is commutative. Such morphisms are generally referred to as S-morphisms in
C. Usually, S-objects are denoted by symbols like X or Y again, and the set of
S-morphisms between them by HomS (X, Y ).
(6) Switching to the dual setting and xing an object R of a category C, we
can dene the category CR of all objects under R. Then Ob(CR ) is the collection
of all morphisms of type R - X where X varies over the objects of C. For
two such objects R - X and R - Y , let Hom(R - X, R - Y ) be
the set of all morphisms X - Y in HomC (X, Y ) such that the diagram
X - Y

R
is commutative. Such morphisms are referred to as morphisms under R or also
as R-morphisms in C. A typical example of such a category is the category
R-Alg of R-algebras under a xed ring R.
Denition 2. Let C be a category and S Ob(C). Let X, Y Ob(CS ). An ob-

ject W CS , together with two S-morphisms p1 : W - X and p2 : W -Y
(referred to as projections onto the factors X, Y ) is called a ber product of X
and Y over S if the following universal property holds:
Given two S-morphisms T - X and T - Y , there is a unique
S-morphism T - W such that the diagram
X
-
6
p1
T - W
p2
?
-
Y
is commutative.
If it exists, the ber product W is unique up to canonical isomorphism and
will be denoted by X S Y .
In particular, we can say that an S-object W together with two S-morphisms

p1 : W - X and p2 : W - Y is a ber product of X and Y over S if the
map
HomS (T, W ) - HomS (T, X) HomS (T, Y ),

h - (p1 h, p2 h),
is bijective for all S-objects T . Suppressing S and replacing CS by any category

C, the latter condition is still meaningful and denes the cartesian product of X
and Y in C; see also 7.2/1. Thus, we can say that the ber products over some
object S of C, as introduced in Denition 2, are just the cartesian products in
CS .
As an example, let us consider the category C = Set of all sets with maps
between sets as morphisms. For a one-point set S, the ber product X S Y of
two sets X, Y is given by the cartesian product

X S Y = X Y = (x, y) ; x X, y Y ,
together with the canonical projections X Y - X and X Y - Y.

More generally, for an arbitrary set S, the ber product of two relative objects
X : X - S and Y : Y - S consists of the subset

X S Y = (x, y) X Y ; X (x) = Y (y) X Y,
together with canonical projections to X and Y where, on the left-hand side, X

and Y have to be interpreted as the relative objects X and Y . Thus, X S Y
is, indeed, a bered product, since, berwise over S, it is just the family of all
cartesian products of type X(s) Y (s) where s varies over S and X(s), Y (s)
are the bers of X, Y over s, i.e. the preimages of s with respect to the maps
X and Y .
Switching to the dual notion of ber products, we arrive at so-called amal-
gamated sums.
Denition 3. Let C be a category and R Ob(C). Let X, Y Ob(CR ). An ob-

ject W CR , together with two R-morphisms 1 : X - W and 2 : Y -W
is called an amalgamated sum of X and Y under R if the following universal
property holds:
Given two R-morphisms X - T and Y - T , there is a unique
R-morphism W - T such that the diagram
X
1
?
-
W - T
-
6
2
Y
is commutative.
4.5 Categories and Functors 141
If it exists, the amalgamated sum W is unique up to canonical isomorphism

and will be denoted by X R Y .
The tensor product A1 R A2 of two R-algebras R - A1 and R - A2

over a ring R, together with the canonical R-algebra homomorphisms
A1 - A1 R A2 , a - a 1,
A2 - A1 R A2 , b - 1 b,
is a typical example of an amalgamated sum; see 4.3/6 for the universal prop-
erty. To illustrate the interplay between amalgamated sums and ber products,
let us consider the category C = Ring of all rings and its dual category C0 . The
latter plays a central role in Algebraic Geometry. Namely, for any ring
R Ob(C), one can consider its associated ane scheme, which is a so-called
ringed space. By abuse of notation, this ringed space is denoted by Spec R and
consists indeed, as main ingredient, of the set of all prime ideals in R, viewed
as a topological space under the Zariski topology. But, in addition, it incorpo-
rates all localizations Rf of R by elements f R, where the elements of Rf
are interpreted as functions on the open set D(f ) = {x Spec R ; f (x) = 0}.
It is crucial that the ring R we started with can be reconstructed from the
ringed space Spec R and that morphisms of ane schemes Spec R - Spec R
correspond bijectively to ring homomorphisms R - R. In other words, the
category of ane schemes, together with their morphisms, can be interpreted
as the dual C0 of the category C of rings. For details see Chapter 6 and, in par-
ticular, 6.6/12. The existence of tensor products in C implies then the existence
of ber products in C0 , in the sense that for R-algebras A1 and A2 we have
Spec A1 Spec R Spec A2 = Spec(A1 R A2 ).
Next let us address the subject of functors between categories, which play
the role of maps within this context.
Denition 4. Let C, D be categories. A covariant functor F : C - D is a

rule that assigns to each object X Ob(C) an object F (X) Ob(D) and to
each morphism X - Y in C a morphism F (X) - F (Y ) in D such that
the following conditions hold :
(i) F (idX ) = idF (X) for all X Ob(C).
(ii) F (g f ) = F (g) F (f ) for composable morphisms g and f in C.
A contravariant functor C - D is a covariant functor C0 - D. In
particular, passing from C to D such a functor inverts directions of arrows.
For example, xing a ring homomorphism R - R , the assignments
M - M R R ,
(f : M - N) - (f idR : M R R - N R R ),
dene a covariant functor R-Mod - R -Mod.

If F : C - D and G : C - D are two (covariant) functors between

categories C and D, we dene a functorial morphism : F - G (also known
as a natural transformation between F and G) as a collection of morphisms
X : F (X) - G(X) in D, where X varies in Ob(C), such that for any mor-
phism f : X - Y in C the diagram
X
F (X) - G(X)
F (f ) G(f )
? ?
Y
F (Y ) - G(Y )
is commutative. Two functors F, G : C -
D are called equivalent or isomor-
phic if there exists a functorial isomorphism : F - G; by the latter we mean
a functorial morphism such that X is an isomorphism for all X Ob(C). Fi-
nally, two categories C, D are called equivalent if there are functors F : C - D
and G : D - C such that G F is equivalent to the identity on C and F G
equivalent to the identity on D.
Exercises
1. Let F : C1 - C2 and G : C2 - C3 be functors between categories. Show that
G F : C1 - C3 is again a functor. Taking into account that there exists the
identity functor idC : C - C on any category C, can we talk about the category
of all categories?
2. For two categories C1 , C2 construct (up to set-theoretical diculties) the cat-
egory Hom(C1 , C2 ) of all (covariant) functors C1 - C2 , where a morphism
: F - G between two such functors is understood as a functorial morphism.
In particular, show for functors F, G, H : C1 - C2 and functorial morphisms
: F - G and : G - H that the formula ( )X = X X for
X Ob(C1 ) can be used to dene the composition of and . Furthermore, note
that there is the identity functorial morphism idF on F given by (idF )X = idF (X)
for any object X in C1 .
As an example, x a group G and write G for the one-point category with mor-
phisms (including their compositions) given by the group G, as well as Ab for the
category of abelian groups. Show that Hom(G, Ab) is the category of G-modules,
i.e. of Z-modules that are equipped with a G-action.
3. Construct amalgamated sums in the categories of sets Set and abelian groups
Ab.
4. Consider a functor F : R-Mod - R-Mod on the category of modules over a
xed ring R by setting F (M ) = HomR (HomR (M, R), R) for objects M of R-Mod
and by dening F in the obvious way on morphisms of R-Mod. Construct a
functorial morphism idR-Mod - F between the identity functor on R-Mod
and the functor F , which is an isomorphism when restricted to the subcategory
of nite free R-modules in R-Mod.
5. Dene the cartesian product of two categories. Show for rings R, R that there is
an equivalence of categories R-Mod R -Mod - (R R )-Mod.
4.6 Faithfully Flat Descent of Modules and their Morphisms 143
6. Given a category C we conclude from Exercise 2 that the functorial morphisms

idC - idC on the identity functor of C form a monoid. Compute this monoid
for the categories Set, Ring, and R-Mod over a given ring R.
4.6 Faithfully Flat Descent of Modules and their Mor-

phisms
Let R and R be two rings and p : R - R a ring homomorphism. The some-

what exotic notation p for a ring homomorphism has been chosen with care;
it is motivated by the algebraic-geometric background of ane schemes. As we
have already pointed out in Section 4.5, the category of ane schemes can be
interpreted as the dual of the category of rings and vice versa. This way, p
corresponds to a well-dened morphism p : Spec R - Spec R between asso-
ciated ane schemes where, as a map of sets on the underlying prime spectra,
p is given by
Spec R p - (p )1 (p ) Spec R.
In fact, our notation suggests to take the morphism p as point of departure and
to interpret p as the associated dual. As we will see in Chapter 6, the ring
homomorphism p admits a natural interpretation as a so-called pull-back of
functions on Spec R to functions on Spec R , just by composition with p. In the
same spirit we dene for R-modules M , hence, objects living on the level of
Spec R, their pull-backs under p by
p M = M R R
and, likewise, for morphisms of R-modules : M - N , their pull-back under

p by
p = idR : M R R - N R R .
The notations p M and p have the advantage that they spell out in an explicit
way, how to view R as an R-module in the occurring tensor products. For
questions of descent, this is of essential importance, as we will see below. Also
note that if p : R - R is decomposable into the product r s of two ring
homomorphisms r and s , there is a canonical isomorphism p M r (s M )
for R-modules M by 4.3/2. In the following such isomorphisms will occur as
identications, just writing p M = r (s M ). In the same way we will proceed
with pull-backs of morphisms.
Now let us start descent theory by discussing the descent of module mor-
phisms.
Proposition 1. For a faithfully at ring homomorphism p : R - R , con-

sider the associated homomorphisms
p1 : R - R R R , a - a 1,
p2 : R - R R R , a - 1 a ,
as well as the composition
q = p1 p = p2 p : R - R R R
and let R = R R R . Then, for any R-modules M, N , the diagram

p p1
-
HomR (M, N ) - HomR (p M, p N ) - HomR (q M, q N )
p2
is exact.
Recall that a diagram of maps between sets

p1
p
- -
-
A B C
p2
is called exact if p induces a bijection
A - ker(p1 , p2 ) := {b B ; p1 (b) = p2 (b)}.
In the case of homomorphisms between abelian groups this is equivalent to the

exactness (in the usual sense) of the sequence
p p1 p2
0 - A - B - C
where ker(p1 p2 ) = ker(p1 , p2 ). In particular, these sequences are exact if and

only if p is injective and im p equals ker(p1 , p2 ).
To a substantial extent, the proof of Proposition 1 will be based on the
following fact:
Lemma 2. In the setting of Proposition 1 the diagram

p p1
-
R - R - R
p2
is exact. It remains exact when tensoring over R with an arbitrary R-module

M.
Proof of Lemma 2. To check the exactness of the above diagram, we can use
4.2/11 and thereby tensor it with a faithfully at R-algebra R - A. Then
the exactness of the diagram
-
A - R R A - R R R R A,
or the corresponding one obtained by tensoring with M over R, has to be

checked. Let A = R R A and set
A = A A A (R R A) A (R R A) R R R R A
where we have used 4.3/2. It follows that the preceding diagram is of the same
type as the one mentioned in the assertion, namely
-
A - A - A ,
just replacing R - R by A - A . Since tensoring with M over R can also

be interpreted as tensoring with M R A over A, it becomes clear that, for the
proof of the lemma, we may replace R without loss of generality by a faithfully
at R-algebra A.
Doing so, apply the preceding argument to the R-algebra R - A given

by p : R -
R . Thereby we are reduced to the diagram
p2 -
R - R R R - ...
But now p2 : R - R R R admits a so-called retraction. Thereby we mean

a homomorphism : R R R - R such that p = idR ; just consider
2
the multiplication map
: R R R - R , a b - a b .
Thus, replacing R by a faithfully at R-algebra, in our case A = R , we may

assume for the diagram
p p1
-
R - R - R
p2
that p admits a retraction . But then p1 and p2 admit retractions as well,

namely the homomorphisms idR and idR , and these retractions are
maintained when tensoring with M over R. Now, to justify the exactness of the
diagram
p p1
-
M - M R R - M R R ,
p2
observe rst that p : M - M R R is injective, since this map admits a

retraction. In addition, we have
im p ker(p1 , p2 ),
due to the relation

p1 p = p2 p . To show the opposite inclusion, choose some
element z = ni=1 xi ai ker(p1 , p2 ), where xi M and ai R . Then the
images
n
n
p1 (z) = xi ai 1, p2 (z) = xi 1 ai
i=1 i=1
coincide, and application of the retraction idM idR of p1 yields

n
n
z= xi 1 (ai ) = p xi (ai ) im p
i=1 i=1
so that im p = ker(p1 , p2 ).
Now we are able to carry out the proof of Proposition 1. To do this we

choose R-modules M, N and look at a commutative diagram of type
- p M -
M -q M
? ? ??
- p N -
N -q N
where the rows are exact by Lemma 2 and where the vertical maps have still
to be specied. We start with a morphism of R-modules : M - N and
consider the morphism of R -modules p : p M - p N obtained from by
tensoring with R . It makes the left-hand square commutative, and we see from
the injectivity of N - p N that p : HomR (M, N ) - HomR (p M, p N )

is injective. Furthermore, the relation p1 p = p2 p implies that the image
im p is contained in the kernel of
p1
-
HomR (p M, p N ) - HomR (q M, q N ).
p2
To show the opposite inclusion, let : p M - p N be a morphism of

R -modules and consider p1 , p2 as vertical maps on the right-hand side of
the above diagram. Then the right-hand square will be commutative if, from
the double arrows, we select just the ones pertaining to p1 or, alternatively,
to p2 . Now if ker(p1 , p2 ), the vertical maps p1 , p2 coincide, and a di-
-
agram chase shows that maps the kernel of p M - q M into the kernel
-
- q N . However, due to the exactness of the rows, this means that
of p N
restricts to a morphism of R-modules : M - N . Since is uniquely

determined by the values it takes on M , we get p = , which establishes the
remaining inclusion.
Next we want to discuss the descent of modules. To do this, x a faithfully

at ring homomorphism p : R - R and look at the functor
p : R-Mod - R -Mod
from the category of R-modules to the category of R -modules, which assigns

to an R-module M the R -module p M = M R R , and to a morphism of
R-modules : M - N the morphism of R -modules p = idR . We
would like to describe the image of this functor. In Proposition 1 we have
already solved this problem for morphisms, but a similar reasoning has still
to be carried out on the level of modules as the objects of our categories. We
will prove the striking fact that p induces an equivalence between the category
of R-modules and the category of R -modules, provided we enrich the latter
modules by some additional structure, so-called descent data.
To dene descent data we consider, in addition to p : R - R and the
homomorphisms
p1 : R - R R R , a - a 1,
p2 : R - R R R , a - 1 a ,
already used above, the homomorphisms

p23 : R R R - R R R R R , a b - 1 a b ,
p13 : R R R - R R R R R , a b - a 1 b ,
p12 : R R R - R R R R R , a b - a b 1,
where, switching to the dual category of ane schemes, the latter maps can be
interpreted as the canonical projections
pi : Spec R Spec R Spec R - Spec R ,

pij : Spec R Spec R Spec R Spec R Spec R - Spec R Spec R Spec R
onto the factors, as indicated by the indices.
Denition 3. As before, let p : R - R be a faithfully at ring homomor-

phism and x an R -module M . A descent datum on M with respect to p is

an isomorphism of R -modules
: p1 M - p M
2
such that the diagram

p12 p23
p12 p1 M - p
12 p2 M

===== p23 p1 M - p
23 p

M

2

p13
p13 p1 M - p
13 p2 M

is commutative (cocycle condition).
Observing the relations
p1 p12 = p1 p13 , resp. p12 p1 = p13 p1 ,

p2 p12 = p1 p23 , resp. p12 p2 = p23 p1 ,
p2 p23 = p2 p13 , resp. p23 p2 = p13 p2 ,
the cocycle condition requires that the dierent possible pull-backs of a descent
datum to the level of R = R R R R R , namely by coecient extension
via the pij , are compatible. To give a trivial example, start with an R-module
M and set M = p M . Then the canonical isomorphism
p1 M = p1 p M - (p p1 ) M = (p p2 ) M p2 p M = p2 M
yields a descent datum on p M , the so-called trivial descent datum. In this case,
all isomorphisms entering into the cocycle condition are canonical so that the
required commutativity holds for trivial reasons.
Denition 4. Let M , N be R -modules with descent data : p1 M - p2 M

and : p1 N - p2 N . A morphism of R -modules f : M - N is called
compatible with the descent data and if the diagram

p1 M - p M
2
p1 f p2 f
? ?

p1 N - p N
2
is commutative.
For us it is of interest that the pairs of type (M , ), consisting of an

R -module M and a descent datum with respect to p on M , form a category

R -Mod-DD if we admit as morphisms only those R -module homomorphisms

that are compatible with the involved descent data.
Theorem 5 (Grothendieck). Let p : R - R be a faithfully at ring homo-

morphism. Then the functor
: R-Mod - R -Mod-DD,
M - p M,
denes an equivalence of categories.
Note that we have already shown in Proposition 1 that the functor is

faithful and, in fact, even fully faithful, which means that for all R-modules M
and N the map
HomR-Mod (M, N ) - HomR -Mod-DD (p M, p N )
given by is injective, and even bijective. To prove that denes an equivalence

of categories, we have to exhibit a functor : R -Mod-DD - R-Mod such
that is equivalent to the identity on R-Mod and is equivalent to the
identity on R -Mod-DD. To do this it is enough to show that is essentially
surjective in the sense that for every object M in R -Mod-DD there is an
object M in R-Mod such that M is isomorphic to (M ). Indeed, choosing for
any object M in R -Mod-DD an R-module M such that M (M ), we see
from the fully faithfulness of that M - M yields a well-dened functor
: R -Mod-DD - R-Mod which, together with , furnishes an equivalence
between the two categories.
Before we actually start the proof of Theorem 5 and show that is es-
sentially surjective, let us develop an alternative description of descent data
that will be quite handy for our purposes. Consider an R -module M and an
isomorphism of R -modules
: p1 M - p M .
2
Then, for any morphisms of ane R-schemes ti : T - Spec R , i = 1, 2,

or dually speaking, for any R-algebra homomorphisms ti : R - S, where
T = Spec S, we obtain via the universal property of ber products
Spec R
- 6
t1
p1
(t1 ,t2 )
T - Spec R
p2
t2
-
?
Spec R
a well-dened morphism (t1 , t2 ) : T - Spec R , which corresponds to the
product map
R - S, a b - t a t b .
1 2
Now, pulling back the isomorphism : p1 M - p2 M via (t1 , t2 ) yields an

isomorphism of S-modules t1 ,t2 that makes the diagram
(t1 ,t2 )
(t1 , t2 ) p1 M - (t1 , t2 ) p2 M

t1 ,t2
t1 M - t M
2
commutative. Likewise, for any three morphisms
ti : T - Spec R , i = 1, 2, 3,
we can consider the canonical morphism
(t1 , t2 , t3 ) : T - Spec R ,
as well as associated diagrams of type

(t1 ,t2 ,t3 ) p12
(t1 , t2 , t3 ) p12 p1 M - (t1 , t2 , t3 ) p12 p2 M

t1 ,t2
t1 M - t M
2 ,
thereby arriving at the following fact:
Lemma 6. An isomorphism of R -modules : p1 M - p M is a descent

2
datum if and only if we have
t1 ,t3 = t2 ,t3 t1 ,t2
for any three morphisms ti : T - Spec R , i = 1, 2, 3, where T varies over all

ane R-schemes, i.e. over the spectra of all R-algebras.
Proof. The relation t1 ,t3 = t2 ,t3 t1 ,t2 is immediately clear from the above
type of diagrams if satises the cocycle condition. To show the converse, apply
the relation mentioned in the lemma to T = Spec R and the possible three
projections ti : T - Spec R .
Now let us begin the proof of Theorem 5 by looking at a special case.
Lemma 7. The assertion of Theorem 5 holds if p : R - R admits a retrac-

tion : R - R (in the sense that p = idR ).
Proof. We consider for T = Spec R the morphisms t, t : T - Spec R given by

p
the identity map id : R - R and the composition R - R - R , and let
M = M where M is an R -module with descent datum : p1 M - p2 M .
Then
f = t,t : M - p M = p M
is an isomorphism of R -modules, and we claim that f is compatible with the
given descent datum on M and the trivial descent datum on p M . This
will show that the functor of Theorem 5 is essentially surjective and thereby
establish the desired equivalence of categories, as we have explained above.
Therefore we have to show that the diagram

p1 M - p M
2
p1 f p2 f
? ?
p1 p M ===== p2 p M
is commutative. To achieve this, observe that the cocycle condition of in the
version of Lemma 6, applied to the morphisms
p1 , p2 , t3 = p p1 = p p2 : Spec R - Spec R ,
yields
p1 ,t3 = p2 ,t3 p1 ,p2 ,
where p1 ,p2 = , since (p1 , p2 ) : Spec R - Spec R is the identity map.
Furthermore, writing (pi , t3 ) for i = 1, 2 as a composition
(pi ,t3 )
Spec R - Spec R Spec R Spec R
-
pi
- (t,t)=(id,p)
Spec R
shows
pi f = pi t,t = pi ,t3 .
Therefore the above cocycle relation reads
p1 f = p2 f
and we see that, indeed, f is compatible with descent data as claimed.
To settle the general case of Theorem 5 by showing that the functor is

essentially surjective, we need a concretion of descent data in terms of so-called
cocartesian diagrams. Again we consider an R -module M and an isomorphism
of R -modules
: p1 M - p2 M ,
in detail
: M R R - R R M ,
where M R R and R R M are viewed as R R R -modules, respecting
positions of factors. Then the canonical maps 1 : M - M R R and
2 : M -
R R M yield a diagram
M R R
1 -
M
2
?
-
R R M
where there is no reason for the latter to be commutative. Since tensor prod-
ucts are signicant only up to canonical isomorphism, we can view as an
identication, thereby arriving at the diagram
1
-
M - M R R ,
1 2
which is cocartesian over

p1
-
R - R R R .
p2
The latter property is indicated by the diagram

-
M - M R R
p1
-
R - R R R
p2
and means the following. A diagram of type

M - M
R - R
with an R -module M and an R -module M is called cocartesian if the
R -bilinear map
R M - M , (a, x) - a (x),
satises the universal property of the tensor product R R M . For a diagram
with multiple horizontal maps, as in our case, cocartesian means that this prop-
erty holds for all arrows at the same position upstairs and downstairs. Thus, in
our case, cocartesian amounts to the condition that M R R is an R -module,
-
which can be interpreted through both maps M - M R R as a coe-
cient extension from M , namely with respect to the maps p1 : R
- R and

p2 : R -
R . Also note that if we consider the cocartesian diagram
-
M - M
p1
-
R - R
p2
associated to an isomorphism of R -modules : M R R - R R M , this

isomorphism can be recovered from the diagram. It corresponds to the identity
map on M and is given by the composition of canonical maps
: M R R - M R R M .
Applying the formalism of cocartesian diagrams to descent data and using

identications of tensor products as discussed above, we arrive at the following
characterization:
Lemma 8. Descent data on M correspond bijectively to cocartesian diagrams

- -
M - M R R -
- M R R R R
p23
-
p1 p13
-
R - R R R - R R R R R
p2 p12
-
where the morphisms of the rst row satisfy relations just as the ones in the
second row, namely
p1 p12 = p1 p13 , resp. p12 p1 = p13 p1 ,

p2 p12 = p1 p23 , resp. p12 p2 = p23 p1 ,
p2 p23 = p2 p13 , resp. p23 p2 = p13 p2 .
For an R-module M , such a descent datum is trivial on M = M R R if and

only if the above diagram can be enlarged by adding a cocartesian diagram
M - M
p
R - R
such that the morphisms in the rst row satisfy relations just as the ones in the
second row, namely p p1 = p p2 and p1 p = p2 p .
Proof. Consider a descent datum : p1 M - p2 M . As explained above, we

can identify p1 M with p2 M via and thereby obtain a cocartesian diagram
-
M - M R R
p1
-
R - R R R .
p2
By a similar identication process, we construct a cocartesian diagram

-
M R R -
- M R R R R
p23
-
p13
R R R - R R R R R .
p12
-
Namely, looking at the cocycle condition satised by , we observe the canonical

identications
() p12 p1 M = p13 p1 M , p12 p2 M = p23 p1 M , p23 p2 M = p13 p2 M
and use the commutativity of the diagram in Denition 3 in order to identify

these (R R R R R )-modules via the isomorphisms p12 , p23 , and p13 .
Thereby we obtain a cocartesian diagram over
-
R R R -
- R R R R R ,
which together with the previous one yields a cocartesian diagram as stated in
the assertion. The additional relations are clear from the construction.
Conversely, look at a cocartesian diagram as specied in the assertion. Then
the general explanations above show that the left part of the diagram determines
an isomorphism of (R R R )-modules : p1 M - p2 M , whereas the right
part together with the required relations implies that satises the cocycle
condition and therefore is a descent datum. Indeed, one can deduce from the
given relations that the canonical isomorphisms () are compatible with the
isomorphisms
p23 (M R R ) p13 (M R R ) p12 (M R R )
derived from the right part of the diagram. From this one easily deduces a
commutative diagram as required for the cocycle condition in Denition 3.
Finally, for an R-module M , a descent datum : p1 M - p2 M on
M = p M is trivial if and only if the canonical diagram

p1 p M
-
M
-
p2 p M
is commutative. The latter amounts to the fact that in the cocartesian diagram
-
M - M - M R R
-
R - R - R R R
both compositions of the rst row coincide and, thus, satisfy the relation spec-
ied in the assertion.
Now we have gathered all auxiliary means in order to nish the proof of
Theorem 5. Considering a descent datum on M , given by a cocartesian diagram
- -
M - M R R -
- M R R R R
- -
R - R R R -
- R R R R R ,
we know from Lemma 2 that the diagram

p p1
-
R - R - R R R ,
p2
remains exact, when we tensor it over R with some R-module M . Thus, if the
above descent datum is induced from an R-module M , then M is necessarily
-
isomorphic to the kernel K = ker(M - M R R ). On the other hand, it is
enough for the proof of Theorem 5 to show that the diagram
-
K - M - M R R
-
R - R - R R R
is cocartesian. To check the latter condition, we can use 4.2/11 and thereby
are allowed to apply a faithfully at coecient extension to the diagram. For
example, taking p : R - R as a coecient extension, we are reduced to the
case where p : R - R admits a retraction. But then, by Lemma 7, there
exists an R-module M that we might insert instead of K in the above diagram
such that the resulting diagram
-
M - M - M R R
-
R - R - R R R
is cocartesian. It follows that M is canonically identied with K and that the
diagram involving K is cocartesian.
As a corollary of Theorem 5, one can work out the technique of faithfully

at descent for algebras as well. One only has to require that the descent data
under consideration are compatible with the algebra structure; see Exercise 3
below. Also let us mention that the classical form of Galois descent can be
viewed as a special case of the faithfully at descent for modules considered
here; see Exercise 5 below.
Exercises
1. Let R - R be an extension of degree > 1 of elds. Show that the functor
R-Mod - R -Mod, M - M R R , is faithful, essentially surjective, but
not fully faithful.
2. Let p : R - R be a faithfully at morphism of rings. Assume there exists
a non-free R-module M such that M = M R R is nite free, say with free
generators x1 , . . . , xn . Show that the morphism of R -modules
: p1 M = M R R - R R M = p M ,
2
obtained by R -linear extension from xi 1 - 1 xi for i = 1, . . . , n, de-

nes a descent datum on M and that (M , ) as object in R -Mod-DD is not

isomorphic to the canonical object p M .

3. Generalize descent theory from modules to algebras. In more precise terms, let
p : R - R be a faithfully at morphism of rings and consider the category
R-Alg of R-algebras, as well as the category R -Alg-DD of R -algebras with
descent data. Adapting the notation of Denition 3, a descent datum on an
R -algebra A is an isomorphism of R -algebras : p1 A - p2 A satisfying the
cocycle condition. Show that the functor A - p A gives rise to an equivalence
of categories R-Alg - R -Alg-DD. Hint: Do not rebuild the whole descent
theory. Use Theorem 5 in conjunction with Proposition 1 instead.
4. Gluing as an example of descent:Consider elements f1 , . . . , fn of a ring R that
generate the unit ideal. Set R = ni=1 Rfi and show:
(a) The canonical morphism R - R is faithfully at.
(b)
n The category R -Mod of R -modules is equivalent to the cartesian product
i=1 Rfi -Mod; see also Exercise4.5/5. Roughly speaking this means that any
R -module is a cartesian product ni=1 Mi with Mi being an Rfi -module for each i
and, likewise, that morphisms of R -modules are cartesian products of morphisms
of Rfi -modules.
(c) The category R-Mod is equivalent to the category R -Mod-G dened as
follows. Its objects are families (Mi )i=1,...,n where each Mi is an Rfi -module,
together with so-called gluing data (see also 7.1/1), by which we mean isomor-
phisms ij : Mi R Rfj - Mj R Rfi of Rfi fj -modules satisfying the cocycle
condition
ik R Rfj = (jk R Rfi ) (ij R Rfk )
for all indices i, j, k; view the occurring maps as morphisms of Rfi fj fk -modules.
Furthermore, a morphism (Mi ) - (Ni ) between two objects of R -Mod-G
with gluing data (ij ) and (ij ) consists of a family of Rfi -module morphisms
i : Mi - Ni , i = 1, . . . , n, that is compatible with the gluing data in the sense
that for all indices i, j the diagram
ij
Mi R Rfj - Mj R Rfi
i Rfj j Rfi
? ij ?
Ni R Rfj - Nj R R f
i
is commutative.
5. Galois descent: Let R - R be a nite Galois extension of elds, write for
the attached Galois group, and x an R -vector space M . Consider actions of
on M such that ( x) = () (x) for elements , coecients R ,
and vectors x M . Show:
(a) Actions of this type correspond bijectively to descent data on M with respect
to R - R .
(b) The xed part M = {x M ; (x) = x for all } of M is an R-vector
space such that the canonical R -linear map M R R - M is an isomorphism.
Hint: If necessary, consult [5], Example 6.2/B, and dualize the arguments.
5. Homological Methods: Ext and Tor
Consider a short exact sequence

() 0 - M - M - M - 0
of modules over some ring R and x an additional R-module E. Then it fol-

lows from the right exactness of tensor products 4.2/1 that the corresponding
sequence
idE idE
0 - M R E - M R E - M R E - 0
obtained by tensoring with E over R is right exact, but maybe not exact in
terms of a short exact sequence. Indeed, unless E is at over R (see 4.2/3), the
map idE might have a non-trivial kernel. For example, let us look at an
exact sequence of type
a
() 0 - R - R - R/(a) - 0
where a indicates multiplication by some element a R that is not a zero

divisor in R. Tensoring with E over R and identifying R R E with E produces
the right exact sequence
a
0 - E - E - E/aE - 0
where, similarly as before, the map a is multiplication by a on E. In par-

ticular, we get
a
ker(E - E) = {x E ; ax = 0},
the latter being called the submodule of a-torsion in E. Clearly, this torsion
module can be non-trivial, as the example E = R/(a) shows.
Returning to the general case of a short exact sequence () as above, one
might ask if there is a natural way to characterize the kernel of the morphism
idE : M R E - M R E.
Of course, keeping E xed for a moment, ker( idE ) will be determined by

and thereby depends on both, M and M . This dependence is most naturally
claried by the so-called long exact Tor sequence 5.2/2. Indeed, the striking
fact we will discuss in 5.1/10 and 5.2/1 is that there exist so-called left derived

158 5. Homological Methods: Ext and Tor
functors TorRi (, E) of the tensor product R E. These are functors on the

category of R-modules such that any short exact sequence () will canonically
lead to an innite exact sequence of R-modules
... - TorR
2 (M , E)
- TorR - TorR - TorR

1 (M , E) 1 (M, E) 1 (M , E)
- M R E - M R E - M R E - 0.
The latter sequence behaves functorially in any reasonable way, namely for exact
sequences of type () on the level of the rst variable, but also for morphisms
on the level of the second variable. In particular, there is an isomorphism

coker TorR - TorR (M , E) ker(M R E - M R E).
1 (M, E) 1
In the special case of a short exact sequence () where M = M = R and

M = R/(a) it follows from 5.2/6 that TorR
1 (R, E) is trivial, since R is at over
R. Hence,

a
- E)
TorR1 (M , E) = Tor1 R/(a), E ker(E
R
is the submodule of a-torsion in E. This is the historical reason for introduc-

ing the terminology of Tor functors for the left derived functors of the tensor
product.
The construction of Tor functors follows a quite universal recipe from ho-
mological algebra. The same methods will be met again in 7.7 where we deal
with Grothendieck cohomology. Let F : R-Mod - R-Mod be a covariant
functor on the category of modules over a certain ring R where F is assumed to
be additive in the sense that it respects the addition of morphisms. Then, given
any R-module M , choose a so-called homological resolution M of M . Thereby
we mean an exact sequence of R-modules
dn+1 dn f
... - Mn+1 - Mn - Mn1 - ... - M0 - M - 0
where, in more strict terms, M is just the sequence

dn+1 dn d0
... - Mn+1 - Mn - Mn1 - ... - M0 - 0.
Then M is exact at all positions except at M0 , unless M is trivial. However, in

any case, M is a complex, which means that the composition of two subsequent
morphisms is always zero. Applying F to the resolution M , we conclude from
the additivity of F that F (M ) is still a complex and we can try to dene the
nth left derived functor Fn of F by

Fn (M ) = ker F (dn ) / im F (dn+1 ) , n N.
In particular, we get F0 = F if F is right exact.

However, such a procedure can only be useful if the resulting modules Fn (M )
are independent of the chosen homological resolution M of M . And, indeed,
5.1 Complexes, Homology, and Cohomology 159
the latter is the case if all members Mn of the resolution M satisfy certain nice
properties, such as being projective. Every free module is projective (see 5.1/7).
Therefore the existence of projective homological resolutions is a triviality. Using
such resolutions, we will explain the general construction of left derived functors
in 5.1, while the particular case of tensor products will be dealt with in 5.2.
For a given covariant additive functor F : R-Mod - R-Mod, the right
n
derived functors F can be constructed similarly as the left derived functors. One
only has to pass to the dual or cohomological situation where all arrows have
been reversed. So for any R-module M we consider a cohomological resolution
M of M given by an exact sequence of R-modules
f dn1 dn
0 - M - M0 - ... - M n1 - Mn - M n+1 - ...
and set
F n (M ) = ker F (dn ) / im F (dn1 ) , n N.
Again, for F n to be well-dened, we have to assume that the members of the
resolution M are suitably chosen, for example, to be injective, which is the
cohomological counterpart of projective. Although the transfer from homologi-
cal to cohomological resolutions is a purely formal procedure reversing arrows,
the existence of injective cohomological resolutions is a non-trivial result that
requires a laborious proof; see 5.3. Finally, for a contravariant additive functor
F : R-Mod - R-Mod its derived functors are constructed viewing it as a
covariant additive functor on the category opposite to the one of R-modules.
Thus, in this case, the left derived functors of F are obtained using cohomo-
logical resolutions, whereas the right derived functors of F use homological
resolutions. As an example, we study in 5.4 the right derived functors of the
Hom functor, namely the so-called Ext functors.
5.1 Complexes, Homology, and Cohomology
Let R be a ring. In the following we will consider R-modules and homomor-

phisms between these, indicating the latter by arrows, as usual. We have already
mentioned in Section 1.5 that a complex, or as we like to say, chain complex of
R-modules is a sequence of R-module homomorphisms
dn+1 dn
... - Mn+1 - Mn - Mn1 - ...
satisfying dn dn+1 = 0 when n varies over Z. One calls

dn
Zn = ker(Mn - Mn1 ) Mn
the submodule of n-cycles,

dn+1
Bn = im(Mn+1 - M n ) Mn
the submodule of n-boundaries, and
Hn = Zn /Bn
the nth homology or the nth homology module of the complex. We will use the
notation M for such a chain complex and write more specically Hn (M ) for
its homology. In many cases, complexes
will satisfy Mn = 0 for all n < 0.
Viewing M as a direct sum nZ Mn , we can combine the homomorphisms
d n : Mn - Mn1 , also known as boundary maps, to yield an R-module ho-
momorphism d : M - M . The latter is said to be of degree 1, since its
application lowers the index of each direct summand by 1. In most cases, we
will just write d instead of dn .
Passing to the dual notion of a chain complex, we arrive at the notion of a
cochain complex M . It is of type
dn1 dn
... - M n1 - Mn - M n+1 - ...
where its n-cocycles are given by

dn
Z n = ker(M n - M n+1 ) M n ,
its n-coboundaries by
dn1
B n = im(M n1 - M n) M n,
and its nth cohomology module by
H n (M ) = Z n /B n .
A homomorphism f : M - N (of degree 0) between chain complexes

M and N is dened as an R-module homomorphism M - N of degree
0, in other words, as a family of R-module homomorphisms fn : Mn - Nn
where the diagram
d
Mn - Mn1
fn fn1
? d
?
Nn - Nn1
is commutative for all n Z. In particular, we can consider sequences of complex
homomorphisms of type
0 - M - M - M - 0,
calling such a sequence exact if, for each level n Z, the sequence
0 - Mn - Mn - Mn - 0
is exact.
Proposition 1. Every exact sequence of chain complexes

f g
0 - M - M - M - 0
gives in a canonical way rise to a so-called long exact homology sequence

Hn (f ) Hn (g)
... - Hn (M ) - Hn (M ) - Hn (M )
Hn1 (f )
- Hn1 (M ) - ...
Proof. Let us start by a preliminary remark and exemplarily consider the com-
plex M where we write in more precise terms
dn+1 dn
... - Mn+1 - Mn - Mn1 - ... .
Then, using im dn+1 ker dn and im dn ker dn1 , the sequence induces a
homomorphism
dn : coker dn+1 - ker dn1 Mn1
where, apparently,
ker dn = ker dn / im dn+1 = Hn (M ),

coker dn = ker dn1 / im dn = Hn1 (M ).
Now let us verify the assertion of the proposition which, basically, is a

generalization of the Snake Lemma 1.5/1. Applying this lemma, we arrive at
the following commutative diagram with exact rows and columns:
0 0 0
? ? ?
0 - ker dn - ker dn - ker dn
? ? ?
fn gn
0 - Mn - Mn - Mn - 0
dn dn d
n
? ? ?
fn1 gn1
0 - Mn1 - Mn1 - Mn1 - 0
? ? ?
coker dn - coker dn - coker dn - 0
? ? ?
0 0 0
Applying the above remark to M , M , and M , induces a diagram
0 0 0
? ? ?
Hn (M ) - Hn (M ) - Hn (M )
? ? ?
coker dn+1 - coker dn+1 - coker dn+1 - 0
dn dn d
n
? ? ?
0 - ker dn1 - ker dn1 - ker dn1
? ? ?
Hn1 (M ) - Hn1 (M ) - Hn1 (M )
? ? ?
0 0 0
where the columns are exact and the same is true for the second and the third
rows. But then the Snake Lemma yields the assertion of the proposition.
Of course, there is a similar assertion for cochain complexes. Moreover, we

can easily derive the following fact from Proposition 1.
Corollary 2. Let
f g
0 - M - M - M - 0
be an exact sequence of chain complexes. If the homology of two of the complexes
M , M , M is trivial, the same holds for the third complex as well.
As an example of a long homology sequence, we will study in Section 5.2 the

so-called long exact Tor sequence associated to an exact sequence of R-modules
0 - M - M - M - 0
and a further R-module E. Due to the right exactness of tensor products (see
4.2/1), such a sequence leads to an exact sequence
M R E - M R E - M R E - 0
and, as we will see, the latter can be enlarged to an exact sequence of type

n (M , E) n (M, E) n (M , E)

n1 (M , E) n1 (M, E) n1 (M , E)
- ...
1 (M , E) 1 (M, E) 1 (M , E)
- M R E - M R E - M R E - 0
which turns out to be a long exact homology sequence in the setting of Propo-
sition 1.
We have already seen in Proposition 1 and its proof that a complex ho-
momorphism f : M - N induces a family Hn (f ) : Hn (M ) - Hn (N )
of homomorphisms on the level of homology. Given two such homomorphisms
f, g : M - N , one is interested in conditions assuring Hn (f ) = Hn (g) for all
n. Sucient for this, but in general not necessary, is the existence of a so-called
homotopy between f and g.
Denition 3. Let f, g : M - N be two homomorphisms of chain complexes.

A homotopy between f and g is a module homomorphism M - N of degree
1, in other words, a system of R-module homomorphisms hn : Mn - Nn+1 ,
such that the maps of the diagram
... - Mn+1 - Mn - Mn1 - ...
hn+1 hn hn1 hn2
f g f g
? ? ?

... - Nn+1 - Nn - Nn1 - ...

satisfy the relation f g = h d + d h. We will call f and g homotopic in this
case.
The following fact is immediately clear by the relation characterizing a

homotopy:
Remark 4. Any two homomorphisms of chain complexes f, g : M - N ,

which are homotopic, induce the same homomorphisms
Hn (f ), Hn (g) : Hn (M ) - Hn (N ), n Z,
on the level of homology.
Denition 5. Let M be an R-module. A homological resolution of M is an

exact sequence of R-modules
f
... - M2 - M1 - M0 - M - 0
where the latter is to be interpreted as a homomorphism of chain complexes

M - M or, in more explicit notation,
... - M2 - M1 - M0 - 0
f
? ? ?
... - 0 - 0 - M - 0,
inducing an isomorphism on the level of homology.
Let F : R-Mod - R-Mod be a covariant functor which is additive in

the sense that, for any R-modules M and N , the map

HomR (M, N ) - HomR F (M ), F (N )
is a group homomorphism. In most cases we will assume that, in addition,

F is right exact, which means that application of F to an exact sequence of
R-modules
M -M - M -0
produces an exact sequence of R-modules
F (M ) - F (M ) - F (M ) - 0
again. As an example we may look at the functor
F: M - M R E,
for a xed R-module E; see 4.2/1. In the following we will construct the so-
called left derived functors of F , the Tor functors in the case of the example.
The process is quite simple. Given an R-module M , resolve it by R-modules
behaving well with respect to F in a certain sense, apply the functor F to the
resolution and consider the homology modules of the resulting complex. For
this to work well, we need projective resolutions; i.e. resolutions by projective
modules which are dened as follows.
Denition 6. An R-module P is called projective if for every epimorphism

: M - M the map HomR (P, M ) - HomR (P, M ), f - f , derived
from is surjective:
P
?

M - M - 0
Proposition 7. Every free module is projective. More precisely, an R-module P

is projective if and only if it is a direct summand of a free one, in other words,
if and only if there is an R-module P such that the direct sum P P is free.
Proof. Clearly, a free R-module F is projective, since homomorphisms F - M

to an arbitrary R-module M can be uniquely constructed by selecting values for
the images of a free generating system of F . In particular, if P is an R-module
such that P P is free, P P will be projective. Then consider a morphism of
R-modules g : P - M and an epimorphism : M - M . Composing g
with the projection P P - P onto the rst factor, this map factors through
: M - M via a morphism f : P P - M and f |P is a factorization of
g through . Hence, P is projective.
Conversely, assume that P is projective and consider a free module F to-
gether with an epimorphism : F - P . Then the short exact sequence

0 - ker - F - P - 0
is split. Indeed, due to the projectivity of P , the identity map idP : P -P
factors through F and, thus, admits a section. Therefore F is isomorphic to the
direct sum P ker and we see that P is a direct summand of a free module.

Let us add that an R-module M is projective and of nite type if and only
if it is locally free of nite rank in the sense of 4.4/2; see Exercise 5 below. Thus,
in 4.4/3, one may add as a fourth equivalent condition M to be projective of
nite type.
Proposition 8. Every R-module M admits a free and, hence, projective homo-

logical resolution.
Proof. Choose a generating system (xi )iI of M and let M0 = R(I) . Then the
homomorphism M0 - M mapping the canonical free generating system of
M0 onto the system (xi )iI is surjective and the construction may be repeated
for the kernel of this map. Continuing this way, we arrive at a free homological
resolution of M .
We need to know in which way dierent projective homological resolutions

of an R-module M might dier.
Lemma 9. Let M - M and N - N be projective resolutions of two

R-modules M and N and let : M - N be an R-module homomorphism.
Then:
(i) There exists a complex homomorphism f : M - N such that the
diagram
M -M
f
? ?
N - N
is commutative.
(ii) If f, f : M - N are complex homomorphisms as in (i), they are
homotopic.
Proof. Starting with assertion (i), we have to construct a complex homomor-

phism f : M - N such that the diagram
... - M2 - M1 - M0 - M - 0
f2 f1 f0
? ? ? ?
... - N2 - N1 - N0 0 - N -
is commutative. To obtain f0 , look at the composition d : M0 - N and lift

it to a homomorphism f0 : M0 - N0 , using the projectivity of M0 . Next, to
dene f1 , we proceed similarly and consider the composition f0 d : M1 - N0 ,

whose image is contained in the kernel ker(N0 - N ). Since the latter coincides
with the image im(N1 - N0 ), we can apply the projectivity of M1 in order
to lift f0 d to a homomorphism f1 : M1 - N1 . Continuing this way, we can
inductively construct homomorphisms fi : Mi - Ni satisfying the required
commutativity conditions.
Turning to assertion (ii), let f, f : M - N be complex homomorphisms
as in (i) and set g = f f . We look at the diagram
... - M2 - M1 - M0 - M - 0
h2 h1 h0 h1
g2 g1 g0 0
? ? ? ?

... - N2 - N1 - N0 - N - 0
where the maps hi are obtained as follows. Start by dening h1 : M - N0 as

the zero map. Then we use the projectivity of M0 to construct h0 : M0 - N1
with
g0 = d h0 = d h0 + h1 d,
observing that im g0 ker(N0 - N ) = im(N1 - N0 ). Next we have to
dene h1 : M1 - N2 such that g1 = d h1 + h0 d. Using the projectivity of
M1 , it is enough to show that
im(g1 h0 d) ker(N1 - N0 ) = im(N2 - N1 ).
However, the latter is clear, since we have d2 = 0 and, thus,
d g1 d h0 d = d g1 (g0 h1 d) d = d g1 g0 d = 0.
Continuing this way, we can construct the required maps hi by induction.
If M - M and M - M are two projective homological resolutions

of some R-module M , then applying Lemma 9 we obtain two complex homo-
morphisms f : M - M and g : M - M , which are compatible with the
identity on M . Furthermore, it follows that g f and f g are homotopic to the
identity on M , respectively M . We will refer to such a setting by saying that
f : M - M and g : M - M dene a homotopy equivalence between M
and M .
Next let us consider a covariant functor
F : R-Mod - R-Mod
on the category of R-modules that is additive. As we have seen the latter means
that, for any R-modules M and N , the map

HomR (M, N ) - HomR F (M ), F (N )
given by F is additive in the sense of being a group homomorphism. Then F

respects direct sums, since a direct sum of type A B can be characterized by
a diagram
p j
A - AB - B
i q
where
p i = idA , q j = idB ,
q i = 0, p j = 0,
i p + j q = idAB .
As an example, let us point out that the functor F = R E, for a xed

R-module E, is additive.
Given any sequence M of R-module homomorphisms, we can apply the
functor F to it, thereby obtaining a sequence of R-module homomorphisms
again; the latter will be denoted by F (M ). If M admits the complex property
and F is additive, F (M ) will be a complex as well. However, in general, F will
not preserve resolutions, even if F is additive.
Proposition and Denition 10. As before, let F be an additive covariant

functor on the category of R-modules. For an arbitrary R-module M , dene
R-modules Fn (M ), n N, as follows: choose a projective homological resolution
M - M and set

Fn (M ) = Hn F (M ) for n N.
Then Fn is canonically a well-dened functor on the category of R-modules, the

so-called nth left derived functor associated to F .
Proof. Fix R-modules M, N , as well as projective homological resolutions

M - M and M - N . Then, applying Lemma 9 (i), we can asso-
ciate to any module homomorphism : M - N a complex homomorphism
f : M - N such that the diagram
M - M
f
? ?
N - N
is commutative. It follows from Lemma 9 (ii) that f is uniquely determined by
this diagram, up to homotopy. Thus, if f : M - N is a second complex
homomorphism making the above diagram commutative, there is a homotopy
between f and f in the sense of Denition 3.
Since F is additive, the images F (M ) and F (N ) will still have the complex
property and, moreover, F will transform the homotopy between f and f into
a homotopy between F (f ) and F (f ). Thus, using Remark 4, the resulting
homomorphism

Hn F (f ) : Hn F (M ) - Hn F (N )
will be independent of the choice of f . Furthermore, we see for M = N that

two projective homological resolutions of one and the same R-module M are
homotopy equivalent and that such an equivalence is respected by the functor
F , due to its additivity. Therefore the complexes F (M ) and F (N ) are ho-
motopy equivalent and the homology modules Hn (F (M )) and Hn (F (N )) can
canonically be identied.
In general, we will assume that the functor F , besides being covariant and
additive, is right exact.
Remark 11. Let F be a covariant additive functor with left derived functors
Fn , n N, on the category of R-modules. Then:
(i) If F is right exact, F0 = F .
(ii) If M is a projective R-module, Fn (M ) = 0 for n 1.
Proof. Let F be right exact and let M be an R-module with a projective reso-
lution M - M , say, given by the exact sequence
... - M1 - M0 - M - 0.
Then applying F yields the exact sequence
F (M1 ) - F (M0 ) - F (M ) - 0
and therefore an isomorphism F0 (M ) - F (M ). On the other hand, if M is

projective, we can take M = (. . . -0 -0 - M0 - 0) with M0 = M
as a projective resolution of M , thus, showing Fn (M ) = 0 for n 1.
Proposition 12 (Long exact homology sequence). Let F be a covariant additive

functor with left derived functors Fn , n N, on the category of R-modules. Then
every exact sequence of R-modules
0 - M - M - M - 0
induces canonically a long homology sequence

... - Fn (M ) - Fn (M ) - Fn (M ) - Fn1 (M ) - ...

... - F0 (M ) - F0 (M ) - F0 (M ) - 0 .
Proof. We start out from projective homological resolutions M - M and

M - M and construct a commutative diagram
.. .. ..
. . .
d2
? ? ?
0 - M1 - M1 M1 - M1 - 0
d1
? ? ?
0 - M0 - M0 M0 - M0 - 0

? ? ?
0 - M - M - M - 0
? ? ?
0 0 0
with exact rows and columns where the left and right columns are given by
the resolutions M - M and M - M . The bottom row consists of the

given exact sequence involving the modules M , M, M and the rows at positions
above the bottom row are the canonical short exact sequences associated to the
direct sums Mn Mn . It remains to specify the maps of the central column.
We want to construct them in such a way that we get a projective resolution

of M . To dene , consider the composition : M0 - M - M and lift

: M0 -
M to a homomorphism : M0 - M , using the projectivity
of M0 . Then the map
: M0 M0 - M, a a - (a ) + (a ),
is an epimorphism making the two lower squares commutative. Furthermore,

the Snake Lemma 1.5/1 shows that the sequence
0 - ker - ker - ker - 0
induced by the row above the last one is exact. Departing from this sequence
and proceeding similarly as we have done with , we can construct a map
d1 : M1 M1 - M M with image ker , making the squares adjacent to
0 0
d1 commutative. Continuing this way, the central column becomes, indeed, a
homological resolution M of M which, in addition, is projective, due to the fact
that Mn and Mn and, hence, Mn Mn are projective. Thus, we have obtained
from the exact sequence
() 0 - M - M - M - 0
an exact sequence
0 - M - M - M - 0
of associated projective resolutions.

Now applying the functor F yields the sequence
() 0 - F (M ) - F (M ) - F (M ) - 0
of complexes which remains exact, since F is additive and, hence, respects direct
sums in the sense that, for all n N, it transforms the canonical exact sequence
0 - Mn - Mn Mn - Mn - 0
into the canonical exact sequence
0 - F (Mn ) - F (Mn ) F (Mn ) - F (Mn ) - 0.
Finally, applying Proposition 1 to the exact sequence () yields the claimed

long exact homology sequence. As shown in Remark 13 below, the latter depends
naturally on the exact sequence ().
Remark 13. In the situation of Proposition 12, the construction of the long
exact homology sequence depends functorially on the exact sequence
0 - M - M - M - 0.
More precisely, a commutative diagram of R-modules
0 - M - M - M - 0
? ? ?
0 - N - N - N - 0
with exact rows canonically yields a commutative diagram
. . . Fn+1 (M ) - Fn (M ) - Fn (M ) - Fn (M ) - Fn1 (M ) . . .
? ? ? ? ?
... Fn+1 (N ) - Fn (N ) - Fn (N ) - Fn (N ) - Fn1 (N ) ...
between associated long exact homology sequences such that the properties of a
functor (from the category of short exact sequences of R-modules to the category
of long exact sequences of R-modules) are satised.
Proof. It follows from Lemma 9 that the vertical maps between the two long
exact homology sequences are unique and depend canonically on the maps be-
tween the modules M , M, M and N , N, N . To show that the squares of the
diagram of the long exact homology sequences are commutative, we must know
that the maps occurring in the Snake Lemma 1.5/1 depend in a functorial way
on the involved modules and homomorphisms. However, the latter is clear, since
the formation of the kernels ker and cokernels coker of module homomorphisms
is functorial.
Exercises
1. Give an example of two complex homomorphisms f, g : M - N that are not
homotopic although they induce the same morphisms Hn (f ) = Hn (g), n Z, on
the level of homology.
2. Show that any direct sum of projective R-modules is projective.
3. Show that every projective R-module is at. Is the converse of this true as well?
4. Let M be a projective R-module and let S R be a multiplicative subset. Show
that the localization M R RS is a projective RS -module.
5. Show that an R-module M is projective of nite type if and only if it is locally
free of nite rank.
Hints: First assume that M is projective of nite type. Deduce that M is a direct
factor of a nite free R-module and conclude that M is of nite presentation.
Now use 4.4/3 (ii) in conjunction with Exercise 3 above to see that M is locally
free of nite rank.
Conversely, consider elements f1 , . . . , fr R generating the unit ideal in R such
that M R Rfi is nite free for all i. Writing R = Rf1 . . . Rfr show that
M R R is projective. Finally, use that the canonical morphism R - R is
faithfully at (as in the proof of 4.4/3) and conclude that M is projective of nite
type.
For further details see Bourbaki [6], II.5.2, Thm. 1 and, in particular, [6], II.3.6,
Prop. 12.
6. The projective dimension pd(M ) of any R-module M is dened as the minimum of
all integers d such that there exists a projective resolution M - M satisfying
Mn = 0 for all n > d. If no such resolution exists, we put pd(M ) = . Show for
R-modules M, N that pd(M N ) = max(pd(M ), pd(N )).
7. Let F : R-Mod - R-Mod be a covariant functor on the category of R-modules
that is additive and right exact. Show that F is exact if and only if the left derived
functors Fn , n > 0, are trivial. The latter is equivalent to the fact that F1 is trivial.
For example, consider a multiplicative subset S R and let F be the functor
that associates to any R-module M its localization MS , viewed as an R-module.
Determine the left derived functors of F .
8. Computing homology via acyclic resolutions: As before, let F be a covariant ad-
ditive and right exact functor on the category of R-modules. Let F0 , F1 , . . . be
its left derived functors. An R-module A is called F -acyclic if Fn (A) = 0 for all
n > 0. Likewise, a homological resolution A - M of some R-module M is
called F -acyclic if all R-modules An are F -acyclic. For such a resolution, show
that there are canonical isomorphisms Fn (M ) - Hn (F (A )), n N, and that
these are functorial in M . Hence, the left derived functors of F can be computed
using F -acyclic homological resolutions.
Hints: Consider a projective resolution P - M and generalize Lemma 9
by showing that there exists a complex homomorphism f : P - A , unique
up to homotopy, such that H0 (f ) equals the identity on M . Constructing the
f n : Pn - An step by step and adapting the Pn if necessary, the fn can be
assumed to be surjective. Then, dening K as the kernel of f , one gets a
short exact sequence of complexes (): 0 - K - P - A - 0.

Note that, due to the F -acyclicity of A , the sequence of complexes F () ob-
tained from () by applying F remains exact. Furthermore, show that K - 0
is an F -acyclic resolution of the zero module and, by induction on n, that all
kernels ker(Kn+1 - Kn ) are F -acyclic. Deduce from this that the complex
F (K ) is exact. Now use the long exact homology sequence (Proposition 1) at-
tached to the exact sequence F () to conclude that the canonical homomorphisms
Hn (F (P )) - Hn (F (A )), n N, are isomorphisms.
For more details see the cohomological version in Lang [17], Thm. XX.6.2.
5.2 The Tor Modules
In the following we want to apply the constructions of Section 5.1 in order to

study the functor
R E : R-Mod - R-Mod, M - M R E
for a xed R-module E. Note that this is a covariant additive functor which is
right exact due to 4.2/1.
Denition 1. The nth left derived functor of R E is denoted by TorR

n (, E).
Thus, if M - M is a projective homological resolution of an R-module M ,
we have
n (M, E) = Hn (M R E),
TorR n N,
and the latter is called the nth Tor module, associated to M and E. In particular,
0 (M, E) = M R E by 5.1/11.
TorR
Rewriting 5.1/12 in our special setting, we arrive at the following fact:
Proposition 2 (First long exact Tor sequence). Let E be an R-module. Then

every short exact sequence of R-modules
0 - M - M - M - 0
induces a long exact sequence
... - TorR
2 (M , E)

1 (M , E) 1 (M, E) 1 (M , E)
- M R E - M R E - M R E - 0.
By its construction, TorR

n (M, E) is a functor in M , but apparently also in E,
since any morphism of R-modules E - E gives rise to a functorial morphism
R E - R E.
5.2 The Tor Modules 173
We say that TorR

n is a bifunctor on the category of R-modules. On the other
hand, to dene TorR
n (M, E), we could just as well interchange the roles of M
and E, by choosing a projective homological resolution E - E of E and
setting
n (M, E) = Hn (M R E ).
TorR
In the following we want to show that Hn (M R E) and Hn (M R E ) are

canonically isomorphic and, hence, that both possible denitions are equivalent.
To achieve this, we need some preparations.
Let M and E be complexes satisfying Mn = En = 0 for n < 0. Then we
arrange the tensor products Mp R Eq as a double complex
.. .. ..
. . .
? ? ?
0 M 2 R E 0 M2 R E1 M2 R E 2 ...
d
? d ? ?
0 M 1 R E 0 M1 R E1 M1 R E 2 ...
? ? ?
0 M 0 R E 0 M0 R E1 M0 R E 2 ...
? ? ?
0 0 0
and look at the associated single complex M R E , which is given by

(M R E )n = Mp R E q , n Z,
p+q=n
together with the boundary maps
dn = d + d : (M R E )n - (M R E )n1
where d = (1)p d on the row with index p of the double complex. To show
that we really get a complex, imagine replacing d by d in the above diagram.
Then the squares become anticommutative in the sense that d d + d d = 0
and we observe that d d = 0, as claimed.
Any complex homomorphisms M - M and E - E induce canonical

complex homomorphisms
M R E - M R E and M R E - M R E
between single complexes, as introduced above. In particular, we will need such

homomorphisms in cases where M - M or E - E are homological

resolutions of given R-modules M and E.
Proposition 3. Let M - M and E - E be projective homological

resolutions of two R-modules M and E. Then the canonical complex homomor-
phisms
M R E - M R E, M R E - M R E
induce canonical isomorphisms
Hn (M R E) Hn (M R E ) - Hn (M R E ), n N.
Corollary 4. The Tor functors can be interpreted as left derived functors of

the tensor product, regardless of the type of functors R E or M R we use.
Consequently, the modules TorR n (M, E), n N, for R-modules M, E can be
dened via the above homology modules.
i (M, ) as the ith left derived functor of the functor M R ,

Interpreting TorR
we can deduce a second Tor sequence from 5.1/12:
Corollary 5 (Second long exact Tor sequence). Let M be an R-module. Then

0 - N - N - N - 0
induces a long exact sequence
... - TorR
2 (M, N )

1 (M, N ) 1 (M, N ) 1 (M, N )
- M R N - M R N - M R N - 0.
Proof of Proposition 3. We restrict ourselves to looking at the complex homo-

morphism M R E - M R E . As the tensor product is symmetric in its
factors, the morphism M R E - M R E can be dealt with in the same
way. Let us start out from the following diagram
.. .. ..
. . .
? ? ?
0 M 2 R E 0 M2 R E1 M2 R E 2 ...
d
? d ? ?
0 M 1 R E 0 M1 R E1 M1 R E 2 ...
? ? ?
0 M 0 R E 0 M0 R E1 M0 R E 2 ...
? ? ?
0 M R E 0 M R E1 M R E 2 ...
? ? ?
0 0 0
where the bottom row is the complex M R E and the upper part coincides
with the double complex associated to M and E , as discussed above. Since a
projective module, as a direct summand of a free module, is at by 4.2/6, we
get from 4.2/4 exactness properties in the diagram as follows:
(i) All rows are exact at positions with column index q > 0, except for the
bottom row.
(ii) All columns are exact.
Now look at the complex homomorphism M R E - M R E and the
induced homomorphisms
n : Hn (M R E ) - Hn (M R E ), n N.
We claim that these, in fact, are isomorphisms. To show that n is surjective,

start with an element x Hn (M R E ) and choose a representative
x Zn (M R E ) M R En .
We want to construct an element in

Zn (M R E ) Mp R E q
p+q=n
which represents a preimage of x. Therefore choose a preimage x0,n M0 R En

of x. In general, the latter will not belong to Zn (M0 R E ), but we can nd an
element x1,n1 M1 R En1 such that
d (x1,n1 ) + (1)0 d (x0,n ) = 0.
This is possible, since x Zn (M R E ) implies that the image of x0,n under

d : M0 R En - M0 R En1
is contained in
d
ker(M0 R En1 - M R En1 ) = im(M1 R En1 - M0 R En1 ).
In the same way we can construct x2,n2 M2 R En2 such that
d (x2,n2 ) + (1)1 d (x1,n1 ) = 0
and so on. The process ends with some element xn,0 Mn R E0 and we see
that, by our construction,
x0,n . . . xn,0 Zn (M R E ).
Then it is easily checked that this element represents a preimage of x with

respect to n : Hn (M R E ) - Hn (M R E ). Hence, n is surjective.
To show that n is injective, consider an element
() x0,n . . . xn,0 Zn (M R E ), xp,q Mp R Eq ,
whose image x in M R En belongs to Bn (M R E ). Then x is just the image

of x0,n under the map M0 R En - M R En . To show
x0,n . . . xn,0 Bn (M R E ),
choose a preimage of x in M R En+1 and, again, a preimage y0,n+1 M0 R En+1

of the latter. Then
x0,n (1)0 d (y0,n+1 )
belongs to the kernel of M0 R En - M R En and therefore admits a preimage
y1,n M1 R En . Using (), we can compute

d x1,n1 (1)1 d (y1,n ) = d (x1,n1 ) + (1)0 d d (y1,n )

= d (x1,n1 ) + (1)0 d x0,n (1)0 d (y0,n+1 )
= d (x1,n1 ) + (1)0 d (x0,n ) = 0
and we see that x1,n1 (1)1 d (y1,n ) admits a preimage y2,n1 M2 R En1 .
Repeating this construction, we nally arrive at an element
y0,n+1 . . . yn+1,0 (M R E )n+1
such that
d(y0,n+1 . . . yn+1,0 ) = x0,n . . . xn,0
and it follows x0,n . . . xn,0 Bn (M R E ), which justies our claim.
Let us point out that there is a more convenient way of proving Proposition 3
by using the advanced technical tool of so-called spectral sequences, namely
the spectral sequence associated to the double complex under consideration.
However, as we will not touch this subject, our proof of the proposition had to
be done by hand.
Next we want to give a characterization of atness in terms of Tor modules.
To begin with, observe the following fact.
Remark 6. Let M and N be R-modules and assume that one of them is pro-
n (M, N ) = 0 for n 1.
jective or, more generally, at. Then TorR
Proof. If one of the modules is projective, we may use 5.1/11 in conjunction

with Corollary 4. If N is just known to be at, the functor R N is exact.
Therefore, if M - M is a projective homological resolution of M , we see
that M R N - M R N is still a homological resolution of M R N , and
we get TorRn (M, N ) = 0 for n 1. Using Corollary 4, the same follows if M is
at.

(i) M is at.
n (M, N ) = 0 for all n 1 and all R-modules N .
(ii) TorR
(iii) TorR
1 (M, N ) = 0 for all R-modules N of nite type.
(vi) TorR1 (M, R/a) = 0 for all nitely generated ideals a R.
Proof. The implication (i) = (ii) is a consequence of Remark 6 and the impli-
cations (ii) = (iii) = (iv) are trivial.
Next assume condition (iv). In order to derive (iii), consider an R-module
N of nite type, say with a set of generators of length s 1. We show by
induction on s that TorR1 (M, N ) = 0. If s = 1, there is an element x N such
that N = Rx. Then the map
R - N, a - ax,
is surjective and its kernel yields an ideal a R such that N R/a. We claim
that TorR R
1 (M, N ) = Tor1 (M, R/a) = 0.
If a is nitely generated, we know TorR 1 (M, N ) = 0 from (iv). Otherwise we
look at the long exact Tor sequence
. . . TorR - TorR (M, R/a) - M R a - M R R = M -
1 (M, R) 1 ....
Since R, as an R-module, is free, we have TorR
1 (M, R) = 0 due to Remark 6,
thereby obtaining an isomorphism
TorR - ker(M R a -
1 (M, R/a) M ).
For every nitely generated ideal a a the composition of canonical maps
M R a - M R a - M

is injective, since we haveTorR
1 (M, R/a )= 0 by (iv). Now if z = ri=1 mi ai
is an element of the kernel of M R a - M , let a = (a1 , . . . , ar ) R be the

ideal which is generated by the elements ai . Then


r
z = mi ai ,
i=1
makes sense as an element of M R a . Clearly, z is a preimage of z and therefore

belongs to the kernel of M R a - M . But then we must have z = 0 and,
hence, z = 0. In other words, M R a - M is injective and we see that
R R
Tor1 (M, N ) = Tor1 (M, R/a) = 0, as claimed. s1
s
Now assume s > 1, say N = i=1 Rxi . Then let N = i=1 Rxi and
consider the exact sequence
0 - N - N - N - 0
where N N/N is generated by a single element, namely the residue

class of xs . By induction hypothesis we can assume TorR
1 (M, N ) = 0 and
R
Tor1 (M, N ) = 0 so that the long exact Tor sequence
- - - -
... TorR
1 (M, N ) TorR
1 (M, N ) TorR
1 (M, N ) ...
shows TorR
1 (M, N ) = 0, as desired.
To pass from condition (iii) to (i), let N - N be an injective homomor-
phism of R-modules and look at the associated exact sequence
0 - N - N - N - 0
where N = N/N . In order to see that M is at, we must show that the
tensorized map M R N - M R N is injective. To do this, assume for a
moment that N is nitely generated. Then N is nitely generated as well and

we have TorR1 (M, N ) = 0 if condition (iii) is satised. Therefore the long Tor
sequence

... - TorR
1 (M, N )
- M R N - M R N - M R N - 0
shows that the sequence
0 - M R N - M R N - M R N - 0
is exact. In particular, M R N - M R N is injective.

If N is not necessarily nitely generated, x any element z = ri=1 mi ni
belonging to the kernel of M R N - M R N . To prove z = 0, we may

replace N by the submodule generated by n1 , . . . , nr . Furthermore, we know
from 4.2/8 that there is a nitely generated submodule N N containing N
such that the image of z is trivial in M R N . Thus z belongs to the kernel of
M R N - M R N and it follows from the special case dealt with above
that z is trivial. Therefore M R N - M R N is injective and, thus, M is
at.
It is easy now to verify the criterion 4.2/9 which we had mentioned without
giving a proof.
Corollary 8. An R-module M is at if and only if for every nitely generated

ideal a R the canonical map a R M - M is injective.
Proof. Assume rst that M is a at R-module. Then, of course, the canonical

map a R M - M derived from the injection a - R remains injective
for all ideals a R. Conversely, assume this condition for all nitely generated
ideals a R. Fixing such an ideal, look at the short exact sequence
0 - a - R - R/a - 0
and the associated long exact Tor sequence
... - TorR - TorR

1 (M, R) 1 (M, R/a)
- a R M - M - M/aM - 0.
We have TorR 1 (M, R) = 0 by Remark 6, since R, as an R-module, is free. There-

fore, if a R M - M is injective, we must have TorR (M, R/a) = 0. However,
1
if the latter is satised for all nitely generated ideals a R, Proposition 7 says
that M is at.

(i) M is at.
(ii) For every short exact sequence
0 - M - M - M - 0
and every R-module N , the tensorized sequence
0 - M R N - M R N - M R N - 0
is exact.
Proof. Let 0 - M -M - M - 0 be a short exact sequence

and assume that M is at. Then Remark 6 yields TorR1 (M , N ) = 0 for any
R-module N and the long exact Tor sequence

TorR
1 (M , N )
- M R N - M R N - M R N - 0
shows that condition (i) implies (ii).

Conversely, assume condition (ii) and look at an exact sequence
0 - M - M - M - 0
where M is free. Then the sequence
0 - M R N - M R N - M R N - 0
is exact for arbitrary R-modules N . Since TorR

1 (M, N ) = 0 by Remark 6, as M
was assumed to be free, the long exact Tor sequence
TorR - TorR (M , N ) - M R N - M R N - M R N - 0
1 (M, N ) 1

shows TorR
1 (M , N ) = 0 so that we get (i) using Proposition 7.
Proposition 10. Let
0 - M - M - M - 0
be a short exact sequence of R-Modules where M is at. Then M is at if and

only if M is at.
Proof. Given an arbitrary R-module N , we can look at the following part of the
long exact Tor sequence
- - -
TorR
2 (M , N ) TorR
1 (M , N ) TorR
1 (M, N ) TorR
1 (M , N )

where TorR R
2 (M , N ) = 0 = Tor1 (M , N ) by Remark 6, since M is at. There-
fore the sequence yields an isomorphism
TorR - TorR (M, N )

1 (M , N ) 1
and we see from Proposition 7 that M is at if and only if M is at.
Exercises
1. For non-zero integers p, q Z show that TorZ1 (Z/(p), Z/(q)) Z/(gcd(p, q)).
2. Let M, N be modules over a principal ideal domain R.
n (M, N ) = 0 for n 2.
(a) Show that TorR
1 (R/(a), M ) for any element a R.
(b) Determine TorR
(c) Determine TorR
1 (Q(R)/R, M ) for the eld of fractions Q(R) of R.
3. Let R be a local ring with maximal ideal m and M an R-module of nite presen-
tation. Show that M is free if and only if TorR
1 (M, R/m) = 0.
4. The formation of Tor modules is compatible with at coecient extension: For a

at ring morphism R - R and R-modules M, N show that there are canonical
isomorphisms of R -modules TorR - TorR (M R R , N R R ),
n (M, N ) R R n
n N. In particular, for any prime ideal p R, there are canonical isomorphisms
TorR - TorR n (Mp , Np ), n N.
p
n (M, N )p
5. Group homology: The group ring ZG of a group G is dened as the free
Z-module generated
by the elements
of G, i.e. Z

G = gG Z g, with multipli-
cation given by ( gG ag g) ( hG bh h) = g,hG ag bh gh. Every Z-module
M can trivially be viewed as a ZG-module by setting g x = x for all g G
and x M .
Given any ZG-module M , the nth homology group, n N, of G with values
Z G
in M is dened by Hn (G, M ) = Torn (Z, M ). Compute Hn (G, M ) for a cyclic
group G.
5.3 Injective Resolutions 181
6. Bourbakis Criterion of atness: Consider an R-module M and an ideal I R.

Assume either that I is nilpotent in the sense that Ir = 0 for a certain expo-
nent
r n> 0, or that R is Noetherian and M ideally separated; the latter means
nN I (a R M ) = 0 for every nitely generated ideal a R. Show that the
following conditions are equivalent:
(a) M is a at R-module.
(b) TorR
1 (M, N ) = 0 for every R-module N satisfying IN = 0.
(c) M/IM is a at R/I-module and TorR
1 (M, R/I) = 0.
(d) M/In M is a at R/In -module for all n 1.
Hint: See Bourbaki [6], III, 5.3.
5.3 Injective Resolutions
Let M and N be R-modules. Then we can look at the covariant additive functor
HomR (M, ) : R-Mod -R-Mod,

E - HomR (M, E),

(f : E - E ) - - f , HomR (M, E) ,
which is easily seen to be left exact in the sense that it transforms exact se-
quences of type
0 - E -E - E
into exact sequences
0 - HomR (M, E ) - HomR (M, E) - HomR (M, E ).
On the other hand, Hom can be viewed as a functor in the rst variable as well,
HomR (, N ) : R-Mod -R-Mod,

E - HomR (E, N ),

(f : E - E) - - f, HomR (E, N ) ,
and we see that HomR (, N ) is an additive, in this case contravariant functor,

which is left exact in the sense that it transforms exact sequences of type
E - E - E - 0
into exact sequences
0 - HomR (E , N ) - HomR (E, N ) - HomR (E , N );
just use the Fundamental Theorem on Homomorphisms 1.4/6.

We already know from Section 5.1 that an R-module P is called projective
if for each surjective morphism of R-modules E - E the associated map
HomR (P, E) - HomR (P, E ) is surjective, a property which is characterized

by the following diagram:

E - E - 0
Passing to the dual diagram
-
6
E E 0
we obtain the notion of an injective R-module.
Denition 1. An R-module I is called injective if for every injective morphism

of R-modules E - E and a given morphism of R-modules E - I, the
latter can always be extended to a morphism E - I, in other words, if any
injection E - E induces a surjection HomR (E, I) - HomR (E , I).
Calling an additive functor exact if it is left and right exact and, thus,
preserves short exact sequences, we see:
Remark 2. (i) An R-module P is projective if and only if the functor

HomR (P, ) is exact.
(ii) An R-module I is injective if and only if the functor HomR (, I) is exact.
Also note that an additive functor preserves general exact sequences as soon
as it preserves short exact ones. This is seen just as in the proof of 4.2/4.
Clearly, the zero module over any ring R is projective, as well as injective.
Also we know that any free module is projective. Just as any direct sum of
projective modules is projective, any cartesian product of injective modules is
injective; cf. Exercise 1 below. Any vector space over a eld is projective, since
it is free. But it is injective as well, because any subspace of a vector space
admits a direct complement. Furthermore, one knows that Q is an injective
Z-module and that any Q-vector space is injective when viewed as a Z-module,
which is a little bit more laborious to prove; see Proposition 5 below. There
are further examples of injective modules; we will explain some of the possible
constructions in a moment.
For additive covariant functors, their left derived functors were dened using
projective homological resolutions. For contravariant functors we need to dualize
this concept. A cohomological resolution of a given R-module M is an exact
sequence of type
f
0 - M - M0 - M1 - M2 - ...,
which we interpret, similarly as for homological resolutions, as a complex ho-

momorphism
M 0 - M - 0 - 0 - ...
f
? ? ? ?
- - - -
M 0 M0 M1 M2 ...
which induces an isomorphism on the level of cohomology modules. The reso-
lution is called injective if all modules M n are injective.
Every R-module admits a projective homological resolution (see 5.1/8),
which is easy to prove. However, the analogue for injective cohomological reso-
lutions is much more demanding and we will spend the remainder of the present
section to establish this result.
Proposition 3. Every R-module M admits an injective cohomological resolu-

tion M - M .
Lemma 4. For every R-module M , there exists an injection M - I into an

injective R-module I.
The proof of Proposition 3 is an easy consequence of the assertion of the

lemma. Indeed, start with an injection M - M 0 of M into an injective
R-module M 0 . Then choose an injection M 0 /M - M 1 into an injective
R-module M 1 so that we get the exact sequence
0 - M - M0 - M 1.
Embedding the cokernel of the map M 0 - M 1 into an injective R-module

2
M and continuing this way, we obtain step by step an injective cohomological
resolution of M .
To carry out the proof of Lemma 4, we need some preparations.
Proposition 5. For an R-module I the following conditions are equivalent:

(i) I is injective.
(ii) Given an ideal a R, every R-linear map a - I extends to an
R-linear map R - I.
Moreover, if R is a principal ideal domain, conditions (i) and (ii) are equiv-
alent to
(iii) I is divisible in the sense that for a R {0} and x I there is always
an element x I such that x = ax .
Proof. The assertion (i) = (ii) is trivial; just apply the dening property of
injective modules to the injection a - R.
Conversely, assume (ii) and consider an injection of R-modules M - M

as well as an R-linear map f : M - I. In order to extend f to M , look at
the set of all pairs (M , f ) where M M is a submodule containing M and

where f: M - I is an R-linear extension of f . Applying Zorns Lemma, this
set contains a maximal element (M , f ) and we claim that M = M . To justify
this we proceed indirectly and assume there is an element y M M . Then
consider the set
a = {r R ; ry M },
which apparently is an ideal in R satisfying
M Ry = ay.
Now, using (ii), the R-linear map
a - I, r - f (ry),
admits an extension g : R - I. Writing x = g(1), we have f (ry) = g(r) = rx

for all r a. In particular, if there is some r R such that ry = 0, then r a
because 0 M and, thus, rx = 0. Therefore we see that
g : Ry - I, ry - rx,
is a well-dened R-linear map coinciding with f on M Ry = ay.

From this we conclude that f : M - I can be extended to an R-linear
map
f : M + Ry - I, z + ry - f (z) + g(ry).
Indeed, f is well-dened, since z, z M and r, r R with z + ry = z + r y
imply z z = (r r)y M Ry and, thus,

f (z) + g(ry) f (z ) + g(r y) = f (z z ) g (r r)y = 0.
However, M M + Ry contradicts the maximality of (M , f ). Hence, we must

have M = M and f : M - I is an extension of f : M - I. It follows that
I is injective.
Finally, if R is a principal ideal domain, let us show that conditions (ii) and
(iii) are equivalent. Indeed, for elements a R {0} and x I consider the
R-linear map
f : (a) - I, ra - rx;
the latter is well-dened, since R is an integral domain. If condition (ii) is given,
f admits an R-linear extension f : R - I and the image f (1) of the unit
element in R yields an element x I satisfying x = ax . Conversely, consider
an ideal a R, say a = (a), and look at an R-linear map f : (a) - I; the
latter is of type ra - rx for r R and some x I. To extend f to an
R-linear map R - I, we may assume a = 0. Then, if I is divisible in the
sense of (iii), there is an element x I such that x = ax and the R-linear map
f: R - I, r - rx ,
extends f . This shows that condition (ii) is satised.
As an application, we repeat that Q, as a divisible Z-module, is injective in

contrast to Z, which is not divisible. Moreover, we can immediately conclude
from Proposition 5:
Corollary 6. As a divisible Z-module, Q/Z is injective. Likewise, Z-modules of

type (Q/Z)X are injective for arbitrary index sets X.
Corollary 7. Every Z-module M can be embedded into an injective Z-module.
Proof. We may assume M = 0, since the zero module is injective itself. Then,
for any y M {0}, consider the submodule Zy M and choose a Z-linear
map
Zy - Q/Z, zy - zu,
where u Q/Z is the residue class of n1 if Zy Z/nZ with n > 0 and where
u = 0 can be chosen arbitrarily if Zy Z. Then extend Zy - Q/Z to a
Z-module homomorphism y : M - Q/Z using the injectivity of Q/Z and
look at the Z-linear map

M - Q/Z, z - (y (z))yM {0} ,
yM {0}
which is injective by its construction.
In particular, the assertion of Lemma 4 is now already clear for Z-modules

and it remains to generalize our argument to the case of arbitrary rings R in
place of Z. In the following, let M be a module over a ring R and let G be
an abelian group, which we will view as a Z-module. Then HomZ (M, G) is an
R-module if we dene the product a for a R and HomZ (M, G) by
(a)(y) = (ay) where y M .
Lemma 8. The canonical map

: HomR M, HomZ (R, G) - HomZ (M, G)

- y - ((y))(1) ,
is an isomorphism of abelian groups.
Proof. By its denition, is additive. To exhibit an inverse of , look at the

additive map
: HomZ (M, G) - HomR (M, HomZ (R, G)),

- y - (a - (ay)) .
The composition is given by

- y - ((y))(1)

- y - (a - ((ay))(1)) = ,
because we have ((ay))(1) = (a(y))(1) = (y)(a 1). Since the composition

is given by

- y - (a - (ay))

- (y - (y)) = ,
we see that and are mutually inverse to each other.
Lemma 9. Let G be an abelian group which in terms of Z-modules is injective.

Then E = HomZ (R, G) is an injective R-module.
Proof. Let M - M be an injection of R-modules and f : M - E an

R-linear map. Then, by the above lemma, f corresponds to a Z-linear map
f : M - G. Using the injectivity of G, the latter extends to a Z-linear map
f : M - G which, again by the lemma, corresponds to an R-linear map
f: M - E. Since f is an extension of f , we see that f is an extension of f
and it follows that E is injective.
Now we can easily nish the proof of Lemma 4. Let M be an R-module. As

in the proof of Corollary 7, we choose a non-trivial Z-linear map Zy - Q/Z
for each y M {0} and extend it to a Z-linear map M - Q/Z, using
Corollary 6. Furthermore, applying Lemma 8 to these maps yields R-linear
maps y : M - HomZ (R, Q/Z) that satisfy y (y) = 0 for all y M {0}.
But then

M - HomZ (R, Q/Z), y - (y (y))yM {0} ,
yM {0}
is an injective R-linear map into an injective R-module. Indeed, the factors

HomZ (R, Q/Z) on the right-hand side are all injective, due to Lemma 9, since
Q/Z as a Z-module is injective.
Exercises
1. Show that any cartesian product of injective R-modules is injective.
2. For a Noetherian ring R show that any direct sum of injective R-modules is
injective. Is this true as well without the Noetherian hypothesis?
3. Let M be an R-module. Show that every injective submodule of M is a direct
summand of M .
5.4 The Ext Modules 187
4. Let M, N be Z-modules where M is injective and N is a torsion module in the

sense that for every x N there is an integer n Z {0} satisfying nx = 0.
Show that M Z N = 0.
5. Consider the so-called p-quasi-cyclic group Z(p ) = Zp1 /Z for a prime num-
ber p. Show that it is divisible and, hence, injective as a Z-module.
6. Let M be a at and N an injective R-module. Show that HomR (M, N ) is an
injective R-module.
5.4 The Ext Modules
In the following we want to introduce Ext functors as right derived functors of

the Hom functor. To do this, let M, N be two R-modules. Choosing a projective
homological resolution M - M of M , we can apply the functor HomR (, N )
to it. Since the functor is contravariant and additive, it transforms M into a
cochain complex
HomR (M , N ) : 0 - HomR (M0 , N ) - HomR (M1 , N ) - ...
and we can dene
ExtnR (M, N ) = H n (HomR (M , N )), n N,
as the nth Ext module associated to M and N . Of course, we have to check
that ExtnR (M, N ) is well-dened. Proceeding as in the proof of 5.1/10, consider
a second projective homological resolution M - M of M . Then we conclude
from 5.1/9 that the chain complexes M and M are homotopy equivalent.
As an additive functor, HomR (, N ) transfers this equivalence into a homotopy
equivalence between HomR (M , N ) and HomR (M , N ). Indeed, as the latter
complexes are cochain complexes, we adapt the denition of a homotopy, known
from 5.1/3 for chain complexes, to our situation as follows:
Denition 1. Let f, g : C - D be homomorphisms of cochain complexes.

A homotopy between f and g is a module homomorphism C - D of degree
1, in other words, a system of R-module homomorphisms h : C n
n - D n1 ,
such that the maps of the diagram
d
... - C n1 - Cn - C n+1 - ...
hn1 hn hn+1 hn2
f g f g
? ? ?

d

... - D n1 - Dn - D n+1 - ...
satisfy the relation f g = hd+dh. Just as in the setting of chain complexes,

f and g will be called homotopic in this case.
Notice that the above diagram coincides with the one given in 5.1/3, except
for the fact that in chain complexes passing through arrows of boundary maps
decreases module indices whereas the contrary is the case in cochain complexes.
In any case, we thereby see that the Ext modules are well-dened. In particular,
we have
Ext0R (M, N ) = HomR (M, N ),
since HomR (, N ) is left exact.
Alternatively, we can take an injective cohomological resolution N - N
of N and consider the cohomology modules
Ext nR (M, N ) = H n (HomR (M, N )), n N.
Also in this case, we must check that these modules are well-dened. This can
be done similarly as for projective homological resolutions. First we need to
prove an analogue of 5.1/9 for homomorphisms of cochain complexes show-
ing that two injective cohomological resolutions N - N and N - N
of N are homotopy equivalent; this is achieved by dualizing the arguments
used in the proof of 5.1/9. Then we use the fact that the covariant additive
functor HomR (M, ) transfers such an equivalence into a homotopy equivalence
between HomR (M, N ) and HomR (M, N ). Thus, indeed, the cohomology mod-
ules Ext nR (M, N ) are well-dened.
In a next step we can compare the modules ExtnR (M, N ) and Ext nR (M, N ).
To do this, we set up the double complex induced from M - M and
N -
N as follows:
.. .. ..
. . .
6 6 6
0 - HomR (M2 , N 0 ) - HomR (M2 , N 1 ) - HomR (M2 , N 2 ) - ...

6 6 6
d

d
6 6 6

6 6 6
0 0 0
The associated single complex HomR (M , N ) is given by

(HomR (M , N ))n = HomR (Mp , N q ), n N,
p+q=n
together with boundary maps dn = d + d where d diers from d by the sign
(1)p on the row with index p. Then we see:
Proposition 2. The canonical complex homomorphisms
HomR (M , N ) - HomR (M , N ) HomR (M, N )
induce isomorphisms
H n (HomR (M , N )) - H n (HomR (M , N )) H n (HomR (M, N ))
for all n N. In particular, ExtnR (M, N ) can be dened by any of these coho-
mology modules.
The proof is done in analogy to the proof of 5.2/3, by dualizing arguments.

Remark 3. Let M, N be R-modules. Then:

(i) Ext0R (M, N ) = HomR (M, N ).
(ii) ExtnR (M, N ) = 0 for n > 0 if M is projective or N is injective.
Proof. The rst assertion has already been mentioned; it follows from the left
exactness of the Hom functor. To verify the second one, we just note that M
given by 0 -M - 0 yields a projective resolution M - M if M is
projective and that, likewise, N given by 0 - N - 0 yields an injective
resolution N - N if N is injective. Then, applying Proposition 2, we are
done.
Next, let us discuss the so-called long exact Ext sequences.
Proposition 4 (Long exact Ext sequences). Let M , N be R-modules. Then

0 - M - M - M - 0
induces a long exact Ext sequence

0 - HomR (M , N ) - HomR (M, N ) - HomR (M , N )

- Ext1R (M , N ) - Ext1R (M, N ) - ...
and every short exact sequence of R-modules
0 - N - N - N - 0
a long exact Ext sequence

0 - HomR (M, N ) - HomR (M, N ) - HomR (M, N )

- Ext1R (M, N ) - Ext1R (M, N ) - ... .
Proof. The rst Ext sequence can be constructed as explained in the proof of
5.1/12, by setting up suitable projective resolutions for M , M , M , applying the
functor HomR (, N ), and using 5.1/1. For the second Ext sequence we proceed
in a similar way, replacing projective by injective resolutions and dualizing
the arguments given in the proof of 5.1/12.
There are two essentially dierent long exact Ext sequences. This corre-
sponds to the fact that Ext can be viewed as a contravariant functor in the rst
and as a covariant functor in the second variable of the Hom functor. In prin-
ciple, a similar situation is faced when dealing with tensor products. However,
in contrast to Hom, the tensor product functor is covariant in both variables
and, furthermore, is symmetric in the sense that the functors R E and E R
are canonically equivalent for any R-module E. This is why the long exact Tor
sequences, as mentioned in 5.2/2 and 5.2/5, are essentially the same.
Corollary 5. For an R-module P the following conditions are equivalent:

(i) P is projective.
(ii) ExtnR (P, N ) = 0 for n > 0 and all R-modules N .
(iii) Ext1R (P, N ) = 0 for all R-modules N .
Proof. The implication (i) = (ii) follows from Remark 3 whereas (ii) = (iii)
is trivial.
To verify (iii) = (i), consider an exact sequence of R-modules
0 - N - N - N - 0,
as well as the associated long exact Ext sequence

0 - HomR (P, N ) - HomR (P, N ) - HomR (P, N )

- Ext1R (P, N ) - Ext1R (P, N ) - ... .
Then, using (iii), the functor HomR (P, ) is exact and we see that P is projective
according to 5.3/2 (i).
Corollary 6. For an R-module I the following conditions are equivalent:

(i) I is injective.
(ii) ExtnR (M, I) = 0 for n > 0 and all R-modules M .
(iii) Ext1R (M, I) = 0 for all R-modules M .
Proof. We use the same arguments as in the proof of Corollary 5, although

from the dual point of view. Again, the implication (i) = (ii) follows from
Remark 3 and (ii) = (iii) is trivial.
To verify (iii) = (i) consider an exact sequence of R-modules
0 - M - M - M - 0
and the associated long exact Ext sequence

0 - HomR (M , I) - HomR (M, I) - HomR (M , I)

- Ext1R (M , I) - Ext1R (M, I) - ... .
Using (iii), we see that the functor HomR (, I) is exact. Therefore I is injective,
due to 5.3/2 (ii).
Exercises
1. Show

ExtnR Mi , N ExtnR (Mi , N ), ExtnR M, Nj ExtnR (M, Nj )
iI iI jJ jJ
for R-modules M, Mi and N, Nj .

2. Let M, N be modules over a principal ideal domain R.
(a) Show that ExtnR (M, N ) = 0 for n 2.
(b) Determine ExtnR (R/(a), M ) for any element a R.
3. Let M, N be R-modules. Show:
(a) If Ext1R (M, N ) = 0 for all R-modules M , then N is injective.
(b) If Ext1R (M, N ) = 0 for all R-modules N , then M is projective.
4. Let M, N be R-modules where R is Noetherian and M is of nite type. Show
for any multiplicative subset S R that there are canonical isomorphisms
ExtnR (M, N )S ExtnRS (MS , NS ), n N. Hint: Use Exercise 4.3/9.
5. Group cohomology: Let G be a group and M a ZG-module (see Exercise 5.2/5).
For n N the nth cohomology group of G with values in M is dened by
H n (G, M ) = ExtnZ G (Z, M ). Compute H n (G, Z) for a cyclic group G.
6. Ext and extensions of modules: Let M, N be R-modules. An extension of M
by N is an exact sequence of R-modules 0 - N - E - M - 0,

simply denoted by E. Two such extensions E, E are called isomorphic if there is
a commutative diagram
0 - N - E - M - 0

?
0 - N - E - M - 0
where E - E is necessarily an isomorphism. Let Ext(M, N ) be the set of

isomorphism classes of such extensions.
Show that there is a bijection Ext(M, N ) - Ext1R (M, N ) as follows. Choose

any exact sequence 0 - K - P - M - 0 with P being projective
(or free). Then there is a commutative diagram of type

0 - K - P - M - 0

? ?
0 - N - E - M - 0
where exists by the projectivity of P and is induced from . Now look at

the long exact Ext sequence associated to the upper row and the module N . It
contains a morphism of R-modules Hom(K, N ) - Ext1 (M, N ); the latter is
R
1
surjective as ExtR (P, N ) = 0, due to the projectivity of P . Show that associating
to E Ext(M, N ) the image of Hom(K, N ) in Ext1R (M, N ) yields a well-
dened map Ext(M, N ) - Ext1R (M, N ).
Conversely, starting with an element Ext1R (M, N ), choose a preimage
Hom(K, N ) with respect to Hom(K, N ) - Ext1 (M, N ) and let E be
R
the amalgamated sum of P and N over K, namely the quotient (P N )/K,
where K is the image of the mapping K - P N, x - (x) (x). Then
we get a commutative diagram

0 - K - P - M - 0

? ?
0 - N - (P N )/K - M - 0
where the lower row usually is referred to as the push-out under of the upper
one. In any case, show that associating to Ext1R (M, N ) the lower row yields
a well-dened map Ext1R (M, N ) - Ext(M, N ).
Show that both maps above are mutually inverse to each other and, thus, dene
a bijection Ext(M, N ) - Ext1R (M, N ). Does this bijection depend on the
choice of the exact sequence 0 - K - P - M - 0?
Give a construction on the level of extensions of M by N that corresponds to the
addition of elements in Ext1R (M, N ).
6. Ane Schemes and Basic Constructions
The rst step in scheme theory is to explain the construction of ane schemes,
namely schemes of type Spec A for a ring A. Such schemes serve as the local
parts from which more general global schemes are obtained via a gluing process.
As we have pointed out already in the introductory section above, Spec A is a
ringed space, i.e. a topological space with a sheaf of rings on it. Let us rst
discuss Spec A as a topological space. The underlying point set consists of the
prime spectrum of A, the set of all prime ideals in A. Furthermore, a subset
Y Spec A is said to be closed if there exists an ideal a A such that
Y = V (a) := {p Spec A ; a p},
and open if its complement is closed in Spec A. We will show in 6.1/1 and 6.1/2
that the open (resp. closed) subsets really dene a topology on Spec A, namely
the so-called Zariski topology. For example, taking A = Z, every prime ideal
p = 0 in Z is maximal and therefore gives rise to a closed subset {p} Spec Z.
More generally, a subset Y Spec Z is closed if and only if it equals Spec Z or
consists of (at most) nitely many closed points. Hence, we can conclude that
the intersection U U of two non-empty open subsets in Spec Z will never be
empty. This shows that the Zariski topology on a prime spectrum Spec A does
not satisfy the Hausdor separation axiom, except for some more or less trivial
cases. However, it is easily seen that Spec A is a Kolmogorov space: given two
dierent points x, y Spec A, at least one of them admits an open neighborhood
not containing the other; see 6.1/8.
The topology of such a Kolmogorov space X = Spec A can be quite patho-
logical. To mention a particular phenomenon that will occur, consider an open
subset U X and a closed point x U in the sense that {x} U is a closed
subset with respect to the topology induced from X on U . So x could be called
a locally closed point of X. If X would satisfy the Hausdor separation axiom,
such a point would automatically be closed in X, since all points of a Hausdor
space are closed. However, in our case where X is just a Kolmogorov space it
can happen, indeed, that x is not closed in X. For example, consider a discrete
valuation ring in the sense of 9.3/3 such as the ring A = Zp , the localization of
Z at a non-zero prime ideal p Z. Then X = Spec Zp contains just two points,
namely the generic point given by the zero ideal 0 Zp and the special point
s given by the maximal ideal pZp Zp . Since s is closed in X, we see that
U = {} will be open in X. Furthermore, U is closed in U , but clearly not

202 6. Ane Schemes and Basic Constructions
closed in X, since the closure of {} is all of X. However, such a phenomenon

cannot happen if A is of nite type over a eld, i.e. a quotient by some ideal of
a polynomial ring in nitely many variables over a eld; see 8.3/6.
Returning to the case of a general ring A, consider an ideal a A generated
by a family (fi )iI of elements in
A. Then the associated Zariski closed subset
V (a) Spec A satises V (a) = iI V (fi ). In particular, passing to comple-
ments in Spec A, we see that any open subset in Spec A is a union of sets of
type D(f ) = Spec A V (f ) for f A. Therefore the latter sets form a basis of
the Zariski topology on Spec A and are referred to as the basic open subsets in
Spec A.
Now x an element f A and let Af be the localization of A by the
multiplicative system that is generated by f . Then we know from 1.2/6 that
the canonical morphism A - Af gives rise to a bijection
Spec Af - D(f ) Spec A, q - q A.
Furthermore, it is not hard to see that this bijection is, in fact, a homeomorphism
with respect to Zariski topologies when the open set D(f ) is equipped with the
restriction of the Zariski topology on Spec A; see 6.2/8. In particular, D(f )
can canonically be identied with Spec Af and this fundamental fact makes it
possible to interpret Af as the ring of functions on D(f ), just as A is the ring
of functions on Spec A. The canonical morphism A - Af from A into its
localization Af plays the role of a restriction morphism, restricting functions on
Spec A to D(f ). More generally, using the universal property of localizations,
any inclusion D(g) D(f ) gives rise to a well-dened restriction morphism
Af - Ag , since the restriction of f A to D(g) is seen to be invertible.
Proceeding like this, a little bit of care is necessary, since a basic open set
D(f ) does not determine its dening element f uniquely. For example, we would
have D(f ) = D(ef n ) for any unit e A and any exponent n > 0. However,
on the level of localizations, the problem does not persist any more. Indeed, if
D(f ) = D(f ), we will see in 6.3, example (4), that the localizations Af and
Af are canonically isomorphic. Furthermore, we show that functions on Spec A
can be dened locally with respect to any open covering of Spec A by basic
open subsets D(fi ) Spec A, i I. More precisely, given elements hi Afi on
D(fi ) that coincide on all overlaps D(fi ) D(fj ) = D(fi fj ), there is a unique
element h A restricting to hi on D(fi ) for all i I. In other words, the
functor D(f ) - Af satises the properties of a sheaf. All this is dealt with in
Section 6.6, where we extend the functor D(f ) - Af to a sheaf with respect
to the Zariski topology on Spec A, its so-called structure sheaf, and thereby
construct Spec A as a ringed space, called the ane scheme associated to A.
For the convenience of the reader, the necessary technical tools such as inductive
and projective limits as well as the technique of sheacation are included in 6.4
and 6.5. Needless to say, the notion of ane schemes allows a straightforward
globalization: a scheme is a ringed space such that each of its points admits an
open neighborhood looking like an ane scheme.
6.1 The Spectrum of a Ring 203
Finally, the notion of a morphism of schemes needs some special atten-

tion. Roughly speaking, one admits only those morphisms of ringed spaces
: X - Y such that the inherent pull-back of functions from Y to X takes
a local function vanishing at a point (x) Y to a local function vanishing at
x X; see 6.6/8 and 6.6/11. Proceeding like this, we show in 6.6/9 that the
set of morphisms Spec A - Spec B between two ane schemes corresponds
bijectively to the set of ring morphisms B - A, a result that is quite essential
from the viewpoint of solutions of polynomial equations, as we have explained
in the introductory section above.
There is some additional material in the present chapter. Namely, we gen-
eralize the construction of the structure sheaf on ane schemes Spec A to
A-modules replacing the ring A, thereby arriving at the notion of quasi-coherent
modules on schemes. Furthermore, the basics of direct and inverse images of
module sheaves are explained in 6.9, including the result 6.9/9 on the quasi-
coherence of direct images.
6.1 The Spectrum of a Ring
Let A be a ring; as before, all rings are assumed to be commutative and to admit
a unit element 1. In Section 1.1 we have already considered the set Spec A of
all prime ideals in A; the latter is called the spectrum or, in more precise terms,
the prime spectrum of A. For the purposes of geometry, the spectrum of a ring
is viewed as a point set and, to underline this point of view, notations like
x Spec A are used for the elements of spectra. However, in situations where it
is more advisable to imagine such points as prime ideals and, thereby, as subsets
in A, it is common practice to use an ideal-like notation such as px instead of
x. Thus, given any point x Spec A, we will use px as a second notation, when
we would like to go back to the original meaning of x being an ideal in A.
For any x Spec A, the localization Ax = Apx is a local ring with maximal
ideal mx = px Ax ; see 1.2/7. It is called the local ring of A at x. Furthermore,
the eld of fractions k(x) = Q(A/px ), which coincides with Ax /mx (see Ex-
ercise 1.2/5), is called the residue eld of A at x; it is related to A via the
canonical maps A - A/px - k(x).
Recall that the elements of A can be interpreted as functions on the spec-
trum Spec A. Just dene f (x) for f A and x Spec A as the residue class of
f in A/px k(x). Thereby every f A determines a map

Spec A - A/px k(x),

xSpec A xSpec A
where f (x) = 0, for some x Spec A, is equivalent to f px and, likewise,

f (x) = 0 to f px . Since px is a prime ideal, an equation (f g)(x) = 0 for two
functions f, g A and a point x Spec A is equivalent to f (x) = 0 or g(x) = 0.
Next consider a subset E A and the associated ideal a = (E) A
generated by E in A. Then
rad(E) = rad(a) = {f A ; f n a for some n N}
is called the nilradical of a or E; cf. 1.3/5. Moreover we set
V (E) = {x Spec A ; f (x) = 0 for all f E}

= {x Spec A ; E px }
and call this the zero set of E; it is also referred to as the variety of E, which
explains the usage of the letter V . For f A we will apply the notations
V (f ) = {x Spec A ; f (x) = 0},

D(f ) = Spec A V (f ) = {x Spec A ; f (x) = 0}
for the zero set of f and its complement in Spec A. The latter is also known as
the domain of f , which explains the usage of the letter D. If we want to keep
track of the ambient spectrum X = Spec A, we write more specically DX (f )
instead of D(f ).
Proposition 1. Consider subsets E, E , and a family of subsets (E ) of a

ring A. Then:
(i) V (0) = Spec A, V (1) = .

(ii) E E = V (E) V (E ).
(iii) V ( E ) = V ( (E )) = V (E ).
(iv) V (EE ) = V (E) V (E ) where EE = {f f ; f E, f E }.
(v) V (E) = V (rad(E)).

Proof. Assertions (i) and (ii) are more or less obvious.
Next, the set V ( E )
in (iii) consists of all points x Spec A where E px , hence, where
E px for all , and therefore coincides with V (E ). Since E
is contained in p x for some x Spec A if and only if the ideal (E )
generated by E is contained in px , we see that, in addition, V ( E )
coincides with V ( (E )).
Turning to assertion (iv), look at a point x Spec A such that x V (E)
and x V (E ). Then we have E px and E px , and there are elements
f E and f E such that f, f px . But then f f px since px is a prime
ideal, and therefore x V (EE ) so that V (EE ) V (E) V (E ). To show the
reverse inclusion, look at a point x V (E), hence, satisfying E px . Then,
clearly, EE px and therefore x V (EE ). This proves V (E) V (EE ) and,
likewise, V (E ) V (EE ).
Finally, assertion (v) is easy to verify. Indeed, E px is equivalent to
rad(E) px , due to the fact that px is a prime ideal.
Corollary 2. Let A be a ring and X = Spec A its spectrum. There exists a

unique topology on X, the so-called Zariski topology, whose closed sets consist
precisely of the subsets of type V (E) X for some subset E A.
Furthermore:
(i) The sets of type D(f ) for f A are open and satisfy the intersection
relation D(f ) D(f ) = D(f f ) for f, f A.
(ii) Every open subset of X is a union of sets of type D(f ). In particular,
the latter form a basis of the Zariski topology on X (which is the reason why
the sets of type D(f ) are referred to as the basic open subsets of X).
Before giving some indications on the proof of the corollary, let us recall the
denition of a topology.
Denition 3. Let X be a set. A topology on X consists of a set T of subsets

in X such that:
(i) , X T .
(ii) If (U ) is a family of elements in T , then U T .
(iii) If (U ) is a nite family of elements in T , then U T .
In the situation of the denition, the elements of T are called the open
subsets of X. Furthermore, a subset V X is called closed if its complement
X V is open. Thus, passing to the complements of open sets, a topology can
alternatively be characterized by the properties of its closed subsets. Doing so,
one has to consider a set T of subsets of X such that:
(i ) X, T .
(ii ) If (V ) is a family of elements in T , then V T .
(iii ) If (V ) is a nite family of elements in T , then V T .
After these explanations, the proof of Corollary 2 can be achieved quite

easily. The assertions (i), (iii), and (iv) of Proposition 1 yield the characterizing
properties for the closed sets of a topology, assuming that we generalize (iv)
by induction to nite unions of sets of type V (E). The sets of type D(f ) for
functions f A, being complements of the sets of type V (f ), are open and
satisfy the claimed intersection property due to Proposition 1 (iv). Finally, an
arbitrary open subset U X = Spec A is the complement of a closed set of
type V (E) for a subset E A. Therefore

V (E) = V (f )
f E
and, hence,
U= D(f ),
f E
which says that every open subset in X is a union of subsets of type D(f ).
Let us consider some examples. The zero ring 0 does not contain any prime
ideal. Therefore its spectrum is empty.
Next, let A be a principal ideal domain, for example A = Z or A = Kt
where K is a eld and t a variable. Then the spectrum X = Spec A consists of
the zero ideal 0 A and of all principal ideals (p) A that are generated by
prime elements p A. Furthermore, the closed subsets of X are of type V (a),
for elements a A. In particular, V (a) = X for a = 0 and V (a) = if a is a unit
in A. In all other cases a admits a non-trivial prime factorization a = p11 . . . prr
with pairwise coprime prime factors pi , exponents i > 0, and a unit A .
A prime ideal (p) A belongs to V (a) if and only if (a) (p), i.e. if and
only if the prime element p divides a. However, the latter can only be the case
if p is associated to one of the prime factors p1 , . . . , pr . In particular, all prime
ideals which are generated by prime elements or, equivalently, all non-zero prime
ideals, give rise to closed points in X = Spec A. Furthermore, the consideration
shows that a subset V Spec A is closed if and only if it coincides with X or ,
or if it consists of nitely many closed points, the latter corresponding to prime
elements in A. Therefore the zero ideal 0 A yields a dense point in X, in the
sense that its closure {0} coincides with X. In particular, if A is not a eld and,
hence, X = {0}, the point 0 X cannot be closed.
Switching to complements, we obtain and X as open sets in X, as well as
the sets of type X {x1 , . . . , xr }, for nitely many closed points x1 , . . . , xr X.
Thus, any non-empty open subset of X will contain the point given by the zero
ideal 0 A, and we thereby see that the Zariski topology on X = Spec A
cannot satisfy the Hausdor separation axiom, unless A is a eld.
Now let us work again with the spectrum X = Spec A of an arbitrary ring
A. So far we have looked at zero sets in X of type V (E) for a subset E A, or
V (a) for an ideal a A. We want to introduce another construction which, in
a certain sense, is an inverse of the mapping V (). Doing so, we associate to a
subset Y X the ideal

I(Y ) = {f A ; f (y) = 0 for all y Y } = py
yY
of all functions in A vanishing on Y ; notice that f py is equivalent to f (y) = 0

and that this implies I({y}) = py for all y X.
Remark 4. Let A be a ring.

(i) Let Y Y
Spec A be subsets. Then I(Y ) I(Y ).
(ii) I( Y ) = I(Y ) for any family (Y ) of subsets in Spec A.
Proof. The assertions follow immediately from the denition of the mapping
I().
Proposition 5. Let A be a ring.

(i) I(V (E)) = rad(E) for any subset E A. In particular, I(V (a)) = a for
ideals a A satisfying a = rad(a).
(ii) Let Y Spec A be a subset. Then V (I(Y )) coincides with the closure
Y of Y with respect to the Zariski topology on Spec A. In particular, we have
Y = V (I(Y )) for closed subsets Y A.
Proof. In the situation of (i), let a be the ideal generated by E. Then

I V (E) = py = p= p
yV (E) pSpec A pSpec A
Ep ap
and, furthermore,
p = rad(a) = rad(E)
pSpec A
ap
by 1.3/6.
To justify (ii), observe rst that Y V (I(Y )), since the ideal I(Y ) of all
functions vanishing on Y admits a zero set which must contain Y . In particular,
the closure Y of Y in Spec A will satisfy Y V (I(Y )), since Y is dened as the
smallest closed subset in Spec A containing Y , i.e. as the intersection of all sets
of type V (E) Spec A such that Y V (E). To show Y = V (I(Y )) it remains
to check that Y V (E) implies V (I(Y )) V (E). Indeed, from Y V (E) we
conclude f (y) = 0 for all y Y and all f E and, hence, E I(Y ). Then
Proposition 1 (ii) yields V (I(Y )) V (E), as desired.
Corollary 6. Let A be a ring and X = Spec A its spectrum.

(i) For x X we have {x} = V (px ). Therefore the closure of a point x X
consists of all points y X such that px py .
(ii) A point x X is closed in X if and only if the ideal px is maximal in
A.
Corollary 7. Let A be a ring and X = Spec A its spectrum. Then the mappings
I -
closed ideals a A
,
subsets in X V satisfying a = rad(a)
Y - I(Y ),
V (a) a,
are inclusion-reversing, bijective, and mutually inverse to each other. Further-

more,
I Y = I(Y ), I Y = rad I(Y )

for a family (Y ) of subsets in X, where for the second equation the subsets
Y are required to be closed in X.
Proof. In view of Proposition 5, all assertions can easily be derived, except

possibly for the second equation, which we will justify in more detail. Since
Y = V (I(Y )) by Proposition 5 (ii), we get from Proposition 1 (iii)

Y = V I(Y )

and therefore
I Y = rad I(Y )

by Proposition 5 (i).
The Zariski topology on the spectrum Spec A of a ring A does not necessarily
satisfy the Hausdor separation axiom, as we have seen above by looking at
principal domains A. However, a weaker version of this axiom holds.
Remark 8. The Zariski topology on X = Spec A yields a Kolmogorov space,

i.e. a topological space satisfying the following separation axiom:
Given two dierent points x, y X, there exists an open neighborhood U of
x such that y U or an open neighborhood U of y such that x U .
Proof. Note that x and y being dierent means px = py , and this is equivalent
to px py or py px . In the rst case we have y V (px ) = {x}. Then X {x}
is an open neighborhood of y which does not contain x.
We have already seen in Corollary 2 that the subsets of type D(f ) X

for elements f A form a basis of the Zariski topology on X. Furthermore,
the equation D(f ) D(f ) = D(f f ) shows that this class is closed under nite
intersection. Of course, a function f A is not uniquely determined by the
subset D(f ) X it denes, except for some rare cases. But we know from
Proposition 5 that V (f ) = V (f ) is equivalent to rad(f ) = rad(f ). Thereby we
see:
Remark 9. The following conditions are equivalent for functions f, f A:

(i) D(f ) = D(f ).
(ii) rad(f ) = rad(f ).
Proposition 10. Let A be a ring and X = Spec A its spectrum. Then, for
any g A, the associated subset D(g) X is quasi-compact. In particular,
X = D(1) is quasi-compact.
Proof. We will see later in 6.2/8 that D(g) with its topology induced from X is
canonically homeomorphic to Spec Ag , the spectrum of the localization of A by
g. Therefore it would be enough to restrict ourselves to the case where g = 1.
However, as this does not really make things easier, we will not proceed like this
and work with a general g instead. Also recall that the notion of quasi-compact
means compact, but without the Hausdor separation axiom. Therefore a subset
U of a topological space X is called quasi-compact if every open covering of U
admits a nite subcover.
Keeping in mind that the sets of type D(f ) form a basis of the Zariski
topology on X, it is enough to show that every covering of D(g) by means of sets
of type D(f ) admits a nite subcover. To check this, consider a family (f )

of elements in A such that D(g) D(f ). Switching to complements in
X yields

V (g) V (f ) = V (a),

where a A is the ideal generated by the elements f . From this we obtain

rad(g) rad(a) by Proposition 5 and there is some integer n N such that
g n a. Since a is generated by the elements f , there are
indices 1 , . . . , r
as well as coecients a1 , . . . , ar A such that g n = ri=1 ai fi . This implies
(g n ) (f1 , . . . , fr ), hence,
V (g) = V (g n ) V (f1 , . . . , fr )
and therefore

r
D(g) D(fi ).
i=1
Thus, the covering (D(f )) of D(g) admits a nite subcover, as claimed.
Proposition 11. Let A be a ring, a A an ideal, and : A - A/a the

canonical surjection. Then the map
a
: Spec A/a - Spec A, p - 1 (p),
induces a homeomorphism of topological spaces Spec A/a - V (a) where

V (a) is equipped with the subspace topology obtained from the Zariski topology
of Spec A.
Corollary 12. For a ring A its spectrum and the spectrum of its reduction
A/rad(A) are canonically homeomorphic.
Proof of Corollary 12. Recall that rad(A), the nilradical of A, is dened as the
nilradical of the zero ideal 0 A. Therefore V (rad(A)) = V (0) = Spec A, and
Proposition 11 applies.
Proof of Proposition 11. First observe that the map a denes a bijection be-
tween Spec A/a and the subset of Spec A consisting of all prime ideals containing
a, hence a bijection Spec A/a - V (a). Taking this map as an identication,
we obtain
Spec A D(f ) V (a) = D((f )) Spec A/a
for elements f A. Since is surjective, the sets of type D(f ) Spec A/a for
f A/a correspond bijectively to the restrictions V (a) D(f ) Spec A for
f A, and this justies the assertion.
Finally, let us consider decompositions of spectra into certain closed subsets.
Denition 13. A topological space X is called irreducible if :

(i) X =
(ii) Given any decomposition X = X1 X2 into closed subsets X1 , X2 , then
X1 = X or X2 = X.
Furthermore, a subset Y of a topological space X is called irreducible if Y
is irreducible under the topology induced from X on Y .
Lemma 14. For a non-empty topological space X the following conditions are
equivalent:
(i) X is irreducible.
(ii) If U1 , U2 X are open and non-empty, then U1 U2 is non-empty.
(iii) If U X is open and non-empty, then U is dense in X, i.e. the closure
U of U in X coincides with X.
(iv) If U X is open, it is connected, i.e. U is not a disjoint union of two
non-empty open subsets of X.
Proof. Let us assume condition (i) and derive (ii). Consider U1 , U2 X open
and non-empty. Then, if U1 U2 were empty, we would get the decomposition
X = (X U1 ) (X U2 ) into the proper closed subsets X U1 , X U2 X.
However, this contradicts (i).
Next assume condition (ii) and let U X be open and non-empty. Then
U and X U are disjoint open subsets in X which, however, is only possible if
X U = and, hence, X = U . Therefore we get (iii).
Assuming (iii), let U X be open. If U is not connected, it is a disjoint
union of two non-empty open subsets U1 , U2 U . Then U1 U2 = implies
U 1 X U2 X , as X U2 is a closed subset in X containing U1 . However,
U1 should be dense in X by (iii). Therefore U must be connected, as claimed in
(iv).
Finally, assume (iv). To derive (i), consider a decomposition X = X1 X2
into closed subsets X1 , X2 . Then we obtain the decomposition
X (X1 X2 ) = (X X1 ) (X X2 )
of the open set X (X1 X2 ) into the disjoint open sets X X1 and X X2 .
Since X (X1 X2 ) is connected according to (iv), we get X X1 = or
X X2 = , hence, X1 = X or X2 = X and, thus, (i).
Proposition 15. Let A be a ring and X = Spec A its spectrum. The following
conditions are equivalent:
(i) X is irreducible as a topological space under the Zariski topology.
(ii) A/rad(A) is an integral domain.
(iii) rad(A) is a prime ideal.
Proof. Using Corollary 12, we may replace A by its reduction A/rad(A) and
thereby assume rad(A) = 0. Now let X be irreducible as in (i). If there are
elements f, g A {0} such that f g = 0, this yields the decomposition
X = V (0) = V (f g) = V (f ) V (g)
into the closed subsets V (f ), V (g) X. But X irreducible implies V (f ) = X or

V (g) = X. Furthermore, from V (f ) = X = V (0) we can conclude that rad(f )
coincides with rad(0) = rad(A) = 0 by Proposition 5 and, hence, that f = 0.
Likewise we get g = 0 from V (g) = X and it follows that A is an integral
domain, as claimed in (ii).
The equivalence between (ii) and (iii) being trivial, assume that A is an
integral domain and let X = X1 X2 be a decomposition of X into proper
closed subsets X1 , X2 X. Then Proposition 5 yields
0 = I(X) = I(X1 ) I(X2 ), I(X1 ) = 0 = I(X2 ).
Choosing fi I(Xi ) {0}, i = 1, 2, we get f1 f2 = 0, in contradiction to the

fact that A was assumed to be an integral domain. Therefore we see that X
must be irreducible, as claimed in (i).
Corollary 16. Let X = Spec A be the spectrum of a ring A and Y X a

closed subset. Then Y is irreducible if and only if I(Y ) is a prime ideal in A.
Proof. Writing a = I(Y ), we have Y = V (a) and a = rad(a) so that we can

use the homeomorphism Spec A/a - Y of Proposition 11. Therefore Y is
irreducible if and only if Spec A/a is irreducible, hence, using Proposition 15,
if and only if A/a is an integral domain. However, the latter is the case if and
only if a is a prime ideal in A.
Corollary 17. Let A be a ring and X = Spec A its spectrum. Then the mappings
I and V of Corollary 7 yield mutually inverse and inclusion-reversing bijections
I -
irreducible closed
prime ideals in A = Spec A,
subsets of X V
Y - I(Y ),
{y} py .
Proof. Given Corollary 7, the assertion is an immediate consequence of Corol-

lary 16, using the fact that V (py ) = {y} by Corollary 6.
Thereby we see that every irreducible closed subset Y Spec A contains

a unique point y such that Y = {y}. This point is known as the generic point
of Y , whereas the points of its closure {y} are referred to as specializations
of y. More precisely, a point x Spec A is called a specialization of a point
y Spec A if x {y}. The latter is equivalent to py px . For example, it

follows from Proposition 15 that Spec A for an integral domain A is irreducible
and therefore admits a unique generic point. The latter corresponds to the zero
ideal in A.
Exercises
1. Show for a ring A that every non-empty closed subset V Spec A contains a
closed point. Deduce that an open subset U Spec A containing all closed points
of Spec A must coincide with Spec A.
2. For an algebraically closed eld K consider the polynomial ring Kt in one
variable t. Show that the set of closed points in Spec Kt can be identied with
K and that there is precisely one non-closed point in Spec Kt, namely the
generic point.
3. For an algebraically closed eld K consider the polynomial ring Kt1 , t2 in vari-
ables t1 , t2 and set X = Spec Kt1 , t2 . Show:
(a) The set of closed points in X can canonically be identied with K 2 .
(b) The non-closed points of X that are dierent from the generic point are given
by the ideals of type (f ) Kt1 , t2 where f Kt1 , t2 is irreducible.
(c) The closure {y} of a point y as in (b) consists of y itself as the generic point
and of the curve {x K 2 ; f (x) = 0}.
4. Let X be a topological space, which is irreducible. Show that every non-empty
open subset U X is irreducible.
!n
5. Show that there is a canonical bijection Spec ni=1 Ai - i=1 Spec Ai for
given rings A1 , . . . , An ; see also Exercise 1.1/6.
6. Let A be an algebra of nite type over a eld and consider a closed subset
Y Spec A. Show that the closed points are dense in Y . Does this remain true
if we replace A by a localization of it?
7. Let A be an algebra of nite type over a eld. Show that Spec A is nite if and
only if A is of nite vector space dimension over K. Hint: A is Noetherian and,
hence, by 2.1/12, contains only nitely many minimal prime ideals. Use this to
reduce the assertion to the case where A is an integral domain. Then apply results
on integral dependence.
8. Let A be an algebra of nite type over a eld F . Provide the spectrum Spm A
of maximal ideals in A with the topology induced from the Zariski topology on
Spec A. Thus, a subset Y Spm A is open (resp. closed) if and only if it is of type
Y Spm A for some open (resp. closed) subset Y Spec A. Show that Spm A
satises the Hausdor separation axiom if F is a nite eld. Is the same true
without the assumption of F being nite?
6.2 Functorial Properties of Spectra
Let : A - A be a ring homomorphism. Then induces for any ideal p A

a monomorphism A/1 (p) - A /p. If p is a prime ideal, A /p is an integral
6.2 Functorial Properties of Spectra 213
domain and the same is true for A/1 (p). Therefore it follows that 1 (p) is
a prime ideal as well and we can state:
Proposition 1. Every ring homomorphism : A - A induces a map

a
: Spec A - Spec A, px - 1 (px )
between associated spectra.
We can be a bit more precise on this fact:
Lemma 2. In the situation of Proposition 1, there is the commutative diagram

A - A
? ?
x
A/1 (px ) - A /px

? x
?
k a (x) - k(x)
for any x Spec A . In particular, for f A and x Spec A we have

x f (a (x)) = (f )(x)
which, in a simplied way, can be read as f a = (f ). Therefore might be

interpreted as the map of composing functions f A with a .
Proof. Implicitly, the upper part of the diagram was used for the construction
of the map a : Spec A - Spec A in Proposition 1, whereas the lower part
is the canonical extension to residue elds, as introduced at the beginning of
Section 6.1.
Proposition 3. As before, consider a ring homomorphism : A - A and

its associated map a : Spec A - Spec A on spectra.
(i) Let E A be a subset. Then (a )1 (V (E)) = V ((E)).
(ii) If a A is an ideal, then a (V (a )) = V (1 (a )).
Proof. We start with assertion (i). The relation x (a )1 (V (E)) for points
x Spec A is equivalent to a (x) V (E) and, hence, to f (a (x)) = 0 for all
f E. Furthermore, it follows from Lemma 2 that f (a (x)) = 0 is equivalent
to (f )(x) = 0. Thus, x (a )1 (V (E)) is equivalent to x V ((E)).
In the situation of (ii) we apply 6.1/5 (ii) and thereby get a (V (a )) = V (a)
for the ideal a = I(a (V (a ))) A. Now f a for elements f A is equivalent
to f vanishing on a (V (a )), or, by Lemma 2, to (f ) vanishing on V (a ) and,
thus, by 6.1/5 (i), to (f ) rad(a ). Therefore we have a = 1 (rad(a )),
and it is easily checked that 1 (rad(a )) = rad(1 (a )). But then we have
V (a) = V (1 (a )) and, hence, a (V (a )) = V (1 (a )).
Corollary 4. In the situation of Proposition 3, let f A. Then

(a )1 D(f ) = D (f ) .
Proof. We have (a )1 (V (f )) = V ((f )) by Proposition 3 (i). Since the forma-

tion of preimages with respect to a is compatible with passing to complements,
the assertion follows.
As a direct consequence, we obtain the following fact:
Corollary 5. The map a : Spec A - Spec A associated to a ring homomor-

phism : A - A is continuous with respect to Zariski topologies on Spec A

and Spec A . Thereby we mean that the preimage of any open (resp. closed )
subset in Spec A is open (resp. closed ) in Spec A .
Next we want to study ring homomorphisms : A - A where the map

a
: Spec A - Spec A is injective and induces a homeomorphism between
Spec A and im a .
Proposition 6. Let : A - A be a ring homomorphism such that every ele-

ment f A is of type f = (f ) h with an element f A and a unit h A .
Then the map a : Spec A - Spec A is injective and denes a homeomor-
phism Spec A - im Spec A where Spec A and Spec A are equipped
a
with their Zariski topologies and im a with the subspace topology induced from
the Zariski topology on Spec A.
Proof. We start by looking at the injectivity of a . Let x, y Spec A satisfy

a
(x) = a (y) and, hence, 1 (px ) = 1 (py ). We claim that then px = py and,
thus, x = y. Indeed, given f px , there exist f A and h A such that
f = (f ) h. Then (f ) = f h1 px and, hence, f 1 (px ) = 1 (py ).
This implies (f ) py and therefore f = (f ) h py so that px py .
Interchanging the roles of x and y, we get px = py and, hence, x = y.
Recalling the denition of the subspace topology, a subset Y im a is
closed (resp. open) with respect to the subspace topology of im a if and only
if there is a closed (resp. open) set Y Spec A such that Y = Y im a .
Since the map a : Spec A - Spec A is continuous, it is immediately clear
that the induced bijection Spec A - im a is continuous as well. To verify
that the latter map is, indeed, a homeomorphism, it is enough to show that,
for any closed subset Y Spec A , there is a closed subset Y Spec A such
that Y = (a )1 (Y ). Such a Y is easy to construct. Indeed, if Y = V (E ) for
some subset E A , the elements of E can be adjusted by suitable units in
A in such a way that we may assume E (A). For E = (E) with a subset
6.2 Functorial Properties of Spectra 215
E A we get from Proposition 3 (i)

Y = V (E ) = V (E) = (a )1 V (E) = (a )1 (Y )
with Y = V (E), as desired.
There are two typical examples of ring homomorphisms : A - A where

the assumption of Proposition 6 is fullled, namely surjections and localizations.
We will discuss both cases separately.
Corollary 7. Let A be a ring and a A an ideal. Then the map

a
: Spec A/a - Spec A
associated to the projection : A - A/a denes a homeomorphism
Spec A/a - V (a) Spec A.
In this case, a is called a closed immersion of spectra.
Proof. The assertion was already shown in 6.1/11. Using im a = V (a) it can
alternatively be derived from Proposition 6.
Corollary 8. Let A be a ring and S A a multiplicative system. Then the

canonical homomorphism : A - AS induces a homeomorphism

a
: Spec AS - D(f ) Spec A.
f S
a a
If im is open in Spec A, we call an open immersion of spectra. For example,
the latter isthe case if S is generated by nitely many elements f1 , . . . , fr A,
since then f S D(f ) = D(f1 . . . fr ).
Proof. Applying Proposition 6 it remains only to determine the image im a .

To do this, we rely on the fact that taking inverse images with respect to
: A - AS yields a bijective correspondence between all prime ideals in
AS and the prime ideals p A satisfying p S = ; see 1.2/6.
However, we
have S px = for a point x Spec A if and only if x f S D(f ) so that
im a = f S D(f ).
Exercises
1. Let : X - Y be a continuous map between topological spaces. Show for any
irreducible subset V X that its image (V ) as well as the closure (V ) are
irreducible in Y .
2. For a morphism of rings : A - A consider the associated map between
a
spectra : Spec A - Spec A. Assume that Spec A is irreducible and let x
be its generic point. Show that a (x) is the generic point of the closure im a .
3. Let : A - A be a morphism of algebras over a eld K and consider the

associated map of spectra a : Spec A - Spec A. Show that the image of any
closed point x Spec A is closed in Spec A, when A is of nite type over K. Is
the latter assumption necessary?
4. Let : A - A be an integral morphism of rings. Show that the associated
map of spectra a : Spec A - Spec A is closed in the sense that it maps closed
subsets of Spec A onto closed subsets of Spec A.
5. For a ring R consider Rt, the polynomial ring in a variable t over R. The
spectrum A1R = Spec Rt is referred to as the ane line over R. The latter is
viewed as a relative object over Spec R via the projection : A1R - Spec R
derived from the canonical injection R - Rt. Assuming that R is a principal
ideal domain, determine the bers of the map . In particular, show:
(a) Given a point s Spec R, the ber 1 (s) is canonically homeomorphic to
the ane line A1k(s) where k(s) is the residue eld of s.
(b) If s corresponds to a prime ideal in R that is generated by a prime element
p R, the ber 1 (s) consists of all ideals in Rt that are of type (p) or
(p, f ) where f is a monic polynomial in Rt whose residue class in (R/pR)t is
irreducible.
(c) For the generic point s Spec R the ber 1 (s) consists of the zero ideal
in Rt and of all ideals of type (f ) where f is primitive in Rt and irreducible
in k(s)t. Note that the residue eld k(s) coincides with the eld of fractions
of R and that a polynomial f Rt is called primitive if the greatest common
divisor of its coecients is 1.
6. Neiles parabola: For a eld K and variables t, t1 , t2 consider the morphism of
K-algebras
: Kt1 , t2 /(t22 t31 ) - Kt, t1 - t2 , t2 - t3 .
Show that the associated map between spectra a is a homeomorphism, although

is injective, but not surjective. Hint: Localizing by the multiplicative system
generated from the residue class t1 yields an isomorphism of K-algebras. Also note
that the plane curve with equation x22 = x31 has been considered by W. Neile; it
is referred to as the semicubical or Neiles parabola.
6.3 Presheaves and Sheaves
It is convenient to use the language of categories and functors, as introduced in

Sect. 4.5, when dealing with sheaves and presheaves.
For a topological space X, let us consider the category Opn(X) of all open
subsets of X, dened as follows. The objects of Opn(X) are, indeed, the open
subsets of X, whereas morphisms U - V between two such subsets U, V X
are given by

if U V
Hom(U, V ) = .
{inclusion map U - V } if U V
6.3 Presheaves and Sheaves 217
Denition 1. Let X be a topological space. A presheaf of sets (resp. groups,

rings, modules over a xed ring R, . . . ) on X is a contravariant functor
F : Opn(X) - Set
into the category of sets (resp. groups, rings, modules over a xed ring R, . . . ).
Thus a presheaf F on X, say of sets, consists of the following data:
(a) sets F(U ) where U varies over all open subsets of X,

(b) morphisms VU : F(V ) - F(U ) where the pair (U, V ) varies over all
inclusions U V of open subsets in X.
The following conditions are required:
(i) UU : F(U ) - F(U ) is the identity map for all open subsets U X,
(ii) VU W
V = U for open subsets U V W X.
W
If X is an open subset of X, the functor F : Opn(X) - Set can be

restricted to the category Opn(X ) of open subsets contained in X . Thereby

we obtain a presheaf again, the latter being denoted by F|X .

Sometimes it is convenient to imagine the elements of F(U ) as functions
on U and, likewise, the maps VU : F(V ) - F(U ) for inclusions U V as
restrictions of functions on V to functions on U . This is why these maps are
usually referred to as restriction morphisms, using the suggestive notation
VU : F(V ) - F(U ), f - f |U .
However, in many cases the elements of F(U ) will not be given as functions on
U in the strict sense of the word. Consequently, we cannot argue in terms of
ordinary functions when dealing with general presheaves.
Concerning presheaves of modules, there exists a slight variant of the sit-
uation covered in Denition 1. Let O be a presheaf of rings on a topological
space X and F a presheaf of abelian groups on X. Then the cartesian product
O F is dened as the functor that associates to an open subset U X the
cartesian product O(U ) F(U ) and to an inclusion U V of open subsets
in X the cartesian product of the restriction morphisms O(V ) - O(U ) and
F(V ) - F(U ). Then a law of composition O F - F is meant as a func-
torial morphism and, thus, consists of maps O(U )F(U ) - F(U ) for U X
open that are compatible with restriction morphisms. Using such a terminology,
an O-module is dened as a presheaf of abelian groups F together with a law of
composition : O F - F such that, for any open subset U X, the map
(U ) : O(U ) F(U ) - F(U ) denes an O(U )-module structure on F(U ).
Denition 2. A presheaf F on a topological space X

is called a sheaf if for
every open subset U X and every covering U = U by open subsets
U X the following hold :
(i) If f, g F(U ) satisfy f |U = g|U for all , then f = g.

(ii) If f F(U ), , satisfy f |U U = f |U U for all , ,
there exists f F(U ) such that f |U = f for all . (Note that f will be
unique by (i).)
Clearly, if F is a sheaf on a topological space X, its restriction F|X to any

open subset X X is a presheaf that is a sheaf again.
Alternatively, we can require in place of conditions (i) and (ii) above that
the diagram
-

F(U ) - F(U ) - F(U U )
(, )
is exact where the map on the left-hand side is given by f - (f |U ) and

the ones on the right-hand side by
(f ) - (f |U U )(, ) ,
(f ) - (f |U U )(, ) .
Also recall that a diagram of type

1

- -
-
A B C
2
is called exact if is injective and its image im coincides with
ker(1 , 2 ) := {b B ; 1 (b) = 2 (b)},
where the latter set is called the kernel of (1 , 2 ). For example, if A, B, C are
abelian groups and , 1 , 2 are group homomorphisms, the above diagram is
exact if and only if the sequence of abelian groups
1 2
0 - A - B - C
is exact.
We will use the remainder of the present section in order to discuss some
examples of presheaves and sheaves.
(0) The zero or trivial sheaf (of abelian groups, rings, or modules) on a
topological space X is obtained by assigning to each open subset in X the zero
group, ring, or module.
(1) The continuous (resp. dierentiable) R-valued functions on R yield a

presheaf of rings and, in fact, a sheaf, since the condition continuous (resp.
dierentiable) is tested locally. To be a bit more explicit, consider X = R as
a topological space with the usual topology and let F(U ) for U R open be
the ring of all continuous (resp. dierentiable) R-valued functions on U . Dene
restriction morphisms as usual by restriction of functions.
(2) Let X be a topological space and G a group. Similarly as before, the

constant G-valued functions on X dene a presheaf of groups. However, since the
condition of a function to be constant cannot be tested locally (on disconnected
sets), this presheaf is not a sheaf in general. On the other hand, the locally
constant G-valued functions on X dene a presheaf which is a sheaf.
(3) Let X be a set which is considered as a topological space under the

discrete topology. (This means that every subset U X is open.) We can
dene a presheaf F on X by

Z if U = X,
F(U ) =
0 if U X,
and by taking the identity on Z as well as the zero maps as restriction mor-
phisms. Then F is not a sheaf if X consists of at least two elements.
(4) Let A be a ring and X = Spec A. Restricting to basic open subsets of

X, consider the category D(X) of all open subsets of type D(f ) X as objects
where f varies in A, with inclusions D(f ) D(g) as morphisms. As a variant
we may also look at the category D (X) which diers from D(X) in so far
as two objects D(f ) and D(g) are viewed as dierent as soon as the functions
f and g are dierent, even if D(f ) and D(g) coincide as subsets of X. Thus,
strictly speaking, the objects of D (X) are given by the elements f A and
the morphisms by the inclusions of type D(f ) D(g). In particular, mapping
f A to the corresponding subset D(f ) X yields a well-dened forgetful
functor F : D (X) - D(X). On the other hand, we can consider a section
of F , namely a functor G : D(X) - D (X) by selecting for each basic open
subset U X an element f A such that U = D(f ). Then it is immediately
clear that F and G dene an equivalence between D (X) and D(X).
There is a natural contravariant functor
OX

: D (X) - Ring,
D(f ) - Af ,
D(f ) D(g) - Ag - Af ,
so to say a presheaf on the family (D(f ))f A of all basic Zariski open subsets
in X, which, in more detail, is dened as follows. Assign to an object D(f )
of D (X) the localization Af of A by the multiplicative system {1, f 1 , f 2 , . . .}.
Then any inclusion D(f ) D(g) gives rise to a well-dened ring homomorphism
Ag - Af . Indeed, D(f ) D(g) is equivalent to V (f ) V (g) and, using
6.1/5 (i), to rad(f ) rad(g), hence, to the existence of elements n N and
a A such that f n = ag. Looking at the canonical map f : A - Af , we
see that f (f )n = f (a)f (g) and, hence, f (g) are units in Af . Thus, by the
universal property of the localization map g : A - Ag , there is a unique
factorization
f
A - Af
-
g fg
-
Ag
by a homomorphism fg : Ag - Af . The latter is the map we want to assign
to the inclusion D(f ) D(g) by means of the functor OX

. To check that OX

f
really is a functor, observe rst that f : Af - Af for any f A is the
identity map. Furthermore, for any chain of inclusions D(f ) D(g) D(h),
the composition
gh fg
Ah - Ag - Af
coincides with fh : Ah - Af . This follows immediately from the commutative
diagram
A
?
h Ag f
-
gh fg
-

fh
Ah - Af
using the universal property of the localization map h : A - Ah . Thus, OX
is a functor, indeed.
Now we want to construct a functor OX : D(X) - Ring such that its
composition with the forgetful functor F : D (X) - D(X) is isomorphic to
OX
. Proceeding in a non-canonical way, we could select for each basic open
subset U X a describing function f A such that U = D(f ), thereby
obtaining a section G : D(X) - D (X) of F . Then OX could be dened
as the composition OX G. Since G F is isomorphic to the identity functor

on D (X), it would follow that OX F is isomorphic to OX

.
However, there is a more satisfying denition of OX , which does not involve
any choices and, thus, is completely natural. For each f A set
S(f ) = {g A ; D(f ) D(g)}.
Then the equation D(g g ) = D(g) D(g ) for elements g, g A (see 6.1/2)
shows that S(f ) is a multiplicative system in A containing all powers of f .
In particular, the universal property of localizations yields a canonical homo-
morphism f : Af - AS(f ) . We claim that the latter is actually an isomor-
phism. Indeed, we know from the discussion of the functor OX
above that,
for any g A, its image f (g) with respect to the canonical homomorphism
f : A - Af is a unit if D(f ) D(g). From this we conclude that f maps

S(f ) into the group of units in Af and, thus, that f must be an isomorphism.
Since D(f ) D(g) implies and, actually, is equivalent to S(f ) S(g), the
theory of localizations yields
OX : D(X) - Ring,
D(f ) - AS(f ) ,
D(f ) D(g) - AS(g) - AS(f ) ,
as a well-dened functor from D(X) to the category of rings. Furthermore, the

isomorphisms of type f : Af - AS(f ) for f A show that the composition
OX F with the forgetful functor F : D (X) - D(X) is isomorphic to the
above considered functor OX . Alluding to this fact, we will ignore the dierence

between Af and AS(f ) in the following by writing Af instead of AS(f ) in most

cases. Later in 6.6/2 we see that the functor OX , which is viewed as a presheaf
on (the basic open sets of) X, is even a sheaf.
Exercises
1. Saturation of multiplicative systems: Let S be a multiplicative system of a ring
A. Dene the saturation of S as the set S of all elements in A that are divided
by an element of S. Show:
(a) S is a multiplicative system in A.
(b) If S is the multiplicative system consisting of all powers of some element
f A, then its saturation S coincides with the multiplicative system S(f ) as
considered above.
(c) The canonical morphism AS - AS is an isomorphism.
2. Give an example of a presheaf F on a topological space X such that F satises

the sheaf condition (i) of Denition 2, but not condition (ii). In the same way,
show that condition (i) is not a consequence of condition (ii).
3. The empty set: For a given topological space X and a set (resp. abelian group,
resp. ring) E, construct a presheaf F of sets (resp. abelian groups, resp. rings)
on X such that F() = E. Show for a sheaf F of sets (resp. abelian groups, resp.
rings) on X that F() is a terminal object in the corresponding category, thus, a
one-point set (resp. the zero group, resp. the zero ring). Hint: A terminal object
of a category C is an object Z such that HomC (Y, Z) is a one-point set for every
object Y in C. Note that, by convention, the cartesian product over the empty
family of objects in F is given by a terminal object of C if the latter exists.
4. Consider the spectrum X = Spec R of a principal ideal domain R. Show that
every open subset in X is basic open and that the presheaf OX dened above is a
sheaf. Hint: Use factoriality arguments in principal ideal domains. Furthermore,
see 6.6/2 for a generalization to arbitrary rings R.
5. Let A be an integral domain and K its eld of fractions. For any point x
in X = Spec A, write Ax for the localization of A by the corresponding prime
ideal px A and view Ax as a subring of K. Furthermore, consider the functor

O : Opn(X) - Ring that is given by O(U ) =
xU Ax on any open sub-
set U X and that, to any inclusion of open sets U V X associates the
canonical inclusion morphism O(V ) - O(U ). Show that O is a sheaf and that
the above constructed presheaf OX can be interpreted as the restriction of the
functor O : Opn(X) - Ring to the subcategory D(X) of Opn(X).
6.4 Inductive and Projective Limits
To dene inductive limits, we need an index set I equipped with a preorder, i.e.
with a binary relation such that
(i) i i for all i I,

(ii) i j and j k implies i k for i, j, k I.
In order to keep things simple, we assume that I is non-empty and, in

addition, directed, namely that
(iii) for any i, j I there exists k I such that i k and j k.
Note that without the latter assumption the proof of Proposition 2 below
would have to be modied while the assertion of Remark 3 could not be main-
tained.
Now let (Gi )iI be a family of sets (resp. groups, rings, modules, etc.), or of
objects in any category C, together with C-morphisms fij : Gi - Gj for all
indices i j in I such that
(a) fii = id for all i I,

(b) fik = fjk fij for indices i j k in I.
We call (Gi )iI together with the morphisms fij an inductive system and
denote it by (Gi , fij )i,jI .
Denition 1. Let (Gi , fij )i,jI be an inductive system of objects in a category

C. An object G of C together with C-morphisms fi : Gi - G, i I, satisfying
fi = fj fij for i j is called an inductive or direct limit of the system
(Gi , fij )i,jI if the following universal property holds:
Let H be an object of C and gi : Gi - H, i I, C-morphisms such that
gi = gj fij for i j. Then there exists a unique C-morphism g : G - H
such that gi = g fi for all i I.
The involved morphisms can be read from the following commutative dia-
gram:
6.4 Inductive and Projective Limits 223
Gj
6
gj
fij
fj
-
-
g
G - H
-
-
fi
gi
Gi
In particular, we see that an inductive limit G of an inductive system (Gi , fij )i,jI
is uniquely determined up to canonical isomorphism if it does exist. Without
spelling out the attached maps explicitly one often writes G = lim Gi , calling

this the inductive limit of the Gi , i I.
When dealing with inductive limits, it is often convenient to restrict the
inductive system (Gi , fij )i,jI to a so-called conal subset I of I. Thereby we
mean a subset I I such that every i I admits an index i I satisfying
i i. Then I is directed again, and it is an easy exercise to show that the
inductive system (Gi , fij )i,jI admits a limit if and only this is the case for the
full system (Gi , fij )i,jI . Furthermore, both limits will canonically coincide if
they exist.
Also let us point out that the notion of inductive systems and their limits
extends to the more general setting, where I is just a collection of indices with
a preorder. In more precise terms, we would talk about a category I, writing
i j for objects i, j if there is a morphism i - j in I. A natural example of
such a collection will occur in Sects. 6.5 and 7.6 where we introduce techniques
of Cech cohomology. Namely, consider a topological space U and take for I the
collection IU of all open coverings of U . Then IU is not a set, since arbitrary
repetitions of covering sets are allowed. However, when restricting to coverings
without repetitions, in other words, to coverings that correspond to subsets
of the power set P(U ), we get a set IU , and the latter is conal in IU with
respect to the preorder given by the renement relation. In particular, any
inductive system indexed by IU will have a limit if its restriction to IU has
a limit. Thereby we are reduced to the case of true index sets, as considered
from the beginning of this section on. Therefore, unless stated otherwise, we
will assume in the following that all inductive (or projective) systems under
consideration are indexed by index sets.
Proposition 2. Inductive limits exist in the categories of sets, groups, rings,

and modules (over a given ring R).
Proof. Starting with the case of an inductive system of sets (Gi , fij )i,jI , we
!
look at the disjoint union G = iI Gi and introduce an equivalence relation
on it as follows. Write x y for elements x, y G where x Gi and y Gj ,
if there is some k I, i, j k, such that fik (x) = fjk (y). We claim that
is an equivalence relation. Of course, is reexive and symmetric, but also
transitive. Indeed, let x Gi , y Gj , and z Gk be elements in G such that

x y and y z. There are indices r I, i, j r, such that fir (x) = fjr (y)
and s I, j, k s, such that fjs (y) = fks (z). Furthermore, using the fact that
I is directed we can nd an index t I such that r, s t. Then the equation
fit (x) = frt (fir (x)) = frt (fjr (y)) = fjt (y)
= fst (fjs (y)) = fst (fks (z)) = fkt (z)
shows x z. Now writing G = G/ , the canonical embeddings fi : Gi - G

induce maps fi : Gi - G which satisfy fi = fj fij for i j, and we claim
that G together with these maps is an inductive limit of (Gi , fij )i,jI .
To justify this, consider a set H together with maps gi : Gi - H that
satisfy gi = gj fij for all indices i j in I. Assuming there is a map g : G - H
satisfying gi = g fi for all i I, we want to show it is unique. Indeed, let
x G and choose a representative x G of x, say x Gi for some i I, which
means fi (x) = x. Then we get g(x) = (g fi )(x) = gi (x) and this shows that
g : G - H is unique if it exists.
To actually construct g : G - H, look at an element x G and choose
a representative x G again, say x Gi for some i I, and set g(x) := gi (x).
If y G is another representative of x, say y Gj , then x and y are equivalent
and there exists an index k I such that i, j k and fik (x) = fjk (y). This
implies
g(x) = gi (x) = gk (fik (x)) = gk (fjk (y)) = gj (y)
and we see that g : G - H is well-dened. Furthermore, it is clear by its
construction that g satises the compatibilities gi = g fi for i I .
Next, let us turn to the construction of inductive limits of groups; induc-
tive limits of rings and modules are dealt with in a similar way. Therefore let
(Gi , fij )i,jI be an inductive system of groups where now, of course, the maps
fij are group homomorphisms. Then the inductive limit G = G/ together
with the associated maps fi : Gi - G exists in the category of sets, as shown
above. We want to equip G with a group structure in such a way that all maps
fi : G i - G become group morphisms. To do this, let x, y G and choose
representatives x, y G, say x Gi and y Gj where fi (x) = x and fj (y) = y.
Then we can nd an index k I with i, j k and set
x y = fk (fik (x) fjk (y)),
knowing that the elements x = fik (x) and y = fjk (y) of Gk form a new set of
representatives for x and y. To show that the product x y is well-dened, look
at another index r I such that there are representatives x , y Gr of x and
y. Then there are indices s, t I satisfying k, r s and k, r t such that
fks (x ) = frs (x ), fkt (y ) = frt (y ).
Of course, we may assume s = t replacing s and t by some t I where s, t t .
Then, using the fact that fks : Gk - Gs and frs : Gr - Gs are group
morphisms, we obtain
fks (x y ) = fks (x ) fks (y ) = frs (x ) frs (y ) = frs (x y )
and it follows that the products x y and x y are equivalent, thus representing
the same element x y G. Therefore the product x y is well-dened in G and
it is easily seen that all maps fi : Gi - G respect such products in the sense
that fi (x y) = fi (x) fi (y) for x, y Gi .
Next we want to show that the above dened product yields, in fact, a group
structure on G and, hence, that all maps fi : Gi - G are group morphisms.
First, there is a well-dened element e G, which is represented by the unit
element ei Gi for any i I; note that any two unit elements ei Gi and
ej Gj are equivalent in G, since group morphisms map unit elements to unit
elements. Furthermore, the group axioms for G follow immediately from those
of the Gi , relying on the fact that for nitely many elements of G there is always
an index i I such that Gi contains representatives of all these elements.
To establish the universal property of the inductive limit consider a group
H and group morphisms gi : Gi - H satisfying gi = gj fij for indices i j
in I. As we already know, there is a unique map of sets g : G - H such that
gi = g fi for all i I. To verify that g is even a group morphism consider two
elements x, y G. Since I is directed, there exists an index i I such that Gi
contains representatives x, y of both, x and y. Then the equations
g(x y) = g(fi (x) fi (y)) = g(fi (x y)) = gi (x y)

= gi (x) gi (y) = g(fi (x)) g(fi (y)) = g(x) g(y)
show that g : G - H is a group morphism, as claimed.
The construction shows that inductive limits are compatible with forgetful
functors as follows:
Remark 3. If G, as in Proposition 2, is an inductive limit in the category of

groups, then, as a set, it is an inductive limit in the category of sets as well. We
say that the inductive limit is compatible with the forgetful functor
Grp - Set,
which associates to a group its underlying set and to a group morphism its
underlying map of sets.
In a similar way the inductive limit is compatible with other forgetful func-
tors, like the one from the category of rings or modules to the category of groups
or sets.
We want to look at a simple example, which is nevertheless quite basic. Let

M be a set. Viewing the nite subsets of M as an index set I, let us use Mi
with i I as a second notation for the subset of M given by i. To dene a
preorder on I, write i j for indices i, j I if Mi Mj . Furthermore, let
fij : Mi - Mj be the associated inclusion map. Since the union of two nite
subsets of M is nite again, it follows that I is directed and that (Mi , fij )i,jI
is an inductive system. We claim that M = lim Mi , more precisely, that M

- M given by the inclusions Mi M is an
together with the maps fi : Mi
inductive limit of the system (Mi , fij )i,jI . To justify this, x a set N and look
at maps gi : Mi - N , i I, where gi = gj fij for i j. Then any maps
gi , gj for arbitrary indices i, j I coincide on Mi Mj and therefore yield a
well-dened map g : M - N , taking into account that M is the union of all
Mi , i I. Of course, we have gi = g fi for all i I and it follows that, indeed,
M is an inductive limit of (Mi , fij )i,jI . In the same way one can show that a
group (resp. a ring, or a module) is the inductive limit of any directed system
of subgroups (resp. subrings, or submodules) of G that cover G. For example,
the system of all nitely generated subgroups (resp. all subrings of nite type
over Z, or all submodules of nite type) of G admits this property.
Next we discuss the concept of projective limits, which is dual to that of
inductive limits. This means that we use the same denition as for the inductive
limit, although arrow directions have to be reversed. Again we need an index set
I with a preorder which satises the conditions (i) and (ii) mentioned at the
beginning of this section. However, we will not assume that I is directed since
this is unnecessary. Now let (Gi )iI be a family of sets (resp. groups, rings,
modules, etc.), or of objects in any category C, together with C-morphisms
fij : Gj - Gi for indices i j in I such that
(a ) fii = id for all i I,

(b ) fik = fij fjk for indices i j k in I.
We call (Gi )iI together with the morphisms fij a projective system and
denote it by (Gi , fij )i,jI .
Denition 4. Let (Gi , fij )i,jI be a projective system of objects in a category

C. An object G of C together with C-morphisms fi : G - Gi , i I, satisfying
fi = fij fj for i j, is called a projective or inverse limit of the system
(Gi , fij )i,jI if the following universal property holds:
Let H be an object of C and gi : H - Gi , i I, C-morphisms such that
gi = fij gj for i j. Then there exists a unique C-morphism g : H -G
such that gi = fi g for all i I.
The involved morphisms can be read from the following commutative dia-
gram:
Gj
-
-
gj
fij
fj
g
H - G
fi
gi - ?
-
Gi
As in the case of inductive limits, a projective limit G of a projective system

(Gi , fij )i,jI is unique up to canonical isomorphism if it does exist. Often one
writes G = lim Gi and calls this the projective limit of the Gi , i I.

Proposition 5. Projective limits exist in the categories of sets, groups, rings,
and modules (over a given ring R).
Proof. Starting
with a projective system of sets (Gi , fij )i,jI , consider the carte-
sian product jI Gj together with the projections

pi : Gj - Gi , (xj )jI - xi ,
jI
where i varies over I. We claim that the subset

G = (xi )iI Gi ; fij (xj ) = xi for all i j in I Gi
iI iI
together with the maps
fi : G - Gi , (xj )jI - xi ,
induced from the projections pi for i I is a projective limit of the system

(Gi , fij )i,jI . This is easy to verify. First observe that fi = fij fj by the de-
nition of G. Now if H is a set together with maps gi : H - Gi , i I, there
is a map
g : H - Gi , x - (gi (x))iI ,
iI
which is uniquely characterized by the relations gi = pi g, i I. If, in addition,

the gi satisfy gi = fij gj for i j, then clearly the image of g lies in G so
that g restricts to a unique map g : H - G satisfying gi = fi g for i I. In
particular, G together with the projections fi : G - Gi is a projective limit
of the system (Gi , fij )i,jI .
Now if (Gi , fij )i,jI
is a projective system of groups (resp. rings or modules),
the cartesian product jI Gj is a group (resp. ring or module) and the same
is true for G, since
all maps - fij are assumed to be homomorphisms. Then the
projections pi : jI Gj Gi restrict to homomorphisms fi : G - Gi
and one shows as above, but now arguing in terms of homomorphisms, that G
together with the homomorphisms fi is a projective limit of (Gi , fij )i,jI .
Just as we have done for inductive limits, let us discuss a simple example of
a projective limit. Consider a family of maps between sets (hi : Xi - S)iI
with common target S, which we want to view as a projective system of sets.
To be formally correct take I = I {0} as index set. So we let I be the disjoint
union of I with another symbol 0 not yet contained in I and set Gi = Xi for
i I as well as G0 = S. Now introduce a preorder on I by writing 0 i for
all i I (and, of course, 0 0 as well as i i for all i I). Then the maps
hi , i I, (together with the identity maps on S and Xi for all i I) yield

a projective system (Gi , fij )i,jI of sets. Analyzing the universal property of
a projective limit of this system, it is convenient to use the category SetS of
relative sets over S. Indeed, a projective limit of (Gi , fij )i,jI will be a relative
object X in SetS together with S-morphisms pi : X - Xi such that the map

HomS (T, X) - HomS (T, Xi ), - (pi )iI
iI
is bijective for any object T in SetS . However, as a

generalization of 4.5/2, this
universal property characterizes the ber product Xi of the Xi over S so
S
that we get
lim Gi = Xi .
S
iI iI
We have looked at inductive and projective limits since we need these con-
cepts for dealing with sheaves and presheaves. An important example of induc-
tive limits is given by the so-called stalks of sheaves and presheaves which we
want to discuss next.
Denition 6. Let X be a topological space equipped with a presheaf (or a sheaf )

F. Then, for any point x X, we call
Fx = lim F(U )

xU
the stalk of F at x, where the inductive limit extends over all open neighborhoods
U X of x.
To describe the involved inductive system in more detail, use the set of all
open neighborhoods of x as index set I and dene a preorder on it by the in-
clusion relation. Then we have to consider the inductive system (F(U ), UV )U,V
where UV : F(U ) - F(V ) is the restriction morphism associated to an inclu-
sion V U of members in I.
As a typical example, consider the sheaf F of all continuous real valued
functions on a topological space X, which is a sheaf of rings. Then there are
canonical homomorphisms
F(U ) - Fx , f - fx ,
for any point x X and any open neighborhood U X of x. The image fx Fx

of a function f F(U ) is called the germ of f at x. From the construction of
inductive limits in the proof of Proposition 2 we can read the following facts:
(i) Every element in Fx can be interpreted as the germ of a continuous
function f F(U ) where U X is a suitable open neighborhood of x.
(ii) If fx , gx Fx are germs of continuous functions f F(U ) and g F(V )
for open neighborhoods U, V X of x, then fx = gx if and only if there is an
open neighborhood W X of x such that W U V and f |W = g|W .
Thus, indeed, the ring Fx may be imagined as the ring of all germs of
continuous real valued functions around x. Let us point out that the example
gives a good picture of the general situation as well. Except for the interpretation
of elements in F(U ) as functions on U , the above characterization of elements
in Fx remains valid for any sheaf or presheaf F on X.
Lemma 7. Let X be a topological space and F a sheaf on it (or a presheaf

satisfying condition 6.3/2 (i)). Then, for any open subset U X, the map

F(U ) - Fx , f - (fx )xU ,
xU
is injective.
The proof is quite simple. Let f, g F(U ) be such that fx = gx for all
x U . Then there exists for every x U an open neighborhood Ux U such
that f |Ux = g|Ux ; use the above condition (ii). Since (Ux )xU is an open covering
of U , we get f = g by condition (i) of 6.3/2.
The lemma makes it possible to describe sheaves via their stalks in a quite
instructive way. Namely, we consider the so-called etale!space of F (from the
French etale meaning spread out), which is given by xX Fx and may be
imagined as the family of all stalks Fx , x X:

Fx - Fy
yX
p
? ?
{x} - X ,
or
Fx !
xX Fx
?
X
x
Given any element f F(U ) on an open subset U X, we can identify f
according to Lemma 7 with the family (fx )xU of all germs of f , hence, with
the map
sf : U - {fx ; x U } Fx , x - fx .
xX
(p sf )(x) = x for all x U and, thus, is a section over U of

The latter satises!
the projection p : xX Fx - X from the etale space associated to F onto
the basis X:
Fx !
xX Fx
f p
?
X
U x
Having this in mind, the elements in F(U ) are usually referred to as sections of
F over!U . However note that, in general, there will be an abundance of sections
of p : xX Fx - X over an open subset U X, in particular, sections that
are not induced by elements of F(U ) and therefore are not related! to the sheaf
F. One can introduce a suitable topology on the etale space xX Fx such that
the elements of F(U ) correspond precisely to the continuous sections over U ,
but we will not apply this point of view in the following.
However, let us point out! that the interpretation of elements from F(U ) as
sections in the etale space xX Fx can be used to canonically associate a sheaf
F to any presheaf F . Indeed, let F be a presheaf on a topological space X.
Then the stalk ! Fx of F is dened at any point x X and we can consider the
etale space p : xX Fx - X over X. For open subsets U X we dene
!
F(U ) as the setof all sections s : U -
xU Fx such that there is an open
covering U = iI Ui together with a family (fi )iI of elements fi F (Ui )
satisfying
s(x) = fi,x for all i I and all x Ui .
! - X that locally on U are
Thus, F(U ) consists of all sections of p : xU Fx
represented by elements in the presheaf F . Without diculties one can check
that F together with the obvious restriction morphisms is a sheaf, the so-called
sheaf associated to the presheaf F ; see 6.5/5 for a more thorough discussion of
associated sheaves.
Exercises
1. Let X be a set and (Ui )iI a family of subsets in X.

(a) Interpret (Ui )iI as a projective system satisfying limiI Ui = iI Ui .

(b) Abandoning the condition on the directness of index sets for inductive
systems,
interpret (Ui )iI as an inductive system satisfying limiI Ui = iI Ui .

2. Interpret the family (Z/nZ)nN as an inductive system of groups with inductive
limit limnN Z/nZ = Q/Z.

3. Tensor products commute with direct limits: Let (Mi )iI be an inductive system
of modules over a ring R. Show for any R-module N that there is a canonical
isomorphism of R-modules (limiI Mi ) R N - limiI (Mi R N ).

4. Exactness of inductive limits: Dene the notion of a short exact sequence
0 - (M )iI - (Mi )iI - (M )iI - 0

i i
of inductive systems of modules over a ring R in a natural way and show that
the resulting sequence 0 - limiI Mi - lim M - lim M - 0
iI i iI i
is exact.
5. Let p be a prime ideal of a ring A. Show that the localizations Af for f A p
form an inductive system and that the the localization Ap can be interpreted as
the inductive limit of this system.
6. Show that any ring A can be interpreted as the projective limit of all its local-
izations Af , f A.
7. a-adic completion: Let a be an ideal of a ring A. Call a subset U A open if for
each element f U there is an integer n N such that f + an U . This way the
powers an , n N, play the role of a fundamental system of neighborhoods of 0. A
sequence (ai )iN of elements in A is called a Cauchy sequence if for every n N
there is an index k N such that ai aj an for all i, j k. Furthermore, such
a sequence is called a zero sequence if for every n N there is an index k N
such that ai an for all i k.
Show:
(a) The open subsets dene, indeed, a topology on A; the latter is called the
a-adic topology.
(b) The set C of Cauchy sequences in A is a ring under componentwise addition
and multiplication; the zero sequences dene anideal N C. There is a canonical
ring homomorphism A - C/N with kernel nN an .
(c) There is a canonical isomorphism C/N - limnN A/an .

(d) The projective limit A = limnN A/an can be viewed as the (separated)

a-adic completion of A. Namely, using the kernels of the canonical morphisms
A - A/an as a fundamental system of neighborhoods of 0 in A, we get a
topology on A such that the image of A - A is dense and every Cauchy se-
quence converges in A. Furthermore, the topology on A induces the given one on
A in the sense that a subset in A is open if and only if it is the preimage of an
open subset in A.
8. For a ring R and a variable t consider the ring Rt of formal power series
in t with
coecients in R. Recall that Rt consists of all formal expressions nN an tn
for arbitrary coecients an R, and that addition as well as multiplication of
such sums are dened as usual. Show that Rt can be interpreted as the (t)-adic
completion (see Exercise 7 above) of the polynomial ring Rt.
9. Sheaf of holomorphic functions: For an open subset U C let OC (U ) be
the C-algebra of holomorphic functions on U . Show that the resulting functor
Opn(C) - C-Alg is a sheaf and that the stalk OC,0 at the origin 0 C can
naturally be identied with the ring of convergent power series

an tn Ct ; an z n is convergent for some z C {0} ,
nN nN
the latter being viewed as a subring of Ct. For the formal power series ring
Ct see Exercise 8 above.
10. Alternative denition of sheaves: An etale sheaf of sets on a topological space X
consists of a topological space F together with a continuous map : F - X
that is a local homeomorphism. Show that the concept of an etale sheaf of sets
is equivalent to the one of sheaf of sets as introduced in 6.3/2.
6.5 Morphisms of Sheaves and Sheacation
To compare dierent presheaves and, likewise, sheaves among each other, we

need to introduce the concept of morphisms between these.
Denition 1. Let X be a topological space and F, G presheaves (resp. sheaves)

on X. A morphism of presheaves (resp. sheaves) : F - G is a functorial
morphism between F and G, the latter being viewed as functors from the category
of open subsets in X to the category of sets (resp. groups, rings, . . . ).
Hence, consists of a collection of morphisms
(U ) : F(U ) - G(U ), U X open,
that are compatible with restriction morphisms in the sense that for open subsets
U V X the diagram
(V )
F(V ) - G(V )
V
U V
U
? ?
(U )
F(U ) - G(U )
is commutative. We call an isomorphism if is a functorial isomorphism,
i.e. if all morphisms (U ) are isomorphisms.
For modules over a presheaf of rings O on X, the concept of a morphism can

be adapted in a natural way. Namely, a morphism of O-modules F - G is
a morphism of presheaves of abelian groups that is compatible with O-module
structures in the sense that the canonical diagram
OF - OG
? ?
F - G
is commutative. For module (pre)sheaves F, G over a (pre)sheaf of rings O, the

set of O-module morphisms is denoted by HomO (F, G).
6.5 Morphisms of Sheaves and Sheacation 233
Remark 2. Let : F - G be a morphism of presheaves on a topological

space X and consider any point x X. Then induces a morphism of stalks
x : Fx - Gx in such a way that for all open neighborhoods U X of x the
canonical diagram
(U )
F(U ) - G(U )
? x
?
Fx - Gx
is commutative.
Proof. Let x X. For open neighborhoods U X of x, consider the composition

(U )
F(U ) - G(U ) - Gx ,
which due to the properties of a morphism of presheaves is compatible with

restriction morphisms of F and G. Using the dening universal property of the
inductive limit Fx = limU x F(U ), the morphisms F(U ) - Gx must factorize
- Gx .
over a unique morphism x : Fx
Proposition 3. Let : F - G be a morphism of sheaves on a topological

space X.
(i) is injective in the sense that (U ) : F(U ) - G(U ) is injective
for all open subsets U X if and only if x : Fx - Gx is injective for all
x X.
(ii) is locally surjective in the sense that for any section g G(U ) over
an open subset U X and a point x U , there is an open neighborhood Ux U
of x such that g|Ux admits a preimage with respect to (Ux ) : F(Ux ) - G(Ux )
if and only if x : Fx - Gx is surjective for all x X.
(iii) is an isomorphism if and only if x : Fx - Gx is bijective for all
x X.
Proof. In all three cases, the only-if parts of the assertions are easily deduced
from the characterization of stalks in 6.4/6 and thereafter. To attack the if parts,
assume rst that x is injective for all x X. Fixing an open subset U X, we
have to show that (U ) : F(U ) - G(U ) is injective as well. To achieve this,
consider elements f, f F(U ) with (U )-images g, g G(U ) that coincide.

Then we have
x (fx ) = gx = gx = x (fx )
and, hence, fx = fx for all x U , since x is injective for all x X. But this
implies f = f by 6.4/7 and we see that (U ) is injective, thereby settling (i).
Next assume that x is surjective for all x X and consider an element
g G(U ) over some open subset U X. Fixing x U , there is an element
fx Fx such that x (fx ) = gx . Now let Ux U be an open neighborhood of x
such that the germ fx is induced by some element f(x) F(Ux ). Then we have

(Ux )(f(x) ) x = gx by Remark 2 and we may even assume (Ux )(f(x) ) = g|Ux
if we replace Ux by a smaller open neighborhood of x. This settles (ii).
Finally assume that x is bijective for all x X. Then is injective by
(i) and locally surjective by (ii). Thus, given a section g G(U ) on some open
subset U X, there exists for every x U an open neighborhood Ux U of
x together with a section f(x) F(Ux ) such that (Ux )(f(x) ) = g|Ux . Since the
map
(Ux Ux ) : F(Ux Ux ) - G(Ux Ux )
is injective for any points x, x U , we must have
f(x) |Ux Ux = f(x ) |Ux Ux
since both quantities left and right are mapped to g|Ux Ux under (Ux Ux ).
Using the fact that F is a sheaf, there is a unique element f F(U ) such that
f |Ux = f(x) for all x U . But then, by the sheaf property of G, we see that
(U ) maps f to g.
Denition 4. Let : F - G be a morphism of presheaves of abelian groups

(resp. presheaves of modules over a presheaf of rings) on a topological space X.
Then the following functors from the category of open subsets U X to the
category of abelian groups are presheaves of abelian groups (resp. presheaves of
O-modules) again:
(i) (ker )pre : U - ker((U )), the kernel of ,
(ii) (im )pre : U - im((U )), the image of ,
(iii) (coker )pre : U - coker((U )) = G(U )/ im((U )), the cokernel of .
If F and G in the situation of the denition are sheaves, it is easily seen

that (ker )pre is a sheaf as well. In this case the kernel is simply denoted by
ker . However, special care is necessary when dealing with image and cokernel
of a morphism of sheaves , since (im )pre and (coker )pre will not be sheaves
in general.
In the following we want to concentrate on the problem of constructing for
a presheaf F a so-called associated sheaf F , which, in a certain sense, is a sheaf
approximating F in a best possible way. A provisional version of this result was
already discussed at the end of Section 6.4.
Proposition 5. Let F be a presheaf on a topological space X. Then there

exists a sheaf F together with a morphism of presheaves F - F such that
the following universal property is satised :
If F - G is a morphism of presheaves into a sheaf G, there exists a
unique morphism of sheaves F - G such that the diagram
F - F
-
G
is commutative.
We call F together with the morphism F - F the sheacation of F,

or the sheaf associated to F.
The most rapid way to establish the above result is to consider sections in
the etale space of F, as we have explained at the end of Section 6.4. However,
here we want to use a more rigorous method derived from Cech cohomology
(see Section 7.6), which has the advantage that it extends to the context of
generalized topologies in the sense of Grothendieck [1].
Let X be a topological space and F a presheaf on it. Furthermore, consider
an open subset U X and an open covering U = (U ) of U by open subsets
U U , hence, satisfying U = U . Then we set

H 0 (U, F) = ker F(U ) - - F(U U )
(, )
where the kernel of the maps on the right-hand side can alternatively be inter-
preted as the limit of the projective system (F(U ) - F(U U )), . By
its denition, H 0 (U, F) consists of all families (f ) F(U ) such that
f |U U = f |U U for all , . In particular, there is a canonical map
F(U ) - H 0 (U, F), f - (f |U ) ,
which is a monomorphism if F satises sheaf condition 6.3/2 (i) and an isomor-

phism if F is a sheaf. Also note that we may view H 0 (U, ) as a functor on the
category of presheaves on X, since any morphism of presheaves : F -G
gives rise to a canonical map H (U, ) : H (U, F)
0 0 - H (U, G).
0
If U = (U ) and V = (V )M are two open coverings of U , we call V a

renement of U if there exists a map : M - such that V U () for all
M . Then there is a canonical map
H 0 (U, F) - H 0 (V, F), (f ) - (f () |V )M ,
and it is easily checked that this map is independent of the map used for
declaring V to be a renement of U. Indeed, let : M - be a second map
such that V U () for all M . It follows V U () U () and, since any
element (f ) H 0 (U, F) satises
f () |U () U () = f () |U () U () ,
we see that f () |V = f () |V for M .

Now x an open subset U X and consider the collection of all open cover-
ings U of U as an index collection IU , equipped with the preorder given by the re-
nement relation. So we write U V for elements in IU if V is a renement of U.
Then IU is directed, since for two open coverings U = (U ) and V = (V )M
of U , we can always construct the product covering U V = (U V )(,)M ,
which is a common renement of U and V. Therefore the family (H 0 (U, F))UIU ,
together with the morphisms induced by renements, form an inductive system,
and we can look at the corresponding limit
F + (U ) = lim H 0 (U, F).

UIU
The latter exists by 6.4/2 and the explanations preceding this result, since we
may replace the collection IU by a conal subset. If U varies over the open
subsets of X, we may interpret F + as a presheaf on X. Indeed, for open subsets
V U X and an open covering U = (U ) of U consider the corresponding
restriction U|V = (U V ) of U to V . Then, using the universal property of
the inductive limit, the morphisms of type
H 0 (U, F) - H 0 (U|V , F), (f ) - (f |U V ) ,
induce morphisms of type
F + (U ) - F + (V ),
which satisfy the conditions of a presheaf on X. Moreover, it is clear that the

maps of type
F(U ) - H 0 (U, F), f - (f |U ) ,
induce a morphism of presheaves F - F + , where for a sheaf F the maps

F(U ) - H (U, F) are isomorphisms and, hence, the same is true for
0
F - F + as well. If F only satises sheaf condition 6.3/2 (i), the maps

F(U ) - H 0 (U, F) are still injective and the resulting map F(U ) - F + (U )
does not lose this property. Unfortunately, F + is not a sheaf in general, but there
is the following result which, in particular, settles the assertion of Proposition 5:
Lemma 6. Let X be a topological space and F a presheaf on X.

(i) The assignment F - F + is compatible with restriction to open subsets
U X, i.e. there are canonical isomorphisms (F + )|U (F|U )+ .
(ii) The presheaf F + satises the sheaf condition 6.3/2 (i).
(iii) Assume that the sheaf condition 6.3/2 (i) holds for F. Then F + sat-
ises the sheaf condition 6.3/2 (ii) as well and, hence, is a sheaf. Further-
more, F - F + is a monomorphism in the sense that the canonical maps
F(U ) - F + (U ) for U X open are injective.
(iv) The presheaf F ++ = (F + )+ is a sheaf, which together with the canonical
map F - F + - F ++ satises the universal property of the associated sheaf
of F.
Proof. Assertion (i) follows directly from the denition of F + . Next, turning to
assertion (ii), x an open subset U X and consider elements f, g F (U )
+
such that there is an open covering U = U where f |U = g|U for all

. Since F + (U ) = limVI H 0 (V, F), there is an open covering V of U
U
such that f, g are represented by elements f , g H 0 (V, F). It follows that the
restrictions f |U and g|U are represented by the elements
f |U , g |U H 0 (V|U , F).
Since we have f |U = g|U for all , there is a renement W of V|U

for every such that the images of f |U and g |U coincide in H 0 (W , F).
Now setting up an open covering V of U by using the coverings W , ,
the images f , g H 0 (V , F) of the elements f , g H 0 (V, F) coincide by
construction and we get f = g as desired.
To verify assertion (iii) assume that F satises the sheaf condition 6.3/2 (i).
Consider an open subset U X and an open covering U = U of U , as
well as elements f F + (U ) such that f |U U = f |U U for all , .
We have to construct f F + (U ) such that f |U = f for all . To do
this choose an open covering V of U for every in such a way that
f F + (U ) is represented by some element f H 0 (V , F). Then consider
the product coverings V V of U U for , and observe that the
restrictions of f , f to U U yield elements f , f H 0 (V V , F) whose
images f |U U and f |U U coincide in F + (U U ). Passing to a suitable
renement W, of V V , we can assume that the images of f , f coincide
already in H 0 (W, , F). Then, by sheaf condition 6.3/2 (i), we see that, in fact,
f , f coincide in H 0 (V V , F) for all , . Setting up an open covering
V of U from the coverings V , , the elements f , , give rise to an
element f H 0 (V, F) and it follows from the construction that f induces an
element f F + (U ) satisfying f |U = f for all . Thus, F + satises the
sheaf condition 6.3/2 (ii). The fact that F - F + is a monomorphism was
already discussed before.
Finally, to obtain (iv) it remains to establish the universal property of an
associated sheaf for the composition F - F + - F ++ . For this it is enough
to check that each morphism : F - G to a sheaf G factors through a unique
morphism + : F + - G. To do this, let U X be open. Using the inductive
limit over all open coverings U = (U ) of U , we conclude from the canonical
commutative diagram
F(U ) - H 0 (U, F)
(U ) H 0 (U,)
? ?

G(U ) - H 0 (U, G)
that : F - G factors through a well-dened morphism + : F + - G.

To see that is uniquely determined by , consider an open set U X and
+
an element f F + (U ). Then there exists an open covering U = (U ) of U

such that f is represented by some element f H 0 (U, F). Furthermore, the
commutative diagram

F(U ) - H 0 (U|U , F) - F + (U )
(U ) H 0 (U|U ,) + (U )
? ? ?

G(U ) - H 0 (U|U , G) - G + (U ) ,
shows for that + (U )(f )|U = + (U )(f |U ) is uniquely determined by .
Since G is a sheaf, + (f ) is unique and we see that + is uniquely determined
by .
In particular, the construction of associated sheaves yields the following

result:
Remark 7. Let F be a presheaf on a topological space X. Then the canonical

morphism F - F + induces for every x X an isomorphism Fx - F +
x
between stalks at x. The same is true for the morphism F - F where F is

the sheaf associated to F.
If : F - G is a morphism between sheaves of abelian groups or of

modules over a sheaf of rings, then, similarly as in Denition 4, we can dene
the following sheaves:
(i) ker = (ker )pre , called the kernel of ,
(ii) im , the sheaf associated to (im )pre , called the image of ,
(iii) coker , the sheaf associated to (coker )pre , called the cokernel of .
As usual, is called a monomorphism if ker = 0. Moreover, is called an
epimorphism if im = G. If : F - G is a monomorphism, we may identify
F with its image in G and thereby view F as a subsheaf of G. In addition, the
quotient G/F makes sense in terms of the cokernel of .
Proposition 8. Let G be a sheaf of abelian groups on a topological space X and

F G a subsheaf. Then:
(i) The quotient in terms of presheaves (G/F)pre , dened as the cokernel
of the attached morphism of presheaves F - G, satises the sheaf condi-
tion 6.3/2 (i).
(ii) The quotient G/F in terms of sheaves, dened as the sheacation of
(G/F)pre , coincides with ((G/F)pre )+ .
(iii) For any section g (G/F)(U ) on some open part U X, there is an
open covering (Ui )iI of U such that the restrictions g|Ui are represented by
sections gi G(Ui ) satisfying (gi gj )|Ui Uj F(Ui Uj ) for all i, j I.
Proof. Consider a section g (G/F)pre (U ) = G(U )/F(U ) on some open subset

U X, represented by a section g G(U ), and assume that there exists an open
covering (Ui )iI of U such that g|Ui = 0 for all i I. Then g|Ui F(Ui ) for all i
and, hence, g F(U ), since F and G are sheaves. In particular, g = 0 and we see
that (G/F)pre satises the sheaf condition 6.3/2 (i). Thus, assertion (i) is clear
while (ii) follows from Lemma 6 (iii). Finally, assertion (iii) is a consequence of
the denition of ((G/F)pre )+ .
Also note that there is the notion of an exact sequence of sheaves of abelian
groups or modules. Namely, a sequence of morphisms

F - F - F
is called exact if im = ker (in the sense of sheaves). There is the following
criterion for exactness:
Proposition 9. A sequence
0 - F - F - F - 0
of sheaves of abelian groups (or of sheaves of modules over a sheaf of rings) on

a topological space X is exact if and only if the sequence of stalks
0 - Fx - Fx - Fx - 0
is exact for all x X.
Proof. First check that the formation of stalks commutes with the formation
of kernels and images of morphisms. Then the assertion is easily derived from
Proposition 3.
Finally, as another application of the technique of sheacation, let us con-

struct the direct sum of a family (Fi )iI of module sheaves over a sheaf of rings
O on a topological space X. A natural way to do this is by considering the
functor
F: U - Fi (U ), U X open.
iI
Clearly, F is an O-module in the setting of presheaves.

To examine if it actually
is a sheaf, we consider an open covering U = U of some open subset
U X and look at the canonical diagram

() Fi (U ) - Fi (U ) - - Fi (U U ).
iI iI , iI
Since direct sums of ordinary modules or abelian groups may be viewed as

restricted cartesian products and since the diagrams

Fi (U ) - Fi (U ) - - Fi (U U ), i I,
,
are exact for all i I, we see that the left map of () is injective with its
image being contained in the kernel of the right double maps. In particular,
F satises the sheaf condition (i) of 6.3/2. On the other hand, any element of
the latter kernel determines a family of sections ri Fi (U ) and such a family
gives rise to an element in iI Fi (U ) only if ri = 0 for almost all i I. If I
is nite, there is no problem and the diagram () will be exact so that F is a
sheaf in this case. Similarly, if is nite, the diagram () is seen to be exact.
But we cannot exclude innite coverings in general, and this is the reason why
F will fail to satisfy the sheaf condition (ii) of 6.3/2 in typical cases where I
is innite. Therefore we are obliged to apply Proposition 5 in order to pass to
the sheacation of F; the latter is of type F + according to Lemma 6, since
F satises already the sheaf condition (i) of 6.3/2. Using Proposition 5 again,
it follows from our construction that for any O-module sheaf G the canonical
morphisms Fi -F - F + give rise to bijections

HomO (F + , G) - HomO (F, G) - HomO (Fi , G).
iI
In particular, the morphisms Fi - F + , i I, satisfy the universal property

of a direct sum. Therefore we see that, from a categorical point of view, F + is
indeed the correct construction for the direct sum of the O-modules Fi in the
setting of module sheaves.
Exercises
1. Give an example of a non-trivial presheaf of abelian groups such that its associ-
ated sheaf is the zero sheaf.
2. Two covariant functors F : D - C and G : C - D between categories C and
D are said to be adjoint if there are bijections Hom(F (Y ), X) Hom(Y, G(X))
that are functorial on objects X in C and Y in D. Interpret the sheacation
functor for presheaves on a topological space as part of a pair of adjoint functors.
3. Let : P - P be a monomorphism of presheaves, i.e. a morphism of pre-
sheaves satisfying (ker )pre = 0. Show that the associated morphism of sheaves
++ : P ++ - P ++ is a monomorphism as well. Conclude for any morphism of
sheaves : F - G that im , the image of in the sense of sheaves, can be
viewed as a subsheaf of G.
4. Inductive limit of sheaves: Let (Fi )iI be an inductive system of sheaves of sets
(resp. abelian groups, resp. rings, etc.) on a topological space X. Dene a presheaf
on X by assigning to each open subset U X the limit limiI Fi (U ). Show that

the associated sheaf, denoted by limiI Fi , satises the universal property of an

inductive limit in the corresponding category of sheaves.
5. Consider the 1-sphere X = {z C ; |z| = 1} as a topological space under the
topology induced from C. Let CX be the sheaf of continuous real valued functions
on X and RX the subsheaf of locally constant functions with values in R. Show
that G = (CX /RX )pre , the quotient in the context of presheaves, is not a sheaf.
6. Skyscraper sheaf : Consider a closed point x of a topological space X together
with an abelian group F . The skyscraper sheaf on X with stalk F at x is dened
by associating to any open subset U X the object F if x U and 0 otherwise.
Show that the resulting presheaf F is a sheaf, indeed, and that the latter is
uniquely characterized up to canonical isomorphism by the fact that Fx = F and
Fy = 0 for all y = x.
7. Sheaf Hom: Consider a topological space X together with sheaves of abelian
groups F, G on it. For any open subset U X set
Hom(F, G)(U ) = Hom(F|U , G|U )
where Hom(F|U , G|U ) denotes the set of sheaf morphisms from the restriction F|U
of F on U to the restriction G|U of G on U . Show that Hom(F, G) is canonically
a presheaf of abelian groups and that the latter is actually a sheaf.
6.6 Construction of Ane Schemes 241
6.6 Construction of Ane Schemes
Let A be a ring and X = Spec A its spectrum. As in example (4) of Sec-

tion 6.3, consider the category D(X) of all open subsets of type D(f ) X
where f A, with inclusions as morphisms. As we have seen, there is a well-
dened contravariant functor
OX : D(X) - Ring,
D(f ) - Af ,
D(f ) D(g) - Ag - Af ,
which associates to an object D(f ) the localization Af and to an inclusion

D(f ) D(g) the canonical morphism Ag - Af . It is viewed as a presheaf
on X or, more precisely, on the basic open subsets of type D(f ) X.
For any element f A, we can look at the open sets D(g) Spec Af given
by elements g Af . To underline the fact that we are considering open subsets
in Spec Af instead of Spec A, let us write Df (g) in place of D(g).
Lemma 1. Let A be a ring and X = Spec A its spectrum. Fixing an element

f A, look at the localization Af and its associated spectrum Xf = Spec Af .
(i) The map a : Xf - D(f ) X induced from the canonical morphism
: A - Af is a homeomorphism .
(ii) The equation a 1 (D(g)) = Df ( (g)) holds for g A. Furthermore,
a 1
induces a bijective correspondence between sets of type D(g) D(f ) where
g A and the open sets of type Df (h) where h Af .
(iii) The restriction of the functor OX : D(X) - Ring to the subcategory
induced on D(f ) is equivalent to the functor OXf : D(Xf ) - Ring.
Proof. Assertion (i) was already proved in 6.2/8 while the rst part of (ii) was
established in 6.2/4. Furthermore, the second part of (ii) is clear as well, since
h = fgn Af for g A and n N yields
fg
Df (h) = Df = Df ( (f g)) = a 1 (D(f g)),
f n+1
and since D(f g) D(f ).

To obtain (iii) use the canonical isomorphism
Af g - (Af ) (g)
for g A, which is a special case of 1.2/10.
Proposition 2. For a ring A and its spectrum X = Spec A, the above functor
OX : D(X) - Ring satises the sheaf properties 6.3/2 (i) and (ii) if we

restrict ourselves to basic open coverings of type D(f ) = D(f ) where
f, f A.

Proof. Consider a basic open covering D(f ) = D(f ) as mentioned in the
assertion. Then we conclude from Lemma 1 that we may assume without loss
of generality f = 1 and, hence, D(f ) = X. Since X is quasi-compact by 6.1/10,
we know that
the covering under consideration admits a nite subcover, say of
type X = ni=1 D(fi ) for functions f1 , . . . , fn A. Let us look at such a covering
rst. In order to check that the rst sheaf property holds, we have to show that
the canonical map
n
A - Afi
i=1
is injective.
To do this, let g A be an element of the kernel. Then for each i,
1 i n,
there is an exponent ri N such that firi g = 0. Furthermore, X = ni=1 D(fi )
is equivalent to
n
= V (fi ) = V (f1 , . . . , fn )
i=1
and by 6.1/5 to A = (f1 , . . . , fr ), or even to A = (f1r1 , . . . , fnrn ), since the radical

of an ideal a A equals the unit ideal if and onlyif a itself is the unit ideal.
Hence, there are elements a1 , . . . , an A such that ni=1 ai firi = 1, and we get

n
n
g= ai firi g = ai 0 = 0.
i=1 i=1
This establishes the rst sheaf property.

The second sheaf property is a little bit more involved. Using the relation
D(fi ) D(fj ) = D(fi fj ) we have to show that the canonical diagram

n
-

n
A - Afi - Afi fj
i=1 i,j=1
is exact. Since the injectivity of the map on the left has just been shown, look
at an element of the kernel of the two maps on the right-hand side, hence, at
an element

n
(gi )i Afi such that gi |D(fi fj ) = gj |D(fi fj ) for all i, j.
i=1
There is a canonical commutative diagram

Afi
-
A Afi fj
-
-
Afj
which we will apply for discussing the coincidence of gi and gj on D(fi fj ).

Namely, for each i there are equations gi = fini hi in Afi for suitable exponents
ni N and elements hi A. Furthermore, from gi |D(fi fj ) = gj |D(fi fj ) we obtain
the equations
nj
fini hi = fj
n
hj and fj j hi fini hj = 0 in Afi fj .
Thus, we can nd some exponent r N such that
n n +r
(fi fj )r (fj j hi fini hj ) = 0 and fj j (fir hi ) = fini +r (fjr hj )
hold in A. Since we are dealing with only nitely many indices i, j, we may
assume that r is independent of i and j. Now, using the fact that f1 , . . . , fn
generate the unit ideal in A, there are elements b1 , . . . , bn A satisfying

n
bi fini +r = 1 in A.
i=1
Then we look at the element

n
g= bi fir hi A
i=1
and claim that g|D(fi ) = gi for all i. Indeed, using the above relations, the
following equalities hold in Afi for a xed i:

n
g= bj fjr hj
j=1

n
= bj fini +r (fjr hj ) fini r
j=1

n
(fir hi ) fini r
n +r
= bj f j j
j=1
n
hi fini
n +r
= b j fj j
j=1
= gi
Therefore g|D(fi ) = gi for all i and we see that the functor OX satises sheaf
properties as claimed, at least if the covering under consideration is nite.
Now consider an open covering U = (D(f )) of X that is not necessarily
nite. As we have already pointed out, U admits a nite subcover U , say cor-
responding to indices 1 , . . . , n . Then look at the canonical commutative
diagram

A - H 0 (U, OX )

?

A - H 0 (U , OX )
where we have used the notation

-
H 0 (U, OX ) = ker OX (D(f )) - OX (D(f f )) ,
(, )
of Section 6.5 and where : H 0 (U, OX ) - H 0 (U , OX ) is the restriction of the

canonical projection

n
OX (D(f )) - OX (D(fi )).
i=1
Note that may also be interpreted as the canonical map obtained in the setting
of Section 6.5 when viewing U as a renement of U. From the nite case, which
has been settled above, we know that is bijective and, in order nish our
proof, the same has to be shown for . The bijectivity of is clear if we know
that is injective. Therefore look at an element (g ) H 0 (U, OX ) whose
-image is trivial. Then the equations
g |D(f )D(fi ) = gi |D(f )D(fi ) , , i = 1, . . . , n,
show that each g trivializes with respect to the nite cover U |D(f ) . Because
we know already that OX satises the sheaf properties for such coverings, g
must be trivial. Hence, indeed, is injective and we are done.
Now we want to extend the functor OX : D(X) - Ring, which so far is

only dened on the basic open subsets of X, to a sheaf of rings OX on all of
X = Spec A. To do this, set
OX (U ) = lim OX (D(f )) = lim Af ,

D(f )U f A with D(f )U
for open subsets U X, where the rst projective limit runs over all open
subsets in X of type D(f ) that are contained in U , and the second one over
all f A such that D(f ) U . Then we conclude from the universal property
of projective limits that OX is a functor on the category of open subsets in X.
Furthermore, if U is a basic open subset, say U = D(g) for some g A, then
OX (U ) = limD(f )U OX (D(f )) is canonically isomorphic to OX (D(g)) = Ag

and we see that, indeed, the restriction of OX to D(X) yields a functor, which
is isomorphic to OX .
Theorem 3. Let A be a ring and X = Spec A its spectrum. The above functor
OX yields a sheaf of rings on X, extending the functor OX : D(X) - Ring
as considered in Proposition 2.
Proof. To check the sheaf conditions for OX , consider an open set U X and
an open covering of U . Rening the latter
if necessary, we will rst assume that
the covering is of the special type U = D(f ) for elements f A. Then,
for arbitrary g A such that D(g) U , there is a canonical commutative

diagram
-

lim Af - Af - Af f

D(f )U ,
? ? -
?
Ag - Agf - Agf f
,
where all squares are commutative if, on the right-hand side, corresponding
horizontal arrows are considered. Interpreting the projective limit as

lim Af = (af )f Af ; af |D(f ) = af for D(f ) D(f ) ,

D(f )U D(f )U
the maps from limD(f )U Af to Af and Ag are given by the projections of

D(f )U Af onto the corresponding factors and it becomes clear that we have

- -
im lim Af Af ker Af - Af f .

D(f )U ,
Furthermore, we know from Proposition 2 that the lower row of the diagram is
exact, since the sets D(gf ) = D(g)D(f ) form an open covering of D(g) U .
We claim that the upper row is exact as well. Indeed, the injectivity of
-
the maps Ag Agf where g varies over all elements in A such that
D(g) U shows that limD(f )U Af -
Af must be injective. Further-
-
more, if (h ) belongs to the kernel of Af - , Af f , this
element
induces for every g A where D(g) U an element of the kernel of
-
A - A and, thus, by the exactness of the lower row, a
gf
, gf f

well-dened element hg Ag . Then (hg )g D(f )U Af represents an element
of limD(f )U Af , which is mapped to (h ) , and we see that the upper row

of the diagram is exact. Since the latter may be identied with the canonical
diagram
-

OX (U ) - OX (D(f )) - OX (D(f ) D(f )),
,
it follows that OX satises the sheaf conditions for the covering (D(f )) of
U , as claimed.
It remains to look at an open covering U = (Ui )iI of U of general type. In
order to show that OX satises sheaf conditions for U, we proceed similarly as
in the proof of Proposition 2. Namely, covering each Ui of U by basic open sets
of type D(f ) X, we obtain a renement U of U which is of the special type
considered in the rst part of our proof. Furthermore, there is a commutative
diagram

OX (U ) - H 0 (U, OX )

?

OX (U ) - H 0 (U , OX )
where and are the canonical maps and is the restriction morphism cor-
responding to the renement U of U. Since U is of the special type dealt
with above, we know already that is bijective, and the same has to be
checked for . To
do this, we show that is injective, looking at an element
(hi ) ker iI OX (Ui ). Then, for xed i I, the component hi is trivial
on each set of U that is contained in Ui . Since these open sets cover Ui and
since OX has already been recognized as being a sheaf with respect to such a
covering, hi must be trivial. Consequently, is injective and we see that OX
satises sheaf properties for the covering U, as claimed.
To simplify our notation, we will write OX again instead of OX , calling this

the structure sheaf of the spectrum X = Spec A. Note that the ring OX (U ) for
any open subset U X may be interpreted as the ring of all functions on U
that locally in a neighborhood of any point x U are of type fg with suitable
global functions f, g A where, of course, f and g may depend on x and satisfy
g(x) = 0.
As the construction method used in the proof of Theorem 3 can be applied
to more general situations, we want to formulate it in more detail.
Lemma 4. Let X be a topological space and B a basis of the topology of X,

i.e. a system of open subsets such that
(i) U, V B = U V B,
(ii) every open subset in X is a union of sets from B.
Furthermore, let O : B - C be a functor from B viewed as a category
with inclusions as morphisms, to a category C (of sets, abelian groups, etc.)
and
assume that O satises the sheaf conditions for coverings of type U = U
where U, U B. Then O can be extended to a sheaf O on all open subsets of
X and this extension is unique up to canonical isomorphism.
Proof. For open subsets U X set
O(U ) = lim O(U )

U B with U U
and apply the arguments given in the proof of Theorem 3 where we have used
the system of all sets of type D(f ) for f A as a basis of the topology of
X = Spec A.
Next we want to discuss a module variant of the structure sheaf OX which

has just been constructed. Doing so, let A be a ring with spectrum X = Spec A
and M an A-module. To associate a sheaf F on X to M , we set

F D(f ) = Mf
for elements f A where Mf = M A Af is the localization of M by the

multiplicative system generated by f in A; see Section 4.3. Just as for OX
one shows that F, as a functor from the category D(X) of all basic open
subsets of type D(f ) X to the category of groups or A-modules, satises
the sheaf conditions of 6.3/2 and, furthermore, that F extends to a sheaf of
A-modules on all open subsets of X. This sheaf is denoted by F again, or by
M in order to underline the dependence on the A-module M . Since Mf is,
in fact, an Af -module for every f A, it is easily checked that the module
structure on F = M extends canonically to an OX -module structure in the
sense of Section 6.3. Using the same notation again, F = M is called the
module sheaf associated to the A-module M or the OX -module associated to
M . Thus, associated modules can be characterized as follows:
Theorem 5. Let A be a ring, X = Spec A its spectrum, and M an A-module.

Then the functor
D(X) - A-Mod, D(f ) - Mf = M A Af ,
mapping an inclusion of basic open subsets D(f ) D(g) to the canonical map
Mg - Mf obtained via localization extends to a sheaf of OX -modules M
on X, called the OX -module sheaf associated to M . The latter is unique up to
canonical isomorphism.
In this context, OX itself may be viewed as the sheaf A associated to A.

Further basic properties of associated modules will be discussed in Section 6.8.
Denition 6. A ringed space is a pair (X, OX ) where X is a topological space

and OX a sheaf of rings on X. We call OX the structure sheaf of the ringed
space.
A morphism of ringed spaces (X, OX ) - (Y, OY ) is a pair (f, f # ) where
f: X - Y is a continuous map and f # : OY - f (OX ) a morphism of
sheaves of rings on Y . Here f (OX ) stands for the sheaf on Y that is given by
V - OX (f 1 (V )) and canonical restriction morphisms. Thus f # consists of
a system of ring homomorphisms
f # (V ) : OY (V ) - OX (f 1 (V )), V Y open,
which are compatible with restriction morphisms.
Note that the composition of morphisms between ringed spaces is dened

in a natural way. Furthermore, if (X, OX ) is a ringed space and U X an open
subset, we can consider its restriction to U , namely the ringed space (U, OX |U ).
The injection U - X may canonically be viewed as a morphism of ringed
spaces; it is a so-called open immersion of ringed spaces.
Of course, for a morphism of ringed spaces (f, f # ) : (X, OX ) - (Y, OY ),

the continuous map f : X - Y is the main ingredient because it determines
how to map the points of X to the points of Y . On the other hand, as we
are considering ringed spaces, namely topological spaces together with rings of
functions on them, it is of interest to know how the functions on Y can be
pulled back to functions on X, like this is done in a true function setting via
composition with f . Since there is no automatic denition of such a pull-back,
the corresponding data have to be provided; this is the role played by f # .
Remark 7. Any morphism of ringed spaces (f, f # ) : (X, OX ) - (Y, OY )

canonically induces ring homomorphisms
fx# : OY,f (x) - OX,x , x X.
Proof. For a point x X and open subsets V Y such that f (x) V , the
compositions
f # (V )
OY (V ) - OX (f 1 (V )) - OX,x
are compatible with restriction morphisms of OY and, thus, induce a ring ho-
momorphism fx# : OY,f (x) - OX,x , as claimed.
A ring is called local if it contains a unique maximal ideal; see 1.2/1. For
example, the localization Ap of a ring A by a prime ideal p A is a local
ring; see 1.2/7. In particular, any eld is a local ring. A ring homomorphism
: A - B between local rings with maximal ideals m A and n B is
called local if (m) n. For example, given a ring homomorphism : A - B
and a prime ideal q B, we know that p = 1 (q) is a prime ideal in A and it
follows that the induced ring homomorphism Ap - Bq is local. However, for
any prime element p Z the canonical inclusion Z(p) - Q is not local where
Z(p) is the localization of Z by the prime ideal (p) Z generated by p.
Denition 8. A locally ringed space is a ringed space (X, OX ) such that the
stalks OX,x at all points x X are local rings.
If (X, OX ) and (Y, OY ) are locally ringed spaces, a morphism of ringed
spaces (f, f # ) : (X, OX ) - (Y, OY ) is called a morphism of locally ringed
spaces if all maps fx# : OY,f (x) - OX,x for x X are local.
Isomorphisms of ringed (resp. locally ringed) spaces are dened as usual in

the categorical sense.
Proposition 9. (i) Let A be a ring, X = Spec A its spectrum, and OX the sheaf
on X associated to A. Then (X, OX ) is a locally ringed space with OX,x = Ax
as stalk at any point x X.
(ii) Let A and B be rings with spectra X = Spec A and Y = Spec B. Then
the canonical map

: Hom (X, OX ), (Y, OY ) - Hom(B, A), (f, f # ) - f # (Y ),
from the set of morphisms of locally ringed spaces (X, OX ) - (Y, OY ) to the
set of ring homomorphisms B - A is bijective; see 7.1/3 for a more general
version of this fact.
(iii) For any morphism of locally ringed spaces (f, f # ) : (X, OX ) - (Y, OY )
as in (ii) and a point x X, the associated map between stalks
fx# : OY,f (x) - OX,x
coincides canonically with the map Bf (x) - Ax obtained via localization from
f # (Y ) : B - A.
Proof. Let us start with assertion (i). Due to Proposition 2 and the construction
in Theorem 3 we know that OX is a sheaf of rings on X. Thus, it remains to
show that the stalk OX,x at a point x X is canonically isomorphic to the
localization Ax . To do this x a point x X. Since the system D(x) of basic
open neighborhoods D(g) of x in X is conal in the system U (x) of all open
neighborhoods of x, in the sense that given any U U (x) there is always a
D(g) D(x) such that D(g) U , we can write
OX,x = lim OX (D(g)) = lim Ag .

D(g)x g(x)=0
Doing so, we want to show that the localization Ax at the prime ideal px A
satises the universal property of an inductive limit of the localizations Ag for
g varying over A px , where g px is another way to say g(x) = 0. First
note that there is a canonical family of ring homomorphisms Ag - Ax for
g A px , which is compatible with restriction morphisms of type Ag - Ag
for x D(g ) D(g). Now consider another family of ring homomorphisms
g : Ag - B for g A px to some ring B and assume that the g are com-
patible with restrictions of type Ag - Ag for x D(g ) D(g). Then, taking
g = 1, we obtain a homomorphism 1 : A - B, which factorizes through all
g : Ag - B for g A px . It follows that each g A px is mapped
via 1 to a unit in B and, furthermore, that all g factor uniquely through a
homomorphism : Ax - B. This shows that Ax is an inductive limit of the
Ag for g A px and that, accordingly, the canonical map OX,x - Ax is an
isomorphism.
To verify assertion (ii) we dene a map

: Hom(B, A) - Hom (X, OX ), (Y, OY ) ,
which will serve as an inverse of . To do this, let : B - A be a ring

homomorphism and consider the corresponding map between spectra
a
: Spec A - Spec B, p - 1 (p).
Now set up a morphism of locally ringed spaces

() = (f, f # ) : (X, OX ) - (Y, OY )
as follows. Let f = a and take

f # D(g)) : OY (D(g) = Bg - A(g) = OX D((g)) = OX f 1 (D(g))
for g B as the canonical map Bg - A(g) derived from : B - A via

localization. More general open subsets V Y can be covered by sets of type
D(g). Therefore we can derive a map f # (V ) : OY (V ) - OX (f 1 (V )) from the
f # (D(g)) for D(g) V by using the sheaf condition 6.3/2 (ii). The compatibil-
ity properties of localizations show that this way we obtain a morphism of ringed
spaces. Furthermore, passing to inductive limits at a point x X, respectively
f (x) Y , we see from (i) that the morphism of stalks fx# : OY,f (x) - OX,x
coincides with the canonical map Bf (x) - Ax , which is local. Hence, (f, f # ) is
a morphism of locally ringed spaces. Also note that this observation will settle
assertion (iii), once we have nished (ii).
Our construction of () = (f, f # ) yields f # (Y ) = for homomorphisms
: B - A so that ( ()) = and, hence, = id. As the remaining
relation = id will be provided by Lemma 10 below, we are done.
Lemma 10. Let A, B be rings with spectra X = Spec A, Y = Spec B and

(f, f # ) : (X, OX ) - (Y, OY ) a morphism between the attached locally ringed
spaces. Then the morphism (f, f # ) is uniquely determined by the ring homo-
morphism f # (Y ) : OY (Y ) - OX (X).
Proof. Writing = f # (Y ), let us show f = a rst. For x X there is the

following commutative diagram
=f # (Y )
B - A
? ?
Bf (x) A

x

fx#
OY,f (x) - OX,x ,
where the homomorphism fx# is local by our assumption and, thus, maps the
maximal ideal generated by pf (x) in OY,f (x) into the maximal ideal generated by
px in OX,x ; use 1.2/7 to see that these ideals are maximal. In other words, we
have
(fx# )1 (px OX,x ) = pf (x) OY,f (x) .
Since px OX,x and pf (x) OY,f (x) admit as preimages in A and B the ideals px and
pf (x) , we conclude that 1 (px ) = pf (x) and, hence, f (x) = a (x), as claimed. In
particular, the map f : X - Y is uniquely determined by the homomorphism
f # (Y ) : OY (Y ) - OX (X).
In order to check that the morphism f # : OY - f OX , consisting of the

homomorphisms f (V ) : OY (V )
# - 1
OX (f (V )) for V Y open, is uniquely
determined by , we consider for such a V the commutative diagram
=f # (Y )
OY (Y ) - OX (X)
? ?
f # (V )
OY (V ) - OX (f 1 (V )) .
In the special case where V = D(g) for some g B, the preceding diagram be-
comes more concrete: we get f 1 (D(g)) = D((g)) from 6.2/4 and the diagram
is of type
=f # (Y )
B -A
? f # (V )
?
Bg - A(g) .
Therefore we see from the universal property of the localization Bg that f # (V )
is uniquely
determined by . If V is more general, choose a basic open covering
V = D(g ) where g B. Then, given b OY (V ), its restrictions
f # (V )(b)|D((g )) = f # (D(g ))(b|D(g ) ), ,
are uniquely determined. Applying the sheaf properties of OX , it follows that

f # (V )(b) is uniquely determined by as well.
Starting out from a ring A, we constructed in Proposition 2 and Theorem 3

the associated sheaf OSpec A on the spectrum Spec A, thereby obtaining a locally
ringed space (Spec A, OSpec A ). Locally ringed spaces of this type will serve as
prototypes for so-called schemes, a notion introduced by A. Grothendieck.
Denition 11. An ane scheme is a locally ringed space (X, OX ) such that
there is an isomorphism of locally ringed spaces (X, OX ) - (Spec A, OSpec A )
for some ring A.
A scheme1 is a locally ringed space (X, OX ) such that there exists an open
covering (Xi )iI of X where (Xi , OX |Xi ) is an ane scheme for all i I. Here
OX |Xi is the restriction of the sheaf OX to the open subset Xi X.
Finally, a morphism of schemes is meant as a morphism of locally ringed
spaces.
In most cases, we will just write X in order to refer to a scheme (X, OX ). The
sheaf OX is called the structure sheaf of X. Likewise, for a morphism of schemes
(f, f # ) : (X, OX ) - (Y, OY ) we will use the simpler notation f : X - Y.
1
The original notion of Grothendieck was prescheme, while the term scheme was reserved
for a prescheme that is separated in the sense of 7.4. However, this convention is not followed
any more today.
If (X, OX ) is a scheme and U X an open subset, we call (U, OX |U ) an open

subscheme of X; it is a scheme again, as is seen using Lemma 1 (iii). Sometimes
the notation (U, OX ) is used in place of OX (U ), calling this ring the ring of
sections of OX over U . If (U, OX |U ) is an ane scheme, U is also referred to
as an ane open subset of X. For any morphism of schemes f : X - Y and
an open subset U X we can dene the restriction of f to U . This is the
morphism of schemes f |U : U - Y , which is meant as the composition of the
open immersion (U, OX |U ) - (X, OX ) with f : (X, OX ) - (Y, OY ). Also
note that any open subset V Y determines an open subset f 1 (V ) X as
well as a unique morphism of schemes f : f 1 (V ) - V making the diagram
f
X - Y
6 6

f
f 1 (V ) - V
commutative.
It is clear that schemes together with their morphisms form a category,
which we will denote by Sch, and that the ane schemes dene a subcategory
of Sch, in fact, a full subcategory, which means that the morphisms X -Y
between two ane schemes X and Y are the same in the category of ane
schemes and the category Sch of all schemes. As usual, the set of morphisms
between two objects X, Y of Sch is denoted by Hom(X, Y ). Furthermore, for a
xed scheme S, we can consider the category Sch/S of relative schemes over S.
As was explained in Section 4.5, the objects of Sch/S are morphisms of schemes
of type X - S where X varies over the objects of Sch and the morphisms
in Sch/S are given by commutative diagrams of type
X - Y
-
S.
We usually talk about S-schemes and S-morphisms, referring to S as the base
scheme of the relative schemes. For two relative schemes X, Y in Sch/S the set
of S-morphisms X - Y is denoted by HomS (X, Y ).
According to our denition, there is a bijective correspondence between
ane schemes and rings. Furthermore, by Proposition 9, morphisms of ane
schemes correspond bijectively to ring homomorphisms. More specically, we
can state:
Proposition 12. The category of ane schemes is equivalent to the opposite

of the category of rings.
For a thorough justication of the assertion we would have to consider

the contravariant functor Spec from the category of rings to the category
of ane schemes that assigns to a ring A the corresponding ane scheme

(Spec A, OSpec A ) and to a ring homomorphism B - A the corresponding
morphism (Spec A, OSpec A ) - (Spec B, OSpec B ) as characterized in Proposi-
tion 9 (ii). On the other hand,
X - OX (X),
X - Y - f # (Y ) : OY (Y ) - OX (X)
sets up a contravariant functor from the category of ane schemes to the cat-
egory of rings, and we see easily from Proposition 9 (ii) that both possible
compositions of these functors are equivalent to the identity.
Let us end this section by some remarks concerning the topology of schemes.
Of course, since ane schemes are Kolmogorov spaces by 6.1/8, the same is
true for arbitrary schemes. Furthermore, since any non-zero ring admits a max-
imal ideal, we know from 6.1/6 that any non-empty ane scheme admits a
closed point. To generalize this assertion to the non-ane case, we need a quasi-
compactness assumption.
Proposition 13. Let X be a non-empty Kolmogorov space that is quasi-

compact. Then X contains a closed point.
Proof. Using the quasi-compactness of X, we see using Zorns Lemma that X,

if it is non-empty, contains a minimal non-empty closed subset Z. We claim
that Z consists of just one point. Indeed, assume there are two dierent points
x, y Z. Then, by the Kolmogorov property, we can nd an open subset U X
containing just one of them, say x U . It follows that x does not belong to
the closure {y} of y in Z and, consequently, that Z cannot be a minimal closed
subset in X.
For schemes X, the argument given in the above proof does not look very
satisfying, since it is not constructive. Instead one might be tempted to look at
a non-empty ane open part U X. Then, by 6.1/6, there is a point x U
that is relatively closed in U . However, from this we cannot conclude that x will
be closed in X, as we have explained already in the introduction of the present
chapter. On the other hand, see 8.3/6 for a class of schemes X where every
point that is relatively closed in some open part of X is already closed in X.
Exercises
1. Let X = Spec Z. Describe the structure sheaf OX of X by exhibiting all restriction
morphisms. Do the same for X = Spec Ct/(t2 t) and X = Spec Ct/(t3 t2 )
where t is a variable.
2. Consider the ane scheme X = Spec Zt and its open subschemes X1 , X2 given
by the basic open subsets D(2), D(t) X where t is a variable. Show that
X1 X2 denes an open subscheme in X that is not ane. Hint: Determine the
ring OX (X1 X2 ).
3. Let X be a scheme containing at most two points. Show that X is ane.
4. Consider a topological space X with three points x0 , x1 , x2 and open subsets
, X0 = {x0 }, X1 = {x0 , x1 }, X2 = {x0 , x2 }, X.
For a eld K and a variable t dene a presheaf OX on X by setting
OX () = 0, OX (X0 ) = K(t), OX (X1 ) = OX (X2 ) = OX (X) = Kt(t)
with canonical restriction morphisms. Show that OX is a sheaf and that (X, OX )
is a scheme that cannot be ane.
5. Disjoint union
! of schemes: Let (Xi )iI be a family of schemes. Dene the disjoint
union X = iI Xi as a ringed space; it is a scheme again. Assuming that all Xi
are non-empty ane, show that X is ane if and only if I is nite. Hint: Use
Exercise 1.1/6.
6. Show that there is a unique morphism of schemes X - Spec Z for any scheme
X. Conclude that the category of schemes is equivalent to the one of relative
schemes over Spec Z.
7. For a scheme X and a point x X show that there exists a canonical morphism
of schemes Spec OX,x - X. Its image consists of all points z X satisfying
x {z}, i.e. such that x is a specialization of z.
8. Given a scheme X and a global section f OX (X), write Xf for the set of all
points x X where f does not vanish; more precisely, x Xf if and only if the
germ fx OX,x is a unit. Show:
(a) Xf is open in X.
(b) Xf = D(f ) if X is ane.
(c) There is a canonical ring morphism f : (OX (X))f - OX (Xf ).
(d) f is injective if X is quasi-compact.

(e) f is an isomorphism if there exists a nite ane open covering (Xi )iI of X
such that all intersections Xi Xj , i, j I, are quasi-compact.
If X is not ane, the ring OX (X) of global sections on X can be quite small.
For example, the projective n-space X = PnK over a eld K, to be constructed
in Section 7.1, satises OX (X) = K. Therefore one cannot expect that the open
sets of type Xf will form a basis of the topology on X.
9. A criterion for ane schemes: Show that a scheme X is ane if and only if
there are global sections f1 , . . . , fr generating the unit ideal in OX (X) with the
property that the subsets Xf1 , . . . , Xfr X as in introduced in Exercise 8 above
are ane. Hint: To establish the if part set A = OX (X) and construct a mor-
phism of schemes : X - Spec A that restricted to each Xf corresponds to
i
the canonical morphism OX (X) - OX (Xf ). Use Exercise 8 to show that
i
induces isomorphisms 1 (Spec Afi ) = Xfi - Spec Afi and, hence, that is
an isomorphism itself.
6.7 The Ane n-Space 255
6.7 The Ane n-Space
Let R be a ring and S = Spec R its spectrum, which we want to view as an ane
scheme, in fact, as a base scheme over which we construct the ane n-space as
a relative S-scheme. To do this, choose variables t1 , . . . , tn and consider
AnS = Spec Rt1 , . . . , tn
as an S-scheme under the morphism AnS - S induced from the canonical

inclusion R - Rt1 , . . . , tn . Then AS is called the ane n-space over S
n
and we will often use the notation AnR as an abbreviation of AnSpec R . For any
R-algebra R let AnS (R ) be the set of all S-morphisms Spec R - AnS ; this is
the set of so-called R -valued points of AnR . Since the set of R-homomorphisms
Rt1 , . . . , tn - R corresponds bijectively to the set (R )n of all n-tuples
with entries in R , via the map - ((t1 ), . . . , (tn )), we see that the set of

R -valued points of AnR is given by
AnR (R ) = (R )n .
In particular, this justies the term ane n-space.

We want to look a bit closer at the ane n-space AnR in the case where R
is a eld, say R = K. Let K be an algebraic closure of K. As we have seen
already at the end of Section 3.2, the map
HomK (Kt1 , . . . , tn , K) - Spec Kt1 , . . . , tn , - ker ,
gives rise to a canonical map
: AnK (K) - Spm Kt1 , . . . , tn
from the set of K-valued points of AnK to the set of closed points of AnK . Indeed, a
morphism Spec K - AnK is given by a K-morphism : Kt1 , . . . , tn - K
and its image im is a subring of K that is integral over K. Then im is a eld
by 3.1/2 and we can conclude that ker is a maximal ideal in Kt1 , . . . , tn .
On the other hand, if m Kt1 , . . . , tn is an arbitrary maximal ideal,
we know from 3.2/4 that the quotient K = Kt1 , . . . , tn /m is a nite eld
extension of K. Therefore we may embed K via a K-homomorphism into K
and it follows that the composition
Kt1 , . . . , tn - Kt1 , . . . , tn /m - K
yields a K-valued point x AnK (K) such that (x) = m. In particular, is

surjective.
To determine the bers of , consider two points x, y AnK (K) given by
K-homomorphisms , : Kt1 , . . . , tn - K and assume (x) = (y). Then
there exists a commutative diagram of K-homomorphisms as follows:
Kt1 , . . . , tn / ker - K

-
-

Kt1 , . . . , tn
- ?
-
Kt1 , . . . , tn / ker - K
Indeed, due to ker = ker the subelds Kt1 , . . . , tn / ker K and
Kt1 , . . . , tn / ker K are isomorphic over K and such a K-isomorphism
can be extended to a K-automorphism of K. Thus, letting the automorphism
group AutK (K) act on K n in a canonical way, we see that two points x, y K n
satisfy (x) = (y) if and only if there exists an automorphism AutK (K)
such that (x) = y. In particular, the map gives rise to a bijection
K n / AutK (K) - Spm Kt1 , . . . , tn Spec Kt1 , . . . , tn .
Since any K-automorphism of K leaves K xed, it follows that induces an

injection
K n - Spm Kt1 , . . . , tn
where the maximal ideal (x) corresponding to a point x = (x1 , . . . , xn ) K n
is given by (t1 x1 , . . . , tn xn ) Kt1 , . . . , tn . Thus, we see that, for an
algebraically closed eld K = K, the map is both, surjective and injective,
hence, denes a bijection
Kn - Spm Kt1 , . . . , tn .
We will see later in Section 7.2 for a general base scheme S that the ane
n-space AnS , like any relative S-scheme, may be imagined as the family of its
bers over S. In our special case the latter are the ane n-spaces Ank(s) where s
varies over the points in S.
Exercises
1. Let S be an ane base scheme. Show that there is a canonical bijection
HomS (T, A1S ) - OT (T ) that is functorial on S-schemes T and, thus, gives
rise to an isomorphism of functors from the category of S-schemes to the category
of sets. Later the assumption ane on S can be removed.
2. Show that the ane n-space AnS over an ane base S can be interpreted as the
n-fold cartesian product (A1S )n in the category of S-schemes. Indeed, for any
S-scheme T there is a canonical bijection HomS (T, AnS ) - HomS (T, A1S )n .
Later we will be able to remove the assumption ane on S.
3. Consider Neiles parabola X = Spec Kt1 , t2 /(t22 t31 ) over a eld K. As we can
conclude from Exercise 6.2/6, there is a morphism of schemes A1K - X that
6.8 Quasi-Coherent Modules 257
is not an isomorphism, although it is bijective as a map of sets. Show that there

cannot exist any isomorphism of schemes between X and the ane line A1K . Hint:
All stalks of the structure sheaf of A1K are discrete valuation rings, while the same
is not true for X.
6.8 Quasi-Coherent Modules
In the following we want to study the category of OX -modules on a scheme X

where by an OX -module we always mean a module sheaf, unless stated other-
wise. We begin with the ane case. So let A be a ring and X = Spec A the
associated ane scheme. As we have explained in Section 6.6, starting out from
an A-module M we can construct an associated OX -module M on X. By 6.6/5
the latter is the sheaf that on basic open subsets of type D(f ) X for f A
is given by the localization

M D(f ) = Mf = M A Af
where the tensor product yields the structure of Mf as an Af -module. For an

arbitrary open subset U X we can take a basic open covering U = (D(f ))
of U given by suitable elements f A and describe the sections over U in terms
of sections on basic open sets by

M (U ) = H 0 (U, M ) M f , OX (U ) = H 0 (U, OX ) Af ;

see Section 6.5 for the notion of H 0 (U, ). Also note that the structure of M (U )
as an OX (U )-module is componentwise induced from the structure of each Mf
as an Af -module. Using this point of view it becomes clear that every morphism
of A-modules : M - N gives rise to an associated morphism of OX -modules
: M - N . On a basic open subset D(f ) X, the latter is given by the
localization f : Mf - Nf of by f . Thereby we see that, in fact, M - M
denes a functor from the category of A-modules to the category of OX -modules.
Also note that A = OX , i.e. the OX -module associated to A with its canonical
A-module structure is the structure sheaf OX viewed as a module over itself.
There is a slight generalization of the construction of associated morphisms.
Consider an A-module M and an OX -module F. Then every morphism of
A-modules : M - F(X) induces a morphism of OX -modules : M - F.
Indeed, for any basic open subset D(f ) X, we know that F D(f) is an
Af -module. Hence, the composition M - F(X) - F D(f ) factors

uniquely over a morphism of Af -modules f : Mf - F D(f ) . As before,
these maps give rise to the desired morphism : M - F. In particular,
the identity map on F(X) induces a well-dened morphism of OX -modules
- F and the latter is an isomorphism if and only if F is associated to
F(X)
some A-module M (which will be isomorphic to F(X)).
Proposition 1. Let M be an A-module and Mf its localization by some ele-

ment f A. Then the restriction M |Xf , where Xf = Spec Af , is canonically
isomorphic to the OXf -module Mf associated to the Af -module Mf .
Proof. This is the module analogue of 6.6/1 (iii); use 1.2/10 in conjunction with
4.3/2.
Proposition 2. For A-modules M and N , the map
HomA (M, N ) - HomOX (M , N ), - ,
is bijective. More generally, for any A-module M and any OX -module F, the
map
HomA M, F(X) - HomO (M , F), - ,
X
is bijective.
Proof. Of course, a left-inverse of the map - is given by - (X)

for morphisms of OX -modules : M - F. The latter is a right-inverse as
well, as follows from the commutative diagram
(X)
M - F(X)
? (D(f )) ?
Mf - F D(f ) ;

namely it shows for any f A that, as a morphism of Af -modules, D(f ) is
uniquely determined by (X).
Corollary 3. The functor M - M from A-modules to OX -modules respects

direct sums, i.e. for any family of A-modules (Mi )iI there is a canonical iso-
morphism

Mi Mi .
iI iI
Proof. As we have shown in Sections 1.4 (Example 4) and 6.5, direct sums of
ordinary modules and of module sheaves are characterized by the same uni-
versal mapping property. Thus, it is enough to refer to the second bijection in
Proposition 2.
Proposition 4. The functor M - M from A-modules to OX -modules is

exact, i.e. for any exact sequence M -M - M of A-modules the
resulting sequence of OX -modules M - M - M is exact as well. In
particular, the functor M - M respects the formation of kernels, images,
and cokernels.
Proof. We know from 4.3/3 that any localization Af of A by an element

f A is at over A. Therefore, if we tensor an exact sequence of A-modules
M - M - M with a localization Af over A, this yields an exact sequence
Mf - Mf - M . In particular, any section of the kernel of M - M
f

living on some open subset of X admits local preimages in M . But then it
follows that the sequence M - M - M is exact.
Corollary 5. A sequence of A-modules M -M - M is exact if and

only if the associated sequence M - M - M is exact.

Proof. Taking into account Proposition 4, we have just to know that an

A-module E is trivial if and only if its associated OX -module E is trivial.
However, this is clear from the construction of associated modules.
Now let F be an arbitrary OX -module. We want to prove the remarkable

fact that the condition of F being associated to an A-module M , in the sense
that F is isomorphic to M , can be checked locally on X.
Theorem 6. Let X be the spectrum of a ring A and F an OX -module. Then

(i) F is associated to an A-module M .
(ii) There exists a family of elements f A such that the basic open subsets
D(f ) cover X and each restriction F|D(f ) is associated to an Af -module M f .
(iii) The following hold for every f A:
(1) given a section r F(X) such that r|D(f ) = 0, there exists n N
such that f n r = 0, and
(2) given a section s F(D(f )), there exist r F(X) and n N such
that r|D(f ) = f n s.
Proof. It is trivial that (i) implies (ii). Therefore assume condition (ii). Since X
is quasi-compact by 6.1/10, there exists a nite family (f ) of elements in
A such that X = D(f ) and each F|D(f ) is associated to an Af -module
M . Now let f A and consider a section r F(X) such that r|D(f ) = 0.
Writing r = r|D(f ) , we get r |D(f f ) = 0 or, in other words, r1 = 0 in (M )f
for all . By the module analogue of 1.2/4 (i), there is an exponent n N
such that f n r = 0 in F(D(f )) = M . Since the index set is nite, we may
assume that n is independent of . But then f n r|D(f ) = f n r = 0 for all
and, hence, since F is a sheaf, f n r must be trivial. Thus, we see that the rst
condition in (iii) is satised.
To derive the second condition in (iii), consider a section s F(D(f )) for
some f A and look at its restriction s = s|D(f f ) F(D(f f )) (M )f on
D(f ) D(f ). Then we can nd an exponent n N such that f n s extends
to a section r F(D(f )) M , where, again, we may assume that n is
independent of . Restricting r and r to D(f ) D(f ) for indices , ,
we see that the dierence
r r F(D(f f )) (M )f (M )f
becomes trivial on D(f f f ) or, equivalently, after localization by f .Using the

module analogue of 1.2/4 (i), it follows that r r is killed by a power of f ,
which we may assume to be independent of the nitely many index pairs (, ).
But then, taking n suciently big from the beginning on, we may assume that all
dierences r r F(D(f f )) are trivial. Since F is a sheaf, there is a global
section r F(X) such that r|D(f ) = r for all so that, in particular,
r|D(f f ) = r |D(f f ) = f n s for all . Using the covering D(f ) = D(f f ) in
conjunction with the sheaf property of F, we obtain r|D(f ) = f n s, as required
in the second condition of (iii).
Finally, let us show that (iii) implies (i). To do this, let M = F(X) and
observe that any f A gives rise to a unique morphism of Af -modules f such
that the diagram
M ======= F(X)
? f
?
Mf - F(D(f ))
with canonical vertical morphisms is commutative. Now, if condition (1) in (iii)
holds, f is injective, and if condition (2) holds, f is surjective. From this one
concludes that F is associated to the A-module M and we are done.
Corollary 7. Let A be a ring and X = Spec A. Furthermore, let M be an

A-module and F an OX -module that is associated to M . Then, for any ane
open subscheme U = Spec B of X, the canonical map M A B - F(U ) is
bijective and F|U is associated to M A B as B-module.
Proof. Since F(U ) is a B-module, the restriction morphism F(X) - F(U )

factors through a canonical morphism of B-modules : M A B - F(U ).
The latter induces a morphism of OU -modules : M
A B - F|U by Propo-
sition 2. Furthermore, for any f A such that D(f ) U , the canonical map
A - B induces an isomorphism Af - Bf via localization and, hence, using
4.3/2, a commutative diagram
M A B - F(U )
? f
?
M A B f - M A Af
where f is an isomorphism. But then is an isomorphism locally on U and,

thus, an isomorphism on U as well.
Next we want to adapt certain special properties of modules over rings to

the situation of modules on schemes. To do this consider for any index set I the
OX -module
(I)

OX = OX
iI
(I)
where, on the right-hand side, OX is meant as a module over itself. Thus, OX
is the direct sum of copies OX parametrized by I; see the end of Section 6.5 for
the direct sum of module sheaves. Note that for a ring A it is common practice
to interpret A(I) as a part of the cartesian product AI , namely as
A(I) = {(ai )iI AI ; ai = 0 for almost all i I}.
This way A(I) is an A-module, which can be interpreted as the direct sum of
copies of A parametrized by I.
Remark 8. Let U = Spec A be an ane open subscheme of a scheme X. Then,

(I)
for any index set I, the restriction of the direct sum OX to U is associated to
(I)
the A-module A(I) , i.e. OX |U A "(I) .
Proof. The construction of direct sums at the end of 6.5 in conjunction with
(I)
6.5/6 (i) shows (OX )|U (OX |U )(I) so that we can apply Corollary 3.
Let us call an OX -module F free if there exists an index set I such that F
is isomorphic to the OX -module (OX )(I) . Furthermore, an OX -module F on a
scheme X is called locally free if every point x X admits an open neighborhood
U such that F|U is free. If more specically F is locally isomorphic to OX , i.e.
if each x X admits an open neighborhood U X such that F|U OX |U ,
then F is called invertible; see 9.2 for a closer discussion of the latter property.
Denition 9. An OX -module F on a scheme X is called quasi-coherent if,

for every point x X, there exists an open neighborhood U X such that the
restriction of F to U admits an exact sequence of type
(OX |U )(J) - (OX |U )(I) - F|U - 0,
called a presentation of F|U .
In particular, the structure sheaf OX is a quasi-coherent module over itself,

on any scheme X.
Theorem 10. Let X be a scheme and F an OX -module. Then the following

(i) F is quasi-coherent.
(ii) Every x X admits an ane open neighborhood U X such that F|U
is associated to some OX (U )-module.
(iii) The restriction F|U to any ane open subscheme U X is associated
to some OX (U )-module.
Proof. In a rst step we show that (i) is equivalent to (ii). For this it is enough
to consider the case where X is ane, say X = Spec A. Assume that F is
isomorphic to the cokernel of some OX -morphism (OX )(J) - (OX )(I) . Then,
by Proposition 2, the morphism must be associated to a morphism of A-modules
: A(J) - A(I) . Since the formation of associated modules respects cokernels
by Proposition 4, we see that F is associated to the A-module coker .
Conversely, assume that F is associated to some A-module M . Then, choos-
ing a presentation
A(J) - A(I) -M -0
and passing to associated OX -modules using Proposition 4, we get a presenta-

tion for F, showing that F is quasi-coherent.
Now assume condition (ii) and consider an ane open subscheme U X.
Then we can nd an open covering U = (U ) of U such that all U are
ane and, in fact, basic open in U and such that F|U is associated to some
OX (U )-module. In this situation it follows from Theorem 6 that F|U is associ-
ated to some OX (U )-module, thereby establishing the implication from (ii) to
(iii). The reverse is trivial.
The good exactness properties of associated modules, as mentioned in the

beginning of this section, extend to quasi-coherent modules as follows:
Proposition 11. Let X be a scheme and
0 - F - F - F - 0
an exact sequence of OX -modules. If two of the modules F, F , F are quasi-

coherent, the same is true for the third one.
Proof. Since quasi-coherence is a local property on X, we may assume that X is

ane, say X = Spec A. Now if F and F are quasi-coherent, they are associated
to A-modules by Theorem 10, namely to M = F (X) and M = F(X). Further-
more, by Proposition 2, the morphism F - F corresponds to a morphism
of A-modules M - M that is injective by Corollary 5. The same result
shows that F is associated to M/M and, hence, that F is quasi-coherent.
Proceeding similarly with the morphism F - F , we can conclude that F
is associated to an A-module if the same is true for F and F .
Now assume that F and F are quasi-coherent, say associated to A-modules
M and M . Then it is easy to see that F is quasi-coherent if we jump to

the cohomological methods of Section 7.7 and use the long exact cohomology
sequence
0 - F (X) - F(X) - F (X) - H 1 (X, F ) - ...
of 7.7/4. Since H 1 (X, F ) = 0 by 7.7/7, the sequence
() 0 - M - M - M - 0,
where M = F(X), is exact. Then the canonical morphism M - F gives rise

to a commutative diagram
- - - -
0 M
M M
0

?
0 - F - F - F - 0
with exact rows, the rst one being exact by Proposition 4. But then one can
show by some diagram chase on the level of sections of M and F that M - F
must be an isomorphism. In particular, F is quasi-coherent, since this is true
for M .
A more direct attack on the quasi-coherence of F demonstrates that, quite
naturally, cohomological methods, namely from Cech cohomology as dealt with
in Section 7.6, come into play. Indeed, a crucial step in the above proof is
to show that the sequence () is exact, and for this only the surjectivity of
F(X) - F (X) has to be checked. Therefore, consider a section f F (X).
Then, locally on X, there are preimages of f in F. In other words, there exists a
(nite) basic open covering U = (Ui )iI of X such that each f |Ui admits a preim-
age fi F(Ui ). In particular, the images of all dierences fij = (fj fi )|Ui Uj
are trivial in F (Ui Uj ) and, hence, viewing F as a subsheaf of F, these be-
long to F (Ui Uj ). Now, using the notion of Cech cochains as introduced in
Section 7.6, the sections fij dene a so-called 1-cochain

f = (fij )i,jI F (Ui Uj ) F(Ui Uj ),
i,jI i,jI
and to nd a preimage of f in F(X), it is enough to nd sections fi F (Ui )

such that (fj fi )|Ui Uj = fij holds in F(Ui Uj ) for all i, j I. Then, indeed,
the dierences fi fi F(Ui ) are preimages of f |Ui coinciding on all overlaps
Ui Uj and, thus, giving rise to a global preimage of f in F(X).
To obtain the desired sections fi F (Ui ), we can use the construction that
will be given in the proof of Proposition 7.6/4. Therefore let us introduce the
modules of q-cochains

C 0 (U, F) = F(Ui ), C 1 (U, F) = F(Uij ),
iI i,jI

C (U, F) =
2
F(Uijk )
i,j,kI
for q = 0, 1, 2, where Uij = Ui Uj and Uijk = Ui Uj Uk . Using a similar

notation for F in place of F, there is a commutative diagram
d0 d1
C 0 (U, F ) - C 1 (U, F ) - C 2 (U, F )

? ? ?
d0 d1
C 0 (U, F) - C 1 (U, F) - C 2 (U, F)
with so-called coboundary maps d0 , d1 , which are given by


d0 : (gi )iI - (gj gi )|Uij ,
i,jI

1
d : (gij )i,jI - (gjk gik + gjk )|Uijk .
i,j,kI
We claim that the 1-cochain f = (fij )i,jI C 1 (U, F ) above is a cocycle in

the sense that it belongs to the kernel of the coboundary map
d1 : C 1 (U, F ) - C 2 (U, F ).
Indeed, look at the above diagram. Since d1 d0 = 0, as is easily veried,

we conclude that d1 (f) = (d1 d0 )((fi )iI ) = 0. Therefore f is a 1-cocycle in
C 1 (U, F), but then also in C 1 (U, F ) and we can conclude from Proposition 7.6/4
that f belongs to the image of d0 : C 0 (U, F ) - C 1 (U, F ). This is what we
need.
To further illustrate the concept of quasi-coherent modules on schemes, let

us discuss some niteness conditions that originate from modules over rings.
Denition 12. An OX -module F on a scheme X is called locally of nite type

if every point x X admits an open neighborhood U X together with an
exact sequence of type (OX |U )(I) - F|U - 0 where I is nite.
Furthermore, F is called locally of nite presentation if every point x X
admits an open neighborhood U X such that there is a presentation
(OX |U )(J) - (OX |U )(I) - F|U - 0
where I and J are nite.
In particular, if F is locally of nite presentation, it is quasi-coherent. The

latter may fail to be true if F is just locally of nite type; cf. Exercise 3 below.
Corollary 13. The following conditions are equivalent for a quasi-coherent

OX -module F on a scheme X:
(i) F is locally of nite type (resp. locally of nite presentation).
(ii) Every x X admits an ane open neighborhood U X such that
F|U is associated to a nite OX (U )-module (resp. an OX (U )-module of nite
presentation).
(iii) The restriction F|U to any ane open subscheme U X is associated
to a nite OX (U )-module (resp. an OX (U )-module of nite presentation).
Proof. If F is quasi-coherent, we see from Proposition 4 and Corollary 5 in

conjunction with Theorem 10 that, for each ane open subscheme U X, the
exact sequences of type (OX |U )(I) - F|U - 0 correspond bijectively to
the exact sequences of type OX (U )(I) - F(U ) - 0. Since the same is true
for presentations, the equivalence of (i) and (ii) follows.
Thus, it only remains to show that (ii) implies (iii). To do this, we may
assume that X is ane, say X = Spec A, and applying Theorem 10 again, that
F is associated to some A-module M . Using (ii) in conjunction with the quasi-

compactness of X, we know that there is a nite covering of X by basic open
sets D(f ) X, = 1, . . . , n, such that F|D(f ) is associated to an Af -module
of nite type (resp. nite presentation). In other words, the localization Mf is
an Af -module of nite type (resp. nite presentation).
Now look at the canonical ring homomorphism

n
: A - Af ,
=1
which is at, since each localization A - Af is at by 4.3/3 (i) and since the
direct sum (here occurring as a nite cartesian product) of at modules is at
by 4.2/6. As the D(f ) cover X, we see thatthis module is, in fact, faithfully
at
by the criterion 4.2/11 (iv). Now, viewing n=1 Mf as a module over n=1 Af ,
this module is of nite type (resp. nite presentation) by our assumption on the
Mf . Since n=1 Mf is obtained from M via tensoring with , it follows from
4.4/1 that M is of nite type (resp. nite presentation). Hence, we get (iii).
Finally, for completeness, let us mention that an OX -module F on a scheme

X is called coherent if the following conditions are satised:
(i) F is locally of nite type.
(ii) If U X is open and : (OX |U )n - F|U is a morphism of
OX |U -modules, then ker is locally of nite type.
In particular, any coherent OX -module F is quasi-coherent. Furthermore,
using the methods of the present section in conjunction with 1.5/7, one can
show that an OX -module F is coherent if and only if, locally on any ane open
part U X, it is associated to an OX (U )-module that is coherent in the sense
of 1.5/8.
Exercises
1. Let A be a ring and X = Spec A. Show that the functor M - M from
A-modules to quasi-coherent OX -modules respects inductive limits.
2. Let A be a ring and M an A-module. Show for X = Spec A and any point
x X that (M )x , the stalk of the associated OX -module M at x, is canonically
isomorphic to the localization Mx of M at x.
3. For a scheme X and a closed point x X consider an OX,x -module F . Let F be
the associated skyscraper sheaf on X satisfying Fx = F and Fy = 0 for y = x;
see Exercise 6.5/6. Show that F is canonically an OX -module and that the latter
is quasi-coherent if and only if Fy = 0 for all y Spec OX,x X dierent from x.
Give examples where F is quasi-coherent, as well as examples where F is locally
of nite type, but not quasi-coherent.
4. Show that the direct sum of quasi-coherent modules on a scheme X is quasi-
coherent. Can we expect the same to be true for quasi-coherent modules on
arbitrary ringed spaces, adapting Denition 9 to this case?
5. Show that the kernel, image, and cokernel of any morphism between quasi-
coherent OX -modules on a scheme X are quasi-coherent.
6. Let A be a Noetherian ring and set X = Spec A. Show that an OX -module is
coherent if and only if it is associated to a nite A-module.
7. Let X be a scheme and F an OX -module that is locally of nite type. For any
point x X where Fx = 0, show that there exists an open neighborhood U of x
satisfying F|U = 0. Conclude that the set {x X ; Fx = 0}, the so-called support
of F, is closed in X.
8. Consider a morphism of quasi-coherent OX -modules f : F - G on a scheme
X and let x be a point where the morphism of stalks fx : Fx - Gx is an
isomorphism. Assume that F is locally of nite type and G locally of nite pre-
sentation and show that there exists an open neighborhood U of x such that
f |U : F|U - G|U is an isomorphism. Hint: Use 1.5/7 in conjunction with Ex-
ercise 7 above.
9. Let A be an integral domain with eld of fractions K. Dene the constant sheaf
KX on X = Spec A by setting KX (U ) = K for each non-empty open subset
U X. Show that KX is a quasi-coherent OX -module that cannot be locally of
nite type and, thus, is not coherent, unless A is a eld.
6.9 Direct and Inverse Images of Module Sheaves
For any morphism of ringed spaces f : X - Y and a sheaf F on X, the

functor
f (F) : V - f 1 (V ), F , V Y open,
denes a sheaf on Y , called the direct image of F under f . If F is an OX -module,
then (f 1 (V ), F), for any V Y open, is an OY (V )-module with respect to
the ring morphism f # (V ) : OY (V ) - OX (f 1 (V )) and we may view f (F) as
an OY -module. Recall that the direct image sheaf f (F) was already considered
in 6.6/6 in the case where F equals the structure sheaf OX .
On the other hand, if G is a sheaf on Y , its inverse image f 1 (G) with
respect to f is obtained via the so-called adjunction formula

HomX f 1 (G), F = HomY G, f (F) ,
which has to be interpreted as a functorial isomorphism between functors from

the category of sheaves F on X to the category of sets. In this setting, f 1 is
called the left adjoint functor of f and, likewise, f the right adjoint functor
of f 1 . However, proceeding like this, some basic questions have to be settled.
Namely, it is by no means clear that f 1 (G) will exist as a sheaf on X and, if
it does, one has to check that it is unique up to canonical isomorphism. Also
note that the denition of f 1 (G) depends a priori on the type of morphisms
we allow for sheaves on X or Y . Indeed, if G is a sheaf of abelian groups on Y
equipped with the structure of an OY -module, we will in general obtain dierent
inverse images f 1 (G), viewing G as a sheaf of abelian groups on Y , or as an
6.9 Direct and Inverse Images of Module Sheaves 267
OY -module. In the rst case, the adjunction formula is considered for sheaves
F of abelian groups on X and for morphisms between such sheaves, whereas
in the second case we are dealing with OX -modules F and morphisms of OX -,
resp. OY -modules. Let us start by looking at the problem of uniqueness.
Lemma 1 (Yoneda). Let A, B be objects of a category C and consider a func-

torial morphism
h : HomC (A, ) - HomC (B, )
between functors from C to the category of sets. Then there is a unique morphism
: B - A inducing h, i.e. for all objects E in C the corresponding map
hE : HomC (A, E) - HomC (B, E) is given by - .
Furthermore, h is a functorial isomorphism if and only if is an isomor-
phism.
Proof. Look at the map hA : HomC (A, A) - HomC (B, A); it sends the identity
morphism id : A - A to a certain morphism : B - A, which we claim is
as stated. Indeed, for any morphism : A - E in C, the functorial morphism
h gives rise to a commutative diagram
hA
HomC (A, A) - HomC (B, A)
? ?
hE
HomC (A, E) - HomC (B, E) ,
where the vertical maps are given by composition with . As we have said, the
identity id HomC (A, A) is mapped horizontally to HomC (B, A) and then
to HomC (B, E), as well as vertically to id = HomC (A, E) and
then to hE () HomC (B, E). This yields
hE () =
and shows that the functorial morphism h is given by composition with .
To derive the uniqueness of , consider two morphisms , HomC (B, A)
inducing h and, hence, satisfying = for all HomC (A, E), where E
varies over all objects in C. Then, for E = A and the identity id HomC (A, E),
we see that id = id and, hence, = .
If h is a functorial isomorphism, its inverse
h1 : HomC (B, ) - HomC (A, )
is given by composition with a well-dened morphism : A - B, as we have
shown above. Using the uniqueness assertions for the compositions and
, we see that is an inverse of . Likewise, one shows for an isomorphism
that its inverse 1 gives rise to an inverse of h.
Proposition 2. Let f : X - Y be a morphism of ringed spaces and G a sheaf

(of sets, abelian groups, or rings) on Y . Then there is a sheaf f 1 (G) (of the
same type) on X, together with a functorial isomorphism

HomX f 1 (G), F - HomY G, f (F)
of functors on the category of sheaves F (of sets, abelian groups, or rings) on

X. Furthermore, f 1 (G) is uniquely determined up to canonical isomorphism;
it is called the inverse image of G with respect to f .
In more precise terms, f 1 (G) is the sheaf associated to the presheaf
U - lim G(V ), U X open,

V f (U )
where the inductive limit extends over all open subsets V Y containing f (U ).
Proof. We have to establish the existence of the left adjoint functor of f , which
can be done under quite general categorical assumptions; see Milne [19], II.2.2,
or HiltonStammbach [16], IX.5.1. However, we want to be more explicit and
prefer to show that the construction specied in the assertion yields a left adjoint
of f , relying on Lemma 1 for the uniqueness assertion. In the following, we
write f 1 (G) for the sheaf associated to the presheaf U - lim G(V )
V f (U )
for U X open.
Let us introduce a map

: HomY G, f (F) - HomX f 1 (G), F
that is functorial in F as follows. A morphism of sheaves : G - f (F)

consists of a family of morphisms

(V ) : G(V ) - f (F)(V ) = F f 1 (V ) , V Y open,
that is compatible with restriction morphisms. Given an open subset U X,

we obtain for open subsets V Y such that U f 1 (V ) morphisms
res
G(V ) - F f 1 (V ) - F(U )
and, thus, a morphism
lim G(V ) - F(U ), U X open,

V f (U )
where the limit extends over all open subsets V Y such that f (U ) V .
Varying U and passing to the associated sheaf on the left-hand side yields a
morphism
f 1 (G) - F,
which we denote by (). It is immediately clear that the resulting map
: HomY (G, f (F)) - HomX (f 1 (G), F) is functorial in F.
To show that is an isomorphism, we construct an inverse map

: HomX f 1 (G), F - HomY G, f (F)
of . A morphism : f 1 (G) - F consists of a system of morphisms
(U ) : f 1 (G)(U ) - F(U ), U X open,
that is compatible with restriction morphisms. Then, for open subsets U X

and V Y satisfying f 1 (V ) U , we can consider the composition
can (U )
G(V ) - lim G(V ) - f 1 (G)(U ) - F(U ),

V f (U )
where can is the canonical morphism of the presheaf U - limV f (U ) G(V )

to its associated sheaf f 1 (G). In particular, taking U = f 1 (V ) for open subsets
V Y , we obtain a system of morphisms

G(V ) - F f 1 (V ) = f (F)(V ), V Y open,
that is compatible with restriction morphisms and, hence, a morphism
G - f (F),
which we denote by ( ). Similarly as before, it is seen that the resulting map

: HomX (f 1 (G), F) - HomY (G, f (F)) is functorial in F. Moreover,
and are mutually inverse and provides the desired isomorphism, showing
that f and f 1 are mutually adjoint.
In Proposition 2, inverse images of module sheaves are not yet covered. To

pass to this case, we need to use tensor products on the level of sheaves. Note
that the tensor product F O G of two modules F, G over a sheaf of rings O
on a topological space X is dened as the O-module sheaf associated to the
presheaf
U - F(U ) O(U ) G(U ), for U X open.
Proposition 3. Let f : X - Y be a morphism of ringed spaces and G an

OY -module sheaf on Y . Then there exists an OX -module sheaf f (G) on X,
together with a functorial isomorphism

HomOX f (G), F - HomO G, f (F)
Y
of functors on the category of OX -module sheaves F on X. Furthermore, f (G)

is uniquely determined up to canonical isomorphism; it is called the inverse
image of the module sheaf G with respect to f .
If f 1 (G) is the inverse image of G, viewed as a sheaf of abelian groups, and
f 1 (OY ) the inverse image of OY as a sheaf of rings, then f 1 (G) is canonically
an f 1 (OY )-module. Furthermore, the morphism OY - f (OX ) given by f
corresponds to a morphism f 1 (OY ) - OX of sheaves of rings inducing an
isomorphism
f (G) f 1 (G) f 1 (OY ) OX .
Proof. The argument given in the proof of Proposition 2 can be extended

to OX -modules F and OY -modules G. Thereby we see that f 1 (G) is an
f 1 (OY )-module giving rise to a functorial isomorphism

Homf 1 (O ) f 1 (G), F - HomO G, f (F) ,
Y Y
acting as adjunction formula. Now, similarly as for ordinary modules (for ex-
ample, see the argument in the proof of Corollary 4 below), the canonical map

Homf 1 (OY ) f 1 (G), F - HomO f 1 (G) 1
X f (OY ) OX , F ,
is a functorial isomorphism. Consequently, f (G) = f 1 (G)f 1 (OY ) OX satises

the adjunction formula

HomOX f (G), F - HomO G, f (F) .
Y
As usual, the uniqueness assertion for f (G) follows from Lemma 1.
The explicit construction of inverse image sheaves in Propositions 2 and 3 in

conjunction with 6.5/6 (i) shows that the formation of the sheaves f 1 (G) and
f (G) with respect to a morphism of ringed spaces f : X - Y is compatible
with restriction to open parts on X and Y , i.e. for open subsets U X and
V Y where f (U ) V and, hence, f restricts to a morphism fU : U - V,
one has 1
f (G) |U fU1 (G|V ), f (G) |U fU (G|V ).
On the other hand, direct images f (F) are only compatible with restriction to
open parts on Y (and their preimages in X), namely

f (F) |V (fV ) (F|f 1 (V ) )
for open subsets V Y and fV : f 1 (V ) - V the morphism induced from

f . Furthermore, let us mention that f , as a functor from the category of
OX -modules to the category of OY -modules, is left exact and that f , as a
functor from the category of OY -modules to the category of OX -modules, is
right exact. Both assertions can be checked directly from our constructions or
derived formally from the adjunction formula.
Looking at the adjunction formula

HomOX f (G), f (G) - HomO G, f (f (G))
Y
for F = f (G) and any OY -module G, we see that the identity morphism
id : f (G) - f (G) corresponds to a canonical morphism f # : G - f (f (G))
of OY -modules. Furthermore, the construction of f 1 (G) and f (G) shows that
f # is induced from the canonical maps

G(V ) - G(V ) OY (V ) OX f 1 (V ) , a - a 1,
on open subsets V Y . In the special case, where G = OY , the morphism f #

coincides with the morphism OY - f (OX ) given by f . In particular, the
chain of functorial isomorphisms

HomOX f (OY ), F HomOY OY , f (F) F (X) HomOX (OX , F)
yields f (OY ) OX , according to Lemma 1.

Analogously, the adjunction formula

HomOX f (f (F)), F - HomO f (F), f (F)
Y
for G = f (F) and any OX -module F shows that the identity morphism
id : f (F) - f (F) corresponds to a canonical morphism f (f (F)) -F
of OX -modules. The latter is induced by the canonical maps

F f 1 (V ) OY (V ) OX f 1 (V ) - F(U )
on open subsets U X and V Y such that f (U ) V .

Finally, we want to discuss direct and inverse images in the case of quasi-
coherent modules on schemes. Restricting to the local situation of a morphism
between ane schemes, we have to describe direct and inverse images of asso-
ciated modules. As usual, for an ane scheme X = Spec A and an A-module
F , the associated OX -module will be denoted by F .
Corollary 4. Let X = Spec A, Y = Spec B be ane schemes and f : X - Y

a morphism given by a ring morphism : B - A. Furthermore, consider an
A-module F and a B-module G.
Then
"
f (F ) F /B , f (G) G
B A,
where F/B is the restriction of F with respect to : B - A; see Section 4.3.
Proof. We start by looking at the direct image f (F ). For g B we have

f (F ) D(g) = F D((g)) = F A A(g)
and furthermore, since A(g) = A B Bg ,
F A A(g) = F A (A B Bg ) = F/B B Bg
"
by 1.3/2. Hence, we can deduce that f (F ) = F /B .
Next, let F be an arbitrary OX -module and consider F = (X, F) as an
A-module. Then, using 6.8/2, the adjunction formula yields

HomOX f (G), F = HomOY (G, f F) = HomB (G, F/B ).
Furthermore, given any B-linear map : G - F/B , we can look at the

B-bilinear map
GA - F, (g, a) - (g) a,
where we use the structure of F as A-module. Passing to the associated tensor
product yields an A-linear map
G B A - F, ga - (g) a,
and it is easily seen that this assignment denes a bijection
HomB (G, F/B ) - HomA (G B A, F ).
Using 6.8/2 again, we derive a bijection

HomOX f (G), F - HomO (G
B A, F),
X
showing f (G) G
B A with the help of Lemma 1.
Since any morphism of schemes f : X - Y is continuous with respect to

the Zariski topology, there exist ane open coverings (Ui )iI of X and (Vi )iI of
Y such that f (Ui ) Vi for all i I. Thereby we can conclude from Corollary 4:
Corollary 5. Let f : X - Y be a morphism of schemes and G a quasi-

coherent OY -module. Then f (G) is a quasi-coherent OX -module.
Corollary 6. Let f : X - Y be a morphism of schemes and G an invertible

(rep. locally free) OY -module. Then f (G) is an invertible (resp. locally free)
OX -module; note that invertible and locally free module sheaves were introduced
within the context of 6.8/9.
Proof. Use Corollary 4 in conjunction with the fact 4.1/9 that free generating
systems of modules are preserved under tensor products.
As can easily be read from Corollary 4, the direct image of an invertible or

locally free OX -module F with respect to a morphism of schemes f : X - Y
is not necessarily locally free any more. However, the weaker condition of quasi-
coherence is preserved if f satises certain niteness conditions.
Denition 7. A morphism of schemes f : X - Y is called quasi-compact

1
if the preimage f (V ) of every quasi-compact open subset V Y is quasi-
compact in X.
Furthermore, f : X - Y is called quasi-separated if the diagonal embed-
ding X/Y : X - X Y X, to be considered in Section 7.4, is quasi-compact.
We have chosen to dene quasi-separatedness via the diagonal morphism

X/Y , since this is the most natural way, although it involves the construction
of ber products to be explained only later in Section 7.2. However, there is
another characterization of quasi-separated morphisms in Remark 8 (iii) below
that is free from ber products. For the purposes of the present section, namely
the proof of the direct image theorem in Proposition 9, it is enough to take this
characterization as a denition, thereby avoiding the use of ber products at
this place.
Remark 8. Let f : X - Y be a morphism of schemes.

(i) An open subset of X (resp. Y ) is quasi-compact if and only if it is a
nite union of ane open subsets in X (resp. Y ).
(ii) f is quasi-compact if and only if there is an ane open covering (Yi )iI
of Y such that the preimages f 1 (Yi ) are quasi-compact.
(iii) f is quasi-separated if and only if there exists an ane open covering
(Yi )iI of Y with the property that, for any ane open subsets U, U X
satisfying f (U ) f (U ) Yi for some i I, the intersection U U is quasi-
compact.
Proof. Assertion (i) is true, since ane open subsets in X are quasi-compact by
6.1/10 and since these sets form a basis of the topology of X.
In the situation of (ii) we start by considering the special case where
f: X - Y is a morphism of ane schemes; let f # : OY (Y ) - OX (X) be
the corresponding ring morphism. Now, if V Y is open and quasi-compact,
it is a nite union of basic open subsets of type D(g) Y for suitable elements
g OY (Y ). Since f 1 (V ) is the union of the preimages f 1 (D(g)) = D(f # (g)),
it is a nite union of ane open subsets in X and, hence, quasi-compact. In
particular, f is quasi-compact.
In a next step we assume that X is quasi-compact and Y is ane. Also in
this case f is quasi-compact, since there is a nite ane open covering (Xi )iI
of X and since every restriction fi : Xi - Y of f is quasi-compact, as we
just have seen. Therefore, if V Y is quasi-compact, the same holds for all
preimages fi1 (V ) and, hence, also for f 1 (V ) = iI fi1 (V ), since it is a nite
union of quasi-compact sets.
In the general case, we look at a quasi-compact open subset V Y and
assume that there is an ane open covering (Yi )iI of Y such that f 1 (Yi ) is
quasi-compact for all i I. Then all restrictions fi : f 1 (Yi ) - Yi of f are
quasi-compact, as we just have seen. Furthermore, there exists a nite ane
open covering (Vj )jJ of V such that for each index j J there is an index i I
satisfying Vj Yi . Then the preimages f 1 (Vj ) are quasi-compact, and f 1 (V ),
as a nite union of quasi-compact sets, is quasi-compact as well. Therefore f
is quasi-compact also in this case. This settles the if part of assertion (ii), the
reverse being trivial.
To verify (iii) we have to make use of ber products as we will construct
them in Section 7.2. Starting with the if part, let us use 7.2/5 and consider the
open covering (f 1 (Yi ) Yi f 1 (Yi ))iI of X Y X. Since we want to apply the
criterion (ii), we may replace Y by any of the Yi and thereby assume that Y
is ane. Then consider the ane open covering (Xi )iI of X consisting of all
ane open subsets in X. It follows from 7.2/4 that the ber products Xi Y Xj
are ane for all i, j I and, furthermore, from 7.2/5 that these form an open
covering of X Y X satisfying 1 X/Y (Xi Y Xj ) = Xi Xj . Since condition (iii)
implies that all these preimages are quasi-compact, X/Y is quasi-compact by
criterion (ii). To settle the only-if part, choose ane open parts U, U X
such that f (U ) f (U ) Yi for some i I. Then U Yi U is an ane open
subscheme in X Y X. Since its preimage with respect to X/Y equals the
intersection U U , the latter must be quasi-compact if X/Y is quasi-compact.
Thus, we are done.
Let us add that any morphism of schemes f : X - Y that is separated in

the sense of 7.4/2 is, in particular, quasi-separated. Namely, separated means
that the diagonal morphism X/Y is a closed immersion in the sense of 7.3/7
and such immersions are ane in the sense that preimages of ane open subsets
are ane again.
Proposition 9. Let f : X - Y be a morphism of schemes and F a quasi-

coherent OX -module where f is quasi-compact and quasi-separated. Then the
direct image f (F) is a quasi-coherent OY -module.
Proof. The problem is local on Y . Therefore we may assume that Y is ane,

say Y = Spec B. If X is ane as well, say X = Spec A, then F is associated
to some A-module F and we have f (F) F " /B according to Corollary 4. In
particular, f (F) is quasi-coherent in this case.
If X is not necessarily ane, we can choose a nite ane open covering
(Xi )iI of X, as f is assumed to be quasi-compact and Y was supposed to
be ane. Furthermore, all intersections Xi Xj are quasi-compact, since f is
quasi-separated; use Remark 8 (iii).
Let us assume for a moment that the intersections Xi Xj are even ane.
We write Fi for the direct image of F|Xi with respect to the composition
f
Xi - X - Y, i I,
as well as Fij for the direct image of F|Xi Xj with respect to the composition
f
X i Xj - X - Y, i, j I.
Then the OY -modules Fi and Fij are quasi-coherent by the special case dealt
with above, and we can look at the exact diagram
-

f (F) - Fi - Fij
iI i,jI
that is given for U = f 1 (V ) with V Y open by the canonical exact diagrams

F(U ) - F(U Xi ) - - F(U Xi Xj )
iI i,jI
related to the sheaf condition of F with respect to the open covering (U Xi )iI
of U . In other words, we have

-
f (F) = ker Fi - Fij
iI i,jI
where the cartesian products can be interpreted as direct sums, since they ex-
tend over nite index sets. Now using the fact that the formation of associated
modules commutes with direct sums (6.8/3), as well as with kernels of module
homomorphisms (6.8/4), we see that f (F) is quasi-coherent.
If the intersections Xi Xj are only known to be quasi-compact, choose for
each pair of indices i, j I a nite ane open covering (Xijk )kJij of Xi Xj

and write Fijk for the direct image of F|Xijk with respect to the composition
f
Xijk - X - Y, i, j I, k Jij .

As before, Fijk is a quasi-coherent OY -module, and the canonical exact diagrams

F(U ) - F(U Xi ) - - F(U Xijk )
iI i,jI,kJij
for U = f 1 (V ) with V Y open give rise to an exact diagram

-

f (F) - Fi -
Fijk .
iI i,jI,kJij
The same reasoning as above shows that f (F) is a quasi-coherent OY -module.

Exercises
1. Let f : X - Y be a morphism of ringed spaces. Show for any sheaf G of abelian
groups on Y and any point x X that there is a canonical isomorphism of stalks
(f 1 G)x Gf (x) . Conclude that the etale space of f 1 G on X may be viewed as
the ber product of the etale space of G over Y with X.
2. Let f : X - Y and g : Y - Z be morphisms of ringed spaces. Show for
sheaves of modules F on X and G on Z that there are canonical isomorphisms
of OZ -modules, resp. OX -modules
(g f ) F g (f F), (g f ) G f (g G).
3. Let f : X - Y be a morphism of ringed spaces. Consider the functors f on

sheaves of abelian groups or modules on X, as well as f 1 on sheaves of abelian
groups on Y , and f on sheaves of modules on Y . Show:
(a) f is left exact.
(b) f 1 is exact.
(c) f is right exact.
4. For a sheaf of rings OX on a topological space X, an OX -ideal is meant as an

OX -submodule sheaf I OX ; see also 7.3/1. Let f : X - Y be a morphism
of ringed spaces. Show for any OY -ideal I OY that the canonical morphism
f 1 I - f 1 OY denes f 1 I as an ideal sheaf in f 1 OY . Moreover, can we
conclude that f I is always an ideal sheaf in f OY = OX ?
5. Consider a topological space X and a subset Z X equipped with the subspace
topology induced from X. Thus, a set V Z is (relatively) open in Z if and
only if it is of type U Z for some open subset U X. Writing i : Z - X for
the inclusion map, show that there is a canonical isomorphism (i F)i(z) Fz for
each z Z and that (i F)x = 0 for points x X not contained in the closure Z
of Z in X. Give an example where Z is dierent from Z and where Fx = 0 for a
point x Z Z.
6. Let X be a topological space with a closed point x. Consider an abelian group F
and view it as a sheaf F of abelian groups on the one-point space {x}. Show for
the inclusion map i : {x} - X that i F equals the skyscraper sheaf on X with
stalk F at x; for the latter see Exercise 6.5/6. Conclude from Proposition 9 for a
scheme X, a closed point x X, and a k(x)-vector space F that the skyscraper
sheaf F with stalk Fx = F is quasi-coherent. See also Exercise 6.8/3.
7. Show that the formation of associated modules on ane schemes respects tensor
products. Conclude that, on any scheme X, the tensor product F OX G of quasi-
coherent OX -modules F, G yields a quasi-coherent OX -module again.
8. Give an example of a morphism of schemes that fails to be quasi-compact, resp.
quasi-separated (in the sense of Remark 8 (iii)).
9. Consider a topological space X and an open subset U X. Let j : U - X be
the inclusion map. Show for any sheaf of abelian groups F on X that the inverse
image sheaf j 1 F coincides with the restriction F|U of F to U .
10. Extending sheaves by zero: As in Exercise 9, let j : U - X be the inclusion of an
open set into a topological space. For any sheaf F of abelian groups on U , dene
its extension by zero j! F (pronounced j lower shriek) as the sheaf associated to
the presheaf given on open subsets U X by

U - F(U ) if U U,
0 otherwise.
Show that j! F is uniquely characterized by the facts that it restricts to F on U

and, furthermore, satises (j! F)x = 0 for all x X U .
11. For a topological space X and a closed subset Z X consider the inclusion maps
i : Z - X and j : U - X where U = X Z. Show for any sheaf F of abelian
groups on X that there is a canonical exact sequence of sheaves of abelian groups
0 - j! (j 1 F) - F - i (i1 F) - 0.
Note that j 1 F = F|U and that, likewise, i1 F may be considered as the re-
striction of F to Z. Hint: Use Exercises 5, 9, and 10 above.
7. Techniques of Global Schemes
Schemes have been dened as ringed spaces with certain additional proper-
ties. Therefore it is reasonable to expect that standard gluing techniques from
the theory of manifolds can be used for the construction of new schemes from
previously established ones. To glue a family of schemes (Xi )iI along certain
overlaps Xi Xj , which we assume to be open in Xi and Xj , we need to
specify these overlaps as open subschemes Xij Xi and Xji Xj for all
i, j I, together with gluing isomorphisms ij : Xij - Xji , which are used
as identications. Of course, the data must be symmetric in the sense that the
ij satisfy ij ji = id (as well as Xii = Xi and ii = id for all i I). However,
the latter is not enough. Given three indices i, j, k I, there are identications
Xij Xi Xik
ij ik
-

jk
Xji Xj Xjk - Xkj Xk Xki
and these must be compatible in the sense that ik = jk ij on the part of
Xi where both sides are dened. The latter is the so-called cocycle condition
for triple overlaps. Indeed, as we will see in 7.1/1, if the cocycle condition is
satised, the Xi can be glued by using the isomorphisms ij as identications.
Furthermore, writing X = iI Xi for the resulting scheme, one can use the
sheaf property of the structure sheaf OX to dene morphisms X - Y by
gluing morphisms Xi - Y that coincide on all overlaps of the Xi ; see 7.1/2.
If the index set I consists of precisely two elements, say I = {1, 2}, the
cocycle condition is automatically fullled and therefore can be neglected. For
example, consider X1 = X2 = Spec K, the ane line A1K over a eld K,
as well as X12 = X21 = Spec K, 1 , the open subscheme of the ane line
obtained by removing its origin, the latter being given by the maximal ideal
() K. Then to construct a scheme X by gluing X1 and X2 along X12 we
can use the identity morphism 12 : X12 - X21 as the gluing isomorphism.
The resulting scheme is denoted by A1K ; it is the ane line with a double
origin. Alternatively, we can use the gluing isomorphism 12 : X12 - X21
given by the isomorphism of K-algebras
K, 1 - K, 1 , - 1 .

278 7. Techniques of Global Schemes
The resulting scheme is the projective line P1K ; see the construction of projective
n-spaces in 7.1 or the more general approach to Proj schemes in 9.1. Also note
that the schemes A1K and P1K are not isomorphic. This is easily seen by looking
at the separatedness condition to be discussed below. The projective line P1K is
separated over K whereas the ane line with double origin A1K is not.
The just described gluing technique is used for several basic constructions in
the setting of schemes. For example, we show in 7.1 how to dene the projective
n-space PnR over some base ring R by gluing n+1 copies of the ane n-space AnR .
Furthermore, we construct the spectrum Spec A of a quasi-coherent OS -algebra
A over some base scheme S; the latter generalizes the construction of the scheme
Spec A associated to a ring A.
Another rather interesting application of the gluing technique is the con-
struction of ber products in the category of schemes, which we deal with in
7.2. As the category of rings admits amalgamated sums in the form of tensor
products (see 4.3/6 and 4.5/3), it follows that its opposite, the category of ane
schemes, admits ber products in the sense of 4.5/2, namely
Spec A1 Spec R Spec A2 = Spec(A1 R A2 )
for algebras A1 , A2 over some ring R. Then, using the characterization 7.1/3 of
morphisms into ane schemes, we can show that ber products in the setting of
ane schemes satisfy the universal property required for ber products in the
larger category of all schemes; see 7.2/4. Thus, it follows that ber products in
the category of ane schemes are ber products in the category of all schemes
as well. It is then a technical but straightforward process to construct ber
products X S Y for arbitrary schemes X, Y over some base scheme S; see
7.2/3. Assuming S ane in a rst step we x ane open coverings (Xi )iI and
(Yj )jJ of X and Y and construct the ber product X S Y by gluing the ber
products Xi S Yj along the overlaps (Xi Xi )S (Yj Yj ) where i, i I and
j, j J; that the latter ber products exist and, in fact, are open subschemes
of Xi S Yj and Xi S Yj follows from the observation made in 7.2/5. Finally,
for general S, one works with respect to an ane open covering (Si )iI of S.
If Xi and Yi are the preimages of Si with respect to the structural morphisms
X - S and Y - S, the ber products Xi S Yi exist by the previous
i
step and X S Y is obtained by gluing the ber products Xi Si Yi along the
overlaps (Xi Xi ) Si Si (Yi Yi ) for i, i I.
It is important to realize that a ber product of schemes X S Y will in
general not serve as a ber product in the category of topological spaces or even
sets. Writing |X| for the point set underlying a scheme X, there is a canonical
map of sets
|X S Y | - |X| |S| |Y |
that is surjective, but not necessarily injective; see 7.2/6. For example, we show
that the ber product Spec C Spec R Spec C consists of two points, whereas
| Spec C| | Spec R| | Spec C| is a one-point set.
Nevertheless, ber products can successfully be used to dene a workable
replacement of the Hausdor separation axiom in the setting of schemes. Recall
that a topological space X is called Hausdor if any two dierent points of

it admit disjoint open neighborhoods. Equivalently, one can require that the
diagonal embedding : X - X X into the cartesian product of X with
itself has closed image when X X is equipped with the product topology.
Passing to a scheme X, say over some base scheme S, the diagonal embedding
X/S : X - X S X of X into the (scheme theoretic) ber product X S X
exists as a morphism in the category of schemes; it is the one that composed
-
with both projections X S X - X yields the identity morphism on X.
Although the topological space underlying X is just a Kolmogorov space (see
6.1/8), it is a surprising fact that the image X/S (X) is closed in X S X in
many signicant cases, for example, if X and S are ane; see 7.4/1. We say
that X is a separated S-scheme in this case. As it turns out, the separatedness
condition is a good adaptation of the Hausdor separation axiom to the setting
of schemes.
Actually, an S-scheme X is called separated if the diagonal embedding
X/S : X - X S X is a closed immersion; see 7.4/2. We will study closed
immersions and locally closed immersions in 7.3. A morphism of ane schemes
f : Spec B - Spec A is called a closed immersion if the corresponding ring
morphism f # : A - B is surjective. Then, writing a = ker f # , there is a
canonical isomorphism A/a - B and we see that Spec B is a scheme living
on the Zariski closed subset V (a) Spec A; it is called a closed subscheme of
Spec A. More generally, a morphism of schemes f : Y - X is called a closed
immersion if for every ane open subscheme U X its preimage f 1 (U ) is an
ane open subscheme in Y such that the resulting morphism f 1 (U ) - U is
a closed immersion in the sense just mentioned. Actually, it is enough to require
this property for U varying over the members of an ane open covering of X;
see 7.3/9. Also note that a morphism of schemes f : Y - X is called a locally
closed immersion if there exists an open subscheme X X such that f factors
through a closed immersion f : Y - X .
It is easy to see that the diagonal embedding X/S : X - X S X associ-
ated to an S-scheme X is always a locally closed immersion; cf. 7.4/1. Moreover,
it is a closed immersion as soon as the image X/S (X) is closed in the ber
product X S X. Therefore the two versions of separatedness as mentioned
above are equivalent; see 7.4/3. As an example, let us show that A1K , the ane
line with double origin over a eld K, is not separated. As above, consider the
ane open covering A1K = X1 X2 by two ane lines Xi = A1K = Spec Ki ,
i = 1, 2. Using the observation 7.2/5 we obtain
A1K K A1K = (X1 K X1 ) (X1 K X2 ) (X2 K X1 ) (X2 K X2 )
as an ane open covering of the ber product of A1K with itself. Furthermore,
we see that the preimages of X1 K X2 and X2 K X1 with respect to the
diagonal embedding : A1K - A1 K A1 coincide both with X1 X2 . Of
K K
course, the latter intersection is the ane line X1 = Spec K1 with its origin
removed, i.e. X1 X2 = Spec K1 , 11 . In particular, we get the following
commutative diagrams
X1 K1
-
6
?
#
X1 X 2 - X1 K X2 K1 , 11
K1 , 2
6
- ?
X2 K2
on the level of schemes, as well as on the level of associated K-algebras; is the

restriction of the diagonal embedding to the preimage of X1 K X2 . We claim
that is not a closed immersion and, hence, that A1K cannot be separated. For
this it is enough to show that # is not surjective, which is easy to see. Indeed,
by our notation the map K1 1 , 11 is meant to be the canonical
- K
one whereas, due to the gluing we have done for constructing A1K , the map
K2 - K1 , 11 is given by 2 - 1 . Thus, 11 does not belong to the
image of # and therefore cannot be a closed immersion. Consequently, A1K
is not separated over K.
In principle, the situation is similar for the projective line P1K . The only
dierence is that the map K2 - K 1 , 11 is given by 2 - 1 .
1
#
This implies that now is surjective, indeed. Since the diagonal embeddings
Xi - Xi K Xi , i = 1, 2, are closed immersions for trivial reasons, it follows
that P1K is a separated K-scheme. We will show more generally in 9.1/18 that
Proj schemes and, in particular, projective n-spaces are separated.
There is another basic technique applicable to schemes, which we have
included in the present chapter, namely cohomology theory. It comes in two
avours. First there is Cech cohomology, which we deal with in 7.6, a theory
that is quite useful for explicit computations. On the other hand we will ex-
plain Grothendieck cohomology in 7.7, which is a derived functor cohomology
using methods similar to the one explained in 5.4 for the construction of Ext
modules. In contrast to Cech cohomology, Grothendieck cohomology admits
nice general properties, like the existence of long cohomology sequences 7.7/4.
In addition, there are quite general situations where both cohomology theories
are compatible so that the advantages of both can be combined; see 7.7/5 and
7.7/6.
Given a topological space X and a sheaf F (of abelian groups) on it, one
considers for any open covering U of X the complex
d0 d1 d2
C (U, F) : 0 - C 0 (U, F) - C 1 (U, F) - C 2 (U, F) - ...
of so-called Cech cochains on U with values in F; see 7.6 for its denition. Then
one is interested in the attached Cech cohomology groups
H q (U, F) = ker dq / im dq1 , q = 0, 1, . . .
For example, if X is a scheme and F its structure sheaf OX , these groups

represent certain invariants of X, at least if one can eliminate the dependence
of the cohomology groups from the covering U. On separated schemes the latter
is possible, indeed, by taking U ane, as follows from Lerays Theorem 7.7/5
in conjunction with 7.4/6 and 7.7/7.
As an example, let us consider the ane 2-space A2K = Spec K1 , 2 over a
eld K and remove the origin from it, namely the closed point 0 given by the
maximal ideal (1 , 2 ) K1 , 2 . The resulting scheme X admits a canonical
ane open covering U, namely
X = Spec K1 , 2 , 21 Spec K1 , 11 , 2 .
To compute the Cech cohomology groups H q (U, F) we use the Cech complex
of so-called alternating cochains; see 7.6/1. The latter is particularly simple in
our case, as it is given by
d0
0 - K1 , 2 , 21 K1 , 11 , 2 - K1 , 11 , 2 , 21 - 0
with d (f1 , f2 ) = f2 f1 . Then
0
H 0 (U, OX ) = ker d0 = OX (X)

due to the sheaf property of OX and, furthermore,

H 1 (U, OX ) = K1 , 11 , 2 , 21 / im d0 = K 1i 2j ,
i,j<0
H (U, OX ) = 0
q
for q > 1.
In particular, H (U, OX ) is a K-vector space of innite dimension and this
1
shows that X cannot be ane. Indeed, for ane schemes and ane open cov-
erings, all higher cohomology groups H q (U, OX ), q > 0, are trivial; use 7.7/7
in conjunction with Lerays Theorem 7.7/5. Of course, that X is not ane can
also be seen from the fact that the inclusion X - A2K , which is a proper
inclusion, nevertheless gives rise to an isomorphism on the level of associated
K-algebras K1 , 2 - OX (X). Finally, let us refer to Serres criterion
7.7/8, which states that the vanishing of rst cohomology groups with values
in quasi-coherent modules characterizes ane schemes.
Also we have included in the present chapter a brief discussion of the Noethe-
rian niteness condition for schemes (see 7.5), after having settled the necessary
prerequisites on associated ideals in 7.3. A scheme X is called locally Noetherian
(see 7.5/3) if every point in X admits an ane open neighborhood Spec A X
where A is Noetherian. The condition has geometric implications for the Zariski
topology as well as consequences on the level of rings of functions. In particular,
if Spec A is any ane open subscheme of a locally Noetherian scheme, then A
will be Noetherian by 7.5/4. For some time it was quite popular to require the
Noetherian condition for any base scheme S whenever relative schemes X/S
were considered. However, we will not proceed like this and follow the strategy
of Grothendieck instead, who preferred to work over a general base S, imposing
any necessary conditions on the structural morphism X - S. Usually this
requires a little bit of extra eort, but has the advantage that the theory re-
mains compatible with general base change. This way impacts originating from
the base and from the morphism can be kept well apart from each other.
7.1 Construction of Schemes by Gluing
According to its denition, any scheme X admits a covering by ane open

subschemes Xi , i I. Therefore we may view X as being obtained by gluing the
ane schemes Xi along their intersections Xi Xj . In the following we want to
study more thoroughly how global schemes can be constructed by gluing known
local parts such as ane schemes along open overlaps. The case where only two
schemes X1 and X2 are involved is easy. We just need a scheme U serving as the
intersection of X1 and X2 . This means, we must be able to view U as an open
subscheme of both X1 and X2 in order to identify X1 with X2 along U . For an
arbitrary number of schemes Xi the situation becomes more complicated, since
triple overlaps and the related cocycle condition have to be taken into account.
Proposition 1. Consider
(i) a family (Xi )iI of schemes,
(ii) open subschemes Xij Xi and isomorphisms ij : Xij - Xji for all
i, j I,
subject to the following conditions:
(a) ij ji = id, Xii = Xi and ii = id for all i, j I.
(b) The isomorphisms ij : Xij - Xji restrict to isomorphisms
(k)
ij : Xij Xik - Xji Xjk
(j) (i) (k)
such that ik = jk ij for all i, j, k I (cocycle condition).
Then the schemes Xi can be glued along the intersections Xij Xji to
set up a scheme X.
More precisely, there is a scheme X together with morphisms i : Xi - X,
i I, such that:
(1) i denes for all i I an isomorphism of Xi onto an open subscheme
Xi X.
(2) X = iI Xi .
(3) i (Xij ) = Xi Xj for all i, j I.
(4) The diagram
Xij
i |Xij
-
ij Xi Xj
-
j |Xji
?
Xji
is commutative for all i, j I.
Moreover, X is unique up to canonical isomorphism.
Proof. The given gluing data can be illustrated as follows:

7.1 Construction of Schemes by Gluing 283
Xi
Xij Xik
Xij Xik
ij ik
(k) (j)
ij ik
Xji Xki
Xji Xjk Xkj Xki
(i)
jk
Xj Xk
Xjk Xkj
jk
The basic idea is to identify the open subschemes Xij Xi and Xji Xj via
the isomorphisms ij : Xij - Xji for all i, j I, thereby obtaining a scheme
X, which is covered by the open subschemes Xi . For this to work properly we
need compatibility of the identication process on triple overlaps, a requirement
expressed !by the cocycle condition. Let us start by dening X as a set; namely,
let X = ( iI Xi )/ , where the relation is given as follows: For x Xi
and y Xj write x y if x Xij and ij (x) = y. Then is an equiva-
lence relation. Indeed, reexivity and symmetry of the relation follow from
condition (a). Furthermore, the transitivity is a consequence of the cocycle con-
dition (b). Indeed, let x Xi , y Xj , and z Xk such that x y and y z.
Then we have
x Xij , y = ij (x) Xji , y Xjk , z = jk (y) Xkj .
In particular, we get y Xji Xjk and, hence, x Xij Xik as well as

z Xkj Xki according to (b). Therefore
(j) (i) (k) (i)
ik (x) = jk ij (x) = jk (y) = z
by the cocycle condition and we see x z. Consequently, is an equivalence

relation, and X is well-dened as a set.
Next consider the canonical maps i : Xi - X, i I, that map any
element of Xi to its associated equivalence class in X. Since ii = id, all i are
injective. Hence, i denes a bijection of Xi onto the subset Xi = i (Xi ) X,
and it is clear that the stated properties (2), (3), and (4) of the assertion hold in
terms of sets. Now let us introduce a topology on X by calling a subset U X
open if i1 (U ) is open in Xi for all i I. This is the nest topology on X such
that all maps i are continuous. Moreover, if there is an index i0 I such that
U Xi0 , then U is open in X if and only if i1 0
(U ) Xi0 is open. Indeed, if
the latter holds, we get for every i I
1
i1 (U ) = i1 (U Xi ) = 1
i,i0 i0 (U )
from (4) and it follows that i1 (U ) is open in Xi , since i1

0
(U ) Xi0 is open
and the map
i,i0
Xi,i0 - Xi ,i - Xi
0 0
is continuous. In particular, all subsets Xi X are open in X and the maps i
induce homeomorphisms Xi - Xi .
It remains to construct the structure sheaf of X. To do this, we trans-
port for every i I the structure sheaf OXi of Xi under the homeomorphism
i : Xi - X to a structure sheaf OX = (i ) (OX ) on X . Then the isomor-
i i i i
phisms ij , now as isomorphisms of locally ringed spaces, induce isomorphisms
- OX |X X
ij : OXj |Xi Xj
# i i j
of sheaves of rings, which we may view as identications by applying the cocycle

condition (b). Indeed, looking at subsets U X and indices i, j I such that
U Xi Xj , we may identify OXj (U ) with OXi (U ) via the isomorphism #
ij (U ).
Proceeding similarly as for the construction of X itself, we would have to set
up equivalence relations from the # ij on the disjoint union of all OXi (U ) for

indices i I such that U Xi and then pass to equivalence classes. Doing so
we obtain a functor OX with values in the category of rings, which is dened on
all open subsets U X such that U Xi for some index i I. Since these sets
form a basis B of the topology on X and since OX satises sheaf conditions
for coverings of sets from B by sets from B, we can extend OX via 6.6/4 to a
sheaf on X, just by setting
OX (U ) = lim OX (U )

U B, U U
for arbitrary open subsets U X. Then it is easily seen that (X, OX ) is a scheme
as stated in the assertion. The latter is unique up to canonical isomorphism,
as can conveniently be checked by using the exact diagram of Proposition 2
below.
Next, let us discuss how to dene scheme morphisms by gluing.
Proposition 2. Let X be a scheme together with a covering by open subschemes

(Xi )iI , and let Y be another scheme. If fi : Xi - Y are morphisms such that
fi |Xi Xj = fj |Xi Xj for all i, j I, there exists a unique morphism f : X - Y
such that f |Xi = fi for all i I. In other words, the canonical diagram

Hom(X, Y ) - Hom(Xi , Y ) - - Hom(Xi Xj , Y )
iI i,jI
is exact.
Proof. In terms of maps between sets, the fi determine to a unique map

f: X - Y such that f |X = fi for all i I. Furthermore, f is continu-
i
ous, since all fi are continuous. Indeed, for V Y open, we have

f 1 (V ) = f 1 (V ) Xi = fi1 (V ),
iI iI
and we see that f 1 (V ) is open in X, since, due to the continuity of the maps
fi , all sets fi1 (V ) are open in Xi and, in particular, in X.
To equip f with the structure of a morphism of schemes, consider for any
open subset V Y the following diagram:

OX f 1 (V )
-
?

OY (V ) - OX f 1 (V ) Xi
iI
- ??

OX f 1 (V ) Xi Xj
i,jI
The vertical maps are the canonical ones; they constitute an exact diagram, due
to the sheaf property of OX . Furthermore, the horizontal map is derived from the
maps fi# : OY - fi OX , i I, and its composition with the two vertical maps
i
yields a unique map OY (V ) - i,jI OX (f 1 (V ) Xi Xj ), due to the com-
patibility fi |Xi Xj = fj |Xi Xj for i, j I. Then, using the exactness of the ver-
tical part of the diagram, we get a unique morphism OY (V ) - OX (f 1 (V ))
making the upper triangle commutative. Letting V vary over the open sets of
Y , the resulting morphisms OY (V ) - OX (f 1 (V )) can be used to dene
a morphism of schemes f : X - Y satisfying f |X = fi for all i I, the
i
uniqueness of f being clear from the construction.
As a rst consequence, let us derive from Proposition 2 the following gen-

eralization of 6.6/9 (ii):
Corollary 3. Let A be a ring and Y a scheme. Then the canonical map

: Hom(Y, Spec A) - Hom A, OY (Y ) , (f, f # ) - f # (Spec A),
from the set of scheme morphisms Y - Spec A to the set of ring homomor-
phisms A - OY (Y ) is bijective. In particular, gives rise to isomorphisms
of functors on schemes Y , or on rings A.
Proof. Fix an ane open covering (Yi )iI of Y and set
B = OY (Y ), Bi = OY (Yi ), Bij = OY (Yij )
for i, j I where Yij = Yi Yj . Then consider the following canonical commu-

tative diagram:
-

Hom(Y, Spec A) - Hom(Yi , Spec A) - Hom(Yij , Spec A)
iI i,jI
? ?
? -

Hom(A, B) - Hom(A, Bi ) - Hom(A, Bij )
iI i,jI
The upper row is the exact one occurring in Proposition 2, whereas the vertical
maps are the canonical ones, as mentioned in the assertion. Furthermore, the
lower row is obtained from the exact row
-

B - Bi - Bij ,
iI i,jI
given by the sheaf property of OY , via the application of Hom(A, ). Since the
latter functor is left exact, the lower row of the above diagram is exact as well.
Now the middle vertical map is bijective, due to 6.6/9 (ii) and the same is
true for the right vertical one if all intersections Yij are ane. In that case
we see by diagram chase that must be bijective as well. For example, this
argument establishes the assertion of the corollary if Y might be viewed as an
open subscheme of some ane scheme Y . In particular, we can assume then
that each Yi is basic open in Y and the same will hold for the intersections Yij .
Passing to the general case, we know from the special case just dealt with
that the right vertical map of the diagram is bijective. But then, as before, the
same is true for the left vertical map and we are done.
For any scheme Y there is a unique ring homomorphism Z - OY (Y ) and,

thus, by Corollary 3, a unique morphism of schemes Y - Spec Z. Therefore
every scheme can canonically be equipped with the structure of a Z-scheme.
As an example of how to apply the gluing techniques discussed above, we
want to construct the ane n-space AnS as a relative S-scheme over an arbitrary
base scheme S. We will achieve this by choosing an ane open covering (Si )iI
of S and by gluing the ane n-spaces AnSi over the base S. To do this in an
eective way, we x a system of variables t = (t1 , . . . , tn ) and introduce the sheaf
OS t of polynomials in t over OS . Then we can interpret the ane n-space AnS
as the spectrum of the sheaf of rings OS t. Explaining this in more detail,
we need some preparations.
Quasi-coherent algebras. Just as in the setting of ordinary rings, a sheaf

of rings A on S together with a morphism OS - A is called an OS -algebra.
The latter amounts to the fact that A(U ), for every open subset U S, is
equipped with the structure of an OS (U )-algebra, in a way that is compatible
with restriction morphisms. If S is ane, say S = Spec R, then, similarly as
done in Section 6.6 for R-modules, we can construct for any R-algebra A its
associated sheaf of OS -algebras A = A, namely the sheaf extending the functor
given on basic open subsets of type D(f ) S for f R by

A D(f ) = A R Rf .
Furthermore, note that, given an OS -algebra B, any morphism of R-algebras
A - B(S) induces by the universal property of tensor products 4.3/6 a
canonical morphism of OS -algebras A - B, whose underlying morphism of
OS -modules coincides with the one considered in 6.8. In particular, we see:
Remark 4. Let A be an OS -algebra on an ane scheme S = Spec R. Then

the canonical morphism A(S) - A is an isomorphism of OS -algebras if and
only if it is an isomorphism of OS -modules.
Based on the construction of associated algebras we can consider quasi-

coherent OS -algebras similarly as in the module case:
Denition 5. An OS -algebra A on a scheme S is called quasi-coherent if there

exists an ane open covering (U ) of S such that A|U , for each , is
associated to some OS (U )-algebra.
The equivalent characterizations of 6.8/10 carry over from the module to

the algebra situation:
Proposition 6. For a scheme S and an OS -algebra A the following conditions

are equivalent:
(i) A is quasi-coherent.
(ii) For every ane open subset U S the restriction A|U is associated to
some OS (U )-algebra.
Proof. The implication from (ii) to (i) is trivial. To verify the reverse, we may
assume U = S, hence, that S is ane, say S = Spec R, and that there is an
ane open covering (U ) of S where A|U , for each , is associated to
some OS (U )-algebra. Then, thinking in terms of OS -modules, we can apply
6.8/10 and thereby see that A is associated to A(S), viewed as an R-module.

Therefore the canonical morphism A(S) - A is an isomorphism in terms of
OS -modules and the same is true by Remark 4 in terms of OS -algebras.
Quasi-coherent polynomial algebras. As an example of a quasi-

coherent OS -algebra on a scheme S we introduce the polynomial algebra
A = OS t in a set of variables t = (t1 , . . . , tn ). To dene this sheaf we pro-
(I)
ceed similarly as for the construction of the free OS -module OS in Section 6.8,
where now I = Nn . Indeed, consider the functor that associates to an open

subset U S the polynomial algebra OS (U )t over OS (U ) and pass to the
associated sheaf of OS -algebras. We denote the latter by OS t and call it the
sheaf of polynomials in t over S. This OS -algebra is characterized by the fact
that the canonical map

HomOS (OS t, B) - B(S)n , - (S)(t1 ), . . . , (S)(tn ) ,
is bijective for any OS -algebra B where HomOS , at this place, indicates the set of
OS -algebra morphisms. Furthermore, note that on every ane open subscheme
U = Spec R S the sheaf OS t is associated to the R-algebra given by the
polynomial ring Rt. In particular, we can state:
Proposition 7. Let S be a scheme and t = (t1 , . . . , tn ) a system of variables.

Then the OS -algebra OS t of polynomials in t over S is quasi-coherent.
The spectrum of a quasi-coherent algebra. Next we want to show

how to obtain from a scheme S and a quasi-coherent OS -algebra A an S-scheme
Spec A, generalizing the construction of ane schemes on spectra of rings. This
is done by gluing ordinary schemes Spec A(U ) for ane open subsets U S.
To explain the process in more detail, choose an ane open covering (Si )iI of
S and consider Xi = Spec A(Si ) for i I as an Si -scheme; let pi : Xi - Si
be the corresponding structural morphism. For any section f OS (Si ) we get
canonical identications

p1
i DSi (f ) = DXi (f ) = Spec A(Si )f = Spec A DSi (f ) ,
using the quasi-coherence of A in conjunction with canonical isomorphisms of

type Af A R Rf for algebras A over a ring R and elements f R; see
the algebra analogue of 4.3/3 (ii). Then we look at the intersections Si Sj for
i, j I and choose ane open coverings

Si Sj = Sij
ij
such that the Sij are basic open in Si and Sj . For example, we could consider
ane open coverings Uij , Uij of Si Sj , where the members of Uij are basic
open in Si and the ones of Uij basic open in Sj . Then we can conclude from
6.6/1 (ii) that the product covering Uij Uij is as desired, namely that its
members are basic open in both, Si and Sj . Thus, by the above interpretation,
the ane schemes p1 1
i (Sij ) and pj (Sij ) are canonically isomorphic for i, j I
and ij , namely to Spec A(Sij ). The same is true over the intersections of
type Sij Sij and we obtain from Proposition 2 canonical isomorphisms
ij : p1 - p1 (Si Sj ),
i (Si Sj ) j i, j I.
Over any ane open subset of the base S these isomorphisms are induced from
identity maps on the level of sections of the OS -algebra A. Using this fact, it
is easy to see that the isomorphisms ij satisfy the conditions of Proposition 1.

Therefore the Xi , i I, can be glued to yield an S-scheme X and we see
from Proposition 2 that the latter is uniquely determined by A, up to canonical
isomorphism over S.
Next consider an S-scheme Y with structural morphism q : Y - S and
an S-morphism f : Y - Spec A. Then the morphism f : OSpec A - f OY
#
given by f induces a family of OS (U )-algebra morphisms

fU : A(U ) - OY q 1 (U ) , U S open,
and, thus, a morphism of OS -algebras f: A - q (OY ). One can show as a

generalization of Corollary 3 that the map
HomS (Y, Spec A) - HomOS (A, q OY ), f - f,
is bijective; see Exercise 4 below. Thereby we obtain a functorial characteri-

zation of Spec A which, departing from the special version of Corollary 3, can
alternatively be used at dierent levels of generality in order to construct Spec A
in a more formal way.
The ane n-space. Applying the above construction to the quasi-

coherent OS -algebra of polynomials in a set of variables t1 , . . . , tn , we are able
to generalize the denition of the ane n-space over a base scheme S, as given
in Section 6.7, by
AnS = Spec OS t1 , . . . , tn .
Later in Section 7.2, when ber products are available, we may alternatively
characterize the ane n-space over S as AnS = AnZ Spec Z S. Also we see that
the canonical injections OS ti - OS t1 , . . . , tn set up a natural bijection
HomS (T, AnS ) - HomS (T, A1 )n ,

S
certainly for S-schemes T that are ane, but also for general S-schemes if we
apply the gluing technique of Proposition 2. This shows that AnS is the n-fold
cartesian product of the ane line A1S in the category of S-schemes.
The projective n-space. We want to discuss another example where

the gluing process of Proposition 1 is applied, namely the construction of the
projective n-space PnS over a scheme S, for n N. In classical analytic geometry,
the projective n-space over a eld K is given by the quotient
Pn (K) = (K n+1 {0})/K .
This means, we consider on K n+1 {0} the equivalence relation given by mul-
tiplication with elements from K and write Pn (K) for the set of associated
equivalence classes. Therefore each element x Pn (K) is represented by a tuple
(x0 , . . . , xn ) K n+1 {0} where, however, the components xi are unique only
up to a common factor K . A good way to express this behavior is to write
x = (x0 : . . . : xn ), where the xi are called the homogeneous coordinates of x.

Quite often one nds the notation x = (x0 , . . . , xn ), although in strict terms it
is not correct.
The projective n-space Pn (K) is covered by the n + 1 subspaces
Ui = {(x0 : . . . : xn ) Pn (K) ; xi = 0}, i = 0, . . . , n,
and each of these can be viewed as an ane n-space under the bijection
Ui - K n, (x0 : . . . : xn ) - (x0 x1 1
i , . . . , 1, . . . xn xi ),
where the symbol 1 at the position with index i means that the entry at this
place has to be discarded. Thus, Pn (K) is covered by n + 1 ane n-spaces K n
and we will base the construction of the projective n-space in terms of schemes
on this fact.
Starting with an ane base scheme S = Spec R, we choose a set of variables
t = (t0 , . . . , tn ) and consider the ring of Laurent polynomials

Rt0 , t1 1
0 , . . . , tn , t n = a t ; a R, a = 0 for almost all Zn+1 ,
Zn+1
which we can view as the localization of the polynomial ring Rt0 , . . . , tn by

the multiplicative system generated by t0 , . . . , tn . Then we introduce for indices
i, j, k = 0, . . . , n the following subrings of Rt0 , t1 1
0 , . . . , t n , tn :
# $
t0 tn
Ai = R , . . . , ,
ti ti
# 1 $
t0 t n tj
Aij = R , . . . , , ,
ti ti ti
# 1 1 $
t0 t n tj tk
Aijk = R , . . . , , , .
ti ti ti ti
Note that we may view Ai as a free polynomial ring over R in the variables
t
t0
ti
, . . . , ti1
ti
, ti+1
ti
, . . . , ttni and, furthermore, Aij as the localization of Ai by tji ,
t
as well as Aijk as the localization of Ai by tji and ttki . Moreover, we have for
i, j, k = 0, . . . , n
Aij = Aji , Aijk = Ajik = Akij .
Then, for i, j, k = 0, . . . , n, look at the schemes Xi = Spec Ai AnR and the
open subschemes
Xij = Spec Aij Xi , Xij Xik = Spec Aijk Xi .
The identication Aij = Aji yields a canonical isomorphism
ij : Xij - Xji ,
which using the identity Aijk = Ajik restricts to an isomorphism

(k)
ij : Xij Xik - Xji Xjk .
Then it follows easily from the above identities between the rings Aij and Aijk
that all prerequisites for the application of Proposition 1 are at hand. In partic-
ular, the Xi can be glued along the intersections Xij to yield a scheme X. All
Xi and Xij are canonically equipped with the structure of an S-scheme, and we
obtain from Proposition 2 a canonical morphism X - S exhibiting X as an
S-scheme. The latter scheme is denoted by PS or PR and is called the projective
n n
n-space over S. Also note that the ring of global sections of the structure sheaf
on X = PnR is given by the intersection
n n # $
t0 tn
OX (X) = Ai = R ,..., = R.
i=0 i=0
ti ti
From this we can conclude that the projective n-space PnR will not be ane in
general since, otherwise, it would be isomorphic to Spec R. For example, if R
is a eld K, then Spec K is a one-point space, whereas PnK for n > 0 contains
more than just one K-valued point, as we will see below.
So let us look at the special case of a eld R = K and let K be a
eld extending K. We want to show that the set of K -valued points of PnK ,
namely PnK (K ) = HomK (Spec K , PnK ), coincides with the ordinary
projective
n-space Pn (K ). To do this, consider a K -valued point of PnK = ni=0 Xi , thus,
a K-morphism x : Spec K - PnK . Then x factors through Xi for some index
i. Therefore x corresponds to a K-homomorphism
# $
t0 tn - K ,
i : Ai = K ,...,
ti ti
and we can view t
t0 n
i : . . . : i
ti ti
as a point of the ordinary projective n-space Pn (K ). If x factors through a
second open part Xj PnK , then x factors through Xi Xj , and i and j
extend to a K-homomorphism
# $
t0 tn t i - K .
ij = ji : Aij = K ,..., ,
ti ti tj
ti
Since tj
is a unit in Aij = Aji , we get
%
t0 t t t t t
n i 0 i n
j : . . . : j = ij ij : . . . : ij ij
tj tj tj ti tj ti

t0 tn
= i : . . . : i .
ti ti
Hence, every point x PnK (K ) gives rise to a well-dened point in Pn (K ),
thereby inducing a map PnK (K ) - Pn (K ). Now, covering Pn by ane
K
n-spaces Xi AnK as done above and using the equation AnK (K ) = (K )n

from Section 6.7, we see immediately that this map is bijective.
We could continue now dening the projective n-space PnS over an arbitrary
base scheme S. A convenient way to do this is within the context of ber
products to be introduced in Section 7.2; just set
PnS = PnZ Spec Z S.
Another possibility is to glue n + 1 ane n-spaces AnS in the way described

above. Also let us refer to Section 9.5 for dening projective schemes of more
general type within the setting of Proj schemes.
Exercises
1. Point functors: For a relative scheme X over some base scheme S consider the
associated functor of points hX : Sch/S - Set, which maps an S-scheme
T to the set hX (T ) = HomS (T, X) of S-morphisms T - X, as well as an
S-morphism T - T to the canonical map Hom S (T, X) - HomS (T , X).
Quite often one writes X(T ) instead of HomS (T, X) and calls this the set of
T -valued points of X. Show:
(a) Let X = Spec OS , for a variable . Then hX is isomorphic to the func-
tor T - OT (T ) on Sch/S. This functor is often referred to as the additive
group and is denoted by Ga . In fact, addition yields a morphism of functors
Ga Ga - Ga dening a functorial group law on Ga ; for the latter term see
the explanations on group schemes in Section 9.6.
(b) Similarly as in (a), let X = Spec OS , 1 . Then hX is isomorphic to the
functor T - OT (T ) on Sch/S. This functor is referred to as the multiplica-
tive group and is denoted by Gm . Multiplication yields a morphism of functors
Gm Gm - Gm dening a functorial group law on Gm .
2. As in Exercise 1, consider point functors hX for schemes X viewing them as
relative schemes in Sch/Z. Let ha
X be the restriction of hX to the full subcategory
of ane schemes in Sch. Show:
(a) Any morphism hX - hX between point functors of schemes X, X is
induced from a scheme morphism X - X . The same is true for isomorphisms.
(b) The assertion of (a) remains true for morphisms ha - ha between point
X X
functors on the category of ane schemes, even if X, X are not necessarily ane.
3. Let R be a discrete valuation ring with eld of fractions K. Consider the multi-
plicative group over R, namely the R-scheme Gm,R = Spec R, 1 . Show that
Gm,R can be viewed as an open subscheme of a larger R-scheme Gm,R such that
the canonical map
HomR (Spec R, Gm,R ) - HomR (Spec K, Gm,R )
becomes bijective. Hint: Glue certain translates of Gm,R along the open sub-
scheme Gm,K = Spec K, 1 Gm,R . Note that Gm,R and the resulting scheme
Gm,R are R-group schemes in the sense of Section 9.6 and that Gm,R is the so-
called Neron model of Gm,K ; see [5], 10.1/5.
4. Let q : Y - S be an S-scheme and A a quasi-coherent OS -algebra. Show that

there is a canonical map HomS (Y, Spec A) - HomO (A, q OY ) and that the
S
latter is bijective. In particular, if q OY is quasi-coherent, conclude that there is
a canonical S-morphism Y - Spec q OY .
5. A morphism of schemes q : Y - S is called ane (see 9.5/1 (i)) if there exists

an ane open covering (Si )iI of S such that f 1 (Si ) is ane for all i I.
Show that q is ane if and only if q OY is a quasi-coherent OS -algebra and the
canonical S-morphism Y - Spec q OY of Exercise 4 is an isomorphism.
6. Morphisms of projective spaces: For an (ane) base scheme S and an integer

m 0, show that there are canonical S-morphisms Pm - Pn that symbolically
S S
(on K-valued points for S consisting of a eld K) are described as
(a) (x0 : . . . : xm ) - (x0 : . . . : xm : 0 : . . . : 0) for m n.
m+d
(b) (x0 : . . . : xm ) - (M0 (x) : . . . : Mn (x)) where M0 , . . . , Mn are the
m
monomials in Zt0 , . . . , tm of a certain degree d > 0 (d-uple embedding).
On the other hand, we can see later that any K-morphism Pm - Pn on a eld
K K
K will be constant if m > n; use 9.3/17 in conjunction with 9.4/4.
7. Give an example showing that, dierent from the ane n-space AnS , the projective
n-space PnS cannot be viewed as an n-fold ber product of P1S with itself. Hint:
For a prime p, set S = Spec Fp and count Fp -valued points in P2S and P1S S P1S ,
assuming that the ber product exists (which actually is the case by 7.2/3).
8. Quotients of schemes by nite groups: Let X be a scheme and a nite subgroup

of the group of automorphisms of X. A quotient of X by is a scheme X together
with a morphism : X - X such that:
(1) = for all .

(2) If : X - Y is a morphism of schemes such that = for all ,
then it admits a unique factorization : X - Y satisfying = .
If the quotient of X by exists, it is unique up to canonical isomorphism and
will be denoted by X/ .
(a) Show that X/ exists if X is ane. Hint: If X = Spec A, set X/ = Spec A
for A the xed ring consisting of all elements in A that are invariant under the
automorphisms given by elements of . Show with the help of 1.3/7 that the
canonical morphism Spec A - Spec A satises the universal property of the
quotient X/ .
(b) In the general case, show that the quotient X/ exists if there is a -invariant
ane open covering of X, i.e. an ane open covering (Xi )iI that satises
(Xi ) = Xi for all and all i I. If the intersection of any two ane
open subschemes in X is ane again (for example, if X is separated; see 7.4/2
and 7.4/6), the condition can be relaxed. In this case X/ exists if each -orbit
in X is contained in an ane open subscheme of X.
(c) In the situation of (b), show that : X - X/ is a quotient in the setting
of sets as well. Hint: Use Exercise 3.3/6.
7.2 Fiber Products
In this section we want to apply the gluing techniques of 7.1/1 in order to

construct ber products of schemes. First, let us recall the notion of cartesian
products for objects in a category. To do this we need to know the cartesian
product of two sets A, B, which is given by the set A B of all pairs (a, b)
such that a A and b B. Then, indeed, A B together with the projections
onto its factors is a cartesian product in the category of sets in the sense of the
denition below.
Denition 1. Let X, Y be objects in a category C. An object W in C together

with two morphisms
p: W - X, q: W - Y,
is called a cartesian product of X and Y in C if for every object T in C (test
object) the map
Hom(T, W ) - Hom(T, X) Hom(T, Y ), f - (p f, q f ),
is bijective; i.e. if for any morphisms
T - X, T - Y
there is always a unique morphism T - W such that the diagram
X
-
6
p
T - W
?
-
Y
is commutative. If W exists, it is unique up to canonical isomorphism and we
write W = X Y . The morphisms p : W - X and q : W - Y are referred
to as the projections onto the factors X and Y of the cartesian product W .
Given an object S in a category C (referred to as a base object) we can

consider the category CS of all relative objects of C over S. As explained in
Section 4.5, the objects in CS are the morphisms of type X - S in C, and
a morphism between two such objects X - S and Y - S is given by a
morphism X - Y in C such that the diagram
X - Y
-
S
7.2 Fiber Products 295
is commutative. Often the category CS is not mentioned explicitly. One just

talks about S-objects and S-morphisms in C, denoting these as usual by X, Y, . . .
and X - Y . As we have pointed out in 4.5/2, the cartesian product of two
objects X, Y in CS is called the ber product of X and Y over S. It is denoted
by W = X S Y if it exists and is characterized by the following commutative
diagram:
X
-
6
p
-
T - W - S
-
q
- ?
Y
Fibred products can also be interpreted within the context of cartesian
diagrams, which are dened as follows:
Denition 2. Let C be a category. A commutative diagram

p
W - X
q
?
?
Y - S
of morphisms in C is called cartesian if the following universal property holds:

Given an object T in C as well as morphisms f : T - X and g : T - Y
in C such that f = g, there exists a unique morphism h : T - W in C
such that p h = f and q h = g:
T
f
h
-
p -
W - X
g
q
? ?
-

Y - S
Viewing the objects X, Y , and W , T in the situation of Denition 2 as

S-schemes, namely under the morphisms , , respectively p = q and
f = g, it becomes clear that such a cartesian diagram shows that W
coincides with the ber product X S Y . Conversely, every ber product of this
type gives rise to a cartesian diagram as above.
Theorem 3. Fibred products exist in the category of schemes.

First, we establish the existence of ber products of ane schemes by tracing

them back to tensor products of algebras. Then we will gradually generalize the
construction, applying the gluing techniques of Section 7.1.
Lemma 4. Fibred products of ane schemes exist in the category of schemes.

More precisely, if A1 and A2 are algebras over a ring R, the ber product of
Spec A1 and Spec A2 over Spec R is given by
Spec A1 Spec R Spec A2 = Spec(A1 R A2 ),
together with the projections
p1 : Spec A1 Spec R Spec A2 - Spec A1 ,

p2 : Spec A1 Spec R Spec A2 - Spec A2
induced from the canonical R-algebra homomorphisms
1 : A 1 - A1 R A2 , 2 : A2 - A1 R A2 .
Proof. Let T be any R-scheme and consider R-morphisms fi : T - Spec Ai ,

i = 1, 2. Using Corollary 3, the latter correspond to morphisms of R-algebras
fi : Ai - OT (T ), i = 1, 2. Furthermore, by 4.3/6 there is a unique morphism
of R-algebras f : A1 R A2 - OT (T ) such that the diagram
A1
f1
1
? -
f
A1 R A2 - OT (T )
6 -
2
f2
A2
is commutative. Then we can apply Corollary 3 again, thereby getting the com-
mutative diagram
Spec A1
-
f1 6
p1
f
T - Spec(A1 R A2 )
f2
p2
- ?
Spec A2
and it follows that Spec(A1 R A2 ) together with the projections p1 , p2 satis-
es the universal property of the ber product in the category of schemes, as
claimed.
Lemma 5. Let 1 : X - S and 2 : Y - S be two S-schemes such that

their ber product W = X S Y with projections
p1 : X S Y - X, p2 : X S Y - Y
exists. If X X and Y Y , as well as S S are open subschemes satisfying

1 (X ) S and 2 (Y ) S , then W = p1 1
1 (X ) p2 (Y ) together with

projections p1 : W - X and p : W
- Y induced from p1 , p2 is a ber

2
product of X and Y over S .
The proof is obvious. Just check the dening universal property for W by
using the one of W = X S Y .
We are now able to carry out the proof of Theorem 3 on the existence of
ber products of schemes. To explain how we will proceed, assume for a moment
that the ber product X S Y is known already:
X
6 1
p1
-

X S Y - S
-
p2
? 2
Y
Choosing ane open coverings (Xi )iI of X and (Yj )jJ of Y , we see that
(p1 1
1 (Xi ) p2 (Yj ))iI,jJ is an open covering of X S Y . The members of this
covering can be viewed in the light of Lemma 5 as ber products of type
Xi S Yj = p1 1
1 (Xi ) p2 (Yj ), i I, j J.
Hence, if S is ane, we conclude from Lemma 4 that (Xi S Yj )iI,jJ is an

ane open covering of X S Y . Furthermore, intersections of any two sets from
this covering can be interpreted as ber products of type
(Xi Xi ) S (Yj Yj ) = p1 1

1 (Xi Xi ) p2 (Yj Yj )
1
= p1 (Xi ) p1 1 1
2 (Yj ) p1 (Xi ) p2 (Yj )
and a similar interpretation is possible on triple overlaps.

To actually prove the existence of the ber product X S Y , let us rst
assume that S is ane. Choosing ane open coverings (Xi )iI and (Yj )jJ of
X and Y as above, the ber products Xi S Yj , i I, j J, exist as ane
schemes according to Lemma 4. Furthermore, applying Lemma 5, we can view
the ber products (Xi Xi ) S (Yj Yj ) for i, i I and j, j J as open
subschemes of both Xi S Yj and Xi S Yj . This yields gluing morphisms as
required in 7.1/1; the cocycle condition follows from the universal property of
ber products by looking at triple overlaps. Hence, the ber products Xi S Yj
can be glued along the intersections (Xi Xi ) S (Yj Yj ) to yield a scheme
X S Y . Due to 7.1/2, the latter is canonically an S-scheme equipped with
projections
p1 : X S Y - X, p2 : X S Y -Y
extending the ones given on Xi S Yj for all i and j. To check that X S Y

satises the universal property of a ber product of X with Y over S, consider
two S-morphisms f1 : T - X and f2 : T - Y , as well as the open covering
1 1
(Tij )iI,jJ of T where Tij = f1 (Xi ) f2 (Yj ). Then the universal property of
the ber products of type Xi S Yj and (Xi Xi ) S (Yj Yj ) yields unique
S-morphisms
Tij - Xi S Y j - X S Y, i I, j J,
that coincide on all intersections of the Tij and, thus, by 7.1/2, give rise to a
unique S-morphism g : T - X S Y such that p g = f for = 1, 2.
Finally, for an arbitrary base scheme S, choose an ane open covering
(Si )iI of S. If the ber product X S Y exists, we see using notations as above
that the ber products
11 (Si ) Si 21 (Si ) = 1 (Si ), i I,
form an open covering of X S Y , with the ber products
11 (Si Si ) Si Si 21 (Si Si ) = 1 (Si Si ), i, i I,
serving as intersections. Keeping this in mind, we can, indeed, get the existence
of the ber product X S Y by gluing the ber products 11 (Si ) Si 21 (Si )
along the intersections 11 (Si Si ) Si Si 21 (Si Si ).
Let us look a bit closer at the points of a ber product X S Y . First

note that the universal property of ber products just says that these products
respect the formation of T -valued points, namely,
(X S Y )(T ) = X(T ) Y (T )
for S-schemes T ; recall that X(T ) = HomS (T, X) is called the set of T -valued
points of an S-scheme X. However, such a nice behavior cannot be expected
from ordinary points. Indeed, for any scheme X let |X| be the set of points
of its underlying topological space. Since any morphism of schemes X -S
induces a map |X| - |S|, it is clear that, given two S-schemes X and Y ,
there is a canonical map
|X S Y | - |X| |S| |Y |.
We want to show that this map will not be injective in general.

Let be a variable. Using the Chinese Remainder Theorem and the prop-
erties of tensor products, there are canonical isomorphisms
R/( 2 + 1) R C C/( 2 + 1) C/( + i) C/( i) C C,

where C C is viewed as a ring under componentwise addition and multiplica-
tion. This ring contains precisely two prime ideals, namely the ideals generated
by (1, 0) and (0, 1). Consequently, the ber product Spec C Spec R Spec C con-
sists of two points. Since Spec R and Spec C are one-point spaces, the canonical
map
| Spec C Spec R Spec C| - | Spec C| | Spec R| | Spec C|
cannot be injective.
Proposition 6. Let X and Y be S-schemes. Then the canonical map

|X S Y | - |X| |S| |Y |
is surjective, but in general not injective.
Proof. To show the surjectivity, consider a point (x, y) |X| |S| |Y |, hence
points x X, y Y lying over a common point s S. Furthermore, let
Spec A X and Spec B Y be ane open neighborhoods of x and y that lie
over an ane open neighborhood Spec R S of s. Then there is the following
commutative diagram
A - A/px - k(x)
6 6 6

-
R - R/ps - k(s) - K = k(x) k(s) k(y) /m
-
? ? ?
B - B/py - k(y)
for a maximal ideal m k(x) k(s) k(y). Passing to spectra, the diagram
Spec k(x) - X
-
? ?
Spec K - Spec k(s) - S
6 6
-
Spec k(y) - Y
yields an S-morphism Spec K - X S Y . Its image z X S Y is a point

lying over x and y and, thus, is mapped to (x, y) |X| |S| |Y |.
Corollary 7. Let f : X - S be a morphism of schemes. For a point s S

look at the canonical morphism Spec k(s) - S and the cartesian diagram
p
Spec k(s) S X - X
? ?
Spec k(s) - S
attached to Spec k(s) S X as a ber product of Spec k(s) with X over S. Then
p induces a homeomorphism
p : Spec X S k(s) - Xs ,
where Xs is the ber f 1 (s) of f over s.
Proof. We know from Proposition 6 that p is surjective. To show that p is, in

fact, a homeomorphism, we may assume by Lemma 5 that X and S are ane,
say X = Spec A and S = Spec R. Now look at the map A - A R k(s) given
by p : Spec k(s) S X - X; it is obtained by tensoring R - k(s) with A
over R. Then every element of A R k(s) is of type a b for some element a
that admits a preimage in A and a unit b A R k(s). Hence, it follows from
6.2/6 that p induces a homeomorphism onto its image, which we have already
recognized as f 1 (s).
Given a relative scheme in the form of an S-scheme X or a morphism of

schemes X - S, we can look at its associated family of bers (Xs )sS , which
are schemes over elds. We thereby get a quite good picture of the relative
scheme X/S, although it must be pointed out that there are some constraints
for families occurring in this way. Indeed, for nice morphisms X - S it
can be observed that the structure of a ber Xs over some point s S will
have inuence on neighboring bers. In addition, a relative scheme X -S
is by no means uniquely characterized by the family of its bers Xs , s S. For
example,
! disregarding connectedness, we can consider the canonical morphism
- S as a relative scheme having same bers as the previous one.
sS Xs
Let us compute the bers of an S-scheme in a simple case.
Let Spec Cx, y/(y 2 2x) be the curve that is given in A2C by the equation
y 2x; the latter is isomorphic to A1C , since Cx, y/(y 2 2x) is isomorphic
2
to Cy. However, this curve can already be dened over Z by considering

X = Spec Zx, y/(y 2 2x) as a Z-scheme. The spectrum of Z consists of the
generic point Spec Z, which corresponds to the zero ideal in Z, and of
the special points p Spec Z that correspond to the principal ideals (p) Z
generated by primes p N. Since k() = Q and k(p) = Fp for primes p, we get:
X = Spec Qx, y/(y 2 2x) A1Q ,
Xp = Spec Fp x, y/(y 2 2x) A1Fp for p = 2,
X2 = Spec F2 x, y/(y 2 ) A1F2 .
Next we want to derive some elementary categorical facts on ber prod-
ucts, which we will formulate for the category of schemes, but which have their
counterparts (except for Remark 11) in any category admitting ber products.
Remark 8. For every S-scheme X the projection X S S - X onto the rst

factor is an isomorphism.
Proof. If T is an S-scheme and f : T - X an S-morphism, then the diagram

X
- 6
f
id
-
f
T - X -S ,
-
id
- ?
S
shows that X satises the universal property of a ber product of X and S over
S.
Remark 9. If f : X - X and g : Y - Y are morphisms of S-schemes,

then f and g induce an S-morphism f g : X S Y - X S Y .
Proof. Use the diagram

f
X -
- X
6 6
f g
X S Y - X S Y
? - ?
g
Y -Y
and the universal property of X S Y .
Remark 10. The ber product is commutative and associative. More precisely,
for S-schemes X, Y , Z there exist canonical isomorphisms
X S Y - Y S X,
(X S Y ) S Z - X S (Y S Z).
Proof. Use the universal property of ber products.
Remark 11. For m, n N and any scheme S there are isomorphisms

AnS AnZ Spec Z S, PnS PnZ Spec Z S,
Am+n
S Am
S S AS .
n
Proof. Use the fact that every ring homomorphism R - R induces canoni-
cally an isomorphism of R -algebras
Rt1 , . . . , tn R R - R t1 , . . . , tn .
In particular, there is a canonical isomorphism
Rt1 , . . . , tm R Rt1 , . . . , tn - Rt1 , . . . , tm+n .
Finally, we want to discuss the concept of base change for relative schemes,
which generalizes the concept of coecient extension known from rings and
modules. The base change functor given by a morphism of (base) schemes
: S - S associates to any S-scheme X an S -scheme XS , namely the
ber product XS = X S S , which is viewed as an S -scheme under the projec-
tion X S S - S . Furthermore, to any morphism of S-schemes f : X - Y
it associates the product morphism
fS = f idS : XS - YS ,
which is an S -morphism. Also note that in the cartesian diagram

X XS
p p
? ?

S S
characterizing XS as the ber product of X and S over S, the projection p is
obtained from p via base change with and the projection from via base
change with p.
Remark 12. Let : S - S be a morphism of schemes. Then, using the

above notations
FS /S : Sch/S - Sch/S
X - XS ,
f - fS ,
is a functor from the category of S-schemes to the category of S -schemes, called

the base change functor attached to the morphism : S - S.
Proposition 13. Let S - S and S - S be morphisms of schemes.

(i) Base change is transitive, i.e. the composition of base change functors
FS /S FS /S is isomorphic to FS /S . In particular, for any S-scheme X there
is a canonical isomorphism
(X S S ) S S - X S S ;
see 4.3/2 for a similar result on the level of rings and modules.
(ii) Base change is compatible with ber products; i.e. for S-schemes X and
Y there is a canonical isomorphism
(X S Y )S - XS S YS
Proof. Starting with assertion (i), write X = X S S and X = X S S . To

interpret X as a ber product of X and S over S, we consider for S-morphisms
u: T - X and v : T - S the following commutative diagram with carte-
sian squares
T
u

X

f
X f
X v

? ? ?

S S S
where the morphisms and exist according to the universal property of ber
products. Indeed, is a unique morphism such that
f = u, = v,
and, likewise, a unique morphism satisfying
f = , = v.
It follows that satises
f f = u, = v
and, furthermore, that is uniquely determined by the latter equations. Hence,

X is a ber product of X and S over S. Applying base change functors to
morphisms we can proceed by a similar argumentation.
The isomorphism in (ii) can easily be derived from (i). We have
(X S S ) S (Y S S ) X S (Y S S ) (X S Y ) S S
due to (i) and the associativity of ber products by Remark 10.
Exercises
1. Let S - S be a morphism of schemes. Show for S -schemes X, Y that there is
a canonical morphism X S Y - X S Y and that the latter is an isomorphism
if S is an open subscheme of S via S - S.

2. Let X, Y be relative schemes over some base scheme S. Show for any Y -scheme
T that the canonical map HomY (T, X S Y ) - HomS (T, X) is bijective.
3. Let X = Spec Kt1 , t2 /(t22 t31 ) be Neiles parabola over a eld K, as in

Exercises 6.2/6 and 6.7/3. Determine the bers of the canonical K-morphism
A1K - X.
4. Consider the ane plane A2K = A1K Spec K A1K together with the projections
p1 , p2 : A1K Spec K A1K - A1 onto its factors. Let be the generic point
K
of AK and write Z = p1 () p1
1 1
2 (). Thus, Z equals the ber of the map
|A1K Spec K A1K | - |A1 | | Spec K| |A1 | over the point (, ). Show that Z is
K K
canonically equipped with the structure of a K-scheme and that the latter is
ane containing an innity of points.
5. Consider the canonical map |A1K Spec K A1K | - |A1K | | Spec K| |A1K | as in Ex-
ercise 4 where now K is assumed to be algebraically closed. Show that the map
is bijective if we restrict ourselves to closed points in A1K Spec K A1K and A1K .
6. Let f : X - Y be a morphism of ane schemes. Show for an ane open
subscheme V Y that its preimage f 1 (V ) is ane as well. In particular, the
intersection of two ane open subschemes in an ane scheme is ane.
7. Let X - S be a surjective morphism of schemes. Show for any base change
morphism S - S that the resulting morphism X S S - S is surjective.
Thus, the surjectivity of scheme morphisms is stable under base change.
7.3 Subschemes and Immersions
Let X be a scheme with structure sheaf OX and U an open subset in X. Then

U together with the sheaf OX |U obtained from OX by restriction to U is a
scheme again, a so-called open subscheme of X, as we have already pointed out
in Section 6.6.
On the other hand, there is a canonical way to introduce closed subschemes
of a scheme X. Namely, let A be a ring and a A an ideal. Then the projection
A - A/a induces a morphism of ane schemes
Spec A/a - Spec A
and we know from 6.2/7 that the latter constitutes a homeomorphism from
Spec A/a onto the closed subset V (a) Spec A. Equipping V (a) with the struc-
ture sheaf induced from A/a we get an ane scheme that we may call a closed
subscheme of Spec A. We will see that this construction can be extended to
arbitrary schemes X in place of Spec A if we start out from a quasi-coherent
ideal I OX replacing the ideal a A.
Finally, both notions of subschemes can be combined. If U - X is an
open subscheme of a scheme X and Z - U a closed subscheme of U , we
arrive at a so-called locally closed subscheme Z - X.
Denition 1. Let X be a topological space and OX a sheaf of rings on X.

An OX -ideal is dened as a sheaf of OX -submodules I OX , i.e. as a sheaf
functor associating to every open subset U X an ideal I(U ) OX (U ) in
7.3 Subschemes and Immersions 305
such a way that the structure of I(U ) as an ideal in OX (U ) is compatible with

restriction morphisms of I and OX .
Now let X = Spec A be an ane scheme. For any A-module M we have

constructed in Section 6.6 the associated sheaf of OX -modules M , the latter
being given on basic open subsets D(f ) Spec A where f A by

M D(f ) = Mf = M A Af .
Applying this construction to an ideal a A, we see from 6.8/4 that the

resulting OX -module I = a is a sheaf of ideals in OX . Therefore, calling a sheaf
of ideals I on an arbitrary scheme X quasi-coherent if it is quasi-coherent in the
sense of OX -modules, the characterization 6.8/10 of quasi-coherent OX -modules
carries over to sheaves of ideals as follows:
Theorem 2. Let X be a scheme and I an OX -ideal. Then the following con-

ditions are equivalent:
(i) I is quasi-coherent.
(ii) Every x X admits an ane open neighborhood U X such that I|U
is associated to an ideal in OX (U ).
(iii) The restriction I|U to any ane open subscheme U X is associated
to an ideal in OX (U ).
Also let us mention the counterpart of 6.8/7 for sheaves of ideals:
Lemma 3. Let A be a ring and X = Spec A. Furthermore, let a be an ideal

in A and I OX the sheaf of ideals associated to a. Then, for any ane open
subscheme U = Spec B of X, the canonical map a A B - I(U ) B is
bijective and, hence, I|U is associated to the ideal aB B.
In the following consider a scheme X and a quasi-coherent ideal I OX . In

order to dene the zero set of I, look at any ane open subscheme U = Spec A in
X. Then I|U is associated to some ideal a A and, as explained in Section 6.1,
we can look at its zero set VU = V (a) U X, which is a closed subset in U .
If U varies over all ane open subschemes of X (or just over the members of
some ane open covering of X), the union of the VU yields a well-dened subset
V (I) X such that V (I) U = VU for all ane open subschemes U X.
Indeed, if U = Spec A and U = Spec A are ane open subschemes in X and
if I is associated to the ideal a A on U and the ideal a A on U , then
Lemma 3 shows for every ane open subscheme U = Spec A U U that
the sets VU U and VU U coincide, as they coincide with V (a ) where a
is the ideal generated by a or, alternatively, a in A . In particular, we see that
V (I) is closed in X, since there exists an open covering (Ui )iI of X such that
V (I) Ui is closed in Ui for all i I.
The denition of V (I) can be phrased in a more elegant way by passing to
the quotient OX /I and by dening V (I) as the support of this sheaf, namely,
V (I) = supp OX /I = {x X ; (OX /I)x = 0}.

Let us show that, indeed, both denitions describe the same subset in X. Using
the exactness of inductive limits (which is easy to obtain by a straightforward
argument) we get
(OX /I)x = OX,x /Ix = Ax /aAx
if U = Spec A is an ane open neighborhood of x in X and I|U is associated
to the ideal a A. Furthermore, Ax /aAx = 0 is equivalent to aAx = Ax = Apx ,
hence to a px , and therefore to x V (a). Consequently, we see that
(supp OX /I) U = V (a)
and both denitions of V (I) coincide.
We want to equip V (I) with a structure sheaf such that V (I) becomes
a scheme. If X is ane, say X = Spec A, we have already indicated how to
proceed. Namely then I is associated to an ideal a A so that V (I) = V (a).
Since the canonical morphism Spec A/a - Spec A induces a homeomor-
phism Spec A/a - V (a) by 6.2/7, we can transport the structure sheaf
of Spec A/a to V (a) and thereby obtain an ane scheme Y with underlying
topological space V (I). If X is not necessarily ane, we cover it by ane open
subschemes U X and equip each V (I) U with the structure of an ane
scheme YU , as just explained. Then we can try to glue the ane schemes YU
to produce a global scheme Y . This functions well, as follows from the con-
struction of the spectrum of a quasi-coherent OX -algebra in Section 7.1. In-
deed, using the easy part of 6.8/11, it is seen that OX /I is a quasi-coherent
OX -algebra. Furthermore, the projection OX - OX /I gives rise to a canon-
ical morphism Spec OX /I - X = Spec OX . The latter induces a homeomor-
phism Spec OX /I - V (I), which corresponds on ane open subschemes
U = Spec A X to the canonical homeomorphism Spec A/a - V (a) of
6.2/7, assuming that I|U is associated to the ideal a A. Thus we can state:
Proposition and Denition 4. Let X be a scheme and I OX a quasi-

coherent ideal. Then:
(i) The zero set V (I) = supp OX /I of I is closed in X.
(ii) The quotient OX /I is a quasi-coherent OX -algebra.
(iii) The canonical map Spec OX /I - X gives rise to a homeomorphism
Spec OX /I - V (I).
In particular, V (I) can be identied with Spec OX /I and thereby is equipped
with the structure of a scheme. Schemes of this type will be referred to as closed
subschemes of X.
There is a reverse construction showing that every closed subset of a scheme

X occurs as the zero set of a quasi-coherent ideal in OX . To explain this, consider
a subset Y X and look at the ideal IY OX that on arbitrary open subsets
U X is given by the functor
U - {h OX (U ) ; h(x) = 0 for all x Y U }.
The equation h(x) = 0 might be read in the residue eld k(x) of x or can
alternatively be interpreted as h|U (x) = 0, for any ane open neighborhood
U of x in U . We call IY the vanishing ideal in OX associated to Y . Clearly,
for any ane open part U = Spec A in X, we have IY (U ) = IA (Y U ) where
IA A is the vanishing ideal in the sense of Section 6.1.
Proposition 5. Let X be a scheme and Y X a closed subset. Then IY OX

is a quasi-coherent ideal satisfying V (IY ) = Y . In particular, Y can be equipped
with the structure of the closed subscheme Spec OX /IY of X in the manner of
Proposition 4. We say that Y is provided with its canonical reduced structure.
The term reduced structure alludes to the fact that the sheaf of rings OX /IY
does not contain non-trivial nilpotent elements. Also note that in general an
ideal I OX satisfying V (I) = Y is not fully characterized by this equation,
since any positive power of I has the same zero set as I itself.
Proof of Proposition 5. It is enough to consider the case where X is ane, say

X = Spec A. Then, being a closed subset of X, we know from 6.1/5 (ii) that Y
equals the zero set of the ideal IA (Y ) A. We claim that IY is associated to
this ideal and, hence, that IY is quasi-coherent. Indeed, consider a basic open
subscheme
) in X where f A. Then Y U coincides with the zero
U = D(f
set V IA (Y ) Af of the ideal generated by IA (Y ) in Af . Hence, using 6.1/5 (i)
and Lemma 6 below, we get

IAf (Y U ) = rad IA (Y ) Af = rad IA (Y ) Af = IA (Y ) Af ,
which means IY (U ) = IA (Y ) Af . This shows that IY is associated to IA (Y )
and, furthermore, that V (IY ) = Y .
Lemma 6. Let A be a ring, a A an ideal and f an element of A. Then we

have rad(aAf ) = (rad a)Af .
Proof. Since the formation of residue rings is compatible with localization, it is

not hard to show that we may replace A by A/a and thereby assume a = 0.
Then, trivially, we have rad(A)Af rad(Af ). To obtain the opposite inclusion,
consider an element h rad(Af ), hence, h Af such that hn = 0 for some
n
n N. If h = far for some a A and r N, we get a1 = 0 in Af and there
exists some s N such that f s an = 0 in A. This shows f a rad(A) and
therefore f1a rad(A) Af so that h rad(A) Af .
Denition 7. A morphism of schemes f : Y - X is called an open (resp.

closed) immersion if there exists an open (resp. closed ) subscheme Y of X such
that f decomposes into the composition
f: Y - Y - X
of an isomorphism Y - Y and the canonical morphism Y
- X.
Lemma 8. For an ane scheme X = Spec A and a morphism of schemes

f: Y - X the following conditions are equivalent:
(i) The morphism f is a closed immersion.
(ii) The scheme Y is ane, say Y = Spec B, and the ring homomorphism
: A - B corresponding to f is surjective.
Proof. If f is a closed immersion, there is a quasi-coherent ideal I OX

such that f induces an isomorphism Y - Spec OX /I. Since X is ane,
I is associated to some ideal a A. It follows that Spec OX /I Spec A/a
and, hence, Y are ane, say Y = Spec B. In particular, the homomorphism
: A - A/a - B is surjective.
Conversely, let Y be ane, say Y = Spec B, and assume that the homo-
morphism : A - B associated to f is surjective. Then, writing a = ker ,
we see that induces an isomorphism A/a - B, and it follows that f fac-
tors through an isomorphism Y - Spec A/a where Spec A/a is viewed as a
closed subscheme of X = Spec A.
Proposition 9. For a morphism of schemes f = (f, f # ) : Y - X the fol-

lowing conditions are equivalent:
(i) f is a closed immersion.
(ii) For every ane open subscheme U X, the morphism f 1 (U ) - U
induced from f is a closed immersion; by Lemma 8 the latter amounts to f 1 (U )
being ane and f # (U ) : OX (U ) - OY (f 1 (U )) surjective.
(iii) There exists an ane open covering (Ui )iI of X such that the morphism
f 1 (Ui ) - Ui induced from f is a closed immersion for every i I; as before,
the latter means f 1 (Ui ) is ane and f # (Ui ) : OX (Ui ) - OY (f 1 (Ui )) is
surjective.
Proof. Assume rst that f is a closed immersion. Then there is a quasi-coherent

ideal I OX such that f induces an isomorphism f : Y - Spec OX /I.
Now if U X is an open subscheme, the ideal I|U OU = OX |U is quasi-
coherent and f restricts to an isomorphism f 1 (U ) - Spec OU /(I|U ).
1
Hence, f (U ) - U is a closed immersion, and the implication (i) = (ii) is
clear.
The implication (ii) = (iii) being trivial, it remains to verify the implica-
tion (iii) = (i). For this it is enough to
show that there exists a quasi-coherent
ideal I OX satisfying I(Ui ) = ker f # (Ui ) for all i I. Indeed, f factors
then through a morphism f : Y - Spec OX /I, followed by the canonical
morphism Spec OX /I - X, and we see from the assumption in (iii) that f
is an isomorphism.
To get hold of such a quasi-coherent OX -ideal I, look at the sequence
0 - I - OX - f OY - 0
where f OY is the direct image sheaf given by the functor U - OY (f 1 (U ))

for open subsets U X (see 5.3) and I the kernel of OX - f (OY ).

Then, clearly, I(Ui ) = ker f # (Ui ) for all i I. Now observe that f OY is
a quasi-coherent OX -module by 6.8/10 since f (OY )|Ui is quasi-coherent for ev-
ery i I; the latter follows from the assumption in (iii) in conjunction with
6.9/4 since f 1 (Ui ) - Ui is a morphism of ane schemes. Furthermore, we
see from 6.8/4 that I, as the kernel of a morphism between quasi-coherent
OX -modules, is quasi-coherent itself, as claimed. We could add that the mor-
phism OX - f OY is an epimorphism and, hence, the above sequence is
exact, although we do not really need this.
Finally, we want to discuss locally closed subschemes of a given scheme X.

To do this, recall that a subset Y of a topological space X is called locally
closed if, for every y Y , there exists an open neighborhood U (y) X such
that Y U (y) is closed in U (y). In particular, Y is then a closed subset of the
open subset yY U (y) X.
Denition 10. Let X be a scheme. A scheme Y is called a (locally closed)

subscheme of X if there exists an open subscheme U X such that Y is
a closed subscheme of U . A morphism of schemes f : Y - X is called a
(locally closed) immersion if f induces an isomorphism Y - Y to a locally

closed subscheme Y of X, i.e. if f decomposes into a composition of a closed
immersion Y - U and an open immersion U - X.
It follows from Proposition 5 that every locally closed subset Y of a scheme

X can be equipped with the structure of a locally closed subscheme of X.
Proposition 11. Let f : Y - X be a locally closed immersion of schemes

and U an open subscheme of X such that f (Y ) is a closed subset of U . Then
f restricts to a closed immersion Y - U.
Proof. In view of Remark 12 below we may assume that U is the largest open
subscheme of X such that f (Y ) is closed in U . Since f is a locally closed
immersion, there is an open subscheme U X such that f restricts to a closed
immersion Y - U ; then necessarily U U . Now choose an ane open
covering (Ui )iI of U such that Ui U or Ui f (Y ) = for all i I. This is
possible, since f (Y ) is closed in U . Then f 1 (Ui ) - Ui is a closed immersion
by Proposition 9 if Ui U . On the other hand, if Ui f (Y ) = , the same is
true because f 1 (Ui ) = . Therefore it follows from Proposition 9 again that f
induces a closed immersion Y - U.
Remark 12. The following conditions on a subset Y of a topological space X

are equivalent:
(i) Y is locally closed in X.
(ii) There exists an open subset U containing Y such that Y is closed in U .
If Y is locally closed in X, there exists a largest open subset U X con-
taining Y such that Y is closed in U .
Proof. If Y X is locally closed, choose for each y Y an open

neighborhood
U (y) X such that Y U (y) is closed in U (y). Then U = yY U (y) is open
in X and Y is closed in U . This establishes the implication (i) = (ii), the
converse being trivial.
If Y is locally closed in X, we consider the set U of all points x X such
that there exists an open neighborhood U (x) X whose restriction Y U (x)
is closed in U (x). Then, apparently, U is the largest open subset in X such that
Y is closed in U . By the way, U = X (Y Y ).
Proposition 13. Open, closed, and locally closed immersions are preserved
under composition and base change.
Proof. Using 7.2/5 it is easily seen that open immersions are preserved under
base change. The same is true for closed immersions in the ane case, due to
the right exactness of the tensor product, and in the general case by applying
Proposition 9. It follows that the case of locally closed immersions is clear as
well.
Next, it is obvious that the composition of two open immersions yields an
open immersion again. The same is true for closed immersions of ane schemes
and, using Proposition 9, also for closed immersions of general schemes. To
show that locally closed immersions are preserved under composition, look at a
g f
composition of morphisms Z -Y - X. It is enough to consider the case
where g is an open immersion and f a closed immersion. Then g(Z) is open in
Y with respect to the topology induced from X and, hence, there is an open
subscheme X X such that g restricts to an isomorphism Z - Y X .
Thereby we get a decomposition
f
f g: Z - Y X - X - X
where f is obtained from f via the base change X - X and, hence, is a

closed immersion. It follows that f g is the composition of a closed immersion
followed by an open immersion and therefore is a locally closed immersion.
Finally, let us add a local characterization of immersions without giving a

proof; see EGA [11], I, 4.2.2.
Proposition 14. Let f : Y - X be a morphism of schemes.

(i) f is an open immersion if and only if f yields a homeomorphism of Y
onto an open subset in X and fy# : OX,f (y) - OY,y is bijective for all y Y .
(ii) f is a closed (resp. locally closed ) immersion if and only if f yields

a homeomorphism of Y onto a closed (resp. locally closed ) subset of X and
fy# : OX,f (y) - OY,y is surjective for all y Y .
Exercises
1. Characterize all subschemes of the ane line A1K over a eld K. In particular,
show that every subscheme of A1K is open or closed, a fact that does not extend
to ane n-spaces of higher dimension.
2. Give an example of a scheme X and of an ideal sheaf I OX where I is not

quasi-coherent. Hint: Use extension by zero, as introduced in Exercise 6.9/10.
3. Let F be an OX -module on a scheme X. Dene the annihilator of F as the

kernel I of the sheaf morphism OX - Hom (F, F) that associates to a
OX
section f OX (U ) on an open subset U X the multiplication by f on F|U ; see
Exercise 6.5/7 for the denition of Hom sheaves. Show that I is an ideal in OX
and that the latter is quasi-coherent if F is quasi-coherent and locally of nite
type. Hint: Reduce to a problem on modules over rings and proceed by induction
on the number of generators.
4. Monomorphisms: A morphism : X - X in some category C is called a

monomorphism if two C-morphisms f, g : T - X coincide as soon as f = g.
(a) Characterize monomorphisms in the category of sets, resp. groups, resp. rings,
resp. modules.
(b) Show that immersions are monomorphisms in the category of schemes.
(c) Give an example of a scheme morphism that is a monomorphism but not an
immersion. Hint: Look at morphisms of type Spec AS - Spec A for a ring A
and a multiplicative system S A.
5. Let X - S be an immersion of schemes. For any point s S, determine the

ber Xs of X over s.
6. Give an example showing that the ber Xs of a relative scheme X - S over

a point s S is not necessarily a subscheme of X.
7. Inverse images of subschemes: Consider a scheme X together with a subscheme

Z X. Show for any scheme morphism f : Y - X that the preimage f 1 (Z)
admits a canonical structure as a subscheme of Y . If Z is an open, resp. closed,
resp. locally closed subscheme in X, the same is true for f 1 (Z) as a subscheme
of Y .
8. Reduced schemes: A scheme X is called reduced if the ring OX (U ) is reduced for

all open subsets U X. Show:
(a) X is reduced if it can be covered by ane open subschemes of type Spec A
where A is reduced.
(b) There exists a unique closed subscheme Xred X that is reduced and point-
wise coincides with X. In terms of topological spaces, the canonical morphism
Xred - X is a homeomorphism.
(c) Every morphism T - X from a reduced scheme T factors uniquely through
Xred .
(d) Every morphism of schemes f : X - Y gives rise to a unique morphism
fred : Xred - Yred such that the obvious diagram is commutative.
7.4 Separated Schemes
Given a relative scheme X over a base scheme S, we can consider the diagonal
morphism : X - X S X, which is characterized by the fact that the
composition pi : X - X with each projection p1 , p2 : X S X - X is
the identity morphism. The image (X) is called the diagonal in X S X.
Note that a morphism of S-schemes : T - X S X factors through
the diagonal morphism : X - X S X if and only if p1 = p2 since
then p1 = p2 coincides with , as can be checked by composing
both morphisms with the projections p1 , p2 . On the other hand, the condition
that p1 coincides with p2 on all points t T is not sucient for such a
factorization. For example, view Spec C as a relative scheme over Spec R. Then,
as we have seen in Section 7.2, the ber product Spec C Spec R Spec C consists
of two points and, hence, the diagonal morphism
: Spec C - Spec C Spec R Spec C
will not be surjective. Therefore the identity map = id on the ber product
Spec C Spec R Spec C cannot be factored through the diagonal morphism ,
although we have (p1 id)(t) = (p2 id)(t) for all points t Spec CSpec R Spec C,
due to the fact that Spec C is a one-point space. In particular, for an arbitrary
S-scheme X, we observe that the obvious inclusion
(X) {z X S X ; p1 (z) = p2 (z)}
will not be an equality in general.
Proposition 1. For any relative scheme X over a base scheme S, the diagonal
morphism : X - X S X is a locally closed immersion. If X and S are
ane, then is even a closed immersion.
Proof. To begin with, let X and S be ane, say X = Spec A and S = Spec R.
Then the structural morphism X - S corresponds to a ring homomorphism
R - A, which equips A with the structure of an R-Algebra. Since the diagonal
morphism corresponds to the multiplication morphism
A R A - A, ab - ab,
which is surjective, we see from 7.3/8 that is a closed immersion.

If X is not necessarily ane, consider a point x X, as well as an open ane
neighborhood U X of x such that U lies over some ane open subscheme
S S. Then we know from 7.2/5 that U S U is an open subscheme of
X S X satisfying 1 X (U S U ) = U . Furthermore, the diagonal morphism
U : U - U S U is a closed immersion by the special case just dealt with.
Hence, if x varies over all points of X, the commutative diagram
7.4 Separated Schemes 313
U
U - U S U

? ?
X
X
- X S X
shows that X is a locally closed immersion.
Denition 2. Let f : X - S be a morphism of schemes. Then f is called

separated if the diagonal morphism : X - X S X is a closed immersion.
A relative S-scheme X is called separated if the structural morphism X - S
is separated. Furthermore, an absolute scheme X is called separated if it is
separated as a Z-scheme.
Proposition 3. An S-scheme X is separated if and only if the image (X) of

the diagonal morphism : X - X S X is closed in X S X.
Proof. Use the fact that is a locally closed immersion by Proposition 1 and,
thus, a closed immersion by 7.3/11.
Recall that a topological space X is said to satisfy the Hausdor separation

axiom if any dierent points x, y X admit disjoint open neighborhoods in X;
the latter is equivalent to the fact that the diagonal is closed in the cartesian
product X X with respect to the product topology. Hence, although schemes
do not satisfy the Hausdor separation axiom in general, the notion of separat-
edness on schemes may be viewed as an adaptation of this axiom to the scheme
case.
We can conclude from Proposition 1:
Remark 4. Let f : X - S be a morphism of ane schemes. Then f is

separated.
Proposition 5. Let X be a relative scheme over an ane base S = Spec R and

let : X - X S X be the corresponding diagonal morphism. Then, for a
given ane open covering (Xi )iI of X, the following conditions are equivalent:
(i) X is separated over S.
(ii) For all i, j I, the diagonal morphism induces a closed immersion
Xi Xj - X i S Xj .
(iii) For all i, j I, the intersection Xi Xj is ane and induces a

surjection OX (Xi ) R OX (Xj ) - OX (Xi Xj ).
Proof. Using 7.2/5 we see that (Xi S Xj )i,jI is an ane open covering of
X S X satisfying 1 (Xi S Xj ) = Xi Xj . Now apply 7.3/8 and 7.3/9.
Corollary 6. Let S be an ane scheme and X a separated S-scheme. Then, if

U and V are ane open subschemes of X, the intersection U V is an ane
open subscheme of X as well.
Proof. It follows from condition (ii) of the above proposition that the diagonal
morphism : X - X S X restricts to a closed immersion U V - U S V .
Since U , V , and S are ane, the ber product U S V is ane by 7.2/4.
Therefore U V is ane by 7.3/8.
We want to give an example of an S-scheme X over an ane base S, for

which the conclusion of Corollary 6 cannot be obtained. In particular, X cannot
be separated then. To construct such a scheme, look at the ane n-space AnK
over some eld K and assume n 2. Let 0 AnK be the origin, i.e. the point
given by the maximal ideal (t1 , . . . , tn ) Kt1 , . . . , tn . Now glue two copies
of AnK by identifying the open subschemes AnK {0} AnK via the identity
morphism id : AnK {0} - An {0}. Thereby we obtain a K-scheme X
K
admitting an ane open covering by two copies X1 , X2 of AnK . In fact, we may
interpret X as the ane n-space over K with a double origin. Then X1 X2
nAK {0} and the latter is not ane for n 2. Indeed, we have
n
coincides with
AK {0} = i=1 D(ti ) and
n

n
Kt1 , . . . , tn t1
i = Kt1 , . . . , tn
i=1
for n 2. This shows that the restriction morphism
OAnK (AnK ) - OAnK (AnK {0})
is bijective and, hence, that AnK {0}, being strictly contained in AnK , cannot
be ane.
Next, let us derive some criteria and properties for separated morphisms; see
also 9.5/17 for the valuative criterion of separatedness. First, we can conclude
from 7.3/9 that the separatedness can be tested locally on the base:
Proposition 7. Let f : X - S be a morphism of schemes and (Si )iI an

open covering of S. Then the following conditions are equivalent:
(i) f is separated.
(ii) The morphism f 1 (Si ) - Si induced from f is separated for all i I.
Later in 9.5/1 a morphism of schemes f : X - S will be called ane

if there exists an ane open covering (Si )iI of S such that f 1 (Si ) is ane
for all i I. Alternatively, we may require that the preimage f 1 (S ) of any
ane open subscheme S S is ane again; see 9.5/3. The link between both
denitions is provided by 6.8/10. Indeed, if A is a quasi-coherent OS -algebra,
the canonical morphism Spec A - S is ane and one can show a converse:
if f : X - S is an ane morphism of schemes, the direct image f OX is a
quasi-coherent OS -algebra and X is canonically isomorphic to Spec f OX ; see

also Exercise 7.1/5. Using the notion of ane morphisms, we can conclude from
Proposition 7 in conjunction with Remark 4 the following assertion:
Corollary 8. Ane morphisms and, in particular, morphisms of ane schemes,

are separated.
Proposition 9. Let S be an ane scheme and f : X - S a morphism of

1
schemes. Assume that for any points s S and x, y f (s) there is always
an ane open subscheme U X such that x, y U . Then f : X - S is
separated.
Proof. Our assumption implies that X S X admits a covering by ane open

subschemes of type U S U where U is ane open in X. Indeed, consider a point
z X S X with projections x, y X. Then these points lie over the same point
s S and, by our assumption on the ber f 1 (s), we can nd an ane open
subscheme U X containing both, x and y. It follows that U S U X S X
is an ane open neighborhood of z and, using Remark 4, that the diagonal
embedding : X - X S X restricts to a closed immersion U - U S U .
Covering X S X by ane open subschemes of type U S U , we conclude from
7.3/9 that is a closed immersion. Consequently, X is separated over S.
Corollary 10. Let f : X - S be a morphism of schemes such that the under-

lying map of topological spaces is injective. Then f is separated. In particular,
every immersion of schemes is separated.
Let us give some applications of separated morphisms.
Lemma 11. Let 1 : X - S and 2 : Y - S be schemes over some base

scheme S. Then, for every morphism of schemes S - T , the canonical dia-
gram

X S Y - X T Y
1 2
? ?
S
S/T
- S T S
is cartesian.
Proof. Let us write

q1 : X S Y - X, q2 : X S Y - Y,
q1 : X T Y - X, q2 : X T Y - Y,
as well as p1 , p2 : S T S - S for the canonical projections. To check that the
above diagram is commutative, observe that the compositions
p1 (1 2 ) = 1 q1 , p2 (1 2 ) = 2 q2
coincide with the structural morphism : X S Y - S and that this yields

the desired commutativity relation (1 2 ) = S/T .
Next, to show that the diagram is cartesian, look at a scheme Z and mor-
phisms
f1 : Z - X T Y, f2 : Z -S
such that (1 2 ) f1 = S/T f2 . Then we get

1 q1 f1 = p1 (1 2 ) f1 = p1 S/T f2 = f2
= p2 S/T f2 = p2 (1 2 ) f1 = 2 q2 f1 .
Therefore the compositions q1 f1 and q2 f1 are S-morphisms and, using the
universal properties of ber products, these induce a well-dened S-morphism
f: Z - X S Y such that f = f1 and f = f2 . Furthermore, f is
uniquely determined by the latter relations, since the compositions q1 f and
q2 f are uniquely determined as the compositions of f1 with the projections
of the ber product X T Y onto its factors X and Y .
Corollary 12. The morphism : X S Y - X T Y of Lemma 11 is a locally

closed immersion. If S - T is separated, is even a closed immersion.
Proof. Immersions and especially closed immersions are stable under base
change; see 7.3/13.
Proposition 13. Let S be a base scheme and f : X - Y an S-morphism.

Then the graph morphism f : X - X S Y given by id : X - X and
f: X - Y is an immersion, even a closed immersion if Y is a separated
S-scheme.
Proof. We consider the cartesian diagram

f
X = X Y Y - X S Y
? ?
Y - Y S Y
obtained in the situation of Lemma 11 by taking for 2 : Y - S the identity
morphism on Y = S and by replacing T by S. Then the assertion follows from
Corollary 12.
Corollary 14. Let f : X - S be a morphism of schemes and : S -X

a section of f , i.e. a morphism satisfying f = idS . Then is an immersion,
even a closed immersion if f is separated.
Proof. Viewing X as an S-scheme under f , we can say that a section of f is

just an S-morphism : S - X. Then Proposition 13 applied to the morphism
shows that
: S - S S X
is a locally closed immersion, and even a closed immersion if X is a separated

S-scheme. However, coincides canonically with in our situation.
Proposition 15. (i) The composition of two separated morphisms is separated

again.
(ii) For separated morphisms of S-schemes f : X - Y and f : X - Y ,
their ber product f f : X S X - Y S Y is separated as well.
(iii) Separated S-morphisms are stable under base change.
(iv) Let f : X - Y and g : Y - Z be morphisms of schemes. If g f
is separated, the same applies to f .
Proof. Starting with assertion (i), look at the cartesian diagram in Lemma 11 for
X = Y . Then if S - T is separated, the upper row of the diagram is a closed
immersion by Corollary 12. Now if the diagonal morphism X - X S X is
a closed immersion, the same is true for the composition
X - X S X - X T X
by 7.3/13 and we see that X -
S -
T is separated if X - S and
S - T have this property.
Assertion (ii) can be derived from (i) and (iii). To obtain (iii), look at
a separated morphism X - S and a base change morphism S - S.
Then, using 7.2/13 and applying the base change to the diagonal morphism
X - X S X yields a morphism
X S S - (X S X) S S = (X S S ) S (X S S ),
which is a diagonal morphism again. Since closed immersions are stable under
base change by 7.3/13, we are done.
To settle (iv), look at the decomposition
f p2
f: X - X Z Y - Y,
where f is the graph morphism (viewing f as a Z-morphism) and p2 the pro-
jection onto the second factor. We know from Proposition 13 that f is a locally
closed immersion. Even better, being injective it is separated by Corollary 10.
Furthermore, p2 may be viewed as a ber product of g f over Z with the
identity on Y , the latter viewed as a Z-scheme via g. Hence, p2 is separated by
(ii). But then, as a composition of two separated morphisms, f is separated by
(i).
Corollary 16. If X is a separated scheme (over Z), then any morphism of

schemes X - Y is separated.
Exercises
1. Show that a scheme morphism f : X - S is a monomorphism (see Exer-
cise 7.3/4) if and only if the diagonal morphism : X - X S X is an iso-
morphism. In particular, monomorphisms are separated.
2. Show that the projective n-space PnR over any ring R is separated.
3. For two S-morphisms f, g : T - X on some base scheme S consider the ber

product T (XS X) X with respect to the morphisms (f, g) : T - X S X given
by f, g and the diagonal morphism : X - XS X; it is called the coincidence
scheme for f and g. Show:
(a) The projection onto the rst factor : T (XS X) X - T is a locally
closed immersion satisfying the following universal property: Every S-morphism
: Z - T such that f = g admits a unique factorization through .
(b) is a closed immersion for all S-schemes T and all S-morphisms f, g if and
only if X is a separated S-scheme.
(c) In particular, assume that X is separated over S, that T is reduced (see
Exercise 7.3/8), and that f coincides with g on a dense open subscheme U X.
Then the morphisms f and g coincide on X.
4. Consider gluing data for schemes, namely schemes Xi for i varying over some
index set I, as well as open subschemes Xij Xi together with isomorphisms
ij : Xij - Xji for i, j I such that conditions (a), (b) of 7.1/1 are satised.
Give a necessary and sucient condition on the gluing data assuring that the
scheme X obtained from gluing the Xi via the intersections Xij is separated.
5. Let X be an S-scheme, for some base scheme S. The separatedness of X cannot

be tested locally on X. But show that the disjoint union of separated S-schemes
is separated.
6. Let f : X - S be a morphism of schemes and consider closed subsets Xi X

and Si S such that f (Xi ) Si where i varies over a nite index set I. Equip
all sets Xi and Si with their reduced structures so that f induces morphisms
fi : Xi - Si , i I, and assume that the Xi cover X. Show that f is separated
if and only if all fi , i I, are separated.
7.5 Noetherian Schemes and their Dimension
The notion of dimension of a scheme X is based on the length of chains of

closed irreducible subsets in X. Recall from 6.1/13 that a topological space X
is said to be irreducible if it is non-empty and cannot be decomposed into a
union X = X1 X2 of two proper closed subsets X1 , X2 X. Let us start with
the following useful observation:
Remark 1. Let X be an irreducible topological space and U a non-empty open

subset of X. Then U is dense in X and irreducible.
Proof. It follows from 6.1/14 (iii) that U is dense in X. Furthermore, consider a

decomposition U = Z1 Z2 into relatively closed subsets Z1 and Z2 of U . This
implies X = Z 1 Z 2 , and we see that Z 1 = X or Z 2 = X since X is irreducible.
But then Zi = Z i U = U for i = 1 or 2 and U is irreducible.
7.5 Noetherian Schemes and their Dimension 319
As we know from 6.1/17 for ane schemes X = Spec A, the map x - {x}
denes a bijection between X as a point set and the set of its irreducible closed
subsets. We want to generalize this fact to arbitrary schemes.
Proposition 2. Let X be a scheme. Then the map

- irreducible closed -
: X , x {x},
subsets of X
is bijective.
Proof. First, observe that for any x X its closure {x} is irreducible and,
hence, that the map is well-dened. Indeed, if {x} = Z1 Z2 for closed
subsets Z1 , Z2 {x}, we must have x Z1 or x Z2 and, therefore, {x} = Z1
or {x} = Z2 .
Now let Z be an irreducible closed subset in X and choose an ane open
subscheme U X meeting Z. Then the bijectivity of can be obtained by
applying 6.1/17 to the closed subset Z U of the ane scheme U , where Z U
is irreducible and dense in Z by Remark 1. Indeed, by 6.1/17 there exists a
point x that is dense in Z U , and we get Z = {x} since Z U is dense in Z.
On the other hand, if x, x are two points that are dense in Z, we must have
x, x Z U and therefore x = x by 6.1/17. This shows that the map is
bijective.
In particular, we see that any irreducible closed subset Z of a scheme X

admits a unique point z satisfying {z} = Z. As in Section 6.1, z is called the
generic point of Z, whereas the points of its closure {z} (including z itself)
are referred to as the specializations of z. Using Zorns Lemma, one shows that
any scheme X admits maximal irreducible subsets; they are necessarily closed.
These subsets cover X and are referred to as the irreducible components of X.
The associated generic points are called the generic points of X.
Denition 3. A scheme X is called locally Noetherian if each point x X

admits an ane open neighborhood U X whose ring of global sections OX (U )
is Noetherian. If, in addition, X is quasi-compact, it is called a Noetherian
scheme.
It is clear that open or closed subschemes of locally Noetherian schemes are

locally Noetherian again. Furthermore, we want to show that the property of
an ane scheme to be (locally) Noetherian is equivalent to the fact that its ring
of global sections is Noetherian.
Proposition 4. An ane scheme X = Spec A is Noetherian if and only if its

ring A is Noetherian.
Proof. The if part of the assertion is trivial. Therefore assume that X is Noethe-
rian. Then X admits a nite covering by ane open subschemes Ui = Spec Ai ,
i = 1, . . . , n, where each Ai is Noetherian. To show that A itself is Noetherian,
look at some ideal a A and let I be the associated ideal in OX . Then the
restriction I|Ui is associated to the ideal aAi Ai by 7.3/3, and the latter is
nitely generated since Ai is Noetherian. But then a is nitely generated by
6.8/13.
Proposition 5. Any Noetherian scheme X consists of nitely many irreducible

components.
Proof. Let us rst consider the case where X is ane, say X = Spec A. Then
A is Noetherian by Proposition 4 and we see from 6.1/17 that the irreducible
components of X correspond bijectively to the minimal prime ideals in A. It is
known that the number of such ideals is nite in a Noetherian ring; just apply
2.1/12 to the zero ideal in A.
To give a more direct argument in the case of a Noetherian ane scheme
X = Spec A, let M be the set of all non-empty closed subsets Z X that cannot
be written as a nite union of irreducible closed subsets of X. If M is non-empty,
Zorns Lemma in conjunction with the Noetherian property shows that there
exists a minimal element Z M . Indeed, any descending chain Z1 Z2 . . .
of elements in M must become stationary, as by 6.1/7 it is equivalent to an
ascending chain of reduced ideals in A, i.e. of ideals that coincide with their
radicals. Then Z cannot be irreducible and there are proper closed subsets
Z1 , Z2 Z such that Z = Z1 Z2 . Now Z1 and Z2 do not belong to M and
therefore are nite unions of irreducible closed subsets in X. Consequently, the
same is true for Z, which, however, is in contradiction with Z M . Therefore
M must be empty and we see that X itself is a nite union of irreducible closed
subsets. Furthermore, Proposition 2 shows that all irreducible components of
X occur as members of this union and, hence, that their number is nite.
Now, let us look at a general Noetherian scheme X. Then we can use its
quasi-compactness in order to nd a nite ane open covering (Ui )iI of X.
Combining Remark 1 with the ane case, there are only nitely many irre-
ducible components of X meeting each Ui . Thus, the number of irreducible
components of X must be nite.
Finally, we want to discuss the dimension of schemes, a notion based on

the dimension of topological spaces. To compute such dimensions, one relies on
the Krull dimension of rings, which we recall below; for more details on the
dimension of rings see Section 2.4.
Denition 6. For a topological space X, the supremum of the lengths n of
chains of irreducible closed subsets
X0 X1 . . . Xn X,
is denoted by dim X and is called the dimension of X.

7.5 Noetherian Schemes and their Dimension 321
The local dimension of X at a point x X, denoted by dimx X, is given by the

inmum over all dimensions dim U where U is an open neighborhood of x in
X.
For a ring A, the supremum of the lengths n of chains
p0 p1 . . . pn A,
where the pi are prime ideals in A, is denoted by dim A and is called the dimen-
sion of A.
The dimension is also referred to as the Krull dimension of X or A.
For example, we see from 6.1/17 that the dimension of an ane scheme
X = Spec A equals the dimension of its ring of global sections A. In particular,
the polynomial ring Kt1 , . . . , tn in n variables over a eld K is of dimension n
(see 2.4/16) and the same is true for the ane n-space AnK . From this one easily
concludes that the projective n-space PnK , as introduced in Section 7.1, is of
dimension n as well. Also note that by convention the supremum over an empty
family of integers is . This way the zero ring and the empty topological
space are said to have dimension .
Denition 7. For a topological space X and an irreducible closed subset Z X,

the supremum of the lengths n of chains
Z = X0 X1 . . . Xn X
where the Xi are irreducible closed subsets of X is denoted by codimX Z and is
called the codimension of Z in X.
For a ring A and a prime ideal p A, the supremum of the lengths n of
chains
p0 p1 . . . pn = p A,
where the pi are prime ideals in A, is denoted by ht p and is called the height
of p.
For an ane scheme X = Spec A and an irreducible closed subscheme

Z X it follows from 6.1/17 again that codimX Z = ht p where p = I(Z) is the
prime ideal in A associated to Z. Furthermore, one knows that the dimension of
any Noetherian local ring is nite (see 2.4/8) and that the height of any prime
ideal p in a Noetherian ring is nite as well (see 2.4/7). From this one easily
deduces that the codimension of any irreducible closed subscheme Z in a locally
Noetherian scheme X is nite.
Exercises
1. A topological space X is called Noetherian if every descending sequence of closed
subsets X Z1 Z2 . . . becomes stationary. Show for a scheme X that
its underlying topological space is Noetherian if X is Noetherian, but that the
converse is not true in general.
2. Show that a topological space X is Noetherian (see Exercise 1) if and only if each
open subset in X is quasi-compact.
3. Let X be a scheme whose underlying topological space is Noetherian; see Exer-
cise 1. Show that X contains a closed point.
4. Let F be a quasi-coherent OX -module on a locally Noetherian scheme X. Show
that F is coherent if and only if it is locally of nite type.
5. Let X be a locally Noetherian scheme. Show that the set of points x X where X
is reduced (resp. integral) in the sense that the stalk OX,x satises rad(OX,x ) = 0
(resp. is an integral domain) is open in X. Hint: Use Exercise 6.8/7.
6. For any scheme X prove dim X = supxX (dim OX,x ).
7. For a discrete valuation ring R, i.e. a principal ideal domain that is a local ring,
consider the scheme S = Spec R. Determine the dimension dim S as well as the
local dimension dims S for all points s S. Do the same for the ane n-space AnS
and the projective n-space PnS . Hint: View AnS and PnS as relative schemes over S
and look at their bers.
7.6 Cech Cohomology
Consider a topological space X and an open covering U = (Ui )iI of it. Further-
more, let us x a presheaf F, say of abelian groups, on X. Setting
Ui0 ...iq = Ui0 . . . Uiq
for indices i0 , . . . , iq I, we dene the group of q-cochains on U with values in

F by
C q (U, F) = F(Ui0 ...iq ).
i0 ...iq I
A cochain g C (U, F) is called alternating if

q
gi(0) ...i(q) = sgn() gi0 ...iq
for any indices i0 , . . . , iq I and any permutation Sq+1 (or, equivalently, for
any transposition Sq+1 ) and if, furthermore, gi0 ...iq = 0 for indices i0 , . . . , iq
that are not pairwise distinct; Sq+1 is the symmetric group of index q + 1. The
alternating q-cochains form a subgroup Caq (U, F) of C q (U, F).
There is a so-called coboundary map
dq : C q (U, F) - C q+1 (U, F),
given by

q+1
(dq g)i0 ...iq+1 = (1)j gi0 ...ij ...iq+1 |Ui0 ...iq+1 ,
j=0
7.6 Cech Cohomology 323
which satises dq+1 dq = 0 and maps alternating cochains into alternating

ones, as is easily veried by looking at transpositions of index tuples (ij means
that the index ij is to be omitted). Thus, we obtain a complex
d0 d1 d2
0 - C 0 (U, F) - C 1 (U, F) - C 2 (U, F) - ...,
which is called the complex of Cech cochains on U with values in F. In short, it

is denoted by C (U, F). Similarly, there is the complex
d0a d1a d2a
0 - Ca0 (U, F) - Ca1 (U, F) - Ca2 (U, F) - ...,
of alternating Cech cochains on U with values in F, denoted by Ca (U, F). As-

sociated to these complexes are the Cech cohomology groups
H q (U, F) = ker dq / im dq1 ,

Haq (U, F) = ker dqa / im dq1
a ,
which are dened for q N (set d1 = 0 and d1

a = 0). There is a canonical
homomorphism
: F(X) - Ca0 (U, F) = C 0 (U, F), g - (g|Ui )iI ,
called the augmentation morphism. It leads to the so-called augmented Cech

complexes
d0 d1 d2
0 - F(X) - C 0 (U, F) - C 1 (U, F) - C 2 (U, F) - ...,
d0a d1a d2a
0 - F(X) - Ca0 (U, F) - Ca1 (U, F) - Ca2 (U, F) - ....
Note that in case F is a sheaf, the augmentation morphism yields an isomor-

phism
F(X) - Ha0 (U, F) = H 0 (U, F).
In particular, we see that the Cech cohomology groups of level 0 do not dier
when working with all cochains or merely with alternating ones. This phe-
nomenon extends to higher cohomology groups as follows:
Lemma 1. Let F be a presheaf on a topological space X and U an open covering

of X. Then the inclusion : Ca (U, F) - C (U, F) induces isomorphisms of
cohomology groups
Haq (U, F) - H q (U, F), q N.
Proof. We will construct a complex homomorphism p : C (U, F) - Ca (U, F)

such that p is the identity on Ca (U, F) and p is homotopic to the identity
on C (U, F) in the sense of 5.4/1. This will imply the assertion.
Digressing for a moment, we consider the sequence
d0 d1 d2 d3
0 F0 F1 F2 ...
where Fq is the free Z-module generated by I q+1 , namely,

Fq = Z (i0 , . . . , iq ),
i0 ,...,iq I
and where dq : Fq - Fq1 for q > 0 is the Z-linear map given by

q
dq (i0 , . . . , iq ) = (1)j (i0 , . . . , ij , . . . , iq ).
j=0
Since dq dq+1 = 0, the modules Fq and maps dq constitute a complex F of

Z-modules. Calling a Z-linear map : Fr - Fs between two modules of F
simplicial if
(i0 , . . . , ir ) Z (j0 , . . . , js )
j0 ,...,js {i0 ,...,ir }
for all i0 , . . . , ir I, it is clear that all maps dq are simplicial.

Now x a total ordering on I. Then we can dene simplicial homomorphisms
q : Fq - Fq as follows. Let i0 , . . . , iq I. If the indices are not pairwise dier-
ent, set q (i0 , . . . , iq ) = 0. Else there is a unique permutation of i0 , . . . , iq such
that i(0) < i(1) < . . . < i(q) , and we set q (i0 , . . . , iq ) = (sgn )(i(0) , . . . , i(q) ).
It is easily seen that the q constitute a complex homomorphism F - F .
We want to show that there is a simplicial homomtopy h in the sense of 5.1/3
between and the identity map id.
To achieve this, we construct by induction simplicial homomorphisms
h q : Fq - Fq+1 , q 0, such that
q idq = hq1 dq + dq+1 hq .
Since 0 id0 is the zero map, we can start with h1 = 0 and h0 = 0. If hq1 is
already constructed for some q 1, a standard calculation shows that
q idq hq1 dq : Fq - Fq
maps Fq into ker dq . Furthermore, this map is simplicial. Thus, xing indices
i0 , . . . , iq I, there is an equation of type
i ,...,i
a := (q idq hq1 dq )(i0 , . . . , iq ) = cj00 ,...,jqq (j0 , . . . , jq )
j0 ,...,jq {i0 ,...,iq }
i ,...,i
with coecients cj00 ,...,jqq Z, and the relation dq a = 0 yields
i ,...,i

q
() cj00 ,...,jqq (1)k (j0 , . . . , jk , . . . , jq ) = 0.
j0 ,...,jq {i0 ,...,iq } k=0
Now let

a =
i ,...,i
cj00 ,...,jqq (i0 , j0 , . . . , jq ).
j0 ,...,jr {i0 ,...,iq }
Then

dq+1 a =
i ,...,i
cj00 ,...,jqq (j0 , . . . , jq )
j0 ,...,jr {i0 ,...,iq }
i ,...,i

q
+ cj00 ,...,jqq (1)k+1 (i0 , j0 , . . . , jk , . . . , jq )
j0 ,...,jr {i0 ,...,iq } k=0
= a.
Indeed, the rst summand is a and the second one is trivial, as follows from
the relation (), transferred from Fq1 to Fq under the Z-linear map given by
(j0 , . . . , jq1 ) - (i0 , j0 , . . . , jq1 ). Thus, dening hq : Fq - Fq+1 by associ-

ating to any tuple (i0 , . . . , iq ) the corresponding element a constructed above,
we see that hq is a simplicial homomorphism as desired.
Turning back to Cech cohomology, observe that any simplicial homomor-
phism : Fr - Fs between two modules of F induces canonically a homo-
morphism : C (U, F)
s - C r (U, F). Indeed, if : Fr - Fs is determined
by the equations

,...,js (j0 , . . . , js ),
aij00,...,i r
(i0 , . . . , ir ) =
j0 ,...,js {i0 ,...,ir }
we set for any g C s (U, F)

,...,js gj0 ,...,js |Ui0 ,...,ir .
aij00,...,ir
(g)i0 ,...,ir =
j0 ,...,js {i0 ,...,ir }
The correspondence - is additive and functorial (contravariant); in par-

ticular, dq equals the coboundary morphism dq1 : C q1 (U, F) - C q (U, F).
Thus, the morphisms q derived from the q above constitute a complex ho-
momorphism : C (U, F) - C (U, F) which is homotopic to the identity
via the morphisms hq1 : C (U, F) - C q1 (U, F). Now observe that q (g) for
q
any g C q (U, F) is characterized by
q (g)i0 ,...,iq = 0
if i0 , . . . , iq I are not pairwise distinct and else by
q (g)i0 ,...,iq = sgn() gi(0) ,...,i(q)
where Sq+1 satises i(0) < . . . < i(q) . From this we easily see that
maps C (U, F) onto Ca (U, F) and restricts to the identity on Ca (U, F). Hence,
it induces a complex homomorphism p : C (U, F) - Ca (U, F) as required for
our proof.
There is an immediate consequence:

Corollary 2. If the covering U consists of n elements, then
H q (U, F) = Haq (U, F) = 0 for q n.
Proof. We have Caq (U, F) = 0 for q n if U consists of n elements.
Now let us consider two open coverings U = (Ui )iI and V = (Vj )jJ of
a topological space X and assume that V is a renement of U. Hence, there
exists a map : J - I such that Vj U (j) for j J. Any such map induces
homomorphisms
q : C q (U, F) - C q (V, F), q 0,
mapping a cochain g C q (U, F) to the cochain q (g) given by

q
(g) j0 ,...,jq = g (j0 ),..., (jq ) |Vj0 ,...,jq .
The maps q constitute a complex homomorphism
: C (U, F) - C (V, F),
as is easily checked, and it is clear that maps alternating cochains to alter-

nating ones.
Although the map : J - I might not be uniquely determined by the
coverings U and V, one can show that the induced maps
H q ( ) : H q (U, F) - H q (V, F), q 0,
are independent of . To do this, let : J - I be a second map satisfying

Vj U (j) for all j J. Then one veries that the homomorphisms
hq : C q (U, F) - C q1 (V, F), q 0,
given by

q1
hq (g) j0 ,...,jq1
= (1)k g (j0 ),..., (jk ), (jk ),..., (jq1 ) |Vj0 ,...,jq1
k=0
dene a homotopy between and . Thus, the maps H q ( ) and H q ( )

must coincide for all q. We will use the notation q (U, V) instead of H q ( ) or
H q ( ). Note that q (U, U) = id and that q (U, W) = q (V, W) q (U, V) if W
is a renement of V and V a renement of U. In particular, we can conclude:
Proposition 3. Assume that the coverings U and V are renements of each

other. Then q (U, V) : H q (U, F) - H q (V, F) is bijective, and its inverse is
q
(V, U) for all q.
In order to obtain cohomology groups on X that do not depend on a cer-

tain covering, we x q and take the inductive limit of the groups H q (U, F)
for U varying over the collection Cov(X) of all open coverings of X. Writ-
ing U V if V is a renement of U we get a preorder on Cov(X), and we
see that Cov(X) is directed, as the product covering (Ui Vj )iI,jJ is a com-
mon
q renement of two
given members (Ui )iI and (Vj )jJ of Cov(X). Therefore
H (U, F), q (U, V) U,VCov(X) is an inductive system. The associated limit
H q (X, F) = lim H q (U, F),

UCov(X)
exists by 6.4/2 and the explanations preceding it. It is called the qth Cech
cohomology group on X with values in F. The above constructions can be carried
out in the same way for alternating cochains, thus, leading to the alternating
Cech cohomology groups
Haq (X, F) = lim Haq (U, F),

UCov(X)
which, however, coincide with the previous ones due to Lemma 1.

As an example, we compute the Cech cohomology groups of a quasi-coherent
module on an ane scheme. Our method is similar to the one applied in 6.6/2.
Proposition 4. Let X be an ane scheme and F a quasi-coherent sheaf of

OX -modules. Then, for any nite covering U = (Ui )iI of X by basic open
subsets in X, the augmented Cech complex
d0 d1 d2
0 - F(X) - C 0 (U, F) - C 1 (U, F) - C 2 (U, F) - ...,
is exact so that H 0 (U, F) = F (X) and H q (U, F) = 0 for all q 1. In particular :

F(X) if q = 0
H (X, F) =
q
0 if q > 0
Proof. The exactness of the augmented Cech complex at C 0 (U, F) simply reects
the sheaf property of F. To show H q (U, F) = 0 for q 1, let X = Spec A. Then
we know from 6.8/10 that F is associated to an A-module M . Furthermore, if
Ui = D(fi ) for elements fi A, the fi generate the unit ideal in A since the Ui
cover X.
Now assume q 1 and consider a cochain g C q (U, F) such that dq g = 0.
We have to construct a cochain g C q1 (U, F) satisfying dq1 g = g. To do
this, look at the relation dq g = 0, which means

q+1
(dq g)i0 ...iq+1 = (1)j gi0 ...ij ...iq+1 |Ui0 ...iq+1 = 0
j=0
for all indices i0 , . . . , iq+1 I. Since Ui0 ,...,iq = D(fi0 . . . fiq ), there exist elements
hi0 ,...,iq M and an integer r N such that gi0 ,...,iq = hi0 ,...,iq (fi0 . . . fiq )r . Then,
writing i instead of i0 and shifting i1 , . . . , iq+1 back to i0 , . . . , iq , we get

q
hi0 ,...,iq (fi0 . . . fiq ) r
= (1)j fir hi,i0 ,...,ij ,...,iq (fi0 . . . fij . . . fiq )r
j=0
on Ui,i0 ,...,iq , for all indices i, i0 , . . . , iq I. Since Ui,i0 ,...,iq = Ui0 ,...,iq D(fi ), there
is some integer n N such that the equation

q
r
() fin+r hi0 ,...,iq (fi0 . . . fiq ) = (1)j fin hi,i0 ,...,ij ,...,iq (fi0 . . . fij . . . fiq )r
j=0
holds already on Ui0 ,...,iq .

Now using the fact that the elements fi , i I, generate the unit ideal in A,
there are elements bi A such that iI bi fin+r = 1. Then we set up a cochain
g C q1 (U, F) by

gi0 ,...,iq1 = bi fin hi,i0 ,...,iq1 (fi0 . . . fiq1 )r
iI
q1
and claim that d g = g. Indeed, using the relation (), we have

q
(dq1 g )i0 ,...,iq = (1)j gi0 ,...,ij ,...,iq |Ui0 ,...,iq
j=0

q

= (1)j bi fin hi,i0 ,...,ij ,...,iq (fi0 . . . fij . . . fiq )r
j=0 iI

= bi fin+r hi0 ,...,iq (fi0 . . . fiq )r
iI
= gi0 ,...,iq
for any indices i0 , . . . , iq I. Thus, we are done.
Exercises
1. Glue two copies of the ane line A1K = Spec K over a eld K via the
K-isomorphisms K, 1 - K, 1 given by - , resp. - 1 .
The resulting K-scheme X comes equipped with an ane open covering U con-
sisting of two ane lines and equals the ane line with a double origin A1K , resp.
the projective line P1K , as explained in the introduction to the present Chapter 7.
Compute the Cech cohomology groups H q (U, OX ), q 0, in both cases.
2. Consider the 1-sphere S 1 = {z C ; |z| = 1} as a topological space under
the topology induced from C. Compute the Cech cohomology groups H q (S 1 , Z),
q 0, with values in the constant sheaf Z on S 1 . Hint: Compute the Cech co-
homology groups H q (Un , Z) where Un , for integers n > 0, consists of the open
sets
t
z = e 2n 2i ; < t < + 2 S 1 , = 0, . . . , 2n 1.
Show that the coverings of this type are conal in Cov(S 1 ).
3. Let C be the sheaf of continuous real valued functions on the 1-sphere S 1 of
Exercise 2. Compute the Cech cohomology groups H q (S 1 , C), q 0.
4. Consider a topological space X with an open covering U = (Ui )iI where the
index set I is provided with a total ordering.
For any sheaf of abelian groups F
on X and integers q 0 set C q (U, F) = i0 <...<iq F(Ui0 ...iq ); the latter is a direct
summand in C q (U, F) such that the coboundary map on C q (U, F) restricts to a
coboundary map C q (U, F) - C q+1 (U, F). Show that the canonical inclusion

of complexes C (U, F) - C (U, F) induces an isomorphism on the level of
cohomology groups.
5. Flasque sheaves: Let X be a topological space. A sheaf F on X is called asque
or abby if for every inclusion of open sets U V the restriction morphism
F(V ) - F(U ) is surjective. Show for any asque sheaf of abelian groups F
on X that the Cech cohomology groups H q (U, F) are trivial for all q 1 and all
open coverings U of X. Hint: Use Exercise 4 in conjunction with the instructions
from [18], Chapter 5, Exc. 2.1.
6. Cup product: Let U = (Ui )iI be an open covering of a topological space X and
F a presheaf of rings on X. Consider C (U, F) = qN C q (U, F) as an abelian
group and dene the so-called cup product on it, via linear extension of the maps

C q (U, F) C q (U, F) - C q+q (U, F), (f, f ) - f f ,
where q, q N and
(f f )i0 ,...,iq+q = fi0 ,...,iq |Ui0 ,...,i fiq ,...,iq+q |Ui0 ,...,i
q+q q+q
for indices i0 , . . . , iq+q I. Admitting rings that are not necessarily commutative,
show that C (U, F) is a ring under the cup product and that the multiplication
induces a ring structure on the Cech cohomology H (U, F) = qN H (U, F).
q
Determine the induced ring structure on H (U, F). 0
7. Cech resolution of a sheaf : Let F be a sheaf of abelian groups on a topological

space X and U = (Ui )iI an open covering of X. For any open subset V X
denote by U V the open covering (Ui V )iI of V . Dene sheaves of abelian
groups C q (U, F), q 0, on X by associating to an open subset V X the group
of q-cochains C q (U V, F|V ). Show that the canonical morphisms
V dqV
F(V ) - C 0 (U V, F|V ), C q (U V, F|V ) - C q+1 (U V, F|V )
give rise to morphisms of sheaves of abelian groups

d0 d1 d2
0 - F - C 0 (U, F) - C 1 (U, F) - C 2 (U, F) - ...
which constitute a resolution of F in the sense that the sequence is exact. Hint:
To check the exactness at positions C q (U, F) for q > 0, show dq dq1 = 0 and
use 6.5/9 in conjunction with the maps q : C q (U, F)x - C q1 (U, F)x between
stalks at points x X that are dened as follows. Fixing a point x X, say
x Ui , start with a germ gx C q (U, F)x induced by some g C q (U V, F|V )

where V X is a suitable open neighborhood of x; we may assume V Ui .
Then let q (g) be the germ at x of the element g C q1 (U V, F|V ) that is
given by gi0 ...iq1 = gii0 ...iq1 . Show that dq1 (g )i0 ...iq = gi0 ...iq dq (g)ii0 ...iq for all
i0 , . . . , iq I.
7.7 Grothendieck Cohomology
Cech cohomology is well suited for computations. However, it lacks certain gen-
eral properties one usually expects from a cohomology theory, like the existence
of long exact cohomology sequences; see 5.4/4 or Theorem 4 below. There is
another approach to cohomology theory following Grothendieck, which does not
have such disadvantages, but at the expense of more serious diculties when
it comes to explicit computations. Fortunately, there are basic situations where
both theories yield the same cohomology groups so that the advantages of both
approaches can be used.
In the present section we will present Grothendiecks approach to cohomol-
ogy for the case of module sheaves on a scheme X. Similarly as exercised for
modules over rings in Chapter 5, the cohomology of OX -modules is dened via
derived functors. The functors we want to consider are the section functor
(X, ) : F - (X, F) = F(X),
which associates to any OX -module F the group of its global sections F(X)
and, for a morphism of schemes : X - Y , the direct image functor
: F - F
where the direct image F of an OX -module F is given by
F : V - (1 (V ), F), V Y open.
Both functors are left exact, and to dene their right derived functors we need
injective resolutions as already discussed in Section 5.3. For brevity, let us write
C for the category of OX -module sheaves.
Denition 1. An object F C is called injective if the functor Hom(, F) is

exact, i.e. if for every short exact sequence
0 - E - E - E - 0
in C the sequence
0 - Hom(E , F) - Hom(E, F) - Hom(E , F) - 0
is exact as well.
7.7 Grothendieck Cohomology 331
As Hom(, F) is left exact, the sequence

0 - Hom(E , F) - Hom(E, F) - Hom(E , F)
will always be exact, and we see that F is injective if and only if for any
monomorphism E - E and any morphism E - F, the latter admits a
(not necessarily unique) extension E - F.
Proposition 2. The category C of OX -modules on a scheme X contains enough

injectives, i.e. for each object F C there is a monomorphism F - I into
an injective object I C.
Proof. Fix an OX,x -module Ix for every x X and construct an OX -module

sheaf I on X by looking at the functor that on open subsets U X is given by

I(U ) = Ix
xU
with canonical restriction morphisms. One might think that the stalk Ix of I at
any point x X will coincide with Ix . But a careful analysis shows that there
is just a canonical map Ix - Ix and that the latter will not be injective in
general. Nevertheless, we claim for every OX -module F that the canonical map

() HomOX (F, I) - HomOX,x (Fx , Ix )
xX
associating to a morphism of OX -modules : F - I the family of induced

morphisms (x : Fx - Ix )xX where Fx is the stalk of F at x, is bijective.
First, the map () is injective since any section f I(U ) over an open subset
U X is uniquely determined by its projections to Ix for x varying over U . On
the other hand, given a family of morphisms (x : Fx - Ix )xX , the canonical
maps x

(U ) : F(U ) - Fx - Ix
xU xU
on open subsets U X yield a preimage of (x )xX in HomOX (F, I).
To settle the assertion of the proposition, x an OX -module F and choose
an injection Fx - Ix into an injective OX,x -module Ix for each x X. This
is possible, since the category of modules over a given ring contains enough
injectives; see 5.3/4. Then, if I is constructed as above, the characterization ()
of OX -morphisms with target I shows that the injections Fx - Ix , x X,
yield a morphism F - I. Clearly, the latter is a monomorphism. Further-
more, using () again, we can conclude that I is injective. Indeed, any injection
of OX -modules F - F induces a family of injections Fx - Fx between
associated stalks, and the commutative diagram
HomOX (F, I) HomOX (F , I)
-

HomOX,x (Fx , Ix ) - HomOX,x (Fx , Ix )
xX xX
shows that the upper row is surjective since the lower row is a cartesian product
of surjections.
Also note that the assertion of Proposition 2 is true for quite general cate-
gories C; cf. Grothendieck [9], Thm. 1.10.1.
Corollary 3. Any object F C admits an injective resolution, i.e. there is an

exact sequence
0 -F - I0 - I1 - ...
with injective objects I i , i = 0, 1, . . .
Recall that, just as in Section 5.4, the above exact sequence should be viewed
as a quasi-isomorphism of complexes
0 - F - 0
?
0 - I0 - I1 - I2 - ...
where the lower row is referred to as an injective resolution of F.
Proof of Corollary 3. We choose an embedding F - I 0 of F into an injective

object I 0 , an embedding I 0 /F - I 1 into an injective object I 1 , then an
embedding I 1 / im I 0 - I 2 into an injective object I 2 , and so on.
Now let us dene right derived functors of the section functor = (X, )
and of the direct image functor , the latter for a morphism of schemes
: X - Y . To apply these functors to an OX -module F, choose an injective
resolution
0 1 2
0 - I0 - I1 - I2 - . . .
of F, apply the functor to it, thereby obtaining a complex of abelian groups

(0 ) (1 ) (2 )
0 - (X, I 0 ) - (X, I 1 ) - (X, I 2 ) - ...,
and take the cohomology of this complex. Then
H q (X, F) = Rq (X, F) = ker (q )/ im (q1 )
for q 0 is called the qth cohomology group of X with values in F. Using the
technique of homotopies as in 5.1/9, one shows that these groups are indepen-
dent of the chosen injective resolution of F and that H q (X, ) = Rq (X, ) is
a functor on C; it is the so-called right derived functor of the section functor
(X, ). Note that R0 (X, ) = (X, ) since the section functor is left exact
on sheaves. For F = OX the cohomology groups H q (X, F) may be viewed as
certain invariants of the scheme X.
In the same way one proceeds with the direct image functor , which might
be viewed as a relative version of the section functor. Applying to the above
injective resolution of F we get a complex of OY -modules
0 1 2
0 - I 0 - I 1 - I 2 - ....
Furthermore, the OY -module
Rq (F) = ker q / im q1
is called the qth direct image of F where q 0. Clearly, R0 (F) equals (F)
as F is a sheaf, and one can show that Rq (F) is the sheaf associated to the
presheaf
Y V - H q (1 (V ), F|1 (V ) );
see [9], 3.7.2.

Of course, the denition of right derived functors works in much more gen-
eral situations for functors that are additive in the sense of Section 5.1. Let
us add that, just as we have explained in 5.1/12 for projective resolutions of
modules, there exist long exact cohomology sequences in our context as well.
Theorem 4. Let be a covariant additive functor on the category C of

OX -modules with values in the category of abelian groups or OX -modules. Then
every exact sequence

0 - F - F - F - 0
of objects in C gives rise to an associated long exact cohomology sequence:
R0 () R0 ()
0 - R0 (F ) - R0 (F) - R0 (F )
R1 () R1 ()
- R1 (F ) - R1 (F) - R1 (F )

- ...
Proof. We translate the proof of 5.1/12 to the dual situation where we replace
projective resolutions by injective ones; a similar situation was faced in 5.4/4.
Thus, we start out from injective cohomological resolutions
F - I , F - I
of F and F and construct a commutative diagram

.. .. ..
. . .
6 6 6
d1
0 - I 1 - I 1 I 1 - I 1 - 0
6 6 6
d0
0 - I 0 - I 0 I 0 - I 0 - 0
6 6 6

0 - F - F - F - 0
6 6 6
0 0 0
with exact rows and columns and where the left and right columns are given
by the selected resolutions of F and F . The bottom row consists of the given
exact sequence involving the modules F , F, F and the rows at positions above
the bottom row are the canonical short exact sequences associated to the direct
sums I q I q . Then the maps , d0 , d1 , . . . of the central column are constructed
to yield an injective resolution of F, relying on the injectivity of the modules
I q and I q for q 0. For this we use the argumentation given in the proof of
5.1/12 which carries over almost literally, just by passing to the dual point of
view.
In particular, the rows of the above diagram yield an exact sequence
0 - I - I - I - 0
of injective resolutions of F , F, and F . Applying the functor to it yields

the sequence
0 - (I ) - (I ) - (I ) - 0,
and the latter remains exact since is additive and, hence, compatible with
direct sums. Then, by 5.1/1, we arrive at the desired long exact cohomology
sequence.
Let us point out that the result 5.1/1, which concerns modules over a ring,
is directly applicable if takes values in the category of abelian groups, since
then we take the long cohomology sequence associated to an exact sequence of
complexes of Z-modules. On the other hand, if is a sheaf functor, we need a
version of 5.1/1 that applies to module sheaves. To obtain such a version, we
have to reprove the Snake Lemma 1.5/1, replacing arguments given in terms of
elements and their images and preimages by those involving the formation of
kernels, cokernels, images, and coimages of module morphisms. We leave this

as an exercise.
Next let us address the problem of computing derived functor cohomol-

ogy. For example, for an injective object I C we have R0 (I) = (I) and
Rq (I) = 0 for q > 0 since we can use 0 - I - 0 as an injective resolution
of I. In general, one can try to compute derived functor cohomology via Cech
cohomology. If F is any OX -module, there is always a canonical morphism
H q (X, F) - H q (X, F)
about which one knows that it is bijective for q = 0, 1 and injective for q = 2.
However, to compute higher cohomology groups via Cech cohomology, one needs
special assumptions. We state the main results without proof; for details see
Godement [7], II, 5.4 and 5.9, or Grothendieck [9], 3.8.
Theorem 5 (Leray). Let U be an open covering of a scheme X and F be an

OX -module. Assume H q (U, F) = 0 for all q > 0 and U any nite intersection
of sets in U. Then the canonical map
H q (U, F) - H q (X, F)
is bijective for all q 0.
Theorem 6 (Cartan). Let X be a scheme, F an OX -module, and S a system

of open subsets of X satisfying the following conditions:
(i) The intersection of two sets in S is in S again.
(ii) Each open covering of some open subset of X admits a renement con-
sisting of sets in S.
(iii) H q (U, F) = 0 for all q > 0 and U S.
Then the canonical homomorphism
H q (X, F) - H q (X, F)
is bijective for all q 0.
For example, let us look at an ane scheme X and let S be the system of
all basic open subsets of X. Then the conditions of Theorem 6 are satised for
the structure sheaf F = OX and for any OX -module F that is associated to an
OX (X)-module; for condition (iii), see 7.6/4. Thus, we can conclude:
Corollary 7. Let X be an ane scheme. Then
H q (X, OX ) = 0 for q > 0.
The same is true for any quasi-coherent OX -module F in place of OX .

Actually, the assertion of Corollary 7 characterizes ane schemes; this is

Serres criterion:
Theorem 8 (Serre). Let X be a quasi-compact scheme. Then the following

(i) X is ane.
(ii) H 1 (X, F) = 0 for all quasi-coherent OX -modules F.
(iii) H 1 (X, I) = 0 for all quasi-coherent ideals I OX .
In order to prepare the proof of the theorem, let us introduce the notion
Xf = {x X ; f (x) = 0}
for any global section f OX (X), assuming that the relation f (x) = 0 is read in
the residue eld k(x) of x. In particular, for any ane open subscheme U X,
the intersection U Xf equals the basic open set DU (f |U ) U where f |U does
not vanish, and we thereby see that Xf is an open subscheme in X.
Lemma 9. Let X be a scheme. Then X is ane if and only if there exist global
sections fi OX (X), i I, such that the following conditions are satised :
(i) The scheme Xfi is ane for all i I.
(ii) The fi , i I, generate the unit ideal in OX (X).
Proof. The only-if part is trivial. Therefore write A = OX (X) and assume that
there exist functions fi A, i I, satisfying conditions (i) and (ii). Now let
Y = Spec A and observe that the identity map id : A - OX (X) gives rise to
a canonical morphism
: X -Y
by 7.1/3. Considering the schemes Yi = Spec Afi as basic open subsets in Y , we

conclude from condition (ii) that the Yi cover Y . Furthermore, 1 (Yi ) = Xfi
and we see from condition (i) that the induced morphisms
i : Xfi - Yi , i I,
are morphisms of ane schemes. In particular, (OX ) is a quasi-coherent

OY -algebra via the morphism # : OY - (OX ) given by . But then we
can apply 7.1/6 to see that (OX ) is associated to an A-algebra, namely to
OX (X) = A. It follows that all i are isomorphisms and, hence, that is an
isomorphism. Thus, we are done.
Lemma 10. Let X be a quasi-compact scheme. Then the following conditions

are equivalent:
(i) X is ane.
(ii) For every exact sequence 0 - F - F - F - 0 of quasi-
coherent OX -modules, the associated sequence of global sections
0 - F (X) - F(X) - F (X) - 0
is exact.
(iii) Condition (ii) holds for all exact sequences of quasi-coherent OX -modules
0 - F - F - F - 0 where F is a submodule of a nite cartesian
product OX .
n
Proof. If X is ane, we know from Corollary 7 that the rst cohomology group
H 1 (X, F) is trivial for any quasi-coherent OX -module F on X. Furthermore, the
long cohomology sequence of Theorem 4 shows that the section functor (X, )
is exact on the category of quasi-coherent OX -modules. This establishes the
implication (i) = (ii).
Since the step from (ii) to (iii) is trivial, it remains to go from (iii) to (i). To
do this, we will use the criterion provided in Lemma 9. So assume condition (iii)
and consider a closed point x X; such a point exists by 6.6/13 if X is non-
empty. Furthermore, let U X be an open neighborhood of x. We claim that
there exists a global section f OX (X) such that x Xf U . To construct
such an f , look at the closed subset Z = X U in X and let I OX be the
ideal of all functions in OX vanishing on Z, as dened within the context of
7.3/5. Furthermore, let Jx OX be the ideal of all functions in OX vanishing
at x. Then I = I Jx is the ideal of all functions in OX vanishing on the
closed subset Z {x} in X. All these ideals are quasi-coherent by 7.3/5 and,
since x Z, the quotient
I/I = I/(I Jx ) (I + Jx )/Jx = OX /Jx
is a quasi-coherent OX -module with support {x}. Now applying (iii) to the

exact sequence 0 - I -I - OX /Jx - 0, there is a global section
f I(X) OX (X) such that f (x) = 0. Since f vanishes on Z, we have
x Xf X Z = U , as claimed. Furthermore, let us point out that we
may assume Xf to be ane, since we may take U to be ane and thereby get
Xf = DU (f |U ).
Now consider the union X of all open subschemes of type Xf X where f
varies over OX (X) under the condition that Xf be ane. Then, by the above
consideration, X is an open subscheme in X containing all closed points. How-
ever, since any non-empty closed subset in X is quasi-compact and therefore
must contain at least one closed point by 6.6/13, we see that X coincides with
many global sections fi OX (X), i I, such
X. In particular, there are nitely
that all Xfi are ane and X = iI Xfi . We claim that the fi generate the unit
ideal in OX (X) and, hence, that Lemma 9 becomes applicable. Indeed, look at
the homomorphism of OX -modules : OX I - OX given by the sections fi ,
i I. The latter is surjective, since for each x X there is an index i I such
that fi is invertible in a neighborhood of x. Hence, extends to a short exact
sequence

0 - F - OX
I - OX - 0.
But then the morphism of OX -modules (X) : OX I

(X) - OX (X) is surjective
by (iii) and, hence, the fi generate the unit ideal in OX (X), as claimed.
Now we are able to carry out the proof of Theorem 8. The implication from
(i) to (ii) follows from Corollary 7, whereas the one from (ii) to (iii) is trivial.
To derive (i) from (iii) we use the criterion given in Lemma 10 (iii). Therefore
let F be a quasi-coherent submodule of some cartesian product OX n
. Using the
canonical ltration 0 OX OX . . . OX , the intersections Fi = F OX
2 n i
,

i = 0, . . . , n, form a ltration of F by quasi-coherent OX -submodules; that
the intersection of two quasi-coherent submodules of some OX -module is quasi-
coherent again needs to be checked on ane open parts of X only and follows

from the criterion 6.8/6 (iii). Furthermore, using 6.8/11, the quotient Fi+1 /Fi
is isomorphic to a quasi-coherent OX -submodule of OX i+1 i
/OX OX or, in other
words, to a quasi-coherent ideal in OX . Then the long cohomology sequence of
Theorem 4 in conjunction with (iii) yields the exact sequence
H 1 (X, Fi ) -
H 1 (X, Fi+1 ) -
H 1 (X, Fi+1 /Fi ) = 0,
showing that H 1 (X, Fi ) = 0 implies H 1 (X, Fi+1

) = 0. Thus, proceeding by

induction, we get H (X, F ) = H (X, Fn ) = 0, which is needed for the applica-
1 1
tion of Lemma 10.
Exercises
1. Let U be an ane open covering of a separated scheme X. Show for any quasi-
coherent OX -module F that the canonical map H q (U, F) - H q (X, F) is bi-
jective for all q 0.
2. Compute the cohomology groups H q (X, OX ), q 0, for the ane line with
a double origin X = A1K and the projective line X = P1K over a eld K, as
considered in Exercise 7.6/1.
3. Let X be a scheme and F an OX -module that is asque; see Exercise 7.6/5. Show
H q (X, F) = 0 for all q > 0.
4. Let X be a separated scheme admitting an ane open covering consisting of n
members. Show for any quasi-coherent OX -module F that H q (X, F) = 0 for all
q n.
5. Let X be the ane plane A2K over a eld K with its origin removed. Compute
H 1 (X, OX ) and conclude that X cannot be ane.
6. Compute the cohomology groups H q (P2K , OP2 ), q 0, for the projective plane
K
over a eld K.
7. Let S be a separated scheme and A a quasi-coherent OS -algebra. Show that
H q (Spec A, OSpec A ) = H q (S, A) for all q 0.
8. Let X be a separated scheme and 0 - F - F - F - 0 a short

exact sequence of OX -modules where F is quasi-coherent. Show that there is an
attached long exact cohomology sequence of Cech cohomology groups:
0 - H 0 (X, F ) - H 0 (X, F) - H 0 (X, F )

- H 1 (X, F ) - H 1 (X, F) - H 1 (X, F )

- ...
Can we avoid the assumption on F to be quasi-coherent?

9. Give an example of a sheaf of OX -modules F on an ane scheme X such that
H 1 (X, F) = 0. Hint: Consider a short exact sequence of OX -modules where the
section functor (X, ) is not right exact and apply the long exact cohomology
sequence.
10. Computing cohomology via acyclic resolutions: For a sheaf of OX -modules F
on a scheme X show that the cohomology groups H q (X, F), q 0, can be
computed replacing injective resolutions by acyclic ones, i.e. using resolutions
0 - F - F0 - F1 - . . . where H q (X, F i ) = 0 for all q 1 and
all i. For example, due to Exercise 3 above we may use asque resolutions. Hint:
Adapt the method of Exercise 5.1/8 to the cohomological situation of module
sheaves. For further details see Lang [17], Thm. XX.6.2.
8. Etale and Smooth Morphisms
Smooth schemes in Algebraic Geometry may be viewed as analogues of

manifolds in Complex Analysis or Differential Geometry. Recall that
a manifold is a ringed space that looks locally like an ane space. For example,
a complex analytic manifold of dimension n is a ringed space (X, OX ) that is
locally isomorphic to open parts of the ane n-space (Cn , OCn ) where OCn is the
sheaf of analytic functions on Cn . Quite often the Inverse Function Theorem
and the Implicit Function Theorem are used in order to show that certain
subobjects of manifolds are manifolds again. As a typical example, consider a
curve C C2 given by an equation of type f (x, y) = y 2 p(x) = 0 where
p Cx is a polynomial of degree 3 having simple roots; in fact, C is a so-
called elliptic curve from which the point at innity has been removed. Due
to the assumption on the polynomial p, the gradient ( f , f ) is non-trivial at all
x y
points of C, and the Implicit Function Theorem shows that C is a submanifold
of C2 . Indeed, assuming fy
(x0 , y0 ) = 0 for a certain point (x0 , y0 ) C, there
exists a neighborhood of (x0 , y0 ) where the equation f (x, y) = 0 is equivalent
to an equation of type y = g(x) for a convergent power series g in x. It follows
that
C - C1 , (x, y) - x,
is an analytic isomorphism locally at (x0 , y0 ). A similar reasoning applies to
points (x0 , y0 ) C where f
x
(x0 , y0 ) = 0, and we thereby see that C is a man-
ifold. However, note that the Inverse and Implicit Function Theorems apply
to settings of analytic (or continuously dierentiable) functions, but not to a
strict polynomial setting. Thus, even if we start with the polynomial equation
f (x, y) = 0, we cannot expect the local solutions y = g(x) or x = h(y) to be
polynomial as well.
Now let us switch to the scheme situation and consider the above curve C
with equation f (x, y) = y 2 p(x) = 0 as a closed subscheme of the ane plane
A2C . Will C be a manifold in the sense of being locally isomorphic to the ane
line A1C ? The answer is negative and should come as no surprise, since there
are two clear indications for this. First, we have seen before that the Implicit
Function Theorem, which we could conveniently use in the analytic context, is
not available in the setting of schemes. Furthermore, the Zariski topologies on
the curve C and on the ane line A1C are rather weak in comparison to the
corresponding complex topologies. Thus, possible local isomorphisms between
C and A1C would be dened on rather large open domains and the latter seems to

342 8. Etale and Smooth Morphisms
be unthinkable from the analytic viewpoint. However, to really prove that there
cannot exist a non-empty open subscheme U C that is isomorphic to an open
subscheme of A1C is not easy. One knows that the function eld of C, namely the
eld of fractions of the ring Cx, y/(y 2 p(x)) (which is an integral domain)
is a function eld of genus 1, whereas the corresponding eld of the ane line
is the purely transcendental extension C(x) of C; the latter is of genus 0. So
we conclude that there cannot exist any local isomorphism between C and the
ane line A1C ; see [15], Exercise I.6.2 for an elementary treatment of a special
example as well as for further information. As a result, we learn that the strict
analogue of a manifold is not a useful notion in the setting of schemes.
On the other hand, we can easily observe that dierential calculus works
quite well on schemes, since for any polynomial P Rti ; i I where R is a
base ring and (ti )iI a family of variables, the formal partial derivatives P
ti
are
dened and satisfy the usual rules. Relying on this fact we dene the smoothness
of schemes via the Jacobian Condition occurring as a requisite in the Implicit
Function Theorem; see 8.5/1. To be more specic, consider a relative scheme
X over some base S. We say that X is smooth of relative dimension r if there
exists an open neighborhood U X at every point x X, together with an
S-morphism j : U - W AnS giving rise to a closed immersion of U into
some open subscheme W AnS , such that the following condition is satised:
If I OW is the sheaf of ideals corresponding to the closed immersion j,
there are n r sections gr+1 , . . . , gn in I that generate I in a neighborhood of
z = j(x) and whose Jacobian matrix satises

gj
rg (z) =nr
ti j=r+1...n
i=1...n
where t1 , . . . , tn are the coordinate functions of AnS .

Of course, the above Jacobian matrix has to be viewed as a matrix with
coecients in the residue eld k(z). Even if the assertion of the Implicit Func-
tion Theorem is not at our disposal, the denition of smooth schemes via the
Jacobian Condition is nevertheless a fully satisfying alternative for the notion
of manifolds as used in the analytic or dierentiable setting.
In order to conveniently handle the smoothness condition, it is necessary to
put the dierential calculus on schemes on a solid basis. We do this by introduc-
ing modules of dierential forms as universal objects characterizing derivations,
a procedure usually referred to as the method of Kahler dierentials. In 8.1
such dierentials are considered on the level of commutative algebra, whereas
the adaptation to sheaves follows in 8.2. Another feature of the smoothness
condition is the fact that smooth S-schemes are automatically locally of nite
presentation. This means that locally over open parts S of the base S such
a scheme can be viewed as a closed subscheme V of some ane n-space AnS
where V is dened by nitely many sections on AnS . Since we will need some
generalities on schemes of locally nite presentation also in later sections, we
have included their basics in 8.3.
8.1 Dierential Forms 343
After these preparations, the treatment of smoothness starts in 8.4 with the
study of unramied schemes, a certain pre-stage for smooth schemes, whereas
the actual theory of smoothness is explained in 8.5. The denition of smoothness
as given above suggests a natural question right away. Namely, the Jacobian
Condition being sucient for the characterization of smoothness, will it also
be necessary so that it becomes a criterion for smoothness? In other words,
consider an S-scheme X with a point x X and an open neighborhood U X
of x, on which we are given any closed immersion j : U - W Am S into some
open part W of an ane m-space Am S . Then, assuming that we know X is a
smooth S-scheme of relative dimension r, will the sheaf of ideals I dening U
as a closed subscheme in W be generated on a neighborhood of z = j(x) by
m r local sections gr+1 , . . . , gm in I satisfying the Jacobian Condition

gj
rg (z) =mr ?
ti
As it will turn out in 8.5/9 the answer is yes. But the proof of this fact is highly
non-trivial. As basic ingredient we need the characterization of smoothness via
the so-called Lifting Property 8.5/8, a fundamental result whose proof fully
justies the elaborate techniques of dierential modules as introduced in 8.1
and 8.2.
Once the characterization of smoothness via the Lifting Property is settled,
the property of a scheme to be smooth becomes quite accessible. For example,
we show in 8.5/13 that an S-scheme X is smooth of relative dimension r if and
only if, locally on open parts U X, its structural morphism X - S is
the composition of an etale morphism U - Ar and the canonical projection
S
ArS - S. Here etale means smooth of relative dimension 0. Note that the
etale morphisms are precisely those morphisms that would be locally invertible
from the viewpoint of the Invertible Function Theorem, a theorem that unfor-
tunately is not at our disposal. In any case, interpreting etale morphisms as the
scheme analogues of locally invertible morphisms, we see that smooth schemes
resemble classical manifolds, indeed, since they are etale locally isomorphic to
ane spaces.
Another useful result on smooth schemes is the Fibre Criterion 8.5/17 stat-
ing that an S-scheme X is smooth if and only if it is at and all its bers
Xs = X S k(s) over points s S are smooth. Furthermore, it is interesting
to know that the smoothness of schemes over a eld k can be characterized in
terms of geometric regularity; see 8.5/15.
8.1 Dierential Forms
In order to study unramied, etale, and smooth morphisms, we need to intro-

duce the sheaf 1X/S of relative dierential forms of degree 1 on an S-scheme
X. In the present section we start with the necessary preparations on modules
of dierential forms in the setting of commutative algebra, working over a base

ring R.
Denition 1. Let A be an R-algebra and M an A-module. An R-derivation

from A to M is an R-linear map d : A - M satisfying the so-called product
rule
d(f g) = f d(g) + gd(f )
for elements f, g A. The set of R-derivations from A to M is naturally an
A-module, which we denote by DerR (A, M ).
Remark 2. Consider an R-derivation d : A - M as above. Then d(r 1) = 0

for all r R.
Proof. Since d is R-linear, we get d(r 1) = r d(1). Furthermore, we have
d(1) = d(1 1) = 1 d(1) + 1 d(1) = d(1) + d(1)
and, thus, d(1) = 0.
Let us consider a fundamental example.
Lemma 3. Let A = RTi ; i I be the polynomial ring in a family of variables

(Ti )iI over R and M an A-module. Then, for each family (xi )iI of elements
in M , there is a unique R-derivation d : A - M such that d(Ti ) = xi for all
i I.
Proof. Since A is generated as an R-algebra by the elements Ti , i I, we see

from Denition 1 in conjunction with Remark 2 that d : A - M is uniquely
determined by the relations d(Ti ) = xi , i I. On the other hand, we can set
P
d(P ) = xi
iI
Ti
for polynomials P A and thereby dene a map d : A - M . By T P

i
we mean
the (formally built) partial derivative of P by the variable Ti , namely given by
P
= n Pn Tin1
Ti n=1

n=0 Pn Ti with coecients Pn R Tj ; j I {i}. Of
n
if we write P =
course, the above sums can contain only nitely many non-zero terms, since,
by its denition, a polynomial P RTi ; i I is a nite linear combination
of monomials in nitely many of the variables Ti . Furthermore, the product
n Pn Tin1 with a factor n N is meant in terms of the Z-algebra structure of
A.
It can be checked in an elementary way that the map d obeys the product
rule and, hence, is a derivation. For this it is enough to show that the partial
derivative by Ti satises the product rule, namely, that
(P Q) Q P
=P +Q
Ti Ti Ti
for polynomials P, Q A. To verify this, write

P = Pm Tim , Q= Qn Tin
mN nN
with coecients Pm , Qn RTj ; j I {i}. Then we have
P

Q

= m Pm Tim1 , = n Qn Tin1
Ti m=1 Ti n=1
and therefore
Q P
P +Q
Ti Ti

= Pm Tim n Qn Tin1 + Qn Tin m Pm Tim1
m=0 n=1 n=0 m=1

= n Pm Qn Tik1 + m Pm Qn Tik1
k=1 m+n=k k=1 m+n=k

(P Q)
= k Pm Qn Tik1 = .
k=1 m+n=k
Ti
Proposition and Denition 4. Let A be an R-algebra. Then there exists an

A-module 1A/R together with an R-derivation dA/R : A - 1A/R satisfying the
following universal property:
For every A-module M the canonical map
: HomA (1A/R , M ) - DerR (A, M ), - dA/R ,
is bijective; in other words, for every R-derivation d : A - M there exists a

unique A-linear map : A/R1 - M such that the diagram
dA/R
A - 1A/R
d
-
M
is commutative.
By the dening universal property, the pair (1A/R , dA/R ) is unique up to

canonical isomorphism. We call 1A/R the module of relative dierential forms
of degree 1 of A over R and dA/R the exterior dierential
& of the R-algebra A.
Furthermore, the n-fold exterior power nA/R = n 1A/R for n N is called
the module of relative dierential forms of degree n of A over R.
Proof. Given an R-derivation dA/R : A - 1 , it is easily checked that

A/R
the composition dA/R for any A-linear map : 1A/R - M yields an
R-derivation from A to M . In particular, the map is well-dened, once the
derivation dA/R has been set up. To actually construct the pair (1A/R , dA/R ),
let us rst look at the case of a free polynomial ring A = RTi ; i I where
(Ti )iI is a family of variables. Set 1A/R = A(I) and write (xi )iI for the family of
canonical free generators of A(I) . Using Lemma 3, there is a unique R-derivation
- 1
dA/R : A A/R such that dA/R (Ti ) = xi for all i I, and we claim that
1
(A/R , dA/R ) enjoys the stated universal property. Indeed, starting out from any
R-derivation d : A - M , we can look at the A-linear map : 1 -M
A/R
given by xi - d(Ti ), i I. Then d = dA/R by the product rule and
Remark 2. Moreover, since 1A/R is generated as an A-module by the elements
dA/R (Ti ), i I, we see that is uniquely determined by the relation d = dA/R .
Therefore we can conclude that, indeed, (1A/R , dA/R ) is the module of relative
dierential forms of A over R.
In the general case, we can interpret A as a quotient of a free polynomial
ring, and it is enough to prove the following assertion:
Lemma 5. Let A be an R-algebra such that the associated module of dierential

forms (1A/R , dA/R ) exists. Then, if A - B is a surjection of R-algebras with
kernel a A, the dierential dA/R induces canonically a commutative diagram
dA/R
A - 1A/R
? d ?
B
B/R
- 1A/R / a 1A/R + A dA/R (a)
where dB/R is an R-derivation of B satisfying the universal property of Propo-
sition 4. In particular,

1B/R = 1A/R / a 1A/R + A dA/R (a)
and (1B/R , dB/R ) is the module of relative dierential forms of B over R.

Proof. First note that 1A/R / a 1A/R + A dA/R (a) is canonically a B-module.
Since
dA/R
- 1 - 1 / a 1
A A/R A/R A/R + A dA/R (a)
is an R-derivation, whose kernel contains the ideal a A, we see immediately

that it factors through an R-derivation
dB/R
B - 1A/R / a 1A/R + A dA/R (a) .
The latter satises the required universal property, as is easily veried by relying
on the universal property of dA/R .
In particular, the explicit construction of the module of dierential forms

in the above proofs shows:
Corollary 6. Let A be an R-algebra that is generated by a family (ti )iI of

elements in A. Then the A-module of dierential forms 1A/R is generated by
the elements dA/R (ti ).
If the ti are variables and A is the free polynomial ring Rti ; i I, then
the dierential forms dA/R (ti ), i I, form a free generating system of 1A/R .
There is another approach to modules of dierential forms, which we want

to present now. Starting out from an R-algebra, consider the multiplication map
: A R A - A, xy - xy,
and set J = ker . Then gives rise to an isomorphism (A R A)/J - A,
whose inverse is induced from any of the two canonical maps
1 : A - A R A, x - x 1,
2 : A - A R A, x - 1 x.
The quotient J/J2 is an (A R A)-module and even an (A R A)/J-module.
Thus, by the above isomorphism, J/J2 may be viewed as an A-module as well
and we can think of the A-module structure of J/J2 as being obtained from its
(A R A)-module structure by restriction of scalars via either of the morphisms
1 , 2 : A - A R A. We claim that the map
A - J, x - 1 x x 1,
induces an R-derivation : A - J/J2 and that (J/J2 , ) is the module of
relative dierential forms of A over R. To justify this, some preparations are
necessary.
Lemma 7. For an R-algebra A let J be the kernel of the multiplication map

: A R A - A.
(i) View A R A as an A-module under the morphism 1 : A - A R A,
x - x 1, and J A R A as a submodule. Then the elements
1 t t 1, t A,
generate J as an A-module.
(ii) If A as an R-algebra is generated by the family (ti )iI , then the elements
1 ti ti 1, i I,
generate the ideal J A R A.
Proof. Clearly, all elements of type 1 t t 1 for t A belong to J. Moreover,

we can write
x y = xy 1 + (x 1)(1 y y 1)
n
for x, y A. Now assume
n that i=1 xi yi for some elements xi , yi A belongs
to J. Then we have i=1 xi yi = 0 and therefore

n
n
xi y i = (xi 1)(1 yi yi 1).
i=1 i=1
In particular, the elements of type 1 t t 1 for t A generate J as an

A-module, as claimed in (i). Furthermore, writing t = xy for elements x, y A,
we obtain
(1 t t 1) = (x 1)(1 y y 1) + (1 x x 1)(1 y),
from which one derives assertion (ii) by means of recursion.
Proposition 8. Let f0 : A - B be a morphism of R-algebras and J B an

ideal such that J2 = 0. View J as an A-module under f0 .
(i) If d : A - J is an R-derivation, then f1 = f0 + d denes a morphism
of R-algebras A - B satisfying f1 f0 mod J.
(ii) If f1 : A - B is a morphism of R-algebras satisfying f1 f0 mod J,
then f1 f0 yields an R-derivation A - J.
Thus, the correspondence f - f f0 sets up a bijection between the set
of R-algebra morphisms f : A - B satisfying f f0 mod J and the set
DerR (A, J) of all R-derivations from A to J.
Proof. An R-linear map f1 : A - B is a morphism of R-algebras if and only if

it satises f1 (1) = 1 and is multiplicative in the sense that f1 (xy) = f1 (x)f1 (y)
for all x, y A. However, if f1 f0 mod J, the equation f1 (1) = 1 follows from
the multiplicativity of f1 . Indeed, using the geometric series in conjunction with
J2 = 0, the congruence f1 (1) f0 (1) = 1 mod J shows that f1 (1) is a unit in B.
The latter must be idempotent and, thus, coincides with 1 if f1 is multiplicative.
Therefore it remains to show that an R-linear map f1 : A - B satisfying
f1 f0 mod J is multiplicative if and only if the dierence d = f1 f0 obeys
the product rule. To check this, let x, y A. Then
f1 (x)f1 (y) = (f0 (x) + d(x))(f0 (y) + d(y)) = f0 (xy) + f0 (x)d(y) + f0 (y)d(x),
since d(x)d(y) J2 = 0. Thus, f1 is multiplicative if and only if
f0 (xy) + f0 (x)d(y) + f0 (y)d(x) = f0 (xy) + d(xy) = f1 (xy)
for all x, y A, i.e. if and only if f0 (x)d(y) + f0 (y)d(x) = d(xy). However, the
latter is equivalent to the fact that d satises the product rule.
Corollary 9. For an R-algebra A let J A R A be the kernel of the multipli-

cation map : A R A - A. Then the map
A - J, x - 1 x x 1,
gives rise to an R-derivation : A - J/J2 .
Proof. We look at the R-algebra morphisms
f 0 , f1 : A - (A R A)/J2
that are induced from the maps
x - x 1, x - 1 x.
Applying Proposition 8 to the ideal J/J2 (A R A)/J2 , we see that the

dierence = f1 f0 is an R-derivation.
Proposition 10. In the situation of Corollary 9 the module of relative dier-

ential forms of degree 1 of A over R is given by the pair (J/J2 , ).
Proof. For an A-module M and an R-derivation d : A - M consider the

R-linear map
: A R A - M, x y - xd(y),
which is even A-linear if we view A R A as an A-module under the morphism
1 : A - A R A given by x - x 1. Then, since d is a derivation, we
obtain for x, y A

(1 x x 1)(1 y y 1)
= (1 xy) (y x) (x y) + (xy 1)
= d(xy) yd(x) xd(y) = 0.
Now observe from Lemma 7 (i) that J2 , as an A-module via 1 : A - A R A,

is generated by all products of type
(1 x x 1)(1 y y 1), x, y A.
Therefore, using the above computation, is trivial on J2 and, hence, induces

by restriction to J an A-linear map : J/J2 - M such that the diagram

A - J/J2
d
-
M
is commutative. Since J/J is generated by the elements (x) for x A, as
2
follows from Lemma 7 (i) again, we see that is unique as an A-linear map
satisfying d = . This shows that, indeed, (J/J2 , ) is the module of relative

dierential forms of degree 1 of A over R.
One can easily see as a corollary that 1A/R must be trivial if the struc-
tural morphism : R - A is a categorical epimorphism, in the sense that
for ring morphisms 1 , 2 : A - B an equality 1 = 2 implies always
1 = 2 . For example, surjective morphisms and localization morphisms are of
this type. Due to the denition of tensor products, the two canonical morphisms
1 , 2 : A - A R A coincide, when composed with : R - A. But then, if
is a categorical epimorphism, we get 1 = 2 and therefore 1A/R = J/J2 = 0
by Lemma 7 (i).
Next we discuss some functorial properties of modules of dierential forms,
starting with the base change functor.
Proposition 11. Let A be an R-algebra and R - R a morphism of rings.

Set A = A R R . Then the exterior dierential dA/R : A - 1A/R induces via

tensoring with R over R an R -derivation
dA /R : A - 1A/R A A ,
and (1A/R A A , dA /R ) is the module of dierential forms of A over R . In

particular, 1A /R 1A/R A A .
Proof. Using the canonical isomorphism
1A/R R R - 1 A (A R R ) = 1 A A
A/R A/R
of 4.3/2, it can be checked by direct computation that dA/R R R is an

R -derivation satisfying the required universal property.
Alternatively, we can base our argument on Proposition 10. To do this, we
look at the canonical exact sequence
0 - J - A R A - A - 0,
which splits as a sequence of R-modules, since the map A - A R A,

x - x 1, may serve as a section of the multiplication map A R A - A.
Therefore the sequence remains exact when applying the tensor product with
R over R. Now view the tensored sequence as the rst row of the canonical
commutative diagram
0 - J R R - A R A R R - A R R - 0
? ? ?
0 - J - A R A - A - 0
where the lower row is the exact sequence associated to the multiplication map
A R A - A . Since the canonical vertical maps in the middle and on
the right are isomorphisms, the same must be true for the vertical map on the
left, and we can conclude that the canonical map J2 R R - J2 is at least
surjective. Thus, tensoring the exact sequence
0 - J2 - J - J/J2 - 0
with R over R, yields a commutative diagram
J2 R R - J R R - J/J2 R R - 0
? ? ?
0 - J2 - J - J /J2 - 0
with exact rows where the left vertical map is surjective and the middle vertical
map is an isomorphism. But then the right vertical map J/J2 R R - J /J2
is an isomorphism as well, as follows by diagram chase or from the Snake
Lemma 1.5/1. Now it can be checked directly from its denition in Corollary 9
that the exterior dierential : A - J/J2 carries over via tensoring with R
over R to the exterior dierential : A
- J /J2 .
Proposition 12. Let f : A - B be a morphism of R-algebras. Then there

is a canonical sequence of B-modules
1A/R A B - 1B/R - 1B/A - 0
and the latter is exact.
Proof. To introduce the morphisms of the sequence, look at the commutative

diagram
dA/R
A - 1A/R
? ?
dB/R
B - 1B/R
where is the unique A-linear map provided by the universal property of the
exterior dierential dA/R . For this to work well, note that dB/R f : A - 1B/R
is an R-derivation from A to 1B/R , the latter being viewed as an A-module via
f: A - B. Since 1
B/R is a B-module as well, we obtain from a B-linear
map A/R A B
1 - 1B/R . Furthermore, the diagram
dB/R
B - 1B/R
dB/A - ?
1B/A
shows the existence of a canonical B-linear map 1B/R - 1B/A so that the
canonical maps of the sequence
() 1A/R A B - 1B/R - 1B/A - 0
are clear.
To show the exactness assertion, we use the fact based on the left exactness
of the functor Hom that () is exact if and only if, for all B-modules M , the
corresponding sequence
0 - HomB (1B/A , M ) - HomB (1B/R , M ) - HomB (1A/R A B, M )
is exact. Since HomB (1A/R A B, M ) = HomA (1A/R , M ) (see Exercise 4.3/8),

we can use Proposition 4 to write the latter sequence as
u v
() 0 - DerA (B, M ) - DerR (B, M ) - DerR (A, M )
where u is the forgetful map interpreting any A-derivation B - M as an

R-derivation and v the composition map of R-derivations B - M with
f : A - B. Then, clearly, u is injective and v u = 0 by Remark 2.
To show that the sequence () is exact, it remains to check the inclusion
ker v im u. To do this, let d : B - M be an R-derivation belonging to
the kernel of v and, thus, satisfying d f = 0. Then d is even an A-derivation,
because for a A and x B the product rule yields
d(a x) = f (a) d(x) + x (d f )(a) = a d(x).
This shows d im u, as required.
If the morphism f : A - B is surjective, we have 1B/A = 0 by Remark 2

and the exact sequence of Proposition 12 can be continued to the left:
Proposition 13. Let f : A - B be a surjection of R-algebras with kernel

a A. Then there is a canonical sequence of B-modules
a/a2 - 1A/R A B - 1B/R - 0
and the latter is exact.
Proof. The exterior dierential dA/R : A - 1A/R yields via restriction to a an

R-linear map a - 1A/R , sending a2 into a 1A/R by the product rule. Thus,
dA/R induces a map a/a2 - 1 /a 1
A/R A/R and we see that the latter is A-
or (A/a)-linear, since for x A and a a we have
dA/R (xa) = xdA/R (a) + adA/R (x) xdA/R (a) + a 1A/R .
Now using the surjectivity of f : A - B, we can identify 1 /a1

A/R A/R with
A/R A B, thereby obtaining a B-linear map a/a2 - 1A/R A B. Combining
1
the latter with the canonical map 1A/R A B - 1B/R from Proposition 12,
we want to show that the resulting sequence
a/a2 - 1A/R A B - 1B/R - 0,
is exact.
Indeed, f being surjective implies 1B/A = 0, as we have already pointed
out. Hence, the map 1A/R A B - 1B/R is surjective by Proposition 12. On
the other hand, this map is already known from Lemma 5. Namely, it coincides
with the canonical surjection

1A/R /a 1A/R - 1 / a 1
A/R A/R + A dA/R (a) ,
whose kernel is generated by the image of dA/R (a) in 1A/R /a1A/R . Consequently,
the sequence mentioned in the assertion is exact.
Proposition 14. Let A and B be two R-algebras. Then there exists a canonical
isomorphism of (A R B)-modules
(1A/R R B) (A R 1B/R ) - 1
AR B/R .
Furthermore, if
1 : A - 1A/R , 2 : B - 1B/R
are the exterior dierentials of A and B, the exterior dierential of A R B
corresponds to the R-linear map
: A R B - (1A/R R B) (A R 1B/R )
given by
xy - 1 (x) y x 2 (y).
Proof. Clearly, as dened in the assertion is R-linear and it is easily checked

that it is even an R-derivation. Namely, for x , x A and y , y B we have:

(x y ) (x y )
= (x x y y )

= 1 (x x ) (y y ) (x x ) 2 (y y )

= x 1 (x ) + x 1 (x ) (y y ) (x x ) y 2 (y ) + y 2 (y )

= (x y ) 1 (x ) y + (x y ) 1 (x ) y

(x y ) x 2 (y ) + (x y ) x 2 (y )

= (x y ) (x y ) + (x y ) (x y )
To show that is the exterior dierential of A R B, consider an R-derivation

d : A R B - M from A R B to some (A R B)-module M . Let
1 : A - A R B, 2 : B - A R B
be the canonical morphisms and write M/A for the A-module obtained from
M by restriction of scalars via 1 . In the same way, let M/B be obtained by
restriction of scalars via 2 . Then
d1 = d 1 : A - M/A , d 2 = d 2 : B - M/B
are R-derivations and the corresponding A-, resp. B-linear maps
1A/R - M/A , 1B/R - M/B
induce (A R B)-linear maps
1 : 1A/R R B - M, 2 : A R 1B/R - M
and, thus, an (A R B)-linear map
= (1 , 2 ) : (1A/R R B) (A R 1B/R ) - M
such that for x , x A and y , y B we have
1 (x ) y x 2 (y ) - d1 (x ) (1 y ) + (x 1) d2 (y ).
Using the product rule for d, one easily checks that = d. Furthermore,
since (1A/R R B) (A R 1B/R ), as an (A R B)-module, is generated by
(A R B), it follows that is uniquely determined by the relation = d
and we see that, indeed, is the exterior dierential of A R B.
Exercises
1. For a eld K, consider the coordinate ring A = Kt1 , t2 /(t22 t31 ) of Neiles
parabola, as in Exercise 6.2/6. Show that the A-module of relative dierential
forms 1A/K can be generated by two elements, but that it is not free. Conclude
once more that the scheme Spec A cannot be isomorphic to the ane line A1K .
2. Let A be an R-algebra, S A a multiplicatively closed system, and AS the cor-
responding localization of A. Show by elementary computation that the module
of relative dierential forms 1AS /R is given by the localization (1A/R )S and that
the map
f sdA/R (f ) f dA/R (s)

d : AS - (1 )S , - ,
A/R
s s2
serves as the exterior dierential of AS over R where dA/R : A - 1A/R denotes
the exterior dierential of A. Conclude that the exact sequence of Proposition 12
reduces to an isomorphism 1A/R A AS - 1AS /R and, in particular, that
1AS /A = 0.
3. Consider a local K-algebra A with maximal ideal m where K is a eld such
that the projection A - A/m restricts to an isomorphism K - A/m.
Show that the exact sequence of Proposition 13 reduces to an isomorphism
m/m2 - 1A/K A K. Hint: Use the direct sum decomposition A = K m.
Modules of relative dierential forms 1L/K for eld extensions L/K:
4. Determine 1L/K for a simple algebraic extension L = K(x) where x is separable,

resp. purely inseparable over K.
5. If L/K is separable algebraic, then 1L/K = 0. Hint: Reduce to the case of nite
separable extensions.
6. If L/K is generated by n elements, say L = K(x1 , . . . , xn ), then 1L/K is generated
over L by the n elements dL/K (x1 ), . . . , dL/K (xn ). Hint: Observe Exercise 2.
7. Let E be an intermediate eld of the extension L/K and (xj )jJ a family of
elements in L such that L = E(xj ; j J). For variables Xj , j J, look at the
canonical morphism of E-algebras : EXj ; j J - L given by Xj - xj
and let (fi )iI be a family of polynomials generating the ideal ker . Show that
the following assertions are equivalent for a given K-derivation : E - V into
some L-vector space V and for a family (vj )jJ of elements in V :
(a) There exists a K-derivation : L - V extending that satises (xj ) = vj
for all j J.
(b) For any polynomial f KXj ; j J write f V Xj ; j J for the
polynomial obtained by transporting coecients with . Then the equations
fi
fi (x) + (x) vj = 0, i I,
Xj
jJ
hold for the tuple x = (xj )jJ .

Moreover, if there exists an extension as in (a), it is unique.
8. Let E be an intermediate eld of the extension L/K. Show that the canonical
map 1E/K E L - 1
L/K of Proposition 12 is injective if and only if every
K-derivation E - L admits an extension as a K-derivation L - L.
9. For a nitely generated eld extension L/K, show that 1L/K = 0 is equivalent to
L/K being separable algebraic. Hint: If 1L/K = 0, choose an intermediate eld
K E L such that E/K is purely transcendental and L/E is algebraic. If
L/E is separable, use Exercises 7 and 8 to show that the extension E/K must
be trivial. Otherwise, derive a contradiction if there should exist an intermediate
eld K E L such that L/E is a non-trivial simple extension, which is purely
inseparable.
10. Show that there exist examples of algebraic extensions L/K that are not separa-
ble, but nevertheless satisfy 1L/K = 0.
11. Assume that L/K is generated by n elements, say L = K(x1 , . . . , xn ). Show that
transgradK L dimL 1L/K n and, moreover, that transgradK L = dimL 1L/K
holds if and only if L/K is separably generated, i.e. if and only if L is separa-
ble algebraic over a purely transcendental extension of K. Hint: After a suitable
renumbering of the elements xi there exists an integer r n such that the dieren-
tial forms dL/K (x1 ), . . . , dL/K (xr ) are a basis of 1L/K . Setting E = K(x1 , . . . , xr )
show 1L/E = 0 and conclude from Exercise 9 that the extension L/E is separable
algebraic.
8.2 Sheaves of Dierential Forms
Now we want to generalize the construction of modules of dierential forms, as

presented in Section 8.1, from rings to schemes. To do this, let us consider a base
scheme S and an S-scheme : X - S. Furthermore, let : X - X S X
be the diagonal embedding, which as we know from 7.4/1, is a locally closed
immersion. In particular, there exists an open subscheme W X S X such
that (X) W and the induced morphism X - W is a closed immersion;
for simplicity the latter will be denoted by again. Let J OW be the
associated quasi-coherent ideal dening (X) as a closed subscheme of W .
Then the OX -module
1X/S = (J /J 2 )
is called the sheaf of relative dierential forms of degree 1 of X over S. Note that
the explicit construction of the inverse image sheaf in 6.9/2 and 6.9/3 shows
that 1X/S is well-dened, independent of the choice of W . For example, we can
assume that W is the largest open subscheme in X S X such that (X) is
closed in W ; cf. 7.3/12. Since J is quasi-coherent, the same is true for J /J 2 ,
and it follows from 6.9/5 that 1X/S is quasi-coherent as well.
To give a description of 1X/S on local ane open parts of X, consider ane
open subschemes V = Spec R S and U = Spec A X such that (U ) V .
Then the morphism U - V induced from corresponds to a ring morphism
R - A equipping A with the structure of an R-algebra. Let J A R A be
the kernel of the multiplication map A R A - A. Then J |U U = J and
V
using 6.9/4 in conjunction with 8.1/10 we obtain isomorphisms
(J /J 2 )|U (J/J2 AR A A) J/J

2 1
A/R .
Note that the tensor product is meant as the coecient extension of J/J2
with respect to the multiplication map A R A - A and that the result-
ing A-module coincides with J/J2 , due to the isomorphism (AR A)/J - A.
Furthermore, we see from 8.1/10 that the exterior dierential dA/R : A - 1A/R
is induced from the morphism of OS -modules
dX/S : OX - 1X/S ,
mapping a section f of OX to the residue class of p2 (f ) p1 (f ) in J /J 2 where

p1 , p2 : X S X - X are the projections onto the factors X. Also in this case
dX/S is called the exterior dierential of X over S. Let us add that, just as in
the case of rings, the sheaf nX/S of relative dierential forms of degree n of X
&
over S, for some n N, is dened as the nth exterior power n 1X/S .
Proposition 1. Let : X - S be an S-scheme. Then, as a quasi-coherent

OX -module, the sheaf of dierential forms 1X/S together with its exterior dif-
ferential dX/S : OX - 1X/S is uniquely characterized up to canonical isomor-
phism by the following universal property:
8.2 Sheaves of Dierential Forms 357
Given ane open subschemes U = Spec A X and V = Spec R S

satisfying (U ) V , there is a unique isomorphism 1X/S |U 1
A/R such
that the restriction of dX/S to U is induced from the exterior dierential
dA/R : A - 1A/R .
Proof. Taking into account the universal property of the modules of dierential
forms 1A/R , it is clear from the above consideration that the sheaf of dieren-
tial forms 1X/S admits the stated property. The uniqueness assertion for 1X/S
follows then by means of a gluing argument.
Applying the above result to the ane n-space X = AnS and choosing a
set of global coordinate functions t1 , . . . , tn on X, we can conclude from 8.1/6
that in this case the OX -module of dierential forms 1X/S is free of rank n,
namely isomorphic to OX n
, and admits the sections dX/S (ti ), i = 1, . . . , n, as
free generators.
Corollary 2. Let A be an algebra over a ring R and dA/R : A - 1

A/R the
exterior dierential of A. Then, for any localization Af of A by some element
f A, there is a canonical isomorphism
1Af /R 1A/R A Af
such that the exterior dierential of Af is given by the map
-
a -
f r dA/R (a) adA/R (f r )
dAf /R : Af 1A/R A Af , .
fr f 2r
Proof. The isomorphism 1Af /R 1A/R A Af is obtained from Proposition 1,

together with the commutative diagram
dA/R
A - 1A/R
? dA /R ?
Af
f
- 1A/R A Af
where the horizontal maps are the exterior dierentials of A and Af . Then,
using d = dAf /R as an abbreviation, we get for a A and r N the equation
a a a
d(a) = d f r
= d(f r
) + f r
d
fr fr fr
and, thus,
a f r d(a) a d(f r )
d = ,
fr f 2r
as claimed.
The assertion of Corollary 2 can just as well be obtained by checking via

elementary computation that the above map dAf /R is, indeed, an exterior dier-
ential of Af ; see Exercise 8.1/2. Then the module of dierential forms 1X/S of a
relative scheme : X - S can be obtained by gluing quasi-coherent modules

A/R where U = Spec A X and V = Spec R S vary over all ane
1
of type
open subschemes of X and S such that (U ) V . Proceeding like this, it is
not necessary to consider the diagonal morphism X/S : X - X S X, and
the results 8.1/7 up to 8.1/10 become dispensable.
Finally, we want to indicate how to generalize the results 8.1/11 up to
8.1/14 to sheaves of dierential forms on schemes. As a general observation, let
us point out that any morphism of S-schemes f : Y - X induces a canonical
morphism f 1X/S - 1 of OY -modules. In the case of ane schemes, say
Y /S
X = Spec A, Y = Spec B, and S = Spec R, the latter corresponds to the mor-
phism of B-modules 1A/R A B - 1B/R given by the canonical commutative
diagram
dA/R
A - 1A/R
f
? ?
dB/R
B - 1B/R .
This way every dierential form 1A/R induces an element 1 1A/R A B
and, taking its image, a dierential form in 1B/R . The same is true for sheaves of
dierential forms. Every section of 1X/S gives rise to a section in f (1X/S )
and, taking its image under the canonical morphism f 1X/S - 1 , to a
Y /S
section of 1Y /S . It is common practice to call , just as well as , the pull-
back of , using the notion f () for both quantities. Of course, if ambiguities
are possible, one has to be careful about specifying the intended type of pull-
back.
Now, xing a base scheme S, it is straightforward how to generalize the
results 8.1/11 up to 8.1/14 by means of gluing techniques to the scheme case:
Proposition 3. For an S-scheme X and a base change morphism S -S

consider the resulting S -scheme X = X S S , as well as the attached projection
p : X - X. Then the canonical morphism
p 1X/S - 1X /S
is an isomorphism.
Proposition 4. Let f : Y - X be a morphism of S-schemes. Then there is

a canonical exact sequence of OY -modules
f 1X/S - 1Y /S - 1Y /X - 0.
Proposition 5. Let X be an S-scheme and j : Y - X a closed subscheme,

given by a quasi-coherent ideal I OX . Then there is a canonical exact sequence
8.2 Sheaves of Dierential Forms 359
of OY -modules
I/I 2 - j 1 - 1Y /S - 0
X/S
where I/I 2 is interpreted as the inverse image j (I/I 2 ).
Proposition 6. Let X1 and X2 be S-schemes and write pi : X1 S X2 - Xi ,

i = 1, 2, for the projections onto the factors of the ber product of X1 and X2 .
Then the canonical morphism of OX1 S X2 -modules
p1 1X1 /S p2 1X2 /S - 1X1 S X2 /S
is an isomorphism.
Exercises
1. Consider a monomorphism of schemes S - S, for example, a locally closed

immersion; see Exercise 7.3/4. Show for any S -scheme X that the canonical
morphism of OX -modules 1X/S - 1 is an isomorphism. Hint: Use Exer-
X/S
cise 7.4/1.
2. Consider the ane line with a double origin X = A1K over a eld K, as described
in the introduction to Chapter 7. Compute the sheaf of dierential forms 1X/K .
3. Consider the projective n-space X = PnS over an ane base scheme S.
(a) Show that the sheaf of dierential forms 1X/S is locally free of rank n in
the sense that each point x X admits an open neighborhood U X where
1X/S |U OUn .
(b) Compute 1X/S for the projective 1-space X = P1S . Show that this module
does not admit non-zero global sections and, hence, that it cannot be free (if S is
non-trivial). Hint: Let S = Spec R. As in Section 7.1, construct P1R by canonically
gluing the ane lines X0 = Spec R tt10 and X1 = Spec R tt01 for variables t0 , t1 .
Furthermore, dene a quasi-coherent OX -module OX (2) (the so-called Serre
twist with index 2, as to be introduced in Section 9.2) by gluing OX0 , the free
module generated by s0 = 1 on X0 , with OX1 , the free module generated by s1 = 1
on X1 , via the equation s1 = ( tt01 )2 s0 on X0 X1 . Show that 1X/S OX (2).
4. Consider the diagonal embedding : X - X S X for a relative scheme X over
some base scheme S. Let W be an open subscheme of X S X such that factors
through a closed immersion X - W and let J OW be the corresponding
quasi-coherent ideal. Show:
(a) 1X/S = 0 is equivalent to J = J 2 .
(b) Assuming that J is locally of nite type, the stalk 1X/S,x at some point x X
vanishes if and only if is a local isomorphism at x, i.e. if and only if there is an
open neighborhood U X of x such that restricts to an isomorphism between
U and an open subscheme of X S X.
For example, the assumption on J to be locally of nite type is fullled, due to
8.1/7 (ii), if X is locally of nite type over S in the sense of 8.3/4.
5. Invariant dierential forms on group schemes: Let G be an S-group scheme over

a base scheme S and : G S G - G the morphism dening its group law, as
well as : S - G the unit section and i : G - G the formation of inverses;
see Section 9.6 for the notion of group schemes. For any S-morphism g : T - G
(i.e. a T -valued point of G with T viewed as a test scheme), we obtain from
G a T -group scheme GT by applying the structural morphism T - S as base
change. Then we can consider the left translation by g on G, or better GT , namely
the morphism
gT id
g : G T - T T GT - GT T GT T
- GT
where T as an index always means base change with respect to T - S; note

that g is an isomorphism.
We say that a global dierential form 1G/S (G) is invariant under left trans-
lation with g if the pull-back of to GT , again denoted by , satises g =
using the identication g 1GT /T 1GT /T furnished by the isomorphism g . Fur-
thermore, we say that is left-invariant if it is invariant under the left translation
by every T -valued point g of G for T varying over all S-schemes. In the same
way, one denes right-invariant dierential forms using right translations on G.
(a) Take for G the additive group Ga,S and the multiplicative group Gm,S intro-
duced in Exercise 7.1/1 and specify an invariant dierential form on G generating
1G/S as a free OG -module.
(b) Passing to S-group schemes of general type, show for every 0 (S, 1G/S )
that there is a unique left-invariant dierential form 1G/S (G) satisfying
= 0 in 1G/S .
(c) Conclude that by extending sections in 1G/S to left-invariant sections of
1G/S we get a canonical isomorphism p 1G/S 1G/S where p : G - S is
the structural morphism of G.
Hints: To establish the uniqueness in (b), use the so-called universal point given
by T = G and the identity morphism T - G. For the existence part in (b)
reduce to the case where 0 is induced by a dierential form 1G/S dened in
a neighborhood of the image of the unit section . Pull back with respect to the

multiplication morphism GS G - G and write = 1 2 according to the
decomposition GS G/S = p1 G/S p2 1G/S of Proposition 6. Then dene as
1 1
(i,id)
the pull-back of 2 with respect to the twisted diagonal morphism G - GS G.
If necessary, consult [5], 4.2, for further details.
8.3 Morphisms of Finite Type and of Finite Presentation
Before discussing unramied, etale, and smooth morphisms we introduce cer-

tain niteness conditions that are satised by such morphisms. On the side of
algebras these are similar to the niteness conditions introduced for modules in
1.5/3.
8.3 Morphisms of Finite Type and of Finite Presentation 361
Denition 1. Let A be an algebra over some ring R and : R - A the

corresponding structural morphism. We call A, as well as , of nite type if
there exists an exact sequence

0 - a - Rt1 , . . . , tn - A - 0
where Rt1 , . . . , tn is a polynomial ring in nitely many variables t1 , . . . , tn over

R and where is a surjection of R-algebras. If, in addition, can be chosen
in such a way that the ideal a = ker is nitely generated, then A and are
called of nite presentation.
For example, we see from 1.2/9 that any localization map
R - Ra1 = Rt/(1 at)
of a ring R into its localization by some element a R is of nite type and even
of nite presentation. As a further observation note:
Lemma 2. Let A be an R-algebra and

0 - a - Rt1 , . . . , tm - A - 0
an exact sequence as considered in Denition 1. Then, if A is of nite presen-

tation, the ideal a = ker is nitely generated.
Proof. If A is of nite presentation, there exists a surjection of R-algebras

: Rt1 , . . . , tn - A where a = ker is nitely generated. Then the mul-
tiplication map
: Rt1 , . . . , tm R Rt1 , . . . , tn - A, pq - (p) (q),
is surjective and we claim that it has a nitely generated kernel. Indeed, look
at the nitely generated ideal Rt1 , . . . , tm R a ker . Dividing it out and
using the right exactness of tensor products 4.2/1, we arrive at a surjection
Rt1 , . . . , tm R A = At1 , . . . , tm - A,
whose kernel is generated by the elements ti 1 1 (ti ), i = 1, . . . , m, as is

easily veried. Therefore ker is nitely generated by 1.5/5 (ii).
Now look at a commutative diagram of type

0 - ker - Rt1 , . . . , tm R Rt1 , . . . , tn - A

- 0

? ?

0 - a - Rt1 , . . . , tm - A - 0
where the middle vertical map sends ti 1 to ti for i = 1, . . . , m and 1 tj
to a -preimage of (tj ) for j = 1, . . . , n. Of course, is surjective. Therefore
the left vertical map ker - a must be surjective as well. But then the
ideal a Rt1 , . . . , tm is nitely generated, since the ideal ker is nitely
generated in Rt1 , . . . , tm R Rt1 , . . . , tn .
Furthermore, we will need some standard facts:
Lemma 3. (i) Ring morphisms of nite type are stable under base change, i.e.
if R - A is a ring morphism of nite type, then for any ring morphism
R - R the induced morphism R - A R R is of nite type. The same
holds for ring morphisms of nite presentation.
(ii) The composition of ring morphisms of nite type is of nite type again.
The same holds for ring morphisms of nite presentation.
Proof. Assertion (i) is a consequence of the right exactness of tensor products

4.2/1. To establish (ii), look at the composition R - A - B of two ring
morphisms and choose surjections : RT - A and : AT - B,

where T and T are systems of variables. Then we can consider the composition
of surjections
id
: RT R RT - A R RT = AT - B.
Now if and are of nite type, we can suppose that T and T are nite
systems of variables and we see that : R - B is of nite type. If and
are of nite presentation, we can suppose that, in addition, ker RT and
ker AT are nitely generated. As ker RT R RT is generated
by (ker ) 1 and ( id)-preimages of generators of ker , it follows that ker
is nitely generated and, hence, that is of nite presentation.
Denition 4. A morphism of schemes f : X - Y is called locally of nite

type (resp. locally of nite presentation) at a point x X if the following holds:
There exist ane open subschemes U X and V Y , say U = Spec A
and V = Spec B, such that x U and f (U ) V , and such that the induced
morphism of rings B - A is of nite type (resp. nite presentation).
We say that f is locally of nite type (resp. locally of nite presentation)
if the corresponding property holds at all points x X.
For example, any closed immersion is locally of nite type and any open
immersion is locally of nite presentation. Furthermore, Lemma 3 shows that
morphisms of locally nite type (resp. locally nite presentation) are preserved
under base change and under composition.
Let us add along the way that a morphism of schemes f is said to be of
nite type if f is locally of nite type and quasi-compact. Moreover, f is said to
be of nite presentation if f is locally of nite presentation, quasi-compact, and
quasi-separated. As usual, we would like to show:
8.3 Morphisms of Finite Type and of Finite Presentation 363
Proposition 5. The following conditions are equivalent for a morphism of

ane schemes f : Spec A - Spec B:
(i) f is locally of nite type (resp. locally of nite presentation).
(ii) The morphism of rings B - A associated to f is of nite type (resp.
of nite presentation).
Proof. Let us write X = Spec A and Y = Spec B. We have only to show that
condition (i) implies (ii). Assume rst that f : X - Y is locally of nite type
(resp. locally of nite presentation) at a certain point x X. Then there are
ane open subschemes U X and V Y where U is a neighborhood of x, such
that f (U ) V and the corresponding morphism of rings : OY (V ) - OX (U )
is of nite type (resp. nite presentation). Choosing a section b B such that
f (x) D(b) V , we may localize by b and thereby assume using Lemma 3 (i)
that V is a basic open subset of Y . Furthermore, since the localization map
B - B b1 is of nite presentation, we see from Lemma 3 (ii) that OX (U )
is even of nite type (resp. nite presentation) over B. In particular, we may
assume V = Y . In a similar way it is possible to replace U by the basic open
subset D(g) attached to a suitable section g A.
Now, assume that f is of nite type at all points x X. Combining the
above argument with the quasi-compactness of X (see 6.1/10), there are nitely
many sections gi A, i = 1, . . . , n, such that the ring morphisms B - Agi1
are of nite type and the D(gi ) cover X or, what is equivalent to the latter,
such that there is an equation ni=1 ai gi = 1 for certain sections ai A. Then
consider a B-subalgebra A A of nite type containing all sections ai , gi and
large enough such that all morphisms i : A gi1 - A gi1 , i = 1, . . . , n,
induced from the inclusion A - A are isomorphisms. To show that the
latter is possible, observe rst that all i are injective, since the localization
morphisms A gi1 are at by 4.3/3. Moreover, Agi1 is a nitely
- A
generated B-algebra. So if h = gar Agi1 for some a A and r N is any of
i
the nitely many generators, we may add a to A . Thereby we can assume that
all generators h belong to A gi1 and, hence, that A gi1 = Agi1 .
By our construction, the morphism f : Spec A - Spec B admits a fac-
torization

Spec A - Spec A - Spec B
with a morphism that is an isomorphism overall basic open subschemes

D(gi ) Spec A , i = 1, . . . , n. Since the equation ni=1 ai gi = 1 persists in A ,
the D(gi ) cover Spec A and it follows that must be an isomorphism. Thus,
we see from 6.6/9 that A coincides with A so that A is of nite type over B.
It remains to consider the case where f is locally of nite presentation. Then,
by the above reasoning, we know already that the ring morphism B - A is
of nite type so that there is an exact sequence

0 - a - Bt1 , . . . , tm - A - 0
where the surjection gives rise to a closed immersion Spec A - Am B . Pro-
ceeding as in the nite type case, there is a nite covering of Spec A by basic
open subsets D(gi ), i = 1, . . . , n, such that Agi1 is of nite presentation over

B. Taking -preimages of the gi in Bt1 , . . . , tm , we extend these sections to
sections on Am
B , denoted by gi again. Thus, using the atness of localization
maps and localizing the above exact sequence by gi yields an exact sequence
i
0 - ai - Bt1 , . . . , tm gi1 - Agi1 - 0,
where ai = a Bt1 , . . . , tm gi1 . We claim that ai is a nitely generated ideal
in Bt1 , . . . , tm gi1 . Knowing this, we see that the quasi-coherent OAm B
-ideal
associated to a is locally of nite type in the sense of 6.8/12. Then it follows from
6.8/13 that a is nitely generated and, hence, that A is of nite presentation
over B.
To show that all ai are nitely generated, look at the canonical commutative
diagram with exact rows
0 - ai - Bt1 , . . . , tm t - Agi1 - 0

? ?
0 - ai - Bt1 , . . . , tm gi1 - Agi1 - 0
where t is an additional variable that is mapped to gi1 ,
horizontally and verti-
cally. The middle vertical map is surjective. Hence, the same is true for the left
vertical one. Furthermore, the kernel ai is nitely generated by Lemma 2, since
Agi1 is of nite presentation over B. But then ai must be nitely generated
as well, and we are done.
Finally, let us mention a particular property on closed points of schemes

that are locally of nite type over a eld.
Proposition 6. Let X be a scheme that is locally of nite type over a eld k.

Then any locally closed point x X is closed, i, e., if x is closed in a certain
open neighborhood U X, then x is already closed in X.
Proof. Assuming that x is locally closed in X, there exists an ane open sub-
scheme U X such that x is closed in U . Hence, the residue extension k(x)/k
is nite by 3.2/4. Now let U be an arbitrary ane open neighborhood of x in
X, say U = Spec A, and let px A be the ideal corresponding to x. Then
A/px k(x) is nite over k and we see from 3.1/2 that A/px must be a eld.
Therefore px is a maximal ideal in A and, consequently, x is closed in U . Thus,
for every ane open part U X the intersection {x} U is closed in U and
this implies that x is a closed point in X.
Exercises
1. For a ring R and a multiplicatively closed subset S R consider the localization
morphism R - RS . Show that R - RS is of nite type if and only if it is
of nite presentation.
8.4 Unramied Morphisms 365
2. Let : A - B be a nite morphism of rings, in the sense that B is a nite

A-module via . Show that B is an A-algebra of nite presentation via if and
only if B is an A-module of nite presentation via .
3. Let : A - B be a morphism of rings that is of nite presentation. Show that
there exists a morphism of rings : A - B such that:
(a) A , resp. B is a subring of A, resp. B.
(b) A is Noetherian and is of nite type.
(c) is obtained from by coecient extension with A over A .
Sometimes this is useful to reduce problems on morphisms of nite presentation
to the Noetherian case.
4. Let X be a scheme that is locally of nite type over a locally Noetherian base
scheme S. Show that X is locally Noetherian and that it is locally of nite pre-
sentation over S.
5. Let f : X - Y and g : Y - Z be morphisms of schemes. Show:
(a) If g f is locally of nite type, the same is true for f .

(b) If g f is locally of nite presentation and g is locally of nite type, then f is
locally of nite presentation.
6. Let X - S be a morphism of schemes that is locally of nite type. Show that
the diagonal morphism X - X S X is locally of nite presentation.
7. Let f : Y - X be a locally closed immersion of schemes and let U X be

an open subscheme such that f factors through a closed immersion Y - U.
Let I OU be the corresponding quasi-coherent ideal. Show that f is locally of
nite presentation if and only if the ideal I OU is locally of nite type.
8.4 Unramied Morphisms
Now we are ready to discuss unramied morphisms of schemes. As will be seen

from Theorem 3 below, such morphisms may be viewed as a certain generaliza-
tion of nite separable eld extensions.
Denition 1. A morphism of schemes f : X - S is called unramied at

a point x X if there exists an open neighborhood U X of x as well as a
closed S-immersion 1 j : U - W AnS into an open subscheme W of some
ane n-space AnS over S such that:
(i) If I OW is the sheaf of ideals associated to the closed immersion j,
there exist nitely many sections generating I in a neighborhood of j(x).
(ii) The dierential forms of type dg for sections g of I where d stands for
the exterior dierential dAnS /S : OAnS /S - 1 n , generate 1 n
AS /S AS /S at j(x).
The morphism f is called unramied if it is unramied at all points of X.
1
An S-immersion is an S-morphism that is an immersion.
Let us explain more closely the terms occurring in the denition. An im-
mersion j : U - W An as required for an open neighborhood U X of x
S
exists if and only if f is locally of nite type at x. In conjunction with condi-
tion (i) this is equivalent to the fact that f is locally of nite presentation at
x. Moreover, in condition (ii) we require that 1An /S is generated at j(x) by all
S
dierential forms obtained from certain local sections of I at j(x). Since 1An /S
S
is a quasi-coherent OAnS /S -module that is locally of nite type by 8.1/6, this
condition may be viewed as a local condition at j(x), or even as a condition on
the level of stalks at j(x), just according to the following lemma:
Lemma 2. Let X be a scheme and F a quasi-coherent OX -module of locally

nite type. Furthermore, let f1 , . . . , fr be local sections of F at a point x X
that are dened in an open neighborhood U X of x. Then the following
(i) There exists an ane open neighborhood U U of x such that the
elements fi |U generate F on U , i.e. such that F(U ) = ri=1 OX (U )fi |U .
(ii) The stalk of F at x satises Fx = ri=1 OX,x fi,x where fi,x is the germ
of fi at x.
(iii) Fx OX,x k(x) = ri=1 k(x)f i,x where k(x) = OX,x /(px ) is the residue
eld of OX at x and f i,x denotes the residue class of the germ fi,x Fx .
If the equivalent conditions of the lemma are met, we say that F is generated
at x by the local sections f1 , . . . , fr .
Proof of the lemma. The implications (i) = (ii) = (iii) are trivial. Moreover,
we have (iii) = (ii) by Nakayamas Lemma in the version of 1.4/11, since F
is assumed to be locally of nite type. Now, to go from (ii) to (i), assume (ii)
and consider an ane open neighborhood U U of x together with sections
g1 , . . . , gs F(U ) generating F on U . Then the germs g1,x , . . . , gs,x Fx
generate the stalk Fx and there are coecients aij,x OX,x such that

r
gj,x = aij,x fi,x , j = 1, . . . , s.
i=1
Taking U small enough, the germs aij,x OX,x extend to sections aij OX (U )
and, shrinking U even more if necessary, we may assume

r
gj = aij fi |U , j = 1, . . . , s.
i=1
Then f1 , . . . , fr generate the OX -module F on U , as required in (i).
If a morphism of schemes f : X - S is unramied at a point x X, it

follows from the preceding lemma that f is unramied in an open neighborhood
of x. Therefore the unramied locus of f , namely, the set of all points in X,
where f is unramied, is open in X. Let us consider an example. For a eld k, a

polynomial p kt in one variable t, and X = S = A1k consider the morphism
f: X - S, corresponding to the morphism of k-algebras
f # : OS (S) = kt - kt = OX (X), t - p.
Let us determine the locus in X where f is unramied. To construct an

S-immersion of X into an ane n-space AnS , we use the graph morphism of
7.4/13, viewing f as a morphism of k-schemes. Thus, writing t instead of t for
the coordinate function on S, we extend f # to a surjection
j# : kt, t - kt, t - p, t - t,
thereby obtaining a closed S-immersion
j = (id, f ) : X = A1k - X k S = A1S ,
which can suggestively be described by x - (x, f (x)). The corresponding

ideal ker j kt, t is generated by p t and we see that f is unramied at
#
a point x X if, as a sucient condition, the dierential form

dp
d(p t ) = dp = dt
dt
generates the module of dierential forms 1A1 /S at j(x); note that dt = 0 by
S
8.1/2, since t is a section living on the base S. Now observe that 1A1 /S is a
S
free OA1S -module, generated by the dierential form dt. Thus, we see that f is
unramied at x, if dp dt
(x) = 0, or in other words, if p p(x), as a polynomial
in k(x)t, does not admit multiple zeros in algebraic extensions of k(x). This
condition is necessary as well, as can be deduced with the help of 8.2/5 from
the characterization of unramied morphisms given below in Theorem 3 (ii);
for details see the proof of Corollary 5.
Moreover, let us point out that any immersion is unramied, as soon as it
is of locally nite presentation. Also it can be shown in a straightforward way
that unramied morphisms are stable under base change, composition, and the
formation of ber products; see Exercise 1.
Now let us state the main characterization theorem for unramied mor-
phisms.
Theorem 3. Let f : X - S be a morphism of schemes that is locally of

nite presentation. Then, for points x X and s = f (x) S, the following
(i) f is unramied at x.
(ii) 1X/S,x = 0.
(iii) The diagonal morphism : X - X S X is a local isomorphism at
x, i.e. there exists an open neighborhood U X of x such that its image (U )
is open in X S X and induces an isomorphism U - (U ).
(iv) Xs = X S Spec k(s) is unramied over Spec k(s) at x.

(v) Consider the morphism between stalks fx# : OS,s - OX,x . Then the
maximal ideal mx OX,x is generated by the image fx (ms ) of the maximal
#
ideal ms OS,s and the induced map between residue elds k(s) - k(x)
denes k(x) as a nite separable extension of k(s).
Before starting
! the proof, let us derive some consequences from the theorem.
We will write iI Xi for the disjoint union of a family of schemes Xi , i I,
just gluing !
these via empty intersections. This way (Xi )iI becomes an open
covering of iI Xi .
Corollary 4. A morphism of schemes X ! - Spec k where k is a eld is

unramied if and only if we have X = iI Spec ki where the morphisms
Spec ki - Spec k correspond to nite separable extensions of elds k - ki .
! - Spec k of the type mentioned in the
Proof. Every morphism iI Spec ki
assertion is unramied by Theorem 3 (v). For the converse it is enough to show
that every ane open subscheme of X is a disjoint union of schemes Spec ki as
stated in the assertion. Thereby we are reduced to the case where X is ane, say
X = Spec A. If X - Spec k is unramied, we see from Theorem 3 (v) that
the stalk OX,x at any point x X is a nite separable eld extension of k. In
particular, every prime ideal of A is maximal and minimal as well. Furthermore,
since X - Spec k is locally of nite type, the same is true by 8.3/5 for A as
a k-algebra. In particular, A is Noetherian by Hilberts Basis Theorem 1.5/14
and we see from 2.1/12 that X consists of only nitely! many closed points, all
of which must be open as well. In particular, X = xX Spec k(x) where, as
seen above, each eld k(x) = OX,x is a nite separable extension of k.
Corollary 5. Let f : X - S be a morphism of schemes that is unramied

at a point x X. Furthermore, let U X be an open neighborhood of x and
j : U - W AnS any closed S-immersion from U into an open subscheme
W AnS . Then the corresponding quasi-coherent ideal I OW is generated in
a neighborhood of j(x) by nitely many sections of I and the dierential forms
of type dg for sections g of I generate the module of dierential forms 1An /S
S
at j(x).
In other words, the dening condition in Denition 1 for the property of f to
be unramied at a point x X, is independent of the choice of the S-immersion
j : U - W AnS .
Proof. Let us consider a closed S-immersion j : U - W with corresponding

quasi-coherent ideal I OW as stated. Since f : X - S is locally of nite
presentation at x, we conclude from Lemma 8.3/2 that I is generated in a
neighborhood of j(x) by nitely many sections of I. Then look at the exact
sequence
I/I 2 - j 1 - 1 -0
W/S U/S
of 8.2/5. Since f is unramied at x X, we have 1U/S,x = 0 according to

Theorem 3 (ii) and the assertion follows from the surjectivity of the map
Ix /Ix2 - (j 1W/S )x = 1AnS /S,x /Ix 1AnS /S,x
in conjunction with Nakayamas Lemma in the version of 1.4/11.
Now we turn to the proof of Theorem 3, starting with the equivalence be-
tween (i) and (ii). Since f : X - S is locally of nite presentation at x,
there is an open neighborhood U X of x together with an S-immersion
j: U - W An into an open subscheme W of some ane n-space An such
S S
that the corresponding quasi-coherent ideal I OW is generated at j(x) by
nitely many sections of I. If f is unramied at x, we may assume that, in
addition, the module of dierential forms 1An /S is generated at j(x) by the
S
dierential forms dg attached to sections g in I. Thus, considering the exact
sequence
d
I/I 2 - j 1 - 1 -0
W/S U/S
of 8.2/5, as well as the corresponding exact sequence of stalks at x (see 6.5/9),

we nd that dx : Ix /Ix2 - j 1 1
W/S,x is surjective and, hence, that U/S,x = 0.
Conversely, if U/S,x is trivial, the morphism Ix /Ix
1 2 - 1
(j W/S )x is surjective
and it follows with the help of Nakayamas Lemma as in the proof of Corollary 5
that f is unramied at x.
Next, in order to establish the equivalence (ii) (iii), decompose the
diagonal embedding X - X S X into a closed immersion : X - W
and an open immersion W - X S X. Let J OW be the quasi-coherent
ideal dening (the image of) X as a closed subscheme of W . Then we have
1X/S = (J /J 2 ) by the denition given in 8.2. Thus, 1X/S is trivial if and
only if (J /J 2 ) is trivial. Now consider an ane open part V W and let
U = 1 (V ) be its restriction to X. Then U is ane open in X. Furthermore,
using 6.9/4, we get
(J /J 2 )(U ) (J /J 2 )(V ) OW (V ) OX (U ) (J /J 2 )(V ),
since J /J 2 is an OW /J -module. Therefore (J /J 2 ) = 0 is equivalent to

J /J 2 = 0 and, thus, to J = J 2 . It follows that 1X/S = 0 is equivalent to
J = J 2.
Let z = (x). Since f : X - S is locally of nite presentation and, in
particular, of nite type at x, we see from 8.1/7 (ii) that J is of nite type on an
open neighborhood of z and the same follows for 1X/S on an open neighborhood
of x. As a consequence, the stalk 1X/S,x at x is trivial if and only if 1X/S is trivial
on an open neighborhood of x. Likewise, Jz = Jz2 is equivalent to the equation
J = J 2 on an open neighborhood of z. Therefore the above considerations
show that 1X/S,x is trivial if and only if Jz = Jz2 . Now let mz OW,z be the
maximal ideal of the stalk of OW at z where Jz mz since z (X). Then we
have Jz2 mz Jz Jz and Jz = Jz2 implies Jz = mz Jz . Moreover, the latter
yields Jz = 0 by Nakayamas Lemma 1.4/10, as Jz is a nite OW,z -module; see

8.1/7 (ii), the argument employed above. Consequently, 1X/S,x = 0 is equivalent
to Jz = 0 and, hence, to the fact that J vanishes on an open neighborhood of
z. However, the latter signies that is a local isomorphism at x.
The implication (i) = (iv) is trivial, since unramied morphisms are com-
patible with base change. For the reverse implication it is enough to derive (ii)
from (iv). To do this, assume that Xs is unramied at x over k(s) and observe
that the implication from (i) to (ii), which has already been established, yields
1Xs /k(s),x = 0 and, hence, using 8.2/3,
1X/S,x /ms 1X/S,x = 1X/S,x OS,s k(s) = 1Xs /k(s),x = 0
where ms is the maximal ideal in OS,s . Then, if mx is the maximal ideal in OX,x ,
the equation
1X/S,x /ms 1X/S,x = 0
yields
1X/S,x /mx 1X/S,x = 0
and therefore 1X/S,x = 0 by Nakayamas Lemma 1.4/10.
Accessing condition (v), let us establish the implication (v) = (ii) rst.
Since the equivalence of conditions (i) through (iv) has already been settled, we
may assume S = Spec k(s). Then OX,x is a eld by condition (v) and we claim
that x gives rise to an open subset {x} X. To justify this, choose an ane
open neighborhood U = Spec A of x in X and let : A - k(x) = OX,x be
the canonical morphism. It follows that U = Spec A with A = A/ ker equals

the closure of {x} in U and that there are inclusions
k(s) - A - k(x).
Observing that the eld k(x) is nite over k(s) by our assumption, we can
conclude from 3.1/2 that A is a eld. Furthermore, since k(x) = OX,x is a
localization of A , we see A = k(x). Hence, x is a closed point in U such that
the corresponding maximal ideal mx A is minimal as well.
Now let p1 , . . . , pr be the remaining minimal prime ideals in A. Their number
is nite by 2.1/12, due to the fact that A is Noetherian; use that A is of nite
type over k(s) by 8.3/5, in conjunction with Hilberts Basis Theorem 1.5/14.
Then pi mx for all i so that x V (p1 ) . . . V (pr ). On the
other hand,
since mx ri=1 pi = rad(A) by 1.3/4, we see that U {x} = ri=1 V (pi ) and,
hence, that the set {x} is open in U , as well as in X. As a consequence, we can
conclude that the stalk 1X/S,x is given by the k(x)-module of dierential forms
1k(x)/k(s) .
By our assumption, k(x)/k(s) is a nite separable extension of elds and
we claim that this implies
1X/S,x = 1k(x)/k(s) = 0.
Indeed, if k is a eld and k /k a nite separable extension, then it is generated by

a primitive element, say k = k(), where the corresponding minimal polynomial
p kt of over k is separable. Using the canonical isomorphism k kt/(p),
we get from 8.1/5

1k /k = 1k[t]/k /(p 1k[t]/k + kt dp) = (k dt)/ k p () dt
where p is the derivative of p. However, p () = 0 since p is separable. Therefore

we get 1k /k = 0.
To nish our proof, we establish the implication (iii) = (v). Let us consider
the ber Xs of f : X - S over the point s, as well as the corresponding stalks
at x and s, thereby obtaining commutative diagrams
X Xs OX,x - OX,x /ms OX,x

6 6
? ?
S Spec k(s) , OS,s - OS,s /ms = k(s)
where ms is the maximal ideal of OS,s . Now, writing mx for the maximal ideal of
OX,x , it is clear that mx = ms OX,x is equivalent to mx /ms OX,x = 0. Furthermore
the residue extension k(x)/k(s) is the same for f : X - S and its ber
f s : Xs - Spec k(s). It follows that f satises condition (v) if and only the
ber fs does. Thus, in view of the fact that condition (iii) is stable under base
change, we may replace the S-scheme X by its ber Xs over s and thereby
assume S = Spec k for a eld k. Moreover, we can pass from X to an ane open
neighborhood of x so that we are dealing with an ane k-scheme X = Spec A
of nite type. Shrinking X even more, we can assume that, in addition, the
diagonal morphism X - X k X is an open immersion.
To derive condition (v) in this situation, it is enough to show that the
k-algebra A = OX (X) is a nite direct product of elds that are nite separable
over k. Fixing an algebraic closure k of k, we see from Lemma 6 below that A
is such a direct product if and only if the k-algebra A k k is a nite direct
product of copies of k. In other words, we may assume that k is algebraically
closed.
Doing so we claim that the Zariski topology of X = Spec A coincides with
the discrete one (where every subset of X is open). Since the closure of any
point of X contains a closed point, it is enough to show that all closed points
of X are open. Therefore consider a closed point z X. Since A is of nite
type over the algebraically closed eld k, we can read k(z) = k from 3.2/4. In
particular, we may interpret z as a k-valued point z : Spec k - X and, hence,
as a section of X over k. Then consider the morphism
h: X - X k X, y - (y, z),
where, using the projections p1 , p2 : X k X - X, the morphism p1 h coin-

cides with the identical morphism id : X - X and p2 h is the composition
z
X - Spec k - X
of the structural morphism of X with the section given by z. Then the

h-preimage of the diagonal (X) X k X coincides with {z}, since any
w X satisfying h(w) (X) yields

w = p1 h(w) = p2 h(w) = z.
Therefore, as claimed, {z} is open in X, since (X) is open in X k X by our

assumption.
Having shown that X carries the discrete topology, all points of X are open
and closed. Since X is ane and, hence, quasi-compact by 6.1/10, it consists
of only nitely many points and, thus, is the disjoint union of open subschemes
concentrated at the points of X. To show that A = OX (X) is a direct product of
copies of k = k we may assume X = {x} where the residue eld k(x) coincides
with k, as shown above. Then A/ rad(A) k and the isomorphisms

A k A / rad(A) k A + A k rad(A) A/ rad(A) k A/ rad(A)
k k k k
show that X k X consists of a single point, just as X does. Furthermore, by

our assumption the diagonal embedding X - X k X or, in other words, the
multiplication map A k A - A is an isomorphism. However, for a non-zero
k-algebra A the latter can only be the case if the structural morphism k - A
is an isomorphism, since otherwise the multiplication map Ak A - A would
have a non-trivial kernel. Thus, X = Spec k and we are done.
Lemma 6. Let A be a k-algebra where k is a eld. Fixing an algebraic closure

k of k, the following conditions are equivalent:
(i) The k-algebra A is a nite direct product of elds that are nite and
separable over k.
(ii) The k-algebra A k k is a nite direct product of copies of k.
Proof. To pass from (i) to (ii), consider a nite separable eld extension k /k.
Choosing a primitive element k , we may assume
r k = k(). Let p kt
be the minimal polynomial of over k and p = j=1 (t j ) its factorization
over k, where the zeros 1 , . . . , r k are pairwise dierent, as p is separable.
Then the Chinese Remainder Theorem yields

r
r
k k k kt/(p) kt/(t j ) k,
j=1 j=1
from which the implication (i) = (ii) is easily derived.

Conversely, if (ii) is given, we have rad(A) = 0, since rad(A k k) = 0, and
the canonical map A - A k k is injective. Furthermore, using 3.1/6, we see
that the latter map is integral, since A k k is generated over A by k and k is
integral over k. Also we know that
dimk A = dimk (A k k) < .

Now if p is a prime ideal in A, the quotient A/p is an integral domain that is nite
and, hence, integral over k. Therefore A/p is a eld by 3.1/2. Furthermore, we
see from the Lying-over Theorem 3.3/2 that every prime ideal of A is restriction
of a prime ideal in A k k. It follows that A, just as A k k, can contain only
nitely many prime ideals p1 , . . . , ps . Since the quotients ki = A/pi are elds,
all pi are maximal as well as minimal, and the Chinese Remainder Theorem in
conjunction with rad(A) = 0 shows A si=1 ki . Then

s
A k k (ki k k)
i=1
implies rad(ki k k) = 0 for all i and we claim that therefore all ki are separable
over k. Indeed, consider an element ki for some i and let p kt be its
minimal polynomial over k. Then, if p = rj=1 (t j )nj is the factorization of
p over k, the inclusion

r
kt/(t j )nj kt/(p) k() k k - ki k k
j=1
shows that all multiplicities nj must be trivial and, hence, that ki /k is separable.
This nishes the step from (ii) to (i).
Exercises
1. Show that unramied morphisms are stable under base change, composition, and
the formation of ber products. Hint: For compositions use the argument in the
proof of 8.3/3 (ii) showing that the composition of two morphisms of locally nite
presentation is of locally nite presentation again.
2. Consider the canonical morphism Kt1 , t2 (t1 t2 ) - Kt1 Kt2 for vari-
ables t1 , t2 over a eld K. Show that the associated morphism of K-schemes is
unramied.
3. Consider Neiles parabola X = Spec Kt1 , t2 /(t22 t31 ) over a eld K and the
attached K-morphism f : A1K - X, as in Exercises 6.2/6 and 6.7/3. Determine
the unramied locus of f .
4. Let f : X - Y be a monomorphism of schemes. Show that f is unramied if
and only if it is locally of nite presentation. Hint: Exercise 7.4/1.
5. Let f : X - Y be a morphism of schemes that is locally of nite presenta-
tion. Show that f is unramied if and only if for each Y -scheme T and each
Y -morphism g : T - X the Y -morphism (g, idT ) : T - X Y T is an open
immersion.
6. For two S-morphisms f, g : T - X on some base scheme S consider the co-
incidence scheme Z = T (XS X) X as in Exercise 7.4/3; it is a locally closed
subscheme of T and even a closed subscheme if X is separated over S. Show that
Z is an open subscheme of T if X is unramied over S.
7. Unramied extensions of discrete valuation rings: Let R - R be an integral

extension of discrete valuation rings. Let K /K be the attached extension of elds
of fractions and k /k the residue extension modulo maximal ideals. Show that the
corresponding scheme morphism Spec R - Spec R is unramied if and only if
the extensions K /K and k /k are nite separable and satisfy K : K = k : k.
Hint: Use Exercise 3.1/8.
8.5 Smooth Morphisms
In the introduction to the present Chapter 8 we have already explained that the
Jacobian Condition known from Differential Geometry provides a rather
natural device for dening smooth morphisms in Algebraic Geometry. In
the following we will discuss this approach in detail.
Denition 1. A morphism of schemes f : X - S is called smooth at a point

x X (of relative dimension r) if there is an open neighborhood U X of x
together with a closed S-immersion j : U - W AnS into an open subscheme
W of some ane n-space AnS such that:
(i) If I OW is the sheaf of ideals corresponding to the closed immersion
j, there are n r sections gr+1 , . . . , gn in I that generate I in a neighborhood
of z = j(x); in particular, we assume r n.
(ii) The residue classes dgr+1 (z), . . . , dgn (z) 1An /S k(z) of the dierential
S
forms dgr+1 , . . . , dgn are linearly independent over k(z).
We say that f is smooth on X if it is smooth at all points of X. Moreover,
f is called etale at x (resp. etale on X) if it is smooth of relative dimension 0
at x (resp. smooth of relative dimension 0 at all points of X).
The notation 1An /S k(z), as used above, is an abbreviation for the

S
k(z)-vector space
1AnS /S,z OAn ,z k(z) 1AnS /S,z /mz 1AnS /S,z

S
where mz OAnS ,z is the maximal ideal and k(z) = OAnS ,z /mz the residue eld of
z. To relate the denition of smoothness to the Jacobian Condition, let t1 , . . . , tn
be the coordinate functions on AnS and imagine the elements gr+1 , . . . , gn occur-
ring in Denition 1 as rational functions in t1 , . . . , tn with coecients living on
the base S. It follows from the construction in the proof of 8.1/4, recalled in
8.1/6, that the dierential forms dt1 , . . . , dtn give rise to a free set of generators
of 1An /S . Thus, there are equations
S

n
dgj = aij dti , j = r + 1, . . . , n,
i=1
with unique sections aij in OS . Using formal partial derivatives, a reasoning

similar to the one in the proof of 8.1/3 shows
8.5 Smooth Morphisms 375

n
gj
dgj = dti , j = r + 1, . . . , n.
i=1
ti
Hence, the linear independence of the residue classes dgr+1 (z), . . . , dgn (z) is
equivalent to the relation

gj
rg (z) = n r
ti
for the Jacobian matrix of gr+1 , . . . , gn at z. Thus, we can say that a morphism
f: X - S is smooth at a point x X if, locally at x, there is an immer-
sion into some ane n-space AnS such that the Jacobian Condition is satised.
However, it is highly non-trivial to show that this condition is satised for every
immersion into an ane m-space Am S as soon as it is satised in one particular
case. This is the so-called Jacobian Criterion in Algebraic Geometry, which we
will prove later in Proposition 9 using as basic ingredient the so-called Lifting
Property for smooth morphisms; see Theorem 8 below.
Note that the structural morphism AnS - S of the ane n-space over a
base scheme S is a trivial example of a smooth morphism. Also observe that a
morphism f : X - S is locally of nite presentation at a point x X if it is
smooth at x. Furthermore, if f is etale at x, it is unramied at x in the sense
of 8.4/1; see also Proposition 6 below.
For an S-scheme X with structural morphism f : X - S, the relative
dimension at a point x X is usually dened as the topological dimension
dimx f = dimx Xs at x of the ber Xs = f 1 (s) over s; for the denition of
local dimensions see 7.5/6. We will show in Proposition 4 below for smooth
morphisms that the relative dimension in this sense coincides with the one as
specied in the setting of Denition 1.
Let us start now proving some elementary facts on smooth morphisms.
Proposition 2. Smooth (resp. etale) morphisms are stable under base change,
composition, and the formation of ber products.
The proof is straightforward from the denition of smooth and etale mor-
phisms; see Exercise 1.
Proposition 3. Let f : X - S be locally of nite presentation. Then the set

of all points x X where f is smooth of relative dimension a given number r,
is open in X.
Proof. Consider a point x X where f is smooth of relative dimension r

and look at a situation as in Denition 1. Then we have an open neighbor-
hood U X of x together with a closed S-immersion j : U - W AnS and
its associated quasi-coherent ideal I OW . Furthermore, there are sections
gr+1 , . . . , gn of I generating I in a neighborhood W W of z = j(x) such that
the residue classes dgr+1 (z), . . . , dgn (z) 1An /S k(z) are linearly independent
S
over k(z). Since the dierential forms dt1 , . . . , dtn attached to the coordinate
functions of AnS yield a free generating system of 1An /S , we can enlarge the
S
system dgr+1 (z), . . . , dgn (z) to a k(z)-basis of 1An /S k(z) by adding r residue
S
classes of the dti , say by adding dt1 (z), . . . , dtr (z). Then 8.4/2 shows that the
dierential forms dt1 , . . . , dtr , dgr+1 , . . . , dgn generate 1An /S at z and therefore
S
also on a certain neighborhood of z, for example on W W . Hence, for any
z W the n residue classes
dt1 (z ), . . . , dtr (z ), dgr+1 (z ), . . . , dgn (z ) 1AnS /S k(z )
will form a k(z )-basis of 1An /S k(z ), which is a k(z )-vector space of dimension
S
n, and we see that the residue classes
dgr+1 (z ), . . . , dgn (z ) 1AnS /S k(z )
are linearly independent for all z W . Thereby we have shown that the
dening property of the smoothness of f at x automatically extends to a certain
neighborhood j1 (W ) of x. Consequently, the set of points in X where f is
smooth of relative dimension r, is open in X.
Proposition 4. Let f : X - S be smooth of relative dimension r at a point

x X. Then, indeed, r coincides with the relative dimension dimx f of f at x.
Proof. Replacing X by a suitable open neighborhood of x, we can assume that

f is smooth of relative dimension r at all its points. Since the smoothness is
preserved by base change due to Proposition 2, we may pass to the ber of f over
s = f (x) and thereby assume that S is the spectrum of a eld k. Furthermore,
we may assume that X is ane, say X = Spec A where A is a k-algebra of nite
type. In particular, A is Noetherian then by 1.5/14 and the same is true for its
localization OX,x .
If A is an integral domain, we know from 3.3/8 that ht m = dim A for all
maximal ideals m A. We then have dimx f = dimx X = dim U for every non-
empty open subscheme U X, since the closed points of U are dense in X;
use Hilberts Nullstellensatz 3.2/6. In the general case, X will consist of nitely
many irreducible components C1 , . . . , Cs (see 7.5/5), so that
dimx X = max dim Ci .

xCi
In particular, there is an index i0 such that dimx X = dim Ci0 , and we can nd
an ane open subscheme U X that is contained in Ci0 , but disjoint from all
other Ci . Since we have required f : X - Spec k to be smooth of relative
dimension r at all its points, we may replace X by U , forgetting about the
initial point x. This way we are reduced to the case where X is irreducible.
Next choose an algebraic closure k of k and apply the base change k/k
to f , thus obtaining a morphism f : X - Spec k that is smooth again by
Proposition 2. Furthermore, we have dim A k k = dim A by 3.3/6. Similarly as
just explained, we may replace X by an irreducible open part. Thus, we have

reduced our original problem to the situation where k is algebraically closed
and X is ane and irreducible. Furthermore, taking for x any closed point in
X, we have dimx X = dim OX,x by 3.3/8 and it remains to show dim OX,x = r.
Now assume that there is a closed immersion j : X - W AnS into
some open subscheme W AnS where the conditions of Denition 1 are sat-
ised. Then there are local sections gr+1 , . . . , gn in I dening X as a closed
subscheme on a neighborhood of z = j(x) in W such that the residue classes
dgr+1 (z), . . . , dgn (z) 1An /S k(z) are linearly independent over k(z). Note
S
that k(z) = k, due to 3.2/4 and the fact that k is algebraically closed.
In particular, the maximal ideal mz OW,z corresponding to z is of type
mz = (t1 c1 , . . . , tn cn ) for the coordinate functions ti on AnS and suitable
constants ci k. Now, looking at the exact sequence
mz /m2z - 1OW,z /k OW,z k(z) - 1k(z)/k = 0
of 8.1/13, the rst map is an isomorphism by reasons of vector space dimensions.

Therefore we can enlarge the system gr+1 , . . . , gn to a system g1 , . . . , gn mz
dening a closed subscheme in a neighborhood of z that is etale over S at z.
Then g1 , . . . , gn generate mz by Nakayamas Lemma in the version of 1.4/11
and we thereby see that gr+1 , . . . , gn are part of a system of parameters in OW,z .
Thus, we conclude from 2.4/13 that dim OX,x = dim OW,z /(gr+1 , . . . , gn ) = r,
as desired.
Proposition 5. Let f : X - S be smooth of relative dimension r at a point

x X. Then X/S is locally free of rank r at x.
1
Proof. Consider a situation as in Denition 1, namely, an open neighborhood

U X of x together with a closed S-immersion j : U - W AnS and with as-
sociated quasi-coherent ideal I OW . Then there are local sections gr+1 , . . . , gn
in I generating I in some neighborhood W W of z = j(x) such that the
residue classes dgr+1 (z), . . . , dgn (z) 1An /S k(z) are linearly independent
S
over k(z). As in the proof of Proposition 3 we may assume that the dierential
forms dt1 , . . . , dtr obtained from the rst r coordinate functions of AnS together
with dgr+1 , . . . , dgn generate 1An /S on some ane open neighborhood of z, say
S
on W . Fixing such generators, we get a short exact sequence of quasi-coherent
OW -modules
0 - R - OW n

- 1 n |W
AS /S
-0
that splits, since 1An /S is free. Then R may be viewed as a quotient of OW n

S
and, hence, is locally of nite type. Furthermore, we get R k(z) = 0 from
our construction, since a split short exact sequence remains exact under coe-
cient extension; for example, use 4.1/8. Now Nakayamas Lemma 1.4/10 yields
Rz = 0 and, hence, that R is trivial in a neighborhood of z. Therefore, on some
neighborhood of z, say on W , the dierential forms dt1 , . . . , dtr , dgr+1 , . . . , dgn
give rise to a free generating system of 1An /S , and the exact sequence
S
I/I 2 - j 1 n - 1X/S - 0
AS /S
from 8.2/5 shows that 1X/S is freely generated on W by the images of the
dierential forms dt1 , . . . , dtr . In particular, 1X/S is locally free at x of rank r.

Proposition 6. The following conditions are equivalent for a morphism of

schemes f : X - S:
(i) f is etale.
(ii) f is smooth and unramied.
Proof. Assume rst that f is etale at some point x X, and consider an open
neighborhood U X of x with a closed S-immersion j : U - W AnS where
I OW is the associated quasi-coherent ideal, as in Denition 1. Furthermore,
let g1 , . . . , gn be local sections generating I on some neighborhood of z = j(x)
and assume that the residue classes dg1 (z), . . . , dgn (z) 1An /S k(z) are linearly
S
independent so that they form a k(z)-basis in 1An /S k(z). Then Nakayamas
S
Lemma in the version of 1.4/11 shows that the dierential forms dg1 , . . . , dgn
generate the module 1An /S at z. Therefore f is unramied at x and, as an etale
S
morphism, smooth at x as well.
Conversely, assume that f is smooth of some relative dimension r and, in
addition, unramied. Then, xing a point x X, there exists an open neigh-
borhood U X together with a closed S-immersion j : U - W AnS and
associated quasi-coherent ideal I OW as in Denition 1, namely, such that I
is generated in a neighborhood of z = j(x) by local sections gr+1 , . . . , gn where
the residue classes dgr+1 (z), . . . , dgn (z) 1An /S k(z) are linearly independent.
S
Since f is unramied at x, the dierential forms dg attached to sections g in
I will generate the module 1An /S at z by 8.4/5 and therefore also the vec-
S
tor space 1An /S k(z), which is of dimension n over k(z). From an equation
n S
g = i=r+1 ai gi with sections ai in OW we conclude

n

n
dg(z) = ai (z)dgi (z) + gi (z)dai (z) = ai (z)dgi (z)
i=r+1 i=r+1
because of gr+1 (z) = . . . = gn (z) = 0, and we see that the residue classes
dgr+1 (z), . . . , dgn (z) are even a basis of 1An /S k(z). Therefore we must have
S
r = 0, and f is etale at x.
Moreover, one can show that the condition etale is equivalent to at

and unramied ; see Corollary 18 below. For example, one easily veries that
a nite separable extension of elds k /k gives rise to an etale morphism
Spec k - Spec k. Then we conclude from 8.4/4 for a morphism X -S
with S the spectrum of a eld that the conditions etale and unramied coin-
cide. On the other hand, the dierence between unramied and etale morphisms
becomes already apparent, when considering immersions. As we know from Sec-

tion 8.4, any immersion of schemes is unramied, provided it is locally of nite
presentation. However, etale immersions are open:
Lemma 7. An immersion f : X - S is etale if and only if it is an open

immersion.
Proof. If f : X - S is an open immersion, we may interpret S as the trivial

ane space AS and X as an open subscheme of A0S . Then the identical morphism
0
j: X - X A0 shows that f is etale at all points x X.

S
Conversely, assume that f is etale. We only need to consider the case where
f is a closed immersion, claiming that in such a situation f is an open immersion
as well. To prove this we can work locally on X. Therefore consider an open
neighborhood U X of some point x X, together with a closed S-immersion
j: U - W An into some open subscheme W of An such that the dening
S S
conditions for f being etale at x in the sense of Denition 1 are satised. Then
we can assume U , S, and W to be ane with rings of global sections A, R, and
A and, furthermore, that there are global sections g1 , . . . , gn OW (W ) = A
generating the quasi-coherent ideal I OW attached to j with the property that
the dierential forms dg1 , . . . , dgn generate 1W/S at z = j(x). Thus, if t1 , . . . , tn
are the coordinate functions of AnS , we have the following commutative diagrams
- - A A
j
U
W
AnS Rt1 , . . . , tn
-

f
?
-
S R
on the level of schemes, respectively, on associated rings of global sections. As
the morphism R - A is surjective, we may modify the variables t1 , . . . , tn in
such a way that their images in A belong to the kernel of A - A and, thus,
to I. Then there are equations

n
t j |W = aij gi , j = 1, . . . , n,
i=1
with coecients aij OW (W ). Using gi (z) = 0 for all i as in the proof of

Proposition 6, the attached equations on the level of dierential forms

n
dtj (z) = aij (z)dgi (z), j = 1, . . . , n,
i=1
show that the determinant det(aij ) is a unit on some neighborhood of z. Hence,

the matrix (aij ) is invertible at z. Shrinking S, U , and W if necessary, we may
even assume that (aij ) is invertible on all of W . Then the elements t1 |W , . . . , tn |W
generate the ideal I on W since the same is true for g1 , . . . , gn . Thus, applying
the base change given by the zero section S - An to the open immersion
S
W - AS , we end up with a morphism U
n - S that, by its construction,
is an open immersion coinciding with the restriction of f to U . Since such a
consideration is possible for all x X, the closed immersion f is open.
Next we discuss the fundamental characterization of unramied, smooth,

and etale morphisms via the so-called Lifting Property.
Theorem 8 (Lifting Property). For a morphism f : X - S of locally nite

presentation the following conditions are equivalent:
(i) f is unramied (resp. smooth, resp. etale).
(ii) For every S-scheme Y where Y is ane and for every closed subscheme
Y Y given by a quasi-coherent ideal J OY satisfying J 2 = 0, the canonical
map
: HomS (Y, X) - HomS (Y , X)
is injective (resp. surjective, resp. bijective).
Proof. We start by looking at unramied morphisms. In this case both condi-

tions (i) and (ii) are local on X and local over S. Therefore we may assume X
and S to be ane, say X = Spec A and S = Spec R. Let B be an R-algebra
and J B an ideal satisfying J2 = 0. Then we know from 8.1/8 that for any
R-algebra morphism : A - B, the mapping - denes a bijection
between the set of R-algebra morphisms : A - B satisfying mod J
and the set DerR (A, J) of all R-derivations from A to J.
If f is unramied, we read 1X/S = 0 from 8.4/3 so that DerR (A, J) = 0. It
follows that any R-algebra morphism : A - B satisfying mod J will
coincide with and, hence, that is injective. To show the converse, consider
the multiplication map A R A - A. Let J A R A be its kernel and
set B = (A R A)/J2 . Then the square of the ideal J/J2 = JB B is trivial
and we can look at the canonical map : HomR (A, B) - HomR (A, B/JB).
Now, if is injective, we get DerR (A, J/J2 ) = 0, as explained above. Since
J/J2 coincides with the module of dierential forms 1A/R by 8.1/10, we get
Hom(1A/R , 1A/R ) = 0 and, thus, 1A/R = 0 so that f is unramied by 8.4/3.
Next we deal with smooth morphisms. By its denition, the smoothness of
the morphism f : X - S in condition (i) can be tested locally on X. That
the same is possible for the corresponding lifting property in (ii) is not obvious
at all and requires a special argument. Postponing this problem for a moment,
let us look at a local situation where f : X - S is a smooth morphism
of relative dimension r and where X and S are ane, say X = Spec A and
S = Spec R. Let j : X - W An be a closed immersion of X into an open
S
ane subscheme W = Spec A of some ane n-space AnS = Spec Rt1 , . . . , tn ,
where W is assumed to be basic open in AnS so that A is a localization of the
polynomial ring Rt1 , . . . , tn . Let I A be the ideal attached to j. Choosing
X small enough, we may assume as in the proof of Proposition 5 that there are
elements g1 , . . . , gn A such that their associated dierential forms dg1 , . . . , dgn
form a set of free generators of 1A /R and, in addition, gr+1 , . . . , gn generate the
ideal I A . Then the residue classes of the latter elements generate I/I2 as a
module over A A /I and it is easily seen that the exact sequence from 8.1/13
gives rise to a short exact sequence
() 0 - I/I2 - 1A /R A A - 1A/R - 0
that is split.
Now let Y = Spec B be an ane S-scheme and Y Y a closed sub-
scheme given by some ideal b B satisfying b2 = 0. We have to show that
every R-algebra morphism : A - B/b lifts to an R-algebra morphism
: A - B. The corresponding problem for A in place of A can be solved,
using the universal properties of polynomial rings and their localizations. For
this to work well, observe that an element b B is a unit if and only if its
residue class b B/b is a unit. The latter is true because all elements of type
1 + B with b are invertible, namely, (1 + )1 = 1 due to 2 = 0.
Thus, there exists an R-algebra morphism : A - B such that the diagram
A - A = A /I
? ?
B - B/b
is commutative. Then we have necessarily (I) b and gives rise to an
A-linear map
: I/I2 - b/b2 = b.
As the above short exact sequence () is split, we see that extends to an
A-linear map : 1A /R A A - b as follows:
0 - I/I2 - 1A /R A A - 1A/R - 0

- ?
b

In particular, induces by composition with the canonical maps
dA /R
A - 1A /R - 1A /R A A
an R-derivation : A - b satisfying |I = |I . Since the image of is

contained in b, we see from 8.1/8 that : A - B is a morphism of
R-algebras giving rise to the desired lifting : A = A /I - B of , due to
the fact that I ker( ).
Now, to derive the lifting property for a smooth morphism of general type
f: X - S, let Y = Spec B be an S-scheme that is ane and let Y Y be a
closed subscheme, given by a quasi-coherent ideal J OY satisfying J 2 = 0.
Furthermore, let : Y - X be an S-morphism. Then, working relatively

with respect to ane open parts in X lying over certain ane open parts in
S and applying the above special case, we can nd a (nite) basic ane open
covering (Yi )iI of Y such that |Yi Y lifts to an S-morphism i : Yi - X for
all i I. As the lifting process is only unique in the unramied case, we cannot
expect that two liftings i , j for i, j I will coincide on the intersection Yi Yj .
To analyze the dierence between i and j , let us simplify our notation and
assume X and S to be ane, say X = Spec A and S = Spec R. Then, writing
Yi Yj = Spec Bij , as well as bij = J (Yi Yj ), the liftings i |Yi Yj and j |Yi Yj
correspond to two R-algebra morphisms A - Bij that are congruent modulo
bij and we see from 8.1/8 that these will dier by an R-derivation A - bij .
Now observing that bij is a Bij /bij -module, due to b2ij = 0, and viewing bij as
an A-module via the morphism A - Bij /bij obtained from : Y - X, we
get

DerR (A, bij ) HomA (1A/R , bij ) HomBij /bij 1A/R A (Bij /bij ), bij .
If we switch back to the previous notation where X and S are not necessarily
ane, the consideration above shows that the obstruction for gluing the mor-
phisms i : Yi - X to a lifting of : Y - X consists of a cocycle on the
basic open covering (Yi Y )iI of Y with values in the sheaf
HomOY ( 1X/S , J ),
which is a quasi-coherent sheaf of OY -modules. To explain the quasi-coherence

in more detail, note that for a scheme Z and sheaves of OZ -modules F, G on
Z, the sheaf HomOZ (F, G) is given by the functor
U - HomOZ |U (F|U , G|U ),
which is canonically an OZ -module. If F and G are quasi-coherent, the compat-

ibility between Hom and localizations for ordinary modules (see Exercise 4.3/9
or Bourbaki [6], II, 2, no. 7, Prop. 19) shows that the sheaf HomOZ (F, G) is
quasi-coherent as well, provided F is locally of nite presentation in the sense
of 6.8/12. In our case, F = 1X/S is quasi-coherent by 6.9/4, since 1X/S is
quasi-coherent. In fact, it is even locally free of nite type, since the same is
true for 1X/S by Proposition 5. Thereby we see that F = 1X/S is locally
of nite presentation and it follows that HomOY ( 1X/S , J ) is quasi-coherent.
Also note that the compatibility result mentioned above is not really needed in
our situation where F is locally free of nite type, since it is trivial then.
Now, HomOY ( 1X/S , J ) being quasi-coherent, we can conclude from 7.6/4
that the rst Cech cohomolgy of this sheaf is trivial and it follows that the
cocycle obtained from the above local liftings i : Yi - X can be resolved.
Therefore these morphisms can be modied by means of derivations as in 8.1/8
to yield local liftings i : Yi - X of that coincide on all overlaps of the cover-
ing (Yi )iI . Thus, they can be glued to dene a global S-morphism : Y -X
providing the desired lifting of .
It remains to show that the surjectivity of in (ii) implies the smoothness

of f . Since condition (ii) is maintained when shrinking X and S, we may work
locally on X. Furthermore, since f is locally of nite presentation, we can assume
that X has been realized as a closed subscheme of some ane n-space AnS , given
by a quasi-coherent sheaf of ideals I OAnS of locally nite type. Then it is
enough to show that the exact sequence from 8.2/5 extends to a short exact
sequence
0 - I/I 2 - 1AnS /S OAn OX - 1X/S - 0

S
that is locally split. Indeed, as it is locally split, the sequence remains exact when
tensoring it with the residue eld k(x) at any point x X. Therefore, choosing
local sections gnr , . . . , gn of I at x such that their images form a basis of the
k(x)-vector space I/I 2 k(x), Nakayamas Lemma in the version of 1.4/11
shows that these elements generate I at x. Furthermore, the residue classes
dgnr (x), . . . , dgn (x) are linearly independent in 1An /S k(x) by construction
S
so that f is smooth of relative dimension r at x.
We would like to establish the above locally split exact sequence more gen-
erally for a smooth S-scheme W in place of AnS . To do this, we assume S, X,
and W to be ane, say S = Spec R, X = Spec A, and W = Spec A where
A = A /I for some nitely generated ideal I A . Applying the surjectivity
of the map in (ii) to the identical morphism id : X - X, we see that the
identical map
= id : A /I - A /I = (A /I2 )/(I/I2 )
admits a morphism of R-algebras
: A /I - A /I2
as a lifting. Therefore the canonical exact sequence of R-modules

0 - I/I2 - A /I2 - A /I - 0,
is split, since may be viewed as an R-linear section of . Furthermore, the

dierence idA /I2 gives rise to a retraction
: A /I2 - I/I2
of the inclusion . Since idA /I2 and are R-algebra morphisms on A /I2
that are congruent modulo I/I2 by construction, we conclude from 8.1/8 that ,
being the dierence of these two maps, is an R-derivation. Composing with the
projection A - A /I2 , it induces an A -module morphism 1 - I/I2
A /R
mapping any dierential form of type da for a I to the residue class a I/I2
associated to a. Therefore 1A /R - I/I2 gives rise to a retraction of the
canonical morphism I/I 2 - A A occurring in the exact sequence of
1
A /R
8.1/13, and we see that the latter sequence extends to a short exact sequence
0 - I/I2 - 1A /R A A - 1A/R - 0
that is split. This settles the implication (ii) = (i) for smooth morphisms.
Finally, the remaining equivalence between (i) and (ii) for etale morphisms
follows with the help of Proposition 6 by combining the corresponding equiva-
lences for unramied and smooth morphisms.
We can show now as a consequence of the Lifting Property of Theorem 8

that the characterizing property for a smooth morphism in Denition 1 is in-
dependent of the choice of the immersion U - W An and that, moreover,
S
the ane n-space AS and its open part W may be replaced by any smooth
n
S-scheme Z.
Proposition 9 (Jacobian Criterion). Let j : X - Z be a closed S-immersion

of S-schemes where j is locally of nite presentation, and let I OZ be the
associated quasi-coherent ideal. Consider a point x X such that Z, as an
S-scheme, is smooth at z = j(x) of relative dimension n. Then the following
(i) X is smooth over S at x of relative dimension r.
(ii) There exists an open neighborhood of x, on which the exact sequence of
8.2/5 yields a short exact sequence
0 - I/I 2 - j 1 - 1X/S - 0
Z/S
that is split. Furthermore, dimk(z) 1X/S k(z) = r.

(iii) Let t1 , . . . , tn and g1 , . . . , gN be local sections in OZ,z such that the as-
sociated dierential forms dt1 , . . . , dtn give rise to a free generating system of
1Z/S,z and g1 , . . . , gN generate the ideal Iz . Then, after a suitable renumbering
of the elements ti and gj , we can assume that gr+1 , . . . , gn generate I and the
dierential forms dt1 , . . . , dtr , dgr+1 , . . . , dgn generate the module of dierential
forms 1Z/S on an open neighborhood of z.
(iv) There exist local sections gr+1 , . . . , gn OZ,z generating Iz such that
dgr+1 (z), . . . , dgn (z) are linearly independent in 1Z/S k(z).
Proof. The implication (i) = (ii) is a consequence of Theorem 8 and its proof.
Indeed, if X is smooth at x of relative dimension r, then the structural morphism
f: X - S satises the Lifting Property on some open neighborhood of x.
As shown in the last step of the proof of Theorem 8, the exact sequence from
8.2/5 extends locally at x to the split exact sequence of (ii) and and we see that
dimk(z) 1X/S k(z) = r using Proposition 5.
As for (ii) = (iii), use the fact that the exact sequence from (ii) remains
exact when tensoring it with the residue eld k(x) = k(z), since it is split locally
at x. Then (iii) is easily obtained with the help of 8.4/2. Furthermore, since 1Z/S
is a locally free OZ -module of rank n by Proposition 5 and since the latter is
generated at z by the dierential forms attached to local sections of OZ,z (see
8.1/6), we can read from 8.4/2 that local sections t1 , . . . , tn OZ,z as in (ii) will
always exist. Therefore the implication from (iii) to (iv) is straightforward.
Finally, the implication (iv) = (i) is clear if, for some open neighbor-
hood V Z of z, there would exist an open S-immersion V - An . As we
S
cannot assume this in the general case, we must rely on the smoothness of Z
at z and choose an open neighborhood V Z of z together with a closed
S-immersion j : V - W Am into an open subscheme W of some ane
S
m-space AS such that the dening conditions for Z to be smooth at z are sat-
m
ised. Thus, if I OW is the quasi-coherent ideal attached to j , there exist

local generators gn+1 , . . . , gm of I at z = j (z) such that the residue classes
dgn+1 (z ), . . . , dgm (z ) are linearly independent in 1W/S k(z ). Now replace

Z by V and X by j1 (V ) so that we may assume V = Z, and choose local

sections gr+1 , . . . , gn in OW extending the sections gr+1 , . . . , gn to some open
neighborhood of z . Then we see:

(a) The sections gr+1 , . . . , gm generate the quasi-coherent ideal attached to
the closed immersion j j : X -Z - W at z .
(b) dgr+1 (z ), . . . , dgm (z ) are linearly independent in 1W/S k(z ).

Assertion (a) follows from the fact that for two surjective ring morphisms
p : A1 - A2 and q : A2 - A3 the kernel of the composition q p is generated
by ker p and arbitrarily chosen p-preimages of generators of ker q. In a similar
way one obtains (b), just tensoring the exact sequence
I /I 2 - j 1
W/S
- 1Z/S - 0
with k(z) = k(z ) and using the properties of the sections gi . Together, (a) and
(b) say that X is smooth at x.
In particular, we can read from conditions (iii) or (iv) that the characterizing
property for X to be smooth of relative dimension r at a point x X is
independent of the choice of the closed S-immersion j : U - W An where
S
U X is an open neighborhood of x and W an open subscheme of some ane
n-space AnS . We may even assume W = AnS for some open subscheme S S
now, as suggested from the fact that X should be locally of nite presentation
at x. However, note that this simpler version is not suitable for introducing the
notion of smoothness in Denition 1, since it does not allow shrinking of X in
a exible way.
Also let us recall from the beginning of this section that the condition on
the linear independence of the residue classes dgr+1 (z), . . . , dgn (z) 1Z/S k(z)
in (iv) can be checked via looking at the Jacobian matrix. Indeed, we know
from Proposition 5 that 1Z/S,z is free of rank n if Z is smooth at z of relative
dimension n. Thus, if dt1 , . . . , dtn are free generators of 1Z/S,z , then for any
given sections gr+1 , . . . , gn OZ,z there are equations

n
dgj = aij dti , j = r + 1, . . . , n,
i=1
with unique coecients aij OZ,z , and the residue classes

n
dgj (z) = aij (z)dti (z), j = r + 1, . . . , n,
i=1
are linearly independent in 1Z/S k(z) if and only if rg(aij (z)) = n r.
Proposition 10. Let f : X - Y be a morphism of S-schemes where X is

smooth at a point x X and Y is smooth at y = f (x). Then the following
(i) f is smooth at x.
(ii) The canonical morphism : (f 1Y /S )x - 1X/S,x is left invertible, i.e.
there is a retraction : 1X/S,x - (f Y /S )x satisfying = id.
id
(iii) The canonical morphism (f 1Y /S ) k(x) - 1X/S k(x) is injective.
Proof. Starting with the implication (i) = (ii), consider the exact sequence

f 1Y /S - 1X/S - 1X/Y - 0
from 8.2/4. If f is smooth at x, then all occurring OX -modules are locally free
of nite type at x, due to Proposition 5, and we have
rg 1X/S,x = rg f (1Y /S ),x + rg 1X/Y,x ,
since relative dimensions are added when composing smooth morphisms. Fur-
thermore, the exact sequence
0 - im - 1X/S - 1X/Y - 0
splits at x since 1X/Y,x is free. Lifting a k(x)-basis of im k(x), we can proceed

as in the proof of Proposition 5 and see with the help of Nakayamas Lemma
that im is locally free of nite type at x. But then the exact sequence
0 - ker - f 1Y /S - im - 0
shows in the same way that ker is locally free of nite type at x as well. Now
observe that the above rank formula yields rg(ker )x = 0. Therefore we get the
exact sequence
0 - (f 1Y /S )x - 1X/S,x - 1X/Y,x - 0,
which splits, since 1X/Y,x is free. Thus, (ii) is clear.

Next, (ii) = (iii) is valid for trivial reasons. To show (iii) = (i), we start
with the special case where Y = AnS . Let t1 , . . . , tn be the coordinate functions
of AnS and let g i = f # (ti ) for i = 1, . . . , n. Then we see from (iii) that the residue
classes dg 1 (x), . . . , dg n (x) are a linearly independent in 1X/S k(x). Now use the
fact that X is smooth at x of some relative dimension r. Shrinking X and S if
necessary and taking them ane, we can nd a closed immersion j : X - Am S.

By the Jacobian Criterion we can assume that the corresponding ideal I OAm S
is generated by global sections hr+1 , . . . , hm , whose associated dierential forms
give rise to a set of linearly independent residue classes dhr+1 (z), . . . , dhm (z) in
1Am /S k(z) where z = j(x). Now look at the morphism
S
idX f
- X S Y
jidY
- Am
S S AS = AY ,
n m
(j, f ) : X
which is a composition of closed immersions; use 7.4/13 in conjunction with

7.4/8, as well as 7.3/13. Choosing extensions g1 , . . . , gn OAm
S
(Am
S ) of g 1 , . . . , g n ,
the m (r n) elements
g 1 t 1 , . . . , gn t n , hr+1 , . . . , hm
will generate the quasi-coherent ideal in OAm

Y
corresponding to the above closed
immersion (j, f ). Since the dierential forms dt1 , . . . , dtn 1Am /Y are trivial,
Y
the dierential forms attached to the above functions are simply
dg1 , . . . , dgn , dhr+1 , . . . , dhm 1Am

Y /Y
,
and we may interpret these as pull-backs of the corresponding dierential forms

in 1Am /S if we use the structural morphism Y - S as base change. Then, by
S
8.2/3, it is enough to show that the residue classes
dg1 (z), . . . , dgn (z), dhr+1 (z), . . . , dhm (z) 1Am

S /S
k(z)
are linearly independent. To verify this, use the fact that X is smooth at x.
Thus, applying the Jacobian Criterion of Proposition 9 yields a sequence
0 - I/I 2 - j 1 m - 1X/S - 0,
AS /S
which is exact and split at x. In particular, the sequence remains split exact
if we tensor it with the residue eld k(x) = k(z). Now observe that by our
assumption (iii) the images of dg1 (z), . . . , dgn (z) in 1X/S k(x) are linearly
independent. Moreover, dhr+1 (z), . . . , dhm (z) are linearly independent by con-
struction. Since the latter residue classes are induced from sections in I, the
exactness of the above sequence, tensored with k(x), shows that the whole sys-
tem is linearly independent. Thus, it follows that f : X - Y is smooth at x
of relative dimension r n.
To deal with the general case, assume X and Y to be smooth at x, resp. y,
namely, of relative dimension r, resp. s. Let h1 , . . . , hs be local sections in OY ,
whose attached dierential forms dh1 , . . . , dhs induce a basis of 1Y /S k(y).
Shrinking X and Y if necessary, we may assume that the hi are global sections
in OY . Then we see from (iii) that there exist local sections gs+1 , . . . , gr in OX
such that
(f dh1 )(x), . . . , (f dhs )(x), dgs+1 (x), . . . , dgr (x)

is a basis of 1X/S k(x). Taking X small enough, we may assume that

gs+1 , . . . , gr are global sections in OX . Then the S-morphisms
g = (gs+1 , . . . , gr ) : X -Ars
S ,
h = (h1 , . . . , hs ) : Y - As
S
induce a commutative diagram

(f,g) p
X - Y S Ars - Y
S
hid
(hf,g) - ?
ArS
where p is the projection onto the rst factor. By the special case dealt with
above, we see that (h f, g) is etale at x and, likewise, that h is etale at y. Then,
by a base change argument, h id will be etale at all points in p1 (y), and it
follows from Lemma 11 below that (f, g) is etale at x. Since f = p (f, g) is a
composition of two smooth morphisms, it is smooth and we are done.
Lemma 11. Let f : X - Y and g : Y - Z be morphisms of schemes

where g is unramied. Then, if g f is unramied (resp. smooth, resp. etale),
the same is true for f .
Proof. Let us rst show that f is locally of nite presentation. To do this, look
at the cartesian diagram
(id,f )
X - X Z Y
f f id
?
?
Y - Y Z Y
derived from 7.4/11 where the morphisms
1 : X - S, 2 : Y - S, S - T
have been taken as
f: X - Y, id : Y - Y, g: Y - Z,
and where is the diagonal morphism attached to g. Since g is locally of nite

type, we can read from 8.1/7 (ii) that is locally of nite presentation. This
property remains unchanged under base change; thus, (id, f ) : X - X Z Y
is locally of nite presentation as well. It follows that f , as a composition of
(id, f ) with the projection p : X Z Y - Y is locally of nite presentation,
because p is obtained from g f via base change with g : Y - Z.
To see that f is unramied, resp. smooth, resp. etale, provided g f admits
this property, we use the Lifting Property from Theorem 8. Let T be a Y -scheme
that is ane and let T T be a closed subscheme, given by a quasi-coherent

ideal I OT satisfying I 2 = 0. Then, if : T - X is a Y -morphism
and : T - X a Z-morphism lifting , it follows that is necessarily a
Y -morphism, since the canonical map
HomZ (T, Y ) - HomZ (T , Y )
is injective, due to the fact that g is unramied. Thereby we can derive without
problems the desired Lifting Property for f if the corresponding one for g f is
known.
Let us formulate Proposition 10 especially for the case of an etale morphism

f: X - Y . Since relative dimensions are added when composing smooth
morphisms, we see with the help of Proposition 5:
Corollary 12. Let f : X - Y be a morphism of S-schemes where X is

smooth at a point x X and Y smooth at y = f (x). Then the following
(i) f is etale at x.
(ii) The canonical morphism (f Y /S )x - 1X/S,x is an isomorphism.
The assertion of the corollary suggests to view etale morphisms as certain

analogues of maps that, for example in the setting of Differential Geome-
try, satisfy the assumptions of the Inverse Function Theorem and thereby are
locally invertible. However, note that etale morphisms are not locally invertible
in Algebraic Geometry, at least not in the general case. Such a behavior
can only be put into eect if the Zariski topology is replaced by the more general
concept of the etale topology. Furthermore, in Differential Geometry the
Implicit Function Theorem suggests that smooth morphisms may be viewed as
local brations by open subsets of ane n-spaces. The same is true in Alge-
braic Geometry if we localize the category of schemes by etale morphisms
in the sense of formally viewing the latter as being invertible.
Proposition 13. For a morphism f : X - S and a point x X the following

(i) f is smooth at x of relative dimension r.
(ii) There exists an open neighborhood U X of x and a commutative
diagram
g
U - ArS
p
f |U
?
-
S,
where g is etale and p is the structural morphism.
Proof. Since the composition of smooth morphisms is smooth again, condition (i)
is a consequence of (ii). Conversely, if f is smooth at x of relative dimension
r, the module of dierential forms 1X/S is locally free at x of rank r and, as
exercised in the proof of Proposition 5, there are local sections g1 , . . . , gr in OX
such that the attached dierential forms dg1 , . . . , dgr generate 1X/S freely at x.
Then g1 , . . . , gr dene an S-morphism g : U - ArS on some open neighborhood
U X of x and Corollary 12 shows that g is etale at x. Using Proposition 3,
we may assume that g is etale on U .
Proposition 14 (Existence of etale quasi-sections). For a smooth morphism

f: X - S and a point s S consider a closed point x Xs of the ber
Xs = X S k(s) of f over s such that the extension k(x)/k(s) is separable.
Then there exists an etale morphism g : S - S together with a point s over
s such that the morphism f : X S S - S obtained from f via base change
with g admits a section h : S - X S S where h(s ) lies over x and satises
k(h(s )) = k(x).
Proof. Let f be smooth at x of relative dimension r and let J OXs be the

quasi-coherent ideal of all functions vanishing at x. Since f is locally of nite
type, the extension k(x)/k(s) is even nite by 3.2/4 and the corresponding
morphism Spec k(x) - Spec k(s) is etale. Viewing Spec k(x) as a closed
subscheme of Xs and using that Xs is smooth over k(s) of relative dimension r,
we conclude from the Jacobian Criterion of Proposition 9 that there exist local
generators g 1 , .., g r of the ideal Jx such that their attached dierential forms
dg 1 , . . . , dg r give rise to a basis of 1Xs /k(s) = 1X/S k(s) at x.
Now lift g 1 , .., g r to local sections g1 , . . . , gr in OX , dened on some open
neighborhood U X of x. Let S be the closed subscheme of U dened by
g1 , . . . , gr . Then we see from the Jacobian Criterion again that g : S - S is

etale at x. Shrinking S if necessary, we may assume by Proposition 3 that g is
etale at all points of S . But then the canonical morphism h : S - X S S

is a section of f : X S S -
S having the desired properties.
The assumption in the above proposition that the ber Xs contains a closed
point x with a separable residue extension k(x)/k(s) is always fullled, provided
Xs is non-empty. Indeed, using Proposition 13 one can show for a smooth scheme
X over a eld k that the set of its closed points x X with separable residue
extension k(x)/k(s) is dense in X.
Next we want to characterize smooth schemes over elds in terms of regu-
larity. Recall from 2.4/18 that a local Noetherian ring A with maximal ideal m
is called regular if m can be generated by dim A elements. Likewise, a locally
Noetherian scheme X is called regular at a point x X if the local ring OX,x is
regular. If OX,x is regular for all points x X, the scheme X is called regular
or non-singular. Since any localization Ap of a regular Noetherian local ring A
by a prime ideal p A is regular again (see Serre [24], Prop. IV.23), it follows
that a locally Noetherian scheme X is regular if and only if it is regular at all
its locally closed points, i.e. at all points that are closed in some open part of
X. Also note that locally closed points are automatically closed if X is locally
of nite type over a eld; see 8.3/6.
Proposition 15. Let X be locally of nite type over a eld k. Then, for a point
x X, the following conditions are equivalent:
(i) X is smooth over k at x.
(ii) There is an open neighborhood U X of x such that U k k is regular
for all eld extensions k /k.
(iii) There is an open neighborhood U X of x and an extension of elds
k /k such that k is perfect and U k k is regular.
(iv) 1X/k,x is generated by dimx X elements.
Proof. Note that X is locally Noetherian by 1.5/14, since it is locally of nite

type over a eld. To show that (i) implies (ii), we apply Proposition 13 and
consider an etale (and, hence, unramied) morphism g : U - Ar , dened on
k
an open neighborhood U X of x. Then we conclude from 8.4/3 for each closed
point u U that the maximal ideal mu OX,u is generated by the maximal
ideal mg(u) OArk ,g(u) and therefore by r elements, since the localization of a
polynomial ring kt1 , . . . , tr at a maximal ideal is regular local ring of dimension
r; see 2.4/17. Furthermore, as U is smooth over k of relative dimension r, we get
dim OX,u = dimu X = r from Proposition 4 so that OX,u is regular by 2.4/18.
But then, since U is locally of nite type over a eld, all its locally closed points
are closed by 8.3/6 and we see that U is regular. The same consideration can be
carried out after base change with any eld extension k /k, which establishes
(ii).
The next step from (ii) to (iii) is trivial. Therefore assume (iii). In order to
derive (iv), we may assume k = k and, hence, that k is perfect. Indeed, if we
can establish the equivalences of the proposition for k in place of k, then (iii)
will imply that X k k is smooth over k at all points lying over x. From this
we can conclude via condition (ii) of the Jacobian Criterion that X is smooth
over k at x, say of relative dimension r. But then 1X/k,x is a free OX,x -module
of rank r by Proposition 5 where r = dimx X by Proposition 4. Consequently,
we are reduced to the case where k is perfect.
Next, observe that all local rings at points u U are integral domains
by 2.4/19, because these are regular local rings. Since local rings belonging to
the intersection of two dierent irreducible components of U cannot be integral
domains, U must be the disjoint union of its irreducible components. Restricting
U to one of these, we may assume that U is irreducible. Then 3.3/8 implies
dimx X = dim OX,u for every closed point u U . Moreover, by the same result,
it is enough to prove that 1X/k,u is generated by dim OX,u elements for all closed
points u U .
Let u U be such a closed point. As k is perfect and X is of locally nite
type over k, the eld k(u) is a nite separable extension of k so that 1k(u)/k = 0
by 8.4/3. Hence, we get an exact sequence
mu /m2u - 1X/k,u k(u) - 1k(u)/k = 0
from 8.1/13. Since X is regular at u, we have dimk(u) mu /m2u = dim OX,u and
(iv) follows from 8.4/2.
It remains to derive (i) from (iv). Since X is locally of nite presentation at
x, there is an open neighborhood U X of x together with a closed immersion
j : U - W Ank where the associated quasi-coherent ideal I OW is locally
of nite type. Look at the exact sequence
I/I 2 - j 1 - 1U/k - 0
W/k
of 8.2/5 and let r = dimx X. Then dimk(x) 1U/k k(x) r by (iv). Thus, using
Nakayamas Lemma in the version of 1.4/11 and observing (iv), there exist local
sections gr+1 , . . . , gn in I at x such that the corresponding dierential forms
dgr+1 , . . . , dgn generate a free direct factor in 1W/k,x of rank n r. Shrinking W
(as well as its closed subscheme U ) if necessary, we may assume that gr+1 , . . . , gn
are dened on all of W and, thus, give rise to a closed subscheme U W where
U U . We may even assume that U is smooth over k of relative dimension r
at all its points. Since r = dimx U by Proposition 4, we see that U is a closed
subscheme in U satisfying dimx U = dimx U . Now let u U be a closed point
that is a specialization of x. Since the implication (i) = (ii) has already been
established, the local ring OU ,u is regular and, hence, by 2.4/19, an integral
domain. Therefore, using 3.3/8, the local dimension dimx U must coincide with
dim OU ,u and we have
dim OU,u dimx U = dimx U = dim OU ,u .
However, the surjection : OU ,u - OU,u corresponding to the closed im-

mersion U -
U , in conjunction with OU ,u being an integral domain shows
dim OU,u < dim OU ,u if has a non-trivial kernel; see 2.4/14. Consequently,
will be an isomorphism and U X is smooth at x over k.
For example, one can read from conditions (iii) and (iv) in Proposition 15
that X is smooth if and only if X is geometrically regular in the sense that
X k k is regular for an algebraic closure k of k. However, note that a locally
Noetherian k-scheme that is regular or, in other terms, non-singular, may fail
to be geometrically regular and therefore might not be smooth. Just look at a
k-scheme Spec k , where k /k is a nite extension of elds that is not separable.
There is a slightly weaker notion: a k-scheme X is called geometrically reduced
if all stalks of the structure sheaf of X k k are reduced in the sense that they
do not contain (non-trivial) nilpotent elements.
Corollary 16. Let X be locally of nite type over a eld k. If X is geometrically

reduced, the smooth locus of X is open and dense in X.
Proof. The smooth locus is open in X by Proposition 3. To show that it is dense

as well, we must show that X is smooth at all its generic points. To do so, let x
be a generic point of X. Restricting X to an open neighborhood of x, we may

assume that X is ane, say X = Spec A. Then A is of nite type over k due
to 8.3/5 and, hence, Noetherian by 1.5/14. Assuming that x is the only generic
point of X, we see using 6.4/7 that A is reduced and, hence, an integral domain
with eld of fractions k(x). In particular, the local dimension dimx X equals
the Krull dimension d = dim A by 3.3/8 and it follows from 3.3/7 that d is the
transcendence degree of the extension of elds k(x)/k.
By our assumption, k(x) k k is reduced for an algebraic closure k of k. In
other words, the extension k(x)/k is separable; cf. [3], 7.3/2 and 7.3/7. But then
we can use [3], 7.4/11, to see that 1k(x)/k is a k(x)-vector space of dimension
d, the transcendence degree of k(x)/k. Since 1X/k k(x) 1k(x)/k by 8.1/11
and since d coincides with the local dimension of X at x, Proposition 15 in
conjunction with 8.4/2 shows that X is smooth at x.
In most of the literature, smoothness is introduced via the so-called Fibre

Criterion. Basically this criterion says that the smoothness of a morphism of
schemes f : X - S can be checked berwise, provided f is at. The latter
means that all local morphisms fx# : OS,f (x) - OX,x at points x X are at
in the sense that OX,x is a at OS,f (x) -module via fx# . This clearly is a local
notion, but also note that a morphism of ane schemes Spec B - Spec A
is at if and only if the corresponding ring homomorphism A - B is at;
use 4.3/5. However, we would like to point out that the concept of atness for
scheme morphisms is much more complicated than the notion of smoothness in
terms of the Jacobian Condition. It is for this reason that we did not base the
denition of smoothness on the Fibre Criterion.
Proposition 17 (Fibre Criterion for Smoothness). Let f : X - S be locally

of nite presentation. Then the following conditions are equivalent for a point
x X and its image s = f (x):
(i) f is smooth at x.
(ii) f is at at x and the ber Xs = X S k(s) is smooth at x over k(s).
Proof. To derive (ii) from (i), we have only to show that the smoothness of f at
x implies that f is at at x. For this we can assume that X and S are ane, say
X = Spec A and S = Spec R and, furthermore, that R is Noetherian. In order to
reduce to the Noetherian assumption, we consider a situation as in Denition 1,
namely a closed immersion j : U - W An from an open neighborhood
S
U X of x into an open subscheme W of some ane n-space AnS . Shrinking
X, we can assume U = X and, in addition, that W is basic open in AnS , say
W = D(h) where h Rt1 , . . . , tn is a global section in the structure sheaf of
AnS . Furthermore, if f is smooth of relative dimension r at x, we can assume
that there are polynomials gr+1 , . . . gn Rt1 , . . . , tn globally generating the
quasi-coherent ideal I OW associated to the closed immersion j. In addition,
we may assume that there are polynomials g1 , . . . , gr Rt1 , . . . , tn such that
the dierential forms dg1 , . . . , dgn generate 1An /S at all points of W . In partic-
S
ular, this way f is assumed to be smooth of relative dimension r at all points of

X. Now choose a Z-subalgebra of nite type R R, large enough to contain
all coecients of the polynomials h, g1 , . . . , gn necessary to describe the above
situation, including some auxiliary polynomials that are used to relate the dif-
ferential forms dgi to the canonical ones dti on W . Then j : X - W An
R

naturally descends to a closed R -immersion X -
W AR showing that
n
f: X - Spec R is obtained via base change with R/R from a smooth mor-
phism f : X
- Spec R . If we can show that f is at on X , it will follow
from arguments of base change like 4.4/1 that f is at on X = X R R and,
in particular, at x. Thereby we are reduced to the case where the base ring R
is a Z-algebra of nite type and, hence, is Noetherian by 1.5/14.
Now assume S = Spec R with R being Noetherian. Starting out from an
auxiliary scheme Z = AnS and using induction, it is enough for the implication
(i) = (ii) to consider an ane situation where Z is a smooth S-scheme that is
already known to be at over S, and where X is a closed subscheme of Z, dened
by a single section g OZ such that the residue class dg(x) 1Z/S k(x) is
non-trivial and, hence, f : X - S is smooth at x. The multiplication by g
denes a morphism of OZ,x -modules : OZ,x - OZ,x giving rise to the short
exact sequence
0 - OZ,x / ker - OZ,x - OX,x - 0.
The latter remains exact when tensoring it over OS,s with the residue eld k(s).
Indeed, since the ber Zs is smooth over k(s) at x, we see that OZ,x k(s)
is regular by Proposition 15 and, hence, an integral domain by 2.4/19. As g
induces a non-zero element in OZ,x k(s), the multiplication by g yields a short
exact sequence
0 - OZ,x k(s) - OZ,x k(s) - OX,x k(s) - 0,
which we may interpret as being obtained from the former one by tensoring
with k(s) over OS,s . But then the long Tor sequence of 5.2/2 yields
O
Tor1 S,s (OX,x , k(s)) = 0,
O
due to the fact that Z is at over S and, hence, Tor1 S,s (OZ,x , k(s)) = 0 by
5.2/7. Therefore we can conclude from Bourbakis Criterion [6], III, 5, no. 2,
Thm. 1 (iii) with J the maximal ideal of OS,s , that OX,x is at over OS,s and,
hence, f : X - S is at at x.
To pass from (ii) to (i), we may assume that S is ane, say S = Spec R, and
that X is a closed subscheme of some ane n-space AnR , say given by a nitely
generated ideal I Rt1 , . . . , tn . Using the fact that the ber Xs is smooth at
x of some relative dimension r, the Jacobian Criterion of Proposition 9 shows
there are elements gr+1 , . . . , gn I such that, locally at x, the induced elements
g r+1 , . . . , g n dene Xs as a closed subscheme of Ank(s) and the residue classes
dg r+1 (x), . . . , dg n (x) are linearly independent in 1An /k(s) k(x) where the
k(s)
latter coincides with 1An /R k(x) by 8.1/11.

R
Now let X be the closed subscheme in AnR that is dened by the ideal
J = (gr+1 , . . . , gn ). Then X is smooth at x of relative dimension r and J I
implies X X . Furthermore, the bers Xs and Xs coincide locally at x. Since
X is at over S at x, the short exact sequence
0 - I/J - Rt1 , . . . , tn /J - Rt1 , . . . , tn /I - 0
remains exact at x when tensoring it over R with k(s); use 5.2/9. Therefore
J/I R k(s) vanishes at x and Nakayamas Lemma 4.10 yields (J/I)x = 0.
Thus, X and X coincide locally at x and we see that X is smooth over S at x
since this is true for X .
We have already pointed out that a nite separable extension of elds gives
rise to an etale morphism between associated schemes. Thus, we conclude from
8.4/4 that a scheme morphism f : X - S with S the spectrum of a eld is
etale at a point x X if and only if it is unramied at x. Keeping this in mind,
the Fibre Criterion for smooth morphisms of Proposition 17 in conjunction with
the one for unramied morphisms of 8.4/3 shows:
Corollary 18. A morphism f : X - S is etale at a point x X if and only

if it is at and unramied at x.
Furthermore, we see with the help of Proposition 15 that a morphism of

schemes f : X - S is smooth if and only if it is locally of nite presentation,
at, and has geometrically regular bers. Such a characterization is often used
in order to get a rapid denition of smooth morphisms.
Also let us mention the fact that every at morphism f : X - S that is
locally of nite presentation, is open in the sense that it maps open subsets of
X onto open subsets of S. Therefore smooth and etale morphisms are open; see
EGA [14], IV, 2.4.6.
Finally, let us refer to the local structure of etale morphisms; see Raynaud
[23], V. For example, relying on this result, it is immediately clear that etale
morphisms are at. In conjunction with Proposition 13 the same follows for
smooth morphisms. This may replace the direct argument we have used to
prove the Fibre Criterion of Proposition 17.
Proposition 19. Let f : X - Y be a morphism of schemes that is etale at a

point x X. Then there exists an ane open neighborhood : Spec A - Y
of f (x) together with a commutative diagram

Spec(At/(p))q - X
can f
?
?
Spec A - Y ,
where p, q At are polynomials in one variable t such that p is monic and the
derivative p of p is invertible on Spec(At/(p))q . Furthermore, is an open
immersion and can denotes the canonical morphism.
The proof is based on a famous result of Zariski (see EGA [14], IV, 18.12.13):
Theorem 20 (Zariskis Main Theorem). Let f : X - Y be quasi-nite and

separated and Y quasi-compact and quasi-separated. Then there exists a com-
mutative diagram
g
X -Z
f - h
Y
where g is an open immersion and h is nite.
Note that a morphism f : X - Y is called quasi-nite if for every y Y

the ber Xy carries the discrete topology. Furthermore, f is called nite (see
9.5/1 and 9.5/3) if for every ane open subscheme V Y the preimage f 1 (V )
is ane and the morphism f # (V ) : OY (V ) - OX (f 1 (V )) is nite in the
1
sense that OX (f (V )) is a nite OY (V )-module via f # (V ).
Exercises
1. Show that smooth morphisms are stable under base change, composition, and the
formation of ber products. Compute relative dimensions in these cases. Hint:
For compositions use the argument in the proof of 8.3/3 (ii) showing that the
composition of two morphisms of locally nite presentation is of locally nite
presentation again.
2. Let A be a smooth algebra over a ring R, in the sense that the associated scheme
morphism Spec A - Spec R is smooth. For a multiplicatively closed subset
S A, give a condition assuring that the localization AS is a smooth R-algebra
as well.
3. As in Exercise 8.4/2 consider the scheme morphism f : X - S associated to
the canonical K-morphism Kt1 , t2 /(t1 t2 ) - Kt1 Kt2 for variables t1 , t2
over a eld K. Show that f is not etale on all of X, although it is unramied.
Determine the etale locus of f .
4. Consider Neiles parabola X = Spec Kt1 , t2 /(t22 t31 ) over a eld K as in Exer-
cise 6.7/3. Determine the smooth locus of X over K.
5. A morphism of schemes f : X - S is called formally unramied (resp. for-
mally smooth, resp. formally etale) if f satises the corresponding lifting prop-
erty, i.e. if for every S-scheme Y where Y is ane and for every closed subscheme
Y Y given by a quasi-coherent ideal J OY of square J 2 = 0, the canonical
map HomS (Y, X) - HomS (Y , X) is injective (resp. surjective, resp. bijective).
Show that such morphisms are stable under base change, composition, and the
formation of ber products.
6. Let X be a scheme that is smooth of relative dimension 1 over a eld. Show that
all local rings OX,x at closed points x X are discrete valuation rings.
7. Let K = k(t) be the function eld in a variable t over a eld k of characteristic
p > 2. Consider the curve X given by the equation y 2 = xp t in the ane plane
A2K , i.e. X = Spec Kt1 , t2 /(t22 tp1 + t) for variables t1 , t2 and show that X is
regular, but not smooth over K. Determine its smooth locus.
8. Let f : X - S be a smooth morphism. If S is regular, show that X is regular
as well. In particular, conclude for any regular ring R, in the sense that all lo-
calizations Rp by prime ideals p R are regular Noetherian local rings, that the
polynomial ring RX1 , . . . , Xn in nitely many variables X1 , . . . , Xn is regular
as well.
9. Consider a smooth scheme X over a eld k. Show that the set of closed points
x X such that the residue extension k(x)/k is separable, is dense in X. Hint:
Reduce the assertion via Proposition 13 to the case where X = Ank . Do this either
by using that at and, hence, etale morphisms are open, or avoid this result by
a weaker direct argument.
10. Descent of smoothness: Let : S - S be a faithfully at morphism of schemes,
in the sense that is at and surjective, and f : X - S a morphism that is lo-
cally of nite presentation. Show that f is smooth if and only if the corresponding
morphism f idS : X S S - S obtained by base change with is smooth.
Hint: Use the results of Section 4.4.
11. Group schemes over a eld : Let G be a k-group scheme of locally nite type over
a eld k; see Section 9.6 for the notion of group schemes. Show that G is smooth
over k if and only if it is geometrically reduced. Give an example of a eld k and
of a k-group scheme of nite type that is reduced, but not smooth.
12. Henselization of a local ring: Fix a local ring R and set S = Spec R. A local
R-algebra is a local ring R with a structural morphism R - R that is lo-
cal. Furthermore, R is called an essentially etale R-algebra if there exists an
etale morphism S - S with a point s S lying over the closed point
s S such that R - R coincides with the inherent morphism between stalks
OS,s - OS ,s . Show for an essentially etale local R-algebra R and any local
R-algebra A that the canonical map HomR (R , A) - Homk (k , kA ) is bijec-
tive where k, k , and kA are the residue elds of R, R , and A. Conclude that
the inductive limit Rh over all essentially etale local R-algebras exists and is a
local R-algebra with residue eld k. It is called the henselization of R and is
characterized by the fact that every etale morphism X - Spec Rh is a local
isomorphism at all points x X over s where the residue extension k(x)/k is
trivial.
9. Projective Schemes and Proper Morphisms
The notion of compactness is fundamental in Topology and far beyond. For

example, given a complex analytic space or manifold, one wishes to construct
a compact closure of it, a so-called compactication. The projective n-space PnC
for instance may be viewed as a compactication of the ane n-space AnC . Quite
often suitable compactications can make the original object more accessible.
A literal translation of the compactness condition to the scheme case leads
to quasi-compactness, as schemes do not satisfy the Hausdor separation axiom.
In particular, all ane schemes, including the ane n-space over an ane base,
are quasi-compact. However, this clearly indicates that quasi-compact schemes
will not be the right analogues of compact spaces in analysis. Indeed, there is
a perfect adaptation of compactness to schemes, the notion of properness, which
will be introduced in 9.5/4. For example, the projective n-space PnS over some
base scheme S is proper, cf. 9.5/9, whereas the ane n-space AnS for n > 0 is not,
unless S is the empty scheme. We will deal with basic material on proper mor-
phisms in Section 9.5, including the valuative criteria on separatedness 9.5/17
and properness 9.5/18. Also we give references for several fundamental theo-
rems like Chows Lemma 9.5/10, the Proper Mapping Theorem 9.5/11, and the
Stein Factorization 9.5/12, whose proofs can readily be studied departing from
the knowledge of the present chapter.
If X is a relative scheme that is proper over its base S (assumed to be ane
in the following), one might ask if X is already projective in the sense that it
admits a closed immersion over S into some projective n-space PnS . In particular,
if the latter is the case, there are global coordinates on X, homogeneous though,
but which can nevertheless serve for a convenient description of X over S.
Due to Chows Lemma 9.5/10, proper schemes can always be dominated by
projective ones, but not all proper schemes are projective. In fact, it is one of
the guiding themes of the present chapter to study conditions assuring that a
given scheme X admits an immersion into the projective n-space PnS and, thus,
can be characterized in terms of homogeneous coordinates. The main criterion
for this is the existence of ample invertible sheaves on X; cf. 9.4/10.
Using an ad hoc gluing technique, the projective n-space PnS has already been
constructed in 7.1. However, to deal conveniently with projective spaces of more
general type, we study in 9.1 so-called Proj schemes attached to graded rings.
In principle, their construction is similar to the construction
of ane schemes,
although the details are quite dierent. So let A = nN An be a graded ring

400 9. Projective Schemes and Proper Morphisms

in the sense of 9.1/1 and write A+ = n>0 An for its so-called irrelevant ideal.
Then the homogeneous prime spectrum of A, denoted by Proj A, is the subset
of the whole prime spectrum Spec A consisting of all graded prime ideals that
do not contain the irrelevant ideal A+ ; see 9.1/2 for the notion of graded ideals.
The set Proj A comes equipped with a natural topology, the Zariski topology
inherited from the Zariski topology of Spec A. Furthermore, the sets of type
D+ (f ) = D(f ) Proj A
for basic open subsets D(f ) Spec A and homogeneous elements f A form a
basis of the Zariski topology on Proj A; see 9.1/12. But most importantly, the
sets D+ (f ), which are referred to as the basic open subsets of Proj A, carry a
natural structure of an ane scheme, namely D+ (f ) = Spec A(f ) where A(f ) is
the homogeneous localization of A by f ; see 9.1/13. Gluing the schemes D+ (f )
along the intersections
D+ (f ) D+ (f ) = D+ (f f ) = Spec A(f f )
we obtain a scheme, namely the scheme Proj A attached to the graded ring A.
For example, we can view the polynomial ring Rt0 , . . . , tn in a set of variables
t0 , . . . , tn over a ring R as a graded ring or, better, as a graded R-algebra by
dening the grading via the total degree function. Then the corresponding Proj
scheme ProjR Rt0 , . . . , tn is just the projective n-space PnR .
From ane schemes we know that any morphism of rings B - A gives
rise to a morphism Spec A - Spec B between associated ane schemes.
A similar fact for morphisms of graded rings and Proj schemes is true only
in some very special situations; see 9.1/20 and 9.1/21 for results of this type.
However, working over some base ring R there is an ingenious way to character-
ize R-morphisms X - Pn from an arbitrary R-scheme X to the projective
R
n-space PR . Recall from 7.1/3 that for any R-scheme Y there is a canonical
n
bijection

HomR (Y, AnR ) - HomR Rt1 , . . . , tn , OY (Y ) OY (Y )n .
Thus, if we are given an R-morphism f : X - PnR , we can cover PnR by n + 1

copies Ui of AnR as in 7.1, say

n
n 't tn (
0
PnR = Ui = Spec R ,...,
i=0 i=0
ti ti
for a set of variables t0 , . . . , tn , and thereby describe each of the restricted mor-
phisms fi : f 1 (Ui ) - Ui by a set of n sections in OX (f 1 (Ui )), namely
the ones obtained via pull-back from the variables tt0i , . . . , ttni (neglecting the
constant ttii = 1) on Ui . The variables ttji satisfy some obvious relations on dou-
ble and triple intersections of the Ui and these are maintained on preimages.
Indeed, a careful analysis of the situation shows that the pull-backs of the vari-
ables tt0i , . . . , ttni can conveniently be denoted as fractions ss0i , . . . , ssni , although
s0 , . . . , sn do not make sense in terms of sections of OX , just as t0 , . . . , tn cannot

be interpreted as sections of the structure sheaf of PnR . However, t0 , . . . , tn make
sense as sections of Serres invertible sheaf OPnR (1), which we construct on PnR ;
use 9.2 and, in particular, 9.2/7 as a reference and note that an invertible sheaf
L on a scheme X is just an OX -module that, locally on X, is isomorphic to OX
itself. Then the pull-back L = f (OPnR (1)) is invertible on X and s0 , . . . , sn make
sense as sections of L, namely as the pull-backs of t0 , . . . , tn . Using such a set-
ting, we get the assertions of 9.4/4 and 9.4/5 showing how to set up a bijection
between R-morphisms X - PnR and equivalence classes of data (L, s0 , . . . , sn )
where L is an invertible sheaf on X and the si form a set of global generators
for L.
Let us mention along the way a very interesting application of Proj schemes,
namely the technique of blowing up, which is a key ingredient for resolving
singularities on schemes. For a ring A and an ideal I A we consider the
direct sum of A-modules A = d
d=0 I and view it as a graded A-algebra where
I d A is the dth power ideal of I and, of course, I 0 = A for d = 0. Then the
canonical scheme morphism : Proj A - Spec A is called the blow-up of the
closed subscheme Spec A/I - Spec A on Spec A. It is interesting to know
that the blow-up morphism is characterized by a universal property involving
invertible ideal sheaves. Namely, the ideal I A generates an invertible ideal
sheaf on Proj A and, furthermore, any scheme morphism Y - Spec A where
I generates an invertible ideal sheaf on Y admits a unique factorization through
: Proj A - Spec A; cf. Exercise 9.2/10.
As a special example, let us look at Neiles parabola X given by the equation
x22 x31 = 0 in the ane plane A2K over a eld K; cf. Exercise 6.2/6. Fixing
variables t, t1 , t2 , we have
X = Spec Kt1 , t2 /(t22 t31 )
and it is not hard to see that the K-algebra morphism
Kt1 , t2 - Kt, t1 - t2 , t2 - t3 ,
induces an identication of A = Kt1 , t2 /(t22 t31 ) with a subring of the poly-

nomial ring Kt, namely
i
A - ci t ; c1 = 0 Kt.
iN
In fact, Kt is the normalization of A, since t is integral over A and Kt is

normal itself; use 3.1/10. Now let : A1K - X be the morphism given by
the inclusion A - Kt. We know that is bijective, although it is not an
isomorphism. Indeed, writing 0 X for the point corresponding to the maximal
ideal m = (t1 , t2 ) = (t2 , t3 ) A, the only preimage above 0 is the origin 0 A1K .
Furthermore, since A - Kt becomes bijective after localization by t2 A,
we see that is an isomorphism above X {0 } and, thus, will be bijective.
Using 2.4/17, all localizations of the polynomial ring Kt by maximal ideals
are regular local rings of Krull dimension 1. This corresponds to the fact that
A1K is non-singular at all its points and the same is true, of course, for X {0 },
due to the isomorphism A1K {0} - X {0 }. However, the point 0 X
is singular, since the local ring Am is not regular. Indeed, t2 denes a system of
parameters in Am , which implies that Am is of Krull dimension 1. On the other
hand, we have A/m K and dimK m/m2 = 2 and this forbids Am to be regular
by 2.4/18. Therefore X has precisely one singularity, namely 0 . The same would
follow from the Jacobian Criterion 8.5/9 in conjunction with 8.5/15, at least if
K is algebraically closed.
To demonstrate that blowing up might be useful for resolving singular-
ities, let us show that the above normalization morphism : A1K -X

is just the blow-up of the point 0 ond X, namely the canonical morphism
Proj A - Spec A where A = d=0 m . Indeed, the irrelevant ideal d=1 m
d
of A is generated by t2 and t3 , viewed as homogeneous elements of degree 1 in A.

Therefore Proj A = D+ (t2 )D+ (t3 ) = D+ (t2 ), since D+ (t2 ) = D+ (t6 ) = D+ (t3 ),
and it follows that Proj A reduces to Spec A(t2 ) where A(t2 ) is the homogeneous
localization of A by t2 . Now consider the canonical inclusions
md - t2d Kt, d N.
Certainly, for d = 0 we get the strict inclusion A - Kt, but all other
inclusions
are bijective! Thus, viewing A as a subring of the graded ring
B = d=0 t 2d
Kt, the homogeneous localization A(t2 ) will coincide with the
homogeneous localization B(t2 ) . If we write t instead of t2 as a homogeneous
element of degree 1 in B, we see that B = Ktt is a polynomial ring in two
variables, the degree function being given by the degree in t . But then it is clear
that the homogeneous localization B(t2 ) = B(t ) coincides with Kt. Therefore
the blow-up of 0 on X is the morphism A1K - X induced from the inclusion
-
A Kt and, thus, coincides with the normalization morphism .
After this digression to the world of blowing up, let us look a bit closer at
the already mentioned criterion 9.4/10 relating ample invertible sheaves to pro-
jective embeddings. We x a relative scheme X over some base ring R assuming
that X is quasi-compact and quasi-separated. An invertible sheaf L on X is
called ample if some tensor power Lm of it admits global generators s0 , . . . , sn
such that each Xsi , dened as the open subscheme of X where Lm is gener-
ated by si , is quasi-ane. The latter means that Xsi is quasi-compact and can
be viewed as an open subscheme of an ane scheme. If, in addition, X is of
nite type, the criterion 9.4/10 asserts that, for m big enough, Lm and the
sections si dene an immersion of X into the projective n-space PnR . As a main
ingredient for the proof we need a careful analysis of quasi-ane schemes; cf.
9.4/7.
There are several interesting applications of the projectivity criterion based
on ample invertible sheaves and we will discuss one of them, namely the pro-
jectivity of abelian varieties. As we will show in 9.6, any abelian variety admits
an ample invertible sheaf and therefore is projective. The construction of such
an invertible sheaf is done in terms of divisors. Therefore we need to discuss
the theory of divisors rst (see 9.3), including their relationship to invertible
9.1 Homogeneous Prime Spectra as Schemes 403
sheaves. This is quite laborious and involves a lot of details, but at the end
we get an equivalence between Weil divisors, Cartier divisors, and invertible
sheaves, provided the scheme that we work on is suciently nice; cf. 9.3/16.
To supply some idea about the nature of abelian varieties, x a eld K.
An abelian variety over K is a K-group scheme A that is proper, smooth, and
irreducible. The term K-group scheme is most conveniently characterized by
the fact that there is a K-morphism : A K A - A dening a group law
on A, in the sense that it induces for each K-scheme T a group structure on
the set of T -valued points HomK (T, A); see 9.6 for more details. The simplest
examples of abelian varieties are elliptic curves (with a rational point); these
exhaust all abelian varieties of dimension 1. Let us show how to access such
elliptic curves over the eld K = C. Consider = Z Z for C R as a
subgroup of the additive group of C, a so-called lattice, and look at the quotient
C/ . The latter is a torus (life belt) from the topological point of view and a
compact Riemann surface (complex analytic manifold of dimension 1) from the
analytic point of view. On C/ lives the so-called Weierstra -function (z)
as a meromorphic function. It satises a well-known dierential equation
(z)2 = 4(z)3 g2 (z) g3
where g2 and g3 depend on and the polynomial 4x3 g2 x g3 has only simple
roots. Furthermore, the map C/ - P2 (C) that is symbolically described
C
by z -
((z) : (z) : 1) gives rise to an isomorphism between C/ and the
submanifold E dened in the complex projective plain P2C (C) by the equation
y 2 z = 4x3 g2 xz 2 g3 z 3 .
Thus, E may be thought to be a subscheme of the projective plain P2C . Fur-
thermore, one can deduce from the addition theorem for (z) that the group
law transported from C/ to E corresponds to a morphism of C-schemes
E C E - E. In other words, E is an abelian variety over C of dimen-
sion 1 and one knows that all abelian varieties over C of dimension 1 are of this
type. This is the classical analytic background of the fact that abelian varieties
here of dimension 1 are projective. In higher dimensions it is still true that
an abelian variety over C, say of dimension d, comes from an analytic quotient
Cd /2d where 2d is a lattice of rank 2d of the additive group Cd . However,
a given quotient of type Cd /2d is algebraizable to become an abelian variety
only if the lattice 2d satises Riemanns period relations; see Mumford [21], I
for more details.
9.1 Homogeneous Prime Spectra as Schemes
In Section 7.1 we have introduced the projective n-space PnS by gluing ane
n-spaces AnS over a base scheme S. Now we want to discuss a more general
construction method for schemes of similar type. Let us start with some prepa-
rations.

Denition 1. A graded ring is a ring A with a decomposition A = nZ An
into a direct sum of subgroups An A such that Am An Am+n for all
m, n Z.
Note that the product Am An is meant as the set of all products a b

where a Am and b An . The elements of An are called homogeneous of
degree n. In particular, the zero element 0 A is homogeneous of every degree
n and it is the only element enjoying this property. Moreover, we have 1 A0
and see that A0 is a subring of A. In most cases we will assume An = 0 for
n < 0, talking then about an N-grading or a grading of type N on A. But the
general case of Z-graded rings will be of interest for us as well, notably when
dealing with localizations. As a typical example of an N-graded ring we can
consider the polynomial ring RX in a set of variables X over some ring R
where the grading is induced from the total degree of polynomials. Likewise,
the corresponding ring RX, X 1 of Laurent polynomials in X is a Z-graded
ring.
In the following let A be a graded ring
of general type. Then every element
f A admits a unique decomposition f = nZ fn into homogeneous elements
fn An where fn = 0 for almost all n Z. We call fn the homogeneous
component of f of degree n. Furthermore, an ideal a A is called graded or
homogeneous if, given any f a, allits homogeneous components fn belong to
a; in other words, if we have a = nZ (a An ). Also note that for a graded
ideal a A the residue ring A/a = nZ An /(a An ) is graded again.
Remark 2. An ideal a A is graded if and only if it is generated by homoge-

neous elements.
Proof. Let a be an ideal in A. Choosing generators for a, we can decompose these

into homogeneous components. If a is graded, all these components belong to a
and therefore generate a. In particular, we see that graded ideals are generated
by homogeneous elements.
Conversely, let (gi )iI be a family of homogeneous elements in A where
gi is homogeneous of degree ni . Then the ideal a = (gi ; i I) generated by
these elements is homogeneous. Indeed, look at some element f a and at a
homogeneous component f of f , say of degree n . To show f a choose an
equation f = iI ai gi with elements ai A where ai = 0 for almost all i I.
Let ai be the homogeneous component of ai of degree n ni . Then, using the
Am An Am+n
condition in
conjunction with the direct sum decomposition
A= A
n=0 n , we get f
= a
g
iI i i and, hence, f a. This shows that a is
graded.
Remark 3. The formation of sums, products, intersections, and radicals of

graded ideals yields graded ideals again.
Proof. The assertion follows from Remark 2 for sums and products of graded
ideals and is clear for intersections, due to the denition of graded ideals. Now
consider a graded ideal a A and a non-trivial element f rad(a) of its radical.
There is an exponent t > 0 such that f t a. Write f = f0 + f1 where f0 A
is the non-zero homogeneous component of highest degree of f and f1 the sum
of all its homogeneous components of lower degree. Then the binomial formula
t
t t t i ti
f = (f0 + f1 ) = ff
i=0
i 0 1
shows that f0t is a homogeneous component of f t , namely of degree tn0 if f0

is homogeneous of degree n0 . Since a is graded, f0t belongs to a and we get
f0 rad(a). Replacing f by f1 = f f0 , we can show by induction that
all homogeneous components of f belong to rad(a) and, hence, that rad(a) is
graded.
Remark 4. A graded ideal p A is prime if and only if f g p for homoge-

neous elements f, g A implies f p or g p.
Proof. We have only to show that a graded ideal is prime as soon as it satises
the prime ideal condition for products of homogeneous elements. Therefore let
p A be a graded ideal and let f, g A such that f g p. Proceeding indirectly,
assume that neither f nor g belongs to p. Then f and g admit homogeneous
components that do not belong to p and we may assume that all homogeneous
components of f and g belonging to p are trivial. Similarly as in the proof of
Remark 3, write f = f0 + f1 and g = g0 + g1 where f0 and g0 are the non-zero
homogeneous components of f and g of highest degree. By our construction,
f0 and g0 do not belong to p. However, f0 g0 is a homogeneous component of
f g p and, thus, contained in p, since p is graded. Now if p satises the prime
ideal condition for products of homogeneous elements, we see that f0 or g0 must
belong to p, contradicting our assumption.

Next we want to look at localizations of graded rings. Let A = nZ An be a
graded ring and f A a homogeneous element, say of degree d, so that f Ad .
Passing to Af , the localization of A by the multiplicative system generated by
f , we introduce the subgroups
a
Af,n = Af ; k N, a An+kd Af , n Z,
fk
and claim that these give rise to a grading on Af . Intuitively speaking, we assign
to a fraction of type fak where a A is homogeneous of some degree deg r the
homogeneous degree n = deg a kd. Note that, even if the grading on A is
of type N, the grading on the localization Af will not preserve this property,
except for trivial cases.
Proposition 5. Let A be a graded ring and f A a homogeneous element. Then

the subgroups
Af,n Af , as introduced above, give rise to the decomposition
Af = nZ Af,n and the latter denes a grading on Af .

Proof. Clearly we have Af = nZ Af,n , as well as Af,m Af,n Af,m+n for
integers m, n Z. Thus, it remains to show that the sum of the subgroups
Af,n Af is direct.
To do this, assume that f is homogeneous of degree d and look at an equation
nZ hn = 0 for some elements hn Af,n where hn is trivial for almost all
n Z. By the denition of Af,n , there exist exponents kn N and homogeneous
elements an An+kn d such that hn = faknn for all n Z, and we may assume that
an and kn are trivial for almost all n Z. Then, choosing max{kn ; n Z}
big enough, we get
f kn an = 0
nZ
as an equation in A where the summands f kn an are homogeneous of degree

d + n. Since A is a graded ring, we must have f kn an = 0 in A and therefore
hn = faknn = 0 for all n Z.
Denition 6. Let A be an N-graded ring and f A an element that is homo-

geneous of degree 1. Then the subring Af,0 Af of all homogeneous elements
of degree 0 in Af is called the homogeneous localization of A by f . The latter
is denoted by A(f ) .
When dealing with homogeneous localizations of type A(f ) , as just intro-

duced, it is reasonable, but not really necessary, to assume the homogeneous
degree d of f to be 1. For example, if A is a graded ring and f A a homo-
geneous element of degree 0, then the subring Af,0 of all homogeneous elements
of degree 0 in Af coincides with the localization (A0 )f of the subring A0 A
by f .
Homogeneous localizations of graded rings play a central role for the con-
struction of projective schemes, similarly as do ordinary localizations of rings
within the context of ane schemes. To prepare the discussion of such schemes,
we want to establish a basic lemma on homogeneous localizations. Also note
that, simplifying our terminology, a graded ring will in the following be tacitly
understood as a graded ring of type N, unless we talk explicitly about a grading
of type Z.
Lemma 7. For a graded ring A consider homogeneous elements f Ad and

g Ae of degrees d, e 1. Then there is a canonical isomorphism
k
- (A(f ) )f e gd , a -
gd g (d1)k a
A(f g)
(f g)k fe f (e+1)k
where a A(d+e)k for k N.

Proof. We look at the canonical isomorphism
: Af g - (Af )f e gd
of 1.2/10, which amounts to rewriting fractions of type (f ag)k as fractions with

a numerator in Af and a power of f e g d as denominator. For example, the
mapping described in the assertion can be used for general a A. Now view
Af as a Z-graded ring. Then f e g d Af is homogeneous of degree 0 and we
see from Proposition 5 that the localization (Af )f e gd is a Z-graded ring again.
Furthermore,
viewing
f g as a Z-graded
A ring as well, it is easily seen that
we have (Af g )n (Af )f e gd n for all n Z; in other words, respects
gradings by maintaining degrees of homogeneous elements. Therefore is an
isomorphism of graded rings and it restricts to an isomorphism between subrings
of elements of degree 0. As the corresponding subring of (Af )f e gd coincides with
the localization (A(f ) )f e gd , we are done.

Now let us associate to a graded ring A = nN An its corresponding

homogeneous prime spectrum Proj A. To dene it, we write A+ = n>0 An
and view this as a graded ideal in A, sometimes referred to as the irrelevant
ideal of A.
Denition 8. Let A be a graded ring. Then
Proj A = {p Spec A ; p graded, A+ p}
is called the homogeneous prime spectrum of A.
Remark 9. Let A be a graded ring and p A a prime ideal such that A+ p.

Then there is the following equivalence for ideals a A:
a p a A+ p.
Proof. Only the implication = needs to be veried. By our assumption there

exists an element f A+ p. Now let a be an ideal in A such that a A+ p
and consider an element a a. Then f a a A+ and, hence, f a p. Since p
is a prime ideal and f p, this implies a p so that a p.
In the situation of the remark we see that p, as a prime ideal in A satisfying

A+ p, is uniquely determined by its intersection p A+ . Thus, we can say
that Proj A consists of all graded prime ideals in A+ , more precisely, of all
intersections pA+ where p A is a graded prime ideal in A such that A+ p.
Next, similarly as we did on ordinary spectra of rings, we can introduce the
Zariski topology on the homogeneous prime spectrum Proj A of a graded ring
A. For any subset E A we set
V+ (E) = {p Proj A ; E p} = V (E) Proj A

where V (E) = {p Spec A ; E p} is the zero set of E in the full spectrum

Spec A, as introduced in Section 6.1. Then V+ (0) = Proj A for the zero ideal
0 A, as well as V+ (A+ ) = . Furthermore, we conclude from Remark 9 that
V+ (E) = V+ (a) = V+ (a A+ )
for a A the graded ideal in A that is generated by (all homogeneous com-

ponents of) the elements of E. If a A is an arbitrary graded ideal in A,
then its radical rad(a) is graded by Remark 3 and the same holds for the ideal
rad+ (a) = rad(a) A+ . Applying Remark 9 again, we get

V+ (a) = V+ rad(a) = V+ rad+ (a) .
Furthermore, using the fact that V+ (E) is the restriction of V (E) to Proj A in
the sense that V+ (E) = V (E) Proj A, we can derive the following analogue of
6.1/1:
Remark 10. Let A be a graded ring. Then, for a family (E ) of subsets in

A, respectively for subsets E, E A, one has

V+ E = V+ (E ), V+ (EE ) = V+ (E) V+ (E ).

Moreover, E E implies V+ (E) V+ (E ).
For any graded ring A we can view its homogeneous prime spectrum Proj A
as a subset of the ordinary spectrum Spec A. In particular, the Zariski topology
on Spec A restricts to a topology on Proj A.
Denition 11. Let A be a graded ring. The restriction of the Zariski topology
from Spec A to Proj A is called the Zariski topology of the homogeneous prime
spectrum Proj A. This way the closed subsets of Proj A are given by the sets of
type V+ (E) for arbitrary subsets E A, while the open subsets of Proj A are
just the complements of the closed ones.
Similarly as in the case of ordinary spectra, we can consider for any f A

the open subset
D+ (f ) = Proj A V+ (f ) = {p Proj A ; f p}
of Proj A; the latter coincides with the restriction of the basic open subset
D(f ) Spec A to Proj A. In particular, the sets of type D+ (f ) form a basis of
the Zariski topology on Proj A. However, this assertion can be improved:
Proposition 12. Let A be a graded ring. Then the sets of type D+ (f ) for
homogeneous elements f A+ form a basis of the Zariski topology on Proj A.
As a variant of this we can observe:
Let d N, d 1. Then the sets of type D+ (f ) for homogeneous elements

f And , where n varies over all integers 1, form a basis of the Zariski
topology on Proj A.
Proof. Let U Proj A be an open subset and a A a graded ideal such that
the complement of U in Proj A is given by the closed subset V+ (a) Proj A.
Then we have

V+ (a) = V+ (a A+ ) = V+ (f ),
f a A+ homogeneous
thus implying

U = Proj(A) V+ (a) = D+ (f ).
f a A+ homogeneous
Hence, every open subset U Proj A is a union of open sets of type D+ (f ) for
homogeneous elements f A+ .
Finally, given an integer d 1, we have D+ (f ) = D+ (f d ) for any element
f A+ that is homogeneous of a certain degree n 1. Since f d And , the
second assertion becomes clear as well.
Next, for any graded ring A, we want to cover the homogeneous prime
spectrum Proj A by ordinary prime spectra of type Spec A(f ) , for homogeneous
elements f A+ . This will enable us to canonically equip Proj A with the
structure of a scheme. As a key fact, we need:
Proposition 13. Let A be a graded ring and f A+ a homogeneous element.

Then the map
f : D+ (f ) - Spec A(f ) ,
p - pAf A(f ) ,
is a homeomorphism of topological spaces.
Proof. First of all, recall from 1.2/5 that the maps
Spec A p - pAf Af , Spec Af q - q A Spec A
yield mutually inverse bijections D(f ) Spec Af . In particular, we thereby see

that f transforms prime ideals from D+ (f ) to prime ideals in A(f ) . Indeed, if
p A is a prime ideal satisfying f p, then the ideal pAf Af is prime and
its intersection with the subring A(f ) Af yields a prime ideal in A(f ) . Next
we want to show:
Let f Ad where d 1. For elements h And , n N, and graded prime
ideals p A satisfying f p, one has
h
() f (p) h p,
fn

h
() f1 D n = D+ (f ) D+ (h) = D+ (f h).
f
Indeed, fhn f (p) is equivalent to h1 pAf and the latter is equivalent to

h p by 1.2/5 (iii). This shows () and equation () is a consequence.
Since the sets of type D( fhn ) form a basis of the Zariski topology on
Spec A(f ) , we conclude that f is continuous. Moreover, if we know that f
is surjective, we can see from () that f is open as well, since the sets of type
D+ (f ) D+ (h) for h And , n 1, form a basis for the restriction of the Zariski
topology from Proj A to D+ (f ); use Proposition 12.
Thus, it remains to show the bijectivity of f . Addressing the injectivity
rst, let p, p D+ (f ) be graded prime ideals such that f (p) = f (p ). Then,
for h A homogeneous of some degree n, we get hd And and we see from ()
that hd , respectively h belongs to p if and only if it belongs to p . Since p and
p are graded, we get p = p .
In order to see that f is surjective, consider a prime ideal q A(f ) and set
xd
pn = x An ; n q
f
for n N. We claim that the pn are the homogeneous components of a graded
prime ideal p A. To justify this we show:
(a) pn is a subgroup in An . Let x, y pn , hence, f n xd , f n y d q. This
implies f 2n (xy)2d q by the binomial formula and therefore f n (xy)d q,
since q is a prime ideal. The latter means x y pn . Also note that 0 pn .
(b) Am pn pm+n for m, n N. Let h Am and x pn , hence, f n xd q.
This implies f (m+n) (hx)d = f m hd f n xd q and therefore hx pm+n .
(c) If x Am and y An are such that xy pm+n , then x pm or
y pn . Let x, y be as stated and assume f (m+n) (xy)d q. Then the prime
ideal condition for q yields f m xd q or f n y d q, hence x pm or y pn .
(d) There exists an integer n > 0 such that pn = An . Since q is a proper
ideal in A(f ) , we must have f d f d = 1 q and, thus, f pd .

Properties (a) and (b) show that p = nN pn is a graded ideal in A.
Furthermore, we see from (c) and (d) in conjunction with Remark 4 that p is
a prime ideal satisfying p D+ (f ). Finally, to show f (p) = q, we use the
following equivalences for elements h And :
h hd h
n
q nd q h pnd n f (p)
f f f
Indeed, the rst equivalence follows from the prime ideal property of q, the
second one from the denition of the subgroup pnd , and the third one from ().
Therefore we see that f (p) = q and, hence, that f is surjective.
Lemma 14. Let A be a graded ring and f, g A+ homogeneous elements of

degree d, respectively e. Then there is a canonical commutative diagram
f g
D+ (f g) - Spec A(f g)

? f
?
D+ (f ) - Spec A(f )
where the maps f g and f are as in Proposition 13 and is the canonical

inclusion. Furthermore, is the open immersion obtained in the manner of
6.2/8 from the localization map
A(f ) - (A(f ) )f e gd A(f g)
where the right-hand isomorphism is the one of Lemma 7.
Proof. Looking at the localization maps
A - Af - Af g ,
we see from 1.2/5 that pAf = pAf g Af for any prime ideal p A satisfying
f g p. If, in addition, p A is graded, we obtain
f (p) = pAf A(f ) , f g (p) = pAf g A(f g)
and, hence,

f g (p) = pAf g A(f g) A(f ) = pAf A(f ) = f (p).
To give an application of the above techniques, we want to deduce an ana-

logue of 6.1/5 for Proj schemes, which will be used later in 9.3/11 for the
characterization of divisors on the projective n-space PnL over a eld L. Doing
so, consider a graded ring A and a subset Y Proj A. Then, similarly as we
did for spectra of ordinary rings, we can consider the ideal
I+ (Y ) = {f A+ ; Y V+ (f )} = I(Y ) A+
where I(Y ) is dened as in the paragraph preceding 6.1/4.
Corollary 15. Let A be a graded ring.

(i) For any subset E A+ , the ideal I+ (V+ (E)) coincides with rad+ (a)
where a is the restriction to A+ of the graded ideal generated by E in A.
(ii) For any subset Y Proj A its Zariski closure coincides with V+ (I+ (Y )).
Proof. In the situation of (i) we have V+ (E) = V+ (a) by Remark 9, and it

remains to show that rad+ (a) is the intersection of all graded prime ideals in
A+ containing a. Replacing A by the quotient A/a with its canonical grading,

we may assume a = 0. Now consider an element f A that is not nilpotent. It
is enough to show that there is a graded prime ideal in A not containing f . Since
not all homogeneous components of f can be nilpotent, we may assume that f
is homogeneous. Then we can consider the homogeneous localization A(f ) and
conclude that it is non-zero, since the ordinary localization Af is non-zero. So
there exists a prime ideal in A(f ) . By Proposition 13 the latter corresponds to
a graded prime ideal p A such that p D+ (f ) and, hence, f p, as desired.
to verify (ii), let a A+ be a graded ideal such that Y V+ (a). Since
Next,
V+ (a) = f a V+ (f ), we get Y V+ (f ) for all f a and, hence, I+ (Y ) a
so that V+ (I+ (Y )) V+ (a) and, in particular, V+ (I+ (Y )) Y . On the other
hand, Y V+ (I+ (Y )) for trivial reasons and we are done.
Now we have collected all necessary tools in order to equip the homo-
geneous prime spectrum Proj A of a graded ring A with the structure of a
scheme. Namely, it follows from Proposition 12 that the sets of type D+ (f )
for homogeneous elements f A+ form an open covering of Proj A and we
can view each of these covering sets as an ane scheme, just by using the
homeomorphism f : D+ (f ) - Spec A(f ) of Proposition 13 and transport-
ing the scheme structure from Spec A(f ) to D+ (f ). Then we apply the method
of 7.1/1 in order to glue the ane schemes D+ (f ) along the intersections
D+ (f ) D+ (g) = D+ (f g) to yield a global scheme. More precisely, we ap-
ply Lemma 14 and view D+ (f g) = Spec A(f g) as an open subscheme of both,
D+ (f ) = Spec A(f ) and D+ (g) = Spec A(g) , namely via the canonical maps
A(f ) - A(f g) A(g)
that are obtained from the corresponding canonical maps between ordinary lo-
calizations. For this to work well, it remains to check the cocycle condition.
The occurring triple intersections of type D+ (f ) D+ (g) D+ (h) may be inter-
preted as schemes of type Spec A(f gh) for homogeneous elements f, g, h A+ .
This way the restrictions of the involved gluing morphisms are reduced to the
identity morphism on Spec A(f gh) , as the same is true in the setting of ordinary
localizations, and we see that the cocycle condition holds for trivial reasons.
Therefore the gluing works well and Proj A becomes a scheme covered by the
open subschemes of type Spec A(f ) for f A+ homogeneous.
Let us point out that there is an alternative construction method, which
looks slightly more elementary. To explain it, we need the following auxiliary
result:
Lemma 16. Let A be a graded ring and f, g A homogeneous elements of

degree 1. Then, for D+ (f ) D+ (g), there is a canonical homomorphism
Ag - Af . The latter respects gradings and maintains degrees of homogeneous
elements. In particular, it restricts to a homomorphism of homogeneous local-
izations A(g) - A(f ) .
Proof. The assumption D+ (f ) D+ (g) implies V+ (f ) V+ (g) and, hence,

rad+ (f ) rad+ (g) by Corollary 15 (i). Thus, there is an equation f n = hg
for some exponent n 1 and a homogeneous element h A. As a result, the
universal property of localizations yields a canonical homomorphism
- Af ,
a - hk
a
Ag ,
gk f nk
where a A. It is easily checked that the latter maps homogeneous elements of
Ag to homogeneous elements in Af of the same degree. In particular, it restricts
to a homomorphism of homogeneous localizations A(g) - A(f ) .
To explain the alternative approach to the scheme Proj A of a graded ring

A, look at the ordinary prime spectrum X = Spec A and start out from the
functor
OX#
: D # (X) - Ring
considered in example (4) of Section 6.3; it associates to any element f A
(interpreted as the basic open subset D(f ) Spec A given by f ) the localization
Af and to any inclusion D(f ) D(g) the canonical map between localizations
Ag - Af . Using the fact that A is a graded ring, we can restrict this functor
to homogeneous elements f A+ . Since any restriction morphism Ag - Af ,
for homogeneous elements f, g A+ , will respect gradings, as we have seen
above, it induces a restriction morphism between homogeneous localizations
A(g) - A(f ) . Thus, for Y = Proj A and D + (Y ) the category of all basic
open subsets in Y with inclusions as morphisms, the functor OX #
gives rise to a
functor
OY : D + (Y ) - Ring
D+ (f ) - A(f ) ,
D+ (f ) D(g+ ) - A(g) - A(f ) .
Even if we are a bit more careful and work with localizations of type AS(f )
instead of Af as exercised in example (4) of Section 6.3, we can use the canonical
isomorphism Af - AS(f ) in order to transport the grading from Af to AS(f ) .
By Lemma 16 the graded ring AS(f ) , as well as its homogeneous part of degree 0,
denoted by (AS(f ) )0 , are well-dened and depend only on the subset D+ (f ) Y .
Then it follows from 6.6/2 and Proposition 13 in conjunction with Lemma 14
that the functor OY : D+ (f ) - (AS(f ) )0 = A(f ) is a sheaf in the setting of
6.6/4. In addition, the latter result shows that OY extends uniquely to a sheaf on
all of Y , denoted by OY again. Moreover, we see for any homogeneous element
f A+ that the locally ringed space (D+ (f ), OY |D+ (f ) ) is isomorphic to the
ane scheme Spec A(f ) . Consequently, Y = (Y, OY ) is a scheme and the ane
schemes Spec A(f ) for f A+ homogeneous form an open covering of it. Thus,
using either method, we can state:
Theorem 17. Let A be a graded ring. Then we can equip its homogeneous prime
spectrum Proj A with the structure of a scheme in such a way that for homo-
geneous elements f A+ the homeomorphisms f : D+ (f ) - Spec A(f ) of

Lemma 13 become isomorphisms of schemes and, thus, give rise to an ane
open covering of Proj A. The resulting scheme is denoted by Proj A again and
is called the Proj scheme associated to A.
A graded ring A together with a ring homomorphism R - A0 A

is called a graded R-algebra; it satises RAn An for all n. In such a case
the scheme Proj A constructed above may be viewed as an R-scheme and we
will write ProjR A for it in order to refer to this fact. As a natural example,
we may consider the polynomial ring A = Rt0 , . . . , td as a graded R-algebra
where the grading is given by the total degree of polynomials. Then the scheme
ProjR A is canonically isomorphic to the projective d-space PdR , as introduced
in Section 7.1. Indeed, the irrelevant ideal A+ A is generated by t0 , . . . , td
and, hence, the scheme Proj A is covered by the ane open subschemes D+ (ti ),
i = 0, . . . , d. The latter are ane d-spaces AdR , since the homogeneous local-
ization A(ti ) coincides with the polynomial ring R tt0i , . . . , ti1
ti
, ti+1
ti
, . . . , ttdi , the
t0 ti1 ti+1 td
fractions ti , . . . , ti , ti , . . . , ti being viewed as variables. Also note that
we have A(ti tj ) (A(ti ) ) tj by Lemma 7 and, hence, that the intersections
ti
D+ (ti ) D+ (tj ) = D+ (ti tj ) are as required in the construction of Section 7.1.
Finally observe that any graded ring A can be viewed as a graded Z-algebra,
thus, equipping Proj A with its canonical structure as a Z-scheme.
The notation Proj A or ProjR A might suggest calling such schemes pro-
jective. However, note that the term projective is reserved for Proj schemes
admitting a closed immersion into a projective d-space for some d N; see
9.5/7.
Proposition 18. For any graded ring A, the corresponding scheme Proj A is
separated (over Z). Likewise, for any ring R and a graded R-algebra A, the
corresponding R-scheme ProjR A is separated.
Proof. We check condition (iii) of 7.4/5 for the ane open covering of ProjR A
consisting of the open subschemes D+ (f ) Spec A(f ) where f varies over all
homogeneous elements in A+ . For two such elements f, g A+ , the intersection
D+ (f ) D+ (g) = D+ (f g) is isomorphic to Spec A(f g) and, hence, ane. Thus,
we have only to show that the multiplication morphism A(f ) R A(g) - A(f g)
is surjective. If f is homogeneous of degree d and g of degree e, the canonical
maps from A(f ) and A(g) to A(f g) may be seen as follows, using Lemma 7:
A(f ) - (A(f ) )f e gd A(f g) - (A(g) )gd f e A(g)
Therefore A(f g) (A(f ) )f e gd is generated (in the sense of algebras) by the

image of A(f ) and the inverse of f e g d . However, the latter inverse coincides
with the image of g d f e A(g) . Therefore it follows that the multiplication
map mentioned above is surjective and, thus, that ProjR A is separated.
Finally, let us discuss some functorial properties of Proj schemes. We know

from the ane case that any ring morphism : A - A gives rise to a mor-
phism between associated ane schemes a : Spec A - Spec A. To obtain a

similar assertion for Proj schemes, assume that A, A are graded rings and that
is a morphism of graded rings in the sense that (An ) An for all n. Looking
at a graded prime ideal p A , its preimage p = 1 (p ) A is graded again
and we thereby see that we can restrict the map a : Spec A - Spec A to

graded prime ideals. However, for p Proj A it can happen that its preimage
p = 1 (p ) A contains the irrelevant ideal A+ of A and, thus, does not be-
long to Proj A. The latter is the case if and only if p contains the image (A+ ).
Thereby we see:
Remark 19. For any morphism of graded rings : A - A , the map

a
: Spec A - Spec A, p - 1 (p ),
restricts to a map

a
: Proj A V+ (A+ ) - Proj A,
satisfying a 1 (D+ (f )) = D+ ((f )) for homogeneous elements f A+ .
Proof. It only remains to work out the formula on the preimages of basic open
subsets of Proj A. To do this, recall that a 1 (D(f )) = D((f )) by 6.2/4. Thus,
it follows
a 1

D+ (f ) = a 1 D(f ) Proj A Proj A V+ ((A+ ))

= D (f ) Proj A V+ ((A+ )) = D (f ) Proj A

= D+ (f ) ,
since V+ ((A+ )) is disjoint from D((f )).
Now consider a homogeneous element f A+ in the situation of the above

remark and write f = (f ). Passing to localizations, induces a morphism
of graded rings f : Af - A and thereby a morphism (f ) : A(f ) - A
f (f )
between homogeneous localizations. The latter in turn gives rise to a morphism
between the associated ane schemes and, using the homeomorphisms of Propo-
sition 13, to a diagram
a
D+ (f ) - D+ (f )
f f
? a ?
Spec A(f ) -
(f )
Spec A(f ) ,
which is easily seen to be commutative. Transporting the structure of scheme
morphism from the lower to the upper row, we obtain a family of scheme mor-
phisms a (f ) : D+ (f ) - D+ (f ), parametrized by the homogeneous elements
f A+ . Furthermore, we claim that 7.1/2 can be applied to glue these mor-

phisms, thereby endowing the map
a
: Proj A V+ ((A+ )) - Proj A
with the structure of a scheme morphism, dened on the open subscheme

Proj A V+ ((A+ )) of Proj A. For this to work well we have to check that for
two homogeneous elements f, g A+ and their images f = (f ) and g = (g)
the corresponding morphisms
a
(f ) : D+ (f ) - D+ (f ), a
(g) : D+ (g ) - D+ (g)
coincide on D+ (f g ). However, that this is the case is deduced with the help of
Lemma 14 from the canonical commutative diagram
Af - Af
? ?
Af g - Af g
and its version for homogeneous localizations. Thus, summing up we can state:
Proposition 20. Let : A - A be a morphism of graded rings. Then there

is an associated scheme morphism

a
: Proj A V+ (A+ ) - Proj A
such that for every homogeneous element f A+ and its image f = (f )

the equation a 1 (D+ (f )) = D+ (f ) holds and the restriction of a to D+ (f )
coincides with the canonical morphism a (f ) : Spec A(f ) - Spec A(f ) in the
a
manner indicated above. In particular, the morphism is ane, using the
terminology introduced in the context of 7.4/8 or 9.5/1.
Corollary 21. In the situation of Proposition 20, assume that : A - A is

a surjective morphism of graded rings with kernel a; the latter is a graded ideal.
Then V+ ((A+ )) = and, hence, gives rise to a morphism
a
: Proj A - Proj A.
Furthermore, a is a closed immersion with image V+ (a).
Proof. Since : A - A is surjective, we have (A+ ) = A and, hence,

+
V+ ((A+ )) = . Furthermore, all localized maps Af - Af are surjective
and the same remains true for the corresponding homogeneous localizations.
Therefore a is a closed immersion by 7.3/9. Since, pointwise, a is induced
from the map a : Spec A - Spec A, which has image V (a), we see that a
must have image V+ (a).
Let us add that, unlike the ane case, a morphism : Proj A - Proj A
between Proj schemes of graded rings A, A is not necessarily induced from
a morphism of graded rings A - A , as there are examples of morphisms
that are not ane. To construct such a morphism, let R be a ring and
view the polynomial rings A = Rt0 and A = Rt0 , . . . , td as graded rings
with respect to the total degree of polynomials. Then Proj A = Spec R, as is
to be shown in Exercise 4. Therefore, viewing A as a graded R-algebra, the
structural morphism Proj A - Spec R may be interpreted as a morphism
: PR = Proj A
d - Proj A. The latter is not ane in general, since PdR is not
necessarily ane, for example if d > 0 and R is a eld; see 7.1.
Exercises
1. Graded rings as polynomial rings. Consider the polynomial ring A = Zti ; i I
in a system of variables (t
i )iI . Show for a family of integers di Z, i I, that
there is a unique grading nZ An on A such that ti Adi for all i I. Conclude
that every graded ring is of type Zti ; i I/a where Zti ; i I is a graded
ring as before and a is a graded ideal.

2. Alternative characterization of homogeneous localizations. Let A = nN An be
(d)
a graded ring. Fixing an integer d > 0, construct a graded ring A(d) = nN An
(d)
as a subring of A by setting An = Adn and show for any homogeneous element
f Ad that the map fxr - x mod (f 1)A(d) denes a ring isomorphism
A(f ) - A /(f 1)A .
(d) (d)

3. Let A = nN An be a graded ring and f A+ a homogeneous element. Show
that D+ (f ) = if and only if f is nilpotent. Conclude that Proj A = is equiv-
alent to the fact that every element f A+ is nilpotent.
4. Let R be a ring and t a variable. View the polynomial ring Rt as a graded
R-algebra with the grading being induced from the degree of polynomials. Show
that there is a canonical isomorphism of schemes Proj Rt Spec R.
5. Consider the projective n-space PnR over a ring R and assume n > 1. Show that
the scheme PnR is not ane, unless R is the zero ring.

6. Show for any graded ring A = nN An that the scheme Proj A is Noetherian if
A is Noetherian. Hint: Use Exercise 2.

7. Let R be a ring and A = nN An a graded R-algebra. Show that ProjR A is an
R-scheme of nite type if A is an R-algebra of nite type. Hint: Use Exercise 2.

8. Consider the graded ring A(d) as in Exercise 2 for a graded ring A = nN An
and an integer d > 0. Show that the inclusion map A(d) - A gives rise to an
isomorphism of schemes Proj A - Proj A(d) .

9. For a graded ring A = nN An , construct in a canonical way a graded ring
A = nN An by setting A0 = Z and An = An for n > 0. Show that the natural

map A - A gives rise to an isomorphism of schemes Proj A - Proj A .

10. Let : A - A be a morphism of graded rings with gradings (An )nN and
(An )nN , and assume that restricts to isomorphisms An - An for su-

ciently large n. Show that gives rise to an isomorphism Proj A - Proj A.

11. Let S be a scheme. Generalize the notion of graded algebras over rings to the
case of sheaves of OS -algebras. Let A = nN An be such a graded OS -algebra
and assume that it is quasi-coherent. Show that all homogeneous components
An are quasi-coherent OS -modules, and construct the Proj scheme ProjS A as a
generalization of Proj schemes for graded algebras.
9.2 Invertible Sheaves and Serre Twists
Let A be a graded ring of general type. A graded A-module is an A-module

M together with a direct sum decomposition M = nZ Mn into abelian sub-
groups Mn of M , where Am Mn Mm+n for all m, n Z. Typically, in our
applications the grading on A will be of type N, whereas on M we will have to
consider a grading of type Z. In particular, we will continue with our habit to
assume gradings on rings to be of type N, unless stated otherwise.
For any graded A-module M and an integer d Z, we dene the graded
A-module M (d) by
M (d)n = Mn+d , n Z.
Thus, without changing the underlying A-module, M (d) is obtained from M
by shifting the indices of the sequence . . . , Mn1 , Mn , Mn+1 , . . . by d steps to
the right. In particular, A can be viewed as a graded module over itself and it
therefore makes sense to view A(d) for any d Z as a graded A-module.
Similarly as for graded rings, one can construct homogeneous localizations
of graded modules. Namely, let A be a graded ring, M a graded A-module
and f A an element that is homogeneous of some degree d 1. Then the
localization Mf may be viewed as a graded Af -module and its homogeneous
part of degree 0, namely
x
M(f ) = Mf,0 = M f ; x M kd ,
fk
is an A(f ) -module. The latter is called the homogeneous localization of M by

"
f . Also note that M(f ) gives rise to a quasi-coherent OD+ (f ) -module M (f ) on
D+ (f ) Spec A(f ) Proj A. Letting f vary over the homogeneous elements
in A+ , we want to show that the resulting modules M "(f ) extend to a global
)
quasi-coherent OProj A -module M on Proj A.
Proposition 1. Let A be a graded ring (of type N) and M a graded A-module.

Let X = Proj A. Then the OD+ (f ) -modules M"(f ) where f varies over all homo-
geneous elements in A+ , can be glued via canonical isomorphisms of type
"
M (f ) |D+ (f g)

M (f g)
- M
"(g) |D+ (f g)
). The latter is called the module (or

to yield a quasi-coherent OX -module M
module sheaf) associated to M on Proj A.
9.2 Invertible Sheaves and Serre Twists 419
If : M - N is a morphism of graded A-modules, i.e. a morphism of

A-modules respecting gradings on M and N , then induces for each homoge-
neous element f A+ an A(f ) -module homomorphism M(f ) - N(f ) . Further-
more, the corresponding OX |D+ (f ) -module homomorphisms M"(f )
-N "(f ) can
be glued to yield an OX -module homomorphism M ) - *
N.
Proof. Given homogeneous elements f, g A+ , say of degrees d and e, there

exist canonical A(f g) -isomorphisms
(M(f ) )f e gd M(f g) - (M(g) )gd f e ,
just as we have seen in 9.1/7; the proof carries over without changes. On the level
of associated modules, these isomorphisms dene gluing morphisms, as stated
in the assertion. To see that the gluing works well, one has to check the cocycle
condition of 7.1/1. This is done in the same way as indicated in the discussion
leading to the proof of 9.1/17. One shows that the gluing morphisms restrict on
triple overlaps of type D+ (f ) D+ (g) D+ (h) = D+ (f gh) to the identity map
on M (f gh) . Alternatively, one can prove a module analogue of 9.1/16 and employ
the second method proposed for the proof of 9.1/17. Finally, the construction
of associated morphisms is done along the lines of the proof of 7.1/2.
We want to show that the formation of associated modules on Proj schemes

respects tensor products. Recall that the tensor product F OX G of two
OX -modules F, G on a scheme X is dened as the OX -module sheaf associ-
ated to the presheaf
U - F(U ) OX (U ) G(U ), for U X open.
If X is ane, say X = Spec A, and F, G are associated to A-modules M , N ,

then the compatibility of tensor products with localizations (use 4.3/2) shows
that F OX G is associated to the tensor product of A-modules M A M .
On the other hand, for a graded ring A and graded A-modules M , N , the
tensor product M A N is a graded A-module again if, for n Z, we call any
nite sum of type

xr ys M A N, x r Mr , ys Ns ,
r+s=n
homogeneous of degree n. To check that we get a well-dened grading this way,

observe the direct sum decomposition

M Z N = Mr Z N s = Mr Z Ns
r,sZ nZ r+s=n
in terms of Z-modules. In particular, working over Z as a graded ring concen-

trated at degree 0, we see that M Z N is a graded Z-module, indeed. Since the
kernel of the canonical map
: M Z N - M A N
is generated by all elements of type (ax) y x (ay) where a A, x M ,

y N , the latter can be generated by homogeneous elements of M Z N . From
this it easily follows that the quotient M Z N/ ker M A N is canonically
a graded A-module.
Proposition 2. Let A be a graded ring whose irrelevant ideal A+ A is

generated by A1 and let M , N be graded A-modules. Then there is a canonical
isomorphism of OX -modules M ) O N * - M A M on X = Proj A.
X
Proof. For any f Ad , d 1, look at the composition
M(f ) A(f ) N(f ) - Mf A f N f - (M A N )f
induced from the inclusions of homogeneous into full localizations on the left-
hand side and from the compatibility of tensor products with localizations on
the right-hand side; use 4.3/2. Since the isomorphism on the right conserves
homogeneous degrees, we end up with a homomorphism of A(f ) -modules
(f ) : M(f ) A(f ) N(f ) - (M A N )(f ) ,
given by
x y -
xy
,
fr fs f r+s
for r, s Z and x Mdr , y Nds . Furthermore, it is clear that, for homogeneous
elements f, g A+ , there is a commutative diagram
(f )
M(f ) A(f ) N(f ) - (M A N )(f )
? (f g)
?
M(gf ) A(f g) N(f g) - (M A N )(f g)
so that, by a gluing argument, the morphisms (f ) give rise to a morphism of

OX -modules
) O N
: M * - M A N .
X
Using the fact that A+ is generated by A1 , we claim that is an isomor-

phism. For this it is enough to show that the above mentioned maps (f ) are
isomorphisms for elements f A1 . In order to construct an inverse of (f ) in
this case, we start with the Z-linear map
: M Z N - M(f ) A(f ) N(f )
that is given for homogeneous elements x Mr and y Ns by
-
x y
xy .
fr fs
Then we see for a A homogeneous, or even a A arbitrary, that
(ax) y x (ay) ker
so that factors through a map
: M A N - M(f ) A(f ) N(f ) ,
which so to speak is linear over the ring morphism

- -
a
A A(f ) , Ar a A(f ) .
fr
Since the latter map sends f to 1 and, thus, factors through the localization
Af , it follows that induces a morphism
f : (M A N )f - M(f ) A(f ) N(f ) ,
given by
xy -
x y

ft fr fs
for r, s, t Z and x Mr , y Ns . Then, restricting f to the homogeneous
localization (M A N )(f ) , we get the desired inverse of (f ) .
In the following, we want to look a bit closer at some examples of associated

modules on Proj schemes, among them Serres twisted sheaves:
Denition 3. Let A be a graded ring and X = Proj A its associated scheme.

The nth twisted Serre sheaf of the structure sheaf OX of X, for n Z, is the
associated to A(n) as graded A-module.
OX -module OX (n) = A(n)
First, let us recall some properties applicable to module sheaves, which we

will need; see also Section 6.8 for a discussion of some of these within the setting
of quasi-coherent modules.
Denition 4. Let X be a scheme and F an OX -module sheaf.

(i) F is called free if there exists an isomorphism OX
(I) - F for some in-
dex set I. It is called free of rank n N if there is an isomorphism OX n - F;
(I)
see Section 6.8 for the denition of the OX -modules OX and OX . n
(ii) F is called locally free (resp. locally free of rank n) if every x X

admits an open neighborhood U X such that the restriction F|U is free (resp.
free of rank n).
(iii) A family of global sections (fi )iI in (X, F) = F (X) is called a system
of generators of F if the OX -module morphism OX
(I) - F given by these
sections is an epimorphism. Furthermore, F is called nitely generated if F
has a nite system of generators.
(iv) F is called locally of nite type if every x X admits an open neigh-
borhood U X such that F|U is nitely generated.
(v) F is called locally of nite presentation if every x X admits an open

neighborhood U X, on which F has a nite presentation, i.e. an exact se-
quence of type
OUm - On - F|U -0
U
for suitable integers m, n N.
In particular, every free OX -module is locally free and every nitely gen-
erated OX -module is locally of nite type. Also it is clear that every free or
locally free OX -module is quasi-coherent. Moreover, every OX -module that is
locally of nite presentation is locally of nite type and quasi-coherent.
Now let us look at the projective d-space X = PdR over some base ring R and
study Serres sheaves OX (n) as introduced in Denition 3. To do this consider
the polynomial ring A = Rt0 , . . . , td over R with variables t0 , . . . , td . The
total degree of polynomials denes a grading on A such that X = ProjR A; see
Section 9.1. Since the variables t0 , . . . , td generate the irrelevant ideal A+ A,
we get V+ (t0 , . . . , td ) = and therefore
X = D+ (t0 ) . . . D+ (td )
so that the schemes D+ (ti ) = Spec A(ti ) AdR with A(ti ) = R tt0i , . . . , ttdi form
an ane open covering of X.
By the denition of Serres sheaf OX (n), its restriction to the open sub-
set D+ (ti ) X is associated to the A(ti ) -module A(n)(ti ) . Here A(ti ) is the
homogeneous localization of A by ti ; it consists of all elements of the ordinary
localization Ati that are homogeneous of degree 0. Likewise, we can interpret the
homogeneous localization A(n)(ti ) of A(n) at ti as the subgroup of all elements
in Ati that are homogeneous of degree n. Then we get
't td ( n
0
A(n)(ti ) = R , . . . , ti ,
ti ti
since multiplication and division by tni sets up an isomorphism of A(ti ) -modules
A(ti ) - A(n)(ti ) . In particular, A(n)(ti ) is a free A(ti ) -module of rank 1,
generated by tni , and it follows that OX (n)|D+ (ti ) is free of rank 1. Consequently,
OX (n) is locally free of rank 1.
In order to determine the group (X, OX (n)) = OX (n)(X) of global sec-
tions of the sheaf OX (n), observe that all localizations of A by monomials in
t0 , . . . , td may be viewed as subrings of Rt0 , t1 1
0 , . . . , td , td , the ring of Laurent
polynomials in t0 , . . . , td over R. Furthermore, canonical restriction morphisms
between such localizations are given by inclusions then. Therefore we get
d 't td ( n
d
Rt0 , . . . , td , t1
0
X, OX (n) = R ,..., ti i =A
i=0
t i t i i=0
and, hence, (X, OX (n)) An . The reverse inclusion An (X, OX (n)) holds
as well, since we can write
t td n
0
f (t0 , . . . , td ) = f ,..., ti
ti ti
for i = 0, . . . , d and any homogeneous polynomial f An . Thus, we obtain

0 ,...,nd N R t0 . . . td
n0 nd
for n 0,
nn0 +...+n d =n
X, OX (n) = An = R for n = 0,

0 for n < 0.
In particular, it follows that an OX -module OX (n) where n = 0 (and R = 0,

d > 0) cannot be free, since otherwise, being locally free of rank 1, it would be
isomorphic to OX and, thus, impose R as its R-module of global sections. On
the other hand, we see for n 0 that tni generates OX (n) on D+ (ti ) and, hence,
that the powers tni , i = 0, . . . , d, yield a system of global generators for OX (n)
on X. Hence, OX (n) is nitely generated. Furthermore, the equation
t n
i
tni = tnj ,
tj
valid on D+ (tj ), allows to conclude more specically that tni generates OX (n) for
n > 0 precisely on D+ (ti ). The reason is that tnj is a free generator of OX (n) on
D+ (tj ). Hence, tni can be a second free generator only at points of D+ (tj ), where
ti
tj
does not vanish. However, the latter is the case precisely on D+ (ti ) D+ (tj );
use 9.1/7 in conjunction with 9.1/14.
Denition 5. Let X be a scheme. An OX -module sheaf L is called an invertible

sheaf (or a line bundle) on X if it is locally free of rank 1; in other words,
there must exist an open covering (Ui )iI of X together with isomorphisms
L|Ui OX |Ui , i I. In particular, L is quasi-coherent in this case.
We have shown above that Serres twisted sheaves OX (n) are locally free of
rank 1 on the projective d-space X = PdR . Thus, they are examples of invertible
sheaves and we will generalize this assertion in Proposition 7 below.
Proposition 6. The isomorphism classes of invertible sheaves on a scheme

X form a commutative group under the group law (L, L ) - L O L , the
X
tensor product of OX -modules. It is denoted by Pic(X), referred to as the Picard
group of X. Furthermore, Pic(X) is isomorphic to the rst cohomology group

H 1 (X, OX ), as we will see below in Corollary 9.
The unit element in Pic(X) is given by the trivial invertible sheaf OX and,
for any L Pic(X), its inverse is L1 = HomOX (L, OX ) where Hom is the
so-called sheaf Hom; see below.
The group Pic(X) is functorial in X. Namely, for any morphism of schemes
f: Y - X the pull-back of invertible sheaves L - f (L) gives rise to a
group homomorphism Pic(X) - Pic(Y ).
Proof. If L and L are invertible, they are locally on X isomorphic to the trivial
sheaf. Using the canonical isomorphisms of type OU OU OU OU on open
subsets U X, we see that L OX L is invertible again. Therefore the ten-
sor product denes, indeed, a law of composition on Pic(X), the collection of
invertible sheaves on X. Clearly, this law is associative and commutative with
OX serving as identity element.
To describe inverses in Pic(X), note that Hom(L, L ) for OX -module sheaves
L, L on X consists of the functor
U - HomOX |U (L|U , L |U ), for U X open,
which is a sheaf of OX -modules. There is a canonical morphism of OX -modules
HomOX (L, OX ) OX L - OX ,
which is given on the level of sections by a - (a). It is easily checked

that this is an isomorphism for L = OX . The same remains true if L is locally
on X isomorphic to OX and, hence, invertible.
Hence, we can conclude that Pic(X) is a group, provided we know that the
isomorphism classes of invertible sheaves on X can be viewed as a set. The latter
is not clear right away, but follows most conveniently from the description of
invertible sheaves by means of Cech 1-cocycles in the style of Proposition 8 and
Corollary 9 below.
To explain the functoriality of Pic(X), consider a morphism of schemes
f: Y - X. Then we see from 6.9/4 that the inverse image f (L) of
any invertible sheaf L on X is invertible on Y . Therefore f induces a map
Pic(X) - Pic(Y ). The latter is a group homomorphism, as follows from the
arguments of 4.3/2 in conjunction with 6.9/4.
Proposition 7. Let A be a graded ring such that the irrelevant ideal A+ A

is generated by A1 and let X = Proj A. Then:
(i) OX (n) is an invertible OX -module for all n Z.
(ii) OX (m) OX OX (n) OX (m + n) for all m, n Z.
(iii) OX (n) HomOX (OX (n), OX ) for all n Z.
Proof. To obtain (i), we proceed similarly as in the case of the projective d-space
PdR . Since A1 generates A+ , we see that V+ (A1 ) = V+ (A+ ) = . Hence, the
system (D+ (f ))f A1 yields an ane open covering of Proj A and it is enough to
show that A(n)(f ) for n Z and elements f A1 is a free A(f ) -module of rank
1. However, the latter is clear, since we may interpret A(n)(f ) as the subgroup
of homogeneous elements of degree n in the localization Af and since
A(f ) - A(n)(f ) , a - af n ,
is an isomorphism of A(f ) -modules.

For assertion (ii) we interpret the trivial isomorphism A A A - A as
an isomorphism of graded A-modules
A(m) A A(n) - A(m + n)
that maintains homogeneous degrees. Then, using Proposition 2, we get the

desired isomorphism
A(m) A(m)
O A(n)
A A(n) A(m + n).
X
Finally, (iii) follows from Proposition 6 in conjunction with (ii).
In particular, we see from Proposition 7 that the map
Z - Pic(X), n - OX (n),
is a group morphism where for the projective d-space X = PdK over a eld K one
can show that it is, in fact, an isomorphism; see 9.3/17. Thus, Serres twisted
sheaves are the only invertible sheaves on X in this case.
Next we want to describe invertible sheaves on a scheme X in terms of Cech

cocycles with values in OX , the sheaf of invertible functions on X. To do this, let
L be an invertible sheaf on X and let U = (Ui )iI be an open covering of X such
that L|Ui is free; the latter amounts to the fact that there is an isomorphism
of OUi -modules OUi - L|Ui , carrying the unit section 1 OX (Ui ) to a
generator fi L(Ui ) of L|Ui as OUi -module. We say that L trivializes with
respect to U, or that L is U-trivial. Then for two indices i, j I both, fi and
fj , generate L on the intersection Ui Uj and there exists an equation of type
fi |Ui Uj = ij fj |Ui Uj with a unit ij OX (Ui Uj ) . Furthermore, on triple
overlaps Ui Uj Uk for i, j, k I we have
fi = ik fk , fi = ij fj , fj = jk fk ,
and, thus,
1
ik = ij jk resp. jk ik ij = 1.
1
The equation jk ik ij = 1 is referred to as the cocycle condition.
If there is an isomorphism OX - L, the unit section 1 OX (X) gives
rise to a global generator f L(X) and there exist elements i OX (Ui ) such
that f |Ui = i fi for all i I. Thereby we obtain on intersections Ui Uj for
i, j I the relation
ij = j i1 .
The equations we have just discussed t nicely into the formalism of Cech
cohomology with values in an abelian sheaf F, as discussed in Section 7.6. For
the convenience of the reader we repeat here the relevant facts, replacing F by

the sheaf OX of invertible functions on X and writing the group law multiplica-
tively. Thus, we look at the Cech complex C (U, OX
) consisting of the groups

of q-cochains on U with values in OX , namely

C q (U, OX )= OX (Ui0 ...iq ) , q N,
i0 ,...,iq I
where Ui0 ,...,iq = Ui0 . . . Uiq , and of the coboundary maps

-
q : C q (U, OX ) C q+1 (U, OX ),

given for a q-cochain = (i0 ,...,iq )i0 ,...,iq I C q (U, OX ) by

q+1
(1)j
q ()i0 ,...,iq+1 = i0 ,...,ij ,...,iq+1 |Ui0 ,...,iq+1 ;
j=0
as usual, the notation ij means that the index ij has to be omitted. In particular,

we get for C 0 (U, OX )
0 ()ij = j |Uij (i |Uij )1 ,

and for C 1 (U, OX )
1 ()ijk = jk |Uijk (ik |Uijk )1 ij |Uijk .
Furthermore, recall that q+1 q = 1 for all q where 1 has to be interpreted as

the unit element of the multiplicative group C q+2 (U, OX ).
Then

Z q (U, OX ) = ker q C q (U, OX )
is called the group of q-cocycles and

B q (U, OX ) = im q1 C q (U, OX )

the group of q-coboundaries in C q (U, OX ) where we put B q (U, OX ) = 1 for

q = 0. Since q+1
= 1, we have B (U, OX ) Z (U, OX ) and one calls
q q q

H q (U, OX ) = Z q (U, OX )/B q (U, OX )

the qth Cech cohomology group on the covering U with values in OX .
Starting out from an invertible sheaf L on X that trivializes with respect
to the open covering U = (Ui )iI of X and from isomorphisms OUi - L|Ui
for i I we have constructed above a 1-cochain = (ij )i,jI that satises
the cocycle condition and, thus, as we can read from the denition of 1 , is a
1-cocycle. If we change the isomorphisms OUi - L|Ui under consideration by
units i OX (Ui ) , we nd out that changes by the 1-coboundary (j i1 |Uij ).
Therefore we can associate to each U-trivial invertible sheaf L on X a well-

dened cohomology class H 1 (U, OX ).

Conversely, starting with a cohomology class H 1 (U, OX ), we can choose
a representative = (ij )i,jI Z (U, OX ) and glue the sheaves OUi with
1
respect to the isomorphisms
ij : OUi |Ui Uj - OU |U U , a - ij a,
j i j
along the intersections Ui Uj to obtain an invertible sheaf L on X. The condi-

tion of to be a 1-cocycle assures that the necessary cocycle condition on triple
overlaps is satised. Moreover, if we change by a 1-coboundary and write L

for the resulting invertible sheaf on X, then such a 1-coboundary yields local
isomorphisms L|Ui - L |Ui that coincide on the intersections Ui Uj and
therefore dene an isomorphism L - L . Thus, we can state:
Proposition 8. Let X be a scheme, U an open covering of X, and PicU (X)

the group of invertible sheaves on X that trivialize with respect to U. Then the
above construction yields an isomorphism of abelian groups
PicU (X) - H 1 (U, O ).

X
Proof. It remains only to show that the described map respects group structures.
This is easily seen. Just observe that if L, L are two U-trivial invertible sheaves
on X and if fi L(Ui ), resp. fi L (Ui ) are sections generating L resp. L
on Ui , then the section fi fi (L OX L )(Ui ) generates the invertible sheaf
L OX L on Ui .
As an example, consider Serres nth twisted sheaf OX (n) on the scheme

X = Proj A of a graded ring A and assume that the irrelevant ideal A+ A is
generated by A1 . Then OX (n) is invertible by Proposition 7 (i) and the proof
of the latter result shows for any f A1 that there is an isomorphism
OX |D+ (f ) - OX (n)|D (f ) , 1 - f n.
+
Thus, choosing elements fi A1 , i I, that generate the irrelevant ideal A+ ,

the basic open sets Ui = D+ (fi ) will cover X and we see that the elements
fin
ij = (A(fi fj ) ) = OX (Ui Uj ) , i, j I,
fjn
dene a 1-cocycle, whose associated invertible sheaf is isomorphic to OX (n).

Also note that this way we can give an alternative proof of Proposition 7 (ii).
Namely, translated to the level of 1-cocycles, the isomorphism
OX (m) OX OX (n) OX (m + n)
amounts to the trivial equations
fim fin f m+n

m
n = im+n , i, j I.
f j fj fj
The assertion of Proposition 8 is easily extended to the full Picard group

Pic(X). Indeed, look at the collection Cov(X) of open coverings of X and
use the renement relation as a preorder. Then Cov(X) is directed, since for
two open coverings U = (Ui )iI and V = (Vj )jJ of X the product covering
U V = (Ui Vj )iI,jJ is a common renement. If V is a renement of U,
there is a canonical inclusion map PicU (X) - PicV (X), and we obtain the
group Pic(X) of all invertible sheaves on X as the inductive limit of the groups
PicU (X) where U varies over the open coverings of X.
A similar inductive limit can be considered on the level of Cech cohomology

groups H 1 (U, OX ), as we have seen in Section 7.6. If V = (Vj )jJ is a renement
of U = (Ui )iI , there is a map : J - I such that Vj U (j) for all j J.
Then
- -
C q (U, OX ) C q (V, OX ), ( (j0 ),..., (jq ) |Vj0 ,...,jq )j0 ,...,jq J ,
gives rise to a well-dened homomorphism

-
q (U, V) : H q (U, OX ) H q (V, OX ),
which is independent of the choice of the renement map : J - I; see Sec-

tion 7.6. The maps q (U, V) constitute an inductive system, and the associated
inductive limit

H q (X, OX ) = lim H q (U, OX )

UCov(X)

exists; it is the qth Cech cohomology group of X with values in OX . Furthermore,
it is easy to see that for a renement V of U the canonical diagram
-
PicU (X) H 1 (U, OX )
? ?
-
PicV (X) H 1 (V, OX )
is commutative. Therefore we conclude from Proposition 8:
Corollary 9. Let X be a scheme. Then the isomorphisms of Proposition 8

induce an isomorphism
Pic(X) - H 1 (X, O )
X
between the Picard group of invertible sheaves on X and the rst Cech cohomol-

ogy group of X with values in OX .

Also note that in place of H 1 (X, OX ) we may use the rst Grothendieck

cohomology group H (X, OX ) as considered in the setting of Section 7.7, since
1
it is known that Cech and Grothendieck cohomology coincide in degree 1.

Finally, we want to discuss the connection between invertible sheaves and
line bundles (in the literal sense of the word). Looking at A1U = A1Z Z U , the
ane line over some base scheme U , call a U -morphism : A1U - A1 linear
U
if it is of type
(t, x) - (x) t , x
for a section OU (U ), i.e. if on ane open parts Spec R U it corresponds
to the morphism of R-algebras
R - R, - (|Spec A ) .

Also note that such a linear U -morphism is an isomorphism if and only if is
a unit in OU (U ) .
Now x a scheme X. Given any X-scheme : L - X, we will use the
notation LU = 1 (U ) for open subsets U X, considering LU as a U -scheme.
A line bundle on X is an X-scheme : L - X with a so-called trivialization
as additional data. Thereby we mean an open covering (Ui )iI of X together
with Ui -isomorphisms
i : LUi - A1 = A1 Z Ui , i I,
Ui Z
such that on each intersection Ui Uj , i, j I, the resulting isomorphism

j i1 : A1Z Z (Ui Uj ) - A1 Z (Ui Uj )
Z
is linear, say of type (t, x) - (ij (x)t, x) for a unit ij OX (Ui Uj ). Fur-
thermore, a morphism L - L between line bundles on X is meant as an
X-morphism that, locally on X, is linear with respect to the given trivializa-
tions on L and L .
The ber of a line bundle L - X over any point x X consists, indeed,
of the ane line A1k(x) = A1Z Z Spec k(x). Moreover, the parametrization of all
these lines over the base X is locally trivial in the sense that it corresponds
locally on X to the situation encountered at the ane 1-space A1U = A1Z Z U .
This is why the term line bundle is applied. However, globally, L can be far
from being isomorphic to the ane line A1X , as the ane lines A1Ui are glued
via linear transformations on the intersections Ui Uj , the latter being given as
multiplication by the units ij OX (Ui Uj ) .
It is easily checked that the family of elements ij , i, j I, forms a 1-cocycle

with values in OX , relative to the covering U = (Ui )iI , and that the isomor-
phism classes of line bundles on X correspond bijectively to the cohomology

classes in H 1 (U, OX ). Therefore we get a bijective correspondence between iso-
morphism classes of invertible sheaves and of line bundles on X. Quite often one
does not make a strict dierence between these two types of objects, by talk-
ing about line bundles in situations where the terminology of invertible sheaves
should be applied. In such cases the line bundle L (in the strict sense) associated
to an invertible sheaf L is referred to as the total space of L.
The correspondence between invertible sheaves and line bundles, as just
described, is not restricted to locally free modules of rank 1. For locally free
modules of higher rank n, one considers on the level of total spaces so-called
vector bundles, which locally look like ane n-spaces AnU . The role of the units
ij is then taken over by invertible matrices with entries in OX (Ui Uj ).
Exercises

1. Let A = nN An be a graded ring. Show that the assumption of Propositions 2
and 7, namely, that the irrelevant ideal A+ A is generated by A1 , is fullled if
and only if A is generated by A1 when viewed as an A0 -algebra.
2. Give an example of a quasi-coherent module sheaf F on a scheme Proj A for a

graded ring A where F is associated to a graded A-module M , but where M is
not uniquely determined by F.
3. Let A be a graded algebra over some ring R and X = ProjR A the associated
scheme. Show for any ring morphism R - R :
(a) A = A R R is canonically a graded R -algebra; set X = Proj A .
(b) Proceeding as in the setting of 9.1/20, the canonical morphism A - AR R
gives rise to a morphism of schemes f : X - X, which may be identied with
the projection X Spec R Spec R - X.
(c) f (OX (n)) OX (n) for all n Z.
4. Consider the projective line X = P1R over a ring R. Describe Serres twisted
sheaves OX (n), n Z, via Cech cocycles and show that there cannot exist any
further invertible sheaves on X, up to isomorphism. See 9.3/17 for a higher-
dimensional version of this fact.
5. Consider the projective line X = P1F2 = ProjF2 F2 t0 , t1 over the eld F2 = Z/2Z.
t2
Show that = t0t+t0
1
+ t0 (t01+t1 ) gives rise to a Cech cocycle with respect to the
open covering of X consisting of the subsets D+ (t0 ) and D+ (t0 + t1 ). Relate the
resulting invertible sheaf to Serres twisted sheaves OX (n), n Z.
6. Let X be a scheme and F an OX -module that is locally of nite presentation.
Show that F is invertible if and only if the stalk Fx is a free OX,x -module of rank
1 at every point x X. Hint: Use 4.4/3.
7. Let A be a graded ring and X = Proj A the associated scheme. Given a sheaf of
OX -modules F, call F(n) = F OX OX (n) for n Z the nth twisted sheaf
of F. Show for a graded A-module M that there are canonical morphisms

M (n) - M )(n), n Z, and that the latter are isomorphisms if the ideal
A+ A is generated by A1 .
8. Let A be a graded ring and X = Proj A as before. Fora sheaf of OX -modules
F dene the associated graded A-module by (F) = nZ (X, F(n)). Inter-
preting the latter as an abelian group, use the canonical isomorphisms of Propo-
sition 7 (ii) to show that (F) is a graded A-module, indeed. Compute (F)
in the case where F = OX (n), n Z, and X is the projective d-space PdR over a
ring R.
9. Let A be a graded ring and X = Proj A as before. Show for graded A-modules M ,
N that the set HomA (M, N ) of A-module morphisms M - N is canonically
a graded A-module if M is nitely generated. Assuming the latter, show that
there is a canonical morphism of OX -modules HomA (M, N )
- Hom (M
OX
), N* ).
Furthermore, show that this morphism is an isomorphism if the ideal A+ A
is generated by A1 and M is an A-module of nite presentation. Hint: Consult
EGA [12], II, 2.5.13, if necessary.
10. Blowing up: Let X be a scheme and I OX a quasi-coherent ideal on X.
Usingthe context of Exercise 9.1/11, write I 0 = OX and view the direct
sum nN I n as a sheaf of graded OX -algebras. Then the canonical morphism
: X = Proj nN I n - X is called the blow-up of I on X, or of the closed
subscheme given by I on X. Show:
9.3 Divisors 431
(a) The blow-up : X - X is an isomorphism if I is invertible in the sense

of being an invertible OX -module.
(b) Let Z be the closed subscheme of X given by the ideal I OX . Then
restricts to an isomorphism X 1 (Z) - X Z.
(c) The ideal IOX that is generated by the inverse image 1 (I) in OX is
invertible, i.e. an invertible OX -module.
(d) Universal property of blowing up: If Y - X is a morphism of schemes
such that the ideal IOY is invertible on Y , there is a unique factorization of
Y - X through the blow-up : X - X. Hint: Consider the case where
X is ane, say X = Spec A, and, to simplify things, where I corresponds to an
ideal I A of nite type, say generated by f0n, . . . , fr . Then use the surjection of
graded A-algebras At0 , . . . , tr - nN I given by ti
- f i and determine
its kernel.
9.3 Divisors
So far we have characterized invertible sheaves in terms of Cech cohomology

and of line bundles. However, in order to actually construct such sheaves, divi-
sors come into play, at least in certain situations. The purpose of the present
section is to develop the basics on divisors and their relationship to invertible
sheaves. As an application, we will show in Section 9.6 how to construct ample
invertible sheaves on abelian varieties, thereby proving that abelian varieties
are projective.
Some local properties of schemes. In order to be able to really work

with divisors, the schemes under consideration should have certain nice prop-
erties, which we introduce next.
Denition 1. Let X be a scheme. Then X is called

(i) integral if it is irreducible and all its stalks OX,x are integral domains,
(ii) reduced if all its stalks OX,x are reduced in the sense that the nilradical
rad(OX,x ) is trivial,
(iii) normal if all its stalks OX,x are normal integral domains (3.1/9),
(iv) factorial if all its stalks OX,x are factorial rings,
(v) regular if all its stalks OX,x are regular Noetherian local rings (2.4/18).
It is clear that the above conditions can be checked locally on X, except

for condition (i), due to the irreducibility requirement. Furthermore, X being
factorial implies normal by 3.1/10 and normal, respectively integral, implies
reduced. Also let us refer to the Theorem of AuslanderBuchsbaum [24], Cor. 4
of Thm. IV.9, stating that regular implies factorial. Moreover, it is of interest
to know how certain of these properties can be characterized globally on ane
schemes.
Remark 2. Let A be a ring and X = Spec A the associated ane scheme.

Then:
(i) X is integral if and only if A is an integral domain.
(ii) X is reduced if and only if A is reduced in the sense that rad(A) = 0.
(iii) If X is irreducible, it is normal if and only if A is a normal integral
domain.
Proof. Assertion (ii) follows from 7.3/6 and from the injection

OX (X) - OX,x , f - (fx )xX ,
xX
of 6.4/7 where, as usual, fx denotes the stalk of f at x. Furthermore, using (i),

we see that (iii) is a consequence of 3.1/11.
Thus, it remains to verify assertion (i). To do this, assume that X is integral
and consider elements f, g A such that f g = 0. Then we get X = V (f ) V (g)
from 6.1/1 (iv), and we may assume X = V (f ) since X is irreducible. But this
implies rad(f ) = rad(A) by 6.1/5 (i) and, furthermore, rad(A) = 0 as we can see
from assertion (ii). Hence, we get f = 0, showing that A is an integral domain.
Conversely, if A is an integral domain, then any localization of A can be
viewed as a subring of its eld of fractions. Thus, all local rings OX,x are integral
domains by 6.6/9. Furthermore, X is irreducible by 6.1/15.
If we want to associate Weil divisors to functions living on a scheme, we

need the concept of so-called discrete valuation rings:
Denition 3. A ring A is called a discrete valuation ring if it is a local principal

ideal domain that is not a eld and, hence, is of Krull dimension 1.
If A is a discrete valuation ring, its maximal ideal is generated by a prime

element p and factorizations of elements in A are reduced to powers of p, up
to units in A. In particular, any element a of the eld of fractions K of A can
uniquely be written as a = p(a) with a unit A and an exponent (a) Z.
Thus, any x K either satises x A or, if the latter is not the case, x1 A.
Therefore a discrete valuation ring is, indeed, a valuation ring in the sense to
be dened later in 9.5/13. Furthermore, the map
: K - Z, a - (a),
is called a valuation of K. Namely, setting (0) = , the following conditions
are satised for elements a, b K:
(i) (a) = a = 0,
(ii) (ab) = (a) + (b),
(iii) (a + b) min (a), (b) .
Also note that for any R, 0 < < 1, the valuation leads to a
non-Archimedean absolute value
9.3 Divisors 433
| |: K - R0 , a - (a) ,
which satises |a + b| max(|a|, |b|), the so-called non-Archimedean triangle

inequality, as derived from condition (iii) above.
In order to relate Noetherian normal rings to discrete valuation rings, we
will apply the theory of primary decompositions for ideals, as dealt with in
Section 2.1. In particular, see 2.1/9 for the denition of Ass(a), the set of prime
ideals associated to an ideal a in a ring A.
Lemma 4. For a Noetherian local integral domain A with maximal ideal m the
(i) A is normal and of dimension 1.
(ii) A is normal and there exists a non-zero element a m such that m is
associated to the ideal aA, i.e. such that m Ass(aA).
(iii) m is a non-zero principal ideal.
(iv) A is a principal ideal domain of dimension 1 and, thus, a discrete valu-
ation ring.
(v) A is regular (2.4/18) and of dimension 1.
Proof. Starting with the implication (i) = (ii), assume that A is a normal and
of dimension 1. Then there are precisely two prime ideals in A, namely 0 and m.
Choosing any non-zero element a m, we know from 1.3/4 that rad(aA) is the
intersection of all prime ideals of A containing a. Since m is the only prime ideal
containing aA, its radical rad(aA) must coincide with m. Then m Ass(Aa) by
2.1/12.
Next let us assume (ii) and derive (iii). If a m is non-zero and m Ass(aA),
there is an element x A such that (aA : x) = m; see 2.1/9. In particular, we
get xm aA and, thus, a1 xm A so that we can view a1 xm as an ideal in
A. We claim that a1 xm m cannot be the case. Indeed, if the latter were true,
we would see from 3.1/4 (iv) that a1 x would be integral over A. However, if
A is normal, this would imply a1 x A and, thus, (aA : x) = A, contradicting
our choice of x. Therefore we conclude a1 xm m and, thus, a1 xm = A. But
then m = x1 aA, which necessarily shows x1 a m and that m is principal.

Now assume (iii) andobserve that
n=0 m
n
= 0. Indeed, if y
A is a gener-
ator of m, we have m n=0 m = n=1 y A =
n n n
n=0 m and, thus,
n
n=0 m = 0
by Nakayamas Lemma 1.4/10. Then look at an arbitrary non-zero ideal a A.
To show that it is principal, x an integer n N such that a mn , but
a mn+1 , and choose z a mn+1 . Using the fact that m is principal, there
exists an equation z = cy n for some element c A m where c must be a unit,
due to the fact that A is a local ring with maximal ideal m. This shows y n a
and, thus, mn = (y n ) a. Since a mn by construction, we have a = (y n ) and
a is principal. Taking into account that dim A = ht m = 1 by 2.4/5, the impli-
cation from (iii) to (iv) is clear. Moreover, that (iv) implies (i) is well-known;
see 3.1/10.
Finally, it follows from the denition of regularity in 2.4/18 that (iv) implies
(v). On the other hand, if A is regular and of dimension 1, its maximal ideal m
is a non-zero principal ideal and we see that (v) implies (iii).
We need another important result on Noetherian normal rings:
Proposition 5. Let A be a Noetherian normal integral domain of dimension

1 with eld of fractions K. Then

Ap = A,
pA prime, ht(p)=1
viewing the localizations Ap as subrings of K.
Proof. First observe for any non-zero element a A that all its associated
prime ideals p Ass(aA) are of height 1. Indeed, passing to the localization Ap
at such a prime ideal p, we can use 2.1/14 to show pAp Ass(aAp ). Then Ap
is a discrete valuation ring by Lemma 4 (ii) and, thus, an integral domain of
dimension 1.
Next, consider an element z = ay K with y, a A such that z Ap
and, hence, y aAp for all prime ideals p A of height 1. Furthermore, let
Ass(aA) consist of the (pairwise distinct) prime ideals p1 , . . . , pr A, and let
aA = ri=1 qi be a primary decomposition of aA where qi is pi -primary for all
i; see 2.1/6. Since all prime ideals p Ass(aA) are of height 1, as shown above,
we get y aAp for all p Ass (aA) = Ass(aA). But then we conclude from
2.1/15 that
r
r
y aApi A = qi = aA
i=1 i=1

and, hence, that z = A. This shows that p Ap where p varies over all prime
y
a
ideals p A of height 1, is contained in A. The reverse inclusion is trivial.
Let us add a reformulation of Proposition 5 in geometric terms. Namely,

consider a rational function f K, the eld of fractions of A, and a point
x Spec A, given by the prime ideal px A. We say that f is dened at
x if f Apx . Furthermore, x is called a point of codimension 1 if px is of
height 1 in A; the latter amounts to the fact that the closure of x in Spec A
is of codimension 1 in the sense of 7.5/7. Thus, the proposition states for a
Noetherian normal integral domain A that a rational function f on Spec A is
dened everywhere, i.e. is a global section of the structure sheaf of Spec A and,
thus, belongs to A, if and only if it is dened at all points of codimension 1.
Meromorphic functions. Let A be a ring. An element a A is called

regular if a is not a zero divisor in A, i.e. if ab = 0 for an element b A implies
b = 0. It is clear that the set S(A) of all regular elements in A is multiplicative.
So we can look at the localization AS(A) , which is called the total quotient ring
9.3 Divisors 435
of A and has the property that the localization map A - AS(A) is injective.
Let us
point out for a Noetherian ring A that the set of regular elements equals
A pAss(A) p where Ass(A) denotes the set of prime ideals in A that are
associated to the zero ideal 0 A; see 2.1/10.
Now let X be a scheme and consider the subsheaf SX OX given on open
subsets U X by
SX (U ) = {g OX (U ) ; gx S(OX,x ) for all x U };
it is called the subsheaf of regular elements in OX . Using 6.4/7 one shows

SX (U ) S(OX (U )), i.e. SX (U ) is a multiplicative subset consisting of regu-
lar elements in OX (U ).
Denition 6. Let X be a scheme and SX OX the subsheaf of regular ele-

ments. The sheaf MX of meromorphic functions on X is the sheaf of rings asso-
ciated to the presheaf MX given on open subsets U X by U - OX (U )SX (U ) .
The global sections of MX are referred to as the meromorphic functions on X.
By its construction, the sheaf of meromorphic functions MX admits a

canonical morphism OX - MX , which is induced from the localization maps
OX (U ) - OX (U )SX (U ) for U X open. We want to show that this morphism
is a monomorphism. Furthermore, although we do not need this later on, we
show that MX (U ) and MX (U ) coincide for any ane open subset U X,
when X is locally Noetherian.
Lemma 7. Let X be a scheme. Then:

(i) The sheaf SX associates to any ane open subset U X the multi-
plicative set of regular elements in OX (U ), i.e. SX (U ) = S(OX (U )).
(ii) The presheaf MX : U - OX (U )SX (U ) on open subsets U X satises
the sheaf condition (i) of 6.3/2.
(iii) The canonical morphism OX - MX is a monomorphism.

(iv) If X is locally Noetherian, MX is a sheaf on the ane open subsets
of X,that is, it satises the sheaf conditions of 6.3/2 for all coverings of type
U = iI Ui where U and the Ui are ane open subsets of X. In particular,
it follows that the canonical map MX (U ) = OX (U )SX (U ) - MX (U ) is an
isomorphism for every ane open subset U X.
Proof. In the situation of (i) assume that U X is ane and let A = OX (U ).

Then for any g S(A) its associated stalk gx OX,x = Ax at a point x U is
regular, since the localization Ax is at over A. Therefore we get S(A) SX (U ).
The reverse inclusion has already been mentioned above; it follows from the
inclusion OX (U ) - xU OX,x of 6.4/7.
To verify (ii), consider an open subset U X and let fg be a fraction
in OX (U )SX (U ) where f OX (U ) and g SX (U ). Furthermore, assume that
U = (Ui )iI is an open covering of U such that the restriction of fg to each Ui is
trivial. Writing fi = f |Ui and gi = g|Ui for i I, we have fgii = 0 in OX (Ui )SX (Ui )
and, hence, since gi SX (Ui ) S(OX (Ui )), in fact fi = 0. But then, using the
sheaf condition 6.3/2 (i) for OX , we get f = 0 and, thus, fg = 0. Therefore the
presheaf under consideration satises the sheaf condition 6.3/2 (i). Furthermore,
using the canonical monomorphism OX - MX , assertion (iii) follows from
6.5/6 (iii).
Finally, to establish (iv), assume that X is locally Noetherian. In view of
6.6/4 it is enough to consider the case where X is ane and to check the sheaf
condition (ii) of 6.3/2 for an ane open covering X = iI Ui . To do this,
consider fractions hi MX (Ui ) such that hi |Ui Uj = hj |Ui Uj for all i, j I.
Writing hi = fgii for suitable sections fi OX (Ui ) and gi SX (Ui ), we would
like to determine a common denominator g OX (X) that could be used for all
fractions hi . Therefore look at the ideals Ii = (OUi : hi ) OUi , i I, consisting
of all sections g OUi such that ghi OUi . Then Ii coincides with the ideal
(gi OUi : fi ) OUi consisting of all sections g OX such that gfi gi OUi .
Hence, Ii can be interpreted as the kernel of the morphism OUi - OUi /gi OUi
induced from multiplication by fi on OUi , and we see from 6.8/4 that Ii is
quasi-coherent for all i I.
Since the fractions hi coincide on all intersections of the Ui and the ideals
Ii do not depend on the particular representation of hi as a fraction fgii , it is
easily seen that the Ii can be glued to yield a quasi-coherent ideal I OX . It
is enough to show that there exists a global section g I(X) that is regular in
OX (X). Indeed, then the sections ghi OX (Ui ) determine a well-dened global
section f OX (X) and, if g is regular, the fraction h = fg yields an element in
MX (X) satisfying h|Ui = hi for all i I. Hence, we see that the sheaf condition
6.3/2 (ii) holds for ane coverings if I(X) contains a regular element.
Proceeding indirectly, let us assume that I(X) does not contain any regular
element of OX (X). Using the fact that OX (X) is Noetherian by 7.5/4, we know
from 2.1/6 that the zero ideal 0 OX (X) is decomposable in the sense that it
admits a primary decomposition. In particular, it follows from 2.1/10 that I(X)
is contained in the union of all prime ideals belonging to Ass(OX (X)), the set
of prime ideals of OX (X) that are associated to the zero ideal. Hence, by 1.3/7,
there is a prime ideal p Ass(OX (X)) such that I(X) p. Then p is of type
(0 : x) for some x OX (X); see 2.1/9. In particular, I(X) is annihilated by x,
and it follows that Ii (Ui ) is annihilated by x|Ui , for every i I. Since Ii (Ui )
contains gi as a regular element of OX (Ui ), we must have x|Ui = 0 and, hence,
x = 0. However, then we would have p = (0 : x) = OX (X), which is impossible.
Thus, I(X) will contain a regular element and we are done.
Let us discuss the sheaf MX of meromorphic functions in some easy situ-

ations. To begin with, consider an ane scheme X = Spec A that is integral.
Then A is an integral domain by Remark 2 and the total quotient ring AS(A)
is just the eld of fractions K of A. Now consider the constant sheaf KX on X
that associates to any non-empty open subset of X the eld K. Using Lemma 7,
the presheaf MX coincides with KX on all basic ane open subsets in X since
9.3 Divisors 437
AS(A ) = K for any non-zero localization A of A. This shows that KX coincides

with the sheaf MX of meromorphic functions on X.
Proposition 8. Let X be a reduced scheme consisting of nitely many irre-

ducible components X1 , . . . , Xn with generic points x1 , . . . , xn . For example, the
latter is the case if X is Noetherian; see Proposition 7.5/5.
Let Ki = k(xi ) be the residue eld of xi . Then the sheaf MX of mero-
morphic functions on X consists of the functor given on open subsets U X
by
n
1 if xi U
U - i (U ) Ki where i (U ) = .
i=1
0 if xi U
In particular, if X is integral, MX is the constant sheaf KX obtained from the
residue eld at the generic point of X.
Proof. Let U = Spec A be an ane open part of X. Then A is reduced by

Remark 2. Since X consists of nitely many irreducible components, the same
is true for U and we thereby see that there are only nitely many minimal
in A, say corresponding to the generic points xi Xi
prime ideals p1 , . . . , pr
for i = 1, . . . , r. Since ri=1 pi = rad(A) = 0 by 1.3/4 and since any shorter
intersection of some of the pi will not yield 0 by 1.3/8, we see that p1 , . . . , pr are
just the prime ideals associated to the zero ideal 0 A. It follows from 2.1/10
that the set S(A) of regular elements in A is given by A ri=1 pi . In particular,
the prime ideals of the total quotient ring AS(A) are induced by p1 , . . . , pr and
therefore are minimal and maximal at the same time. But then the Chinese
Remainder Theorem yields an isomorphism

r
AS(A) k(xi ).
i=1
n
-
This consideration shows that the functor U i=1 i (U ) Ki is a sheaf
coinciding on ane open parts of X with the sheaf MX of meromorphic func-
tions. But then both sheaves must be the same.
Weil and Cartier divisors. Recall from 7.5/3 the notion of a locally
Noetherian scheme and from 7.5/7 the notion of codimension for closed subsets
of schemes.
Denition 9. Let X be a locally Noetherian scheme. A prime divisor on X is

an irreducible closed subset D X of codimension 1. The set of prime divisors
on X is denoted by PD(X).
A Weil divisor (or just referred to as a divisor) on X is a formal linear
combination
nD D
DPD(X)
with coecients nD Z where the sum is required to be locally nite in the

following sense: each x X admits an open neighborhood U X such that
nD = 0 for almost all D PD(X) satisfying D U = . Then
/ /
/ /
/ n D / := D
/ D /
DPD(X) DPD(X), nD =0
is a closed subset in X, called the support of the divisor.

The Weil divisors onX form an abelian group, which is denoted by Div(X).
Furthermore, a divisor DPD(X) nD D Div(X) is called eective if nD 0
for all D PD(X).
We may view Div(X) as a subgroup of ZPD(X) , the group of all families

(nD )DPD(X) of integers nD Z. Also note that this way Div(X) contains
Z(PD(X)) , the group of all families (nD )DPD(X) , where nD = 0 for almost all
D PD(X), and that Div(X) = Z(PD(X)) if X is quasi-compact. By convention
we get Div(X) = 0 for PD(X) = .
Now assume X to be a Noetherian scheme that is integral and normal.
We want to associate a Weil divisor to any invertible meromorphic function
f MX (X) = K where K is the residue eld at the generic point of X. Let
D be a prime divisor on X and x D its generic point. Then, using 7.5/1,
one shows for any ane open part U X containing x that the restriction
D U is a prime divisor on U . Therefore the local ring OX,x is a Noetherian
normal integral domain of dimension 1 with eld of fractions K and, hence,
by Lemma 4, a discrete valuation ring. Let D : K - Z be the associated
valuation and set nD = D (f ). We claim that nD = 0 for almost all prime
divisors D on X. Indeed, choose a non-empty ane open subset U X, say
U = Spec A. If f |U = g 1 h for elements g, h A, we see that f restricts to a unit

in OX on the complement of the closed subset Z = (X U ) V (g) V (h) in X.
Since Z consists of nitely many irreducible components by Proposition 7.5/5,
all of them of codimension 1 due to the fact that X is irreducible, there are
only nitely many prime divisors of X contained in Z. Therefore the sum

(f ) = nD D
DPD(X)
where PD(X) denotes the set of prime divisors in X, has at most nitely many
non-trivial coecients nD and, hence, is a divisor on X. The latter is called
the principal divisor attached to f . It is clear that the principal divisors form a
subgroup in the group of Weil divisors Div(X).
Denition 10. Let X be a Noetherian scheme that is integral and normal. Let
K = MX (X) be its eld of meromorphic functions; see Proposition 8.
(i) Consider
a meromorphic function f K . Then the above dened divi-
sor (f ) = DPD(X) nD D is called the principal divisor associated to f . The
principal divisors form a subgroup of the group Div(X) of all divisors on X.
9.3 Divisors 439
(ii) Two divisors D1 , D2 Div(X) are called linearly equivalent if D1 D2

is a principal divisor.
(iii) The quotient of Div(X) by the subgroup of principal divisors is called
the divisor class group of X; it is denoted by Cl(X).
Let us look at the special case where X = PnL = ProjL Lt0 , . . . , tn and L
is a eld. Then X is integral and normal, the latter due to the fact that poly-
nomial rings over elds are factorial and, hence, normal. Furthermore, we will
use the basic fact that any prime factorization f = p1 . . . pr of a homogeneous
polynomial f Lt0 , . . . , tn is homogeneous in the sense that all prime factors
pi are homogeneous. Indeed, the product of the leading homogeneous parts of
the pi yields the leading homogeneous part of f and if f is homogeneous, the
same must be true for the pi .
Lemma 11. Let X = PnL = ProjL Lt0 , . . . , tn where n > 0 and L is a eld.
Then the set PD(X) of prime divisors on X corresponds bijectively to the set
of principal ideals in Lt0 , . . . , tn that are generated by homogeneous prime
polynomials.
Proof. We can conclude from 9.1/15 that the maps
Y - I+ (Y ), a - V+ (a),
dene mutually inverse bijections between closed subsets in X and graded ideals
a Lt0 , . . . , tn + satisfying a = rad+ (a). Furthermore, note that any graded
ideal a Lt0 , . . . , tn is automatically contained in the irrelevant ideal, un-
less it is the unit ideal. Combining this fact with the argument given in the
proof of 6.1/15, one shows that a closed subset Y X is irreducible if and
only if its associated ideal I+ (Y ) is a graded prime ideal in Lt0 , . . . , tn + .
In particular, the prime divisors on X correspond bijectively to the graded
prime ideals p Lt0 , . . . , tn of height ht(p) = 1; note that then automatically
(t0 , . . . , tn ) p, since n > 0 implies ht(t0 , . . . , tn ) 2. We claim that these
prime ideals are precisely the ones that are generated by a homogeneous prime
element p Lt0 , . . . , tn .
Indeed, if p is any non-zero graded prime ideal in Lt0 , . . . , tn , consider a
homogeneous element f = 0 in p. Using the homogeneous prime factorization
of f , we see that p must contain a homogeneous prime factor p of f and,
hence, that p contains the graded prime ideal (p). From this we conclude that
p coincides with (p) if its height is 1 and, furthermore, that any ideal generated
by a homogeneous prime element in Lt0 , . . . , tn must be of height 1. Thus, we
are done.
In particular, the preceding result enables us to dene the degree of a prime

divisor D on X = PnL by deg D = deg p if D = V+ (p) for a homogeneous prime
element p Lt0 , . . . , tn .
Proposition 12. For a eld L and X = PnL , n > 0, the degree morphism

deg : Div(X) - Z, nD D - nD deg D,
DPD(X) DPD(X)
is trivial on principal divisors and induces an isomorphism Cl(X) - Z.
Proof. We will freely use the bijective correspondence of Lemma 11 between

prime divisors on X and principal ideals in Lt0 , . . . , tn that are generated by
homogeneous prime polynomials. Looking at a hyperplane like V+ (t0 ) X, we
see that deg is surjective. Next observe that the eld MX (X) of meromorphic
functions on X coincides with the eld of all fractions in L(t0 , . . . , tn ) that are
homogeneous of degree 0; see Proposition 8. Therefore, if f = g 1 h MX (X)
for homogeneous polynomials g, h Lt0 , . . . , tn , then the homogeneous
prime factorization applied to g and h yields a prime factorization of type
f = p11 . . . prr with a unit L , pairwise non-equivalent homogeneous prime
elements pi Lt0 , . . . , tn , and exponents i Z such that ri=1 i deg pi = 0.
For i = 1, . . . , r let Di be the prime divisor associated to the polynomial
pi . We claim that the principal divisor (f ) equals ri=1 i Di and, hence, has
degree 0. To verify this, let D be any prime divisor on X, say obtained from a
homogeneous prime polynomial p Lt0 , . . . , tn of degree d, and let x be the
generic point of D. Then there is an index j {0, . . . , n} such that x D+ (tj ),
and we see from 9.1/13 that the corresponding prime ideal px Lt0 , . . . , tn (tj )
is generated by td j p. Thus, using Lemma 4, the local ring OX,x is a discrete
deg pi
valuation ring with maximal ideal generated by td j p. Now observe that tj pi
for any i may be viewed as an element of Lt0 , . . . , tn (tj ) OX,x and that this
element generates the maximal ideal in OX,x if the prime elements p and pi dier
deg pi
at most by a unit. On the other hand, if the latter is not r the case, tj pi will
be invertible in OX,x . This shows that, indeed, (f ) = i=1 i Di as claimed and
that the degree morphism deg : Div(X) - Z is trivial on principal divisors.
It remains to show that any divisor Q = DPD(X) nD D of degree 0
is principal. To do this, choose for each prime divisor D PD(X) a homo-
geneous prime polynomial fD Lt0 , . . . , tn such that D = V+ (fD ). Then
nD
f = DPD(X) D f is of degree 0 and, hence, a meromorphic function in
MX (X) . Since (f ) = Q by a reasoning as exercised above, we are done.
The concept of associating to a meromorphic function a Weil divisor is

generalized by the concept of Cartier divisors. To dene these on a scheme X,
we denote by MX MX the subsheaf of invertible meromorphic functions

on X and, likewise, by OX OX the subsheaf of invertible functions of the
structure sheaf of X.
Denition 13. Let X be a scheme. Then CaDiv(X) = (X, MX /OX

) is called
the group of Cartier divisors on X. Thus, using 6.5/8, a Cartier divisor f on
X is represented by a family of sections fi MX (Ui ) where (Ui )iI is an open
9.3 Divisors 441
covering of X such that (fj1 fi )|Ui Uj OX

(Ui Uj ) for all i, j I. If, in

addition, fi OX (Ui ) MX (Ui ), the divisor f is called eective.
In analogy with Weil divisors, the group law on (X, MX /OX
) is written
additively.
A Cartier divisor is called principal if it belongs to the image of the canonical
map (X, MX ) - (X, M /O ). Two Cartier divisors in CaDiv(X) are
X X
called linearly equivalent if their dierence is principal.
The quotient of CaDiv(X) by the subgroup of principal divisors is denoted
by CaCl(X) and is called the group of classes of Cartier divisors.
For a Noetherian scheme X that is integral and normal, one can produce
Weil divisors from Cartier divisors. Indeed, consider a Cartier divisor f in
CaDiv(X), say represented by sections fi (X, MX (Ui )) where (Ui )iI is
an open covering of X and fj1 fi OX
(Ui Uj ) for all i, j I. Now let D be a
prime divisor on X and let x be its generic point. Then we know from Lemma 4
that OX,x is a discrete valuation ring. Let D be the associated valuation on its
eld of fractions and set nD (f ) = D (fi ) for any i I such that Ui = . Then
nD (f ) is well-dened since MX is a constant sheaf by Proposition 8 and since
the dierence fj1 fi is invertible in OX (Ui Uj ). Furthermore, as discussed for
principal Weil divisors, we have nD = 0 for almost all D PD(X). Thus, associ-
ating to any f CaDiv(X) the corresponding Weil divisor DPD(X) nD (f ) D
yields a canonical group homomorphism : CaDiv(X) - Div(X), mapping
principal to principal and eective to eective divisors.
Proposition 14. Let X be a Noetherian scheme that is integral and normal.

Then the canonical homomorphism : CaDiv(X) - Div(X) satises the fol-
lowing assertions:
(i) The image (f ) Div(X) of a divisor f CaDiv(X) is eective if and
only if f is eective.
(ii) is injective, maps principal divisors bijectively to principal divisors, and
induces an injective homomorphism
CaCl(X) - Cl(X).
(iii) If, in addition, X is factorial, the map CaCl(X) - Cl(X) of (ii) is

surjective and, hence, an isomorphism.
Proof. First, observe that the sheaf of meromorphic functions MX equals the
constant sheaf KX for K the residue eld at the generic point of X. More
precisely, K coincides with the eld of fractions of OX (U ) for any non-empty
ane open subset U X or even of any stalk OX,x at a point x X; see
Proposition 8. To verify our assertions, consider a Cartier divisor f CaDiv(X),
given by sections fi MX (Ui ) for some open covering (Ui )iI of X. Rening it
if necessary, we may assume that the covering (Ui )iI is ane, say Ui = Spec Ai
for i I. Since X is integral, all Ai are integral domains by Remark 2 and
7.5/1. In addition, they are Noetherian by 7.5/4 and normal by Remark 2

again. Now assume that the Weil divisor (f ) = DPD(X) nD (f ) D associated
to f is eective. Then all values nD (f ) for D varying over the prime divisors
of X are non-negative, and we conclude from Proposition 5 that fi Ai for all
i I. Therefore f is eective. In the same way we get fi1 Ai and, hence,
fi Ai = OX
(Ui ) if both f and f are eective. In particular, f is trivial
if (f ) is trivial and, hence, CaDiv(X) - Div(X) is injective. Furthermore,
combining the latter with the denition of principal divisors, we can conclude
that principal Cartier divisors are mapped bijectively to principal Weil divisors.
Now assume that, in addition, X is factorial. It remains to show that the
homomorphism CaCl(X) - Cl(X) is surjective. For this it is enough to
show that any prime divisor D on X belongs to the image of this map. Indeed,
let I OX be the vanishing ideal of D; the latter is quasi-coherent by 7.3/5.
Then, for any x D, the ideal Ix OX,x is a prime ideal of height 1 and, hence,
generated by a single prime element since OX,x is factorial. On the other hand,
we have Ix = OX,x for x X D. Furthermore, for any ane open subset
U X, the ideal I|U is associated to a nitely generated ideal in OX (U ),
since this ring is Noetherian by Proposition 7.5/4. Therefore, choosing for each
x X a generator of the ideal Ix OX,x and extending it to a section fx on
a suitably small ane open neighborhood Ux X, we can use the implication
(ii) = (i) of 8.4/2 and thereby assume that I|Ux is generated by the section
fx OX (Ux ). But then the family f = (fx )xX denes a Cartier divisor on X,
whose associated Weil divisor is D. Thus, we are done.
Finally we want to discuss the relationship between Cartier divisors and

invertible sheaves. Let us consider a Cartier divisor D represented by a family
of sections fi MX (Ui ) where (Ui )iI is an open covering of X. Then we can
dene a subsheaf OX (D) MX by the condition that OX (D)|Ui = fi1 OX |Ui .
Indeed, since fi and fj , for any indices i, j I, dier on Ui Uj by a unit
in OX
(Ui Uj ), the subsheaves fi1 OX |Ui MX |Ui coincide on overlaps of
the Ui and therefore can be glued to produce a well-dened subsheaf OX (D)
of MX . If D is eective and fi OX (Ui ) for all i I, the sheaf OX (D) may
be interpreted as the sheaf of meromorphic functions having poles of a type
not worse than indicated by the divisor D. In particular, the unit section of OX
then gives rise to a global section in OX (D) and, hence, to a monomorphism
OX - OX (D).
It is easily seen for general D that OX (D) is independent of the meromorphic
functions fi representing D and, furthermore, since the fi are units in MX , that
OX (D) is an invertible sheaf. Thereby we get a canonical map
CaDiv(X) - Pic(X), D - OX (D),
which is easily seen to be a group homomorphism. Indeed, the tensor product of

two invertible subsheaves L, L MX , say with free generators fi , fi MX (Ui )
on an open covering (Ui )iI of X, is given by the subsheaf L L MX that
on each Ui is generated by fi fi .
9.3 Divisors 443
Proposition 15. Let X be a scheme. Then the above dened homomorphism

CaDiv(X) - Pic(X) induces a monomorphism
CaCl(X) - Pic(X).
Its image consists of all invertible subsheaves of MX .
Furthermore, if X is Noetherian and reduced, any invertible OX -module
can be viewed as a subsheaf of MX and, hence, CaCl(X) - Pic(X) is an
isomorphism in this case.
Proof. The rst part is easy to see. Let D be a Cartier divisor on X, represented
by a family of sections fi MX (Ui ) where (Ui )iI is an open covering of X.
Then the associated invertible sheaf OX (D) is a subsheaf of MX and is trivial if
and only if it admits a global generator f MX (X). If the latter is the case, f
diers on Ui from fi1 by a unit in OX
(Ui ), and D is the principal Cartier divisor
1
given by f . In particular, OX (D) is trivial if and only if D is principal so
that CaDiv(X) - Pic(X) induces a monomorphism CaCl(X) - Pic(X).
Furthermore, if L is an invertible subsheaf of MX , choose a trivializing open
covering (Ui )iI of X and let fi generate L on Ui . We may assume that all Ui
are ane and, furthermore, using Lemma 7, that fi MX (Ui ) = (Ai )S(Ai ) for
Ai = OX (Ui ). Then we can write fi = s1 i ai for some regular element si S(Ai )
and some ai Ai . Since fi has the property that cfi = 0 for c Ai implies
c = 0, we see that the same is true for ai . Therefore ai S(Ai ) and, hence, fi
is invertible in (Ai )S(Ai ) . Thereby we conclude fi MX (Ui ), and it follows that
the fi1 give rise to a Cartier divisor on X, whose associated invertible sheaf
coincides with L.
Now assume that X is Noetherian and reduced and let L be an invertible
sheaf on X. We have to show that L may be viewed as a subsheaf of the sheaf
MX of meromorphic functions on X. Let x1 , . . . , xn be the generic points of X
and let Ui be an ane open neighborhood of xi such that Ui is disjoint from
all irreducible components {xj } for i = j. Choosing Ui small enough, we may
assume that L is trivial on each Ui . Then Ui {xi } for all i, and we see that
U = ni=1 Ui is a disjoint union of ane open subsets of X, containing all generic
points of X. Furthermore, L|U is trivial.
Let : U - X be the inclusion map and consider the direct image (L|U )
of the restriction of L to U . It is an OX -module and consists of the functor
U - L(U U ), for U X open.
Furthermore, the restriction morphisms L(U ) - L(U U ) dene a morphism
of OX -modules L - (L|U ), which we claim is injective. Indeed, this can be
checked locally on X and, since L is locally trivial, we may replace L by OX .
Then it is enough to show that all restriction maps OX (U ) - OX (U U )
for U X ane open are injective. However, the latter is easily seen to be
true, since a function on a reduced ane scheme U vanishes if and only if it
vanishes at the generic points of U ; this follows from 1.3/4. Now look at the
chain of inclusions
L - (L|U ) (OX |U ) - (MX |U )
and observe that the canonical morphism MX - (MX |U ) is an isomor-

phism by Proposition 8. Therefore L can be viewed as a subsheaf of MX and
we are done.
Corollary 16. Let X be a Noetherian scheme that is integral and factorial.

Then the groups
Cl(X) of classes of Weil divisors on X,
CaCl(X) of classes of Cartier divisors on X, and
Pic(X) of isomorphism classes of invertible sheaves on X
are isomorphic in a canonical way.
Proof. Use Propositions 14 and 15.
Corollary 17. Let L be a eld, and consider the projective n-space X = PnL for
n > 0. Then the degree morphism deg : Div(X) - Z gives rise to canonical
isomorphisms
Pic(X) CaCl(X) Cl(X) Z,
such that Serres twisted sheaf OX (n) Pic(X) corresponds to n Z for all
integers n. In particular, up to isomorphism, Serres twisted sheaves are the only
invertible sheaves on X = PnL .
Proof. Using 9.2/7, as well as Corollary 16 in conjunction with Proposition 12,

it remains to show that Serres twisted sheaf OX (1) corresponds to a divisor
D Div(X) of degree 1. To do this, write X = ProjL Lt0 , . . . , tn and look
at the prime divisor D = V+ (t0 ), which is a divisor of degree 1 and, hence,
corresponds to 1 Z under the isomorphism of Proposition 12. The Cartier
divisor corresponding to D in the setting of Proposition 14 is given by the
meromorphic functions fi MX (D+ (ti )), where fi = t1 i t0 for i = 0, . . . , n.
Then OX (D) is generated by fi1 = ti t1 0 on D (t
+ i ), and multiplication by t0 in
the eld of fractions of Lt0 , . . . , tn shows that OX (D) is isomorphic to OX (1).

Exercises
1. Let X be a Noetherian normal scheme or, more generally, a Noetherian scheme
such that all stalks OX,x at points x X are integral domains. Show that X is
the disjoint union of its irreducible components, the latter being open in X.
2. Show Cl(AnK ) = 0 for the ane n-space AnK over a eld K. Hint: Use the fact
that the polynomial ring Kt1 , . . . , tn is factorial and, hence, that every prime
ideal of height 1 is principal.
3. Let A be a Noetherian integral domain. As a generalization of Exercise 2 show
that A is factorial if and only if the scheme X = Spec A is normal and satises
Cl(X) = 0. Hint: For the only-if part proceed as in Exercise 2. For the if part
9.3 Divisors 445
show that every prime ideal p A of height 1 is principal and that all principal
ideals generated by irreducible elements are prime.
4. Let X be a Noetherian scheme that is integral and normal. Furthermore, consider
a non-trivial open subscheme U X and set Z = X U . Show:
(a) The map D - DU on prime divisors D PD(X) gives rise to a surjective
homomorphism Cl(X) - Cl(U ), which is an isomorphism if codimX Z 2
for every irreducible component Z of Z.
(b) If Z is irreducible and of codimension 1 in X, there is a canonical exact
sequence Z - Cl(X) - Cl(U ) - 0.
(c) Discuss the case where X is the projective n-space PnK = ProjK t0 , . . . , tn
over a eld K and where U = D+ (t0 ) = AnK .
5. Rational maps: Let X, Y be schemes. Two morphisms f : U - Y and
g: V - Y given on dense open subschemes U, V X are called equivalent
if there exists an open dense subscheme W U V such that f |W = g|W . The
associated equivalence classes are called rational maps from X to Y and are de-
noted by dashed arrows X - Y . A rational map h : X - Y is said to be
dened at a point x X if there exists a representative f : U - Y of h where
x U . Show:
(a) The intersection of two dense open subsets in X is dense open in X again.
Hence, the above dened relation is an equivalence relation, indeed.
(b) Let h : X - Y be a rational map and assume that X is Noetherian and
normal and Y ane. Then, if h is dened at all points of codimension 1, it is
represented by a unique scheme morphism X - Y.
6. Compute the group of Cartier divisor classes CaCl(X) for Neiles parabola
X = Spec Kt1 , t2 /(t22 t31 ) over a eld K. Show that it is isomorphic to the
additive group of K. Can we talk about Cl(X)? What about Pic(X)? Hint: Con-
sider the canonical morphism : A1K - X as in Exercise 6.7/3, the so-called
normalization morphism of X, and use the exact sequence of sheaves of abelian
groups 0 - OA 1 /OX - M /O
X X
- M / O 1
X A
- 0.
K K
7. Show Pic(A1K ) = Z for the ane line with double origin over a eld K.
8. Canonical sheaves on projective spaces: Consider the projective n-space PnR over
some ring R and let n = nPn /R be the sheaf of relative dierential forms of
R
degree n; it is called the canonical sheaf on PnR and is dened as the nth exterior
power of the sheaf of relative dierential forms of degree 1 (see Section 8.2). Show
that n is an invertible sheaf on PnR and describe n in terms of Serres twisted
sheaves. Furthermore, in the case of a eld R, specify a Weil divisor D inducing
n via the correspondence of Corollary 16.
9. Algebraic varieties over elds via schemes: An algebraic variety over a eld K
is dened as a separated integral K-scheme of nite type that remains integral
when extending the base K to an algebraic closure K. This way the algebraic
varieties over K may be viewed as a full subcategory of the category of K-schemes.
Show that the ber product X K Y of two algebraic varieties over K is an
algebraic variety again and, thus, that the ber product of K-schemes gives rise
to a cartesian product in the category of algebraic varieties over K.
9.4 Global Sections of Invertible Sheaves
Let X be a scheme and L an invertible sheaf on X. The aim of the present

section is to study morphisms from X or parts of it into a projective n-space
PnR . Let us start by some preparations. For any global section l (X, L) we
set
V (l) := supp L/OX l = {x X ; OX,x l Lx }
and call this the zero set of l.1 Thus, its complement
Xl = {x X ; OX,x l = Lx }
consists of all points x X where l can be viewed as a generator of the stalk

Lx . Recall from 9.2/4 that a family of global sections li (X, L), i I, is
a set of generators of L if the morphism of OX -modules OX
(I) - L given by
these sections is an epimorphism. Using 6.5/9, the latter is equivalent
to the
fact that the li generate Lx at every point x X, or even to X = iI Xli .
Remark 1. Let l be a global section of an invertible sheaf L on a scheme X.

(i) The zero set V (l) is closed in X and, hence, its complement Xl is open
in X.
(ii) The canonical morphism OX - L given by multiplication with the
section l is an isomorphism when restricted to Xl . Thus, indeed, we can say
that l generates L on Xl .
Proof. The rst assertion can be tested locally on X. So we may assume that
X is ane, say X = Spec A, and L = OX . Then we can write
V (l) = {x Spec A ; l is not a unit in OX,x = Ax }.
It follows from 1.2/7 in conjunction with 1.2/5 (iii) that l induces a non-unit in
Ax if and only if l is contained in the prime ideal px A given by x. Therefore
V (l) coincides with the ordinary zero set
V (l) = {x X ; l px },
and we see that V (l) is closed in X. Hence, its complement Xl is open in X.

The second assertion is a consequence of 6.5/3.
Remark 2. For an invertible sheaf L on a scheme X consider the open sub-

scheme Xl X associated to a global section l (X, L). Furthermore, let F
be a quasi-coherent OX -module.
(i) If X is quasi-compact, then for any global section r (X, F) satisfying
r|Xl = 0, there exists an exponent i N such that r li = 0 in (X, F Li )
1
Note that for L = OX this notion of zero set coincides with the one introduced in
Sections. 6.1 and 7.3; see the proof of Remark 1.
9.4 Global Sections of Invertible Sheaves 447
where Li is the ith tensor power of L and, accordingly, li is to be interpreted

as the ith tensor power of l.
(ii) If X is quasi-compact and quasi-separated, then for any local section
s (Xl , F), there exists an exponent i N such that s li |Xl extends to a
global section r (X, F Li ).
Proof. The argument is similar to the one used in the proof of 6.8/6. To begin
with, assume that X is ane, say X = Spec A, and that L is trivial. Then
L OX and tensoring with sections of L amounts to multiplying with the
corresponding sections in OX . In particular, we may view l as an element of
A. Since F is a quasi-coherent OX -module on the ane scheme Spec A, we
know from 6.8/10 that it is associated to the A-module F = F(X). Hence,
the restriction map F(X) - F(Xl ) can be identied with the canonical
localization map F - F A Al . The kernel of the latter consists of all elements
in F that are annihilated by a certain power of l and, furthermore, every element
of F A Al admits an extension to F if we multiply by a certain power of l; see
also 6.8/6. This settles assertions (i) and (ii) if X is ane.
In the general case, we choose an ane open covering (Uj )jJ of X trivial-
izing L. Taking X to be quasi-compact, we may assume that J is nite. Now
let r (X, F) such that r|Xl = 0. Then r|Xl Uj = 0 for each j J and, by
the ane special case, we can nd an exponent i > 0 such that r li |Uj = 0.
Since J is nite, we may assume that i is independent of j so that r li = 0 on
all of X. Thus, assertion (i) is clear.
On the other hand, given s (Xl , F), the ane special case shows that
each restriction s|Xl Uj can be extended to a section rj (Uj , F Li ) by
tensoring with a certain power li where we can choose i independent of j. Then
rj and rj will coincide on Xl Uj Uj for any indices j, j J. Since Uj Uj is
quasi-compact in the situation of (ii), we may apply (i) to the scheme Uj Uj in
place of X. Thus, enlarging i, we can assume that rj and rj coincide on Uj Uj .
But then the rj dene a section r (X, F Li ) such that r|Xl = s li |Xl .
This settles assertion (ii).
Remark 3. Let f : X - Y be a morphism of schemes. For an invertible

sheaf G on Y consider the canonical morphism
f# : G - f (f (G)),
also referred to as pull-back of sections with respect to f . Then:

(i) For any global section t (Y, G) its pull-back s = f # (Y )(t) is a global
section in f (G).
(ii) In the situation of (i) we have f 1 (V (t)) = V (s) for the zero sets of the
global sections t and s = f # (Y )(t), as well as f 1 (Yt ) = Xs .
(iii) If t0 , . . . , tn (Y, G) are global generators of G, then the pull-backs
si = f # (Y )(ti ), i = 0, . . . , n, form a system of global generators for f (G).
Proof. Assertion (i) is clear by the denition of direct image sheaves. To check
(ii), we may work locally on Y and X and thereby assume that X and Y are
ane. Moreover, we may assume G = OY . Then f # coincides with the canonical
map f # : OY - f OX and the assertion follows from 6.2/4. In the same way
we assume for assertion (iii) that X and Y are ane, and that G = OY . If
t0 , . . . , tn OY (Y ) generate OY as OY -module, the ti generate the unit ideal
in OY (Y ). This implies that the pull-backs s0 , . . . , sn generate the unit ideal in
OX (X) and, hence, that the si form a set of generators for f (OY ) = OX .
Now xing a base ring R (or, more generally, a base scheme S), we start
looking at R-scheme morphisms of type
f: X - PnR = ProjR Rt0 , . . . , tn .
To simplify our notation, we will sometimes write P in place of PnR for the
projective n-space. As usual, let OP (1) be Serres twist of the structure sheaf
OP . Then we know from 9.2/7 that OP (1) is an invertible sheaf on PnR , and the
same is true for L = f (OP (1)) on X by 6.9/6. In particular, we can look at the
pull-back morphism
f # : OP (1) - f (f (OP (1)) = f (L)
and apply Remark 3. Thus, the global generators t0 , . . . , tn (PnR , OP (1)) give
rise via f # to global generators s0 , . . . , sn of L such that
f 1 (PnR,ti ) = Xsi , i = 0, . . . , n,
where PnR,ti = D+ (ti ) AnR , as we have seen in Section 9.2. It follows that X
is covered by the open subsets Xsi , just as PnR is covered by the open subsets
PnR,ti . Thus, the morphism f : X - Pn can be interpreted as being obtained
R
by gluing the induced morphisms
f i : Xs i - AnR = PnR,ti , i = 0, . . . , n.
Using 7.1/3, the latter correspond to R-algebra morphisms

't tn (
fi# : R
0
,..., - (Xsi , OX ), i = 0, . . . , n,
ti ti
and we claim that, suggestively, fi# can be characterized by

tj -
sj
, j = 0, . . . , n.
ti si
Indeed, ti generates OP (1) on PnR,ti , and we can write
tj
tj = ti , j = 0, . . . , n,
ti
tj
interpreting the fractions ti
as sections in OP (PnR,ti ). Furthermore, applying the
pull-back
f # : OP (1) - f (L)
and using the compatibility with the pull-back on the level of structure sheaves
f # : OP - f (OX )
yields the equations

t
j
sj = fi# si , j = 0, . . . , n.
ti
Since si is a free generator of the restriction of L to Xsi , there are unique sections
t
cij OX (Xsi ), j = 0, . . . , n, satisfying sj = cij si on Xsi , namely cij = fi# ( tji ),
sj
and it makes sense to write si in place of cij . Thus, summing up, we can assert
that any R-morphism f : X - PnR provides global generators s0 , . . . , sn of the
invertible sheaf L = f (OP (1)), from which f can be recovered. In the following
we want to show that, more generally, one can actually construct morphisms
into the projective n-space PnR from global generators of invertible sheaves.
Theorem 4. Let X be an R-scheme and L an invertible sheaf on X with global

sections s0 , . . . , sn (X, L) that generate L. Then there is a unique morphism
of R-schemes
- Pn = Proj R
f: X R R t 0 , . . . , tn
,
together with an isomorphism of OX -modules f (OP (1)) - L, such that

f # (ti ) = si for i = 0, . . . , n under the resulting pull-back morphism

f # : OP (1) - f f (OP (1)) - f (L).
It follows that f 1 (PnR,ti ) = Xsi for i = 0, . . . , n, and that f is obtained by

gluing the R-morphisms
fi : Xsi - Pn ,
R,ti
given by the R-algebra morphisms

't tn ( - t0 tn s0 sn
0
fi# : R , . . . , (Xsi , OX ), ,..., - ,..., .
ti ti ti ti si si
Proof. If there is an R-morphism f : X - Pn enjoying the stated properties,

R
then, as we have seen above, it is obtained by gluing the R-morphisms
fi : X s i - PnR,ti , i = 0, . . . , n,
given by
't tn ( t0 tn s0 sn
fi# : R
0
,..., - (Xsi , OX ), ,..., - ,..., .
ti ti ti ti si si
This implies the uniqueness assertion. On the other hand, in order to actually
construct f from the given data, we look at the morphisms fi : Xsi - PnR,ti as
dened above and show with the help of 7.1/2 that these can be glued to yield
a morphism f : X - Pn as required. First, the open subsets Xs cover X,
R i
since the sections s0 , . . . , sn form a set of global generators of L. Next, observe
that PnR,ti PnR,tj , for any indices i and j, coincides with the basic open part
D+ (ti tj ) PnR,ti , where tji is invertible and, hence, that fi1 (PnR,ti PnR,tj ) is the
t
s
open part in Xsi where sji is invertible. However, the latter is just Xsi Xsj so
that we get fi1 (PnR,ti PnR,tj ) = Xsi Xsj . Furthermore, it is easy to see that the
restrictions of fi and fj to Xsi Xsj coincide, due to the canonical commutative
diagram
't tn ( tk s
0
R ,..., - Xs Xs , OX , - k ,
ti ti t itj i

j
ti si

't t ( tk s
0
R ,...,
n - Xsj Xsi , OX , - k .
ti
tj tj tj tj sj
In particular, the global sections s0 , . . . , sn (X, L) give rise to a well-dened
R-morphism f : X - Pn satisfying f 1 (Pn ) = Xs for all i.
R R,ti i
Now consider the inverse image L = f (OP (1)). The global sections
t0 , . . . , tn of OP (1) give rise to global sections s0 , . . . , sn (X, L ), which form a
set of generators of L . Furthermore, L trivializes with respect to the open cov-
s
ering U = (Xsi )i=0,...,n of X, and we see that the cocycles ( si ), ( ssji ) C 1 (U, OX )
j

coincide. Hence, using 9.2/8, we obtain an isomorphism L L of invertible
sheaves on X, as desired.
Corollary 5. For any R-scheme X, the R-morphisms X - Pn correspond

R
bijectively to the equivalence classes of data (L, s0 , . . . , sn ), where L is an in-
vertible sheaf on X with global sections s0 , . . . , sn (X, L). Two such data
(L, s0 , . . . , sn ) and (L , s0 , . . . , sn ) are called equivalent if there is an isomor-
phism of OX -modules L - L mapping si to si for all i.
Corollary 6. Let X be an R-scheme and L an invertible sheaf on X with global

sections s0 , . . . , sn (X, L). Then there is a well-dened R-morphism
f: U - PnR = ProjR Rt0 , . . . , tn ,
n
for U = i=0 Xsi the open part of X, where the sections si generate L.
Next we want to study so-called ample and very ample invertible sheaves,
which are convenient for constructing immersions into projective n-spaces. As
a technical tool, we need to introduce quasi-ane schemes.
Proposition and Denition 7. For a quasi-compact scheme U the following

(i) There exists an open immersion U - X into an ane scheme X.

(ii) The canonical morphism U - Spec (U, OU ) derived from the iden-
tity map on (U, OU ) in the manner of 7.1/3 is an open immersion.
(iii) There exists a locally closed immersion U - X into an ane
scheme X.
A quasi-compact scheme U verifying the equivalent conditions (i), (ii), and
(iii) is called quasi-ane.
Proof. Starting with the implication (i) = (ii), let U be a quasi-compact

scheme, X = Spec A an ane scheme, and : U - X an open immersion. We
write B = (U, OU ) for the ring of global sections of the structure sheaf of U .
Then we see from 7.1/3 that is characterized by a ring morphism A - B.
Now, considering the morphism Spec B - Spec A associated to A -B
in the sense of 6.6/9, as well as the canonical morphism U - Spec B associ-
ated to the identity map id : B - (U, OU ) in the sense of 7.1/3, we get the
factorization
: U - Spec B - Spec A
of and it remains to show that U - Spec B is an open immersion.

To do this, x elements f1 , . . . , fr A such that U = ri=1 D(fi ). Then the
sheaf property of OU or OX gives rise to the exact diagram

r
-

r
(U, OU ) - Afi - Afi fj ,
i=1 i,j=1
and the latter remains exact if we tensor it over A with any localization Ag of
A by an element g A; use that Ag is at over A by 4.3/3. As we are dealing
with nite cartesian products, these can be interpreted as direct sums, which
commute with tensor products. Thus, for any g A, we arrive at the exact
diagram
r
-
r
(U, OU ) A Ag - Agfi - Agfi fj ,
i=1 i,j=1
showing that the canonical map (U, OU ) A Ag - (U D(g), OU ) is an

isomorphism. Therefore, if we look at the decomposition
U D(g) - Spec Bg - Spec Ag
of over the open set D(g) Spec A, we see that U D(g) - Spec Bg
corresponds to an isomorphism Bg - (U D(g), OU ) on the level of
global sections. This implies that U D(g) - Spec Bg is an isomorphism,
too, if U D(g) is ane. For example, the latter is the case for D(g) U , since
then U D(g) = D(g).
Using this for g = f1 , . . . , fr , we see that U is locally an open subscheme
of Spec B, and we can conclude that U - Spec B is an open immersion, as
stated in (ii).
The implication (ii) = (iii) being trivial, it remains to derive (i) from
(iii). To do so let : U - X be a locally closed immersion where U is quasi-
compact and X is ane, say X = Spec A. Then we can replace X by the so-
called schematic closure of U in X and thereby assume that the ring morphism
A - (U, OU ) corresponding to is injective. Being a bit more precise,
let a A be the kernel of A - (U, OU ). Then admits a factorization

U - Spec A/a - Spec A by 7.1/3 and we can interpret as being obtained
from by means of the base change Spec A/a - Spec A. Thus, is a locally
closed immersion again, since immersions are stable under base change (7.3/13).
Consequently, we can replace by and thereby assume that the corresponding
ring morphism A - (U, OU ) is injective.
Now let U X be an open subscheme such that decomposes into a closed
immersion U - U and the open immersion U - X. Since U is quasi-

compact, we may assume that U is quasi-compact as well, say U = ri=1 D(fi )
for elements fi A. Then consider the morphisms
i : U D(fi ) - U D(fi ) = D(fi ), i = 1, . . . , r,
induced from and write B = (U, OU ). By the same argument as applied
above, there is a canonical bijection B A Afi - (U D(fi ), OU ), which
we will use as an identication. The morphisms i are closed immersions of
ane schemes, since U - U is a closed immersion and the D(fi ) are ane.
Therefore i corresponds to a surjection # - B A Af . Since A -B
i : Afi i
was assumed to be injective and Afi is at over A, it follows that #

i is injective
as well and therefore bijective. Consequently, all i are isomorphisms and it
follows that U - U is an isomorphism. But then : U - X is an open
immersion and we are done.
Let R be a ring and X an R-scheme. Recall from 8.3/4 that X is called

locally of nite type over R if each x X admits an ane open neighborhood
U = Spec A X such that the corresponding ring homomorphism R -A
is of nite type in the sense of 8.3/1, namely, that it extends to a surjection
Rt1 , . . . , tn - A for some polynomial variables t1 , . . . , tn .2 Moreover, X is
called of nite type over R if, in addition, it is quasi-compact. Also note that a
morphism of ane schemes Spec A - Spec R is (locally) of nite type if and
only if the associated ring morphism R - A is of nite type; see 8.3/5.
Denition 8. Let X be an R-scheme that is quasi-compact and quasi-separated.

An invertible OX -module L is called ample if there is an integer m > 0 together
with nitely many global sections s0 , . . . , sn of Lm , the m-fold tensor product
of L, such that:
(1) s0 , . . . , sn generate Lm on X.
(2) Xsi is quasi-ane for i = 0, . . . , n.
2
Actually, we would have to consider U over some ane open part of Spec R, for instance
over a basic open part D(f ) = Spec Rf for some f R. However, since Rf = Rt/(1 tf )
for a variable t, the morphism Rf - A is of nite type if and only if the composition
R - Rf - A is of nite type.
Denition 9. Let X be an R-scheme. An invertible OX -module L is called

very ample if, for some n N, there is an immersion f : X - Pn satisfying
R

f (OP (1)) L.
See Remark 16 below for some simple examples of ample and very ample
invertible sheaves.
Theorem 10. Let X be an R-scheme that is quasi-compact and quasi-separated

and L an invertible OX -module. If L is very ample, then L is ample as well.
If, more specically, X is of nite type and quasi-separated over R, the
(i) L is ample.
(ii) There is an integer m > 0 such that Lm is very ample.
(iii) There is an integer m0 > 0 such that Lm is very ample for all m m0 .
Proof. To start with, assume that L is very ample and let f : X - Pn be

R

an immersion satisfying f (OP (1)) L. If we write PR = ProjR Rt0 , . . . , tn
n
as usual, then t0 , . . . , tn are global generators of OP (1) and their pull-backs

si = f # (ti ), i = 0, . . . , n are global generators of f (OP (1)) L. Moreover, we
can look at the immersions fi : Xsi - Pn , i = 0, . . . , n, induced from f ,
R,ti
and we claim that the open subschemes Xsi X are quasi-compact. Since X is
quasi-compact by assumption, there exists a nite ane open covering (Uj )jJ
of X such that each restriction L|Uj is trivial. Then, using an isomorphism
L|Uj - OUj , we see that Xsi Uj is a basic open subset in Uj and, hence, is
ane. Thus, as a nite union of ane open sets, Xsi is quasi-compact. Applying
Proposition 7, it follows that Xsi = f 1 (PnR,ti ) is quasi-ane, and we see that
L is ample. In the same way we can conclude that L is ample if a power Lm
for some exponent m > 0 is very ample.
To establish the remaining implications (i) = (ii) = (iii), we need a

crucial auxiliary construction, which we will explain independently of the spe-
cial situation considered in Theorem 10. Let f : X - S = Spec R be a
quasi-compact and quasi-separated morphism of schemes and L an invertible
OX -module. For m N we denote by Xm X the open subscheme consisting
of all points x Xwhere global sections from (X,Lm ) generate the module
Lm . Then Xm = l (X,Lm ) Xl and we set X = m>0 Xm . Furthermore, we
m
view the direct sum m0 (X, L ) as a (commutative) graded R-algebra,
taking the tensor product of global sections in the sheaves Lm as multipli-
cation. This multiplication is commutative, indeed, as is easily traced back to
the case where L is the trivial sheaf OX and where the tensor product of sec-
tions is given by the multiplication of OX as a sheaf of rings. In particular, the
homogeneous prime spectrum

P = ProjR (X, Lm )
m0

is dened as an R-scheme and, for any m > 0 and l (X, Lm ), we can
consider the ane open subscheme

Pl = Spec (X, Lm ) P,
(l)
m0
which is the basic open subscheme D+ (l) P given by l.
Lemma 11. There exists a canonical R-morphism : X - P , satisfying

1 m
(Pl ) = Xl for all l (X, L ), m > 0. It is constructed by gluing the
R-morphisms
l : X l - Pl , l (X, Lm ), m > 0,
induced from the R-algebra morphisms

- (Xl , OX ),
l : (Pl , OP )
#
/
r r/
- / ,
li li / Xl
where the latter are isomorphisms. Here r varies over (X, Lim ), and lri |Xl on
the right-hand side is to be interpreted as the quotient of r|Xl by li |Xl , i.e. it is
the section in OX (Xl ) that multiplied with the generator li of Lim on Xl = Xli
yields the section r|Xl (Xl , Lim ).
Proof. First, one checks that there are well-dened morphisms of R-algebras
m
l , as stated. Next, for any global sections l, l in powers L
# , Lm , there is a
canonical commutative diagram
#
(Pl , OP ) l - (Xl , OX )
res res
? # ?
ll
-
(Pll , OP ) (Xll , OX ) ,
and it is easily seen from this that the morphisms l can be glued to yield
a well-dened R-morphism : X - P . Furthermore, one concludes from
# l m l m
l ( lm ) = lm |Xl in conjunction with 9.1/7 that 1 l (Pll ) = Xl Xl = Xll and
then, by a covering argument that 1 (Pl ) = Xl .
It remains to show that the morphisms # l are isomorphisms. Starting with
the injectivity, consider an element r (X, Lim ) such that # l ( li ) = li |Xl = 0.
r r
Then r|Xl = 0 and by Remark 2 (i) there is an exponent j N such that rlj = 0.
Hence, we get
r rlj
i
= i+j = 0
l l
as an equation in (Pl , OP ).
To show the surjectivity of the # l , consider a section t (Xl , OX ). Using
Remark 2 (ii), there is an exponent i N such that tli |Xl (Xl , OX Lmi )
extends to a global section r (X, Lmi ). However, then lri |Xl = t, and we are
done.
As a corollary, we conclude from Lemma 11:
Lemma 12. Let X be an R-scheme that is quasi-compact and quasi-separated

and L an invertible OX -module. Then L is ample if and only if the morphism
: X - P considered above satises the following conditions:
(i) X = X .
(ii) is an open immersion.
Proof. First assume that L is ample. Then there exist global generators
s0 , . . . , sn (X, Lm ) of Lm for some integer m > 0 such that all Xsi are
quasi-ane. This implies Xm = X and, in particular, X = X. Moreover, we
conclude from Lemma 11 in conjunction with Proposition 7 that the morphisms
si : Xsi - Ps are open immersions. Using 1 (Ps ) = Xs , it follows that
i i i
is an open immersion so that conditions (i) and (ii) are satised.
Conversely, if X = X , as required in (i), we can use the quasi-compactness
of X to obtain global sections l1 , . . . , lr in certain powers Lm1 , . . . , Lmr of L
such that X = Xl1 . . . Xlr . Taking suitable powers of the li , we end up
with global sections in Lm1 ...mr , which generate this sheaf. Therefore Lm for
some exponent m > 0 admits a set of global generators s0 , . . . , sn . Moreover,
if : X - P is an open immersion, as required in (ii), all restrictions
1
si : (Psi ) = Xsi - Ps are open immersions. Thus, all Xs are quasi-
i i
ane by Proposition 7 and Lemma 11, provided we can show that Xsi is quasi-
compact. However, the latter is easy to achieve; the argument is the same as
the one used in the beginning of the proof of Theorem 10. Choose an ane
open subscheme U X where L is trivial and identify L|U with OX |U . Then
Xsi U coincides with the basic open subset D(si |U ) of U , showing that this
is an ane open subscheme in X. If we consider a nite ane open covering
of X trivializing L, we see that Xsi is a nite union of ane open subschemes
of X and therefore quasi-compact. Thus, conditions (i) and (ii) imply that L is
ample.
Now we come back to the proof of Theorem 10, turning to the implication
(i) = (ii). We assume that L is ample and, to simplify things a bit, we rst
m
consider the special case where m0 (X, L ), as a graded R-algebra, is
generated by certain global sections s0 , . . . , sn (X, L). Then these sections
must generate L as an OX -module as well. Indeed, otherwise the si would have
a common zero in X, and the same would be true for all monomials in the
si , viewing these as global sections in appropriate powers Lm of L. Since, by
our assumption, (X, Lm ) is generated as an R-module by monomials of total
degree m in the si , we see that Lm , for any exponent m > 0, could not have
a set of global generators. However, this contradicts our assumption of L being
ample. Therefore the sections s0 , . . . , sn must generate L.
Now look at the R-morphism : X - P of Lemma 11. Since L is

ample, we have X = X according to Lemma 12, and is an open immersion.
Composing with the closed immersion P - PnR given by the surjection of
graded R-algebras

Rt0 , . . . , tn - (X, Lm ), ti - si ,
mN
as in 9.1/21, we arrive at an immersion f : X - PnR . Apparently, f is dened

in the manner of Theorem 4 by the generators s0 , . . . , sn of L. Thus, we have
f (OP (1)) L and L is very ample.
Also in the general case we can basically
proceed like this, with some es-
sential modications though. We replace mN (X, Lm ) by a suitable subal-
gebra A that is generated over R by nitely many homogeneous elements of a
certain degree > 0. Then X has to be dened relative to A, and the resulting
morphism
X - Proj A = P
R
is to be studied. For this to work well, A has to be big enough such that we
obtain X = X and, in addition, have isomorphisms
(Pl , OP ) - (Xl , OX )
as in Lemma 11. To put all this into eect, we have to rely in a crucial way on
the fact that X is of nite type over R.
Explaining this in more detail, assume that L is ample and choose an expo-
nent m0 > 0 such that Lm0 is generated by global sections, say by the elements
of a nite subset L (X, Lm0 ). Then we have

X= Xl
lL
and we may assume that the open subschemes Xl X are quasi-ane for l L.
Thus, applying Proposition 7 in conjunction with Lemma 11, the canonical
morphism Xl - Spec OX (Xl ) Pl is an open immersion for every l L.
Moreover, we claim that all Xl may be assumed to be ane. Indeed, this is seen
by writing Xl as a nite union of basic open sets of type D(g) Spec OX (Xl ),
for suitable functions g OX (Xl ), and by trying to extend the elements g
to global sections of X. As shown in Remark 2 (ii), this is, as a rule, only
possible after multiplication by a certain power of l so that one obtains a global
extension lg (X, Lim0 ) of g li for some i > 0. If we set lg = llg , we see that
Xlg = Xl D(g) = D(g) is ane. Therefore, choosing i big enough such that
the powers li are sucient for treating the nitely many quasi-ane schemes
Xl , l L, as well as the nitely many ane open subschemes D(g) Xl needed
to cover the Xl , we can nd a nite subset L (X, Lm0 ) for m0 > 0 big
enough such that the elements l L generate Lm0 and such that Xl is ane
for each l L.
Now we use that X and, in particular, the open subschemes Xl X, l L,

are of nite type over R. We have already pointed out in 8.3/5 that an ane
R-scheme Spec A is of (locally) nite type if and only of if the corresponding
ring morphism R - A is of nite type. In our case, we know that Xl is
ane for all l L and, hence, that each ring OX (Xl ) is of nite type over R,
in other words, a nitely generated R-algebra. For l L let Tl OX (Xl ) be
a nite subset generating OX (Xl ) as an R-algebra. Choosing a suciently big
exponent i > 0, we can extend all products of type t li , where l L and t Tl ,
to global sections hl,t (X, Lim0 ); see Remark 2 (ii). Then we can replace L
by the subset {li ; l L} (X, Lim0 ), observing that Xli = Xl . Writing m0
in place of im0 , we obtain nite subsets L (X, Lm0 ) and Tl OX (Xl ) for
each l L such that:

(a) X = lL Xl where Xl is ane for all l L.
(b) OX (Xl ), for each l L, is generated as an R-algebra by the elements

| for global sections hl,t (X, Lm0 ).
h
t Tl , and we have t = l,t l Xl

Now consider the graded R-subalgebra in mN (X, Lm ) that is generated by
all elements l, hl,t (X, Lm0 ) where l varies in L and t in Tl , namely

A = Rl, hl,t ; l L, t Tl (X, Lm ).
mN
Let P = ProjR A and consider the ane open subschemes Pl = Spec A(l) for
l L. We try to compose the morphism from Lemmata 11 and 12 with
the morphism P V+ (A+ ) - P constructed
via 9.1/20 from the inclusion
- m
mN (X, L

morphism of graded R-algebras A ). Since we obtain
(X) lL Pl P V+ (A+ ) from (i), such a composition is possible, indeed,
and we thereby arrive at a morphism

: X - P V+ (A+ ) - P .
We want to show that is an open immersion, similarly as in Lemmata 11 and

12 above.
Lemma 13. The morphism : X - P is characterized by the R-algebra

morphisms
/
r r //
#
l : (Pl , OP ) - (Xl , OX ), - , l L,
li li /Xl
and the latter are isomorphisms. Thus, restricts for l L to an isomorphism

l : Xl = 1 - P and we can conclude that is an open immersion
(Pl ) l
with image lL Pl .
Proof. Since Xl is ane for all l L and since these schemes cover X, it is
enough to show that #
l is an isomorphism for all l L. The injectivity of
#
l is obtained as in the proof of Lemma 11. To verify the surjectivity x an
element in (Xl , OX ). It is a polynomial in the elements t Tl with coecients
in R. Therefore it is enough to show that each t Tl admits a preimage in
(Pl , OP ). However, this is clear from our construction, since
/
hl,t h /
- l,t / = t
l l /Xl
for l L and t Tl .
It is now easy to see that, for an ample invertible sheaf L in the situation
m0
of Theorem 10, a certain power L where m0 > 0 becomes very ample. The
graded R-subalgebra A mN (X, Lm ) is generated by certain homoge-
neous elements of degree m0 , namely the elements l, hl,t considered above, which
we will denote by a0 , . . . , aN in the following. Non-trivial homogeneous elements
can only exist in degrees divisible by m0 . Therefore we can modify the degree
by dividing out m0 , a process leaving P = ProjR A untouched. Then
Rt0 , . . . , tN - A = Ra0 , . . . , aN , ti - ai ,
is a surjection of graded R-algebras and the latter gives rise to a closed

immersion P - PN R ; see 9.1/21. Composition with the open immersion
: X - P yields an immersion f : X - PN R , which by its construction
coincides with the morphism X - PN that is determined in the manner of
R
Theorem 4 by the global generators l, hl,t (X, Lm0 ) where l L and t Tl .
Hence, f (OP (1)) Lm0 and we see that Lm0 is very ample. In particular,
the implication (i) = (ii) of Theorem 10 is clear.
To nish the proof of Theorem 10, it remains to show that for an invertible
sheaf L with a very ample power Lm , m > 0, we can always nd an exponent
m0 > 0 such that Lm is very ample for all m m0 . In a rst step, we study
the corresponding question for the existence of global generating systems.
Lemma 14. Let X be an R-scheme that is quasi-compact and quasi-separated

and L an ample invertible OX -module. If F is a quasi-coherent OX -module of
nite type, there exists an exponent m0 > 0 such that F Lm for all m m0
is generated by a nite number of global sections.
Proof. Choose some exponent m0 > 0 and a nite subset L (X, Lm0 )
such that Lm0 is generated by the global sections belonging to L. Then, as
shown above for the implication (i) = (ii) of Theorem 10, we may assume
that the open subschemes Xl X are ane for all l L. It follows that F|Xl
is associated to a nite OX (Xl )-module and, using Remark 2 (ii), it is possible
to extend a nite generating system of F|Xl to a system of global sections in
(X, F Lk0 m0 ) for suitable k0 > 0, just by multiplication with global sections
in Lm0 . It follows that F Lk0 m0 and, thus, also F Lkm0 for all k k0
are generated by nitely many global sections.
The same argument works for F L, . . . , F L(m0 1) as quasi-coherent

OX -modules in place of F. Choosing k0 universally for all these modules, the
assertion of the lemma follows for k0 m0 in place of m0 .
Finally we will need:
Lemma 15. Let X be an R-scheme and L a very ample invertible sheaf on X.

If L is a second invertible sheaf on X that is generated by nitely many global
sections, then L L is very ample.
Proof. We choose an immersion f : X - Pn satisfying f (OP (1)) L. Fur-

R
- n
thermore, let f : X PR be a morphism satisfying f (OP (1)) L . Then
we consider the morphism
X/R f f
(f, f ) : X - X R X - PnR R PnR
where the latter may be interpreted as the composition of the graph morphism
f : X - X R Pn with the morphism f id : X R Pn - Pn R Pn .
R R R R
Note that f is an immersion by 7.4/13 and, furthermore, that f id is an
immersion by 7.3/13, since f has this property. Therefore, using 7.3/13 again,

(f, f ) is an immersion itself. Writing p, p for the projections from PnR R PnR
onto its factors, we get

(f, f ) p OP (1) f OP (1) L,

(f, f ) p OP (1) f OP (1) L ,
and, hence,
(f, f ) p OP (1) p OP (1) L L .
Now we use the so-called Segre embedding

: PnR R PnR - Pnn +n+n
,
R
which is dened by the (n + 1)(n + 1) sections

ti tj (PnR R PnR , p OP (1) p OP (1)), i = 0, . . . , n, j = 0, . . . , n ,
where the ti , resp. tj , are (the pull-backs of) the canonical generators of the rst

Serre twists of the structure sheaves of PnR , resp. PnR . Then, if tij , i = 0, . . . , n,
+n+n

j = 0, . . . , n , are the canonical generators of the rst Serre twist of PnnR ,
we can describe locally by the R-algebra morphisms
't tnn ( 't tn ( ' t t (
R
00
,..., - R
0
R R 0 , . . . , n ,
,...,
tij tij ti ti tj tj

t00 tnn t t t t
- 0 0 , . . . , n n .
,...,
tij tij ti tj ti tj
In particular, we see that the Segre embedding is a closed immersion. Therefore

(f, f ) : X - Pnn
R
+n+n
is an immersion satisfying ( (f, f )) (OP (1)) L L and it follows that

L L is very ample.
Now it is easy to derive the implication (ii) = (iii) of Theorem 10 and

thereby to nish the proof of the theorem. Assume that Lm1 is very ample for
some exponent m1 > 0. Since L is necessarily ample, as we have seen already,
we can use Lemma 14 and thereby get an exponent m2 > 0 such that Lm is
generated by global sections for m m2 . Then Lemma 15 shows that Lm is
very ample for m m1 + m2 .
We end this section by discussing some easy examples of ample and very
ample invertible sheaves.
Remark 16. Let R be a non-trivial ring and X an R-scheme.

(i) If X is ane, every invertible OX -module L is ample. If, in addition, X
is of nite type over R, then every invertible OX -module L is even very ample.
(ii) Let X = PnR , n > 0. The dth Serre twist OX (d), for d Z, is ample if
and only if d > 0. Furthermore, OX (d) is even very ample for d > 0.
Proof. (i) If X is ane, say X = Spec A, and L is an invertible OX -module,

then L is associated to an A-module L. Choose elements f0 , . . . , fr A such
that X is covered by the basic open sets D(fi ) and L|D(fi ) is trivial for all
i, hence, isomorphic to OX |D(fi ) . Multiplying a generating element of L|D(fi )
with a suciently high power of fi , we obtain an element si L = (X, L)
generating L on D(fi ) and such that Xsi = D(fi ). Then s0 , . . . , sr is a set of
global generators of L such that Xsi is ane for all i. It follows that L is ample.
If X is an ane R-scheme of nite type, it is easy to see that the structure
sheaf OX is very ample. Indeed, choose a closed immersion X - AnR and
compose it with the open immersion AnR - PnR = ProjR Rt0 , . . . , tn onto
the basic open subset D+ (t0 ). Then OP (1)|D+ (t0 ) is trivial and, hence, its pull-
back to X is trivial. Thus, OX is very ample, and it follows from Lemma 15
that every invertible OX -module L is very ample, since L OX L.
(ii) Finally, let X = PnR = ProjR Rt0 , . . . , tn . Then OX (1) is very ample
for trivial reasons and we see from Lemma 15 that OX (d) is very ample for
d > 0. On the other hand, OX (d) cannot be ample for d < 0, since OX (d) does
not admit non-zero global sections in this case, as we have seen in Section 9.2.
But also OX (0) = OX is not ample, due to (X, OX ) = R. Indeed, every
R-morphism PnR - Pm dened via global sections s0 , . . . , sm (X, OX )
R
must factor through the structural morphism PnR - Spec R and therefore
will not be injective for n > 0. Hence, it cannot be an immersion.
Exercises
1. Alternative characterization of ample invertible sheaves: Let L be an invertible
sheaf on an
R-scheme Xm that is quasi-compact and quasi-separated. Furthermore,
let A = mN (X, L ) be the associated graded R-algebra. Show that the
(a) L is ample.
(b) The open subsets Xf X for f A+ homogeneous constitute a basis of the
topology on X.
(c) The ane open subschemes of X that are of type Xf for f A+ homogeneous
cover X.
This shows that the notion of ample invertible sheaves as given in Denition 8 is
compatible with the one used in EGA [12], II, 4.5.3.
2. Let X be a scheme over a ring R.
(a) Show that X is separated over R as soon as there exists a very ample invertible
sheaf on it.
(b) Assume that X is of nite type and quasi-separated over R. Conclude that X
is separated over R as soon as there exists an ample invertible sheaf on it.
3. Let Pm - Pn be a non-constant morphism of projective spaces over R. Show
R R
that m n.
4. Consider invertible sheaves L, L on an R-scheme X that is quasi-compact and
quasi-separated. Show:
(a) If L is ample, there exists an exponent m > 0 such that Lm L is ample. In
fact, we may take m = 1 if L is generated by global sections.
(b) If L and L are ample, the same is true for L L .
5. Consider the projective n-space PnK = ProjK Kt0 , . . . , tn over a eld K. Show
that every K-automorphism : PnK - Pn is linear in the sense that it is
K
induced from a K-automorphism of the K-vector space OP (1) = Kt0 . . . Ktn .
6. d-uple embedding: Consider the projective m-space Pm R = ProjR R t0 , . . . , tm
overR and
its invertible sheaf L = OP (d) for some d N. Let M0 , . . . , Mn be
the m+d
m monomials in t0 , . . . , tm of degree d. Show that the Mi give rise to
an R-morphism : Pm - Pn , the so-called d-uple or Veronese embedding.
R R
Describe in terms of coordinates and show that it is a closed immersion for
d > 0. Conclude once more that OP (d) is very ample if d > 0.
7. Let A be a graded ring (of type N), whose irrelevant ideal A+ is generated by
nitely many elements in A1 . For a quasi-coherent OX -module
F on X = Proj A
consider the associated graded A-module (F) = nZ (X, F(n) where

F(n) = F OX (n); see Exercise 9.2/8. Show that there is a canonical morphism
of OX -modules
(F)
- F and that the latter is an isomorphism.
8. Consider the projective n-space PnR = ProjR Rt0 , . . . , tn over some ring R. Show:
(a) Given a graded ideal I Rt0 , . . . , tn , the residue class ring Rt0 , . . . , tn /I
is canonically a graded R-algebra and the projection map
Rt0 , . . . , tn - Rt0 , . . . , tn /I

gives rise to a closed immersion of R-schemes
ProjR Rt0 , . . . , tn /I - Proj Rt0 , . . . , tn .

R
(b) Every closed immersion Z - Proj Rt0 , . . . , tn of R-schemes is of the

R
type as described in (a). Hint: Use Exercise 7.
9. Consider the the Segre embedding : P1R R P1R - P3 for projective spaces
R
over a ring R. Specify a graded ideal I Rt0 , t1 , t2 , t3 such that the closed
subscheme Z P3R dened by is obtained in the manner of Exercise 8. The
subscheme Z is called a quadric surface in P3R .
Let A, B be graded algebras (of type N) over a ring R. Then the direct sum
10.
nN An R Bn is canonically a graded R-algebra, called the cartesian product
of A and B; it is denoted by A R B. Assuming that the irrelevant ideals A+ A
and B+ B are generated by homogeneous elements of degree 1, show that there
is a canonical isomorphism ProjR (A R B) - (ProjR A) R (ProjR B). For
example, consider the case of free polynomial rings in nitely many variables over
R, say A = Rt0 , . . . , tm and B = Rt0 , . . . , tn . Then the elements ti tj where
i = 0, . . . , m, j = 0, . . . , n dene global generators of the rst Serre twist O(1)
for ProjR (A R B), and the resulting morphism
ProjR Rt0 , . . . , tm R ProjR Rt0 , . . . , tn - Pmn+m+n

R
is just the Segre embedding. Hint: Use localization arguments similar to the ones
employed in the proof of 9.2/2.
9.5 Proper Morphisms
The notation Proj A for the Proj scheme of a graded ring A might suggest the
term projective for such schemes, alluding to the denition of the projective
n-space PnR = ProjR t0 , . . . , tn over some base ring R. However, in reality,
an R-scheme X is called projective only if it is of type ProjR A for a graded
R-algebra A of nite type, generated by nitely many homogeneous elements
of degree 1. The latter is equivalent to the fact that ProjR A admits a closed
immersion into some projective n-space PnR ; see Proposition 7 below. As a gen-
eralization of projective schemes, we will study proper schemes and morphisms
in the present section.
Denition 1. Let f : X - Y be a morphism of schemes.

(i) f is called ane (resp. quasi-ane) if there exists an ane open covering
(Vi )iI of Y such that f 1 (Vi ) is ane (resp. quasi-ane) for all i I.
(ii) f is called nite if there exists an ane open covering (Vi )iI of Y such
that f 1 (Vi ) is ane and the morphism f # (Vi ) : OY (Vi ) - OX (f 1 (Vi )) is
nite for all i I in the sense that OX (f 1 (Vi )) is a nite OY (Vi )-module.
It is clear that ane morphisms are quasi-ane and that nite morphisms
are ane. For example, every closed immersion is nite and, in particular, ane.
9.5 Proper Morphisms 463
Remark 2. Ane and quasi-ane morphisms are quasi-compact and separated.
Proof. Ane and quasi-ane morphisms are quasi-compact by 6.9/8, since

ane schemes are quasi-compact by 6.1/10 and quasi-ane schemes are quasi-
compact by denition; see 9.4/7. Furthermore, ane morphisms are separated
by 7.4/8. It remains to consider a quasi-ane morphism of schemes f : X - Y
and to show that it is separated. Using 7.4/7 we may assume that Y is
ane and X is quasi-ane. Then f decomposes into the canonical morphism
X - Spec OX (X), which is an open immersion, and the ane morphism
Spec OX (X) - Spec OY (Y ) = Y derived from the f . Both are separated,
the rst one by 7.4/10, and, hence, their composition is separated by 7.4/15.

Proposition 3. Let f : X - Y be a morphism of schemes, where Y is ane.

Then:
(i) f is ane if and only if X is ane.
(ii) f is quasi-ane if and only if X is quasi-ane.
(iii) f is nite if and only if X is ane and the corresponding morphism
OY (Y ) - OX (X) is nite.
Proof. The if parts of the dierent assertions are trivial. To verify the only-if
part of (i), let f be ane. Then we see from 6.9/4 or 6.9/9 that f (OX ) is a
quasi-coherent OY -module and, hence, by 7.1/4, a quasi-coherent OY -algebra.
Recalling the construction of the spectrum of a quasi-coherent algebra in Sec-
tion 7.1, it follows that X as a Y -scheme is isomorphic to Spec f (OX ). Since
f (OX ) is associated to OX (X) as an OY (Y )-algebra by 7.1/6, X is ane.
In the situation of (ii) write A = OX (X) and B = OY (Y ) and look at the
canonical decomposition
f: X - Spec A - Spec B = Y.
For any g B we can consider the basic open subset D(g) = Spec Bg Spec B,
its preimage Spec A B Bg Spec A, and its preimage in X, which we denote
by Xg so that f , restricted to preimages over Spec Bg , decomposes into
fg : X g - Spec A B Bg - Spec Bg .
We claim that the ring morphism of global sections A B Bg - OX (Xg )

associated to Xg - Spec A B Bg is an isomorphism. Indeed, choose an
ane open covering (Ui )iI of X, where we may assume that I is nite, since
f and, hence, X are quasi-compact. Furthermore, we know from 7.4/5 that all
intersections Ui Uj are ane for i, j I, since f is separated by Remark 2.
Then look at the exact diagram

A = OX (X) - OX (Ui ) - - OX (Ui Uj )
iI i,jI
and observe that it remains exact when tensoring it with Bg over B, since I is
nite. This shows that the morphism A B Bg - OX (Xg ) is an isomorphism,
as claimed.
Now, in order to show that X is quasi-ane, we use freely the characterizing
properties of quasi-ane schemes, as mentioned in 9.4/7. It remains to show
that X - Spec A is an open immersion. For this it is enough to see that
Xg - Spec A B Bg is an open immersion for a set of elements g such that
the D(g) cover Y = Spec B. However, that the latter is true follows from the
fact that f is quasi-ane. Indeed, given a point y Spec B, there is an ane
open neighborhood V Spec B such that f 1 (V ) is quasi-ane. Choosing a
basic open subset D(g) V containing y, the preimage Xg = f 1 (D(g)) is still
quasi-ane. So the morphism Xg - Spec AB Bg will be an open immersion,
since it induces an isomorphism on the level of global sections. Therefore we are
done.
Finally, concerning (iii), if f is nite, X is ane by (i). Furthermore, f (OX )
is a quasi-coherent OY -module that is locally of nite type in the sense of 6.8/12.
But then OX (X) = f (OX )(Y ) is a nite module over OY (Y ) by 6.8/13.
Let us add without proof that the properties of a morphism of schemes to

be ane, quasi-ane, or nite are stable under composition of morphisms, base
change, and ber products.
Denition 4. Let f : X - Y be a morphism of schemes.

(i) f is called closed if the image f (Z) of every closed subset Z X is
closed in Y .
(ii) f is called universally closed if for every base change Y - Y the
resulting morphism f id : X Y Y -
Y is closed.
(iii) f as well as X as a Y -scheme are called proper if f is separated, of
nite type, and universally closed.
It is immediately clear that for a scheme morphism f : X - Y the above

conditions (i), (ii), and (iii) can be tested locally over Y . Furthermore, using
7.4/15 it is easily seen that the properties for a morphism to be universally
closed or proper are stable under composition of morphisms, base change, and
ber products.
Remark 5. Every nite morphism of schemes is proper.
Proof. Let f : X - Y be a nite morphism of schemes. Since the properness

of f can be tested locally on Y , we may assume that X and Y are ane,
say X = Spec A and Y = Spec B. Then, of course, f is of nite type and
separated. Furthermore, since nite morphisms are preserved under base change,
it is enough to show that f is closed.
To check the latter, observe that the ring morphism f # : B - A attached
to f is nite and, hence, integral. Now let Z X be a closed subset, say
Z = V (a) for some ideal a A, and consider the ideal b = (f # )1 (a) in B.

Then we get f (Z) V (b) and we claim that f (Z) = V (b). Indeed, look at the
commutative diagram
f#
B - A
? ?
f#
B/b - A/a ,
induced from f # , where f # , just as f # , is nite and, hence, integral. Therefore
we conclude from the Lying-over Theorem 3.3/2 that for every prime ideal
q B/b there is a prime ideal p A/a such that (f # )1 (p) = q. Consequently,
there exists for every y V (b) a preimage x Z. This shows f (Z) = V (b) so
that f is closed.
f g
Proposition 6. Let X -Y - Z be morphisms of schemes such that the
composition g f is proper. Then:
(i) If g is separated, f is proper.
(ii) If g is separated and of nite type and if f is surjective, then g is proper.
Proof. Starting with assertion (i), we see from 7.4/15 (iv) that f is separated,
since g f is separated. Furthermore, write f as a composition
f f
f : X = X Y Y - X Z Y - Z Z Y = Y
where f is the canonical morphism and f is the morphism obtained from

g f : X - Z via base change with g : Y - Z. Since g is separated, we see
from 7.4/12 that f is a closed immersion. Moreover, f is closed, since g f is

universally closed. Therefore f is closed and the same argument in conjunction

with the fact that separated and proper morphisms are stable under base change
shows that f is, in fact, universally closed.
In addition, these considerations show that f is quasi-compact. Indeed,
being a closed immersion, f is quasi-compact and f is obtained from the
quasi-compact morphism g f : X - Z via base change. Thereby it only
remains to check that f is locally of nite type, which, however, is clear, since
a morphism of rings B - A is of nite type as soon as there is a morphism
of rings C - B such that the composition C - B - A is of nite type.
In the situation of (ii) it is only to show that g is universally closed. To do
this, look at a closed subset F Y . Then we get g(F ) = (g f )(f 1 (F )) from
the surjectivity of f and it follows that, if f 1 (F ) is closed in X, its image under
g f is closed in Z. In particular, g is closed. Now observe that the assumptions
in (ii) are preserved under base change. Indeed, concerning the surjectivity of
f we may look at bers over some point y Y and thereby assume that Y
consists of a eld, say Y = Spec K. Furthermore, it is enough to consider a base
change on Y that is given by an extension of elds K /K. But then, due to the
fact that the extension K /K is faithfully at, f K K will be surjective if f
is. Thus, the assumptions in (ii) are preserved under base change and we can
show as before that g is universally closed.
Next we want to introduce projective morphisms and to discuss their rela-

tionship to proper morphisms.
Proposition and Denition 7. For a scheme X over some base ring R the
(i) There exists an R-morphism X - Pn that is a closed immersion.
R
an R-isomorphism X - ProjR A where A is a graded
(ii) There exists
R-algebra, say A = i=0 Ai , that is generated by nitely many elements in A1 .
If the above conditions are met, the R-scheme X and its structural morphism
X - Spec R are called projective.
Proof. Assume rst that condition (ii) is given and let f0 , . . . , f1 A1 generate
A as an R-algebra. Then
Rt0 , . . . , tn - A, ti - fi ,
is a surjection of graded R-algebras inducing a closed immersion X - Pn ;

R
see 9.1/21.
Conversely, if there exists a closed immersion X - PnR over R, then X may
be viewed as a closed subscheme of PnR and, thus, by 7.3/4, is given by a quasi-
coherent ideal I OPnR . We will prove below in Lemma 8 that I is associated to
a graded ideal I Rt0 , . . . , tn . Then the construction of associated modules
in 9.2/1 in conjunction with 7.3/4 shows that we get X ProjR Rt0 , . . . , tn /I,
where A = Rt0 , . . . , tn /I is a graded R-algebra as required in (ii).

Lemma 8. Let R be a ring, A = i=0 Ai a graded R-algebra, and I a quasi-
coherent ideal of the structure sheaf OX where X = ProjR A.
Assume that A is generated by nitely many elements f0 , . . . , fn A1 . Then
there exists a graded ideal I A, whose associated ideal I OX , as dened in
9.2/1, coincides with I.
Proof. Let I|D+ (fi ) be associated to the ideal ai A(fi ) , and let I A be the
ideal generated by all homogeneous elements g A such that g Ar implies
g
ai for all i = 0, . . . , n.
fir
We claim that I is as required. First observe that the ideal I(fi ) A(fi ) induced
from I is contained in ai for all i. On the other hand, if h belongs to aj for
some index j, it is of type h = fgr for some g Ar , and we would like to get
j
g
fir
ai for all i = 0, . . . , n. Since the ideals ai and aj , for any i, j, restrict to the
same ideal on D+ (fi ) D+ (fj ), we see that fgr ai holds after restricting both
i
quantities to D+ (fi ) D+ (fj ). Since

OX D+ (fi ) D+ (fj ) = A(fi fj ) = (A(fi ) ) fj
fi
by 9.1/7, it follows that sections on D+ (fi ) D+ (fj ) can be extended to D+ (fi )

f
by multiplying with a certain power of fji . Thus, enlarging r, we may assume
g
fir
ai for all i = 0, . . . , n. From this we conclude g I and, hence, I(fi ) = ai
for all i.
Theorem 9. Every projective scheme X over a ring R is proper.
Proof. Using Proposition 7, we know X ProjR A for a graded R-algebra

A = i=0 Ai that is generated by nitely many elements f1 , . . . , fn A1 .
In particular, ProjR A is separated over R by 9.1/18 and quasi-compact, as it
is covered by the ane open subschemes Spec A(fi ) , i = 0, . . . , n. Since each
homogeneous localization A(fi ) , viewed as an R-algebra, is generated by the
fractions ff0i , . . . , ffni , it follows that ProjR A is locally of nite type over R and,
thus, all in all, of nite type over R.
It remains to show that the structural morphism p : ProjR A - Spec R
is universally closed. Since projective morphisms are stable under ane base
change (note that we have considered projective n-spaces over an ane base
only) and since the closedness of a subset of a scheme Y can be checked relative
to an open covering of Y , it is enough to show that p is closed. Furthermore,
if Z ProjR A is a closed subset, we may apply 7.3/5 and view Z as a closed
subscheme of ProjR A. Then Z is projective over R again and we thereby see
that it is enough to show that the image of p : ProjR A - Spec R is closed
in Spec R.
To check the latter, consider a point y Spec R with residue eld k(y).
Then y im p is equivalent to

p1 (y) = (ProjR A) R k(y) = Projk(y) A R k(y) =
and we claim that this is equivalent to the existence of an index i0 N satisfying
Ai R k(y) = 0 for all i i0 .
Indeed, from 9.1/15 (i) we conclude that Projk(y) (A R k(y)) = if and only
if rad+ (0) = A+ R k(y) for the zero ideal 0 A R k(y) and, hence, if and
only if all elements of A+ R k(y) are nilpotent. Since A+ is a nitely generated
ideal in A, the latter amounts to the existence of an index i0 N such that
Ai R k(y) = 0 for i i0 .
Now use the fact that A is generated by nitely many elements in A1
and, hence, that each Ai is a nitely generated R-module. Then Nakayamas
Lemma 1.4/10 implies that Ai R k(y) = 0 is equivalent to Ai R Ry = 0 and,
thus, to y suppR Ai where
suppR Ai = {x Spec R ; Ai R Rx = 0}
is the support of Ai as an R-module. Since A+ is generated by A1 as an

R-algebra, we have
suppR A1 suppR A2 . . .
and thereby obtain

y im p y suppR Ai .
i=1
Since the support of an R-module of nite type is closed in Spec R and the
intersection of closed subsets is closed again, we deduce that im p is closed in
Spec R.
For completeness, let us show that the support suppR M of an R-module of
nite type is closed in Spec R. Namely, we claim for the so-called annihilator
ideal
a = {a R ; aM = 0} R
of M that suppR M = V (a). Indeed, by denition we have x suppR M if and
only if Mx = M R Rx = 0. Fixing generators m1 , . . . , mr M , we see that
Mx = 0 if and only if, for every i = 1, . . . , r there is some si R px such that
si mi = 0. Thus, replacing si by the product s = s1 . . . sr , it follows that Mx = 0
if and only if there exists some s R px such that sM = 0 and, hence, s a.
Therefore x suppR M is equivalent to a px and, thus, to x V (a).
Next we want to give some outline on how to dene projective schemes

over base schemes S that are not necessarily ane; for further details see EGA
[12], in particular II, 4 and 5.5. Up to now we have dealt with the property of
a morphism of schemes X - S to be projective only in the case where
S is ane. To remove this special assumption, the notion of Proj schemes is
generalized from Proj A for a graded ring A = m=0 Am to ProjS A for a quasi-
coherent graded OS -algebra A = m=0 A m over arbitrary base schemes S; see
also Exercise 9.1/11. This is done similarly as for the construction of the scheme
SpecS A, the spectrum of a quasi-coherent OS -algebra A, by choosing an ane
open covering (Si )iI of S and gluing the Proj schemes

ProjSi A(Si ) = ProjSi Am (Si ), i I,
m=0
in a canonical way. Applying this concept to the construction of projective

spaces, we need to dene for a module M over some ring R its associated
symmetric R-algebra SymR (M ). To do this, look at the tensor algebra generated
by M over R,

TR (M ) = M m ,
m=0
where we write M m for the m-fold tensor product M R . . . R M . The mul-

tiplication on TR (M ) is given by the tensor product where the latter is not
commutative, except for some special cases. However, we enforce commuta-
tivity by dening SymR (M ) as the quotient of TR (M ) by the two-sided ideal
generated by all elements of type x y y x where x, y M . Then SymR (M )

is a graded R-Algebra, now with commutative multiplication. For m N we
call the R-module Symm R (M ) of all elements in SymR (M ) that are homogeneous
of degree m the mth symmetric power of M over R. For example, SymR (Rn ) is
just the polynomial ring Rt1 , . . . , tn in n variables t1 , . . . , tn over R.
Since the construction of SymR (M ) is compatible with localization by el-
ements in R, symmetric algebras can more generally be considered for quasi-
coherent modules on a base scheme S. This way one obtains from any quasi-
coherent OS -module E the corresponding symmetric algebra SymOS (E), which
is a quasi-coherent graded OS -algebra, and from the latter the attached Proj
scheme PS (E) = ProjS (SymOS (E)) over S. For example, taking E = OSn+1 we
see that PS (E) is just the projective n-space PnS . The construction of PS (E)
is of particular interest for locally free OS -modules E. Then we talk about a
projective space bundle on S. It is easily seen that the denition of projective
morphisms via Proposition 7 can be generalized if we
(i) replace the projective n-space PnR by the scheme PS (E) attached to a
quasi-coherent OS -module of locally nite type E and
(ii) pass to graded quasi-coherent OS -algebras A that, locally on S, are
generated by nitely many sections in A1 .
Every Proj scheme PS (E) constructed from a quasi-coherent OS -module E
carries an invertible sheaf OP (1), the rst Serre twist, which is dened similarly
as in Section 9.4. In particular, the denition of very ample invertible sheaves
from 9.4/9 can be carried over to the case of schemes over a not necessarily ane
base. If p : X - S is a morphism of schemes equipping X with the structure of
an S-scheme, an invertible sheaf L on X is called very ample relative to S if there
is a quasi-coherent OS -module E together with an S-morphism f : X - PS (E)
that is an S-immersion satisfying f (OP (1)) L. For an ane base S and a
morphism of nite type p : X - S one can show that this denition coincides
with the one given in 9.4/9; see EGA [12], II, 4.4.7.
Also the property of being ample can be dened relative to a not necessarily
ane base S. Namely, an invertible sheaf L on an S-scheme X, given by a
morphism p : X - S that is quasi-compact and quasi-separated, is called
ample relative to S if there exists an ane open covering (Si )iI of S such
that L|Si is ample on p1 (Si ) for all i. Also this denition is for an ane base
S equivalent to the one given in 9.4/8 (see EGA [12], II, 4.6.6), and one can
show for a quasi-compact base S and a quasi-separated morphism of nite type
p : X - S that the correspondence between ample and very ample invertible
sheaves from 9.4/10 generalizes to the relative case; see EGA [12], II, 4.6.11.
We end this survey by mentioning without proof some important results on
proper morphisms.
Theorem 10 (Chows Lemma, EGA [12], II, 5.6.1, 5.6.2). Let f : X -Y

be a proper morphism of schemes where Y is Noetherian. Then there exists a
morphism g : X - X such that:
(i) f g : X - Y is projective,
(ii) g : X - X is surjective and there is a dense open subscheme U X

such that g 1 (U ) is dense in X and g induces an isomorphism g 1 (U ) - U .
Theorem 11 (Proper Mapping Theorem, EGA [13], III, 3.2.1). Consider a

proper morphism of schemes f : X - Y where Y is locally Noetherian. Then
for any coherent OX -module F (see the end of 6.8), the higher direct images
Rq f (F), q 0, as dened in 7.7, are coherent as well.
Theorem 12 (Stein Factorization, EGA [13], III, 4.3.1). Let f : X - Y be

a proper morphism of schemes where Y is locally Noetherian. Then there exists
a factorization
f
X - Y
-
g h
-
Y
where g is proper, surjective, and has connected bers, and where h is nite.
Moreover, f OX is a quasi-coherent OY -algebra such that Y = Spec f OX .
Finally we want to show that the properness (and separatedness) of scheme

morphisms can be characterized using techniques of valuation rings. In other
words we want to prove the so-called valuative criteria for separated and proper
morphisms. Note that discrete valuation rings have already been dened in
9.3/3. However, we will need a more general version of valuation rings here.
Denition 13. Let R be an integral domain with eld of fractions K. Then R

is called a valuation ring if for x K we have x R or, otherwise, x1 R.
Since discrete valuation rings are unique factorization domains, it is easily

seen that discrete valuation rings are valuation rings, indeed. Every valuation
ring R is a local ring, because any two elements x, y R will satisfy (x) (y) or
(y) (x). As usual, the unique maximal ideal m R consists of all non-units in
R, hence, of 0 and of all elements x R {0} such that x1 R. In particular,
the spectrum Spec R contains a unique generic point , which corresponds to
the zero ideal 0 R, and a unique closed point s, which corresponds to the
maximal ideal m R. Furthermore, the inclusion R K into the eld of
fractions of R yields a canonical morphism i : Spec K - Spec R. We will
use it to dene for morphisms of schemes X - Y and Spec R - Y the
restriction map
: HomY (Spec R, X) - HomY (Spec K, X)
given by composition with i : Spec K - Spec R. For h HomY (Spec R, X)

its associated image h = (h) is characterized by the commutative diagram
h
Spec K -X
-
h
i f
? ?
Spec R - Y .
Proposition 14. Let f : X - Y and Spec R - Y be morphisms of schemes

where R is a valuation ring with eld of fractions K. Furthermore, let
be the composition with the canonical morphism i : Spec K - Spec R. Then

(i) If f is separated, is injective.
(ii) If f is proper, is bijective.
Proof. As usual, let be the generic point and s the closed point of Spec R.
Starting with assertion (i), let h, h : Spec R - X be two Y -morphisms with
same image h = (h) = (h ) in HomY (Spec K, X) and consider the morphism
(h, h ) : Spec R - X Y X
given by h and h . By our assumption, its restriction to Spec K factors via h

through the diagonal morphism X/Y : X - X Y X. Therefore (h, h ) maps
into the diagonal = X/Y (X). Now if f is separated, is closed in X Y X
and we can conclude from the continuity of (h, h ) that the image of s belongs
to so that h(s) = h (s). Choosing an ane open neighborhood Spec A X
of h(s) = h (s), it follows that both, h and h , map Spec R into Spec A, thereby
giving rise to the commutative diagram
h#
K A
6
i# h#
h#

R .
The latter shows that h coincides with h , since i# is injective.
To verify assertion (ii), consider a morphism h HomY (Spec K, X). Writing
X = X Y Spec R, we get a commutative diagram
h : Spec K - X - X
f f
i
-
? ?
Spec R - Y
and it is enough to show that f : X - Spec R admits a section extending the
morphism Spec K - X . To obtain such a section, let Z X be the closure
of (the image of) Spec K in X and view Z as a subscheme of X , equipping

it with its reduced structure; see 7.3/5. Then Z is irreducible by 7.5/2 and,
in fact, integral. Indeed, all its non-empty ane open parts are irreducible by
7.5/1 and, thus, by the reducedness in conjunction with 6.1/15 and 9.3/2, are of
type Spec A for an integral domain A. Furthermore, f is assumed to be proper
and, hence, universally closed. Therefore f is closed and it follows that f (Z) is
closed in Spec R. But then we must have f (Z) = Spec R, since f (Z) contains
the generic point Spec R. Now if is the generic point of Z and s Z a
point over s Spec R, there is a canonical commutative diagram
OZ,
OZ,s
6 6
fs#

K
R
of injective homomorphisms, which we view as inclusions. Furthermore, fs# is

local and we claim that, actually, fs#
is an isomorphism. Indeed, otherwise there
would be an element a OZ,s that is not contained in R. By the valuation ring

property of R the inverse a1 would belong to the maximal ideal m R and a
would be a unit in OZ,s . However, this is impossible if fs#
is local. Consequently,
fs#
is an isomorphism. Using this fact, we see for any ane open neighborhood
Spec A Z of s that the localization map A - OZ,s denes a section of
Z - Spec R and it follows that X - Spec R admits a section as well.
The valuative criteria for separatedness and properness say that, under cer-
tain niteness conditions, separated and proper morphisms can be characterized
by the conditions given in Proposition 14 (i) and (ii). To prepare the proof of
this fact, we start by some auxiliary results.
Lemma 15. Let f : X - Y be a quasi-compact morphism of schemes. Then

(i) f (X) is closed in Y .
(ii) f (X) is stable under specialization, i.e. given any point f (X) and a
specialization y {}, then y f (X).
Proof. Clearly, if f (X) is closed in Y , it is stable under specialization. Conversely,

assume that f (X) is stable under specialization. Since the closedness of f (X)
can be tested locally on Y , we may assume Y to be ane, say Y = Spec B.
Furthermore, we may equip X with its reduced structure as constructed in 7.3/5
and then assume that Y is reduced and satises f (X) = Y . Indeed, consider
the ring morphism f # : B - OX (X) corresponding to f and replace B by
B/ ker f # .
Since X is quasi-compact by assumption, it admits a nite ane open cov-
ering (Xi )iI . Writing Yi = f (Xi ) we obtain

Yi = f (Xi ) = f (Xi ) = f (X) = Y,
iI iI iI
and we may view the closed subsets Yi as closed subschemes of Y , namely of

type Yi = Spec Bi where Bi = B/I(Yi ).
Now, in order to show f (X) = Y and, thus, the closedness of f (X), consider
a point y Y , say y Yi for some index i I. Then there exists a generic
point Yi such that y {}, and it is enough to show f (Xi ), since then
f (X) implies y f (X) by our assumption. Therefore consider the ring
morphism Bi - Ai corresponding to Xi - Yi and localize it at . Since
Bi, is at over Bi we arrive at a monomorphism Bi, - Ai Bi Bi, . Moreover,
Bi is reduced and the same is true for Bi, . Thus, it follows that Bi, is a eld,
as it is a reduced ring with a unique prime ideal, namely the one corresponding
to . In particular, Ai Bi Bi, cannot be zero and, thus, there exists a prime
ideal p Ai Bi Bi, . Looking at the commutative diagram
Bi - Ai
? ?
Bi, - Ai Bi Bi, ,
we get p Bi, = 0. But then the preimage of p in Ai is a prime ideal p

corresponding to a point Xi that satises f ( ) = . In particular, it
follows f (Xi ), as claimed.
We need an auxiliary result showing how valuation rings come into play
when dealing with local rings.
Lemma 16. Let K be a eld and R K a local ring with maximal ideal m R.
Then there exists a valuation ring R with eld of fractions K dominating R,
i.e. such that R R and m R = m for the maximal ideal m of R .
Proof. Let k = R/m be the residue eld of R. Furthermore, choose an algebraic

closure k of k and consider the canonical map : R - R/m - k. Let

be the set of all pairs (R , ) where R is a subring of K containing R and
: R - k a homomorphism extending . Then is a non-empty partially
ordered set, writing (R , ) (R , ) whenever R R and |R = . By
Zorns Lemma, there is a maximal element (R , ) and we claim that R
is a valuation ring with eld of fractions K dominating R.
It is easy to see that R is a local ring with maximal ideal m = ker .
Indeed, k is an integral domain and, hence, m = ker a prime ideal. Then,
extending : R - k to the localization R , we get R = R from the
m m
maximality of (R , ). In particular, we have m R = ker = m and it follows

that R is a local ring dominating R.

Thus, it remains to show that R is a valuation ring with eld of fractions
K. To achieve this, consider an element x K and assume rst that x is
integral over R . Then the inclusion R - R x is nite by 3.1/4, and

there is a prime ideal p R x such that p R = m ; see the Lying-over
Theorem 3.3/2. The resulting inclusion R /m - R x/p is nite as well
and even a nite extension of elds, since R /m is a eld; see 3.1/2. Therefore
the embedding R /m - k induced from : R - k admits an extension
-

R x/p k, and the resulting homomorphism R x - R x/p - k

extends : R - k. But then the maximality of (R , ) yields R
x = R

and, thus, x R .
It remains to show for an arbitrary element x K that x R or, otherwise,
x R . Proceeding indirectly, we may assume by the above consideration that
1
neither x nor x1 are integral over R . In particular, x = 0 and we must have

x1 R x. Otherwise there would be an equation of type
x1 = a0 + a1 x + . . . + an xn
with coecients ai R and this would lead to an integral equation for x1 over
R if we multiply with xn . Therefore x cannot be a unit in R x and there
exists a maximal ideal m R x containing x. Now look at the composition
: R - R x - R x/m .
Since x m , we see that is surjective, mapping R onto the eld R x/m .

Therefore the kernel of is a maximal ideal in R and, hence, must coincide with
m , because R is a local ring with maximal ideal m . But then induces an iso-
morphism of elds : R /m - R x/m and it follows that : R - k
admits an extension to R x. However, since (R , ) is maximal in , we get
R x = R and therefore x R , contradicting our choice of x. Thus, R is a
valuation ring with eld of fractions K.
Proposition 17 (Valuative criterion of separatedness). Let f : X - Y be a

quasi-separated morphism of schemes. Then the following conditions are equiv-
alent:
(i) f is separated.
(ii) For every morphism of schemes Spec R - Y where R is a valuation
ring with eld of fractions K, the map
induced from i : Spec K - Spec R is injective.
Proof. According to Proposition 14 we have only to show that condition (ii)

implies the separatedness of f . So assume that (ii) is given. Since f is quasi-
separated, the diagonal embedding X/Y : X - X Y X is quasi-compact.
Hence, using Lemma 15, it is enough to show that the diagonal = X/Y (X)
in X Y X is stable under specialization. To achieve this, x a point x
specializing into some point x0 X Y X so that x0 Z = {x}. Applying
7.3/5, we consider Z as a subscheme in X Y X, providing it with its reduced
structure. Then, similarly as shown in the proof of Proposition 14, the scheme Z
is integral. Thus, the local ring OZ,x0 is a subring of the eld K = OZ,x = k(x)
and there is a valuation ring R with eld of fractions K dominating OZ,x0 ; see
Lemma 16. Let be the generic and s the closed point of Spec R. Then the
inclusion map OZ,x0 - R together with the localization map A - OZ,x
0
corresponding to any ane open neighborhood Spec A Z of x0 induce a
Y -morphism
: Spec R - Z - X Y X,
- x, s - x0 ,
such that the restriction of to Spec K factors through the diagonal embedding
- X Y X. Composing with the projections X Y X -
X/Y : X -X

we get two morphisms h, h : Spec R - X restricting to one and the same
morphism h : Spec K - X and, furthermore, making the diagram
h
Spec K - X
-
-
i h f
h
? ?
Spec R - Y
commutative. Then we conclude h = h from (ii) and see that the above mor-
phism , which can also be written in the form
(h, h ) : Spec R - Z - X Y X,
factors through the diagonal embedding X/Y . In particular, x0 = (h, h )(s)

and is stable under specialization.
Proposition 18 (Valuative criterion of properness). Let f : X - Y be a

quasi-separated scheme morphism of nite type. Then the following conditions
are equivalent:
(i) f is proper.
(ii) For every morphism of schemes Spec R - Y where R is a valuation
ring with eld of fractions K, the map
induced from i : Spec K - Spec R is bijective.
Proof. According to Proposition 14 we have only to show that condition (ii)

implies the properness of f . So assume that (ii) is given. Then f is separated
by Proposition 17 and it remains to show that f is universally closed. Since
our assumptions on f , as well as condition (ii), are stable under base change
Y - Y , it is enough to show that f is closed. So let V X be a closed
subset. Then we may view V as a closed subscheme of X, equipping it with its
reduced structure as obtained in 7.3/5. The composition V - X - Y is

quasi-separated and of nite type again, and a reasoning on ane open parts
of X shows that also condition (ii) carries over to this morphism. Thus, writing
X in place of V again, it is enough to show that f (X) is closed in Y .
Since f : X - Y is quasi-compact, Lemma 15 becomes applicable and it
is only to show that f (X) is stable under specialization. To do this we proceed
similarly as in the proof of Proposition 17. For a point x X and its image
y = f (x) f (X), look at a specialization y0 {y} and view Z = {y} as
a closed subscheme in Y , providing it with its reduced structure, as in 7.3/5.
Then the local ring OZ,y0 is a subring of the eld OZ,y = k(y), the latter being
a subeld of the eld K = k(x). By Lemma 16 we can nd a valuation ring R
with eld of fractions K dominating OZ,y0 . If is the generic and s the closed
point of Spec R, the inclusion OZ,y0 - R induces a morphism
Spec R - Z - Y,
- y, s - y0 ,
such that the composition with i : Spec K - Spec R lifts to a Y -morphism

h : Spec K - X. Then, by condition (ii), the latter extends to a Y -morphism
h : Spec R - X so that necessarily y0 = (f h)(s) f (X). Hence, f (X) is
stable under specialization and therefore closed in Y .
Let us point out that it is enough for the valuative criteria in Propositions 17
and 18 to require condition (ii) just for discrete valuation rings in place of
general valuation rings if we require some additional niteness conditions for
f: X - Y . Namely, we need that Y is locally Noetherian and, especially for
the separatedness in Proposition 17, that f is, in addition, locally of nite type;
see EGA [12], II, 7.2.3 and 7.3.8.
Exercises
1. Show that the structural morphism of the ane n-space AnR over a ring R is not
universally closed, unless n = 0 or R = 0. Excluding these trivial cases, deduce
that AnR cannot be proper over R.
2. Let f : X - Y be a proper morphism between ane schemes. Show that f
is nite. Hint: Since X is of nite type over Y , there exists a closed immersion
X - AnY into some ane n-space over Y . Reduce by induction to the case
where n = 1. Then interpret A1Y as the open part D+ (t1 ) of the projective line
P1Y = ProjR Rt0 , t1 where R = OY (Y ) and show that the resulting morphism
X - P1Y is proper. Conclude that the image of X is closed in P1Y and that the
associated graded ideal I Rt0 , t1 satises (t0 , t1 ) rad(I, t1 ). Finally, derive
an integral equation for tt01 over R modulo the ideal induced from I in R tt01 .
3. Let K be a eld and X a proper K-scheme that is irreducible. Show that every
K-morphism X - Y into an ane K-scheme Y of nite type is constant in
the sense that the image of f consists of precisely one point. Hint: Use Exercise 2.
9.6 Abelian Varieties are Projective 477
4. Let S be an ane base scheme and X a proper S-scheme, as well as Y a separated

S-scheme. Show that an open S-immersion X - Y is an isomorphism as soon
as Y is connected.
5. Let S be an ane base scheme and X a proper S-scheme. Show that X is pro-
jective if and only if there exists an ample invertible sheaf on X.
6. Let R be a discrete valuation ring with eld of fractions K. Show:
(a) The canonical map HomR (Spec R, P1R ) - HomR (Spec K, P1 ) is bijective.
R
(b) The canonical map HomR (Spec R, A1R ) - HomR (Spec K, A1 ) is injective,
R
but not bijective.
(c) There exists a separated R-scheme X such that X R K A1K and the map
HomR (Spec R, X) - HomR (Spec K, X) is bijective.
(d) Any R-scheme X as in (c) cannot be of nite type.
7. Use the valuative criterion of properness to show once more that the projective
n-space PnR over a ring R is proper.
8. Give an example of a morphism of schemes X - S with a point s S such
that the ber Xs of X over s is proper, but where for each open neighborhood
S S of s the induced morphism X S S - S will not be proper.
9. Let f : X - S be a proper morphism of schemes and let L be an invertible
sheaf on X that is ample relative to f . Show for any global section l (X, L)
that f restricts to an ane morphism Xl - S. In particular, Xl is ane if S
is ane.
10. Let X be a scheme of nite type over the eld of complex numbers C. Then
X(C), the set of C-valued points of X, carries a natural topology inherited from
the topology of C, the so-called complex topology. Indeed, if U is an ane open
piece in X, choose a closed immersion U - AnC into some ane n-space and
provide U (C) with the restriction of the complex topology on AnC (C) = Cn .
Granting the assertion of Chows Lemma, show:
(a) The complex topology on X(C) is well-dened. Any morphism X - Y
between C-schemes of nite type yields a continuous map X(C) - Y (C).
Furthermore, the complex topology respects cartesian products.
(b) X is separated over C if and only X(C) is Hausdor.
(c) If X is proper over C if and only if X(C) is compact.
Hint: Consult Mumford [20], 1, 10, but pay attention to the fact that our de-
nition of properness leaves the context of schemes of nite type over C.
9.6 Abelian Varieties are Projective
As an application of the theory of ample and very ample invertible sheaves from
Section 9.4 we want to prove that any abelian variety A over a eld K admits a
closed immersion A - PnK into some projective n-space, in other words, that
abelian varieties are projective in the sense of 9.5/7. Indeed, we will show that
certain general properties of abelian varieties guarantee the existence of ample
and, hence, very ample invertible sheaves. Of course, we cannot supply a full
treatment of abelian varieties at this place. We will restrict ourselves to a few
basic facts and refer to the excellent book of Mumford [21] for more details. In
the following let K be a eld with algebraic closure K.
Denition 1. An abelian variety over K is a proper smooth K-group scheme

that is irreducible.
Let us explain the terminology applied in the denition. A K-group scheme

is a scheme A together with K-morphisms
: A K A -A (group law),
: Spec K -A (unit section),
i: A -A (inverse)
such that the standard group axioms are satised. Using the structural mor-
phism p : A - Spec K, we thereby mean that the following diagrams are
commutative:
id
A K A K A - A K A
id (associativity)
?
?
A K A - A
(p,id)
A - Spec K K A
id id (unit section)
?
?
A A K A
(i,id)
A - A K A
p (inverse)
? e
?
Spec K - A
Since the occurring products are ber products in the category of schemes and
not just ordinary cartesian products of sets, we cannot claim that the group
law of a K-group scheme A denes a group structure on its underlying set.
However, induces for every K-scheme Z a true group structure on the set of
Z-valued points A(Z) = HomK (Z, A), namely by associating to a pair (x, y) of
points in A(Z) the composition
(x,y)
Z - A K A - A.
Furthermore, the formation of the group A(Z) is functorial in Z in the sense that
any K-morphism Z - Z gives rise to a group morphism A(Z) - A(Z ).
On the other hand, it is easily checked that a functorial group law on the point
functor A() equips A with the structure of a K-group scheme. Of course, we

may replace K by an arbitrary base scheme S, thereby arriving at the notion
of S-group schemes.
The notion of properness has been introduced in 9.5/4. It means that the
structural morphism p : A - Spec K is of nite type, separated, and univer-
sally closed. For the property of smoothness see 8.5/1. It follows from 8.5/15
in conjunction with 2.4/19 that all stalks OA,x of a smooth K-group scheme A
are integral domains. Since abelian varieties are required to be irreducible, they
give rise to integral schemes. Also let us mention that for K-group schemes of
nite type smooth is equivalent to geometrically reduced, which means that all
stalks of the structure sheaf of AK K are reduced. In addition, let us point out
that for K-group schemes of nite type the property irreducible can be checked
after base change with K/K so that we may replace irreducible by geometrically
irreducible.
As one of the rst results on abelian varieties one shows:
Proposition 2 ([21], II.4, Question 4 (ii)). The group law on an abelian variety
A is commutative, i.e. the diagram

A K A - A K A

? ?
A ========= A
where is the morphism interchanging factors, is commutative.
Relying on this fact, the group law on an abelian variety is usually written
additively. For any K-valued point x A(K) one can consider its associated
translation x on A, given by z - x + z. More precisely, we thereby mean the
K-morphism
(p,id) xid
x : A - Spec K K A - A K A - A.
In a similar way it is possible to translate with a K-valued point, although before
doing so we must apply the base change K/K, replacing A by the K-group
scheme A K K. We will derive the projectivity of abelian varieties from the
following basic result:
Theorem 3 (Theorem of the Square, [21], II.6, Cor. 4). Let A be an abelian
variety over K and L an invertible sheaf on A. Then

x+y (L) L x (L) y (L)
for K-valued points x, y A(K).
To construct an ample invertible sheaf on an abelian variety A, we use the

equivalence between classes of Weil divisors, classes of Cartier divisors, and iso-
morphism classes of invertible sheaves on A, as discussed in 9.3/16. For the
necessary premises, observe rst that A is of nite type over K and, thus,
Noetherian, by Hilberts Basis Theorem 1.5/14. Furthermore, the smoothness
of A implies that all local rings of A are regular (8.5/15) and therefore factorial
as well; see the Theorem of AuslanderBuchsbaum in [24], Cor. 4 of Thm. IV.9.
Also note that the above equivalence between classes of Weil or Cartier divisors
and invertible sheaves is compatible with translations. Indeed, for a prime divi-
sor D PD(A) and a point x A(K) the pull-back x1 (D) = x (D) under the
translation x is a prime divisor again, and we can dene the pull-back under
x of divisors in Div(A) by

x nD D = nD x1 (D).
DPD(X) DPD(X)
Extending this construction to Cartier divisors via the equivalence of 9.3/16,

one easily shows that there is a canonical isomorphism
OX (x (D)) x (OX (D))
on the level of associated invertible sheaves.
Proposition 4. Let A be an abelian variety over K. Then there exists a closed

immersion A - PnK into some projective n-space over K.
Proof. As we will see, it is enough to construct an ample invertible sheaf on A. To

do so, choose a non-empty ane open subset U A. Since A is Noetherian, its
complement3 A U is a nite union of irreducible closed subsets Z1 , . . . , Zr A.
In fact, the Zi are prime divisors on A, as one can conclude from EGA [14], IV,
21.12.7. However, it is not really necessary to use this result. One can enlarge
each Zi to a 1-codimensional irreducible closed subset Di A; just apply the
niteness result on Krull dimensions for local Noetherian rings 2.4/8 to suitable
ane open parts of A. Thereby we arrive at an eective Weil divisor
D = D 1 + . . . + Dr
r
on A with support |D| = i=1 Di , whose complement A |D| is contained in
U and, thus, is at least quasi-ane.
In order to apply the Theorem of the Square to the invertible sheaf associ-
ated to D, we need to translate D under the group law of A. Doing so, we will
identify K-valued points of A with their corresponding closed points in A. Also
we will use for an algebraically closed eld K = K that the K-valued points of
A are dense in any open or closed subset of A. This is a consequence of Hilberts
Nullstellensatz 3.2/6.
Lemma 5. If K is algebraically closed, there exists for every z A(K) a point

x A(K) such that
z x (|D|) x (|D|).
3
Observe that in this section we use a bold version for the set theoretic minus sign; this
is to avoid the interference with minus signs attached to the group law on A.
Proof. First, z (|D| + x) (|D| x) is equivalent to
x (z |D|) (|D| z).
Next observe that Z = (z |D|) (|D| z) is a nite union of prime divisors in

A and, hence, must be strictly contained in A, since A is irreducible. Therefore
the complement of Z in A is open and non-empty. In particular, it contains a
K-valued point by Hilberts Nullstellensatz 3.2/6.
Now let I(D) OA be the sheaf of ideals consisting of all functions in OA

vanishing on |D|. Then, as in the proof of 9.3/14 (iii), one shows that I(D)
is an invertible OA -module. Picking local generators of I(D), we pass to the
associated Cartier divisor, and further to the attached invertible sheaf on A;
the latter is the inverse OA (D) of I(D). Then observe that the injection
I(D) - OA may be tensored with OA (D) and thereby yields an injection of
OA -modules
OA - OA (D).
In particular, the image of the unit section in OA gives rise to a global section
s (A, OA (D)). The latter generates the sheaf OA (D) precisely at those points
where I(D) coincides with OA , namely, on the complement A |D|. The same
considerations are valid for the pull-backs of D with respect to translations x
given by K-valued points x A(K).
We claim that the invertible sheaf L = OA (D) is ample on A. By a descent
argument for ample invertible sheaves, such as the one mentioned in [5], 6.1/7,
we may reduce the assertion to the case of an algebraically closed eld K.
Therefore we assume K to be algebraically closed in the following. Applying
Theorem 3 for y = x, we obtain isomorphisms
L2 x (L) x

(L) OA (x (D)) OA (x

(D))
for arbitrary x A(K). From what we have seen above, we know that
OA (x (D)) admits a global section s generating this sheaf precisely on the com-
plement
Ux = A |x (D)| = A x (|D|).

Likewise, OA (x (D)) admits a global section s generating this sheaf precisely
on the complement
Ux = A
|x (D)| = A x (|D|).
Then s s generates the tensor product OA (x (D)) OA (x

(D)) precisely on

Ux Ux = A x (|D|) x (|D|)
and this open subset of A is quasi-ane, since Ux and Ux enjoy this property
as translates of A |D| U . Indeed, Ux Ux is contained in the intersection
x (U ) x (U ) and the latter is ane by 7.4/6, since U is ane and A is
separated over K.
Now we read from Lemma 5 that for every K-valued point z A(K)
there is some x A(K) such that z does not belong to the support of the
divisor x (D) + x

(D). Thus, there exists for every z A(K) a global section
sz (A, L ) such that Asz is quasi-ane and L2 is generated in z by sz ,
2
thus showing z Asz . Since A is quasi-compact, it follows that A is covered by

nitely many open subsets of type Asz and we see that L is ample.
Then it follows from 9.4/10 that a certain power Lm is very ample (where
it can be shown that m = 3 is sucient). Thus, there exists an immersion
f : A - PnK into a certain projective n-space satisfying f (OP (1)) Lm . As
PnK is separated over K and the composition A - PnK - Spec K is proper,
f is proper as well; see 9.5/6. In particular, f (A) is closed in PnK and, hence, f
is a closed immersion by 7.3/11. Thus, A is projective and we are done.
Exercises
1. Let G be an S-group scheme over some base scheme S, given by the group law
: G S G - G, a left unit : S - G, and a left inverse i : G - G, as
specied above. Show that just as in the case of abstract groups, and i enjoy
the properties of a right unit and of a right inverse as well.
2. Let G be a K-group scheme over a eld K. Show that G is a separated K-scheme.
Hint: The unit section : Spec K - G corresponds to a closed point e G, and
the diagonal G K G equals the preimage of e with respect to the morphism
G K G - G that is symbolically characterized by (x, y) - x1 y.
3. For an integer n N consider the functor Gln : Sch - Set that associates to a
scheme T the set of OT (T )-linear automorphisms of (OT (T ))n . Show that there
is a natural law of composition : Gln Gln - Gln equipping Gln with the
structure of a functor from Sch to the category of groups Grp. Show that Gln is
representable by an ane Z-group scheme, i.e. that there exists an ane Z-group
scheme G whose associated functor of points Sch - Grp is isomorphic to Gln .
The resulting group scheme G is denoted by Gln again and is called the general
linear group of index n.
4. Show that abelian varieties are commutative group schemes. Hint: Fix an abelian
variety A over a eld K and write AT for the T -group scheme derived from A via
a base change T - Spec K. Then, for a T -valued point g : T - A, consider
the T -morphism g : AT - AT that is the composition of the left translation
with g and the right translation with g 1 on AT . So g can symbolically be
described by x - gxg 1 . Observe that the unit section T : T - AT is a
closed immersion; let IT OAT be the associated quasi-coherent ideal. Show
that the quotients ITm /ITm+1 are free OT -modules for all m N. Observing that
g leaves the unit section T invariant, deduce for every m that g gives rise
to an OT -linear automorphism of ITm /ITm+1 . Based on such automorphisms, use
Exercise 3 to construct morphisms of K-group schemes A - Gln,K for suitable
integers n N. Conclude from Exercise 9.5/3 that the latter are constant and,
nally, that g will be the identical morphism on AT .
5. Elliptic curves via the Theorem of RiemannRoch: Let C be a smooth projective
curve over a eld K which, for simplicity, is supposed to be algebraically closed.
In this context the term curve means an integral K-scheme of dimension 1, au-
tomatically of nite type in our situation. Then the equivalence between classes
of Weil divisors, Cartier divisors, and invertible sheaves of 9.3/16 is valid on C
and we can consider a so-called canonical divisor K, namely, a Weil divisor cor-
responding to the invertible sheaf of dierential forms 1C/K on C. Furthermore,
the famous Theorem of RiemannRoch applies to C; see for example Hartshorne
[15], IV.1.3. It states that there exists an integer g 0, the so-called genus of C,
such that for every Weil divisor D on C the following equation holds:

dimK C, OC (D) dimK C, OC (K D) = deg D g + 1
Here OC (D) is the invertible sheaf corresponding

for OC (K D),
to D, likewise
and the degree of D is given by deg D = si=1 ni if D = si=1 ni Di for prime
divisors Di on C. Smooth projective curves of genus 1 are called elliptic curves.
Show:
(a) The projective line P1K is a smooth projective curve of genus 0.
(b) Let C be the closed subscheme of the projective plane P2K = ProjK Kx, y, z
that is given by an equation of type y 2 z = x3 + xz 2 + z 3 where , K and
4 3 + 27 2 = 0. Assume char K = 2, 3. Then C is an elliptic curve.
(c) Consider an elliptic curve C and x a closed point O C, for example, the
point with homogeneous coordinates (0 : 1 : 0) in the situation of (b). Then the
map C(K) - Div(C), P - (P ) (O), where (P ) and (O) stand for the
prime divisors represented by P and O, gives rise to a bijection between the set
of K-valued points of C and the group of Weil divisors on C of degree 0. In
particular, C(K) can be equipped with a group structure such that O becomes
the unit element.
In the situation of (c) one can show that the group structure on C(K) is induced
from a structure on C as a K-group scheme. In particular, elliptic curves are
examples of abelian varieties; in fact, the only examples of abelian varieties in
dimension 1. Consult the book of Silverman [25] for further details on elliptic
curves.
Literature
1. M. Artin: Grothendieck Topologies, Notes on a seminar by M. Artin, Harvard

University (1962)
2. M. F. Atiyah, I. G. Macdonald: Introduction to Commutative Algebra,
AddisonWesley (1969)
3. S. Bosch: Algebra, Springer, 7. Auflage (2009)
4. S. Bosch, W. Lutkebohmert: Formal and rigid geometry I. Rigid spaces. Math.
Ann. 295, 291317 (1993)
5. S. Bosch, W. Lutkebohmert, M. Raynaud: Neron Models, Springer (1990)
6. N. Bourbaki: Algebre Commutative, Chap. IIV, Masson (1985)
7. R. Godement: Theorie des Faisceaux, Hermann Paris (1964)
8. U. Gortz, T. Wedhorn: Algebraic Geometry I. Schemes with Examples and
Exercises. Vieweg + Teubner (2010)
9. A. Grothendieck: Sur quelques points dalgebre homologique, Tohoku Math. J.
9, 119221 (1957)
10. A. Grothendieck: Fondements de la Geometrie Algebrique. Sem. Bourbaki, exp.
149 (1956/57), 182 (1958/59), 190 (159/60), 195 (159/60), 212 (1960/61), 221
(1960/61), 232 (1961/62), 236 (1961/62), Benjamin, New York (1966)
11. A. Grothendieck and J. A. Dieudonne: Elements de Geometrie Algebrique I,
Springer (1971)
12. A. Grothendieck and J. A. Dieudonne: Elements de Geometrie Algebrique II,
Publ. Math. 8 (1961)
13. A. Grothendieck and J. A. Dieudonne: Elements de Geometrie Algebrique III,
Publ. Math. 11, 17 (1961, 1963)
14. A. Grothendieck and J. A. Dieudonne: Elements de Geometrie Algebrique IV,
Publ. Math. 20, 24, 28, 32 (1964, 1965, 1966, 1967)
15. R. Hartshorne: Algebraic Geometry, Springer (1977)
16. P. J. Hilton and U. Stammbach: A Course in Homological Algebra, Springer
(1971)
17. S. Lang: Algebra, 3rd ed., Addison-Wesley (1993)
18. Q. Liu: Algebraic Geometry and Arithmetic Curves, Oxford University Press
(2002)
19. J. S. Milne: Etale Cohomology, Princeton University Press (1980)
20. D. Mumford: The Red Book of Varieties and Schemes, Lecture Notes in Math-
ematics 1358, Springer (1999), preliminary version Harvard (1967)
21. D. Mumford: Abelian Varieties, Oxford University Press (1970)
22. M. Nagata: Local Rings, Interscience Publishers (1962)
23. M. Raynaud: Anneaux Locaux Henseliens, Lect. Notes in Math. 169 (1970)

DOI 10.1007/978-1-4471-4829-6, Springer-Verlag London 2013
486 Literature
24. J.-P. Serre: Algebre Locale Multiplicites, Lect. Notes in Math. 11 (1965)
25. J. H. Silverman: The Arithmetic of Elliptic Curves, Springer (1986)
26. A. Weil: Foundations of Algebraic Geometry, Amer. Math. Soc. Colloquium
Publ. 29 (1946) (revised and enlarged edition 1962)
Glossary of Notations
empty set
N natural numbers, including zero
Z ring of integers
Q eld of rational numbers
R eld of real numbers
C eld of complex numbers
RX polynomial ring 10
0 zero ring 10
R
group of units of a ring 10
iI Rai ideal generated by a family of elements 11
(a)
principal ideal 11
iI ai sum of ideals 12
iI ai intersection of ideals 12
n
i=1 ai product of ideals 12
(a : b) ideal quotient 12
Ann(b) annihilator of an ideal 12
ker kernel of a ring morphism 12
im image of a ring morphism 12
R/a residue class ring 12
Spec R prime spectrum of a ring 16
Spm R maximal spectrum of a ring 16
px prime ideal given by a point of a spectrum 16
V (a) zero set of an ideal 16
D(f ) basic open set of a spectrum 16
a map on spectra induced from a ring morphism 17
Z(p) localization of the ring of integers at a prime 18
RS localization of a ring 20
S 1 R localization of a ring 20
Q(R) eld of fractions of an integral domain 21
K(X) rational function eld in one variable 21
Rp localization of a ring at a prime ideal 21
Rf localization of a ring by some element 21
Rf 1 localization of a ring by some element 21
aRS extension of an ideal to a localization of a ring 21
bR restriction of an ideal 21
j(R) Jacobson radical of a ring 26
mx maximal ideal attached to a point 26

488 Glossary of Notations
rad(R) nilradical of a ring 27

j(a) Jacobson radical of an ideal 28
rad(a) nilradical of an ideal 28
I(Y ) ideal of functions vanishing on a set 29
ker kernel of a module morphism 32
im image of a module morphism 33
M/N residue class module 33

Rxi submodule generated by a family of elements 34
iI
N sum of submodules 34
iI i
iI Ni direct sum of submodules 34
xi1 . . . xis element of a direct sum of modules 34

M direct product of a family of modules 34
iI i
iI Mi direct sum of modules 35
R(I) direct sum of copies of a ring 35
HomR (M, N ) set of module morphisms 35
aM submodule induced from an ideal 36

iI Mi intersection of submodules 36
coker f cokernel of a module morphism 40
RX ring of formal power series 54

a = ri=1 qi primary decomposition of an ideal 59
Ass(a) set of prime ideals associated to an ideal 61
Z(a) set of zero divisors modulo an ideal 62
Ass (a) set of isolated prime ideals associated to an ideal 63
p(n) symbolic power of a prime ideal 65

R = iN ai graded ring constructed from an ideal 72

M = iN Mi graded module 72
dim R Krull dimension of a ring 74
ht p height of a prime ideal 74
ht a height of an ideal 74
coht a coheight of an ideal 74
ij Kroneckers delta 87
R xed ring 90
Vmax (a) zero set of an ideal 94
VK (a) zero set of an ideal 95
PR restricted ideal 96
S 1 p extended ideal in a localization 96
AutK (K ) group of automorphisms 98
transgradK (Q(A)) transcendence degree of a eld of fractions 101
M R N tensor product of modules 108
xy tensor product of two elements 108
M1 R . . . R Mn tensor product of modules 112

E/R restriction of a module 113
hE Hom functor 115
BilR (M N, E) module of bilinear maps 115
MS localization of a module 125
Mp localization of a module at a prime ideal 125
Mf localization of a module by some ring element 125
Ob(C) objects of a category 138
Glossary of Notations 489
Hom(X, Y ) morphisms between objects of a category 138

idX identity morphism 138
HomC (X, Y ) morphisms between objects of a category 138
Set category of sets 138
Grp category of groups 138
Ring category of rings 138
R-Mod category of modules over a ring 138
C0 dual category 138
Cop opposite category 139
CS category of relative objects 139
HomS (X, Y ) set of relative morphisms 139
CR category of relative objects 139
R-Alg category of algebras under a xed ring 139
X S Y ber product in a category 139
X R Y amalgamated sum in a category 141
Spec R ane scheme associated to a ring 141
F (g) image of a morphism under a functor 141
X morphisms making up a functorial morphism 142
p M pull-back of a module under a morphism 143
p pull-back of a morphism under a morphism 143
R -Mod-DD category of modules with descent data 148
dn boundary map of a chain complex 159
Zn submodule of cycles 159
Bn submodule of boundaries 159
Hn homology module 160
M chain complex 160
Hn (M ) homology of a chain complex 160
M cochain complex 160
Zn submodule of cocycles 160
Bn submodule of coboundaries 160
H n (M ) cohomology of a cochain complex 160
Hn (f ) homology morphism attached to a complex morphism 163
TorRn (M, E) Tor module 172
M R E single complex associated to a tensor product of complexes 173
HomR (M, ) Hom functor on a category of modules 181
HomR (, N ) Hom functor on a category of modules 181
Z(p ) p-quasi-cyclic group 187
ExtnR (M, N ) Ext module, via resolutions in the rst variable 187
Ext nR (M, N ) Ext module, via resolutions in the second variable 188
HomR (M , N ) single complex associated to the Hom of complexes 188
Ext(M, N ) extensions of modules 191
Ax localization of a ring at a point 203
k(x) residue eld at a point 203
rad(E) nilradical of the ideal generated by a set 204
V (E) zero set 204
V (f ) zero set of an element 204
D(f ), DX (f ) domain of an element 204
I(Y ) vanishing ideal 206
a map between spectra induced from a ring morphism 213

Opn(X) category of open subsets of a topological space 216
VU restriction morphism on a sheaf or presheaf 217
F|X restriction of a sheaf or presheaf 217
f |U restriction of an element of a sheaf or presheaf 217
OF cartesian product of functors 217
ker(1 , 2 ) kernel of a pair of maps 218
D(X) category of basic open subsets of a spectrum 219
D (X) variant of the category of basic open subsets 219

OX variant of the structural presheaf on a spectrum 219
OX structural presheaf on a spectrum 220
S(f ) multiplicative system generated by a basic open set 220
AS(f ) localization attached to a basic open set 220
(Gi , fij )i,jI inductive system 222
lim Gi inductive limit 223
P(U ) power set 223

(Gi , fij )i,jI projective system 226
lim Gi projective limit 227
SetS category of relative sets 228

Fx stalk of a sheaf at a point 228
fx germ of a function 228
HomO (F, G) set of morphisms between module sheaves 232
(ker )pre kernel of a morphism of presheaves 234
(im )pre image of a morphism of presheaves 234
(coker )pre cokernel of a morphism of presheaves 234
H 0 (U, F) Cech cohomology group 235
F+ partial sheacation of a presheaf 236
F ++ sheacation of a presheaf 236
ker kernel of a morphism of sheaves 238
im image of a morphism of sheaves 238
coker cokernel of a morphism of sheaves 238
Df (g) domain of an element on a basic open set 241
OX structure sheaf 246
M module sheaf associated to a module over a ring 247
(X, OX ) ringed space 247
(f, f # ) morphism of ringed spaces 247
f (OX ) direct image of structure sheaf 247
fx# morphism on the level of stalks 248
(U, OX ) ring of sections 252
Sch category of schemes 252
Hom(X, Y ) set of morphisms between schemes 252
Sch/S category of relative schemes 252
HomS (X, Y ) set of morphisms between relative schemes 252
AnR , AnS ane n-space 255
AnS (R ) set of points of the ane n-space with values in a ring 255
f (F) direct image sheaf 266
f 1 (G) inverse image sheaf 266
f 1 inverse image functor 266
f direct image functor 266

F O G tensor product of module sheaves 269
f (G) inverse image of a module sheaf 269
j! F extension of a sheaf by zero 276
A sheaf associated to an algebra 287
OS t sheaf of polynomials 287
HomOS set of morphisms between sheaves of algebras 288
Spec A spectrum of a quasi-coherent algebra 288
AnS ane n-space over a general base 289
Pn (K) points of the projective n-space 289
(x0 : . . . : xn ) point of a projective n-space 290
PnS , PnR projective n-space over a base scheme 291
PnK (K ) points of the projective n-space with values in a eld 291
hX point functor on the category of schemes 292
Ga additive group as a functor 292
Gm multiplicative group as a functor 292
ha
X point functor on the category of ane schemes 292
Gm,R multiplicative group over an ane base 292
X/ quotient of a scheme by a nite group 293
V (I) zero set of an ideal sheaf 305
supp OX /I support of a module sheaf 306
IY vanishing ideal sheaf of a set 307
dim X dimension of a topological space 320
dimx X local dimension of a topological space at a point 321
codimX Z codimension of an irreducible subset in a topological space 321
Ui0 ...iq intersection of open sets 322
C q (U, F) group of Cech cochains 322
Sq+1 symmetric group 322
Caq (U, F) group of alternating Cech cochains 322
dq Cech coboundary map 322
C (U, F) Cech complex 323
Ca (U, F) alternating Cech complex 323
H q (U, F) Cech cohomology group 323
Haq (U, F) alternating Cech cohomology group 323
q (U, V) morphism between Cech complexes induced by renement 326
Cov(X) collection of open coverings of a topological space 327
UV renement relation for coverings 327
Xf open subset of a scheme where a given function is non-zero 336
DerR (A, M ) module of derivations 344
1A/R module of relative dierential forms of degree 1 346
dA/R exterior dierential 346
nA/R module of relative dierential forms of higher degree 346
1X/S sheaf of relative dierential forms on a scheme 356
dX/S exterior dierential 356
nX/S sheaf of relative dierential forms of higher degree 356
f () pull-back of a dierential form 358
g left translation on a group scheme 360
dimx f relative dimension of a scheme morphism at a point 375

nZ An graded ring 404
A(f ) homogeneous localization of a graded ring 406
A+ irrelevant ideal of a graded ring 407
Proj A homogeneous prime spectrum of a graded ring 407
V+ (E) zero set on a homogeneous prime spectrum 407
rad+ (a) restricted radical in a graded ring 408
D+ (f ) basic open subset of a homogeneous prime spectrum 408
I+ (Y ) restricted vanishing ideal 411
D + (Y ) category of basic open subsets of a homogeneous spectrum 413
Proj A Proj scheme associated to a graded ring 414
ProjR A relative Proj scheme associated to a graded ring 414
ProjS A Proj scheme of a quasi-coherent sheaf of algebras 418
M (d) graded module obtained by shifting degrees 418
M(f ) homogeneous localization of a graded module 418
)
M module sheaf associated to a graded module 418
OX (n) Serre twist of the structure sheaf 421
Pic(X) Picard group of a scheme 423
HomOX (L, OX ) Hom sheaf of morphisms between module sheaves 423
PicU (X) Picard group relative to an open covering 427
UV product covering 427
S(A) set of regular elements of a ring 434
S X OX subsheaf of regular elements 435
MX sheaf of meromorphic functions on a scheme 435
MX presheaf inducing the sheaf of meromorphic functions 435
KX constant sheaf 436
PD(X)
set of prime divisors on a scheme 437
DPD(X) nD D Weil divisor on a scheme 437
Div(X) group of Weil divisors on a scheme 438
(f ) principal divisor attached to a meromorphic function 438
Cl(X) Weil divisor class group on a scheme 439
deg D degree of a Weil divisor on a projective space 439
MX sheaf of invertible meromorphic functions 440
OX sheaf of invertible functions of a structure sheaf 440
CaDiv(X) group of Cartier divisors on a scheme 440
CaCl(X) Cartier divisor class group on a scheme 441
OX (D) subsheaf of meromorphic functions associated to a divisor 442
X - Y rational map of schemes 445
V (l) zero set of a global section of an invertible sheaf 446
Xl set where a global section generates an invertible sheaf 446
Li tensor power of an invertible sheaf 447
X set where global sections generate powers of a sheaf 453
Pl basic open subscheme of a particular Proj scheme 454
A R B cartesian product of graded R-algebras 462
suppR M support of a module 467
A= m=0 Am quasi-coherent sheaf of graded algebras 468
SymR (M ) symmetric algebra of a module 468
TR (M ) tensor algebra of a module 468
M m tensor power of a module 468
Symm R (M ) homogeneous part of a symmetric power of a module 469

SymOS (E) symmetric algebra of a module sheaf 469
PS (E) Proj scheme of the symmetric algebra of a module sheaf 469
Gln general linear group as group scheme 482
deg D degree of a Weil divisor on a curve 483
Index
abelian variety, 402., 478. annihilator, 12, 311, 468

of dimension 1, 483 anticommutative diagram, 173
acyclic arrow.
module, 171 See morphism
resolution, 171, 339 Artinian
additive module, 66.
functor, 158, 164, 166 ring, 56, 66.
group, as functor, 292 associated
adic graded module, 430
completion, 231 module sheaf, 247, 418.
topology, 71, 231 prime ideal, 56, 61
adjoint functor, 240, 266. embedded, 63
adjunction, 266. isolated, 63, 75
ane sheaf of algebras, 287
line, 216 sheaf to a presheaf, 230, 234.
with double origin, 277 augmentation, 323
morphism of schemes, 314, 462. augmented Cech complex, 323
open set, 252 automorphism
scheme, 141, 202, 251, 254, 336 of modules, 31
n-space, 255., 286, 289, 399 of rings, 10
algebra, 32, 128
base
of nite presentation, 361.
change, 302
of nite type, 55, 84, 91, 361.
for dierential modules, 350
algebraic
scheme, 252
curve, 483
basic open set, 16, 202, 205, 400, 408
equation, 83 bifunctor, 173
extension, 83, 85 bilinear map, 107
variety, 445 blowing up, 401., 430
algebraically independent, 91 boundary
alternating Cech element, 160
cochain, 322 map, 160
cohomology group, 323, 327
complex, 323 canonical
amalgamated sum, 140, 192 divisor, 483
ample invertible sheaf, 399, 402, 452, sheaf, 445
469, 477, 477. Cartans Theorem, 335
alternative characterization, 461 cartesian

496 Index
diagram, 295 CohenSeidenberg Theorems, 96.

product, 140, 294 coherent
of graded algebras, 462 module, 9, 47, 50.
Cartier divisor, 440. sheaf, 265
class group, 441 ring, 47, 53
categorical cohomological resolution, 159, 182
epimorphism, 350 cohomology.
monomorphism, 311 See also Cech cohomology
category, 131, 138. modules, 160
of ane schemes, 141 of ane schemes, 335
of relative objects, 139 of schemes, 280, 332
Cauchy sequence, 231 via derived functors, 330.
Cech coincidence scheme, 318, 373
cochain, 322 cokernel
cohomology, 263, 322., 425 of a module morphism, 40
of ane schemes, 335 of a presheaf morphism, 234
cohomology group, 323, 327 of a sheaf morphism, 238
complex, 323 collection, 138
resolution, 329 compactication, 399
chain complex
complex, 159 morphism.
condition See morphism of complexes
ascending, 48, 67 of modules, 38
descending, 66 complex topology on a scheme, 477
Chinese Remainder Theorem, 17 connected, 210
Chows Lemma, 399, 469 constant sheaf, 266, 436
closed continuous map, 214
immersion, 215, 279, 307. contravariant functor, 141
morphism of schemes, 464 convergent power series, 231
point, 201 coset
set, 205, 472 of an ideal, 12
subscheme, 279, 304, 306 of a submodule, 33
coboundary covariant functor, 104, 141
element, 160 Cramers rule, 83, 87
map, 160, 322 cup product, 329
cocartesian diagram, 151 curve, 483
cochain complex, 160 cycle, 159
cocycle, 160
condition decomposable ideal, 59
for Cech cocycles, 425 Dedekind, R., 7
for descent data, 107, 147 Dedekind domain, 136
for gluing, 156, 277, 282 degree
codimension of a set, 321, 434 of a complex morphism, 160
coecient of a divisor, 439., 483
extension for modules, 103 derivation, 344
ring, 7 derived functor, 332.
conal system, 223 descent, 143., 397
coheight, 74 datum, 107, 146.
Index 497
trivial, 147 elliptic curve, 341, 403, 483

of module properties, 131. function eld, 342
of modules, 103, 146. embedded prime ideal.
of morphisms, 143 See associated prime ideal
Theorem of Grothendieck, 148 endomorphism
diagonal of modules, 31
embedding. of rings, 10
See diagonal morphism epimorphism
morphism, 279, 312 in a category, 350
dierential form, of modules, 31
See module, resp. sheaf of dierential of rings, 10
forms of sheaves, 238
dimension equivalence
of a ring, 55, 69, 74, 84, 100, 321 of categories, 142
of a topological space, 320 of functors, 142
of polynomial rings, 78. essentially
of the empty topological space, 321 etale algebra, 397
of the zero ring, 321 surjective functor, 148
direct etale
image morphism, 343
functor, 330 morphism of schemes, 374.
sheaf, 247, 266 quasi-sections, 390
limit, 222. scheme, 374.
product topology, 389
of modules, 34 etale
sum sheaf, 232
of modules, 35 space, 229, 275
of sheaves, 239 Euclidean domain, 11
of submodules, 34 exact
directed set, 222 diagram, 144, 218
discrete functor, 182
topology, 219 sequence
valuation ring. of complexes, 160
See valuation ring, discrete of modules, 38
disjoint union of schemes, 254 of sheaves, 238
divisor, 402, 431, 437. Ext
class group, 439 functor, 159, 187.
dominating ring, 473 module, 187
double sequence, 189
complex, 173, 188 extension
origin, on ane line, 277 by zero, 276
dual category, 138 of coecients, 124
of modules, 191
eective exterior dierential, 346, 356
Cartier divisor, 441
Weil divisor, 438 factorial
eigenvalue, 103 ring, 57, 80, 84, 89, 444
eigenvector, 103 scheme, 431
498 Index
faithful functor, 148 etale, 396

faithfully at smooth, 396
module, 105, 121 unramied, 396
morphism formal power series, 231
of rings, 121 free
of schemes, 397 element, 36
ber of a relative scheme, 300 module, 36
ber product, 139. sheaf, 261, 421
categorical facts, 300. resolution, 165
of schemes, 278, 295. full subcategory, 252
points, 298 fully faithful functor, 148
Fibre Criterion for smoothness, 343, 393 functor, 131, 138
eld of fractions, 21 of base change, 302
ltration, 72 of points on a scheme, 292
stable, 72 functorial
nite group law, 292, 478, 482
morphism isomorphism, 142
of rings, 83, 85 morphism, 142
of schemes, 462. properties
presentation, of dierential modules, 350.
See also module, resp. algebra, resp. of immersions, 310
morphism, resp. scheme of separated morphisms, 317
of a module, 43
of a module sheaf, 422 general linear group, 482
type, generators
See module, resp. algebra, resp. mor- of a module, 34
phism, resp. scheme of a module sheaf, 421
nitely generated of an ideal, 11
ideal, 11 generic point, 201, 211, 319
module. genus of a curve, 483
See module of nite type geometrically
module sheaf, 421 reduced scheme, 392
Five Lemma, 53 regular scheme, 392
xed germ, 228
eld, 98 gluing
ring, 90, 102, 293 morphisms, 284
abby sheaf, 329 schemes, 277., 282
asque sheaf, 329 gluing data
at for modules, 156
module, 105, 117 for morphisms, 284
morphism for schemes, 277, 282
of rings, 117 Going-down Theorem, 84, 98
of schemes, 393 Going-up Theorem, 84, 97
atness graded
Bourbaki Criterion, 181 algebra, 400
local characterization, 127 ideal, 404
via Tor modules, 177. module, 418.
formally prime ideal, 400
Index 499
ring, 73, 399, 404. of a presheaf morphism, 234

of type N, 404 of a ring morphism, 12
of type Z, 404 of a sheaf morphism, 238
grading, 404 immersion, 312, 316
graph morphism, 316 Implicit Function Theorem, 341., 389
Grothendieck cohomology, 330.. inductive
See also cohomology limit, 222., 240
group of tensor products, 120
ring, 180 system, 222
scheme, 360, 397, 403, 478, 482 injective
module, 159, 182
Hausdor separation axiom, 71, 201, 313 sheaf, 330
analogue for schemes, 278 resolution, 183., 332
height, 74, 321 integral
henselization of a local ring, 397 algebraic number, 88
Hilberts
closure, 84, 88
Basis Theorem, 9, 50
domain, 10
Nullstellensatz, 84, 94, 95
element, 85
Hilbert polynomial, 80
equation, 83, 85
Hom functor, 115, 187, 331
extension, 83, 85
left exactness, 115, 181
morphism, 83, 85
homogeneous
scheme, 431
component, 404
integrally closed ring, 88
degree, 404
intersection
element, 404
of ideals, 12
ideal, 404
of modules, 36
localization, 400, 406., 417
of quasi-coherent modules, 338
of a module, 418
invariant dierential form, 360
prime
inverse
ideal, 405
image
spectrum, 400., 407
module sheaf, 269
homological resolution, 158, 163
of a subscheme, 311
homology
sheaf, 266.
group, 180
limit, 226
module, 160
homomorphism. Inverse Function Theorem, 341., 389
See morphism invertible
homotopy element, 10
between complex morphisms, 163 sheaf, 261, 401, 423., 442.
equivalence, 166 as Cech cocycle, 425
Hom sheaf, 423 irreducible
component, 56, 319
ideal, 7, 11 ideal, 59
quotient, 12 set, 30, 210, 318.
ideally separated module, 181 topological space, 210
idempotent element, 7 irrelevant ideal, 400, 407
identity morphism, 138 isolated prime ideal.
image See associated prime ideal
of a module morphism, 33 isomorphism
500 Index
in a category, 138 algebra, 397

of functors, 142 dimension, 321
of modules, 31 generators of a module sheaf, 366
of presheaves, 232 morphism of local rings, 248
of rings, 10 ring, 9, 18, 22, 248
of sheaves, 232 localization, 9
at a point, 203
Jacobian of a graded ring, 405
Condition, 342, 374 of a module, 125.
Criterion, 343, 375, 384. of a ring, 19.
matrix, 342 localization morphism, 26
Jacobson locally
radical, 9, 26. closed
ring, 84, 93, 102 immersion, 279, 309
JordanHolder sequence, 70 point, 201, 364
subscheme, 304, 309
Kahler dierential, 342
free
kernel
of a module morphism, 33 module, 106, 133
of a pair of maps, 218 module sheaf, 261, 421
of a presheaf morphism, 234 Noetherian scheme, 281, 319
of a ring morphism, 12 of nite presentation,
of a sheaf morphism, 238 See module sheaf, resp. scheme, resp.
Kolmogorov space, 201, 208, 253 morphism
Krull dimension. of nite type,
See dimension See module sheaf, resp. scheme, resp.
Krulls morphism
Dimension Theorem, 56, 57, 76 ringed space, 248
Intersection Theorem, 57, 71, 73, 75 long exact
Principal Ideal Theorem, 75, 77 homology sequence, 161, 168
Kummer, E., 7 Tor sequence, 157, 162, 172, 174
Lying-over Theorem, 84, 97
left
derived functor, 158, 164, 167 manifold, 341
exact functor, 181 maximal
Lemma ideal, 13
of ArtinRees, 57, 71 spectrum.
of Nakayama, 9, 37, 38 See spectrum
length of a chain, 74 meromorphic function, 435.
Lerays Theorem, 335 module, 7, 31
Lifting Property, 343, 375, 380 of dierential forms, 342, 346.
linear for eld extensions, 355
equivalence of nite presentation, 9, 43
of Cartier divisors, 441 of nite type, 9, 34, 43
of Weil divisors, 439 module sheaf, 217, 257.
map, 107 locally of nite presentation, 264, 422
U -morphism, 428 locally of nite type, 264, 421
line bundle, 423, 428, 429 monomorphism
local in a category, 311
Index 501
of modules, 31 normal
of rings, 10 extension, 98
of sheaves, 238 integral domain, 84, 89
morphism scheme, 431
between projective spaces, 293
between Proj schemes, 416 object of a category, 138
compatible with descent data, 147 open
in a category, 138 immersion, 307
of algebras, 32, 128 of ringed spaces, 247
of complexes, 160 of spectra, 215
of graded rings, 415 set, 205
of locally ringed spaces, 248 subscheme, 252
opposite category, 139
of modules, 31
of presheaves, 232. parameters, 57, 74, 78
of ringed spaces, 247 Picard group, 137, 423
of rings, 10 point
of nite presentation, 361. of codimension 1, 434
of nite type, 361. with value, 255, 291, 292, 298
of schemes, 203, 251 polynomial ring, 7
locally of nite presentation, 362. preorder, 222
locally of nite type, 362., 452 prescheme, 251
of nite presentation, 362 presentation of a module sheaf, 261
of nite type, 362, 452. presheaf, 217.
of sheaves, 232. of constant functions, 219
multiplicative of modules, 217
group, 292 primary
as functor, 292 decomposition
system, 19 of ideals, 55, 59.
of modules, 66
Nakayamas Lemma. ideal, 22, 55, 58.
See Lemma of Nakayama module, 66
natural transformation, 142 part of an ideal, 65
Neiles parabola, 216, 256, 304, 354, 373, prime
396, 401, 445 divisor, 75, 437
Neron model, 292 ideal, 13.
nilpotent element, 7, 27 See also associated prime ideal
nilradical, 27., 204 spectrum.
Noethers Normalization Lemma, 84, 91, See spectrum
96 principal
Noetherian Cartier divisor, 441
module, 9, 47. ideal, 11
ring, 9, 47, 50, 55 domain, 11, 17, 18
scheme, 319 Weil divisor, 438
topological space, 321 product
non-Archimedean covering, 427
absolute value, 432 of functors, 217
triangle inequality, 433 of ideals, 12
non-singular scheme, 390 rule for derivations, 344
502 Index
projection, 139 rational

projective function eld, 21
dimension, 171 map, 445
embedding, 402 reduced
limit, 226. ideal, 28
line, 278 ring, 27
module, 123, 159, 164, 181 scheme, 311, 431
morphism of schemes, 466. structure on a subscheme, 307
resolution, 164 renement of a covering, 235, 326
scheme, 399, 414, 466., 477, 480 regular
n-space, 278, 289., 399., 414, 439, element, 434
460 local ring, 57, 75, 80
homogeneous coordinates, 290 scheme, 390, 431
space bundle, 469 relative
system, 226 dimension, 374376
Proj scheme, 399, 414. object, 139
proper scheme, 252
morphism of schemes, 464. residue class
scheme, 399, 464. eld, 18, 203
Proper Mapping Theorem, 470 module, 33
pseudo-coherent module, 52 ring, 12
pull-back, 143 resolution, 163
of a dierential form, 358 resolution of singularities, 402
of sections, 447 restricted ideal, 96
push-out, 192 restriction
morphism, 217
quadric surface, 462 of a morphism, 252
quasi-ane of a ringed space, 247
morphism of schemes, 462. of coecients, 123
scheme, 402, 451 retraction, 39, 137, 145
quasi-coherent right
ideal sheaf, 305 derived functor, 159
module sheaf, 261. exact functor, 164
sheaf exactness of tensor products, 105, 113
of algebras, 287 ring, 7.
of graded algebras, 468 commutative with unit element, 11
quasi-compact of algebraic numbers, 55
morphism of schemes, 272 of formal power series, 54
set, 208 of fractions, 9
quasi-cyclic group, 187 of Laurent polynomials, 26
quasi-separated ringed space, 141, 201, 247
morphism of schemes, 272
quasi Galois extension, 98 saturation
quotient of a multiplicative system, 221
of a scheme by a group, 293 schematic closure, 452
of modules, 33 scheme, 201., 251..
of sheaves, 238 See also morphism of schemes
ring, 12 locally of nite presentation, 342
Index 503
of nite presentation, 362 stalk, 228.

of nite type, 362 Stein Factorization, 470
section, 39 structural
functor, 330 morphism, 32
of a sheaf, 230, 252 presheaf on a spectrum, 219.
Segre embedding, 459, 462 structure sheaf, 202, 246, 247, 251
separated submodule, 32
morphism, 313. subring, 11
scheme, 279, 313. subscheme, 304.
sequence of modules, 38 subsheaf, 238
Serres of regular elements, 435
criterion, 336 subspace topology, 214
twisted sheaves, 359, 401, 421, 427, sum
430, 469 of ideals, 12
sheaf, 217. of submodules, 34
of algebras, 286 support
of continuous functions, 218, 228 of a divisor, 438
of dierentiable functions, 218 of a module, 468
of dierential forms, 356. sheaf, 266
of holomorphic functions, 231 of a sheaf, 305
of ideals, 276, 304 symbolic power, 65, 75
of locally constant functions, 219 symmetric
of modules. algebra, 468
See module sheaf group, 322
of polynomials, 287 power of a module, 469
sheaf Hom, 240
sheacation, 235 tensor
short exact sequence of modules, 39 algebra, 468
simple module, 70 tensor product, 103, 104
simplicial homomorphism, 324 as amalgamated sum, 141
single complex associated to a double of algebras, 128.
complex, 173, 188 of graded modules, 419
skyscraper sheaf, 240, 276 of modules, 107.
smooth of module sheaves, 269, 419
algebra, 396 of morphisms, 113
locus, 375, 392 Theorem
morphism of schemes, 374. of AuslanderBuchsbaum, 431
scheme, 342, 374. of RiemannRoch, 483
smoothness and regularity, 390. of the Square, 479
Snake Lemma, 9, 40, 162 on Homomorphisms, 8, 13, 33
snake morphism, 41 on Isomorphisms, 33
specialization, 211, 319 topological space, 205
special point, 201 connected, 210
spectral sequence, 176 Tor
spectrum functor, 164
of a quasi-coherent algebra, 278, 288 module, 172.
of a ring, 8, 9, 16, 201, 203. and coecient extension, 180
split exact sequence of modules, 39 torsion-free module, 119
504 Index
torsion module, 157, 187 morphism of schemes, 365.

total scheme, 343, 365.
quotient ring, 21, 434 d-uple embedding, 293, 461
space of an invertible sheaf, 429
transcendence degree, 101 valuation, 432
translation on a group scheme, 360, 479 ring, 90, 470.
trivialization, 429 discrete, 18, 84, 201, 432
trivial sheaf, 218 valuative criterion
U-trivial invertible sheaf, 427 of properness, 475
twisted sheaf. of separatedness, 474
See Serres twisted sheaf vanishing ideal, 29, 206, 307
variable, 8
unit, 10 variety, 445
ideal, 11 vector bundle, 429
universal Veronese embedding, 461
point, 360 very ample invertible sheaf, 453, 469,
property 477
of amalgamated sums, 140
of blowing up, 431 Weil divisor, 437.
of coecient extension, 130
of direct products, 37 Yoneda Lemma, 267
of direct sums, 35, 37
of ber products, 139 Zariskis Main Theorem, 396
of localizations, 23. Zariski topology, 16, 201, 204, 408
of polynomial rings, 8 zero
of residue class modules, 33 divisor, 7, 10, 62
of residue class rings, 13 ideal, 11
of tensor products, 107 sequence, 231
universally closed morphism, 464 set, 16, 29, 30, 204, 305
unramied of a section, 446
locus, 366 sheaf, 218

Bosch - Algebraic Geometry and Commutative Algebra PDF

Uploaded by

Copyright:

Available Formats

Bosch - Algebraic Geometry and Commutative Algebra PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Bosch - Algebraic Geometry and Commutative Algebra PDF

Uploaded by

Copyright:

Available Formats

Universitext

Universitext is a series of textbooks that presents material from a wide variety

For further volumes:

ISSN 0172-5939 ISSN 2191-6675 (electronic)

Library of Congress Control Number: 2012953696

Springer-Verlag London 2013

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

The domain of Algebraic Geometry is a fascinating branch of Mathematics

Therefore it is not my intention to cover as many topics as possible in my book.

Preliminary versions of my manuscripts on Commutative Algebra and

Munster, December 2011 Siegfried Bosch

Part A. Commutative Algebra . . . . . . . . . . . . . . . . . . . . . 1

1 Rings and Modules . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7

2 The Theory of Noetherian Rings . . . . . . . . . . . . . . . . . . . . 55

4 Extension of Coecients and Descent . . . . . . . . . . . . . . . . . 103

5 Homological Methods: Ext and Tor . . . . . . . . . . . . . . . . . . . 157

Part B. Algebraic Geometry . . . . . . . . . . . . . . . . . . . . . . 193

6 Ane Schemes and Basic Constructions . . . . . . . . . . . . . . . . 201

7 Techniques of Global Schemes . . . . . . . . . . . . . . . . . . . . . 277

8 Etale and Smooth Morphisms . . . . . . . . . . . . . . . . . . . . . . 341

9 Projective Schemes and Proper Morphisms . . . . . . . . . . . . . . 399

Glossary of Notations . . . . . . . . . . . . . . . . . . . . . . . . . . . . 487

S. Bosch, Algebraic Geometry and Commutative Algebra, Universitext, 7

aXin11 . . . Xinrr - (a)xni11 . . . xnirr

1.1 Rings and Ideals

Let us recall the denition of a ring.

Denition 1. A set R together with two laws of composition + (addition)

(ii) The multiplication is associative, i.e.

(iii) There exists a unit element in R, which means an element 1 R such

We list some important examples of rings:

An element a of a ring R is called invertible or a unit if there exists some

is a group with respect to the multiplication given on R.

Denition 2. A map : R - R between rings is called a ring homomor-

an epimorphism as a surjective homomorphism.1 A subset R of a ring R is

Convention. From now on, a ring is meant as a commutative ring with 1.

Denition 3. Let R be a ring. A subset a R is called an ideal in R if :

Furthermore, an ideal a R is called nitely generated if, as before, it admits

is again an ideal in R; it is called the sum of the ideals ai . Furthermore, the

of all ideals ai is an ideal in R. For nitely many ideals a1 , . . . , an R we can

which is dened as the ideal generated by all products a1 . . . an where ai ai

Finally, for ideals a, b R their ideal quotient is the ideal given by

Proposition 4. Let : R - R be a morphism of rings. Then

is an ideal in R and im = (R) is a subring of R .

R/a = {r + a ; r R}, where r + a = {r + a ; a a},

of all (additive) cosets of a in R and dene two laws of composition on it by

is a surjective ring homomorphism satisfying ker = a, as desired. The ring

Proposition 5 (Fundamental Theorem on Homomorphisms). Let : R - R

Corollary 6. Let : R - R be a surjection between rings. Then induces

Next we want to discuss prime ideals and maximal ideals.

Denition 7. Let R be a ring.

Proposition 8. Let R be a ring and a R an ideal. Then:

Lemma 9. An ideal m R is maximal if and only if the zero ideal 0 R/m

Lemma 10. The zero ideal 0 R of a ring R is maximal if and only if R is

Proof of Lemma 9. Let : R - R/m be the canonical homomorphism. Then

In order to illustrate the assertion of Proposition 8, consider a prime p Z,

which substitutes X1 , 0 for the variables X1 , X2 . Since KX1  is an integral

Proposition 11. Any ring R = 0 contains a maximal ideal.

is a proper ideal in R. In order to justify this, observe that b contains at least

which substitutes X1 , 0 for the variables X1 , X2 . Since KX1 is an integral

RS - RT /(1 fi ti ; i I).

Proof. The canonical ring homomorphism : R - RT /(1 fi ti ; i I)

mx = {f KX1 , . . . , Xn ; f (x) = 0},